xref: /openbmc/linux/arch/x86/kvm/vmx/vmx.c (revision 5388b581)
1 // SPDX-License-Identifier: GPL-2.0-only
2 /*
3  * Kernel-based Virtual Machine driver for Linux
4  *
5  * This module enables machines with Intel VT-x extensions to run virtual
6  * machines without emulation or binary translation.
7  *
8  * Copyright (C) 2006 Qumranet, Inc.
9  * Copyright 2010 Red Hat, Inc. and/or its affiliates.
10  *
11  * Authors:
12  *   Avi Kivity   <avi@qumranet.com>
13  *   Yaniv Kamay  <yaniv@qumranet.com>
14  */
15 
16 #include <linux/frame.h>
17 #include <linux/highmem.h>
18 #include <linux/hrtimer.h>
19 #include <linux/kernel.h>
20 #include <linux/kvm_host.h>
21 #include <linux/module.h>
22 #include <linux/moduleparam.h>
23 #include <linux/mod_devicetable.h>
24 #include <linux/mm.h>
25 #include <linux/sched.h>
26 #include <linux/sched/smt.h>
27 #include <linux/slab.h>
28 #include <linux/tboot.h>
29 #include <linux/trace_events.h>
30 
31 #include <asm/apic.h>
32 #include <asm/asm.h>
33 #include <asm/cpu.h>
34 #include <asm/debugreg.h>
35 #include <asm/desc.h>
36 #include <asm/fpu/internal.h>
37 #include <asm/io.h>
38 #include <asm/irq_remapping.h>
39 #include <asm/kexec.h>
40 #include <asm/perf_event.h>
41 #include <asm/mce.h>
42 #include <asm/mmu_context.h>
43 #include <asm/mshyperv.h>
44 #include <asm/spec-ctrl.h>
45 #include <asm/virtext.h>
46 #include <asm/vmx.h>
47 
48 #include "capabilities.h"
49 #include "cpuid.h"
50 #include "evmcs.h"
51 #include "irq.h"
52 #include "kvm_cache_regs.h"
53 #include "lapic.h"
54 #include "mmu.h"
55 #include "nested.h"
56 #include "ops.h"
57 #include "pmu.h"
58 #include "trace.h"
59 #include "vmcs.h"
60 #include "vmcs12.h"
61 #include "vmx.h"
62 #include "x86.h"
63 
64 MODULE_AUTHOR("Qumranet");
65 MODULE_LICENSE("GPL");
66 
67 static const struct x86_cpu_id vmx_cpu_id[] = {
68 	X86_FEATURE_MATCH(X86_FEATURE_VMX),
69 	{}
70 };
71 MODULE_DEVICE_TABLE(x86cpu, vmx_cpu_id);
72 
73 bool __read_mostly enable_vpid = 1;
74 module_param_named(vpid, enable_vpid, bool, 0444);
75 
76 static bool __read_mostly enable_vnmi = 1;
77 module_param_named(vnmi, enable_vnmi, bool, S_IRUGO);
78 
79 bool __read_mostly flexpriority_enabled = 1;
80 module_param_named(flexpriority, flexpriority_enabled, bool, S_IRUGO);
81 
82 bool __read_mostly enable_ept = 1;
83 module_param_named(ept, enable_ept, bool, S_IRUGO);
84 
85 bool __read_mostly enable_unrestricted_guest = 1;
86 module_param_named(unrestricted_guest,
87 			enable_unrestricted_guest, bool, S_IRUGO);
88 
89 bool __read_mostly enable_ept_ad_bits = 1;
90 module_param_named(eptad, enable_ept_ad_bits, bool, S_IRUGO);
91 
92 static bool __read_mostly emulate_invalid_guest_state = true;
93 module_param(emulate_invalid_guest_state, bool, S_IRUGO);
94 
95 static bool __read_mostly fasteoi = 1;
96 module_param(fasteoi, bool, S_IRUGO);
97 
98 static bool __read_mostly enable_apicv = 1;
99 module_param(enable_apicv, bool, S_IRUGO);
100 
101 /*
102  * If nested=1, nested virtualization is supported, i.e., guests may use
103  * VMX and be a hypervisor for its own guests. If nested=0, guests may not
104  * use VMX instructions.
105  */
106 static bool __read_mostly nested = 1;
107 module_param(nested, bool, S_IRUGO);
108 
109 bool __read_mostly enable_pml = 1;
110 module_param_named(pml, enable_pml, bool, S_IRUGO);
111 
112 static bool __read_mostly dump_invalid_vmcs = 0;
113 module_param(dump_invalid_vmcs, bool, 0644);
114 
115 #define MSR_BITMAP_MODE_X2APIC		1
116 #define MSR_BITMAP_MODE_X2APIC_APICV	2
117 
118 #define KVM_VMX_TSC_MULTIPLIER_MAX     0xffffffffffffffffULL
119 
120 /* Guest_tsc -> host_tsc conversion requires 64-bit division.  */
121 static int __read_mostly cpu_preemption_timer_multi;
122 static bool __read_mostly enable_preemption_timer = 1;
123 #ifdef CONFIG_X86_64
124 module_param_named(preemption_timer, enable_preemption_timer, bool, S_IRUGO);
125 #endif
126 
127 #define KVM_VM_CR0_ALWAYS_OFF (X86_CR0_NW | X86_CR0_CD)
128 #define KVM_VM_CR0_ALWAYS_ON_UNRESTRICTED_GUEST X86_CR0_NE
129 #define KVM_VM_CR0_ALWAYS_ON				\
130 	(KVM_VM_CR0_ALWAYS_ON_UNRESTRICTED_GUEST | 	\
131 	 X86_CR0_WP | X86_CR0_PG | X86_CR0_PE)
132 #define KVM_CR4_GUEST_OWNED_BITS				      \
133 	(X86_CR4_PVI | X86_CR4_DE | X86_CR4_PCE | X86_CR4_OSFXSR      \
134 	 | X86_CR4_OSXMMEXCPT | X86_CR4_LA57 | X86_CR4_TSD)
135 
136 #define KVM_VM_CR4_ALWAYS_ON_UNRESTRICTED_GUEST X86_CR4_VMXE
137 #define KVM_PMODE_VM_CR4_ALWAYS_ON (X86_CR4_PAE | X86_CR4_VMXE)
138 #define KVM_RMODE_VM_CR4_ALWAYS_ON (X86_CR4_VME | X86_CR4_PAE | X86_CR4_VMXE)
139 
140 #define RMODE_GUEST_OWNED_EFLAGS_BITS (~(X86_EFLAGS_IOPL | X86_EFLAGS_VM))
141 
142 #define MSR_IA32_RTIT_STATUS_MASK (~(RTIT_STATUS_FILTEREN | \
143 	RTIT_STATUS_CONTEXTEN | RTIT_STATUS_TRIGGEREN | \
144 	RTIT_STATUS_ERROR | RTIT_STATUS_STOPPED | \
145 	RTIT_STATUS_BYTECNT))
146 
147 #define MSR_IA32_RTIT_OUTPUT_BASE_MASK \
148 	(~((1UL << cpuid_query_maxphyaddr(vcpu)) - 1) | 0x7f)
149 
150 /*
151  * These 2 parameters are used to config the controls for Pause-Loop Exiting:
152  * ple_gap:    upper bound on the amount of time between two successive
153  *             executions of PAUSE in a loop. Also indicate if ple enabled.
154  *             According to test, this time is usually smaller than 128 cycles.
155  * ple_window: upper bound on the amount of time a guest is allowed to execute
156  *             in a PAUSE loop. Tests indicate that most spinlocks are held for
157  *             less than 2^12 cycles
158  * Time is measured based on a counter that runs at the same rate as the TSC,
159  * refer SDM volume 3b section 21.6.13 & 22.1.3.
160  */
161 static unsigned int ple_gap = KVM_DEFAULT_PLE_GAP;
162 module_param(ple_gap, uint, 0444);
163 
164 static unsigned int ple_window = KVM_VMX_DEFAULT_PLE_WINDOW;
165 module_param(ple_window, uint, 0444);
166 
167 /* Default doubles per-vcpu window every exit. */
168 static unsigned int ple_window_grow = KVM_DEFAULT_PLE_WINDOW_GROW;
169 module_param(ple_window_grow, uint, 0444);
170 
171 /* Default resets per-vcpu window every exit to ple_window. */
172 static unsigned int ple_window_shrink = KVM_DEFAULT_PLE_WINDOW_SHRINK;
173 module_param(ple_window_shrink, uint, 0444);
174 
175 /* Default is to compute the maximum so we can never overflow. */
176 static unsigned int ple_window_max        = KVM_VMX_DEFAULT_PLE_WINDOW_MAX;
177 module_param(ple_window_max, uint, 0444);
178 
179 /* Default is SYSTEM mode, 1 for host-guest mode */
180 int __read_mostly pt_mode = PT_MODE_SYSTEM;
181 module_param(pt_mode, int, S_IRUGO);
182 
183 static DEFINE_STATIC_KEY_FALSE(vmx_l1d_should_flush);
184 static DEFINE_STATIC_KEY_FALSE(vmx_l1d_flush_cond);
185 static DEFINE_MUTEX(vmx_l1d_flush_mutex);
186 
187 /* Storage for pre module init parameter parsing */
188 static enum vmx_l1d_flush_state __read_mostly vmentry_l1d_flush_param = VMENTER_L1D_FLUSH_AUTO;
189 
190 static const struct {
191 	const char *option;
192 	bool for_parse;
193 } vmentry_l1d_param[] = {
194 	[VMENTER_L1D_FLUSH_AUTO]	 = {"auto", true},
195 	[VMENTER_L1D_FLUSH_NEVER]	 = {"never", true},
196 	[VMENTER_L1D_FLUSH_COND]	 = {"cond", true},
197 	[VMENTER_L1D_FLUSH_ALWAYS]	 = {"always", true},
198 	[VMENTER_L1D_FLUSH_EPT_DISABLED] = {"EPT disabled", false},
199 	[VMENTER_L1D_FLUSH_NOT_REQUIRED] = {"not required", false},
200 };
201 
202 #define L1D_CACHE_ORDER 4
203 static void *vmx_l1d_flush_pages;
204 
205 static int vmx_setup_l1d_flush(enum vmx_l1d_flush_state l1tf)
206 {
207 	struct page *page;
208 	unsigned int i;
209 
210 	if (!boot_cpu_has_bug(X86_BUG_L1TF)) {
211 		l1tf_vmx_mitigation = VMENTER_L1D_FLUSH_NOT_REQUIRED;
212 		return 0;
213 	}
214 
215 	if (!enable_ept) {
216 		l1tf_vmx_mitigation = VMENTER_L1D_FLUSH_EPT_DISABLED;
217 		return 0;
218 	}
219 
220 	if (boot_cpu_has(X86_FEATURE_ARCH_CAPABILITIES)) {
221 		u64 msr;
222 
223 		rdmsrl(MSR_IA32_ARCH_CAPABILITIES, msr);
224 		if (msr & ARCH_CAP_SKIP_VMENTRY_L1DFLUSH) {
225 			l1tf_vmx_mitigation = VMENTER_L1D_FLUSH_NOT_REQUIRED;
226 			return 0;
227 		}
228 	}
229 
230 	/* If set to auto use the default l1tf mitigation method */
231 	if (l1tf == VMENTER_L1D_FLUSH_AUTO) {
232 		switch (l1tf_mitigation) {
233 		case L1TF_MITIGATION_OFF:
234 			l1tf = VMENTER_L1D_FLUSH_NEVER;
235 			break;
236 		case L1TF_MITIGATION_FLUSH_NOWARN:
237 		case L1TF_MITIGATION_FLUSH:
238 		case L1TF_MITIGATION_FLUSH_NOSMT:
239 			l1tf = VMENTER_L1D_FLUSH_COND;
240 			break;
241 		case L1TF_MITIGATION_FULL:
242 		case L1TF_MITIGATION_FULL_FORCE:
243 			l1tf = VMENTER_L1D_FLUSH_ALWAYS;
244 			break;
245 		}
246 	} else if (l1tf_mitigation == L1TF_MITIGATION_FULL_FORCE) {
247 		l1tf = VMENTER_L1D_FLUSH_ALWAYS;
248 	}
249 
250 	if (l1tf != VMENTER_L1D_FLUSH_NEVER && !vmx_l1d_flush_pages &&
251 	    !boot_cpu_has(X86_FEATURE_FLUSH_L1D)) {
252 		/*
253 		 * This allocation for vmx_l1d_flush_pages is not tied to a VM
254 		 * lifetime and so should not be charged to a memcg.
255 		 */
256 		page = alloc_pages(GFP_KERNEL, L1D_CACHE_ORDER);
257 		if (!page)
258 			return -ENOMEM;
259 		vmx_l1d_flush_pages = page_address(page);
260 
261 		/*
262 		 * Initialize each page with a different pattern in
263 		 * order to protect against KSM in the nested
264 		 * virtualization case.
265 		 */
266 		for (i = 0; i < 1u << L1D_CACHE_ORDER; ++i) {
267 			memset(vmx_l1d_flush_pages + i * PAGE_SIZE, i + 1,
268 			       PAGE_SIZE);
269 		}
270 	}
271 
272 	l1tf_vmx_mitigation = l1tf;
273 
274 	if (l1tf != VMENTER_L1D_FLUSH_NEVER)
275 		static_branch_enable(&vmx_l1d_should_flush);
276 	else
277 		static_branch_disable(&vmx_l1d_should_flush);
278 
279 	if (l1tf == VMENTER_L1D_FLUSH_COND)
280 		static_branch_enable(&vmx_l1d_flush_cond);
281 	else
282 		static_branch_disable(&vmx_l1d_flush_cond);
283 	return 0;
284 }
285 
286 static int vmentry_l1d_flush_parse(const char *s)
287 {
288 	unsigned int i;
289 
290 	if (s) {
291 		for (i = 0; i < ARRAY_SIZE(vmentry_l1d_param); i++) {
292 			if (vmentry_l1d_param[i].for_parse &&
293 			    sysfs_streq(s, vmentry_l1d_param[i].option))
294 				return i;
295 		}
296 	}
297 	return -EINVAL;
298 }
299 
300 static int vmentry_l1d_flush_set(const char *s, const struct kernel_param *kp)
301 {
302 	int l1tf, ret;
303 
304 	l1tf = vmentry_l1d_flush_parse(s);
305 	if (l1tf < 0)
306 		return l1tf;
307 
308 	if (!boot_cpu_has(X86_BUG_L1TF))
309 		return 0;
310 
311 	/*
312 	 * Has vmx_init() run already? If not then this is the pre init
313 	 * parameter parsing. In that case just store the value and let
314 	 * vmx_init() do the proper setup after enable_ept has been
315 	 * established.
316 	 */
317 	if (l1tf_vmx_mitigation == VMENTER_L1D_FLUSH_AUTO) {
318 		vmentry_l1d_flush_param = l1tf;
319 		return 0;
320 	}
321 
322 	mutex_lock(&vmx_l1d_flush_mutex);
323 	ret = vmx_setup_l1d_flush(l1tf);
324 	mutex_unlock(&vmx_l1d_flush_mutex);
325 	return ret;
326 }
327 
328 static int vmentry_l1d_flush_get(char *s, const struct kernel_param *kp)
329 {
330 	if (WARN_ON_ONCE(l1tf_vmx_mitigation >= ARRAY_SIZE(vmentry_l1d_param)))
331 		return sprintf(s, "???\n");
332 
333 	return sprintf(s, "%s\n", vmentry_l1d_param[l1tf_vmx_mitigation].option);
334 }
335 
336 static const struct kernel_param_ops vmentry_l1d_flush_ops = {
337 	.set = vmentry_l1d_flush_set,
338 	.get = vmentry_l1d_flush_get,
339 };
340 module_param_cb(vmentry_l1d_flush, &vmentry_l1d_flush_ops, NULL, 0644);
341 
342 static bool guest_state_valid(struct kvm_vcpu *vcpu);
343 static u32 vmx_segment_access_rights(struct kvm_segment *var);
344 static __always_inline void vmx_disable_intercept_for_msr(unsigned long *msr_bitmap,
345 							  u32 msr, int type);
346 
347 void vmx_vmexit(void);
348 
349 #define vmx_insn_failed(fmt...)		\
350 do {					\
351 	WARN_ONCE(1, fmt);		\
352 	pr_warn_ratelimited(fmt);	\
353 } while (0)
354 
355 asmlinkage void vmread_error(unsigned long field, bool fault)
356 {
357 	if (fault)
358 		kvm_spurious_fault();
359 	else
360 		vmx_insn_failed("kvm: vmread failed: field=%lx\n", field);
361 }
362 
363 noinline void vmwrite_error(unsigned long field, unsigned long value)
364 {
365 	vmx_insn_failed("kvm: vmwrite failed: field=%lx val=%lx err=%d\n",
366 			field, value, vmcs_read32(VM_INSTRUCTION_ERROR));
367 }
368 
369 noinline void vmclear_error(struct vmcs *vmcs, u64 phys_addr)
370 {
371 	vmx_insn_failed("kvm: vmclear failed: %p/%llx\n", vmcs, phys_addr);
372 }
373 
374 noinline void vmptrld_error(struct vmcs *vmcs, u64 phys_addr)
375 {
376 	vmx_insn_failed("kvm: vmptrld failed: %p/%llx\n", vmcs, phys_addr);
377 }
378 
379 noinline void invvpid_error(unsigned long ext, u16 vpid, gva_t gva)
380 {
381 	vmx_insn_failed("kvm: invvpid failed: ext=0x%lx vpid=%u gva=0x%lx\n",
382 			ext, vpid, gva);
383 }
384 
385 noinline void invept_error(unsigned long ext, u64 eptp, gpa_t gpa)
386 {
387 	vmx_insn_failed("kvm: invept failed: ext=0x%lx eptp=%llx gpa=0x%llx\n",
388 			ext, eptp, gpa);
389 }
390 
391 static DEFINE_PER_CPU(struct vmcs *, vmxarea);
392 DEFINE_PER_CPU(struct vmcs *, current_vmcs);
393 /*
394  * We maintain a per-CPU linked-list of VMCS loaded on that CPU. This is needed
395  * when a CPU is brought down, and we need to VMCLEAR all VMCSs loaded on it.
396  */
397 static DEFINE_PER_CPU(struct list_head, loaded_vmcss_on_cpu);
398 
399 /*
400  * We maintian a per-CPU linked-list of vCPU, so in wakeup_handler() we
401  * can find which vCPU should be waken up.
402  */
403 static DEFINE_PER_CPU(struct list_head, blocked_vcpu_on_cpu);
404 static DEFINE_PER_CPU(spinlock_t, blocked_vcpu_on_cpu_lock);
405 
406 static DECLARE_BITMAP(vmx_vpid_bitmap, VMX_NR_VPIDS);
407 static DEFINE_SPINLOCK(vmx_vpid_lock);
408 
409 struct vmcs_config vmcs_config;
410 struct vmx_capability vmx_capability;
411 
412 #define VMX_SEGMENT_FIELD(seg)					\
413 	[VCPU_SREG_##seg] = {                                   \
414 		.selector = GUEST_##seg##_SELECTOR,		\
415 		.base = GUEST_##seg##_BASE,		   	\
416 		.limit = GUEST_##seg##_LIMIT,		   	\
417 		.ar_bytes = GUEST_##seg##_AR_BYTES,	   	\
418 	}
419 
420 static const struct kvm_vmx_segment_field {
421 	unsigned selector;
422 	unsigned base;
423 	unsigned limit;
424 	unsigned ar_bytes;
425 } kvm_vmx_segment_fields[] = {
426 	VMX_SEGMENT_FIELD(CS),
427 	VMX_SEGMENT_FIELD(DS),
428 	VMX_SEGMENT_FIELD(ES),
429 	VMX_SEGMENT_FIELD(FS),
430 	VMX_SEGMENT_FIELD(GS),
431 	VMX_SEGMENT_FIELD(SS),
432 	VMX_SEGMENT_FIELD(TR),
433 	VMX_SEGMENT_FIELD(LDTR),
434 };
435 
436 u64 host_efer;
437 static unsigned long host_idt_base;
438 
439 /*
440  * Though SYSCALL is only supported in 64-bit mode on Intel CPUs, kvm
441  * will emulate SYSCALL in legacy mode if the vendor string in guest
442  * CPUID.0:{EBX,ECX,EDX} is "AuthenticAMD" or "AMDisbetter!" To
443  * support this emulation, IA32_STAR must always be included in
444  * vmx_msr_index[], even in i386 builds.
445  */
446 const u32 vmx_msr_index[] = {
447 #ifdef CONFIG_X86_64
448 	MSR_SYSCALL_MASK, MSR_LSTAR, MSR_CSTAR,
449 #endif
450 	MSR_EFER, MSR_TSC_AUX, MSR_STAR,
451 	MSR_IA32_TSX_CTRL,
452 };
453 
454 #if IS_ENABLED(CONFIG_HYPERV)
455 static bool __read_mostly enlightened_vmcs = true;
456 module_param(enlightened_vmcs, bool, 0444);
457 
458 /* check_ept_pointer() should be under protection of ept_pointer_lock. */
459 static void check_ept_pointer_match(struct kvm *kvm)
460 {
461 	struct kvm_vcpu *vcpu;
462 	u64 tmp_eptp = INVALID_PAGE;
463 	int i;
464 
465 	kvm_for_each_vcpu(i, vcpu, kvm) {
466 		if (!VALID_PAGE(tmp_eptp)) {
467 			tmp_eptp = to_vmx(vcpu)->ept_pointer;
468 		} else if (tmp_eptp != to_vmx(vcpu)->ept_pointer) {
469 			to_kvm_vmx(kvm)->ept_pointers_match
470 				= EPT_POINTERS_MISMATCH;
471 			return;
472 		}
473 	}
474 
475 	to_kvm_vmx(kvm)->ept_pointers_match = EPT_POINTERS_MATCH;
476 }
477 
478 static int kvm_fill_hv_flush_list_func(struct hv_guest_mapping_flush_list *flush,
479 		void *data)
480 {
481 	struct kvm_tlb_range *range = data;
482 
483 	return hyperv_fill_flush_guest_mapping_list(flush, range->start_gfn,
484 			range->pages);
485 }
486 
487 static inline int __hv_remote_flush_tlb_with_range(struct kvm *kvm,
488 		struct kvm_vcpu *vcpu, struct kvm_tlb_range *range)
489 {
490 	u64 ept_pointer = to_vmx(vcpu)->ept_pointer;
491 
492 	/*
493 	 * FLUSH_GUEST_PHYSICAL_ADDRESS_SPACE hypercall needs address
494 	 * of the base of EPT PML4 table, strip off EPT configuration
495 	 * information.
496 	 */
497 	if (range)
498 		return hyperv_flush_guest_mapping_range(ept_pointer & PAGE_MASK,
499 				kvm_fill_hv_flush_list_func, (void *)range);
500 	else
501 		return hyperv_flush_guest_mapping(ept_pointer & PAGE_MASK);
502 }
503 
504 static int hv_remote_flush_tlb_with_range(struct kvm *kvm,
505 		struct kvm_tlb_range *range)
506 {
507 	struct kvm_vcpu *vcpu;
508 	int ret = 0, i;
509 
510 	spin_lock(&to_kvm_vmx(kvm)->ept_pointer_lock);
511 
512 	if (to_kvm_vmx(kvm)->ept_pointers_match == EPT_POINTERS_CHECK)
513 		check_ept_pointer_match(kvm);
514 
515 	if (to_kvm_vmx(kvm)->ept_pointers_match != EPT_POINTERS_MATCH) {
516 		kvm_for_each_vcpu(i, vcpu, kvm) {
517 			/* If ept_pointer is invalid pointer, bypass flush request. */
518 			if (VALID_PAGE(to_vmx(vcpu)->ept_pointer))
519 				ret |= __hv_remote_flush_tlb_with_range(
520 					kvm, vcpu, range);
521 		}
522 	} else {
523 		ret = __hv_remote_flush_tlb_with_range(kvm,
524 				kvm_get_vcpu(kvm, 0), range);
525 	}
526 
527 	spin_unlock(&to_kvm_vmx(kvm)->ept_pointer_lock);
528 	return ret;
529 }
530 static int hv_remote_flush_tlb(struct kvm *kvm)
531 {
532 	return hv_remote_flush_tlb_with_range(kvm, NULL);
533 }
534 
535 static int hv_enable_direct_tlbflush(struct kvm_vcpu *vcpu)
536 {
537 	struct hv_enlightened_vmcs *evmcs;
538 	struct hv_partition_assist_pg **p_hv_pa_pg =
539 			&vcpu->kvm->arch.hyperv.hv_pa_pg;
540 	/*
541 	 * Synthetic VM-Exit is not enabled in current code and so All
542 	 * evmcs in singe VM shares same assist page.
543 	 */
544 	if (!*p_hv_pa_pg)
545 		*p_hv_pa_pg = kzalloc(PAGE_SIZE, GFP_KERNEL);
546 
547 	if (!*p_hv_pa_pg)
548 		return -ENOMEM;
549 
550 	evmcs = (struct hv_enlightened_vmcs *)to_vmx(vcpu)->loaded_vmcs->vmcs;
551 
552 	evmcs->partition_assist_page =
553 		__pa(*p_hv_pa_pg);
554 	evmcs->hv_vm_id = (unsigned long)vcpu->kvm;
555 	evmcs->hv_enlightenments_control.nested_flush_hypercall = 1;
556 
557 	return 0;
558 }
559 
560 #endif /* IS_ENABLED(CONFIG_HYPERV) */
561 
562 /*
563  * Comment's format: document - errata name - stepping - processor name.
564  * Refer from
565  * https://www.virtualbox.org/svn/vbox/trunk/src/VBox/VMM/VMMR0/HMR0.cpp
566  */
567 static u32 vmx_preemption_cpu_tfms[] = {
568 /* 323344.pdf - BA86   - D0 - Xeon 7500 Series */
569 0x000206E6,
570 /* 323056.pdf - AAX65  - C2 - Xeon L3406 */
571 /* 322814.pdf - AAT59  - C2 - i7-600, i5-500, i5-400 and i3-300 Mobile */
572 /* 322911.pdf - AAU65  - C2 - i5-600, i3-500 Desktop and Pentium G6950 */
573 0x00020652,
574 /* 322911.pdf - AAU65  - K0 - i5-600, i3-500 Desktop and Pentium G6950 */
575 0x00020655,
576 /* 322373.pdf - AAO95  - B1 - Xeon 3400 Series */
577 /* 322166.pdf - AAN92  - B1 - i7-800 and i5-700 Desktop */
578 /*
579  * 320767.pdf - AAP86  - B1 -
580  * i7-900 Mobile Extreme, i7-800 and i7-700 Mobile
581  */
582 0x000106E5,
583 /* 321333.pdf - AAM126 - C0 - Xeon 3500 */
584 0x000106A0,
585 /* 321333.pdf - AAM126 - C1 - Xeon 3500 */
586 0x000106A1,
587 /* 320836.pdf - AAJ124 - C0 - i7-900 Desktop Extreme and i7-900 Desktop */
588 0x000106A4,
589  /* 321333.pdf - AAM126 - D0 - Xeon 3500 */
590  /* 321324.pdf - AAK139 - D0 - Xeon 5500 */
591  /* 320836.pdf - AAJ124 - D0 - i7-900 Extreme and i7-900 Desktop */
592 0x000106A5,
593  /* Xeon E3-1220 V2 */
594 0x000306A8,
595 };
596 
597 static inline bool cpu_has_broken_vmx_preemption_timer(void)
598 {
599 	u32 eax = cpuid_eax(0x00000001), i;
600 
601 	/* Clear the reserved bits */
602 	eax &= ~(0x3U << 14 | 0xfU << 28);
603 	for (i = 0; i < ARRAY_SIZE(vmx_preemption_cpu_tfms); i++)
604 		if (eax == vmx_preemption_cpu_tfms[i])
605 			return true;
606 
607 	return false;
608 }
609 
610 static inline bool cpu_need_virtualize_apic_accesses(struct kvm_vcpu *vcpu)
611 {
612 	return flexpriority_enabled && lapic_in_kernel(vcpu);
613 }
614 
615 static inline bool report_flexpriority(void)
616 {
617 	return flexpriority_enabled;
618 }
619 
620 static inline int __find_msr_index(struct vcpu_vmx *vmx, u32 msr)
621 {
622 	int i;
623 
624 	for (i = 0; i < vmx->nmsrs; ++i)
625 		if (vmx_msr_index[vmx->guest_msrs[i].index] == msr)
626 			return i;
627 	return -1;
628 }
629 
630 struct shared_msr_entry *find_msr_entry(struct vcpu_vmx *vmx, u32 msr)
631 {
632 	int i;
633 
634 	i = __find_msr_index(vmx, msr);
635 	if (i >= 0)
636 		return &vmx->guest_msrs[i];
637 	return NULL;
638 }
639 
640 static int vmx_set_guest_msr(struct vcpu_vmx *vmx, struct shared_msr_entry *msr, u64 data)
641 {
642 	int ret = 0;
643 
644 	u64 old_msr_data = msr->data;
645 	msr->data = data;
646 	if (msr - vmx->guest_msrs < vmx->save_nmsrs) {
647 		preempt_disable();
648 		ret = kvm_set_shared_msr(msr->index, msr->data,
649 					 msr->mask);
650 		preempt_enable();
651 		if (ret)
652 			msr->data = old_msr_data;
653 	}
654 	return ret;
655 }
656 
657 void loaded_vmcs_init(struct loaded_vmcs *loaded_vmcs)
658 {
659 	vmcs_clear(loaded_vmcs->vmcs);
660 	if (loaded_vmcs->shadow_vmcs && loaded_vmcs->launched)
661 		vmcs_clear(loaded_vmcs->shadow_vmcs);
662 	loaded_vmcs->cpu = -1;
663 	loaded_vmcs->launched = 0;
664 }
665 
666 #ifdef CONFIG_KEXEC_CORE
667 /*
668  * This bitmap is used to indicate whether the vmclear
669  * operation is enabled on all cpus. All disabled by
670  * default.
671  */
672 static cpumask_t crash_vmclear_enabled_bitmap = CPU_MASK_NONE;
673 
674 static inline void crash_enable_local_vmclear(int cpu)
675 {
676 	cpumask_set_cpu(cpu, &crash_vmclear_enabled_bitmap);
677 }
678 
679 static inline void crash_disable_local_vmclear(int cpu)
680 {
681 	cpumask_clear_cpu(cpu, &crash_vmclear_enabled_bitmap);
682 }
683 
684 static inline int crash_local_vmclear_enabled(int cpu)
685 {
686 	return cpumask_test_cpu(cpu, &crash_vmclear_enabled_bitmap);
687 }
688 
689 static void crash_vmclear_local_loaded_vmcss(void)
690 {
691 	int cpu = raw_smp_processor_id();
692 	struct loaded_vmcs *v;
693 
694 	if (!crash_local_vmclear_enabled(cpu))
695 		return;
696 
697 	list_for_each_entry(v, &per_cpu(loaded_vmcss_on_cpu, cpu),
698 			    loaded_vmcss_on_cpu_link)
699 		vmcs_clear(v->vmcs);
700 }
701 #else
702 static inline void crash_enable_local_vmclear(int cpu) { }
703 static inline void crash_disable_local_vmclear(int cpu) { }
704 #endif /* CONFIG_KEXEC_CORE */
705 
706 static void __loaded_vmcs_clear(void *arg)
707 {
708 	struct loaded_vmcs *loaded_vmcs = arg;
709 	int cpu = raw_smp_processor_id();
710 
711 	if (loaded_vmcs->cpu != cpu)
712 		return; /* vcpu migration can race with cpu offline */
713 	if (per_cpu(current_vmcs, cpu) == loaded_vmcs->vmcs)
714 		per_cpu(current_vmcs, cpu) = NULL;
715 	crash_disable_local_vmclear(cpu);
716 	list_del(&loaded_vmcs->loaded_vmcss_on_cpu_link);
717 
718 	/*
719 	 * we should ensure updating loaded_vmcs->loaded_vmcss_on_cpu_link
720 	 * is before setting loaded_vmcs->vcpu to -1 which is done in
721 	 * loaded_vmcs_init. Otherwise, other cpu can see vcpu = -1 fist
722 	 * then adds the vmcs into percpu list before it is deleted.
723 	 */
724 	smp_wmb();
725 
726 	loaded_vmcs_init(loaded_vmcs);
727 	crash_enable_local_vmclear(cpu);
728 }
729 
730 void loaded_vmcs_clear(struct loaded_vmcs *loaded_vmcs)
731 {
732 	int cpu = loaded_vmcs->cpu;
733 
734 	if (cpu != -1)
735 		smp_call_function_single(cpu,
736 			 __loaded_vmcs_clear, loaded_vmcs, 1);
737 }
738 
739 static bool vmx_segment_cache_test_set(struct vcpu_vmx *vmx, unsigned seg,
740 				       unsigned field)
741 {
742 	bool ret;
743 	u32 mask = 1 << (seg * SEG_FIELD_NR + field);
744 
745 	if (!kvm_register_is_available(&vmx->vcpu, VCPU_EXREG_SEGMENTS)) {
746 		kvm_register_mark_available(&vmx->vcpu, VCPU_EXREG_SEGMENTS);
747 		vmx->segment_cache.bitmask = 0;
748 	}
749 	ret = vmx->segment_cache.bitmask & mask;
750 	vmx->segment_cache.bitmask |= mask;
751 	return ret;
752 }
753 
754 static u16 vmx_read_guest_seg_selector(struct vcpu_vmx *vmx, unsigned seg)
755 {
756 	u16 *p = &vmx->segment_cache.seg[seg].selector;
757 
758 	if (!vmx_segment_cache_test_set(vmx, seg, SEG_FIELD_SEL))
759 		*p = vmcs_read16(kvm_vmx_segment_fields[seg].selector);
760 	return *p;
761 }
762 
763 static ulong vmx_read_guest_seg_base(struct vcpu_vmx *vmx, unsigned seg)
764 {
765 	ulong *p = &vmx->segment_cache.seg[seg].base;
766 
767 	if (!vmx_segment_cache_test_set(vmx, seg, SEG_FIELD_BASE))
768 		*p = vmcs_readl(kvm_vmx_segment_fields[seg].base);
769 	return *p;
770 }
771 
772 static u32 vmx_read_guest_seg_limit(struct vcpu_vmx *vmx, unsigned seg)
773 {
774 	u32 *p = &vmx->segment_cache.seg[seg].limit;
775 
776 	if (!vmx_segment_cache_test_set(vmx, seg, SEG_FIELD_LIMIT))
777 		*p = vmcs_read32(kvm_vmx_segment_fields[seg].limit);
778 	return *p;
779 }
780 
781 static u32 vmx_read_guest_seg_ar(struct vcpu_vmx *vmx, unsigned seg)
782 {
783 	u32 *p = &vmx->segment_cache.seg[seg].ar;
784 
785 	if (!vmx_segment_cache_test_set(vmx, seg, SEG_FIELD_AR))
786 		*p = vmcs_read32(kvm_vmx_segment_fields[seg].ar_bytes);
787 	return *p;
788 }
789 
790 void update_exception_bitmap(struct kvm_vcpu *vcpu)
791 {
792 	u32 eb;
793 
794 	eb = (1u << PF_VECTOR) | (1u << UD_VECTOR) | (1u << MC_VECTOR) |
795 	     (1u << DB_VECTOR) | (1u << AC_VECTOR);
796 	/*
797 	 * Guest access to VMware backdoor ports could legitimately
798 	 * trigger #GP because of TSS I/O permission bitmap.
799 	 * We intercept those #GP and allow access to them anyway
800 	 * as VMware does.
801 	 */
802 	if (enable_vmware_backdoor)
803 		eb |= (1u << GP_VECTOR);
804 	if ((vcpu->guest_debug &
805 	     (KVM_GUESTDBG_ENABLE | KVM_GUESTDBG_USE_SW_BP)) ==
806 	    (KVM_GUESTDBG_ENABLE | KVM_GUESTDBG_USE_SW_BP))
807 		eb |= 1u << BP_VECTOR;
808 	if (to_vmx(vcpu)->rmode.vm86_active)
809 		eb = ~0;
810 	if (enable_ept)
811 		eb &= ~(1u << PF_VECTOR); /* bypass_guest_pf = 0 */
812 
813 	/* When we are running a nested L2 guest and L1 specified for it a
814 	 * certain exception bitmap, we must trap the same exceptions and pass
815 	 * them to L1. When running L2, we will only handle the exceptions
816 	 * specified above if L1 did not want them.
817 	 */
818 	if (is_guest_mode(vcpu))
819 		eb |= get_vmcs12(vcpu)->exception_bitmap;
820 
821 	vmcs_write32(EXCEPTION_BITMAP, eb);
822 }
823 
824 /*
825  * Check if MSR is intercepted for currently loaded MSR bitmap.
826  */
827 static bool msr_write_intercepted(struct kvm_vcpu *vcpu, u32 msr)
828 {
829 	unsigned long *msr_bitmap;
830 	int f = sizeof(unsigned long);
831 
832 	if (!cpu_has_vmx_msr_bitmap())
833 		return true;
834 
835 	msr_bitmap = to_vmx(vcpu)->loaded_vmcs->msr_bitmap;
836 
837 	if (msr <= 0x1fff) {
838 		return !!test_bit(msr, msr_bitmap + 0x800 / f);
839 	} else if ((msr >= 0xc0000000) && (msr <= 0xc0001fff)) {
840 		msr &= 0x1fff;
841 		return !!test_bit(msr, msr_bitmap + 0xc00 / f);
842 	}
843 
844 	return true;
845 }
846 
847 static void clear_atomic_switch_msr_special(struct vcpu_vmx *vmx,
848 		unsigned long entry, unsigned long exit)
849 {
850 	vm_entry_controls_clearbit(vmx, entry);
851 	vm_exit_controls_clearbit(vmx, exit);
852 }
853 
854 int vmx_find_msr_index(struct vmx_msrs *m, u32 msr)
855 {
856 	unsigned int i;
857 
858 	for (i = 0; i < m->nr; ++i) {
859 		if (m->val[i].index == msr)
860 			return i;
861 	}
862 	return -ENOENT;
863 }
864 
865 static void clear_atomic_switch_msr(struct vcpu_vmx *vmx, unsigned msr)
866 {
867 	int i;
868 	struct msr_autoload *m = &vmx->msr_autoload;
869 
870 	switch (msr) {
871 	case MSR_EFER:
872 		if (cpu_has_load_ia32_efer()) {
873 			clear_atomic_switch_msr_special(vmx,
874 					VM_ENTRY_LOAD_IA32_EFER,
875 					VM_EXIT_LOAD_IA32_EFER);
876 			return;
877 		}
878 		break;
879 	case MSR_CORE_PERF_GLOBAL_CTRL:
880 		if (cpu_has_load_perf_global_ctrl()) {
881 			clear_atomic_switch_msr_special(vmx,
882 					VM_ENTRY_LOAD_IA32_PERF_GLOBAL_CTRL,
883 					VM_EXIT_LOAD_IA32_PERF_GLOBAL_CTRL);
884 			return;
885 		}
886 		break;
887 	}
888 	i = vmx_find_msr_index(&m->guest, msr);
889 	if (i < 0)
890 		goto skip_guest;
891 	--m->guest.nr;
892 	m->guest.val[i] = m->guest.val[m->guest.nr];
893 	vmcs_write32(VM_ENTRY_MSR_LOAD_COUNT, m->guest.nr);
894 
895 skip_guest:
896 	i = vmx_find_msr_index(&m->host, msr);
897 	if (i < 0)
898 		return;
899 
900 	--m->host.nr;
901 	m->host.val[i] = m->host.val[m->host.nr];
902 	vmcs_write32(VM_EXIT_MSR_LOAD_COUNT, m->host.nr);
903 }
904 
905 static void add_atomic_switch_msr_special(struct vcpu_vmx *vmx,
906 		unsigned long entry, unsigned long exit,
907 		unsigned long guest_val_vmcs, unsigned long host_val_vmcs,
908 		u64 guest_val, u64 host_val)
909 {
910 	vmcs_write64(guest_val_vmcs, guest_val);
911 	if (host_val_vmcs != HOST_IA32_EFER)
912 		vmcs_write64(host_val_vmcs, host_val);
913 	vm_entry_controls_setbit(vmx, entry);
914 	vm_exit_controls_setbit(vmx, exit);
915 }
916 
917 static void add_atomic_switch_msr(struct vcpu_vmx *vmx, unsigned msr,
918 				  u64 guest_val, u64 host_val, bool entry_only)
919 {
920 	int i, j = 0;
921 	struct msr_autoload *m = &vmx->msr_autoload;
922 
923 	switch (msr) {
924 	case MSR_EFER:
925 		if (cpu_has_load_ia32_efer()) {
926 			add_atomic_switch_msr_special(vmx,
927 					VM_ENTRY_LOAD_IA32_EFER,
928 					VM_EXIT_LOAD_IA32_EFER,
929 					GUEST_IA32_EFER,
930 					HOST_IA32_EFER,
931 					guest_val, host_val);
932 			return;
933 		}
934 		break;
935 	case MSR_CORE_PERF_GLOBAL_CTRL:
936 		if (cpu_has_load_perf_global_ctrl()) {
937 			add_atomic_switch_msr_special(vmx,
938 					VM_ENTRY_LOAD_IA32_PERF_GLOBAL_CTRL,
939 					VM_EXIT_LOAD_IA32_PERF_GLOBAL_CTRL,
940 					GUEST_IA32_PERF_GLOBAL_CTRL,
941 					HOST_IA32_PERF_GLOBAL_CTRL,
942 					guest_val, host_val);
943 			return;
944 		}
945 		break;
946 	case MSR_IA32_PEBS_ENABLE:
947 		/* PEBS needs a quiescent period after being disabled (to write
948 		 * a record).  Disabling PEBS through VMX MSR swapping doesn't
949 		 * provide that period, so a CPU could write host's record into
950 		 * guest's memory.
951 		 */
952 		wrmsrl(MSR_IA32_PEBS_ENABLE, 0);
953 	}
954 
955 	i = vmx_find_msr_index(&m->guest, msr);
956 	if (!entry_only)
957 		j = vmx_find_msr_index(&m->host, msr);
958 
959 	if ((i < 0 && m->guest.nr == NR_LOADSTORE_MSRS) ||
960 		(j < 0 &&  m->host.nr == NR_LOADSTORE_MSRS)) {
961 		printk_once(KERN_WARNING "Not enough msr switch entries. "
962 				"Can't add msr %x\n", msr);
963 		return;
964 	}
965 	if (i < 0) {
966 		i = m->guest.nr++;
967 		vmcs_write32(VM_ENTRY_MSR_LOAD_COUNT, m->guest.nr);
968 	}
969 	m->guest.val[i].index = msr;
970 	m->guest.val[i].value = guest_val;
971 
972 	if (entry_only)
973 		return;
974 
975 	if (j < 0) {
976 		j = m->host.nr++;
977 		vmcs_write32(VM_EXIT_MSR_LOAD_COUNT, m->host.nr);
978 	}
979 	m->host.val[j].index = msr;
980 	m->host.val[j].value = host_val;
981 }
982 
983 static bool update_transition_efer(struct vcpu_vmx *vmx, int efer_offset)
984 {
985 	u64 guest_efer = vmx->vcpu.arch.efer;
986 	u64 ignore_bits = 0;
987 
988 	/* Shadow paging assumes NX to be available.  */
989 	if (!enable_ept)
990 		guest_efer |= EFER_NX;
991 
992 	/*
993 	 * LMA and LME handled by hardware; SCE meaningless outside long mode.
994 	 */
995 	ignore_bits |= EFER_SCE;
996 #ifdef CONFIG_X86_64
997 	ignore_bits |= EFER_LMA | EFER_LME;
998 	/* SCE is meaningful only in long mode on Intel */
999 	if (guest_efer & EFER_LMA)
1000 		ignore_bits &= ~(u64)EFER_SCE;
1001 #endif
1002 
1003 	/*
1004 	 * On EPT, we can't emulate NX, so we must switch EFER atomically.
1005 	 * On CPUs that support "load IA32_EFER", always switch EFER
1006 	 * atomically, since it's faster than switching it manually.
1007 	 */
1008 	if (cpu_has_load_ia32_efer() ||
1009 	    (enable_ept && ((vmx->vcpu.arch.efer ^ host_efer) & EFER_NX))) {
1010 		if (!(guest_efer & EFER_LMA))
1011 			guest_efer &= ~EFER_LME;
1012 		if (guest_efer != host_efer)
1013 			add_atomic_switch_msr(vmx, MSR_EFER,
1014 					      guest_efer, host_efer, false);
1015 		else
1016 			clear_atomic_switch_msr(vmx, MSR_EFER);
1017 		return false;
1018 	} else {
1019 		clear_atomic_switch_msr(vmx, MSR_EFER);
1020 
1021 		guest_efer &= ~ignore_bits;
1022 		guest_efer |= host_efer & ignore_bits;
1023 
1024 		vmx->guest_msrs[efer_offset].data = guest_efer;
1025 		vmx->guest_msrs[efer_offset].mask = ~ignore_bits;
1026 
1027 		return true;
1028 	}
1029 }
1030 
1031 #ifdef CONFIG_X86_32
1032 /*
1033  * On 32-bit kernels, VM exits still load the FS and GS bases from the
1034  * VMCS rather than the segment table.  KVM uses this helper to figure
1035  * out the current bases to poke them into the VMCS before entry.
1036  */
1037 static unsigned long segment_base(u16 selector)
1038 {
1039 	struct desc_struct *table;
1040 	unsigned long v;
1041 
1042 	if (!(selector & ~SEGMENT_RPL_MASK))
1043 		return 0;
1044 
1045 	table = get_current_gdt_ro();
1046 
1047 	if ((selector & SEGMENT_TI_MASK) == SEGMENT_LDT) {
1048 		u16 ldt_selector = kvm_read_ldt();
1049 
1050 		if (!(ldt_selector & ~SEGMENT_RPL_MASK))
1051 			return 0;
1052 
1053 		table = (struct desc_struct *)segment_base(ldt_selector);
1054 	}
1055 	v = get_desc_base(&table[selector >> 3]);
1056 	return v;
1057 }
1058 #endif
1059 
1060 static inline void pt_load_msr(struct pt_ctx *ctx, u32 addr_range)
1061 {
1062 	u32 i;
1063 
1064 	wrmsrl(MSR_IA32_RTIT_STATUS, ctx->status);
1065 	wrmsrl(MSR_IA32_RTIT_OUTPUT_BASE, ctx->output_base);
1066 	wrmsrl(MSR_IA32_RTIT_OUTPUT_MASK, ctx->output_mask);
1067 	wrmsrl(MSR_IA32_RTIT_CR3_MATCH, ctx->cr3_match);
1068 	for (i = 0; i < addr_range; i++) {
1069 		wrmsrl(MSR_IA32_RTIT_ADDR0_A + i * 2, ctx->addr_a[i]);
1070 		wrmsrl(MSR_IA32_RTIT_ADDR0_B + i * 2, ctx->addr_b[i]);
1071 	}
1072 }
1073 
1074 static inline void pt_save_msr(struct pt_ctx *ctx, u32 addr_range)
1075 {
1076 	u32 i;
1077 
1078 	rdmsrl(MSR_IA32_RTIT_STATUS, ctx->status);
1079 	rdmsrl(MSR_IA32_RTIT_OUTPUT_BASE, ctx->output_base);
1080 	rdmsrl(MSR_IA32_RTIT_OUTPUT_MASK, ctx->output_mask);
1081 	rdmsrl(MSR_IA32_RTIT_CR3_MATCH, ctx->cr3_match);
1082 	for (i = 0; i < addr_range; i++) {
1083 		rdmsrl(MSR_IA32_RTIT_ADDR0_A + i * 2, ctx->addr_a[i]);
1084 		rdmsrl(MSR_IA32_RTIT_ADDR0_B + i * 2, ctx->addr_b[i]);
1085 	}
1086 }
1087 
1088 static void pt_guest_enter(struct vcpu_vmx *vmx)
1089 {
1090 	if (pt_mode == PT_MODE_SYSTEM)
1091 		return;
1092 
1093 	/*
1094 	 * GUEST_IA32_RTIT_CTL is already set in the VMCS.
