xref: /openbmc/linux/arch/x86/kvm/vmx/vmx.c (revision 519a8a6c)
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 #include <linux/entry-kvm.h>
31 
32 #include <asm/apic.h>
33 #include <asm/asm.h>
34 #include <asm/cpu.h>
35 #include <asm/cpu_device_id.h>
36 #include <asm/debugreg.h>
37 #include <asm/desc.h>
38 #include <asm/fpu/internal.h>
39 #include <asm/io.h>
40 #include <asm/irq_remapping.h>
41 #include <asm/kexec.h>
42 #include <asm/perf_event.h>
43 #include <asm/mce.h>
44 #include <asm/mmu_context.h>
45 #include <asm/mshyperv.h>
46 #include <asm/mwait.h>
47 #include <asm/spec-ctrl.h>
48 #include <asm/virtext.h>
49 #include <asm/vmx.h>
50 
51 #include "capabilities.h"
52 #include "cpuid.h"
53 #include "evmcs.h"
54 #include "irq.h"
55 #include "kvm_cache_regs.h"
56 #include "lapic.h"
57 #include "mmu.h"
58 #include "nested.h"
59 #include "ops.h"
60 #include "pmu.h"
61 #include "trace.h"
62 #include "vmcs.h"
63 #include "vmcs12.h"
64 #include "vmx.h"
65 #include "x86.h"
66 
67 MODULE_AUTHOR("Qumranet");
68 MODULE_LICENSE("GPL");
69 
70 #ifdef MODULE
71 static const struct x86_cpu_id vmx_cpu_id[] = {
72 	X86_MATCH_FEATURE(X86_FEATURE_VMX, NULL),
73 	{}
74 };
75 MODULE_DEVICE_TABLE(x86cpu, vmx_cpu_id);
76 #endif
77 
78 bool __read_mostly enable_vpid = 1;
79 module_param_named(vpid, enable_vpid, bool, 0444);
80 
81 static bool __read_mostly enable_vnmi = 1;
82 module_param_named(vnmi, enable_vnmi, bool, S_IRUGO);
83 
84 bool __read_mostly flexpriority_enabled = 1;
85 module_param_named(flexpriority, flexpriority_enabled, bool, S_IRUGO);
86 
87 bool __read_mostly enable_ept = 1;
88 module_param_named(ept, enable_ept, bool, S_IRUGO);
89 
90 bool __read_mostly enable_unrestricted_guest = 1;
91 module_param_named(unrestricted_guest,
92 			enable_unrestricted_guest, bool, S_IRUGO);
93 
94 bool __read_mostly enable_ept_ad_bits = 1;
95 module_param_named(eptad, enable_ept_ad_bits, bool, S_IRUGO);
96 
97 static bool __read_mostly emulate_invalid_guest_state = true;
98 module_param(emulate_invalid_guest_state, bool, S_IRUGO);
99 
100 static bool __read_mostly fasteoi = 1;
101 module_param(fasteoi, bool, S_IRUGO);
102 
103 bool __read_mostly enable_apicv = 1;
104 module_param(enable_apicv, bool, S_IRUGO);
105 
106 /*
107  * If nested=1, nested virtualization is supported, i.e., guests may use
108  * VMX and be a hypervisor for its own guests. If nested=0, guests may not
109  * use VMX instructions.
110  */
111 static bool __read_mostly nested = 1;
112 module_param(nested, bool, S_IRUGO);
113 
114 bool __read_mostly enable_pml = 1;
115 module_param_named(pml, enable_pml, bool, S_IRUGO);
116 
117 static bool __read_mostly dump_invalid_vmcs = 0;
118 module_param(dump_invalid_vmcs, bool, 0644);
119 
120 #define MSR_BITMAP_MODE_X2APIC		1
121 #define MSR_BITMAP_MODE_X2APIC_APICV	2
122 
123 #define KVM_VMX_TSC_MULTIPLIER_MAX     0xffffffffffffffffULL
124 
125 /* Guest_tsc -> host_tsc conversion requires 64-bit division.  */
126 static int __read_mostly cpu_preemption_timer_multi;
127 static bool __read_mostly enable_preemption_timer = 1;
128 #ifdef CONFIG_X86_64
129 module_param_named(preemption_timer, enable_preemption_timer, bool, S_IRUGO);
130 #endif
131 
132 #define KVM_VM_CR0_ALWAYS_OFF (X86_CR0_NW | X86_CR0_CD)
133 #define KVM_VM_CR0_ALWAYS_ON_UNRESTRICTED_GUEST X86_CR0_NE
134 #define KVM_VM_CR0_ALWAYS_ON				\
135 	(KVM_VM_CR0_ALWAYS_ON_UNRESTRICTED_GUEST | 	\
136 	 X86_CR0_WP | X86_CR0_PG | X86_CR0_PE)
137 
138 #define KVM_VM_CR4_ALWAYS_ON_UNRESTRICTED_GUEST X86_CR4_VMXE
139 #define KVM_PMODE_VM_CR4_ALWAYS_ON (X86_CR4_PAE | X86_CR4_VMXE)
140 #define KVM_RMODE_VM_CR4_ALWAYS_ON (X86_CR4_VME | X86_CR4_PAE | X86_CR4_VMXE)
141 
142 #define RMODE_GUEST_OWNED_EFLAGS_BITS (~(X86_EFLAGS_IOPL | X86_EFLAGS_VM))
143 
144 #define MSR_IA32_RTIT_STATUS_MASK (~(RTIT_STATUS_FILTEREN | \
145 	RTIT_STATUS_CONTEXTEN | RTIT_STATUS_TRIGGEREN | \
146 	RTIT_STATUS_ERROR | RTIT_STATUS_STOPPED | \
147 	RTIT_STATUS_BYTECNT))
148 
149 #define MSR_IA32_RTIT_OUTPUT_BASE_MASK \
150 	(~((1UL << cpuid_query_maxphyaddr(vcpu)) - 1) | 0x7f)
151 
152 /*
153  * These 2 parameters are used to config the controls for Pause-Loop Exiting:
154  * ple_gap:    upper bound on the amount of time between two successive
155  *             executions of PAUSE in a loop. Also indicate if ple enabled.
156  *             According to test, this time is usually smaller than 128 cycles.
157  * ple_window: upper bound on the amount of time a guest is allowed to execute
158  *             in a PAUSE loop. Tests indicate that most spinlocks are held for
159  *             less than 2^12 cycles
160  * Time is measured based on a counter that runs at the same rate as the TSC,
161  * refer SDM volume 3b section 21.6.13 & 22.1.3.
162  */
163 static unsigned int ple_gap = KVM_DEFAULT_PLE_GAP;
164 module_param(ple_gap, uint, 0444);
165 
166 static unsigned int ple_window = KVM_VMX_DEFAULT_PLE_WINDOW;
167 module_param(ple_window, uint, 0444);
168 
169 /* Default doubles per-vcpu window every exit. */
170 static unsigned int ple_window_grow = KVM_DEFAULT_PLE_WINDOW_GROW;
171 module_param(ple_window_grow, uint, 0444);
172 
173 /* Default resets per-vcpu window every exit to ple_window. */
174 static unsigned int ple_window_shrink = KVM_DEFAULT_PLE_WINDOW_SHRINK;
175 module_param(ple_window_shrink, uint, 0444);
176 
177 /* Default is to compute the maximum so we can never overflow. */
178 static unsigned int ple_window_max        = KVM_VMX_DEFAULT_PLE_WINDOW_MAX;
179 module_param(ple_window_max, uint, 0444);
180 
181 /* Default is SYSTEM mode, 1 for host-guest mode */
182 int __read_mostly pt_mode = PT_MODE_SYSTEM;
183 module_param(pt_mode, int, S_IRUGO);
184 
185 static DEFINE_STATIC_KEY_FALSE(vmx_l1d_should_flush);
186 static DEFINE_STATIC_KEY_FALSE(vmx_l1d_flush_cond);
187 static DEFINE_MUTEX(vmx_l1d_flush_mutex);
188 
189 /* Storage for pre module init parameter parsing */
190 static enum vmx_l1d_flush_state __read_mostly vmentry_l1d_flush_param = VMENTER_L1D_FLUSH_AUTO;
191 
192 static const struct {
193 	const char *option;
194 	bool for_parse;
195 } vmentry_l1d_param[] = {
196 	[VMENTER_L1D_FLUSH_AUTO]	 = {"auto", true},
197 	[VMENTER_L1D_FLUSH_NEVER]	 = {"never", true},
198 	[VMENTER_L1D_FLUSH_COND]	 = {"cond", true},
199 	[VMENTER_L1D_FLUSH_ALWAYS]	 = {"always", true},
200 	[VMENTER_L1D_FLUSH_EPT_DISABLED] = {"EPT disabled", false},
201 	[VMENTER_L1D_FLUSH_NOT_REQUIRED] = {"not required", false},
202 };
203 
204 #define L1D_CACHE_ORDER 4
205 static void *vmx_l1d_flush_pages;
206 
207 static int vmx_setup_l1d_flush(enum vmx_l1d_flush_state l1tf)
208 {
209 	struct page *page;
210 	unsigned int i;
211 
212 	if (!boot_cpu_has_bug(X86_BUG_L1TF)) {
213 		l1tf_vmx_mitigation = VMENTER_L1D_FLUSH_NOT_REQUIRED;
214 		return 0;
215 	}
216 
217 	if (!enable_ept) {
218 		l1tf_vmx_mitigation = VMENTER_L1D_FLUSH_EPT_DISABLED;
219 		return 0;
220 	}
221 
222 	if (boot_cpu_has(X86_FEATURE_ARCH_CAPABILITIES)) {
223 		u64 msr;
224 
225 		rdmsrl(MSR_IA32_ARCH_CAPABILITIES, msr);
226 		if (msr & ARCH_CAP_SKIP_VMENTRY_L1DFLUSH) {
227 			l1tf_vmx_mitigation = VMENTER_L1D_FLUSH_NOT_REQUIRED;
228 			return 0;
229 		}
230 	}
231 
232 	/* If set to auto use the default l1tf mitigation method */
233 	if (l1tf == VMENTER_L1D_FLUSH_AUTO) {
234 		switch (l1tf_mitigation) {
235 		case L1TF_MITIGATION_OFF:
236 			l1tf = VMENTER_L1D_FLUSH_NEVER;
237 			break;
238 		case L1TF_MITIGATION_FLUSH_NOWARN:
239 		case L1TF_MITIGATION_FLUSH:
240 		case L1TF_MITIGATION_FLUSH_NOSMT:
241 			l1tf = VMENTER_L1D_FLUSH_COND;
242 			break;
243 		case L1TF_MITIGATION_FULL:
244 		case L1TF_MITIGATION_FULL_FORCE:
245 			l1tf = VMENTER_L1D_FLUSH_ALWAYS;
246 			break;
247 		}
248 	} else if (l1tf_mitigation == L1TF_MITIGATION_FULL_FORCE) {
249 		l1tf = VMENTER_L1D_FLUSH_ALWAYS;
250 	}
251 
252 	if (l1tf != VMENTER_L1D_FLUSH_NEVER && !vmx_l1d_flush_pages &&
253 	    !boot_cpu_has(X86_FEATURE_FLUSH_L1D)) {
254 		/*
255 		 * This allocation for vmx_l1d_flush_pages is not tied to a VM
256 		 * lifetime and so should not be charged to a memcg.
257 		 */
258 		page = alloc_pages(GFP_KERNEL, L1D_CACHE_ORDER);
259 		if (!page)
260 			return -ENOMEM;
261 		vmx_l1d_flush_pages = page_address(page);
262 
263 		/*
264 		 * Initialize each page with a different pattern in
265 		 * order to protect against KSM in the nested
266 		 * virtualization case.
267 		 */
268 		for (i = 0; i < 1u << L1D_CACHE_ORDER; ++i) {
269 			memset(vmx_l1d_flush_pages + i * PAGE_SIZE, i + 1,
270 			       PAGE_SIZE);
271 		}
272 	}
273 
274 	l1tf_vmx_mitigation = l1tf;
275 
276 	if (l1tf != VMENTER_L1D_FLUSH_NEVER)
277 		static_branch_enable(&vmx_l1d_should_flush);
278 	else
279 		static_branch_disable(&vmx_l1d_should_flush);
280 
281 	if (l1tf == VMENTER_L1D_FLUSH_COND)
282 		static_branch_enable(&vmx_l1d_flush_cond);
283 	else
284 		static_branch_disable(&vmx_l1d_flush_cond);
285 	return 0;
286 }
287 
288 static int vmentry_l1d_flush_parse(const char *s)
289 {
290 	unsigned int i;
291 
292 	if (s) {
293 		for (i = 0; i < ARRAY_SIZE(vmentry_l1d_param); i++) {
294 			if (vmentry_l1d_param[i].for_parse &&
295 			    sysfs_streq(s, vmentry_l1d_param[i].option))
296 				return i;
297 		}
298 	}
299 	return -EINVAL;
300 }
301 
302 static int vmentry_l1d_flush_set(const char *s, const struct kernel_param *kp)
303 {
304 	int l1tf, ret;
305 
306 	l1tf = vmentry_l1d_flush_parse(s);
307 	if (l1tf < 0)
308 		return l1tf;
309 
310 	if (!boot_cpu_has(X86_BUG_L1TF))
311 		return 0;
312 
313 	/*
314 	 * Has vmx_init() run already? If not then this is the pre init
315 	 * parameter parsing. In that case just store the value and let
316 	 * vmx_init() do the proper setup after enable_ept has been
317 	 * established.
318 	 */
319 	if (l1tf_vmx_mitigation == VMENTER_L1D_FLUSH_AUTO) {
320 		vmentry_l1d_flush_param = l1tf;
321 		return 0;
322 	}
323 
324 	mutex_lock(&vmx_l1d_flush_mutex);
325 	ret = vmx_setup_l1d_flush(l1tf);
326 	mutex_unlock(&vmx_l1d_flush_mutex);
327 	return ret;
328 }
329 
330 static int vmentry_l1d_flush_get(char *s, const struct kernel_param *kp)
331 {
332 	if (WARN_ON_ONCE(l1tf_vmx_mitigation >= ARRAY_SIZE(vmentry_l1d_param)))
333 		return sprintf(s, "???\n");
334 
335 	return sprintf(s, "%s\n", vmentry_l1d_param[l1tf_vmx_mitigation].option);
336 }
337 
338 static const struct kernel_param_ops vmentry_l1d_flush_ops = {
339 	.set = vmentry_l1d_flush_set,
340 	.get = vmentry_l1d_flush_get,
341 };
342 module_param_cb(vmentry_l1d_flush, &vmentry_l1d_flush_ops, NULL, 0644);
343 
344 static bool guest_state_valid(struct kvm_vcpu *vcpu);
345 static u32 vmx_segment_access_rights(struct kvm_segment *var);
346 static __always_inline void vmx_disable_intercept_for_msr(unsigned long *msr_bitmap,
347 							  u32 msr, int type);
348 
349 void vmx_vmexit(void);
350 
351 #define vmx_insn_failed(fmt...)		\
352 do {					\
353 	WARN_ONCE(1, fmt);		\
354 	pr_warn_ratelimited(fmt);	\
355 } while (0)
356 
357 asmlinkage void vmread_error(unsigned long field, bool fault)
358 {
359 	if (fault)
360 		kvm_spurious_fault();
361 	else
362 		vmx_insn_failed("kvm: vmread failed: field=%lx\n", field);
363 }
364 
365 noinline void vmwrite_error(unsigned long field, unsigned long value)
366 {
367 	vmx_insn_failed("kvm: vmwrite failed: field=%lx val=%lx err=%d\n",
368 			field, value, vmcs_read32(VM_INSTRUCTION_ERROR));
369 }
370 
371 noinline void vmclear_error(struct vmcs *vmcs, u64 phys_addr)
372 {
373 	vmx_insn_failed("kvm: vmclear failed: %p/%llx\n", vmcs, phys_addr);
374 }
375 
376 noinline void vmptrld_error(struct vmcs *vmcs, u64 phys_addr)
377 {
378 	vmx_insn_failed("kvm: vmptrld failed: %p/%llx\n", vmcs, phys_addr);
379 }
380 
381 noinline void invvpid_error(unsigned long ext, u16 vpid, gva_t gva)
382 {
383 	vmx_insn_failed("kvm: invvpid failed: ext=0x%lx vpid=%u gva=0x%lx\n",
384 			ext, vpid, gva);
385 }
386 
387 noinline void invept_error(unsigned long ext, u64 eptp, gpa_t gpa)
388 {
389 	vmx_insn_failed("kvm: invept failed: ext=0x%lx eptp=%llx gpa=0x%llx\n",
390 			ext, eptp, gpa);
391 }
392 
393 static DEFINE_PER_CPU(struct vmcs *, vmxarea);
394 DEFINE_PER_CPU(struct vmcs *, current_vmcs);
395 /*
396  * We maintain a per-CPU linked-list of VMCS loaded on that CPU. This is needed
397  * when a CPU is brought down, and we need to VMCLEAR all VMCSs loaded on it.
398  */
399 static DEFINE_PER_CPU(struct list_head, loaded_vmcss_on_cpu);
400 
401 /*
402  * We maintian a per-CPU linked-list of vCPU, so in wakeup_handler() we
403  * can find which vCPU should be waken up.
404  */
405 static DEFINE_PER_CPU(struct list_head, blocked_vcpu_on_cpu);
406 static DEFINE_PER_CPU(spinlock_t, blocked_vcpu_on_cpu_lock);
407 
408 static DECLARE_BITMAP(vmx_vpid_bitmap, VMX_NR_VPIDS);
409 static DEFINE_SPINLOCK(vmx_vpid_lock);
410 
411 struct vmcs_config vmcs_config;
412 struct vmx_capability vmx_capability;
413 
414 #define VMX_SEGMENT_FIELD(seg)					\
415 	[VCPU_SREG_##seg] = {                                   \
416 		.selector = GUEST_##seg##_SELECTOR,		\
417 		.base = GUEST_##seg##_BASE,		   	\
418 		.limit = GUEST_##seg##_LIMIT,		   	\
419 		.ar_bytes = GUEST_##seg##_AR_BYTES,	   	\
420 	}
421 
422 static const struct kvm_vmx_segment_field {
423 	unsigned selector;
424 	unsigned base;
425 	unsigned limit;
426 	unsigned ar_bytes;
427 } kvm_vmx_segment_fields[] = {
428 	VMX_SEGMENT_FIELD(CS),
429 	VMX_SEGMENT_FIELD(DS),
430 	VMX_SEGMENT_FIELD(ES),
431 	VMX_SEGMENT_FIELD(FS),
432 	VMX_SEGMENT_FIELD(GS),
433 	VMX_SEGMENT_FIELD(SS),
434 	VMX_SEGMENT_FIELD(TR),
435 	VMX_SEGMENT_FIELD(LDTR),
436 };
437 
438 static inline void vmx_segment_cache_clear(struct vcpu_vmx *vmx)
439 {
440 	vmx->segment_cache.bitmask = 0;
441 }
442 
443 static unsigned long host_idt_base;
444 
445 /*
446  * Though SYSCALL is only supported in 64-bit mode on Intel CPUs, kvm
447  * will emulate SYSCALL in legacy mode if the vendor string in guest
448  * CPUID.0:{EBX,ECX,EDX} is "AuthenticAMD" or "AMDisbetter!" To
449  * support this emulation, IA32_STAR must always be included in
450  * vmx_msr_index[], even in i386 builds.
451  */
452 const u32 vmx_msr_index[] = {
453 #ifdef CONFIG_X86_64
454 	MSR_SYSCALL_MASK, MSR_LSTAR, MSR_CSTAR,
455 #endif
456 	MSR_EFER, MSR_TSC_AUX, MSR_STAR,
457 	MSR_IA32_TSX_CTRL,
458 };
459 
460 #if IS_ENABLED(CONFIG_HYPERV)
461 static bool __read_mostly enlightened_vmcs = true;
462 module_param(enlightened_vmcs, bool, 0444);
463 
464 /* check_ept_pointer() should be under protection of ept_pointer_lock. */
465 static void check_ept_pointer_match(struct kvm *kvm)
466 {
467 	struct kvm_vcpu *vcpu;
468 	u64 tmp_eptp = INVALID_PAGE;
469 	int i;
470 
471 	kvm_for_each_vcpu(i, vcpu, kvm) {
472 		if (!VALID_PAGE(tmp_eptp)) {
473 			tmp_eptp = to_vmx(vcpu)->ept_pointer;
474 		} else if (tmp_eptp != to_vmx(vcpu)->ept_pointer) {
475 			to_kvm_vmx(kvm)->ept_pointers_match
476 				= EPT_POINTERS_MISMATCH;
477 			return;
478 		}
479 	}
480 
481 	to_kvm_vmx(kvm)->ept_pointers_match = EPT_POINTERS_MATCH;
482 }
483 
484 static int kvm_fill_hv_flush_list_func(struct hv_guest_mapping_flush_list *flush,
485 		void *data)
486 {
487 	struct kvm_tlb_range *range = data;
488 
489 	return hyperv_fill_flush_guest_mapping_list(flush, range->start_gfn,
490 			range->pages);
491 }
492 
493 static inline int __hv_remote_flush_tlb_with_range(struct kvm *kvm,
494 		struct kvm_vcpu *vcpu, struct kvm_tlb_range *range)
495 {
496 	u64 ept_pointer = to_vmx(vcpu)->ept_pointer;
497 
498 	/*
499 	 * FLUSH_GUEST_PHYSICAL_ADDRESS_SPACE hypercall needs address
500 	 * of the base of EPT PML4 table, strip off EPT configuration
501 	 * information.
502 	 */
503 	if (range)
504 		return hyperv_flush_guest_mapping_range(ept_pointer & PAGE_MASK,
505 				kvm_fill_hv_flush_list_func, (void *)range);
506 	else
507 		return hyperv_flush_guest_mapping(ept_pointer & PAGE_MASK);
508 }
509 
510 static int hv_remote_flush_tlb_with_range(struct kvm *kvm,
511 		struct kvm_tlb_range *range)
512 {
513 	struct kvm_vcpu *vcpu;
514 	int ret = 0, i;
515 
516 	spin_lock(&to_kvm_vmx(kvm)->ept_pointer_lock);
517 
518 	if (to_kvm_vmx(kvm)->ept_pointers_match == EPT_POINTERS_CHECK)
519 		check_ept_pointer_match(kvm);
520 
521 	if (to_kvm_vmx(kvm)->ept_pointers_match != EPT_POINTERS_MATCH) {
522 		kvm_for_each_vcpu(i, vcpu, kvm) {
523 			/* If ept_pointer is invalid pointer, bypass flush request. */
524 			if (VALID_PAGE(to_vmx(vcpu)->ept_pointer))
525 				ret |= __hv_remote_flush_tlb_with_range(
526 					kvm, vcpu, range);
527 		}
528 	} else {
529 		ret = __hv_remote_flush_tlb_with_range(kvm,
530 				kvm_get_vcpu(kvm, 0), range);
531 	}
532 
533 	spin_unlock(&to_kvm_vmx(kvm)->ept_pointer_lock);
534 	return ret;
535 }
536 static int hv_remote_flush_tlb(struct kvm *kvm)
537 {
538 	return hv_remote_flush_tlb_with_range(kvm, NULL);
539 }
540 
541 static int hv_enable_direct_tlbflush(struct kvm_vcpu *vcpu)
542 {
543 	struct hv_enlightened_vmcs *evmcs;
544 	struct hv_partition_assist_pg **p_hv_pa_pg =
545 			&vcpu->kvm->arch.hyperv.hv_pa_pg;
546 	/*
547 	 * Synthetic VM-Exit is not enabled in current code and so All
548 	 * evmcs in singe VM shares same assist page.
549 	 */
550 	if (!*p_hv_pa_pg)
551 		*p_hv_pa_pg = kzalloc(PAGE_SIZE, GFP_KERNEL);
552 
553 	if (!*p_hv_pa_pg)
554 		return -ENOMEM;
555 
556 	evmcs = (struct hv_enlightened_vmcs *)to_vmx(vcpu)->loaded_vmcs->vmcs;
557 
558 	evmcs->partition_assist_page =
559 		__pa(*p_hv_pa_pg);
560 	evmcs->hv_vm_id = (unsigned long)vcpu->kvm;
561 	evmcs->hv_enlightenments_control.nested_flush_hypercall = 1;
562 
563 	return 0;
564 }
565 
566 #endif /* IS_ENABLED(CONFIG_HYPERV) */
567 
568 /*
569  * Comment's format: document - errata name - stepping - processor name.
570  * Refer from
571  * https://www.virtualbox.org/svn/vbox/trunk/src/VBox/VMM/VMMR0/HMR0.cpp
572  */
573 static u32 vmx_preemption_cpu_tfms[] = {
574 /* 323344.pdf - BA86   - D0 - Xeon 7500 Series */
575 0x000206E6,
576 /* 323056.pdf - AAX65  - C2 - Xeon L3406 */
577 /* 322814.pdf - AAT59  - C2 - i7-600, i5-500, i5-400 and i3-300 Mobile */
578 /* 322911.pdf - AAU65  - C2 - i5-600, i3-500 Desktop and Pentium G6950 */
579 0x00020652,
580 /* 322911.pdf - AAU65  - K0 - i5-600, i3-500 Desktop and Pentium G6950 */
581 0x00020655,
582 /* 322373.pdf - AAO95  - B1 - Xeon 3400 Series */
583 /* 322166.pdf - AAN92  - B1 - i7-800 and i5-700 Desktop */
584 /*
585  * 320767.pdf - AAP86  - B1 -
586  * i7-900 Mobile Extreme, i7-800 and i7-700 Mobile
587  */
588 0x000106E5,
589 /* 321333.pdf - AAM126 - C0 - Xeon 3500 */
590 0x000106A0,
591 /* 321333.pdf - AAM126 - C1 - Xeon 3500 */
592 0x000106A1,
593 /* 320836.pdf - AAJ124 - C0 - i7-900 Desktop Extreme and i7-900 Desktop */
594 0x000106A4,
595  /* 321333.pdf - AAM126 - D0 - Xeon 3500 */
596  /* 321324.pdf - AAK139 - D0 - Xeon 5500 */
597  /* 320836.pdf - AAJ124 - D0 - i7-900 Extreme and i7-900 Desktop */
598 0x000106A5,
599  /* Xeon E3-1220 V2 */
600 0x000306A8,
601 };
602 
603 static inline bool cpu_has_broken_vmx_preemption_timer(void)
604 {
605 	u32 eax = cpuid_eax(0x00000001), i;
606 
607 	/* Clear the reserved bits */
608 	eax &= ~(0x3U << 14 | 0xfU << 28);
609 	for (i = 0; i < ARRAY_SIZE(vmx_preemption_cpu_tfms); i++)
610 		if (eax == vmx_preemption_cpu_tfms[i])
611 			return true;
612 
613 	return false;
614 }
615 
616 static inline bool cpu_need_virtualize_apic_accesses(struct kvm_vcpu *vcpu)
617 {
618 	return flexpriority_enabled && lapic_in_kernel(vcpu);
619 }
620 
621 static inline bool report_flexpriority(void)
622 {
623 	return flexpriority_enabled;
624 }
625 
626 static inline int __find_msr_index(struct vcpu_vmx *vmx, u32 msr)
627 {
628 	int i;
629 
630 	for (i = 0; i < vmx->nmsrs; ++i)
631 		if (vmx_msr_index[vmx->guest_msrs[i].index] == msr)
632 			return i;
633 	return -1;
634 }
635 
636 struct shared_msr_entry *find_msr_entry(struct vcpu_vmx *vmx, u32 msr)
637 {
638 	int i;
639 
640 	i = __find_msr_index(vmx, msr);
641 	if (i >= 0)
642 		return &vmx->guest_msrs[i];
643 	return NULL;
644 }
645 
646 static int vmx_set_guest_msr(struct vcpu_vmx *vmx, struct shared_msr_entry *msr, u64 data)
647 {
648 	int ret = 0;
649 
650 	u64 old_msr_data = msr->data;
651 	msr->data = data;
652 	if (msr - vmx->guest_msrs < vmx->save_nmsrs) {
653 		preempt_disable();
654 		ret = kvm_set_shared_msr(msr->index, msr->data,
655 					 msr->mask);
656 		preempt_enable();
657 		if (ret)
658 			msr->data = old_msr_data;
659 	}
660 	return ret;
661 }
662 
663 #ifdef CONFIG_KEXEC_CORE
664 static void crash_vmclear_local_loaded_vmcss(void)
665 {
666 	int cpu = raw_smp_processor_id();
667 	struct loaded_vmcs *v;
668 
669 	list_for_each_entry(v, &per_cpu(loaded_vmcss_on_cpu, cpu),
670 			    loaded_vmcss_on_cpu_link)
671 		vmcs_clear(v->vmcs);
672 }
673 #endif /* CONFIG_KEXEC_CORE */
674 
675 static void __loaded_vmcs_clear(void *arg)
676 {
677 	struct loaded_vmcs *loaded_vmcs = arg;
678 	int cpu = raw_smp_processor_id();
679 
680 	if (loaded_vmcs->cpu != cpu)
681 		return; /* vcpu migration can race with cpu offline */
682 	if (per_cpu(current_vmcs, cpu) == loaded_vmcs->vmcs)
683 		per_cpu(current_vmcs, cpu) = NULL;
684 
685 	vmcs_clear(loaded_vmcs->vmcs);
686 	if (loaded_vmcs->shadow_vmcs && loaded_vmcs->launched)
687 		vmcs_clear(loaded_vmcs->shadow_vmcs);
688 
689 	list_del(&loaded_vmcs->loaded_vmcss_on_cpu_link);
690 
691 	/*
692 	 * Ensure all writes to loaded_vmcs, including deleting it from its
693 	 * current percpu list, complete before setting loaded_vmcs->vcpu to
694 	 * -1, otherwise a different cpu can see vcpu == -1 first and add
695 	 * loaded_vmcs to its percpu list before it's deleted from this cpu's
696 	 * list. Pairs with the smp_rmb() in vmx_vcpu_load_vmcs().
697 	 */
698 	smp_wmb();
699 
700 	loaded_vmcs->cpu = -1;
701 	loaded_vmcs->launched = 0;
702 }
703 
704 void loaded_vmcs_clear(struct loaded_vmcs *loaded_vmcs)
705 {
706 	int cpu = loaded_vmcs->cpu;
707 
708 	if (cpu != -1)
709 		smp_call_function_single(cpu,
710 			 __loaded_vmcs_clear, loaded_vmcs, 1);
711 }
712 
713 static bool vmx_segment_cache_test_set(struct vcpu_vmx *vmx, unsigned seg,
714 				       unsigned field)
715 {
716 	bool ret;
717 	u32 mask = 1 << (seg * SEG_FIELD_NR + field);
718 
719 	if (!kvm_register_is_available(&vmx->vcpu, VCPU_EXREG_SEGMENTS)) {
720 		kvm_register_mark_available(&vmx->vcpu, VCPU_EXREG_SEGMENTS);
721 		vmx->segment_cache.bitmask = 0;
722 	}
723 	ret = vmx->segment_cache.bitmask & mask;
724 	vmx->segment_cache.bitmask |= mask;
725 	return ret;
726 }
727 
728 static u16 vmx_read_guest_seg_selector(struct vcpu_vmx *vmx, unsigned seg)
729 {
730 	u16 *p = &vmx->segment_cache.seg[seg].selector;
731 
732 	if (!vmx_segment_cache_test_set(vmx, seg, SEG_FIELD_SEL))
733 		*p = vmcs_read16(kvm_vmx_segment_fields[seg].selector);
734 	return *p;
735 }
736 
737 static ulong vmx_read_guest_seg_base(struct vcpu_vmx *vmx, unsigned seg)
738 {
739 	ulong *p = &vmx->segment_cache.seg[seg].base;
740 
741 	if (!vmx_segment_cache_test_set(vmx, seg, SEG_FIELD_BASE))
742 		*p = vmcs_readl(kvm_vmx_segment_fields[seg].base);
743 	return *p;
744 }
745 
746 static u32 vmx_read_guest_seg_limit(struct vcpu_vmx *vmx, unsigned seg)
747 {
748 	u32 *p = &vmx->segment_cache.seg[seg].limit;
749 
750 	if (!vmx_segment_cache_test_set(vmx, seg, SEG_FIELD_LIMIT))
751 		*p = vmcs_read32(kvm_vmx_segment_fields[seg].limit);
752 	return *p;
753 }
754 
755 static u32 vmx_read_guest_seg_ar(struct vcpu_vmx *vmx, unsigned seg)
756 {
757 	u32 *p = &vmx->segment_cache.seg[seg].ar;
758 
759 	if (!vmx_segment_cache_test_set(vmx, seg, SEG_FIELD_AR))
760 		*p = vmcs_read32(kvm_vmx_segment_fields[seg].ar_bytes);
761 	return *p;
762 }
763 
764 void update_exception_bitmap(struct kvm_vcpu *vcpu)
765 {
766 	u32 eb;
767 
768 	eb = (1u << PF_VECTOR) | (1u << UD_VECTOR) | (1u << MC_VECTOR) |
769 	     (1u << DB_VECTOR) | (1u << AC_VECTOR);
770 	/*
771 	 * Guest access to VMware backdoor ports could legitimately
772 	 * trigger #GP because of TSS I/O permission bitmap.
773 	 * We intercept those #GP and allow access to them anyway
774 	 * as VMware does.
775 	 */
776 	if (enable_vmware_backdoor)
777 		eb |= (1u << GP_VECTOR);
778 	if ((vcpu->guest_debug &
779 	     (KVM_GUESTDBG_ENABLE | KVM_GUESTDBG_USE_SW_BP)) ==
780 	    (KVM_GUESTDBG_ENABLE | KVM_GUESTDBG_USE_SW_BP))
781 		eb |= 1u << BP_VECTOR;
782 	if (to_vmx(vcpu)->rmode.vm86_active)
783 		eb = ~0;
784 	if (enable_ept)
785 		eb &= ~(1u << PF_VECTOR);
786 
787 	/* When we are running a nested L2 guest and L1 specified for it a
788 	 * certain exception bitmap, we must trap the same exceptions and pass
789 	 * them to L1. When running L2, we will only handle the exceptions
790 	 * specified above if L1 did not want them.
791 	 */
792 	if (is_guest_mode(vcpu))
793 		eb |= get_vmcs12(vcpu)->exception_bitmap;
794 
795 	vmcs_write32(EXCEPTION_BITMAP, eb);
796 }
797 
798 /*
799  * Check if MSR is intercepted for currently loaded MSR bitmap.
800  */
801 static bool msr_write_intercepted(struct kvm_vcpu *vcpu, u32 msr)
802 {
803 	unsigned long *msr_bitmap;
804 	int f = sizeof(unsigned long);
805 
806 	if (!cpu_has_vmx_msr_bitmap())
807 		return true;
808 
809 	msr_bitmap = to_vmx(vcpu)->loaded_vmcs->msr_bitmap;
810 
811 	if (msr <= 0x1fff) {
812 		return !!test_bit(msr, msr_bitmap + 0x800 / f);
813 	} else if ((msr >= 0xc0000000) && (msr <= 0xc0001fff)) {
814 		msr &= 0x1fff;
815 		return !!test_bit(msr, msr_bitmap + 0xc00 / f);
816 	}
817 
818 	return true;
819 }
820 
821 static void clear_atomic_switch_msr_special(struct vcpu_vmx *vmx,
822 		unsigned long entry, unsigned long exit)
823 {
824 	vm_entry_controls_clearbit(vmx, entry);
825 	vm_exit_controls_clearbit(vmx, exit);
826 }
827 
828 int vmx_find_msr_index(struct vmx_msrs *m, u32 msr)
829 {
830 	unsigned int i;
831 
832 	for (i = 0; i < m->nr; ++i) {
833 		if (m->val[i].index == msr)
834 			return i;
835 	}
836 	return -ENOENT;
837 }
838 
839 static void clear_atomic_switch_msr(struct vcpu_vmx *vmx, unsigned msr)
840 {
841 	int i;
842 	struct msr_autoload *m = &vmx->msr_autoload;
843 
844 	switch (msr) {
845 	case MSR_EFER:
846 		if (cpu_has_load_ia32_efer()) {
847 			clear_atomic_switch_msr_special(vmx,
848 					VM_ENTRY_LOAD_IA32_EFER,
849 					VM_EXIT_LOAD_IA32_EFER);
850 			return;
851 		}
852 		break;
853 	case MSR_CORE_PERF_GLOBAL_CTRL:
854 		if (cpu_has_load_perf_global_ctrl()) {
855 			clear_atomic_switch_msr_special(vmx,
856 					VM_ENTRY_LOAD_IA32_PERF_GLOBAL_CTRL,
857 					VM_EXIT_LOAD_IA32_PERF_GLOBAL_CTRL);
858 			return;
859 		}
860 		break;
861 	}
862 	i = vmx_find_msr_index(&m->guest, msr);
863 	if (i < 0)
864 		goto skip_guest;
865 	--m->guest.nr;
866 	m->guest.val[i] = m->guest.val[m->guest.nr];
867 	vmcs_write32(VM_ENTRY_MSR_LOAD_COUNT, m->guest.nr);
868 
869 skip_guest:
870 	i = vmx_find_msr_index(&m->host, msr);
871 	if (i < 0)
872 		return;
873 
874 	--m->host.nr;
875 	m->host.val[i] = m->host.val[m->host.nr];
876 	vmcs_write32(VM_EXIT_MSR_LOAD_COUNT, m->host.nr);
877 }
878 
879 static void add_atomic_switch_msr_special(struct vcpu_vmx *vmx,
880 		unsigned long entry, unsigned long exit,
881 		unsigned long guest_val_vmcs, unsigned long host_val_vmcs,
882 		u64 guest_val, u64 host_val)
883 {
884 	vmcs_write64(guest_val_vmcs, guest_val);
885 	if (host_val_vmcs != HOST_IA32_EFER)
886 		vmcs_write64(host_val_vmcs, host_val);
887 	vm_entry_controls_setbit(vmx, entry);
888 	vm_exit_controls_setbit(vmx, exit);
889 }
890 
891 static void add_atomic_switch_msr(struct vcpu_vmx *vmx, unsigned msr,
892 				  u64 guest_val, u64 host_val, bool entry_only)
893 {
894 	int i, j = 0;
895 	struct msr_autoload *m = &vmx->msr_autoload;
896 
897 	switch (msr) {
898 	case MSR_EFER:
899 		if (cpu_has_load_ia32_efer()) {
900 			add_atomic_switch_msr_special(vmx,
901 					VM_ENTRY_LOAD_IA32_EFER,
902 					VM_EXIT_LOAD_IA32_EFER,
903 					GUEST_IA32_EFER,
904 					HOST_IA32_EFER,
905 					guest_val, host_val);
906 			return;
907 		}
908 		break;
909 	case MSR_CORE_PERF_GLOBAL_CTRL:
910 		if (cpu_has_load_perf_global_ctrl()) {
911 			add_atomic_switch_msr_special(vmx,
912 					VM_ENTRY_LOAD_IA32_PERF_GLOBAL_CTRL,
913 					VM_EXIT_LOAD_IA32_PERF_GLOBAL_CTRL,
914 					GUEST_IA32_PERF_GLOBAL_CTRL,
915 					HOST_IA32_PERF_GLOBAL_CTRL,
916 					guest_val, host_val);
917 			return;
918 		}
919 		break;
920 	case MSR_IA32_PEBS_ENABLE:
921 		/* PEBS needs a quiescent period after being disabled (to write
922 		 * a record).  Disabling PEBS through VMX MSR swapping doesn't
923 		 * provide that period, so a CPU could write host's record into
924 		 * guest's memory.
925 		 */
926 		wrmsrl(MSR_IA32_PEBS_ENABLE, 0);
927 	}
928 
929 	i = vmx_find_msr_index(&m->guest, msr);
930 	if (!entry_only)
931 		j = vmx_find_msr_index(&m->host, msr);
932 
933 	if ((i < 0 && m->guest.nr == NR_LOADSTORE_MSRS) ||
934 		(j < 0 &&  m->host.nr == NR_LOADSTORE_MSRS)) {
935 		printk_once(KERN_WARNING "Not enough msr switch entries. "
936 				"Can't add msr %x\n", msr);
937 		return;
938 	}
939 	if (i < 0) {
940 		i = m->guest.nr++;
941 		vmcs_write32(VM_ENTRY_MSR_LOAD_COUNT, m->guest.nr);
942 	}
943 	m->guest.val[i].index = msr;
944 	m->guest.val[i].value = guest_val;
945 
946 	if (entry_only)
947 		return;
948 
949 	if (j < 0) {
950 		j = m->host.nr++;
951 		vmcs_write32(VM_EXIT_MSR_LOAD_COUNT, m->host.nr);
952 	}
953 	m->host.val[j].index = msr;
954 	m->host.val[j].value = host_val;
955 }
956 
957 static bool update_transition_efer(struct vcpu_vmx *vmx, int efer_offset)
958 {
959 	u64 guest_efer = vmx->vcpu.arch.efer;
960 	u64 ignore_bits = 0;
961 
962 	/* Shadow paging assumes NX to be available.  */
963 	if (!enable_ept)
964 		guest_efer |= EFER_NX;
965 
966 	/*
967 	 * LMA and LME handled by hardware; SCE meaningless outside long mode.
968 	 */
969 	ignore_bits |= EFER_SCE;
970 #ifdef CONFIG_X86_64
971 	ignore_bits |= EFER_LMA | EFER_LME;
972 	/* SCE is meaningful only in long mode on Intel */
973 	if (guest_efer & EFER_LMA)
974 		ignore_bits &= ~(u64)EFER_SCE;
975 #endif
976 
977 	/*
978 	 * On EPT, we can't emulate NX, so we must switch EFER atomically.
979 	 * On CPUs that support "load IA32_EFER", always switch EFER
980 	 * atomically, since it's faster than switching it manually.
981 	 */
982 	if (cpu_has_load_ia32_efer() ||
983 	    (enable_ept && ((vmx->vcpu.arch.efer ^ host_efer) & EFER_NX))) {
984 		if (!(guest_efer & EFER_LMA))
985 			guest_efer &= ~EFER_LME;
986 		if (guest_efer != host_efer)
987 			add_atomic_switch_msr(vmx, MSR_EFER,
988 					      guest_efer, host_efer, false);
989 		else
990 			clear_atomic_switch_msr(vmx, MSR_EFER);
991 		return false;
992 	} else {
993 		clear_atomic_switch_msr(vmx, MSR_EFER);
994 
995 		guest_efer &= ~ignore_bits;
996 		guest_efer |= host_efer & ignore_bits;
997 
998 		vmx->guest_msrs[efer_offset].data = guest_efer;
999 		vmx->guest_msrs[efer_offset].mask = ~ignore_bits;
1000 
1001 		return true;
1002 	}
1003 }
1004 
1005 #ifdef CONFIG_X86_32
1006 /*
1007  * On 32-bit kernels, VM exits still load the FS and GS bases from the
1008  * VMCS rather than the segment table.  KVM uses this helper to figure
1009  * out the current bases to poke them into the VMCS before entry.
1010  */
1011 static unsigned long segment_base(u16 selector)
1012 {
1013 	struct desc_struct *table;
1014 	unsigned long v;
1015 
1016 	if (!(selector & ~SEGMENT_RPL_MASK))
1017 		return 0;
1018 
1019 	table = get_current_gdt_ro();
1020 
1021 	if ((selector & SEGMENT_TI_MASK) == SEGMENT_LDT) {
1022 		u16 ldt_selector = kvm_read_ldt();
1023 
1024 		if (!(ldt_selector & ~SEGMENT_RPL_MASK))
1025 			return 0;
1026 
1027 		table = (struct desc_struct *)segment_base(ldt_selector);
1028 	}
1029 	v = get_desc_base(&table[selector >> 3]);
1030 	return v;
1031 }
1032 #endif
1033 
1034 static inline bool pt_can_write_msr(struct vcpu_vmx *vmx)
1035 {
1036 	return vmx_pt_mode_is_host_guest() &&
1037 	       !(vmx->pt_desc.guest.ctl & RTIT_CTL_TRACEEN);
1038 }
1039 
1040 static inline void pt_load_msr(struct pt_ctx *ctx, u32 addr_range)
1041 {
1042 	u32 i;
1043 
1044 	wrmsrl(MSR_IA32_RTIT_STATUS, ctx->status);
1045 	wrmsrl(MSR_IA32_RTIT_OUTPUT_BASE, ctx->output_base);
1046 	wrmsrl(MSR_IA32_RTIT_OUTPUT_MASK, ctx->output_mask);
1047 	wrmsrl(MSR_IA32_RTIT_CR3_MATCH, ctx->cr3_match);
1048 	for (i = 0; i < addr_range; i++) {
1049 		wrmsrl(MSR_IA32_RTIT_ADDR0_A + i * 2, ctx->addr_a[i]);
1050 		wrmsrl(MSR_IA32_RTIT_ADDR0_B + i * 2, ctx->addr_b[i]);
1051 	}
1052 }
1053 
1054 static inline void pt_save_msr(struct pt_ctx *ctx, u32 addr_range)
1055 {
1056 	u32 i;
1057 
1058 	rdmsrl(MSR_IA32_RTIT_STATUS, ctx->status);
1059 	rdmsrl(MSR_IA32_RTIT_OUTPUT_BASE, ctx->output_base);
1060 	rdmsrl(MSR_IA32_RTIT_OUTPUT_MASK, ctx->output_mask);
1061 	rdmsrl(MSR_IA32_RTIT_CR3_MATCH, ctx->cr3_match);
1062 	for (i = 0; i < addr_range; i++) {
1063 		rdmsrl(MSR_IA32_RTIT_ADDR0_A + i * 2, ctx->addr_a[i]);
1064 		rdmsrl(MSR_IA32_RTIT_ADDR0_B + i * 2, ctx->addr_b[i]);
1065 	}
1066 }
1067 
1068 static void pt_guest_enter(struct vcpu_vmx *vmx)
1069 {
1070 	if (vmx_pt_mode_is_system())
1071 		return;
1072 
1073 	/*
1074 	 * GUEST_IA32_RTIT_CTL is already set in the VMCS.
1075 	 * Save host state before VM entry.
1076 	 */
1077 	rdmsrl(MSR_IA32_RTIT_CTL, vmx->pt_desc.host.ctl);
1078 	if (vmx->pt_desc.guest.ctl & RTIT_CTL_TRACEEN) {
1079 		wrmsrl(MSR_IA32_RTIT_CTL, 0);
1080 		pt_save_msr(&vmx->pt_desc.host, vmx->pt_desc.addr_range);
1081 		pt_load_msr(&vmx->pt_desc.guest, vmx->pt_desc.addr_range);
1082 	}
1083 }
1084 
1085 static void pt_guest_exit(struct vcpu_vmx *vmx)
1086 {
1087 	if (vmx_pt_mode_is_system())
1088 		return;
1089 
1090 	if (vmx->pt_desc.guest.ctl & RTIT_CTL_TRACEEN) {
1091 		pt_save_msr(&vmx->pt_desc.guest, vmx->pt_desc.addr_range);
1092 		pt_load_msr(&vmx->pt_desc.host, vmx->pt_desc.addr_range);
1093 	}
1094 
1095 	/* Reload host state (IA32_RTIT_CTL will be cleared on VM exit). */
1096 	wrmsrl(MSR_IA32_RTIT_CTL, vmx->pt_desc.host.ctl);
1097 }
1098 
1099 void vmx_set_host_fs_gs(struct vmcs_host_state *host, u16 fs_sel, u16 gs_sel,
1100 			unsigned long fs_base, unsigned long gs_base)
1101 {
1102 	if (unlikely(fs_sel != host->fs_sel)) {
1103 		if (!(fs_sel & 7))
1104 			vmcs_write16(HOST_FS_SELECTOR, fs_sel);
1105 		else
1106 			vmcs_write16(HOST_FS_SELECTOR, 0);
1107 		host->fs_sel = fs_sel;
1108 	}
1109 	if (unlikely(gs_sel != host->gs_sel)) {
1110 		if (!(gs_sel & 7))
1111 			vmcs_write16(HOST_GS_SELECTOR, gs_sel);
1112 		else
1113 			vmcs_write16(HOST_GS_SELECTOR, 0);
1114 		host->gs_sel = gs_sel;
1115 	}
1116 	if (unlikely(fs_base != host->fs_base)) {
1117 		vmcs_writel(HOST_FS_BASE, fs_base);
1118 		host->fs_base = fs_base;
1119 	}
1120 	if (unlikely(gs_base != host->gs_base)) {
1121 		vmcs_writel(HOST_GS_BASE, gs_base);
1122 		host->gs_base = gs_base;
1123 	}
1124 }
1125 
1126 void vmx_prepare_switch_to_guest(struct kvm_vcpu *vcpu)
1127 {
1128 	struct vcpu_vmx *vmx = to_vmx(vcpu);
1129 	struct vmcs_host_state *host_state;
1130 #ifdef CONFIG_X86_64
1131 	int cpu = raw_smp_processor_id();
1132 #endif
1133 	unsigned long fs_base, gs_base;
1134 	u16 fs_sel, gs_sel;
1135 	int i;
1136 
1137 	vmx->req_immediate_exit = false;
1138 
1139 	/*
1140 	 * Note that guest MSRs to be saved/restored can also be changed
1141 	 * when guest state is loaded. This happens when guest transitions
1142 	 * to/from long-mode by setting MSR_EFER.LMA.
