xref: /openbmc/linux/arch/x86/kvm/vmx/vmx.c (revision fcbd8037f7df694aa7bfb7ce82c0c7f5e53e7b7b)
1 // SPDX-License-Identifier: GPL-2.0-only
2 /*
3  * Kernel-based Virtual Machine driver for Linux
4  *
5  * This module enables machines with Intel VT-x extensions to run virtual
6  * machines without emulation or binary translation.
7  *
8  * Copyright (C) 2006 Qumranet, Inc.
9  * Copyright 2010 Red Hat, Inc. and/or its affiliates.
10  *
11  * Authors:
12  *   Avi Kivity   <avi@qumranet.com>
13  *   Yaniv Kamay  <yaniv@qumranet.com>
14  */
15 
16 #include <linux/frame.h>
17 #include <linux/highmem.h>
18 #include <linux/hrtimer.h>
19 #include <linux/kernel.h>
20 #include <linux/kvm_host.h>
21 #include <linux/module.h>
22 #include <linux/moduleparam.h>
23 #include <linux/mod_devicetable.h>
24 #include <linux/mm.h>
25 #include <linux/sched.h>
26 #include <linux/sched/smt.h>
27 #include <linux/slab.h>
28 #include <linux/tboot.h>
29 #include <linux/trace_events.h>
30 
31 #include <asm/apic.h>
32 #include <asm/asm.h>
33 #include <asm/cpu.h>
34 #include <asm/debugreg.h>
35 #include <asm/desc.h>
36 #include <asm/fpu/internal.h>
37 #include <asm/io.h>
38 #include <asm/irq_remapping.h>
39 #include <asm/kexec.h>
40 #include <asm/perf_event.h>
41 #include <asm/mce.h>
42 #include <asm/mmu_context.h>
43 #include <asm/mshyperv.h>
44 #include <asm/spec-ctrl.h>
45 #include <asm/virtext.h>
46 #include <asm/vmx.h>
47 
48 #include "capabilities.h"
49 #include "cpuid.h"
50 #include "evmcs.h"
51 #include "irq.h"
52 #include "kvm_cache_regs.h"
53 #include "lapic.h"
54 #include "mmu.h"
55 #include "nested.h"
56 #include "ops.h"
57 #include "pmu.h"
58 #include "trace.h"
59 #include "vmcs.h"
60 #include "vmcs12.h"
61 #include "vmx.h"
62 #include "x86.h"
63 
64 MODULE_AUTHOR("Qumranet");
65 MODULE_LICENSE("GPL");
66 
67 static const struct x86_cpu_id vmx_cpu_id[] = {
68 	X86_FEATURE_MATCH(X86_FEATURE_VMX),
69 	{}
70 };
71 MODULE_DEVICE_TABLE(x86cpu, vmx_cpu_id);
72 
73 bool __read_mostly enable_vpid = 1;
74 module_param_named(vpid, enable_vpid, bool, 0444);
75 
76 static bool __read_mostly enable_vnmi = 1;
77 module_param_named(vnmi, enable_vnmi, bool, S_IRUGO);
78 
79 bool __read_mostly flexpriority_enabled = 1;
80 module_param_named(flexpriority, flexpriority_enabled, bool, S_IRUGO);
81 
82 bool __read_mostly enable_ept = 1;
83 module_param_named(ept, enable_ept, bool, S_IRUGO);
84 
85 bool __read_mostly enable_unrestricted_guest = 1;
86 module_param_named(unrestricted_guest,
87 			enable_unrestricted_guest, bool, S_IRUGO);
88 
89 bool __read_mostly enable_ept_ad_bits = 1;
90 module_param_named(eptad, enable_ept_ad_bits, bool, S_IRUGO);
91 
92 static bool __read_mostly emulate_invalid_guest_state = true;
93 module_param(emulate_invalid_guest_state, bool, S_IRUGO);
94 
95 static bool __read_mostly fasteoi = 1;
96 module_param(fasteoi, bool, S_IRUGO);
97 
98 static bool __read_mostly enable_apicv = 1;
99 module_param(enable_apicv, bool, S_IRUGO);
100 
101 /*
102  * If nested=1, nested virtualization is supported, i.e., guests may use
103  * VMX and be a hypervisor for its own guests. If nested=0, guests may not
104  * use VMX instructions.
105  */
106 static bool __read_mostly nested = 1;
107 module_param(nested, bool, S_IRUGO);
108 
109 static u64 __read_mostly host_xss;
110 
111 bool __read_mostly enable_pml = 1;
112 module_param_named(pml, enable_pml, bool, S_IRUGO);
113 
114 static bool __read_mostly dump_invalid_vmcs = 0;
115 module_param(dump_invalid_vmcs, bool, 0644);
116 
117 #define MSR_BITMAP_MODE_X2APIC		1
118 #define MSR_BITMAP_MODE_X2APIC_APICV	2
119 
120 #define KVM_VMX_TSC_MULTIPLIER_MAX     0xffffffffffffffffULL
121 
122 /* Guest_tsc -> host_tsc conversion requires 64-bit division.  */
123 static int __read_mostly cpu_preemption_timer_multi;
124 static bool __read_mostly enable_preemption_timer = 1;
125 #ifdef CONFIG_X86_64
126 module_param_named(preemption_timer, enable_preemption_timer, bool, S_IRUGO);
127 #endif
128 
129 #define KVM_VM_CR0_ALWAYS_OFF (X86_CR0_NW | X86_CR0_CD)
130 #define KVM_VM_CR0_ALWAYS_ON_UNRESTRICTED_GUEST X86_CR0_NE
131 #define KVM_VM_CR0_ALWAYS_ON				\
132 	(KVM_VM_CR0_ALWAYS_ON_UNRESTRICTED_GUEST | 	\
133 	 X86_CR0_WP | X86_CR0_PG | X86_CR0_PE)
134 #define KVM_CR4_GUEST_OWNED_BITS				      \
135 	(X86_CR4_PVI | X86_CR4_DE | X86_CR4_PCE | X86_CR4_OSFXSR      \
136 	 | X86_CR4_OSXMMEXCPT | X86_CR4_LA57 | X86_CR4_TSD)
137 
138 #define KVM_VM_CR4_ALWAYS_ON_UNRESTRICTED_GUEST X86_CR4_VMXE
139 #define KVM_PMODE_VM_CR4_ALWAYS_ON (X86_CR4_PAE | X86_CR4_VMXE)
140 #define KVM_RMODE_VM_CR4_ALWAYS_ON (X86_CR4_VME | X86_CR4_PAE | X86_CR4_VMXE)
141 
142 #define RMODE_GUEST_OWNED_EFLAGS_BITS (~(X86_EFLAGS_IOPL | X86_EFLAGS_VM))
143 
144 #define MSR_IA32_RTIT_STATUS_MASK (~(RTIT_STATUS_FILTEREN | \
145 	RTIT_STATUS_CONTEXTEN | RTIT_STATUS_TRIGGEREN | \
146 	RTIT_STATUS_ERROR | RTIT_STATUS_STOPPED | \
147 	RTIT_STATUS_BYTECNT))
148 
149 #define MSR_IA32_RTIT_OUTPUT_BASE_MASK \
150 	(~((1UL << cpuid_query_maxphyaddr(vcpu)) - 1) | 0x7f)
151 
152 /*
153  * These 2 parameters are used to config the controls for Pause-Loop Exiting:
154  * ple_gap:    upper bound on the amount of time between two successive
155  *             executions of PAUSE in a loop. Also indicate if ple enabled.
156  *             According to test, this time is usually smaller than 128 cycles.
157  * ple_window: upper bound on the amount of time a guest is allowed to execute
158  *             in a PAUSE loop. Tests indicate that most spinlocks are held for
159  *             less than 2^12 cycles
160  * Time is measured based on a counter that runs at the same rate as the TSC,
161  * refer SDM volume 3b section 21.6.13 & 22.1.3.
162  */
163 static unsigned int ple_gap = KVM_DEFAULT_PLE_GAP;
164 module_param(ple_gap, uint, 0444);
165 
166 static unsigned int ple_window = KVM_VMX_DEFAULT_PLE_WINDOW;
167 module_param(ple_window, uint, 0444);
168 
169 /* Default doubles per-vcpu window every exit. */
170 static unsigned int ple_window_grow = KVM_DEFAULT_PLE_WINDOW_GROW;
171 module_param(ple_window_grow, uint, 0444);
172 
173 /* Default resets per-vcpu window every exit to ple_window. */
174 static unsigned int ple_window_shrink = KVM_DEFAULT_PLE_WINDOW_SHRINK;
175 module_param(ple_window_shrink, uint, 0444);
176 
177 /* Default is to compute the maximum so we can never overflow. */
178 static unsigned int ple_window_max        = KVM_VMX_DEFAULT_PLE_WINDOW_MAX;
179 module_param(ple_window_max, uint, 0444);
180 
181 /* Default is SYSTEM mode, 1 for host-guest mode */
182 int __read_mostly pt_mode = PT_MODE_SYSTEM;
183 module_param(pt_mode, int, S_IRUGO);
184 
185 static DEFINE_STATIC_KEY_FALSE(vmx_l1d_should_flush);
186 static DEFINE_STATIC_KEY_FALSE(vmx_l1d_flush_cond);
187 static DEFINE_MUTEX(vmx_l1d_flush_mutex);
188 
189 /* Storage for pre module init parameter parsing */
190 static enum vmx_l1d_flush_state __read_mostly vmentry_l1d_flush_param = VMENTER_L1D_FLUSH_AUTO;
191 
192 static const struct {
193 	const char *option;
194 	bool for_parse;
195 } vmentry_l1d_param[] = {
196 	[VMENTER_L1D_FLUSH_AUTO]	 = {"auto", true},
197 	[VMENTER_L1D_FLUSH_NEVER]	 = {"never", true},
198 	[VMENTER_L1D_FLUSH_COND]	 = {"cond", true},
199 	[VMENTER_L1D_FLUSH_ALWAYS]	 = {"always", true},
200 	[VMENTER_L1D_FLUSH_EPT_DISABLED] = {"EPT disabled", false},
201 	[VMENTER_L1D_FLUSH_NOT_REQUIRED] = {"not required", false},
202 };
203 
204 #define L1D_CACHE_ORDER 4
205 static void *vmx_l1d_flush_pages;
206 
207 static int vmx_setup_l1d_flush(enum vmx_l1d_flush_state l1tf)
208 {
209 	struct page *page;
210 	unsigned int i;
211 
212 	if (!boot_cpu_has_bug(X86_BUG_L1TF)) {
213 		l1tf_vmx_mitigation = VMENTER_L1D_FLUSH_NOT_REQUIRED;
214 		return 0;
215 	}
216 
217 	if (!enable_ept) {
218 		l1tf_vmx_mitigation = VMENTER_L1D_FLUSH_EPT_DISABLED;
219 		return 0;
220 	}
221 
222 	if (boot_cpu_has(X86_FEATURE_ARCH_CAPABILITIES)) {
223 		u64 msr;
224 
225 		rdmsrl(MSR_IA32_ARCH_CAPABILITIES, msr);
226 		if (msr & ARCH_CAP_SKIP_VMENTRY_L1DFLUSH) {
227 			l1tf_vmx_mitigation = VMENTER_L1D_FLUSH_NOT_REQUIRED;
228 			return 0;
229 		}
230 	}
231 
232 	/* If set to auto use the default l1tf mitigation method */
233 	if (l1tf == VMENTER_L1D_FLUSH_AUTO) {
234 		switch (l1tf_mitigation) {
235 		case L1TF_MITIGATION_OFF:
236 			l1tf = VMENTER_L1D_FLUSH_NEVER;
237 			break;
238 		case L1TF_MITIGATION_FLUSH_NOWARN:
239 		case L1TF_MITIGATION_FLUSH:
240 		case L1TF_MITIGATION_FLUSH_NOSMT:
241 			l1tf = VMENTER_L1D_FLUSH_COND;
242 			break;
243 		case L1TF_MITIGATION_FULL:
244 		case L1TF_MITIGATION_FULL_FORCE:
245 			l1tf = VMENTER_L1D_FLUSH_ALWAYS;
246 			break;
247 		}
248 	} else if (l1tf_mitigation == L1TF_MITIGATION_FULL_FORCE) {
249 		l1tf = VMENTER_L1D_FLUSH_ALWAYS;
250 	}
251 
252 	if (l1tf != VMENTER_L1D_FLUSH_NEVER && !vmx_l1d_flush_pages &&
253 	    !boot_cpu_has(X86_FEATURE_FLUSH_L1D)) {
254 		/*
255 		 * This allocation for vmx_l1d_flush_pages is not tied to a VM
256 		 * lifetime and so should not be charged to a memcg.
257 		 */
258 		page = alloc_pages(GFP_KERNEL, L1D_CACHE_ORDER);
259 		if (!page)
260 			return -ENOMEM;
261 		vmx_l1d_flush_pages = page_address(page);
262 
263 		/*
264 		 * Initialize each page with a different pattern in
265 		 * order to protect against KSM in the nested
266 		 * virtualization case.
267 		 */
268 		for (i = 0; i < 1u << L1D_CACHE_ORDER; ++i) {
269 			memset(vmx_l1d_flush_pages + i * PAGE_SIZE, i + 1,
270 			       PAGE_SIZE);
271 		}
272 	}
273 
274 	l1tf_vmx_mitigation = l1tf;
275 
276 	if (l1tf != VMENTER_L1D_FLUSH_NEVER)
277 		static_branch_enable(&vmx_l1d_should_flush);
278 	else
279 		static_branch_disable(&vmx_l1d_should_flush);
280 
281 	if (l1tf == VMENTER_L1D_FLUSH_COND)
282 		static_branch_enable(&vmx_l1d_flush_cond);
283 	else
284 		static_branch_disable(&vmx_l1d_flush_cond);
285 	return 0;
286 }
287 
288 static int vmentry_l1d_flush_parse(const char *s)
289 {
290 	unsigned int i;
291 
292 	if (s) {
293 		for (i = 0; i < ARRAY_SIZE(vmentry_l1d_param); i++) {
294 			if (vmentry_l1d_param[i].for_parse &&
295 			    sysfs_streq(s, vmentry_l1d_param[i].option))
296 				return i;
297 		}
298 	}
299 	return -EINVAL;
300 }
301 
302 static int vmentry_l1d_flush_set(const char *s, const struct kernel_param *kp)
303 {
304 	int l1tf, ret;
305 
306 	l1tf = vmentry_l1d_flush_parse(s);
307 	if (l1tf < 0)
308 		return l1tf;
309 
310 	if (!boot_cpu_has(X86_BUG_L1TF))
311 		return 0;
312 
313 	/*
314 	 * Has vmx_init() run already? If not then this is the pre init
315 	 * parameter parsing. In that case just store the value and let
316 	 * vmx_init() do the proper setup after enable_ept has been
317 	 * established.
318 	 */
319 	if (l1tf_vmx_mitigation == VMENTER_L1D_FLUSH_AUTO) {
320 		vmentry_l1d_flush_param = l1tf;
321 		return 0;
322 	}
323 
324 	mutex_lock(&vmx_l1d_flush_mutex);
325 	ret = vmx_setup_l1d_flush(l1tf);
326 	mutex_unlock(&vmx_l1d_flush_mutex);
327 	return ret;
328 }
329 
330 static int vmentry_l1d_flush_get(char *s, const struct kernel_param *kp)
331 {
332 	if (WARN_ON_ONCE(l1tf_vmx_mitigation >= ARRAY_SIZE(vmentry_l1d_param)))
333 		return sprintf(s, "???\n");
334 
335 	return sprintf(s, "%s\n", vmentry_l1d_param[l1tf_vmx_mitigation].option);
336 }
337 
338 static const struct kernel_param_ops vmentry_l1d_flush_ops = {
339 	.set = vmentry_l1d_flush_set,
340 	.get = vmentry_l1d_flush_get,
341 };
342 module_param_cb(vmentry_l1d_flush, &vmentry_l1d_flush_ops, NULL, 0644);
343 
344 static bool guest_state_valid(struct kvm_vcpu *vcpu);
345 static u32 vmx_segment_access_rights(struct kvm_segment *var);
346 static __always_inline void vmx_disable_intercept_for_msr(unsigned long *msr_bitmap,
347 							  u32 msr, int type);
348 
349 void vmx_vmexit(void);
350 
351 #define vmx_insn_failed(fmt...)		\
352 do {					\
353 	WARN_ONCE(1, fmt);		\
354 	pr_warn_ratelimited(fmt);	\
355 } while (0)
356 
357 asmlinkage void vmread_error(unsigned long field, bool fault)
358 {
359 	if (fault)
360 		kvm_spurious_fault();
361 	else
362 		vmx_insn_failed("kvm: vmread failed: field=%lx\n", field);
363 }
364 
365 noinline void vmwrite_error(unsigned long field, unsigned long value)
366 {
367 	vmx_insn_failed("kvm: vmwrite failed: field=%lx val=%lx err=%d\n",
368 			field, value, vmcs_read32(VM_INSTRUCTION_ERROR));
369 }
370 
371 noinline void vmclear_error(struct vmcs *vmcs, u64 phys_addr)
372 {
373 	vmx_insn_failed("kvm: vmclear failed: %p/%llx\n", vmcs, phys_addr);
374 }
375 
376 noinline void vmptrld_error(struct vmcs *vmcs, u64 phys_addr)
377 {
378 	vmx_insn_failed("kvm: vmptrld failed: %p/%llx\n", vmcs, phys_addr);
379 }
380 
381 noinline void invvpid_error(unsigned long ext, u16 vpid, gva_t gva)
382 {
383 	vmx_insn_failed("kvm: invvpid failed: ext=0x%lx vpid=%u gva=0x%lx\n",
384 			ext, vpid, gva);
385 }
386 
387 noinline void invept_error(unsigned long ext, u64 eptp, gpa_t gpa)
388 {
389 	vmx_insn_failed("kvm: invept failed: ext=0x%lx eptp=%llx gpa=0x%llx\n",
390 			ext, eptp, gpa);
391 }
392 
393 static DEFINE_PER_CPU(struct vmcs *, vmxarea);
394 DEFINE_PER_CPU(struct vmcs *, current_vmcs);
395 /*
396  * We maintain a per-CPU linked-list of VMCS loaded on that CPU. This is needed
397  * when a CPU is brought down, and we need to VMCLEAR all VMCSs loaded on it.
398  */
399 static DEFINE_PER_CPU(struct list_head, loaded_vmcss_on_cpu);
400 
401 /*
402  * We maintian a per-CPU linked-list of vCPU, so in wakeup_handler() we
403  * can find which vCPU should be waken up.
404  */
405 static DEFINE_PER_CPU(struct list_head, blocked_vcpu_on_cpu);
406 static DEFINE_PER_CPU(spinlock_t, blocked_vcpu_on_cpu_lock);
407 
408 static DECLARE_BITMAP(vmx_vpid_bitmap, VMX_NR_VPIDS);
409 static DEFINE_SPINLOCK(vmx_vpid_lock);
410 
411 struct vmcs_config vmcs_config;
412 struct vmx_capability vmx_capability;
413 
414 #define VMX_SEGMENT_FIELD(seg)					\
415 	[VCPU_SREG_##seg] = {                                   \
416 		.selector = GUEST_##seg##_SELECTOR,		\
417 		.base = GUEST_##seg##_BASE,		   	\
418 		.limit = GUEST_##seg##_LIMIT,		   	\
419 		.ar_bytes = GUEST_##seg##_AR_BYTES,	   	\
420 	}
421 
422 static const struct kvm_vmx_segment_field {
423 	unsigned selector;
424 	unsigned base;
425 	unsigned limit;
426 	unsigned ar_bytes;
427 } kvm_vmx_segment_fields[] = {
428 	VMX_SEGMENT_FIELD(CS),
429 	VMX_SEGMENT_FIELD(DS),
430 	VMX_SEGMENT_FIELD(ES),
431 	VMX_SEGMENT_FIELD(FS),
432 	VMX_SEGMENT_FIELD(GS),
433 	VMX_SEGMENT_FIELD(SS),
434 	VMX_SEGMENT_FIELD(TR),
435 	VMX_SEGMENT_FIELD(LDTR),
436 };
437 
438 u64 host_efer;
439 static unsigned long host_idt_base;
440 
441 /*
442  * Though SYSCALL is only supported in 64-bit mode on Intel CPUs, kvm
443  * will emulate SYSCALL in legacy mode if the vendor string in guest
444  * CPUID.0:{EBX,ECX,EDX} is "AuthenticAMD" or "AMDisbetter!" To
445  * support this emulation, IA32_STAR must always be included in
446  * vmx_msr_index[], even in i386 builds.
447  */
448 const u32 vmx_msr_index[] = {
449 #ifdef CONFIG_X86_64
450 	MSR_SYSCALL_MASK, MSR_LSTAR, MSR_CSTAR,
451 #endif
452 	MSR_EFER, MSR_TSC_AUX, MSR_STAR,
453 };
454 
455 #if IS_ENABLED(CONFIG_HYPERV)
456 static bool __read_mostly enlightened_vmcs = true;
457 module_param(enlightened_vmcs, bool, 0444);
458 
459 /* check_ept_pointer() should be under protection of ept_pointer_lock. */
460 static void check_ept_pointer_match(struct kvm *kvm)
461 {
462 	struct kvm_vcpu *vcpu;
463 	u64 tmp_eptp = INVALID_PAGE;
464 	int i;
465 
466 	kvm_for_each_vcpu(i, vcpu, kvm) {
467 		if (!VALID_PAGE(tmp_eptp)) {
468 			tmp_eptp = to_vmx(vcpu)->ept_pointer;
469 		} else if (tmp_eptp != to_vmx(vcpu)->ept_pointer) {
470 			to_kvm_vmx(kvm)->ept_pointers_match
471 				= EPT_POINTERS_MISMATCH;
472 			return;
473 		}
474 	}
475 
476 	to_kvm_vmx(kvm)->ept_pointers_match = EPT_POINTERS_MATCH;
477 }
478 
479 static int kvm_fill_hv_flush_list_func(struct hv_guest_mapping_flush_list *flush,
480 		void *data)
481 {
482 	struct kvm_tlb_range *range = data;
483 
484 	return hyperv_fill_flush_guest_mapping_list(flush, range->start_gfn,
485 			range->pages);
486 }
487 
488 static inline int __hv_remote_flush_tlb_with_range(struct kvm *kvm,
489 		struct kvm_vcpu *vcpu, struct kvm_tlb_range *range)
490 {
491 	u64 ept_pointer = to_vmx(vcpu)->ept_pointer;
492 
493 	/*
494 	 * FLUSH_GUEST_PHYSICAL_ADDRESS_SPACE hypercall needs address
495 	 * of the base of EPT PML4 table, strip off EPT configuration
496 	 * information.
497 	 */
498 	if (range)
499 		return hyperv_flush_guest_mapping_range(ept_pointer & PAGE_MASK,
500 				kvm_fill_hv_flush_list_func, (void *)range);
501 	else
502 		return hyperv_flush_guest_mapping(ept_pointer & PAGE_MASK);
503 }
504 
505 static int hv_remote_flush_tlb_with_range(struct kvm *kvm,
506 		struct kvm_tlb_range *range)
507 {
508 	struct kvm_vcpu *vcpu;
509 	int ret = 0, i;
510 
511 	spin_lock(&to_kvm_vmx(kvm)->ept_pointer_lock);
512 
513 	if (to_kvm_vmx(kvm)->ept_pointers_match == EPT_POINTERS_CHECK)
514 		check_ept_pointer_match(kvm);
515 
516 	if (to_kvm_vmx(kvm)->ept_pointers_match != EPT_POINTERS_MATCH) {
517 		kvm_for_each_vcpu(i, vcpu, kvm) {
518 			/* If ept_pointer is invalid pointer, bypass flush request. */
519 			if (VALID_PAGE(to_vmx(vcpu)->ept_pointer))
520 				ret |= __hv_remote_flush_tlb_with_range(
521 					kvm, vcpu, range);
522 		}
523 	} else {
524 		ret = __hv_remote_flush_tlb_with_range(kvm,
525 				kvm_get_vcpu(kvm, 0), range);
526 	}
527 
528 	spin_unlock(&to_kvm_vmx(kvm)->ept_pointer_lock);
529 	return ret;
530 }
531 static int hv_remote_flush_tlb(struct kvm *kvm)
532 {
533 	return hv_remote_flush_tlb_with_range(kvm, NULL);
534 }
535 
536 static int hv_enable_direct_tlbflush(struct kvm_vcpu *vcpu)
537 {
538 	struct hv_enlightened_vmcs *evmcs;
539 	struct hv_partition_assist_pg **p_hv_pa_pg =
540 			&vcpu->kvm->arch.hyperv.hv_pa_pg;
541 	/*
542 	 * Synthetic VM-Exit is not enabled in current code and so All
543 	 * evmcs in singe VM shares same assist page.
544 	 */
545 	if (!*p_hv_pa_pg)
546 		*p_hv_pa_pg = kzalloc(PAGE_SIZE, GFP_KERNEL);
547 
548 	if (!*p_hv_pa_pg)
549 		return -ENOMEM;
550 
551 	evmcs = (struct hv_enlightened_vmcs *)to_vmx(vcpu)->loaded_vmcs->vmcs;
552 
553 	evmcs->partition_assist_page =
554 		__pa(*p_hv_pa_pg);
555 	evmcs->hv_vm_id = (unsigned long)vcpu->kvm;
556 	evmcs->hv_enlightenments_control.nested_flush_hypercall = 1;
557 
558 	return 0;
559 }
560 
561 #endif /* IS_ENABLED(CONFIG_HYPERV) */
562 
563 /*
564  * Comment's format: document - errata name - stepping - processor name.
565  * Refer from
566  * https://www.virtualbox.org/svn/vbox/trunk/src/VBox/VMM/VMMR0/HMR0.cpp
567  */
568 static u32 vmx_preemption_cpu_tfms[] = {
569 /* 323344.pdf - BA86   - D0 - Xeon 7500 Series */
570 0x000206E6,
571 /* 323056.pdf - AAX65  - C2 - Xeon L3406 */
572 /* 322814.pdf - AAT59  - C2 - i7-600, i5-500, i5-400 and i3-300 Mobile */
573 /* 322911.pdf - AAU65  - C2 - i5-600, i3-500 Desktop and Pentium G6950 */
574 0x00020652,
575 /* 322911.pdf - AAU65  - K0 - i5-600, i3-500 Desktop and Pentium G6950 */
576 0x00020655,
577 /* 322373.pdf - AAO95  - B1 - Xeon 3400 Series */
578 /* 322166.pdf - AAN92  - B1 - i7-800 and i5-700 Desktop */
579 /*
580  * 320767.pdf - AAP86  - B1 -
581  * i7-900 Mobile Extreme, i7-800 and i7-700 Mobile
582  */
583 0x000106E5,
584 /* 321333.pdf - AAM126 - C0 - Xeon 3500 */
585 0x000106A0,
586 /* 321333.pdf - AAM126 - C1 - Xeon 3500 */
587 0x000106A1,
588 /* 320836.pdf - AAJ124 - C0 - i7-900 Desktop Extreme and i7-900 Desktop */
589 0x000106A4,
590  /* 321333.pdf - AAM126 - D0 - Xeon 3500 */
591  /* 321324.pdf - AAK139 - D0 - Xeon 5500 */
592  /* 320836.pdf - AAJ124 - D0 - i7-900 Extreme and i7-900 Desktop */
593 0x000106A5,
594  /* Xeon E3-1220 V2 */
595 0x000306A8,
596 };
597 
598 static inline bool cpu_has_broken_vmx_preemption_timer(void)
599 {
600 	u32 eax = cpuid_eax(0x00000001), i;
601 
602 	/* Clear the reserved bits */
603 	eax &= ~(0x3U << 14 | 0xfU << 28);
604 	for (i = 0; i < ARRAY_SIZE(vmx_preemption_cpu_tfms); i++)
605 		if (eax == vmx_preemption_cpu_tfms[i])
606 			return true;
607 
608 	return false;
609 }
610 
611 static inline bool cpu_need_virtualize_apic_accesses(struct kvm_vcpu *vcpu)
612 {
613 	return flexpriority_enabled && lapic_in_kernel(vcpu);
614 }
615 
616 static inline bool report_flexpriority(void)
617 {
618 	return flexpriority_enabled;
619 }
620 
621 static inline int __find_msr_index(struct vcpu_vmx *vmx, u32 msr)
622 {
623 	int i;
624 
625 	for (i = 0; i < vmx->nmsrs; ++i)
626 		if (vmx_msr_index[vmx->guest_msrs[i].index] == msr)
627 			return i;
628 	return -1;
629 }
630 
631 struct shared_msr_entry *find_msr_entry(struct vcpu_vmx *vmx, u32 msr)
632 {
633 	int i;
634 
635 	i = __find_msr_index(vmx, msr);
636 	if (i >= 0)
637 		return &vmx->guest_msrs[i];
638 	return NULL;
639 }
640 
641 void loaded_vmcs_init(struct loaded_vmcs *loaded_vmcs)
642 {
643 	vmcs_clear(loaded_vmcs->vmcs);
644 	if (loaded_vmcs->shadow_vmcs && loaded_vmcs->launched)
645 		vmcs_clear(loaded_vmcs->shadow_vmcs);
646 	loaded_vmcs->cpu = -1;
647 	loaded_vmcs->launched = 0;
648 }
649 
650 #ifdef CONFIG_KEXEC_CORE
651 /*
652  * This bitmap is used to indicate whether the vmclear
653  * operation is enabled on all cpus. All disabled by
654  * default.
655  */
656 static cpumask_t crash_vmclear_enabled_bitmap = CPU_MASK_NONE;
657 
658 static inline void crash_enable_local_vmclear(int cpu)
659 {
660 	cpumask_set_cpu(cpu, &crash_vmclear_enabled_bitmap);
661 }
662 
663 static inline void crash_disable_local_vmclear(int cpu)
664 {
665 	cpumask_clear_cpu(cpu, &crash_vmclear_enabled_bitmap);
666 }
667 
668 static inline int crash_local_vmclear_enabled(int cpu)
669 {
670 	return cpumask_test_cpu(cpu, &crash_vmclear_enabled_bitmap);
671 }
672 
673 static void crash_vmclear_local_loaded_vmcss(void)
674 {
675 	int cpu = raw_smp_processor_id();
676 	struct loaded_vmcs *v;
677 
678 	if (!crash_local_vmclear_enabled(cpu))
679 		return;
680 
681 	list_for_each_entry(v, &per_cpu(loaded_vmcss_on_cpu, cpu),
682 			    loaded_vmcss_on_cpu_link)
683 		vmcs_clear(v->vmcs);
684 }
685 #else
686 static inline void crash_enable_local_vmclear(int cpu) { }
687 static inline void crash_disable_local_vmclear(int cpu) { }
688 #endif /* CONFIG_KEXEC_CORE */
689 
690 static void __loaded_vmcs_clear(void *arg)
691 {
692 	struct loaded_vmcs *loaded_vmcs = arg;
693 	int cpu = raw_smp_processor_id();
694 
695 	if (loaded_vmcs->cpu != cpu)
696 		return; /* vcpu migration can race with cpu offline */
697 	if (per_cpu(current_vmcs, cpu) == loaded_vmcs->vmcs)
698 		per_cpu(current_vmcs, cpu) = NULL;
699 	crash_disable_local_vmclear(cpu);
700 	list_del(&loaded_vmcs->loaded_vmcss_on_cpu_link);
701 
702 	/*
703 	 * we should ensure updating loaded_vmcs->loaded_vmcss_on_cpu_link
704 	 * is before setting loaded_vmcs->vcpu to -1 which is done in
705 	 * loaded_vmcs_init. Otherwise, other cpu can see vcpu = -1 fist
706 	 * then adds the vmcs into percpu list before it is deleted.
707 	 */
708 	smp_wmb();
709 
710 	loaded_vmcs_init(loaded_vmcs);
711 	crash_enable_local_vmclear(cpu);
712 }
713 
714 void loaded_vmcs_clear(struct loaded_vmcs *loaded_vmcs)
715 {
716 	int cpu = loaded_vmcs->cpu;
717 
718 	if (cpu != -1)
719 		smp_call_function_single(cpu,
720 			 __loaded_vmcs_clear, loaded_vmcs, 1);
721 }
722 
723 static bool vmx_segment_cache_test_set(struct vcpu_vmx *vmx, unsigned seg,
724 				       unsigned field)
725 {
726 	bool ret;
727 	u32 mask = 1 << (seg * SEG_FIELD_NR + field);
728 
729 	if (!(vmx->vcpu.arch.regs_avail & (1 << VCPU_EXREG_SEGMENTS))) {
730 		vmx->vcpu.arch.regs_avail |= (1 << VCPU_EXREG_SEGMENTS);
731 		vmx->segment_cache.bitmask = 0;
732 	}
733 	ret = vmx->segment_cache.bitmask & mask;
734 	vmx->segment_cache.bitmask |= mask;
735 	return ret;
736 }
737 
738 static u16 vmx_read_guest_seg_selector(struct vcpu_vmx *vmx, unsigned seg)
739 {
740 	u16 *p = &vmx->segment_cache.seg[seg].selector;
741 
742 	if (!vmx_segment_cache_test_set(vmx, seg, SEG_FIELD_SEL))
743 		*p = vmcs_read16(kvm_vmx_segment_fields[seg].selector);
744 	return *p;
745 }
746 
747 static ulong vmx_read_guest_seg_base(struct vcpu_vmx *vmx, unsigned seg)
748 {
749 	ulong *p = &vmx->segment_cache.seg[seg].base;
750 
751 	if (!vmx_segment_cache_test_set(vmx, seg, SEG_FIELD_BASE))
752 		*p = vmcs_readl(kvm_vmx_segment_fields[seg].base);
753 	return *p;
754 }
755 
756 static u32 vmx_read_guest_seg_limit(struct vcpu_vmx *vmx, unsigned seg)
757 {
758 	u32 *p = &vmx->segment_cache.seg[seg].limit;
759 
760 	if (!vmx_segment_cache_test_set(vmx, seg, SEG_FIELD_LIMIT))
761 		*p = vmcs_read32(kvm_vmx_segment_fields[seg].limit);
762 	return *p;
763 }
764 
765 static u32 vmx_read_guest_seg_ar(struct vcpu_vmx *vmx, unsigned seg)
766 {
767 	u32 *p = &vmx->segment_cache.seg[seg].ar;
768 
769 	if (!vmx_segment_cache_test_set(vmx, seg, SEG_FIELD_AR))
770 		*p = vmcs_read32(kvm_vmx_segment_fields[seg].ar_bytes);
771 	return *p;
772 }
773 
774 void update_exception_bitmap(struct kvm_vcpu *vcpu)
775 {
776 	u32 eb;
777 
778 	eb = (1u << PF_VECTOR) | (1u << UD_VECTOR) | (1u << MC_VECTOR) |
779 	     (1u << DB_VECTOR) | (1u << AC_VECTOR);
780 	/*
781 	 * Guest access to VMware backdoor ports could legitimately
782 	 * trigger #GP because of TSS I/O permission bitmap.
783 	 * We intercept those #GP and allow access to them anyway
784 	 * as VMware does.
785 	 */
786 	if (enable_vmware_backdoor)
787 		eb |= (1u << GP_VECTOR);
788 	if ((vcpu->guest_debug &
789 	     (KVM_GUESTDBG_ENABLE | KVM_GUESTDBG_USE_SW_BP)) ==
790 	    (KVM_GUESTDBG_ENABLE | KVM_GUESTDBG_USE_SW_BP))
791 		eb |= 1u << BP_VECTOR;
792 	if (to_vmx(vcpu)->rmode.vm86_active)
793 		eb = ~0;
794 	if (enable_ept)
795 		eb &= ~(1u << PF_VECTOR); /* bypass_guest_pf = 0 */
796 
797 	/* When we are running a nested L2 guest and L1 specified for it a
798 	 * certain exception bitmap, we must trap the same exceptions and pass
799 	 * them to L1. When running L2, we will only handle the exceptions
800 	 * specified above if L1 did not want them.
801 	 */
802 	if (is_guest_mode(vcpu))
803 		eb |= get_vmcs12(vcpu)->exception_bitmap;
804 
805 	vmcs_write32(EXCEPTION_BITMAP, eb);
806 }
807 
808 /*
809  * Check if MSR is intercepted for currently loaded MSR bitmap.
810  */
811 static bool msr_write_intercepted(struct kvm_vcpu *vcpu, u32 msr)
812 {
813 	unsigned long *msr_bitmap;
814 	int f = sizeof(unsigned long);
815 
816 	if (!cpu_has_vmx_msr_bitmap())
817 		return true;
818 
819 	msr_bitmap = to_vmx(vcpu)->loaded_vmcs->msr_bitmap;
820 
821 	if (msr <= 0x1fff) {
822 		return !!test_bit(msr, msr_bitmap + 0x800 / f);
823 	} else if ((msr >= 0xc0000000) && (msr <= 0xc0001fff)) {
824 		msr &= 0x1fff;
825 		return !!test_bit(msr, msr_bitmap + 0xc00 / f);
826 	}
827 
828 	return true;
829 }
830 
831 static void clear_atomic_switch_msr_special(struct vcpu_vmx *vmx,
832 		unsigned long entry, unsigned long exit)
833 {
834 	vm_entry_controls_clearbit(vmx, entry);
835 	vm_exit_controls_clearbit(vmx, exit);
836 }
837 
838 static int find_msr(struct vmx_msrs *m, unsigned int msr)
839 {
840 	unsigned int i;
841 
842 	for (i = 0; i < m->nr; ++i) {
843 		if (m->val[i].index == msr)
844 			return i;
845 	}
846 	return -ENOENT;
847 }
848 
849 static void clear_atomic_switch_msr(struct vcpu_vmx *vmx, unsigned msr)
850 {
851 	int i;
852 	struct msr_autoload *m = &vmx->msr_autoload;
853 
854 	switch (msr) {
855 	case MSR_EFER:
856 		if (cpu_has_load_ia32_efer()) {
857 			clear_atomic_switch_msr_special(vmx,
858 					VM_ENTRY_LOAD_IA32_EFER,
859 					VM_EXIT_LOAD_IA32_EFER);
860 			return;
861 		}
862 		break;
863 	case MSR_CORE_PERF_GLOBAL_CTRL:
864 		if (cpu_has_load_perf_global_ctrl()) {
865 			clear_atomic_switch_msr_special(vmx,
866 					VM_ENTRY_LOAD_IA32_PERF_GLOBAL_CTRL,
867 					VM_EXIT_LOAD_IA32_PERF_GLOBAL_CTRL);
868 			return;
869 		}
870 		break;
871 	}
872 	i = find_msr(&m->guest, msr);
873 	if (i < 0)
874 		goto skip_guest;
875 	--m->guest.nr;
876 	m->guest.val[i] = m->guest.val[m->guest.nr];
877 	vmcs_write32(VM_ENTRY_MSR_LOAD_COUNT, m->guest.nr);
878 
879 skip_guest:
880 	i = find_msr(&m->host, msr);
881 	if (i < 0)
882 		return;
883 
884 	--m->host.nr;
885 	m->host.val[i] = m->host.val[m->host.nr];
886 	vmcs_write32(VM_EXIT_MSR_LOAD_COUNT, m->host.nr);
887 }
888 
889 static void add_atomic_switch_msr_special(struct vcpu_vmx *vmx,
890 		unsigned long entry, unsigned long exit,
891 		unsigned long guest_val_vmcs, unsigned long host_val_vmcs,
892 		u64 guest_val, u64 host_val)
893 {
894 	vmcs_write64(guest_val_vmcs, guest_val);
895 	if (host_val_vmcs != HOST_IA32_EFER)
896 		vmcs_write64(host_val_vmcs, host_val);
897 	vm_entry_controls_setbit(vmx, entry);
898 	vm_exit_controls_setbit(vmx, exit);
899 }
900 
901 static void add_atomic_switch_msr(struct vcpu_vmx *vmx, unsigned msr,
902 				  u64 guest_val, u64 host_val, bool entry_only)
903 {
904 	int i, j = 0;
905 	struct msr_autoload *m = &vmx->msr_autoload;
906 
907 	switch (msr) {
908 	case MSR_EFER:
909 		if (cpu_has_load_ia32_efer()) {
910 			add_atomic_switch_msr_special(vmx,
911 					VM_ENTRY_LOAD_IA32_EFER,
912 					VM_EXIT_LOAD_IA32_EFER,
913 					GUEST_IA32_EFER,
914 					HOST_IA32_EFER,
915 					guest_val, host_val);
916 			return;
917 		}
918 		break;
919 	case MSR_CORE_PERF_GLOBAL_CTRL:
920 		if (cpu_has_load_perf_global_ctrl()) {
921 			add_atomic_switch_msr_special(vmx,
922 					VM_ENTRY_LOAD_IA32_PERF_GLOBAL_CTRL,
923 					VM_EXIT_LOAD_IA32_PERF_GLOBAL_CTRL,
924 					GUEST_IA32_PERF_GLOBAL_CTRL,
925 					HOST_IA32_PERF_GLOBAL_CTRL,
926 					guest_val, host_val);
927 			return;
928 		}
929 		break;
930 	case MSR_IA32_PEBS_ENABLE:
931 		/* PEBS needs a quiescent period after being disabled (to write
932 		 * a record).  Disabling PEBS through VMX MSR swapping doesn't
933 		 * provide that period, so a CPU could write host's record into
934 		 * guest's memory.
935 		 */
936 		wrmsrl(MSR_IA32_PEBS_ENABLE, 0);
937 	}
938 
939 	i = find_msr(&m->guest, msr);
940 	if (!entry_only)
941 		j = find_msr(&m->host, msr);
942 
943 	if ((i < 0 && m->guest.nr == NR_AUTOLOAD_MSRS) ||
944 		(j < 0 &&  m->host.nr == NR_AUTOLOAD_MSRS)) {
945 		printk_once(KERN_WARNING "Not enough msr switch entries. "
946 				"Can't add msr %x\n", msr);
947 		return;
948 	}
949 	if (i < 0) {
950 		i = m->guest.nr++;
951 		vmcs_write32(VM_ENTRY_MSR_LOAD_COUNT, m->guest.nr);
952 	}
953 	m->guest.val[i].index = msr;
954 	m->guest.val[i].value = guest_val;
955 
956 	if (entry_only)
957 		return;
958 
959 	if (j < 0) {
960 		j = m->host.nr++;
961 		vmcs_write32(VM_EXIT_MSR_LOAD_COUNT, m->host.nr);
962 	}
963 	m->host.val[j].index = msr;
964 	m->host.val[j].value = host_val;
965 }
966 
967 static bool update_transition_efer(struct vcpu_vmx *vmx, int efer_offset)
968 {
969 	u64 guest_efer = vmx->vcpu.arch.efer;
970 	u64 ignore_bits = 0;
971 
972 	/* Shadow paging assumes NX to be available.  */
973 	if (!enable_ept)
974 		guest_efer |= EFER_NX;
975 
976 	/*
977 	 * LMA and LME handled by hardware; SCE meaningless outside long mode.
978 	 */
979 	ignore_bits |= EFER_SCE;
980 #ifdef CONFIG_X86_64
981 	ignore_bits |= EFER_LMA | EFER_LME;
982 	/* SCE is meaningful only in long mode on Intel */
983 	if (guest_efer & EFER_LMA)
984 		ignore_bits &= ~(u64)EFER_SCE;
985 #endif
986 
987 	/*
988 	 * On EPT, we can't emulate NX, so we must switch EFER atomically.
989 	 * On CPUs that support "load IA32_EFER", always switch EFER
990 	 * atomically, since it's faster than switching it manually.
991 	 */
992 	if (cpu_has_load_ia32_efer() ||
993 	    (enable_ept && ((vmx->vcpu.arch.efer ^ host_efer) & EFER_NX))) {
994 		if (!(guest_efer & EFER_LMA))
995 			guest_efer &= ~EFER_LME;
996 		if (guest_efer != host_efer)
997 			add_atomic_switch_msr(vmx, MSR_EFER,
998 					      guest_efer, host_efer, false);
999 		else
1000 			clear_atomic_switch_msr(vmx, MSR_EFER);
1001 		return false;
1002 	} else {
1003 		clear_atomic_switch_msr(vmx, MSR_EFER);
1004 
1005 		guest_efer &= ~ignore_bits;
1006 		guest_efer |= host_efer & ignore_bits;
1007 
1008 		vmx->guest_msrs[efer_offset].data = guest_efer;
1009 		vmx->guest_msrs[efer_offset].mask = ~ignore_bits;
1010 
1011 		return true;
1012 	}
1013 }
1014 
1015 #ifdef CONFIG_X86_32
1016 /*
1017  * On 32-bit kernels, VM exits still load the FS and GS bases from the
1018  * VMCS rather than the segment table.  KVM uses this helper to figure
1019  * out the current bases to poke them into the VMCS before entry.
1020  */
1021 static unsigned long segment_base(u16 selector)
1022 {
1023 	struct desc_struct *table;
1024 	unsigned long v;
1025 
1026 	if (!(selector & ~SEGMENT_RPL_MASK))
1027 		return 0;
1028 
1029 	table = get_current_gdt_ro();
1030 
1031 	if ((selector & SEGMENT_TI_MASK) == SEGMENT_LDT) {
1032 		u16 ldt_selector = kvm_read_ldt();
1033 
1034 		if (!(ldt_selector & ~SEGMENT_RPL_MASK))
1035 			return 0;
1036 
1037 		table = (struct desc_struct *)segment_base(ldt_selector);
1038 	}
1039 	v = get_desc_base(&table[selector >> 3]);
1040 	return v;
1041 }
1042 #endif
1043 
1044 static inline void pt_load_msr(struct pt_ctx *ctx, u32 addr_range)
1045 {
1046 	u32 i;
1047 
1048 	wrmsrl(MSR_IA32_RTIT_STATUS, ctx->status);
1049 	wrmsrl(MSR_IA32_RTIT_OUTPUT_BASE, ctx->output_base);
1050 	wrmsrl(MSR_IA32_RTIT_OUTPUT_MASK, ctx->output_mask);
1051 	wrmsrl(MSR_IA32_RTIT_CR3_MATCH, ctx->cr3_match);
1052 	for (i = 0; i < addr_range; i++) {
1053 		wrmsrl(MSR_IA32_RTIT_ADDR0_A + i * 2, ctx->addr_a[i]);
1054 		wrmsrl(MSR_IA32_RTIT_ADDR0_B + i * 2, ctx->addr_b[i]);
1055 	}
1056 }
1057 
1058 static inline void pt_save_msr(struct pt_ctx *ctx, u32 addr_range)
1059 {
1060 	u32 i;
1061 
1062 	rdmsrl(MSR_IA32_RTIT_STATUS, ctx->status);
1063 	rdmsrl(MSR_IA32_RTIT_OUTPUT_BASE, ctx->output_base);
1064 	rdmsrl(MSR_IA32_RTIT_OUTPUT_MASK, ctx->output_mask);
1065 	rdmsrl(MSR_IA32_RTIT_CR3_MATCH, ctx->cr3_match);
1066 	for (i = 0; i < addr_range; i++) {
1067 		rdmsrl(MSR_IA32_RTIT_ADDR0_A + i * 2, ctx->addr_a[i]);
1068 		rdmsrl(MSR_IA32_RTIT_ADDR0_B + i * 2, ctx->addr_b[i]);
1069 	}
1070 }
1071 
1072 static void pt_guest_enter(struct vcpu_vmx *vmx)
1073 {
1074 	if (pt_mode == PT_MODE_SYSTEM)
1075 		return;
1076 
1077 	/*
1078 	 * GUEST_IA32_RTIT_CTL is already set in the VMCS.