1095 	 * Save host state before VM entry.
1096 	 */
1097 	rdmsrl(MSR_IA32_RTIT_CTL, vmx->pt_desc.host.ctl);
1098 	if (vmx->pt_desc.guest.ctl & RTIT_CTL_TRACEEN) {
1099 		wrmsrl(MSR_IA32_RTIT_CTL, 0);
1100 		pt_save_msr(&vmx->pt_desc.host, vmx->pt_desc.addr_range);
1101 		pt_load_msr(&vmx->pt_desc.guest, vmx->pt_desc.addr_range);
1102 	}
1103 }
1104 
1105 static void pt_guest_exit(struct vcpu_vmx *vmx)
1106 {
1107 	if (pt_mode == PT_MODE_SYSTEM)
1108 		return;
1109 
1110 	if (vmx->pt_desc.guest.ctl & RTIT_CTL_TRACEEN) {
1111 		pt_save_msr(&vmx->pt_desc.guest, vmx->pt_desc.addr_range);
1112 		pt_load_msr(&vmx->pt_desc.host, vmx->pt_desc.addr_range);
1113 	}
1114 
1115 	/* Reload host state (IA32_RTIT_CTL will be cleared on VM exit). */
1116 	wrmsrl(MSR_IA32_RTIT_CTL, vmx->pt_desc.host.ctl);
1117 }
1118 
1119 void vmx_set_host_fs_gs(struct vmcs_host_state *host, u16 fs_sel, u16 gs_sel,
1120 			unsigned long fs_base, unsigned long gs_base)
1121 {
1122 	if (unlikely(fs_sel != host->fs_sel)) {
1123 		if (!(fs_sel & 7))
1124 			vmcs_write16(HOST_FS_SELECTOR, fs_sel);
1125 		else
1126 			vmcs_write16(HOST_FS_SELECTOR, 0);
1127 		host->fs_sel = fs_sel;
1128 	}
1129 	if (unlikely(gs_sel != host->gs_sel)) {
1130 		if (!(gs_sel & 7))
1131 			vmcs_write16(HOST_GS_SELECTOR, gs_sel);
1132 		else
1133 			vmcs_write16(HOST_GS_SELECTOR, 0);
1134 		host->gs_sel = gs_sel;
1135 	}
1136 	if (unlikely(fs_base != host->fs_base)) {
1137 		vmcs_writel(HOST_FS_BASE, fs_base);
1138 		host->fs_base = fs_base;
1139 	}
1140 	if (unlikely(gs_base != host->gs_base)) {
1141 		vmcs_writel(HOST_GS_BASE, gs_base);
1142 		host->gs_base = gs_base;
1143 	}
1144 }
1145 
1146 void vmx_prepare_switch_to_guest(struct kvm_vcpu *vcpu)
1147 {
1148 	struct vcpu_vmx *vmx = to_vmx(vcpu);
1149 	struct vmcs_host_state *host_state;
1150 #ifdef CONFIG_X86_64
1151 	int cpu = raw_smp_processor_id();
1152 #endif
1153 	unsigned long fs_base, gs_base;
1154 	u16 fs_sel, gs_sel;
1155 	int i;
1156 
1157 	vmx->req_immediate_exit = false;
1158 
1159 	/*
1160 	 * Note that guest MSRs to be saved/restored can also be changed
1161 	 * when guest state is loaded. This happens when guest transitions
1162 	 * to/from long-mode by setting MSR_EFER.LMA.
1163 	 */
1164 	if (!vmx->guest_msrs_ready) {
1165 		vmx->guest_msrs_ready = true;
1166 		for (i = 0; i < vmx->save_nmsrs; ++i)
1167 			kvm_set_shared_msr(vmx->guest_msrs[i].index,
1168 					   vmx->guest_msrs[i].data,
1169 					   vmx->guest_msrs[i].mask);
1170 
1171 	}
1172 	if (vmx->guest_state_loaded)
1173 		return;
1174 
1175 	host_state = &vmx->loaded_vmcs->host_state;
1176 
1177 	/*
1178 	 * Set host fs and gs selectors.  Unfortunately, 22.2.3 does not
1179 	 * allow segment selectors with cpl > 0 or ti == 1.
1180 	 */
1181 	host_state->ldt_sel = kvm_read_ldt();
1182 
1183 #ifdef CONFIG_X86_64
1184 	savesegment(ds, host_state->ds_sel);
1185 	savesegment(es, host_state->es_sel);
1186 
1187 	gs_base = cpu_kernelmode_gs_base(cpu);
1188 	if (likely(is_64bit_mm(current->mm))) {
1189 		save_fsgs_for_kvm();
1190 		fs_sel = current->thread.fsindex;
1191 		gs_sel = current->thread.gsindex;
1192 		fs_base = current->thread.fsbase;
1193 		vmx->msr_host_kernel_gs_base = current->thread.gsbase;
1194 	} else {
1195 		savesegment(fs, fs_sel);
1196 		savesegment(gs, gs_sel);
1197 		fs_base = read_msr(MSR_FS_BASE);
1198 		vmx->msr_host_kernel_gs_base = read_msr(MSR_KERNEL_GS_BASE);
1199 	}
1200 
1201 	wrmsrl(MSR_KERNEL_GS_BASE, vmx->msr_guest_kernel_gs_base);
1202 #else
1203 	savesegment(fs, fs_sel);
1204 	savesegment(gs, gs_sel);
1205 	fs_base = segment_base(fs_sel);
1206 	gs_base = segment_base(gs_sel);
1207 #endif
1208 
1209 	vmx_set_host_fs_gs(host_state, fs_sel, gs_sel, fs_base, gs_base);
1210 	vmx->guest_state_loaded = true;
1211 }
1212 
1213 static void vmx_prepare_switch_to_host(struct vcpu_vmx *vmx)
1214 {
1215 	struct vmcs_host_state *host_state;
1216 
1217 	if (!vmx->guest_state_loaded)
1218 		return;
1219 
1220 	host_state = &vmx->loaded_vmcs->host_state;
1221 
1222 	++vmx->vcpu.stat.host_state_reload;
1223 
1224 #ifdef CONFIG_X86_64
1225 	rdmsrl(MSR_KERNEL_GS_BASE, vmx->msr_guest_kernel_gs_base);
1226 #endif
1227 	if (host_state->ldt_sel || (host_state->gs_sel & 7)) {
1228 		kvm_load_ldt(host_state->ldt_sel);
1229 #ifdef CONFIG_X86_64
1230 		load_gs_index(host_state->gs_sel);
1231 #else
1232 		loadsegment(gs, host_state->gs_sel);
1233 #endif
1234 	}
1235 	if (host_state->fs_sel & 7)
1236 		loadsegment(fs, host_state->fs_sel);
1237 #ifdef CONFIG_X86_64
1238 	if (unlikely(host_state->ds_sel | host_state->es_sel)) {
1239 		loadsegment(ds, host_state->ds_sel);
1240 		loadsegment(es, host_state->es_sel);
1241 	}
1242 #endif
1243 	invalidate_tss_limit();
1244 #ifdef CONFIG_X86_64
1245 	wrmsrl(MSR_KERNEL_GS_BASE, vmx->msr_host_kernel_gs_base);
1246 #endif
1247 	load_fixmap_gdt(raw_smp_processor_id());
1248 	vmx->guest_state_loaded = false;
1249 	vmx->guest_msrs_ready = false;
1250 }
1251 
1252 #ifdef CONFIG_X86_64
1253 static u64 vmx_read_guest_kernel_gs_base(struct vcpu_vmx *vmx)
1254 {
1255 	preempt_disable();
1256 	if (vmx->guest_state_loaded)
1257 		rdmsrl(MSR_KERNEL_GS_BASE, vmx->msr_guest_kernel_gs_base);
1258 	preempt_enable();
1259 	return vmx->msr_guest_kernel_gs_base;
1260 }
1261 
1262 static void vmx_write_guest_kernel_gs_base(struct vcpu_vmx *vmx, u64 data)
1263 {
1264 	preempt_disable();
1265 	if (vmx->guest_state_loaded)
1266 		wrmsrl(MSR_KERNEL_GS_BASE, data);
1267 	preempt_enable();
1268 	vmx->msr_guest_kernel_gs_base = data;
1269 }
1270 #endif
1271 
1272 static void vmx_vcpu_pi_load(struct kvm_vcpu *vcpu, int cpu)
1273 {
1274 	struct pi_desc *pi_desc = vcpu_to_pi_desc(vcpu);
1275 	struct pi_desc old, new;
1276 	unsigned int dest;
1277 
1278 	/*
1279 	 * In case of hot-plug or hot-unplug, we may have to undo
1280 	 * vmx_vcpu_pi_put even if there is no assigned device.  And we
1281 	 * always keep PI.NDST up to date for simplicity: it makes the
1282 	 * code easier, and CPU migration is not a fast path.
1283 	 */
1284 	if (!pi_test_sn(pi_desc) && vcpu->cpu == cpu)
1285 		return;
1286 
1287 	/*
1288 	 * If the 'nv' field is POSTED_INTR_WAKEUP_VECTOR, do not change
1289 	 * PI.NDST: pi_post_block is the one expected to change PID.NDST and the
1290 	 * wakeup handler expects the vCPU to be on the blocked_vcpu_list that
1291 	 * matches PI.NDST. Otherwise, a vcpu may not be able to be woken up
1292 	 * correctly.
1293 	 */
1294 	if (pi_desc->nv == POSTED_INTR_WAKEUP_VECTOR || vcpu->cpu == cpu) {
1295 		pi_clear_sn(pi_desc);
1296 		goto after_clear_sn;
1297 	}
1298 
1299 	/* The full case.  */
1300 	do {
1301 		old.control = new.control = pi_desc->control;
1302 
1303 		dest = cpu_physical_id(cpu);
1304 
1305 		if (x2apic_enabled())
1306 			new.ndst = dest;
1307 		else
1308 			new.ndst = (dest << 8) & 0xFF00;
1309 
1310 		new.sn = 0;
1311 	} while (cmpxchg64(&pi_desc->control, old.control,
1312 			   new.control) != old.control);
1313 
1314 after_clear_sn:
1315 
1316 	/*
1317 	 * Clear SN before reading the bitmap.  The VT-d firmware
1318 	 * writes the bitmap and reads SN atomically (5.2.3 in the
1319 	 * spec), so it doesn't really have a memory barrier that
1320 	 * pairs with this, but we cannot do that and we need one.
1321 	 */
1322 	smp_mb__after_atomic();
1323 
1324 	if (!pi_is_pir_empty(pi_desc))
1325 		pi_set_on(pi_desc);
1326 }
1327 
1328 void vmx_vcpu_load_vmcs(struct kvm_vcpu *vcpu, int cpu)
1329 {
1330 	struct vcpu_vmx *vmx = to_vmx(vcpu);
1331 	bool already_loaded = vmx->loaded_vmcs->cpu == cpu;
1332 
1333 	if (!already_loaded) {
1334 		loaded_vmcs_clear(vmx->loaded_vmcs);
1335 		local_irq_disable();
1336 		crash_disable_local_vmclear(cpu);
1337 
1338 		/*
1339 		 * Read loaded_vmcs->cpu should be before fetching
1340 		 * loaded_vmcs->loaded_vmcss_on_cpu_link.
1341 		 * See the comments in __loaded_vmcs_clear().
1342 		 */
1343 		smp_rmb();
1344 
1345 		list_add(&vmx->loaded_vmcs->loaded_vmcss_on_cpu_link,
1346 			 &per_cpu(loaded_vmcss_on_cpu, cpu));
1347 		crash_enable_local_vmclear(cpu);
1348 		local_irq_enable();
1349 	}
1350 
1351 	if (per_cpu(current_vmcs, cpu) != vmx->loaded_vmcs->vmcs) {
1352 		per_cpu(current_vmcs, cpu) = vmx->loaded_vmcs->vmcs;
1353 		vmcs_load(vmx->loaded_vmcs->vmcs);
1354 		indirect_branch_prediction_barrier();
1355 	}
1356 
1357 	if (!already_loaded) {
1358 		void *gdt = get_current_gdt_ro();
1359 		unsigned long sysenter_esp;
1360 
1361 		kvm_make_request(KVM_REQ_TLB_FLUSH, vcpu);
1362 
1363 		/*
1364 		 * Linux uses per-cpu TSS and GDT, so set these when switching
1365 		 * processors.  See 22.2.4.
1366 		 */
1367 		vmcs_writel(HOST_TR_BASE,
1368 			    (unsigned long)&get_cpu_entry_area(cpu)->tss.x86_tss);
1369 		vmcs_writel(HOST_GDTR_BASE, (unsigned long)gdt);   /* 22.2.4 */
1370 
1371 		rdmsrl(MSR_IA32_SYSENTER_ESP, sysenter_esp);
1372 		vmcs_writel(HOST_IA32_SYSENTER_ESP, sysenter_esp); /* 22.2.3 */
1373 
1374 		vmx->loaded_vmcs->cpu = cpu;
1375 	}
1376 
1377 	/* Setup TSC multiplier */
1378 	if (kvm_has_tsc_control &&
1379 	    vmx->current_tsc_ratio != vcpu->arch.tsc_scaling_ratio)
1380 		decache_tsc_multiplier(vmx);
1381 }
1382 
1383 /*
1384  * Switches to specified vcpu, until a matching vcpu_put(), but assumes
1385  * vcpu mutex is already taken.
1386  */
1387 void vmx_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
1388 {
1389 	struct vcpu_vmx *vmx = to_vmx(vcpu);
1390 
1391 	vmx_vcpu_load_vmcs(vcpu, cpu);
1392 
1393 	vmx_vcpu_pi_load(vcpu, cpu);
1394 
1395 	vmx->host_pkru = read_pkru();
1396 	vmx->host_debugctlmsr = get_debugctlmsr();
1397 }
1398 
1399 static void vmx_vcpu_pi_put(struct kvm_vcpu *vcpu)
1400 {
1401 	struct pi_desc *pi_desc = vcpu_to_pi_desc(vcpu);
1402 
1403 	if (!kvm_arch_has_assigned_device(vcpu->kvm) ||
1404 		!irq_remapping_cap(IRQ_POSTING_CAP)  ||
1405 		!kvm_vcpu_apicv_active(vcpu))
1406 		return;
1407 
1408 	/* Set SN when the vCPU is preempted */
1409 	if (vcpu->preempted)
1410 		pi_set_sn(pi_desc);
1411 }
1412 
1413 static void vmx_vcpu_put(struct kvm_vcpu *vcpu)
1414 {
1415 	vmx_vcpu_pi_put(vcpu);
1416 
1417 	vmx_prepare_switch_to_host(to_vmx(vcpu));
1418 }
1419 
1420 static bool emulation_required(struct kvm_vcpu *vcpu)
1421 {
1422 	return emulate_invalid_guest_state && !guest_state_valid(vcpu);
1423 }
1424 
1425 static void vmx_decache_cr0_guest_bits(struct kvm_vcpu *vcpu);
1426 
1427 unsigned long vmx_get_rflags(struct kvm_vcpu *vcpu)
1428 {
1429 	struct vcpu_vmx *vmx = to_vmx(vcpu);
1430 	unsigned long rflags, save_rflags;
1431 
1432 	if (!kvm_register_is_available(vcpu, VCPU_EXREG_RFLAGS)) {
1433 		kvm_register_mark_available(vcpu, VCPU_EXREG_RFLAGS);
1434 		rflags = vmcs_readl(GUEST_RFLAGS);
1435 		if (vmx->rmode.vm86_active) {
1436 			rflags &= RMODE_GUEST_OWNED_EFLAGS_BITS;
1437 			save_rflags = vmx->rmode.save_rflags;
1438 			rflags |= save_rflags & ~RMODE_GUEST_OWNED_EFLAGS_BITS;
1439 		}
1440 		vmx->rflags = rflags;
1441 	}
1442 	return vmx->rflags;
1443 }
1444 
1445 void vmx_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags)
1446 {
1447 	struct vcpu_vmx *vmx = to_vmx(vcpu);
1448 	unsigned long old_rflags;
1449 
1450 	if (enable_unrestricted_guest) {
1451 		kvm_register_mark_available(vcpu, VCPU_EXREG_RFLAGS);
1452 		vmx->rflags = rflags;
1453 		vmcs_writel(GUEST_RFLAGS, rflags);
1454 		return;
1455 	}
1456 
1457 	old_rflags = vmx_get_rflags(vcpu);
1458 	vmx->rflags = rflags;
1459 	if (vmx->rmode.vm86_active) {
1460 		vmx->rmode.save_rflags = rflags;
1461 		rflags |= X86_EFLAGS_IOPL | X86_EFLAGS_VM;
1462 	}
1463 	vmcs_writel(GUEST_RFLAGS, rflags);
1464 
1465 	if ((old_rflags ^ vmx->rflags) & X86_EFLAGS_VM)
1466 		vmx->emulation_required = emulation_required(vcpu);
1467 }
1468 
1469 u32 vmx_get_interrupt_shadow(struct kvm_vcpu *vcpu)
1470 {
1471 	u32 interruptibility = vmcs_read32(GUEST_INTERRUPTIBILITY_INFO);
1472 	int ret = 0;
1473 
1474 	if (interruptibility & GUEST_INTR_STATE_STI)
1475 		ret |= KVM_X86_SHADOW_INT_STI;
1476 	if (interruptibility & GUEST_INTR_STATE_MOV_SS)
1477 		ret |= KVM_X86_SHADOW_INT_MOV_SS;
1478 
1479 	return ret;
1480 }
1481 
1482 void vmx_set_interrupt_shadow(struct kvm_vcpu *vcpu, int mask)
1483 {
1484 	u32 interruptibility_old = vmcs_read32(GUEST_INTERRUPTIBILITY_INFO);
1485 	u32 interruptibility = interruptibility_old;
1486 
1487 	interruptibility &= ~(GUEST_INTR_STATE_STI | GUEST_INTR_STATE_MOV_SS);
1488 
1489 	if (mask & KVM_X86_SHADOW_INT_MOV_SS)
1490 		interruptibility |= GUEST_INTR_STATE_MOV_SS;
1491 	else if (mask & KVM_X86_SHADOW_INT_STI)
1492 		interruptibility |= GUEST_INTR_STATE_STI;
1493 
1494 	if ((interruptibility != interruptibility_old))
1495 		vmcs_write32(GUEST_INTERRUPTIBILITY_INFO, interruptibility);
1496 }
1497 
1498 static int vmx_rtit_ctl_check(struct kvm_vcpu *vcpu, u64 data)
1499 {
1500 	struct vcpu_vmx *vmx = to_vmx(vcpu);
1501 	unsigned long value;
1502 
1503 	/*
1504 	 * Any MSR write that attempts to change bits marked reserved will
1505 	 * case a #GP fault.
1506 	 */
1507 	if (data & vmx->pt_desc.ctl_bitmask)
1508 		return 1;
1509 
1510 	/*
1511 	 * Any attempt to modify IA32_RTIT_CTL while TraceEn is set will
1512 	 * result in a #GP unless the same write also clears TraceEn.
1513 	 */
1514 	if ((vmx->pt_desc.guest.ctl & RTIT_CTL_TRACEEN) &&
1515 		((vmx->pt_desc.guest.ctl ^ data) & ~RTIT_CTL_TRACEEN))
1516 		return 1;
1517 
1518 	/*
1519 	 * WRMSR to IA32_RTIT_CTL that sets TraceEn but clears this bit
1520 	 * and FabricEn would cause #GP, if
1521 	 * CPUID.(EAX=14H, ECX=0):ECX.SNGLRGNOUT[bit 2] = 0
1522 	 */
1523 	if ((data & RTIT_CTL_TRACEEN) && !(data & RTIT_CTL_TOPA) &&
1524 		!(data & RTIT_CTL_FABRIC_EN) &&
1525 		!intel_pt_validate_cap(vmx->pt_desc.caps,
1526 					PT_CAP_single_range_output))
1527 		return 1;
1528 
1529 	/*
1530 	 * MTCFreq, CycThresh and PSBFreq encodings check, any MSR write that
1531 	 * utilize encodings marked reserved will casue a #GP fault.
1532 	 */
1533 	value = intel_pt_validate_cap(vmx->pt_desc.caps, PT_CAP_mtc_periods);
1534 	if (intel_pt_validate_cap(vmx->pt_desc.caps, PT_CAP_mtc) &&
1535 			!test_bit((data & RTIT_CTL_MTC_RANGE) >>
1536 			RTIT_CTL_MTC_RANGE_OFFSET, &value))
1537 		return 1;
1538 	value = intel_pt_validate_cap(vmx->pt_desc.caps,
1539 						PT_CAP_cycle_thresholds);
1540 	if (intel_pt_validate_cap(vmx->pt_desc.caps, PT_CAP_psb_cyc) &&
1541 			!test_bit((data & RTIT_CTL_CYC_THRESH) >>
1542 			RTIT_CTL_CYC_THRESH_OFFSET, &value))
1543 		return 1;
1544 	value = intel_pt_validate_cap(vmx->pt_desc.caps, PT_CAP_psb_periods);
1545 	if (intel_pt_validate_cap(vmx->pt_desc.caps, PT_CAP_psb_cyc) &&
1546 			!test_bit((data & RTIT_CTL_PSB_FREQ) >>
1547 			RTIT_CTL_PSB_FREQ_OFFSET, &value))
1548 		return 1;
1549 
1550 	/*
1551 	 * If ADDRx_CFG is reserved or the encodings is >2 will
1552 	 * cause a #GP fault.
1553 	 */
1554 	value = (data & RTIT_CTL_ADDR0) >> RTIT_CTL_ADDR0_OFFSET;
1555 	if ((value && (vmx->pt_desc.addr_range < 1)) || (value > 2))
1556 		return 1;
1557 	value = (data & RTIT_CTL_ADDR1) >> RTIT_CTL_ADDR1_OFFSET;
1558 	if ((value && (vmx->pt_desc.addr_range < 2)) || (value > 2))
1559 		return 1;
1560 	value = (data & RTIT_CTL_ADDR2) >> RTIT_CTL_ADDR2_OFFSET;
1561 	if ((value && (vmx->pt_desc.addr_range < 3)) || (value > 2))
1562 		return 1;
1563 	value = (data & RTIT_CTL_ADDR3) >> RTIT_CTL_ADDR3_OFFSET;
1564 	if ((value && (vmx->pt_desc.addr_range < 4)) || (value > 2))
1565 		return 1;
1566 
1567 	return 0;
1568 }
1569 
1570 static int skip_emulated_instruction(struct kvm_vcpu *vcpu)
1571 {
1572 	unsigned long rip;
1573 
1574 	/*
1575 	 * Using VMCS.VM_EXIT_INSTRUCTION_LEN on EPT misconfig depends on
1576 	 * undefined behavior: Intel's SDM doesn't mandate the VMCS field be
1577 	 * set when EPT misconfig occurs.  In practice, real hardware updates
1578 	 * VM_EXIT_INSTRUCTION_LEN on EPT misconfig, but other hypervisors
1579 	 * (namely Hyper-V) don't set it due to it being undefined behavior,
1580 	 * i.e. we end up advancing IP with some random value.
1581 	 */
1582 	if (!static_cpu_has(X86_FEATURE_HYPERVISOR) ||
1583 	    to_vmx(vcpu)->exit_reason != EXIT_REASON_EPT_MISCONFIG) {
1584 		rip = kvm_rip_read(vcpu);
1585 		rip += vmcs_read32(VM_EXIT_INSTRUCTION_LEN);
1586 		kvm_rip_write(vcpu, rip);
1587 	} else {
1588 		if (!kvm_emulate_instruction(vcpu, EMULTYPE_SKIP))
1589 			return 0;
1590 	}
1591 
1592 	/* skipping an emulated instruction also counts */
1593 	vmx_set_interrupt_shadow(vcpu, 0);
1594 
1595 	return 1;
1596 }
1597 
1598 static void vmx_clear_hlt(struct kvm_vcpu *vcpu)
1599 {
1600 	/*
1601 	 * Ensure that we clear the HLT state in the VMCS.  We don't need to
1602 	 * explicitly skip the instruction because if the HLT state is set,
1603 	 * then the instruction is already executing and RIP has already been
1604 	 * advanced.
1605 	 */
1606 	if (kvm_hlt_in_guest(vcpu->kvm) &&
1607 			vmcs_read32(GUEST_ACTIVITY_STATE) == GUEST_ACTIVITY_HLT)
1608 		vmcs_write32(GUEST_ACTIVITY_STATE, GUEST_ACTIVITY_ACTIVE);
1609 }
1610 
1611 static void vmx_queue_exception(struct kvm_vcpu *vcpu)
1612 {
1613 	struct vcpu_vmx *vmx = to_vmx(vcpu);
1614 	unsigned nr = vcpu->arch.exception.nr;
1615 	bool has_error_code = vcpu->arch.exception.has_error_code;
1616 	u32 error_code = vcpu->arch.exception.error_code;
1617 	u32 intr_info = nr | INTR_INFO_VALID_MASK;
1618 
1619 	kvm_deliver_exception_payload(vcpu);
1620 
1621 	if (has_error_code) {
1622 		vmcs_write32(VM_ENTRY_EXCEPTION_ERROR_CODE, error_code);
1623 		intr_info |= INTR_INFO_DELIVER_CODE_MASK;
1624 	}
1625 
1626 	if (vmx->rmode.vm86_active) {
1627 		int inc_eip = 0;
1628 		if (kvm_exception_is_soft(nr))
1629 			inc_eip = vcpu->arch.event_exit_inst_len;
1630 		kvm_inject_realmode_interrupt(vcpu, nr, inc_eip);
1631 		return;
1632 	}
1633 
1634 	WARN_ON_ONCE(vmx->emulation_required);
1635 
1636 	if (kvm_exception_is_soft(nr)) {
1637 		vmcs_write32(VM_ENTRY_INSTRUCTION_LEN,
1638 			     vmx->vcpu.arch.event_exit_inst_len);
1639 		intr_info |= INTR_TYPE_SOFT_EXCEPTION;
1640 	} else
1641 		intr_info |= INTR_TYPE_HARD_EXCEPTION;
1642 
1643 	vmcs_write32(VM_ENTRY_INTR_INFO_FIELD, intr_info);
1644 
1645 	vmx_clear_hlt(vcpu);
1646 }
1647 
1648 static bool vmx_rdtscp_supported(void)
1649 {
1650 	return cpu_has_vmx_rdtscp();
1651 }
1652 
1653 static bool vmx_invpcid_supported(void)
1654 {
1655 	return cpu_has_vmx_invpcid();
1656 }
1657 
1658 /*
1659  * Swap MSR entry in host/guest MSR entry array.
1660  */
1661 static void move_msr_up(struct vcpu_vmx *vmx, int from, int to)
1662 {
1663 	struct shared_msr_entry tmp;
1664 
1665 	tmp = vmx->guest_msrs[to];
1666 	vmx->guest_msrs[to] = vmx->guest_msrs[from];
1667 	vmx->guest_msrs[from] = tmp;
1668 }
1669 
1670 /*
1671  * Set up the vmcs to automatically save and restore system
1672  * msrs.  Don't touch the 64-bit msrs if the guest is in legacy
1673  * mode, as fiddling with msrs is very expensive.
1674  */
1675 static void setup_msrs(struct vcpu_vmx *vmx)
1676 {
1677 	int save_nmsrs, index;
1678 
1679 	save_nmsrs = 0;
1680 #ifdef CONFIG_X86_64
1681 	/*
1682 	 * The SYSCALL MSRs are only needed on long mode guests, and only
1683 	 * when EFER.SCE is set.
1684 	 */
1685 	if (is_long_mode(&vmx->vcpu) && (vmx->vcpu.arch.efer & EFER_SCE)) {
1686 		index = __find_msr_index(vmx, MSR_STAR);
1687 		if (index >= 0)
1688 			move_msr_up(vmx, index, save_nmsrs++);
1689 		index = __find_msr_index(vmx, MSR_LSTAR);
1690 		if (index >= 0)
1691 			move_msr_up(vmx, index, save_nmsrs++);
1692 		index = __find_msr_index(vmx, MSR_SYSCALL_MASK);
1693 		if (index >= 0)
1694 			move_msr_up(vmx, index, save_nmsrs++);
1695 	}
1696 #endif
1697 	index = __find_msr_index(vmx, MSR_EFER);
1698 	if (index >= 0 && update_transition_efer(vmx, index))
1699 		move_msr_up(vmx, index, save_nmsrs++);
1700 	index = __find_msr_index(vmx, MSR_TSC_AUX);
1701 	if (index >= 0 && guest_cpuid_has(&vmx->vcpu, X86_FEATURE_RDTSCP))
1702 		move_msr_up(vmx, index, save_nmsrs++);
1703 	index = __find_msr_index(vmx, MSR_IA32_TSX_CTRL);
1704 	if (index >= 0)
1705 		move_msr_up(vmx, index, save_nmsrs++);
1706 
1707 	vmx->save_nmsrs = save_nmsrs;
1708 	vmx->guest_msrs_ready = false;
1709 
1710 	if (cpu_has_vmx_msr_bitmap())
1711 		vmx_update_msr_bitmap(&vmx->vcpu);
1712 }
1713 
1714 static u64 vmx_read_l1_tsc_offset(struct kvm_vcpu *vcpu)
1715 {
1716 	struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
1717 
1718 	if (is_guest_mode(vcpu) &&
1719 	    (vmcs12->cpu_based_vm_exec_control & CPU_BASED_USE_TSC_OFFSETING))
1720 		return vcpu->arch.tsc_offset - vmcs12->tsc_offset;
1721 
1722 	return vcpu->arch.tsc_offset;
1723 }
1724 
1725 static u64 vmx_write_l1_tsc_offset(struct kvm_vcpu *vcpu, u64 offset)
1726 {
1727 	struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
1728 	u64 g_tsc_offset = 0;
1729 
1730 	/*
1731 	 * We're here if L1 chose not to trap WRMSR to TSC. According
1732 	 * to the spec, this should set L1's TSC; The offset that L1
1733 	 * set for L2 remains unchanged, and still needs to be added
1734 	 * to the newly set TSC to get L2's TSC.
1735 	 */
1736 	if (is_guest_mode(vcpu) &&
1737 	    (vmcs12->cpu_based_vm_exec_control & CPU_BASED_USE_TSC_OFFSETING))
1738 		g_tsc_offset = vmcs12->tsc_offset;
1739 
1740 	trace_kvm_write_tsc_offset(vcpu->vcpu_id,
1741 				   vcpu->arch.tsc_offset - g_tsc_offset,
1742 				   offset);
1743 	vmcs_write64(TSC_OFFSET, offset + g_tsc_offset);
1744 	return offset + g_tsc_offset;
1745 }
1746 
1747 /*
1748  * nested_vmx_allowed() checks whether a guest should be allowed to use VMX
1749  * instructions and MSRs (i.e., nested VMX). Nested VMX is disabled for
1750  * all guests if the "nested" module option is off, and can also be disabled
1751  * for a single guest by disabling its VMX cpuid bit.
1752  */
1753 bool nested_vmx_allowed(struct kvm_vcpu *vcpu)
1754 {
1755 	return nested && guest_cpuid_has(vcpu, X86_FEATURE_VMX);
1756 }
1757 
1758 static inline bool vmx_feature_control_msr_valid(struct kvm_vcpu *vcpu,
1759 						 uint64_t val)
1760 {
1761 	uint64_t valid_bits = to_vmx(vcpu)->msr_ia32_feature_control_valid_bits;
1762 
1763 	return !(val & ~valid_bits);
1764 }
1765 
1766 static int vmx_get_msr_feature(struct kvm_msr_entry *msr)
1767 {
1768 	switch (msr->index) {
1769 	case MSR_IA32_VMX_BASIC ... MSR_IA32_VMX_VMFUNC:
1770 		if (!nested)
1771 			return 1;
1772 		return vmx_get_vmx_msr(&vmcs_config.nested, msr->index, &msr->data);
1773 	default:
1774 		return 1;
1775 	}
1776 
1777 	return 0;
1778 }
1779 
1780 /*
1781  * Reads an msr value (of 'msr_index') into 'pdata'.
1782  * Returns 0 on success, non-0 otherwise.
1783  * Assumes vcpu_load() was already called.
1784  */
1785 static int vmx_get_msr(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
1786 {
1787 	struct vcpu_vmx *vmx = to_vmx(vcpu);
1788 	struct shared_msr_entry *msr;
1789 	u32 index;
1790 
1791 	switch (msr_info->index) {
1792 #ifdef CONFIG_X86_64
1793 	case MSR_FS_BASE:
1794 		msr_info->data = vmcs_readl(GUEST_FS_BASE);
1795 		break;
1796 	case MSR_GS_BASE:
1797 		msr_info->data = vmcs_readl(GUEST_GS_BASE);
1798 		break;
1799 	case MSR_KERNEL_GS_BASE:
1800 		msr_info->data = vmx_read_guest_kernel_gs_base(vmx);
1801 		break;
1802 #endif
1803 	case MSR_EFER:
1804 		return kvm_get_msr_common(vcpu, msr_info);
1805 	case MSR_IA32_TSX_CTRL:
1806 		if (!msr_info->host_initiated &&
1807 		    !(vcpu->arch.arch_capabilities & ARCH_CAP_TSX_CTRL_MSR))
1808 			return 1;
1809 		goto find_shared_msr;
1810 	case MSR_IA32_UMWAIT_CONTROL:
1811 		if (!msr_info->host_initiated && !vmx_has_waitpkg(vmx))
1812 			return 1;
1813 
1814 		msr_info->data = vmx->msr_ia32_umwait_control;
1815 		break;
1816 	case MSR_IA32_SPEC_CTRL:
1817 		if (!msr_info->host_initiated &&
1818 		    !guest_cpuid_has(vcpu, X86_FEATURE_SPEC_CTRL))
1819 			return 1;
1820 
1821 		msr_info->data = to_vmx(vcpu)->spec_ctrl;
1822 		break;
1823 	case MSR_IA32_SYSENTER_CS:
1824 		msr_info->data = vmcs_read32(GUEST_SYSENTER_CS);
1825 		break;
1826 	case MSR_IA32_SYSENTER_EIP:
1827 		msr_info->data = vmcs_readl(GUEST_SYSENTER_EIP);
1828 		break;
1829 	case MSR_IA32_SYSENTER_ESP:
1830 		msr_info->data = vmcs_readl(GUEST_SYSENTER_ESP);
1831 		break;
1832 	case MSR_IA32_BNDCFGS:
1833 		if (!kvm_mpx_supported() ||
1834 		    (!msr_info->host_initiated &&
1835 		     !guest_cpuid_has(vcpu, X86_FEATURE_MPX)))
1836 			return 1;
1837 		msr_info->data = vmcs_read64(GUEST_BNDCFGS);
1838 		break;
1839 	case MSR_IA32_MCG_EXT_CTL:
1840 		if (!msr_info->host_initiated &&
1841 		    !(vmx->msr_ia32_feature_control &
1842 		      FEATURE_CONTROL_LMCE))
1843 			return 1;
1844 		msr_info->data = vcpu->arch.mcg_ext_ctl;
1845 		break;
1846 	case MSR_IA32_FEATURE_CONTROL:
1847 		msr_info->data = vmx->msr_ia32_feature_control;
1848 		break;
1849 	case MSR_IA32_VMX_BASIC ... MSR_IA32_VMX_VMFUNC:
1850 		if (!nested_vmx_allowed(vcpu))
1851 			return 1;
1852 		return vmx_get_vmx_msr(&vmx->nested.msrs, msr_info->index,
1853 				       &msr_info->data);
1854 	case MSR_IA32_RTIT_CTL:
1855 		if (pt_mode != PT_MODE_HOST_GUEST)
1856 			return 1;
1857 		msr_info->data = vmx->pt_desc.guest.ctl;
1858 		break;
1859 	case MSR_IA32_RTIT_STATUS:
1860 		if (pt_mode != PT_MODE_HOST_GUEST)
1861 			return 1;
1862 		msr_info->data = vmx->pt_desc.guest.status;
1863 		break;
1864 	case MSR_IA32_RTIT_CR3_MATCH:
1865 		if ((pt_mode != PT_MODE_HOST_GUEST) ||
1866 			!intel_pt_validate_cap(vmx->pt_desc.caps,
1867 						PT_CAP_cr3_filtering))
1868 			return 1;
1869 		msr_info->data = vmx->pt_desc.guest.cr3_match;
1870 		break;
1871 	case MSR_IA32_RTIT_OUTPUT_BASE:
1872 		if ((pt_mode != PT_MODE_HOST_GUEST) ||
1873 			(!intel_pt_validate_cap(vmx->pt_desc.caps,
1874 					PT_CAP_topa_output) &&
1875 			 !intel_pt_validate_cap(vmx->pt_desc.caps,
1876 					PT_CAP_single_range_output)))
1877 			return 1;
1878 		msr_info->data = vmx->pt_desc.guest.output_base;
1879 		break;
1880 	case MSR_IA32_RTIT_OUTPUT_MASK:
1881 		if ((pt_mode != PT_MODE_HOST_GUEST) ||
1882 			(!intel_pt_validate_cap(vmx->pt_desc.caps,
1883 					PT_CAP_topa_output) &&
1884 			 !intel_pt_validate_cap(vmx->pt_desc.caps,
1885 					PT_CAP_single_range_output)))
1886 			return 1;
1887 		msr_info->data = vmx->pt_desc.guest.output_mask;
1888 		break;
1889 	case MSR_IA32_RTIT_ADDR0_A ... MSR_IA32_RTIT_ADDR3_B:
1890 		index = msr_info->index - MSR_IA32_RTIT_ADDR0_A;
1891 		if ((pt_mode != PT_MODE_HOST_GUEST) ||
1892 			(index >= 2 * intel_pt_validate_cap(vmx->pt_desc.caps,
1893 					PT_CAP_num_address_ranges)))
1894 			return 1;
1895 		if (index % 2)
1896 			msr_info->data = vmx->pt_desc.guest.addr_b[index / 2];
1897 		else
1898 			msr_info->data = vmx->pt_desc.guest.addr_a[index / 2];
1899 		break;
1900 	case MSR_TSC_AUX:
1901 		if (!msr_info->host_initiated &&
1902 		    !guest_cpuid_has(vcpu, X86_FEATURE_RDTSCP))
1903 			return 1;
1904 		goto find_shared_msr;
1905 	default:
1906 	find_shared_msr:
1907 		msr = find_msr_entry(vmx, msr_info->index);
1908 		if (msr) {
1909 			msr_info->data = msr->data;
1910 			break;
1911 		}
1912 		return kvm_get_msr_common(vcpu, msr_info);
1913 	}
1914 
1915 	return 0;
1916 }
1917 
1918 /*
1919  * Writes msr value into into the appropriate "register".
1920  * Returns 0 on success, non-0 otherwise.
1921  * Assumes vcpu_load() was already called.
1922  */
1923 static int vmx_set_msr(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
1924 {
1925 	struct vcpu_vmx *vmx = to_vmx(vcpu);
1926 	struct shared_msr_entry *msr;
1927 	int ret = 0;
1928 	u32 msr_index = msr_info->index;
1929 	u64 data = msr_info->data;
1930 	u32 index;
1931 
1932 	switch (msr_index) {
1933 	case MSR_EFER:
1934 		ret = kvm_set_msr_common(vcpu, msr_info);
1935 		break;
1936 #ifdef CONFIG_X86_64
1937 	case MSR_FS_BASE:
1938 		vmx_segment_cache_clear(vmx);
1939 		vmcs_writel(GUEST_FS_BASE, data);
1940 		break;
1941 	case MSR_GS_BASE:
1942 		vmx_segment_cache_clear(vmx);
1943 		vmcs_writel(GUEST_GS_BASE, data);
1944 		break;
1945 	case MSR_KERNEL_GS_BASE:
1946 		vmx_write_guest_kernel_gs_base(vmx, data);
1947 		break;
1948 #endif
1949 	case MSR_IA32_SYSENTER_CS:
1950 		if (is_guest_mode(vcpu))
1951 			get_vmcs12(vcpu)->guest_sysenter_cs = data;
1952 		vmcs_write32(GUEST_SYSENTER_CS, data);
1953 		break;
1954 	case MSR_IA32_SYSENTER_EIP:
1955 		if (is_guest_mode(vcpu))
1956 			get_vmcs12(vcpu)->guest_sysenter_eip = data;
1957 		vmcs_writel(GUEST_SYSENTER_EIP, data);
1958 		break;
1959 	case MSR_IA32_SYSENTER_ESP:
1960 		if (is_guest_mode(vcpu))
1961 			get_vmcs12(vcpu)->guest_sysenter_esp = data;
1962 		vmcs_writel(GUEST_SYSENTER_ESP, data);
1963 		break;
1964 	case MSR_IA32_DEBUGCTLMSR:
1965 		if (is_guest_mode(vcpu) && get_vmcs12(vcpu)->vm_exit_controls &
1966 						VM_EXIT_SAVE_DEBUG_CONTROLS)
1967 			get_vmcs12(vcpu)->guest_ia32_debugctl = data;
1968 
1969 		ret = kvm_set_msr_common(vcpu, msr_info);
1970 		break;
1971 
1972 	case MSR_IA32_BNDCFGS:
1973 		if (!kvm_mpx_supported() ||
1974 		    (!msr_info->host_initiated &&
1975 		     !guest_cpuid_has(vcpu, X86_FEATURE_MPX)))
1976 			return 1;
1977 		if (is_noncanonical_address(data & PAGE_MASK, vcpu) ||
1978 		    (data & MSR_IA32_BNDCFGS_RSVD))
1979 			return 1;
1980 		vmcs_write64(GUEST_BNDCFGS, data);
1981 		break;
1982 	case MSR_IA32_UMWAIT_CONTROL:
1983 		if (!msr_info->host_initiated && !vmx_has_waitpkg(vmx))
1984 			return 1;
1985 
1986 		/* The reserved bit 1 and non-32 bit [63:32] should be zero */
1987 		if (data & (BIT_ULL(1) | GENMASK_ULL(63, 32)))
1988 			return 1;
1989 
1990 		vmx->msr_ia32_umwait_control = data;
1991 		break;
1992 	case MSR_IA32_SPEC_CTRL:
1993 		if (!msr_info->host_initiated &&
1994 		    !guest_cpuid_has(vcpu, X86_FEATURE_SPEC_CTRL))
1995 			return 1;
1996 
1997 		/* The STIBP bit doesn't fault even if it's not advertised */
1998 		if (data & ~(SPEC_CTRL_IBRS | SPEC_CTRL_STIBP | SPEC_CTRL_SSBD))
1999 			return 1;
2000 
2001 		vmx->spec_ctrl = data;
2002 
2003 		if (!data)
2004 			break;
2005 
2006 		/*
2007 		 * For non-nested:
2008 		 * When it's written (to non-zero) for the first time, pass
2009 		 * it through.
2010 		 *
2011 		 * For nested:
2012 		 * The handling of the MSR bitmap for L2 guests is done in
2013 		 * nested_vmx_merge_msr_bitmap. We should not touch the
2014 		 * vmcs02.msr_bitmap here since it gets completely overwritten
2015 		 * in the merging. We update the vmcs01 here for L1 as well
2016 		 * since it will end up touching the MSR anyway now.
2017 		 */
2018 		vmx_disable_intercept_for_msr(vmx->vmcs01.msr_bitmap,
2019 					      MSR_IA32_SPEC_CTRL,
2020 					      MSR_TYPE_RW);
2021 		break;
2022 	case MSR_IA32_TSX_CTRL:
2023 		if (!msr_info->host_initiated &&
2024 		    !(vcpu->arch.arch_capabilities & ARCH_CAP_TSX_CTRL_MSR))
2025 			return 1;
2026 		if (data & ~(TSX_CTRL_RTM_DISABLE | TSX_CTRL_CPUID_CLEAR))
2027 			return 1;
2028 		goto find_shared_msr;
2029 	case MSR_IA32_PRED_CMD:
2030 		if (!msr_info->host_initiated &&
2031 		    !guest_cpuid_has(vcpu, X86_FEATURE_SPEC_CTRL))
2032 			return 1;
2033 
2034 		if (data & ~PRED_CMD_IBPB)
2035 			return 1;
2036 
2037 		if (!data)
2038 			break;
2039 
2040 		wrmsrl(MSR_IA32_PRED_CMD, PRED_CMD_IBPB);
2041 
2042 		/*
2043 		 * For non-nested:
2044 		 * When it's written (to non-zero) for the first time, pass
2045 		 * it through.
2046 		 *
2047 		 * For nested:
2048 		 * The handling of the MSR bitmap for L2 guests is done in
2049 		 * nested_vmx_merge_msr_bitmap. We should not touch the
2050 		 * vmcs02.msr_bitmap here since it gets completely overwritten
2051 		 * in the merging.
