xref: /openbmc/linux/arch/x86/kvm/vmx/vmx.c (revision cd99b9eb)
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 	/*
1517 	 * Unlike CR0 and CR4, RFLAGS handling requires checking if the vCPU
1518 	 * is an unrestricted guest in order to mark L2 as needing emulation
1519 	 * if L1 runs L2 as a restricted guest.
1520 	 */
1521 	if (is_unrestricted_guest(vcpu)) {
1522 		kvm_register_mark_available(vcpu, VCPU_EXREG_RFLAGS);
1523 		vmx->rflags = rflags;
1524 		vmcs_writel(GUEST_RFLAGS, rflags);
1525 		return;
1526 	}
1527 
1528 	old_rflags = vmx_get_rflags(vcpu);
1529 	vmx->rflags = rflags;
1530 	if (vmx->rmode.vm86_active) {
1531 		vmx->rmode.save_rflags = rflags;
1532 		rflags |= X86_EFLAGS_IOPL | X86_EFLAGS_VM;
1533 	}
1534 	vmcs_writel(GUEST_RFLAGS, rflags);
1535 
1536 	if ((old_rflags ^ vmx->rflags) & X86_EFLAGS_VM)
1537 		vmx->emulation_required = vmx_emulation_required(vcpu);
1538 }
1539 
1540 static bool vmx_get_if_flag(struct kvm_vcpu *vcpu)
1541 {
1542 	return vmx_get_rflags(vcpu) & X86_EFLAGS_IF;
1543 }
1544 
1545 u32 vmx_get_interrupt_shadow(struct kvm_vcpu *vcpu)
1546 {
1547 	u32 interruptibility = vmcs_read32(GUEST_INTERRUPTIBILITY_INFO);
1548 	int ret = 0;
1549 
1550 	if (interruptibility & GUEST_INTR_STATE_STI)
1551 		ret |= KVM_X86_SHADOW_INT_STI;
1552 	if (interruptibility & GUEST_INTR_STATE_MOV_SS)
1553 		ret |= KVM_X86_SHADOW_INT_MOV_SS;
1554 
1555 	return ret;
1556 }
1557 
1558 void vmx_set_interrupt_shadow(struct kvm_vcpu *vcpu, int mask)
1559 {
1560 	u32 interruptibility_old = vmcs_read32(GUEST_INTERRUPTIBILITY_INFO);
1561 	u32 interruptibility = interruptibility_old;
1562 
1563 	interruptibility &= ~(GUEST_INTR_STATE_STI | GUEST_INTR_STATE_MOV_SS);
1564 
1565 	if (mask & KVM_X86_SHADOW_INT_MOV_SS)
1566 		interruptibility |= GUEST_INTR_STATE_MOV_SS;
1567 	else if (mask & KVM_X86_SHADOW_INT_STI)
1568 		interruptibility |= GUEST_INTR_STATE_STI;
1569 
1570 	if ((interruptibility != interruptibility_old))
1571 		vmcs_write32(GUEST_INTERRUPTIBILITY_INFO, interruptibility);
1572 }
1573 
1574 static int vmx_rtit_ctl_check(struct kvm_vcpu *vcpu, u64 data)
1575 {
1576 	struct vcpu_vmx *vmx = to_vmx(vcpu);
1577 	unsigned long value;
1578 
1579 	/*
1580 	 * Any MSR write that attempts to change bits marked reserved will
1581 	 * case a #GP fault.
1582 	 */
1583 	if (data & vmx->pt_desc.ctl_bitmask)
1584 		return 1;
1585 
1586 	/*
1587 	 * Any attempt to modify IA32_RTIT_CTL while TraceEn is set will
1588 	 * result in a #GP unless the same write also clears TraceEn.
1589 	 */
1590 	if ((vmx->pt_desc.guest.ctl & RTIT_CTL_TRACEEN) &&
1591 		((vmx->pt_desc.guest.ctl ^ data) & ~RTIT_CTL_TRACEEN))
1592 		return 1;
1593 
1594 	/*
1595 	 * WRMSR to IA32_RTIT_CTL that sets TraceEn but clears this bit
1596 	 * and FabricEn would cause #GP, if
1597 	 * CPUID.(EAX=14H, ECX=0):ECX.SNGLRGNOUT[bit 2] = 0
1598 	 */
1599 	if ((data & RTIT_CTL_TRACEEN) && !(data & RTIT_CTL_TOPA) &&
1600 		!(data & RTIT_CTL_FABRIC_EN) &&
1601 		!intel_pt_validate_cap(vmx->pt_desc.caps,
1602 					PT_CAP_single_range_output))
1603 		return 1;
1604 
1605 	/*
1606 	 * MTCFreq, CycThresh and PSBFreq encodings check, any MSR write that
1607 	 * utilize encodings marked reserved will cause a #GP fault.
1608 	 */
1609 	value = intel_pt_validate_cap(vmx->pt_desc.caps, PT_CAP_mtc_periods);
1610 	if (intel_pt_validate_cap(vmx->pt_desc.caps, PT_CAP_mtc) &&
1611 			!test_bit((data & RTIT_CTL_MTC_RANGE) >>
1612 			RTIT_CTL_MTC_RANGE_OFFSET, &value))
1613 		return 1;
1614 	value = intel_pt_validate_cap(vmx->pt_desc.caps,
1615 						PT_CAP_cycle_thresholds);
1616 	if (intel_pt_validate_cap(vmx->pt_desc.caps, PT_CAP_psb_cyc) &&
1617 			!test_bit((data & RTIT_CTL_CYC_THRESH) >>
1618 			RTIT_CTL_CYC_THRESH_OFFSET, &value))
1619 		return 1;
1620 	value = intel_pt_validate_cap(vmx->pt_desc.caps, PT_CAP_psb_periods);
1621 	if (intel_pt_validate_cap(vmx->pt_desc.caps, PT_CAP_psb_cyc) &&
1622 			!test_bit((data & RTIT_CTL_PSB_FREQ) >>
1623 			RTIT_CTL_PSB_FREQ_OFFSET, &value))
1624 		return 1;
1625 
1626 	/*
1627 	 * If ADDRx_CFG is reserved or the encodings is >2 will
1628 	 * cause a #GP fault.
1629 	 */
1630 	value = (data & RTIT_CTL_ADDR0) >> RTIT_CTL_ADDR0_OFFSET;
1631 	if ((value && (vmx->pt_desc.num_address_ranges < 1)) || (value > 2))
1632 		return 1;
1633 	value = (data & RTIT_CTL_ADDR1) >> RTIT_CTL_ADDR1_OFFSET;
1634 	if ((value && (vmx->pt_desc.num_address_ranges < 2)) || (value > 2))
1635 		return 1;
1636 	value = (data & RTIT_CTL_ADDR2) >> RTIT_CTL_ADDR2_OFFSET;
1637 	if ((value && (vmx->pt_desc.num_address_ranges < 3)) || (value > 2))
1638 		return 1;
1639 	value = (data & RTIT_CTL_ADDR3) >> RTIT_CTL_ADDR3_OFFSET;
1640 	if ((value && (vmx->pt_desc.num_address_ranges < 4)) || (value > 2))
1641 		return 1;
1642 
1643 	return 0;
1644 }
1645 
1646 static bool vmx_can_emulate_instruction(struct kvm_vcpu *vcpu, int emul_type,
1647 					void *insn, int insn_len)
1648 {
1649 	/*
1650 	 * Emulation of instructions in SGX enclaves is impossible as RIP does
1651 	 * not point at the failing instruction, and even if it did, the code
1652 	 * stream is inaccessible.  Inject #UD instead of exiting to userspace
1653 	 * so that guest userspace can't DoS the guest simply by triggering
1654 	 * emulation (enclaves are CPL3 only).
1655 	 */
1656 	if (to_vmx(vcpu)->exit_reason.enclave_mode) {
1657 		kvm_queue_exception(vcpu, UD_VECTOR);
1658 		return false;
1659 	}
1660 	return true;
1661 }
1662 
1663 static int skip_emulated_instruction(struct kvm_vcpu *vcpu)
1664 {
1665 	union vmx_exit_reason exit_reason = to_vmx(vcpu)->exit_reason;
1666 	unsigned long rip, orig_rip;
1667 	u32 instr_len;
1668 
1669 	/*
1670 	 * Using VMCS.VM_EXIT_INSTRUCTION_LEN on EPT misconfig depends on
1671 	 * undefined behavior: Intel's SDM doesn't mandate the VMCS field be
1672 	 * set when EPT misconfig occurs.  In practice, real hardware updates
1673 	 * VM_EXIT_INSTRUCTION_LEN on EPT misconfig, but other hypervisors
1674 	 * (namely Hyper-V) don't set it due to it being undefined behavior,
1675 	 * i.e. we end up advancing IP with some random value.
1676 	 */
1677 	if (!static_cpu_has(X86_FEATURE_HYPERVISOR) ||
1678 	    exit_reason.basic != EXIT_REASON_EPT_MISCONFIG) {
1679 		instr_len = vmcs_read32(VM_EXIT_INSTRUCTION_LEN);
1680 
1681 		/*
1682 		 * Emulating an enclave's instructions isn't supported as KVM
1683 		 * cannot access the enclave's memory or its true RIP, e.g. the
1684 		 * vmcs.GUEST_RIP points at the exit point of the enclave, not
1685 		 * the RIP that actually triggered the VM-Exit.  But, because
1686 		 * most instructions that cause VM-Exit will #UD in an enclave,
1687 		 * most instruction-based VM-Exits simply do not occur.
1688 		 *
1689 		 * There are a few exceptions, notably the debug instructions
1690 		 * INT1ICEBRK and INT3, as they are allowed in debug enclaves
1691 		 * and generate #DB/#BP as expected, which KVM might intercept.
1692 		 * But again, the CPU does the dirty work and saves an instr
1693 		 * length of zero so VMMs don't shoot themselves in the foot.
1694 		 * WARN if KVM tries to skip a non-zero length instruction on
1695 		 * a VM-Exit from an enclave.
1696 		 */
1697 		if (!instr_len)
1698 			goto rip_updated;
1699 
1700 		WARN_ONCE(exit_reason.enclave_mode,
1701 			  "skipping instruction after SGX enclave VM-Exit");
1702 
1703 		orig_rip = kvm_rip_read(vcpu);
1704 		rip = orig_rip + instr_len;
1705 #ifdef CONFIG_X86_64
1706 		/*
1707 		 * We need to mask out the high 32 bits of RIP if not in 64-bit
1708 		 * mode, but just finding out that we are in 64-bit mode is
1709 		 * quite expensive.  Only do it if there was a carry.
1710 		 */
1711 		if (unlikely(((rip ^ orig_rip) >> 31) == 3) && !is_64_bit_mode(vcpu))
1712 			rip = (u32)rip;
1713 #endif
1714 		kvm_rip_write(vcpu, rip);
1715 	} else {
1716 		if (!kvm_emulate_instruction(vcpu, EMULTYPE_SKIP))
1717 			return 0;
1718 	}
1719 
1720 rip_updated:
1721 	/* skipping an emulated instruction also counts */
1722 	vmx_set_interrupt_shadow(vcpu, 0);
1723 
1724 	return 1;
1725 }
1726 
1727 /*
1728  * Recognizes a pending MTF VM-exit and records the nested state for later
1729  * delivery.
1730  */
1731 static void vmx_update_emulated_instruction(struct kvm_vcpu *vcpu)
1732 {
1733 	struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
1734 	struct vcpu_vmx *vmx = to_vmx(vcpu);
1735 
1736 	if (!is_guest_mode(vcpu))
1737 		return;
1738 
1739 	/*
1740 	 * Per the SDM, MTF takes priority over debug-trap exceptions besides
1741 	 * TSS T-bit traps and ICEBP (INT1).  KVM doesn't emulate T-bit traps
1742 	 * or ICEBP (in the emulator proper), and skipping of ICEBP after an
1743 	 * intercepted #DB deliberately avoids single-step #DB and MTF updates
1744 	 * as ICEBP is higher priority than both.  As instruction emulation is
1745 	 * completed at this point (i.e. KVM is at the instruction boundary),
1746 	 * any #DB exception pending delivery must be a debug-trap of lower
1747 	 * priority than MTF.  Record the pending MTF state to be delivered in
1748 	 * vmx_check_nested_events().
1749 	 */
1750 	if (nested_cpu_has_mtf(vmcs12) &&
1751 	    (!vcpu->arch.exception.pending ||
1752 	     vcpu->arch.exception.vector == DB_VECTOR) &&
1753 	    (!vcpu->arch.exception_vmexit.pending ||
1754 	     vcpu->arch.exception_vmexit.vector == DB_VECTOR)) {
1755 		vmx->nested.mtf_pending = true;
1756 		kvm_make_request(KVM_REQ_EVENT, vcpu);
1757 	} else {
1758 		vmx->nested.mtf_pending = false;
1759 	}
1760 }
1761 
1762 static int vmx_skip_emulated_instruction(struct kvm_vcpu *vcpu)
1763 {
1764 	vmx_update_emulated_instruction(vcpu);
1765 	return skip_emulated_instruction(vcpu);
1766 }
1767 
1768 static void vmx_clear_hlt(struct kvm_vcpu *vcpu)
1769 {
1770 	/*
1771 	 * Ensure that we clear the HLT state in the VMCS.  We don't need to
1772 	 * explicitly skip the instruction because if the HLT state is set,
1773 	 * then the instruction is already executing and RIP has already been
1774 	 * advanced.
1775 	 */
1776 	if (kvm_hlt_in_guest(vcpu->kvm) &&
1777 			vmcs_read32(GUEST_ACTIVITY_STATE) == GUEST_ACTIVITY_HLT)
1778 		vmcs_write32(GUEST_ACTIVITY_STATE, GUEST_ACTIVITY_ACTIVE);
1779 }
1780 
1781 static void vmx_inject_exception(struct kvm_vcpu *vcpu)
1782 {
1783 	struct kvm_queued_exception *ex = &vcpu->arch.exception;
1784 	u32 intr_info = ex->vector | INTR_INFO_VALID_MASK;
1785 	struct vcpu_vmx *vmx = to_vmx(vcpu);
1786 
1787 	kvm_deliver_exception_payload(vcpu, ex);
1788 
1789 	if (ex->has_error_code) {
1790 		/*
1791 		 * Despite the error code being architecturally defined as 32
1792 		 * bits, and the VMCS field being 32 bits, Intel CPUs and thus
1793 		 * VMX don't actually supporting setting bits 31:16.  Hardware
1794 		 * will (should) never provide a bogus error code, but AMD CPUs
1795 		 * do generate error codes with bits 31:16 set, and so KVM's
1796 		 * ABI lets userspace shove in arbitrary 32-bit values.  Drop
1797 		 * the upper bits to avoid VM-Fail, losing information that
1798 		 * does't really exist is preferable to killing the VM.
1799 		 */
1800 		vmcs_write32(VM_ENTRY_EXCEPTION_ERROR_CODE, (u16)ex->error_code);
1801 		intr_info |= INTR_INFO_DELIVER_CODE_MASK;
1802 	}
1803 
1804 	if (vmx->rmode.vm86_active) {
1805 		int inc_eip = 0;
1806 		if (kvm_exception_is_soft(ex->vector))
1807 			inc_eip = vcpu->arch.event_exit_inst_len;
1808 		kvm_inject_realmode_interrupt(vcpu, ex->vector, inc_eip);
1809 		return;
1810 	}
1811 
1812 	WARN_ON_ONCE(vmx->emulation_required);
1813 
1814 	if (kvm_exception_is_soft(ex->vector)) {
1815 		vmcs_write32(VM_ENTRY_INSTRUCTION_LEN,
1816 			     vmx->vcpu.arch.event_exit_inst_len);
1817 		intr_info |= INTR_TYPE_SOFT_EXCEPTION;
1818 	} else
1819 		intr_info |= INTR_TYPE_HARD_EXCEPTION;
1820 
1821 	vmcs_write32(VM_ENTRY_INTR_INFO_FIELD, intr_info);
1822 
1823 	vmx_clear_hlt(vcpu);
1824 }
1825 
1826 static void vmx_setup_uret_msr(struct vcpu_vmx *vmx, unsigned int msr,
1827 			       bool load_into_hardware)
1828 {
1829 	struct vmx_uret_msr *uret_msr;
1830 
1831 	uret_msr = vmx_find_uret_msr(vmx, msr);
1832 	if (!uret_msr)
1833 		return;
1834 
1835 	uret_msr->load_into_hardware = load_into_hardware;
1836 }
1837 
1838 /*
1839  * Configuring user return MSRs to automatically save, load, and restore MSRs
1840  * that need to be shoved into hardware when running the guest.  Note, omitting
1841  * an MSR here does _NOT_ mean it's not emulated, only that it will not be
1842  * loaded into hardware when running the guest.
1843  */
1844 static void vmx_setup_uret_msrs(struct vcpu_vmx *vmx)
1845 {
1846 #ifdef CONFIG_X86_64
1847 	bool load_syscall_msrs;
1848 
1849 	/*
1850 	 * The SYSCALL MSRs are only needed on long mode guests, and only
1851 	 * when EFER.SCE is set.
1852 	 */
1853 	load_syscall_msrs = is_long_mode(&vmx->vcpu) &&
1854 			    (vmx->vcpu.arch.efer & EFER_SCE);
1855 
1856 	vmx_setup_uret_msr(vmx, MSR_STAR, load_syscall_msrs);
1857 	vmx_setup_uret_msr(vmx, MSR_LSTAR, load_syscall_msrs);
1858 	vmx_setup_uret_msr(vmx, MSR_SYSCALL_MASK, load_syscall_msrs);
1859 #endif
1860 	vmx_setup_uret_msr(vmx, MSR_EFER, update_transition_efer(vmx));
1861 
1862 	vmx_setup_uret_msr(vmx, MSR_TSC_AUX,
1863 			   guest_cpuid_has(&vmx->vcpu, X86_FEATURE_RDTSCP) ||
1864 			   guest_cpuid_has(&vmx->vcpu, X86_FEATURE_RDPID));
1865 
1866 	/*
1867 	 * hle=0, rtm=0, tsx_ctrl=1 can be found with some combinations of new
1868 	 * kernel and old userspace.  If those guests run on a tsx=off host, do
1869 	 * allow guests to use TSX_CTRL, but don't change the value in hardware
1870 	 * so that TSX remains always disabled.
1871 	 */
1872 	vmx_setup_uret_msr(vmx, MSR_IA32_TSX_CTRL, boot_cpu_has(X86_FEATURE_RTM));
1873 
1874 	/*
1875 	 * The set of MSRs to load may have changed, reload MSRs before the
1876 	 * next VM-Enter.
1877 	 */
1878 	vmx->guest_uret_msrs_loaded = false;
1879 }
1880 
1881 u64 vmx_get_l2_tsc_offset(struct kvm_vcpu *vcpu)
1882 {
1883 	struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
1884 
1885 	if (nested_cpu_has(vmcs12, CPU_BASED_USE_TSC_OFFSETTING))
1886 		return vmcs12->tsc_offset;
1887 
1888 	return 0;
1889 }
1890 
1891 u64 vmx_get_l2_tsc_multiplier(struct kvm_vcpu *vcpu)
1892 {
1893 	struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
1894 
1895 	if (nested_cpu_has(vmcs12, CPU_BASED_USE_TSC_OFFSETTING) &&
1896 	    nested_cpu_has2(vmcs12, SECONDARY_EXEC_TSC_SCALING))
1897 		return vmcs12->tsc_multiplier;
1898 
1899 	return kvm_caps.default_tsc_scaling_ratio;
1900 }
1901 
1902 static void vmx_write_tsc_offset(struct kvm_vcpu *vcpu, u64 offset)
1903 {
1904 	vmcs_write64(TSC_OFFSET, offset);
1905 }
1906 
1907 static void vmx_write_tsc_multiplier(struct kvm_vcpu *vcpu, u64 multiplier)
1908 {
1909 	vmcs_write64(TSC_MULTIPLIER, multiplier);
1910 }
1911 
1912 /*
1913  * nested_vmx_allowed() checks whether a guest should be allowed to use VMX
1914  * instructions and MSRs (i.e., nested VMX). Nested VMX is disabled for
1915  * all guests if the "nested" module option is off, and can also be disabled
1916  * for a single guest by disabling its VMX cpuid bit.
1917  */
1918 bool nested_vmx_allowed(struct kvm_vcpu *vcpu)
1919 {
1920 	return nested && guest_cpuid_has(vcpu, X86_FEATURE_VMX);
1921 }
1922 
1923 /*
1924  * Userspace is allowed to set any supported IA32_FEATURE_CONTROL regardless of
1925  * guest CPUID.  Note, KVM allows userspace to set "VMX in SMX" to maintain
1926  * backwards compatibility even though KVM doesn't support emulating SMX.  And
1927  * because userspace set "VMX in SMX", the guest must also be allowed to set it,
1928  * e.g. if the MSR is left unlocked and the guest does a RMW operation.
1929  */
1930 #define KVM_SUPPORTED_FEATURE_CONTROL  (FEAT_CTL_LOCKED			 | \
1931 					FEAT_CTL_VMX_ENABLED_INSIDE_SMX	 | \
1932 					FEAT_CTL_VMX_ENABLED_OUTSIDE_SMX | \
1933 					FEAT_CTL_SGX_LC_ENABLED		 | \
1934 					FEAT_CTL_SGX_ENABLED		 | \
1935 					FEAT_CTL_LMCE_ENABLED)
1936 
1937 static inline bool is_vmx_feature_control_msr_valid(struct vcpu_vmx *vmx,
1938 						    struct msr_data *msr)
1939 {
1940 	uint64_t valid_bits;
1941 
1942 	/*
1943 	 * Ensure KVM_SUPPORTED_FEATURE_CONTROL is updated when new bits are
1944 	 * exposed to the guest.
1945 	 */
1946 	WARN_ON_ONCE(vmx->msr_ia32_feature_control_valid_bits &
1947 		     ~KVM_SUPPORTED_FEATURE_CONTROL);
1948 
1949 	if (!msr->host_initiated &&
1950 	    (vmx->msr_ia32_feature_control & FEAT_CTL_LOCKED))
1951 		return false;
1952 
1953 	if (msr->host_initiated)
1954 		valid_bits = KVM_SUPPORTED_FEATURE_CONTROL;
1955 	else
1956 		valid_bits = vmx->msr_ia32_feature_control_valid_bits;
1957 
1958 	return !(msr->data & ~valid_bits);
1959 }
1960 
1961 static int vmx_get_msr_feature(struct kvm_msr_entry *msr)
1962 {
1963 	switch (msr->index) {
1964 	case KVM_FIRST_EMULATED_VMX_MSR ... KVM_LAST_EMULATED_VMX_MSR:
1965 		if (!nested)
1966 			return 1;
1967 		return vmx_get_vmx_msr(&vmcs_config.nested, msr->index, &msr->data);
1968 	default:
1969 		return KVM_MSR_RET_INVALID;
1970 	}
1971 }
1972 
1973 /*
1974  * Reads an msr value (of 'msr_info->index') into 'msr_info->data'.
1975  * Returns 0 on success, non-0 otherwise.
1976  * Assumes vcpu_load() was already called.
1977  */
1978 static int vmx_get_msr(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
1979 {
1980 	struct vcpu_vmx *vmx = to_vmx(vcpu);
1981 	struct vmx_uret_msr *msr;
1982 	u32 index;
1983 
1984 	switch (msr_info->index) {
1985 #ifdef CONFIG_X86_64
1986 	case MSR_FS_BASE:
1987 		msr_info->data = vmcs_readl(GUEST_FS_BASE);
1988 		break;
1989 	case MSR_GS_BASE:
1990 		msr_info->data = vmcs_readl(GUEST_GS_BASE);
1991 		break;
1992 	case MSR_KERNEL_GS_BASE:
1993 		msr_info->data = vmx_read_guest_kernel_gs_base(vmx);
1994 		break;
1995 #endif
1996 	case MSR_EFER:
1997 		return kvm_get_msr_common(vcpu, msr_info);
1998 	case MSR_IA32_TSX_CTRL:
1999 		if (!msr_info->host_initiated &&
2000 		    !(vcpu->arch.arch_capabilities & ARCH_CAP_TSX_CTRL_MSR))
2001 			return 1;
2002 		goto find_uret_msr;
2003 	case MSR_IA32_UMWAIT_CONTROL:
2004 		if (!msr_info->host_initiated && !vmx_has_waitpkg(vmx))
2005 			return 1;
2006 
2007 		msr_info->data = vmx->msr_ia32_umwait_control;
2008 		break;
2009 	case MSR_IA32_SPEC_CTRL:
2010 		if (!msr_info->host_initiated &&
2011 		    !guest_has_spec_ctrl_msr(vcpu))
2012 			return 1;
2013 
2014 		msr_info->data = to_vmx(vcpu)->spec_ctrl;
2015 		break;
2016 	case MSR_IA32_SYSENTER_CS:
2017 		msr_info->data = vmcs_read32(GUEST_SYSENTER_CS);
2018 		break;
2019 	case MSR_IA32_SYSENTER_EIP:
2020 		msr_info->data = vmcs_readl(GUEST_SYSENTER_EIP);
2021 		break;
2022 	case MSR_IA32_SYSENTER_ESP:
2023 		msr_info->data = vmcs_readl(GUEST_SYSENTER_ESP);
2024 		break;
2025 	case MSR_IA32_BNDCFGS:
2026 		if (!kvm_mpx_supported() ||
2027 		    (!msr_info->host_initiated &&
2028 		     !guest_cpuid_has(vcpu, X86_FEATURE_MPX)))
2029 			return 1;
2030 		msr_info->data = vmcs_read64(GUEST_BNDCFGS);
2031 		break;
2032 	case MSR_IA32_MCG_EXT_CTL:
2033 		if (!msr_info->host_initiated &&
2034 		    !(vmx->msr_ia32_feature_control &
2035 		      FEAT_CTL_LMCE_ENABLED))
2036 			return 1;
2037 		msr_info->data = vcpu->arch.mcg_ext_ctl;
2038 		break;
2039 	case MSR_IA32_FEAT_CTL:
2040 		msr_info->data = vmx->msr_ia32_feature_control;
2041 		break;
2042 	case MSR_IA32_SGXLEPUBKEYHASH0 ... MSR_IA32_SGXLEPUBKEYHASH3:
2043 		if (!msr_info->host_initiated &&
2044 		    !guest_cpuid_has(vcpu, X86_FEATURE_SGX_LC))
2045 			return 1;
2046 		msr_info->data = to_vmx(vcpu)->msr_ia32_sgxlepubkeyhash
2047 			[msr_info->index - MSR_IA32_SGXLEPUBKEYHASH0];
2048 		break;
2049 	case KVM_FIRST_EMULATED_VMX_MSR ... KVM_LAST_EMULATED_VMX_MSR:
2050 		if (!nested_vmx_allowed(vcpu))
2051 			return 1;
2052 		if (vmx_get_vmx_msr(&vmx->nested.msrs, msr_info->index,
2053 				    &msr_info->data))
2054 			return 1;
2055 		/*
2056 		 * Enlightened VMCS v1 doesn't have certain VMCS fields but
2057 		 * instead of just ignoring the features, different Hyper-V
2058 		 * versions are either trying to use them and fail or do some
2059 		 * sanity checking and refuse to boot. Filter all unsupported
2060 		 * features out.
2061 		 */
2062 		if (!msr_info->host_initiated && guest_cpuid_has_evmcs(vcpu))
2063 			nested_evmcs_filter_control_msr(vcpu, msr_info->index,
2064 							&msr_info->data);
2065 		break;
2066 	case MSR_IA32_RTIT_CTL:
2067 		if (!vmx_pt_mode_is_host_guest())
2068 			return 1;
2069 		msr_info->data = vmx->pt_desc.guest.ctl;
2070 		break;
2071 	case MSR_IA32_RTIT_STATUS:
2072 		if (!vmx_pt_mode_is_host_guest())
2073 			return 1;
2074 		msr_info->data = vmx->pt_desc.guest.status;
2075 		break;
2076 	case MSR_IA32_RTIT_CR3_MATCH:
2077 		if (!vmx_pt_mode_is_host_guest() ||
2078 			!intel_pt_validate_cap(vmx->pt_desc.caps,
2079 						PT_CAP_cr3_filtering))
2080 			return 1;
2081 		msr_info->data = vmx->pt_desc.guest.cr3_match;
2082 		break;
2083 	case MSR_IA32_RTIT_OUTPUT_BASE:
2084 		if (!vmx_pt_mode_is_host_guest() ||
2085 			(!intel_pt_validate_cap(vmx->pt_desc.caps,
2086 					PT_CAP_topa_output) &&
2087 			 !intel_pt_validate_cap(vmx->pt_desc.caps,
2088 					PT_CAP_single_range_output)))
2089 			return 1;
2090 		msr_info->data = vmx->pt_desc.guest.output_base;
2091 		break;
2092 	case MSR_IA32_RTIT_OUTPUT_MASK:
2093 		if (!vmx_pt_mode_is_host_guest() ||
2094 			(!intel_pt_validate_cap(vmx->pt_desc.caps,
2095 					PT_CAP_topa_output) &&
2096 			 !intel_pt_validate_cap(vmx->pt_desc.caps,
2097 					PT_CAP_single_range_output)))
2098 			return 1;
2099 		msr_info->data = vmx->pt_desc.guest.output_mask;
2100 		break;
2101 	case MSR_IA32_RTIT_ADDR0_A ... MSR_IA32_RTIT_ADDR3_B:
2102 		index = msr_info->index - MSR_IA32_RTIT_ADDR0_A;
2103 		if (!vmx_pt_mode_is_host_guest() ||
2104 		    (index >= 2 * vmx->pt_desc.num_address_ranges))
2105 			return 1;
2106 		if (index % 2)
2107 			msr_info->data = vmx->pt_desc.guest.addr_b[index / 2];
2108 		else
2109 			msr_info->data = vmx->pt_desc.guest.addr_a[index / 2];
2110 		break;
2111 	case MSR_IA32_DEBUGCTLMSR:
2112 		msr_info->data = vmcs_read64(GUEST_IA32_DEBUGCTL);
2113 		break;
2114 	default:
2115 	find_uret_msr:
2116 		msr = vmx_find_uret_msr(vmx, msr_info->index);
2117 		if (msr) {
2118 			msr_info->data = msr->data;
2119 			break;
2120 		}
2121 		return kvm_get_msr_common(vcpu, msr_info);
2122 	}
2123 
2124 	return 0;
2125 }
2126 
2127 static u64 nested_vmx_truncate_sysenter_addr(struct kvm_vcpu *vcpu,
2128 						    u64 data)
2129 {
2130 #ifdef CONFIG_X86_64
2131 	if (!guest_cpuid_has(vcpu, X86_FEATURE_LM))
2132 		return (u32)data;
2133 #endif
2134 	return (unsigned long)data;
2135 }
2136 
2137 static u64 vmx_get_supported_debugctl(struct kvm_vcpu *vcpu, bool host_initiated)
2138 {
2139 	u64 debugctl = 0;
2140 
2141 	if (boot_cpu_has(X86_FEATURE_BUS_LOCK_DETECT) &&
2142 	    (host_initiated || guest_cpuid_has(vcpu, X86_FEATURE_BUS_LOCK_DETECT)))
2143 		debugctl |= DEBUGCTLMSR_BUS_LOCK_DETECT;
2144 
2145 	if ((kvm_caps.supported_perf_cap & PMU_CAP_LBR_FMT) &&
2146 	    (host_initiated || intel_pmu_lbr_is_enabled(vcpu)))
2147 		debugctl |= DEBUGCTLMSR_LBR | DEBUGCTLMSR_FREEZE_LBRS_ON_PMI;
2148 
2149 	return debugctl;
2150 }
2151 
2152 /*
2153  * Writes msr value into the appropriate "register".