1143 	 */
1144 	if (!vmx->guest_msrs_ready) {
1145 		vmx->guest_msrs_ready = true;
1146 		for (i = 0; i < vmx->save_nmsrs; ++i)
1147 			kvm_set_shared_msr(vmx->guest_msrs[i].index,
1148 					   vmx->guest_msrs[i].data,
1149 					   vmx->guest_msrs[i].mask);
1150 
1151 	}
1152 
1153     	if (vmx->nested.need_vmcs12_to_shadow_sync)
1154 		nested_sync_vmcs12_to_shadow(vcpu);
1155 
1156 	if (vmx->guest_state_loaded)
1157 		return;
1158 
1159 	host_state = &vmx->loaded_vmcs->host_state;
1160 
1161 	/*
1162 	 * Set host fs and gs selectors.  Unfortunately, 22.2.3 does not
1163 	 * allow segment selectors with cpl > 0 or ti == 1.
1164 	 */
1165 	host_state->ldt_sel = kvm_read_ldt();
1166 
1167 #ifdef CONFIG_X86_64
1168 	savesegment(ds, host_state->ds_sel);
1169 	savesegment(es, host_state->es_sel);
1170 
1171 	gs_base = cpu_kernelmode_gs_base(cpu);
1172 	if (likely(is_64bit_mm(current->mm))) {
1173 		current_save_fsgs();
1174 		fs_sel = current->thread.fsindex;
1175 		gs_sel = current->thread.gsindex;
1176 		fs_base = current->thread.fsbase;
1177 		vmx->msr_host_kernel_gs_base = current->thread.gsbase;
1178 	} else {
1179 		savesegment(fs, fs_sel);
1180 		savesegment(gs, gs_sel);
1181 		fs_base = read_msr(MSR_FS_BASE);
1182 		vmx->msr_host_kernel_gs_base = read_msr(MSR_KERNEL_GS_BASE);
1183 	}
1184 
1185 	wrmsrl(MSR_KERNEL_GS_BASE, vmx->msr_guest_kernel_gs_base);
1186 #else
1187 	savesegment(fs, fs_sel);
1188 	savesegment(gs, gs_sel);
1189 	fs_base = segment_base(fs_sel);
1190 	gs_base = segment_base(gs_sel);
1191 #endif
1192 
1193 	vmx_set_host_fs_gs(host_state, fs_sel, gs_sel, fs_base, gs_base);
1194 	vmx->guest_state_loaded = true;
1195 }
1196 
1197 static void vmx_prepare_switch_to_host(struct vcpu_vmx *vmx)
1198 {
1199 	struct vmcs_host_state *host_state;
1200 
1201 	if (!vmx->guest_state_loaded)
1202 		return;
1203 
1204 	host_state = &vmx->loaded_vmcs->host_state;
1205 
1206 	++vmx->vcpu.stat.host_state_reload;
1207 
1208 #ifdef CONFIG_X86_64
1209 	rdmsrl(MSR_KERNEL_GS_BASE, vmx->msr_guest_kernel_gs_base);
1210 #endif
1211 	if (host_state->ldt_sel || (host_state->gs_sel & 7)) {
1212 		kvm_load_ldt(host_state->ldt_sel);
1213 #ifdef CONFIG_X86_64
1214 		load_gs_index(host_state->gs_sel);
1215 #else
1216 		loadsegment(gs, host_state->gs_sel);
1217 #endif
1218 	}
1219 	if (host_state->fs_sel & 7)
1220 		loadsegment(fs, host_state->fs_sel);
1221 #ifdef CONFIG_X86_64
1222 	if (unlikely(host_state->ds_sel | host_state->es_sel)) {
1223 		loadsegment(ds, host_state->ds_sel);
1224 		loadsegment(es, host_state->es_sel);
1225 	}
1226 #endif
1227 	invalidate_tss_limit();
1228 #ifdef CONFIG_X86_64
1229 	wrmsrl(MSR_KERNEL_GS_BASE, vmx->msr_host_kernel_gs_base);
1230 #endif
1231 	load_fixmap_gdt(raw_smp_processor_id());
1232 	vmx->guest_state_loaded = false;
1233 	vmx->guest_msrs_ready = false;
1234 }
1235 
1236 #ifdef CONFIG_X86_64
1237 static u64 vmx_read_guest_kernel_gs_base(struct vcpu_vmx *vmx)
1238 {
1239 	preempt_disable();
1240 	if (vmx->guest_state_loaded)
1241 		rdmsrl(MSR_KERNEL_GS_BASE, vmx->msr_guest_kernel_gs_base);
1242 	preempt_enable();
1243 	return vmx->msr_guest_kernel_gs_base;
1244 }
1245 
1246 static void vmx_write_guest_kernel_gs_base(struct vcpu_vmx *vmx, u64 data)
1247 {
1248 	preempt_disable();
1249 	if (vmx->guest_state_loaded)
1250 		wrmsrl(MSR_KERNEL_GS_BASE, data);
1251 	preempt_enable();
1252 	vmx->msr_guest_kernel_gs_base = data;
1253 }
1254 #endif
1255 
1256 static void vmx_vcpu_pi_load(struct kvm_vcpu *vcpu, int cpu)
1257 {
1258 	struct pi_desc *pi_desc = vcpu_to_pi_desc(vcpu);
1259 	struct pi_desc old, new;
1260 	unsigned int dest;
1261 
1262 	/*
1263 	 * In case of hot-plug or hot-unplug, we may have to undo
1264 	 * vmx_vcpu_pi_put even if there is no assigned device.  And we
1265 	 * always keep PI.NDST up to date for simplicity: it makes the
1266 	 * code easier, and CPU migration is not a fast path.
1267 	 */
1268 	if (!pi_test_sn(pi_desc) && vcpu->cpu == cpu)
1269 		return;
1270 
1271 	/*
1272 	 * If the 'nv' field is POSTED_INTR_WAKEUP_VECTOR, do not change
1273 	 * PI.NDST: pi_post_block is the one expected to change PID.NDST and the
1274 	 * wakeup handler expects the vCPU to be on the blocked_vcpu_list that
1275 	 * matches PI.NDST. Otherwise, a vcpu may not be able to be woken up
1276 	 * correctly.
1277 	 */
1278 	if (pi_desc->nv == POSTED_INTR_WAKEUP_VECTOR || vcpu->cpu == cpu) {
1279 		pi_clear_sn(pi_desc);
1280 		goto after_clear_sn;
1281 	}
1282 
1283 	/* The full case.  */
1284 	do {
1285 		old.control = new.control = pi_desc->control;
1286 
1287 		dest = cpu_physical_id(cpu);
1288 
1289 		if (x2apic_enabled())
1290 			new.ndst = dest;
1291 		else
1292 			new.ndst = (dest << 8) & 0xFF00;
1293 
1294 		new.sn = 0;
1295 	} while (cmpxchg64(&pi_desc->control, old.control,
1296 			   new.control) != old.control);
1297 
1298 after_clear_sn:
1299 
1300 	/*
1301 	 * Clear SN before reading the bitmap.  The VT-d firmware
1302 	 * writes the bitmap and reads SN atomically (5.2.3 in the
1303 	 * spec), so it doesn't really have a memory barrier that
1304 	 * pairs with this, but we cannot do that and we need one.
1305 	 */
1306 	smp_mb__after_atomic();
1307 
1308 	if (!pi_is_pir_empty(pi_desc))
1309 		pi_set_on(pi_desc);
1310 }
1311 
1312 void vmx_vcpu_load_vmcs(struct kvm_vcpu *vcpu, int cpu,
1313 			struct loaded_vmcs *buddy)
1314 {
1315 	struct vcpu_vmx *vmx = to_vmx(vcpu);
1316 	bool already_loaded = vmx->loaded_vmcs->cpu == cpu;
1317 	struct vmcs *prev;
1318 
1319 	if (!already_loaded) {
1320 		loaded_vmcs_clear(vmx->loaded_vmcs);
1321 		local_irq_disable();
1322 
1323 		/*
1324 		 * Ensure loaded_vmcs->cpu is read before adding loaded_vmcs to
1325 		 * this cpu's percpu list, otherwise it may not yet be deleted
1326 		 * from its previous cpu's percpu list.  Pairs with the
1327 		 * smb_wmb() in __loaded_vmcs_clear().
1328 		 */
1329 		smp_rmb();
1330 
1331 		list_add(&vmx->loaded_vmcs->loaded_vmcss_on_cpu_link,
1332 			 &per_cpu(loaded_vmcss_on_cpu, cpu));
1333 		local_irq_enable();
1334 	}
1335 
1336 	prev = per_cpu(current_vmcs, cpu);
1337 	if (prev != vmx->loaded_vmcs->vmcs) {
1338 		per_cpu(current_vmcs, cpu) = vmx->loaded_vmcs->vmcs;
1339 		vmcs_load(vmx->loaded_vmcs->vmcs);
1340 
1341 		/*
1342 		 * No indirect branch prediction barrier needed when switching
1343 		 * the active VMCS within a guest, e.g. on nested VM-Enter.
1344 		 * The L1 VMM can protect itself with retpolines, IBPB or IBRS.
1345 		 */
1346 		if (!buddy || WARN_ON_ONCE(buddy->vmcs != prev))
1347 			indirect_branch_prediction_barrier();
1348 	}
1349 
1350 	if (!already_loaded) {
1351 		void *gdt = get_current_gdt_ro();
1352 		unsigned long sysenter_esp;
1353 
1354 		/*
1355 		 * Flush all EPTP/VPID contexts, the new pCPU may have stale
1356 		 * TLB entries from its previous association with the vCPU.
1357 		 */
1358 		kvm_make_request(KVM_REQ_TLB_FLUSH, vcpu);
1359 
1360 		/*
1361 		 * Linux uses per-cpu TSS and GDT, so set these when switching
1362 		 * processors.  See 22.2.4.
1363 		 */
1364 		vmcs_writel(HOST_TR_BASE,
1365 			    (unsigned long)&get_cpu_entry_area(cpu)->tss.x86_tss);
1366 		vmcs_writel(HOST_GDTR_BASE, (unsigned long)gdt);   /* 22.2.4 */
1367 
1368 		rdmsrl(MSR_IA32_SYSENTER_ESP, sysenter_esp);
1369 		vmcs_writel(HOST_IA32_SYSENTER_ESP, sysenter_esp); /* 22.2.3 */
1370 
1371 		vmx->loaded_vmcs->cpu = cpu;
1372 	}
1373 
1374 	/* Setup TSC multiplier */
1375 	if (kvm_has_tsc_control &&
1376 	    vmx->current_tsc_ratio != vcpu->arch.tsc_scaling_ratio)
1377 		decache_tsc_multiplier(vmx);
1378 }
1379 
1380 /*
1381  * Switches to specified vcpu, until a matching vcpu_put(), but assumes
1382  * vcpu mutex is already taken.
1383  */
1384 static void vmx_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
1385 {
1386 	struct vcpu_vmx *vmx = to_vmx(vcpu);
1387 
1388 	vmx_vcpu_load_vmcs(vcpu, cpu, NULL);
1389 
1390 	vmx_vcpu_pi_load(vcpu, cpu);
1391 
1392 	vmx->host_debugctlmsr = get_debugctlmsr();
1393 }
1394 
1395 static void vmx_vcpu_pi_put(struct kvm_vcpu *vcpu)
1396 {
1397 	struct pi_desc *pi_desc = vcpu_to_pi_desc(vcpu);
1398 
1399 	if (!kvm_arch_has_assigned_device(vcpu->kvm) ||
1400 		!irq_remapping_cap(IRQ_POSTING_CAP)  ||
1401 		!kvm_vcpu_apicv_active(vcpu))
1402 		return;
1403 
1404 	/* Set SN when the vCPU is preempted */
1405 	if (vcpu->preempted)
1406 		pi_set_sn(pi_desc);
1407 }
1408 
1409 static void vmx_vcpu_put(struct kvm_vcpu *vcpu)
1410 {
1411 	vmx_vcpu_pi_put(vcpu);
1412 
1413 	vmx_prepare_switch_to_host(to_vmx(vcpu));
1414 }
1415 
1416 static bool emulation_required(struct kvm_vcpu *vcpu)
1417 {
1418 	return emulate_invalid_guest_state && !guest_state_valid(vcpu);
1419 }
1420 
1421 unsigned long vmx_get_rflags(struct kvm_vcpu *vcpu)
1422 {
1423 	struct vcpu_vmx *vmx = to_vmx(vcpu);
1424 	unsigned long rflags, save_rflags;
1425 
1426 	if (!kvm_register_is_available(vcpu, VCPU_EXREG_RFLAGS)) {
1427 		kvm_register_mark_available(vcpu, VCPU_EXREG_RFLAGS);
1428 		rflags = vmcs_readl(GUEST_RFLAGS);
1429 		if (vmx->rmode.vm86_active) {
1430 			rflags &= RMODE_GUEST_OWNED_EFLAGS_BITS;
1431 			save_rflags = vmx->rmode.save_rflags;
1432 			rflags |= save_rflags & ~RMODE_GUEST_OWNED_EFLAGS_BITS;
1433 		}
1434 		vmx->rflags = rflags;
1435 	}
1436 	return vmx->rflags;
1437 }
1438 
1439 void vmx_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags)
1440 {
1441 	struct vcpu_vmx *vmx = to_vmx(vcpu);
1442 	unsigned long old_rflags;
1443 
1444 	if (enable_unrestricted_guest) {
1445 		kvm_register_mark_available(vcpu, VCPU_EXREG_RFLAGS);
1446 		vmx->rflags = rflags;
1447 		vmcs_writel(GUEST_RFLAGS, rflags);
1448 		return;
1449 	}
1450 
1451 	old_rflags = vmx_get_rflags(vcpu);
1452 	vmx->rflags = rflags;
1453 	if (vmx->rmode.vm86_active) {
1454 		vmx->rmode.save_rflags = rflags;
1455 		rflags |= X86_EFLAGS_IOPL | X86_EFLAGS_VM;
1456 	}
1457 	vmcs_writel(GUEST_RFLAGS, rflags);
1458 
1459 	if ((old_rflags ^ vmx->rflags) & X86_EFLAGS_VM)
1460 		vmx->emulation_required = emulation_required(vcpu);
1461 }
1462 
1463 u32 vmx_get_interrupt_shadow(struct kvm_vcpu *vcpu)
1464 {
1465 	u32 interruptibility = vmcs_read32(GUEST_INTERRUPTIBILITY_INFO);
1466 	int ret = 0;
1467 
1468 	if (interruptibility & GUEST_INTR_STATE_STI)
1469 		ret |= KVM_X86_SHADOW_INT_STI;
1470 	if (interruptibility & GUEST_INTR_STATE_MOV_SS)
1471 		ret |= KVM_X86_SHADOW_INT_MOV_SS;
1472 
1473 	return ret;
1474 }
1475 
1476 void vmx_set_interrupt_shadow(struct kvm_vcpu *vcpu, int mask)
1477 {
1478 	u32 interruptibility_old = vmcs_read32(GUEST_INTERRUPTIBILITY_INFO);
1479 	u32 interruptibility = interruptibility_old;
1480 
1481 	interruptibility &= ~(GUEST_INTR_STATE_STI | GUEST_INTR_STATE_MOV_SS);
1482 
1483 	if (mask & KVM_X86_SHADOW_INT_MOV_SS)
1484 		interruptibility |= GUEST_INTR_STATE_MOV_SS;
1485 	else if (mask & KVM_X86_SHADOW_INT_STI)
1486 		interruptibility |= GUEST_INTR_STATE_STI;
1487 
1488 	if ((interruptibility != interruptibility_old))
1489 		vmcs_write32(GUEST_INTERRUPTIBILITY_INFO, interruptibility);
1490 }
1491 
1492 static int vmx_rtit_ctl_check(struct kvm_vcpu *vcpu, u64 data)
1493 {
1494 	struct vcpu_vmx *vmx = to_vmx(vcpu);
1495 	unsigned long value;
1496 
1497 	/*
1498 	 * Any MSR write that attempts to change bits marked reserved will
1499 	 * case a #GP fault.
1500 	 */
1501 	if (data & vmx->pt_desc.ctl_bitmask)
1502 		return 1;
1503 
1504 	/*
1505 	 * Any attempt to modify IA32_RTIT_CTL while TraceEn is set will
1506 	 * result in a #GP unless the same write also clears TraceEn.
1507 	 */
1508 	if ((vmx->pt_desc.guest.ctl & RTIT_CTL_TRACEEN) &&
1509 		((vmx->pt_desc.guest.ctl ^ data) & ~RTIT_CTL_TRACEEN))
1510 		return 1;
1511 
1512 	/*
1513 	 * WRMSR to IA32_RTIT_CTL that sets TraceEn but clears this bit
1514 	 * and FabricEn would cause #GP, if
1515 	 * CPUID.(EAX=14H, ECX=0):ECX.SNGLRGNOUT[bit 2] = 0
1516 	 */
1517 	if ((data & RTIT_CTL_TRACEEN) && !(data & RTIT_CTL_TOPA) &&
1518 		!(data & RTIT_CTL_FABRIC_EN) &&
1519 		!intel_pt_validate_cap(vmx->pt_desc.caps,
1520 					PT_CAP_single_range_output))
1521 		return 1;
1522 
1523 	/*
1524 	 * MTCFreq, CycThresh and PSBFreq encodings check, any MSR write that
1525 	 * utilize encodings marked reserved will casue a #GP fault.
1526 	 */
1527 	value = intel_pt_validate_cap(vmx->pt_desc.caps, PT_CAP_mtc_periods);
1528 	if (intel_pt_validate_cap(vmx->pt_desc.caps, PT_CAP_mtc) &&
1529 			!test_bit((data & RTIT_CTL_MTC_RANGE) >>
1530 			RTIT_CTL_MTC_RANGE_OFFSET, &value))
1531 		return 1;
1532 	value = intel_pt_validate_cap(vmx->pt_desc.caps,
1533 						PT_CAP_cycle_thresholds);
1534 	if (intel_pt_validate_cap(vmx->pt_desc.caps, PT_CAP_psb_cyc) &&
1535 			!test_bit((data & RTIT_CTL_CYC_THRESH) >>
1536 			RTIT_CTL_CYC_THRESH_OFFSET, &value))
1537 		return 1;
1538 	value = intel_pt_validate_cap(vmx->pt_desc.caps, PT_CAP_psb_periods);
1539 	if (intel_pt_validate_cap(vmx->pt_desc.caps, PT_CAP_psb_cyc) &&
1540 			!test_bit((data & RTIT_CTL_PSB_FREQ) >>
1541 			RTIT_CTL_PSB_FREQ_OFFSET, &value))
1542 		return 1;
1543 
1544 	/*
1545 	 * If ADDRx_CFG is reserved or the encodings is >2 will
1546 	 * cause a #GP fault.
1547 	 */
1548 	value = (data & RTIT_CTL_ADDR0) >> RTIT_CTL_ADDR0_OFFSET;
1549 	if ((value && (vmx->pt_desc.addr_range < 1)) || (value > 2))
1550 		return 1;
1551 	value = (data & RTIT_CTL_ADDR1) >> RTIT_CTL_ADDR1_OFFSET;
1552 	if ((value && (vmx->pt_desc.addr_range < 2)) || (value > 2))
1553 		return 1;
1554 	value = (data & RTIT_CTL_ADDR2) >> RTIT_CTL_ADDR2_OFFSET;
1555 	if ((value && (vmx->pt_desc.addr_range < 3)) || (value > 2))
1556 		return 1;
1557 	value = (data & RTIT_CTL_ADDR3) >> RTIT_CTL_ADDR3_OFFSET;
1558 	if ((value && (vmx->pt_desc.addr_range < 4)) || (value > 2))
1559 		return 1;
1560 
1561 	return 0;
1562 }
1563 
1564 static int skip_emulated_instruction(struct kvm_vcpu *vcpu)
1565 {
1566 	unsigned long rip, orig_rip;
1567 
1568 	/*
1569 	 * Using VMCS.VM_EXIT_INSTRUCTION_LEN on EPT misconfig depends on
1570 	 * undefined behavior: Intel's SDM doesn't mandate the VMCS field be
1571 	 * set when EPT misconfig occurs.  In practice, real hardware updates
1572 	 * VM_EXIT_INSTRUCTION_LEN on EPT misconfig, but other hypervisors
1573 	 * (namely Hyper-V) don't set it due to it being undefined behavior,
1574 	 * i.e. we end up advancing IP with some random value.
1575 	 */
1576 	if (!static_cpu_has(X86_FEATURE_HYPERVISOR) ||
1577 	    to_vmx(vcpu)->exit_reason != EXIT_REASON_EPT_MISCONFIG) {
1578 		orig_rip = kvm_rip_read(vcpu);
1579 		rip = orig_rip + vmcs_read32(VM_EXIT_INSTRUCTION_LEN);
1580 #ifdef CONFIG_X86_64
1581 		/*
1582 		 * We need to mask out the high 32 bits of RIP if not in 64-bit
1583 		 * mode, but just finding out that we are in 64-bit mode is
1584 		 * quite expensive.  Only do it if there was a carry.
1585 		 */
1586 		if (unlikely(((rip ^ orig_rip) >> 31) == 3) && !is_64_bit_mode(vcpu))
1587 			rip = (u32)rip;
1588 #endif
1589 		kvm_rip_write(vcpu, rip);
1590 	} else {
1591 		if (!kvm_emulate_instruction(vcpu, EMULTYPE_SKIP))
1592 			return 0;
1593 	}
1594 
1595 	/* skipping an emulated instruction also counts */
1596 	vmx_set_interrupt_shadow(vcpu, 0);
1597 
1598 	return 1;
1599 }
1600 
1601 /*
1602  * Handles kvm_read/write_guest_virt*() result and either injects #PF or returns
1603  * KVM_EXIT_INTERNAL_ERROR for cases not currently handled by KVM. Return value
1604  * indicates whether exit to userspace is needed.
1605  */
1606 int vmx_handle_memory_failure(struct kvm_vcpu *vcpu, int r,
1607 			      struct x86_exception *e)
1608 {
1609 	if (r == X86EMUL_PROPAGATE_FAULT) {
1610 		kvm_inject_emulated_page_fault(vcpu, e);
1611 		return 1;
1612 	}
1613 
1614 	/*
1615 	 * In case kvm_read/write_guest_virt*() failed with X86EMUL_IO_NEEDED
1616 	 * while handling a VMX instruction KVM could've handled the request
1617 	 * correctly by exiting to userspace and performing I/O but there
1618 	 * doesn't seem to be a real use-case behind such requests, just return
1619 	 * KVM_EXIT_INTERNAL_ERROR for now.
1620 	 */
1621 	vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
1622 	vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_EMULATION;
1623 	vcpu->run->internal.ndata = 0;
1624 
1625 	return 0;
1626 }
1627 
1628 /*
1629  * Recognizes a pending MTF VM-exit and records the nested state for later
1630  * delivery.
1631  */
1632 static void vmx_update_emulated_instruction(struct kvm_vcpu *vcpu)
1633 {
1634 	struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
1635 	struct vcpu_vmx *vmx = to_vmx(vcpu);
1636 
1637 	if (!is_guest_mode(vcpu))
1638 		return;
1639 
1640 	/*
1641 	 * Per the SDM, MTF takes priority over debug-trap exceptions besides
1642 	 * T-bit traps. As instruction emulation is completed (i.e. at the
1643 	 * instruction boundary), any #DB exception pending delivery must be a
1644 	 * debug-trap. Record the pending MTF state to be delivered in
1645 	 * vmx_check_nested_events().
1646 	 */
1647 	if (nested_cpu_has_mtf(vmcs12) &&
1648 	    (!vcpu->arch.exception.pending ||
1649 	     vcpu->arch.exception.nr == DB_VECTOR))
1650 		vmx->nested.mtf_pending = true;
1651 	else
1652 		vmx->nested.mtf_pending = false;
1653 }
1654 
1655 static int vmx_skip_emulated_instruction(struct kvm_vcpu *vcpu)
1656 {
1657 	vmx_update_emulated_instruction(vcpu);
1658 	return skip_emulated_instruction(vcpu);
1659 }
1660 
1661 static void vmx_clear_hlt(struct kvm_vcpu *vcpu)
1662 {
1663 	/*
1664 	 * Ensure that we clear the HLT state in the VMCS.  We don't need to
1665 	 * explicitly skip the instruction because if the HLT state is set,
1666 	 * then the instruction is already executing and RIP has already been
1667 	 * advanced.
1668 	 */
1669 	if (kvm_hlt_in_guest(vcpu->kvm) &&
1670 			vmcs_read32(GUEST_ACTIVITY_STATE) == GUEST_ACTIVITY_HLT)
1671 		vmcs_write32(GUEST_ACTIVITY_STATE, GUEST_ACTIVITY_ACTIVE);
1672 }
1673 
1674 static void vmx_queue_exception(struct kvm_vcpu *vcpu)
1675 {
1676 	struct vcpu_vmx *vmx = to_vmx(vcpu);
1677 	unsigned nr = vcpu->arch.exception.nr;
1678 	bool has_error_code = vcpu->arch.exception.has_error_code;
1679 	u32 error_code = vcpu->arch.exception.error_code;
1680 	u32 intr_info = nr | INTR_INFO_VALID_MASK;
1681 
1682 	kvm_deliver_exception_payload(vcpu);
1683 
1684 	if (has_error_code) {
1685 		vmcs_write32(VM_ENTRY_EXCEPTION_ERROR_CODE, error_code);
1686 		intr_info |= INTR_INFO_DELIVER_CODE_MASK;
1687 	}
1688 
1689 	if (vmx->rmode.vm86_active) {
1690 		int inc_eip = 0;
1691 		if (kvm_exception_is_soft(nr))
1692 			inc_eip = vcpu->arch.event_exit_inst_len;
1693 		kvm_inject_realmode_interrupt(vcpu, nr, inc_eip);
1694 		return;
1695 	}
1696 
1697 	WARN_ON_ONCE(vmx->emulation_required);
1698 
1699 	if (kvm_exception_is_soft(nr)) {
1700 		vmcs_write32(VM_ENTRY_INSTRUCTION_LEN,
1701 			     vmx->vcpu.arch.event_exit_inst_len);
1702 		intr_info |= INTR_TYPE_SOFT_EXCEPTION;
1703 	} else
1704 		intr_info |= INTR_TYPE_HARD_EXCEPTION;
1705 
1706 	vmcs_write32(VM_ENTRY_INTR_INFO_FIELD, intr_info);
1707 
1708 	vmx_clear_hlt(vcpu);
1709 }
1710 
1711 /*
1712  * Swap MSR entry in host/guest MSR entry array.
1713  */
1714 static void move_msr_up(struct vcpu_vmx *vmx, int from, int to)
1715 {
1716 	struct shared_msr_entry tmp;
1717 
1718 	tmp = vmx->guest_msrs[to];
1719 	vmx->guest_msrs[to] = vmx->guest_msrs[from];
1720 	vmx->guest_msrs[from] = tmp;
1721 }
1722 
1723 /*
1724  * Set up the vmcs to automatically save and restore system
1725  * msrs.  Don't touch the 64-bit msrs if the guest is in legacy
1726  * mode, as fiddling with msrs is very expensive.
1727  */
1728 static void setup_msrs(struct vcpu_vmx *vmx)
1729 {
1730 	int save_nmsrs, index;
1731 
1732 	save_nmsrs = 0;
1733 #ifdef CONFIG_X86_64
1734 	/*
1735 	 * The SYSCALL MSRs are only needed on long mode guests, and only
1736 	 * when EFER.SCE is set.
1737 	 */
1738 	if (is_long_mode(&vmx->vcpu) && (vmx->vcpu.arch.efer & EFER_SCE)) {
1739 		index = __find_msr_index(vmx, MSR_STAR);
1740 		if (index >= 0)
1741 			move_msr_up(vmx, index, save_nmsrs++);
1742 		index = __find_msr_index(vmx, MSR_LSTAR);
1743 		if (index >= 0)
1744 			move_msr_up(vmx, index, save_nmsrs++);
1745 		index = __find_msr_index(vmx, MSR_SYSCALL_MASK);
1746 		if (index >= 0)
1747 			move_msr_up(vmx, index, save_nmsrs++);
1748 	}
1749 #endif
1750 	index = __find_msr_index(vmx, MSR_EFER);
1751 	if (index >= 0 && update_transition_efer(vmx, index))
1752 		move_msr_up(vmx, index, save_nmsrs++);
1753 	index = __find_msr_index(vmx, MSR_TSC_AUX);
1754 	if (index >= 0 && guest_cpuid_has(&vmx->vcpu, X86_FEATURE_RDTSCP))
1755 		move_msr_up(vmx, index, save_nmsrs++);
1756 	index = __find_msr_index(vmx, MSR_IA32_TSX_CTRL);
1757 	if (index >= 0)
1758 		move_msr_up(vmx, index, save_nmsrs++);
1759 
1760 	vmx->save_nmsrs = save_nmsrs;
1761 	vmx->guest_msrs_ready = false;
1762 
1763 	if (cpu_has_vmx_msr_bitmap())
1764 		vmx_update_msr_bitmap(&vmx->vcpu);
1765 }
1766 
1767 static u64 vmx_write_l1_tsc_offset(struct kvm_vcpu *vcpu, u64 offset)
1768 {
1769 	struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
1770 	u64 g_tsc_offset = 0;
1771 
1772 	/*
1773 	 * We're here if L1 chose not to trap WRMSR to TSC. According
1774 	 * to the spec, this should set L1's TSC; The offset that L1
1775 	 * set for L2 remains unchanged, and still needs to be added
1776 	 * to the newly set TSC to get L2's TSC.
1777 	 */
1778 	if (is_guest_mode(vcpu) &&
1779 	    (vmcs12->cpu_based_vm_exec_control & CPU_BASED_USE_TSC_OFFSETTING))
1780 		g_tsc_offset = vmcs12->tsc_offset;
1781 
1782 	trace_kvm_write_tsc_offset(vcpu->vcpu_id,
1783 				   vcpu->arch.tsc_offset - g_tsc_offset,
1784 				   offset);
1785 	vmcs_write64(TSC_OFFSET, offset + g_tsc_offset);
1786 	return offset + g_tsc_offset;
1787 }
1788 
1789 /*
1790  * nested_vmx_allowed() checks whether a guest should be allowed to use VMX
1791  * instructions and MSRs (i.e., nested VMX). Nested VMX is disabled for
1792  * all guests if the "nested" module option is off, and can also be disabled
1793  * for a single guest by disabling its VMX cpuid bit.
1794  */
1795 bool nested_vmx_allowed(struct kvm_vcpu *vcpu)
1796 {
1797 	return nested && guest_cpuid_has(vcpu, X86_FEATURE_VMX);
1798 }
1799 
1800 static inline bool vmx_feature_control_msr_valid(struct kvm_vcpu *vcpu,
1801 						 uint64_t val)
1802 {
1803 	uint64_t valid_bits = to_vmx(vcpu)->msr_ia32_feature_control_valid_bits;
1804 
1805 	return !(val & ~valid_bits);
1806 }
1807 
1808 static int vmx_get_msr_feature(struct kvm_msr_entry *msr)
1809 {
1810 	switch (msr->index) {
1811 	case MSR_IA32_VMX_BASIC ... MSR_IA32_VMX_VMFUNC:
1812 		if (!nested)
1813 			return 1;
1814 		return vmx_get_vmx_msr(&vmcs_config.nested, msr->index, &msr->data);
1815 	case MSR_IA32_PERF_CAPABILITIES:
1816 		msr->data = vmx_get_perf_capabilities();
1817 		return 0;
1818 	default:
1819 		return 1;
1820 	}
1821 }
1822 
1823 /*
1824  * Reads an msr value (of 'msr_index') into 'pdata'.
1825  * Returns 0 on success, non-0 otherwise.
1826  * Assumes vcpu_load() was already called.
1827  */
1828 static int vmx_get_msr(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
1829 {
1830 	struct vcpu_vmx *vmx = to_vmx(vcpu);
1831 	struct shared_msr_entry *msr;
1832 	u32 index;
1833 
1834 	switch (msr_info->index) {
1835 #ifdef CONFIG_X86_64
1836 	case MSR_FS_BASE:
1837 		msr_info->data = vmcs_readl(GUEST_FS_BASE);
1838 		break;
1839 	case MSR_GS_BASE:
1840 		msr_info->data = vmcs_readl(GUEST_GS_BASE);
1841 		break;
1842 	case MSR_KERNEL_GS_BASE:
1843 		msr_info->data = vmx_read_guest_kernel_gs_base(vmx);
1844 		break;
1845 #endif
1846 	case MSR_EFER:
1847 		return kvm_get_msr_common(vcpu, msr_info);
1848 	case MSR_IA32_TSX_CTRL:
1849 		if (!msr_info->host_initiated &&
1850 		    !(vcpu->arch.arch_capabilities & ARCH_CAP_TSX_CTRL_MSR))
1851 			return 1;
1852 		goto find_shared_msr;
1853 	case MSR_IA32_UMWAIT_CONTROL:
1854 		if (!msr_info->host_initiated && !vmx_has_waitpkg(vmx))
1855 			return 1;
1856 
1857 		msr_info->data = vmx->msr_ia32_umwait_control;
1858 		break;
1859 	case MSR_IA32_SPEC_CTRL:
1860 		if (!msr_info->host_initiated &&
1861 		    !guest_cpuid_has(vcpu, X86_FEATURE_SPEC_CTRL))
1862 			return 1;
1863 
1864 		msr_info->data = to_vmx(vcpu)->spec_ctrl;
1865 		break;
1866 	case MSR_IA32_SYSENTER_CS:
1867 		msr_info->data = vmcs_read32(GUEST_SYSENTER_CS);
1868 		break;
1869 	case MSR_IA32_SYSENTER_EIP:
1870 		msr_info->data = vmcs_readl(GUEST_SYSENTER_EIP);
1871 		break;
1872 	case MSR_IA32_SYSENTER_ESP:
1873 		msr_info->data = vmcs_readl(GUEST_SYSENTER_ESP);
1874 		break;
1875 	case MSR_IA32_BNDCFGS:
1876 		if (!kvm_mpx_supported() ||
1877 		    (!msr_info->host_initiated &&
1878 		     !guest_cpuid_has(vcpu, X86_FEATURE_MPX)))
1879 			return 1;
1880 		msr_info->data = vmcs_read64(GUEST_BNDCFGS);
1881 		break;
1882 	case MSR_IA32_MCG_EXT_CTL:
1883 		if (!msr_info->host_initiated &&
1884 		    !(vmx->msr_ia32_feature_control &
1885 		      FEAT_CTL_LMCE_ENABLED))
1886 			return 1;
1887 		msr_info->data = vcpu->arch.mcg_ext_ctl;
1888 		break;
1889 	case MSR_IA32_FEAT_CTL:
1890 		msr_info->data = vmx->msr_ia32_feature_control;
1891 		break;
1892 	case MSR_IA32_VMX_BASIC ... MSR_IA32_VMX_VMFUNC:
1893 		if (!nested_vmx_allowed(vcpu))
1894 			return 1;
1895 		if (vmx_get_vmx_msr(&vmx->nested.msrs, msr_info->index,
1896 				    &msr_info->data))
1897 			return 1;
1898 		/*
1899 		 * Enlightened VMCS v1 doesn't have certain fields, but buggy
1900 		 * Hyper-V versions are still trying to use corresponding
1901 		 * features when they are exposed. Filter out the essential
1902 		 * minimum.
1903 		 */
1904 		if (!msr_info->host_initiated &&
1905 		    vmx->nested.enlightened_vmcs_enabled)
1906 			nested_evmcs_filter_control_msr(msr_info->index,
1907 							&msr_info->data);
1908 		break;
1909 	case MSR_IA32_RTIT_CTL:
1910 		if (!vmx_pt_mode_is_host_guest())
1911 			return 1;
1912 		msr_info->data = vmx->pt_desc.guest.ctl;
1913 		break;
1914 	case MSR_IA32_RTIT_STATUS:
1915 		if (!vmx_pt_mode_is_host_guest())
1916 			return 1;
1917 		msr_info->data = vmx->pt_desc.guest.status;
1918 		break;
1919 	case MSR_IA32_RTIT_CR3_MATCH:
1920 		if (!vmx_pt_mode_is_host_guest() ||
1921 			!intel_pt_validate_cap(vmx->pt_desc.caps,
1922 						PT_CAP_cr3_filtering))
1923 			return 1;
1924 		msr_info->data = vmx->pt_desc.guest.cr3_match;
1925 		break;
1926 	case MSR_IA32_RTIT_OUTPUT_BASE:
1927 		if (!vmx_pt_mode_is_host_guest() ||
1928 			(!intel_pt_validate_cap(vmx->pt_desc.caps,
1929 					PT_CAP_topa_output) &&
1930 			 !intel_pt_validate_cap(vmx->pt_desc.caps,
1931 					PT_CAP_single_range_output)))
1932 			return 1;
1933 		msr_info->data = vmx->pt_desc.guest.output_base;
1934 		break;
1935 	case MSR_IA32_RTIT_OUTPUT_MASK:
1936 		if (!vmx_pt_mode_is_host_guest() ||
1937 			(!intel_pt_validate_cap(vmx->pt_desc.caps,
1938 					PT_CAP_topa_output) &&
1939 			 !intel_pt_validate_cap(vmx->pt_desc.caps,
1940 					PT_CAP_single_range_output)))
1941 			return 1;
1942 		msr_info->data = vmx->pt_desc.guest.output_mask;
1943 		break;
1944 	case MSR_IA32_RTIT_ADDR0_A ... MSR_IA32_RTIT_ADDR3_B:
1945 		index = msr_info->index - MSR_IA32_RTIT_ADDR0_A;
1946 		if (!vmx_pt_mode_is_host_guest() ||
1947 			(index >= 2 * intel_pt_validate_cap(vmx->pt_desc.caps,
1948 					PT_CAP_num_address_ranges)))
1949 			return 1;
1950 		if (index % 2)
1951 			msr_info->data = vmx->pt_desc.guest.addr_b[index / 2];
1952 		else
1953 			msr_info->data = vmx->pt_desc.guest.addr_a[index / 2];
1954 		break;
1955 	case MSR_TSC_AUX:
1956 		if (!msr_info->host_initiated &&
1957 		    !guest_cpuid_has(vcpu, X86_FEATURE_RDTSCP))
1958 			return 1;
1959 		goto find_shared_msr;
1960 	default:
1961 	find_shared_msr:
1962 		msr = find_msr_entry(vmx, msr_info->index);
1963 		if (msr) {
1964 			msr_info->data = msr->data;
1965 			break;
1966 		}
1967 		return kvm_get_msr_common(vcpu, msr_info);
1968 	}
1969 
1970 	return 0;
1971 }
1972 
1973 static u64 nested_vmx_truncate_sysenter_addr(struct kvm_vcpu *vcpu,
1974 						    u64 data)
1975 {
1976 #ifdef CONFIG_X86_64
1977 	if (!guest_cpuid_has(vcpu, X86_FEATURE_LM))
1978 		return (u32)data;
1979 #endif
1980 	return (unsigned long)data;
1981 }
1982 
1983 /*
1984  * Writes msr value into the appropriate "register".
1985  * Returns 0 on success, non-0 otherwise.
1986  * Assumes vcpu_load() was already called.
1987  */
1988 static int vmx_set_msr(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
1989 {
1990 	struct vcpu_vmx *vmx = to_vmx(vcpu);
1991 	struct shared_msr_entry *msr;
1992 	int ret = 0;
1993 	u32 msr_index = msr_info->index;
1994 	u64 data = msr_info->data;
1995 	u32 index;
1996 
1997 	switch (msr_index) {
1998 	case MSR_EFER:
1999 		ret = kvm_set_msr_common(vcpu, msr_info);
2000 		break;
2001 #ifdef CONFIG_X86_64
2002 	case MSR_FS_BASE:
2003 		vmx_segment_cache_clear(vmx);
2004 		vmcs_writel(GUEST_FS_BASE, data);
2005 		break;
2006 	case MSR_GS_BASE:
2007 		vmx_segment_cache_clear(vmx);
2008 		vmcs_writel(GUEST_GS_BASE, data);
2009 		break;
2010 	case MSR_KERNEL_GS_BASE:
2011 		vmx_write_guest_kernel_gs_base(vmx, data);
2012 		break;
2013 #endif
2014 	case MSR_IA32_SYSENTER_CS:
2015 		if (is_guest_mode(vcpu))
2016 			get_vmcs12(vcpu)->guest_sysenter_cs = data;
2017 		vmcs_write32(GUEST_SYSENTER_CS, data);
2018 		break;
2019 	case MSR_IA32_SYSENTER_EIP:
2020 		if (is_guest_mode(vcpu)) {
2021 			data = nested_vmx_truncate_sysenter_addr(vcpu, data);
2022 			get_vmcs12(vcpu)->guest_sysenter_eip = data;
2023 		}
2024 		vmcs_writel(GUEST_SYSENTER_EIP, data);
2025 		break;
2026 	case MSR_IA32_SYSENTER_ESP:
2027 		if (is_guest_mode(vcpu)) {
2028 			data = nested_vmx_truncate_sysenter_addr(vcpu, data);
2029 			get_vmcs12(vcpu)->guest_sysenter_esp = data;
2030 		}
2031 		vmcs_writel(GUEST_SYSENTER_ESP, data);
2032 		break;
2033 	case MSR_IA32_DEBUGCTLMSR:
2034 		if (is_guest_mode(vcpu) && get_vmcs12(vcpu)->vm_exit_controls &
2035 						VM_EXIT_SAVE_DEBUG_CONTROLS)
2036 			get_vmcs12(vcpu)->guest_ia32_debugctl = data;
2037 
2038 		ret = kvm_set_msr_common(vcpu, msr_info);
2039 		break;
2040 
2041 	case MSR_IA32_BNDCFGS:
2042 		if (!kvm_mpx_supported() ||
2043 		    (!msr_info->host_initiated &&
2044 		     !guest_cpuid_has(vcpu, X86_FEATURE_MPX)))
2045 			return 1;
2046 		if (is_noncanonical_address(data & PAGE_MASK, vcpu) ||
2047 		    (data & MSR_IA32_BNDCFGS_RSVD))
2048 			return 1;
2049 		vmcs_write64(GUEST_BNDCFGS, data);
2050 		break;
2051 	case MSR_IA32_UMWAIT_CONTROL:
2052 		if (!msr_info->host_initiated && !vmx_has_waitpkg(vmx))
2053 			return 1;
2054 
2055 		/* The reserved bit 1 and non-32 bit [63:32] should be zero */
2056 		if (data & (BIT_ULL(1) | GENMASK_ULL(63, 32)))
2057 			return 1;
2058 
2059 		vmx->msr_ia32_umwait_control = data;
2060 		break;
2061 	case MSR_IA32_SPEC_CTRL:
2062 		if (!msr_info->host_initiated &&
2063 		    !guest_cpuid_has(vcpu, X86_FEATURE_SPEC_CTRL))
2064 			return 1;
2065 
2066 		if (data & ~kvm_spec_ctrl_valid_bits(vcpu))
2067 			return 1;
2068 
2069 		vmx->spec_ctrl = data;
2070 		if (!data)
2071 			break;
2072 
2073 		/*
2074 		 * For non-nested:
2075 		 * When it's written (to non-zero) for the first time, pass
2076 		 * it through.
2077 		 *
2078 		 * For nested:
2079 		 * The handling of the MSR bitmap for L2 guests is done in
2080 		 * nested_vmx_prepare_msr_bitmap. We should not touch the
2081 		 * vmcs02.msr_bitmap here since it gets completely overwritten
2082 		 * in the merging. We update the vmcs01 here for L1 as well
2083 		 * since it will end up touching the MSR anyway now.
2084 		 */
2085 		vmx_disable_intercept_for_msr(vmx->vmcs01.msr_bitmap,
2086 					      MSR_IA32_SPEC_CTRL,
2087 					      MSR_TYPE_RW);
2088 		break;
2089 	case MSR_IA32_TSX_CTRL:
2090 		if (!msr_info->host_initiated &&
2091 		    !(vcpu->arch.arch_capabilities & ARCH_CAP_TSX_CTRL_MSR))
2092 			return 1;
2093 		if (data & ~(TSX_CTRL_RTM_DISABLE | TSX_CTRL_CPUID_CLEAR))
2094 			return 1;
2095 		goto find_shared_msr;
2096 	case MSR_IA32_PRED_CMD:
2097 		if (!msr_info->host_initiated &&
2098 		    !guest_cpuid_has(vcpu, X86_FEATURE_SPEC_CTRL))
2099 			return 1;
2100 
2101 		if (data & ~PRED_CMD_IBPB)
2102 			return 1;
2103 		if (!boot_cpu_has(X86_FEATURE_SPEC_CTRL))
2104 			return 1;
2105 		if (!data)
2106 			break;
2107 
2108 		wrmsrl(MSR_IA32_PRED_CMD, PRED_CMD_IBPB);
2109 
2110 		/*
2111 		 * For non-nested:
2112 		 * When it's written (to non-zero) for the first time, pass
2113 		 * it through.
2114 		 *
2115 		 * For nested:
2116 		 * The handling of the MSR bitmap for L2 guests is done in
2117 		 * nested_vmx_prepare_msr_bitmap. We should not touch the
2118 		 * vmcs02.msr_bitmap here since it gets completely overwritten
2119 		 * in the merging.