1079 	 * Save host state before VM entry.
1080 	 */
1081 	rdmsrl(MSR_IA32_RTIT_CTL, vmx->pt_desc.host.ctl);
1082 	if (vmx->pt_desc.guest.ctl & RTIT_CTL_TRACEEN) {
1083 		wrmsrl(MSR_IA32_RTIT_CTL, 0);
1084 		pt_save_msr(&vmx->pt_desc.host, vmx->pt_desc.addr_range);
1085 		pt_load_msr(&vmx->pt_desc.guest, vmx->pt_desc.addr_range);
1086 	}
1087 }
1088 
1089 static void pt_guest_exit(struct vcpu_vmx *vmx)
1090 {
1091 	if (pt_mode == PT_MODE_SYSTEM)
1092 		return;
1093 
1094 	if (vmx->pt_desc.guest.ctl & RTIT_CTL_TRACEEN) {
1095 		pt_save_msr(&vmx->pt_desc.guest, vmx->pt_desc.addr_range);
1096 		pt_load_msr(&vmx->pt_desc.host, vmx->pt_desc.addr_range);
1097 	}
1098 
1099 	/* Reload host state (IA32_RTIT_CTL will be cleared on VM exit). */
1100 	wrmsrl(MSR_IA32_RTIT_CTL, vmx->pt_desc.host.ctl);
1101 }
1102 
1103 void vmx_set_host_fs_gs(struct vmcs_host_state *host, u16 fs_sel, u16 gs_sel,
1104 			unsigned long fs_base, unsigned long gs_base)
1105 {
1106 	if (unlikely(fs_sel != host->fs_sel)) {
1107 		if (!(fs_sel & 7))
1108 			vmcs_write16(HOST_FS_SELECTOR, fs_sel);
1109 		else
1110 			vmcs_write16(HOST_FS_SELECTOR, 0);
1111 		host->fs_sel = fs_sel;
1112 	}
1113 	if (unlikely(gs_sel != host->gs_sel)) {
1114 		if (!(gs_sel & 7))
1115 			vmcs_write16(HOST_GS_SELECTOR, gs_sel);
1116 		else
1117 			vmcs_write16(HOST_GS_SELECTOR, 0);
1118 		host->gs_sel = gs_sel;
1119 	}
1120 	if (unlikely(fs_base != host->fs_base)) {
1121 		vmcs_writel(HOST_FS_BASE, fs_base);
1122 		host->fs_base = fs_base;
1123 	}
1124 	if (unlikely(gs_base != host->gs_base)) {
1125 		vmcs_writel(HOST_GS_BASE, gs_base);
1126 		host->gs_base = gs_base;
1127 	}
1128 }
1129 
1130 void vmx_prepare_switch_to_guest(struct kvm_vcpu *vcpu)
1131 {
1132 	struct vcpu_vmx *vmx = to_vmx(vcpu);
1133 	struct vmcs_host_state *host_state;
1134 #ifdef CONFIG_X86_64
1135 	int cpu = raw_smp_processor_id();
1136 #endif
1137 	unsigned long fs_base, gs_base;
1138 	u16 fs_sel, gs_sel;
1139 	int i;
1140 
1141 	vmx->req_immediate_exit = false;
1142 
1143 	/*
1144 	 * Note that guest MSRs to be saved/restored can also be changed
1145 	 * when guest state is loaded. This happens when guest transitions
1146 	 * to/from long-mode by setting MSR_EFER.LMA.
1147 	 */
1148 	if (!vmx->guest_msrs_ready) {
1149 		vmx->guest_msrs_ready = true;
1150 		for (i = 0; i < vmx->save_nmsrs; ++i)
1151 			kvm_set_shared_msr(vmx->guest_msrs[i].index,
1152 					   vmx->guest_msrs[i].data,
1153 					   vmx->guest_msrs[i].mask);
1154 
1155 	}
1156 	if (vmx->guest_state_loaded)
1157 		return;
1158 
1159 	host_state = &vmx->loaded_vmcs->host_state;
1160 
1161 	/*
1162 	 * Set host fs and gs selectors.  Unfortunately, 22.2.3 does not
1163 	 * allow segment selectors with cpl > 0 or ti == 1.
1164 	 */
1165 	host_state->ldt_sel = kvm_read_ldt();
1166 
1167 #ifdef CONFIG_X86_64
1168 	savesegment(ds, host_state->ds_sel);
1169 	savesegment(es, host_state->es_sel);
1170 
1171 	gs_base = cpu_kernelmode_gs_base(cpu);
1172 	if (likely(is_64bit_mm(current->mm))) {
1173 		save_fsgs_for_kvm();
1174 		fs_sel = current->thread.fsindex;
1175 		gs_sel = current->thread.gsindex;
1176 		fs_base = current->thread.fsbase;
1177 		vmx->msr_host_kernel_gs_base = current->thread.gsbase;
1178 	} else {
1179 		savesegment(fs, fs_sel);
1180 		savesegment(gs, gs_sel);
1181 		fs_base = read_msr(MSR_FS_BASE);
1182 		vmx->msr_host_kernel_gs_base = read_msr(MSR_KERNEL_GS_BASE);
1183 	}
1184 
1185 	wrmsrl(MSR_KERNEL_GS_BASE, vmx->msr_guest_kernel_gs_base);
1186 #else
1187 	savesegment(fs, fs_sel);
1188 	savesegment(gs, gs_sel);
1189 	fs_base = segment_base(fs_sel);
1190 	gs_base = segment_base(gs_sel);
1191 #endif
1192 
1193 	vmx_set_host_fs_gs(host_state, fs_sel, gs_sel, fs_base, gs_base);
1194 	vmx->guest_state_loaded = true;
1195 }
1196 
1197 static void vmx_prepare_switch_to_host(struct vcpu_vmx *vmx)
1198 {
1199 	struct vmcs_host_state *host_state;
1200 
1201 	if (!vmx->guest_state_loaded)
1202 		return;
1203 
1204 	host_state = &vmx->loaded_vmcs->host_state;
1205 
1206 	++vmx->vcpu.stat.host_state_reload;
1207 
1208 #ifdef CONFIG_X86_64
1209 	rdmsrl(MSR_KERNEL_GS_BASE, vmx->msr_guest_kernel_gs_base);
1210 #endif
1211 	if (host_state->ldt_sel || (host_state->gs_sel & 7)) {
1212 		kvm_load_ldt(host_state->ldt_sel);
1213 #ifdef CONFIG_X86_64
1214 		load_gs_index(host_state->gs_sel);
1215 #else
1216 		loadsegment(gs, host_state->gs_sel);
1217 #endif
1218 	}
1219 	if (host_state->fs_sel & 7)
1220 		loadsegment(fs, host_state->fs_sel);
1221 #ifdef CONFIG_X86_64
1222 	if (unlikely(host_state->ds_sel | host_state->es_sel)) {
1223 		loadsegment(ds, host_state->ds_sel);
1224 		loadsegment(es, host_state->es_sel);
1225 	}
1226 #endif
1227 	invalidate_tss_limit();
1228 #ifdef CONFIG_X86_64
1229 	wrmsrl(MSR_KERNEL_GS_BASE, vmx->msr_host_kernel_gs_base);
1230 #endif
1231 	load_fixmap_gdt(raw_smp_processor_id());
1232 	vmx->guest_state_loaded = false;
1233 	vmx->guest_msrs_ready = false;
1234 }
1235 
1236 #ifdef CONFIG_X86_64
1237 static u64 vmx_read_guest_kernel_gs_base(struct vcpu_vmx *vmx)
1238 {
1239 	preempt_disable();
1240 	if (vmx->guest_state_loaded)
1241 		rdmsrl(MSR_KERNEL_GS_BASE, vmx->msr_guest_kernel_gs_base);
1242 	preempt_enable();
1243 	return vmx->msr_guest_kernel_gs_base;
1244 }
1245 
1246 static void vmx_write_guest_kernel_gs_base(struct vcpu_vmx *vmx, u64 data)
1247 {
1248 	preempt_disable();
1249 	if (vmx->guest_state_loaded)
1250 		wrmsrl(MSR_KERNEL_GS_BASE, data);
1251 	preempt_enable();
1252 	vmx->msr_guest_kernel_gs_base = data;
1253 }
1254 #endif
1255 
1256 static void vmx_vcpu_pi_load(struct kvm_vcpu *vcpu, int cpu)
1257 {
1258 	struct pi_desc *pi_desc = vcpu_to_pi_desc(vcpu);
1259 	struct pi_desc old, new;
1260 	unsigned int dest;
1261 
1262 	/*
1263 	 * In case of hot-plug or hot-unplug, we may have to undo
1264 	 * vmx_vcpu_pi_put even if there is no assigned device.  And we
1265 	 * always keep PI.NDST up to date for simplicity: it makes the
1266 	 * code easier, and CPU migration is not a fast path.
1267 	 */
1268 	if (!pi_test_sn(pi_desc) && vcpu->cpu == cpu)
1269 		return;
1270 
1271 	/* The full case.  */
1272 	do {
1273 		old.control = new.control = pi_desc->control;
1274 
1275 		dest = cpu_physical_id(cpu);
1276 
1277 		if (x2apic_enabled())
1278 			new.ndst = dest;
1279 		else
1280 			new.ndst = (dest << 8) & 0xFF00;
1281 
1282 		new.sn = 0;
1283 	} while (cmpxchg64(&pi_desc->control, old.control,
1284 			   new.control) != old.control);
1285 
1286 	/*
1287 	 * Clear SN before reading the bitmap.  The VT-d firmware
1288 	 * writes the bitmap and reads SN atomically (5.2.3 in the
1289 	 * spec), so it doesn't really have a memory barrier that
1290 	 * pairs with this, but we cannot do that and we need one.
1291 	 */
1292 	smp_mb__after_atomic();
1293 
1294 	if (!bitmap_empty((unsigned long *)pi_desc->pir, NR_VECTORS))
1295 		pi_set_on(pi_desc);
1296 }
1297 
1298 void vmx_vcpu_load_vmcs(struct kvm_vcpu *vcpu, int cpu)
1299 {
1300 	struct vcpu_vmx *vmx = to_vmx(vcpu);
1301 	bool already_loaded = vmx->loaded_vmcs->cpu == cpu;
1302 
1303 	if (!already_loaded) {
1304 		loaded_vmcs_clear(vmx->loaded_vmcs);
1305 		local_irq_disable();
1306 		crash_disable_local_vmclear(cpu);
1307 
1308 		/*
1309 		 * Read loaded_vmcs->cpu should be before fetching
1310 		 * loaded_vmcs->loaded_vmcss_on_cpu_link.
1311 		 * See the comments in __loaded_vmcs_clear().
1312 		 */
1313 		smp_rmb();
1314 
1315 		list_add(&vmx->loaded_vmcs->loaded_vmcss_on_cpu_link,
1316 			 &per_cpu(loaded_vmcss_on_cpu, cpu));
1317 		crash_enable_local_vmclear(cpu);
1318 		local_irq_enable();
1319 	}
1320 
1321 	if (per_cpu(current_vmcs, cpu) != vmx->loaded_vmcs->vmcs) {
1322 		per_cpu(current_vmcs, cpu) = vmx->loaded_vmcs->vmcs;
1323 		vmcs_load(vmx->loaded_vmcs->vmcs);
1324 		indirect_branch_prediction_barrier();
1325 	}
1326 
1327 	if (!already_loaded) {
1328 		void *gdt = get_current_gdt_ro();
1329 		unsigned long sysenter_esp;
1330 
1331 		kvm_make_request(KVM_REQ_TLB_FLUSH, vcpu);
1332 
1333 		/*
1334 		 * Linux uses per-cpu TSS and GDT, so set these when switching
1335 		 * processors.  See 22.2.4.
1336 		 */
1337 		vmcs_writel(HOST_TR_BASE,
1338 			    (unsigned long)&get_cpu_entry_area(cpu)->tss.x86_tss);
1339 		vmcs_writel(HOST_GDTR_BASE, (unsigned long)gdt);   /* 22.2.4 */
1340 
1341 		/*
1342 		 * VM exits change the host TR limit to 0x67 after a VM
1343 		 * exit.  This is okay, since 0x67 covers everything except
1344 		 * the IO bitmap and have have code to handle the IO bitmap
1345 		 * being lost after a VM exit.
1346 		 */
1347 		BUILD_BUG_ON(IO_BITMAP_OFFSET - 1 != 0x67);
1348 
1349 		rdmsrl(MSR_IA32_SYSENTER_ESP, sysenter_esp);
1350 		vmcs_writel(HOST_IA32_SYSENTER_ESP, sysenter_esp); /* 22.2.3 */
1351 
1352 		vmx->loaded_vmcs->cpu = cpu;
1353 	}
1354 
1355 	/* Setup TSC multiplier */
1356 	if (kvm_has_tsc_control &&
1357 	    vmx->current_tsc_ratio != vcpu->arch.tsc_scaling_ratio)
1358 		decache_tsc_multiplier(vmx);
1359 }
1360 
1361 /*
1362  * Switches to specified vcpu, until a matching vcpu_put(), but assumes
1363  * vcpu mutex is already taken.
1364  */
1365 void vmx_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
1366 {
1367 	struct vcpu_vmx *vmx = to_vmx(vcpu);
1368 
1369 	vmx_vcpu_load_vmcs(vcpu, cpu);
1370 
1371 	vmx_vcpu_pi_load(vcpu, cpu);
1372 
1373 	vmx->host_pkru = read_pkru();
1374 	vmx->host_debugctlmsr = get_debugctlmsr();
1375 }
1376 
1377 static void vmx_vcpu_pi_put(struct kvm_vcpu *vcpu)
1378 {
1379 	struct pi_desc *pi_desc = vcpu_to_pi_desc(vcpu);
1380 
1381 	if (!kvm_arch_has_assigned_device(vcpu->kvm) ||
1382 		!irq_remapping_cap(IRQ_POSTING_CAP)  ||
1383 		!kvm_vcpu_apicv_active(vcpu))
1384 		return;
1385 
1386 	/* Set SN when the vCPU is preempted */
1387 	if (vcpu->preempted)
1388 		pi_set_sn(pi_desc);
1389 }
1390 
1391 static void vmx_vcpu_put(struct kvm_vcpu *vcpu)
1392 {
1393 	vmx_vcpu_pi_put(vcpu);
1394 
1395 	vmx_prepare_switch_to_host(to_vmx(vcpu));
1396 }
1397 
1398 static bool emulation_required(struct kvm_vcpu *vcpu)
1399 {
1400 	return emulate_invalid_guest_state && !guest_state_valid(vcpu);
1401 }
1402 
1403 static void vmx_decache_cr0_guest_bits(struct kvm_vcpu *vcpu);
1404 
1405 unsigned long vmx_get_rflags(struct kvm_vcpu *vcpu)
1406 {
1407 	unsigned long rflags, save_rflags;
1408 
1409 	if (!test_bit(VCPU_EXREG_RFLAGS, (ulong *)&vcpu->arch.regs_avail)) {
1410 		__set_bit(VCPU_EXREG_RFLAGS, (ulong *)&vcpu->arch.regs_avail);
1411 		rflags = vmcs_readl(GUEST_RFLAGS);
1412 		if (to_vmx(vcpu)->rmode.vm86_active) {
1413 			rflags &= RMODE_GUEST_OWNED_EFLAGS_BITS;
1414 			save_rflags = to_vmx(vcpu)->rmode.save_rflags;
1415 			rflags |= save_rflags & ~RMODE_GUEST_OWNED_EFLAGS_BITS;
1416 		}
1417 		to_vmx(vcpu)->rflags = rflags;
1418 	}
1419 	return to_vmx(vcpu)->rflags;
1420 }
1421 
1422 void vmx_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags)
1423 {
1424 	unsigned long old_rflags = vmx_get_rflags(vcpu);
1425 
1426 	__set_bit(VCPU_EXREG_RFLAGS, (ulong *)&vcpu->arch.regs_avail);
1427 	to_vmx(vcpu)->rflags = rflags;
1428 	if (to_vmx(vcpu)->rmode.vm86_active) {
1429 		to_vmx(vcpu)->rmode.save_rflags = rflags;
1430 		rflags |= X86_EFLAGS_IOPL | X86_EFLAGS_VM;
1431 	}
1432 	vmcs_writel(GUEST_RFLAGS, rflags);
1433 
1434 	if ((old_rflags ^ to_vmx(vcpu)->rflags) & X86_EFLAGS_VM)
1435 		to_vmx(vcpu)->emulation_required = emulation_required(vcpu);
1436 }
1437 
1438 u32 vmx_get_interrupt_shadow(struct kvm_vcpu *vcpu)
1439 {
1440 	u32 interruptibility = vmcs_read32(GUEST_INTERRUPTIBILITY_INFO);
1441 	int ret = 0;
1442 
1443 	if (interruptibility & GUEST_INTR_STATE_STI)
1444 		ret |= KVM_X86_SHADOW_INT_STI;
1445 	if (interruptibility & GUEST_INTR_STATE_MOV_SS)
1446 		ret |= KVM_X86_SHADOW_INT_MOV_SS;
1447 
1448 	return ret;
1449 }
1450 
1451 void vmx_set_interrupt_shadow(struct kvm_vcpu *vcpu, int mask)
1452 {
1453 	u32 interruptibility_old = vmcs_read32(GUEST_INTERRUPTIBILITY_INFO);
1454 	u32 interruptibility = interruptibility_old;
1455 
1456 	interruptibility &= ~(GUEST_INTR_STATE_STI | GUEST_INTR_STATE_MOV_SS);
1457 
1458 	if (mask & KVM_X86_SHADOW_INT_MOV_SS)
1459 		interruptibility |= GUEST_INTR_STATE_MOV_SS;
1460 	else if (mask & KVM_X86_SHADOW_INT_STI)
1461 		interruptibility |= GUEST_INTR_STATE_STI;
1462 
1463 	if ((interruptibility != interruptibility_old))
1464 		vmcs_write32(GUEST_INTERRUPTIBILITY_INFO, interruptibility);
1465 }
1466 
1467 static int vmx_rtit_ctl_check(struct kvm_vcpu *vcpu, u64 data)
1468 {
1469 	struct vcpu_vmx *vmx = to_vmx(vcpu);
1470 	unsigned long value;
1471 
1472 	/*
1473 	 * Any MSR write that attempts to change bits marked reserved will
1474 	 * case a #GP fault.
1475 	 */
1476 	if (data & vmx->pt_desc.ctl_bitmask)
1477 		return 1;
1478 
1479 	/*
1480 	 * Any attempt to modify IA32_RTIT_CTL while TraceEn is set will
1481 	 * result in a #GP unless the same write also clears TraceEn.
1482 	 */
1483 	if ((vmx->pt_desc.guest.ctl & RTIT_CTL_TRACEEN) &&
1484 		((vmx->pt_desc.guest.ctl ^ data) & ~RTIT_CTL_TRACEEN))
1485 		return 1;
1486 
1487 	/*
1488 	 * WRMSR to IA32_RTIT_CTL that sets TraceEn but clears this bit
1489 	 * and FabricEn would cause #GP, if
1490 	 * CPUID.(EAX=14H, ECX=0):ECX.SNGLRGNOUT[bit 2] = 0
1491 	 */
1492 	if ((data & RTIT_CTL_TRACEEN) && !(data & RTIT_CTL_TOPA) &&
1493 		!(data & RTIT_CTL_FABRIC_EN) &&
1494 		!intel_pt_validate_cap(vmx->pt_desc.caps,
1495 					PT_CAP_single_range_output))
1496 		return 1;
1497 
1498 	/*
1499 	 * MTCFreq, CycThresh and PSBFreq encodings check, any MSR write that
1500 	 * utilize encodings marked reserved will casue a #GP fault.
1501 	 */
1502 	value = intel_pt_validate_cap(vmx->pt_desc.caps, PT_CAP_mtc_periods);
1503 	if (intel_pt_validate_cap(vmx->pt_desc.caps, PT_CAP_mtc) &&
1504 			!test_bit((data & RTIT_CTL_MTC_RANGE) >>
1505 			RTIT_CTL_MTC_RANGE_OFFSET, &value))
1506 		return 1;
1507 	value = intel_pt_validate_cap(vmx->pt_desc.caps,
1508 						PT_CAP_cycle_thresholds);
1509 	if (intel_pt_validate_cap(vmx->pt_desc.caps, PT_CAP_psb_cyc) &&
1510 			!test_bit((data & RTIT_CTL_CYC_THRESH) >>
1511 			RTIT_CTL_CYC_THRESH_OFFSET, &value))
1512 		return 1;
1513 	value = intel_pt_validate_cap(vmx->pt_desc.caps, PT_CAP_psb_periods);
1514 	if (intel_pt_validate_cap(vmx->pt_desc.caps, PT_CAP_psb_cyc) &&
1515 			!test_bit((data & RTIT_CTL_PSB_FREQ) >>
1516 			RTIT_CTL_PSB_FREQ_OFFSET, &value))
1517 		return 1;
1518 
1519 	/*
1520 	 * If ADDRx_CFG is reserved or the encodings is >2 will
1521 	 * cause a #GP fault.
1522 	 */
1523 	value = (data & RTIT_CTL_ADDR0) >> RTIT_CTL_ADDR0_OFFSET;
1524 	if ((value && (vmx->pt_desc.addr_range < 1)) || (value > 2))
1525 		return 1;
1526 	value = (data & RTIT_CTL_ADDR1) >> RTIT_CTL_ADDR1_OFFSET;
1527 	if ((value && (vmx->pt_desc.addr_range < 2)) || (value > 2))
1528 		return 1;
1529 	value = (data & RTIT_CTL_ADDR2) >> RTIT_CTL_ADDR2_OFFSET;
1530 	if ((value && (vmx->pt_desc.addr_range < 3)) || (value > 2))
1531 		return 1;
1532 	value = (data & RTIT_CTL_ADDR3) >> RTIT_CTL_ADDR3_OFFSET;
1533 	if ((value && (vmx->pt_desc.addr_range < 4)) || (value > 2))
1534 		return 1;
1535 
1536 	return 0;
1537 }
1538 
1539 static int skip_emulated_instruction(struct kvm_vcpu *vcpu)
1540 {
1541 	unsigned long rip;
1542 
1543 	/*
1544 	 * Using VMCS.VM_EXIT_INSTRUCTION_LEN on EPT misconfig depends on
1545 	 * undefined behavior: Intel's SDM doesn't mandate the VMCS field be
1546 	 * set when EPT misconfig occurs.  In practice, real hardware updates
1547 	 * VM_EXIT_INSTRUCTION_LEN on EPT misconfig, but other hypervisors
1548 	 * (namely Hyper-V) don't set it due to it being undefined behavior,
1549 	 * i.e. we end up advancing IP with some random value.
1550 	 */
1551 	if (!static_cpu_has(X86_FEATURE_HYPERVISOR) ||
1552 	    to_vmx(vcpu)->exit_reason != EXIT_REASON_EPT_MISCONFIG) {
1553 		rip = kvm_rip_read(vcpu);
1554 		rip += vmcs_read32(VM_EXIT_INSTRUCTION_LEN);
1555 		kvm_rip_write(vcpu, rip);
1556 	} else {
1557 		if (!kvm_emulate_instruction(vcpu, EMULTYPE_SKIP))
1558 			return 0;
1559 	}
1560 
1561 	/* skipping an emulated instruction also counts */
1562 	vmx_set_interrupt_shadow(vcpu, 0);
1563 
1564 	return 1;
1565 }
1566 
1567 static void vmx_clear_hlt(struct kvm_vcpu *vcpu)
1568 {
1569 	/*
1570 	 * Ensure that we clear the HLT state in the VMCS.  We don't need to
1571 	 * explicitly skip the instruction because if the HLT state is set,
1572 	 * then the instruction is already executing and RIP has already been
1573 	 * advanced.
1574 	 */
1575 	if (kvm_hlt_in_guest(vcpu->kvm) &&
1576 			vmcs_read32(GUEST_ACTIVITY_STATE) == GUEST_ACTIVITY_HLT)
1577 		vmcs_write32(GUEST_ACTIVITY_STATE, GUEST_ACTIVITY_ACTIVE);
1578 }
1579 
1580 static void vmx_queue_exception(struct kvm_vcpu *vcpu)
1581 {
1582 	struct vcpu_vmx *vmx = to_vmx(vcpu);
1583 	unsigned nr = vcpu->arch.exception.nr;
1584 	bool has_error_code = vcpu->arch.exception.has_error_code;
1585 	u32 error_code = vcpu->arch.exception.error_code;
1586 	u32 intr_info = nr | INTR_INFO_VALID_MASK;
1587 
1588 	kvm_deliver_exception_payload(vcpu);
1589 
1590 	if (has_error_code) {
1591 		vmcs_write32(VM_ENTRY_EXCEPTION_ERROR_CODE, error_code);
1592 		intr_info |= INTR_INFO_DELIVER_CODE_MASK;
1593 	}
1594 
1595 	if (vmx->rmode.vm86_active) {
1596 		int inc_eip = 0;
1597 		if (kvm_exception_is_soft(nr))
1598 			inc_eip = vcpu->arch.event_exit_inst_len;
1599 		kvm_inject_realmode_interrupt(vcpu, nr, inc_eip);
1600 		return;
1601 	}
1602 
1603 	WARN_ON_ONCE(vmx->emulation_required);
1604 
1605 	if (kvm_exception_is_soft(nr)) {
1606 		vmcs_write32(VM_ENTRY_INSTRUCTION_LEN,
1607 			     vmx->vcpu.arch.event_exit_inst_len);
1608 		intr_info |= INTR_TYPE_SOFT_EXCEPTION;
1609 	} else
1610 		intr_info |= INTR_TYPE_HARD_EXCEPTION;
1611 
1612 	vmcs_write32(VM_ENTRY_INTR_INFO_FIELD, intr_info);
1613 
1614 	vmx_clear_hlt(vcpu);
1615 }
1616 
1617 static bool vmx_rdtscp_supported(void)
1618 {
1619 	return cpu_has_vmx_rdtscp();
1620 }
1621 
1622 static bool vmx_invpcid_supported(void)
1623 {
1624 	return cpu_has_vmx_invpcid();
1625 }
1626 
1627 /*
1628  * Swap MSR entry in host/guest MSR entry array.
1629  */
1630 static void move_msr_up(struct vcpu_vmx *vmx, int from, int to)
1631 {
1632 	struct shared_msr_entry tmp;
1633 
1634 	tmp = vmx->guest_msrs[to];
1635 	vmx->guest_msrs[to] = vmx->guest_msrs[from];
1636 	vmx->guest_msrs[from] = tmp;
1637 }
1638 
1639 /*
1640  * Set up the vmcs to automatically save and restore system
1641  * msrs.  Don't touch the 64-bit msrs if the guest is in legacy
1642  * mode, as fiddling with msrs is very expensive.
1643  */
1644 static void setup_msrs(struct vcpu_vmx *vmx)
1645 {
1646 	int save_nmsrs, index;
1647 
1648 	save_nmsrs = 0;
1649 #ifdef CONFIG_X86_64
1650 	/*
1651 	 * The SYSCALL MSRs are only needed on long mode guests, and only
1652 	 * when EFER.SCE is set.
1653 	 */
1654 	if (is_long_mode(&vmx->vcpu) && (vmx->vcpu.arch.efer & EFER_SCE)) {
1655 		index = __find_msr_index(vmx, MSR_STAR);
1656 		if (index >= 0)
1657 			move_msr_up(vmx, index, save_nmsrs++);
1658 		index = __find_msr_index(vmx, MSR_LSTAR);
1659 		if (index >= 0)
1660 			move_msr_up(vmx, index, save_nmsrs++);
1661 		index = __find_msr_index(vmx, MSR_SYSCALL_MASK);
1662 		if (index >= 0)
1663 			move_msr_up(vmx, index, save_nmsrs++);
1664 	}
1665 #endif
1666 	index = __find_msr_index(vmx, MSR_EFER);
1667 	if (index >= 0 && update_transition_efer(vmx, index))
1668 		move_msr_up(vmx, index, save_nmsrs++);
1669 	index = __find_msr_index(vmx, MSR_TSC_AUX);
1670 	if (index >= 0 && guest_cpuid_has(&vmx->vcpu, X86_FEATURE_RDTSCP))
1671 		move_msr_up(vmx, index, save_nmsrs++);
1672 
1673 	vmx->save_nmsrs = save_nmsrs;
1674 	vmx->guest_msrs_ready = false;
1675 
1676 	if (cpu_has_vmx_msr_bitmap())
1677 		vmx_update_msr_bitmap(&vmx->vcpu);
1678 }
1679 
1680 static u64 vmx_read_l1_tsc_offset(struct kvm_vcpu *vcpu)
1681 {
1682 	struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
1683 
1684 	if (is_guest_mode(vcpu) &&
1685 	    (vmcs12->cpu_based_vm_exec_control & CPU_BASED_USE_TSC_OFFSETING))
1686 		return vcpu->arch.tsc_offset - vmcs12->tsc_offset;
1687 
1688 	return vcpu->arch.tsc_offset;
1689 }
1690 
1691 static u64 vmx_write_l1_tsc_offset(struct kvm_vcpu *vcpu, u64 offset)
1692 {
1693 	struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
1694 	u64 g_tsc_offset = 0;
1695 
1696 	/*
1697 	 * We're here if L1 chose not to trap WRMSR to TSC. According
1698 	 * to the spec, this should set L1's TSC; The offset that L1
1699 	 * set for L2 remains unchanged, and still needs to be added
1700 	 * to the newly set TSC to get L2's TSC.
1701 	 */
1702 	if (is_guest_mode(vcpu) &&
1703 	    (vmcs12->cpu_based_vm_exec_control & CPU_BASED_USE_TSC_OFFSETING))
1704 		g_tsc_offset = vmcs12->tsc_offset;
1705 
1706 	trace_kvm_write_tsc_offset(vcpu->vcpu_id,
1707 				   vcpu->arch.tsc_offset - g_tsc_offset,
1708 				   offset);
1709 	vmcs_write64(TSC_OFFSET, offset + g_tsc_offset);
1710 	return offset + g_tsc_offset;
1711 }
1712 
1713 /*
1714  * nested_vmx_allowed() checks whether a guest should be allowed to use VMX
1715  * instructions and MSRs (i.e., nested VMX). Nested VMX is disabled for
1716  * all guests if the "nested" module option is off, and can also be disabled
1717  * for a single guest by disabling its VMX cpuid bit.
1718  */
1719 bool nested_vmx_allowed(struct kvm_vcpu *vcpu)
1720 {
1721 	return nested && guest_cpuid_has(vcpu, X86_FEATURE_VMX);
1722 }
1723 
1724 static inline bool vmx_feature_control_msr_valid(struct kvm_vcpu *vcpu,
1725 						 uint64_t val)
1726 {
1727 	uint64_t valid_bits = to_vmx(vcpu)->msr_ia32_feature_control_valid_bits;
1728 
1729 	return !(val & ~valid_bits);
1730 }
1731 
1732 static int vmx_get_msr_feature(struct kvm_msr_entry *msr)
1733 {
1734 	switch (msr->index) {
1735 	case MSR_IA32_VMX_BASIC ... MSR_IA32_VMX_VMFUNC:
1736 		if (!nested)
1737 			return 1;
1738 		return vmx_get_vmx_msr(&vmcs_config.nested, msr->index, &msr->data);
1739 	default:
1740 		return 1;
1741 	}
1742 
1743 	return 0;
1744 }
1745 
1746 /*
1747  * Reads an msr value (of 'msr_index') into 'pdata'.
1748  * Returns 0 on success, non-0 otherwise.
1749  * Assumes vcpu_load() was already called.
1750  */
1751 static int vmx_get_msr(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
1752 {
1753 	struct vcpu_vmx *vmx = to_vmx(vcpu);
1754 	struct shared_msr_entry *msr;
1755 	u32 index;
1756 
1757 	switch (msr_info->index) {
1758 #ifdef CONFIG_X86_64
1759 	case MSR_FS_BASE:
1760 		msr_info->data = vmcs_readl(GUEST_FS_BASE);
1761 		break;
1762 	case MSR_GS_BASE:
1763 		msr_info->data = vmcs_readl(GUEST_GS_BASE);
1764 		break;
1765 	case MSR_KERNEL_GS_BASE:
1766 		msr_info->data = vmx_read_guest_kernel_gs_base(vmx);
1767 		break;
1768 #endif
1769 	case MSR_EFER:
1770 		return kvm_get_msr_common(vcpu, msr_info);
1771 	case MSR_IA32_UMWAIT_CONTROL:
1772 		if (!msr_info->host_initiated && !vmx_has_waitpkg(vmx))
1773 			return 1;
1774 
1775 		msr_info->data = vmx->msr_ia32_umwait_control;
1776 		break;
1777 	case MSR_IA32_SPEC_CTRL:
1778 		if (!msr_info->host_initiated &&
1779 		    !guest_cpuid_has(vcpu, X86_FEATURE_SPEC_CTRL))
1780 			return 1;
1781 
1782 		msr_info->data = to_vmx(vcpu)->spec_ctrl;
1783 		break;
1784 	case MSR_IA32_SYSENTER_CS:
1785 		msr_info->data = vmcs_read32(GUEST_SYSENTER_CS);
1786 		break;
1787 	case MSR_IA32_SYSENTER_EIP:
1788 		msr_info->data = vmcs_readl(GUEST_SYSENTER_EIP);
1789 		break;
1790 	case MSR_IA32_SYSENTER_ESP:
1791 		msr_info->data = vmcs_readl(GUEST_SYSENTER_ESP);
1792 		break;
1793 	case MSR_IA32_BNDCFGS:
1794 		if (!kvm_mpx_supported() ||
1795 		    (!msr_info->host_initiated &&
1796 		     !guest_cpuid_has(vcpu, X86_FEATURE_MPX)))
1797 			return 1;
1798 		msr_info->data = vmcs_read64(GUEST_BNDCFGS);
1799 		break;
1800 	case MSR_IA32_MCG_EXT_CTL:
1801 		if (!msr_info->host_initiated &&
1802 		    !(vmx->msr_ia32_feature_control &
1803 		      FEATURE_CONTROL_LMCE))
1804 			return 1;
1805 		msr_info->data = vcpu->arch.mcg_ext_ctl;
1806 		break;
1807 	case MSR_IA32_FEATURE_CONTROL:
1808 		msr_info->data = vmx->msr_ia32_feature_control;
1809 		break;
1810 	case MSR_IA32_VMX_BASIC ... MSR_IA32_VMX_VMFUNC:
1811 		if (!nested_vmx_allowed(vcpu))
1812 			return 1;
1813 		return vmx_get_vmx_msr(&vmx->nested.msrs, msr_info->index,
1814 				       &msr_info->data);
1815 	case MSR_IA32_XSS:
1816 		if (!vmx_xsaves_supported() ||
1817 		    (!msr_info->host_initiated &&
1818 		     !(guest_cpuid_has(vcpu, X86_FEATURE_XSAVE) &&
1819 		       guest_cpuid_has(vcpu, X86_FEATURE_XSAVES))))
1820 			return 1;
1821 		msr_info->data = vcpu->arch.ia32_xss;
1822 		break;
1823 	case MSR_IA32_RTIT_CTL:
1824 		if (pt_mode != PT_MODE_HOST_GUEST)
1825 			return 1;
1826 		msr_info->data = vmx->pt_desc.guest.ctl;
1827 		break;
1828 	case MSR_IA32_RTIT_STATUS:
1829 		if (pt_mode != PT_MODE_HOST_GUEST)
1830 			return 1;
1831 		msr_info->data = vmx->pt_desc.guest.status;
1832 		break;
1833 	case MSR_IA32_RTIT_CR3_MATCH:
1834 		if ((pt_mode != PT_MODE_HOST_GUEST) ||
1835 			!intel_pt_validate_cap(vmx->pt_desc.caps,
1836 						PT_CAP_cr3_filtering))
1837 			return 1;
1838 		msr_info->data = vmx->pt_desc.guest.cr3_match;
1839 		break;
1840 	case MSR_IA32_RTIT_OUTPUT_BASE:
1841 		if ((pt_mode != PT_MODE_HOST_GUEST) ||
1842 			(!intel_pt_validate_cap(vmx->pt_desc.caps,
1843 					PT_CAP_topa_output) &&
1844 			 !intel_pt_validate_cap(vmx->pt_desc.caps,
1845 					PT_CAP_single_range_output)))
1846 			return 1;
1847 		msr_info->data = vmx->pt_desc.guest.output_base;
1848 		break;
1849 	case MSR_IA32_RTIT_OUTPUT_MASK:
1850 		if ((pt_mode != PT_MODE_HOST_GUEST) ||
1851 			(!intel_pt_validate_cap(vmx->pt_desc.caps,
1852 					PT_CAP_topa_output) &&
1853 			 !intel_pt_validate_cap(vmx->pt_desc.caps,
1854 					PT_CAP_single_range_output)))
1855 			return 1;
1856 		msr_info->data = vmx->pt_desc.guest.output_mask;
1857 		break;
1858 	case MSR_IA32_RTIT_ADDR0_A ... MSR_IA32_RTIT_ADDR3_B:
1859 		index = msr_info->index - MSR_IA32_RTIT_ADDR0_A;
1860 		if ((pt_mode != PT_MODE_HOST_GUEST) ||
1861 			(index >= 2 * intel_pt_validate_cap(vmx->pt_desc.caps,
1862 					PT_CAP_num_address_ranges)))
1863 			return 1;
1864 		if (index % 2)
1865 			msr_info->data = vmx->pt_desc.guest.addr_b[index / 2];
1866 		else
1867 			msr_info->data = vmx->pt_desc.guest.addr_a[index / 2];
1868 		break;
1869 	case MSR_TSC_AUX:
1870 		if (!msr_info->host_initiated &&
1871 		    !guest_cpuid_has(vcpu, X86_FEATURE_RDTSCP))
1872 			return 1;
1873 		/* Else, falls through */
1874 	default:
1875 		msr = find_msr_entry(vmx, msr_info->index);
1876 		if (msr) {
1877 			msr_info->data = msr->data;
1878 			break;
1879 		}
1880 		return kvm_get_msr_common(vcpu, msr_info);
1881 	}
1882 
1883 	return 0;
1884 }
1885 
1886 /*
1887  * Writes msr value into into the appropriate "register".
1888  * Returns 0 on success, non-0 otherwise.
1889  * Assumes vcpu_load() was already called.
1890  */
1891 static int vmx_set_msr(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
1892 {
1893 	struct vcpu_vmx *vmx = to_vmx(vcpu);
1894 	struct shared_msr_entry *msr;
1895 	int ret = 0;
1896 	u32 msr_index = msr_info->index;
1897 	u64 data = msr_info->data;
1898 	u32 index;
1899 
1900 	switch (msr_index) {
1901 	case MSR_EFER:
1902 		ret = kvm_set_msr_common(vcpu, msr_info);
1903 		break;
1904 #ifdef CONFIG_X86_64
1905 	case MSR_FS_BASE:
1906 		vmx_segment_cache_clear(vmx);
1907 		vmcs_writel(GUEST_FS_BASE, data);
1908 		break;
1909 	case MSR_GS_BASE:
1910 		vmx_segment_cache_clear(vmx);
1911 		vmcs_writel(GUEST_GS_BASE, data);
1912 		break;
1913 	case MSR_KERNEL_GS_BASE:
1914 		vmx_write_guest_kernel_gs_base(vmx, data);
1915 		break;
1916 #endif
1917 	case MSR_IA32_SYSENTER_CS:
1918 		if (is_guest_mode(vcpu))
1919 			get_vmcs12(vcpu)->guest_sysenter_cs = data;
1920 		vmcs_write32(GUEST_SYSENTER_CS, data);
1921 		break;
1922 	case MSR_IA32_SYSENTER_EIP:
1923 		if (is_guest_mode(vcpu))
1924 			get_vmcs12(vcpu)->guest_sysenter_eip = data;
1925 		vmcs_writel(GUEST_SYSENTER_EIP, data);
1926 		break;
1927 	case MSR_IA32_SYSENTER_ESP:
1928 		if (is_guest_mode(vcpu))
1929 			get_vmcs12(vcpu)->guest_sysenter_esp = data;
1930 		vmcs_writel(GUEST_SYSENTER_ESP, data);
1931 		break;
1932 	case MSR_IA32_DEBUGCTLMSR:
1933 		if (is_guest_mode(vcpu) && get_vmcs12(vcpu)->vm_exit_controls &
1934 						VM_EXIT_SAVE_DEBUG_CONTROLS)
1935 			get_vmcs12(vcpu)->guest_ia32_debugctl = data;
1936 
1937 		ret = kvm_set_msr_common(vcpu, msr_info);
1938 		break;
1939 
1940 	case MSR_IA32_BNDCFGS:
1941 		if (!kvm_mpx_supported() ||
1942 		    (!msr_info->host_initiated &&
1943 		     !guest_cpuid_has(vcpu, X86_FEATURE_MPX)))
1944 			return 1;
1945 		if (is_noncanonical_address(data & PAGE_MASK, vcpu) ||
1946 		    (data & MSR_IA32_BNDCFGS_RSVD))
1947 			return 1;
1948 		vmcs_write64(GUEST_BNDCFGS, data);
1949 		break;
1950 	case MSR_IA32_UMWAIT_CONTROL:
1951 		if (!msr_info->host_initiated && !vmx_has_waitpkg(vmx))
1952 			return 1;
1953 
1954 		/* The reserved bit 1 and non-32 bit [63:32] should be zero */
1955 		if (data & (BIT_ULL(1) | GENMASK_ULL(63, 32)))
1956 			return 1;
1957 
1958 		vmx->msr_ia32_umwait_control = data;
1959 		break;
1960 	case MSR_IA32_SPEC_CTRL:
1961 		if (!msr_info->host_initiated &&
1962 		    !guest_cpuid_has(vcpu, X86_FEATURE_SPEC_CTRL))
1963 			return 1;
1964 
1965 		/* The STIBP bit doesn't fault even if it's not advertised */
1966 		if (data & ~(SPEC_CTRL_IBRS | SPEC_CTRL_STIBP | SPEC_CTRL_SSBD))
1967 			return 1;
1968 
1969 		vmx->spec_ctrl = data;
1970 
1971 		if (!data)
1972 			break;
1973 
1974 		/*
1975 		 * For non-nested:
1976 		 * When it's written (to non-zero) for the first time, pass
1977 		 * it through.
1978 		 *
1979 		 * For nested:
1980 		 * The handling of the MSR bitmap for L2 guests is done in
1981 		 * nested_vmx_merge_msr_bitmap. We should not touch the
1982 		 * vmcs02.msr_bitmap here since it gets completely overwritten
1983 		 * in the merging. We update the vmcs01 here for L1 as well
1984 		 * since it will end up touching the MSR anyway now.
1985 		 */
1986 		vmx_disable_intercept_for_msr(vmx->vmcs01.msr_bitmap,
1987 					      MSR_IA32_SPEC_CTRL,
1988 					      MSR_TYPE_RW);
1989 		break;
1990 	case MSR_IA32_PRED_CMD:
1991 		if (!msr_info->host_initiated &&
1992 		    !guest_cpuid_has(vcpu, X86_FEATURE_SPEC_CTRL))
1993 			return 1;
1994 
1995 		if (data & ~PRED_CMD_IBPB)
1996 			return 1;
1997 
1998 		if (!data)
1999 			break;
2000 
2001 		wrmsrl(MSR_IA32_PRED_CMD, PRED_CMD_IBPB);
2002 
2003 		/*
2004 		 * For non-nested:
2005 		 * When it's written (to non-zero) for the first time, pass
2006 		 * it through.
2007 		 *
2008 		 * For nested:
2009 		 * The handling of the MSR bitmap for L2 guests is done in
2010 		 * nested_vmx_merge_msr_bitmap. We should not touch the
2011 		 * vmcs02.msr_bitmap here since it gets completely overwritten
2012 		 * in the merging.
2013 		 */
2014 		vmx_disable_intercept_for_msr(vmx->vmcs01.msr_bitmap, MSR_IA32_PRED_CMD,
2015 					      MSR_TYPE_W);
2016 		break;
2017 	case MSR_IA32_CR_PAT:
2018 		if (!kvm_pat_valid(data))
2019 			return 1;
2020 
2021 		if (is_guest_mode(vcpu) &&
2022 		    get_vmcs12(vcpu)->vm_exit_controls & VM_EXIT_SAVE_IA32_PAT)
2023 			get_vmcs12(vcpu)->guest_ia32_pat = data;
2024 
2025 		if (vmcs_config.vmentry_ctrl & VM_ENTRY_LOAD_IA32_PAT) {
2026 			vmcs_write64(GUEST_IA32_PAT, data);
2027 			vcpu->arch.pat = data;
2028 			break;
2029 		}
2030 		ret = kvm_set_msr_common(vcpu, msr_info);
2031 		break;
2032 	case MSR_IA32_TSC_ADJUST:
2033 		ret = kvm_set_msr_common(vcpu, msr_info);
2034 		break;
2035 	case MSR_IA32_MCG_EXT_CTL:
2036 		if ((!msr_info->host_initiated &&
2037 		     !(to_vmx(vcpu)->msr_ia32_feature_control &
2038 		       FEATURE_CONTROL_LMCE)) ||
2039 		    (data & ~MCG_EXT_CTL_LMCE_EN))
2040 			return 1;
2041 		vcpu->arch.mcg_ext_ctl = data;
2042 		break;
2043 	case MSR_IA32_FEATURE_CONTROL:
2044 		if (!vmx_feature_control_msr_valid(vcpu, data) ||
2045 		    (to_vmx(vcpu)->msr_ia32_feature_control &
2046 		     FEATURE_CONTROL_LOCKED && !msr_info->host_initiated))
2047 			return 1;
2048 		vmx->msr_ia32_feature_control = data;
2049 		if (msr_info->host_initiated && data == 0)
2050 			vmx_leave_nested(vcpu);
2051 		break;
2052 	case MSR_IA32_VMX_BASIC ... MSR_IA32_VMX_VMFUNC:
2053 		if (!msr_info->host_initiated)
2054 			return 1; /* they are read-only */
2055 		if (!nested_vmx_allowed(vcpu))
2056 			return 1;
2057 		return vmx_set_vmx_msr(vcpu, msr_index, data);
2058 	case MSR_IA32_XSS:
2059 		if (!vmx_xsaves_supported() ||
2060 		    (!msr_info->host_initiated &&
2061 		     !(guest_cpuid_has(vcpu, X86_FEATURE_XSAVE) &&
2062 		       guest_cpuid_has(vcpu, X86_FEATURE_XSAVES))))
2063 			return 1;
2064 		/*
2065 		 * The only supported bit as of Skylake is bit 8, but
2066 		 * it is not supported on KVM.