2154  * Returns 0 on success, non-0 otherwise.
2155  * Assumes vcpu_load() was already called.
2156  */
2157 static int vmx_set_msr(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
2158 {
2159 	struct vcpu_vmx *vmx = to_vmx(vcpu);
2160 	struct vmx_uret_msr *msr;
2161 	int ret = 0;
2162 	u32 msr_index = msr_info->index;
2163 	u64 data = msr_info->data;
2164 	u32 index;
2165 
2166 	switch (msr_index) {
2167 	case MSR_EFER:
2168 		ret = kvm_set_msr_common(vcpu, msr_info);
2169 		break;
2170 #ifdef CONFIG_X86_64
2171 	case MSR_FS_BASE:
2172 		vmx_segment_cache_clear(vmx);
2173 		vmcs_writel(GUEST_FS_BASE, data);
2174 		break;
2175 	case MSR_GS_BASE:
2176 		vmx_segment_cache_clear(vmx);
2177 		vmcs_writel(GUEST_GS_BASE, data);
2178 		break;
2179 	case MSR_KERNEL_GS_BASE:
2180 		vmx_write_guest_kernel_gs_base(vmx, data);
2181 		break;
2182 	case MSR_IA32_XFD:
2183 		ret = kvm_set_msr_common(vcpu, msr_info);
2184 		/*
2185 		 * Always intercepting WRMSR could incur non-negligible
2186 		 * overhead given xfd might be changed frequently in
2187 		 * guest context switch. Disable write interception
2188 		 * upon the first write with a non-zero value (indicating
2189 		 * potential usage on dynamic xfeatures). Also update
2190 		 * exception bitmap to trap #NM for proper virtualization
2191 		 * of guest xfd_err.
2192 		 */
2193 		if (!ret && data) {
2194 			vmx_disable_intercept_for_msr(vcpu, MSR_IA32_XFD,
2195 						      MSR_TYPE_RW);
2196 			vcpu->arch.xfd_no_write_intercept = true;
2197 			vmx_update_exception_bitmap(vcpu);
2198 		}
2199 		break;
2200 #endif
2201 	case MSR_IA32_SYSENTER_CS:
2202 		if (is_guest_mode(vcpu))
2203 			get_vmcs12(vcpu)->guest_sysenter_cs = data;
2204 		vmcs_write32(GUEST_SYSENTER_CS, data);
2205 		break;
2206 	case MSR_IA32_SYSENTER_EIP:
2207 		if (is_guest_mode(vcpu)) {
2208 			data = nested_vmx_truncate_sysenter_addr(vcpu, data);
2209 			get_vmcs12(vcpu)->guest_sysenter_eip = data;
2210 		}
2211 		vmcs_writel(GUEST_SYSENTER_EIP, data);
2212 		break;
2213 	case MSR_IA32_SYSENTER_ESP:
2214 		if (is_guest_mode(vcpu)) {
2215 			data = nested_vmx_truncate_sysenter_addr(vcpu, data);
2216 			get_vmcs12(vcpu)->guest_sysenter_esp = data;
2217 		}
2218 		vmcs_writel(GUEST_SYSENTER_ESP, data);
2219 		break;
2220 	case MSR_IA32_DEBUGCTLMSR: {
2221 		u64 invalid;
2222 
2223 		invalid = data & ~vmx_get_supported_debugctl(vcpu, msr_info->host_initiated);
2224 		if (invalid & (DEBUGCTLMSR_BTF|DEBUGCTLMSR_LBR)) {
2225 			kvm_pr_unimpl_wrmsr(vcpu, msr_index, data);
2226 			data &= ~(DEBUGCTLMSR_BTF|DEBUGCTLMSR_LBR);
2227 			invalid &= ~(DEBUGCTLMSR_BTF|DEBUGCTLMSR_LBR);
2228 		}
2229 
2230 		if (invalid)
2231 			return 1;
2232 
2233 		if (is_guest_mode(vcpu) && get_vmcs12(vcpu)->vm_exit_controls &
2234 						VM_EXIT_SAVE_DEBUG_CONTROLS)
2235 			get_vmcs12(vcpu)->guest_ia32_debugctl = data;
2236 
2237 		vmcs_write64(GUEST_IA32_DEBUGCTL, data);
2238 		if (intel_pmu_lbr_is_enabled(vcpu) && !to_vmx(vcpu)->lbr_desc.event &&
2239 		    (data & DEBUGCTLMSR_LBR))
2240 			intel_pmu_create_guest_lbr_event(vcpu);
2241 		return 0;
2242 	}
2243 	case MSR_IA32_BNDCFGS:
2244 		if (!kvm_mpx_supported() ||
2245 		    (!msr_info->host_initiated &&
2246 		     !guest_cpuid_has(vcpu, X86_FEATURE_MPX)))
2247 			return 1;
2248 		if (is_noncanonical_address(data & PAGE_MASK, vcpu) ||
2249 		    (data & MSR_IA32_BNDCFGS_RSVD))
2250 			return 1;
2251 
2252 		if (is_guest_mode(vcpu) &&
2253 		    ((vmx->nested.msrs.entry_ctls_high & VM_ENTRY_LOAD_BNDCFGS) ||
2254 		     (vmx->nested.msrs.exit_ctls_high & VM_EXIT_CLEAR_BNDCFGS)))
2255 			get_vmcs12(vcpu)->guest_bndcfgs = data;
2256 
2257 		vmcs_write64(GUEST_BNDCFGS, data);
2258 		break;
2259 	case MSR_IA32_UMWAIT_CONTROL:
2260 		if (!msr_info->host_initiated && !vmx_has_waitpkg(vmx))
2261 			return 1;
2262 
2263 		/* The reserved bit 1 and non-32 bit [63:32] should be zero */
2264 		if (data & (BIT_ULL(1) | GENMASK_ULL(63, 32)))
2265 			return 1;
2266 
2267 		vmx->msr_ia32_umwait_control = data;
2268 		break;
2269 	case MSR_IA32_SPEC_CTRL:
2270 		if (!msr_info->host_initiated &&
2271 		    !guest_has_spec_ctrl_msr(vcpu))
2272 			return 1;
2273 
2274 		if (kvm_spec_ctrl_test_value(data))
2275 			return 1;
2276 
2277 		vmx->spec_ctrl = data;
2278 		if (!data)
2279 			break;
2280 
2281 		/*
2282 		 * For non-nested:
2283 		 * When it's written (to non-zero) for the first time, pass
2284 		 * it through.
2285 		 *
2286 		 * For nested:
2287 		 * The handling of the MSR bitmap for L2 guests is done in
2288 		 * nested_vmx_prepare_msr_bitmap. We should not touch the
2289 		 * vmcs02.msr_bitmap here since it gets completely overwritten
2290 		 * in the merging. We update the vmcs01 here for L1 as well
2291 		 * since it will end up touching the MSR anyway now.
2292 		 */
2293 		vmx_disable_intercept_for_msr(vcpu,
2294 					      MSR_IA32_SPEC_CTRL,
2295 					      MSR_TYPE_RW);
2296 		break;
2297 	case MSR_IA32_TSX_CTRL:
2298 		if (!msr_info->host_initiated &&
2299 		    !(vcpu->arch.arch_capabilities & ARCH_CAP_TSX_CTRL_MSR))
2300 			return 1;
2301 		if (data & ~(TSX_CTRL_RTM_DISABLE | TSX_CTRL_CPUID_CLEAR))
2302 			return 1;
2303 		goto find_uret_msr;
2304 	case MSR_IA32_CR_PAT:
2305 		ret = kvm_set_msr_common(vcpu, msr_info);
2306 		if (ret)
2307 			break;
2308 
2309 		if (is_guest_mode(vcpu) &&
2310 		    get_vmcs12(vcpu)->vm_exit_controls & VM_EXIT_SAVE_IA32_PAT)
2311 			get_vmcs12(vcpu)->guest_ia32_pat = data;
2312 
2313 		if (vmcs_config.vmentry_ctrl & VM_ENTRY_LOAD_IA32_PAT)
2314 			vmcs_write64(GUEST_IA32_PAT, data);
2315 		break;
2316 	case MSR_IA32_MCG_EXT_CTL:
2317 		if ((!msr_info->host_initiated &&
2318 		     !(to_vmx(vcpu)->msr_ia32_feature_control &
2319 		       FEAT_CTL_LMCE_ENABLED)) ||
2320 		    (data & ~MCG_EXT_CTL_LMCE_EN))
2321 			return 1;
2322 		vcpu->arch.mcg_ext_ctl = data;
2323 		break;
2324 	case MSR_IA32_FEAT_CTL:
2325 		if (!is_vmx_feature_control_msr_valid(vmx, msr_info))
2326 			return 1;
2327 
2328 		vmx->msr_ia32_feature_control = data;
2329 		if (msr_info->host_initiated && data == 0)
2330 			vmx_leave_nested(vcpu);
2331 
2332 		/* SGX may be enabled/disabled by guest's firmware */
2333 		vmx_write_encls_bitmap(vcpu, NULL);
2334 		break;
2335 	case MSR_IA32_SGXLEPUBKEYHASH0 ... MSR_IA32_SGXLEPUBKEYHASH3:
2336 		/*
2337 		 * On real hardware, the LE hash MSRs are writable before
2338 		 * the firmware sets bit 0 in MSR 0x7a ("activating" SGX),
2339 		 * at which point SGX related bits in IA32_FEATURE_CONTROL
2340 		 * become writable.
2341 		 *
2342 		 * KVM does not emulate SGX activation for simplicity, so
2343 		 * allow writes to the LE hash MSRs if IA32_FEATURE_CONTROL
2344 		 * is unlocked.  This is technically not architectural
2345 		 * behavior, but it's close enough.
2346 		 */
2347 		if (!msr_info->host_initiated &&
2348 		    (!guest_cpuid_has(vcpu, X86_FEATURE_SGX_LC) ||
2349 		    ((vmx->msr_ia32_feature_control & FEAT_CTL_LOCKED) &&
2350 		    !(vmx->msr_ia32_feature_control & FEAT_CTL_SGX_LC_ENABLED))))
2351 			return 1;
2352 		vmx->msr_ia32_sgxlepubkeyhash
2353 			[msr_index - MSR_IA32_SGXLEPUBKEYHASH0] = data;
2354 		break;
2355 	case KVM_FIRST_EMULATED_VMX_MSR ... KVM_LAST_EMULATED_VMX_MSR:
2356 		if (!msr_info->host_initiated)
2357 			return 1; /* they are read-only */
2358 		if (!nested_vmx_allowed(vcpu))
2359 			return 1;
2360 		return vmx_set_vmx_msr(vcpu, msr_index, data);
2361 	case MSR_IA32_RTIT_CTL:
2362 		if (!vmx_pt_mode_is_host_guest() ||
2363 			vmx_rtit_ctl_check(vcpu, data) ||
2364 			vmx->nested.vmxon)
2365 			return 1;
2366 		vmcs_write64(GUEST_IA32_RTIT_CTL, data);
2367 		vmx->pt_desc.guest.ctl = data;
2368 		pt_update_intercept_for_msr(vcpu);
2369 		break;
2370 	case MSR_IA32_RTIT_STATUS:
2371 		if (!pt_can_write_msr(vmx))
2372 			return 1;
2373 		if (data & MSR_IA32_RTIT_STATUS_MASK)
2374 			return 1;
2375 		vmx->pt_desc.guest.status = data;
2376 		break;
2377 	case MSR_IA32_RTIT_CR3_MATCH:
2378 		if (!pt_can_write_msr(vmx))
2379 			return 1;
2380 		if (!intel_pt_validate_cap(vmx->pt_desc.caps,
2381 					   PT_CAP_cr3_filtering))
2382 			return 1;
2383 		vmx->pt_desc.guest.cr3_match = data;
2384 		break;
2385 	case MSR_IA32_RTIT_OUTPUT_BASE:
2386 		if (!pt_can_write_msr(vmx))
2387 			return 1;
2388 		if (!intel_pt_validate_cap(vmx->pt_desc.caps,
2389 					   PT_CAP_topa_output) &&
2390 		    !intel_pt_validate_cap(vmx->pt_desc.caps,
2391 					   PT_CAP_single_range_output))
2392 			return 1;
2393 		if (!pt_output_base_valid(vcpu, data))
2394 			return 1;
2395 		vmx->pt_desc.guest.output_base = data;
2396 		break;
2397 	case MSR_IA32_RTIT_OUTPUT_MASK:
2398 		if (!pt_can_write_msr(vmx))
2399 			return 1;
2400 		if (!intel_pt_validate_cap(vmx->pt_desc.caps,
2401 					   PT_CAP_topa_output) &&
2402 		    !intel_pt_validate_cap(vmx->pt_desc.caps,
2403 					   PT_CAP_single_range_output))
2404 			return 1;
2405 		vmx->pt_desc.guest.output_mask = data;
2406 		break;
2407 	case MSR_IA32_RTIT_ADDR0_A ... MSR_IA32_RTIT_ADDR3_B:
2408 		if (!pt_can_write_msr(vmx))
2409 			return 1;
2410 		index = msr_info->index - MSR_IA32_RTIT_ADDR0_A;
2411 		if (index >= 2 * vmx->pt_desc.num_address_ranges)
2412 			return 1;
2413 		if (is_noncanonical_address(data, vcpu))
2414 			return 1;
2415 		if (index % 2)
2416 			vmx->pt_desc.guest.addr_b[index / 2] = data;
2417 		else
2418 			vmx->pt_desc.guest.addr_a[index / 2] = data;
2419 		break;
2420 	case MSR_IA32_PERF_CAPABILITIES:
2421 		if (data && !vcpu_to_pmu(vcpu)->version)
2422 			return 1;
2423 		if (data & PMU_CAP_LBR_FMT) {
2424 			if ((data & PMU_CAP_LBR_FMT) !=
2425 			    (kvm_caps.supported_perf_cap & PMU_CAP_LBR_FMT))
2426 				return 1;
2427 			if (!cpuid_model_is_consistent(vcpu))
2428 				return 1;
2429 		}
2430 		if (data & PERF_CAP_PEBS_FORMAT) {
2431 			if ((data & PERF_CAP_PEBS_MASK) !=
2432 			    (kvm_caps.supported_perf_cap & PERF_CAP_PEBS_MASK))
2433 				return 1;
2434 			if (!guest_cpuid_has(vcpu, X86_FEATURE_DS))
2435 				return 1;
2436 			if (!guest_cpuid_has(vcpu, X86_FEATURE_DTES64))
2437 				return 1;
2438 			if (!cpuid_model_is_consistent(vcpu))
2439 				return 1;
2440 		}
2441 		ret = kvm_set_msr_common(vcpu, msr_info);
2442 		break;
2443 
2444 	default:
2445 	find_uret_msr:
2446 		msr = vmx_find_uret_msr(vmx, msr_index);
2447 		if (msr)
2448 			ret = vmx_set_guest_uret_msr(vmx, msr, data);
2449 		else
2450 			ret = kvm_set_msr_common(vcpu, msr_info);
2451 	}
2452 
2453 	/* FB_CLEAR may have changed, also update the FB_CLEAR_DIS behavior */
2454 	if (msr_index == MSR_IA32_ARCH_CAPABILITIES)
2455 		vmx_update_fb_clear_dis(vcpu, vmx);
2456 
2457 	return ret;
2458 }
2459 
2460 static void vmx_cache_reg(struct kvm_vcpu *vcpu, enum kvm_reg reg)
2461 {
2462 	unsigned long guest_owned_bits;
2463 
2464 	kvm_register_mark_available(vcpu, reg);
2465 
2466 	switch (reg) {
2467 	case VCPU_REGS_RSP:
2468 		vcpu->arch.regs[VCPU_REGS_RSP] = vmcs_readl(GUEST_RSP);
2469 		break;
2470 	case VCPU_REGS_RIP:
2471 		vcpu->arch.regs[VCPU_REGS_RIP] = vmcs_readl(GUEST_RIP);
2472 		break;
2473 	case VCPU_EXREG_PDPTR:
2474 		if (enable_ept)
2475 			ept_save_pdptrs(vcpu);
2476 		break;
2477 	case VCPU_EXREG_CR0:
2478 		guest_owned_bits = vcpu->arch.cr0_guest_owned_bits;
2479 
2480 		vcpu->arch.cr0 &= ~guest_owned_bits;
2481 		vcpu->arch.cr0 |= vmcs_readl(GUEST_CR0) & guest_owned_bits;
2482 		break;
2483 	case VCPU_EXREG_CR3:
2484 		/*
2485 		 * When intercepting CR3 loads, e.g. for shadowing paging, KVM's
2486 		 * CR3 is loaded into hardware, not the guest's CR3.
2487 		 */
2488 		if (!(exec_controls_get(to_vmx(vcpu)) & CPU_BASED_CR3_LOAD_EXITING))
2489 			vcpu->arch.cr3 = vmcs_readl(GUEST_CR3);
2490 		break;
2491 	case VCPU_EXREG_CR4:
2492 		guest_owned_bits = vcpu->arch.cr4_guest_owned_bits;
2493 
2494 		vcpu->arch.cr4 &= ~guest_owned_bits;
2495 		vcpu->arch.cr4 |= vmcs_readl(GUEST_CR4) & guest_owned_bits;
2496 		break;
2497 	default:
2498 		KVM_BUG_ON(1, vcpu->kvm);
2499 		break;
2500 	}
2501 }
2502 
2503 /*
2504  * There is no X86_FEATURE for SGX yet, but anyway we need to query CPUID
2505  * directly instead of going through cpu_has(), to ensure KVM is trapping
2506  * ENCLS whenever it's supported in hardware.  It does not matter whether
2507  * the host OS supports or has enabled SGX.
2508  */
2509 static bool cpu_has_sgx(void)
2510 {
2511 	return cpuid_eax(0) >= 0x12 && (cpuid_eax(0x12) & BIT(0));
2512 }
2513 
2514 /*
2515  * Some cpus support VM_{ENTRY,EXIT}_IA32_PERF_GLOBAL_CTRL but they
2516  * can't be used due to errata where VM Exit may incorrectly clear
2517  * IA32_PERF_GLOBAL_CTRL[34:32]. Work around the errata by using the
2518  * MSR load mechanism to switch IA32_PERF_GLOBAL_CTRL.
2519  */
2520 static bool cpu_has_perf_global_ctrl_bug(void)
2521 {
2522 	if (boot_cpu_data.x86 == 0x6) {
2523 		switch (boot_cpu_data.x86_model) {
2524 		case INTEL_FAM6_NEHALEM_EP:	/* AAK155 */
2525 		case INTEL_FAM6_NEHALEM:	/* AAP115 */
2526 		case INTEL_FAM6_WESTMERE:	/* AAT100 */
2527 		case INTEL_FAM6_WESTMERE_EP:	/* BC86,AAY89,BD102 */
2528 		case INTEL_FAM6_NEHALEM_EX:	/* BA97 */
2529 			return true;
2530 		default:
2531 			break;
2532 		}
2533 	}
2534 
2535 	return false;
2536 }
2537 
2538 static int adjust_vmx_controls(u32 ctl_min, u32 ctl_opt, u32 msr, u32 *result)
2539 {
2540 	u32 vmx_msr_low, vmx_msr_high;
2541 	u32 ctl = ctl_min | ctl_opt;
2542 
2543 	rdmsr(msr, vmx_msr_low, vmx_msr_high);
2544 
2545 	ctl &= vmx_msr_high; /* bit == 0 in high word ==> must be zero */
2546 	ctl |= vmx_msr_low;  /* bit == 1 in low word  ==> must be one  */
2547 
2548 	/* Ensure minimum (required) set of control bits are supported. */
2549 	if (ctl_min & ~ctl)
2550 		return -EIO;
2551 
2552 	*result = ctl;
2553 	return 0;
2554 }
2555 
2556 static u64 adjust_vmx_controls64(u64 ctl_opt, u32 msr)
2557 {
2558 	u64 allowed;
2559 
2560 	rdmsrl(msr, allowed);
2561 
2562 	return  ctl_opt & allowed;
2563 }
2564 
2565 static int setup_vmcs_config(struct vmcs_config *vmcs_conf,
2566 			     struct vmx_capability *vmx_cap)
2567 {
2568 	u32 vmx_msr_low, vmx_msr_high;
2569 	u32 _pin_based_exec_control = 0;
2570 	u32 _cpu_based_exec_control = 0;
2571 	u32 _cpu_based_2nd_exec_control = 0;
2572 	u64 _cpu_based_3rd_exec_control = 0;
2573 	u32 _vmexit_control = 0;
2574 	u32 _vmentry_control = 0;
2575 	u64 misc_msr;
2576 	int i;
2577 
2578 	/*
2579 	 * LOAD/SAVE_DEBUG_CONTROLS are absent because both are mandatory.
2580 	 * SAVE_IA32_PAT and SAVE_IA32_EFER are absent because KVM always
2581 	 * intercepts writes to PAT and EFER, i.e. never enables those controls.
2582 	 */
2583 	struct {
2584 		u32 entry_control;
2585 		u32 exit_control;
2586 	} const vmcs_entry_exit_pairs[] = {
2587 		{ VM_ENTRY_LOAD_IA32_PERF_GLOBAL_CTRL,	VM_EXIT_LOAD_IA32_PERF_GLOBAL_CTRL },
2588 		{ VM_ENTRY_LOAD_IA32_PAT,		VM_EXIT_LOAD_IA32_PAT },
2589 		{ VM_ENTRY_LOAD_IA32_EFER,		VM_EXIT_LOAD_IA32_EFER },
2590 		{ VM_ENTRY_LOAD_BNDCFGS,		VM_EXIT_CLEAR_BNDCFGS },
2591 		{ VM_ENTRY_LOAD_IA32_RTIT_CTL,		VM_EXIT_CLEAR_IA32_RTIT_CTL },
2592 	};
2593 
2594 	memset(vmcs_conf, 0, sizeof(*vmcs_conf));
2595 
2596 	if (adjust_vmx_controls(KVM_REQUIRED_VMX_CPU_BASED_VM_EXEC_CONTROL,
2597 				KVM_OPTIONAL_VMX_CPU_BASED_VM_EXEC_CONTROL,
2598 				MSR_IA32_VMX_PROCBASED_CTLS,
2599 				&_cpu_based_exec_control))
2600 		return -EIO;
2601 	if (_cpu_based_exec_control & CPU_BASED_ACTIVATE_SECONDARY_CONTROLS) {
2602 		if (adjust_vmx_controls(KVM_REQUIRED_VMX_SECONDARY_VM_EXEC_CONTROL,
2603 					KVM_OPTIONAL_VMX_SECONDARY_VM_EXEC_CONTROL,
2604 					MSR_IA32_VMX_PROCBASED_CTLS2,
2605 					&_cpu_based_2nd_exec_control))
2606 			return -EIO;
2607 	}
2608 #ifndef CONFIG_X86_64
2609 	if (!(_cpu_based_2nd_exec_control &
2610 				SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES))
2611 		_cpu_based_exec_control &= ~CPU_BASED_TPR_SHADOW;
2612 #endif
2613 
2614 	if (!(_cpu_based_exec_control & CPU_BASED_TPR_SHADOW))
2615 		_cpu_based_2nd_exec_control &= ~(
2616 				SECONDARY_EXEC_APIC_REGISTER_VIRT |
2617 				SECONDARY_EXEC_VIRTUALIZE_X2APIC_MODE |
2618 				SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY);
2619 
2620 	rdmsr_safe(MSR_IA32_VMX_EPT_VPID_CAP,
2621 		&vmx_cap->ept, &vmx_cap->vpid);
2622 
2623 	if (!(_cpu_based_2nd_exec_control & SECONDARY_EXEC_ENABLE_EPT) &&
2624 	    vmx_cap->ept) {
2625 		pr_warn_once("EPT CAP should not exist if not support "
2626 				"1-setting enable EPT VM-execution control\n");
2627 
2628 		if (error_on_inconsistent_vmcs_config)
2629 			return -EIO;
2630 
2631 		vmx_cap->ept = 0;
2632 	}
2633 	if (!(_cpu_based_2nd_exec_control & SECONDARY_EXEC_ENABLE_VPID) &&
2634 	    vmx_cap->vpid) {
2635 		pr_warn_once("VPID CAP should not exist if not support "
2636 				"1-setting enable VPID VM-execution control\n");
2637 
2638 		if (error_on_inconsistent_vmcs_config)
2639 			return -EIO;
2640 
2641 		vmx_cap->vpid = 0;
2642 	}
2643 
2644 	if (!cpu_has_sgx())
2645 		_cpu_based_2nd_exec_control &= ~SECONDARY_EXEC_ENCLS_EXITING;
2646 
2647 	if (_cpu_based_exec_control & CPU_BASED_ACTIVATE_TERTIARY_CONTROLS)
2648 		_cpu_based_3rd_exec_control =
2649 			adjust_vmx_controls64(KVM_OPTIONAL_VMX_TERTIARY_VM_EXEC_CONTROL,
2650 					      MSR_IA32_VMX_PROCBASED_CTLS3);
2651 
2652 	if (adjust_vmx_controls(KVM_REQUIRED_VMX_VM_EXIT_CONTROLS,
2653 				KVM_OPTIONAL_VMX_VM_EXIT_CONTROLS,
2654 				MSR_IA32_VMX_EXIT_CTLS,
2655 				&_vmexit_control))
2656 		return -EIO;
2657 
2658 	if (adjust_vmx_controls(KVM_REQUIRED_VMX_PIN_BASED_VM_EXEC_CONTROL,
2659 				KVM_OPTIONAL_VMX_PIN_BASED_VM_EXEC_CONTROL,
2660 				MSR_IA32_VMX_PINBASED_CTLS,
2661 				&_pin_based_exec_control))
2662 		return -EIO;
2663 
2664 	if (cpu_has_broken_vmx_preemption_timer())
2665 		_pin_based_exec_control &= ~PIN_BASED_VMX_PREEMPTION_TIMER;
2666 	if (!(_cpu_based_2nd_exec_control &
2667 		SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY))
2668 		_pin_based_exec_control &= ~PIN_BASED_POSTED_INTR;
2669 
2670 	if (adjust_vmx_controls(KVM_REQUIRED_VMX_VM_ENTRY_CONTROLS,
2671 				KVM_OPTIONAL_VMX_VM_ENTRY_CONTROLS,
2672 				MSR_IA32_VMX_ENTRY_CTLS,
2673 				&_vmentry_control))
2674 		return -EIO;
2675 
2676 	for (i = 0; i < ARRAY_SIZE(vmcs_entry_exit_pairs); i++) {
2677 		u32 n_ctrl = vmcs_entry_exit_pairs[i].entry_control;
2678 		u32 x_ctrl = vmcs_entry_exit_pairs[i].exit_control;
2679 
2680 		if (!(_vmentry_control & n_ctrl) == !(_vmexit_control & x_ctrl))
2681 			continue;
2682 
2683 		pr_warn_once("Inconsistent VM-Entry/VM-Exit pair, entry = %x, exit = %x\n",
2684 			     _vmentry_control & n_ctrl, _vmexit_control & x_ctrl);
2685 
2686 		if (error_on_inconsistent_vmcs_config)
2687 			return -EIO;
2688 
2689 		_vmentry_control &= ~n_ctrl;
2690 		_vmexit_control &= ~x_ctrl;
2691 	}
2692 
2693 	rdmsr(MSR_IA32_VMX_BASIC, vmx_msr_low, vmx_msr_high);
2694 
2695 	/* IA-32 SDM Vol 3B: VMCS size is never greater than 4kB. */
2696 	if ((vmx_msr_high & 0x1fff) > PAGE_SIZE)
2697 		return -EIO;
2698 
2699 #ifdef CONFIG_X86_64
2700 	/* IA-32 SDM Vol 3B: 64-bit CPUs always have VMX_BASIC_MSR[48]==0. */
2701 	if (vmx_msr_high & (1u<<16))
2702 		return -EIO;
2703 #endif
2704 
2705 	/* Require Write-Back (WB) memory type for VMCS accesses. */
2706 	if (((vmx_msr_high >> 18) & 15) != 6)
2707 		return -EIO;
2708 
2709 	rdmsrl(MSR_IA32_VMX_MISC, misc_msr);
2710 
2711 	vmcs_conf->size = vmx_msr_high & 0x1fff;
2712 	vmcs_conf->basic_cap = vmx_msr_high & ~0x1fff;
2713 
2714 	vmcs_conf->revision_id = vmx_msr_low;
2715 
2716 	vmcs_conf->pin_based_exec_ctrl = _pin_based_exec_control;
2717 	vmcs_conf->cpu_based_exec_ctrl = _cpu_based_exec_control;
2718 	vmcs_conf->cpu_based_2nd_exec_ctrl = _cpu_based_2nd_exec_control;
2719 	vmcs_conf->cpu_based_3rd_exec_ctrl = _cpu_based_3rd_exec_control;
2720 	vmcs_conf->vmexit_ctrl         = _vmexit_control;
2721 	vmcs_conf->vmentry_ctrl        = _vmentry_control;
2722 	vmcs_conf->misc	= misc_msr;
2723 
2724 #if IS_ENABLED(CONFIG_HYPERV)
2725 	if (enlightened_vmcs)
2726 		evmcs_sanitize_exec_ctrls(vmcs_conf);
2727 #endif
2728 
2729 	return 0;
2730 }
2731 
2732 static bool kvm_is_vmx_supported(void)
2733 {
2734 	int cpu = raw_smp_processor_id();
2735 
2736 	if (!cpu_has_vmx()) {
2737 		pr_err("VMX not supported by CPU %d\n", cpu);
2738 		return false;
2739 	}
2740 
2741 	if (!this_cpu_has(X86_FEATURE_MSR_IA32_FEAT_CTL) ||
2742 	    !this_cpu_has(X86_FEATURE_VMX)) {
2743 		pr_err("VMX not enabled (by BIOS) in MSR_IA32_FEAT_CTL on CPU %d\n", cpu);
2744 		return false;
2745 	}
2746 
2747 	return true;
2748 }
2749 
2750 static int vmx_check_processor_compat(void)
2751 {
2752 	int cpu = raw_smp_processor_id();
2753 	struct vmcs_config vmcs_conf;
2754 	struct vmx_capability vmx_cap;
2755 
2756 	if (!kvm_is_vmx_supported())
2757 		return -EIO;
2758 
2759 	if (setup_vmcs_config(&vmcs_conf, &vmx_cap) < 0) {
2760 		pr_err("Failed to setup VMCS config on CPU %d\n", cpu);
2761 		return -EIO;
2762 	}
2763 	if (nested)
2764 		nested_vmx_setup_ctls_msrs(&vmcs_conf, vmx_cap.ept);
2765 	if (memcmp(&vmcs_config, &vmcs_conf, sizeof(struct vmcs_config))) {
2766 		pr_err("Inconsistent VMCS config on CPU %d\n", cpu);
2767 		return -EIO;
2768 	}
2769 	return 0;
2770 }
2771 
2772 static int kvm_cpu_vmxon(u64 vmxon_pointer)
2773 {
2774 	u64 msr;
2775 
2776 	cr4_set_bits(X86_CR4_VMXE);
2777 
2778 	asm_volatile_goto("1: vmxon %[vmxon_pointer]\n\t"
2779 			  _ASM_EXTABLE(1b, %l[fault])
2780 			  : : [vmxon_pointer] "m"(vmxon_pointer)
2781 			  : : fault);
2782 	return 0;
2783 
2784 fault:
2785 	WARN_ONCE(1, "VMXON faulted, MSR_IA32_FEAT_CTL (0x3a) = 0x%llx\n",
2786 		  rdmsrl_safe(MSR_IA32_FEAT_CTL, &msr) ? 0xdeadbeef : msr);
2787 	cr4_clear_bits(X86_CR4_VMXE);
2788 
2789 	return -EFAULT;
2790 }
2791 
2792 static int vmx_hardware_enable(void)
2793 {
2794 	int cpu = raw_smp_processor_id();
2795 	u64 phys_addr = __pa(per_cpu(vmxarea, cpu));
2796 	int r;
2797 
2798 	if (cr4_read_shadow() & X86_CR4_VMXE)
2799 		return -EBUSY;
2800 
2801 	/*
2802 	 * This can happen if we hot-added a CPU but failed to allocate
2803 	 * VP assist page for it.