2052 		 */
2053 		vmx_disable_intercept_for_msr(vmx->vmcs01.msr_bitmap, MSR_IA32_PRED_CMD,
2054 					      MSR_TYPE_W);
2055 		break;
2056 	case MSR_IA32_CR_PAT:
2057 		if (!kvm_pat_valid(data))
2058 			return 1;
2059 
2060 		if (is_guest_mode(vcpu) &&
2061 		    get_vmcs12(vcpu)->vm_exit_controls & VM_EXIT_SAVE_IA32_PAT)
2062 			get_vmcs12(vcpu)->guest_ia32_pat = data;
2063 
2064 		if (vmcs_config.vmentry_ctrl & VM_ENTRY_LOAD_IA32_PAT) {
2065 			vmcs_write64(GUEST_IA32_PAT, data);
2066 			vcpu->arch.pat = data;
2067 			break;
2068 		}
2069 		ret = kvm_set_msr_common(vcpu, msr_info);
2070 		break;
2071 	case MSR_IA32_TSC_ADJUST:
2072 		ret = kvm_set_msr_common(vcpu, msr_info);
2073 		break;
2074 	case MSR_IA32_MCG_EXT_CTL:
2075 		if ((!msr_info->host_initiated &&
2076 		     !(to_vmx(vcpu)->msr_ia32_feature_control &
2077 		       FEATURE_CONTROL_LMCE)) ||
2078 		    (data & ~MCG_EXT_CTL_LMCE_EN))
2079 			return 1;
2080 		vcpu->arch.mcg_ext_ctl = data;
2081 		break;
2082 	case MSR_IA32_FEATURE_CONTROL:
2083 		if (!vmx_feature_control_msr_valid(vcpu, data) ||
2084 		    (to_vmx(vcpu)->msr_ia32_feature_control &
2085 		     FEATURE_CONTROL_LOCKED && !msr_info->host_initiated))
2086 			return 1;
2087 		vmx->msr_ia32_feature_control = data;
2088 		if (msr_info->host_initiated && data == 0)
2089 			vmx_leave_nested(vcpu);
2090 		break;
2091 	case MSR_IA32_VMX_BASIC ... MSR_IA32_VMX_VMFUNC:
2092 		if (!msr_info->host_initiated)
2093 			return 1; /* they are read-only */
2094 		if (!nested_vmx_allowed(vcpu))
2095 			return 1;
2096 		return vmx_set_vmx_msr(vcpu, msr_index, data);
2097 	case MSR_IA32_RTIT_CTL:
2098 		if ((pt_mode != PT_MODE_HOST_GUEST) ||
2099 			vmx_rtit_ctl_check(vcpu, data) ||
2100 			vmx->nested.vmxon)
2101 			return 1;
2102 		vmcs_write64(GUEST_IA32_RTIT_CTL, data);
2103 		vmx->pt_desc.guest.ctl = data;
2104 		pt_update_intercept_for_msr(vmx);
2105 		break;
2106 	case MSR_IA32_RTIT_STATUS:
2107 		if ((pt_mode != PT_MODE_HOST_GUEST) ||
2108 			(vmx->pt_desc.guest.ctl & RTIT_CTL_TRACEEN) ||
2109 			(data & MSR_IA32_RTIT_STATUS_MASK))
2110 			return 1;
2111 		vmx->pt_desc.guest.status = data;
2112 		break;
2113 	case MSR_IA32_RTIT_CR3_MATCH:
2114 		if ((pt_mode != PT_MODE_HOST_GUEST) ||
2115 			(vmx->pt_desc.guest.ctl & RTIT_CTL_TRACEEN) ||
2116 			!intel_pt_validate_cap(vmx->pt_desc.caps,
2117 						PT_CAP_cr3_filtering))
2118 			return 1;
2119 		vmx->pt_desc.guest.cr3_match = data;
2120 		break;
2121 	case MSR_IA32_RTIT_OUTPUT_BASE:
2122 		if ((pt_mode != PT_MODE_HOST_GUEST) ||
2123 			(vmx->pt_desc.guest.ctl & RTIT_CTL_TRACEEN) ||
2124 			(!intel_pt_validate_cap(vmx->pt_desc.caps,
2125 					PT_CAP_topa_output) &&
2126 			 !intel_pt_validate_cap(vmx->pt_desc.caps,
2127 					PT_CAP_single_range_output)) ||
2128 			(data & MSR_IA32_RTIT_OUTPUT_BASE_MASK))
2129 			return 1;
2130 		vmx->pt_desc.guest.output_base = data;
2131 		break;
2132 	case MSR_IA32_RTIT_OUTPUT_MASK:
2133 		if ((pt_mode != PT_MODE_HOST_GUEST) ||
2134 			(vmx->pt_desc.guest.ctl & RTIT_CTL_TRACEEN) ||
2135 			(!intel_pt_validate_cap(vmx->pt_desc.caps,
2136 					PT_CAP_topa_output) &&
2137 			 !intel_pt_validate_cap(vmx->pt_desc.caps,
2138 					PT_CAP_single_range_output)))
2139 			return 1;
2140 		vmx->pt_desc.guest.output_mask = data;
2141 		break;
2142 	case MSR_IA32_RTIT_ADDR0_A ... MSR_IA32_RTIT_ADDR3_B:
2143 		index = msr_info->index - MSR_IA32_RTIT_ADDR0_A;
2144 		if ((pt_mode != PT_MODE_HOST_GUEST) ||
2145 			(vmx->pt_desc.guest.ctl & RTIT_CTL_TRACEEN) ||
2146 			(index >= 2 * intel_pt_validate_cap(vmx->pt_desc.caps,
2147 					PT_CAP_num_address_ranges)))
2148 			return 1;
2149 		if (index % 2)
2150 			vmx->pt_desc.guest.addr_b[index / 2] = data;
2151 		else
2152 			vmx->pt_desc.guest.addr_a[index / 2] = data;
2153 		break;
2154 	case MSR_TSC_AUX:
2155 		if (!msr_info->host_initiated &&
2156 		    !guest_cpuid_has(vcpu, X86_FEATURE_RDTSCP))
2157 			return 1;
2158 		/* Check reserved bit, higher 32 bits should be zero */
2159 		if ((data >> 32) != 0)
2160 			return 1;
2161 		goto find_shared_msr;
2162 
2163 	default:
2164 	find_shared_msr:
2165 		msr = find_msr_entry(vmx, msr_index);
2166 		if (msr)
2167 			ret = vmx_set_guest_msr(vmx, msr, data);
2168 		else
2169 			ret = kvm_set_msr_common(vcpu, msr_info);
2170 	}
2171 
2172 	return ret;
2173 }
2174 
2175 static void vmx_cache_reg(struct kvm_vcpu *vcpu, enum kvm_reg reg)
2176 {
2177 	kvm_register_mark_available(vcpu, reg);
2178 
2179 	switch (reg) {
2180 	case VCPU_REGS_RSP:
2181 		vcpu->arch.regs[VCPU_REGS_RSP] = vmcs_readl(GUEST_RSP);
2182 		break;
2183 	case VCPU_REGS_RIP:
2184 		vcpu->arch.regs[VCPU_REGS_RIP] = vmcs_readl(GUEST_RIP);
2185 		break;
2186 	case VCPU_EXREG_PDPTR:
2187 		if (enable_ept)
2188 			ept_save_pdptrs(vcpu);
2189 		break;
2190 	case VCPU_EXREG_CR3:
2191 		if (enable_unrestricted_guest || (enable_ept && is_paging(vcpu)))
2192 			vcpu->arch.cr3 = vmcs_readl(GUEST_CR3);
2193 		break;
2194 	default:
2195 		WARN_ON_ONCE(1);
2196 		break;
2197 	}
2198 }
2199 
2200 static __init int cpu_has_kvm_support(void)
2201 {
2202 	return cpu_has_vmx();
2203 }
2204 
2205 static __init int vmx_disabled_by_bios(void)
2206 {
2207 	u64 msr;
2208 
2209 	rdmsrl(MSR_IA32_FEATURE_CONTROL, msr);
2210 	if (msr & FEATURE_CONTROL_LOCKED) {
2211 		/* launched w/ TXT and VMX disabled */
2212 		if (!(msr & FEATURE_CONTROL_VMXON_ENABLED_INSIDE_SMX)
2213 			&& tboot_enabled())
2214 			return 1;
2215 		/* launched w/o TXT and VMX only enabled w/ TXT */
2216 		if (!(msr & FEATURE_CONTROL_VMXON_ENABLED_OUTSIDE_SMX)
2217 			&& (msr & FEATURE_CONTROL_VMXON_ENABLED_INSIDE_SMX)
2218 			&& !tboot_enabled()) {
2219 			printk(KERN_WARNING "kvm: disable TXT in the BIOS or "
2220 				"activate TXT before enabling KVM\n");
2221 			return 1;
2222 		}
2223 		/* launched w/o TXT and VMX disabled */
2224 		if (!(msr & FEATURE_CONTROL_VMXON_ENABLED_OUTSIDE_SMX)
2225 			&& !tboot_enabled())
2226 			return 1;
2227 	}
2228 
2229 	return 0;
2230 }
2231 
2232 static void kvm_cpu_vmxon(u64 addr)
2233 {
2234 	cr4_set_bits(X86_CR4_VMXE);
2235 	intel_pt_handle_vmx(1);
2236 
2237 	asm volatile ("vmxon %0" : : "m"(addr));
2238 }
2239 
2240 static int hardware_enable(void)
2241 {
2242 	int cpu = raw_smp_processor_id();
2243 	u64 phys_addr = __pa(per_cpu(vmxarea, cpu));
2244 	u64 old, test_bits;
2245 
2246 	if (cr4_read_shadow() & X86_CR4_VMXE)
2247 		return -EBUSY;
2248 
2249 	/*
2250 	 * This can happen if we hot-added a CPU but failed to allocate
2251 	 * VP assist page for it.
2252 	 */
2253 	if (static_branch_unlikely(&enable_evmcs) &&
2254 	    !hv_get_vp_assist_page(cpu))
2255 		return -EFAULT;
2256 
2257 	INIT_LIST_HEAD(&per_cpu(loaded_vmcss_on_cpu, cpu));
2258 	INIT_LIST_HEAD(&per_cpu(blocked_vcpu_on_cpu, cpu));
2259 	spin_lock_init(&per_cpu(blocked_vcpu_on_cpu_lock, cpu));
2260 
2261 	/*
2262 	 * Now we can enable the vmclear operation in kdump
2263 	 * since the loaded_vmcss_on_cpu list on this cpu
2264 	 * has been initialized.
2265 	 *
2266 	 * Though the cpu is not in VMX operation now, there
2267 	 * is no problem to enable the vmclear operation
2268 	 * for the loaded_vmcss_on_cpu list is empty!
2269 	 */
2270 	crash_enable_local_vmclear(cpu);
2271 
2272 	rdmsrl(MSR_IA32_FEATURE_CONTROL, old);
2273 
2274 	test_bits = FEATURE_CONTROL_LOCKED;
2275 	test_bits |= FEATURE_CONTROL_VMXON_ENABLED_OUTSIDE_SMX;
2276 	if (tboot_enabled())
2277 		test_bits |= FEATURE_CONTROL_VMXON_ENABLED_INSIDE_SMX;
2278 
2279 	if ((old & test_bits) != test_bits) {
2280 		/* enable and lock */
2281 		wrmsrl(MSR_IA32_FEATURE_CONTROL, old | test_bits);
2282 	}
2283 	kvm_cpu_vmxon(phys_addr);
2284 	if (enable_ept)
2285 		ept_sync_global();
2286 
2287 	return 0;
2288 }
2289 
2290 static void vmclear_local_loaded_vmcss(void)
2291 {
2292 	int cpu = raw_smp_processor_id();
2293 	struct loaded_vmcs *v, *n;
2294 
2295 	list_for_each_entry_safe(v, n, &per_cpu(loaded_vmcss_on_cpu, cpu),
2296 				 loaded_vmcss_on_cpu_link)
2297 		__loaded_vmcs_clear(v);
2298 }
2299 
2300 
2301 /* Just like cpu_vmxoff(), but with the __kvm_handle_fault_on_reboot()
2302  * tricks.
2303  */
2304 static void kvm_cpu_vmxoff(void)
2305 {
2306 	asm volatile (__ex("vmxoff"));
2307 
2308 	intel_pt_handle_vmx(0);
2309 	cr4_clear_bits(X86_CR4_VMXE);
2310 }
2311 
2312 static void hardware_disable(void)
2313 {
2314 	vmclear_local_loaded_vmcss();
2315 	kvm_cpu_vmxoff();
2316 }
2317 
2318 static __init int adjust_vmx_controls(u32 ctl_min, u32 ctl_opt,
2319 				      u32 msr, u32 *result)
2320 {
2321 	u32 vmx_msr_low, vmx_msr_high;
2322 	u32 ctl = ctl_min | ctl_opt;
2323 
2324 	rdmsr(msr, vmx_msr_low, vmx_msr_high);
2325 
2326 	ctl &= vmx_msr_high; /* bit == 0 in high word ==> must be zero */
2327 	ctl |= vmx_msr_low;  /* bit == 1 in low word  ==> must be one  */
2328 
2329 	/* Ensure minimum (required) set of control bits are supported. */
2330 	if (ctl_min & ~ctl)
2331 		return -EIO;
2332 
2333 	*result = ctl;
2334 	return 0;
2335 }
2336 
2337 static __init int setup_vmcs_config(struct vmcs_config *vmcs_conf,
2338 				    struct vmx_capability *vmx_cap)
2339 {
2340 	u32 vmx_msr_low, vmx_msr_high;
2341 	u32 min, opt, min2, opt2;
2342 	u32 _pin_based_exec_control = 0;
2343 	u32 _cpu_based_exec_control = 0;
2344 	u32 _cpu_based_2nd_exec_control = 0;
2345 	u32 _vmexit_control = 0;
2346 	u32 _vmentry_control = 0;
2347 
2348 	memset(vmcs_conf, 0, sizeof(*vmcs_conf));
2349 	min = CPU_BASED_HLT_EXITING |
2350 #ifdef CONFIG_X86_64
2351 	      CPU_BASED_CR8_LOAD_EXITING |
2352 	      CPU_BASED_CR8_STORE_EXITING |
2353 #endif
2354 	      CPU_BASED_CR3_LOAD_EXITING |
2355 	      CPU_BASED_CR3_STORE_EXITING |
2356 	      CPU_BASED_UNCOND_IO_EXITING |
2357 	      CPU_BASED_MOV_DR_EXITING |
2358 	      CPU_BASED_USE_TSC_OFFSETING |
2359 	      CPU_BASED_MWAIT_EXITING |
2360 	      CPU_BASED_MONITOR_EXITING |
2361 	      CPU_BASED_INVLPG_EXITING |
2362 	      CPU_BASED_RDPMC_EXITING;
2363 
2364 	opt = CPU_BASED_TPR_SHADOW |
2365 	      CPU_BASED_USE_MSR_BITMAPS |
2366 	      CPU_BASED_ACTIVATE_SECONDARY_CONTROLS;
2367 	if (adjust_vmx_controls(min, opt, MSR_IA32_VMX_PROCBASED_CTLS,
2368 				&_cpu_based_exec_control) < 0)
2369 		return -EIO;
2370 #ifdef CONFIG_X86_64
2371 	if ((_cpu_based_exec_control & CPU_BASED_TPR_SHADOW))
2372 		_cpu_based_exec_control &= ~CPU_BASED_CR8_LOAD_EXITING &
2373 					   ~CPU_BASED_CR8_STORE_EXITING;
2374 #endif
2375 	if (_cpu_based_exec_control & CPU_BASED_ACTIVATE_SECONDARY_CONTROLS) {
2376 		min2 = 0;
2377 		opt2 = SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES |
2378 			SECONDARY_EXEC_VIRTUALIZE_X2APIC_MODE |
2379 			SECONDARY_EXEC_WBINVD_EXITING |
2380 			SECONDARY_EXEC_ENABLE_VPID |
2381 			SECONDARY_EXEC_ENABLE_EPT |
2382 			SECONDARY_EXEC_UNRESTRICTED_GUEST |
2383 			SECONDARY_EXEC_PAUSE_LOOP_EXITING |
2384 			SECONDARY_EXEC_DESC |
2385 			SECONDARY_EXEC_RDTSCP |
2386 			SECONDARY_EXEC_ENABLE_INVPCID |
2387 			SECONDARY_EXEC_APIC_REGISTER_VIRT |
2388 			SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY |
2389 			SECONDARY_EXEC_SHADOW_VMCS |
2390 			SECONDARY_EXEC_XSAVES |
2391 			SECONDARY_EXEC_RDSEED_EXITING |
2392 			SECONDARY_EXEC_RDRAND_EXITING |
2393 			SECONDARY_EXEC_ENABLE_PML |
2394 			SECONDARY_EXEC_TSC_SCALING |
2395 			SECONDARY_EXEC_ENABLE_USR_WAIT_PAUSE |
2396 			SECONDARY_EXEC_PT_USE_GPA |
2397 			SECONDARY_EXEC_PT_CONCEAL_VMX |
2398 			SECONDARY_EXEC_ENABLE_VMFUNC |
2399 			SECONDARY_EXEC_ENCLS_EXITING;
2400 		if (adjust_vmx_controls(min2, opt2,
2401 					MSR_IA32_VMX_PROCBASED_CTLS2,
2402 					&_cpu_based_2nd_exec_control) < 0)
2403 			return -EIO;
2404 	}
2405 #ifndef CONFIG_X86_64
2406 	if (!(_cpu_based_2nd_exec_control &
2407 				SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES))
2408 		_cpu_based_exec_control &= ~CPU_BASED_TPR_SHADOW;
2409 #endif
2410 
2411 	if (!(_cpu_based_exec_control & CPU_BASED_TPR_SHADOW))
2412 		_cpu_based_2nd_exec_control &= ~(
2413 				SECONDARY_EXEC_APIC_REGISTER_VIRT |
2414 				SECONDARY_EXEC_VIRTUALIZE_X2APIC_MODE |
2415 				SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY);
2416 
2417 	rdmsr_safe(MSR_IA32_VMX_EPT_VPID_CAP,
2418 		&vmx_cap->ept, &vmx_cap->vpid);
2419 
2420 	if (_cpu_based_2nd_exec_control & SECONDARY_EXEC_ENABLE_EPT) {
2421 		/* CR3 accesses and invlpg don't need to cause VM Exits when EPT
2422 		   enabled */
2423 		_cpu_based_exec_control &= ~(CPU_BASED_CR3_LOAD_EXITING |
2424 					     CPU_BASED_CR3_STORE_EXITING |
2425 					     CPU_BASED_INVLPG_EXITING);
2426 	} else if (vmx_cap->ept) {
2427 		vmx_cap->ept = 0;
2428 		pr_warn_once("EPT CAP should not exist if not support "
2429 				"1-setting enable EPT VM-execution control\n");
2430 	}
2431 	if (!(_cpu_based_2nd_exec_control & SECONDARY_EXEC_ENABLE_VPID) &&
2432 		vmx_cap->vpid) {
2433 		vmx_cap->vpid = 0;
2434 		pr_warn_once("VPID CAP should not exist if not support "
2435 				"1-setting enable VPID VM-execution control\n");
2436 	}
2437 
2438 	min = VM_EXIT_SAVE_DEBUG_CONTROLS | VM_EXIT_ACK_INTR_ON_EXIT;
2439 #ifdef CONFIG_X86_64
2440 	min |= VM_EXIT_HOST_ADDR_SPACE_SIZE;
2441 #endif
2442 	opt = VM_EXIT_LOAD_IA32_PERF_GLOBAL_CTRL |
2443 	      VM_EXIT_LOAD_IA32_PAT |
2444 	      VM_EXIT_LOAD_IA32_EFER |
2445 	      VM_EXIT_CLEAR_BNDCFGS |
2446 	      VM_EXIT_PT_CONCEAL_PIP |
2447 	      VM_EXIT_CLEAR_IA32_RTIT_CTL;
2448 	if (adjust_vmx_controls(min, opt, MSR_IA32_VMX_EXIT_CTLS,
2449 				&_vmexit_control) < 0)
2450 		return -EIO;
2451 
2452 	min = PIN_BASED_EXT_INTR_MASK | PIN_BASED_NMI_EXITING;
2453 	opt = PIN_BASED_VIRTUAL_NMIS | PIN_BASED_POSTED_INTR |
2454 		 PIN_BASED_VMX_PREEMPTION_TIMER;
2455 	if (adjust_vmx_controls(min, opt, MSR_IA32_VMX_PINBASED_CTLS,
2456 				&_pin_based_exec_control) < 0)
2457 		return -EIO;
2458 
2459 	if (cpu_has_broken_vmx_preemption_timer())
2460 		_pin_based_exec_control &= ~PIN_BASED_VMX_PREEMPTION_TIMER;
2461 	if (!(_cpu_based_2nd_exec_control &
2462 		SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY))
2463 		_pin_based_exec_control &= ~PIN_BASED_POSTED_INTR;
2464 
2465 	min = VM_ENTRY_LOAD_DEBUG_CONTROLS;
2466 	opt = VM_ENTRY_LOAD_IA32_PERF_GLOBAL_CTRL |
2467 	      VM_ENTRY_LOAD_IA32_PAT |
2468 	      VM_ENTRY_LOAD_IA32_EFER |
2469 	      VM_ENTRY_LOAD_BNDCFGS |
2470 	      VM_ENTRY_PT_CONCEAL_PIP |
2471 	      VM_ENTRY_LOAD_IA32_RTIT_CTL;
2472 	if (adjust_vmx_controls(min, opt, MSR_IA32_VMX_ENTRY_CTLS,
2473 				&_vmentry_control) < 0)
2474 		return -EIO;
2475 
2476 	/*
2477 	 * Some cpus support VM_{ENTRY,EXIT}_IA32_PERF_GLOBAL_CTRL but they
2478 	 * can't be used due to an errata where VM Exit may incorrectly clear
2479 	 * IA32_PERF_GLOBAL_CTRL[34:32].  Workaround the errata by using the
2480 	 * MSR load mechanism to switch IA32_PERF_GLOBAL_CTRL.
2481 	 */
2482 	if (boot_cpu_data.x86 == 0x6) {
2483 		switch (boot_cpu_data.x86_model) {
2484 		case 26: /* AAK155 */
2485 		case 30: /* AAP115 */
2486 		case 37: /* AAT100 */
2487 		case 44: /* BC86,AAY89,BD102 */
2488 		case 46: /* BA97 */
2489 			_vmentry_control &= ~VM_ENTRY_LOAD_IA32_PERF_GLOBAL_CTRL;
2490 			_vmexit_control &= ~VM_EXIT_LOAD_IA32_PERF_GLOBAL_CTRL;
2491 			pr_warn_once("kvm: VM_EXIT_LOAD_IA32_PERF_GLOBAL_CTRL "
2492 					"does not work properly. Using workaround\n");
2493 			break;
2494 		default:
2495 			break;
2496 		}
2497 	}
2498 
2499 
2500 	rdmsr(MSR_IA32_VMX_BASIC, vmx_msr_low, vmx_msr_high);
2501 
2502 	/* IA-32 SDM Vol 3B: VMCS size is never greater than 4kB. */
2503 	if ((vmx_msr_high & 0x1fff) > PAGE_SIZE)
2504 		return -EIO;
2505 
2506 #ifdef CONFIG_X86_64
2507 	/* IA-32 SDM Vol 3B: 64-bit CPUs always have VMX_BASIC_MSR[48]==0. */
2508 	if (vmx_msr_high & (1u<<16))
2509 		return -EIO;
2510 #endif
2511 
2512 	/* Require Write-Back (WB) memory type for VMCS accesses. */
2513 	if (((vmx_msr_high >> 18) & 15) != 6)
2514 		return -EIO;
2515 
2516 	vmcs_conf->size = vmx_msr_high & 0x1fff;
2517 	vmcs_conf->order = get_order(vmcs_conf->size);
2518 	vmcs_conf->basic_cap = vmx_msr_high & ~0x1fff;
2519 
2520 	vmcs_conf->revision_id = vmx_msr_low;
2521 
2522 	vmcs_conf->pin_based_exec_ctrl = _pin_based_exec_control;
2523 	vmcs_conf->cpu_based_exec_ctrl = _cpu_based_exec_control;
2524 	vmcs_conf->cpu_based_2nd_exec_ctrl = _cpu_based_2nd_exec_control;
2525 	vmcs_conf->vmexit_ctrl         = _vmexit_control;
2526 	vmcs_conf->vmentry_ctrl        = _vmentry_control;
2527 
2528 	if (static_branch_unlikely(&enable_evmcs))
2529 		evmcs_sanitize_exec_ctrls(vmcs_conf);
2530 
2531 	return 0;
2532 }
2533 
2534 struct vmcs *alloc_vmcs_cpu(bool shadow, int cpu, gfp_t flags)
2535 {
2536 	int node = cpu_to_node(cpu);
2537 	struct page *pages;
2538 	struct vmcs *vmcs;
2539 
2540 	pages = __alloc_pages_node(node, flags, vmcs_config.order);
2541 	if (!pages)
2542 		return NULL;
2543 	vmcs = page_address(pages);
2544 	memset(vmcs, 0, vmcs_config.size);
2545 
2546 	/* KVM supports Enlightened VMCS v1 only */
2547 	if (static_branch_unlikely(&enable_evmcs))
2548 		vmcs->hdr.revision_id = KVM_EVMCS_VERSION;
2549 	else
2550 		vmcs->hdr.revision_id = vmcs_config.revision_id;
2551 
2552 	if (shadow)
2553 		vmcs->hdr.shadow_vmcs = 1;
2554 	return vmcs;
2555 }
2556 
2557 void free_vmcs(struct vmcs *vmcs)
2558 {
2559 	free_pages((unsigned long)vmcs, vmcs_config.order);
2560 }
2561 
2562 /*
2563  * Free a VMCS, but before that VMCLEAR it on the CPU where it was last loaded
2564  */
2565 void free_loaded_vmcs(struct loaded_vmcs *loaded_vmcs)
2566 {
2567 	if (!loaded_vmcs->vmcs)
2568 		return;
2569 	loaded_vmcs_clear(loaded_vmcs);
2570 	free_vmcs(loaded_vmcs->vmcs);
2571 	loaded_vmcs->vmcs = NULL;
2572 	if (loaded_vmcs->msr_bitmap)
2573 		free_page((unsigned long)loaded_vmcs->msr_bitmap);
2574 	WARN_ON(loaded_vmcs->shadow_vmcs != NULL);
2575 }
2576 
2577 int alloc_loaded_vmcs(struct loaded_vmcs *loaded_vmcs)
2578 {
2579 	loaded_vmcs->vmcs = alloc_vmcs(false);
2580 	if (!loaded_vmcs->vmcs)
2581 		return -ENOMEM;
2582 
2583 	loaded_vmcs->shadow_vmcs = NULL;
2584 	loaded_vmcs->hv_timer_soft_disabled = false;
2585 	loaded_vmcs_init(loaded_vmcs);
2586 
2587 	if (cpu_has_vmx_msr_bitmap()) {
2588 		loaded_vmcs->msr_bitmap = (unsigned long *)
2589 				__get_free_page(GFP_KERNEL_ACCOUNT);
2590 		if (!loaded_vmcs->msr_bitmap)
2591 			goto out_vmcs;
2592 		memset(loaded_vmcs->msr_bitmap, 0xff, PAGE_SIZE);
2593 
2594 		if (IS_ENABLED(CONFIG_HYPERV) &&
2595 		    static_branch_unlikely(&enable_evmcs) &&
2596 		    (ms_hyperv.nested_features & HV_X64_NESTED_MSR_BITMAP)) {
2597 			struct hv_enlightened_vmcs *evmcs =
2598 				(struct hv_enlightened_vmcs *)loaded_vmcs->vmcs;
2599 
2600 			evmcs->hv_enlightenments_control.msr_bitmap = 1;
2601 		}
2602 	}
2603 
2604 	memset(&loaded_vmcs->host_state, 0, sizeof(struct vmcs_host_state));
2605 	memset(&loaded_vmcs->controls_shadow, 0,
2606 		sizeof(struct vmcs_controls_shadow));
2607 
2608 	return 0;
2609 
2610 out_vmcs:
2611 	free_loaded_vmcs(loaded_vmcs);
2612 	return -ENOMEM;
2613 }
2614 
2615 static void free_kvm_area(void)
2616 {
2617 	int cpu;
2618 
2619 	for_each_possible_cpu(cpu) {
2620 		free_vmcs(per_cpu(vmxarea, cpu));
2621 		per_cpu(vmxarea, cpu) = NULL;
2622 	}
2623 }
2624 
2625 static __init int alloc_kvm_area(void)
2626 {
2627 	int cpu;
2628 
2629 	for_each_possible_cpu(cpu) {
2630 		struct vmcs *vmcs;
2631 
2632 		vmcs = alloc_vmcs_cpu(false, cpu, GFP_KERNEL);
2633 		if (!vmcs) {
2634 			free_kvm_area();
2635 			return -ENOMEM;
2636 		}
2637 
2638 		/*
2639 		 * When eVMCS is enabled, alloc_vmcs_cpu() sets
2640 		 * vmcs->revision_id to KVM_EVMCS_VERSION instead of
2641 		 * revision_id reported by MSR_IA32_VMX_BASIC.
2642 		 *
2643 		 * However, even though not explicitly documented by
2644 		 * TLFS, VMXArea passed as VMXON argument should
2645 		 * still be marked with revision_id reported by
2646 		 * physical CPU.
2647 		 */
2648 		if (static_branch_unlikely(&enable_evmcs))
2649 			vmcs->hdr.revision_id = vmcs_config.revision_id;
2650 
2651 		per_cpu(vmxarea, cpu) = vmcs;
2652 	}
2653 	return 0;
2654 }
2655 
2656 static void fix_pmode_seg(struct kvm_vcpu *vcpu, int seg,
2657 		struct kvm_segment *save)
2658 {
2659 	if (!emulate_invalid_guest_state) {
2660 		/*
2661 		 * CS and SS RPL should be equal during guest entry according
2662 		 * to VMX spec, but in reality it is not always so. Since vcpu
2663 		 * is in the middle of the transition from real mode to
2664 		 * protected mode it is safe to assume that RPL 0 is a good
2665 		 * default value.
2666 		 */
2667 		if (seg == VCPU_SREG_CS || seg == VCPU_SREG_SS)
2668 			save->selector &= ~SEGMENT_RPL_MASK;
2669 		save->dpl = save->selector & SEGMENT_RPL_MASK;
2670 		save->s = 1;
2671 	}
2672 	vmx_set_segment(vcpu, save, seg);
2673 }
2674 
2675 static void enter_pmode(struct kvm_vcpu *vcpu)
2676 {
2677 	unsigned long flags;
2678 	struct vcpu_vmx *vmx = to_vmx(vcpu);
2679 
2680 	/*
2681 	 * Update real mode segment cache. It may be not up-to-date if sement
2682 	 * register was written while vcpu was in a guest mode.
2683 	 */
2684 	vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_ES], VCPU_SREG_ES);
2685 	vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_DS], VCPU_SREG_DS);
2686 	vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_FS], VCPU_SREG_FS);
2687 	vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_GS], VCPU_SREG_GS);
2688 	vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_SS], VCPU_SREG_SS);
2689 	vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_CS], VCPU_SREG_CS);
2690 
2691 	vmx->rmode.vm86_active = 0;
2692 
2693 	vmx_segment_cache_clear(vmx);
2694 
2695 	vmx_set_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_TR], VCPU_SREG_TR);
2696 
2697 	flags = vmcs_readl(GUEST_RFLAGS);
2698 	flags &= RMODE_GUEST_OWNED_EFLAGS_BITS;
2699 	flags |= vmx->rmode.save_rflags & ~RMODE_GUEST_OWNED_EFLAGS_BITS;
2700 	vmcs_writel(GUEST_RFLAGS, flags);
2701 
2702 	vmcs_writel(GUEST_CR4, (vmcs_readl(GUEST_CR4) & ~X86_CR4_VME) |
2703 			(vmcs_readl(CR4_READ_SHADOW) & X86_CR4_VME));
2704 
2705 	update_exception_bitmap(vcpu);
2706 
2707 	fix_pmode_seg(vcpu, VCPU_SREG_CS, &vmx->rmode.segs[VCPU_SREG_CS]);
2708 	fix_pmode_seg(vcpu, VCPU_SREG_SS, &vmx->rmode.segs[VCPU_SREG_SS]);
2709 	fix_pmode_seg(vcpu, VCPU_SREG_ES, &vmx->rmode.segs[VCPU_SREG_ES]);
2710 	fix_pmode_seg(vcpu, VCPU_SREG_DS, &vmx->rmode.segs[VCPU_SREG_DS]);
2711 	fix_pmode_seg(vcpu, VCPU_SREG_FS, &vmx->rmode.segs[VCPU_SREG_FS]);
2712 	fix_pmode_seg(vcpu, VCPU_SREG_GS, &vmx->rmode.segs[VCPU_SREG_GS]);
2713 }
2714 
2715 static void fix_rmode_seg(int seg, struct kvm_segment *save)
2716 {
2717 	const struct kvm_vmx_segment_field *sf = &kvm_vmx_segment_fields[seg];
2718 	struct kvm_segment var = *save;
2719 
2720 	var.dpl = 0x3;
2721 	if (seg == VCPU_SREG_CS)
2722 		var.type = 0x3;
2723 
2724 	if (!emulate_invalid_guest_state) {
2725 		var.selector = var.base >> 4;
2726 		var.base = var.base & 0xffff0;
2727 		var.limit = 0xffff;
2728 		var.g = 0;
2729 		var.db = 0;
2730 		var.present = 1;
2731 		var.s = 1;
2732 		var.l = 0;
2733 		var.unusable = 0;
2734 		var.type = 0x3;
2735 		var.avl = 0;
2736 		if (save->base & 0xf)
2737 			printk_once(KERN_WARNING "kvm: segment base is not "
2738 					"paragraph aligned when entering "
2739 					"protected mode (seg=%d)", seg);
2740 	}
2741 
2742 	vmcs_write16(sf->selector, var.selector);
2743 	vmcs_writel(sf->base, var.base);
2744 	vmcs_write32(sf->limit, var.limit);
2745 	vmcs_write32(sf->ar_bytes, vmx_segment_access_rights(&var));
2746 }
2747 
2748 static void enter_rmode(struct kvm_vcpu *vcpu)
2749 {
2750 	unsigned long flags;
2751 	struct vcpu_vmx *vmx = to_vmx(vcpu);
2752 	struct kvm_vmx *kvm_vmx = to_kvm_vmx(vcpu->kvm);
2753 
2754 	vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_TR], VCPU_SREG_TR);
2755 	vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_ES], VCPU_SREG_ES);
2756 	vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_DS], VCPU_SREG_DS);
2757 	vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_FS], VCPU_SREG_FS);
2758 	vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_GS], VCPU_SREG_GS);
2759 	vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_SS], VCPU_SREG_SS);
2760 	vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_CS], VCPU_SREG_CS);
2761 
2762 	vmx->rmode.vm86_active = 1;
2763 
2764 	/*
2765 	 * Very old userspace does not call KVM_SET_TSS_ADDR before entering
2766 	 * vcpu. Warn the user that an update is overdue.
2767 	 */
2768 	if (!kvm_vmx->tss_addr)
2769 		printk_once(KERN_WARNING "kvm: KVM_SET_TSS_ADDR need to be "
2770 			     "called before entering vcpu\n");
2771 
2772 	vmx_segment_cache_clear(vmx);
2773 
2774 	vmcs_writel(GUEST_TR_BASE, kvm_vmx->tss_addr);
2775 	vmcs_write32(GUEST_TR_LIMIT, RMODE_TSS_SIZE - 1);
2776 	vmcs_write32(GUEST_TR_AR_BYTES, 0x008b);
2777 
2778 	flags = vmcs_readl(GUEST_RFLAGS);
2779 	vmx->rmode.save_rflags = flags;
2780 
2781 	flags |= X86_EFLAGS_IOPL | X86_EFLAGS_VM;
2782 
2783 	vmcs_writel(GUEST_RFLAGS, flags);
2784 	vmcs_writel(GUEST_CR4, vmcs_readl(GUEST_CR4) | X86_CR4_VME);
2785 	update_exception_bitmap(vcpu);
2786 
2787 	fix_rmode_seg(VCPU_SREG_SS, &vmx->rmode.segs[VCPU_SREG_SS]);
2788 	fix_rmode_seg(VCPU_SREG_CS, &vmx->rmode.segs[VCPU_SREG_CS]);
2789 	fix_rmode_seg(VCPU_SREG_ES, &vmx->rmode.segs[VCPU_SREG_ES]);
2790 	fix_rmode_seg(VCPU_SREG_DS, &vmx->rmode.segs[VCPU_SREG_DS]);
2791 	fix_rmode_seg(VCPU_SREG_GS, &vmx->rmode.segs[VCPU_SREG_GS]);
2792 	fix_rmode_seg(VCPU_SREG_FS, &vmx->rmode.segs[VCPU_SREG_FS]);
2793 
2794 	kvm_mmu_reset_context(vcpu);
2795 }
2796 
2797 void vmx_set_efer(struct kvm_vcpu *vcpu, u64 efer)
2798 {
2799 	struct vcpu_vmx *vmx = to_vmx(vcpu);
2800 	struct shared_msr_entry *msr = find_msr_entry(vmx, MSR_EFER);
2801 
2802 	if (!msr)
2803 		return;
2804 
2805 	vcpu->arch.efer = efer;
2806 	if (efer & EFER_LMA) {
2807 		vm_entry_controls_setbit(to_vmx(vcpu), VM_ENTRY_IA32E_MODE);
2808 		msr->data = efer;
2809 	} else {
2810 		vm_entry_controls_clearbit(to_vmx(vcpu), VM_ENTRY_IA32E_MODE);
2811 
2812 		msr->data = efer & ~EFER_LME;
2813 	}
2814 	setup_msrs(vmx);
2815 }
2816 
2817 #ifdef CONFIG_X86_64
2818 
2819 static void enter_lmode(struct kvm_vcpu *vcpu)
2820 {
2821 	u32 guest_tr_ar;
2822 
2823 	vmx_segment_cache_clear(to_vmx(vcpu));
2824 
2825 	guest_tr_ar = vmcs_read32(GUEST_TR_AR_BYTES);
2826 	if ((guest_tr_ar & VMX_AR_TYPE_MASK) != VMX_AR_TYPE_BUSY_64_TSS) {
2827 		pr_debug_ratelimited("%s: tss fixup for long mode. \n",
2828 				     __func__);
2829 		vmcs_write32(GUEST_TR_AR_BYTES,
2830 			     (guest_tr_ar & ~VMX_AR_TYPE_MASK)
2831 			     | VMX_AR_TYPE_BUSY_64_TSS);
2832 	}
2833 	vmx_set_efer(vcpu, vcpu->arch.efer | EFER_LMA);
2834 }
2835 
2836 static void exit_lmode(struct kvm_vcpu *vcpu)
2837 {
2838 	vm_entry_controls_clearbit(to_vmx(vcpu), VM_ENTRY_IA32E_MODE);
2839 	vmx_set_efer(vcpu, vcpu->arch.efer & ~EFER_LMA);
2840 }
2841 
2842 #endif
2843 
2844 static void vmx_flush_tlb_gva(struct kvm_vcpu *vcpu, gva_t addr)
2845 {
2846 	int vpid = to_vmx(vcpu)->vpid;
2847 
2848 	if (!vpid_sync_vcpu_addr(vpid, addr))
2849 		vpid_sync_context(vpid);
2850 
2851 	/*
2852 	 * If VPIDs are not supported or enabled, then the above is a no-op.
2853 	 * But we don't really need a TLB flush in that case anyway, because
2854 	 * each VM entry/exit includes an implicit flush when VPID is 0.
2855 	 */
2856 }
2857 
2858 static void vmx_decache_cr0_guest_bits(struct kvm_vcpu *vcpu)
2859 {
2860 	ulong cr0_guest_owned_bits = vcpu->arch.cr0_guest_owned_bits;
2861 
2862 	vcpu->arch.cr0 &= ~cr0_guest_owned_bits;
2863 	vcpu->arch.cr0 |= vmcs_readl(GUEST_CR0) & cr0_guest_owned_bits;
2864 }
2865 
2866 static void vmx_decache_cr4_guest_bits(struct kvm_vcpu *vcpu)
2867 {
2868 	ulong cr4_guest_owned_bits = vcpu->arch.cr4_guest_owned_bits;
2869 
2870 	vcpu->arch.cr4 &= ~cr4_guest_owned_bits;
2871 	vcpu->arch.cr4 |= vmcs_readl(GUEST_CR4) & cr4_guest_owned_bits;
2872 }
2873 
2874 static void ept_load_pdptrs(struct kvm_vcpu *vcpu)
2875 {
2876 	struct kvm_mmu *mmu = vcpu->arch.walk_mmu;
2877 
2878 	if (!kvm_register_is_dirty(vcpu, VCPU_EXREG_PDPTR))
2879 		return;
2880 
2881 	if (is_pae_paging(vcpu)) {
2882 		vmcs_write64(GUEST_PDPTR0, mmu->pdptrs[0]);
2883 		vmcs_write64(GUEST_PDPTR1, mmu->pdptrs[1]);
2884 		vmcs_write64(GUEST_PDPTR2, mmu->pdptrs[2]);
2885 		vmcs_write64(GUEST_PDPTR3, mmu->pdptrs[3]);
2886 	}
2887 }
2888 
2889 void ept_save_pdptrs(struct kvm_vcpu *vcpu)
2890 {
2891 	struct kvm_mmu *mmu = vcpu->arch.walk_mmu;
2892 
2893 	if (is_pae_paging(vcpu)) {
2894 		mmu->pdptrs[0] = vmcs_read64(GUEST_PDPTR0);
2895 		mmu->pdptrs[1] = vmcs_read64(GUEST_PDPTR1);
2896 		mmu->pdptrs[2] = vmcs_read64(GUEST_PDPTR2);
2897 		mmu->pdptrs[3] = vmcs_read64(GUEST_PDPTR3);
2898 	}
2899 
2900 	kvm_register_mark_dirty(vcpu, VCPU_EXREG_PDPTR);
2901 }
2902 
2903 static void ept_update_paging_mode_cr0(unsigned long *hw_cr0,
2904 					unsigned long cr0,
2905 					struct kvm_vcpu *vcpu)
2906 {
2907 	struct vcpu_vmx *vmx = to_vmx(vcpu);
2908 
2909 	if (!kvm_register_is_available(vcpu, VCPU_EXREG_CR3))
2910 		vmx_cache_reg(vcpu, VCPU_EXREG_CR3);
2911 	if (!(cr0 & X86_CR0_PG)) {
2912 		/* From paging/starting to nonpaging */
2913 		exec_controls_setbit(vmx, CPU_BASED_CR3_LOAD_EXITING |
2914 					  CPU_BASED_CR3_STORE_EXITING);
2915 		vcpu->arch.cr0 = cr0;
2916 		vmx_set_cr4(vcpu, kvm_read_cr4(vcpu));
2917 	} else if (!is_paging(vcpu)) {
2918 		/* From nonpaging to paging */
2919 		exec_controls_clearbit(vmx, CPU_BASED_CR3_LOAD_EXITING |
2920 					    CPU_BASED_CR3_STORE_EXITING);
2921 		vcpu->arch.cr0 = cr0;
2922 		vmx_set_cr4(vcpu, kvm_read_cr4(vcpu));
2923 	}
2924 
2925 	if (!(cr0 & X86_CR0_WP))
2926 		*hw_cr0 &= ~X86_CR0_WP;
2927 }
2928 
2929 void vmx_set_cr0(struct kvm_vcpu *vcpu, unsigned long cr0)
2930 {
2931 	struct vcpu_vmx *vmx = to_vmx(vcpu);
2932 	unsigned long hw_cr0;
2933 
2934 	hw_cr0 = (cr0 & ~KVM_VM_CR0_ALWAYS_OFF);
2935 	if (enable_unrestricted_guest)
2936 		hw_cr0 |= KVM_VM_CR0_ALWAYS_ON_UNRESTRICTED_GUEST;
2937 	else {
2938 		hw_cr0 |= KVM_VM_CR0_ALWAYS_ON;
2939 
2940 		if (vmx->rmode.vm86_active && (cr0 & X86_CR0_PE))
2941 			enter_pmode(vcpu);
2942 
2943 		if (!vmx->rmode.vm86_active && !(cr0 & X86_CR0_PE))
2944 			enter_rmode(vcpu);
2945 	}
2946 
2947 #ifdef CONFIG_X86_64
2948 	if (vcpu->arch.efer & EFER_LME) {
2949 		if (!is_paging(vcpu) && (cr0 & X86_CR0_PG))
2950 			enter_lmode(vcpu);
2951 		if (is_paging(vcpu) && !(cr0 & X86_CR0_PG))
2952 			exit_lmode(vcpu);
2953 	}
2954 #endif
2955 
2956 	if (enable_ept && !enable_unrestricted_guest)
2957 		ept_update_paging_mode_cr0(&hw_cr0, cr0, vcpu);
2958 
2959 	vmcs_writel(CR0_READ_SHADOW, cr0);
2960 	vmcs_writel(GUEST_CR0, hw_cr0);
2961 	vcpu->arch.cr0 = cr0;
2962 
2963 	/* depends on vcpu->arch.cr0 to be set to a new value */
2964 	vmx->emulation_required = emulation_required(vcpu);
2965 }
2966 
2967 static int get_ept_level(struct kvm_vcpu *vcpu)
2968 {
2969 	if (cpu_has_vmx_ept_5levels() && (cpuid_maxphyaddr(vcpu) > 48))
2970 		return 5;
2971 	return 4;
2972 }
2973 
2974 u64 construct_eptp(struct kvm_vcpu *vcpu, unsigned long root_hpa)
2975 {
2976 	u64 eptp = VMX_EPTP_MT_WB;
2977 
2978 	eptp |= (get_ept_level(vcpu) == 5) ? VMX_EPTP_PWL_5 : VMX_EPTP_PWL_4;
2979 
2980 	if (enable_ept_ad_bits &&
2981 	    (!is_guest_mode(vcpu) || nested_ept_ad_enabled(vcpu)))
2982 		eptp |= VMX_EPTP_AD_ENABLE_BIT;
2983 	eptp |= (root_hpa & PAGE_MASK);
2984 
2985 	return eptp;
2986 }
2987 
2988 void vmx_set_cr3(struct kvm_vcpu *vcpu, unsigned long cr3)
2989 {
2990 	struct kvm *kvm = vcpu->kvm;
2991 	bool update_guest_cr3 = true;
2992 	unsigned long guest_cr3;
2993 	u64 eptp;
2994 
2995 	guest_cr3 = cr3;
2996 	if (enable_ept) {
2997 		eptp = construct_eptp(vcpu, cr3);
2998 		vmcs_write64(EPT_POINTER, eptp);
2999 
3000 		if (kvm_x86_ops->tlb_remote_flush) {
3001 			spin_lock(&to_kvm_vmx(kvm)->ept_pointer_lock);
3002 			to_vmx(vcpu)->ept_pointer = eptp;
3003 			to_kvm_vmx(kvm)->ept_pointers_match
3004 				= EPT_POINTERS_CHECK;
3005 			spin_unlock(&to_kvm_vmx(kvm)->ept_pointer_lock);
3006 		}
3007 
3008 		/* Loading vmcs02.GUEST_CR3 is handled by nested VM-Enter. */
3009 		if (is_guest_mode(vcpu))
3010 			update_guest_cr3 = false;
3011 		else if (!enable_unrestricted_guest && !is_paging(vcpu))
3012 			guest_cr3 = to_kvm_vmx(kvm)->ept_identity_map_addr;
3013 		else if (test_bit(VCPU_EXREG_CR3, (ulong *)&vcpu->arch.regs_avail))
3014 			guest_cr3 = vcpu->arch.cr3;
3015 		else /* vmcs01.GUEST_CR3 is already up-to-date. */
3016 			update_guest_cr3 = false;
3017 		ept_load_pdptrs(vcpu);
3018 	}
3019 
3020 	if (update_guest_cr3)
3021 		vmcs_writel(GUEST_CR3, guest_cr3);
3022 }
3023 
3024 int vmx_set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4)
3025 {
3026 	struct vcpu_vmx *vmx = to_vmx(vcpu);
3027 	/*
3028 	 * Pass through host's Machine Check Enable value to hw_cr4, which
3029 	 * is in force while we are in guest mode.  Do not let guests control
3030 	 * this bit, even if host CR4.MCE == 0.
