xref: /openbmc/linux/arch/x86/kvm/vmx/vmx.c (revision 88021836)
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 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
16 
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/objtool.h>
26 #include <linux/sched.h>
27 #include <linux/sched/smt.h>
28 #include <linux/slab.h>
29 #include <linux/tboot.h>
30 #include <linux/trace_events.h>
31 #include <linux/entry-kvm.h>
32 
33 #include <asm/apic.h>
34 #include <asm/asm.h>
35 #include <asm/cpu.h>
36 #include <asm/cpu_device_id.h>
37 #include <asm/debugreg.h>
38 #include <asm/desc.h>
39 #include <asm/fpu/api.h>
40 #include <asm/fpu/xstate.h>
41 #include <asm/idtentry.h>
42 #include <asm/io.h>
43 #include <asm/irq_remapping.h>
44 #include <asm/kexec.h>
45 #include <asm/perf_event.h>
46 #include <asm/mmu_context.h>
47 #include <asm/mshyperv.h>
48 #include <asm/mwait.h>
49 #include <asm/spec-ctrl.h>
50 #include <asm/virtext.h>
51 #include <asm/vmx.h>
52 
53 #include "capabilities.h"
54 #include "cpuid.h"
55 #include "hyperv.h"
56 #include "kvm_onhyperv.h"
57 #include "irq.h"
58 #include "kvm_cache_regs.h"
59 #include "lapic.h"
60 #include "mmu.h"
61 #include "nested.h"
62 #include "pmu.h"
63 #include "sgx.h"
64 #include "trace.h"
65 #include "vmcs.h"
66 #include "vmcs12.h"
67 #include "vmx.h"
68 #include "x86.h"
69 #include "smm.h"
70 
71 MODULE_AUTHOR("Qumranet");
72 MODULE_LICENSE("GPL");
73 
74 #ifdef MODULE
75 static const struct x86_cpu_id vmx_cpu_id[] = {
76 	X86_MATCH_FEATURE(X86_FEATURE_VMX, NULL),
77 	{}
78 };
79 MODULE_DEVICE_TABLE(x86cpu, vmx_cpu_id);
80 #endif
81 
82 bool __read_mostly enable_vpid = 1;
83 module_param_named(vpid, enable_vpid, bool, 0444);
84 
85 static bool __read_mostly enable_vnmi = 1;
86 module_param_named(vnmi, enable_vnmi, bool, S_IRUGO);
87 
88 bool __read_mostly flexpriority_enabled = 1;
89 module_param_named(flexpriority, flexpriority_enabled, bool, S_IRUGO);
90 
91 bool __read_mostly enable_ept = 1;
92 module_param_named(ept, enable_ept, bool, S_IRUGO);
93 
94 bool __read_mostly enable_unrestricted_guest = 1;
95 module_param_named(unrestricted_guest,
96 			enable_unrestricted_guest, bool, S_IRUGO);
97 
98 bool __read_mostly enable_ept_ad_bits = 1;
99 module_param_named(eptad, enable_ept_ad_bits, bool, S_IRUGO);
100 
101 static bool __read_mostly emulate_invalid_guest_state = true;
102 module_param(emulate_invalid_guest_state, bool, S_IRUGO);
103 
104 static bool __read_mostly fasteoi = 1;
105 module_param(fasteoi, bool, S_IRUGO);
106 
107 module_param(enable_apicv, bool, S_IRUGO);
108 
109 bool __read_mostly enable_ipiv = true;
110 module_param(enable_ipiv, bool, 0444);
111 
112 /*
113  * If nested=1, nested virtualization is supported, i.e., guests may use
114  * VMX and be a hypervisor for its own guests. If nested=0, guests may not
115  * use VMX instructions.
116  */
117 static bool __read_mostly nested = 1;
118 module_param(nested, bool, S_IRUGO);
119 
120 bool __read_mostly enable_pml = 1;
121 module_param_named(pml, enable_pml, bool, S_IRUGO);
122 
123 static bool __read_mostly error_on_inconsistent_vmcs_config = true;
124 module_param(error_on_inconsistent_vmcs_config, bool, 0444);
125 
126 static bool __read_mostly dump_invalid_vmcs = 0;
127 module_param(dump_invalid_vmcs, bool, 0644);
128 
129 #define MSR_BITMAP_MODE_X2APIC		1
130 #define MSR_BITMAP_MODE_X2APIC_APICV	2
131 
132 #define KVM_VMX_TSC_MULTIPLIER_MAX     0xffffffffffffffffULL
133 
134 /* Guest_tsc -> host_tsc conversion requires 64-bit division.  */
135 static int __read_mostly cpu_preemption_timer_multi;
136 static bool __read_mostly enable_preemption_timer = 1;
137 #ifdef CONFIG_X86_64
138 module_param_named(preemption_timer, enable_preemption_timer, bool, S_IRUGO);
139 #endif
140 
141 extern bool __read_mostly allow_smaller_maxphyaddr;
142 module_param(allow_smaller_maxphyaddr, bool, S_IRUGO);
143 
144 #define KVM_VM_CR0_ALWAYS_OFF (X86_CR0_NW | X86_CR0_CD)
145 #define KVM_VM_CR0_ALWAYS_ON_UNRESTRICTED_GUEST X86_CR0_NE
146 #define KVM_VM_CR0_ALWAYS_ON				\
147 	(KVM_VM_CR0_ALWAYS_ON_UNRESTRICTED_GUEST | X86_CR0_PG | X86_CR0_PE)
148 
149 #define KVM_VM_CR4_ALWAYS_ON_UNRESTRICTED_GUEST X86_CR4_VMXE
150 #define KVM_PMODE_VM_CR4_ALWAYS_ON (X86_CR4_PAE | X86_CR4_VMXE)
151 #define KVM_RMODE_VM_CR4_ALWAYS_ON (X86_CR4_VME | X86_CR4_PAE | X86_CR4_VMXE)
152 
153 #define RMODE_GUEST_OWNED_EFLAGS_BITS (~(X86_EFLAGS_IOPL | X86_EFLAGS_VM))
154 
155 #define MSR_IA32_RTIT_STATUS_MASK (~(RTIT_STATUS_FILTEREN | \
156 	RTIT_STATUS_CONTEXTEN | RTIT_STATUS_TRIGGEREN | \
157 	RTIT_STATUS_ERROR | RTIT_STATUS_STOPPED | \
158 	RTIT_STATUS_BYTECNT))
159 
160 /*
161  * List of MSRs that can be directly passed to the guest.
162  * In addition to these x2apic and PT MSRs are handled specially.
163  */
164 static u32 vmx_possible_passthrough_msrs[MAX_POSSIBLE_PASSTHROUGH_MSRS] = {
165 	MSR_IA32_SPEC_CTRL,
166 	MSR_IA32_PRED_CMD,
167 	MSR_IA32_FLUSH_CMD,
168 	MSR_IA32_TSC,
169 #ifdef CONFIG_X86_64
170 	MSR_FS_BASE,
171 	MSR_GS_BASE,
172 	MSR_KERNEL_GS_BASE,
173 	MSR_IA32_XFD,
174 	MSR_IA32_XFD_ERR,
175 #endif
176 	MSR_IA32_SYSENTER_CS,
177 	MSR_IA32_SYSENTER_ESP,
178 	MSR_IA32_SYSENTER_EIP,
179 	MSR_CORE_C1_RES,
180 	MSR_CORE_C3_RESIDENCY,
181 	MSR_CORE_C6_RESIDENCY,
182 	MSR_CORE_C7_RESIDENCY,
183 };
184 
185 /*
186  * These 2 parameters are used to config the controls for Pause-Loop Exiting:
187  * ple_gap:    upper bound on the amount of time between two successive
188  *             executions of PAUSE in a loop. Also indicate if ple enabled.
189  *             According to test, this time is usually smaller than 128 cycles.
190  * ple_window: upper bound on the amount of time a guest is allowed to execute
191  *             in a PAUSE loop. Tests indicate that most spinlocks are held for
192  *             less than 2^12 cycles
193  * Time is measured based on a counter that runs at the same rate as the TSC,
194  * refer SDM volume 3b section 21.6.13 & 22.1.3.
195  */
196 static unsigned int ple_gap = KVM_DEFAULT_PLE_GAP;
197 module_param(ple_gap, uint, 0444);
198 
199 static unsigned int ple_window = KVM_VMX_DEFAULT_PLE_WINDOW;
200 module_param(ple_window, uint, 0444);
201 
202 /* Default doubles per-vcpu window every exit. */
203 static unsigned int ple_window_grow = KVM_DEFAULT_PLE_WINDOW_GROW;
204 module_param(ple_window_grow, uint, 0444);
205 
206 /* Default resets per-vcpu window every exit to ple_window. */
207 static unsigned int ple_window_shrink = KVM_DEFAULT_PLE_WINDOW_SHRINK;
208 module_param(ple_window_shrink, uint, 0444);
209 
210 /* Default is to compute the maximum so we can never overflow. */
211 static unsigned int ple_window_max        = KVM_VMX_DEFAULT_PLE_WINDOW_MAX;
212 module_param(ple_window_max, uint, 0444);
213 
214 /* Default is SYSTEM mode, 1 for host-guest mode */
215 int __read_mostly pt_mode = PT_MODE_SYSTEM;
216 module_param(pt_mode, int, S_IRUGO);
217 
218 static DEFINE_STATIC_KEY_FALSE(vmx_l1d_should_flush);
219 static DEFINE_STATIC_KEY_FALSE(vmx_l1d_flush_cond);
220 static DEFINE_MUTEX(vmx_l1d_flush_mutex);
221 
222 /* Storage for pre module init parameter parsing */
223 static enum vmx_l1d_flush_state __read_mostly vmentry_l1d_flush_param = VMENTER_L1D_FLUSH_AUTO;
224 
225 static const struct {
226 	const char *option;
227 	bool for_parse;
228 } vmentry_l1d_param[] = {
229 	[VMENTER_L1D_FLUSH_AUTO]	 = {"auto", true},
230 	[VMENTER_L1D_FLUSH_NEVER]	 = {"never", true},
231 	[VMENTER_L1D_FLUSH_COND]	 = {"cond", true},
232 	[VMENTER_L1D_FLUSH_ALWAYS]	 = {"always", true},
233 	[VMENTER_L1D_FLUSH_EPT_DISABLED] = {"EPT disabled", false},
234 	[VMENTER_L1D_FLUSH_NOT_REQUIRED] = {"not required", false},
235 };
236 
237 #define L1D_CACHE_ORDER 4
238 static void *vmx_l1d_flush_pages;
239 
240 /* Control for disabling CPU Fill buffer clear */
241 static bool __read_mostly vmx_fb_clear_ctrl_available;
242 
243 static int vmx_setup_l1d_flush(enum vmx_l1d_flush_state l1tf)
244 {
245 	struct page *page;
246 	unsigned int i;
247 
248 	if (!boot_cpu_has_bug(X86_BUG_L1TF)) {
249 		l1tf_vmx_mitigation = VMENTER_L1D_FLUSH_NOT_REQUIRED;
250 		return 0;
251 	}
252 
253 	if (!enable_ept) {
254 		l1tf_vmx_mitigation = VMENTER_L1D_FLUSH_EPT_DISABLED;
255 		return 0;
256 	}
257 
258 	if (boot_cpu_has(X86_FEATURE_ARCH_CAPABILITIES)) {
259 		u64 msr;
260 
261 		rdmsrl(MSR_IA32_ARCH_CAPABILITIES, msr);
262 		if (msr & ARCH_CAP_SKIP_VMENTRY_L1DFLUSH) {
263 			l1tf_vmx_mitigation = VMENTER_L1D_FLUSH_NOT_REQUIRED;
264 			return 0;
265 		}
266 	}
267 
268 	/* If set to auto use the default l1tf mitigation method */
269 	if (l1tf == VMENTER_L1D_FLUSH_AUTO) {
270 		switch (l1tf_mitigation) {
271 		case L1TF_MITIGATION_OFF:
272 			l1tf = VMENTER_L1D_FLUSH_NEVER;
273 			break;
274 		case L1TF_MITIGATION_FLUSH_NOWARN:
275 		case L1TF_MITIGATION_FLUSH:
276 		case L1TF_MITIGATION_FLUSH_NOSMT:
277 			l1tf = VMENTER_L1D_FLUSH_COND;
278 			break;
279 		case L1TF_MITIGATION_FULL:
280 		case L1TF_MITIGATION_FULL_FORCE:
281 			l1tf = VMENTER_L1D_FLUSH_ALWAYS;
282 			break;
283 		}
284 	} else if (l1tf_mitigation == L1TF_MITIGATION_FULL_FORCE) {
285 		l1tf = VMENTER_L1D_FLUSH_ALWAYS;
286 	}
287 
288 	if (l1tf != VMENTER_L1D_FLUSH_NEVER && !vmx_l1d_flush_pages &&
289 	    !boot_cpu_has(X86_FEATURE_FLUSH_L1D)) {
290 		/*
291 		 * This allocation for vmx_l1d_flush_pages is not tied to a VM
292 		 * lifetime and so should not be charged to a memcg.
293 		 */
294 		page = alloc_pages(GFP_KERNEL, L1D_CACHE_ORDER);
295 		if (!page)
296 			return -ENOMEM;
297 		vmx_l1d_flush_pages = page_address(page);
298 
299 		/*
300 		 * Initialize each page with a different pattern in
301 		 * order to protect against KSM in the nested
302 		 * virtualization case.
303 		 */
304 		for (i = 0; i < 1u << L1D_CACHE_ORDER; ++i) {
305 			memset(vmx_l1d_flush_pages + i * PAGE_SIZE, i + 1,
306 			       PAGE_SIZE);
307 		}
308 	}
309 
310 	l1tf_vmx_mitigation = l1tf;
311 
312 	if (l1tf != VMENTER_L1D_FLUSH_NEVER)
313 		static_branch_enable(&vmx_l1d_should_flush);
314 	else
315 		static_branch_disable(&vmx_l1d_should_flush);
316 
317 	if (l1tf == VMENTER_L1D_FLUSH_COND)
318 		static_branch_enable(&vmx_l1d_flush_cond);
319 	else
320 		static_branch_disable(&vmx_l1d_flush_cond);
321 	return 0;
322 }
323 
324 static int vmentry_l1d_flush_parse(const char *s)
325 {
326 	unsigned int i;
327 
328 	if (s) {
329 		for (i = 0; i < ARRAY_SIZE(vmentry_l1d_param); i++) {
330 			if (vmentry_l1d_param[i].for_parse &&
331 			    sysfs_streq(s, vmentry_l1d_param[i].option))
332 				return i;
333 		}
334 	}
335 	return -EINVAL;
336 }
337 
338 static int vmentry_l1d_flush_set(const char *s, const struct kernel_param *kp)
339 {
340 	int l1tf, ret;
341 
342 	l1tf = vmentry_l1d_flush_parse(s);
343 	if (l1tf < 0)
344 		return l1tf;
345 
346 	if (!boot_cpu_has(X86_BUG_L1TF))
347 		return 0;
348 
349 	/*
350 	 * Has vmx_init() run already? If not then this is the pre init
351 	 * parameter parsing. In that case just store the value and let
352 	 * vmx_init() do the proper setup after enable_ept has been
353 	 * established.
354 	 */
355 	if (l1tf_vmx_mitigation == VMENTER_L1D_FLUSH_AUTO) {
356 		vmentry_l1d_flush_param = l1tf;
357 		return 0;
358 	}
359 
360 	mutex_lock(&vmx_l1d_flush_mutex);
361 	ret = vmx_setup_l1d_flush(l1tf);
362 	mutex_unlock(&vmx_l1d_flush_mutex);
363 	return ret;
364 }
365 
366 static int vmentry_l1d_flush_get(char *s, const struct kernel_param *kp)
367 {
368 	if (WARN_ON_ONCE(l1tf_vmx_mitigation >= ARRAY_SIZE(vmentry_l1d_param)))
369 		return sprintf(s, "???\n");
370 
371 	return sprintf(s, "%s\n", vmentry_l1d_param[l1tf_vmx_mitigation].option);
372 }
373 
374 static void vmx_setup_fb_clear_ctrl(void)
375 {
376 	u64 msr;
377 
378 	if (boot_cpu_has(X86_FEATURE_ARCH_CAPABILITIES) &&
379 	    !boot_cpu_has_bug(X86_BUG_MDS) &&
380 	    !boot_cpu_has_bug(X86_BUG_TAA)) {
381 		rdmsrl(MSR_IA32_ARCH_CAPABILITIES, msr);
382 		if (msr & ARCH_CAP_FB_CLEAR_CTRL)
383 			vmx_fb_clear_ctrl_available = true;
384 	}
385 }
386 
387 static __always_inline void vmx_disable_fb_clear(struct vcpu_vmx *vmx)
388 {
389 	u64 msr;
390 
391 	if (!vmx->disable_fb_clear)
392 		return;
393 
394 	msr = __rdmsr(MSR_IA32_MCU_OPT_CTRL);
395 	msr |= FB_CLEAR_DIS;
396 	native_wrmsrl(MSR_IA32_MCU_OPT_CTRL, msr);
397 	/* Cache the MSR value to avoid reading it later */
398 	vmx->msr_ia32_mcu_opt_ctrl = msr;
399 }
400 
401 static __always_inline void vmx_enable_fb_clear(struct vcpu_vmx *vmx)
402 {
403 	if (!vmx->disable_fb_clear)
404 		return;
405 
406 	vmx->msr_ia32_mcu_opt_ctrl &= ~FB_CLEAR_DIS;
407 	native_wrmsrl(MSR_IA32_MCU_OPT_CTRL, vmx->msr_ia32_mcu_opt_ctrl);
408 }
409 
410 static void vmx_update_fb_clear_dis(struct kvm_vcpu *vcpu, struct vcpu_vmx *vmx)
411 {
412 	vmx->disable_fb_clear = vmx_fb_clear_ctrl_available;
413 
414 	/*
415 	 * If guest will not execute VERW, there is no need to set FB_CLEAR_DIS
416 	 * at VMEntry. Skip the MSR read/write when a guest has no use case to
417 	 * execute VERW.
418 	 */
419 	if ((vcpu->arch.arch_capabilities & ARCH_CAP_FB_CLEAR) ||
420 	   ((vcpu->arch.arch_capabilities & ARCH_CAP_MDS_NO) &&
421 	    (vcpu->arch.arch_capabilities & ARCH_CAP_TAA_NO) &&
422 	    (vcpu->arch.arch_capabilities & ARCH_CAP_PSDP_NO) &&
423 	    (vcpu->arch.arch_capabilities & ARCH_CAP_FBSDP_NO) &&
424 	    (vcpu->arch.arch_capabilities & ARCH_CAP_SBDR_SSDP_NO)))
425 		vmx->disable_fb_clear = false;
426 }
427 
428 static const struct kernel_param_ops vmentry_l1d_flush_ops = {
429 	.set = vmentry_l1d_flush_set,
430 	.get = vmentry_l1d_flush_get,
431 };
432 module_param_cb(vmentry_l1d_flush, &vmentry_l1d_flush_ops, NULL, 0644);
433 
434 static u32 vmx_segment_access_rights(struct kvm_segment *var);
435 
436 void vmx_vmexit(void);
437 
438 #define vmx_insn_failed(fmt...)		\
439 do {					\
440 	WARN_ONCE(1, fmt);		\
441 	pr_warn_ratelimited(fmt);	\
442 } while (0)
443 
444 noinline void vmread_error(unsigned long field)
445 {
446 	vmx_insn_failed("vmread failed: field=%lx\n", field);
447 }
448 
449 #ifndef CONFIG_CC_HAS_ASM_GOTO_OUTPUT
450 noinstr void vmread_error_trampoline2(unsigned long field, bool fault)
451 {
452 	if (fault) {
453 		kvm_spurious_fault();
454 	} else {
455 		instrumentation_begin();
456 		vmread_error(field);
457 		instrumentation_end();
458 	}
459 }
460 #endif
461 
462 noinline void vmwrite_error(unsigned long field, unsigned long value)
463 {
464 	vmx_insn_failed("vmwrite failed: field=%lx val=%lx err=%u\n",
465 			field, value, vmcs_read32(VM_INSTRUCTION_ERROR));
466 }
467 
468 noinline void vmclear_error(struct vmcs *vmcs, u64 phys_addr)
469 {
470 	vmx_insn_failed("vmclear failed: %p/%llx err=%u\n",
471 			vmcs, phys_addr, vmcs_read32(VM_INSTRUCTION_ERROR));
472 }
473 
474 noinline void vmptrld_error(struct vmcs *vmcs, u64 phys_addr)
475 {
476 	vmx_insn_failed("vmptrld failed: %p/%llx err=%u\n",
477 			vmcs, phys_addr, vmcs_read32(VM_INSTRUCTION_ERROR));
478 }
479 
480 noinline void invvpid_error(unsigned long ext, u16 vpid, gva_t gva)
481 {
482 	vmx_insn_failed("invvpid failed: ext=0x%lx vpid=%u gva=0x%lx\n",
483 			ext, vpid, gva);
484 }
485 
486 noinline void invept_error(unsigned long ext, u64 eptp, gpa_t gpa)
487 {
488 	vmx_insn_failed("invept failed: ext=0x%lx eptp=%llx gpa=0x%llx\n",
489 			ext, eptp, gpa);
490 }
491 
492 static DEFINE_PER_CPU(struct vmcs *, vmxarea);
493 DEFINE_PER_CPU(struct vmcs *, current_vmcs);
494 /*
495  * We maintain a per-CPU linked-list of VMCS loaded on that CPU. This is needed
496  * when a CPU is brought down, and we need to VMCLEAR all VMCSs loaded on it.
497  */
498 static DEFINE_PER_CPU(struct list_head, loaded_vmcss_on_cpu);
499 
500 static DECLARE_BITMAP(vmx_vpid_bitmap, VMX_NR_VPIDS);
501 static DEFINE_SPINLOCK(vmx_vpid_lock);
502 
503 struct vmcs_config vmcs_config __ro_after_init;
504 struct vmx_capability vmx_capability __ro_after_init;
505 
506 #define VMX_SEGMENT_FIELD(seg)					\
507 	[VCPU_SREG_##seg] = {                                   \
508 		.selector = GUEST_##seg##_SELECTOR,		\
509 		.base = GUEST_##seg##_BASE,		   	\
510 		.limit = GUEST_##seg##_LIMIT,		   	\
511 		.ar_bytes = GUEST_##seg##_AR_BYTES,	   	\
512 	}
513 
514 static const struct kvm_vmx_segment_field {
515 	unsigned selector;
516 	unsigned base;
517 	unsigned limit;
518 	unsigned ar_bytes;
519 } kvm_vmx_segment_fields[] = {
520 	VMX_SEGMENT_FIELD(CS),
521 	VMX_SEGMENT_FIELD(DS),
522 	VMX_SEGMENT_FIELD(ES),
523 	VMX_SEGMENT_FIELD(FS),
524 	VMX_SEGMENT_FIELD(GS),
525 	VMX_SEGMENT_FIELD(SS),
526 	VMX_SEGMENT_FIELD(TR),
527 	VMX_SEGMENT_FIELD(LDTR),
528 };
529 
530 static inline void vmx_segment_cache_clear(struct vcpu_vmx *vmx)
531 {
532 	vmx->segment_cache.bitmask = 0;
533 }
534 
535 static unsigned long host_idt_base;
536 
537 #if IS_ENABLED(CONFIG_HYPERV)
538 static struct kvm_x86_ops vmx_x86_ops __initdata;
539 
540 static bool __read_mostly enlightened_vmcs = true;
541 module_param(enlightened_vmcs, bool, 0444);
542 
543 static int hv_enable_l2_tlb_flush(struct kvm_vcpu *vcpu)
544 {
545 	struct hv_enlightened_vmcs *evmcs;
546 	struct hv_partition_assist_pg **p_hv_pa_pg =
547 			&to_kvm_hv(vcpu->kvm)->hv_pa_pg;
548 	/*
549 	 * Synthetic VM-Exit is not enabled in current code and so All
550 	 * evmcs in singe VM shares same assist page.
551 	 */
552 	if (!*p_hv_pa_pg)
553 		*p_hv_pa_pg = kzalloc(PAGE_SIZE, GFP_KERNEL_ACCOUNT);
554 
555 	if (!*p_hv_pa_pg)
556 		return -ENOMEM;
557 
558 	evmcs = (struct hv_enlightened_vmcs *)to_vmx(vcpu)->loaded_vmcs->vmcs;
559 
560 	evmcs->partition_assist_page =
561 		__pa(*p_hv_pa_pg);
562 	evmcs->hv_vm_id = (unsigned long)vcpu->kvm;
563 	evmcs->hv_enlightenments_control.nested_flush_hypercall = 1;
564 
565 	return 0;
566 }
567 
568 static __init void hv_init_evmcs(void)
569 {
570 	int cpu;
571 
572 	if (!enlightened_vmcs)
573 		return;
574 
575 	/*
576 	 * Enlightened VMCS usage should be recommended and the host needs
577 	 * to support eVMCS v1 or above.
578 	 */
579 	if (ms_hyperv.hints & HV_X64_ENLIGHTENED_VMCS_RECOMMENDED &&
580 	    (ms_hyperv.nested_features & HV_X64_ENLIGHTENED_VMCS_VERSION) >=
581 	     KVM_EVMCS_VERSION) {
582 
583 		/* Check that we have assist pages on all online CPUs */
584 		for_each_online_cpu(cpu) {
585 			if (!hv_get_vp_assist_page(cpu)) {
586 				enlightened_vmcs = false;
587 				break;
588 			}
589 		}
590 
591 		if (enlightened_vmcs) {
592 			pr_info("Using Hyper-V Enlightened VMCS\n");
593 			static_branch_enable(&__kvm_is_using_evmcs);
594 		}
595 
596 		if (ms_hyperv.nested_features & HV_X64_NESTED_DIRECT_FLUSH)
597 			vmx_x86_ops.enable_l2_tlb_flush
598 				= hv_enable_l2_tlb_flush;
599 
600 	} else {
601 		enlightened_vmcs = false;
602 	}
603 }
604 
605 static void hv_reset_evmcs(void)
606 {
607 	struct hv_vp_assist_page *vp_ap;
608 
609 	if (!kvm_is_using_evmcs())
610 		return;
611 
612 	/*
613 	 * KVM should enable eVMCS if and only if all CPUs have a VP assist
614 	 * page, and should reject CPU onlining if eVMCS is enabled the CPU
615 	 * doesn't have a VP assist page allocated.
616 	 */
617 	vp_ap = hv_get_vp_assist_page(smp_processor_id());
618 	if (WARN_ON_ONCE(!vp_ap))
619 		return;
620 
621 	/*
622 	 * Reset everything to support using non-enlightened VMCS access later
623 	 * (e.g. when we reload the module with enlightened_vmcs=0)
624 	 */
625 	vp_ap->nested_control.features.directhypercall = 0;
626 	vp_ap->current_nested_vmcs = 0;
627 	vp_ap->enlighten_vmentry = 0;
628 }
629 
630 #else /* IS_ENABLED(CONFIG_HYPERV) */
631 static void hv_init_evmcs(void) {}
632 static void hv_reset_evmcs(void) {}
633 #endif /* IS_ENABLED(CONFIG_HYPERV) */
634 
635 /*
636  * Comment's format: document - errata name - stepping - processor name.
637  * Refer from
638  * https://www.virtualbox.org/svn/vbox/trunk/src/VBox/VMM/VMMR0/HMR0.cpp
639  */
640 static u32 vmx_preemption_cpu_tfms[] = {
641 /* 323344.pdf - BA86   - D0 - Xeon 7500 Series */
642 0x000206E6,
643 /* 323056.pdf - AAX65  - C2 - Xeon L3406 */
644 /* 322814.pdf - AAT59  - C2 - i7-600, i5-500, i5-400 and i3-300 Mobile */
645 /* 322911.pdf - AAU65  - C2 - i5-600, i3-500 Desktop and Pentium G6950 */
646 0x00020652,
647 /* 322911.pdf - AAU65  - K0 - i5-600, i3-500 Desktop and Pentium G6950 */
648 0x00020655,
649 /* 322373.pdf - AAO95  - B1 - Xeon 3400 Series */
650 /* 322166.pdf - AAN92  - B1 - i7-800 and i5-700 Desktop */
651 /*
652  * 320767.pdf - AAP86  - B1 -
653  * i7-900 Mobile Extreme, i7-800 and i7-700 Mobile
654  */
655 0x000106E5,
656 /* 321333.pdf - AAM126 - C0 - Xeon 3500 */
657 0x000106A0,
658 /* 321333.pdf - AAM126 - C1 - Xeon 3500 */
659 0x000106A1,
660 /* 320836.pdf - AAJ124 - C0 - i7-900 Desktop Extreme and i7-900 Desktop */
661 0x000106A4,
662  /* 321333.pdf - AAM126 - D0 - Xeon 3500 */
663  /* 321324.pdf - AAK139 - D0 - Xeon 5500 */
664  /* 320836.pdf - AAJ124 - D0 - i7-900 Extreme and i7-900 Desktop */
665 0x000106A5,
666  /* Xeon E3-1220 V2 */
667 0x000306A8,
668 };
669 
670 static inline bool cpu_has_broken_vmx_preemption_timer(void)
671 {
672 	u32 eax = cpuid_eax(0x00000001), i;
673 
674 	/* Clear the reserved bits */
675 	eax &= ~(0x3U << 14 | 0xfU << 28);
676 	for (i = 0; i < ARRAY_SIZE(vmx_preemption_cpu_tfms); i++)
677 		if (eax == vmx_preemption_cpu_tfms[i])
678 			return true;
679 
680 	return false;
681 }
682 
683 static inline bool cpu_need_virtualize_apic_accesses(struct kvm_vcpu *vcpu)
684 {
685 	return flexpriority_enabled && lapic_in_kernel(vcpu);
686 }
687 
688 static int possible_passthrough_msr_slot(u32 msr)
689 {
690 	u32 i;
691 
692 	for (i = 0; i < ARRAY_SIZE(vmx_possible_passthrough_msrs); i++)
693 		if (vmx_possible_passthrough_msrs[i] == msr)
694 			return i;
695 
696 	return -ENOENT;
697 }
698 
699 static bool is_valid_passthrough_msr(u32 msr)
700 {
701 	bool r;
702 
703 	switch (msr) {
704 	case 0x800 ... 0x8ff:
705 		/* x2APIC MSRs. These are handled in vmx_update_msr_bitmap_x2apic() */
706 		return true;
707 	case MSR_IA32_RTIT_STATUS:
708 	case MSR_IA32_RTIT_OUTPUT_BASE:
709 	case MSR_IA32_RTIT_OUTPUT_MASK:
710 	case MSR_IA32_RTIT_CR3_MATCH:
711 	case MSR_IA32_RTIT_ADDR0_A ... MSR_IA32_RTIT_ADDR3_B:
712 		/* PT MSRs. These are handled in pt_update_intercept_for_msr() */
713 	case MSR_LBR_SELECT:
714 	case MSR_LBR_TOS:
715 	case MSR_LBR_INFO_0 ... MSR_LBR_INFO_0 + 31:
716 	case MSR_LBR_NHM_FROM ... MSR_LBR_NHM_FROM + 31:
717 	case MSR_LBR_NHM_TO ... MSR_LBR_NHM_TO + 31:
718 	case MSR_LBR_CORE_FROM ... MSR_LBR_CORE_FROM + 8:
719 	case MSR_LBR_CORE_TO ... MSR_LBR_CORE_TO + 8:
720 		/* LBR MSRs. These are handled in vmx_update_intercept_for_lbr_msrs() */
721 		return true;
722 	}
723 
724 	r = possible_passthrough_msr_slot(msr) != -ENOENT;
725 
726 	WARN(!r, "Invalid MSR %x, please adapt vmx_possible_passthrough_msrs[]", msr);
727 
728 	return r;
729 }
730 
731 struct vmx_uret_msr *vmx_find_uret_msr(struct vcpu_vmx *vmx, u32 msr)
732 {
733 	int i;
734 
735 	i = kvm_find_user_return_msr(msr);
736 	if (i >= 0)
737 		return &vmx->guest_uret_msrs[i];
738 	return NULL;
739 }
740 
741 static int vmx_set_guest_uret_msr(struct vcpu_vmx *vmx,
742 				  struct vmx_uret_msr *msr, u64 data)
743 {
744 	unsigned int slot = msr - vmx->guest_uret_msrs;
745 	int ret = 0;
746 
747 	if (msr->load_into_hardware) {
748 		preempt_disable();
749 		ret = kvm_set_user_return_msr(slot, data, msr->mask);
750 		preempt_enable();
751 	}
752 	if (!ret)
753 		msr->data = data;
754 	return ret;
755 }
756 
757 #ifdef CONFIG_KEXEC_CORE
758 static void crash_vmclear_local_loaded_vmcss(void)
759 {
760 	int cpu = raw_smp_processor_id();
761 	struct loaded_vmcs *v;
762 
763 	list_for_each_entry(v, &per_cpu(loaded_vmcss_on_cpu, cpu),
764 			    loaded_vmcss_on_cpu_link)
765 		vmcs_clear(v->vmcs);
766 }
767 #endif /* CONFIG_KEXEC_CORE */
768 
769 static void __loaded_vmcs_clear(void *arg)
770 {
771 	struct loaded_vmcs *loaded_vmcs = arg;
772 	int cpu = raw_smp_processor_id();
773 
774 	if (loaded_vmcs->cpu != cpu)
775 		return; /* vcpu migration can race with cpu offline */
776 	if (per_cpu(current_vmcs, cpu) == loaded_vmcs->vmcs)
777 		per_cpu(current_vmcs, cpu) = NULL;
778 
779 	vmcs_clear(loaded_vmcs->vmcs);
780 	if (loaded_vmcs->shadow_vmcs && loaded_vmcs->launched)
781 		vmcs_clear(loaded_vmcs->shadow_vmcs);
782 
783 	list_del(&loaded_vmcs->loaded_vmcss_on_cpu_link);
784 
785 	/*
786 	 * Ensure all writes to loaded_vmcs, including deleting it from its
787 	 * current percpu list, complete before setting loaded_vmcs->cpu to
788 	 * -1, otherwise a different cpu can see loaded_vmcs->cpu == -1 first
789 	 * and add loaded_vmcs to its percpu list before it's deleted from this
790 	 * cpu's list. Pairs with the smp_rmb() in vmx_vcpu_load_vmcs().
791 	 */
792 	smp_wmb();
793 
794 	loaded_vmcs->cpu = -1;
795 	loaded_vmcs->launched = 0;
796 }
797 
798 void loaded_vmcs_clear(struct loaded_vmcs *loaded_vmcs)
799 {
800 	int cpu = loaded_vmcs->cpu;
801 
802 	if (cpu != -1)
803 		smp_call_function_single(cpu,
804 			 __loaded_vmcs_clear, loaded_vmcs, 1);
805 }
806 
807 static bool vmx_segment_cache_test_set(struct vcpu_vmx *vmx, unsigned seg,
808 				       unsigned field)
809 {
810 	bool ret;
811 	u32 mask = 1 << (seg * SEG_FIELD_NR + field);
812 
813 	if (!kvm_register_is_available(&vmx->vcpu, VCPU_EXREG_SEGMENTS)) {
814 		kvm_register_mark_available(&vmx->vcpu, VCPU_EXREG_SEGMENTS);
815 		vmx->segment_cache.bitmask = 0;
816 	}
817 	ret = vmx->segment_cache.bitmask & mask;
818 	vmx->segment_cache.bitmask |= mask;
819 	return ret;
820 }
821 
822 static u16 vmx_read_guest_seg_selector(struct vcpu_vmx *vmx, unsigned seg)
823 {
824 	u16 *p = &vmx->segment_cache.seg[seg].selector;
825 
826 	if (!vmx_segment_cache_test_set(vmx, seg, SEG_FIELD_SEL))
827 		*p = vmcs_read16(kvm_vmx_segment_fields[seg].selector);
828 	return *p;
829 }
830 
831 static ulong vmx_read_guest_seg_base(struct vcpu_vmx *vmx, unsigned seg)
832 {
833 	ulong *p = &vmx->segment_cache.seg[seg].base;
834 
835 	if (!vmx_segment_cache_test_set(vmx, seg, SEG_FIELD_BASE))
836 		*p = vmcs_readl(kvm_vmx_segment_fields[seg].base);
837 	return *p;
838 }
839 
840 static u32 vmx_read_guest_seg_limit(struct vcpu_vmx *vmx, unsigned seg)
841 {
842 	u32 *p = &vmx->segment_cache.seg[seg].limit;
843 
844 	if (!vmx_segment_cache_test_set(vmx, seg, SEG_FIELD_LIMIT))
845 		*p = vmcs_read32(kvm_vmx_segment_fields[seg].limit);
846 	return *p;
847 }
848 
849 static u32 vmx_read_guest_seg_ar(struct vcpu_vmx *vmx, unsigned seg)
850 {
851 	u32 *p = &vmx->segment_cache.seg[seg].ar;
852 
853 	if (!vmx_segment_cache_test_set(vmx, seg, SEG_FIELD_AR))
854 		*p = vmcs_read32(kvm_vmx_segment_fields[seg].ar_bytes);
855 	return *p;
856 }
857 
858 void vmx_update_exception_bitmap(struct kvm_vcpu *vcpu)
859 {
860 	u32 eb;
861 
862 	eb = (1u << PF_VECTOR) | (1u << UD_VECTOR) | (1u << MC_VECTOR) |
863 	     (1u << DB_VECTOR) | (1u << AC_VECTOR);
864 	/*
865 	 * Guest access to VMware backdoor ports could legitimately
866 	 * trigger #GP because of TSS I/O permission bitmap.
867 	 * We intercept those #GP and allow access to them anyway
868 	 * as VMware does.
869 	 */
870 	if (enable_vmware_backdoor)
871 		eb |= (1u << GP_VECTOR);
872 	if ((vcpu->guest_debug &
873 	     (KVM_GUESTDBG_ENABLE | KVM_GUESTDBG_USE_SW_BP)) ==
874 	    (KVM_GUESTDBG_ENABLE | KVM_GUESTDBG_USE_SW_BP))
875 		eb |= 1u << BP_VECTOR;
876 	if (to_vmx(vcpu)->rmode.vm86_active)
877 		eb = ~0;
878 	if (!vmx_need_pf_intercept(vcpu))
879 		eb &= ~(1u << PF_VECTOR);
880 
881 	/* When we are running a nested L2 guest and L1 specified for it a
882 	 * certain exception bitmap, we must trap the same exceptions and pass
883 	 * them to L1. When running L2, we will only handle the exceptions
884 	 * specified above if L1 did not want them.
885 	 */
886 	if (is_guest_mode(vcpu))
887 		eb |= get_vmcs12(vcpu)->exception_bitmap;
888 	else {
889 		int mask = 0, match = 0;
890 
891 		if (enable_ept && (eb & (1u << PF_VECTOR))) {
892 			/*
893 			 * If EPT is enabled, #PF is currently only intercepted
894 			 * if MAXPHYADDR is smaller on the guest than on the
895 			 * host.  In that case we only care about present,
896 			 * non-reserved faults.  For vmcs02, however, PFEC_MASK
897 			 * and PFEC_MATCH are set in prepare_vmcs02_rare.
898 			 */
899 			mask = PFERR_PRESENT_MASK | PFERR_RSVD_MASK;
900 			match = PFERR_PRESENT_MASK;
901 		}
902 		vmcs_write32(PAGE_FAULT_ERROR_CODE_MASK, mask);
903 		vmcs_write32(PAGE_FAULT_ERROR_CODE_MATCH, match);
904 	}
905 
906 	/*
907 	 * Disabling xfd interception indicates that dynamic xfeatures
908 	 * might be used in the guest. Always trap #NM in this case
909 	 * to save guest xfd_err timely.
910 	 */
911 	if (vcpu->arch.xfd_no_write_intercept)
912 		eb |= (1u << NM_VECTOR);
913 
914 	vmcs_write32(EXCEPTION_BITMAP, eb);
915 }
916 
917 /*
918  * Check if MSR is intercepted for currently loaded MSR bitmap.
919  */
920 static bool msr_write_intercepted(struct vcpu_vmx *vmx, u32 msr)
921 {
922 	if (!(exec_controls_get(vmx) & CPU_BASED_USE_MSR_BITMAPS))
923 		return true;
924 
925 	return vmx_test_msr_bitmap_write(vmx->loaded_vmcs->msr_bitmap, msr);
926 }
927 
928 unsigned int __vmx_vcpu_run_flags(struct vcpu_vmx *vmx)
929 {
930 	unsigned int flags = 0;
931 
932 	if (vmx->loaded_vmcs->launched)
933 		flags |= VMX_RUN_VMRESUME;
934 
935 	/*
936 	 * If writes to the SPEC_CTRL MSR aren't intercepted, the guest is free
937 	 * to change it directly without causing a vmexit.  In that case read
938 	 * it after vmexit and store it in vmx->spec_ctrl.
939 	 */
940 	if (!msr_write_intercepted(vmx, MSR_IA32_SPEC_CTRL))
941 		flags |= VMX_RUN_SAVE_SPEC_CTRL;
942 
943 	return flags;
944 }
945 
946 static __always_inline void clear_atomic_switch_msr_special(struct vcpu_vmx *vmx,
947 		unsigned long entry, unsigned long exit)
948 {
949 	vm_entry_controls_clearbit(vmx, entry);
950 	vm_exit_controls_clearbit(vmx, exit);
951 }
952 
953 int vmx_find_loadstore_msr_slot(struct vmx_msrs *m, u32 msr)
954 {
955 	unsigned int i;
956 
957 	for (i = 0; i < m->nr; ++i) {
958 		if (m->val[i].index == msr)
959 			return i;
960 	}
961 	return -ENOENT;
962 }
963 
964 static void clear_atomic_switch_msr(struct vcpu_vmx *vmx, unsigned msr)
965 {
966 	int i;
967 	struct msr_autoload *m = &vmx->msr_autoload;
968 
969 	switch (msr) {
970 	case MSR_EFER:
971 		if (cpu_has_load_ia32_efer()) {
972 			clear_atomic_switch_msr_special(vmx,
973 					VM_ENTRY_LOAD_IA32_EFER,
974 					VM_EXIT_LOAD_IA32_EFER);
975 			return;
976 		}
977 		break;
978 	case MSR_CORE_PERF_GLOBAL_CTRL:
979 		if (cpu_has_load_perf_global_ctrl()) {
980 			clear_atomic_switch_msr_special(vmx,
981 					VM_ENTRY_LOAD_IA32_PERF_GLOBAL_CTRL,
982 					VM_EXIT_LOAD_IA32_PERF_GLOBAL_CTRL);
983 			return;
984 		}
985 		break;
986 	}
987 	i = vmx_find_loadstore_msr_slot(&m->guest, msr);
988 	if (i < 0)
989 		goto skip_guest;
990 	--m->guest.nr;
991 	m->guest.val[i] = m->guest.val[m->guest.nr];
992 	vmcs_write32(VM_ENTRY_MSR_LOAD_COUNT, m->guest.nr);
993 
994 skip_guest:
995 	i = vmx_find_loadstore_msr_slot(&m->host, msr);
996 	if (i < 0)
997 		return;
998 
999 	--m->host.nr;
1000 	m->host.val[i] = m->host.val[m->host.nr];
1001 	vmcs_write32(VM_EXIT_MSR_LOAD_COUNT, m->host.nr);
1002 }
1003 
1004 static __always_inline void add_atomic_switch_msr_special(struct vcpu_vmx *vmx,
1005 		unsigned long entry, unsigned long exit,
1006 		unsigned long guest_val_vmcs, unsigned long host_val_vmcs,
1007 		u64 guest_val, u64 host_val)
1008 {
1009 	vmcs_write64(guest_val_vmcs, guest_val);
1010 	if (host_val_vmcs != HOST_IA32_EFER)
1011 		vmcs_write64(host_val_vmcs, host_val);
1012 	vm_entry_controls_setbit(vmx, entry);
1013 	vm_exit_controls_setbit(vmx, exit);
1014 }
1015 
1016 static void add_atomic_switch_msr(struct vcpu_vmx *vmx, unsigned msr,
1017 				  u64 guest_val, u64 host_val, bool entry_only)
1018 {
1019 	int i, j = 0;
1020 	struct msr_autoload *m = &vmx->msr_autoload;
1021 
1022 	switch (msr) {
1023 	case MSR_EFER:
1024 		if (cpu_has_load_ia32_efer()) {
1025 			add_atomic_switch_msr_special(vmx,
1026 					VM_ENTRY_LOAD_IA32_EFER,
1027 					VM_EXIT_LOAD_IA32_EFER,
1028 					GUEST_IA32_EFER,
1029 					HOST_IA32_EFER,
1030 					guest_val, host_val);
1031 			return;
1032 		}
1033 		break;
1034 	case MSR_CORE_PERF_GLOBAL_CTRL:
1035 		if (cpu_has_load_perf_global_ctrl()) {
1036 			add_atomic_switch_msr_special(vmx,
1037 					VM_ENTRY_LOAD_IA32_PERF_GLOBAL_CTRL,
1038 					VM_EXIT_LOAD_IA32_PERF_GLOBAL_CTRL,
1039 					GUEST_IA32_PERF_GLOBAL_CTRL,
1040 					HOST_IA32_PERF_GLOBAL_CTRL,
1041 					guest_val, host_val);
1042 			return;
1043 		}
1044 		break;
1045 	case MSR_IA32_PEBS_ENABLE:
1046 		/* PEBS needs a quiescent period after being disabled (to write
1047 		 * a record).  Disabling PEBS through VMX MSR swapping doesn't
1048 		 * provide that period, so a CPU could write host's record into
1049 		 * guest's memory.
1050 		 */
1051 		wrmsrl(MSR_IA32_PEBS_ENABLE, 0);
1052 	}
1053 
1054 	i = vmx_find_loadstore_msr_slot(&m->guest, msr);
1055 	if (!entry_only)
1056 		j = vmx_find_loadstore_msr_slot(&m->host, msr);
1057 
1058 	if ((i < 0 && m->guest.nr == MAX_NR_LOADSTORE_MSRS) ||
1059 	    (j < 0 &&  m->host.nr == MAX_NR_LOADSTORE_MSRS)) {
1060 		printk_once(KERN_WARNING "Not enough msr switch entries. "
1061 				"Can't add msr %x\n", msr);
1062 		return;
1063 	}
1064 	if (i < 0) {
1065 		i = m->guest.nr++;
1066 		vmcs_write32(VM_ENTRY_MSR_LOAD_COUNT, m->guest.nr);
1067 	}
1068 	m->guest.val[i].index = msr;
1069 	m->guest.val[i].value = guest_val;
1070 
1071 	if (entry_only)
1072 		return;
1073 
1074 	if (j < 0) {
1075 		j = m->host.nr++;
1076 		vmcs_write32(VM_EXIT_MSR_LOAD_COUNT, m->host.nr);
1077 	}
1078 	m->host.val[j].index = msr;
1079 	m->host.val[j].value = host_val;
1080 }
1081 
1082 static bool update_transition_efer(struct vcpu_vmx *vmx)
1083 {
1084 	u64 guest_efer = vmx->vcpu.arch.efer;
1085 	u64 ignore_bits = 0;
1086 	int i;
1087 
1088 	/* Shadow paging assumes NX to be available.  */
1089 	if (!enable_ept)
1090 		guest_efer |= EFER_NX;
1091 
1092 	/*
1093 	 * LMA and LME handled by hardware; SCE meaningless outside long mode.
1094 	 */
1095 	ignore_bits |= EFER_SCE;
1096 #ifdef CONFIG_X86_64
1097 	ignore_bits |= EFER_LMA | EFER_LME;
1098 	/* SCE is meaningful only in long mode on Intel */
1099 	if (guest_efer & EFER_LMA)
1100 		ignore_bits &= ~(u64)EFER_SCE;
1101 #endif
1102 
1103 	/*
1104 	 * On EPT, we can't emulate NX, so we must switch EFER atomically.
1105 	 * On CPUs that support "load IA32_EFER", always switch EFER
1106 	 * atomically, since it's faster than switching it manually.
1107 	 */
1108 	if (cpu_has_load_ia32_efer() ||
1109 	    (enable_ept && ((vmx->vcpu.arch.efer ^ host_efer) & EFER_NX))) {
1110 		if (!(guest_efer & EFER_LMA))
1111 			guest_efer &= ~EFER_LME;
1112 		if (guest_efer != host_efer)
1113 			add_atomic_switch_msr(vmx, MSR_EFER,
1114 					      guest_efer, host_efer, false);
1115 		else
1116 			clear_atomic_switch_msr(vmx, MSR_EFER);
1117 		return false;
1118 	}
1119 
1120 	i = kvm_find_user_return_msr(MSR_EFER);
1121 	if (i < 0)
1122 		return false;
1123 
1124 	clear_atomic_switch_msr(vmx, MSR_EFER);
1125 
1126 	guest_efer &= ~ignore_bits;
1127 	guest_efer |= host_efer & ignore_bits;
1128 
1129 	vmx->guest_uret_msrs[i].data = guest_efer;
1130 	vmx->guest_uret_msrs[i].mask = ~ignore_bits;
1131 
1132 	return true;
1133 }
1134 
1135 #ifdef CONFIG_X86_32
1136 /*
1137  * On 32-bit kernels, VM exits still load the FS and GS bases from the
1138  * VMCS rather than the segment table.  KVM uses this helper to figure
1139  * out the current bases to poke them into the VMCS before entry.
1140  */
1141 static unsigned long segment_base(u16 selector)
1142 {
1143 	struct desc_struct *table;
1144 	unsigned long v;
1145 
1146 	if (!(selector & ~SEGMENT_RPL_MASK))
1147 		return 0;
1148 
1149 	table = get_current_gdt_ro();
1150 
1151 	if ((selector & SEGMENT_TI_MASK) == SEGMENT_LDT) {
1152 		u16 ldt_selector = kvm_read_ldt();
1153 
1154 		if (!(ldt_selector & ~SEGMENT_RPL_MASK))
1155 			return 0;
1156 
1157 		table = (struct desc_struct *)segment_base(ldt_selector);
1158 	}
1159 	v = get_desc_base(&table[selector >> 3]);
1160 	return v;
1161 }
1162 #endif
1163 
1164 static inline bool pt_can_write_msr(struct vcpu_vmx *vmx)
1165 {
1166 	return vmx_pt_mode_is_host_guest() &&
1167 	       !(vmx->pt_desc.guest.ctl & RTIT_CTL_TRACEEN);
1168 }
1169 
1170 static inline bool pt_output_base_valid(struct kvm_vcpu *vcpu, u64 base)
1171 {
1172 	/* The base must be 128-byte aligned and a legal physical address. */
1173 	return kvm_vcpu_is_legal_aligned_gpa(vcpu, base, 128);
1174 }
1175 
1176 static inline void pt_load_msr(struct pt_ctx *ctx, u32 addr_range)
1177 {
1178 	u32 i;
1179 
1180 	wrmsrl(MSR_IA32_RTIT_STATUS, ctx->status);
1181 	wrmsrl(MSR_IA32_RTIT_OUTPUT_BASE, ctx->output_base);
1182 	wrmsrl(MSR_IA32_RTIT_OUTPUT_MASK, ctx->output_mask);
1183 	wrmsrl(MSR_IA32_RTIT_CR3_MATCH, ctx->cr3_match);
1184 	for (i = 0; i < addr_range; i++) {
1185 		wrmsrl(MSR_IA32_RTIT_ADDR0_A + i * 2, ctx->addr_a[i]);
1186 		wrmsrl(MSR_IA32_RTIT_ADDR0_B + i * 2, ctx->addr_b[i]);
1187 	}
1188 }
1189 
1190 static inline void pt_save_msr(struct pt_ctx *ctx, u32 addr_range)
1191 {
1192 	u32 i;
1193 
1194 	rdmsrl(MSR_IA32_RTIT_STATUS, ctx->status);
1195 	rdmsrl(MSR_IA32_RTIT_OUTPUT_BASE, ctx->output_base);
1196 	rdmsrl(MSR_IA32_RTIT_OUTPUT_MASK, ctx->output_mask);
1197 	rdmsrl(MSR_IA32_RTIT_CR3_MATCH, ctx->cr3_match);
1198 	for (i = 0; i < addr_range; i++) {
1199 		rdmsrl(MSR_IA32_RTIT_ADDR0_A + i * 2, ctx->addr_a[i]);
1200 		rdmsrl(MSR_IA32_RTIT_ADDR0_B + i * 2, ctx->addr_b[i]);
1201 	}
1202 }
1203 
1204 static void pt_guest_enter(struct vcpu_vmx *vmx)
1205 {
1206 	if (vmx_pt_mode_is_system())
1207 		return;
1208 
1209 	/*
1210 	 * GUEST_IA32_RTIT_CTL is already set in the VMCS.