2067 		 */
2068 		if (data != 0)
2069 			return 1;
2070 		vcpu->arch.ia32_xss = data;
2071 		if (vcpu->arch.ia32_xss != host_xss)
2072 			add_atomic_switch_msr(vmx, MSR_IA32_XSS,
2073 				vcpu->arch.ia32_xss, host_xss, false);
2074 		else
2075 			clear_atomic_switch_msr(vmx, MSR_IA32_XSS);
2076 		break;
2077 	case MSR_IA32_RTIT_CTL:
2078 		if ((pt_mode != PT_MODE_HOST_GUEST) ||
2079 			vmx_rtit_ctl_check(vcpu, data) ||
2080 			vmx->nested.vmxon)
2081 			return 1;
2082 		vmcs_write64(GUEST_IA32_RTIT_CTL, data);
2083 		vmx->pt_desc.guest.ctl = data;
2084 		pt_update_intercept_for_msr(vmx);
2085 		break;
2086 	case MSR_IA32_RTIT_STATUS:
2087 		if ((pt_mode != PT_MODE_HOST_GUEST) ||
2088 			(vmx->pt_desc.guest.ctl & RTIT_CTL_TRACEEN) ||
2089 			(data & MSR_IA32_RTIT_STATUS_MASK))
2090 			return 1;
2091 		vmx->pt_desc.guest.status = data;
2092 		break;
2093 	case MSR_IA32_RTIT_CR3_MATCH:
2094 		if ((pt_mode != PT_MODE_HOST_GUEST) ||
2095 			(vmx->pt_desc.guest.ctl & RTIT_CTL_TRACEEN) ||
2096 			!intel_pt_validate_cap(vmx->pt_desc.caps,
2097 						PT_CAP_cr3_filtering))
2098 			return 1;
2099 		vmx->pt_desc.guest.cr3_match = data;
2100 		break;
2101 	case MSR_IA32_RTIT_OUTPUT_BASE:
2102 		if ((pt_mode != PT_MODE_HOST_GUEST) ||
2103 			(vmx->pt_desc.guest.ctl & RTIT_CTL_TRACEEN) ||
2104 			(!intel_pt_validate_cap(vmx->pt_desc.caps,
2105 					PT_CAP_topa_output) &&
2106 			 !intel_pt_validate_cap(vmx->pt_desc.caps,
2107 					PT_CAP_single_range_output)) ||
2108 			(data & MSR_IA32_RTIT_OUTPUT_BASE_MASK))
2109 			return 1;
2110 		vmx->pt_desc.guest.output_base = data;
2111 		break;
2112 	case MSR_IA32_RTIT_OUTPUT_MASK:
2113 		if ((pt_mode != PT_MODE_HOST_GUEST) ||
2114 			(vmx->pt_desc.guest.ctl & RTIT_CTL_TRACEEN) ||
2115 			(!intel_pt_validate_cap(vmx->pt_desc.caps,
2116 					PT_CAP_topa_output) &&
2117 			 !intel_pt_validate_cap(vmx->pt_desc.caps,
2118 					PT_CAP_single_range_output)))
2119 			return 1;
2120 		vmx->pt_desc.guest.output_mask = data;
2121 		break;
2122 	case MSR_IA32_RTIT_ADDR0_A ... MSR_IA32_RTIT_ADDR3_B:
2123 		index = msr_info->index - MSR_IA32_RTIT_ADDR0_A;
2124 		if ((pt_mode != PT_MODE_HOST_GUEST) ||
2125 			(vmx->pt_desc.guest.ctl & RTIT_CTL_TRACEEN) ||
2126 			(index >= 2 * intel_pt_validate_cap(vmx->pt_desc.caps,
2127 					PT_CAP_num_address_ranges)))
2128 			return 1;
2129 		if (index % 2)
2130 			vmx->pt_desc.guest.addr_b[index / 2] = data;
2131 		else
2132 			vmx->pt_desc.guest.addr_a[index / 2] = data;
2133 		break;
2134 	case MSR_TSC_AUX:
2135 		if (!msr_info->host_initiated &&
2136 		    !guest_cpuid_has(vcpu, X86_FEATURE_RDTSCP))
2137 			return 1;
2138 		/* Check reserved bit, higher 32 bits should be zero */
2139 		if ((data >> 32) != 0)
2140 			return 1;
2141 		/* Else, falls through */
2142 	default:
2143 		msr = find_msr_entry(vmx, msr_index);
2144 		if (msr) {
2145 			u64 old_msr_data = msr->data;
2146 			msr->data = data;
2147 			if (msr - vmx->guest_msrs < vmx->save_nmsrs) {
2148 				preempt_disable();
2149 				ret = kvm_set_shared_msr(msr->index, msr->data,
2150 							 msr->mask);
2151 				preempt_enable();
2152 				if (ret)
2153 					msr->data = old_msr_data;
2154 			}
2155 			break;
2156 		}
2157 		ret = kvm_set_msr_common(vcpu, msr_info);
2158 	}
2159 
2160 	return ret;
2161 }
2162 
2163 static void vmx_cache_reg(struct kvm_vcpu *vcpu, enum kvm_reg reg)
2164 {
2165 	__set_bit(reg, (unsigned long *)&vcpu->arch.regs_avail);
2166 	switch (reg) {
2167 	case VCPU_REGS_RSP:
2168 		vcpu->arch.regs[VCPU_REGS_RSP] = vmcs_readl(GUEST_RSP);
2169 		break;
2170 	case VCPU_REGS_RIP:
2171 		vcpu->arch.regs[VCPU_REGS_RIP] = vmcs_readl(GUEST_RIP);
2172 		break;
2173 	case VCPU_EXREG_PDPTR:
2174 		if (enable_ept)
2175 			ept_save_pdptrs(vcpu);
2176 		break;
2177 	default:
2178 		break;
2179 	}
2180 }
2181 
2182 static __init int cpu_has_kvm_support(void)
2183 {
2184 	return cpu_has_vmx();
2185 }
2186 
2187 static __init int vmx_disabled_by_bios(void)
2188 {
2189 	u64 msr;
2190 
2191 	rdmsrl(MSR_IA32_FEATURE_CONTROL, msr);
2192 	if (msr & FEATURE_CONTROL_LOCKED) {
2193 		/* launched w/ TXT and VMX disabled */
2194 		if (!(msr & FEATURE_CONTROL_VMXON_ENABLED_INSIDE_SMX)
2195 			&& tboot_enabled())
2196 			return 1;
2197 		/* launched w/o TXT and VMX only enabled w/ TXT */
2198 		if (!(msr & FEATURE_CONTROL_VMXON_ENABLED_OUTSIDE_SMX)
2199 			&& (msr & FEATURE_CONTROL_VMXON_ENABLED_INSIDE_SMX)
2200 			&& !tboot_enabled()) {
2201 			printk(KERN_WARNING "kvm: disable TXT in the BIOS or "
2202 				"activate TXT before enabling KVM\n");
2203 			return 1;
2204 		}
2205 		/* launched w/o TXT and VMX disabled */
2206 		if (!(msr & FEATURE_CONTROL_VMXON_ENABLED_OUTSIDE_SMX)
2207 			&& !tboot_enabled())
2208 			return 1;
2209 	}
2210 
2211 	return 0;
2212 }
2213 
2214 static void kvm_cpu_vmxon(u64 addr)
2215 {
2216 	cr4_set_bits(X86_CR4_VMXE);
2217 	intel_pt_handle_vmx(1);
2218 
2219 	asm volatile ("vmxon %0" : : "m"(addr));
2220 }
2221 
2222 static int hardware_enable(void)
2223 {
2224 	int cpu = raw_smp_processor_id();
2225 	u64 phys_addr = __pa(per_cpu(vmxarea, cpu));
2226 	u64 old, test_bits;
2227 
2228 	if (cr4_read_shadow() & X86_CR4_VMXE)
2229 		return -EBUSY;
2230 
2231 	/*
2232 	 * This can happen if we hot-added a CPU but failed to allocate
2233 	 * VP assist page for it.
2234 	 */
2235 	if (static_branch_unlikely(&enable_evmcs) &&
2236 	    !hv_get_vp_assist_page(cpu))
2237 		return -EFAULT;
2238 
2239 	INIT_LIST_HEAD(&per_cpu(loaded_vmcss_on_cpu, cpu));
2240 	INIT_LIST_HEAD(&per_cpu(blocked_vcpu_on_cpu, cpu));
2241 	spin_lock_init(&per_cpu(blocked_vcpu_on_cpu_lock, cpu));
2242 
2243 	/*
2244 	 * Now we can enable the vmclear operation in kdump
2245 	 * since the loaded_vmcss_on_cpu list on this cpu
2246 	 * has been initialized.
2247 	 *
2248 	 * Though the cpu is not in VMX operation now, there
2249 	 * is no problem to enable the vmclear operation
2250 	 * for the loaded_vmcss_on_cpu list is empty!
2251 	 */
2252 	crash_enable_local_vmclear(cpu);
2253 
2254 	rdmsrl(MSR_IA32_FEATURE_CONTROL, old);
2255 
2256 	test_bits = FEATURE_CONTROL_LOCKED;
2257 	test_bits |= FEATURE_CONTROL_VMXON_ENABLED_OUTSIDE_SMX;
2258 	if (tboot_enabled())
2259 		test_bits |= FEATURE_CONTROL_VMXON_ENABLED_INSIDE_SMX;
2260 
2261 	if ((old & test_bits) != test_bits) {
2262 		/* enable and lock */
2263 		wrmsrl(MSR_IA32_FEATURE_CONTROL, old | test_bits);
2264 	}
2265 	kvm_cpu_vmxon(phys_addr);
2266 	if (enable_ept)
2267 		ept_sync_global();
2268 
2269 	return 0;
2270 }
2271 
2272 static void vmclear_local_loaded_vmcss(void)
2273 {
2274 	int cpu = raw_smp_processor_id();
2275 	struct loaded_vmcs *v, *n;
2276 
2277 	list_for_each_entry_safe(v, n, &per_cpu(loaded_vmcss_on_cpu, cpu),
2278 				 loaded_vmcss_on_cpu_link)
2279 		__loaded_vmcs_clear(v);
2280 }
2281 
2282 
2283 /* Just like cpu_vmxoff(), but with the __kvm_handle_fault_on_reboot()
2284  * tricks.
2285  */
2286 static void kvm_cpu_vmxoff(void)
2287 {
2288 	asm volatile (__ex("vmxoff"));
2289 
2290 	intel_pt_handle_vmx(0);
2291 	cr4_clear_bits(X86_CR4_VMXE);
2292 }
2293 
2294 static void hardware_disable(void)
2295 {
2296 	vmclear_local_loaded_vmcss();
2297 	kvm_cpu_vmxoff();
2298 }
2299 
2300 static __init int adjust_vmx_controls(u32 ctl_min, u32 ctl_opt,
2301 				      u32 msr, u32 *result)
2302 {
2303 	u32 vmx_msr_low, vmx_msr_high;
2304 	u32 ctl = ctl_min | ctl_opt;
2305 
2306 	rdmsr(msr, vmx_msr_low, vmx_msr_high);
2307 
2308 	ctl &= vmx_msr_high; /* bit == 0 in high word ==> must be zero */
2309 	ctl |= vmx_msr_low;  /* bit == 1 in low word  ==> must be one  */
2310 
2311 	/* Ensure minimum (required) set of control bits are supported. */
2312 	if (ctl_min & ~ctl)
2313 		return -EIO;
2314 
2315 	*result = ctl;
2316 	return 0;
2317 }
2318 
2319 static __init int setup_vmcs_config(struct vmcs_config *vmcs_conf,
2320 				    struct vmx_capability *vmx_cap)
2321 {
2322 	u32 vmx_msr_low, vmx_msr_high;
2323 	u32 min, opt, min2, opt2;
2324 	u32 _pin_based_exec_control = 0;
2325 	u32 _cpu_based_exec_control = 0;
2326 	u32 _cpu_based_2nd_exec_control = 0;
2327 	u32 _vmexit_control = 0;
2328 	u32 _vmentry_control = 0;
2329 
2330 	memset(vmcs_conf, 0, sizeof(*vmcs_conf));
2331 	min = CPU_BASED_HLT_EXITING |
2332 #ifdef CONFIG_X86_64
2333 	      CPU_BASED_CR8_LOAD_EXITING |
2334 	      CPU_BASED_CR8_STORE_EXITING |
2335 #endif
2336 	      CPU_BASED_CR3_LOAD_EXITING |
2337 	      CPU_BASED_CR3_STORE_EXITING |
2338 	      CPU_BASED_UNCOND_IO_EXITING |
2339 	      CPU_BASED_MOV_DR_EXITING |
2340 	      CPU_BASED_USE_TSC_OFFSETING |
2341 	      CPU_BASED_MWAIT_EXITING |
2342 	      CPU_BASED_MONITOR_EXITING |
2343 	      CPU_BASED_INVLPG_EXITING |
2344 	      CPU_BASED_RDPMC_EXITING;
2345 
2346 	opt = CPU_BASED_TPR_SHADOW |
2347 	      CPU_BASED_USE_MSR_BITMAPS |
2348 	      CPU_BASED_ACTIVATE_SECONDARY_CONTROLS;
2349 	if (adjust_vmx_controls(min, opt, MSR_IA32_VMX_PROCBASED_CTLS,
2350 				&_cpu_based_exec_control) < 0)
2351 		return -EIO;
2352 #ifdef CONFIG_X86_64
2353 	if ((_cpu_based_exec_control & CPU_BASED_TPR_SHADOW))
2354 		_cpu_based_exec_control &= ~CPU_BASED_CR8_LOAD_EXITING &
2355 					   ~CPU_BASED_CR8_STORE_EXITING;
2356 #endif
2357 	if (_cpu_based_exec_control & CPU_BASED_ACTIVATE_SECONDARY_CONTROLS) {
2358 		min2 = 0;
2359 		opt2 = SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES |
2360 			SECONDARY_EXEC_VIRTUALIZE_X2APIC_MODE |
2361 			SECONDARY_EXEC_WBINVD_EXITING |
2362 			SECONDARY_EXEC_ENABLE_VPID |
2363 			SECONDARY_EXEC_ENABLE_EPT |
2364 			SECONDARY_EXEC_UNRESTRICTED_GUEST |
2365 			SECONDARY_EXEC_PAUSE_LOOP_EXITING |
2366 			SECONDARY_EXEC_DESC |
2367 			SECONDARY_EXEC_RDTSCP |
2368 			SECONDARY_EXEC_ENABLE_INVPCID |
2369 			SECONDARY_EXEC_APIC_REGISTER_VIRT |
2370 			SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY |
2371 			SECONDARY_EXEC_SHADOW_VMCS |
2372 			SECONDARY_EXEC_XSAVES |
2373 			SECONDARY_EXEC_RDSEED_EXITING |
2374 			SECONDARY_EXEC_RDRAND_EXITING |
2375 			SECONDARY_EXEC_ENABLE_PML |
2376 			SECONDARY_EXEC_TSC_SCALING |
2377 			SECONDARY_EXEC_ENABLE_USR_WAIT_PAUSE |
2378 			SECONDARY_EXEC_PT_USE_GPA |
2379 			SECONDARY_EXEC_PT_CONCEAL_VMX |
2380 			SECONDARY_EXEC_ENABLE_VMFUNC |
2381 			SECONDARY_EXEC_ENCLS_EXITING;
2382 		if (adjust_vmx_controls(min2, opt2,
2383 					MSR_IA32_VMX_PROCBASED_CTLS2,
2384 					&_cpu_based_2nd_exec_control) < 0)
2385 			return -EIO;
2386 	}
2387 #ifndef CONFIG_X86_64
2388 	if (!(_cpu_based_2nd_exec_control &
2389 				SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES))
2390 		_cpu_based_exec_control &= ~CPU_BASED_TPR_SHADOW;
2391 #endif
2392 
2393 	if (!(_cpu_based_exec_control & CPU_BASED_TPR_SHADOW))
2394 		_cpu_based_2nd_exec_control &= ~(
2395 				SECONDARY_EXEC_APIC_REGISTER_VIRT |
2396 				SECONDARY_EXEC_VIRTUALIZE_X2APIC_MODE |
2397 				SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY);
2398 
2399 	rdmsr_safe(MSR_IA32_VMX_EPT_VPID_CAP,
2400 		&vmx_cap->ept, &vmx_cap->vpid);
2401 
2402 	if (_cpu_based_2nd_exec_control & SECONDARY_EXEC_ENABLE_EPT) {
2403 		/* CR3 accesses and invlpg don't need to cause VM Exits when EPT
2404 		   enabled */
2405 		_cpu_based_exec_control &= ~(CPU_BASED_CR3_LOAD_EXITING |
2406 					     CPU_BASED_CR3_STORE_EXITING |
2407 					     CPU_BASED_INVLPG_EXITING);
2408 	} else if (vmx_cap->ept) {
2409 		vmx_cap->ept = 0;
2410 		pr_warn_once("EPT CAP should not exist if not support "
2411 				"1-setting enable EPT VM-execution control\n");
2412 	}
2413 	if (!(_cpu_based_2nd_exec_control & SECONDARY_EXEC_ENABLE_VPID) &&
2414 		vmx_cap->vpid) {
2415 		vmx_cap->vpid = 0;
2416 		pr_warn_once("VPID CAP should not exist if not support "
2417 				"1-setting enable VPID VM-execution control\n");
2418 	}
2419 
2420 	min = VM_EXIT_SAVE_DEBUG_CONTROLS | VM_EXIT_ACK_INTR_ON_EXIT;
2421 #ifdef CONFIG_X86_64
2422 	min |= VM_EXIT_HOST_ADDR_SPACE_SIZE;
2423 #endif
2424 	opt = VM_EXIT_LOAD_IA32_PERF_GLOBAL_CTRL |
2425 	      VM_EXIT_LOAD_IA32_PAT |
2426 	      VM_EXIT_LOAD_IA32_EFER |
2427 	      VM_EXIT_CLEAR_BNDCFGS |
2428 	      VM_EXIT_PT_CONCEAL_PIP |
2429 	      VM_EXIT_CLEAR_IA32_RTIT_CTL;
2430 	if (adjust_vmx_controls(min, opt, MSR_IA32_VMX_EXIT_CTLS,
2431 				&_vmexit_control) < 0)
2432 		return -EIO;
2433 
2434 	min = PIN_BASED_EXT_INTR_MASK | PIN_BASED_NMI_EXITING;
2435 	opt = PIN_BASED_VIRTUAL_NMIS | PIN_BASED_POSTED_INTR |
2436 		 PIN_BASED_VMX_PREEMPTION_TIMER;
2437 	if (adjust_vmx_controls(min, opt, MSR_IA32_VMX_PINBASED_CTLS,
2438 				&_pin_based_exec_control) < 0)
2439 		return -EIO;
2440 
2441 	if (cpu_has_broken_vmx_preemption_timer())
2442 		_pin_based_exec_control &= ~PIN_BASED_VMX_PREEMPTION_TIMER;
2443 	if (!(_cpu_based_2nd_exec_control &
2444 		SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY))
2445 		_pin_based_exec_control &= ~PIN_BASED_POSTED_INTR;
2446 
2447 	min = VM_ENTRY_LOAD_DEBUG_CONTROLS;
2448 	opt = VM_ENTRY_LOAD_IA32_PERF_GLOBAL_CTRL |
2449 	      VM_ENTRY_LOAD_IA32_PAT |
2450 	      VM_ENTRY_LOAD_IA32_EFER |
2451 	      VM_ENTRY_LOAD_BNDCFGS |
2452 	      VM_ENTRY_PT_CONCEAL_PIP |
2453 	      VM_ENTRY_LOAD_IA32_RTIT_CTL;
2454 	if (adjust_vmx_controls(min, opt, MSR_IA32_VMX_ENTRY_CTLS,
2455 				&_vmentry_control) < 0)
2456 		return -EIO;
2457 
2458 	/*
2459 	 * Some cpus support VM_{ENTRY,EXIT}_IA32_PERF_GLOBAL_CTRL but they
2460 	 * can't be used due to an errata where VM Exit may incorrectly clear
2461 	 * IA32_PERF_GLOBAL_CTRL[34:32].  Workaround the errata by using the
2462 	 * MSR load mechanism to switch IA32_PERF_GLOBAL_CTRL.
2463 	 */
2464 	if (boot_cpu_data.x86 == 0x6) {
2465 		switch (boot_cpu_data.x86_model) {
2466 		case 26: /* AAK155 */
2467 		case 30: /* AAP115 */
2468 		case 37: /* AAT100 */
2469 		case 44: /* BC86,AAY89,BD102 */
2470 		case 46: /* BA97 */
2471 			_vmentry_control &= ~VM_ENTRY_LOAD_IA32_PERF_GLOBAL_CTRL;
2472 			_vmexit_control &= ~VM_EXIT_LOAD_IA32_PERF_GLOBAL_CTRL;
2473 			pr_warn_once("kvm: VM_EXIT_LOAD_IA32_PERF_GLOBAL_CTRL "
2474 					"does not work properly. Using workaround\n");
2475 			break;
2476 		default:
2477 			break;
2478 		}
2479 	}
2480 
2481 
2482 	rdmsr(MSR_IA32_VMX_BASIC, vmx_msr_low, vmx_msr_high);
2483 
2484 	/* IA-32 SDM Vol 3B: VMCS size is never greater than 4kB. */
2485 	if ((vmx_msr_high & 0x1fff) > PAGE_SIZE)
2486 		return -EIO;
2487 
2488 #ifdef CONFIG_X86_64
2489 	/* IA-32 SDM Vol 3B: 64-bit CPUs always have VMX_BASIC_MSR[48]==0. */
2490 	if (vmx_msr_high & (1u<<16))
2491 		return -EIO;
2492 #endif
2493 
2494 	/* Require Write-Back (WB) memory type for VMCS accesses. */
2495 	if (((vmx_msr_high >> 18) & 15) != 6)
2496 		return -EIO;
2497 
2498 	vmcs_conf->size = vmx_msr_high & 0x1fff;
2499 	vmcs_conf->order = get_order(vmcs_conf->size);
2500 	vmcs_conf->basic_cap = vmx_msr_high & ~0x1fff;
2501 
2502 	vmcs_conf->revision_id = vmx_msr_low;
2503 
2504 	vmcs_conf->pin_based_exec_ctrl = _pin_based_exec_control;
2505 	vmcs_conf->cpu_based_exec_ctrl = _cpu_based_exec_control;
2506 	vmcs_conf->cpu_based_2nd_exec_ctrl = _cpu_based_2nd_exec_control;
2507 	vmcs_conf->vmexit_ctrl         = _vmexit_control;
2508 	vmcs_conf->vmentry_ctrl        = _vmentry_control;
2509 
2510 	if (static_branch_unlikely(&enable_evmcs))
2511 		evmcs_sanitize_exec_ctrls(vmcs_conf);
2512 
2513 	return 0;
2514 }
2515 
2516 struct vmcs *alloc_vmcs_cpu(bool shadow, int cpu, gfp_t flags)
2517 {
2518 	int node = cpu_to_node(cpu);
2519 	struct page *pages;
2520 	struct vmcs *vmcs;
2521 
2522 	pages = __alloc_pages_node(node, flags, vmcs_config.order);
2523 	if (!pages)
2524 		return NULL;
2525 	vmcs = page_address(pages);
2526 	memset(vmcs, 0, vmcs_config.size);
2527 
2528 	/* KVM supports Enlightened VMCS v1 only */
2529 	if (static_branch_unlikely(&enable_evmcs))
2530 		vmcs->hdr.revision_id = KVM_EVMCS_VERSION;
2531 	else
2532 		vmcs->hdr.revision_id = vmcs_config.revision_id;
2533 
2534 	if (shadow)
2535 		vmcs->hdr.shadow_vmcs = 1;
2536 	return vmcs;
2537 }
2538 
2539 void free_vmcs(struct vmcs *vmcs)
2540 {
2541 	free_pages((unsigned long)vmcs, vmcs_config.order);
2542 }
2543 
2544 /*
2545  * Free a VMCS, but before that VMCLEAR it on the CPU where it was last loaded
2546  */
2547 void free_loaded_vmcs(struct loaded_vmcs *loaded_vmcs)
2548 {
2549 	if (!loaded_vmcs->vmcs)
2550 		return;
2551 	loaded_vmcs_clear(loaded_vmcs);
2552 	free_vmcs(loaded_vmcs->vmcs);
2553 	loaded_vmcs->vmcs = NULL;
2554 	if (loaded_vmcs->msr_bitmap)
2555 		free_page((unsigned long)loaded_vmcs->msr_bitmap);
2556 	WARN_ON(loaded_vmcs->shadow_vmcs != NULL);
2557 }
2558 
2559 int alloc_loaded_vmcs(struct loaded_vmcs *loaded_vmcs)
2560 {
2561 	loaded_vmcs->vmcs = alloc_vmcs(false);
2562 	if (!loaded_vmcs->vmcs)
2563 		return -ENOMEM;
2564 
2565 	loaded_vmcs->shadow_vmcs = NULL;
2566 	loaded_vmcs->hv_timer_soft_disabled = false;
2567 	loaded_vmcs_init(loaded_vmcs);
2568 
2569 	if (cpu_has_vmx_msr_bitmap()) {
2570 		loaded_vmcs->msr_bitmap = (unsigned long *)
2571 				__get_free_page(GFP_KERNEL_ACCOUNT);
2572 		if (!loaded_vmcs->msr_bitmap)
2573 			goto out_vmcs;
2574 		memset(loaded_vmcs->msr_bitmap, 0xff, PAGE_SIZE);
2575 
2576 		if (IS_ENABLED(CONFIG_HYPERV) &&
2577 		    static_branch_unlikely(&enable_evmcs) &&
2578 		    (ms_hyperv.nested_features & HV_X64_NESTED_MSR_BITMAP)) {
2579 			struct hv_enlightened_vmcs *evmcs =
2580 				(struct hv_enlightened_vmcs *)loaded_vmcs->vmcs;
2581 
2582 			evmcs->hv_enlightenments_control.msr_bitmap = 1;
2583 		}
2584 	}
2585 
2586 	memset(&loaded_vmcs->host_state, 0, sizeof(struct vmcs_host_state));
2587 	memset(&loaded_vmcs->controls_shadow, 0,
2588 		sizeof(struct vmcs_controls_shadow));
2589 
2590 	return 0;
2591 
2592 out_vmcs:
2593 	free_loaded_vmcs(loaded_vmcs);
2594 	return -ENOMEM;
2595 }
2596 
2597 static void free_kvm_area(void)
2598 {
2599 	int cpu;
2600 
2601 	for_each_possible_cpu(cpu) {
2602 		free_vmcs(per_cpu(vmxarea, cpu));
2603 		per_cpu(vmxarea, cpu) = NULL;
2604 	}
2605 }
2606 
2607 static __init int alloc_kvm_area(void)
2608 {
2609 	int cpu;
2610 
2611 	for_each_possible_cpu(cpu) {
2612 		struct vmcs *vmcs;
2613 
2614 		vmcs = alloc_vmcs_cpu(false, cpu, GFP_KERNEL);
2615 		if (!vmcs) {
2616 			free_kvm_area();
2617 			return -ENOMEM;
2618 		}
2619 
2620 		/*
2621 		 * When eVMCS is enabled, alloc_vmcs_cpu() sets
2622 		 * vmcs->revision_id to KVM_EVMCS_VERSION instead of
2623 		 * revision_id reported by MSR_IA32_VMX_BASIC.
2624 		 *
2625 		 * However, even though not explicitly documented by
2626 		 * TLFS, VMXArea passed as VMXON argument should
2627 		 * still be marked with revision_id reported by
2628 		 * physical CPU.
2629 		 */
2630 		if (static_branch_unlikely(&enable_evmcs))
2631 			vmcs->hdr.revision_id = vmcs_config.revision_id;
2632 
2633 		per_cpu(vmxarea, cpu) = vmcs;
2634 	}
2635 	return 0;
2636 }
2637 
2638 static void fix_pmode_seg(struct kvm_vcpu *vcpu, int seg,
2639 		struct kvm_segment *save)
2640 {
2641 	if (!emulate_invalid_guest_state) {
2642 		/*
2643 		 * CS and SS RPL should be equal during guest entry according
2644 		 * to VMX spec, but in reality it is not always so. Since vcpu
2645 		 * is in the middle of the transition from real mode to
2646 		 * protected mode it is safe to assume that RPL 0 is a good
2647 		 * default value.
2648 		 */
2649 		if (seg == VCPU_SREG_CS || seg == VCPU_SREG_SS)
2650 			save->selector &= ~SEGMENT_RPL_MASK;
2651 		save->dpl = save->selector & SEGMENT_RPL_MASK;
2652 		save->s = 1;
2653 	}
2654 	vmx_set_segment(vcpu, save, seg);
2655 }
2656 
2657 static void enter_pmode(struct kvm_vcpu *vcpu)
2658 {
2659 	unsigned long flags;
2660 	struct vcpu_vmx *vmx = to_vmx(vcpu);
2661 
2662 	/*
2663 	 * Update real mode segment cache. It may be not up-to-date if sement
2664 	 * register was written while vcpu was in a guest mode.
2665 	 */
2666 	vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_ES], VCPU_SREG_ES);
2667 	vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_DS], VCPU_SREG_DS);
2668 	vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_FS], VCPU_SREG_FS);
2669 	vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_GS], VCPU_SREG_GS);
2670 	vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_SS], VCPU_SREG_SS);
2671 	vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_CS], VCPU_SREG_CS);
2672 
2673 	vmx->rmode.vm86_active = 0;
2674 
2675 	vmx_segment_cache_clear(vmx);
2676 
2677 	vmx_set_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_TR], VCPU_SREG_TR);
2678 
2679 	flags = vmcs_readl(GUEST_RFLAGS);
2680 	flags &= RMODE_GUEST_OWNED_EFLAGS_BITS;
2681 	flags |= vmx->rmode.save_rflags & ~RMODE_GUEST_OWNED_EFLAGS_BITS;
2682 	vmcs_writel(GUEST_RFLAGS, flags);
2683 
2684 	vmcs_writel(GUEST_CR4, (vmcs_readl(GUEST_CR4) & ~X86_CR4_VME) |
2685 			(vmcs_readl(CR4_READ_SHADOW) & X86_CR4_VME));
2686 
2687 	update_exception_bitmap(vcpu);
2688 
2689 	fix_pmode_seg(vcpu, VCPU_SREG_CS, &vmx->rmode.segs[VCPU_SREG_CS]);
2690 	fix_pmode_seg(vcpu, VCPU_SREG_SS, &vmx->rmode.segs[VCPU_SREG_SS]);
2691 	fix_pmode_seg(vcpu, VCPU_SREG_ES, &vmx->rmode.segs[VCPU_SREG_ES]);
2692 	fix_pmode_seg(vcpu, VCPU_SREG_DS, &vmx->rmode.segs[VCPU_SREG_DS]);
2693 	fix_pmode_seg(vcpu, VCPU_SREG_FS, &vmx->rmode.segs[VCPU_SREG_FS]);
2694 	fix_pmode_seg(vcpu, VCPU_SREG_GS, &vmx->rmode.segs[VCPU_SREG_GS]);
2695 }
2696 
2697 static void fix_rmode_seg(int seg, struct kvm_segment *save)
2698 {
2699 	const struct kvm_vmx_segment_field *sf = &kvm_vmx_segment_fields[seg];
2700 	struct kvm_segment var = *save;
2701 
2702 	var.dpl = 0x3;
2703 	if (seg == VCPU_SREG_CS)
2704 		var.type = 0x3;
2705 
2706 	if (!emulate_invalid_guest_state) {
2707 		var.selector = var.base >> 4;
2708 		var.base = var.base & 0xffff0;
2709 		var.limit = 0xffff;
2710 		var.g = 0;
2711 		var.db = 0;
2712 		var.present = 1;
2713 		var.s = 1;
2714 		var.l = 0;
2715 		var.unusable = 0;
2716 		var.type = 0x3;
2717 		var.avl = 0;
2718 		if (save->base & 0xf)
2719 			printk_once(KERN_WARNING "kvm: segment base is not "
2720 					"paragraph aligned when entering "
2721 					"protected mode (seg=%d)", seg);
2722 	}
2723 
2724 	vmcs_write16(sf->selector, var.selector);
2725 	vmcs_writel(sf->base, var.base);
2726 	vmcs_write32(sf->limit, var.limit);
2727 	vmcs_write32(sf->ar_bytes, vmx_segment_access_rights(&var));
2728 }
2729 
2730 static void enter_rmode(struct kvm_vcpu *vcpu)
2731 {
2732 	unsigned long flags;
2733 	struct vcpu_vmx *vmx = to_vmx(vcpu);
2734 	struct kvm_vmx *kvm_vmx = to_kvm_vmx(vcpu->kvm);
2735 
2736 	vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_TR], VCPU_SREG_TR);
2737 	vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_ES], VCPU_SREG_ES);
2738 	vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_DS], VCPU_SREG_DS);
2739 	vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_FS], VCPU_SREG_FS);
2740 	vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_GS], VCPU_SREG_GS);
2741 	vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_SS], VCPU_SREG_SS);
2742 	vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_CS], VCPU_SREG_CS);
2743 
2744 	vmx->rmode.vm86_active = 1;
2745 
2746 	/*
2747 	 * Very old userspace does not call KVM_SET_TSS_ADDR before entering
2748 	 * vcpu. Warn the user that an update is overdue.
2749 	 */
2750 	if (!kvm_vmx->tss_addr)
2751 		printk_once(KERN_WARNING "kvm: KVM_SET_TSS_ADDR need to be "
2752 			     "called before entering vcpu\n");
2753 
2754 	vmx_segment_cache_clear(vmx);
2755 
2756 	vmcs_writel(GUEST_TR_BASE, kvm_vmx->tss_addr);
2757 	vmcs_write32(GUEST_TR_LIMIT, RMODE_TSS_SIZE - 1);
2758 	vmcs_write32(GUEST_TR_AR_BYTES, 0x008b);
2759 
2760 	flags = vmcs_readl(GUEST_RFLAGS);
2761 	vmx->rmode.save_rflags = flags;
2762 
2763 	flags |= X86_EFLAGS_IOPL | X86_EFLAGS_VM;
2764 
2765 	vmcs_writel(GUEST_RFLAGS, flags);
2766 	vmcs_writel(GUEST_CR4, vmcs_readl(GUEST_CR4) | X86_CR4_VME);
2767 	update_exception_bitmap(vcpu);
2768 
2769 	fix_rmode_seg(VCPU_SREG_SS, &vmx->rmode.segs[VCPU_SREG_SS]);
2770 	fix_rmode_seg(VCPU_SREG_CS, &vmx->rmode.segs[VCPU_SREG_CS]);
2771 	fix_rmode_seg(VCPU_SREG_ES, &vmx->rmode.segs[VCPU_SREG_ES]);
2772 	fix_rmode_seg(VCPU_SREG_DS, &vmx->rmode.segs[VCPU_SREG_DS]);
2773 	fix_rmode_seg(VCPU_SREG_GS, &vmx->rmode.segs[VCPU_SREG_GS]);
2774 	fix_rmode_seg(VCPU_SREG_FS, &vmx->rmode.segs[VCPU_SREG_FS]);
2775 
2776 	kvm_mmu_reset_context(vcpu);
2777 }
2778 
2779 void vmx_set_efer(struct kvm_vcpu *vcpu, u64 efer)
2780 {
2781 	struct vcpu_vmx *vmx = to_vmx(vcpu);
2782 	struct shared_msr_entry *msr = find_msr_entry(vmx, MSR_EFER);
2783 
2784 	if (!msr)
2785 		return;
2786 
2787 	vcpu->arch.efer = efer;
2788 	if (efer & EFER_LMA) {
2789 		vm_entry_controls_setbit(to_vmx(vcpu), VM_ENTRY_IA32E_MODE);
2790 		msr->data = efer;
2791 	} else {
2792 		vm_entry_controls_clearbit(to_vmx(vcpu), VM_ENTRY_IA32E_MODE);
2793 
2794 		msr->data = efer & ~EFER_LME;
2795 	}
2796 	setup_msrs(vmx);
2797 }
2798 
2799 #ifdef CONFIG_X86_64
2800 
2801 static void enter_lmode(struct kvm_vcpu *vcpu)
2802 {
2803 	u32 guest_tr_ar;
2804 
2805 	vmx_segment_cache_clear(to_vmx(vcpu));
2806 
2807 	guest_tr_ar = vmcs_read32(GUEST_TR_AR_BYTES);
2808 	if ((guest_tr_ar & VMX_AR_TYPE_MASK) != VMX_AR_TYPE_BUSY_64_TSS) {
2809 		pr_debug_ratelimited("%s: tss fixup for long mode. \n",
2810 				     __func__);
2811 		vmcs_write32(GUEST_TR_AR_BYTES,
2812 			     (guest_tr_ar & ~VMX_AR_TYPE_MASK)
2813 			     | VMX_AR_TYPE_BUSY_64_TSS);
2814 	}
2815 	vmx_set_efer(vcpu, vcpu->arch.efer | EFER_LMA);
2816 }
2817 
2818 static void exit_lmode(struct kvm_vcpu *vcpu)
2819 {
2820 	vm_entry_controls_clearbit(to_vmx(vcpu), VM_ENTRY_IA32E_MODE);
2821 	vmx_set_efer(vcpu, vcpu->arch.efer & ~EFER_LMA);
2822 }
2823 
2824 #endif
2825 
2826 static void vmx_flush_tlb_gva(struct kvm_vcpu *vcpu, gva_t addr)
2827 {
2828 	int vpid = to_vmx(vcpu)->vpid;
2829 
2830 	if (!vpid_sync_vcpu_addr(vpid, addr))
2831 		vpid_sync_context(vpid);
2832 
2833 	/*
2834 	 * If VPIDs are not supported or enabled, then the above is a no-op.
2835 	 * But we don't really need a TLB flush in that case anyway, because
2836 	 * each VM entry/exit includes an implicit flush when VPID is 0.
2837 	 */
2838 }
2839 
2840 static void vmx_decache_cr0_guest_bits(struct kvm_vcpu *vcpu)
2841 {
2842 	ulong cr0_guest_owned_bits = vcpu->arch.cr0_guest_owned_bits;
2843 
2844 	vcpu->arch.cr0 &= ~cr0_guest_owned_bits;
2845 	vcpu->arch.cr0 |= vmcs_readl(GUEST_CR0) & cr0_guest_owned_bits;
2846 }
2847 
2848 static void vmx_decache_cr3(struct kvm_vcpu *vcpu)
2849 {
2850 	if (enable_unrestricted_guest || (enable_ept && is_paging(vcpu)))
2851 		vcpu->arch.cr3 = vmcs_readl(GUEST_CR3);
2852 	__set_bit(VCPU_EXREG_CR3, (ulong *)&vcpu->arch.regs_avail);
2853 }
2854 
2855 static void vmx_decache_cr4_guest_bits(struct kvm_vcpu *vcpu)
2856 {
2857 	ulong cr4_guest_owned_bits = vcpu->arch.cr4_guest_owned_bits;
2858 
2859 	vcpu->arch.cr4 &= ~cr4_guest_owned_bits;
2860 	vcpu->arch.cr4 |= vmcs_readl(GUEST_CR4) & cr4_guest_owned_bits;
2861 }
2862 
2863 static void ept_load_pdptrs(struct kvm_vcpu *vcpu)
2864 {
2865 	struct kvm_mmu *mmu = vcpu->arch.walk_mmu;
2866 
2867 	if (!test_bit(VCPU_EXREG_PDPTR,
2868 		      (unsigned long *)&vcpu->arch.regs_dirty))
2869 		return;
2870 
2871 	if (is_pae_paging(vcpu)) {
2872 		vmcs_write64(GUEST_PDPTR0, mmu->pdptrs[0]);
2873 		vmcs_write64(GUEST_PDPTR1, mmu->pdptrs[1]);
2874 		vmcs_write64(GUEST_PDPTR2, mmu->pdptrs[2]);
2875 		vmcs_write64(GUEST_PDPTR3, mmu->pdptrs[3]);
2876 	}
2877 }
2878 
2879 void ept_save_pdptrs(struct kvm_vcpu *vcpu)
2880 {
2881 	struct kvm_mmu *mmu = vcpu->arch.walk_mmu;
2882 
2883 	if (is_pae_paging(vcpu)) {
2884 		mmu->pdptrs[0] = vmcs_read64(GUEST_PDPTR0);
2885 		mmu->pdptrs[1] = vmcs_read64(GUEST_PDPTR1);
2886 		mmu->pdptrs[2] = vmcs_read64(GUEST_PDPTR2);
2887 		mmu->pdptrs[3] = vmcs_read64(GUEST_PDPTR3);
2888 	}
2889 
2890 	__set_bit(VCPU_EXREG_PDPTR,
2891 		  (unsigned long *)&vcpu->arch.regs_avail);
2892 	__set_bit(VCPU_EXREG_PDPTR,
2893 		  (unsigned long *)&vcpu->arch.regs_dirty);
2894 }
2895 
2896 static void ept_update_paging_mode_cr0(unsigned long *hw_cr0,
2897 					unsigned long cr0,
2898 					struct kvm_vcpu *vcpu)
2899 {
2900 	struct vcpu_vmx *vmx = to_vmx(vcpu);
2901 
2902 	if (!test_bit(VCPU_EXREG_CR3, (ulong *)&vcpu->arch.regs_avail))
2903 		vmx_decache_cr3(vcpu);
2904 	if (!(cr0 & X86_CR0_PG)) {
2905 		/* From paging/starting to nonpaging */
2906 		exec_controls_setbit(vmx, CPU_BASED_CR3_LOAD_EXITING |
2907 					  CPU_BASED_CR3_STORE_EXITING);
2908 		vcpu->arch.cr0 = cr0;
2909 		vmx_set_cr4(vcpu, kvm_read_cr4(vcpu));
2910 	} else if (!is_paging(vcpu)) {
2911 		/* From nonpaging to paging */
2912 		exec_controls_clearbit(vmx, CPU_BASED_CR3_LOAD_EXITING |
2913 					    CPU_BASED_CR3_STORE_EXITING);
2914 		vcpu->arch.cr0 = cr0;
2915 		vmx_set_cr4(vcpu, kvm_read_cr4(vcpu));
2916 	}
2917 
2918 	if (!(cr0 & X86_CR0_WP))
2919 		*hw_cr0 &= ~X86_CR0_WP;
2920 }
2921 
2922 void vmx_set_cr0(struct kvm_vcpu *vcpu, unsigned long cr0)
2923 {
2924 	struct vcpu_vmx *vmx = to_vmx(vcpu);
2925 	unsigned long hw_cr0;
2926 
2927 	hw_cr0 = (cr0 & ~KVM_VM_CR0_ALWAYS_OFF);
2928 	if (enable_unrestricted_guest)
2929 		hw_cr0 |= KVM_VM_CR0_ALWAYS_ON_UNRESTRICTED_GUEST;
2930 	else {
2931 		hw_cr0 |= KVM_VM_CR0_ALWAYS_ON;
2932 
2933 		if (vmx->rmode.vm86_active && (cr0 & X86_CR0_PE))
2934 			enter_pmode(vcpu);
2935 
2936 		if (!vmx->rmode.vm86_active && !(cr0 & X86_CR0_PE))
2937 			enter_rmode(vcpu);
2938 	}
2939 
2940 #ifdef CONFIG_X86_64
2941 	if (vcpu->arch.efer & EFER_LME) {
2942 		if (!is_paging(vcpu) && (cr0 & X86_CR0_PG))
2943 			enter_lmode(vcpu);
2944 		if (is_paging(vcpu) && !(cr0 & X86_CR0_PG))
2945 			exit_lmode(vcpu);
2946 	}
2947 #endif
2948 
2949 	if (enable_ept && !enable_unrestricted_guest)
2950 		ept_update_paging_mode_cr0(&hw_cr0, cr0, vcpu);
2951 
2952 	vmcs_writel(CR0_READ_SHADOW, cr0);
2953 	vmcs_writel(GUEST_CR0, hw_cr0);
2954 	vcpu->arch.cr0 = cr0;
2955 
2956 	/* depends on vcpu->arch.cr0 to be set to a new value */
2957 	vmx->emulation_required = emulation_required(vcpu);
2958 }
2959 
2960 static int get_ept_level(struct kvm_vcpu *vcpu)
2961 {
2962 	if (cpu_has_vmx_ept_5levels() && (cpuid_maxphyaddr(vcpu) > 48))
2963 		return 5;
2964 	return 4;
2965 }
2966 
2967 u64 construct_eptp(struct kvm_vcpu *vcpu, unsigned long root_hpa)
2968 {
2969 	u64 eptp = VMX_EPTP_MT_WB;
2970 
2971 	eptp |= (get_ept_level(vcpu) == 5) ? VMX_EPTP_PWL_5 : VMX_EPTP_PWL_4;
2972 
2973 	if (enable_ept_ad_bits &&
2974 	    (!is_guest_mode(vcpu) || nested_ept_ad_enabled(vcpu)))
2975 		eptp |= VMX_EPTP_AD_ENABLE_BIT;
2976 	eptp |= (root_hpa & PAGE_MASK);
2977 
2978 	return eptp;
2979 }
2980 
2981 void vmx_set_cr3(struct kvm_vcpu *vcpu, unsigned long cr3)
2982 {
2983 	struct kvm *kvm = vcpu->kvm;
2984 	unsigned long guest_cr3;
2985 	u64 eptp;
2986 
2987 	guest_cr3 = cr3;
2988 	if (enable_ept) {
2989 		eptp = construct_eptp(vcpu, cr3);
2990 		vmcs_write64(EPT_POINTER, eptp);
2991 
2992 		if (kvm_x86_ops->tlb_remote_flush) {
2993 			spin_lock(&to_kvm_vmx(kvm)->ept_pointer_lock);
2994 			to_vmx(vcpu)->ept_pointer = eptp;
2995 			to_kvm_vmx(kvm)->ept_pointers_match
2996 				= EPT_POINTERS_CHECK;
2997 			spin_unlock(&to_kvm_vmx(kvm)->ept_pointer_lock);
2998 		}
2999 
3000 		if (enable_unrestricted_guest || is_paging(vcpu) ||
3001 		    is_guest_mode(vcpu))
3002 			guest_cr3 = kvm_read_cr3(vcpu);
3003 		else
3004 			guest_cr3 = to_kvm_vmx(kvm)->ept_identity_map_addr;
3005 		ept_load_pdptrs(vcpu);
3006 	}
3007 
3008 	vmcs_writel(GUEST_CR3, guest_cr3);
3009 }
3010 
3011 int vmx_set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4)
3012 {
3013 	struct vcpu_vmx *vmx = to_vmx(vcpu);
3014 	/*
3015 	 * Pass through host's Machine Check Enable value to hw_cr4, which
3016 	 * is in force while we are in guest mode.  Do not let guests control
3017 	 * this bit, even if host CR4.MCE == 0.