3031 	 */
3032 	unsigned long hw_cr4;
3033 
3034 	hw_cr4 = (cr4_read_shadow() & X86_CR4_MCE) | (cr4 & ~X86_CR4_MCE);
3035 	if (enable_unrestricted_guest)
3036 		hw_cr4 |= KVM_VM_CR4_ALWAYS_ON_UNRESTRICTED_GUEST;
3037 	else if (vmx->rmode.vm86_active)
3038 		hw_cr4 |= KVM_RMODE_VM_CR4_ALWAYS_ON;
3039 	else
3040 		hw_cr4 |= KVM_PMODE_VM_CR4_ALWAYS_ON;
3041 
3042 	if (!boot_cpu_has(X86_FEATURE_UMIP) && vmx_umip_emulated()) {
3043 		if (cr4 & X86_CR4_UMIP) {
3044 			secondary_exec_controls_setbit(vmx, SECONDARY_EXEC_DESC);
3045 			hw_cr4 &= ~X86_CR4_UMIP;
3046 		} else if (!is_guest_mode(vcpu) ||
3047 			!nested_cpu_has2(get_vmcs12(vcpu), SECONDARY_EXEC_DESC)) {
3048 			secondary_exec_controls_clearbit(vmx, SECONDARY_EXEC_DESC);
3049 		}
3050 	}
3051 
3052 	if (cr4 & X86_CR4_VMXE) {
3053 		/*
3054 		 * To use VMXON (and later other VMX instructions), a guest
3055 		 * must first be able to turn on cr4.VMXE (see handle_vmon()).
3056 		 * So basically the check on whether to allow nested VMX
3057 		 * is here.  We operate under the default treatment of SMM,
3058 		 * so VMX cannot be enabled under SMM.
3059 		 */
3060 		if (!nested_vmx_allowed(vcpu) || is_smm(vcpu))
3061 			return 1;
3062 	}
3063 
3064 	if (vmx->nested.vmxon && !nested_cr4_valid(vcpu, cr4))
3065 		return 1;
3066 
3067 	vcpu->arch.cr4 = cr4;
3068 
3069 	if (!enable_unrestricted_guest) {
3070 		if (enable_ept) {
3071 			if (!is_paging(vcpu)) {
3072 				hw_cr4 &= ~X86_CR4_PAE;
3073 				hw_cr4 |= X86_CR4_PSE;
3074 			} else if (!(cr4 & X86_CR4_PAE)) {
3075 				hw_cr4 &= ~X86_CR4_PAE;
3076 			}
3077 		}
3078 
3079 		/*
3080 		 * SMEP/SMAP/PKU is disabled if CPU is in non-paging mode in
3081 		 * hardware.  To emulate this behavior, SMEP/SMAP/PKU needs
3082 		 * to be manually disabled when guest switches to non-paging
3083 		 * mode.
3084 		 *
3085 		 * If !enable_unrestricted_guest, the CPU is always running
3086 		 * with CR0.PG=1 and CR4 needs to be modified.
3087 		 * If enable_unrestricted_guest, the CPU automatically
3088 		 * disables SMEP/SMAP/PKU when the guest sets CR0.PG=0.
3089 		 */
3090 		if (!is_paging(vcpu))
3091 			hw_cr4 &= ~(X86_CR4_SMEP | X86_CR4_SMAP | X86_CR4_PKE);
3092 	}
3093 
3094 	vmcs_writel(CR4_READ_SHADOW, cr4);
3095 	vmcs_writel(GUEST_CR4, hw_cr4);
3096 	return 0;
3097 }
3098 
3099 void vmx_get_segment(struct kvm_vcpu *vcpu, struct kvm_segment *var, int seg)
3100 {
3101 	struct vcpu_vmx *vmx = to_vmx(vcpu);
3102 	u32 ar;
3103 
3104 	if (vmx->rmode.vm86_active && seg != VCPU_SREG_LDTR) {
3105 		*var = vmx->rmode.segs[seg];
3106 		if (seg == VCPU_SREG_TR
3107 		    || var->selector == vmx_read_guest_seg_selector(vmx, seg))
3108 			return;
3109 		var->base = vmx_read_guest_seg_base(vmx, seg);
3110 		var->selector = vmx_read_guest_seg_selector(vmx, seg);
3111 		return;
3112 	}
3113 	var->base = vmx_read_guest_seg_base(vmx, seg);
3114 	var->limit = vmx_read_guest_seg_limit(vmx, seg);
3115 	var->selector = vmx_read_guest_seg_selector(vmx, seg);
3116 	ar = vmx_read_guest_seg_ar(vmx, seg);
3117 	var->unusable = (ar >> 16) & 1;
3118 	var->type = ar & 15;
3119 	var->s = (ar >> 4) & 1;
3120 	var->dpl = (ar >> 5) & 3;
3121 	/*
3122 	 * Some userspaces do not preserve unusable property. Since usable
3123 	 * segment has to be present according to VMX spec we can use present
3124 	 * property to amend userspace bug by making unusable segment always
3125 	 * nonpresent. vmx_segment_access_rights() already marks nonpresent
3126 	 * segment as unusable.
3127 	 */
3128 	var->present = !var->unusable;
3129 	var->avl = (ar >> 12) & 1;
3130 	var->l = (ar >> 13) & 1;
3131 	var->db = (ar >> 14) & 1;
3132 	var->g = (ar >> 15) & 1;
3133 }
3134 
3135 static u64 vmx_get_segment_base(struct kvm_vcpu *vcpu, int seg)
3136 {
3137 	struct kvm_segment s;
3138 
3139 	if (to_vmx(vcpu)->rmode.vm86_active) {
3140 		vmx_get_segment(vcpu, &s, seg);
3141 		return s.base;
3142 	}
3143 	return vmx_read_guest_seg_base(to_vmx(vcpu), seg);
3144 }
3145 
3146 int vmx_get_cpl(struct kvm_vcpu *vcpu)
3147 {
3148 	struct vcpu_vmx *vmx = to_vmx(vcpu);
3149 
3150 	if (unlikely(vmx->rmode.vm86_active))
3151 		return 0;
3152 	else {
3153 		int ar = vmx_read_guest_seg_ar(vmx, VCPU_SREG_SS);
3154 		return VMX_AR_DPL(ar);
3155 	}
3156 }
3157 
3158 static u32 vmx_segment_access_rights(struct kvm_segment *var)
3159 {
3160 	u32 ar;
3161 
3162 	if (var->unusable || !var->present)
3163 		ar = 1 << 16;
3164 	else {
3165 		ar = var->type & 15;
3166 		ar |= (var->s & 1) << 4;
3167 		ar |= (var->dpl & 3) << 5;
3168 		ar |= (var->present & 1) << 7;
3169 		ar |= (var->avl & 1) << 12;
3170 		ar |= (var->l & 1) << 13;
3171 		ar |= (var->db & 1) << 14;
3172 		ar |= (var->g & 1) << 15;
3173 	}
3174 
3175 	return ar;
3176 }
3177 
3178 void vmx_set_segment(struct kvm_vcpu *vcpu, struct kvm_segment *var, int seg)
3179 {
3180 	struct vcpu_vmx *vmx = to_vmx(vcpu);
3181 	const struct kvm_vmx_segment_field *sf = &kvm_vmx_segment_fields[seg];
3182 
3183 	vmx_segment_cache_clear(vmx);
3184 
3185 	if (vmx->rmode.vm86_active && seg != VCPU_SREG_LDTR) {
3186 		vmx->rmode.segs[seg] = *var;
3187 		if (seg == VCPU_SREG_TR)
3188 			vmcs_write16(sf->selector, var->selector);
3189 		else if (var->s)
3190 			fix_rmode_seg(seg, &vmx->rmode.segs[seg]);
3191 		goto out;
3192 	}
3193 
3194 	vmcs_writel(sf->base, var->base);
3195 	vmcs_write32(sf->limit, var->limit);
3196 	vmcs_write16(sf->selector, var->selector);
3197 
3198 	/*
3199 	 *   Fix the "Accessed" bit in AR field of segment registers for older
3200 	 * qemu binaries.
3201 	 *   IA32 arch specifies that at the time of processor reset the
3202 	 * "Accessed" bit in the AR field of segment registers is 1. And qemu
3203 	 * is setting it to 0 in the userland code. This causes invalid guest
3204 	 * state vmexit when "unrestricted guest" mode is turned on.
3205 	 *    Fix for this setup issue in cpu_reset is being pushed in the qemu
3206 	 * tree. Newer qemu binaries with that qemu fix would not need this
3207 	 * kvm hack.
3208 	 */
3209 	if (enable_unrestricted_guest && (seg != VCPU_SREG_LDTR))
3210 		var->type |= 0x1; /* Accessed */
3211 
3212 	vmcs_write32(sf->ar_bytes, vmx_segment_access_rights(var));
3213 
3214 out:
3215 	vmx->emulation_required = emulation_required(vcpu);
3216 }
3217 
3218 static void vmx_get_cs_db_l_bits(struct kvm_vcpu *vcpu, int *db, int *l)
3219 {
3220 	u32 ar = vmx_read_guest_seg_ar(to_vmx(vcpu), VCPU_SREG_CS);
3221 
3222 	*db = (ar >> 14) & 1;
3223 	*l = (ar >> 13) & 1;
3224 }
3225 
3226 static void vmx_get_idt(struct kvm_vcpu *vcpu, struct desc_ptr *dt)
3227 {
3228 	dt->size = vmcs_read32(GUEST_IDTR_LIMIT);
3229 	dt->address = vmcs_readl(GUEST_IDTR_BASE);
3230 }
3231 
3232 static void vmx_set_idt(struct kvm_vcpu *vcpu, struct desc_ptr *dt)
3233 {
3234 	vmcs_write32(GUEST_IDTR_LIMIT, dt->size);
3235 	vmcs_writel(GUEST_IDTR_BASE, dt->address);
3236 }
3237 
3238 static void vmx_get_gdt(struct kvm_vcpu *vcpu, struct desc_ptr *dt)
3239 {
3240 	dt->size = vmcs_read32(GUEST_GDTR_LIMIT);
3241 	dt->address = vmcs_readl(GUEST_GDTR_BASE);
3242 }
3243 
3244 static void vmx_set_gdt(struct kvm_vcpu *vcpu, struct desc_ptr *dt)
3245 {
3246 	vmcs_write32(GUEST_GDTR_LIMIT, dt->size);
3247 	vmcs_writel(GUEST_GDTR_BASE, dt->address);
3248 }
3249 
3250 static bool rmode_segment_valid(struct kvm_vcpu *vcpu, int seg)
3251 {
3252 	struct kvm_segment var;
3253 	u32 ar;
3254 
3255 	vmx_get_segment(vcpu, &var, seg);
3256 	var.dpl = 0x3;
3257 	if (seg == VCPU_SREG_CS)
3258 		var.type = 0x3;
3259 	ar = vmx_segment_access_rights(&var);
3260 
3261 	if (var.base != (var.selector << 4))
3262 		return false;
3263 	if (var.limit != 0xffff)
3264 		return false;
3265 	if (ar != 0xf3)
3266 		return false;
3267 
3268 	return true;
3269 }
3270 
3271 static bool code_segment_valid(struct kvm_vcpu *vcpu)
3272 {
3273 	struct kvm_segment cs;
3274 	unsigned int cs_rpl;
3275 
3276 	vmx_get_segment(vcpu, &cs, VCPU_SREG_CS);
3277 	cs_rpl = cs.selector & SEGMENT_RPL_MASK;
3278 
3279 	if (cs.unusable)
3280 		return false;
3281 	if (~cs.type & (VMX_AR_TYPE_CODE_MASK|VMX_AR_TYPE_ACCESSES_MASK))
3282 		return false;
3283 	if (!cs.s)
3284 		return false;
3285 	if (cs.type & VMX_AR_TYPE_WRITEABLE_MASK) {
3286 		if (cs.dpl > cs_rpl)
3287 			return false;
3288 	} else {
3289 		if (cs.dpl != cs_rpl)
3290 			return false;
3291 	}
3292 	if (!cs.present)
3293 		return false;
3294 
3295 	/* TODO: Add Reserved field check, this'll require a new member in the kvm_segment_field structure */
3296 	return true;
3297 }
3298 
3299 static bool stack_segment_valid(struct kvm_vcpu *vcpu)
3300 {
3301 	struct kvm_segment ss;
3302 	unsigned int ss_rpl;
3303 
3304 	vmx_get_segment(vcpu, &ss, VCPU_SREG_SS);
3305 	ss_rpl = ss.selector & SEGMENT_RPL_MASK;
3306 
3307 	if (ss.unusable)
3308 		return true;
3309 	if (ss.type != 3 && ss.type != 7)
3310 		return false;
3311 	if (!ss.s)
3312 		return false;
3313 	if (ss.dpl != ss_rpl) /* DPL != RPL */
3314 		return false;
3315 	if (!ss.present)
3316 		return false;
3317 
3318 	return true;
3319 }
3320 
3321 static bool data_segment_valid(struct kvm_vcpu *vcpu, int seg)
3322 {
3323 	struct kvm_segment var;
3324 	unsigned int rpl;
3325 
3326 	vmx_get_segment(vcpu, &var, seg);
3327 	rpl = var.selector & SEGMENT_RPL_MASK;
3328 
3329 	if (var.unusable)
3330 		return true;
3331 	if (!var.s)
3332 		return false;
3333 	if (!var.present)
3334 		return false;
3335 	if (~var.type & (VMX_AR_TYPE_CODE_MASK|VMX_AR_TYPE_WRITEABLE_MASK)) {
3336 		if (var.dpl < rpl) /* DPL < RPL */
3337 			return false;
3338 	}
3339 
3340 	/* TODO: Add other members to kvm_segment_field to allow checking for other access
3341 	 * rights flags
3342 	 */
3343 	return true;
3344 }
3345 
3346 static bool tr_valid(struct kvm_vcpu *vcpu)
3347 {
3348 	struct kvm_segment tr;
3349 
3350 	vmx_get_segment(vcpu, &tr, VCPU_SREG_TR);
3351 
3352 	if (tr.unusable)
3353 		return false;
3354 	if (tr.selector & SEGMENT_TI_MASK)	/* TI = 1 */
3355 		return false;
3356 	if (tr.type != 3 && tr.type != 11) /* TODO: Check if guest is in IA32e mode */
3357 		return false;
3358 	if (!tr.present)
3359 		return false;
3360 
3361 	return true;
3362 }
3363 
3364 static bool ldtr_valid(struct kvm_vcpu *vcpu)
3365 {
3366 	struct kvm_segment ldtr;
3367 
3368 	vmx_get_segment(vcpu, &ldtr, VCPU_SREG_LDTR);
3369 
3370 	if (ldtr.unusable)
3371 		return true;
3372 	if (ldtr.selector & SEGMENT_TI_MASK)	/* TI = 1 */
3373 		return false;
3374 	if (ldtr.type != 2)
3375 		return false;
3376 	if (!ldtr.present)
3377 		return false;
3378 
3379 	return true;
3380 }
3381 
3382 static bool cs_ss_rpl_check(struct kvm_vcpu *vcpu)
3383 {
3384 	struct kvm_segment cs, ss;
3385 
3386 	vmx_get_segment(vcpu, &cs, VCPU_SREG_CS);
3387 	vmx_get_segment(vcpu, &ss, VCPU_SREG_SS);
3388 
3389 	return ((cs.selector & SEGMENT_RPL_MASK) ==
3390 		 (ss.selector & SEGMENT_RPL_MASK));
3391 }
3392 
3393 /*
3394  * Check if guest state is valid. Returns true if valid, false if
3395  * not.
3396  * We assume that registers are always usable
3397  */
3398 static bool guest_state_valid(struct kvm_vcpu *vcpu)
3399 {
3400 	if (enable_unrestricted_guest)
3401 		return true;
3402 
3403 	/* real mode guest state checks */
3404 	if (!is_protmode(vcpu) || (vmx_get_rflags(vcpu) & X86_EFLAGS_VM)) {
3405 		if (!rmode_segment_valid(vcpu, VCPU_SREG_CS))
3406 			return false;
3407 		if (!rmode_segment_valid(vcpu, VCPU_SREG_SS))
3408 			return false;
3409 		if (!rmode_segment_valid(vcpu, VCPU_SREG_DS))
3410 			return false;
3411 		if (!rmode_segment_valid(vcpu, VCPU_SREG_ES))
3412 			return false;
3413 		if (!rmode_segment_valid(vcpu, VCPU_SREG_FS))
3414 			return false;
3415 		if (!rmode_segment_valid(vcpu, VCPU_SREG_GS))
3416 			return false;
3417 	} else {
3418 	/* protected mode guest state checks */
3419 		if (!cs_ss_rpl_check(vcpu))
3420 			return false;
3421 		if (!code_segment_valid(vcpu))
3422 			return false;
3423 		if (!stack_segment_valid(vcpu))
3424 			return false;
3425 		if (!data_segment_valid(vcpu, VCPU_SREG_DS))
3426 			return false;
3427 		if (!data_segment_valid(vcpu, VCPU_SREG_ES))
3428 			return false;
3429 		if (!data_segment_valid(vcpu, VCPU_SREG_FS))
3430 			return false;
3431 		if (!data_segment_valid(vcpu, VCPU_SREG_GS))
3432 			return false;
3433 		if (!tr_valid(vcpu))
3434 			return false;
3435 		if (!ldtr_valid(vcpu))
3436 			return false;
3437 	}
3438 	/* TODO:
3439 	 * - Add checks on RIP
3440 	 * - Add checks on RFLAGS
3441 	 */
3442 
3443 	return true;
3444 }
3445 
3446 static int init_rmode_tss(struct kvm *kvm)
3447 {
3448 	gfn_t fn;
3449 	u16 data = 0;
3450 	int idx, r;
3451 
3452 	idx = srcu_read_lock(&kvm->srcu);
3453 	fn = to_kvm_vmx(kvm)->tss_addr >> PAGE_SHIFT;
3454 	r = kvm_clear_guest_page(kvm, fn, 0, PAGE_SIZE);
3455 	if (r < 0)
3456 		goto out;
3457 	data = TSS_BASE_SIZE + TSS_REDIRECTION_SIZE;
3458 	r = kvm_write_guest_page(kvm, fn++, &data,
3459 			TSS_IOPB_BASE_OFFSET, sizeof(u16));
3460 	if (r < 0)
3461 		goto out;
3462 	r = kvm_clear_guest_page(kvm, fn++, 0, PAGE_SIZE);
3463 	if (r < 0)
3464 		goto out;
3465 	r = kvm_clear_guest_page(kvm, fn, 0, PAGE_SIZE);
3466 	if (r < 0)
3467 		goto out;
3468 	data = ~0;
3469 	r = kvm_write_guest_page(kvm, fn, &data,
3470 				 RMODE_TSS_SIZE - 2 * PAGE_SIZE - 1,
3471 				 sizeof(u8));
3472 out:
3473 	srcu_read_unlock(&kvm->srcu, idx);
3474 	return r;
3475 }
3476 
3477 static int init_rmode_identity_map(struct kvm *kvm)
3478 {
3479 	struct kvm_vmx *kvm_vmx = to_kvm_vmx(kvm);
3480 	int i, idx, r = 0;
3481 	kvm_pfn_t identity_map_pfn;
3482 	u32 tmp;
3483 
3484 	/* Protect kvm_vmx->ept_identity_pagetable_done. */
3485 	mutex_lock(&kvm->slots_lock);
3486 
3487 	if (likely(kvm_vmx->ept_identity_pagetable_done))
3488 		goto out2;
3489 
3490 	if (!kvm_vmx->ept_identity_map_addr)
3491 		kvm_vmx->ept_identity_map_addr = VMX_EPT_IDENTITY_PAGETABLE_ADDR;
3492 	identity_map_pfn = kvm_vmx->ept_identity_map_addr >> PAGE_SHIFT;
3493 
3494 	r = __x86_set_memory_region(kvm, IDENTITY_PAGETABLE_PRIVATE_MEMSLOT,
3495 				    kvm_vmx->ept_identity_map_addr, PAGE_SIZE);
3496 	if (r < 0)
3497 		goto out2;
3498 
3499 	idx = srcu_read_lock(&kvm->srcu);
3500 	r = kvm_clear_guest_page(kvm, identity_map_pfn, 0, PAGE_SIZE);
3501 	if (r < 0)
3502 		goto out;
3503 	/* Set up identity-mapping pagetable for EPT in real mode */
3504 	for (i = 0; i < PT32_ENT_PER_PAGE; i++) {
3505 		tmp = (i << 22) + (_PAGE_PRESENT | _PAGE_RW | _PAGE_USER |
3506 			_PAGE_ACCESSED | _PAGE_DIRTY | _PAGE_PSE);
3507 		r = kvm_write_guest_page(kvm, identity_map_pfn,
3508 				&tmp, i * sizeof(tmp), sizeof(tmp));
3509 		if (r < 0)
3510 			goto out;
3511 	}
3512 	kvm_vmx->ept_identity_pagetable_done = true;
3513 
3514 out:
3515 	srcu_read_unlock(&kvm->srcu, idx);
3516 
3517 out2:
3518 	mutex_unlock(&kvm->slots_lock);
3519 	return r;
3520 }
3521 
3522 static void seg_setup(int seg)
3523 {
3524 	const struct kvm_vmx_segment_field *sf = &kvm_vmx_segment_fields[seg];
3525 	unsigned int ar;
3526 
3527 	vmcs_write16(sf->selector, 0);
3528 	vmcs_writel(sf->base, 0);
3529 	vmcs_write32(sf->limit, 0xffff);
3530 	ar = 0x93;
3531 	if (seg == VCPU_SREG_CS)
3532 		ar |= 0x08; /* code segment */
3533 
3534 	vmcs_write32(sf->ar_bytes, ar);
3535 }
3536 
3537 static int alloc_apic_access_page(struct kvm *kvm)
3538 {
3539 	struct page *page;
3540 	int r = 0;
3541 
3542 	mutex_lock(&kvm->slots_lock);
3543 	if (kvm->arch.apic_access_page_done)
3544 		goto out;
3545 	r = __x86_set_memory_region(kvm, APIC_ACCESS_PAGE_PRIVATE_MEMSLOT,
3546 				    APIC_DEFAULT_PHYS_BASE, PAGE_SIZE);
3547 	if (r)
3548 		goto out;
3549 
3550 	page = gfn_to_page(kvm, APIC_DEFAULT_PHYS_BASE >> PAGE_SHIFT);
3551 	if (is_error_page(page)) {
3552 		r = -EFAULT;
3553 		goto out;
3554 	}
3555 
3556 	/*
3557 	 * Do not pin the page in memory, so that memory hot-unplug
3558 	 * is able to migrate it.
3559 	 */
3560 	put_page(page);
3561 	kvm->arch.apic_access_page_done = true;
3562 out:
3563 	mutex_unlock(&kvm->slots_lock);
3564 	return r;
3565 }
3566 
3567 int allocate_vpid(void)
3568 {
3569 	int vpid;
3570 
3571 	if (!enable_vpid)
3572 		return 0;
3573 	spin_lock(&vmx_vpid_lock);
3574 	vpid = find_first_zero_bit(vmx_vpid_bitmap, VMX_NR_VPIDS);
3575 	if (vpid < VMX_NR_VPIDS)
3576 		__set_bit(vpid, vmx_vpid_bitmap);
3577 	else
3578 		vpid = 0;
3579 	spin_unlock(&vmx_vpid_lock);
3580 	return vpid;
3581 }
3582 
3583 void free_vpid(int vpid)
3584 {
3585 	if (!enable_vpid || vpid == 0)
3586 		return;
3587 	spin_lock(&vmx_vpid_lock);
3588 	__clear_bit(vpid, vmx_vpid_bitmap);
3589 	spin_unlock(&vmx_vpid_lock);
3590 }
3591 
3592 static __always_inline void vmx_disable_intercept_for_msr(unsigned long *msr_bitmap,
3593 							  u32 msr, int type)
3594 {
3595 	int f = sizeof(unsigned long);
3596 
3597 	if (!cpu_has_vmx_msr_bitmap())
3598 		return;
3599 
3600 	if (static_branch_unlikely(&enable_evmcs))
3601 		evmcs_touch_msr_bitmap();
3602 
3603 	/*
3604 	 * See Intel PRM Vol. 3, 20.6.9 (MSR-Bitmap Address). Early manuals
3605 	 * have the write-low and read-high bitmap offsets the wrong way round.
3606 	 * We can control MSRs 0x00000000-0x00001fff and 0xc0000000-0xc0001fff.
3607 	 */
3608 	if (msr <= 0x1fff) {
3609 		if (type & MSR_TYPE_R)
3610 			/* read-low */
3611 			__clear_bit(msr, msr_bitmap + 0x000 / f);
3612 
3613 		if (type & MSR_TYPE_W)
3614 			/* write-low */
3615 			__clear_bit(msr, msr_bitmap + 0x800 / f);
3616 
3617 	} else if ((msr >= 0xc0000000) && (msr <= 0xc0001fff)) {
3618 		msr &= 0x1fff;
3619 		if (type & MSR_TYPE_R)
3620 			/* read-high */
3621 			__clear_bit(msr, msr_bitmap + 0x400 / f);
3622 
3623 		if (type & MSR_TYPE_W)
3624 			/* write-high */
3625 			__clear_bit(msr, msr_bitmap + 0xc00 / f);
3626 
3627 	}
3628 }
3629 
3630 static __always_inline void vmx_enable_intercept_for_msr(unsigned long *msr_bitmap,
3631 							 u32 msr, int type)
3632 {
3633 	int f = sizeof(unsigned long);
3634 
3635 	if (!cpu_has_vmx_msr_bitmap())
3636 		return;
3637 
3638 	if (static_branch_unlikely(&enable_evmcs))
3639 		evmcs_touch_msr_bitmap();
3640 
3641 	/*
3642 	 * See Intel PRM Vol. 3, 20.6.9 (MSR-Bitmap Address). Early manuals
3643 	 * have the write-low and read-high bitmap offsets the wrong way round.
3644 	 * We can control MSRs 0x00000000-0x00001fff and 0xc0000000-0xc0001fff.
3645 	 */
3646 	if (msr <= 0x1fff) {
3647 		if (type & MSR_TYPE_R)
3648 			/* read-low */
3649 			__set_bit(msr, msr_bitmap + 0x000 / f);
3650 
3651 		if (type & MSR_TYPE_W)
3652 			/* write-low */
3653 			__set_bit(msr, msr_bitmap + 0x800 / f);
3654 
3655 	} else if ((msr >= 0xc0000000) && (msr <= 0xc0001fff)) {
3656 		msr &= 0x1fff;
3657 		if (type & MSR_TYPE_R)
3658 			/* read-high */
3659 			__set_bit(msr, msr_bitmap + 0x400 / f);
3660 
3661 		if (type & MSR_TYPE_W)
3662 			/* write-high */
3663 			__set_bit(msr, msr_bitmap + 0xc00 / f);
3664 
3665 	}
3666 }
3667 
3668 static __always_inline void vmx_set_intercept_for_msr(unsigned long *msr_bitmap,
3669 			     			      u32 msr, int type, bool value)
3670 {
3671 	if (value)
3672 		vmx_enable_intercept_for_msr(msr_bitmap, msr, type);
3673 	else
3674 		vmx_disable_intercept_for_msr(msr_bitmap, msr, type);
3675 }
3676 
3677 static u8 vmx_msr_bitmap_mode(struct kvm_vcpu *vcpu)
3678 {
3679 	u8 mode = 0;
3680 
3681 	if (cpu_has_secondary_exec_ctrls() &&
3682 	    (secondary_exec_controls_get(to_vmx(vcpu)) &
3683 	     SECONDARY_EXEC_VIRTUALIZE_X2APIC_MODE)) {
3684 		mode |= MSR_BITMAP_MODE_X2APIC;
3685 		if (enable_apicv && kvm_vcpu_apicv_active(vcpu))
3686 			mode |= MSR_BITMAP_MODE_X2APIC_APICV;
3687 	}
3688 
3689 	return mode;
3690 }
3691 
3692 static void vmx_update_msr_bitmap_x2apic(unsigned long *msr_bitmap,
3693 					 u8 mode)
3694 {
3695 	int msr;
3696 
3697 	for (msr = 0x800; msr <= 0x8ff; msr += BITS_PER_LONG) {
3698 		unsigned word = msr / BITS_PER_LONG;
3699 		msr_bitmap[word] = (mode & MSR_BITMAP_MODE_X2APIC_APICV) ? 0 : ~0;
3700 		msr_bitmap[word + (0x800 / sizeof(long))] = ~0;
3701 	}
3702 
3703 	if (mode & MSR_BITMAP_MODE_X2APIC) {
3704 		/*
3705 		 * TPR reads and writes can be virtualized even if virtual interrupt
3706 		 * delivery is not in use.
3707 		 */
3708 		vmx_disable_intercept_for_msr(msr_bitmap, X2APIC_MSR(APIC_TASKPRI), MSR_TYPE_RW);
3709 		if (mode & MSR_BITMAP_MODE_X2APIC_APICV) {
3710 			vmx_enable_intercept_for_msr(msr_bitmap, X2APIC_MSR(APIC_TMCCT), MSR_TYPE_R);
3711 			vmx_disable_intercept_for_msr(msr_bitmap, X2APIC_MSR(APIC_EOI), MSR_TYPE_W);
3712 			vmx_disable_intercept_for_msr(msr_bitmap, X2APIC_MSR(APIC_SELF_IPI), MSR_TYPE_W);
3713 		}
3714 	}
3715 }
3716 
3717 void vmx_update_msr_bitmap(struct kvm_vcpu *vcpu)
3718 {
3719 	struct vcpu_vmx *vmx = to_vmx(vcpu);
3720 	unsigned long *msr_bitmap = vmx->vmcs01.msr_bitmap;
3721 	u8 mode = vmx_msr_bitmap_mode(vcpu);
3722 	u8 changed = mode ^ vmx->msr_bitmap_mode;
3723 
3724 	if (!changed)
3725 		return;
3726 
3727 	if (changed & (MSR_BITMAP_MODE_X2APIC | MSR_BITMAP_MODE_X2APIC_APICV))
3728 		vmx_update_msr_bitmap_x2apic(msr_bitmap, mode);
3729 
3730 	vmx->msr_bitmap_mode = mode;
3731 }
3732 
3733 void pt_update_intercept_for_msr(struct vcpu_vmx *vmx)
3734 {
3735 	unsigned long *msr_bitmap = vmx->vmcs01.msr_bitmap;
3736 	bool flag = !(vmx->pt_desc.guest.ctl & RTIT_CTL_TRACEEN);
3737 	u32 i;
3738 
3739 	vmx_set_intercept_for_msr(msr_bitmap, MSR_IA32_RTIT_STATUS,
3740 							MSR_TYPE_RW, flag);
3741 	vmx_set_intercept_for_msr(msr_bitmap, MSR_IA32_RTIT_OUTPUT_BASE,
3742 							MSR_TYPE_RW, flag);
3743 	vmx_set_intercept_for_msr(msr_bitmap, MSR_IA32_RTIT_OUTPUT_MASK,
3744 							MSR_TYPE_RW, flag);
3745 	vmx_set_intercept_for_msr(msr_bitmap, MSR_IA32_RTIT_CR3_MATCH,
3746 							MSR_TYPE_RW, flag);
3747 	for (i = 0; i < vmx->pt_desc.addr_range; i++) {
3748 		vmx_set_intercept_for_msr(msr_bitmap,
3749 			MSR_IA32_RTIT_ADDR0_A + i * 2, MSR_TYPE_RW, flag);
3750 		vmx_set_intercept_for_msr(msr_bitmap,
3751 			MSR_IA32_RTIT_ADDR0_B + i * 2, MSR_TYPE_RW, flag);
3752 	}
3753 }
3754 
3755 static bool vmx_get_enable_apicv(struct kvm *kvm)
3756 {
3757 	return enable_apicv;
3758 }
3759 
3760 static bool vmx_guest_apic_has_interrupt(struct kvm_vcpu *vcpu)
3761 {
3762 	struct vcpu_vmx *vmx = to_vmx(vcpu);
3763 	void *vapic_page;
3764 	u32 vppr;
3765 	int rvi;
3766 
3767 	if (WARN_ON_ONCE(!is_guest_mode(vcpu)) ||
3768 		!nested_cpu_has_vid(get_vmcs12(vcpu)) ||
3769 		WARN_ON_ONCE(!vmx->nested.virtual_apic_map.gfn))
3770 		return false;
3771 
3772 	rvi = vmx_get_rvi();
3773 
3774 	vapic_page = vmx->nested.virtual_apic_map.hva;
3775 	vppr = *((u32 *)(vapic_page + APIC_PROCPRI));
3776 
3777 	return ((rvi & 0xf0) > (vppr & 0xf0));
3778 }
3779 
3780 static inline bool kvm_vcpu_trigger_posted_interrupt(struct kvm_vcpu *vcpu,
3781 						     bool nested)
3782 {
3783 #ifdef CONFIG_SMP
3784 	int pi_vec = nested ? POSTED_INTR_NESTED_VECTOR : POSTED_INTR_VECTOR;
3785 
3786 	if (vcpu->mode == IN_GUEST_MODE) {
3787 		/*
3788 		 * The vector of interrupt to be delivered to vcpu had
3789 		 * been set in PIR before this function.
3790 		 *
3791 		 * Following cases will be reached in this block, and
3792 		 * we always send a notification event in all cases as
3793 		 * explained below.
3794 		 *
3795 		 * Case 1: vcpu keeps in non-root mode. Sending a
3796 		 * notification event posts the interrupt to vcpu.
3797 		 *
3798 		 * Case 2: vcpu exits to root mode and is still
3799 		 * runnable. PIR will be synced to vIRR before the
3800 		 * next vcpu entry. Sending a notification event in
3801 		 * this case has no effect, as vcpu is not in root
3802 		 * mode.
3803 		 *
3804 		 * Case 3: vcpu exits to root mode and is blocked.
3805 		 * vcpu_block() has already synced PIR to vIRR and
3806 		 * never blocks vcpu if vIRR is not cleared. Therefore,
3807 		 * a blocked vcpu here does not wait for any requested
3808 		 * interrupts in PIR, and sending a notification event
3809 		 * which has no effect is safe here.
3810 		 */
3811 
3812 		apic->send_IPI_mask(get_cpu_mask(vcpu->cpu), pi_vec);
3813 		return true;
3814 	}
3815 #endif
3816 	return false;
3817 }
3818 
3819 static int vmx_deliver_nested_posted_interrupt(struct kvm_vcpu *vcpu,
3820 						int vector)
3821 {
3822 	struct vcpu_vmx *vmx = to_vmx(vcpu);
3823 
3824 	if (is_guest_mode(vcpu) &&
3825 	    vector == vmx->nested.posted_intr_nv) {
3826 		/*
3827 		 * If a posted intr is not recognized by hardware,
3828 		 * we will accomplish it in the next vmentry.
3829 		 */
3830 		vmx->nested.pi_pending = true;
3831 		kvm_make_request(KVM_REQ_EVENT, vcpu);
3832 		/* the PIR and ON have been set by L1. */
3833 		if (!kvm_vcpu_trigger_posted_interrupt(vcpu, true))
3834 			kvm_vcpu_kick(vcpu);
3835 		return 0;
3836 	}
3837 	return -1;
3838 }
3839 /*
3840  * Send interrupt to vcpu via posted interrupt way.
3841  * 1. If target vcpu is running(non-root mode), send posted interrupt
3842  * notification to vcpu and hardware will sync PIR to vIRR atomically.
3843  * 2. If target vcpu isn't running(root mode), kick it to pick up the
3844  * interrupt from PIR in next vmentry.
3845  */
3846 static void vmx_deliver_posted_interrupt(struct kvm_vcpu *vcpu, int vector)
3847 {
3848 	struct vcpu_vmx *vmx = to_vmx(vcpu);
3849 	int r;
3850 
3851 	r = vmx_deliver_nested_posted_interrupt(vcpu, vector);
3852 	if (!r)
3853 		return;
3854 
3855 	if (pi_test_and_set_pir(vector, &vmx->pi_desc))
3856 		return;
3857 
3858 	/* If a previous notification has sent the IPI, nothing to do.  */
3859 	if (pi_test_and_set_on(&vmx->pi_desc))
3860 		return;
3861 
3862 	if (!kvm_vcpu_trigger_posted_interrupt(vcpu, false))
3863 		kvm_vcpu_kick(vcpu);
3864 }
3865 
3866 /*
3867  * Set up the vmcs's constant host-state fields, i.e., host-state fields that
3868  * will not change in the lifetime of the guest.
3869  * Note that host-state that does change is set elsewhere. E.g., host-state
3870  * that is set differently for each CPU is set in vmx_vcpu_load(), not here.
3871  */
3872 void vmx_set_constant_host_state(struct vcpu_vmx *vmx)
3873 {
3874 	u32 low32, high32;
3875 	unsigned long tmpl;
3876 	unsigned long cr0, cr3, cr4;
3877 
3878 	cr0 = read_cr0();
3879 	WARN_ON(cr0 & X86_CR0_TS);
3880 	vmcs_writel(HOST_CR0, cr0);  /* 22.2.3 */
3881 
3882 	/*
3883 	 * Save the most likely value for this task's CR3 in the VMCS.
3884 	 * We can't use __get_current_cr3_fast() because we're not atomic.
3885 	 */
3886 	cr3 = __read_cr3();
3887 	vmcs_writel(HOST_CR3, cr3);		/* 22.2.3  FIXME: shadow tables */
3888 	vmx->loaded_vmcs->host_state.cr3 = cr3;
3889 
3890 	/* Save the most likely value for this task's CR4 in the VMCS. */
3891 	cr4 = cr4_read_shadow();
3892 	vmcs_writel(HOST_CR4, cr4);			/* 22.2.3, 22.2.5 */
3893 	vmx->loaded_vmcs->host_state.cr4 = cr4;
3894 
3895 	vmcs_write16(HOST_CS_SELECTOR, __KERNEL_CS);  /* 22.2.4 */
3896 #ifdef CONFIG_X86_64
3897 	/*
3898 	 * Load null selectors, so we can avoid reloading them in
3899 	 * vmx_prepare_switch_to_host(), in case userspace uses
3900 	 * the null selectors too (the expected case).
3901 	 */
3902 	vmcs_write16(HOST_DS_SELECTOR, 0);
3903 	vmcs_write16(HOST_ES_SELECTOR, 0);
3904 #else
3905 	vmcs_write16(HOST_DS_SELECTOR, __KERNEL_DS);  /* 22.2.4 */
3906 	vmcs_write16(HOST_ES_SELECTOR, __KERNEL_DS);  /* 22.2.4 */
3907 #endif
3908 	vmcs_write16(HOST_SS_SELECTOR, __KERNEL_DS);  /* 22.2.4 */
3909 	vmcs_write16(HOST_TR_SELECTOR, GDT_ENTRY_TSS*8);  /* 22.2.4 */
3910 
3911 	vmcs_writel(HOST_IDTR_BASE, host_idt_base);   /* 22.2.4 */
3912 
3913 	vmcs_writel(HOST_RIP, (unsigned long)vmx_vmexit); /* 22.2.5 */
3914 
3915 	rdmsr(MSR_IA32_SYSENTER_CS, low32, high32);
3916 	vmcs_write32(HOST_IA32_SYSENTER_CS, low32);
3917 	rdmsrl(MSR_IA32_SYSENTER_EIP, tmpl);
3918 	vmcs_writel(HOST_IA32_SYSENTER_EIP, tmpl);   /* 22.2.3 */
3919 
3920 	if (vmcs_config.vmexit_ctrl & VM_EXIT_LOAD_IA32_PAT) {
3921 		rdmsr(MSR_IA32_CR_PAT, low32, high32);
3922 		vmcs_write64(HOST_IA32_PAT, low32 | ((u64) high32 << 32));
3923 	}
3924 
3925 	if (cpu_has_load_ia32_efer())
3926 		vmcs_write64(HOST_IA32_EFER, host_efer);
3927 }
3928 
3929 void set_cr4_guest_host_mask(struct vcpu_vmx *vmx)
3930 {
3931 	vmx->vcpu.arch.cr4_guest_owned_bits = KVM_CR4_GUEST_OWNED_BITS;
3932 	if (enable_ept)
3933 		vmx->vcpu.arch.cr4_guest_owned_bits |= X86_CR4_PGE;
3934 	if (is_guest_mode(&vmx->vcpu))
3935 		vmx->vcpu.arch.cr4_guest_owned_bits &=
3936 			~get_vmcs12(&vmx->vcpu)->cr4_guest_host_mask;
3937 	vmcs_writel(CR4_GUEST_HOST_MASK, ~vmx->vcpu.arch.cr4_guest_owned_bits);
3938 }
3939 
3940 u32 vmx_pin_based_exec_ctrl(struct vcpu_vmx *vmx)
3941 {
3942 	u32 pin_based_exec_ctrl = vmcs_config.pin_based_exec_ctrl;
3943 
3944 	if (!kvm_vcpu_apicv_active(&vmx->vcpu))
3945 		pin_based_exec_ctrl &= ~PIN_BASED_POSTED_INTR;
3946 
3947 	if (!enable_vnmi)
3948 		pin_based_exec_ctrl &= ~PIN_BASED_VIRTUAL_NMIS;
3949 
3950 	if (!enable_preemption_timer)
3951 		pin_based_exec_ctrl &= ~PIN_BASED_VMX_PREEMPTION_TIMER;
3952 
3953 	return pin_based_exec_ctrl;
3954 }
3955 
3956 static void vmx_refresh_apicv_exec_ctrl(struct kvm_vcpu *vcpu)
3957 {
3958 	struct vcpu_vmx *vmx = to_vmx(vcpu);
3959 
3960 	pin_controls_set(vmx, vmx_pin_based_exec_ctrl(vmx));
3961 	if (cpu_has_secondary_exec_ctrls()) {
3962 		if (kvm_vcpu_apicv_active(vcpu))
3963 			secondary_exec_controls_setbit(vmx,
3964 				      SECONDARY_EXEC_APIC_REGISTER_VIRT |
3965 				      SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY);
3966 		else
3967 			secondary_exec_controls_clearbit(vmx,
3968 					SECONDARY_EXEC_APIC_REGISTER_VIRT |
3969 					SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY);
3970 	}
3971 
3972 	if (cpu_has_vmx_msr_bitmap())
3973 		vmx_update_msr_bitmap(vcpu);
3974 }
3975 
3976 u32 vmx_exec_control(struct vcpu_vmx *vmx)
3977 {
3978 	u32 exec_control = vmcs_config.cpu_based_exec_ctrl;
3979 
3980 	if (vmx->vcpu.arch.switch_db_regs & KVM_DEBUGREG_WONT_EXIT)
3981 		exec_control &= ~CPU_BASED_MOV_DR_EXITING;
3982 
3983 	if (!cpu_need_tpr_shadow(&vmx->vcpu)) {
3984 		exec_control &= ~CPU_BASED_TPR_SHADOW;
3985 #ifdef CONFIG_X86_64
3986 		exec_control |= CPU_BASED_CR8_STORE_EXITING |
3987 				CPU_BASED_CR8_LOAD_EXITING;
3988 #endif
3989 	}
3990 	if (!enable_ept)
3991 		exec_control |= CPU_BASED_CR3_STORE_EXITING |
3992 				CPU_BASED_CR3_LOAD_EXITING  |
3993 				CPU_BASED_INVLPG_EXITING;
3994 	if (kvm_mwait_in_guest(vmx->vcpu.kvm))
3995 		exec_control &= ~(CPU_BASED_MWAIT_EXITING |
3996 				CPU_BASED_MONITOR_EXITING);
3997 	if (kvm_hlt_in_guest(vmx->vcpu.kvm))
3998 		exec_control &= ~CPU_BASED_HLT_EXITING;
3999 	return exec_control;
4000 }
4001 
4002 
4003 static void vmx_compute_secondary_exec_control(struct vcpu_vmx *vmx)
4004 {
4005 	struct kvm_vcpu *vcpu = &vmx->vcpu;
4006 
4007 	u32 exec_control = vmcs_config.cpu_based_2nd_exec_ctrl;
4008 
4009 	if (pt_mode == PT_MODE_SYSTEM)
4010 		exec_control &= ~(SECONDARY_EXEC_PT_USE_GPA | SECONDARY_EXEC_PT_CONCEAL_VMX);
4011 	if (!cpu_need_virtualize_apic_accesses(vcpu))
4012 		exec_control &= ~SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES;
4013 	if (vmx->vpid == 0)
4014 		exec_control &= ~SECONDARY_EXEC_ENABLE_VPID;
4015 	if (!enable_ept) {
4016 		exec_control &= ~SECONDARY_EXEC_ENABLE_EPT;
4017 		enable_unrestricted_guest = 0;
4018 	}
4019 	if (!enable_unrestricted_guest)
4020 		exec_control &= ~SECONDARY_EXEC_UNRESTRICTED_GUEST;
4021 	if (kvm_pause_in_guest(vmx->vcpu.kvm))
4022 		exec_control &= ~SECONDARY_EXEC_PAUSE_LOOP_EXITING;
4023 	if (!kvm_vcpu_apicv_active(vcpu))
4024 		exec_control &= ~(SECONDARY_EXEC_APIC_REGISTER_VIRT |
4025 				  SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY);
4026 	exec_control &= ~SECONDARY_EXEC_VIRTUALIZE_X2APIC_MODE;
4027 
4028 	/* SECONDARY_EXEC_DESC is enabled/disabled on writes to CR4.UMIP,
4029 	 * in vmx_set_cr4.  */
4030 	exec_control &= ~SECONDARY_EXEC_DESC;
4031 
4032 	/* SECONDARY_EXEC_SHADOW_VMCS is enabled when L1 executes VMPTRLD
4033 	   (handle_vmptrld).