2804 	 */
2805 	if (kvm_is_using_evmcs() && !hv_get_vp_assist_page(cpu))
2806 		return -EFAULT;
2807 
2808 	intel_pt_handle_vmx(1);
2809 
2810 	r = kvm_cpu_vmxon(phys_addr);
2811 	if (r) {
2812 		intel_pt_handle_vmx(0);
2813 		return r;
2814 	}
2815 
2816 	if (enable_ept)
2817 		ept_sync_global();
2818 
2819 	return 0;
2820 }
2821 
2822 static void vmclear_local_loaded_vmcss(void)
2823 {
2824 	int cpu = raw_smp_processor_id();
2825 	struct loaded_vmcs *v, *n;
2826 
2827 	list_for_each_entry_safe(v, n, &per_cpu(loaded_vmcss_on_cpu, cpu),
2828 				 loaded_vmcss_on_cpu_link)
2829 		__loaded_vmcs_clear(v);
2830 }
2831 
2832 static void vmx_hardware_disable(void)
2833 {
2834 	vmclear_local_loaded_vmcss();
2835 
2836 	if (cpu_vmxoff())
2837 		kvm_spurious_fault();
2838 
2839 	hv_reset_evmcs();
2840 
2841 	intel_pt_handle_vmx(0);
2842 }
2843 
2844 struct vmcs *alloc_vmcs_cpu(bool shadow, int cpu, gfp_t flags)
2845 {
2846 	int node = cpu_to_node(cpu);
2847 	struct page *pages;
2848 	struct vmcs *vmcs;
2849 
2850 	pages = __alloc_pages_node(node, flags, 0);
2851 	if (!pages)
2852 		return NULL;
2853 	vmcs = page_address(pages);
2854 	memset(vmcs, 0, vmcs_config.size);
2855 
2856 	/* KVM supports Enlightened VMCS v1 only */
2857 	if (kvm_is_using_evmcs())
2858 		vmcs->hdr.revision_id = KVM_EVMCS_VERSION;
2859 	else
2860 		vmcs->hdr.revision_id = vmcs_config.revision_id;
2861 
2862 	if (shadow)
2863 		vmcs->hdr.shadow_vmcs = 1;
2864 	return vmcs;
2865 }
2866 
2867 void free_vmcs(struct vmcs *vmcs)
2868 {
2869 	free_page((unsigned long)vmcs);
2870 }
2871 
2872 /*
2873  * Free a VMCS, but before that VMCLEAR it on the CPU where it was last loaded
2874  */
2875 void free_loaded_vmcs(struct loaded_vmcs *loaded_vmcs)
2876 {
2877 	if (!loaded_vmcs->vmcs)
2878 		return;
2879 	loaded_vmcs_clear(loaded_vmcs);
2880 	free_vmcs(loaded_vmcs->vmcs);
2881 	loaded_vmcs->vmcs = NULL;
2882 	if (loaded_vmcs->msr_bitmap)
2883 		free_page((unsigned long)loaded_vmcs->msr_bitmap);
2884 	WARN_ON(loaded_vmcs->shadow_vmcs != NULL);
2885 }
2886 
2887 int alloc_loaded_vmcs(struct loaded_vmcs *loaded_vmcs)
2888 {
2889 	loaded_vmcs->vmcs = alloc_vmcs(false);
2890 	if (!loaded_vmcs->vmcs)
2891 		return -ENOMEM;
2892 
2893 	vmcs_clear(loaded_vmcs->vmcs);
2894 
2895 	loaded_vmcs->shadow_vmcs = NULL;
2896 	loaded_vmcs->hv_timer_soft_disabled = false;
2897 	loaded_vmcs->cpu = -1;
2898 	loaded_vmcs->launched = 0;
2899 
2900 	if (cpu_has_vmx_msr_bitmap()) {
2901 		loaded_vmcs->msr_bitmap = (unsigned long *)
2902 				__get_free_page(GFP_KERNEL_ACCOUNT);
2903 		if (!loaded_vmcs->msr_bitmap)
2904 			goto out_vmcs;
2905 		memset(loaded_vmcs->msr_bitmap, 0xff, PAGE_SIZE);
2906 	}
2907 
2908 	memset(&loaded_vmcs->host_state, 0, sizeof(struct vmcs_host_state));
2909 	memset(&loaded_vmcs->controls_shadow, 0,
2910 		sizeof(struct vmcs_controls_shadow));
2911 
2912 	return 0;
2913 
2914 out_vmcs:
2915 	free_loaded_vmcs(loaded_vmcs);
2916 	return -ENOMEM;
2917 }
2918 
2919 static void free_kvm_area(void)
2920 {
2921 	int cpu;
2922 
2923 	for_each_possible_cpu(cpu) {
2924 		free_vmcs(per_cpu(vmxarea, cpu));
2925 		per_cpu(vmxarea, cpu) = NULL;
2926 	}
2927 }
2928 
2929 static __init int alloc_kvm_area(void)
2930 {
2931 	int cpu;
2932 
2933 	for_each_possible_cpu(cpu) {
2934 		struct vmcs *vmcs;
2935 
2936 		vmcs = alloc_vmcs_cpu(false, cpu, GFP_KERNEL);
2937 		if (!vmcs) {
2938 			free_kvm_area();
2939 			return -ENOMEM;
2940 		}
2941 
2942 		/*
2943 		 * When eVMCS is enabled, alloc_vmcs_cpu() sets
2944 		 * vmcs->revision_id to KVM_EVMCS_VERSION instead of
2945 		 * revision_id reported by MSR_IA32_VMX_BASIC.
2946 		 *
2947 		 * However, even though not explicitly documented by
2948 		 * TLFS, VMXArea passed as VMXON argument should
2949 		 * still be marked with revision_id reported by
2950 		 * physical CPU.
2951 		 */
2952 		if (kvm_is_using_evmcs())
2953 			vmcs->hdr.revision_id = vmcs_config.revision_id;
2954 
2955 		per_cpu(vmxarea, cpu) = vmcs;
2956 	}
2957 	return 0;
2958 }
2959 
2960 static void fix_pmode_seg(struct kvm_vcpu *vcpu, int seg,
2961 		struct kvm_segment *save)
2962 {
2963 	if (!emulate_invalid_guest_state) {
2964 		/*
2965 		 * CS and SS RPL should be equal during guest entry according
2966 		 * to VMX spec, but in reality it is not always so. Since vcpu
2967 		 * is in the middle of the transition from real mode to
2968 		 * protected mode it is safe to assume that RPL 0 is a good
2969 		 * default value.
2970 		 */
2971 		if (seg == VCPU_SREG_CS || seg == VCPU_SREG_SS)
2972 			save->selector &= ~SEGMENT_RPL_MASK;
2973 		save->dpl = save->selector & SEGMENT_RPL_MASK;
2974 		save->s = 1;
2975 	}
2976 	__vmx_set_segment(vcpu, save, seg);
2977 }
2978 
2979 static void enter_pmode(struct kvm_vcpu *vcpu)
2980 {
2981 	unsigned long flags;
2982 	struct vcpu_vmx *vmx = to_vmx(vcpu);
2983 
2984 	/*
2985 	 * Update real mode segment cache. It may be not up-to-date if segment
2986 	 * register was written while vcpu was in a guest mode.
2987 	 */
2988 	vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_ES], VCPU_SREG_ES);
2989 	vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_DS], VCPU_SREG_DS);
2990 	vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_FS], VCPU_SREG_FS);
2991 	vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_GS], VCPU_SREG_GS);
2992 	vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_SS], VCPU_SREG_SS);
2993 	vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_CS], VCPU_SREG_CS);
2994 
2995 	vmx->rmode.vm86_active = 0;
2996 
2997 	__vmx_set_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_TR], VCPU_SREG_TR);
2998 
2999 	flags = vmcs_readl(GUEST_RFLAGS);
3000 	flags &= RMODE_GUEST_OWNED_EFLAGS_BITS;
3001 	flags |= vmx->rmode.save_rflags & ~RMODE_GUEST_OWNED_EFLAGS_BITS;
3002 	vmcs_writel(GUEST_RFLAGS, flags);
3003 
3004 	vmcs_writel(GUEST_CR4, (vmcs_readl(GUEST_CR4) & ~X86_CR4_VME) |
3005 			(vmcs_readl(CR4_READ_SHADOW) & X86_CR4_VME));
3006 
3007 	vmx_update_exception_bitmap(vcpu);
3008 
3009 	fix_pmode_seg(vcpu, VCPU_SREG_CS, &vmx->rmode.segs[VCPU_SREG_CS]);
3010 	fix_pmode_seg(vcpu, VCPU_SREG_SS, &vmx->rmode.segs[VCPU_SREG_SS]);
3011 	fix_pmode_seg(vcpu, VCPU_SREG_ES, &vmx->rmode.segs[VCPU_SREG_ES]);
3012 	fix_pmode_seg(vcpu, VCPU_SREG_DS, &vmx->rmode.segs[VCPU_SREG_DS]);
3013 	fix_pmode_seg(vcpu, VCPU_SREG_FS, &vmx->rmode.segs[VCPU_SREG_FS]);
3014 	fix_pmode_seg(vcpu, VCPU_SREG_GS, &vmx->rmode.segs[VCPU_SREG_GS]);
3015 }
3016 
3017 static void fix_rmode_seg(int seg, struct kvm_segment *save)
3018 {
3019 	const struct kvm_vmx_segment_field *sf = &kvm_vmx_segment_fields[seg];
3020 	struct kvm_segment var = *save;
3021 
3022 	var.dpl = 0x3;
3023 	if (seg == VCPU_SREG_CS)
3024 		var.type = 0x3;
3025 
3026 	if (!emulate_invalid_guest_state) {
3027 		var.selector = var.base >> 4;
3028 		var.base = var.base & 0xffff0;
3029 		var.limit = 0xffff;
3030 		var.g = 0;
3031 		var.db = 0;
3032 		var.present = 1;
3033 		var.s = 1;
3034 		var.l = 0;
3035 		var.unusable = 0;
3036 		var.type = 0x3;
3037 		var.avl = 0;
3038 		if (save->base & 0xf)
3039 			pr_warn_once("segment base is not paragraph aligned "
3040 				     "when entering protected mode (seg=%d)", seg);
3041 	}
3042 
3043 	vmcs_write16(sf->selector, var.selector);
3044 	vmcs_writel(sf->base, var.base);
3045 	vmcs_write32(sf->limit, var.limit);
3046 	vmcs_write32(sf->ar_bytes, vmx_segment_access_rights(&var));
3047 }
3048 
3049 static void enter_rmode(struct kvm_vcpu *vcpu)
3050 {
3051 	unsigned long flags;
3052 	struct vcpu_vmx *vmx = to_vmx(vcpu);
3053 	struct kvm_vmx *kvm_vmx = to_kvm_vmx(vcpu->kvm);
3054 
3055 	/*
3056 	 * KVM should never use VM86 to virtualize Real Mode when L2 is active,
3057 	 * as using VM86 is unnecessary if unrestricted guest is enabled, and
3058 	 * if unrestricted guest is disabled, VM-Enter (from L1) with CR0.PG=0
3059 	 * should VM-Fail and KVM should reject userspace attempts to stuff
3060 	 * CR0.PG=0 when L2 is active.
3061 	 */
3062 	WARN_ON_ONCE(is_guest_mode(vcpu));
3063 
3064 	vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_TR], VCPU_SREG_TR);
3065 	vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_ES], VCPU_SREG_ES);
3066 	vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_DS], VCPU_SREG_DS);
3067 	vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_FS], VCPU_SREG_FS);
3068 	vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_GS], VCPU_SREG_GS);
3069 	vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_SS], VCPU_SREG_SS);
3070 	vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_CS], VCPU_SREG_CS);
3071 
3072 	vmx->rmode.vm86_active = 1;
3073 
3074 	/*
3075 	 * Very old userspace does not call KVM_SET_TSS_ADDR before entering
3076 	 * vcpu. Warn the user that an update is overdue.
3077 	 */
3078 	if (!kvm_vmx->tss_addr)
3079 		pr_warn_once("KVM_SET_TSS_ADDR needs to be called before running vCPU\n");
3080 
3081 	vmx_segment_cache_clear(vmx);
3082 
3083 	vmcs_writel(GUEST_TR_BASE, kvm_vmx->tss_addr);
3084 	vmcs_write32(GUEST_TR_LIMIT, RMODE_TSS_SIZE - 1);
3085 	vmcs_write32(GUEST_TR_AR_BYTES, 0x008b);
3086 
3087 	flags = vmcs_readl(GUEST_RFLAGS);
3088 	vmx->rmode.save_rflags = flags;
3089 
3090 	flags |= X86_EFLAGS_IOPL | X86_EFLAGS_VM;
3091 
3092 	vmcs_writel(GUEST_RFLAGS, flags);
3093 	vmcs_writel(GUEST_CR4, vmcs_readl(GUEST_CR4) | X86_CR4_VME);
3094 	vmx_update_exception_bitmap(vcpu);
3095 
3096 	fix_rmode_seg(VCPU_SREG_SS, &vmx->rmode.segs[VCPU_SREG_SS]);
3097 	fix_rmode_seg(VCPU_SREG_CS, &vmx->rmode.segs[VCPU_SREG_CS]);
3098 	fix_rmode_seg(VCPU_SREG_ES, &vmx->rmode.segs[VCPU_SREG_ES]);
3099 	fix_rmode_seg(VCPU_SREG_DS, &vmx->rmode.segs[VCPU_SREG_DS]);
3100 	fix_rmode_seg(VCPU_SREG_GS, &vmx->rmode.segs[VCPU_SREG_GS]);
3101 	fix_rmode_seg(VCPU_SREG_FS, &vmx->rmode.segs[VCPU_SREG_FS]);
3102 }
3103 
3104 int vmx_set_efer(struct kvm_vcpu *vcpu, u64 efer)
3105 {
3106 	struct vcpu_vmx *vmx = to_vmx(vcpu);
3107 
3108 	/* Nothing to do if hardware doesn't support EFER. */
3109 	if (!vmx_find_uret_msr(vmx, MSR_EFER))
3110 		return 0;
3111 
3112 	vcpu->arch.efer = efer;
3113 #ifdef CONFIG_X86_64
3114 	if (efer & EFER_LMA)
3115 		vm_entry_controls_setbit(vmx, VM_ENTRY_IA32E_MODE);
3116 	else
3117 		vm_entry_controls_clearbit(vmx, VM_ENTRY_IA32E_MODE);
3118 #else
3119 	if (KVM_BUG_ON(efer & EFER_LMA, vcpu->kvm))
3120 		return 1;
3121 #endif
3122 
3123 	vmx_setup_uret_msrs(vmx);
3124 	return 0;
3125 }
3126 
3127 #ifdef CONFIG_X86_64
3128 
3129 static void enter_lmode(struct kvm_vcpu *vcpu)
3130 {
3131 	u32 guest_tr_ar;
3132 
3133 	vmx_segment_cache_clear(to_vmx(vcpu));
3134 
3135 	guest_tr_ar = vmcs_read32(GUEST_TR_AR_BYTES);
3136 	if ((guest_tr_ar & VMX_AR_TYPE_MASK) != VMX_AR_TYPE_BUSY_64_TSS) {
3137 		pr_debug_ratelimited("%s: tss fixup for long mode. \n",
3138 				     __func__);
3139 		vmcs_write32(GUEST_TR_AR_BYTES,
3140 			     (guest_tr_ar & ~VMX_AR_TYPE_MASK)
3141 			     | VMX_AR_TYPE_BUSY_64_TSS);
3142 	}
3143 	vmx_set_efer(vcpu, vcpu->arch.efer | EFER_LMA);
3144 }
3145 
3146 static void exit_lmode(struct kvm_vcpu *vcpu)
3147 {
3148 	vmx_set_efer(vcpu, vcpu->arch.efer & ~EFER_LMA);
3149 }
3150 
3151 #endif
3152 
3153 static void vmx_flush_tlb_all(struct kvm_vcpu *vcpu)
3154 {
3155 	struct vcpu_vmx *vmx = to_vmx(vcpu);
3156 
3157 	/*
3158 	 * INVEPT must be issued when EPT is enabled, irrespective of VPID, as
3159 	 * the CPU is not required to invalidate guest-physical mappings on
3160 	 * VM-Entry, even if VPID is disabled.  Guest-physical mappings are
3161 	 * associated with the root EPT structure and not any particular VPID
3162 	 * (INVVPID also isn't required to invalidate guest-physical mappings).
3163 	 */
3164 	if (enable_ept) {
3165 		ept_sync_global();
3166 	} else if (enable_vpid) {
3167 		if (cpu_has_vmx_invvpid_global()) {
3168 			vpid_sync_vcpu_global();
3169 		} else {
3170 			vpid_sync_vcpu_single(vmx->vpid);
3171 			vpid_sync_vcpu_single(vmx->nested.vpid02);
3172 		}
3173 	}
3174 }
3175 
3176 static inline int vmx_get_current_vpid(struct kvm_vcpu *vcpu)
3177 {
3178 	if (is_guest_mode(vcpu))
3179 		return nested_get_vpid02(vcpu);
3180 	return to_vmx(vcpu)->vpid;
3181 }
3182 
3183 static void vmx_flush_tlb_current(struct kvm_vcpu *vcpu)
3184 {
3185 	struct kvm_mmu *mmu = vcpu->arch.mmu;
3186 	u64 root_hpa = mmu->root.hpa;
3187 
3188 	/* No flush required if the current context is invalid. */
3189 	if (!VALID_PAGE(root_hpa))
3190 		return;
3191 
3192 	if (enable_ept)
3193 		ept_sync_context(construct_eptp(vcpu, root_hpa,
3194 						mmu->root_role.level));
3195 	else
3196 		vpid_sync_context(vmx_get_current_vpid(vcpu));
3197 }
3198 
3199 static void vmx_flush_tlb_gva(struct kvm_vcpu *vcpu, gva_t addr)
3200 {
3201 	/*
3202 	 * vpid_sync_vcpu_addr() is a nop if vpid==0, see the comment in
3203 	 * vmx_flush_tlb_guest() for an explanation of why this is ok.
3204 	 */
3205 	vpid_sync_vcpu_addr(vmx_get_current_vpid(vcpu), addr);
3206 }
3207 
3208 static void vmx_flush_tlb_guest(struct kvm_vcpu *vcpu)
3209 {
3210 	/*
3211 	 * vpid_sync_context() is a nop if vpid==0, e.g. if enable_vpid==0 or a
3212 	 * vpid couldn't be allocated for this vCPU.  VM-Enter and VM-Exit are
3213 	 * required to flush GVA->{G,H}PA mappings from the TLB if vpid is
3214 	 * disabled (VM-Enter with vpid enabled and vpid==0 is disallowed),
3215 	 * i.e. no explicit INVVPID is necessary.
3216 	 */
3217 	vpid_sync_context(vmx_get_current_vpid(vcpu));
3218 }
3219 
3220 void vmx_ept_load_pdptrs(struct kvm_vcpu *vcpu)
3221 {
3222 	struct kvm_mmu *mmu = vcpu->arch.walk_mmu;
3223 
3224 	if (!kvm_register_is_dirty(vcpu, VCPU_EXREG_PDPTR))
3225 		return;
3226 
3227 	if (is_pae_paging(vcpu)) {
3228 		vmcs_write64(GUEST_PDPTR0, mmu->pdptrs[0]);
3229 		vmcs_write64(GUEST_PDPTR1, mmu->pdptrs[1]);
3230 		vmcs_write64(GUEST_PDPTR2, mmu->pdptrs[2]);
3231 		vmcs_write64(GUEST_PDPTR3, mmu->pdptrs[3]);
3232 	}
3233 }
3234 
3235 void ept_save_pdptrs(struct kvm_vcpu *vcpu)
3236 {
3237 	struct kvm_mmu *mmu = vcpu->arch.walk_mmu;
3238 
3239 	if (WARN_ON_ONCE(!is_pae_paging(vcpu)))
3240 		return;
3241 
3242 	mmu->pdptrs[0] = vmcs_read64(GUEST_PDPTR0);
3243 	mmu->pdptrs[1] = vmcs_read64(GUEST_PDPTR1);
3244 	mmu->pdptrs[2] = vmcs_read64(GUEST_PDPTR2);
3245 	mmu->pdptrs[3] = vmcs_read64(GUEST_PDPTR3);
3246 
3247 	kvm_register_mark_available(vcpu, VCPU_EXREG_PDPTR);
3248 }
3249 
3250 #define CR3_EXITING_BITS (CPU_BASED_CR3_LOAD_EXITING | \
3251 			  CPU_BASED_CR3_STORE_EXITING)
3252 
3253 static bool vmx_is_valid_cr0(struct kvm_vcpu *vcpu, unsigned long cr0)
3254 {
3255 	if (is_guest_mode(vcpu))
3256 		return nested_guest_cr0_valid(vcpu, cr0);
3257 
3258 	if (to_vmx(vcpu)->nested.vmxon)
3259 		return nested_host_cr0_valid(vcpu, cr0);
3260 
3261 	return true;
3262 }
3263 
3264 void vmx_set_cr0(struct kvm_vcpu *vcpu, unsigned long cr0)
3265 {
3266 	struct vcpu_vmx *vmx = to_vmx(vcpu);
3267 	unsigned long hw_cr0, old_cr0_pg;
3268 	u32 tmp;
3269 
3270 	old_cr0_pg = kvm_read_cr0_bits(vcpu, X86_CR0_PG);
3271 
3272 	hw_cr0 = (cr0 & ~KVM_VM_CR0_ALWAYS_OFF);
3273 	if (enable_unrestricted_guest)
3274 		hw_cr0 |= KVM_VM_CR0_ALWAYS_ON_UNRESTRICTED_GUEST;
3275 	else {
3276 		hw_cr0 |= KVM_VM_CR0_ALWAYS_ON;
3277 		if (!enable_ept)
3278 			hw_cr0 |= X86_CR0_WP;
3279 
3280 		if (vmx->rmode.vm86_active && (cr0 & X86_CR0_PE))
3281 			enter_pmode(vcpu);
3282 
3283 		if (!vmx->rmode.vm86_active && !(cr0 & X86_CR0_PE))
3284 			enter_rmode(vcpu);
3285 	}
3286 
3287 	vmcs_writel(CR0_READ_SHADOW, cr0);
3288 	vmcs_writel(GUEST_CR0, hw_cr0);
3289 	vcpu->arch.cr0 = cr0;
3290 	kvm_register_mark_available(vcpu, VCPU_EXREG_CR0);
3291 
3292 #ifdef CONFIG_X86_64
3293 	if (vcpu->arch.efer & EFER_LME) {
3294 		if (!old_cr0_pg && (cr0 & X86_CR0_PG))
3295 			enter_lmode(vcpu);
3296 		else if (old_cr0_pg && !(cr0 & X86_CR0_PG))
3297 			exit_lmode(vcpu);
3298 	}
3299 #endif
3300 
3301 	if (enable_ept && !enable_unrestricted_guest) {
3302 		/*
3303 		 * Ensure KVM has an up-to-date snapshot of the guest's CR3.  If
3304 		 * the below code _enables_ CR3 exiting, vmx_cache_reg() will
3305 		 * (correctly) stop reading vmcs.GUEST_CR3 because it thinks
3306 		 * KVM's CR3 is installed.
3307 		 */
3308 		if (!kvm_register_is_available(vcpu, VCPU_EXREG_CR3))
3309 			vmx_cache_reg(vcpu, VCPU_EXREG_CR3);
3310 
3311 		/*
3312 		 * When running with EPT but not unrestricted guest, KVM must
3313 		 * intercept CR3 accesses when paging is _disabled_.  This is
3314 		 * necessary because restricted guests can't actually run with
3315 		 * paging disabled, and so KVM stuffs its own CR3 in order to
3316 		 * run the guest when identity mapped page tables.
3317 		 *
3318 		 * Do _NOT_ check the old CR0.PG, e.g. to optimize away the
3319 		 * update, it may be stale with respect to CR3 interception,
3320 		 * e.g. after nested VM-Enter.
3321 		 *
3322 		 * Lastly, honor L1's desires, i.e. intercept CR3 loads and/or
3323 		 * stores to forward them to L1, even if KVM does not need to
3324 		 * intercept them to preserve its identity mapped page tables.
3325 		 */
3326 		if (!(cr0 & X86_CR0_PG)) {
3327 			exec_controls_setbit(vmx, CR3_EXITING_BITS);
3328 		} else if (!is_guest_mode(vcpu)) {
3329 			exec_controls_clearbit(vmx, CR3_EXITING_BITS);
3330 		} else {
3331 			tmp = exec_controls_get(vmx);
3332 			tmp &= ~CR3_EXITING_BITS;
3333 			tmp |= get_vmcs12(vcpu)->cpu_based_vm_exec_control & CR3_EXITING_BITS;
3334 			exec_controls_set(vmx, tmp);
3335 		}
3336 
3337 		/* Note, vmx_set_cr4() consumes the new vcpu->arch.cr0. */
3338 		if ((old_cr0_pg ^ cr0) & X86_CR0_PG)
3339 			vmx_set_cr4(vcpu, kvm_read_cr4(vcpu));
3340 
3341 		/*
3342 		 * When !CR0_PG -> CR0_PG, vcpu->arch.cr3 becomes active, but
3343 		 * GUEST_CR3 is still vmx->ept_identity_map_addr if EPT + !URG.
3344 		 */
3345 		if (!(old_cr0_pg & X86_CR0_PG) && (cr0 & X86_CR0_PG))
3346 			kvm_register_mark_dirty(vcpu, VCPU_EXREG_CR3);
3347 	}
3348 
3349 	/* depends on vcpu->arch.cr0 to be set to a new value */
3350 	vmx->emulation_required = vmx_emulation_required(vcpu);
3351 }
3352 
3353 static int vmx_get_max_tdp_level(void)
3354 {
3355 	if (cpu_has_vmx_ept_5levels())
3356 		return 5;
3357 	return 4;
3358 }
3359 
3360 u64 construct_eptp(struct kvm_vcpu *vcpu, hpa_t root_hpa, int root_level)
3361 {
3362 	u64 eptp = VMX_EPTP_MT_WB;
3363 
3364 	eptp |= (root_level == 5) ? VMX_EPTP_PWL_5 : VMX_EPTP_PWL_4;
3365 
3366 	if (enable_ept_ad_bits &&
3367 	    (!is_guest_mode(vcpu) || nested_ept_ad_enabled(vcpu)))
3368 		eptp |= VMX_EPTP_AD_ENABLE_BIT;
3369 	eptp |= root_hpa;
3370 
3371 	return eptp;
3372 }
3373 
3374 static void vmx_load_mmu_pgd(struct kvm_vcpu *vcpu, hpa_t root_hpa,
3375 			     int root_level)
3376 {
3377 	struct kvm *kvm = vcpu->kvm;
3378 	bool update_guest_cr3 = true;
3379 	unsigned long guest_cr3;
3380 	u64 eptp;
3381 
3382 	if (enable_ept) {
3383 		eptp = construct_eptp(vcpu, root_hpa, root_level);
3384 		vmcs_write64(EPT_POINTER, eptp);
3385 
3386 		hv_track_root_tdp(vcpu, root_hpa);
3387 
3388 		if (!enable_unrestricted_guest && !is_paging(vcpu))
3389 			guest_cr3 = to_kvm_vmx(kvm)->ept_identity_map_addr;
3390 		else if (kvm_register_is_dirty(vcpu, VCPU_EXREG_CR3))
3391 			guest_cr3 = vcpu->arch.cr3;
3392 		else /* vmcs.GUEST_CR3 is already up-to-date. */
3393 			update_guest_cr3 = false;
3394 		vmx_ept_load_pdptrs(vcpu);
3395 	} else {
3396 		guest_cr3 = root_hpa | kvm_get_active_pcid(vcpu);
3397 	}
3398 
3399 	if (update_guest_cr3)
3400 		vmcs_writel(GUEST_CR3, guest_cr3);
3401 }
3402 
3403 
3404 static bool vmx_is_valid_cr4(struct kvm_vcpu *vcpu, unsigned long cr4)
3405 {
3406 	/*
3407 	 * We operate under the default treatment of SMM, so VMX cannot be
3408 	 * enabled under SMM.  Note, whether or not VMXE is allowed at all,
3409 	 * i.e. is a reserved bit, is handled by common x86 code.
3410 	 */
3411 	if ((cr4 & X86_CR4_VMXE) && is_smm(vcpu))
3412 		return false;
3413 
3414 	if (to_vmx(vcpu)->nested.vmxon && !nested_cr4_valid(vcpu, cr4))
3415 		return false;
3416 
3417 	return true;
3418 }
3419 
3420 void vmx_set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4)
3421 {
3422 	unsigned long old_cr4 = kvm_read_cr4(vcpu);
3423 	struct vcpu_vmx *vmx = to_vmx(vcpu);
3424 	unsigned long hw_cr4;
3425 
3426 	/*
3427 	 * Pass through host's Machine Check Enable value to hw_cr4, which
3428 	 * is in force while we are in guest mode.  Do not let guests control
3429 	 * this bit, even if host CR4.MCE == 0.
3430 	 */
3431 	hw_cr4 = (cr4_read_shadow() & X86_CR4_MCE) | (cr4 & ~X86_CR4_MCE);
3432 	if (enable_unrestricted_guest)
3433 		hw_cr4 |= KVM_VM_CR4_ALWAYS_ON_UNRESTRICTED_GUEST;
3434 	else if (vmx->rmode.vm86_active)
3435 		hw_cr4 |= KVM_RMODE_VM_CR4_ALWAYS_ON;
3436 	else
3437 		hw_cr4 |= KVM_PMODE_VM_CR4_ALWAYS_ON;
3438 
3439 	if (vmx_umip_emulated()) {
3440 		if (cr4 & X86_CR4_UMIP) {
3441 			secondary_exec_controls_setbit(vmx, SECONDARY_EXEC_DESC);
3442 			hw_cr4 &= ~X86_CR4_UMIP;
3443 		} else if (!is_guest_mode(vcpu) ||
3444 			!nested_cpu_has2(get_vmcs12(vcpu), SECONDARY_EXEC_DESC)) {
3445 			secondary_exec_controls_clearbit(vmx, SECONDARY_EXEC_DESC);
3446 		}
3447 	}
3448 
3449 	vcpu->arch.cr4 = cr4;
3450 	kvm_register_mark_available(vcpu, VCPU_EXREG_CR4);
3451 
3452 	if (!enable_unrestricted_guest) {
3453 		if (enable_ept) {
3454 			if (!is_paging(vcpu)) {
3455 				hw_cr4 &= ~X86_CR4_PAE;
3456 				hw_cr4 |= X86_CR4_PSE;
3457 			} else if (!(cr4 & X86_CR4_PAE)) {
3458 				hw_cr4 &= ~X86_CR4_PAE;
3459 			}
3460 		}
3461 
3462 		/*
3463 		 * SMEP/SMAP/PKU is disabled if CPU is in non-paging mode in
3464 		 * hardware.  To emulate this behavior, SMEP/SMAP/PKU needs
3465 		 * to be manually disabled when guest switches to non-paging
3466 		 * mode.
