xref: /openbmc/linux/arch/x86/kvm/vmx/vmx.c (revision aa5b395b)
1 // SPDX-License-Identifier: GPL-2.0-only
2 /*
3  * Kernel-based Virtual Machine driver for Linux
4  *
5  * This module enables machines with Intel VT-x extensions to run virtual
6  * machines without emulation or binary translation.
7  *
8  * Copyright (C) 2006 Qumranet, Inc.
9  * Copyright 2010 Red Hat, Inc. and/or its affiliates.
10  *
11  * Authors:
12  *   Avi Kivity   <avi@qumranet.com>
13  *   Yaniv Kamay  <yaniv@qumranet.com>
14  */
15 
16 #include <linux/frame.h>
17 #include <linux/highmem.h>
18 #include <linux/hrtimer.h>
19 #include <linux/kernel.h>
20 #include <linux/kvm_host.h>
21 #include <linux/module.h>
22 #include <linux/moduleparam.h>
23 #include <linux/mod_devicetable.h>
24 #include <linux/mm.h>
25 #include <linux/sched.h>
26 #include <linux/sched/smt.h>
27 #include <linux/slab.h>
28 #include <linux/tboot.h>
29 #include <linux/trace_events.h>
30 
31 #include <asm/apic.h>
32 #include <asm/asm.h>
33 #include <asm/cpu.h>
34 #include <asm/debugreg.h>
35 #include <asm/desc.h>
36 #include <asm/fpu/internal.h>
37 #include <asm/io.h>
38 #include <asm/irq_remapping.h>
39 #include <asm/kexec.h>
40 #include <asm/perf_event.h>
41 #include <asm/mce.h>
42 #include <asm/mmu_context.h>
43 #include <asm/mshyperv.h>
44 #include <asm/spec-ctrl.h>
45 #include <asm/virtext.h>
46 #include <asm/vmx.h>
47 
48 #include "capabilities.h"
49 #include "cpuid.h"
50 #include "evmcs.h"
51 #include "irq.h"
52 #include "kvm_cache_regs.h"
53 #include "lapic.h"
54 #include "mmu.h"
55 #include "nested.h"
56 #include "ops.h"
57 #include "pmu.h"
58 #include "trace.h"
59 #include "vmcs.h"
60 #include "vmcs12.h"
61 #include "vmx.h"
62 #include "x86.h"
63 
64 MODULE_AUTHOR("Qumranet");
65 MODULE_LICENSE("GPL");
66 
67 static const struct x86_cpu_id vmx_cpu_id[] = {
68 	X86_FEATURE_MATCH(X86_FEATURE_VMX),
69 	{}
70 };
71 MODULE_DEVICE_TABLE(x86cpu, vmx_cpu_id);
72 
73 bool __read_mostly enable_vpid = 1;
74 module_param_named(vpid, enable_vpid, bool, 0444);
75 
76 static bool __read_mostly enable_vnmi = 1;
77 module_param_named(vnmi, enable_vnmi, bool, S_IRUGO);
78 
79 bool __read_mostly flexpriority_enabled = 1;
80 module_param_named(flexpriority, flexpriority_enabled, bool, S_IRUGO);
81 
82 bool __read_mostly enable_ept = 1;
83 module_param_named(ept, enable_ept, bool, S_IRUGO);
84 
85 bool __read_mostly enable_unrestricted_guest = 1;
86 module_param_named(unrestricted_guest,
87 			enable_unrestricted_guest, bool, S_IRUGO);
88 
89 bool __read_mostly enable_ept_ad_bits = 1;
90 module_param_named(eptad, enable_ept_ad_bits, bool, S_IRUGO);
91 
92 static bool __read_mostly emulate_invalid_guest_state = true;
93 module_param(emulate_invalid_guest_state, bool, S_IRUGO);
94 
95 static bool __read_mostly fasteoi = 1;
96 module_param(fasteoi, bool, S_IRUGO);
97 
98 static bool __read_mostly enable_apicv = 1;
99 module_param(enable_apicv, bool, S_IRUGO);
100 
101 /*
102  * If nested=1, nested virtualization is supported, i.e., guests may use
103  * VMX and be a hypervisor for its own guests. If nested=0, guests may not
104  * use VMX instructions.
105  */
106 static bool __read_mostly nested = 1;
107 module_param(nested, bool, S_IRUGO);
108 
109 bool __read_mostly enable_pml = 1;
110 module_param_named(pml, enable_pml, bool, S_IRUGO);
111 
112 static bool __read_mostly dump_invalid_vmcs = 0;
113 module_param(dump_invalid_vmcs, bool, 0644);
114 
115 #define MSR_BITMAP_MODE_X2APIC		1
116 #define MSR_BITMAP_MODE_X2APIC_APICV	2
117 
118 #define KVM_VMX_TSC_MULTIPLIER_MAX     0xffffffffffffffffULL
119 
120 /* Guest_tsc -> host_tsc conversion requires 64-bit division.  */
121 static int __read_mostly cpu_preemption_timer_multi;
122 static bool __read_mostly enable_preemption_timer = 1;
123 #ifdef CONFIG_X86_64
124 module_param_named(preemption_timer, enable_preemption_timer, bool, S_IRUGO);
125 #endif
126 
127 #define KVM_VM_CR0_ALWAYS_OFF (X86_CR0_NW | X86_CR0_CD)
128 #define KVM_VM_CR0_ALWAYS_ON_UNRESTRICTED_GUEST X86_CR0_NE
129 #define KVM_VM_CR0_ALWAYS_ON				\
130 	(KVM_VM_CR0_ALWAYS_ON_UNRESTRICTED_GUEST | 	\
131 	 X86_CR0_WP | X86_CR0_PG | X86_CR0_PE)
132 #define KVM_CR4_GUEST_OWNED_BITS				      \
133 	(X86_CR4_PVI | X86_CR4_DE | X86_CR4_PCE | X86_CR4_OSFXSR      \
134 	 | X86_CR4_OSXMMEXCPT | X86_CR4_LA57 | X86_CR4_TSD)
135 
136 #define KVM_VM_CR4_ALWAYS_ON_UNRESTRICTED_GUEST X86_CR4_VMXE
137 #define KVM_PMODE_VM_CR4_ALWAYS_ON (X86_CR4_PAE | X86_CR4_VMXE)
138 #define KVM_RMODE_VM_CR4_ALWAYS_ON (X86_CR4_VME | X86_CR4_PAE | X86_CR4_VMXE)
139 
140 #define RMODE_GUEST_OWNED_EFLAGS_BITS (~(X86_EFLAGS_IOPL | X86_EFLAGS_VM))
141 
142 #define MSR_IA32_RTIT_STATUS_MASK (~(RTIT_STATUS_FILTEREN | \
143 	RTIT_STATUS_CONTEXTEN | RTIT_STATUS_TRIGGEREN | \
144 	RTIT_STATUS_ERROR | RTIT_STATUS_STOPPED | \
145 	RTIT_STATUS_BYTECNT))
146 
147 #define MSR_IA32_RTIT_OUTPUT_BASE_MASK \
148 	(~((1UL << cpuid_query_maxphyaddr(vcpu)) - 1) | 0x7f)
149 
150 /*
151  * These 2 parameters are used to config the controls for Pause-Loop Exiting:
152  * ple_gap:    upper bound on the amount of time between two successive
153  *             executions of PAUSE in a loop. Also indicate if ple enabled.
154  *             According to test, this time is usually smaller than 128 cycles.
155  * ple_window: upper bound on the amount of time a guest is allowed to execute
156  *             in a PAUSE loop. Tests indicate that most spinlocks are held for
157  *             less than 2^12 cycles
158  * Time is measured based on a counter that runs at the same rate as the TSC,
159  * refer SDM volume 3b section 21.6.13 & 22.1.3.
160  */
161 static unsigned int ple_gap = KVM_DEFAULT_PLE_GAP;
162 module_param(ple_gap, uint, 0444);
163 
164 static unsigned int ple_window = KVM_VMX_DEFAULT_PLE_WINDOW;
165 module_param(ple_window, uint, 0444);
166 
167 /* Default doubles per-vcpu window every exit. */
168 static unsigned int ple_window_grow = KVM_DEFAULT_PLE_WINDOW_GROW;
169 module_param(ple_window_grow, uint, 0444);
170 
171 /* Default resets per-vcpu window every exit to ple_window. */
172 static unsigned int ple_window_shrink = KVM_DEFAULT_PLE_WINDOW_SHRINK;
173 module_param(ple_window_shrink, uint, 0444);
174 
175 /* Default is to compute the maximum so we can never overflow. */
176 static unsigned int ple_window_max        = KVM_VMX_DEFAULT_PLE_WINDOW_MAX;
177 module_param(ple_window_max, uint, 0444);
178 
179 /* Default is SYSTEM mode, 1 for host-guest mode */
180 int __read_mostly pt_mode = PT_MODE_SYSTEM;
181 module_param(pt_mode, int, S_IRUGO);
182 
183 static DEFINE_STATIC_KEY_FALSE(vmx_l1d_should_flush);
184 static DEFINE_STATIC_KEY_FALSE(vmx_l1d_flush_cond);
185 static DEFINE_MUTEX(vmx_l1d_flush_mutex);
186 
187 /* Storage for pre module init parameter parsing */
188 static enum vmx_l1d_flush_state __read_mostly vmentry_l1d_flush_param = VMENTER_L1D_FLUSH_AUTO;
189 
190 static const struct {
191 	const char *option;
192 	bool for_parse;
193 } vmentry_l1d_param[] = {
194 	[VMENTER_L1D_FLUSH_AUTO]	 = {"auto", true},
195 	[VMENTER_L1D_FLUSH_NEVER]	 = {"never", true},
196 	[VMENTER_L1D_FLUSH_COND]	 = {"cond", true},
197 	[VMENTER_L1D_FLUSH_ALWAYS]	 = {"always", true},
198 	[VMENTER_L1D_FLUSH_EPT_DISABLED] = {"EPT disabled", false},
199 	[VMENTER_L1D_FLUSH_NOT_REQUIRED] = {"not required", false},
200 };
201 
202 #define L1D_CACHE_ORDER 4
203 static void *vmx_l1d_flush_pages;
204 
205 static int vmx_setup_l1d_flush(enum vmx_l1d_flush_state l1tf)
206 {
207 	struct page *page;
208 	unsigned int i;
209 
210 	if (!boot_cpu_has_bug(X86_BUG_L1TF)) {
211 		l1tf_vmx_mitigation = VMENTER_L1D_FLUSH_NOT_REQUIRED;
212 		return 0;
213 	}
214 
215 	if (!enable_ept) {
216 		l1tf_vmx_mitigation = VMENTER_L1D_FLUSH_EPT_DISABLED;
217 		return 0;
218 	}
219 
220 	if (boot_cpu_has(X86_FEATURE_ARCH_CAPABILITIES)) {
221 		u64 msr;
222 
223 		rdmsrl(MSR_IA32_ARCH_CAPABILITIES, msr);
224 		if (msr & ARCH_CAP_SKIP_VMENTRY_L1DFLUSH) {
225 			l1tf_vmx_mitigation = VMENTER_L1D_FLUSH_NOT_REQUIRED;
226 			return 0;
227 		}
228 	}
229 
230 	/* If set to auto use the default l1tf mitigation method */
231 	if (l1tf == VMENTER_L1D_FLUSH_AUTO) {
232 		switch (l1tf_mitigation) {
233 		case L1TF_MITIGATION_OFF:
234 			l1tf = VMENTER_L1D_FLUSH_NEVER;
235 			break;
236 		case L1TF_MITIGATION_FLUSH_NOWARN:
237 		case L1TF_MITIGATION_FLUSH:
238 		case L1TF_MITIGATION_FLUSH_NOSMT:
239 			l1tf = VMENTER_L1D_FLUSH_COND;
240 			break;
241 		case L1TF_MITIGATION_FULL:
242 		case L1TF_MITIGATION_FULL_FORCE:
243 			l1tf = VMENTER_L1D_FLUSH_ALWAYS;
244 			break;
245 		}
246 	} else if (l1tf_mitigation == L1TF_MITIGATION_FULL_FORCE) {
247 		l1tf = VMENTER_L1D_FLUSH_ALWAYS;
248 	}
249 
250 	if (l1tf != VMENTER_L1D_FLUSH_NEVER && !vmx_l1d_flush_pages &&
251 	    !boot_cpu_has(X86_FEATURE_FLUSH_L1D)) {
252 		/*
253 		 * This allocation for vmx_l1d_flush_pages is not tied to a VM
254 		 * lifetime and so should not be charged to a memcg.
255 		 */
256 		page = alloc_pages(GFP_KERNEL, L1D_CACHE_ORDER);
257 		if (!page)
258 			return -ENOMEM;
259 		vmx_l1d_flush_pages = page_address(page);
260 
261 		/*
262 		 * Initialize each page with a different pattern in
263 		 * order to protect against KSM in the nested
264 		 * virtualization case.
265 		 */
266 		for (i = 0; i < 1u << L1D_CACHE_ORDER; ++i) {
267 			memset(vmx_l1d_flush_pages + i * PAGE_SIZE, i + 1,
268 			       PAGE_SIZE);
269 		}
270 	}
271 
272 	l1tf_vmx_mitigation = l1tf;
273 
274 	if (l1tf != VMENTER_L1D_FLUSH_NEVER)
275 		static_branch_enable(&vmx_l1d_should_flush);
276 	else
277 		static_branch_disable(&vmx_l1d_should_flush);
278 
279 	if (l1tf == VMENTER_L1D_FLUSH_COND)
280 		static_branch_enable(&vmx_l1d_flush_cond);
281 	else
282 		static_branch_disable(&vmx_l1d_flush_cond);
283 	return 0;
284 }
285 
286 static int vmentry_l1d_flush_parse(const char *s)
287 {
288 	unsigned int i;
289 
290 	if (s) {
291 		for (i = 0; i < ARRAY_SIZE(vmentry_l1d_param); i++) {
292 			if (vmentry_l1d_param[i].for_parse &&
293 			    sysfs_streq(s, vmentry_l1d_param[i].option))
294 				return i;
295 		}
296 	}
297 	return -EINVAL;
298 }
299 
300 static int vmentry_l1d_flush_set(const char *s, const struct kernel_param *kp)
301 {
302 	int l1tf, ret;
303 
304 	l1tf = vmentry_l1d_flush_parse(s);
305 	if (l1tf < 0)
306 		return l1tf;
307 
308 	if (!boot_cpu_has(X86_BUG_L1TF))
309 		return 0;
310 
311 	/*
312 	 * Has vmx_init() run already? If not then this is the pre init
313 	 * parameter parsing. In that case just store the value and let
314 	 * vmx_init() do the proper setup after enable_ept has been
315 	 * established.
316 	 */
317 	if (l1tf_vmx_mitigation == VMENTER_L1D_FLUSH_AUTO) {
318 		vmentry_l1d_flush_param = l1tf;
319 		return 0;
320 	}
321 
322 	mutex_lock(&vmx_l1d_flush_mutex);
323 	ret = vmx_setup_l1d_flush(l1tf);
324 	mutex_unlock(&vmx_l1d_flush_mutex);
325 	return ret;
326 }
327 
328 static int vmentry_l1d_flush_get(char *s, const struct kernel_param *kp)
329 {
330 	if (WARN_ON_ONCE(l1tf_vmx_mitigation >= ARRAY_SIZE(vmentry_l1d_param)))
331 		return sprintf(s, "???\n");
332 
333 	return sprintf(s, "%s\n", vmentry_l1d_param[l1tf_vmx_mitigation].option);
334 }
335 
336 static const struct kernel_param_ops vmentry_l1d_flush_ops = {
337 	.set = vmentry_l1d_flush_set,
338 	.get = vmentry_l1d_flush_get,
339 };
340 module_param_cb(vmentry_l1d_flush, &vmentry_l1d_flush_ops, NULL, 0644);
341 
342 static bool guest_state_valid(struct kvm_vcpu *vcpu);
343 static u32 vmx_segment_access_rights(struct kvm_segment *var);
344 static __always_inline void vmx_disable_intercept_for_msr(unsigned long *msr_bitmap,
345 							  u32 msr, int type);
346 
347 void vmx_vmexit(void);
348 
349 #define vmx_insn_failed(fmt...)		\
350 do {					\
351 	WARN_ONCE(1, fmt);		\
352 	pr_warn_ratelimited(fmt);	\
353 } while (0)
354 
355 asmlinkage void vmread_error(unsigned long field, bool fault)
356 {
357 	if (fault)
358 		kvm_spurious_fault();
359 	else
360 		vmx_insn_failed("kvm: vmread failed: field=%lx\n", field);
361 }
362 
363 noinline void vmwrite_error(unsigned long field, unsigned long value)
364 {
365 	vmx_insn_failed("kvm: vmwrite failed: field=%lx val=%lx err=%d\n",
366 			field, value, vmcs_read32(VM_INSTRUCTION_ERROR));
367 }
368 
369 noinline void vmclear_error(struct vmcs *vmcs, u64 phys_addr)
370 {
371 	vmx_insn_failed("kvm: vmclear failed: %p/%llx\n", vmcs, phys_addr);
372 }
373 
374 noinline void vmptrld_error(struct vmcs *vmcs, u64 phys_addr)
375 {
376 	vmx_insn_failed("kvm: vmptrld failed: %p/%llx\n", vmcs, phys_addr);
377 }
378 
379 noinline void invvpid_error(unsigned long ext, u16 vpid, gva_t gva)
380 {
381 	vmx_insn_failed("kvm: invvpid failed: ext=0x%lx vpid=%u gva=0x%lx\n",
382 			ext, vpid, gva);
383 }
384 
385 noinline void invept_error(unsigned long ext, u64 eptp, gpa_t gpa)
386 {
387 	vmx_insn_failed("kvm: invept failed: ext=0x%lx eptp=%llx gpa=0x%llx\n",
388 			ext, eptp, gpa);
389 }
390 
391 static DEFINE_PER_CPU(struct vmcs *, vmxarea);
392 DEFINE_PER_CPU(struct vmcs *, current_vmcs);
393 /*
394  * We maintain a per-CPU linked-list of VMCS loaded on that CPU. This is needed
395  * when a CPU is brought down, and we need to VMCLEAR all VMCSs loaded on it.
396  */
397 static DEFINE_PER_CPU(struct list_head, loaded_vmcss_on_cpu);
398 
399 /*
400  * We maintian a per-CPU linked-list of vCPU, so in wakeup_handler() we
401  * can find which vCPU should be waken up.
402  */
403 static DEFINE_PER_CPU(struct list_head, blocked_vcpu_on_cpu);
404 static DEFINE_PER_CPU(spinlock_t, blocked_vcpu_on_cpu_lock);
405 
406 static DECLARE_BITMAP(vmx_vpid_bitmap, VMX_NR_VPIDS);
407 static DEFINE_SPINLOCK(vmx_vpid_lock);
408 
409 struct vmcs_config vmcs_config;
410 struct vmx_capability vmx_capability;
411 
412 #define VMX_SEGMENT_FIELD(seg)					\
413 	[VCPU_SREG_##seg] = {                                   \
414 		.selector = GUEST_##seg##_SELECTOR,		\
415 		.base = GUEST_##seg##_BASE,		   	\
416 		.limit = GUEST_##seg##_LIMIT,		   	\
417 		.ar_bytes = GUEST_##seg##_AR_BYTES,	   	\
418 	}
419 
420 static const struct kvm_vmx_segment_field {
421 	unsigned selector;
422 	unsigned base;
423 	unsigned limit;
424 	unsigned ar_bytes;
425 } kvm_vmx_segment_fields[] = {
426 	VMX_SEGMENT_FIELD(CS),
427 	VMX_SEGMENT_FIELD(DS),
428 	VMX_SEGMENT_FIELD(ES),
429 	VMX_SEGMENT_FIELD(FS),
430 	VMX_SEGMENT_FIELD(GS),
431 	VMX_SEGMENT_FIELD(SS),
432 	VMX_SEGMENT_FIELD(TR),
433 	VMX_SEGMENT_FIELD(LDTR),
434 };
435 
436 u64 host_efer;
437 static unsigned long host_idt_base;
438 
439 /*
440  * Though SYSCALL is only supported in 64-bit mode on Intel CPUs, kvm
441  * will emulate SYSCALL in legacy mode if the vendor string in guest
442  * CPUID.0:{EBX,ECX,EDX} is "AuthenticAMD" or "AMDisbetter!" To
443  * support this emulation, IA32_STAR must always be included in
444  * vmx_msr_index[], even in i386 builds.
445  */
446 const u32 vmx_msr_index[] = {
447 #ifdef CONFIG_X86_64
448 	MSR_SYSCALL_MASK, MSR_LSTAR, MSR_CSTAR,
449 #endif
450 	MSR_EFER, MSR_TSC_AUX, MSR_STAR,
451 	MSR_IA32_TSX_CTRL,
452 };
453 
454 #if IS_ENABLED(CONFIG_HYPERV)
455 static bool __read_mostly enlightened_vmcs = true;
456 module_param(enlightened_vmcs, bool, 0444);
457 
458 /* check_ept_pointer() should be under protection of ept_pointer_lock. */
459 static void check_ept_pointer_match(struct kvm *kvm)
460 {
461 	struct kvm_vcpu *vcpu;
462 	u64 tmp_eptp = INVALID_PAGE;
463 	int i;
464 
465 	kvm_for_each_vcpu(i, vcpu, kvm) {
466 		if (!VALID_PAGE(tmp_eptp)) {
467 			tmp_eptp = to_vmx(vcpu)->ept_pointer;
468 		} else if (tmp_eptp != to_vmx(vcpu)->ept_pointer) {
469 			to_kvm_vmx(kvm)->ept_pointers_match
470 				= EPT_POINTERS_MISMATCH;
471 			return;
472 		}
473 	}
474 
475 	to_kvm_vmx(kvm)->ept_pointers_match = EPT_POINTERS_MATCH;
476 }
477 
478 static int kvm_fill_hv_flush_list_func(struct hv_guest_mapping_flush_list *flush,
479 		void *data)
480 {
481 	struct kvm_tlb_range *range = data;
482 
483 	return hyperv_fill_flush_guest_mapping_list(flush, range->start_gfn,
484 			range->pages);
485 }
486 
487 static inline int __hv_remote_flush_tlb_with_range(struct kvm *kvm,
488 		struct kvm_vcpu *vcpu, struct kvm_tlb_range *range)
489 {
490 	u64 ept_pointer = to_vmx(vcpu)->ept_pointer;
491 
492 	/*
493 	 * FLUSH_GUEST_PHYSICAL_ADDRESS_SPACE hypercall needs address
494 	 * of the base of EPT PML4 table, strip off EPT configuration
495 	 * information.
496 	 */
497 	if (range)
498 		return hyperv_flush_guest_mapping_range(ept_pointer & PAGE_MASK,
499 				kvm_fill_hv_flush_list_func, (void *)range);
500 	else
501 		return hyperv_flush_guest_mapping(ept_pointer & PAGE_MASK);
502 }
503 
504 static int hv_remote_flush_tlb_with_range(struct kvm *kvm,
505 		struct kvm_tlb_range *range)
506 {
507 	struct kvm_vcpu *vcpu;
508 	int ret = 0, i;
509 
510 	spin_lock(&to_kvm_vmx(kvm)->ept_pointer_lock);
511 
512 	if (to_kvm_vmx(kvm)->ept_pointers_match == EPT_POINTERS_CHECK)
513 		check_ept_pointer_match(kvm);
514 
515 	if (to_kvm_vmx(kvm)->ept_pointers_match != EPT_POINTERS_MATCH) {
516 		kvm_for_each_vcpu(i, vcpu, kvm) {
517 			/* If ept_pointer is invalid pointer, bypass flush request. */
518 			if (VALID_PAGE(to_vmx(vcpu)->ept_pointer))
519 				ret |= __hv_remote_flush_tlb_with_range(
520 					kvm, vcpu, range);
521 		}
522 	} else {
523 		ret = __hv_remote_flush_tlb_with_range(kvm,
524 				kvm_get_vcpu(kvm, 0), range);
525 	}
526 
527 	spin_unlock(&to_kvm_vmx(kvm)->ept_pointer_lock);
528 	return ret;
529 }
530 static int hv_remote_flush_tlb(struct kvm *kvm)
531 {
532 	return hv_remote_flush_tlb_with_range(kvm, NULL);
533 }
534 
535 static int hv_enable_direct_tlbflush(struct kvm_vcpu *vcpu)
536 {
537 	struct hv_enlightened_vmcs *evmcs;
538 	struct hv_partition_assist_pg **p_hv_pa_pg =
539 			&vcpu->kvm->arch.hyperv.hv_pa_pg;
540 	/*
541 	 * Synthetic VM-Exit is not enabled in current code and so All
542 	 * evmcs in singe VM shares same assist page.
543 	 */
544 	if (!*p_hv_pa_pg)
545 		*p_hv_pa_pg = kzalloc(PAGE_SIZE, GFP_KERNEL);
546 
547 	if (!*p_hv_pa_pg)
548 		return -ENOMEM;
549 
550 	evmcs = (struct hv_enlightened_vmcs *)to_vmx(vcpu)->loaded_vmcs->vmcs;
551 
552 	evmcs->partition_assist_page =
553 		__pa(*p_hv_pa_pg);
554 	evmcs->hv_vm_id = (unsigned long)vcpu->kvm;
555 	evmcs->hv_enlightenments_control.nested_flush_hypercall = 1;
556 
557 	return 0;
558 }
559 
560 #endif /* IS_ENABLED(CONFIG_HYPERV) */
561 
562 /*
563  * Comment's format: document - errata name - stepping - processor name.
564  * Refer from
565  * https://www.virtualbox.org/svn/vbox/trunk/src/VBox/VMM/VMMR0/HMR0.cpp
566  */
567 static u32 vmx_preemption_cpu_tfms[] = {
568 /* 323344.pdf - BA86   - D0 - Xeon 7500 Series */
569 0x000206E6,
570 /* 323056.pdf - AAX65  - C2 - Xeon L3406 */
571 /* 322814.pdf - AAT59  - C2 - i7-600, i5-500, i5-400 and i3-300 Mobile */
572 /* 322911.pdf - AAU65  - C2 - i5-600, i3-500 Desktop and Pentium G6950 */
573 0x00020652,
574 /* 322911.pdf - AAU65  - K0 - i5-600, i3-500 Desktop and Pentium G6950 */
575 0x00020655,
576 /* 322373.pdf - AAO95  - B1 - Xeon 3400 Series */
577 /* 322166.pdf - AAN92  - B1 - i7-800 and i5-700 Desktop */
578 /*
579  * 320767.pdf - AAP86  - B1 -
580  * i7-900 Mobile Extreme, i7-800 and i7-700 Mobile
581  */
582 0x000106E5,
583 /* 321333.pdf - AAM126 - C0 - Xeon 3500 */
584 0x000106A0,
585 /* 321333.pdf - AAM126 - C1 - Xeon 3500 */
586 0x000106A1,
587 /* 320836.pdf - AAJ124 - C0 - i7-900 Desktop Extreme and i7-900 Desktop */
588 0x000106A4,
589  /* 321333.pdf - AAM126 - D0 - Xeon 3500 */
590  /* 321324.pdf - AAK139 - D0 - Xeon 5500 */
591  /* 320836.pdf - AAJ124 - D0 - i7-900 Extreme and i7-900 Desktop */
592 0x000106A5,
593  /* Xeon E3-1220 V2 */
594 0x000306A8,
595 };
596 
597 static inline bool cpu_has_broken_vmx_preemption_timer(void)
598 {
599 	u32 eax = cpuid_eax(0x00000001), i;
600 
601 	/* Clear the reserved bits */
602 	eax &= ~(0x3U << 14 | 0xfU << 28);
603 	for (i = 0; i < ARRAY_SIZE(vmx_preemption_cpu_tfms); i++)
604 		if (eax == vmx_preemption_cpu_tfms[i])
605 			return true;
606 
607 	return false;
608 }
609 
610 static inline bool cpu_need_virtualize_apic_accesses(struct kvm_vcpu *vcpu)
611 {
612 	return flexpriority_enabled && lapic_in_kernel(vcpu);
613 }
614 
615 static inline bool report_flexpriority(void)
616 {
617 	return flexpriority_enabled;
618 }
619 
620 static inline int __find_msr_index(struct vcpu_vmx *vmx, u32 msr)
621 {
622 	int i;
623 
624 	for (i = 0; i < vmx->nmsrs; ++i)
625 		if (vmx_msr_index[vmx->guest_msrs[i].index] == msr)
626 			return i;
627 	return -1;
628 }
629 
630 struct shared_msr_entry *find_msr_entry(struct vcpu_vmx *vmx, u32 msr)
631 {
632 	int i;
633 
634 	i = __find_msr_index(vmx, msr);
635 	if (i >= 0)
636 		return &vmx->guest_msrs[i];
637 	return NULL;
638 }
639 
640 static int vmx_set_guest_msr(struct vcpu_vmx *vmx, struct shared_msr_entry *msr, u64 data)
641 {
642 	int ret = 0;
643 
644 	u64 old_msr_data = msr->data;
645 	msr->data = data;
646 	if (msr - vmx->guest_msrs < vmx->save_nmsrs) {
647 		preempt_disable();
648 		ret = kvm_set_shared_msr(msr->index, msr->data,
649 					 msr->mask);
650 		preempt_enable();
651 		if (ret)
652 			msr->data = old_msr_data;
653 	}
654 	return ret;
655 }
656 
657 void loaded_vmcs_init(struct loaded_vmcs *loaded_vmcs)
658 {
659 	vmcs_clear(loaded_vmcs->vmcs);
660 	if (loaded_vmcs->shadow_vmcs && loaded_vmcs->launched)
661 		vmcs_clear(loaded_vmcs->shadow_vmcs);
662 	loaded_vmcs->cpu = -1;
663 	loaded_vmcs->launched = 0;
664 }
665 
666 #ifdef CONFIG_KEXEC_CORE
667 /*
668  * This bitmap is used to indicate whether the vmclear
669  * operation is enabled on all cpus. All disabled by
670  * default.
671  */
672 static cpumask_t crash_vmclear_enabled_bitmap = CPU_MASK_NONE;
673 
674 static inline void crash_enable_local_vmclear(int cpu)
675 {
676 	cpumask_set_cpu(cpu, &crash_vmclear_enabled_bitmap);
677 }
678 
679 static inline void crash_disable_local_vmclear(int cpu)
680 {
681 	cpumask_clear_cpu(cpu, &crash_vmclear_enabled_bitmap);
682 }
683 
684 static inline int crash_local_vmclear_enabled(int cpu)
685 {
686 	return cpumask_test_cpu(cpu, &crash_vmclear_enabled_bitmap);
687 }
688 
689 static void crash_vmclear_local_loaded_vmcss(void)
690 {
691 	int cpu = raw_smp_processor_id();
692 	struct loaded_vmcs *v;
693 
694 	if (!crash_local_vmclear_enabled(cpu))
695 		return;
696 
697 	list_for_each_entry(v, &per_cpu(loaded_vmcss_on_cpu, cpu),
698 			    loaded_vmcss_on_cpu_link)
699 		vmcs_clear(v->vmcs);
700 }
701 #else
702 static inline void crash_enable_local_vmclear(int cpu) { }
703 static inline void crash_disable_local_vmclear(int cpu) { }
704 #endif /* CONFIG_KEXEC_CORE */
705 
706 static void __loaded_vmcs_clear(void *arg)
707 {
708 	struct loaded_vmcs *loaded_vmcs = arg;
709 	int cpu = raw_smp_processor_id();
710 
711 	if (loaded_vmcs->cpu != cpu)
712 		return; /* vcpu migration can race with cpu offline */
713 	if (per_cpu(current_vmcs, cpu) == loaded_vmcs->vmcs)
714 		per_cpu(current_vmcs, cpu) = NULL;
715 	crash_disable_local_vmclear(cpu);
716 	list_del(&loaded_vmcs->loaded_vmcss_on_cpu_link);
717 
718 	/*
719 	 * we should ensure updating loaded_vmcs->loaded_vmcss_on_cpu_link
720 	 * is before setting loaded_vmcs->vcpu to -1 which is done in
721 	 * loaded_vmcs_init. Otherwise, other cpu can see vcpu = -1 fist
722 	 * then adds the vmcs into percpu list before it is deleted.
723 	 */
724 	smp_wmb();
725 
726 	loaded_vmcs_init(loaded_vmcs);
727 	crash_enable_local_vmclear(cpu);
728 }
729 
730 void loaded_vmcs_clear(struct loaded_vmcs *loaded_vmcs)
731 {
732 	int cpu = loaded_vmcs->cpu;
733 
734 	if (cpu != -1)
735 		smp_call_function_single(cpu,
736 			 __loaded_vmcs_clear, loaded_vmcs, 1);
737 }
738 
739 static bool vmx_segment_cache_test_set(struct vcpu_vmx *vmx, unsigned seg,
740 				       unsigned field)
741 {
742 	bool ret;
743 	u32 mask = 1 << (seg * SEG_FIELD_NR + field);
744 
745 	if (!kvm_register_is_available(&vmx->vcpu, VCPU_EXREG_SEGMENTS)) {
746 		kvm_register_mark_available(&vmx->vcpu, VCPU_EXREG_SEGMENTS);
747 		vmx->segment_cache.bitmask = 0;
748 	}
749 	ret = vmx->segment_cache.bitmask & mask;
750 	vmx->segment_cache.bitmask |= mask;
751 	return ret;
752 }
753 
754 static u16 vmx_read_guest_seg_selector(struct vcpu_vmx *vmx, unsigned seg)
755 {
756 	u16 *p = &vmx->segment_cache.seg[seg].selector;
757 
758 	if (!vmx_segment_cache_test_set(vmx, seg, SEG_FIELD_SEL))
759 		*p = vmcs_read16(kvm_vmx_segment_fields[seg].selector);
760 	return *p;
761 }
762 
763 static ulong vmx_read_guest_seg_base(struct vcpu_vmx *vmx, unsigned seg)
764 {
765 	ulong *p = &vmx->segment_cache.seg[seg].base;
766 
767 	if (!vmx_segment_cache_test_set(vmx, seg, SEG_FIELD_BASE))
768 		*p = vmcs_readl(kvm_vmx_segment_fields[seg].base);
769 	return *p;
770 }
771 
772 static u32 vmx_read_guest_seg_limit(struct vcpu_vmx *vmx, unsigned seg)
773 {
774 	u32 *p = &vmx->segment_cache.seg[seg].limit;
775 
776 	if (!vmx_segment_cache_test_set(vmx, seg, SEG_FIELD_LIMIT))
777 		*p = vmcs_read32(kvm_vmx_segment_fields[seg].limit);
778 	return *p;
779 }
780 
781 static u32 vmx_read_guest_seg_ar(struct vcpu_vmx *vmx, unsigned seg)
782 {
783 	u32 *p = &vmx->segment_cache.seg[seg].ar;
784 
785 	if (!vmx_segment_cache_test_set(vmx, seg, SEG_FIELD_AR))
786 		*p = vmcs_read32(kvm_vmx_segment_fields[seg].ar_bytes);
787 	return *p;
788 }
789 
790 void update_exception_bitmap(struct kvm_vcpu *vcpu)
791 {
792 	u32 eb;
793 
794 	eb = (1u << PF_VECTOR) | (1u << UD_VECTOR) | (1u << MC_VECTOR) |
795 	     (1u << DB_VECTOR) | (1u << AC_VECTOR);
796 	/*
797 	 * Guest access to VMware backdoor ports could legitimately
798 	 * trigger #GP because of TSS I/O permission bitmap.
799 	 * We intercept those #GP and allow access to them anyway
800 	 * as VMware does.
801 	 */
802 	if (enable_vmware_backdoor)
803 		eb |= (1u << GP_VECTOR);
804 	if ((vcpu->guest_debug &
805 	     (KVM_GUESTDBG_ENABLE | KVM_GUESTDBG_USE_SW_BP)) ==
806 	    (KVM_GUESTDBG_ENABLE | KVM_GUESTDBG_USE_SW_BP))
807 		eb |= 1u << BP_VECTOR;
808 	if (to_vmx(vcpu)->rmode.vm86_active)
809 		eb = ~0;
810 	if (enable_ept)
811 		eb &= ~(1u << PF_VECTOR); /* bypass_guest_pf = 0 */
812 
813 	/* When we are running a nested L2 guest and L1 specified for it a
814 	 * certain exception bitmap, we must trap the same exceptions and pass
815 	 * them to L1. When running L2, we will only handle the exceptions
816 	 * specified above if L1 did not want them.
817 	 */
818 	if (is_guest_mode(vcpu))
819 		eb |= get_vmcs12(vcpu)->exception_bitmap;
820 
821 	vmcs_write32(EXCEPTION_BITMAP, eb);
822 }
823 
824 /*
825  * Check if MSR is intercepted for currently loaded MSR bitmap.
826  */
827 static bool msr_write_intercepted(struct kvm_vcpu *vcpu, u32 msr)
828 {
829 	unsigned long *msr_bitmap;
830 	int f = sizeof(unsigned long);
831 
832 	if (!cpu_has_vmx_msr_bitmap())
833 		return true;
834 
835 	msr_bitmap = to_vmx(vcpu)->loaded_vmcs->msr_bitmap;
836 
837 	if (msr <= 0x1fff) {
838 		return !!test_bit(msr, msr_bitmap + 0x800 / f);
839 	} else if ((msr >= 0xc0000000) && (msr <= 0xc0001fff)) {
840 		msr &= 0x1fff;
841 		return !!test_bit(msr, msr_bitmap + 0xc00 / f);
842 	}
843 
844 	return true;
845 }
846 
847 static void clear_atomic_switch_msr_special(struct vcpu_vmx *vmx,
848 		unsigned long entry, unsigned long exit)
849 {
850 	vm_entry_controls_clearbit(vmx, entry);
851 	vm_exit_controls_clearbit(vmx, exit);
852 }
853 
854 int vmx_find_msr_index(struct vmx_msrs *m, u32 msr)
855 {
856 	unsigned int i;
857 
858 	for (i = 0; i < m->nr; ++i) {
859 		if (m->val[i].index == msr)
860 			return i;
861 	}
862 	return -ENOENT;
863 }
864 
865 static void clear_atomic_switch_msr(struct vcpu_vmx *vmx, unsigned msr)
866 {
867 	int i;
868 	struct msr_autoload *m = &vmx->msr_autoload;
869 
870 	switch (msr) {
871 	case MSR_EFER:
872 		if (cpu_has_load_ia32_efer()) {
873 			clear_atomic_switch_msr_special(vmx,
874 					VM_ENTRY_LOAD_IA32_EFER,
875 					VM_EXIT_LOAD_IA32_EFER);
876 			return;
877 		}
878 		break;
879 	case MSR_CORE_PERF_GLOBAL_CTRL:
880 		if (cpu_has_load_perf_global_ctrl()) {
881 			clear_atomic_switch_msr_special(vmx,
882 					VM_ENTRY_LOAD_IA32_PERF_GLOBAL_CTRL,
883 					VM_EXIT_LOAD_IA32_PERF_GLOBAL_CTRL);
884 			return;
885 		}
886 		break;
887 	}
888 	i = vmx_find_msr_index(&m->guest, msr);
889 	if (i < 0)
890 		goto skip_guest;
891 	--m->guest.nr;
892 	m->guest.val[i] = m->guest.val[m->guest.nr];
893 	vmcs_write32(VM_ENTRY_MSR_LOAD_COUNT, m->guest.nr);
894 
895 skip_guest:
896 	i = vmx_find_msr_index(&m->host, msr);
897 	if (i < 0)
898 		return;
899 
900 	--m->host.nr;
901 	m->host.val[i] = m->host.val[m->host.nr];
902 	vmcs_write32(VM_EXIT_MSR_LOAD_COUNT, m->host.nr);
903 }
904 
905 static void add_atomic_switch_msr_special(struct vcpu_vmx *vmx,
906 		unsigned long entry, unsigned long exit,
907 		unsigned long guest_val_vmcs, unsigned long host_val_vmcs,
908 		u64 guest_val, u64 host_val)
909 {
910 	vmcs_write64(guest_val_vmcs, guest_val);
911 	if (host_val_vmcs != HOST_IA32_EFER)
912 		vmcs_write64(host_val_vmcs, host_val);
913 	vm_entry_controls_setbit(vmx, entry);
914 	vm_exit_controls_setbit(vmx, exit);
915 }
916 
917 static void add_atomic_switch_msr(struct vcpu_vmx *vmx, unsigned msr,
918 				  u64 guest_val, u64 host_val, bool entry_only)
919 {
920 	int i, j = 0;
921 	struct msr_autoload *m = &vmx->msr_autoload;
922 
923 	switch (msr) {
924 	case MSR_EFER:
925 		if (cpu_has_load_ia32_efer()) {
926 			add_atomic_switch_msr_special(vmx,
927 					VM_ENTRY_LOAD_IA32_EFER,
928 					VM_EXIT_LOAD_IA32_EFER,
929 					GUEST_IA32_EFER,
930 					HOST_IA32_EFER,
931 					guest_val, host_val);
932 			return;
933 		}
934 		break;
935 	case MSR_CORE_PERF_GLOBAL_CTRL:
936 		if (cpu_has_load_perf_global_ctrl()) {
937 			add_atomic_switch_msr_special(vmx,
938 					VM_ENTRY_LOAD_IA32_PERF_GLOBAL_CTRL,
939 					VM_EXIT_LOAD_IA32_PERF_GLOBAL_CTRL,
940 					GUEST_IA32_PERF_GLOBAL_CTRL,
941 					HOST_IA32_PERF_GLOBAL_CTRL,
942 					guest_val, host_val);
943 			return;
944 		}
945 		break;
946 	case MSR_IA32_PEBS_ENABLE:
947 		/* PEBS needs a quiescent period after being disabled (to write
948 		 * a record).  Disabling PEBS through VMX MSR swapping doesn't
949 		 * provide that period, so a CPU could write host's record into
950 		 * guest's memory.
951 		 */
952 		wrmsrl(MSR_IA32_PEBS_ENABLE, 0);
953 	}
954 
955 	i = vmx_find_msr_index(&m->guest, msr);
956 	if (!entry_only)
957 		j = vmx_find_msr_index(&m->host, msr);
958 
959 	if ((i < 0 && m->guest.nr == NR_LOADSTORE_MSRS) ||
960 		(j < 0 &&  m->host.nr == NR_LOADSTORE_MSRS)) {
961 		printk_once(KERN_WARNING "Not enough msr switch entries. "
962 				"Can't add msr %x\n", msr);
963 		return;
964 	}
965 	if (i < 0) {
966 		i = m->guest.nr++;
967 		vmcs_write32(VM_ENTRY_MSR_LOAD_COUNT, m->guest.nr);
968 	}
969 	m->guest.val[i].index = msr;
970 	m->guest.val[i].value = guest_val;
971 
972 	if (entry_only)
973 		return;
974 
975 	if (j < 0) {
976 		j = m->host.nr++;
977 		vmcs_write32(VM_EXIT_MSR_LOAD_COUNT, m->host.nr);
978 	}
979 	m->host.val[j].index = msr;
980 	m->host.val[j].value = host_val;
981 }
982 
983 static bool update_transition_efer(struct vcpu_vmx *vmx, int efer_offset)
984 {
985 	u64 guest_efer = vmx->vcpu.arch.efer;
986 	u64 ignore_bits = 0;
987 
988 	/* Shadow paging assumes NX to be available.  */
989 	if (!enable_ept)
990 		guest_efer |= EFER_NX;
991 
992 	/*
993 	 * LMA and LME handled by hardware; SCE meaningless outside long mode.
994 	 */
995 	ignore_bits |= EFER_SCE;
996 #ifdef CONFIG_X86_64
997 	ignore_bits |= EFER_LMA | EFER_LME;
998 	/* SCE is meaningful only in long mode on Intel */
999 	if (guest_efer & EFER_LMA)
1000 		ignore_bits &= ~(u64)EFER_SCE;
1001 #endif
1002 
1003 	/*
1004 	 * On EPT, we can't emulate NX, so we must switch EFER atomically.
1005 	 * On CPUs that support "load IA32_EFER", always switch EFER
1006 	 * atomically, since it's faster than switching it manually.
1007 	 */
1008 	if (cpu_has_load_ia32_efer() ||
1009 	    (enable_ept && ((vmx->vcpu.arch.efer ^ host_efer) & EFER_NX))) {
1010 		if (!(guest_efer & EFER_LMA))
1011 			guest_efer &= ~EFER_LME;
1012 		if (guest_efer != host_efer)
1013 			add_atomic_switch_msr(vmx, MSR_EFER,
1014 					      guest_efer, host_efer, false);
1015 		else
1016 			clear_atomic_switch_msr(vmx, MSR_EFER);
1017 		return false;
1018 	} else {
1019 		clear_atomic_switch_msr(vmx, MSR_EFER);
1020 
1021 		guest_efer &= ~ignore_bits;
1022 		guest_efer |= host_efer & ignore_bits;
1023 
1024 		vmx->guest_msrs[efer_offset].data = guest_efer;
1025 		vmx->guest_msrs[efer_offset].mask = ~ignore_bits;
1026 
1027 		return true;
1028 	}
1029 }
1030 
1031 #ifdef CONFIG_X86_32
1032 /*
1033  * On 32-bit kernels, VM exits still load the FS and GS bases from the
1034  * VMCS rather than the segment table.  KVM uses this helper to figure
1035  * out the current bases to poke them into the VMCS before entry.