4034 	   We can NOT enable shadow_vmcs here because we don't have yet
4035 	   a current VMCS12
4036 	*/
4037 	exec_control &= ~SECONDARY_EXEC_SHADOW_VMCS;
4038 
4039 	if (!enable_pml)
4040 		exec_control &= ~SECONDARY_EXEC_ENABLE_PML;
4041 
4042 	if (vmx_xsaves_supported()) {
4043 		/* Exposing XSAVES only when XSAVE is exposed */
4044 		bool xsaves_enabled =
4045 			guest_cpuid_has(vcpu, X86_FEATURE_XSAVE) &&
4046 			guest_cpuid_has(vcpu, X86_FEATURE_XSAVES);
4047 
4048 		vcpu->arch.xsaves_enabled = xsaves_enabled;
4049 
4050 		if (!xsaves_enabled)
4051 			exec_control &= ~SECONDARY_EXEC_XSAVES;
4052 
4053 		if (nested) {
4054 			if (xsaves_enabled)
4055 				vmx->nested.msrs.secondary_ctls_high |=
4056 					SECONDARY_EXEC_XSAVES;
4057 			else
4058 				vmx->nested.msrs.secondary_ctls_high &=
4059 					~SECONDARY_EXEC_XSAVES;
4060 		}
4061 	}
4062 
4063 	if (vmx_rdtscp_supported()) {
4064 		bool rdtscp_enabled = guest_cpuid_has(vcpu, X86_FEATURE_RDTSCP);
4065 		if (!rdtscp_enabled)
4066 			exec_control &= ~SECONDARY_EXEC_RDTSCP;
4067 
4068 		if (nested) {
4069 			if (rdtscp_enabled)
4070 				vmx->nested.msrs.secondary_ctls_high |=
4071 					SECONDARY_EXEC_RDTSCP;
4072 			else
4073 				vmx->nested.msrs.secondary_ctls_high &=
4074 					~SECONDARY_EXEC_RDTSCP;
4075 		}
4076 	}
4077 
4078 	if (vmx_invpcid_supported()) {
4079 		/* Exposing INVPCID only when PCID is exposed */
4080 		bool invpcid_enabled =
4081 			guest_cpuid_has(vcpu, X86_FEATURE_INVPCID) &&
4082 			guest_cpuid_has(vcpu, X86_FEATURE_PCID);
4083 
4084 		if (!invpcid_enabled) {
4085 			exec_control &= ~SECONDARY_EXEC_ENABLE_INVPCID;
4086 			guest_cpuid_clear(vcpu, X86_FEATURE_INVPCID);
4087 		}
4088 
4089 		if (nested) {
4090 			if (invpcid_enabled)
4091 				vmx->nested.msrs.secondary_ctls_high |=
4092 					SECONDARY_EXEC_ENABLE_INVPCID;
4093 			else
4094 				vmx->nested.msrs.secondary_ctls_high &=
4095 					~SECONDARY_EXEC_ENABLE_INVPCID;
4096 		}
4097 	}
4098 
4099 	if (vmx_rdrand_supported()) {
4100 		bool rdrand_enabled = guest_cpuid_has(vcpu, X86_FEATURE_RDRAND);
4101 		if (rdrand_enabled)
4102 			exec_control &= ~SECONDARY_EXEC_RDRAND_EXITING;
4103 
4104 		if (nested) {
4105 			if (rdrand_enabled)
4106 				vmx->nested.msrs.secondary_ctls_high |=
4107 					SECONDARY_EXEC_RDRAND_EXITING;
4108 			else
4109 				vmx->nested.msrs.secondary_ctls_high &=
4110 					~SECONDARY_EXEC_RDRAND_EXITING;
4111 		}
4112 	}
4113 
4114 	if (vmx_rdseed_supported()) {
4115 		bool rdseed_enabled = guest_cpuid_has(vcpu, X86_FEATURE_RDSEED);
4116 		if (rdseed_enabled)
4117 			exec_control &= ~SECONDARY_EXEC_RDSEED_EXITING;
4118 
4119 		if (nested) {
4120 			if (rdseed_enabled)
4121 				vmx->nested.msrs.secondary_ctls_high |=
4122 					SECONDARY_EXEC_RDSEED_EXITING;
4123 			else
4124 				vmx->nested.msrs.secondary_ctls_high &=
4125 					~SECONDARY_EXEC_RDSEED_EXITING;
4126 		}
4127 	}
4128 
4129 	if (vmx_waitpkg_supported()) {
4130 		bool waitpkg_enabled =
4131 			guest_cpuid_has(vcpu, X86_FEATURE_WAITPKG);
4132 
4133 		if (!waitpkg_enabled)
4134 			exec_control &= ~SECONDARY_EXEC_ENABLE_USR_WAIT_PAUSE;
4135 
4136 		if (nested) {
4137 			if (waitpkg_enabled)
4138 				vmx->nested.msrs.secondary_ctls_high |=
4139 					SECONDARY_EXEC_ENABLE_USR_WAIT_PAUSE;
4140 			else
4141 				vmx->nested.msrs.secondary_ctls_high &=
4142 					~SECONDARY_EXEC_ENABLE_USR_WAIT_PAUSE;
4143 		}
4144 	}
4145 
4146 	vmx->secondary_exec_control = exec_control;
4147 }
4148 
4149 static void ept_set_mmio_spte_mask(void)
4150 {
4151 	/*
4152 	 * EPT Misconfigurations can be generated if the value of bits 2:0
4153 	 * of an EPT paging-structure entry is 110b (write/execute).
4154 	 */
4155 	kvm_mmu_set_mmio_spte_mask(VMX_EPT_RWX_MASK,
4156 				   VMX_EPT_MISCONFIG_WX_VALUE, 0);
4157 }
4158 
4159 #define VMX_XSS_EXIT_BITMAP 0
4160 
4161 /*
4162  * Noting that the initialization of Guest-state Area of VMCS is in
4163  * vmx_vcpu_reset().
4164  */
4165 static void init_vmcs(struct vcpu_vmx *vmx)
4166 {
4167 	if (nested)
4168 		nested_vmx_set_vmcs_shadowing_bitmap();
4169 
4170 	if (cpu_has_vmx_msr_bitmap())
4171 		vmcs_write64(MSR_BITMAP, __pa(vmx->vmcs01.msr_bitmap));
4172 
4173 	vmcs_write64(VMCS_LINK_POINTER, -1ull); /* 22.3.1.5 */
4174 
4175 	/* Control */
4176 	pin_controls_set(vmx, vmx_pin_based_exec_ctrl(vmx));
4177 
4178 	exec_controls_set(vmx, vmx_exec_control(vmx));
4179 
4180 	if (cpu_has_secondary_exec_ctrls()) {
4181 		vmx_compute_secondary_exec_control(vmx);
4182 		secondary_exec_controls_set(vmx, vmx->secondary_exec_control);
4183 	}
4184 
4185 	if (kvm_vcpu_apicv_active(&vmx->vcpu)) {
4186 		vmcs_write64(EOI_EXIT_BITMAP0, 0);
4187 		vmcs_write64(EOI_EXIT_BITMAP1, 0);
4188 		vmcs_write64(EOI_EXIT_BITMAP2, 0);
4189 		vmcs_write64(EOI_EXIT_BITMAP3, 0);
4190 
4191 		vmcs_write16(GUEST_INTR_STATUS, 0);
4192 
4193 		vmcs_write16(POSTED_INTR_NV, POSTED_INTR_VECTOR);
4194 		vmcs_write64(POSTED_INTR_DESC_ADDR, __pa((&vmx->pi_desc)));
4195 	}
4196 
4197 	if (!kvm_pause_in_guest(vmx->vcpu.kvm)) {
4198 		vmcs_write32(PLE_GAP, ple_gap);
4199 		vmx->ple_window = ple_window;
4200 		vmx->ple_window_dirty = true;
4201 	}
4202 
4203 	vmcs_write32(PAGE_FAULT_ERROR_CODE_MASK, 0);
4204 	vmcs_write32(PAGE_FAULT_ERROR_CODE_MATCH, 0);
4205 	vmcs_write32(CR3_TARGET_COUNT, 0);           /* 22.2.1 */
4206 
4207 	vmcs_write16(HOST_FS_SELECTOR, 0);            /* 22.2.4 */
4208 	vmcs_write16(HOST_GS_SELECTOR, 0);            /* 22.2.4 */
4209 	vmx_set_constant_host_state(vmx);
4210 	vmcs_writel(HOST_FS_BASE, 0); /* 22.2.4 */
4211 	vmcs_writel(HOST_GS_BASE, 0); /* 22.2.4 */
4212 
4213 	if (cpu_has_vmx_vmfunc())
4214 		vmcs_write64(VM_FUNCTION_CONTROL, 0);
4215 
4216 	vmcs_write32(VM_EXIT_MSR_STORE_COUNT, 0);
4217 	vmcs_write32(VM_EXIT_MSR_LOAD_COUNT, 0);
4218 	vmcs_write64(VM_EXIT_MSR_LOAD_ADDR, __pa(vmx->msr_autoload.host.val));
4219 	vmcs_write32(VM_ENTRY_MSR_LOAD_COUNT, 0);
4220 	vmcs_write64(VM_ENTRY_MSR_LOAD_ADDR, __pa(vmx->msr_autoload.guest.val));
4221 
4222 	if (vmcs_config.vmentry_ctrl & VM_ENTRY_LOAD_IA32_PAT)
4223 		vmcs_write64(GUEST_IA32_PAT, vmx->vcpu.arch.pat);
4224 
4225 	vm_exit_controls_set(vmx, vmx_vmexit_ctrl());
4226 
4227 	/* 22.2.1, 20.8.1 */
4228 	vm_entry_controls_set(vmx, vmx_vmentry_ctrl());
4229 
4230 	vmx->vcpu.arch.cr0_guest_owned_bits = X86_CR0_TS;
4231 	vmcs_writel(CR0_GUEST_HOST_MASK, ~X86_CR0_TS);
4232 
4233 	set_cr4_guest_host_mask(vmx);
4234 
4235 	if (vmx->vpid != 0)
4236 		vmcs_write16(VIRTUAL_PROCESSOR_ID, vmx->vpid);
4237 
4238 	if (vmx_xsaves_supported())
4239 		vmcs_write64(XSS_EXIT_BITMAP, VMX_XSS_EXIT_BITMAP);
4240 
4241 	if (enable_pml) {
4242 		vmcs_write64(PML_ADDRESS, page_to_phys(vmx->pml_pg));
4243 		vmcs_write16(GUEST_PML_INDEX, PML_ENTITY_NUM - 1);
4244 	}
4245 
4246 	if (cpu_has_vmx_encls_vmexit())
4247 		vmcs_write64(ENCLS_EXITING_BITMAP, -1ull);
4248 
4249 	if (pt_mode == PT_MODE_HOST_GUEST) {
4250 		memset(&vmx->pt_desc, 0, sizeof(vmx->pt_desc));
4251 		/* Bit[6~0] are forced to 1, writes are ignored. */
4252 		vmx->pt_desc.guest.output_mask = 0x7F;
4253 		vmcs_write64(GUEST_IA32_RTIT_CTL, 0);
4254 	}
4255 }
4256 
4257 static void vmx_vcpu_reset(struct kvm_vcpu *vcpu, bool init_event)
4258 {
4259 	struct vcpu_vmx *vmx = to_vmx(vcpu);
4260 	struct msr_data apic_base_msr;
4261 	u64 cr0;
4262 
4263 	vmx->rmode.vm86_active = 0;
4264 	vmx->spec_ctrl = 0;
4265 
4266 	vmx->msr_ia32_umwait_control = 0;
4267 
4268 	vcpu->arch.microcode_version = 0x100000000ULL;
4269 	vmx->vcpu.arch.regs[VCPU_REGS_RDX] = get_rdx_init_val();
4270 	vmx->hv_deadline_tsc = -1;
4271 	kvm_set_cr8(vcpu, 0);
4272 
4273 	if (!init_event) {
4274 		apic_base_msr.data = APIC_DEFAULT_PHYS_BASE |
4275 				     MSR_IA32_APICBASE_ENABLE;
4276 		if (kvm_vcpu_is_reset_bsp(vcpu))
4277 			apic_base_msr.data |= MSR_IA32_APICBASE_BSP;
4278 		apic_base_msr.host_initiated = true;
4279 		kvm_set_apic_base(vcpu, &apic_base_msr);
4280 	}
4281 
4282 	vmx_segment_cache_clear(vmx);
4283 
4284 	seg_setup(VCPU_SREG_CS);
4285 	vmcs_write16(GUEST_CS_SELECTOR, 0xf000);
4286 	vmcs_writel(GUEST_CS_BASE, 0xffff0000ul);
4287 
4288 	seg_setup(VCPU_SREG_DS);
4289 	seg_setup(VCPU_SREG_ES);
4290 	seg_setup(VCPU_SREG_FS);
4291 	seg_setup(VCPU_SREG_GS);
4292 	seg_setup(VCPU_SREG_SS);
4293 
4294 	vmcs_write16(GUEST_TR_SELECTOR, 0);
4295 	vmcs_writel(GUEST_TR_BASE, 0);
4296 	vmcs_write32(GUEST_TR_LIMIT, 0xffff);
4297 	vmcs_write32(GUEST_TR_AR_BYTES, 0x008b);
4298 
4299 	vmcs_write16(GUEST_LDTR_SELECTOR, 0);
4300 	vmcs_writel(GUEST_LDTR_BASE, 0);
4301 	vmcs_write32(GUEST_LDTR_LIMIT, 0xffff);
4302 	vmcs_write32(GUEST_LDTR_AR_BYTES, 0x00082);
4303 
4304 	if (!init_event) {
4305 		vmcs_write32(GUEST_SYSENTER_CS, 0);
4306 		vmcs_writel(GUEST_SYSENTER_ESP, 0);
4307 		vmcs_writel(GUEST_SYSENTER_EIP, 0);
4308 		vmcs_write64(GUEST_IA32_DEBUGCTL, 0);
4309 	}
4310 
4311 	kvm_set_rflags(vcpu, X86_EFLAGS_FIXED);
4312 	kvm_rip_write(vcpu, 0xfff0);
4313 
4314 	vmcs_writel(GUEST_GDTR_BASE, 0);
4315 	vmcs_write32(GUEST_GDTR_LIMIT, 0xffff);
4316 
4317 	vmcs_writel(GUEST_IDTR_BASE, 0);
4318 	vmcs_write32(GUEST_IDTR_LIMIT, 0xffff);
4319 
4320 	vmcs_write32(GUEST_ACTIVITY_STATE, GUEST_ACTIVITY_ACTIVE);
4321 	vmcs_write32(GUEST_INTERRUPTIBILITY_INFO, 0);
4322 	vmcs_writel(GUEST_PENDING_DBG_EXCEPTIONS, 0);
4323 	if (kvm_mpx_supported())
4324 		vmcs_write64(GUEST_BNDCFGS, 0);
4325 
4326 	setup_msrs(vmx);
4327 
4328 	vmcs_write32(VM_ENTRY_INTR_INFO_FIELD, 0);  /* 22.2.1 */
4329 
4330 	if (cpu_has_vmx_tpr_shadow() && !init_event) {
4331 		vmcs_write64(VIRTUAL_APIC_PAGE_ADDR, 0);
4332 		if (cpu_need_tpr_shadow(vcpu))
4333 			vmcs_write64(VIRTUAL_APIC_PAGE_ADDR,
4334 				     __pa(vcpu->arch.apic->regs));
4335 		vmcs_write32(TPR_THRESHOLD, 0);
4336 	}
4337 
4338 	kvm_make_request(KVM_REQ_APIC_PAGE_RELOAD, vcpu);
4339 
4340 	cr0 = X86_CR0_NW | X86_CR0_CD | X86_CR0_ET;
4341 	vmx->vcpu.arch.cr0 = cr0;
4342 	vmx_set_cr0(vcpu, cr0); /* enter rmode */
4343 	vmx_set_cr4(vcpu, 0);
4344 	vmx_set_efer(vcpu, 0);
4345 
4346 	update_exception_bitmap(vcpu);
4347 
4348 	vpid_sync_context(vmx->vpid);
4349 	if (init_event)
4350 		vmx_clear_hlt(vcpu);
4351 }
4352 
4353 static void enable_irq_window(struct kvm_vcpu *vcpu)
4354 {
4355 	exec_controls_setbit(to_vmx(vcpu), CPU_BASED_VIRTUAL_INTR_PENDING);
4356 }
4357 
4358 static void enable_nmi_window(struct kvm_vcpu *vcpu)
4359 {
4360 	if (!enable_vnmi ||
4361 	    vmcs_read32(GUEST_INTERRUPTIBILITY_INFO) & GUEST_INTR_STATE_STI) {
4362 		enable_irq_window(vcpu);
4363 		return;
4364 	}
4365 
4366 	exec_controls_setbit(to_vmx(vcpu), CPU_BASED_VIRTUAL_NMI_PENDING);
4367 }
4368 
4369 static void vmx_inject_irq(struct kvm_vcpu *vcpu)
4370 {
4371 	struct vcpu_vmx *vmx = to_vmx(vcpu);
4372 	uint32_t intr;
4373 	int irq = vcpu->arch.interrupt.nr;
4374 
4375 	trace_kvm_inj_virq(irq);
4376 
4377 	++vcpu->stat.irq_injections;
4378 	if (vmx->rmode.vm86_active) {
4379 		int inc_eip = 0;
4380 		if (vcpu->arch.interrupt.soft)
4381 			inc_eip = vcpu->arch.event_exit_inst_len;
4382 		kvm_inject_realmode_interrupt(vcpu, irq, inc_eip);
4383 		return;
4384 	}
4385 	intr = irq | INTR_INFO_VALID_MASK;
4386 	if (vcpu->arch.interrupt.soft) {
4387 		intr |= INTR_TYPE_SOFT_INTR;
4388 		vmcs_write32(VM_ENTRY_INSTRUCTION_LEN,
4389 			     vmx->vcpu.arch.event_exit_inst_len);
4390 	} else
4391 		intr |= INTR_TYPE_EXT_INTR;
4392 	vmcs_write32(VM_ENTRY_INTR_INFO_FIELD, intr);
4393 
4394 	vmx_clear_hlt(vcpu);
4395 }
4396 
4397 static void vmx_inject_nmi(struct kvm_vcpu *vcpu)
4398 {
4399 	struct vcpu_vmx *vmx = to_vmx(vcpu);
4400 
4401 	if (!enable_vnmi) {
4402 		/*
4403 		 * Tracking the NMI-blocked state in software is built upon
4404 		 * finding the next open IRQ window. This, in turn, depends on
4405 		 * well-behaving guests: They have to keep IRQs disabled at
4406 		 * least as long as the NMI handler runs. Otherwise we may
4407 		 * cause NMI nesting, maybe breaking the guest. But as this is
4408 		 * highly unlikely, we can live with the residual risk.
4409 		 */
4410 		vmx->loaded_vmcs->soft_vnmi_blocked = 1;
4411 		vmx->loaded_vmcs->vnmi_blocked_time = 0;
4412 	}
4413 
4414 	++vcpu->stat.nmi_injections;
4415 	vmx->loaded_vmcs->nmi_known_unmasked = false;
4416 
4417 	if (vmx->rmode.vm86_active) {
4418 		kvm_inject_realmode_interrupt(vcpu, NMI_VECTOR, 0);
4419 		return;
4420 	}
4421 
4422 	vmcs_write32(VM_ENTRY_INTR_INFO_FIELD,
4423 			INTR_TYPE_NMI_INTR | INTR_INFO_VALID_MASK | NMI_VECTOR);
4424 
4425 	vmx_clear_hlt(vcpu);
4426 }
4427 
4428 bool vmx_get_nmi_mask(struct kvm_vcpu *vcpu)
4429 {
4430 	struct vcpu_vmx *vmx = to_vmx(vcpu);
4431 	bool masked;
4432 
4433 	if (!enable_vnmi)
4434 		return vmx->loaded_vmcs->soft_vnmi_blocked;
4435 	if (vmx->loaded_vmcs->nmi_known_unmasked)
4436 		return false;
4437 	masked = vmcs_read32(GUEST_INTERRUPTIBILITY_INFO) & GUEST_INTR_STATE_NMI;
4438 	vmx->loaded_vmcs->nmi_known_unmasked = !masked;
4439 	return masked;
4440 }
4441 
4442 void vmx_set_nmi_mask(struct kvm_vcpu *vcpu, bool masked)
4443 {
4444 	struct vcpu_vmx *vmx = to_vmx(vcpu);
4445 
4446 	if (!enable_vnmi) {
4447 		if (vmx->loaded_vmcs->soft_vnmi_blocked != masked) {
4448 			vmx->loaded_vmcs->soft_vnmi_blocked = masked;
4449 			vmx->loaded_vmcs->vnmi_blocked_time = 0;
4450 		}
4451 	} else {
4452 		vmx->loaded_vmcs->nmi_known_unmasked = !masked;
4453 		if (masked)
4454 			vmcs_set_bits(GUEST_INTERRUPTIBILITY_INFO,
4455 				      GUEST_INTR_STATE_NMI);
4456 		else
4457 			vmcs_clear_bits(GUEST_INTERRUPTIBILITY_INFO,
4458 					GUEST_INTR_STATE_NMI);
4459 	}
4460 }
4461 
4462 static int vmx_nmi_allowed(struct kvm_vcpu *vcpu)
4463 {
4464 	if (to_vmx(vcpu)->nested.nested_run_pending)
4465 		return 0;
4466 
4467 	if (!enable_vnmi &&
4468 	    to_vmx(vcpu)->loaded_vmcs->soft_vnmi_blocked)
4469 		return 0;
4470 
4471 	return	!(vmcs_read32(GUEST_INTERRUPTIBILITY_INFO) &
4472 		  (GUEST_INTR_STATE_MOV_SS | GUEST_INTR_STATE_STI
4473 		   | GUEST_INTR_STATE_NMI));
4474 }
4475 
4476 static int vmx_interrupt_allowed(struct kvm_vcpu *vcpu)
4477 {
4478 	return (!to_vmx(vcpu)->nested.nested_run_pending &&
4479 		vmcs_readl(GUEST_RFLAGS) & X86_EFLAGS_IF) &&
4480 		!(vmcs_read32(GUEST_INTERRUPTIBILITY_INFO) &
4481 			(GUEST_INTR_STATE_STI | GUEST_INTR_STATE_MOV_SS));
4482 }
4483 
4484 static int vmx_set_tss_addr(struct kvm *kvm, unsigned int addr)
4485 {
4486 	int ret;
4487 
4488 	if (enable_unrestricted_guest)
4489 		return 0;
4490 
4491 	ret = x86_set_memory_region(kvm, TSS_PRIVATE_MEMSLOT, addr,
4492 				    PAGE_SIZE * 3);
4493 	if (ret)
4494 		return ret;
4495 	to_kvm_vmx(kvm)->tss_addr = addr;
4496 	return init_rmode_tss(kvm);
4497 }
4498 
4499 static int vmx_set_identity_map_addr(struct kvm *kvm, u64 ident_addr)
4500 {
4501 	to_kvm_vmx(kvm)->ept_identity_map_addr = ident_addr;
4502 	return 0;
4503 }
4504 
4505 static bool rmode_exception(struct kvm_vcpu *vcpu, int vec)
4506 {
4507 	switch (vec) {
4508 	case BP_VECTOR:
4509 		/*
4510 		 * Update instruction length as we may reinject the exception
4511 		 * from user space while in guest debugging mode.
4512 		 */
4513 		to_vmx(vcpu)->vcpu.arch.event_exit_inst_len =
4514 			vmcs_read32(VM_EXIT_INSTRUCTION_LEN);
4515 		if (vcpu->guest_debug & KVM_GUESTDBG_USE_SW_BP)
4516 			return false;
4517 		/* fall through */
4518 	case DB_VECTOR:
4519 		if (vcpu->guest_debug &
4520 			(KVM_GUESTDBG_SINGLESTEP | KVM_GUESTDBG_USE_HW_BP))
4521 			return false;
4522 		/* fall through */
4523 	case DE_VECTOR:
4524 	case OF_VECTOR:
4525 	case BR_VECTOR:
4526 	case UD_VECTOR:
4527 	case DF_VECTOR:
4528 	case SS_VECTOR:
4529 	case GP_VECTOR:
4530 	case MF_VECTOR:
4531 		return true;
4532 	break;
4533 	}
4534 	return false;
4535 }
4536 
4537 static int handle_rmode_exception(struct kvm_vcpu *vcpu,
4538 				  int vec, u32 err_code)
4539 {
4540 	/*
4541 	 * Instruction with address size override prefix opcode 0x67
4542 	 * Cause the #SS fault with 0 error code in VM86 mode.
4543 	 */
4544 	if (((vec == GP_VECTOR) || (vec == SS_VECTOR)) && err_code == 0) {
4545 		if (kvm_emulate_instruction(vcpu, 0)) {
4546 			if (vcpu->arch.halt_request) {
4547 				vcpu->arch.halt_request = 0;
4548 				return kvm_vcpu_halt(vcpu);
4549 			}
4550 			return 1;
4551 		}
4552 		return 0;
4553 	}
4554 
4555 	/*
4556 	 * Forward all other exceptions that are valid in real mode.
4557 	 * FIXME: Breaks guest debugging in real mode, needs to be fixed with
4558 	 *        the required debugging infrastructure rework.
4559 	 */
4560 	kvm_queue_exception(vcpu, vec);
4561 	return 1;
4562 }
4563 
4564 /*
4565  * Trigger machine check on the host. We assume all the MSRs are already set up
4566  * by the CPU and that we still run on the same CPU as the MCE occurred on.
4567  * We pass a fake environment to the machine check handler because we want
4568  * the guest to be always treated like user space, no matter what context
4569  * it used internally.
4570  */
4571 static void kvm_machine_check(void)
4572 {
4573 #if defined(CONFIG_X86_MCE) && defined(CONFIG_X86_64)
4574 	struct pt_regs regs = {
4575 		.cs = 3, /* Fake ring 3 no matter what the guest ran on */
4576 		.flags = X86_EFLAGS_IF,
4577 	};
4578 
4579 	do_machine_check(&regs, 0);
4580 #endif
4581 }
4582 
4583 static int handle_machine_check(struct kvm_vcpu *vcpu)
4584 {
4585 	/* handled by vmx_vcpu_run() */
4586 	return 1;
4587 }
4588 
4589 static int handle_exception_nmi(struct kvm_vcpu *vcpu)
4590 {
4591 	struct vcpu_vmx *vmx = to_vmx(vcpu);
4592 	struct kvm_run *kvm_run = vcpu->run;
4593 	u32 intr_info, ex_no, error_code;
4594 	unsigned long cr2, rip, dr6;
4595 	u32 vect_info;
4596 
4597 	vect_info = vmx->idt_vectoring_info;
4598 	intr_info = vmx->exit_intr_info;
4599 
4600 	if (is_machine_check(intr_info) || is_nmi(intr_info))
4601 		return 1; /* handled by handle_exception_nmi_irqoff() */
4602 
4603 	if (is_invalid_opcode(intr_info))
4604 		return handle_ud(vcpu);
4605 
4606 	error_code = 0;
4607 	if (intr_info & INTR_INFO_DELIVER_CODE_MASK)
4608 		error_code = vmcs_read32(VM_EXIT_INTR_ERROR_CODE);
4609 
4610 	if (!vmx->rmode.vm86_active && is_gp_fault(intr_info)) {
4611 		WARN_ON_ONCE(!enable_vmware_backdoor);
4612 
4613 		/*
4614 		 * VMware backdoor emulation on #GP interception only handles
4615 		 * IN{S}, OUT{S}, and RDPMC, none of which generate a non-zero
4616 		 * error code on #GP.
4617 		 */
4618 		if (error_code) {
4619 			kvm_queue_exception_e(vcpu, GP_VECTOR, error_code);
4620 			return 1;
4621 		}
4622 		return kvm_emulate_instruction(vcpu, EMULTYPE_VMWARE_GP);
4623 	}
4624 
4625 	/*
4626 	 * The #PF with PFEC.RSVD = 1 indicates the guest is accessing
4627 	 * MMIO, it is better to report an internal error.
4628 	 * See the comments in vmx_handle_exit.
4629 	 */
4630 	if ((vect_info & VECTORING_INFO_VALID_MASK) &&
4631 	    !(is_page_fault(intr_info) && !(error_code & PFERR_RSVD_MASK))) {
4632 		vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
4633 		vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_SIMUL_EX;
4634 		vcpu->run->internal.ndata = 3;
4635 		vcpu->run->internal.data[0] = vect_info;
4636 		vcpu->run->internal.data[1] = intr_info;
4637 		vcpu->run->internal.data[2] = error_code;
4638 		return 0;
4639 	}
4640 
4641 	if (is_page_fault(intr_info)) {
4642 		cr2 = vmcs_readl(EXIT_QUALIFICATION);
4643 		/* EPT won't cause page fault directly */
4644 		WARN_ON_ONCE(!vcpu->arch.apf.host_apf_reason && enable_ept);
4645 		return kvm_handle_page_fault(vcpu, error_code, cr2, NULL, 0);
4646 	}
4647 
4648 	ex_no = intr_info & INTR_INFO_VECTOR_MASK;
4649 
4650 	if (vmx->rmode.vm86_active && rmode_exception(vcpu, ex_no))
4651 		return handle_rmode_exception(vcpu, ex_no, error_code);
4652 
4653 	switch (ex_no) {
4654 	case AC_VECTOR:
4655 		kvm_queue_exception_e(vcpu, AC_VECTOR, error_code);
4656 		return 1;
4657 	case DB_VECTOR:
4658 		dr6 = vmcs_readl(EXIT_QUALIFICATION);
4659 		if (!(vcpu->guest_debug &
4660 		      (KVM_GUESTDBG_SINGLESTEP | KVM_GUESTDBG_USE_HW_BP))) {
4661 			vcpu->arch.dr6 &= ~DR_TRAP_BITS;
4662 			vcpu->arch.dr6 |= dr6 | DR6_RTM;
4663 			if (is_icebp(intr_info))
4664 				WARN_ON(!skip_emulated_instruction(vcpu));
4665 
4666 			kvm_queue_exception(vcpu, DB_VECTOR);
4667 			return 1;
4668 		}
4669 		kvm_run->debug.arch.dr6 = dr6 | DR6_FIXED_1;
4670 		kvm_run->debug.arch.dr7 = vmcs_readl(GUEST_DR7);
4671 		/* fall through */
4672 	case BP_VECTOR:
4673 		/*
4674 		 * Update instruction length as we may reinject #BP from
4675 		 * user space while in guest debugging mode. Reading it for
4676 		 * #DB as well causes no harm, it is not used in that case.
4677 		 */
4678 		vmx->vcpu.arch.event_exit_inst_len =
4679 			vmcs_read32(VM_EXIT_INSTRUCTION_LEN);
4680 		kvm_run->exit_reason = KVM_EXIT_DEBUG;
4681 		rip = kvm_rip_read(vcpu);
4682 		kvm_run->debug.arch.pc = vmcs_readl(GUEST_CS_BASE) + rip;
4683 		kvm_run->debug.arch.exception = ex_no;
4684 		break;
4685 	default:
4686 		kvm_run->exit_reason = KVM_EXIT_EXCEPTION;
4687 		kvm_run->ex.exception = ex_no;
4688 		kvm_run->ex.error_code = error_code;
4689 		break;
4690 	}
4691 	return 0;
4692 }
4693 
4694 static __always_inline int handle_external_interrupt(struct kvm_vcpu *vcpu)
4695 {
4696 	++vcpu->stat.irq_exits;
4697 	return 1;
4698 }
4699 
4700 static int handle_triple_fault(struct kvm_vcpu *vcpu)
4701 {
4702 	vcpu->run->exit_reason = KVM_EXIT_SHUTDOWN;
4703 	vcpu->mmio_needed = 0;
4704 	return 0;
4705 }
4706 
4707 static int handle_io(struct kvm_vcpu *vcpu)
4708 {
4709 	unsigned long exit_qualification;
4710 	int size, in, string;
4711 	unsigned port;
4712 
4713 	exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
4714 	string = (exit_qualification & 16) != 0;
4715 
4716 	++vcpu->stat.io_exits;
4717 
4718 	if (string)
4719 		return kvm_emulate_instruction(vcpu, 0);
4720 
4721 	port = exit_qualification >> 16;
4722 	size = (exit_qualification & 7) + 1;
4723 	in = (exit_qualification & 8) != 0;
4724 
4725 	return kvm_fast_pio(vcpu, size, port, in);
4726 }
4727 
4728 static void
4729 vmx_patch_hypercall(struct kvm_vcpu *vcpu, unsigned char *hypercall)
4730 {
4731 	/*
4732 	 * Patch in the VMCALL instruction:
4733 	 */
4734 	hypercall[0] = 0x0f;
4735 	hypercall[1] = 0x01;
4736 	hypercall[2] = 0xc1;
4737 }
4738 
4739 /* called to set cr0 as appropriate for a mov-to-cr0 exit. */
4740 static int handle_set_cr0(struct kvm_vcpu *vcpu, unsigned long val)
4741 {
4742 	if (is_guest_mode(vcpu)) {
4743 		struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
4744 		unsigned long orig_val = val;
4745 
4746 		/*
4747 		 * We get here when L2 changed cr0 in a way that did not change
4748 		 * any of L1's shadowed bits (see nested_vmx_exit_handled_cr),
4749 		 * but did change L0 shadowed bits. So we first calculate the
4750 		 * effective cr0 value that L1 would like to write into the
4751 		 * hardware. It consists of the L2-owned bits from the new
4752 		 * value combined with the L1-owned bits from L1's guest_cr0.
4753 		 */
4754 		val = (val & ~vmcs12->cr0_guest_host_mask) |
4755 			(vmcs12->guest_cr0 & vmcs12->cr0_guest_host_mask);
4756 
4757 		if (!nested_guest_cr0_valid(vcpu, val))
4758 			return 1;
4759 
4760 		if (kvm_set_cr0(vcpu, val))
4761 			return 1;
4762 		vmcs_writel(CR0_READ_SHADOW, orig_val);
4763 		return 0;
4764 	} else {
4765 		if (to_vmx(vcpu)->nested.vmxon &&
4766 		    !nested_host_cr0_valid(vcpu, val))
4767 			return 1;
4768 
4769 		return kvm_set_cr0(vcpu, val);
4770 	}
4771 }
4772 
4773 static int handle_set_cr4(struct kvm_vcpu *vcpu, unsigned long val)
4774 {
4775 	if (is_guest_mode(vcpu)) {
4776 		struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
4777 		unsigned long orig_val = val;
4778 
4779 		/* analogously to handle_set_cr0 */
4780 		val = (val & ~vmcs12->cr4_guest_host_mask) |
4781 			(vmcs12->guest_cr4 & vmcs12->cr4_guest_host_mask);
4782 		if (kvm_set_cr4(vcpu, val))
4783 			return 1;
4784 		vmcs_writel(CR4_READ_SHADOW, orig_val);
4785 		return 0;
4786 	} else
4787 		return kvm_set_cr4(vcpu, val);
4788 }
4789 
4790 static int handle_desc(struct kvm_vcpu *vcpu)
4791 {
4792 	WARN_ON(!(vcpu->arch.cr4 & X86_CR4_UMIP));
4793 	return kvm_emulate_instruction(vcpu, 0);
4794 }
4795 
4796 static int handle_cr(struct kvm_vcpu *vcpu)
4797 {
4798 	unsigned long exit_qualification, val;
4799 	int cr;
4800 	int reg;
4801 	int err;
4802 	int ret;
4803 
4804 	exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
4805 	cr = exit_qualification & 15;
4806 	reg = (exit_qualification >> 8) & 15;
4807 	switch ((exit_qualification >> 4) & 3) {
4808 	case 0: /* mov to cr */
4809 		val = kvm_register_readl(vcpu, reg);
4810 		trace_kvm_cr_write(cr, val);
4811 		switch (cr) {
4812 		case 0:
4813 			err = handle_set_cr0(vcpu, val);
4814 			return kvm_complete_insn_gp(vcpu, err);
4815 		case 3:
4816 			WARN_ON_ONCE(enable_unrestricted_guest);
4817 			err = kvm_set_cr3(vcpu, val);
4818 			return kvm_complete_insn_gp(vcpu, err);
4819 		case 4:
4820 			err = handle_set_cr4(vcpu, val);
4821 			return kvm_complete_insn_gp(vcpu, err);
4822 		case 8: {
4823 				u8 cr8_prev = kvm_get_cr8(vcpu);
4824 				u8 cr8 = (u8)val;
4825 				err = kvm_set_cr8(vcpu, cr8);
4826 				ret = kvm_complete_insn_gp(vcpu, err);
4827 				if (lapic_in_kernel(vcpu))
4828 					return ret;
4829 				if (cr8_prev <= cr8)
4830 					return ret;
4831 				/*
4832 				 * TODO: we might be squashing a
4833 				 * KVM_GUESTDBG_SINGLESTEP-triggered
4834 				 * KVM_EXIT_DEBUG here.
4835 				 */
4836 				vcpu->run->exit_reason = KVM_EXIT_SET_TPR;
4837 				return 0;
4838 			}
4839 		}
4840 		break;
4841 	case 2: /* clts */
4842 		WARN_ONCE(1, "Guest should always own CR0.TS");
4843 		vmx_set_cr0(vcpu, kvm_read_cr0_bits(vcpu, ~X86_CR0_TS));
4844 		trace_kvm_cr_write(0, kvm_read_cr0(vcpu));
4845 		return kvm_skip_emulated_instruction(vcpu);
4846 	case 1: /*mov from cr*/
4847 		switch (cr) {
4848 		case 3:
4849 			WARN_ON_ONCE(enable_unrestricted_guest);
4850 			val = kvm_read_cr3(vcpu);
4851 			kvm_register_write(vcpu, reg, val);
4852 			trace_kvm_cr_read(cr, val);
4853 			return kvm_skip_emulated_instruction(vcpu);
4854 		case 8:
4855 			val = kvm_get_cr8(vcpu);
4856 			kvm_register_write(vcpu, reg, val);
4857 			trace_kvm_cr_read(cr, val);
4858 			return kvm_skip_emulated_instruction(vcpu);
4859 		}
4860 		break;
4861 	case 3: /* lmsw */
4862 		val = (exit_qualification >> LMSW_SOURCE_DATA_SHIFT) & 0x0f;
4863 		trace_kvm_cr_write(0, (kvm_read_cr0(vcpu) & ~0xful) | val);
4864 		kvm_lmsw(vcpu, val);
4865 
4866 		return kvm_skip_emulated_instruction(vcpu);
4867 	default:
4868 		break;
4869 	}
4870 	vcpu->run->exit_reason = 0;
4871 	vcpu_unimpl(vcpu, "unhandled control register: op %d cr %d\n",
4872 	       (int)(exit_qualification >> 4) & 3, cr);
4873 	return 0;
4874 }
4875 
4876 static int handle_dr(struct kvm_vcpu *vcpu)
4877 {
4878 	unsigned long exit_qualification;
4879 	int dr, dr7, reg;
4880 
4881 	exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
4882 	dr = exit_qualification & DEBUG_REG_ACCESS_NUM;
4883 
4884 	/* First, if DR does not exist, trigger UD */
4885 	if (!kvm_require_dr(vcpu, dr))
4886 		return 1;
4887 
4888 	/* Do not handle if the CPL > 0, will trigger GP on re-entry */
4889 	if (!kvm_require_cpl(vcpu, 0))
4890 		return 1;
4891 	dr7 = vmcs_readl(GUEST_DR7);
4892 	if (dr7 & DR7_GD) {
4893 		/*
4894 		 * As the vm-exit takes precedence over the debug trap, we
4895 		 * need to emulate the latter, either for the host or the
4896 		 * guest debugging itself.
4897 		 */
4898 		if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP) {
4899 			vcpu->run->debug.arch.dr6 = vcpu->arch.dr6;
4900 			vcpu->run->debug.arch.dr7 = dr7;
4901 			vcpu->run->debug.arch.pc = kvm_get_linear_rip(vcpu);
4902 			vcpu->run->debug.arch.exception = DB_VECTOR;
4903 			vcpu->run->exit_reason = KVM_EXIT_DEBUG;
4904 			return 0;
4905 		} else {
4906 			vcpu->arch.dr6 &= ~DR_TRAP_BITS;
4907 			vcpu->arch.dr6 |= DR6_BD | DR6_RTM;
4908 			kvm_queue_exception(vcpu, DB_VECTOR);
4909 			return 1;
4910 		}
4911 	}
4912 
4913 	if (vcpu->guest_debug == 0) {
4914 		exec_controls_clearbit(to_vmx(vcpu), CPU_BASED_MOV_DR_EXITING);
4915 
4916 		/*
4917 		 * No more DR vmexits; force a reload of the debug registers
4918 		 * and reenter on this instruction.  The next vmexit will
4919 		 * retrieve the full state of the debug registers.
4920 		 */
4921 		vcpu->arch.switch_db_regs |= KVM_DEBUGREG_WONT_EXIT;
4922 		return 1;
4923 	}
4924 
4925 	reg = DEBUG_REG_ACCESS_REG(exit_qualification);
4926 	if (exit_qualification & TYPE_MOV_FROM_DR) {
4927 		unsigned long val;
4928 
4929 		if (kvm_get_dr(vcpu, dr, &val))
4930 			return 1;
4931 		kvm_register_write(vcpu, reg, val);
4932 	} else
4933 		if (kvm_set_dr(vcpu, dr, kvm_register_readl(vcpu, reg)))
4934 			return 1;
4935 
4936 	return kvm_skip_emulated_instruction(vcpu);
4937 }
4938 
4939 static u64 vmx_get_dr6(struct kvm_vcpu *vcpu)
4940 {
4941 	return vcpu->arch.dr6;
4942 }
4943 
4944 static void vmx_set_dr6(struct kvm_vcpu *vcpu, unsigned long val)
4945 {
4946 }
4947 
4948 static void vmx_sync_dirty_debug_regs(struct kvm_vcpu *vcpu)
4949 {
4950 	get_debugreg(vcpu->arch.db[0], 0);
4951 	get_debugreg(vcpu->arch.db[1], 1);
4952 	get_debugreg(vcpu->arch.db[2], 2);
4953 	get_debugreg(vcpu->arch.db[3], 3);
4954 	get_debugreg(vcpu->arch.dr6, 6);
4955 	vcpu->arch.dr7 = vmcs_readl(GUEST_DR7);
4956 
4957 	vcpu->arch.switch_db_regs &= ~KVM_DEBUGREG_WONT_EXIT;
4958 	exec_controls_setbit(to_vmx(vcpu), CPU_BASED_MOV_DR_EXITING);
4959 }
4960 
4961 static void vmx_set_dr7(struct kvm_vcpu *vcpu, unsigned long val)
4962 {
4963 	vmcs_writel(GUEST_DR7, val);
4964 }
4965 
4966 static int handle_tpr_below_threshold(struct kvm_vcpu *vcpu)
4967 {
4968 	kvm_apic_update_ppr(vcpu);
4969 	return 1;
4970 }
4971 
4972 static int handle_interrupt_window(struct kvm_vcpu *vcpu)
4973 {
4974 	exec_controls_clearbit(to_vmx(vcpu), CPU_BASED_VIRTUAL_INTR_PENDING);
4975 
4976 	kvm_make_request(KVM_REQ_EVENT, vcpu);
4977 
4978 	++vcpu->stat.irq_window_exits;
4979 	return 1;
4980 }
4981 
4982 static int handle_vmcall(struct kvm_vcpu *vcpu)
4983 {
4984 	return kvm_emulate_hypercall(vcpu);
4985 }
4986 
4987 static int handle_invd(struct kvm_vcpu *vcpu)
4988 {
4989 	return kvm_emulate_instruction(vcpu, 0);
4990 }
4991 
4992 static int handle_invlpg(struct kvm_vcpu *vcpu)
4993 {
4994 	unsigned long exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
4995 
4996 	kvm_mmu_invlpg(vcpu, exit_qualification);
4997 	return kvm_skip_emulated_instruction(vcpu);
4998 }
4999 
5000 static int handle_rdpmc(struct kvm_vcpu *vcpu)
5001 {
5002 	int err;
5003 
5004 	err = kvm_rdpmc(vcpu);
5005 	return kvm_complete_insn_gp(vcpu, err);
5006 }
5007 
5008 static int handle_wbinvd(struct kvm_vcpu *vcpu)
5009 {
5010 	return kvm_emulate_wbinvd(vcpu);
5011 }
5012 
5013 static int handle_xsetbv(struct kvm_vcpu *vcpu)
5014 {
5015 	u64 new_bv = kvm_read_edx_eax(vcpu);
5016 	u32 index = kvm_rcx_read(vcpu);
5017 
5018 	if (kvm_set_xcr(vcpu, index, new_bv) == 0)
5019 		return kvm_skip_emulated_instruction(vcpu);
5020 	return 1;
5021 }
5022 
5023 static int handle_apic_access(struct kvm_vcpu *vcpu)
5024 {
5025 	if (likely(fasteoi)) {
5026 		unsigned long exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
5027 		int access_type, offset;
5028 
5029 		access_type = exit_qualification & APIC_ACCESS_TYPE;
5030 		offset = exit_qualification & APIC_ACCESS_OFFSET;
5031 		/*
5032 		 * Sane guest uses MOV to write EOI, with written value
5033 		 * not cared. So make a short-circuit here by avoiding
5034 		 * heavy instruction emulation.
5035 		 */
5036 		if ((access_type == TYPE_LINEAR_APIC_INST_WRITE) &&
5037 		    (offset == APIC_EOI)) {
5038 			kvm_lapic_set_eoi(vcpu);
5039 			return kvm_skip_emulated_instruction(vcpu);
5040 		}
5041 	}
5042 	return kvm_emulate_instruction(vcpu, 0);
5043 }
5044 
5045 static int handle_apic_eoi_induced(struct kvm_vcpu *vcpu)
5046 {
5047 	unsigned long exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
5048 	int vector = exit_qualification & 0xff;
5049 
5050 	/* EOI-induced VM exit is trap-like and thus no need to adjust IP */
5051 	kvm_apic_set_eoi_accelerated(vcpu, vector);
5052 	return 1;
5053 }
5054 
5055 static int handle_apic_write(struct kvm_vcpu *vcpu)
5056 {
5057 	unsigned long exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
5058 	u32 offset = exit_qualification & 0xfff;
5059 
5060 	/* APIC-write VM exit is trap-like and thus no need to adjust IP */
5061 	kvm_apic_write_nodecode(vcpu, offset);
5062 	return 1;
5063 }
5064 
5065 static int handle_task_switch(struct kvm_vcpu *vcpu)
5066 {
5067 	struct vcpu_vmx *vmx = to_vmx(vcpu);
5068 	unsigned long exit_qualification;
5069 	bool has_error_code = false;
5070 	u32 error_code = 0;
5071 	u16 tss_selector;
5072 	int reason, type, idt_v, idt_index;
5073 
5074 	idt_v = (vmx->idt_vectoring_info & VECTORING_INFO_VALID_MASK);
5075 	idt_index = (vmx->idt_vectoring_info & VECTORING_INFO_VECTOR_MASK);
5076 	type = (vmx->idt_vectoring_info & VECTORING_INFO_TYPE_MASK);
5077 
5078 	exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
5079 
5080 	reason = (u32)exit_qualification >> 30;
5081 	if (reason == TASK_SWITCH_GATE && idt_v) {
5082 		switch (type) {
5083 		case INTR_TYPE_NMI_INTR:
5084 			vcpu->arch.nmi_injected = false;
5085 			vmx_set_nmi_mask(vcpu, true);
5086 			break;
5087 		case INTR_TYPE_EXT_INTR:
5088 		case INTR_TYPE_SOFT_INTR:
5089 			kvm_clear_interrupt_queue(vcpu);
5090 			break;
5091 		case INTR_TYPE_HARD_EXCEPTION:
5092 			if (vmx->idt_vectoring_info &
5093 			    VECTORING_INFO_DELIVER_CODE_MASK) {
5094 				has_error_code = true;
5095 				error_code =
5096 					vmcs_read32(IDT_VECTORING_ERROR_CODE);
5097 			}
5098 			/* fall through */
5099 		case INTR_TYPE_SOFT_EXCEPTION:
5100 			kvm_clear_exception_queue(vcpu);
5101 			break;
5102 		default:
5103 			break;
5104 		}
5105 	}
5106 	tss_selector = exit_qualification;
5107 
5108 	if (!idt_v || (type != INTR_TYPE_HARD_EXCEPTION &&
5109 		       type != INTR_TYPE_EXT_INTR &&
5110 		       type != INTR_TYPE_NMI_INTR))
5111 		WARN_ON(!skip_emulated_instruction(vcpu));
5112 
5113 	/*
5114 	 * TODO: What about debug traps on tss switch?
5115 	 *       Are we supposed to inject them and update dr6?
5116 	 */
5117 	return kvm_task_switch(vcpu, tss_selector,
5118 			       type == INTR_TYPE_SOFT_INTR ? idt_index : -1,
5119 			       reason, has_error_code, error_code);
5120 }
5121 
5122 static int handle_ept_violation(struct kvm_vcpu *vcpu)
5123 {
5124 	unsigned long exit_qualification;
5125 	gpa_t gpa;
5126 	u64 error_code;
5127 
5128 	exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
5129 
5130 	/*
5131 	 * EPT violation happened while executing iret from NMI,
5132 	 * "blocked by NMI" bit has to be set before next VM entry.