1211 	 * Save host state before VM entry.
1212 	 */
1213 	rdmsrl(MSR_IA32_RTIT_CTL, vmx->pt_desc.host.ctl);
1214 	if (vmx->pt_desc.guest.ctl & RTIT_CTL_TRACEEN) {
1215 		wrmsrl(MSR_IA32_RTIT_CTL, 0);
1216 		pt_save_msr(&vmx->pt_desc.host, vmx->pt_desc.num_address_ranges);
1217 		pt_load_msr(&vmx->pt_desc.guest, vmx->pt_desc.num_address_ranges);
1218 	}
1219 }
1220 
1221 static void pt_guest_exit(struct vcpu_vmx *vmx)
1222 {
1223 	if (vmx_pt_mode_is_system())
1224 		return;
1225 
1226 	if (vmx->pt_desc.guest.ctl & RTIT_CTL_TRACEEN) {
1227 		pt_save_msr(&vmx->pt_desc.guest, vmx->pt_desc.num_address_ranges);
1228 		pt_load_msr(&vmx->pt_desc.host, vmx->pt_desc.num_address_ranges);
1229 	}
1230 
1231 	/*
1232 	 * KVM requires VM_EXIT_CLEAR_IA32_RTIT_CTL to expose PT to the guest,
1233 	 * i.e. RTIT_CTL is always cleared on VM-Exit.  Restore it if necessary.
1234 	 */
1235 	if (vmx->pt_desc.host.ctl)
1236 		wrmsrl(MSR_IA32_RTIT_CTL, vmx->pt_desc.host.ctl);
1237 }
1238 
1239 void vmx_set_host_fs_gs(struct vmcs_host_state *host, u16 fs_sel, u16 gs_sel,
1240 			unsigned long fs_base, unsigned long gs_base)
1241 {
1242 	if (unlikely(fs_sel != host->fs_sel)) {
1243 		if (!(fs_sel & 7))
1244 			vmcs_write16(HOST_FS_SELECTOR, fs_sel);
1245 		else
1246 			vmcs_write16(HOST_FS_SELECTOR, 0);
1247 		host->fs_sel = fs_sel;
1248 	}
1249 	if (unlikely(gs_sel != host->gs_sel)) {
1250 		if (!(gs_sel & 7))
1251 			vmcs_write16(HOST_GS_SELECTOR, gs_sel);
1252 		else
1253 			vmcs_write16(HOST_GS_SELECTOR, 0);
1254 		host->gs_sel = gs_sel;
1255 	}
1256 	if (unlikely(fs_base != host->fs_base)) {
1257 		vmcs_writel(HOST_FS_BASE, fs_base);
1258 		host->fs_base = fs_base;
1259 	}
1260 	if (unlikely(gs_base != host->gs_base)) {
1261 		vmcs_writel(HOST_GS_BASE, gs_base);
1262 		host->gs_base = gs_base;
1263 	}
1264 }
1265 
1266 void vmx_prepare_switch_to_guest(struct kvm_vcpu *vcpu)
1267 {
1268 	struct vcpu_vmx *vmx = to_vmx(vcpu);
1269 	struct vmcs_host_state *host_state;
1270 #ifdef CONFIG_X86_64
1271 	int cpu = raw_smp_processor_id();
1272 #endif
1273 	unsigned long fs_base, gs_base;
1274 	u16 fs_sel, gs_sel;
1275 	int i;
1276 
1277 	vmx->req_immediate_exit = false;
1278 
1279 	/*
1280 	 * Note that guest MSRs to be saved/restored can also be changed
1281 	 * when guest state is loaded. This happens when guest transitions
1282 	 * to/from long-mode by setting MSR_EFER.LMA.
1283 	 */
1284 	if (!vmx->guest_uret_msrs_loaded) {
1285 		vmx->guest_uret_msrs_loaded = true;
1286 		for (i = 0; i < kvm_nr_uret_msrs; ++i) {
1287 			if (!vmx->guest_uret_msrs[i].load_into_hardware)
1288 				continue;
1289 
1290 			kvm_set_user_return_msr(i,
1291 						vmx->guest_uret_msrs[i].data,
1292 						vmx->guest_uret_msrs[i].mask);
1293 		}
1294 	}
1295 
1296 	if (vmx->nested.need_vmcs12_to_shadow_sync)
1297 		nested_sync_vmcs12_to_shadow(vcpu);
1298 
1299 	if (vmx->guest_state_loaded)
1300 		return;
1301 
1302 	host_state = &vmx->loaded_vmcs->host_state;
1303 
1304 	/*
1305 	 * Set host fs and gs selectors.  Unfortunately, 22.2.3 does not
1306 	 * allow segment selectors with cpl > 0 or ti == 1.
1307 	 */
1308 	host_state->ldt_sel = kvm_read_ldt();
1309 
1310 #ifdef CONFIG_X86_64
1311 	savesegment(ds, host_state->ds_sel);
1312 	savesegment(es, host_state->es_sel);
1313 
1314 	gs_base = cpu_kernelmode_gs_base(cpu);
1315 	if (likely(is_64bit_mm(current->mm))) {
1316 		current_save_fsgs();
1317 		fs_sel = current->thread.fsindex;
1318 		gs_sel = current->thread.gsindex;
1319 		fs_base = current->thread.fsbase;
1320 		vmx->msr_host_kernel_gs_base = current->thread.gsbase;
1321 	} else {
1322 		savesegment(fs, fs_sel);
1323 		savesegment(gs, gs_sel);
1324 		fs_base = read_msr(MSR_FS_BASE);
1325 		vmx->msr_host_kernel_gs_base = read_msr(MSR_KERNEL_GS_BASE);
1326 	}
1327 
1328 	wrmsrl(MSR_KERNEL_GS_BASE, vmx->msr_guest_kernel_gs_base);
1329 #else
1330 	savesegment(fs, fs_sel);
1331 	savesegment(gs, gs_sel);
1332 	fs_base = segment_base(fs_sel);
1333 	gs_base = segment_base(gs_sel);
1334 #endif
1335 
1336 	vmx_set_host_fs_gs(host_state, fs_sel, gs_sel, fs_base, gs_base);
1337 	vmx->guest_state_loaded = true;
1338 }
1339 
1340 static void vmx_prepare_switch_to_host(struct vcpu_vmx *vmx)
1341 {
1342 	struct vmcs_host_state *host_state;
1343 
1344 	if (!vmx->guest_state_loaded)
1345 		return;
1346 
1347 	host_state = &vmx->loaded_vmcs->host_state;
1348 
1349 	++vmx->vcpu.stat.host_state_reload;
1350 
1351 #ifdef CONFIG_X86_64
1352 	rdmsrl(MSR_KERNEL_GS_BASE, vmx->msr_guest_kernel_gs_base);
1353 #endif
1354 	if (host_state->ldt_sel || (host_state->gs_sel & 7)) {
1355 		kvm_load_ldt(host_state->ldt_sel);
1356 #ifdef CONFIG_X86_64
1357 		load_gs_index(host_state->gs_sel);
1358 #else
1359 		loadsegment(gs, host_state->gs_sel);
1360 #endif
1361 	}
1362 	if (host_state->fs_sel & 7)
1363 		loadsegment(fs, host_state->fs_sel);
1364 #ifdef CONFIG_X86_64
1365 	if (unlikely(host_state->ds_sel | host_state->es_sel)) {
1366 		loadsegment(ds, host_state->ds_sel);
1367 		loadsegment(es, host_state->es_sel);
1368 	}
1369 #endif
1370 	invalidate_tss_limit();
1371 #ifdef CONFIG_X86_64
1372 	wrmsrl(MSR_KERNEL_GS_BASE, vmx->msr_host_kernel_gs_base);
1373 #endif
1374 	load_fixmap_gdt(raw_smp_processor_id());
1375 	vmx->guest_state_loaded = false;
1376 	vmx->guest_uret_msrs_loaded = false;
1377 }
1378 
1379 #ifdef CONFIG_X86_64
1380 static u64 vmx_read_guest_kernel_gs_base(struct vcpu_vmx *vmx)
1381 {
1382 	preempt_disable();
1383 	if (vmx->guest_state_loaded)
1384 		rdmsrl(MSR_KERNEL_GS_BASE, vmx->msr_guest_kernel_gs_base);
1385 	preempt_enable();
1386 	return vmx->msr_guest_kernel_gs_base;
1387 }
1388 
1389 static void vmx_write_guest_kernel_gs_base(struct vcpu_vmx *vmx, u64 data)
1390 {
1391 	preempt_disable();
1392 	if (vmx->guest_state_loaded)
1393 		wrmsrl(MSR_KERNEL_GS_BASE, data);
1394 	preempt_enable();
1395 	vmx->msr_guest_kernel_gs_base = data;
1396 }
1397 #endif
1398 
1399 void vmx_vcpu_load_vmcs(struct kvm_vcpu *vcpu, int cpu,
1400 			struct loaded_vmcs *buddy)
1401 {
1402 	struct vcpu_vmx *vmx = to_vmx(vcpu);
1403 	bool already_loaded = vmx->loaded_vmcs->cpu == cpu;
1404 	struct vmcs *prev;
1405 
1406 	if (!already_loaded) {
1407 		loaded_vmcs_clear(vmx->loaded_vmcs);
1408 		local_irq_disable();
1409 
1410 		/*
1411 		 * Ensure loaded_vmcs->cpu is read before adding loaded_vmcs to
1412 		 * this cpu's percpu list, otherwise it may not yet be deleted
1413 		 * from its previous cpu's percpu list.  Pairs with the
1414 		 * smb_wmb() in __loaded_vmcs_clear().
1415 		 */
1416 		smp_rmb();
1417 
1418 		list_add(&vmx->loaded_vmcs->loaded_vmcss_on_cpu_link,
1419 			 &per_cpu(loaded_vmcss_on_cpu, cpu));
1420 		local_irq_enable();
1421 	}
1422 
1423 	prev = per_cpu(current_vmcs, cpu);
1424 	if (prev != vmx->loaded_vmcs->vmcs) {
1425 		per_cpu(current_vmcs, cpu) = vmx->loaded_vmcs->vmcs;
1426 		vmcs_load(vmx->loaded_vmcs->vmcs);
1427 
1428 		/*
1429 		 * No indirect branch prediction barrier needed when switching
1430 		 * the active VMCS within a vCPU, unless IBRS is advertised to
1431 		 * the vCPU.  To minimize the number of IBPBs executed, KVM
1432 		 * performs IBPB on nested VM-Exit (a single nested transition
1433 		 * may switch the active VMCS multiple times).
1434 		 */
1435 		if (!buddy || WARN_ON_ONCE(buddy->vmcs != prev))
1436 			indirect_branch_prediction_barrier();
1437 	}
1438 
1439 	if (!already_loaded) {
1440 		void *gdt = get_current_gdt_ro();
1441 
1442 		/*
1443 		 * Flush all EPTP/VPID contexts, the new pCPU may have stale
1444 		 * TLB entries from its previous association with the vCPU.
1445 		 */
1446 		kvm_make_request(KVM_REQ_TLB_FLUSH, vcpu);
1447 
1448 		/*
1449 		 * Linux uses per-cpu TSS and GDT, so set these when switching
1450 		 * processors.  See 22.2.4.
1451 		 */
1452 		vmcs_writel(HOST_TR_BASE,
1453 			    (unsigned long)&get_cpu_entry_area(cpu)->tss.x86_tss);
1454 		vmcs_writel(HOST_GDTR_BASE, (unsigned long)gdt);   /* 22.2.4 */
1455 
1456 		if (IS_ENABLED(CONFIG_IA32_EMULATION) || IS_ENABLED(CONFIG_X86_32)) {
1457 			/* 22.2.3 */
1458 			vmcs_writel(HOST_IA32_SYSENTER_ESP,
1459 				    (unsigned long)(cpu_entry_stack(cpu) + 1));
1460 		}
1461 
1462 		vmx->loaded_vmcs->cpu = cpu;
1463 	}
1464 }
1465 
1466 /*
1467  * Switches to specified vcpu, until a matching vcpu_put(), but assumes
1468  * vcpu mutex is already taken.
1469  */
1470 static void vmx_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
1471 {
1472 	struct vcpu_vmx *vmx = to_vmx(vcpu);
1473 
1474 	vmx_vcpu_load_vmcs(vcpu, cpu, NULL);
1475 
1476 	vmx_vcpu_pi_load(vcpu, cpu);
1477 
1478 	vmx->host_debugctlmsr = get_debugctlmsr();
1479 }
1480 
1481 static void vmx_vcpu_put(struct kvm_vcpu *vcpu)
1482 {
1483 	vmx_vcpu_pi_put(vcpu);
1484 
1485 	vmx_prepare_switch_to_host(to_vmx(vcpu));
1486 }
1487 
1488 bool vmx_emulation_required(struct kvm_vcpu *vcpu)
1489 {
1490 	return emulate_invalid_guest_state && !vmx_guest_state_valid(vcpu);
1491 }
1492 
1493 unsigned long vmx_get_rflags(struct kvm_vcpu *vcpu)
1494 {
1495 	struct vcpu_vmx *vmx = to_vmx(vcpu);
1496 	unsigned long rflags, save_rflags;
1497 
1498 	if (!kvm_register_is_available(vcpu, VCPU_EXREG_RFLAGS)) {
1499 		kvm_register_mark_available(vcpu, VCPU_EXREG_RFLAGS);
1500 		rflags = vmcs_readl(GUEST_RFLAGS);
1501 		if (vmx->rmode.vm86_active) {
1502 			rflags &= RMODE_GUEST_OWNED_EFLAGS_BITS;
1503 			save_rflags = vmx->rmode.save_rflags;
1504 			rflags |= save_rflags & ~RMODE_GUEST_OWNED_EFLAGS_BITS;
1505 		}
1506 		vmx->rflags = rflags;
1507 	}
1508 	return vmx->rflags;
1509 }
1510 
1511 void vmx_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags)
1512 {
1513 	struct vcpu_vmx *vmx = to_vmx(vcpu);
1514 	unsigned long old_rflags;
1515 
1516 	if (is_unrestricted_guest(vcpu)) {
1517 		kvm_register_mark_available(vcpu, VCPU_EXREG_RFLAGS);
1518 		vmx->rflags = rflags;
1519 		vmcs_writel(GUEST_RFLAGS, rflags);
1520 		return;
1521 	}
1522 
1523 	old_rflags = vmx_get_rflags(vcpu);
1524 	vmx->rflags = rflags;
1525 	if (vmx->rmode.vm86_active) {
1526 		vmx->rmode.save_rflags = rflags;
1527 		rflags |= X86_EFLAGS_IOPL | X86_EFLAGS_VM;
1528 	}
1529 	vmcs_writel(GUEST_RFLAGS, rflags);
1530 
1531 	if ((old_rflags ^ vmx->rflags) & X86_EFLAGS_VM)
1532 		vmx->emulation_required = vmx_emulation_required(vcpu);
1533 }
1534 
1535 static bool vmx_get_if_flag(struct kvm_vcpu *vcpu)
1536 {
1537 	return vmx_get_rflags(vcpu) & X86_EFLAGS_IF;
1538 }
1539 
1540 u32 vmx_get_interrupt_shadow(struct kvm_vcpu *vcpu)
1541 {
1542 	u32 interruptibility = vmcs_read32(GUEST_INTERRUPTIBILITY_INFO);
1543 	int ret = 0;
1544 
1545 	if (interruptibility & GUEST_INTR_STATE_STI)
1546 		ret |= KVM_X86_SHADOW_INT_STI;
1547 	if (interruptibility & GUEST_INTR_STATE_MOV_SS)
1548 		ret |= KVM_X86_SHADOW_INT_MOV_SS;
1549 
1550 	return ret;
1551 }
1552 
1553 void vmx_set_interrupt_shadow(struct kvm_vcpu *vcpu, int mask)
1554 {
1555 	u32 interruptibility_old = vmcs_read32(GUEST_INTERRUPTIBILITY_INFO);
1556 	u32 interruptibility = interruptibility_old;
1557 
1558 	interruptibility &= ~(GUEST_INTR_STATE_STI | GUEST_INTR_STATE_MOV_SS);
1559 
1560 	if (mask & KVM_X86_SHADOW_INT_MOV_SS)
1561 		interruptibility |= GUEST_INTR_STATE_MOV_SS;
1562 	else if (mask & KVM_X86_SHADOW_INT_STI)
1563 		interruptibility |= GUEST_INTR_STATE_STI;
1564 
1565 	if ((interruptibility != interruptibility_old))
1566 		vmcs_write32(GUEST_INTERRUPTIBILITY_INFO, interruptibility);
1567 }
1568 
1569 static int vmx_rtit_ctl_check(struct kvm_vcpu *vcpu, u64 data)
1570 {
1571 	struct vcpu_vmx *vmx = to_vmx(vcpu);
1572 	unsigned long value;
1573 
1574 	/*
1575 	 * Any MSR write that attempts to change bits marked reserved will
1576 	 * case a #GP fault.
1577 	 */
1578 	if (data & vmx->pt_desc.ctl_bitmask)
1579 		return 1;
1580 
1581 	/*
1582 	 * Any attempt to modify IA32_RTIT_CTL while TraceEn is set will
1583 	 * result in a #GP unless the same write also clears TraceEn.
1584 	 */
1585 	if ((vmx->pt_desc.guest.ctl & RTIT_CTL_TRACEEN) &&
1586 		((vmx->pt_desc.guest.ctl ^ data) & ~RTIT_CTL_TRACEEN))
1587 		return 1;
1588 
1589 	/*
1590 	 * WRMSR to IA32_RTIT_CTL that sets TraceEn but clears this bit
1591 	 * and FabricEn would cause #GP, if
1592 	 * CPUID.(EAX=14H, ECX=0):ECX.SNGLRGNOUT[bit 2] = 0
1593 	 */
1594 	if ((data & RTIT_CTL_TRACEEN) && !(data & RTIT_CTL_TOPA) &&
1595 		!(data & RTIT_CTL_FABRIC_EN) &&
1596 		!intel_pt_validate_cap(vmx->pt_desc.caps,
1597 					PT_CAP_single_range_output))
1598 		return 1;
1599 
1600 	/*
1601 	 * MTCFreq, CycThresh and PSBFreq encodings check, any MSR write that
1602 	 * utilize encodings marked reserved will cause a #GP fault.
1603 	 */
1604 	value = intel_pt_validate_cap(vmx->pt_desc.caps, PT_CAP_mtc_periods);
1605 	if (intel_pt_validate_cap(vmx->pt_desc.caps, PT_CAP_mtc) &&
1606 			!test_bit((data & RTIT_CTL_MTC_RANGE) >>
1607 			RTIT_CTL_MTC_RANGE_OFFSET, &value))
1608 		return 1;
1609 	value = intel_pt_validate_cap(vmx->pt_desc.caps,
1610 						PT_CAP_cycle_thresholds);
1611 	if (intel_pt_validate_cap(vmx->pt_desc.caps, PT_CAP_psb_cyc) &&
1612 			!test_bit((data & RTIT_CTL_CYC_THRESH) >>
1613 			RTIT_CTL_CYC_THRESH_OFFSET, &value))
1614 		return 1;
1615 	value = intel_pt_validate_cap(vmx->pt_desc.caps, PT_CAP_psb_periods);
1616 	if (intel_pt_validate_cap(vmx->pt_desc.caps, PT_CAP_psb_cyc) &&
1617 			!test_bit((data & RTIT_CTL_PSB_FREQ) >>
1618 			RTIT_CTL_PSB_FREQ_OFFSET, &value))
1619 		return 1;
1620 
1621 	/*
1622 	 * If ADDRx_CFG is reserved or the encodings is >2 will
1623 	 * cause a #GP fault.
1624 	 */
1625 	value = (data & RTIT_CTL_ADDR0) >> RTIT_CTL_ADDR0_OFFSET;
1626 	if ((value && (vmx->pt_desc.num_address_ranges < 1)) || (value > 2))
1627 		return 1;
1628 	value = (data & RTIT_CTL_ADDR1) >> RTIT_CTL_ADDR1_OFFSET;
1629 	if ((value && (vmx->pt_desc.num_address_ranges < 2)) || (value > 2))
1630 		return 1;
1631 	value = (data & RTIT_CTL_ADDR2) >> RTIT_CTL_ADDR2_OFFSET;
1632 	if ((value && (vmx->pt_desc.num_address_ranges < 3)) || (value > 2))
1633 		return 1;
1634 	value = (data & RTIT_CTL_ADDR3) >> RTIT_CTL_ADDR3_OFFSET;
1635 	if ((value && (vmx->pt_desc.num_address_ranges < 4)) || (value > 2))
1636 		return 1;
1637 
1638 	return 0;
1639 }
1640 
1641 static bool vmx_can_emulate_instruction(struct kvm_vcpu *vcpu, int emul_type,
1642 					void *insn, int insn_len)
1643 {
1644 	/*
1645 	 * Emulation of instructions in SGX enclaves is impossible as RIP does
1646 	 * not point at the failing instruction, and even if it did, the code
1647 	 * stream is inaccessible.  Inject #UD instead of exiting to userspace
1648 	 * so that guest userspace can't DoS the guest simply by triggering
1649 	 * emulation (enclaves are CPL3 only).
1650 	 */
1651 	if (to_vmx(vcpu)->exit_reason.enclave_mode) {
1652 		kvm_queue_exception(vcpu, UD_VECTOR);
1653 		return false;
1654 	}
1655 	return true;
1656 }
1657 
1658 static int skip_emulated_instruction(struct kvm_vcpu *vcpu)
1659 {
1660 	union vmx_exit_reason exit_reason = to_vmx(vcpu)->exit_reason;
1661 	unsigned long rip, orig_rip;
1662 	u32 instr_len;
1663 
1664 	/*
1665 	 * Using VMCS.VM_EXIT_INSTRUCTION_LEN on EPT misconfig depends on
1666 	 * undefined behavior: Intel's SDM doesn't mandate the VMCS field be
1667 	 * set when EPT misconfig occurs.  In practice, real hardware updates
1668 	 * VM_EXIT_INSTRUCTION_LEN on EPT misconfig, but other hypervisors
1669 	 * (namely Hyper-V) don't set it due to it being undefined behavior,
1670 	 * i.e. we end up advancing IP with some random value.
1671 	 */
1672 	if (!static_cpu_has(X86_FEATURE_HYPERVISOR) ||
1673 	    exit_reason.basic != EXIT_REASON_EPT_MISCONFIG) {
1674 		instr_len = vmcs_read32(VM_EXIT_INSTRUCTION_LEN);
1675 
1676 		/*
1677 		 * Emulating an enclave's instructions isn't supported as KVM
1678 		 * cannot access the enclave's memory or its true RIP, e.g. the
1679 		 * vmcs.GUEST_RIP points at the exit point of the enclave, not
1680 		 * the RIP that actually triggered the VM-Exit.  But, because
1681 		 * most instructions that cause VM-Exit will #UD in an enclave,
1682 		 * most instruction-based VM-Exits simply do not occur.
1683 		 *
1684 		 * There are a few exceptions, notably the debug instructions
1685 		 * INT1ICEBRK and INT3, as they are allowed in debug enclaves
1686 		 * and generate #DB/#BP as expected, which KVM might intercept.
1687 		 * But again, the CPU does the dirty work and saves an instr
1688 		 * length of zero so VMMs don't shoot themselves in the foot.
1689 		 * WARN if KVM tries to skip a non-zero length instruction on
1690 		 * a VM-Exit from an enclave.
1691 		 */
1692 		if (!instr_len)
1693 			goto rip_updated;
1694 
1695 		WARN_ONCE(exit_reason.enclave_mode,
1696 			  "skipping instruction after SGX enclave VM-Exit");
1697 
1698 		orig_rip = kvm_rip_read(vcpu);
1699 		rip = orig_rip + instr_len;
1700 #ifdef CONFIG_X86_64
1701 		/*
1702 		 * We need to mask out the high 32 bits of RIP if not in 64-bit
1703 		 * mode, but just finding out that we are in 64-bit mode is
1704 		 * quite expensive.  Only do it if there was a carry.
1705 		 */
1706 		if (unlikely(((rip ^ orig_rip) >> 31) == 3) && !is_64_bit_mode(vcpu))
1707 			rip = (u32)rip;
1708 #endif
1709 		kvm_rip_write(vcpu, rip);
1710 	} else {
1711 		if (!kvm_emulate_instruction(vcpu, EMULTYPE_SKIP))
1712 			return 0;
1713 	}
1714 
1715 rip_updated:
1716 	/* skipping an emulated instruction also counts */
1717 	vmx_set_interrupt_shadow(vcpu, 0);
1718 
1719 	return 1;
1720 }
1721 
1722 /*
1723  * Recognizes a pending MTF VM-exit and records the nested state for later
1724  * delivery.
1725  */
1726 static void vmx_update_emulated_instruction(struct kvm_vcpu *vcpu)
1727 {
1728 	struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
1729 	struct vcpu_vmx *vmx = to_vmx(vcpu);
1730 
1731 	if (!is_guest_mode(vcpu))
1732 		return;
1733 
1734 	/*
1735 	 * Per the SDM, MTF takes priority over debug-trap exceptions besides
1736 	 * TSS T-bit traps and ICEBP (INT1).  KVM doesn't emulate T-bit traps
1737 	 * or ICEBP (in the emulator proper), and skipping of ICEBP after an
1738 	 * intercepted #DB deliberately avoids single-step #DB and MTF updates
1739 	 * as ICEBP is higher priority than both.  As instruction emulation is
1740 	 * completed at this point (i.e. KVM is at the instruction boundary),
1741 	 * any #DB exception pending delivery must be a debug-trap of lower
1742 	 * priority than MTF.  Record the pending MTF state to be delivered in
1743 	 * vmx_check_nested_events().
1744 	 */
1745 	if (nested_cpu_has_mtf(vmcs12) &&
1746 	    (!vcpu->arch.exception.pending ||
1747 	     vcpu->arch.exception.vector == DB_VECTOR) &&
1748 	    (!vcpu->arch.exception_vmexit.pending ||
1749 	     vcpu->arch.exception_vmexit.vector == DB_VECTOR)) {
1750 		vmx->nested.mtf_pending = true;
1751 		kvm_make_request(KVM_REQ_EVENT, vcpu);
1752 	} else {
1753 		vmx->nested.mtf_pending = false;
1754 	}
1755 }
1756 
1757 static int vmx_skip_emulated_instruction(struct kvm_vcpu *vcpu)
1758 {
1759 	vmx_update_emulated_instruction(vcpu);
1760 	return skip_emulated_instruction(vcpu);
1761 }
1762 
1763 static void vmx_clear_hlt(struct kvm_vcpu *vcpu)
1764 {
1765 	/*
1766 	 * Ensure that we clear the HLT state in the VMCS.  We don't need to
1767 	 * explicitly skip the instruction because if the HLT state is set,
1768 	 * then the instruction is already executing and RIP has already been
1769 	 * advanced.
1770 	 */
1771 	if (kvm_hlt_in_guest(vcpu->kvm) &&
1772 			vmcs_read32(GUEST_ACTIVITY_STATE) == GUEST_ACTIVITY_HLT)
1773 		vmcs_write32(GUEST_ACTIVITY_STATE, GUEST_ACTIVITY_ACTIVE);
1774 }
1775 
1776 static void vmx_inject_exception(struct kvm_vcpu *vcpu)
1777 {
1778 	struct kvm_queued_exception *ex = &vcpu->arch.exception;
1779 	u32 intr_info = ex->vector | INTR_INFO_VALID_MASK;
1780 	struct vcpu_vmx *vmx = to_vmx(vcpu);
1781 
1782 	kvm_deliver_exception_payload(vcpu, ex);
1783 
1784 	if (ex->has_error_code) {
1785 		/*
1786 		 * Despite the error code being architecturally defined as 32
1787 		 * bits, and the VMCS field being 32 bits, Intel CPUs and thus
1788 		 * VMX don't actually supporting setting bits 31:16.  Hardware
1789 		 * will (should) never provide a bogus error code, but AMD CPUs
1790 		 * do generate error codes with bits 31:16 set, and so KVM's
1791 		 * ABI lets userspace shove in arbitrary 32-bit values.  Drop
1792 		 * the upper bits to avoid VM-Fail, losing information that
1793 		 * does't really exist is preferable to killing the VM.
1794 		 */
1795 		vmcs_write32(VM_ENTRY_EXCEPTION_ERROR_CODE, (u16)ex->error_code);
1796 		intr_info |= INTR_INFO_DELIVER_CODE_MASK;
1797 	}
1798 
1799 	if (vmx->rmode.vm86_active) {
1800 		int inc_eip = 0;
1801 		if (kvm_exception_is_soft(ex->vector))
1802 			inc_eip = vcpu->arch.event_exit_inst_len;
1803 		kvm_inject_realmode_interrupt(vcpu, ex->vector, inc_eip);
1804 		return;
1805 	}
1806 
1807 	WARN_ON_ONCE(vmx->emulation_required);
1808 
1809 	if (kvm_exception_is_soft(ex->vector)) {
1810 		vmcs_write32(VM_ENTRY_INSTRUCTION_LEN,
1811 			     vmx->vcpu.arch.event_exit_inst_len);
1812 		intr_info |= INTR_TYPE_SOFT_EXCEPTION;
1813 	} else
1814 		intr_info |= INTR_TYPE_HARD_EXCEPTION;
1815 
1816 	vmcs_write32(VM_ENTRY_INTR_INFO_FIELD, intr_info);
1817 
1818 	vmx_clear_hlt(vcpu);
1819 }
1820 
1821 static void vmx_setup_uret_msr(struct vcpu_vmx *vmx, unsigned int msr,
1822 			       bool load_into_hardware)
1823 {
1824 	struct vmx_uret_msr *uret_msr;
1825 
1826 	uret_msr = vmx_find_uret_msr(vmx, msr);
1827 	if (!uret_msr)
1828 		return;
1829 
1830 	uret_msr->load_into_hardware = load_into_hardware;
1831 }
1832 
1833 /*
1834  * Configuring user return MSRs to automatically save, load, and restore MSRs
1835  * that need to be shoved into hardware when running the guest.  Note, omitting
1836  * an MSR here does _NOT_ mean it's not emulated, only that it will not be
1837  * loaded into hardware when running the guest.
1838  */
1839 static void vmx_setup_uret_msrs(struct vcpu_vmx *vmx)
1840 {
1841 #ifdef CONFIG_X86_64
1842 	bool load_syscall_msrs;
1843 
1844 	/*
1845 	 * The SYSCALL MSRs are only needed on long mode guests, and only
1846 	 * when EFER.SCE is set.
1847 	 */
1848 	load_syscall_msrs = is_long_mode(&vmx->vcpu) &&
1849 			    (vmx->vcpu.arch.efer & EFER_SCE);
1850 
1851 	vmx_setup_uret_msr(vmx, MSR_STAR, load_syscall_msrs);
1852 	vmx_setup_uret_msr(vmx, MSR_LSTAR, load_syscall_msrs);
1853 	vmx_setup_uret_msr(vmx, MSR_SYSCALL_MASK, load_syscall_msrs);
1854 #endif
1855 	vmx_setup_uret_msr(vmx, MSR_EFER, update_transition_efer(vmx));
1856 
1857 	vmx_setup_uret_msr(vmx, MSR_TSC_AUX,
1858 			   guest_cpuid_has(&vmx->vcpu, X86_FEATURE_RDTSCP) ||
1859 			   guest_cpuid_has(&vmx->vcpu, X86_FEATURE_RDPID));
1860 
1861 	/*
1862 	 * hle=0, rtm=0, tsx_ctrl=1 can be found with some combinations of new
1863 	 * kernel and old userspace.  If those guests run on a tsx=off host, do
1864 	 * allow guests to use TSX_CTRL, but don't change the value in hardware
1865 	 * so that TSX remains always disabled.
1866 	 */
1867 	vmx_setup_uret_msr(vmx, MSR_IA32_TSX_CTRL, boot_cpu_has(X86_FEATURE_RTM));
1868 
1869 	/*
1870 	 * The set of MSRs to load may have changed, reload MSRs before the
1871 	 * next VM-Enter.
1872 	 */
1873 	vmx->guest_uret_msrs_loaded = false;
1874 }
1875 
1876 u64 vmx_get_l2_tsc_offset(struct kvm_vcpu *vcpu)
1877 {
1878 	struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
1879 
1880 	if (nested_cpu_has(vmcs12, CPU_BASED_USE_TSC_OFFSETTING))
1881 		return vmcs12->tsc_offset;
1882 
1883 	return 0;
1884 }
1885 
1886 u64 vmx_get_l2_tsc_multiplier(struct kvm_vcpu *vcpu)
1887 {
1888 	struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
1889 
1890 	if (nested_cpu_has(vmcs12, CPU_BASED_USE_TSC_OFFSETTING) &&
1891 	    nested_cpu_has2(vmcs12, SECONDARY_EXEC_TSC_SCALING))
1892 		return vmcs12->tsc_multiplier;
1893 
1894 	return kvm_caps.default_tsc_scaling_ratio;
1895 }
1896 
1897 static void vmx_write_tsc_offset(struct kvm_vcpu *vcpu, u64 offset)
1898 {
1899 	vmcs_write64(TSC_OFFSET, offset);
1900 }
1901 
1902 static void vmx_write_tsc_multiplier(struct kvm_vcpu *vcpu, u64 multiplier)
1903 {
1904 	vmcs_write64(TSC_MULTIPLIER, multiplier);
1905 }
1906 
1907 /*
1908  * nested_vmx_allowed() checks whether a guest should be allowed to use VMX
1909  * instructions and MSRs (i.e., nested VMX). Nested VMX is disabled for
1910  * all guests if the "nested" module option is off, and can also be disabled
1911  * for a single guest by disabling its VMX cpuid bit.
1912  */
1913 bool nested_vmx_allowed(struct kvm_vcpu *vcpu)
1914 {
1915 	return nested && guest_cpuid_has(vcpu, X86_FEATURE_VMX);
1916 }
1917 
1918 /*
1919  * Userspace is allowed to set any supported IA32_FEATURE_CONTROL regardless of
1920  * guest CPUID.  Note, KVM allows userspace to set "VMX in SMX" to maintain
1921  * backwards compatibility even though KVM doesn't support emulating SMX.  And
1922  * because userspace set "VMX in SMX", the guest must also be allowed to set it,
1923  * e.g. if the MSR is left unlocked and the guest does a RMW operation.
1924  */
1925 #define KVM_SUPPORTED_FEATURE_CONTROL  (FEAT_CTL_LOCKED			 | \
1926 					FEAT_CTL_VMX_ENABLED_INSIDE_SMX	 | \
1927 					FEAT_CTL_VMX_ENABLED_OUTSIDE_SMX | \
1928 					FEAT_CTL_SGX_LC_ENABLED		 | \
1929 					FEAT_CTL_SGX_ENABLED		 | \
1930 					FEAT_CTL_LMCE_ENABLED)
1931 
1932 static inline bool is_vmx_feature_control_msr_valid(struct vcpu_vmx *vmx,
1933 						    struct msr_data *msr)
1934 {
1935 	uint64_t valid_bits;
1936 
1937 	/*
1938 	 * Ensure KVM_SUPPORTED_FEATURE_CONTROL is updated when new bits are
1939 	 * exposed to the guest.
1940 	 */
1941 	WARN_ON_ONCE(vmx->msr_ia32_feature_control_valid_bits &
1942 		     ~KVM_SUPPORTED_FEATURE_CONTROL);
1943 
1944 	if (!msr->host_initiated &&
1945 	    (vmx->msr_ia32_feature_control & FEAT_CTL_LOCKED))
1946 		return false;
1947 
1948 	if (msr->host_initiated)
1949 		valid_bits = KVM_SUPPORTED_FEATURE_CONTROL;
1950 	else
1951 		valid_bits = vmx->msr_ia32_feature_control_valid_bits;
1952 
1953 	return !(msr->data & ~valid_bits);
1954 }
1955 
1956 static int vmx_get_msr_feature(struct kvm_msr_entry *msr)
1957 {
1958 	switch (msr->index) {
1959 	case KVM_FIRST_EMULATED_VMX_MSR ... KVM_LAST_EMULATED_VMX_MSR:
1960 		if (!nested)
1961 			return 1;
1962 		return vmx_get_vmx_msr(&vmcs_config.nested, msr->index, &msr->data);
1963 	default:
1964 		return KVM_MSR_RET_INVALID;
1965 	}
1966 }
1967 
1968 /*
1969  * Reads an msr value (of 'msr_info->index') into 'msr_info->data'.
1970  * Returns 0 on success, non-0 otherwise.
1971  * Assumes vcpu_load() was already called.
1972  */
1973 static int vmx_get_msr(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
1974 {
1975 	struct vcpu_vmx *vmx = to_vmx(vcpu);
1976 	struct vmx_uret_msr *msr;
1977 	u32 index;
1978 
1979 	switch (msr_info->index) {
1980 #ifdef CONFIG_X86_64
1981 	case MSR_FS_BASE:
1982 		msr_info->data = vmcs_readl(GUEST_FS_BASE);
1983 		break;
1984 	case MSR_GS_BASE:
1985 		msr_info->data = vmcs_readl(GUEST_GS_BASE);
1986 		break;
1987 	case MSR_KERNEL_GS_BASE:
1988 		msr_info->data = vmx_read_guest_kernel_gs_base(vmx);
1989 		break;
1990 #endif
1991 	case MSR_EFER:
1992 		return kvm_get_msr_common(vcpu, msr_info);
1993 	case MSR_IA32_TSX_CTRL:
1994 		if (!msr_info->host_initiated &&
1995 		    !(vcpu->arch.arch_capabilities & ARCH_CAP_TSX_CTRL_MSR))
1996 			return 1;
1997 		goto find_uret_msr;
1998 	case MSR_IA32_UMWAIT_CONTROL:
1999 		if (!msr_info->host_initiated && !vmx_has_waitpkg(vmx))
2000 			return 1;
2001 
2002 		msr_info->data = vmx->msr_ia32_umwait_control;
2003 		break;
2004 	case MSR_IA32_SPEC_CTRL:
2005 		if (!msr_info->host_initiated &&
2006 		    !guest_has_spec_ctrl_msr(vcpu))
2007 			return 1;
2008 
2009 		msr_info->data = to_vmx(vcpu)->spec_ctrl;
2010 		break;
2011 	case MSR_IA32_SYSENTER_CS:
2012 		msr_info->data = vmcs_read32(GUEST_SYSENTER_CS);
2013 		break;
2014 	case MSR_IA32_SYSENTER_EIP:
2015 		msr_info->data = vmcs_readl(GUEST_SYSENTER_EIP);
2016 		break;
2017 	case MSR_IA32_SYSENTER_ESP:
2018 		msr_info->data = vmcs_readl(GUEST_SYSENTER_ESP);
2019 		break;
2020 	case MSR_IA32_BNDCFGS:
2021 		if (!kvm_mpx_supported() ||
2022 		    (!msr_info->host_initiated &&
2023 		     !guest_cpuid_has(vcpu, X86_FEATURE_MPX)))
2024 			return 1;
2025 		msr_info->data = vmcs_read64(GUEST_BNDCFGS);
2026 		break;
2027 	case MSR_IA32_MCG_EXT_CTL:
2028 		if (!msr_info->host_initiated &&
2029 		    !(vmx->msr_ia32_feature_control &
2030 		      FEAT_CTL_LMCE_ENABLED))
2031 			return 1;
2032 		msr_info->data = vcpu->arch.mcg_ext_ctl;
2033 		break;
2034 	case MSR_IA32_FEAT_CTL:
2035 		msr_info->data = vmx->msr_ia32_feature_control;
2036 		break;
2037 	case MSR_IA32_SGXLEPUBKEYHASH0 ... MSR_IA32_SGXLEPUBKEYHASH3:
2038 		if (!msr_info->host_initiated &&
2039 		    !guest_cpuid_has(vcpu, X86_FEATURE_SGX_LC))
2040 			return 1;
2041 		msr_info->data = to_vmx(vcpu)->msr_ia32_sgxlepubkeyhash
2042 			[msr_info->index - MSR_IA32_SGXLEPUBKEYHASH0];
2043 		break;
2044 	case KVM_FIRST_EMULATED_VMX_MSR ... KVM_LAST_EMULATED_VMX_MSR:
2045 		if (!nested_vmx_allowed(vcpu))
2046 			return 1;
2047 		if (vmx_get_vmx_msr(&vmx->nested.msrs, msr_info->index,
2048 				    &msr_info->data))
2049 			return 1;
2050 		/*
2051 		 * Enlightened VMCS v1 doesn't have certain VMCS fields but
2052 		 * instead of just ignoring the features, different Hyper-V
2053 		 * versions are either trying to use them and fail or do some
2054 		 * sanity checking and refuse to boot. Filter all unsupported
2055 		 * features out.
2056 		 */
2057 		if (!msr_info->host_initiated && guest_cpuid_has_evmcs(vcpu))
2058 			nested_evmcs_filter_control_msr(vcpu, msr_info->index,
2059 							&msr_info->data);
2060 		break;
2061 	case MSR_IA32_RTIT_CTL:
2062 		if (!vmx_pt_mode_is_host_guest())
2063 			return 1;
2064 		msr_info->data = vmx->pt_desc.guest.ctl;
2065 		break;
2066 	case MSR_IA32_RTIT_STATUS:
2067 		if (!vmx_pt_mode_is_host_guest())
2068 			return 1;
2069 		msr_info->data = vmx->pt_desc.guest.status;
2070 		break;
2071 	case MSR_IA32_RTIT_CR3_MATCH:
2072 		if (!vmx_pt_mode_is_host_guest() ||
2073 			!intel_pt_validate_cap(vmx->pt_desc.caps,
2074 						PT_CAP_cr3_filtering))
2075 			return 1;
2076 		msr_info->data = vmx->pt_desc.guest.cr3_match;
2077 		break;
2078 	case MSR_IA32_RTIT_OUTPUT_BASE:
2079 		if (!vmx_pt_mode_is_host_guest() ||
2080 			(!intel_pt_validate_cap(vmx->pt_desc.caps,
2081 					PT_CAP_topa_output) &&
2082 			 !intel_pt_validate_cap(vmx->pt_desc.caps,
2083 					PT_CAP_single_range_output)))
2084 			return 1;
2085 		msr_info->data = vmx->pt_desc.guest.output_base;
2086 		break;
2087 	case MSR_IA32_RTIT_OUTPUT_MASK:
2088 		if (!vmx_pt_mode_is_host_guest() ||
2089 			(!intel_pt_validate_cap(vmx->pt_desc.caps,
2090 					PT_CAP_topa_output) &&
2091 			 !intel_pt_validate_cap(vmx->pt_desc.caps,
2092 					PT_CAP_single_range_output)))
2093 			return 1;
2094 		msr_info->data = vmx->pt_desc.guest.output_mask;
2095 		break;
2096 	case MSR_IA32_RTIT_ADDR0_A ... MSR_IA32_RTIT_ADDR3_B:
2097 		index = msr_info->index - MSR_IA32_RTIT_ADDR0_A;
2098 		if (!vmx_pt_mode_is_host_guest() ||
2099 		    (index >= 2 * vmx->pt_desc.num_address_ranges))
2100 			return 1;
2101 		if (index % 2)
2102 			msr_info->data = vmx->pt_desc.guest.addr_b[index / 2];
2103 		else
2104 			msr_info->data = vmx->pt_desc.guest.addr_a[index / 2];
2105 		break;
2106 	case MSR_IA32_DEBUGCTLMSR:
2107 		msr_info->data = vmcs_read64(GUEST_IA32_DEBUGCTL);
2108 		break;
2109 	default:
2110 	find_uret_msr:
2111 		msr = vmx_find_uret_msr(vmx, msr_info->index);
2112 		if (msr) {
2113 			msr_info->data = msr->data;
2114 			break;
2115 		}
2116 		return kvm_get_msr_common(vcpu, msr_info);
2117 	}
2118 
2119 	return 0;
2120 }
2121 
2122 static u64 nested_vmx_truncate_sysenter_addr(struct kvm_vcpu *vcpu,
2123 						    u64 data)
2124 {
2125 #ifdef CONFIG_X86_64
2126 	if (!guest_cpuid_has(vcpu, X86_FEATURE_LM))
2127 		return (u32)data;
2128 #endif
2129 	return (unsigned long)data;
2130 }
2131 
2132 static u64 vmx_get_supported_debugctl(struct kvm_vcpu *vcpu, bool host_initiated)
2133 {
2134 	u64 debugctl = 0;
2135 
2136 	if (boot_cpu_has(X86_FEATURE_BUS_LOCK_DETECT) &&
2137 	    (host_initiated || guest_cpuid_has(vcpu, X86_FEATURE_BUS_LOCK_DETECT)))
2138 		debugctl |= DEBUGCTLMSR_BUS_LOCK_DETECT;
2139 
2140 	if ((kvm_caps.supported_perf_cap & PMU_CAP_LBR_FMT) &&
2141 	    (host_initiated || intel_pmu_lbr_is_enabled(vcpu)))
2142 		debugctl |= DEBUGCTLMSR_LBR | DEBUGCTLMSR_FREEZE_LBRS_ON_PMI;
2143 
2144 	return debugctl;
2145 }
2146 
2147 /*
2148  * Writes msr value into the appropriate "register".
2149  * Returns 0 on success, non-0 otherwise.
2150  * Assumes vcpu_load() was already called.
2151  */
2152 static int vmx_set_msr(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
2153 {
2154 	struct vcpu_vmx *vmx = to_vmx(vcpu);
2155 	struct vmx_uret_msr *msr;
2156 	int ret = 0;
2157 	u32 msr_index = msr_info->index;
2158 	u64 data = msr_info->data;
2159 	u32 index;
2160 
2161 	switch (msr_index) {
2162 	case MSR_EFER:
2163 		ret = kvm_set_msr_common(vcpu, msr_info);
2164 		break;
2165 #ifdef CONFIG_X86_64
2166 	case MSR_FS_BASE:
2167 		vmx_segment_cache_clear(vmx);
2168 		vmcs_writel(GUEST_FS_BASE, data);
2169 		break;
2170 	case MSR_GS_BASE:
2171 		vmx_segment_cache_clear(vmx);
2172 		vmcs_writel(GUEST_GS_BASE, data);
2173 		break;
2174 	case MSR_KERNEL_GS_BASE:
2175 		vmx_write_guest_kernel_gs_base(vmx, data);
2176 		break;
2177 	case MSR_IA32_XFD:
2178 		ret = kvm_set_msr_common(vcpu, msr_info);
2179 		/*
2180 		 * Always intercepting WRMSR could incur non-negligible
2181 		 * overhead given xfd might be changed frequently in
2182 		 * guest context switch. Disable write interception
2183 		 * upon the first write with a non-zero value (indicating
2184 		 * potential usage on dynamic xfeatures). Also update
2185 		 * exception bitmap to trap #NM for proper virtualization
2186 		 * of guest xfd_err.
2187 		 */
2188 		if (!ret && data) {
2189 			vmx_disable_intercept_for_msr(vcpu, MSR_IA32_XFD,
2190 						      MSR_TYPE_RW);
2191 			vcpu->arch.xfd_no_write_intercept = true;
2192 			vmx_update_exception_bitmap(vcpu);
2193 		}
2194 		break;
2195 #endif
2196 	case MSR_IA32_SYSENTER_CS:
2197 		if (is_guest_mode(vcpu))
2198 			get_vmcs12(vcpu)->guest_sysenter_cs = data;
2199 		vmcs_write32(GUEST_SYSENTER_CS, data);
2200 		break;
2201 	case MSR_IA32_SYSENTER_EIP:
2202 		if (is_guest_mode(vcpu)) {
2203 			data = nested_vmx_truncate_sysenter_addr(vcpu, data);
2204 			get_vmcs12(vcpu)->guest_sysenter_eip = data;
2205 		}
2206 		vmcs_writel(GUEST_SYSENTER_EIP, data);
2207 		break;
2208 	case MSR_IA32_SYSENTER_ESP:
2209 		if (is_guest_mode(vcpu)) {
2210 			data = nested_vmx_truncate_sysenter_addr(vcpu, data);
2211 			get_vmcs12(vcpu)->guest_sysenter_esp = data;
2212 		}
2213 		vmcs_writel(GUEST_SYSENTER_ESP, data);
2214 		break;
2215 	case MSR_IA32_DEBUGCTLMSR: {
2216 		u64 invalid;
2217 
2218 		invalid = data & ~vmx_get_supported_debugctl(vcpu, msr_info->host_initiated);
2219 		if (invalid & (DEBUGCTLMSR_BTF|DEBUGCTLMSR_LBR)) {
2220 			kvm_pr_unimpl_wrmsr(vcpu, msr_index, data);
2221 			data &= ~(DEBUGCTLMSR_BTF|DEBUGCTLMSR_LBR);
2222 			invalid &= ~(DEBUGCTLMSR_BTF|DEBUGCTLMSR_LBR);
2223 		}
2224 
2225 		if (invalid)
2226 			return 1;
2227 
2228 		if (is_guest_mode(vcpu) && get_vmcs12(vcpu)->vm_exit_controls &
2229 						VM_EXIT_SAVE_DEBUG_CONTROLS)
2230 			get_vmcs12(vcpu)->guest_ia32_debugctl = data;
2231 
2232 		vmcs_write64(GUEST_IA32_DEBUGCTL, data);
2233 		if (intel_pmu_lbr_is_enabled(vcpu) && !to_vmx(vcpu)->lbr_desc.event &&
2234 		    (data & DEBUGCTLMSR_LBR))
2235 			intel_pmu_create_guest_lbr_event(vcpu);
2236 		return 0;
2237 	}
2238 	case MSR_IA32_BNDCFGS:
2239 		if (!kvm_mpx_supported() ||
2240 		    (!msr_info->host_initiated &&
2241 		     !guest_cpuid_has(vcpu, X86_FEATURE_MPX)))
2242 			return 1;
2243 		if (is_noncanonical_address(data & PAGE_MASK, vcpu) ||
2244 		    (data & MSR_IA32_BNDCFGS_RSVD))
2245 			return 1;
2246 
2247 		if (is_guest_mode(vcpu) &&
2248 		    ((vmx->nested.msrs.entry_ctls_high & VM_ENTRY_LOAD_BNDCFGS) ||
2249 		     (vmx->nested.msrs.exit_ctls_high & VM_EXIT_CLEAR_BNDCFGS)))
2250 			get_vmcs12(vcpu)->guest_bndcfgs = data;
2251 
2252 		vmcs_write64(GUEST_BNDCFGS, data);
2253 		break;
2254 	case MSR_IA32_UMWAIT_CONTROL:
2255 		if (!msr_info->host_initiated && !vmx_has_waitpkg(vmx))
2256 			return 1;
2257 
2258 		/* The reserved bit 1 and non-32 bit [63:32] should be zero */
2259 		if (data & (BIT_ULL(1) | GENMASK_ULL(63, 32)))
2260 			return 1;
2261 
2262 		vmx->msr_ia32_umwait_control = data;
2263 		break;
2264 	case MSR_IA32_SPEC_CTRL:
2265 		if (!msr_info->host_initiated &&
2266 		    !guest_has_spec_ctrl_msr(vcpu))
2267 			return 1;
2268 
2269 		if (kvm_spec_ctrl_test_value(data))
2270 			return 1;
2271 
2272 		vmx->spec_ctrl = data;
2273 		if (!data)
2274 			break;
2275 
2276 		/*
2277 		 * For non-nested:
2278 		 * When it's written (to non-zero) for the first time, pass
2279 		 * it through.