1036  */
1037 static unsigned long segment_base(u16 selector)
1038 {
1039 	struct desc_struct *table;
1040 	unsigned long v;
1041 
1042 	if (!(selector & ~SEGMENT_RPL_MASK))
1043 		return 0;
1044 
1045 	table = get_current_gdt_ro();
1046 
1047 	if ((selector & SEGMENT_TI_MASK) == SEGMENT_LDT) {
1048 		u16 ldt_selector = kvm_read_ldt();
1049 
1050 		if (!(ldt_selector & ~SEGMENT_RPL_MASK))
1051 			return 0;
1052 
1053 		table = (struct desc_struct *)segment_base(ldt_selector);
1054 	}
1055 	v = get_desc_base(&table[selector >> 3]);
1056 	return v;
1057 }
1058 #endif
1059 
1060 static inline void pt_load_msr(struct pt_ctx *ctx, u32 addr_range)
1061 {
1062 	u32 i;
1063 
1064 	wrmsrl(MSR_IA32_RTIT_STATUS, ctx->status);
1065 	wrmsrl(MSR_IA32_RTIT_OUTPUT_BASE, ctx->output_base);
1066 	wrmsrl(MSR_IA32_RTIT_OUTPUT_MASK, ctx->output_mask);
1067 	wrmsrl(MSR_IA32_RTIT_CR3_MATCH, ctx->cr3_match);
1068 	for (i = 0; i < addr_range; i++) {
1069 		wrmsrl(MSR_IA32_RTIT_ADDR0_A + i * 2, ctx->addr_a[i]);
1070 		wrmsrl(MSR_IA32_RTIT_ADDR0_B + i * 2, ctx->addr_b[i]);
1071 	}
1072 }
1073 
1074 static inline void pt_save_msr(struct pt_ctx *ctx, u32 addr_range)
1075 {
1076 	u32 i;
1077 
1078 	rdmsrl(MSR_IA32_RTIT_STATUS, ctx->status);
1079 	rdmsrl(MSR_IA32_RTIT_OUTPUT_BASE, ctx->output_base);
1080 	rdmsrl(MSR_IA32_RTIT_OUTPUT_MASK, ctx->output_mask);
1081 	rdmsrl(MSR_IA32_RTIT_CR3_MATCH, ctx->cr3_match);
1082 	for (i = 0; i < addr_range; i++) {
1083 		rdmsrl(MSR_IA32_RTIT_ADDR0_A + i * 2, ctx->addr_a[i]);
1084 		rdmsrl(MSR_IA32_RTIT_ADDR0_B + i * 2, ctx->addr_b[i]);
1085 	}
1086 }
1087 
1088 static void pt_guest_enter(struct vcpu_vmx *vmx)
1089 {
1090 	if (pt_mode == PT_MODE_SYSTEM)
1091 		return;
1092 
1093 	/*
1094 	 * GUEST_IA32_RTIT_CTL is already set in the VMCS.
1095 	 * Save host state before VM entry.
1096 	 */
1097 	rdmsrl(MSR_IA32_RTIT_CTL, vmx->pt_desc.host.ctl);
1098 	if (vmx->pt_desc.guest.ctl & RTIT_CTL_TRACEEN) {
1099 		wrmsrl(MSR_IA32_RTIT_CTL, 0);
1100 		pt_save_msr(&vmx->pt_desc.host, vmx->pt_desc.addr_range);
1101 		pt_load_msr(&vmx->pt_desc.guest, vmx->pt_desc.addr_range);
1102 	}
1103 }
1104 
1105 static void pt_guest_exit(struct vcpu_vmx *vmx)
1106 {
1107 	if (pt_mode == PT_MODE_SYSTEM)
1108 		return;
1109 
1110 	if (vmx->pt_desc.guest.ctl & RTIT_CTL_TRACEEN) {
1111 		pt_save_msr(&vmx->pt_desc.guest, vmx->pt_desc.addr_range);
1112 		pt_load_msr(&vmx->pt_desc.host, vmx->pt_desc.addr_range);
1113 	}
1114 
1115 	/* Reload host state (IA32_RTIT_CTL will be cleared on VM exit). */
1116 	wrmsrl(MSR_IA32_RTIT_CTL, vmx->pt_desc.host.ctl);
1117 }
1118 
1119 void vmx_set_host_fs_gs(struct vmcs_host_state *host, u16 fs_sel, u16 gs_sel,
1120 			unsigned long fs_base, unsigned long gs_base)
1121 {
1122 	if (unlikely(fs_sel != host->fs_sel)) {
1123 		if (!(fs_sel & 7))
1124 			vmcs_write16(HOST_FS_SELECTOR, fs_sel);
1125 		else
1126 			vmcs_write16(HOST_FS_SELECTOR, 0);
1127 		host->fs_sel = fs_sel;
1128 	}
1129 	if (unlikely(gs_sel != host->gs_sel)) {
1130 		if (!(gs_sel & 7))
1131 			vmcs_write16(HOST_GS_SELECTOR, gs_sel);
1132 		else
1133 			vmcs_write16(HOST_GS_SELECTOR, 0);
1134 		host->gs_sel = gs_sel;
1135 	}
1136 	if (unlikely(fs_base != host->fs_base)) {
1137 		vmcs_writel(HOST_FS_BASE, fs_base);
1138 		host->fs_base = fs_base;
1139 	}
1140 	if (unlikely(gs_base != host->gs_base)) {
1141 		vmcs_writel(HOST_GS_BASE, gs_base);
1142 		host->gs_base = gs_base;
1143 	}
1144 }
1145 
1146 void vmx_prepare_switch_to_guest(struct kvm_vcpu *vcpu)
1147 {
1148 	struct vcpu_vmx *vmx = to_vmx(vcpu);
1149 	struct vmcs_host_state *host_state;
1150 #ifdef CONFIG_X86_64
1151 	int cpu = raw_smp_processor_id();
1152 #endif
1153 	unsigned long fs_base, gs_base;
1154 	u16 fs_sel, gs_sel;
1155 	int i;
1156 
1157 	vmx->req_immediate_exit = false;
1158 
1159 	/*
1160 	 * Note that guest MSRs to be saved/restored can also be changed
1161 	 * when guest state is loaded. This happens when guest transitions
1162 	 * to/from long-mode by setting MSR_EFER.LMA.
1163 	 */
1164 	if (!vmx->guest_msrs_ready) {
1165 		vmx->guest_msrs_ready = true;
1166 		for (i = 0; i < vmx->save_nmsrs; ++i)
1167 			kvm_set_shared_msr(vmx->guest_msrs[i].index,
1168 					   vmx->guest_msrs[i].data,
1169 					   vmx->guest_msrs[i].mask);
1170 
1171 	}
1172 	if (vmx->guest_state_loaded)
1173 		return;
1174 
1175 	host_state = &vmx->loaded_vmcs->host_state;
1176 
1177 	/*
1178 	 * Set host fs and gs selectors.  Unfortunately, 22.2.3 does not
1179 	 * allow segment selectors with cpl > 0 or ti == 1.
1180 	 */
1181 	host_state->ldt_sel = kvm_read_ldt();
1182 
1183 #ifdef CONFIG_X86_64
1184 	savesegment(ds, host_state->ds_sel);
1185 	savesegment(es, host_state->es_sel);
1186 
1187 	gs_base = cpu_kernelmode_gs_base(cpu);
1188 	if (likely(is_64bit_mm(current->mm))) {
1189 		save_fsgs_for_kvm();
1190 		fs_sel = current->thread.fsindex;
1191 		gs_sel = current->thread.gsindex;
1192 		fs_base = current->thread.fsbase;
1193 		vmx->msr_host_kernel_gs_base = current->thread.gsbase;
1194 	} else {
1195 		savesegment(fs, fs_sel);
1196 		savesegment(gs, gs_sel);
1197 		fs_base = read_msr(MSR_FS_BASE);
1198 		vmx->msr_host_kernel_gs_base = read_msr(MSR_KERNEL_GS_BASE);
1199 	}
1200 
1201 	wrmsrl(MSR_KERNEL_GS_BASE, vmx->msr_guest_kernel_gs_base);
1202 #else
1203 	savesegment(fs, fs_sel);
1204 	savesegment(gs, gs_sel);
1205 	fs_base = segment_base(fs_sel);
1206 	gs_base = segment_base(gs_sel);
1207 #endif
1208 
1209 	vmx_set_host_fs_gs(host_state, fs_sel, gs_sel, fs_base, gs_base);
1210 	vmx->guest_state_loaded = true;
1211 }
1212 
1213 static void vmx_prepare_switch_to_host(struct vcpu_vmx *vmx)
1214 {
1215 	struct vmcs_host_state *host_state;
1216 
1217 	if (!vmx->guest_state_loaded)
1218 		return;
1219 
1220 	host_state = &vmx->loaded_vmcs->host_state;
1221 
1222 	++vmx->vcpu.stat.host_state_reload;
1223 
1224 #ifdef CONFIG_X86_64
1225 	rdmsrl(MSR_KERNEL_GS_BASE, vmx->msr_guest_kernel_gs_base);
1226 #endif
1227 	if (host_state->ldt_sel || (host_state->gs_sel & 7)) {
1228 		kvm_load_ldt(host_state->ldt_sel);
1229 #ifdef CONFIG_X86_64
1230 		load_gs_index(host_state->gs_sel);
1231 #else
1232 		loadsegment(gs, host_state->gs_sel);
1233 #endif
1234 	}
1235 	if (host_state->fs_sel & 7)
1236 		loadsegment(fs, host_state->fs_sel);
1237 #ifdef CONFIG_X86_64
1238 	if (unlikely(host_state->ds_sel | host_state->es_sel)) {
1239 		loadsegment(ds, host_state->ds_sel);
1240 		loadsegment(es, host_state->es_sel);
1241 	}
1242 #endif
1243 	invalidate_tss_limit();
1244 #ifdef CONFIG_X86_64
1245 	wrmsrl(MSR_KERNEL_GS_BASE, vmx->msr_host_kernel_gs_base);
1246 #endif
1247 	load_fixmap_gdt(raw_smp_processor_id());
1248 	vmx->guest_state_loaded = false;
1249 	vmx->guest_msrs_ready = false;
1250 }
1251 
1252 #ifdef CONFIG_X86_64
1253 static u64 vmx_read_guest_kernel_gs_base(struct vcpu_vmx *vmx)
1254 {
1255 	preempt_disable();
1256 	if (vmx->guest_state_loaded)
1257 		rdmsrl(MSR_KERNEL_GS_BASE, vmx->msr_guest_kernel_gs_base);
1258 	preempt_enable();
1259 	return vmx->msr_guest_kernel_gs_base;
1260 }
1261 
1262 static void vmx_write_guest_kernel_gs_base(struct vcpu_vmx *vmx, u64 data)
1263 {
1264 	preempt_disable();
1265 	if (vmx->guest_state_loaded)
1266 		wrmsrl(MSR_KERNEL_GS_BASE, data);
1267 	preempt_enable();
1268 	vmx->msr_guest_kernel_gs_base = data;
1269 }
1270 #endif
1271 
1272 static void vmx_vcpu_pi_load(struct kvm_vcpu *vcpu, int cpu)
1273 {
1274 	struct pi_desc *pi_desc = vcpu_to_pi_desc(vcpu);
1275 	struct pi_desc old, new;
1276 	unsigned int dest;
1277 
1278 	/*
1279 	 * In case of hot-plug or hot-unplug, we may have to undo
1280 	 * vmx_vcpu_pi_put even if there is no assigned device.  And we
1281 	 * always keep PI.NDST up to date for simplicity: it makes the
1282 	 * code easier, and CPU migration is not a fast path.
1283 	 */
1284 	if (!pi_test_sn(pi_desc) && vcpu->cpu == cpu)
1285 		return;
1286 
1287 	/*
1288 	 * If the 'nv' field is POSTED_INTR_WAKEUP_VECTOR, do not change
1289 	 * PI.NDST: pi_post_block is the one expected to change PID.NDST and the
1290 	 * wakeup handler expects the vCPU to be on the blocked_vcpu_list that
1291 	 * matches PI.NDST. Otherwise, a vcpu may not be able to be woken up
1292 	 * correctly.
1293 	 */
1294 	if (pi_desc->nv == POSTED_INTR_WAKEUP_VECTOR || vcpu->cpu == cpu) {
1295 		pi_clear_sn(pi_desc);
1296 		goto after_clear_sn;
1297 	}
1298 
1299 	/* The full case.  */
1300 	do {
1301 		old.control = new.control = pi_desc->control;
1302 
1303 		dest = cpu_physical_id(cpu);
1304 
1305 		if (x2apic_enabled())
1306 			new.ndst = dest;
1307 		else
1308 			new.ndst = (dest << 8) & 0xFF00;
1309 
1310 		new.sn = 0;
1311 	} while (cmpxchg64(&pi_desc->control, old.control,
1312 			   new.control) != old.control);
1313 
1314 after_clear_sn:
1315 
1316 	/*
1317 	 * Clear SN before reading the bitmap.  The VT-d firmware
1318 	 * writes the bitmap and reads SN atomically (5.2.3 in the
1319 	 * spec), so it doesn't really have a memory barrier that
1320 	 * pairs with this, but we cannot do that and we need one.
1321 	 */
1322 	smp_mb__after_atomic();
1323 
1324 	if (!pi_is_pir_empty(pi_desc))
1325 		pi_set_on(pi_desc);
1326 }
1327 
1328 void vmx_vcpu_load_vmcs(struct kvm_vcpu *vcpu, int cpu)
1329 {
1330 	struct vcpu_vmx *vmx = to_vmx(vcpu);
1331 	bool already_loaded = vmx->loaded_vmcs->cpu == cpu;
1332 
1333 	if (!already_loaded) {
1334 		loaded_vmcs_clear(vmx->loaded_vmcs);
1335 		local_irq_disable();
1336 		crash_disable_local_vmclear(cpu);
1337 
1338 		/*
1339 		 * Read loaded_vmcs->cpu should be before fetching
1340 		 * loaded_vmcs->loaded_vmcss_on_cpu_link.
1341 		 * See the comments in __loaded_vmcs_clear().
1342 		 */
1343 		smp_rmb();
1344 
1345 		list_add(&vmx->loaded_vmcs->loaded_vmcss_on_cpu_link,
1346 			 &per_cpu(loaded_vmcss_on_cpu, cpu));
1347 		crash_enable_local_vmclear(cpu);
1348 		local_irq_enable();
1349 	}
1350 
1351 	if (per_cpu(current_vmcs, cpu) != vmx->loaded_vmcs->vmcs) {
1352 		per_cpu(current_vmcs, cpu) = vmx->loaded_vmcs->vmcs;
1353 		vmcs_load(vmx->loaded_vmcs->vmcs);
1354 		indirect_branch_prediction_barrier();
1355 	}
1356 
1357 	if (!already_loaded) {
1358 		void *gdt = get_current_gdt_ro();
1359 		unsigned long sysenter_esp;
1360 
1361 		kvm_make_request(KVM_REQ_TLB_FLUSH, vcpu);
1362 
1363 		/*
1364 		 * Linux uses per-cpu TSS and GDT, so set these when switching
1365 		 * processors.  See 22.2.4.
1366 		 */
1367 		vmcs_writel(HOST_TR_BASE,
1368 			    (unsigned long)&get_cpu_entry_area(cpu)->tss.x86_tss);
1369 		vmcs_writel(HOST_GDTR_BASE, (unsigned long)gdt);   /* 22.2.4 */
1370 
1371 		rdmsrl(MSR_IA32_SYSENTER_ESP, sysenter_esp);
1372 		vmcs_writel(HOST_IA32_SYSENTER_ESP, sysenter_esp); /* 22.2.3 */
1373 
1374 		vmx->loaded_vmcs->cpu = cpu;
1375 	}
1376 
1377 	/* Setup TSC multiplier */
1378 	if (kvm_has_tsc_control &&
1379 	    vmx->current_tsc_ratio != vcpu->arch.tsc_scaling_ratio)
1380 		decache_tsc_multiplier(vmx);
1381 }
1382 
1383 /*
1384  * Switches to specified vcpu, until a matching vcpu_put(), but assumes
1385  * vcpu mutex is already taken.
1386  */
1387 void vmx_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
1388 {
1389 	struct vcpu_vmx *vmx = to_vmx(vcpu);
1390 
1391 	vmx_vcpu_load_vmcs(vcpu, cpu);
1392 
1393 	vmx_vcpu_pi_load(vcpu, cpu);
1394 
1395 	vmx->host_pkru = read_pkru();
1396 	vmx->host_debugctlmsr = get_debugctlmsr();
1397 }
1398 
1399 static void vmx_vcpu_pi_put(struct kvm_vcpu *vcpu)
1400 {
1401 	struct pi_desc *pi_desc = vcpu_to_pi_desc(vcpu);
1402 
1403 	if (!kvm_arch_has_assigned_device(vcpu->kvm) ||
1404 		!irq_remapping_cap(IRQ_POSTING_CAP)  ||
1405 		!kvm_vcpu_apicv_active(vcpu))
1406 		return;
1407 
1408 	/* Set SN when the vCPU is preempted */
1409 	if (vcpu->preempted)
1410 		pi_set_sn(pi_desc);
1411 }
1412 
1413 static void vmx_vcpu_put(struct kvm_vcpu *vcpu)
1414 {
1415 	vmx_vcpu_pi_put(vcpu);
1416 
1417 	vmx_prepare_switch_to_host(to_vmx(vcpu));
1418 }
1419 
1420 static bool emulation_required(struct kvm_vcpu *vcpu)
1421 {
1422 	return emulate_invalid_guest_state && !guest_state_valid(vcpu);
1423 }
1424 
1425 static void vmx_decache_cr0_guest_bits(struct kvm_vcpu *vcpu);
1426 
1427 unsigned long vmx_get_rflags(struct kvm_vcpu *vcpu)
1428 {
1429 	struct vcpu_vmx *vmx = to_vmx(vcpu);
1430 	unsigned long rflags, save_rflags;
1431 
1432 	if (!kvm_register_is_available(vcpu, VCPU_EXREG_RFLAGS)) {
1433 		kvm_register_mark_available(vcpu, VCPU_EXREG_RFLAGS);
1434 		rflags = vmcs_readl(GUEST_RFLAGS);
1435 		if (vmx->rmode.vm86_active) {
1436 			rflags &= RMODE_GUEST_OWNED_EFLAGS_BITS;
1437 			save_rflags = vmx->rmode.save_rflags;
1438 			rflags |= save_rflags & ~RMODE_GUEST_OWNED_EFLAGS_BITS;
1439 		}
1440 		vmx->rflags = rflags;
1441 	}
1442 	return vmx->rflags;
1443 }
1444 
1445 void vmx_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags)
1446 {
1447 	struct vcpu_vmx *vmx = to_vmx(vcpu);
1448 	unsigned long old_rflags;
1449 
1450 	if (enable_unrestricted_guest) {
1451 		kvm_register_mark_available(vcpu, VCPU_EXREG_RFLAGS);
1452 		vmx->rflags = rflags;
1453 		vmcs_writel(GUEST_RFLAGS, rflags);
1454 		return;
1455 	}
1456 
1457 	old_rflags = vmx_get_rflags(vcpu);
1458 	vmx->rflags = rflags;
1459 	if (vmx->rmode.vm86_active) {
1460 		vmx->rmode.save_rflags = rflags;
1461 		rflags |= X86_EFLAGS_IOPL | X86_EFLAGS_VM;
1462 	}
1463 	vmcs_writel(GUEST_RFLAGS, rflags);
1464 
1465 	if ((old_rflags ^ vmx->rflags) & X86_EFLAGS_VM)
1466 		vmx->emulation_required = emulation_required(vcpu);
1467 }
1468 
1469 u32 vmx_get_interrupt_shadow(struct kvm_vcpu *vcpu)
1470 {
1471 	u32 interruptibility = vmcs_read32(GUEST_INTERRUPTIBILITY_INFO);
1472 	int ret = 0;
1473 
1474 	if (interruptibility & GUEST_INTR_STATE_STI)
1475 		ret |= KVM_X86_SHADOW_INT_STI;
1476 	if (interruptibility & GUEST_INTR_STATE_MOV_SS)
1477 		ret |= KVM_X86_SHADOW_INT_MOV_SS;
1478 
1479 	return ret;
1480 }
1481 
1482 void vmx_set_interrupt_shadow(struct kvm_vcpu *vcpu, int mask)
1483 {
1484 	u32 interruptibility_old = vmcs_read32(GUEST_INTERRUPTIBILITY_INFO);
1485 	u32 interruptibility = interruptibility_old;
1486 
1487 	interruptibility &= ~(GUEST_INTR_STATE_STI | GUEST_INTR_STATE_MOV_SS);
1488 
1489 	if (mask & KVM_X86_SHADOW_INT_MOV_SS)
1490 		interruptibility |= GUEST_INTR_STATE_MOV_SS;
1491 	else if (mask & KVM_X86_SHADOW_INT_STI)
1492 		interruptibility |= GUEST_INTR_STATE_STI;
1493 
1494 	if ((interruptibility != interruptibility_old))
1495 		vmcs_write32(GUEST_INTERRUPTIBILITY_INFO, interruptibility);
1496 }
1497 
1498 static int vmx_rtit_ctl_check(struct kvm_vcpu *vcpu, u64 data)
1499 {
1500 	struct vcpu_vmx *vmx = to_vmx(vcpu);
1501 	unsigned long value;
1502 
1503 	/*
1504 	 * Any MSR write that attempts to change bits marked reserved will
1505 	 * case a #GP fault.
1506 	 */
1507 	if (data & vmx->pt_desc.ctl_bitmask)
1508 		return 1;
1509 
1510 	/*
1511 	 * Any attempt to modify IA32_RTIT_CTL while TraceEn is set will
1512 	 * result in a #GP unless the same write also clears TraceEn.
1513 	 */
1514 	if ((vmx->pt_desc.guest.ctl & RTIT_CTL_TRACEEN) &&
1515 		((vmx->pt_desc.guest.ctl ^ data) & ~RTIT_CTL_TRACEEN))
1516 		return 1;
1517 
1518 	/*
1519 	 * WRMSR to IA32_RTIT_CTL that sets TraceEn but clears this bit
1520 	 * and FabricEn would cause #GP, if
1521 	 * CPUID.(EAX=14H, ECX=0):ECX.SNGLRGNOUT[bit 2] = 0
1522 	 */
1523 	if ((data & RTIT_CTL_TRACEEN) && !(data & RTIT_CTL_TOPA) &&
1524 		!(data & RTIT_CTL_FABRIC_EN) &&
1525 		!intel_pt_validate_cap(vmx->pt_desc.caps,
1526 					PT_CAP_single_range_output))
1527 		return 1;
1528 
1529 	/*
1530 	 * MTCFreq, CycThresh and PSBFreq encodings check, any MSR write that
1531 	 * utilize encodings marked reserved will casue a #GP fault.
1532 	 */
1533 	value = intel_pt_validate_cap(vmx->pt_desc.caps, PT_CAP_mtc_periods);
1534 	if (intel_pt_validate_cap(vmx->pt_desc.caps, PT_CAP_mtc) &&
1535 			!test_bit((data & RTIT_CTL_MTC_RANGE) >>
1536 			RTIT_CTL_MTC_RANGE_OFFSET, &value))
1537 		return 1;
1538 	value = intel_pt_validate_cap(vmx->pt_desc.caps,
1539 						PT_CAP_cycle_thresholds);
1540 	if (intel_pt_validate_cap(vmx->pt_desc.caps, PT_CAP_psb_cyc) &&
1541 			!test_bit((data & RTIT_CTL_CYC_THRESH) >>
1542 			RTIT_CTL_CYC_THRESH_OFFSET, &value))
1543 		return 1;
1544 	value = intel_pt_validate_cap(vmx->pt_desc.caps, PT_CAP_psb_periods);
1545 	if (intel_pt_validate_cap(vmx->pt_desc.caps, PT_CAP_psb_cyc) &&
1546 			!test_bit((data & RTIT_CTL_PSB_FREQ) >>
1547 			RTIT_CTL_PSB_FREQ_OFFSET, &value))
1548 		return 1;
1549 
1550 	/*
1551 	 * If ADDRx_CFG is reserved or the encodings is >2 will
1552 	 * cause a #GP fault.
1553 	 */
1554 	value = (data & RTIT_CTL_ADDR0) >> RTIT_CTL_ADDR0_OFFSET;
1555 	if ((value && (vmx->pt_desc.addr_range < 1)) || (value > 2))
1556 		return 1;
1557 	value = (data & RTIT_CTL_ADDR1) >> RTIT_CTL_ADDR1_OFFSET;
1558 	if ((value && (vmx->pt_desc.addr_range < 2)) || (value > 2))
1559 		return 1;
1560 	value = (data & RTIT_CTL_ADDR2) >> RTIT_CTL_ADDR2_OFFSET;
1561 	if ((value && (vmx->pt_desc.addr_range < 3)) || (value > 2))
1562 		return 1;
1563 	value = (data & RTIT_CTL_ADDR3) >> RTIT_CTL_ADDR3_OFFSET;
1564 	if ((value && (vmx->pt_desc.addr_range < 4)) || (value > 2))
1565 		return 1;
1566 
1567 	return 0;
1568 }
1569 
1570 static int skip_emulated_instruction(struct kvm_vcpu *vcpu)
1571 {
1572 	unsigned long rip;
1573 
1574 	/*
1575 	 * Using VMCS.VM_EXIT_INSTRUCTION_LEN on EPT misconfig depends on
1576 	 * undefined behavior: Intel's SDM doesn't mandate the VMCS field be
1577 	 * set when EPT misconfig occurs.  In practice, real hardware updates
1578 	 * VM_EXIT_INSTRUCTION_LEN on EPT misconfig, but other hypervisors
1579 	 * (namely Hyper-V) don't set it due to it being undefined behavior,
1580 	 * i.e. we end up advancing IP with some random value.
1581 	 */
1582 	if (!static_cpu_has(X86_FEATURE_HYPERVISOR) ||
1583 	    to_vmx(vcpu)->exit_reason != EXIT_REASON_EPT_MISCONFIG) {
1584 		rip = kvm_rip_read(vcpu);
1585 		rip += vmcs_read32(VM_EXIT_INSTRUCTION_LEN);
1586 		kvm_rip_write(vcpu, rip);
1587 	} else {
1588 		if (!kvm_emulate_instruction(vcpu, EMULTYPE_SKIP))
1589 			return 0;
1590 	}
1591 
1592 	/* skipping an emulated instruction also counts */
1593 	vmx_set_interrupt_shadow(vcpu, 0);
1594 
1595 	return 1;
1596 }
1597 
1598 static void vmx_clear_hlt(struct kvm_vcpu *vcpu)
1599 {
1600 	/*
1601 	 * Ensure that we clear the HLT state in the VMCS.  We don't need to
1602 	 * explicitly skip the instruction because if the HLT state is set,
1603 	 * then the instruction is already executing and RIP has already been
1604 	 * advanced.
1605 	 */
1606 	if (kvm_hlt_in_guest(vcpu->kvm) &&
1607 			vmcs_read32(GUEST_ACTIVITY_STATE) == GUEST_ACTIVITY_HLT)
1608 		vmcs_write32(GUEST_ACTIVITY_STATE, GUEST_ACTIVITY_ACTIVE);
1609 }
1610 
1611 static void vmx_queue_exception(struct kvm_vcpu *vcpu)
1612 {
1613 	struct vcpu_vmx *vmx = to_vmx(vcpu);
1614 	unsigned nr = vcpu->arch.exception.nr;
1615 	bool has_error_code = vcpu->arch.exception.has_error_code;
1616 	u32 error_code = vcpu->arch.exception.error_code;
1617 	u32 intr_info = nr | INTR_INFO_VALID_MASK;
1618 
1619 	kvm_deliver_exception_payload(vcpu);
1620 
1621 	if (has_error_code) {
1622 		vmcs_write32(VM_ENTRY_EXCEPTION_ERROR_CODE, error_code);
1623 		intr_info |= INTR_INFO_DELIVER_CODE_MASK;
1624 	}
1625 
1626 	if (vmx->rmode.vm86_active) {
1627 		int inc_eip = 0;
1628 		if (kvm_exception_is_soft(nr))
1629 			inc_eip = vcpu->arch.event_exit_inst_len;
1630 		kvm_inject_realmode_interrupt(vcpu, nr, inc_eip);
1631 		return;
1632 	}
1633 
1634 	WARN_ON_ONCE(vmx->emulation_required);
1635 
1636 	if (kvm_exception_is_soft(nr)) {
1637 		vmcs_write32(VM_ENTRY_INSTRUCTION_LEN,
1638 			     vmx->vcpu.arch.event_exit_inst_len);
1639 		intr_info |= INTR_TYPE_SOFT_EXCEPTION;
1640 	} else
1641 		intr_info |= INTR_TYPE_HARD_EXCEPTION;
1642 
1643 	vmcs_write32(VM_ENTRY_INTR_INFO_FIELD, intr_info);
1644 
1645 	vmx_clear_hlt(vcpu);
1646 }
1647 
1648 static bool vmx_rdtscp_supported(void)
1649 {
1650 	return cpu_has_vmx_rdtscp();
1651 }
1652 
1653 static bool vmx_invpcid_supported(void)
1654 {
1655 	return cpu_has_vmx_invpcid();
1656 }
1657 
1658 /*
1659  * Swap MSR entry in host/guest MSR entry array.
1660  */
1661 static void move_msr_up(struct vcpu_vmx *vmx, int from, int to)
1662 {
1663 	struct shared_msr_entry tmp;
1664 
1665 	tmp = vmx->guest_msrs[to];
1666 	vmx->guest_msrs[to] = vmx->guest_msrs[from];
1667 	vmx->guest_msrs[from] = tmp;
1668 }
1669 
1670 /*
1671  * Set up the vmcs to automatically save and restore system
1672  * msrs.  Don't touch the 64-bit msrs if the guest is in legacy
1673  * mode, as fiddling with msrs is very expensive.
1674  */
1675 static void setup_msrs(struct vcpu_vmx *vmx)
1676 {
1677 	int save_nmsrs, index;
1678 
1679 	save_nmsrs = 0;
1680 #ifdef CONFIG_X86_64
1681 	/*
1682 	 * The SYSCALL MSRs are only needed on long mode guests, and only
1683 	 * when EFER.SCE is set.
1684 	 */
1685 	if (is_long_mode(&vmx->vcpu) && (vmx->vcpu.arch.efer & EFER_SCE)) {
1686 		index = __find_msr_index(vmx, MSR_STAR);
1687 		if (index >= 0)
1688 			move_msr_up(vmx, index, save_nmsrs++);
1689 		index = __find_msr_index(vmx, MSR_LSTAR);
1690 		if (index >= 0)
1691 			move_msr_up(vmx, index, save_nmsrs++);
1692 		index = __find_msr_index(vmx, MSR_SYSCALL_MASK);
1693 		if (index >= 0)
1694 			move_msr_up(vmx, index, save_nmsrs++);
1695 	}
1696 #endif
1697 	index = __find_msr_index(vmx, MSR_EFER);
1698 	if (index >= 0 && update_transition_efer(vmx, index))
1699 		move_msr_up(vmx, index, save_nmsrs++);
1700 	index = __find_msr_index(vmx, MSR_TSC_AUX);
1701 	if (index >= 0 && guest_cpuid_has(&vmx->vcpu, X86_FEATURE_RDTSCP))
1702 		move_msr_up(vmx, index, save_nmsrs++);
1703 	index = __find_msr_index(vmx, MSR_IA32_TSX_CTRL);
1704 	if (index >= 0)
1705 		move_msr_up(vmx, index, save_nmsrs++);
1706 
1707 	vmx->save_nmsrs = save_nmsrs;
1708 	vmx->guest_msrs_ready = false;
1709 
1710 	if (cpu_has_vmx_msr_bitmap())
1711 		vmx_update_msr_bitmap(&vmx->vcpu);
1712 }
1713 
1714 static u64 vmx_read_l1_tsc_offset(struct kvm_vcpu *vcpu)
1715 {
1716 	struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
1717 
1718 	if (is_guest_mode(vcpu) &&
1719 	    (vmcs12->cpu_based_vm_exec_control & CPU_BASED_USE_TSC_OFFSETING))
1720 		return vcpu->arch.tsc_offset - vmcs12->tsc_offset;
1721 
1722 	return vcpu->arch.tsc_offset;
1723 }
1724 
1725 static u64 vmx_write_l1_tsc_offset(struct kvm_vcpu *vcpu, u64 offset)
1726 {
1727 	struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
1728 	u64 g_tsc_offset = 0;
1729 
1730 	/*
1731 	 * We're here if L1 chose not to trap WRMSR to TSC. According
1732 	 * to the spec, this should set L1's TSC; The offset that L1
1733 	 * set for L2 remains unchanged, and still needs to be added
1734 	 * to the newly set TSC to get L2's TSC.
1735 	 */
1736 	if (is_guest_mode(vcpu) &&
1737 	    (vmcs12->cpu_based_vm_exec_control & CPU_BASED_USE_TSC_OFFSETING))
1738 		g_tsc_offset = vmcs12->tsc_offset;
1739 
1740 	trace_kvm_write_tsc_offset(vcpu->vcpu_id,
1741 				   vcpu->arch.tsc_offset - g_tsc_offset,
1742 				   offset);
1743 	vmcs_write64(TSC_OFFSET, offset + g_tsc_offset);
1744 	return offset + g_tsc_offset;
1745 }
1746 
1747 /*
1748  * nested_vmx_allowed() checks whether a guest should be allowed to use VMX
1749  * instructions and MSRs (i.e., nested VMX). Nested VMX is disabled for
1750  * all guests if the "nested" module option is off, and can also be disabled
1751  * for a single guest by disabling its VMX cpuid bit.
1752  */
1753 bool nested_vmx_allowed(struct kvm_vcpu *vcpu)
1754 {
1755 	return nested && guest_cpuid_has(vcpu, X86_FEATURE_VMX);
1756 }
1757 
1758 static inline bool vmx_feature_control_msr_valid(struct kvm_vcpu *vcpu,
1759 						 uint64_t val)
1760 {
1761 	uint64_t valid_bits = to_vmx(vcpu)->msr_ia32_feature_control_valid_bits;
1762 
1763 	return !(val & ~valid_bits);
1764 }
1765 
1766 static int vmx_get_msr_feature(struct kvm_msr_entry *msr)
1767 {
1768 	switch (msr->index) {
1769 	case MSR_IA32_VMX_BASIC ... MSR_IA32_VMX_VMFUNC:
1770 		if (!nested)
1771 			return 1;
1772 		return vmx_get_vmx_msr(&vmcs_config.nested, msr->index, &msr->data);
1773 	default:
1774 		return 1;
1775 	}
1776 
1777 	return 0;
1778 }
1779 
1780 /*
1781  * Reads an msr value (of 'msr_index') into 'pdata'.
1782  * Returns 0 on success, non-0 otherwise.
1783  * Assumes vcpu_load() was already called.
1784  */
1785 static int vmx_get_msr(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
1786 {
1787 	struct vcpu_vmx *vmx = to_vmx(vcpu);
1788 	struct shared_msr_entry *msr;
1789 	u32 index;
1790 
1791 	switch (msr_info->index) {
1792 #ifdef CONFIG_X86_64
1793 	case MSR_FS_BASE:
1794 		msr_info->data = vmcs_readl(GUEST_FS_BASE);
1795 		break;
1796 	case MSR_GS_BASE:
1797 		msr_info->data = vmcs_readl(GUEST_GS_BASE);
1798 		break;
1799 	case MSR_KERNEL_GS_BASE:
1800 		msr_info->data = vmx_read_guest_kernel_gs_base(vmx);
1801 		break;
1802 #endif
1803 	case MSR_EFER:
1804 		return kvm_get_msr_common(vcpu, msr_info);
1805 	case MSR_IA32_TSX_CTRL:
1806 		if (!msr_info->host_initiated &&
1807 		    !(vcpu->arch.arch_capabilities & ARCH_CAP_TSX_CTRL_MSR))
1808 			return 1;
1809 		goto find_shared_msr;
1810 	case MSR_IA32_UMWAIT_CONTROL:
1811 		if (!msr_info->host_initiated && !vmx_has_waitpkg(vmx))
1812 			return 1;
1813 
1814 		msr_info->data = vmx->msr_ia32_umwait_control;
1815 		break;
1816 	case MSR_IA32_SPEC_CTRL:
1817 		if (!msr_info->host_initiated &&
1818 		    !guest_cpuid_has(vcpu, X86_FEATURE_SPEC_CTRL))
1819 			return 1;
1820 
1821 		msr_info->data = to_vmx(vcpu)->spec_ctrl;
1822 		break;
1823 	case MSR_IA32_SYSENTER_CS:
1824 		msr_info->data = vmcs_read32(GUEST_SYSENTER_CS);
1825 		break;
1826 	case MSR_IA32_SYSENTER_EIP:
1827 		msr_info->data = vmcs_readl(GUEST_SYSENTER_EIP);
1828 		break;
1829 	case MSR_IA32_SYSENTER_ESP:
1830 		msr_info->data = vmcs_readl(GUEST_SYSENTER_ESP);
1831 		break;
1832 	case MSR_IA32_BNDCFGS:
1833 		if (!kvm_mpx_supported() ||
1834 		    (!msr_info->host_initiated &&
1835 		     !guest_cpuid_has(vcpu, X86_FEATURE_MPX)))
1836 			return 1;
1837 		msr_info->data = vmcs_read64(GUEST_BNDCFGS);
1838 		break;
1839 	case MSR_IA32_MCG_EXT_CTL:
1840 		if (!msr_info->host_initiated &&
1841 		    !(vmx->msr_ia32_feature_control &
1842 		      FEAT_CTL_LMCE_ENABLED))
1843 			return 1;
1844 		msr_info->data = vcpu->arch.mcg_ext_ctl;
1845 		break;
1846 	case MSR_IA32_FEAT_CTL:
1847 		msr_info->data = vmx->msr_ia32_feature_control;
1848 		break;
1849 	case MSR_IA32_VMX_BASIC ... MSR_IA32_VMX_VMFUNC:
1850 		if (!nested_vmx_allowed(vcpu))
1851 			return 1;
1852 		return vmx_get_vmx_msr(&vmx->nested.msrs, msr_info->index,
1853 				       &msr_info->data);
1854 	case MSR_IA32_RTIT_CTL:
1855 		if (pt_mode != PT_MODE_HOST_GUEST)
1856 			return 1;
1857 		msr_info->data = vmx->pt_desc.guest.ctl;
1858 		break;
1859 	case MSR_IA32_RTIT_STATUS:
1860 		if (pt_mode != PT_MODE_HOST_GUEST)
1861 			return 1;
1862 		msr_info->data = vmx->pt_desc.guest.status;
1863 		break;
1864 	case MSR_IA32_RTIT_CR3_MATCH:
1865 		if ((pt_mode != PT_MODE_HOST_GUEST) ||
1866 			!intel_pt_validate_cap(vmx->pt_desc.caps,
1867 						PT_CAP_cr3_filtering))
1868 			return 1;
1869 		msr_info->data = vmx->pt_desc.guest.cr3_match;
1870 		break;
1871 	case MSR_IA32_RTIT_OUTPUT_BASE:
1872 		if ((pt_mode != PT_MODE_HOST_GUEST) ||
1873 			(!intel_pt_validate_cap(vmx->pt_desc.caps,
1874 					PT_CAP_topa_output) &&
1875 			 !intel_pt_validate_cap(vmx->pt_desc.caps,
1876 					PT_CAP_single_range_output)))
1877 			return 1;
1878 		msr_info->data = vmx->pt_desc.guest.output_base;
1879 		break;
1880 	case MSR_IA32_RTIT_OUTPUT_MASK:
1881 		if ((pt_mode != PT_MODE_HOST_GUEST) ||
1882 			(!intel_pt_validate_cap(vmx->pt_desc.caps,
1883 					PT_CAP_topa_output) &&
1884 			 !intel_pt_validate_cap(vmx->pt_desc.caps,
1885 					PT_CAP_single_range_output)))
1886 			return 1;
1887 		msr_info->data = vmx->pt_desc.guest.output_mask;
1888 		break;
1889 	case MSR_IA32_RTIT_ADDR0_A ... MSR_IA32_RTIT_ADDR3_B:
1890 		index = msr_info->index - MSR_IA32_RTIT_ADDR0_A;
1891 		if ((pt_mode != PT_MODE_HOST_GUEST) ||
1892 			(index >= 2 * intel_pt_validate_cap(vmx->pt_desc.caps,
1893 					PT_CAP_num_address_ranges)))
1894 			return 1;
1895 		if (index % 2)
1896 			msr_info->data = vmx->pt_desc.guest.addr_b[index / 2];
1897 		else
1898 			msr_info->data = vmx->pt_desc.guest.addr_a[index / 2];
1899 		break;
1900 	case MSR_TSC_AUX:
1901 		if (!msr_info->host_initiated &&
1902 		    !guest_cpuid_has(vcpu, X86_FEATURE_RDTSCP))
1903 			return 1;
1904 		goto find_shared_msr;
1905 	default:
1906 	find_shared_msr:
1907 		msr = find_msr_entry(vmx, msr_info->index);
1908 		if (msr) {
1909 			msr_info->data = msr->data;
1910 			break;
1911 		}
1912 		return kvm_get_msr_common(vcpu, msr_info);
1913 	}
1914 
1915 	return 0;
1916 }
1917 
1918 /*
1919  * Writes msr value into into the appropriate "register".
1920  * Returns 0 on success, non-0 otherwise.
1921  * Assumes vcpu_load() was already called.
1922  */
1923 static int vmx_set_msr(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
1924 {
1925 	struct vcpu_vmx *vmx = to_vmx(vcpu);
1926 	struct shared_msr_entry *msr;
1927 	int ret = 0;
1928 	u32 msr_index = msr_info->index;
1929 	u64 data = msr_info->data;
1930 	u32 index;
1931 
1932 	switch (msr_index) {
1933 	case MSR_EFER:
1934 		ret = kvm_set_msr_common(vcpu, msr_info);
1935 		break;
1936 #ifdef CONFIG_X86_64
1937 	case MSR_FS_BASE:
1938 		vmx_segment_cache_clear(vmx);
1939 		vmcs_writel(GUEST_FS_BASE, data);
1940 		break;
1941 	case MSR_GS_BASE:
1942 		vmx_segment_cache_clear(vmx);
1943 		vmcs_writel(GUEST_GS_BASE, data);
1944 		break;
1945 	case MSR_KERNEL_GS_BASE:
1946 		vmx_write_guest_kernel_gs_base(vmx, data);
1947 		break;
1948 #endif
1949 	case MSR_IA32_SYSENTER_CS:
1950 		if (is_guest_mode(vcpu))
1951 			get_vmcs12(vcpu)->guest_sysenter_cs = data;
1952 		vmcs_write32(GUEST_SYSENTER_CS, data);
1953 		break;
1954 	case MSR_IA32_SYSENTER_EIP:
1955 		if (is_guest_mode(vcpu))
1956 			get_vmcs12(vcpu)->guest_sysenter_eip = data;
1957 		vmcs_writel(GUEST_SYSENTER_EIP, data);
1958 		break;
1959 	case MSR_IA32_SYSENTER_ESP:
1960 		if (is_guest_mode(vcpu))
1961 			get_vmcs12(vcpu)->guest_sysenter_esp = data;
1962 		vmcs_writel(GUEST_SYSENTER_ESP, data);
1963 		break;
1964 	case MSR_IA32_DEBUGCTLMSR:
1965 		if (is_guest_mode(vcpu) && get_vmcs12(vcpu)->vm_exit_controls &
1966 						VM_EXIT_SAVE_DEBUG_CONTROLS)
1967 			get_vmcs12(vcpu)->guest_ia32_debugctl = data;
1968 
1969 		ret = kvm_set_msr_common(vcpu, msr_info);
1970 		break;
1971 
1972 	case MSR_IA32_BNDCFGS:
1973 		if (!kvm_mpx_supported() ||
1974 		    (!msr_info->host_initiated &&
1975 		     !guest_cpuid_has(vcpu, X86_FEATURE_MPX)))
1976 			return 1;
1977 		if (is_noncanonical_address(data & PAGE_MASK, vcpu) ||
1978 		    (data & MSR_IA32_BNDCFGS_RSVD))
1979 			return 1;
1980 		vmcs_write64(GUEST_BNDCFGS, data);
1981 		break;
1982 	case MSR_IA32_UMWAIT_CONTROL:
1983 		if (!msr_info->host_initiated && !vmx_has_waitpkg(vmx))
1984 			return 1;
1985 
1986 		/* The reserved bit 1 and non-32 bit [63:32] should be zero */
1987 		if (data & (BIT_ULL(1) | GENMASK_ULL(63, 32)))
1988 			return 1;
1989 
1990 		vmx->msr_ia32_umwait_control = data;
1991 		break;
1992 	case MSR_IA32_SPEC_CTRL:
1993 		if (!msr_info->host_initiated &&
1994 		    !guest_cpuid_has(vcpu, X86_FEATURE_SPEC_CTRL))
1995 			return 1;
1996 
1997 		/* The STIBP bit doesn't fault even if it's not advertised */
1998 		if (data & ~(SPEC_CTRL_IBRS | SPEC_CTRL_STIBP | SPEC_CTRL_SSBD))
1999 			return 1;
2000 
2001 		vmx->spec_ctrl = data;
2002 
2003 		if (!data)
2004 			break;
2005 
2006 		/*
2007 		 * For non-nested:
2008 		 * When it's written (to non-zero) for the first time, pass
2009 		 * it through.
2010 		 *
2011 		 * For nested:
2012 		 * The handling of the MSR bitmap for L2 guests is done in
2013 		 * nested_vmx_merge_msr_bitmap. We should not touch the
2014 		 * vmcs02.msr_bitmap here since it gets completely overwritten
2015 		 * in the merging. We update the vmcs01 here for L1 as well
2016 		 * since it will end up touching the MSR anyway now.
2017 		 */
2018 		vmx_disable_intercept_for_msr(vmx->vmcs01.msr_bitmap,
2019 					      MSR_IA32_SPEC_CTRL,
2020 					      MSR_TYPE_RW);
2021 		break;
2022 	case MSR_IA32_TSX_CTRL:
2023 		if (!msr_info->host_initiated &&
2024 		    !(vcpu->arch.arch_capabilities & ARCH_CAP_TSX_CTRL_MSR))
2025 			return 1;
2026 		if (data & ~(TSX_CTRL_RTM_DISABLE | TSX_CTRL_CPUID_CLEAR))
2027 			return 1;
2028 		goto find_shared_msr;
2029 	case MSR_IA32_PRED_CMD:
2030 		if (!msr_info->host_initiated &&
2031 		    !guest_cpuid_has(vcpu, X86_FEATURE_SPEC_CTRL))
2032 			return 1;
2033 
2034 		if (data & ~PRED_CMD_IBPB)
2035 			return 1;
2036 
2037 		if (!data)
2038 			break;
2039 
2040 		wrmsrl(MSR_IA32_PRED_CMD, PRED_CMD_IBPB);
2041 
2042 		/*
2043 		 * For non-nested:
2044 		 * When it's written (to non-zero) for the first time, pass
2045 		 * it through.