3018 	 */
3019 	unsigned long hw_cr4;
3020 
3021 	hw_cr4 = (cr4_read_shadow() & X86_CR4_MCE) | (cr4 & ~X86_CR4_MCE);
3022 	if (enable_unrestricted_guest)
3023 		hw_cr4 |= KVM_VM_CR4_ALWAYS_ON_UNRESTRICTED_GUEST;
3024 	else if (vmx->rmode.vm86_active)
3025 		hw_cr4 |= KVM_RMODE_VM_CR4_ALWAYS_ON;
3026 	else
3027 		hw_cr4 |= KVM_PMODE_VM_CR4_ALWAYS_ON;
3028 
3029 	if (!boot_cpu_has(X86_FEATURE_UMIP) && vmx_umip_emulated()) {
3030 		if (cr4 & X86_CR4_UMIP) {
3031 			secondary_exec_controls_setbit(vmx, SECONDARY_EXEC_DESC);
3032 			hw_cr4 &= ~X86_CR4_UMIP;
3033 		} else if (!is_guest_mode(vcpu) ||
3034 			!nested_cpu_has2(get_vmcs12(vcpu), SECONDARY_EXEC_DESC)) {
3035 			secondary_exec_controls_clearbit(vmx, SECONDARY_EXEC_DESC);
3036 		}
3037 	}
3038 
3039 	if (cr4 & X86_CR4_VMXE) {
3040 		/*
3041 		 * To use VMXON (and later other VMX instructions), a guest
3042 		 * must first be able to turn on cr4.VMXE (see handle_vmon()).
3043 		 * So basically the check on whether to allow nested VMX
3044 		 * is here.  We operate under the default treatment of SMM,
3045 		 * so VMX cannot be enabled under SMM.
3046 		 */
3047 		if (!nested_vmx_allowed(vcpu) || is_smm(vcpu))
3048 			return 1;
3049 	}
3050 
3051 	if (vmx->nested.vmxon && !nested_cr4_valid(vcpu, cr4))
3052 		return 1;
3053 
3054 	vcpu->arch.cr4 = cr4;
3055 
3056 	if (!enable_unrestricted_guest) {
3057 		if (enable_ept) {
3058 			if (!is_paging(vcpu)) {
3059 				hw_cr4 &= ~X86_CR4_PAE;
3060 				hw_cr4 |= X86_CR4_PSE;
3061 			} else if (!(cr4 & X86_CR4_PAE)) {
3062 				hw_cr4 &= ~X86_CR4_PAE;
3063 			}
3064 		}
3065 
3066 		/*
3067 		 * SMEP/SMAP/PKU is disabled if CPU is in non-paging mode in
3068 		 * hardware.  To emulate this behavior, SMEP/SMAP/PKU needs
3069 		 * to be manually disabled when guest switches to non-paging
3070 		 * mode.
3071 		 *
3072 		 * If !enable_unrestricted_guest, the CPU is always running
3073 		 * with CR0.PG=1 and CR4 needs to be modified.
3074 		 * If enable_unrestricted_guest, the CPU automatically
3075 		 * disables SMEP/SMAP/PKU when the guest sets CR0.PG=0.
3076 		 */
3077 		if (!is_paging(vcpu))
3078 			hw_cr4 &= ~(X86_CR4_SMEP | X86_CR4_SMAP | X86_CR4_PKE);
3079 	}
3080 
3081 	vmcs_writel(CR4_READ_SHADOW, cr4);
3082 	vmcs_writel(GUEST_CR4, hw_cr4);
3083 	return 0;
3084 }
3085 
3086 void vmx_get_segment(struct kvm_vcpu *vcpu, struct kvm_segment *var, int seg)
3087 {
3088 	struct vcpu_vmx *vmx = to_vmx(vcpu);
3089 	u32 ar;
3090 
3091 	if (vmx->rmode.vm86_active && seg != VCPU_SREG_LDTR) {
3092 		*var = vmx->rmode.segs[seg];
3093 		if (seg == VCPU_SREG_TR
3094 		    || var->selector == vmx_read_guest_seg_selector(vmx, seg))
3095 			return;
3096 		var->base = vmx_read_guest_seg_base(vmx, seg);
3097 		var->selector = vmx_read_guest_seg_selector(vmx, seg);
3098 		return;
3099 	}
3100 	var->base = vmx_read_guest_seg_base(vmx, seg);
3101 	var->limit = vmx_read_guest_seg_limit(vmx, seg);
3102 	var->selector = vmx_read_guest_seg_selector(vmx, seg);
3103 	ar = vmx_read_guest_seg_ar(vmx, seg);
3104 	var->unusable = (ar >> 16) & 1;
3105 	var->type = ar & 15;
3106 	var->s = (ar >> 4) & 1;
3107 	var->dpl = (ar >> 5) & 3;
3108 	/*
3109 	 * Some userspaces do not preserve unusable property. Since usable
3110 	 * segment has to be present according to VMX spec we can use present
3111 	 * property to amend userspace bug by making unusable segment always
3112 	 * nonpresent. vmx_segment_access_rights() already marks nonpresent
3113 	 * segment as unusable.
3114 	 */
3115 	var->present = !var->unusable;
3116 	var->avl = (ar >> 12) & 1;
3117 	var->l = (ar >> 13) & 1;
3118 	var->db = (ar >> 14) & 1;
3119 	var->g = (ar >> 15) & 1;
3120 }
3121 
3122 static u64 vmx_get_segment_base(struct kvm_vcpu *vcpu, int seg)
3123 {
3124 	struct kvm_segment s;
3125 
3126 	if (to_vmx(vcpu)->rmode.vm86_active) {
3127 		vmx_get_segment(vcpu, &s, seg);
3128 		return s.base;
3129 	}
3130 	return vmx_read_guest_seg_base(to_vmx(vcpu), seg);
3131 }
3132 
3133 int vmx_get_cpl(struct kvm_vcpu *vcpu)
3134 {
3135 	struct vcpu_vmx *vmx = to_vmx(vcpu);
3136 
3137 	if (unlikely(vmx->rmode.vm86_active))
3138 		return 0;
3139 	else {
3140 		int ar = vmx_read_guest_seg_ar(vmx, VCPU_SREG_SS);
3141 		return VMX_AR_DPL(ar);
3142 	}
3143 }
3144 
3145 static u32 vmx_segment_access_rights(struct kvm_segment *var)
3146 {
3147 	u32 ar;
3148 
3149 	if (var->unusable || !var->present)
3150 		ar = 1 << 16;
3151 	else {
3152 		ar = var->type & 15;
3153 		ar |= (var->s & 1) << 4;
3154 		ar |= (var->dpl & 3) << 5;
3155 		ar |= (var->present & 1) << 7;
3156 		ar |= (var->avl & 1) << 12;
3157 		ar |= (var->l & 1) << 13;
3158 		ar |= (var->db & 1) << 14;
3159 		ar |= (var->g & 1) << 15;
3160 	}
3161 
3162 	return ar;
3163 }
3164 
3165 void vmx_set_segment(struct kvm_vcpu *vcpu, struct kvm_segment *var, int seg)
3166 {
3167 	struct vcpu_vmx *vmx = to_vmx(vcpu);
3168 	const struct kvm_vmx_segment_field *sf = &kvm_vmx_segment_fields[seg];
3169 
3170 	vmx_segment_cache_clear(vmx);
3171 
3172 	if (vmx->rmode.vm86_active && seg != VCPU_SREG_LDTR) {
3173 		vmx->rmode.segs[seg] = *var;
3174 		if (seg == VCPU_SREG_TR)
3175 			vmcs_write16(sf->selector, var->selector);
3176 		else if (var->s)
3177 			fix_rmode_seg(seg, &vmx->rmode.segs[seg]);
3178 		goto out;
3179 	}
3180 
3181 	vmcs_writel(sf->base, var->base);
3182 	vmcs_write32(sf->limit, var->limit);
3183 	vmcs_write16(sf->selector, var->selector);
3184 
3185 	/*
3186 	 *   Fix the "Accessed" bit in AR field of segment registers for older
3187 	 * qemu binaries.
3188 	 *   IA32 arch specifies that at the time of processor reset the
3189 	 * "Accessed" bit in the AR field of segment registers is 1. And qemu
3190 	 * is setting it to 0 in the userland code. This causes invalid guest
3191 	 * state vmexit when "unrestricted guest" mode is turned on.
3192 	 *    Fix for this setup issue in cpu_reset is being pushed in the qemu
3193 	 * tree. Newer qemu binaries with that qemu fix would not need this
3194 	 * kvm hack.
3195 	 */
3196 	if (enable_unrestricted_guest && (seg != VCPU_SREG_LDTR))
3197 		var->type |= 0x1; /* Accessed */
3198 
3199 	vmcs_write32(sf->ar_bytes, vmx_segment_access_rights(var));
3200 
3201 out:
3202 	vmx->emulation_required = emulation_required(vcpu);
3203 }
3204 
3205 static void vmx_get_cs_db_l_bits(struct kvm_vcpu *vcpu, int *db, int *l)
3206 {
3207 	u32 ar = vmx_read_guest_seg_ar(to_vmx(vcpu), VCPU_SREG_CS);
3208 
3209 	*db = (ar >> 14) & 1;
3210 	*l = (ar >> 13) & 1;
3211 }
3212 
3213 static void vmx_get_idt(struct kvm_vcpu *vcpu, struct desc_ptr *dt)
3214 {
3215 	dt->size = vmcs_read32(GUEST_IDTR_LIMIT);
3216 	dt->address = vmcs_readl(GUEST_IDTR_BASE);
3217 }
3218 
3219 static void vmx_set_idt(struct kvm_vcpu *vcpu, struct desc_ptr *dt)
3220 {
3221 	vmcs_write32(GUEST_IDTR_LIMIT, dt->size);
3222 	vmcs_writel(GUEST_IDTR_BASE, dt->address);
3223 }
3224 
3225 static void vmx_get_gdt(struct kvm_vcpu *vcpu, struct desc_ptr *dt)
3226 {
3227 	dt->size = vmcs_read32(GUEST_GDTR_LIMIT);
3228 	dt->address = vmcs_readl(GUEST_GDTR_BASE);
3229 }
3230 
3231 static void vmx_set_gdt(struct kvm_vcpu *vcpu, struct desc_ptr *dt)
3232 {
3233 	vmcs_write32(GUEST_GDTR_LIMIT, dt->size);
3234 	vmcs_writel(GUEST_GDTR_BASE, dt->address);
3235 }
3236 
3237 static bool rmode_segment_valid(struct kvm_vcpu *vcpu, int seg)
3238 {
3239 	struct kvm_segment var;
3240 	u32 ar;
3241 
3242 	vmx_get_segment(vcpu, &var, seg);
3243 	var.dpl = 0x3;
3244 	if (seg == VCPU_SREG_CS)
3245 		var.type = 0x3;
3246 	ar = vmx_segment_access_rights(&var);
3247 
3248 	if (var.base != (var.selector << 4))
3249 		return false;
3250 	if (var.limit != 0xffff)
3251 		return false;
3252 	if (ar != 0xf3)
3253 		return false;
3254 
3255 	return true;
3256 }
3257 
3258 static bool code_segment_valid(struct kvm_vcpu *vcpu)
3259 {
3260 	struct kvm_segment cs;
3261 	unsigned int cs_rpl;
3262 
3263 	vmx_get_segment(vcpu, &cs, VCPU_SREG_CS);
3264 	cs_rpl = cs.selector & SEGMENT_RPL_MASK;
3265 
3266 	if (cs.unusable)
3267 		return false;
3268 	if (~cs.type & (VMX_AR_TYPE_CODE_MASK|VMX_AR_TYPE_ACCESSES_MASK))
3269 		return false;
3270 	if (!cs.s)
3271 		return false;
3272 	if (cs.type & VMX_AR_TYPE_WRITEABLE_MASK) {
3273 		if (cs.dpl > cs_rpl)
3274 			return false;
3275 	} else {
3276 		if (cs.dpl != cs_rpl)
3277 			return false;
3278 	}
3279 	if (!cs.present)
3280 		return false;
3281 
3282 	/* TODO: Add Reserved field check, this'll require a new member in the kvm_segment_field structure */
3283 	return true;
3284 }
3285 
3286 static bool stack_segment_valid(struct kvm_vcpu *vcpu)
3287 {
3288 	struct kvm_segment ss;
3289 	unsigned int ss_rpl;
3290 
3291 	vmx_get_segment(vcpu, &ss, VCPU_SREG_SS);
3292 	ss_rpl = ss.selector & SEGMENT_RPL_MASK;
3293 
3294 	if (ss.unusable)
3295 		return true;
3296 	if (ss.type != 3 && ss.type != 7)
3297 		return false;
3298 	if (!ss.s)
3299 		return false;
3300 	if (ss.dpl != ss_rpl) /* DPL != RPL */
3301 		return false;
3302 	if (!ss.present)
3303 		return false;
3304 
3305 	return true;
3306 }
3307 
3308 static bool data_segment_valid(struct kvm_vcpu *vcpu, int seg)
3309 {
3310 	struct kvm_segment var;
3311 	unsigned int rpl;
3312 
3313 	vmx_get_segment(vcpu, &var, seg);
3314 	rpl = var.selector & SEGMENT_RPL_MASK;
3315 
3316 	if (var.unusable)
3317 		return true;
3318 	if (!var.s)
3319 		return false;
3320 	if (!var.present)
3321 		return false;
3322 	if (~var.type & (VMX_AR_TYPE_CODE_MASK|VMX_AR_TYPE_WRITEABLE_MASK)) {
3323 		if (var.dpl < rpl) /* DPL < RPL */
3324 			return false;
3325 	}
3326 
3327 	/* TODO: Add other members to kvm_segment_field to allow checking for other access
3328 	 * rights flags
3329 	 */
3330 	return true;
3331 }
3332 
3333 static bool tr_valid(struct kvm_vcpu *vcpu)
3334 {
3335 	struct kvm_segment tr;
3336 
3337 	vmx_get_segment(vcpu, &tr, VCPU_SREG_TR);
3338 
3339 	if (tr.unusable)
3340 		return false;
3341 	if (tr.selector & SEGMENT_TI_MASK)	/* TI = 1 */
3342 		return false;
3343 	if (tr.type != 3 && tr.type != 11) /* TODO: Check if guest is in IA32e mode */
3344 		return false;
3345 	if (!tr.present)
3346 		return false;
3347 
3348 	return true;
3349 }
3350 
3351 static bool ldtr_valid(struct kvm_vcpu *vcpu)
3352 {
3353 	struct kvm_segment ldtr;
3354 
3355 	vmx_get_segment(vcpu, &ldtr, VCPU_SREG_LDTR);
3356 
3357 	if (ldtr.unusable)
3358 		return true;
3359 	if (ldtr.selector & SEGMENT_TI_MASK)	/* TI = 1 */
3360 		return false;
3361 	if (ldtr.type != 2)
3362 		return false;
3363 	if (!ldtr.present)
3364 		return false;
3365 
3366 	return true;
3367 }
3368 
3369 static bool cs_ss_rpl_check(struct kvm_vcpu *vcpu)
3370 {
3371 	struct kvm_segment cs, ss;
3372 
3373 	vmx_get_segment(vcpu, &cs, VCPU_SREG_CS);
3374 	vmx_get_segment(vcpu, &ss, VCPU_SREG_SS);
3375 
3376 	return ((cs.selector & SEGMENT_RPL_MASK) ==
3377 		 (ss.selector & SEGMENT_RPL_MASK));
3378 }
3379 
3380 /*
3381  * Check if guest state is valid. Returns true if valid, false if
3382  * not.
3383  * We assume that registers are always usable
3384  */
3385 static bool guest_state_valid(struct kvm_vcpu *vcpu)
3386 {
3387 	if (enable_unrestricted_guest)
3388 		return true;
3389 
3390 	/* real mode guest state checks */
3391 	if (!is_protmode(vcpu) || (vmx_get_rflags(vcpu) & X86_EFLAGS_VM)) {
3392 		if (!rmode_segment_valid(vcpu, VCPU_SREG_CS))
3393 			return false;
3394 		if (!rmode_segment_valid(vcpu, VCPU_SREG_SS))
3395 			return false;
3396 		if (!rmode_segment_valid(vcpu, VCPU_SREG_DS))
3397 			return false;
3398 		if (!rmode_segment_valid(vcpu, VCPU_SREG_ES))
3399 			return false;
3400 		if (!rmode_segment_valid(vcpu, VCPU_SREG_FS))
3401 			return false;
3402 		if (!rmode_segment_valid(vcpu, VCPU_SREG_GS))
3403 			return false;
3404 	} else {
3405 	/* protected mode guest state checks */
3406 		if (!cs_ss_rpl_check(vcpu))
3407 			return false;
3408 		if (!code_segment_valid(vcpu))
3409 			return false;
3410 		if (!stack_segment_valid(vcpu))
3411 			return false;
3412 		if (!data_segment_valid(vcpu, VCPU_SREG_DS))
3413 			return false;
3414 		if (!data_segment_valid(vcpu, VCPU_SREG_ES))
3415 			return false;
3416 		if (!data_segment_valid(vcpu, VCPU_SREG_FS))
3417 			return false;
3418 		if (!data_segment_valid(vcpu, VCPU_SREG_GS))
3419 			return false;
3420 		if (!tr_valid(vcpu))
3421 			return false;
3422 		if (!ldtr_valid(vcpu))
3423 			return false;
3424 	}
3425 	/* TODO:
3426 	 * - Add checks on RIP
3427 	 * - Add checks on RFLAGS
3428 	 */
3429 
3430 	return true;
3431 }
3432 
3433 static int init_rmode_tss(struct kvm *kvm)
3434 {
3435 	gfn_t fn;
3436 	u16 data = 0;
3437 	int idx, r;
3438 
3439 	idx = srcu_read_lock(&kvm->srcu);
3440 	fn = to_kvm_vmx(kvm)->tss_addr >> PAGE_SHIFT;
3441 	r = kvm_clear_guest_page(kvm, fn, 0, PAGE_SIZE);
3442 	if (r < 0)
3443 		goto out;
3444 	data = TSS_BASE_SIZE + TSS_REDIRECTION_SIZE;
3445 	r = kvm_write_guest_page(kvm, fn++, &data,
3446 			TSS_IOPB_BASE_OFFSET, sizeof(u16));
3447 	if (r < 0)
3448 		goto out;
3449 	r = kvm_clear_guest_page(kvm, fn++, 0, PAGE_SIZE);
3450 	if (r < 0)
3451 		goto out;
3452 	r = kvm_clear_guest_page(kvm, fn, 0, PAGE_SIZE);
3453 	if (r < 0)
3454 		goto out;
3455 	data = ~0;
3456 	r = kvm_write_guest_page(kvm, fn, &data,
3457 				 RMODE_TSS_SIZE - 2 * PAGE_SIZE - 1,
3458 				 sizeof(u8));
3459 out:
3460 	srcu_read_unlock(&kvm->srcu, idx);
3461 	return r;
3462 }
3463 
3464 static int init_rmode_identity_map(struct kvm *kvm)
3465 {
3466 	struct kvm_vmx *kvm_vmx = to_kvm_vmx(kvm);
3467 	int i, idx, r = 0;
3468 	kvm_pfn_t identity_map_pfn;
3469 	u32 tmp;
3470 
3471 	/* Protect kvm_vmx->ept_identity_pagetable_done. */
3472 	mutex_lock(&kvm->slots_lock);
3473 
3474 	if (likely(kvm_vmx->ept_identity_pagetable_done))
3475 		goto out2;
3476 
3477 	if (!kvm_vmx->ept_identity_map_addr)
3478 		kvm_vmx->ept_identity_map_addr = VMX_EPT_IDENTITY_PAGETABLE_ADDR;
3479 	identity_map_pfn = kvm_vmx->ept_identity_map_addr >> PAGE_SHIFT;
3480 
3481 	r = __x86_set_memory_region(kvm, IDENTITY_PAGETABLE_PRIVATE_MEMSLOT,
3482 				    kvm_vmx->ept_identity_map_addr, PAGE_SIZE);
3483 	if (r < 0)
3484 		goto out2;
3485 
3486 	idx = srcu_read_lock(&kvm->srcu);
3487 	r = kvm_clear_guest_page(kvm, identity_map_pfn, 0, PAGE_SIZE);
3488 	if (r < 0)
3489 		goto out;
3490 	/* Set up identity-mapping pagetable for EPT in real mode */
3491 	for (i = 0; i < PT32_ENT_PER_PAGE; i++) {
3492 		tmp = (i << 22) + (_PAGE_PRESENT | _PAGE_RW | _PAGE_USER |
3493 			_PAGE_ACCESSED | _PAGE_DIRTY | _PAGE_PSE);
3494 		r = kvm_write_guest_page(kvm, identity_map_pfn,
3495 				&tmp, i * sizeof(tmp), sizeof(tmp));
3496 		if (r < 0)
3497 			goto out;
3498 	}
3499 	kvm_vmx->ept_identity_pagetable_done = true;
3500 
3501 out:
3502 	srcu_read_unlock(&kvm->srcu, idx);
3503 
3504 out2:
3505 	mutex_unlock(&kvm->slots_lock);
3506 	return r;
3507 }
3508 
3509 static void seg_setup(int seg)
3510 {
3511 	const struct kvm_vmx_segment_field *sf = &kvm_vmx_segment_fields[seg];
3512 	unsigned int ar;
3513 
3514 	vmcs_write16(sf->selector, 0);
3515 	vmcs_writel(sf->base, 0);
3516 	vmcs_write32(sf->limit, 0xffff);
3517 	ar = 0x93;
3518 	if (seg == VCPU_SREG_CS)
3519 		ar |= 0x08; /* code segment */
3520 
3521 	vmcs_write32(sf->ar_bytes, ar);
3522 }
3523 
3524 static int alloc_apic_access_page(struct kvm *kvm)
3525 {
3526 	struct page *page;
3527 	int r = 0;
3528 
3529 	mutex_lock(&kvm->slots_lock);
3530 	if (kvm->arch.apic_access_page_done)
3531 		goto out;
3532 	r = __x86_set_memory_region(kvm, APIC_ACCESS_PAGE_PRIVATE_MEMSLOT,
3533 				    APIC_DEFAULT_PHYS_BASE, PAGE_SIZE);
3534 	if (r)
3535 		goto out;
3536 
3537 	page = gfn_to_page(kvm, APIC_DEFAULT_PHYS_BASE >> PAGE_SHIFT);
3538 	if (is_error_page(page)) {
3539 		r = -EFAULT;
3540 		goto out;
3541 	}
3542 
3543 	/*
3544 	 * Do not pin the page in memory, so that memory hot-unplug
3545 	 * is able to migrate it.
3546 	 */
3547 	put_page(page);
3548 	kvm->arch.apic_access_page_done = true;
3549 out:
3550 	mutex_unlock(&kvm->slots_lock);
3551 	return r;
3552 }
3553 
3554 int allocate_vpid(void)
3555 {
3556 	int vpid;
3557 
3558 	if (!enable_vpid)
3559 		return 0;
3560 	spin_lock(&vmx_vpid_lock);
3561 	vpid = find_first_zero_bit(vmx_vpid_bitmap, VMX_NR_VPIDS);
3562 	if (vpid < VMX_NR_VPIDS)
3563 		__set_bit(vpid, vmx_vpid_bitmap);
3564 	else
3565 		vpid = 0;
3566 	spin_unlock(&vmx_vpid_lock);
3567 	return vpid;
3568 }
3569 
3570 void free_vpid(int vpid)
3571 {
3572 	if (!enable_vpid || vpid == 0)
3573 		return;
3574 	spin_lock(&vmx_vpid_lock);
3575 	__clear_bit(vpid, vmx_vpid_bitmap);
3576 	spin_unlock(&vmx_vpid_lock);
3577 }
3578 
3579 static __always_inline void vmx_disable_intercept_for_msr(unsigned long *msr_bitmap,
3580 							  u32 msr, int type)
3581 {
3582 	int f = sizeof(unsigned long);
3583 
3584 	if (!cpu_has_vmx_msr_bitmap())
3585 		return;
3586 
3587 	if (static_branch_unlikely(&enable_evmcs))
3588 		evmcs_touch_msr_bitmap();
3589 
3590 	/*
3591 	 * See Intel PRM Vol. 3, 20.6.9 (MSR-Bitmap Address). Early manuals
3592 	 * have the write-low and read-high bitmap offsets the wrong way round.
3593 	 * We can control MSRs 0x00000000-0x00001fff and 0xc0000000-0xc0001fff.
3594 	 */
3595 	if (msr <= 0x1fff) {
3596 		if (type & MSR_TYPE_R)
3597 			/* read-low */
3598 			__clear_bit(msr, msr_bitmap + 0x000 / f);
3599 
3600 		if (type & MSR_TYPE_W)
3601 			/* write-low */
3602 			__clear_bit(msr, msr_bitmap + 0x800 / f);
3603 
3604 	} else if ((msr >= 0xc0000000) && (msr <= 0xc0001fff)) {
3605 		msr &= 0x1fff;
3606 		if (type & MSR_TYPE_R)
3607 			/* read-high */
3608 			__clear_bit(msr, msr_bitmap + 0x400 / f);
3609 
3610 		if (type & MSR_TYPE_W)
3611 			/* write-high */
3612 			__clear_bit(msr, msr_bitmap + 0xc00 / f);
3613 
3614 	}
3615 }
3616 
3617 static __always_inline void vmx_enable_intercept_for_msr(unsigned long *msr_bitmap,
3618 							 u32 msr, int type)
3619 {
3620 	int f = sizeof(unsigned long);
3621 
3622 	if (!cpu_has_vmx_msr_bitmap())
3623 		return;
3624 
3625 	if (static_branch_unlikely(&enable_evmcs))
3626 		evmcs_touch_msr_bitmap();
3627 
3628 	/*
3629 	 * See Intel PRM Vol. 3, 20.6.9 (MSR-Bitmap Address). Early manuals
3630 	 * have the write-low and read-high bitmap offsets the wrong way round.
3631 	 * We can control MSRs 0x00000000-0x00001fff and 0xc0000000-0xc0001fff.
3632 	 */
3633 	if (msr <= 0x1fff) {
3634 		if (type & MSR_TYPE_R)
3635 			/* read-low */
3636 			__set_bit(msr, msr_bitmap + 0x000 / f);
3637 
3638 		if (type & MSR_TYPE_W)
3639 			/* write-low */
3640 			__set_bit(msr, msr_bitmap + 0x800 / f);
3641 
3642 	} else if ((msr >= 0xc0000000) && (msr <= 0xc0001fff)) {
3643 		msr &= 0x1fff;
3644 		if (type & MSR_TYPE_R)
3645 			/* read-high */
3646 			__set_bit(msr, msr_bitmap + 0x400 / f);
3647 
3648 		if (type & MSR_TYPE_W)
3649 			/* write-high */
3650 			__set_bit(msr, msr_bitmap + 0xc00 / f);
3651 
3652 	}
3653 }
3654 
3655 static __always_inline void vmx_set_intercept_for_msr(unsigned long *msr_bitmap,
3656 			     			      u32 msr, int type, bool value)
3657 {
3658 	if (value)
3659 		vmx_enable_intercept_for_msr(msr_bitmap, msr, type);
3660 	else
3661 		vmx_disable_intercept_for_msr(msr_bitmap, msr, type);
3662 }
3663 
3664 static u8 vmx_msr_bitmap_mode(struct kvm_vcpu *vcpu)
3665 {
3666 	u8 mode = 0;
3667 
3668 	if (cpu_has_secondary_exec_ctrls() &&
3669 	    (secondary_exec_controls_get(to_vmx(vcpu)) &
3670 	     SECONDARY_EXEC_VIRTUALIZE_X2APIC_MODE)) {
3671 		mode |= MSR_BITMAP_MODE_X2APIC;
3672 		if (enable_apicv && kvm_vcpu_apicv_active(vcpu))
3673 			mode |= MSR_BITMAP_MODE_X2APIC_APICV;
3674 	}
3675 
3676 	return mode;
3677 }
3678 
3679 static void vmx_update_msr_bitmap_x2apic(unsigned long *msr_bitmap,
3680 					 u8 mode)
3681 {
3682 	int msr;
3683 
3684 	for (msr = 0x800; msr <= 0x8ff; msr += BITS_PER_LONG) {
3685 		unsigned word = msr / BITS_PER_LONG;
3686 		msr_bitmap[word] = (mode & MSR_BITMAP_MODE_X2APIC_APICV) ? 0 : ~0;
3687 		msr_bitmap[word + (0x800 / sizeof(long))] = ~0;
3688 	}
3689 
3690 	if (mode & MSR_BITMAP_MODE_X2APIC) {
3691 		/*
3692 		 * TPR reads and writes can be virtualized even if virtual interrupt
3693 		 * delivery is not in use.
3694 		 */
3695 		vmx_disable_intercept_for_msr(msr_bitmap, X2APIC_MSR(APIC_TASKPRI), MSR_TYPE_RW);
3696 		if (mode & MSR_BITMAP_MODE_X2APIC_APICV) {
3697 			vmx_enable_intercept_for_msr(msr_bitmap, X2APIC_MSR(APIC_TMCCT), MSR_TYPE_R);
3698 			vmx_disable_intercept_for_msr(msr_bitmap, X2APIC_MSR(APIC_EOI), MSR_TYPE_W);
3699 			vmx_disable_intercept_for_msr(msr_bitmap, X2APIC_MSR(APIC_SELF_IPI), MSR_TYPE_W);
3700 		}
3701 	}
3702 }
3703 
3704 void vmx_update_msr_bitmap(struct kvm_vcpu *vcpu)
3705 {
3706 	struct vcpu_vmx *vmx = to_vmx(vcpu);
3707 	unsigned long *msr_bitmap = vmx->vmcs01.msr_bitmap;
3708 	u8 mode = vmx_msr_bitmap_mode(vcpu);
3709 	u8 changed = mode ^ vmx->msr_bitmap_mode;
3710 
3711 	if (!changed)
3712 		return;
3713 
3714 	if (changed & (MSR_BITMAP_MODE_X2APIC | MSR_BITMAP_MODE_X2APIC_APICV))
3715 		vmx_update_msr_bitmap_x2apic(msr_bitmap, mode);
3716 
3717 	vmx->msr_bitmap_mode = mode;
3718 }
3719 
3720 void pt_update_intercept_for_msr(struct vcpu_vmx *vmx)
3721 {
3722 	unsigned long *msr_bitmap = vmx->vmcs01.msr_bitmap;
3723 	bool flag = !(vmx->pt_desc.guest.ctl & RTIT_CTL_TRACEEN);
3724 	u32 i;
3725 
3726 	vmx_set_intercept_for_msr(msr_bitmap, MSR_IA32_RTIT_STATUS,
3727 							MSR_TYPE_RW, flag);
3728 	vmx_set_intercept_for_msr(msr_bitmap, MSR_IA32_RTIT_OUTPUT_BASE,
3729 							MSR_TYPE_RW, flag);
3730 	vmx_set_intercept_for_msr(msr_bitmap, MSR_IA32_RTIT_OUTPUT_MASK,
3731 							MSR_TYPE_RW, flag);
3732 	vmx_set_intercept_for_msr(msr_bitmap, MSR_IA32_RTIT_CR3_MATCH,
3733 							MSR_TYPE_RW, flag);
3734 	for (i = 0; i < vmx->pt_desc.addr_range; i++) {
3735 		vmx_set_intercept_for_msr(msr_bitmap,
3736 			MSR_IA32_RTIT_ADDR0_A + i * 2, MSR_TYPE_RW, flag);
3737 		vmx_set_intercept_for_msr(msr_bitmap,
3738 			MSR_IA32_RTIT_ADDR0_B + i * 2, MSR_TYPE_RW, flag);
3739 	}
3740 }
3741 
3742 static bool vmx_get_enable_apicv(struct kvm_vcpu *vcpu)
3743 {
3744 	return enable_apicv;
3745 }
3746 
3747 static bool vmx_guest_apic_has_interrupt(struct kvm_vcpu *vcpu)
3748 {
3749 	struct vcpu_vmx *vmx = to_vmx(vcpu);
3750 	void *vapic_page;
3751 	u32 vppr;
3752 	int rvi;
3753 
3754 	if (WARN_ON_ONCE(!is_guest_mode(vcpu)) ||
3755 		!nested_cpu_has_vid(get_vmcs12(vcpu)) ||
3756 		WARN_ON_ONCE(!vmx->nested.virtual_apic_map.gfn))
3757 		return false;
3758 
3759 	rvi = vmx_get_rvi();
3760 
3761 	vapic_page = vmx->nested.virtual_apic_map.hva;
3762 	vppr = *((u32 *)(vapic_page + APIC_PROCPRI));
3763 
3764 	return ((rvi & 0xf0) > (vppr & 0xf0));
3765 }
3766 
3767 static inline bool kvm_vcpu_trigger_posted_interrupt(struct kvm_vcpu *vcpu,
3768 						     bool nested)
3769 {
3770 #ifdef CONFIG_SMP
3771 	int pi_vec = nested ? POSTED_INTR_NESTED_VECTOR : POSTED_INTR_VECTOR;
3772 
3773 	if (vcpu->mode == IN_GUEST_MODE) {
3774 		/*
3775 		 * The vector of interrupt to be delivered to vcpu had
3776 		 * been set in PIR before this function.
3777 		 *
3778 		 * Following cases will be reached in this block, and
3779 		 * we always send a notification event in all cases as
3780 		 * explained below.
3781 		 *
3782 		 * Case 1: vcpu keeps in non-root mode. Sending a
3783 		 * notification event posts the interrupt to vcpu.
3784 		 *
3785 		 * Case 2: vcpu exits to root mode and is still
3786 		 * runnable. PIR will be synced to vIRR before the
3787 		 * next vcpu entry. Sending a notification event in
3788 		 * this case has no effect, as vcpu is not in root
3789 		 * mode.
3790 		 *
3791 		 * Case 3: vcpu exits to root mode and is blocked.
3792 		 * vcpu_block() has already synced PIR to vIRR and
3793 		 * never blocks vcpu if vIRR is not cleared. Therefore,
3794 		 * a blocked vcpu here does not wait for any requested
3795 		 * interrupts in PIR, and sending a notification event
3796 		 * which has no effect is safe here.
3797 		 */
3798 
3799 		apic->send_IPI_mask(get_cpu_mask(vcpu->cpu), pi_vec);
3800 		return true;
3801 	}
3802 #endif
3803 	return false;
3804 }
3805 
3806 static int vmx_deliver_nested_posted_interrupt(struct kvm_vcpu *vcpu,
3807 						int vector)
3808 {
3809 	struct vcpu_vmx *vmx = to_vmx(vcpu);
3810 
3811 	if (is_guest_mode(vcpu) &&
3812 	    vector == vmx->nested.posted_intr_nv) {
3813 		/*
3814 		 * If a posted intr is not recognized by hardware,
3815 		 * we will accomplish it in the next vmentry.
3816 		 */
3817 		vmx->nested.pi_pending = true;
3818 		kvm_make_request(KVM_REQ_EVENT, vcpu);
3819 		/* the PIR and ON have been set by L1. */
3820 		if (!kvm_vcpu_trigger_posted_interrupt(vcpu, true))
3821 			kvm_vcpu_kick(vcpu);
3822 		return 0;
3823 	}
3824 	return -1;
3825 }
3826 /*
3827  * Send interrupt to vcpu via posted interrupt way.
3828  * 1. If target vcpu is running(non-root mode), send posted interrupt
3829  * notification to vcpu and hardware will sync PIR to vIRR atomically.
3830  * 2. If target vcpu isn't running(root mode), kick it to pick up the
3831  * interrupt from PIR in next vmentry.
3832  */
3833 static void vmx_deliver_posted_interrupt(struct kvm_vcpu *vcpu, int vector)
3834 {
3835 	struct vcpu_vmx *vmx = to_vmx(vcpu);
3836 	int r;
3837 
3838 	r = vmx_deliver_nested_posted_interrupt(vcpu, vector);
3839 	if (!r)
3840 		return;
3841 
3842 	if (pi_test_and_set_pir(vector, &vmx->pi_desc))
3843 		return;
3844 
3845 	/* If a previous notification has sent the IPI, nothing to do.  */
3846 	if (pi_test_and_set_on(&vmx->pi_desc))
3847 		return;
3848 
3849 	if (!kvm_vcpu_trigger_posted_interrupt(vcpu, false))
3850 		kvm_vcpu_kick(vcpu);
3851 }
3852 
3853 /*
3854  * Set up the vmcs's constant host-state fields, i.e., host-state fields that
3855  * will not change in the lifetime of the guest.
3856  * Note that host-state that does change is set elsewhere. E.g., host-state
3857  * that is set differently for each CPU is set in vmx_vcpu_load(), not here.
3858  */
3859 void vmx_set_constant_host_state(struct vcpu_vmx *vmx)
3860 {
3861 	u32 low32, high32;
3862 	unsigned long tmpl;
3863 	unsigned long cr0, cr3, cr4;
3864 
3865 	cr0 = read_cr0();
3866 	WARN_ON(cr0 & X86_CR0_TS);
3867 	vmcs_writel(HOST_CR0, cr0);  /* 22.2.3 */
3868 
3869 	/*
3870 	 * Save the most likely value for this task's CR3 in the VMCS.
3871 	 * We can't use __get_current_cr3_fast() because we're not atomic.
3872 	 */
3873 	cr3 = __read_cr3();
3874 	vmcs_writel(HOST_CR3, cr3);		/* 22.2.3  FIXME: shadow tables */
3875 	vmx->loaded_vmcs->host_state.cr3 = cr3;
3876 
3877 	/* Save the most likely value for this task's CR4 in the VMCS. */
3878 	cr4 = cr4_read_shadow();
3879 	vmcs_writel(HOST_CR4, cr4);			/* 22.2.3, 22.2.5 */
3880 	vmx->loaded_vmcs->host_state.cr4 = cr4;
3881 
3882 	vmcs_write16(HOST_CS_SELECTOR, __KERNEL_CS);  /* 22.2.4 */
3883 #ifdef CONFIG_X86_64
3884 	/*
3885 	 * Load null selectors, so we can avoid reloading them in
3886 	 * vmx_prepare_switch_to_host(), in case userspace uses
3887 	 * the null selectors too (the expected case).
3888 	 */
3889 	vmcs_write16(HOST_DS_SELECTOR, 0);
3890 	vmcs_write16(HOST_ES_SELECTOR, 0);
3891 #else
3892 	vmcs_write16(HOST_DS_SELECTOR, __KERNEL_DS);  /* 22.2.4 */
3893 	vmcs_write16(HOST_ES_SELECTOR, __KERNEL_DS);  /* 22.2.4 */
3894 #endif
3895 	vmcs_write16(HOST_SS_SELECTOR, __KERNEL_DS);  /* 22.2.4 */
3896 	vmcs_write16(HOST_TR_SELECTOR, GDT_ENTRY_TSS*8);  /* 22.2.4 */
3897 
3898 	vmcs_writel(HOST_IDTR_BASE, host_idt_base);   /* 22.2.4 */
3899 
3900 	vmcs_writel(HOST_RIP, (unsigned long)vmx_vmexit); /* 22.2.5 */
3901 
3902 	rdmsr(MSR_IA32_SYSENTER_CS, low32, high32);
3903 	vmcs_write32(HOST_IA32_SYSENTER_CS, low32);
3904 	rdmsrl(MSR_IA32_SYSENTER_EIP, tmpl);
3905 	vmcs_writel(HOST_IA32_SYSENTER_EIP, tmpl);   /* 22.2.3 */
3906 
3907 	if (vmcs_config.vmexit_ctrl & VM_EXIT_LOAD_IA32_PAT) {
3908 		rdmsr(MSR_IA32_CR_PAT, low32, high32);
3909 		vmcs_write64(HOST_IA32_PAT, low32 | ((u64) high32 << 32));
3910 	}
3911 
3912 	if (cpu_has_load_ia32_efer())
3913 		vmcs_write64(HOST_IA32_EFER, host_efer);
3914 }
3915 
3916 void set_cr4_guest_host_mask(struct vcpu_vmx *vmx)
3917 {
3918 	vmx->vcpu.arch.cr4_guest_owned_bits = KVM_CR4_GUEST_OWNED_BITS;
3919 	if (enable_ept)
3920 		vmx->vcpu.arch.cr4_guest_owned_bits |= X86_CR4_PGE;
3921 	if (is_guest_mode(&vmx->vcpu))
3922 		vmx->vcpu.arch.cr4_guest_owned_bits &=
3923 			~get_vmcs12(&vmx->vcpu)->cr4_guest_host_mask;
3924 	vmcs_writel(CR4_GUEST_HOST_MASK, ~vmx->vcpu.arch.cr4_guest_owned_bits);
3925 }
3926 
3927 u32 vmx_pin_based_exec_ctrl(struct vcpu_vmx *vmx)
3928 {
3929 	u32 pin_based_exec_ctrl = vmcs_config.pin_based_exec_ctrl;
3930 
3931 	if (!kvm_vcpu_apicv_active(&vmx->vcpu))
3932 		pin_based_exec_ctrl &= ~PIN_BASED_POSTED_INTR;
3933 
3934 	if (!enable_vnmi)
3935 		pin_based_exec_ctrl &= ~PIN_BASED_VIRTUAL_NMIS;
3936 
3937 	if (!enable_preemption_timer)
3938 		pin_based_exec_ctrl &= ~PIN_BASED_VMX_PREEMPTION_TIMER;
3939 
3940 	return pin_based_exec_ctrl;
3941 }
3942 
3943 static void vmx_refresh_apicv_exec_ctrl(struct kvm_vcpu *vcpu)
3944 {
3945 	struct vcpu_vmx *vmx = to_vmx(vcpu);
3946 
3947 	pin_controls_set(vmx, vmx_pin_based_exec_ctrl(vmx));
3948 	if (cpu_has_secondary_exec_ctrls()) {
3949 		if (kvm_vcpu_apicv_active(vcpu))
3950 			secondary_exec_controls_setbit(vmx,
3951 				      SECONDARY_EXEC_APIC_REGISTER_VIRT |
3952 				      SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY);
3953 		else
3954 			secondary_exec_controls_clearbit(vmx,
3955 					SECONDARY_EXEC_APIC_REGISTER_VIRT |
3956 					SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY);
3957 	}
3958 
3959 	if (cpu_has_vmx_msr_bitmap())
3960 		vmx_update_msr_bitmap(vcpu);
3961 }
3962 
3963 u32 vmx_exec_control(struct vcpu_vmx *vmx)
3964 {
3965 	u32 exec_control = vmcs_config.cpu_based_exec_ctrl;
3966 
3967 	if (vmx->vcpu.arch.switch_db_regs & KVM_DEBUGREG_WONT_EXIT)
3968 		exec_control &= ~CPU_BASED_MOV_DR_EXITING;
3969 
3970 	if (!cpu_need_tpr_shadow(&vmx->vcpu)) {
3971 		exec_control &= ~CPU_BASED_TPR_SHADOW;
3972 #ifdef CONFIG_X86_64
3973 		exec_control |= CPU_BASED_CR8_STORE_EXITING |
3974 				CPU_BASED_CR8_LOAD_EXITING;
3975 #endif
3976 	}
3977 	if (!enable_ept)
3978 		exec_control |= CPU_BASED_CR3_STORE_EXITING |
3979 				CPU_BASED_CR3_LOAD_EXITING  |
3980 				CPU_BASED_INVLPG_EXITING;
3981 	if (kvm_mwait_in_guest(vmx->vcpu.kvm))
3982 		exec_control &= ~(CPU_BASED_MWAIT_EXITING |
3983 				CPU_BASED_MONITOR_EXITING);
3984 	if (kvm_hlt_in_guest(vmx->vcpu.kvm))
3985 		exec_control &= ~CPU_BASED_HLT_EXITING;
3986 	return exec_control;
3987 }
3988 
3989 
3990 static void vmx_compute_secondary_exec_control(struct vcpu_vmx *vmx)
3991 {
3992 	struct kvm_vcpu *vcpu = &vmx->vcpu;
3993 
3994 	u32 exec_control = vmcs_config.cpu_based_2nd_exec_ctrl;
3995 
3996 	if (pt_mode == PT_MODE_SYSTEM)
3997 		exec_control &= ~(SECONDARY_EXEC_PT_USE_GPA | SECONDARY_EXEC_PT_CONCEAL_VMX);
3998 	if (!cpu_need_virtualize_apic_accesses(vcpu))
3999 		exec_control &= ~SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES;
4000 	if (vmx->vpid == 0)
4001 		exec_control &= ~SECONDARY_EXEC_ENABLE_VPID;
4002 	if (!enable_ept) {
4003 		exec_control &= ~SECONDARY_EXEC_ENABLE_EPT;
4004 		enable_unrestricted_guest = 0;
4005 	}
4006 	if (!enable_unrestricted_guest)
4007 		exec_control &= ~SECONDARY_EXEC_UNRESTRICTED_GUEST;
4008 	if (kvm_pause_in_guest(vmx->vcpu.kvm))
4009 		exec_control &= ~SECONDARY_EXEC_PAUSE_LOOP_EXITING;
4010 	if (!kvm_vcpu_apicv_active(vcpu))
4011 		exec_control &= ~(SECONDARY_EXEC_APIC_REGISTER_VIRT |
4012 				  SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY);
4013 	exec_control &= ~SECONDARY_EXEC_VIRTUALIZE_X2APIC_MODE;
4014 
4015 	/* SECONDARY_EXEC_DESC is enabled/disabled on writes to CR4.UMIP,
4016 	 * in vmx_set_cr4.  */
4017 	exec_control &= ~SECONDARY_EXEC_DESC;
4018 
4019 	/* SECONDARY_EXEC_SHADOW_VMCS is enabled when L1 executes VMPTRLD
4020 	   (handle_vmptrld).