3467 		 *
3468 		 * If !enable_unrestricted_guest, the CPU is always running
3469 		 * with CR0.PG=1 and CR4 needs to be modified.
3470 		 * If enable_unrestricted_guest, the CPU automatically
3471 		 * disables SMEP/SMAP/PKU when the guest sets CR0.PG=0.
3472 		 */
3473 		if (!is_paging(vcpu))
3474 			hw_cr4 &= ~(X86_CR4_SMEP | X86_CR4_SMAP | X86_CR4_PKE);
3475 	}
3476 
3477 	vmcs_writel(CR4_READ_SHADOW, cr4);
3478 	vmcs_writel(GUEST_CR4, hw_cr4);
3479 
3480 	if ((cr4 ^ old_cr4) & (X86_CR4_OSXSAVE | X86_CR4_PKE))
3481 		kvm_update_cpuid_runtime(vcpu);
3482 }
3483 
3484 void vmx_get_segment(struct kvm_vcpu *vcpu, struct kvm_segment *var, int seg)
3485 {
3486 	struct vcpu_vmx *vmx = to_vmx(vcpu);
3487 	u32 ar;
3488 
3489 	if (vmx->rmode.vm86_active && seg != VCPU_SREG_LDTR) {
3490 		*var = vmx->rmode.segs[seg];
3491 		if (seg == VCPU_SREG_TR
3492 		    || var->selector == vmx_read_guest_seg_selector(vmx, seg))
3493 			return;
3494 		var->base = vmx_read_guest_seg_base(vmx, seg);
3495 		var->selector = vmx_read_guest_seg_selector(vmx, seg);
3496 		return;
3497 	}
3498 	var->base = vmx_read_guest_seg_base(vmx, seg);
3499 	var->limit = vmx_read_guest_seg_limit(vmx, seg);
3500 	var->selector = vmx_read_guest_seg_selector(vmx, seg);
3501 	ar = vmx_read_guest_seg_ar(vmx, seg);
3502 	var->unusable = (ar >> 16) & 1;
3503 	var->type = ar & 15;
3504 	var->s = (ar >> 4) & 1;
3505 	var->dpl = (ar >> 5) & 3;
3506 	/*
3507 	 * Some userspaces do not preserve unusable property. Since usable
3508 	 * segment has to be present according to VMX spec we can use present
3509 	 * property to amend userspace bug by making unusable segment always
3510 	 * nonpresent. vmx_segment_access_rights() already marks nonpresent
3511 	 * segment as unusable.
3512 	 */
3513 	var->present = !var->unusable;
3514 	var->avl = (ar >> 12) & 1;
3515 	var->l = (ar >> 13) & 1;
3516 	var->db = (ar >> 14) & 1;
3517 	var->g = (ar >> 15) & 1;
3518 }
3519 
3520 static u64 vmx_get_segment_base(struct kvm_vcpu *vcpu, int seg)
3521 {
3522 	struct kvm_segment s;
3523 
3524 	if (to_vmx(vcpu)->rmode.vm86_active) {
3525 		vmx_get_segment(vcpu, &s, seg);
3526 		return s.base;
3527 	}
3528 	return vmx_read_guest_seg_base(to_vmx(vcpu), seg);
3529 }
3530 
3531 int vmx_get_cpl(struct kvm_vcpu *vcpu)
3532 {
3533 	struct vcpu_vmx *vmx = to_vmx(vcpu);
3534 
3535 	if (unlikely(vmx->rmode.vm86_active))
3536 		return 0;
3537 	else {
3538 		int ar = vmx_read_guest_seg_ar(vmx, VCPU_SREG_SS);
3539 		return VMX_AR_DPL(ar);
3540 	}
3541 }
3542 
3543 static u32 vmx_segment_access_rights(struct kvm_segment *var)
3544 {
3545 	u32 ar;
3546 
3547 	ar = var->type & 15;
3548 	ar |= (var->s & 1) << 4;
3549 	ar |= (var->dpl & 3) << 5;
3550 	ar |= (var->present & 1) << 7;
3551 	ar |= (var->avl & 1) << 12;
3552 	ar |= (var->l & 1) << 13;
3553 	ar |= (var->db & 1) << 14;
3554 	ar |= (var->g & 1) << 15;
3555 	ar |= (var->unusable || !var->present) << 16;
3556 
3557 	return ar;
3558 }
3559 
3560 void __vmx_set_segment(struct kvm_vcpu *vcpu, struct kvm_segment *var, int seg)
3561 {
3562 	struct vcpu_vmx *vmx = to_vmx(vcpu);
3563 	const struct kvm_vmx_segment_field *sf = &kvm_vmx_segment_fields[seg];
3564 
3565 	vmx_segment_cache_clear(vmx);
3566 
3567 	if (vmx->rmode.vm86_active && seg != VCPU_SREG_LDTR) {
3568 		vmx->rmode.segs[seg] = *var;
3569 		if (seg == VCPU_SREG_TR)
3570 			vmcs_write16(sf->selector, var->selector);
3571 		else if (var->s)
3572 			fix_rmode_seg(seg, &vmx->rmode.segs[seg]);
3573 		return;
3574 	}
3575 
3576 	vmcs_writel(sf->base, var->base);
3577 	vmcs_write32(sf->limit, var->limit);
3578 	vmcs_write16(sf->selector, var->selector);
3579 
3580 	/*
3581 	 *   Fix the "Accessed" bit in AR field of segment registers for older
3582 	 * qemu binaries.
3583 	 *   IA32 arch specifies that at the time of processor reset the
3584 	 * "Accessed" bit in the AR field of segment registers is 1. And qemu
3585 	 * is setting it to 0 in the userland code. This causes invalid guest
3586 	 * state vmexit when "unrestricted guest" mode is turned on.
3587 	 *    Fix for this setup issue in cpu_reset is being pushed in the qemu
3588 	 * tree. Newer qemu binaries with that qemu fix would not need this
3589 	 * kvm hack.
3590 	 */
3591 	if (is_unrestricted_guest(vcpu) && (seg != VCPU_SREG_LDTR))
3592 		var->type |= 0x1; /* Accessed */
3593 
3594 	vmcs_write32(sf->ar_bytes, vmx_segment_access_rights(var));
3595 }
3596 
3597 static void vmx_set_segment(struct kvm_vcpu *vcpu, struct kvm_segment *var, int seg)
3598 {
3599 	__vmx_set_segment(vcpu, var, seg);
3600 
3601 	to_vmx(vcpu)->emulation_required = vmx_emulation_required(vcpu);
3602 }
3603 
3604 static void vmx_get_cs_db_l_bits(struct kvm_vcpu *vcpu, int *db, int *l)
3605 {
3606 	u32 ar = vmx_read_guest_seg_ar(to_vmx(vcpu), VCPU_SREG_CS);
3607 
3608 	*db = (ar >> 14) & 1;
3609 	*l = (ar >> 13) & 1;
3610 }
3611 
3612 static void vmx_get_idt(struct kvm_vcpu *vcpu, struct desc_ptr *dt)
3613 {
3614 	dt->size = vmcs_read32(GUEST_IDTR_LIMIT);
3615 	dt->address = vmcs_readl(GUEST_IDTR_BASE);
3616 }
3617 
3618 static void vmx_set_idt(struct kvm_vcpu *vcpu, struct desc_ptr *dt)
3619 {
3620 	vmcs_write32(GUEST_IDTR_LIMIT, dt->size);
3621 	vmcs_writel(GUEST_IDTR_BASE, dt->address);
3622 }
3623 
3624 static void vmx_get_gdt(struct kvm_vcpu *vcpu, struct desc_ptr *dt)
3625 {
3626 	dt->size = vmcs_read32(GUEST_GDTR_LIMIT);
3627 	dt->address = vmcs_readl(GUEST_GDTR_BASE);
3628 }
3629 
3630 static void vmx_set_gdt(struct kvm_vcpu *vcpu, struct desc_ptr *dt)
3631 {
3632 	vmcs_write32(GUEST_GDTR_LIMIT, dt->size);
3633 	vmcs_writel(GUEST_GDTR_BASE, dt->address);
3634 }
3635 
3636 static bool rmode_segment_valid(struct kvm_vcpu *vcpu, int seg)
3637 {
3638 	struct kvm_segment var;
3639 	u32 ar;
3640 
3641 	vmx_get_segment(vcpu, &var, seg);
3642 	var.dpl = 0x3;
3643 	if (seg == VCPU_SREG_CS)
3644 		var.type = 0x3;
3645 	ar = vmx_segment_access_rights(&var);
3646 
3647 	if (var.base != (var.selector << 4))
3648 		return false;
3649 	if (var.limit != 0xffff)
3650 		return false;
3651 	if (ar != 0xf3)
3652 		return false;
3653 
3654 	return true;
3655 }
3656 
3657 static bool code_segment_valid(struct kvm_vcpu *vcpu)
3658 {
3659 	struct kvm_segment cs;
3660 	unsigned int cs_rpl;
3661 
3662 	vmx_get_segment(vcpu, &cs, VCPU_SREG_CS);
3663 	cs_rpl = cs.selector & SEGMENT_RPL_MASK;
3664 
3665 	if (cs.unusable)
3666 		return false;
3667 	if (~cs.type & (VMX_AR_TYPE_CODE_MASK|VMX_AR_TYPE_ACCESSES_MASK))
3668 		return false;
3669 	if (!cs.s)
3670 		return false;
3671 	if (cs.type & VMX_AR_TYPE_WRITEABLE_MASK) {
3672 		if (cs.dpl > cs_rpl)
3673 			return false;
3674 	} else {
3675 		if (cs.dpl != cs_rpl)
3676 			return false;
3677 	}
3678 	if (!cs.present)
3679 		return false;
3680 
3681 	/* TODO: Add Reserved field check, this'll require a new member in the kvm_segment_field structure */
3682 	return true;
3683 }
3684 
3685 static bool stack_segment_valid(struct kvm_vcpu *vcpu)
3686 {
3687 	struct kvm_segment ss;
3688 	unsigned int ss_rpl;
3689 
3690 	vmx_get_segment(vcpu, &ss, VCPU_SREG_SS);
3691 	ss_rpl = ss.selector & SEGMENT_RPL_MASK;
3692 
3693 	if (ss.unusable)
3694 		return true;
3695 	if (ss.type != 3 && ss.type != 7)
3696 		return false;
3697 	if (!ss.s)
3698 		return false;
3699 	if (ss.dpl != ss_rpl) /* DPL != RPL */
3700 		return false;
3701 	if (!ss.present)
3702 		return false;
3703 
3704 	return true;
3705 }
3706 
3707 static bool data_segment_valid(struct kvm_vcpu *vcpu, int seg)
3708 {
3709 	struct kvm_segment var;
3710 	unsigned int rpl;
3711 
3712 	vmx_get_segment(vcpu, &var, seg);
3713 	rpl = var.selector & SEGMENT_RPL_MASK;
3714 
3715 	if (var.unusable)
3716 		return true;
3717 	if (!var.s)
3718 		return false;
3719 	if (!var.present)
3720 		return false;
3721 	if (~var.type & (VMX_AR_TYPE_CODE_MASK|VMX_AR_TYPE_WRITEABLE_MASK)) {
3722 		if (var.dpl < rpl) /* DPL < RPL */
3723 			return false;
3724 	}
3725 
3726 	/* TODO: Add other members to kvm_segment_field to allow checking for other access
3727 	 * rights flags
3728 	 */
3729 	return true;
3730 }
3731 
3732 static bool tr_valid(struct kvm_vcpu *vcpu)
3733 {
3734 	struct kvm_segment tr;
3735 
3736 	vmx_get_segment(vcpu, &tr, VCPU_SREG_TR);
3737 
3738 	if (tr.unusable)
3739 		return false;
3740 	if (tr.selector & SEGMENT_TI_MASK)	/* TI = 1 */
3741 		return false;
3742 	if (tr.type != 3 && tr.type != 11) /* TODO: Check if guest is in IA32e mode */
3743 		return false;
3744 	if (!tr.present)
3745 		return false;
3746 
3747 	return true;
3748 }
3749 
3750 static bool ldtr_valid(struct kvm_vcpu *vcpu)
3751 {
3752 	struct kvm_segment ldtr;
3753 
3754 	vmx_get_segment(vcpu, &ldtr, VCPU_SREG_LDTR);
3755 
3756 	if (ldtr.unusable)
3757 		return true;
3758 	if (ldtr.selector & SEGMENT_TI_MASK)	/* TI = 1 */
3759 		return false;
3760 	if (ldtr.type != 2)
3761 		return false;
3762 	if (!ldtr.present)
3763 		return false;
3764 
3765 	return true;
3766 }
3767 
3768 static bool cs_ss_rpl_check(struct kvm_vcpu *vcpu)
3769 {
3770 	struct kvm_segment cs, ss;
3771 
3772 	vmx_get_segment(vcpu, &cs, VCPU_SREG_CS);
3773 	vmx_get_segment(vcpu, &ss, VCPU_SREG_SS);
3774 
3775 	return ((cs.selector & SEGMENT_RPL_MASK) ==
3776 		 (ss.selector & SEGMENT_RPL_MASK));
3777 }
3778 
3779 /*
3780  * Check if guest state is valid. Returns true if valid, false if
3781  * not.
3782  * We assume that registers are always usable
3783  */
3784 bool __vmx_guest_state_valid(struct kvm_vcpu *vcpu)
3785 {
3786 	/* real mode guest state checks */
3787 	if (!is_protmode(vcpu) || (vmx_get_rflags(vcpu) & X86_EFLAGS_VM)) {
3788 		if (!rmode_segment_valid(vcpu, VCPU_SREG_CS))
3789 			return false;
3790 		if (!rmode_segment_valid(vcpu, VCPU_SREG_SS))
3791 			return false;
3792 		if (!rmode_segment_valid(vcpu, VCPU_SREG_DS))
3793 			return false;
3794 		if (!rmode_segment_valid(vcpu, VCPU_SREG_ES))
3795 			return false;
3796 		if (!rmode_segment_valid(vcpu, VCPU_SREG_FS))
3797 			return false;
3798 		if (!rmode_segment_valid(vcpu, VCPU_SREG_GS))
3799 			return false;
3800 	} else {
3801 	/* protected mode guest state checks */
3802 		if (!cs_ss_rpl_check(vcpu))
3803 			return false;
3804 		if (!code_segment_valid(vcpu))
3805 			return false;
3806 		if (!stack_segment_valid(vcpu))
3807 			return false;
3808 		if (!data_segment_valid(vcpu, VCPU_SREG_DS))
3809 			return false;
3810 		if (!data_segment_valid(vcpu, VCPU_SREG_ES))
3811 			return false;
3812 		if (!data_segment_valid(vcpu, VCPU_SREG_FS))
3813 			return false;
3814 		if (!data_segment_valid(vcpu, VCPU_SREG_GS))
3815 			return false;
3816 		if (!tr_valid(vcpu))
3817 			return false;
3818 		if (!ldtr_valid(vcpu))
3819 			return false;
3820 	}
3821 	/* TODO:
3822 	 * - Add checks on RIP
3823 	 * - Add checks on RFLAGS
3824 	 */
3825 
3826 	return true;
3827 }
3828 
3829 static int init_rmode_tss(struct kvm *kvm, void __user *ua)
3830 {
3831 	const void *zero_page = (const void *) __va(page_to_phys(ZERO_PAGE(0)));
3832 	u16 data;
3833 	int i;
3834 
3835 	for (i = 0; i < 3; i++) {
3836 		if (__copy_to_user(ua + PAGE_SIZE * i, zero_page, PAGE_SIZE))
3837 			return -EFAULT;
3838 	}
3839 
3840 	data = TSS_BASE_SIZE + TSS_REDIRECTION_SIZE;
3841 	if (__copy_to_user(ua + TSS_IOPB_BASE_OFFSET, &data, sizeof(u16)))
3842 		return -EFAULT;
3843 
3844 	data = ~0;
3845 	if (__copy_to_user(ua + RMODE_TSS_SIZE - 1, &data, sizeof(u8)))
3846 		return -EFAULT;
3847 
3848 	return 0;
3849 }
3850 
3851 static int init_rmode_identity_map(struct kvm *kvm)
3852 {
3853 	struct kvm_vmx *kvm_vmx = to_kvm_vmx(kvm);
3854 	int i, r = 0;
3855 	void __user *uaddr;
3856 	u32 tmp;
3857 
3858 	/* Protect kvm_vmx->ept_identity_pagetable_done. */
3859 	mutex_lock(&kvm->slots_lock);
3860 
3861 	if (likely(kvm_vmx->ept_identity_pagetable_done))
3862 		goto out;
3863 
3864 	if (!kvm_vmx->ept_identity_map_addr)
3865 		kvm_vmx->ept_identity_map_addr = VMX_EPT_IDENTITY_PAGETABLE_ADDR;
3866 
3867 	uaddr = __x86_set_memory_region(kvm,
3868 					IDENTITY_PAGETABLE_PRIVATE_MEMSLOT,
3869 					kvm_vmx->ept_identity_map_addr,
3870 					PAGE_SIZE);
3871 	if (IS_ERR(uaddr)) {
3872 		r = PTR_ERR(uaddr);
3873 		goto out;
3874 	}
3875 
3876 	/* Set up identity-mapping pagetable for EPT in real mode */
3877 	for (i = 0; i < (PAGE_SIZE / sizeof(tmp)); i++) {
3878 		tmp = (i << 22) + (_PAGE_PRESENT | _PAGE_RW | _PAGE_USER |
3879 			_PAGE_ACCESSED | _PAGE_DIRTY | _PAGE_PSE);
3880 		if (__copy_to_user(uaddr + i * sizeof(tmp), &tmp, sizeof(tmp))) {
3881 			r = -EFAULT;
3882 			goto out;
3883 		}
3884 	}
3885 	kvm_vmx->ept_identity_pagetable_done = true;
3886 
3887 out:
3888 	mutex_unlock(&kvm->slots_lock);
3889 	return r;
3890 }
3891 
3892 static void seg_setup(int seg)
3893 {
3894 	const struct kvm_vmx_segment_field *sf = &kvm_vmx_segment_fields[seg];
3895 	unsigned int ar;
3896 
3897 	vmcs_write16(sf->selector, 0);
3898 	vmcs_writel(sf->base, 0);
3899 	vmcs_write32(sf->limit, 0xffff);
3900 	ar = 0x93;
3901 	if (seg == VCPU_SREG_CS)
3902 		ar |= 0x08; /* code segment */
3903 
3904 	vmcs_write32(sf->ar_bytes, ar);
3905 }
3906 
3907 int allocate_vpid(void)
3908 {
3909 	int vpid;
3910 
3911 	if (!enable_vpid)
3912 		return 0;
3913 	spin_lock(&vmx_vpid_lock);
3914 	vpid = find_first_zero_bit(vmx_vpid_bitmap, VMX_NR_VPIDS);
3915 	if (vpid < VMX_NR_VPIDS)
3916 		__set_bit(vpid, vmx_vpid_bitmap);
3917 	else
3918 		vpid = 0;
3919 	spin_unlock(&vmx_vpid_lock);
3920 	return vpid;
3921 }
3922 
3923 void free_vpid(int vpid)
3924 {
3925 	if (!enable_vpid || vpid == 0)
3926 		return;
3927 	spin_lock(&vmx_vpid_lock);
3928 	__clear_bit(vpid, vmx_vpid_bitmap);
3929 	spin_unlock(&vmx_vpid_lock);
3930 }
3931 
3932 static void vmx_msr_bitmap_l01_changed(struct vcpu_vmx *vmx)
3933 {
3934 	/*
3935 	 * When KVM is a nested hypervisor on top of Hyper-V and uses
3936 	 * 'Enlightened MSR Bitmap' feature L0 needs to know that MSR
3937 	 * bitmap has changed.
3938 	 */
3939 	if (kvm_is_using_evmcs()) {
3940 		struct hv_enlightened_vmcs *evmcs = (void *)vmx->vmcs01.vmcs;
3941 
3942 		if (evmcs->hv_enlightenments_control.msr_bitmap)
3943 			evmcs->hv_clean_fields &=
3944 				~HV_VMX_ENLIGHTENED_CLEAN_FIELD_MSR_BITMAP;
3945 	}
3946 
3947 	vmx->nested.force_msr_bitmap_recalc = true;
3948 }
3949 
3950 void vmx_disable_intercept_for_msr(struct kvm_vcpu *vcpu, u32 msr, int type)
3951 {
3952 	struct vcpu_vmx *vmx = to_vmx(vcpu);
3953 	unsigned long *msr_bitmap = vmx->vmcs01.msr_bitmap;
3954 
3955 	if (!cpu_has_vmx_msr_bitmap())
3956 		return;
3957 
3958 	vmx_msr_bitmap_l01_changed(vmx);
3959 
3960 	/*
3961 	 * Mark the desired intercept state in shadow bitmap, this is needed
3962 	 * for resync when the MSR filters change.
3963 	*/
3964 	if (is_valid_passthrough_msr(msr)) {
3965 		int idx = possible_passthrough_msr_slot(msr);
3966 
3967 		if (idx != -ENOENT) {
3968 			if (type & MSR_TYPE_R)
3969 				clear_bit(idx, vmx->shadow_msr_intercept.read);
3970 			if (type & MSR_TYPE_W)
3971 				clear_bit(idx, vmx->shadow_msr_intercept.write);
3972 		}
3973 	}
3974 
3975 	if ((type & MSR_TYPE_R) &&
3976 	    !kvm_msr_allowed(vcpu, msr, KVM_MSR_FILTER_READ)) {
3977 		vmx_set_msr_bitmap_read(msr_bitmap, msr);
3978 		type &= ~MSR_TYPE_R;
3979 	}
3980 
3981 	if ((type & MSR_TYPE_W) &&
3982 	    !kvm_msr_allowed(vcpu, msr, KVM_MSR_FILTER_WRITE)) {
3983 		vmx_set_msr_bitmap_write(msr_bitmap, msr);
3984 		type &= ~MSR_TYPE_W;
3985 	}
3986 
3987 	if (type & MSR_TYPE_R)
3988 		vmx_clear_msr_bitmap_read(msr_bitmap, msr);
3989 
3990 	if (type & MSR_TYPE_W)
3991 		vmx_clear_msr_bitmap_write(msr_bitmap, msr);
3992 }
3993 
3994 void vmx_enable_intercept_for_msr(struct kvm_vcpu *vcpu, u32 msr, int type)
3995 {
3996 	struct vcpu_vmx *vmx = to_vmx(vcpu);
3997 	unsigned long *msr_bitmap = vmx->vmcs01.msr_bitmap;
3998 
3999 	if (!cpu_has_vmx_msr_bitmap())
4000 		return;
4001 
4002 	vmx_msr_bitmap_l01_changed(vmx);
4003 
4004 	/*
4005 	 * Mark the desired intercept state in shadow bitmap, this is needed
4006 	 * for resync when the MSR filter changes.
4007 	*/
4008 	if (is_valid_passthrough_msr(msr)) {
4009 		int idx = possible_passthrough_msr_slot(msr);
4010 
4011 		if (idx != -ENOENT) {
4012 			if (type & MSR_TYPE_R)
4013 				set_bit(idx, vmx->shadow_msr_intercept.read);
4014 			if (type & MSR_TYPE_W)
4015 				set_bit(idx, vmx->shadow_msr_intercept.write);
4016 		}
4017 	}
4018 
4019 	if (type & MSR_TYPE_R)
4020 		vmx_set_msr_bitmap_read(msr_bitmap, msr);
4021 
4022 	if (type & MSR_TYPE_W)
4023 		vmx_set_msr_bitmap_write(msr_bitmap, msr);
4024 }
4025 
4026 static void vmx_update_msr_bitmap_x2apic(struct kvm_vcpu *vcpu)
4027 {
4028 	/*
4029 	 * x2APIC indices for 64-bit accesses into the RDMSR and WRMSR halves
4030 	 * of the MSR bitmap.  KVM emulates APIC registers up through 0x3f0,
4031 	 * i.e. MSR 0x83f, and so only needs to dynamically manipulate 64 bits.
4032 	 */
4033 	const int read_idx = APIC_BASE_MSR / BITS_PER_LONG_LONG;
4034 	const int write_idx = read_idx + (0x800 / sizeof(u64));
4035 	struct vcpu_vmx *vmx = to_vmx(vcpu);
4036 	u64 *msr_bitmap = (u64 *)vmx->vmcs01.msr_bitmap;
4037 	u8 mode;
4038 
4039 	if (!cpu_has_vmx_msr_bitmap() || WARN_ON_ONCE(!lapic_in_kernel(vcpu)))
4040 		return;
4041 
4042 	if (cpu_has_secondary_exec_ctrls() &&
4043 	    (secondary_exec_controls_get(vmx) &
4044 	     SECONDARY_EXEC_VIRTUALIZE_X2APIC_MODE)) {
4045 		mode = MSR_BITMAP_MODE_X2APIC;
4046 		if (enable_apicv && kvm_vcpu_apicv_active(vcpu))
4047 			mode |= MSR_BITMAP_MODE_X2APIC_APICV;
4048 	} else {
4049 		mode = 0;
4050 	}
4051 
4052 	if (mode == vmx->x2apic_msr_bitmap_mode)
4053 		return;
4054 
4055 	vmx->x2apic_msr_bitmap_mode = mode;
4056 
4057 	/*
4058 	 * Reset the bitmap for MSRs 0x800 - 0x83f.  Leave AMD's uber-extended
4059 	 * registers (0x840 and above) intercepted, KVM doesn't support them.
4060 	 * Intercept all writes by default and poke holes as needed.  Pass
4061 	 * through reads for all valid registers by default in x2APIC+APICv
4062 	 * mode, only the current timer count needs on-demand emulation by KVM.
4063 	 */
4064 	if (mode & MSR_BITMAP_MODE_X2APIC_APICV)
4065 		msr_bitmap[read_idx] = ~kvm_lapic_readable_reg_mask(vcpu->arch.apic);
4066 	else
4067 		msr_bitmap[read_idx] = ~0ull;
4068 	msr_bitmap[write_idx] = ~0ull;
4069 
4070 	/*
4071 	 * TPR reads and writes can be virtualized even if virtual interrupt
4072 	 * delivery is not in use.
4073 	 */
4074 	vmx_set_intercept_for_msr(vcpu, X2APIC_MSR(APIC_TASKPRI), MSR_TYPE_RW,
4075 				  !(mode & MSR_BITMAP_MODE_X2APIC));
4076 
4077 	if (mode & MSR_BITMAP_MODE_X2APIC_APICV) {
4078 		vmx_enable_intercept_for_msr(vcpu, X2APIC_MSR(APIC_TMCCT), MSR_TYPE_RW);
4079 		vmx_disable_intercept_for_msr(vcpu, X2APIC_MSR(APIC_EOI), MSR_TYPE_W);
4080 		vmx_disable_intercept_for_msr(vcpu, X2APIC_MSR(APIC_SELF_IPI), MSR_TYPE_W);
4081 		if (enable_ipiv)
4082 			vmx_disable_intercept_for_msr(vcpu, X2APIC_MSR(APIC_ICR), MSR_TYPE_RW);
4083 	}
4084 }
4085 
4086 void pt_update_intercept_for_msr(struct kvm_vcpu *vcpu)
4087 {
4088 	struct vcpu_vmx *vmx = to_vmx(vcpu);
4089 	bool flag = !(vmx->pt_desc.guest.ctl & RTIT_CTL_TRACEEN);
4090 	u32 i;
4091 
4092 	vmx_set_intercept_for_msr(vcpu, MSR_IA32_RTIT_STATUS, MSR_TYPE_RW, flag);
4093 	vmx_set_intercept_for_msr(vcpu, MSR_IA32_RTIT_OUTPUT_BASE, MSR_TYPE_RW, flag);
4094 	vmx_set_intercept_for_msr(vcpu, MSR_IA32_RTIT_OUTPUT_MASK, MSR_TYPE_RW, flag);
4095 	vmx_set_intercept_for_msr(vcpu, MSR_IA32_RTIT_CR3_MATCH, MSR_TYPE_RW, flag);
4096 	for (i = 0; i < vmx->pt_desc.num_address_ranges; i++) {
4097 		vmx_set_intercept_for_msr(vcpu, MSR_IA32_RTIT_ADDR0_A + i * 2, MSR_TYPE_RW, flag);
4098 		vmx_set_intercept_for_msr(vcpu, MSR_IA32_RTIT_ADDR0_B + i * 2, MSR_TYPE_RW, flag);
4099 	}
4100 }
4101 
4102 static bool vmx_guest_apic_has_interrupt(struct kvm_vcpu *vcpu)
4103 {
4104 	struct vcpu_vmx *vmx = to_vmx(vcpu);
4105 	void *vapic_page;
4106 	u32 vppr;
4107 	int rvi;
4108 
4109 	if (WARN_ON_ONCE(!is_guest_mode(vcpu)) ||
4110 		!nested_cpu_has_vid(get_vmcs12(vcpu)) ||
4111 		WARN_ON_ONCE(!vmx->nested.virtual_apic_map.gfn))
4112 		return false;
4113 
4114 	rvi = vmx_get_rvi();
4115 
4116 	vapic_page = vmx->nested.virtual_apic_map.hva;
4117 	vppr = *((u32 *)(vapic_page + APIC_PROCPRI));
4118 
4119 	return ((rvi & 0xf0) > (vppr & 0xf0));
4120 }
4121 
4122 static void vmx_msr_filter_changed(struct kvm_vcpu *vcpu)
4123 {
4124 	struct vcpu_vmx *vmx = to_vmx(vcpu);
4125 	u32 i;
4126 
4127 	/*
4128 	 * Redo intercept permissions for MSRs that KVM is passing through to
4129 	 * the guest.  Disabling interception will check the new MSR filter and
4130 	 * ensure that KVM enables interception if usersepace wants to filter
4131 	 * the MSR.  MSRs that KVM is already intercepting don't need to be
4132 	 * refreshed since KVM is going to intercept them regardless of what
4133 	 * userspace wants.
4134 	 */
4135 	for (i = 0; i < ARRAY_SIZE(vmx_possible_passthrough_msrs); i++) {
4136 		u32 msr = vmx_possible_passthrough_msrs[i];
4137 
4138 		if (!test_bit(i, vmx->shadow_msr_intercept.read))
4139 			vmx_disable_intercept_for_msr(vcpu, msr, MSR_TYPE_R);
4140 
4141 		if (!test_bit(i, vmx->shadow_msr_intercept.write))
4142 			vmx_disable_intercept_for_msr(vcpu, msr, MSR_TYPE_W);
4143 	}
4144 
4145 	/* PT MSRs can be passed through iff PT is exposed to the guest. */
4146 	if (vmx_pt_mode_is_host_guest())
4147 		pt_update_intercept_for_msr(vcpu);
4148 }
4149 
4150 static inline void kvm_vcpu_trigger_posted_interrupt(struct kvm_vcpu *vcpu,
4151 						     int pi_vec)
4152 {
4153 #ifdef CONFIG_SMP
4154 	if (vcpu->mode == IN_GUEST_MODE) {
4155 		/*
4156 		 * The vector of the virtual has already been set in the PIR.