2046 		 *
2047 		 * For nested:
2048 		 * The handling of the MSR bitmap for L2 guests is done in
2049 		 * nested_vmx_merge_msr_bitmap. We should not touch the
2050 		 * vmcs02.msr_bitmap here since it gets completely overwritten
2051 		 * in the merging.
2052 		 */
2053 		vmx_disable_intercept_for_msr(vmx->vmcs01.msr_bitmap, MSR_IA32_PRED_CMD,
2054 					      MSR_TYPE_W);
2055 		break;
2056 	case MSR_IA32_CR_PAT:
2057 		if (!kvm_pat_valid(data))
2058 			return 1;
2059 
2060 		if (is_guest_mode(vcpu) &&
2061 		    get_vmcs12(vcpu)->vm_exit_controls & VM_EXIT_SAVE_IA32_PAT)
2062 			get_vmcs12(vcpu)->guest_ia32_pat = data;
2063 
2064 		if (vmcs_config.vmentry_ctrl & VM_ENTRY_LOAD_IA32_PAT) {
2065 			vmcs_write64(GUEST_IA32_PAT, data);
2066 			vcpu->arch.pat = data;
2067 			break;
2068 		}
2069 		ret = kvm_set_msr_common(vcpu, msr_info);
2070 		break;
2071 	case MSR_IA32_TSC_ADJUST:
2072 		ret = kvm_set_msr_common(vcpu, msr_info);
2073 		break;
2074 	case MSR_IA32_MCG_EXT_CTL:
2075 		if ((!msr_info->host_initiated &&
2076 		     !(to_vmx(vcpu)->msr_ia32_feature_control &
2077 		       FEAT_CTL_LMCE_ENABLED)) ||
2078 		    (data & ~MCG_EXT_CTL_LMCE_EN))
2079 			return 1;
2080 		vcpu->arch.mcg_ext_ctl = data;
2081 		break;
2082 	case MSR_IA32_FEAT_CTL:
2083 		if (!vmx_feature_control_msr_valid(vcpu, data) ||
2084 		    (to_vmx(vcpu)->msr_ia32_feature_control &
2085 		     FEAT_CTL_LOCKED && !msr_info->host_initiated))
2086 			return 1;
2087 		vmx->msr_ia32_feature_control = data;
2088 		if (msr_info->host_initiated && data == 0)
2089 			vmx_leave_nested(vcpu);
2090 		break;
2091 	case MSR_IA32_VMX_BASIC ... MSR_IA32_VMX_VMFUNC:
2092 		if (!msr_info->host_initiated)
2093 			return 1; /* they are read-only */
2094 		if (!nested_vmx_allowed(vcpu))
2095 			return 1;
2096 		return vmx_set_vmx_msr(vcpu, msr_index, data);
2097 	case MSR_IA32_RTIT_CTL:
2098 		if ((pt_mode != PT_MODE_HOST_GUEST) ||
2099 			vmx_rtit_ctl_check(vcpu, data) ||
2100 			vmx->nested.vmxon)
2101 			return 1;
2102 		vmcs_write64(GUEST_IA32_RTIT_CTL, data);
2103 		vmx->pt_desc.guest.ctl = data;
2104 		pt_update_intercept_for_msr(vmx);
2105 		break;
2106 	case MSR_IA32_RTIT_STATUS:
2107 		if ((pt_mode != PT_MODE_HOST_GUEST) ||
2108 			(vmx->pt_desc.guest.ctl & RTIT_CTL_TRACEEN) ||
2109 			(data & MSR_IA32_RTIT_STATUS_MASK))
2110 			return 1;
2111 		vmx->pt_desc.guest.status = data;
2112 		break;
2113 	case MSR_IA32_RTIT_CR3_MATCH:
2114 		if ((pt_mode != PT_MODE_HOST_GUEST) ||
2115 			(vmx->pt_desc.guest.ctl & RTIT_CTL_TRACEEN) ||
2116 			!intel_pt_validate_cap(vmx->pt_desc.caps,
2117 						PT_CAP_cr3_filtering))
2118 			return 1;
2119 		vmx->pt_desc.guest.cr3_match = data;
2120 		break;
2121 	case MSR_IA32_RTIT_OUTPUT_BASE:
2122 		if ((pt_mode != PT_MODE_HOST_GUEST) ||
2123 			(vmx->pt_desc.guest.ctl & RTIT_CTL_TRACEEN) ||
2124 			(!intel_pt_validate_cap(vmx->pt_desc.caps,
2125 					PT_CAP_topa_output) &&
2126 			 !intel_pt_validate_cap(vmx->pt_desc.caps,
2127 					PT_CAP_single_range_output)) ||
2128 			(data & MSR_IA32_RTIT_OUTPUT_BASE_MASK))
2129 			return 1;
2130 		vmx->pt_desc.guest.output_base = data;
2131 		break;
2132 	case MSR_IA32_RTIT_OUTPUT_MASK:
2133 		if ((pt_mode != PT_MODE_HOST_GUEST) ||
2134 			(vmx->pt_desc.guest.ctl & RTIT_CTL_TRACEEN) ||
2135 			(!intel_pt_validate_cap(vmx->pt_desc.caps,
2136 					PT_CAP_topa_output) &&
2137 			 !intel_pt_validate_cap(vmx->pt_desc.caps,
2138 					PT_CAP_single_range_output)))
2139 			return 1;
2140 		vmx->pt_desc.guest.output_mask = data;
2141 		break;
2142 	case MSR_IA32_RTIT_ADDR0_A ... MSR_IA32_RTIT_ADDR3_B:
2143 		index = msr_info->index - MSR_IA32_RTIT_ADDR0_A;
2144 		if ((pt_mode != PT_MODE_HOST_GUEST) ||
2145 			(vmx->pt_desc.guest.ctl & RTIT_CTL_TRACEEN) ||
2146 			(index >= 2 * intel_pt_validate_cap(vmx->pt_desc.caps,
2147 					PT_CAP_num_address_ranges)))
2148 			return 1;
2149 		if (index % 2)
2150 			vmx->pt_desc.guest.addr_b[index / 2] = data;
2151 		else
2152 			vmx->pt_desc.guest.addr_a[index / 2] = data;
2153 		break;
2154 	case MSR_TSC_AUX:
2155 		if (!msr_info->host_initiated &&
2156 		    !guest_cpuid_has(vcpu, X86_FEATURE_RDTSCP))
2157 			return 1;
2158 		/* Check reserved bit, higher 32 bits should be zero */
2159 		if ((data >> 32) != 0)
2160 			return 1;
2161 		goto find_shared_msr;
2162 
2163 	default:
2164 	find_shared_msr:
2165 		msr = find_msr_entry(vmx, msr_index);
2166 		if (msr)
2167 			ret = vmx_set_guest_msr(vmx, msr, data);
2168 		else
2169 			ret = kvm_set_msr_common(vcpu, msr_info);
2170 	}
2171 
2172 	return ret;
2173 }
2174 
2175 static void vmx_cache_reg(struct kvm_vcpu *vcpu, enum kvm_reg reg)
2176 {
2177 	kvm_register_mark_available(vcpu, reg);
2178 
2179 	switch (reg) {
2180 	case VCPU_REGS_RSP:
2181 		vcpu->arch.regs[VCPU_REGS_RSP] = vmcs_readl(GUEST_RSP);
2182 		break;
2183 	case VCPU_REGS_RIP:
2184 		vcpu->arch.regs[VCPU_REGS_RIP] = vmcs_readl(GUEST_RIP);
2185 		break;
2186 	case VCPU_EXREG_PDPTR:
2187 		if (enable_ept)
2188 			ept_save_pdptrs(vcpu);
2189 		break;
2190 	case VCPU_EXREG_CR3:
2191 		if (enable_unrestricted_guest || (enable_ept && is_paging(vcpu)))
2192 			vcpu->arch.cr3 = vmcs_readl(GUEST_CR3);
2193 		break;
2194 	default:
2195 		WARN_ON_ONCE(1);
2196 		break;
2197 	}
2198 }
2199 
2200 static __init int cpu_has_kvm_support(void)
2201 {
2202 	return cpu_has_vmx();
2203 }
2204 
2205 static __init int vmx_disabled_by_bios(void)
2206 {
2207 	return !boot_cpu_has(X86_FEATURE_MSR_IA32_FEAT_CTL) ||
2208 	       !boot_cpu_has(X86_FEATURE_VMX);
2209 }
2210 
2211 static void kvm_cpu_vmxon(u64 addr)
2212 {
2213 	cr4_set_bits(X86_CR4_VMXE);
2214 	intel_pt_handle_vmx(1);
2215 
2216 	asm volatile ("vmxon %0" : : "m"(addr));
2217 }
2218 
2219 static int hardware_enable(void)
2220 {
2221 	int cpu = raw_smp_processor_id();
2222 	u64 phys_addr = __pa(per_cpu(vmxarea, cpu));
2223 
2224 	if (cr4_read_shadow() & X86_CR4_VMXE)
2225 		return -EBUSY;
2226 
2227 	/*
2228 	 * This can happen if we hot-added a CPU but failed to allocate
2229 	 * VP assist page for it.
2230 	 */
2231 	if (static_branch_unlikely(&enable_evmcs) &&
2232 	    !hv_get_vp_assist_page(cpu))
2233 		return -EFAULT;
2234 
2235 	INIT_LIST_HEAD(&per_cpu(loaded_vmcss_on_cpu, cpu));
2236 	INIT_LIST_HEAD(&per_cpu(blocked_vcpu_on_cpu, cpu));
2237 	spin_lock_init(&per_cpu(blocked_vcpu_on_cpu_lock, cpu));
2238 
2239 	/*
2240 	 * Now we can enable the vmclear operation in kdump
2241 	 * since the loaded_vmcss_on_cpu list on this cpu
2242 	 * has been initialized.
2243 	 *
2244 	 * Though the cpu is not in VMX operation now, there
2245 	 * is no problem to enable the vmclear operation
2246 	 * for the loaded_vmcss_on_cpu list is empty!
2247 	 */
2248 	crash_enable_local_vmclear(cpu);
2249 
2250 	kvm_cpu_vmxon(phys_addr);
2251 	if (enable_ept)
2252 		ept_sync_global();
2253 
2254 	return 0;
2255 }
2256 
2257 static void vmclear_local_loaded_vmcss(void)
2258 {
2259 	int cpu = raw_smp_processor_id();
2260 	struct loaded_vmcs *v, *n;
2261 
2262 	list_for_each_entry_safe(v, n, &per_cpu(loaded_vmcss_on_cpu, cpu),
2263 				 loaded_vmcss_on_cpu_link)
2264 		__loaded_vmcs_clear(v);
2265 }
2266 
2267 
2268 /* Just like cpu_vmxoff(), but with the __kvm_handle_fault_on_reboot()
2269  * tricks.
2270  */
2271 static void kvm_cpu_vmxoff(void)
2272 {
2273 	asm volatile (__ex("vmxoff"));
2274 
2275 	intel_pt_handle_vmx(0);
2276 	cr4_clear_bits(X86_CR4_VMXE);
2277 }
2278 
2279 static void hardware_disable(void)
2280 {
2281 	vmclear_local_loaded_vmcss();
2282 	kvm_cpu_vmxoff();
2283 }
2284 
2285 static __init int adjust_vmx_controls(u32 ctl_min, u32 ctl_opt,
2286 				      u32 msr, u32 *result)
2287 {
2288 	u32 vmx_msr_low, vmx_msr_high;
2289 	u32 ctl = ctl_min | ctl_opt;
2290 
2291 	rdmsr(msr, vmx_msr_low, vmx_msr_high);
2292 
2293 	ctl &= vmx_msr_high; /* bit == 0 in high word ==> must be zero */
2294 	ctl |= vmx_msr_low;  /* bit == 1 in low word  ==> must be one  */
2295 
2296 	/* Ensure minimum (required) set of control bits are supported. */
2297 	if (ctl_min & ~ctl)
2298 		return -EIO;
2299 
2300 	*result = ctl;
2301 	return 0;
2302 }
2303 
2304 static __init int setup_vmcs_config(struct vmcs_config *vmcs_conf,
2305 				    struct vmx_capability *vmx_cap)
2306 {
2307 	u32 vmx_msr_low, vmx_msr_high;
2308 	u32 min, opt, min2, opt2;
2309 	u32 _pin_based_exec_control = 0;
2310 	u32 _cpu_based_exec_control = 0;
2311 	u32 _cpu_based_2nd_exec_control = 0;
2312 	u32 _vmexit_control = 0;
2313 	u32 _vmentry_control = 0;
2314 
2315 	memset(vmcs_conf, 0, sizeof(*vmcs_conf));
2316 	min = CPU_BASED_HLT_EXITING |
2317 #ifdef CONFIG_X86_64
2318 	      CPU_BASED_CR8_LOAD_EXITING |
2319 	      CPU_BASED_CR8_STORE_EXITING |
2320 #endif
2321 	      CPU_BASED_CR3_LOAD_EXITING |
2322 	      CPU_BASED_CR3_STORE_EXITING |
2323 	      CPU_BASED_UNCOND_IO_EXITING |
2324 	      CPU_BASED_MOV_DR_EXITING |
2325 	      CPU_BASED_USE_TSC_OFFSETING |
2326 	      CPU_BASED_MWAIT_EXITING |
2327 	      CPU_BASED_MONITOR_EXITING |
2328 	      CPU_BASED_INVLPG_EXITING |
2329 	      CPU_BASED_RDPMC_EXITING;
2330 
2331 	opt = CPU_BASED_TPR_SHADOW |
2332 	      CPU_BASED_USE_MSR_BITMAPS |
2333 	      CPU_BASED_ACTIVATE_SECONDARY_CONTROLS;
2334 	if (adjust_vmx_controls(min, opt, MSR_IA32_VMX_PROCBASED_CTLS,
2335 				&_cpu_based_exec_control) < 0)
2336 		return -EIO;
2337 #ifdef CONFIG_X86_64
2338 	if ((_cpu_based_exec_control & CPU_BASED_TPR_SHADOW))
2339 		_cpu_based_exec_control &= ~CPU_BASED_CR8_LOAD_EXITING &
2340 					   ~CPU_BASED_CR8_STORE_EXITING;
2341 #endif
2342 	if (_cpu_based_exec_control & CPU_BASED_ACTIVATE_SECONDARY_CONTROLS) {
2343 		min2 = 0;
2344 		opt2 = SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES |
2345 			SECONDARY_EXEC_VIRTUALIZE_X2APIC_MODE |
2346 			SECONDARY_EXEC_WBINVD_EXITING |
2347 			SECONDARY_EXEC_ENABLE_VPID |
2348 			SECONDARY_EXEC_ENABLE_EPT |
2349 			SECONDARY_EXEC_UNRESTRICTED_GUEST |
2350 			SECONDARY_EXEC_PAUSE_LOOP_EXITING |
2351 			SECONDARY_EXEC_DESC |
2352 			SECONDARY_EXEC_RDTSCP |
2353 			SECONDARY_EXEC_ENABLE_INVPCID |
2354 			SECONDARY_EXEC_APIC_REGISTER_VIRT |
2355 			SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY |
2356 			SECONDARY_EXEC_SHADOW_VMCS |
2357 			SECONDARY_EXEC_XSAVES |
2358 			SECONDARY_EXEC_RDSEED_EXITING |
2359 			SECONDARY_EXEC_RDRAND_EXITING |
2360 			SECONDARY_EXEC_ENABLE_PML |
2361 			SECONDARY_EXEC_TSC_SCALING |
2362 			SECONDARY_EXEC_ENABLE_USR_WAIT_PAUSE |
2363 			SECONDARY_EXEC_PT_USE_GPA |
2364 			SECONDARY_EXEC_PT_CONCEAL_VMX |
2365 			SECONDARY_EXEC_ENABLE_VMFUNC |
2366 			SECONDARY_EXEC_ENCLS_EXITING;
2367 		if (adjust_vmx_controls(min2, opt2,
2368 					MSR_IA32_VMX_PROCBASED_CTLS2,
2369 					&_cpu_based_2nd_exec_control) < 0)
2370 			return -EIO;
2371 	}
2372 #ifndef CONFIG_X86_64
2373 	if (!(_cpu_based_2nd_exec_control &
2374 				SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES))
2375 		_cpu_based_exec_control &= ~CPU_BASED_TPR_SHADOW;
2376 #endif
2377 
2378 	if (!(_cpu_based_exec_control & CPU_BASED_TPR_SHADOW))
2379 		_cpu_based_2nd_exec_control &= ~(
2380 				SECONDARY_EXEC_APIC_REGISTER_VIRT |
2381 				SECONDARY_EXEC_VIRTUALIZE_X2APIC_MODE |
2382 				SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY);
2383 
2384 	rdmsr_safe(MSR_IA32_VMX_EPT_VPID_CAP,
2385 		&vmx_cap->ept, &vmx_cap->vpid);
2386 
2387 	if (_cpu_based_2nd_exec_control & SECONDARY_EXEC_ENABLE_EPT) {
2388 		/* CR3 accesses and invlpg don't need to cause VM Exits when EPT
2389 		   enabled */
2390 		_cpu_based_exec_control &= ~(CPU_BASED_CR3_LOAD_EXITING |
2391 					     CPU_BASED_CR3_STORE_EXITING |
2392 					     CPU_BASED_INVLPG_EXITING);
2393 	} else if (vmx_cap->ept) {
2394 		vmx_cap->ept = 0;
2395 		pr_warn_once("EPT CAP should not exist if not support "
2396 				"1-setting enable EPT VM-execution control\n");
2397 	}
2398 	if (!(_cpu_based_2nd_exec_control & SECONDARY_EXEC_ENABLE_VPID) &&
2399 		vmx_cap->vpid) {
2400 		vmx_cap->vpid = 0;
2401 		pr_warn_once("VPID CAP should not exist if not support "
2402 				"1-setting enable VPID VM-execution control\n");
2403 	}
2404 
2405 	min = VM_EXIT_SAVE_DEBUG_CONTROLS | VM_EXIT_ACK_INTR_ON_EXIT;
2406 #ifdef CONFIG_X86_64
2407 	min |= VM_EXIT_HOST_ADDR_SPACE_SIZE;
2408 #endif
2409 	opt = VM_EXIT_LOAD_IA32_PERF_GLOBAL_CTRL |
2410 	      VM_EXIT_LOAD_IA32_PAT |
2411 	      VM_EXIT_LOAD_IA32_EFER |
2412 	      VM_EXIT_CLEAR_BNDCFGS |
2413 	      VM_EXIT_PT_CONCEAL_PIP |
2414 	      VM_EXIT_CLEAR_IA32_RTIT_CTL;
2415 	if (adjust_vmx_controls(min, opt, MSR_IA32_VMX_EXIT_CTLS,
2416 				&_vmexit_control) < 0)
2417 		return -EIO;
2418 
2419 	min = PIN_BASED_EXT_INTR_MASK | PIN_BASED_NMI_EXITING;
2420 	opt = PIN_BASED_VIRTUAL_NMIS | PIN_BASED_POSTED_INTR |
2421 		 PIN_BASED_VMX_PREEMPTION_TIMER;
2422 	if (adjust_vmx_controls(min, opt, MSR_IA32_VMX_PINBASED_CTLS,
2423 				&_pin_based_exec_control) < 0)
2424 		return -EIO;
2425 
2426 	if (cpu_has_broken_vmx_preemption_timer())
2427 		_pin_based_exec_control &= ~PIN_BASED_VMX_PREEMPTION_TIMER;
2428 	if (!(_cpu_based_2nd_exec_control &
2429 		SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY))
2430 		_pin_based_exec_control &= ~PIN_BASED_POSTED_INTR;
2431 
2432 	min = VM_ENTRY_LOAD_DEBUG_CONTROLS;
2433 	opt = VM_ENTRY_LOAD_IA32_PERF_GLOBAL_CTRL |
2434 	      VM_ENTRY_LOAD_IA32_PAT |
2435 	      VM_ENTRY_LOAD_IA32_EFER |
2436 	      VM_ENTRY_LOAD_BNDCFGS |
2437 	      VM_ENTRY_PT_CONCEAL_PIP |
2438 	      VM_ENTRY_LOAD_IA32_RTIT_CTL;
2439 	if (adjust_vmx_controls(min, opt, MSR_IA32_VMX_ENTRY_CTLS,
2440 				&_vmentry_control) < 0)
2441 		return -EIO;
2442 
2443 	/*
2444 	 * Some cpus support VM_{ENTRY,EXIT}_IA32_PERF_GLOBAL_CTRL but they
2445 	 * can't be used due to an errata where VM Exit may incorrectly clear
2446 	 * IA32_PERF_GLOBAL_CTRL[34:32].  Workaround the errata by using the
2447 	 * MSR load mechanism to switch IA32_PERF_GLOBAL_CTRL.
2448 	 */
2449 	if (boot_cpu_data.x86 == 0x6) {
2450 		switch (boot_cpu_data.x86_model) {
2451 		case 26: /* AAK155 */
2452 		case 30: /* AAP115 */
2453 		case 37: /* AAT100 */
2454 		case 44: /* BC86,AAY89,BD102 */
2455 		case 46: /* BA97 */
2456 			_vmentry_control &= ~VM_ENTRY_LOAD_IA32_PERF_GLOBAL_CTRL;
2457 			_vmexit_control &= ~VM_EXIT_LOAD_IA32_PERF_GLOBAL_CTRL;
2458 			pr_warn_once("kvm: VM_EXIT_LOAD_IA32_PERF_GLOBAL_CTRL "
2459 					"does not work properly. Using workaround\n");
2460 			break;
2461 		default:
2462 			break;
2463 		}
2464 	}
2465 
2466 
2467 	rdmsr(MSR_IA32_VMX_BASIC, vmx_msr_low, vmx_msr_high);
2468 
2469 	/* IA-32 SDM Vol 3B: VMCS size is never greater than 4kB. */
2470 	if ((vmx_msr_high & 0x1fff) > PAGE_SIZE)
2471 		return -EIO;
2472 
2473 #ifdef CONFIG_X86_64
2474 	/* IA-32 SDM Vol 3B: 64-bit CPUs always have VMX_BASIC_MSR[48]==0. */
2475 	if (vmx_msr_high & (1u<<16))
2476 		return -EIO;
2477 #endif
2478 
2479 	/* Require Write-Back (WB) memory type for VMCS accesses. */
2480 	if (((vmx_msr_high >> 18) & 15) != 6)
2481 		return -EIO;
2482 
2483 	vmcs_conf->size = vmx_msr_high & 0x1fff;
2484 	vmcs_conf->order = get_order(vmcs_conf->size);
2485 	vmcs_conf->basic_cap = vmx_msr_high & ~0x1fff;
2486 
2487 	vmcs_conf->revision_id = vmx_msr_low;
2488 
2489 	vmcs_conf->pin_based_exec_ctrl = _pin_based_exec_control;
2490 	vmcs_conf->cpu_based_exec_ctrl = _cpu_based_exec_control;
2491 	vmcs_conf->cpu_based_2nd_exec_ctrl = _cpu_based_2nd_exec_control;
2492 	vmcs_conf->vmexit_ctrl         = _vmexit_control;
2493 	vmcs_conf->vmentry_ctrl        = _vmentry_control;
2494 
2495 	if (static_branch_unlikely(&enable_evmcs))
2496 		evmcs_sanitize_exec_ctrls(vmcs_conf);
2497 
2498 	return 0;
2499 }
2500 
2501 struct vmcs *alloc_vmcs_cpu(bool shadow, int cpu, gfp_t flags)
2502 {
2503 	int node = cpu_to_node(cpu);
2504 	struct page *pages;
2505 	struct vmcs *vmcs;
2506 
2507 	pages = __alloc_pages_node(node, flags, vmcs_config.order);
2508 	if (!pages)
2509 		return NULL;
2510 	vmcs = page_address(pages);
2511 	memset(vmcs, 0, vmcs_config.size);
2512 
2513 	/* KVM supports Enlightened VMCS v1 only */
2514 	if (static_branch_unlikely(&enable_evmcs))
2515 		vmcs->hdr.revision_id = KVM_EVMCS_VERSION;
2516 	else
2517 		vmcs->hdr.revision_id = vmcs_config.revision_id;
2518 
2519 	if (shadow)
2520 		vmcs->hdr.shadow_vmcs = 1;
2521 	return vmcs;
2522 }
2523 
2524 void free_vmcs(struct vmcs *vmcs)
2525 {
2526 	free_pages((unsigned long)vmcs, vmcs_config.order);
2527 }
2528 
2529 /*
2530  * Free a VMCS, but before that VMCLEAR it on the CPU where it was last loaded
2531  */
2532 void free_loaded_vmcs(struct loaded_vmcs *loaded_vmcs)
2533 {
2534 	if (!loaded_vmcs->vmcs)
2535 		return;
2536 	loaded_vmcs_clear(loaded_vmcs);
2537 	free_vmcs(loaded_vmcs->vmcs);
2538 	loaded_vmcs->vmcs = NULL;
2539 	if (loaded_vmcs->msr_bitmap)
2540 		free_page((unsigned long)loaded_vmcs->msr_bitmap);
2541 	WARN_ON(loaded_vmcs->shadow_vmcs != NULL);
2542 }
2543 
2544 int alloc_loaded_vmcs(struct loaded_vmcs *loaded_vmcs)
2545 {
2546 	loaded_vmcs->vmcs = alloc_vmcs(false);
2547 	if (!loaded_vmcs->vmcs)
2548 		return -ENOMEM;
2549 
2550 	loaded_vmcs->shadow_vmcs = NULL;
2551 	loaded_vmcs->hv_timer_soft_disabled = false;
2552 	loaded_vmcs_init(loaded_vmcs);
2553 
2554 	if (cpu_has_vmx_msr_bitmap()) {
2555 		loaded_vmcs->msr_bitmap = (unsigned long *)
2556 				__get_free_page(GFP_KERNEL_ACCOUNT);
2557 		if (!loaded_vmcs->msr_bitmap)
2558 			goto out_vmcs;
2559 		memset(loaded_vmcs->msr_bitmap, 0xff, PAGE_SIZE);
2560 
2561 		if (IS_ENABLED(CONFIG_HYPERV) &&
2562 		    static_branch_unlikely(&enable_evmcs) &&
2563 		    (ms_hyperv.nested_features & HV_X64_NESTED_MSR_BITMAP)) {
2564 			struct hv_enlightened_vmcs *evmcs =
2565 				(struct hv_enlightened_vmcs *)loaded_vmcs->vmcs;
2566 
2567 			evmcs->hv_enlightenments_control.msr_bitmap = 1;
2568 		}
2569 	}
2570 
2571 	memset(&loaded_vmcs->host_state, 0, sizeof(struct vmcs_host_state));
2572 	memset(&loaded_vmcs->controls_shadow, 0,
2573 		sizeof(struct vmcs_controls_shadow));
2574 
2575 	return 0;
2576 
2577 out_vmcs:
2578 	free_loaded_vmcs(loaded_vmcs);
2579 	return -ENOMEM;
2580 }
2581 
2582 static void free_kvm_area(void)
2583 {
2584 	int cpu;
2585 
2586 	for_each_possible_cpu(cpu) {
2587 		free_vmcs(per_cpu(vmxarea, cpu));
2588 		per_cpu(vmxarea, cpu) = NULL;
2589 	}
2590 }
2591 
2592 static __init int alloc_kvm_area(void)
2593 {
2594 	int cpu;
2595 
2596 	for_each_possible_cpu(cpu) {
2597 		struct vmcs *vmcs;
2598 
2599 		vmcs = alloc_vmcs_cpu(false, cpu, GFP_KERNEL);
2600 		if (!vmcs) {
2601 			free_kvm_area();
2602 			return -ENOMEM;
2603 		}
2604 
2605 		/*
2606 		 * When eVMCS is enabled, alloc_vmcs_cpu() sets
2607 		 * vmcs->revision_id to KVM_EVMCS_VERSION instead of
2608 		 * revision_id reported by MSR_IA32_VMX_BASIC.
2609 		 *
2610 		 * However, even though not explicitly documented by
2611 		 * TLFS, VMXArea passed as VMXON argument should
2612 		 * still be marked with revision_id reported by
2613 		 * physical CPU.
2614 		 */
2615 		if (static_branch_unlikely(&enable_evmcs))
2616 			vmcs->hdr.revision_id = vmcs_config.revision_id;
2617 
2618 		per_cpu(vmxarea, cpu) = vmcs;
2619 	}
2620 	return 0;
2621 }
2622 
2623 static void fix_pmode_seg(struct kvm_vcpu *vcpu, int seg,
2624 		struct kvm_segment *save)
2625 {
2626 	if (!emulate_invalid_guest_state) {
2627 		/*
2628 		 * CS and SS RPL should be equal during guest entry according
2629 		 * to VMX spec, but in reality it is not always so. Since vcpu
2630 		 * is in the middle of the transition from real mode to
2631 		 * protected mode it is safe to assume that RPL 0 is a good
2632 		 * default value.
2633 		 */
2634 		if (seg == VCPU_SREG_CS || seg == VCPU_SREG_SS)
2635 			save->selector &= ~SEGMENT_RPL_MASK;
2636 		save->dpl = save->selector & SEGMENT_RPL_MASK;
2637 		save->s = 1;
2638 	}
2639 	vmx_set_segment(vcpu, save, seg);
2640 }
2641 
2642 static void enter_pmode(struct kvm_vcpu *vcpu)
2643 {
2644 	unsigned long flags;
2645 	struct vcpu_vmx *vmx = to_vmx(vcpu);
2646 
2647 	/*
2648 	 * Update real mode segment cache. It may be not up-to-date if sement
2649 	 * register was written while vcpu was in a guest mode.
2650 	 */
2651 	vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_ES], VCPU_SREG_ES);
2652 	vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_DS], VCPU_SREG_DS);
2653 	vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_FS], VCPU_SREG_FS);
2654 	vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_GS], VCPU_SREG_GS);
2655 	vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_SS], VCPU_SREG_SS);
2656 	vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_CS], VCPU_SREG_CS);
2657 
2658 	vmx->rmode.vm86_active = 0;
2659 
2660 	vmx_segment_cache_clear(vmx);
2661 
2662 	vmx_set_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_TR], VCPU_SREG_TR);
2663 
2664 	flags = vmcs_readl(GUEST_RFLAGS);
2665 	flags &= RMODE_GUEST_OWNED_EFLAGS_BITS;
2666 	flags |= vmx->rmode.save_rflags & ~RMODE_GUEST_OWNED_EFLAGS_BITS;
2667 	vmcs_writel(GUEST_RFLAGS, flags);
2668 
2669 	vmcs_writel(GUEST_CR4, (vmcs_readl(GUEST_CR4) & ~X86_CR4_VME) |
2670 			(vmcs_readl(CR4_READ_SHADOW) & X86_CR4_VME));
2671 
2672 	update_exception_bitmap(vcpu);
2673 
2674 	fix_pmode_seg(vcpu, VCPU_SREG_CS, &vmx->rmode.segs[VCPU_SREG_CS]);
2675 	fix_pmode_seg(vcpu, VCPU_SREG_SS, &vmx->rmode.segs[VCPU_SREG_SS]);
2676 	fix_pmode_seg(vcpu, VCPU_SREG_ES, &vmx->rmode.segs[VCPU_SREG_ES]);
2677 	fix_pmode_seg(vcpu, VCPU_SREG_DS, &vmx->rmode.segs[VCPU_SREG_DS]);
2678 	fix_pmode_seg(vcpu, VCPU_SREG_FS, &vmx->rmode.segs[VCPU_SREG_FS]);
2679 	fix_pmode_seg(vcpu, VCPU_SREG_GS, &vmx->rmode.segs[VCPU_SREG_GS]);
2680 }
2681 
2682 static void fix_rmode_seg(int seg, struct kvm_segment *save)
2683 {
2684 	const struct kvm_vmx_segment_field *sf = &kvm_vmx_segment_fields[seg];
2685 	struct kvm_segment var = *save;
2686 
2687 	var.dpl = 0x3;
2688 	if (seg == VCPU_SREG_CS)
2689 		var.type = 0x3;
2690 
2691 	if (!emulate_invalid_guest_state) {
2692 		var.selector = var.base >> 4;
2693 		var.base = var.base & 0xffff0;
2694 		var.limit = 0xffff;
2695 		var.g = 0;
2696 		var.db = 0;
2697 		var.present = 1;
2698 		var.s = 1;
2699 		var.l = 0;
2700 		var.unusable = 0;
2701 		var.type = 0x3;
2702 		var.avl = 0;
2703 		if (save->base & 0xf)
2704 			printk_once(KERN_WARNING "kvm: segment base is not "
2705 					"paragraph aligned when entering "
2706 					"protected mode (seg=%d)", seg);
2707 	}
2708 
2709 	vmcs_write16(sf->selector, var.selector);
2710 	vmcs_writel(sf->base, var.base);
2711 	vmcs_write32(sf->limit, var.limit);
2712 	vmcs_write32(sf->ar_bytes, vmx_segment_access_rights(&var));
2713 }
2714 
2715 static void enter_rmode(struct kvm_vcpu *vcpu)
2716 {
2717 	unsigned long flags;
2718 	struct vcpu_vmx *vmx = to_vmx(vcpu);
2719 	struct kvm_vmx *kvm_vmx = to_kvm_vmx(vcpu->kvm);
2720 
2721 	vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_TR], VCPU_SREG_TR);
2722 	vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_ES], VCPU_SREG_ES);
2723 	vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_DS], VCPU_SREG_DS);
2724 	vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_FS], VCPU_SREG_FS);
2725 	vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_GS], VCPU_SREG_GS);
2726 	vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_SS], VCPU_SREG_SS);
2727 	vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_CS], VCPU_SREG_CS);
2728 
2729 	vmx->rmode.vm86_active = 1;
2730 
2731 	/*
2732 	 * Very old userspace does not call KVM_SET_TSS_ADDR before entering
2733 	 * vcpu. Warn the user that an update is overdue.
2734 	 */
2735 	if (!kvm_vmx->tss_addr)
2736 		printk_once(KERN_WARNING "kvm: KVM_SET_TSS_ADDR need to be "
2737 			     "called before entering vcpu\n");
2738 
2739 	vmx_segment_cache_clear(vmx);
2740 
2741 	vmcs_writel(GUEST_TR_BASE, kvm_vmx->tss_addr);
2742 	vmcs_write32(GUEST_TR_LIMIT, RMODE_TSS_SIZE - 1);
2743 	vmcs_write32(GUEST_TR_AR_BYTES, 0x008b);
2744 
2745 	flags = vmcs_readl(GUEST_RFLAGS);
2746 	vmx->rmode.save_rflags = flags;
2747 
2748 	flags |= X86_EFLAGS_IOPL | X86_EFLAGS_VM;
2749 
2750 	vmcs_writel(GUEST_RFLAGS, flags);
2751 	vmcs_writel(GUEST_CR4, vmcs_readl(GUEST_CR4) | X86_CR4_VME);
2752 	update_exception_bitmap(vcpu);
2753 
2754 	fix_rmode_seg(VCPU_SREG_SS, &vmx->rmode.segs[VCPU_SREG_SS]);
2755 	fix_rmode_seg(VCPU_SREG_CS, &vmx->rmode.segs[VCPU_SREG_CS]);
2756 	fix_rmode_seg(VCPU_SREG_ES, &vmx->rmode.segs[VCPU_SREG_ES]);
2757 	fix_rmode_seg(VCPU_SREG_DS, &vmx->rmode.segs[VCPU_SREG_DS]);
2758 	fix_rmode_seg(VCPU_SREG_GS, &vmx->rmode.segs[VCPU_SREG_GS]);
2759 	fix_rmode_seg(VCPU_SREG_FS, &vmx->rmode.segs[VCPU_SREG_FS]);
2760 
2761 	kvm_mmu_reset_context(vcpu);
2762 }
2763 
2764 void vmx_set_efer(struct kvm_vcpu *vcpu, u64 efer)
2765 {
2766 	struct vcpu_vmx *vmx = to_vmx(vcpu);
2767 	struct shared_msr_entry *msr = find_msr_entry(vmx, MSR_EFER);
2768 
2769 	if (!msr)
2770 		return;
2771 
2772 	vcpu->arch.efer = efer;
2773 	if (efer & EFER_LMA) {
2774 		vm_entry_controls_setbit(to_vmx(vcpu), VM_ENTRY_IA32E_MODE);
2775 		msr->data = efer;
2776 	} else {
2777 		vm_entry_controls_clearbit(to_vmx(vcpu), VM_ENTRY_IA32E_MODE);
2778 
2779 		msr->data = efer & ~EFER_LME;
2780 	}
2781 	setup_msrs(vmx);
2782 }
2783 
2784 #ifdef CONFIG_X86_64
2785 
2786 static void enter_lmode(struct kvm_vcpu *vcpu)
2787 {
2788 	u32 guest_tr_ar;
2789 
2790 	vmx_segment_cache_clear(to_vmx(vcpu));
2791 
2792 	guest_tr_ar = vmcs_read32(GUEST_TR_AR_BYTES);
2793 	if ((guest_tr_ar & VMX_AR_TYPE_MASK) != VMX_AR_TYPE_BUSY_64_TSS) {
2794 		pr_debug_ratelimited("%s: tss fixup for long mode. \n",
2795 				     __func__);
2796 		vmcs_write32(GUEST_TR_AR_BYTES,
2797 			     (guest_tr_ar & ~VMX_AR_TYPE_MASK)
2798 			     | VMX_AR_TYPE_BUSY_64_TSS);
2799 	}
2800 	vmx_set_efer(vcpu, vcpu->arch.efer | EFER_LMA);
2801 }
2802 
2803 static void exit_lmode(struct kvm_vcpu *vcpu)
2804 {
2805 	vm_entry_controls_clearbit(to_vmx(vcpu), VM_ENTRY_IA32E_MODE);
2806 	vmx_set_efer(vcpu, vcpu->arch.efer & ~EFER_LMA);
2807 }
2808 
2809 #endif
2810 
2811 static void vmx_flush_tlb_gva(struct kvm_vcpu *vcpu, gva_t addr)
2812 {
2813 	int vpid = to_vmx(vcpu)->vpid;
2814 
2815 	if (!vpid_sync_vcpu_addr(vpid, addr))
2816 		vpid_sync_context(vpid);
2817 
2818 	/*
2819 	 * If VPIDs are not supported or enabled, then the above is a no-op.
2820 	 * But we don't really need a TLB flush in that case anyway, because
2821 	 * each VM entry/exit includes an implicit flush when VPID is 0.
2822 	 */
2823 }
2824 
2825 static void vmx_decache_cr0_guest_bits(struct kvm_vcpu *vcpu)
2826 {
2827 	ulong cr0_guest_owned_bits = vcpu->arch.cr0_guest_owned_bits;
2828 
2829 	vcpu->arch.cr0 &= ~cr0_guest_owned_bits;
2830 	vcpu->arch.cr0 |= vmcs_readl(GUEST_CR0) & cr0_guest_owned_bits;
2831 }
2832 
2833 static void vmx_decache_cr4_guest_bits(struct kvm_vcpu *vcpu)
2834 {
2835 	ulong cr4_guest_owned_bits = vcpu->arch.cr4_guest_owned_bits;
2836 
2837 	vcpu->arch.cr4 &= ~cr4_guest_owned_bits;
2838 	vcpu->arch.cr4 |= vmcs_readl(GUEST_CR4) & cr4_guest_owned_bits;
2839 }
2840 
2841 static void ept_load_pdptrs(struct kvm_vcpu *vcpu)
2842 {
2843 	struct kvm_mmu *mmu = vcpu->arch.walk_mmu;
2844 
2845 	if (!kvm_register_is_dirty(vcpu, VCPU_EXREG_PDPTR))
2846 		return;
2847 
2848 	if (is_pae_paging(vcpu)) {
2849 		vmcs_write64(GUEST_PDPTR0, mmu->pdptrs[0]);
2850 		vmcs_write64(GUEST_PDPTR1, mmu->pdptrs[1]);
2851 		vmcs_write64(GUEST_PDPTR2, mmu->pdptrs[2]);
2852 		vmcs_write64(GUEST_PDPTR3, mmu->pdptrs[3]);
2853 	}
2854 }
2855 
2856 void ept_save_pdptrs(struct kvm_vcpu *vcpu)
2857 {
2858 	struct kvm_mmu *mmu = vcpu->arch.walk_mmu;
2859 
2860 	if (is_pae_paging(vcpu)) {
2861 		mmu->pdptrs[0] = vmcs_read64(GUEST_PDPTR0);
2862 		mmu->pdptrs[1] = vmcs_read64(GUEST_PDPTR1);
2863 		mmu->pdptrs[2] = vmcs_read64(GUEST_PDPTR2);
2864 		mmu->pdptrs[3] = vmcs_read64(GUEST_PDPTR3);
2865 	}
2866 
2867 	kvm_register_mark_dirty(vcpu, VCPU_EXREG_PDPTR);
2868 }
2869 
2870 static void ept_update_paging_mode_cr0(unsigned long *hw_cr0,
2871 					unsigned long cr0,
2872 					struct kvm_vcpu *vcpu)
2873 {
2874 	struct vcpu_vmx *vmx = to_vmx(vcpu);
2875 
2876 	if (!kvm_register_is_available(vcpu, VCPU_EXREG_CR3))
2877 		vmx_cache_reg(vcpu, VCPU_EXREG_CR3);
2878 	if (!(cr0 & X86_CR0_PG)) {
2879 		/* From paging/starting to nonpaging */
2880 		exec_controls_setbit(vmx, CPU_BASED_CR3_LOAD_EXITING |
2881 					  CPU_BASED_CR3_STORE_EXITING);
2882 		vcpu->arch.cr0 = cr0;
2883 		vmx_set_cr4(vcpu, kvm_read_cr4(vcpu));
2884 	} else if (!is_paging(vcpu)) {
2885 		/* From nonpaging to paging */
2886 		exec_controls_clearbit(vmx, CPU_BASED_CR3_LOAD_EXITING |
2887 					    CPU_BASED_CR3_STORE_EXITING);
2888 		vcpu->arch.cr0 = cr0;
2889 		vmx_set_cr4(vcpu, kvm_read_cr4(vcpu));
2890 	}
2891 
2892 	if (!(cr0 & X86_CR0_WP))
2893 		*hw_cr0 &= ~X86_CR0_WP;
2894 }
2895 
2896 void vmx_set_cr0(struct kvm_vcpu *vcpu, unsigned long cr0)
2897 {
2898 	struct vcpu_vmx *vmx = to_vmx(vcpu);
2899 	unsigned long hw_cr0;
2900 
2901 	hw_cr0 = (cr0 & ~KVM_VM_CR0_ALWAYS_OFF);
2902 	if (enable_unrestricted_guest)
2903 		hw_cr0 |= KVM_VM_CR0_ALWAYS_ON_UNRESTRICTED_GUEST;
2904 	else {
2905 		hw_cr0 |= KVM_VM_CR0_ALWAYS_ON;
2906 
2907 		if (vmx->rmode.vm86_active && (cr0 & X86_CR0_PE))
2908 			enter_pmode(vcpu);
2909 
2910 		if (!vmx->rmode.vm86_active && !(cr0 & X86_CR0_PE))
2911 			enter_rmode(vcpu);
2912 	}
2913 
2914 #ifdef CONFIG_X86_64
2915 	if (vcpu->arch.efer & EFER_LME) {
2916 		if (!is_paging(vcpu) && (cr0 & X86_CR0_PG))
2917 			enter_lmode(vcpu);
2918 		if (is_paging(vcpu) && !(cr0 & X86_CR0_PG))
2919 			exit_lmode(vcpu);
2920 	}
2921 #endif
2922 
2923 	if (enable_ept && !enable_unrestricted_guest)
2924 		ept_update_paging_mode_cr0(&hw_cr0, cr0, vcpu);
2925 
2926 	vmcs_writel(CR0_READ_SHADOW, cr0);
2927 	vmcs_writel(GUEST_CR0, hw_cr0);
2928 	vcpu->arch.cr0 = cr0;
2929 
2930 	/* depends on vcpu->arch.cr0 to be set to a new value */
2931 	vmx->emulation_required = emulation_required(vcpu);
2932 }
2933 
2934 static int get_ept_level(struct kvm_vcpu *vcpu)
2935 {
2936 	if (cpu_has_vmx_ept_5levels() && (cpuid_maxphyaddr(vcpu) > 48))
2937 		return 5;
2938 	return 4;
2939 }
2940 
2941 u64 construct_eptp(struct kvm_vcpu *vcpu, unsigned long root_hpa)
2942 {
2943 	u64 eptp = VMX_EPTP_MT_WB;
2944 
2945 	eptp |= (get_ept_level(vcpu) == 5) ? VMX_EPTP_PWL_5 : VMX_EPTP_PWL_4;
2946 
2947 	if (enable_ept_ad_bits &&
2948 	    (!is_guest_mode(vcpu) || nested_ept_ad_enabled(vcpu)))
2949 		eptp |= VMX_EPTP_AD_ENABLE_BIT;
2950 	eptp |= (root_hpa & PAGE_MASK);
2951 
2952 	return eptp;
2953 }
2954 
2955 void vmx_set_cr3(struct kvm_vcpu *vcpu, unsigned long cr3)
2956 {
2957 	struct kvm *kvm = vcpu->kvm;
2958 	bool update_guest_cr3 = true;
2959 	unsigned long guest_cr3;
2960 	u64 eptp;
2961 
2962 	guest_cr3 = cr3;
2963 	if (enable_ept) {
2964 		eptp = construct_eptp(vcpu, cr3);
2965 		vmcs_write64(EPT_POINTER, eptp);
2966 
2967 		if (kvm_x86_ops->tlb_remote_flush) {
2968 			spin_lock(&to_kvm_vmx(kvm)->ept_pointer_lock);
2969 			to_vmx(vcpu)->ept_pointer = eptp;
2970 			to_kvm_vmx(kvm)->ept_pointers_match
2971 				= EPT_POINTERS_CHECK;
2972 			spin_unlock(&to_kvm_vmx(kvm)->ept_pointer_lock);
2973 		}
2974 
2975 		/* Loading vmcs02.GUEST_CR3 is handled by nested VM-Enter. */
2976 		if (is_guest_mode(vcpu))
2977 			update_guest_cr3 = false;
2978 		else if (!enable_unrestricted_guest && !is_paging(vcpu))
2979 			guest_cr3 = to_kvm_vmx(kvm)->ept_identity_map_addr;
2980 		else if (test_bit(VCPU_EXREG_CR3, (ulong *)&vcpu->arch.regs_avail))
2981 			guest_cr3 = vcpu->arch.cr3;
2982 		else /* vmcs01.GUEST_CR3 is already up-to-date. */
2983 			update_guest_cr3 = false;
2984 		ept_load_pdptrs(vcpu);
2985 	}
2986 
2987 	if (update_guest_cr3)
2988 		vmcs_writel(GUEST_CR3, guest_cr3);
2989 }
2990 
2991 int vmx_set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4)
2992 {
2993 	struct vcpu_vmx *vmx = to_vmx(vcpu);
2994 	/*
2995 	 * Pass through host's Machine Check Enable value to hw_cr4, which
2996 	 * is in force while we are in guest mode.  Do not let guests control
2997 	 * this bit, even if host CR4.MCE == 0.