2280 		 *
2281 		 * For nested:
2282 		 * The handling of the MSR bitmap for L2 guests is done in
2283 		 * nested_vmx_prepare_msr_bitmap. We should not touch the
2284 		 * vmcs02.msr_bitmap here since it gets completely overwritten
2285 		 * in the merging. We update the vmcs01 here for L1 as well
2286 		 * since it will end up touching the MSR anyway now.
2287 		 */
2288 		vmx_disable_intercept_for_msr(vcpu,
2289 					      MSR_IA32_SPEC_CTRL,
2290 					      MSR_TYPE_RW);
2291 		break;
2292 	case MSR_IA32_TSX_CTRL:
2293 		if (!msr_info->host_initiated &&
2294 		    !(vcpu->arch.arch_capabilities & ARCH_CAP_TSX_CTRL_MSR))
2295 			return 1;
2296 		if (data & ~(TSX_CTRL_RTM_DISABLE | TSX_CTRL_CPUID_CLEAR))
2297 			return 1;
2298 		goto find_uret_msr;
2299 	case MSR_IA32_CR_PAT:
2300 		ret = kvm_set_msr_common(vcpu, msr_info);
2301 		if (ret)
2302 			break;
2303 
2304 		if (is_guest_mode(vcpu) &&
2305 		    get_vmcs12(vcpu)->vm_exit_controls & VM_EXIT_SAVE_IA32_PAT)
2306 			get_vmcs12(vcpu)->guest_ia32_pat = data;
2307 
2308 		if (vmcs_config.vmentry_ctrl & VM_ENTRY_LOAD_IA32_PAT)
2309 			vmcs_write64(GUEST_IA32_PAT, data);
2310 		break;
2311 	case MSR_IA32_MCG_EXT_CTL:
2312 		if ((!msr_info->host_initiated &&
2313 		     !(to_vmx(vcpu)->msr_ia32_feature_control &
2314 		       FEAT_CTL_LMCE_ENABLED)) ||
2315 		    (data & ~MCG_EXT_CTL_LMCE_EN))
2316 			return 1;
2317 		vcpu->arch.mcg_ext_ctl = data;
2318 		break;
2319 	case MSR_IA32_FEAT_CTL:
2320 		if (!is_vmx_feature_control_msr_valid(vmx, msr_info))
2321 			return 1;
2322 
2323 		vmx->msr_ia32_feature_control = data;
2324 		if (msr_info->host_initiated && data == 0)
2325 			vmx_leave_nested(vcpu);
2326 
2327 		/* SGX may be enabled/disabled by guest's firmware */
2328 		vmx_write_encls_bitmap(vcpu, NULL);
2329 		break;
2330 	case MSR_IA32_SGXLEPUBKEYHASH0 ... MSR_IA32_SGXLEPUBKEYHASH3:
2331 		/*
2332 		 * On real hardware, the LE hash MSRs are writable before
2333 		 * the firmware sets bit 0 in MSR 0x7a ("activating" SGX),
2334 		 * at which point SGX related bits in IA32_FEATURE_CONTROL
2335 		 * become writable.
2336 		 *
2337 		 * KVM does not emulate SGX activation for simplicity, so
2338 		 * allow writes to the LE hash MSRs if IA32_FEATURE_CONTROL
2339 		 * is unlocked.  This is technically not architectural
2340 		 * behavior, but it's close enough.
2341 		 */
2342 		if (!msr_info->host_initiated &&
2343 		    (!guest_cpuid_has(vcpu, X86_FEATURE_SGX_LC) ||
2344 		    ((vmx->msr_ia32_feature_control & FEAT_CTL_LOCKED) &&
2345 		    !(vmx->msr_ia32_feature_control & FEAT_CTL_SGX_LC_ENABLED))))
2346 			return 1;
2347 		vmx->msr_ia32_sgxlepubkeyhash
2348 			[msr_index - MSR_IA32_SGXLEPUBKEYHASH0] = data;
2349 		break;
2350 	case KVM_FIRST_EMULATED_VMX_MSR ... KVM_LAST_EMULATED_VMX_MSR:
2351 		if (!msr_info->host_initiated)
2352 			return 1; /* they are read-only */
2353 		if (!nested_vmx_allowed(vcpu))
2354 			return 1;
2355 		return vmx_set_vmx_msr(vcpu, msr_index, data);
2356 	case MSR_IA32_RTIT_CTL:
2357 		if (!vmx_pt_mode_is_host_guest() ||
2358 			vmx_rtit_ctl_check(vcpu, data) ||
2359 			vmx->nested.vmxon)
2360 			return 1;
2361 		vmcs_write64(GUEST_IA32_RTIT_CTL, data);
2362 		vmx->pt_desc.guest.ctl = data;
2363 		pt_update_intercept_for_msr(vcpu);
2364 		break;
2365 	case MSR_IA32_RTIT_STATUS:
2366 		if (!pt_can_write_msr(vmx))
2367 			return 1;
2368 		if (data & MSR_IA32_RTIT_STATUS_MASK)
2369 			return 1;
2370 		vmx->pt_desc.guest.status = data;
2371 		break;
2372 	case MSR_IA32_RTIT_CR3_MATCH:
2373 		if (!pt_can_write_msr(vmx))
2374 			return 1;
2375 		if (!intel_pt_validate_cap(vmx->pt_desc.caps,
2376 					   PT_CAP_cr3_filtering))
2377 			return 1;
2378 		vmx->pt_desc.guest.cr3_match = data;
2379 		break;
2380 	case MSR_IA32_RTIT_OUTPUT_BASE:
2381 		if (!pt_can_write_msr(vmx))
2382 			return 1;
2383 		if (!intel_pt_validate_cap(vmx->pt_desc.caps,
2384 					   PT_CAP_topa_output) &&
2385 		    !intel_pt_validate_cap(vmx->pt_desc.caps,
2386 					   PT_CAP_single_range_output))
2387 			return 1;
2388 		if (!pt_output_base_valid(vcpu, data))
2389 			return 1;
2390 		vmx->pt_desc.guest.output_base = data;
2391 		break;
2392 	case MSR_IA32_RTIT_OUTPUT_MASK:
2393 		if (!pt_can_write_msr(vmx))
2394 			return 1;
2395 		if (!intel_pt_validate_cap(vmx->pt_desc.caps,
2396 					   PT_CAP_topa_output) &&
2397 		    !intel_pt_validate_cap(vmx->pt_desc.caps,
2398 					   PT_CAP_single_range_output))
2399 			return 1;
2400 		vmx->pt_desc.guest.output_mask = data;
2401 		break;
2402 	case MSR_IA32_RTIT_ADDR0_A ... MSR_IA32_RTIT_ADDR3_B:
2403 		if (!pt_can_write_msr(vmx))
2404 			return 1;
2405 		index = msr_info->index - MSR_IA32_RTIT_ADDR0_A;
2406 		if (index >= 2 * vmx->pt_desc.num_address_ranges)
2407 			return 1;
2408 		if (is_noncanonical_address(data, vcpu))
2409 			return 1;
2410 		if (index % 2)
2411 			vmx->pt_desc.guest.addr_b[index / 2] = data;
2412 		else
2413 			vmx->pt_desc.guest.addr_a[index / 2] = data;
2414 		break;
2415 	case MSR_IA32_PERF_CAPABILITIES:
2416 		if (data && !vcpu_to_pmu(vcpu)->version)
2417 			return 1;
2418 		if (data & PMU_CAP_LBR_FMT) {
2419 			if ((data & PMU_CAP_LBR_FMT) !=
2420 			    (kvm_caps.supported_perf_cap & PMU_CAP_LBR_FMT))
2421 				return 1;
2422 			if (!cpuid_model_is_consistent(vcpu))
2423 				return 1;
2424 		}
2425 		if (data & PERF_CAP_PEBS_FORMAT) {
2426 			if ((data & PERF_CAP_PEBS_MASK) !=
2427 			    (kvm_caps.supported_perf_cap & PERF_CAP_PEBS_MASK))
2428 				return 1;
2429 			if (!guest_cpuid_has(vcpu, X86_FEATURE_DS))
2430 				return 1;
2431 			if (!guest_cpuid_has(vcpu, X86_FEATURE_DTES64))
2432 				return 1;
2433 			if (!cpuid_model_is_consistent(vcpu))
2434 				return 1;
2435 		}
2436 		ret = kvm_set_msr_common(vcpu, msr_info);
2437 		break;
2438 
2439 	default:
2440 	find_uret_msr:
2441 		msr = vmx_find_uret_msr(vmx, msr_index);
2442 		if (msr)
2443 			ret = vmx_set_guest_uret_msr(vmx, msr, data);
2444 		else
2445 			ret = kvm_set_msr_common(vcpu, msr_info);
2446 	}
2447 
2448 	/* FB_CLEAR may have changed, also update the FB_CLEAR_DIS behavior */
2449 	if (msr_index == MSR_IA32_ARCH_CAPABILITIES)
2450 		vmx_update_fb_clear_dis(vcpu, vmx);
2451 
2452 	return ret;
2453 }
2454 
2455 static void vmx_cache_reg(struct kvm_vcpu *vcpu, enum kvm_reg reg)
2456 {
2457 	unsigned long guest_owned_bits;
2458 
2459 	kvm_register_mark_available(vcpu, reg);
2460 
2461 	switch (reg) {
2462 	case VCPU_REGS_RSP:
2463 		vcpu->arch.regs[VCPU_REGS_RSP] = vmcs_readl(GUEST_RSP);
2464 		break;
2465 	case VCPU_REGS_RIP:
2466 		vcpu->arch.regs[VCPU_REGS_RIP] = vmcs_readl(GUEST_RIP);
2467 		break;
2468 	case VCPU_EXREG_PDPTR:
2469 		if (enable_ept)
2470 			ept_save_pdptrs(vcpu);
2471 		break;
2472 	case VCPU_EXREG_CR0:
2473 		guest_owned_bits = vcpu->arch.cr0_guest_owned_bits;
2474 
2475 		vcpu->arch.cr0 &= ~guest_owned_bits;
2476 		vcpu->arch.cr0 |= vmcs_readl(GUEST_CR0) & guest_owned_bits;
2477 		break;
2478 	case VCPU_EXREG_CR3:
2479 		/*
2480 		 * When intercepting CR3 loads, e.g. for shadowing paging, KVM's
2481 		 * CR3 is loaded into hardware, not the guest's CR3.
2482 		 */
2483 		if (!(exec_controls_get(to_vmx(vcpu)) & CPU_BASED_CR3_LOAD_EXITING))
2484 			vcpu->arch.cr3 = vmcs_readl(GUEST_CR3);
2485 		break;
2486 	case VCPU_EXREG_CR4:
2487 		guest_owned_bits = vcpu->arch.cr4_guest_owned_bits;
2488 
2489 		vcpu->arch.cr4 &= ~guest_owned_bits;
2490 		vcpu->arch.cr4 |= vmcs_readl(GUEST_CR4) & guest_owned_bits;
2491 		break;
2492 	default:
2493 		KVM_BUG_ON(1, vcpu->kvm);
2494 		break;
2495 	}
2496 }
2497 
2498 /*
2499  * There is no X86_FEATURE for SGX yet, but anyway we need to query CPUID
2500  * directly instead of going through cpu_has(), to ensure KVM is trapping
2501  * ENCLS whenever it's supported in hardware.  It does not matter whether
2502  * the host OS supports or has enabled SGX.
2503  */
2504 static bool cpu_has_sgx(void)
2505 {
2506 	return cpuid_eax(0) >= 0x12 && (cpuid_eax(0x12) & BIT(0));
2507 }
2508 
2509 /*
2510  * Some cpus support VM_{ENTRY,EXIT}_IA32_PERF_GLOBAL_CTRL but they
2511  * can't be used due to errata where VM Exit may incorrectly clear
2512  * IA32_PERF_GLOBAL_CTRL[34:32]. Work around the errata by using the
2513  * MSR load mechanism to switch IA32_PERF_GLOBAL_CTRL.
2514  */
2515 static bool cpu_has_perf_global_ctrl_bug(void)
2516 {
2517 	if (boot_cpu_data.x86 == 0x6) {
2518 		switch (boot_cpu_data.x86_model) {
2519 		case INTEL_FAM6_NEHALEM_EP:	/* AAK155 */
2520 		case INTEL_FAM6_NEHALEM:	/* AAP115 */
2521 		case INTEL_FAM6_WESTMERE:	/* AAT100 */
2522 		case INTEL_FAM6_WESTMERE_EP:	/* BC86,AAY89,BD102 */
2523 		case INTEL_FAM6_NEHALEM_EX:	/* BA97 */
2524 			return true;
2525 		default:
2526 			break;
2527 		}
2528 	}
2529 
2530 	return false;
2531 }
2532 
2533 static int adjust_vmx_controls(u32 ctl_min, u32 ctl_opt, u32 msr, u32 *result)
2534 {
2535 	u32 vmx_msr_low, vmx_msr_high;
2536 	u32 ctl = ctl_min | ctl_opt;
2537 
2538 	rdmsr(msr, vmx_msr_low, vmx_msr_high);
2539 
2540 	ctl &= vmx_msr_high; /* bit == 0 in high word ==> must be zero */
2541 	ctl |= vmx_msr_low;  /* bit == 1 in low word  ==> must be one  */
2542 
2543 	/* Ensure minimum (required) set of control bits are supported. */
2544 	if (ctl_min & ~ctl)
2545 		return -EIO;
2546 
2547 	*result = ctl;
2548 	return 0;
2549 }
2550 
2551 static u64 adjust_vmx_controls64(u64 ctl_opt, u32 msr)
2552 {
2553 	u64 allowed;
2554 
2555 	rdmsrl(msr, allowed);
2556 
2557 	return  ctl_opt & allowed;
2558 }
2559 
2560 static int setup_vmcs_config(struct vmcs_config *vmcs_conf,
2561 			     struct vmx_capability *vmx_cap)
2562 {
2563 	u32 vmx_msr_low, vmx_msr_high;
2564 	u32 _pin_based_exec_control = 0;
2565 	u32 _cpu_based_exec_control = 0;
2566 	u32 _cpu_based_2nd_exec_control = 0;
2567 	u64 _cpu_based_3rd_exec_control = 0;
2568 	u32 _vmexit_control = 0;
2569 	u32 _vmentry_control = 0;
2570 	u64 misc_msr;
2571 	int i;
2572 
2573 	/*
2574 	 * LOAD/SAVE_DEBUG_CONTROLS are absent because both are mandatory.
2575 	 * SAVE_IA32_PAT and SAVE_IA32_EFER are absent because KVM always
2576 	 * intercepts writes to PAT and EFER, i.e. never enables those controls.
2577 	 */
2578 	struct {
2579 		u32 entry_control;
2580 		u32 exit_control;
2581 	} const vmcs_entry_exit_pairs[] = {
2582 		{ VM_ENTRY_LOAD_IA32_PERF_GLOBAL_CTRL,	VM_EXIT_LOAD_IA32_PERF_GLOBAL_CTRL },
2583 		{ VM_ENTRY_LOAD_IA32_PAT,		VM_EXIT_LOAD_IA32_PAT },
2584 		{ VM_ENTRY_LOAD_IA32_EFER,		VM_EXIT_LOAD_IA32_EFER },
2585 		{ VM_ENTRY_LOAD_BNDCFGS,		VM_EXIT_CLEAR_BNDCFGS },
2586 		{ VM_ENTRY_LOAD_IA32_RTIT_CTL,		VM_EXIT_CLEAR_IA32_RTIT_CTL },
2587 	};
2588 
2589 	memset(vmcs_conf, 0, sizeof(*vmcs_conf));
2590 
2591 	if (adjust_vmx_controls(KVM_REQUIRED_VMX_CPU_BASED_VM_EXEC_CONTROL,
2592 				KVM_OPTIONAL_VMX_CPU_BASED_VM_EXEC_CONTROL,
2593 				MSR_IA32_VMX_PROCBASED_CTLS,
2594 				&_cpu_based_exec_control))
2595 		return -EIO;
2596 	if (_cpu_based_exec_control & CPU_BASED_ACTIVATE_SECONDARY_CONTROLS) {
2597 		if (adjust_vmx_controls(KVM_REQUIRED_VMX_SECONDARY_VM_EXEC_CONTROL,
2598 					KVM_OPTIONAL_VMX_SECONDARY_VM_EXEC_CONTROL,
2599 					MSR_IA32_VMX_PROCBASED_CTLS2,
2600 					&_cpu_based_2nd_exec_control))
2601 			return -EIO;
2602 	}
2603 #ifndef CONFIG_X86_64
2604 	if (!(_cpu_based_2nd_exec_control &
2605 				SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES))
2606 		_cpu_based_exec_control &= ~CPU_BASED_TPR_SHADOW;
2607 #endif
2608 
2609 	if (!(_cpu_based_exec_control & CPU_BASED_TPR_SHADOW))
2610 		_cpu_based_2nd_exec_control &= ~(
2611 				SECONDARY_EXEC_APIC_REGISTER_VIRT |
2612 				SECONDARY_EXEC_VIRTUALIZE_X2APIC_MODE |
2613 				SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY);
2614 
2615 	rdmsr_safe(MSR_IA32_VMX_EPT_VPID_CAP,
2616 		&vmx_cap->ept, &vmx_cap->vpid);
2617 
2618 	if (!(_cpu_based_2nd_exec_control & SECONDARY_EXEC_ENABLE_EPT) &&
2619 	    vmx_cap->ept) {
2620 		pr_warn_once("EPT CAP should not exist if not support "
2621 				"1-setting enable EPT VM-execution control\n");
2622 
2623 		if (error_on_inconsistent_vmcs_config)
2624 			return -EIO;
2625 
2626 		vmx_cap->ept = 0;
2627 	}
2628 	if (!(_cpu_based_2nd_exec_control & SECONDARY_EXEC_ENABLE_VPID) &&
2629 	    vmx_cap->vpid) {
2630 		pr_warn_once("VPID CAP should not exist if not support "
2631 				"1-setting enable VPID VM-execution control\n");
2632 
2633 		if (error_on_inconsistent_vmcs_config)
2634 			return -EIO;
2635 
2636 		vmx_cap->vpid = 0;
2637 	}
2638 
2639 	if (!cpu_has_sgx())
2640 		_cpu_based_2nd_exec_control &= ~SECONDARY_EXEC_ENCLS_EXITING;
2641 
2642 	if (_cpu_based_exec_control & CPU_BASED_ACTIVATE_TERTIARY_CONTROLS)
2643 		_cpu_based_3rd_exec_control =
2644 			adjust_vmx_controls64(KVM_OPTIONAL_VMX_TERTIARY_VM_EXEC_CONTROL,
2645 					      MSR_IA32_VMX_PROCBASED_CTLS3);
2646 
2647 	if (adjust_vmx_controls(KVM_REQUIRED_VMX_VM_EXIT_CONTROLS,
2648 				KVM_OPTIONAL_VMX_VM_EXIT_CONTROLS,
2649 				MSR_IA32_VMX_EXIT_CTLS,
2650 				&_vmexit_control))
2651 		return -EIO;
2652 
2653 	if (adjust_vmx_controls(KVM_REQUIRED_VMX_PIN_BASED_VM_EXEC_CONTROL,
2654 				KVM_OPTIONAL_VMX_PIN_BASED_VM_EXEC_CONTROL,
2655 				MSR_IA32_VMX_PINBASED_CTLS,
2656 				&_pin_based_exec_control))
2657 		return -EIO;
2658 
2659 	if (cpu_has_broken_vmx_preemption_timer())
2660 		_pin_based_exec_control &= ~PIN_BASED_VMX_PREEMPTION_TIMER;
2661 	if (!(_cpu_based_2nd_exec_control &
2662 		SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY))
2663 		_pin_based_exec_control &= ~PIN_BASED_POSTED_INTR;
2664 
2665 	if (adjust_vmx_controls(KVM_REQUIRED_VMX_VM_ENTRY_CONTROLS,
2666 				KVM_OPTIONAL_VMX_VM_ENTRY_CONTROLS,
2667 				MSR_IA32_VMX_ENTRY_CTLS,
2668 				&_vmentry_control))
2669 		return -EIO;
2670 
2671 	for (i = 0; i < ARRAY_SIZE(vmcs_entry_exit_pairs); i++) {
2672 		u32 n_ctrl = vmcs_entry_exit_pairs[i].entry_control;
2673 		u32 x_ctrl = vmcs_entry_exit_pairs[i].exit_control;
2674 
2675 		if (!(_vmentry_control & n_ctrl) == !(_vmexit_control & x_ctrl))
2676 			continue;
2677 
2678 		pr_warn_once("Inconsistent VM-Entry/VM-Exit pair, entry = %x, exit = %x\n",
2679 			     _vmentry_control & n_ctrl, _vmexit_control & x_ctrl);
2680 
2681 		if (error_on_inconsistent_vmcs_config)
2682 			return -EIO;
2683 
2684 		_vmentry_control &= ~n_ctrl;
2685 		_vmexit_control &= ~x_ctrl;
2686 	}
2687 
2688 	rdmsr(MSR_IA32_VMX_BASIC, vmx_msr_low, vmx_msr_high);
2689 
2690 	/* IA-32 SDM Vol 3B: VMCS size is never greater than 4kB. */
2691 	if ((vmx_msr_high & 0x1fff) > PAGE_SIZE)
2692 		return -EIO;
2693 
2694 #ifdef CONFIG_X86_64
2695 	/* IA-32 SDM Vol 3B: 64-bit CPUs always have VMX_BASIC_MSR[48]==0. */
2696 	if (vmx_msr_high & (1u<<16))
2697 		return -EIO;
2698 #endif
2699 
2700 	/* Require Write-Back (WB) memory type for VMCS accesses. */
2701 	if (((vmx_msr_high >> 18) & 15) != 6)
2702 		return -EIO;
2703 
2704 	rdmsrl(MSR_IA32_VMX_MISC, misc_msr);
2705 
2706 	vmcs_conf->size = vmx_msr_high & 0x1fff;
2707 	vmcs_conf->basic_cap = vmx_msr_high & ~0x1fff;
2708 
2709 	vmcs_conf->revision_id = vmx_msr_low;
2710 
2711 	vmcs_conf->pin_based_exec_ctrl = _pin_based_exec_control;
2712 	vmcs_conf->cpu_based_exec_ctrl = _cpu_based_exec_control;
2713 	vmcs_conf->cpu_based_2nd_exec_ctrl = _cpu_based_2nd_exec_control;
2714 	vmcs_conf->cpu_based_3rd_exec_ctrl = _cpu_based_3rd_exec_control;
2715 	vmcs_conf->vmexit_ctrl         = _vmexit_control;
2716 	vmcs_conf->vmentry_ctrl        = _vmentry_control;
2717 	vmcs_conf->misc	= misc_msr;
2718 
2719 #if IS_ENABLED(CONFIG_HYPERV)
2720 	if (enlightened_vmcs)
2721 		evmcs_sanitize_exec_ctrls(vmcs_conf);
2722 #endif
2723 
2724 	return 0;
2725 }
2726 
2727 static bool kvm_is_vmx_supported(void)
2728 {
2729 	int cpu = raw_smp_processor_id();
2730 
2731 	if (!cpu_has_vmx()) {
2732 		pr_err("VMX not supported by CPU %d\n", cpu);
2733 		return false;
2734 	}
2735 
2736 	if (!this_cpu_has(X86_FEATURE_MSR_IA32_FEAT_CTL) ||
2737 	    !this_cpu_has(X86_FEATURE_VMX)) {
2738 		pr_err("VMX not enabled (by BIOS) in MSR_IA32_FEAT_CTL on CPU %d\n", cpu);
2739 		return false;
2740 	}
2741 
2742 	return true;
2743 }
2744 
2745 static int vmx_check_processor_compat(void)
2746 {
2747 	int cpu = raw_smp_processor_id();
2748 	struct vmcs_config vmcs_conf;
2749 	struct vmx_capability vmx_cap;
2750 
2751 	if (!kvm_is_vmx_supported())
2752 		return -EIO;
2753 
2754 	if (setup_vmcs_config(&vmcs_conf, &vmx_cap) < 0) {
2755 		pr_err("Failed to setup VMCS config on CPU %d\n", cpu);
2756 		return -EIO;
2757 	}
2758 	if (nested)
2759 		nested_vmx_setup_ctls_msrs(&vmcs_conf, vmx_cap.ept);
2760 	if (memcmp(&vmcs_config, &vmcs_conf, sizeof(struct vmcs_config))) {
2761 		pr_err("Inconsistent VMCS config on CPU %d\n", cpu);
2762 		return -EIO;
2763 	}
2764 	return 0;
2765 }
2766 
2767 static int kvm_cpu_vmxon(u64 vmxon_pointer)
2768 {
2769 	u64 msr;
2770 
2771 	cr4_set_bits(X86_CR4_VMXE);
2772 
2773 	asm_volatile_goto("1: vmxon %[vmxon_pointer]\n\t"
2774 			  _ASM_EXTABLE(1b, %l[fault])
2775 			  : : [vmxon_pointer] "m"(vmxon_pointer)
2776 			  : : fault);
2777 	return 0;
2778 
2779 fault:
2780 	WARN_ONCE(1, "VMXON faulted, MSR_IA32_FEAT_CTL (0x3a) = 0x%llx\n",
2781 		  rdmsrl_safe(MSR_IA32_FEAT_CTL, &msr) ? 0xdeadbeef : msr);
2782 	cr4_clear_bits(X86_CR4_VMXE);
2783 
2784 	return -EFAULT;
2785 }
2786 
2787 static int vmx_hardware_enable(void)
2788 {
2789 	int cpu = raw_smp_processor_id();
2790 	u64 phys_addr = __pa(per_cpu(vmxarea, cpu));
2791 	int r;
2792 
2793 	if (cr4_read_shadow() & X86_CR4_VMXE)
2794 		return -EBUSY;
2795 
2796 	/*
2797 	 * This can happen if we hot-added a CPU but failed to allocate
2798 	 * VP assist page for it.
2799 	 */
2800 	if (kvm_is_using_evmcs() && !hv_get_vp_assist_page(cpu))
2801 		return -EFAULT;
2802 
2803 	intel_pt_handle_vmx(1);
2804 
2805 	r = kvm_cpu_vmxon(phys_addr);
2806 	if (r) {
2807 		intel_pt_handle_vmx(0);
2808 		return r;
2809 	}
2810 
2811 	if (enable_ept)
2812 		ept_sync_global();
2813 
2814 	return 0;
2815 }
2816 
2817 static void vmclear_local_loaded_vmcss(void)
2818 {
2819 	int cpu = raw_smp_processor_id();
2820 	struct loaded_vmcs *v, *n;
2821 
2822 	list_for_each_entry_safe(v, n, &per_cpu(loaded_vmcss_on_cpu, cpu),
2823 				 loaded_vmcss_on_cpu_link)
2824 		__loaded_vmcs_clear(v);
2825 }
2826 
2827 static void vmx_hardware_disable(void)
2828 {
2829 	vmclear_local_loaded_vmcss();
2830 
2831 	if (cpu_vmxoff())
2832 		kvm_spurious_fault();
2833 
2834 	hv_reset_evmcs();
2835 
2836 	intel_pt_handle_vmx(0);
2837 }
2838 
2839 struct vmcs *alloc_vmcs_cpu(bool shadow, int cpu, gfp_t flags)
2840 {
2841 	int node = cpu_to_node(cpu);
2842 	struct page *pages;
2843 	struct vmcs *vmcs;
2844 
2845 	pages = __alloc_pages_node(node, flags, 0);
2846 	if (!pages)
2847 		return NULL;
2848 	vmcs = page_address(pages);
2849 	memset(vmcs, 0, vmcs_config.size);
2850 
2851 	/* KVM supports Enlightened VMCS v1 only */
2852 	if (kvm_is_using_evmcs())
2853 		vmcs->hdr.revision_id = KVM_EVMCS_VERSION;
2854 	else
2855 		vmcs->hdr.revision_id = vmcs_config.revision_id;
2856 
2857 	if (shadow)
2858 		vmcs->hdr.shadow_vmcs = 1;
2859 	return vmcs;
2860 }
2861 
2862 void free_vmcs(struct vmcs *vmcs)
2863 {
2864 	free_page((unsigned long)vmcs);
2865 }
2866 
2867 /*
2868  * Free a VMCS, but before that VMCLEAR it on the CPU where it was last loaded
2869  */
2870 void free_loaded_vmcs(struct loaded_vmcs *loaded_vmcs)
2871 {
2872 	if (!loaded_vmcs->vmcs)
2873 		return;
2874 	loaded_vmcs_clear(loaded_vmcs);
2875 	free_vmcs(loaded_vmcs->vmcs);
2876 	loaded_vmcs->vmcs = NULL;
2877 	if (loaded_vmcs->msr_bitmap)
2878 		free_page((unsigned long)loaded_vmcs->msr_bitmap);
2879 	WARN_ON(loaded_vmcs->shadow_vmcs != NULL);
2880 }
2881 
2882 int alloc_loaded_vmcs(struct loaded_vmcs *loaded_vmcs)
2883 {
2884 	loaded_vmcs->vmcs = alloc_vmcs(false);
2885 	if (!loaded_vmcs->vmcs)
2886 		return -ENOMEM;
2887 
2888 	vmcs_clear(loaded_vmcs->vmcs);
2889 
2890 	loaded_vmcs->shadow_vmcs = NULL;
2891 	loaded_vmcs->hv_timer_soft_disabled = false;
2892 	loaded_vmcs->cpu = -1;
2893 	loaded_vmcs->launched = 0;
2894 
2895 	if (cpu_has_vmx_msr_bitmap()) {
2896 		loaded_vmcs->msr_bitmap = (unsigned long *)
2897 				__get_free_page(GFP_KERNEL_ACCOUNT);
2898 		if (!loaded_vmcs->msr_bitmap)
2899 			goto out_vmcs;
2900 		memset(loaded_vmcs->msr_bitmap, 0xff, PAGE_SIZE);
2901 	}
2902 
2903 	memset(&loaded_vmcs->host_state, 0, sizeof(struct vmcs_host_state));
2904 	memset(&loaded_vmcs->controls_shadow, 0,
2905 		sizeof(struct vmcs_controls_shadow));
2906 
2907 	return 0;
2908 
2909 out_vmcs:
2910 	free_loaded_vmcs(loaded_vmcs);
2911 	return -ENOMEM;
2912 }
2913 
2914 static void free_kvm_area(void)
2915 {
2916 	int cpu;
2917 
2918 	for_each_possible_cpu(cpu) {
2919 		free_vmcs(per_cpu(vmxarea, cpu));
2920 		per_cpu(vmxarea, cpu) = NULL;
2921 	}
2922 }
2923 
2924 static __init int alloc_kvm_area(void)
2925 {
2926 	int cpu;
2927 
2928 	for_each_possible_cpu(cpu) {
2929 		struct vmcs *vmcs;
2930 
2931 		vmcs = alloc_vmcs_cpu(false, cpu, GFP_KERNEL);
2932 		if (!vmcs) {
2933 			free_kvm_area();
2934 			return -ENOMEM;
2935 		}
2936 
2937 		/*
2938 		 * When eVMCS is enabled, alloc_vmcs_cpu() sets
2939 		 * vmcs->revision_id to KVM_EVMCS_VERSION instead of
2940 		 * revision_id reported by MSR_IA32_VMX_BASIC.
2941 		 *
2942 		 * However, even though not explicitly documented by
2943 		 * TLFS, VMXArea passed as VMXON argument should
2944 		 * still be marked with revision_id reported by
2945 		 * physical CPU.
2946 		 */
2947 		if (kvm_is_using_evmcs())
2948 			vmcs->hdr.revision_id = vmcs_config.revision_id;
2949 
2950 		per_cpu(vmxarea, cpu) = vmcs;
2951 	}
2952 	return 0;
2953 }
2954 
2955 static void fix_pmode_seg(struct kvm_vcpu *vcpu, int seg,
2956 		struct kvm_segment *save)
2957 {
2958 	if (!emulate_invalid_guest_state) {
2959 		/*
2960 		 * CS and SS RPL should be equal during guest entry according
2961 		 * to VMX spec, but in reality it is not always so. Since vcpu
2962 		 * is in the middle of the transition from real mode to
2963 		 * protected mode it is safe to assume that RPL 0 is a good
2964 		 * default value.
2965 		 */
2966 		if (seg == VCPU_SREG_CS || seg == VCPU_SREG_SS)
2967 			save->selector &= ~SEGMENT_RPL_MASK;
2968 		save->dpl = save->selector & SEGMENT_RPL_MASK;
2969 		save->s = 1;
2970 	}
2971 	__vmx_set_segment(vcpu, save, seg);
2972 }
2973 
2974 static void enter_pmode(struct kvm_vcpu *vcpu)
2975 {
2976 	unsigned long flags;
2977 	struct vcpu_vmx *vmx = to_vmx(vcpu);
2978 
2979 	/*
2980 	 * Update real mode segment cache. It may be not up-to-date if segment
2981 	 * register was written while vcpu was in a guest mode.
2982 	 */
2983 	vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_ES], VCPU_SREG_ES);
2984 	vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_DS], VCPU_SREG_DS);
2985 	vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_FS], VCPU_SREG_FS);
2986 	vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_GS], VCPU_SREG_GS);
2987 	vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_SS], VCPU_SREG_SS);
2988 	vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_CS], VCPU_SREG_CS);
2989 
2990 	vmx->rmode.vm86_active = 0;
2991 
2992 	__vmx_set_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_TR], VCPU_SREG_TR);
2993 
2994 	flags = vmcs_readl(GUEST_RFLAGS);
2995 	flags &= RMODE_GUEST_OWNED_EFLAGS_BITS;
2996 	flags |= vmx->rmode.save_rflags & ~RMODE_GUEST_OWNED_EFLAGS_BITS;
2997 	vmcs_writel(GUEST_RFLAGS, flags);
2998 
2999 	vmcs_writel(GUEST_CR4, (vmcs_readl(GUEST_CR4) & ~X86_CR4_VME) |
3000 			(vmcs_readl(CR4_READ_SHADOW) & X86_CR4_VME));
3001 
3002 	vmx_update_exception_bitmap(vcpu);
3003 
3004 	fix_pmode_seg(vcpu, VCPU_SREG_CS, &vmx->rmode.segs[VCPU_SREG_CS]);
3005 	fix_pmode_seg(vcpu, VCPU_SREG_SS, &vmx->rmode.segs[VCPU_SREG_SS]);
3006 	fix_pmode_seg(vcpu, VCPU_SREG_ES, &vmx->rmode.segs[VCPU_SREG_ES]);
3007 	fix_pmode_seg(vcpu, VCPU_SREG_DS, &vmx->rmode.segs[VCPU_SREG_DS]);
3008 	fix_pmode_seg(vcpu, VCPU_SREG_FS, &vmx->rmode.segs[VCPU_SREG_FS]);
3009 	fix_pmode_seg(vcpu, VCPU_SREG_GS, &vmx->rmode.segs[VCPU_SREG_GS]);
3010 }
3011 
3012 static void fix_rmode_seg(int seg, struct kvm_segment *save)
3013 {
3014 	const struct kvm_vmx_segment_field *sf = &kvm_vmx_segment_fields[seg];
3015 	struct kvm_segment var = *save;
3016 
3017 	var.dpl = 0x3;
3018 	if (seg == VCPU_SREG_CS)
3019 		var.type = 0x3;
3020 
3021 	if (!emulate_invalid_guest_state) {
3022 		var.selector = var.base >> 4;
3023 		var.base = var.base & 0xffff0;
3024 		var.limit = 0xffff;
3025 		var.g = 0;
3026 		var.db = 0;
3027 		var.present = 1;
3028 		var.s = 1;
3029 		var.l = 0;
3030 		var.unusable = 0;
3031 		var.type = 0x3;
3032 		var.avl = 0;
3033 		if (save->base & 0xf)
3034 			pr_warn_once("segment base is not paragraph aligned "
3035 				     "when entering protected mode (seg=%d)", seg);
3036 	}
3037 
3038 	vmcs_write16(sf->selector, var.selector);
3039 	vmcs_writel(sf->base, var.base);
3040 	vmcs_write32(sf->limit, var.limit);
3041 	vmcs_write32(sf->ar_bytes, vmx_segment_access_rights(&var));
3042 }
3043 
3044 static void enter_rmode(struct kvm_vcpu *vcpu)
3045 {
3046 	unsigned long flags;
3047 	struct vcpu_vmx *vmx = to_vmx(vcpu);
3048 	struct kvm_vmx *kvm_vmx = to_kvm_vmx(vcpu->kvm);
3049 
3050 	vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_TR], VCPU_SREG_TR);
3051 	vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_ES], VCPU_SREG_ES);
3052 	vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_DS], VCPU_SREG_DS);
3053 	vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_FS], VCPU_SREG_FS);
3054 	vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_GS], VCPU_SREG_GS);
3055 	vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_SS], VCPU_SREG_SS);
3056 	vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_CS], VCPU_SREG_CS);
3057 
3058 	vmx->rmode.vm86_active = 1;
3059 
3060 	/*
3061 	 * Very old userspace does not call KVM_SET_TSS_ADDR before entering
3062 	 * vcpu. Warn the user that an update is overdue.
3063 	 */
3064 	if (!kvm_vmx->tss_addr)
3065 		pr_warn_once("KVM_SET_TSS_ADDR needs to be called before running vCPU\n");
3066 
3067 	vmx_segment_cache_clear(vmx);
3068 
3069 	vmcs_writel(GUEST_TR_BASE, kvm_vmx->tss_addr);
3070 	vmcs_write32(GUEST_TR_LIMIT, RMODE_TSS_SIZE - 1);
3071 	vmcs_write32(GUEST_TR_AR_BYTES, 0x008b);
3072 
3073 	flags = vmcs_readl(GUEST_RFLAGS);
3074 	vmx->rmode.save_rflags = flags;
3075 
3076 	flags |= X86_EFLAGS_IOPL | X86_EFLAGS_VM;
3077 
3078 	vmcs_writel(GUEST_RFLAGS, flags);
3079 	vmcs_writel(GUEST_CR4, vmcs_readl(GUEST_CR4) | X86_CR4_VME);
3080 	vmx_update_exception_bitmap(vcpu);
3081 
3082 	fix_rmode_seg(VCPU_SREG_SS, &vmx->rmode.segs[VCPU_SREG_SS]);
3083 	fix_rmode_seg(VCPU_SREG_CS, &vmx->rmode.segs[VCPU_SREG_CS]);
3084 	fix_rmode_seg(VCPU_SREG_ES, &vmx->rmode.segs[VCPU_SREG_ES]);
3085 	fix_rmode_seg(VCPU_SREG_DS, &vmx->rmode.segs[VCPU_SREG_DS]);
3086 	fix_rmode_seg(VCPU_SREG_GS, &vmx->rmode.segs[VCPU_SREG_GS]);
3087 	fix_rmode_seg(VCPU_SREG_FS, &vmx->rmode.segs[VCPU_SREG_FS]);
3088 }
3089 
3090 int vmx_set_efer(struct kvm_vcpu *vcpu, u64 efer)
3091 {
3092 	struct vcpu_vmx *vmx = to_vmx(vcpu);
3093 
3094 	/* Nothing to do if hardware doesn't support EFER. */
3095 	if (!vmx_find_uret_msr(vmx, MSR_EFER))
3096 		return 0;
3097 
3098 	vcpu->arch.efer = efer;
3099 #ifdef CONFIG_X86_64
3100 	if (efer & EFER_LMA)
3101 		vm_entry_controls_setbit(vmx, VM_ENTRY_IA32E_MODE);
3102 	else
3103 		vm_entry_controls_clearbit(vmx, VM_ENTRY_IA32E_MODE);
3104 #else
3105 	if (KVM_BUG_ON(efer & EFER_LMA, vcpu->kvm))
3106 		return 1;
3107 #endif
3108 
3109 	vmx_setup_uret_msrs(vmx);
3110 	return 0;
3111 }
3112 
3113 #ifdef CONFIG_X86_64
3114 
3115 static void enter_lmode(struct kvm_vcpu *vcpu)
3116 {
3117 	u32 guest_tr_ar;
3118 
3119 	vmx_segment_cache_clear(to_vmx(vcpu));
3120 
3121 	guest_tr_ar = vmcs_read32(GUEST_TR_AR_BYTES);
3122 	if ((guest_tr_ar & VMX_AR_TYPE_MASK) != VMX_AR_TYPE_BUSY_64_TSS) {
3123 		pr_debug_ratelimited("%s: tss fixup for long mode. \n",
3124 				     __func__);
3125 		vmcs_write32(GUEST_TR_AR_BYTES,
3126 			     (guest_tr_ar & ~VMX_AR_TYPE_MASK)
3127 			     | VMX_AR_TYPE_BUSY_64_TSS);
3128 	}
3129 	vmx_set_efer(vcpu, vcpu->arch.efer | EFER_LMA);
3130 }
3131 
3132 static void exit_lmode(struct kvm_vcpu *vcpu)
3133 {
3134 	vmx_set_efer(vcpu, vcpu->arch.efer & ~EFER_LMA);
3135 }
3136 
3137 #endif
3138 
3139 static void vmx_flush_tlb_all(struct kvm_vcpu *vcpu)
3140 {
3141 	struct vcpu_vmx *vmx = to_vmx(vcpu);
3142 
3143 	/*
3144 	 * INVEPT must be issued when EPT is enabled, irrespective of VPID, as
3145 	 * the CPU is not required to invalidate guest-physical mappings on
3146 	 * VM-Entry, even if VPID is disabled.  Guest-physical mappings are
3147 	 * associated with the root EPT structure and not any particular VPID
3148 	 * (INVVPID also isn't required to invalidate guest-physical mappings).
3149 	 */
3150 	if (enable_ept) {
3151 		ept_sync_global();
3152 	} else if (enable_vpid) {
3153 		if (cpu_has_vmx_invvpid_global()) {
3154 			vpid_sync_vcpu_global();
3155 		} else {
3156 			vpid_sync_vcpu_single(vmx->vpid);
3157 			vpid_sync_vcpu_single(vmx->nested.vpid02);
3158 		}
3159 	}
3160 }
3161 
3162 static inline int vmx_get_current_vpid(struct kvm_vcpu *vcpu)
3163 {
3164 	if (is_guest_mode(vcpu))
3165 		return nested_get_vpid02(vcpu);
3166 	return to_vmx(vcpu)->vpid;
3167 }
3168 
3169 static void vmx_flush_tlb_current(struct kvm_vcpu *vcpu)
3170 {
3171 	struct kvm_mmu *mmu = vcpu->arch.mmu;
3172 	u64 root_hpa = mmu->root.hpa;
3173 
3174 	/* No flush required if the current context is invalid. */
3175 	if (!VALID_PAGE(root_hpa))
3176 		return;
3177 
3178 	if (enable_ept)
3179 		ept_sync_context(construct_eptp(vcpu, root_hpa,
3180 						mmu->root_role.level));
3181 	else
3182 		vpid_sync_context(vmx_get_current_vpid(vcpu));
3183 }
3184 
3185 static void vmx_flush_tlb_gva(struct kvm_vcpu *vcpu, gva_t addr)
3186 {
3187 	/*
3188 	 * vpid_sync_vcpu_addr() is a nop if vpid==0, see the comment in
3189 	 * vmx_flush_tlb_guest() for an explanation of why this is ok.
3190 	 */
3191 	vpid_sync_vcpu_addr(vmx_get_current_vpid(vcpu), addr);
3192 }
3193 
3194 static void vmx_flush_tlb_guest(struct kvm_vcpu *vcpu)
3195 {
3196 	/*
3197 	 * vpid_sync_context() is a nop if vpid==0, e.g. if enable_vpid==0 or a
3198 	 * vpid couldn't be allocated for this vCPU.  VM-Enter and VM-Exit are
3199 	 * required to flush GVA->{G,H}PA mappings from the TLB if vpid is
3200 	 * disabled (VM-Enter with vpid enabled and vpid==0 is disallowed),
3201 	 * i.e. no explicit INVVPID is necessary.
3202 	 */
3203 	vpid_sync_context(vmx_get_current_vpid(vcpu));
3204 }
3205 
3206 void vmx_ept_load_pdptrs(struct kvm_vcpu *vcpu)
3207 {
3208 	struct kvm_mmu *mmu = vcpu->arch.walk_mmu;
3209 
3210 	if (!kvm_register_is_dirty(vcpu, VCPU_EXREG_PDPTR))
3211 		return;
3212 
3213 	if (is_pae_paging(vcpu)) {
3214 		vmcs_write64(GUEST_PDPTR0, mmu->pdptrs[0]);
3215 		vmcs_write64(GUEST_PDPTR1, mmu->pdptrs[1]);
3216 		vmcs_write64(GUEST_PDPTR2, mmu->pdptrs[2]);
3217 		vmcs_write64(GUEST_PDPTR3, mmu->pdptrs[3]);
3218 	}
3219 }
3220 
3221 void ept_save_pdptrs(struct kvm_vcpu *vcpu)
3222 {
3223 	struct kvm_mmu *mmu = vcpu->arch.walk_mmu;
3224 
3225 	if (WARN_ON_ONCE(!is_pae_paging(vcpu)))
3226 		return;
3227 
3228 	mmu->pdptrs[0] = vmcs_read64(GUEST_PDPTR0);
3229 	mmu->pdptrs[1] = vmcs_read64(GUEST_PDPTR1);
3230 	mmu->pdptrs[2] = vmcs_read64(GUEST_PDPTR2);
3231 	mmu->pdptrs[3] = vmcs_read64(GUEST_PDPTR3);
3232 
3233 	kvm_register_mark_available(vcpu, VCPU_EXREG_PDPTR);
3234 }
3235 
3236 #define CR3_EXITING_BITS (CPU_BASED_CR3_LOAD_EXITING | \
3237 			  CPU_BASED_CR3_STORE_EXITING)
3238 
3239 void vmx_set_cr0(struct kvm_vcpu *vcpu, unsigned long cr0)
3240 {
3241 	struct vcpu_vmx *vmx = to_vmx(vcpu);
3242 	unsigned long hw_cr0, old_cr0_pg;
3243 	u32 tmp;
3244 
3245 	old_cr0_pg = kvm_read_cr0_bits(vcpu, X86_CR0_PG);
3246 
3247 	hw_cr0 = (cr0 & ~KVM_VM_CR0_ALWAYS_OFF);
3248 	if (is_unrestricted_guest(vcpu))
3249 		hw_cr0 |= KVM_VM_CR0_ALWAYS_ON_UNRESTRICTED_GUEST;
3250 	else {
3251 		hw_cr0 |= KVM_VM_CR0_ALWAYS_ON;
3252 		if (!enable_ept)
3253 			hw_cr0 |= X86_CR0_WP;
3254 
3255 		if (vmx->rmode.vm86_active && (cr0 & X86_CR0_PE))
3256 			enter_pmode(vcpu);
3257 
3258 		if (!vmx->rmode.vm86_active && !(cr0 & X86_CR0_PE))
3259 			enter_rmode(vcpu);
3260 	}
3261 
3262 	vmcs_writel(CR0_READ_SHADOW, cr0);
3263 	vmcs_writel(GUEST_CR0, hw_cr0);
3264 	vcpu->arch.cr0 = cr0;
3265 	kvm_register_mark_available(vcpu, VCPU_EXREG_CR0);
3266 
3267 #ifdef CONFIG_X86_64
3268 	if (vcpu->arch.efer & EFER_LME) {
3269 		if (!old_cr0_pg && (cr0 & X86_CR0_PG))
3270 			enter_lmode(vcpu);
3271 		else if (old_cr0_pg && !(cr0 & X86_CR0_PG))
3272 			exit_lmode(vcpu);
3273 	}
3274 #endif
3275 
3276 	if (enable_ept && !is_unrestricted_guest(vcpu)) {
3277 		/*
3278 		 * Ensure KVM has an up-to-date snapshot of the guest's CR3.  If
3279 		 * the below code _enables_ CR3 exiting, vmx_cache_reg() will
3280 		 * (correctly) stop reading vmcs.GUEST_CR3 because it thinks
3281 		 * KVM's CR3 is installed.
3282 		 */
3283 		if (!kvm_register_is_available(vcpu, VCPU_EXREG_CR3))
3284 			vmx_cache_reg(vcpu, VCPU_EXREG_CR3);
3285 
3286 		/*
3287 		 * When running with EPT but not unrestricted guest, KVM must
3288 		 * intercept CR3 accesses when paging is _disabled_.  This is
3289 		 * necessary because restricted guests can't actually run with
3290 		 * paging disabled, and so KVM stuffs its own CR3 in order to
3291 		 * run the guest when identity mapped page tables.
3292 		 *
3293 		 * Do _NOT_ check the old CR0.PG, e.g. to optimize away the
3294 		 * update, it may be stale with respect to CR3 interception,
3295 		 * e.g. after nested VM-Enter.
3296 		 *
3297 		 * Lastly, honor L1's desires, i.e. intercept CR3 loads and/or
3298 		 * stores to forward them to L1, even if KVM does not need to
3299 		 * intercept them to preserve its identity mapped page tables.
3300 		 */
3301 		if (!(cr0 & X86_CR0_PG)) {
3302 			exec_controls_setbit(vmx, CR3_EXITING_BITS);
3303 		} else if (!is_guest_mode(vcpu)) {
3304 			exec_controls_clearbit(vmx, CR3_EXITING_BITS);
3305 		} else {
3306 			tmp = exec_controls_get(vmx);
3307 			tmp &= ~CR3_EXITING_BITS;
3308 			tmp |= get_vmcs12(vcpu)->cpu_based_vm_exec_control & CR3_EXITING_BITS;
3309 			exec_controls_set(vmx, tmp);
3310 		}
3311 
3312 		/* Note, vmx_set_cr4() consumes the new vcpu->arch.cr0. */
3313 		if ((old_cr0_pg ^ cr0) & X86_CR0_PG)
3314 			vmx_set_cr4(vcpu, kvm_read_cr4(vcpu));
3315 
3316 		/*
3317 		 * When !CR0_PG -> CR0_PG, vcpu->arch.cr3 becomes active, but
3318 		 * GUEST_CR3 is still vmx->ept_identity_map_addr if EPT + !URG.
3319 		 */
3320 		if (!(old_cr0_pg & X86_CR0_PG) && (cr0 & X86_CR0_PG))
3321 			kvm_register_mark_dirty(vcpu, VCPU_EXREG_CR3);
3322 	}
3323 
3324 	/* depends on vcpu->arch.cr0 to be set to a new value */
3325 	vmx->emulation_required = vmx_emulation_required(vcpu);
3326 }
3327 
3328 static int vmx_get_max_tdp_level(void)
3329 {
3330 	if (cpu_has_vmx_ept_5levels())
3331 		return 5;
3332 	return 4;
3333 }
3334 
3335 u64 construct_eptp(struct kvm_vcpu *vcpu, hpa_t root_hpa, int root_level)
3336 {
3337 	u64 eptp = VMX_EPTP_MT_WB;
3338 
3339 	eptp |= (root_level == 5) ? VMX_EPTP_PWL_5 : VMX_EPTP_PWL_4;
3340 
3341 	if (enable_ept_ad_bits &&
3342 	    (!is_guest_mode(vcpu) || nested_ept_ad_enabled(vcpu)))
3343 		eptp |= VMX_EPTP_AD_ENABLE_BIT;
3344 	eptp |= root_hpa;
3345 
3346 	return eptp;
3347 }
3348 
3349 static void vmx_load_mmu_pgd(struct kvm_vcpu *vcpu, hpa_t root_hpa,
3350 			     int root_level)
3351 {
3352 	struct kvm *kvm = vcpu->kvm;
3353 	bool update_guest_cr3 = true;
3354 	unsigned long guest_cr3;
3355 	u64 eptp;
3356 
3357 	if (enable_ept) {
3358 		eptp = construct_eptp(vcpu, root_hpa, root_level);
3359 		vmcs_write64(EPT_POINTER, eptp);
3360 
3361 		hv_track_root_tdp(vcpu, root_hpa);
3362 
3363 		if (!enable_unrestricted_guest && !is_paging(vcpu))
3364 			guest_cr3 = to_kvm_vmx(kvm)->ept_identity_map_addr;
3365 		else if (kvm_register_is_dirty(vcpu, VCPU_EXREG_CR3))
3366 			guest_cr3 = vcpu->arch.cr3;
3367 		else /* vmcs.GUEST_CR3 is already up-to-date. */
3368 			update_guest_cr3 = false;
3369 		vmx_ept_load_pdptrs(vcpu);
3370 	} else {
3371 		guest_cr3 = root_hpa | kvm_get_active_pcid(vcpu);
3372 	}
3373 
3374 	if (update_guest_cr3)
3375 		vmcs_writel(GUEST_CR3, guest_cr3);
3376 }
3377 
3378 
3379 static bool vmx_is_valid_cr4(struct kvm_vcpu *vcpu, unsigned long cr4)
3380 {
3381 	/*
3382 	 * We operate under the default treatment of SMM, so VMX cannot be
3383 	 * enabled under SMM.  Note, whether or not VMXE is allowed at all,
3384 	 * i.e. is a reserved bit, is handled by common x86 code.