4157 		 * Send a notification event to deliver the virtual interrupt
4158 		 * unless the vCPU is the currently running vCPU, i.e. the
4159 		 * event is being sent from a fastpath VM-Exit handler, in
4160 		 * which case the PIR will be synced to the vIRR before
4161 		 * re-entering the guest.
4162 		 *
4163 		 * When the target is not the running vCPU, the following
4164 		 * possibilities emerge:
4165 		 *
4166 		 * Case 1: vCPU stays in non-root mode. Sending a notification
4167 		 * event posts the interrupt to the vCPU.
4168 		 *
4169 		 * Case 2: vCPU exits to root mode and is still runnable. The
4170 		 * PIR will be synced to the vIRR before re-entering the guest.
4171 		 * Sending a notification event is ok as the host IRQ handler
4172 		 * will ignore the spurious event.
4173 		 *
4174 		 * Case 3: vCPU exits to root mode and is blocked. vcpu_block()
4175 		 * has already synced PIR to vIRR and never blocks the vCPU if
4176 		 * the vIRR is not empty. Therefore, a blocked vCPU here does
4177 		 * not wait for any requested interrupts in PIR, and sending a
4178 		 * notification event also results in a benign, spurious event.
4179 		 */
4180 
4181 		if (vcpu != kvm_get_running_vcpu())
4182 			__apic_send_IPI_mask(get_cpu_mask(vcpu->cpu), pi_vec);
4183 		return;
4184 	}
4185 #endif
4186 	/*
4187 	 * The vCPU isn't in the guest; wake the vCPU in case it is blocking,
4188 	 * otherwise do nothing as KVM will grab the highest priority pending
4189 	 * IRQ via ->sync_pir_to_irr() in vcpu_enter_guest().
4190 	 */
4191 	kvm_vcpu_wake_up(vcpu);
4192 }
4193 
4194 static int vmx_deliver_nested_posted_interrupt(struct kvm_vcpu *vcpu,
4195 						int vector)
4196 {
4197 	struct vcpu_vmx *vmx = to_vmx(vcpu);
4198 
4199 	if (is_guest_mode(vcpu) &&
4200 	    vector == vmx->nested.posted_intr_nv) {
4201 		/*
4202 		 * If a posted intr is not recognized by hardware,
4203 		 * we will accomplish it in the next vmentry.
4204 		 */
4205 		vmx->nested.pi_pending = true;
4206 		kvm_make_request(KVM_REQ_EVENT, vcpu);
4207 
4208 		/*
4209 		 * This pairs with the smp_mb_*() after setting vcpu->mode in
4210 		 * vcpu_enter_guest() to guarantee the vCPU sees the event
4211 		 * request if triggering a posted interrupt "fails" because
4212 		 * vcpu->mode != IN_GUEST_MODE.  The extra barrier is needed as
4213 		 * the smb_wmb() in kvm_make_request() only ensures everything
4214 		 * done before making the request is visible when the request
4215 		 * is visible, it doesn't ensure ordering between the store to
4216 		 * vcpu->requests and the load from vcpu->mode.
4217 		 */
4218 		smp_mb__after_atomic();
4219 
4220 		/* the PIR and ON have been set by L1. */
4221 		kvm_vcpu_trigger_posted_interrupt(vcpu, POSTED_INTR_NESTED_VECTOR);
4222 		return 0;
4223 	}
4224 	return -1;
4225 }
4226 /*
4227  * Send interrupt to vcpu via posted interrupt way.
4228  * 1. If target vcpu is running(non-root mode), send posted interrupt
4229  * notification to vcpu and hardware will sync PIR to vIRR atomically.
4230  * 2. If target vcpu isn't running(root mode), kick it to pick up the
4231  * interrupt from PIR in next vmentry.
4232  */
4233 static int vmx_deliver_posted_interrupt(struct kvm_vcpu *vcpu, int vector)
4234 {
4235 	struct vcpu_vmx *vmx = to_vmx(vcpu);
4236 	int r;
4237 
4238 	r = vmx_deliver_nested_posted_interrupt(vcpu, vector);
4239 	if (!r)
4240 		return 0;
4241 
4242 	/* Note, this is called iff the local APIC is in-kernel. */
4243 	if (!vcpu->arch.apic->apicv_active)
4244 		return -1;
4245 
4246 	if (pi_test_and_set_pir(vector, &vmx->pi_desc))
4247 		return 0;
4248 
4249 	/* If a previous notification has sent the IPI, nothing to do.  */
4250 	if (pi_test_and_set_on(&vmx->pi_desc))
4251 		return 0;
4252 
4253 	/*
4254 	 * The implied barrier in pi_test_and_set_on() pairs with the smp_mb_*()
4255 	 * after setting vcpu->mode in vcpu_enter_guest(), thus the vCPU is
4256 	 * guaranteed to see PID.ON=1 and sync the PIR to IRR if triggering a
4257 	 * posted interrupt "fails" because vcpu->mode != IN_GUEST_MODE.
4258 	 */
4259 	kvm_vcpu_trigger_posted_interrupt(vcpu, POSTED_INTR_VECTOR);
4260 	return 0;
4261 }
4262 
4263 static void vmx_deliver_interrupt(struct kvm_lapic *apic, int delivery_mode,
4264 				  int trig_mode, int vector)
4265 {
4266 	struct kvm_vcpu *vcpu = apic->vcpu;
4267 
4268 	if (vmx_deliver_posted_interrupt(vcpu, vector)) {
4269 		kvm_lapic_set_irr(vector, apic);
4270 		kvm_make_request(KVM_REQ_EVENT, vcpu);
4271 		kvm_vcpu_kick(vcpu);
4272 	} else {
4273 		trace_kvm_apicv_accept_irq(vcpu->vcpu_id, delivery_mode,
4274 					   trig_mode, vector);
4275 	}
4276 }
4277 
4278 /*
4279  * Set up the vmcs's constant host-state fields, i.e., host-state fields that
4280  * will not change in the lifetime of the guest.
4281  * Note that host-state that does change is set elsewhere. E.g., host-state
4282  * that is set differently for each CPU is set in vmx_vcpu_load(), not here.
4283  */
4284 void vmx_set_constant_host_state(struct vcpu_vmx *vmx)
4285 {
4286 	u32 low32, high32;
4287 	unsigned long tmpl;
4288 	unsigned long cr0, cr3, cr4;
4289 
4290 	cr0 = read_cr0();
4291 	WARN_ON(cr0 & X86_CR0_TS);
4292 	vmcs_writel(HOST_CR0, cr0);  /* 22.2.3 */
4293 
4294 	/*
4295 	 * Save the most likely value for this task's CR3 in the VMCS.
4296 	 * We can't use __get_current_cr3_fast() because we're not atomic.
4297 	 */
4298 	cr3 = __read_cr3();
4299 	vmcs_writel(HOST_CR3, cr3);		/* 22.2.3  FIXME: shadow tables */
4300 	vmx->loaded_vmcs->host_state.cr3 = cr3;
4301 
4302 	/* Save the most likely value for this task's CR4 in the VMCS. */
4303 	cr4 = cr4_read_shadow();
4304 	vmcs_writel(HOST_CR4, cr4);			/* 22.2.3, 22.2.5 */
4305 	vmx->loaded_vmcs->host_state.cr4 = cr4;
4306 
4307 	vmcs_write16(HOST_CS_SELECTOR, __KERNEL_CS);  /* 22.2.4 */
4308 #ifdef CONFIG_X86_64
4309 	/*
4310 	 * Load null selectors, so we can avoid reloading them in
4311 	 * vmx_prepare_switch_to_host(), in case userspace uses
4312 	 * the null selectors too (the expected case).
4313 	 */
4314 	vmcs_write16(HOST_DS_SELECTOR, 0);
4315 	vmcs_write16(HOST_ES_SELECTOR, 0);
4316 #else
4317 	vmcs_write16(HOST_DS_SELECTOR, __KERNEL_DS);  /* 22.2.4 */
4318 	vmcs_write16(HOST_ES_SELECTOR, __KERNEL_DS);  /* 22.2.4 */
4319 #endif
4320 	vmcs_write16(HOST_SS_SELECTOR, __KERNEL_DS);  /* 22.2.4 */
4321 	vmcs_write16(HOST_TR_SELECTOR, GDT_ENTRY_TSS*8);  /* 22.2.4 */
4322 
4323 	vmcs_writel(HOST_IDTR_BASE, host_idt_base);   /* 22.2.4 */
4324 
4325 	vmcs_writel(HOST_RIP, (unsigned long)vmx_vmexit); /* 22.2.5 */
4326 
4327 	rdmsr(MSR_IA32_SYSENTER_CS, low32, high32);
4328 	vmcs_write32(HOST_IA32_SYSENTER_CS, low32);
4329 
4330 	/*
4331 	 * SYSENTER is used for 32-bit system calls on either 32-bit or
4332 	 * 64-bit kernels.  It is always zero If neither is allowed, otherwise
4333 	 * vmx_vcpu_load_vmcs loads it with the per-CPU entry stack (and may
4334 	 * have already done so!).
4335 	 */
4336 	if (!IS_ENABLED(CONFIG_IA32_EMULATION) && !IS_ENABLED(CONFIG_X86_32))
4337 		vmcs_writel(HOST_IA32_SYSENTER_ESP, 0);
4338 
4339 	rdmsrl(MSR_IA32_SYSENTER_EIP, tmpl);
4340 	vmcs_writel(HOST_IA32_SYSENTER_EIP, tmpl);   /* 22.2.3 */
4341 
4342 	if (vmcs_config.vmexit_ctrl & VM_EXIT_LOAD_IA32_PAT) {
4343 		rdmsr(MSR_IA32_CR_PAT, low32, high32);
4344 		vmcs_write64(HOST_IA32_PAT, low32 | ((u64) high32 << 32));
4345 	}
4346 
4347 	if (cpu_has_load_ia32_efer())
4348 		vmcs_write64(HOST_IA32_EFER, host_efer);
4349 }
4350 
4351 void set_cr4_guest_host_mask(struct vcpu_vmx *vmx)
4352 {
4353 	struct kvm_vcpu *vcpu = &vmx->vcpu;
4354 
4355 	vcpu->arch.cr4_guest_owned_bits = KVM_POSSIBLE_CR4_GUEST_BITS &
4356 					  ~vcpu->arch.cr4_guest_rsvd_bits;
4357 	if (!enable_ept) {
4358 		vcpu->arch.cr4_guest_owned_bits &= ~X86_CR4_TLBFLUSH_BITS;
4359 		vcpu->arch.cr4_guest_owned_bits &= ~X86_CR4_PDPTR_BITS;
4360 	}
4361 	if (is_guest_mode(&vmx->vcpu))
4362 		vcpu->arch.cr4_guest_owned_bits &=
4363 			~get_vmcs12(vcpu)->cr4_guest_host_mask;
4364 	vmcs_writel(CR4_GUEST_HOST_MASK, ~vcpu->arch.cr4_guest_owned_bits);
4365 }
4366 
4367 static u32 vmx_pin_based_exec_ctrl(struct vcpu_vmx *vmx)
4368 {
4369 	u32 pin_based_exec_ctrl = vmcs_config.pin_based_exec_ctrl;
4370 
4371 	if (!kvm_vcpu_apicv_active(&vmx->vcpu))
4372 		pin_based_exec_ctrl &= ~PIN_BASED_POSTED_INTR;
4373 
4374 	if (!enable_vnmi)
4375 		pin_based_exec_ctrl &= ~PIN_BASED_VIRTUAL_NMIS;
4376 
4377 	if (!enable_preemption_timer)
4378 		pin_based_exec_ctrl &= ~PIN_BASED_VMX_PREEMPTION_TIMER;
4379 
4380 	return pin_based_exec_ctrl;
4381 }
4382 
4383 static u32 vmx_vmentry_ctrl(void)
4384 {
4385 	u32 vmentry_ctrl = vmcs_config.vmentry_ctrl;
4386 
4387 	if (vmx_pt_mode_is_system())
4388 		vmentry_ctrl &= ~(VM_ENTRY_PT_CONCEAL_PIP |
4389 				  VM_ENTRY_LOAD_IA32_RTIT_CTL);
4390 	/*
4391 	 * IA32e mode, and loading of EFER and PERF_GLOBAL_CTRL are toggled dynamically.
4392 	 */
4393 	vmentry_ctrl &= ~(VM_ENTRY_LOAD_IA32_PERF_GLOBAL_CTRL |
4394 			  VM_ENTRY_LOAD_IA32_EFER |
4395 			  VM_ENTRY_IA32E_MODE);
4396 
4397 	if (cpu_has_perf_global_ctrl_bug())
4398 		vmentry_ctrl &= ~VM_ENTRY_LOAD_IA32_PERF_GLOBAL_CTRL;
4399 
4400 	return vmentry_ctrl;
4401 }
4402 
4403 static u32 vmx_vmexit_ctrl(void)
4404 {
4405 	u32 vmexit_ctrl = vmcs_config.vmexit_ctrl;
4406 
4407 	/*
4408 	 * Not used by KVM and never set in vmcs01 or vmcs02, but emulated for
4409 	 * nested virtualization and thus allowed to be set in vmcs12.
4410 	 */
4411 	vmexit_ctrl &= ~(VM_EXIT_SAVE_IA32_PAT | VM_EXIT_SAVE_IA32_EFER |
4412 			 VM_EXIT_SAVE_VMX_PREEMPTION_TIMER);
4413 
4414 	if (vmx_pt_mode_is_system())
4415 		vmexit_ctrl &= ~(VM_EXIT_PT_CONCEAL_PIP |
4416 				 VM_EXIT_CLEAR_IA32_RTIT_CTL);
4417 
4418 	if (cpu_has_perf_global_ctrl_bug())
4419 		vmexit_ctrl &= ~VM_EXIT_LOAD_IA32_PERF_GLOBAL_CTRL;
4420 
4421 	/* Loading of EFER and PERF_GLOBAL_CTRL are toggled dynamically */
4422 	return vmexit_ctrl &
4423 		~(VM_EXIT_LOAD_IA32_PERF_GLOBAL_CTRL | VM_EXIT_LOAD_IA32_EFER);
4424 }
4425 
4426 static void vmx_refresh_apicv_exec_ctrl(struct kvm_vcpu *vcpu)
4427 {
4428 	struct vcpu_vmx *vmx = to_vmx(vcpu);
4429 
4430 	if (is_guest_mode(vcpu)) {
4431 		vmx->nested.update_vmcs01_apicv_status = true;
4432 		return;
4433 	}
4434 
4435 	pin_controls_set(vmx, vmx_pin_based_exec_ctrl(vmx));
4436 
4437 	if (kvm_vcpu_apicv_active(vcpu)) {
4438 		secondary_exec_controls_setbit(vmx,
4439 					       SECONDARY_EXEC_APIC_REGISTER_VIRT |
4440 					       SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY);
4441 		if (enable_ipiv)
4442 			tertiary_exec_controls_setbit(vmx, TERTIARY_EXEC_IPI_VIRT);
4443 	} else {
4444 		secondary_exec_controls_clearbit(vmx,
4445 						 SECONDARY_EXEC_APIC_REGISTER_VIRT |
4446 						 SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY);
4447 		if (enable_ipiv)
4448 			tertiary_exec_controls_clearbit(vmx, TERTIARY_EXEC_IPI_VIRT);
4449 	}
4450 
4451 	vmx_update_msr_bitmap_x2apic(vcpu);
4452 }
4453 
4454 static u32 vmx_exec_control(struct vcpu_vmx *vmx)
4455 {
4456 	u32 exec_control = vmcs_config.cpu_based_exec_ctrl;
4457 
4458 	/*
4459 	 * Not used by KVM, but fully supported for nesting, i.e. are allowed in
4460 	 * vmcs12 and propagated to vmcs02 when set in vmcs12.
4461 	 */
4462 	exec_control &= ~(CPU_BASED_RDTSC_EXITING |
4463 			  CPU_BASED_USE_IO_BITMAPS |
4464 			  CPU_BASED_MONITOR_TRAP_FLAG |
4465 			  CPU_BASED_PAUSE_EXITING);
4466 
4467 	/* INTR_WINDOW_EXITING and NMI_WINDOW_EXITING are toggled dynamically */
4468 	exec_control &= ~(CPU_BASED_INTR_WINDOW_EXITING |
4469 			  CPU_BASED_NMI_WINDOW_EXITING);
4470 
4471 	if (vmx->vcpu.arch.switch_db_regs & KVM_DEBUGREG_WONT_EXIT)
4472 		exec_control &= ~CPU_BASED_MOV_DR_EXITING;
4473 
4474 	if (!cpu_need_tpr_shadow(&vmx->vcpu))
4475 		exec_control &= ~CPU_BASED_TPR_SHADOW;
4476 
4477 #ifdef CONFIG_X86_64
4478 	if (exec_control & CPU_BASED_TPR_SHADOW)
4479 		exec_control &= ~(CPU_BASED_CR8_LOAD_EXITING |
4480 				  CPU_BASED_CR8_STORE_EXITING);
4481 	else
4482 		exec_control |= CPU_BASED_CR8_STORE_EXITING |
4483 				CPU_BASED_CR8_LOAD_EXITING;
4484 #endif
4485 	/* No need to intercept CR3 access or INVPLG when using EPT. */
4486 	if (enable_ept)
4487 		exec_control &= ~(CPU_BASED_CR3_LOAD_EXITING |
4488 				  CPU_BASED_CR3_STORE_EXITING |
4489 				  CPU_BASED_INVLPG_EXITING);
4490 	if (kvm_mwait_in_guest(vmx->vcpu.kvm))
4491 		exec_control &= ~(CPU_BASED_MWAIT_EXITING |
4492 				CPU_BASED_MONITOR_EXITING);
4493 	if (kvm_hlt_in_guest(vmx->vcpu.kvm))
4494 		exec_control &= ~CPU_BASED_HLT_EXITING;
4495 	return exec_control;
4496 }
4497 
4498 static u64 vmx_tertiary_exec_control(struct vcpu_vmx *vmx)
4499 {
4500 	u64 exec_control = vmcs_config.cpu_based_3rd_exec_ctrl;
4501 
4502 	/*
4503 	 * IPI virtualization relies on APICv. Disable IPI virtualization if
4504 	 * APICv is inhibited.
4505 	 */
4506 	if (!enable_ipiv || !kvm_vcpu_apicv_active(&vmx->vcpu))
4507 		exec_control &= ~TERTIARY_EXEC_IPI_VIRT;
4508 
4509 	return exec_control;
4510 }
4511 
4512 /*
4513  * Adjust a single secondary execution control bit to intercept/allow an
4514  * instruction in the guest.  This is usually done based on whether or not a
4515  * feature has been exposed to the guest in order to correctly emulate faults.
4516  */
4517 static inline void
4518 vmx_adjust_secondary_exec_control(struct vcpu_vmx *vmx, u32 *exec_control,
4519 				  u32 control, bool enabled, bool exiting)
4520 {
4521 	/*
4522 	 * If the control is for an opt-in feature, clear the control if the
4523 	 * feature is not exposed to the guest, i.e. not enabled.  If the
4524 	 * control is opt-out, i.e. an exiting control, clear the control if
4525 	 * the feature _is_ exposed to the guest, i.e. exiting/interception is
4526 	 * disabled for the associated instruction.  Note, the caller is
4527 	 * responsible presetting exec_control to set all supported bits.
4528 	 */
4529 	if (enabled == exiting)
4530 		*exec_control &= ~control;
4531 
4532 	/*
4533 	 * Update the nested MSR settings so that a nested VMM can/can't set
4534 	 * controls for features that are/aren't exposed to the guest.
4535 	 */
4536 	if (nested) {
4537 		/*
4538 		 * All features that can be added or removed to VMX MSRs must
4539 		 * be supported in the first place for nested virtualization.
4540 		 */
4541 		if (WARN_ON_ONCE(!(vmcs_config.nested.secondary_ctls_high & control)))
4542 			enabled = false;
4543 
4544 		if (enabled)
4545 			vmx->nested.msrs.secondary_ctls_high |= control;
4546 		else
4547 			vmx->nested.msrs.secondary_ctls_high &= ~control;
4548 	}
4549 }
4550 
4551 /*
4552  * Wrapper macro for the common case of adjusting a secondary execution control
4553  * based on a single guest CPUID bit, with a dedicated feature bit.  This also
4554  * verifies that the control is actually supported by KVM and hardware.
4555  */
4556 #define vmx_adjust_sec_exec_control(vmx, exec_control, name, feat_name, ctrl_name, exiting) \
4557 ({									 \
4558 	bool __enabled;							 \
4559 									 \
4560 	if (cpu_has_vmx_##name()) {					 \
4561 		__enabled = guest_cpuid_has(&(vmx)->vcpu,		 \
4562 					    X86_FEATURE_##feat_name);	 \
4563 		vmx_adjust_secondary_exec_control(vmx, exec_control,	 \
4564 			SECONDARY_EXEC_##ctrl_name, __enabled, exiting); \
4565 	}								 \
4566 })
4567 
4568 /* More macro magic for ENABLE_/opt-in versus _EXITING/opt-out controls. */
4569 #define vmx_adjust_sec_exec_feature(vmx, exec_control, lname, uname) \
4570 	vmx_adjust_sec_exec_control(vmx, exec_control, lname, uname, ENABLE_##uname, false)
4571 
4572 #define vmx_adjust_sec_exec_exiting(vmx, exec_control, lname, uname) \
4573 	vmx_adjust_sec_exec_control(vmx, exec_control, lname, uname, uname##_EXITING, true)
4574 
4575 static u32 vmx_secondary_exec_control(struct vcpu_vmx *vmx)
4576 {
4577 	struct kvm_vcpu *vcpu = &vmx->vcpu;
4578 
4579 	u32 exec_control = vmcs_config.cpu_based_2nd_exec_ctrl;
4580 
4581 	if (vmx_pt_mode_is_system())
4582 		exec_control &= ~(SECONDARY_EXEC_PT_USE_GPA | SECONDARY_EXEC_PT_CONCEAL_VMX);
4583 	if (!cpu_need_virtualize_apic_accesses(vcpu))
4584 		exec_control &= ~SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES;
4585 	if (vmx->vpid == 0)
4586 		exec_control &= ~SECONDARY_EXEC_ENABLE_VPID;
4587 	if (!enable_ept) {
4588 		exec_control &= ~SECONDARY_EXEC_ENABLE_EPT;
4589 		enable_unrestricted_guest = 0;
4590 	}
4591 	if (!enable_unrestricted_guest)
4592 		exec_control &= ~SECONDARY_EXEC_UNRESTRICTED_GUEST;
4593 	if (kvm_pause_in_guest(vmx->vcpu.kvm))
4594 		exec_control &= ~SECONDARY_EXEC_PAUSE_LOOP_EXITING;
4595 	if (!kvm_vcpu_apicv_active(vcpu))
4596 		exec_control &= ~(SECONDARY_EXEC_APIC_REGISTER_VIRT |
4597 				  SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY);
4598 	exec_control &= ~SECONDARY_EXEC_VIRTUALIZE_X2APIC_MODE;
4599 
4600 	/*
4601 	 * KVM doesn't support VMFUNC for L1, but the control is set in KVM's
4602 	 * base configuration as KVM emulates VMFUNC[EPTP_SWITCHING] for L2.
4603 	 */
4604 	exec_control &= ~SECONDARY_EXEC_ENABLE_VMFUNC;
4605 
4606 	/* SECONDARY_EXEC_DESC is enabled/disabled on writes to CR4.UMIP,
4607 	 * in vmx_set_cr4.  */
4608 	exec_control &= ~SECONDARY_EXEC_DESC;
4609 
4610 	/* SECONDARY_EXEC_SHADOW_VMCS is enabled when L1 executes VMPTRLD
4611 	   (handle_vmptrld).
4612 	   We can NOT enable shadow_vmcs here because we don't have yet
4613 	   a current VMCS12
4614 	*/
4615 	exec_control &= ~SECONDARY_EXEC_SHADOW_VMCS;
4616 
4617 	/*
4618 	 * PML is enabled/disabled when dirty logging of memsmlots changes, but
4619 	 * it needs to be set here when dirty logging is already active, e.g.
4620 	 * if this vCPU was created after dirty logging was enabled.
4621 	 */
4622 	if (!enable_pml || !atomic_read(&vcpu->kvm->nr_memslots_dirty_logging))
4623 		exec_control &= ~SECONDARY_EXEC_ENABLE_PML;
4624 
4625 	if (cpu_has_vmx_xsaves()) {
4626 		/* Exposing XSAVES only when XSAVE is exposed */
4627 		bool xsaves_enabled =
4628 			boot_cpu_has(X86_FEATURE_XSAVE) &&
4629 			guest_cpuid_has(vcpu, X86_FEATURE_XSAVE) &&
4630 			guest_cpuid_has(vcpu, X86_FEATURE_XSAVES);
4631 
4632 		vcpu->arch.xsaves_enabled = xsaves_enabled;
4633 
4634 		vmx_adjust_secondary_exec_control(vmx, &exec_control,
4635 						  SECONDARY_EXEC_XSAVES,
4636 						  xsaves_enabled, false);
4637 	}
4638 
4639 	/*
4640 	 * RDPID is also gated by ENABLE_RDTSCP, turn on the control if either
4641 	 * feature is exposed to the guest.  This creates a virtualization hole
4642 	 * if both are supported in hardware but only one is exposed to the
4643 	 * guest, but letting the guest execute RDTSCP or RDPID when either one
4644 	 * is advertised is preferable to emulating the advertised instruction
4645 	 * in KVM on #UD, and obviously better than incorrectly injecting #UD.
4646 	 */
4647 	if (cpu_has_vmx_rdtscp()) {
4648 		bool rdpid_or_rdtscp_enabled =
4649 			guest_cpuid_has(vcpu, X86_FEATURE_RDTSCP) ||
4650 			guest_cpuid_has(vcpu, X86_FEATURE_RDPID);
4651 
4652 		vmx_adjust_secondary_exec_control(vmx, &exec_control,
4653 						  SECONDARY_EXEC_ENABLE_RDTSCP,
4654 						  rdpid_or_rdtscp_enabled, false);
4655 	}
4656 	vmx_adjust_sec_exec_feature(vmx, &exec_control, invpcid, INVPCID);
4657 
4658 	vmx_adjust_sec_exec_exiting(vmx, &exec_control, rdrand, RDRAND);
4659 	vmx_adjust_sec_exec_exiting(vmx, &exec_control, rdseed, RDSEED);
4660 
4661 	vmx_adjust_sec_exec_control(vmx, &exec_control, waitpkg, WAITPKG,
4662 				    ENABLE_USR_WAIT_PAUSE, false);
4663 
4664 	if (!vcpu->kvm->arch.bus_lock_detection_enabled)
4665 		exec_control &= ~SECONDARY_EXEC_BUS_LOCK_DETECTION;
4666 
4667 	if (!kvm_notify_vmexit_enabled(vcpu->kvm))
4668 		exec_control &= ~SECONDARY_EXEC_NOTIFY_VM_EXITING;
4669 
4670 	return exec_control;
4671 }
4672 
4673 static inline int vmx_get_pid_table_order(struct kvm *kvm)
4674 {
4675 	return get_order(kvm->arch.max_vcpu_ids * sizeof(*to_kvm_vmx(kvm)->pid_table));
4676 }
4677 
4678 static int vmx_alloc_ipiv_pid_table(struct kvm *kvm)
4679 {
4680 	struct page *pages;
4681 	struct kvm_vmx *kvm_vmx = to_kvm_vmx(kvm);
4682 
4683 	if (!irqchip_in_kernel(kvm) || !enable_ipiv)
4684 		return 0;
4685 
4686 	if (kvm_vmx->pid_table)
4687 		return 0;
4688 
4689 	pages = alloc_pages(GFP_KERNEL_ACCOUNT | __GFP_ZERO,
4690 			    vmx_get_pid_table_order(kvm));
4691 	if (!pages)
4692 		return -ENOMEM;
4693 
4694 	kvm_vmx->pid_table = (void *)page_address(pages);
4695 	return 0;
4696 }
4697 
4698 static int vmx_vcpu_precreate(struct kvm *kvm)
4699 {
4700 	return vmx_alloc_ipiv_pid_table(kvm);
4701 }
4702 
4703 #define VMX_XSS_EXIT_BITMAP 0
4704 
4705 static void init_vmcs(struct vcpu_vmx *vmx)
4706 {
4707 	struct kvm *kvm = vmx->vcpu.kvm;
4708 	struct kvm_vmx *kvm_vmx = to_kvm_vmx(kvm);
4709 
4710 	if (nested)
4711 		nested_vmx_set_vmcs_shadowing_bitmap();
4712 
4713 	if (cpu_has_vmx_msr_bitmap())
4714 		vmcs_write64(MSR_BITMAP, __pa(vmx->vmcs01.msr_bitmap));
4715 
4716 	vmcs_write64(VMCS_LINK_POINTER, INVALID_GPA); /* 22.3.1.5 */
4717 
4718 	/* Control */
4719 	pin_controls_set(vmx, vmx_pin_based_exec_ctrl(vmx));
4720 
4721 	exec_controls_set(vmx, vmx_exec_control(vmx));
4722 
4723 	if (cpu_has_secondary_exec_ctrls())
4724 		secondary_exec_controls_set(vmx, vmx_secondary_exec_control(vmx));
4725 
4726 	if (cpu_has_tertiary_exec_ctrls())
4727 		tertiary_exec_controls_set(vmx, vmx_tertiary_exec_control(vmx));
4728 
4729 	if (enable_apicv && lapic_in_kernel(&vmx->vcpu)) {
4730 		vmcs_write64(EOI_EXIT_BITMAP0, 0);
4731 		vmcs_write64(EOI_EXIT_BITMAP1, 0);
4732 		vmcs_write64(EOI_EXIT_BITMAP2, 0);
4733 		vmcs_write64(EOI_EXIT_BITMAP3, 0);
4734 
4735 		vmcs_write16(GUEST_INTR_STATUS, 0);
4736 
4737 		vmcs_write16(POSTED_INTR_NV, POSTED_INTR_VECTOR);
4738 		vmcs_write64(POSTED_INTR_DESC_ADDR, __pa((&vmx->pi_desc)));
4739 	}
4740 
4741 	if (vmx_can_use_ipiv(&vmx->vcpu)) {
4742 		vmcs_write64(PID_POINTER_TABLE, __pa(kvm_vmx->pid_table));
4743 		vmcs_write16(LAST_PID_POINTER_INDEX, kvm->arch.max_vcpu_ids - 1);
4744 	}
4745 
4746 	if (!kvm_pause_in_guest(kvm)) {
4747 		vmcs_write32(PLE_GAP, ple_gap);
4748 		vmx->ple_window = ple_window;
4749 		vmx->ple_window_dirty = true;
4750 	}
4751 
4752 	if (kvm_notify_vmexit_enabled(kvm))
4753 		vmcs_write32(NOTIFY_WINDOW, kvm->arch.notify_window);
4754 
4755 	vmcs_write32(PAGE_FAULT_ERROR_CODE_MASK, 0);
4756 	vmcs_write32(PAGE_FAULT_ERROR_CODE_MATCH, 0);
4757 	vmcs_write32(CR3_TARGET_COUNT, 0);           /* 22.2.1 */
4758 
4759 	vmcs_write16(HOST_FS_SELECTOR, 0);            /* 22.2.4 */
4760 	vmcs_write16(HOST_GS_SELECTOR, 0);            /* 22.2.4 */
4761 	vmx_set_constant_host_state(vmx);
4762 	vmcs_writel(HOST_FS_BASE, 0); /* 22.2.4 */
4763 	vmcs_writel(HOST_GS_BASE, 0); /* 22.2.4 */
4764 
4765 	if (cpu_has_vmx_vmfunc())
4766 		vmcs_write64(VM_FUNCTION_CONTROL, 0);
4767 
4768 	vmcs_write32(VM_EXIT_MSR_STORE_COUNT, 0);
4769 	vmcs_write32(VM_EXIT_MSR_LOAD_COUNT, 0);
4770 	vmcs_write64(VM_EXIT_MSR_LOAD_ADDR, __pa(vmx->msr_autoload.host.val));
4771 	vmcs_write32(VM_ENTRY_MSR_LOAD_COUNT, 0);
4772 	vmcs_write64(VM_ENTRY_MSR_LOAD_ADDR, __pa(vmx->msr_autoload.guest.val));
4773 
4774 	if (vmcs_config.vmentry_ctrl & VM_ENTRY_LOAD_IA32_PAT)
4775 		vmcs_write64(GUEST_IA32_PAT, vmx->vcpu.arch.pat);
4776 
4777 	vm_exit_controls_set(vmx, vmx_vmexit_ctrl());
4778 
4779 	/* 22.2.1, 20.8.1 */
4780 	vm_entry_controls_set(vmx, vmx_vmentry_ctrl());
4781 
4782 	vmx->vcpu.arch.cr0_guest_owned_bits = vmx_l1_guest_owned_cr0_bits();
4783 	vmcs_writel(CR0_GUEST_HOST_MASK, ~vmx->vcpu.arch.cr0_guest_owned_bits);
4784 
4785 	set_cr4_guest_host_mask(vmx);
4786 
4787 	if (vmx->vpid != 0)
4788 		vmcs_write16(VIRTUAL_PROCESSOR_ID, vmx->vpid);
4789 
4790 	if (cpu_has_vmx_xsaves())
4791 		vmcs_write64(XSS_EXIT_BITMAP, VMX_XSS_EXIT_BITMAP);
4792 
4793 	if (enable_pml) {
4794 		vmcs_write64(PML_ADDRESS, page_to_phys(vmx->pml_pg));
4795 		vmcs_write16(GUEST_PML_INDEX, PML_ENTITY_NUM - 1);
4796 	}
4797 
4798 	vmx_write_encls_bitmap(&vmx->vcpu, NULL);
4799 
4800 	if (vmx_pt_mode_is_host_guest()) {
4801 		memset(&vmx->pt_desc, 0, sizeof(vmx->pt_desc));
4802 		/* Bit[6~0] are forced to 1, writes are ignored. */
4803 		vmx->pt_desc.guest.output_mask = 0x7F;
4804 		vmcs_write64(GUEST_IA32_RTIT_CTL, 0);
4805 	}
4806 
4807 	vmcs_write32(GUEST_SYSENTER_CS, 0);
4808 	vmcs_writel(GUEST_SYSENTER_ESP, 0);
4809 	vmcs_writel(GUEST_SYSENTER_EIP, 0);
4810 	vmcs_write64(GUEST_IA32_DEBUGCTL, 0);
4811 
4812 	if (cpu_has_vmx_tpr_shadow()) {
4813 		vmcs_write64(VIRTUAL_APIC_PAGE_ADDR, 0);
4814 		if (cpu_need_tpr_shadow(&vmx->vcpu))
4815 			vmcs_write64(VIRTUAL_APIC_PAGE_ADDR,
4816 				     __pa(vmx->vcpu.arch.apic->regs));
4817 		vmcs_write32(TPR_THRESHOLD, 0);
4818 	}
4819 
4820 	vmx_setup_uret_msrs(vmx);
4821 }
4822 
4823 static void __vmx_vcpu_reset(struct kvm_vcpu *vcpu)
4824 {
4825 	struct vcpu_vmx *vmx = to_vmx(vcpu);
4826 
4827 	init_vmcs(vmx);
4828 
4829 	if (nested)
4830 		memcpy(&vmx->nested.msrs, &vmcs_config.nested, sizeof(vmx->nested.msrs));
4831 
4832 	vcpu_setup_sgx_lepubkeyhash(vcpu);
4833 
4834 	vmx->nested.posted_intr_nv = -1;
4835 	vmx->nested.vmxon_ptr = INVALID_GPA;
4836 	vmx->nested.current_vmptr = INVALID_GPA;
4837 	vmx->nested.hv_evmcs_vmptr = EVMPTR_INVALID;
4838 
4839 	vcpu->arch.microcode_version = 0x100000000ULL;
4840 	vmx->msr_ia32_feature_control_valid_bits = FEAT_CTL_LOCKED;
4841 
4842 	/*
4843 	 * Enforce invariant: pi_desc.nv is always either POSTED_INTR_VECTOR
4844 	 * or POSTED_INTR_WAKEUP_VECTOR.