5133 	 * There are errata that may cause this bit to not be set:
5134 	 * AAK134, BY25.
5135 	 */
5136 	if (!(to_vmx(vcpu)->idt_vectoring_info & VECTORING_INFO_VALID_MASK) &&
5137 			enable_vnmi &&
5138 			(exit_qualification & INTR_INFO_UNBLOCK_NMI))
5139 		vmcs_set_bits(GUEST_INTERRUPTIBILITY_INFO, GUEST_INTR_STATE_NMI);
5140 
5141 	gpa = vmcs_read64(GUEST_PHYSICAL_ADDRESS);
5142 	trace_kvm_page_fault(gpa, exit_qualification);
5143 
5144 	/* Is it a read fault? */
5145 	error_code = (exit_qualification & EPT_VIOLATION_ACC_READ)
5146 		     ? PFERR_USER_MASK : 0;
5147 	/* Is it a write fault? */
5148 	error_code |= (exit_qualification & EPT_VIOLATION_ACC_WRITE)
5149 		      ? PFERR_WRITE_MASK : 0;
5150 	/* Is it a fetch fault? */
5151 	error_code |= (exit_qualification & EPT_VIOLATION_ACC_INSTR)
5152 		      ? PFERR_FETCH_MASK : 0;
5153 	/* ept page table entry is present? */
5154 	error_code |= (exit_qualification &
5155 		       (EPT_VIOLATION_READABLE | EPT_VIOLATION_WRITABLE |
5156 			EPT_VIOLATION_EXECUTABLE))
5157 		      ? PFERR_PRESENT_MASK : 0;
5158 
5159 	error_code |= (exit_qualification & 0x100) != 0 ?
5160 	       PFERR_GUEST_FINAL_MASK : PFERR_GUEST_PAGE_MASK;
5161 
5162 	vcpu->arch.exit_qualification = exit_qualification;
5163 	return kvm_mmu_page_fault(vcpu, gpa, error_code, NULL, 0);
5164 }
5165 
5166 static int handle_ept_misconfig(struct kvm_vcpu *vcpu)
5167 {
5168 	gpa_t gpa;
5169 
5170 	/*
5171 	 * A nested guest cannot optimize MMIO vmexits, because we have an
5172 	 * nGPA here instead of the required GPA.
5173 	 */
5174 	gpa = vmcs_read64(GUEST_PHYSICAL_ADDRESS);
5175 	if (!is_guest_mode(vcpu) &&
5176 	    !kvm_io_bus_write(vcpu, KVM_FAST_MMIO_BUS, gpa, 0, NULL)) {
5177 		trace_kvm_fast_mmio(gpa);
5178 		return kvm_skip_emulated_instruction(vcpu);
5179 	}
5180 
5181 	return kvm_mmu_page_fault(vcpu, gpa, PFERR_RSVD_MASK, NULL, 0);
5182 }
5183 
5184 static int handle_nmi_window(struct kvm_vcpu *vcpu)
5185 {
5186 	WARN_ON_ONCE(!enable_vnmi);
5187 	exec_controls_clearbit(to_vmx(vcpu), CPU_BASED_VIRTUAL_NMI_PENDING);
5188 	++vcpu->stat.nmi_window_exits;
5189 	kvm_make_request(KVM_REQ_EVENT, vcpu);
5190 
5191 	return 1;
5192 }
5193 
5194 static int handle_invalid_guest_state(struct kvm_vcpu *vcpu)
5195 {
5196 	struct vcpu_vmx *vmx = to_vmx(vcpu);
5197 	bool intr_window_requested;
5198 	unsigned count = 130;
5199 
5200 	/*
5201 	 * We should never reach the point where we are emulating L2
5202 	 * due to invalid guest state as that means we incorrectly
5203 	 * allowed a nested VMEntry with an invalid vmcs12.
5204 	 */
5205 	WARN_ON_ONCE(vmx->emulation_required && vmx->nested.nested_run_pending);
5206 
5207 	intr_window_requested = exec_controls_get(vmx) &
5208 				CPU_BASED_VIRTUAL_INTR_PENDING;
5209 
5210 	while (vmx->emulation_required && count-- != 0) {
5211 		if (intr_window_requested && vmx_interrupt_allowed(vcpu))
5212 			return handle_interrupt_window(&vmx->vcpu);
5213 
5214 		if (kvm_test_request(KVM_REQ_EVENT, vcpu))
5215 			return 1;
5216 
5217 		if (!kvm_emulate_instruction(vcpu, 0))
5218 			return 0;
5219 
5220 		if (vmx->emulation_required && !vmx->rmode.vm86_active &&
5221 		    vcpu->arch.exception.pending) {
5222 			vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
5223 			vcpu->run->internal.suberror =
5224 						KVM_INTERNAL_ERROR_EMULATION;
5225 			vcpu->run->internal.ndata = 0;
5226 			return 0;
5227 		}
5228 
5229 		if (vcpu->arch.halt_request) {
5230 			vcpu->arch.halt_request = 0;
5231 			return kvm_vcpu_halt(vcpu);
5232 		}
5233 
5234 		/*
5235 		 * Note, return 1 and not 0, vcpu_run() is responsible for
5236 		 * morphing the pending signal into the proper return code.
5237 		 */
5238 		if (signal_pending(current))
5239 			return 1;
5240 
5241 		if (need_resched())
5242 			schedule();
5243 	}
5244 
5245 	return 1;
5246 }
5247 
5248 static void grow_ple_window(struct kvm_vcpu *vcpu)
5249 {
5250 	struct vcpu_vmx *vmx = to_vmx(vcpu);
5251 	unsigned int old = vmx->ple_window;
5252 
5253 	vmx->ple_window = __grow_ple_window(old, ple_window,
5254 					    ple_window_grow,
5255 					    ple_window_max);
5256 
5257 	if (vmx->ple_window != old) {
5258 		vmx->ple_window_dirty = true;
5259 		trace_kvm_ple_window_update(vcpu->vcpu_id,
5260 					    vmx->ple_window, old);
5261 	}
5262 }
5263 
5264 static void shrink_ple_window(struct kvm_vcpu *vcpu)
5265 {
5266 	struct vcpu_vmx *vmx = to_vmx(vcpu);
5267 	unsigned int old = vmx->ple_window;
5268 
5269 	vmx->ple_window = __shrink_ple_window(old, ple_window,
5270 					      ple_window_shrink,
5271 					      ple_window);
5272 
5273 	if (vmx->ple_window != old) {
5274 		vmx->ple_window_dirty = true;
5275 		trace_kvm_ple_window_update(vcpu->vcpu_id,
5276 					    vmx->ple_window, old);
5277 	}
5278 }
5279 
5280 /*
5281  * Handler for POSTED_INTERRUPT_WAKEUP_VECTOR.
5282  */
5283 static void wakeup_handler(void)
5284 {
5285 	struct kvm_vcpu *vcpu;
5286 	int cpu = smp_processor_id();
5287 
5288 	spin_lock(&per_cpu(blocked_vcpu_on_cpu_lock, cpu));
5289 	list_for_each_entry(vcpu, &per_cpu(blocked_vcpu_on_cpu, cpu),
5290 			blocked_vcpu_list) {
5291 		struct pi_desc *pi_desc = vcpu_to_pi_desc(vcpu);
5292 
5293 		if (pi_test_on(pi_desc) == 1)
5294 			kvm_vcpu_kick(vcpu);
5295 	}
5296 	spin_unlock(&per_cpu(blocked_vcpu_on_cpu_lock, cpu));
5297 }
5298 
5299 static void vmx_enable_tdp(void)
5300 {
5301 	kvm_mmu_set_mask_ptes(VMX_EPT_READABLE_MASK,
5302 		enable_ept_ad_bits ? VMX_EPT_ACCESS_BIT : 0ull,
5303 		enable_ept_ad_bits ? VMX_EPT_DIRTY_BIT : 0ull,
5304 		0ull, VMX_EPT_EXECUTABLE_MASK,
5305 		cpu_has_vmx_ept_execute_only() ? 0ull : VMX_EPT_READABLE_MASK,
5306 		VMX_EPT_RWX_MASK, 0ull);
5307 
5308 	ept_set_mmio_spte_mask();
5309 	kvm_enable_tdp();
5310 }
5311 
5312 /*
5313  * Indicate a busy-waiting vcpu in spinlock. We do not enable the PAUSE
5314  * exiting, so only get here on cpu with PAUSE-Loop-Exiting.
5315  */
5316 static int handle_pause(struct kvm_vcpu *vcpu)
5317 {
5318 	if (!kvm_pause_in_guest(vcpu->kvm))
5319 		grow_ple_window(vcpu);
5320 
5321 	/*
5322 	 * Intel sdm vol3 ch-25.1.3 says: The "PAUSE-loop exiting"
5323 	 * VM-execution control is ignored if CPL > 0. OTOH, KVM
5324 	 * never set PAUSE_EXITING and just set PLE if supported,
5325 	 * so the vcpu must be CPL=0 if it gets a PAUSE exit.
5326 	 */
5327 	kvm_vcpu_on_spin(vcpu, true);
5328 	return kvm_skip_emulated_instruction(vcpu);
5329 }
5330 
5331 static int handle_nop(struct kvm_vcpu *vcpu)
5332 {
5333 	return kvm_skip_emulated_instruction(vcpu);
5334 }
5335 
5336 static int handle_mwait(struct kvm_vcpu *vcpu)
5337 {
5338 	printk_once(KERN_WARNING "kvm: MWAIT instruction emulated as NOP!\n");
5339 	return handle_nop(vcpu);
5340 }
5341 
5342 static int handle_invalid_op(struct kvm_vcpu *vcpu)
5343 {
5344 	kvm_queue_exception(vcpu, UD_VECTOR);
5345 	return 1;
5346 }
5347 
5348 static int handle_monitor_trap(struct kvm_vcpu *vcpu)
5349 {
5350 	return 1;
5351 }
5352 
5353 static int handle_monitor(struct kvm_vcpu *vcpu)
5354 {
5355 	printk_once(KERN_WARNING "kvm: MONITOR instruction emulated as NOP!\n");
5356 	return handle_nop(vcpu);
5357 }
5358 
5359 static int handle_invpcid(struct kvm_vcpu *vcpu)
5360 {
5361 	u32 vmx_instruction_info;
5362 	unsigned long type;
5363 	bool pcid_enabled;
5364 	gva_t gva;
5365 	struct x86_exception e;
5366 	unsigned i;
5367 	unsigned long roots_to_free = 0;
5368 	struct {
5369 		u64 pcid;
5370 		u64 gla;
5371 	} operand;
5372 
5373 	if (!guest_cpuid_has(vcpu, X86_FEATURE_INVPCID)) {
5374 		kvm_queue_exception(vcpu, UD_VECTOR);
5375 		return 1;
5376 	}
5377 
5378 	vmx_instruction_info = vmcs_read32(VMX_INSTRUCTION_INFO);
5379 	type = kvm_register_readl(vcpu, (vmx_instruction_info >> 28) & 0xf);
5380 
5381 	if (type > 3) {
5382 		kvm_inject_gp(vcpu, 0);
5383 		return 1;
5384 	}
5385 
5386 	/* According to the Intel instruction reference, the memory operand
5387 	 * is read even if it isn't needed (e.g., for type==all)
5388 	 */
5389 	if (get_vmx_mem_address(vcpu, vmcs_readl(EXIT_QUALIFICATION),
5390 				vmx_instruction_info, false,
5391 				sizeof(operand), &gva))
5392 		return 1;
5393 
5394 	if (kvm_read_guest_virt(vcpu, gva, &operand, sizeof(operand), &e)) {
5395 		kvm_inject_page_fault(vcpu, &e);
5396 		return 1;
5397 	}
5398 
5399 	if (operand.pcid >> 12 != 0) {
5400 		kvm_inject_gp(vcpu, 0);
5401 		return 1;
5402 	}
5403 
5404 	pcid_enabled = kvm_read_cr4_bits(vcpu, X86_CR4_PCIDE);
5405 
5406 	switch (type) {
5407 	case INVPCID_TYPE_INDIV_ADDR:
5408 		if ((!pcid_enabled && (operand.pcid != 0)) ||
5409 		    is_noncanonical_address(operand.gla, vcpu)) {
5410 			kvm_inject_gp(vcpu, 0);
5411 			return 1;
5412 		}
5413 		kvm_mmu_invpcid_gva(vcpu, operand.gla, operand.pcid);
5414 		return kvm_skip_emulated_instruction(vcpu);
5415 
5416 	case INVPCID_TYPE_SINGLE_CTXT:
5417 		if (!pcid_enabled && (operand.pcid != 0)) {
5418 			kvm_inject_gp(vcpu, 0);
5419 			return 1;
5420 		}
5421 
5422 		if (kvm_get_active_pcid(vcpu) == operand.pcid) {
5423 			kvm_mmu_sync_roots(vcpu);
5424 			kvm_make_request(KVM_REQ_TLB_FLUSH, vcpu);
5425 		}
5426 
5427 		for (i = 0; i < KVM_MMU_NUM_PREV_ROOTS; i++)
5428 			if (kvm_get_pcid(vcpu, vcpu->arch.mmu->prev_roots[i].cr3)
5429 			    == operand.pcid)
5430 				roots_to_free |= KVM_MMU_ROOT_PREVIOUS(i);
5431 
5432 		kvm_mmu_free_roots(vcpu, vcpu->arch.mmu, roots_to_free);
5433 		/*
5434 		 * If neither the current cr3 nor any of the prev_roots use the
5435 		 * given PCID, then nothing needs to be done here because a
5436 		 * resync will happen anyway before switching to any other CR3.
5437 		 */
5438 
5439 		return kvm_skip_emulated_instruction(vcpu);
5440 
5441 	case INVPCID_TYPE_ALL_NON_GLOBAL:
5442 		/*
5443 		 * Currently, KVM doesn't mark global entries in the shadow
5444 		 * page tables, so a non-global flush just degenerates to a
5445 		 * global flush. If needed, we could optimize this later by
5446 		 * keeping track of global entries in shadow page tables.
5447 		 */
5448 
5449 		/* fall-through */
5450 	case INVPCID_TYPE_ALL_INCL_GLOBAL:
5451 		kvm_mmu_unload(vcpu);
5452 		return kvm_skip_emulated_instruction(vcpu);
5453 
5454 	default:
5455 		BUG(); /* We have already checked above that type <= 3 */
5456 	}
5457 }
5458 
5459 static int handle_pml_full(struct kvm_vcpu *vcpu)
5460 {
5461 	unsigned long exit_qualification;
5462 
5463 	trace_kvm_pml_full(vcpu->vcpu_id);
5464 
5465 	exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
5466 
5467 	/*
5468 	 * PML buffer FULL happened while executing iret from NMI,
5469 	 * "blocked by NMI" bit has to be set before next VM entry.
5470 	 */
5471 	if (!(to_vmx(vcpu)->idt_vectoring_info & VECTORING_INFO_VALID_MASK) &&
5472 			enable_vnmi &&
5473 			(exit_qualification & INTR_INFO_UNBLOCK_NMI))
5474 		vmcs_set_bits(GUEST_INTERRUPTIBILITY_INFO,
5475 				GUEST_INTR_STATE_NMI);
5476 
5477 	/*
5478 	 * PML buffer already flushed at beginning of VMEXIT. Nothing to do
5479 	 * here.., and there's no userspace involvement needed for PML.
5480 	 */
5481 	return 1;
5482 }
5483 
5484 static int handle_preemption_timer(struct kvm_vcpu *vcpu)
5485 {
5486 	struct vcpu_vmx *vmx = to_vmx(vcpu);
5487 
5488 	if (!vmx->req_immediate_exit &&
5489 	    !unlikely(vmx->loaded_vmcs->hv_timer_soft_disabled))
5490 		kvm_lapic_expired_hv_timer(vcpu);
5491 
5492 	return 1;
5493 }
5494 
5495 /*
5496  * When nested=0, all VMX instruction VM Exits filter here.  The handlers
5497  * are overwritten by nested_vmx_setup() when nested=1.
5498  */
5499 static int handle_vmx_instruction(struct kvm_vcpu *vcpu)
5500 {
5501 	kvm_queue_exception(vcpu, UD_VECTOR);
5502 	return 1;
5503 }
5504 
5505 static int handle_encls(struct kvm_vcpu *vcpu)
5506 {
5507 	/*
5508 	 * SGX virtualization is not yet supported.  There is no software
5509 	 * enable bit for SGX, so we have to trap ENCLS and inject a #UD
5510 	 * to prevent the guest from executing ENCLS.
5511 	 */
5512 	kvm_queue_exception(vcpu, UD_VECTOR);
5513 	return 1;
5514 }
5515 
5516 /*
5517  * The exit handlers return 1 if the exit was handled fully and guest execution
5518  * may resume.  Otherwise they set the kvm_run parameter to indicate what needs
5519  * to be done to userspace and return 0.
5520  */
5521 static int (*kvm_vmx_exit_handlers[])(struct kvm_vcpu *vcpu) = {
5522 	[EXIT_REASON_EXCEPTION_NMI]           = handle_exception_nmi,
5523 	[EXIT_REASON_EXTERNAL_INTERRUPT]      = handle_external_interrupt,
5524 	[EXIT_REASON_TRIPLE_FAULT]            = handle_triple_fault,
5525 	[EXIT_REASON_NMI_WINDOW]	      = handle_nmi_window,
5526 	[EXIT_REASON_IO_INSTRUCTION]          = handle_io,
5527 	[EXIT_REASON_CR_ACCESS]               = handle_cr,
5528 	[EXIT_REASON_DR_ACCESS]               = handle_dr,
5529 	[EXIT_REASON_CPUID]                   = kvm_emulate_cpuid,
5530 	[EXIT_REASON_MSR_READ]                = kvm_emulate_rdmsr,
5531 	[EXIT_REASON_MSR_WRITE]               = kvm_emulate_wrmsr,
5532 	[EXIT_REASON_PENDING_INTERRUPT]       = handle_interrupt_window,
5533 	[EXIT_REASON_HLT]                     = kvm_emulate_halt,
5534 	[EXIT_REASON_INVD]		      = handle_invd,
5535 	[EXIT_REASON_INVLPG]		      = handle_invlpg,
5536 	[EXIT_REASON_RDPMC]                   = handle_rdpmc,
5537 	[EXIT_REASON_VMCALL]                  = handle_vmcall,
5538 	[EXIT_REASON_VMCLEAR]		      = handle_vmx_instruction,
5539 	[EXIT_REASON_VMLAUNCH]		      = handle_vmx_instruction,
5540 	[EXIT_REASON_VMPTRLD]		      = handle_vmx_instruction,
5541 	[EXIT_REASON_VMPTRST]		      = handle_vmx_instruction,
5542 	[EXIT_REASON_VMREAD]		      = handle_vmx_instruction,
5543 	[EXIT_REASON_VMRESUME]		      = handle_vmx_instruction,
5544 	[EXIT_REASON_VMWRITE]		      = handle_vmx_instruction,
5545 	[EXIT_REASON_VMOFF]		      = handle_vmx_instruction,
5546 	[EXIT_REASON_VMON]		      = handle_vmx_instruction,
5547 	[EXIT_REASON_TPR_BELOW_THRESHOLD]     = handle_tpr_below_threshold,
5548 	[EXIT_REASON_APIC_ACCESS]             = handle_apic_access,
5549 	[EXIT_REASON_APIC_WRITE]              = handle_apic_write,
5550 	[EXIT_REASON_EOI_INDUCED]             = handle_apic_eoi_induced,
5551 	[EXIT_REASON_WBINVD]                  = handle_wbinvd,
5552 	[EXIT_REASON_XSETBV]                  = handle_xsetbv,
5553 	[EXIT_REASON_TASK_SWITCH]             = handle_task_switch,
5554 	[EXIT_REASON_MCE_DURING_VMENTRY]      = handle_machine_check,
5555 	[EXIT_REASON_GDTR_IDTR]		      = handle_desc,
5556 	[EXIT_REASON_LDTR_TR]		      = handle_desc,
5557 	[EXIT_REASON_EPT_VIOLATION]	      = handle_ept_violation,
5558 	[EXIT_REASON_EPT_MISCONFIG]           = handle_ept_misconfig,
5559 	[EXIT_REASON_PAUSE_INSTRUCTION]       = handle_pause,
5560 	[EXIT_REASON_MWAIT_INSTRUCTION]	      = handle_mwait,
5561 	[EXIT_REASON_MONITOR_TRAP_FLAG]       = handle_monitor_trap,
5562 	[EXIT_REASON_MONITOR_INSTRUCTION]     = handle_monitor,
5563 	[EXIT_REASON_INVEPT]                  = handle_vmx_instruction,
5564 	[EXIT_REASON_INVVPID]                 = handle_vmx_instruction,
5565 	[EXIT_REASON_RDRAND]                  = handle_invalid_op,
5566 	[EXIT_REASON_RDSEED]                  = handle_invalid_op,
5567 	[EXIT_REASON_PML_FULL]		      = handle_pml_full,
5568 	[EXIT_REASON_INVPCID]                 = handle_invpcid,
5569 	[EXIT_REASON_VMFUNC]		      = handle_vmx_instruction,
5570 	[EXIT_REASON_PREEMPTION_TIMER]	      = handle_preemption_timer,
5571 	[EXIT_REASON_ENCLS]		      = handle_encls,
5572 };
5573 
5574 static const int kvm_vmx_max_exit_handlers =
5575 	ARRAY_SIZE(kvm_vmx_exit_handlers);
5576 
5577 static void vmx_get_exit_info(struct kvm_vcpu *vcpu, u64 *info1, u64 *info2)
5578 {
5579 	*info1 = vmcs_readl(EXIT_QUALIFICATION);
5580 	*info2 = vmcs_read32(VM_EXIT_INTR_INFO);
5581 }
5582 
5583 static void vmx_destroy_pml_buffer(struct vcpu_vmx *vmx)
5584 {
5585 	if (vmx->pml_pg) {
5586 		__free_page(vmx->pml_pg);
5587 		vmx->pml_pg = NULL;
5588 	}
5589 }
5590 
5591 static void vmx_flush_pml_buffer(struct kvm_vcpu *vcpu)
5592 {
5593 	struct vcpu_vmx *vmx = to_vmx(vcpu);
5594 	u64 *pml_buf;
5595 	u16 pml_idx;
5596 
5597 	pml_idx = vmcs_read16(GUEST_PML_INDEX);
5598 
5599 	/* Do nothing if PML buffer is empty */
5600 	if (pml_idx == (PML_ENTITY_NUM - 1))
5601 		return;
5602 
5603 	/* PML index always points to next available PML buffer entity */
5604 	if (pml_idx >= PML_ENTITY_NUM)
5605 		pml_idx = 0;
5606 	else
5607 		pml_idx++;
5608 
5609 	pml_buf = page_address(vmx->pml_pg);
5610 	for (; pml_idx < PML_ENTITY_NUM; pml_idx++) {
5611 		u64 gpa;
5612 
5613 		gpa = pml_buf[pml_idx];
5614 		WARN_ON(gpa & (PAGE_SIZE - 1));
5615 		kvm_vcpu_mark_page_dirty(vcpu, gpa >> PAGE_SHIFT);
5616 	}
5617 
5618 	/* reset PML index */
5619 	vmcs_write16(GUEST_PML_INDEX, PML_ENTITY_NUM - 1);
5620 }
5621 
5622 /*
5623  * Flush all vcpus' PML buffer and update logged GPAs to dirty_bitmap.
5624  * Called before reporting dirty_bitmap to userspace.
5625  */
5626 static void kvm_flush_pml_buffers(struct kvm *kvm)
5627 {
5628 	int i;
5629 	struct kvm_vcpu *vcpu;
5630 	/*
5631 	 * We only need to kick vcpu out of guest mode here, as PML buffer
5632 	 * is flushed at beginning of all VMEXITs, and it's obvious that only
5633 	 * vcpus running in guest are possible to have unflushed GPAs in PML
5634 	 * buffer.
5635 	 */
5636 	kvm_for_each_vcpu(i, vcpu, kvm)
5637 		kvm_vcpu_kick(vcpu);
5638 }
5639 
5640 static void vmx_dump_sel(char *name, uint32_t sel)
5641 {
5642 	pr_err("%s sel=0x%04x, attr=0x%05x, limit=0x%08x, base=0x%016lx\n",
5643 	       name, vmcs_read16(sel),
5644 	       vmcs_read32(sel + GUEST_ES_AR_BYTES - GUEST_ES_SELECTOR),
5645 	       vmcs_read32(sel + GUEST_ES_LIMIT - GUEST_ES_SELECTOR),
5646 	       vmcs_readl(sel + GUEST_ES_BASE - GUEST_ES_SELECTOR));
5647 }
5648 
5649 static void vmx_dump_dtsel(char *name, uint32_t limit)
5650 {
5651 	pr_err("%s                           limit=0x%08x, base=0x%016lx\n",
5652 	       name, vmcs_read32(limit),
5653 	       vmcs_readl(limit + GUEST_GDTR_BASE - GUEST_GDTR_LIMIT));
5654 }
5655 
5656 void dump_vmcs(void)
5657 {
5658 	u32 vmentry_ctl, vmexit_ctl;
5659 	u32 cpu_based_exec_ctrl, pin_based_exec_ctrl, secondary_exec_control;
5660 	unsigned long cr4;
5661 	u64 efer;
5662 	int i, n;
5663 
5664 	if (!dump_invalid_vmcs) {
5665 		pr_warn_ratelimited("set kvm_intel.dump_invalid_vmcs=1 to dump internal KVM state.\n");
5666 		return;
5667 	}
5668 
5669 	vmentry_ctl = vmcs_read32(VM_ENTRY_CONTROLS);
5670 	vmexit_ctl = vmcs_read32(VM_EXIT_CONTROLS);
5671 	cpu_based_exec_ctrl = vmcs_read32(CPU_BASED_VM_EXEC_CONTROL);
5672 	pin_based_exec_ctrl = vmcs_read32(PIN_BASED_VM_EXEC_CONTROL);
5673 	cr4 = vmcs_readl(GUEST_CR4);
5674 	efer = vmcs_read64(GUEST_IA32_EFER);
5675 	secondary_exec_control = 0;
5676 	if (cpu_has_secondary_exec_ctrls())
5677 		secondary_exec_control = vmcs_read32(SECONDARY_VM_EXEC_CONTROL);
5678 
5679 	pr_err("*** Guest State ***\n");
5680 	pr_err("CR0: actual=0x%016lx, shadow=0x%016lx, gh_mask=%016lx\n",
5681 	       vmcs_readl(GUEST_CR0), vmcs_readl(CR0_READ_SHADOW),
5682 	       vmcs_readl(CR0_GUEST_HOST_MASK));
5683 	pr_err("CR4: actual=0x%016lx, shadow=0x%016lx, gh_mask=%016lx\n",
5684 	       cr4, vmcs_readl(CR4_READ_SHADOW), vmcs_readl(CR4_GUEST_HOST_MASK));
5685 	pr_err("CR3 = 0x%016lx\n", vmcs_readl(GUEST_CR3));
5686 	if ((secondary_exec_control & SECONDARY_EXEC_ENABLE_EPT) &&
5687 	    (cr4 & X86_CR4_PAE) && !(efer & EFER_LMA))
5688 	{
5689 		pr_err("PDPTR0 = 0x%016llx  PDPTR1 = 0x%016llx\n",
5690 		       vmcs_read64(GUEST_PDPTR0), vmcs_read64(GUEST_PDPTR1));
5691 		pr_err("PDPTR2 = 0x%016llx  PDPTR3 = 0x%016llx\n",
5692 		       vmcs_read64(GUEST_PDPTR2), vmcs_read64(GUEST_PDPTR3));
5693 	}
5694 	pr_err("RSP = 0x%016lx  RIP = 0x%016lx\n",
5695 	       vmcs_readl(GUEST_RSP), vmcs_readl(GUEST_RIP));
5696 	pr_err("RFLAGS=0x%08lx         DR7 = 0x%016lx\n",
5697 	       vmcs_readl(GUEST_RFLAGS), vmcs_readl(GUEST_DR7));
5698 	pr_err("Sysenter RSP=%016lx CS:RIP=%04x:%016lx\n",
5699 	       vmcs_readl(GUEST_SYSENTER_ESP),
5700 	       vmcs_read32(GUEST_SYSENTER_CS), vmcs_readl(GUEST_SYSENTER_EIP));
5701 	vmx_dump_sel("CS:  ", GUEST_CS_SELECTOR);
5702 	vmx_dump_sel("DS:  ", GUEST_DS_SELECTOR);
5703 	vmx_dump_sel("SS:  ", GUEST_SS_SELECTOR);
5704 	vmx_dump_sel("ES:  ", GUEST_ES_SELECTOR);
5705 	vmx_dump_sel("FS:  ", GUEST_FS_SELECTOR);
5706 	vmx_dump_sel("GS:  ", GUEST_GS_SELECTOR);
5707 	vmx_dump_dtsel("GDTR:", GUEST_GDTR_LIMIT);
5708 	vmx_dump_sel("LDTR:", GUEST_LDTR_SELECTOR);
5709 	vmx_dump_dtsel("IDTR:", GUEST_IDTR_LIMIT);
5710 	vmx_dump_sel("TR:  ", GUEST_TR_SELECTOR);
5711 	if ((vmexit_ctl & (VM_EXIT_SAVE_IA32_PAT | VM_EXIT_SAVE_IA32_EFER)) ||
5712 	    (vmentry_ctl & (VM_ENTRY_LOAD_IA32_PAT | VM_ENTRY_LOAD_IA32_EFER)))
5713 		pr_err("EFER =     0x%016llx  PAT = 0x%016llx\n",
5714 		       efer, vmcs_read64(GUEST_IA32_PAT));
5715 	pr_err("DebugCtl = 0x%016llx  DebugExceptions = 0x%016lx\n",
5716 	       vmcs_read64(GUEST_IA32_DEBUGCTL),
5717 	       vmcs_readl(GUEST_PENDING_DBG_EXCEPTIONS));
5718 	if (cpu_has_load_perf_global_ctrl() &&
5719 	    vmentry_ctl & VM_ENTRY_LOAD_IA32_PERF_GLOBAL_CTRL)
5720 		pr_err("PerfGlobCtl = 0x%016llx\n",
5721 		       vmcs_read64(GUEST_IA32_PERF_GLOBAL_CTRL));
5722 	if (vmentry_ctl & VM_ENTRY_LOAD_BNDCFGS)
5723 		pr_err("BndCfgS = 0x%016llx\n", vmcs_read64(GUEST_BNDCFGS));
5724 	pr_err("Interruptibility = %08x  ActivityState = %08x\n",
5725 	       vmcs_read32(GUEST_INTERRUPTIBILITY_INFO),
5726 	       vmcs_read32(GUEST_ACTIVITY_STATE));
5727 	if (secondary_exec_control & SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY)
5728 		pr_err("InterruptStatus = %04x\n",
5729 		       vmcs_read16(GUEST_INTR_STATUS));
5730 
5731 	pr_err("*** Host State ***\n");
5732 	pr_err("RIP = 0x%016lx  RSP = 0x%016lx\n",
5733 	       vmcs_readl(HOST_RIP), vmcs_readl(HOST_RSP));
5734 	pr_err("CS=%04x SS=%04x DS=%04x ES=%04x FS=%04x GS=%04x TR=%04x\n",
5735 	       vmcs_read16(HOST_CS_SELECTOR), vmcs_read16(HOST_SS_SELECTOR),
5736 	       vmcs_read16(HOST_DS_SELECTOR), vmcs_read16(HOST_ES_SELECTOR),
5737 	       vmcs_read16(HOST_FS_SELECTOR), vmcs_read16(HOST_GS_SELECTOR),
5738 	       vmcs_read16(HOST_TR_SELECTOR));
5739 	pr_err("FSBase=%016lx GSBase=%016lx TRBase=%016lx\n",
5740 	       vmcs_readl(HOST_FS_BASE), vmcs_readl(HOST_GS_BASE),
5741 	       vmcs_readl(HOST_TR_BASE));
5742 	pr_err("GDTBase=%016lx IDTBase=%016lx\n",
5743 	       vmcs_readl(HOST_GDTR_BASE), vmcs_readl(HOST_IDTR_BASE));
5744 	pr_err("CR0=%016lx CR3=%016lx CR4=%016lx\n",
5745 	       vmcs_readl(HOST_CR0), vmcs_readl(HOST_CR3),
5746 	       vmcs_readl(HOST_CR4));
5747 	pr_err("Sysenter RSP=%016lx CS:RIP=%04x:%016lx\n",
5748 	       vmcs_readl(HOST_IA32_SYSENTER_ESP),
5749 	       vmcs_read32(HOST_IA32_SYSENTER_CS),
5750 	       vmcs_readl(HOST_IA32_SYSENTER_EIP));
5751 	if (vmexit_ctl & (VM_EXIT_LOAD_IA32_PAT | VM_EXIT_LOAD_IA32_EFER))
5752 		pr_err("EFER = 0x%016llx  PAT = 0x%016llx\n",
5753 		       vmcs_read64(HOST_IA32_EFER),
5754 		       vmcs_read64(HOST_IA32_PAT));
5755 	if (cpu_has_load_perf_global_ctrl() &&
5756 	    vmexit_ctl & VM_EXIT_LOAD_IA32_PERF_GLOBAL_CTRL)
5757 		pr_err("PerfGlobCtl = 0x%016llx\n",
5758 		       vmcs_read64(HOST_IA32_PERF_GLOBAL_CTRL));
5759 
5760 	pr_err("*** Control State ***\n");
5761 	pr_err("PinBased=%08x CPUBased=%08x SecondaryExec=%08x\n",
5762 	       pin_based_exec_ctrl, cpu_based_exec_ctrl, secondary_exec_control);
5763 	pr_err("EntryControls=%08x ExitControls=%08x\n", vmentry_ctl, vmexit_ctl);
5764 	pr_err("ExceptionBitmap=%08x PFECmask=%08x PFECmatch=%08x\n",
5765 	       vmcs_read32(EXCEPTION_BITMAP),
5766 	       vmcs_read32(PAGE_FAULT_ERROR_CODE_MASK),
5767 	       vmcs_read32(PAGE_FAULT_ERROR_CODE_MATCH));
5768 	pr_err("VMEntry: intr_info=%08x errcode=%08x ilen=%08x\n",
5769 	       vmcs_read32(VM_ENTRY_INTR_INFO_FIELD),
5770 	       vmcs_read32(VM_ENTRY_EXCEPTION_ERROR_CODE),
5771 	       vmcs_read32(VM_ENTRY_INSTRUCTION_LEN));
5772 	pr_err("VMExit: intr_info=%08x errcode=%08x ilen=%08x\n",
5773 	       vmcs_read32(VM_EXIT_INTR_INFO),
5774 	       vmcs_read32(VM_EXIT_INTR_ERROR_CODE),
5775 	       vmcs_read32(VM_EXIT_INSTRUCTION_LEN));
5776 	pr_err("        reason=%08x qualification=%016lx\n",
5777 	       vmcs_read32(VM_EXIT_REASON), vmcs_readl(EXIT_QUALIFICATION));
5778 	pr_err("IDTVectoring: info=%08x errcode=%08x\n",
5779 	       vmcs_read32(IDT_VECTORING_INFO_FIELD),
5780 	       vmcs_read32(IDT_VECTORING_ERROR_CODE));
5781 	pr_err("TSC Offset = 0x%016llx\n", vmcs_read64(TSC_OFFSET));
5782 	if (secondary_exec_control & SECONDARY_EXEC_TSC_SCALING)
5783 		pr_err("TSC Multiplier = 0x%016llx\n",
5784 		       vmcs_read64(TSC_MULTIPLIER));
5785 	if (cpu_based_exec_ctrl & CPU_BASED_TPR_SHADOW) {
5786 		if (secondary_exec_control & SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY) {
5787 			u16 status = vmcs_read16(GUEST_INTR_STATUS);
5788 			pr_err("SVI|RVI = %02x|%02x ", status >> 8, status & 0xff);
5789 		}
5790 		pr_cont("TPR Threshold = 0x%02x\n", vmcs_read32(TPR_THRESHOLD));
5791 		if (secondary_exec_control & SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES)
5792 			pr_err("APIC-access addr = 0x%016llx ", vmcs_read64(APIC_ACCESS_ADDR));
5793 		pr_cont("virt-APIC addr = 0x%016llx\n", vmcs_read64(VIRTUAL_APIC_PAGE_ADDR));
5794 	}
5795 	if (pin_based_exec_ctrl & PIN_BASED_POSTED_INTR)
5796 		pr_err("PostedIntrVec = 0x%02x\n", vmcs_read16(POSTED_INTR_NV));
5797 	if ((secondary_exec_control & SECONDARY_EXEC_ENABLE_EPT))
5798 		pr_err("EPT pointer = 0x%016llx\n", vmcs_read64(EPT_POINTER));
5799 	n = vmcs_read32(CR3_TARGET_COUNT);
5800 	for (i = 0; i + 1 < n; i += 4)
5801 		pr_err("CR3 target%u=%016lx target%u=%016lx\n",
5802 		       i, vmcs_readl(CR3_TARGET_VALUE0 + i * 2),
5803 		       i + 1, vmcs_readl(CR3_TARGET_VALUE0 + i * 2 + 2));
5804 	if (i < n)
5805 		pr_err("CR3 target%u=%016lx\n",
5806 		       i, vmcs_readl(CR3_TARGET_VALUE0 + i * 2));
5807 	if (secondary_exec_control & SECONDARY_EXEC_PAUSE_LOOP_EXITING)
5808 		pr_err("PLE Gap=%08x Window=%08x\n",
5809 		       vmcs_read32(PLE_GAP), vmcs_read32(PLE_WINDOW));
5810 	if (secondary_exec_control & SECONDARY_EXEC_ENABLE_VPID)
5811 		pr_err("Virtual processor ID = 0x%04x\n",
5812 		       vmcs_read16(VIRTUAL_PROCESSOR_ID));
5813 }
5814 
5815 /*
5816  * The guest has exited.  See if we can fix it or if we need userspace
5817  * assistance.
5818  */
5819 static int vmx_handle_exit(struct kvm_vcpu *vcpu)
5820 {
5821 	struct vcpu_vmx *vmx = to_vmx(vcpu);
5822 	u32 exit_reason = vmx->exit_reason;
5823 	u32 vectoring_info = vmx->idt_vectoring_info;
5824 
5825 	trace_kvm_exit(exit_reason, vcpu, KVM_ISA_VMX);
5826 
5827 	/*
5828 	 * Flush logged GPAs PML buffer, this will make dirty_bitmap more
5829 	 * updated. Another good is, in kvm_vm_ioctl_get_dirty_log, before
5830 	 * querying dirty_bitmap, we only need to kick all vcpus out of guest
5831 	 * mode as if vcpus is in root mode, the PML buffer must has been
5832 	 * flushed already.
5833 	 */
5834 	if (enable_pml)
5835 		vmx_flush_pml_buffer(vcpu);
5836 
5837 	/* If guest state is invalid, start emulating */
5838 	if (vmx->emulation_required)
5839 		return handle_invalid_guest_state(vcpu);
5840 
5841 	if (is_guest_mode(vcpu) && nested_vmx_exit_reflected(vcpu, exit_reason))
5842 		return nested_vmx_reflect_vmexit(vcpu, exit_reason);
5843 
5844 	if (exit_reason & VMX_EXIT_REASONS_FAILED_VMENTRY) {
5845 		dump_vmcs();
5846 		vcpu->run->exit_reason = KVM_EXIT_FAIL_ENTRY;
5847 		vcpu->run->fail_entry.hardware_entry_failure_reason
5848 			= exit_reason;
5849 		return 0;
5850 	}
5851 
5852 	if (unlikely(vmx->fail)) {
5853 		dump_vmcs();
5854 		vcpu->run->exit_reason = KVM_EXIT_FAIL_ENTRY;
5855 		vcpu->run->fail_entry.hardware_entry_failure_reason
5856 			= vmcs_read32(VM_INSTRUCTION_ERROR);
5857 		return 0;
5858 	}
5859 
5860 	/*
5861 	 * Note:
5862 	 * Do not try to fix EXIT_REASON_EPT_MISCONFIG if it caused by
5863 	 * delivery event since it indicates guest is accessing MMIO.
5864 	 * The vm-exit can be triggered again after return to guest that
5865 	 * will cause infinite loop.
5866 	 */
5867 	if ((vectoring_info & VECTORING_INFO_VALID_MASK) &&
5868 			(exit_reason != EXIT_REASON_EXCEPTION_NMI &&
5869 			exit_reason != EXIT_REASON_EPT_VIOLATION &&
5870 			exit_reason != EXIT_REASON_PML_FULL &&
5871 			exit_reason != EXIT_REASON_TASK_SWITCH)) {
5872 		vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
5873 		vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_DELIVERY_EV;
5874 		vcpu->run->internal.ndata = 3;
5875 		vcpu->run->internal.data[0] = vectoring_info;
5876 		vcpu->run->internal.data[1] = exit_reason;
5877 		vcpu->run->internal.data[2] = vcpu->arch.exit_qualification;
5878 		if (exit_reason == EXIT_REASON_EPT_MISCONFIG) {
5879 			vcpu->run->internal.ndata++;
5880 			vcpu->run->internal.data[3] =
5881 				vmcs_read64(GUEST_PHYSICAL_ADDRESS);
5882 		}
5883 		return 0;
5884 	}
5885 
5886 	if (unlikely(!enable_vnmi &&
5887 		     vmx->loaded_vmcs->soft_vnmi_blocked)) {
5888 		if (vmx_interrupt_allowed(vcpu)) {
5889 			vmx->loaded_vmcs->soft_vnmi_blocked = 0;
5890 		} else if (vmx->loaded_vmcs->vnmi_blocked_time > 1000000000LL &&
5891 			   vcpu->arch.nmi_pending) {
5892 			/*
5893 			 * This CPU don't support us in finding the end of an
5894 			 * NMI-blocked window if the guest runs with IRQs
5895 			 * disabled. So we pull the trigger after 1 s of
5896 			 * futile waiting, but inform the user about this.
5897 			 */
5898 			printk(KERN_WARNING "%s: Breaking out of NMI-blocked "
5899 			       "state on VCPU %d after 1 s timeout\n",
5900 			       __func__, vcpu->vcpu_id);
5901 			vmx->loaded_vmcs->soft_vnmi_blocked = 0;
5902 		}
5903 	}
5904 
5905 	if (exit_reason < kvm_vmx_max_exit_handlers
5906 	    && kvm_vmx_exit_handlers[exit_reason]) {
5907 #ifdef CONFIG_RETPOLINE
5908 		if (exit_reason == EXIT_REASON_MSR_WRITE)
5909 			return kvm_emulate_wrmsr(vcpu);
5910 		else if (exit_reason == EXIT_REASON_PREEMPTION_TIMER)
5911 			return handle_preemption_timer(vcpu);
5912 		else if (exit_reason == EXIT_REASON_PENDING_INTERRUPT)
5913 			return handle_interrupt_window(vcpu);
5914 		else if (exit_reason == EXIT_REASON_EXTERNAL_INTERRUPT)
5915 			return handle_external_interrupt(vcpu);
5916 		else if (exit_reason == EXIT_REASON_HLT)
5917 			return kvm_emulate_halt(vcpu);
5918 		else if (exit_reason == EXIT_REASON_EPT_MISCONFIG)
5919 			return handle_ept_misconfig(vcpu);
5920 #endif
5921 		return kvm_vmx_exit_handlers[exit_reason](vcpu);
5922 	} else {
5923 		vcpu_unimpl(vcpu, "vmx: unexpected exit reason 0x%x\n",
5924 				exit_reason);
5925 		dump_vmcs();
5926 		vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
5927 		vcpu->run->internal.suberror =
5928 			KVM_INTERNAL_ERROR_UNEXPECTED_EXIT_REASON;
5929 		vcpu->run->internal.ndata = 1;
5930 		vcpu->run->internal.data[0] = exit_reason;
5931 		return 0;
5932 	}
5933 }
5934 
5935 /*
5936  * Software based L1D cache flush which is used when microcode providing
5937  * the cache control MSR is not loaded.
5938  *
5939  * The L1D cache is 32 KiB on Nehalem and later microarchitectures, but to
5940  * flush it is required to read in 64 KiB because the replacement algorithm
5941  * is not exactly LRU. This could be sized at runtime via topology
5942  * information but as all relevant affected CPUs have 32KiB L1D cache size
5943  * there is no point in doing so.
5944  */
5945 static void vmx_l1d_flush(struct kvm_vcpu *vcpu)
5946 {
5947 	int size = PAGE_SIZE << L1D_CACHE_ORDER;
5948 
5949 	/*
5950 	 * This code is only executed when the the flush mode is 'cond' or
5951 	 * 'always'
5952 	 */
5953 	if (static_branch_likely(&vmx_l1d_flush_cond)) {
5954 		bool flush_l1d;
5955 
5956 		/*
5957 		 * Clear the per-vcpu flush bit, it gets set again
5958 		 * either from vcpu_run() or from one of the unsafe
5959 		 * VMEXIT handlers.
5960 		 */
5961 		flush_l1d = vcpu->arch.l1tf_flush_l1d;
5962 		vcpu->arch.l1tf_flush_l1d = false;
5963 
5964 		/*
5965 		 * Clear the per-cpu flush bit, it gets set again from
5966 		 * the interrupt handlers.