2998 	 */
2999 	unsigned long hw_cr4;
3000 
3001 	hw_cr4 = (cr4_read_shadow() & X86_CR4_MCE) | (cr4 & ~X86_CR4_MCE);
3002 	if (enable_unrestricted_guest)
3003 		hw_cr4 |= KVM_VM_CR4_ALWAYS_ON_UNRESTRICTED_GUEST;
3004 	else if (vmx->rmode.vm86_active)
3005 		hw_cr4 |= KVM_RMODE_VM_CR4_ALWAYS_ON;
3006 	else
3007 		hw_cr4 |= KVM_PMODE_VM_CR4_ALWAYS_ON;
3008 
3009 	if (!boot_cpu_has(X86_FEATURE_UMIP) && vmx_umip_emulated()) {
3010 		if (cr4 & X86_CR4_UMIP) {
3011 			secondary_exec_controls_setbit(vmx, SECONDARY_EXEC_DESC);
3012 			hw_cr4 &= ~X86_CR4_UMIP;
3013 		} else if (!is_guest_mode(vcpu) ||
3014 			!nested_cpu_has2(get_vmcs12(vcpu), SECONDARY_EXEC_DESC)) {
3015 			secondary_exec_controls_clearbit(vmx, SECONDARY_EXEC_DESC);
3016 		}
3017 	}
3018 
3019 	if (cr4 & X86_CR4_VMXE) {
3020 		/*
3021 		 * To use VMXON (and later other VMX instructions), a guest
3022 		 * must first be able to turn on cr4.VMXE (see handle_vmon()).
3023 		 * So basically the check on whether to allow nested VMX
3024 		 * is here.  We operate under the default treatment of SMM,
3025 		 * so VMX cannot be enabled under SMM.
3026 		 */
3027 		if (!nested_vmx_allowed(vcpu) || is_smm(vcpu))
3028 			return 1;
3029 	}
3030 
3031 	if (vmx->nested.vmxon && !nested_cr4_valid(vcpu, cr4))
3032 		return 1;
3033 
3034 	vcpu->arch.cr4 = cr4;
3035 
3036 	if (!enable_unrestricted_guest) {
3037 		if (enable_ept) {
3038 			if (!is_paging(vcpu)) {
3039 				hw_cr4 &= ~X86_CR4_PAE;
3040 				hw_cr4 |= X86_CR4_PSE;
3041 			} else if (!(cr4 & X86_CR4_PAE)) {
3042 				hw_cr4 &= ~X86_CR4_PAE;
3043 			}
3044 		}
3045 
3046 		/*
3047 		 * SMEP/SMAP/PKU is disabled if CPU is in non-paging mode in
3048 		 * hardware.  To emulate this behavior, SMEP/SMAP/PKU needs
3049 		 * to be manually disabled when guest switches to non-paging
3050 		 * mode.
3051 		 *
3052 		 * If !enable_unrestricted_guest, the CPU is always running
3053 		 * with CR0.PG=1 and CR4 needs to be modified.
3054 		 * If enable_unrestricted_guest, the CPU automatically
3055 		 * disables SMEP/SMAP/PKU when the guest sets CR0.PG=0.
3056 		 */
3057 		if (!is_paging(vcpu))
3058 			hw_cr4 &= ~(X86_CR4_SMEP | X86_CR4_SMAP | X86_CR4_PKE);
3059 	}
3060 
3061 	vmcs_writel(CR4_READ_SHADOW, cr4);
3062 	vmcs_writel(GUEST_CR4, hw_cr4);
3063 	return 0;
3064 }
3065 
3066 void vmx_get_segment(struct kvm_vcpu *vcpu, struct kvm_segment *var, int seg)
3067 {
3068 	struct vcpu_vmx *vmx = to_vmx(vcpu);
3069 	u32 ar;
3070 
3071 	if (vmx->rmode.vm86_active && seg != VCPU_SREG_LDTR) {
3072 		*var = vmx->rmode.segs[seg];
3073 		if (seg == VCPU_SREG_TR
3074 		    || var->selector == vmx_read_guest_seg_selector(vmx, seg))
3075 			return;
3076 		var->base = vmx_read_guest_seg_base(vmx, seg);
3077 		var->selector = vmx_read_guest_seg_selector(vmx, seg);
3078 		return;
3079 	}
3080 	var->base = vmx_read_guest_seg_base(vmx, seg);
3081 	var->limit = vmx_read_guest_seg_limit(vmx, seg);
3082 	var->selector = vmx_read_guest_seg_selector(vmx, seg);
3083 	ar = vmx_read_guest_seg_ar(vmx, seg);
3084 	var->unusable = (ar >> 16) & 1;
3085 	var->type = ar & 15;
3086 	var->s = (ar >> 4) & 1;
3087 	var->dpl = (ar >> 5) & 3;
3088 	/*
3089 	 * Some userspaces do not preserve unusable property. Since usable
3090 	 * segment has to be present according to VMX spec we can use present
3091 	 * property to amend userspace bug by making unusable segment always
3092 	 * nonpresent. vmx_segment_access_rights() already marks nonpresent
3093 	 * segment as unusable.
3094 	 */
3095 	var->present = !var->unusable;
3096 	var->avl = (ar >> 12) & 1;
3097 	var->l = (ar >> 13) & 1;
3098 	var->db = (ar >> 14) & 1;
3099 	var->g = (ar >> 15) & 1;
3100 }
3101 
3102 static u64 vmx_get_segment_base(struct kvm_vcpu *vcpu, int seg)
3103 {
3104 	struct kvm_segment s;
3105 
3106 	if (to_vmx(vcpu)->rmode.vm86_active) {
3107 		vmx_get_segment(vcpu, &s, seg);
3108 		return s.base;
3109 	}
3110 	return vmx_read_guest_seg_base(to_vmx(vcpu), seg);
3111 }
3112 
3113 int vmx_get_cpl(struct kvm_vcpu *vcpu)
3114 {
3115 	struct vcpu_vmx *vmx = to_vmx(vcpu);
3116 
3117 	if (unlikely(vmx->rmode.vm86_active))
3118 		return 0;
3119 	else {
3120 		int ar = vmx_read_guest_seg_ar(vmx, VCPU_SREG_SS);
3121 		return VMX_AR_DPL(ar);
3122 	}
3123 }
3124 
3125 static u32 vmx_segment_access_rights(struct kvm_segment *var)
3126 {
3127 	u32 ar;
3128 
3129 	if (var->unusable || !var->present)
3130 		ar = 1 << 16;
3131 	else {
3132 		ar = var->type & 15;
3133 		ar |= (var->s & 1) << 4;
3134 		ar |= (var->dpl & 3) << 5;
3135 		ar |= (var->present & 1) << 7;
3136 		ar |= (var->avl & 1) << 12;
3137 		ar |= (var->l & 1) << 13;
3138 		ar |= (var->db & 1) << 14;
3139 		ar |= (var->g & 1) << 15;
3140 	}
3141 
3142 	return ar;
3143 }
3144 
3145 void vmx_set_segment(struct kvm_vcpu *vcpu, struct kvm_segment *var, int seg)
3146 {
3147 	struct vcpu_vmx *vmx = to_vmx(vcpu);
3148 	const struct kvm_vmx_segment_field *sf = &kvm_vmx_segment_fields[seg];
3149 
3150 	vmx_segment_cache_clear(vmx);
3151 
3152 	if (vmx->rmode.vm86_active && seg != VCPU_SREG_LDTR) {
3153 		vmx->rmode.segs[seg] = *var;
3154 		if (seg == VCPU_SREG_TR)
3155 			vmcs_write16(sf->selector, var->selector);
3156 		else if (var->s)
3157 			fix_rmode_seg(seg, &vmx->rmode.segs[seg]);
3158 		goto out;
3159 	}
3160 
3161 	vmcs_writel(sf->base, var->base);
3162 	vmcs_write32(sf->limit, var->limit);
3163 	vmcs_write16(sf->selector, var->selector);
3164 
3165 	/*
3166 	 *   Fix the "Accessed" bit in AR field of segment registers for older
3167 	 * qemu binaries.
3168 	 *   IA32 arch specifies that at the time of processor reset the
3169 	 * "Accessed" bit in the AR field of segment registers is 1. And qemu
3170 	 * is setting it to 0 in the userland code. This causes invalid guest
3171 	 * state vmexit when "unrestricted guest" mode is turned on.
3172 	 *    Fix for this setup issue in cpu_reset is being pushed in the qemu
3173 	 * tree. Newer qemu binaries with that qemu fix would not need this
3174 	 * kvm hack.
3175 	 */
3176 	if (enable_unrestricted_guest && (seg != VCPU_SREG_LDTR))
3177 		var->type |= 0x1; /* Accessed */
3178 
3179 	vmcs_write32(sf->ar_bytes, vmx_segment_access_rights(var));
3180 
3181 out:
3182 	vmx->emulation_required = emulation_required(vcpu);
3183 }
3184 
3185 static void vmx_get_cs_db_l_bits(struct kvm_vcpu *vcpu, int *db, int *l)
3186 {
3187 	u32 ar = vmx_read_guest_seg_ar(to_vmx(vcpu), VCPU_SREG_CS);
3188 
3189 	*db = (ar >> 14) & 1;
3190 	*l = (ar >> 13) & 1;
3191 }
3192 
3193 static void vmx_get_idt(struct kvm_vcpu *vcpu, struct desc_ptr *dt)
3194 {
3195 	dt->size = vmcs_read32(GUEST_IDTR_LIMIT);
3196 	dt->address = vmcs_readl(GUEST_IDTR_BASE);
3197 }
3198 
3199 static void vmx_set_idt(struct kvm_vcpu *vcpu, struct desc_ptr *dt)
3200 {
3201 	vmcs_write32(GUEST_IDTR_LIMIT, dt->size);
3202 	vmcs_writel(GUEST_IDTR_BASE, dt->address);
3203 }
3204 
3205 static void vmx_get_gdt(struct kvm_vcpu *vcpu, struct desc_ptr *dt)
3206 {
3207 	dt->size = vmcs_read32(GUEST_GDTR_LIMIT);
3208 	dt->address = vmcs_readl(GUEST_GDTR_BASE);
3209 }
3210 
3211 static void vmx_set_gdt(struct kvm_vcpu *vcpu, struct desc_ptr *dt)
3212 {
3213 	vmcs_write32(GUEST_GDTR_LIMIT, dt->size);
3214 	vmcs_writel(GUEST_GDTR_BASE, dt->address);
3215 }
3216 
3217 static bool rmode_segment_valid(struct kvm_vcpu *vcpu, int seg)
3218 {
3219 	struct kvm_segment var;
3220 	u32 ar;
3221 
3222 	vmx_get_segment(vcpu, &var, seg);
3223 	var.dpl = 0x3;
3224 	if (seg == VCPU_SREG_CS)
3225 		var.type = 0x3;
3226 	ar = vmx_segment_access_rights(&var);
3227 
3228 	if (var.base != (var.selector << 4))
3229 		return false;
3230 	if (var.limit != 0xffff)
3231 		return false;
3232 	if (ar != 0xf3)
3233 		return false;
3234 
3235 	return true;
3236 }
3237 
3238 static bool code_segment_valid(struct kvm_vcpu *vcpu)
3239 {
3240 	struct kvm_segment cs;
3241 	unsigned int cs_rpl;
3242 
3243 	vmx_get_segment(vcpu, &cs, VCPU_SREG_CS);
3244 	cs_rpl = cs.selector & SEGMENT_RPL_MASK;
3245 
3246 	if (cs.unusable)
3247 		return false;
3248 	if (~cs.type & (VMX_AR_TYPE_CODE_MASK|VMX_AR_TYPE_ACCESSES_MASK))
3249 		return false;
3250 	if (!cs.s)
3251 		return false;
3252 	if (cs.type & VMX_AR_TYPE_WRITEABLE_MASK) {
3253 		if (cs.dpl > cs_rpl)
3254 			return false;
3255 	} else {
3256 		if (cs.dpl != cs_rpl)
3257 			return false;
3258 	}
3259 	if (!cs.present)
3260 		return false;
3261 
3262 	/* TODO: Add Reserved field check, this'll require a new member in the kvm_segment_field structure */
3263 	return true;
3264 }
3265 
3266 static bool stack_segment_valid(struct kvm_vcpu *vcpu)
3267 {
3268 	struct kvm_segment ss;
3269 	unsigned int ss_rpl;
3270 
3271 	vmx_get_segment(vcpu, &ss, VCPU_SREG_SS);
3272 	ss_rpl = ss.selector & SEGMENT_RPL_MASK;
3273 
3274 	if (ss.unusable)
3275 		return true;
3276 	if (ss.type != 3 && ss.type != 7)
3277 		return false;
3278 	if (!ss.s)
3279 		return false;
3280 	if (ss.dpl != ss_rpl) /* DPL != RPL */
3281 		return false;
3282 	if (!ss.present)
3283 		return false;
3284 
3285 	return true;
3286 }
3287 
3288 static bool data_segment_valid(struct kvm_vcpu *vcpu, int seg)
3289 {
3290 	struct kvm_segment var;
3291 	unsigned int rpl;
3292 
3293 	vmx_get_segment(vcpu, &var, seg);
3294 	rpl = var.selector & SEGMENT_RPL_MASK;
3295 
3296 	if (var.unusable)
3297 		return true;
3298 	if (!var.s)
3299 		return false;
3300 	if (!var.present)
3301 		return false;
3302 	if (~var.type & (VMX_AR_TYPE_CODE_MASK|VMX_AR_TYPE_WRITEABLE_MASK)) {
3303 		if (var.dpl < rpl) /* DPL < RPL */
3304 			return false;
3305 	}
3306 
3307 	/* TODO: Add other members to kvm_segment_field to allow checking for other access
3308 	 * rights flags
3309 	 */
3310 	return true;
3311 }
3312 
3313 static bool tr_valid(struct kvm_vcpu *vcpu)
3314 {
3315 	struct kvm_segment tr;
3316 
3317 	vmx_get_segment(vcpu, &tr, VCPU_SREG_TR);
3318 
3319 	if (tr.unusable)
3320 		return false;
3321 	if (tr.selector & SEGMENT_TI_MASK)	/* TI = 1 */
3322 		return false;
3323 	if (tr.type != 3 && tr.type != 11) /* TODO: Check if guest is in IA32e mode */
3324 		return false;
3325 	if (!tr.present)
3326 		return false;
3327 
3328 	return true;
3329 }
3330 
3331 static bool ldtr_valid(struct kvm_vcpu *vcpu)
3332 {
3333 	struct kvm_segment ldtr;
3334 
3335 	vmx_get_segment(vcpu, &ldtr, VCPU_SREG_LDTR);
3336 
3337 	if (ldtr.unusable)
3338 		return true;
3339 	if (ldtr.selector & SEGMENT_TI_MASK)	/* TI = 1 */
3340 		return false;
3341 	if (ldtr.type != 2)
3342 		return false;
3343 	if (!ldtr.present)
3344 		return false;
3345 
3346 	return true;
3347 }
3348 
3349 static bool cs_ss_rpl_check(struct kvm_vcpu *vcpu)
3350 {
3351 	struct kvm_segment cs, ss;
3352 
3353 	vmx_get_segment(vcpu, &cs, VCPU_SREG_CS);
3354 	vmx_get_segment(vcpu, &ss, VCPU_SREG_SS);
3355 
3356 	return ((cs.selector & SEGMENT_RPL_MASK) ==
3357 		 (ss.selector & SEGMENT_RPL_MASK));
3358 }
3359 
3360 /*
3361  * Check if guest state is valid. Returns true if valid, false if
3362  * not.
3363  * We assume that registers are always usable
3364  */
3365 static bool guest_state_valid(struct kvm_vcpu *vcpu)
3366 {
3367 	if (enable_unrestricted_guest)
3368 		return true;
3369 
3370 	/* real mode guest state checks */
3371 	if (!is_protmode(vcpu) || (vmx_get_rflags(vcpu) & X86_EFLAGS_VM)) {
3372 		if (!rmode_segment_valid(vcpu, VCPU_SREG_CS))
3373 			return false;
3374 		if (!rmode_segment_valid(vcpu, VCPU_SREG_SS))
3375 			return false;
3376 		if (!rmode_segment_valid(vcpu, VCPU_SREG_DS))
3377 			return false;
3378 		if (!rmode_segment_valid(vcpu, VCPU_SREG_ES))
3379 			return false;
3380 		if (!rmode_segment_valid(vcpu, VCPU_SREG_FS))
3381 			return false;
3382 		if (!rmode_segment_valid(vcpu, VCPU_SREG_GS))
3383 			return false;
3384 	} else {
3385 	/* protected mode guest state checks */
3386 		if (!cs_ss_rpl_check(vcpu))
3387 			return false;
3388 		if (!code_segment_valid(vcpu))
3389 			return false;
3390 		if (!stack_segment_valid(vcpu))
3391 			return false;
3392 		if (!data_segment_valid(vcpu, VCPU_SREG_DS))
3393 			return false;
3394 		if (!data_segment_valid(vcpu, VCPU_SREG_ES))
3395 			return false;
3396 		if (!data_segment_valid(vcpu, VCPU_SREG_FS))
3397 			return false;
3398 		if (!data_segment_valid(vcpu, VCPU_SREG_GS))
3399 			return false;
3400 		if (!tr_valid(vcpu))
3401 			return false;
3402 		if (!ldtr_valid(vcpu))
3403 			return false;
3404 	}
3405 	/* TODO:
3406 	 * - Add checks on RIP
3407 	 * - Add checks on RFLAGS
3408 	 */
3409 
3410 	return true;
3411 }
3412 
3413 static int init_rmode_tss(struct kvm *kvm)
3414 {
3415 	gfn_t fn;
3416 	u16 data = 0;
3417 	int idx, r;
3418 
3419 	idx = srcu_read_lock(&kvm->srcu);
3420 	fn = to_kvm_vmx(kvm)->tss_addr >> PAGE_SHIFT;
3421 	r = kvm_clear_guest_page(kvm, fn, 0, PAGE_SIZE);
3422 	if (r < 0)
3423 		goto out;
3424 	data = TSS_BASE_SIZE + TSS_REDIRECTION_SIZE;
3425 	r = kvm_write_guest_page(kvm, fn++, &data,
3426 			TSS_IOPB_BASE_OFFSET, sizeof(u16));
3427 	if (r < 0)
3428 		goto out;
3429 	r = kvm_clear_guest_page(kvm, fn++, 0, PAGE_SIZE);
3430 	if (r < 0)
3431 		goto out;
3432 	r = kvm_clear_guest_page(kvm, fn, 0, PAGE_SIZE);
3433 	if (r < 0)
3434 		goto out;
3435 	data = ~0;
3436 	r = kvm_write_guest_page(kvm, fn, &data,
3437 				 RMODE_TSS_SIZE - 2 * PAGE_SIZE - 1,
3438 				 sizeof(u8));
3439 out:
3440 	srcu_read_unlock(&kvm->srcu, idx);
3441 	return r;
3442 }
3443 
3444 static int init_rmode_identity_map(struct kvm *kvm)
3445 {
3446 	struct kvm_vmx *kvm_vmx = to_kvm_vmx(kvm);
3447 	int i, idx, r = 0;
3448 	kvm_pfn_t identity_map_pfn;
3449 	u32 tmp;
3450 
3451 	/* Protect kvm_vmx->ept_identity_pagetable_done. */
3452 	mutex_lock(&kvm->slots_lock);
3453 
3454 	if (likely(kvm_vmx->ept_identity_pagetable_done))
3455 		goto out2;
3456 
3457 	if (!kvm_vmx->ept_identity_map_addr)
3458 		kvm_vmx->ept_identity_map_addr = VMX_EPT_IDENTITY_PAGETABLE_ADDR;
3459 	identity_map_pfn = kvm_vmx->ept_identity_map_addr >> PAGE_SHIFT;
3460 
3461 	r = __x86_set_memory_region(kvm, IDENTITY_PAGETABLE_PRIVATE_MEMSLOT,
3462 				    kvm_vmx->ept_identity_map_addr, PAGE_SIZE);
3463 	if (r < 0)
3464 		goto out2;
3465 
3466 	idx = srcu_read_lock(&kvm->srcu);
3467 	r = kvm_clear_guest_page(kvm, identity_map_pfn, 0, PAGE_SIZE);
3468 	if (r < 0)
3469 		goto out;
3470 	/* Set up identity-mapping pagetable for EPT in real mode */
3471 	for (i = 0; i < PT32_ENT_PER_PAGE; i++) {
3472 		tmp = (i << 22) + (_PAGE_PRESENT | _PAGE_RW | _PAGE_USER |
3473 			_PAGE_ACCESSED | _PAGE_DIRTY | _PAGE_PSE);
3474 		r = kvm_write_guest_page(kvm, identity_map_pfn,
3475 				&tmp, i * sizeof(tmp), sizeof(tmp));
3476 		if (r < 0)
3477 			goto out;
3478 	}
3479 	kvm_vmx->ept_identity_pagetable_done = true;
3480 
3481 out:
3482 	srcu_read_unlock(&kvm->srcu, idx);
3483 
3484 out2:
3485 	mutex_unlock(&kvm->slots_lock);
3486 	return r;
3487 }
3488 
3489 static void seg_setup(int seg)
3490 {
3491 	const struct kvm_vmx_segment_field *sf = &kvm_vmx_segment_fields[seg];
3492 	unsigned int ar;
3493 
3494 	vmcs_write16(sf->selector, 0);
3495 	vmcs_writel(sf->base, 0);
3496 	vmcs_write32(sf->limit, 0xffff);
3497 	ar = 0x93;
3498 	if (seg == VCPU_SREG_CS)
3499 		ar |= 0x08; /* code segment */
3500 
3501 	vmcs_write32(sf->ar_bytes, ar);
3502 }
3503 
3504 static int alloc_apic_access_page(struct kvm *kvm)
3505 {
3506 	struct page *page;
3507 	int r = 0;
3508 
3509 	mutex_lock(&kvm->slots_lock);
3510 	if (kvm->arch.apic_access_page_done)
3511 		goto out;
3512 	r = __x86_set_memory_region(kvm, APIC_ACCESS_PAGE_PRIVATE_MEMSLOT,
3513 				    APIC_DEFAULT_PHYS_BASE, PAGE_SIZE);
3514 	if (r)
3515 		goto out;
3516 
3517 	page = gfn_to_page(kvm, APIC_DEFAULT_PHYS_BASE >> PAGE_SHIFT);
3518 	if (is_error_page(page)) {
3519 		r = -EFAULT;
3520 		goto out;
3521 	}
3522 
3523 	/*
3524 	 * Do not pin the page in memory, so that memory hot-unplug
3525 	 * is able to migrate it.
3526 	 */
3527 	put_page(page);
3528 	kvm->arch.apic_access_page_done = true;
3529 out:
3530 	mutex_unlock(&kvm->slots_lock);
3531 	return r;
3532 }
3533 
3534 int allocate_vpid(void)
3535 {
3536 	int vpid;
3537 
3538 	if (!enable_vpid)
3539 		return 0;
3540 	spin_lock(&vmx_vpid_lock);
3541 	vpid = find_first_zero_bit(vmx_vpid_bitmap, VMX_NR_VPIDS);
3542 	if (vpid < VMX_NR_VPIDS)
3543 		__set_bit(vpid, vmx_vpid_bitmap);
3544 	else
3545 		vpid = 0;
3546 	spin_unlock(&vmx_vpid_lock);
3547 	return vpid;
3548 }
3549 
3550 void free_vpid(int vpid)
3551 {
3552 	if (!enable_vpid || vpid == 0)
3553 		return;
3554 	spin_lock(&vmx_vpid_lock);
3555 	__clear_bit(vpid, vmx_vpid_bitmap);
3556 	spin_unlock(&vmx_vpid_lock);
3557 }
3558 
3559 static __always_inline void vmx_disable_intercept_for_msr(unsigned long *msr_bitmap,
3560 							  u32 msr, int type)
3561 {
3562 	int f = sizeof(unsigned long);
3563 
3564 	if (!cpu_has_vmx_msr_bitmap())
3565 		return;
3566 
3567 	if (static_branch_unlikely(&enable_evmcs))
3568 		evmcs_touch_msr_bitmap();
3569 
3570 	/*
3571 	 * See Intel PRM Vol. 3, 20.6.9 (MSR-Bitmap Address). Early manuals
3572 	 * have the write-low and read-high bitmap offsets the wrong way round.
3573 	 * We can control MSRs 0x00000000-0x00001fff and 0xc0000000-0xc0001fff.
3574 	 */
3575 	if (msr <= 0x1fff) {
3576 		if (type & MSR_TYPE_R)
3577 			/* read-low */
3578 			__clear_bit(msr, msr_bitmap + 0x000 / f);
3579 
3580 		if (type & MSR_TYPE_W)
3581 			/* write-low */
3582 			__clear_bit(msr, msr_bitmap + 0x800 / f);
3583 
3584 	} else if ((msr >= 0xc0000000) && (msr <= 0xc0001fff)) {
3585 		msr &= 0x1fff;
3586 		if (type & MSR_TYPE_R)
3587 			/* read-high */
3588 			__clear_bit(msr, msr_bitmap + 0x400 / f);
3589 
3590 		if (type & MSR_TYPE_W)
3591 			/* write-high */
3592 			__clear_bit(msr, msr_bitmap + 0xc00 / f);
3593 
3594 	}
3595 }
3596 
3597 static __always_inline void vmx_enable_intercept_for_msr(unsigned long *msr_bitmap,
3598 							 u32 msr, int type)
3599 {
3600 	int f = sizeof(unsigned long);
3601 
3602 	if (!cpu_has_vmx_msr_bitmap())
3603 		return;
3604 
3605 	if (static_branch_unlikely(&enable_evmcs))
3606 		evmcs_touch_msr_bitmap();
3607 
3608 	/*
3609 	 * See Intel PRM Vol. 3, 20.6.9 (MSR-Bitmap Address). Early manuals
3610 	 * have the write-low and read-high bitmap offsets the wrong way round.
3611 	 * We can control MSRs 0x00000000-0x00001fff and 0xc0000000-0xc0001fff.
3612 	 */
3613 	if (msr <= 0x1fff) {
3614 		if (type & MSR_TYPE_R)
3615 			/* read-low */
3616 			__set_bit(msr, msr_bitmap + 0x000 / f);
3617 
3618 		if (type & MSR_TYPE_W)
3619 			/* write-low */
3620 			__set_bit(msr, msr_bitmap + 0x800 / f);
3621 
3622 	} else if ((msr >= 0xc0000000) && (msr <= 0xc0001fff)) {
3623 		msr &= 0x1fff;
3624 		if (type & MSR_TYPE_R)
3625 			/* read-high */
3626 			__set_bit(msr, msr_bitmap + 0x400 / f);
3627 
3628 		if (type & MSR_TYPE_W)
3629 			/* write-high */
3630 			__set_bit(msr, msr_bitmap + 0xc00 / f);
3631 
3632 	}
3633 }
3634 
3635 static __always_inline void vmx_set_intercept_for_msr(unsigned long *msr_bitmap,
3636 			     			      u32 msr, int type, bool value)
3637 {
3638 	if (value)
3639 		vmx_enable_intercept_for_msr(msr_bitmap, msr, type);
3640 	else
3641 		vmx_disable_intercept_for_msr(msr_bitmap, msr, type);
3642 }
3643 
3644 static u8 vmx_msr_bitmap_mode(struct kvm_vcpu *vcpu)
3645 {
3646 	u8 mode = 0;
3647 
3648 	if (cpu_has_secondary_exec_ctrls() &&
3649 	    (secondary_exec_controls_get(to_vmx(vcpu)) &
3650 	     SECONDARY_EXEC_VIRTUALIZE_X2APIC_MODE)) {
3651 		mode |= MSR_BITMAP_MODE_X2APIC;
3652 		if (enable_apicv && kvm_vcpu_apicv_active(vcpu))
3653 			mode |= MSR_BITMAP_MODE_X2APIC_APICV;
3654 	}
3655 
3656 	return mode;
3657 }
3658 
3659 static void vmx_update_msr_bitmap_x2apic(unsigned long *msr_bitmap,
3660 					 u8 mode)
3661 {
3662 	int msr;
3663 
3664 	for (msr = 0x800; msr <= 0x8ff; msr += BITS_PER_LONG) {
3665 		unsigned word = msr / BITS_PER_LONG;
3666 		msr_bitmap[word] = (mode & MSR_BITMAP_MODE_X2APIC_APICV) ? 0 : ~0;
3667 		msr_bitmap[word + (0x800 / sizeof(long))] = ~0;
3668 	}
3669 
3670 	if (mode & MSR_BITMAP_MODE_X2APIC) {
3671 		/*
3672 		 * TPR reads and writes can be virtualized even if virtual interrupt
3673 		 * delivery is not in use.
3674 		 */
3675 		vmx_disable_intercept_for_msr(msr_bitmap, X2APIC_MSR(APIC_TASKPRI), MSR_TYPE_RW);
3676 		if (mode & MSR_BITMAP_MODE_X2APIC_APICV) {
3677 			vmx_enable_intercept_for_msr(msr_bitmap, X2APIC_MSR(APIC_TMCCT), MSR_TYPE_R);
3678 			vmx_disable_intercept_for_msr(msr_bitmap, X2APIC_MSR(APIC_EOI), MSR_TYPE_W);
3679 			vmx_disable_intercept_for_msr(msr_bitmap, X2APIC_MSR(APIC_SELF_IPI), MSR_TYPE_W);
3680 		}
3681 	}
3682 }
3683 
3684 void vmx_update_msr_bitmap(struct kvm_vcpu *vcpu)
3685 {
3686 	struct vcpu_vmx *vmx = to_vmx(vcpu);
3687 	unsigned long *msr_bitmap = vmx->vmcs01.msr_bitmap;
3688 	u8 mode = vmx_msr_bitmap_mode(vcpu);
3689 	u8 changed = mode ^ vmx->msr_bitmap_mode;
3690 
3691 	if (!changed)
3692 		return;
3693 
3694 	if (changed & (MSR_BITMAP_MODE_X2APIC | MSR_BITMAP_MODE_X2APIC_APICV))
3695 		vmx_update_msr_bitmap_x2apic(msr_bitmap, mode);
3696 
3697 	vmx->msr_bitmap_mode = mode;
3698 }
3699 
3700 void pt_update_intercept_for_msr(struct vcpu_vmx *vmx)
3701 {
3702 	unsigned long *msr_bitmap = vmx->vmcs01.msr_bitmap;
3703 	bool flag = !(vmx->pt_desc.guest.ctl & RTIT_CTL_TRACEEN);
3704 	u32 i;
3705 
3706 	vmx_set_intercept_for_msr(msr_bitmap, MSR_IA32_RTIT_STATUS,
3707 							MSR_TYPE_RW, flag);
3708 	vmx_set_intercept_for_msr(msr_bitmap, MSR_IA32_RTIT_OUTPUT_BASE,
3709 							MSR_TYPE_RW, flag);
3710 	vmx_set_intercept_for_msr(msr_bitmap, MSR_IA32_RTIT_OUTPUT_MASK,
3711 							MSR_TYPE_RW, flag);
3712 	vmx_set_intercept_for_msr(msr_bitmap, MSR_IA32_RTIT_CR3_MATCH,
3713 							MSR_TYPE_RW, flag);
3714 	for (i = 0; i < vmx->pt_desc.addr_range; i++) {
3715 		vmx_set_intercept_for_msr(msr_bitmap,
3716 			MSR_IA32_RTIT_ADDR0_A + i * 2, MSR_TYPE_RW, flag);
3717 		vmx_set_intercept_for_msr(msr_bitmap,
3718 			MSR_IA32_RTIT_ADDR0_B + i * 2, MSR_TYPE_RW, flag);
3719 	}
3720 }
3721 
3722 static bool vmx_get_enable_apicv(struct kvm *kvm)
3723 {
3724 	return enable_apicv;
3725 }
3726 
3727 static bool vmx_guest_apic_has_interrupt(struct kvm_vcpu *vcpu)
3728 {
3729 	struct vcpu_vmx *vmx = to_vmx(vcpu);
3730 	void *vapic_page;
3731 	u32 vppr;
3732 	int rvi;
3733 
3734 	if (WARN_ON_ONCE(!is_guest_mode(vcpu)) ||
3735 		!nested_cpu_has_vid(get_vmcs12(vcpu)) ||
3736 		WARN_ON_ONCE(!vmx->nested.virtual_apic_map.gfn))
3737 		return false;
3738 
3739 	rvi = vmx_get_rvi();
3740 
3741 	vapic_page = vmx->nested.virtual_apic_map.hva;
3742 	vppr = *((u32 *)(vapic_page + APIC_PROCPRI));
3743 
3744 	return ((rvi & 0xf0) > (vppr & 0xf0));
3745 }
3746 
3747 static inline bool kvm_vcpu_trigger_posted_interrupt(struct kvm_vcpu *vcpu,
3748 						     bool nested)
3749 {
3750 #ifdef CONFIG_SMP
3751 	int pi_vec = nested ? POSTED_INTR_NESTED_VECTOR : POSTED_INTR_VECTOR;
3752 
3753 	if (vcpu->mode == IN_GUEST_MODE) {
3754 		/*
3755 		 * The vector of interrupt to be delivered to vcpu had
3756 		 * been set in PIR before this function.
3757 		 *
3758 		 * Following cases will be reached in this block, and
3759 		 * we always send a notification event in all cases as
3760 		 * explained below.
3761 		 *
3762 		 * Case 1: vcpu keeps in non-root mode. Sending a
3763 		 * notification event posts the interrupt to vcpu.
3764 		 *
3765 		 * Case 2: vcpu exits to root mode and is still
3766 		 * runnable. PIR will be synced to vIRR before the
3767 		 * next vcpu entry. Sending a notification event in
3768 		 * this case has no effect, as vcpu is not in root
3769 		 * mode.
3770 		 *
3771 		 * Case 3: vcpu exits to root mode and is blocked.
3772 		 * vcpu_block() has already synced PIR to vIRR and
3773 		 * never blocks vcpu if vIRR is not cleared. Therefore,
3774 		 * a blocked vcpu here does not wait for any requested
3775 		 * interrupts in PIR, and sending a notification event
3776 		 * which has no effect is safe here.
3777 		 */
3778 
3779 		apic->send_IPI_mask(get_cpu_mask(vcpu->cpu), pi_vec);
3780 		return true;
3781 	}
3782 #endif
3783 	return false;
3784 }
3785 
3786 static int vmx_deliver_nested_posted_interrupt(struct kvm_vcpu *vcpu,
3787 						int vector)
3788 {
3789 	struct vcpu_vmx *vmx = to_vmx(vcpu);
3790 
3791 	if (is_guest_mode(vcpu) &&
3792 	    vector == vmx->nested.posted_intr_nv) {
3793 		/*
3794 		 * If a posted intr is not recognized by hardware,
3795 		 * we will accomplish it in the next vmentry.
3796 		 */
3797 		vmx->nested.pi_pending = true;
3798 		kvm_make_request(KVM_REQ_EVENT, vcpu);
3799 		/* the PIR and ON have been set by L1. */
3800 		if (!kvm_vcpu_trigger_posted_interrupt(vcpu, true))
3801 			kvm_vcpu_kick(vcpu);
3802 		return 0;
3803 	}
3804 	return -1;
3805 }
3806 /*
3807  * Send interrupt to vcpu via posted interrupt way.
3808  * 1. If target vcpu is running(non-root mode), send posted interrupt
3809  * notification to vcpu and hardware will sync PIR to vIRR atomically.
3810  * 2. If target vcpu isn't running(root mode), kick it to pick up the
3811  * interrupt from PIR in next vmentry.
3812  */
3813 static void vmx_deliver_posted_interrupt(struct kvm_vcpu *vcpu, int vector)
3814 {
3815 	struct vcpu_vmx *vmx = to_vmx(vcpu);
3816 	int r;
3817 
3818 	r = vmx_deliver_nested_posted_interrupt(vcpu, vector);
3819 	if (!r)
3820 		return;
3821 
3822 	if (pi_test_and_set_pir(vector, &vmx->pi_desc))
3823 		return;
3824 
3825 	/* If a previous notification has sent the IPI, nothing to do.  */
3826 	if (pi_test_and_set_on(&vmx->pi_desc))
3827 		return;
3828 
3829 	if (!kvm_vcpu_trigger_posted_interrupt(vcpu, false))
3830 		kvm_vcpu_kick(vcpu);
3831 }
3832 
3833 /*
3834  * Set up the vmcs's constant host-state fields, i.e., host-state fields that
3835  * will not change in the lifetime of the guest.
3836  * Note that host-state that does change is set elsewhere. E.g., host-state
3837  * that is set differently for each CPU is set in vmx_vcpu_load(), not here.
3838  */
3839 void vmx_set_constant_host_state(struct vcpu_vmx *vmx)
3840 {
3841 	u32 low32, high32;
3842 	unsigned long tmpl;
3843 	unsigned long cr0, cr3, cr4;
3844 
3845 	cr0 = read_cr0();
3846 	WARN_ON(cr0 & X86_CR0_TS);
3847 	vmcs_writel(HOST_CR0, cr0);  /* 22.2.3 */
3848 
3849 	/*
3850 	 * Save the most likely value for this task's CR3 in the VMCS.
3851 	 * We can't use __get_current_cr3_fast() because we're not atomic.
3852 	 */
3853 	cr3 = __read_cr3();
3854 	vmcs_writel(HOST_CR3, cr3);		/* 22.2.3  FIXME: shadow tables */
3855 	vmx->loaded_vmcs->host_state.cr3 = cr3;
3856 
3857 	/* Save the most likely value for this task's CR4 in the VMCS. */
3858 	cr4 = cr4_read_shadow();
3859 	vmcs_writel(HOST_CR4, cr4);			/* 22.2.3, 22.2.5 */
3860 	vmx->loaded_vmcs->host_state.cr4 = cr4;
3861 
3862 	vmcs_write16(HOST_CS_SELECTOR, __KERNEL_CS);  /* 22.2.4 */
3863 #ifdef CONFIG_X86_64
3864 	/*
3865 	 * Load null selectors, so we can avoid reloading them in
3866 	 * vmx_prepare_switch_to_host(), in case userspace uses
3867 	 * the null selectors too (the expected case).
3868 	 */
3869 	vmcs_write16(HOST_DS_SELECTOR, 0);
3870 	vmcs_write16(HOST_ES_SELECTOR, 0);
3871 #else
3872 	vmcs_write16(HOST_DS_SELECTOR, __KERNEL_DS);  /* 22.2.4 */
3873 	vmcs_write16(HOST_ES_SELECTOR, __KERNEL_DS);  /* 22.2.4 */
3874 #endif
3875 	vmcs_write16(HOST_SS_SELECTOR, __KERNEL_DS);  /* 22.2.4 */
3876 	vmcs_write16(HOST_TR_SELECTOR, GDT_ENTRY_TSS*8);  /* 22.2.4 */
3877 
3878 	vmcs_writel(HOST_IDTR_BASE, host_idt_base);   /* 22.2.4 */
3879 
3880 	vmcs_writel(HOST_RIP, (unsigned long)vmx_vmexit); /* 22.2.5 */
3881 
3882 	rdmsr(MSR_IA32_SYSENTER_CS, low32, high32);
3883 	vmcs_write32(HOST_IA32_SYSENTER_CS, low32);
3884 	rdmsrl(MSR_IA32_SYSENTER_EIP, tmpl);
3885 	vmcs_writel(HOST_IA32_SYSENTER_EIP, tmpl);   /* 22.2.3 */
3886 
3887 	if (vmcs_config.vmexit_ctrl & VM_EXIT_LOAD_IA32_PAT) {
3888 		rdmsr(MSR_IA32_CR_PAT, low32, high32);
3889 		vmcs_write64(HOST_IA32_PAT, low32 | ((u64) high32 << 32));
3890 	}
3891 
3892 	if (cpu_has_load_ia32_efer())
3893 		vmcs_write64(HOST_IA32_EFER, host_efer);
3894 }
3895 
3896 void set_cr4_guest_host_mask(struct vcpu_vmx *vmx)
3897 {
3898 	vmx->vcpu.arch.cr4_guest_owned_bits = KVM_CR4_GUEST_OWNED_BITS;
3899 	if (enable_ept)
3900 		vmx->vcpu.arch.cr4_guest_owned_bits |= X86_CR4_PGE;
3901 	if (is_guest_mode(&vmx->vcpu))
3902 		vmx->vcpu.arch.cr4_guest_owned_bits &=
3903 			~get_vmcs12(&vmx->vcpu)->cr4_guest_host_mask;
3904 	vmcs_writel(CR4_GUEST_HOST_MASK, ~vmx->vcpu.arch.cr4_guest_owned_bits);
3905 }
3906 
3907 u32 vmx_pin_based_exec_ctrl(struct vcpu_vmx *vmx)
3908 {
3909 	u32 pin_based_exec_ctrl = vmcs_config.pin_based_exec_ctrl;
3910 
3911 	if (!kvm_vcpu_apicv_active(&vmx->vcpu))
3912 		pin_based_exec_ctrl &= ~PIN_BASED_POSTED_INTR;
3913 
3914 	if (!enable_vnmi)
3915 		pin_based_exec_ctrl &= ~PIN_BASED_VIRTUAL_NMIS;
3916 
3917 	if (!enable_preemption_timer)
3918 		pin_based_exec_ctrl &= ~PIN_BASED_VMX_PREEMPTION_TIMER;
3919 
3920 	return pin_based_exec_ctrl;
3921 }
3922 
3923 static void vmx_refresh_apicv_exec_ctrl(struct kvm_vcpu *vcpu)
3924 {
3925 	struct vcpu_vmx *vmx = to_vmx(vcpu);
3926 
3927 	pin_controls_set(vmx, vmx_pin_based_exec_ctrl(vmx));
3928 	if (cpu_has_secondary_exec_ctrls()) {
3929 		if (kvm_vcpu_apicv_active(vcpu))
3930 			secondary_exec_controls_setbit(vmx,
3931 				      SECONDARY_EXEC_APIC_REGISTER_VIRT |
3932 				      SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY);
3933 		else
3934 			secondary_exec_controls_clearbit(vmx,
3935 					SECONDARY_EXEC_APIC_REGISTER_VIRT |
3936 					SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY);
3937 	}
3938 
3939 	if (cpu_has_vmx_msr_bitmap())
3940 		vmx_update_msr_bitmap(vcpu);
3941 }
3942 
3943 u32 vmx_exec_control(struct vcpu_vmx *vmx)
3944 {
3945 	u32 exec_control = vmcs_config.cpu_based_exec_ctrl;
3946 
3947 	if (vmx->vcpu.arch.switch_db_regs & KVM_DEBUGREG_WONT_EXIT)
3948 		exec_control &= ~CPU_BASED_MOV_DR_EXITING;
3949 
3950 	if (!cpu_need_tpr_shadow(&vmx->vcpu)) {
3951 		exec_control &= ~CPU_BASED_TPR_SHADOW;
3952 #ifdef CONFIG_X86_64
3953 		exec_control |= CPU_BASED_CR8_STORE_EXITING |
3954 				CPU_BASED_CR8_LOAD_EXITING;
3955 #endif
3956 	}
3957 	if (!enable_ept)
3958 		exec_control |= CPU_BASED_CR3_STORE_EXITING |
3959 				CPU_BASED_CR3_LOAD_EXITING  |
3960 				CPU_BASED_INVLPG_EXITING;
3961 	if (kvm_mwait_in_guest(vmx->vcpu.kvm))
3962 		exec_control &= ~(CPU_BASED_MWAIT_EXITING |
3963 				CPU_BASED_MONITOR_EXITING);
3964 	if (kvm_hlt_in_guest(vmx->vcpu.kvm))
3965 		exec_control &= ~CPU_BASED_HLT_EXITING;
3966 	return exec_control;
3967 }
3968 
3969 
3970 static void vmx_compute_secondary_exec_control(struct vcpu_vmx *vmx)
3971 {
3972 	struct kvm_vcpu *vcpu = &vmx->vcpu;
3973 
3974 	u32 exec_control = vmcs_config.cpu_based_2nd_exec_ctrl;
3975 
3976 	if (pt_mode == PT_MODE_SYSTEM)
3977 		exec_control &= ~(SECONDARY_EXEC_PT_USE_GPA | SECONDARY_EXEC_PT_CONCEAL_VMX);
3978 	if (!cpu_need_virtualize_apic_accesses(vcpu))
3979 		exec_control &= ~SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES;
3980 	if (vmx->vpid == 0)
3981 		exec_control &= ~SECONDARY_EXEC_ENABLE_VPID;
3982 	if (!enable_ept) {
3983 		exec_control &= ~SECONDARY_EXEC_ENABLE_EPT;
3984 		enable_unrestricted_guest = 0;
3985 	}
3986 	if (!enable_unrestricted_guest)
3987 		exec_control &= ~SECONDARY_EXEC_UNRESTRICTED_GUEST;
3988 	if (kvm_pause_in_guest(vmx->vcpu.kvm))
3989 		exec_control &= ~SECONDARY_EXEC_PAUSE_LOOP_EXITING;
3990 	if (!kvm_vcpu_apicv_active(vcpu))
3991 		exec_control &= ~(SECONDARY_EXEC_APIC_REGISTER_VIRT |
3992 				  SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY);
3993 	exec_control &= ~SECONDARY_EXEC_VIRTUALIZE_X2APIC_MODE;
3994 
3995 	/* SECONDARY_EXEC_DESC is enabled/disabled on writes to CR4.UMIP,
3996 	 * in vmx_set_cr4.  */
3997 	exec_control &= ~SECONDARY_EXEC_DESC;
3998 
3999 	/* SECONDARY_EXEC_SHADOW_VMCS is enabled when L1 executes VMPTRLD
4000 	   (handle_vmptrld).