4845 	 */
4846 	vmx->pi_desc.nv = POSTED_INTR_VECTOR;
4847 	vmx->pi_desc.sn = 1;
4848 }
4849 
4850 static void vmx_vcpu_reset(struct kvm_vcpu *vcpu, bool init_event)
4851 {
4852 	struct vcpu_vmx *vmx = to_vmx(vcpu);
4853 
4854 	if (!init_event)
4855 		__vmx_vcpu_reset(vcpu);
4856 
4857 	vmx->rmode.vm86_active = 0;
4858 	vmx->spec_ctrl = 0;
4859 
4860 	vmx->msr_ia32_umwait_control = 0;
4861 
4862 	vmx->hv_deadline_tsc = -1;
4863 	kvm_set_cr8(vcpu, 0);
4864 
4865 	vmx_segment_cache_clear(vmx);
4866 	kvm_register_mark_available(vcpu, VCPU_EXREG_SEGMENTS);
4867 
4868 	seg_setup(VCPU_SREG_CS);
4869 	vmcs_write16(GUEST_CS_SELECTOR, 0xf000);
4870 	vmcs_writel(GUEST_CS_BASE, 0xffff0000ul);
4871 
4872 	seg_setup(VCPU_SREG_DS);
4873 	seg_setup(VCPU_SREG_ES);
4874 	seg_setup(VCPU_SREG_FS);
4875 	seg_setup(VCPU_SREG_GS);
4876 	seg_setup(VCPU_SREG_SS);
4877 
4878 	vmcs_write16(GUEST_TR_SELECTOR, 0);
4879 	vmcs_writel(GUEST_TR_BASE, 0);
4880 	vmcs_write32(GUEST_TR_LIMIT, 0xffff);
4881 	vmcs_write32(GUEST_TR_AR_BYTES, 0x008b);
4882 
4883 	vmcs_write16(GUEST_LDTR_SELECTOR, 0);
4884 	vmcs_writel(GUEST_LDTR_BASE, 0);
4885 	vmcs_write32(GUEST_LDTR_LIMIT, 0xffff);
4886 	vmcs_write32(GUEST_LDTR_AR_BYTES, 0x00082);
4887 
4888 	vmcs_writel(GUEST_GDTR_BASE, 0);
4889 	vmcs_write32(GUEST_GDTR_LIMIT, 0xffff);
4890 
4891 	vmcs_writel(GUEST_IDTR_BASE, 0);
4892 	vmcs_write32(GUEST_IDTR_LIMIT, 0xffff);
4893 
4894 	vmcs_write32(GUEST_ACTIVITY_STATE, GUEST_ACTIVITY_ACTIVE);
4895 	vmcs_write32(GUEST_INTERRUPTIBILITY_INFO, 0);
4896 	vmcs_writel(GUEST_PENDING_DBG_EXCEPTIONS, 0);
4897 	if (kvm_mpx_supported())
4898 		vmcs_write64(GUEST_BNDCFGS, 0);
4899 
4900 	vmcs_write32(VM_ENTRY_INTR_INFO_FIELD, 0);  /* 22.2.1 */
4901 
4902 	kvm_make_request(KVM_REQ_APIC_PAGE_RELOAD, vcpu);
4903 
4904 	vpid_sync_context(vmx->vpid);
4905 
4906 	vmx_update_fb_clear_dis(vcpu, vmx);
4907 }
4908 
4909 static void vmx_enable_irq_window(struct kvm_vcpu *vcpu)
4910 {
4911 	exec_controls_setbit(to_vmx(vcpu), CPU_BASED_INTR_WINDOW_EXITING);
4912 }
4913 
4914 static void vmx_enable_nmi_window(struct kvm_vcpu *vcpu)
4915 {
4916 	if (!enable_vnmi ||
4917 	    vmcs_read32(GUEST_INTERRUPTIBILITY_INFO) & GUEST_INTR_STATE_STI) {
4918 		vmx_enable_irq_window(vcpu);
4919 		return;
4920 	}
4921 
4922 	exec_controls_setbit(to_vmx(vcpu), CPU_BASED_NMI_WINDOW_EXITING);
4923 }
4924 
4925 static void vmx_inject_irq(struct kvm_vcpu *vcpu, bool reinjected)
4926 {
4927 	struct vcpu_vmx *vmx = to_vmx(vcpu);
4928 	uint32_t intr;
4929 	int irq = vcpu->arch.interrupt.nr;
4930 
4931 	trace_kvm_inj_virq(irq, vcpu->arch.interrupt.soft, reinjected);
4932 
4933 	++vcpu->stat.irq_injections;
4934 	if (vmx->rmode.vm86_active) {
4935 		int inc_eip = 0;
4936 		if (vcpu->arch.interrupt.soft)
4937 			inc_eip = vcpu->arch.event_exit_inst_len;
4938 		kvm_inject_realmode_interrupt(vcpu, irq, inc_eip);
4939 		return;
4940 	}
4941 	intr = irq | INTR_INFO_VALID_MASK;
4942 	if (vcpu->arch.interrupt.soft) {
4943 		intr |= INTR_TYPE_SOFT_INTR;
4944 		vmcs_write32(VM_ENTRY_INSTRUCTION_LEN,
4945 			     vmx->vcpu.arch.event_exit_inst_len);
4946 	} else
4947 		intr |= INTR_TYPE_EXT_INTR;
4948 	vmcs_write32(VM_ENTRY_INTR_INFO_FIELD, intr);
4949 
4950 	vmx_clear_hlt(vcpu);
4951 }
4952 
4953 static void vmx_inject_nmi(struct kvm_vcpu *vcpu)
4954 {
4955 	struct vcpu_vmx *vmx = to_vmx(vcpu);
4956 
4957 	if (!enable_vnmi) {
4958 		/*
4959 		 * Tracking the NMI-blocked state in software is built upon
4960 		 * finding the next open IRQ window. This, in turn, depends on
4961 		 * well-behaving guests: They have to keep IRQs disabled at
4962 		 * least as long as the NMI handler runs. Otherwise we may
4963 		 * cause NMI nesting, maybe breaking the guest. But as this is
4964 		 * highly unlikely, we can live with the residual risk.
4965 		 */
4966 		vmx->loaded_vmcs->soft_vnmi_blocked = 1;
4967 		vmx->loaded_vmcs->vnmi_blocked_time = 0;
4968 	}
4969 
4970 	++vcpu->stat.nmi_injections;
4971 	vmx->loaded_vmcs->nmi_known_unmasked = false;
4972 
4973 	if (vmx->rmode.vm86_active) {
4974 		kvm_inject_realmode_interrupt(vcpu, NMI_VECTOR, 0);
4975 		return;
4976 	}
4977 
4978 	vmcs_write32(VM_ENTRY_INTR_INFO_FIELD,
4979 			INTR_TYPE_NMI_INTR | INTR_INFO_VALID_MASK | NMI_VECTOR);
4980 
4981 	vmx_clear_hlt(vcpu);
4982 }
4983 
4984 bool vmx_get_nmi_mask(struct kvm_vcpu *vcpu)
4985 {
4986 	struct vcpu_vmx *vmx = to_vmx(vcpu);
4987 	bool masked;
4988 
4989 	if (!enable_vnmi)
4990 		return vmx->loaded_vmcs->soft_vnmi_blocked;
4991 	if (vmx->loaded_vmcs->nmi_known_unmasked)
4992 		return false;
4993 	masked = vmcs_read32(GUEST_INTERRUPTIBILITY_INFO) & GUEST_INTR_STATE_NMI;
4994 	vmx->loaded_vmcs->nmi_known_unmasked = !masked;
4995 	return masked;
4996 }
4997 
4998 void vmx_set_nmi_mask(struct kvm_vcpu *vcpu, bool masked)
4999 {
5000 	struct vcpu_vmx *vmx = to_vmx(vcpu);
5001 
5002 	if (!enable_vnmi) {
5003 		if (vmx->loaded_vmcs->soft_vnmi_blocked != masked) {
5004 			vmx->loaded_vmcs->soft_vnmi_blocked = masked;
5005 			vmx->loaded_vmcs->vnmi_blocked_time = 0;
5006 		}
5007 	} else {
5008 		vmx->loaded_vmcs->nmi_known_unmasked = !masked;
5009 		if (masked)
5010 			vmcs_set_bits(GUEST_INTERRUPTIBILITY_INFO,
5011 				      GUEST_INTR_STATE_NMI);
5012 		else
5013 			vmcs_clear_bits(GUEST_INTERRUPTIBILITY_INFO,
5014 					GUEST_INTR_STATE_NMI);
5015 	}
5016 }
5017 
5018 bool vmx_nmi_blocked(struct kvm_vcpu *vcpu)
5019 {
5020 	if (is_guest_mode(vcpu) && nested_exit_on_nmi(vcpu))
5021 		return false;
5022 
5023 	if (!enable_vnmi && to_vmx(vcpu)->loaded_vmcs->soft_vnmi_blocked)
5024 		return true;
5025 
5026 	return (vmcs_read32(GUEST_INTERRUPTIBILITY_INFO) &
5027 		(GUEST_INTR_STATE_MOV_SS | GUEST_INTR_STATE_STI |
5028 		 GUEST_INTR_STATE_NMI));
5029 }
5030 
5031 static int vmx_nmi_allowed(struct kvm_vcpu *vcpu, bool for_injection)
5032 {
5033 	if (to_vmx(vcpu)->nested.nested_run_pending)
5034 		return -EBUSY;
5035 
5036 	/* An NMI must not be injected into L2 if it's supposed to VM-Exit.  */
5037 	if (for_injection && is_guest_mode(vcpu) && nested_exit_on_nmi(vcpu))
5038 		return -EBUSY;
5039 
5040 	return !vmx_nmi_blocked(vcpu);
5041 }
5042 
5043 bool vmx_interrupt_blocked(struct kvm_vcpu *vcpu)
5044 {
5045 	if (is_guest_mode(vcpu) && nested_exit_on_intr(vcpu))
5046 		return false;
5047 
5048 	return !(vmx_get_rflags(vcpu) & X86_EFLAGS_IF) ||
5049 	       (vmcs_read32(GUEST_INTERRUPTIBILITY_INFO) &
5050 		(GUEST_INTR_STATE_STI | GUEST_INTR_STATE_MOV_SS));
5051 }
5052 
5053 static int vmx_interrupt_allowed(struct kvm_vcpu *vcpu, bool for_injection)
5054 {
5055 	if (to_vmx(vcpu)->nested.nested_run_pending)
5056 		return -EBUSY;
5057 
5058 	/*
5059 	 * An IRQ must not be injected into L2 if it's supposed to VM-Exit,
5060 	 * e.g. if the IRQ arrived asynchronously after checking nested events.
5061 	 */
5062 	if (for_injection && is_guest_mode(vcpu) && nested_exit_on_intr(vcpu))
5063 		return -EBUSY;
5064 
5065 	return !vmx_interrupt_blocked(vcpu);
5066 }
5067 
5068 static int vmx_set_tss_addr(struct kvm *kvm, unsigned int addr)
5069 {
5070 	void __user *ret;
5071 
5072 	if (enable_unrestricted_guest)
5073 		return 0;
5074 
5075 	mutex_lock(&kvm->slots_lock);
5076 	ret = __x86_set_memory_region(kvm, TSS_PRIVATE_MEMSLOT, addr,
5077 				      PAGE_SIZE * 3);
5078 	mutex_unlock(&kvm->slots_lock);
5079 
5080 	if (IS_ERR(ret))
5081 		return PTR_ERR(ret);
5082 
5083 	to_kvm_vmx(kvm)->tss_addr = addr;
5084 
5085 	return init_rmode_tss(kvm, ret);
5086 }
5087 
5088 static int vmx_set_identity_map_addr(struct kvm *kvm, u64 ident_addr)
5089 {
5090 	to_kvm_vmx(kvm)->ept_identity_map_addr = ident_addr;
5091 	return 0;
5092 }
5093 
5094 static bool rmode_exception(struct kvm_vcpu *vcpu, int vec)
5095 {
5096 	switch (vec) {
5097 	case BP_VECTOR:
5098 		/*
5099 		 * Update instruction length as we may reinject the exception
5100 		 * from user space while in guest debugging mode.
5101 		 */
5102 		to_vmx(vcpu)->vcpu.arch.event_exit_inst_len =
5103 			vmcs_read32(VM_EXIT_INSTRUCTION_LEN);
5104 		if (vcpu->guest_debug & KVM_GUESTDBG_USE_SW_BP)
5105 			return false;
5106 		fallthrough;
5107 	case DB_VECTOR:
5108 		return !(vcpu->guest_debug &
5109 			(KVM_GUESTDBG_SINGLESTEP | KVM_GUESTDBG_USE_HW_BP));
5110 	case DE_VECTOR:
5111 	case OF_VECTOR:
5112 	case BR_VECTOR:
5113 	case UD_VECTOR:
5114 	case DF_VECTOR:
5115 	case SS_VECTOR:
5116 	case GP_VECTOR:
5117 	case MF_VECTOR:
5118 		return true;
5119 	}
5120 	return false;
5121 }
5122 
5123 static int handle_rmode_exception(struct kvm_vcpu *vcpu,
5124 				  int vec, u32 err_code)
5125 {
5126 	/*
5127 	 * Instruction with address size override prefix opcode 0x67
5128 	 * Cause the #SS fault with 0 error code in VM86 mode.
5129 	 */
5130 	if (((vec == GP_VECTOR) || (vec == SS_VECTOR)) && err_code == 0) {
5131 		if (kvm_emulate_instruction(vcpu, 0)) {
5132 			if (vcpu->arch.halt_request) {
5133 				vcpu->arch.halt_request = 0;
5134 				return kvm_emulate_halt_noskip(vcpu);
5135 			}
5136 			return 1;
5137 		}
5138 		return 0;
5139 	}
5140 
5141 	/*
5142 	 * Forward all other exceptions that are valid in real mode.
5143 	 * FIXME: Breaks guest debugging in real mode, needs to be fixed with
5144 	 *        the required debugging infrastructure rework.
5145 	 */
5146 	kvm_queue_exception(vcpu, vec);
5147 	return 1;
5148 }
5149 
5150 static int handle_machine_check(struct kvm_vcpu *vcpu)
5151 {
5152 	/* handled by vmx_vcpu_run() */
5153 	return 1;
5154 }
5155 
5156 /*
5157  * If the host has split lock detection disabled, then #AC is
5158  * unconditionally injected into the guest, which is the pre split lock
5159  * detection behaviour.
5160  *
5161  * If the host has split lock detection enabled then #AC is
5162  * only injected into the guest when:
5163  *  - Guest CPL == 3 (user mode)
5164  *  - Guest has #AC detection enabled in CR0
5165  *  - Guest EFLAGS has AC bit set
5166  */
5167 bool vmx_guest_inject_ac(struct kvm_vcpu *vcpu)
5168 {
5169 	if (!boot_cpu_has(X86_FEATURE_SPLIT_LOCK_DETECT))
5170 		return true;
5171 
5172 	return vmx_get_cpl(vcpu) == 3 && kvm_is_cr0_bit_set(vcpu, X86_CR0_AM) &&
5173 	       (kvm_get_rflags(vcpu) & X86_EFLAGS_AC);
5174 }
5175 
5176 static int handle_exception_nmi(struct kvm_vcpu *vcpu)
5177 {
5178 	struct vcpu_vmx *vmx = to_vmx(vcpu);
5179 	struct kvm_run *kvm_run = vcpu->run;
5180 	u32 intr_info, ex_no, error_code;
5181 	unsigned long cr2, dr6;
5182 	u32 vect_info;
5183 
5184 	vect_info = vmx->idt_vectoring_info;
5185 	intr_info = vmx_get_intr_info(vcpu);
5186 
5187 	/*
5188 	 * Machine checks are handled by handle_exception_irqoff(), or by
5189 	 * vmx_vcpu_run() if a #MC occurs on VM-Entry.  NMIs are handled by
5190 	 * vmx_vcpu_enter_exit().
5191 	 */
5192 	if (is_machine_check(intr_info) || is_nmi(intr_info))
5193 		return 1;
5194 
5195 	/*
5196 	 * Queue the exception here instead of in handle_nm_fault_irqoff().
5197 	 * This ensures the nested_vmx check is not skipped so vmexit can
5198 	 * be reflected to L1 (when it intercepts #NM) before reaching this
5199 	 * point.
5200 	 */
5201 	if (is_nm_fault(intr_info)) {
5202 		kvm_queue_exception(vcpu, NM_VECTOR);
5203 		return 1;
5204 	}
5205 
5206 	if (is_invalid_opcode(intr_info))
5207 		return handle_ud(vcpu);
5208 
5209 	error_code = 0;
5210 	if (intr_info & INTR_INFO_DELIVER_CODE_MASK)
5211 		error_code = vmcs_read32(VM_EXIT_INTR_ERROR_CODE);
5212 
5213 	if (!vmx->rmode.vm86_active && is_gp_fault(intr_info)) {
5214 		WARN_ON_ONCE(!enable_vmware_backdoor);
5215 
5216 		/*
5217 		 * VMware backdoor emulation on #GP interception only handles
5218 		 * IN{S}, OUT{S}, and RDPMC, none of which generate a non-zero
5219 		 * error code on #GP.
5220 		 */
5221 		if (error_code) {
5222 			kvm_queue_exception_e(vcpu, GP_VECTOR, error_code);
5223 			return 1;
5224 		}
5225 		return kvm_emulate_instruction(vcpu, EMULTYPE_VMWARE_GP);
5226 	}
5227 
5228 	/*
5229 	 * The #PF with PFEC.RSVD = 1 indicates the guest is accessing
5230 	 * MMIO, it is better to report an internal error.
5231 	 * See the comments in vmx_handle_exit.
5232 	 */
5233 	if ((vect_info & VECTORING_INFO_VALID_MASK) &&
5234 	    !(is_page_fault(intr_info) && !(error_code & PFERR_RSVD_MASK))) {
5235 		vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
5236 		vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_SIMUL_EX;
5237 		vcpu->run->internal.ndata = 4;
5238 		vcpu->run->internal.data[0] = vect_info;
5239 		vcpu->run->internal.data[1] = intr_info;
5240 		vcpu->run->internal.data[2] = error_code;
5241 		vcpu->run->internal.data[3] = vcpu->arch.last_vmentry_cpu;
5242 		return 0;
5243 	}
5244 
5245 	if (is_page_fault(intr_info)) {
5246 		cr2 = vmx_get_exit_qual(vcpu);
5247 		if (enable_ept && !vcpu->arch.apf.host_apf_flags) {
5248 			/*
5249 			 * EPT will cause page fault only if we need to
5250 			 * detect illegal GPAs.
5251 			 */
5252 			WARN_ON_ONCE(!allow_smaller_maxphyaddr);
5253 			kvm_fixup_and_inject_pf_error(vcpu, cr2, error_code);
5254 			return 1;
5255 		} else
5256 			return kvm_handle_page_fault(vcpu, error_code, cr2, NULL, 0);
5257 	}
5258 
5259 	ex_no = intr_info & INTR_INFO_VECTOR_MASK;
5260 
5261 	if (vmx->rmode.vm86_active && rmode_exception(vcpu, ex_no))
5262 		return handle_rmode_exception(vcpu, ex_no, error_code);
5263 
5264 	switch (ex_no) {
5265 	case DB_VECTOR:
5266 		dr6 = vmx_get_exit_qual(vcpu);
5267 		if (!(vcpu->guest_debug &
5268 		      (KVM_GUESTDBG_SINGLESTEP | KVM_GUESTDBG_USE_HW_BP))) {
5269 			/*
5270 			 * If the #DB was due to ICEBP, a.k.a. INT1, skip the
5271 			 * instruction.  ICEBP generates a trap-like #DB, but
5272 			 * despite its interception control being tied to #DB,
5273 			 * is an instruction intercept, i.e. the VM-Exit occurs
5274 			 * on the ICEBP itself.  Use the inner "skip" helper to
5275 			 * avoid single-step #DB and MTF updates, as ICEBP is
5276 			 * higher priority.  Note, skipping ICEBP still clears
5277 			 * STI and MOVSS blocking.
5278 			 *
5279 			 * For all other #DBs, set vmcs.PENDING_DBG_EXCEPTIONS.BS
5280 			 * if single-step is enabled in RFLAGS and STI or MOVSS
5281 			 * blocking is active, as the CPU doesn't set the bit
5282 			 * on VM-Exit due to #DB interception.  VM-Entry has a
5283 			 * consistency check that a single-step #DB is pending
5284 			 * in this scenario as the previous instruction cannot
5285 			 * have toggled RFLAGS.TF 0=>1 (because STI and POP/MOV
5286 			 * don't modify RFLAGS), therefore the one instruction
5287 			 * delay when activating single-step breakpoints must
5288 			 * have already expired.  Note, the CPU sets/clears BS
5289 			 * as appropriate for all other VM-Exits types.
5290 			 */
5291 			if (is_icebp(intr_info))
5292 				WARN_ON(!skip_emulated_instruction(vcpu));
5293 			else if ((vmx_get_rflags(vcpu) & X86_EFLAGS_TF) &&
5294 				 (vmcs_read32(GUEST_INTERRUPTIBILITY_INFO) &
5295 				  (GUEST_INTR_STATE_STI | GUEST_INTR_STATE_MOV_SS)))
5296 				vmcs_writel(GUEST_PENDING_DBG_EXCEPTIONS,
5297 					    vmcs_readl(GUEST_PENDING_DBG_EXCEPTIONS) | DR6_BS);
5298 
5299 			kvm_queue_exception_p(vcpu, DB_VECTOR, dr6);
5300 			return 1;
5301 		}
5302 		kvm_run->debug.arch.dr6 = dr6 | DR6_ACTIVE_LOW;
5303 		kvm_run->debug.arch.dr7 = vmcs_readl(GUEST_DR7);
5304 		fallthrough;
5305 	case BP_VECTOR:
5306 		/*
5307 		 * Update instruction length as we may reinject #BP from
5308 		 * user space while in guest debugging mode. Reading it for
5309 		 * #DB as well causes no harm, it is not used in that case.
5310 		 */
5311 		vmx->vcpu.arch.event_exit_inst_len =
5312 			vmcs_read32(VM_EXIT_INSTRUCTION_LEN);
5313 		kvm_run->exit_reason = KVM_EXIT_DEBUG;
5314 		kvm_run->debug.arch.pc = kvm_get_linear_rip(vcpu);
5315 		kvm_run->debug.arch.exception = ex_no;
5316 		break;
5317 	case AC_VECTOR:
5318 		if (vmx_guest_inject_ac(vcpu)) {
5319 			kvm_queue_exception_e(vcpu, AC_VECTOR, error_code);
5320 			return 1;
5321 		}
5322 
5323 		/*
5324 		 * Handle split lock. Depending on detection mode this will
5325 		 * either warn and disable split lock detection for this
5326 		 * task or force SIGBUS on it.
5327 		 */
5328 		if (handle_guest_split_lock(kvm_rip_read(vcpu)))
5329 			return 1;
5330 		fallthrough;
5331 	default:
5332 		kvm_run->exit_reason = KVM_EXIT_EXCEPTION;
5333 		kvm_run->ex.exception = ex_no;
5334 		kvm_run->ex.error_code = error_code;
5335 		break;
5336 	}
5337 	return 0;
5338 }
5339 
5340 static __always_inline int handle_external_interrupt(struct kvm_vcpu *vcpu)
5341 {
5342 	++vcpu->stat.irq_exits;
5343 	return 1;
5344 }
5345 
5346 static int handle_triple_fault(struct kvm_vcpu *vcpu)
5347 {
5348 	vcpu->run->exit_reason = KVM_EXIT_SHUTDOWN;
5349 	vcpu->mmio_needed = 0;
5350 	return 0;
5351 }
5352 
5353 static int handle_io(struct kvm_vcpu *vcpu)
5354 {
5355 	unsigned long exit_qualification;
5356 	int size, in, string;
5357 	unsigned port;
5358 
5359 	exit_qualification = vmx_get_exit_qual(vcpu);
5360 	string = (exit_qualification & 16) != 0;
5361 
5362 	++vcpu->stat.io_exits;
5363 
5364 	if (string)
5365 		return kvm_emulate_instruction(vcpu, 0);
5366 
5367 	port = exit_qualification >> 16;
5368 	size = (exit_qualification & 7) + 1;
5369 	in = (exit_qualification & 8) != 0;
5370 
5371 	return kvm_fast_pio(vcpu, size, port, in);
5372 }
5373 
5374 static void
5375 vmx_patch_hypercall(struct kvm_vcpu *vcpu, unsigned char *hypercall)
5376 {
5377 	/*
5378 	 * Patch in the VMCALL instruction:
5379 	 */
5380 	hypercall[0] = 0x0f;
5381 	hypercall[1] = 0x01;
5382 	hypercall[2] = 0xc1;
5383 }
5384 
5385 /* called to set cr0 as appropriate for a mov-to-cr0 exit. */
5386 static int handle_set_cr0(struct kvm_vcpu *vcpu, unsigned long val)
5387 {
5388 	if (is_guest_mode(vcpu)) {
5389 		struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
5390 		unsigned long orig_val = val;
5391 
5392 		/*
5393 		 * We get here when L2 changed cr0 in a way that did not change
5394 		 * any of L1's shadowed bits (see nested_vmx_exit_handled_cr),
5395 		 * but did change L0 shadowed bits. So we first calculate the
5396 		 * effective cr0 value that L1 would like to write into the
5397 		 * hardware. It consists of the L2-owned bits from the new
5398 		 * value combined with the L1-owned bits from L1's guest_cr0.
5399 		 */
5400 		val = (val & ~vmcs12->cr0_guest_host_mask) |
5401 			(vmcs12->guest_cr0 & vmcs12->cr0_guest_host_mask);
5402 
5403 		if (kvm_set_cr0(vcpu, val))
5404 			return 1;
5405 		vmcs_writel(CR0_READ_SHADOW, orig_val);
5406 		return 0;
5407 	} else {
5408 		return kvm_set_cr0(vcpu, val);
5409 	}
5410 }
5411 
5412 static int handle_set_cr4(struct kvm_vcpu *vcpu, unsigned long val)
5413 {
5414 	if (is_guest_mode(vcpu)) {
5415 		struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
5416 		unsigned long orig_val = val;
5417 
5418 		/* analogously to handle_set_cr0 */
5419 		val = (val & ~vmcs12->cr4_guest_host_mask) |
5420 			(vmcs12->guest_cr4 & vmcs12->cr4_guest_host_mask);
5421 		if (kvm_set_cr4(vcpu, val))
5422 			return 1;
5423 		vmcs_writel(CR4_READ_SHADOW, orig_val);
5424 		return 0;
5425 	} else
5426 		return kvm_set_cr4(vcpu, val);
5427 }
5428 
5429 static int handle_desc(struct kvm_vcpu *vcpu)
5430 {
5431 	/*
5432 	 * UMIP emulation relies on intercepting writes to CR4.UMIP, i.e. this
5433 	 * and other code needs to be updated if UMIP can be guest owned.