2120 		 */
2121 		vmx_disable_intercept_for_msr(vmx->vmcs01.msr_bitmap, MSR_IA32_PRED_CMD,
2122 					      MSR_TYPE_W);
2123 		break;
2124 	case MSR_IA32_CR_PAT:
2125 		if (!kvm_pat_valid(data))
2126 			return 1;
2127 
2128 		if (is_guest_mode(vcpu) &&
2129 		    get_vmcs12(vcpu)->vm_exit_controls & VM_EXIT_SAVE_IA32_PAT)
2130 			get_vmcs12(vcpu)->guest_ia32_pat = data;
2131 
2132 		if (vmcs_config.vmentry_ctrl & VM_ENTRY_LOAD_IA32_PAT) {
2133 			vmcs_write64(GUEST_IA32_PAT, data);
2134 			vcpu->arch.pat = data;
2135 			break;
2136 		}
2137 		ret = kvm_set_msr_common(vcpu, msr_info);
2138 		break;
2139 	case MSR_IA32_TSC_ADJUST:
2140 		ret = kvm_set_msr_common(vcpu, msr_info);
2141 		break;
2142 	case MSR_IA32_MCG_EXT_CTL:
2143 		if ((!msr_info->host_initiated &&
2144 		     !(to_vmx(vcpu)->msr_ia32_feature_control &
2145 		       FEAT_CTL_LMCE_ENABLED)) ||
2146 		    (data & ~MCG_EXT_CTL_LMCE_EN))
2147 			return 1;
2148 		vcpu->arch.mcg_ext_ctl = data;
2149 		break;
2150 	case MSR_IA32_FEAT_CTL:
2151 		if (!vmx_feature_control_msr_valid(vcpu, data) ||
2152 		    (to_vmx(vcpu)->msr_ia32_feature_control &
2153 		     FEAT_CTL_LOCKED && !msr_info->host_initiated))
2154 			return 1;
2155 		vmx->msr_ia32_feature_control = data;
2156 		if (msr_info->host_initiated && data == 0)
2157 			vmx_leave_nested(vcpu);
2158 		break;
2159 	case MSR_IA32_VMX_BASIC ... MSR_IA32_VMX_VMFUNC:
2160 		if (!msr_info->host_initiated)
2161 			return 1; /* they are read-only */
2162 		if (!nested_vmx_allowed(vcpu))
2163 			return 1;
2164 		return vmx_set_vmx_msr(vcpu, msr_index, data);
2165 	case MSR_IA32_RTIT_CTL:
2166 		if (!vmx_pt_mode_is_host_guest() ||
2167 			vmx_rtit_ctl_check(vcpu, data) ||
2168 			vmx->nested.vmxon)
2169 			return 1;
2170 		vmcs_write64(GUEST_IA32_RTIT_CTL, data);
2171 		vmx->pt_desc.guest.ctl = data;
2172 		pt_update_intercept_for_msr(vmx);
2173 		break;
2174 	case MSR_IA32_RTIT_STATUS:
2175 		if (!pt_can_write_msr(vmx))
2176 			return 1;
2177 		if (data & MSR_IA32_RTIT_STATUS_MASK)
2178 			return 1;
2179 		vmx->pt_desc.guest.status = data;
2180 		break;
2181 	case MSR_IA32_RTIT_CR3_MATCH:
2182 		if (!pt_can_write_msr(vmx))
2183 			return 1;
2184 		if (!intel_pt_validate_cap(vmx->pt_desc.caps,
2185 					   PT_CAP_cr3_filtering))
2186 			return 1;
2187 		vmx->pt_desc.guest.cr3_match = data;
2188 		break;
2189 	case MSR_IA32_RTIT_OUTPUT_BASE:
2190 		if (!pt_can_write_msr(vmx))
2191 			return 1;
2192 		if (!intel_pt_validate_cap(vmx->pt_desc.caps,
2193 					   PT_CAP_topa_output) &&
2194 		    !intel_pt_validate_cap(vmx->pt_desc.caps,
2195 					   PT_CAP_single_range_output))
2196 			return 1;
2197 		if (data & MSR_IA32_RTIT_OUTPUT_BASE_MASK)
2198 			return 1;
2199 		vmx->pt_desc.guest.output_base = data;
2200 		break;
2201 	case MSR_IA32_RTIT_OUTPUT_MASK:
2202 		if (!pt_can_write_msr(vmx))
2203 			return 1;
2204 		if (!intel_pt_validate_cap(vmx->pt_desc.caps,
2205 					   PT_CAP_topa_output) &&
2206 		    !intel_pt_validate_cap(vmx->pt_desc.caps,
2207 					   PT_CAP_single_range_output))
2208 			return 1;
2209 		vmx->pt_desc.guest.output_mask = data;
2210 		break;
2211 	case MSR_IA32_RTIT_ADDR0_A ... MSR_IA32_RTIT_ADDR3_B:
2212 		if (!pt_can_write_msr(vmx))
2213 			return 1;
2214 		index = msr_info->index - MSR_IA32_RTIT_ADDR0_A;
2215 		if (index >= 2 * intel_pt_validate_cap(vmx->pt_desc.caps,
2216 						       PT_CAP_num_address_ranges))
2217 			return 1;
2218 		if (is_noncanonical_address(data, vcpu))
2219 			return 1;
2220 		if (index % 2)
2221 			vmx->pt_desc.guest.addr_b[index / 2] = data;
2222 		else
2223 			vmx->pt_desc.guest.addr_a[index / 2] = data;
2224 		break;
2225 	case MSR_TSC_AUX:
2226 		if (!msr_info->host_initiated &&
2227 		    !guest_cpuid_has(vcpu, X86_FEATURE_RDTSCP))
2228 			return 1;
2229 		/* Check reserved bit, higher 32 bits should be zero */
2230 		if ((data >> 32) != 0)
2231 			return 1;
2232 		goto find_shared_msr;
2233 
2234 	default:
2235 	find_shared_msr:
2236 		msr = find_msr_entry(vmx, msr_index);
2237 		if (msr)
2238 			ret = vmx_set_guest_msr(vmx, msr, data);
2239 		else
2240 			ret = kvm_set_msr_common(vcpu, msr_info);
2241 	}
2242 
2243 	return ret;
2244 }
2245 
2246 static void vmx_cache_reg(struct kvm_vcpu *vcpu, enum kvm_reg reg)
2247 {
2248 	unsigned long guest_owned_bits;
2249 
2250 	kvm_register_mark_available(vcpu, reg);
2251 
2252 	switch (reg) {
2253 	case VCPU_REGS_RSP:
2254 		vcpu->arch.regs[VCPU_REGS_RSP] = vmcs_readl(GUEST_RSP);
2255 		break;
2256 	case VCPU_REGS_RIP:
2257 		vcpu->arch.regs[VCPU_REGS_RIP] = vmcs_readl(GUEST_RIP);
2258 		break;
2259 	case VCPU_EXREG_PDPTR:
2260 		if (enable_ept)
2261 			ept_save_pdptrs(vcpu);
2262 		break;
2263 	case VCPU_EXREG_CR0:
2264 		guest_owned_bits = vcpu->arch.cr0_guest_owned_bits;
2265 
2266 		vcpu->arch.cr0 &= ~guest_owned_bits;
2267 		vcpu->arch.cr0 |= vmcs_readl(GUEST_CR0) & guest_owned_bits;
2268 		break;
2269 	case VCPU_EXREG_CR3:
2270 		if (enable_unrestricted_guest || (enable_ept && is_paging(vcpu)))
2271 			vcpu->arch.cr3 = vmcs_readl(GUEST_CR3);
2272 		break;
2273 	case VCPU_EXREG_CR4:
2274 		guest_owned_bits = vcpu->arch.cr4_guest_owned_bits;
2275 
2276 		vcpu->arch.cr4 &= ~guest_owned_bits;
2277 		vcpu->arch.cr4 |= vmcs_readl(GUEST_CR4) & guest_owned_bits;
2278 		break;
2279 	default:
2280 		WARN_ON_ONCE(1);
2281 		break;
2282 	}
2283 }
2284 
2285 static __init int cpu_has_kvm_support(void)
2286 {
2287 	return cpu_has_vmx();
2288 }
2289 
2290 static __init int vmx_disabled_by_bios(void)
2291 {
2292 	return !boot_cpu_has(X86_FEATURE_MSR_IA32_FEAT_CTL) ||
2293 	       !boot_cpu_has(X86_FEATURE_VMX);
2294 }
2295 
2296 static int kvm_cpu_vmxon(u64 vmxon_pointer)
2297 {
2298 	u64 msr;
2299 
2300 	cr4_set_bits(X86_CR4_VMXE);
2301 	intel_pt_handle_vmx(1);
2302 
2303 	asm_volatile_goto("1: vmxon %[vmxon_pointer]\n\t"
2304 			  _ASM_EXTABLE(1b, %l[fault])
2305 			  : : [vmxon_pointer] "m"(vmxon_pointer)
2306 			  : : fault);
2307 	return 0;
2308 
2309 fault:
2310 	WARN_ONCE(1, "VMXON faulted, MSR_IA32_FEAT_CTL (0x3a) = 0x%llx\n",
2311 		  rdmsrl_safe(MSR_IA32_FEAT_CTL, &msr) ? 0xdeadbeef : msr);
2312 	intel_pt_handle_vmx(0);
2313 	cr4_clear_bits(X86_CR4_VMXE);
2314 
2315 	return -EFAULT;
2316 }
2317 
2318 static int hardware_enable(void)
2319 {
2320 	int cpu = raw_smp_processor_id();
2321 	u64 phys_addr = __pa(per_cpu(vmxarea, cpu));
2322 	int r;
2323 
2324 	if (cr4_read_shadow() & X86_CR4_VMXE)
2325 		return -EBUSY;
2326 
2327 	/*
2328 	 * This can happen if we hot-added a CPU but failed to allocate
2329 	 * VP assist page for it.
2330 	 */
2331 	if (static_branch_unlikely(&enable_evmcs) &&
2332 	    !hv_get_vp_assist_page(cpu))
2333 		return -EFAULT;
2334 
2335 	r = kvm_cpu_vmxon(phys_addr);
2336 	if (r)
2337 		return r;
2338 
2339 	if (enable_ept)
2340 		ept_sync_global();
2341 
2342 	return 0;
2343 }
2344 
2345 static void vmclear_local_loaded_vmcss(void)
2346 {
2347 	int cpu = raw_smp_processor_id();
2348 	struct loaded_vmcs *v, *n;
2349 
2350 	list_for_each_entry_safe(v, n, &per_cpu(loaded_vmcss_on_cpu, cpu),
2351 				 loaded_vmcss_on_cpu_link)
2352 		__loaded_vmcs_clear(v);
2353 }
2354 
2355 
2356 /* Just like cpu_vmxoff(), but with the __kvm_handle_fault_on_reboot()
2357  * tricks.
2358  */
2359 static void kvm_cpu_vmxoff(void)
2360 {
2361 	asm volatile (__ex("vmxoff"));
2362 
2363 	intel_pt_handle_vmx(0);
2364 	cr4_clear_bits(X86_CR4_VMXE);
2365 }
2366 
2367 static void hardware_disable(void)
2368 {
2369 	vmclear_local_loaded_vmcss();
2370 	kvm_cpu_vmxoff();
2371 }
2372 
2373 /*
2374  * There is no X86_FEATURE for SGX yet, but anyway we need to query CPUID
2375  * directly instead of going through cpu_has(), to ensure KVM is trapping
2376  * ENCLS whenever it's supported in hardware.  It does not matter whether
2377  * the host OS supports or has enabled SGX.
2378  */
2379 static bool cpu_has_sgx(void)
2380 {
2381 	return cpuid_eax(0) >= 0x12 && (cpuid_eax(0x12) & BIT(0));
2382 }
2383 
2384 static __init int adjust_vmx_controls(u32 ctl_min, u32 ctl_opt,
2385 				      u32 msr, u32 *result)
2386 {
2387 	u32 vmx_msr_low, vmx_msr_high;
2388 	u32 ctl = ctl_min | ctl_opt;
2389 
2390 	rdmsr(msr, vmx_msr_low, vmx_msr_high);
2391 
2392 	ctl &= vmx_msr_high; /* bit == 0 in high word ==> must be zero */
2393 	ctl |= vmx_msr_low;  /* bit == 1 in low word  ==> must be one  */
2394 
2395 	/* Ensure minimum (required) set of control bits are supported. */
2396 	if (ctl_min & ~ctl)
2397 		return -EIO;
2398 
2399 	*result = ctl;
2400 	return 0;
2401 }
2402 
2403 static __init int setup_vmcs_config(struct vmcs_config *vmcs_conf,
2404 				    struct vmx_capability *vmx_cap)
2405 {
2406 	u32 vmx_msr_low, vmx_msr_high;
2407 	u32 min, opt, min2, opt2;
2408 	u32 _pin_based_exec_control = 0;
2409 	u32 _cpu_based_exec_control = 0;
2410 	u32 _cpu_based_2nd_exec_control = 0;
2411 	u32 _vmexit_control = 0;
2412 	u32 _vmentry_control = 0;
2413 
2414 	memset(vmcs_conf, 0, sizeof(*vmcs_conf));
2415 	min = CPU_BASED_HLT_EXITING |
2416 #ifdef CONFIG_X86_64
2417 	      CPU_BASED_CR8_LOAD_EXITING |
2418 	      CPU_BASED_CR8_STORE_EXITING |
2419 #endif
2420 	      CPU_BASED_CR3_LOAD_EXITING |
2421 	      CPU_BASED_CR3_STORE_EXITING |
2422 	      CPU_BASED_UNCOND_IO_EXITING |
2423 	      CPU_BASED_MOV_DR_EXITING |
2424 	      CPU_BASED_USE_TSC_OFFSETTING |
2425 	      CPU_BASED_MWAIT_EXITING |
2426 	      CPU_BASED_MONITOR_EXITING |
2427 	      CPU_BASED_INVLPG_EXITING |
2428 	      CPU_BASED_RDPMC_EXITING;
2429 
2430 	opt = CPU_BASED_TPR_SHADOW |
2431 	      CPU_BASED_USE_MSR_BITMAPS |
2432 	      CPU_BASED_ACTIVATE_SECONDARY_CONTROLS;
2433 	if (adjust_vmx_controls(min, opt, MSR_IA32_VMX_PROCBASED_CTLS,
2434 				&_cpu_based_exec_control) < 0)
2435 		return -EIO;
2436 #ifdef CONFIG_X86_64
2437 	if ((_cpu_based_exec_control & CPU_BASED_TPR_SHADOW))
2438 		_cpu_based_exec_control &= ~CPU_BASED_CR8_LOAD_EXITING &
2439 					   ~CPU_BASED_CR8_STORE_EXITING;
2440 #endif
2441 	if (_cpu_based_exec_control & CPU_BASED_ACTIVATE_SECONDARY_CONTROLS) {
2442 		min2 = 0;
2443 		opt2 = SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES |
2444 			SECONDARY_EXEC_VIRTUALIZE_X2APIC_MODE |
2445 			SECONDARY_EXEC_WBINVD_EXITING |
2446 			SECONDARY_EXEC_ENABLE_VPID |
2447 			SECONDARY_EXEC_ENABLE_EPT |
2448 			SECONDARY_EXEC_UNRESTRICTED_GUEST |
2449 			SECONDARY_EXEC_PAUSE_LOOP_EXITING |
2450 			SECONDARY_EXEC_DESC |
2451 			SECONDARY_EXEC_RDTSCP |
2452 			SECONDARY_EXEC_ENABLE_INVPCID |
2453 			SECONDARY_EXEC_APIC_REGISTER_VIRT |
2454 			SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY |
2455 			SECONDARY_EXEC_SHADOW_VMCS |
2456 			SECONDARY_EXEC_XSAVES |
2457 			SECONDARY_EXEC_RDSEED_EXITING |
2458 			SECONDARY_EXEC_RDRAND_EXITING |
2459 			SECONDARY_EXEC_ENABLE_PML |
2460 			SECONDARY_EXEC_TSC_SCALING |
2461 			SECONDARY_EXEC_ENABLE_USR_WAIT_PAUSE |
2462 			SECONDARY_EXEC_PT_USE_GPA |
2463 			SECONDARY_EXEC_PT_CONCEAL_VMX |
2464 			SECONDARY_EXEC_ENABLE_VMFUNC;
2465 		if (cpu_has_sgx())
2466 			opt2 |= SECONDARY_EXEC_ENCLS_EXITING;
2467 		if (adjust_vmx_controls(min2, opt2,
2468 					MSR_IA32_VMX_PROCBASED_CTLS2,
2469 					&_cpu_based_2nd_exec_control) < 0)
2470 			return -EIO;
2471 	}
2472 #ifndef CONFIG_X86_64
2473 	if (!(_cpu_based_2nd_exec_control &
2474 				SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES))
2475 		_cpu_based_exec_control &= ~CPU_BASED_TPR_SHADOW;
2476 #endif
2477 
2478 	if (!(_cpu_based_exec_control & CPU_BASED_TPR_SHADOW))
2479 		_cpu_based_2nd_exec_control &= ~(
2480 				SECONDARY_EXEC_APIC_REGISTER_VIRT |
2481 				SECONDARY_EXEC_VIRTUALIZE_X2APIC_MODE |
2482 				SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY);
2483 
2484 	rdmsr_safe(MSR_IA32_VMX_EPT_VPID_CAP,
2485 		&vmx_cap->ept, &vmx_cap->vpid);
2486 
2487 	if (_cpu_based_2nd_exec_control & SECONDARY_EXEC_ENABLE_EPT) {
2488 		/* CR3 accesses and invlpg don't need to cause VM Exits when EPT
2489 		   enabled */
2490 		_cpu_based_exec_control &= ~(CPU_BASED_CR3_LOAD_EXITING |
2491 					     CPU_BASED_CR3_STORE_EXITING |
2492 					     CPU_BASED_INVLPG_EXITING);
2493 	} else if (vmx_cap->ept) {
2494 		vmx_cap->ept = 0;
2495 		pr_warn_once("EPT CAP should not exist if not support "
2496 				"1-setting enable EPT VM-execution control\n");
2497 	}
2498 	if (!(_cpu_based_2nd_exec_control & SECONDARY_EXEC_ENABLE_VPID) &&
2499 		vmx_cap->vpid) {
2500 		vmx_cap->vpid = 0;
2501 		pr_warn_once("VPID CAP should not exist if not support "
2502 				"1-setting enable VPID VM-execution control\n");
2503 	}
2504 
2505 	min = VM_EXIT_SAVE_DEBUG_CONTROLS | VM_EXIT_ACK_INTR_ON_EXIT;
2506 #ifdef CONFIG_X86_64
2507 	min |= VM_EXIT_HOST_ADDR_SPACE_SIZE;
2508 #endif
2509 	opt = VM_EXIT_LOAD_IA32_PERF_GLOBAL_CTRL |
2510 	      VM_EXIT_LOAD_IA32_PAT |
2511 	      VM_EXIT_LOAD_IA32_EFER |
2512 	      VM_EXIT_CLEAR_BNDCFGS |
2513 	      VM_EXIT_PT_CONCEAL_PIP |
2514 	      VM_EXIT_CLEAR_IA32_RTIT_CTL;
2515 	if (adjust_vmx_controls(min, opt, MSR_IA32_VMX_EXIT_CTLS,
2516 				&_vmexit_control) < 0)
2517 		return -EIO;
2518 
2519 	min = PIN_BASED_EXT_INTR_MASK | PIN_BASED_NMI_EXITING;
2520 	opt = PIN_BASED_VIRTUAL_NMIS | PIN_BASED_POSTED_INTR |
2521 		 PIN_BASED_VMX_PREEMPTION_TIMER;
2522 	if (adjust_vmx_controls(min, opt, MSR_IA32_VMX_PINBASED_CTLS,
2523 				&_pin_based_exec_control) < 0)
2524 		return -EIO;
2525 
2526 	if (cpu_has_broken_vmx_preemption_timer())
2527 		_pin_based_exec_control &= ~PIN_BASED_VMX_PREEMPTION_TIMER;
2528 	if (!(_cpu_based_2nd_exec_control &
2529 		SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY))
2530 		_pin_based_exec_control &= ~PIN_BASED_POSTED_INTR;
2531 
2532 	min = VM_ENTRY_LOAD_DEBUG_CONTROLS;
2533 	opt = VM_ENTRY_LOAD_IA32_PERF_GLOBAL_CTRL |
2534 	      VM_ENTRY_LOAD_IA32_PAT |
2535 	      VM_ENTRY_LOAD_IA32_EFER |
2536 	      VM_ENTRY_LOAD_BNDCFGS |
2537 	      VM_ENTRY_PT_CONCEAL_PIP |
2538 	      VM_ENTRY_LOAD_IA32_RTIT_CTL;
2539 	if (adjust_vmx_controls(min, opt, MSR_IA32_VMX_ENTRY_CTLS,
2540 				&_vmentry_control) < 0)
2541 		return -EIO;
2542 
2543 	/*
2544 	 * Some cpus support VM_{ENTRY,EXIT}_IA32_PERF_GLOBAL_CTRL but they
2545 	 * can't be used due to an errata where VM Exit may incorrectly clear
2546 	 * IA32_PERF_GLOBAL_CTRL[34:32].  Workaround the errata by using the
2547 	 * MSR load mechanism to switch IA32_PERF_GLOBAL_CTRL.
2548 	 */
2549 	if (boot_cpu_data.x86 == 0x6) {
2550 		switch (boot_cpu_data.x86_model) {
2551 		case 26: /* AAK155 */
2552 		case 30: /* AAP115 */
2553 		case 37: /* AAT100 */
2554 		case 44: /* BC86,AAY89,BD102 */
2555 		case 46: /* BA97 */
2556 			_vmentry_control &= ~VM_ENTRY_LOAD_IA32_PERF_GLOBAL_CTRL;
2557 			_vmexit_control &= ~VM_EXIT_LOAD_IA32_PERF_GLOBAL_CTRL;
2558 			pr_warn_once("kvm: VM_EXIT_LOAD_IA32_PERF_GLOBAL_CTRL "
2559 					"does not work properly. Using workaround\n");
2560 			break;
2561 		default:
2562 			break;
2563 		}
2564 	}
2565 
2566 
2567 	rdmsr(MSR_IA32_VMX_BASIC, vmx_msr_low, vmx_msr_high);
2568 
2569 	/* IA-32 SDM Vol 3B: VMCS size is never greater than 4kB. */
2570 	if ((vmx_msr_high & 0x1fff) > PAGE_SIZE)
2571 		return -EIO;
2572 
2573 #ifdef CONFIG_X86_64
2574 	/* IA-32 SDM Vol 3B: 64-bit CPUs always have VMX_BASIC_MSR[48]==0. */
2575 	if (vmx_msr_high & (1u<<16))
2576 		return -EIO;
2577 #endif
2578 
2579 	/* Require Write-Back (WB) memory type for VMCS accesses. */
2580 	if (((vmx_msr_high >> 18) & 15) != 6)
2581 		return -EIO;
2582 
2583 	vmcs_conf->size = vmx_msr_high & 0x1fff;
2584 	vmcs_conf->order = get_order(vmcs_conf->size);
2585 	vmcs_conf->basic_cap = vmx_msr_high & ~0x1fff;
2586 
2587 	vmcs_conf->revision_id = vmx_msr_low;
2588 
2589 	vmcs_conf->pin_based_exec_ctrl = _pin_based_exec_control;
2590 	vmcs_conf->cpu_based_exec_ctrl = _cpu_based_exec_control;
2591 	vmcs_conf->cpu_based_2nd_exec_ctrl = _cpu_based_2nd_exec_control;
2592 	vmcs_conf->vmexit_ctrl         = _vmexit_control;
2593 	vmcs_conf->vmentry_ctrl        = _vmentry_control;
2594 
2595 	if (static_branch_unlikely(&enable_evmcs))
2596 		evmcs_sanitize_exec_ctrls(vmcs_conf);
2597 
2598 	return 0;
2599 }
2600 
2601 struct vmcs *alloc_vmcs_cpu(bool shadow, int cpu, gfp_t flags)
2602 {
2603 	int node = cpu_to_node(cpu);
2604 	struct page *pages;
2605 	struct vmcs *vmcs;
2606 
2607 	pages = __alloc_pages_node(node, flags, vmcs_config.order);
2608 	if (!pages)
2609 		return NULL;
2610 	vmcs = page_address(pages);
2611 	memset(vmcs, 0, vmcs_config.size);
2612 
2613 	/* KVM supports Enlightened VMCS v1 only */
2614 	if (static_branch_unlikely(&enable_evmcs))
2615 		vmcs->hdr.revision_id = KVM_EVMCS_VERSION;
2616 	else
2617 		vmcs->hdr.revision_id = vmcs_config.revision_id;
2618 
2619 	if (shadow)
2620 		vmcs->hdr.shadow_vmcs = 1;
2621 	return vmcs;
2622 }
2623 
2624 void free_vmcs(struct vmcs *vmcs)
2625 {
2626 	free_pages((unsigned long)vmcs, vmcs_config.order);
2627 }
2628 
2629 /*
2630  * Free a VMCS, but before that VMCLEAR it on the CPU where it was last loaded
2631  */
2632 void free_loaded_vmcs(struct loaded_vmcs *loaded_vmcs)
2633 {
2634 	if (!loaded_vmcs->vmcs)
2635 		return;
2636 	loaded_vmcs_clear(loaded_vmcs);
2637 	free_vmcs(loaded_vmcs->vmcs);
2638 	loaded_vmcs->vmcs = NULL;
2639 	if (loaded_vmcs->msr_bitmap)
2640 		free_page((unsigned long)loaded_vmcs->msr_bitmap);
2641 	WARN_ON(loaded_vmcs->shadow_vmcs != NULL);
2642 }
2643 
2644 int alloc_loaded_vmcs(struct loaded_vmcs *loaded_vmcs)
2645 {
2646 	loaded_vmcs->vmcs = alloc_vmcs(false);
2647 	if (!loaded_vmcs->vmcs)
2648 		return -ENOMEM;
2649 
2650 	vmcs_clear(loaded_vmcs->vmcs);
2651 
2652 	loaded_vmcs->shadow_vmcs = NULL;
2653 	loaded_vmcs->hv_timer_soft_disabled = false;
2654 	loaded_vmcs->cpu = -1;
2655 	loaded_vmcs->launched = 0;
2656 
2657 	if (cpu_has_vmx_msr_bitmap()) {
2658 		loaded_vmcs->msr_bitmap = (unsigned long *)
2659 				__get_free_page(GFP_KERNEL_ACCOUNT);
2660 		if (!loaded_vmcs->msr_bitmap)
2661 			goto out_vmcs;
2662 		memset(loaded_vmcs->msr_bitmap, 0xff, PAGE_SIZE);
2663 
2664 		if (IS_ENABLED(CONFIG_HYPERV) &&
2665 		    static_branch_unlikely(&enable_evmcs) &&
2666 		    (ms_hyperv.nested_features & HV_X64_NESTED_MSR_BITMAP)) {
2667 			struct hv_enlightened_vmcs *evmcs =
2668 				(struct hv_enlightened_vmcs *)loaded_vmcs->vmcs;
2669 
2670 			evmcs->hv_enlightenments_control.msr_bitmap = 1;
2671 		}
2672 	}
2673 
2674 	memset(&loaded_vmcs->host_state, 0, sizeof(struct vmcs_host_state));
2675 	memset(&loaded_vmcs->controls_shadow, 0,
2676 		sizeof(struct vmcs_controls_shadow));
2677 
2678 	return 0;
2679 
2680 out_vmcs:
2681 	free_loaded_vmcs(loaded_vmcs);
2682 	return -ENOMEM;
2683 }
2684 
2685 static void free_kvm_area(void)
2686 {
2687 	int cpu;
2688 
2689 	for_each_possible_cpu(cpu) {
2690 		free_vmcs(per_cpu(vmxarea, cpu));
2691 		per_cpu(vmxarea, cpu) = NULL;
2692 	}
2693 }
2694 
2695 static __init int alloc_kvm_area(void)
2696 {
2697 	int cpu;
2698 
2699 	for_each_possible_cpu(cpu) {
2700 		struct vmcs *vmcs;
2701 
2702 		vmcs = alloc_vmcs_cpu(false, cpu, GFP_KERNEL);
2703 		if (!vmcs) {
2704 			free_kvm_area();
2705 			return -ENOMEM;
2706 		}
2707 
2708 		/*
2709 		 * When eVMCS is enabled, alloc_vmcs_cpu() sets
2710 		 * vmcs->revision_id to KVM_EVMCS_VERSION instead of
2711 		 * revision_id reported by MSR_IA32_VMX_BASIC.
2712 		 *
2713 		 * However, even though not explicitly documented by
2714 		 * TLFS, VMXArea passed as VMXON argument should
2715 		 * still be marked with revision_id reported by
2716 		 * physical CPU.
2717 		 */
2718 		if (static_branch_unlikely(&enable_evmcs))
2719 			vmcs->hdr.revision_id = vmcs_config.revision_id;
2720 
2721 		per_cpu(vmxarea, cpu) = vmcs;
2722 	}
2723 	return 0;
2724 }
2725 
2726 static void fix_pmode_seg(struct kvm_vcpu *vcpu, int seg,
2727 		struct kvm_segment *save)
2728 {
2729 	if (!emulate_invalid_guest_state) {
2730 		/*
2731 		 * CS and SS RPL should be equal during guest entry according
2732 		 * to VMX spec, but in reality it is not always so. Since vcpu
2733 		 * is in the middle of the transition from real mode to
2734 		 * protected mode it is safe to assume that RPL 0 is a good
2735 		 * default value.
2736 		 */
2737 		if (seg == VCPU_SREG_CS || seg == VCPU_SREG_SS)
2738 			save->selector &= ~SEGMENT_RPL_MASK;
2739 		save->dpl = save->selector & SEGMENT_RPL_MASK;
2740 		save->s = 1;
2741 	}
2742 	vmx_set_segment(vcpu, save, seg);
2743 }
2744 
2745 static void enter_pmode(struct kvm_vcpu *vcpu)
2746 {
2747 	unsigned long flags;
2748 	struct vcpu_vmx *vmx = to_vmx(vcpu);
2749 
2750 	/*
2751 	 * Update real mode segment cache. It may be not up-to-date if sement
2752 	 * register was written while vcpu was in a guest mode.
2753 	 */
2754 	vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_ES], VCPU_SREG_ES);
2755 	vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_DS], VCPU_SREG_DS);
2756 	vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_FS], VCPU_SREG_FS);
2757 	vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_GS], VCPU_SREG_GS);
2758 	vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_SS], VCPU_SREG_SS);
2759 	vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_CS], VCPU_SREG_CS);
2760 
2761 	vmx->rmode.vm86_active = 0;
2762 
2763 	vmx_set_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_TR], VCPU_SREG_TR);
2764 
2765 	flags = vmcs_readl(GUEST_RFLAGS);
2766 	flags &= RMODE_GUEST_OWNED_EFLAGS_BITS;
2767 	flags |= vmx->rmode.save_rflags & ~RMODE_GUEST_OWNED_EFLAGS_BITS;
2768 	vmcs_writel(GUEST_RFLAGS, flags);
2769 
2770 	vmcs_writel(GUEST_CR4, (vmcs_readl(GUEST_CR4) & ~X86_CR4_VME) |
2771 			(vmcs_readl(CR4_READ_SHADOW) & X86_CR4_VME));
2772 
2773 	update_exception_bitmap(vcpu);
2774 
2775 	fix_pmode_seg(vcpu, VCPU_SREG_CS, &vmx->rmode.segs[VCPU_SREG_CS]);
2776 	fix_pmode_seg(vcpu, VCPU_SREG_SS, &vmx->rmode.segs[VCPU_SREG_SS]);
2777 	fix_pmode_seg(vcpu, VCPU_SREG_ES, &vmx->rmode.segs[VCPU_SREG_ES]);
2778 	fix_pmode_seg(vcpu, VCPU_SREG_DS, &vmx->rmode.segs[VCPU_SREG_DS]);
2779 	fix_pmode_seg(vcpu, VCPU_SREG_FS, &vmx->rmode.segs[VCPU_SREG_FS]);
2780 	fix_pmode_seg(vcpu, VCPU_SREG_GS, &vmx->rmode.segs[VCPU_SREG_GS]);
2781 }
2782 
2783 static void fix_rmode_seg(int seg, struct kvm_segment *save)
2784 {
2785 	const struct kvm_vmx_segment_field *sf = &kvm_vmx_segment_fields[seg];
2786 	struct kvm_segment var = *save;
2787 
2788 	var.dpl = 0x3;
2789 	if (seg == VCPU_SREG_CS)
2790 		var.type = 0x3;
2791 
2792 	if (!emulate_invalid_guest_state) {
2793 		var.selector = var.base >> 4;
2794 		var.base = var.base & 0xffff0;
2795 		var.limit = 0xffff;
2796 		var.g = 0;
2797 		var.db = 0;
2798 		var.present = 1;
2799 		var.s = 1;
2800 		var.l = 0;
2801 		var.unusable = 0;
2802 		var.type = 0x3;
2803 		var.avl = 0;
2804 		if (save->base & 0xf)
2805 			printk_once(KERN_WARNING "kvm: segment base is not "
2806 					"paragraph aligned when entering "
2807 					"protected mode (seg=%d)", seg);
2808 	}
2809 
2810 	vmcs_write16(sf->selector, var.selector);
2811 	vmcs_writel(sf->base, var.base);
2812 	vmcs_write32(sf->limit, var.limit);
2813 	vmcs_write32(sf->ar_bytes, vmx_segment_access_rights(&var));
2814 }
2815 
2816 static void enter_rmode(struct kvm_vcpu *vcpu)
2817 {
2818 	unsigned long flags;
2819 	struct vcpu_vmx *vmx = to_vmx(vcpu);
2820 	struct kvm_vmx *kvm_vmx = to_kvm_vmx(vcpu->kvm);
2821 
2822 	vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_TR], VCPU_SREG_TR);
2823 	vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_ES], VCPU_SREG_ES);
2824 	vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_DS], VCPU_SREG_DS);
2825 	vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_FS], VCPU_SREG_FS);
2826 	vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_GS], VCPU_SREG_GS);
2827 	vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_SS], VCPU_SREG_SS);
2828 	vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_CS], VCPU_SREG_CS);
2829 
2830 	vmx->rmode.vm86_active = 1;
2831 
2832 	/*
2833 	 * Very old userspace does not call KVM_SET_TSS_ADDR before entering
2834 	 * vcpu. Warn the user that an update is overdue.
2835 	 */
2836 	if (!kvm_vmx->tss_addr)
2837 		printk_once(KERN_WARNING "kvm: KVM_SET_TSS_ADDR need to be "
2838 			     "called before entering vcpu\n");
2839 
2840 	vmx_segment_cache_clear(vmx);
2841 
2842 	vmcs_writel(GUEST_TR_BASE, kvm_vmx->tss_addr);
2843 	vmcs_write32(GUEST_TR_LIMIT, RMODE_TSS_SIZE - 1);
2844 	vmcs_write32(GUEST_TR_AR_BYTES, 0x008b);
2845 
2846 	flags = vmcs_readl(GUEST_RFLAGS);
2847 	vmx->rmode.save_rflags = flags;
2848 
2849 	flags |= X86_EFLAGS_IOPL | X86_EFLAGS_VM;
2850 
2851 	vmcs_writel(GUEST_RFLAGS, flags);
2852 	vmcs_writel(GUEST_CR4, vmcs_readl(GUEST_CR4) | X86_CR4_VME);
2853 	update_exception_bitmap(vcpu);
2854 
2855 	fix_rmode_seg(VCPU_SREG_SS, &vmx->rmode.segs[VCPU_SREG_SS]);
2856 	fix_rmode_seg(VCPU_SREG_CS, &vmx->rmode.segs[VCPU_SREG_CS]);
2857 	fix_rmode_seg(VCPU_SREG_ES, &vmx->rmode.segs[VCPU_SREG_ES]);
2858 	fix_rmode_seg(VCPU_SREG_DS, &vmx->rmode.segs[VCPU_SREG_DS]);
2859 	fix_rmode_seg(VCPU_SREG_GS, &vmx->rmode.segs[VCPU_SREG_GS]);
2860 	fix_rmode_seg(VCPU_SREG_FS, &vmx->rmode.segs[VCPU_SREG_FS]);
2861 
2862 	kvm_mmu_reset_context(vcpu);
2863 }
2864 
2865 void vmx_set_efer(struct kvm_vcpu *vcpu, u64 efer)
2866 {
2867 	struct vcpu_vmx *vmx = to_vmx(vcpu);
2868 	struct shared_msr_entry *msr = find_msr_entry(vmx, MSR_EFER);
2869 
2870 	if (!msr)
2871 		return;
2872 
2873 	vcpu->arch.efer = efer;
2874 	if (efer & EFER_LMA) {
2875 		vm_entry_controls_setbit(to_vmx(vcpu), VM_ENTRY_IA32E_MODE);
2876 		msr->data = efer;
2877 	} else {
2878 		vm_entry_controls_clearbit(to_vmx(vcpu), VM_ENTRY_IA32E_MODE);
2879 
2880 		msr->data = efer & ~EFER_LME;
2881 	}
2882 	setup_msrs(vmx);
2883 }
2884 
2885 #ifdef CONFIG_X86_64
2886 
2887 static void enter_lmode(struct kvm_vcpu *vcpu)
2888 {
2889 	u32 guest_tr_ar;
2890 
2891 	vmx_segment_cache_clear(to_vmx(vcpu));
2892 
2893 	guest_tr_ar = vmcs_read32(GUEST_TR_AR_BYTES);
2894 	if ((guest_tr_ar & VMX_AR_TYPE_MASK) != VMX_AR_TYPE_BUSY_64_TSS) {
2895 		pr_debug_ratelimited("%s: tss fixup for long mode. \n",
2896 				     __func__);
2897 		vmcs_write32(GUEST_TR_AR_BYTES,
2898 			     (guest_tr_ar & ~VMX_AR_TYPE_MASK)
2899 			     | VMX_AR_TYPE_BUSY_64_TSS);
2900 	}
2901 	vmx_set_efer(vcpu, vcpu->arch.efer | EFER_LMA);
2902 }
2903 
2904 static void exit_lmode(struct kvm_vcpu *vcpu)
2905 {
2906 	vm_entry_controls_clearbit(to_vmx(vcpu), VM_ENTRY_IA32E_MODE);
2907 	vmx_set_efer(vcpu, vcpu->arch.efer & ~EFER_LMA);
2908 }
2909 
2910 #endif
2911 
2912 static void vmx_flush_tlb_all(struct kvm_vcpu *vcpu)
2913 {
2914 	struct vcpu_vmx *vmx = to_vmx(vcpu);
2915 
2916 	/*
2917 	 * INVEPT must be issued when EPT is enabled, irrespective of VPID, as
2918 	 * the CPU is not required to invalidate guest-physical mappings on
2919 	 * VM-Entry, even if VPID is disabled.  Guest-physical mappings are
2920 	 * associated with the root EPT structure and not any particular VPID
2921 	 * (INVVPID also isn't required to invalidate guest-physical mappings).
2922 	 */
2923 	if (enable_ept) {
2924 		ept_sync_global();
2925 	} else if (enable_vpid) {
2926 		if (cpu_has_vmx_invvpid_global()) {
2927 			vpid_sync_vcpu_global();
2928 		} else {
2929 			vpid_sync_vcpu_single(vmx->vpid);
2930 			vpid_sync_vcpu_single(vmx->nested.vpid02);
2931 		}
2932 	}
2933 }
2934 
2935 static void vmx_flush_tlb_current(struct kvm_vcpu *vcpu)
2936 {
2937 	u64 root_hpa = vcpu->arch.mmu->root_hpa;
2938 
2939 	/* No flush required if the current context is invalid. */
2940 	if (!VALID_PAGE(root_hpa))
2941 		return;
2942 
2943 	if (enable_ept)
2944 		ept_sync_context(construct_eptp(vcpu, root_hpa));
2945 	else if (!is_guest_mode(vcpu))
2946 		vpid_sync_context(to_vmx(vcpu)->vpid);
2947 	else
2948 		vpid_sync_context(nested_get_vpid02(vcpu));
2949 }
2950 
2951 static void vmx_flush_tlb_gva(struct kvm_vcpu *vcpu, gva_t addr)
2952 {
2953 	/*
2954 	 * vpid_sync_vcpu_addr() is a nop if vmx->vpid==0, see the comment in
2955 	 * vmx_flush_tlb_guest() for an explanation of why this is ok.
2956 	 */
2957 	vpid_sync_vcpu_addr(to_vmx(vcpu)->vpid, addr);
2958 }
2959 
2960 static void vmx_flush_tlb_guest(struct kvm_vcpu *vcpu)
2961 {
2962 	/*
2963 	 * vpid_sync_context() is a nop if vmx->vpid==0, e.g. if enable_vpid==0
2964 	 * or a vpid couldn't be allocated for this vCPU.  VM-Enter and VM-Exit
2965 	 * are required to flush GVA->{G,H}PA mappings from the TLB if vpid is
2966 	 * disabled (VM-Enter with vpid enabled and vpid==0 is disallowed),
2967 	 * i.e. no explicit INVVPID is necessary.
2968 	 */
2969 	vpid_sync_context(to_vmx(vcpu)->vpid);
2970 }
2971 
2972 static void ept_load_pdptrs(struct kvm_vcpu *vcpu)
2973 {
2974 	struct kvm_mmu *mmu = vcpu->arch.walk_mmu;
2975 
2976 	if (!kvm_register_is_dirty(vcpu, VCPU_EXREG_PDPTR))
2977 		return;
2978 
2979 	if (is_pae_paging(vcpu)) {
2980 		vmcs_write64(GUEST_PDPTR0, mmu->pdptrs[0]);
2981 		vmcs_write64(GUEST_PDPTR1, mmu->pdptrs[1]);
2982 		vmcs_write64(GUEST_PDPTR2, mmu->pdptrs[2]);
2983 		vmcs_write64(GUEST_PDPTR3, mmu->pdptrs[3]);
2984 	}
2985 }
2986 
2987 void ept_save_pdptrs(struct kvm_vcpu *vcpu)
2988 {
2989 	struct kvm_mmu *mmu = vcpu->arch.walk_mmu;
2990 
2991 	if (WARN_ON_ONCE(!is_pae_paging(vcpu)))
2992 		return;
2993 
2994 	mmu->pdptrs[0] = vmcs_read64(GUEST_PDPTR0);
2995 	mmu->pdptrs[1] = vmcs_read64(GUEST_PDPTR1);
2996 	mmu->pdptrs[2] = vmcs_read64(GUEST_PDPTR2);
2997 	mmu->pdptrs[3] = vmcs_read64(GUEST_PDPTR3);
2998 
2999 	kvm_register_mark_dirty(vcpu, VCPU_EXREG_PDPTR);
3000 }
3001 
3002 static void ept_update_paging_mode_cr0(unsigned long *hw_cr0,
3003 					unsigned long cr0,
3004 					struct kvm_vcpu *vcpu)
3005 {
3006 	struct vcpu_vmx *vmx = to_vmx(vcpu);
3007 
3008 	if (!kvm_register_is_available(vcpu, VCPU_EXREG_CR3))
3009 		vmx_cache_reg(vcpu, VCPU_EXREG_CR3);
3010 	if (!(cr0 & X86_CR0_PG)) {
3011 		/* From paging/starting to nonpaging */
3012 		exec_controls_setbit(vmx, CPU_BASED_CR3_LOAD_EXITING |
3013 					  CPU_BASED_CR3_STORE_EXITING);
3014 		vcpu->arch.cr0 = cr0;
3015 		vmx_set_cr4(vcpu, kvm_read_cr4(vcpu));
3016 	} else if (!is_paging(vcpu)) {
3017 		/* From nonpaging to paging */
3018 		exec_controls_clearbit(vmx, CPU_BASED_CR3_LOAD_EXITING |
3019 					    CPU_BASED_CR3_STORE_EXITING);
3020 		vcpu->arch.cr0 = cr0;
3021 		vmx_set_cr4(vcpu, kvm_read_cr4(vcpu));
3022 	}
3023 
3024 	if (!(cr0 & X86_CR0_WP))
3025 		*hw_cr0 &= ~X86_CR0_WP;
3026 }
3027 
3028 void vmx_set_cr0(struct kvm_vcpu *vcpu, unsigned long cr0)
3029 {
3030 	struct vcpu_vmx *vmx = to_vmx(vcpu);
3031 	unsigned long hw_cr0;
3032 
3033 	hw_cr0 = (cr0 & ~KVM_VM_CR0_ALWAYS_OFF);
3034 	if (enable_unrestricted_guest)
3035 		hw_cr0 |= KVM_VM_CR0_ALWAYS_ON_UNRESTRICTED_GUEST;
3036 	else {
3037 		hw_cr0 |= KVM_VM_CR0_ALWAYS_ON;
3038 
3039 		if (vmx->rmode.vm86_active && (cr0 & X86_CR0_PE))
3040 			enter_pmode(vcpu);
3041 
3042 		if (!vmx->rmode.vm86_active && !(cr0 & X86_CR0_PE))
3043 			enter_rmode(vcpu);
3044 	}
3045 
3046 #ifdef CONFIG_X86_64
3047 	if (vcpu->arch.efer & EFER_LME) {
3048 		if (!is_paging(vcpu) && (cr0 & X86_CR0_PG))
3049 			enter_lmode(vcpu);
3050 		if (is_paging(vcpu) && !(cr0 & X86_CR0_PG))
3051 			exit_lmode(vcpu);
3052 	}
3053 #endif
3054 
3055 	if (enable_ept && !enable_unrestricted_guest)
3056 		ept_update_paging_mode_cr0(&hw_cr0, cr0, vcpu);
3057 
3058 	vmcs_writel(CR0_READ_SHADOW, cr0);
3059 	vmcs_writel(GUEST_CR0, hw_cr0);
3060 	vcpu->arch.cr0 = cr0;
3061 	kvm_register_mark_available(vcpu, VCPU_EXREG_CR0);
3062 
3063 	/* depends on vcpu->arch.cr0 to be set to a new value */
3064 	vmx->emulation_required = emulation_required(vcpu);
3065 }
3066 
3067 static int vmx_get_tdp_level(struct kvm_vcpu *vcpu)
3068 {
3069 	if (cpu_has_vmx_ept_5levels() && (cpuid_maxphyaddr(vcpu) > 48))
3070 		return 5;
3071 	return 4;
3072 }
3073 
3074 static int get_ept_level(struct kvm_vcpu *vcpu)
3075 {
3076 	if (is_guest_mode(vcpu) && nested_cpu_has_ept(get_vmcs12(vcpu)))
3077 		return vmx_eptp_page_walk_level(nested_ept_get_eptp(vcpu));
3078 
3079 	return vmx_get_tdp_level(vcpu);
3080 }
3081 
3082 u64 construct_eptp(struct kvm_vcpu *vcpu, unsigned long root_hpa)
3083 {
3084 	u64 eptp = VMX_EPTP_MT_WB;
3085 
3086 	eptp |= (get_ept_level(vcpu) == 5) ? VMX_EPTP_PWL_5 : VMX_EPTP_PWL_4;
3087 
3088 	if (enable_ept_ad_bits &&
3089 	    (!is_guest_mode(vcpu) || nested_ept_ad_enabled(vcpu)))
3090 		eptp |= VMX_EPTP_AD_ENABLE_BIT;
3091 	eptp |= (root_hpa & PAGE_MASK);
3092 
3093 	return eptp;
3094 }
3095 
3096 void vmx_load_mmu_pgd(struct kvm_vcpu *vcpu, unsigned long pgd)
3097 {
3098 	struct kvm *kvm = vcpu->kvm;
3099 	bool update_guest_cr3 = true;
3100 	unsigned long guest_cr3;
3101 	u64 eptp;
3102 
3103 	if (enable_ept) {
3104 		eptp = construct_eptp(vcpu, pgd);
3105 		vmcs_write64(EPT_POINTER, eptp);
3106 
3107 		if (kvm_x86_ops.tlb_remote_flush) {
3108 			spin_lock(&to_kvm_vmx(kvm)->ept_pointer_lock);
3109 			to_vmx(vcpu)->ept_pointer = eptp;
3110 			to_kvm_vmx(kvm)->ept_pointers_match
3111 				= EPT_POINTERS_CHECK;
3112 			spin_unlock(&to_kvm_vmx(kvm)->ept_pointer_lock);
3113 		}
3114 
3115 		if (!enable_unrestricted_guest && !is_paging(vcpu))
3116 			guest_cr3 = to_kvm_vmx(kvm)->ept_identity_map_addr;
3117 		else if (test_bit(VCPU_EXREG_CR3, (ulong *)&vcpu->arch.regs_avail))
3118 			guest_cr3 = vcpu->arch.cr3;
3119 		else /* vmcs01.GUEST_CR3 is already up-to-date. */
3120 			update_guest_cr3 = false;
3121 		ept_load_pdptrs(vcpu);
3122 	} else {
3123 		guest_cr3 = pgd;
3124 	}
3125 
3126 	if (update_guest_cr3)
3127 		vmcs_writel(GUEST_CR3, guest_cr3);
3128 }
3129 
3130 int vmx_set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4)
3131 {
3132 	struct vcpu_vmx *vmx = to_vmx(vcpu);
3133 	/*
3134 	 * Pass through host's Machine Check Enable value to hw_cr4, which
3135 	 * is in force while we are in guest mode.  Do not let guests control
3136 	 * this bit, even if host CR4.MCE == 0.