3385 	 */
3386 	if ((cr4 & X86_CR4_VMXE) && is_smm(vcpu))
3387 		return false;
3388 
3389 	if (to_vmx(vcpu)->nested.vmxon && !nested_cr4_valid(vcpu, cr4))
3390 		return false;
3391 
3392 	return true;
3393 }
3394 
3395 void vmx_set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4)
3396 {
3397 	unsigned long old_cr4 = kvm_read_cr4(vcpu);
3398 	struct vcpu_vmx *vmx = to_vmx(vcpu);
3399 	unsigned long hw_cr4;
3400 
3401 	/*
3402 	 * Pass through host's Machine Check Enable value to hw_cr4, which
3403 	 * is in force while we are in guest mode.  Do not let guests control
3404 	 * this bit, even if host CR4.MCE == 0.
3405 	 */
3406 	hw_cr4 = (cr4_read_shadow() & X86_CR4_MCE) | (cr4 & ~X86_CR4_MCE);
3407 	if (is_unrestricted_guest(vcpu))
3408 		hw_cr4 |= KVM_VM_CR4_ALWAYS_ON_UNRESTRICTED_GUEST;
3409 	else if (vmx->rmode.vm86_active)
3410 		hw_cr4 |= KVM_RMODE_VM_CR4_ALWAYS_ON;
3411 	else
3412 		hw_cr4 |= KVM_PMODE_VM_CR4_ALWAYS_ON;
3413 
3414 	if (vmx_umip_emulated()) {
3415 		if (cr4 & X86_CR4_UMIP) {
3416 			secondary_exec_controls_setbit(vmx, SECONDARY_EXEC_DESC);
3417 			hw_cr4 &= ~X86_CR4_UMIP;
3418 		} else if (!is_guest_mode(vcpu) ||
3419 			!nested_cpu_has2(get_vmcs12(vcpu), SECONDARY_EXEC_DESC)) {
3420 			secondary_exec_controls_clearbit(vmx, SECONDARY_EXEC_DESC);
3421 		}
3422 	}
3423 
3424 	vcpu->arch.cr4 = cr4;
3425 	kvm_register_mark_available(vcpu, VCPU_EXREG_CR4);
3426 
3427 	if (!is_unrestricted_guest(vcpu)) {
3428 		if (enable_ept) {
3429 			if (!is_paging(vcpu)) {
3430 				hw_cr4 &= ~X86_CR4_PAE;
3431 				hw_cr4 |= X86_CR4_PSE;
3432 			} else if (!(cr4 & X86_CR4_PAE)) {
3433 				hw_cr4 &= ~X86_CR4_PAE;
3434 			}
3435 		}
3436 
3437 		/*
3438 		 * SMEP/SMAP/PKU is disabled if CPU is in non-paging mode in
3439 		 * hardware.  To emulate this behavior, SMEP/SMAP/PKU needs
3440 		 * to be manually disabled when guest switches to non-paging
3441 		 * mode.
3442 		 *
3443 		 * If !enable_unrestricted_guest, the CPU is always running
3444 		 * with CR0.PG=1 and CR4 needs to be modified.
3445 		 * If enable_unrestricted_guest, the CPU automatically
3446 		 * disables SMEP/SMAP/PKU when the guest sets CR0.PG=0.
3447 		 */
3448 		if (!is_paging(vcpu))
3449 			hw_cr4 &= ~(X86_CR4_SMEP | X86_CR4_SMAP | X86_CR4_PKE);
3450 	}
3451 
3452 	vmcs_writel(CR4_READ_SHADOW, cr4);
3453 	vmcs_writel(GUEST_CR4, hw_cr4);
3454 
3455 	if ((cr4 ^ old_cr4) & (X86_CR4_OSXSAVE | X86_CR4_PKE))
3456 		kvm_update_cpuid_runtime(vcpu);
3457 }
3458 
3459 void vmx_get_segment(struct kvm_vcpu *vcpu, struct kvm_segment *var, int seg)
3460 {
3461 	struct vcpu_vmx *vmx = to_vmx(vcpu);
3462 	u32 ar;
3463 
3464 	if (vmx->rmode.vm86_active && seg != VCPU_SREG_LDTR) {
3465 		*var = vmx->rmode.segs[seg];
3466 		if (seg == VCPU_SREG_TR
3467 		    || var->selector == vmx_read_guest_seg_selector(vmx, seg))
3468 			return;
3469 		var->base = vmx_read_guest_seg_base(vmx, seg);
3470 		var->selector = vmx_read_guest_seg_selector(vmx, seg);
3471 		return;
3472 	}
3473 	var->base = vmx_read_guest_seg_base(vmx, seg);
3474 	var->limit = vmx_read_guest_seg_limit(vmx, seg);
3475 	var->selector = vmx_read_guest_seg_selector(vmx, seg);
3476 	ar = vmx_read_guest_seg_ar(vmx, seg);
3477 	var->unusable = (ar >> 16) & 1;
3478 	var->type = ar & 15;
3479 	var->s = (ar >> 4) & 1;
3480 	var->dpl = (ar >> 5) & 3;
3481 	/*
3482 	 * Some userspaces do not preserve unusable property. Since usable
3483 	 * segment has to be present according to VMX spec we can use present
3484 	 * property to amend userspace bug by making unusable segment always
3485 	 * nonpresent. vmx_segment_access_rights() already marks nonpresent
3486 	 * segment as unusable.
3487 	 */
3488 	var->present = !var->unusable;
3489 	var->avl = (ar >> 12) & 1;
3490 	var->l = (ar >> 13) & 1;
3491 	var->db = (ar >> 14) & 1;
3492 	var->g = (ar >> 15) & 1;
3493 }
3494 
3495 static u64 vmx_get_segment_base(struct kvm_vcpu *vcpu, int seg)
3496 {
3497 	struct kvm_segment s;
3498 
3499 	if (to_vmx(vcpu)->rmode.vm86_active) {
3500 		vmx_get_segment(vcpu, &s, seg);
3501 		return s.base;
3502 	}
3503 	return vmx_read_guest_seg_base(to_vmx(vcpu), seg);
3504 }
3505 
3506 int vmx_get_cpl(struct kvm_vcpu *vcpu)
3507 {
3508 	struct vcpu_vmx *vmx = to_vmx(vcpu);
3509 
3510 	if (unlikely(vmx->rmode.vm86_active))
3511 		return 0;
3512 	else {
3513 		int ar = vmx_read_guest_seg_ar(vmx, VCPU_SREG_SS);
3514 		return VMX_AR_DPL(ar);
3515 	}
3516 }
3517 
3518 static u32 vmx_segment_access_rights(struct kvm_segment *var)
3519 {
3520 	u32 ar;
3521 
3522 	ar = var->type & 15;
3523 	ar |= (var->s & 1) << 4;
3524 	ar |= (var->dpl & 3) << 5;
3525 	ar |= (var->present & 1) << 7;
3526 	ar |= (var->avl & 1) << 12;
3527 	ar |= (var->l & 1) << 13;
3528 	ar |= (var->db & 1) << 14;
3529 	ar |= (var->g & 1) << 15;
3530 	ar |= (var->unusable || !var->present) << 16;
3531 
3532 	return ar;
3533 }
3534 
3535 void __vmx_set_segment(struct kvm_vcpu *vcpu, struct kvm_segment *var, int seg)
3536 {
3537 	struct vcpu_vmx *vmx = to_vmx(vcpu);
3538 	const struct kvm_vmx_segment_field *sf = &kvm_vmx_segment_fields[seg];
3539 
3540 	vmx_segment_cache_clear(vmx);
3541 
3542 	if (vmx->rmode.vm86_active && seg != VCPU_SREG_LDTR) {
3543 		vmx->rmode.segs[seg] = *var;
3544 		if (seg == VCPU_SREG_TR)
3545 			vmcs_write16(sf->selector, var->selector);
3546 		else if (var->s)
3547 			fix_rmode_seg(seg, &vmx->rmode.segs[seg]);
3548 		return;
3549 	}
3550 
3551 	vmcs_writel(sf->base, var->base);
3552 	vmcs_write32(sf->limit, var->limit);
3553 	vmcs_write16(sf->selector, var->selector);
3554 
3555 	/*
3556 	 *   Fix the "Accessed" bit in AR field of segment registers for older
3557 	 * qemu binaries.
3558 	 *   IA32 arch specifies that at the time of processor reset the
3559 	 * "Accessed" bit in the AR field of segment registers is 1. And qemu
3560 	 * is setting it to 0 in the userland code. This causes invalid guest
3561 	 * state vmexit when "unrestricted guest" mode is turned on.
3562 	 *    Fix for this setup issue in cpu_reset is being pushed in the qemu
3563 	 * tree. Newer qemu binaries with that qemu fix would not need this
3564 	 * kvm hack.
3565 	 */
3566 	if (is_unrestricted_guest(vcpu) && (seg != VCPU_SREG_LDTR))
3567 		var->type |= 0x1; /* Accessed */
3568 
3569 	vmcs_write32(sf->ar_bytes, vmx_segment_access_rights(var));
3570 }
3571 
3572 static void vmx_set_segment(struct kvm_vcpu *vcpu, struct kvm_segment *var, int seg)
3573 {
3574 	__vmx_set_segment(vcpu, var, seg);
3575 
3576 	to_vmx(vcpu)->emulation_required = vmx_emulation_required(vcpu);
3577 }
3578 
3579 static void vmx_get_cs_db_l_bits(struct kvm_vcpu *vcpu, int *db, int *l)
3580 {
3581 	u32 ar = vmx_read_guest_seg_ar(to_vmx(vcpu), VCPU_SREG_CS);
3582 
3583 	*db = (ar >> 14) & 1;
3584 	*l = (ar >> 13) & 1;
3585 }
3586 
3587 static void vmx_get_idt(struct kvm_vcpu *vcpu, struct desc_ptr *dt)
3588 {
3589 	dt->size = vmcs_read32(GUEST_IDTR_LIMIT);
3590 	dt->address = vmcs_readl(GUEST_IDTR_BASE);
3591 }
3592 
3593 static void vmx_set_idt(struct kvm_vcpu *vcpu, struct desc_ptr *dt)
3594 {
3595 	vmcs_write32(GUEST_IDTR_LIMIT, dt->size);
3596 	vmcs_writel(GUEST_IDTR_BASE, dt->address);
3597 }
3598 
3599 static void vmx_get_gdt(struct kvm_vcpu *vcpu, struct desc_ptr *dt)
3600 {
3601 	dt->size = vmcs_read32(GUEST_GDTR_LIMIT);
3602 	dt->address = vmcs_readl(GUEST_GDTR_BASE);
3603 }
3604 
3605 static void vmx_set_gdt(struct kvm_vcpu *vcpu, struct desc_ptr *dt)
3606 {
3607 	vmcs_write32(GUEST_GDTR_LIMIT, dt->size);
3608 	vmcs_writel(GUEST_GDTR_BASE, dt->address);
3609 }
3610 
3611 static bool rmode_segment_valid(struct kvm_vcpu *vcpu, int seg)
3612 {
3613 	struct kvm_segment var;
3614 	u32 ar;
3615 
3616 	vmx_get_segment(vcpu, &var, seg);
3617 	var.dpl = 0x3;
3618 	if (seg == VCPU_SREG_CS)
3619 		var.type = 0x3;
3620 	ar = vmx_segment_access_rights(&var);
3621 
3622 	if (var.base != (var.selector << 4))
3623 		return false;
3624 	if (var.limit != 0xffff)
3625 		return false;
3626 	if (ar != 0xf3)
3627 		return false;
3628 
3629 	return true;
3630 }
3631 
3632 static bool code_segment_valid(struct kvm_vcpu *vcpu)
3633 {
3634 	struct kvm_segment cs;
3635 	unsigned int cs_rpl;
3636 
3637 	vmx_get_segment(vcpu, &cs, VCPU_SREG_CS);
3638 	cs_rpl = cs.selector & SEGMENT_RPL_MASK;
3639 
3640 	if (cs.unusable)
3641 		return false;
3642 	if (~cs.type & (VMX_AR_TYPE_CODE_MASK|VMX_AR_TYPE_ACCESSES_MASK))
3643 		return false;
3644 	if (!cs.s)
3645 		return false;
3646 	if (cs.type & VMX_AR_TYPE_WRITEABLE_MASK) {
3647 		if (cs.dpl > cs_rpl)
3648 			return false;
3649 	} else {
3650 		if (cs.dpl != cs_rpl)
3651 			return false;
3652 	}
3653 	if (!cs.present)
3654 		return false;
3655 
3656 	/* TODO: Add Reserved field check, this'll require a new member in the kvm_segment_field structure */
3657 	return true;
3658 }
3659 
3660 static bool stack_segment_valid(struct kvm_vcpu *vcpu)
3661 {
3662 	struct kvm_segment ss;
3663 	unsigned int ss_rpl;
3664 
3665 	vmx_get_segment(vcpu, &ss, VCPU_SREG_SS);
3666 	ss_rpl = ss.selector & SEGMENT_RPL_MASK;
3667 
3668 	if (ss.unusable)
3669 		return true;
3670 	if (ss.type != 3 && ss.type != 7)
3671 		return false;
3672 	if (!ss.s)
3673 		return false;
3674 	if (ss.dpl != ss_rpl) /* DPL != RPL */
3675 		return false;
3676 	if (!ss.present)
3677 		return false;
3678 
3679 	return true;
3680 }
3681 
3682 static bool data_segment_valid(struct kvm_vcpu *vcpu, int seg)
3683 {
3684 	struct kvm_segment var;
3685 	unsigned int rpl;
3686 
3687 	vmx_get_segment(vcpu, &var, seg);
3688 	rpl = var.selector & SEGMENT_RPL_MASK;
3689 
3690 	if (var.unusable)
3691 		return true;
3692 	if (!var.s)
3693 		return false;
3694 	if (!var.present)
3695 		return false;
3696 	if (~var.type & (VMX_AR_TYPE_CODE_MASK|VMX_AR_TYPE_WRITEABLE_MASK)) {
3697 		if (var.dpl < rpl) /* DPL < RPL */
3698 			return false;
3699 	}
3700 
3701 	/* TODO: Add other members to kvm_segment_field to allow checking for other access
3702 	 * rights flags
3703 	 */
3704 	return true;
3705 }
3706 
3707 static bool tr_valid(struct kvm_vcpu *vcpu)
3708 {
3709 	struct kvm_segment tr;
3710 
3711 	vmx_get_segment(vcpu, &tr, VCPU_SREG_TR);
3712 
3713 	if (tr.unusable)
3714 		return false;
3715 	if (tr.selector & SEGMENT_TI_MASK)	/* TI = 1 */
3716 		return false;
3717 	if (tr.type != 3 && tr.type != 11) /* TODO: Check if guest is in IA32e mode */
3718 		return false;
3719 	if (!tr.present)
3720 		return false;
3721 
3722 	return true;
3723 }
3724 
3725 static bool ldtr_valid(struct kvm_vcpu *vcpu)
3726 {
3727 	struct kvm_segment ldtr;
3728 
3729 	vmx_get_segment(vcpu, &ldtr, VCPU_SREG_LDTR);
3730 
3731 	if (ldtr.unusable)
3732 		return true;
3733 	if (ldtr.selector & SEGMENT_TI_MASK)	/* TI = 1 */
3734 		return false;
3735 	if (ldtr.type != 2)
3736 		return false;
3737 	if (!ldtr.present)
3738 		return false;
3739 
3740 	return true;
3741 }
3742 
3743 static bool cs_ss_rpl_check(struct kvm_vcpu *vcpu)
3744 {
3745 	struct kvm_segment cs, ss;
3746 
3747 	vmx_get_segment(vcpu, &cs, VCPU_SREG_CS);
3748 	vmx_get_segment(vcpu, &ss, VCPU_SREG_SS);
3749 
3750 	return ((cs.selector & SEGMENT_RPL_MASK) ==
3751 		 (ss.selector & SEGMENT_RPL_MASK));
3752 }
3753 
3754 /*
3755  * Check if guest state is valid. Returns true if valid, false if
3756  * not.
3757  * We assume that registers are always usable
3758  */
3759 bool __vmx_guest_state_valid(struct kvm_vcpu *vcpu)
3760 {
3761 	/* real mode guest state checks */
3762 	if (!is_protmode(vcpu) || (vmx_get_rflags(vcpu) & X86_EFLAGS_VM)) {
3763 		if (!rmode_segment_valid(vcpu, VCPU_SREG_CS))
3764 			return false;
3765 		if (!rmode_segment_valid(vcpu, VCPU_SREG_SS))
3766 			return false;
3767 		if (!rmode_segment_valid(vcpu, VCPU_SREG_DS))
3768 			return false;
3769 		if (!rmode_segment_valid(vcpu, VCPU_SREG_ES))
3770 			return false;
3771 		if (!rmode_segment_valid(vcpu, VCPU_SREG_FS))
3772 			return false;
3773 		if (!rmode_segment_valid(vcpu, VCPU_SREG_GS))
3774 			return false;
3775 	} else {
3776 	/* protected mode guest state checks */
3777 		if (!cs_ss_rpl_check(vcpu))
3778 			return false;
3779 		if (!code_segment_valid(vcpu))
3780 			return false;
3781 		if (!stack_segment_valid(vcpu))
3782 			return false;
3783 		if (!data_segment_valid(vcpu, VCPU_SREG_DS))
3784 			return false;
3785 		if (!data_segment_valid(vcpu, VCPU_SREG_ES))
3786 			return false;
3787 		if (!data_segment_valid(vcpu, VCPU_SREG_FS))
3788 			return false;
3789 		if (!data_segment_valid(vcpu, VCPU_SREG_GS))
3790 			return false;
3791 		if (!tr_valid(vcpu))
3792 			return false;
3793 		if (!ldtr_valid(vcpu))
3794 			return false;
3795 	}
3796 	/* TODO:
3797 	 * - Add checks on RIP
3798 	 * - Add checks on RFLAGS
3799 	 */
3800 
3801 	return true;
3802 }
3803 
3804 static int init_rmode_tss(struct kvm *kvm, void __user *ua)
3805 {
3806 	const void *zero_page = (const void *) __va(page_to_phys(ZERO_PAGE(0)));
3807 	u16 data;
3808 	int i;
3809 
3810 	for (i = 0; i < 3; i++) {
3811 		if (__copy_to_user(ua + PAGE_SIZE * i, zero_page, PAGE_SIZE))
3812 			return -EFAULT;
3813 	}
3814 
3815 	data = TSS_BASE_SIZE + TSS_REDIRECTION_SIZE;
3816 	if (__copy_to_user(ua + TSS_IOPB_BASE_OFFSET, &data, sizeof(u16)))
3817 		return -EFAULT;
3818 
3819 	data = ~0;
3820 	if (__copy_to_user(ua + RMODE_TSS_SIZE - 1, &data, sizeof(u8)))
3821 		return -EFAULT;
3822 
3823 	return 0;
3824 }
3825 
3826 static int init_rmode_identity_map(struct kvm *kvm)
3827 {
3828 	struct kvm_vmx *kvm_vmx = to_kvm_vmx(kvm);
3829 	int i, r = 0;
3830 	void __user *uaddr;
3831 	u32 tmp;
3832 
3833 	/* Protect kvm_vmx->ept_identity_pagetable_done. */
3834 	mutex_lock(&kvm->slots_lock);
3835 
3836 	if (likely(kvm_vmx->ept_identity_pagetable_done))
3837 		goto out;
3838 
3839 	if (!kvm_vmx->ept_identity_map_addr)
3840 		kvm_vmx->ept_identity_map_addr = VMX_EPT_IDENTITY_PAGETABLE_ADDR;
3841 
3842 	uaddr = __x86_set_memory_region(kvm,
3843 					IDENTITY_PAGETABLE_PRIVATE_MEMSLOT,
3844 					kvm_vmx->ept_identity_map_addr,
3845 					PAGE_SIZE);
3846 	if (IS_ERR(uaddr)) {
3847 		r = PTR_ERR(uaddr);
3848 		goto out;
3849 	}
3850 
3851 	/* Set up identity-mapping pagetable for EPT in real mode */
3852 	for (i = 0; i < (PAGE_SIZE / sizeof(tmp)); i++) {
3853 		tmp = (i << 22) + (_PAGE_PRESENT | _PAGE_RW | _PAGE_USER |
3854 			_PAGE_ACCESSED | _PAGE_DIRTY | _PAGE_PSE);
3855 		if (__copy_to_user(uaddr + i * sizeof(tmp), &tmp, sizeof(tmp))) {
3856 			r = -EFAULT;
3857 			goto out;
3858 		}
3859 	}
3860 	kvm_vmx->ept_identity_pagetable_done = true;
3861 
3862 out:
3863 	mutex_unlock(&kvm->slots_lock);
3864 	return r;
3865 }
3866 
3867 static void seg_setup(int seg)
3868 {
3869 	const struct kvm_vmx_segment_field *sf = &kvm_vmx_segment_fields[seg];
3870 	unsigned int ar;
3871 
3872 	vmcs_write16(sf->selector, 0);
3873 	vmcs_writel(sf->base, 0);
3874 	vmcs_write32(sf->limit, 0xffff);
3875 	ar = 0x93;
3876 	if (seg == VCPU_SREG_CS)
3877 		ar |= 0x08; /* code segment */
3878 
3879 	vmcs_write32(sf->ar_bytes, ar);
3880 }
3881 
3882 int allocate_vpid(void)
3883 {
3884 	int vpid;
3885 
3886 	if (!enable_vpid)
3887 		return 0;
3888 	spin_lock(&vmx_vpid_lock);
3889 	vpid = find_first_zero_bit(vmx_vpid_bitmap, VMX_NR_VPIDS);
3890 	if (vpid < VMX_NR_VPIDS)
3891 		__set_bit(vpid, vmx_vpid_bitmap);
3892 	else
3893 		vpid = 0;
3894 	spin_unlock(&vmx_vpid_lock);
3895 	return vpid;
3896 }
3897 
3898 void free_vpid(int vpid)
3899 {
3900 	if (!enable_vpid || vpid == 0)
3901 		return;
3902 	spin_lock(&vmx_vpid_lock);
3903 	__clear_bit(vpid, vmx_vpid_bitmap);
3904 	spin_unlock(&vmx_vpid_lock);
3905 }
3906 
3907 static void vmx_msr_bitmap_l01_changed(struct vcpu_vmx *vmx)
3908 {
3909 	/*
3910 	 * When KVM is a nested hypervisor on top of Hyper-V and uses
3911 	 * 'Enlightened MSR Bitmap' feature L0 needs to know that MSR
3912 	 * bitmap has changed.
3913 	 */
3914 	if (kvm_is_using_evmcs()) {
3915 		struct hv_enlightened_vmcs *evmcs = (void *)vmx->vmcs01.vmcs;
3916 
3917 		if (evmcs->hv_enlightenments_control.msr_bitmap)
3918 			evmcs->hv_clean_fields &=
3919 				~HV_VMX_ENLIGHTENED_CLEAN_FIELD_MSR_BITMAP;
3920 	}
3921 
3922 	vmx->nested.force_msr_bitmap_recalc = true;
3923 }
3924 
3925 void vmx_disable_intercept_for_msr(struct kvm_vcpu *vcpu, u32 msr, int type)
3926 {
3927 	struct vcpu_vmx *vmx = to_vmx(vcpu);
3928 	unsigned long *msr_bitmap = vmx->vmcs01.msr_bitmap;
3929 
3930 	if (!cpu_has_vmx_msr_bitmap())
3931 		return;
3932 
3933 	vmx_msr_bitmap_l01_changed(vmx);
3934 
3935 	/*
3936 	 * Mark the desired intercept state in shadow bitmap, this is needed
3937 	 * for resync when the MSR filters change.
3938 	*/
3939 	if (is_valid_passthrough_msr(msr)) {
3940 		int idx = possible_passthrough_msr_slot(msr);
3941 
3942 		if (idx != -ENOENT) {
3943 			if (type & MSR_TYPE_R)
3944 				clear_bit(idx, vmx->shadow_msr_intercept.read);
3945 			if (type & MSR_TYPE_W)
3946 				clear_bit(idx, vmx->shadow_msr_intercept.write);
3947 		}
3948 	}
3949 
3950 	if ((type & MSR_TYPE_R) &&
3951 	    !kvm_msr_allowed(vcpu, msr, KVM_MSR_FILTER_READ)) {
3952 		vmx_set_msr_bitmap_read(msr_bitmap, msr);
3953 		type &= ~MSR_TYPE_R;
3954 	}
3955 
3956 	if ((type & MSR_TYPE_W) &&
3957 	    !kvm_msr_allowed(vcpu, msr, KVM_MSR_FILTER_WRITE)) {
3958 		vmx_set_msr_bitmap_write(msr_bitmap, msr);
3959 		type &= ~MSR_TYPE_W;
3960 	}
3961 
3962 	if (type & MSR_TYPE_R)
3963 		vmx_clear_msr_bitmap_read(msr_bitmap, msr);
3964 
3965 	if (type & MSR_TYPE_W)
3966 		vmx_clear_msr_bitmap_write(msr_bitmap, msr);
3967 }
3968 
3969 void vmx_enable_intercept_for_msr(struct kvm_vcpu *vcpu, u32 msr, int type)
3970 {
3971 	struct vcpu_vmx *vmx = to_vmx(vcpu);
3972 	unsigned long *msr_bitmap = vmx->vmcs01.msr_bitmap;
3973 
3974 	if (!cpu_has_vmx_msr_bitmap())
3975 		return;
3976 
3977 	vmx_msr_bitmap_l01_changed(vmx);
3978 
3979 	/*
3980 	 * Mark the desired intercept state in shadow bitmap, this is needed
3981 	 * for resync when the MSR filter changes.
3982 	*/
3983 	if (is_valid_passthrough_msr(msr)) {
3984 		int idx = possible_passthrough_msr_slot(msr);
3985 
3986 		if (idx != -ENOENT) {
3987 			if (type & MSR_TYPE_R)
3988 				set_bit(idx, vmx->shadow_msr_intercept.read);
3989 			if (type & MSR_TYPE_W)
3990 				set_bit(idx, vmx->shadow_msr_intercept.write);
3991 		}
3992 	}
3993 
3994 	if (type & MSR_TYPE_R)
3995 		vmx_set_msr_bitmap_read(msr_bitmap, msr);
3996 
3997 	if (type & MSR_TYPE_W)
3998 		vmx_set_msr_bitmap_write(msr_bitmap, msr);
3999 }
4000 
4001 static void vmx_update_msr_bitmap_x2apic(struct kvm_vcpu *vcpu)
4002 {
4003 	/*
4004 	 * x2APIC indices for 64-bit accesses into the RDMSR and WRMSR halves
4005 	 * of the MSR bitmap.  KVM emulates APIC registers up through 0x3f0,
4006 	 * i.e. MSR 0x83f, and so only needs to dynamically manipulate 64 bits.
4007 	 */
4008 	const int read_idx = APIC_BASE_MSR / BITS_PER_LONG_LONG;
4009 	const int write_idx = read_idx + (0x800 / sizeof(u64));
4010 	struct vcpu_vmx *vmx = to_vmx(vcpu);
4011 	u64 *msr_bitmap = (u64 *)vmx->vmcs01.msr_bitmap;
4012 	u8 mode;
4013 
4014 	if (!cpu_has_vmx_msr_bitmap() || WARN_ON_ONCE(!lapic_in_kernel(vcpu)))
4015 		return;
4016 
4017 	if (cpu_has_secondary_exec_ctrls() &&
4018 	    (secondary_exec_controls_get(vmx) &
4019 	     SECONDARY_EXEC_VIRTUALIZE_X2APIC_MODE)) {
4020 		mode = MSR_BITMAP_MODE_X2APIC;
4021 		if (enable_apicv && kvm_vcpu_apicv_active(vcpu))
4022 			mode |= MSR_BITMAP_MODE_X2APIC_APICV;
4023 	} else {
4024 		mode = 0;
4025 	}
4026 
4027 	if (mode == vmx->x2apic_msr_bitmap_mode)
4028 		return;
4029 
4030 	vmx->x2apic_msr_bitmap_mode = mode;
4031 
4032 	/*
4033 	 * Reset the bitmap for MSRs 0x800 - 0x83f.  Leave AMD's uber-extended
4034 	 * registers (0x840 and above) intercepted, KVM doesn't support them.
4035 	 * Intercept all writes by default and poke holes as needed.  Pass
4036 	 * through reads for all valid registers by default in x2APIC+APICv
4037 	 * mode, only the current timer count needs on-demand emulation by KVM.
4038 	 */
4039 	if (mode & MSR_BITMAP_MODE_X2APIC_APICV)
4040 		msr_bitmap[read_idx] = ~kvm_lapic_readable_reg_mask(vcpu->arch.apic);
4041 	else
4042 		msr_bitmap[read_idx] = ~0ull;
4043 	msr_bitmap[write_idx] = ~0ull;
4044 
4045 	/*
4046 	 * TPR reads and writes can be virtualized even if virtual interrupt
4047 	 * delivery is not in use.
4048 	 */
4049 	vmx_set_intercept_for_msr(vcpu, X2APIC_MSR(APIC_TASKPRI), MSR_TYPE_RW,
4050 				  !(mode & MSR_BITMAP_MODE_X2APIC));
4051 
4052 	if (mode & MSR_BITMAP_MODE_X2APIC_APICV) {
4053 		vmx_enable_intercept_for_msr(vcpu, X2APIC_MSR(APIC_TMCCT), MSR_TYPE_RW);
4054 		vmx_disable_intercept_for_msr(vcpu, X2APIC_MSR(APIC_EOI), MSR_TYPE_W);
4055 		vmx_disable_intercept_for_msr(vcpu, X2APIC_MSR(APIC_SELF_IPI), MSR_TYPE_W);
4056 		if (enable_ipiv)
4057 			vmx_disable_intercept_for_msr(vcpu, X2APIC_MSR(APIC_ICR), MSR_TYPE_RW);
4058 	}
4059 }
4060 
4061 void pt_update_intercept_for_msr(struct kvm_vcpu *vcpu)
4062 {
4063 	struct vcpu_vmx *vmx = to_vmx(vcpu);
4064 	bool flag = !(vmx->pt_desc.guest.ctl & RTIT_CTL_TRACEEN);
4065 	u32 i;
4066 
4067 	vmx_set_intercept_for_msr(vcpu, MSR_IA32_RTIT_STATUS, MSR_TYPE_RW, flag);
4068 	vmx_set_intercept_for_msr(vcpu, MSR_IA32_RTIT_OUTPUT_BASE, MSR_TYPE_RW, flag);
4069 	vmx_set_intercept_for_msr(vcpu, MSR_IA32_RTIT_OUTPUT_MASK, MSR_TYPE_RW, flag);
4070 	vmx_set_intercept_for_msr(vcpu, MSR_IA32_RTIT_CR3_MATCH, MSR_TYPE_RW, flag);
4071 	for (i = 0; i < vmx->pt_desc.num_address_ranges; i++) {
4072 		vmx_set_intercept_for_msr(vcpu, MSR_IA32_RTIT_ADDR0_A + i * 2, MSR_TYPE_RW, flag);
4073 		vmx_set_intercept_for_msr(vcpu, MSR_IA32_RTIT_ADDR0_B + i * 2, MSR_TYPE_RW, flag);
4074 	}
4075 }
4076 
4077 static bool vmx_guest_apic_has_interrupt(struct kvm_vcpu *vcpu)
4078 {
4079 	struct vcpu_vmx *vmx = to_vmx(vcpu);
4080 	void *vapic_page;
4081 	u32 vppr;
4082 	int rvi;
4083 
4084 	if (WARN_ON_ONCE(!is_guest_mode(vcpu)) ||
4085 		!nested_cpu_has_vid(get_vmcs12(vcpu)) ||
4086 		WARN_ON_ONCE(!vmx->nested.virtual_apic_map.gfn))
4087 		return false;
4088 
4089 	rvi = vmx_get_rvi();
4090 
4091 	vapic_page = vmx->nested.virtual_apic_map.hva;
4092 	vppr = *((u32 *)(vapic_page + APIC_PROCPRI));
4093 
4094 	return ((rvi & 0xf0) > (vppr & 0xf0));
4095 }
4096 
4097 static void vmx_msr_filter_changed(struct kvm_vcpu *vcpu)
4098 {
4099 	struct vcpu_vmx *vmx = to_vmx(vcpu);
4100 	u32 i;
4101 
4102 	/*
4103 	 * Redo intercept permissions for MSRs that KVM is passing through to
4104 	 * the guest.  Disabling interception will check the new MSR filter and
4105 	 * ensure that KVM enables interception if usersepace wants to filter
4106 	 * the MSR.  MSRs that KVM is already intercepting don't need to be
4107 	 * refreshed since KVM is going to intercept them regardless of what
4108 	 * userspace wants.
4109 	 */
4110 	for (i = 0; i < ARRAY_SIZE(vmx_possible_passthrough_msrs); i++) {
4111 		u32 msr = vmx_possible_passthrough_msrs[i];
4112 
4113 		if (!test_bit(i, vmx->shadow_msr_intercept.read))
4114 			vmx_disable_intercept_for_msr(vcpu, msr, MSR_TYPE_R);
4115 
4116 		if (!test_bit(i, vmx->shadow_msr_intercept.write))
4117 			vmx_disable_intercept_for_msr(vcpu, msr, MSR_TYPE_W);
4118 	}
4119 
4120 	/* PT MSRs can be passed through iff PT is exposed to the guest. */
4121 	if (vmx_pt_mode_is_host_guest())
4122 		pt_update_intercept_for_msr(vcpu);
4123 }
4124 
4125 static inline void kvm_vcpu_trigger_posted_interrupt(struct kvm_vcpu *vcpu,
4126 						     int pi_vec)
4127 {
4128 #ifdef CONFIG_SMP
4129 	if (vcpu->mode == IN_GUEST_MODE) {
4130 		/*
4131 		 * The vector of the virtual has already been set in the PIR.
4132 		 * Send a notification event to deliver the virtual interrupt
4133 		 * unless the vCPU is the currently running vCPU, i.e. the
4134 		 * event is being sent from a fastpath VM-Exit handler, in
4135 		 * which case the PIR will be synced to the vIRR before
4136 		 * re-entering the guest.
4137 		 *
4138 		 * When the target is not the running vCPU, the following
4139 		 * possibilities emerge:
4140 		 *
4141 		 * Case 1: vCPU stays in non-root mode. Sending a notification
4142 		 * event posts the interrupt to the vCPU.
4143 		 *
4144 		 * Case 2: vCPU exits to root mode and is still runnable. The
4145 		 * PIR will be synced to the vIRR before re-entering the guest.
4146 		 * Sending a notification event is ok as the host IRQ handler
4147 		 * will ignore the spurious event.
4148 		 *
4149 		 * Case 3: vCPU exits to root mode and is blocked. vcpu_block()
4150 		 * has already synced PIR to vIRR and never blocks the vCPU if
4151 		 * the vIRR is not empty. Therefore, a blocked vCPU here does
4152 		 * not wait for any requested interrupts in PIR, and sending a
4153 		 * notification event also results in a benign, spurious event.
4154 		 */
4155 
4156 		if (vcpu != kvm_get_running_vcpu())
4157 			apic->send_IPI_mask(get_cpu_mask(vcpu->cpu), pi_vec);
4158 		return;
4159 	}
4160 #endif
4161 	/*
4162 	 * The vCPU isn't in the guest; wake the vCPU in case it is blocking,
4163 	 * otherwise do nothing as KVM will grab the highest priority pending
4164 	 * IRQ via ->sync_pir_to_irr() in vcpu_enter_guest().
4165 	 */
4166 	kvm_vcpu_wake_up(vcpu);
4167 }
4168 
4169 static int vmx_deliver_nested_posted_interrupt(struct kvm_vcpu *vcpu,
4170 						int vector)
4171 {
4172 	struct vcpu_vmx *vmx = to_vmx(vcpu);
4173 
4174 	if (is_guest_mode(vcpu) &&
4175 	    vector == vmx->nested.posted_intr_nv) {
4176 		/*
4177 		 * If a posted intr is not recognized by hardware,
4178 		 * we will accomplish it in the next vmentry.
4179 		 */
4180 		vmx->nested.pi_pending = true;
4181 		kvm_make_request(KVM_REQ_EVENT, vcpu);
4182 
4183 		/*
4184 		 * This pairs with the smp_mb_*() after setting vcpu->mode in
4185 		 * vcpu_enter_guest() to guarantee the vCPU sees the event
4186 		 * request if triggering a posted interrupt "fails" because
4187 		 * vcpu->mode != IN_GUEST_MODE.  The extra barrier is needed as
4188 		 * the smb_wmb() in kvm_make_request() only ensures everything
4189 		 * done before making the request is visible when the request
4190 		 * is visible, it doesn't ensure ordering between the store to
4191 		 * vcpu->requests and the load from vcpu->mode.
4192 		 */
4193 		smp_mb__after_atomic();
4194 
4195 		/* the PIR and ON have been set by L1. */
4196 		kvm_vcpu_trigger_posted_interrupt(vcpu, POSTED_INTR_NESTED_VECTOR);
4197 		return 0;
4198 	}
4199 	return -1;
4200 }
4201 /*
4202  * Send interrupt to vcpu via posted interrupt way.
4203  * 1. If target vcpu is running(non-root mode), send posted interrupt
4204  * notification to vcpu and hardware will sync PIR to vIRR atomically.
4205  * 2. If target vcpu isn't running(root mode), kick it to pick up the
4206  * interrupt from PIR in next vmentry.
4207  */
4208 static int vmx_deliver_posted_interrupt(struct kvm_vcpu *vcpu, int vector)
4209 {
4210 	struct vcpu_vmx *vmx = to_vmx(vcpu);
4211 	int r;
4212 
4213 	r = vmx_deliver_nested_posted_interrupt(vcpu, vector);
4214 	if (!r)
4215 		return 0;
4216 
4217 	/* Note, this is called iff the local APIC is in-kernel. */
4218 	if (!vcpu->arch.apic->apicv_active)
4219 		return -1;
4220 
4221 	if (pi_test_and_set_pir(vector, &vmx->pi_desc))
4222 		return 0;
4223 
4224 	/* If a previous notification has sent the IPI, nothing to do.  */
4225 	if (pi_test_and_set_on(&vmx->pi_desc))
4226 		return 0;
4227 
4228 	/*
4229 	 * The implied barrier in pi_test_and_set_on() pairs with the smp_mb_*()
4230 	 * after setting vcpu->mode in vcpu_enter_guest(), thus the vCPU is
4231 	 * guaranteed to see PID.ON=1 and sync the PIR to IRR if triggering a
4232 	 * posted interrupt "fails" because vcpu->mode != IN_GUEST_MODE.
4233 	 */
4234 	kvm_vcpu_trigger_posted_interrupt(vcpu, POSTED_INTR_VECTOR);
4235 	return 0;
4236 }
4237 
4238 static void vmx_deliver_interrupt(struct kvm_lapic *apic, int delivery_mode,
4239 				  int trig_mode, int vector)
4240 {
4241 	struct kvm_vcpu *vcpu = apic->vcpu;
4242 
4243 	if (vmx_deliver_posted_interrupt(vcpu, vector)) {
4244 		kvm_lapic_set_irr(vector, apic);
4245 		kvm_make_request(KVM_REQ_EVENT, vcpu);
4246 		kvm_vcpu_kick(vcpu);
4247 	} else {
4248 		trace_kvm_apicv_accept_irq(vcpu->vcpu_id, delivery_mode,
4249 					   trig_mode, vector);
4250 	}
4251 }
4252 
4253 /*
4254  * Set up the vmcs's constant host-state fields, i.e., host-state fields that
4255  * will not change in the lifetime of the guest.
4256  * Note that host-state that does change is set elsewhere. E.g., host-state
4257  * that is set differently for each CPU is set in vmx_vcpu_load(), not here.
4258  */
4259 void vmx_set_constant_host_state(struct vcpu_vmx *vmx)
4260 {
4261 	u32 low32, high32;
4262 	unsigned long tmpl;
4263 	unsigned long cr0, cr3, cr4;
4264 
4265 	cr0 = read_cr0();
4266 	WARN_ON(cr0 & X86_CR0_TS);
4267 	vmcs_writel(HOST_CR0, cr0);  /* 22.2.3 */
4268 
4269 	/*
4270 	 * Save the most likely value for this task's CR3 in the VMCS.
4271 	 * We can't use __get_current_cr3_fast() because we're not atomic.
4272 	 */
4273 	cr3 = __read_cr3();
4274 	vmcs_writel(HOST_CR3, cr3);		/* 22.2.3  FIXME: shadow tables */
4275 	vmx->loaded_vmcs->host_state.cr3 = cr3;
4276 
4277 	/* Save the most likely value for this task's CR4 in the VMCS. */
4278 	cr4 = cr4_read_shadow();
4279 	vmcs_writel(HOST_CR4, cr4);			/* 22.2.3, 22.2.5 */
4280 	vmx->loaded_vmcs->host_state.cr4 = cr4;
4281 
4282 	vmcs_write16(HOST_CS_SELECTOR, __KERNEL_CS);  /* 22.2.4 */
4283 #ifdef CONFIG_X86_64
4284 	/*
4285 	 * Load null selectors, so we can avoid reloading them in
4286 	 * vmx_prepare_switch_to_host(), in case userspace uses
4287 	 * the null selectors too (the expected case).
4288 	 */
4289 	vmcs_write16(HOST_DS_SELECTOR, 0);
4290 	vmcs_write16(HOST_ES_SELECTOR, 0);
4291 #else
4292 	vmcs_write16(HOST_DS_SELECTOR, __KERNEL_DS);  /* 22.2.4 */
4293 	vmcs_write16(HOST_ES_SELECTOR, __KERNEL_DS);  /* 22.2.4 */
4294 #endif
4295 	vmcs_write16(HOST_SS_SELECTOR, __KERNEL_DS);  /* 22.2.4 */
4296 	vmcs_write16(HOST_TR_SELECTOR, GDT_ENTRY_TSS*8);  /* 22.2.4 */
4297 
4298 	vmcs_writel(HOST_IDTR_BASE, host_idt_base);   /* 22.2.4 */
4299 
4300 	vmcs_writel(HOST_RIP, (unsigned long)vmx_vmexit); /* 22.2.5 */
4301 
4302 	rdmsr(MSR_IA32_SYSENTER_CS, low32, high32);
4303 	vmcs_write32(HOST_IA32_SYSENTER_CS, low32);
4304 
4305 	/*
4306 	 * SYSENTER is used for 32-bit system calls on either 32-bit or
4307 	 * 64-bit kernels.  It is always zero If neither is allowed, otherwise
4308 	 * vmx_vcpu_load_vmcs loads it with the per-CPU entry stack (and may
4309 	 * have already done so!).
4310 	 */
4311 	if (!IS_ENABLED(CONFIG_IA32_EMULATION) && !IS_ENABLED(CONFIG_X86_32))
4312 		vmcs_writel(HOST_IA32_SYSENTER_ESP, 0);
4313 
4314 	rdmsrl(MSR_IA32_SYSENTER_EIP, tmpl);
4315 	vmcs_writel(HOST_IA32_SYSENTER_EIP, tmpl);   /* 22.2.3 */
4316 
4317 	if (vmcs_config.vmexit_ctrl & VM_EXIT_LOAD_IA32_PAT) {
4318 		rdmsr(MSR_IA32_CR_PAT, low32, high32);
4319 		vmcs_write64(HOST_IA32_PAT, low32 | ((u64) high32 << 32));
4320 	}
4321 
4322 	if (cpu_has_load_ia32_efer())
4323 		vmcs_write64(HOST_IA32_EFER, host_efer);
4324 }
4325 
4326 void set_cr4_guest_host_mask(struct vcpu_vmx *vmx)
4327 {
4328 	struct kvm_vcpu *vcpu = &vmx->vcpu;
4329 
4330 	vcpu->arch.cr4_guest_owned_bits = KVM_POSSIBLE_CR4_GUEST_BITS &
4331 					  ~vcpu->arch.cr4_guest_rsvd_bits;
4332 	if (!enable_ept) {
4333 		vcpu->arch.cr4_guest_owned_bits &= ~X86_CR4_TLBFLUSH_BITS;
4334 		vcpu->arch.cr4_guest_owned_bits &= ~X86_CR4_PDPTR_BITS;
4335 	}
4336 	if (is_guest_mode(&vmx->vcpu))
4337 		vcpu->arch.cr4_guest_owned_bits &=
4338 			~get_vmcs12(vcpu)->cr4_guest_host_mask;
4339 	vmcs_writel(CR4_GUEST_HOST_MASK, ~vcpu->arch.cr4_guest_owned_bits);
4340 }
4341 
4342 static u32 vmx_pin_based_exec_ctrl(struct vcpu_vmx *vmx)
4343 {
4344 	u32 pin_based_exec_ctrl = vmcs_config.pin_based_exec_ctrl;
4345 
4346 	if (!kvm_vcpu_apicv_active(&vmx->vcpu))
4347 		pin_based_exec_ctrl &= ~PIN_BASED_POSTED_INTR;
4348 
4349 	if (!enable_vnmi)
4350 		pin_based_exec_ctrl &= ~PIN_BASED_VIRTUAL_NMIS;
4351 
4352 	if (!enable_preemption_timer)
4353 		pin_based_exec_ctrl &= ~PIN_BASED_VMX_PREEMPTION_TIMER;
4354 
4355 	return pin_based_exec_ctrl;
4356 }
4357 
4358 static u32 vmx_vmentry_ctrl(void)
4359 {
4360 	u32 vmentry_ctrl = vmcs_config.vmentry_ctrl;
4361 
4362 	if (vmx_pt_mode_is_system())
4363 		vmentry_ctrl &= ~(VM_ENTRY_PT_CONCEAL_PIP |
4364 				  VM_ENTRY_LOAD_IA32_RTIT_CTL);
4365 	/*
4366 	 * IA32e mode, and loading of EFER and PERF_GLOBAL_CTRL are toggled dynamically.
4367 	 */
4368 	vmentry_ctrl &= ~(VM_ENTRY_LOAD_IA32_PERF_GLOBAL_CTRL |
4369 			  VM_ENTRY_LOAD_IA32_EFER |
4370 			  VM_ENTRY_IA32E_MODE);
4371 
4372 	if (cpu_has_perf_global_ctrl_bug())
4373 		vmentry_ctrl &= ~VM_ENTRY_LOAD_IA32_PERF_GLOBAL_CTRL;
4374 
4375 	return vmentry_ctrl;
4376 }
4377 
4378 static u32 vmx_vmexit_ctrl(void)
4379 {
4380 	u32 vmexit_ctrl = vmcs_config.vmexit_ctrl;
4381 
4382 	/*
4383 	 * Not used by KVM and never set in vmcs01 or vmcs02, but emulated for
4384 	 * nested virtualization and thus allowed to be set in vmcs12.
4385 	 */
4386 	vmexit_ctrl &= ~(VM_EXIT_SAVE_IA32_PAT | VM_EXIT_SAVE_IA32_EFER |
4387 			 VM_EXIT_SAVE_VMX_PREEMPTION_TIMER);
4388 
4389 	if (vmx_pt_mode_is_system())
4390 		vmexit_ctrl &= ~(VM_EXIT_PT_CONCEAL_PIP |
4391 				 VM_EXIT_CLEAR_IA32_RTIT_CTL);
4392 
4393 	if (cpu_has_perf_global_ctrl_bug())
4394 		vmexit_ctrl &= ~VM_EXIT_LOAD_IA32_PERF_GLOBAL_CTRL;
4395 
4396 	/* Loading of EFER and PERF_GLOBAL_CTRL are toggled dynamically */
4397 	return vmexit_ctrl &
4398 		~(VM_EXIT_LOAD_IA32_PERF_GLOBAL_CTRL | VM_EXIT_LOAD_IA32_EFER);
4399 }
4400 
4401 static void vmx_refresh_apicv_exec_ctrl(struct kvm_vcpu *vcpu)
4402 {
4403 	struct vcpu_vmx *vmx = to_vmx(vcpu);
4404 
4405 	if (is_guest_mode(vcpu)) {
4406 		vmx->nested.update_vmcs01_apicv_status = true;
4407 		return;
4408 	}
4409 
4410 	pin_controls_set(vmx, vmx_pin_based_exec_ctrl(vmx));
4411 
4412 	if (kvm_vcpu_apicv_active(vcpu)) {
4413 		secondary_exec_controls_setbit(vmx,
4414 					       SECONDARY_EXEC_APIC_REGISTER_VIRT |
4415 					       SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY);
4416 		if (enable_ipiv)
4417 			tertiary_exec_controls_setbit(vmx, TERTIARY_EXEC_IPI_VIRT);
4418 	} else {
4419 		secondary_exec_controls_clearbit(vmx,
4420 						 SECONDARY_EXEC_APIC_REGISTER_VIRT |
4421 						 SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY);
4422 		if (enable_ipiv)
4423 			tertiary_exec_controls_clearbit(vmx, TERTIARY_EXEC_IPI_VIRT);
4424 	}
4425 
4426 	vmx_update_msr_bitmap_x2apic(vcpu);
4427 }
4428 
4429 static u32 vmx_exec_control(struct vcpu_vmx *vmx)
4430 {
4431 	u32 exec_control = vmcs_config.cpu_based_exec_ctrl;
4432 
4433 	/*
4434 	 * Not used by KVM, but fully supported for nesting, i.e. are allowed in
4435 	 * vmcs12 and propagated to vmcs02 when set in vmcs12.
4436 	 */
4437 	exec_control &= ~(CPU_BASED_RDTSC_EXITING |
4438 			  CPU_BASED_USE_IO_BITMAPS |
4439 			  CPU_BASED_MONITOR_TRAP_FLAG |
4440 			  CPU_BASED_PAUSE_EXITING);
4441 
4442 	/* INTR_WINDOW_EXITING and NMI_WINDOW_EXITING are toggled dynamically */
4443 	exec_control &= ~(CPU_BASED_INTR_WINDOW_EXITING |
4444 			  CPU_BASED_NMI_WINDOW_EXITING);
4445 
4446 	if (vmx->vcpu.arch.switch_db_regs & KVM_DEBUGREG_WONT_EXIT)
4447 		exec_control &= ~CPU_BASED_MOV_DR_EXITING;
4448 
4449 	if (!cpu_need_tpr_shadow(&vmx->vcpu))
4450 		exec_control &= ~CPU_BASED_TPR_SHADOW;
4451 
4452 #ifdef CONFIG_X86_64
4453 	if (exec_control & CPU_BASED_TPR_SHADOW)
4454 		exec_control &= ~(CPU_BASED_CR8_LOAD_EXITING |
4455 				  CPU_BASED_CR8_STORE_EXITING);
4456 	else
4457 		exec_control |= CPU_BASED_CR8_STORE_EXITING |
4458 				CPU_BASED_CR8_LOAD_EXITING;
4459 #endif
4460 	/* No need to intercept CR3 access or INVPLG when using EPT. */
4461 	if (enable_ept)
4462 		exec_control &= ~(CPU_BASED_CR3_LOAD_EXITING |
4463 				  CPU_BASED_CR3_STORE_EXITING |
4464 				  CPU_BASED_INVLPG_EXITING);
4465 	if (kvm_mwait_in_guest(vmx->vcpu.kvm))
4466 		exec_control &= ~(CPU_BASED_MWAIT_EXITING |
4467 				CPU_BASED_MONITOR_EXITING);
4468 	if (kvm_hlt_in_guest(vmx->vcpu.kvm))
4469 		exec_control &= ~CPU_BASED_HLT_EXITING;
4470 	return exec_control;
4471 }
4472 
4473 static u64 vmx_tertiary_exec_control(struct vcpu_vmx *vmx)
4474 {
4475 	u64 exec_control = vmcs_config.cpu_based_3rd_exec_ctrl;
4476 
4477 	/*
4478 	 * IPI virtualization relies on APICv. Disable IPI virtualization if
4479 	 * APICv is inhibited.