5434 	 */
5435 	BUILD_BUG_ON(KVM_POSSIBLE_CR4_GUEST_BITS & X86_CR4_UMIP);
5436 
5437 	WARN_ON_ONCE(!kvm_is_cr4_bit_set(vcpu, X86_CR4_UMIP));
5438 	return kvm_emulate_instruction(vcpu, 0);
5439 }
5440 
5441 static int handle_cr(struct kvm_vcpu *vcpu)
5442 {
5443 	unsigned long exit_qualification, val;
5444 	int cr;
5445 	int reg;
5446 	int err;
5447 	int ret;
5448 
5449 	exit_qualification = vmx_get_exit_qual(vcpu);
5450 	cr = exit_qualification & 15;
5451 	reg = (exit_qualification >> 8) & 15;
5452 	switch ((exit_qualification >> 4) & 3) {
5453 	case 0: /* mov to cr */
5454 		val = kvm_register_read(vcpu, reg);
5455 		trace_kvm_cr_write(cr, val);
5456 		switch (cr) {
5457 		case 0:
5458 			err = handle_set_cr0(vcpu, val);
5459 			return kvm_complete_insn_gp(vcpu, err);
5460 		case 3:
5461 			WARN_ON_ONCE(enable_unrestricted_guest);
5462 
5463 			err = kvm_set_cr3(vcpu, val);
5464 			return kvm_complete_insn_gp(vcpu, err);
5465 		case 4:
5466 			err = handle_set_cr4(vcpu, val);
5467 			return kvm_complete_insn_gp(vcpu, err);
5468 		case 8: {
5469 				u8 cr8_prev = kvm_get_cr8(vcpu);
5470 				u8 cr8 = (u8)val;
5471 				err = kvm_set_cr8(vcpu, cr8);
5472 				ret = kvm_complete_insn_gp(vcpu, err);
5473 				if (lapic_in_kernel(vcpu))
5474 					return ret;
5475 				if (cr8_prev <= cr8)
5476 					return ret;
5477 				/*
5478 				 * TODO: we might be squashing a
5479 				 * KVM_GUESTDBG_SINGLESTEP-triggered
5480 				 * KVM_EXIT_DEBUG here.
5481 				 */
5482 				vcpu->run->exit_reason = KVM_EXIT_SET_TPR;
5483 				return 0;
5484 			}
5485 		}
5486 		break;
5487 	case 2: /* clts */
5488 		KVM_BUG(1, vcpu->kvm, "Guest always owns CR0.TS");
5489 		return -EIO;
5490 	case 1: /*mov from cr*/
5491 		switch (cr) {
5492 		case 3:
5493 			WARN_ON_ONCE(enable_unrestricted_guest);
5494 
5495 			val = kvm_read_cr3(vcpu);
5496 			kvm_register_write(vcpu, reg, val);
5497 			trace_kvm_cr_read(cr, val);
5498 			return kvm_skip_emulated_instruction(vcpu);
5499 		case 8:
5500 			val = kvm_get_cr8(vcpu);
5501 			kvm_register_write(vcpu, reg, val);
5502 			trace_kvm_cr_read(cr, val);
5503 			return kvm_skip_emulated_instruction(vcpu);
5504 		}
5505 		break;
5506 	case 3: /* lmsw */
5507 		val = (exit_qualification >> LMSW_SOURCE_DATA_SHIFT) & 0x0f;
5508 		trace_kvm_cr_write(0, (kvm_read_cr0_bits(vcpu, ~0xful) | val));
5509 		kvm_lmsw(vcpu, val);
5510 
5511 		return kvm_skip_emulated_instruction(vcpu);
5512 	default:
5513 		break;
5514 	}
5515 	vcpu->run->exit_reason = 0;
5516 	vcpu_unimpl(vcpu, "unhandled control register: op %d cr %d\n",
5517 	       (int)(exit_qualification >> 4) & 3, cr);
5518 	return 0;
5519 }
5520 
5521 static int handle_dr(struct kvm_vcpu *vcpu)
5522 {
5523 	unsigned long exit_qualification;
5524 	int dr, dr7, reg;
5525 	int err = 1;
5526 
5527 	exit_qualification = vmx_get_exit_qual(vcpu);
5528 	dr = exit_qualification & DEBUG_REG_ACCESS_NUM;
5529 
5530 	/* First, if DR does not exist, trigger UD */
5531 	if (!kvm_require_dr(vcpu, dr))
5532 		return 1;
5533 
5534 	if (vmx_get_cpl(vcpu) > 0)
5535 		goto out;
5536 
5537 	dr7 = vmcs_readl(GUEST_DR7);
5538 	if (dr7 & DR7_GD) {
5539 		/*
5540 		 * As the vm-exit takes precedence over the debug trap, we
5541 		 * need to emulate the latter, either for the host or the
5542 		 * guest debugging itself.
5543 		 */
5544 		if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP) {
5545 			vcpu->run->debug.arch.dr6 = DR6_BD | DR6_ACTIVE_LOW;
5546 			vcpu->run->debug.arch.dr7 = dr7;
5547 			vcpu->run->debug.arch.pc = kvm_get_linear_rip(vcpu);
5548 			vcpu->run->debug.arch.exception = DB_VECTOR;
5549 			vcpu->run->exit_reason = KVM_EXIT_DEBUG;
5550 			return 0;
5551 		} else {
5552 			kvm_queue_exception_p(vcpu, DB_VECTOR, DR6_BD);
5553 			return 1;
5554 		}
5555 	}
5556 
5557 	if (vcpu->guest_debug == 0) {
5558 		exec_controls_clearbit(to_vmx(vcpu), CPU_BASED_MOV_DR_EXITING);
5559 
5560 		/*
5561 		 * No more DR vmexits; force a reload of the debug registers
5562 		 * and reenter on this instruction.  The next vmexit will
5563 		 * retrieve the full state of the debug registers.
5564 		 */
5565 		vcpu->arch.switch_db_regs |= KVM_DEBUGREG_WONT_EXIT;
5566 		return 1;
5567 	}
5568 
5569 	reg = DEBUG_REG_ACCESS_REG(exit_qualification);
5570 	if (exit_qualification & TYPE_MOV_FROM_DR) {
5571 		unsigned long val;
5572 
5573 		kvm_get_dr(vcpu, dr, &val);
5574 		kvm_register_write(vcpu, reg, val);
5575 		err = 0;
5576 	} else {
5577 		err = kvm_set_dr(vcpu, dr, kvm_register_read(vcpu, reg));
5578 	}
5579 
5580 out:
5581 	return kvm_complete_insn_gp(vcpu, err);
5582 }
5583 
5584 static void vmx_sync_dirty_debug_regs(struct kvm_vcpu *vcpu)
5585 {
5586 	get_debugreg(vcpu->arch.db[0], 0);
5587 	get_debugreg(vcpu->arch.db[1], 1);
5588 	get_debugreg(vcpu->arch.db[2], 2);
5589 	get_debugreg(vcpu->arch.db[3], 3);
5590 	get_debugreg(vcpu->arch.dr6, 6);
5591 	vcpu->arch.dr7 = vmcs_readl(GUEST_DR7);
5592 
5593 	vcpu->arch.switch_db_regs &= ~KVM_DEBUGREG_WONT_EXIT;
5594 	exec_controls_setbit(to_vmx(vcpu), CPU_BASED_MOV_DR_EXITING);
5595 
5596 	/*
5597 	 * exc_debug expects dr6 to be cleared after it runs, avoid that it sees
5598 	 * a stale dr6 from the guest.
5599 	 */
5600 	set_debugreg(DR6_RESERVED, 6);
5601 }
5602 
5603 static void vmx_set_dr7(struct kvm_vcpu *vcpu, unsigned long val)
5604 {
5605 	vmcs_writel(GUEST_DR7, val);
5606 }
5607 
5608 static int handle_tpr_below_threshold(struct kvm_vcpu *vcpu)
5609 {
5610 	kvm_apic_update_ppr(vcpu);
5611 	return 1;
5612 }
5613 
5614 static int handle_interrupt_window(struct kvm_vcpu *vcpu)
5615 {
5616 	exec_controls_clearbit(to_vmx(vcpu), CPU_BASED_INTR_WINDOW_EXITING);
5617 
5618 	kvm_make_request(KVM_REQ_EVENT, vcpu);
5619 
5620 	++vcpu->stat.irq_window_exits;
5621 	return 1;
5622 }
5623 
5624 static int handle_invlpg(struct kvm_vcpu *vcpu)
5625 {
5626 	unsigned long exit_qualification = vmx_get_exit_qual(vcpu);
5627 
5628 	kvm_mmu_invlpg(vcpu, exit_qualification);
5629 	return kvm_skip_emulated_instruction(vcpu);
5630 }
5631 
5632 static int handle_apic_access(struct kvm_vcpu *vcpu)
5633 {
5634 	if (likely(fasteoi)) {
5635 		unsigned long exit_qualification = vmx_get_exit_qual(vcpu);
5636 		int access_type, offset;
5637 
5638 		access_type = exit_qualification & APIC_ACCESS_TYPE;
5639 		offset = exit_qualification & APIC_ACCESS_OFFSET;
5640 		/*
5641 		 * Sane guest uses MOV to write EOI, with written value
5642 		 * not cared. So make a short-circuit here by avoiding
5643 		 * heavy instruction emulation.
5644 		 */
5645 		if ((access_type == TYPE_LINEAR_APIC_INST_WRITE) &&
5646 		    (offset == APIC_EOI)) {
5647 			kvm_lapic_set_eoi(vcpu);
5648 			return kvm_skip_emulated_instruction(vcpu);
5649 		}
5650 	}
5651 	return kvm_emulate_instruction(vcpu, 0);
5652 }
5653 
5654 static int handle_apic_eoi_induced(struct kvm_vcpu *vcpu)
5655 {
5656 	unsigned long exit_qualification = vmx_get_exit_qual(vcpu);
5657 	int vector = exit_qualification & 0xff;
5658 
5659 	/* EOI-induced VM exit is trap-like and thus no need to adjust IP */
5660 	kvm_apic_set_eoi_accelerated(vcpu, vector);
5661 	return 1;
5662 }
5663 
5664 static int handle_apic_write(struct kvm_vcpu *vcpu)
5665 {
5666 	unsigned long exit_qualification = vmx_get_exit_qual(vcpu);
5667 
5668 	/*
5669 	 * APIC-write VM-Exit is trap-like, KVM doesn't need to advance RIP and
5670 	 * hardware has done any necessary aliasing, offset adjustments, etc...
5671 	 * for the access.  I.e. the correct value has already been  written to
5672 	 * the vAPIC page for the correct 16-byte chunk.  KVM needs only to
5673 	 * retrieve the register value and emulate the access.
5674 	 */
5675 	u32 offset = exit_qualification & 0xff0;
5676 
5677 	kvm_apic_write_nodecode(vcpu, offset);
5678 	return 1;
5679 }
5680 
5681 static int handle_task_switch(struct kvm_vcpu *vcpu)
5682 {
5683 	struct vcpu_vmx *vmx = to_vmx(vcpu);
5684 	unsigned long exit_qualification;
5685 	bool has_error_code = false;
5686 	u32 error_code = 0;
5687 	u16 tss_selector;
5688 	int reason, type, idt_v, idt_index;
5689 
5690 	idt_v = (vmx->idt_vectoring_info & VECTORING_INFO_VALID_MASK);
5691 	idt_index = (vmx->idt_vectoring_info & VECTORING_INFO_VECTOR_MASK);
5692 	type = (vmx->idt_vectoring_info & VECTORING_INFO_TYPE_MASK);
5693 
5694 	exit_qualification = vmx_get_exit_qual(vcpu);
5695 
5696 	reason = (u32)exit_qualification >> 30;
5697 	if (reason == TASK_SWITCH_GATE && idt_v) {
5698 		switch (type) {
5699 		case INTR_TYPE_NMI_INTR:
5700 			vcpu->arch.nmi_injected = false;
5701 			vmx_set_nmi_mask(vcpu, true);
5702 			break;
5703 		case INTR_TYPE_EXT_INTR:
5704 		case INTR_TYPE_SOFT_INTR:
5705 			kvm_clear_interrupt_queue(vcpu);
5706 			break;
5707 		case INTR_TYPE_HARD_EXCEPTION:
5708 			if (vmx->idt_vectoring_info &
5709 			    VECTORING_INFO_DELIVER_CODE_MASK) {
5710 				has_error_code = true;
5711 				error_code =
5712 					vmcs_read32(IDT_VECTORING_ERROR_CODE);
5713 			}
5714 			fallthrough;
5715 		case INTR_TYPE_SOFT_EXCEPTION:
5716 			kvm_clear_exception_queue(vcpu);
5717 			break;
5718 		default:
5719 			break;
5720 		}
5721 	}
5722 	tss_selector = exit_qualification;
5723 
5724 	if (!idt_v || (type != INTR_TYPE_HARD_EXCEPTION &&
5725 		       type != INTR_TYPE_EXT_INTR &&
5726 		       type != INTR_TYPE_NMI_INTR))
5727 		WARN_ON(!skip_emulated_instruction(vcpu));
5728 
5729 	/*
5730 	 * TODO: What about debug traps on tss switch?
5731 	 *       Are we supposed to inject them and update dr6?
5732 	 */
5733 	return kvm_task_switch(vcpu, tss_selector,
5734 			       type == INTR_TYPE_SOFT_INTR ? idt_index : -1,
5735 			       reason, has_error_code, error_code);
5736 }
5737 
5738 static int handle_ept_violation(struct kvm_vcpu *vcpu)
5739 {
5740 	unsigned long exit_qualification;
5741 	gpa_t gpa;
5742 	u64 error_code;
5743 
5744 	exit_qualification = vmx_get_exit_qual(vcpu);
5745 
5746 	/*
5747 	 * EPT violation happened while executing iret from NMI,
5748 	 * "blocked by NMI" bit has to be set before next VM entry.
5749 	 * There are errata that may cause this bit to not be set:
5750 	 * AAK134, BY25.
5751 	 */
5752 	if (!(to_vmx(vcpu)->idt_vectoring_info & VECTORING_INFO_VALID_MASK) &&
5753 			enable_vnmi &&
5754 			(exit_qualification & INTR_INFO_UNBLOCK_NMI))
5755 		vmcs_set_bits(GUEST_INTERRUPTIBILITY_INFO, GUEST_INTR_STATE_NMI);
5756 
5757 	gpa = vmcs_read64(GUEST_PHYSICAL_ADDRESS);
5758 	trace_kvm_page_fault(vcpu, gpa, exit_qualification);
5759 
5760 	/* Is it a read fault? */
5761 	error_code = (exit_qualification & EPT_VIOLATION_ACC_READ)
5762 		     ? PFERR_USER_MASK : 0;
5763 	/* Is it a write fault? */
5764 	error_code |= (exit_qualification & EPT_VIOLATION_ACC_WRITE)
5765 		      ? PFERR_WRITE_MASK : 0;
5766 	/* Is it a fetch fault? */
5767 	error_code |= (exit_qualification & EPT_VIOLATION_ACC_INSTR)
5768 		      ? PFERR_FETCH_MASK : 0;
5769 	/* ept page table entry is present? */
5770 	error_code |= (exit_qualification & EPT_VIOLATION_RWX_MASK)
5771 		      ? PFERR_PRESENT_MASK : 0;
5772 
5773 	error_code |= (exit_qualification & EPT_VIOLATION_GVA_TRANSLATED) != 0 ?
5774 	       PFERR_GUEST_FINAL_MASK : PFERR_GUEST_PAGE_MASK;
5775 
5776 	vcpu->arch.exit_qualification = exit_qualification;
5777 
5778 	/*
5779 	 * Check that the GPA doesn't exceed physical memory limits, as that is
5780 	 * a guest page fault.  We have to emulate the instruction here, because
5781 	 * if the illegal address is that of a paging structure, then
5782 	 * EPT_VIOLATION_ACC_WRITE bit is set.  Alternatively, if supported we
5783 	 * would also use advanced VM-exit information for EPT violations to
5784 	 * reconstruct the page fault error code.
5785 	 */
5786 	if (unlikely(allow_smaller_maxphyaddr && kvm_vcpu_is_illegal_gpa(vcpu, gpa)))
5787 		return kvm_emulate_instruction(vcpu, 0);
5788 
5789 	return kvm_mmu_page_fault(vcpu, gpa, error_code, NULL, 0);
5790 }
5791 
5792 static int handle_ept_misconfig(struct kvm_vcpu *vcpu)
5793 {
5794 	gpa_t gpa;
5795 
5796 	if (!vmx_can_emulate_instruction(vcpu, EMULTYPE_PF, NULL, 0))
5797 		return 1;
5798 
5799 	/*
5800 	 * A nested guest cannot optimize MMIO vmexits, because we have an
5801 	 * nGPA here instead of the required GPA.
5802 	 */
5803 	gpa = vmcs_read64(GUEST_PHYSICAL_ADDRESS);
5804 	if (!is_guest_mode(vcpu) &&
5805 	    !kvm_io_bus_write(vcpu, KVM_FAST_MMIO_BUS, gpa, 0, NULL)) {
5806 		trace_kvm_fast_mmio(gpa);
5807 		return kvm_skip_emulated_instruction(vcpu);
5808 	}
5809 
5810 	return kvm_mmu_page_fault(vcpu, gpa, PFERR_RSVD_MASK, NULL, 0);
5811 }
5812 
5813 static int handle_nmi_window(struct kvm_vcpu *vcpu)
5814 {
5815 	if (KVM_BUG_ON(!enable_vnmi, vcpu->kvm))
5816 		return -EIO;
5817 
5818 	exec_controls_clearbit(to_vmx(vcpu), CPU_BASED_NMI_WINDOW_EXITING);
5819 	++vcpu->stat.nmi_window_exits;
5820 	kvm_make_request(KVM_REQ_EVENT, vcpu);
5821 
5822 	return 1;
5823 }
5824 
5825 static bool vmx_emulation_required_with_pending_exception(struct kvm_vcpu *vcpu)
5826 {
5827 	struct vcpu_vmx *vmx = to_vmx(vcpu);
5828 
5829 	return vmx->emulation_required && !vmx->rmode.vm86_active &&
5830 	       (kvm_is_exception_pending(vcpu) || vcpu->arch.exception.injected);
5831 }
5832 
5833 static int handle_invalid_guest_state(struct kvm_vcpu *vcpu)
5834 {
5835 	struct vcpu_vmx *vmx = to_vmx(vcpu);
5836 	bool intr_window_requested;
5837 	unsigned count = 130;
5838 
5839 	intr_window_requested = exec_controls_get(vmx) &
5840 				CPU_BASED_INTR_WINDOW_EXITING;
5841 
5842 	while (vmx->emulation_required && count-- != 0) {
5843 		if (intr_window_requested && !vmx_interrupt_blocked(vcpu))
5844 			return handle_interrupt_window(&vmx->vcpu);
5845 
5846 		if (kvm_test_request(KVM_REQ_EVENT, vcpu))
5847 			return 1;
5848 
5849 		if (!kvm_emulate_instruction(vcpu, 0))
5850 			return 0;
5851 
5852 		if (vmx_emulation_required_with_pending_exception(vcpu)) {
5853 			kvm_prepare_emulation_failure_exit(vcpu);
5854 			return 0;
5855 		}
5856 
5857 		if (vcpu->arch.halt_request) {
5858 			vcpu->arch.halt_request = 0;
5859 			return kvm_emulate_halt_noskip(vcpu);
5860 		}
5861 
5862 		/*
5863 		 * Note, return 1 and not 0, vcpu_run() will invoke
5864 		 * xfer_to_guest_mode() which will create a proper return
5865 		 * code.
5866 		 */
5867 		if (__xfer_to_guest_mode_work_pending())
5868 			return 1;
5869 	}
5870 
5871 	return 1;
5872 }
5873 
5874 static int vmx_vcpu_pre_run(struct kvm_vcpu *vcpu)
5875 {
5876 	if (vmx_emulation_required_with_pending_exception(vcpu)) {
5877 		kvm_prepare_emulation_failure_exit(vcpu);
5878 		return 0;
5879 	}
5880 
5881 	return 1;
5882 }
5883 
5884 static void grow_ple_window(struct kvm_vcpu *vcpu)
5885 {
5886 	struct vcpu_vmx *vmx = to_vmx(vcpu);
5887 	unsigned int old = vmx->ple_window;
5888 
5889 	vmx->ple_window = __grow_ple_window(old, ple_window,
5890 					    ple_window_grow,
5891 					    ple_window_max);
5892 
5893 	if (vmx->ple_window != old) {
5894 		vmx->ple_window_dirty = true;
5895 		trace_kvm_ple_window_update(vcpu->vcpu_id,
5896 					    vmx->ple_window, old);
5897 	}
5898 }
5899 
5900 static void shrink_ple_window(struct kvm_vcpu *vcpu)
5901 {
5902 	struct vcpu_vmx *vmx = to_vmx(vcpu);
5903 	unsigned int old = vmx->ple_window;
5904 
5905 	vmx->ple_window = __shrink_ple_window(old, ple_window,
5906 					      ple_window_shrink,
5907 					      ple_window);
5908 
5909 	if (vmx->ple_window != old) {
5910 		vmx->ple_window_dirty = true;
5911 		trace_kvm_ple_window_update(vcpu->vcpu_id,
5912 					    vmx->ple_window, old);
5913 	}
5914 }
5915 
5916 /*
5917  * Indicate a busy-waiting vcpu in spinlock. We do not enable the PAUSE
5918  * exiting, so only get here on cpu with PAUSE-Loop-Exiting.
5919  */
5920 static int handle_pause(struct kvm_vcpu *vcpu)
5921 {
5922 	if (!kvm_pause_in_guest(vcpu->kvm))
5923 		grow_ple_window(vcpu);
5924 
5925 	/*
5926 	 * Intel sdm vol3 ch-25.1.3 says: The "PAUSE-loop exiting"
5927 	 * VM-execution control is ignored if CPL > 0. OTOH, KVM
5928 	 * never set PAUSE_EXITING and just set PLE if supported,
5929 	 * so the vcpu must be CPL=0 if it gets a PAUSE exit.
5930 	 */
5931 	kvm_vcpu_on_spin(vcpu, true);
5932 	return kvm_skip_emulated_instruction(vcpu);
5933 }
5934 
5935 static int handle_monitor_trap(struct kvm_vcpu *vcpu)
5936 {
5937 	return 1;
5938 }
5939 
5940 static int handle_invpcid(struct kvm_vcpu *vcpu)
5941 {
5942 	u32 vmx_instruction_info;
5943 	unsigned long type;
5944 	gva_t gva;
5945 	struct {
5946 		u64 pcid;
5947 		u64 gla;
5948 	} operand;
5949 	int gpr_index;
5950 
5951 	if (!guest_cpuid_has(vcpu, X86_FEATURE_INVPCID)) {
5952 		kvm_queue_exception(vcpu, UD_VECTOR);
5953 		return 1;
5954 	}
5955 
5956 	vmx_instruction_info = vmcs_read32(VMX_INSTRUCTION_INFO);
5957 	gpr_index = vmx_get_instr_info_reg2(vmx_instruction_info);
5958 	type = kvm_register_read(vcpu, gpr_index);
5959 
5960 	/* According to the Intel instruction reference, the memory operand
5961 	 * is read even if it isn't needed (e.g., for type==all)
5962 	 */
5963 	if (get_vmx_mem_address(vcpu, vmx_get_exit_qual(vcpu),
5964 				vmx_instruction_info, false,
5965 				sizeof(operand), &gva))
5966 		return 1;
5967 
5968 	return kvm_handle_invpcid(vcpu, type, gva);
5969 }
5970 
5971 static int handle_pml_full(struct kvm_vcpu *vcpu)
5972 {
5973 	unsigned long exit_qualification;
5974 
5975 	trace_kvm_pml_full(vcpu->vcpu_id);
5976 
5977 	exit_qualification = vmx_get_exit_qual(vcpu);
5978 
5979 	/*
5980 	 * PML buffer FULL happened while executing iret from NMI,
5981 	 * "blocked by NMI" bit has to be set before next VM entry.
5982 	 */
5983 	if (!(to_vmx(vcpu)->idt_vectoring_info & VECTORING_INFO_VALID_MASK) &&
5984 			enable_vnmi &&
5985 			(exit_qualification & INTR_INFO_UNBLOCK_NMI))
5986 		vmcs_set_bits(GUEST_INTERRUPTIBILITY_INFO,
5987 				GUEST_INTR_STATE_NMI);
5988 
5989 	/*
5990 	 * PML buffer already flushed at beginning of VMEXIT. Nothing to do
5991 	 * here.., and there's no userspace involvement needed for PML.
5992 	 */
5993 	return 1;
5994 }
5995 
5996 static fastpath_t handle_fastpath_preemption_timer(struct kvm_vcpu *vcpu)
5997 {
5998 	struct vcpu_vmx *vmx = to_vmx(vcpu);
5999 
6000 	if (!vmx->req_immediate_exit &&
6001 	    !unlikely(vmx->loaded_vmcs->hv_timer_soft_disabled)) {
6002 		kvm_lapic_expired_hv_timer(vcpu);
6003 		return EXIT_FASTPATH_REENTER_GUEST;
6004 	}
6005 
6006 	return EXIT_FASTPATH_NONE;
6007 }
6008 
6009 static int handle_preemption_timer(struct kvm_vcpu *vcpu)
6010 {
6011 	handle_fastpath_preemption_timer(vcpu);
6012 	return 1;
6013 }
6014 
6015 /*
6016  * When nested=0, all VMX instruction VM Exits filter here.  The handlers
6017  * are overwritten by nested_vmx_setup() when nested=1.
6018  */
6019 static int handle_vmx_instruction(struct kvm_vcpu *vcpu)
6020 {
6021 	kvm_queue_exception(vcpu, UD_VECTOR);
6022 	return 1;
6023 }
6024 
6025 #ifndef CONFIG_X86_SGX_KVM
6026 static int handle_encls(struct kvm_vcpu *vcpu)
6027 {
6028 	/*
6029 	 * SGX virtualization is disabled.  There is no software enable bit for
6030 	 * SGX, so KVM intercepts all ENCLS leafs and injects a #UD to prevent
6031 	 * the guest from executing ENCLS (when SGX is supported by hardware).
6032 	 */
6033 	kvm_queue_exception(vcpu, UD_VECTOR);
6034 	return 1;
6035 }
6036 #endif /* CONFIG_X86_SGX_KVM */
6037 
6038 static int handle_bus_lock_vmexit(struct kvm_vcpu *vcpu)
6039 {
6040 	/*
6041 	 * Hardware may or may not set the BUS_LOCK_DETECTED flag on BUS_LOCK
6042 	 * VM-Exits. Unconditionally set the flag here and leave the handling to
6043 	 * vmx_handle_exit().
6044 	 */
6045 	to_vmx(vcpu)->exit_reason.bus_lock_detected = true;
6046 	return 1;
6047 }
6048 
6049 static int handle_notify(struct kvm_vcpu *vcpu)
6050 {
6051 	unsigned long exit_qual = vmx_get_exit_qual(vcpu);
6052 	bool context_invalid = exit_qual & NOTIFY_VM_CONTEXT_INVALID;
6053 
6054 	++vcpu->stat.notify_window_exits;
6055 
6056 	/*
6057 	 * Notify VM exit happened while executing iret from NMI,
6058 	 * "blocked by NMI" bit has to be set before next VM entry.
6059 	 */
6060 	if (enable_vnmi && (exit_qual & INTR_INFO_UNBLOCK_NMI))
6061 		vmcs_set_bits(GUEST_INTERRUPTIBILITY_INFO,
6062 			      GUEST_INTR_STATE_NMI);
6063 
6064 	if (vcpu->kvm->arch.notify_vmexit_flags & KVM_X86_NOTIFY_VMEXIT_USER ||
6065 	    context_invalid) {
6066 		vcpu->run->exit_reason = KVM_EXIT_NOTIFY;
6067 		vcpu->run->notify.flags = context_invalid ?
6068 					  KVM_NOTIFY_CONTEXT_INVALID : 0;
6069 		return 0;
6070 	}
6071 
6072 	return 1;
6073 }
6074 
6075 /*
6076  * The exit handlers return 1 if the exit was handled fully and guest execution
6077  * may resume.  Otherwise they set the kvm_run parameter to indicate what needs
6078  * to be done to userspace and return 0.