3137 	 */
3138 	unsigned long hw_cr4;
3139 
3140 	hw_cr4 = (cr4_read_shadow() & X86_CR4_MCE) | (cr4 & ~X86_CR4_MCE);
3141 	if (enable_unrestricted_guest)
3142 		hw_cr4 |= KVM_VM_CR4_ALWAYS_ON_UNRESTRICTED_GUEST;
3143 	else if (vmx->rmode.vm86_active)
3144 		hw_cr4 |= KVM_RMODE_VM_CR4_ALWAYS_ON;
3145 	else
3146 		hw_cr4 |= KVM_PMODE_VM_CR4_ALWAYS_ON;
3147 
3148 	if (!boot_cpu_has(X86_FEATURE_UMIP) && vmx_umip_emulated()) {
3149 		if (cr4 & X86_CR4_UMIP) {
3150 			secondary_exec_controls_setbit(vmx, SECONDARY_EXEC_DESC);
3151 			hw_cr4 &= ~X86_CR4_UMIP;
3152 		} else if (!is_guest_mode(vcpu) ||
3153 			!nested_cpu_has2(get_vmcs12(vcpu), SECONDARY_EXEC_DESC)) {
3154 			secondary_exec_controls_clearbit(vmx, SECONDARY_EXEC_DESC);
3155 		}
3156 	}
3157 
3158 	if (cr4 & X86_CR4_VMXE) {
3159 		/*
3160 		 * To use VMXON (and later other VMX instructions), a guest
3161 		 * must first be able to turn on cr4.VMXE (see handle_vmon()).
3162 		 * So basically the check on whether to allow nested VMX
3163 		 * is here.  We operate under the default treatment of SMM,
3164 		 * so VMX cannot be enabled under SMM.
3165 		 */
3166 		if (!nested_vmx_allowed(vcpu) || is_smm(vcpu))
3167 			return 1;
3168 	}
3169 
3170 	if (vmx->nested.vmxon && !nested_cr4_valid(vcpu, cr4))
3171 		return 1;
3172 
3173 	vcpu->arch.cr4 = cr4;
3174 	kvm_register_mark_available(vcpu, VCPU_EXREG_CR4);
3175 
3176 	if (!enable_unrestricted_guest) {
3177 		if (enable_ept) {
3178 			if (!is_paging(vcpu)) {
3179 				hw_cr4 &= ~X86_CR4_PAE;
3180 				hw_cr4 |= X86_CR4_PSE;
3181 			} else if (!(cr4 & X86_CR4_PAE)) {
3182 				hw_cr4 &= ~X86_CR4_PAE;
3183 			}
3184 		}
3185 
3186 		/*
3187 		 * SMEP/SMAP/PKU is disabled if CPU is in non-paging mode in
3188 		 * hardware.  To emulate this behavior, SMEP/SMAP/PKU needs
3189 		 * to be manually disabled when guest switches to non-paging
3190 		 * mode.
3191 		 *
3192 		 * If !enable_unrestricted_guest, the CPU is always running
3193 		 * with CR0.PG=1 and CR4 needs to be modified.
3194 		 * If enable_unrestricted_guest, the CPU automatically
3195 		 * disables SMEP/SMAP/PKU when the guest sets CR0.PG=0.
3196 		 */
3197 		if (!is_paging(vcpu))
3198 			hw_cr4 &= ~(X86_CR4_SMEP | X86_CR4_SMAP | X86_CR4_PKE);
3199 	}
3200 
3201 	vmcs_writel(CR4_READ_SHADOW, cr4);
3202 	vmcs_writel(GUEST_CR4, hw_cr4);
3203 	return 0;
3204 }
3205 
3206 void vmx_get_segment(struct kvm_vcpu *vcpu, struct kvm_segment *var, int seg)
3207 {
3208 	struct vcpu_vmx *vmx = to_vmx(vcpu);
3209 	u32 ar;
3210 
3211 	if (vmx->rmode.vm86_active && seg != VCPU_SREG_LDTR) {
3212 		*var = vmx->rmode.segs[seg];
3213 		if (seg == VCPU_SREG_TR
3214 		    || var->selector == vmx_read_guest_seg_selector(vmx, seg))
3215 			return;
3216 		var->base = vmx_read_guest_seg_base(vmx, seg);
3217 		var->selector = vmx_read_guest_seg_selector(vmx, seg);
3218 		return;
3219 	}
3220 	var->base = vmx_read_guest_seg_base(vmx, seg);
3221 	var->limit = vmx_read_guest_seg_limit(vmx, seg);
3222 	var->selector = vmx_read_guest_seg_selector(vmx, seg);
3223 	ar = vmx_read_guest_seg_ar(vmx, seg);
3224 	var->unusable = (ar >> 16) & 1;
3225 	var->type = ar & 15;
3226 	var->s = (ar >> 4) & 1;
3227 	var->dpl = (ar >> 5) & 3;
3228 	/*
3229 	 * Some userspaces do not preserve unusable property. Since usable
3230 	 * segment has to be present according to VMX spec we can use present
3231 	 * property to amend userspace bug by making unusable segment always
3232 	 * nonpresent. vmx_segment_access_rights() already marks nonpresent
3233 	 * segment as unusable.
3234 	 */
3235 	var->present = !var->unusable;
3236 	var->avl = (ar >> 12) & 1;
3237 	var->l = (ar >> 13) & 1;
3238 	var->db = (ar >> 14) & 1;
3239 	var->g = (ar >> 15) & 1;
3240 }
3241 
3242 static u64 vmx_get_segment_base(struct kvm_vcpu *vcpu, int seg)
3243 {
3244 	struct kvm_segment s;
3245 
3246 	if (to_vmx(vcpu)->rmode.vm86_active) {
3247 		vmx_get_segment(vcpu, &s, seg);
3248 		return s.base;
3249 	}
3250 	return vmx_read_guest_seg_base(to_vmx(vcpu), seg);
3251 }
3252 
3253 int vmx_get_cpl(struct kvm_vcpu *vcpu)
3254 {
3255 	struct vcpu_vmx *vmx = to_vmx(vcpu);
3256 
3257 	if (unlikely(vmx->rmode.vm86_active))
3258 		return 0;
3259 	else {
3260 		int ar = vmx_read_guest_seg_ar(vmx, VCPU_SREG_SS);
3261 		return VMX_AR_DPL(ar);
3262 	}
3263 }
3264 
3265 static u32 vmx_segment_access_rights(struct kvm_segment *var)
3266 {
3267 	u32 ar;
3268 
3269 	if (var->unusable || !var->present)
3270 		ar = 1 << 16;
3271 	else {
3272 		ar = var->type & 15;
3273 		ar |= (var->s & 1) << 4;
3274 		ar |= (var->dpl & 3) << 5;
3275 		ar |= (var->present & 1) << 7;
3276 		ar |= (var->avl & 1) << 12;
3277 		ar |= (var->l & 1) << 13;
3278 		ar |= (var->db & 1) << 14;
3279 		ar |= (var->g & 1) << 15;
3280 	}
3281 
3282 	return ar;
3283 }
3284 
3285 void vmx_set_segment(struct kvm_vcpu *vcpu, struct kvm_segment *var, int seg)
3286 {
3287 	struct vcpu_vmx *vmx = to_vmx(vcpu);
3288 	const struct kvm_vmx_segment_field *sf = &kvm_vmx_segment_fields[seg];
3289 
3290 	vmx_segment_cache_clear(vmx);
3291 
3292 	if (vmx->rmode.vm86_active && seg != VCPU_SREG_LDTR) {
3293 		vmx->rmode.segs[seg] = *var;
3294 		if (seg == VCPU_SREG_TR)
3295 			vmcs_write16(sf->selector, var->selector);
3296 		else if (var->s)
3297 			fix_rmode_seg(seg, &vmx->rmode.segs[seg]);
3298 		goto out;
3299 	}
3300 
3301 	vmcs_writel(sf->base, var->base);
3302 	vmcs_write32(sf->limit, var->limit);
3303 	vmcs_write16(sf->selector, var->selector);
3304 
3305 	/*
3306 	 *   Fix the "Accessed" bit in AR field of segment registers for older
3307 	 * qemu binaries.
3308 	 *   IA32 arch specifies that at the time of processor reset the
3309 	 * "Accessed" bit in the AR field of segment registers is 1. And qemu
3310 	 * is setting it to 0 in the userland code. This causes invalid guest
3311 	 * state vmexit when "unrestricted guest" mode is turned on.
3312 	 *    Fix for this setup issue in cpu_reset is being pushed in the qemu
3313 	 * tree. Newer qemu binaries with that qemu fix would not need this
3314 	 * kvm hack.
3315 	 */
3316 	if (enable_unrestricted_guest && (seg != VCPU_SREG_LDTR))
3317 		var->type |= 0x1; /* Accessed */
3318 
3319 	vmcs_write32(sf->ar_bytes, vmx_segment_access_rights(var));
3320 
3321 out:
3322 	vmx->emulation_required = emulation_required(vcpu);
3323 }
3324 
3325 static void vmx_get_cs_db_l_bits(struct kvm_vcpu *vcpu, int *db, int *l)
3326 {
3327 	u32 ar = vmx_read_guest_seg_ar(to_vmx(vcpu), VCPU_SREG_CS);
3328 
3329 	*db = (ar >> 14) & 1;
3330 	*l = (ar >> 13) & 1;
3331 }
3332 
3333 static void vmx_get_idt(struct kvm_vcpu *vcpu, struct desc_ptr *dt)
3334 {
3335 	dt->size = vmcs_read32(GUEST_IDTR_LIMIT);
3336 	dt->address = vmcs_readl(GUEST_IDTR_BASE);
3337 }
3338 
3339 static void vmx_set_idt(struct kvm_vcpu *vcpu, struct desc_ptr *dt)
3340 {
3341 	vmcs_write32(GUEST_IDTR_LIMIT, dt->size);
3342 	vmcs_writel(GUEST_IDTR_BASE, dt->address);
3343 }
3344 
3345 static void vmx_get_gdt(struct kvm_vcpu *vcpu, struct desc_ptr *dt)
3346 {
3347 	dt->size = vmcs_read32(GUEST_GDTR_LIMIT);
3348 	dt->address = vmcs_readl(GUEST_GDTR_BASE);
3349 }
3350 
3351 static void vmx_set_gdt(struct kvm_vcpu *vcpu, struct desc_ptr *dt)
3352 {
3353 	vmcs_write32(GUEST_GDTR_LIMIT, dt->size);
3354 	vmcs_writel(GUEST_GDTR_BASE, dt->address);
3355 }
3356 
3357 static bool rmode_segment_valid(struct kvm_vcpu *vcpu, int seg)
3358 {
3359 	struct kvm_segment var;
3360 	u32 ar;
3361 
3362 	vmx_get_segment(vcpu, &var, seg);
3363 	var.dpl = 0x3;
3364 	if (seg == VCPU_SREG_CS)
3365 		var.type = 0x3;
3366 	ar = vmx_segment_access_rights(&var);
3367 
3368 	if (var.base != (var.selector << 4))
3369 		return false;
3370 	if (var.limit != 0xffff)
3371 		return false;
3372 	if (ar != 0xf3)
3373 		return false;
3374 
3375 	return true;
3376 }
3377 
3378 static bool code_segment_valid(struct kvm_vcpu *vcpu)
3379 {
3380 	struct kvm_segment cs;
3381 	unsigned int cs_rpl;
3382 
3383 	vmx_get_segment(vcpu, &cs, VCPU_SREG_CS);
3384 	cs_rpl = cs.selector & SEGMENT_RPL_MASK;
3385 
3386 	if (cs.unusable)
3387 		return false;
3388 	if (~cs.type & (VMX_AR_TYPE_CODE_MASK|VMX_AR_TYPE_ACCESSES_MASK))
3389 		return false;
3390 	if (!cs.s)
3391 		return false;
3392 	if (cs.type & VMX_AR_TYPE_WRITEABLE_MASK) {
3393 		if (cs.dpl > cs_rpl)
3394 			return false;
3395 	} else {
3396 		if (cs.dpl != cs_rpl)
3397 			return false;
3398 	}
3399 	if (!cs.present)
3400 		return false;
3401 
3402 	/* TODO: Add Reserved field check, this'll require a new member in the kvm_segment_field structure */
3403 	return true;
3404 }
3405 
3406 static bool stack_segment_valid(struct kvm_vcpu *vcpu)
3407 {
3408 	struct kvm_segment ss;
3409 	unsigned int ss_rpl;
3410 
3411 	vmx_get_segment(vcpu, &ss, VCPU_SREG_SS);
3412 	ss_rpl = ss.selector & SEGMENT_RPL_MASK;
3413 
3414 	if (ss.unusable)
3415 		return true;
3416 	if (ss.type != 3 && ss.type != 7)
3417 		return false;
3418 	if (!ss.s)
3419 		return false;
3420 	if (ss.dpl != ss_rpl) /* DPL != RPL */
3421 		return false;
3422 	if (!ss.present)
3423 		return false;
3424 
3425 	return true;
3426 }
3427 
3428 static bool data_segment_valid(struct kvm_vcpu *vcpu, int seg)
3429 {
3430 	struct kvm_segment var;
3431 	unsigned int rpl;
3432 
3433 	vmx_get_segment(vcpu, &var, seg);
3434 	rpl = var.selector & SEGMENT_RPL_MASK;
3435 
3436 	if (var.unusable)
3437 		return true;
3438 	if (!var.s)
3439 		return false;
3440 	if (!var.present)
3441 		return false;
3442 	if (~var.type & (VMX_AR_TYPE_CODE_MASK|VMX_AR_TYPE_WRITEABLE_MASK)) {
3443 		if (var.dpl < rpl) /* DPL < RPL */
3444 			return false;
3445 	}
3446 
3447 	/* TODO: Add other members to kvm_segment_field to allow checking for other access
3448 	 * rights flags
3449 	 */
3450 	return true;
3451 }
3452 
3453 static bool tr_valid(struct kvm_vcpu *vcpu)
3454 {
3455 	struct kvm_segment tr;
3456 
3457 	vmx_get_segment(vcpu, &tr, VCPU_SREG_TR);
3458 
3459 	if (tr.unusable)
3460 		return false;
3461 	if (tr.selector & SEGMENT_TI_MASK)	/* TI = 1 */
3462 		return false;
3463 	if (tr.type != 3 && tr.type != 11) /* TODO: Check if guest is in IA32e mode */
3464 		return false;
3465 	if (!tr.present)
3466 		return false;
3467 
3468 	return true;
3469 }
3470 
3471 static bool ldtr_valid(struct kvm_vcpu *vcpu)
3472 {
3473 	struct kvm_segment ldtr;
3474 
3475 	vmx_get_segment(vcpu, &ldtr, VCPU_SREG_LDTR);
3476 
3477 	if (ldtr.unusable)
3478 		return true;
3479 	if (ldtr.selector & SEGMENT_TI_MASK)	/* TI = 1 */
3480 		return false;
3481 	if (ldtr.type != 2)
3482 		return false;
3483 	if (!ldtr.present)
3484 		return false;
3485 
3486 	return true;
3487 }
3488 
3489 static bool cs_ss_rpl_check(struct kvm_vcpu *vcpu)
3490 {
3491 	struct kvm_segment cs, ss;
3492 
3493 	vmx_get_segment(vcpu, &cs, VCPU_SREG_CS);
3494 	vmx_get_segment(vcpu, &ss, VCPU_SREG_SS);
3495 
3496 	return ((cs.selector & SEGMENT_RPL_MASK) ==
3497 		 (ss.selector & SEGMENT_RPL_MASK));
3498 }
3499 
3500 /*
3501  * Check if guest state is valid. Returns true if valid, false if
3502  * not.
3503  * We assume that registers are always usable
3504  */
3505 static bool guest_state_valid(struct kvm_vcpu *vcpu)
3506 {
3507 	if (enable_unrestricted_guest)
3508 		return true;
3509 
3510 	/* real mode guest state checks */
3511 	if (!is_protmode(vcpu) || (vmx_get_rflags(vcpu) & X86_EFLAGS_VM)) {
3512 		if (!rmode_segment_valid(vcpu, VCPU_SREG_CS))
3513 			return false;
3514 		if (!rmode_segment_valid(vcpu, VCPU_SREG_SS))
3515 			return false;
3516 		if (!rmode_segment_valid(vcpu, VCPU_SREG_DS))
3517 			return false;
3518 		if (!rmode_segment_valid(vcpu, VCPU_SREG_ES))
3519 			return false;
3520 		if (!rmode_segment_valid(vcpu, VCPU_SREG_FS))
3521 			return false;
3522 		if (!rmode_segment_valid(vcpu, VCPU_SREG_GS))
3523 			return false;
3524 	} else {
3525 	/* protected mode guest state checks */
3526 		if (!cs_ss_rpl_check(vcpu))
3527 			return false;
3528 		if (!code_segment_valid(vcpu))
3529 			return false;
3530 		if (!stack_segment_valid(vcpu))
3531 			return false;
3532 		if (!data_segment_valid(vcpu, VCPU_SREG_DS))
3533 			return false;
3534 		if (!data_segment_valid(vcpu, VCPU_SREG_ES))
3535 			return false;
3536 		if (!data_segment_valid(vcpu, VCPU_SREG_FS))
3537 			return false;
3538 		if (!data_segment_valid(vcpu, VCPU_SREG_GS))
3539 			return false;
3540 		if (!tr_valid(vcpu))
3541 			return false;
3542 		if (!ldtr_valid(vcpu))
3543 			return false;
3544 	}
3545 	/* TODO:
3546 	 * - Add checks on RIP
3547 	 * - Add checks on RFLAGS
3548 	 */
3549 
3550 	return true;
3551 }
3552 
3553 static int init_rmode_tss(struct kvm *kvm)
3554 {
3555 	gfn_t fn;
3556 	u16 data = 0;
3557 	int idx, r;
3558 
3559 	idx = srcu_read_lock(&kvm->srcu);
3560 	fn = to_kvm_vmx(kvm)->tss_addr >> PAGE_SHIFT;
3561 	r = kvm_clear_guest_page(kvm, fn, 0, PAGE_SIZE);
3562 	if (r < 0)
3563 		goto out;
3564 	data = TSS_BASE_SIZE + TSS_REDIRECTION_SIZE;
3565 	r = kvm_write_guest_page(kvm, fn++, &data,
3566 			TSS_IOPB_BASE_OFFSET, sizeof(u16));
3567 	if (r < 0)
3568 		goto out;
3569 	r = kvm_clear_guest_page(kvm, fn++, 0, PAGE_SIZE);
3570 	if (r < 0)
3571 		goto out;
3572 	r = kvm_clear_guest_page(kvm, fn, 0, PAGE_SIZE);
3573 	if (r < 0)
3574 		goto out;
3575 	data = ~0;
3576 	r = kvm_write_guest_page(kvm, fn, &data,
3577 				 RMODE_TSS_SIZE - 2 * PAGE_SIZE - 1,
3578 				 sizeof(u8));
3579 out:
3580 	srcu_read_unlock(&kvm->srcu, idx);
3581 	return r;
3582 }
3583 
3584 static int init_rmode_identity_map(struct kvm *kvm)
3585 {
3586 	struct kvm_vmx *kvm_vmx = to_kvm_vmx(kvm);
3587 	int i, r = 0;
3588 	kvm_pfn_t identity_map_pfn;
3589 	u32 tmp;
3590 
3591 	/* Protect kvm_vmx->ept_identity_pagetable_done. */
3592 	mutex_lock(&kvm->slots_lock);
3593 
3594 	if (likely(kvm_vmx->ept_identity_pagetable_done))
3595 		goto out;
3596 
3597 	if (!kvm_vmx->ept_identity_map_addr)
3598 		kvm_vmx->ept_identity_map_addr = VMX_EPT_IDENTITY_PAGETABLE_ADDR;
3599 	identity_map_pfn = kvm_vmx->ept_identity_map_addr >> PAGE_SHIFT;
3600 
3601 	r = __x86_set_memory_region(kvm, IDENTITY_PAGETABLE_PRIVATE_MEMSLOT,
3602 				    kvm_vmx->ept_identity_map_addr, PAGE_SIZE);
3603 	if (r < 0)
3604 		goto out;
3605 
3606 	r = kvm_clear_guest_page(kvm, identity_map_pfn, 0, PAGE_SIZE);
3607 	if (r < 0)
3608 		goto out;
3609 	/* Set up identity-mapping pagetable for EPT in real mode */
3610 	for (i = 0; i < PT32_ENT_PER_PAGE; i++) {
3611 		tmp = (i << 22) + (_PAGE_PRESENT | _PAGE_RW | _PAGE_USER |
3612 			_PAGE_ACCESSED | _PAGE_DIRTY | _PAGE_PSE);
3613 		r = kvm_write_guest_page(kvm, identity_map_pfn,
3614 				&tmp, i * sizeof(tmp), sizeof(tmp));
3615 		if (r < 0)
3616 			goto out;
3617 	}
3618 	kvm_vmx->ept_identity_pagetable_done = true;
3619 
3620 out:
3621 	mutex_unlock(&kvm->slots_lock);
3622 	return r;
3623 }
3624 
3625 static void seg_setup(int seg)
3626 {
3627 	const struct kvm_vmx_segment_field *sf = &kvm_vmx_segment_fields[seg];
3628 	unsigned int ar;
3629 
3630 	vmcs_write16(sf->selector, 0);
3631 	vmcs_writel(sf->base, 0);
3632 	vmcs_write32(sf->limit, 0xffff);
3633 	ar = 0x93;
3634 	if (seg == VCPU_SREG_CS)
3635 		ar |= 0x08; /* code segment */
3636 
3637 	vmcs_write32(sf->ar_bytes, ar);
3638 }
3639 
3640 static int alloc_apic_access_page(struct kvm *kvm)
3641 {
3642 	struct page *page;
3643 	int r = 0;
3644 
3645 	mutex_lock(&kvm->slots_lock);
3646 	if (kvm->arch.apic_access_page_done)
3647 		goto out;
3648 	r = __x86_set_memory_region(kvm, APIC_ACCESS_PAGE_PRIVATE_MEMSLOT,
3649 				    APIC_DEFAULT_PHYS_BASE, PAGE_SIZE);
3650 	if (r)
3651 		goto out;
3652 
3653 	page = gfn_to_page(kvm, APIC_DEFAULT_PHYS_BASE >> PAGE_SHIFT);
3654 	if (is_error_page(page)) {
3655 		r = -EFAULT;
3656 		goto out;
3657 	}
3658 
3659 	/*
3660 	 * Do not pin the page in memory, so that memory hot-unplug
3661 	 * is able to migrate it.
3662 	 */
3663 	put_page(page);
3664 	kvm->arch.apic_access_page_done = true;
3665 out:
3666 	mutex_unlock(&kvm->slots_lock);
3667 	return r;
3668 }
3669 
3670 int allocate_vpid(void)
3671 {
3672 	int vpid;
3673 
3674 	if (!enable_vpid)
3675 		return 0;
3676 	spin_lock(&vmx_vpid_lock);
3677 	vpid = find_first_zero_bit(vmx_vpid_bitmap, VMX_NR_VPIDS);
3678 	if (vpid < VMX_NR_VPIDS)
3679 		__set_bit(vpid, vmx_vpid_bitmap);
3680 	else
3681 		vpid = 0;
3682 	spin_unlock(&vmx_vpid_lock);
3683 	return vpid;
3684 }
3685 
3686 void free_vpid(int vpid)
3687 {
3688 	if (!enable_vpid || vpid == 0)
3689 		return;
3690 	spin_lock(&vmx_vpid_lock);
3691 	__clear_bit(vpid, vmx_vpid_bitmap);
3692 	spin_unlock(&vmx_vpid_lock);
3693 }
3694 
3695 static __always_inline void vmx_disable_intercept_for_msr(unsigned long *msr_bitmap,
3696 							  u32 msr, int type)
3697 {
3698 	int f = sizeof(unsigned long);
3699 
3700 	if (!cpu_has_vmx_msr_bitmap())
3701 		return;
3702 
3703 	if (static_branch_unlikely(&enable_evmcs))
3704 		evmcs_touch_msr_bitmap();
3705 
3706 	/*
3707 	 * See Intel PRM Vol. 3, 20.6.9 (MSR-Bitmap Address). Early manuals
3708 	 * have the write-low and read-high bitmap offsets the wrong way round.
3709 	 * We can control MSRs 0x00000000-0x00001fff and 0xc0000000-0xc0001fff.
3710 	 */
3711 	if (msr <= 0x1fff) {
3712 		if (type & MSR_TYPE_R)
3713 			/* read-low */
3714 			__clear_bit(msr, msr_bitmap + 0x000 / f);
3715 
3716 		if (type & MSR_TYPE_W)
3717 			/* write-low */
3718 			__clear_bit(msr, msr_bitmap + 0x800 / f);
3719 
3720 	} else if ((msr >= 0xc0000000) && (msr <= 0xc0001fff)) {
3721 		msr &= 0x1fff;
3722 		if (type & MSR_TYPE_R)
3723 			/* read-high */
3724 			__clear_bit(msr, msr_bitmap + 0x400 / f);
3725 
3726 		if (type & MSR_TYPE_W)
3727 			/* write-high */
3728 			__clear_bit(msr, msr_bitmap + 0xc00 / f);
3729 
3730 	}
3731 }
3732 
3733 static __always_inline void vmx_enable_intercept_for_msr(unsigned long *msr_bitmap,
3734 							 u32 msr, int type)
3735 {
3736 	int f = sizeof(unsigned long);
3737 
3738 	if (!cpu_has_vmx_msr_bitmap())
3739 		return;
3740 
3741 	if (static_branch_unlikely(&enable_evmcs))
3742 		evmcs_touch_msr_bitmap();
3743 
3744 	/*
3745 	 * See Intel PRM Vol. 3, 20.6.9 (MSR-Bitmap Address). Early manuals
3746 	 * have the write-low and read-high bitmap offsets the wrong way round.
3747 	 * We can control MSRs 0x00000000-0x00001fff and 0xc0000000-0xc0001fff.
3748 	 */
3749 	if (msr <= 0x1fff) {
3750 		if (type & MSR_TYPE_R)
3751 			/* read-low */
3752 			__set_bit(msr, msr_bitmap + 0x000 / f);
3753 
3754 		if (type & MSR_TYPE_W)
3755 			/* write-low */
3756 			__set_bit(msr, msr_bitmap + 0x800 / f);
3757 
3758 	} else if ((msr >= 0xc0000000) && (msr <= 0xc0001fff)) {
3759 		msr &= 0x1fff;
3760 		if (type & MSR_TYPE_R)
3761 			/* read-high */
3762 			__set_bit(msr, msr_bitmap + 0x400 / f);
3763 
3764 		if (type & MSR_TYPE_W)
3765 			/* write-high */
3766 			__set_bit(msr, msr_bitmap + 0xc00 / f);
3767 
3768 	}
3769 }
3770 
3771 static __always_inline void vmx_set_intercept_for_msr(unsigned long *msr_bitmap,
3772 			     			      u32 msr, int type, bool value)
3773 {
3774 	if (value)
3775 		vmx_enable_intercept_for_msr(msr_bitmap, msr, type);
3776 	else
3777 		vmx_disable_intercept_for_msr(msr_bitmap, msr, type);
3778 }
3779 
3780 static u8 vmx_msr_bitmap_mode(struct kvm_vcpu *vcpu)
3781 {
3782 	u8 mode = 0;
3783 
3784 	if (cpu_has_secondary_exec_ctrls() &&
3785 	    (secondary_exec_controls_get(to_vmx(vcpu)) &
3786 	     SECONDARY_EXEC_VIRTUALIZE_X2APIC_MODE)) {
3787 		mode |= MSR_BITMAP_MODE_X2APIC;
3788 		if (enable_apicv && kvm_vcpu_apicv_active(vcpu))
3789 			mode |= MSR_BITMAP_MODE_X2APIC_APICV;
3790 	}
3791 
3792 	return mode;
3793 }
3794 
3795 static void vmx_update_msr_bitmap_x2apic(unsigned long *msr_bitmap,
3796 					 u8 mode)
3797 {
3798 	int msr;
3799 
3800 	for (msr = 0x800; msr <= 0x8ff; msr += BITS_PER_LONG) {
3801 		unsigned word = msr / BITS_PER_LONG;
3802 		msr_bitmap[word] = (mode & MSR_BITMAP_MODE_X2APIC_APICV) ? 0 : ~0;
3803 		msr_bitmap[word + (0x800 / sizeof(long))] = ~0;
3804 	}
3805 
3806 	if (mode & MSR_BITMAP_MODE_X2APIC) {
3807 		/*
3808 		 * TPR reads and writes can be virtualized even if virtual interrupt
3809 		 * delivery is not in use.
3810 		 */
3811 		vmx_disable_intercept_for_msr(msr_bitmap, X2APIC_MSR(APIC_TASKPRI), MSR_TYPE_RW);
3812 		if (mode & MSR_BITMAP_MODE_X2APIC_APICV) {
3813 			vmx_enable_intercept_for_msr(msr_bitmap, X2APIC_MSR(APIC_TMCCT), MSR_TYPE_R);
3814 			vmx_disable_intercept_for_msr(msr_bitmap, X2APIC_MSR(APIC_EOI), MSR_TYPE_W);
3815 			vmx_disable_intercept_for_msr(msr_bitmap, X2APIC_MSR(APIC_SELF_IPI), MSR_TYPE_W);
3816 		}
3817 	}
3818 }
3819 
3820 void vmx_update_msr_bitmap(struct kvm_vcpu *vcpu)
3821 {
3822 	struct vcpu_vmx *vmx = to_vmx(vcpu);
3823 	unsigned long *msr_bitmap = vmx->vmcs01.msr_bitmap;
3824 	u8 mode = vmx_msr_bitmap_mode(vcpu);
3825 	u8 changed = mode ^ vmx->msr_bitmap_mode;
3826 
3827 	if (!changed)
3828 		return;
3829 
3830 	if (changed & (MSR_BITMAP_MODE_X2APIC | MSR_BITMAP_MODE_X2APIC_APICV))
3831 		vmx_update_msr_bitmap_x2apic(msr_bitmap, mode);
3832 
3833 	vmx->msr_bitmap_mode = mode;
3834 }
3835 
3836 void pt_update_intercept_for_msr(struct vcpu_vmx *vmx)
3837 {
3838 	unsigned long *msr_bitmap = vmx->vmcs01.msr_bitmap;
3839 	bool flag = !(vmx->pt_desc.guest.ctl & RTIT_CTL_TRACEEN);
3840 	u32 i;
3841 
3842 	vmx_set_intercept_for_msr(msr_bitmap, MSR_IA32_RTIT_STATUS,
3843 							MSR_TYPE_RW, flag);
3844 	vmx_set_intercept_for_msr(msr_bitmap, MSR_IA32_RTIT_OUTPUT_BASE,
3845 							MSR_TYPE_RW, flag);
3846 	vmx_set_intercept_for_msr(msr_bitmap, MSR_IA32_RTIT_OUTPUT_MASK,
3847 							MSR_TYPE_RW, flag);
3848 	vmx_set_intercept_for_msr(msr_bitmap, MSR_IA32_RTIT_CR3_MATCH,
3849 							MSR_TYPE_RW, flag);
3850 	for (i = 0; i < vmx->pt_desc.addr_range; i++) {
3851 		vmx_set_intercept_for_msr(msr_bitmap,
3852 			MSR_IA32_RTIT_ADDR0_A + i * 2, MSR_TYPE_RW, flag);
3853 		vmx_set_intercept_for_msr(msr_bitmap,
3854 			MSR_IA32_RTIT_ADDR0_B + i * 2, MSR_TYPE_RW, flag);
3855 	}
3856 }
3857 
3858 static bool vmx_guest_apic_has_interrupt(struct kvm_vcpu *vcpu)
3859 {
3860 	struct vcpu_vmx *vmx = to_vmx(vcpu);
3861 	void *vapic_page;
3862 	u32 vppr;
3863 	int rvi;
3864 
3865 	if (WARN_ON_ONCE(!is_guest_mode(vcpu)) ||
3866 		!nested_cpu_has_vid(get_vmcs12(vcpu)) ||
3867 		WARN_ON_ONCE(!vmx->nested.virtual_apic_map.gfn))
3868 		return false;
3869 
3870 	rvi = vmx_get_rvi();
3871 
3872 	vapic_page = vmx->nested.virtual_apic_map.hva;
3873 	vppr = *((u32 *)(vapic_page + APIC_PROCPRI));
3874 
3875 	return ((rvi & 0xf0) > (vppr & 0xf0));
3876 }
3877 
3878 static inline bool kvm_vcpu_trigger_posted_interrupt(struct kvm_vcpu *vcpu,
3879 						     bool nested)
3880 {
3881 #ifdef CONFIG_SMP
3882 	int pi_vec = nested ? POSTED_INTR_NESTED_VECTOR : POSTED_INTR_VECTOR;
3883 
3884 	if (vcpu->mode == IN_GUEST_MODE) {
3885 		/*
3886 		 * The vector of interrupt to be delivered to vcpu had
3887 		 * been set in PIR before this function.
3888 		 *
3889 		 * Following cases will be reached in this block, and
3890 		 * we always send a notification event in all cases as
3891 		 * explained below.
3892 		 *
3893 		 * Case 1: vcpu keeps in non-root mode. Sending a
3894 		 * notification event posts the interrupt to vcpu.
3895 		 *
3896 		 * Case 2: vcpu exits to root mode and is still
3897 		 * runnable. PIR will be synced to vIRR before the
3898 		 * next vcpu entry. Sending a notification event in
3899 		 * this case has no effect, as vcpu is not in root
3900 		 * mode.
3901 		 *
3902 		 * Case 3: vcpu exits to root mode and is blocked.
3903 		 * vcpu_block() has already synced PIR to vIRR and
3904 		 * never blocks vcpu if vIRR is not cleared. Therefore,
3905 		 * a blocked vcpu here does not wait for any requested
3906 		 * interrupts in PIR, and sending a notification event
3907 		 * which has no effect is safe here.
3908 		 */
3909 
3910 		apic->send_IPI_mask(get_cpu_mask(vcpu->cpu), pi_vec);
3911 		return true;
3912 	}
3913 #endif
3914 	return false;
3915 }
3916 
3917 static int vmx_deliver_nested_posted_interrupt(struct kvm_vcpu *vcpu,
3918 						int vector)
3919 {
3920 	struct vcpu_vmx *vmx = to_vmx(vcpu);
3921 
3922 	if (is_guest_mode(vcpu) &&
3923 	    vector == vmx->nested.posted_intr_nv) {
3924 		/*
3925 		 * If a posted intr is not recognized by hardware,
3926 		 * we will accomplish it in the next vmentry.
3927 		 */
3928 		vmx->nested.pi_pending = true;
3929 		kvm_make_request(KVM_REQ_EVENT, vcpu);
3930 		/* the PIR and ON have been set by L1. */
3931 		if (!kvm_vcpu_trigger_posted_interrupt(vcpu, true))
3932 			kvm_vcpu_kick(vcpu);
3933 		return 0;
3934 	}
3935 	return -1;
3936 }
3937 /*
3938  * Send interrupt to vcpu via posted interrupt way.
3939  * 1. If target vcpu is running(non-root mode), send posted interrupt
3940  * notification to vcpu and hardware will sync PIR to vIRR atomically.
3941  * 2. If target vcpu isn't running(root mode), kick it to pick up the
3942  * interrupt from PIR in next vmentry.
3943  */
3944 static int vmx_deliver_posted_interrupt(struct kvm_vcpu *vcpu, int vector)
3945 {
3946 	struct vcpu_vmx *vmx = to_vmx(vcpu);
3947 	int r;
3948 
3949 	r = vmx_deliver_nested_posted_interrupt(vcpu, vector);
3950 	if (!r)
3951 		return 0;
3952 
3953 	if (!vcpu->arch.apicv_active)
3954 		return -1;
3955 
3956 	if (pi_test_and_set_pir(vector, &vmx->pi_desc))
3957 		return 0;
3958 
3959 	/* If a previous notification has sent the IPI, nothing to do.  */
3960 	if (pi_test_and_set_on(&vmx->pi_desc))
3961 		return 0;
3962 
3963 	if (vcpu != kvm_get_running_vcpu() &&
3964 	    !kvm_vcpu_trigger_posted_interrupt(vcpu, false))
3965 		kvm_vcpu_kick(vcpu);
3966 
3967 	return 0;
3968 }
3969 
3970 /*
3971  * Set up the vmcs's constant host-state fields, i.e., host-state fields that
3972  * will not change in the lifetime of the guest.
3973  * Note that host-state that does change is set elsewhere. E.g., host-state
3974  * that is set differently for each CPU is set in vmx_vcpu_load(), not here.
3975  */
3976 void vmx_set_constant_host_state(struct vcpu_vmx *vmx)
3977 {
3978 	u32 low32, high32;
3979 	unsigned long tmpl;
3980 	unsigned long cr0, cr3, cr4;
3981 
3982 	cr0 = read_cr0();
3983 	WARN_ON(cr0 & X86_CR0_TS);
3984 	vmcs_writel(HOST_CR0, cr0);  /* 22.2.3 */
3985 
3986 	/*
3987 	 * Save the most likely value for this task's CR3 in the VMCS.
3988 	 * We can't use __get_current_cr3_fast() because we're not atomic.
3989 	 */
3990 	cr3 = __read_cr3();
3991 	vmcs_writel(HOST_CR3, cr3);		/* 22.2.3  FIXME: shadow tables */
3992 	vmx->loaded_vmcs->host_state.cr3 = cr3;
3993 
3994 	/* Save the most likely value for this task's CR4 in the VMCS. */
3995 	cr4 = cr4_read_shadow();
3996 	vmcs_writel(HOST_CR4, cr4);			/* 22.2.3, 22.2.5 */
3997 	vmx->loaded_vmcs->host_state.cr4 = cr4;
3998 
3999 	vmcs_write16(HOST_CS_SELECTOR, __KERNEL_CS);  /* 22.2.4 */
4000 #ifdef CONFIG_X86_64
4001 	/*
4002 	 * Load null selectors, so we can avoid reloading them in
4003 	 * vmx_prepare_switch_to_host(), in case userspace uses
4004 	 * the null selectors too (the expected case).
4005 	 */
4006 	vmcs_write16(HOST_DS_SELECTOR, 0);
4007 	vmcs_write16(HOST_ES_SELECTOR, 0);
4008 #else
4009 	vmcs_write16(HOST_DS_SELECTOR, __KERNEL_DS);  /* 22.2.4 */
4010 	vmcs_write16(HOST_ES_SELECTOR, __KERNEL_DS);  /* 22.2.4 */
4011 #endif
4012 	vmcs_write16(HOST_SS_SELECTOR, __KERNEL_DS);  /* 22.2.4 */
4013 	vmcs_write16(HOST_TR_SELECTOR, GDT_ENTRY_TSS*8);  /* 22.2.4 */
4014 
4015 	vmcs_writel(HOST_IDTR_BASE, host_idt_base);   /* 22.2.4 */
4016 
4017 	vmcs_writel(HOST_RIP, (unsigned long)vmx_vmexit); /* 22.2.5 */
4018 
4019 	rdmsr(MSR_IA32_SYSENTER_CS, low32, high32);
4020 	vmcs_write32(HOST_IA32_SYSENTER_CS, low32);
4021 	rdmsrl(MSR_IA32_SYSENTER_EIP, tmpl);
4022 	vmcs_writel(HOST_IA32_SYSENTER_EIP, tmpl);   /* 22.2.3 */
4023 
4024 	if (vmcs_config.vmexit_ctrl & VM_EXIT_LOAD_IA32_PAT) {
4025 		rdmsr(MSR_IA32_CR_PAT, low32, high32);
4026 		vmcs_write64(HOST_IA32_PAT, low32 | ((u64) high32 << 32));
4027 	}
4028 
4029 	if (cpu_has_load_ia32_efer())
4030 		vmcs_write64(HOST_IA32_EFER, host_efer);
4031 }
4032 
4033 void set_cr4_guest_host_mask(struct vcpu_vmx *vmx)
4034 {
4035 	vmx->vcpu.arch.cr4_guest_owned_bits = KVM_POSSIBLE_CR4_GUEST_BITS;
4036 	if (!enable_ept)
4037 		vmx->vcpu.arch.cr4_guest_owned_bits &= ~X86_CR4_PGE;
4038 	if (is_guest_mode(&vmx->vcpu))
4039 		vmx->vcpu.arch.cr4_guest_owned_bits &=
4040 			~get_vmcs12(&vmx->vcpu)->cr4_guest_host_mask;
4041 	vmcs_writel(CR4_GUEST_HOST_MASK, ~vmx->vcpu.arch.cr4_guest_owned_bits);
4042 }
4043 
4044 u32 vmx_pin_based_exec_ctrl(struct vcpu_vmx *vmx)
4045 {
4046 	u32 pin_based_exec_ctrl = vmcs_config.pin_based_exec_ctrl;
4047 
4048 	if (!kvm_vcpu_apicv_active(&vmx->vcpu))
4049 		pin_based_exec_ctrl &= ~PIN_BASED_POSTED_INTR;
4050 
4051 	if (!enable_vnmi)
4052 		pin_based_exec_ctrl &= ~PIN_BASED_VIRTUAL_NMIS;
4053 
4054 	if (!enable_preemption_timer)
4055 		pin_based_exec_ctrl &= ~PIN_BASED_VMX_PREEMPTION_TIMER;
4056 
4057 	return pin_based_exec_ctrl;
4058 }
4059 
4060 static void vmx_refresh_apicv_exec_ctrl(struct kvm_vcpu *vcpu)
4061 {
4062 	struct vcpu_vmx *vmx = to_vmx(vcpu);
4063 
4064 	pin_controls_set(vmx, vmx_pin_based_exec_ctrl(vmx));
4065 	if (cpu_has_secondary_exec_ctrls()) {
4066 		if (kvm_vcpu_apicv_active(vcpu))
4067 			secondary_exec_controls_setbit(vmx,
4068 				      SECONDARY_EXEC_APIC_REGISTER_VIRT |
4069 				      SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY);
4070 		else
4071 			secondary_exec_controls_clearbit(vmx,
4072 					SECONDARY_EXEC_APIC_REGISTER_VIRT |
4073 					SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY);
4074 	}
4075 
4076 	if (cpu_has_vmx_msr_bitmap())
4077 		vmx_update_msr_bitmap(vcpu);
4078 }
4079 
4080 u32 vmx_exec_control(struct vcpu_vmx *vmx)
4081 {
4082 	u32 exec_control = vmcs_config.cpu_based_exec_ctrl;
4083 
4084 	if (vmx->vcpu.arch.switch_db_regs & KVM_DEBUGREG_WONT_EXIT)
4085 		exec_control &= ~CPU_BASED_MOV_DR_EXITING;
4086 
4087 	if (!cpu_need_tpr_shadow(&vmx->vcpu)) {
4088 		exec_control &= ~CPU_BASED_TPR_SHADOW;
4089 #ifdef CONFIG_X86_64
4090 		exec_control |= CPU_BASED_CR8_STORE_EXITING |
4091 				CPU_BASED_CR8_LOAD_EXITING;
4092 #endif
4093 	}
4094 	if (!enable_ept)
4095 		exec_control |= CPU_BASED_CR3_STORE_EXITING |
4096 				CPU_BASED_CR3_LOAD_EXITING  |
4097 				CPU_BASED_INVLPG_EXITING;
4098 	if (kvm_mwait_in_guest(vmx->vcpu.kvm))
4099 		exec_control &= ~(CPU_BASED_MWAIT_EXITING |
4100 				CPU_BASED_MONITOR_EXITING);
4101 	if (kvm_hlt_in_guest(vmx->vcpu.kvm))
4102 		exec_control &= ~CPU_BASED_HLT_EXITING;
4103 	return exec_control;
4104 }
4105 
4106 
4107 static void vmx_compute_secondary_exec_control(struct vcpu_vmx *vmx)
4108 {
4109 	struct kvm_vcpu *vcpu = &vmx->vcpu;
4110 
4111 	u32 exec_control = vmcs_config.cpu_based_2nd_exec_ctrl;
4112 
4113 	if (vmx_pt_mode_is_system())
4114 		exec_control &= ~(SECONDARY_EXEC_PT_USE_GPA | SECONDARY_EXEC_PT_CONCEAL_VMX);
4115 	if (!cpu_need_virtualize_apic_accesses(vcpu))
4116 		exec_control &= ~SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES;
4117 	if (vmx->vpid == 0)
4118 		exec_control &= ~SECONDARY_EXEC_ENABLE_VPID;
4119 	if (!enable_ept) {
4120 		exec_control &= ~SECONDARY_EXEC_ENABLE_EPT;
4121 		enable_unrestricted_guest = 0;
4122 	}
4123 	if (!enable_unrestricted_guest)
4124 		exec_control &= ~SECONDARY_EXEC_UNRESTRICTED_GUEST;
4125 	if (kvm_pause_in_guest(vmx->vcpu.kvm))
4126 		exec_control &= ~SECONDARY_EXEC_PAUSE_LOOP_EXITING;
4127 	if (!kvm_vcpu_apicv_active(vcpu))
4128 		exec_control &= ~(SECONDARY_EXEC_APIC_REGISTER_VIRT |
4129 				  SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY);
4130 	exec_control &= ~SECONDARY_EXEC_VIRTUALIZE_X2APIC_MODE;
4131 
4132 	/* SECONDARY_EXEC_DESC is enabled/disabled on writes to CR4.UMIP,
4133 	 * in vmx_set_cr4.  */
4134 	exec_control &= ~SECONDARY_EXEC_DESC;
4135 
4136 	/* SECONDARY_EXEC_SHADOW_VMCS is enabled when L1 executes VMPTRLD
4137 	   (handle_vmptrld).
4138 	   We can NOT enable shadow_vmcs here because we don't have yet
4139 	   a current VMCS12
4140 	*/
4141 	exec_control &= ~SECONDARY_EXEC_SHADOW_VMCS;
4142 
4143 	if (!enable_pml)
4144 		exec_control &= ~SECONDARY_EXEC_ENABLE_PML;
4145 
4146 	if (vmx_xsaves_supported()) {
4147 		/* Exposing XSAVES only when XSAVE is exposed */
4148 		bool xsaves_enabled =
4149 			boot_cpu_has(X86_FEATURE_XSAVE) &&
4150 			guest_cpuid_has(vcpu, X86_FEATURE_XSAVE) &&
4151 			guest_cpuid_has(vcpu, X86_FEATURE_XSAVES);
4152 
4153 		vcpu->arch.xsaves_enabled = xsaves_enabled;
4154 
4155 		if (!xsaves_enabled)
4156 			exec_control &= ~SECONDARY_EXEC_XSAVES;
4157 
4158 		if (nested) {
4159 			if (xsaves_enabled)
4160 				vmx->nested.msrs.secondary_ctls_high |=
4161 					SECONDARY_EXEC_XSAVES;
4162 			else
4163 				vmx->nested.msrs.secondary_ctls_high &=
4164 					~SECONDARY_EXEC_XSAVES;
4165 		}
4166 	}
4167 
4168 	if (cpu_has_vmx_rdtscp()) {
4169 		bool rdtscp_enabled = guest_cpuid_has(vcpu, X86_FEATURE_RDTSCP);
4170 		if (!rdtscp_enabled)
4171 			exec_control &= ~SECONDARY_EXEC_RDTSCP;
4172 
4173 		if (nested) {
4174 			if (rdtscp_enabled)
4175 				vmx->nested.msrs.secondary_ctls_high |=
4176 					SECONDARY_EXEC_RDTSCP;
4177 			else
4178 				vmx->nested.msrs.secondary_ctls_high &=
4179 					~SECONDARY_EXEC_RDTSCP;
4180 		}
4181 	}
4182 
4183 	if (cpu_has_vmx_invpcid()) {
4184 		/* Exposing INVPCID only when PCID is exposed */
4185 		bool invpcid_enabled =
4186 			guest_cpuid_has(vcpu, X86_FEATURE_INVPCID) &&
4187 			guest_cpuid_has(vcpu, X86_FEATURE_PCID);
4188 
4189 		if (!invpcid_enabled) {
4190 			exec_control &= ~SECONDARY_EXEC_ENABLE_INVPCID;
4191 			guest_cpuid_clear(vcpu, X86_FEATURE_INVPCID);
4192 		}
4193 
4194 		if (nested) {
4195 			if (invpcid_enabled)
4196 				vmx->nested.msrs.secondary_ctls_high |=
4197 					SECONDARY_EXEC_ENABLE_INVPCID;
4198 			else
4199 				vmx->nested.msrs.secondary_ctls_high &=
4200 					~SECONDARY_EXEC_ENABLE_INVPCID;
4201 		}
4202 	}
4203 
4204 	if (vmx_rdrand_supported()) {
4205 		bool rdrand_enabled = guest_cpuid_has(vcpu, X86_FEATURE_RDRAND);
4206 		if (rdrand_enabled)
4207 			exec_control &= ~SECONDARY_EXEC_RDRAND_EXITING;
4208 
4209 		if (nested) {
4210 			if (rdrand_enabled)
4211 				vmx->nested.msrs.secondary_ctls_high |=
4212 					SECONDARY_EXEC_RDRAND_EXITING;
4213 			else
4214 				vmx->nested.msrs.secondary_ctls_high &=
4215 					~SECONDARY_EXEC_RDRAND_EXITING;
4216 		}
4217 	}
4218 
4219 	if (vmx_rdseed_supported()) {
4220 		bool rdseed_enabled = guest_cpuid_has(vcpu, X86_FEATURE_RDSEED);
4221 		if (rdseed_enabled)
4222 			exec_control &= ~SECONDARY_EXEC_RDSEED_EXITING;
4223 
4224 		if (nested) {
4225 			if (rdseed_enabled)
4226 				vmx->nested.msrs.secondary_ctls_high |=
4227 					SECONDARY_EXEC_RDSEED_EXITING;
4228 			else
4229 				vmx->nested.msrs.secondary_ctls_high &=
4230 					~SECONDARY_EXEC_RDSEED_EXITING;
4231 		}
4232 	}
4233 
4234 	if (vmx_waitpkg_supported()) {
4235 		bool waitpkg_enabled =
4236 			guest_cpuid_has(vcpu, X86_FEATURE_WAITPKG);
4237 
4238 		if (!waitpkg_enabled)
4239 			exec_control &= ~SECONDARY_EXEC_ENABLE_USR_WAIT_PAUSE;
4240 
4241 		if (nested) {
4242 			if (waitpkg_enabled)
4243 				vmx->nested.msrs.secondary_ctls_high |=
4244 					SECONDARY_EXEC_ENABLE_USR_WAIT_PAUSE;
4245 			else
4246 				vmx->nested.msrs.secondary_ctls_high &=
4247 					~SECONDARY_EXEC_ENABLE_USR_WAIT_PAUSE;
4248 		}
4249 	}
4250 
4251 	vmx->secondary_exec_control = exec_control;
4252 }
4253 
4254 static void ept_set_mmio_spte_mask(void)
4255 {
4256 	/*
4257 	 * EPT Misconfigurations can be generated if the value of bits 2:0
4258 	 * of an EPT paging-structure entry is 110b (write/execute).