4021 	   We can NOT enable shadow_vmcs here because we don't have yet
4022 	   a current VMCS12
4023 	*/
4024 	exec_control &= ~SECONDARY_EXEC_SHADOW_VMCS;
4025 
4026 	if (!enable_pml)
4027 		exec_control &= ~SECONDARY_EXEC_ENABLE_PML;
4028 
4029 	if (vmx_xsaves_supported()) {
4030 		/* Exposing XSAVES only when XSAVE is exposed */
4031 		bool xsaves_enabled =
4032 			guest_cpuid_has(vcpu, X86_FEATURE_XSAVE) &&
4033 			guest_cpuid_has(vcpu, X86_FEATURE_XSAVES);
4034 
4035 		if (!xsaves_enabled)
4036 			exec_control &= ~SECONDARY_EXEC_XSAVES;
4037 
4038 		if (nested) {
4039 			if (xsaves_enabled)
4040 				vmx->nested.msrs.secondary_ctls_high |=
4041 					SECONDARY_EXEC_XSAVES;
4042 			else
4043 				vmx->nested.msrs.secondary_ctls_high &=
4044 					~SECONDARY_EXEC_XSAVES;
4045 		}
4046 	}
4047 
4048 	if (vmx_rdtscp_supported()) {
4049 		bool rdtscp_enabled = guest_cpuid_has(vcpu, X86_FEATURE_RDTSCP);
4050 		if (!rdtscp_enabled)
4051 			exec_control &= ~SECONDARY_EXEC_RDTSCP;
4052 
4053 		if (nested) {
4054 			if (rdtscp_enabled)
4055 				vmx->nested.msrs.secondary_ctls_high |=
4056 					SECONDARY_EXEC_RDTSCP;
4057 			else
4058 				vmx->nested.msrs.secondary_ctls_high &=
4059 					~SECONDARY_EXEC_RDTSCP;
4060 		}
4061 	}
4062 
4063 	if (vmx_invpcid_supported()) {
4064 		/* Exposing INVPCID only when PCID is exposed */
4065 		bool invpcid_enabled =
4066 			guest_cpuid_has(vcpu, X86_FEATURE_INVPCID) &&
4067 			guest_cpuid_has(vcpu, X86_FEATURE_PCID);
4068 
4069 		if (!invpcid_enabled) {
4070 			exec_control &= ~SECONDARY_EXEC_ENABLE_INVPCID;
4071 			guest_cpuid_clear(vcpu, X86_FEATURE_INVPCID);
4072 		}
4073 
4074 		if (nested) {
4075 			if (invpcid_enabled)
4076 				vmx->nested.msrs.secondary_ctls_high |=
4077 					SECONDARY_EXEC_ENABLE_INVPCID;
4078 			else
4079 				vmx->nested.msrs.secondary_ctls_high &=
4080 					~SECONDARY_EXEC_ENABLE_INVPCID;
4081 		}
4082 	}
4083 
4084 	if (vmx_rdrand_supported()) {
4085 		bool rdrand_enabled = guest_cpuid_has(vcpu, X86_FEATURE_RDRAND);
4086 		if (rdrand_enabled)
4087 			exec_control &= ~SECONDARY_EXEC_RDRAND_EXITING;
4088 
4089 		if (nested) {
4090 			if (rdrand_enabled)
4091 				vmx->nested.msrs.secondary_ctls_high |=
4092 					SECONDARY_EXEC_RDRAND_EXITING;
4093 			else
4094 				vmx->nested.msrs.secondary_ctls_high &=
4095 					~SECONDARY_EXEC_RDRAND_EXITING;
4096 		}
4097 	}
4098 
4099 	if (vmx_rdseed_supported()) {
4100 		bool rdseed_enabled = guest_cpuid_has(vcpu, X86_FEATURE_RDSEED);
4101 		if (rdseed_enabled)
4102 			exec_control &= ~SECONDARY_EXEC_RDSEED_EXITING;
4103 
4104 		if (nested) {
4105 			if (rdseed_enabled)
4106 				vmx->nested.msrs.secondary_ctls_high |=
4107 					SECONDARY_EXEC_RDSEED_EXITING;
4108 			else
4109 				vmx->nested.msrs.secondary_ctls_high &=
4110 					~SECONDARY_EXEC_RDSEED_EXITING;
4111 		}
4112 	}
4113 
4114 	if (vmx_waitpkg_supported()) {
4115 		bool waitpkg_enabled =
4116 			guest_cpuid_has(vcpu, X86_FEATURE_WAITPKG);
4117 
4118 		if (!waitpkg_enabled)
4119 			exec_control &= ~SECONDARY_EXEC_ENABLE_USR_WAIT_PAUSE;
4120 
4121 		if (nested) {
4122 			if (waitpkg_enabled)
4123 				vmx->nested.msrs.secondary_ctls_high |=
4124 					SECONDARY_EXEC_ENABLE_USR_WAIT_PAUSE;
4125 			else
4126 				vmx->nested.msrs.secondary_ctls_high &=
4127 					~SECONDARY_EXEC_ENABLE_USR_WAIT_PAUSE;
4128 		}
4129 	}
4130 
4131 	vmx->secondary_exec_control = exec_control;
4132 }
4133 
4134 static void ept_set_mmio_spte_mask(void)
4135 {
4136 	/*
4137 	 * EPT Misconfigurations can be generated if the value of bits 2:0
4138 	 * of an EPT paging-structure entry is 110b (write/execute).
4139 	 */
4140 	kvm_mmu_set_mmio_spte_mask(VMX_EPT_RWX_MASK,
4141 				   VMX_EPT_MISCONFIG_WX_VALUE, 0);
4142 }
4143 
4144 #define VMX_XSS_EXIT_BITMAP 0
4145 
4146 /*
4147  * Sets up the vmcs for emulated real mode.
4148  */
4149 static void vmx_vcpu_setup(struct vcpu_vmx *vmx)
4150 {
4151 	int i;
4152 
4153 	if (nested)
4154 		nested_vmx_vcpu_setup();
4155 
4156 	if (cpu_has_vmx_msr_bitmap())
4157 		vmcs_write64(MSR_BITMAP, __pa(vmx->vmcs01.msr_bitmap));
4158 
4159 	vmcs_write64(VMCS_LINK_POINTER, -1ull); /* 22.3.1.5 */
4160 
4161 	/* Control */
4162 	pin_controls_set(vmx, vmx_pin_based_exec_ctrl(vmx));
4163 	vmx->hv_deadline_tsc = -1;
4164 
4165 	exec_controls_set(vmx, vmx_exec_control(vmx));
4166 
4167 	if (cpu_has_secondary_exec_ctrls()) {
4168 		vmx_compute_secondary_exec_control(vmx);
4169 		secondary_exec_controls_set(vmx, vmx->secondary_exec_control);
4170 	}
4171 
4172 	if (kvm_vcpu_apicv_active(&vmx->vcpu)) {
4173 		vmcs_write64(EOI_EXIT_BITMAP0, 0);
4174 		vmcs_write64(EOI_EXIT_BITMAP1, 0);
4175 		vmcs_write64(EOI_EXIT_BITMAP2, 0);
4176 		vmcs_write64(EOI_EXIT_BITMAP3, 0);
4177 
4178 		vmcs_write16(GUEST_INTR_STATUS, 0);
4179 
4180 		vmcs_write16(POSTED_INTR_NV, POSTED_INTR_VECTOR);
4181 		vmcs_write64(POSTED_INTR_DESC_ADDR, __pa((&vmx->pi_desc)));
4182 	}
4183 
4184 	if (!kvm_pause_in_guest(vmx->vcpu.kvm)) {
4185 		vmcs_write32(PLE_GAP, ple_gap);
4186 		vmx->ple_window = ple_window;
4187 		vmx->ple_window_dirty = true;
4188 	}
4189 
4190 	vmcs_write32(PAGE_FAULT_ERROR_CODE_MASK, 0);
4191 	vmcs_write32(PAGE_FAULT_ERROR_CODE_MATCH, 0);
4192 	vmcs_write32(CR3_TARGET_COUNT, 0);           /* 22.2.1 */
4193 
4194 	vmcs_write16(HOST_FS_SELECTOR, 0);            /* 22.2.4 */
4195 	vmcs_write16(HOST_GS_SELECTOR, 0);            /* 22.2.4 */
4196 	vmx_set_constant_host_state(vmx);
4197 	vmcs_writel(HOST_FS_BASE, 0); /* 22.2.4 */
4198 	vmcs_writel(HOST_GS_BASE, 0); /* 22.2.4 */
4199 
4200 	if (cpu_has_vmx_vmfunc())
4201 		vmcs_write64(VM_FUNCTION_CONTROL, 0);
4202 
4203 	vmcs_write32(VM_EXIT_MSR_STORE_COUNT, 0);
4204 	vmcs_write32(VM_EXIT_MSR_LOAD_COUNT, 0);
4205 	vmcs_write64(VM_EXIT_MSR_LOAD_ADDR, __pa(vmx->msr_autoload.host.val));
4206 	vmcs_write32(VM_ENTRY_MSR_LOAD_COUNT, 0);
4207 	vmcs_write64(VM_ENTRY_MSR_LOAD_ADDR, __pa(vmx->msr_autoload.guest.val));
4208 
4209 	if (vmcs_config.vmentry_ctrl & VM_ENTRY_LOAD_IA32_PAT)
4210 		vmcs_write64(GUEST_IA32_PAT, vmx->vcpu.arch.pat);
4211 
4212 	for (i = 0; i < ARRAY_SIZE(vmx_msr_index); ++i) {
4213 		u32 index = vmx_msr_index[i];
4214 		u32 data_low, data_high;
4215 		int j = vmx->nmsrs;
4216 
4217 		if (rdmsr_safe(index, &data_low, &data_high) < 0)
4218 			continue;
4219 		if (wrmsr_safe(index, data_low, data_high) < 0)
4220 			continue;
4221 		vmx->guest_msrs[j].index = i;
4222 		vmx->guest_msrs[j].data = 0;
4223 		vmx->guest_msrs[j].mask = -1ull;
4224 		++vmx->nmsrs;
4225 	}
4226 
4227 	vm_exit_controls_set(vmx, vmx_vmexit_ctrl());
4228 
4229 	/* 22.2.1, 20.8.1 */
4230 	vm_entry_controls_set(vmx, vmx_vmentry_ctrl());
4231 
4232 	vmx->vcpu.arch.cr0_guest_owned_bits = X86_CR0_TS;
4233 	vmcs_writel(CR0_GUEST_HOST_MASK, ~X86_CR0_TS);
4234 
4235 	set_cr4_guest_host_mask(vmx);
4236 
4237 	if (vmx_xsaves_supported())
4238 		vmcs_write64(XSS_EXIT_BITMAP, VMX_XSS_EXIT_BITMAP);
4239 
4240 	if (enable_pml) {
4241 		vmcs_write64(PML_ADDRESS, page_to_phys(vmx->pml_pg));
4242 		vmcs_write16(GUEST_PML_INDEX, PML_ENTITY_NUM - 1);
4243 	}
4244 
4245 	if (cpu_has_vmx_encls_vmexit())
4246 		vmcs_write64(ENCLS_EXITING_BITMAP, -1ull);
4247 
4248 	if (pt_mode == PT_MODE_HOST_GUEST) {
4249 		memset(&vmx->pt_desc, 0, sizeof(vmx->pt_desc));
4250 		/* Bit[6~0] are forced to 1, writes are ignored. */
4251 		vmx->pt_desc.guest.output_mask = 0x7F;
4252 		vmcs_write64(GUEST_IA32_RTIT_CTL, 0);
4253 	}
4254 }
4255 
4256 static void vmx_vcpu_reset(struct kvm_vcpu *vcpu, bool init_event)
4257 {
4258 	struct vcpu_vmx *vmx = to_vmx(vcpu);
4259 	struct msr_data apic_base_msr;
4260 	u64 cr0;
4261 
4262 	vmx->rmode.vm86_active = 0;
4263 	vmx->spec_ctrl = 0;
4264 
4265 	vmx->msr_ia32_umwait_control = 0;
4266 
4267 	vcpu->arch.microcode_version = 0x100000000ULL;
4268 	vmx->vcpu.arch.regs[VCPU_REGS_RDX] = get_rdx_init_val();
4269 	vmx->hv_deadline_tsc = -1;
4270 	kvm_set_cr8(vcpu, 0);
4271 
4272 	if (!init_event) {
4273 		apic_base_msr.data = APIC_DEFAULT_PHYS_BASE |
4274 				     MSR_IA32_APICBASE_ENABLE;
4275 		if (kvm_vcpu_is_reset_bsp(vcpu))
4276 			apic_base_msr.data |= MSR_IA32_APICBASE_BSP;
4277 		apic_base_msr.host_initiated = true;
4278 		kvm_set_apic_base(vcpu, &apic_base_msr);
4279 	}
4280 
4281 	vmx_segment_cache_clear(vmx);
4282 
4283 	seg_setup(VCPU_SREG_CS);
4284 	vmcs_write16(GUEST_CS_SELECTOR, 0xf000);
4285 	vmcs_writel(GUEST_CS_BASE, 0xffff0000ul);
4286 
4287 	seg_setup(VCPU_SREG_DS);
4288 	seg_setup(VCPU_SREG_ES);
4289 	seg_setup(VCPU_SREG_FS);
4290 	seg_setup(VCPU_SREG_GS);
4291 	seg_setup(VCPU_SREG_SS);
4292 
4293 	vmcs_write16(GUEST_TR_SELECTOR, 0);
4294 	vmcs_writel(GUEST_TR_BASE, 0);
4295 	vmcs_write32(GUEST_TR_LIMIT, 0xffff);
4296 	vmcs_write32(GUEST_TR_AR_BYTES, 0x008b);
4297 
4298 	vmcs_write16(GUEST_LDTR_SELECTOR, 0);
4299 	vmcs_writel(GUEST_LDTR_BASE, 0);
4300 	vmcs_write32(GUEST_LDTR_LIMIT, 0xffff);
4301 	vmcs_write32(GUEST_LDTR_AR_BYTES, 0x00082);
4302 
4303 	if (!init_event) {
4304 		vmcs_write32(GUEST_SYSENTER_CS, 0);
4305 		vmcs_writel(GUEST_SYSENTER_ESP, 0);
4306 		vmcs_writel(GUEST_SYSENTER_EIP, 0);
4307 		vmcs_write64(GUEST_IA32_DEBUGCTL, 0);
4308 	}
4309 
4310 	kvm_set_rflags(vcpu, X86_EFLAGS_FIXED);
4311 	kvm_rip_write(vcpu, 0xfff0);
4312 
4313 	vmcs_writel(GUEST_GDTR_BASE, 0);
4314 	vmcs_write32(GUEST_GDTR_LIMIT, 0xffff);
4315 
4316 	vmcs_writel(GUEST_IDTR_BASE, 0);
4317 	vmcs_write32(GUEST_IDTR_LIMIT, 0xffff);
4318 
4319 	vmcs_write32(GUEST_ACTIVITY_STATE, GUEST_ACTIVITY_ACTIVE);
4320 	vmcs_write32(GUEST_INTERRUPTIBILITY_INFO, 0);
4321 	vmcs_writel(GUEST_PENDING_DBG_EXCEPTIONS, 0);
4322 	if (kvm_mpx_supported())
4323 		vmcs_write64(GUEST_BNDCFGS, 0);
4324 
4325 	setup_msrs(vmx);
4326 
4327 	vmcs_write32(VM_ENTRY_INTR_INFO_FIELD, 0);  /* 22.2.1 */
4328 
4329 	if (cpu_has_vmx_tpr_shadow() && !init_event) {
4330 		vmcs_write64(VIRTUAL_APIC_PAGE_ADDR, 0);
4331 		if (cpu_need_tpr_shadow(vcpu))
4332 			vmcs_write64(VIRTUAL_APIC_PAGE_ADDR,
4333 				     __pa(vcpu->arch.apic->regs));
4334 		vmcs_write32(TPR_THRESHOLD, 0);
4335 	}
4336 
4337 	kvm_make_request(KVM_REQ_APIC_PAGE_RELOAD, vcpu);
4338 
4339 	if (vmx->vpid != 0)
4340 		vmcs_write16(VIRTUAL_PROCESSOR_ID, vmx->vpid);
4341 
4342 	cr0 = X86_CR0_NW | X86_CR0_CD | X86_CR0_ET;
4343 	vmx->vcpu.arch.cr0 = cr0;
4344 	vmx_set_cr0(vcpu, cr0); /* enter rmode */
4345 	vmx_set_cr4(vcpu, 0);
4346 	vmx_set_efer(vcpu, 0);
4347 
4348 	update_exception_bitmap(vcpu);
4349 
4350 	vpid_sync_context(vmx->vpid);
4351 	if (init_event)
4352 		vmx_clear_hlt(vcpu);
4353 }
4354 
4355 static void enable_irq_window(struct kvm_vcpu *vcpu)
4356 {
4357 	exec_controls_setbit(to_vmx(vcpu), CPU_BASED_VIRTUAL_INTR_PENDING);
4358 }
4359 
4360 static void enable_nmi_window(struct kvm_vcpu *vcpu)
4361 {
4362 	if (!enable_vnmi ||
4363 	    vmcs_read32(GUEST_INTERRUPTIBILITY_INFO) & GUEST_INTR_STATE_STI) {
4364 		enable_irq_window(vcpu);
4365 		return;
4366 	}
4367 
4368 	exec_controls_setbit(to_vmx(vcpu), CPU_BASED_VIRTUAL_NMI_PENDING);
4369 }
4370 
4371 static void vmx_inject_irq(struct kvm_vcpu *vcpu)
4372 {
4373 	struct vcpu_vmx *vmx = to_vmx(vcpu);
4374 	uint32_t intr;
4375 	int irq = vcpu->arch.interrupt.nr;
4376 
4377 	trace_kvm_inj_virq(irq);
4378 
4379 	++vcpu->stat.irq_injections;
4380 	if (vmx->rmode.vm86_active) {
4381 		int inc_eip = 0;
4382 		if (vcpu->arch.interrupt.soft)
4383 			inc_eip = vcpu->arch.event_exit_inst_len;
4384 		kvm_inject_realmode_interrupt(vcpu, irq, inc_eip);
4385 		return;
4386 	}
4387 	intr = irq | INTR_INFO_VALID_MASK;
4388 	if (vcpu->arch.interrupt.soft) {
4389 		intr |= INTR_TYPE_SOFT_INTR;
4390 		vmcs_write32(VM_ENTRY_INSTRUCTION_LEN,
4391 			     vmx->vcpu.arch.event_exit_inst_len);
4392 	} else
4393 		intr |= INTR_TYPE_EXT_INTR;
4394 	vmcs_write32(VM_ENTRY_INTR_INFO_FIELD, intr);
4395 
4396 	vmx_clear_hlt(vcpu);
4397 }
4398 
4399 static void vmx_inject_nmi(struct kvm_vcpu *vcpu)
4400 {
4401 	struct vcpu_vmx *vmx = to_vmx(vcpu);
4402 
4403 	if (!enable_vnmi) {
4404 		/*
4405 		 * Tracking the NMI-blocked state in software is built upon
4406 		 * finding the next open IRQ window. This, in turn, depends on
4407 		 * well-behaving guests: They have to keep IRQs disabled at
4408 		 * least as long as the NMI handler runs. Otherwise we may
4409 		 * cause NMI nesting, maybe breaking the guest. But as this is
4410 		 * highly unlikely, we can live with the residual risk.
4411 		 */
4412 		vmx->loaded_vmcs->soft_vnmi_blocked = 1;
4413 		vmx->loaded_vmcs->vnmi_blocked_time = 0;
4414 	}
4415 
4416 	++vcpu->stat.nmi_injections;
4417 	vmx->loaded_vmcs->nmi_known_unmasked = false;
4418 
4419 	if (vmx->rmode.vm86_active) {
4420 		kvm_inject_realmode_interrupt(vcpu, NMI_VECTOR, 0);
4421 		return;
4422 	}
4423 
4424 	vmcs_write32(VM_ENTRY_INTR_INFO_FIELD,
4425 			INTR_TYPE_NMI_INTR | INTR_INFO_VALID_MASK | NMI_VECTOR);
4426 
4427 	vmx_clear_hlt(vcpu);
4428 }
4429 
4430 bool vmx_get_nmi_mask(struct kvm_vcpu *vcpu)
4431 {
4432 	struct vcpu_vmx *vmx = to_vmx(vcpu);
4433 	bool masked;
4434 
4435 	if (!enable_vnmi)
4436 		return vmx->loaded_vmcs->soft_vnmi_blocked;
4437 	if (vmx->loaded_vmcs->nmi_known_unmasked)
4438 		return false;
4439 	masked = vmcs_read32(GUEST_INTERRUPTIBILITY_INFO) & GUEST_INTR_STATE_NMI;
4440 	vmx->loaded_vmcs->nmi_known_unmasked = !masked;
4441 	return masked;
4442 }
4443 
4444 void vmx_set_nmi_mask(struct kvm_vcpu *vcpu, bool masked)
4445 {
4446 	struct vcpu_vmx *vmx = to_vmx(vcpu);
4447 
4448 	if (!enable_vnmi) {
4449 		if (vmx->loaded_vmcs->soft_vnmi_blocked != masked) {
4450 			vmx->loaded_vmcs->soft_vnmi_blocked = masked;
4451 			vmx->loaded_vmcs->vnmi_blocked_time = 0;
4452 		}
4453 	} else {
4454 		vmx->loaded_vmcs->nmi_known_unmasked = !masked;
4455 		if (masked)
4456 			vmcs_set_bits(GUEST_INTERRUPTIBILITY_INFO,
4457 				      GUEST_INTR_STATE_NMI);
4458 		else
4459 			vmcs_clear_bits(GUEST_INTERRUPTIBILITY_INFO,
4460 					GUEST_INTR_STATE_NMI);
4461 	}
4462 }
4463 
4464 static int vmx_nmi_allowed(struct kvm_vcpu *vcpu)
4465 {
4466 	if (to_vmx(vcpu)->nested.nested_run_pending)
4467 		return 0;
4468 
4469 	if (!enable_vnmi &&
4470 	    to_vmx(vcpu)->loaded_vmcs->soft_vnmi_blocked)
4471 		return 0;
4472 
4473 	return	!(vmcs_read32(GUEST_INTERRUPTIBILITY_INFO) &
4474 		  (GUEST_INTR_STATE_MOV_SS | GUEST_INTR_STATE_STI
4475 		   | GUEST_INTR_STATE_NMI));
4476 }
4477 
4478 static int vmx_interrupt_allowed(struct kvm_vcpu *vcpu)
4479 {
4480 	return (!to_vmx(vcpu)->nested.nested_run_pending &&
4481 		vmcs_readl(GUEST_RFLAGS) & X86_EFLAGS_IF) &&
4482 		!(vmcs_read32(GUEST_INTERRUPTIBILITY_INFO) &
4483 			(GUEST_INTR_STATE_STI | GUEST_INTR_STATE_MOV_SS));
4484 }
4485 
4486 static int vmx_set_tss_addr(struct kvm *kvm, unsigned int addr)
4487 {
4488 	int ret;
4489 
4490 	if (enable_unrestricted_guest)
4491 		return 0;
4492 
4493 	ret = x86_set_memory_region(kvm, TSS_PRIVATE_MEMSLOT, addr,
4494 				    PAGE_SIZE * 3);
4495 	if (ret)
4496 		return ret;
4497 	to_kvm_vmx(kvm)->tss_addr = addr;
4498 	return init_rmode_tss(kvm);
4499 }
4500 
4501 static int vmx_set_identity_map_addr(struct kvm *kvm, u64 ident_addr)
4502 {
4503 	to_kvm_vmx(kvm)->ept_identity_map_addr = ident_addr;
4504 	return 0;
4505 }
4506 
4507 static bool rmode_exception(struct kvm_vcpu *vcpu, int vec)
4508 {
4509 	switch (vec) {
4510 	case BP_VECTOR:
4511 		/*
4512 		 * Update instruction length as we may reinject the exception
4513 		 * from user space while in guest debugging mode.
4514 		 */
4515 		to_vmx(vcpu)->vcpu.arch.event_exit_inst_len =
4516 			vmcs_read32(VM_EXIT_INSTRUCTION_LEN);
4517 		if (vcpu->guest_debug & KVM_GUESTDBG_USE_SW_BP)
4518 			return false;
4519 		/* fall through */
4520 	case DB_VECTOR:
4521 		if (vcpu->guest_debug &
4522 			(KVM_GUESTDBG_SINGLESTEP | KVM_GUESTDBG_USE_HW_BP))
4523 			return false;
4524 		/* fall through */
4525 	case DE_VECTOR:
4526 	case OF_VECTOR:
4527 	case BR_VECTOR:
4528 	case UD_VECTOR:
4529 	case DF_VECTOR:
4530 	case SS_VECTOR:
4531 	case GP_VECTOR:
4532 	case MF_VECTOR:
4533 		return true;
4534 	break;
4535 	}
4536 	return false;
4537 }
4538 
4539 static int handle_rmode_exception(struct kvm_vcpu *vcpu,
4540 				  int vec, u32 err_code)
4541 {
4542 	/*
4543 	 * Instruction with address size override prefix opcode 0x67
4544 	 * Cause the #SS fault with 0 error code in VM86 mode.
4545 	 */
4546 	if (((vec == GP_VECTOR) || (vec == SS_VECTOR)) && err_code == 0) {
4547 		if (kvm_emulate_instruction(vcpu, 0)) {
4548 			if (vcpu->arch.halt_request) {
4549 				vcpu->arch.halt_request = 0;
4550 				return kvm_vcpu_halt(vcpu);
4551 			}
4552 			return 1;
4553 		}
4554 		return 0;
4555 	}
4556 
4557 	/*
4558 	 * Forward all other exceptions that are valid in real mode.
4559 	 * FIXME: Breaks guest debugging in real mode, needs to be fixed with
4560 	 *        the required debugging infrastructure rework.
4561 	 */
4562 	kvm_queue_exception(vcpu, vec);
4563 	return 1;
4564 }
4565 
4566 /*
4567  * Trigger machine check on the host. We assume all the MSRs are already set up
4568  * by the CPU and that we still run on the same CPU as the MCE occurred on.
4569  * We pass a fake environment to the machine check handler because we want
4570  * the guest to be always treated like user space, no matter what context
4571  * it used internally.
4572  */
4573 static void kvm_machine_check(void)
4574 {
4575 #if defined(CONFIG_X86_MCE) && defined(CONFIG_X86_64)
4576 	struct pt_regs regs = {
4577 		.cs = 3, /* Fake ring 3 no matter what the guest ran on */
4578 		.flags = X86_EFLAGS_IF,
4579 	};
4580 
4581 	do_machine_check(&regs, 0);
4582 #endif
4583 }
4584 
4585 static int handle_machine_check(struct kvm_vcpu *vcpu)
4586 {
4587 	/* handled by vmx_vcpu_run() */
4588 	return 1;
4589 }
4590 
4591 static int handle_exception_nmi(struct kvm_vcpu *vcpu)
4592 {
4593 	struct vcpu_vmx *vmx = to_vmx(vcpu);
4594 	struct kvm_run *kvm_run = vcpu->run;
4595 	u32 intr_info, ex_no, error_code;
4596 	unsigned long cr2, rip, dr6;
4597 	u32 vect_info;
4598 
4599 	vect_info = vmx->idt_vectoring_info;
4600 	intr_info = vmx->exit_intr_info;
4601 
4602 	if (is_machine_check(intr_info) || is_nmi(intr_info))
4603 		return 1; /* handled by handle_exception_nmi_irqoff() */
4604 
4605 	if (is_invalid_opcode(intr_info))
4606 		return handle_ud(vcpu);
4607 
4608 	error_code = 0;
4609 	if (intr_info & INTR_INFO_DELIVER_CODE_MASK)
4610 		error_code = vmcs_read32(VM_EXIT_INTR_ERROR_CODE);
4611 
4612 	if (!vmx->rmode.vm86_active && is_gp_fault(intr_info)) {
4613 		WARN_ON_ONCE(!enable_vmware_backdoor);
4614 
4615 		/*
4616 		 * VMware backdoor emulation on #GP interception only handles
4617 		 * IN{S}, OUT{S}, and RDPMC, none of which generate a non-zero
4618 		 * error code on #GP.
4619 		 */
4620 		if (error_code) {
4621 			kvm_queue_exception_e(vcpu, GP_VECTOR, error_code);
4622 			return 1;
4623 		}
4624 		return kvm_emulate_instruction(vcpu, EMULTYPE_VMWARE_GP);
4625 	}
4626 
4627 	/*
4628 	 * The #PF with PFEC.RSVD = 1 indicates the guest is accessing
4629 	 * MMIO, it is better to report an internal error.
4630 	 * See the comments in vmx_handle_exit.
4631 	 */
4632 	if ((vect_info & VECTORING_INFO_VALID_MASK) &&
4633 	    !(is_page_fault(intr_info) && !(error_code & PFERR_RSVD_MASK))) {
4634 		vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
4635 		vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_SIMUL_EX;
4636 		vcpu->run->internal.ndata = 3;
4637 		vcpu->run->internal.data[0] = vect_info;
4638 		vcpu->run->internal.data[1] = intr_info;
4639 		vcpu->run->internal.data[2] = error_code;
4640 		return 0;
4641 	}
4642 
4643 	if (is_page_fault(intr_info)) {
4644 		cr2 = vmcs_readl(EXIT_QUALIFICATION);
4645 		/* EPT won't cause page fault directly */
4646 		WARN_ON_ONCE(!vcpu->arch.apf.host_apf_reason && enable_ept);
4647 		return kvm_handle_page_fault(vcpu, error_code, cr2, NULL, 0);
4648 	}
4649 
4650 	ex_no = intr_info & INTR_INFO_VECTOR_MASK;
4651 
4652 	if (vmx->rmode.vm86_active && rmode_exception(vcpu, ex_no))
4653 		return handle_rmode_exception(vcpu, ex_no, error_code);
4654 
4655 	switch (ex_no) {
4656 	case AC_VECTOR:
4657 		kvm_queue_exception_e(vcpu, AC_VECTOR, error_code);
4658 		return 1;
4659 	case DB_VECTOR:
4660 		dr6 = vmcs_readl(EXIT_QUALIFICATION);
4661 		if (!(vcpu->guest_debug &
4662 		      (KVM_GUESTDBG_SINGLESTEP | KVM_GUESTDBG_USE_HW_BP))) {
4663 			vcpu->arch.dr6 &= ~DR_TRAP_BITS;
4664 			vcpu->arch.dr6 |= dr6 | DR6_RTM;
4665 			if (is_icebp(intr_info))
4666 				WARN_ON(!skip_emulated_instruction(vcpu));
4667 
4668 			kvm_queue_exception(vcpu, DB_VECTOR);
4669 			return 1;
4670 		}
4671 		kvm_run->debug.arch.dr6 = dr6 | DR6_FIXED_1;
4672 		kvm_run->debug.arch.dr7 = vmcs_readl(GUEST_DR7);
4673 		/* fall through */
4674 	case BP_VECTOR:
4675 		/*
4676 		 * Update instruction length as we may reinject #BP from
4677 		 * user space while in guest debugging mode. Reading it for
4678 		 * #DB as well causes no harm, it is not used in that case.
4679 		 */
4680 		vmx->vcpu.arch.event_exit_inst_len =
4681 			vmcs_read32(VM_EXIT_INSTRUCTION_LEN);
4682 		kvm_run->exit_reason = KVM_EXIT_DEBUG;
4683 		rip = kvm_rip_read(vcpu);
4684 		kvm_run->debug.arch.pc = vmcs_readl(GUEST_CS_BASE) + rip;
4685 		kvm_run->debug.arch.exception = ex_no;
4686 		break;
4687 	default:
4688 		kvm_run->exit_reason = KVM_EXIT_EXCEPTION;
4689 		kvm_run->ex.exception = ex_no;
4690 		kvm_run->ex.error_code = error_code;
4691 		break;
4692 	}
4693 	return 0;
4694 }
4695 
4696 static int handle_external_interrupt(struct kvm_vcpu *vcpu)
4697 {
4698 	++vcpu->stat.irq_exits;
4699 	return 1;
4700 }
4701 
4702 static int handle_triple_fault(struct kvm_vcpu *vcpu)
4703 {
4704 	vcpu->run->exit_reason = KVM_EXIT_SHUTDOWN;
4705 	vcpu->mmio_needed = 0;
4706 	return 0;
4707 }
4708 
4709 static int handle_io(struct kvm_vcpu *vcpu)
4710 {
4711 	unsigned long exit_qualification;
4712 	int size, in, string;
4713 	unsigned port;
4714 
4715 	exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
4716 	string = (exit_qualification & 16) != 0;
4717 
4718 	++vcpu->stat.io_exits;
4719 
4720 	if (string)
4721 		return kvm_emulate_instruction(vcpu, 0);
4722 
4723 	port = exit_qualification >> 16;
4724 	size = (exit_qualification & 7) + 1;
4725 	in = (exit_qualification & 8) != 0;
4726 
4727 	return kvm_fast_pio(vcpu, size, port, in);
4728 }
4729 
4730 static void
4731 vmx_patch_hypercall(struct kvm_vcpu *vcpu, unsigned char *hypercall)
4732 {
4733 	/*
4734 	 * Patch in the VMCALL instruction:
4735 	 */
4736 	hypercall[0] = 0x0f;
4737 	hypercall[1] = 0x01;
4738 	hypercall[2] = 0xc1;
4739 }
4740 
4741 /* called to set cr0 as appropriate for a mov-to-cr0 exit. */
4742 static int handle_set_cr0(struct kvm_vcpu *vcpu, unsigned long val)
4743 {
4744 	if (is_guest_mode(vcpu)) {
4745 		struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
4746 		unsigned long orig_val = val;
4747 
4748 		/*
4749 		 * We get here when L2 changed cr0 in a way that did not change
4750 		 * any of L1's shadowed bits (see nested_vmx_exit_handled_cr),
4751 		 * but did change L0 shadowed bits. So we first calculate the
4752 		 * effective cr0 value that L1 would like to write into the
4753 		 * hardware. It consists of the L2-owned bits from the new
4754 		 * value combined with the L1-owned bits from L1's guest_cr0.
4755 		 */
4756 		val = (val & ~vmcs12->cr0_guest_host_mask) |
4757 			(vmcs12->guest_cr0 & vmcs12->cr0_guest_host_mask);
4758 
4759 		if (!nested_guest_cr0_valid(vcpu, val))
4760 			return 1;
4761 
4762 		if (kvm_set_cr0(vcpu, val))
4763 			return 1;
4764 		vmcs_writel(CR0_READ_SHADOW, orig_val);
4765 		return 0;
4766 	} else {
4767 		if (to_vmx(vcpu)->nested.vmxon &&
4768 		    !nested_host_cr0_valid(vcpu, val))
4769 			return 1;
4770 
4771 		return kvm_set_cr0(vcpu, val);
4772 	}
4773 }
4774 
4775 static int handle_set_cr4(struct kvm_vcpu *vcpu, unsigned long val)
4776 {
4777 	if (is_guest_mode(vcpu)) {
4778 		struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
4779 		unsigned long orig_val = val;
4780 
4781 		/* analogously to handle_set_cr0 */
4782 		val = (val & ~vmcs12->cr4_guest_host_mask) |
4783 			(vmcs12->guest_cr4 & vmcs12->cr4_guest_host_mask);
4784 		if (kvm_set_cr4(vcpu, val))
4785 			return 1;
4786 		vmcs_writel(CR4_READ_SHADOW, orig_val);
4787 		return 0;
4788 	} else
4789 		return kvm_set_cr4(vcpu, val);
4790 }
4791 
4792 static int handle_desc(struct kvm_vcpu *vcpu)
4793 {
4794 	WARN_ON(!(vcpu->arch.cr4 & X86_CR4_UMIP));
4795 	return kvm_emulate_instruction(vcpu, 0);
4796 }
4797 
4798 static int handle_cr(struct kvm_vcpu *vcpu)
4799 {
4800 	unsigned long exit_qualification, val;
4801 	int cr;
4802 	int reg;
4803 	int err;
4804 	int ret;
4805 
4806 	exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
4807 	cr = exit_qualification & 15;
4808 	reg = (exit_qualification >> 8) & 15;
4809 	switch ((exit_qualification >> 4) & 3) {
4810 	case 0: /* mov to cr */
4811 		val = kvm_register_readl(vcpu, reg);
4812 		trace_kvm_cr_write(cr, val);
4813 		switch (cr) {
4814 		case 0:
4815 			err = handle_set_cr0(vcpu, val);
4816 			return kvm_complete_insn_gp(vcpu, err);
4817 		case 3:
4818 			WARN_ON_ONCE(enable_unrestricted_guest);
4819 			err = kvm_set_cr3(vcpu, val);
4820 			return kvm_complete_insn_gp(vcpu, err);
4821 		case 4:
4822 			err = handle_set_cr4(vcpu, val);
4823 			return kvm_complete_insn_gp(vcpu, err);
4824 		case 8: {
4825 				u8 cr8_prev = kvm_get_cr8(vcpu);
4826 				u8 cr8 = (u8)val;
4827 				err = kvm_set_cr8(vcpu, cr8);
4828 				ret = kvm_complete_insn_gp(vcpu, err);
4829 				if (lapic_in_kernel(vcpu))
4830 					return ret;
4831 				if (cr8_prev <= cr8)
4832 					return ret;
4833 				/*
4834 				 * TODO: we might be squashing a
4835 				 * KVM_GUESTDBG_SINGLESTEP-triggered
4836 				 * KVM_EXIT_DEBUG here.
4837 				 */
4838 				vcpu->run->exit_reason = KVM_EXIT_SET_TPR;
4839 				return 0;
4840 			}
4841 		}
4842 		break;
4843 	case 2: /* clts */
4844 		WARN_ONCE(1, "Guest should always own CR0.TS");
4845 		vmx_set_cr0(vcpu, kvm_read_cr0_bits(vcpu, ~X86_CR0_TS));
4846 		trace_kvm_cr_write(0, kvm_read_cr0(vcpu));
4847 		return kvm_skip_emulated_instruction(vcpu);
4848 	case 1: /*mov from cr*/
4849 		switch (cr) {
4850 		case 3:
4851 			WARN_ON_ONCE(enable_unrestricted_guest);
4852 			val = kvm_read_cr3(vcpu);
4853 			kvm_register_write(vcpu, reg, val);
4854 			trace_kvm_cr_read(cr, val);
4855 			return kvm_skip_emulated_instruction(vcpu);
4856 		case 8:
4857 			val = kvm_get_cr8(vcpu);
4858 			kvm_register_write(vcpu, reg, val);
4859 			trace_kvm_cr_read(cr, val);
4860 			return kvm_skip_emulated_instruction(vcpu);
4861 		}
4862 		break;
4863 	case 3: /* lmsw */
4864 		val = (exit_qualification >> LMSW_SOURCE_DATA_SHIFT) & 0x0f;
4865 		trace_kvm_cr_write(0, (kvm_read_cr0(vcpu) & ~0xful) | val);
4866 		kvm_lmsw(vcpu, val);
4867 
4868 		return kvm_skip_emulated_instruction(vcpu);
4869 	default:
4870 		break;
4871 	}
4872 	vcpu->run->exit_reason = 0;
4873 	vcpu_unimpl(vcpu, "unhandled control register: op %d cr %d\n",
4874 	       (int)(exit_qualification >> 4) & 3, cr);
4875 	return 0;
4876 }
4877 
4878 static int handle_dr(struct kvm_vcpu *vcpu)
4879 {
4880 	unsigned long exit_qualification;
4881 	int dr, dr7, reg;
4882 
4883 	exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
4884 	dr = exit_qualification & DEBUG_REG_ACCESS_NUM;
4885 
4886 	/* First, if DR does not exist, trigger UD */
4887 	if (!kvm_require_dr(vcpu, dr))
4888 		return 1;
4889 
4890 	/* Do not handle if the CPL > 0, will trigger GP on re-entry */
4891 	if (!kvm_require_cpl(vcpu, 0))
4892 		return 1;
4893 	dr7 = vmcs_readl(GUEST_DR7);
4894 	if (dr7 & DR7_GD) {
4895 		/*
4896 		 * As the vm-exit takes precedence over the debug trap, we
4897 		 * need to emulate the latter, either for the host or the
4898 		 * guest debugging itself.
4899 		 */
4900 		if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP) {
4901 			vcpu->run->debug.arch.dr6 = vcpu->arch.dr6;
4902 			vcpu->run->debug.arch.dr7 = dr7;
4903 			vcpu->run->debug.arch.pc = kvm_get_linear_rip(vcpu);
4904 			vcpu->run->debug.arch.exception = DB_VECTOR;
4905 			vcpu->run->exit_reason = KVM_EXIT_DEBUG;
4906 			return 0;
4907 		} else {
4908 			vcpu->arch.dr6 &= ~DR_TRAP_BITS;
4909 			vcpu->arch.dr6 |= DR6_BD | DR6_RTM;
4910 			kvm_queue_exception(vcpu, DB_VECTOR);
4911 			return 1;
4912 		}
4913 	}
4914 
4915 	if (vcpu->guest_debug == 0) {
4916 		exec_controls_clearbit(to_vmx(vcpu), CPU_BASED_MOV_DR_EXITING);
4917 
4918 		/*
4919 		 * No more DR vmexits; force a reload of the debug registers
4920 		 * and reenter on this instruction.  The next vmexit will
4921 		 * retrieve the full state of the debug registers.
4922 		 */
4923 		vcpu->arch.switch_db_regs |= KVM_DEBUGREG_WONT_EXIT;
4924 		return 1;
4925 	}
4926 
4927 	reg = DEBUG_REG_ACCESS_REG(exit_qualification);
4928 	if (exit_qualification & TYPE_MOV_FROM_DR) {
4929 		unsigned long val;
4930 
4931 		if (kvm_get_dr(vcpu, dr, &val))
4932 			return 1;
4933 		kvm_register_write(vcpu, reg, val);
4934 	} else
4935 		if (kvm_set_dr(vcpu, dr, kvm_register_readl(vcpu, reg)))
4936 			return 1;
4937 
4938 	return kvm_skip_emulated_instruction(vcpu);
4939 }
4940 
4941 static u64 vmx_get_dr6(struct kvm_vcpu *vcpu)
4942 {
4943 	return vcpu->arch.dr6;
4944 }
4945 
4946 static void vmx_set_dr6(struct kvm_vcpu *vcpu, unsigned long val)
4947 {
4948 }
4949 
4950 static void vmx_sync_dirty_debug_regs(struct kvm_vcpu *vcpu)
4951 {
4952 	get_debugreg(vcpu->arch.db[0], 0);
4953 	get_debugreg(vcpu->arch.db[1], 1);
4954 	get_debugreg(vcpu->arch.db[2], 2);
4955 	get_debugreg(vcpu->arch.db[3], 3);
4956 	get_debugreg(vcpu->arch.dr6, 6);
4957 	vcpu->arch.dr7 = vmcs_readl(GUEST_DR7);
4958 
4959 	vcpu->arch.switch_db_regs &= ~KVM_DEBUGREG_WONT_EXIT;
4960 	exec_controls_setbit(to_vmx(vcpu), CPU_BASED_MOV_DR_EXITING);
4961 }
4962 
4963 static void vmx_set_dr7(struct kvm_vcpu *vcpu, unsigned long val)
4964 {
4965 	vmcs_writel(GUEST_DR7, val);
4966 }
4967 
4968 static int handle_cpuid(struct kvm_vcpu *vcpu)
4969 {
4970 	return kvm_emulate_cpuid(vcpu);
4971 }
4972 
4973 static int handle_rdmsr(struct kvm_vcpu *vcpu)
4974 {
4975 	return kvm_emulate_rdmsr(vcpu);
4976 }
4977 
4978 static int handle_wrmsr(struct kvm_vcpu *vcpu)
4979 {
4980 	return kvm_emulate_wrmsr(vcpu);
4981 }
4982 
4983 static int handle_tpr_below_threshold(struct kvm_vcpu *vcpu)
4984 {
4985 	kvm_apic_update_ppr(vcpu);
4986 	return 1;
4987 }
4988 
4989 static int handle_interrupt_window(struct kvm_vcpu *vcpu)
4990 {
4991 	exec_controls_clearbit(to_vmx(vcpu), CPU_BASED_VIRTUAL_INTR_PENDING);
4992 
4993 	kvm_make_request(KVM_REQ_EVENT, vcpu);
4994 
4995 	++vcpu->stat.irq_window_exits;
4996 	return 1;
4997 }
4998 
4999 static int handle_halt(struct kvm_vcpu *vcpu)
5000 {
5001 	return kvm_emulate_halt(vcpu);
5002 }
5003 
5004 static int handle_vmcall(struct kvm_vcpu *vcpu)
5005 {
5006 	return kvm_emulate_hypercall(vcpu);
5007 }
5008 
5009 static int handle_invd(struct kvm_vcpu *vcpu)
5010 {
5011 	return kvm_emulate_instruction(vcpu, 0);
5012 }
5013 
5014 static int handle_invlpg(struct kvm_vcpu *vcpu)
5015 {
5016 	unsigned long exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
5017 
5018 	kvm_mmu_invlpg(vcpu, exit_qualification);
5019 	return kvm_skip_emulated_instruction(vcpu);
5020 }
5021 
5022 static int handle_rdpmc(struct kvm_vcpu *vcpu)
5023 {
5024 	int err;
5025 
5026 	err = kvm_rdpmc(vcpu);
5027 	return kvm_complete_insn_gp(vcpu, err);
5028 }
5029 
5030 static int handle_wbinvd(struct kvm_vcpu *vcpu)
5031 {
5032 	return kvm_emulate_wbinvd(vcpu);
5033 }
5034 
5035 static int handle_xsetbv(struct kvm_vcpu *vcpu)
5036 {
5037 	u64 new_bv = kvm_read_edx_eax(vcpu);
5038 	u32 index = kvm_rcx_read(vcpu);
5039 
5040 	if (kvm_set_xcr(vcpu, index, new_bv) == 0)
5041 		return kvm_skip_emulated_instruction(vcpu);
5042 	return 1;
5043 }
5044 
5045 static int handle_apic_access(struct kvm_vcpu *vcpu)
5046 {
5047 	if (likely(fasteoi)) {
5048 		unsigned long exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
5049 		int access_type, offset;
5050 
5051 		access_type = exit_qualification & APIC_ACCESS_TYPE;
5052 		offset = exit_qualification & APIC_ACCESS_OFFSET;
5053 		/*
5054 		 * Sane guest uses MOV to write EOI, with written value
5055 		 * not cared. So make a short-circuit here by avoiding
5056 		 * heavy instruction emulation.
5057 		 */
5058 		if ((access_type == TYPE_LINEAR_APIC_INST_WRITE) &&
5059 		    (offset == APIC_EOI)) {
5060 			kvm_lapic_set_eoi(vcpu);
5061 			return kvm_skip_emulated_instruction(vcpu);
5062 		}
5063 	}
5064 	return kvm_emulate_instruction(vcpu, 0);
5065 }
5066 
5067 static int handle_apic_eoi_induced(struct kvm_vcpu *vcpu)
5068 {
5069 	unsigned long exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
5070 	int vector = exit_qualification & 0xff;
5071 
5072 	/* EOI-induced VM exit is trap-like and thus no need to adjust IP */
5073 	kvm_apic_set_eoi_accelerated(vcpu, vector);
5074 	return 1;
5075 }
5076 
5077 static int handle_apic_write(struct kvm_vcpu *vcpu)
5078 {
5079 	unsigned long exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
5080 	u32 offset = exit_qualification & 0xfff;
5081 
5082 	/* APIC-write VM exit is trap-like and thus no need to adjust IP */
5083 	kvm_apic_write_nodecode(vcpu, offset);
5084 	return 1;
5085 }
5086 
5087 static int handle_task_switch(struct kvm_vcpu *vcpu)
5088 {
5089 	struct vcpu_vmx *vmx = to_vmx(vcpu);
5090 	unsigned long exit_qualification;
5091 	bool has_error_code = false;
5092 	u32 error_code = 0;
5093 	u16 tss_selector;
5094 	int reason, type, idt_v, idt_index;
5095 
5096 	idt_v = (vmx->idt_vectoring_info & VECTORING_INFO_VALID_MASK);
5097 	idt_index = (vmx->idt_vectoring_info & VECTORING_INFO_VECTOR_MASK);
5098 	type = (vmx->idt_vectoring_info & VECTORING_INFO_TYPE_MASK);
5099 
5100 	exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
5101 
5102 	reason = (u32)exit_qualification >> 30;
5103 	if (reason == TASK_SWITCH_GATE && idt_v) {
5104 		switch (type) {
5105 		case INTR_TYPE_NMI_INTR:
5106 			vcpu->arch.nmi_injected = false;
5107 			vmx_set_nmi_mask(vcpu, true);
5108 			break;
5109 		case INTR_TYPE_EXT_INTR:
5110 		case INTR_TYPE_SOFT_INTR:
5111 			kvm_clear_interrupt_queue(vcpu);
5112 			break;
5113 		case INTR_TYPE_HARD_EXCEPTION:
5114 			if (vmx->idt_vectoring_info &
5115 			    VECTORING_INFO_DELIVER_CODE_MASK) {
5116 				has_error_code = true;
5117 				error_code =
5118 					vmcs_read32(IDT_VECTORING_ERROR_CODE);
5119 			}
5120 			/* fall through */
5121 		case INTR_TYPE_SOFT_EXCEPTION:
5122 			kvm_clear_exception_queue(vcpu);
5123 			break;
5124 		default:
5125 			break;
5126 		}
5127 	}
5128 	tss_selector = exit_qualification;
5129 
5130 	if (!idt_v || (type != INTR_TYPE_HARD_EXCEPTION &&
5131 		       type != INTR_TYPE_EXT_INTR &&
5132 		       type != INTR_TYPE_NMI_INTR))
5133 		WARN_ON(!skip_emulated_instruction(vcpu));
5134 
5135 	/*
5136 	 * TODO: What about debug traps on tss switch?