4480 	 */
4481 	if (!enable_ipiv || !kvm_vcpu_apicv_active(&vmx->vcpu))
4482 		exec_control &= ~TERTIARY_EXEC_IPI_VIRT;
4483 
4484 	return exec_control;
4485 }
4486 
4487 /*
4488  * Adjust a single secondary execution control bit to intercept/allow an
4489  * instruction in the guest.  This is usually done based on whether or not a
4490  * feature has been exposed to the guest in order to correctly emulate faults.
4491  */
4492 static inline void
4493 vmx_adjust_secondary_exec_control(struct vcpu_vmx *vmx, u32 *exec_control,
4494 				  u32 control, bool enabled, bool exiting)
4495 {
4496 	/*
4497 	 * If the control is for an opt-in feature, clear the control if the
4498 	 * feature is not exposed to the guest, i.e. not enabled.  If the
4499 	 * control is opt-out, i.e. an exiting control, clear the control if
4500 	 * the feature _is_ exposed to the guest, i.e. exiting/interception is
4501 	 * disabled for the associated instruction.  Note, the caller is
4502 	 * responsible presetting exec_control to set all supported bits.
4503 	 */
4504 	if (enabled == exiting)
4505 		*exec_control &= ~control;
4506 
4507 	/*
4508 	 * Update the nested MSR settings so that a nested VMM can/can't set
4509 	 * controls for features that are/aren't exposed to the guest.
4510 	 */
4511 	if (nested) {
4512 		/*
4513 		 * All features that can be added or removed to VMX MSRs must
4514 		 * be supported in the first place for nested virtualization.
4515 		 */
4516 		if (WARN_ON_ONCE(!(vmcs_config.nested.secondary_ctls_high & control)))
4517 			enabled = false;
4518 
4519 		if (enabled)
4520 			vmx->nested.msrs.secondary_ctls_high |= control;
4521 		else
4522 			vmx->nested.msrs.secondary_ctls_high &= ~control;
4523 	}
4524 }
4525 
4526 /*
4527  * Wrapper macro for the common case of adjusting a secondary execution control
4528  * based on a single guest CPUID bit, with a dedicated feature bit.  This also
4529  * verifies that the control is actually supported by KVM and hardware.
4530  */
4531 #define vmx_adjust_sec_exec_control(vmx, exec_control, name, feat_name, ctrl_name, exiting) \
4532 ({									 \
4533 	bool __enabled;							 \
4534 									 \
4535 	if (cpu_has_vmx_##name()) {					 \
4536 		__enabled = guest_cpuid_has(&(vmx)->vcpu,		 \
4537 					    X86_FEATURE_##feat_name);	 \
4538 		vmx_adjust_secondary_exec_control(vmx, exec_control,	 \
4539 			SECONDARY_EXEC_##ctrl_name, __enabled, exiting); \
4540 	}								 \
4541 })
4542 
4543 /* More macro magic for ENABLE_/opt-in versus _EXITING/opt-out controls. */
4544 #define vmx_adjust_sec_exec_feature(vmx, exec_control, lname, uname) \
4545 	vmx_adjust_sec_exec_control(vmx, exec_control, lname, uname, ENABLE_##uname, false)
4546 
4547 #define vmx_adjust_sec_exec_exiting(vmx, exec_control, lname, uname) \
4548 	vmx_adjust_sec_exec_control(vmx, exec_control, lname, uname, uname##_EXITING, true)
4549 
4550 static u32 vmx_secondary_exec_control(struct vcpu_vmx *vmx)
4551 {
4552 	struct kvm_vcpu *vcpu = &vmx->vcpu;
4553 
4554 	u32 exec_control = vmcs_config.cpu_based_2nd_exec_ctrl;
4555 
4556 	if (vmx_pt_mode_is_system())
4557 		exec_control &= ~(SECONDARY_EXEC_PT_USE_GPA | SECONDARY_EXEC_PT_CONCEAL_VMX);
4558 	if (!cpu_need_virtualize_apic_accesses(vcpu))
4559 		exec_control &= ~SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES;
4560 	if (vmx->vpid == 0)
4561 		exec_control &= ~SECONDARY_EXEC_ENABLE_VPID;
4562 	if (!enable_ept) {
4563 		exec_control &= ~SECONDARY_EXEC_ENABLE_EPT;
4564 		enable_unrestricted_guest = 0;
4565 	}
4566 	if (!enable_unrestricted_guest)
4567 		exec_control &= ~SECONDARY_EXEC_UNRESTRICTED_GUEST;
4568 	if (kvm_pause_in_guest(vmx->vcpu.kvm))
4569 		exec_control &= ~SECONDARY_EXEC_PAUSE_LOOP_EXITING;
4570 	if (!kvm_vcpu_apicv_active(vcpu))
4571 		exec_control &= ~(SECONDARY_EXEC_APIC_REGISTER_VIRT |
4572 				  SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY);
4573 	exec_control &= ~SECONDARY_EXEC_VIRTUALIZE_X2APIC_MODE;
4574 
4575 	/*
4576 	 * KVM doesn't support VMFUNC for L1, but the control is set in KVM's
4577 	 * base configuration as KVM emulates VMFUNC[EPTP_SWITCHING] for L2.
4578 	 */
4579 	exec_control &= ~SECONDARY_EXEC_ENABLE_VMFUNC;
4580 
4581 	/* SECONDARY_EXEC_DESC is enabled/disabled on writes to CR4.UMIP,
4582 	 * in vmx_set_cr4.  */
4583 	exec_control &= ~SECONDARY_EXEC_DESC;
4584 
4585 	/* SECONDARY_EXEC_SHADOW_VMCS is enabled when L1 executes VMPTRLD
4586 	   (handle_vmptrld).
4587 	   We can NOT enable shadow_vmcs here because we don't have yet
4588 	   a current VMCS12
4589 	*/
4590 	exec_control &= ~SECONDARY_EXEC_SHADOW_VMCS;
4591 
4592 	/*
4593 	 * PML is enabled/disabled when dirty logging of memsmlots changes, but
4594 	 * it needs to be set here when dirty logging is already active, e.g.
4595 	 * if this vCPU was created after dirty logging was enabled.
4596 	 */
4597 	if (!enable_pml || !atomic_read(&vcpu->kvm->nr_memslots_dirty_logging))
4598 		exec_control &= ~SECONDARY_EXEC_ENABLE_PML;
4599 
4600 	if (cpu_has_vmx_xsaves()) {
4601 		/* Exposing XSAVES only when XSAVE is exposed */
4602 		bool xsaves_enabled =
4603 			boot_cpu_has(X86_FEATURE_XSAVE) &&
4604 			guest_cpuid_has(vcpu, X86_FEATURE_XSAVE) &&
4605 			guest_cpuid_has(vcpu, X86_FEATURE_XSAVES);
4606 
4607 		vcpu->arch.xsaves_enabled = xsaves_enabled;
4608 
4609 		vmx_adjust_secondary_exec_control(vmx, &exec_control,
4610 						  SECONDARY_EXEC_XSAVES,
4611 						  xsaves_enabled, false);
4612 	}
4613 
4614 	/*
4615 	 * RDPID is also gated by ENABLE_RDTSCP, turn on the control if either
4616 	 * feature is exposed to the guest.  This creates a virtualization hole
4617 	 * if both are supported in hardware but only one is exposed to the
4618 	 * guest, but letting the guest execute RDTSCP or RDPID when either one
4619 	 * is advertised is preferable to emulating the advertised instruction
4620 	 * in KVM on #UD, and obviously better than incorrectly injecting #UD.
4621 	 */
4622 	if (cpu_has_vmx_rdtscp()) {
4623 		bool rdpid_or_rdtscp_enabled =
4624 			guest_cpuid_has(vcpu, X86_FEATURE_RDTSCP) ||
4625 			guest_cpuid_has(vcpu, X86_FEATURE_RDPID);
4626 
4627 		vmx_adjust_secondary_exec_control(vmx, &exec_control,
4628 						  SECONDARY_EXEC_ENABLE_RDTSCP,
4629 						  rdpid_or_rdtscp_enabled, false);
4630 	}
4631 	vmx_adjust_sec_exec_feature(vmx, &exec_control, invpcid, INVPCID);
4632 
4633 	vmx_adjust_sec_exec_exiting(vmx, &exec_control, rdrand, RDRAND);
4634 	vmx_adjust_sec_exec_exiting(vmx, &exec_control, rdseed, RDSEED);
4635 
4636 	vmx_adjust_sec_exec_control(vmx, &exec_control, waitpkg, WAITPKG,
4637 				    ENABLE_USR_WAIT_PAUSE, false);
4638 
4639 	if (!vcpu->kvm->arch.bus_lock_detection_enabled)
4640 		exec_control &= ~SECONDARY_EXEC_BUS_LOCK_DETECTION;
4641 
4642 	if (!kvm_notify_vmexit_enabled(vcpu->kvm))
4643 		exec_control &= ~SECONDARY_EXEC_NOTIFY_VM_EXITING;
4644 
4645 	return exec_control;
4646 }
4647 
4648 static inline int vmx_get_pid_table_order(struct kvm *kvm)
4649 {
4650 	return get_order(kvm->arch.max_vcpu_ids * sizeof(*to_kvm_vmx(kvm)->pid_table));
4651 }
4652 
4653 static int vmx_alloc_ipiv_pid_table(struct kvm *kvm)
4654 {
4655 	struct page *pages;
4656 	struct kvm_vmx *kvm_vmx = to_kvm_vmx(kvm);
4657 
4658 	if (!irqchip_in_kernel(kvm) || !enable_ipiv)
4659 		return 0;
4660 
4661 	if (kvm_vmx->pid_table)
4662 		return 0;
4663 
4664 	pages = alloc_pages(GFP_KERNEL_ACCOUNT | __GFP_ZERO,
4665 			    vmx_get_pid_table_order(kvm));
4666 	if (!pages)
4667 		return -ENOMEM;
4668 
4669 	kvm_vmx->pid_table = (void *)page_address(pages);
4670 	return 0;
4671 }
4672 
4673 static int vmx_vcpu_precreate(struct kvm *kvm)
4674 {
4675 	return vmx_alloc_ipiv_pid_table(kvm);
4676 }
4677 
4678 #define VMX_XSS_EXIT_BITMAP 0
4679 
4680 static void init_vmcs(struct vcpu_vmx *vmx)
4681 {
4682 	struct kvm *kvm = vmx->vcpu.kvm;
4683 	struct kvm_vmx *kvm_vmx = to_kvm_vmx(kvm);
4684 
4685 	if (nested)
4686 		nested_vmx_set_vmcs_shadowing_bitmap();
4687 
4688 	if (cpu_has_vmx_msr_bitmap())
4689 		vmcs_write64(MSR_BITMAP, __pa(vmx->vmcs01.msr_bitmap));
4690 
4691 	vmcs_write64(VMCS_LINK_POINTER, INVALID_GPA); /* 22.3.1.5 */
4692 
4693 	/* Control */
4694 	pin_controls_set(vmx, vmx_pin_based_exec_ctrl(vmx));
4695 
4696 	exec_controls_set(vmx, vmx_exec_control(vmx));
4697 
4698 	if (cpu_has_secondary_exec_ctrls())
4699 		secondary_exec_controls_set(vmx, vmx_secondary_exec_control(vmx));
4700 
4701 	if (cpu_has_tertiary_exec_ctrls())
4702 		tertiary_exec_controls_set(vmx, vmx_tertiary_exec_control(vmx));
4703 
4704 	if (enable_apicv && lapic_in_kernel(&vmx->vcpu)) {
4705 		vmcs_write64(EOI_EXIT_BITMAP0, 0);
4706 		vmcs_write64(EOI_EXIT_BITMAP1, 0);
4707 		vmcs_write64(EOI_EXIT_BITMAP2, 0);
4708 		vmcs_write64(EOI_EXIT_BITMAP3, 0);
4709 
4710 		vmcs_write16(GUEST_INTR_STATUS, 0);
4711 
4712 		vmcs_write16(POSTED_INTR_NV, POSTED_INTR_VECTOR);
4713 		vmcs_write64(POSTED_INTR_DESC_ADDR, __pa((&vmx->pi_desc)));
4714 	}
4715 
4716 	if (vmx_can_use_ipiv(&vmx->vcpu)) {
4717 		vmcs_write64(PID_POINTER_TABLE, __pa(kvm_vmx->pid_table));
4718 		vmcs_write16(LAST_PID_POINTER_INDEX, kvm->arch.max_vcpu_ids - 1);
4719 	}
4720 
4721 	if (!kvm_pause_in_guest(kvm)) {
4722 		vmcs_write32(PLE_GAP, ple_gap);
4723 		vmx->ple_window = ple_window;
4724 		vmx->ple_window_dirty = true;
4725 	}
4726 
4727 	if (kvm_notify_vmexit_enabled(kvm))
4728 		vmcs_write32(NOTIFY_WINDOW, kvm->arch.notify_window);
4729 
4730 	vmcs_write32(PAGE_FAULT_ERROR_CODE_MASK, 0);
4731 	vmcs_write32(PAGE_FAULT_ERROR_CODE_MATCH, 0);
4732 	vmcs_write32(CR3_TARGET_COUNT, 0);           /* 22.2.1 */
4733 
4734 	vmcs_write16(HOST_FS_SELECTOR, 0);            /* 22.2.4 */
4735 	vmcs_write16(HOST_GS_SELECTOR, 0);            /* 22.2.4 */
4736 	vmx_set_constant_host_state(vmx);
4737 	vmcs_writel(HOST_FS_BASE, 0); /* 22.2.4 */
4738 	vmcs_writel(HOST_GS_BASE, 0); /* 22.2.4 */
4739 
4740 	if (cpu_has_vmx_vmfunc())
4741 		vmcs_write64(VM_FUNCTION_CONTROL, 0);
4742 
4743 	vmcs_write32(VM_EXIT_MSR_STORE_COUNT, 0);
4744 	vmcs_write32(VM_EXIT_MSR_LOAD_COUNT, 0);
4745 	vmcs_write64(VM_EXIT_MSR_LOAD_ADDR, __pa(vmx->msr_autoload.host.val));
4746 	vmcs_write32(VM_ENTRY_MSR_LOAD_COUNT, 0);
4747 	vmcs_write64(VM_ENTRY_MSR_LOAD_ADDR, __pa(vmx->msr_autoload.guest.val));
4748 
4749 	if (vmcs_config.vmentry_ctrl & VM_ENTRY_LOAD_IA32_PAT)
4750 		vmcs_write64(GUEST_IA32_PAT, vmx->vcpu.arch.pat);
4751 
4752 	vm_exit_controls_set(vmx, vmx_vmexit_ctrl());
4753 
4754 	/* 22.2.1, 20.8.1 */
4755 	vm_entry_controls_set(vmx, vmx_vmentry_ctrl());
4756 
4757 	vmx->vcpu.arch.cr0_guest_owned_bits = vmx_l1_guest_owned_cr0_bits();
4758 	vmcs_writel(CR0_GUEST_HOST_MASK, ~vmx->vcpu.arch.cr0_guest_owned_bits);
4759 
4760 	set_cr4_guest_host_mask(vmx);
4761 
4762 	if (vmx->vpid != 0)
4763 		vmcs_write16(VIRTUAL_PROCESSOR_ID, vmx->vpid);
4764 
4765 	if (cpu_has_vmx_xsaves())
4766 		vmcs_write64(XSS_EXIT_BITMAP, VMX_XSS_EXIT_BITMAP);
4767 
4768 	if (enable_pml) {
4769 		vmcs_write64(PML_ADDRESS, page_to_phys(vmx->pml_pg));
4770 		vmcs_write16(GUEST_PML_INDEX, PML_ENTITY_NUM - 1);
4771 	}
4772 
4773 	vmx_write_encls_bitmap(&vmx->vcpu, NULL);
4774 
4775 	if (vmx_pt_mode_is_host_guest()) {
4776 		memset(&vmx->pt_desc, 0, sizeof(vmx->pt_desc));
4777 		/* Bit[6~0] are forced to 1, writes are ignored. */
4778 		vmx->pt_desc.guest.output_mask = 0x7F;
4779 		vmcs_write64(GUEST_IA32_RTIT_CTL, 0);
4780 	}
4781 
4782 	vmcs_write32(GUEST_SYSENTER_CS, 0);
4783 	vmcs_writel(GUEST_SYSENTER_ESP, 0);
4784 	vmcs_writel(GUEST_SYSENTER_EIP, 0);
4785 	vmcs_write64(GUEST_IA32_DEBUGCTL, 0);
4786 
4787 	if (cpu_has_vmx_tpr_shadow()) {
4788 		vmcs_write64(VIRTUAL_APIC_PAGE_ADDR, 0);
4789 		if (cpu_need_tpr_shadow(&vmx->vcpu))
4790 			vmcs_write64(VIRTUAL_APIC_PAGE_ADDR,
4791 				     __pa(vmx->vcpu.arch.apic->regs));
4792 		vmcs_write32(TPR_THRESHOLD, 0);
4793 	}
4794 
4795 	vmx_setup_uret_msrs(vmx);
4796 }
4797 
4798 static void __vmx_vcpu_reset(struct kvm_vcpu *vcpu)
4799 {
4800 	struct vcpu_vmx *vmx = to_vmx(vcpu);
4801 
4802 	init_vmcs(vmx);
4803 
4804 	if (nested)
4805 		memcpy(&vmx->nested.msrs, &vmcs_config.nested, sizeof(vmx->nested.msrs));
4806 
4807 	vcpu_setup_sgx_lepubkeyhash(vcpu);
4808 
4809 	vmx->nested.posted_intr_nv = -1;
4810 	vmx->nested.vmxon_ptr = INVALID_GPA;
4811 	vmx->nested.current_vmptr = INVALID_GPA;
4812 	vmx->nested.hv_evmcs_vmptr = EVMPTR_INVALID;
4813 
4814 	vcpu->arch.microcode_version = 0x100000000ULL;
4815 	vmx->msr_ia32_feature_control_valid_bits = FEAT_CTL_LOCKED;
4816 
4817 	/*
4818 	 * Enforce invariant: pi_desc.nv is always either POSTED_INTR_VECTOR
4819 	 * or POSTED_INTR_WAKEUP_VECTOR.
4820 	 */
4821 	vmx->pi_desc.nv = POSTED_INTR_VECTOR;
4822 	vmx->pi_desc.sn = 1;
4823 }
4824 
4825 static void vmx_vcpu_reset(struct kvm_vcpu *vcpu, bool init_event)
4826 {
4827 	struct vcpu_vmx *vmx = to_vmx(vcpu);
4828 
4829 	if (!init_event)
4830 		__vmx_vcpu_reset(vcpu);
4831 
4832 	vmx->rmode.vm86_active = 0;
4833 	vmx->spec_ctrl = 0;
4834 
4835 	vmx->msr_ia32_umwait_control = 0;
4836 
4837 	vmx->hv_deadline_tsc = -1;
4838 	kvm_set_cr8(vcpu, 0);
4839 
4840 	vmx_segment_cache_clear(vmx);
4841 	kvm_register_mark_available(vcpu, VCPU_EXREG_SEGMENTS);
4842 
4843 	seg_setup(VCPU_SREG_CS);
4844 	vmcs_write16(GUEST_CS_SELECTOR, 0xf000);
4845 	vmcs_writel(GUEST_CS_BASE, 0xffff0000ul);
4846 
4847 	seg_setup(VCPU_SREG_DS);
4848 	seg_setup(VCPU_SREG_ES);
4849 	seg_setup(VCPU_SREG_FS);
4850 	seg_setup(VCPU_SREG_GS);
4851 	seg_setup(VCPU_SREG_SS);
4852 
4853 	vmcs_write16(GUEST_TR_SELECTOR, 0);
4854 	vmcs_writel(GUEST_TR_BASE, 0);
4855 	vmcs_write32(GUEST_TR_LIMIT, 0xffff);
4856 	vmcs_write32(GUEST_TR_AR_BYTES, 0x008b);
4857 
4858 	vmcs_write16(GUEST_LDTR_SELECTOR, 0);
4859 	vmcs_writel(GUEST_LDTR_BASE, 0);
4860 	vmcs_write32(GUEST_LDTR_LIMIT, 0xffff);
4861 	vmcs_write32(GUEST_LDTR_AR_BYTES, 0x00082);
4862 
4863 	vmcs_writel(GUEST_GDTR_BASE, 0);
4864 	vmcs_write32(GUEST_GDTR_LIMIT, 0xffff);
4865 
4866 	vmcs_writel(GUEST_IDTR_BASE, 0);
4867 	vmcs_write32(GUEST_IDTR_LIMIT, 0xffff);
4868 
4869 	vmcs_write32(GUEST_ACTIVITY_STATE, GUEST_ACTIVITY_ACTIVE);
4870 	vmcs_write32(GUEST_INTERRUPTIBILITY_INFO, 0);
4871 	vmcs_writel(GUEST_PENDING_DBG_EXCEPTIONS, 0);
4872 	if (kvm_mpx_supported())
4873 		vmcs_write64(GUEST_BNDCFGS, 0);
4874 
4875 	vmcs_write32(VM_ENTRY_INTR_INFO_FIELD, 0);  /* 22.2.1 */
4876 
4877 	kvm_make_request(KVM_REQ_APIC_PAGE_RELOAD, vcpu);
4878 
4879 	vpid_sync_context(vmx->vpid);
4880 
4881 	vmx_update_fb_clear_dis(vcpu, vmx);
4882 }
4883 
4884 static void vmx_enable_irq_window(struct kvm_vcpu *vcpu)
4885 {
4886 	exec_controls_setbit(to_vmx(vcpu), CPU_BASED_INTR_WINDOW_EXITING);
4887 }
4888 
4889 static void vmx_enable_nmi_window(struct kvm_vcpu *vcpu)
4890 {
4891 	if (!enable_vnmi ||
4892 	    vmcs_read32(GUEST_INTERRUPTIBILITY_INFO) & GUEST_INTR_STATE_STI) {
4893 		vmx_enable_irq_window(vcpu);
4894 		return;
4895 	}
4896 
4897 	exec_controls_setbit(to_vmx(vcpu), CPU_BASED_NMI_WINDOW_EXITING);
4898 }
4899 
4900 static void vmx_inject_irq(struct kvm_vcpu *vcpu, bool reinjected)
4901 {
4902 	struct vcpu_vmx *vmx = to_vmx(vcpu);
4903 	uint32_t intr;
4904 	int irq = vcpu->arch.interrupt.nr;
4905 
4906 	trace_kvm_inj_virq(irq, vcpu->arch.interrupt.soft, reinjected);
4907 
4908 	++vcpu->stat.irq_injections;
4909 	if (vmx->rmode.vm86_active) {
4910 		int inc_eip = 0;
4911 		if (vcpu->arch.interrupt.soft)
4912 			inc_eip = vcpu->arch.event_exit_inst_len;
4913 		kvm_inject_realmode_interrupt(vcpu, irq, inc_eip);
4914 		return;
4915 	}
4916 	intr = irq | INTR_INFO_VALID_MASK;
4917 	if (vcpu->arch.interrupt.soft) {
4918 		intr |= INTR_TYPE_SOFT_INTR;
4919 		vmcs_write32(VM_ENTRY_INSTRUCTION_LEN,
4920 			     vmx->vcpu.arch.event_exit_inst_len);
4921 	} else
4922 		intr |= INTR_TYPE_EXT_INTR;
4923 	vmcs_write32(VM_ENTRY_INTR_INFO_FIELD, intr);
4924 
4925 	vmx_clear_hlt(vcpu);
4926 }
4927 
4928 static void vmx_inject_nmi(struct kvm_vcpu *vcpu)
4929 {
4930 	struct vcpu_vmx *vmx = to_vmx(vcpu);
4931 
4932 	if (!enable_vnmi) {
4933 		/*
4934 		 * Tracking the NMI-blocked state in software is built upon
4935 		 * finding the next open IRQ window. This, in turn, depends on
4936 		 * well-behaving guests: They have to keep IRQs disabled at
4937 		 * least as long as the NMI handler runs. Otherwise we may
4938 		 * cause NMI nesting, maybe breaking the guest. But as this is
4939 		 * highly unlikely, we can live with the residual risk.
4940 		 */
4941 		vmx->loaded_vmcs->soft_vnmi_blocked = 1;
4942 		vmx->loaded_vmcs->vnmi_blocked_time = 0;
4943 	}
4944 
4945 	++vcpu->stat.nmi_injections;
4946 	vmx->loaded_vmcs->nmi_known_unmasked = false;
4947 
4948 	if (vmx->rmode.vm86_active) {
4949 		kvm_inject_realmode_interrupt(vcpu, NMI_VECTOR, 0);
4950 		return;
4951 	}
4952 
4953 	vmcs_write32(VM_ENTRY_INTR_INFO_FIELD,
4954 			INTR_TYPE_NMI_INTR | INTR_INFO_VALID_MASK | NMI_VECTOR);
4955 
4956 	vmx_clear_hlt(vcpu);
4957 }
4958 
4959 bool vmx_get_nmi_mask(struct kvm_vcpu *vcpu)
4960 {
4961 	struct vcpu_vmx *vmx = to_vmx(vcpu);
4962 	bool masked;
4963 
4964 	if (!enable_vnmi)
4965 		return vmx->loaded_vmcs->soft_vnmi_blocked;
4966 	if (vmx->loaded_vmcs->nmi_known_unmasked)
4967 		return false;
4968 	masked = vmcs_read32(GUEST_INTERRUPTIBILITY_INFO) & GUEST_INTR_STATE_NMI;
4969 	vmx->loaded_vmcs->nmi_known_unmasked = !masked;
4970 	return masked;
4971 }
4972 
4973 void vmx_set_nmi_mask(struct kvm_vcpu *vcpu, bool masked)
4974 {
4975 	struct vcpu_vmx *vmx = to_vmx(vcpu);
4976 
4977 	if (!enable_vnmi) {
4978 		if (vmx->loaded_vmcs->soft_vnmi_blocked != masked) {
4979 			vmx->loaded_vmcs->soft_vnmi_blocked = masked;
4980 			vmx->loaded_vmcs->vnmi_blocked_time = 0;
4981 		}
4982 	} else {
4983 		vmx->loaded_vmcs->nmi_known_unmasked = !masked;
4984 		if (masked)
4985 			vmcs_set_bits(GUEST_INTERRUPTIBILITY_INFO,
4986 				      GUEST_INTR_STATE_NMI);
4987 		else
4988 			vmcs_clear_bits(GUEST_INTERRUPTIBILITY_INFO,
4989 					GUEST_INTR_STATE_NMI);
4990 	}
4991 }
4992 
4993 bool vmx_nmi_blocked(struct kvm_vcpu *vcpu)
4994 {
4995 	if (is_guest_mode(vcpu) && nested_exit_on_nmi(vcpu))
4996 		return false;
4997 
4998 	if (!enable_vnmi && to_vmx(vcpu)->loaded_vmcs->soft_vnmi_blocked)
4999 		return true;
5000 
5001 	return (vmcs_read32(GUEST_INTERRUPTIBILITY_INFO) &
5002 		(GUEST_INTR_STATE_MOV_SS | GUEST_INTR_STATE_STI |
5003 		 GUEST_INTR_STATE_NMI));
5004 }
5005 
5006 static int vmx_nmi_allowed(struct kvm_vcpu *vcpu, bool for_injection)
5007 {
5008 	if (to_vmx(vcpu)->nested.nested_run_pending)
5009 		return -EBUSY;
5010 
5011 	/* An NMI must not be injected into L2 if it's supposed to VM-Exit.  */
5012 	if (for_injection && is_guest_mode(vcpu) && nested_exit_on_nmi(vcpu))
5013 		return -EBUSY;
5014 
5015 	return !vmx_nmi_blocked(vcpu);
5016 }
5017 
5018 bool vmx_interrupt_blocked(struct kvm_vcpu *vcpu)
5019 {
5020 	if (is_guest_mode(vcpu) && nested_exit_on_intr(vcpu))
5021 		return false;
5022 
5023 	return !(vmx_get_rflags(vcpu) & X86_EFLAGS_IF) ||
5024 	       (vmcs_read32(GUEST_INTERRUPTIBILITY_INFO) &
5025 		(GUEST_INTR_STATE_STI | GUEST_INTR_STATE_MOV_SS));
5026 }
5027 
5028 static int vmx_interrupt_allowed(struct kvm_vcpu *vcpu, bool for_injection)
5029 {
5030 	if (to_vmx(vcpu)->nested.nested_run_pending)
5031 		return -EBUSY;
5032 
5033 	/*
5034 	 * An IRQ must not be injected into L2 if it's supposed to VM-Exit,
5035 	 * e.g. if the IRQ arrived asynchronously after checking nested events.
5036 	 */
5037 	if (for_injection && is_guest_mode(vcpu) && nested_exit_on_intr(vcpu))
5038 		return -EBUSY;
5039 
5040 	return !vmx_interrupt_blocked(vcpu);
5041 }
5042 
5043 static int vmx_set_tss_addr(struct kvm *kvm, unsigned int addr)
5044 {
5045 	void __user *ret;
5046 
5047 	if (enable_unrestricted_guest)
5048 		return 0;
5049 
5050 	mutex_lock(&kvm->slots_lock);
5051 	ret = __x86_set_memory_region(kvm, TSS_PRIVATE_MEMSLOT, addr,
5052 				      PAGE_SIZE * 3);
5053 	mutex_unlock(&kvm->slots_lock);
5054 
5055 	if (IS_ERR(ret))
5056 		return PTR_ERR(ret);
5057 
5058 	to_kvm_vmx(kvm)->tss_addr = addr;
5059 
5060 	return init_rmode_tss(kvm, ret);
5061 }
5062 
5063 static int vmx_set_identity_map_addr(struct kvm *kvm, u64 ident_addr)
5064 {
5065 	to_kvm_vmx(kvm)->ept_identity_map_addr = ident_addr;
5066 	return 0;
5067 }
5068 
5069 static bool rmode_exception(struct kvm_vcpu *vcpu, int vec)
5070 {
5071 	switch (vec) {
5072 	case BP_VECTOR:
5073 		/*
5074 		 * Update instruction length as we may reinject the exception
5075 		 * from user space while in guest debugging mode.
5076 		 */
5077 		to_vmx(vcpu)->vcpu.arch.event_exit_inst_len =
5078 			vmcs_read32(VM_EXIT_INSTRUCTION_LEN);
5079 		if (vcpu->guest_debug & KVM_GUESTDBG_USE_SW_BP)
5080 			return false;
5081 		fallthrough;
5082 	case DB_VECTOR:
5083 		return !(vcpu->guest_debug &
5084 			(KVM_GUESTDBG_SINGLESTEP | KVM_GUESTDBG_USE_HW_BP));
5085 	case DE_VECTOR:
5086 	case OF_VECTOR:
5087 	case BR_VECTOR:
5088 	case UD_VECTOR:
5089 	case DF_VECTOR:
5090 	case SS_VECTOR:
5091 	case GP_VECTOR:
5092 	case MF_VECTOR:
5093 		return true;
5094 	}
5095 	return false;
5096 }
5097 
5098 static int handle_rmode_exception(struct kvm_vcpu *vcpu,
5099 				  int vec, u32 err_code)
5100 {
5101 	/*
5102 	 * Instruction with address size override prefix opcode 0x67
5103 	 * Cause the #SS fault with 0 error code in VM86 mode.
5104 	 */
5105 	if (((vec == GP_VECTOR) || (vec == SS_VECTOR)) && err_code == 0) {
5106 		if (kvm_emulate_instruction(vcpu, 0)) {
5107 			if (vcpu->arch.halt_request) {
5108 				vcpu->arch.halt_request = 0;
5109 				return kvm_emulate_halt_noskip(vcpu);
5110 			}
5111 			return 1;
5112 		}
5113 		return 0;
5114 	}
5115 
5116 	/*
5117 	 * Forward all other exceptions that are valid in real mode.
5118 	 * FIXME: Breaks guest debugging in real mode, needs to be fixed with
5119 	 *        the required debugging infrastructure rework.
5120 	 */
5121 	kvm_queue_exception(vcpu, vec);
5122 	return 1;
5123 }
5124 
5125 static int handle_machine_check(struct kvm_vcpu *vcpu)
5126 {
5127 	/* handled by vmx_vcpu_run() */
5128 	return 1;
5129 }
5130 
5131 /*
5132  * If the host has split lock detection disabled, then #AC is
5133  * unconditionally injected into the guest, which is the pre split lock
5134  * detection behaviour.
5135  *
5136  * If the host has split lock detection enabled then #AC is
5137  * only injected into the guest when:
5138  *  - Guest CPL == 3 (user mode)
5139  *  - Guest has #AC detection enabled in CR0
5140  *  - Guest EFLAGS has AC bit set
5141  */
5142 bool vmx_guest_inject_ac(struct kvm_vcpu *vcpu)
5143 {
5144 	if (!boot_cpu_has(X86_FEATURE_SPLIT_LOCK_DETECT))
5145 		return true;
5146 
5147 	return vmx_get_cpl(vcpu) == 3 && kvm_is_cr0_bit_set(vcpu, X86_CR0_AM) &&
5148 	       (kvm_get_rflags(vcpu) & X86_EFLAGS_AC);
5149 }
5150 
5151 static int handle_exception_nmi(struct kvm_vcpu *vcpu)
5152 {
5153 	struct vcpu_vmx *vmx = to_vmx(vcpu);
5154 	struct kvm_run *kvm_run = vcpu->run;
5155 	u32 intr_info, ex_no, error_code;
5156 	unsigned long cr2, dr6;
5157 	u32 vect_info;
5158 
5159 	vect_info = vmx->idt_vectoring_info;
5160 	intr_info = vmx_get_intr_info(vcpu);
5161 
5162 	/*
5163 	 * Machine checks are handled by handle_exception_irqoff(), or by
5164 	 * vmx_vcpu_run() if a #MC occurs on VM-Entry.  NMIs are handled by
5165 	 * vmx_vcpu_enter_exit().
5166 	 */
5167 	if (is_machine_check(intr_info) || is_nmi(intr_info))
5168 		return 1;
5169 
5170 	/*
5171 	 * Queue the exception here instead of in handle_nm_fault_irqoff().
5172 	 * This ensures the nested_vmx check is not skipped so vmexit can
5173 	 * be reflected to L1 (when it intercepts #NM) before reaching this
5174 	 * point.
5175 	 */
5176 	if (is_nm_fault(intr_info)) {
5177 		kvm_queue_exception(vcpu, NM_VECTOR);
5178 		return 1;
5179 	}
5180 
5181 	if (is_invalid_opcode(intr_info))
5182 		return handle_ud(vcpu);
5183 
5184 	error_code = 0;
5185 	if (intr_info & INTR_INFO_DELIVER_CODE_MASK)
5186 		error_code = vmcs_read32(VM_EXIT_INTR_ERROR_CODE);
5187 
5188 	if (!vmx->rmode.vm86_active && is_gp_fault(intr_info)) {
5189 		WARN_ON_ONCE(!enable_vmware_backdoor);
5190 
5191 		/*
5192 		 * VMware backdoor emulation on #GP interception only handles
5193 		 * IN{S}, OUT{S}, and RDPMC, none of which generate a non-zero
5194 		 * error code on #GP.
5195 		 */
5196 		if (error_code) {
5197 			kvm_queue_exception_e(vcpu, GP_VECTOR, error_code);
5198 			return 1;
5199 		}
5200 		return kvm_emulate_instruction(vcpu, EMULTYPE_VMWARE_GP);
5201 	}
5202 
5203 	/*
5204 	 * The #PF with PFEC.RSVD = 1 indicates the guest is accessing
5205 	 * MMIO, it is better to report an internal error.
5206 	 * See the comments in vmx_handle_exit.
5207 	 */
5208 	if ((vect_info & VECTORING_INFO_VALID_MASK) &&
5209 	    !(is_page_fault(intr_info) && !(error_code & PFERR_RSVD_MASK))) {
5210 		vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
5211 		vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_SIMUL_EX;
5212 		vcpu->run->internal.ndata = 4;
5213 		vcpu->run->internal.data[0] = vect_info;
5214 		vcpu->run->internal.data[1] = intr_info;
5215 		vcpu->run->internal.data[2] = error_code;
5216 		vcpu->run->internal.data[3] = vcpu->arch.last_vmentry_cpu;
5217 		return 0;
5218 	}
5219 
5220 	if (is_page_fault(intr_info)) {
5221 		cr2 = vmx_get_exit_qual(vcpu);
5222 		if (enable_ept && !vcpu->arch.apf.host_apf_flags) {
5223 			/*
5224 			 * EPT will cause page fault only if we need to
5225 			 * detect illegal GPAs.
5226 			 */
5227 			WARN_ON_ONCE(!allow_smaller_maxphyaddr);
5228 			kvm_fixup_and_inject_pf_error(vcpu, cr2, error_code);
5229 			return 1;
5230 		} else
5231 			return kvm_handle_page_fault(vcpu, error_code, cr2, NULL, 0);
5232 	}
5233 
5234 	ex_no = intr_info & INTR_INFO_VECTOR_MASK;
5235 
5236 	if (vmx->rmode.vm86_active && rmode_exception(vcpu, ex_no))
5237 		return handle_rmode_exception(vcpu, ex_no, error_code);
5238 
5239 	switch (ex_no) {
5240 	case DB_VECTOR:
5241 		dr6 = vmx_get_exit_qual(vcpu);
5242 		if (!(vcpu->guest_debug &
5243 		      (KVM_GUESTDBG_SINGLESTEP | KVM_GUESTDBG_USE_HW_BP))) {
5244 			/*
5245 			 * If the #DB was due to ICEBP, a.k.a. INT1, skip the
5246 			 * instruction.  ICEBP generates a trap-like #DB, but
5247 			 * despite its interception control being tied to #DB,
5248 			 * is an instruction intercept, i.e. the VM-Exit occurs
5249 			 * on the ICEBP itself.  Use the inner "skip" helper to
5250 			 * avoid single-step #DB and MTF updates, as ICEBP is
5251 			 * higher priority.  Note, skipping ICEBP still clears
5252 			 * STI and MOVSS blocking.
5253 			 *
5254 			 * For all other #DBs, set vmcs.PENDING_DBG_EXCEPTIONS.BS
5255 			 * if single-step is enabled in RFLAGS and STI or MOVSS
5256 			 * blocking is active, as the CPU doesn't set the bit
5257 			 * on VM-Exit due to #DB interception.  VM-Entry has a
5258 			 * consistency check that a single-step #DB is pending
5259 			 * in this scenario as the previous instruction cannot
5260 			 * have toggled RFLAGS.TF 0=>1 (because STI and POP/MOV
5261 			 * don't modify RFLAGS), therefore the one instruction
5262 			 * delay when activating single-step breakpoints must
5263 			 * have already expired.  Note, the CPU sets/clears BS
5264 			 * as appropriate for all other VM-Exits types.
5265 			 */
5266 			if (is_icebp(intr_info))
5267 				WARN_ON(!skip_emulated_instruction(vcpu));
5268 			else if ((vmx_get_rflags(vcpu) & X86_EFLAGS_TF) &&
5269 				 (vmcs_read32(GUEST_INTERRUPTIBILITY_INFO) &
5270 				  (GUEST_INTR_STATE_STI | GUEST_INTR_STATE_MOV_SS)))
5271 				vmcs_writel(GUEST_PENDING_DBG_EXCEPTIONS,
5272 					    vmcs_readl(GUEST_PENDING_DBG_EXCEPTIONS) | DR6_BS);
5273 
5274 			kvm_queue_exception_p(vcpu, DB_VECTOR, dr6);
5275 			return 1;
5276 		}
5277 		kvm_run->debug.arch.dr6 = dr6 | DR6_ACTIVE_LOW;
5278 		kvm_run->debug.arch.dr7 = vmcs_readl(GUEST_DR7);
5279 		fallthrough;
5280 	case BP_VECTOR:
5281 		/*
5282 		 * Update instruction length as we may reinject #BP from
5283 		 * user space while in guest debugging mode. Reading it for
5284 		 * #DB as well causes no harm, it is not used in that case.
5285 		 */
5286 		vmx->vcpu.arch.event_exit_inst_len =
5287 			vmcs_read32(VM_EXIT_INSTRUCTION_LEN);
5288 		kvm_run->exit_reason = KVM_EXIT_DEBUG;
5289 		kvm_run->debug.arch.pc = kvm_get_linear_rip(vcpu);
5290 		kvm_run->debug.arch.exception = ex_no;
5291 		break;
5292 	case AC_VECTOR:
5293 		if (vmx_guest_inject_ac(vcpu)) {
5294 			kvm_queue_exception_e(vcpu, AC_VECTOR, error_code);
5295 			return 1;
5296 		}
5297 
5298 		/*
5299 		 * Handle split lock. Depending on detection mode this will
5300 		 * either warn and disable split lock detection for this
5301 		 * task or force SIGBUS on it.
5302 		 */
5303 		if (handle_guest_split_lock(kvm_rip_read(vcpu)))
5304 			return 1;
5305 		fallthrough;
5306 	default:
5307 		kvm_run->exit_reason = KVM_EXIT_EXCEPTION;
5308 		kvm_run->ex.exception = ex_no;
5309 		kvm_run->ex.error_code = error_code;
5310 		break;
5311 	}
5312 	return 0;
5313 }
5314 
5315 static __always_inline int handle_external_interrupt(struct kvm_vcpu *vcpu)
5316 {
5317 	++vcpu->stat.irq_exits;
5318 	return 1;
5319 }
5320 
5321 static int handle_triple_fault(struct kvm_vcpu *vcpu)
5322 {
5323 	vcpu->run->exit_reason = KVM_EXIT_SHUTDOWN;
5324 	vcpu->mmio_needed = 0;
5325 	return 0;
5326 }
5327 
5328 static int handle_io(struct kvm_vcpu *vcpu)
5329 {
5330 	unsigned long exit_qualification;
5331 	int size, in, string;
5332 	unsigned port;
5333 
5334 	exit_qualification = vmx_get_exit_qual(vcpu);
5335 	string = (exit_qualification & 16) != 0;
5336 
5337 	++vcpu->stat.io_exits;
5338 
5339 	if (string)
5340 		return kvm_emulate_instruction(vcpu, 0);
5341 
5342 	port = exit_qualification >> 16;
5343 	size = (exit_qualification & 7) + 1;
5344 	in = (exit_qualification & 8) != 0;
5345 
5346 	return kvm_fast_pio(vcpu, size, port, in);
5347 }
5348 
5349 static void
5350 vmx_patch_hypercall(struct kvm_vcpu *vcpu, unsigned char *hypercall)
5351 {
5352 	/*
5353 	 * Patch in the VMCALL instruction:
5354 	 */
5355 	hypercall[0] = 0x0f;
5356 	hypercall[1] = 0x01;
5357 	hypercall[2] = 0xc1;
5358 }
5359 
5360 /* called to set cr0 as appropriate for a mov-to-cr0 exit. */
5361 static int handle_set_cr0(struct kvm_vcpu *vcpu, unsigned long val)
5362 {
5363 	if (is_guest_mode(vcpu)) {
5364 		struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
5365 		unsigned long orig_val = val;
5366 
5367 		/*
5368 		 * We get here when L2 changed cr0 in a way that did not change
5369 		 * any of L1's shadowed bits (see nested_vmx_exit_handled_cr),
5370 		 * but did change L0 shadowed bits. So we first calculate the
5371 		 * effective cr0 value that L1 would like to write into the
5372 		 * hardware. It consists of the L2-owned bits from the new
5373 		 * value combined with the L1-owned bits from L1's guest_cr0.
5374 		 */
5375 		val = (val & ~vmcs12->cr0_guest_host_mask) |
5376 			(vmcs12->guest_cr0 & vmcs12->cr0_guest_host_mask);
5377 
5378 		if (!nested_guest_cr0_valid(vcpu, val))
5379 			return 1;
5380 
5381 		if (kvm_set_cr0(vcpu, val))
5382 			return 1;
5383 		vmcs_writel(CR0_READ_SHADOW, orig_val);
5384 		return 0;
5385 	} else {
5386 		if (to_vmx(vcpu)->nested.vmxon &&
5387 		    !nested_host_cr0_valid(vcpu, val))
5388 			return 1;
5389 
5390 		return kvm_set_cr0(vcpu, val);
5391 	}
5392 }
5393 
5394 static int handle_set_cr4(struct kvm_vcpu *vcpu, unsigned long val)
5395 {
5396 	if (is_guest_mode(vcpu)) {
5397 		struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
5398 		unsigned long orig_val = val;
5399 
5400 		/* analogously to handle_set_cr0 */
5401 		val = (val & ~vmcs12->cr4_guest_host_mask) |
5402 			(vmcs12->guest_cr4 & vmcs12->cr4_guest_host_mask);
5403 		if (kvm_set_cr4(vcpu, val))
5404 			return 1;
5405 		vmcs_writel(CR4_READ_SHADOW, orig_val);
5406 		return 0;
5407 	} else
5408 		return kvm_set_cr4(vcpu, val);
5409 }
5410 
5411 static int handle_desc(struct kvm_vcpu *vcpu)
5412 {
5413 	/*
5414 	 * UMIP emulation relies on intercepting writes to CR4.UMIP, i.e. this
5415 	 * and other code needs to be updated if UMIP can be guest owned.
5416 	 */
5417 	BUILD_BUG_ON(KVM_POSSIBLE_CR4_GUEST_BITS & X86_CR4_UMIP);
5418 
5419 	WARN_ON_ONCE(!kvm_is_cr4_bit_set(vcpu, X86_CR4_UMIP));
5420 	return kvm_emulate_instruction(vcpu, 0);
5421 }
5422 
5423 static int handle_cr(struct kvm_vcpu *vcpu)
5424 {
5425 	unsigned long exit_qualification, val;
5426 	int cr;
5427 	int reg;
5428 	int err;
5429 	int ret;
5430 
5431 	exit_qualification = vmx_get_exit_qual(vcpu);
5432 	cr = exit_qualification & 15;
5433 	reg = (exit_qualification >> 8) & 15;
5434 	switch ((exit_qualification >> 4) & 3) {
5435 	case 0: /* mov to cr */
5436 		val = kvm_register_read(vcpu, reg);
5437 		trace_kvm_cr_write(cr, val);
5438 		switch (cr) {
5439 		case 0:
5440 			err = handle_set_cr0(vcpu, val);
5441 			return kvm_complete_insn_gp(vcpu, err);
5442 		case 3:
5443 			WARN_ON_ONCE(enable_unrestricted_guest);
5444 
5445 			err = kvm_set_cr3(vcpu, val);
5446 			return kvm_complete_insn_gp(vcpu, err);
5447 		case 4:
5448 			err = handle_set_cr4(vcpu, val);
5449 			return kvm_complete_insn_gp(vcpu, err);
5450 		case 8: {
5451 				u8 cr8_prev = kvm_get_cr8(vcpu);
5452 				u8 cr8 = (u8)val;
5453 				err = kvm_set_cr8(vcpu, cr8);
5454 				ret = kvm_complete_insn_gp(vcpu, err);
5455 				if (lapic_in_kernel(vcpu))
5456 					return ret;
5457 				if (cr8_prev <= cr8)
5458 					return ret;
5459 				/*
5460 				 * TODO: we might be squashing a
5461 				 * KVM_GUESTDBG_SINGLESTEP-triggered
5462 				 * KVM_EXIT_DEBUG here.
5463 				 */
5464 				vcpu->run->exit_reason = KVM_EXIT_SET_TPR;
5465 				return 0;
5466 			}
5467 		}
5468 		break;
5469 	case 2: /* clts */
5470 		KVM_BUG(1, vcpu->kvm, "Guest always owns CR0.TS");
5471 		return -EIO;
5472 	case 1: /*mov from cr*/
5473 		switch (cr) {
5474 		case 3:
5475 			WARN_ON_ONCE(enable_unrestricted_guest);
5476 
5477 			val = kvm_read_cr3(vcpu);
5478 			kvm_register_write(vcpu, reg, val);
5479 			trace_kvm_cr_read(cr, val);
5480 			return kvm_skip_emulated_instruction(vcpu);
5481 		case 8:
5482 			val = kvm_get_cr8(vcpu);
5483 			kvm_register_write(vcpu, reg, val);
5484 			trace_kvm_cr_read(cr, val);
5485 			return kvm_skip_emulated_instruction(vcpu);
5486 		}
5487 		break;
5488 	case 3: /* lmsw */
5489 		val = (exit_qualification >> LMSW_SOURCE_DATA_SHIFT) & 0x0f;
5490 		trace_kvm_cr_write(0, (kvm_read_cr0_bits(vcpu, ~0xful) | val));
5491 		kvm_lmsw(vcpu, val);
5492 
5493 		return kvm_skip_emulated_instruction(vcpu);
5494 	default:
5495 		break;
5496 	}
5497 	vcpu->run->exit_reason = 0;
5498 	vcpu_unimpl(vcpu, "unhandled control register: op %d cr %d\n",
5499 	       (int)(exit_qualification >> 4) & 3, cr);
5500 	return 0;
5501 }
5502 
5503 static int handle_dr(struct kvm_vcpu *vcpu)
5504 {
5505 	unsigned long exit_qualification;
5506 	int dr, dr7, reg;
5507 	int err = 1;
5508 
5509 	exit_qualification = vmx_get_exit_qual(vcpu);
5510 	dr = exit_qualification & DEBUG_REG_ACCESS_NUM;
5511 
5512 	/* First, if DR does not exist, trigger UD */
5513 	if (!kvm_require_dr(vcpu, dr))
5514 		return 1;
5515 
5516 	if (vmx_get_cpl(vcpu) > 0)
5517 		goto out;
5518 
5519 	dr7 = vmcs_readl(GUEST_DR7);
5520 	if (dr7 & DR7_GD) {
5521 		/*
5522 		 * As the vm-exit takes precedence over the debug trap, we
5523 		 * need to emulate the latter, either for the host or the
5524 		 * guest debugging itself.
5525 		 */
5526 		if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP) {
5527 			vcpu->run->debug.arch.dr6 = DR6_BD | DR6_ACTIVE_LOW;
5528 			vcpu->run->debug.arch.dr7 = dr7;
5529 			vcpu->run->debug.arch.pc = kvm_get_linear_rip(vcpu);
5530 			vcpu->run->debug.arch.exception = DB_VECTOR;
5531 			vcpu->run->exit_reason = KVM_EXIT_DEBUG;
5532 			return 0;
5533 		} else {
5534 			kvm_queue_exception_p(vcpu, DB_VECTOR, DR6_BD);
5535 			return 1;
5536 		}
5537 	}
5538 
5539 	if (vcpu->guest_debug == 0) {
5540 		exec_controls_clearbit(to_vmx(vcpu), CPU_BASED_MOV_DR_EXITING);
5541 
5542 		/*
5543 		 * No more DR vmexits; force a reload of the debug registers
5544 		 * and reenter on this instruction.  The next vmexit will
5545 		 * retrieve the full state of the debug registers.