5967 		 */
5968 		flush_l1d |= kvm_get_cpu_l1tf_flush_l1d();
5969 		kvm_clear_cpu_l1tf_flush_l1d();
5970 
5971 		if (!flush_l1d)
5972 			return;
5973 	}
5974 
5975 	vcpu->stat.l1d_flush++;
5976 
5977 	if (static_cpu_has(X86_FEATURE_FLUSH_L1D)) {
5978 		wrmsrl(MSR_IA32_FLUSH_CMD, L1D_FLUSH);
5979 		return;
5980 	}
5981 
5982 	asm volatile(
5983 		/* First ensure the pages are in the TLB */
5984 		"xorl	%%eax, %%eax\n"
5985 		".Lpopulate_tlb:\n\t"
5986 		"movzbl	(%[flush_pages], %%" _ASM_AX "), %%ecx\n\t"
5987 		"addl	$4096, %%eax\n\t"
5988 		"cmpl	%%eax, %[size]\n\t"
5989 		"jne	.Lpopulate_tlb\n\t"
5990 		"xorl	%%eax, %%eax\n\t"
5991 		"cpuid\n\t"
5992 		/* Now fill the cache */
5993 		"xorl	%%eax, %%eax\n"
5994 		".Lfill_cache:\n"
5995 		"movzbl	(%[flush_pages], %%" _ASM_AX "), %%ecx\n\t"
5996 		"addl	$64, %%eax\n\t"
5997 		"cmpl	%%eax, %[size]\n\t"
5998 		"jne	.Lfill_cache\n\t"
5999 		"lfence\n"
6000 		:: [flush_pages] "r" (vmx_l1d_flush_pages),
6001 		    [size] "r" (size)
6002 		: "eax", "ebx", "ecx", "edx");
6003 }
6004 
6005 static void update_cr8_intercept(struct kvm_vcpu *vcpu, int tpr, int irr)
6006 {
6007 	struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
6008 	int tpr_threshold;
6009 
6010 	if (is_guest_mode(vcpu) &&
6011 		nested_cpu_has(vmcs12, CPU_BASED_TPR_SHADOW))
6012 		return;
6013 
6014 	tpr_threshold = (irr == -1 || tpr < irr) ? 0 : irr;
6015 	if (is_guest_mode(vcpu))
6016 		to_vmx(vcpu)->nested.l1_tpr_threshold = tpr_threshold;
6017 	else
6018 		vmcs_write32(TPR_THRESHOLD, tpr_threshold);
6019 }
6020 
6021 void vmx_set_virtual_apic_mode(struct kvm_vcpu *vcpu)
6022 {
6023 	struct vcpu_vmx *vmx = to_vmx(vcpu);
6024 	u32 sec_exec_control;
6025 
6026 	if (!lapic_in_kernel(vcpu))
6027 		return;
6028 
6029 	if (!flexpriority_enabled &&
6030 	    !cpu_has_vmx_virtualize_x2apic_mode())
6031 		return;
6032 
6033 	/* Postpone execution until vmcs01 is the current VMCS. */
6034 	if (is_guest_mode(vcpu)) {
6035 		vmx->nested.change_vmcs01_virtual_apic_mode = true;
6036 		return;
6037 	}
6038 
6039 	sec_exec_control = secondary_exec_controls_get(vmx);
6040 	sec_exec_control &= ~(SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES |
6041 			      SECONDARY_EXEC_VIRTUALIZE_X2APIC_MODE);
6042 
6043 	switch (kvm_get_apic_mode(vcpu)) {
6044 	case LAPIC_MODE_INVALID:
6045 		WARN_ONCE(true, "Invalid local APIC state");
6046 	case LAPIC_MODE_DISABLED:
6047 		break;
6048 	case LAPIC_MODE_XAPIC:
6049 		if (flexpriority_enabled) {
6050 			sec_exec_control |=
6051 				SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES;
6052 			vmx_flush_tlb(vcpu, true);
6053 		}
6054 		break;
6055 	case LAPIC_MODE_X2APIC:
6056 		if (cpu_has_vmx_virtualize_x2apic_mode())
6057 			sec_exec_control |=
6058 				SECONDARY_EXEC_VIRTUALIZE_X2APIC_MODE;
6059 		break;
6060 	}
6061 	secondary_exec_controls_set(vmx, sec_exec_control);
6062 
6063 	vmx_update_msr_bitmap(vcpu);
6064 }
6065 
6066 static void vmx_set_apic_access_page_addr(struct kvm_vcpu *vcpu, hpa_t hpa)
6067 {
6068 	if (!is_guest_mode(vcpu)) {
6069 		vmcs_write64(APIC_ACCESS_ADDR, hpa);
6070 		vmx_flush_tlb(vcpu, true);
6071 	}
6072 }
6073 
6074 static void vmx_hwapic_isr_update(struct kvm_vcpu *vcpu, int max_isr)
6075 {
6076 	u16 status;
6077 	u8 old;
6078 
6079 	if (max_isr == -1)
6080 		max_isr = 0;
6081 
6082 	status = vmcs_read16(GUEST_INTR_STATUS);
6083 	old = status >> 8;
6084 	if (max_isr != old) {
6085 		status &= 0xff;
6086 		status |= max_isr << 8;
6087 		vmcs_write16(GUEST_INTR_STATUS, status);
6088 	}
6089 }
6090 
6091 static void vmx_set_rvi(int vector)
6092 {
6093 	u16 status;
6094 	u8 old;
6095 
6096 	if (vector == -1)
6097 		vector = 0;
6098 
6099 	status = vmcs_read16(GUEST_INTR_STATUS);
6100 	old = (u8)status & 0xff;
6101 	if ((u8)vector != old) {
6102 		status &= ~0xff;
6103 		status |= (u8)vector;
6104 		vmcs_write16(GUEST_INTR_STATUS, status);
6105 	}
6106 }
6107 
6108 static void vmx_hwapic_irr_update(struct kvm_vcpu *vcpu, int max_irr)
6109 {
6110 	/*
6111 	 * When running L2, updating RVI is only relevant when
6112 	 * vmcs12 virtual-interrupt-delivery enabled.
6113 	 * However, it can be enabled only when L1 also
6114 	 * intercepts external-interrupts and in that case
6115 	 * we should not update vmcs02 RVI but instead intercept
6116 	 * interrupt. Therefore, do nothing when running L2.
6117 	 */
6118 	if (!is_guest_mode(vcpu))
6119 		vmx_set_rvi(max_irr);
6120 }
6121 
6122 static int vmx_sync_pir_to_irr(struct kvm_vcpu *vcpu)
6123 {
6124 	struct vcpu_vmx *vmx = to_vmx(vcpu);
6125 	int max_irr;
6126 	bool max_irr_updated;
6127 
6128 	WARN_ON(!vcpu->arch.apicv_active);
6129 	if (pi_test_on(&vmx->pi_desc)) {
6130 		pi_clear_on(&vmx->pi_desc);
6131 		/*
6132 		 * IOMMU can write to PID.ON, so the barrier matters even on UP.
6133 		 * But on x86 this is just a compiler barrier anyway.
6134 		 */
6135 		smp_mb__after_atomic();
6136 		max_irr_updated =
6137 			kvm_apic_update_irr(vcpu, vmx->pi_desc.pir, &max_irr);
6138 
6139 		/*
6140 		 * If we are running L2 and L1 has a new pending interrupt
6141 		 * which can be injected, we should re-evaluate
6142 		 * what should be done with this new L1 interrupt.
6143 		 * If L1 intercepts external-interrupts, we should
6144 		 * exit from L2 to L1. Otherwise, interrupt should be
6145 		 * delivered directly to L2.
6146 		 */
6147 		if (is_guest_mode(vcpu) && max_irr_updated) {
6148 			if (nested_exit_on_intr(vcpu))
6149 				kvm_vcpu_exiting_guest_mode(vcpu);
6150 			else
6151 				kvm_make_request(KVM_REQ_EVENT, vcpu);
6152 		}
6153 	} else {
6154 		max_irr = kvm_lapic_find_highest_irr(vcpu);
6155 	}
6156 	vmx_hwapic_irr_update(vcpu, max_irr);
6157 	return max_irr;
6158 }
6159 
6160 static bool vmx_dy_apicv_has_pending_interrupt(struct kvm_vcpu *vcpu)
6161 {
6162 	struct pi_desc *pi_desc = vcpu_to_pi_desc(vcpu);
6163 
6164 	return pi_test_on(pi_desc) ||
6165 		(pi_test_sn(pi_desc) && !pi_is_pir_empty(pi_desc));
6166 }
6167 
6168 static void vmx_load_eoi_exitmap(struct kvm_vcpu *vcpu, u64 *eoi_exit_bitmap)
6169 {
6170 	if (!kvm_vcpu_apicv_active(vcpu))
6171 		return;
6172 
6173 	vmcs_write64(EOI_EXIT_BITMAP0, eoi_exit_bitmap[0]);
6174 	vmcs_write64(EOI_EXIT_BITMAP1, eoi_exit_bitmap[1]);
6175 	vmcs_write64(EOI_EXIT_BITMAP2, eoi_exit_bitmap[2]);
6176 	vmcs_write64(EOI_EXIT_BITMAP3, eoi_exit_bitmap[3]);
6177 }
6178 
6179 static void vmx_apicv_post_state_restore(struct kvm_vcpu *vcpu)
6180 {
6181 	struct vcpu_vmx *vmx = to_vmx(vcpu);
6182 
6183 	pi_clear_on(&vmx->pi_desc);
6184 	memset(vmx->pi_desc.pir, 0, sizeof(vmx->pi_desc.pir));
6185 }
6186 
6187 static void handle_exception_nmi_irqoff(struct vcpu_vmx *vmx)
6188 {
6189 	vmx->exit_intr_info = vmcs_read32(VM_EXIT_INTR_INFO);
6190 
6191 	/* if exit due to PF check for async PF */
6192 	if (is_page_fault(vmx->exit_intr_info))
6193 		vmx->vcpu.arch.apf.host_apf_reason = kvm_read_and_reset_pf_reason();
6194 
6195 	/* Handle machine checks before interrupts are enabled */
6196 	if (is_machine_check(vmx->exit_intr_info))
6197 		kvm_machine_check();
6198 
6199 	/* We need to handle NMIs before interrupts are enabled */
6200 	if (is_nmi(vmx->exit_intr_info)) {
6201 		kvm_before_interrupt(&vmx->vcpu);
6202 		asm("int $2");
6203 		kvm_after_interrupt(&vmx->vcpu);
6204 	}
6205 }
6206 
6207 static void handle_external_interrupt_irqoff(struct kvm_vcpu *vcpu)
6208 {
6209 	unsigned int vector;
6210 	unsigned long entry;
6211 #ifdef CONFIG_X86_64
6212 	unsigned long tmp;
6213 #endif
6214 	gate_desc *desc;
6215 	u32 intr_info;
6216 
6217 	intr_info = vmcs_read32(VM_EXIT_INTR_INFO);
6218 	if (WARN_ONCE(!is_external_intr(intr_info),
6219 	    "KVM: unexpected VM-Exit interrupt info: 0x%x", intr_info))
6220 		return;
6221 
6222 	vector = intr_info & INTR_INFO_VECTOR_MASK;
6223 	desc = (gate_desc *)host_idt_base + vector;
6224 	entry = gate_offset(desc);
6225 
6226 	kvm_before_interrupt(vcpu);
6227 
6228 	asm volatile(
6229 #ifdef CONFIG_X86_64
6230 		"mov %%" _ASM_SP ", %[sp]\n\t"
6231 		"and $0xfffffffffffffff0, %%" _ASM_SP "\n\t"
6232 		"push $%c[ss]\n\t"
6233 		"push %[sp]\n\t"
6234 #endif
6235 		"pushf\n\t"
6236 		__ASM_SIZE(push) " $%c[cs]\n\t"
6237 		CALL_NOSPEC
6238 		:
6239 #ifdef CONFIG_X86_64
6240 		[sp]"=&r"(tmp),
6241 #endif
6242 		ASM_CALL_CONSTRAINT
6243 		:
6244 		THUNK_TARGET(entry),
6245 		[ss]"i"(__KERNEL_DS),
6246 		[cs]"i"(__KERNEL_CS)
6247 	);
6248 
6249 	kvm_after_interrupt(vcpu);
6250 }
6251 STACK_FRAME_NON_STANDARD(handle_external_interrupt_irqoff);
6252 
6253 static void vmx_handle_exit_irqoff(struct kvm_vcpu *vcpu)
6254 {
6255 	struct vcpu_vmx *vmx = to_vmx(vcpu);
6256 
6257 	if (vmx->exit_reason == EXIT_REASON_EXTERNAL_INTERRUPT)
6258 		handle_external_interrupt_irqoff(vcpu);
6259 	else if (vmx->exit_reason == EXIT_REASON_EXCEPTION_NMI)
6260 		handle_exception_nmi_irqoff(vmx);
6261 }
6262 
6263 static bool vmx_has_emulated_msr(int index)
6264 {
6265 	switch (index) {
6266 	case MSR_IA32_SMBASE:
6267 		/*
6268 		 * We cannot do SMM unless we can run the guest in big
6269 		 * real mode.
6270 		 */
6271 		return enable_unrestricted_guest || emulate_invalid_guest_state;
6272 	case MSR_IA32_VMX_BASIC ... MSR_IA32_VMX_VMFUNC:
6273 		return nested;
6274 	case MSR_AMD64_VIRT_SPEC_CTRL:
6275 		/* This is AMD only.  */
6276 		return false;
6277 	default:
6278 		return true;
6279 	}
6280 }
6281 
6282 static bool vmx_pt_supported(void)
6283 {
6284 	return pt_mode == PT_MODE_HOST_GUEST;
6285 }
6286 
6287 static void vmx_recover_nmi_blocking(struct vcpu_vmx *vmx)
6288 {
6289 	u32 exit_intr_info;
6290 	bool unblock_nmi;
6291 	u8 vector;
6292 	bool idtv_info_valid;
6293 
6294 	idtv_info_valid = vmx->idt_vectoring_info & VECTORING_INFO_VALID_MASK;
6295 
6296 	if (enable_vnmi) {
6297 		if (vmx->loaded_vmcs->nmi_known_unmasked)
6298 			return;
6299 		/*
6300 		 * Can't use vmx->exit_intr_info since we're not sure what
6301 		 * the exit reason is.
6302 		 */
6303 		exit_intr_info = vmcs_read32(VM_EXIT_INTR_INFO);
6304 		unblock_nmi = (exit_intr_info & INTR_INFO_UNBLOCK_NMI) != 0;
6305 		vector = exit_intr_info & INTR_INFO_VECTOR_MASK;
6306 		/*
6307 		 * SDM 3: 27.7.1.2 (September 2008)
6308 		 * Re-set bit "block by NMI" before VM entry if vmexit caused by
6309 		 * a guest IRET fault.
6310 		 * SDM 3: 23.2.2 (September 2008)
6311 		 * Bit 12 is undefined in any of the following cases:
6312 		 *  If the VM exit sets the valid bit in the IDT-vectoring
6313 		 *   information field.
6314 		 *  If the VM exit is due to a double fault.
6315 		 */
6316 		if ((exit_intr_info & INTR_INFO_VALID_MASK) && unblock_nmi &&
6317 		    vector != DF_VECTOR && !idtv_info_valid)
6318 			vmcs_set_bits(GUEST_INTERRUPTIBILITY_INFO,
6319 				      GUEST_INTR_STATE_NMI);
6320 		else
6321 			vmx->loaded_vmcs->nmi_known_unmasked =
6322 				!(vmcs_read32(GUEST_INTERRUPTIBILITY_INFO)
6323 				  & GUEST_INTR_STATE_NMI);
6324 	} else if (unlikely(vmx->loaded_vmcs->soft_vnmi_blocked))
6325 		vmx->loaded_vmcs->vnmi_blocked_time +=
6326 			ktime_to_ns(ktime_sub(ktime_get(),
6327 					      vmx->loaded_vmcs->entry_time));
6328 }
6329 
6330 static void __vmx_complete_interrupts(struct kvm_vcpu *vcpu,
6331 				      u32 idt_vectoring_info,
6332 				      int instr_len_field,
6333 				      int error_code_field)
6334 {
6335 	u8 vector;
6336 	int type;
6337 	bool idtv_info_valid;
6338 
6339 	idtv_info_valid = idt_vectoring_info & VECTORING_INFO_VALID_MASK;
6340 
6341 	vcpu->arch.nmi_injected = false;
6342 	kvm_clear_exception_queue(vcpu);
6343 	kvm_clear_interrupt_queue(vcpu);
6344 
6345 	if (!idtv_info_valid)
6346 		return;
6347 
6348 	kvm_make_request(KVM_REQ_EVENT, vcpu);
6349 
6350 	vector = idt_vectoring_info & VECTORING_INFO_VECTOR_MASK;
6351 	type = idt_vectoring_info & VECTORING_INFO_TYPE_MASK;
6352 
6353 	switch (type) {
6354 	case INTR_TYPE_NMI_INTR:
6355 		vcpu->arch.nmi_injected = true;
6356 		/*
6357 		 * SDM 3: 27.7.1.2 (September 2008)
6358 		 * Clear bit "block by NMI" before VM entry if a NMI
6359 		 * delivery faulted.
6360 		 */
6361 		vmx_set_nmi_mask(vcpu, false);
6362 		break;
6363 	case INTR_TYPE_SOFT_EXCEPTION:
6364 		vcpu->arch.event_exit_inst_len = vmcs_read32(instr_len_field);
6365 		/* fall through */
6366 	case INTR_TYPE_HARD_EXCEPTION:
6367 		if (idt_vectoring_info & VECTORING_INFO_DELIVER_CODE_MASK) {
6368 			u32 err = vmcs_read32(error_code_field);
6369 			kvm_requeue_exception_e(vcpu, vector, err);
6370 		} else
6371 			kvm_requeue_exception(vcpu, vector);
6372 		break;
6373 	case INTR_TYPE_SOFT_INTR:
6374 		vcpu->arch.event_exit_inst_len = vmcs_read32(instr_len_field);
6375 		/* fall through */
6376 	case INTR_TYPE_EXT_INTR:
6377 		kvm_queue_interrupt(vcpu, vector, type == INTR_TYPE_SOFT_INTR);
6378 		break;
6379 	default:
6380 		break;
6381 	}
6382 }
6383 
6384 static void vmx_complete_interrupts(struct vcpu_vmx *vmx)
6385 {
6386 	__vmx_complete_interrupts(&vmx->vcpu, vmx->idt_vectoring_info,
6387 				  VM_EXIT_INSTRUCTION_LEN,
6388 				  IDT_VECTORING_ERROR_CODE);
6389 }
6390 
6391 static void vmx_cancel_injection(struct kvm_vcpu *vcpu)
6392 {
6393 	__vmx_complete_interrupts(vcpu,
6394 				  vmcs_read32(VM_ENTRY_INTR_INFO_FIELD),
6395 				  VM_ENTRY_INSTRUCTION_LEN,
6396 				  VM_ENTRY_EXCEPTION_ERROR_CODE);
6397 
6398 	vmcs_write32(VM_ENTRY_INTR_INFO_FIELD, 0);
6399 }
6400 
6401 static void atomic_switch_perf_msrs(struct vcpu_vmx *vmx)
6402 {
6403 	int i, nr_msrs;
6404 	struct perf_guest_switch_msr *msrs;
6405 
6406 	msrs = perf_guest_get_msrs(&nr_msrs);
6407 
6408 	if (!msrs)
6409 		return;
6410 
6411 	for (i = 0; i < nr_msrs; i++)
6412 		if (msrs[i].host == msrs[i].guest)
6413 			clear_atomic_switch_msr(vmx, msrs[i].msr);
6414 		else
6415 			add_atomic_switch_msr(vmx, msrs[i].msr, msrs[i].guest,
6416 					msrs[i].host, false);
6417 }
6418 
6419 static void atomic_switch_umwait_control_msr(struct vcpu_vmx *vmx)
6420 {
6421 	u32 host_umwait_control;
6422 
6423 	if (!vmx_has_waitpkg(vmx))
6424 		return;
6425 
6426 	host_umwait_control = get_umwait_control_msr();
6427 
6428 	if (vmx->msr_ia32_umwait_control != host_umwait_control)
6429 		add_atomic_switch_msr(vmx, MSR_IA32_UMWAIT_CONTROL,
6430 			vmx->msr_ia32_umwait_control,
6431 			host_umwait_control, false);
6432 	else
6433 		clear_atomic_switch_msr(vmx, MSR_IA32_UMWAIT_CONTROL);
6434 }
6435 
6436 static void vmx_update_hv_timer(struct kvm_vcpu *vcpu)
6437 {
6438 	struct vcpu_vmx *vmx = to_vmx(vcpu);
6439 	u64 tscl;
6440 	u32 delta_tsc;
6441 
6442 	if (vmx->req_immediate_exit) {
6443 		vmcs_write32(VMX_PREEMPTION_TIMER_VALUE, 0);
6444 		vmx->loaded_vmcs->hv_timer_soft_disabled = false;
6445 	} else if (vmx->hv_deadline_tsc != -1) {
6446 		tscl = rdtsc();
6447 		if (vmx->hv_deadline_tsc > tscl)
6448 			/* set_hv_timer ensures the delta fits in 32-bits */
6449 			delta_tsc = (u32)((vmx->hv_deadline_tsc - tscl) >>
6450 				cpu_preemption_timer_multi);
6451 		else
6452 			delta_tsc = 0;
6453 
6454 		vmcs_write32(VMX_PREEMPTION_TIMER_VALUE, delta_tsc);
6455 		vmx->loaded_vmcs->hv_timer_soft_disabled = false;
6456 	} else if (!vmx->loaded_vmcs->hv_timer_soft_disabled) {
6457 		vmcs_write32(VMX_PREEMPTION_TIMER_VALUE, -1);
6458 		vmx->loaded_vmcs->hv_timer_soft_disabled = true;
6459 	}
6460 }
6461 
6462 void vmx_update_host_rsp(struct vcpu_vmx *vmx, unsigned long host_rsp)
6463 {
6464 	if (unlikely(host_rsp != vmx->loaded_vmcs->host_state.rsp)) {
6465 		vmx->loaded_vmcs->host_state.rsp = host_rsp;
6466 		vmcs_writel(HOST_RSP, host_rsp);
6467 	}
6468 }
6469 
6470 bool __vmx_vcpu_run(struct vcpu_vmx *vmx, unsigned long *regs, bool launched);
6471 
6472 static void vmx_vcpu_run(struct kvm_vcpu *vcpu)
6473 {
6474 	struct vcpu_vmx *vmx = to_vmx(vcpu);
6475 	unsigned long cr3, cr4;
6476 
6477 	/* Record the guest's net vcpu time for enforced NMI injections. */
6478 	if (unlikely(!enable_vnmi &&
6479 		     vmx->loaded_vmcs->soft_vnmi_blocked))
6480 		vmx->loaded_vmcs->entry_time = ktime_get();
6481 
6482 	/* Don't enter VMX if guest state is invalid, let the exit handler
6483 	   start emulation until we arrive back to a valid state */
6484 	if (vmx->emulation_required)
6485 		return;
6486 
6487 	if (vmx->ple_window_dirty) {
6488 		vmx->ple_window_dirty = false;
6489 		vmcs_write32(PLE_WINDOW, vmx->ple_window);
6490 	}
6491 
6492 	if (vmx->nested.need_vmcs12_to_shadow_sync)
6493 		nested_sync_vmcs12_to_shadow(vcpu);
6494 
6495 	if (kvm_register_is_dirty(vcpu, VCPU_REGS_RSP))
6496 		vmcs_writel(GUEST_RSP, vcpu->arch.regs[VCPU_REGS_RSP]);
6497 	if (kvm_register_is_dirty(vcpu, VCPU_REGS_RIP))
6498 		vmcs_writel(GUEST_RIP, vcpu->arch.regs[VCPU_REGS_RIP]);
6499 
6500 	cr3 = __get_current_cr3_fast();
6501 	if (unlikely(cr3 != vmx->loaded_vmcs->host_state.cr3)) {
6502 		vmcs_writel(HOST_CR3, cr3);
6503 		vmx->loaded_vmcs->host_state.cr3 = cr3;
6504 	}
6505 
6506 	cr4 = cr4_read_shadow();
6507 	if (unlikely(cr4 != vmx->loaded_vmcs->host_state.cr4)) {
6508 		vmcs_writel(HOST_CR4, cr4);
6509 		vmx->loaded_vmcs->host_state.cr4 = cr4;
6510 	}
6511 
6512 	/* When single-stepping over STI and MOV SS, we must clear the
6513 	 * corresponding interruptibility bits in the guest state. Otherwise
6514 	 * vmentry fails as it then expects bit 14 (BS) in pending debug
6515 	 * exceptions being set, but that's not correct for the guest debugging
6516 	 * case. */
6517 	if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)
6518 		vmx_set_interrupt_shadow(vcpu, 0);
6519 
6520 	kvm_load_guest_xsave_state(vcpu);
6521 
6522 	if (static_cpu_has(X86_FEATURE_PKU) &&
6523 	    kvm_read_cr4_bits(vcpu, X86_CR4_PKE) &&
6524 	    vcpu->arch.pkru != vmx->host_pkru)
6525 		__write_pkru(vcpu->arch.pkru);
6526 
6527 	pt_guest_enter(vmx);
6528 
6529 	atomic_switch_perf_msrs(vmx);
6530 	atomic_switch_umwait_control_msr(vmx);
6531 
6532 	if (enable_preemption_timer)
6533 		vmx_update_hv_timer(vcpu);
6534 
6535 	if (lapic_in_kernel(vcpu) &&
6536 		vcpu->arch.apic->lapic_timer.timer_advance_ns)
6537 		kvm_wait_lapic_expire(vcpu);
6538 
6539 	/*
6540 	 * If this vCPU has touched SPEC_CTRL, restore the guest's value if
6541 	 * it's non-zero. Since vmentry is serialising on affected CPUs, there
6542 	 * is no need to worry about the conditional branch over the wrmsr
6543 	 * being speculatively taken.
6544 	 */
6545 	x86_spec_ctrl_set_guest(vmx->spec_ctrl, 0);
6546 
6547 	/* L1D Flush includes CPU buffer clear to mitigate MDS */
6548 	if (static_branch_unlikely(&vmx_l1d_should_flush))
6549 		vmx_l1d_flush(vcpu);
6550 	else if (static_branch_unlikely(&mds_user_clear))
6551 		mds_clear_cpu_buffers();
6552 
6553 	if (vcpu->arch.cr2 != read_cr2())
6554 		write_cr2(vcpu->arch.cr2);
6555 
6556 	vmx->fail = __vmx_vcpu_run(vmx, (unsigned long *)&vcpu->arch.regs,
6557 				   vmx->loaded_vmcs->launched);
6558 
6559 	vcpu->arch.cr2 = read_cr2();
6560 
6561 	/*
6562 	 * We do not use IBRS in the kernel. If this vCPU has used the
6563 	 * SPEC_CTRL MSR it may have left it on; save the value and
6564 	 * turn it off. This is much more efficient than blindly adding
6565 	 * it to the atomic save/restore list. Especially as the former
6566 	 * (Saving guest MSRs on vmexit) doesn't even exist in KVM.
6567 	 *
6568 	 * For non-nested case:
6569 	 * If the L01 MSR bitmap does not intercept the MSR, then we need to
6570 	 * save it.
6571 	 *
6572 	 * For nested case:
6573 	 * If the L02 MSR bitmap does not intercept the MSR, then we need to
6574 	 * save it.
6575 	 */
6576 	if (unlikely(!msr_write_intercepted(vcpu, MSR_IA32_SPEC_CTRL)))
6577 		vmx->spec_ctrl = native_read_msr(MSR_IA32_SPEC_CTRL);
6578 
6579 	x86_spec_ctrl_restore_host(vmx->spec_ctrl, 0);
6580 
6581 	/* All fields are clean at this point */
6582 	if (static_branch_unlikely(&enable_evmcs))
6583 		current_evmcs->hv_clean_fields |=
6584 			HV_VMX_ENLIGHTENED_CLEAN_FIELD_ALL;
6585 
6586 	if (static_branch_unlikely(&enable_evmcs))
6587 		current_evmcs->hv_vp_id = vcpu->arch.hyperv.vp_index;
6588 
6589 	/* MSR_IA32_DEBUGCTLMSR is zeroed on vmexit. Restore it if needed */
6590 	if (vmx->host_debugctlmsr)
6591 		update_debugctlmsr(vmx->host_debugctlmsr);
6592 
6593 #ifndef CONFIG_X86_64
6594 	/*
6595 	 * The sysexit path does not restore ds/es, so we must set them to
6596 	 * a reasonable value ourselves.
6597 	 *
6598 	 * We can't defer this to vmx_prepare_switch_to_host() since that
6599 	 * function may be executed in interrupt context, which saves and
6600 	 * restore segments around it, nullifying its effect.
6601 	 */
6602 	loadsegment(ds, __USER_DS);
6603 	loadsegment(es, __USER_DS);
6604 #endif
6605 
6606 	vcpu->arch.regs_avail = ~((1 << VCPU_REGS_RIP) | (1 << VCPU_REGS_RSP)
6607 				  | (1 << VCPU_EXREG_RFLAGS)
6608 				  | (1 << VCPU_EXREG_PDPTR)
6609 				  | (1 << VCPU_EXREG_SEGMENTS)
6610 				  | (1 << VCPU_EXREG_CR3));
6611 	vcpu->arch.regs_dirty = 0;
6612 
6613 	pt_guest_exit(vmx);
6614 
6615 	/*
6616 	 * eager fpu is enabled if PKEY is supported and CR4 is switched
6617 	 * back on host, so it is safe to read guest PKRU from current
6618 	 * XSAVE.
6619 	 */
6620 	if (static_cpu_has(X86_FEATURE_PKU) &&
6621 	    kvm_read_cr4_bits(vcpu, X86_CR4_PKE)) {
6622 		vcpu->arch.pkru = rdpkru();
6623 		if (vcpu->arch.pkru != vmx->host_pkru)
6624 			__write_pkru(vmx->host_pkru);
6625 	}
6626 
6627 	kvm_load_host_xsave_state(vcpu);
6628 
6629 	vmx->nested.nested_run_pending = 0;
6630 	vmx->idt_vectoring_info = 0;
6631 
6632 	vmx->exit_reason = vmx->fail ? 0xdead : vmcs_read32(VM_EXIT_REASON);
6633 	if ((u16)vmx->exit_reason == EXIT_REASON_MCE_DURING_VMENTRY)
6634 		kvm_machine_check();
6635 
6636 	if (vmx->fail || (vmx->exit_reason & VMX_EXIT_REASONS_FAILED_VMENTRY))
6637 		return;
6638 
6639 	vmx->loaded_vmcs->launched = 1;
6640 	vmx->idt_vectoring_info = vmcs_read32(IDT_VECTORING_INFO_FIELD);
6641 
6642 	vmx_recover_nmi_blocking(vmx);
6643 	vmx_complete_interrupts(vmx);
6644 }
6645 
6646 static struct kvm *vmx_vm_alloc(void)
6647 {
6648 	struct kvm_vmx *kvm_vmx = __vmalloc(sizeof(struct kvm_vmx),
6649 					    GFP_KERNEL_ACCOUNT | __GFP_ZERO,
6650 					    PAGE_KERNEL);
6651 	return &kvm_vmx->kvm;
6652 }
6653 
6654 static void vmx_vm_free(struct kvm *kvm)
6655 {
6656 	kfree(kvm->arch.hyperv.hv_pa_pg);
6657 	vfree(to_kvm_vmx(kvm));
6658 }
6659 
6660 static void vmx_free_vcpu(struct kvm_vcpu *vcpu)
6661 {
6662 	struct vcpu_vmx *vmx = to_vmx(vcpu);
6663 
6664 	if (enable_pml)
6665 		vmx_destroy_pml_buffer(vmx);
6666 	free_vpid(vmx->vpid);
6667 	nested_vmx_free_vcpu(vcpu);
6668 	free_loaded_vmcs(vmx->loaded_vmcs);
6669 	kvm_vcpu_uninit(vcpu);
6670 	kmem_cache_free(x86_fpu_cache, vmx->vcpu.arch.user_fpu);
6671 	kmem_cache_free(x86_fpu_cache, vmx->vcpu.arch.guest_fpu);
6672 	kmem_cache_free(kvm_vcpu_cache, vmx);
6673 }
6674 
6675 static struct kvm_vcpu *vmx_create_vcpu(struct kvm *kvm, unsigned int id)
6676 {
6677 	int err;
6678 	struct vcpu_vmx *vmx;
6679 	unsigned long *msr_bitmap;
6680 	int i, cpu;
6681 
6682 	BUILD_BUG_ON_MSG(offsetof(struct vcpu_vmx, vcpu) != 0,
6683 		"struct kvm_vcpu must be at offset 0 for arch usercopy region");
6684 
6685 	vmx = kmem_cache_zalloc(kvm_vcpu_cache, GFP_KERNEL_ACCOUNT);
6686 	if (!vmx)
6687 		return ERR_PTR(-ENOMEM);
6688 
6689 	vmx->vcpu.arch.user_fpu = kmem_cache_zalloc(x86_fpu_cache,
6690 			GFP_KERNEL_ACCOUNT);
6691 	if (!vmx->vcpu.arch.user_fpu) {
6692 		printk(KERN_ERR "kvm: failed to allocate kvm userspace's fpu\n");
6693 		err = -ENOMEM;
6694 		goto free_partial_vcpu;
6695 	}
6696 
6697 	vmx->vcpu.arch.guest_fpu = kmem_cache_zalloc(x86_fpu_cache,
6698 			GFP_KERNEL_ACCOUNT);
6699 	if (!vmx->vcpu.arch.guest_fpu) {
6700 		printk(KERN_ERR "kvm: failed to allocate vcpu's fpu\n");
6701 		err = -ENOMEM;
6702 		goto free_user_fpu;
6703 	}
6704 
6705 	vmx->vpid = allocate_vpid();
6706 
6707 	err = kvm_vcpu_init(&vmx->vcpu, kvm, id);
6708 	if (err)
6709 		goto free_vcpu;
6710 
6711 	err = -ENOMEM;
6712 
6713 	/*
6714 	 * If PML is turned on, failure on enabling PML just results in failure
6715 	 * of creating the vcpu, therefore we can simplify PML logic (by
6716 	 * avoiding dealing with cases, such as enabling PML partially on vcpus
6717 	 * for the guest, etc.
6718 	 */
6719 	if (enable_pml) {
6720 		vmx->pml_pg = alloc_page(GFP_KERNEL_ACCOUNT | __GFP_ZERO);
6721 		if (!vmx->pml_pg)
6722 			goto uninit_vcpu;
6723 	}
6724 
6725 	BUILD_BUG_ON(ARRAY_SIZE(vmx_msr_index) != NR_SHARED_MSRS);
6726 
6727 	for (i = 0; i < ARRAY_SIZE(vmx_msr_index); ++i) {
6728 		u32 index = vmx_msr_index[i];
6729 		u32 data_low, data_high;
6730 		int j = vmx->nmsrs;
6731 
6732 		if (rdmsr_safe(index, &data_low, &data_high) < 0)
6733 			continue;
6734 		if (wrmsr_safe(index, data_low, data_high) < 0)
6735 			continue;
6736 
6737 		vmx->guest_msrs[j].index = i;
6738 		vmx->guest_msrs[j].data = 0;
6739 		switch (index) {
6740 		case MSR_IA32_TSX_CTRL:
6741 			/*
6742 			 * No need to pass TSX_CTRL_CPUID_CLEAR through, so
6743 			 * let's avoid changing CPUID bits under the host
6744 			 * kernel's feet.
6745 			 */
6746 			vmx->guest_msrs[j].mask = ~(u64)TSX_CTRL_CPUID_CLEAR;
6747 			break;
6748 		default:
6749 			vmx->guest_msrs[j].mask = -1ull;
6750 			break;
6751 		}
6752 		++vmx->nmsrs;
6753 	}
6754 
6755 	err = alloc_loaded_vmcs(&vmx->vmcs01);
6756 	if (err < 0)
6757 		goto free_pml;
6758 
6759 	msr_bitmap = vmx->vmcs01.msr_bitmap;
6760 	vmx_disable_intercept_for_msr(msr_bitmap, MSR_IA32_TSC, MSR_TYPE_R);
6761 	vmx_disable_intercept_for_msr(msr_bitmap, MSR_FS_BASE, MSR_TYPE_RW);
6762 	vmx_disable_intercept_for_msr(msr_bitmap, MSR_GS_BASE, MSR_TYPE_RW);
6763 	vmx_disable_intercept_for_msr(msr_bitmap, MSR_KERNEL_GS_BASE, MSR_TYPE_RW);
6764 	vmx_disable_intercept_for_msr(msr_bitmap, MSR_IA32_SYSENTER_CS, MSR_TYPE_RW);
6765 	vmx_disable_intercept_for_msr(msr_bitmap, MSR_IA32_SYSENTER_ESP, MSR_TYPE_RW);
6766 	vmx_disable_intercept_for_msr(msr_bitmap, MSR_IA32_SYSENTER_EIP, MSR_TYPE_RW);
6767 	if (kvm_cstate_in_guest(kvm)) {
6768 		vmx_disable_intercept_for_msr(msr_bitmap, MSR_CORE_C1_RES, MSR_TYPE_R);
6769 		vmx_disable_intercept_for_msr(msr_bitmap, MSR_CORE_C3_RESIDENCY, MSR_TYPE_R);
6770 		vmx_disable_intercept_for_msr(msr_bitmap, MSR_CORE_C6_RESIDENCY, MSR_TYPE_R);
6771 		vmx_disable_intercept_for_msr(msr_bitmap, MSR_CORE_C7_RESIDENCY, MSR_TYPE_R);
6772 	}
6773 	vmx->msr_bitmap_mode = 0;
6774 
6775 	vmx->loaded_vmcs = &vmx->vmcs01;
6776 	cpu = get_cpu();
6777 	vmx_vcpu_load(&vmx->vcpu, cpu);
6778 	vmx->vcpu.cpu = cpu;
6779 	init_vmcs(vmx);
6780 	vmx_vcpu_put(&vmx->vcpu);
6781 	put_cpu();
6782 	if (cpu_need_virtualize_apic_accesses(&vmx->vcpu)) {
6783 		err = alloc_apic_access_page(kvm);
6784 		if (err)
6785 			goto free_vmcs;
6786 	}
6787 
6788 	if (enable_ept && !enable_unrestricted_guest) {
6789 		err = init_rmode_identity_map(kvm);
6790 		if (err)
6791 			goto free_vmcs;
6792 	}
6793 
6794 	if (nested)
6795 		nested_vmx_setup_ctls_msrs(&vmx->nested.msrs,
6796 					   vmx_capability.ept,
6797 					   kvm_vcpu_apicv_active(&vmx->vcpu));
6798 	else
6799 		memset(&vmx->nested.msrs, 0, sizeof(vmx->nested.msrs));
6800 
6801 	vmx->nested.posted_intr_nv = -1;
6802 	vmx->nested.current_vmptr = -1ull;
6803 
6804 	vmx->msr_ia32_feature_control_valid_bits = FEATURE_CONTROL_LOCKED;
6805 
6806 	/*
6807 	 * Enforce invariant: pi_desc.nv is always either POSTED_INTR_VECTOR
6808 	 * or POSTED_INTR_WAKEUP_VECTOR.
6809 	 */
6810 	vmx->pi_desc.nv = POSTED_INTR_VECTOR;
6811 	vmx->pi_desc.sn = 1;
6812 
6813 	vmx->ept_pointer = INVALID_PAGE;
6814 
6815 	return &vmx->vcpu;
6816 
6817 free_vmcs:
6818 	free_loaded_vmcs(vmx->loaded_vmcs);
6819 free_pml:
6820 	vmx_destroy_pml_buffer(vmx);
6821 uninit_vcpu:
6822 	kvm_vcpu_uninit(&vmx->vcpu);
6823 free_vcpu:
6824 	free_vpid(vmx->vpid);
6825 	kmem_cache_free(x86_fpu_cache, vmx->vcpu.arch.guest_fpu);
6826 free_user_fpu:
6827 	kmem_cache_free(x86_fpu_cache, vmx->vcpu.arch.user_fpu);
6828 free_partial_vcpu:
6829 	kmem_cache_free(kvm_vcpu_cache, vmx);
6830 	return ERR_PTR(err);
6831 }
6832 
6833 #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"
6834 #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"
6835 
6836 static int vmx_vm_init(struct kvm *kvm)
6837 {
6838 	spin_lock_init(&to_kvm_vmx(kvm)->ept_pointer_lock);
6839 
6840 	if (!ple_gap)
6841 		kvm->arch.pause_in_guest = true;
6842 
6843 	if (boot_cpu_has(X86_BUG_L1TF) && enable_ept) {
6844 		switch (l1tf_mitigation) {
6845 		case L1TF_MITIGATION_OFF:
6846 		case L1TF_MITIGATION_FLUSH_NOWARN:
6847 			/* 'I explicitly don't care' is set */
6848 			break;
6849 		case L1TF_MITIGATION_FLUSH:
6850 		case L1TF_MITIGATION_FLUSH_NOSMT:
6851 		case L1TF_MITIGATION_FULL:
6852 			/*
6853 			 * Warn upon starting the first VM in a potentially
6854 			 * insecure environment.
6855 			 */
6856 			if (sched_smt_active())
6857 				pr_warn_once(L1TF_MSG_SMT);
6858 			if (l1tf_vmx_mitigation == VMENTER_L1D_FLUSH_NEVER)
6859 				pr_warn_once(L1TF_MSG_L1D);
6860 			break;
6861 		case L1TF_MITIGATION_FULL_FORCE:
6862 			/* Flush is enforced */
6863 			break;
6864 		}
6865 	}
6866 	return 0;
6867 }
6868 
6869 static int __init vmx_check_processor_compat(void)
6870 {
6871 	struct vmcs_config vmcs_conf;
6872 	struct vmx_capability vmx_cap;
6873 
6874 	if (setup_vmcs_config(&vmcs_conf, &vmx_cap) < 0)
6875 		return -EIO;
6876 	if (nested)
6877 		nested_vmx_setup_ctls_msrs(&vmcs_conf.nested, vmx_cap.ept,
6878 					   enable_apicv);
6879 	if (memcmp(&vmcs_config, &vmcs_conf, sizeof(struct vmcs_config)) != 0) {
6880 		printk(KERN_ERR "kvm: CPU %d feature inconsistency!\n",
6881 				smp_processor_id());
6882 		return -EIO;
6883 	}
6884 	return 0;
6885 }
6886 
6887 static u64 vmx_get_mt_mask(struct kvm_vcpu *vcpu, gfn_t gfn, bool is_mmio)
6888 {
6889 	u8 cache;
6890 	u64 ipat = 0;
6891 
6892 	/* For VT-d and EPT combination
6893 	 * 1. MMIO: always map as UC
6894 	 * 2. EPT with VT-d:
6895 	 *   a. VT-d without snooping control feature: can't guarantee the
6896 	 *	result, try to trust guest.
6897 	 *   b. VT-d with snooping control feature: snooping control feature of
6898 	 *	VT-d engine can guarantee the cache correctness. Just set it
6899 	 *	to WB to keep consistent with host. So the same as item 3.
6900 	 * 3. EPT without VT-d: always map as WB and set IPAT=1 to keep
6901 	 *    consistent with host MTRR
6902 	 */
6903 	if (is_mmio) {
6904 		cache = MTRR_TYPE_UNCACHABLE;
6905 		goto exit;
6906 	}
6907 
6908 	if (!kvm_arch_has_noncoherent_dma(vcpu->kvm)) {
6909 		ipat = VMX_EPT_IPAT_BIT;
6910 		cache = MTRR_TYPE_WRBACK;
6911 		goto exit;
6912 	}
6913 
6914 	if (kvm_read_cr0(vcpu) & X86_CR0_CD) {
6915 		ipat = VMX_EPT_IPAT_BIT;
6916 		if (kvm_check_has_quirk(vcpu->kvm, KVM_X86_QUIRK_CD_NW_CLEARED))
6917 			cache = MTRR_TYPE_WRBACK;
6918 		else
6919 			cache = MTRR_TYPE_UNCACHABLE;
6920 		goto exit;
6921 	}
6922 
6923 	cache = kvm_mtrr_get_guest_memory_type(vcpu, gfn);
6924 
6925 exit:
6926 	return (cache << VMX_EPT_MT_EPTE_SHIFT) | ipat;
6927 }
6928 
6929 static int vmx_get_lpage_level(void)
6930 {
6931 	if (enable_ept && !cpu_has_vmx_ept_1g_page())
6932 		return PT_DIRECTORY_LEVEL;
6933 	else
6934 		/* For shadow and EPT supported 1GB page */
6935 		return PT_PDPE_LEVEL;
6936 }
6937 
6938 static void vmcs_set_secondary_exec_control(struct vcpu_vmx *vmx)
6939 {
6940 	/*
6941 	 * These bits in the secondary execution controls field
6942 	 * are dynamic, the others are mostly based on the hypervisor
6943 	 * architecture and the guest's CPUID.  Do not touch the
6944 	 * dynamic bits.
6945 	 */
6946 	u32 mask =
6947 		SECONDARY_EXEC_SHADOW_VMCS |
6948 		SECONDARY_EXEC_VIRTUALIZE_X2APIC_MODE |
6949 		SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES |
6950 		SECONDARY_EXEC_DESC;
6951 
6952 	u32 new_ctl = vmx->secondary_exec_control;
6953 	u32 cur_ctl = secondary_exec_controls_get(vmx);
6954 
6955 	secondary_exec_controls_set(vmx, (new_ctl & ~mask) | (cur_ctl & mask));
6956 }
6957 
6958 /*
6959  * Generate MSR_IA32_VMX_CR{0,4}_FIXED1 according to CPUID. Only set bits
6960  * (indicating "allowed-1") if they are supported in the guest's CPUID.