4001 	   We can NOT enable shadow_vmcs here because we don't have yet
4002 	   a current VMCS12
4003 	*/
4004 	exec_control &= ~SECONDARY_EXEC_SHADOW_VMCS;
4005 
4006 	if (!enable_pml)
4007 		exec_control &= ~SECONDARY_EXEC_ENABLE_PML;
4008 
4009 	if (vmx_xsaves_supported()) {
4010 		/* Exposing XSAVES only when XSAVE is exposed */
4011 		bool xsaves_enabled =
4012 			guest_cpuid_has(vcpu, X86_FEATURE_XSAVE) &&
4013 			guest_cpuid_has(vcpu, X86_FEATURE_XSAVES);
4014 
4015 		vcpu->arch.xsaves_enabled = xsaves_enabled;
4016 
4017 		if (!xsaves_enabled)
4018 			exec_control &= ~SECONDARY_EXEC_XSAVES;
4019 
4020 		if (nested) {
4021 			if (xsaves_enabled)
4022 				vmx->nested.msrs.secondary_ctls_high |=
4023 					SECONDARY_EXEC_XSAVES;
4024 			else
4025 				vmx->nested.msrs.secondary_ctls_high &=
4026 					~SECONDARY_EXEC_XSAVES;
4027 		}
4028 	}
4029 
4030 	if (vmx_rdtscp_supported()) {
4031 		bool rdtscp_enabled = guest_cpuid_has(vcpu, X86_FEATURE_RDTSCP);
4032 		if (!rdtscp_enabled)
4033 			exec_control &= ~SECONDARY_EXEC_RDTSCP;
4034 
4035 		if (nested) {
4036 			if (rdtscp_enabled)
4037 				vmx->nested.msrs.secondary_ctls_high |=
4038 					SECONDARY_EXEC_RDTSCP;
4039 			else
4040 				vmx->nested.msrs.secondary_ctls_high &=
4041 					~SECONDARY_EXEC_RDTSCP;
4042 		}
4043 	}
4044 
4045 	if (vmx_invpcid_supported()) {
4046 		/* Exposing INVPCID only when PCID is exposed */
4047 		bool invpcid_enabled =
4048 			guest_cpuid_has(vcpu, X86_FEATURE_INVPCID) &&
4049 			guest_cpuid_has(vcpu, X86_FEATURE_PCID);
4050 
4051 		if (!invpcid_enabled) {
4052 			exec_control &= ~SECONDARY_EXEC_ENABLE_INVPCID;
4053 			guest_cpuid_clear(vcpu, X86_FEATURE_INVPCID);
4054 		}
4055 
4056 		if (nested) {
4057 			if (invpcid_enabled)
4058 				vmx->nested.msrs.secondary_ctls_high |=
4059 					SECONDARY_EXEC_ENABLE_INVPCID;
4060 			else
4061 				vmx->nested.msrs.secondary_ctls_high &=
4062 					~SECONDARY_EXEC_ENABLE_INVPCID;
4063 		}
4064 	}
4065 
4066 	if (vmx_rdrand_supported()) {
4067 		bool rdrand_enabled = guest_cpuid_has(vcpu, X86_FEATURE_RDRAND);
4068 		if (rdrand_enabled)
4069 			exec_control &= ~SECONDARY_EXEC_RDRAND_EXITING;
4070 
4071 		if (nested) {
4072 			if (rdrand_enabled)
4073 				vmx->nested.msrs.secondary_ctls_high |=
4074 					SECONDARY_EXEC_RDRAND_EXITING;
4075 			else
4076 				vmx->nested.msrs.secondary_ctls_high &=
4077 					~SECONDARY_EXEC_RDRAND_EXITING;
4078 		}
4079 	}
4080 
4081 	if (vmx_rdseed_supported()) {
4082 		bool rdseed_enabled = guest_cpuid_has(vcpu, X86_FEATURE_RDSEED);
4083 		if (rdseed_enabled)
4084 			exec_control &= ~SECONDARY_EXEC_RDSEED_EXITING;
4085 
4086 		if (nested) {
4087 			if (rdseed_enabled)
4088 				vmx->nested.msrs.secondary_ctls_high |=
4089 					SECONDARY_EXEC_RDSEED_EXITING;
4090 			else
4091 				vmx->nested.msrs.secondary_ctls_high &=
4092 					~SECONDARY_EXEC_RDSEED_EXITING;
4093 		}
4094 	}
4095 
4096 	if (vmx_waitpkg_supported()) {
4097 		bool waitpkg_enabled =
4098 			guest_cpuid_has(vcpu, X86_FEATURE_WAITPKG);
4099 
4100 		if (!waitpkg_enabled)
4101 			exec_control &= ~SECONDARY_EXEC_ENABLE_USR_WAIT_PAUSE;
4102 
4103 		if (nested) {
4104 			if (waitpkg_enabled)
4105 				vmx->nested.msrs.secondary_ctls_high |=
4106 					SECONDARY_EXEC_ENABLE_USR_WAIT_PAUSE;
4107 			else
4108 				vmx->nested.msrs.secondary_ctls_high &=
4109 					~SECONDARY_EXEC_ENABLE_USR_WAIT_PAUSE;
4110 		}
4111 	}
4112 
4113 	vmx->secondary_exec_control = exec_control;
4114 }
4115 
4116 static void ept_set_mmio_spte_mask(void)
4117 {
4118 	/*
4119 	 * EPT Misconfigurations can be generated if the value of bits 2:0
4120 	 * of an EPT paging-structure entry is 110b (write/execute).
4121 	 */
4122 	kvm_mmu_set_mmio_spte_mask(VMX_EPT_RWX_MASK,
4123 				   VMX_EPT_MISCONFIG_WX_VALUE, 0);
4124 }
4125 
4126 #define VMX_XSS_EXIT_BITMAP 0
4127 
4128 /*
4129  * Noting that the initialization of Guest-state Area of VMCS is in
4130  * vmx_vcpu_reset().
4131  */
4132 static void init_vmcs(struct vcpu_vmx *vmx)
4133 {
4134 	if (nested)
4135 		nested_vmx_set_vmcs_shadowing_bitmap();
4136 
4137 	if (cpu_has_vmx_msr_bitmap())
4138 		vmcs_write64(MSR_BITMAP, __pa(vmx->vmcs01.msr_bitmap));
4139 
4140 	vmcs_write64(VMCS_LINK_POINTER, -1ull); /* 22.3.1.5 */
4141 
4142 	/* Control */
4143 	pin_controls_set(vmx, vmx_pin_based_exec_ctrl(vmx));
4144 
4145 	exec_controls_set(vmx, vmx_exec_control(vmx));
4146 
4147 	if (cpu_has_secondary_exec_ctrls()) {
4148 		vmx_compute_secondary_exec_control(vmx);
4149 		secondary_exec_controls_set(vmx, vmx->secondary_exec_control);
4150 	}
4151 
4152 	if (kvm_vcpu_apicv_active(&vmx->vcpu)) {
4153 		vmcs_write64(EOI_EXIT_BITMAP0, 0);
4154 		vmcs_write64(EOI_EXIT_BITMAP1, 0);
4155 		vmcs_write64(EOI_EXIT_BITMAP2, 0);
4156 		vmcs_write64(EOI_EXIT_BITMAP3, 0);
4157 
4158 		vmcs_write16(GUEST_INTR_STATUS, 0);
4159 
4160 		vmcs_write16(POSTED_INTR_NV, POSTED_INTR_VECTOR);
4161 		vmcs_write64(POSTED_INTR_DESC_ADDR, __pa((&vmx->pi_desc)));
4162 	}
4163 
4164 	if (!kvm_pause_in_guest(vmx->vcpu.kvm)) {
4165 		vmcs_write32(PLE_GAP, ple_gap);
4166 		vmx->ple_window = ple_window;
4167 		vmx->ple_window_dirty = true;
4168 	}
4169 
4170 	vmcs_write32(PAGE_FAULT_ERROR_CODE_MASK, 0);
4171 	vmcs_write32(PAGE_FAULT_ERROR_CODE_MATCH, 0);
4172 	vmcs_write32(CR3_TARGET_COUNT, 0);           /* 22.2.1 */
4173 
4174 	vmcs_write16(HOST_FS_SELECTOR, 0);            /* 22.2.4 */
4175 	vmcs_write16(HOST_GS_SELECTOR, 0);            /* 22.2.4 */
4176 	vmx_set_constant_host_state(vmx);
4177 	vmcs_writel(HOST_FS_BASE, 0); /* 22.2.4 */
4178 	vmcs_writel(HOST_GS_BASE, 0); /* 22.2.4 */
4179 
4180 	if (cpu_has_vmx_vmfunc())
4181 		vmcs_write64(VM_FUNCTION_CONTROL, 0);
4182 
4183 	vmcs_write32(VM_EXIT_MSR_STORE_COUNT, 0);
4184 	vmcs_write32(VM_EXIT_MSR_LOAD_COUNT, 0);
4185 	vmcs_write64(VM_EXIT_MSR_LOAD_ADDR, __pa(vmx->msr_autoload.host.val));
4186 	vmcs_write32(VM_ENTRY_MSR_LOAD_COUNT, 0);
4187 	vmcs_write64(VM_ENTRY_MSR_LOAD_ADDR, __pa(vmx->msr_autoload.guest.val));
4188 
4189 	if (vmcs_config.vmentry_ctrl & VM_ENTRY_LOAD_IA32_PAT)
4190 		vmcs_write64(GUEST_IA32_PAT, vmx->vcpu.arch.pat);
4191 
4192 	vm_exit_controls_set(vmx, vmx_vmexit_ctrl());
4193 
4194 	/* 22.2.1, 20.8.1 */
4195 	vm_entry_controls_set(vmx, vmx_vmentry_ctrl());
4196 
4197 	vmx->vcpu.arch.cr0_guest_owned_bits = X86_CR0_TS;
4198 	vmcs_writel(CR0_GUEST_HOST_MASK, ~X86_CR0_TS);
4199 
4200 	set_cr4_guest_host_mask(vmx);
4201 
4202 	if (vmx->vpid != 0)
4203 		vmcs_write16(VIRTUAL_PROCESSOR_ID, vmx->vpid);
4204 
4205 	if (vmx_xsaves_supported())
4206 		vmcs_write64(XSS_EXIT_BITMAP, VMX_XSS_EXIT_BITMAP);
4207 
4208 	if (enable_pml) {
4209 		vmcs_write64(PML_ADDRESS, page_to_phys(vmx->pml_pg));
4210 		vmcs_write16(GUEST_PML_INDEX, PML_ENTITY_NUM - 1);
4211 	}
4212 
4213 	if (cpu_has_vmx_encls_vmexit())
4214 		vmcs_write64(ENCLS_EXITING_BITMAP, -1ull);
4215 
4216 	if (pt_mode == PT_MODE_HOST_GUEST) {
4217 		memset(&vmx->pt_desc, 0, sizeof(vmx->pt_desc));
4218 		/* Bit[6~0] are forced to 1, writes are ignored. */
4219 		vmx->pt_desc.guest.output_mask = 0x7F;
4220 		vmcs_write64(GUEST_IA32_RTIT_CTL, 0);
4221 	}
4222 }
4223 
4224 static void vmx_vcpu_reset(struct kvm_vcpu *vcpu, bool init_event)
4225 {
4226 	struct vcpu_vmx *vmx = to_vmx(vcpu);
4227 	struct msr_data apic_base_msr;
4228 	u64 cr0;
4229 
4230 	vmx->rmode.vm86_active = 0;
4231 	vmx->spec_ctrl = 0;
4232 
4233 	vmx->msr_ia32_umwait_control = 0;
4234 
4235 	vcpu->arch.microcode_version = 0x100000000ULL;
4236 	vmx->vcpu.arch.regs[VCPU_REGS_RDX] = get_rdx_init_val();
4237 	vmx->hv_deadline_tsc = -1;
4238 	kvm_set_cr8(vcpu, 0);
4239 
4240 	if (!init_event) {
4241 		apic_base_msr.data = APIC_DEFAULT_PHYS_BASE |
4242 				     MSR_IA32_APICBASE_ENABLE;
4243 		if (kvm_vcpu_is_reset_bsp(vcpu))
4244 			apic_base_msr.data |= MSR_IA32_APICBASE_BSP;
4245 		apic_base_msr.host_initiated = true;
4246 		kvm_set_apic_base(vcpu, &apic_base_msr);
4247 	}
4248 
4249 	vmx_segment_cache_clear(vmx);
4250 
4251 	seg_setup(VCPU_SREG_CS);
4252 	vmcs_write16(GUEST_CS_SELECTOR, 0xf000);
4253 	vmcs_writel(GUEST_CS_BASE, 0xffff0000ul);
4254 
4255 	seg_setup(VCPU_SREG_DS);
4256 	seg_setup(VCPU_SREG_ES);
4257 	seg_setup(VCPU_SREG_FS);
4258 	seg_setup(VCPU_SREG_GS);
4259 	seg_setup(VCPU_SREG_SS);
4260 
4261 	vmcs_write16(GUEST_TR_SELECTOR, 0);
4262 	vmcs_writel(GUEST_TR_BASE, 0);
4263 	vmcs_write32(GUEST_TR_LIMIT, 0xffff);
4264 	vmcs_write32(GUEST_TR_AR_BYTES, 0x008b);
4265 
4266 	vmcs_write16(GUEST_LDTR_SELECTOR, 0);
4267 	vmcs_writel(GUEST_LDTR_BASE, 0);
4268 	vmcs_write32(GUEST_LDTR_LIMIT, 0xffff);
4269 	vmcs_write32(GUEST_LDTR_AR_BYTES, 0x00082);
4270 
4271 	if (!init_event) {
4272 		vmcs_write32(GUEST_SYSENTER_CS, 0);
4273 		vmcs_writel(GUEST_SYSENTER_ESP, 0);
4274 		vmcs_writel(GUEST_SYSENTER_EIP, 0);
4275 		vmcs_write64(GUEST_IA32_DEBUGCTL, 0);
4276 	}
4277 
4278 	kvm_set_rflags(vcpu, X86_EFLAGS_FIXED);
4279 	kvm_rip_write(vcpu, 0xfff0);
4280 
4281 	vmcs_writel(GUEST_GDTR_BASE, 0);
4282 	vmcs_write32(GUEST_GDTR_LIMIT, 0xffff);
4283 
4284 	vmcs_writel(GUEST_IDTR_BASE, 0);
4285 	vmcs_write32(GUEST_IDTR_LIMIT, 0xffff);
4286 
4287 	vmcs_write32(GUEST_ACTIVITY_STATE, GUEST_ACTIVITY_ACTIVE);
4288 	vmcs_write32(GUEST_INTERRUPTIBILITY_INFO, 0);
4289 	vmcs_writel(GUEST_PENDING_DBG_EXCEPTIONS, 0);
4290 	if (kvm_mpx_supported())
4291 		vmcs_write64(GUEST_BNDCFGS, 0);
4292 
4293 	setup_msrs(vmx);
4294 
4295 	vmcs_write32(VM_ENTRY_INTR_INFO_FIELD, 0);  /* 22.2.1 */
4296 
4297 	if (cpu_has_vmx_tpr_shadow() && !init_event) {
4298 		vmcs_write64(VIRTUAL_APIC_PAGE_ADDR, 0);
4299 		if (cpu_need_tpr_shadow(vcpu))
4300 			vmcs_write64(VIRTUAL_APIC_PAGE_ADDR,
4301 				     __pa(vcpu->arch.apic->regs));
4302 		vmcs_write32(TPR_THRESHOLD, 0);
4303 	}
4304 
4305 	kvm_make_request(KVM_REQ_APIC_PAGE_RELOAD, vcpu);
4306 
4307 	cr0 = X86_CR0_NW | X86_CR0_CD | X86_CR0_ET;
4308 	vmx->vcpu.arch.cr0 = cr0;
4309 	vmx_set_cr0(vcpu, cr0); /* enter rmode */
4310 	vmx_set_cr4(vcpu, 0);
4311 	vmx_set_efer(vcpu, 0);
4312 
4313 	update_exception_bitmap(vcpu);
4314 
4315 	vpid_sync_context(vmx->vpid);
4316 	if (init_event)
4317 		vmx_clear_hlt(vcpu);
4318 }
4319 
4320 static void enable_irq_window(struct kvm_vcpu *vcpu)
4321 {
4322 	exec_controls_setbit(to_vmx(vcpu), CPU_BASED_VIRTUAL_INTR_PENDING);
4323 }
4324 
4325 static void enable_nmi_window(struct kvm_vcpu *vcpu)
4326 {
4327 	if (!enable_vnmi ||
4328 	    vmcs_read32(GUEST_INTERRUPTIBILITY_INFO) & GUEST_INTR_STATE_STI) {
4329 		enable_irq_window(vcpu);
4330 		return;
4331 	}
4332 
4333 	exec_controls_setbit(to_vmx(vcpu), CPU_BASED_VIRTUAL_NMI_PENDING);
4334 }
4335 
4336 static void vmx_inject_irq(struct kvm_vcpu *vcpu)
4337 {
4338 	struct vcpu_vmx *vmx = to_vmx(vcpu);
4339 	uint32_t intr;
4340 	int irq = vcpu->arch.interrupt.nr;
4341 
4342 	trace_kvm_inj_virq(irq);
4343 
4344 	++vcpu->stat.irq_injections;
4345 	if (vmx->rmode.vm86_active) {
4346 		int inc_eip = 0;
4347 		if (vcpu->arch.interrupt.soft)
4348 			inc_eip = vcpu->arch.event_exit_inst_len;
4349 		kvm_inject_realmode_interrupt(vcpu, irq, inc_eip);
4350 		return;
4351 	}
4352 	intr = irq | INTR_INFO_VALID_MASK;
4353 	if (vcpu->arch.interrupt.soft) {
4354 		intr |= INTR_TYPE_SOFT_INTR;
4355 		vmcs_write32(VM_ENTRY_INSTRUCTION_LEN,
4356 			     vmx->vcpu.arch.event_exit_inst_len);
4357 	} else
4358 		intr |= INTR_TYPE_EXT_INTR;
4359 	vmcs_write32(VM_ENTRY_INTR_INFO_FIELD, intr);
4360 
4361 	vmx_clear_hlt(vcpu);
4362 }
4363 
4364 static void vmx_inject_nmi(struct kvm_vcpu *vcpu)
4365 {
4366 	struct vcpu_vmx *vmx = to_vmx(vcpu);
4367 
4368 	if (!enable_vnmi) {
4369 		/*
4370 		 * Tracking the NMI-blocked state in software is built upon
4371 		 * finding the next open IRQ window. This, in turn, depends on
4372 		 * well-behaving guests: They have to keep IRQs disabled at
4373 		 * least as long as the NMI handler runs. Otherwise we may
4374 		 * cause NMI nesting, maybe breaking the guest. But as this is
4375 		 * highly unlikely, we can live with the residual risk.
4376 		 */
4377 		vmx->loaded_vmcs->soft_vnmi_blocked = 1;
4378 		vmx->loaded_vmcs->vnmi_blocked_time = 0;
4379 	}
4380 
4381 	++vcpu->stat.nmi_injections;
4382 	vmx->loaded_vmcs->nmi_known_unmasked = false;
4383 
4384 	if (vmx->rmode.vm86_active) {
4385 		kvm_inject_realmode_interrupt(vcpu, NMI_VECTOR, 0);
4386 		return;
4387 	}
4388 
4389 	vmcs_write32(VM_ENTRY_INTR_INFO_FIELD,
4390 			INTR_TYPE_NMI_INTR | INTR_INFO_VALID_MASK | NMI_VECTOR);
4391 
4392 	vmx_clear_hlt(vcpu);
4393 }
4394 
4395 bool vmx_get_nmi_mask(struct kvm_vcpu *vcpu)
4396 {
4397 	struct vcpu_vmx *vmx = to_vmx(vcpu);
4398 	bool masked;
4399 
4400 	if (!enable_vnmi)
4401 		return vmx->loaded_vmcs->soft_vnmi_blocked;
4402 	if (vmx->loaded_vmcs->nmi_known_unmasked)
4403 		return false;
4404 	masked = vmcs_read32(GUEST_INTERRUPTIBILITY_INFO) & GUEST_INTR_STATE_NMI;
4405 	vmx->loaded_vmcs->nmi_known_unmasked = !masked;
4406 	return masked;
4407 }
4408 
4409 void vmx_set_nmi_mask(struct kvm_vcpu *vcpu, bool masked)
4410 {
4411 	struct vcpu_vmx *vmx = to_vmx(vcpu);
4412 
4413 	if (!enable_vnmi) {
4414 		if (vmx->loaded_vmcs->soft_vnmi_blocked != masked) {
4415 			vmx->loaded_vmcs->soft_vnmi_blocked = masked;
4416 			vmx->loaded_vmcs->vnmi_blocked_time = 0;
4417 		}
4418 	} else {
4419 		vmx->loaded_vmcs->nmi_known_unmasked = !masked;
4420 		if (masked)
4421 			vmcs_set_bits(GUEST_INTERRUPTIBILITY_INFO,
4422 				      GUEST_INTR_STATE_NMI);
4423 		else
4424 			vmcs_clear_bits(GUEST_INTERRUPTIBILITY_INFO,
4425 					GUEST_INTR_STATE_NMI);
4426 	}
4427 }
4428 
4429 static int vmx_nmi_allowed(struct kvm_vcpu *vcpu)
4430 {
4431 	if (to_vmx(vcpu)->nested.nested_run_pending)
4432 		return 0;
4433 
4434 	if (!enable_vnmi &&
4435 	    to_vmx(vcpu)->loaded_vmcs->soft_vnmi_blocked)
4436 		return 0;
4437 
4438 	return	!(vmcs_read32(GUEST_INTERRUPTIBILITY_INFO) &
4439 		  (GUEST_INTR_STATE_MOV_SS | GUEST_INTR_STATE_STI
4440 		   | GUEST_INTR_STATE_NMI));
4441 }
4442 
4443 static int vmx_interrupt_allowed(struct kvm_vcpu *vcpu)
4444 {
4445 	return (!to_vmx(vcpu)->nested.nested_run_pending &&
4446 		vmcs_readl(GUEST_RFLAGS) & X86_EFLAGS_IF) &&
4447 		!(vmcs_read32(GUEST_INTERRUPTIBILITY_INFO) &
4448 			(GUEST_INTR_STATE_STI | GUEST_INTR_STATE_MOV_SS));
4449 }
4450 
4451 static int vmx_set_tss_addr(struct kvm *kvm, unsigned int addr)
4452 {
4453 	int ret;
4454 
4455 	if (enable_unrestricted_guest)
4456 		return 0;
4457 
4458 	ret = x86_set_memory_region(kvm, TSS_PRIVATE_MEMSLOT, addr,
4459 				    PAGE_SIZE * 3);
4460 	if (ret)
4461 		return ret;
4462 	to_kvm_vmx(kvm)->tss_addr = addr;
4463 	return init_rmode_tss(kvm);
4464 }
4465 
4466 static int vmx_set_identity_map_addr(struct kvm *kvm, u64 ident_addr)
4467 {
4468 	to_kvm_vmx(kvm)->ept_identity_map_addr = ident_addr;
4469 	return 0;
4470 }
4471 
4472 static bool rmode_exception(struct kvm_vcpu *vcpu, int vec)
4473 {
4474 	switch (vec) {
4475 	case BP_VECTOR:
4476 		/*
4477 		 * Update instruction length as we may reinject the exception
4478 		 * from user space while in guest debugging mode.
4479 		 */
4480 		to_vmx(vcpu)->vcpu.arch.event_exit_inst_len =
4481 			vmcs_read32(VM_EXIT_INSTRUCTION_LEN);
4482 		if (vcpu->guest_debug & KVM_GUESTDBG_USE_SW_BP)
4483 			return false;
4484 		/* fall through */
4485 	case DB_VECTOR:
4486 		if (vcpu->guest_debug &
4487 			(KVM_GUESTDBG_SINGLESTEP | KVM_GUESTDBG_USE_HW_BP))
4488 			return false;
4489 		/* fall through */
4490 	case DE_VECTOR:
4491 	case OF_VECTOR:
4492 	case BR_VECTOR:
4493 	case UD_VECTOR:
4494 	case DF_VECTOR:
4495 	case SS_VECTOR:
4496 	case GP_VECTOR:
4497 	case MF_VECTOR:
4498 		return true;
4499 	break;
4500 	}
4501 	return false;
4502 }
4503 
4504 static int handle_rmode_exception(struct kvm_vcpu *vcpu,
4505 				  int vec, u32 err_code)
4506 {
4507 	/*
4508 	 * Instruction with address size override prefix opcode 0x67
4509 	 * Cause the #SS fault with 0 error code in VM86 mode.
4510 	 */
4511 	if (((vec == GP_VECTOR) || (vec == SS_VECTOR)) && err_code == 0) {
4512 		if (kvm_emulate_instruction(vcpu, 0)) {
4513 			if (vcpu->arch.halt_request) {
4514 				vcpu->arch.halt_request = 0;
4515 				return kvm_vcpu_halt(vcpu);
4516 			}
4517 			return 1;
4518 		}
4519 		return 0;
4520 	}
4521 
4522 	/*
4523 	 * Forward all other exceptions that are valid in real mode.
4524 	 * FIXME: Breaks guest debugging in real mode, needs to be fixed with
4525 	 *        the required debugging infrastructure rework.
4526 	 */
4527 	kvm_queue_exception(vcpu, vec);
4528 	return 1;
4529 }
4530 
4531 /*
4532  * Trigger machine check on the host. We assume all the MSRs are already set up
4533  * by the CPU and that we still run on the same CPU as the MCE occurred on.
4534  * We pass a fake environment to the machine check handler because we want
4535  * the guest to be always treated like user space, no matter what context
4536  * it used internally.
4537  */
4538 static void kvm_machine_check(void)
4539 {
4540 #if defined(CONFIG_X86_MCE) && defined(CONFIG_X86_64)
4541 	struct pt_regs regs = {
4542 		.cs = 3, /* Fake ring 3 no matter what the guest ran on */
4543 		.flags = X86_EFLAGS_IF,
4544 	};
4545 
4546 	do_machine_check(&regs, 0);
4547 #endif
4548 }
4549 
4550 static int handle_machine_check(struct kvm_vcpu *vcpu)
4551 {
4552 	/* handled by vmx_vcpu_run() */
4553 	return 1;
4554 }
4555 
4556 static int handle_exception_nmi(struct kvm_vcpu *vcpu)
4557 {
4558 	struct vcpu_vmx *vmx = to_vmx(vcpu);
4559 	struct kvm_run *kvm_run = vcpu->run;
4560 	u32 intr_info, ex_no, error_code;
4561 	unsigned long cr2, rip, dr6;
4562 	u32 vect_info;
4563 
4564 	vect_info = vmx->idt_vectoring_info;
4565 	intr_info = vmx->exit_intr_info;
4566 
4567 	if (is_machine_check(intr_info) || is_nmi(intr_info))
4568 		return 1; /* handled by handle_exception_nmi_irqoff() */
4569 
4570 	if (is_invalid_opcode(intr_info))
4571 		return handle_ud(vcpu);
4572 
4573 	error_code = 0;
4574 	if (intr_info & INTR_INFO_DELIVER_CODE_MASK)
4575 		error_code = vmcs_read32(VM_EXIT_INTR_ERROR_CODE);
4576 
4577 	if (!vmx->rmode.vm86_active && is_gp_fault(intr_info)) {
4578 		WARN_ON_ONCE(!enable_vmware_backdoor);
4579 
4580 		/*
4581 		 * VMware backdoor emulation on #GP interception only handles
4582 		 * IN{S}, OUT{S}, and RDPMC, none of which generate a non-zero
4583 		 * error code on #GP.
4584 		 */
4585 		if (error_code) {
4586 			kvm_queue_exception_e(vcpu, GP_VECTOR, error_code);
4587 			return 1;
4588 		}
4589 		return kvm_emulate_instruction(vcpu, EMULTYPE_VMWARE_GP);
4590 	}
4591 
4592 	/*
4593 	 * The #PF with PFEC.RSVD = 1 indicates the guest is accessing
4594 	 * MMIO, it is better to report an internal error.
4595 	 * See the comments in vmx_handle_exit.
4596 	 */
4597 	if ((vect_info & VECTORING_INFO_VALID_MASK) &&
4598 	    !(is_page_fault(intr_info) && !(error_code & PFERR_RSVD_MASK))) {
4599 		vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
4600 		vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_SIMUL_EX;
4601 		vcpu->run->internal.ndata = 3;
4602 		vcpu->run->internal.data[0] = vect_info;
4603 		vcpu->run->internal.data[1] = intr_info;
4604 		vcpu->run->internal.data[2] = error_code;
4605 		return 0;
4606 	}
4607 
4608 	if (is_page_fault(intr_info)) {
4609 		cr2 = vmcs_readl(EXIT_QUALIFICATION);
4610 		/* EPT won't cause page fault directly */
4611 		WARN_ON_ONCE(!vcpu->arch.apf.host_apf_reason && enable_ept);
4612 		return kvm_handle_page_fault(vcpu, error_code, cr2, NULL, 0);
4613 	}
4614 
4615 	ex_no = intr_info & INTR_INFO_VECTOR_MASK;
4616 
4617 	if (vmx->rmode.vm86_active && rmode_exception(vcpu, ex_no))
4618 		return handle_rmode_exception(vcpu, ex_no, error_code);
4619 
4620 	switch (ex_no) {
4621 	case AC_VECTOR:
4622 		kvm_queue_exception_e(vcpu, AC_VECTOR, error_code);
4623 		return 1;
4624 	case DB_VECTOR:
4625 		dr6 = vmcs_readl(EXIT_QUALIFICATION);
4626 		if (!(vcpu->guest_debug &
4627 		      (KVM_GUESTDBG_SINGLESTEP | KVM_GUESTDBG_USE_HW_BP))) {
4628 			vcpu->arch.dr6 &= ~DR_TRAP_BITS;
4629 			vcpu->arch.dr6 |= dr6 | DR6_RTM;
4630 			if (is_icebp(intr_info))
4631 				WARN_ON(!skip_emulated_instruction(vcpu));
4632 
4633 			kvm_queue_exception(vcpu, DB_VECTOR);
4634 			return 1;
4635 		}
4636 		kvm_run->debug.arch.dr6 = dr6 | DR6_FIXED_1;
4637 		kvm_run->debug.arch.dr7 = vmcs_readl(GUEST_DR7);
4638 		/* fall through */
4639 	case BP_VECTOR:
4640 		/*
4641 		 * Update instruction length as we may reinject #BP from
4642 		 * user space while in guest debugging mode. Reading it for
4643 		 * #DB as well causes no harm, it is not used in that case.
4644 		 */
4645 		vmx->vcpu.arch.event_exit_inst_len =
4646 			vmcs_read32(VM_EXIT_INSTRUCTION_LEN);
4647 		kvm_run->exit_reason = KVM_EXIT_DEBUG;
4648 		rip = kvm_rip_read(vcpu);
4649 		kvm_run->debug.arch.pc = vmcs_readl(GUEST_CS_BASE) + rip;
4650 		kvm_run->debug.arch.exception = ex_no;
4651 		break;
4652 	default:
4653 		kvm_run->exit_reason = KVM_EXIT_EXCEPTION;
4654 		kvm_run->ex.exception = ex_no;
4655 		kvm_run->ex.error_code = error_code;
4656 		break;
4657 	}
4658 	return 0;
4659 }
4660 
4661 static __always_inline int handle_external_interrupt(struct kvm_vcpu *vcpu)
4662 {
4663 	++vcpu->stat.irq_exits;
4664 	return 1;
4665 }
4666 
4667 static int handle_triple_fault(struct kvm_vcpu *vcpu)
4668 {
4669 	vcpu->run->exit_reason = KVM_EXIT_SHUTDOWN;
4670 	vcpu->mmio_needed = 0;
4671 	return 0;
4672 }
4673 
4674 static int handle_io(struct kvm_vcpu *vcpu)
4675 {
4676 	unsigned long exit_qualification;
4677 	int size, in, string;
4678 	unsigned port;
4679 
4680 	exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
4681 	string = (exit_qualification & 16) != 0;
4682 
4683 	++vcpu->stat.io_exits;
4684 
4685 	if (string)
4686 		return kvm_emulate_instruction(vcpu, 0);
4687 
4688 	port = exit_qualification >> 16;
4689 	size = (exit_qualification & 7) + 1;
4690 	in = (exit_qualification & 8) != 0;
4691 
4692 	return kvm_fast_pio(vcpu, size, port, in);
4693 }
4694 
4695 static void
4696 vmx_patch_hypercall(struct kvm_vcpu *vcpu, unsigned char *hypercall)
4697 {
4698 	/*
4699 	 * Patch in the VMCALL instruction:
4700 	 */
4701 	hypercall[0] = 0x0f;
4702 	hypercall[1] = 0x01;
4703 	hypercall[2] = 0xc1;
4704 }
4705 
4706 /* called to set cr0 as appropriate for a mov-to-cr0 exit. */
4707 static int handle_set_cr0(struct kvm_vcpu *vcpu, unsigned long val)
4708 {
4709 	if (is_guest_mode(vcpu)) {
4710 		struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
4711 		unsigned long orig_val = val;
4712 
4713 		/*
4714 		 * We get here when L2 changed cr0 in a way that did not change
4715 		 * any of L1's shadowed bits (see nested_vmx_exit_handled_cr),
4716 		 * but did change L0 shadowed bits. So we first calculate the
4717 		 * effective cr0 value that L1 would like to write into the
4718 		 * hardware. It consists of the L2-owned bits from the new
4719 		 * value combined with the L1-owned bits from L1's guest_cr0.
4720 		 */
4721 		val = (val & ~vmcs12->cr0_guest_host_mask) |
4722 			(vmcs12->guest_cr0 & vmcs12->cr0_guest_host_mask);
4723 
4724 		if (!nested_guest_cr0_valid(vcpu, val))
4725 			return 1;
4726 
4727 		if (kvm_set_cr0(vcpu, val))
4728 			return 1;
4729 		vmcs_writel(CR0_READ_SHADOW, orig_val);
4730 		return 0;
4731 	} else {
4732 		if (to_vmx(vcpu)->nested.vmxon &&
4733 		    !nested_host_cr0_valid(vcpu, val))
4734 			return 1;
4735 
4736 		return kvm_set_cr0(vcpu, val);
4737 	}
4738 }
4739 
4740 static int handle_set_cr4(struct kvm_vcpu *vcpu, unsigned long val)
4741 {
4742 	if (is_guest_mode(vcpu)) {
4743 		struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
4744 		unsigned long orig_val = val;
4745 
4746 		/* analogously to handle_set_cr0 */
4747 		val = (val & ~vmcs12->cr4_guest_host_mask) |
4748 			(vmcs12->guest_cr4 & vmcs12->cr4_guest_host_mask);
4749 		if (kvm_set_cr4(vcpu, val))
4750 			return 1;
4751 		vmcs_writel(CR4_READ_SHADOW, orig_val);
4752 		return 0;
4753 	} else
4754 		return kvm_set_cr4(vcpu, val);
4755 }
4756 
4757 static int handle_desc(struct kvm_vcpu *vcpu)
4758 {
4759 	WARN_ON(!(vcpu->arch.cr4 & X86_CR4_UMIP));
4760 	return kvm_emulate_instruction(vcpu, 0);
4761 }
4762 
4763 static int handle_cr(struct kvm_vcpu *vcpu)
4764 {
4765 	unsigned long exit_qualification, val;
4766 	int cr;
4767 	int reg;
4768 	int err;
4769 	int ret;
4770 
4771 	exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
4772 	cr = exit_qualification & 15;
4773 	reg = (exit_qualification >> 8) & 15;
4774 	switch ((exit_qualification >> 4) & 3) {
4775 	case 0: /* mov to cr */
4776 		val = kvm_register_readl(vcpu, reg);
4777 		trace_kvm_cr_write(cr, val);
4778 		switch (cr) {
4779 		case 0:
4780 			err = handle_set_cr0(vcpu, val);
4781 			return kvm_complete_insn_gp(vcpu, err);
4782 		case 3:
4783 			WARN_ON_ONCE(enable_unrestricted_guest);
4784 			err = kvm_set_cr3(vcpu, val);
4785 			return kvm_complete_insn_gp(vcpu, err);
4786 		case 4:
4787 			err = handle_set_cr4(vcpu, val);
4788 			return kvm_complete_insn_gp(vcpu, err);
4789 		case 8: {
4790 				u8 cr8_prev = kvm_get_cr8(vcpu);
4791 				u8 cr8 = (u8)val;
4792 				err = kvm_set_cr8(vcpu, cr8);
4793 				ret = kvm_complete_insn_gp(vcpu, err);
4794 				if (lapic_in_kernel(vcpu))
4795 					return ret;
4796 				if (cr8_prev <= cr8)
4797 					return ret;
4798 				/*
4799 				 * TODO: we might be squashing a
4800 				 * KVM_GUESTDBG_SINGLESTEP-triggered
4801 				 * KVM_EXIT_DEBUG here.
4802 				 */
4803 				vcpu->run->exit_reason = KVM_EXIT_SET_TPR;
4804 				return 0;
4805 			}
4806 		}
4807 		break;
4808 	case 2: /* clts */
4809 		WARN_ONCE(1, "Guest should always own CR0.TS");
4810 		vmx_set_cr0(vcpu, kvm_read_cr0_bits(vcpu, ~X86_CR0_TS));
4811 		trace_kvm_cr_write(0, kvm_read_cr0(vcpu));
4812 		return kvm_skip_emulated_instruction(vcpu);
4813 	case 1: /*mov from cr*/
4814 		switch (cr) {
4815 		case 3:
4816 			WARN_ON_ONCE(enable_unrestricted_guest);
4817 			val = kvm_read_cr3(vcpu);
4818 			kvm_register_write(vcpu, reg, val);
4819 			trace_kvm_cr_read(cr, val);
4820 			return kvm_skip_emulated_instruction(vcpu);
4821 		case 8:
4822 			val = kvm_get_cr8(vcpu);
4823 			kvm_register_write(vcpu, reg, val);
4824 			trace_kvm_cr_read(cr, val);
4825 			return kvm_skip_emulated_instruction(vcpu);
4826 		}
4827 		break;
4828 	case 3: /* lmsw */
4829 		val = (exit_qualification >> LMSW_SOURCE_DATA_SHIFT) & 0x0f;
4830 		trace_kvm_cr_write(0, (kvm_read_cr0(vcpu) & ~0xful) | val);
4831 		kvm_lmsw(vcpu, val);
4832 
4833 		return kvm_skip_emulated_instruction(vcpu);
4834 	default:
4835 		break;
4836 	}
4837 	vcpu->run->exit_reason = 0;
4838 	vcpu_unimpl(vcpu, "unhandled control register: op %d cr %d\n",
4839 	       (int)(exit_qualification >> 4) & 3, cr);
4840 	return 0;
4841 }
4842 
4843 static int handle_dr(struct kvm_vcpu *vcpu)
4844 {
4845 	unsigned long exit_qualification;
4846 	int dr, dr7, reg;
4847 
4848 	exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
4849 	dr = exit_qualification & DEBUG_REG_ACCESS_NUM;
4850 
4851 	/* First, if DR does not exist, trigger UD */
4852 	if (!kvm_require_dr(vcpu, dr))
4853 		return 1;
4854 
4855 	/* Do not handle if the CPL > 0, will trigger GP on re-entry */
4856 	if (!kvm_require_cpl(vcpu, 0))
4857 		return 1;
4858 	dr7 = vmcs_readl(GUEST_DR7);
4859 	if (dr7 & DR7_GD) {
4860 		/*
4861 		 * As the vm-exit takes precedence over the debug trap, we
4862 		 * need to emulate the latter, either for the host or the
4863 		 * guest debugging itself.
4864 		 */
4865 		if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP) {
4866 			vcpu->run->debug.arch.dr6 = vcpu->arch.dr6;
4867 			vcpu->run->debug.arch.dr7 = dr7;
4868 			vcpu->run->debug.arch.pc = kvm_get_linear_rip(vcpu);
4869 			vcpu->run->debug.arch.exception = DB_VECTOR;
4870 			vcpu->run->exit_reason = KVM_EXIT_DEBUG;
4871 			return 0;
4872 		} else {
4873 			vcpu->arch.dr6 &= ~DR_TRAP_BITS;
4874 			vcpu->arch.dr6 |= DR6_BD | DR6_RTM;
4875 			kvm_queue_exception(vcpu, DB_VECTOR);
4876 			return 1;
4877 		}
4878 	}
4879 
4880 	if (vcpu->guest_debug == 0) {
4881 		exec_controls_clearbit(to_vmx(vcpu), CPU_BASED_MOV_DR_EXITING);
4882 
4883 		/*
4884 		 * No more DR vmexits; force a reload of the debug registers
4885 		 * and reenter on this instruction.  The next vmexit will
4886 		 * retrieve the full state of the debug registers.
4887 		 */
4888 		vcpu->arch.switch_db_regs |= KVM_DEBUGREG_WONT_EXIT;
4889 		return 1;
4890 	}
4891 
4892 	reg = DEBUG_REG_ACCESS_REG(exit_qualification);
4893 	if (exit_qualification & TYPE_MOV_FROM_DR) {
4894 		unsigned long val;
4895 
4896 		if (kvm_get_dr(vcpu, dr, &val))
4897 			return 1;
4898 		kvm_register_write(vcpu, reg, val);
4899 	} else
4900 		if (kvm_set_dr(vcpu, dr, kvm_register_readl(vcpu, reg)))
4901 			return 1;
4902 
4903 	return kvm_skip_emulated_instruction(vcpu);
4904 }
4905 
4906 static u64 vmx_get_dr6(struct kvm_vcpu *vcpu)
4907 {
4908 	return vcpu->arch.dr6;
4909 }
4910 
4911 static void vmx_set_dr6(struct kvm_vcpu *vcpu, unsigned long val)
4912 {
4913 }
4914 
4915 static void vmx_sync_dirty_debug_regs(struct kvm_vcpu *vcpu)
4916 {
4917 	get_debugreg(vcpu->arch.db[0], 0);
4918 	get_debugreg(vcpu->arch.db[1], 1);
4919 	get_debugreg(vcpu->arch.db[2], 2);
4920 	get_debugreg(vcpu->arch.db[3], 3);
4921 	get_debugreg(vcpu->arch.dr6, 6);
4922 	vcpu->arch.dr7 = vmcs_readl(GUEST_DR7);
4923 
4924 	vcpu->arch.switch_db_regs &= ~KVM_DEBUGREG_WONT_EXIT;
4925 	exec_controls_setbit(to_vmx(vcpu), CPU_BASED_MOV_DR_EXITING);
4926 }
4927 
4928 static void vmx_set_dr7(struct kvm_vcpu *vcpu, unsigned long val)
4929 {
4930 	vmcs_writel(GUEST_DR7, val);
4931 }
4932 
4933 static int handle_tpr_below_threshold(struct kvm_vcpu *vcpu)
4934 {
4935 	kvm_apic_update_ppr(vcpu);
4936 	return 1;
4937 }
4938 
4939 static int handle_interrupt_window(struct kvm_vcpu *vcpu)
4940 {
4941 	exec_controls_clearbit(to_vmx(vcpu), CPU_BASED_VIRTUAL_INTR_PENDING);
4942 
4943 	kvm_make_request(KVM_REQ_EVENT, vcpu);
4944 
4945 	++vcpu->stat.irq_window_exits;
4946 	return 1;
4947 }
4948 
4949 static int handle_vmcall(struct kvm_vcpu *vcpu)
4950 {
4951 	return kvm_emulate_hypercall(vcpu);
4952 }
4953 
4954 static int handle_invd(struct kvm_vcpu *vcpu)
4955 {
4956 	return kvm_emulate_instruction(vcpu, 0);
4957 }
4958 
4959 static int handle_invlpg(struct kvm_vcpu *vcpu)
4960 {
4961 	unsigned long exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
4962 
4963 	kvm_mmu_invlpg(vcpu, exit_qualification);
4964 	return kvm_skip_emulated_instruction(vcpu);
4965 }
4966 
4967 static int handle_rdpmc(struct kvm_vcpu *vcpu)
4968 {
4969 	int err;
4970 
4971 	err = kvm_rdpmc(vcpu);
4972 	return kvm_complete_insn_gp(vcpu, err);
4973 }
4974 
4975 static int handle_wbinvd(struct kvm_vcpu *vcpu)
4976 {
4977 	return kvm_emulate_wbinvd(vcpu);
4978 }
4979 
4980 static int handle_xsetbv(struct kvm_vcpu *vcpu)
4981 {
4982 	u64 new_bv = kvm_read_edx_eax(vcpu);
4983 	u32 index = kvm_rcx_read(vcpu);
4984 
4985 	if (kvm_set_xcr(vcpu, index, new_bv) == 0)
4986 		return kvm_skip_emulated_instruction(vcpu);
4987 	return 1;
4988 }
4989 
4990 static int handle_apic_access(struct kvm_vcpu *vcpu)
4991 {
4992 	if (likely(fasteoi)) {
4993 		unsigned long exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
4994 		int access_type, offset;
4995 
4996 		access_type = exit_qualification & APIC_ACCESS_TYPE;
4997 		offset = exit_qualification & APIC_ACCESS_OFFSET;
4998 		/*
4999 		 * Sane guest uses MOV to write EOI, with written value
5000 		 * not cared. So make a short-circuit here by avoiding
5001 		 * heavy instruction emulation.