5546 		 */
5547 		vcpu->arch.switch_db_regs |= KVM_DEBUGREG_WONT_EXIT;
5548 		return 1;
5549 	}
5550 
5551 	reg = DEBUG_REG_ACCESS_REG(exit_qualification);
5552 	if (exit_qualification & TYPE_MOV_FROM_DR) {
5553 		unsigned long val;
5554 
5555 		kvm_get_dr(vcpu, dr, &val);
5556 		kvm_register_write(vcpu, reg, val);
5557 		err = 0;
5558 	} else {
5559 		err = kvm_set_dr(vcpu, dr, kvm_register_read(vcpu, reg));
5560 	}
5561 
5562 out:
5563 	return kvm_complete_insn_gp(vcpu, err);
5564 }
5565 
5566 static void vmx_sync_dirty_debug_regs(struct kvm_vcpu *vcpu)
5567 {
5568 	get_debugreg(vcpu->arch.db[0], 0);
5569 	get_debugreg(vcpu->arch.db[1], 1);
5570 	get_debugreg(vcpu->arch.db[2], 2);
5571 	get_debugreg(vcpu->arch.db[3], 3);
5572 	get_debugreg(vcpu->arch.dr6, 6);
5573 	vcpu->arch.dr7 = vmcs_readl(GUEST_DR7);
5574 
5575 	vcpu->arch.switch_db_regs &= ~KVM_DEBUGREG_WONT_EXIT;
5576 	exec_controls_setbit(to_vmx(vcpu), CPU_BASED_MOV_DR_EXITING);
5577 
5578 	/*
5579 	 * exc_debug expects dr6 to be cleared after it runs, avoid that it sees
5580 	 * a stale dr6 from the guest.
5581 	 */
5582 	set_debugreg(DR6_RESERVED, 6);
5583 }
5584 
5585 static void vmx_set_dr7(struct kvm_vcpu *vcpu, unsigned long val)
5586 {
5587 	vmcs_writel(GUEST_DR7, val);
5588 }
5589 
5590 static int handle_tpr_below_threshold(struct kvm_vcpu *vcpu)
5591 {
5592 	kvm_apic_update_ppr(vcpu);
5593 	return 1;
5594 }
5595 
5596 static int handle_interrupt_window(struct kvm_vcpu *vcpu)
5597 {
5598 	exec_controls_clearbit(to_vmx(vcpu), CPU_BASED_INTR_WINDOW_EXITING);
5599 
5600 	kvm_make_request(KVM_REQ_EVENT, vcpu);
5601 
5602 	++vcpu->stat.irq_window_exits;
5603 	return 1;
5604 }
5605 
5606 static int handle_invlpg(struct kvm_vcpu *vcpu)
5607 {
5608 	unsigned long exit_qualification = vmx_get_exit_qual(vcpu);
5609 
5610 	kvm_mmu_invlpg(vcpu, exit_qualification);
5611 	return kvm_skip_emulated_instruction(vcpu);
5612 }
5613 
5614 static int handle_apic_access(struct kvm_vcpu *vcpu)
5615 {
5616 	if (likely(fasteoi)) {
5617 		unsigned long exit_qualification = vmx_get_exit_qual(vcpu);
5618 		int access_type, offset;
5619 
5620 		access_type = exit_qualification & APIC_ACCESS_TYPE;
5621 		offset = exit_qualification & APIC_ACCESS_OFFSET;
5622 		/*
5623 		 * Sane guest uses MOV to write EOI, with written value
5624 		 * not cared. So make a short-circuit here by avoiding
5625 		 * heavy instruction emulation.
5626 		 */
5627 		if ((access_type == TYPE_LINEAR_APIC_INST_WRITE) &&
5628 		    (offset == APIC_EOI)) {
5629 			kvm_lapic_set_eoi(vcpu);
5630 			return kvm_skip_emulated_instruction(vcpu);
5631 		}
5632 	}
5633 	return kvm_emulate_instruction(vcpu, 0);
5634 }
5635 
5636 static int handle_apic_eoi_induced(struct kvm_vcpu *vcpu)
5637 {
5638 	unsigned long exit_qualification = vmx_get_exit_qual(vcpu);
5639 	int vector = exit_qualification & 0xff;
5640 
5641 	/* EOI-induced VM exit is trap-like and thus no need to adjust IP */
5642 	kvm_apic_set_eoi_accelerated(vcpu, vector);
5643 	return 1;
5644 }
5645 
5646 static int handle_apic_write(struct kvm_vcpu *vcpu)
5647 {
5648 	unsigned long exit_qualification = vmx_get_exit_qual(vcpu);
5649 
5650 	/*
5651 	 * APIC-write VM-Exit is trap-like, KVM doesn't need to advance RIP and
5652 	 * hardware has done any necessary aliasing, offset adjustments, etc...
5653 	 * for the access.  I.e. the correct value has already been  written to
5654 	 * the vAPIC page for the correct 16-byte chunk.  KVM needs only to
5655 	 * retrieve the register value and emulate the access.
5656 	 */
5657 	u32 offset = exit_qualification & 0xff0;
5658 
5659 	kvm_apic_write_nodecode(vcpu, offset);
5660 	return 1;
5661 }
5662 
5663 static int handle_task_switch(struct kvm_vcpu *vcpu)
5664 {
5665 	struct vcpu_vmx *vmx = to_vmx(vcpu);
5666 	unsigned long exit_qualification;
5667 	bool has_error_code = false;
5668 	u32 error_code = 0;
5669 	u16 tss_selector;
5670 	int reason, type, idt_v, idt_index;
5671 
5672 	idt_v = (vmx->idt_vectoring_info & VECTORING_INFO_VALID_MASK);
5673 	idt_index = (vmx->idt_vectoring_info & VECTORING_INFO_VECTOR_MASK);
5674 	type = (vmx->idt_vectoring_info & VECTORING_INFO_TYPE_MASK);
5675 
5676 	exit_qualification = vmx_get_exit_qual(vcpu);
5677 
5678 	reason = (u32)exit_qualification >> 30;
5679 	if (reason == TASK_SWITCH_GATE && idt_v) {
5680 		switch (type) {
5681 		case INTR_TYPE_NMI_INTR:
5682 			vcpu->arch.nmi_injected = false;
5683 			vmx_set_nmi_mask(vcpu, true);
5684 			break;
5685 		case INTR_TYPE_EXT_INTR:
5686 		case INTR_TYPE_SOFT_INTR:
5687 			kvm_clear_interrupt_queue(vcpu);
5688 			break;
5689 		case INTR_TYPE_HARD_EXCEPTION:
5690 			if (vmx->idt_vectoring_info &
5691 			    VECTORING_INFO_DELIVER_CODE_MASK) {
5692 				has_error_code = true;
5693 				error_code =
5694 					vmcs_read32(IDT_VECTORING_ERROR_CODE);
5695 			}
5696 			fallthrough;
5697 		case INTR_TYPE_SOFT_EXCEPTION:
5698 			kvm_clear_exception_queue(vcpu);
5699 			break;
5700 		default:
5701 			break;
5702 		}
5703 	}
5704 	tss_selector = exit_qualification;
5705 
5706 	if (!idt_v || (type != INTR_TYPE_HARD_EXCEPTION &&
5707 		       type != INTR_TYPE_EXT_INTR &&
5708 		       type != INTR_TYPE_NMI_INTR))
5709 		WARN_ON(!skip_emulated_instruction(vcpu));
5710 
5711 	/*
5712 	 * TODO: What about debug traps on tss switch?
5713 	 *       Are we supposed to inject them and update dr6?
5714 	 */
5715 	return kvm_task_switch(vcpu, tss_selector,
5716 			       type == INTR_TYPE_SOFT_INTR ? idt_index : -1,
5717 			       reason, has_error_code, error_code);
5718 }
5719 
5720 static int handle_ept_violation(struct kvm_vcpu *vcpu)
5721 {
5722 	unsigned long exit_qualification;
5723 	gpa_t gpa;
5724 	u64 error_code;
5725 
5726 	exit_qualification = vmx_get_exit_qual(vcpu);
5727 
5728 	/*
5729 	 * EPT violation happened while executing iret from NMI,
5730 	 * "blocked by NMI" bit has to be set before next VM entry.
5731 	 * There are errata that may cause this bit to not be set:
5732 	 * AAK134, BY25.
5733 	 */
5734 	if (!(to_vmx(vcpu)->idt_vectoring_info & VECTORING_INFO_VALID_MASK) &&
5735 			enable_vnmi &&
5736 			(exit_qualification & INTR_INFO_UNBLOCK_NMI))
5737 		vmcs_set_bits(GUEST_INTERRUPTIBILITY_INFO, GUEST_INTR_STATE_NMI);
5738 
5739 	gpa = vmcs_read64(GUEST_PHYSICAL_ADDRESS);
5740 	trace_kvm_page_fault(vcpu, gpa, exit_qualification);
5741 
5742 	/* Is it a read fault? */
5743 	error_code = (exit_qualification & EPT_VIOLATION_ACC_READ)
5744 		     ? PFERR_USER_MASK : 0;
5745 	/* Is it a write fault? */
5746 	error_code |= (exit_qualification & EPT_VIOLATION_ACC_WRITE)
5747 		      ? PFERR_WRITE_MASK : 0;
5748 	/* Is it a fetch fault? */
5749 	error_code |= (exit_qualification & EPT_VIOLATION_ACC_INSTR)
5750 		      ? PFERR_FETCH_MASK : 0;
5751 	/* ept page table entry is present? */
5752 	error_code |= (exit_qualification & EPT_VIOLATION_RWX_MASK)
5753 		      ? PFERR_PRESENT_MASK : 0;
5754 
5755 	error_code |= (exit_qualification & EPT_VIOLATION_GVA_TRANSLATED) != 0 ?
5756 	       PFERR_GUEST_FINAL_MASK : PFERR_GUEST_PAGE_MASK;
5757 
5758 	vcpu->arch.exit_qualification = exit_qualification;
5759 
5760 	/*
5761 	 * Check that the GPA doesn't exceed physical memory limits, as that is
5762 	 * a guest page fault.  We have to emulate the instruction here, because
5763 	 * if the illegal address is that of a paging structure, then
5764 	 * EPT_VIOLATION_ACC_WRITE bit is set.  Alternatively, if supported we
5765 	 * would also use advanced VM-exit information for EPT violations to
5766 	 * reconstruct the page fault error code.
5767 	 */
5768 	if (unlikely(allow_smaller_maxphyaddr && kvm_vcpu_is_illegal_gpa(vcpu, gpa)))
5769 		return kvm_emulate_instruction(vcpu, 0);
5770 
5771 	return kvm_mmu_page_fault(vcpu, gpa, error_code, NULL, 0);
5772 }
5773 
5774 static int handle_ept_misconfig(struct kvm_vcpu *vcpu)
5775 {
5776 	gpa_t gpa;
5777 
5778 	if (!vmx_can_emulate_instruction(vcpu, EMULTYPE_PF, NULL, 0))
5779 		return 1;
5780 
5781 	/*
5782 	 * A nested guest cannot optimize MMIO vmexits, because we have an
5783 	 * nGPA here instead of the required GPA.
5784 	 */
5785 	gpa = vmcs_read64(GUEST_PHYSICAL_ADDRESS);
5786 	if (!is_guest_mode(vcpu) &&
5787 	    !kvm_io_bus_write(vcpu, KVM_FAST_MMIO_BUS, gpa, 0, NULL)) {
5788 		trace_kvm_fast_mmio(gpa);
5789 		return kvm_skip_emulated_instruction(vcpu);
5790 	}
5791 
5792 	return kvm_mmu_page_fault(vcpu, gpa, PFERR_RSVD_MASK, NULL, 0);
5793 }
5794 
5795 static int handle_nmi_window(struct kvm_vcpu *vcpu)
5796 {
5797 	if (KVM_BUG_ON(!enable_vnmi, vcpu->kvm))
5798 		return -EIO;
5799 
5800 	exec_controls_clearbit(to_vmx(vcpu), CPU_BASED_NMI_WINDOW_EXITING);
5801 	++vcpu->stat.nmi_window_exits;
5802 	kvm_make_request(KVM_REQ_EVENT, vcpu);
5803 
5804 	return 1;
5805 }
5806 
5807 static bool vmx_emulation_required_with_pending_exception(struct kvm_vcpu *vcpu)
5808 {
5809 	struct vcpu_vmx *vmx = to_vmx(vcpu);
5810 
5811 	return vmx->emulation_required && !vmx->rmode.vm86_active &&
5812 	       (kvm_is_exception_pending(vcpu) || vcpu->arch.exception.injected);
5813 }
5814 
5815 static int handle_invalid_guest_state(struct kvm_vcpu *vcpu)
5816 {
5817 	struct vcpu_vmx *vmx = to_vmx(vcpu);
5818 	bool intr_window_requested;
5819 	unsigned count = 130;
5820 
5821 	intr_window_requested = exec_controls_get(vmx) &
5822 				CPU_BASED_INTR_WINDOW_EXITING;
5823 
5824 	while (vmx->emulation_required && count-- != 0) {
5825 		if (intr_window_requested && !vmx_interrupt_blocked(vcpu))
5826 			return handle_interrupt_window(&vmx->vcpu);
5827 
5828 		if (kvm_test_request(KVM_REQ_EVENT, vcpu))
5829 			return 1;
5830 
5831 		if (!kvm_emulate_instruction(vcpu, 0))
5832 			return 0;
5833 
5834 		if (vmx_emulation_required_with_pending_exception(vcpu)) {
5835 			kvm_prepare_emulation_failure_exit(vcpu);
5836 			return 0;
5837 		}
5838 
5839 		if (vcpu->arch.halt_request) {
5840 			vcpu->arch.halt_request = 0;
5841 			return kvm_emulate_halt_noskip(vcpu);
5842 		}
5843 
5844 		/*
5845 		 * Note, return 1 and not 0, vcpu_run() will invoke
5846 		 * xfer_to_guest_mode() which will create a proper return
5847 		 * code.
5848 		 */
5849 		if (__xfer_to_guest_mode_work_pending())
5850 			return 1;
5851 	}
5852 
5853 	return 1;
5854 }
5855 
5856 static int vmx_vcpu_pre_run(struct kvm_vcpu *vcpu)
5857 {
5858 	if (vmx_emulation_required_with_pending_exception(vcpu)) {
5859 		kvm_prepare_emulation_failure_exit(vcpu);
5860 		return 0;
5861 	}
5862 
5863 	return 1;
5864 }
5865 
5866 static void grow_ple_window(struct kvm_vcpu *vcpu)
5867 {
5868 	struct vcpu_vmx *vmx = to_vmx(vcpu);
5869 	unsigned int old = vmx->ple_window;
5870 
5871 	vmx->ple_window = __grow_ple_window(old, ple_window,
5872 					    ple_window_grow,
5873 					    ple_window_max);
5874 
5875 	if (vmx->ple_window != old) {
5876 		vmx->ple_window_dirty = true;
5877 		trace_kvm_ple_window_update(vcpu->vcpu_id,
5878 					    vmx->ple_window, old);
5879 	}
5880 }
5881 
5882 static void shrink_ple_window(struct kvm_vcpu *vcpu)
5883 {
5884 	struct vcpu_vmx *vmx = to_vmx(vcpu);
5885 	unsigned int old = vmx->ple_window;
5886 
5887 	vmx->ple_window = __shrink_ple_window(old, ple_window,
5888 					      ple_window_shrink,
5889 					      ple_window);
5890 
5891 	if (vmx->ple_window != old) {
5892 		vmx->ple_window_dirty = true;
5893 		trace_kvm_ple_window_update(vcpu->vcpu_id,
5894 					    vmx->ple_window, old);
5895 	}
5896 }
5897 
5898 /*
5899  * Indicate a busy-waiting vcpu in spinlock. We do not enable the PAUSE
5900  * exiting, so only get here on cpu with PAUSE-Loop-Exiting.
5901  */
5902 static int handle_pause(struct kvm_vcpu *vcpu)
5903 {
5904 	if (!kvm_pause_in_guest(vcpu->kvm))
5905 		grow_ple_window(vcpu);
5906 
5907 	/*
5908 	 * Intel sdm vol3 ch-25.1.3 says: The "PAUSE-loop exiting"
5909 	 * VM-execution control is ignored if CPL > 0. OTOH, KVM
5910 	 * never set PAUSE_EXITING and just set PLE if supported,
5911 	 * so the vcpu must be CPL=0 if it gets a PAUSE exit.
5912 	 */
5913 	kvm_vcpu_on_spin(vcpu, true);
5914 	return kvm_skip_emulated_instruction(vcpu);
5915 }
5916 
5917 static int handle_monitor_trap(struct kvm_vcpu *vcpu)
5918 {
5919 	return 1;
5920 }
5921 
5922 static int handle_invpcid(struct kvm_vcpu *vcpu)
5923 {
5924 	u32 vmx_instruction_info;
5925 	unsigned long type;
5926 	gva_t gva;
5927 	struct {
5928 		u64 pcid;
5929 		u64 gla;
5930 	} operand;
5931 	int gpr_index;
5932 
5933 	if (!guest_cpuid_has(vcpu, X86_FEATURE_INVPCID)) {
5934 		kvm_queue_exception(vcpu, UD_VECTOR);
5935 		return 1;
5936 	}
5937 
5938 	vmx_instruction_info = vmcs_read32(VMX_INSTRUCTION_INFO);
5939 	gpr_index = vmx_get_instr_info_reg2(vmx_instruction_info);
5940 	type = kvm_register_read(vcpu, gpr_index);
5941 
5942 	/* According to the Intel instruction reference, the memory operand
5943 	 * is read even if it isn't needed (e.g., for type==all)
5944 	 */
5945 	if (get_vmx_mem_address(vcpu, vmx_get_exit_qual(vcpu),
5946 				vmx_instruction_info, false,
5947 				sizeof(operand), &gva))
5948 		return 1;
5949 
5950 	return kvm_handle_invpcid(vcpu, type, gva);
5951 }
5952 
5953 static int handle_pml_full(struct kvm_vcpu *vcpu)
5954 {
5955 	unsigned long exit_qualification;
5956 
5957 	trace_kvm_pml_full(vcpu->vcpu_id);
5958 
5959 	exit_qualification = vmx_get_exit_qual(vcpu);
5960 
5961 	/*
5962 	 * PML buffer FULL happened while executing iret from NMI,
5963 	 * "blocked by NMI" bit has to be set before next VM entry.
5964 	 */
5965 	if (!(to_vmx(vcpu)->idt_vectoring_info & VECTORING_INFO_VALID_MASK) &&
5966 			enable_vnmi &&
5967 			(exit_qualification & INTR_INFO_UNBLOCK_NMI))
5968 		vmcs_set_bits(GUEST_INTERRUPTIBILITY_INFO,
5969 				GUEST_INTR_STATE_NMI);
5970 
5971 	/*
5972 	 * PML buffer already flushed at beginning of VMEXIT. Nothing to do
5973 	 * here.., and there's no userspace involvement needed for PML.
5974 	 */
5975 	return 1;
5976 }
5977 
5978 static fastpath_t handle_fastpath_preemption_timer(struct kvm_vcpu *vcpu)
5979 {
5980 	struct vcpu_vmx *vmx = to_vmx(vcpu);
5981 
5982 	if (!vmx->req_immediate_exit &&
5983 	    !unlikely(vmx->loaded_vmcs->hv_timer_soft_disabled)) {
5984 		kvm_lapic_expired_hv_timer(vcpu);
5985 		return EXIT_FASTPATH_REENTER_GUEST;
5986 	}
5987 
5988 	return EXIT_FASTPATH_NONE;
5989 }
5990 
5991 static int handle_preemption_timer(struct kvm_vcpu *vcpu)
5992 {
5993 	handle_fastpath_preemption_timer(vcpu);
5994 	return 1;
5995 }
5996 
5997 /*
5998  * When nested=0, all VMX instruction VM Exits filter here.  The handlers
5999  * are overwritten by nested_vmx_setup() when nested=1.
6000  */
6001 static int handle_vmx_instruction(struct kvm_vcpu *vcpu)
6002 {
6003 	kvm_queue_exception(vcpu, UD_VECTOR);
6004 	return 1;
6005 }
6006 
6007 #ifndef CONFIG_X86_SGX_KVM
6008 static int handle_encls(struct kvm_vcpu *vcpu)
6009 {
6010 	/*
6011 	 * SGX virtualization is disabled.  There is no software enable bit for
6012 	 * SGX, so KVM intercepts all ENCLS leafs and injects a #UD to prevent
6013 	 * the guest from executing ENCLS (when SGX is supported by hardware).
6014 	 */
6015 	kvm_queue_exception(vcpu, UD_VECTOR);
6016 	return 1;
6017 }
6018 #endif /* CONFIG_X86_SGX_KVM */
6019 
6020 static int handle_bus_lock_vmexit(struct kvm_vcpu *vcpu)
6021 {
6022 	/*
6023 	 * Hardware may or may not set the BUS_LOCK_DETECTED flag on BUS_LOCK
6024 	 * VM-Exits. Unconditionally set the flag here and leave the handling to
6025 	 * vmx_handle_exit().
6026 	 */
6027 	to_vmx(vcpu)->exit_reason.bus_lock_detected = true;
6028 	return 1;
6029 }
6030 
6031 static int handle_notify(struct kvm_vcpu *vcpu)
6032 {
6033 	unsigned long exit_qual = vmx_get_exit_qual(vcpu);
6034 	bool context_invalid = exit_qual & NOTIFY_VM_CONTEXT_INVALID;
6035 
6036 	++vcpu->stat.notify_window_exits;
6037 
6038 	/*
6039 	 * Notify VM exit happened while executing iret from NMI,
6040 	 * "blocked by NMI" bit has to be set before next VM entry.
6041 	 */
6042 	if (enable_vnmi && (exit_qual & INTR_INFO_UNBLOCK_NMI))
6043 		vmcs_set_bits(GUEST_INTERRUPTIBILITY_INFO,
6044 			      GUEST_INTR_STATE_NMI);
6045 
6046 	if (vcpu->kvm->arch.notify_vmexit_flags & KVM_X86_NOTIFY_VMEXIT_USER ||
6047 	    context_invalid) {
6048 		vcpu->run->exit_reason = KVM_EXIT_NOTIFY;
6049 		vcpu->run->notify.flags = context_invalid ?
6050 					  KVM_NOTIFY_CONTEXT_INVALID : 0;
6051 		return 0;
6052 	}
6053 
6054 	return 1;
6055 }
6056 
6057 /*
6058  * The exit handlers return 1 if the exit was handled fully and guest execution
6059  * may resume.  Otherwise they set the kvm_run parameter to indicate what needs
6060  * to be done to userspace and return 0.
6061  */
6062 static int (*kvm_vmx_exit_handlers[])(struct kvm_vcpu *vcpu) = {
6063 	[EXIT_REASON_EXCEPTION_NMI]           = handle_exception_nmi,
6064 	[EXIT_REASON_EXTERNAL_INTERRUPT]      = handle_external_interrupt,
6065 	[EXIT_REASON_TRIPLE_FAULT]            = handle_triple_fault,
6066 	[EXIT_REASON_NMI_WINDOW]	      = handle_nmi_window,
6067 	[EXIT_REASON_IO_INSTRUCTION]          = handle_io,
6068 	[EXIT_REASON_CR_ACCESS]               = handle_cr,
6069 	[EXIT_REASON_DR_ACCESS]               = handle_dr,
6070 	[EXIT_REASON_CPUID]                   = kvm_emulate_cpuid,
6071 	[EXIT_REASON_MSR_READ]                = kvm_emulate_rdmsr,
6072 	[EXIT_REASON_MSR_WRITE]               = kvm_emulate_wrmsr,
6073 	[EXIT_REASON_INTERRUPT_WINDOW]        = handle_interrupt_window,
6074 	[EXIT_REASON_HLT]                     = kvm_emulate_halt,
6075 	[EXIT_REASON_INVD]		      = kvm_emulate_invd,
6076 	[EXIT_REASON_INVLPG]		      = handle_invlpg,
6077 	[EXIT_REASON_RDPMC]                   = kvm_emulate_rdpmc,
6078 	[EXIT_REASON_VMCALL]                  = kvm_emulate_hypercall,
6079 	[EXIT_REASON_VMCLEAR]		      = handle_vmx_instruction,
6080 	[EXIT_REASON_VMLAUNCH]		      = handle_vmx_instruction,
6081 	[EXIT_REASON_VMPTRLD]		      = handle_vmx_instruction,
6082 	[EXIT_REASON_VMPTRST]		      = handle_vmx_instruction,
6083 	[EXIT_REASON_VMREAD]		      = handle_vmx_instruction,
6084 	[EXIT_REASON_VMRESUME]		      = handle_vmx_instruction,
6085 	[EXIT_REASON_VMWRITE]		      = handle_vmx_instruction,
6086 	[EXIT_REASON_VMOFF]		      = handle_vmx_instruction,
6087 	[EXIT_REASON_VMON]		      = handle_vmx_instruction,
6088 	[EXIT_REASON_TPR_BELOW_THRESHOLD]     = handle_tpr_below_threshold,
6089 	[EXIT_REASON_APIC_ACCESS]             = handle_apic_access,
6090 	[EXIT_REASON_APIC_WRITE]              = handle_apic_write,
6091 	[EXIT_REASON_EOI_INDUCED]             = handle_apic_eoi_induced,
6092 	[EXIT_REASON_WBINVD]                  = kvm_emulate_wbinvd,
6093 	[EXIT_REASON_XSETBV]                  = kvm_emulate_xsetbv,
6094 	[EXIT_REASON_TASK_SWITCH]             = handle_task_switch,
6095 	[EXIT_REASON_MCE_DURING_VMENTRY]      = handle_machine_check,
6096 	[EXIT_REASON_GDTR_IDTR]		      = handle_desc,
6097 	[EXIT_REASON_LDTR_TR]		      = handle_desc,
6098 	[EXIT_REASON_EPT_VIOLATION]	      = handle_ept_violation,
6099 	[EXIT_REASON_EPT_MISCONFIG]           = handle_ept_misconfig,
6100 	[EXIT_REASON_PAUSE_INSTRUCTION]       = handle_pause,
6101 	[EXIT_REASON_MWAIT_INSTRUCTION]	      = kvm_emulate_mwait,
6102 	[EXIT_REASON_MONITOR_TRAP_FLAG]       = handle_monitor_trap,
6103 	[EXIT_REASON_MONITOR_INSTRUCTION]     = kvm_emulate_monitor,
6104 	[EXIT_REASON_INVEPT]                  = handle_vmx_instruction,
6105 	[EXIT_REASON_INVVPID]                 = handle_vmx_instruction,
6106 	[EXIT_REASON_RDRAND]                  = kvm_handle_invalid_op,
6107 	[EXIT_REASON_RDSEED]                  = kvm_handle_invalid_op,
6108 	[EXIT_REASON_PML_FULL]		      = handle_pml_full,
6109 	[EXIT_REASON_INVPCID]                 = handle_invpcid,
6110 	[EXIT_REASON_VMFUNC]		      = handle_vmx_instruction,
6111 	[EXIT_REASON_PREEMPTION_TIMER]	      = handle_preemption_timer,
6112 	[EXIT_REASON_ENCLS]		      = handle_encls,
6113 	[EXIT_REASON_BUS_LOCK]                = handle_bus_lock_vmexit,
6114 	[EXIT_REASON_NOTIFY]		      = handle_notify,
6115 };
6116 
6117 static const int kvm_vmx_max_exit_handlers =
6118 	ARRAY_SIZE(kvm_vmx_exit_handlers);
6119 
6120 static void vmx_get_exit_info(struct kvm_vcpu *vcpu, u32 *reason,
6121 			      u64 *info1, u64 *info2,
6122 			      u32 *intr_info, u32 *error_code)
6123 {
6124 	struct vcpu_vmx *vmx = to_vmx(vcpu);
6125 
6126 	*reason = vmx->exit_reason.full;
6127 	*info1 = vmx_get_exit_qual(vcpu);
6128 	if (!(vmx->exit_reason.failed_vmentry)) {
6129 		*info2 = vmx->idt_vectoring_info;
6130 		*intr_info = vmx_get_intr_info(vcpu);
6131 		if (is_exception_with_error_code(*intr_info))
6132 			*error_code = vmcs_read32(VM_EXIT_INTR_ERROR_CODE);
6133 		else
6134 			*error_code = 0;
6135 	} else {
6136 		*info2 = 0;
6137 		*intr_info = 0;
6138 		*error_code = 0;
6139 	}
6140 }
6141 
6142 static void vmx_destroy_pml_buffer(struct vcpu_vmx *vmx)
6143 {
6144 	if (vmx->pml_pg) {
6145 		__free_page(vmx->pml_pg);
6146 		vmx->pml_pg = NULL;
6147 	}
6148 }
6149 
6150 static void vmx_flush_pml_buffer(struct kvm_vcpu *vcpu)
6151 {
6152 	struct vcpu_vmx *vmx = to_vmx(vcpu);
6153 	u64 *pml_buf;
6154 	u16 pml_idx;
6155 
6156 	pml_idx = vmcs_read16(GUEST_PML_INDEX);
6157 
6158 	/* Do nothing if PML buffer is empty */
6159 	if (pml_idx == (PML_ENTITY_NUM - 1))
6160 		return;
6161 
6162 	/* PML index always points to next available PML buffer entity */
6163 	if (pml_idx >= PML_ENTITY_NUM)
6164 		pml_idx = 0;
6165 	else
6166 		pml_idx++;
6167 
6168 	pml_buf = page_address(vmx->pml_pg);
6169 	for (; pml_idx < PML_ENTITY_NUM; pml_idx++) {
6170 		u64 gpa;
6171 
6172 		gpa = pml_buf[pml_idx];
6173 		WARN_ON(gpa & (PAGE_SIZE - 1));
6174 		kvm_vcpu_mark_page_dirty(vcpu, gpa >> PAGE_SHIFT);
6175 	}
6176 
6177 	/* reset PML index */
6178 	vmcs_write16(GUEST_PML_INDEX, PML_ENTITY_NUM - 1);
6179 }
6180 
6181 static void vmx_dump_sel(char *name, uint32_t sel)
6182 {
6183 	pr_err("%s sel=0x%04x, attr=0x%05x, limit=0x%08x, base=0x%016lx\n",
6184 	       name, vmcs_read16(sel),
6185 	       vmcs_read32(sel + GUEST_ES_AR_BYTES - GUEST_ES_SELECTOR),
6186 	       vmcs_read32(sel + GUEST_ES_LIMIT - GUEST_ES_SELECTOR),
6187 	       vmcs_readl(sel + GUEST_ES_BASE - GUEST_ES_SELECTOR));
6188 }
6189 
6190 static void vmx_dump_dtsel(char *name, uint32_t limit)
6191 {
6192 	pr_err("%s                           limit=0x%08x, base=0x%016lx\n",
6193 	       name, vmcs_read32(limit),
6194 	       vmcs_readl(limit + GUEST_GDTR_BASE - GUEST_GDTR_LIMIT));
6195 }
6196 
6197 static void vmx_dump_msrs(char *name, struct vmx_msrs *m)
6198 {
6199 	unsigned int i;
6200 	struct vmx_msr_entry *e;
6201 
6202 	pr_err("MSR %s:\n", name);
6203 	for (i = 0, e = m->val; i < m->nr; ++i, ++e)
6204 		pr_err("  %2d: msr=0x%08x value=0x%016llx\n", i, e->index, e->value);
6205 }
6206 
6207 void dump_vmcs(struct kvm_vcpu *vcpu)
6208 {
6209 	struct vcpu_vmx *vmx = to_vmx(vcpu);
6210 	u32 vmentry_ctl, vmexit_ctl;
6211 	u32 cpu_based_exec_ctrl, pin_based_exec_ctrl, secondary_exec_control;
6212 	u64 tertiary_exec_control;
6213 	unsigned long cr4;
6214 	int efer_slot;
6215 
6216 	if (!dump_invalid_vmcs) {
6217 		pr_warn_ratelimited("set kvm_intel.dump_invalid_vmcs=1 to dump internal KVM state.\n");
6218 		return;
6219 	}
6220 
6221 	vmentry_ctl = vmcs_read32(VM_ENTRY_CONTROLS);
6222 	vmexit_ctl = vmcs_read32(VM_EXIT_CONTROLS);
6223 	cpu_based_exec_ctrl = vmcs_read32(CPU_BASED_VM_EXEC_CONTROL);
6224 	pin_based_exec_ctrl = vmcs_read32(PIN_BASED_VM_EXEC_CONTROL);
6225 	cr4 = vmcs_readl(GUEST_CR4);
6226 
6227 	if (cpu_has_secondary_exec_ctrls())
6228 		secondary_exec_control = vmcs_read32(SECONDARY_VM_EXEC_CONTROL);
6229 	else
6230 		secondary_exec_control = 0;
6231 
6232 	if (cpu_has_tertiary_exec_ctrls())
6233 		tertiary_exec_control = vmcs_read64(TERTIARY_VM_EXEC_CONTROL);
6234 	else
6235 		tertiary_exec_control = 0;
6236 
6237 	pr_err("VMCS %p, last attempted VM-entry on CPU %d\n",
6238 	       vmx->loaded_vmcs->vmcs, vcpu->arch.last_vmentry_cpu);
6239 	pr_err("*** Guest State ***\n");
6240 	pr_err("CR0: actual=0x%016lx, shadow=0x%016lx, gh_mask=%016lx\n",
6241 	       vmcs_readl(GUEST_CR0), vmcs_readl(CR0_READ_SHADOW),
6242 	       vmcs_readl(CR0_GUEST_HOST_MASK));
6243 	pr_err("CR4: actual=0x%016lx, shadow=0x%016lx, gh_mask=%016lx\n",
6244 	       cr4, vmcs_readl(CR4_READ_SHADOW), vmcs_readl(CR4_GUEST_HOST_MASK));
6245 	pr_err("CR3 = 0x%016lx\n", vmcs_readl(GUEST_CR3));
6246 	if (cpu_has_vmx_ept()) {
6247 		pr_err("PDPTR0 = 0x%016llx  PDPTR1 = 0x%016llx\n",
6248 		       vmcs_read64(GUEST_PDPTR0), vmcs_read64(GUEST_PDPTR1));
6249 		pr_err("PDPTR2 = 0x%016llx  PDPTR3 = 0x%016llx\n",
6250 		       vmcs_read64(GUEST_PDPTR2), vmcs_read64(GUEST_PDPTR3));
6251 	}
6252 	pr_err("RSP = 0x%016lx  RIP = 0x%016lx\n",
6253 	       vmcs_readl(GUEST_RSP), vmcs_readl(GUEST_RIP));
6254 	pr_err("RFLAGS=0x%08lx         DR7 = 0x%016lx\n",
6255 	       vmcs_readl(GUEST_RFLAGS), vmcs_readl(GUEST_DR7));
6256 	pr_err("Sysenter RSP=%016lx CS:RIP=%04x:%016lx\n",
6257 	       vmcs_readl(GUEST_SYSENTER_ESP),
6258 	       vmcs_read32(GUEST_SYSENTER_CS), vmcs_readl(GUEST_SYSENTER_EIP));
6259 	vmx_dump_sel("CS:  ", GUEST_CS_SELECTOR);
6260 	vmx_dump_sel("DS:  ", GUEST_DS_SELECTOR);
6261 	vmx_dump_sel("SS:  ", GUEST_SS_SELECTOR);
6262 	vmx_dump_sel("ES:  ", GUEST_ES_SELECTOR);
6263 	vmx_dump_sel("FS:  ", GUEST_FS_SELECTOR);
6264 	vmx_dump_sel("GS:  ", GUEST_GS_SELECTOR);
6265 	vmx_dump_dtsel("GDTR:", GUEST_GDTR_LIMIT);
6266 	vmx_dump_sel("LDTR:", GUEST_LDTR_SELECTOR);
6267 	vmx_dump_dtsel("IDTR:", GUEST_IDTR_LIMIT);
6268 	vmx_dump_sel("TR:  ", GUEST_TR_SELECTOR);
6269 	efer_slot = vmx_find_loadstore_msr_slot(&vmx->msr_autoload.guest, MSR_EFER);
6270 	if (vmentry_ctl & VM_ENTRY_LOAD_IA32_EFER)
6271 		pr_err("EFER= 0x%016llx\n", vmcs_read64(GUEST_IA32_EFER));
6272 	else if (efer_slot >= 0)
6273 		pr_err("EFER= 0x%016llx (autoload)\n",
6274 		       vmx->msr_autoload.guest.val[efer_slot].value);
6275 	else if (vmentry_ctl & VM_ENTRY_IA32E_MODE)
6276 		pr_err("EFER= 0x%016llx (effective)\n",
6277 		       vcpu->arch.efer | (EFER_LMA | EFER_LME));
6278 	else
6279 		pr_err("EFER= 0x%016llx (effective)\n",
6280 		       vcpu->arch.efer & ~(EFER_LMA | EFER_LME));
6281 	if (vmentry_ctl & VM_ENTRY_LOAD_IA32_PAT)
6282 		pr_err("PAT = 0x%016llx\n", vmcs_read64(GUEST_IA32_PAT));
6283 	pr_err("DebugCtl = 0x%016llx  DebugExceptions = 0x%016lx\n",
6284 	       vmcs_read64(GUEST_IA32_DEBUGCTL),
6285 	       vmcs_readl(GUEST_PENDING_DBG_EXCEPTIONS));
6286 	if (cpu_has_load_perf_global_ctrl() &&
6287 	    vmentry_ctl & VM_ENTRY_LOAD_IA32_PERF_GLOBAL_CTRL)
6288 		pr_err("PerfGlobCtl = 0x%016llx\n",
6289 		       vmcs_read64(GUEST_IA32_PERF_GLOBAL_CTRL));
6290 	if (vmentry_ctl & VM_ENTRY_LOAD_BNDCFGS)
6291 		pr_err("BndCfgS = 0x%016llx\n", vmcs_read64(GUEST_BNDCFGS));
6292 	pr_err("Interruptibility = %08x  ActivityState = %08x\n",
6293 	       vmcs_read32(GUEST_INTERRUPTIBILITY_INFO),
6294 	       vmcs_read32(GUEST_ACTIVITY_STATE));
6295 	if (secondary_exec_control & SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY)
6296 		pr_err("InterruptStatus = %04x\n",
6297 		       vmcs_read16(GUEST_INTR_STATUS));
6298 	if (vmcs_read32(VM_ENTRY_MSR_LOAD_COUNT) > 0)
6299 		vmx_dump_msrs("guest autoload", &vmx->msr_autoload.guest);
6300 	if (vmcs_read32(VM_EXIT_MSR_STORE_COUNT) > 0)
6301 		vmx_dump_msrs("guest autostore", &vmx->msr_autostore.guest);
6302 
6303 	pr_err("*** Host State ***\n");
6304 	pr_err("RIP = 0x%016lx  RSP = 0x%016lx\n",
6305 	       vmcs_readl(HOST_RIP), vmcs_readl(HOST_RSP));
6306 	pr_err("CS=%04x SS=%04x DS=%04x ES=%04x FS=%04x GS=%04x TR=%04x\n",
6307 	       vmcs_read16(HOST_CS_SELECTOR), vmcs_read16(HOST_SS_SELECTOR),
6308 	       vmcs_read16(HOST_DS_SELECTOR), vmcs_read16(HOST_ES_SELECTOR),
6309 	       vmcs_read16(HOST_FS_SELECTOR), vmcs_read16(HOST_GS_SELECTOR),
6310 	       vmcs_read16(HOST_TR_SELECTOR));
6311 	pr_err("FSBase=%016lx GSBase=%016lx TRBase=%016lx\n",
6312 	       vmcs_readl(HOST_FS_BASE), vmcs_readl(HOST_GS_BASE),
6313 	       vmcs_readl(HOST_TR_BASE));
6314 	pr_err("GDTBase=%016lx IDTBase=%016lx\n",
6315 	       vmcs_readl(HOST_GDTR_BASE), vmcs_readl(HOST_IDTR_BASE));
6316 	pr_err("CR0=%016lx CR3=%016lx CR4=%016lx\n",
6317 	       vmcs_readl(HOST_CR0), vmcs_readl(HOST_CR3),
6318 	       vmcs_readl(HOST_CR4));
6319 	pr_err("Sysenter RSP=%016lx CS:RIP=%04x:%016lx\n",
6320 	       vmcs_readl(HOST_IA32_SYSENTER_ESP),
6321 	       vmcs_read32(HOST_IA32_SYSENTER_CS),
6322 	       vmcs_readl(HOST_IA32_SYSENTER_EIP));
6323 	if (vmexit_ctl & VM_EXIT_LOAD_IA32_EFER)
6324 		pr_err("EFER= 0x%016llx\n", vmcs_read64(HOST_IA32_EFER));
6325 	if (vmexit_ctl & VM_EXIT_LOAD_IA32_PAT)
6326 		pr_err("PAT = 0x%016llx\n", vmcs_read64(HOST_IA32_PAT));
6327 	if (cpu_has_load_perf_global_ctrl() &&
6328 	    vmexit_ctl & VM_EXIT_LOAD_IA32_PERF_GLOBAL_CTRL)
6329 		pr_err("PerfGlobCtl = 0x%016llx\n",
6330 		       vmcs_read64(HOST_IA32_PERF_GLOBAL_CTRL));
6331 	if (vmcs_read32(VM_EXIT_MSR_LOAD_COUNT) > 0)
6332 		vmx_dump_msrs("host autoload", &vmx->msr_autoload.host);
6333 
6334 	pr_err("*** Control State ***\n");
6335 	pr_err("CPUBased=0x%08x SecondaryExec=0x%08x TertiaryExec=0x%016llx\n",
6336 	       cpu_based_exec_ctrl, secondary_exec_control, tertiary_exec_control);
6337 	pr_err("PinBased=0x%08x EntryControls=%08x ExitControls=%08x\n",
6338 	       pin_based_exec_ctrl, vmentry_ctl, vmexit_ctl);
6339 	pr_err("ExceptionBitmap=%08x PFECmask=%08x PFECmatch=%08x\n",
6340 	       vmcs_read32(EXCEPTION_BITMAP),
6341 	       vmcs_read32(PAGE_FAULT_ERROR_CODE_MASK),
6342 	       vmcs_read32(PAGE_FAULT_ERROR_CODE_MATCH));
6343 	pr_err("VMEntry: intr_info=%08x errcode=%08x ilen=%08x\n",
6344 	       vmcs_read32(VM_ENTRY_INTR_INFO_FIELD),
6345 	       vmcs_read32(VM_ENTRY_EXCEPTION_ERROR_CODE),
6346 	       vmcs_read32(VM_ENTRY_INSTRUCTION_LEN));
6347 	pr_err("VMExit: intr_info=%08x errcode=%08x ilen=%08x\n",
6348 	       vmcs_read32(VM_EXIT_INTR_INFO),
6349 	       vmcs_read32(VM_EXIT_INTR_ERROR_CODE),
6350 	       vmcs_read32(VM_EXIT_INSTRUCTION_LEN));
6351 	pr_err("        reason=%08x qualification=%016lx\n",
6352 	       vmcs_read32(VM_EXIT_REASON), vmcs_readl(EXIT_QUALIFICATION));
6353 	pr_err("IDTVectoring: info=%08x errcode=%08x\n",
6354 	       vmcs_read32(IDT_VECTORING_INFO_FIELD),
6355 	       vmcs_read32(IDT_VECTORING_ERROR_CODE));
6356 	pr_err("TSC Offset = 0x%016llx\n", vmcs_read64(TSC_OFFSET));
6357 	if (secondary_exec_control & SECONDARY_EXEC_TSC_SCALING)
6358 		pr_err("TSC Multiplier = 0x%016llx\n",
6359 		       vmcs_read64(TSC_MULTIPLIER));
6360 	if (cpu_based_exec_ctrl & CPU_BASED_TPR_SHADOW) {
6361 		if (secondary_exec_control & SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY) {
6362 			u16 status = vmcs_read16(GUEST_INTR_STATUS);
6363 			pr_err("SVI|RVI = %02x|%02x ", status >> 8, status & 0xff);
6364 		}
6365 		pr_cont("TPR Threshold = 0x%02x\n", vmcs_read32(TPR_THRESHOLD));
6366 		if (secondary_exec_control & SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES)
6367 			pr_err("APIC-access addr = 0x%016llx ", vmcs_read64(APIC_ACCESS_ADDR));
6368 		pr_cont("virt-APIC addr = 0x%016llx\n", vmcs_read64(VIRTUAL_APIC_PAGE_ADDR));
6369 	}
6370 	if (pin_based_exec_ctrl & PIN_BASED_POSTED_INTR)
6371 		pr_err("PostedIntrVec = 0x%02x\n", vmcs_read16(POSTED_INTR_NV));
6372 	if ((secondary_exec_control & SECONDARY_EXEC_ENABLE_EPT))
6373 		pr_err("EPT pointer = 0x%016llx\n", vmcs_read64(EPT_POINTER));
6374 	if (secondary_exec_control & SECONDARY_EXEC_PAUSE_LOOP_EXITING)
6375 		pr_err("PLE Gap=%08x Window=%08x\n",
6376 		       vmcs_read32(PLE_GAP), vmcs_read32(PLE_WINDOW));
6377 	if (secondary_exec_control & SECONDARY_EXEC_ENABLE_VPID)
6378 		pr_err("Virtual processor ID = 0x%04x\n",
6379 		       vmcs_read16(VIRTUAL_PROCESSOR_ID));
6380 }
6381 
6382 /*
6383  * The guest has exited.  See if we can fix it or if we need userspace
6384  * assistance.
6385  */
6386 static int __vmx_handle_exit(struct kvm_vcpu *vcpu, fastpath_t exit_fastpath)
6387 {
6388 	struct vcpu_vmx *vmx = to_vmx(vcpu);
6389 	union vmx_exit_reason exit_reason = vmx->exit_reason;
6390 	u32 vectoring_info = vmx->idt_vectoring_info;
6391 	u16 exit_handler_index;
6392 
6393 	/*
6394 	 * Flush logged GPAs PML buffer, this will make dirty_bitmap more
6395 	 * updated. Another good is, in kvm_vm_ioctl_get_dirty_log, before
6396 	 * querying dirty_bitmap, we only need to kick all vcpus out of guest
6397 	 * mode as if vcpus is in root mode, the PML buffer must has been
6398 	 * flushed already.  Note, PML is never enabled in hardware while
6399 	 * running L2.
6400 	 */
6401 	if (enable_pml && !is_guest_mode(vcpu))
6402 		vmx_flush_pml_buffer(vcpu);
6403 
6404 	/*
6405 	 * KVM should never reach this point with a pending nested VM-Enter.
6406 	 * More specifically, short-circuiting VM-Entry to emulate L2 due to
6407 	 * invalid guest state should never happen as that means KVM knowingly
6408 	 * allowed a nested VM-Enter with an invalid vmcs12.  More below.
6409 	 */
6410 	if (KVM_BUG_ON(vmx->nested.nested_run_pending, vcpu->kvm))
6411 		return -EIO;
6412 
6413 	if (is_guest_mode(vcpu)) {
6414 		/*
6415 		 * PML is never enabled when running L2, bail immediately if a
6416 		 * PML full exit occurs as something is horribly wrong.
6417 		 */
6418 		if (exit_reason.basic == EXIT_REASON_PML_FULL)
6419 			goto unexpected_vmexit;
6420 
6421 		/*
6422 		 * The host physical addresses of some pages of guest memory
6423 		 * are loaded into the vmcs02 (e.g. vmcs12's Virtual APIC
6424 		 * Page). The CPU may write to these pages via their host
6425 		 * physical address while L2 is running, bypassing any
6426 		 * address-translation-based dirty tracking (e.g. EPT write
6427 		 * protection).
6428 		 *
6429 		 * Mark them dirty on every exit from L2 to prevent them from
6430 		 * getting out of sync with dirty tracking.
6431 		 */
6432 		nested_mark_vmcs12_pages_dirty(vcpu);
6433 
6434 		/*
6435 		 * Synthesize a triple fault if L2 state is invalid.  In normal
6436 		 * operation, nested VM-Enter rejects any attempt to enter L2
6437 		 * with invalid state.  However, those checks are skipped if
6438 		 * state is being stuffed via RSM or KVM_SET_NESTED_STATE.  If
6439 		 * L2 state is invalid, it means either L1 modified SMRAM state
6440 		 * or userspace provided bad state.  Synthesize TRIPLE_FAULT as
6441 		 * doing so is architecturally allowed in the RSM case, and is
6442 		 * the least awful solution for the userspace case without
6443 		 * risking false positives.
6444 		 */
6445 		if (vmx->emulation_required) {
6446 			nested_vmx_vmexit(vcpu, EXIT_REASON_TRIPLE_FAULT, 0, 0);
6447 			return 1;
6448 		}
6449 
6450 		if (nested_vmx_reflect_vmexit(vcpu))
6451 			return 1;
6452 	}
6453 
6454 	/* If guest state is invalid, start emulating.  L2 is handled above. */
6455 	if (vmx->emulation_required)
6456 		return handle_invalid_guest_state(vcpu);
6457 
6458 	if (exit_reason.failed_vmentry) {
6459 		dump_vmcs(vcpu);
6460 		vcpu->run->exit_reason = KVM_EXIT_FAIL_ENTRY;
6461 		vcpu->run->fail_entry.hardware_entry_failure_reason
6462 			= exit_reason.full;
6463 		vcpu->run->fail_entry.cpu = vcpu->arch.last_vmentry_cpu;
6464 		return 0;
6465 	}
6466 
6467 	if (unlikely(vmx->fail)) {
6468 		dump_vmcs(vcpu);
6469 		vcpu->run->exit_reason = KVM_EXIT_FAIL_ENTRY;
6470 		vcpu->run->fail_entry.hardware_entry_failure_reason
6471 			= vmcs_read32(VM_INSTRUCTION_ERROR);
6472 		vcpu->run->fail_entry.cpu = vcpu->arch.last_vmentry_cpu;
6473 		return 0;
6474 	}
6475 
6476 	/*
6477 	 * Note:
6478 	 * Do not try to fix EXIT_REASON_EPT_MISCONFIG if it caused by
6479 	 * delivery event since it indicates guest is accessing MMIO.
6480 	 * The vm-exit can be triggered again after return to guest that
6481 	 * will cause infinite loop.
6482 	 */
6483 	if ((vectoring_info & VECTORING_INFO_VALID_MASK) &&
6484 	    (exit_reason.basic != EXIT_REASON_EXCEPTION_NMI &&
6485 	     exit_reason.basic != EXIT_REASON_EPT_VIOLATION &&
6486 	     exit_reason.basic != EXIT_REASON_PML_FULL &&
6487 	     exit_reason.basic != EXIT_REASON_APIC_ACCESS &&
6488 	     exit_reason.basic != EXIT_REASON_TASK_SWITCH &&
6489 	     exit_reason.basic != EXIT_REASON_NOTIFY)) {
6490 		int ndata = 3;
6491 
6492 		vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
6493 		vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_DELIVERY_EV;
6494 		vcpu->run->internal.data[0] = vectoring_info;
6495 		vcpu->run->internal.data[1] = exit_reason.full;
6496 		vcpu->run->internal.data[2] = vcpu->arch.exit_qualification;
6497 		if (exit_reason.basic == EXIT_REASON_EPT_MISCONFIG) {
6498 			vcpu->run->internal.data[ndata++] =
6499 				vmcs_read64(GUEST_PHYSICAL_ADDRESS);
6500 		}
6501 		vcpu->run->internal.data[ndata++] = vcpu->arch.last_vmentry_cpu;
6502 		vcpu->run->internal.ndata = ndata;
6503 		return 0;
6504 	}
6505 
6506 	if (unlikely(!enable_vnmi &&
6507 		     vmx->loaded_vmcs->soft_vnmi_blocked)) {
6508 		if (!vmx_interrupt_blocked(vcpu)) {
6509 			vmx->loaded_vmcs->soft_vnmi_blocked = 0;
6510 		} else if (vmx->loaded_vmcs->vnmi_blocked_time > 1000000000LL &&
6511 			   vcpu->arch.nmi_pending) {
6512 			/*
6513 			 * This CPU don't support us in finding the end of an
6514 			 * NMI-blocked window if the guest runs with IRQs
6515 			 * disabled. So we pull the trigger after 1 s of
6516 			 * futile waiting, but inform the user about this.