5137 	 *       Are we supposed to inject them and update dr6?
5138 	 */
5139 	return kvm_task_switch(vcpu, tss_selector,
5140 			       type == INTR_TYPE_SOFT_INTR ? idt_index : -1,
5141 			       reason, has_error_code, error_code);
5142 }
5143 
5144 static int handle_ept_violation(struct kvm_vcpu *vcpu)
5145 {
5146 	unsigned long exit_qualification;
5147 	gpa_t gpa;
5148 	u64 error_code;
5149 
5150 	exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
5151 
5152 	/*
5153 	 * EPT violation happened while executing iret from NMI,
5154 	 * "blocked by NMI" bit has to be set before next VM entry.
5155 	 * There are errata that may cause this bit to not be set:
5156 	 * AAK134, BY25.
5157 	 */
5158 	if (!(to_vmx(vcpu)->idt_vectoring_info & VECTORING_INFO_VALID_MASK) &&
5159 			enable_vnmi &&
5160 			(exit_qualification & INTR_INFO_UNBLOCK_NMI))
5161 		vmcs_set_bits(GUEST_INTERRUPTIBILITY_INFO, GUEST_INTR_STATE_NMI);
5162 
5163 	gpa = vmcs_read64(GUEST_PHYSICAL_ADDRESS);
5164 	trace_kvm_page_fault(gpa, exit_qualification);
5165 
5166 	/* Is it a read fault? */
5167 	error_code = (exit_qualification & EPT_VIOLATION_ACC_READ)
5168 		     ? PFERR_USER_MASK : 0;
5169 	/* Is it a write fault? */
5170 	error_code |= (exit_qualification & EPT_VIOLATION_ACC_WRITE)
5171 		      ? PFERR_WRITE_MASK : 0;
5172 	/* Is it a fetch fault? */
5173 	error_code |= (exit_qualification & EPT_VIOLATION_ACC_INSTR)
5174 		      ? PFERR_FETCH_MASK : 0;
5175 	/* ept page table entry is present? */
5176 	error_code |= (exit_qualification &
5177 		       (EPT_VIOLATION_READABLE | EPT_VIOLATION_WRITABLE |
5178 			EPT_VIOLATION_EXECUTABLE))
5179 		      ? PFERR_PRESENT_MASK : 0;
5180 
5181 	error_code |= (exit_qualification & 0x100) != 0 ?
5182 	       PFERR_GUEST_FINAL_MASK : PFERR_GUEST_PAGE_MASK;
5183 
5184 	vcpu->arch.exit_qualification = exit_qualification;
5185 	return kvm_mmu_page_fault(vcpu, gpa, error_code, NULL, 0);
5186 }
5187 
5188 static int handle_ept_misconfig(struct kvm_vcpu *vcpu)
5189 {
5190 	gpa_t gpa;
5191 
5192 	/*
5193 	 * A nested guest cannot optimize MMIO vmexits, because we have an
5194 	 * nGPA here instead of the required GPA.
5195 	 */
5196 	gpa = vmcs_read64(GUEST_PHYSICAL_ADDRESS);
5197 	if (!is_guest_mode(vcpu) &&
5198 	    !kvm_io_bus_write(vcpu, KVM_FAST_MMIO_BUS, gpa, 0, NULL)) {
5199 		trace_kvm_fast_mmio(gpa);
5200 		return kvm_skip_emulated_instruction(vcpu);
5201 	}
5202 
5203 	return kvm_mmu_page_fault(vcpu, gpa, PFERR_RSVD_MASK, NULL, 0);
5204 }
5205 
5206 static int handle_nmi_window(struct kvm_vcpu *vcpu)
5207 {
5208 	WARN_ON_ONCE(!enable_vnmi);
5209 	exec_controls_clearbit(to_vmx(vcpu), CPU_BASED_VIRTUAL_NMI_PENDING);
5210 	++vcpu->stat.nmi_window_exits;
5211 	kvm_make_request(KVM_REQ_EVENT, vcpu);
5212 
5213 	return 1;
5214 }
5215 
5216 static int handle_invalid_guest_state(struct kvm_vcpu *vcpu)
5217 {
5218 	struct vcpu_vmx *vmx = to_vmx(vcpu);
5219 	bool intr_window_requested;
5220 	unsigned count = 130;
5221 
5222 	/*
5223 	 * We should never reach the point where we are emulating L2
5224 	 * due to invalid guest state as that means we incorrectly
5225 	 * allowed a nested VMEntry with an invalid vmcs12.
5226 	 */
5227 	WARN_ON_ONCE(vmx->emulation_required && vmx->nested.nested_run_pending);
5228 
5229 	intr_window_requested = exec_controls_get(vmx) &
5230 				CPU_BASED_VIRTUAL_INTR_PENDING;
5231 
5232 	while (vmx->emulation_required && count-- != 0) {
5233 		if (intr_window_requested && vmx_interrupt_allowed(vcpu))
5234 			return handle_interrupt_window(&vmx->vcpu);
5235 
5236 		if (kvm_test_request(KVM_REQ_EVENT, vcpu))
5237 			return 1;
5238 
5239 		if (!kvm_emulate_instruction(vcpu, 0))
5240 			return 0;
5241 
5242 		if (vmx->emulation_required && !vmx->rmode.vm86_active &&
5243 		    vcpu->arch.exception.pending) {
5244 			vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
5245 			vcpu->run->internal.suberror =
5246 						KVM_INTERNAL_ERROR_EMULATION;
5247 			vcpu->run->internal.ndata = 0;
5248 			return 0;
5249 		}
5250 
5251 		if (vcpu->arch.halt_request) {
5252 			vcpu->arch.halt_request = 0;
5253 			return kvm_vcpu_halt(vcpu);
5254 		}
5255 
5256 		/*
5257 		 * Note, return 1 and not 0, vcpu_run() is responsible for
5258 		 * morphing the pending signal into the proper return code.
5259 		 */
5260 		if (signal_pending(current))
5261 			return 1;
5262 
5263 		if (need_resched())
5264 			schedule();
5265 	}
5266 
5267 	return 1;
5268 }
5269 
5270 static void grow_ple_window(struct kvm_vcpu *vcpu)
5271 {
5272 	struct vcpu_vmx *vmx = to_vmx(vcpu);
5273 	unsigned int old = vmx->ple_window;
5274 
5275 	vmx->ple_window = __grow_ple_window(old, ple_window,
5276 					    ple_window_grow,
5277 					    ple_window_max);
5278 
5279 	if (vmx->ple_window != old) {
5280 		vmx->ple_window_dirty = true;
5281 		trace_kvm_ple_window_update(vcpu->vcpu_id,
5282 					    vmx->ple_window, old);
5283 	}
5284 }
5285 
5286 static void shrink_ple_window(struct kvm_vcpu *vcpu)
5287 {
5288 	struct vcpu_vmx *vmx = to_vmx(vcpu);
5289 	unsigned int old = vmx->ple_window;
5290 
5291 	vmx->ple_window = __shrink_ple_window(old, ple_window,
5292 					      ple_window_shrink,
5293 					      ple_window);
5294 
5295 	if (vmx->ple_window != old) {
5296 		vmx->ple_window_dirty = true;
5297 		trace_kvm_ple_window_update(vcpu->vcpu_id,
5298 					    vmx->ple_window, old);
5299 	}
5300 }
5301 
5302 /*
5303  * Handler for POSTED_INTERRUPT_WAKEUP_VECTOR.
5304  */
5305 static void wakeup_handler(void)
5306 {
5307 	struct kvm_vcpu *vcpu;
5308 	int cpu = smp_processor_id();
5309 
5310 	spin_lock(&per_cpu(blocked_vcpu_on_cpu_lock, cpu));
5311 	list_for_each_entry(vcpu, &per_cpu(blocked_vcpu_on_cpu, cpu),
5312 			blocked_vcpu_list) {
5313 		struct pi_desc *pi_desc = vcpu_to_pi_desc(vcpu);
5314 
5315 		if (pi_test_on(pi_desc) == 1)
5316 			kvm_vcpu_kick(vcpu);
5317 	}
5318 	spin_unlock(&per_cpu(blocked_vcpu_on_cpu_lock, cpu));
5319 }
5320 
5321 static void vmx_enable_tdp(void)
5322 {
5323 	kvm_mmu_set_mask_ptes(VMX_EPT_READABLE_MASK,
5324 		enable_ept_ad_bits ? VMX_EPT_ACCESS_BIT : 0ull,
5325 		enable_ept_ad_bits ? VMX_EPT_DIRTY_BIT : 0ull,
5326 		0ull, VMX_EPT_EXECUTABLE_MASK,
5327 		cpu_has_vmx_ept_execute_only() ? 0ull : VMX_EPT_READABLE_MASK,
5328 		VMX_EPT_RWX_MASK, 0ull);
5329 
5330 	ept_set_mmio_spte_mask();
5331 	kvm_enable_tdp();
5332 }
5333 
5334 /*
5335  * Indicate a busy-waiting vcpu in spinlock. We do not enable the PAUSE
5336  * exiting, so only get here on cpu with PAUSE-Loop-Exiting.
5337  */
5338 static int handle_pause(struct kvm_vcpu *vcpu)
5339 {
5340 	if (!kvm_pause_in_guest(vcpu->kvm))
5341 		grow_ple_window(vcpu);
5342 
5343 	/*
5344 	 * Intel sdm vol3 ch-25.1.3 says: The "PAUSE-loop exiting"
5345 	 * VM-execution control is ignored if CPL > 0. OTOH, KVM
5346 	 * never set PAUSE_EXITING and just set PLE if supported,
5347 	 * so the vcpu must be CPL=0 if it gets a PAUSE exit.
5348 	 */
5349 	kvm_vcpu_on_spin(vcpu, true);
5350 	return kvm_skip_emulated_instruction(vcpu);
5351 }
5352 
5353 static int handle_nop(struct kvm_vcpu *vcpu)
5354 {
5355 	return kvm_skip_emulated_instruction(vcpu);
5356 }
5357 
5358 static int handle_mwait(struct kvm_vcpu *vcpu)
5359 {
5360 	printk_once(KERN_WARNING "kvm: MWAIT instruction emulated as NOP!\n");
5361 	return handle_nop(vcpu);
5362 }
5363 
5364 static int handle_invalid_op(struct kvm_vcpu *vcpu)
5365 {
5366 	kvm_queue_exception(vcpu, UD_VECTOR);
5367 	return 1;
5368 }
5369 
5370 static int handle_monitor_trap(struct kvm_vcpu *vcpu)
5371 {
5372 	return 1;
5373 }
5374 
5375 static int handle_monitor(struct kvm_vcpu *vcpu)
5376 {
5377 	printk_once(KERN_WARNING "kvm: MONITOR instruction emulated as NOP!\n");
5378 	return handle_nop(vcpu);
5379 }
5380 
5381 static int handle_invpcid(struct kvm_vcpu *vcpu)
5382 {
5383 	u32 vmx_instruction_info;
5384 	unsigned long type;
5385 	bool pcid_enabled;
5386 	gva_t gva;
5387 	struct x86_exception e;
5388 	unsigned i;
5389 	unsigned long roots_to_free = 0;
5390 	struct {
5391 		u64 pcid;
5392 		u64 gla;
5393 	} operand;
5394 
5395 	if (!guest_cpuid_has(vcpu, X86_FEATURE_INVPCID)) {
5396 		kvm_queue_exception(vcpu, UD_VECTOR);
5397 		return 1;
5398 	}
5399 
5400 	vmx_instruction_info = vmcs_read32(VMX_INSTRUCTION_INFO);
5401 	type = kvm_register_readl(vcpu, (vmx_instruction_info >> 28) & 0xf);
5402 
5403 	if (type > 3) {
5404 		kvm_inject_gp(vcpu, 0);
5405 		return 1;
5406 	}
5407 
5408 	/* According to the Intel instruction reference, the memory operand
5409 	 * is read even if it isn't needed (e.g., for type==all)
5410 	 */
5411 	if (get_vmx_mem_address(vcpu, vmcs_readl(EXIT_QUALIFICATION),
5412 				vmx_instruction_info, false,
5413 				sizeof(operand), &gva))
5414 		return 1;
5415 
5416 	if (kvm_read_guest_virt(vcpu, gva, &operand, sizeof(operand), &e)) {
5417 		kvm_inject_page_fault(vcpu, &e);
5418 		return 1;
5419 	}
5420 
5421 	if (operand.pcid >> 12 != 0) {
5422 		kvm_inject_gp(vcpu, 0);
5423 		return 1;
5424 	}
5425 
5426 	pcid_enabled = kvm_read_cr4_bits(vcpu, X86_CR4_PCIDE);
5427 
5428 	switch (type) {
5429 	case INVPCID_TYPE_INDIV_ADDR:
5430 		if ((!pcid_enabled && (operand.pcid != 0)) ||
5431 		    is_noncanonical_address(operand.gla, vcpu)) {
5432 			kvm_inject_gp(vcpu, 0);
5433 			return 1;
5434 		}
5435 		kvm_mmu_invpcid_gva(vcpu, operand.gla, operand.pcid);
5436 		return kvm_skip_emulated_instruction(vcpu);
5437 
5438 	case INVPCID_TYPE_SINGLE_CTXT:
5439 		if (!pcid_enabled && (operand.pcid != 0)) {
5440 			kvm_inject_gp(vcpu, 0);
5441 			return 1;
5442 		}
5443 
5444 		if (kvm_get_active_pcid(vcpu) == operand.pcid) {
5445 			kvm_mmu_sync_roots(vcpu);
5446 			kvm_make_request(KVM_REQ_TLB_FLUSH, vcpu);
5447 		}
5448 
5449 		for (i = 0; i < KVM_MMU_NUM_PREV_ROOTS; i++)
5450 			if (kvm_get_pcid(vcpu, vcpu->arch.mmu->prev_roots[i].cr3)
5451 			    == operand.pcid)
5452 				roots_to_free |= KVM_MMU_ROOT_PREVIOUS(i);
5453 
5454 		kvm_mmu_free_roots(vcpu, vcpu->arch.mmu, roots_to_free);
5455 		/*
5456 		 * If neither the current cr3 nor any of the prev_roots use the
5457 		 * given PCID, then nothing needs to be done here because a
5458 		 * resync will happen anyway before switching to any other CR3.
5459 		 */
5460 
5461 		return kvm_skip_emulated_instruction(vcpu);
5462 
5463 	case INVPCID_TYPE_ALL_NON_GLOBAL:
5464 		/*
5465 		 * Currently, KVM doesn't mark global entries in the shadow
5466 		 * page tables, so a non-global flush just degenerates to a
5467 		 * global flush. If needed, we could optimize this later by
5468 		 * keeping track of global entries in shadow page tables.
5469 		 */
5470 
5471 		/* fall-through */
5472 	case INVPCID_TYPE_ALL_INCL_GLOBAL:
5473 		kvm_mmu_unload(vcpu);
5474 		return kvm_skip_emulated_instruction(vcpu);
5475 
5476 	default:
5477 		BUG(); /* We have already checked above that type <= 3 */
5478 	}
5479 }
5480 
5481 static int handle_pml_full(struct kvm_vcpu *vcpu)
5482 {
5483 	unsigned long exit_qualification;
5484 
5485 	trace_kvm_pml_full(vcpu->vcpu_id);
5486 
5487 	exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
5488 
5489 	/*
5490 	 * PML buffer FULL happened while executing iret from NMI,
5491 	 * "blocked by NMI" bit has to be set before next VM entry.
5492 	 */
5493 	if (!(to_vmx(vcpu)->idt_vectoring_info & VECTORING_INFO_VALID_MASK) &&
5494 			enable_vnmi &&
5495 			(exit_qualification & INTR_INFO_UNBLOCK_NMI))
5496 		vmcs_set_bits(GUEST_INTERRUPTIBILITY_INFO,
5497 				GUEST_INTR_STATE_NMI);
5498 
5499 	/*
5500 	 * PML buffer already flushed at beginning of VMEXIT. Nothing to do
5501 	 * here.., and there's no userspace involvement needed for PML.
5502 	 */
5503 	return 1;
5504 }
5505 
5506 static int handle_preemption_timer(struct kvm_vcpu *vcpu)
5507 {
5508 	struct vcpu_vmx *vmx = to_vmx(vcpu);
5509 
5510 	if (!vmx->req_immediate_exit &&
5511 	    !unlikely(vmx->loaded_vmcs->hv_timer_soft_disabled))
5512 		kvm_lapic_expired_hv_timer(vcpu);
5513 
5514 	return 1;
5515 }
5516 
5517 /*
5518  * When nested=0, all VMX instruction VM Exits filter here.  The handlers
5519  * are overwritten by nested_vmx_setup() when nested=1.
5520  */
5521 static int handle_vmx_instruction(struct kvm_vcpu *vcpu)
5522 {
5523 	kvm_queue_exception(vcpu, UD_VECTOR);
5524 	return 1;
5525 }
5526 
5527 static int handle_encls(struct kvm_vcpu *vcpu)
5528 {
5529 	/*
5530 	 * SGX virtualization is not yet supported.  There is no software
5531 	 * enable bit for SGX, so we have to trap ENCLS and inject a #UD
5532 	 * to prevent the guest from executing ENCLS.
5533 	 */
5534 	kvm_queue_exception(vcpu, UD_VECTOR);
5535 	return 1;
5536 }
5537 
5538 /*
5539  * The exit handlers return 1 if the exit was handled fully and guest execution
5540  * may resume.  Otherwise they set the kvm_run parameter to indicate what needs
5541  * to be done to userspace and return 0.
5542  */
5543 static int (*kvm_vmx_exit_handlers[])(struct kvm_vcpu *vcpu) = {
5544 	[EXIT_REASON_EXCEPTION_NMI]           = handle_exception_nmi,
5545 	[EXIT_REASON_EXTERNAL_INTERRUPT]      = handle_external_interrupt,
5546 	[EXIT_REASON_TRIPLE_FAULT]            = handle_triple_fault,
5547 	[EXIT_REASON_NMI_WINDOW]	      = handle_nmi_window,
5548 	[EXIT_REASON_IO_INSTRUCTION]          = handle_io,
5549 	[EXIT_REASON_CR_ACCESS]               = handle_cr,
5550 	[EXIT_REASON_DR_ACCESS]               = handle_dr,
5551 	[EXIT_REASON_CPUID]                   = handle_cpuid,
5552 	[EXIT_REASON_MSR_READ]                = handle_rdmsr,
5553 	[EXIT_REASON_MSR_WRITE]               = handle_wrmsr,
5554 	[EXIT_REASON_PENDING_INTERRUPT]       = handle_interrupt_window,
5555 	[EXIT_REASON_HLT]                     = handle_halt,
5556 	[EXIT_REASON_INVD]		      = handle_invd,
5557 	[EXIT_REASON_INVLPG]		      = handle_invlpg,
5558 	[EXIT_REASON_RDPMC]                   = handle_rdpmc,
5559 	[EXIT_REASON_VMCALL]                  = handle_vmcall,
5560 	[EXIT_REASON_VMCLEAR]		      = handle_vmx_instruction,
5561 	[EXIT_REASON_VMLAUNCH]		      = handle_vmx_instruction,
5562 	[EXIT_REASON_VMPTRLD]		      = handle_vmx_instruction,
5563 	[EXIT_REASON_VMPTRST]		      = handle_vmx_instruction,
5564 	[EXIT_REASON_VMREAD]		      = handle_vmx_instruction,
5565 	[EXIT_REASON_VMRESUME]		      = handle_vmx_instruction,
5566 	[EXIT_REASON_VMWRITE]		      = handle_vmx_instruction,
5567 	[EXIT_REASON_VMOFF]		      = handle_vmx_instruction,
5568 	[EXIT_REASON_VMON]		      = handle_vmx_instruction,
5569 	[EXIT_REASON_TPR_BELOW_THRESHOLD]     = handle_tpr_below_threshold,
5570 	[EXIT_REASON_APIC_ACCESS]             = handle_apic_access,
5571 	[EXIT_REASON_APIC_WRITE]              = handle_apic_write,
5572 	[EXIT_REASON_EOI_INDUCED]             = handle_apic_eoi_induced,
5573 	[EXIT_REASON_WBINVD]                  = handle_wbinvd,
5574 	[EXIT_REASON_XSETBV]                  = handle_xsetbv,
5575 	[EXIT_REASON_TASK_SWITCH]             = handle_task_switch,
5576 	[EXIT_REASON_MCE_DURING_VMENTRY]      = handle_machine_check,
5577 	[EXIT_REASON_GDTR_IDTR]		      = handle_desc,
5578 	[EXIT_REASON_LDTR_TR]		      = handle_desc,
5579 	[EXIT_REASON_EPT_VIOLATION]	      = handle_ept_violation,
5580 	[EXIT_REASON_EPT_MISCONFIG]           = handle_ept_misconfig,
5581 	[EXIT_REASON_PAUSE_INSTRUCTION]       = handle_pause,
5582 	[EXIT_REASON_MWAIT_INSTRUCTION]	      = handle_mwait,
5583 	[EXIT_REASON_MONITOR_TRAP_FLAG]       = handle_monitor_trap,
5584 	[EXIT_REASON_MONITOR_INSTRUCTION]     = handle_monitor,
5585 	[EXIT_REASON_INVEPT]                  = handle_vmx_instruction,
5586 	[EXIT_REASON_INVVPID]                 = handle_vmx_instruction,
5587 	[EXIT_REASON_RDRAND]                  = handle_invalid_op,
5588 	[EXIT_REASON_RDSEED]                  = handle_invalid_op,
5589 	[EXIT_REASON_PML_FULL]		      = handle_pml_full,
5590 	[EXIT_REASON_INVPCID]                 = handle_invpcid,
5591 	[EXIT_REASON_VMFUNC]		      = handle_vmx_instruction,
5592 	[EXIT_REASON_PREEMPTION_TIMER]	      = handle_preemption_timer,
5593 	[EXIT_REASON_ENCLS]		      = handle_encls,
5594 };
5595 
5596 static const int kvm_vmx_max_exit_handlers =
5597 	ARRAY_SIZE(kvm_vmx_exit_handlers);
5598 
5599 static void vmx_get_exit_info(struct kvm_vcpu *vcpu, u64 *info1, u64 *info2)
5600 {
5601 	*info1 = vmcs_readl(EXIT_QUALIFICATION);
5602 	*info2 = vmcs_read32(VM_EXIT_INTR_INFO);
5603 }
5604 
5605 static void vmx_destroy_pml_buffer(struct vcpu_vmx *vmx)
5606 {
5607 	if (vmx->pml_pg) {
5608 		__free_page(vmx->pml_pg);
5609 		vmx->pml_pg = NULL;
5610 	}
5611 }
5612 
5613 static void vmx_flush_pml_buffer(struct kvm_vcpu *vcpu)
5614 {
5615 	struct vcpu_vmx *vmx = to_vmx(vcpu);
5616 	u64 *pml_buf;
5617 	u16 pml_idx;
5618 
5619 	pml_idx = vmcs_read16(GUEST_PML_INDEX);
5620 
5621 	/* Do nothing if PML buffer is empty */
5622 	if (pml_idx == (PML_ENTITY_NUM - 1))
5623 		return;
5624 
5625 	/* PML index always points to next available PML buffer entity */
5626 	if (pml_idx >= PML_ENTITY_NUM)
5627 		pml_idx = 0;
5628 	else
5629 		pml_idx++;
5630 
5631 	pml_buf = page_address(vmx->pml_pg);
5632 	for (; pml_idx < PML_ENTITY_NUM; pml_idx++) {
5633 		u64 gpa;
5634 
5635 		gpa = pml_buf[pml_idx];
5636 		WARN_ON(gpa & (PAGE_SIZE - 1));
5637 		kvm_vcpu_mark_page_dirty(vcpu, gpa >> PAGE_SHIFT);
5638 	}
5639 
5640 	/* reset PML index */
5641 	vmcs_write16(GUEST_PML_INDEX, PML_ENTITY_NUM - 1);
5642 }
5643 
5644 /*
5645  * Flush all vcpus' PML buffer and update logged GPAs to dirty_bitmap.
5646  * Called before reporting dirty_bitmap to userspace.
5647  */
5648 static void kvm_flush_pml_buffers(struct kvm *kvm)
5649 {
5650 	int i;
5651 	struct kvm_vcpu *vcpu;
5652 	/*
5653 	 * We only need to kick vcpu out of guest mode here, as PML buffer
5654 	 * is flushed at beginning of all VMEXITs, and it's obvious that only
5655 	 * vcpus running in guest are possible to have unflushed GPAs in PML
5656 	 * buffer.
5657 	 */
5658 	kvm_for_each_vcpu(i, vcpu, kvm)
5659 		kvm_vcpu_kick(vcpu);
5660 }
5661 
5662 static void vmx_dump_sel(char *name, uint32_t sel)
5663 {
5664 	pr_err("%s sel=0x%04x, attr=0x%05x, limit=0x%08x, base=0x%016lx\n",
5665 	       name, vmcs_read16(sel),
5666 	       vmcs_read32(sel + GUEST_ES_AR_BYTES - GUEST_ES_SELECTOR),
5667 	       vmcs_read32(sel + GUEST_ES_LIMIT - GUEST_ES_SELECTOR),
5668 	       vmcs_readl(sel + GUEST_ES_BASE - GUEST_ES_SELECTOR));
5669 }
5670 
5671 static void vmx_dump_dtsel(char *name, uint32_t limit)
5672 {
5673 	pr_err("%s                           limit=0x%08x, base=0x%016lx\n",
5674 	       name, vmcs_read32(limit),
5675 	       vmcs_readl(limit + GUEST_GDTR_BASE - GUEST_GDTR_LIMIT));
5676 }
5677 
5678 void dump_vmcs(void)
5679 {
5680 	u32 vmentry_ctl, vmexit_ctl;
5681 	u32 cpu_based_exec_ctrl, pin_based_exec_ctrl, secondary_exec_control;
5682 	unsigned long cr4;
5683 	u64 efer;
5684 	int i, n;
5685 
5686 	if (!dump_invalid_vmcs) {
5687 		pr_warn_ratelimited("set kvm_intel.dump_invalid_vmcs=1 to dump internal KVM state.\n");
5688 		return;
5689 	}
5690 
5691 	vmentry_ctl = vmcs_read32(VM_ENTRY_CONTROLS);
5692 	vmexit_ctl = vmcs_read32(VM_EXIT_CONTROLS);
5693 	cpu_based_exec_ctrl = vmcs_read32(CPU_BASED_VM_EXEC_CONTROL);
5694 	pin_based_exec_ctrl = vmcs_read32(PIN_BASED_VM_EXEC_CONTROL);
5695 	cr4 = vmcs_readl(GUEST_CR4);
5696 	efer = vmcs_read64(GUEST_IA32_EFER);
5697 	secondary_exec_control = 0;
5698 	if (cpu_has_secondary_exec_ctrls())
5699 		secondary_exec_control = vmcs_read32(SECONDARY_VM_EXEC_CONTROL);
5700 
5701 	pr_err("*** Guest State ***\n");
5702 	pr_err("CR0: actual=0x%016lx, shadow=0x%016lx, gh_mask=%016lx\n",
5703 	       vmcs_readl(GUEST_CR0), vmcs_readl(CR0_READ_SHADOW),
5704 	       vmcs_readl(CR0_GUEST_HOST_MASK));
5705 	pr_err("CR4: actual=0x%016lx, shadow=0x%016lx, gh_mask=%016lx\n",
5706 	       cr4, vmcs_readl(CR4_READ_SHADOW), vmcs_readl(CR4_GUEST_HOST_MASK));
5707 	pr_err("CR3 = 0x%016lx\n", vmcs_readl(GUEST_CR3));
5708 	if ((secondary_exec_control & SECONDARY_EXEC_ENABLE_EPT) &&
5709 	    (cr4 & X86_CR4_PAE) && !(efer & EFER_LMA))
5710 	{
5711 		pr_err("PDPTR0 = 0x%016llx  PDPTR1 = 0x%016llx\n",
5712 		       vmcs_read64(GUEST_PDPTR0), vmcs_read64(GUEST_PDPTR1));
5713 		pr_err("PDPTR2 = 0x%016llx  PDPTR3 = 0x%016llx\n",
5714 		       vmcs_read64(GUEST_PDPTR2), vmcs_read64(GUEST_PDPTR3));
5715 	}
5716 	pr_err("RSP = 0x%016lx  RIP = 0x%016lx\n",
5717 	       vmcs_readl(GUEST_RSP), vmcs_readl(GUEST_RIP));
5718 	pr_err("RFLAGS=0x%08lx         DR7 = 0x%016lx\n",
5719 	       vmcs_readl(GUEST_RFLAGS), vmcs_readl(GUEST_DR7));
5720 	pr_err("Sysenter RSP=%016lx CS:RIP=%04x:%016lx\n",
5721 	       vmcs_readl(GUEST_SYSENTER_ESP),
5722 	       vmcs_read32(GUEST_SYSENTER_CS), vmcs_readl(GUEST_SYSENTER_EIP));
5723 	vmx_dump_sel("CS:  ", GUEST_CS_SELECTOR);
5724 	vmx_dump_sel("DS:  ", GUEST_DS_SELECTOR);
5725 	vmx_dump_sel("SS:  ", GUEST_SS_SELECTOR);
5726 	vmx_dump_sel("ES:  ", GUEST_ES_SELECTOR);
5727 	vmx_dump_sel("FS:  ", GUEST_FS_SELECTOR);
5728 	vmx_dump_sel("GS:  ", GUEST_GS_SELECTOR);
5729 	vmx_dump_dtsel("GDTR:", GUEST_GDTR_LIMIT);
5730 	vmx_dump_sel("LDTR:", GUEST_LDTR_SELECTOR);
5731 	vmx_dump_dtsel("IDTR:", GUEST_IDTR_LIMIT);
5732 	vmx_dump_sel("TR:  ", GUEST_TR_SELECTOR);
5733 	if ((vmexit_ctl & (VM_EXIT_SAVE_IA32_PAT | VM_EXIT_SAVE_IA32_EFER)) ||
5734 	    (vmentry_ctl & (VM_ENTRY_LOAD_IA32_PAT | VM_ENTRY_LOAD_IA32_EFER)))
5735 		pr_err("EFER =     0x%016llx  PAT = 0x%016llx\n",
5736 		       efer, vmcs_read64(GUEST_IA32_PAT));
5737 	pr_err("DebugCtl = 0x%016llx  DebugExceptions = 0x%016lx\n",
5738 	       vmcs_read64(GUEST_IA32_DEBUGCTL),
5739 	       vmcs_readl(GUEST_PENDING_DBG_EXCEPTIONS));
5740 	if (cpu_has_load_perf_global_ctrl() &&
5741 	    vmentry_ctl & VM_ENTRY_LOAD_IA32_PERF_GLOBAL_CTRL)
5742 		pr_err("PerfGlobCtl = 0x%016llx\n",
5743 		       vmcs_read64(GUEST_IA32_PERF_GLOBAL_CTRL));
5744 	if (vmentry_ctl & VM_ENTRY_LOAD_BNDCFGS)
5745 		pr_err("BndCfgS = 0x%016llx\n", vmcs_read64(GUEST_BNDCFGS));
5746 	pr_err("Interruptibility = %08x  ActivityState = %08x\n",
5747 	       vmcs_read32(GUEST_INTERRUPTIBILITY_INFO),
5748 	       vmcs_read32(GUEST_ACTIVITY_STATE));
5749 	if (secondary_exec_control & SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY)
5750 		pr_err("InterruptStatus = %04x\n",
5751 		       vmcs_read16(GUEST_INTR_STATUS));
5752 
5753 	pr_err("*** Host State ***\n");
5754 	pr_err("RIP = 0x%016lx  RSP = 0x%016lx\n",
5755 	       vmcs_readl(HOST_RIP), vmcs_readl(HOST_RSP));
5756 	pr_err("CS=%04x SS=%04x DS=%04x ES=%04x FS=%04x GS=%04x TR=%04x\n",
5757 	       vmcs_read16(HOST_CS_SELECTOR), vmcs_read16(HOST_SS_SELECTOR),
5758 	       vmcs_read16(HOST_DS_SELECTOR), vmcs_read16(HOST_ES_SELECTOR),
5759 	       vmcs_read16(HOST_FS_SELECTOR), vmcs_read16(HOST_GS_SELECTOR),
5760 	       vmcs_read16(HOST_TR_SELECTOR));
5761 	pr_err("FSBase=%016lx GSBase=%016lx TRBase=%016lx\n",
5762 	       vmcs_readl(HOST_FS_BASE), vmcs_readl(HOST_GS_BASE),
5763 	       vmcs_readl(HOST_TR_BASE));
5764 	pr_err("GDTBase=%016lx IDTBase=%016lx\n",
5765 	       vmcs_readl(HOST_GDTR_BASE), vmcs_readl(HOST_IDTR_BASE));
5766 	pr_err("CR0=%016lx CR3=%016lx CR4=%016lx\n",
5767 	       vmcs_readl(HOST_CR0), vmcs_readl(HOST_CR3),
5768 	       vmcs_readl(HOST_CR4));
5769 	pr_err("Sysenter RSP=%016lx CS:RIP=%04x:%016lx\n",
5770 	       vmcs_readl(HOST_IA32_SYSENTER_ESP),
5771 	       vmcs_read32(HOST_IA32_SYSENTER_CS),
5772 	       vmcs_readl(HOST_IA32_SYSENTER_EIP));
5773 	if (vmexit_ctl & (VM_EXIT_LOAD_IA32_PAT | VM_EXIT_LOAD_IA32_EFER))
5774 		pr_err("EFER = 0x%016llx  PAT = 0x%016llx\n",
5775 		       vmcs_read64(HOST_IA32_EFER),
5776 		       vmcs_read64(HOST_IA32_PAT));
5777 	if (cpu_has_load_perf_global_ctrl() &&
5778 	    vmexit_ctl & VM_EXIT_LOAD_IA32_PERF_GLOBAL_CTRL)
5779 		pr_err("PerfGlobCtl = 0x%016llx\n",
5780 		       vmcs_read64(HOST_IA32_PERF_GLOBAL_CTRL));
5781 
5782 	pr_err("*** Control State ***\n");
5783 	pr_err("PinBased=%08x CPUBased=%08x SecondaryExec=%08x\n",
5784 	       pin_based_exec_ctrl, cpu_based_exec_ctrl, secondary_exec_control);
5785 	pr_err("EntryControls=%08x ExitControls=%08x\n", vmentry_ctl, vmexit_ctl);
5786 	pr_err("ExceptionBitmap=%08x PFECmask=%08x PFECmatch=%08x\n",
5787 	       vmcs_read32(EXCEPTION_BITMAP),
5788 	       vmcs_read32(PAGE_FAULT_ERROR_CODE_MASK),
5789 	       vmcs_read32(PAGE_FAULT_ERROR_CODE_MATCH));
5790 	pr_err("VMEntry: intr_info=%08x errcode=%08x ilen=%08x\n",
5791 	       vmcs_read32(VM_ENTRY_INTR_INFO_FIELD),
5792 	       vmcs_read32(VM_ENTRY_EXCEPTION_ERROR_CODE),
5793 	       vmcs_read32(VM_ENTRY_INSTRUCTION_LEN));
5794 	pr_err("VMExit: intr_info=%08x errcode=%08x ilen=%08x\n",
5795 	       vmcs_read32(VM_EXIT_INTR_INFO),
5796 	       vmcs_read32(VM_EXIT_INTR_ERROR_CODE),
5797 	       vmcs_read32(VM_EXIT_INSTRUCTION_LEN));
5798 	pr_err("        reason=%08x qualification=%016lx\n",
5799 	       vmcs_read32(VM_EXIT_REASON), vmcs_readl(EXIT_QUALIFICATION));
5800 	pr_err("IDTVectoring: info=%08x errcode=%08x\n",
5801 	       vmcs_read32(IDT_VECTORING_INFO_FIELD),
5802 	       vmcs_read32(IDT_VECTORING_ERROR_CODE));
5803 	pr_err("TSC Offset = 0x%016llx\n", vmcs_read64(TSC_OFFSET));
5804 	if (secondary_exec_control & SECONDARY_EXEC_TSC_SCALING)
5805 		pr_err("TSC Multiplier = 0x%016llx\n",
5806 		       vmcs_read64(TSC_MULTIPLIER));
5807 	if (cpu_based_exec_ctrl & CPU_BASED_TPR_SHADOW) {
5808 		if (secondary_exec_control & SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY) {
5809 			u16 status = vmcs_read16(GUEST_INTR_STATUS);
5810 			pr_err("SVI|RVI = %02x|%02x ", status >> 8, status & 0xff);
5811 		}
5812 		pr_cont("TPR Threshold = 0x%02x\n", vmcs_read32(TPR_THRESHOLD));
5813 		if (secondary_exec_control & SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES)
5814 			pr_err("APIC-access addr = 0x%016llx ", vmcs_read64(APIC_ACCESS_ADDR));
5815 		pr_cont("virt-APIC addr = 0x%016llx\n", vmcs_read64(VIRTUAL_APIC_PAGE_ADDR));
5816 	}
5817 	if (pin_based_exec_ctrl & PIN_BASED_POSTED_INTR)
5818 		pr_err("PostedIntrVec = 0x%02x\n", vmcs_read16(POSTED_INTR_NV));
5819 	if ((secondary_exec_control & SECONDARY_EXEC_ENABLE_EPT))
5820 		pr_err("EPT pointer = 0x%016llx\n", vmcs_read64(EPT_POINTER));
5821 	n = vmcs_read32(CR3_TARGET_COUNT);
5822 	for (i = 0; i + 1 < n; i += 4)
5823 		pr_err("CR3 target%u=%016lx target%u=%016lx\n",
5824 		       i, vmcs_readl(CR3_TARGET_VALUE0 + i * 2),
5825 		       i + 1, vmcs_readl(CR3_TARGET_VALUE0 + i * 2 + 2));
5826 	if (i < n)
5827 		pr_err("CR3 target%u=%016lx\n",
5828 		       i, vmcs_readl(CR3_TARGET_VALUE0 + i * 2));
5829 	if (secondary_exec_control & SECONDARY_EXEC_PAUSE_LOOP_EXITING)
5830 		pr_err("PLE Gap=%08x Window=%08x\n",
5831 		       vmcs_read32(PLE_GAP), vmcs_read32(PLE_WINDOW));
5832 	if (secondary_exec_control & SECONDARY_EXEC_ENABLE_VPID)
5833 		pr_err("Virtual processor ID = 0x%04x\n",
5834 		       vmcs_read16(VIRTUAL_PROCESSOR_ID));
5835 }
5836 
5837 /*
5838  * The guest has exited.  See if we can fix it or if we need userspace
5839  * assistance.
5840  */
5841 static int vmx_handle_exit(struct kvm_vcpu *vcpu)
5842 {
5843 	struct vcpu_vmx *vmx = to_vmx(vcpu);
5844 	u32 exit_reason = vmx->exit_reason;
5845 	u32 vectoring_info = vmx->idt_vectoring_info;
5846 
5847 	trace_kvm_exit(exit_reason, vcpu, KVM_ISA_VMX);
5848 
5849 	/*
5850 	 * Flush logged GPAs PML buffer, this will make dirty_bitmap more
5851 	 * updated. Another good is, in kvm_vm_ioctl_get_dirty_log, before
5852 	 * querying dirty_bitmap, we only need to kick all vcpus out of guest
5853 	 * mode as if vcpus is in root mode, the PML buffer must has been
5854 	 * flushed already.
5855 	 */
5856 	if (enable_pml)
5857 		vmx_flush_pml_buffer(vcpu);
5858 
5859 	/* If guest state is invalid, start emulating */
5860 	if (vmx->emulation_required)
5861 		return handle_invalid_guest_state(vcpu);
5862 
5863 	if (is_guest_mode(vcpu) && nested_vmx_exit_reflected(vcpu, exit_reason))
5864 		return nested_vmx_reflect_vmexit(vcpu, exit_reason);
5865 
5866 	if (exit_reason & VMX_EXIT_REASONS_FAILED_VMENTRY) {
5867 		dump_vmcs();
5868 		vcpu->run->exit_reason = KVM_EXIT_FAIL_ENTRY;
5869 		vcpu->run->fail_entry.hardware_entry_failure_reason
5870 			= exit_reason;
5871 		return 0;
5872 	}
5873 
5874 	if (unlikely(vmx->fail)) {
5875 		dump_vmcs();
5876 		vcpu->run->exit_reason = KVM_EXIT_FAIL_ENTRY;
5877 		vcpu->run->fail_entry.hardware_entry_failure_reason
5878 			= vmcs_read32(VM_INSTRUCTION_ERROR);
5879 		return 0;
5880 	}
5881 
5882 	/*
5883 	 * Note:
5884 	 * Do not try to fix EXIT_REASON_EPT_MISCONFIG if it caused by
5885 	 * delivery event since it indicates guest is accessing MMIO.
5886 	 * The vm-exit can be triggered again after return to guest that
5887 	 * will cause infinite loop.
5888 	 */
5889 	if ((vectoring_info & VECTORING_INFO_VALID_MASK) &&
5890 			(exit_reason != EXIT_REASON_EXCEPTION_NMI &&
5891 			exit_reason != EXIT_REASON_EPT_VIOLATION &&
5892 			exit_reason != EXIT_REASON_PML_FULL &&
5893 			exit_reason != EXIT_REASON_TASK_SWITCH)) {
5894 		vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
5895 		vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_DELIVERY_EV;
5896 		vcpu->run->internal.ndata = 3;
5897 		vcpu->run->internal.data[0] = vectoring_info;
5898 		vcpu->run->internal.data[1] = exit_reason;
5899 		vcpu->run->internal.data[2] = vcpu->arch.exit_qualification;
5900 		if (exit_reason == EXIT_REASON_EPT_MISCONFIG) {
5901 			vcpu->run->internal.ndata++;
5902 			vcpu->run->internal.data[3] =
5903 				vmcs_read64(GUEST_PHYSICAL_ADDRESS);
5904 		}
5905 		return 0;
5906 	}
5907 
5908 	if (unlikely(!enable_vnmi &&
5909 		     vmx->loaded_vmcs->soft_vnmi_blocked)) {
5910 		if (vmx_interrupt_allowed(vcpu)) {
5911 			vmx->loaded_vmcs->soft_vnmi_blocked = 0;
5912 		} else if (vmx->loaded_vmcs->vnmi_blocked_time > 1000000000LL &&
5913 			   vcpu->arch.nmi_pending) {
5914 			/*
5915 			 * This CPU don't support us in finding the end of an
5916 			 * NMI-blocked window if the guest runs with IRQs
5917 			 * disabled. So we pull the trigger after 1 s of
5918 			 * futile waiting, but inform the user about this.
5919 			 */
5920 			printk(KERN_WARNING "%s: Breaking out of NMI-blocked "
5921 			       "state on VCPU %d after 1 s timeout\n",
5922 			       __func__, vcpu->vcpu_id);
5923 			vmx->loaded_vmcs->soft_vnmi_blocked = 0;
5924 		}
5925 	}
5926 
5927 	if (exit_reason < kvm_vmx_max_exit_handlers
5928 	    && kvm_vmx_exit_handlers[exit_reason])
5929 		return kvm_vmx_exit_handlers[exit_reason](vcpu);
5930 	else {
5931 		vcpu_unimpl(vcpu, "vmx: unexpected exit reason 0x%x\n",
5932 				exit_reason);
5933 		dump_vmcs();
5934 		vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
5935 		vcpu->run->internal.suberror =
5936 			KVM_INTERNAL_ERROR_UNEXPECTED_EXIT_REASON;
5937 		vcpu->run->internal.ndata = 1;
5938 		vcpu->run->internal.data[0] = exit_reason;
5939 		return 0;
5940 	}
5941 }
5942 
5943 /*
5944  * Software based L1D cache flush which is used when microcode providing
5945  * the cache control MSR is not loaded.
5946  *
5947  * The L1D cache is 32 KiB on Nehalem and later microarchitectures, but to
5948  * flush it is required to read in 64 KiB because the replacement algorithm
5949  * is not exactly LRU. This could be sized at runtime via topology
5950  * information but as all relevant affected CPUs have 32KiB L1D cache size
5951  * there is no point in doing so.
5952  */
5953 static void vmx_l1d_flush(struct kvm_vcpu *vcpu)
5954 {
5955 	int size = PAGE_SIZE << L1D_CACHE_ORDER;
5956 
5957 	/*
5958 	 * This code is only executed when the the flush mode is 'cond' or
5959 	 * 'always'
5960 	 */
5961 	if (static_branch_likely(&vmx_l1d_flush_cond)) {
5962 		bool flush_l1d;
5963 
5964 		/*
5965 		 * Clear the per-vcpu flush bit, it gets set again
5966 		 * either from vcpu_run() or from one of the unsafe
5967 		 * VMEXIT handlers.
5968 		 */
5969 		flush_l1d = vcpu->arch.l1tf_flush_l1d;
5970 		vcpu->arch.l1tf_flush_l1d = false;
5971 
5972 		/*
5973 		 * Clear the per-cpu flush bit, it gets set again from
5974 		 * the interrupt handlers.