4259 	 */
4260 	kvm_mmu_set_mmio_spte_mask(VMX_EPT_MISCONFIG_WX_VALUE, 0);
4261 }
4262 
4263 #define VMX_XSS_EXIT_BITMAP 0
4264 
4265 /*
4266  * Noting that the initialization of Guest-state Area of VMCS is in
4267  * vmx_vcpu_reset().
4268  */
4269 static void init_vmcs(struct vcpu_vmx *vmx)
4270 {
4271 	if (nested)
4272 		nested_vmx_set_vmcs_shadowing_bitmap();
4273 
4274 	if (cpu_has_vmx_msr_bitmap())
4275 		vmcs_write64(MSR_BITMAP, __pa(vmx->vmcs01.msr_bitmap));
4276 
4277 	vmcs_write64(VMCS_LINK_POINTER, -1ull); /* 22.3.1.5 */
4278 
4279 	/* Control */
4280 	pin_controls_set(vmx, vmx_pin_based_exec_ctrl(vmx));
4281 
4282 	exec_controls_set(vmx, vmx_exec_control(vmx));
4283 
4284 	if (cpu_has_secondary_exec_ctrls()) {
4285 		vmx_compute_secondary_exec_control(vmx);
4286 		secondary_exec_controls_set(vmx, vmx->secondary_exec_control);
4287 	}
4288 
4289 	if (kvm_vcpu_apicv_active(&vmx->vcpu)) {
4290 		vmcs_write64(EOI_EXIT_BITMAP0, 0);
4291 		vmcs_write64(EOI_EXIT_BITMAP1, 0);
4292 		vmcs_write64(EOI_EXIT_BITMAP2, 0);
4293 		vmcs_write64(EOI_EXIT_BITMAP3, 0);
4294 
4295 		vmcs_write16(GUEST_INTR_STATUS, 0);
4296 
4297 		vmcs_write16(POSTED_INTR_NV, POSTED_INTR_VECTOR);
4298 		vmcs_write64(POSTED_INTR_DESC_ADDR, __pa((&vmx->pi_desc)));
4299 	}
4300 
4301 	if (!kvm_pause_in_guest(vmx->vcpu.kvm)) {
4302 		vmcs_write32(PLE_GAP, ple_gap);
4303 		vmx->ple_window = ple_window;
4304 		vmx->ple_window_dirty = true;
4305 	}
4306 
4307 	vmcs_write32(PAGE_FAULT_ERROR_CODE_MASK, 0);
4308 	vmcs_write32(PAGE_FAULT_ERROR_CODE_MATCH, 0);
4309 	vmcs_write32(CR3_TARGET_COUNT, 0);           /* 22.2.1 */
4310 
4311 	vmcs_write16(HOST_FS_SELECTOR, 0);            /* 22.2.4 */
4312 	vmcs_write16(HOST_GS_SELECTOR, 0);            /* 22.2.4 */
4313 	vmx_set_constant_host_state(vmx);
4314 	vmcs_writel(HOST_FS_BASE, 0); /* 22.2.4 */
4315 	vmcs_writel(HOST_GS_BASE, 0); /* 22.2.4 */
4316 
4317 	if (cpu_has_vmx_vmfunc())
4318 		vmcs_write64(VM_FUNCTION_CONTROL, 0);
4319 
4320 	vmcs_write32(VM_EXIT_MSR_STORE_COUNT, 0);
4321 	vmcs_write32(VM_EXIT_MSR_LOAD_COUNT, 0);
4322 	vmcs_write64(VM_EXIT_MSR_LOAD_ADDR, __pa(vmx->msr_autoload.host.val));
4323 	vmcs_write32(VM_ENTRY_MSR_LOAD_COUNT, 0);
4324 	vmcs_write64(VM_ENTRY_MSR_LOAD_ADDR, __pa(vmx->msr_autoload.guest.val));
4325 
4326 	if (vmcs_config.vmentry_ctrl & VM_ENTRY_LOAD_IA32_PAT)
4327 		vmcs_write64(GUEST_IA32_PAT, vmx->vcpu.arch.pat);
4328 
4329 	vm_exit_controls_set(vmx, vmx_vmexit_ctrl());
4330 
4331 	/* 22.2.1, 20.8.1 */
4332 	vm_entry_controls_set(vmx, vmx_vmentry_ctrl());
4333 
4334 	vmx->vcpu.arch.cr0_guest_owned_bits = KVM_POSSIBLE_CR0_GUEST_BITS;
4335 	vmcs_writel(CR0_GUEST_HOST_MASK, ~vmx->vcpu.arch.cr0_guest_owned_bits);
4336 
4337 	set_cr4_guest_host_mask(vmx);
4338 
4339 	if (vmx->vpid != 0)
4340 		vmcs_write16(VIRTUAL_PROCESSOR_ID, vmx->vpid);
4341 
4342 	if (vmx_xsaves_supported())
4343 		vmcs_write64(XSS_EXIT_BITMAP, VMX_XSS_EXIT_BITMAP);
4344 
4345 	if (enable_pml) {
4346 		vmcs_write64(PML_ADDRESS, page_to_phys(vmx->pml_pg));
4347 		vmcs_write16(GUEST_PML_INDEX, PML_ENTITY_NUM - 1);
4348 	}
4349 
4350 	if (cpu_has_vmx_encls_vmexit())
4351 		vmcs_write64(ENCLS_EXITING_BITMAP, -1ull);
4352 
4353 	if (vmx_pt_mode_is_host_guest()) {
4354 		memset(&vmx->pt_desc, 0, sizeof(vmx->pt_desc));
4355 		/* Bit[6~0] are forced to 1, writes are ignored. */
4356 		vmx->pt_desc.guest.output_mask = 0x7F;
4357 		vmcs_write64(GUEST_IA32_RTIT_CTL, 0);
4358 	}
4359 }
4360 
4361 static void vmx_vcpu_reset(struct kvm_vcpu *vcpu, bool init_event)
4362 {
4363 	struct vcpu_vmx *vmx = to_vmx(vcpu);
4364 	struct msr_data apic_base_msr;
4365 	u64 cr0;
4366 
4367 	vmx->rmode.vm86_active = 0;
4368 	vmx->spec_ctrl = 0;
4369 
4370 	vmx->msr_ia32_umwait_control = 0;
4371 
4372 	vmx->vcpu.arch.regs[VCPU_REGS_RDX] = get_rdx_init_val();
4373 	vmx->hv_deadline_tsc = -1;
4374 	kvm_set_cr8(vcpu, 0);
4375 
4376 	if (!init_event) {
4377 		apic_base_msr.data = APIC_DEFAULT_PHYS_BASE |
4378 				     MSR_IA32_APICBASE_ENABLE;
4379 		if (kvm_vcpu_is_reset_bsp(vcpu))
4380 			apic_base_msr.data |= MSR_IA32_APICBASE_BSP;
4381 		apic_base_msr.host_initiated = true;
4382 		kvm_set_apic_base(vcpu, &apic_base_msr);
4383 	}
4384 
4385 	vmx_segment_cache_clear(vmx);
4386 
4387 	seg_setup(VCPU_SREG_CS);
4388 	vmcs_write16(GUEST_CS_SELECTOR, 0xf000);
4389 	vmcs_writel(GUEST_CS_BASE, 0xffff0000ul);
4390 
4391 	seg_setup(VCPU_SREG_DS);
4392 	seg_setup(VCPU_SREG_ES);
4393 	seg_setup(VCPU_SREG_FS);
4394 	seg_setup(VCPU_SREG_GS);
4395 	seg_setup(VCPU_SREG_SS);
4396 
4397 	vmcs_write16(GUEST_TR_SELECTOR, 0);
4398 	vmcs_writel(GUEST_TR_BASE, 0);
4399 	vmcs_write32(GUEST_TR_LIMIT, 0xffff);
4400 	vmcs_write32(GUEST_TR_AR_BYTES, 0x008b);
4401 
4402 	vmcs_write16(GUEST_LDTR_SELECTOR, 0);
4403 	vmcs_writel(GUEST_LDTR_BASE, 0);
4404 	vmcs_write32(GUEST_LDTR_LIMIT, 0xffff);
4405 	vmcs_write32(GUEST_LDTR_AR_BYTES, 0x00082);
4406 
4407 	if (!init_event) {
4408 		vmcs_write32(GUEST_SYSENTER_CS, 0);
4409 		vmcs_writel(GUEST_SYSENTER_ESP, 0);
4410 		vmcs_writel(GUEST_SYSENTER_EIP, 0);
4411 		vmcs_write64(GUEST_IA32_DEBUGCTL, 0);
4412 	}
4413 
4414 	kvm_set_rflags(vcpu, X86_EFLAGS_FIXED);
4415 	kvm_rip_write(vcpu, 0xfff0);
4416 
4417 	vmcs_writel(GUEST_GDTR_BASE, 0);
4418 	vmcs_write32(GUEST_GDTR_LIMIT, 0xffff);
4419 
4420 	vmcs_writel(GUEST_IDTR_BASE, 0);
4421 	vmcs_write32(GUEST_IDTR_LIMIT, 0xffff);
4422 
4423 	vmcs_write32(GUEST_ACTIVITY_STATE, GUEST_ACTIVITY_ACTIVE);
4424 	vmcs_write32(GUEST_INTERRUPTIBILITY_INFO, 0);
4425 	vmcs_writel(GUEST_PENDING_DBG_EXCEPTIONS, 0);
4426 	if (kvm_mpx_supported())
4427 		vmcs_write64(GUEST_BNDCFGS, 0);
4428 
4429 	setup_msrs(vmx);
4430 
4431 	vmcs_write32(VM_ENTRY_INTR_INFO_FIELD, 0);  /* 22.2.1 */
4432 
4433 	if (cpu_has_vmx_tpr_shadow() && !init_event) {
4434 		vmcs_write64(VIRTUAL_APIC_PAGE_ADDR, 0);
4435 		if (cpu_need_tpr_shadow(vcpu))
4436 			vmcs_write64(VIRTUAL_APIC_PAGE_ADDR,
4437 				     __pa(vcpu->arch.apic->regs));
4438 		vmcs_write32(TPR_THRESHOLD, 0);
4439 	}
4440 
4441 	kvm_make_request(KVM_REQ_APIC_PAGE_RELOAD, vcpu);
4442 
4443 	cr0 = X86_CR0_NW | X86_CR0_CD | X86_CR0_ET;
4444 	vmx->vcpu.arch.cr0 = cr0;
4445 	vmx_set_cr0(vcpu, cr0); /* enter rmode */
4446 	vmx_set_cr4(vcpu, 0);
4447 	vmx_set_efer(vcpu, 0);
4448 
4449 	update_exception_bitmap(vcpu);
4450 
4451 	vpid_sync_context(vmx->vpid);
4452 	if (init_event)
4453 		vmx_clear_hlt(vcpu);
4454 }
4455 
4456 static void enable_irq_window(struct kvm_vcpu *vcpu)
4457 {
4458 	exec_controls_setbit(to_vmx(vcpu), CPU_BASED_INTR_WINDOW_EXITING);
4459 }
4460 
4461 static void enable_nmi_window(struct kvm_vcpu *vcpu)
4462 {
4463 	if (!enable_vnmi ||
4464 	    vmcs_read32(GUEST_INTERRUPTIBILITY_INFO) & GUEST_INTR_STATE_STI) {
4465 		enable_irq_window(vcpu);
4466 		return;
4467 	}
4468 
4469 	exec_controls_setbit(to_vmx(vcpu), CPU_BASED_NMI_WINDOW_EXITING);
4470 }
4471 
4472 static void vmx_inject_irq(struct kvm_vcpu *vcpu)
4473 {
4474 	struct vcpu_vmx *vmx = to_vmx(vcpu);
4475 	uint32_t intr;
4476 	int irq = vcpu->arch.interrupt.nr;
4477 
4478 	trace_kvm_inj_virq(irq);
4479 
4480 	++vcpu->stat.irq_injections;
4481 	if (vmx->rmode.vm86_active) {
4482 		int inc_eip = 0;
4483 		if (vcpu->arch.interrupt.soft)
4484 			inc_eip = vcpu->arch.event_exit_inst_len;
4485 		kvm_inject_realmode_interrupt(vcpu, irq, inc_eip);
4486 		return;
4487 	}
4488 	intr = irq | INTR_INFO_VALID_MASK;
4489 	if (vcpu->arch.interrupt.soft) {
4490 		intr |= INTR_TYPE_SOFT_INTR;
4491 		vmcs_write32(VM_ENTRY_INSTRUCTION_LEN,
4492 			     vmx->vcpu.arch.event_exit_inst_len);
4493 	} else
4494 		intr |= INTR_TYPE_EXT_INTR;
4495 	vmcs_write32(VM_ENTRY_INTR_INFO_FIELD, intr);
4496 
4497 	vmx_clear_hlt(vcpu);
4498 }
4499 
4500 static void vmx_inject_nmi(struct kvm_vcpu *vcpu)
4501 {
4502 	struct vcpu_vmx *vmx = to_vmx(vcpu);
4503 
4504 	if (!enable_vnmi) {
4505 		/*
4506 		 * Tracking the NMI-blocked state in software is built upon
4507 		 * finding the next open IRQ window. This, in turn, depends on
4508 		 * well-behaving guests: They have to keep IRQs disabled at
4509 		 * least as long as the NMI handler runs. Otherwise we may
4510 		 * cause NMI nesting, maybe breaking the guest. But as this is
4511 		 * highly unlikely, we can live with the residual risk.
4512 		 */
4513 		vmx->loaded_vmcs->soft_vnmi_blocked = 1;
4514 		vmx->loaded_vmcs->vnmi_blocked_time = 0;
4515 	}
4516 
4517 	++vcpu->stat.nmi_injections;
4518 	vmx->loaded_vmcs->nmi_known_unmasked = false;
4519 
4520 	if (vmx->rmode.vm86_active) {
4521 		kvm_inject_realmode_interrupt(vcpu, NMI_VECTOR, 0);
4522 		return;
4523 	}
4524 
4525 	vmcs_write32(VM_ENTRY_INTR_INFO_FIELD,
4526 			INTR_TYPE_NMI_INTR | INTR_INFO_VALID_MASK | NMI_VECTOR);
4527 
4528 	vmx_clear_hlt(vcpu);
4529 }
4530 
4531 bool vmx_get_nmi_mask(struct kvm_vcpu *vcpu)
4532 {
4533 	struct vcpu_vmx *vmx = to_vmx(vcpu);
4534 	bool masked;
4535 
4536 	if (!enable_vnmi)
4537 		return vmx->loaded_vmcs->soft_vnmi_blocked;
4538 	if (vmx->loaded_vmcs->nmi_known_unmasked)
4539 		return false;
4540 	masked = vmcs_read32(GUEST_INTERRUPTIBILITY_INFO) & GUEST_INTR_STATE_NMI;
4541 	vmx->loaded_vmcs->nmi_known_unmasked = !masked;
4542 	return masked;
4543 }
4544 
4545 void vmx_set_nmi_mask(struct kvm_vcpu *vcpu, bool masked)
4546 {
4547 	struct vcpu_vmx *vmx = to_vmx(vcpu);
4548 
4549 	if (!enable_vnmi) {
4550 		if (vmx->loaded_vmcs->soft_vnmi_blocked != masked) {
4551 			vmx->loaded_vmcs->soft_vnmi_blocked = masked;
4552 			vmx->loaded_vmcs->vnmi_blocked_time = 0;
4553 		}
4554 	} else {
4555 		vmx->loaded_vmcs->nmi_known_unmasked = !masked;
4556 		if (masked)
4557 			vmcs_set_bits(GUEST_INTERRUPTIBILITY_INFO,
4558 				      GUEST_INTR_STATE_NMI);
4559 		else
4560 			vmcs_clear_bits(GUEST_INTERRUPTIBILITY_INFO,
4561 					GUEST_INTR_STATE_NMI);
4562 	}
4563 }
4564 
4565 bool vmx_nmi_blocked(struct kvm_vcpu *vcpu)
4566 {
4567 	if (is_guest_mode(vcpu) && nested_exit_on_nmi(vcpu))
4568 		return false;
4569 
4570 	if (!enable_vnmi && to_vmx(vcpu)->loaded_vmcs->soft_vnmi_blocked)
4571 		return true;
4572 
4573 	return (vmcs_read32(GUEST_INTERRUPTIBILITY_INFO) &
4574 		(GUEST_INTR_STATE_MOV_SS | GUEST_INTR_STATE_STI |
4575 		 GUEST_INTR_STATE_NMI));
4576 }
4577 
4578 static int vmx_nmi_allowed(struct kvm_vcpu *vcpu, bool for_injection)
4579 {
4580 	if (to_vmx(vcpu)->nested.nested_run_pending)
4581 		return -EBUSY;
4582 
4583 	/* An NMI must not be injected into L2 if it's supposed to VM-Exit.  */
4584 	if (for_injection && is_guest_mode(vcpu) && nested_exit_on_nmi(vcpu))
4585 		return -EBUSY;
4586 
4587 	return !vmx_nmi_blocked(vcpu);
4588 }
4589 
4590 bool vmx_interrupt_blocked(struct kvm_vcpu *vcpu)
4591 {
4592 	if (is_guest_mode(vcpu) && nested_exit_on_intr(vcpu))
4593 		return false;
4594 
4595 	return !(vmx_get_rflags(vcpu) & X86_EFLAGS_IF) ||
4596 	       (vmcs_read32(GUEST_INTERRUPTIBILITY_INFO) &
4597 		(GUEST_INTR_STATE_STI | GUEST_INTR_STATE_MOV_SS));
4598 }
4599 
4600 static int vmx_interrupt_allowed(struct kvm_vcpu *vcpu, bool for_injection)
4601 {
4602 	if (to_vmx(vcpu)->nested.nested_run_pending)
4603 		return -EBUSY;
4604 
4605        /*
4606         * An IRQ must not be injected into L2 if it's supposed to VM-Exit,
4607         * e.g. if the IRQ arrived asynchronously after checking nested events.
4608         */
4609 	if (for_injection && is_guest_mode(vcpu) && nested_exit_on_intr(vcpu))
4610 		return -EBUSY;
4611 
4612 	return !vmx_interrupt_blocked(vcpu);
4613 }
4614 
4615 static int vmx_set_tss_addr(struct kvm *kvm, unsigned int addr)
4616 {
4617 	int ret;
4618 
4619 	if (enable_unrestricted_guest)
4620 		return 0;
4621 
4622 	mutex_lock(&kvm->slots_lock);
4623 	ret = __x86_set_memory_region(kvm, TSS_PRIVATE_MEMSLOT, addr,
4624 				      PAGE_SIZE * 3);
4625 	mutex_unlock(&kvm->slots_lock);
4626 
4627 	if (ret)
4628 		return ret;
4629 	to_kvm_vmx(kvm)->tss_addr = addr;
4630 	return init_rmode_tss(kvm);
4631 }
4632 
4633 static int vmx_set_identity_map_addr(struct kvm *kvm, u64 ident_addr)
4634 {
4635 	to_kvm_vmx(kvm)->ept_identity_map_addr = ident_addr;
4636 	return 0;
4637 }
4638 
4639 static bool rmode_exception(struct kvm_vcpu *vcpu, int vec)
4640 {
4641 	switch (vec) {
4642 	case BP_VECTOR:
4643 		/*
4644 		 * Update instruction length as we may reinject the exception
4645 		 * from user space while in guest debugging mode.
4646 		 */
4647 		to_vmx(vcpu)->vcpu.arch.event_exit_inst_len =
4648 			vmcs_read32(VM_EXIT_INSTRUCTION_LEN);
4649 		if (vcpu->guest_debug & KVM_GUESTDBG_USE_SW_BP)
4650 			return false;
4651 		/* fall through */
4652 	case DB_VECTOR:
4653 		return !(vcpu->guest_debug &
4654 			(KVM_GUESTDBG_SINGLESTEP | KVM_GUESTDBG_USE_HW_BP));
4655 	case DE_VECTOR:
4656 	case OF_VECTOR:
4657 	case BR_VECTOR:
4658 	case UD_VECTOR:
4659 	case DF_VECTOR:
4660 	case SS_VECTOR:
4661 	case GP_VECTOR:
4662 	case MF_VECTOR:
4663 		return true;
4664 	}
4665 	return false;
4666 }
4667 
4668 static int handle_rmode_exception(struct kvm_vcpu *vcpu,
4669 				  int vec, u32 err_code)
4670 {
4671 	/*
4672 	 * Instruction with address size override prefix opcode 0x67
4673 	 * Cause the #SS fault with 0 error code in VM86 mode.
4674 	 */
4675 	if (((vec == GP_VECTOR) || (vec == SS_VECTOR)) && err_code == 0) {
4676 		if (kvm_emulate_instruction(vcpu, 0)) {
4677 			if (vcpu->arch.halt_request) {
4678 				vcpu->arch.halt_request = 0;
4679 				return kvm_vcpu_halt(vcpu);
4680 			}
4681 			return 1;
4682 		}
4683 		return 0;
4684 	}
4685 
4686 	/*
4687 	 * Forward all other exceptions that are valid in real mode.
4688 	 * FIXME: Breaks guest debugging in real mode, needs to be fixed with
4689 	 *        the required debugging infrastructure rework.
4690 	 */
4691 	kvm_queue_exception(vcpu, vec);
4692 	return 1;
4693 }
4694 
4695 /*
4696  * Trigger machine check on the host. We assume all the MSRs are already set up
4697  * by the CPU and that we still run on the same CPU as the MCE occurred on.
4698  * We pass a fake environment to the machine check handler because we want
4699  * the guest to be always treated like user space, no matter what context
4700  * it used internally.
4701  */
4702 static void kvm_machine_check(void)
4703 {
4704 #if defined(CONFIG_X86_MCE)
4705 	struct pt_regs regs = {
4706 		.cs = 3, /* Fake ring 3 no matter what the guest ran on */
4707 		.flags = X86_EFLAGS_IF,
4708 	};
4709 
4710 	do_machine_check(&regs);
4711 #endif
4712 }
4713 
4714 static int handle_machine_check(struct kvm_vcpu *vcpu)
4715 {
4716 	/* handled by vmx_vcpu_run() */
4717 	return 1;
4718 }
4719 
4720 /*
4721  * If the host has split lock detection disabled, then #AC is
4722  * unconditionally injected into the guest, which is the pre split lock
4723  * detection behaviour.
4724  *
4725  * If the host has split lock detection enabled then #AC is
4726  * only injected into the guest when:
4727  *  - Guest CPL == 3 (user mode)
4728  *  - Guest has #AC detection enabled in CR0
4729  *  - Guest EFLAGS has AC bit set
4730  */
4731 static inline bool guest_inject_ac(struct kvm_vcpu *vcpu)
4732 {
4733 	if (!boot_cpu_has(X86_FEATURE_SPLIT_LOCK_DETECT))
4734 		return true;
4735 
4736 	return vmx_get_cpl(vcpu) == 3 && kvm_read_cr0_bits(vcpu, X86_CR0_AM) &&
4737 	       (kvm_get_rflags(vcpu) & X86_EFLAGS_AC);
4738 }
4739 
4740 static int handle_exception_nmi(struct kvm_vcpu *vcpu)
4741 {
4742 	struct vcpu_vmx *vmx = to_vmx(vcpu);
4743 	struct kvm_run *kvm_run = vcpu->run;
4744 	u32 intr_info, ex_no, error_code;
4745 	unsigned long cr2, rip, dr6;
4746 	u32 vect_info;
4747 
4748 	vect_info = vmx->idt_vectoring_info;
4749 	intr_info = vmx_get_intr_info(vcpu);
4750 
4751 	if (is_machine_check(intr_info) || is_nmi(intr_info))
4752 		return 1; /* handled by handle_exception_nmi_irqoff() */
4753 
4754 	if (is_invalid_opcode(intr_info))
4755 		return handle_ud(vcpu);
4756 
4757 	error_code = 0;
4758 	if (intr_info & INTR_INFO_DELIVER_CODE_MASK)
4759 		error_code = vmcs_read32(VM_EXIT_INTR_ERROR_CODE);
4760 
4761 	if (!vmx->rmode.vm86_active && is_gp_fault(intr_info)) {
4762 		WARN_ON_ONCE(!enable_vmware_backdoor);
4763 
4764 		/*
4765 		 * VMware backdoor emulation on #GP interception only handles
4766 		 * IN{S}, OUT{S}, and RDPMC, none of which generate a non-zero
4767 		 * error code on #GP.
4768 		 */
4769 		if (error_code) {
4770 			kvm_queue_exception_e(vcpu, GP_VECTOR, error_code);
4771 			return 1;
4772 		}
4773 		return kvm_emulate_instruction(vcpu, EMULTYPE_VMWARE_GP);
4774 	}
4775 
4776 	/*
4777 	 * The #PF with PFEC.RSVD = 1 indicates the guest is accessing
4778 	 * MMIO, it is better to report an internal error.
4779 	 * See the comments in vmx_handle_exit.
4780 	 */
4781 	if ((vect_info & VECTORING_INFO_VALID_MASK) &&
4782 	    !(is_page_fault(intr_info) && !(error_code & PFERR_RSVD_MASK))) {
4783 		vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
4784 		vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_SIMUL_EX;
4785 		vcpu->run->internal.ndata = 3;
4786 		vcpu->run->internal.data[0] = vect_info;
4787 		vcpu->run->internal.data[1] = intr_info;
4788 		vcpu->run->internal.data[2] = error_code;
4789 		return 0;
4790 	}
4791 
4792 	if (is_page_fault(intr_info)) {
4793 		cr2 = vmx_get_exit_qual(vcpu);
4794 		/* EPT won't cause page fault directly */
4795 		WARN_ON_ONCE(!vcpu->arch.apf.host_apf_flags && enable_ept);
4796 		return kvm_handle_page_fault(vcpu, error_code, cr2, NULL, 0);
4797 	}
4798 
4799 	ex_no = intr_info & INTR_INFO_VECTOR_MASK;
4800 
4801 	if (vmx->rmode.vm86_active && rmode_exception(vcpu, ex_no))
4802 		return handle_rmode_exception(vcpu, ex_no, error_code);
4803 
4804 	switch (ex_no) {
4805 	case DB_VECTOR:
4806 		dr6 = vmx_get_exit_qual(vcpu);
4807 		if (!(vcpu->guest_debug &
4808 		      (KVM_GUESTDBG_SINGLESTEP | KVM_GUESTDBG_USE_HW_BP))) {
4809 			if (is_icebp(intr_info))
4810 				WARN_ON(!skip_emulated_instruction(vcpu));
4811 
4812 			kvm_queue_exception_p(vcpu, DB_VECTOR, dr6);
4813 			return 1;
4814 		}
4815 		kvm_run->debug.arch.dr6 = dr6 | DR6_FIXED_1 | DR6_RTM;
4816 		kvm_run->debug.arch.dr7 = vmcs_readl(GUEST_DR7);
4817 		/* fall through */
4818 	case BP_VECTOR:
4819 		/*
4820 		 * Update instruction length as we may reinject #BP from
4821 		 * user space while in guest debugging mode. Reading it for
4822 		 * #DB as well causes no harm, it is not used in that case.
4823 		 */
4824 		vmx->vcpu.arch.event_exit_inst_len =
4825 			vmcs_read32(VM_EXIT_INSTRUCTION_LEN);
4826 		kvm_run->exit_reason = KVM_EXIT_DEBUG;
4827 		rip = kvm_rip_read(vcpu);
4828 		kvm_run->debug.arch.pc = vmcs_readl(GUEST_CS_BASE) + rip;
4829 		kvm_run->debug.arch.exception = ex_no;
4830 		break;
4831 	case AC_VECTOR:
4832 		if (guest_inject_ac(vcpu)) {
4833 			kvm_queue_exception_e(vcpu, AC_VECTOR, error_code);
4834 			return 1;
4835 		}
4836 
4837 		/*
4838 		 * Handle split lock. Depending on detection mode this will
4839 		 * either warn and disable split lock detection for this
4840 		 * task or force SIGBUS on it.
4841 		 */
4842 		if (handle_guest_split_lock(kvm_rip_read(vcpu)))
4843 			return 1;
4844 		fallthrough;
4845 	default:
4846 		kvm_run->exit_reason = KVM_EXIT_EXCEPTION;
4847 		kvm_run->ex.exception = ex_no;
4848 		kvm_run->ex.error_code = error_code;
4849 		break;
4850 	}
4851 	return 0;
4852 }
4853 
4854 static __always_inline int handle_external_interrupt(struct kvm_vcpu *vcpu)
4855 {
4856 	++vcpu->stat.irq_exits;
4857 	return 1;
4858 }
4859 
4860 static int handle_triple_fault(struct kvm_vcpu *vcpu)
4861 {
4862 	vcpu->run->exit_reason = KVM_EXIT_SHUTDOWN;
4863 	vcpu->mmio_needed = 0;
4864 	return 0;
4865 }
4866 
4867 static int handle_io(struct kvm_vcpu *vcpu)
4868 {
4869 	unsigned long exit_qualification;
4870 	int size, in, string;
4871 	unsigned port;
4872 
4873 	exit_qualification = vmx_get_exit_qual(vcpu);
4874 	string = (exit_qualification & 16) != 0;
4875 
4876 	++vcpu->stat.io_exits;
4877 
4878 	if (string)
4879 		return kvm_emulate_instruction(vcpu, 0);
4880 
4881 	port = exit_qualification >> 16;
4882 	size = (exit_qualification & 7) + 1;
4883 	in = (exit_qualification & 8) != 0;
4884 
4885 	return kvm_fast_pio(vcpu, size, port, in);
4886 }
4887 
4888 static void
4889 vmx_patch_hypercall(struct kvm_vcpu *vcpu, unsigned char *hypercall)
4890 {
4891 	/*
4892 	 * Patch in the VMCALL instruction:
4893 	 */
4894 	hypercall[0] = 0x0f;
4895 	hypercall[1] = 0x01;
4896 	hypercall[2] = 0xc1;
4897 }
4898 
4899 /* called to set cr0 as appropriate for a mov-to-cr0 exit. */
4900 static int handle_set_cr0(struct kvm_vcpu *vcpu, unsigned long val)
4901 {
4902 	if (is_guest_mode(vcpu)) {
4903 		struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
4904 		unsigned long orig_val = val;
4905 
4906 		/*
4907 		 * We get here when L2 changed cr0 in a way that did not change
4908 		 * any of L1's shadowed bits (see nested_vmx_exit_handled_cr),
4909 		 * but did change L0 shadowed bits. So we first calculate the
4910 		 * effective cr0 value that L1 would like to write into the
4911 		 * hardware. It consists of the L2-owned bits from the new
4912 		 * value combined with the L1-owned bits from L1's guest_cr0.
4913 		 */
4914 		val = (val & ~vmcs12->cr0_guest_host_mask) |
4915 			(vmcs12->guest_cr0 & vmcs12->cr0_guest_host_mask);
4916 
4917 		if (!nested_guest_cr0_valid(vcpu, val))
4918 			return 1;
4919 
4920 		if (kvm_set_cr0(vcpu, val))
4921 			return 1;
4922 		vmcs_writel(CR0_READ_SHADOW, orig_val);
4923 		return 0;
4924 	} else {
4925 		if (to_vmx(vcpu)->nested.vmxon &&
4926 		    !nested_host_cr0_valid(vcpu, val))
4927 			return 1;
4928 
4929 		return kvm_set_cr0(vcpu, val);
4930 	}
4931 }
4932 
4933 static int handle_set_cr4(struct kvm_vcpu *vcpu, unsigned long val)
4934 {
4935 	if (is_guest_mode(vcpu)) {
4936 		struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
4937 		unsigned long orig_val = val;
4938 
4939 		/* analogously to handle_set_cr0 */
4940 		val = (val & ~vmcs12->cr4_guest_host_mask) |
4941 			(vmcs12->guest_cr4 & vmcs12->cr4_guest_host_mask);
4942 		if (kvm_set_cr4(vcpu, val))
4943 			return 1;
4944 		vmcs_writel(CR4_READ_SHADOW, orig_val);
4945 		return 0;
4946 	} else
4947 		return kvm_set_cr4(vcpu, val);
4948 }
4949 
4950 static int handle_desc(struct kvm_vcpu *vcpu)
4951 {
4952 	WARN_ON(!(vcpu->arch.cr4 & X86_CR4_UMIP));
4953 	return kvm_emulate_instruction(vcpu, 0);
4954 }
4955 
4956 static int handle_cr(struct kvm_vcpu *vcpu)
4957 {
4958 	unsigned long exit_qualification, val;
4959 	int cr;
4960 	int reg;
4961 	int err;
4962 	int ret;
4963 
4964 	exit_qualification = vmx_get_exit_qual(vcpu);
4965 	cr = exit_qualification & 15;
4966 	reg = (exit_qualification >> 8) & 15;
4967 	switch ((exit_qualification >> 4) & 3) {
4968 	case 0: /* mov to cr */
4969 		val = kvm_register_readl(vcpu, reg);
4970 		trace_kvm_cr_write(cr, val);
4971 		switch (cr) {
4972 		case 0:
4973 			err = handle_set_cr0(vcpu, val);
4974 			return kvm_complete_insn_gp(vcpu, err);
4975 		case 3:
4976 			WARN_ON_ONCE(enable_unrestricted_guest);
4977 			err = kvm_set_cr3(vcpu, val);
4978 			return kvm_complete_insn_gp(vcpu, err);
4979 		case 4:
4980 			err = handle_set_cr4(vcpu, val);
4981 			return kvm_complete_insn_gp(vcpu, err);
4982 		case 8: {
4983 				u8 cr8_prev = kvm_get_cr8(vcpu);
4984 				u8 cr8 = (u8)val;
4985 				err = kvm_set_cr8(vcpu, cr8);
4986 				ret = kvm_complete_insn_gp(vcpu, err);
4987 				if (lapic_in_kernel(vcpu))
4988 					return ret;
4989 				if (cr8_prev <= cr8)
4990 					return ret;
4991 				/*
4992 				 * TODO: we might be squashing a
4993 				 * KVM_GUESTDBG_SINGLESTEP-triggered
4994 				 * KVM_EXIT_DEBUG here.
4995 				 */
4996 				vcpu->run->exit_reason = KVM_EXIT_SET_TPR;
4997 				return 0;
4998 			}
4999 		}
5000 		break;
5001 	case 2: /* clts */
5002 		WARN_ONCE(1, "Guest should always own CR0.TS");
5003 		vmx_set_cr0(vcpu, kvm_read_cr0_bits(vcpu, ~X86_CR0_TS));
5004 		trace_kvm_cr_write(0, kvm_read_cr0(vcpu));
5005 		return kvm_skip_emulated_instruction(vcpu);
5006 	case 1: /*mov from cr*/
5007 		switch (cr) {
5008 		case 3:
5009 			WARN_ON_ONCE(enable_unrestricted_guest);
5010 			val = kvm_read_cr3(vcpu);
5011 			kvm_register_write(vcpu, reg, val);
5012 			trace_kvm_cr_read(cr, val);
5013 			return kvm_skip_emulated_instruction(vcpu);
5014 		case 8:
5015 			val = kvm_get_cr8(vcpu);
5016 			kvm_register_write(vcpu, reg, val);
5017 			trace_kvm_cr_read(cr, val);
5018 			return kvm_skip_emulated_instruction(vcpu);
5019 		}
5020 		break;
5021 	case 3: /* lmsw */
5022 		val = (exit_qualification >> LMSW_SOURCE_DATA_SHIFT) & 0x0f;
5023 		trace_kvm_cr_write(0, (kvm_read_cr0(vcpu) & ~0xful) | val);
5024 		kvm_lmsw(vcpu, val);
5025 
5026 		return kvm_skip_emulated_instruction(vcpu);
5027 	default:
5028 		break;
5029 	}
5030 	vcpu->run->exit_reason = 0;
5031 	vcpu_unimpl(vcpu, "unhandled control register: op %d cr %d\n",
5032 	       (int)(exit_qualification >> 4) & 3, cr);
5033 	return 0;
5034 }
5035 
5036 static int handle_dr(struct kvm_vcpu *vcpu)
5037 {
5038 	unsigned long exit_qualification;
5039 	int dr, dr7, reg;
5040 
5041 	exit_qualification = vmx_get_exit_qual(vcpu);
5042 	dr = exit_qualification & DEBUG_REG_ACCESS_NUM;
5043 
5044 	/* First, if DR does not exist, trigger UD */
5045 	if (!kvm_require_dr(vcpu, dr))
5046 		return 1;
5047 
5048 	/* Do not handle if the CPL > 0, will trigger GP on re-entry */
5049 	if (!kvm_require_cpl(vcpu, 0))
5050 		return 1;
5051 	dr7 = vmcs_readl(GUEST_DR7);
5052 	if (dr7 & DR7_GD) {
5053 		/*
5054 		 * As the vm-exit takes precedence over the debug trap, we
5055 		 * need to emulate the latter, either for the host or the
5056 		 * guest debugging itself.
5057 		 */
5058 		if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP) {
5059 			vcpu->run->debug.arch.dr6 = DR6_BD | DR6_RTM | DR6_FIXED_1;
5060 			vcpu->run->debug.arch.dr7 = dr7;
5061 			vcpu->run->debug.arch.pc = kvm_get_linear_rip(vcpu);
5062 			vcpu->run->debug.arch.exception = DB_VECTOR;
5063 			vcpu->run->exit_reason = KVM_EXIT_DEBUG;
5064 			return 0;
5065 		} else {
5066 			kvm_queue_exception_p(vcpu, DB_VECTOR, DR6_BD);
5067 			return 1;
5068 		}
5069 	}
5070 
5071 	if (vcpu->guest_debug == 0) {
5072 		exec_controls_clearbit(to_vmx(vcpu), CPU_BASED_MOV_DR_EXITING);
5073 
5074 		/*
5075 		 * No more DR vmexits; force a reload of the debug registers
5076 		 * and reenter on this instruction.  The next vmexit will
5077 		 * retrieve the full state of the debug registers.
5078 		 */
5079 		vcpu->arch.switch_db_regs |= KVM_DEBUGREG_WONT_EXIT;
5080 		return 1;
5081 	}
5082 
5083 	reg = DEBUG_REG_ACCESS_REG(exit_qualification);
5084 	if (exit_qualification & TYPE_MOV_FROM_DR) {
5085 		unsigned long val;
5086 
5087 		if (kvm_get_dr(vcpu, dr, &val))
5088 			return 1;
5089 		kvm_register_write(vcpu, reg, val);
5090 	} else
5091 		if (kvm_set_dr(vcpu, dr, kvm_register_readl(vcpu, reg)))
5092 			return 1;
5093 
5094 	return kvm_skip_emulated_instruction(vcpu);
5095 }
5096 
5097 static void vmx_sync_dirty_debug_regs(struct kvm_vcpu *vcpu)
5098 {
5099 	get_debugreg(vcpu->arch.db[0], 0);
5100 	get_debugreg(vcpu->arch.db[1], 1);
5101 	get_debugreg(vcpu->arch.db[2], 2);
5102 	get_debugreg(vcpu->arch.db[3], 3);
5103 	get_debugreg(vcpu->arch.dr6, 6);
5104 	vcpu->arch.dr7 = vmcs_readl(GUEST_DR7);
5105 
5106 	vcpu->arch.switch_db_regs &= ~KVM_DEBUGREG_WONT_EXIT;
5107 	exec_controls_setbit(to_vmx(vcpu), CPU_BASED_MOV_DR_EXITING);
5108 }
5109 
5110 static void vmx_set_dr7(struct kvm_vcpu *vcpu, unsigned long val)
5111 {
5112 	vmcs_writel(GUEST_DR7, val);
5113 }
5114 
5115 static int handle_tpr_below_threshold(struct kvm_vcpu *vcpu)
5116 {
5117 	kvm_apic_update_ppr(vcpu);
5118 	return 1;
5119 }
5120 
5121 static int handle_interrupt_window(struct kvm_vcpu *vcpu)
5122 {
5123 	exec_controls_clearbit(to_vmx(vcpu), CPU_BASED_INTR_WINDOW_EXITING);
5124 
5125 	kvm_make_request(KVM_REQ_EVENT, vcpu);
5126 
5127 	++vcpu->stat.irq_window_exits;
5128 	return 1;
5129 }
5130 
5131 static int handle_vmcall(struct kvm_vcpu *vcpu)
5132 {
5133 	return kvm_emulate_hypercall(vcpu);
5134 }
5135 
5136 static int handle_invd(struct kvm_vcpu *vcpu)
5137 {
5138 	return kvm_emulate_instruction(vcpu, 0);
5139 }
5140 
5141 static int handle_invlpg(struct kvm_vcpu *vcpu)
5142 {
5143 	unsigned long exit_qualification = vmx_get_exit_qual(vcpu);
5144 
5145 	kvm_mmu_invlpg(vcpu, exit_qualification);
5146 	return kvm_skip_emulated_instruction(vcpu);
5147 }
5148 
5149 static int handle_rdpmc(struct kvm_vcpu *vcpu)
5150 {
5151 	int err;
5152 
5153 	err = kvm_rdpmc(vcpu);
5154 	return kvm_complete_insn_gp(vcpu, err);
5155 }
5156 
5157 static int handle_wbinvd(struct kvm_vcpu *vcpu)
5158 {
5159 	return kvm_emulate_wbinvd(vcpu);
5160 }
5161 
5162 static int handle_xsetbv(struct kvm_vcpu *vcpu)
5163 {
5164 	u64 new_bv = kvm_read_edx_eax(vcpu);
5165 	u32 index = kvm_rcx_read(vcpu);
5166 
5167 	if (kvm_set_xcr(vcpu, index, new_bv) == 0)
5168 		return kvm_skip_emulated_instruction(vcpu);
5169 	return 1;
5170 }
5171 
5172 static int handle_apic_access(struct kvm_vcpu *vcpu)
5173 {
5174 	if (likely(fasteoi)) {
5175 		unsigned long exit_qualification = vmx_get_exit_qual(vcpu);
5176 		int access_type, offset;
5177 
5178 		access_type = exit_qualification & APIC_ACCESS_TYPE;
5179 		offset = exit_qualification & APIC_ACCESS_OFFSET;
5180 		/*
5181 		 * Sane guest uses MOV to write EOI, with written value
5182 		 * not cared. So make a short-circuit here by avoiding
5183 		 * heavy instruction emulation.
5184 		 */
5185 		if ((access_type == TYPE_LINEAR_APIC_INST_WRITE) &&
5186 		    (offset == APIC_EOI)) {
5187 			kvm_lapic_set_eoi(vcpu);
5188 			return kvm_skip_emulated_instruction(vcpu);
5189 		}
5190 	}
5191 	return kvm_emulate_instruction(vcpu, 0);
5192 }
5193 
5194 static int handle_apic_eoi_induced(struct kvm_vcpu *vcpu)
5195 {
5196 	unsigned long exit_qualification = vmx_get_exit_qual(vcpu);
5197 	int vector = exit_qualification & 0xff;
5198 
5199 	/* EOI-induced VM exit is trap-like and thus no need to adjust IP */
5200 	kvm_apic_set_eoi_accelerated(vcpu, vector);
5201 	return 1;
5202 }
5203 
5204 static int handle_apic_write(struct kvm_vcpu *vcpu)
5205 {
5206 	unsigned long exit_qualification = vmx_get_exit_qual(vcpu);
5207 	u32 offset = exit_qualification & 0xfff;
5208 
5209 	/* APIC-write VM exit is trap-like and thus no need to adjust IP */
5210 	kvm_apic_write_nodecode(vcpu, offset);
5211 	return 1;
5212 }
5213 
5214 static int handle_task_switch(struct kvm_vcpu *vcpu)
5215 {
5216 	struct vcpu_vmx *vmx = to_vmx(vcpu);
5217 	unsigned long exit_qualification;
5218 	bool has_error_code = false;
5219 	u32 error_code = 0;
5220 	u16 tss_selector;
5221 	int reason, type, idt_v, idt_index;
5222 
5223 	idt_v = (vmx->idt_vectoring_info & VECTORING_INFO_VALID_MASK);
5224 	idt_index = (vmx->idt_vectoring_info & VECTORING_INFO_VECTOR_MASK);
5225 	type = (vmx->idt_vectoring_info & VECTORING_INFO_TYPE_MASK);
5226 
5227 	exit_qualification = vmx_get_exit_qual(vcpu);
5228 
5229 	reason = (u32)exit_qualification >> 30;
5230 	if (reason == TASK_SWITCH_GATE && idt_v) {
5231 		switch (type) {
5232 		case INTR_TYPE_NMI_INTR:
5233 			vcpu->arch.nmi_injected = false;
5234 			vmx_set_nmi_mask(vcpu, true);
5235 			break;
5236 		case INTR_TYPE_EXT_INTR:
5237 		case INTR_TYPE_SOFT_INTR:
5238 			kvm_clear_interrupt_queue(vcpu);
5239 			break;
5240 		case INTR_TYPE_HARD_EXCEPTION:
5241 			if (vmx->idt_vectoring_info &
5242 			    VECTORING_INFO_DELIVER_CODE_MASK) {
5243 				has_error_code = true;
5244 				error_code =
5245 					vmcs_read32(IDT_VECTORING_ERROR_CODE);
5246 			}
5247 			/* fall through */
5248 		case INTR_TYPE_SOFT_EXCEPTION:
5249 			kvm_clear_exception_queue(vcpu);
5250 			break;
5251 		default:
5252 			break;
5253 		}
5254 	}
5255 	tss_selector = exit_qualification;
5256 
5257 	if (!idt_v || (type != INTR_TYPE_HARD_EXCEPTION &&
5258 		       type != INTR_TYPE_EXT_INTR &&
5259 		       type != INTR_TYPE_NMI_INTR))
5260 		WARN_ON(!skip_emulated_instruction(vcpu));
5261 
5262 	/*
5263 	 * TODO: What about debug traps on tss switch?
5264 	 *       Are we supposed to inject them and update dr6?
5265 	 */
5266 	return kvm_task_switch(vcpu, tss_selector,
5267 			       type == INTR_TYPE_SOFT_INTR ? idt_index : -1,
5268 			       reason, has_error_code, error_code);
5269 }
5270 
5271 static int handle_ept_violation(struct kvm_vcpu *vcpu)
5272 {
5273 	unsigned long exit_qualification;
5274 	gpa_t gpa;
5275 	u64 error_code;
5276 
5277 	exit_qualification = vmx_get_exit_qual(vcpu);
5278 
5279 	/*
5280 	 * EPT violation happened while executing iret from NMI,
5281 	 * "blocked by NMI" bit has to be set before next VM entry.