5002 		 */
5003 		if ((access_type == TYPE_LINEAR_APIC_INST_WRITE) &&
5004 		    (offset == APIC_EOI)) {
5005 			kvm_lapic_set_eoi(vcpu);
5006 			return kvm_skip_emulated_instruction(vcpu);
5007 		}
5008 	}
5009 	return kvm_emulate_instruction(vcpu, 0);
5010 }
5011 
5012 static int handle_apic_eoi_induced(struct kvm_vcpu *vcpu)
5013 {
5014 	unsigned long exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
5015 	int vector = exit_qualification & 0xff;
5016 
5017 	/* EOI-induced VM exit is trap-like and thus no need to adjust IP */
5018 	kvm_apic_set_eoi_accelerated(vcpu, vector);
5019 	return 1;
5020 }
5021 
5022 static int handle_apic_write(struct kvm_vcpu *vcpu)
5023 {
5024 	unsigned long exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
5025 	u32 offset = exit_qualification & 0xfff;
5026 
5027 	/* APIC-write VM exit is trap-like and thus no need to adjust IP */
5028 	kvm_apic_write_nodecode(vcpu, offset);
5029 	return 1;
5030 }
5031 
5032 static int handle_task_switch(struct kvm_vcpu *vcpu)
5033 {
5034 	struct vcpu_vmx *vmx = to_vmx(vcpu);
5035 	unsigned long exit_qualification;
5036 	bool has_error_code = false;
5037 	u32 error_code = 0;
5038 	u16 tss_selector;
5039 	int reason, type, idt_v, idt_index;
5040 
5041 	idt_v = (vmx->idt_vectoring_info & VECTORING_INFO_VALID_MASK);
5042 	idt_index = (vmx->idt_vectoring_info & VECTORING_INFO_VECTOR_MASK);
5043 	type = (vmx->idt_vectoring_info & VECTORING_INFO_TYPE_MASK);
5044 
5045 	exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
5046 
5047 	reason = (u32)exit_qualification >> 30;
5048 	if (reason == TASK_SWITCH_GATE && idt_v) {
5049 		switch (type) {
5050 		case INTR_TYPE_NMI_INTR:
5051 			vcpu->arch.nmi_injected = false;
5052 			vmx_set_nmi_mask(vcpu, true);
5053 			break;
5054 		case INTR_TYPE_EXT_INTR:
5055 		case INTR_TYPE_SOFT_INTR:
5056 			kvm_clear_interrupt_queue(vcpu);
5057 			break;
5058 		case INTR_TYPE_HARD_EXCEPTION:
5059 			if (vmx->idt_vectoring_info &
5060 			    VECTORING_INFO_DELIVER_CODE_MASK) {
5061 				has_error_code = true;
5062 				error_code =
5063 					vmcs_read32(IDT_VECTORING_ERROR_CODE);
5064 			}
5065 			/* fall through */
5066 		case INTR_TYPE_SOFT_EXCEPTION:
5067 			kvm_clear_exception_queue(vcpu);
5068 			break;
5069 		default:
5070 			break;
5071 		}
5072 	}
5073 	tss_selector = exit_qualification;
5074 
5075 	if (!idt_v || (type != INTR_TYPE_HARD_EXCEPTION &&
5076 		       type != INTR_TYPE_EXT_INTR &&
5077 		       type != INTR_TYPE_NMI_INTR))
5078 		WARN_ON(!skip_emulated_instruction(vcpu));
5079 
5080 	/*
5081 	 * TODO: What about debug traps on tss switch?
5082 	 *       Are we supposed to inject them and update dr6?
5083 	 */
5084 	return kvm_task_switch(vcpu, tss_selector,
5085 			       type == INTR_TYPE_SOFT_INTR ? idt_index : -1,
5086 			       reason, has_error_code, error_code);
5087 }
5088 
5089 static int handle_ept_violation(struct kvm_vcpu *vcpu)
5090 {
5091 	unsigned long exit_qualification;
5092 	gpa_t gpa;
5093 	u64 error_code;
5094 
5095 	exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
5096 
5097 	/*
5098 	 * EPT violation happened while executing iret from NMI,
5099 	 * "blocked by NMI" bit has to be set before next VM entry.
5100 	 * There are errata that may cause this bit to not be set:
5101 	 * AAK134, BY25.
5102 	 */
5103 	if (!(to_vmx(vcpu)->idt_vectoring_info & VECTORING_INFO_VALID_MASK) &&
5104 			enable_vnmi &&
5105 			(exit_qualification & INTR_INFO_UNBLOCK_NMI))
5106 		vmcs_set_bits(GUEST_INTERRUPTIBILITY_INFO, GUEST_INTR_STATE_NMI);
5107 
5108 	gpa = vmcs_read64(GUEST_PHYSICAL_ADDRESS);
5109 	trace_kvm_page_fault(gpa, exit_qualification);
5110 
5111 	/* Is it a read fault? */
5112 	error_code = (exit_qualification & EPT_VIOLATION_ACC_READ)
5113 		     ? PFERR_USER_MASK : 0;
5114 	/* Is it a write fault? */
5115 	error_code |= (exit_qualification & EPT_VIOLATION_ACC_WRITE)
5116 		      ? PFERR_WRITE_MASK : 0;
5117 	/* Is it a fetch fault? */
5118 	error_code |= (exit_qualification & EPT_VIOLATION_ACC_INSTR)
5119 		      ? PFERR_FETCH_MASK : 0;
5120 	/* ept page table entry is present? */
5121 	error_code |= (exit_qualification &
5122 		       (EPT_VIOLATION_READABLE | EPT_VIOLATION_WRITABLE |
5123 			EPT_VIOLATION_EXECUTABLE))
5124 		      ? PFERR_PRESENT_MASK : 0;
5125 
5126 	error_code |= (exit_qualification & 0x100) != 0 ?
5127 	       PFERR_GUEST_FINAL_MASK : PFERR_GUEST_PAGE_MASK;
5128 
5129 	vcpu->arch.exit_qualification = exit_qualification;
5130 	return kvm_mmu_page_fault(vcpu, gpa, error_code, NULL, 0);
5131 }
5132 
5133 static int handle_ept_misconfig(struct kvm_vcpu *vcpu)
5134 {
5135 	gpa_t gpa;
5136 
5137 	/*
5138 	 * A nested guest cannot optimize MMIO vmexits, because we have an
5139 	 * nGPA here instead of the required GPA.
5140 	 */
5141 	gpa = vmcs_read64(GUEST_PHYSICAL_ADDRESS);
5142 	if (!is_guest_mode(vcpu) &&
5143 	    !kvm_io_bus_write(vcpu, KVM_FAST_MMIO_BUS, gpa, 0, NULL)) {
5144 		trace_kvm_fast_mmio(gpa);
5145 		return kvm_skip_emulated_instruction(vcpu);
5146 	}
5147 
5148 	return kvm_mmu_page_fault(vcpu, gpa, PFERR_RSVD_MASK, NULL, 0);
5149 }
5150 
5151 static int handle_nmi_window(struct kvm_vcpu *vcpu)
5152 {
5153 	WARN_ON_ONCE(!enable_vnmi);
5154 	exec_controls_clearbit(to_vmx(vcpu), CPU_BASED_VIRTUAL_NMI_PENDING);
5155 	++vcpu->stat.nmi_window_exits;
5156 	kvm_make_request(KVM_REQ_EVENT, vcpu);
5157 
5158 	return 1;
5159 }
5160 
5161 static int handle_invalid_guest_state(struct kvm_vcpu *vcpu)
5162 {
5163 	struct vcpu_vmx *vmx = to_vmx(vcpu);
5164 	bool intr_window_requested;
5165 	unsigned count = 130;
5166 
5167 	/*
5168 	 * We should never reach the point where we are emulating L2
5169 	 * due to invalid guest state as that means we incorrectly
5170 	 * allowed a nested VMEntry with an invalid vmcs12.
5171 	 */
5172 	WARN_ON_ONCE(vmx->emulation_required && vmx->nested.nested_run_pending);
5173 
5174 	intr_window_requested = exec_controls_get(vmx) &
5175 				CPU_BASED_VIRTUAL_INTR_PENDING;
5176 
5177 	while (vmx->emulation_required && count-- != 0) {
5178 		if (intr_window_requested && vmx_interrupt_allowed(vcpu))
5179 			return handle_interrupt_window(&vmx->vcpu);
5180 
5181 		if (kvm_test_request(KVM_REQ_EVENT, vcpu))
5182 			return 1;
5183 
5184 		if (!kvm_emulate_instruction(vcpu, 0))
5185 			return 0;
5186 
5187 		if (vmx->emulation_required && !vmx->rmode.vm86_active &&
5188 		    vcpu->arch.exception.pending) {
5189 			vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
5190 			vcpu->run->internal.suberror =
5191 						KVM_INTERNAL_ERROR_EMULATION;
5192 			vcpu->run->internal.ndata = 0;
5193 			return 0;
5194 		}
5195 
5196 		if (vcpu->arch.halt_request) {
5197 			vcpu->arch.halt_request = 0;
5198 			return kvm_vcpu_halt(vcpu);
5199 		}
5200 
5201 		/*
5202 		 * Note, return 1 and not 0, vcpu_run() is responsible for
5203 		 * morphing the pending signal into the proper return code.
5204 		 */
5205 		if (signal_pending(current))
5206 			return 1;
5207 
5208 		if (need_resched())
5209 			schedule();
5210 	}
5211 
5212 	return 1;
5213 }
5214 
5215 static void grow_ple_window(struct kvm_vcpu *vcpu)
5216 {
5217 	struct vcpu_vmx *vmx = to_vmx(vcpu);
5218 	unsigned int old = vmx->ple_window;
5219 
5220 	vmx->ple_window = __grow_ple_window(old, ple_window,
5221 					    ple_window_grow,
5222 					    ple_window_max);
5223 
5224 	if (vmx->ple_window != old) {
5225 		vmx->ple_window_dirty = true;
5226 		trace_kvm_ple_window_update(vcpu->vcpu_id,
5227 					    vmx->ple_window, old);
5228 	}
5229 }
5230 
5231 static void shrink_ple_window(struct kvm_vcpu *vcpu)
5232 {
5233 	struct vcpu_vmx *vmx = to_vmx(vcpu);
5234 	unsigned int old = vmx->ple_window;
5235 
5236 	vmx->ple_window = __shrink_ple_window(old, ple_window,
5237 					      ple_window_shrink,
5238 					      ple_window);
5239 
5240 	if (vmx->ple_window != old) {
5241 		vmx->ple_window_dirty = true;
5242 		trace_kvm_ple_window_update(vcpu->vcpu_id,
5243 					    vmx->ple_window, old);
5244 	}
5245 }
5246 
5247 /*
5248  * Handler for POSTED_INTERRUPT_WAKEUP_VECTOR.
5249  */
5250 static void wakeup_handler(void)
5251 {
5252 	struct kvm_vcpu *vcpu;
5253 	int cpu = smp_processor_id();
5254 
5255 	spin_lock(&per_cpu(blocked_vcpu_on_cpu_lock, cpu));
5256 	list_for_each_entry(vcpu, &per_cpu(blocked_vcpu_on_cpu, cpu),
5257 			blocked_vcpu_list) {
5258 		struct pi_desc *pi_desc = vcpu_to_pi_desc(vcpu);
5259 
5260 		if (pi_test_on(pi_desc) == 1)
5261 			kvm_vcpu_kick(vcpu);
5262 	}
5263 	spin_unlock(&per_cpu(blocked_vcpu_on_cpu_lock, cpu));
5264 }
5265 
5266 static void vmx_enable_tdp(void)
5267 {
5268 	kvm_mmu_set_mask_ptes(VMX_EPT_READABLE_MASK,
5269 		enable_ept_ad_bits ? VMX_EPT_ACCESS_BIT : 0ull,
5270 		enable_ept_ad_bits ? VMX_EPT_DIRTY_BIT : 0ull,
5271 		0ull, VMX_EPT_EXECUTABLE_MASK,
5272 		cpu_has_vmx_ept_execute_only() ? 0ull : VMX_EPT_READABLE_MASK,
5273 		VMX_EPT_RWX_MASK, 0ull);
5274 
5275 	ept_set_mmio_spte_mask();
5276 	kvm_enable_tdp();
5277 }
5278 
5279 /*
5280  * Indicate a busy-waiting vcpu in spinlock. We do not enable the PAUSE
5281  * exiting, so only get here on cpu with PAUSE-Loop-Exiting.
5282  */
5283 static int handle_pause(struct kvm_vcpu *vcpu)
5284 {
5285 	if (!kvm_pause_in_guest(vcpu->kvm))
5286 		grow_ple_window(vcpu);
5287 
5288 	/*
5289 	 * Intel sdm vol3 ch-25.1.3 says: The "PAUSE-loop exiting"
5290 	 * VM-execution control is ignored if CPL > 0. OTOH, KVM
5291 	 * never set PAUSE_EXITING and just set PLE if supported,
5292 	 * so the vcpu must be CPL=0 if it gets a PAUSE exit.
5293 	 */
5294 	kvm_vcpu_on_spin(vcpu, true);
5295 	return kvm_skip_emulated_instruction(vcpu);
5296 }
5297 
5298 static int handle_nop(struct kvm_vcpu *vcpu)
5299 {
5300 	return kvm_skip_emulated_instruction(vcpu);
5301 }
5302 
5303 static int handle_mwait(struct kvm_vcpu *vcpu)
5304 {
5305 	printk_once(KERN_WARNING "kvm: MWAIT instruction emulated as NOP!\n");
5306 	return handle_nop(vcpu);
5307 }
5308 
5309 static int handle_invalid_op(struct kvm_vcpu *vcpu)
5310 {
5311 	kvm_queue_exception(vcpu, UD_VECTOR);
5312 	return 1;
5313 }
5314 
5315 static int handle_monitor_trap(struct kvm_vcpu *vcpu)
5316 {
5317 	return 1;
5318 }
5319 
5320 static int handle_monitor(struct kvm_vcpu *vcpu)
5321 {
5322 	printk_once(KERN_WARNING "kvm: MONITOR instruction emulated as NOP!\n");
5323 	return handle_nop(vcpu);
5324 }
5325 
5326 static int handle_invpcid(struct kvm_vcpu *vcpu)
5327 {
5328 	u32 vmx_instruction_info;
5329 	unsigned long type;
5330 	bool pcid_enabled;
5331 	gva_t gva;
5332 	struct x86_exception e;
5333 	unsigned i;
5334 	unsigned long roots_to_free = 0;
5335 	struct {
5336 		u64 pcid;
5337 		u64 gla;
5338 	} operand;
5339 
5340 	if (!guest_cpuid_has(vcpu, X86_FEATURE_INVPCID)) {
5341 		kvm_queue_exception(vcpu, UD_VECTOR);
5342 		return 1;
5343 	}
5344 
5345 	vmx_instruction_info = vmcs_read32(VMX_INSTRUCTION_INFO);
5346 	type = kvm_register_readl(vcpu, (vmx_instruction_info >> 28) & 0xf);
5347 
5348 	if (type > 3) {
5349 		kvm_inject_gp(vcpu, 0);
5350 		return 1;
5351 	}
5352 
5353 	/* According to the Intel instruction reference, the memory operand
5354 	 * is read even if it isn't needed (e.g., for type==all)
5355 	 */
5356 	if (get_vmx_mem_address(vcpu, vmcs_readl(EXIT_QUALIFICATION),
5357 				vmx_instruction_info, false,
5358 				sizeof(operand), &gva))
5359 		return 1;
5360 
5361 	if (kvm_read_guest_virt(vcpu, gva, &operand, sizeof(operand), &e)) {
5362 		kvm_inject_page_fault(vcpu, &e);
5363 		return 1;
5364 	}
5365 
5366 	if (operand.pcid >> 12 != 0) {
5367 		kvm_inject_gp(vcpu, 0);
5368 		return 1;
5369 	}
5370 
5371 	pcid_enabled = kvm_read_cr4_bits(vcpu, X86_CR4_PCIDE);
5372 
5373 	switch (type) {
5374 	case INVPCID_TYPE_INDIV_ADDR:
5375 		if ((!pcid_enabled && (operand.pcid != 0)) ||
5376 		    is_noncanonical_address(operand.gla, vcpu)) {
5377 			kvm_inject_gp(vcpu, 0);
5378 			return 1;
5379 		}
5380 		kvm_mmu_invpcid_gva(vcpu, operand.gla, operand.pcid);
5381 		return kvm_skip_emulated_instruction(vcpu);
5382 
5383 	case INVPCID_TYPE_SINGLE_CTXT:
5384 		if (!pcid_enabled && (operand.pcid != 0)) {
5385 			kvm_inject_gp(vcpu, 0);
5386 			return 1;
5387 		}
5388 
5389 		if (kvm_get_active_pcid(vcpu) == operand.pcid) {
5390 			kvm_mmu_sync_roots(vcpu);
5391 			kvm_make_request(KVM_REQ_TLB_FLUSH, vcpu);
5392 		}
5393 
5394 		for (i = 0; i < KVM_MMU_NUM_PREV_ROOTS; i++)
5395 			if (kvm_get_pcid(vcpu, vcpu->arch.mmu->prev_roots[i].cr3)
5396 			    == operand.pcid)
5397 				roots_to_free |= KVM_MMU_ROOT_PREVIOUS(i);
5398 
5399 		kvm_mmu_free_roots(vcpu, vcpu->arch.mmu, roots_to_free);
5400 		/*
5401 		 * If neither the current cr3 nor any of the prev_roots use the
5402 		 * given PCID, then nothing needs to be done here because a
5403 		 * resync will happen anyway before switching to any other CR3.
5404 		 */
5405 
5406 		return kvm_skip_emulated_instruction(vcpu);
5407 
5408 	case INVPCID_TYPE_ALL_NON_GLOBAL:
5409 		/*
5410 		 * Currently, KVM doesn't mark global entries in the shadow
5411 		 * page tables, so a non-global flush just degenerates to a
5412 		 * global flush. If needed, we could optimize this later by
5413 		 * keeping track of global entries in shadow page tables.
5414 		 */
5415 
5416 		/* fall-through */
5417 	case INVPCID_TYPE_ALL_INCL_GLOBAL:
5418 		kvm_mmu_unload(vcpu);
5419 		return kvm_skip_emulated_instruction(vcpu);
5420 
5421 	default:
5422 		BUG(); /* We have already checked above that type <= 3 */
5423 	}
5424 }
5425 
5426 static int handle_pml_full(struct kvm_vcpu *vcpu)
5427 {
5428 	unsigned long exit_qualification;
5429 
5430 	trace_kvm_pml_full(vcpu->vcpu_id);
5431 
5432 	exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
5433 
5434 	/*
5435 	 * PML buffer FULL happened while executing iret from NMI,
5436 	 * "blocked by NMI" bit has to be set before next VM entry.
5437 	 */
5438 	if (!(to_vmx(vcpu)->idt_vectoring_info & VECTORING_INFO_VALID_MASK) &&
5439 			enable_vnmi &&
5440 			(exit_qualification & INTR_INFO_UNBLOCK_NMI))
5441 		vmcs_set_bits(GUEST_INTERRUPTIBILITY_INFO,
5442 				GUEST_INTR_STATE_NMI);
5443 
5444 	/*
5445 	 * PML buffer already flushed at beginning of VMEXIT. Nothing to do
5446 	 * here.., and there's no userspace involvement needed for PML.
5447 	 */
5448 	return 1;
5449 }
5450 
5451 static int handle_preemption_timer(struct kvm_vcpu *vcpu)
5452 {
5453 	struct vcpu_vmx *vmx = to_vmx(vcpu);
5454 
5455 	if (!vmx->req_immediate_exit &&
5456 	    !unlikely(vmx->loaded_vmcs->hv_timer_soft_disabled))
5457 		kvm_lapic_expired_hv_timer(vcpu);
5458 
5459 	return 1;
5460 }
5461 
5462 /*
5463  * When nested=0, all VMX instruction VM Exits filter here.  The handlers
5464  * are overwritten by nested_vmx_setup() when nested=1.
5465  */
5466 static int handle_vmx_instruction(struct kvm_vcpu *vcpu)
5467 {
5468 	kvm_queue_exception(vcpu, UD_VECTOR);
5469 	return 1;
5470 }
5471 
5472 static int handle_encls(struct kvm_vcpu *vcpu)
5473 {
5474 	/*
5475 	 * SGX virtualization is not yet supported.  There is no software
5476 	 * enable bit for SGX, so we have to trap ENCLS and inject a #UD
5477 	 * to prevent the guest from executing ENCLS.
5478 	 */
5479 	kvm_queue_exception(vcpu, UD_VECTOR);
5480 	return 1;
5481 }
5482 
5483 /*
5484  * The exit handlers return 1 if the exit was handled fully and guest execution
5485  * may resume.  Otherwise they set the kvm_run parameter to indicate what needs
5486  * to be done to userspace and return 0.
5487  */
5488 static int (*kvm_vmx_exit_handlers[])(struct kvm_vcpu *vcpu) = {
5489 	[EXIT_REASON_EXCEPTION_NMI]           = handle_exception_nmi,
5490 	[EXIT_REASON_EXTERNAL_INTERRUPT]      = handle_external_interrupt,
5491 	[EXIT_REASON_TRIPLE_FAULT]            = handle_triple_fault,
5492 	[EXIT_REASON_NMI_WINDOW]	      = handle_nmi_window,
5493 	[EXIT_REASON_IO_INSTRUCTION]          = handle_io,
5494 	[EXIT_REASON_CR_ACCESS]               = handle_cr,
5495 	[EXIT_REASON_DR_ACCESS]               = handle_dr,
5496 	[EXIT_REASON_CPUID]                   = kvm_emulate_cpuid,
5497 	[EXIT_REASON_MSR_READ]                = kvm_emulate_rdmsr,
5498 	[EXIT_REASON_MSR_WRITE]               = kvm_emulate_wrmsr,
5499 	[EXIT_REASON_PENDING_INTERRUPT]       = handle_interrupt_window,
5500 	[EXIT_REASON_HLT]                     = kvm_emulate_halt,
5501 	[EXIT_REASON_INVD]		      = handle_invd,
5502 	[EXIT_REASON_INVLPG]		      = handle_invlpg,
5503 	[EXIT_REASON_RDPMC]                   = handle_rdpmc,
5504 	[EXIT_REASON_VMCALL]                  = handle_vmcall,
5505 	[EXIT_REASON_VMCLEAR]		      = handle_vmx_instruction,
5506 	[EXIT_REASON_VMLAUNCH]		      = handle_vmx_instruction,
5507 	[EXIT_REASON_VMPTRLD]		      = handle_vmx_instruction,
5508 	[EXIT_REASON_VMPTRST]		      = handle_vmx_instruction,
5509 	[EXIT_REASON_VMREAD]		      = handle_vmx_instruction,
5510 	[EXIT_REASON_VMRESUME]		      = handle_vmx_instruction,
5511 	[EXIT_REASON_VMWRITE]		      = handle_vmx_instruction,
5512 	[EXIT_REASON_VMOFF]		      = handle_vmx_instruction,
5513 	[EXIT_REASON_VMON]		      = handle_vmx_instruction,
5514 	[EXIT_REASON_TPR_BELOW_THRESHOLD]     = handle_tpr_below_threshold,
5515 	[EXIT_REASON_APIC_ACCESS]             = handle_apic_access,
5516 	[EXIT_REASON_APIC_WRITE]              = handle_apic_write,
5517 	[EXIT_REASON_EOI_INDUCED]             = handle_apic_eoi_induced,
5518 	[EXIT_REASON_WBINVD]                  = handle_wbinvd,
5519 	[EXIT_REASON_XSETBV]                  = handle_xsetbv,
5520 	[EXIT_REASON_TASK_SWITCH]             = handle_task_switch,
5521 	[EXIT_REASON_MCE_DURING_VMENTRY]      = handle_machine_check,
5522 	[EXIT_REASON_GDTR_IDTR]		      = handle_desc,
5523 	[EXIT_REASON_LDTR_TR]		      = handle_desc,
5524 	[EXIT_REASON_EPT_VIOLATION]	      = handle_ept_violation,
5525 	[EXIT_REASON_EPT_MISCONFIG]           = handle_ept_misconfig,
5526 	[EXIT_REASON_PAUSE_INSTRUCTION]       = handle_pause,
5527 	[EXIT_REASON_MWAIT_INSTRUCTION]	      = handle_mwait,
5528 	[EXIT_REASON_MONITOR_TRAP_FLAG]       = handle_monitor_trap,
5529 	[EXIT_REASON_MONITOR_INSTRUCTION]     = handle_monitor,
5530 	[EXIT_REASON_INVEPT]                  = handle_vmx_instruction,
5531 	[EXIT_REASON_INVVPID]                 = handle_vmx_instruction,
5532 	[EXIT_REASON_RDRAND]                  = handle_invalid_op,
5533 	[EXIT_REASON_RDSEED]                  = handle_invalid_op,
5534 	[EXIT_REASON_PML_FULL]		      = handle_pml_full,
5535 	[EXIT_REASON_INVPCID]                 = handle_invpcid,
5536 	[EXIT_REASON_VMFUNC]		      = handle_vmx_instruction,
5537 	[EXIT_REASON_PREEMPTION_TIMER]	      = handle_preemption_timer,
5538 	[EXIT_REASON_ENCLS]		      = handle_encls,
5539 };
5540 
5541 static const int kvm_vmx_max_exit_handlers =
5542 	ARRAY_SIZE(kvm_vmx_exit_handlers);
5543 
5544 static void vmx_get_exit_info(struct kvm_vcpu *vcpu, u64 *info1, u64 *info2)
5545 {
5546 	*info1 = vmcs_readl(EXIT_QUALIFICATION);
5547 	*info2 = vmcs_read32(VM_EXIT_INTR_INFO);
5548 }
5549 
5550 static void vmx_destroy_pml_buffer(struct vcpu_vmx *vmx)
5551 {
5552 	if (vmx->pml_pg) {
5553 		__free_page(vmx->pml_pg);
5554 		vmx->pml_pg = NULL;
5555 	}
5556 }
5557 
5558 static void vmx_flush_pml_buffer(struct kvm_vcpu *vcpu)
5559 {
5560 	struct vcpu_vmx *vmx = to_vmx(vcpu);
5561 	u64 *pml_buf;
5562 	u16 pml_idx;
5563 
5564 	pml_idx = vmcs_read16(GUEST_PML_INDEX);
5565 
5566 	/* Do nothing if PML buffer is empty */
5567 	if (pml_idx == (PML_ENTITY_NUM - 1))
5568 		return;
5569 
5570 	/* PML index always points to next available PML buffer entity */
5571 	if (pml_idx >= PML_ENTITY_NUM)
5572 		pml_idx = 0;
5573 	else
5574 		pml_idx++;
5575 
5576 	pml_buf = page_address(vmx->pml_pg);
5577 	for (; pml_idx < PML_ENTITY_NUM; pml_idx++) {
5578 		u64 gpa;
5579 
5580 		gpa = pml_buf[pml_idx];
5581 		WARN_ON(gpa & (PAGE_SIZE - 1));
5582 		kvm_vcpu_mark_page_dirty(vcpu, gpa >> PAGE_SHIFT);
5583 	}
5584 
5585 	/* reset PML index */
5586 	vmcs_write16(GUEST_PML_INDEX, PML_ENTITY_NUM - 1);
5587 }
5588 
5589 /*
5590  * Flush all vcpus' PML buffer and update logged GPAs to dirty_bitmap.
5591  * Called before reporting dirty_bitmap to userspace.
5592  */
5593 static void kvm_flush_pml_buffers(struct kvm *kvm)
5594 {
5595 	int i;
5596 	struct kvm_vcpu *vcpu;
5597 	/*
5598 	 * We only need to kick vcpu out of guest mode here, as PML buffer
5599 	 * is flushed at beginning of all VMEXITs, and it's obvious that only
5600 	 * vcpus running in guest are possible to have unflushed GPAs in PML
5601 	 * buffer.
5602 	 */
5603 	kvm_for_each_vcpu(i, vcpu, kvm)
5604 		kvm_vcpu_kick(vcpu);
5605 }
5606 
5607 static void vmx_dump_sel(char *name, uint32_t sel)
5608 {
5609 	pr_err("%s sel=0x%04x, attr=0x%05x, limit=0x%08x, base=0x%016lx\n",
5610 	       name, vmcs_read16(sel),
5611 	       vmcs_read32(sel + GUEST_ES_AR_BYTES - GUEST_ES_SELECTOR),
5612 	       vmcs_read32(sel + GUEST_ES_LIMIT - GUEST_ES_SELECTOR),
5613 	       vmcs_readl(sel + GUEST_ES_BASE - GUEST_ES_SELECTOR));
5614 }
5615 
5616 static void vmx_dump_dtsel(char *name, uint32_t limit)
5617 {
5618 	pr_err("%s                           limit=0x%08x, base=0x%016lx\n",
5619 	       name, vmcs_read32(limit),
5620 	       vmcs_readl(limit + GUEST_GDTR_BASE - GUEST_GDTR_LIMIT));
5621 }
5622 
5623 void dump_vmcs(void)
5624 {
5625 	u32 vmentry_ctl, vmexit_ctl;
5626 	u32 cpu_based_exec_ctrl, pin_based_exec_ctrl, secondary_exec_control;
5627 	unsigned long cr4;
5628 	u64 efer;
5629 	int i, n;
5630 
5631 	if (!dump_invalid_vmcs) {
5632 		pr_warn_ratelimited("set kvm_intel.dump_invalid_vmcs=1 to dump internal KVM state.\n");
5633 		return;
5634 	}
5635 
5636 	vmentry_ctl = vmcs_read32(VM_ENTRY_CONTROLS);
5637 	vmexit_ctl = vmcs_read32(VM_EXIT_CONTROLS);
5638 	cpu_based_exec_ctrl = vmcs_read32(CPU_BASED_VM_EXEC_CONTROL);
5639 	pin_based_exec_ctrl = vmcs_read32(PIN_BASED_VM_EXEC_CONTROL);
5640 	cr4 = vmcs_readl(GUEST_CR4);
5641 	efer = vmcs_read64(GUEST_IA32_EFER);
5642 	secondary_exec_control = 0;
5643 	if (cpu_has_secondary_exec_ctrls())
5644 		secondary_exec_control = vmcs_read32(SECONDARY_VM_EXEC_CONTROL);
5645 
5646 	pr_err("*** Guest State ***\n");
5647 	pr_err("CR0: actual=0x%016lx, shadow=0x%016lx, gh_mask=%016lx\n",
5648 	       vmcs_readl(GUEST_CR0), vmcs_readl(CR0_READ_SHADOW),
5649 	       vmcs_readl(CR0_GUEST_HOST_MASK));
5650 	pr_err("CR4: actual=0x%016lx, shadow=0x%016lx, gh_mask=%016lx\n",
5651 	       cr4, vmcs_readl(CR4_READ_SHADOW), vmcs_readl(CR4_GUEST_HOST_MASK));
5652 	pr_err("CR3 = 0x%016lx\n", vmcs_readl(GUEST_CR3));
5653 	if ((secondary_exec_control & SECONDARY_EXEC_ENABLE_EPT) &&
5654 	    (cr4 & X86_CR4_PAE) && !(efer & EFER_LMA))
5655 	{
5656 		pr_err("PDPTR0 = 0x%016llx  PDPTR1 = 0x%016llx\n",
5657 		       vmcs_read64(GUEST_PDPTR0), vmcs_read64(GUEST_PDPTR1));
5658 		pr_err("PDPTR2 = 0x%016llx  PDPTR3 = 0x%016llx\n",
5659 		       vmcs_read64(GUEST_PDPTR2), vmcs_read64(GUEST_PDPTR3));
5660 	}
5661 	pr_err("RSP = 0x%016lx  RIP = 0x%016lx\n",
5662 	       vmcs_readl(GUEST_RSP), vmcs_readl(GUEST_RIP));
5663 	pr_err("RFLAGS=0x%08lx         DR7 = 0x%016lx\n",
5664 	       vmcs_readl(GUEST_RFLAGS), vmcs_readl(GUEST_DR7));
5665 	pr_err("Sysenter RSP=%016lx CS:RIP=%04x:%016lx\n",
5666 	       vmcs_readl(GUEST_SYSENTER_ESP),
5667 	       vmcs_read32(GUEST_SYSENTER_CS), vmcs_readl(GUEST_SYSENTER_EIP));
5668 	vmx_dump_sel("CS:  ", GUEST_CS_SELECTOR);
5669 	vmx_dump_sel("DS:  ", GUEST_DS_SELECTOR);
5670 	vmx_dump_sel("SS:  ", GUEST_SS_SELECTOR);
5671 	vmx_dump_sel("ES:  ", GUEST_ES_SELECTOR);
5672 	vmx_dump_sel("FS:  ", GUEST_FS_SELECTOR);
5673 	vmx_dump_sel("GS:  ", GUEST_GS_SELECTOR);
5674 	vmx_dump_dtsel("GDTR:", GUEST_GDTR_LIMIT);
5675 	vmx_dump_sel("LDTR:", GUEST_LDTR_SELECTOR);
5676 	vmx_dump_dtsel("IDTR:", GUEST_IDTR_LIMIT);
5677 	vmx_dump_sel("TR:  ", GUEST_TR_SELECTOR);
5678 	if ((vmexit_ctl & (VM_EXIT_SAVE_IA32_PAT | VM_EXIT_SAVE_IA32_EFER)) ||
5679 	    (vmentry_ctl & (VM_ENTRY_LOAD_IA32_PAT | VM_ENTRY_LOAD_IA32_EFER)))
5680 		pr_err("EFER =     0x%016llx  PAT = 0x%016llx\n",
5681 		       efer, vmcs_read64(GUEST_IA32_PAT));
5682 	pr_err("DebugCtl = 0x%016llx  DebugExceptions = 0x%016lx\n",
5683 	       vmcs_read64(GUEST_IA32_DEBUGCTL),
5684 	       vmcs_readl(GUEST_PENDING_DBG_EXCEPTIONS));
5685 	if (cpu_has_load_perf_global_ctrl() &&
5686 	    vmentry_ctl & VM_ENTRY_LOAD_IA32_PERF_GLOBAL_CTRL)
5687 		pr_err("PerfGlobCtl = 0x%016llx\n",
5688 		       vmcs_read64(GUEST_IA32_PERF_GLOBAL_CTRL));
5689 	if (vmentry_ctl & VM_ENTRY_LOAD_BNDCFGS)
5690 		pr_err("BndCfgS = 0x%016llx\n", vmcs_read64(GUEST_BNDCFGS));
5691 	pr_err("Interruptibility = %08x  ActivityState = %08x\n",
5692 	       vmcs_read32(GUEST_INTERRUPTIBILITY_INFO),
5693 	       vmcs_read32(GUEST_ACTIVITY_STATE));
5694 	if (secondary_exec_control & SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY)
5695 		pr_err("InterruptStatus = %04x\n",
5696 		       vmcs_read16(GUEST_INTR_STATUS));
5697 
5698 	pr_err("*** Host State ***\n");
5699 	pr_err("RIP = 0x%016lx  RSP = 0x%016lx\n",
5700 	       vmcs_readl(HOST_RIP), vmcs_readl(HOST_RSP));
5701 	pr_err("CS=%04x SS=%04x DS=%04x ES=%04x FS=%04x GS=%04x TR=%04x\n",
5702 	       vmcs_read16(HOST_CS_SELECTOR), vmcs_read16(HOST_SS_SELECTOR),
5703 	       vmcs_read16(HOST_DS_SELECTOR), vmcs_read16(HOST_ES_SELECTOR),
5704 	       vmcs_read16(HOST_FS_SELECTOR), vmcs_read16(HOST_GS_SELECTOR),
5705 	       vmcs_read16(HOST_TR_SELECTOR));
5706 	pr_err("FSBase=%016lx GSBase=%016lx TRBase=%016lx\n",
5707 	       vmcs_readl(HOST_FS_BASE), vmcs_readl(HOST_GS_BASE),
5708 	       vmcs_readl(HOST_TR_BASE));
5709 	pr_err("GDTBase=%016lx IDTBase=%016lx\n",
5710 	       vmcs_readl(HOST_GDTR_BASE), vmcs_readl(HOST_IDTR_BASE));
5711 	pr_err("CR0=%016lx CR3=%016lx CR4=%016lx\n",
5712 	       vmcs_readl(HOST_CR0), vmcs_readl(HOST_CR3),
5713 	       vmcs_readl(HOST_CR4));
5714 	pr_err("Sysenter RSP=%016lx CS:RIP=%04x:%016lx\n",
5715 	       vmcs_readl(HOST_IA32_SYSENTER_ESP),
5716 	       vmcs_read32(HOST_IA32_SYSENTER_CS),
5717 	       vmcs_readl(HOST_IA32_SYSENTER_EIP));
5718 	if (vmexit_ctl & (VM_EXIT_LOAD_IA32_PAT | VM_EXIT_LOAD_IA32_EFER))
5719 		pr_err("EFER = 0x%016llx  PAT = 0x%016llx\n",
5720 		       vmcs_read64(HOST_IA32_EFER),
5721 		       vmcs_read64(HOST_IA32_PAT));
5722 	if (cpu_has_load_perf_global_ctrl() &&
5723 	    vmexit_ctl & VM_EXIT_LOAD_IA32_PERF_GLOBAL_CTRL)
5724 		pr_err("PerfGlobCtl = 0x%016llx\n",
5725 		       vmcs_read64(HOST_IA32_PERF_GLOBAL_CTRL));
5726 
5727 	pr_err("*** Control State ***\n");
5728 	pr_err("PinBased=%08x CPUBased=%08x SecondaryExec=%08x\n",
5729 	       pin_based_exec_ctrl, cpu_based_exec_ctrl, secondary_exec_control);
5730 	pr_err("EntryControls=%08x ExitControls=%08x\n", vmentry_ctl, vmexit_ctl);
5731 	pr_err("ExceptionBitmap=%08x PFECmask=%08x PFECmatch=%08x\n",
5732 	       vmcs_read32(EXCEPTION_BITMAP),
5733 	       vmcs_read32(PAGE_FAULT_ERROR_CODE_MASK),
5734 	       vmcs_read32(PAGE_FAULT_ERROR_CODE_MATCH));
5735 	pr_err("VMEntry: intr_info=%08x errcode=%08x ilen=%08x\n",
5736 	       vmcs_read32(VM_ENTRY_INTR_INFO_FIELD),
5737 	       vmcs_read32(VM_ENTRY_EXCEPTION_ERROR_CODE),
5738 	       vmcs_read32(VM_ENTRY_INSTRUCTION_LEN));
5739 	pr_err("VMExit: intr_info=%08x errcode=%08x ilen=%08x\n",
5740 	       vmcs_read32(VM_EXIT_INTR_INFO),
5741 	       vmcs_read32(VM_EXIT_INTR_ERROR_CODE),
5742 	       vmcs_read32(VM_EXIT_INSTRUCTION_LEN));
5743 	pr_err("        reason=%08x qualification=%016lx\n",
5744 	       vmcs_read32(VM_EXIT_REASON), vmcs_readl(EXIT_QUALIFICATION));
5745 	pr_err("IDTVectoring: info=%08x errcode=%08x\n",
5746 	       vmcs_read32(IDT_VECTORING_INFO_FIELD),
5747 	       vmcs_read32(IDT_VECTORING_ERROR_CODE));
5748 	pr_err("TSC Offset = 0x%016llx\n", vmcs_read64(TSC_OFFSET));
5749 	if (secondary_exec_control & SECONDARY_EXEC_TSC_SCALING)
5750 		pr_err("TSC Multiplier = 0x%016llx\n",
5751 		       vmcs_read64(TSC_MULTIPLIER));
5752 	if (cpu_based_exec_ctrl & CPU_BASED_TPR_SHADOW) {
5753 		if (secondary_exec_control & SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY) {
5754 			u16 status = vmcs_read16(GUEST_INTR_STATUS);
5755 			pr_err("SVI|RVI = %02x|%02x ", status >> 8, status & 0xff);
5756 		}
5757 		pr_cont("TPR Threshold = 0x%02x\n", vmcs_read32(TPR_THRESHOLD));
5758 		if (secondary_exec_control & SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES)
5759 			pr_err("APIC-access addr = 0x%016llx ", vmcs_read64(APIC_ACCESS_ADDR));
5760 		pr_cont("virt-APIC addr = 0x%016llx\n", vmcs_read64(VIRTUAL_APIC_PAGE_ADDR));
5761 	}
5762 	if (pin_based_exec_ctrl & PIN_BASED_POSTED_INTR)
5763 		pr_err("PostedIntrVec = 0x%02x\n", vmcs_read16(POSTED_INTR_NV));
5764 	if ((secondary_exec_control & SECONDARY_EXEC_ENABLE_EPT))
5765 		pr_err("EPT pointer = 0x%016llx\n", vmcs_read64(EPT_POINTER));
5766 	n = vmcs_read32(CR3_TARGET_COUNT);
5767 	for (i = 0; i + 1 < n; i += 4)
5768 		pr_err("CR3 target%u=%016lx target%u=%016lx\n",
5769 		       i, vmcs_readl(CR3_TARGET_VALUE0 + i * 2),
5770 		       i + 1, vmcs_readl(CR3_TARGET_VALUE0 + i * 2 + 2));
5771 	if (i < n)
5772 		pr_err("CR3 target%u=%016lx\n",
5773 		       i, vmcs_readl(CR3_TARGET_VALUE0 + i * 2));
5774 	if (secondary_exec_control & SECONDARY_EXEC_PAUSE_LOOP_EXITING)
5775 		pr_err("PLE Gap=%08x Window=%08x\n",
5776 		       vmcs_read32(PLE_GAP), vmcs_read32(PLE_WINDOW));
5777 	if (secondary_exec_control & SECONDARY_EXEC_ENABLE_VPID)
5778 		pr_err("Virtual processor ID = 0x%04x\n",
5779 		       vmcs_read16(VIRTUAL_PROCESSOR_ID));
5780 }
5781 
5782 /*
5783  * The guest has exited.  See if we can fix it or if we need userspace
5784  * assistance.
5785  */
5786 static int vmx_handle_exit(struct kvm_vcpu *vcpu)
5787 {
5788 	struct vcpu_vmx *vmx = to_vmx(vcpu);
5789 	u32 exit_reason = vmx->exit_reason;
5790 	u32 vectoring_info = vmx->idt_vectoring_info;
5791 
5792 	trace_kvm_exit(exit_reason, vcpu, KVM_ISA_VMX);
5793 
5794 	/*
5795 	 * Flush logged GPAs PML buffer, this will make dirty_bitmap more
5796 	 * updated. Another good is, in kvm_vm_ioctl_get_dirty_log, before
5797 	 * querying dirty_bitmap, we only need to kick all vcpus out of guest
5798 	 * mode as if vcpus is in root mode, the PML buffer must has been
5799 	 * flushed already.
5800 	 */
5801 	if (enable_pml)
5802 		vmx_flush_pml_buffer(vcpu);
5803 
5804 	/* If guest state is invalid, start emulating */
5805 	if (vmx->emulation_required)
5806 		return handle_invalid_guest_state(vcpu);
5807 
5808 	if (is_guest_mode(vcpu) && nested_vmx_exit_reflected(vcpu, exit_reason))
5809 		return nested_vmx_reflect_vmexit(vcpu, exit_reason);
5810 
5811 	if (exit_reason & VMX_EXIT_REASONS_FAILED_VMENTRY) {
5812 		dump_vmcs();
5813 		vcpu->run->exit_reason = KVM_EXIT_FAIL_ENTRY;
5814 		vcpu->run->fail_entry.hardware_entry_failure_reason
5815 			= exit_reason;
5816 		return 0;
5817 	}
5818 
5819 	if (unlikely(vmx->fail)) {
5820 		dump_vmcs();
5821 		vcpu->run->exit_reason = KVM_EXIT_FAIL_ENTRY;
5822 		vcpu->run->fail_entry.hardware_entry_failure_reason
5823 			= vmcs_read32(VM_INSTRUCTION_ERROR);
5824 		return 0;
5825 	}
5826 
5827 	/*
5828 	 * Note:
5829 	 * Do not try to fix EXIT_REASON_EPT_MISCONFIG if it caused by
5830 	 * delivery event since it indicates guest is accessing MMIO.
5831 	 * The vm-exit can be triggered again after return to guest that
5832 	 * will cause infinite loop.
5833 	 */
5834 	if ((vectoring_info & VECTORING_INFO_VALID_MASK) &&
5835 			(exit_reason != EXIT_REASON_EXCEPTION_NMI &&
5836 			exit_reason != EXIT_REASON_EPT_VIOLATION &&
5837 			exit_reason != EXIT_REASON_PML_FULL &&
5838 			exit_reason != EXIT_REASON_TASK_SWITCH)) {
5839 		vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
5840 		vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_DELIVERY_EV;
5841 		vcpu->run->internal.ndata = 3;
5842 		vcpu->run->internal.data[0] = vectoring_info;
5843 		vcpu->run->internal.data[1] = exit_reason;
5844 		vcpu->run->internal.data[2] = vcpu->arch.exit_qualification;
5845 		if (exit_reason == EXIT_REASON_EPT_MISCONFIG) {
5846 			vcpu->run->internal.ndata++;
5847 			vcpu->run->internal.data[3] =
5848 				vmcs_read64(GUEST_PHYSICAL_ADDRESS);
5849 		}
5850 		return 0;
5851 	}
5852 
5853 	if (unlikely(!enable_vnmi &&
5854 		     vmx->loaded_vmcs->soft_vnmi_blocked)) {
5855 		if (vmx_interrupt_allowed(vcpu)) {
5856 			vmx->loaded_vmcs->soft_vnmi_blocked = 0;
5857 		} else if (vmx->loaded_vmcs->vnmi_blocked_time > 1000000000LL &&
5858 			   vcpu->arch.nmi_pending) {
5859 			/*
5860 			 * This CPU don't support us in finding the end of an
5861 			 * NMI-blocked window if the guest runs with IRQs
5862 			 * disabled. So we pull the trigger after 1 s of
5863 			 * futile waiting, but inform the user about this.