5975 		 */
5976 		flush_l1d |= kvm_get_cpu_l1tf_flush_l1d();
5977 		kvm_clear_cpu_l1tf_flush_l1d();
5978 
5979 		if (!flush_l1d)
5980 			return;
5981 	}
5982 
5983 	vcpu->stat.l1d_flush++;
5984 
5985 	if (static_cpu_has(X86_FEATURE_FLUSH_L1D)) {
5986 		wrmsrl(MSR_IA32_FLUSH_CMD, L1D_FLUSH);
5987 		return;
5988 	}
5989 
5990 	asm volatile(
5991 		/* First ensure the pages are in the TLB */
5992 		"xorl	%%eax, %%eax\n"
5993 		".Lpopulate_tlb:\n\t"
5994 		"movzbl	(%[flush_pages], %%" _ASM_AX "), %%ecx\n\t"
5995 		"addl	$4096, %%eax\n\t"
5996 		"cmpl	%%eax, %[size]\n\t"
5997 		"jne	.Lpopulate_tlb\n\t"
5998 		"xorl	%%eax, %%eax\n\t"
5999 		"cpuid\n\t"
6000 		/* Now fill the cache */
6001 		"xorl	%%eax, %%eax\n"
6002 		".Lfill_cache:\n"
6003 		"movzbl	(%[flush_pages], %%" _ASM_AX "), %%ecx\n\t"
6004 		"addl	$64, %%eax\n\t"
6005 		"cmpl	%%eax, %[size]\n\t"
6006 		"jne	.Lfill_cache\n\t"
6007 		"lfence\n"
6008 		:: [flush_pages] "r" (vmx_l1d_flush_pages),
6009 		    [size] "r" (size)
6010 		: "eax", "ebx", "ecx", "edx");
6011 }
6012 
6013 static void update_cr8_intercept(struct kvm_vcpu *vcpu, int tpr, int irr)
6014 {
6015 	struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
6016 
6017 	if (is_guest_mode(vcpu) &&
6018 		nested_cpu_has(vmcs12, CPU_BASED_TPR_SHADOW))
6019 		return;
6020 
6021 	if (irr == -1 || tpr < irr) {
6022 		vmcs_write32(TPR_THRESHOLD, 0);
6023 		return;
6024 	}
6025 
6026 	vmcs_write32(TPR_THRESHOLD, irr);
6027 }
6028 
6029 void vmx_set_virtual_apic_mode(struct kvm_vcpu *vcpu)
6030 {
6031 	struct vcpu_vmx *vmx = to_vmx(vcpu);
6032 	u32 sec_exec_control;
6033 
6034 	if (!lapic_in_kernel(vcpu))
6035 		return;
6036 
6037 	if (!flexpriority_enabled &&
6038 	    !cpu_has_vmx_virtualize_x2apic_mode())
6039 		return;
6040 
6041 	/* Postpone execution until vmcs01 is the current VMCS. */
6042 	if (is_guest_mode(vcpu)) {
6043 		vmx->nested.change_vmcs01_virtual_apic_mode = true;
6044 		return;
6045 	}
6046 
6047 	sec_exec_control = secondary_exec_controls_get(vmx);
6048 	sec_exec_control &= ~(SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES |
6049 			      SECONDARY_EXEC_VIRTUALIZE_X2APIC_MODE);
6050 
6051 	switch (kvm_get_apic_mode(vcpu)) {
6052 	case LAPIC_MODE_INVALID:
6053 		WARN_ONCE(true, "Invalid local APIC state");
6054 	case LAPIC_MODE_DISABLED:
6055 		break;
6056 	case LAPIC_MODE_XAPIC:
6057 		if (flexpriority_enabled) {
6058 			sec_exec_control |=
6059 				SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES;
6060 			vmx_flush_tlb(vcpu, true);
6061 		}
6062 		break;
6063 	case LAPIC_MODE_X2APIC:
6064 		if (cpu_has_vmx_virtualize_x2apic_mode())
6065 			sec_exec_control |=
6066 				SECONDARY_EXEC_VIRTUALIZE_X2APIC_MODE;
6067 		break;
6068 	}
6069 	secondary_exec_controls_set(vmx, sec_exec_control);
6070 
6071 	vmx_update_msr_bitmap(vcpu);
6072 }
6073 
6074 static void vmx_set_apic_access_page_addr(struct kvm_vcpu *vcpu, hpa_t hpa)
6075 {
6076 	if (!is_guest_mode(vcpu)) {
6077 		vmcs_write64(APIC_ACCESS_ADDR, hpa);
6078 		vmx_flush_tlb(vcpu, true);
6079 	}
6080 }
6081 
6082 static void vmx_hwapic_isr_update(struct kvm_vcpu *vcpu, int max_isr)
6083 {
6084 	u16 status;
6085 	u8 old;
6086 
6087 	if (max_isr == -1)
6088 		max_isr = 0;
6089 
6090 	status = vmcs_read16(GUEST_INTR_STATUS);
6091 	old = status >> 8;
6092 	if (max_isr != old) {
6093 		status &= 0xff;
6094 		status |= max_isr << 8;
6095 		vmcs_write16(GUEST_INTR_STATUS, status);
6096 	}
6097 }
6098 
6099 static void vmx_set_rvi(int vector)
6100 {
6101 	u16 status;
6102 	u8 old;
6103 
6104 	if (vector == -1)
6105 		vector = 0;
6106 
6107 	status = vmcs_read16(GUEST_INTR_STATUS);
6108 	old = (u8)status & 0xff;
6109 	if ((u8)vector != old) {
6110 		status &= ~0xff;
6111 		status |= (u8)vector;
6112 		vmcs_write16(GUEST_INTR_STATUS, status);
6113 	}
6114 }
6115 
6116 static void vmx_hwapic_irr_update(struct kvm_vcpu *vcpu, int max_irr)
6117 {
6118 	/*
6119 	 * When running L2, updating RVI is only relevant when
6120 	 * vmcs12 virtual-interrupt-delivery enabled.
6121 	 * However, it can be enabled only when L1 also
6122 	 * intercepts external-interrupts and in that case
6123 	 * we should not update vmcs02 RVI but instead intercept
6124 	 * interrupt. Therefore, do nothing when running L2.
6125 	 */
6126 	if (!is_guest_mode(vcpu))
6127 		vmx_set_rvi(max_irr);
6128 }
6129 
6130 static int vmx_sync_pir_to_irr(struct kvm_vcpu *vcpu)
6131 {
6132 	struct vcpu_vmx *vmx = to_vmx(vcpu);
6133 	int max_irr;
6134 	bool max_irr_updated;
6135 
6136 	WARN_ON(!vcpu->arch.apicv_active);
6137 	if (pi_test_on(&vmx->pi_desc)) {
6138 		pi_clear_on(&vmx->pi_desc);
6139 		/*
6140 		 * IOMMU can write to PIR.ON, so the barrier matters even on UP.
6141 		 * But on x86 this is just a compiler barrier anyway.
6142 		 */
6143 		smp_mb__after_atomic();
6144 		max_irr_updated =
6145 			kvm_apic_update_irr(vcpu, vmx->pi_desc.pir, &max_irr);
6146 
6147 		/*
6148 		 * If we are running L2 and L1 has a new pending interrupt
6149 		 * which can be injected, we should re-evaluate
6150 		 * what should be done with this new L1 interrupt.
6151 		 * If L1 intercepts external-interrupts, we should
6152 		 * exit from L2 to L1. Otherwise, interrupt should be
6153 		 * delivered directly to L2.
6154 		 */
6155 		if (is_guest_mode(vcpu) && max_irr_updated) {
6156 			if (nested_exit_on_intr(vcpu))
6157 				kvm_vcpu_exiting_guest_mode(vcpu);
6158 			else
6159 				kvm_make_request(KVM_REQ_EVENT, vcpu);
6160 		}
6161 	} else {
6162 		max_irr = kvm_lapic_find_highest_irr(vcpu);
6163 	}
6164 	vmx_hwapic_irr_update(vcpu, max_irr);
6165 	return max_irr;
6166 }
6167 
6168 static bool vmx_dy_apicv_has_pending_interrupt(struct kvm_vcpu *vcpu)
6169 {
6170 	return pi_test_on(vcpu_to_pi_desc(vcpu));
6171 }
6172 
6173 static void vmx_load_eoi_exitmap(struct kvm_vcpu *vcpu, u64 *eoi_exit_bitmap)
6174 {
6175 	if (!kvm_vcpu_apicv_active(vcpu))
6176 		return;
6177 
6178 	vmcs_write64(EOI_EXIT_BITMAP0, eoi_exit_bitmap[0]);
6179 	vmcs_write64(EOI_EXIT_BITMAP1, eoi_exit_bitmap[1]);
6180 	vmcs_write64(EOI_EXIT_BITMAP2, eoi_exit_bitmap[2]);
6181 	vmcs_write64(EOI_EXIT_BITMAP3, eoi_exit_bitmap[3]);
6182 }
6183 
6184 static void vmx_apicv_post_state_restore(struct kvm_vcpu *vcpu)
6185 {
6186 	struct vcpu_vmx *vmx = to_vmx(vcpu);
6187 
6188 	pi_clear_on(&vmx->pi_desc);
6189 	memset(vmx->pi_desc.pir, 0, sizeof(vmx->pi_desc.pir));
6190 }
6191 
6192 static void handle_exception_nmi_irqoff(struct vcpu_vmx *vmx)
6193 {
6194 	vmx->exit_intr_info = vmcs_read32(VM_EXIT_INTR_INFO);
6195 
6196 	/* if exit due to PF check for async PF */
6197 	if (is_page_fault(vmx->exit_intr_info))
6198 		vmx->vcpu.arch.apf.host_apf_reason = kvm_read_and_reset_pf_reason();
6199 
6200 	/* Handle machine checks before interrupts are enabled */
6201 	if (is_machine_check(vmx->exit_intr_info))
6202 		kvm_machine_check();
6203 
6204 	/* We need to handle NMIs before interrupts are enabled */
6205 	if (is_nmi(vmx->exit_intr_info)) {
6206 		kvm_before_interrupt(&vmx->vcpu);
6207 		asm("int $2");
6208 		kvm_after_interrupt(&vmx->vcpu);
6209 	}
6210 }
6211 
6212 static void handle_external_interrupt_irqoff(struct kvm_vcpu *vcpu)
6213 {
6214 	unsigned int vector;
6215 	unsigned long entry;
6216 #ifdef CONFIG_X86_64
6217 	unsigned long tmp;
6218 #endif
6219 	gate_desc *desc;
6220 	u32 intr_info;
6221 
6222 	intr_info = vmcs_read32(VM_EXIT_INTR_INFO);
6223 	if (WARN_ONCE(!is_external_intr(intr_info),
6224 	    "KVM: unexpected VM-Exit interrupt info: 0x%x", intr_info))
6225 		return;
6226 
6227 	vector = intr_info & INTR_INFO_VECTOR_MASK;
6228 	desc = (gate_desc *)host_idt_base + vector;
6229 	entry = gate_offset(desc);
6230 
6231 	kvm_before_interrupt(vcpu);
6232 
6233 	asm volatile(
6234 #ifdef CONFIG_X86_64
6235 		"mov %%" _ASM_SP ", %[sp]\n\t"
6236 		"and $0xfffffffffffffff0, %%" _ASM_SP "\n\t"
6237 		"push $%c[ss]\n\t"
6238 		"push %[sp]\n\t"
6239 #endif
6240 		"pushf\n\t"
6241 		__ASM_SIZE(push) " $%c[cs]\n\t"
6242 		CALL_NOSPEC
6243 		:
6244 #ifdef CONFIG_X86_64
6245 		[sp]"=&r"(tmp),
6246 #endif
6247 		ASM_CALL_CONSTRAINT
6248 		:
6249 		THUNK_TARGET(entry),
6250 		[ss]"i"(__KERNEL_DS),
6251 		[cs]"i"(__KERNEL_CS)
6252 	);
6253 
6254 	kvm_after_interrupt(vcpu);
6255 }
6256 STACK_FRAME_NON_STANDARD(handle_external_interrupt_irqoff);
6257 
6258 static void vmx_handle_exit_irqoff(struct kvm_vcpu *vcpu)
6259 {
6260 	struct vcpu_vmx *vmx = to_vmx(vcpu);
6261 
6262 	if (vmx->exit_reason == EXIT_REASON_EXTERNAL_INTERRUPT)
6263 		handle_external_interrupt_irqoff(vcpu);
6264 	else if (vmx->exit_reason == EXIT_REASON_EXCEPTION_NMI)
6265 		handle_exception_nmi_irqoff(vmx);
6266 }
6267 
6268 static bool vmx_has_emulated_msr(int index)
6269 {
6270 	switch (index) {
6271 	case MSR_IA32_SMBASE:
6272 		/*
6273 		 * We cannot do SMM unless we can run the guest in big
6274 		 * real mode.
6275 		 */
6276 		return enable_unrestricted_guest || emulate_invalid_guest_state;
6277 	case MSR_IA32_VMX_BASIC ... MSR_IA32_VMX_VMFUNC:
6278 		return nested;
6279 	case MSR_AMD64_VIRT_SPEC_CTRL:
6280 		/* This is AMD only.  */
6281 		return false;
6282 	default:
6283 		return true;
6284 	}
6285 }
6286 
6287 static bool vmx_pt_supported(void)
6288 {
6289 	return pt_mode == PT_MODE_HOST_GUEST;
6290 }
6291 
6292 static void vmx_recover_nmi_blocking(struct vcpu_vmx *vmx)
6293 {
6294 	u32 exit_intr_info;
6295 	bool unblock_nmi;
6296 	u8 vector;
6297 	bool idtv_info_valid;
6298 
6299 	idtv_info_valid = vmx->idt_vectoring_info & VECTORING_INFO_VALID_MASK;
6300 
6301 	if (enable_vnmi) {
6302 		if (vmx->loaded_vmcs->nmi_known_unmasked)
6303 			return;
6304 		/*
6305 		 * Can't use vmx->exit_intr_info since we're not sure what
6306 		 * the exit reason is.
6307 		 */
6308 		exit_intr_info = vmcs_read32(VM_EXIT_INTR_INFO);
6309 		unblock_nmi = (exit_intr_info & INTR_INFO_UNBLOCK_NMI) != 0;
6310 		vector = exit_intr_info & INTR_INFO_VECTOR_MASK;
6311 		/*
6312 		 * SDM 3: 27.7.1.2 (September 2008)
6313 		 * Re-set bit "block by NMI" before VM entry if vmexit caused by
6314 		 * a guest IRET fault.
6315 		 * SDM 3: 23.2.2 (September 2008)
6316 		 * Bit 12 is undefined in any of the following cases:
6317 		 *  If the VM exit sets the valid bit in the IDT-vectoring
6318 		 *   information field.
6319 		 *  If the VM exit is due to a double fault.
6320 		 */
6321 		if ((exit_intr_info & INTR_INFO_VALID_MASK) && unblock_nmi &&
6322 		    vector != DF_VECTOR && !idtv_info_valid)
6323 			vmcs_set_bits(GUEST_INTERRUPTIBILITY_INFO,
6324 				      GUEST_INTR_STATE_NMI);
6325 		else
6326 			vmx->loaded_vmcs->nmi_known_unmasked =
6327 				!(vmcs_read32(GUEST_INTERRUPTIBILITY_INFO)
6328 				  & GUEST_INTR_STATE_NMI);
6329 	} else if (unlikely(vmx->loaded_vmcs->soft_vnmi_blocked))
6330 		vmx->loaded_vmcs->vnmi_blocked_time +=
6331 			ktime_to_ns(ktime_sub(ktime_get(),
6332 					      vmx->loaded_vmcs->entry_time));
6333 }
6334 
6335 static void __vmx_complete_interrupts(struct kvm_vcpu *vcpu,
6336 				      u32 idt_vectoring_info,
6337 				      int instr_len_field,
6338 				      int error_code_field)
6339 {
6340 	u8 vector;
6341 	int type;
6342 	bool idtv_info_valid;
6343 
6344 	idtv_info_valid = idt_vectoring_info & VECTORING_INFO_VALID_MASK;
6345 
6346 	vcpu->arch.nmi_injected = false;
6347 	kvm_clear_exception_queue(vcpu);
6348 	kvm_clear_interrupt_queue(vcpu);
6349 
6350 	if (!idtv_info_valid)
6351 		return;
6352 
6353 	kvm_make_request(KVM_REQ_EVENT, vcpu);
6354 
6355 	vector = idt_vectoring_info & VECTORING_INFO_VECTOR_MASK;
6356 	type = idt_vectoring_info & VECTORING_INFO_TYPE_MASK;
6357 
6358 	switch (type) {
6359 	case INTR_TYPE_NMI_INTR:
6360 		vcpu->arch.nmi_injected = true;
6361 		/*
6362 		 * SDM 3: 27.7.1.2 (September 2008)
6363 		 * Clear bit "block by NMI" before VM entry if a NMI
6364 		 * delivery faulted.
6365 		 */
6366 		vmx_set_nmi_mask(vcpu, false);
6367 		break;
6368 	case INTR_TYPE_SOFT_EXCEPTION:
6369 		vcpu->arch.event_exit_inst_len = vmcs_read32(instr_len_field);
6370 		/* fall through */
6371 	case INTR_TYPE_HARD_EXCEPTION:
6372 		if (idt_vectoring_info & VECTORING_INFO_DELIVER_CODE_MASK) {
6373 			u32 err = vmcs_read32(error_code_field);
6374 			kvm_requeue_exception_e(vcpu, vector, err);
6375 		} else
6376 			kvm_requeue_exception(vcpu, vector);
6377 		break;
6378 	case INTR_TYPE_SOFT_INTR:
6379 		vcpu->arch.event_exit_inst_len = vmcs_read32(instr_len_field);
6380 		/* fall through */
6381 	case INTR_TYPE_EXT_INTR:
6382 		kvm_queue_interrupt(vcpu, vector, type == INTR_TYPE_SOFT_INTR);
6383 		break;
6384 	default:
6385 		break;
6386 	}
6387 }
6388 
6389 static void vmx_complete_interrupts(struct vcpu_vmx *vmx)
6390 {
6391 	__vmx_complete_interrupts(&vmx->vcpu, vmx->idt_vectoring_info,
6392 				  VM_EXIT_INSTRUCTION_LEN,
6393 				  IDT_VECTORING_ERROR_CODE);
6394 }
6395 
6396 static void vmx_cancel_injection(struct kvm_vcpu *vcpu)
6397 {
6398 	__vmx_complete_interrupts(vcpu,
6399 				  vmcs_read32(VM_ENTRY_INTR_INFO_FIELD),
6400 				  VM_ENTRY_INSTRUCTION_LEN,
6401 				  VM_ENTRY_EXCEPTION_ERROR_CODE);
6402 
6403 	vmcs_write32(VM_ENTRY_INTR_INFO_FIELD, 0);
6404 }
6405 
6406 static void atomic_switch_perf_msrs(struct vcpu_vmx *vmx)
6407 {
6408 	int i, nr_msrs;
6409 	struct perf_guest_switch_msr *msrs;
6410 
6411 	msrs = perf_guest_get_msrs(&nr_msrs);
6412 
6413 	if (!msrs)
6414 		return;
6415 
6416 	for (i = 0; i < nr_msrs; i++)
6417 		if (msrs[i].host == msrs[i].guest)
6418 			clear_atomic_switch_msr(vmx, msrs[i].msr);
6419 		else
6420 			add_atomic_switch_msr(vmx, msrs[i].msr, msrs[i].guest,
6421 					msrs[i].host, false);
6422 }
6423 
6424 static void atomic_switch_umwait_control_msr(struct vcpu_vmx *vmx)
6425 {
6426 	u32 host_umwait_control;
6427 
6428 	if (!vmx_has_waitpkg(vmx))
6429 		return;
6430 
6431 	host_umwait_control = get_umwait_control_msr();
6432 
6433 	if (vmx->msr_ia32_umwait_control != host_umwait_control)
6434 		add_atomic_switch_msr(vmx, MSR_IA32_UMWAIT_CONTROL,
6435 			vmx->msr_ia32_umwait_control,
6436 			host_umwait_control, false);
6437 	else
6438 		clear_atomic_switch_msr(vmx, MSR_IA32_UMWAIT_CONTROL);
6439 }
6440 
6441 static void vmx_update_hv_timer(struct kvm_vcpu *vcpu)
6442 {
6443 	struct vcpu_vmx *vmx = to_vmx(vcpu);
6444 	u64 tscl;
6445 	u32 delta_tsc;
6446 
6447 	if (vmx->req_immediate_exit) {
6448 		vmcs_write32(VMX_PREEMPTION_TIMER_VALUE, 0);
6449 		vmx->loaded_vmcs->hv_timer_soft_disabled = false;
6450 	} else if (vmx->hv_deadline_tsc != -1) {
6451 		tscl = rdtsc();
6452 		if (vmx->hv_deadline_tsc > tscl)
6453 			/* set_hv_timer ensures the delta fits in 32-bits */
6454 			delta_tsc = (u32)((vmx->hv_deadline_tsc - tscl) >>
6455 				cpu_preemption_timer_multi);
6456 		else
6457 			delta_tsc = 0;
6458 
6459 		vmcs_write32(VMX_PREEMPTION_TIMER_VALUE, delta_tsc);
6460 		vmx->loaded_vmcs->hv_timer_soft_disabled = false;
6461 	} else if (!vmx->loaded_vmcs->hv_timer_soft_disabled) {
6462 		vmcs_write32(VMX_PREEMPTION_TIMER_VALUE, -1);
6463 		vmx->loaded_vmcs->hv_timer_soft_disabled = true;
6464 	}
6465 }
6466 
6467 void vmx_update_host_rsp(struct vcpu_vmx *vmx, unsigned long host_rsp)
6468 {
6469 	if (unlikely(host_rsp != vmx->loaded_vmcs->host_state.rsp)) {
6470 		vmx->loaded_vmcs->host_state.rsp = host_rsp;
6471 		vmcs_writel(HOST_RSP, host_rsp);
6472 	}
6473 }
6474 
6475 bool __vmx_vcpu_run(struct vcpu_vmx *vmx, unsigned long *regs, bool launched);
6476 
6477 static void vmx_vcpu_run(struct kvm_vcpu *vcpu)
6478 {
6479 	struct vcpu_vmx *vmx = to_vmx(vcpu);
6480 	unsigned long cr3, cr4;
6481 
6482 	/* Record the guest's net vcpu time for enforced NMI injections. */
6483 	if (unlikely(!enable_vnmi &&
6484 		     vmx->loaded_vmcs->soft_vnmi_blocked))
6485 		vmx->loaded_vmcs->entry_time = ktime_get();
6486 
6487 	/* Don't enter VMX if guest state is invalid, let the exit handler
6488 	   start emulation until we arrive back to a valid state */
6489 	if (vmx->emulation_required)
6490 		return;
6491 
6492 	if (vmx->ple_window_dirty) {
6493 		vmx->ple_window_dirty = false;
6494 		vmcs_write32(PLE_WINDOW, vmx->ple_window);
6495 	}
6496 
6497 	if (vmx->nested.need_vmcs12_to_shadow_sync)
6498 		nested_sync_vmcs12_to_shadow(vcpu);
6499 
6500 	if (test_bit(VCPU_REGS_RSP, (unsigned long *)&vcpu->arch.regs_dirty))
6501 		vmcs_writel(GUEST_RSP, vcpu->arch.regs[VCPU_REGS_RSP]);
6502 	if (test_bit(VCPU_REGS_RIP, (unsigned long *)&vcpu->arch.regs_dirty))
6503 		vmcs_writel(GUEST_RIP, vcpu->arch.regs[VCPU_REGS_RIP]);
6504 
6505 	cr3 = __get_current_cr3_fast();
6506 	if (unlikely(cr3 != vmx->loaded_vmcs->host_state.cr3)) {
6507 		vmcs_writel(HOST_CR3, cr3);
6508 		vmx->loaded_vmcs->host_state.cr3 = cr3;
6509 	}
6510 
6511 	cr4 = cr4_read_shadow();
6512 	if (unlikely(cr4 != vmx->loaded_vmcs->host_state.cr4)) {
6513 		vmcs_writel(HOST_CR4, cr4);
6514 		vmx->loaded_vmcs->host_state.cr4 = cr4;
6515 	}
6516 
6517 	/* When single-stepping over STI and MOV SS, we must clear the
6518 	 * corresponding interruptibility bits in the guest state. Otherwise
6519 	 * vmentry fails as it then expects bit 14 (BS) in pending debug
6520 	 * exceptions being set, but that's not correct for the guest debugging
6521 	 * case. */
6522 	if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)
6523 		vmx_set_interrupt_shadow(vcpu, 0);
6524 
6525 	kvm_load_guest_xcr0(vcpu);
6526 
6527 	if (static_cpu_has(X86_FEATURE_PKU) &&
6528 	    kvm_read_cr4_bits(vcpu, X86_CR4_PKE) &&
6529 	    vcpu->arch.pkru != vmx->host_pkru)
6530 		__write_pkru(vcpu->arch.pkru);
6531 
6532 	pt_guest_enter(vmx);
6533 
6534 	atomic_switch_perf_msrs(vmx);
6535 	atomic_switch_umwait_control_msr(vmx);
6536 
6537 	if (enable_preemption_timer)
6538 		vmx_update_hv_timer(vcpu);
6539 
6540 	if (lapic_in_kernel(vcpu) &&
6541 		vcpu->arch.apic->lapic_timer.timer_advance_ns)
6542 		kvm_wait_lapic_expire(vcpu);
6543 
6544 	/*
6545 	 * If this vCPU has touched SPEC_CTRL, restore the guest's value if
6546 	 * it's non-zero. Since vmentry is serialising on affected CPUs, there
6547 	 * is no need to worry about the conditional branch over the wrmsr
6548 	 * being speculatively taken.
6549 	 */
6550 	x86_spec_ctrl_set_guest(vmx->spec_ctrl, 0);
6551 
6552 	/* L1D Flush includes CPU buffer clear to mitigate MDS */
6553 	if (static_branch_unlikely(&vmx_l1d_should_flush))
6554 		vmx_l1d_flush(vcpu);
6555 	else if (static_branch_unlikely(&mds_user_clear))
6556 		mds_clear_cpu_buffers();
6557 
6558 	if (vcpu->arch.cr2 != read_cr2())
6559 		write_cr2(vcpu->arch.cr2);
6560 
6561 	vmx->fail = __vmx_vcpu_run(vmx, (unsigned long *)&vcpu->arch.regs,
6562 				   vmx->loaded_vmcs->launched);
6563 
6564 	vcpu->arch.cr2 = read_cr2();
6565 
6566 	/*
6567 	 * We do not use IBRS in the kernel. If this vCPU has used the
6568 	 * SPEC_CTRL MSR it may have left it on; save the value and
6569 	 * turn it off. This is much more efficient than blindly adding
6570 	 * it to the atomic save/restore list. Especially as the former
6571 	 * (Saving guest MSRs on vmexit) doesn't even exist in KVM.
6572 	 *
6573 	 * For non-nested case:
6574 	 * If the L01 MSR bitmap does not intercept the MSR, then we need to
6575 	 * save it.
6576 	 *
6577 	 * For nested case:
6578 	 * If the L02 MSR bitmap does not intercept the MSR, then we need to
6579 	 * save it.
6580 	 */
6581 	if (unlikely(!msr_write_intercepted(vcpu, MSR_IA32_SPEC_CTRL)))
6582 		vmx->spec_ctrl = native_read_msr(MSR_IA32_SPEC_CTRL);
6583 
6584 	x86_spec_ctrl_restore_host(vmx->spec_ctrl, 0);
6585 
6586 	/* All fields are clean at this point */
6587 	if (static_branch_unlikely(&enable_evmcs))
6588 		current_evmcs->hv_clean_fields |=
6589 			HV_VMX_ENLIGHTENED_CLEAN_FIELD_ALL;
6590 
6591 	if (static_branch_unlikely(&enable_evmcs))
6592 		current_evmcs->hv_vp_id = vcpu->arch.hyperv.vp_index;
6593 
6594 	/* MSR_IA32_DEBUGCTLMSR is zeroed on vmexit. Restore it if needed */
6595 	if (vmx->host_debugctlmsr)
6596 		update_debugctlmsr(vmx->host_debugctlmsr);
6597 
6598 #ifndef CONFIG_X86_64
6599 	/*
6600 	 * The sysexit path does not restore ds/es, so we must set them to
6601 	 * a reasonable value ourselves.
6602 	 *
6603 	 * We can't defer this to vmx_prepare_switch_to_host() since that
6604 	 * function may be executed in interrupt context, which saves and
6605 	 * restore segments around it, nullifying its effect.
6606 	 */
6607 	loadsegment(ds, __USER_DS);
6608 	loadsegment(es, __USER_DS);
6609 #endif
6610 
6611 	vcpu->arch.regs_avail = ~((1 << VCPU_REGS_RIP) | (1 << VCPU_REGS_RSP)
6612 				  | (1 << VCPU_EXREG_RFLAGS)
6613 				  | (1 << VCPU_EXREG_PDPTR)
6614 				  | (1 << VCPU_EXREG_SEGMENTS)
6615 				  | (1 << VCPU_EXREG_CR3));
6616 	vcpu->arch.regs_dirty = 0;
6617 
6618 	pt_guest_exit(vmx);
6619 
6620 	/*
6621 	 * eager fpu is enabled if PKEY is supported and CR4 is switched
6622 	 * back on host, so it is safe to read guest PKRU from current
6623 	 * XSAVE.
6624 	 */
6625 	if (static_cpu_has(X86_FEATURE_PKU) &&
6626 	    kvm_read_cr4_bits(vcpu, X86_CR4_PKE)) {
6627 		vcpu->arch.pkru = rdpkru();
6628 		if (vcpu->arch.pkru != vmx->host_pkru)
6629 			__write_pkru(vmx->host_pkru);
6630 	}
6631 
6632 	kvm_put_guest_xcr0(vcpu);
6633 
6634 	vmx->nested.nested_run_pending = 0;
6635 	vmx->idt_vectoring_info = 0;
6636 
6637 	vmx->exit_reason = vmx->fail ? 0xdead : vmcs_read32(VM_EXIT_REASON);
6638 	if ((u16)vmx->exit_reason == EXIT_REASON_MCE_DURING_VMENTRY)
6639 		kvm_machine_check();
6640 
6641 	if (vmx->fail || (vmx->exit_reason & VMX_EXIT_REASONS_FAILED_VMENTRY))
6642 		return;
6643 
6644 	vmx->loaded_vmcs->launched = 1;
6645 	vmx->idt_vectoring_info = vmcs_read32(IDT_VECTORING_INFO_FIELD);
6646 
6647 	vmx_recover_nmi_blocking(vmx);
6648 	vmx_complete_interrupts(vmx);
6649 }
6650 
6651 static struct kvm *vmx_vm_alloc(void)
6652 {
6653 	struct kvm_vmx *kvm_vmx = __vmalloc(sizeof(struct kvm_vmx),
6654 					    GFP_KERNEL_ACCOUNT | __GFP_ZERO,
6655 					    PAGE_KERNEL);
6656 	return &kvm_vmx->kvm;
6657 }
6658 
6659 static void vmx_vm_free(struct kvm *kvm)
6660 {
6661 	kfree(kvm->arch.hyperv.hv_pa_pg);
6662 	vfree(to_kvm_vmx(kvm));
6663 }
6664 
6665 static void vmx_free_vcpu(struct kvm_vcpu *vcpu)
6666 {
6667 	struct vcpu_vmx *vmx = to_vmx(vcpu);
6668 
6669 	if (enable_pml)
6670 		vmx_destroy_pml_buffer(vmx);
6671 	free_vpid(vmx->vpid);
6672 	nested_vmx_free_vcpu(vcpu);
6673 	free_loaded_vmcs(vmx->loaded_vmcs);
6674 	kfree(vmx->guest_msrs);
6675 	kvm_vcpu_uninit(vcpu);
6676 	kmem_cache_free(x86_fpu_cache, vmx->vcpu.arch.user_fpu);
6677 	kmem_cache_free(x86_fpu_cache, vmx->vcpu.arch.guest_fpu);
6678 	kmem_cache_free(kvm_vcpu_cache, vmx);
6679 }
6680 
6681 static struct kvm_vcpu *vmx_create_vcpu(struct kvm *kvm, unsigned int id)
6682 {
6683 	int err;
6684 	struct vcpu_vmx *vmx;
6685 	unsigned long *msr_bitmap;
6686 	int cpu;
6687 
6688 	BUILD_BUG_ON_MSG(offsetof(struct vcpu_vmx, vcpu) != 0,
6689 		"struct kvm_vcpu must be at offset 0 for arch usercopy region");
6690 
6691 	vmx = kmem_cache_zalloc(kvm_vcpu_cache, GFP_KERNEL_ACCOUNT);
6692 	if (!vmx)
6693 		return ERR_PTR(-ENOMEM);
6694 
6695 	vmx->vcpu.arch.user_fpu = kmem_cache_zalloc(x86_fpu_cache,
6696 			GFP_KERNEL_ACCOUNT);
6697 	if (!vmx->vcpu.arch.user_fpu) {
6698 		printk(KERN_ERR "kvm: failed to allocate kvm userspace's fpu\n");
6699 		err = -ENOMEM;
6700 		goto free_partial_vcpu;
6701 	}
6702 
6703 	vmx->vcpu.arch.guest_fpu = kmem_cache_zalloc(x86_fpu_cache,
6704 			GFP_KERNEL_ACCOUNT);
6705 	if (!vmx->vcpu.arch.guest_fpu) {
6706 		printk(KERN_ERR "kvm: failed to allocate vcpu's fpu\n");
6707 		err = -ENOMEM;
6708 		goto free_user_fpu;
6709 	}
6710 
6711 	vmx->vpid = allocate_vpid();
6712 
6713 	err = kvm_vcpu_init(&vmx->vcpu, kvm, id);
6714 	if (err)
6715 		goto free_vcpu;
6716 
6717 	err = -ENOMEM;
6718 
6719 	/*
6720 	 * If PML is turned on, failure on enabling PML just results in failure
6721 	 * of creating the vcpu, therefore we can simplify PML logic (by
6722 	 * avoiding dealing with cases, such as enabling PML partially on vcpus
6723 	 * for the guest, etc.
6724 	 */
6725 	if (enable_pml) {
6726 		vmx->pml_pg = alloc_page(GFP_KERNEL_ACCOUNT | __GFP_ZERO);
6727 		if (!vmx->pml_pg)
6728 			goto uninit_vcpu;
6729 	}
6730 
6731 	vmx->guest_msrs = kmalloc(PAGE_SIZE, GFP_KERNEL_ACCOUNT);
6732 	BUILD_BUG_ON(ARRAY_SIZE(vmx_msr_index) * sizeof(vmx->guest_msrs[0])
6733 		     > PAGE_SIZE);
6734 
6735 	if (!vmx->guest_msrs)
6736 		goto free_pml;
6737 
6738 	err = alloc_loaded_vmcs(&vmx->vmcs01);
6739 	if (err < 0)
6740 		goto free_msrs;
6741 
6742 	msr_bitmap = vmx->vmcs01.msr_bitmap;
6743 	vmx_disable_intercept_for_msr(msr_bitmap, MSR_IA32_TSC, MSR_TYPE_R);
6744 	vmx_disable_intercept_for_msr(msr_bitmap, MSR_FS_BASE, MSR_TYPE_RW);
6745 	vmx_disable_intercept_for_msr(msr_bitmap, MSR_GS_BASE, MSR_TYPE_RW);
6746 	vmx_disable_intercept_for_msr(msr_bitmap, MSR_KERNEL_GS_BASE, MSR_TYPE_RW);
6747 	vmx_disable_intercept_for_msr(msr_bitmap, MSR_IA32_SYSENTER_CS, MSR_TYPE_RW);
6748 	vmx_disable_intercept_for_msr(msr_bitmap, MSR_IA32_SYSENTER_ESP, MSR_TYPE_RW);
6749 	vmx_disable_intercept_for_msr(msr_bitmap, MSR_IA32_SYSENTER_EIP, MSR_TYPE_RW);
6750 	if (kvm_cstate_in_guest(kvm)) {
6751 		vmx_disable_intercept_for_msr(msr_bitmap, MSR_CORE_C1_RES, MSR_TYPE_R);
6752 		vmx_disable_intercept_for_msr(msr_bitmap, MSR_CORE_C3_RESIDENCY, MSR_TYPE_R);
6753 		vmx_disable_intercept_for_msr(msr_bitmap, MSR_CORE_C6_RESIDENCY, MSR_TYPE_R);
6754 		vmx_disable_intercept_for_msr(msr_bitmap, MSR_CORE_C7_RESIDENCY, MSR_TYPE_R);
6755 	}
6756 	vmx->msr_bitmap_mode = 0;
6757 
6758 	vmx->loaded_vmcs = &vmx->vmcs01;
6759 	cpu = get_cpu();
6760 	vmx_vcpu_load(&vmx->vcpu, cpu);
6761 	vmx->vcpu.cpu = cpu;
6762 	vmx_vcpu_setup(vmx);
6763 	vmx_vcpu_put(&vmx->vcpu);
6764 	put_cpu();
6765 	if (cpu_need_virtualize_apic_accesses(&vmx->vcpu)) {
6766 		err = alloc_apic_access_page(kvm);
6767 		if (err)
6768 			goto free_vmcs;
6769 	}
6770 
6771 	if (enable_ept && !enable_unrestricted_guest) {
6772 		err = init_rmode_identity_map(kvm);
6773 		if (err)
6774 			goto free_vmcs;
6775 	}
6776 
6777 	if (nested)
6778 		nested_vmx_setup_ctls_msrs(&vmx->nested.msrs,
6779 					   vmx_capability.ept,
6780 					   kvm_vcpu_apicv_active(&vmx->vcpu));
6781 	else
6782 		memset(&vmx->nested.msrs, 0, sizeof(vmx->nested.msrs));
6783 
6784 	vmx->nested.posted_intr_nv = -1;
6785 	vmx->nested.current_vmptr = -1ull;
6786 
6787 	vmx->msr_ia32_feature_control_valid_bits = FEATURE_CONTROL_LOCKED;
6788 
6789 	/*
6790 	 * Enforce invariant: pi_desc.nv is always either POSTED_INTR_VECTOR
6791 	 * or POSTED_INTR_WAKEUP_VECTOR.
6792 	 */
6793 	vmx->pi_desc.nv = POSTED_INTR_VECTOR;
6794 	vmx->pi_desc.sn = 1;
6795 
6796 	vmx->ept_pointer = INVALID_PAGE;
6797 
6798 	return &vmx->vcpu;
6799 
6800 free_vmcs:
6801 	free_loaded_vmcs(vmx->loaded_vmcs);
6802 free_msrs:
6803 	kfree(vmx->guest_msrs);
6804 free_pml:
6805 	vmx_destroy_pml_buffer(vmx);
6806 uninit_vcpu:
6807 	kvm_vcpu_uninit(&vmx->vcpu);
6808 free_vcpu:
6809 	free_vpid(vmx->vpid);
6810 	kmem_cache_free(x86_fpu_cache, vmx->vcpu.arch.guest_fpu);
6811 free_user_fpu:
6812 	kmem_cache_free(x86_fpu_cache, vmx->vcpu.arch.user_fpu);
6813 free_partial_vcpu:
6814 	kmem_cache_free(kvm_vcpu_cache, vmx);
6815 	return ERR_PTR(err);
6816 }
6817 
6818 #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"
6819 #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"
6820 
6821 static int vmx_vm_init(struct kvm *kvm)
6822 {
6823 	spin_lock_init(&to_kvm_vmx(kvm)->ept_pointer_lock);
6824 
6825 	if (!ple_gap)
6826 		kvm->arch.pause_in_guest = true;
6827 
6828 	if (boot_cpu_has(X86_BUG_L1TF) && enable_ept) {
6829 		switch (l1tf_mitigation) {
6830 		case L1TF_MITIGATION_OFF:
6831 		case L1TF_MITIGATION_FLUSH_NOWARN:
6832 			/* 'I explicitly don't care' is set */
6833 			break;
6834 		case L1TF_MITIGATION_FLUSH:
6835 		case L1TF_MITIGATION_FLUSH_NOSMT:
6836 		case L1TF_MITIGATION_FULL:
6837 			/*
6838 			 * Warn upon starting the first VM in a potentially
6839 			 * insecure environment.
6840 			 */
6841 			if (sched_smt_active())
6842 				pr_warn_once(L1TF_MSG_SMT);
6843 			if (l1tf_vmx_mitigation == VMENTER_L1D_FLUSH_NEVER)
6844 				pr_warn_once(L1TF_MSG_L1D);
6845 			break;
6846 		case L1TF_MITIGATION_FULL_FORCE:
6847 			/* Flush is enforced */
6848 			break;
6849 		}
6850 	}
6851 	return 0;
6852 }
6853 
6854 static int __init vmx_check_processor_compat(void)
6855 {
6856 	struct vmcs_config vmcs_conf;
6857 	struct vmx_capability vmx_cap;
6858 
6859 	if (setup_vmcs_config(&vmcs_conf, &vmx_cap) < 0)
6860 		return -EIO;
6861 	if (nested)
6862 		nested_vmx_setup_ctls_msrs(&vmcs_conf.nested, vmx_cap.ept,
6863 					   enable_apicv);
6864 	if (memcmp(&vmcs_config, &vmcs_conf, sizeof(struct vmcs_config)) != 0) {
6865 		printk(KERN_ERR "kvm: CPU %d feature inconsistency!\n",
6866 				smp_processor_id());
6867 		return -EIO;
6868 	}
6869 	return 0;
6870 }
6871 
6872 static u64 vmx_get_mt_mask(struct kvm_vcpu *vcpu, gfn_t gfn, bool is_mmio)
6873 {
6874 	u8 cache;
6875 	u64 ipat = 0;
6876 
6877 	/* For VT-d and EPT combination
6878 	 * 1. MMIO: always map as UC
6879 	 * 2. EPT with VT-d:
6880 	 *   a. VT-d without snooping control feature: can't guarantee the
6881 	 *	result, try to trust guest.
6882 	 *   b. VT-d with snooping control feature: snooping control feature of
6883 	 *	VT-d engine can guarantee the cache correctness. Just set it
6884 	 *	to WB to keep consistent with host. So the same as item 3.
6885 	 * 3. EPT without VT-d: always map as WB and set IPAT=1 to keep
6886 	 *    consistent with host MTRR
6887 	 */
6888 	if (is_mmio) {
6889 		cache = MTRR_TYPE_UNCACHABLE;
6890 		goto exit;
6891 	}
6892 
6893 	if (!kvm_arch_has_noncoherent_dma(vcpu->kvm)) {
6894 		ipat = VMX_EPT_IPAT_BIT;
6895 		cache = MTRR_TYPE_WRBACK;
6896 		goto exit;
6897 	}
6898 
6899 	if (kvm_read_cr0(vcpu) & X86_CR0_CD) {
6900 		ipat = VMX_EPT_IPAT_BIT;
6901 		if (kvm_check_has_quirk(vcpu->kvm, KVM_X86_QUIRK_CD_NW_CLEARED))
6902 			cache = MTRR_TYPE_WRBACK;
6903 		else
6904 			cache = MTRR_TYPE_UNCACHABLE;
6905 		goto exit;
6906 	}
6907 
6908 	cache = kvm_mtrr_get_guest_memory_type(vcpu, gfn);
6909 
6910 exit:
6911 	return (cache << VMX_EPT_MT_EPTE_SHIFT) | ipat;
6912 }
6913 
6914 static int vmx_get_lpage_level(void)
6915 {
6916 	if (enable_ept && !cpu_has_vmx_ept_1g_page())
6917 		return PT_DIRECTORY_LEVEL;
6918 	else
6919 		/* For shadow and EPT supported 1GB page */
6920 		return PT_PDPE_LEVEL;
6921 }
6922 
6923 static void vmcs_set_secondary_exec_control(struct vcpu_vmx *vmx)
6924 {
6925 	/*
6926 	 * These bits in the secondary execution controls field
6927 	 * are dynamic, the others are mostly based on the hypervisor
6928 	 * architecture and the guest's CPUID.  Do not touch the
6929 	 * dynamic bits.
6930 	 */
6931 	u32 mask =
6932 		SECONDARY_EXEC_SHADOW_VMCS |
6933 		SECONDARY_EXEC_VIRTUALIZE_X2APIC_MODE |
6934 		SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES |
6935 		SECONDARY_EXEC_DESC;
6936 
6937 	u32 new_ctl = vmx->secondary_exec_control;
6938 	u32 cur_ctl = secondary_exec_controls_get(vmx);
6939 
6940 	secondary_exec_controls_set(vmx, (new_ctl & ~mask) | (cur_ctl & mask));
6941 }
6942 
6943 /*
6944  * Generate MSR_IA32_VMX_CR{0,4}_FIXED1 according to CPUID. Only set bits
6945  * (indicating "allowed-1") if they are supported in the guest's CPUID.