5282 	 * There are errata that may cause this bit to not be set:
5283 	 * AAK134, BY25.
5284 	 */
5285 	if (!(to_vmx(vcpu)->idt_vectoring_info & VECTORING_INFO_VALID_MASK) &&
5286 			enable_vnmi &&
5287 			(exit_qualification & INTR_INFO_UNBLOCK_NMI))
5288 		vmcs_set_bits(GUEST_INTERRUPTIBILITY_INFO, GUEST_INTR_STATE_NMI);
5289 
5290 	gpa = vmcs_read64(GUEST_PHYSICAL_ADDRESS);
5291 	trace_kvm_page_fault(gpa, exit_qualification);
5292 
5293 	/* Is it a read fault? */
5294 	error_code = (exit_qualification & EPT_VIOLATION_ACC_READ)
5295 		     ? PFERR_USER_MASK : 0;
5296 	/* Is it a write fault? */
5297 	error_code |= (exit_qualification & EPT_VIOLATION_ACC_WRITE)
5298 		      ? PFERR_WRITE_MASK : 0;
5299 	/* Is it a fetch fault? */
5300 	error_code |= (exit_qualification & EPT_VIOLATION_ACC_INSTR)
5301 		      ? PFERR_FETCH_MASK : 0;
5302 	/* ept page table entry is present? */
5303 	error_code |= (exit_qualification &
5304 		       (EPT_VIOLATION_READABLE | EPT_VIOLATION_WRITABLE |
5305 			EPT_VIOLATION_EXECUTABLE))
5306 		      ? PFERR_PRESENT_MASK : 0;
5307 
5308 	error_code |= (exit_qualification & 0x100) != 0 ?
5309 	       PFERR_GUEST_FINAL_MASK : PFERR_GUEST_PAGE_MASK;
5310 
5311 	vcpu->arch.exit_qualification = exit_qualification;
5312 	return kvm_mmu_page_fault(vcpu, gpa, error_code, NULL, 0);
5313 }
5314 
5315 static int handle_ept_misconfig(struct kvm_vcpu *vcpu)
5316 {
5317 	gpa_t gpa;
5318 
5319 	/*
5320 	 * A nested guest cannot optimize MMIO vmexits, because we have an
5321 	 * nGPA here instead of the required GPA.
5322 	 */
5323 	gpa = vmcs_read64(GUEST_PHYSICAL_ADDRESS);
5324 	if (!is_guest_mode(vcpu) &&
5325 	    !kvm_io_bus_write(vcpu, KVM_FAST_MMIO_BUS, gpa, 0, NULL)) {
5326 		trace_kvm_fast_mmio(gpa);
5327 		return kvm_skip_emulated_instruction(vcpu);
5328 	}
5329 
5330 	return kvm_mmu_page_fault(vcpu, gpa, PFERR_RSVD_MASK, NULL, 0);
5331 }
5332 
5333 static int handle_nmi_window(struct kvm_vcpu *vcpu)
5334 {
5335 	WARN_ON_ONCE(!enable_vnmi);
5336 	exec_controls_clearbit(to_vmx(vcpu), CPU_BASED_NMI_WINDOW_EXITING);
5337 	++vcpu->stat.nmi_window_exits;
5338 	kvm_make_request(KVM_REQ_EVENT, vcpu);
5339 
5340 	return 1;
5341 }
5342 
5343 static int handle_invalid_guest_state(struct kvm_vcpu *vcpu)
5344 {
5345 	struct vcpu_vmx *vmx = to_vmx(vcpu);
5346 	bool intr_window_requested;
5347 	unsigned count = 130;
5348 
5349 	intr_window_requested = exec_controls_get(vmx) &
5350 				CPU_BASED_INTR_WINDOW_EXITING;
5351 
5352 	while (vmx->emulation_required && count-- != 0) {
5353 		if (intr_window_requested && !vmx_interrupt_blocked(vcpu))
5354 			return handle_interrupt_window(&vmx->vcpu);
5355 
5356 		if (kvm_test_request(KVM_REQ_EVENT, vcpu))
5357 			return 1;
5358 
5359 		if (!kvm_emulate_instruction(vcpu, 0))
5360 			return 0;
5361 
5362 		if (vmx->emulation_required && !vmx->rmode.vm86_active &&
5363 		    vcpu->arch.exception.pending) {
5364 			vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
5365 			vcpu->run->internal.suberror =
5366 						KVM_INTERNAL_ERROR_EMULATION;
5367 			vcpu->run->internal.ndata = 0;
5368 			return 0;
5369 		}
5370 
5371 		if (vcpu->arch.halt_request) {
5372 			vcpu->arch.halt_request = 0;
5373 			return kvm_vcpu_halt(vcpu);
5374 		}
5375 
5376 		/*
5377 		 * Note, return 1 and not 0, vcpu_run() will invoke
5378 		 * xfer_to_guest_mode() which will create a proper return
5379 		 * code.
5380 		 */
5381 		if (__xfer_to_guest_mode_work_pending())
5382 			return 1;
5383 	}
5384 
5385 	return 1;
5386 }
5387 
5388 static void grow_ple_window(struct kvm_vcpu *vcpu)
5389 {
5390 	struct vcpu_vmx *vmx = to_vmx(vcpu);
5391 	unsigned int old = vmx->ple_window;
5392 
5393 	vmx->ple_window = __grow_ple_window(old, ple_window,
5394 					    ple_window_grow,
5395 					    ple_window_max);
5396 
5397 	if (vmx->ple_window != old) {
5398 		vmx->ple_window_dirty = true;
5399 		trace_kvm_ple_window_update(vcpu->vcpu_id,
5400 					    vmx->ple_window, old);
5401 	}
5402 }
5403 
5404 static void shrink_ple_window(struct kvm_vcpu *vcpu)
5405 {
5406 	struct vcpu_vmx *vmx = to_vmx(vcpu);
5407 	unsigned int old = vmx->ple_window;
5408 
5409 	vmx->ple_window = __shrink_ple_window(old, ple_window,
5410 					      ple_window_shrink,
5411 					      ple_window);
5412 
5413 	if (vmx->ple_window != old) {
5414 		vmx->ple_window_dirty = true;
5415 		trace_kvm_ple_window_update(vcpu->vcpu_id,
5416 					    vmx->ple_window, old);
5417 	}
5418 }
5419 
5420 /*
5421  * Handler for POSTED_INTERRUPT_WAKEUP_VECTOR.
5422  */
5423 static void wakeup_handler(void)
5424 {
5425 	struct kvm_vcpu *vcpu;
5426 	int cpu = smp_processor_id();
5427 
5428 	spin_lock(&per_cpu(blocked_vcpu_on_cpu_lock, cpu));
5429 	list_for_each_entry(vcpu, &per_cpu(blocked_vcpu_on_cpu, cpu),
5430 			blocked_vcpu_list) {
5431 		struct pi_desc *pi_desc = vcpu_to_pi_desc(vcpu);
5432 
5433 		if (pi_test_on(pi_desc) == 1)
5434 			kvm_vcpu_kick(vcpu);
5435 	}
5436 	spin_unlock(&per_cpu(blocked_vcpu_on_cpu_lock, cpu));
5437 }
5438 
5439 static void vmx_enable_tdp(void)
5440 {
5441 	kvm_mmu_set_mask_ptes(VMX_EPT_READABLE_MASK,
5442 		enable_ept_ad_bits ? VMX_EPT_ACCESS_BIT : 0ull,
5443 		enable_ept_ad_bits ? VMX_EPT_DIRTY_BIT : 0ull,
5444 		0ull, VMX_EPT_EXECUTABLE_MASK,
5445 		cpu_has_vmx_ept_execute_only() ? 0ull : VMX_EPT_READABLE_MASK,
5446 		VMX_EPT_RWX_MASK, 0ull);
5447 
5448 	ept_set_mmio_spte_mask();
5449 }
5450 
5451 /*
5452  * Indicate a busy-waiting vcpu in spinlock. We do not enable the PAUSE
5453  * exiting, so only get here on cpu with PAUSE-Loop-Exiting.
5454  */
5455 static int handle_pause(struct kvm_vcpu *vcpu)
5456 {
5457 	if (!kvm_pause_in_guest(vcpu->kvm))
5458 		grow_ple_window(vcpu);
5459 
5460 	/*
5461 	 * Intel sdm vol3 ch-25.1.3 says: The "PAUSE-loop exiting"
5462 	 * VM-execution control is ignored if CPL > 0. OTOH, KVM
5463 	 * never set PAUSE_EXITING and just set PLE if supported,
5464 	 * so the vcpu must be CPL=0 if it gets a PAUSE exit.
5465 	 */
5466 	kvm_vcpu_on_spin(vcpu, true);
5467 	return kvm_skip_emulated_instruction(vcpu);
5468 }
5469 
5470 static int handle_nop(struct kvm_vcpu *vcpu)
5471 {
5472 	return kvm_skip_emulated_instruction(vcpu);
5473 }
5474 
5475 static int handle_mwait(struct kvm_vcpu *vcpu)
5476 {
5477 	printk_once(KERN_WARNING "kvm: MWAIT instruction emulated as NOP!\n");
5478 	return handle_nop(vcpu);
5479 }
5480 
5481 static int handle_invalid_op(struct kvm_vcpu *vcpu)
5482 {
5483 	kvm_queue_exception(vcpu, UD_VECTOR);
5484 	return 1;
5485 }
5486 
5487 static int handle_monitor_trap(struct kvm_vcpu *vcpu)
5488 {
5489 	return 1;
5490 }
5491 
5492 static int handle_monitor(struct kvm_vcpu *vcpu)
5493 {
5494 	printk_once(KERN_WARNING "kvm: MONITOR instruction emulated as NOP!\n");
5495 	return handle_nop(vcpu);
5496 }
5497 
5498 static int handle_invpcid(struct kvm_vcpu *vcpu)
5499 {
5500 	u32 vmx_instruction_info;
5501 	unsigned long type;
5502 	bool pcid_enabled;
5503 	gva_t gva;
5504 	struct x86_exception e;
5505 	unsigned i;
5506 	unsigned long roots_to_free = 0;
5507 	struct {
5508 		u64 pcid;
5509 		u64 gla;
5510 	} operand;
5511 	int r;
5512 
5513 	if (!guest_cpuid_has(vcpu, X86_FEATURE_INVPCID)) {
5514 		kvm_queue_exception(vcpu, UD_VECTOR);
5515 		return 1;
5516 	}
5517 
5518 	vmx_instruction_info = vmcs_read32(VMX_INSTRUCTION_INFO);
5519 	type = kvm_register_readl(vcpu, (vmx_instruction_info >> 28) & 0xf);
5520 
5521 	if (type > 3) {
5522 		kvm_inject_gp(vcpu, 0);
5523 		return 1;
5524 	}
5525 
5526 	/* According to the Intel instruction reference, the memory operand
5527 	 * is read even if it isn't needed (e.g., for type==all)
5528 	 */
5529 	if (get_vmx_mem_address(vcpu, vmx_get_exit_qual(vcpu),
5530 				vmx_instruction_info, false,
5531 				sizeof(operand), &gva))
5532 		return 1;
5533 
5534 	r = kvm_read_guest_virt(vcpu, gva, &operand, sizeof(operand), &e);
5535 	if (r != X86EMUL_CONTINUE)
5536 		return vmx_handle_memory_failure(vcpu, r, &e);
5537 
5538 	if (operand.pcid >> 12 != 0) {
5539 		kvm_inject_gp(vcpu, 0);
5540 		return 1;
5541 	}
5542 
5543 	pcid_enabled = kvm_read_cr4_bits(vcpu, X86_CR4_PCIDE);
5544 
5545 	switch (type) {
5546 	case INVPCID_TYPE_INDIV_ADDR:
5547 		if ((!pcid_enabled && (operand.pcid != 0)) ||
5548 		    is_noncanonical_address(operand.gla, vcpu)) {
5549 			kvm_inject_gp(vcpu, 0);
5550 			return 1;
5551 		}
5552 		kvm_mmu_invpcid_gva(vcpu, operand.gla, operand.pcid);
5553 		return kvm_skip_emulated_instruction(vcpu);
5554 
5555 	case INVPCID_TYPE_SINGLE_CTXT:
5556 		if (!pcid_enabled && (operand.pcid != 0)) {
5557 			kvm_inject_gp(vcpu, 0);
5558 			return 1;
5559 		}
5560 
5561 		if (kvm_get_active_pcid(vcpu) == operand.pcid) {
5562 			kvm_mmu_sync_roots(vcpu);
5563 			kvm_make_request(KVM_REQ_TLB_FLUSH_CURRENT, vcpu);
5564 		}
5565 
5566 		for (i = 0; i < KVM_MMU_NUM_PREV_ROOTS; i++)
5567 			if (kvm_get_pcid(vcpu, vcpu->arch.mmu->prev_roots[i].pgd)
5568 			    == operand.pcid)
5569 				roots_to_free |= KVM_MMU_ROOT_PREVIOUS(i);
5570 
5571 		kvm_mmu_free_roots(vcpu, vcpu->arch.mmu, roots_to_free);
5572 		/*
5573 		 * If neither the current cr3 nor any of the prev_roots use the
5574 		 * given PCID, then nothing needs to be done here because a
5575 		 * resync will happen anyway before switching to any other CR3.
5576 		 */
5577 
5578 		return kvm_skip_emulated_instruction(vcpu);
5579 
5580 	case INVPCID_TYPE_ALL_NON_GLOBAL:
5581 		/*
5582 		 * Currently, KVM doesn't mark global entries in the shadow
5583 		 * page tables, so a non-global flush just degenerates to a
5584 		 * global flush. If needed, we could optimize this later by
5585 		 * keeping track of global entries in shadow page tables.
5586 		 */
5587 
5588 		/* fall-through */
5589 	case INVPCID_TYPE_ALL_INCL_GLOBAL:
5590 		kvm_mmu_unload(vcpu);
5591 		return kvm_skip_emulated_instruction(vcpu);
5592 
5593 	default:
5594 		BUG(); /* We have already checked above that type <= 3 */
5595 	}
5596 }
5597 
5598 static int handle_pml_full(struct kvm_vcpu *vcpu)
5599 {
5600 	unsigned long exit_qualification;
5601 
5602 	trace_kvm_pml_full(vcpu->vcpu_id);
5603 
5604 	exit_qualification = vmx_get_exit_qual(vcpu);
5605 
5606 	/*
5607 	 * PML buffer FULL happened while executing iret from NMI,
5608 	 * "blocked by NMI" bit has to be set before next VM entry.
5609 	 */
5610 	if (!(to_vmx(vcpu)->idt_vectoring_info & VECTORING_INFO_VALID_MASK) &&
5611 			enable_vnmi &&
5612 			(exit_qualification & INTR_INFO_UNBLOCK_NMI))
5613 		vmcs_set_bits(GUEST_INTERRUPTIBILITY_INFO,
5614 				GUEST_INTR_STATE_NMI);
5615 
5616 	/*
5617 	 * PML buffer already flushed at beginning of VMEXIT. Nothing to do
5618 	 * here.., and there's no userspace involvement needed for PML.
5619 	 */
5620 	return 1;
5621 }
5622 
5623 static fastpath_t handle_fastpath_preemption_timer(struct kvm_vcpu *vcpu)
5624 {
5625 	struct vcpu_vmx *vmx = to_vmx(vcpu);
5626 
5627 	if (!vmx->req_immediate_exit &&
5628 	    !unlikely(vmx->loaded_vmcs->hv_timer_soft_disabled)) {
5629 		kvm_lapic_expired_hv_timer(vcpu);
5630 		return EXIT_FASTPATH_REENTER_GUEST;
5631 	}
5632 
5633 	return EXIT_FASTPATH_NONE;
5634 }
5635 
5636 static int handle_preemption_timer(struct kvm_vcpu *vcpu)
5637 {
5638 	handle_fastpath_preemption_timer(vcpu);
5639 	return 1;
5640 }
5641 
5642 /*
5643  * When nested=0, all VMX instruction VM Exits filter here.  The handlers
5644  * are overwritten by nested_vmx_setup() when nested=1.
5645  */
5646 static int handle_vmx_instruction(struct kvm_vcpu *vcpu)
5647 {
5648 	kvm_queue_exception(vcpu, UD_VECTOR);
5649 	return 1;
5650 }
5651 
5652 static int handle_encls(struct kvm_vcpu *vcpu)
5653 {
5654 	/*
5655 	 * SGX virtualization is not yet supported.  There is no software
5656 	 * enable bit for SGX, so we have to trap ENCLS and inject a #UD
5657 	 * to prevent the guest from executing ENCLS.
5658 	 */
5659 	kvm_queue_exception(vcpu, UD_VECTOR);
5660 	return 1;
5661 }
5662 
5663 /*
5664  * The exit handlers return 1 if the exit was handled fully and guest execution
5665  * may resume.  Otherwise they set the kvm_run parameter to indicate what needs
5666  * to be done to userspace and return 0.
5667  */
5668 static int (*kvm_vmx_exit_handlers[])(struct kvm_vcpu *vcpu) = {
5669 	[EXIT_REASON_EXCEPTION_NMI]           = handle_exception_nmi,
5670 	[EXIT_REASON_EXTERNAL_INTERRUPT]      = handle_external_interrupt,
5671 	[EXIT_REASON_TRIPLE_FAULT]            = handle_triple_fault,
5672 	[EXIT_REASON_NMI_WINDOW]	      = handle_nmi_window,
5673 	[EXIT_REASON_IO_INSTRUCTION]          = handle_io,
5674 	[EXIT_REASON_CR_ACCESS]               = handle_cr,
5675 	[EXIT_REASON_DR_ACCESS]               = handle_dr,
5676 	[EXIT_REASON_CPUID]                   = kvm_emulate_cpuid,
5677 	[EXIT_REASON_MSR_READ]                = kvm_emulate_rdmsr,
5678 	[EXIT_REASON_MSR_WRITE]               = kvm_emulate_wrmsr,
5679 	[EXIT_REASON_INTERRUPT_WINDOW]        = handle_interrupt_window,
5680 	[EXIT_REASON_HLT]                     = kvm_emulate_halt,
5681 	[EXIT_REASON_INVD]		      = handle_invd,
5682 	[EXIT_REASON_INVLPG]		      = handle_invlpg,
5683 	[EXIT_REASON_RDPMC]                   = handle_rdpmc,
5684 	[EXIT_REASON_VMCALL]                  = handle_vmcall,
5685 	[EXIT_REASON_VMCLEAR]		      = handle_vmx_instruction,
5686 	[EXIT_REASON_VMLAUNCH]		      = handle_vmx_instruction,
5687 	[EXIT_REASON_VMPTRLD]		      = handle_vmx_instruction,
5688 	[EXIT_REASON_VMPTRST]		      = handle_vmx_instruction,
5689 	[EXIT_REASON_VMREAD]		      = handle_vmx_instruction,
5690 	[EXIT_REASON_VMRESUME]		      = handle_vmx_instruction,
5691 	[EXIT_REASON_VMWRITE]		      = handle_vmx_instruction,
5692 	[EXIT_REASON_VMOFF]		      = handle_vmx_instruction,
5693 	[EXIT_REASON_VMON]		      = handle_vmx_instruction,
5694 	[EXIT_REASON_TPR_BELOW_THRESHOLD]     = handle_tpr_below_threshold,
5695 	[EXIT_REASON_APIC_ACCESS]             = handle_apic_access,
5696 	[EXIT_REASON_APIC_WRITE]              = handle_apic_write,
5697 	[EXIT_REASON_EOI_INDUCED]             = handle_apic_eoi_induced,
5698 	[EXIT_REASON_WBINVD]                  = handle_wbinvd,
5699 	[EXIT_REASON_XSETBV]                  = handle_xsetbv,
5700 	[EXIT_REASON_TASK_SWITCH]             = handle_task_switch,
5701 	[EXIT_REASON_MCE_DURING_VMENTRY]      = handle_machine_check,
5702 	[EXIT_REASON_GDTR_IDTR]		      = handle_desc,
5703 	[EXIT_REASON_LDTR_TR]		      = handle_desc,
5704 	[EXIT_REASON_EPT_VIOLATION]	      = handle_ept_violation,
5705 	[EXIT_REASON_EPT_MISCONFIG]           = handle_ept_misconfig,
5706 	[EXIT_REASON_PAUSE_INSTRUCTION]       = handle_pause,
5707 	[EXIT_REASON_MWAIT_INSTRUCTION]	      = handle_mwait,
5708 	[EXIT_REASON_MONITOR_TRAP_FLAG]       = handle_monitor_trap,
5709 	[EXIT_REASON_MONITOR_INSTRUCTION]     = handle_monitor,
5710 	[EXIT_REASON_INVEPT]                  = handle_vmx_instruction,
5711 	[EXIT_REASON_INVVPID]                 = handle_vmx_instruction,
5712 	[EXIT_REASON_RDRAND]                  = handle_invalid_op,
5713 	[EXIT_REASON_RDSEED]                  = handle_invalid_op,
5714 	[EXIT_REASON_PML_FULL]		      = handle_pml_full,
5715 	[EXIT_REASON_INVPCID]                 = handle_invpcid,
5716 	[EXIT_REASON_VMFUNC]		      = handle_vmx_instruction,
5717 	[EXIT_REASON_PREEMPTION_TIMER]	      = handle_preemption_timer,
5718 	[EXIT_REASON_ENCLS]		      = handle_encls,
5719 };
5720 
5721 static const int kvm_vmx_max_exit_handlers =
5722 	ARRAY_SIZE(kvm_vmx_exit_handlers);
5723 
5724 static void vmx_get_exit_info(struct kvm_vcpu *vcpu, u64 *info1, u64 *info2)
5725 {
5726 	*info1 = vmx_get_exit_qual(vcpu);
5727 	*info2 = vmx_get_intr_info(vcpu);
5728 }
5729 
5730 static void vmx_destroy_pml_buffer(struct vcpu_vmx *vmx)
5731 {
5732 	if (vmx->pml_pg) {
5733 		__free_page(vmx->pml_pg);
5734 		vmx->pml_pg = NULL;
5735 	}
5736 }
5737 
5738 static void vmx_flush_pml_buffer(struct kvm_vcpu *vcpu)
5739 {
5740 	struct vcpu_vmx *vmx = to_vmx(vcpu);
5741 	u64 *pml_buf;
5742 	u16 pml_idx;
5743 
5744 	pml_idx = vmcs_read16(GUEST_PML_INDEX);
5745 
5746 	/* Do nothing if PML buffer is empty */
5747 	if (pml_idx == (PML_ENTITY_NUM - 1))
5748 		return;
5749 
5750 	/* PML index always points to next available PML buffer entity */
5751 	if (pml_idx >= PML_ENTITY_NUM)
5752 		pml_idx = 0;
5753 	else
5754 		pml_idx++;
5755 
5756 	pml_buf = page_address(vmx->pml_pg);
5757 	for (; pml_idx < PML_ENTITY_NUM; pml_idx++) {
5758 		u64 gpa;
5759 
5760 		gpa = pml_buf[pml_idx];
5761 		WARN_ON(gpa & (PAGE_SIZE - 1));
5762 		kvm_vcpu_mark_page_dirty(vcpu, gpa >> PAGE_SHIFT);
5763 	}
5764 
5765 	/* reset PML index */
5766 	vmcs_write16(GUEST_PML_INDEX, PML_ENTITY_NUM - 1);
5767 }
5768 
5769 /*
5770  * Flush all vcpus' PML buffer and update logged GPAs to dirty_bitmap.
5771  * Called before reporting dirty_bitmap to userspace.
5772  */
5773 static void kvm_flush_pml_buffers(struct kvm *kvm)
5774 {
5775 	int i;
5776 	struct kvm_vcpu *vcpu;
5777 	/*
5778 	 * We only need to kick vcpu out of guest mode here, as PML buffer
5779 	 * is flushed at beginning of all VMEXITs, and it's obvious that only
5780 	 * vcpus running in guest are possible to have unflushed GPAs in PML
5781 	 * buffer.
5782 	 */
5783 	kvm_for_each_vcpu(i, vcpu, kvm)
5784 		kvm_vcpu_kick(vcpu);
5785 }
5786 
5787 static void vmx_dump_sel(char *name, uint32_t sel)
5788 {
5789 	pr_err("%s sel=0x%04x, attr=0x%05x, limit=0x%08x, base=0x%016lx\n",
5790 	       name, vmcs_read16(sel),
5791 	       vmcs_read32(sel + GUEST_ES_AR_BYTES - GUEST_ES_SELECTOR),
5792 	       vmcs_read32(sel + GUEST_ES_LIMIT - GUEST_ES_SELECTOR),
5793 	       vmcs_readl(sel + GUEST_ES_BASE - GUEST_ES_SELECTOR));
5794 }
5795 
5796 static void vmx_dump_dtsel(char *name, uint32_t limit)
5797 {
5798 	pr_err("%s                           limit=0x%08x, base=0x%016lx\n",
5799 	       name, vmcs_read32(limit),
5800 	       vmcs_readl(limit + GUEST_GDTR_BASE - GUEST_GDTR_LIMIT));
5801 }
5802 
5803 void dump_vmcs(void)
5804 {
5805 	u32 vmentry_ctl, vmexit_ctl;
5806 	u32 cpu_based_exec_ctrl, pin_based_exec_ctrl, secondary_exec_control;
5807 	unsigned long cr4;
5808 	u64 efer;
5809 
5810 	if (!dump_invalid_vmcs) {
5811 		pr_warn_ratelimited("set kvm_intel.dump_invalid_vmcs=1 to dump internal KVM state.\n");
5812 		return;
5813 	}
5814 
5815 	vmentry_ctl = vmcs_read32(VM_ENTRY_CONTROLS);
5816 	vmexit_ctl = vmcs_read32(VM_EXIT_CONTROLS);
5817 	cpu_based_exec_ctrl = vmcs_read32(CPU_BASED_VM_EXEC_CONTROL);
5818 	pin_based_exec_ctrl = vmcs_read32(PIN_BASED_VM_EXEC_CONTROL);
5819 	cr4 = vmcs_readl(GUEST_CR4);
5820 	efer = vmcs_read64(GUEST_IA32_EFER);
5821 	secondary_exec_control = 0;
5822 	if (cpu_has_secondary_exec_ctrls())
5823 		secondary_exec_control = vmcs_read32(SECONDARY_VM_EXEC_CONTROL);
5824 
5825 	pr_err("*** Guest State ***\n");
5826 	pr_err("CR0: actual=0x%016lx, shadow=0x%016lx, gh_mask=%016lx\n",
5827 	       vmcs_readl(GUEST_CR0), vmcs_readl(CR0_READ_SHADOW),
5828 	       vmcs_readl(CR0_GUEST_HOST_MASK));
5829 	pr_err("CR4: actual=0x%016lx, shadow=0x%016lx, gh_mask=%016lx\n",
5830 	       cr4, vmcs_readl(CR4_READ_SHADOW), vmcs_readl(CR4_GUEST_HOST_MASK));
5831 	pr_err("CR3 = 0x%016lx\n", vmcs_readl(GUEST_CR3));
5832 	if ((secondary_exec_control & SECONDARY_EXEC_ENABLE_EPT) &&
5833 	    (cr4 & X86_CR4_PAE) && !(efer & EFER_LMA))
5834 	{
5835 		pr_err("PDPTR0 = 0x%016llx  PDPTR1 = 0x%016llx\n",
5836 		       vmcs_read64(GUEST_PDPTR0), vmcs_read64(GUEST_PDPTR1));
5837 		pr_err("PDPTR2 = 0x%016llx  PDPTR3 = 0x%016llx\n",
5838 		       vmcs_read64(GUEST_PDPTR2), vmcs_read64(GUEST_PDPTR3));
5839 	}
5840 	pr_err("RSP = 0x%016lx  RIP = 0x%016lx\n",
5841 	       vmcs_readl(GUEST_RSP), vmcs_readl(GUEST_RIP));
5842 	pr_err("RFLAGS=0x%08lx         DR7 = 0x%016lx\n",
5843 	       vmcs_readl(GUEST_RFLAGS), vmcs_readl(GUEST_DR7));
5844 	pr_err("Sysenter RSP=%016lx CS:RIP=%04x:%016lx\n",
5845 	       vmcs_readl(GUEST_SYSENTER_ESP),
5846 	       vmcs_read32(GUEST_SYSENTER_CS), vmcs_readl(GUEST_SYSENTER_EIP));
5847 	vmx_dump_sel("CS:  ", GUEST_CS_SELECTOR);
5848 	vmx_dump_sel("DS:  ", GUEST_DS_SELECTOR);
5849 	vmx_dump_sel("SS:  ", GUEST_SS_SELECTOR);
5850 	vmx_dump_sel("ES:  ", GUEST_ES_SELECTOR);
5851 	vmx_dump_sel("FS:  ", GUEST_FS_SELECTOR);
5852 	vmx_dump_sel("GS:  ", GUEST_GS_SELECTOR);
5853 	vmx_dump_dtsel("GDTR:", GUEST_GDTR_LIMIT);
5854 	vmx_dump_sel("LDTR:", GUEST_LDTR_SELECTOR);
5855 	vmx_dump_dtsel("IDTR:", GUEST_IDTR_LIMIT);
5856 	vmx_dump_sel("TR:  ", GUEST_TR_SELECTOR);
5857 	if ((vmexit_ctl & (VM_EXIT_SAVE_IA32_PAT | VM_EXIT_SAVE_IA32_EFER)) ||
5858 	    (vmentry_ctl & (VM_ENTRY_LOAD_IA32_PAT | VM_ENTRY_LOAD_IA32_EFER)))
5859 		pr_err("EFER =     0x%016llx  PAT = 0x%016llx\n",
5860 		       efer, vmcs_read64(GUEST_IA32_PAT));
5861 	pr_err("DebugCtl = 0x%016llx  DebugExceptions = 0x%016lx\n",
5862 	       vmcs_read64(GUEST_IA32_DEBUGCTL),
5863 	       vmcs_readl(GUEST_PENDING_DBG_EXCEPTIONS));
5864 	if (cpu_has_load_perf_global_ctrl() &&
5865 	    vmentry_ctl & VM_ENTRY_LOAD_IA32_PERF_GLOBAL_CTRL)
5866 		pr_err("PerfGlobCtl = 0x%016llx\n",
5867 		       vmcs_read64(GUEST_IA32_PERF_GLOBAL_CTRL));
5868 	if (vmentry_ctl & VM_ENTRY_LOAD_BNDCFGS)
5869 		pr_err("BndCfgS = 0x%016llx\n", vmcs_read64(GUEST_BNDCFGS));
5870 	pr_err("Interruptibility = %08x  ActivityState = %08x\n",
5871 	       vmcs_read32(GUEST_INTERRUPTIBILITY_INFO),
5872 	       vmcs_read32(GUEST_ACTIVITY_STATE));
5873 	if (secondary_exec_control & SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY)
5874 		pr_err("InterruptStatus = %04x\n",
5875 		       vmcs_read16(GUEST_INTR_STATUS));
5876 
5877 	pr_err("*** Host State ***\n");
5878 	pr_err("RIP = 0x%016lx  RSP = 0x%016lx\n",
5879 	       vmcs_readl(HOST_RIP), vmcs_readl(HOST_RSP));
5880 	pr_err("CS=%04x SS=%04x DS=%04x ES=%04x FS=%04x GS=%04x TR=%04x\n",
5881 	       vmcs_read16(HOST_CS_SELECTOR), vmcs_read16(HOST_SS_SELECTOR),
5882 	       vmcs_read16(HOST_DS_SELECTOR), vmcs_read16(HOST_ES_SELECTOR),
5883 	       vmcs_read16(HOST_FS_SELECTOR), vmcs_read16(HOST_GS_SELECTOR),
5884 	       vmcs_read16(HOST_TR_SELECTOR));
5885 	pr_err("FSBase=%016lx GSBase=%016lx TRBase=%016lx\n",
5886 	       vmcs_readl(HOST_FS_BASE), vmcs_readl(HOST_GS_BASE),
5887 	       vmcs_readl(HOST_TR_BASE));
5888 	pr_err("GDTBase=%016lx IDTBase=%016lx\n",
5889 	       vmcs_readl(HOST_GDTR_BASE), vmcs_readl(HOST_IDTR_BASE));
5890 	pr_err("CR0=%016lx CR3=%016lx CR4=%016lx\n",
5891 	       vmcs_readl(HOST_CR0), vmcs_readl(HOST_CR3),
5892 	       vmcs_readl(HOST_CR4));
5893 	pr_err("Sysenter RSP=%016lx CS:RIP=%04x:%016lx\n",
5894 	       vmcs_readl(HOST_IA32_SYSENTER_ESP),
5895 	       vmcs_read32(HOST_IA32_SYSENTER_CS),
5896 	       vmcs_readl(HOST_IA32_SYSENTER_EIP));
5897 	if (vmexit_ctl & (VM_EXIT_LOAD_IA32_PAT | VM_EXIT_LOAD_IA32_EFER))
5898 		pr_err("EFER = 0x%016llx  PAT = 0x%016llx\n",
5899 		       vmcs_read64(HOST_IA32_EFER),
5900 		       vmcs_read64(HOST_IA32_PAT));
5901 	if (cpu_has_load_perf_global_ctrl() &&
5902 	    vmexit_ctl & VM_EXIT_LOAD_IA32_PERF_GLOBAL_CTRL)
5903 		pr_err("PerfGlobCtl = 0x%016llx\n",
5904 		       vmcs_read64(HOST_IA32_PERF_GLOBAL_CTRL));
5905 
5906 	pr_err("*** Control State ***\n");
5907 	pr_err("PinBased=%08x CPUBased=%08x SecondaryExec=%08x\n",
5908 	       pin_based_exec_ctrl, cpu_based_exec_ctrl, secondary_exec_control);
5909 	pr_err("EntryControls=%08x ExitControls=%08x\n", vmentry_ctl, vmexit_ctl);
5910 	pr_err("ExceptionBitmap=%08x PFECmask=%08x PFECmatch=%08x\n",
5911 	       vmcs_read32(EXCEPTION_BITMAP),
5912 	       vmcs_read32(PAGE_FAULT_ERROR_CODE_MASK),
5913 	       vmcs_read32(PAGE_FAULT_ERROR_CODE_MATCH));
5914 	pr_err("VMEntry: intr_info=%08x errcode=%08x ilen=%08x\n",
5915 	       vmcs_read32(VM_ENTRY_INTR_INFO_FIELD),
5916 	       vmcs_read32(VM_ENTRY_EXCEPTION_ERROR_CODE),
5917 	       vmcs_read32(VM_ENTRY_INSTRUCTION_LEN));
5918 	pr_err("VMExit: intr_info=%08x errcode=%08x ilen=%08x\n",
5919 	       vmcs_read32(VM_EXIT_INTR_INFO),
5920 	       vmcs_read32(VM_EXIT_INTR_ERROR_CODE),
5921 	       vmcs_read32(VM_EXIT_INSTRUCTION_LEN));
5922 	pr_err("        reason=%08x qualification=%016lx\n",
5923 	       vmcs_read32(VM_EXIT_REASON), vmcs_readl(EXIT_QUALIFICATION));
5924 	pr_err("IDTVectoring: info=%08x errcode=%08x\n",
5925 	       vmcs_read32(IDT_VECTORING_INFO_FIELD),
5926 	       vmcs_read32(IDT_VECTORING_ERROR_CODE));
5927 	pr_err("TSC Offset = 0x%016llx\n", vmcs_read64(TSC_OFFSET));
5928 	if (secondary_exec_control & SECONDARY_EXEC_TSC_SCALING)
5929 		pr_err("TSC Multiplier = 0x%016llx\n",
5930 		       vmcs_read64(TSC_MULTIPLIER));
5931 	if (cpu_based_exec_ctrl & CPU_BASED_TPR_SHADOW) {
5932 		if (secondary_exec_control & SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY) {
5933 			u16 status = vmcs_read16(GUEST_INTR_STATUS);
5934 			pr_err("SVI|RVI = %02x|%02x ", status >> 8, status & 0xff);
5935 		}
5936 		pr_cont("TPR Threshold = 0x%02x\n", vmcs_read32(TPR_THRESHOLD));
5937 		if (secondary_exec_control & SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES)
5938 			pr_err("APIC-access addr = 0x%016llx ", vmcs_read64(APIC_ACCESS_ADDR));
5939 		pr_cont("virt-APIC addr = 0x%016llx\n", vmcs_read64(VIRTUAL_APIC_PAGE_ADDR));
5940 	}
5941 	if (pin_based_exec_ctrl & PIN_BASED_POSTED_INTR)
5942 		pr_err("PostedIntrVec = 0x%02x\n", vmcs_read16(POSTED_INTR_NV));
5943 	if ((secondary_exec_control & SECONDARY_EXEC_ENABLE_EPT))
5944 		pr_err("EPT pointer = 0x%016llx\n", vmcs_read64(EPT_POINTER));
5945 	if (secondary_exec_control & SECONDARY_EXEC_PAUSE_LOOP_EXITING)
5946 		pr_err("PLE Gap=%08x Window=%08x\n",
5947 		       vmcs_read32(PLE_GAP), vmcs_read32(PLE_WINDOW));
5948 	if (secondary_exec_control & SECONDARY_EXEC_ENABLE_VPID)
5949 		pr_err("Virtual processor ID = 0x%04x\n",
5950 		       vmcs_read16(VIRTUAL_PROCESSOR_ID));
5951 }
5952 
5953 /*
5954  * The guest has exited.  See if we can fix it or if we need userspace
5955  * assistance.
5956  */
5957 static int vmx_handle_exit(struct kvm_vcpu *vcpu, fastpath_t exit_fastpath)
5958 {
5959 	struct vcpu_vmx *vmx = to_vmx(vcpu);
5960 	u32 exit_reason = vmx->exit_reason;
5961 	u32 vectoring_info = vmx->idt_vectoring_info;
5962 
5963 	/*
5964 	 * Flush logged GPAs PML buffer, this will make dirty_bitmap more
5965 	 * updated. Another good is, in kvm_vm_ioctl_get_dirty_log, before
5966 	 * querying dirty_bitmap, we only need to kick all vcpus out of guest
5967 	 * mode as if vcpus is in root mode, the PML buffer must has been
5968 	 * flushed already.
5969 	 */
5970 	if (enable_pml)
5971 		vmx_flush_pml_buffer(vcpu);
5972 
5973 	/*
5974 	 * We should never reach this point with a pending nested VM-Enter, and
5975 	 * more specifically emulation of L2 due to invalid guest state (see
5976 	 * below) should never happen as that means we incorrectly allowed a
5977 	 * nested VM-Enter with an invalid vmcs12.
5978 	 */
5979 	WARN_ON_ONCE(vmx->nested.nested_run_pending);
5980 
5981 	/* If guest state is invalid, start emulating */
5982 	if (vmx->emulation_required)
5983 		return handle_invalid_guest_state(vcpu);
5984 
5985 	if (is_guest_mode(vcpu)) {
5986 		/*
5987 		 * The host physical addresses of some pages of guest memory
5988 		 * are loaded into the vmcs02 (e.g. vmcs12's Virtual APIC
5989 		 * Page). The CPU may write to these pages via their host
5990 		 * physical address while L2 is running, bypassing any
5991 		 * address-translation-based dirty tracking (e.g. EPT write
5992 		 * protection).
5993 		 *
5994 		 * Mark them dirty on every exit from L2 to prevent them from
5995 		 * getting out of sync with dirty tracking.
5996 		 */
5997 		nested_mark_vmcs12_pages_dirty(vcpu);
5998 
5999 		if (nested_vmx_reflect_vmexit(vcpu))
6000 			return 1;
6001 	}
6002 
6003 	if (exit_reason & VMX_EXIT_REASONS_FAILED_VMENTRY) {
6004 		dump_vmcs();
6005 		vcpu->run->exit_reason = KVM_EXIT_FAIL_ENTRY;
6006 		vcpu->run->fail_entry.hardware_entry_failure_reason
6007 			= exit_reason;
6008 		return 0;
6009 	}
6010 
6011 	if (unlikely(vmx->fail)) {
6012 		dump_vmcs();
6013 		vcpu->run->exit_reason = KVM_EXIT_FAIL_ENTRY;
6014 		vcpu->run->fail_entry.hardware_entry_failure_reason
6015 			= vmcs_read32(VM_INSTRUCTION_ERROR);
6016 		return 0;
6017 	}
6018 
6019 	/*
6020 	 * Note:
6021 	 * Do not try to fix EXIT_REASON_EPT_MISCONFIG if it caused by
6022 	 * delivery event since it indicates guest is accessing MMIO.
6023 	 * The vm-exit can be triggered again after return to guest that
6024 	 * will cause infinite loop.
6025 	 */
6026 	if ((vectoring_info & VECTORING_INFO_VALID_MASK) &&
6027 			(exit_reason != EXIT_REASON_EXCEPTION_NMI &&
6028 			exit_reason != EXIT_REASON_EPT_VIOLATION &&
6029 			exit_reason != EXIT_REASON_PML_FULL &&
6030 			exit_reason != EXIT_REASON_TASK_SWITCH)) {
6031 		vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
6032 		vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_DELIVERY_EV;
6033 		vcpu->run->internal.ndata = 3;
6034 		vcpu->run->internal.data[0] = vectoring_info;
6035 		vcpu->run->internal.data[1] = exit_reason;
6036 		vcpu->run->internal.data[2] = vcpu->arch.exit_qualification;
6037 		if (exit_reason == EXIT_REASON_EPT_MISCONFIG) {
6038 			vcpu->run->internal.ndata++;
6039 			vcpu->run->internal.data[3] =
6040 				vmcs_read64(GUEST_PHYSICAL_ADDRESS);
6041 		}
6042 		return 0;
6043 	}
6044 
6045 	if (unlikely(!enable_vnmi &&
6046 		     vmx->loaded_vmcs->soft_vnmi_blocked)) {
6047 		if (!vmx_interrupt_blocked(vcpu)) {
6048 			vmx->loaded_vmcs->soft_vnmi_blocked = 0;
6049 		} else if (vmx->loaded_vmcs->vnmi_blocked_time > 1000000000LL &&
6050 			   vcpu->arch.nmi_pending) {
6051 			/*
6052 			 * This CPU don't support us in finding the end of an
6053 			 * NMI-blocked window if the guest runs with IRQs
6054 			 * disabled. So we pull the trigger after 1 s of
6055 			 * futile waiting, but inform the user about this.
6056 			 */
6057 			printk(KERN_WARNING "%s: Breaking out of NMI-blocked "
6058 			       "state on VCPU %d after 1 s timeout\n",
6059 			       __func__, vcpu->vcpu_id);
6060 			vmx->loaded_vmcs->soft_vnmi_blocked = 0;
6061 		}
6062 	}
6063 
6064 	if (exit_fastpath != EXIT_FASTPATH_NONE)
6065 		return 1;
6066 
6067 	if (exit_reason >= kvm_vmx_max_exit_handlers)
6068 		goto unexpected_vmexit;
6069 #ifdef CONFIG_RETPOLINE
6070 	if (exit_reason == EXIT_REASON_MSR_WRITE)
6071 		return kvm_emulate_wrmsr(vcpu);
6072 	else if (exit_reason == EXIT_REASON_PREEMPTION_TIMER)
6073 		return handle_preemption_timer(vcpu);
6074 	else if (exit_reason == EXIT_REASON_INTERRUPT_WINDOW)
6075 		return handle_interrupt_window(vcpu);
6076 	else if (exit_reason == EXIT_REASON_EXTERNAL_INTERRUPT)
6077 		return handle_external_interrupt(vcpu);
6078 	else if (exit_reason == EXIT_REASON_HLT)
6079 		return kvm_emulate_halt(vcpu);
6080 	else if (exit_reason == EXIT_REASON_EPT_MISCONFIG)
6081 		return handle_ept_misconfig(vcpu);
6082 #endif
6083 
6084 	exit_reason = array_index_nospec(exit_reason,
6085 					 kvm_vmx_max_exit_handlers);
6086 	if (!kvm_vmx_exit_handlers[exit_reason])
6087 		goto unexpected_vmexit;
6088 
6089 	return kvm_vmx_exit_handlers[exit_reason](vcpu);
6090 
6091 unexpected_vmexit:
6092 	vcpu_unimpl(vcpu, "vmx: unexpected exit reason 0x%x\n", exit_reason);
6093 	dump_vmcs();
6094 	vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
6095 	vcpu->run->internal.suberror =
6096 			KVM_INTERNAL_ERROR_UNEXPECTED_EXIT_REASON;
6097 	vcpu->run->internal.ndata = 1;
6098 	vcpu->run->internal.data[0] = exit_reason;
6099 	return 0;
6100 }
6101 
6102 /*
6103  * Software based L1D cache flush which is used when microcode providing
6104  * the cache control MSR is not loaded.
6105  *
6106  * The L1D cache is 32 KiB on Nehalem and later microarchitectures, but to
6107  * flush it is required to read in 64 KiB because the replacement algorithm
6108  * is not exactly LRU. This could be sized at runtime via topology
6109  * information but as all relevant affected CPUs have 32KiB L1D cache size
6110  * there is no point in doing so.
6111  */
6112 static void vmx_l1d_flush(struct kvm_vcpu *vcpu)
6113 {
6114 	int size = PAGE_SIZE << L1D_CACHE_ORDER;
6115 
6116 	/*
6117 	 * This code is only executed when the the flush mode is 'cond' or
6118 	 * 'always'
6119 	 */
6120 	if (static_branch_likely(&vmx_l1d_flush_cond)) {
6121 		bool flush_l1d;
6122 
6123 		/*
6124 		 * Clear the per-vcpu flush bit, it gets set again
6125 		 * either from vcpu_run() or from one of the unsafe
6126 		 * VMEXIT handlers.
6127 		 */
6128 		flush_l1d = vcpu->arch.l1tf_flush_l1d;
6129 		vcpu->arch.l1tf_flush_l1d = false;
6130 
6131 		/*
6132 		 * Clear the per-cpu flush bit, it gets set again from
6133 		 * the interrupt handlers.
6134 		 */
6135 		flush_l1d |= kvm_get_cpu_l1tf_flush_l1d();
6136 		kvm_clear_cpu_l1tf_flush_l1d();
6137 
6138 		if (!flush_l1d)
6139 			return;
6140 	}
6141 
6142 	vcpu->stat.l1d_flush++;
6143 
6144 	if (static_cpu_has(X86_FEATURE_FLUSH_L1D)) {
6145 		wrmsrl(MSR_IA32_FLUSH_CMD, L1D_FLUSH);
6146 		return;
6147 	}
6148 
6149 	asm volatile(
6150 		/* First ensure the pages are in the TLB */
6151 		"xorl	%%eax, %%eax\n"
6152 		".Lpopulate_tlb:\n\t"
6153 		"movzbl	(%[flush_pages], %%" _ASM_AX "), %%ecx\n\t"
6154 		"addl	$4096, %%eax\n\t"
6155 		"cmpl	%%eax, %[size]\n\t"
6156 		"jne	.Lpopulate_tlb\n\t"
6157 		"xorl	%%eax, %%eax\n\t"
6158 		"cpuid\n\t"
6159 		/* Now fill the cache */
6160 		"xorl	%%eax, %%eax\n"
6161 		".Lfill_cache:\n"
6162 		"movzbl	(%[flush_pages], %%" _ASM_AX "), %%ecx\n\t"
6163 		"addl	$64, %%eax\n\t"
6164 		"cmpl	%%eax, %[size]\n\t"
6165 		"jne	.Lfill_cache\n\t"
6166 		"lfence\n"
6167 		:: [flush_pages] "r" (vmx_l1d_flush_pages),
6168 		    [size] "r" (size)
6169 		: "eax", "ebx", "ecx", "edx");
6170 }
6171 
6172 static void update_cr8_intercept(struct kvm_vcpu *vcpu, int tpr, int irr)
6173 {
6174 	struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
6175 	int tpr_threshold;
6176 
6177 	if (is_guest_mode(vcpu) &&
6178 		nested_cpu_has(vmcs12, CPU_BASED_TPR_SHADOW))
6179 		return;
6180 
6181 	tpr_threshold = (irr == -1 || tpr < irr) ? 0 : irr;
6182 	if (is_guest_mode(vcpu))
6183 		to_vmx(vcpu)->nested.l1_tpr_threshold = tpr_threshold;
6184 	else
6185 		vmcs_write32(TPR_THRESHOLD, tpr_threshold);
6186 }
6187 
6188 void vmx_set_virtual_apic_mode(struct kvm_vcpu *vcpu)
6189 {
6190 	struct vcpu_vmx *vmx = to_vmx(vcpu);
6191 	u32 sec_exec_control;
6192 
6193 	if (!lapic_in_kernel(vcpu))
6194 		return;
6195 
6196 	if (!flexpriority_enabled &&
6197 	    !cpu_has_vmx_virtualize_x2apic_mode())
6198 		return;
6199 
6200 	/* Postpone execution until vmcs01 is the current VMCS. */
6201 	if (is_guest_mode(vcpu)) {
6202 		vmx->nested.change_vmcs01_virtual_apic_mode = true;
6203 		return;
6204 	}
6205 
6206 	sec_exec_control = secondary_exec_controls_get(vmx);
6207 	sec_exec_control &= ~(SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES |
6208 			      SECONDARY_EXEC_VIRTUALIZE_X2APIC_MODE);
6209 
6210 	switch (kvm_get_apic_mode(vcpu)) {
6211 	case LAPIC_MODE_INVALID:
6212 		WARN_ONCE(true, "Invalid local APIC state");
6213 	case LAPIC_MODE_DISABLED:
6214 		break;
6215 	case LAPIC_MODE_XAPIC:
6216 		if (flexpriority_enabled) {
6217 			sec_exec_control |=
6218 				SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES;
6219 			kvm_make_request(KVM_REQ_APIC_PAGE_RELOAD, vcpu);
6220 
6221 			/*
6222 			 * Flush the TLB, reloading the APIC access page will
6223 			 * only do so if its physical address has changed, but
6224 			 * the guest may have inserted a non-APIC mapping into
6225 			 * the TLB while the APIC access page was disabled.