5864 			 */
5865 			printk(KERN_WARNING "%s: Breaking out of NMI-blocked "
5866 			       "state on VCPU %d after 1 s timeout\n",
5867 			       __func__, vcpu->vcpu_id);
5868 			vmx->loaded_vmcs->soft_vnmi_blocked = 0;
5869 		}
5870 	}
5871 
5872 	if (exit_reason < kvm_vmx_max_exit_handlers
5873 	    && kvm_vmx_exit_handlers[exit_reason]) {
5874 #ifdef CONFIG_RETPOLINE
5875 		if (exit_reason == EXIT_REASON_MSR_WRITE)
5876 			return kvm_emulate_wrmsr(vcpu);
5877 		else if (exit_reason == EXIT_REASON_PREEMPTION_TIMER)
5878 			return handle_preemption_timer(vcpu);
5879 		else if (exit_reason == EXIT_REASON_PENDING_INTERRUPT)
5880 			return handle_interrupt_window(vcpu);
5881 		else if (exit_reason == EXIT_REASON_EXTERNAL_INTERRUPT)
5882 			return handle_external_interrupt(vcpu);
5883 		else if (exit_reason == EXIT_REASON_HLT)
5884 			return kvm_emulate_halt(vcpu);
5885 		else if (exit_reason == EXIT_REASON_EPT_MISCONFIG)
5886 			return handle_ept_misconfig(vcpu);
5887 #endif
5888 		return kvm_vmx_exit_handlers[exit_reason](vcpu);
5889 	} else {
5890 		vcpu_unimpl(vcpu, "vmx: unexpected exit reason 0x%x\n",
5891 				exit_reason);
5892 		dump_vmcs();
5893 		vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
5894 		vcpu->run->internal.suberror =
5895 			KVM_INTERNAL_ERROR_UNEXPECTED_EXIT_REASON;
5896 		vcpu->run->internal.ndata = 1;
5897 		vcpu->run->internal.data[0] = exit_reason;
5898 		return 0;
5899 	}
5900 }
5901 
5902 /*
5903  * Software based L1D cache flush which is used when microcode providing
5904  * the cache control MSR is not loaded.
5905  *
5906  * The L1D cache is 32 KiB on Nehalem and later microarchitectures, but to
5907  * flush it is required to read in 64 KiB because the replacement algorithm
5908  * is not exactly LRU. This could be sized at runtime via topology
5909  * information but as all relevant affected CPUs have 32KiB L1D cache size
5910  * there is no point in doing so.
5911  */
5912 static void vmx_l1d_flush(struct kvm_vcpu *vcpu)
5913 {
5914 	int size = PAGE_SIZE << L1D_CACHE_ORDER;
5915 
5916 	/*
5917 	 * This code is only executed when the the flush mode is 'cond' or
5918 	 * 'always'
5919 	 */
5920 	if (static_branch_likely(&vmx_l1d_flush_cond)) {
5921 		bool flush_l1d;
5922 
5923 		/*
5924 		 * Clear the per-vcpu flush bit, it gets set again
5925 		 * either from vcpu_run() or from one of the unsafe
5926 		 * VMEXIT handlers.
5927 		 */
5928 		flush_l1d = vcpu->arch.l1tf_flush_l1d;
5929 		vcpu->arch.l1tf_flush_l1d = false;
5930 
5931 		/*
5932 		 * Clear the per-cpu flush bit, it gets set again from
5933 		 * the interrupt handlers.
5934 		 */
5935 		flush_l1d |= kvm_get_cpu_l1tf_flush_l1d();
5936 		kvm_clear_cpu_l1tf_flush_l1d();
5937 
5938 		if (!flush_l1d)
5939 			return;
5940 	}
5941 
5942 	vcpu->stat.l1d_flush++;
5943 
5944 	if (static_cpu_has(X86_FEATURE_FLUSH_L1D)) {
5945 		wrmsrl(MSR_IA32_FLUSH_CMD, L1D_FLUSH);
5946 		return;
5947 	}
5948 
5949 	asm volatile(
5950 		/* First ensure the pages are in the TLB */
5951 		"xorl	%%eax, %%eax\n"
5952 		".Lpopulate_tlb:\n\t"
5953 		"movzbl	(%[flush_pages], %%" _ASM_AX "), %%ecx\n\t"
5954 		"addl	$4096, %%eax\n\t"
5955 		"cmpl	%%eax, %[size]\n\t"
5956 		"jne	.Lpopulate_tlb\n\t"
5957 		"xorl	%%eax, %%eax\n\t"
5958 		"cpuid\n\t"
5959 		/* Now fill the cache */
5960 		"xorl	%%eax, %%eax\n"
5961 		".Lfill_cache:\n"
5962 		"movzbl	(%[flush_pages], %%" _ASM_AX "), %%ecx\n\t"
5963 		"addl	$64, %%eax\n\t"
5964 		"cmpl	%%eax, %[size]\n\t"
5965 		"jne	.Lfill_cache\n\t"
5966 		"lfence\n"
5967 		:: [flush_pages] "r" (vmx_l1d_flush_pages),
5968 		    [size] "r" (size)
5969 		: "eax", "ebx", "ecx", "edx");
5970 }
5971 
5972 static void update_cr8_intercept(struct kvm_vcpu *vcpu, int tpr, int irr)
5973 {
5974 	struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
5975 	int tpr_threshold;
5976 
5977 	if (is_guest_mode(vcpu) &&
5978 		nested_cpu_has(vmcs12, CPU_BASED_TPR_SHADOW))
5979 		return;
5980 
5981 	tpr_threshold = (irr == -1 || tpr < irr) ? 0 : irr;
5982 	if (is_guest_mode(vcpu))
5983 		to_vmx(vcpu)->nested.l1_tpr_threshold = tpr_threshold;
5984 	else
5985 		vmcs_write32(TPR_THRESHOLD, tpr_threshold);
5986 }
5987 
5988 void vmx_set_virtual_apic_mode(struct kvm_vcpu *vcpu)
5989 {
5990 	struct vcpu_vmx *vmx = to_vmx(vcpu);
5991 	u32 sec_exec_control;
5992 
5993 	if (!lapic_in_kernel(vcpu))
5994 		return;
5995 
5996 	if (!flexpriority_enabled &&
5997 	    !cpu_has_vmx_virtualize_x2apic_mode())
5998 		return;
5999 
6000 	/* Postpone execution until vmcs01 is the current VMCS. */
6001 	if (is_guest_mode(vcpu)) {
6002 		vmx->nested.change_vmcs01_virtual_apic_mode = true;
6003 		return;
6004 	}
6005 
6006 	sec_exec_control = secondary_exec_controls_get(vmx);
6007 	sec_exec_control &= ~(SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES |
6008 			      SECONDARY_EXEC_VIRTUALIZE_X2APIC_MODE);
6009 
6010 	switch (kvm_get_apic_mode(vcpu)) {
6011 	case LAPIC_MODE_INVALID:
6012 		WARN_ONCE(true, "Invalid local APIC state");
6013 	case LAPIC_MODE_DISABLED:
6014 		break;
6015 	case LAPIC_MODE_XAPIC:
6016 		if (flexpriority_enabled) {
6017 			sec_exec_control |=
6018 				SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES;
6019 			vmx_flush_tlb(vcpu, true);
6020 		}
6021 		break;
6022 	case LAPIC_MODE_X2APIC:
6023 		if (cpu_has_vmx_virtualize_x2apic_mode())
6024 			sec_exec_control |=
6025 				SECONDARY_EXEC_VIRTUALIZE_X2APIC_MODE;
6026 		break;
6027 	}
6028 	secondary_exec_controls_set(vmx, sec_exec_control);
6029 
6030 	vmx_update_msr_bitmap(vcpu);
6031 }
6032 
6033 static void vmx_set_apic_access_page_addr(struct kvm_vcpu *vcpu, hpa_t hpa)
6034 {
6035 	if (!is_guest_mode(vcpu)) {
6036 		vmcs_write64(APIC_ACCESS_ADDR, hpa);
6037 		vmx_flush_tlb(vcpu, true);
6038 	}
6039 }
6040 
6041 static void vmx_hwapic_isr_update(struct kvm_vcpu *vcpu, int max_isr)
6042 {
6043 	u16 status;
6044 	u8 old;
6045 
6046 	if (max_isr == -1)
6047 		max_isr = 0;
6048 
6049 	status = vmcs_read16(GUEST_INTR_STATUS);
6050 	old = status >> 8;
6051 	if (max_isr != old) {
6052 		status &= 0xff;
6053 		status |= max_isr << 8;
6054 		vmcs_write16(GUEST_INTR_STATUS, status);
6055 	}
6056 }
6057 
6058 static void vmx_set_rvi(int vector)
6059 {
6060 	u16 status;
6061 	u8 old;
6062 
6063 	if (vector == -1)
6064 		vector = 0;
6065 
6066 	status = vmcs_read16(GUEST_INTR_STATUS);
6067 	old = (u8)status & 0xff;
6068 	if ((u8)vector != old) {
6069 		status &= ~0xff;
6070 		status |= (u8)vector;
6071 		vmcs_write16(GUEST_INTR_STATUS, status);
6072 	}
6073 }
6074 
6075 static void vmx_hwapic_irr_update(struct kvm_vcpu *vcpu, int max_irr)
6076 {
6077 	/*
6078 	 * When running L2, updating RVI is only relevant when
6079 	 * vmcs12 virtual-interrupt-delivery enabled.
6080 	 * However, it can be enabled only when L1 also
6081 	 * intercepts external-interrupts and in that case
6082 	 * we should not update vmcs02 RVI but instead intercept
6083 	 * interrupt. Therefore, do nothing when running L2.
6084 	 */
6085 	if (!is_guest_mode(vcpu))
6086 		vmx_set_rvi(max_irr);
6087 }
6088 
6089 static int vmx_sync_pir_to_irr(struct kvm_vcpu *vcpu)
6090 {
6091 	struct vcpu_vmx *vmx = to_vmx(vcpu);
6092 	int max_irr;
6093 	bool max_irr_updated;
6094 
6095 	WARN_ON(!vcpu->arch.apicv_active);
6096 	if (pi_test_on(&vmx->pi_desc)) {
6097 		pi_clear_on(&vmx->pi_desc);
6098 		/*
6099 		 * IOMMU can write to PID.ON, so the barrier matters even on UP.
6100 		 * But on x86 this is just a compiler barrier anyway.
6101 		 */
6102 		smp_mb__after_atomic();
6103 		max_irr_updated =
6104 			kvm_apic_update_irr(vcpu, vmx->pi_desc.pir, &max_irr);
6105 
6106 		/*
6107 		 * If we are running L2 and L1 has a new pending interrupt
6108 		 * which can be injected, we should re-evaluate
6109 		 * what should be done with this new L1 interrupt.
6110 		 * If L1 intercepts external-interrupts, we should
6111 		 * exit from L2 to L1. Otherwise, interrupt should be
6112 		 * delivered directly to L2.
6113 		 */
6114 		if (is_guest_mode(vcpu) && max_irr_updated) {
6115 			if (nested_exit_on_intr(vcpu))
6116 				kvm_vcpu_exiting_guest_mode(vcpu);
6117 			else
6118 				kvm_make_request(KVM_REQ_EVENT, vcpu);
6119 		}
6120 	} else {
6121 		max_irr = kvm_lapic_find_highest_irr(vcpu);
6122 	}
6123 	vmx_hwapic_irr_update(vcpu, max_irr);
6124 	return max_irr;
6125 }
6126 
6127 static bool vmx_dy_apicv_has_pending_interrupt(struct kvm_vcpu *vcpu)
6128 {
6129 	struct pi_desc *pi_desc = vcpu_to_pi_desc(vcpu);
6130 
6131 	return pi_test_on(pi_desc) ||
6132 		(pi_test_sn(pi_desc) && !pi_is_pir_empty(pi_desc));
6133 }
6134 
6135 static void vmx_load_eoi_exitmap(struct kvm_vcpu *vcpu, u64 *eoi_exit_bitmap)
6136 {
6137 	if (!kvm_vcpu_apicv_active(vcpu))
6138 		return;
6139 
6140 	vmcs_write64(EOI_EXIT_BITMAP0, eoi_exit_bitmap[0]);
6141 	vmcs_write64(EOI_EXIT_BITMAP1, eoi_exit_bitmap[1]);
6142 	vmcs_write64(EOI_EXIT_BITMAP2, eoi_exit_bitmap[2]);
6143 	vmcs_write64(EOI_EXIT_BITMAP3, eoi_exit_bitmap[3]);
6144 }
6145 
6146 static void vmx_apicv_post_state_restore(struct kvm_vcpu *vcpu)
6147 {
6148 	struct vcpu_vmx *vmx = to_vmx(vcpu);
6149 
6150 	pi_clear_on(&vmx->pi_desc);
6151 	memset(vmx->pi_desc.pir, 0, sizeof(vmx->pi_desc.pir));
6152 }
6153 
6154 static void handle_exception_nmi_irqoff(struct vcpu_vmx *vmx)
6155 {
6156 	vmx->exit_intr_info = vmcs_read32(VM_EXIT_INTR_INFO);
6157 
6158 	/* if exit due to PF check for async PF */
6159 	if (is_page_fault(vmx->exit_intr_info))
6160 		vmx->vcpu.arch.apf.host_apf_reason = kvm_read_and_reset_pf_reason();
6161 
6162 	/* Handle machine checks before interrupts are enabled */
6163 	if (is_machine_check(vmx->exit_intr_info))
6164 		kvm_machine_check();
6165 
6166 	/* We need to handle NMIs before interrupts are enabled */
6167 	if (is_nmi(vmx->exit_intr_info)) {
6168 		kvm_before_interrupt(&vmx->vcpu);
6169 		asm("int $2");
6170 		kvm_after_interrupt(&vmx->vcpu);
6171 	}
6172 }
6173 
6174 static void handle_external_interrupt_irqoff(struct kvm_vcpu *vcpu)
6175 {
6176 	unsigned int vector;
6177 	unsigned long entry;
6178 #ifdef CONFIG_X86_64
6179 	unsigned long tmp;
6180 #endif
6181 	gate_desc *desc;
6182 	u32 intr_info;
6183 
6184 	intr_info = vmcs_read32(VM_EXIT_INTR_INFO);
6185 	if (WARN_ONCE(!is_external_intr(intr_info),
6186 	    "KVM: unexpected VM-Exit interrupt info: 0x%x", intr_info))
6187 		return;
6188 
6189 	vector = intr_info & INTR_INFO_VECTOR_MASK;
6190 	desc = (gate_desc *)host_idt_base + vector;
6191 	entry = gate_offset(desc);
6192 
6193 	kvm_before_interrupt(vcpu);
6194 
6195 	asm volatile(
6196 #ifdef CONFIG_X86_64
6197 		"mov %%" _ASM_SP ", %[sp]\n\t"
6198 		"and $0xfffffffffffffff0, %%" _ASM_SP "\n\t"
6199 		"push $%c[ss]\n\t"
6200 		"push %[sp]\n\t"
6201 #endif
6202 		"pushf\n\t"
6203 		__ASM_SIZE(push) " $%c[cs]\n\t"
6204 		CALL_NOSPEC
6205 		:
6206 #ifdef CONFIG_X86_64
6207 		[sp]"=&r"(tmp),
6208 #endif
6209 		ASM_CALL_CONSTRAINT
6210 		:
6211 		THUNK_TARGET(entry),
6212 		[ss]"i"(__KERNEL_DS),
6213 		[cs]"i"(__KERNEL_CS)
6214 	);
6215 
6216 	kvm_after_interrupt(vcpu);
6217 }
6218 STACK_FRAME_NON_STANDARD(handle_external_interrupt_irqoff);
6219 
6220 static void vmx_handle_exit_irqoff(struct kvm_vcpu *vcpu)
6221 {
6222 	struct vcpu_vmx *vmx = to_vmx(vcpu);
6223 
6224 	if (vmx->exit_reason == EXIT_REASON_EXTERNAL_INTERRUPT)
6225 		handle_external_interrupt_irqoff(vcpu);
6226 	else if (vmx->exit_reason == EXIT_REASON_EXCEPTION_NMI)
6227 		handle_exception_nmi_irqoff(vmx);
6228 }
6229 
6230 static bool vmx_has_emulated_msr(int index)
6231 {
6232 	switch (index) {
6233 	case MSR_IA32_SMBASE:
6234 		/*
6235 		 * We cannot do SMM unless we can run the guest in big
6236 		 * real mode.
6237 		 */
6238 		return enable_unrestricted_guest || emulate_invalid_guest_state;
6239 	case MSR_IA32_VMX_BASIC ... MSR_IA32_VMX_VMFUNC:
6240 		return nested;
6241 	case MSR_AMD64_VIRT_SPEC_CTRL:
6242 		/* This is AMD only.  */
6243 		return false;
6244 	default:
6245 		return true;
6246 	}
6247 }
6248 
6249 static bool vmx_pt_supported(void)
6250 {
6251 	return pt_mode == PT_MODE_HOST_GUEST;
6252 }
6253 
6254 static void vmx_recover_nmi_blocking(struct vcpu_vmx *vmx)
6255 {
6256 	u32 exit_intr_info;
6257 	bool unblock_nmi;
6258 	u8 vector;
6259 	bool idtv_info_valid;
6260 
6261 	idtv_info_valid = vmx->idt_vectoring_info & VECTORING_INFO_VALID_MASK;
6262 
6263 	if (enable_vnmi) {
6264 		if (vmx->loaded_vmcs->nmi_known_unmasked)
6265 			return;
6266 		/*
6267 		 * Can't use vmx->exit_intr_info since we're not sure what
6268 		 * the exit reason is.
6269 		 */
6270 		exit_intr_info = vmcs_read32(VM_EXIT_INTR_INFO);
6271 		unblock_nmi = (exit_intr_info & INTR_INFO_UNBLOCK_NMI) != 0;
6272 		vector = exit_intr_info & INTR_INFO_VECTOR_MASK;
6273 		/*
6274 		 * SDM 3: 27.7.1.2 (September 2008)
6275 		 * Re-set bit "block by NMI" before VM entry if vmexit caused by
6276 		 * a guest IRET fault.
6277 		 * SDM 3: 23.2.2 (September 2008)
6278 		 * Bit 12 is undefined in any of the following cases:
6279 		 *  If the VM exit sets the valid bit in the IDT-vectoring
6280 		 *   information field.
6281 		 *  If the VM exit is due to a double fault.
6282 		 */
6283 		if ((exit_intr_info & INTR_INFO_VALID_MASK) && unblock_nmi &&
6284 		    vector != DF_VECTOR && !idtv_info_valid)
6285 			vmcs_set_bits(GUEST_INTERRUPTIBILITY_INFO,
6286 				      GUEST_INTR_STATE_NMI);
6287 		else
6288 			vmx->loaded_vmcs->nmi_known_unmasked =
6289 				!(vmcs_read32(GUEST_INTERRUPTIBILITY_INFO)
6290 				  & GUEST_INTR_STATE_NMI);
6291 	} else if (unlikely(vmx->loaded_vmcs->soft_vnmi_blocked))
6292 		vmx->loaded_vmcs->vnmi_blocked_time +=
6293 			ktime_to_ns(ktime_sub(ktime_get(),
6294 					      vmx->loaded_vmcs->entry_time));
6295 }
6296 
6297 static void __vmx_complete_interrupts(struct kvm_vcpu *vcpu,
6298 				      u32 idt_vectoring_info,
6299 				      int instr_len_field,
6300 				      int error_code_field)
6301 {
6302 	u8 vector;
6303 	int type;
6304 	bool idtv_info_valid;
6305 
6306 	idtv_info_valid = idt_vectoring_info & VECTORING_INFO_VALID_MASK;
6307 
6308 	vcpu->arch.nmi_injected = false;
6309 	kvm_clear_exception_queue(vcpu);
6310 	kvm_clear_interrupt_queue(vcpu);
6311 
6312 	if (!idtv_info_valid)
6313 		return;
6314 
6315 	kvm_make_request(KVM_REQ_EVENT, vcpu);
6316 
6317 	vector = idt_vectoring_info & VECTORING_INFO_VECTOR_MASK;
6318 	type = idt_vectoring_info & VECTORING_INFO_TYPE_MASK;
6319 
6320 	switch (type) {
6321 	case INTR_TYPE_NMI_INTR:
6322 		vcpu->arch.nmi_injected = true;
6323 		/*
6324 		 * SDM 3: 27.7.1.2 (September 2008)
6325 		 * Clear bit "block by NMI" before VM entry if a NMI
6326 		 * delivery faulted.
6327 		 */
6328 		vmx_set_nmi_mask(vcpu, false);
6329 		break;
6330 	case INTR_TYPE_SOFT_EXCEPTION:
6331 		vcpu->arch.event_exit_inst_len = vmcs_read32(instr_len_field);
6332 		/* fall through */
6333 	case INTR_TYPE_HARD_EXCEPTION:
6334 		if (idt_vectoring_info & VECTORING_INFO_DELIVER_CODE_MASK) {
6335 			u32 err = vmcs_read32(error_code_field);
6336 			kvm_requeue_exception_e(vcpu, vector, err);
6337 		} else
6338 			kvm_requeue_exception(vcpu, vector);
6339 		break;
6340 	case INTR_TYPE_SOFT_INTR:
6341 		vcpu->arch.event_exit_inst_len = vmcs_read32(instr_len_field);
6342 		/* fall through */
6343 	case INTR_TYPE_EXT_INTR:
6344 		kvm_queue_interrupt(vcpu, vector, type == INTR_TYPE_SOFT_INTR);
6345 		break;
6346 	default:
6347 		break;
6348 	}
6349 }
6350 
6351 static void vmx_complete_interrupts(struct vcpu_vmx *vmx)
6352 {
6353 	__vmx_complete_interrupts(&vmx->vcpu, vmx->idt_vectoring_info,
6354 				  VM_EXIT_INSTRUCTION_LEN,
6355 				  IDT_VECTORING_ERROR_CODE);
6356 }
6357 
6358 static void vmx_cancel_injection(struct kvm_vcpu *vcpu)
6359 {
6360 	__vmx_complete_interrupts(vcpu,
6361 				  vmcs_read32(VM_ENTRY_INTR_INFO_FIELD),
6362 				  VM_ENTRY_INSTRUCTION_LEN,
6363 				  VM_ENTRY_EXCEPTION_ERROR_CODE);
6364 
6365 	vmcs_write32(VM_ENTRY_INTR_INFO_FIELD, 0);
6366 }
6367 
6368 static void atomic_switch_perf_msrs(struct vcpu_vmx *vmx)
6369 {
6370 	int i, nr_msrs;
6371 	struct perf_guest_switch_msr *msrs;
6372 
6373 	msrs = perf_guest_get_msrs(&nr_msrs);
6374 
6375 	if (!msrs)
6376 		return;
6377 
6378 	for (i = 0; i < nr_msrs; i++)
6379 		if (msrs[i].host == msrs[i].guest)
6380 			clear_atomic_switch_msr(vmx, msrs[i].msr);
6381 		else
6382 			add_atomic_switch_msr(vmx, msrs[i].msr, msrs[i].guest,
6383 					msrs[i].host, false);
6384 }
6385 
6386 static void atomic_switch_umwait_control_msr(struct vcpu_vmx *vmx)
6387 {
6388 	u32 host_umwait_control;
6389 
6390 	if (!vmx_has_waitpkg(vmx))
6391 		return;
6392 
6393 	host_umwait_control = get_umwait_control_msr();
6394 
6395 	if (vmx->msr_ia32_umwait_control != host_umwait_control)
6396 		add_atomic_switch_msr(vmx, MSR_IA32_UMWAIT_CONTROL,
6397 			vmx->msr_ia32_umwait_control,
6398 			host_umwait_control, false);
6399 	else
6400 		clear_atomic_switch_msr(vmx, MSR_IA32_UMWAIT_CONTROL);
6401 }
6402 
6403 static void vmx_update_hv_timer(struct kvm_vcpu *vcpu)
6404 {
6405 	struct vcpu_vmx *vmx = to_vmx(vcpu);
6406 	u64 tscl;
6407 	u32 delta_tsc;
6408 
6409 	if (vmx->req_immediate_exit) {
6410 		vmcs_write32(VMX_PREEMPTION_TIMER_VALUE, 0);
6411 		vmx->loaded_vmcs->hv_timer_soft_disabled = false;
6412 	} else if (vmx->hv_deadline_tsc != -1) {
6413 		tscl = rdtsc();
6414 		if (vmx->hv_deadline_tsc > tscl)
6415 			/* set_hv_timer ensures the delta fits in 32-bits */
6416 			delta_tsc = (u32)((vmx->hv_deadline_tsc - tscl) >>
6417 				cpu_preemption_timer_multi);
6418 		else
6419 			delta_tsc = 0;
6420 
6421 		vmcs_write32(VMX_PREEMPTION_TIMER_VALUE, delta_tsc);
6422 		vmx->loaded_vmcs->hv_timer_soft_disabled = false;
6423 	} else if (!vmx->loaded_vmcs->hv_timer_soft_disabled) {
6424 		vmcs_write32(VMX_PREEMPTION_TIMER_VALUE, -1);
6425 		vmx->loaded_vmcs->hv_timer_soft_disabled = true;
6426 	}
6427 }
6428 
6429 void vmx_update_host_rsp(struct vcpu_vmx *vmx, unsigned long host_rsp)
6430 {
6431 	if (unlikely(host_rsp != vmx->loaded_vmcs->host_state.rsp)) {
6432 		vmx->loaded_vmcs->host_state.rsp = host_rsp;
6433 		vmcs_writel(HOST_RSP, host_rsp);
6434 	}
6435 }
6436 
6437 bool __vmx_vcpu_run(struct vcpu_vmx *vmx, unsigned long *regs, bool launched);
6438 
6439 static void vmx_vcpu_run(struct kvm_vcpu *vcpu)
6440 {
6441 	struct vcpu_vmx *vmx = to_vmx(vcpu);
6442 	unsigned long cr3, cr4;
6443 
6444 	/* Record the guest's net vcpu time for enforced NMI injections. */
6445 	if (unlikely(!enable_vnmi &&
6446 		     vmx->loaded_vmcs->soft_vnmi_blocked))
6447 		vmx->loaded_vmcs->entry_time = ktime_get();
6448 
6449 	/* Don't enter VMX if guest state is invalid, let the exit handler
6450 	   start emulation until we arrive back to a valid state */
6451 	if (vmx->emulation_required)
6452 		return;
6453 
6454 	if (vmx->ple_window_dirty) {
6455 		vmx->ple_window_dirty = false;
6456 		vmcs_write32(PLE_WINDOW, vmx->ple_window);
6457 	}
6458 
6459 	if (vmx->nested.need_vmcs12_to_shadow_sync)
6460 		nested_sync_vmcs12_to_shadow(vcpu);
6461 
6462 	if (kvm_register_is_dirty(vcpu, VCPU_REGS_RSP))
6463 		vmcs_writel(GUEST_RSP, vcpu->arch.regs[VCPU_REGS_RSP]);
6464 	if (kvm_register_is_dirty(vcpu, VCPU_REGS_RIP))
6465 		vmcs_writel(GUEST_RIP, vcpu->arch.regs[VCPU_REGS_RIP]);
6466 
6467 	cr3 = __get_current_cr3_fast();
6468 	if (unlikely(cr3 != vmx->loaded_vmcs->host_state.cr3)) {
6469 		vmcs_writel(HOST_CR3, cr3);
6470 		vmx->loaded_vmcs->host_state.cr3 = cr3;
6471 	}
6472 
6473 	cr4 = cr4_read_shadow();
6474 	if (unlikely(cr4 != vmx->loaded_vmcs->host_state.cr4)) {
6475 		vmcs_writel(HOST_CR4, cr4);
6476 		vmx->loaded_vmcs->host_state.cr4 = cr4;
6477 	}
6478 
6479 	/* When single-stepping over STI and MOV SS, we must clear the
6480 	 * corresponding interruptibility bits in the guest state. Otherwise
6481 	 * vmentry fails as it then expects bit 14 (BS) in pending debug
6482 	 * exceptions being set, but that's not correct for the guest debugging
6483 	 * case. */
6484 	if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)
6485 		vmx_set_interrupt_shadow(vcpu, 0);
6486 
6487 	kvm_load_guest_xsave_state(vcpu);
6488 
6489 	if (static_cpu_has(X86_FEATURE_PKU) &&
6490 	    kvm_read_cr4_bits(vcpu, X86_CR4_PKE) &&
6491 	    vcpu->arch.pkru != vmx->host_pkru)
6492 		__write_pkru(vcpu->arch.pkru);
6493 
6494 	pt_guest_enter(vmx);
6495 
6496 	atomic_switch_perf_msrs(vmx);
6497 	atomic_switch_umwait_control_msr(vmx);
6498 
6499 	if (enable_preemption_timer)
6500 		vmx_update_hv_timer(vcpu);
6501 
6502 	if (lapic_in_kernel(vcpu) &&
6503 		vcpu->arch.apic->lapic_timer.timer_advance_ns)
6504 		kvm_wait_lapic_expire(vcpu);
6505 
6506 	/*
6507 	 * If this vCPU has touched SPEC_CTRL, restore the guest's value if
6508 	 * it's non-zero. Since vmentry is serialising on affected CPUs, there
6509 	 * is no need to worry about the conditional branch over the wrmsr
6510 	 * being speculatively taken.
6511 	 */
6512 	x86_spec_ctrl_set_guest(vmx->spec_ctrl, 0);
6513 
6514 	/* L1D Flush includes CPU buffer clear to mitigate MDS */
6515 	if (static_branch_unlikely(&vmx_l1d_should_flush))
6516 		vmx_l1d_flush(vcpu);
6517 	else if (static_branch_unlikely(&mds_user_clear))
6518 		mds_clear_cpu_buffers();
6519 
6520 	if (vcpu->arch.cr2 != read_cr2())
6521 		write_cr2(vcpu->arch.cr2);
6522 
6523 	vmx->fail = __vmx_vcpu_run(vmx, (unsigned long *)&vcpu->arch.regs,
6524 				   vmx->loaded_vmcs->launched);
6525 
6526 	vcpu->arch.cr2 = read_cr2();
6527 
6528 	/*
6529 	 * We do not use IBRS in the kernel. If this vCPU has used the
6530 	 * SPEC_CTRL MSR it may have left it on; save the value and
6531 	 * turn it off. This is much more efficient than blindly adding
6532 	 * it to the atomic save/restore list. Especially as the former
6533 	 * (Saving guest MSRs on vmexit) doesn't even exist in KVM.
6534 	 *
6535 	 * For non-nested case:
6536 	 * If the L01 MSR bitmap does not intercept the MSR, then we need to
6537 	 * save it.
6538 	 *
6539 	 * For nested case:
6540 	 * If the L02 MSR bitmap does not intercept the MSR, then we need to
6541 	 * save it.
6542 	 */
6543 	if (unlikely(!msr_write_intercepted(vcpu, MSR_IA32_SPEC_CTRL)))
6544 		vmx->spec_ctrl = native_read_msr(MSR_IA32_SPEC_CTRL);
6545 
6546 	x86_spec_ctrl_restore_host(vmx->spec_ctrl, 0);
6547 
6548 	/* All fields are clean at this point */
6549 	if (static_branch_unlikely(&enable_evmcs))
6550 		current_evmcs->hv_clean_fields |=
6551 			HV_VMX_ENLIGHTENED_CLEAN_FIELD_ALL;
6552 
6553 	if (static_branch_unlikely(&enable_evmcs))
6554 		current_evmcs->hv_vp_id = vcpu->arch.hyperv.vp_index;
6555 
6556 	/* MSR_IA32_DEBUGCTLMSR is zeroed on vmexit. Restore it if needed */
6557 	if (vmx->host_debugctlmsr)
6558 		update_debugctlmsr(vmx->host_debugctlmsr);
6559 
6560 #ifndef CONFIG_X86_64
6561 	/*
6562 	 * The sysexit path does not restore ds/es, so we must set them to
6563 	 * a reasonable value ourselves.
6564 	 *
6565 	 * We can't defer this to vmx_prepare_switch_to_host() since that
6566 	 * function may be executed in interrupt context, which saves and
6567 	 * restore segments around it, nullifying its effect.
6568 	 */
6569 	loadsegment(ds, __USER_DS);
6570 	loadsegment(es, __USER_DS);
6571 #endif
6572 
6573 	vcpu->arch.regs_avail = ~((1 << VCPU_REGS_RIP) | (1 << VCPU_REGS_RSP)
6574 				  | (1 << VCPU_EXREG_RFLAGS)
6575 				  | (1 << VCPU_EXREG_PDPTR)
6576 				  | (1 << VCPU_EXREG_SEGMENTS)
6577 				  | (1 << VCPU_EXREG_CR3));
6578 	vcpu->arch.regs_dirty = 0;
6579 
6580 	pt_guest_exit(vmx);
6581 
6582 	/*
6583 	 * eager fpu is enabled if PKEY is supported and CR4 is switched
6584 	 * back on host, so it is safe to read guest PKRU from current
6585 	 * XSAVE.
6586 	 */
6587 	if (static_cpu_has(X86_FEATURE_PKU) &&
6588 	    kvm_read_cr4_bits(vcpu, X86_CR4_PKE)) {
6589 		vcpu->arch.pkru = rdpkru();
6590 		if (vcpu->arch.pkru != vmx->host_pkru)
6591 			__write_pkru(vmx->host_pkru);
6592 	}
6593 
6594 	kvm_load_host_xsave_state(vcpu);
6595 
6596 	vmx->nested.nested_run_pending = 0;
6597 	vmx->idt_vectoring_info = 0;
6598 
6599 	vmx->exit_reason = vmx->fail ? 0xdead : vmcs_read32(VM_EXIT_REASON);
6600 	if ((u16)vmx->exit_reason == EXIT_REASON_MCE_DURING_VMENTRY)
6601 		kvm_machine_check();
6602 
6603 	if (vmx->fail || (vmx->exit_reason & VMX_EXIT_REASONS_FAILED_VMENTRY))
6604 		return;
6605 
6606 	vmx->loaded_vmcs->launched = 1;
6607 	vmx->idt_vectoring_info = vmcs_read32(IDT_VECTORING_INFO_FIELD);
6608 
6609 	vmx_recover_nmi_blocking(vmx);
6610 	vmx_complete_interrupts(vmx);
6611 }
6612 
6613 static struct kvm *vmx_vm_alloc(void)
6614 {
6615 	struct kvm_vmx *kvm_vmx = __vmalloc(sizeof(struct kvm_vmx),
6616 					    GFP_KERNEL_ACCOUNT | __GFP_ZERO,
6617 					    PAGE_KERNEL);
6618 	return &kvm_vmx->kvm;
6619 }
6620 
6621 static void vmx_vm_free(struct kvm *kvm)
6622 {
6623 	kfree(kvm->arch.hyperv.hv_pa_pg);
6624 	vfree(to_kvm_vmx(kvm));
6625 }
6626 
6627 static void vmx_free_vcpu(struct kvm_vcpu *vcpu)
6628 {
6629 	struct vcpu_vmx *vmx = to_vmx(vcpu);
6630 
6631 	if (enable_pml)
6632 		vmx_destroy_pml_buffer(vmx);
6633 	free_vpid(vmx->vpid);
6634 	nested_vmx_free_vcpu(vcpu);
6635 	free_loaded_vmcs(vmx->loaded_vmcs);
6636 	kvm_vcpu_uninit(vcpu);
6637 	kmem_cache_free(x86_fpu_cache, vmx->vcpu.arch.user_fpu);
6638 	kmem_cache_free(x86_fpu_cache, vmx->vcpu.arch.guest_fpu);
6639 	kmem_cache_free(kvm_vcpu_cache, vmx);
6640 }
6641 
6642 static struct kvm_vcpu *vmx_create_vcpu(struct kvm *kvm, unsigned int id)
6643 {
6644 	int err;
6645 	struct vcpu_vmx *vmx;
6646 	unsigned long *msr_bitmap;
6647 	int i, cpu;
6648 
6649 	BUILD_BUG_ON_MSG(offsetof(struct vcpu_vmx, vcpu) != 0,
6650 		"struct kvm_vcpu must be at offset 0 for arch usercopy region");
6651 
6652 	vmx = kmem_cache_zalloc(kvm_vcpu_cache, GFP_KERNEL_ACCOUNT);
6653 	if (!vmx)
6654 		return ERR_PTR(-ENOMEM);
6655 
6656 	vmx->vcpu.arch.user_fpu = kmem_cache_zalloc(x86_fpu_cache,
6657 			GFP_KERNEL_ACCOUNT);
6658 	if (!vmx->vcpu.arch.user_fpu) {
6659 		printk(KERN_ERR "kvm: failed to allocate kvm userspace's fpu\n");
6660 		err = -ENOMEM;
6661 		goto free_partial_vcpu;
6662 	}
6663 
6664 	vmx->vcpu.arch.guest_fpu = kmem_cache_zalloc(x86_fpu_cache,
6665 			GFP_KERNEL_ACCOUNT);
6666 	if (!vmx->vcpu.arch.guest_fpu) {
6667 		printk(KERN_ERR "kvm: failed to allocate vcpu's fpu\n");
6668 		err = -ENOMEM;
6669 		goto free_user_fpu;
6670 	}
6671 
6672 	vmx->vpid = allocate_vpid();
6673 
6674 	err = kvm_vcpu_init(&vmx->vcpu, kvm, id);
6675 	if (err)
6676 		goto free_vcpu;
6677 
6678 	err = -ENOMEM;
6679 
6680 	/*
6681 	 * If PML is turned on, failure on enabling PML just results in failure
6682 	 * of creating the vcpu, therefore we can simplify PML logic (by
6683 	 * avoiding dealing with cases, such as enabling PML partially on vcpus
6684 	 * for the guest, etc.
6685 	 */
6686 	if (enable_pml) {
6687 		vmx->pml_pg = alloc_page(GFP_KERNEL_ACCOUNT | __GFP_ZERO);
6688 		if (!vmx->pml_pg)
6689 			goto uninit_vcpu;
6690 	}
6691 
6692 	BUILD_BUG_ON(ARRAY_SIZE(vmx_msr_index) != NR_SHARED_MSRS);
6693 
6694 	for (i = 0; i < ARRAY_SIZE(vmx_msr_index); ++i) {
6695 		u32 index = vmx_msr_index[i];
6696 		u32 data_low, data_high;
6697 		int j = vmx->nmsrs;
6698 
6699 		if (rdmsr_safe(index, &data_low, &data_high) < 0)
6700 			continue;
6701 		if (wrmsr_safe(index, data_low, data_high) < 0)
6702 			continue;
6703 
6704 		vmx->guest_msrs[j].index = i;
6705 		vmx->guest_msrs[j].data = 0;
6706 		switch (index) {
6707 		case MSR_IA32_TSX_CTRL:
6708 			/*
6709 			 * No need to pass TSX_CTRL_CPUID_CLEAR through, so
6710 			 * let's avoid changing CPUID bits under the host
6711 			 * kernel's feet.
6712 			 */
6713 			vmx->guest_msrs[j].mask = ~(u64)TSX_CTRL_CPUID_CLEAR;
6714 			break;
6715 		default:
6716 			vmx->guest_msrs[j].mask = -1ull;
6717 			break;
6718 		}
6719 		++vmx->nmsrs;
6720 	}
6721 
6722 	err = alloc_loaded_vmcs(&vmx->vmcs01);
6723 	if (err < 0)
6724 		goto free_pml;
6725 
6726 	msr_bitmap = vmx->vmcs01.msr_bitmap;
6727 	vmx_disable_intercept_for_msr(msr_bitmap, MSR_IA32_TSC, MSR_TYPE_R);
6728 	vmx_disable_intercept_for_msr(msr_bitmap, MSR_FS_BASE, MSR_TYPE_RW);
6729 	vmx_disable_intercept_for_msr(msr_bitmap, MSR_GS_BASE, MSR_TYPE_RW);
6730 	vmx_disable_intercept_for_msr(msr_bitmap, MSR_KERNEL_GS_BASE, MSR_TYPE_RW);
6731 	vmx_disable_intercept_for_msr(msr_bitmap, MSR_IA32_SYSENTER_CS, MSR_TYPE_RW);
6732 	vmx_disable_intercept_for_msr(msr_bitmap, MSR_IA32_SYSENTER_ESP, MSR_TYPE_RW);
6733 	vmx_disable_intercept_for_msr(msr_bitmap, MSR_IA32_SYSENTER_EIP, MSR_TYPE_RW);
6734 	if (kvm_cstate_in_guest(kvm)) {
6735 		vmx_disable_intercept_for_msr(msr_bitmap, MSR_CORE_C1_RES, MSR_TYPE_R);
6736 		vmx_disable_intercept_for_msr(msr_bitmap, MSR_CORE_C3_RESIDENCY, MSR_TYPE_R);
6737 		vmx_disable_intercept_for_msr(msr_bitmap, MSR_CORE_C6_RESIDENCY, MSR_TYPE_R);
6738 		vmx_disable_intercept_for_msr(msr_bitmap, MSR_CORE_C7_RESIDENCY, MSR_TYPE_R);
6739 	}
6740 	vmx->msr_bitmap_mode = 0;
6741 
6742 	vmx->loaded_vmcs = &vmx->vmcs01;
6743 	cpu = get_cpu();
6744 	vmx_vcpu_load(&vmx->vcpu, cpu);
6745 	vmx->vcpu.cpu = cpu;
6746 	init_vmcs(vmx);
6747 	vmx_vcpu_put(&vmx->vcpu);
6748 	put_cpu();
6749 	if (cpu_need_virtualize_apic_accesses(&vmx->vcpu)) {
6750 		err = alloc_apic_access_page(kvm);
6751 		if (err)
6752 			goto free_vmcs;
6753 	}
6754 
6755 	if (enable_ept && !enable_unrestricted_guest) {
6756 		err = init_rmode_identity_map(kvm);
6757 		if (err)
6758 			goto free_vmcs;
6759 	}
6760 
6761 	if (nested)
6762 		nested_vmx_setup_ctls_msrs(&vmx->nested.msrs,
6763 					   vmx_capability.ept,
6764 					   kvm_vcpu_apicv_active(&vmx->vcpu));
6765 	else
6766 		memset(&vmx->nested.msrs, 0, sizeof(vmx->nested.msrs));
6767 
6768 	vmx->nested.posted_intr_nv = -1;
6769 	vmx->nested.current_vmptr = -1ull;
6770 
6771 	vmx->msr_ia32_feature_control_valid_bits = FEAT_CTL_LOCKED;
6772 
6773 	/*
6774 	 * Enforce invariant: pi_desc.nv is always either POSTED_INTR_VECTOR
6775 	 * or POSTED_INTR_WAKEUP_VECTOR.
6776 	 */
6777 	vmx->pi_desc.nv = POSTED_INTR_VECTOR;
6778 	vmx->pi_desc.sn = 1;
6779 
6780 	vmx->ept_pointer = INVALID_PAGE;
6781 
6782 	return &vmx->vcpu;
6783 
6784 free_vmcs:
6785 	free_loaded_vmcs(vmx->loaded_vmcs);
6786 free_pml:
6787 	vmx_destroy_pml_buffer(vmx);
6788 uninit_vcpu:
6789 	kvm_vcpu_uninit(&vmx->vcpu);
6790 free_vcpu:
6791 	free_vpid(vmx->vpid);
6792 	kmem_cache_free(x86_fpu_cache, vmx->vcpu.arch.guest_fpu);
6793 free_user_fpu:
6794 	kmem_cache_free(x86_fpu_cache, vmx->vcpu.arch.user_fpu);
6795 free_partial_vcpu:
6796 	kmem_cache_free(kvm_vcpu_cache, vmx);
6797 	return ERR_PTR(err);
6798 }
6799 
6800 #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"
6801 #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"
6802 
6803 static int vmx_vm_init(struct kvm *kvm)
6804 {
6805 	spin_lock_init(&to_kvm_vmx(kvm)->ept_pointer_lock);
6806 
6807 	if (!ple_gap)
6808 		kvm->arch.pause_in_guest = true;
6809 
6810 	if (boot_cpu_has(X86_BUG_L1TF) && enable_ept) {
6811 		switch (l1tf_mitigation) {
6812 		case L1TF_MITIGATION_OFF:
6813 		case L1TF_MITIGATION_FLUSH_NOWARN:
6814 			/* 'I explicitly don't care' is set */
6815 			break;
6816 		case L1TF_MITIGATION_FLUSH:
6817 		case L1TF_MITIGATION_FLUSH_NOSMT:
6818 		case L1TF_MITIGATION_FULL:
6819 			/*
6820 			 * Warn upon starting the first VM in a potentially
6821 			 * insecure environment.
6822 			 */
6823 			if (sched_smt_active())
6824 				pr_warn_once(L1TF_MSG_SMT);
6825 			if (l1tf_vmx_mitigation == VMENTER_L1D_FLUSH_NEVER)
6826 				pr_warn_once(L1TF_MSG_L1D);
6827 			break;
6828 		case L1TF_MITIGATION_FULL_FORCE:
6829 			/* Flush is enforced */
6830 			break;
6831 		}
6832 	}
6833 	return 0;
6834 }
6835 
6836 static int __init vmx_check_processor_compat(void)
6837 {
6838 	struct vmcs_config vmcs_conf;
6839 	struct vmx_capability vmx_cap;
6840 
6841 	if (!this_cpu_has(X86_FEATURE_MSR_IA32_FEAT_CTL) ||
6842 	    !this_cpu_has(X86_FEATURE_VMX)) {
6843 		pr_err("kvm: VMX is disabled on CPU %d\n", smp_processor_id());
6844 		return -EIO;
6845 	}
6846 
6847 	if (setup_vmcs_config(&vmcs_conf, &vmx_cap) < 0)
6848 		return -EIO;
6849 	if (nested)
6850 		nested_vmx_setup_ctls_msrs(&vmcs_conf.nested, vmx_cap.ept,
6851 					   enable_apicv);
6852 	if (memcmp(&vmcs_config, &vmcs_conf, sizeof(struct vmcs_config)) != 0) {
6853 		printk(KERN_ERR "kvm: CPU %d feature inconsistency!\n",
6854 				smp_processor_id());
6855 		return -EIO;
6856 	}
6857 	return 0;
6858 }
6859 
6860 static u64 vmx_get_mt_mask(struct kvm_vcpu *vcpu, gfn_t gfn, bool is_mmio)
6861 {
6862 	u8 cache;
6863 	u64 ipat = 0;
6864 
6865 	/* For VT-d and EPT combination
6866 	 * 1. MMIO: always map as UC
6867 	 * 2. EPT with VT-d:
6868 	 *   a. VT-d without snooping control feature: can't guarantee the
6869 	 *	result, try to trust guest.