6961  */
6962 static void nested_vmx_cr_fixed1_bits_update(struct kvm_vcpu *vcpu)
6963 {
6964 	struct vcpu_vmx *vmx = to_vmx(vcpu);
6965 	struct kvm_cpuid_entry2 *entry;
6966 
6967 	vmx->nested.msrs.cr0_fixed1 = 0xffffffff;
6968 	vmx->nested.msrs.cr4_fixed1 = X86_CR4_PCE;
6969 
6970 #define cr4_fixed1_update(_cr4_mask, _reg, _cpuid_mask) do {		\
6971 	if (entry && (entry->_reg & (_cpuid_mask)))			\
6972 		vmx->nested.msrs.cr4_fixed1 |= (_cr4_mask);	\
6973 } while (0)
6974 
6975 	entry = kvm_find_cpuid_entry(vcpu, 0x1, 0);
6976 	cr4_fixed1_update(X86_CR4_VME,        edx, bit(X86_FEATURE_VME));
6977 	cr4_fixed1_update(X86_CR4_PVI,        edx, bit(X86_FEATURE_VME));
6978 	cr4_fixed1_update(X86_CR4_TSD,        edx, bit(X86_FEATURE_TSC));
6979 	cr4_fixed1_update(X86_CR4_DE,         edx, bit(X86_FEATURE_DE));
6980 	cr4_fixed1_update(X86_CR4_PSE,        edx, bit(X86_FEATURE_PSE));
6981 	cr4_fixed1_update(X86_CR4_PAE,        edx, bit(X86_FEATURE_PAE));
6982 	cr4_fixed1_update(X86_CR4_MCE,        edx, bit(X86_FEATURE_MCE));
6983 	cr4_fixed1_update(X86_CR4_PGE,        edx, bit(X86_FEATURE_PGE));
6984 	cr4_fixed1_update(X86_CR4_OSFXSR,     edx, bit(X86_FEATURE_FXSR));
6985 	cr4_fixed1_update(X86_CR4_OSXMMEXCPT, edx, bit(X86_FEATURE_XMM));
6986 	cr4_fixed1_update(X86_CR4_VMXE,       ecx, bit(X86_FEATURE_VMX));
6987 	cr4_fixed1_update(X86_CR4_SMXE,       ecx, bit(X86_FEATURE_SMX));
6988 	cr4_fixed1_update(X86_CR4_PCIDE,      ecx, bit(X86_FEATURE_PCID));
6989 	cr4_fixed1_update(X86_CR4_OSXSAVE,    ecx, bit(X86_FEATURE_XSAVE));
6990 
6991 	entry = kvm_find_cpuid_entry(vcpu, 0x7, 0);
6992 	cr4_fixed1_update(X86_CR4_FSGSBASE,   ebx, bit(X86_FEATURE_FSGSBASE));
6993 	cr4_fixed1_update(X86_CR4_SMEP,       ebx, bit(X86_FEATURE_SMEP));
6994 	cr4_fixed1_update(X86_CR4_SMAP,       ebx, bit(X86_FEATURE_SMAP));
6995 	cr4_fixed1_update(X86_CR4_PKE,        ecx, bit(X86_FEATURE_PKU));
6996 	cr4_fixed1_update(X86_CR4_UMIP,       ecx, bit(X86_FEATURE_UMIP));
6997 	cr4_fixed1_update(X86_CR4_LA57,       ecx, bit(X86_FEATURE_LA57));
6998 
6999 #undef cr4_fixed1_update
7000 }
7001 
7002 static void nested_vmx_entry_exit_ctls_update(struct kvm_vcpu *vcpu)
7003 {
7004 	struct vcpu_vmx *vmx = to_vmx(vcpu);
7005 
7006 	if (kvm_mpx_supported()) {
7007 		bool mpx_enabled = guest_cpuid_has(vcpu, X86_FEATURE_MPX);
7008 
7009 		if (mpx_enabled) {
7010 			vmx->nested.msrs.entry_ctls_high |= VM_ENTRY_LOAD_BNDCFGS;
7011 			vmx->nested.msrs.exit_ctls_high |= VM_EXIT_CLEAR_BNDCFGS;
7012 		} else {
7013 			vmx->nested.msrs.entry_ctls_high &= ~VM_ENTRY_LOAD_BNDCFGS;
7014 			vmx->nested.msrs.exit_ctls_high &= ~VM_EXIT_CLEAR_BNDCFGS;
7015 		}
7016 	}
7017 }
7018 
7019 static void update_intel_pt_cfg(struct kvm_vcpu *vcpu)
7020 {
7021 	struct vcpu_vmx *vmx = to_vmx(vcpu);
7022 	struct kvm_cpuid_entry2 *best = NULL;
7023 	int i;
7024 
7025 	for (i = 0; i < PT_CPUID_LEAVES; i++) {
7026 		best = kvm_find_cpuid_entry(vcpu, 0x14, i);
7027 		if (!best)
7028 			return;
7029 		vmx->pt_desc.caps[CPUID_EAX + i*PT_CPUID_REGS_NUM] = best->eax;
7030 		vmx->pt_desc.caps[CPUID_EBX + i*PT_CPUID_REGS_NUM] = best->ebx;
7031 		vmx->pt_desc.caps[CPUID_ECX + i*PT_CPUID_REGS_NUM] = best->ecx;
7032 		vmx->pt_desc.caps[CPUID_EDX + i*PT_CPUID_REGS_NUM] = best->edx;
7033 	}
7034 
7035 	/* Get the number of configurable Address Ranges for filtering */
7036 	vmx->pt_desc.addr_range = intel_pt_validate_cap(vmx->pt_desc.caps,
7037 						PT_CAP_num_address_ranges);
7038 
7039 	/* Initialize and clear the no dependency bits */
7040 	vmx->pt_desc.ctl_bitmask = ~(RTIT_CTL_TRACEEN | RTIT_CTL_OS |
7041 			RTIT_CTL_USR | RTIT_CTL_TSC_EN | RTIT_CTL_DISRETC);
7042 
7043 	/*
7044 	 * If CPUID.(EAX=14H,ECX=0):EBX[0]=1 CR3Filter can be set otherwise
7045 	 * will inject an #GP
7046 	 */
7047 	if (intel_pt_validate_cap(vmx->pt_desc.caps, PT_CAP_cr3_filtering))
7048 		vmx->pt_desc.ctl_bitmask &= ~RTIT_CTL_CR3EN;
7049 
7050 	/*
7051 	 * If CPUID.(EAX=14H,ECX=0):EBX[1]=1 CYCEn, CycThresh and
7052 	 * PSBFreq can be set
7053 	 */
7054 	if (intel_pt_validate_cap(vmx->pt_desc.caps, PT_CAP_psb_cyc))
7055 		vmx->pt_desc.ctl_bitmask &= ~(RTIT_CTL_CYCLEACC |
7056 				RTIT_CTL_CYC_THRESH | RTIT_CTL_PSB_FREQ);
7057 
7058 	/*
7059 	 * If CPUID.(EAX=14H,ECX=0):EBX[3]=1 MTCEn BranchEn and
7060 	 * MTCFreq can be set
7061 	 */
7062 	if (intel_pt_validate_cap(vmx->pt_desc.caps, PT_CAP_mtc))
7063 		vmx->pt_desc.ctl_bitmask &= ~(RTIT_CTL_MTC_EN |
7064 				RTIT_CTL_BRANCH_EN | RTIT_CTL_MTC_RANGE);
7065 
7066 	/* If CPUID.(EAX=14H,ECX=0):EBX[4]=1 FUPonPTW and PTWEn can be set */
7067 	if (intel_pt_validate_cap(vmx->pt_desc.caps, PT_CAP_ptwrite))
7068 		vmx->pt_desc.ctl_bitmask &= ~(RTIT_CTL_FUP_ON_PTW |
7069 							RTIT_CTL_PTW_EN);
7070 
7071 	/* If CPUID.(EAX=14H,ECX=0):EBX[5]=1 PwrEvEn can be set */
7072 	if (intel_pt_validate_cap(vmx->pt_desc.caps, PT_CAP_power_event_trace))
7073 		vmx->pt_desc.ctl_bitmask &= ~RTIT_CTL_PWR_EVT_EN;
7074 
7075 	/* If CPUID.(EAX=14H,ECX=0):ECX[0]=1 ToPA can be set */
7076 	if (intel_pt_validate_cap(vmx->pt_desc.caps, PT_CAP_topa_output))
7077 		vmx->pt_desc.ctl_bitmask &= ~RTIT_CTL_TOPA;
7078 
7079 	/* If CPUID.(EAX=14H,ECX=0):ECX[3]=1 FabircEn can be set */
7080 	if (intel_pt_validate_cap(vmx->pt_desc.caps, PT_CAP_output_subsys))
7081 		vmx->pt_desc.ctl_bitmask &= ~RTIT_CTL_FABRIC_EN;
7082 
7083 	/* unmask address range configure area */
7084 	for (i = 0; i < vmx->pt_desc.addr_range; i++)
7085 		vmx->pt_desc.ctl_bitmask &= ~(0xfULL << (32 + i * 4));
7086 }
7087 
7088 static void vmx_cpuid_update(struct kvm_vcpu *vcpu)
7089 {
7090 	struct vcpu_vmx *vmx = to_vmx(vcpu);
7091 
7092 	/* xsaves_enabled is recomputed in vmx_compute_secondary_exec_control(). */
7093 	vcpu->arch.xsaves_enabled = false;
7094 
7095 	if (cpu_has_secondary_exec_ctrls()) {
7096 		vmx_compute_secondary_exec_control(vmx);
7097 		vmcs_set_secondary_exec_control(vmx);
7098 	}
7099 
7100 	if (nested_vmx_allowed(vcpu))
7101 		to_vmx(vcpu)->msr_ia32_feature_control_valid_bits |=
7102 			FEATURE_CONTROL_VMXON_ENABLED_INSIDE_SMX |
7103 			FEATURE_CONTROL_VMXON_ENABLED_OUTSIDE_SMX;
7104 	else
7105 		to_vmx(vcpu)->msr_ia32_feature_control_valid_bits &=
7106 			~(FEATURE_CONTROL_VMXON_ENABLED_INSIDE_SMX |
7107 			  FEATURE_CONTROL_VMXON_ENABLED_OUTSIDE_SMX);
7108 
7109 	if (nested_vmx_allowed(vcpu)) {
7110 		nested_vmx_cr_fixed1_bits_update(vcpu);
7111 		nested_vmx_entry_exit_ctls_update(vcpu);
7112 	}
7113 
7114 	if (boot_cpu_has(X86_FEATURE_INTEL_PT) &&
7115 			guest_cpuid_has(vcpu, X86_FEATURE_INTEL_PT))
7116 		update_intel_pt_cfg(vcpu);
7117 
7118 	if (boot_cpu_has(X86_FEATURE_RTM)) {
7119 		struct shared_msr_entry *msr;
7120 		msr = find_msr_entry(vmx, MSR_IA32_TSX_CTRL);
7121 		if (msr) {
7122 			bool enabled = guest_cpuid_has(vcpu, X86_FEATURE_RTM);
7123 			vmx_set_guest_msr(vmx, msr, enabled ? 0 : TSX_CTRL_RTM_DISABLE);
7124 		}
7125 	}
7126 }
7127 
7128 static void vmx_set_supported_cpuid(u32 func, struct kvm_cpuid_entry2 *entry)
7129 {
7130 	if (func == 1 && nested)
7131 		entry->ecx |= bit(X86_FEATURE_VMX);
7132 }
7133 
7134 static void vmx_request_immediate_exit(struct kvm_vcpu *vcpu)
7135 {
7136 	to_vmx(vcpu)->req_immediate_exit = true;
7137 }
7138 
7139 static int vmx_check_intercept(struct kvm_vcpu *vcpu,
7140 			       struct x86_instruction_info *info,
7141 			       enum x86_intercept_stage stage)
7142 {
7143 	struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
7144 	struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
7145 
7146 	/*
7147 	 * RDPID causes #UD if disabled through secondary execution controls.
7148 	 * Because it is marked as EmulateOnUD, we need to intercept it here.
7149 	 */
7150 	if (info->intercept == x86_intercept_rdtscp &&
7151 	    !nested_cpu_has2(vmcs12, SECONDARY_EXEC_RDTSCP)) {
7152 		ctxt->exception.vector = UD_VECTOR;
7153 		ctxt->exception.error_code_valid = false;
7154 		return X86EMUL_PROPAGATE_FAULT;
7155 	}
7156 
7157 	/* TODO: check more intercepts... */
7158 	return X86EMUL_CONTINUE;
7159 }
7160 
7161 #ifdef CONFIG_X86_64
7162 /* (a << shift) / divisor, return 1 if overflow otherwise 0 */
7163 static inline int u64_shl_div_u64(u64 a, unsigned int shift,
7164 				  u64 divisor, u64 *result)
7165 {
7166 	u64 low = a << shift, high = a >> (64 - shift);
7167 
7168 	/* To avoid the overflow on divq */
7169 	if (high >= divisor)
7170 		return 1;
7171 
7172 	/* Low hold the result, high hold rem which is discarded */
7173 	asm("divq %2\n\t" : "=a" (low), "=d" (high) :
7174 	    "rm" (divisor), "0" (low), "1" (high));
7175 	*result = low;
7176 
7177 	return 0;
7178 }
7179 
7180 static int vmx_set_hv_timer(struct kvm_vcpu *vcpu, u64 guest_deadline_tsc,
7181 			    bool *expired)
7182 {
7183 	struct vcpu_vmx *vmx;
7184 	u64 tscl, guest_tscl, delta_tsc, lapic_timer_advance_cycles;
7185 	struct kvm_timer *ktimer = &vcpu->arch.apic->lapic_timer;
7186 
7187 	if (kvm_mwait_in_guest(vcpu->kvm) ||
7188 		kvm_can_post_timer_interrupt(vcpu))
7189 		return -EOPNOTSUPP;
7190 
7191 	vmx = to_vmx(vcpu);
7192 	tscl = rdtsc();
7193 	guest_tscl = kvm_read_l1_tsc(vcpu, tscl);
7194 	delta_tsc = max(guest_deadline_tsc, guest_tscl) - guest_tscl;
7195 	lapic_timer_advance_cycles = nsec_to_cycles(vcpu,
7196 						    ktimer->timer_advance_ns);
7197 
7198 	if (delta_tsc > lapic_timer_advance_cycles)
7199 		delta_tsc -= lapic_timer_advance_cycles;
7200 	else
7201 		delta_tsc = 0;
7202 
7203 	/* Convert to host delta tsc if tsc scaling is enabled */
7204 	if (vcpu->arch.tsc_scaling_ratio != kvm_default_tsc_scaling_ratio &&
7205 	    delta_tsc && u64_shl_div_u64(delta_tsc,
7206 				kvm_tsc_scaling_ratio_frac_bits,
7207 				vcpu->arch.tsc_scaling_ratio, &delta_tsc))
7208 		return -ERANGE;
7209 
7210 	/*
7211 	 * If the delta tsc can't fit in the 32 bit after the multi shift,
7212 	 * we can't use the preemption timer.
7213 	 * It's possible that it fits on later vmentries, but checking
7214 	 * on every vmentry is costly so we just use an hrtimer.
7215 	 */
7216 	if (delta_tsc >> (cpu_preemption_timer_multi + 32))
7217 		return -ERANGE;
7218 
7219 	vmx->hv_deadline_tsc = tscl + delta_tsc;
7220 	*expired = !delta_tsc;
7221 	return 0;
7222 }
7223 
7224 static void vmx_cancel_hv_timer(struct kvm_vcpu *vcpu)
7225 {
7226 	to_vmx(vcpu)->hv_deadline_tsc = -1;
7227 }
7228 #endif
7229 
7230 static void vmx_sched_in(struct kvm_vcpu *vcpu, int cpu)
7231 {
7232 	if (!kvm_pause_in_guest(vcpu->kvm))
7233 		shrink_ple_window(vcpu);
7234 }
7235 
7236 static void vmx_slot_enable_log_dirty(struct kvm *kvm,
7237 				     struct kvm_memory_slot *slot)
7238 {
7239 	kvm_mmu_slot_leaf_clear_dirty(kvm, slot);
7240 	kvm_mmu_slot_largepage_remove_write_access(kvm, slot);
7241 }
7242 
7243 static void vmx_slot_disable_log_dirty(struct kvm *kvm,
7244 				       struct kvm_memory_slot *slot)
7245 {
7246 	kvm_mmu_slot_set_dirty(kvm, slot);
7247 }
7248 
7249 static void vmx_flush_log_dirty(struct kvm *kvm)
7250 {
7251 	kvm_flush_pml_buffers(kvm);
7252 }
7253 
7254 static int vmx_write_pml_buffer(struct kvm_vcpu *vcpu)
7255 {
7256 	struct vmcs12 *vmcs12;
7257 	struct vcpu_vmx *vmx = to_vmx(vcpu);
7258 	gpa_t gpa, dst;
7259 
7260 	if (is_guest_mode(vcpu)) {
7261 		WARN_ON_ONCE(vmx->nested.pml_full);
7262 
7263 		/*
7264 		 * Check if PML is enabled for the nested guest.
7265 		 * Whether eptp bit 6 is set is already checked
7266 		 * as part of A/D emulation.
7267 		 */
7268 		vmcs12 = get_vmcs12(vcpu);
7269 		if (!nested_cpu_has_pml(vmcs12))
7270 			return 0;
7271 
7272 		if (vmcs12->guest_pml_index >= PML_ENTITY_NUM) {
7273 			vmx->nested.pml_full = true;
7274 			return 1;
7275 		}
7276 
7277 		gpa = vmcs_read64(GUEST_PHYSICAL_ADDRESS) & ~0xFFFull;
7278 		dst = vmcs12->pml_address + sizeof(u64) * vmcs12->guest_pml_index;
7279 
7280 		if (kvm_write_guest_page(vcpu->kvm, gpa_to_gfn(dst), &gpa,
7281 					 offset_in_page(dst), sizeof(gpa)))
7282 			return 0;
7283 
7284 		vmcs12->guest_pml_index--;
7285 	}
7286 
7287 	return 0;
7288 }
7289 
7290 static void vmx_enable_log_dirty_pt_masked(struct kvm *kvm,
7291 					   struct kvm_memory_slot *memslot,
7292 					   gfn_t offset, unsigned long mask)
7293 {
7294 	kvm_mmu_clear_dirty_pt_masked(kvm, memslot, offset, mask);
7295 }
7296 
7297 static void __pi_post_block(struct kvm_vcpu *vcpu)
7298 {
7299 	struct pi_desc *pi_desc = vcpu_to_pi_desc(vcpu);
7300 	struct pi_desc old, new;
7301 	unsigned int dest;
7302 
7303 	do {
7304 		old.control = new.control = pi_desc->control;
7305 		WARN(old.nv != POSTED_INTR_WAKEUP_VECTOR,
7306 		     "Wakeup handler not enabled while the VCPU is blocked\n");
7307 
7308 		dest = cpu_physical_id(vcpu->cpu);
7309 
7310 		if (x2apic_enabled())
7311 			new.ndst = dest;
7312 		else
7313 			new.ndst = (dest << 8) & 0xFF00;
7314 
7315 		/* set 'NV' to 'notification vector' */
7316 		new.nv = POSTED_INTR_VECTOR;
7317 	} while (cmpxchg64(&pi_desc->control, old.control,
7318 			   new.control) != old.control);
7319 
7320 	if (!WARN_ON_ONCE(vcpu->pre_pcpu == -1)) {
7321 		spin_lock(&per_cpu(blocked_vcpu_on_cpu_lock, vcpu->pre_pcpu));
7322 		list_del(&vcpu->blocked_vcpu_list);
7323 		spin_unlock(&per_cpu(blocked_vcpu_on_cpu_lock, vcpu->pre_pcpu));
7324 		vcpu->pre_pcpu = -1;
7325 	}
7326 }
7327 
7328 /*
7329  * This routine does the following things for vCPU which is going
7330  * to be blocked if VT-d PI is enabled.
7331  * - Store the vCPU to the wakeup list, so when interrupts happen
7332  *   we can find the right vCPU to wake up.
7333  * - Change the Posted-interrupt descriptor as below:
7334  *      'NDST' <-- vcpu->pre_pcpu
7335  *      'NV' <-- POSTED_INTR_WAKEUP_VECTOR
7336  * - If 'ON' is set during this process, which means at least one
7337  *   interrupt is posted for this vCPU, we cannot block it, in
7338  *   this case, return 1, otherwise, return 0.
7339  *
7340  */
7341 static int pi_pre_block(struct kvm_vcpu *vcpu)
7342 {
7343 	unsigned int dest;
7344 	struct pi_desc old, new;
7345 	struct pi_desc *pi_desc = vcpu_to_pi_desc(vcpu);
7346 
7347 	if (!kvm_arch_has_assigned_device(vcpu->kvm) ||
7348 		!irq_remapping_cap(IRQ_POSTING_CAP)  ||
7349 		!kvm_vcpu_apicv_active(vcpu))
7350 		return 0;
7351 
7352 	WARN_ON(irqs_disabled());
7353 	local_irq_disable();
7354 	if (!WARN_ON_ONCE(vcpu->pre_pcpu != -1)) {
7355 		vcpu->pre_pcpu = vcpu->cpu;
7356 		spin_lock(&per_cpu(blocked_vcpu_on_cpu_lock, vcpu->pre_pcpu));
7357 		list_add_tail(&vcpu->blocked_vcpu_list,
7358 			      &per_cpu(blocked_vcpu_on_cpu,
7359 				       vcpu->pre_pcpu));
7360 		spin_unlock(&per_cpu(blocked_vcpu_on_cpu_lock, vcpu->pre_pcpu));
7361 	}
7362 
7363 	do {
7364 		old.control = new.control = pi_desc->control;
7365 
7366 		WARN((pi_desc->sn == 1),
7367 		     "Warning: SN field of posted-interrupts "
7368 		     "is set before blocking\n");
7369 
7370 		/*
7371 		 * Since vCPU can be preempted during this process,
7372 		 * vcpu->cpu could be different with pre_pcpu, we
7373 		 * need to set pre_pcpu as the destination of wakeup
7374 		 * notification event, then we can find the right vCPU
7375 		 * to wakeup in wakeup handler if interrupts happen
7376 		 * when the vCPU is in blocked state.
7377 		 */
7378 		dest = cpu_physical_id(vcpu->pre_pcpu);
7379 
7380 		if (x2apic_enabled())
7381 			new.ndst = dest;
7382 		else
7383 			new.ndst = (dest << 8) & 0xFF00;
7384 
7385 		/* set 'NV' to 'wakeup vector' */
7386 		new.nv = POSTED_INTR_WAKEUP_VECTOR;
7387 	} while (cmpxchg64(&pi_desc->control, old.control,
7388 			   new.control) != old.control);
7389 
7390 	/* We should not block the vCPU if an interrupt is posted for it.  */
7391 	if (pi_test_on(pi_desc) == 1)
7392 		__pi_post_block(vcpu);
7393 
7394 	local_irq_enable();
7395 	return (vcpu->pre_pcpu == -1);
7396 }
7397 
7398 static int vmx_pre_block(struct kvm_vcpu *vcpu)
7399 {
7400 	if (pi_pre_block(vcpu))
7401 		return 1;
7402 
7403 	if (kvm_lapic_hv_timer_in_use(vcpu))
7404 		kvm_lapic_switch_to_sw_timer(vcpu);
7405 
7406 	return 0;
7407 }
7408 
7409 static void pi_post_block(struct kvm_vcpu *vcpu)
7410 {
7411 	if (vcpu->pre_pcpu == -1)
7412 		return;
7413 
7414 	WARN_ON(irqs_disabled());
7415 	local_irq_disable();
7416 	__pi_post_block(vcpu);
7417 	local_irq_enable();
7418 }
7419 
7420 static void vmx_post_block(struct kvm_vcpu *vcpu)
7421 {
7422 	if (kvm_x86_ops->set_hv_timer)
7423 		kvm_lapic_switch_to_hv_timer(vcpu);
7424 
7425 	pi_post_block(vcpu);
7426 }
7427 
7428 /*
7429  * vmx_update_pi_irte - set IRTE for Posted-Interrupts
7430  *
7431  * @kvm: kvm
7432  * @host_irq: host irq of the interrupt
7433  * @guest_irq: gsi of the interrupt
7434  * @set: set or unset PI
7435  * returns 0 on success, < 0 on failure
7436  */
7437 static int vmx_update_pi_irte(struct kvm *kvm, unsigned int host_irq,
7438 			      uint32_t guest_irq, bool set)
7439 {
7440 	struct kvm_kernel_irq_routing_entry *e;
7441 	struct kvm_irq_routing_table *irq_rt;
7442 	struct kvm_lapic_irq irq;
7443 	struct kvm_vcpu *vcpu;
7444 	struct vcpu_data vcpu_info;
7445 	int idx, ret = 0;
7446 
7447 	if (!kvm_arch_has_assigned_device(kvm) ||
7448 		!irq_remapping_cap(IRQ_POSTING_CAP) ||
7449 		!kvm_vcpu_apicv_active(kvm->vcpus[0]))
7450 		return 0;
7451 
7452 	idx = srcu_read_lock(&kvm->irq_srcu);
7453 	irq_rt = srcu_dereference(kvm->irq_routing, &kvm->irq_srcu);
7454 	if (guest_irq >= irq_rt->nr_rt_entries ||
7455 	    hlist_empty(&irq_rt->map[guest_irq])) {
7456 		pr_warn_once("no route for guest_irq %u/%u (broken user space?)\n",
7457 			     guest_irq, irq_rt->nr_rt_entries);
7458 		goto out;
7459 	}
7460 
7461 	hlist_for_each_entry(e, &irq_rt->map[guest_irq], link) {
7462 		if (e->type != KVM_IRQ_ROUTING_MSI)
7463 			continue;
7464 		/*
7465 		 * VT-d PI cannot support posting multicast/broadcast
7466 		 * interrupts to a vCPU, we still use interrupt remapping
7467 		 * for these kind of interrupts.
7468 		 *
7469 		 * For lowest-priority interrupts, we only support
7470 		 * those with single CPU as the destination, e.g. user
7471 		 * configures the interrupts via /proc/irq or uses
7472 		 * irqbalance to make the interrupts single-CPU.
7473 		 *
7474 		 * We will support full lowest-priority interrupt later.
7475 		 *
7476 		 * In addition, we can only inject generic interrupts using
7477 		 * the PI mechanism, refuse to route others through it.
7478 		 */
7479 
7480 		kvm_set_msi_irq(kvm, e, &irq);
7481 		if (!kvm_intr_is_single_vcpu(kvm, &irq, &vcpu) ||
7482 		    !kvm_irq_is_postable(&irq)) {
7483 			/*
7484 			 * Make sure the IRTE is in remapped mode if
7485 			 * we don't handle it in posted mode.
7486 			 */
7487 			ret = irq_set_vcpu_affinity(host_irq, NULL);
7488 			if (ret < 0) {
7489 				printk(KERN_INFO
7490 				   "failed to back to remapped mode, irq: %u\n",
7491 				   host_irq);
7492 				goto out;
7493 			}
7494 
7495 			continue;
7496 		}
7497 
7498 		vcpu_info.pi_desc_addr = __pa(vcpu_to_pi_desc(vcpu));
7499 		vcpu_info.vector = irq.vector;
7500 
7501 		trace_kvm_pi_irte_update(host_irq, vcpu->vcpu_id, e->gsi,
7502 				vcpu_info.vector, vcpu_info.pi_desc_addr, set);
7503 
7504 		if (set)
7505 			ret = irq_set_vcpu_affinity(host_irq, &vcpu_info);
7506 		else
7507 			ret = irq_set_vcpu_affinity(host_irq, NULL);
7508 
7509 		if (ret < 0) {
7510 			printk(KERN_INFO "%s: failed to update PI IRTE\n",
7511 					__func__);
7512 			goto out;
7513 		}
7514 	}
7515 
7516 	ret = 0;
7517 out:
7518 	srcu_read_unlock(&kvm->irq_srcu, idx);
7519 	return ret;
7520 }
7521 
7522 static void vmx_setup_mce(struct kvm_vcpu *vcpu)
7523 {
7524 	if (vcpu->arch.mcg_cap & MCG_LMCE_P)
7525 		to_vmx(vcpu)->msr_ia32_feature_control_valid_bits |=
7526 			FEATURE_CONTROL_LMCE;
7527 	else
7528 		to_vmx(vcpu)->msr_ia32_feature_control_valid_bits &=
7529 			~FEATURE_CONTROL_LMCE;
7530 }
7531 
7532 static int vmx_smi_allowed(struct kvm_vcpu *vcpu)
7533 {
7534 	/* we need a nested vmexit to enter SMM, postpone if run is pending */
7535 	if (to_vmx(vcpu)->nested.nested_run_pending)
7536 		return 0;
7537 	return 1;
7538 }
7539 
7540 static int vmx_pre_enter_smm(struct kvm_vcpu *vcpu, char *smstate)
7541 {
7542 	struct vcpu_vmx *vmx = to_vmx(vcpu);
7543 
7544 	vmx->nested.smm.guest_mode = is_guest_mode(vcpu);
7545 	if (vmx->nested.smm.guest_mode)
7546 		nested_vmx_vmexit(vcpu, -1, 0, 0);
7547 
7548 	vmx->nested.smm.vmxon = vmx->nested.vmxon;
7549 	vmx->nested.vmxon = false;
7550 	vmx_clear_hlt(vcpu);
7551 	return 0;
7552 }
7553 
7554 static int vmx_pre_leave_smm(struct kvm_vcpu *vcpu, const char *smstate)
7555 {
7556 	struct vcpu_vmx *vmx = to_vmx(vcpu);
7557 	int ret;
7558 
7559 	if (vmx->nested.smm.vmxon) {
7560 		vmx->nested.vmxon = true;
7561 		vmx->nested.smm.vmxon = false;
7562 	}
7563 
7564 	if (vmx->nested.smm.guest_mode) {
7565 		ret = nested_vmx_enter_non_root_mode(vcpu, false);
7566 		if (ret)
7567 			return ret;
7568 
7569 		vmx->nested.smm.guest_mode = false;
7570 	}
7571 	return 0;
7572 }
7573 
7574 static int enable_smi_window(struct kvm_vcpu *vcpu)
7575 {
7576 	return 0;
7577 }
7578 
7579 static bool vmx_need_emulation_on_page_fault(struct kvm_vcpu *vcpu)
7580 {
7581 	return false;
7582 }
7583 
7584 static bool vmx_apic_init_signal_blocked(struct kvm_vcpu *vcpu)
7585 {
7586 	return to_vmx(vcpu)->nested.vmxon;
7587 }
7588 
7589 static __init int hardware_setup(void)
7590 {
7591 	unsigned long host_bndcfgs;
7592 	struct desc_ptr dt;
7593 	int r, i;
7594 
7595 	rdmsrl_safe(MSR_EFER, &host_efer);
7596 
7597 	store_idt(&dt);
7598 	host_idt_base = dt.address;
7599 
7600 	for (i = 0; i < ARRAY_SIZE(vmx_msr_index); ++i)
7601 		kvm_define_shared_msr(i, vmx_msr_index[i]);
7602 
7603 	if (setup_vmcs_config(&vmcs_config, &vmx_capability) < 0)
7604 		return -EIO;
7605 
7606 	if (boot_cpu_has(X86_FEATURE_NX))
7607 		kvm_enable_efer_bits(EFER_NX);
7608 
7609 	if (boot_cpu_has(X86_FEATURE_MPX)) {
7610 		rdmsrl(MSR_IA32_BNDCFGS, host_bndcfgs);
7611 		WARN_ONCE(host_bndcfgs, "KVM: BNDCFGS in host will be lost");
7612 	}
7613 
7614 	if (!cpu_has_vmx_vpid() || !cpu_has_vmx_invvpid() ||
7615 	    !(cpu_has_vmx_invvpid_single() || cpu_has_vmx_invvpid_global()))
7616 		enable_vpid = 0;
7617 
7618 	if (!cpu_has_vmx_ept() ||
7619 	    !cpu_has_vmx_ept_4levels() ||
7620 	    !cpu_has_vmx_ept_mt_wb() ||
7621 	    !cpu_has_vmx_invept_global())
7622 		enable_ept = 0;
7623 
7624 	if (!cpu_has_vmx_ept_ad_bits() || !enable_ept)
7625 		enable_ept_ad_bits = 0;
7626 
7627 	if (!cpu_has_vmx_unrestricted_guest() || !enable_ept)
7628 		enable_unrestricted_guest = 0;
7629 
7630 	if (!cpu_has_vmx_flexpriority())
7631 		flexpriority_enabled = 0;
7632 
7633 	if (!cpu_has_virtual_nmis())
7634 		enable_vnmi = 0;
7635 
7636 	/*
7637 	 * set_apic_access_page_addr() is used to reload apic access
7638 	 * page upon invalidation.  No need to do anything if not
7639 	 * using the APIC_ACCESS_ADDR VMCS field.
7640 	 */
7641 	if (!flexpriority_enabled)
7642 		kvm_x86_ops->set_apic_access_page_addr = NULL;
7643 
7644 	if (!cpu_has_vmx_tpr_shadow())
7645 		kvm_x86_ops->update_cr8_intercept = NULL;
7646 
7647 	if (enable_ept && !cpu_has_vmx_ept_2m_page())
7648 		kvm_disable_largepages();
7649 
7650 #if IS_ENABLED(CONFIG_HYPERV)
7651 	if (ms_hyperv.nested_features & HV_X64_NESTED_GUEST_MAPPING_FLUSH
7652 	    && enable_ept) {
7653 		kvm_x86_ops->tlb_remote_flush = hv_remote_flush_tlb;
7654 		kvm_x86_ops->tlb_remote_flush_with_range =
7655 				hv_remote_flush_tlb_with_range;
7656 	}
7657 #endif
7658 
7659 	if (!cpu_has_vmx_ple()) {
7660 		ple_gap = 0;
7661 		ple_window = 0;
7662 		ple_window_grow = 0;
7663 		ple_window_max = 0;
7664 		ple_window_shrink = 0;
7665 	}
7666 
7667 	if (!cpu_has_vmx_apicv()) {
7668 		enable_apicv = 0;
7669 		kvm_x86_ops->sync_pir_to_irr = NULL;
7670 	}
7671 
7672 	if (cpu_has_vmx_tsc_scaling()) {
7673 		kvm_has_tsc_control = true;
7674 		kvm_max_tsc_scaling_ratio = KVM_VMX_TSC_MULTIPLIER_MAX;
7675 		kvm_tsc_scaling_ratio_frac_bits = 48;
7676 	}
7677 
7678 	set_bit(0, vmx_vpid_bitmap); /* 0 is reserved for host */
7679 
7680 	if (enable_ept)
7681 		vmx_enable_tdp();
7682 	else
7683 		kvm_disable_tdp();
7684 
7685 	/*
7686 	 * Only enable PML when hardware supports PML feature, and both EPT
7687 	 * and EPT A/D bit features are enabled -- PML depends on them to work.
7688 	 */
7689 	if (!enable_ept || !enable_ept_ad_bits || !cpu_has_vmx_pml())
7690 		enable_pml = 0;
7691 
7692 	if (!enable_pml) {
7693 		kvm_x86_ops->slot_enable_log_dirty = NULL;
7694 		kvm_x86_ops->slot_disable_log_dirty = NULL;
7695 		kvm_x86_ops->flush_log_dirty = NULL;
7696 		kvm_x86_ops->enable_log_dirty_pt_masked = NULL;
7697 	}
7698 
7699 	if (!cpu_has_vmx_preemption_timer())
7700 		enable_preemption_timer = false;
7701 
7702 	if (enable_preemption_timer) {
7703 		u64 use_timer_freq = 5000ULL * 1000 * 1000;
7704 		u64 vmx_msr;
7705 
7706 		rdmsrl(MSR_IA32_VMX_MISC, vmx_msr);
7707 		cpu_preemption_timer_multi =
7708 			vmx_msr & VMX_MISC_PREEMPTION_TIMER_RATE_MASK;
7709 
7710 		if (tsc_khz)
7711 			use_timer_freq = (u64)tsc_khz * 1000;
7712 		use_timer_freq >>= cpu_preemption_timer_multi;
7713 
7714 		/*
7715 		 * KVM "disables" the preemption timer by setting it to its max
7716 		 * value.  Don't use the timer if it might cause spurious exits
7717 		 * at a rate faster than 0.1 Hz (of uninterrupted guest time).
7718 		 */
7719 		if (use_timer_freq > 0xffffffffu / 10)
7720 			enable_preemption_timer = false;
7721 	}
7722 
7723 	if (!enable_preemption_timer) {
7724 		kvm_x86_ops->set_hv_timer = NULL;
7725 		kvm_x86_ops->cancel_hv_timer = NULL;
7726 		kvm_x86_ops->request_immediate_exit = __kvm_request_immediate_exit;
7727 	}
7728 
7729 	kvm_set_posted_intr_wakeup_handler(wakeup_handler);
7730 
7731 	kvm_mce_cap_supported |= MCG_LMCE_P;
7732 
7733 	if (pt_mode != PT_MODE_SYSTEM && pt_mode != PT_MODE_HOST_GUEST)
7734 		return -EINVAL;
7735 	if (!enable_ept || !cpu_has_vmx_intel_pt())
7736 		pt_mode = PT_MODE_SYSTEM;
7737 
7738 	if (nested) {
7739 		nested_vmx_setup_ctls_msrs(&vmcs_config.nested,
7740 					   vmx_capability.ept, enable_apicv);
7741 
7742 		r = nested_vmx_hardware_setup(kvm_vmx_exit_handlers);
7743 		if (r)
7744 			return r;
7745 	}
7746 
7747 	r = alloc_kvm_area();
7748 	if (r)
7749 		nested_vmx_hardware_unsetup();
7750 	return r;
7751 }
7752 
7753 static __exit void hardware_unsetup(void)
7754 {
7755 	if (nested)
7756 		nested_vmx_hardware_unsetup();
7757 
7758 	free_kvm_area();
7759 }
7760 
7761 static struct kvm_x86_ops vmx_x86_ops __ro_after_init = {
7762 	.cpu_has_kvm_support = cpu_has_kvm_support,
7763 	.disabled_by_bios = vmx_disabled_by_bios,
7764 	.hardware_setup = hardware_setup,
7765 	.hardware_unsetup = hardware_unsetup,
7766 	.check_processor_compatibility = vmx_check_processor_compat,
7767 	.hardware_enable = hardware_enable,
7768 	.hardware_disable = hardware_disable,
7769 	.cpu_has_accelerated_tpr = report_flexpriority,
7770 	.has_emulated_msr = vmx_has_emulated_msr,
7771 
7772 	.vm_init = vmx_vm_init,
7773 	.vm_alloc = vmx_vm_alloc,
7774 	.vm_free = vmx_vm_free,
7775 
7776 	.vcpu_create = vmx_create_vcpu,
7777 	.vcpu_free = vmx_free_vcpu,
7778 	.vcpu_reset = vmx_vcpu_reset,
7779 
7780 	.prepare_guest_switch = vmx_prepare_switch_to_guest,
7781 	.vcpu_load = vmx_vcpu_load,
7782 	.vcpu_put = vmx_vcpu_put,
7783 
7784 	.update_bp_intercept = update_exception_bitmap,
7785 	.get_msr_feature = vmx_get_msr_feature,
7786 	.get_msr = vmx_get_msr,
7787 	.set_msr = vmx_set_msr,
7788 	.get_segment_base = vmx_get_segment_base,
7789 	.get_segment = vmx_get_segment,
7790 	.set_segment = vmx_set_segment,
7791 	.get_cpl = vmx_get_cpl,
7792 	.get_cs_db_l_bits = vmx_get_cs_db_l_bits,
7793 	.decache_cr0_guest_bits = vmx_decache_cr0_guest_bits,
7794 	.decache_cr4_guest_bits = vmx_decache_cr4_guest_bits,
7795 	.set_cr0 = vmx_set_cr0,
7796 	.set_cr3 = vmx_set_cr3,
7797 	.set_cr4 = vmx_set_cr4,
7798 	.set_efer = vmx_set_efer,
7799 	.get_idt = vmx_get_idt,
7800 	.set_idt = vmx_set_idt,
7801 	.get_gdt = vmx_get_gdt,
7802 	.set_gdt = vmx_set_gdt,
7803 	.get_dr6 = vmx_get_dr6,
7804 	.set_dr6 = vmx_set_dr6,
7805 	.set_dr7 = vmx_set_dr7,
7806 	.sync_dirty_debug_regs = vmx_sync_dirty_debug_regs,
7807 	.cache_reg = vmx_cache_reg,
7808 	.get_rflags = vmx_get_rflags,
7809 	.set_rflags = vmx_set_rflags,
7810 
7811 	.tlb_flush = vmx_flush_tlb,
7812 	.tlb_flush_gva = vmx_flush_tlb_gva,
7813 
7814 	.run = vmx_vcpu_run,
7815 	.handle_exit = vmx_handle_exit,
7816 	.skip_emulated_instruction = skip_emulated_instruction,
7817 	.set_interrupt_shadow = vmx_set_interrupt_shadow,
7818 	.get_interrupt_shadow = vmx_get_interrupt_shadow,
7819 	.patch_hypercall = vmx_patch_hypercall,
7820 	.set_irq = vmx_inject_irq,
7821 	.set_nmi = vmx_inject_nmi,
7822 	.queue_exception = vmx_queue_exception,
7823 	.cancel_injection = vmx_cancel_injection,
7824 	.interrupt_allowed = vmx_interrupt_allowed,
7825 	.nmi_allowed = vmx_nmi_allowed,
7826 	.get_nmi_mask = vmx_get_nmi_mask,
7827 	.set_nmi_mask = vmx_set_nmi_mask,
7828 	.enable_nmi_window = enable_nmi_window,
7829 	.enable_irq_window = enable_irq_window,
7830 	.update_cr8_intercept = update_cr8_intercept,
7831 	.set_virtual_apic_mode = vmx_set_virtual_apic_mode,
7832 	.set_apic_access_page_addr = vmx_set_apic_access_page_addr,
7833 	.get_enable_apicv = vmx_get_enable_apicv,
7834 	.refresh_apicv_exec_ctrl = vmx_refresh_apicv_exec_ctrl,
7835 	.load_eoi_exitmap = vmx_load_eoi_exitmap,
7836 	.apicv_post_state_restore = vmx_apicv_post_state_restore,
7837 	.hwapic_irr_update = vmx_hwapic_irr_update,
7838 	.hwapic_isr_update = vmx_hwapic_isr_update,
7839 	.guest_apic_has_interrupt = vmx_guest_apic_has_interrupt,
7840 	.sync_pir_to_irr = vmx_sync_pir_to_irr,
7841 	.deliver_posted_interrupt = vmx_deliver_posted_interrupt,
7842 	.dy_apicv_has_pending_interrupt = vmx_dy_apicv_has_pending_interrupt,
7843 
7844 	.set_tss_addr = vmx_set_tss_addr,
7845 	.set_identity_map_addr = vmx_set_identity_map_addr,
7846 	.get_tdp_level = get_ept_level,
7847 	.get_mt_mask = vmx_get_mt_mask,
7848 
7849 	.get_exit_info = vmx_get_exit_info,
7850 
7851 	.get_lpage_level = vmx_get_lpage_level,
7852 
7853 	.cpuid_update = vmx_cpuid_update,
7854 
7855 	.rdtscp_supported = vmx_rdtscp_supported,
7856 	.invpcid_supported = vmx_invpcid_supported,
7857 
7858 	.set_supported_cpuid = vmx_set_supported_cpuid,
7859 
7860 	.has_wbinvd_exit = cpu_has_vmx_wbinvd_exit,
7861 
7862 	.read_l1_tsc_offset = vmx_read_l1_tsc_offset,
7863 	.write_l1_tsc_offset = vmx_write_l1_tsc_offset,
7864 
7865 	.set_tdp_cr3 = vmx_set_cr3,
7866 
7867 	.check_intercept = vmx_check_intercept,
7868 	.handle_exit_irqoff = vmx_handle_exit_irqoff,
7869 	.mpx_supported = vmx_mpx_supported,
7870 	.xsaves_supported = vmx_xsaves_supported,
7871 	.umip_emulated = vmx_umip_emulated,
7872 	.pt_supported = vmx_pt_supported,
7873 
7874 	.request_immediate_exit = vmx_request_immediate_exit,
7875 
7876 	.sched_in = vmx_sched_in,
7877 
7878 	.slot_enable_log_dirty = vmx_slot_enable_log_dirty,
7879 	.slot_disable_log_dirty = vmx_slot_disable_log_dirty,
7880 	.flush_log_dirty = vmx_flush_log_dirty,
7881 	.enable_log_dirty_pt_masked = vmx_enable_log_dirty_pt_masked,
7882 	.write_log_dirty = vmx_write_pml_buffer,
7883 
7884 	.pre_block = vmx_pre_block,
7885 	.post_block = vmx_post_block,
7886 
7887 	.pmu_ops = &intel_pmu_ops,
7888 
7889 	.update_pi_irte = vmx_update_pi_irte,
7890 
7891 #ifdef CONFIG_X86_64
7892 	.set_hv_timer = vmx_set_hv_timer,
7893 	.cancel_hv_timer = vmx_cancel_hv_timer,
7894 #endif
7895 
7896 	.setup_mce = vmx_setup_mce,
7897 
7898 	.smi_allowed = vmx_smi_allowed,
7899 	.pre_enter_smm = vmx_pre_enter_smm,
7900 	.pre_leave_smm = vmx_pre_leave_smm,
7901 	.enable_smi_window = enable_smi_window,
7902 
7903 	.check_nested_events = NULL,
7904 	.get_nested_state = NULL,
7905 	.set_nested_state = NULL,
7906 	.get_vmcs12_pages = NULL,
7907 	.nested_enable_evmcs = NULL,
7908 	.nested_get_evmcs_version = NULL,
7909 	.need_emulation_on_page_fault = vmx_need_emulation_on_page_fault,
7910 	.apic_init_signal_blocked = vmx_apic_init_signal_blocked,
7911 };
7912 
7913 static void vmx_cleanup_l1d_flush(void)
7914 {
7915 	if (vmx_l1d_flush_pages) {
7916 		free_pages((unsigned long)vmx_l1d_flush_pages, L1D_CACHE_ORDER);
7917 		vmx_l1d_flush_pages = NULL;
7918 	}
7919 	/* Restore state so sysfs ignores VMX */
7920 	l1tf_vmx_mitigation = VMENTER_L1D_FLUSH_AUTO;
7921 }
7922 
7923 static void vmx_exit(void)
7924 {
7925 #ifdef CONFIG_KEXEC_CORE
7926 	RCU_INIT_POINTER(crash_vmclear_loaded_vmcss, NULL);
7927 	synchronize_rcu();
7928 #endif
7929 
7930 	kvm_exit();
7931 
7932 #if IS_ENABLED(CONFIG_HYPERV)
7933 	if (static_branch_unlikely(&enable_evmcs)) {
7934 		int cpu;
7935 		struct hv_vp_assist_page *vp_ap;
7936 		/*
7937 		 * Reset everything to support using non-enlightened VMCS
7938 		 * access later (e.g. when we reload the module with
7939 		 * enlightened_vmcs=0)
7940 		 */
7941 		for_each_online_cpu(cpu) {
7942 			vp_ap =	hv_get_vp_assist_page(cpu);
7943 
7944 			if (!vp_ap)
7945 				continue;
7946 
7947 			vp_ap->nested_control.features.directhypercall = 0;
7948 			vp_ap->current_nested_vmcs = 0;
7949 			vp_ap->enlighten_vmentry = 0;
7950 		}
7951 
7952 		static_branch_disable(&enable_evmcs);
7953 	}
7954 #endif
7955 	vmx_cleanup_l1d_flush();
7956 }
7957 module_exit(vmx_exit);
7958 
7959 static int __init vmx_init(void)
7960 {
7961 	int r;
7962 
7963 #if IS_ENABLED(CONFIG_HYPERV)
7964 	/*
7965 	 * Enlightened VMCS usage should be recommended and the host needs
7966 	 * to support eVMCS v1 or above. We can also disable eVMCS support
7967 	 * with module parameter.
7968 	 */
7969 	if (enlightened_vmcs &&
7970 	    ms_hyperv.hints & HV_X64_ENLIGHTENED_VMCS_RECOMMENDED &&
7971 	    (ms_hyperv.nested_features & HV_X64_ENLIGHTENED_VMCS_VERSION) >=
7972 	    KVM_EVMCS_VERSION) {
7973 		int cpu;
7974 
7975 		/* Check that we have assist pages on all online CPUs */
7976 		for_each_online_cpu(cpu) {
7977 			if (!hv_get_vp_assist_page(cpu)) {
7978 				enlightened_vmcs = false;
7979 				break;
7980 			}
7981 		}
7982 
7983 		if (enlightened_vmcs) {
7984 			pr_info("KVM: vmx: using Hyper-V Enlightened VMCS\n");
7985 			static_branch_enable(&enable_evmcs);
7986 		}
7987 
7988 		if (ms_hyperv.nested_features & HV_X64_NESTED_DIRECT_FLUSH)
7989 			vmx_x86_ops.enable_direct_tlbflush
7990 				= hv_enable_direct_tlbflush;
7991 
7992 	} else {
7993 		enlightened_vmcs = false;
7994 	}
7995 #endif
7996 
7997 	r = kvm_init(&vmx_x86_ops, sizeof(struct vcpu_vmx),
7998 		     __alignof__(struct vcpu_vmx), THIS_MODULE);
7999 	if (r)
8000 		return r;
8001 
8002 	/*
8003 	 * Must be called after kvm_init() so enable_ept is properly set
8004 	 * up. Hand the parameter mitigation value in which was stored in
8005 	 * the pre module init parser. If no parameter was given, it will
8006 	 * contain 'auto' which will be turned into the default 'cond'
8007 	 * mitigation mode.
8008 	 */
8009 	r = vmx_setup_l1d_flush(vmentry_l1d_flush_param);
8010 	if (r) {
8011 		vmx_exit();
8012 		return r;
8013 	}
8014 
8015 #ifdef CONFIG_KEXEC_CORE
8016 	rcu_assign_pointer(crash_vmclear_loaded_vmcss,
8017 			   crash_vmclear_local_loaded_vmcss);
8018 #endif
8019 	vmx_check_vmcs12_offsets();
8020 
8021 	return 0;
8022 }
8023 module_init(vmx_init);
8024