6226 			 */
6227 			kvm_make_request(KVM_REQ_TLB_FLUSH_CURRENT, vcpu);
6228 		}
6229 		break;
6230 	case LAPIC_MODE_X2APIC:
6231 		if (cpu_has_vmx_virtualize_x2apic_mode())
6232 			sec_exec_control |=
6233 				SECONDARY_EXEC_VIRTUALIZE_X2APIC_MODE;
6234 		break;
6235 	}
6236 	secondary_exec_controls_set(vmx, sec_exec_control);
6237 
6238 	vmx_update_msr_bitmap(vcpu);
6239 }
6240 
6241 static void vmx_set_apic_access_page_addr(struct kvm_vcpu *vcpu)
6242 {
6243 	struct page *page;
6244 
6245 	/* Defer reload until vmcs01 is the current VMCS. */
6246 	if (is_guest_mode(vcpu)) {
6247 		to_vmx(vcpu)->nested.reload_vmcs01_apic_access_page = true;
6248 		return;
6249 	}
6250 
6251 	if (!(secondary_exec_controls_get(to_vmx(vcpu)) &
6252 	    SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES))
6253 		return;
6254 
6255 	page = gfn_to_page(vcpu->kvm, APIC_DEFAULT_PHYS_BASE >> PAGE_SHIFT);
6256 	if (is_error_page(page))
6257 		return;
6258 
6259 	vmcs_write64(APIC_ACCESS_ADDR, page_to_phys(page));
6260 	vmx_flush_tlb_current(vcpu);
6261 
6262 	/*
6263 	 * Do not pin apic access page in memory, the MMU notifier
6264 	 * will call us again if it is migrated or swapped out.
6265 	 */
6266 	put_page(page);
6267 }
6268 
6269 static void vmx_hwapic_isr_update(struct kvm_vcpu *vcpu, int max_isr)
6270 {
6271 	u16 status;
6272 	u8 old;
6273 
6274 	if (max_isr == -1)
6275 		max_isr = 0;
6276 
6277 	status = vmcs_read16(GUEST_INTR_STATUS);
6278 	old = status >> 8;
6279 	if (max_isr != old) {
6280 		status &= 0xff;
6281 		status |= max_isr << 8;
6282 		vmcs_write16(GUEST_INTR_STATUS, status);
6283 	}
6284 }
6285 
6286 static void vmx_set_rvi(int vector)
6287 {
6288 	u16 status;
6289 	u8 old;
6290 
6291 	if (vector == -1)
6292 		vector = 0;
6293 
6294 	status = vmcs_read16(GUEST_INTR_STATUS);
6295 	old = (u8)status & 0xff;
6296 	if ((u8)vector != old) {
6297 		status &= ~0xff;
6298 		status |= (u8)vector;
6299 		vmcs_write16(GUEST_INTR_STATUS, status);
6300 	}
6301 }
6302 
6303 static void vmx_hwapic_irr_update(struct kvm_vcpu *vcpu, int max_irr)
6304 {
6305 	/*
6306 	 * When running L2, updating RVI is only relevant when
6307 	 * vmcs12 virtual-interrupt-delivery enabled.
6308 	 * However, it can be enabled only when L1 also
6309 	 * intercepts external-interrupts and in that case
6310 	 * we should not update vmcs02 RVI but instead intercept
6311 	 * interrupt. Therefore, do nothing when running L2.
6312 	 */
6313 	if (!is_guest_mode(vcpu))
6314 		vmx_set_rvi(max_irr);
6315 }
6316 
6317 static int vmx_sync_pir_to_irr(struct kvm_vcpu *vcpu)
6318 {
6319 	struct vcpu_vmx *vmx = to_vmx(vcpu);
6320 	int max_irr;
6321 	bool max_irr_updated;
6322 
6323 	WARN_ON(!vcpu->arch.apicv_active);
6324 	if (pi_test_on(&vmx->pi_desc)) {
6325 		pi_clear_on(&vmx->pi_desc);
6326 		/*
6327 		 * IOMMU can write to PID.ON, so the barrier matters even on UP.
6328 		 * But on x86 this is just a compiler barrier anyway.
6329 		 */
6330 		smp_mb__after_atomic();
6331 		max_irr_updated =
6332 			kvm_apic_update_irr(vcpu, vmx->pi_desc.pir, &max_irr);
6333 
6334 		/*
6335 		 * If we are running L2 and L1 has a new pending interrupt
6336 		 * which can be injected, we should re-evaluate
6337 		 * what should be done with this new L1 interrupt.
6338 		 * If L1 intercepts external-interrupts, we should
6339 		 * exit from L2 to L1. Otherwise, interrupt should be
6340 		 * delivered directly to L2.
6341 		 */
6342 		if (is_guest_mode(vcpu) && max_irr_updated) {
6343 			if (nested_exit_on_intr(vcpu))
6344 				kvm_vcpu_exiting_guest_mode(vcpu);
6345 			else
6346 				kvm_make_request(KVM_REQ_EVENT, vcpu);
6347 		}
6348 	} else {
6349 		max_irr = kvm_lapic_find_highest_irr(vcpu);
6350 	}
6351 	vmx_hwapic_irr_update(vcpu, max_irr);
6352 	return max_irr;
6353 }
6354 
6355 static bool vmx_dy_apicv_has_pending_interrupt(struct kvm_vcpu *vcpu)
6356 {
6357 	struct pi_desc *pi_desc = vcpu_to_pi_desc(vcpu);
6358 
6359 	return pi_test_on(pi_desc) ||
6360 		(pi_test_sn(pi_desc) && !pi_is_pir_empty(pi_desc));
6361 }
6362 
6363 static void vmx_load_eoi_exitmap(struct kvm_vcpu *vcpu, u64 *eoi_exit_bitmap)
6364 {
6365 	if (!kvm_vcpu_apicv_active(vcpu))
6366 		return;
6367 
6368 	vmcs_write64(EOI_EXIT_BITMAP0, eoi_exit_bitmap[0]);
6369 	vmcs_write64(EOI_EXIT_BITMAP1, eoi_exit_bitmap[1]);
6370 	vmcs_write64(EOI_EXIT_BITMAP2, eoi_exit_bitmap[2]);
6371 	vmcs_write64(EOI_EXIT_BITMAP3, eoi_exit_bitmap[3]);
6372 }
6373 
6374 static void vmx_apicv_post_state_restore(struct kvm_vcpu *vcpu)
6375 {
6376 	struct vcpu_vmx *vmx = to_vmx(vcpu);
6377 
6378 	pi_clear_on(&vmx->pi_desc);
6379 	memset(vmx->pi_desc.pir, 0, sizeof(vmx->pi_desc.pir));
6380 }
6381 
6382 static void handle_exception_nmi_irqoff(struct vcpu_vmx *vmx)
6383 {
6384 	u32 intr_info = vmx_get_intr_info(&vmx->vcpu);
6385 
6386 	/* if exit due to PF check for async PF */
6387 	if (is_page_fault(intr_info)) {
6388 		vmx->vcpu.arch.apf.host_apf_flags = kvm_read_and_reset_apf_flags();
6389 	/* Handle machine checks before interrupts are enabled */
6390 	} else if (is_machine_check(intr_info)) {
6391 		kvm_machine_check();
6392 	/* We need to handle NMIs before interrupts are enabled */
6393 	} else if (is_nmi(intr_info)) {
6394 		kvm_before_interrupt(&vmx->vcpu);
6395 		asm("int $2");
6396 		kvm_after_interrupt(&vmx->vcpu);
6397 	}
6398 }
6399 
6400 static void handle_external_interrupt_irqoff(struct kvm_vcpu *vcpu)
6401 {
6402 	unsigned int vector;
6403 	unsigned long entry;
6404 #ifdef CONFIG_X86_64
6405 	unsigned long tmp;
6406 #endif
6407 	gate_desc *desc;
6408 	u32 intr_info = vmx_get_intr_info(vcpu);
6409 
6410 	if (WARN_ONCE(!is_external_intr(intr_info),
6411 	    "KVM: unexpected VM-Exit interrupt info: 0x%x", intr_info))
6412 		return;
6413 
6414 	vector = intr_info & INTR_INFO_VECTOR_MASK;
6415 	desc = (gate_desc *)host_idt_base + vector;
6416 	entry = gate_offset(desc);
6417 
6418 	kvm_before_interrupt(vcpu);
6419 
6420 	asm volatile(
6421 #ifdef CONFIG_X86_64
6422 		"mov %%rsp, %[sp]\n\t"
6423 		"and $-16, %%rsp\n\t"
6424 		"push %[ss]\n\t"
6425 		"push %[sp]\n\t"
6426 #endif
6427 		"pushf\n\t"
6428 		"push %[cs]\n\t"
6429 		CALL_NOSPEC
6430 		:
6431 #ifdef CONFIG_X86_64
6432 		[sp]"=&r"(tmp),
6433 #endif
6434 		ASM_CALL_CONSTRAINT
6435 		:
6436 		[thunk_target]"r"(entry),
6437 #ifdef CONFIG_X86_64
6438 		[ss]"i"(__KERNEL_DS),
6439 #endif
6440 		[cs]"i"(__KERNEL_CS)
6441 	);
6442 
6443 	kvm_after_interrupt(vcpu);
6444 }
6445 STACK_FRAME_NON_STANDARD(handle_external_interrupt_irqoff);
6446 
6447 static void vmx_handle_exit_irqoff(struct kvm_vcpu *vcpu)
6448 {
6449 	struct vcpu_vmx *vmx = to_vmx(vcpu);
6450 
6451 	if (vmx->exit_reason == EXIT_REASON_EXTERNAL_INTERRUPT)
6452 		handle_external_interrupt_irqoff(vcpu);
6453 	else if (vmx->exit_reason == EXIT_REASON_EXCEPTION_NMI)
6454 		handle_exception_nmi_irqoff(vmx);
6455 }
6456 
6457 static bool vmx_has_emulated_msr(u32 index)
6458 {
6459 	switch (index) {
6460 	case MSR_IA32_SMBASE:
6461 		/*
6462 		 * We cannot do SMM unless we can run the guest in big
6463 		 * real mode.
6464 		 */
6465 		return enable_unrestricted_guest || emulate_invalid_guest_state;
6466 	case MSR_IA32_VMX_BASIC ... MSR_IA32_VMX_VMFUNC:
6467 		return nested;
6468 	case MSR_AMD64_VIRT_SPEC_CTRL:
6469 		/* This is AMD only.  */
6470 		return false;
6471 	default:
6472 		return true;
6473 	}
6474 }
6475 
6476 static void vmx_recover_nmi_blocking(struct vcpu_vmx *vmx)
6477 {
6478 	u32 exit_intr_info;
6479 	bool unblock_nmi;
6480 	u8 vector;
6481 	bool idtv_info_valid;
6482 
6483 	idtv_info_valid = vmx->idt_vectoring_info & VECTORING_INFO_VALID_MASK;
6484 
6485 	if (enable_vnmi) {
6486 		if (vmx->loaded_vmcs->nmi_known_unmasked)
6487 			return;
6488 
6489 		exit_intr_info = vmx_get_intr_info(&vmx->vcpu);
6490 		unblock_nmi = (exit_intr_info & INTR_INFO_UNBLOCK_NMI) != 0;
6491 		vector = exit_intr_info & INTR_INFO_VECTOR_MASK;
6492 		/*
6493 		 * SDM 3: 27.7.1.2 (September 2008)
6494 		 * Re-set bit "block by NMI" before VM entry if vmexit caused by
6495 		 * a guest IRET fault.
6496 		 * SDM 3: 23.2.2 (September 2008)
6497 		 * Bit 12 is undefined in any of the following cases:
6498 		 *  If the VM exit sets the valid bit in the IDT-vectoring
6499 		 *   information field.
6500 		 *  If the VM exit is due to a double fault.
6501 		 */
6502 		if ((exit_intr_info & INTR_INFO_VALID_MASK) && unblock_nmi &&
6503 		    vector != DF_VECTOR && !idtv_info_valid)
6504 			vmcs_set_bits(GUEST_INTERRUPTIBILITY_INFO,
6505 				      GUEST_INTR_STATE_NMI);
6506 		else
6507 			vmx->loaded_vmcs->nmi_known_unmasked =
6508 				!(vmcs_read32(GUEST_INTERRUPTIBILITY_INFO)
6509 				  & GUEST_INTR_STATE_NMI);
6510 	} else if (unlikely(vmx->loaded_vmcs->soft_vnmi_blocked))
6511 		vmx->loaded_vmcs->vnmi_blocked_time +=
6512 			ktime_to_ns(ktime_sub(ktime_get(),
6513 					      vmx->loaded_vmcs->entry_time));
6514 }
6515 
6516 static void __vmx_complete_interrupts(struct kvm_vcpu *vcpu,
6517 				      u32 idt_vectoring_info,
6518 				      int instr_len_field,
6519 				      int error_code_field)
6520 {
6521 	u8 vector;
6522 	int type;
6523 	bool idtv_info_valid;
6524 
6525 	idtv_info_valid = idt_vectoring_info & VECTORING_INFO_VALID_MASK;
6526 
6527 	vcpu->arch.nmi_injected = false;
6528 	kvm_clear_exception_queue(vcpu);
6529 	kvm_clear_interrupt_queue(vcpu);
6530 
6531 	if (!idtv_info_valid)
6532 		return;
6533 
6534 	kvm_make_request(KVM_REQ_EVENT, vcpu);
6535 
6536 	vector = idt_vectoring_info & VECTORING_INFO_VECTOR_MASK;
6537 	type = idt_vectoring_info & VECTORING_INFO_TYPE_MASK;
6538 
6539 	switch (type) {
6540 	case INTR_TYPE_NMI_INTR:
6541 		vcpu->arch.nmi_injected = true;
6542 		/*
6543 		 * SDM 3: 27.7.1.2 (September 2008)
6544 		 * Clear bit "block by NMI" before VM entry if a NMI
6545 		 * delivery faulted.
6546 		 */
6547 		vmx_set_nmi_mask(vcpu, false);
6548 		break;
6549 	case INTR_TYPE_SOFT_EXCEPTION:
6550 		vcpu->arch.event_exit_inst_len = vmcs_read32(instr_len_field);
6551 		/* fall through */
6552 	case INTR_TYPE_HARD_EXCEPTION:
6553 		if (idt_vectoring_info & VECTORING_INFO_DELIVER_CODE_MASK) {
6554 			u32 err = vmcs_read32(error_code_field);
6555 			kvm_requeue_exception_e(vcpu, vector, err);
6556 		} else
6557 			kvm_requeue_exception(vcpu, vector);
6558 		break;
6559 	case INTR_TYPE_SOFT_INTR:
6560 		vcpu->arch.event_exit_inst_len = vmcs_read32(instr_len_field);
6561 		/* fall through */
6562 	case INTR_TYPE_EXT_INTR:
6563 		kvm_queue_interrupt(vcpu, vector, type == INTR_TYPE_SOFT_INTR);
6564 		break;
6565 	default:
6566 		break;
6567 	}
6568 }
6569 
6570 static void vmx_complete_interrupts(struct vcpu_vmx *vmx)
6571 {
6572 	__vmx_complete_interrupts(&vmx->vcpu, vmx->idt_vectoring_info,
6573 				  VM_EXIT_INSTRUCTION_LEN,
6574 				  IDT_VECTORING_ERROR_CODE);
6575 }
6576 
6577 static void vmx_cancel_injection(struct kvm_vcpu *vcpu)
6578 {
6579 	__vmx_complete_interrupts(vcpu,
6580 				  vmcs_read32(VM_ENTRY_INTR_INFO_FIELD),
6581 				  VM_ENTRY_INSTRUCTION_LEN,
6582 				  VM_ENTRY_EXCEPTION_ERROR_CODE);
6583 
6584 	vmcs_write32(VM_ENTRY_INTR_INFO_FIELD, 0);
6585 }
6586 
6587 static void atomic_switch_perf_msrs(struct vcpu_vmx *vmx)
6588 {
6589 	int i, nr_msrs;
6590 	struct perf_guest_switch_msr *msrs;
6591 
6592 	msrs = perf_guest_get_msrs(&nr_msrs);
6593 
6594 	if (!msrs)
6595 		return;
6596 
6597 	for (i = 0; i < nr_msrs; i++)
6598 		if (msrs[i].host == msrs[i].guest)
6599 			clear_atomic_switch_msr(vmx, msrs[i].msr);
6600 		else
6601 			add_atomic_switch_msr(vmx, msrs[i].msr, msrs[i].guest,
6602 					msrs[i].host, false);
6603 }
6604 
6605 static void vmx_update_hv_timer(struct kvm_vcpu *vcpu)
6606 {
6607 	struct vcpu_vmx *vmx = to_vmx(vcpu);
6608 	u64 tscl;
6609 	u32 delta_tsc;
6610 
6611 	if (vmx->req_immediate_exit) {
6612 		vmcs_write32(VMX_PREEMPTION_TIMER_VALUE, 0);
6613 		vmx->loaded_vmcs->hv_timer_soft_disabled = false;
6614 	} else if (vmx->hv_deadline_tsc != -1) {
6615 		tscl = rdtsc();
6616 		if (vmx->hv_deadline_tsc > tscl)
6617 			/* set_hv_timer ensures the delta fits in 32-bits */
6618 			delta_tsc = (u32)((vmx->hv_deadline_tsc - tscl) >>
6619 				cpu_preemption_timer_multi);
6620 		else
6621 			delta_tsc = 0;
6622 
6623 		vmcs_write32(VMX_PREEMPTION_TIMER_VALUE, delta_tsc);
6624 		vmx->loaded_vmcs->hv_timer_soft_disabled = false;
6625 	} else if (!vmx->loaded_vmcs->hv_timer_soft_disabled) {
6626 		vmcs_write32(VMX_PREEMPTION_TIMER_VALUE, -1);
6627 		vmx->loaded_vmcs->hv_timer_soft_disabled = true;
6628 	}
6629 }
6630 
6631 void vmx_update_host_rsp(struct vcpu_vmx *vmx, unsigned long host_rsp)
6632 {
6633 	if (unlikely(host_rsp != vmx->loaded_vmcs->host_state.rsp)) {
6634 		vmx->loaded_vmcs->host_state.rsp = host_rsp;
6635 		vmcs_writel(HOST_RSP, host_rsp);
6636 	}
6637 }
6638 
6639 static fastpath_t vmx_exit_handlers_fastpath(struct kvm_vcpu *vcpu)
6640 {
6641 	switch (to_vmx(vcpu)->exit_reason) {
6642 	case EXIT_REASON_MSR_WRITE:
6643 		return handle_fastpath_set_msr_irqoff(vcpu);
6644 	case EXIT_REASON_PREEMPTION_TIMER:
6645 		return handle_fastpath_preemption_timer(vcpu);
6646 	default:
6647 		return EXIT_FASTPATH_NONE;
6648 	}
6649 }
6650 
6651 bool __vmx_vcpu_run(struct vcpu_vmx *vmx, unsigned long *regs, bool launched);
6652 
6653 static fastpath_t vmx_vcpu_run(struct kvm_vcpu *vcpu)
6654 {
6655 	fastpath_t exit_fastpath;
6656 	struct vcpu_vmx *vmx = to_vmx(vcpu);
6657 	unsigned long cr3, cr4;
6658 
6659 reenter_guest:
6660 	/* Record the guest's net vcpu time for enforced NMI injections. */
6661 	if (unlikely(!enable_vnmi &&
6662 		     vmx->loaded_vmcs->soft_vnmi_blocked))
6663 		vmx->loaded_vmcs->entry_time = ktime_get();
6664 
6665 	/* Don't enter VMX if guest state is invalid, let the exit handler
6666 	   start emulation until we arrive back to a valid state */
6667 	if (vmx->emulation_required)
6668 		return EXIT_FASTPATH_NONE;
6669 
6670 	if (vmx->ple_window_dirty) {
6671 		vmx->ple_window_dirty = false;
6672 		vmcs_write32(PLE_WINDOW, vmx->ple_window);
6673 	}
6674 
6675 	/*
6676 	 * We did this in prepare_switch_to_guest, because it needs to
6677 	 * be within srcu_read_lock.
6678 	 */
6679 	WARN_ON_ONCE(vmx->nested.need_vmcs12_to_shadow_sync);
6680 
6681 	if (kvm_register_is_dirty(vcpu, VCPU_REGS_RSP))
6682 		vmcs_writel(GUEST_RSP, vcpu->arch.regs[VCPU_REGS_RSP]);
6683 	if (kvm_register_is_dirty(vcpu, VCPU_REGS_RIP))
6684 		vmcs_writel(GUEST_RIP, vcpu->arch.regs[VCPU_REGS_RIP]);
6685 
6686 	cr3 = __get_current_cr3_fast();
6687 	if (unlikely(cr3 != vmx->loaded_vmcs->host_state.cr3)) {
6688 		vmcs_writel(HOST_CR3, cr3);
6689 		vmx->loaded_vmcs->host_state.cr3 = cr3;
6690 	}
6691 
6692 	cr4 = cr4_read_shadow();
6693 	if (unlikely(cr4 != vmx->loaded_vmcs->host_state.cr4)) {
6694 		vmcs_writel(HOST_CR4, cr4);
6695 		vmx->loaded_vmcs->host_state.cr4 = cr4;
6696 	}
6697 
6698 	/* When single-stepping over STI and MOV SS, we must clear the
6699 	 * corresponding interruptibility bits in the guest state. Otherwise
6700 	 * vmentry fails as it then expects bit 14 (BS) in pending debug
6701 	 * exceptions being set, but that's not correct for the guest debugging
6702 	 * case. */
6703 	if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)
6704 		vmx_set_interrupt_shadow(vcpu, 0);
6705 
6706 	kvm_load_guest_xsave_state(vcpu);
6707 
6708 	pt_guest_enter(vmx);
6709 
6710 	atomic_switch_perf_msrs(vmx);
6711 
6712 	if (enable_preemption_timer)
6713 		vmx_update_hv_timer(vcpu);
6714 
6715 	if (lapic_in_kernel(vcpu) &&
6716 		vcpu->arch.apic->lapic_timer.timer_advance_ns)
6717 		kvm_wait_lapic_expire(vcpu);
6718 
6719 	/*
6720 	 * If this vCPU has touched SPEC_CTRL, restore the guest's value if
6721 	 * it's non-zero. Since vmentry is serialising on affected CPUs, there
6722 	 * is no need to worry about the conditional branch over the wrmsr
6723 	 * being speculatively taken.
6724 	 */
6725 	x86_spec_ctrl_set_guest(vmx->spec_ctrl, 0);
6726 
6727 	/* L1D Flush includes CPU buffer clear to mitigate MDS */
6728 	if (static_branch_unlikely(&vmx_l1d_should_flush))
6729 		vmx_l1d_flush(vcpu);
6730 	else if (static_branch_unlikely(&mds_user_clear))
6731 		mds_clear_cpu_buffers();
6732 
6733 	if (vcpu->arch.cr2 != read_cr2())
6734 		write_cr2(vcpu->arch.cr2);
6735 
6736 	vmx->fail = __vmx_vcpu_run(vmx, (unsigned long *)&vcpu->arch.regs,
6737 				   vmx->loaded_vmcs->launched);
6738 
6739 	vcpu->arch.cr2 = read_cr2();
6740 
6741 	/*
6742 	 * We do not use IBRS in the kernel. If this vCPU has used the
6743 	 * SPEC_CTRL MSR it may have left it on; save the value and
6744 	 * turn it off. This is much more efficient than blindly adding
6745 	 * it to the atomic save/restore list. Especially as the former
6746 	 * (Saving guest MSRs on vmexit) doesn't even exist in KVM.
6747 	 *
6748 	 * For non-nested case:
6749 	 * If the L01 MSR bitmap does not intercept the MSR, then we need to
6750 	 * save it.
6751 	 *
6752 	 * For nested case:
6753 	 * If the L02 MSR bitmap does not intercept the MSR, then we need to
6754 	 * save it.
6755 	 */
6756 	if (unlikely(!msr_write_intercepted(vcpu, MSR_IA32_SPEC_CTRL)))
6757 		vmx->spec_ctrl = native_read_msr(MSR_IA32_SPEC_CTRL);
6758 
6759 	x86_spec_ctrl_restore_host(vmx->spec_ctrl, 0);
6760 
6761 	/* All fields are clean at this point */
6762 	if (static_branch_unlikely(&enable_evmcs))
6763 		current_evmcs->hv_clean_fields |=
6764 			HV_VMX_ENLIGHTENED_CLEAN_FIELD_ALL;
6765 
6766 	if (static_branch_unlikely(&enable_evmcs))
6767 		current_evmcs->hv_vp_id = vcpu->arch.hyperv.vp_index;
6768 
6769 	/* MSR_IA32_DEBUGCTLMSR is zeroed on vmexit. Restore it if needed */
6770 	if (vmx->host_debugctlmsr)
6771 		update_debugctlmsr(vmx->host_debugctlmsr);
6772 
6773 #ifndef CONFIG_X86_64
6774 	/*
6775 	 * The sysexit path does not restore ds/es, so we must set them to
6776 	 * a reasonable value ourselves.
6777 	 *
6778 	 * We can't defer this to vmx_prepare_switch_to_host() since that
6779 	 * function may be executed in interrupt context, which saves and
6780 	 * restore segments around it, nullifying its effect.
6781 	 */
6782 	loadsegment(ds, __USER_DS);
6783 	loadsegment(es, __USER_DS);
6784 #endif
6785 
6786 	vmx_register_cache_reset(vcpu);
6787 
6788 	pt_guest_exit(vmx);
6789 
6790 	kvm_load_host_xsave_state(vcpu);
6791 
6792 	vmx->nested.nested_run_pending = 0;
6793 	vmx->idt_vectoring_info = 0;
6794 
6795 	if (unlikely(vmx->fail)) {
6796 		vmx->exit_reason = 0xdead;
6797 		return EXIT_FASTPATH_NONE;
6798 	}
6799 
6800 	vmx->exit_reason = vmcs_read32(VM_EXIT_REASON);
6801 	if (unlikely((u16)vmx->exit_reason == EXIT_REASON_MCE_DURING_VMENTRY))
6802 		kvm_machine_check();
6803 
6804 	trace_kvm_exit(vmx->exit_reason, vcpu, KVM_ISA_VMX);
6805 
6806 	if (unlikely(vmx->exit_reason & VMX_EXIT_REASONS_FAILED_VMENTRY))
6807 		return EXIT_FASTPATH_NONE;
6808 
6809 	vmx->loaded_vmcs->launched = 1;
6810 	vmx->idt_vectoring_info = vmcs_read32(IDT_VECTORING_INFO_FIELD);
6811 
6812 	vmx_recover_nmi_blocking(vmx);
6813 	vmx_complete_interrupts(vmx);
6814 
6815 	if (is_guest_mode(vcpu))
6816 		return EXIT_FASTPATH_NONE;
6817 
6818 	exit_fastpath = vmx_exit_handlers_fastpath(vcpu);
6819 	if (exit_fastpath == EXIT_FASTPATH_REENTER_GUEST) {
6820 		if (!kvm_vcpu_exit_request(vcpu)) {
6821 			/*
6822 			 * FIXME: this goto should be a loop in vcpu_enter_guest,
6823 			 * but it would incur the cost of a retpoline for now.
6824 			 * Revisit once static calls are available.
6825 			 */
6826 			if (vcpu->arch.apicv_active)
6827 				vmx_sync_pir_to_irr(vcpu);
6828 			goto reenter_guest;
6829 		}
6830 		exit_fastpath = EXIT_FASTPATH_EXIT_HANDLED;
6831 	}
6832 
6833 	return exit_fastpath;
6834 }
6835 
6836 static void vmx_free_vcpu(struct kvm_vcpu *vcpu)
6837 {
6838 	struct vcpu_vmx *vmx = to_vmx(vcpu);
6839 
6840 	if (enable_pml)
6841 		vmx_destroy_pml_buffer(vmx);
6842 	free_vpid(vmx->vpid);
6843 	nested_vmx_free_vcpu(vcpu);
6844 	free_loaded_vmcs(vmx->loaded_vmcs);
6845 }
6846 
6847 static int vmx_create_vcpu(struct kvm_vcpu *vcpu)
6848 {
6849 	struct vcpu_vmx *vmx;
6850 	unsigned long *msr_bitmap;
6851 	int i, cpu, err;
6852 
6853 	BUILD_BUG_ON(offsetof(struct vcpu_vmx, vcpu) != 0);
6854 	vmx = to_vmx(vcpu);
6855 
6856 	err = -ENOMEM;
6857 
6858 	vmx->vpid = allocate_vpid();
6859 
6860 	/*
6861 	 * If PML is turned on, failure on enabling PML just results in failure
6862 	 * of creating the vcpu, therefore we can simplify PML logic (by
6863 	 * avoiding dealing with cases, such as enabling PML partially on vcpus
6864 	 * for the guest), etc.
6865 	 */
6866 	if (enable_pml) {
6867 		vmx->pml_pg = alloc_page(GFP_KERNEL_ACCOUNT | __GFP_ZERO);
6868 		if (!vmx->pml_pg)
6869 			goto free_vpid;
6870 	}
6871 
6872 	BUILD_BUG_ON(ARRAY_SIZE(vmx_msr_index) != NR_SHARED_MSRS);
6873 
6874 	for (i = 0; i < ARRAY_SIZE(vmx_msr_index); ++i) {
6875 		u32 index = vmx_msr_index[i];
6876 		u32 data_low, data_high;
6877 		int j = vmx->nmsrs;
6878 
6879 		if (rdmsr_safe(index, &data_low, &data_high) < 0)
6880 			continue;
6881 		if (wrmsr_safe(index, data_low, data_high) < 0)
6882 			continue;
6883 
6884 		vmx->guest_msrs[j].index = i;
6885 		vmx->guest_msrs[j].data = 0;
6886 		switch (index) {
6887 		case MSR_IA32_TSX_CTRL:
6888 			/*
6889 			 * No need to pass TSX_CTRL_CPUID_CLEAR through, so
6890 			 * let's avoid changing CPUID bits under the host
6891 			 * kernel's feet.
6892 			 */
6893 			vmx->guest_msrs[j].mask = ~(u64)TSX_CTRL_CPUID_CLEAR;
6894 			break;
6895 		default:
6896 			vmx->guest_msrs[j].mask = -1ull;
6897 			break;
6898 		}
6899 		++vmx->nmsrs;
6900 	}
6901 
6902 	err = alloc_loaded_vmcs(&vmx->vmcs01);
6903 	if (err < 0)
6904 		goto free_pml;
6905 
6906 	msr_bitmap = vmx->vmcs01.msr_bitmap;
6907 	vmx_disable_intercept_for_msr(msr_bitmap, MSR_IA32_TSC, MSR_TYPE_R);
6908 	vmx_disable_intercept_for_msr(msr_bitmap, MSR_FS_BASE, MSR_TYPE_RW);
6909 	vmx_disable_intercept_for_msr(msr_bitmap, MSR_GS_BASE, MSR_TYPE_RW);
6910 	vmx_disable_intercept_for_msr(msr_bitmap, MSR_KERNEL_GS_BASE, MSR_TYPE_RW);
6911 	vmx_disable_intercept_for_msr(msr_bitmap, MSR_IA32_SYSENTER_CS, MSR_TYPE_RW);
6912 	vmx_disable_intercept_for_msr(msr_bitmap, MSR_IA32_SYSENTER_ESP, MSR_TYPE_RW);
6913 	vmx_disable_intercept_for_msr(msr_bitmap, MSR_IA32_SYSENTER_EIP, MSR_TYPE_RW);
6914 	if (kvm_cstate_in_guest(vcpu->kvm)) {
6915 		vmx_disable_intercept_for_msr(msr_bitmap, MSR_CORE_C1_RES, MSR_TYPE_R);
6916 		vmx_disable_intercept_for_msr(msr_bitmap, MSR_CORE_C3_RESIDENCY, MSR_TYPE_R);
6917 		vmx_disable_intercept_for_msr(msr_bitmap, MSR_CORE_C6_RESIDENCY, MSR_TYPE_R);
6918 		vmx_disable_intercept_for_msr(msr_bitmap, MSR_CORE_C7_RESIDENCY, MSR_TYPE_R);
6919 	}
6920 	vmx->msr_bitmap_mode = 0;
6921 
6922 	vmx->loaded_vmcs = &vmx->vmcs01;
6923 	cpu = get_cpu();
6924 	vmx_vcpu_load(vcpu, cpu);
6925 	vcpu->cpu = cpu;
6926 	init_vmcs(vmx);
6927 	vmx_vcpu_put(vcpu);
6928 	put_cpu();
6929 	if (cpu_need_virtualize_apic_accesses(vcpu)) {
6930 		err = alloc_apic_access_page(vcpu->kvm);
6931 		if (err)
6932 			goto free_vmcs;
6933 	}
6934 
6935 	if (enable_ept && !enable_unrestricted_guest) {
6936 		err = init_rmode_identity_map(vcpu->kvm);
6937 		if (err)
6938 			goto free_vmcs;
6939 	}
6940 
6941 	if (nested)
6942 		nested_vmx_setup_ctls_msrs(&vmx->nested.msrs,
6943 					   vmx_capability.ept);
6944 	else
6945 		memset(&vmx->nested.msrs, 0, sizeof(vmx->nested.msrs));
6946 
6947 	vmx->nested.posted_intr_nv = -1;
6948 	vmx->nested.current_vmptr = -1ull;
6949 
6950 	vcpu->arch.microcode_version = 0x100000000ULL;
6951 	vmx->msr_ia32_feature_control_valid_bits = FEAT_CTL_LOCKED;
6952 
6953 	/*
6954 	 * Enforce invariant: pi_desc.nv is always either POSTED_INTR_VECTOR
6955 	 * or POSTED_INTR_WAKEUP_VECTOR.
6956 	 */
6957 	vmx->pi_desc.nv = POSTED_INTR_VECTOR;
6958 	vmx->pi_desc.sn = 1;
6959 
6960 	vmx->ept_pointer = INVALID_PAGE;
6961 
6962 	return 0;
6963 
6964 free_vmcs:
6965 	free_loaded_vmcs(vmx->loaded_vmcs);
6966 free_pml:
6967 	vmx_destroy_pml_buffer(vmx);
6968 free_vpid:
6969 	free_vpid(vmx->vpid);
6970 	return err;
6971 }
6972 
6973 #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"
6974 #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"
6975 
6976 static int vmx_vm_init(struct kvm *kvm)
6977 {
6978 	spin_lock_init(&to_kvm_vmx(kvm)->ept_pointer_lock);
6979 
6980 	if (!ple_gap)
6981 		kvm->arch.pause_in_guest = true;
6982 
6983 	if (boot_cpu_has(X86_BUG_L1TF) && enable_ept) {
6984 		switch (l1tf_mitigation) {
6985 		case L1TF_MITIGATION_OFF:
6986 		case L1TF_MITIGATION_FLUSH_NOWARN:
6987 			/* 'I explicitly don't care' is set */
6988 			break;
6989 		case L1TF_MITIGATION_FLUSH:
6990 		case L1TF_MITIGATION_FLUSH_NOSMT:
6991 		case L1TF_MITIGATION_FULL:
6992 			/*
6993 			 * Warn upon starting the first VM in a potentially
6994 			 * insecure environment.
6995 			 */
6996 			if (sched_smt_active())
6997 				pr_warn_once(L1TF_MSG_SMT);
6998 			if (l1tf_vmx_mitigation == VMENTER_L1D_FLUSH_NEVER)
6999 				pr_warn_once(L1TF_MSG_L1D);
7000 			break;
7001 		case L1TF_MITIGATION_FULL_FORCE:
7002 			/* Flush is enforced */
7003 			break;
7004 		}
7005 	}
7006 	kvm_apicv_init(kvm, enable_apicv);
7007 	return 0;
7008 }
7009 
7010 static int __init vmx_check_processor_compat(void)
7011 {
7012 	struct vmcs_config vmcs_conf;
7013 	struct vmx_capability vmx_cap;
7014 
7015 	if (!this_cpu_has(X86_FEATURE_MSR_IA32_FEAT_CTL) ||
7016 	    !this_cpu_has(X86_FEATURE_VMX)) {
7017 		pr_err("kvm: VMX is disabled on CPU %d\n", smp_processor_id());
7018 		return -EIO;
7019 	}
7020 
7021 	if (setup_vmcs_config(&vmcs_conf, &vmx_cap) < 0)
7022 		return -EIO;
7023 	if (nested)
7024 		nested_vmx_setup_ctls_msrs(&vmcs_conf.nested, vmx_cap.ept);
7025 	if (memcmp(&vmcs_config, &vmcs_conf, sizeof(struct vmcs_config)) != 0) {
7026 		printk(KERN_ERR "kvm: CPU %d feature inconsistency!\n",
7027 				smp_processor_id());
7028 		return -EIO;
7029 	}
7030 	return 0;
7031 }
7032 
7033 static u64 vmx_get_mt_mask(struct kvm_vcpu *vcpu, gfn_t gfn, bool is_mmio)
7034 {
7035 	u8 cache;
7036 	u64 ipat = 0;
7037 
7038 	/* We wanted to honor guest CD/MTRR/PAT, but doing so could result in
7039 	 * memory aliases with conflicting memory types and sometimes MCEs.
7040 	 * We have to be careful as to what are honored and when.
7041 	 *
7042 	 * For MMIO, guest CD/MTRR are ignored.  The EPT memory type is set to
7043 	 * UC.  The effective memory type is UC or WC depending on guest PAT.
7044 	 * This was historically the source of MCEs and we want to be
7045 	 * conservative.
7046 	 *
7047 	 * When there is no need to deal with noncoherent DMA (e.g., no VT-d
7048 	 * or VT-d has snoop control), guest CD/MTRR/PAT are all ignored.  The
7049 	 * EPT memory type is set to WB.  The effective memory type is forced
7050 	 * WB.
7051 	 *
7052 	 * Otherwise, we trust guest.  Guest CD/MTRR/PAT are all honored.  The
7053 	 * EPT memory type is used to emulate guest CD/MTRR.
7054 	 */
7055 
7056 	if (is_mmio) {
7057 		cache = MTRR_TYPE_UNCACHABLE;
7058 		goto exit;
7059 	}
7060 
7061 	if (!kvm_arch_has_noncoherent_dma(vcpu->kvm)) {
7062 		ipat = VMX_EPT_IPAT_BIT;
7063 		cache = MTRR_TYPE_WRBACK;
7064 		goto exit;
7065 	}
7066 
7067 	if (kvm_read_cr0(vcpu) & X86_CR0_CD) {
7068 		ipat = VMX_EPT_IPAT_BIT;
7069 		if (kvm_check_has_quirk(vcpu->kvm, KVM_X86_QUIRK_CD_NW_CLEARED))
7070 			cache = MTRR_TYPE_WRBACK;
7071 		else
7072 			cache = MTRR_TYPE_UNCACHABLE;
7073 		goto exit;
7074 	}
7075 
7076 	cache = kvm_mtrr_get_guest_memory_type(vcpu, gfn);
7077 
7078 exit:
7079 	return (cache << VMX_EPT_MT_EPTE_SHIFT) | ipat;
7080 }
7081 
7082 static void vmcs_set_secondary_exec_control(struct vcpu_vmx *vmx)
7083 {
7084 	/*
7085 	 * These bits in the secondary execution controls field
7086 	 * are dynamic, the others are mostly based on the hypervisor
7087 	 * architecture and the guest's CPUID.  Do not touch the
7088 	 * dynamic bits.
7089 	 */
7090 	u32 mask =
7091 		SECONDARY_EXEC_SHADOW_VMCS |
7092 		SECONDARY_EXEC_VIRTUALIZE_X2APIC_MODE |
7093 		SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES |
7094 		SECONDARY_EXEC_DESC;
7095 
7096 	u32 new_ctl = vmx->secondary_exec_control;
7097 	u32 cur_ctl = secondary_exec_controls_get(vmx);
7098 
7099 	secondary_exec_controls_set(vmx, (new_ctl & ~mask) | (cur_ctl & mask));
7100 }
7101 
7102 /*
7103  * Generate MSR_IA32_VMX_CR{0,4}_FIXED1 according to CPUID. Only set bits
7104  * (indicating "allowed-1") if they are supported in the guest's CPUID.