6870 	 *   b. VT-d with snooping control feature: snooping control feature of
6871 	 *	VT-d engine can guarantee the cache correctness. Just set it
6872 	 *	to WB to keep consistent with host. So the same as item 3.
6873 	 * 3. EPT without VT-d: always map as WB and set IPAT=1 to keep
6874 	 *    consistent with host MTRR
6875 	 */
6876 	if (is_mmio) {
6877 		cache = MTRR_TYPE_UNCACHABLE;
6878 		goto exit;
6879 	}
6880 
6881 	if (!kvm_arch_has_noncoherent_dma(vcpu->kvm)) {
6882 		ipat = VMX_EPT_IPAT_BIT;
6883 		cache = MTRR_TYPE_WRBACK;
6884 		goto exit;
6885 	}
6886 
6887 	if (kvm_read_cr0(vcpu) & X86_CR0_CD) {
6888 		ipat = VMX_EPT_IPAT_BIT;
6889 		if (kvm_check_has_quirk(vcpu->kvm, KVM_X86_QUIRK_CD_NW_CLEARED))
6890 			cache = MTRR_TYPE_WRBACK;
6891 		else
6892 			cache = MTRR_TYPE_UNCACHABLE;
6893 		goto exit;
6894 	}
6895 
6896 	cache = kvm_mtrr_get_guest_memory_type(vcpu, gfn);
6897 
6898 exit:
6899 	return (cache << VMX_EPT_MT_EPTE_SHIFT) | ipat;
6900 }
6901 
6902 static int vmx_get_lpage_level(void)
6903 {
6904 	if (enable_ept && !cpu_has_vmx_ept_1g_page())
6905 		return PT_DIRECTORY_LEVEL;
6906 	else
6907 		/* For shadow and EPT supported 1GB page */
6908 		return PT_PDPE_LEVEL;
6909 }
6910 
6911 static void vmcs_set_secondary_exec_control(struct vcpu_vmx *vmx)
6912 {
6913 	/*
6914 	 * These bits in the secondary execution controls field
6915 	 * are dynamic, the others are mostly based on the hypervisor
6916 	 * architecture and the guest's CPUID.  Do not touch the
6917 	 * dynamic bits.
6918 	 */
6919 	u32 mask =
6920 		SECONDARY_EXEC_SHADOW_VMCS |
6921 		SECONDARY_EXEC_VIRTUALIZE_X2APIC_MODE |
6922 		SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES |
6923 		SECONDARY_EXEC_DESC;
6924 
6925 	u32 new_ctl = vmx->secondary_exec_control;
6926 	u32 cur_ctl = secondary_exec_controls_get(vmx);
6927 
6928 	secondary_exec_controls_set(vmx, (new_ctl & ~mask) | (cur_ctl & mask));
6929 }
6930 
6931 /*
6932  * Generate MSR_IA32_VMX_CR{0,4}_FIXED1 according to CPUID. Only set bits
6933  * (indicating "allowed-1") if they are supported in the guest's CPUID.
6934  */
6935 static void nested_vmx_cr_fixed1_bits_update(struct kvm_vcpu *vcpu)
6936 {
6937 	struct vcpu_vmx *vmx = to_vmx(vcpu);
6938 	struct kvm_cpuid_entry2 *entry;
6939 
6940 	vmx->nested.msrs.cr0_fixed1 = 0xffffffff;
6941 	vmx->nested.msrs.cr4_fixed1 = X86_CR4_PCE;
6942 
6943 #define cr4_fixed1_update(_cr4_mask, _reg, _cpuid_mask) do {		\
6944 	if (entry && (entry->_reg & (_cpuid_mask)))			\
6945 		vmx->nested.msrs.cr4_fixed1 |= (_cr4_mask);	\
6946 } while (0)
6947 
6948 	entry = kvm_find_cpuid_entry(vcpu, 0x1, 0);
6949 	cr4_fixed1_update(X86_CR4_VME,        edx, bit(X86_FEATURE_VME));
6950 	cr4_fixed1_update(X86_CR4_PVI,        edx, bit(X86_FEATURE_VME));
6951 	cr4_fixed1_update(X86_CR4_TSD,        edx, bit(X86_FEATURE_TSC));
6952 	cr4_fixed1_update(X86_CR4_DE,         edx, bit(X86_FEATURE_DE));
6953 	cr4_fixed1_update(X86_CR4_PSE,        edx, bit(X86_FEATURE_PSE));
6954 	cr4_fixed1_update(X86_CR4_PAE,        edx, bit(X86_FEATURE_PAE));
6955 	cr4_fixed1_update(X86_CR4_MCE,        edx, bit(X86_FEATURE_MCE));
6956 	cr4_fixed1_update(X86_CR4_PGE,        edx, bit(X86_FEATURE_PGE));
6957 	cr4_fixed1_update(X86_CR4_OSFXSR,     edx, bit(X86_FEATURE_FXSR));
6958 	cr4_fixed1_update(X86_CR4_OSXMMEXCPT, edx, bit(X86_FEATURE_XMM));
6959 	cr4_fixed1_update(X86_CR4_VMXE,       ecx, bit(X86_FEATURE_VMX));
6960 	cr4_fixed1_update(X86_CR4_SMXE,       ecx, bit(X86_FEATURE_SMX));
6961 	cr4_fixed1_update(X86_CR4_PCIDE,      ecx, bit(X86_FEATURE_PCID));
6962 	cr4_fixed1_update(X86_CR4_OSXSAVE,    ecx, bit(X86_FEATURE_XSAVE));
6963 
6964 	entry = kvm_find_cpuid_entry(vcpu, 0x7, 0);
6965 	cr4_fixed1_update(X86_CR4_FSGSBASE,   ebx, bit(X86_FEATURE_FSGSBASE));
6966 	cr4_fixed1_update(X86_CR4_SMEP,       ebx, bit(X86_FEATURE_SMEP));
6967 	cr4_fixed1_update(X86_CR4_SMAP,       ebx, bit(X86_FEATURE_SMAP));
6968 	cr4_fixed1_update(X86_CR4_PKE,        ecx, bit(X86_FEATURE_PKU));
6969 	cr4_fixed1_update(X86_CR4_UMIP,       ecx, bit(X86_FEATURE_UMIP));
6970 	cr4_fixed1_update(X86_CR4_LA57,       ecx, bit(X86_FEATURE_LA57));
6971 
6972 #undef cr4_fixed1_update
6973 }
6974 
6975 static void nested_vmx_entry_exit_ctls_update(struct kvm_vcpu *vcpu)
6976 {
6977 	struct vcpu_vmx *vmx = to_vmx(vcpu);
6978 
6979 	if (kvm_mpx_supported()) {
6980 		bool mpx_enabled = guest_cpuid_has(vcpu, X86_FEATURE_MPX);
6981 
6982 		if (mpx_enabled) {
6983 			vmx->nested.msrs.entry_ctls_high |= VM_ENTRY_LOAD_BNDCFGS;
6984 			vmx->nested.msrs.exit_ctls_high |= VM_EXIT_CLEAR_BNDCFGS;
6985 		} else {
6986 			vmx->nested.msrs.entry_ctls_high &= ~VM_ENTRY_LOAD_BNDCFGS;
6987 			vmx->nested.msrs.exit_ctls_high &= ~VM_EXIT_CLEAR_BNDCFGS;
6988 		}
6989 	}
6990 }
6991 
6992 static void update_intel_pt_cfg(struct kvm_vcpu *vcpu)
6993 {
6994 	struct vcpu_vmx *vmx = to_vmx(vcpu);
6995 	struct kvm_cpuid_entry2 *best = NULL;
6996 	int i;
6997 
6998 	for (i = 0; i < PT_CPUID_LEAVES; i++) {
6999 		best = kvm_find_cpuid_entry(vcpu, 0x14, i);
7000 		if (!best)
7001 			return;
7002 		vmx->pt_desc.caps[CPUID_EAX + i*PT_CPUID_REGS_NUM] = best->eax;
7003 		vmx->pt_desc.caps[CPUID_EBX + i*PT_CPUID_REGS_NUM] = best->ebx;
7004 		vmx->pt_desc.caps[CPUID_ECX + i*PT_CPUID_REGS_NUM] = best->ecx;
7005 		vmx->pt_desc.caps[CPUID_EDX + i*PT_CPUID_REGS_NUM] = best->edx;
7006 	}
7007 
7008 	/* Get the number of configurable Address Ranges for filtering */
7009 	vmx->pt_desc.addr_range = intel_pt_validate_cap(vmx->pt_desc.caps,
7010 						PT_CAP_num_address_ranges);
7011 
7012 	/* Initialize and clear the no dependency bits */
7013 	vmx->pt_desc.ctl_bitmask = ~(RTIT_CTL_TRACEEN | RTIT_CTL_OS |
7014 			RTIT_CTL_USR | RTIT_CTL_TSC_EN | RTIT_CTL_DISRETC);
7015 
7016 	/*
7017 	 * If CPUID.(EAX=14H,ECX=0):EBX[0]=1 CR3Filter can be set otherwise
7018 	 * will inject an #GP
7019 	 */
7020 	if (intel_pt_validate_cap(vmx->pt_desc.caps, PT_CAP_cr3_filtering))
7021 		vmx->pt_desc.ctl_bitmask &= ~RTIT_CTL_CR3EN;
7022 
7023 	/*
7024 	 * If CPUID.(EAX=14H,ECX=0):EBX[1]=1 CYCEn, CycThresh and
7025 	 * PSBFreq can be set
7026 	 */
7027 	if (intel_pt_validate_cap(vmx->pt_desc.caps, PT_CAP_psb_cyc))
7028 		vmx->pt_desc.ctl_bitmask &= ~(RTIT_CTL_CYCLEACC |
7029 				RTIT_CTL_CYC_THRESH | RTIT_CTL_PSB_FREQ);
7030 
7031 	/*
7032 	 * If CPUID.(EAX=14H,ECX=0):EBX[3]=1 MTCEn BranchEn and
7033 	 * MTCFreq can be set
7034 	 */
7035 	if (intel_pt_validate_cap(vmx->pt_desc.caps, PT_CAP_mtc))
7036 		vmx->pt_desc.ctl_bitmask &= ~(RTIT_CTL_MTC_EN |
7037 				RTIT_CTL_BRANCH_EN | RTIT_CTL_MTC_RANGE);
7038 
7039 	/* If CPUID.(EAX=14H,ECX=0):EBX[4]=1 FUPonPTW and PTWEn can be set */
7040 	if (intel_pt_validate_cap(vmx->pt_desc.caps, PT_CAP_ptwrite))
7041 		vmx->pt_desc.ctl_bitmask &= ~(RTIT_CTL_FUP_ON_PTW |
7042 							RTIT_CTL_PTW_EN);
7043 
7044 	/* If CPUID.(EAX=14H,ECX=0):EBX[5]=1 PwrEvEn can be set */
7045 	if (intel_pt_validate_cap(vmx->pt_desc.caps, PT_CAP_power_event_trace))
7046 		vmx->pt_desc.ctl_bitmask &= ~RTIT_CTL_PWR_EVT_EN;
7047 
7048 	/* If CPUID.(EAX=14H,ECX=0):ECX[0]=1 ToPA can be set */
7049 	if (intel_pt_validate_cap(vmx->pt_desc.caps, PT_CAP_topa_output))
7050 		vmx->pt_desc.ctl_bitmask &= ~RTIT_CTL_TOPA;
7051 
7052 	/* If CPUID.(EAX=14H,ECX=0):ECX[3]=1 FabircEn can be set */
7053 	if (intel_pt_validate_cap(vmx->pt_desc.caps, PT_CAP_output_subsys))
7054 		vmx->pt_desc.ctl_bitmask &= ~RTIT_CTL_FABRIC_EN;
7055 
7056 	/* unmask address range configure area */
7057 	for (i = 0; i < vmx->pt_desc.addr_range; i++)
7058 		vmx->pt_desc.ctl_bitmask &= ~(0xfULL << (32 + i * 4));
7059 }
7060 
7061 static void vmx_cpuid_update(struct kvm_vcpu *vcpu)
7062 {
7063 	struct vcpu_vmx *vmx = to_vmx(vcpu);
7064 
7065 	/* xsaves_enabled is recomputed in vmx_compute_secondary_exec_control(). */
7066 	vcpu->arch.xsaves_enabled = false;
7067 
7068 	if (cpu_has_secondary_exec_ctrls()) {
7069 		vmx_compute_secondary_exec_control(vmx);
7070 		vmcs_set_secondary_exec_control(vmx);
7071 	}
7072 
7073 	if (nested_vmx_allowed(vcpu))
7074 		to_vmx(vcpu)->msr_ia32_feature_control_valid_bits |=
7075 			FEAT_CTL_VMX_ENABLED_INSIDE_SMX |
7076 			FEAT_CTL_VMX_ENABLED_OUTSIDE_SMX;
7077 	else
7078 		to_vmx(vcpu)->msr_ia32_feature_control_valid_bits &=
7079 			~(FEAT_CTL_VMX_ENABLED_INSIDE_SMX |
7080 			  FEAT_CTL_VMX_ENABLED_OUTSIDE_SMX);
7081 
7082 	if (nested_vmx_allowed(vcpu)) {
7083 		nested_vmx_cr_fixed1_bits_update(vcpu);
7084 		nested_vmx_entry_exit_ctls_update(vcpu);
7085 	}
7086 
7087 	if (boot_cpu_has(X86_FEATURE_INTEL_PT) &&
7088 			guest_cpuid_has(vcpu, X86_FEATURE_INTEL_PT))
7089 		update_intel_pt_cfg(vcpu);
7090 
7091 	if (boot_cpu_has(X86_FEATURE_RTM)) {
7092 		struct shared_msr_entry *msr;
7093 		msr = find_msr_entry(vmx, MSR_IA32_TSX_CTRL);
7094 		if (msr) {
7095 			bool enabled = guest_cpuid_has(vcpu, X86_FEATURE_RTM);
7096 			vmx_set_guest_msr(vmx, msr, enabled ? 0 : TSX_CTRL_RTM_DISABLE);
7097 		}
7098 	}
7099 }
7100 
7101 static void vmx_set_supported_cpuid(u32 func, struct kvm_cpuid_entry2 *entry)
7102 {
7103 	if (func == 1 && nested)
7104 		entry->ecx |= bit(X86_FEATURE_VMX);
7105 }
7106 
7107 static void vmx_request_immediate_exit(struct kvm_vcpu *vcpu)
7108 {
7109 	to_vmx(vcpu)->req_immediate_exit = true;
7110 }
7111 
7112 static int vmx_check_intercept(struct kvm_vcpu *vcpu,
7113 			       struct x86_instruction_info *info,
7114 			       enum x86_intercept_stage stage)
7115 {
7116 	struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
7117 	struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
7118 
7119 	/*
7120 	 * RDPID causes #UD if disabled through secondary execution controls.
7121 	 * Because it is marked as EmulateOnUD, we need to intercept it here.
7122 	 */
7123 	if (info->intercept == x86_intercept_rdtscp &&
7124 	    !nested_cpu_has2(vmcs12, SECONDARY_EXEC_RDTSCP)) {
7125 		ctxt->exception.vector = UD_VECTOR;
7126 		ctxt->exception.error_code_valid = false;
7127 		return X86EMUL_PROPAGATE_FAULT;
7128 	}
7129 
7130 	/* TODO: check more intercepts... */
7131 	return X86EMUL_CONTINUE;
7132 }
7133 
7134 #ifdef CONFIG_X86_64
7135 /* (a << shift) / divisor, return 1 if overflow otherwise 0 */
7136 static inline int u64_shl_div_u64(u64 a, unsigned int shift,
7137 				  u64 divisor, u64 *result)
7138 {
7139 	u64 low = a << shift, high = a >> (64 - shift);
7140 
7141 	/* To avoid the overflow on divq */
7142 	if (high >= divisor)
7143 		return 1;
7144 
7145 	/* Low hold the result, high hold rem which is discarded */
7146 	asm("divq %2\n\t" : "=a" (low), "=d" (high) :
7147 	    "rm" (divisor), "0" (low), "1" (high));
7148 	*result = low;
7149 
7150 	return 0;
7151 }
7152 
7153 static int vmx_set_hv_timer(struct kvm_vcpu *vcpu, u64 guest_deadline_tsc,
7154 			    bool *expired)
7155 {
7156 	struct vcpu_vmx *vmx;
7157 	u64 tscl, guest_tscl, delta_tsc, lapic_timer_advance_cycles;
7158 	struct kvm_timer *ktimer = &vcpu->arch.apic->lapic_timer;
7159 
7160 	if (kvm_mwait_in_guest(vcpu->kvm) ||
7161 		kvm_can_post_timer_interrupt(vcpu))
7162 		return -EOPNOTSUPP;
7163 
7164 	vmx = to_vmx(vcpu);
7165 	tscl = rdtsc();
7166 	guest_tscl = kvm_read_l1_tsc(vcpu, tscl);
7167 	delta_tsc = max(guest_deadline_tsc, guest_tscl) - guest_tscl;
7168 	lapic_timer_advance_cycles = nsec_to_cycles(vcpu,
7169 						    ktimer->timer_advance_ns);
7170 
7171 	if (delta_tsc > lapic_timer_advance_cycles)
7172 		delta_tsc -= lapic_timer_advance_cycles;
7173 	else
7174 		delta_tsc = 0;
7175 
7176 	/* Convert to host delta tsc if tsc scaling is enabled */
7177 	if (vcpu->arch.tsc_scaling_ratio != kvm_default_tsc_scaling_ratio &&
7178 	    delta_tsc && u64_shl_div_u64(delta_tsc,
7179 				kvm_tsc_scaling_ratio_frac_bits,
7180 				vcpu->arch.tsc_scaling_ratio, &delta_tsc))
7181 		return -ERANGE;
7182 
7183 	/*
7184 	 * If the delta tsc can't fit in the 32 bit after the multi shift,
7185 	 * we can't use the preemption timer.
7186 	 * It's possible that it fits on later vmentries, but checking
7187 	 * on every vmentry is costly so we just use an hrtimer.
7188 	 */
7189 	if (delta_tsc >> (cpu_preemption_timer_multi + 32))
7190 		return -ERANGE;
7191 
7192 	vmx->hv_deadline_tsc = tscl + delta_tsc;
7193 	*expired = !delta_tsc;
7194 	return 0;
7195 }
7196 
7197 static void vmx_cancel_hv_timer(struct kvm_vcpu *vcpu)
7198 {
7199 	to_vmx(vcpu)->hv_deadline_tsc = -1;
7200 }
7201 #endif
7202 
7203 static void vmx_sched_in(struct kvm_vcpu *vcpu, int cpu)
7204 {
7205 	if (!kvm_pause_in_guest(vcpu->kvm))
7206 		shrink_ple_window(vcpu);
7207 }
7208 
7209 static void vmx_slot_enable_log_dirty(struct kvm *kvm,
7210 				     struct kvm_memory_slot *slot)
7211 {
7212 	kvm_mmu_slot_leaf_clear_dirty(kvm, slot);
7213 	kvm_mmu_slot_largepage_remove_write_access(kvm, slot);
7214 }
7215 
7216 static void vmx_slot_disable_log_dirty(struct kvm *kvm,
7217 				       struct kvm_memory_slot *slot)
7218 {
7219 	kvm_mmu_slot_set_dirty(kvm, slot);
7220 }
7221 
7222 static void vmx_flush_log_dirty(struct kvm *kvm)
7223 {
7224 	kvm_flush_pml_buffers(kvm);
7225 }
7226 
7227 static int vmx_write_pml_buffer(struct kvm_vcpu *vcpu)
7228 {
7229 	struct vmcs12 *vmcs12;
7230 	struct vcpu_vmx *vmx = to_vmx(vcpu);
7231 	gpa_t gpa, dst;
7232 
7233 	if (is_guest_mode(vcpu)) {
7234 		WARN_ON_ONCE(vmx->nested.pml_full);
7235 
7236 		/*
7237 		 * Check if PML is enabled for the nested guest.
7238 		 * Whether eptp bit 6 is set is already checked
7239 		 * as part of A/D emulation.
7240 		 */
7241 		vmcs12 = get_vmcs12(vcpu);
7242 		if (!nested_cpu_has_pml(vmcs12))
7243 			return 0;
7244 
7245 		if (vmcs12->guest_pml_index >= PML_ENTITY_NUM) {
7246 			vmx->nested.pml_full = true;
7247 			return 1;
7248 		}
7249 
7250 		gpa = vmcs_read64(GUEST_PHYSICAL_ADDRESS) & ~0xFFFull;
7251 		dst = vmcs12->pml_address + sizeof(u64) * vmcs12->guest_pml_index;
7252 
7253 		if (kvm_write_guest_page(vcpu->kvm, gpa_to_gfn(dst), &gpa,
7254 					 offset_in_page(dst), sizeof(gpa)))
7255 			return 0;
7256 
7257 		vmcs12->guest_pml_index--;
7258 	}
7259 
7260 	return 0;
7261 }
7262 
7263 static void vmx_enable_log_dirty_pt_masked(struct kvm *kvm,
7264 					   struct kvm_memory_slot *memslot,
7265 					   gfn_t offset, unsigned long mask)
7266 {
7267 	kvm_mmu_clear_dirty_pt_masked(kvm, memslot, offset, mask);
7268 }
7269 
7270 static void __pi_post_block(struct kvm_vcpu *vcpu)
7271 {
7272 	struct pi_desc *pi_desc = vcpu_to_pi_desc(vcpu);
7273 	struct pi_desc old, new;
7274 	unsigned int dest;
7275 
7276 	do {
7277 		old.control = new.control = pi_desc->control;
7278 		WARN(old.nv != POSTED_INTR_WAKEUP_VECTOR,
7279 		     "Wakeup handler not enabled while the VCPU is blocked\n");
7280 
7281 		dest = cpu_physical_id(vcpu->cpu);
7282 
7283 		if (x2apic_enabled())
7284 			new.ndst = dest;
7285 		else
7286 			new.ndst = (dest << 8) & 0xFF00;
7287 
7288 		/* set 'NV' to 'notification vector' */
7289 		new.nv = POSTED_INTR_VECTOR;
7290 	} while (cmpxchg64(&pi_desc->control, old.control,
7291 			   new.control) != old.control);
7292 
7293 	if (!WARN_ON_ONCE(vcpu->pre_pcpu == -1)) {
7294 		spin_lock(&per_cpu(blocked_vcpu_on_cpu_lock, vcpu->pre_pcpu));
7295 		list_del(&vcpu->blocked_vcpu_list);
7296 		spin_unlock(&per_cpu(blocked_vcpu_on_cpu_lock, vcpu->pre_pcpu));
7297 		vcpu->pre_pcpu = -1;
7298 	}
7299 }
7300 
7301 /*
7302  * This routine does the following things for vCPU which is going
7303  * to be blocked if VT-d PI is enabled.
7304  * - Store the vCPU to the wakeup list, so when interrupts happen
7305  *   we can find the right vCPU to wake up.
7306  * - Change the Posted-interrupt descriptor as below:
7307  *      'NDST' <-- vcpu->pre_pcpu
7308  *      'NV' <-- POSTED_INTR_WAKEUP_VECTOR
7309  * - If 'ON' is set during this process, which means at least one
7310  *   interrupt is posted for this vCPU, we cannot block it, in
7311  *   this case, return 1, otherwise, return 0.
7312  *
7313  */
7314 static int pi_pre_block(struct kvm_vcpu *vcpu)
7315 {
7316 	unsigned int dest;
7317 	struct pi_desc old, new;
7318 	struct pi_desc *pi_desc = vcpu_to_pi_desc(vcpu);
7319 
7320 	if (!kvm_arch_has_assigned_device(vcpu->kvm) ||
7321 		!irq_remapping_cap(IRQ_POSTING_CAP)  ||
7322 		!kvm_vcpu_apicv_active(vcpu))
7323 		return 0;
7324 
7325 	WARN_ON(irqs_disabled());
7326 	local_irq_disable();
7327 	if (!WARN_ON_ONCE(vcpu->pre_pcpu != -1)) {
7328 		vcpu->pre_pcpu = vcpu->cpu;
7329 		spin_lock(&per_cpu(blocked_vcpu_on_cpu_lock, vcpu->pre_pcpu));
7330 		list_add_tail(&vcpu->blocked_vcpu_list,
7331 			      &per_cpu(blocked_vcpu_on_cpu,
7332 				       vcpu->pre_pcpu));
7333 		spin_unlock(&per_cpu(blocked_vcpu_on_cpu_lock, vcpu->pre_pcpu));
7334 	}
7335 
7336 	do {
7337 		old.control = new.control = pi_desc->control;
7338 
7339 		WARN((pi_desc->sn == 1),
7340 		     "Warning: SN field of posted-interrupts "
7341 		     "is set before blocking\n");
7342 
7343 		/*
7344 		 * Since vCPU can be preempted during this process,
7345 		 * vcpu->cpu could be different with pre_pcpu, we
7346 		 * need to set pre_pcpu as the destination of wakeup
7347 		 * notification event, then we can find the right vCPU
7348 		 * to wakeup in wakeup handler if interrupts happen
7349 		 * when the vCPU is in blocked state.
7350 		 */
7351 		dest = cpu_physical_id(vcpu->pre_pcpu);
7352 
7353 		if (x2apic_enabled())
7354 			new.ndst = dest;
7355 		else
7356 			new.ndst = (dest << 8) & 0xFF00;
7357 
7358 		/* set 'NV' to 'wakeup vector' */
7359 		new.nv = POSTED_INTR_WAKEUP_VECTOR;
7360 	} while (cmpxchg64(&pi_desc->control, old.control,
7361 			   new.control) != old.control);
7362 
7363 	/* We should not block the vCPU if an interrupt is posted for it.  */
7364 	if (pi_test_on(pi_desc) == 1)
7365 		__pi_post_block(vcpu);
7366 
7367 	local_irq_enable();
7368 	return (vcpu->pre_pcpu == -1);
7369 }
7370 
7371 static int vmx_pre_block(struct kvm_vcpu *vcpu)
7372 {
7373 	if (pi_pre_block(vcpu))
7374 		return 1;
7375 
7376 	if (kvm_lapic_hv_timer_in_use(vcpu))
7377 		kvm_lapic_switch_to_sw_timer(vcpu);
7378 
7379 	return 0;
7380 }
7381 
7382 static void pi_post_block(struct kvm_vcpu *vcpu)
7383 {
7384 	if (vcpu->pre_pcpu == -1)
7385 		return;
7386 
7387 	WARN_ON(irqs_disabled());
7388 	local_irq_disable();
7389 	__pi_post_block(vcpu);
7390 	local_irq_enable();
7391 }
7392 
7393 static void vmx_post_block(struct kvm_vcpu *vcpu)
7394 {
7395 	if (kvm_x86_ops->set_hv_timer)
7396 		kvm_lapic_switch_to_hv_timer(vcpu);
7397 
7398 	pi_post_block(vcpu);
7399 }
7400 
7401 /*
7402  * vmx_update_pi_irte - set IRTE for Posted-Interrupts
7403  *
7404  * @kvm: kvm
7405  * @host_irq: host irq of the interrupt
7406  * @guest_irq: gsi of the interrupt
7407  * @set: set or unset PI
7408  * returns 0 on success, < 0 on failure
7409  */
7410 static int vmx_update_pi_irte(struct kvm *kvm, unsigned int host_irq,
7411 			      uint32_t guest_irq, bool set)
7412 {
7413 	struct kvm_kernel_irq_routing_entry *e;
7414 	struct kvm_irq_routing_table *irq_rt;
7415 	struct kvm_lapic_irq irq;
7416 	struct kvm_vcpu *vcpu;
7417 	struct vcpu_data vcpu_info;
7418 	int idx, ret = 0;
7419 
7420 	if (!kvm_arch_has_assigned_device(kvm) ||
7421 		!irq_remapping_cap(IRQ_POSTING_CAP) ||
7422 		!kvm_vcpu_apicv_active(kvm->vcpus[0]))
7423 		return 0;
7424 
7425 	idx = srcu_read_lock(&kvm->irq_srcu);
7426 	irq_rt = srcu_dereference(kvm->irq_routing, &kvm->irq_srcu);
7427 	if (guest_irq >= irq_rt->nr_rt_entries ||
7428 	    hlist_empty(&irq_rt->map[guest_irq])) {
7429 		pr_warn_once("no route for guest_irq %u/%u (broken user space?)\n",
7430 			     guest_irq, irq_rt->nr_rt_entries);
7431 		goto out;
7432 	}
7433 
7434 	hlist_for_each_entry(e, &irq_rt->map[guest_irq], link) {
7435 		if (e->type != KVM_IRQ_ROUTING_MSI)
7436 			continue;
7437 		/*
7438 		 * VT-d PI cannot support posting multicast/broadcast
7439 		 * interrupts to a vCPU, we still use interrupt remapping
7440 		 * for these kind of interrupts.
7441 		 *
7442 		 * For lowest-priority interrupts, we only support
7443 		 * those with single CPU as the destination, e.g. user
7444 		 * configures the interrupts via /proc/irq or uses
7445 		 * irqbalance to make the interrupts single-CPU.
7446 		 *
7447 		 * We will support full lowest-priority interrupt later.
7448 		 *
7449 		 * In addition, we can only inject generic interrupts using
7450 		 * the PI mechanism, refuse to route others through it.
7451 		 */
7452 
7453 		kvm_set_msi_irq(kvm, e, &irq);
7454 		if (!kvm_intr_is_single_vcpu(kvm, &irq, &vcpu) ||
7455 		    !kvm_irq_is_postable(&irq)) {
7456 			/*
7457 			 * Make sure the IRTE is in remapped mode if
7458 			 * we don't handle it in posted mode.
7459 			 */
7460 			ret = irq_set_vcpu_affinity(host_irq, NULL);
7461 			if (ret < 0) {
7462 				printk(KERN_INFO
7463 				   "failed to back to remapped mode, irq: %u\n",
7464 				   host_irq);
7465 				goto out;
7466 			}
7467 
7468 			continue;
7469 		}
7470 
7471 		vcpu_info.pi_desc_addr = __pa(vcpu_to_pi_desc(vcpu));
7472 		vcpu_info.vector = irq.vector;
7473 
7474 		trace_kvm_pi_irte_update(host_irq, vcpu->vcpu_id, e->gsi,
7475 				vcpu_info.vector, vcpu_info.pi_desc_addr, set);
7476 
7477 		if (set)
7478 			ret = irq_set_vcpu_affinity(host_irq, &vcpu_info);
7479 		else
7480 			ret = irq_set_vcpu_affinity(host_irq, NULL);
7481 
7482 		if (ret < 0) {
7483 			printk(KERN_INFO "%s: failed to update PI IRTE\n",
7484 					__func__);
7485 			goto out;
7486 		}
7487 	}
7488 
7489 	ret = 0;
7490 out:
7491 	srcu_read_unlock(&kvm->irq_srcu, idx);
7492 	return ret;
7493 }
7494 
7495 static void vmx_setup_mce(struct kvm_vcpu *vcpu)
7496 {
7497 	if (vcpu->arch.mcg_cap & MCG_LMCE_P)
7498 		to_vmx(vcpu)->msr_ia32_feature_control_valid_bits |=
7499 			FEAT_CTL_LMCE_ENABLED;
7500 	else
7501 		to_vmx(vcpu)->msr_ia32_feature_control_valid_bits &=
7502 			~FEAT_CTL_LMCE_ENABLED;
7503 }
7504 
7505 static int vmx_smi_allowed(struct kvm_vcpu *vcpu)
7506 {
7507 	/* we need a nested vmexit to enter SMM, postpone if run is pending */
7508 	if (to_vmx(vcpu)->nested.nested_run_pending)
7509 		return 0;
7510 	return 1;
7511 }
7512 
7513 static int vmx_pre_enter_smm(struct kvm_vcpu *vcpu, char *smstate)
7514 {
7515 	struct vcpu_vmx *vmx = to_vmx(vcpu);
7516 
7517 	vmx->nested.smm.guest_mode = is_guest_mode(vcpu);
7518 	if (vmx->nested.smm.guest_mode)
7519 		nested_vmx_vmexit(vcpu, -1, 0, 0);
7520 
7521 	vmx->nested.smm.vmxon = vmx->nested.vmxon;
7522 	vmx->nested.vmxon = false;
7523 	vmx_clear_hlt(vcpu);
7524 	return 0;
7525 }
7526 
7527 static int vmx_pre_leave_smm(struct kvm_vcpu *vcpu, const char *smstate)
7528 {
7529 	struct vcpu_vmx *vmx = to_vmx(vcpu);
7530 	int ret;
7531 
7532 	if (vmx->nested.smm.vmxon) {
7533 		vmx->nested.vmxon = true;
7534 		vmx->nested.smm.vmxon = false;
7535 	}
7536 
7537 	if (vmx->nested.smm.guest_mode) {
7538 		ret = nested_vmx_enter_non_root_mode(vcpu, false);
7539 		if (ret)
7540 			return ret;
7541 
7542 		vmx->nested.smm.guest_mode = false;
7543 	}
7544 	return 0;
7545 }
7546 
7547 static int enable_smi_window(struct kvm_vcpu *vcpu)
7548 {
7549 	return 0;
7550 }
7551 
7552 static bool vmx_need_emulation_on_page_fault(struct kvm_vcpu *vcpu)
7553 {
7554 	return false;
7555 }
7556 
7557 static bool vmx_apic_init_signal_blocked(struct kvm_vcpu *vcpu)
7558 {
7559 	return to_vmx(vcpu)->nested.vmxon;
7560 }
7561 
7562 static __init int hardware_setup(void)
7563 {
7564 	unsigned long host_bndcfgs;
7565 	struct desc_ptr dt;
7566 	int r, i;
7567 
7568 	rdmsrl_safe(MSR_EFER, &host_efer);
7569 
7570 	store_idt(&dt);
7571 	host_idt_base = dt.address;
7572 
7573 	for (i = 0; i < ARRAY_SIZE(vmx_msr_index); ++i)
7574 		kvm_define_shared_msr(i, vmx_msr_index[i]);
7575 
7576 	if (setup_vmcs_config(&vmcs_config, &vmx_capability) < 0)
7577 		return -EIO;
7578 
7579 	if (boot_cpu_has(X86_FEATURE_NX))
7580 		kvm_enable_efer_bits(EFER_NX);
7581 
7582 	if (boot_cpu_has(X86_FEATURE_MPX)) {
7583 		rdmsrl(MSR_IA32_BNDCFGS, host_bndcfgs);
7584 		WARN_ONCE(host_bndcfgs, "KVM: BNDCFGS in host will be lost");
7585 	}
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_cr4_guest_bits = vmx_decache_cr4_guest_bits,
7768 	.set_cr0 = vmx_set_cr0,
7769 	.set_cr3 = vmx_set_cr3,
7770 	.set_cr4 = vmx_set_cr4,
7771 	.set_efer = vmx_set_efer,
7772 	.get_idt = vmx_get_idt,
7773 	.set_idt = vmx_set_idt,
7774 	.get_gdt = vmx_get_gdt,
7775 	.set_gdt = vmx_set_gdt,
7776 	.get_dr6 = vmx_get_dr6,
7777 	.set_dr6 = vmx_set_dr6,
7778 	.set_dr7 = vmx_set_dr7,
7779 	.sync_dirty_debug_regs = vmx_sync_dirty_debug_regs,
7780 	.cache_reg = vmx_cache_reg,
7781 	.get_rflags = vmx_get_rflags,
7782 	.set_rflags = vmx_set_rflags,
7783 
7784 	.tlb_flush = vmx_flush_tlb,
7785 	.tlb_flush_gva = vmx_flush_tlb_gva,
7786 
7787 	.run = vmx_vcpu_run,
7788 	.handle_exit = vmx_handle_exit,
7789 	.skip_emulated_instruction = skip_emulated_instruction,
7790 	.set_interrupt_shadow = vmx_set_interrupt_shadow,
7791 	.get_interrupt_shadow = vmx_get_interrupt_shadow,
7792 	.patch_hypercall = vmx_patch_hypercall,
7793 	.set_irq = vmx_inject_irq,
7794 	.set_nmi = vmx_inject_nmi,
7795 	.queue_exception = vmx_queue_exception,
7796 	.cancel_injection = vmx_cancel_injection,
7797 	.interrupt_allowed = vmx_interrupt_allowed,
7798 	.nmi_allowed = vmx_nmi_allowed,
7799 	.get_nmi_mask = vmx_get_nmi_mask,
7800 	.set_nmi_mask = vmx_set_nmi_mask,
7801 	.enable_nmi_window = enable_nmi_window,
7802 	.enable_irq_window = enable_irq_window,
7803 	.update_cr8_intercept = update_cr8_intercept,
7804 	.set_virtual_apic_mode = vmx_set_virtual_apic_mode,
7805 	.set_apic_access_page_addr = vmx_set_apic_access_page_addr,
7806 	.get_enable_apicv = vmx_get_enable_apicv,
7807 	.refresh_apicv_exec_ctrl = vmx_refresh_apicv_exec_ctrl,
7808 	.load_eoi_exitmap = vmx_load_eoi_exitmap,
7809 	.apicv_post_state_restore = vmx_apicv_post_state_restore,
7810 	.hwapic_irr_update = vmx_hwapic_irr_update,
7811 	.hwapic_isr_update = vmx_hwapic_isr_update,
7812 	.guest_apic_has_interrupt = vmx_guest_apic_has_interrupt,
7813 	.sync_pir_to_irr = vmx_sync_pir_to_irr,
7814 	.deliver_posted_interrupt = vmx_deliver_posted_interrupt,
7815 	.dy_apicv_has_pending_interrupt = vmx_dy_apicv_has_pending_interrupt,
7816 
7817 	.set_tss_addr = vmx_set_tss_addr,
7818 	.set_identity_map_addr = vmx_set_identity_map_addr,
7819 	.get_tdp_level = get_ept_level,
7820 	.get_mt_mask = vmx_get_mt_mask,
7821 
7822 	.get_exit_info = vmx_get_exit_info,
7823 
7824 	.get_lpage_level = vmx_get_lpage_level,
7825 
7826 	.cpuid_update = vmx_cpuid_update,
7827 
7828 	.rdtscp_supported = vmx_rdtscp_supported,
7829 	.invpcid_supported = vmx_invpcid_supported,
7830 
7831 	.set_supported_cpuid = vmx_set_supported_cpuid,
7832 
7833 	.has_wbinvd_exit = cpu_has_vmx_wbinvd_exit,
7834 
7835 	.read_l1_tsc_offset = vmx_read_l1_tsc_offset,
7836 	.write_l1_tsc_offset = vmx_write_l1_tsc_offset,
7837 
7838 	.set_tdp_cr3 = vmx_set_cr3,
7839 
7840 	.check_intercept = vmx_check_intercept,
7841 	.handle_exit_irqoff = vmx_handle_exit_irqoff,
7842 	.mpx_supported = vmx_mpx_supported,
7843 	.xsaves_supported = vmx_xsaves_supported,
7844 	.umip_emulated = vmx_umip_emulated,
7845 	.pt_supported = vmx_pt_supported,
7846 
7847 	.request_immediate_exit = vmx_request_immediate_exit,
7848 
7849 	.sched_in = vmx_sched_in,
7850 
7851 	.slot_enable_log_dirty = vmx_slot_enable_log_dirty,
7852 	.slot_disable_log_dirty = vmx_slot_disable_log_dirty,
7853 	.flush_log_dirty = vmx_flush_log_dirty,
7854 	.enable_log_dirty_pt_masked = vmx_enable_log_dirty_pt_masked,
7855 	.write_log_dirty = vmx_write_pml_buffer,
7856 
7857 	.pre_block = vmx_pre_block,
7858 	.post_block = vmx_post_block,
7859 
7860 	.pmu_ops = &intel_pmu_ops,
7861 
7862 	.update_pi_irte = vmx_update_pi_irte,
7863 
7864 #ifdef CONFIG_X86_64
7865 	.set_hv_timer = vmx_set_hv_timer,
7866 	.cancel_hv_timer = vmx_cancel_hv_timer,
7867 #endif
7868 
7869 	.setup_mce = vmx_setup_mce,
7870 
7871 	.smi_allowed = vmx_smi_allowed,
7872 	.pre_enter_smm = vmx_pre_enter_smm,
7873 	.pre_leave_smm = vmx_pre_leave_smm,
7874 	.enable_smi_window = enable_smi_window,
7875 
7876 	.check_nested_events = NULL,
7877 	.get_nested_state = NULL,
7878 	.set_nested_state = NULL,
7879 	.get_vmcs12_pages = NULL,
7880 	.nested_enable_evmcs = NULL,
7881 	.nested_get_evmcs_version = NULL,
7882 	.need_emulation_on_page_fault = vmx_need_emulation_on_page_fault,
7883 	.apic_init_signal_blocked = vmx_apic_init_signal_blocked,
7884 };
7885 
7886 static void vmx_cleanup_l1d_flush(void)
7887 {
7888 	if (vmx_l1d_flush_pages) {
7889 		free_pages((unsigned long)vmx_l1d_flush_pages, L1D_CACHE_ORDER);
7890 		vmx_l1d_flush_pages = NULL;
7891 	}
7892 	/* Restore state so sysfs ignores VMX */
7893 	l1tf_vmx_mitigation = VMENTER_L1D_FLUSH_AUTO;
7894 }
7895 
7896 static void vmx_exit(void)
7897 {
7898 #ifdef CONFIG_KEXEC_CORE
7899 	RCU_INIT_POINTER(crash_vmclear_loaded_vmcss, NULL);
7900 	synchronize_rcu();
7901 #endif
7902 
7903 	kvm_exit();
7904 
7905 #if IS_ENABLED(CONFIG_HYPERV)
7906 	if (static_branch_unlikely(&enable_evmcs)) {
7907 		int cpu;
7908 		struct hv_vp_assist_page *vp_ap;
7909 		/*
7910 		 * Reset everything to support using non-enlightened VMCS
7911 		 * access later (e.g. when we reload the module with
7912 		 * enlightened_vmcs=0)
7913 		 */
7914 		for_each_online_cpu(cpu) {
7915 			vp_ap =	hv_get_vp_assist_page(cpu);
7916 
7917 			if (!vp_ap)
7918 				continue;
7919 
7920 			vp_ap->nested_control.features.directhypercall = 0;
7921 			vp_ap->current_nested_vmcs = 0;
7922 			vp_ap->enlighten_vmentry = 0;
7923 		}
7924 
7925 		static_branch_disable(&enable_evmcs);
7926 	}
7927 #endif
7928 	vmx_cleanup_l1d_flush();
7929 }
7930 module_exit(vmx_exit);
7931 
7932 static int __init vmx_init(void)
7933 {
7934 	int r;
7935 
7936 #if IS_ENABLED(CONFIG_HYPERV)
7937 	/*
7938 	 * Enlightened VMCS usage should be recommended and the host needs
7939 	 * to support eVMCS v1 or above. We can also disable eVMCS support
7940 	 * with module parameter.
7941 	 */
7942 	if (enlightened_vmcs &&
7943 	    ms_hyperv.hints & HV_X64_ENLIGHTENED_VMCS_RECOMMENDED &&
7944 	    (ms_hyperv.nested_features & HV_X64_ENLIGHTENED_VMCS_VERSION) >=
7945 	    KVM_EVMCS_VERSION) {
7946 		int cpu;
7947 
7948 		/* Check that we have assist pages on all online CPUs */
7949 		for_each_online_cpu(cpu) {
7950 			if (!hv_get_vp_assist_page(cpu)) {
7951 				enlightened_vmcs = false;
7952 				break;
7953 			}
7954 		}
7955 
7956 		if (enlightened_vmcs) {
7957 			pr_info("KVM: vmx: using Hyper-V Enlightened VMCS\n");
7958 			static_branch_enable(&enable_evmcs);
7959 		}
7960 
7961 		if (ms_hyperv.nested_features & HV_X64_NESTED_DIRECT_FLUSH)
7962 			vmx_x86_ops.enable_direct_tlbflush
7963 				= hv_enable_direct_tlbflush;
7964 
7965 	} else {
7966 		enlightened_vmcs = false;
7967 	}
7968 #endif
7969 
7970 	r = kvm_init(&vmx_x86_ops, sizeof(struct vcpu_vmx),
7971 		     __alignof__(struct vcpu_vmx), THIS_MODULE);
7972 	if (r)
7973 		return r;
7974 
7975 	/*
7976 	 * Must be called after kvm_init() so enable_ept is properly set
7977 	 * up. Hand the parameter mitigation value in which was stored in
7978 	 * the pre module init parser. If no parameter was given, it will
7979 	 * contain 'auto' which will be turned into the default 'cond'
7980 	 * mitigation mode.
7981 	 */
7982 	r = vmx_setup_l1d_flush(vmentry_l1d_flush_param);
7983 	if (r) {
7984 		vmx_exit();
7985 		return r;
7986 	}
7987 
7988 #ifdef CONFIG_KEXEC_CORE
7989 	rcu_assign_pointer(crash_vmclear_loaded_vmcss,
7990 			   crash_vmclear_local_loaded_vmcss);
7991 #endif
7992 	vmx_check_vmcs12_offsets();
7993 
7994 	return 0;
7995 }
7996 module_init(vmx_init);
7997