6517 			 */
6518 			printk(KERN_WARNING "%s: Breaking out of NMI-blocked "
6519 			       "state on VCPU %d after 1 s timeout\n",
6520 			       __func__, vcpu->vcpu_id);
6521 			vmx->loaded_vmcs->soft_vnmi_blocked = 0;
6522 		}
6523 	}
6524 
6525 	if (exit_fastpath != EXIT_FASTPATH_NONE)
6526 		return 1;
6527 
6528 	if (exit_reason.basic >= kvm_vmx_max_exit_handlers)
6529 		goto unexpected_vmexit;
6530 #ifdef CONFIG_RETPOLINE
6531 	if (exit_reason.basic == EXIT_REASON_MSR_WRITE)
6532 		return kvm_emulate_wrmsr(vcpu);
6533 	else if (exit_reason.basic == EXIT_REASON_PREEMPTION_TIMER)
6534 		return handle_preemption_timer(vcpu);
6535 	else if (exit_reason.basic == EXIT_REASON_INTERRUPT_WINDOW)
6536 		return handle_interrupt_window(vcpu);
6537 	else if (exit_reason.basic == EXIT_REASON_EXTERNAL_INTERRUPT)
6538 		return handle_external_interrupt(vcpu);
6539 	else if (exit_reason.basic == EXIT_REASON_HLT)
6540 		return kvm_emulate_halt(vcpu);
6541 	else if (exit_reason.basic == EXIT_REASON_EPT_MISCONFIG)
6542 		return handle_ept_misconfig(vcpu);
6543 #endif
6544 
6545 	exit_handler_index = array_index_nospec((u16)exit_reason.basic,
6546 						kvm_vmx_max_exit_handlers);
6547 	if (!kvm_vmx_exit_handlers[exit_handler_index])
6548 		goto unexpected_vmexit;
6549 
6550 	return kvm_vmx_exit_handlers[exit_handler_index](vcpu);
6551 
6552 unexpected_vmexit:
6553 	vcpu_unimpl(vcpu, "vmx: unexpected exit reason 0x%x\n",
6554 		    exit_reason.full);
6555 	dump_vmcs(vcpu);
6556 	vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
6557 	vcpu->run->internal.suberror =
6558 			KVM_INTERNAL_ERROR_UNEXPECTED_EXIT_REASON;
6559 	vcpu->run->internal.ndata = 2;
6560 	vcpu->run->internal.data[0] = exit_reason.full;
6561 	vcpu->run->internal.data[1] = vcpu->arch.last_vmentry_cpu;
6562 	return 0;
6563 }
6564 
6565 static int vmx_handle_exit(struct kvm_vcpu *vcpu, fastpath_t exit_fastpath)
6566 {
6567 	int ret = __vmx_handle_exit(vcpu, exit_fastpath);
6568 
6569 	/*
6570 	 * Exit to user space when bus lock detected to inform that there is
6571 	 * a bus lock in guest.
6572 	 */
6573 	if (to_vmx(vcpu)->exit_reason.bus_lock_detected) {
6574 		if (ret > 0)
6575 			vcpu->run->exit_reason = KVM_EXIT_X86_BUS_LOCK;
6576 
6577 		vcpu->run->flags |= KVM_RUN_X86_BUS_LOCK;
6578 		return 0;
6579 	}
6580 	return ret;
6581 }
6582 
6583 /*
6584  * Software based L1D cache flush which is used when microcode providing
6585  * the cache control MSR is not loaded.
6586  *
6587  * The L1D cache is 32 KiB on Nehalem and later microarchitectures, but to
6588  * flush it is required to read in 64 KiB because the replacement algorithm
6589  * is not exactly LRU. This could be sized at runtime via topology
6590  * information but as all relevant affected CPUs have 32KiB L1D cache size
6591  * there is no point in doing so.
6592  */
6593 static noinstr void vmx_l1d_flush(struct kvm_vcpu *vcpu)
6594 {
6595 	int size = PAGE_SIZE << L1D_CACHE_ORDER;
6596 
6597 	/*
6598 	 * This code is only executed when the flush mode is 'cond' or
6599 	 * 'always'
6600 	 */
6601 	if (static_branch_likely(&vmx_l1d_flush_cond)) {
6602 		bool flush_l1d;
6603 
6604 		/*
6605 		 * Clear the per-vcpu flush bit, it gets set again
6606 		 * either from vcpu_run() or from one of the unsafe
6607 		 * VMEXIT handlers.
6608 		 */
6609 		flush_l1d = vcpu->arch.l1tf_flush_l1d;
6610 		vcpu->arch.l1tf_flush_l1d = false;
6611 
6612 		/*
6613 		 * Clear the per-cpu flush bit, it gets set again from
6614 		 * the interrupt handlers.
6615 		 */
6616 		flush_l1d |= kvm_get_cpu_l1tf_flush_l1d();
6617 		kvm_clear_cpu_l1tf_flush_l1d();
6618 
6619 		if (!flush_l1d)
6620 			return;
6621 	}
6622 
6623 	vcpu->stat.l1d_flush++;
6624 
6625 	if (static_cpu_has(X86_FEATURE_FLUSH_L1D)) {
6626 		native_wrmsrl(MSR_IA32_FLUSH_CMD, L1D_FLUSH);
6627 		return;
6628 	}
6629 
6630 	asm volatile(
6631 		/* First ensure the pages are in the TLB */
6632 		"xorl	%%eax, %%eax\n"
6633 		".Lpopulate_tlb:\n\t"
6634 		"movzbl	(%[flush_pages], %%" _ASM_AX "), %%ecx\n\t"
6635 		"addl	$4096, %%eax\n\t"
6636 		"cmpl	%%eax, %[size]\n\t"
6637 		"jne	.Lpopulate_tlb\n\t"
6638 		"xorl	%%eax, %%eax\n\t"
6639 		"cpuid\n\t"
6640 		/* Now fill the cache */
6641 		"xorl	%%eax, %%eax\n"
6642 		".Lfill_cache:\n"
6643 		"movzbl	(%[flush_pages], %%" _ASM_AX "), %%ecx\n\t"
6644 		"addl	$64, %%eax\n\t"
6645 		"cmpl	%%eax, %[size]\n\t"
6646 		"jne	.Lfill_cache\n\t"
6647 		"lfence\n"
6648 		:: [flush_pages] "r" (vmx_l1d_flush_pages),
6649 		    [size] "r" (size)
6650 		: "eax", "ebx", "ecx", "edx");
6651 }
6652 
6653 static void vmx_update_cr8_intercept(struct kvm_vcpu *vcpu, int tpr, int irr)
6654 {
6655 	struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
6656 	int tpr_threshold;
6657 
6658 	if (is_guest_mode(vcpu) &&
6659 		nested_cpu_has(vmcs12, CPU_BASED_TPR_SHADOW))
6660 		return;
6661 
6662 	tpr_threshold = (irr == -1 || tpr < irr) ? 0 : irr;
6663 	if (is_guest_mode(vcpu))
6664 		to_vmx(vcpu)->nested.l1_tpr_threshold = tpr_threshold;
6665 	else
6666 		vmcs_write32(TPR_THRESHOLD, tpr_threshold);
6667 }
6668 
6669 void vmx_set_virtual_apic_mode(struct kvm_vcpu *vcpu)
6670 {
6671 	struct vcpu_vmx *vmx = to_vmx(vcpu);
6672 	u32 sec_exec_control;
6673 
6674 	if (!lapic_in_kernel(vcpu))
6675 		return;
6676 
6677 	if (!flexpriority_enabled &&
6678 	    !cpu_has_vmx_virtualize_x2apic_mode())
6679 		return;
6680 
6681 	/* Postpone execution until vmcs01 is the current VMCS. */
6682 	if (is_guest_mode(vcpu)) {
6683 		vmx->nested.change_vmcs01_virtual_apic_mode = true;
6684 		return;
6685 	}
6686 
6687 	sec_exec_control = secondary_exec_controls_get(vmx);
6688 	sec_exec_control &= ~(SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES |
6689 			      SECONDARY_EXEC_VIRTUALIZE_X2APIC_MODE);
6690 
6691 	switch (kvm_get_apic_mode(vcpu)) {
6692 	case LAPIC_MODE_INVALID:
6693 		WARN_ONCE(true, "Invalid local APIC state");
6694 		break;
6695 	case LAPIC_MODE_DISABLED:
6696 		break;
6697 	case LAPIC_MODE_XAPIC:
6698 		if (flexpriority_enabled) {
6699 			sec_exec_control |=
6700 				SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES;
6701 			kvm_make_request(KVM_REQ_APIC_PAGE_RELOAD, vcpu);
6702 
6703 			/*
6704 			 * Flush the TLB, reloading the APIC access page will
6705 			 * only do so if its physical address has changed, but
6706 			 * the guest may have inserted a non-APIC mapping into
6707 			 * the TLB while the APIC access page was disabled.
6708 			 */
6709 			kvm_make_request(KVM_REQ_TLB_FLUSH_CURRENT, vcpu);
6710 		}
6711 		break;
6712 	case LAPIC_MODE_X2APIC:
6713 		if (cpu_has_vmx_virtualize_x2apic_mode())
6714 			sec_exec_control |=
6715 				SECONDARY_EXEC_VIRTUALIZE_X2APIC_MODE;
6716 		break;
6717 	}
6718 	secondary_exec_controls_set(vmx, sec_exec_control);
6719 
6720 	vmx_update_msr_bitmap_x2apic(vcpu);
6721 }
6722 
6723 static void vmx_set_apic_access_page_addr(struct kvm_vcpu *vcpu)
6724 {
6725 	const gfn_t gfn = APIC_DEFAULT_PHYS_BASE >> PAGE_SHIFT;
6726 	struct kvm *kvm = vcpu->kvm;
6727 	struct kvm_memslots *slots = kvm_memslots(kvm);
6728 	struct kvm_memory_slot *slot;
6729 	unsigned long mmu_seq;
6730 	kvm_pfn_t pfn;
6731 
6732 	/* Defer reload until vmcs01 is the current VMCS. */
6733 	if (is_guest_mode(vcpu)) {
6734 		to_vmx(vcpu)->nested.reload_vmcs01_apic_access_page = true;
6735 		return;
6736 	}
6737 
6738 	if (!(secondary_exec_controls_get(to_vmx(vcpu)) &
6739 	    SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES))
6740 		return;
6741 
6742 	/*
6743 	 * Grab the memslot so that the hva lookup for the mmu_notifier retry
6744 	 * is guaranteed to use the same memslot as the pfn lookup, i.e. rely
6745 	 * on the pfn lookup's validation of the memslot to ensure a valid hva
6746 	 * is used for the retry check.
6747 	 */
6748 	slot = id_to_memslot(slots, APIC_ACCESS_PAGE_PRIVATE_MEMSLOT);
6749 	if (!slot || slot->flags & KVM_MEMSLOT_INVALID)
6750 		return;
6751 
6752 	/*
6753 	 * Ensure that the mmu_notifier sequence count is read before KVM
6754 	 * retrieves the pfn from the primary MMU.  Note, the memslot is
6755 	 * protected by SRCU, not the mmu_notifier.  Pairs with the smp_wmb()
6756 	 * in kvm_mmu_invalidate_end().
6757 	 */
6758 	mmu_seq = kvm->mmu_invalidate_seq;
6759 	smp_rmb();
6760 
6761 	/*
6762 	 * No need to retry if the memslot does not exist or is invalid.  KVM
6763 	 * controls the APIC-access page memslot, and only deletes the memslot
6764 	 * if APICv is permanently inhibited, i.e. the memslot won't reappear.
6765 	 */
6766 	pfn = gfn_to_pfn_memslot(slot, gfn);
6767 	if (is_error_noslot_pfn(pfn))
6768 		return;
6769 
6770 	read_lock(&vcpu->kvm->mmu_lock);
6771 	if (mmu_invalidate_retry_hva(kvm, mmu_seq,
6772 				     gfn_to_hva_memslot(slot, gfn))) {
6773 		kvm_make_request(KVM_REQ_APIC_PAGE_RELOAD, vcpu);
6774 		read_unlock(&vcpu->kvm->mmu_lock);
6775 		goto out;
6776 	}
6777 
6778 	vmcs_write64(APIC_ACCESS_ADDR, pfn_to_hpa(pfn));
6779 	read_unlock(&vcpu->kvm->mmu_lock);
6780 
6781 	vmx_flush_tlb_current(vcpu);
6782 
6783 out:
6784 	/*
6785 	 * Do not pin apic access page in memory, the MMU notifier
6786 	 * will call us again if it is migrated or swapped out.
6787 	 */
6788 	kvm_release_pfn_clean(pfn);
6789 }
6790 
6791 static void vmx_hwapic_isr_update(int max_isr)
6792 {
6793 	u16 status;
6794 	u8 old;
6795 
6796 	if (max_isr == -1)
6797 		max_isr = 0;
6798 
6799 	status = vmcs_read16(GUEST_INTR_STATUS);
6800 	old = status >> 8;
6801 	if (max_isr != old) {
6802 		status &= 0xff;
6803 		status |= max_isr << 8;
6804 		vmcs_write16(GUEST_INTR_STATUS, status);
6805 	}
6806 }
6807 
6808 static void vmx_set_rvi(int vector)
6809 {
6810 	u16 status;
6811 	u8 old;
6812 
6813 	if (vector == -1)
6814 		vector = 0;
6815 
6816 	status = vmcs_read16(GUEST_INTR_STATUS);
6817 	old = (u8)status & 0xff;
6818 	if ((u8)vector != old) {
6819 		status &= ~0xff;
6820 		status |= (u8)vector;
6821 		vmcs_write16(GUEST_INTR_STATUS, status);
6822 	}
6823 }
6824 
6825 static void vmx_hwapic_irr_update(struct kvm_vcpu *vcpu, int max_irr)
6826 {
6827 	/*
6828 	 * When running L2, updating RVI is only relevant when
6829 	 * vmcs12 virtual-interrupt-delivery enabled.
6830 	 * However, it can be enabled only when L1 also
6831 	 * intercepts external-interrupts and in that case
6832 	 * we should not update vmcs02 RVI but instead intercept
6833 	 * interrupt. Therefore, do nothing when running L2.
6834 	 */
6835 	if (!is_guest_mode(vcpu))
6836 		vmx_set_rvi(max_irr);
6837 }
6838 
6839 static int vmx_sync_pir_to_irr(struct kvm_vcpu *vcpu)
6840 {
6841 	struct vcpu_vmx *vmx = to_vmx(vcpu);
6842 	int max_irr;
6843 	bool got_posted_interrupt;
6844 
6845 	if (KVM_BUG_ON(!enable_apicv, vcpu->kvm))
6846 		return -EIO;
6847 
6848 	if (pi_test_on(&vmx->pi_desc)) {
6849 		pi_clear_on(&vmx->pi_desc);
6850 		/*
6851 		 * IOMMU can write to PID.ON, so the barrier matters even on UP.
6852 		 * But on x86 this is just a compiler barrier anyway.
6853 		 */
6854 		smp_mb__after_atomic();
6855 		got_posted_interrupt =
6856 			kvm_apic_update_irr(vcpu, vmx->pi_desc.pir, &max_irr);
6857 	} else {
6858 		max_irr = kvm_lapic_find_highest_irr(vcpu);
6859 		got_posted_interrupt = false;
6860 	}
6861 
6862 	/*
6863 	 * Newly recognized interrupts are injected via either virtual interrupt
6864 	 * delivery (RVI) or KVM_REQ_EVENT.  Virtual interrupt delivery is
6865 	 * disabled in two cases:
6866 	 *
6867 	 * 1) If L2 is running and the vCPU has a new pending interrupt.  If L1
6868 	 * wants to exit on interrupts, KVM_REQ_EVENT is needed to synthesize a
6869 	 * VM-Exit to L1.  If L1 doesn't want to exit, the interrupt is injected
6870 	 * into L2, but KVM doesn't use virtual interrupt delivery to inject
6871 	 * interrupts into L2, and so KVM_REQ_EVENT is again needed.
6872 	 *
6873 	 * 2) If APICv is disabled for this vCPU, assigned devices may still
6874 	 * attempt to post interrupts.  The posted interrupt vector will cause
6875 	 * a VM-Exit and the subsequent entry will call sync_pir_to_irr.
6876 	 */
6877 	if (!is_guest_mode(vcpu) && kvm_vcpu_apicv_active(vcpu))
6878 		vmx_set_rvi(max_irr);
6879 	else if (got_posted_interrupt)
6880 		kvm_make_request(KVM_REQ_EVENT, vcpu);
6881 
6882 	return max_irr;
6883 }
6884 
6885 static void vmx_load_eoi_exitmap(struct kvm_vcpu *vcpu, u64 *eoi_exit_bitmap)
6886 {
6887 	if (!kvm_vcpu_apicv_active(vcpu))
6888 		return;
6889 
6890 	vmcs_write64(EOI_EXIT_BITMAP0, eoi_exit_bitmap[0]);
6891 	vmcs_write64(EOI_EXIT_BITMAP1, eoi_exit_bitmap[1]);
6892 	vmcs_write64(EOI_EXIT_BITMAP2, eoi_exit_bitmap[2]);
6893 	vmcs_write64(EOI_EXIT_BITMAP3, eoi_exit_bitmap[3]);
6894 }
6895 
6896 static void vmx_apicv_post_state_restore(struct kvm_vcpu *vcpu)
6897 {
6898 	struct vcpu_vmx *vmx = to_vmx(vcpu);
6899 
6900 	pi_clear_on(&vmx->pi_desc);
6901 	memset(vmx->pi_desc.pir, 0, sizeof(vmx->pi_desc.pir));
6902 }
6903 
6904 void vmx_do_interrupt_irqoff(unsigned long entry);
6905 void vmx_do_nmi_irqoff(void);
6906 
6907 static void handle_nm_fault_irqoff(struct kvm_vcpu *vcpu)
6908 {
6909 	/*
6910 	 * Save xfd_err to guest_fpu before interrupt is enabled, so the
6911 	 * MSR value is not clobbered by the host activity before the guest
6912 	 * has chance to consume it.
6913 	 *
6914 	 * Do not blindly read xfd_err here, since this exception might
6915 	 * be caused by L1 interception on a platform which doesn't
6916 	 * support xfd at all.
6917 	 *
6918 	 * Do it conditionally upon guest_fpu::xfd. xfd_err matters
6919 	 * only when xfd contains a non-zero value.
6920 	 *
6921 	 * Queuing exception is done in vmx_handle_exit. See comment there.
6922 	 */
6923 	if (vcpu->arch.guest_fpu.fpstate->xfd)
6924 		rdmsrl(MSR_IA32_XFD_ERR, vcpu->arch.guest_fpu.xfd_err);
6925 }
6926 
6927 static void handle_exception_irqoff(struct vcpu_vmx *vmx)
6928 {
6929 	u32 intr_info = vmx_get_intr_info(&vmx->vcpu);
6930 
6931 	/* if exit due to PF check for async PF */
6932 	if (is_page_fault(intr_info))
6933 		vmx->vcpu.arch.apf.host_apf_flags = kvm_read_and_reset_apf_flags();
6934 	/* if exit due to NM, handle before interrupts are enabled */
6935 	else if (is_nm_fault(intr_info))
6936 		handle_nm_fault_irqoff(&vmx->vcpu);
6937 	/* Handle machine checks before interrupts are enabled */
6938 	else if (is_machine_check(intr_info))
6939 		kvm_machine_check();
6940 }
6941 
6942 static void handle_external_interrupt_irqoff(struct kvm_vcpu *vcpu)
6943 {
6944 	u32 intr_info = vmx_get_intr_info(vcpu);
6945 	unsigned int vector = intr_info & INTR_INFO_VECTOR_MASK;
6946 	gate_desc *desc = (gate_desc *)host_idt_base + vector;
6947 
6948 	if (KVM_BUG(!is_external_intr(intr_info), vcpu->kvm,
6949 	    "unexpected VM-Exit interrupt info: 0x%x", intr_info))
6950 		return;
6951 
6952 	kvm_before_interrupt(vcpu, KVM_HANDLING_IRQ);
6953 	vmx_do_interrupt_irqoff(gate_offset(desc));
6954 	kvm_after_interrupt(vcpu);
6955 
6956 	vcpu->arch.at_instruction_boundary = true;
6957 }
6958 
6959 static void vmx_handle_exit_irqoff(struct kvm_vcpu *vcpu)
6960 {
6961 	struct vcpu_vmx *vmx = to_vmx(vcpu);
6962 
6963 	if (vmx->emulation_required)
6964 		return;
6965 
6966 	if (vmx->exit_reason.basic == EXIT_REASON_EXTERNAL_INTERRUPT)
6967 		handle_external_interrupt_irqoff(vcpu);
6968 	else if (vmx->exit_reason.basic == EXIT_REASON_EXCEPTION_NMI)
6969 		handle_exception_irqoff(vmx);
6970 }
6971 
6972 /*
6973  * The kvm parameter can be NULL (module initialization, or invocation before
6974  * VM creation). Be sure to check the kvm parameter before using it.
6975  */
6976 static bool vmx_has_emulated_msr(struct kvm *kvm, u32 index)
6977 {
6978 	switch (index) {
6979 	case MSR_IA32_SMBASE:
6980 		if (!IS_ENABLED(CONFIG_KVM_SMM))
6981 			return false;
6982 		/*
6983 		 * We cannot do SMM unless we can run the guest in big
6984 		 * real mode.
6985 		 */
6986 		return enable_unrestricted_guest || emulate_invalid_guest_state;
6987 	case KVM_FIRST_EMULATED_VMX_MSR ... KVM_LAST_EMULATED_VMX_MSR:
6988 		return nested;
6989 	case MSR_AMD64_VIRT_SPEC_CTRL:
6990 	case MSR_AMD64_TSC_RATIO:
6991 		/* This is AMD only.  */
6992 		return false;
6993 	default:
6994 		return true;
6995 	}
6996 }
6997 
6998 static void vmx_recover_nmi_blocking(struct vcpu_vmx *vmx)
6999 {
7000 	u32 exit_intr_info;
7001 	bool unblock_nmi;
7002 	u8 vector;
7003 	bool idtv_info_valid;
7004 
7005 	idtv_info_valid = vmx->idt_vectoring_info & VECTORING_INFO_VALID_MASK;
7006 
7007 	if (enable_vnmi) {
7008 		if (vmx->loaded_vmcs->nmi_known_unmasked)
7009 			return;
7010 
7011 		exit_intr_info = vmx_get_intr_info(&vmx->vcpu);
7012 		unblock_nmi = (exit_intr_info & INTR_INFO_UNBLOCK_NMI) != 0;
7013 		vector = exit_intr_info & INTR_INFO_VECTOR_MASK;
7014 		/*
7015 		 * SDM 3: 27.7.1.2 (September 2008)
7016 		 * Re-set bit "block by NMI" before VM entry if vmexit caused by
7017 		 * a guest IRET fault.
7018 		 * SDM 3: 23.2.2 (September 2008)
7019 		 * Bit 12 is undefined in any of the following cases:
7020 		 *  If the VM exit sets the valid bit in the IDT-vectoring
7021 		 *   information field.
7022 		 *  If the VM exit is due to a double fault.
7023 		 */
7024 		if ((exit_intr_info & INTR_INFO_VALID_MASK) && unblock_nmi &&
7025 		    vector != DF_VECTOR && !idtv_info_valid)
7026 			vmcs_set_bits(GUEST_INTERRUPTIBILITY_INFO,
7027 				      GUEST_INTR_STATE_NMI);
7028 		else
7029 			vmx->loaded_vmcs->nmi_known_unmasked =
7030 				!(vmcs_read32(GUEST_INTERRUPTIBILITY_INFO)
7031 				  & GUEST_INTR_STATE_NMI);
7032 	} else if (unlikely(vmx->loaded_vmcs->soft_vnmi_blocked))
7033 		vmx->loaded_vmcs->vnmi_blocked_time +=
7034 			ktime_to_ns(ktime_sub(ktime_get(),
7035 					      vmx->loaded_vmcs->entry_time));
7036 }
7037 
7038 static void __vmx_complete_interrupts(struct kvm_vcpu *vcpu,
7039 				      u32 idt_vectoring_info,
7040 				      int instr_len_field,
7041 				      int error_code_field)
7042 {
7043 	u8 vector;
7044 	int type;
7045 	bool idtv_info_valid;
7046 
7047 	idtv_info_valid = idt_vectoring_info & VECTORING_INFO_VALID_MASK;
7048 
7049 	vcpu->arch.nmi_injected = false;
7050 	kvm_clear_exception_queue(vcpu);
7051 	kvm_clear_interrupt_queue(vcpu);
7052 
7053 	if (!idtv_info_valid)
7054 		return;
7055 
7056 	kvm_make_request(KVM_REQ_EVENT, vcpu);
7057 
7058 	vector = idt_vectoring_info & VECTORING_INFO_VECTOR_MASK;
7059 	type = idt_vectoring_info & VECTORING_INFO_TYPE_MASK;
7060 
7061 	switch (type) {
7062 	case INTR_TYPE_NMI_INTR:
7063 		vcpu->arch.nmi_injected = true;
7064 		/*
7065 		 * SDM 3: 27.7.1.2 (September 2008)
7066 		 * Clear bit "block by NMI" before VM entry if a NMI
7067 		 * delivery faulted.
7068 		 */
7069 		vmx_set_nmi_mask(vcpu, false);
7070 		break;
7071 	case INTR_TYPE_SOFT_EXCEPTION:
7072 		vcpu->arch.event_exit_inst_len = vmcs_read32(instr_len_field);
7073 		fallthrough;
7074 	case INTR_TYPE_HARD_EXCEPTION:
7075 		if (idt_vectoring_info & VECTORING_INFO_DELIVER_CODE_MASK) {
7076 			u32 err = vmcs_read32(error_code_field);
7077 			kvm_requeue_exception_e(vcpu, vector, err);
7078 		} else
7079 			kvm_requeue_exception(vcpu, vector);
7080 		break;
7081 	case INTR_TYPE_SOFT_INTR:
7082 		vcpu->arch.event_exit_inst_len = vmcs_read32(instr_len_field);
7083 		fallthrough;
7084 	case INTR_TYPE_EXT_INTR:
7085 		kvm_queue_interrupt(vcpu, vector, type == INTR_TYPE_SOFT_INTR);
7086 		break;
7087 	default:
7088 		break;
7089 	}
7090 }
7091 
7092 static void vmx_complete_interrupts(struct vcpu_vmx *vmx)
7093 {
7094 	__vmx_complete_interrupts(&vmx->vcpu, vmx->idt_vectoring_info,
7095 				  VM_EXIT_INSTRUCTION_LEN,
7096 				  IDT_VECTORING_ERROR_CODE);
7097 }
7098 
7099 static void vmx_cancel_injection(struct kvm_vcpu *vcpu)
7100 {
7101 	__vmx_complete_interrupts(vcpu,
7102 				  vmcs_read32(VM_ENTRY_INTR_INFO_FIELD),
7103 				  VM_ENTRY_INSTRUCTION_LEN,
7104 				  VM_ENTRY_EXCEPTION_ERROR_CODE);
7105 
7106 	vmcs_write32(VM_ENTRY_INTR_INFO_FIELD, 0);
7107 }
7108 
7109 static void atomic_switch_perf_msrs(struct vcpu_vmx *vmx)
7110 {
7111 	int i, nr_msrs;
7112 	struct perf_guest_switch_msr *msrs;
7113 	struct kvm_pmu *pmu = vcpu_to_pmu(&vmx->vcpu);
7114 
7115 	pmu->host_cross_mapped_mask = 0;
7116 	if (pmu->pebs_enable & pmu->global_ctrl)
7117 		intel_pmu_cross_mapped_check(pmu);
7118 
7119 	/* Note, nr_msrs may be garbage if perf_guest_get_msrs() returns NULL. */
7120 	msrs = perf_guest_get_msrs(&nr_msrs, (void *)pmu);
7121 	if (!msrs)
7122 		return;
7123 
7124 	for (i = 0; i < nr_msrs; i++)
7125 		if (msrs[i].host == msrs[i].guest)
7126 			clear_atomic_switch_msr(vmx, msrs[i].msr);
7127 		else
7128 			add_atomic_switch_msr(vmx, msrs[i].msr, msrs[i].guest,
7129 					msrs[i].host, false);
7130 }
7131 
7132 static void vmx_update_hv_timer(struct kvm_vcpu *vcpu)
7133 {
7134 	struct vcpu_vmx *vmx = to_vmx(vcpu);
7135 	u64 tscl;
7136 	u32 delta_tsc;
7137 
7138 	if (vmx->req_immediate_exit) {
7139 		vmcs_write32(VMX_PREEMPTION_TIMER_VALUE, 0);
7140 		vmx->loaded_vmcs->hv_timer_soft_disabled = false;
7141 	} else if (vmx->hv_deadline_tsc != -1) {
7142 		tscl = rdtsc();
7143 		if (vmx->hv_deadline_tsc > tscl)
7144 			/* set_hv_timer ensures the delta fits in 32-bits */
7145 			delta_tsc = (u32)((vmx->hv_deadline_tsc - tscl) >>
7146 				cpu_preemption_timer_multi);
7147 		else
7148 			delta_tsc = 0;
7149 
7150 		vmcs_write32(VMX_PREEMPTION_TIMER_VALUE, delta_tsc);
7151 		vmx->loaded_vmcs->hv_timer_soft_disabled = false;
7152 	} else if (!vmx->loaded_vmcs->hv_timer_soft_disabled) {
7153 		vmcs_write32(VMX_PREEMPTION_TIMER_VALUE, -1);
7154 		vmx->loaded_vmcs->hv_timer_soft_disabled = true;
7155 	}
7156 }
7157 
7158 void noinstr vmx_update_host_rsp(struct vcpu_vmx *vmx, unsigned long host_rsp)
7159 {
7160 	if (unlikely(host_rsp != vmx->loaded_vmcs->host_state.rsp)) {
7161 		vmx->loaded_vmcs->host_state.rsp = host_rsp;
7162 		vmcs_writel(HOST_RSP, host_rsp);
7163 	}
7164 }
7165 
7166 void noinstr vmx_spec_ctrl_restore_host(struct vcpu_vmx *vmx,
7167 					unsigned int flags)
7168 {
7169 	u64 hostval = this_cpu_read(x86_spec_ctrl_current);
7170 
7171 	if (!cpu_feature_enabled(X86_FEATURE_MSR_SPEC_CTRL))
7172 		return;
7173 
7174 	if (flags & VMX_RUN_SAVE_SPEC_CTRL)
7175 		vmx->spec_ctrl = __rdmsr(MSR_IA32_SPEC_CTRL);
7176 
7177 	/*
7178 	 * If the guest/host SPEC_CTRL values differ, restore the host value.
7179 	 *
7180 	 * For legacy IBRS, the IBRS bit always needs to be written after
7181 	 * transitioning from a less privileged predictor mode, regardless of
7182 	 * whether the guest/host values differ.
7183 	 */
7184 	if (cpu_feature_enabled(X86_FEATURE_KERNEL_IBRS) ||
7185 	    vmx->spec_ctrl != hostval)
7186 		native_wrmsrl(MSR_IA32_SPEC_CTRL, hostval);
7187 
7188 	barrier_nospec();
7189 }
7190 
7191 static fastpath_t vmx_exit_handlers_fastpath(struct kvm_vcpu *vcpu)
7192 {
7193 	switch (to_vmx(vcpu)->exit_reason.basic) {
7194 	case EXIT_REASON_MSR_WRITE:
7195 		return handle_fastpath_set_msr_irqoff(vcpu);
7196 	case EXIT_REASON_PREEMPTION_TIMER:
7197 		return handle_fastpath_preemption_timer(vcpu);
7198 	default:
7199 		return EXIT_FASTPATH_NONE;
7200 	}
7201 }
7202 
7203 static noinstr void vmx_vcpu_enter_exit(struct kvm_vcpu *vcpu,
7204 					unsigned int flags)
7205 {
7206 	struct vcpu_vmx *vmx = to_vmx(vcpu);
7207 
7208 	guest_state_enter_irqoff();
7209 
7210 	/* L1D Flush includes CPU buffer clear to mitigate MDS */
7211 	if (static_branch_unlikely(&vmx_l1d_should_flush))
7212 		vmx_l1d_flush(vcpu);
7213 	else if (static_branch_unlikely(&mds_user_clear))
7214 		mds_clear_cpu_buffers();
7215 	else if (static_branch_unlikely(&mmio_stale_data_clear) &&
7216 		 kvm_arch_has_assigned_device(vcpu->kvm))
7217 		mds_clear_cpu_buffers();
7218 
7219 	vmx_disable_fb_clear(vmx);
7220 
7221 	if (vcpu->arch.cr2 != native_read_cr2())
7222 		native_write_cr2(vcpu->arch.cr2);
7223 
7224 	vmx->fail = __vmx_vcpu_run(vmx, (unsigned long *)&vcpu->arch.regs,
7225 				   flags);
7226 
7227 	vcpu->arch.cr2 = native_read_cr2();
7228 
7229 	vmx_enable_fb_clear(vmx);
7230 
7231 	if (unlikely(vmx->fail))
7232 		vmx->exit_reason.full = 0xdead;
7233 	else
7234 		vmx->exit_reason.full = vmcs_read32(VM_EXIT_REASON);
7235 
7236 	if ((u16)vmx->exit_reason.basic == EXIT_REASON_EXCEPTION_NMI &&
7237 	    is_nmi(vmx_get_intr_info(vcpu))) {
7238 		kvm_before_interrupt(vcpu, KVM_HANDLING_NMI);
7239 		vmx_do_nmi_irqoff();
7240 		kvm_after_interrupt(vcpu);
7241 	}
7242 
7243 	guest_state_exit_irqoff();
7244 }
7245 
7246 static fastpath_t vmx_vcpu_run(struct kvm_vcpu *vcpu)
7247 {
7248 	struct vcpu_vmx *vmx = to_vmx(vcpu);
7249 	unsigned long cr3, cr4;
7250 
7251 	/* Record the guest's net vcpu time for enforced NMI injections. */
7252 	if (unlikely(!enable_vnmi &&
7253 		     vmx->loaded_vmcs->soft_vnmi_blocked))
7254 		vmx->loaded_vmcs->entry_time = ktime_get();
7255 
7256 	/*
7257 	 * Don't enter VMX if guest state is invalid, let the exit handler
7258 	 * start emulation until we arrive back to a valid state.  Synthesize a
7259 	 * consistency check VM-Exit due to invalid guest state and bail.
7260 	 */
7261 	if (unlikely(vmx->emulation_required)) {
7262 		vmx->fail = 0;
7263 
7264 		vmx->exit_reason.full = EXIT_REASON_INVALID_STATE;
7265 		vmx->exit_reason.failed_vmentry = 1;
7266 		kvm_register_mark_available(vcpu, VCPU_EXREG_EXIT_INFO_1);
7267 		vmx->exit_qualification = ENTRY_FAIL_DEFAULT;
7268 		kvm_register_mark_available(vcpu, VCPU_EXREG_EXIT_INFO_2);
7269 		vmx->exit_intr_info = 0;
7270 		return EXIT_FASTPATH_NONE;
7271 	}
7272 
7273 	trace_kvm_entry(vcpu);
7274 
7275 	if (vmx->ple_window_dirty) {
7276 		vmx->ple_window_dirty = false;
7277 		vmcs_write32(PLE_WINDOW, vmx->ple_window);
7278 	}
7279 
7280 	/*
7281 	 * We did this in prepare_switch_to_guest, because it needs to
7282 	 * be within srcu_read_lock.
7283 	 */
7284 	WARN_ON_ONCE(vmx->nested.need_vmcs12_to_shadow_sync);
7285 
7286 	if (kvm_register_is_dirty(vcpu, VCPU_REGS_RSP))
7287 		vmcs_writel(GUEST_RSP, vcpu->arch.regs[VCPU_REGS_RSP]);
7288 	if (kvm_register_is_dirty(vcpu, VCPU_REGS_RIP))
7289 		vmcs_writel(GUEST_RIP, vcpu->arch.regs[VCPU_REGS_RIP]);
7290 	vcpu->arch.regs_dirty = 0;
7291 
7292 	/*
7293 	 * Refresh vmcs.HOST_CR3 if necessary.  This must be done immediately
7294 	 * prior to VM-Enter, as the kernel may load a new ASID (PCID) any time
7295 	 * it switches back to the current->mm, which can occur in KVM context
7296 	 * when switching to a temporary mm to patch kernel code, e.g. if KVM
7297 	 * toggles a static key while handling a VM-Exit.
7298 	 */
7299 	cr3 = __get_current_cr3_fast();
7300 	if (unlikely(cr3 != vmx->loaded_vmcs->host_state.cr3)) {
7301 		vmcs_writel(HOST_CR3, cr3);
7302 		vmx->loaded_vmcs->host_state.cr3 = cr3;
7303 	}
7304 
7305 	cr4 = cr4_read_shadow();
7306 	if (unlikely(cr4 != vmx->loaded_vmcs->host_state.cr4)) {
7307 		vmcs_writel(HOST_CR4, cr4);
7308 		vmx->loaded_vmcs->host_state.cr4 = cr4;
7309 	}
7310 
7311 	/* When KVM_DEBUGREG_WONT_EXIT, dr6 is accessible in guest. */
7312 	if (unlikely(vcpu->arch.switch_db_regs & KVM_DEBUGREG_WONT_EXIT))
7313 		set_debugreg(vcpu->arch.dr6, 6);
7314 
7315 	/* When single-stepping over STI and MOV SS, we must clear the
7316 	 * corresponding interruptibility bits in the guest state. Otherwise
7317 	 * vmentry fails as it then expects bit 14 (BS) in pending debug
7318 	 * exceptions being set, but that's not correct for the guest debugging
7319 	 * case. */
7320 	if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)
7321 		vmx_set_interrupt_shadow(vcpu, 0);
7322 
7323 	kvm_load_guest_xsave_state(vcpu);
7324 
7325 	pt_guest_enter(vmx);
7326 
7327 	atomic_switch_perf_msrs(vmx);
7328 	if (intel_pmu_lbr_is_enabled(vcpu))
7329 		vmx_passthrough_lbr_msrs(vcpu);
7330 
7331 	if (enable_preemption_timer)
7332 		vmx_update_hv_timer(vcpu);
7333 
7334 	kvm_wait_lapic_expire(vcpu);
7335 
7336 	/* The actual VMENTER/EXIT is in the .noinstr.text section. */
7337 	vmx_vcpu_enter_exit(vcpu, __vmx_vcpu_run_flags(vmx));
7338 
7339 	/* All fields are clean at this point */
7340 	if (kvm_is_using_evmcs()) {
7341 		current_evmcs->hv_clean_fields |=
7342 			HV_VMX_ENLIGHTENED_CLEAN_FIELD_ALL;
7343 
7344 		current_evmcs->hv_vp_id = kvm_hv_get_vpindex(vcpu);
7345 	}
7346 
7347 	/* MSR_IA32_DEBUGCTLMSR is zeroed on vmexit. Restore it if needed */
7348 	if (vmx->host_debugctlmsr)
7349 		update_debugctlmsr(vmx->host_debugctlmsr);
7350 
7351 #ifndef CONFIG_X86_64
7352 	/*
7353 	 * The sysexit path does not restore ds/es, so we must set them to
7354 	 * a reasonable value ourselves.
7355 	 *
7356 	 * We can't defer this to vmx_prepare_switch_to_host() since that
7357 	 * function may be executed in interrupt context, which saves and
7358 	 * restore segments around it, nullifying its effect.
7359 	 */
7360 	loadsegment(ds, __USER_DS);
7361 	loadsegment(es, __USER_DS);
7362 #endif
7363 
7364 	vcpu->arch.regs_avail &= ~VMX_REGS_LAZY_LOAD_SET;
7365 
7366 	pt_guest_exit(vmx);
7367 
7368 	kvm_load_host_xsave_state(vcpu);
7369 
7370 	if (is_guest_mode(vcpu)) {
7371 		/*
7372 		 * Track VMLAUNCH/VMRESUME that have made past guest state
7373 		 * checking.
7374 		 */
7375 		if (vmx->nested.nested_run_pending &&
7376 		    !vmx->exit_reason.failed_vmentry)
7377 			++vcpu->stat.nested_run;
7378 
7379 		vmx->nested.nested_run_pending = 0;
7380 	}
7381 
7382 	vmx->idt_vectoring_info = 0;
7383 
7384 	if (unlikely(vmx->fail))
7385 		return EXIT_FASTPATH_NONE;
7386 
7387 	if (unlikely((u16)vmx->exit_reason.basic == EXIT_REASON_MCE_DURING_VMENTRY))
7388 		kvm_machine_check();
7389 
7390 	if (likely(!vmx->exit_reason.failed_vmentry))
7391 		vmx->idt_vectoring_info = vmcs_read32(IDT_VECTORING_INFO_FIELD);
7392 
7393 	trace_kvm_exit(vcpu, KVM_ISA_VMX);
7394 
7395 	if (unlikely(vmx->exit_reason.failed_vmentry))
7396 		return EXIT_FASTPATH_NONE;
7397 
7398 	vmx->loaded_vmcs->launched = 1;
7399 
7400 	vmx_recover_nmi_blocking(vmx);
7401 	vmx_complete_interrupts(vmx);
7402 
7403 	if (is_guest_mode(vcpu))
7404 		return EXIT_FASTPATH_NONE;
7405 
7406 	return vmx_exit_handlers_fastpath(vcpu);
7407 }
7408 
7409 static void vmx_vcpu_free(struct kvm_vcpu *vcpu)
7410 {
7411 	struct vcpu_vmx *vmx = to_vmx(vcpu);
7412 
7413 	if (enable_pml)
7414 		vmx_destroy_pml_buffer(vmx);
7415 	free_vpid(vmx->vpid);
7416 	nested_vmx_free_vcpu(vcpu);
7417 	free_loaded_vmcs(vmx->loaded_vmcs);
7418 }
7419 
7420 static int vmx_vcpu_create(struct kvm_vcpu *vcpu)
7421 {
7422 	struct vmx_uret_msr *tsx_ctrl;
7423 	struct vcpu_vmx *vmx;
7424 	int i, err;
7425 
7426 	BUILD_BUG_ON(offsetof(struct vcpu_vmx, vcpu) != 0);
7427 	vmx = to_vmx(vcpu);
7428 
7429 	INIT_LIST_HEAD(&vmx->pi_wakeup_list);
7430 
7431 	err = -ENOMEM;
7432 
7433 	vmx->vpid = allocate_vpid();
7434 
7435 	/*
7436 	 * If PML is turned on, failure on enabling PML just results in failure
7437 	 * of creating the vcpu, therefore we can simplify PML logic (by
7438 	 * avoiding dealing with cases, such as enabling PML partially on vcpus
7439 	 * for the guest), etc.
7440 	 */
7441 	if (enable_pml) {
7442 		vmx->pml_pg = alloc_page(GFP_KERNEL_ACCOUNT | __GFP_ZERO);
7443 		if (!vmx->pml_pg)
7444 			goto free_vpid;
7445 	}
7446 
7447 	for (i = 0; i < kvm_nr_uret_msrs; ++i)
7448 		vmx->guest_uret_msrs[i].mask = -1ull;
7449 	if (boot_cpu_has(X86_FEATURE_RTM)) {
7450 		/*
7451 		 * TSX_CTRL_CPUID_CLEAR is handled in the CPUID interception.
7452 		 * Keep the host value unchanged to avoid changing CPUID bits
7453 		 * under the host kernel's feet.
7454 		 */
7455 		tsx_ctrl = vmx_find_uret_msr(vmx, MSR_IA32_TSX_CTRL);
7456 		if (tsx_ctrl)
7457 			tsx_ctrl->mask = ~(u64)TSX_CTRL_CPUID_CLEAR;
7458 	}
7459 
7460 	err = alloc_loaded_vmcs(&vmx->vmcs01);
7461 	if (err < 0)
7462 		goto free_pml;
7463 
7464 	/*
7465 	 * Use Hyper-V 'Enlightened MSR Bitmap' feature when KVM runs as a
7466 	 * nested (L1) hypervisor and Hyper-V in L0 supports it. Enable the
7467 	 * feature only for vmcs01, KVM currently isn't equipped to realize any
7468 	 * performance benefits from enabling it for vmcs02.
7469 	 */
7470 	if (kvm_is_using_evmcs() &&
7471 	    (ms_hyperv.nested_features & HV_X64_NESTED_MSR_BITMAP)) {
7472 		struct hv_enlightened_vmcs *evmcs = (void *)vmx->vmcs01.vmcs;
7473 
7474 		evmcs->hv_enlightenments_control.msr_bitmap = 1;
7475 	}
7476 
7477 	/* The MSR bitmap starts with all ones */
7478 	bitmap_fill(vmx->shadow_msr_intercept.read, MAX_POSSIBLE_PASSTHROUGH_MSRS);
7479 	bitmap_fill(vmx->shadow_msr_intercept.write, MAX_POSSIBLE_PASSTHROUGH_MSRS);
7480 
7481 	vmx_disable_intercept_for_msr(vcpu, MSR_IA32_TSC, MSR_TYPE_R);
7482 #ifdef CONFIG_X86_64
7483 	vmx_disable_intercept_for_msr(vcpu, MSR_FS_BASE, MSR_TYPE_RW);
7484 	vmx_disable_intercept_for_msr(vcpu, MSR_GS_BASE, MSR_TYPE_RW);
7485 	vmx_disable_intercept_for_msr(vcpu, MSR_KERNEL_GS_BASE, MSR_TYPE_RW);
7486 #endif
7487 	vmx_disable_intercept_for_msr(vcpu, MSR_IA32_SYSENTER_CS, MSR_TYPE_RW);
7488 	vmx_disable_intercept_for_msr(vcpu, MSR_IA32_SYSENTER_ESP, MSR_TYPE_RW);
7489 	vmx_disable_intercept_for_msr(vcpu, MSR_IA32_SYSENTER_EIP, MSR_TYPE_RW);
7490 	if (kvm_cstate_in_guest(vcpu->kvm)) {
7491 		vmx_disable_intercept_for_msr(vcpu, MSR_CORE_C1_RES, MSR_TYPE_R);
7492 		vmx_disable_intercept_for_msr(vcpu, MSR_CORE_C3_RESIDENCY, MSR_TYPE_R);
7493 		vmx_disable_intercept_for_msr(vcpu, MSR_CORE_C6_RESIDENCY, MSR_TYPE_R);
7494 		vmx_disable_intercept_for_msr(vcpu, MSR_CORE_C7_RESIDENCY, MSR_TYPE_R);
7495 	}
7496 
7497 	vmx->loaded_vmcs = &vmx->vmcs01;
7498 
7499 	if (cpu_need_virtualize_apic_accesses(vcpu)) {
7500 		err = kvm_alloc_apic_access_page(vcpu->kvm);
7501 		if (err)
7502 			goto free_vmcs;
7503 	}
7504 
7505 	if (enable_ept && !enable_unrestricted_guest) {
7506 		err = init_rmode_identity_map(vcpu->kvm);
7507 		if (err)
7508 			goto free_vmcs;
7509 	}
7510 
7511 	if (vmx_can_use_ipiv(vcpu))
7512 		WRITE_ONCE(to_kvm_vmx(vcpu->kvm)->pid_table[vcpu->vcpu_id],
7513 			   __pa(&vmx->pi_desc) | PID_TABLE_ENTRY_VALID);
7514 
7515 	return 0;
7516 
7517 free_vmcs:
7518 	free_loaded_vmcs(vmx->loaded_vmcs);
7519 free_pml:
7520 	vmx_destroy_pml_buffer(vmx);
7521 free_vpid:
7522 	free_vpid(vmx->vpid);
7523 	return err;
7524 }
7525 
7526 #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"
7527 #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"
7528 
7529 static int vmx_vm_init(struct kvm *kvm)
7530 {
7531 	if (!ple_gap)
7532 		kvm->arch.pause_in_guest = true;
7533 
7534 	if (boot_cpu_has(X86_BUG_L1TF) && enable_ept) {
7535 		switch (l1tf_mitigation) {
7536 		case L1TF_MITIGATION_OFF:
7537 		case L1TF_MITIGATION_FLUSH_NOWARN:
7538 			/* 'I explicitly don't care' is set */
7539 			break;
7540 		case L1TF_MITIGATION_FLUSH:
7541 		case L1TF_MITIGATION_FLUSH_NOSMT:
7542 		case L1TF_MITIGATION_FULL:
7543 			/*
7544 			 * Warn upon starting the first VM in a potentially
7545 			 * insecure environment.
7546 			 */
7547 			if (sched_smt_active())
7548 				pr_warn_once(L1TF_MSG_SMT);
7549 			if (l1tf_vmx_mitigation == VMENTER_L1D_FLUSH_NEVER)
7550 				pr_warn_once(L1TF_MSG_L1D);
7551 			break;
7552 		case L1TF_MITIGATION_FULL_FORCE:
7553 			/* Flush is enforced */
7554 			break;
7555 		}
7556 	}
7557 	return 0;
7558 }
7559 
7560 static u8 vmx_get_mt_mask(struct kvm_vcpu *vcpu, gfn_t gfn, bool is_mmio)
7561 {
7562 	u8 cache;
7563 
7564 	/* We wanted to honor guest CD/MTRR/PAT, but doing so could result in
7565 	 * memory aliases with conflicting memory types and sometimes MCEs.
7566 	 * We have to be careful as to what are honored and when.
7567 	 *
7568 	 * For MMIO, guest CD/MTRR are ignored.  The EPT memory type is set to
7569 	 * UC.  The effective memory type is UC or WC depending on guest PAT.
7570 	 * This was historically the source of MCEs and we want to be
7571 	 * conservative.
7572 	 *
7573 	 * When there is no need to deal with noncoherent DMA (e.g., no VT-d
7574 	 * or VT-d has snoop control), guest CD/MTRR/PAT are all ignored.  The
7575 	 * EPT memory type is set to WB.  The effective memory type is forced
7576 	 * WB.
7577 	 *
7578 	 * Otherwise, we trust guest.  Guest CD/MTRR/PAT are all honored.  The
7579 	 * EPT memory type is used to emulate guest CD/MTRR.
7580 	 */
7581 
7582 	if (is_mmio)
7583 		return MTRR_TYPE_UNCACHABLE << VMX_EPT_MT_EPTE_SHIFT;
7584 
7585 	if (!kvm_arch_has_noncoherent_dma(vcpu->kvm))
7586 		return (MTRR_TYPE_WRBACK << VMX_EPT_MT_EPTE_SHIFT) | VMX_EPT_IPAT_BIT;
7587 
7588 	if (kvm_read_cr0_bits(vcpu, X86_CR0_CD)) {
7589 		if (kvm_check_has_quirk(vcpu->kvm, KVM_X86_QUIRK_CD_NW_CLEARED))
7590 			cache = MTRR_TYPE_WRBACK;
7591 		else
7592 			cache = MTRR_TYPE_UNCACHABLE;
7593 
7594 		return (cache << VMX_EPT_MT_EPTE_SHIFT) | VMX_EPT_IPAT_BIT;
7595 	}
7596 
7597 	return kvm_mtrr_get_guest_memory_type(vcpu, gfn) << VMX_EPT_MT_EPTE_SHIFT;
7598 }
7599 
7600 static void vmcs_set_secondary_exec_control(struct vcpu_vmx *vmx, u32 new_ctl)
7601 {
7602 	/*
7603 	 * These bits in the secondary execution controls field
7604 	 * are dynamic, the others are mostly based on the hypervisor
7605 	 * architecture and the guest's CPUID.  Do not touch the
7606 	 * dynamic bits.
7607 	 */
7608 	u32 mask =
7609 		SECONDARY_EXEC_SHADOW_VMCS |
7610 		SECONDARY_EXEC_VIRTUALIZE_X2APIC_MODE |
7611 		SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES |
7612 		SECONDARY_EXEC_DESC;
7613 
7614 	u32 cur_ctl = secondary_exec_controls_get(vmx);
7615 
7616 	secondary_exec_controls_set(vmx, (new_ctl & ~mask) | (cur_ctl & mask));
7617 }
7618 
7619 /*
7620  * Generate MSR_IA32_VMX_CR{0,4}_FIXED1 according to CPUID. Only set bits
7621  * (indicating "allowed-1") if they are supported in the guest's CPUID.