7105  */
7106 static void nested_vmx_cr_fixed1_bits_update(struct kvm_vcpu *vcpu)
7107 {
7108 	struct vcpu_vmx *vmx = to_vmx(vcpu);
7109 	struct kvm_cpuid_entry2 *entry;
7110 
7111 	vmx->nested.msrs.cr0_fixed1 = 0xffffffff;
7112 	vmx->nested.msrs.cr4_fixed1 = X86_CR4_PCE;
7113 
7114 #define cr4_fixed1_update(_cr4_mask, _reg, _cpuid_mask) do {		\
7115 	if (entry && (entry->_reg & (_cpuid_mask)))			\
7116 		vmx->nested.msrs.cr4_fixed1 |= (_cr4_mask);	\
7117 } while (0)
7118 
7119 	entry = kvm_find_cpuid_entry(vcpu, 0x1, 0);
7120 	cr4_fixed1_update(X86_CR4_VME,        edx, feature_bit(VME));
7121 	cr4_fixed1_update(X86_CR4_PVI,        edx, feature_bit(VME));
7122 	cr4_fixed1_update(X86_CR4_TSD,        edx, feature_bit(TSC));
7123 	cr4_fixed1_update(X86_CR4_DE,         edx, feature_bit(DE));
7124 	cr4_fixed1_update(X86_CR4_PSE,        edx, feature_bit(PSE));
7125 	cr4_fixed1_update(X86_CR4_PAE,        edx, feature_bit(PAE));
7126 	cr4_fixed1_update(X86_CR4_MCE,        edx, feature_bit(MCE));
7127 	cr4_fixed1_update(X86_CR4_PGE,        edx, feature_bit(PGE));
7128 	cr4_fixed1_update(X86_CR4_OSFXSR,     edx, feature_bit(FXSR));
7129 	cr4_fixed1_update(X86_CR4_OSXMMEXCPT, edx, feature_bit(XMM));
7130 	cr4_fixed1_update(X86_CR4_VMXE,       ecx, feature_bit(VMX));
7131 	cr4_fixed1_update(X86_CR4_SMXE,       ecx, feature_bit(SMX));
7132 	cr4_fixed1_update(X86_CR4_PCIDE,      ecx, feature_bit(PCID));
7133 	cr4_fixed1_update(X86_CR4_OSXSAVE,    ecx, feature_bit(XSAVE));
7134 
7135 	entry = kvm_find_cpuid_entry(vcpu, 0x7, 0);
7136 	cr4_fixed1_update(X86_CR4_FSGSBASE,   ebx, feature_bit(FSGSBASE));
7137 	cr4_fixed1_update(X86_CR4_SMEP,       ebx, feature_bit(SMEP));
7138 	cr4_fixed1_update(X86_CR4_SMAP,       ebx, feature_bit(SMAP));
7139 	cr4_fixed1_update(X86_CR4_PKE,        ecx, feature_bit(PKU));
7140 	cr4_fixed1_update(X86_CR4_UMIP,       ecx, feature_bit(UMIP));
7141 	cr4_fixed1_update(X86_CR4_LA57,       ecx, feature_bit(LA57));
7142 
7143 #undef cr4_fixed1_update
7144 }
7145 
7146 static void nested_vmx_entry_exit_ctls_update(struct kvm_vcpu *vcpu)
7147 {
7148 	struct vcpu_vmx *vmx = to_vmx(vcpu);
7149 
7150 	if (kvm_mpx_supported()) {
7151 		bool mpx_enabled = guest_cpuid_has(vcpu, X86_FEATURE_MPX);
7152 
7153 		if (mpx_enabled) {
7154 			vmx->nested.msrs.entry_ctls_high |= VM_ENTRY_LOAD_BNDCFGS;
7155 			vmx->nested.msrs.exit_ctls_high |= VM_EXIT_CLEAR_BNDCFGS;
7156 		} else {
7157 			vmx->nested.msrs.entry_ctls_high &= ~VM_ENTRY_LOAD_BNDCFGS;
7158 			vmx->nested.msrs.exit_ctls_high &= ~VM_EXIT_CLEAR_BNDCFGS;
7159 		}
7160 	}
7161 }
7162 
7163 static void update_intel_pt_cfg(struct kvm_vcpu *vcpu)
7164 {
7165 	struct vcpu_vmx *vmx = to_vmx(vcpu);
7166 	struct kvm_cpuid_entry2 *best = NULL;
7167 	int i;
7168 
7169 	for (i = 0; i < PT_CPUID_LEAVES; i++) {
7170 		best = kvm_find_cpuid_entry(vcpu, 0x14, i);
7171 		if (!best)
7172 			return;
7173 		vmx->pt_desc.caps[CPUID_EAX + i*PT_CPUID_REGS_NUM] = best->eax;
7174 		vmx->pt_desc.caps[CPUID_EBX + i*PT_CPUID_REGS_NUM] = best->ebx;
7175 		vmx->pt_desc.caps[CPUID_ECX + i*PT_CPUID_REGS_NUM] = best->ecx;
7176 		vmx->pt_desc.caps[CPUID_EDX + i*PT_CPUID_REGS_NUM] = best->edx;
7177 	}
7178 
7179 	/* Get the number of configurable Address Ranges for filtering */
7180 	vmx->pt_desc.addr_range = intel_pt_validate_cap(vmx->pt_desc.caps,
7181 						PT_CAP_num_address_ranges);
7182 
7183 	/* Initialize and clear the no dependency bits */
7184 	vmx->pt_desc.ctl_bitmask = ~(RTIT_CTL_TRACEEN | RTIT_CTL_OS |
7185 			RTIT_CTL_USR | RTIT_CTL_TSC_EN | RTIT_CTL_DISRETC);
7186 
7187 	/*
7188 	 * If CPUID.(EAX=14H,ECX=0):EBX[0]=1 CR3Filter can be set otherwise
7189 	 * will inject an #GP
7190 	 */
7191 	if (intel_pt_validate_cap(vmx->pt_desc.caps, PT_CAP_cr3_filtering))
7192 		vmx->pt_desc.ctl_bitmask &= ~RTIT_CTL_CR3EN;
7193 
7194 	/*
7195 	 * If CPUID.(EAX=14H,ECX=0):EBX[1]=1 CYCEn, CycThresh and
7196 	 * PSBFreq can be set
7197 	 */
7198 	if (intel_pt_validate_cap(vmx->pt_desc.caps, PT_CAP_psb_cyc))
7199 		vmx->pt_desc.ctl_bitmask &= ~(RTIT_CTL_CYCLEACC |
7200 				RTIT_CTL_CYC_THRESH | RTIT_CTL_PSB_FREQ);
7201 
7202 	/*
7203 	 * If CPUID.(EAX=14H,ECX=0):EBX[3]=1 MTCEn BranchEn and
7204 	 * MTCFreq can be set
7205 	 */
7206 	if (intel_pt_validate_cap(vmx->pt_desc.caps, PT_CAP_mtc))
7207 		vmx->pt_desc.ctl_bitmask &= ~(RTIT_CTL_MTC_EN |
7208 				RTIT_CTL_BRANCH_EN | RTIT_CTL_MTC_RANGE);
7209 
7210 	/* If CPUID.(EAX=14H,ECX=0):EBX[4]=1 FUPonPTW and PTWEn can be set */
7211 	if (intel_pt_validate_cap(vmx->pt_desc.caps, PT_CAP_ptwrite))
7212 		vmx->pt_desc.ctl_bitmask &= ~(RTIT_CTL_FUP_ON_PTW |
7213 							RTIT_CTL_PTW_EN);
7214 
7215 	/* If CPUID.(EAX=14H,ECX=0):EBX[5]=1 PwrEvEn can be set */
7216 	if (intel_pt_validate_cap(vmx->pt_desc.caps, PT_CAP_power_event_trace))
7217 		vmx->pt_desc.ctl_bitmask &= ~RTIT_CTL_PWR_EVT_EN;
7218 
7219 	/* If CPUID.(EAX=14H,ECX=0):ECX[0]=1 ToPA can be set */
7220 	if (intel_pt_validate_cap(vmx->pt_desc.caps, PT_CAP_topa_output))
7221 		vmx->pt_desc.ctl_bitmask &= ~RTIT_CTL_TOPA;
7222 
7223 	/* If CPUID.(EAX=14H,ECX=0):ECX[3]=1 FabircEn can be set */
7224 	if (intel_pt_validate_cap(vmx->pt_desc.caps, PT_CAP_output_subsys))
7225 		vmx->pt_desc.ctl_bitmask &= ~RTIT_CTL_FABRIC_EN;
7226 
7227 	/* unmask address range configure area */
7228 	for (i = 0; i < vmx->pt_desc.addr_range; i++)
7229 		vmx->pt_desc.ctl_bitmask &= ~(0xfULL << (32 + i * 4));
7230 }
7231 
7232 static void vmx_cpuid_update(struct kvm_vcpu *vcpu)
7233 {
7234 	struct vcpu_vmx *vmx = to_vmx(vcpu);
7235 
7236 	/* xsaves_enabled is recomputed in vmx_compute_secondary_exec_control(). */
7237 	vcpu->arch.xsaves_enabled = false;
7238 
7239 	if (cpu_has_secondary_exec_ctrls()) {
7240 		vmx_compute_secondary_exec_control(vmx);
7241 		vmcs_set_secondary_exec_control(vmx);
7242 	}
7243 
7244 	if (nested_vmx_allowed(vcpu))
7245 		to_vmx(vcpu)->msr_ia32_feature_control_valid_bits |=
7246 			FEAT_CTL_VMX_ENABLED_INSIDE_SMX |
7247 			FEAT_CTL_VMX_ENABLED_OUTSIDE_SMX;
7248 	else
7249 		to_vmx(vcpu)->msr_ia32_feature_control_valid_bits &=
7250 			~(FEAT_CTL_VMX_ENABLED_INSIDE_SMX |
7251 			  FEAT_CTL_VMX_ENABLED_OUTSIDE_SMX);
7252 
7253 	if (nested_vmx_allowed(vcpu)) {
7254 		nested_vmx_cr_fixed1_bits_update(vcpu);
7255 		nested_vmx_entry_exit_ctls_update(vcpu);
7256 	}
7257 
7258 	if (boot_cpu_has(X86_FEATURE_INTEL_PT) &&
7259 			guest_cpuid_has(vcpu, X86_FEATURE_INTEL_PT))
7260 		update_intel_pt_cfg(vcpu);
7261 
7262 	if (boot_cpu_has(X86_FEATURE_RTM)) {
7263 		struct shared_msr_entry *msr;
7264 		msr = find_msr_entry(vmx, MSR_IA32_TSX_CTRL);
7265 		if (msr) {
7266 			bool enabled = guest_cpuid_has(vcpu, X86_FEATURE_RTM);
7267 			vmx_set_guest_msr(vmx, msr, enabled ? 0 : TSX_CTRL_RTM_DISABLE);
7268 		}
7269 	}
7270 }
7271 
7272 static __init void vmx_set_cpu_caps(void)
7273 {
7274 	kvm_set_cpu_caps();
7275 
7276 	/* CPUID 0x1 */
7277 	if (nested)
7278 		kvm_cpu_cap_set(X86_FEATURE_VMX);
7279 
7280 	/* CPUID 0x7 */
7281 	if (kvm_mpx_supported())
7282 		kvm_cpu_cap_check_and_set(X86_FEATURE_MPX);
7283 	if (cpu_has_vmx_invpcid())
7284 		kvm_cpu_cap_check_and_set(X86_FEATURE_INVPCID);
7285 	if (vmx_pt_mode_is_host_guest())
7286 		kvm_cpu_cap_check_and_set(X86_FEATURE_INTEL_PT);
7287 
7288 	if (vmx_umip_emulated())
7289 		kvm_cpu_cap_set(X86_FEATURE_UMIP);
7290 
7291 	/* CPUID 0xD.1 */
7292 	supported_xss = 0;
7293 	if (!vmx_xsaves_supported())
7294 		kvm_cpu_cap_clear(X86_FEATURE_XSAVES);
7295 
7296 	/* CPUID 0x80000001 */
7297 	if (!cpu_has_vmx_rdtscp())
7298 		kvm_cpu_cap_clear(X86_FEATURE_RDTSCP);
7299 
7300 	if (vmx_waitpkg_supported())
7301 		kvm_cpu_cap_check_and_set(X86_FEATURE_WAITPKG);
7302 }
7303 
7304 static void vmx_request_immediate_exit(struct kvm_vcpu *vcpu)
7305 {
7306 	to_vmx(vcpu)->req_immediate_exit = true;
7307 }
7308 
7309 static int vmx_check_intercept_io(struct kvm_vcpu *vcpu,
7310 				  struct x86_instruction_info *info)
7311 {
7312 	struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
7313 	unsigned short port;
7314 	bool intercept;
7315 	int size;
7316 
7317 	if (info->intercept == x86_intercept_in ||
7318 	    info->intercept == x86_intercept_ins) {
7319 		port = info->src_val;
7320 		size = info->dst_bytes;
7321 	} else {
7322 		port = info->dst_val;
7323 		size = info->src_bytes;
7324 	}
7325 
7326 	/*
7327 	 * If the 'use IO bitmaps' VM-execution control is 0, IO instruction
7328 	 * VM-exits depend on the 'unconditional IO exiting' VM-execution
7329 	 * control.
7330 	 *
7331 	 * Otherwise, IO instruction VM-exits are controlled by the IO bitmaps.
7332 	 */
7333 	if (!nested_cpu_has(vmcs12, CPU_BASED_USE_IO_BITMAPS))
7334 		intercept = nested_cpu_has(vmcs12,
7335 					   CPU_BASED_UNCOND_IO_EXITING);
7336 	else
7337 		intercept = nested_vmx_check_io_bitmaps(vcpu, port, size);
7338 
7339 	/* FIXME: produce nested vmexit and return X86EMUL_INTERCEPTED.  */
7340 	return intercept ? X86EMUL_UNHANDLEABLE : X86EMUL_CONTINUE;
7341 }
7342 
7343 static int vmx_check_intercept(struct kvm_vcpu *vcpu,
7344 			       struct x86_instruction_info *info,
7345 			       enum x86_intercept_stage stage,
7346 			       struct x86_exception *exception)
7347 {
7348 	struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
7349 
7350 	switch (info->intercept) {
7351 	/*
7352 	 * RDPID causes #UD if disabled through secondary execution controls.
7353 	 * Because it is marked as EmulateOnUD, we need to intercept it here.
7354 	 */
7355 	case x86_intercept_rdtscp:
7356 		if (!nested_cpu_has2(vmcs12, SECONDARY_EXEC_RDTSCP)) {
7357 			exception->vector = UD_VECTOR;
7358 			exception->error_code_valid = false;
7359 			return X86EMUL_PROPAGATE_FAULT;
7360 		}
7361 		break;
7362 
7363 	case x86_intercept_in:
7364 	case x86_intercept_ins:
7365 	case x86_intercept_out:
7366 	case x86_intercept_outs:
7367 		return vmx_check_intercept_io(vcpu, info);
7368 
7369 	case x86_intercept_lgdt:
7370 	case x86_intercept_lidt:
7371 	case x86_intercept_lldt:
7372 	case x86_intercept_ltr:
7373 	case x86_intercept_sgdt:
7374 	case x86_intercept_sidt:
7375 	case x86_intercept_sldt:
7376 	case x86_intercept_str:
7377 		if (!nested_cpu_has2(vmcs12, SECONDARY_EXEC_DESC))
7378 			return X86EMUL_CONTINUE;
7379 
7380 		/* FIXME: produce nested vmexit and return X86EMUL_INTERCEPTED.  */
7381 		break;
7382 
7383 	/* TODO: check more intercepts... */
7384 	default:
7385 		break;
7386 	}
7387 
7388 	return X86EMUL_UNHANDLEABLE;
7389 }
7390 
7391 #ifdef CONFIG_X86_64
7392 /* (a << shift) / divisor, return 1 if overflow otherwise 0 */
7393 static inline int u64_shl_div_u64(u64 a, unsigned int shift,
7394 				  u64 divisor, u64 *result)
7395 {
7396 	u64 low = a << shift, high = a >> (64 - shift);
7397 
7398 	/* To avoid the overflow on divq */
7399 	if (high >= divisor)
7400 		return 1;
7401 
7402 	/* Low hold the result, high hold rem which is discarded */
7403 	asm("divq %2\n\t" : "=a" (low), "=d" (high) :
7404 	    "rm" (divisor), "0" (low), "1" (high));
7405 	*result = low;
7406 
7407 	return 0;
7408 }
7409 
7410 static int vmx_set_hv_timer(struct kvm_vcpu *vcpu, u64 guest_deadline_tsc,
7411 			    bool *expired)
7412 {
7413 	struct vcpu_vmx *vmx;
7414 	u64 tscl, guest_tscl, delta_tsc, lapic_timer_advance_cycles;
7415 	struct kvm_timer *ktimer = &vcpu->arch.apic->lapic_timer;
7416 
7417 	vmx = to_vmx(vcpu);
7418 	tscl = rdtsc();
7419 	guest_tscl = kvm_read_l1_tsc(vcpu, tscl);
7420 	delta_tsc = max(guest_deadline_tsc, guest_tscl) - guest_tscl;
7421 	lapic_timer_advance_cycles = nsec_to_cycles(vcpu,
7422 						    ktimer->timer_advance_ns);
7423 
7424 	if (delta_tsc > lapic_timer_advance_cycles)
7425 		delta_tsc -= lapic_timer_advance_cycles;
7426 	else
7427 		delta_tsc = 0;
7428 
7429 	/* Convert to host delta tsc if tsc scaling is enabled */
7430 	if (vcpu->arch.tsc_scaling_ratio != kvm_default_tsc_scaling_ratio &&
7431 	    delta_tsc && u64_shl_div_u64(delta_tsc,
7432 				kvm_tsc_scaling_ratio_frac_bits,
7433 				vcpu->arch.tsc_scaling_ratio, &delta_tsc))
7434 		return -ERANGE;
7435 
7436 	/*
7437 	 * If the delta tsc can't fit in the 32 bit after the multi shift,
7438 	 * we can't use the preemption timer.
7439 	 * It's possible that it fits on later vmentries, but checking
7440 	 * on every vmentry is costly so we just use an hrtimer.
7441 	 */
7442 	if (delta_tsc >> (cpu_preemption_timer_multi + 32))
7443 		return -ERANGE;
7444 
7445 	vmx->hv_deadline_tsc = tscl + delta_tsc;
7446 	*expired = !delta_tsc;
7447 	return 0;
7448 }
7449 
7450 static void vmx_cancel_hv_timer(struct kvm_vcpu *vcpu)
7451 {
7452 	to_vmx(vcpu)->hv_deadline_tsc = -1;
7453 }
7454 #endif
7455 
7456 static void vmx_sched_in(struct kvm_vcpu *vcpu, int cpu)
7457 {
7458 	if (!kvm_pause_in_guest(vcpu->kvm))
7459 		shrink_ple_window(vcpu);
7460 }
7461 
7462 static void vmx_slot_enable_log_dirty(struct kvm *kvm,
7463 				     struct kvm_memory_slot *slot)
7464 {
7465 	if (!kvm_dirty_log_manual_protect_and_init_set(kvm))
7466 		kvm_mmu_slot_leaf_clear_dirty(kvm, slot);
7467 	kvm_mmu_slot_largepage_remove_write_access(kvm, slot);
7468 }
7469 
7470 static void vmx_slot_disable_log_dirty(struct kvm *kvm,
7471 				       struct kvm_memory_slot *slot)
7472 {
7473 	kvm_mmu_slot_set_dirty(kvm, slot);
7474 }
7475 
7476 static void vmx_flush_log_dirty(struct kvm *kvm)
7477 {
7478 	kvm_flush_pml_buffers(kvm);
7479 }
7480 
7481 static int vmx_write_pml_buffer(struct kvm_vcpu *vcpu, gpa_t gpa)
7482 {
7483 	struct vmcs12 *vmcs12;
7484 	struct vcpu_vmx *vmx = to_vmx(vcpu);
7485 	gpa_t dst;
7486 
7487 	if (is_guest_mode(vcpu)) {
7488 		WARN_ON_ONCE(vmx->nested.pml_full);
7489 
7490 		/*
7491 		 * Check if PML is enabled for the nested guest.
7492 		 * Whether eptp bit 6 is set is already checked
7493 		 * as part of A/D emulation.
7494 		 */
7495 		vmcs12 = get_vmcs12(vcpu);
7496 		if (!nested_cpu_has_pml(vmcs12))
7497 			return 0;
7498 
7499 		if (vmcs12->guest_pml_index >= PML_ENTITY_NUM) {
7500 			vmx->nested.pml_full = true;
7501 			return 1;
7502 		}
7503 
7504 		gpa &= ~0xFFFull;
7505 		dst = vmcs12->pml_address + sizeof(u64) * vmcs12->guest_pml_index;
7506 
7507 		if (kvm_write_guest_page(vcpu->kvm, gpa_to_gfn(dst), &gpa,
7508 					 offset_in_page(dst), sizeof(gpa)))
7509 			return 0;
7510 
7511 		vmcs12->guest_pml_index--;
7512 	}
7513 
7514 	return 0;
7515 }
7516 
7517 static void vmx_enable_log_dirty_pt_masked(struct kvm *kvm,
7518 					   struct kvm_memory_slot *memslot,
7519 					   gfn_t offset, unsigned long mask)
7520 {
7521 	kvm_mmu_clear_dirty_pt_masked(kvm, memslot, offset, mask);
7522 }
7523 
7524 static void __pi_post_block(struct kvm_vcpu *vcpu)
7525 {
7526 	struct pi_desc *pi_desc = vcpu_to_pi_desc(vcpu);
7527 	struct pi_desc old, new;
7528 	unsigned int dest;
7529 
7530 	do {
7531 		old.control = new.control = pi_desc->control;
7532 		WARN(old.nv != POSTED_INTR_WAKEUP_VECTOR,
7533 		     "Wakeup handler not enabled while the VCPU is blocked\n");
7534 
7535 		dest = cpu_physical_id(vcpu->cpu);
7536 
7537 		if (x2apic_enabled())
7538 			new.ndst = dest;
7539 		else
7540 			new.ndst = (dest << 8) & 0xFF00;
7541 
7542 		/* set 'NV' to 'notification vector' */
7543 		new.nv = POSTED_INTR_VECTOR;
7544 	} while (cmpxchg64(&pi_desc->control, old.control,
7545 			   new.control) != old.control);
7546 
7547 	if (!WARN_ON_ONCE(vcpu->pre_pcpu == -1)) {
7548 		spin_lock(&per_cpu(blocked_vcpu_on_cpu_lock, vcpu->pre_pcpu));
7549 		list_del(&vcpu->blocked_vcpu_list);
7550 		spin_unlock(&per_cpu(blocked_vcpu_on_cpu_lock, vcpu->pre_pcpu));
7551 		vcpu->pre_pcpu = -1;
7552 	}
7553 }
7554 
7555 /*
7556  * This routine does the following things for vCPU which is going
7557  * to be blocked if VT-d PI is enabled.
7558  * - Store the vCPU to the wakeup list, so when interrupts happen
7559  *   we can find the right vCPU to wake up.
7560  * - Change the Posted-interrupt descriptor as below:
7561  *      'NDST' <-- vcpu->pre_pcpu
7562  *      'NV' <-- POSTED_INTR_WAKEUP_VECTOR
7563  * - If 'ON' is set during this process, which means at least one
7564  *   interrupt is posted for this vCPU, we cannot block it, in
7565  *   this case, return 1, otherwise, return 0.
7566  *
7567  */
7568 static int pi_pre_block(struct kvm_vcpu *vcpu)
7569 {
7570 	unsigned int dest;
7571 	struct pi_desc old, new;
7572 	struct pi_desc *pi_desc = vcpu_to_pi_desc(vcpu);
7573 
7574 	if (!kvm_arch_has_assigned_device(vcpu->kvm) ||
7575 		!irq_remapping_cap(IRQ_POSTING_CAP)  ||
7576 		!kvm_vcpu_apicv_active(vcpu))
7577 		return 0;
7578 
7579 	WARN_ON(irqs_disabled());
7580 	local_irq_disable();
7581 	if (!WARN_ON_ONCE(vcpu->pre_pcpu != -1)) {
7582 		vcpu->pre_pcpu = vcpu->cpu;
7583 		spin_lock(&per_cpu(blocked_vcpu_on_cpu_lock, vcpu->pre_pcpu));
7584 		list_add_tail(&vcpu->blocked_vcpu_list,
7585 			      &per_cpu(blocked_vcpu_on_cpu,
7586 				       vcpu->pre_pcpu));
7587 		spin_unlock(&per_cpu(blocked_vcpu_on_cpu_lock, vcpu->pre_pcpu));
7588 	}
7589 
7590 	do {
7591 		old.control = new.control = pi_desc->control;
7592 
7593 		WARN((pi_desc->sn == 1),
7594 		     "Warning: SN field of posted-interrupts "
7595 		     "is set before blocking\n");
7596 
7597 		/*
7598 		 * Since vCPU can be preempted during this process,
7599 		 * vcpu->cpu could be different with pre_pcpu, we
7600 		 * need to set pre_pcpu as the destination of wakeup
7601 		 * notification event, then we can find the right vCPU
7602 		 * to wakeup in wakeup handler if interrupts happen
7603 		 * when the vCPU is in blocked state.
7604 		 */
7605 		dest = cpu_physical_id(vcpu->pre_pcpu);
7606 
7607 		if (x2apic_enabled())
7608 			new.ndst = dest;
7609 		else
7610 			new.ndst = (dest << 8) & 0xFF00;
7611 
7612 		/* set 'NV' to 'wakeup vector' */
7613 		new.nv = POSTED_INTR_WAKEUP_VECTOR;
7614 	} while (cmpxchg64(&pi_desc->control, old.control,
7615 			   new.control) != old.control);
7616 
7617 	/* We should not block the vCPU if an interrupt is posted for it.  */
7618 	if (pi_test_on(pi_desc) == 1)
7619 		__pi_post_block(vcpu);
7620 
7621 	local_irq_enable();
7622 	return (vcpu->pre_pcpu == -1);
7623 }
7624 
7625 static int vmx_pre_block(struct kvm_vcpu *vcpu)
7626 {
7627 	if (pi_pre_block(vcpu))
7628 		return 1;
7629 
7630 	if (kvm_lapic_hv_timer_in_use(vcpu))
7631 		kvm_lapic_switch_to_sw_timer(vcpu);
7632 
7633 	return 0;
7634 }
7635 
7636 static void pi_post_block(struct kvm_vcpu *vcpu)
7637 {
7638 	if (vcpu->pre_pcpu == -1)
7639 		return;
7640 
7641 	WARN_ON(irqs_disabled());
7642 	local_irq_disable();
7643 	__pi_post_block(vcpu);
7644 	local_irq_enable();
7645 }
7646 
7647 static void vmx_post_block(struct kvm_vcpu *vcpu)
7648 {
7649 	if (kvm_x86_ops.set_hv_timer)
7650 		kvm_lapic_switch_to_hv_timer(vcpu);
7651 
7652 	pi_post_block(vcpu);
7653 }
7654 
7655 /*
7656  * vmx_update_pi_irte - set IRTE for Posted-Interrupts
7657  *
7658  * @kvm: kvm
7659  * @host_irq: host irq of the interrupt
7660  * @guest_irq: gsi of the interrupt
7661  * @set: set or unset PI
7662  * returns 0 on success, < 0 on failure
7663  */
7664 static int vmx_update_pi_irte(struct kvm *kvm, unsigned int host_irq,
7665 			      uint32_t guest_irq, bool set)
7666 {
7667 	struct kvm_kernel_irq_routing_entry *e;
7668 	struct kvm_irq_routing_table *irq_rt;
7669 	struct kvm_lapic_irq irq;
7670 	struct kvm_vcpu *vcpu;
7671 	struct vcpu_data vcpu_info;
7672 	int idx, ret = 0;
7673 
7674 	if (!kvm_arch_has_assigned_device(kvm) ||
7675 		!irq_remapping_cap(IRQ_POSTING_CAP) ||
7676 		!kvm_vcpu_apicv_active(kvm->vcpus[0]))
7677 		return 0;
7678 
7679 	idx = srcu_read_lock(&kvm->irq_srcu);
7680 	irq_rt = srcu_dereference(kvm->irq_routing, &kvm->irq_srcu);
7681 	if (guest_irq >= irq_rt->nr_rt_entries ||
7682 	    hlist_empty(&irq_rt->map[guest_irq])) {
7683 		pr_warn_once("no route for guest_irq %u/%u (broken user space?)\n",
7684 			     guest_irq, irq_rt->nr_rt_entries);
7685 		goto out;
7686 	}
7687 
7688 	hlist_for_each_entry(e, &irq_rt->map[guest_irq], link) {
7689 		if (e->type != KVM_IRQ_ROUTING_MSI)
7690 			continue;
7691 		/*
7692 		 * VT-d PI cannot support posting multicast/broadcast
7693 		 * interrupts to a vCPU, we still use interrupt remapping
7694 		 * for these kind of interrupts.
7695 		 *
7696 		 * For lowest-priority interrupts, we only support
7697 		 * those with single CPU as the destination, e.g. user
7698 		 * configures the interrupts via /proc/irq or uses
7699 		 * irqbalance to make the interrupts single-CPU.
7700 		 *
7701 		 * We will support full lowest-priority interrupt later.
7702 		 *
7703 		 * In addition, we can only inject generic interrupts using
7704 		 * the PI mechanism, refuse to route others through it.
7705 		 */
7706 
7707 		kvm_set_msi_irq(kvm, e, &irq);
7708 		if (!kvm_intr_is_single_vcpu(kvm, &irq, &vcpu) ||
7709 		    !kvm_irq_is_postable(&irq)) {
7710 			/*
7711 			 * Make sure the IRTE is in remapped mode if
7712 			 * we don't handle it in posted mode.
7713 			 */
7714 			ret = irq_set_vcpu_affinity(host_irq, NULL);
7715 			if (ret < 0) {
7716 				printk(KERN_INFO
7717 				   "failed to back to remapped mode, irq: %u\n",
7718 				   host_irq);
7719 				goto out;
7720 			}
7721 
7722 			continue;
7723 		}
7724 
7725 		vcpu_info.pi_desc_addr = __pa(vcpu_to_pi_desc(vcpu));
7726 		vcpu_info.vector = irq.vector;
7727 
7728 		trace_kvm_pi_irte_update(host_irq, vcpu->vcpu_id, e->gsi,
7729 				vcpu_info.vector, vcpu_info.pi_desc_addr, set);
7730 
7731 		if (set)
7732 			ret = irq_set_vcpu_affinity(host_irq, &vcpu_info);
7733 		else
7734 			ret = irq_set_vcpu_affinity(host_irq, NULL);
7735 
7736 		if (ret < 0) {
7737 			printk(KERN_INFO "%s: failed to update PI IRTE\n",
7738 					__func__);
7739 			goto out;
7740 		}
7741 	}
7742 
7743 	ret = 0;
7744 out:
7745 	srcu_read_unlock(&kvm->irq_srcu, idx);
7746 	return ret;
7747 }
7748 
7749 static void vmx_setup_mce(struct kvm_vcpu *vcpu)
7750 {
7751 	if (vcpu->arch.mcg_cap & MCG_LMCE_P)
7752 		to_vmx(vcpu)->msr_ia32_feature_control_valid_bits |=
7753 			FEAT_CTL_LMCE_ENABLED;
7754 	else
7755 		to_vmx(vcpu)->msr_ia32_feature_control_valid_bits &=
7756 			~FEAT_CTL_LMCE_ENABLED;
7757 }
7758 
7759 static int vmx_smi_allowed(struct kvm_vcpu *vcpu, bool for_injection)
7760 {
7761 	/* we need a nested vmexit to enter SMM, postpone if run is pending */
7762 	if (to_vmx(vcpu)->nested.nested_run_pending)
7763 		return -EBUSY;
7764 	return !is_smm(vcpu);
7765 }
7766 
7767 static int vmx_pre_enter_smm(struct kvm_vcpu *vcpu, char *smstate)
7768 {
7769 	struct vcpu_vmx *vmx = to_vmx(vcpu);
7770 
7771 	vmx->nested.smm.guest_mode = is_guest_mode(vcpu);
7772 	if (vmx->nested.smm.guest_mode)
7773 		nested_vmx_vmexit(vcpu, -1, 0, 0);
7774 
7775 	vmx->nested.smm.vmxon = vmx->nested.vmxon;
7776 	vmx->nested.vmxon = false;
7777 	vmx_clear_hlt(vcpu);
7778 	return 0;
7779 }
7780 
7781 static int vmx_pre_leave_smm(struct kvm_vcpu *vcpu, const char *smstate)
7782 {
7783 	struct vcpu_vmx *vmx = to_vmx(vcpu);
7784 	int ret;
7785 
7786 	if (vmx->nested.smm.vmxon) {
7787 		vmx->nested.vmxon = true;
7788 		vmx->nested.smm.vmxon = false;
7789 	}
7790 
7791 	if (vmx->nested.smm.guest_mode) {
7792 		ret = nested_vmx_enter_non_root_mode(vcpu, false);
7793 		if (ret)
7794 			return ret;
7795 
7796 		vmx->nested.smm.guest_mode = false;
7797 	}
7798 	return 0;
7799 }
7800 
7801 static void enable_smi_window(struct kvm_vcpu *vcpu)
7802 {
7803 	/* RSM will cause a vmexit anyway.  */
7804 }
7805 
7806 static bool vmx_need_emulation_on_page_fault(struct kvm_vcpu *vcpu)
7807 {
7808 	return false;
7809 }
7810 
7811 static bool vmx_apic_init_signal_blocked(struct kvm_vcpu *vcpu)
7812 {
7813 	return to_vmx(vcpu)->nested.vmxon;
7814 }
7815 
7816 static void vmx_migrate_timers(struct kvm_vcpu *vcpu)
7817 {
7818 	if (is_guest_mode(vcpu)) {
7819 		struct hrtimer *timer = &to_vmx(vcpu)->nested.preemption_timer;
7820 
7821 		if (hrtimer_try_to_cancel(timer) == 1)
7822 			hrtimer_start_expires(timer, HRTIMER_MODE_ABS_PINNED);
7823 	}
7824 }
7825 
7826 static void hardware_unsetup(void)
7827 {
7828 	if (nested)
7829 		nested_vmx_hardware_unsetup();
7830 
7831 	free_kvm_area();
7832 }
7833 
7834 static bool vmx_check_apicv_inhibit_reasons(ulong bit)
7835 {
7836 	ulong supported = BIT(APICV_INHIBIT_REASON_DISABLE) |
7837 			  BIT(APICV_INHIBIT_REASON_HYPERV);
7838 
7839 	return supported & BIT(bit);
7840 }
7841 
7842 static struct kvm_x86_ops vmx_x86_ops __initdata = {
7843 	.hardware_unsetup = hardware_unsetup,
7844 
7845 	.hardware_enable = hardware_enable,
7846 	.hardware_disable = hardware_disable,
7847 	.cpu_has_accelerated_tpr = report_flexpriority,
7848 	.has_emulated_msr = vmx_has_emulated_msr,
7849 
7850 	.vm_size = sizeof(struct kvm_vmx),
7851 	.vm_init = vmx_vm_init,
7852 
7853 	.vcpu_create = vmx_create_vcpu,
7854 	.vcpu_free = vmx_free_vcpu,
7855 	.vcpu_reset = vmx_vcpu_reset,
7856 
7857 	.prepare_guest_switch = vmx_prepare_switch_to_guest,
7858 	.vcpu_load = vmx_vcpu_load,
7859 	.vcpu_put = vmx_vcpu_put,
7860 
7861 	.update_bp_intercept = update_exception_bitmap,
7862 	.get_msr_feature = vmx_get_msr_feature,
7863 	.get_msr = vmx_get_msr,
7864 	.set_msr = vmx_set_msr,
7865 	.get_segment_base = vmx_get_segment_base,
7866 	.get_segment = vmx_get_segment,
7867 	.set_segment = vmx_set_segment,
7868 	.get_cpl = vmx_get_cpl,
7869 	.get_cs_db_l_bits = vmx_get_cs_db_l_bits,
7870 	.set_cr0 = vmx_set_cr0,
7871 	.set_cr4 = vmx_set_cr4,
7872 	.set_efer = vmx_set_efer,
7873 	.get_idt = vmx_get_idt,
7874 	.set_idt = vmx_set_idt,
7875 	.get_gdt = vmx_get_gdt,
7876 	.set_gdt = vmx_set_gdt,
7877 	.set_dr7 = vmx_set_dr7,
7878 	.sync_dirty_debug_regs = vmx_sync_dirty_debug_regs,
7879 	.cache_reg = vmx_cache_reg,
7880 	.get_rflags = vmx_get_rflags,
7881 	.set_rflags = vmx_set_rflags,
7882 
7883 	.tlb_flush_all = vmx_flush_tlb_all,
7884 	.tlb_flush_current = vmx_flush_tlb_current,
7885 	.tlb_flush_gva = vmx_flush_tlb_gva,
7886 	.tlb_flush_guest = vmx_flush_tlb_guest,
7887 
7888 	.run = vmx_vcpu_run,
7889 	.handle_exit = vmx_handle_exit,
7890 	.skip_emulated_instruction = vmx_skip_emulated_instruction,
7891 	.update_emulated_instruction = vmx_update_emulated_instruction,
7892 	.set_interrupt_shadow = vmx_set_interrupt_shadow,
7893 	.get_interrupt_shadow = vmx_get_interrupt_shadow,
7894 	.patch_hypercall = vmx_patch_hypercall,
7895 	.set_irq = vmx_inject_irq,
7896 	.set_nmi = vmx_inject_nmi,
7897 	.queue_exception = vmx_queue_exception,
7898 	.cancel_injection = vmx_cancel_injection,
7899 	.interrupt_allowed = vmx_interrupt_allowed,
7900 	.nmi_allowed = vmx_nmi_allowed,
7901 	.get_nmi_mask = vmx_get_nmi_mask,
7902 	.set_nmi_mask = vmx_set_nmi_mask,
7903 	.enable_nmi_window = enable_nmi_window,
7904 	.enable_irq_window = enable_irq_window,
7905 	.update_cr8_intercept = update_cr8_intercept,
7906 	.set_virtual_apic_mode = vmx_set_virtual_apic_mode,
7907 	.set_apic_access_page_addr = vmx_set_apic_access_page_addr,
7908 	.refresh_apicv_exec_ctrl = vmx_refresh_apicv_exec_ctrl,
7909 	.load_eoi_exitmap = vmx_load_eoi_exitmap,
7910 	.apicv_post_state_restore = vmx_apicv_post_state_restore,
7911 	.check_apicv_inhibit_reasons = vmx_check_apicv_inhibit_reasons,
7912 	.hwapic_irr_update = vmx_hwapic_irr_update,
7913 	.hwapic_isr_update = vmx_hwapic_isr_update,
7914 	.guest_apic_has_interrupt = vmx_guest_apic_has_interrupt,
7915 	.sync_pir_to_irr = vmx_sync_pir_to_irr,
7916 	.deliver_posted_interrupt = vmx_deliver_posted_interrupt,
7917 	.dy_apicv_has_pending_interrupt = vmx_dy_apicv_has_pending_interrupt,
7918 
7919 	.set_tss_addr = vmx_set_tss_addr,
7920 	.set_identity_map_addr = vmx_set_identity_map_addr,
7921 	.get_tdp_level = vmx_get_tdp_level,
7922 	.get_mt_mask = vmx_get_mt_mask,
7923 
7924 	.get_exit_info = vmx_get_exit_info,
7925 
7926 	.cpuid_update = vmx_cpuid_update,
7927 
7928 	.has_wbinvd_exit = cpu_has_vmx_wbinvd_exit,
7929 
7930 	.write_l1_tsc_offset = vmx_write_l1_tsc_offset,
7931 
7932 	.load_mmu_pgd = vmx_load_mmu_pgd,
7933 
7934 	.check_intercept = vmx_check_intercept,
7935 	.handle_exit_irqoff = vmx_handle_exit_irqoff,
7936 
7937 	.request_immediate_exit = vmx_request_immediate_exit,
7938 
7939 	.sched_in = vmx_sched_in,
7940 
7941 	.slot_enable_log_dirty = vmx_slot_enable_log_dirty,
7942 	.slot_disable_log_dirty = vmx_slot_disable_log_dirty,
7943 	.flush_log_dirty = vmx_flush_log_dirty,
7944 	.enable_log_dirty_pt_masked = vmx_enable_log_dirty_pt_masked,
7945 	.write_log_dirty = vmx_write_pml_buffer,
7946 
7947 	.pre_block = vmx_pre_block,
7948 	.post_block = vmx_post_block,
7949 
7950 	.pmu_ops = &intel_pmu_ops,
7951 	.nested_ops = &vmx_nested_ops,
7952 
7953 	.update_pi_irte = vmx_update_pi_irte,
7954 
7955 #ifdef CONFIG_X86_64
7956 	.set_hv_timer = vmx_set_hv_timer,
7957 	.cancel_hv_timer = vmx_cancel_hv_timer,
7958 #endif
7959 
7960 	.setup_mce = vmx_setup_mce,
7961 
7962 	.smi_allowed = vmx_smi_allowed,
7963 	.pre_enter_smm = vmx_pre_enter_smm,
7964 	.pre_leave_smm = vmx_pre_leave_smm,
7965 	.enable_smi_window = enable_smi_window,
7966 
7967 	.need_emulation_on_page_fault = vmx_need_emulation_on_page_fault,
7968 	.apic_init_signal_blocked = vmx_apic_init_signal_blocked,
7969 	.migrate_timers = vmx_migrate_timers,
7970 };
7971 
7972 static __init int hardware_setup(void)
7973 {
7974 	unsigned long host_bndcfgs;
7975 	struct desc_ptr dt;
7976 	int r, i, ept_lpage_level;
7977 
7978 	store_idt(&dt);
7979 	host_idt_base = dt.address;
7980 
7981 	for (i = 0; i < ARRAY_SIZE(vmx_msr_index); ++i)
7982 		kvm_define_shared_msr(i, vmx_msr_index[i]);
7983 
7984 	if (setup_vmcs_config(&vmcs_config, &vmx_capability) < 0)
7985 		return -EIO;
7986 
7987 	if (boot_cpu_has(X86_FEATURE_NX))
7988 		kvm_enable_efer_bits(EFER_NX);
7989 
7990 	if (boot_cpu_has(X86_FEATURE_MPX)) {
7991 		rdmsrl(MSR_IA32_BNDCFGS, host_bndcfgs);
7992 		WARN_ONCE(host_bndcfgs, "KVM: BNDCFGS in host will be lost");
7993 	}
7994 
7995 	if (!cpu_has_vmx_mpx())
7996 		supported_xcr0 &= ~(XFEATURE_MASK_BNDREGS |
7997 				    XFEATURE_MASK_BNDCSR);
7998 
7999 	if (!cpu_has_vmx_vpid() || !cpu_has_vmx_invvpid() ||
8000 	    !(cpu_has_vmx_invvpid_single() || cpu_has_vmx_invvpid_global()))
8001 		enable_vpid = 0;
8002 
8003 	if (!cpu_has_vmx_ept() ||
8004 	    !cpu_has_vmx_ept_4levels() ||
8005 	    !cpu_has_vmx_ept_mt_wb() ||
8006 	    !cpu_has_vmx_invept_global())
8007 		enable_ept = 0;
8008 
8009 	if (!cpu_has_vmx_ept_ad_bits() || !enable_ept)
8010 		enable_ept_ad_bits = 0;
8011 
8012 	if (!cpu_has_vmx_unrestricted_guest() || !enable_ept)
8013 		enable_unrestricted_guest = 0;
8014 
8015 	if (!cpu_has_vmx_flexpriority())
8016 		flexpriority_enabled = 0;
8017 
8018 	if (!cpu_has_virtual_nmis())
8019 		enable_vnmi = 0;
8020 
8021 	/*
8022 	 * set_apic_access_page_addr() is used to reload apic access
8023 	 * page upon invalidation.  No need to do anything if not
8024 	 * using the APIC_ACCESS_ADDR VMCS field.
8025 	 */
8026 	if (!flexpriority_enabled)
8027 		vmx_x86_ops.set_apic_access_page_addr = NULL;
8028 
8029 	if (!cpu_has_vmx_tpr_shadow())
8030 		vmx_x86_ops.update_cr8_intercept = NULL;
8031 
8032 #if IS_ENABLED(CONFIG_HYPERV)
8033 	if (ms_hyperv.nested_features & HV_X64_NESTED_GUEST_MAPPING_FLUSH
8034 	    && enable_ept) {
8035 		vmx_x86_ops.tlb_remote_flush = hv_remote_flush_tlb;
8036 		vmx_x86_ops.tlb_remote_flush_with_range =
8037 				hv_remote_flush_tlb_with_range;
8038 	}
8039 #endif
8040 
8041 	if (!cpu_has_vmx_ple()) {
8042 		ple_gap = 0;
8043 		ple_window = 0;
8044 		ple_window_grow = 0;
8045 		ple_window_max = 0;
8046 		ple_window_shrink = 0;
8047 	}
8048 
8049 	if (!cpu_has_vmx_apicv()) {
8050 		enable_apicv = 0;
8051 		vmx_x86_ops.sync_pir_to_irr = NULL;
8052 	}
8053 
8054 	if (cpu_has_vmx_tsc_scaling()) {
8055 		kvm_has_tsc_control = true;
8056 		kvm_max_tsc_scaling_ratio = KVM_VMX_TSC_MULTIPLIER_MAX;
8057 		kvm_tsc_scaling_ratio_frac_bits = 48;
8058 	}
8059 
8060 	set_bit(0, vmx_vpid_bitmap); /* 0 is reserved for host */
8061 
8062 	if (enable_ept)
8063 		vmx_enable_tdp();
8064 
8065 	if (!enable_ept)
8066 		ept_lpage_level = 0;
8067 	else if (cpu_has_vmx_ept_1g_page())
8068 		ept_lpage_level = PG_LEVEL_1G;
8069 	else if (cpu_has_vmx_ept_2m_page())
8070 		ept_lpage_level = PG_LEVEL_2M;
8071 	else
8072 		ept_lpage_level = PG_LEVEL_4K;
8073 	kvm_configure_mmu(enable_ept, ept_lpage_level);
8074 
8075 	/*
8076 	 * Only enable PML when hardware supports PML feature, and both EPT
8077 	 * and EPT A/D bit features are enabled -- PML depends on them to work.
8078 	 */
8079 	if (!enable_ept || !enable_ept_ad_bits || !cpu_has_vmx_pml())
8080 		enable_pml = 0;
8081 
8082 	if (!enable_pml) {
8083 		vmx_x86_ops.slot_enable_log_dirty = NULL;
8084 		vmx_x86_ops.slot_disable_log_dirty = NULL;
8085 		vmx_x86_ops.flush_log_dirty = NULL;
8086 		vmx_x86_ops.enable_log_dirty_pt_masked = NULL;
8087 	}
8088 
8089 	if (!cpu_has_vmx_preemption_timer())
8090 		enable_preemption_timer = false;
8091 
8092 	if (enable_preemption_timer) {
8093 		u64 use_timer_freq = 5000ULL * 1000 * 1000;
8094 		u64 vmx_msr;
8095 
8096 		rdmsrl(MSR_IA32_VMX_MISC, vmx_msr);
8097 		cpu_preemption_timer_multi =
8098 			vmx_msr & VMX_MISC_PREEMPTION_TIMER_RATE_MASK;
8099 
8100 		if (tsc_khz)
8101 			use_timer_freq = (u64)tsc_khz * 1000;
8102 		use_timer_freq >>= cpu_preemption_timer_multi;
8103 
8104 		/*
8105 		 * KVM "disables" the preemption timer by setting it to its max
8106 		 * value.  Don't use the timer if it might cause spurious exits
8107 		 * at a rate faster than 0.1 Hz (of uninterrupted guest time).
8108 		 */
8109 		if (use_timer_freq > 0xffffffffu / 10)
8110 			enable_preemption_timer = false;
8111 	}
8112 
8113 	if (!enable_preemption_timer) {
8114 		vmx_x86_ops.set_hv_timer = NULL;
8115 		vmx_x86_ops.cancel_hv_timer = NULL;
8116 		vmx_x86_ops.request_immediate_exit = __kvm_request_immediate_exit;
8117 	}
8118 
8119 	kvm_set_posted_intr_wakeup_handler(wakeup_handler);
8120 
8121 	kvm_mce_cap_supported |= MCG_LMCE_P;
8122 
8123 	if (pt_mode != PT_MODE_SYSTEM && pt_mode != PT_MODE_HOST_GUEST)
8124 		return -EINVAL;
8125 	if (!enable_ept || !cpu_has_vmx_intel_pt())
8126 		pt_mode = PT_MODE_SYSTEM;
8127 
8128 	if (nested) {
8129 		nested_vmx_setup_ctls_msrs(&vmcs_config.nested,
8130 					   vmx_capability.ept);
8131 
8132 		r = nested_vmx_hardware_setup(kvm_vmx_exit_handlers);
8133 		if (r)
8134 			return r;
8135 	}
8136 
8137 	vmx_set_cpu_caps();
8138 
8139 	r = alloc_kvm_area();
8140 	if (r)
8141 		nested_vmx_hardware_unsetup();
8142 	return r;
8143 }
8144 
8145 static struct kvm_x86_init_ops vmx_init_ops __initdata = {
8146 	.cpu_has_kvm_support = cpu_has_kvm_support,
8147 	.disabled_by_bios = vmx_disabled_by_bios,
8148 	.check_processor_compatibility = vmx_check_processor_compat,
8149 	.hardware_setup = hardware_setup,
8150 
8151 	.runtime_ops = &vmx_x86_ops,
8152 };
8153 
8154 static void vmx_cleanup_l1d_flush(void)
8155 {
8156 	if (vmx_l1d_flush_pages) {
8157 		free_pages((unsigned long)vmx_l1d_flush_pages, L1D_CACHE_ORDER);
8158 		vmx_l1d_flush_pages = NULL;
8159 	}
8160 	/* Restore state so sysfs ignores VMX */
8161 	l1tf_vmx_mitigation = VMENTER_L1D_FLUSH_AUTO;
8162 }
8163 
8164 static void vmx_exit(void)
8165 {
8166 #ifdef CONFIG_KEXEC_CORE
8167 	RCU_INIT_POINTER(crash_vmclear_loaded_vmcss, NULL);
8168 	synchronize_rcu();
8169 #endif
8170 
8171 	kvm_exit();
8172 
8173 #if IS_ENABLED(CONFIG_HYPERV)
8174 	if (static_branch_unlikely(&enable_evmcs)) {
8175 		int cpu;
8176 		struct hv_vp_assist_page *vp_ap;
8177 		/*
8178 		 * Reset everything to support using non-enlightened VMCS
8179 		 * access later (e.g. when we reload the module with
8180 		 * enlightened_vmcs=0)
8181 		 */
8182 		for_each_online_cpu(cpu) {
8183 			vp_ap =	hv_get_vp_assist_page(cpu);
8184 
8185 			if (!vp_ap)
8186 				continue;
8187 
8188 			vp_ap->nested_control.features.directhypercall = 0;
8189 			vp_ap->current_nested_vmcs = 0;
8190 			vp_ap->enlighten_vmentry = 0;
8191 		}
8192 
8193 		static_branch_disable(&enable_evmcs);
8194 	}
8195 #endif
8196 	vmx_cleanup_l1d_flush();
8197 }
8198 module_exit(vmx_exit);
8199 
8200 static int __init vmx_init(void)
8201 {
8202 	int r, cpu;
8203 
8204 #if IS_ENABLED(CONFIG_HYPERV)
8205 	/*
8206 	 * Enlightened VMCS usage should be recommended and the host needs
8207 	 * to support eVMCS v1 or above. We can also disable eVMCS support
8208 	 * with module parameter.
8209 	 */
8210 	if (enlightened_vmcs &&
8211 	    ms_hyperv.hints & HV_X64_ENLIGHTENED_VMCS_RECOMMENDED &&
8212 	    (ms_hyperv.nested_features & HV_X64_ENLIGHTENED_VMCS_VERSION) >=
8213 	    KVM_EVMCS_VERSION) {
8214 		int cpu;
8215 
8216 		/* Check that we have assist pages on all online CPUs */
8217 		for_each_online_cpu(cpu) {
8218 			if (!hv_get_vp_assist_page(cpu)) {
8219 				enlightened_vmcs = false;
8220 				break;
8221 			}
8222 		}
8223 
8224 		if (enlightened_vmcs) {
8225 			pr_info("KVM: vmx: using Hyper-V Enlightened VMCS\n");
8226 			static_branch_enable(&enable_evmcs);
8227 		}
8228 
8229 		if (ms_hyperv.nested_features & HV_X64_NESTED_DIRECT_FLUSH)
8230 			vmx_x86_ops.enable_direct_tlbflush
8231 				= hv_enable_direct_tlbflush;
8232 
8233 	} else {
8234 		enlightened_vmcs = false;
8235 	}
8236 #endif
8237 
8238 	r = kvm_init(&vmx_init_ops, sizeof(struct vcpu_vmx),
8239 		     __alignof__(struct vcpu_vmx), THIS_MODULE);
8240 	if (r)
8241 		return r;
8242 
8243 	/*
8244 	 * Must be called after kvm_init() so enable_ept is properly set
8245 	 * up. Hand the parameter mitigation value in which was stored in
8246 	 * the pre module init parser. If no parameter was given, it will
8247 	 * contain 'auto' which will be turned into the default 'cond'
8248 	 * mitigation mode.
8249 	 */
8250 	r = vmx_setup_l1d_flush(vmentry_l1d_flush_param);
8251 	if (r) {
8252 		vmx_exit();
8253 		return r;
8254 	}
8255 
8256 	for_each_possible_cpu(cpu) {
8257 		INIT_LIST_HEAD(&per_cpu(loaded_vmcss_on_cpu, cpu));
8258 		INIT_LIST_HEAD(&per_cpu(blocked_vcpu_on_cpu, cpu));
8259 		spin_lock_init(&per_cpu(blocked_vcpu_on_cpu_lock, cpu));
8260 	}
8261 
8262 #ifdef CONFIG_KEXEC_CORE
8263 	rcu_assign_pointer(crash_vmclear_loaded_vmcss,
8264 			   crash_vmclear_local_loaded_vmcss);
8265 #endif
8266 	vmx_check_vmcs12_offsets();
8267 
8268 	return 0;
8269 }
8270 module_init(vmx_init);
8271