6946  */
6947 static void nested_vmx_cr_fixed1_bits_update(struct kvm_vcpu *vcpu)
6948 {
6949 	struct vcpu_vmx *vmx = to_vmx(vcpu);
6950 	struct kvm_cpuid_entry2 *entry;
6951 
6952 	vmx->nested.msrs.cr0_fixed1 = 0xffffffff;
6953 	vmx->nested.msrs.cr4_fixed1 = X86_CR4_PCE;
6954 
6955 #define cr4_fixed1_update(_cr4_mask, _reg, _cpuid_mask) do {		\
6956 	if (entry && (entry->_reg & (_cpuid_mask)))			\
6957 		vmx->nested.msrs.cr4_fixed1 |= (_cr4_mask);	\
6958 } while (0)
6959 
6960 	entry = kvm_find_cpuid_entry(vcpu, 0x1, 0);
6961 	cr4_fixed1_update(X86_CR4_VME,        edx, bit(X86_FEATURE_VME));
6962 	cr4_fixed1_update(X86_CR4_PVI,        edx, bit(X86_FEATURE_VME));
6963 	cr4_fixed1_update(X86_CR4_TSD,        edx, bit(X86_FEATURE_TSC));
6964 	cr4_fixed1_update(X86_CR4_DE,         edx, bit(X86_FEATURE_DE));
6965 	cr4_fixed1_update(X86_CR4_PSE,        edx, bit(X86_FEATURE_PSE));
6966 	cr4_fixed1_update(X86_CR4_PAE,        edx, bit(X86_FEATURE_PAE));
6967 	cr4_fixed1_update(X86_CR4_MCE,        edx, bit(X86_FEATURE_MCE));
6968 	cr4_fixed1_update(X86_CR4_PGE,        edx, bit(X86_FEATURE_PGE));
6969 	cr4_fixed1_update(X86_CR4_OSFXSR,     edx, bit(X86_FEATURE_FXSR));
6970 	cr4_fixed1_update(X86_CR4_OSXMMEXCPT, edx, bit(X86_FEATURE_XMM));
6971 	cr4_fixed1_update(X86_CR4_VMXE,       ecx, bit(X86_FEATURE_VMX));
6972 	cr4_fixed1_update(X86_CR4_SMXE,       ecx, bit(X86_FEATURE_SMX));
6973 	cr4_fixed1_update(X86_CR4_PCIDE,      ecx, bit(X86_FEATURE_PCID));
6974 	cr4_fixed1_update(X86_CR4_OSXSAVE,    ecx, bit(X86_FEATURE_XSAVE));
6975 
6976 	entry = kvm_find_cpuid_entry(vcpu, 0x7, 0);
6977 	cr4_fixed1_update(X86_CR4_FSGSBASE,   ebx, bit(X86_FEATURE_FSGSBASE));
6978 	cr4_fixed1_update(X86_CR4_SMEP,       ebx, bit(X86_FEATURE_SMEP));
6979 	cr4_fixed1_update(X86_CR4_SMAP,       ebx, bit(X86_FEATURE_SMAP));
6980 	cr4_fixed1_update(X86_CR4_PKE,        ecx, bit(X86_FEATURE_PKU));
6981 	cr4_fixed1_update(X86_CR4_UMIP,       ecx, bit(X86_FEATURE_UMIP));
6982 
6983 #undef cr4_fixed1_update
6984 }
6985 
6986 static void nested_vmx_entry_exit_ctls_update(struct kvm_vcpu *vcpu)
6987 {
6988 	struct vcpu_vmx *vmx = to_vmx(vcpu);
6989 
6990 	if (kvm_mpx_supported()) {
6991 		bool mpx_enabled = guest_cpuid_has(vcpu, X86_FEATURE_MPX);
6992 
6993 		if (mpx_enabled) {
6994 			vmx->nested.msrs.entry_ctls_high |= VM_ENTRY_LOAD_BNDCFGS;
6995 			vmx->nested.msrs.exit_ctls_high |= VM_EXIT_CLEAR_BNDCFGS;
6996 		} else {
6997 			vmx->nested.msrs.entry_ctls_high &= ~VM_ENTRY_LOAD_BNDCFGS;
6998 			vmx->nested.msrs.exit_ctls_high &= ~VM_EXIT_CLEAR_BNDCFGS;
6999 		}
7000 	}
7001 }
7002 
7003 static void update_intel_pt_cfg(struct kvm_vcpu *vcpu)
7004 {
7005 	struct vcpu_vmx *vmx = to_vmx(vcpu);
7006 	struct kvm_cpuid_entry2 *best = NULL;
7007 	int i;
7008 
7009 	for (i = 0; i < PT_CPUID_LEAVES; i++) {
7010 		best = kvm_find_cpuid_entry(vcpu, 0x14, i);
7011 		if (!best)
7012 			return;
7013 		vmx->pt_desc.caps[CPUID_EAX + i*PT_CPUID_REGS_NUM] = best->eax;
7014 		vmx->pt_desc.caps[CPUID_EBX + i*PT_CPUID_REGS_NUM] = best->ebx;
7015 		vmx->pt_desc.caps[CPUID_ECX + i*PT_CPUID_REGS_NUM] = best->ecx;
7016 		vmx->pt_desc.caps[CPUID_EDX + i*PT_CPUID_REGS_NUM] = best->edx;
7017 	}
7018 
7019 	/* Get the number of configurable Address Ranges for filtering */
7020 	vmx->pt_desc.addr_range = intel_pt_validate_cap(vmx->pt_desc.caps,
7021 						PT_CAP_num_address_ranges);
7022 
7023 	/* Initialize and clear the no dependency bits */
7024 	vmx->pt_desc.ctl_bitmask = ~(RTIT_CTL_TRACEEN | RTIT_CTL_OS |
7025 			RTIT_CTL_USR | RTIT_CTL_TSC_EN | RTIT_CTL_DISRETC);
7026 
7027 	/*
7028 	 * If CPUID.(EAX=14H,ECX=0):EBX[0]=1 CR3Filter can be set otherwise
7029 	 * will inject an #GP
7030 	 */
7031 	if (intel_pt_validate_cap(vmx->pt_desc.caps, PT_CAP_cr3_filtering))
7032 		vmx->pt_desc.ctl_bitmask &= ~RTIT_CTL_CR3EN;
7033 
7034 	/*
7035 	 * If CPUID.(EAX=14H,ECX=0):EBX[1]=1 CYCEn, CycThresh and
7036 	 * PSBFreq can be set
7037 	 */
7038 	if (intel_pt_validate_cap(vmx->pt_desc.caps, PT_CAP_psb_cyc))
7039 		vmx->pt_desc.ctl_bitmask &= ~(RTIT_CTL_CYCLEACC |
7040 				RTIT_CTL_CYC_THRESH | RTIT_CTL_PSB_FREQ);
7041 
7042 	/*
7043 	 * If CPUID.(EAX=14H,ECX=0):EBX[3]=1 MTCEn BranchEn and
7044 	 * MTCFreq can be set
7045 	 */
7046 	if (intel_pt_validate_cap(vmx->pt_desc.caps, PT_CAP_mtc))
7047 		vmx->pt_desc.ctl_bitmask &= ~(RTIT_CTL_MTC_EN |
7048 				RTIT_CTL_BRANCH_EN | RTIT_CTL_MTC_RANGE);
7049 
7050 	/* If CPUID.(EAX=14H,ECX=0):EBX[4]=1 FUPonPTW and PTWEn can be set */
7051 	if (intel_pt_validate_cap(vmx->pt_desc.caps, PT_CAP_ptwrite))
7052 		vmx->pt_desc.ctl_bitmask &= ~(RTIT_CTL_FUP_ON_PTW |
7053 							RTIT_CTL_PTW_EN);
7054 
7055 	/* If CPUID.(EAX=14H,ECX=0):EBX[5]=1 PwrEvEn can be set */
7056 	if (intel_pt_validate_cap(vmx->pt_desc.caps, PT_CAP_power_event_trace))
7057 		vmx->pt_desc.ctl_bitmask &= ~RTIT_CTL_PWR_EVT_EN;
7058 
7059 	/* If CPUID.(EAX=14H,ECX=0):ECX[0]=1 ToPA can be set */
7060 	if (intel_pt_validate_cap(vmx->pt_desc.caps, PT_CAP_topa_output))
7061 		vmx->pt_desc.ctl_bitmask &= ~RTIT_CTL_TOPA;
7062 
7063 	/* If CPUID.(EAX=14H,ECX=0):ECX[3]=1 FabircEn can be set */
7064 	if (intel_pt_validate_cap(vmx->pt_desc.caps, PT_CAP_output_subsys))
7065 		vmx->pt_desc.ctl_bitmask &= ~RTIT_CTL_FABRIC_EN;
7066 
7067 	/* unmask address range configure area */
7068 	for (i = 0; i < vmx->pt_desc.addr_range; i++)
7069 		vmx->pt_desc.ctl_bitmask &= ~(0xfULL << (32 + i * 4));
7070 }
7071 
7072 static void vmx_cpuid_update(struct kvm_vcpu *vcpu)
7073 {
7074 	struct vcpu_vmx *vmx = to_vmx(vcpu);
7075 
7076 	if (cpu_has_secondary_exec_ctrls()) {
7077 		vmx_compute_secondary_exec_control(vmx);
7078 		vmcs_set_secondary_exec_control(vmx);
7079 	}
7080 
7081 	if (nested_vmx_allowed(vcpu))
7082 		to_vmx(vcpu)->msr_ia32_feature_control_valid_bits |=
7083 			FEATURE_CONTROL_VMXON_ENABLED_OUTSIDE_SMX;
7084 	else
7085 		to_vmx(vcpu)->msr_ia32_feature_control_valid_bits &=
7086 			~FEATURE_CONTROL_VMXON_ENABLED_OUTSIDE_SMX;
7087 
7088 	if (nested_vmx_allowed(vcpu)) {
7089 		nested_vmx_cr_fixed1_bits_update(vcpu);
7090 		nested_vmx_entry_exit_ctls_update(vcpu);
7091 	}
7092 
7093 	if (boot_cpu_has(X86_FEATURE_INTEL_PT) &&
7094 			guest_cpuid_has(vcpu, X86_FEATURE_INTEL_PT))
7095 		update_intel_pt_cfg(vcpu);
7096 }
7097 
7098 static void vmx_set_supported_cpuid(u32 func, struct kvm_cpuid_entry2 *entry)
7099 {
7100 	if (func == 1 && nested)
7101 		entry->ecx |= bit(X86_FEATURE_VMX);
7102 }
7103 
7104 static void vmx_request_immediate_exit(struct kvm_vcpu *vcpu)
7105 {
7106 	to_vmx(vcpu)->req_immediate_exit = true;
7107 }
7108 
7109 static int vmx_check_intercept(struct kvm_vcpu *vcpu,
7110 			       struct x86_instruction_info *info,
7111 			       enum x86_intercept_stage stage)
7112 {
7113 	struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
7114 	struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
7115 
7116 	/*
7117 	 * RDPID causes #UD if disabled through secondary execution controls.
7118 	 * Because it is marked as EmulateOnUD, we need to intercept it here.
7119 	 */
7120 	if (info->intercept == x86_intercept_rdtscp &&
7121 	    !nested_cpu_has2(vmcs12, SECONDARY_EXEC_RDTSCP)) {
7122 		ctxt->exception.vector = UD_VECTOR;
7123 		ctxt->exception.error_code_valid = false;
7124 		return X86EMUL_PROPAGATE_FAULT;
7125 	}
7126 
7127 	/* TODO: check more intercepts... */
7128 	return X86EMUL_CONTINUE;
7129 }
7130 
7131 #ifdef CONFIG_X86_64
7132 /* (a << shift) / divisor, return 1 if overflow otherwise 0 */
7133 static inline int u64_shl_div_u64(u64 a, unsigned int shift,
7134 				  u64 divisor, u64 *result)
7135 {
7136 	u64 low = a << shift, high = a >> (64 - shift);
7137 
7138 	/* To avoid the overflow on divq */
7139 	if (high >= divisor)
7140 		return 1;
7141 
7142 	/* Low hold the result, high hold rem which is discarded */
7143 	asm("divq %2\n\t" : "=a" (low), "=d" (high) :
7144 	    "rm" (divisor), "0" (low), "1" (high));
7145 	*result = low;
7146 
7147 	return 0;
7148 }
7149 
7150 static int vmx_set_hv_timer(struct kvm_vcpu *vcpu, u64 guest_deadline_tsc,
7151 			    bool *expired)
7152 {
7153 	struct vcpu_vmx *vmx;
7154 	u64 tscl, guest_tscl, delta_tsc, lapic_timer_advance_cycles;
7155 	struct kvm_timer *ktimer = &vcpu->arch.apic->lapic_timer;
7156 
7157 	if (kvm_mwait_in_guest(vcpu->kvm) ||
7158 		kvm_can_post_timer_interrupt(vcpu))
7159 		return -EOPNOTSUPP;
7160 
7161 	vmx = to_vmx(vcpu);
7162 	tscl = rdtsc();
7163 	guest_tscl = kvm_read_l1_tsc(vcpu, tscl);
7164 	delta_tsc = max(guest_deadline_tsc, guest_tscl) - guest_tscl;
7165 	lapic_timer_advance_cycles = nsec_to_cycles(vcpu,
7166 						    ktimer->timer_advance_ns);
7167 
7168 	if (delta_tsc > lapic_timer_advance_cycles)
7169 		delta_tsc -= lapic_timer_advance_cycles;
7170 	else
7171 		delta_tsc = 0;
7172 
7173 	/* Convert to host delta tsc if tsc scaling is enabled */
7174 	if (vcpu->arch.tsc_scaling_ratio != kvm_default_tsc_scaling_ratio &&
7175 	    delta_tsc && u64_shl_div_u64(delta_tsc,
7176 				kvm_tsc_scaling_ratio_frac_bits,
7177 				vcpu->arch.tsc_scaling_ratio, &delta_tsc))
7178 		return -ERANGE;
7179 
7180 	/*
7181 	 * If the delta tsc can't fit in the 32 bit after the multi shift,
7182 	 * we can't use the preemption timer.
7183 	 * It's possible that it fits on later vmentries, but checking
7184 	 * on every vmentry is costly so we just use an hrtimer.
7185 	 */
7186 	if (delta_tsc >> (cpu_preemption_timer_multi + 32))
7187 		return -ERANGE;
7188 
7189 	vmx->hv_deadline_tsc = tscl + delta_tsc;
7190 	*expired = !delta_tsc;
7191 	return 0;
7192 }
7193 
7194 static void vmx_cancel_hv_timer(struct kvm_vcpu *vcpu)
7195 {
7196 	to_vmx(vcpu)->hv_deadline_tsc = -1;
7197 }
7198 #endif
7199 
7200 static void vmx_sched_in(struct kvm_vcpu *vcpu, int cpu)
7201 {
7202 	if (!kvm_pause_in_guest(vcpu->kvm))
7203 		shrink_ple_window(vcpu);
7204 }
7205 
7206 static void vmx_slot_enable_log_dirty(struct kvm *kvm,
7207 				     struct kvm_memory_slot *slot)
7208 {
7209 	kvm_mmu_slot_leaf_clear_dirty(kvm, slot);
7210 	kvm_mmu_slot_largepage_remove_write_access(kvm, slot);
7211 }
7212 
7213 static void vmx_slot_disable_log_dirty(struct kvm *kvm,
7214 				       struct kvm_memory_slot *slot)
7215 {
7216 	kvm_mmu_slot_set_dirty(kvm, slot);
7217 }
7218 
7219 static void vmx_flush_log_dirty(struct kvm *kvm)
7220 {
7221 	kvm_flush_pml_buffers(kvm);
7222 }
7223 
7224 static int vmx_write_pml_buffer(struct kvm_vcpu *vcpu)
7225 {
7226 	struct vmcs12 *vmcs12;
7227 	struct vcpu_vmx *vmx = to_vmx(vcpu);
7228 	gpa_t gpa, dst;
7229 
7230 	if (is_guest_mode(vcpu)) {
7231 		WARN_ON_ONCE(vmx->nested.pml_full);
7232 
7233 		/*
7234 		 * Check if PML is enabled for the nested guest.
7235 		 * Whether eptp bit 6 is set is already checked
7236 		 * as part of A/D emulation.
7237 		 */
7238 		vmcs12 = get_vmcs12(vcpu);
7239 		if (!nested_cpu_has_pml(vmcs12))
7240 			return 0;
7241 
7242 		if (vmcs12->guest_pml_index >= PML_ENTITY_NUM) {
7243 			vmx->nested.pml_full = true;
7244 			return 1;
7245 		}
7246 
7247 		gpa = vmcs_read64(GUEST_PHYSICAL_ADDRESS) & ~0xFFFull;
7248 		dst = vmcs12->pml_address + sizeof(u64) * vmcs12->guest_pml_index;
7249 
7250 		if (kvm_write_guest_page(vcpu->kvm, gpa_to_gfn(dst), &gpa,
7251 					 offset_in_page(dst), sizeof(gpa)))
7252 			return 0;
7253 
7254 		vmcs12->guest_pml_index--;
7255 	}
7256 
7257 	return 0;
7258 }
7259 
7260 static void vmx_enable_log_dirty_pt_masked(struct kvm *kvm,
7261 					   struct kvm_memory_slot *memslot,
7262 					   gfn_t offset, unsigned long mask)
7263 {
7264 	kvm_mmu_clear_dirty_pt_masked(kvm, memslot, offset, mask);
7265 }
7266 
7267 static void __pi_post_block(struct kvm_vcpu *vcpu)
7268 {
7269 	struct pi_desc *pi_desc = vcpu_to_pi_desc(vcpu);
7270 	struct pi_desc old, new;
7271 	unsigned int dest;
7272 
7273 	do {
7274 		old.control = new.control = pi_desc->control;
7275 		WARN(old.nv != POSTED_INTR_WAKEUP_VECTOR,
7276 		     "Wakeup handler not enabled while the VCPU is blocked\n");
7277 
7278 		dest = cpu_physical_id(vcpu->cpu);
7279 
7280 		if (x2apic_enabled())
7281 			new.ndst = dest;
7282 		else
7283 			new.ndst = (dest << 8) & 0xFF00;
7284 
7285 		/* set 'NV' to 'notification vector' */
7286 		new.nv = POSTED_INTR_VECTOR;
7287 	} while (cmpxchg64(&pi_desc->control, old.control,
7288 			   new.control) != old.control);
7289 
7290 	if (!WARN_ON_ONCE(vcpu->pre_pcpu == -1)) {
7291 		spin_lock(&per_cpu(blocked_vcpu_on_cpu_lock, vcpu->pre_pcpu));
7292 		list_del(&vcpu->blocked_vcpu_list);
7293 		spin_unlock(&per_cpu(blocked_vcpu_on_cpu_lock, vcpu->pre_pcpu));
7294 		vcpu->pre_pcpu = -1;
7295 	}
7296 }
7297 
7298 /*
7299  * This routine does the following things for vCPU which is going
7300  * to be blocked if VT-d PI is enabled.
7301  * - Store the vCPU to the wakeup list, so when interrupts happen
7302  *   we can find the right vCPU to wake up.
7303  * - Change the Posted-interrupt descriptor as below:
7304  *      'NDST' <-- vcpu->pre_pcpu
7305  *      'NV' <-- POSTED_INTR_WAKEUP_VECTOR
7306  * - If 'ON' is set during this process, which means at least one
7307  *   interrupt is posted for this vCPU, we cannot block it, in
7308  *   this case, return 1, otherwise, return 0.
7309  *
7310  */
7311 static int pi_pre_block(struct kvm_vcpu *vcpu)
7312 {
7313 	unsigned int dest;
7314 	struct pi_desc old, new;
7315 	struct pi_desc *pi_desc = vcpu_to_pi_desc(vcpu);
7316 
7317 	if (!kvm_arch_has_assigned_device(vcpu->kvm) ||
7318 		!irq_remapping_cap(IRQ_POSTING_CAP)  ||
7319 		!kvm_vcpu_apicv_active(vcpu))
7320 		return 0;
7321 
7322 	WARN_ON(irqs_disabled());
7323 	local_irq_disable();
7324 	if (!WARN_ON_ONCE(vcpu->pre_pcpu != -1)) {
7325 		vcpu->pre_pcpu = vcpu->cpu;
7326 		spin_lock(&per_cpu(blocked_vcpu_on_cpu_lock, vcpu->pre_pcpu));
7327 		list_add_tail(&vcpu->blocked_vcpu_list,
7328 			      &per_cpu(blocked_vcpu_on_cpu,
7329 				       vcpu->pre_pcpu));
7330 		spin_unlock(&per_cpu(blocked_vcpu_on_cpu_lock, vcpu->pre_pcpu));
7331 	}
7332 
7333 	do {
7334 		old.control = new.control = pi_desc->control;
7335 
7336 		WARN((pi_desc->sn == 1),
7337 		     "Warning: SN field of posted-interrupts "
7338 		     "is set before blocking\n");
7339 
7340 		/*
7341 		 * Since vCPU can be preempted during this process,
7342 		 * vcpu->cpu could be different with pre_pcpu, we
7343 		 * need to set pre_pcpu as the destination of wakeup
7344 		 * notification event, then we can find the right vCPU
7345 		 * to wakeup in wakeup handler if interrupts happen
7346 		 * when the vCPU is in blocked state.
7347 		 */
7348 		dest = cpu_physical_id(vcpu->pre_pcpu);
7349 
7350 		if (x2apic_enabled())
7351 			new.ndst = dest;
7352 		else
7353 			new.ndst = (dest << 8) & 0xFF00;
7354 
7355 		/* set 'NV' to 'wakeup vector' */
7356 		new.nv = POSTED_INTR_WAKEUP_VECTOR;
7357 	} while (cmpxchg64(&pi_desc->control, old.control,
7358 			   new.control) != old.control);
7359 
7360 	/* We should not block the vCPU if an interrupt is posted for it.  */
7361 	if (pi_test_on(pi_desc) == 1)
7362 		__pi_post_block(vcpu);
7363 
7364 	local_irq_enable();
7365 	return (vcpu->pre_pcpu == -1);
7366 }
7367 
7368 static int vmx_pre_block(struct kvm_vcpu *vcpu)
7369 {
7370 	if (pi_pre_block(vcpu))
7371 		return 1;
7372 
7373 	if (kvm_lapic_hv_timer_in_use(vcpu))
7374 		kvm_lapic_switch_to_sw_timer(vcpu);
7375 
7376 	return 0;
7377 }
7378 
7379 static void pi_post_block(struct kvm_vcpu *vcpu)
7380 {
7381 	if (vcpu->pre_pcpu == -1)
7382 		return;
7383 
7384 	WARN_ON(irqs_disabled());
7385 	local_irq_disable();
7386 	__pi_post_block(vcpu);
7387 	local_irq_enable();
7388 }
7389 
7390 static void vmx_post_block(struct kvm_vcpu *vcpu)
7391 {
7392 	if (kvm_x86_ops->set_hv_timer)
7393 		kvm_lapic_switch_to_hv_timer(vcpu);
7394 
7395 	pi_post_block(vcpu);
7396 }
7397 
7398 /*
7399  * vmx_update_pi_irte - set IRTE for Posted-Interrupts
7400  *
7401  * @kvm: kvm
7402  * @host_irq: host irq of the interrupt
7403  * @guest_irq: gsi of the interrupt
7404  * @set: set or unset PI
7405  * returns 0 on success, < 0 on failure
7406  */
7407 static int vmx_update_pi_irte(struct kvm *kvm, unsigned int host_irq,
7408 			      uint32_t guest_irq, bool set)
7409 {
7410 	struct kvm_kernel_irq_routing_entry *e;
7411 	struct kvm_irq_routing_table *irq_rt;
7412 	struct kvm_lapic_irq irq;
7413 	struct kvm_vcpu *vcpu;
7414 	struct vcpu_data vcpu_info;
7415 	int idx, ret = 0;
7416 
7417 	if (!kvm_arch_has_assigned_device(kvm) ||
7418 		!irq_remapping_cap(IRQ_POSTING_CAP) ||
7419 		!kvm_vcpu_apicv_active(kvm->vcpus[0]))
7420 		return 0;
7421 
7422 	idx = srcu_read_lock(&kvm->irq_srcu);
7423 	irq_rt = srcu_dereference(kvm->irq_routing, &kvm->irq_srcu);
7424 	if (guest_irq >= irq_rt->nr_rt_entries ||
7425 	    hlist_empty(&irq_rt->map[guest_irq])) {
7426 		pr_warn_once("no route for guest_irq %u/%u (broken user space?)\n",
7427 			     guest_irq, irq_rt->nr_rt_entries);
7428 		goto out;
7429 	}
7430 
7431 	hlist_for_each_entry(e, &irq_rt->map[guest_irq], link) {
7432 		if (e->type != KVM_IRQ_ROUTING_MSI)
7433 			continue;
7434 		/*
7435 		 * VT-d PI cannot support posting multicast/broadcast
7436 		 * interrupts to a vCPU, we still use interrupt remapping
7437 		 * for these kind of interrupts.
7438 		 *
7439 		 * For lowest-priority interrupts, we only support
7440 		 * those with single CPU as the destination, e.g. user
7441 		 * configures the interrupts via /proc/irq or uses
7442 		 * irqbalance to make the interrupts single-CPU.
7443 		 *
7444 		 * We will support full lowest-priority interrupt later.
7445 		 *
7446 		 * In addition, we can only inject generic interrupts using
7447 		 * the PI mechanism, refuse to route others through it.
7448 		 */
7449 
7450 		kvm_set_msi_irq(kvm, e, &irq);
7451 		if (!kvm_intr_is_single_vcpu(kvm, &irq, &vcpu) ||
7452 		    !kvm_irq_is_postable(&irq)) {
7453 			/*
7454 			 * Make sure the IRTE is in remapped mode if
7455 			 * we don't handle it in posted mode.
7456 			 */
7457 			ret = irq_set_vcpu_affinity(host_irq, NULL);
7458 			if (ret < 0) {
7459 				printk(KERN_INFO
7460 				   "failed to back to remapped mode, irq: %u\n",
7461 				   host_irq);
7462 				goto out;
7463 			}
7464 
7465 			continue;
7466 		}
7467 
7468 		vcpu_info.pi_desc_addr = __pa(vcpu_to_pi_desc(vcpu));
7469 		vcpu_info.vector = irq.vector;
7470 
7471 		trace_kvm_pi_irte_update(host_irq, vcpu->vcpu_id, e->gsi,
7472 				vcpu_info.vector, vcpu_info.pi_desc_addr, set);
7473 
7474 		if (set)
7475 			ret = irq_set_vcpu_affinity(host_irq, &vcpu_info);
7476 		else
7477 			ret = irq_set_vcpu_affinity(host_irq, NULL);
7478 
7479 		if (ret < 0) {
7480 			printk(KERN_INFO "%s: failed to update PI IRTE\n",
7481 					__func__);
7482 			goto out;
7483 		}
7484 	}
7485 
7486 	ret = 0;
7487 out:
7488 	srcu_read_unlock(&kvm->irq_srcu, idx);
7489 	return ret;
7490 }
7491 
7492 static void vmx_setup_mce(struct kvm_vcpu *vcpu)
7493 {
7494 	if (vcpu->arch.mcg_cap & MCG_LMCE_P)
7495 		to_vmx(vcpu)->msr_ia32_feature_control_valid_bits |=
7496 			FEATURE_CONTROL_LMCE;
7497 	else
7498 		to_vmx(vcpu)->msr_ia32_feature_control_valid_bits &=
7499 			~FEATURE_CONTROL_LMCE;
7500 }
7501 
7502 static int vmx_smi_allowed(struct kvm_vcpu *vcpu)
7503 {
7504 	/* we need a nested vmexit to enter SMM, postpone if run is pending */
7505 	if (to_vmx(vcpu)->nested.nested_run_pending)
7506 		return 0;
7507 	return 1;
7508 }
7509 
7510 static int vmx_pre_enter_smm(struct kvm_vcpu *vcpu, char *smstate)
7511 {
7512 	struct vcpu_vmx *vmx = to_vmx(vcpu);
7513 
7514 	vmx->nested.smm.guest_mode = is_guest_mode(vcpu);
7515 	if (vmx->nested.smm.guest_mode)
7516 		nested_vmx_vmexit(vcpu, -1, 0, 0);
7517 
7518 	vmx->nested.smm.vmxon = vmx->nested.vmxon;
7519 	vmx->nested.vmxon = false;
7520 	vmx_clear_hlt(vcpu);
7521 	return 0;
7522 }
7523 
7524 static int vmx_pre_leave_smm(struct kvm_vcpu *vcpu, const char *smstate)
7525 {
7526 	struct vcpu_vmx *vmx = to_vmx(vcpu);
7527 	int ret;
7528 
7529 	if (vmx->nested.smm.vmxon) {
7530 		vmx->nested.vmxon = true;
7531 		vmx->nested.smm.vmxon = false;
7532 	}
7533 
7534 	if (vmx->nested.smm.guest_mode) {
7535 		ret = nested_vmx_enter_non_root_mode(vcpu, false);
7536 		if (ret)
7537 			return ret;
7538 
7539 		vmx->nested.smm.guest_mode = false;
7540 	}
7541 	return 0;
7542 }
7543 
7544 static int enable_smi_window(struct kvm_vcpu *vcpu)
7545 {
7546 	return 0;
7547 }
7548 
7549 static bool vmx_need_emulation_on_page_fault(struct kvm_vcpu *vcpu)
7550 {
7551 	return false;
7552 }
7553 
7554 static bool vmx_apic_init_signal_blocked(struct kvm_vcpu *vcpu)
7555 {
7556 	return to_vmx(vcpu)->nested.vmxon;
7557 }
7558 
7559 static __init int hardware_setup(void)
7560 {
7561 	unsigned long host_bndcfgs;
7562 	struct desc_ptr dt;
7563 	int r, i;
7564 
7565 	rdmsrl_safe(MSR_EFER, &host_efer);
7566 
7567 	store_idt(&dt);
7568 	host_idt_base = dt.address;
7569 
7570 	for (i = 0; i < ARRAY_SIZE(vmx_msr_index); ++i)
7571 		kvm_define_shared_msr(i, vmx_msr_index[i]);
7572 
7573 	if (setup_vmcs_config(&vmcs_config, &vmx_capability) < 0)
7574 		return -EIO;
7575 
7576 	if (boot_cpu_has(X86_FEATURE_NX))
7577 		kvm_enable_efer_bits(EFER_NX);
7578 
7579 	if (boot_cpu_has(X86_FEATURE_MPX)) {
7580 		rdmsrl(MSR_IA32_BNDCFGS, host_bndcfgs);
7581 		WARN_ONCE(host_bndcfgs, "KVM: BNDCFGS in host will be lost");
7582 	}
7583 
7584 	if (boot_cpu_has(X86_FEATURE_XSAVES))
7585 		rdmsrl(MSR_IA32_XSS, host_xss);
7586 
7587 	if (!cpu_has_vmx_vpid() || !cpu_has_vmx_invvpid() ||
7588 	    !(cpu_has_vmx_invvpid_single() || cpu_has_vmx_invvpid_global()))
7589 		enable_vpid = 0;
7590 
7591 	if (!cpu_has_vmx_ept() ||
7592 	    !cpu_has_vmx_ept_4levels() ||
7593 	    !cpu_has_vmx_ept_mt_wb() ||
7594 	    !cpu_has_vmx_invept_global())
7595 		enable_ept = 0;
7596 
7597 	if (!cpu_has_vmx_ept_ad_bits() || !enable_ept)
7598 		enable_ept_ad_bits = 0;
7599 
7600 	if (!cpu_has_vmx_unrestricted_guest() || !enable_ept)
7601 		enable_unrestricted_guest = 0;
7602 
7603 	if (!cpu_has_vmx_flexpriority())
7604 		flexpriority_enabled = 0;
7605 
7606 	if (!cpu_has_virtual_nmis())
7607 		enable_vnmi = 0;
7608 
7609 	/*
7610 	 * set_apic_access_page_addr() is used to reload apic access
7611 	 * page upon invalidation.  No need to do anything if not
7612 	 * using the APIC_ACCESS_ADDR VMCS field.
7613 	 */
7614 	if (!flexpriority_enabled)
7615 		kvm_x86_ops->set_apic_access_page_addr = NULL;
7616 
7617 	if (!cpu_has_vmx_tpr_shadow())
7618 		kvm_x86_ops->update_cr8_intercept = NULL;
7619 
7620 	if (enable_ept && !cpu_has_vmx_ept_2m_page())
7621 		kvm_disable_largepages();
7622 
7623 #if IS_ENABLED(CONFIG_HYPERV)
7624 	if (ms_hyperv.nested_features & HV_X64_NESTED_GUEST_MAPPING_FLUSH
7625 	    && enable_ept) {
7626 		kvm_x86_ops->tlb_remote_flush = hv_remote_flush_tlb;
7627 		kvm_x86_ops->tlb_remote_flush_with_range =
7628 				hv_remote_flush_tlb_with_range;
7629 	}
7630 #endif
7631 
7632 	if (!cpu_has_vmx_ple()) {
7633 		ple_gap = 0;
7634 		ple_window = 0;
7635 		ple_window_grow = 0;
7636 		ple_window_max = 0;
7637 		ple_window_shrink = 0;
7638 	}
7639 
7640 	if (!cpu_has_vmx_apicv()) {
7641 		enable_apicv = 0;
7642 		kvm_x86_ops->sync_pir_to_irr = NULL;
7643 	}
7644 
7645 	if (cpu_has_vmx_tsc_scaling()) {
7646 		kvm_has_tsc_control = true;
7647 		kvm_max_tsc_scaling_ratio = KVM_VMX_TSC_MULTIPLIER_MAX;
7648 		kvm_tsc_scaling_ratio_frac_bits = 48;
7649 	}
7650 
7651 	set_bit(0, vmx_vpid_bitmap); /* 0 is reserved for host */
7652 
7653 	if (enable_ept)
7654 		vmx_enable_tdp();
7655 	else
7656 		kvm_disable_tdp();
7657 
7658 	/*
7659 	 * Only enable PML when hardware supports PML feature, and both EPT
7660 	 * and EPT A/D bit features are enabled -- PML depends on them to work.
7661 	 */
7662 	if (!enable_ept || !enable_ept_ad_bits || !cpu_has_vmx_pml())
7663 		enable_pml = 0;
7664 
7665 	if (!enable_pml) {
7666 		kvm_x86_ops->slot_enable_log_dirty = NULL;
7667 		kvm_x86_ops->slot_disable_log_dirty = NULL;
7668 		kvm_x86_ops->flush_log_dirty = NULL;
7669 		kvm_x86_ops->enable_log_dirty_pt_masked = NULL;
7670 	}
7671 
7672 	if (!cpu_has_vmx_preemption_timer())
7673 		enable_preemption_timer = false;
7674 
7675 	if (enable_preemption_timer) {
7676 		u64 use_timer_freq = 5000ULL * 1000 * 1000;
7677 		u64 vmx_msr;
7678 
7679 		rdmsrl(MSR_IA32_VMX_MISC, vmx_msr);
7680 		cpu_preemption_timer_multi =
7681 			vmx_msr & VMX_MISC_PREEMPTION_TIMER_RATE_MASK;
7682 
7683 		if (tsc_khz)
7684 			use_timer_freq = (u64)tsc_khz * 1000;
7685 		use_timer_freq >>= cpu_preemption_timer_multi;
7686 
7687 		/*
7688 		 * KVM "disables" the preemption timer by setting it to its max
7689 		 * value.  Don't use the timer if it might cause spurious exits
7690 		 * at a rate faster than 0.1 Hz (of uninterrupted guest time).
7691 		 */
7692 		if (use_timer_freq > 0xffffffffu / 10)
7693 			enable_preemption_timer = false;
7694 	}
7695 
7696 	if (!enable_preemption_timer) {
7697 		kvm_x86_ops->set_hv_timer = NULL;
7698 		kvm_x86_ops->cancel_hv_timer = NULL;
7699 		kvm_x86_ops->request_immediate_exit = __kvm_request_immediate_exit;
7700 	}
7701 
7702 	kvm_set_posted_intr_wakeup_handler(wakeup_handler);
7703 
7704 	kvm_mce_cap_supported |= MCG_LMCE_P;
7705 
7706 	if (pt_mode != PT_MODE_SYSTEM && pt_mode != PT_MODE_HOST_GUEST)
7707 		return -EINVAL;
7708 	if (!enable_ept || !cpu_has_vmx_intel_pt())
7709 		pt_mode = PT_MODE_SYSTEM;
7710 
7711 	if (nested) {
7712 		nested_vmx_setup_ctls_msrs(&vmcs_config.nested,
7713 					   vmx_capability.ept, enable_apicv);
7714 
7715 		r = nested_vmx_hardware_setup(kvm_vmx_exit_handlers);
7716 		if (r)
7717 			return r;
7718 	}
7719 
7720 	r = alloc_kvm_area();
7721 	if (r)
7722 		nested_vmx_hardware_unsetup();
7723 	return r;
7724 }
7725 
7726 static __exit void hardware_unsetup(void)
7727 {
7728 	if (nested)
7729 		nested_vmx_hardware_unsetup();
7730 
7731 	free_kvm_area();
7732 }
7733 
7734 static struct kvm_x86_ops vmx_x86_ops __ro_after_init = {
7735 	.cpu_has_kvm_support = cpu_has_kvm_support,
7736 	.disabled_by_bios = vmx_disabled_by_bios,
7737 	.hardware_setup = hardware_setup,
7738 	.hardware_unsetup = hardware_unsetup,
7739 	.check_processor_compatibility = vmx_check_processor_compat,
7740 	.hardware_enable = hardware_enable,
7741 	.hardware_disable = hardware_disable,
7742 	.cpu_has_accelerated_tpr = report_flexpriority,
7743 	.has_emulated_msr = vmx_has_emulated_msr,
7744 
7745 	.vm_init = vmx_vm_init,
7746 	.vm_alloc = vmx_vm_alloc,
7747 	.vm_free = vmx_vm_free,
7748 
7749 	.vcpu_create = vmx_create_vcpu,
7750 	.vcpu_free = vmx_free_vcpu,
7751 	.vcpu_reset = vmx_vcpu_reset,
7752 
7753 	.prepare_guest_switch = vmx_prepare_switch_to_guest,
7754 	.vcpu_load = vmx_vcpu_load,
7755 	.vcpu_put = vmx_vcpu_put,
7756 
7757 	.update_bp_intercept = update_exception_bitmap,
7758 	.get_msr_feature = vmx_get_msr_feature,
7759 	.get_msr = vmx_get_msr,
7760 	.set_msr = vmx_set_msr,
7761 	.get_segment_base = vmx_get_segment_base,
7762 	.get_segment = vmx_get_segment,
7763 	.set_segment = vmx_set_segment,
7764 	.get_cpl = vmx_get_cpl,
7765 	.get_cs_db_l_bits = vmx_get_cs_db_l_bits,
7766 	.decache_cr0_guest_bits = vmx_decache_cr0_guest_bits,
7767 	.decache_cr3 = vmx_decache_cr3,
7768 	.decache_cr4_guest_bits = vmx_decache_cr4_guest_bits,
7769 	.set_cr0 = vmx_set_cr0,
7770 	.set_cr3 = vmx_set_cr3,
7771 	.set_cr4 = vmx_set_cr4,
7772 	.set_efer = vmx_set_efer,
7773 	.get_idt = vmx_get_idt,
7774 	.set_idt = vmx_set_idt,
7775 	.get_gdt = vmx_get_gdt,
7776 	.set_gdt = vmx_set_gdt,
7777 	.get_dr6 = vmx_get_dr6,
7778 	.set_dr6 = vmx_set_dr6,
7779 	.set_dr7 = vmx_set_dr7,
7780 	.sync_dirty_debug_regs = vmx_sync_dirty_debug_regs,
7781 	.cache_reg = vmx_cache_reg,
7782 	.get_rflags = vmx_get_rflags,
7783 	.set_rflags = vmx_set_rflags,
7784 
7785 	.tlb_flush = vmx_flush_tlb,
7786 	.tlb_flush_gva = vmx_flush_tlb_gva,
7787 
7788 	.run = vmx_vcpu_run,
7789 	.handle_exit = vmx_handle_exit,
7790 	.skip_emulated_instruction = skip_emulated_instruction,
7791 	.set_interrupt_shadow = vmx_set_interrupt_shadow,
7792 	.get_interrupt_shadow = vmx_get_interrupt_shadow,
7793 	.patch_hypercall = vmx_patch_hypercall,
7794 	.set_irq = vmx_inject_irq,
7795 	.set_nmi = vmx_inject_nmi,
7796 	.queue_exception = vmx_queue_exception,
7797 	.cancel_injection = vmx_cancel_injection,
7798 	.interrupt_allowed = vmx_interrupt_allowed,
7799 	.nmi_allowed = vmx_nmi_allowed,
7800 	.get_nmi_mask = vmx_get_nmi_mask,
7801 	.set_nmi_mask = vmx_set_nmi_mask,
7802 	.enable_nmi_window = enable_nmi_window,
7803 	.enable_irq_window = enable_irq_window,
7804 	.update_cr8_intercept = update_cr8_intercept,
7805 	.set_virtual_apic_mode = vmx_set_virtual_apic_mode,
7806 	.set_apic_access_page_addr = vmx_set_apic_access_page_addr,
7807 	.get_enable_apicv = vmx_get_enable_apicv,
7808 	.refresh_apicv_exec_ctrl = vmx_refresh_apicv_exec_ctrl,
7809 	.load_eoi_exitmap = vmx_load_eoi_exitmap,
7810 	.apicv_post_state_restore = vmx_apicv_post_state_restore,
7811 	.hwapic_irr_update = vmx_hwapic_irr_update,
7812 	.hwapic_isr_update = vmx_hwapic_isr_update,
7813 	.guest_apic_has_interrupt = vmx_guest_apic_has_interrupt,
7814 	.sync_pir_to_irr = vmx_sync_pir_to_irr,
7815 	.deliver_posted_interrupt = vmx_deliver_posted_interrupt,
7816 	.dy_apicv_has_pending_interrupt = vmx_dy_apicv_has_pending_interrupt,
7817 
7818 	.set_tss_addr = vmx_set_tss_addr,
7819 	.set_identity_map_addr = vmx_set_identity_map_addr,
7820 	.get_tdp_level = get_ept_level,
7821 	.get_mt_mask = vmx_get_mt_mask,
7822 
7823 	.get_exit_info = vmx_get_exit_info,
7824 
7825 	.get_lpage_level = vmx_get_lpage_level,
7826 
7827 	.cpuid_update = vmx_cpuid_update,
7828 
7829 	.rdtscp_supported = vmx_rdtscp_supported,
7830 	.invpcid_supported = vmx_invpcid_supported,
7831 
7832 	.set_supported_cpuid = vmx_set_supported_cpuid,
7833 
7834 	.has_wbinvd_exit = cpu_has_vmx_wbinvd_exit,
7835 
7836 	.read_l1_tsc_offset = vmx_read_l1_tsc_offset,
7837 	.write_l1_tsc_offset = vmx_write_l1_tsc_offset,
7838 
7839 	.set_tdp_cr3 = vmx_set_cr3,
7840 
7841 	.check_intercept = vmx_check_intercept,
7842 	.handle_exit_irqoff = vmx_handle_exit_irqoff,
7843 	.mpx_supported = vmx_mpx_supported,
7844 	.xsaves_supported = vmx_xsaves_supported,
7845 	.umip_emulated = vmx_umip_emulated,
7846 	.pt_supported = vmx_pt_supported,
7847 
7848 	.request_immediate_exit = vmx_request_immediate_exit,
7849 
7850 	.sched_in = vmx_sched_in,
7851 
7852 	.slot_enable_log_dirty = vmx_slot_enable_log_dirty,
7853 	.slot_disable_log_dirty = vmx_slot_disable_log_dirty,
7854 	.flush_log_dirty = vmx_flush_log_dirty,
7855 	.enable_log_dirty_pt_masked = vmx_enable_log_dirty_pt_masked,
7856 	.write_log_dirty = vmx_write_pml_buffer,
7857 
7858 	.pre_block = vmx_pre_block,
7859 	.post_block = vmx_post_block,
7860 
7861 	.pmu_ops = &intel_pmu_ops,
7862 
7863 	.update_pi_irte = vmx_update_pi_irte,
7864 
7865 #ifdef CONFIG_X86_64
7866 	.set_hv_timer = vmx_set_hv_timer,
7867 	.cancel_hv_timer = vmx_cancel_hv_timer,
7868 #endif
7869 
7870 	.setup_mce = vmx_setup_mce,
7871 
7872 	.smi_allowed = vmx_smi_allowed,
7873 	.pre_enter_smm = vmx_pre_enter_smm,
7874 	.pre_leave_smm = vmx_pre_leave_smm,
7875 	.enable_smi_window = enable_smi_window,
7876 
7877 	.check_nested_events = NULL,
7878 	.get_nested_state = NULL,
7879 	.set_nested_state = NULL,
7880 	.get_vmcs12_pages = NULL,
7881 	.nested_enable_evmcs = NULL,
7882 	.nested_get_evmcs_version = NULL,
7883 	.need_emulation_on_page_fault = vmx_need_emulation_on_page_fault,
7884 	.apic_init_signal_blocked = vmx_apic_init_signal_blocked,
7885 };
7886 
7887 static void vmx_cleanup_l1d_flush(void)
7888 {
7889 	if (vmx_l1d_flush_pages) {
7890 		free_pages((unsigned long)vmx_l1d_flush_pages, L1D_CACHE_ORDER);
7891 		vmx_l1d_flush_pages = NULL;
7892 	}
7893 	/* Restore state so sysfs ignores VMX */
7894 	l1tf_vmx_mitigation = VMENTER_L1D_FLUSH_AUTO;
7895 }
7896 
7897 static void vmx_exit(void)
7898 {
7899 #ifdef CONFIG_KEXEC_CORE
7900 	RCU_INIT_POINTER(crash_vmclear_loaded_vmcss, NULL);
7901 	synchronize_rcu();
7902 #endif
7903 
7904 	kvm_exit();
7905 
7906 #if IS_ENABLED(CONFIG_HYPERV)
7907 	if (static_branch_unlikely(&enable_evmcs)) {
7908 		int cpu;
7909 		struct hv_vp_assist_page *vp_ap;
7910 		/*
7911 		 * Reset everything to support using non-enlightened VMCS
7912 		 * access later (e.g. when we reload the module with
7913 		 * enlightened_vmcs=0)
7914 		 */
7915 		for_each_online_cpu(cpu) {
7916 			vp_ap =	hv_get_vp_assist_page(cpu);
7917 
7918 			if (!vp_ap)
7919 				continue;
7920 
7921 			vp_ap->nested_control.features.directhypercall = 0;
7922 			vp_ap->current_nested_vmcs = 0;
7923 			vp_ap->enlighten_vmentry = 0;
7924 		}
7925 
7926 		static_branch_disable(&enable_evmcs);
7927 	}
7928 #endif
7929 	vmx_cleanup_l1d_flush();
7930 }
7931 module_exit(vmx_exit);
7932 
7933 static int __init vmx_init(void)
7934 {
7935 	int r;
7936 
7937 #if IS_ENABLED(CONFIG_HYPERV)
7938 	/*
7939 	 * Enlightened VMCS usage should be recommended and the host needs
7940 	 * to support eVMCS v1 or above. We can also disable eVMCS support
7941 	 * with module parameter.
7942 	 */
7943 	if (enlightened_vmcs &&
7944 	    ms_hyperv.hints & HV_X64_ENLIGHTENED_VMCS_RECOMMENDED &&
7945 	    (ms_hyperv.nested_features & HV_X64_ENLIGHTENED_VMCS_VERSION) >=
7946 	    KVM_EVMCS_VERSION) {
7947 		int cpu;
7948 
7949 		/* Check that we have assist pages on all online CPUs */
7950 		for_each_online_cpu(cpu) {
7951 			if (!hv_get_vp_assist_page(cpu)) {
7952 				enlightened_vmcs = false;
7953 				break;
7954 			}
7955 		}
7956 
7957 		if (enlightened_vmcs) {
7958 			pr_info("KVM: vmx: using Hyper-V Enlightened VMCS\n");
7959 			static_branch_enable(&enable_evmcs);
7960 		}
7961 
7962 		if (ms_hyperv.nested_features & HV_X64_NESTED_DIRECT_FLUSH)
7963 			vmx_x86_ops.enable_direct_tlbflush
7964 				= hv_enable_direct_tlbflush;
7965 
7966 	} else {
7967 		enlightened_vmcs = false;
7968 	}
7969 #endif
7970 
7971 	r = kvm_init(&vmx_x86_ops, sizeof(struct vcpu_vmx),
7972 		     __alignof__(struct vcpu_vmx), THIS_MODULE);
7973 	if (r)
7974 		return r;
7975 
7976 	/*
7977 	 * Must be called after kvm_init() so enable_ept is properly set
7978 	 * up. Hand the parameter mitigation value in which was stored in
7979 	 * the pre module init parser. If no parameter was given, it will
7980 	 * contain 'auto' which will be turned into the default 'cond'
7981 	 * mitigation mode.
7982 	 */
7983 	r = vmx_setup_l1d_flush(vmentry_l1d_flush_param);
7984 	if (r) {
7985 		vmx_exit();
7986 		return r;
7987 	}
7988 
7989 #ifdef CONFIG_KEXEC_CORE
7990 	rcu_assign_pointer(crash_vmclear_loaded_vmcss,
7991 			   crash_vmclear_local_loaded_vmcss);
7992 #endif
7993 	vmx_check_vmcs12_offsets();
7994 
7995 	return 0;
7996 }
7997 module_init(vmx_init);
7998