7622  */
7623 static void nested_vmx_cr_fixed1_bits_update(struct kvm_vcpu *vcpu)
7624 {
7625 	struct vcpu_vmx *vmx = to_vmx(vcpu);
7626 	struct kvm_cpuid_entry2 *entry;
7627 
7628 	vmx->nested.msrs.cr0_fixed1 = 0xffffffff;
7629 	vmx->nested.msrs.cr4_fixed1 = X86_CR4_PCE;
7630 
7631 #define cr4_fixed1_update(_cr4_mask, _reg, _cpuid_mask) do {		\
7632 	if (entry && (entry->_reg & (_cpuid_mask)))			\
7633 		vmx->nested.msrs.cr4_fixed1 |= (_cr4_mask);	\
7634 } while (0)
7635 
7636 	entry = kvm_find_cpuid_entry(vcpu, 0x1);
7637 	cr4_fixed1_update(X86_CR4_VME,        edx, feature_bit(VME));
7638 	cr4_fixed1_update(X86_CR4_PVI,        edx, feature_bit(VME));
7639 	cr4_fixed1_update(X86_CR4_TSD,        edx, feature_bit(TSC));
7640 	cr4_fixed1_update(X86_CR4_DE,         edx, feature_bit(DE));
7641 	cr4_fixed1_update(X86_CR4_PSE,        edx, feature_bit(PSE));
7642 	cr4_fixed1_update(X86_CR4_PAE,        edx, feature_bit(PAE));
7643 	cr4_fixed1_update(X86_CR4_MCE,        edx, feature_bit(MCE));
7644 	cr4_fixed1_update(X86_CR4_PGE,        edx, feature_bit(PGE));
7645 	cr4_fixed1_update(X86_CR4_OSFXSR,     edx, feature_bit(FXSR));
7646 	cr4_fixed1_update(X86_CR4_OSXMMEXCPT, edx, feature_bit(XMM));
7647 	cr4_fixed1_update(X86_CR4_VMXE,       ecx, feature_bit(VMX));
7648 	cr4_fixed1_update(X86_CR4_SMXE,       ecx, feature_bit(SMX));
7649 	cr4_fixed1_update(X86_CR4_PCIDE,      ecx, feature_bit(PCID));
7650 	cr4_fixed1_update(X86_CR4_OSXSAVE,    ecx, feature_bit(XSAVE));
7651 
7652 	entry = kvm_find_cpuid_entry_index(vcpu, 0x7, 0);
7653 	cr4_fixed1_update(X86_CR4_FSGSBASE,   ebx, feature_bit(FSGSBASE));
7654 	cr4_fixed1_update(X86_CR4_SMEP,       ebx, feature_bit(SMEP));
7655 	cr4_fixed1_update(X86_CR4_SMAP,       ebx, feature_bit(SMAP));
7656 	cr4_fixed1_update(X86_CR4_PKE,        ecx, feature_bit(PKU));
7657 	cr4_fixed1_update(X86_CR4_UMIP,       ecx, feature_bit(UMIP));
7658 	cr4_fixed1_update(X86_CR4_LA57,       ecx, feature_bit(LA57));
7659 
7660 #undef cr4_fixed1_update
7661 }
7662 
7663 static void update_intel_pt_cfg(struct kvm_vcpu *vcpu)
7664 {
7665 	struct vcpu_vmx *vmx = to_vmx(vcpu);
7666 	struct kvm_cpuid_entry2 *best = NULL;
7667 	int i;
7668 
7669 	for (i = 0; i < PT_CPUID_LEAVES; i++) {
7670 		best = kvm_find_cpuid_entry_index(vcpu, 0x14, i);
7671 		if (!best)
7672 			return;
7673 		vmx->pt_desc.caps[CPUID_EAX + i*PT_CPUID_REGS_NUM] = best->eax;
7674 		vmx->pt_desc.caps[CPUID_EBX + i*PT_CPUID_REGS_NUM] = best->ebx;
7675 		vmx->pt_desc.caps[CPUID_ECX + i*PT_CPUID_REGS_NUM] = best->ecx;
7676 		vmx->pt_desc.caps[CPUID_EDX + i*PT_CPUID_REGS_NUM] = best->edx;
7677 	}
7678 
7679 	/* Get the number of configurable Address Ranges for filtering */
7680 	vmx->pt_desc.num_address_ranges = intel_pt_validate_cap(vmx->pt_desc.caps,
7681 						PT_CAP_num_address_ranges);
7682 
7683 	/* Initialize and clear the no dependency bits */
7684 	vmx->pt_desc.ctl_bitmask = ~(RTIT_CTL_TRACEEN | RTIT_CTL_OS |
7685 			RTIT_CTL_USR | RTIT_CTL_TSC_EN | RTIT_CTL_DISRETC |
7686 			RTIT_CTL_BRANCH_EN);
7687 
7688 	/*
7689 	 * If CPUID.(EAX=14H,ECX=0):EBX[0]=1 CR3Filter can be set otherwise
7690 	 * will inject an #GP
7691 	 */
7692 	if (intel_pt_validate_cap(vmx->pt_desc.caps, PT_CAP_cr3_filtering))
7693 		vmx->pt_desc.ctl_bitmask &= ~RTIT_CTL_CR3EN;
7694 
7695 	/*
7696 	 * If CPUID.(EAX=14H,ECX=0):EBX[1]=1 CYCEn, CycThresh and
7697 	 * PSBFreq can be set
7698 	 */
7699 	if (intel_pt_validate_cap(vmx->pt_desc.caps, PT_CAP_psb_cyc))
7700 		vmx->pt_desc.ctl_bitmask &= ~(RTIT_CTL_CYCLEACC |
7701 				RTIT_CTL_CYC_THRESH | RTIT_CTL_PSB_FREQ);
7702 
7703 	/*
7704 	 * If CPUID.(EAX=14H,ECX=0):EBX[3]=1 MTCEn and MTCFreq can be set
7705 	 */
7706 	if (intel_pt_validate_cap(vmx->pt_desc.caps, PT_CAP_mtc))
7707 		vmx->pt_desc.ctl_bitmask &= ~(RTIT_CTL_MTC_EN |
7708 					      RTIT_CTL_MTC_RANGE);
7709 
7710 	/* If CPUID.(EAX=14H,ECX=0):EBX[4]=1 FUPonPTW and PTWEn can be set */
7711 	if (intel_pt_validate_cap(vmx->pt_desc.caps, PT_CAP_ptwrite))
7712 		vmx->pt_desc.ctl_bitmask &= ~(RTIT_CTL_FUP_ON_PTW |
7713 							RTIT_CTL_PTW_EN);
7714 
7715 	/* If CPUID.(EAX=14H,ECX=0):EBX[5]=1 PwrEvEn can be set */
7716 	if (intel_pt_validate_cap(vmx->pt_desc.caps, PT_CAP_power_event_trace))
7717 		vmx->pt_desc.ctl_bitmask &= ~RTIT_CTL_PWR_EVT_EN;
7718 
7719 	/* If CPUID.(EAX=14H,ECX=0):ECX[0]=1 ToPA can be set */
7720 	if (intel_pt_validate_cap(vmx->pt_desc.caps, PT_CAP_topa_output))
7721 		vmx->pt_desc.ctl_bitmask &= ~RTIT_CTL_TOPA;
7722 
7723 	/* If CPUID.(EAX=14H,ECX=0):ECX[3]=1 FabricEn can be set */
7724 	if (intel_pt_validate_cap(vmx->pt_desc.caps, PT_CAP_output_subsys))
7725 		vmx->pt_desc.ctl_bitmask &= ~RTIT_CTL_FABRIC_EN;
7726 
7727 	/* unmask address range configure area */
7728 	for (i = 0; i < vmx->pt_desc.num_address_ranges; i++)
7729 		vmx->pt_desc.ctl_bitmask &= ~(0xfULL << (32 + i * 4));
7730 }
7731 
7732 static void vmx_vcpu_after_set_cpuid(struct kvm_vcpu *vcpu)
7733 {
7734 	struct vcpu_vmx *vmx = to_vmx(vcpu);
7735 
7736 	/* xsaves_enabled is recomputed in vmx_compute_secondary_exec_control(). */
7737 	vcpu->arch.xsaves_enabled = false;
7738 
7739 	vmx_setup_uret_msrs(vmx);
7740 
7741 	if (cpu_has_secondary_exec_ctrls())
7742 		vmcs_set_secondary_exec_control(vmx,
7743 						vmx_secondary_exec_control(vmx));
7744 
7745 	if (nested_vmx_allowed(vcpu))
7746 		vmx->msr_ia32_feature_control_valid_bits |=
7747 			FEAT_CTL_VMX_ENABLED_INSIDE_SMX |
7748 			FEAT_CTL_VMX_ENABLED_OUTSIDE_SMX;
7749 	else
7750 		vmx->msr_ia32_feature_control_valid_bits &=
7751 			~(FEAT_CTL_VMX_ENABLED_INSIDE_SMX |
7752 			  FEAT_CTL_VMX_ENABLED_OUTSIDE_SMX);
7753 
7754 	if (nested_vmx_allowed(vcpu))
7755 		nested_vmx_cr_fixed1_bits_update(vcpu);
7756 
7757 	if (boot_cpu_has(X86_FEATURE_INTEL_PT) &&
7758 			guest_cpuid_has(vcpu, X86_FEATURE_INTEL_PT))
7759 		update_intel_pt_cfg(vcpu);
7760 
7761 	if (boot_cpu_has(X86_FEATURE_RTM)) {
7762 		struct vmx_uret_msr *msr;
7763 		msr = vmx_find_uret_msr(vmx, MSR_IA32_TSX_CTRL);
7764 		if (msr) {
7765 			bool enabled = guest_cpuid_has(vcpu, X86_FEATURE_RTM);
7766 			vmx_set_guest_uret_msr(vmx, msr, enabled ? 0 : TSX_CTRL_RTM_DISABLE);
7767 		}
7768 	}
7769 
7770 	if (kvm_cpu_cap_has(X86_FEATURE_XFD))
7771 		vmx_set_intercept_for_msr(vcpu, MSR_IA32_XFD_ERR, MSR_TYPE_R,
7772 					  !guest_cpuid_has(vcpu, X86_FEATURE_XFD));
7773 
7774 	if (boot_cpu_has(X86_FEATURE_IBPB))
7775 		vmx_set_intercept_for_msr(vcpu, MSR_IA32_PRED_CMD, MSR_TYPE_W,
7776 					  !guest_has_pred_cmd_msr(vcpu));
7777 
7778 	if (boot_cpu_has(X86_FEATURE_FLUSH_L1D))
7779 		vmx_set_intercept_for_msr(vcpu, MSR_IA32_FLUSH_CMD, MSR_TYPE_W,
7780 					  !guest_cpuid_has(vcpu, X86_FEATURE_FLUSH_L1D));
7781 
7782 	set_cr4_guest_host_mask(vmx);
7783 
7784 	vmx_write_encls_bitmap(vcpu, NULL);
7785 	if (guest_cpuid_has(vcpu, X86_FEATURE_SGX))
7786 		vmx->msr_ia32_feature_control_valid_bits |= FEAT_CTL_SGX_ENABLED;
7787 	else
7788 		vmx->msr_ia32_feature_control_valid_bits &= ~FEAT_CTL_SGX_ENABLED;
7789 
7790 	if (guest_cpuid_has(vcpu, X86_FEATURE_SGX_LC))
7791 		vmx->msr_ia32_feature_control_valid_bits |=
7792 			FEAT_CTL_SGX_LC_ENABLED;
7793 	else
7794 		vmx->msr_ia32_feature_control_valid_bits &=
7795 			~FEAT_CTL_SGX_LC_ENABLED;
7796 
7797 	/* Refresh #PF interception to account for MAXPHYADDR changes. */
7798 	vmx_update_exception_bitmap(vcpu);
7799 }
7800 
7801 static u64 vmx_get_perf_capabilities(void)
7802 {
7803 	u64 perf_cap = PMU_CAP_FW_WRITES;
7804 	struct x86_pmu_lbr lbr;
7805 	u64 host_perf_cap = 0;
7806 
7807 	if (!enable_pmu)
7808 		return 0;
7809 
7810 	if (boot_cpu_has(X86_FEATURE_PDCM))
7811 		rdmsrl(MSR_IA32_PERF_CAPABILITIES, host_perf_cap);
7812 
7813 	if (!cpu_feature_enabled(X86_FEATURE_ARCH_LBR)) {
7814 		x86_perf_get_lbr(&lbr);
7815 		if (lbr.nr)
7816 			perf_cap |= host_perf_cap & PMU_CAP_LBR_FMT;
7817 	}
7818 
7819 	if (vmx_pebs_supported()) {
7820 		perf_cap |= host_perf_cap & PERF_CAP_PEBS_MASK;
7821 		if ((perf_cap & PERF_CAP_PEBS_FORMAT) < 4)
7822 			perf_cap &= ~PERF_CAP_PEBS_BASELINE;
7823 	}
7824 
7825 	return perf_cap;
7826 }
7827 
7828 static __init void vmx_set_cpu_caps(void)
7829 {
7830 	kvm_set_cpu_caps();
7831 
7832 	/* CPUID 0x1 */
7833 	if (nested)
7834 		kvm_cpu_cap_set(X86_FEATURE_VMX);
7835 
7836 	/* CPUID 0x7 */
7837 	if (kvm_mpx_supported())
7838 		kvm_cpu_cap_check_and_set(X86_FEATURE_MPX);
7839 	if (!cpu_has_vmx_invpcid())
7840 		kvm_cpu_cap_clear(X86_FEATURE_INVPCID);
7841 	if (vmx_pt_mode_is_host_guest())
7842 		kvm_cpu_cap_check_and_set(X86_FEATURE_INTEL_PT);
7843 	if (vmx_pebs_supported()) {
7844 		kvm_cpu_cap_check_and_set(X86_FEATURE_DS);
7845 		kvm_cpu_cap_check_and_set(X86_FEATURE_DTES64);
7846 	}
7847 
7848 	if (!enable_pmu)
7849 		kvm_cpu_cap_clear(X86_FEATURE_PDCM);
7850 	kvm_caps.supported_perf_cap = vmx_get_perf_capabilities();
7851 
7852 	if (!enable_sgx) {
7853 		kvm_cpu_cap_clear(X86_FEATURE_SGX);
7854 		kvm_cpu_cap_clear(X86_FEATURE_SGX_LC);
7855 		kvm_cpu_cap_clear(X86_FEATURE_SGX1);
7856 		kvm_cpu_cap_clear(X86_FEATURE_SGX2);
7857 	}
7858 
7859 	if (vmx_umip_emulated())
7860 		kvm_cpu_cap_set(X86_FEATURE_UMIP);
7861 
7862 	/* CPUID 0xD.1 */
7863 	kvm_caps.supported_xss = 0;
7864 	if (!cpu_has_vmx_xsaves())
7865 		kvm_cpu_cap_clear(X86_FEATURE_XSAVES);
7866 
7867 	/* CPUID 0x80000001 and 0x7 (RDPID) */
7868 	if (!cpu_has_vmx_rdtscp()) {
7869 		kvm_cpu_cap_clear(X86_FEATURE_RDTSCP);
7870 		kvm_cpu_cap_clear(X86_FEATURE_RDPID);
7871 	}
7872 
7873 	if (cpu_has_vmx_waitpkg())
7874 		kvm_cpu_cap_check_and_set(X86_FEATURE_WAITPKG);
7875 }
7876 
7877 static void vmx_request_immediate_exit(struct kvm_vcpu *vcpu)
7878 {
7879 	to_vmx(vcpu)->req_immediate_exit = true;
7880 }
7881 
7882 static int vmx_check_intercept_io(struct kvm_vcpu *vcpu,
7883 				  struct x86_instruction_info *info)
7884 {
7885 	struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
7886 	unsigned short port;
7887 	bool intercept;
7888 	int size;
7889 
7890 	if (info->intercept == x86_intercept_in ||
7891 	    info->intercept == x86_intercept_ins) {
7892 		port = info->src_val;
7893 		size = info->dst_bytes;
7894 	} else {
7895 		port = info->dst_val;
7896 		size = info->src_bytes;
7897 	}
7898 
7899 	/*
7900 	 * If the 'use IO bitmaps' VM-execution control is 0, IO instruction
7901 	 * VM-exits depend on the 'unconditional IO exiting' VM-execution
7902 	 * control.
7903 	 *
7904 	 * Otherwise, IO instruction VM-exits are controlled by the IO bitmaps.
7905 	 */
7906 	if (!nested_cpu_has(vmcs12, CPU_BASED_USE_IO_BITMAPS))
7907 		intercept = nested_cpu_has(vmcs12,
7908 					   CPU_BASED_UNCOND_IO_EXITING);
7909 	else
7910 		intercept = nested_vmx_check_io_bitmaps(vcpu, port, size);
7911 
7912 	/* FIXME: produce nested vmexit and return X86EMUL_INTERCEPTED.  */
7913 	return intercept ? X86EMUL_UNHANDLEABLE : X86EMUL_CONTINUE;
7914 }
7915 
7916 static int vmx_check_intercept(struct kvm_vcpu *vcpu,
7917 			       struct x86_instruction_info *info,
7918 			       enum x86_intercept_stage stage,
7919 			       struct x86_exception *exception)
7920 {
7921 	struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
7922 
7923 	switch (info->intercept) {
7924 	/*
7925 	 * RDPID causes #UD if disabled through secondary execution controls.
7926 	 * Because it is marked as EmulateOnUD, we need to intercept it here.
7927 	 * Note, RDPID is hidden behind ENABLE_RDTSCP.
7928 	 */
7929 	case x86_intercept_rdpid:
7930 		if (!nested_cpu_has2(vmcs12, SECONDARY_EXEC_ENABLE_RDTSCP)) {
7931 			exception->vector = UD_VECTOR;
7932 			exception->error_code_valid = false;
7933 			return X86EMUL_PROPAGATE_FAULT;
7934 		}
7935 		break;
7936 
7937 	case x86_intercept_in:
7938 	case x86_intercept_ins:
7939 	case x86_intercept_out:
7940 	case x86_intercept_outs:
7941 		return vmx_check_intercept_io(vcpu, info);
7942 
7943 	case x86_intercept_lgdt:
7944 	case x86_intercept_lidt:
7945 	case x86_intercept_lldt:
7946 	case x86_intercept_ltr:
7947 	case x86_intercept_sgdt:
7948 	case x86_intercept_sidt:
7949 	case x86_intercept_sldt:
7950 	case x86_intercept_str:
7951 		if (!nested_cpu_has2(vmcs12, SECONDARY_EXEC_DESC))
7952 			return X86EMUL_CONTINUE;
7953 
7954 		/* FIXME: produce nested vmexit and return X86EMUL_INTERCEPTED.  */
7955 		break;
7956 
7957 	case x86_intercept_pause:
7958 		/*
7959 		 * PAUSE is a single-byte NOP with a REPE prefix, i.e. collides
7960 		 * with vanilla NOPs in the emulator.  Apply the interception
7961 		 * check only to actual PAUSE instructions.  Don't check
7962 		 * PAUSE-loop-exiting, software can't expect a given PAUSE to
7963 		 * exit, i.e. KVM is within its rights to allow L2 to execute
7964 		 * the PAUSE.
7965 		 */
7966 		if ((info->rep_prefix != REPE_PREFIX) ||
7967 		    !nested_cpu_has2(vmcs12, CPU_BASED_PAUSE_EXITING))
7968 			return X86EMUL_CONTINUE;
7969 
7970 		break;
7971 
7972 	/* TODO: check more intercepts... */
7973 	default:
7974 		break;
7975 	}
7976 
7977 	return X86EMUL_UNHANDLEABLE;
7978 }
7979 
7980 #ifdef CONFIG_X86_64
7981 /* (a << shift) / divisor, return 1 if overflow otherwise 0 */
7982 static inline int u64_shl_div_u64(u64 a, unsigned int shift,
7983 				  u64 divisor, u64 *result)
7984 {
7985 	u64 low = a << shift, high = a >> (64 - shift);
7986 
7987 	/* To avoid the overflow on divq */
7988 	if (high >= divisor)
7989 		return 1;
7990 
7991 	/* Low hold the result, high hold rem which is discarded */
7992 	asm("divq %2\n\t" : "=a" (low), "=d" (high) :
7993 	    "rm" (divisor), "0" (low), "1" (high));
7994 	*result = low;
7995 
7996 	return 0;
7997 }
7998 
7999 static int vmx_set_hv_timer(struct kvm_vcpu *vcpu, u64 guest_deadline_tsc,
8000 			    bool *expired)
8001 {
8002 	struct vcpu_vmx *vmx;
8003 	u64 tscl, guest_tscl, delta_tsc, lapic_timer_advance_cycles;
8004 	struct kvm_timer *ktimer = &vcpu->arch.apic->lapic_timer;
8005 
8006 	vmx = to_vmx(vcpu);
8007 	tscl = rdtsc();
8008 	guest_tscl = kvm_read_l1_tsc(vcpu, tscl);
8009 	delta_tsc = max(guest_deadline_tsc, guest_tscl) - guest_tscl;
8010 	lapic_timer_advance_cycles = nsec_to_cycles(vcpu,
8011 						    ktimer->timer_advance_ns);
8012 
8013 	if (delta_tsc > lapic_timer_advance_cycles)
8014 		delta_tsc -= lapic_timer_advance_cycles;
8015 	else
8016 		delta_tsc = 0;
8017 
8018 	/* Convert to host delta tsc if tsc scaling is enabled */
8019 	if (vcpu->arch.l1_tsc_scaling_ratio != kvm_caps.default_tsc_scaling_ratio &&
8020 	    delta_tsc && u64_shl_div_u64(delta_tsc,
8021 				kvm_caps.tsc_scaling_ratio_frac_bits,
8022 				vcpu->arch.l1_tsc_scaling_ratio, &delta_tsc))
8023 		return -ERANGE;
8024 
8025 	/*
8026 	 * If the delta tsc can't fit in the 32 bit after the multi shift,
8027 	 * we can't use the preemption timer.
8028 	 * It's possible that it fits on later vmentries, but checking
8029 	 * on every vmentry is costly so we just use an hrtimer.
8030 	 */
8031 	if (delta_tsc >> (cpu_preemption_timer_multi + 32))
8032 		return -ERANGE;
8033 
8034 	vmx->hv_deadline_tsc = tscl + delta_tsc;
8035 	*expired = !delta_tsc;
8036 	return 0;
8037 }
8038 
8039 static void vmx_cancel_hv_timer(struct kvm_vcpu *vcpu)
8040 {
8041 	to_vmx(vcpu)->hv_deadline_tsc = -1;
8042 }
8043 #endif
8044 
8045 static void vmx_sched_in(struct kvm_vcpu *vcpu, int cpu)
8046 {
8047 	if (!kvm_pause_in_guest(vcpu->kvm))
8048 		shrink_ple_window(vcpu);
8049 }
8050 
8051 void vmx_update_cpu_dirty_logging(struct kvm_vcpu *vcpu)
8052 {
8053 	struct vcpu_vmx *vmx = to_vmx(vcpu);
8054 
8055 	if (WARN_ON_ONCE(!enable_pml))
8056 		return;
8057 
8058 	if (is_guest_mode(vcpu)) {
8059 		vmx->nested.update_vmcs01_cpu_dirty_logging = true;
8060 		return;
8061 	}
8062 
8063 	/*
8064 	 * Note, nr_memslots_dirty_logging can be changed concurrent with this
8065 	 * code, but in that case another update request will be made and so
8066 	 * the guest will never run with a stale PML value.
8067 	 */
8068 	if (atomic_read(&vcpu->kvm->nr_memslots_dirty_logging))
8069 		secondary_exec_controls_setbit(vmx, SECONDARY_EXEC_ENABLE_PML);
8070 	else
8071 		secondary_exec_controls_clearbit(vmx, SECONDARY_EXEC_ENABLE_PML);
8072 }
8073 
8074 static void vmx_setup_mce(struct kvm_vcpu *vcpu)
8075 {
8076 	if (vcpu->arch.mcg_cap & MCG_LMCE_P)
8077 		to_vmx(vcpu)->msr_ia32_feature_control_valid_bits |=
8078 			FEAT_CTL_LMCE_ENABLED;
8079 	else
8080 		to_vmx(vcpu)->msr_ia32_feature_control_valid_bits &=
8081 			~FEAT_CTL_LMCE_ENABLED;
8082 }
8083 
8084 #ifdef CONFIG_KVM_SMM
8085 static int vmx_smi_allowed(struct kvm_vcpu *vcpu, bool for_injection)
8086 {
8087 	/* we need a nested vmexit to enter SMM, postpone if run is pending */
8088 	if (to_vmx(vcpu)->nested.nested_run_pending)
8089 		return -EBUSY;
8090 	return !is_smm(vcpu);
8091 }
8092 
8093 static int vmx_enter_smm(struct kvm_vcpu *vcpu, union kvm_smram *smram)
8094 {
8095 	struct vcpu_vmx *vmx = to_vmx(vcpu);
8096 
8097 	/*
8098 	 * TODO: Implement custom flows for forcing the vCPU out/in of L2 on
8099 	 * SMI and RSM.  Using the common VM-Exit + VM-Enter routines is wrong
8100 	 * SMI and RSM only modify state that is saved and restored via SMRAM.
8101 	 * E.g. most MSRs are left untouched, but many are modified by VM-Exit
8102 	 * and VM-Enter, and thus L2's values may be corrupted on SMI+RSM.
8103 	 */
8104 	vmx->nested.smm.guest_mode = is_guest_mode(vcpu);
8105 	if (vmx->nested.smm.guest_mode)
8106 		nested_vmx_vmexit(vcpu, -1, 0, 0);
8107 
8108 	vmx->nested.smm.vmxon = vmx->nested.vmxon;
8109 	vmx->nested.vmxon = false;
8110 	vmx_clear_hlt(vcpu);
8111 	return 0;
8112 }
8113 
8114 static int vmx_leave_smm(struct kvm_vcpu *vcpu, const union kvm_smram *smram)
8115 {
8116 	struct vcpu_vmx *vmx = to_vmx(vcpu);
8117 	int ret;
8118 
8119 	if (vmx->nested.smm.vmxon) {
8120 		vmx->nested.vmxon = true;
8121 		vmx->nested.smm.vmxon = false;
8122 	}
8123 
8124 	if (vmx->nested.smm.guest_mode) {
8125 		ret = nested_vmx_enter_non_root_mode(vcpu, false);
8126 		if (ret)
8127 			return ret;
8128 
8129 		vmx->nested.nested_run_pending = 1;
8130 		vmx->nested.smm.guest_mode = false;
8131 	}
8132 	return 0;
8133 }
8134 
8135 static void vmx_enable_smi_window(struct kvm_vcpu *vcpu)
8136 {
8137 	/* RSM will cause a vmexit anyway.  */
8138 }
8139 #endif
8140 
8141 static bool vmx_apic_init_signal_blocked(struct kvm_vcpu *vcpu)
8142 {
8143 	return to_vmx(vcpu)->nested.vmxon && !is_guest_mode(vcpu);
8144 }
8145 
8146 static void vmx_migrate_timers(struct kvm_vcpu *vcpu)
8147 {
8148 	if (is_guest_mode(vcpu)) {
8149 		struct hrtimer *timer = &to_vmx(vcpu)->nested.preemption_timer;
8150 
8151 		if (hrtimer_try_to_cancel(timer) == 1)
8152 			hrtimer_start_expires(timer, HRTIMER_MODE_ABS_PINNED);
8153 	}
8154 }
8155 
8156 static void vmx_hardware_unsetup(void)
8157 {
8158 	kvm_set_posted_intr_wakeup_handler(NULL);
8159 
8160 	if (nested)
8161 		nested_vmx_hardware_unsetup();
8162 
8163 	free_kvm_area();
8164 }
8165 
8166 #define VMX_REQUIRED_APICV_INHIBITS			\
8167 (							\
8168 	BIT(APICV_INHIBIT_REASON_DISABLE)|		\
8169 	BIT(APICV_INHIBIT_REASON_ABSENT) |		\
8170 	BIT(APICV_INHIBIT_REASON_HYPERV) |		\
8171 	BIT(APICV_INHIBIT_REASON_BLOCKIRQ) |		\
8172 	BIT(APICV_INHIBIT_REASON_PHYSICAL_ID_ALIASED) |	\
8173 	BIT(APICV_INHIBIT_REASON_APIC_ID_MODIFIED) |	\
8174 	BIT(APICV_INHIBIT_REASON_APIC_BASE_MODIFIED)	\
8175 )
8176 
8177 static void vmx_vm_destroy(struct kvm *kvm)
8178 {
8179 	struct kvm_vmx *kvm_vmx = to_kvm_vmx(kvm);
8180 
8181 	free_pages((unsigned long)kvm_vmx->pid_table, vmx_get_pid_table_order(kvm));
8182 }
8183 
8184 static struct kvm_x86_ops vmx_x86_ops __initdata = {
8185 	.name = KBUILD_MODNAME,
8186 
8187 	.check_processor_compatibility = vmx_check_processor_compat,
8188 
8189 	.hardware_unsetup = vmx_hardware_unsetup,
8190 
8191 	.hardware_enable = vmx_hardware_enable,
8192 	.hardware_disable = vmx_hardware_disable,
8193 	.has_emulated_msr = vmx_has_emulated_msr,
8194 
8195 	.vm_size = sizeof(struct kvm_vmx),
8196 	.vm_init = vmx_vm_init,
8197 	.vm_destroy = vmx_vm_destroy,
8198 
8199 	.vcpu_precreate = vmx_vcpu_precreate,
8200 	.vcpu_create = vmx_vcpu_create,
8201 	.vcpu_free = vmx_vcpu_free,
8202 	.vcpu_reset = vmx_vcpu_reset,
8203 
8204 	.prepare_switch_to_guest = vmx_prepare_switch_to_guest,
8205 	.vcpu_load = vmx_vcpu_load,
8206 	.vcpu_put = vmx_vcpu_put,
8207 
8208 	.update_exception_bitmap = vmx_update_exception_bitmap,
8209 	.get_msr_feature = vmx_get_msr_feature,
8210 	.get_msr = vmx_get_msr,
8211 	.set_msr = vmx_set_msr,
8212 	.get_segment_base = vmx_get_segment_base,
8213 	.get_segment = vmx_get_segment,
8214 	.set_segment = vmx_set_segment,
8215 	.get_cpl = vmx_get_cpl,
8216 	.get_cs_db_l_bits = vmx_get_cs_db_l_bits,
8217 	.set_cr0 = vmx_set_cr0,
8218 	.is_valid_cr4 = vmx_is_valid_cr4,
8219 	.set_cr4 = vmx_set_cr4,
8220 	.set_efer = vmx_set_efer,
8221 	.get_idt = vmx_get_idt,
8222 	.set_idt = vmx_set_idt,
8223 	.get_gdt = vmx_get_gdt,
8224 	.set_gdt = vmx_set_gdt,
8225 	.set_dr7 = vmx_set_dr7,
8226 	.sync_dirty_debug_regs = vmx_sync_dirty_debug_regs,
8227 	.cache_reg = vmx_cache_reg,
8228 	.get_rflags = vmx_get_rflags,
8229 	.set_rflags = vmx_set_rflags,
8230 	.get_if_flag = vmx_get_if_flag,
8231 
8232 	.flush_tlb_all = vmx_flush_tlb_all,
8233 	.flush_tlb_current = vmx_flush_tlb_current,
8234 	.flush_tlb_gva = vmx_flush_tlb_gva,
8235 	.flush_tlb_guest = vmx_flush_tlb_guest,
8236 
8237 	.vcpu_pre_run = vmx_vcpu_pre_run,
8238 	.vcpu_run = vmx_vcpu_run,
8239 	.handle_exit = vmx_handle_exit,
8240 	.skip_emulated_instruction = vmx_skip_emulated_instruction,
8241 	.update_emulated_instruction = vmx_update_emulated_instruction,
8242 	.set_interrupt_shadow = vmx_set_interrupt_shadow,
8243 	.get_interrupt_shadow = vmx_get_interrupt_shadow,
8244 	.patch_hypercall = vmx_patch_hypercall,
8245 	.inject_irq = vmx_inject_irq,
8246 	.inject_nmi = vmx_inject_nmi,
8247 	.inject_exception = vmx_inject_exception,
8248 	.cancel_injection = vmx_cancel_injection,
8249 	.interrupt_allowed = vmx_interrupt_allowed,
8250 	.nmi_allowed = vmx_nmi_allowed,
8251 	.get_nmi_mask = vmx_get_nmi_mask,
8252 	.set_nmi_mask = vmx_set_nmi_mask,
8253 	.enable_nmi_window = vmx_enable_nmi_window,
8254 	.enable_irq_window = vmx_enable_irq_window,
8255 	.update_cr8_intercept = vmx_update_cr8_intercept,
8256 	.set_virtual_apic_mode = vmx_set_virtual_apic_mode,
8257 	.set_apic_access_page_addr = vmx_set_apic_access_page_addr,
8258 	.refresh_apicv_exec_ctrl = vmx_refresh_apicv_exec_ctrl,
8259 	.load_eoi_exitmap = vmx_load_eoi_exitmap,
8260 	.apicv_post_state_restore = vmx_apicv_post_state_restore,
8261 	.required_apicv_inhibits = VMX_REQUIRED_APICV_INHIBITS,
8262 	.hwapic_irr_update = vmx_hwapic_irr_update,
8263 	.hwapic_isr_update = vmx_hwapic_isr_update,
8264 	.guest_apic_has_interrupt = vmx_guest_apic_has_interrupt,
8265 	.sync_pir_to_irr = vmx_sync_pir_to_irr,
8266 	.deliver_interrupt = vmx_deliver_interrupt,
8267 	.dy_apicv_has_pending_interrupt = pi_has_pending_interrupt,
8268 
8269 	.set_tss_addr = vmx_set_tss_addr,
8270 	.set_identity_map_addr = vmx_set_identity_map_addr,
8271 	.get_mt_mask = vmx_get_mt_mask,
8272 
8273 	.get_exit_info = vmx_get_exit_info,
8274 
8275 	.vcpu_after_set_cpuid = vmx_vcpu_after_set_cpuid,
8276 
8277 	.has_wbinvd_exit = cpu_has_vmx_wbinvd_exit,
8278 
8279 	.get_l2_tsc_offset = vmx_get_l2_tsc_offset,
8280 	.get_l2_tsc_multiplier = vmx_get_l2_tsc_multiplier,
8281 	.write_tsc_offset = vmx_write_tsc_offset,
8282 	.write_tsc_multiplier = vmx_write_tsc_multiplier,
8283 
8284 	.load_mmu_pgd = vmx_load_mmu_pgd,
8285 
8286 	.check_intercept = vmx_check_intercept,
8287 	.handle_exit_irqoff = vmx_handle_exit_irqoff,
8288 
8289 	.request_immediate_exit = vmx_request_immediate_exit,
8290 
8291 	.sched_in = vmx_sched_in,
8292 
8293 	.cpu_dirty_log_size = PML_ENTITY_NUM,
8294 	.update_cpu_dirty_logging = vmx_update_cpu_dirty_logging,
8295 
8296 	.nested_ops = &vmx_nested_ops,
8297 
8298 	.pi_update_irte = vmx_pi_update_irte,
8299 	.pi_start_assignment = vmx_pi_start_assignment,
8300 
8301 #ifdef CONFIG_X86_64
8302 	.set_hv_timer = vmx_set_hv_timer,
8303 	.cancel_hv_timer = vmx_cancel_hv_timer,
8304 #endif
8305 
8306 	.setup_mce = vmx_setup_mce,
8307 
8308 #ifdef CONFIG_KVM_SMM
8309 	.smi_allowed = vmx_smi_allowed,
8310 	.enter_smm = vmx_enter_smm,
8311 	.leave_smm = vmx_leave_smm,
8312 	.enable_smi_window = vmx_enable_smi_window,
8313 #endif
8314 
8315 	.can_emulate_instruction = vmx_can_emulate_instruction,
8316 	.apic_init_signal_blocked = vmx_apic_init_signal_blocked,
8317 	.migrate_timers = vmx_migrate_timers,
8318 
8319 	.msr_filter_changed = vmx_msr_filter_changed,
8320 	.complete_emulated_msr = kvm_complete_insn_gp,
8321 
8322 	.vcpu_deliver_sipi_vector = kvm_vcpu_deliver_sipi_vector,
8323 };
8324 
8325 static unsigned int vmx_handle_intel_pt_intr(void)
8326 {
8327 	struct kvm_vcpu *vcpu = kvm_get_running_vcpu();
8328 
8329 	/* '0' on failure so that the !PT case can use a RET0 static call. */
8330 	if (!vcpu || !kvm_handling_nmi_from_guest(vcpu))
8331 		return 0;
8332 
8333 	kvm_make_request(KVM_REQ_PMI, vcpu);
8334 	__set_bit(MSR_CORE_PERF_GLOBAL_OVF_CTRL_TRACE_TOPA_PMI_BIT,
8335 		  (unsigned long *)&vcpu->arch.pmu.global_status);
8336 	return 1;
8337 }
8338 
8339 static __init void vmx_setup_user_return_msrs(void)
8340 {
8341 
8342 	/*
8343 	 * Though SYSCALL is only supported in 64-bit mode on Intel CPUs, kvm
8344 	 * will emulate SYSCALL in legacy mode if the vendor string in guest
8345 	 * CPUID.0:{EBX,ECX,EDX} is "AuthenticAMD" or "AMDisbetter!" To
8346 	 * support this emulation, MSR_STAR is included in the list for i386,
8347 	 * but is never loaded into hardware.  MSR_CSTAR is also never loaded
8348 	 * into hardware and is here purely for emulation purposes.
8349 	 */
8350 	const u32 vmx_uret_msrs_list[] = {
8351 	#ifdef CONFIG_X86_64
8352 		MSR_SYSCALL_MASK, MSR_LSTAR, MSR_CSTAR,
8353 	#endif
8354 		MSR_EFER, MSR_TSC_AUX, MSR_STAR,
8355 		MSR_IA32_TSX_CTRL,
8356 	};
8357 	int i;
8358 
8359 	BUILD_BUG_ON(ARRAY_SIZE(vmx_uret_msrs_list) != MAX_NR_USER_RETURN_MSRS);
8360 
8361 	for (i = 0; i < ARRAY_SIZE(vmx_uret_msrs_list); ++i)
8362 		kvm_add_user_return_msr(vmx_uret_msrs_list[i]);
8363 }
8364 
8365 static void __init vmx_setup_me_spte_mask(void)
8366 {
8367 	u64 me_mask = 0;
8368 
8369 	/*
8370 	 * kvm_get_shadow_phys_bits() returns shadow_phys_bits.  Use
8371 	 * the former to avoid exposing shadow_phys_bits.
8372 	 *
8373 	 * On pre-MKTME system, boot_cpu_data.x86_phys_bits equals to
8374 	 * shadow_phys_bits.  On MKTME and/or TDX capable systems,
8375 	 * boot_cpu_data.x86_phys_bits holds the actual physical address
8376 	 * w/o the KeyID bits, and shadow_phys_bits equals to MAXPHYADDR
8377 	 * reported by CPUID.  Those bits between are KeyID bits.
8378 	 */
8379 	if (boot_cpu_data.x86_phys_bits != kvm_get_shadow_phys_bits())
8380 		me_mask = rsvd_bits(boot_cpu_data.x86_phys_bits,
8381 			kvm_get_shadow_phys_bits() - 1);
8382 	/*
8383 	 * Unlike SME, host kernel doesn't support setting up any
8384 	 * MKTME KeyID on Intel platforms.  No memory encryption
8385 	 * bits should be included into the SPTE.
8386 	 */
8387 	kvm_mmu_set_me_spte_mask(0, me_mask);
8388 }
8389 
8390 static struct kvm_x86_init_ops vmx_init_ops __initdata;
8391 
8392 static __init int hardware_setup(void)
8393 {
8394 	unsigned long host_bndcfgs;
8395 	struct desc_ptr dt;
8396 	int r;
8397 
8398 	store_idt(&dt);
8399 	host_idt_base = dt.address;
8400 
8401 	vmx_setup_user_return_msrs();
8402 
8403 	if (setup_vmcs_config(&vmcs_config, &vmx_capability) < 0)
8404 		return -EIO;
8405 
8406 	if (cpu_has_perf_global_ctrl_bug())
8407 		pr_warn_once("VM_EXIT_LOAD_IA32_PERF_GLOBAL_CTRL "
8408 			     "does not work properly. Using workaround\n");
8409 
8410 	if (boot_cpu_has(X86_FEATURE_NX))
8411 		kvm_enable_efer_bits(EFER_NX);
8412 
8413 	if (boot_cpu_has(X86_FEATURE_MPX)) {
8414 		rdmsrl(MSR_IA32_BNDCFGS, host_bndcfgs);
8415 		WARN_ONCE(host_bndcfgs, "BNDCFGS in host will be lost");
8416 	}
8417 
8418 	if (!cpu_has_vmx_mpx())
8419 		kvm_caps.supported_xcr0 &= ~(XFEATURE_MASK_BNDREGS |
8420 					     XFEATURE_MASK_BNDCSR);
8421 
8422 	if (!cpu_has_vmx_vpid() || !cpu_has_vmx_invvpid() ||
8423 	    !(cpu_has_vmx_invvpid_single() || cpu_has_vmx_invvpid_global()))
8424 		enable_vpid = 0;
8425 
8426 	if (!cpu_has_vmx_ept() ||
8427 	    !cpu_has_vmx_ept_4levels() ||
8428 	    !cpu_has_vmx_ept_mt_wb() ||
8429 	    !cpu_has_vmx_invept_global())
8430 		enable_ept = 0;
8431 
8432 	/* NX support is required for shadow paging. */
8433 	if (!enable_ept && !boot_cpu_has(X86_FEATURE_NX)) {
8434 		pr_err_ratelimited("NX (Execute Disable) not supported\n");
8435 		return -EOPNOTSUPP;
8436 	}
8437 
8438 	if (!cpu_has_vmx_ept_ad_bits() || !enable_ept)
8439 		enable_ept_ad_bits = 0;
8440 
8441 	if (!cpu_has_vmx_unrestricted_guest() || !enable_ept)
8442 		enable_unrestricted_guest = 0;
8443 
8444 	if (!cpu_has_vmx_flexpriority())
8445 		flexpriority_enabled = 0;
8446 
8447 	if (!cpu_has_virtual_nmis())
8448 		enable_vnmi = 0;
8449 
8450 #ifdef CONFIG_X86_SGX_KVM
8451 	if (!cpu_has_vmx_encls_vmexit())
8452 		enable_sgx = false;
8453 #endif
8454 
8455 	/*
8456 	 * set_apic_access_page_addr() is used to reload apic access
8457 	 * page upon invalidation.  No need to do anything if not
8458 	 * using the APIC_ACCESS_ADDR VMCS field.
8459 	 */
8460 	if (!flexpriority_enabled)
8461 		vmx_x86_ops.set_apic_access_page_addr = NULL;
8462 
8463 	if (!cpu_has_vmx_tpr_shadow())
8464 		vmx_x86_ops.update_cr8_intercept = NULL;
8465 
8466 #if IS_ENABLED(CONFIG_HYPERV)
8467 	if (ms_hyperv.nested_features & HV_X64_NESTED_GUEST_MAPPING_FLUSH
8468 	    && enable_ept) {
8469 		vmx_x86_ops.flush_remote_tlbs = hv_flush_remote_tlbs;
8470 		vmx_x86_ops.flush_remote_tlbs_range = hv_flush_remote_tlbs_range;
8471 	}
8472 #endif
8473 
8474 	if (!cpu_has_vmx_ple()) {
8475 		ple_gap = 0;
8476 		ple_window = 0;
8477 		ple_window_grow = 0;
8478 		ple_window_max = 0;
8479 		ple_window_shrink = 0;
8480 	}
8481 
8482 	if (!cpu_has_vmx_apicv())
8483 		enable_apicv = 0;
8484 	if (!enable_apicv)
8485 		vmx_x86_ops.sync_pir_to_irr = NULL;
8486 
8487 	if (!enable_apicv || !cpu_has_vmx_ipiv())
8488 		enable_ipiv = false;
8489 
8490 	if (cpu_has_vmx_tsc_scaling())
8491 		kvm_caps.has_tsc_control = true;
8492 
8493 	kvm_caps.max_tsc_scaling_ratio = KVM_VMX_TSC_MULTIPLIER_MAX;
8494 	kvm_caps.tsc_scaling_ratio_frac_bits = 48;
8495 	kvm_caps.has_bus_lock_exit = cpu_has_vmx_bus_lock_detection();
8496 	kvm_caps.has_notify_vmexit = cpu_has_notify_vmexit();
8497 
8498 	set_bit(0, vmx_vpid_bitmap); /* 0 is reserved for host */
8499 
8500 	if (enable_ept)
8501 		kvm_mmu_set_ept_masks(enable_ept_ad_bits,
8502 				      cpu_has_vmx_ept_execute_only());
8503 
8504 	/*
8505 	 * Setup shadow_me_value/shadow_me_mask to include MKTME KeyID
8506 	 * bits to shadow_zero_check.
8507 	 */
8508 	vmx_setup_me_spte_mask();
8509 
8510 	kvm_configure_mmu(enable_ept, 0, vmx_get_max_tdp_level(),
8511 			  ept_caps_to_lpage_level(vmx_capability.ept));
8512 
8513 	/*
8514 	 * Only enable PML when hardware supports PML feature, and both EPT
8515 	 * and EPT A/D bit features are enabled -- PML depends on them to work.
8516 	 */
8517 	if (!enable_ept || !enable_ept_ad_bits || !cpu_has_vmx_pml())
8518 		enable_pml = 0;
8519 
8520 	if (!enable_pml)
8521 		vmx_x86_ops.cpu_dirty_log_size = 0;
8522 
8523 	if (!cpu_has_vmx_preemption_timer())
8524 		enable_preemption_timer = false;
8525 
8526 	if (enable_preemption_timer) {
8527 		u64 use_timer_freq = 5000ULL * 1000 * 1000;
8528 
8529 		cpu_preemption_timer_multi =
8530 			vmcs_config.misc & VMX_MISC_PREEMPTION_TIMER_RATE_MASK;
8531 
8532 		if (tsc_khz)
8533 			use_timer_freq = (u64)tsc_khz * 1000;
8534 		use_timer_freq >>= cpu_preemption_timer_multi;
8535 
8536 		/*
8537 		 * KVM "disables" the preemption timer by setting it to its max
8538 		 * value.  Don't use the timer if it might cause spurious exits
8539 		 * at a rate faster than 0.1 Hz (of uninterrupted guest time).
8540 		 */
8541 		if (use_timer_freq > 0xffffffffu / 10)
8542 			enable_preemption_timer = false;
8543 	}
8544 
8545 	if (!enable_preemption_timer) {
8546 		vmx_x86_ops.set_hv_timer = NULL;
8547 		vmx_x86_ops.cancel_hv_timer = NULL;
8548 		vmx_x86_ops.request_immediate_exit = __kvm_request_immediate_exit;
8549 	}
8550 
8551 	kvm_caps.supported_mce_cap |= MCG_LMCE_P;
8552 	kvm_caps.supported_mce_cap |= MCG_CMCI_P;
8553 
8554 	if (pt_mode != PT_MODE_SYSTEM && pt_mode != PT_MODE_HOST_GUEST)
8555 		return -EINVAL;
8556 	if (!enable_ept || !enable_pmu || !cpu_has_vmx_intel_pt())
8557 		pt_mode = PT_MODE_SYSTEM;
8558 	if (pt_mode == PT_MODE_HOST_GUEST)
8559 		vmx_init_ops.handle_intel_pt_intr = vmx_handle_intel_pt_intr;
8560 	else
8561 		vmx_init_ops.handle_intel_pt_intr = NULL;
8562 
8563 	setup_default_sgx_lepubkeyhash();
8564 
8565 	if (nested) {
8566 		nested_vmx_setup_ctls_msrs(&vmcs_config, vmx_capability.ept);
8567 
8568 		r = nested_vmx_hardware_setup(kvm_vmx_exit_handlers);
8569 		if (r)
8570 			return r;
8571 	}
8572 
8573 	vmx_set_cpu_caps();
8574 
8575 	r = alloc_kvm_area();
8576 	if (r && nested)
8577 		nested_vmx_hardware_unsetup();
8578 
8579 	kvm_set_posted_intr_wakeup_handler(pi_wakeup_handler);
8580 
8581 	return r;
8582 }
8583 
8584 static struct kvm_x86_init_ops vmx_init_ops __initdata = {
8585 	.hardware_setup = hardware_setup,
8586 	.handle_intel_pt_intr = NULL,
8587 
8588 	.runtime_ops = &vmx_x86_ops,
8589 	.pmu_ops = &intel_pmu_ops,
8590 };
8591 
8592 static void vmx_cleanup_l1d_flush(void)
8593 {
8594 	if (vmx_l1d_flush_pages) {
8595 		free_pages((unsigned long)vmx_l1d_flush_pages, L1D_CACHE_ORDER);
8596 		vmx_l1d_flush_pages = NULL;
8597 	}
8598 	/* Restore state so sysfs ignores VMX */
8599 	l1tf_vmx_mitigation = VMENTER_L1D_FLUSH_AUTO;
8600 }
8601 
8602 static void __vmx_exit(void)
8603 {
8604 	allow_smaller_maxphyaddr = false;
8605 
8606 #ifdef CONFIG_KEXEC_CORE
8607 	RCU_INIT_POINTER(crash_vmclear_loaded_vmcss, NULL);
8608 	synchronize_rcu();
8609 #endif
8610 	vmx_cleanup_l1d_flush();
8611 }
8612 
8613 static void vmx_exit(void)
8614 {
8615 	kvm_exit();
8616 	kvm_x86_vendor_exit();
8617 
8618 	__vmx_exit();
8619 }
8620 module_exit(vmx_exit);
8621 
8622 static int __init vmx_init(void)
8623 {
8624 	int r, cpu;
8625 
8626 	if (!kvm_is_vmx_supported())
8627 		return -EOPNOTSUPP;
8628 
8629 	/*
8630 	 * Note, hv_init_evmcs() touches only VMX knobs, i.e. there's nothing
8631 	 * to unwind if a later step fails.
8632 	 */
8633 	hv_init_evmcs();
8634 
8635 	r = kvm_x86_vendor_init(&vmx_init_ops);
8636 	if (r)
8637 		return r;
8638 
8639 	/*
8640 	 * Must be called after common x86 init so enable_ept is properly set
8641 	 * up. Hand the parameter mitigation value in which was stored in
8642 	 * the pre module init parser. If no parameter was given, it will
8643 	 * contain 'auto' which will be turned into the default 'cond'
8644 	 * mitigation mode.
8645 	 */
8646 	r = vmx_setup_l1d_flush(vmentry_l1d_flush_param);
8647 	if (r)
8648 		goto err_l1d_flush;
8649 
8650 	vmx_setup_fb_clear_ctrl();
8651 
8652 	for_each_possible_cpu(cpu) {
8653 		INIT_LIST_HEAD(&per_cpu(loaded_vmcss_on_cpu, cpu));
8654 
8655 		pi_init_cpu(cpu);
8656 	}
8657 
8658 #ifdef CONFIG_KEXEC_CORE
8659 	rcu_assign_pointer(crash_vmclear_loaded_vmcss,
8660 			   crash_vmclear_local_loaded_vmcss);
8661 #endif
8662 	vmx_check_vmcs12_offsets();
8663 
8664 	/*
8665 	 * Shadow paging doesn't have a (further) performance penalty
8666 	 * from GUEST_MAXPHYADDR < HOST_MAXPHYADDR so enable it
8667 	 * by default
8668 	 */
8669 	if (!enable_ept)
8670 		allow_smaller_maxphyaddr = true;
8671 
8672 	/*
8673 	 * Common KVM initialization _must_ come last, after this, /dev/kvm is
8674 	 * exposed to userspace!
8675 	 */
8676 	r = kvm_init(sizeof(struct vcpu_vmx), __alignof__(struct vcpu_vmx),
8677 		     THIS_MODULE);
8678 	if (r)
8679 		goto err_kvm_init;
8680 
8681 	return 0;
8682 
8683 err_kvm_init:
8684 	__vmx_exit();
8685 err_l1d_flush:
8686 	kvm_x86_vendor_exit();
8687 	return r;
8688 }
8689 module_init(vmx_init);
8690