6079  */
6080 static int (*kvm_vmx_exit_handlers[])(struct kvm_vcpu *vcpu) = {
6081 	[EXIT_REASON_EXCEPTION_NMI]           = handle_exception_nmi,
6082 	[EXIT_REASON_EXTERNAL_INTERRUPT]      = handle_external_interrupt,
6083 	[EXIT_REASON_TRIPLE_FAULT]            = handle_triple_fault,
6084 	[EXIT_REASON_NMI_WINDOW]	      = handle_nmi_window,
6085 	[EXIT_REASON_IO_INSTRUCTION]          = handle_io,
6086 	[EXIT_REASON_CR_ACCESS]               = handle_cr,
6087 	[EXIT_REASON_DR_ACCESS]               = handle_dr,
6088 	[EXIT_REASON_CPUID]                   = kvm_emulate_cpuid,
6089 	[EXIT_REASON_MSR_READ]                = kvm_emulate_rdmsr,
6090 	[EXIT_REASON_MSR_WRITE]               = kvm_emulate_wrmsr,
6091 	[EXIT_REASON_INTERRUPT_WINDOW]        = handle_interrupt_window,
6092 	[EXIT_REASON_HLT]                     = kvm_emulate_halt,
6093 	[EXIT_REASON_INVD]		      = kvm_emulate_invd,
6094 	[EXIT_REASON_INVLPG]		      = handle_invlpg,
6095 	[EXIT_REASON_RDPMC]                   = kvm_emulate_rdpmc,
6096 	[EXIT_REASON_VMCALL]                  = kvm_emulate_hypercall,
6097 	[EXIT_REASON_VMCLEAR]		      = handle_vmx_instruction,
6098 	[EXIT_REASON_VMLAUNCH]		      = handle_vmx_instruction,
6099 	[EXIT_REASON_VMPTRLD]		      = handle_vmx_instruction,
6100 	[EXIT_REASON_VMPTRST]		      = handle_vmx_instruction,
6101 	[EXIT_REASON_VMREAD]		      = handle_vmx_instruction,
6102 	[EXIT_REASON_VMRESUME]		      = handle_vmx_instruction,
6103 	[EXIT_REASON_VMWRITE]		      = handle_vmx_instruction,
6104 	[EXIT_REASON_VMOFF]		      = handle_vmx_instruction,
6105 	[EXIT_REASON_VMON]		      = handle_vmx_instruction,
6106 	[EXIT_REASON_TPR_BELOW_THRESHOLD]     = handle_tpr_below_threshold,
6107 	[EXIT_REASON_APIC_ACCESS]             = handle_apic_access,
6108 	[EXIT_REASON_APIC_WRITE]              = handle_apic_write,
6109 	[EXIT_REASON_EOI_INDUCED]             = handle_apic_eoi_induced,
6110 	[EXIT_REASON_WBINVD]                  = kvm_emulate_wbinvd,
6111 	[EXIT_REASON_XSETBV]                  = kvm_emulate_xsetbv,
6112 	[EXIT_REASON_TASK_SWITCH]             = handle_task_switch,
6113 	[EXIT_REASON_MCE_DURING_VMENTRY]      = handle_machine_check,
6114 	[EXIT_REASON_GDTR_IDTR]		      = handle_desc,
6115 	[EXIT_REASON_LDTR_TR]		      = handle_desc,
6116 	[EXIT_REASON_EPT_VIOLATION]	      = handle_ept_violation,
6117 	[EXIT_REASON_EPT_MISCONFIG]           = handle_ept_misconfig,
6118 	[EXIT_REASON_PAUSE_INSTRUCTION]       = handle_pause,
6119 	[EXIT_REASON_MWAIT_INSTRUCTION]	      = kvm_emulate_mwait,
6120 	[EXIT_REASON_MONITOR_TRAP_FLAG]       = handle_monitor_trap,
6121 	[EXIT_REASON_MONITOR_INSTRUCTION]     = kvm_emulate_monitor,
6122 	[EXIT_REASON_INVEPT]                  = handle_vmx_instruction,
6123 	[EXIT_REASON_INVVPID]                 = handle_vmx_instruction,
6124 	[EXIT_REASON_RDRAND]                  = kvm_handle_invalid_op,
6125 	[EXIT_REASON_RDSEED]                  = kvm_handle_invalid_op,
6126 	[EXIT_REASON_PML_FULL]		      = handle_pml_full,
6127 	[EXIT_REASON_INVPCID]                 = handle_invpcid,
6128 	[EXIT_REASON_VMFUNC]		      = handle_vmx_instruction,
6129 	[EXIT_REASON_PREEMPTION_TIMER]	      = handle_preemption_timer,
6130 	[EXIT_REASON_ENCLS]		      = handle_encls,
6131 	[EXIT_REASON_BUS_LOCK]                = handle_bus_lock_vmexit,
6132 	[EXIT_REASON_NOTIFY]		      = handle_notify,
6133 };
6134 
6135 static const int kvm_vmx_max_exit_handlers =
6136 	ARRAY_SIZE(kvm_vmx_exit_handlers);
6137 
6138 static void vmx_get_exit_info(struct kvm_vcpu *vcpu, u32 *reason,
6139 			      u64 *info1, u64 *info2,
6140 			      u32 *intr_info, u32 *error_code)
6141 {
6142 	struct vcpu_vmx *vmx = to_vmx(vcpu);
6143 
6144 	*reason = vmx->exit_reason.full;
6145 	*info1 = vmx_get_exit_qual(vcpu);
6146 	if (!(vmx->exit_reason.failed_vmentry)) {
6147 		*info2 = vmx->idt_vectoring_info;
6148 		*intr_info = vmx_get_intr_info(vcpu);
6149 		if (is_exception_with_error_code(*intr_info))
6150 			*error_code = vmcs_read32(VM_EXIT_INTR_ERROR_CODE);
6151 		else
6152 			*error_code = 0;
6153 	} else {
6154 		*info2 = 0;
6155 		*intr_info = 0;
6156 		*error_code = 0;
6157 	}
6158 }
6159 
6160 static void vmx_destroy_pml_buffer(struct vcpu_vmx *vmx)
6161 {
6162 	if (vmx->pml_pg) {
6163 		__free_page(vmx->pml_pg);
6164 		vmx->pml_pg = NULL;
6165 	}
6166 }
6167 
6168 static void vmx_flush_pml_buffer(struct kvm_vcpu *vcpu)
6169 {
6170 	struct vcpu_vmx *vmx = to_vmx(vcpu);
6171 	u64 *pml_buf;
6172 	u16 pml_idx;
6173 
6174 	pml_idx = vmcs_read16(GUEST_PML_INDEX);
6175 
6176 	/* Do nothing if PML buffer is empty */
6177 	if (pml_idx == (PML_ENTITY_NUM - 1))
6178 		return;
6179 
6180 	/* PML index always points to next available PML buffer entity */
6181 	if (pml_idx >= PML_ENTITY_NUM)
6182 		pml_idx = 0;
6183 	else
6184 		pml_idx++;
6185 
6186 	pml_buf = page_address(vmx->pml_pg);
6187 	for (; pml_idx < PML_ENTITY_NUM; pml_idx++) {
6188 		u64 gpa;
6189 
6190 		gpa = pml_buf[pml_idx];
6191 		WARN_ON(gpa & (PAGE_SIZE - 1));
6192 		kvm_vcpu_mark_page_dirty(vcpu, gpa >> PAGE_SHIFT);
6193 	}
6194 
6195 	/* reset PML index */
6196 	vmcs_write16(GUEST_PML_INDEX, PML_ENTITY_NUM - 1);
6197 }
6198 
6199 static void vmx_dump_sel(char *name, uint32_t sel)
6200 {
6201 	pr_err("%s sel=0x%04x, attr=0x%05x, limit=0x%08x, base=0x%016lx\n",
6202 	       name, vmcs_read16(sel),
6203 	       vmcs_read32(sel + GUEST_ES_AR_BYTES - GUEST_ES_SELECTOR),
6204 	       vmcs_read32(sel + GUEST_ES_LIMIT - GUEST_ES_SELECTOR),
6205 	       vmcs_readl(sel + GUEST_ES_BASE - GUEST_ES_SELECTOR));
6206 }
6207 
6208 static void vmx_dump_dtsel(char *name, uint32_t limit)
6209 {
6210 	pr_err("%s                           limit=0x%08x, base=0x%016lx\n",
6211 	       name, vmcs_read32(limit),
6212 	       vmcs_readl(limit + GUEST_GDTR_BASE - GUEST_GDTR_LIMIT));
6213 }
6214 
6215 static void vmx_dump_msrs(char *name, struct vmx_msrs *m)
6216 {
6217 	unsigned int i;
6218 	struct vmx_msr_entry *e;
6219 
6220 	pr_err("MSR %s:\n", name);
6221 	for (i = 0, e = m->val; i < m->nr; ++i, ++e)
6222 		pr_err("  %2d: msr=0x%08x value=0x%016llx\n", i, e->index, e->value);
6223 }
6224 
6225 void dump_vmcs(struct kvm_vcpu *vcpu)
6226 {
6227 	struct vcpu_vmx *vmx = to_vmx(vcpu);
6228 	u32 vmentry_ctl, vmexit_ctl;
6229 	u32 cpu_based_exec_ctrl, pin_based_exec_ctrl, secondary_exec_control;
6230 	u64 tertiary_exec_control;
6231 	unsigned long cr4;
6232 	int efer_slot;
6233 
6234 	if (!dump_invalid_vmcs) {
6235 		pr_warn_ratelimited("set kvm_intel.dump_invalid_vmcs=1 to dump internal KVM state.\n");
6236 		return;
6237 	}
6238 
6239 	vmentry_ctl = vmcs_read32(VM_ENTRY_CONTROLS);
6240 	vmexit_ctl = vmcs_read32(VM_EXIT_CONTROLS);
6241 	cpu_based_exec_ctrl = vmcs_read32(CPU_BASED_VM_EXEC_CONTROL);
6242 	pin_based_exec_ctrl = vmcs_read32(PIN_BASED_VM_EXEC_CONTROL);
6243 	cr4 = vmcs_readl(GUEST_CR4);
6244 
6245 	if (cpu_has_secondary_exec_ctrls())
6246 		secondary_exec_control = vmcs_read32(SECONDARY_VM_EXEC_CONTROL);
6247 	else
6248 		secondary_exec_control = 0;
6249 
6250 	if (cpu_has_tertiary_exec_ctrls())
6251 		tertiary_exec_control = vmcs_read64(TERTIARY_VM_EXEC_CONTROL);
6252 	else
6253 		tertiary_exec_control = 0;
6254 
6255 	pr_err("VMCS %p, last attempted VM-entry on CPU %d\n",
6256 	       vmx->loaded_vmcs->vmcs, vcpu->arch.last_vmentry_cpu);
6257 	pr_err("*** Guest State ***\n");
6258 	pr_err("CR0: actual=0x%016lx, shadow=0x%016lx, gh_mask=%016lx\n",
6259 	       vmcs_readl(GUEST_CR0), vmcs_readl(CR0_READ_SHADOW),
6260 	       vmcs_readl(CR0_GUEST_HOST_MASK));
6261 	pr_err("CR4: actual=0x%016lx, shadow=0x%016lx, gh_mask=%016lx\n",
6262 	       cr4, vmcs_readl(CR4_READ_SHADOW), vmcs_readl(CR4_GUEST_HOST_MASK));
6263 	pr_err("CR3 = 0x%016lx\n", vmcs_readl(GUEST_CR3));
6264 	if (cpu_has_vmx_ept()) {
6265 		pr_err("PDPTR0 = 0x%016llx  PDPTR1 = 0x%016llx\n",
6266 		       vmcs_read64(GUEST_PDPTR0), vmcs_read64(GUEST_PDPTR1));
6267 		pr_err("PDPTR2 = 0x%016llx  PDPTR3 = 0x%016llx\n",
6268 		       vmcs_read64(GUEST_PDPTR2), vmcs_read64(GUEST_PDPTR3));
6269 	}
6270 	pr_err("RSP = 0x%016lx  RIP = 0x%016lx\n",
6271 	       vmcs_readl(GUEST_RSP), vmcs_readl(GUEST_RIP));
6272 	pr_err("RFLAGS=0x%08lx         DR7 = 0x%016lx\n",
6273 	       vmcs_readl(GUEST_RFLAGS), vmcs_readl(GUEST_DR7));
6274 	pr_err("Sysenter RSP=%016lx CS:RIP=%04x:%016lx\n",
6275 	       vmcs_readl(GUEST_SYSENTER_ESP),
6276 	       vmcs_read32(GUEST_SYSENTER_CS), vmcs_readl(GUEST_SYSENTER_EIP));
6277 	vmx_dump_sel("CS:  ", GUEST_CS_SELECTOR);
6278 	vmx_dump_sel("DS:  ", GUEST_DS_SELECTOR);
6279 	vmx_dump_sel("SS:  ", GUEST_SS_SELECTOR);
6280 	vmx_dump_sel("ES:  ", GUEST_ES_SELECTOR);
6281 	vmx_dump_sel("FS:  ", GUEST_FS_SELECTOR);
6282 	vmx_dump_sel("GS:  ", GUEST_GS_SELECTOR);
6283 	vmx_dump_dtsel("GDTR:", GUEST_GDTR_LIMIT);
6284 	vmx_dump_sel("LDTR:", GUEST_LDTR_SELECTOR);
6285 	vmx_dump_dtsel("IDTR:", GUEST_IDTR_LIMIT);
6286 	vmx_dump_sel("TR:  ", GUEST_TR_SELECTOR);
6287 	efer_slot = vmx_find_loadstore_msr_slot(&vmx->msr_autoload.guest, MSR_EFER);
6288 	if (vmentry_ctl & VM_ENTRY_LOAD_IA32_EFER)
6289 		pr_err("EFER= 0x%016llx\n", vmcs_read64(GUEST_IA32_EFER));
6290 	else if (efer_slot >= 0)
6291 		pr_err("EFER= 0x%016llx (autoload)\n",
6292 		       vmx->msr_autoload.guest.val[efer_slot].value);
6293 	else if (vmentry_ctl & VM_ENTRY_IA32E_MODE)
6294 		pr_err("EFER= 0x%016llx (effective)\n",
6295 		       vcpu->arch.efer | (EFER_LMA | EFER_LME));
6296 	else
6297 		pr_err("EFER= 0x%016llx (effective)\n",
6298 		       vcpu->arch.efer & ~(EFER_LMA | EFER_LME));
6299 	if (vmentry_ctl & VM_ENTRY_LOAD_IA32_PAT)
6300 		pr_err("PAT = 0x%016llx\n", vmcs_read64(GUEST_IA32_PAT));
6301 	pr_err("DebugCtl = 0x%016llx  DebugExceptions = 0x%016lx\n",
6302 	       vmcs_read64(GUEST_IA32_DEBUGCTL),
6303 	       vmcs_readl(GUEST_PENDING_DBG_EXCEPTIONS));
6304 	if (cpu_has_load_perf_global_ctrl() &&
6305 	    vmentry_ctl & VM_ENTRY_LOAD_IA32_PERF_GLOBAL_CTRL)
6306 		pr_err("PerfGlobCtl = 0x%016llx\n",
6307 		       vmcs_read64(GUEST_IA32_PERF_GLOBAL_CTRL));
6308 	if (vmentry_ctl & VM_ENTRY_LOAD_BNDCFGS)
6309 		pr_err("BndCfgS = 0x%016llx\n", vmcs_read64(GUEST_BNDCFGS));
6310 	pr_err("Interruptibility = %08x  ActivityState = %08x\n",
6311 	       vmcs_read32(GUEST_INTERRUPTIBILITY_INFO),
6312 	       vmcs_read32(GUEST_ACTIVITY_STATE));
6313 	if (secondary_exec_control & SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY)
6314 		pr_err("InterruptStatus = %04x\n",
6315 		       vmcs_read16(GUEST_INTR_STATUS));
6316 	if (vmcs_read32(VM_ENTRY_MSR_LOAD_COUNT) > 0)
6317 		vmx_dump_msrs("guest autoload", &vmx->msr_autoload.guest);
6318 	if (vmcs_read32(VM_EXIT_MSR_STORE_COUNT) > 0)
6319 		vmx_dump_msrs("guest autostore", &vmx->msr_autostore.guest);
6320 
6321 	pr_err("*** Host State ***\n");
6322 	pr_err("RIP = 0x%016lx  RSP = 0x%016lx\n",
6323 	       vmcs_readl(HOST_RIP), vmcs_readl(HOST_RSP));
6324 	pr_err("CS=%04x SS=%04x DS=%04x ES=%04x FS=%04x GS=%04x TR=%04x\n",
6325 	       vmcs_read16(HOST_CS_SELECTOR), vmcs_read16(HOST_SS_SELECTOR),
6326 	       vmcs_read16(HOST_DS_SELECTOR), vmcs_read16(HOST_ES_SELECTOR),
6327 	       vmcs_read16(HOST_FS_SELECTOR), vmcs_read16(HOST_GS_SELECTOR),
6328 	       vmcs_read16(HOST_TR_SELECTOR));
6329 	pr_err("FSBase=%016lx GSBase=%016lx TRBase=%016lx\n",
6330 	       vmcs_readl(HOST_FS_BASE), vmcs_readl(HOST_GS_BASE),
6331 	       vmcs_readl(HOST_TR_BASE));
6332 	pr_err("GDTBase=%016lx IDTBase=%016lx\n",
6333 	       vmcs_readl(HOST_GDTR_BASE), vmcs_readl(HOST_IDTR_BASE));
6334 	pr_err("CR0=%016lx CR3=%016lx CR4=%016lx\n",
6335 	       vmcs_readl(HOST_CR0), vmcs_readl(HOST_CR3),
6336 	       vmcs_readl(HOST_CR4));
6337 	pr_err("Sysenter RSP=%016lx CS:RIP=%04x:%016lx\n",
6338 	       vmcs_readl(HOST_IA32_SYSENTER_ESP),
6339 	       vmcs_read32(HOST_IA32_SYSENTER_CS),
6340 	       vmcs_readl(HOST_IA32_SYSENTER_EIP));
6341 	if (vmexit_ctl & VM_EXIT_LOAD_IA32_EFER)
6342 		pr_err("EFER= 0x%016llx\n", vmcs_read64(HOST_IA32_EFER));
6343 	if (vmexit_ctl & VM_EXIT_LOAD_IA32_PAT)
6344 		pr_err("PAT = 0x%016llx\n", vmcs_read64(HOST_IA32_PAT));
6345 	if (cpu_has_load_perf_global_ctrl() &&
6346 	    vmexit_ctl & VM_EXIT_LOAD_IA32_PERF_GLOBAL_CTRL)
6347 		pr_err("PerfGlobCtl = 0x%016llx\n",
6348 		       vmcs_read64(HOST_IA32_PERF_GLOBAL_CTRL));
6349 	if (vmcs_read32(VM_EXIT_MSR_LOAD_COUNT) > 0)
6350 		vmx_dump_msrs("host autoload", &vmx->msr_autoload.host);
6351 
6352 	pr_err("*** Control State ***\n");
6353 	pr_err("CPUBased=0x%08x SecondaryExec=0x%08x TertiaryExec=0x%016llx\n",
6354 	       cpu_based_exec_ctrl, secondary_exec_control, tertiary_exec_control);
6355 	pr_err("PinBased=0x%08x EntryControls=%08x ExitControls=%08x\n",
6356 	       pin_based_exec_ctrl, vmentry_ctl, vmexit_ctl);
6357 	pr_err("ExceptionBitmap=%08x PFECmask=%08x PFECmatch=%08x\n",
6358 	       vmcs_read32(EXCEPTION_BITMAP),
6359 	       vmcs_read32(PAGE_FAULT_ERROR_CODE_MASK),
6360 	       vmcs_read32(PAGE_FAULT_ERROR_CODE_MATCH));
6361 	pr_err("VMEntry: intr_info=%08x errcode=%08x ilen=%08x\n",
6362 	       vmcs_read32(VM_ENTRY_INTR_INFO_FIELD),
6363 	       vmcs_read32(VM_ENTRY_EXCEPTION_ERROR_CODE),
6364 	       vmcs_read32(VM_ENTRY_INSTRUCTION_LEN));
6365 	pr_err("VMExit: intr_info=%08x errcode=%08x ilen=%08x\n",
6366 	       vmcs_read32(VM_EXIT_INTR_INFO),
6367 	       vmcs_read32(VM_EXIT_INTR_ERROR_CODE),
6368 	       vmcs_read32(VM_EXIT_INSTRUCTION_LEN));
6369 	pr_err("        reason=%08x qualification=%016lx\n",
6370 	       vmcs_read32(VM_EXIT_REASON), vmcs_readl(EXIT_QUALIFICATION));
6371 	pr_err("IDTVectoring: info=%08x errcode=%08x\n",
6372 	       vmcs_read32(IDT_VECTORING_INFO_FIELD),
6373 	       vmcs_read32(IDT_VECTORING_ERROR_CODE));
6374 	pr_err("TSC Offset = 0x%016llx\n", vmcs_read64(TSC_OFFSET));
6375 	if (secondary_exec_control & SECONDARY_EXEC_TSC_SCALING)
6376 		pr_err("TSC Multiplier = 0x%016llx\n",
6377 		       vmcs_read64(TSC_MULTIPLIER));
6378 	if (cpu_based_exec_ctrl & CPU_BASED_TPR_SHADOW) {
6379 		if (secondary_exec_control & SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY) {
6380 			u16 status = vmcs_read16(GUEST_INTR_STATUS);
6381 			pr_err("SVI|RVI = %02x|%02x ", status >> 8, status & 0xff);
6382 		}
6383 		pr_cont("TPR Threshold = 0x%02x\n", vmcs_read32(TPR_THRESHOLD));
6384 		if (secondary_exec_control & SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES)
6385 			pr_err("APIC-access addr = 0x%016llx ", vmcs_read64(APIC_ACCESS_ADDR));
6386 		pr_cont("virt-APIC addr = 0x%016llx\n", vmcs_read64(VIRTUAL_APIC_PAGE_ADDR));
6387 	}
6388 	if (pin_based_exec_ctrl & PIN_BASED_POSTED_INTR)
6389 		pr_err("PostedIntrVec = 0x%02x\n", vmcs_read16(POSTED_INTR_NV));
6390 	if ((secondary_exec_control & SECONDARY_EXEC_ENABLE_EPT))
6391 		pr_err("EPT pointer = 0x%016llx\n", vmcs_read64(EPT_POINTER));
6392 	if (secondary_exec_control & SECONDARY_EXEC_PAUSE_LOOP_EXITING)
6393 		pr_err("PLE Gap=%08x Window=%08x\n",
6394 		       vmcs_read32(PLE_GAP), vmcs_read32(PLE_WINDOW));
6395 	if (secondary_exec_control & SECONDARY_EXEC_ENABLE_VPID)
6396 		pr_err("Virtual processor ID = 0x%04x\n",
6397 		       vmcs_read16(VIRTUAL_PROCESSOR_ID));
6398 }
6399 
6400 /*
6401  * The guest has exited.  See if we can fix it or if we need userspace
6402  * assistance.
6403  */
6404 static int __vmx_handle_exit(struct kvm_vcpu *vcpu, fastpath_t exit_fastpath)
6405 {
6406 	struct vcpu_vmx *vmx = to_vmx(vcpu);
6407 	union vmx_exit_reason exit_reason = vmx->exit_reason;
6408 	u32 vectoring_info = vmx->idt_vectoring_info;
6409 	u16 exit_handler_index;
6410 
6411 	/*
6412 	 * Flush logged GPAs PML buffer, this will make dirty_bitmap more
6413 	 * updated. Another good is, in kvm_vm_ioctl_get_dirty_log, before
6414 	 * querying dirty_bitmap, we only need to kick all vcpus out of guest
6415 	 * mode as if vcpus is in root mode, the PML buffer must has been
6416 	 * flushed already.  Note, PML is never enabled in hardware while
6417 	 * running L2.
6418 	 */
6419 	if (enable_pml && !is_guest_mode(vcpu))
6420 		vmx_flush_pml_buffer(vcpu);
6421 
6422 	/*
6423 	 * KVM should never reach this point with a pending nested VM-Enter.
6424 	 * More specifically, short-circuiting VM-Entry to emulate L2 due to
6425 	 * invalid guest state should never happen as that means KVM knowingly
6426 	 * allowed a nested VM-Enter with an invalid vmcs12.  More below.
6427 	 */
6428 	if (KVM_BUG_ON(vmx->nested.nested_run_pending, vcpu->kvm))
6429 		return -EIO;
6430 
6431 	if (is_guest_mode(vcpu)) {
6432 		/*
6433 		 * PML is never enabled when running L2, bail immediately if a
6434 		 * PML full exit occurs as something is horribly wrong.
6435 		 */
6436 		if (exit_reason.basic == EXIT_REASON_PML_FULL)
6437 			goto unexpected_vmexit;
6438 
6439 		/*
6440 		 * The host physical addresses of some pages of guest memory
6441 		 * are loaded into the vmcs02 (e.g. vmcs12's Virtual APIC
6442 		 * Page). The CPU may write to these pages via their host
6443 		 * physical address while L2 is running, bypassing any
6444 		 * address-translation-based dirty tracking (e.g. EPT write
6445 		 * protection).
6446 		 *
6447 		 * Mark them dirty on every exit from L2 to prevent them from
6448 		 * getting out of sync with dirty tracking.
6449 		 */
6450 		nested_mark_vmcs12_pages_dirty(vcpu);
6451 
6452 		/*
6453 		 * Synthesize a triple fault if L2 state is invalid.  In normal
6454 		 * operation, nested VM-Enter rejects any attempt to enter L2
6455 		 * with invalid state.  However, those checks are skipped if
6456 		 * state is being stuffed via RSM or KVM_SET_NESTED_STATE.  If
6457 		 * L2 state is invalid, it means either L1 modified SMRAM state
6458 		 * or userspace provided bad state.  Synthesize TRIPLE_FAULT as
6459 		 * doing so is architecturally allowed in the RSM case, and is
6460 		 * the least awful solution for the userspace case without
6461 		 * risking false positives.
6462 		 */
6463 		if (vmx->emulation_required) {
6464 			nested_vmx_vmexit(vcpu, EXIT_REASON_TRIPLE_FAULT, 0, 0);
6465 			return 1;
6466 		}
6467 
6468 		if (nested_vmx_reflect_vmexit(vcpu))
6469 			return 1;
6470 	}
6471 
6472 	/* If guest state is invalid, start emulating.  L2 is handled above. */
6473 	if (vmx->emulation_required)
6474 		return handle_invalid_guest_state(vcpu);
6475 
6476 	if (exit_reason.failed_vmentry) {
6477 		dump_vmcs(vcpu);
6478 		vcpu->run->exit_reason = KVM_EXIT_FAIL_ENTRY;
6479 		vcpu->run->fail_entry.hardware_entry_failure_reason
6480 			= exit_reason.full;
6481 		vcpu->run->fail_entry.cpu = vcpu->arch.last_vmentry_cpu;
6482 		return 0;
6483 	}
6484 
6485 	if (unlikely(vmx->fail)) {
6486 		dump_vmcs(vcpu);
6487 		vcpu->run->exit_reason = KVM_EXIT_FAIL_ENTRY;
6488 		vcpu->run->fail_entry.hardware_entry_failure_reason
6489 			= vmcs_read32(VM_INSTRUCTION_ERROR);
6490 		vcpu->run->fail_entry.cpu = vcpu->arch.last_vmentry_cpu;
6491 		return 0;
6492 	}
6493 
6494 	/*
6495 	 * Note:
6496 	 * Do not try to fix EXIT_REASON_EPT_MISCONFIG if it caused by
6497 	 * delivery event since it indicates guest is accessing MMIO.
6498 	 * The vm-exit can be triggered again after return to guest that
6499 	 * will cause infinite loop.
6500 	 */
6501 	if ((vectoring_info & VECTORING_INFO_VALID_MASK) &&
6502 	    (exit_reason.basic != EXIT_REASON_EXCEPTION_NMI &&
6503 	     exit_reason.basic != EXIT_REASON_EPT_VIOLATION &&
6504 	     exit_reason.basic != EXIT_REASON_PML_FULL &&
6505 	     exit_reason.basic != EXIT_REASON_APIC_ACCESS &&
6506 	     exit_reason.basic != EXIT_REASON_TASK_SWITCH &&
6507 	     exit_reason.basic != EXIT_REASON_NOTIFY)) {
6508 		int ndata = 3;
6509 
6510 		vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
6511 		vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_DELIVERY_EV;
6512 		vcpu->run->internal.data[0] = vectoring_info;
6513 		vcpu->run->internal.data[1] = exit_reason.full;
6514 		vcpu->run->internal.data[2] = vcpu->arch.exit_qualification;
6515 		if (exit_reason.basic == EXIT_REASON_EPT_MISCONFIG) {
6516 			vcpu->run->internal.data[ndata++] =
6517 				vmcs_read64(GUEST_PHYSICAL_ADDRESS);
6518 		}
6519 		vcpu->run->internal.data[ndata++] = vcpu->arch.last_vmentry_cpu;
6520 		vcpu->run->internal.ndata = ndata;
6521 		return 0;
6522 	}
6523 
6524 	if (unlikely(!enable_vnmi &&
6525 		     vmx->loaded_vmcs->soft_vnmi_blocked)) {
6526 		if (!vmx_interrupt_blocked(vcpu)) {
6527 			vmx->loaded_vmcs->soft_vnmi_blocked = 0;
6528 		} else if (vmx->loaded_vmcs->vnmi_blocked_time > 1000000000LL &&
6529 			   vcpu->arch.nmi_pending) {
6530 			/*
6531 			 * This CPU don't support us in finding the end of an
6532 			 * NMI-blocked window if the guest runs with IRQs
6533 			 * disabled. So we pull the trigger after 1 s of
6534 			 * futile waiting, but inform the user about this.
6535 			 */
6536 			printk(KERN_WARNING "%s: Breaking out of NMI-blocked "
6537 			       "state on VCPU %d after 1 s timeout\n",
6538 			       __func__, vcpu->vcpu_id);
6539 			vmx->loaded_vmcs->soft_vnmi_blocked = 0;
6540 		}
6541 	}
6542 
6543 	if (exit_fastpath != EXIT_FASTPATH_NONE)
6544 		return 1;
6545 
6546 	if (exit_reason.basic >= kvm_vmx_max_exit_handlers)
6547 		goto unexpected_vmexit;
6548 #ifdef CONFIG_RETPOLINE
6549 	if (exit_reason.basic == EXIT_REASON_MSR_WRITE)
6550 		return kvm_emulate_wrmsr(vcpu);
6551 	else if (exit_reason.basic == EXIT_REASON_PREEMPTION_TIMER)
6552 		return handle_preemption_timer(vcpu);
6553 	else if (exit_reason.basic == EXIT_REASON_INTERRUPT_WINDOW)
6554 		return handle_interrupt_window(vcpu);
6555 	else if (exit_reason.basic == EXIT_REASON_EXTERNAL_INTERRUPT)
6556 		return handle_external_interrupt(vcpu);
6557 	else if (exit_reason.basic == EXIT_REASON_HLT)
6558 		return kvm_emulate_halt(vcpu);
6559 	else if (exit_reason.basic == EXIT_REASON_EPT_MISCONFIG)
6560 		return handle_ept_misconfig(vcpu);
6561 #endif
6562 
6563 	exit_handler_index = array_index_nospec((u16)exit_reason.basic,
6564 						kvm_vmx_max_exit_handlers);
6565 	if (!kvm_vmx_exit_handlers[exit_handler_index])
6566 		goto unexpected_vmexit;
6567 
6568 	return kvm_vmx_exit_handlers[exit_handler_index](vcpu);
6569 
6570 unexpected_vmexit:
6571 	vcpu_unimpl(vcpu, "vmx: unexpected exit reason 0x%x\n",
6572 		    exit_reason.full);
6573 	dump_vmcs(vcpu);
6574 	vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
6575 	vcpu->run->internal.suberror =
6576 			KVM_INTERNAL_ERROR_UNEXPECTED_EXIT_REASON;
6577 	vcpu->run->internal.ndata = 2;
6578 	vcpu->run->internal.data[0] = exit_reason.full;
6579 	vcpu->run->internal.data[1] = vcpu->arch.last_vmentry_cpu;
6580 	return 0;
6581 }
6582 
6583 static int vmx_handle_exit(struct kvm_vcpu *vcpu, fastpath_t exit_fastpath)
6584 {
6585 	int ret = __vmx_handle_exit(vcpu, exit_fastpath);
6586 
6587 	/*
6588 	 * Exit to user space when bus lock detected to inform that there is
6589 	 * a bus lock in guest.
6590 	 */
6591 	if (to_vmx(vcpu)->exit_reason.bus_lock_detected) {
6592 		if (ret > 0)
6593 			vcpu->run->exit_reason = KVM_EXIT_X86_BUS_LOCK;
6594 
6595 		vcpu->run->flags |= KVM_RUN_X86_BUS_LOCK;
6596 		return 0;
6597 	}
6598 	return ret;
6599 }
6600 
6601 /*
6602  * Software based L1D cache flush which is used when microcode providing
6603  * the cache control MSR is not loaded.
6604  *
6605  * The L1D cache is 32 KiB on Nehalem and later microarchitectures, but to
6606  * flush it is required to read in 64 KiB because the replacement algorithm
6607  * is not exactly LRU. This could be sized at runtime via topology
6608  * information but as all relevant affected CPUs have 32KiB L1D cache size
6609  * there is no point in doing so.
6610  */
6611 static noinstr void vmx_l1d_flush(struct kvm_vcpu *vcpu)
6612 {
6613 	int size = PAGE_SIZE << L1D_CACHE_ORDER;
6614 
6615 	/*
6616 	 * This code is only executed when the flush mode is 'cond' or
6617 	 * 'always'
6618 	 */
6619 	if (static_branch_likely(&vmx_l1d_flush_cond)) {
6620 		bool flush_l1d;
6621 
6622 		/*
6623 		 * Clear the per-vcpu flush bit, it gets set again
6624 		 * either from vcpu_run() or from one of the unsafe
6625 		 * VMEXIT handlers.
6626 		 */
6627 		flush_l1d = vcpu->arch.l1tf_flush_l1d;
6628 		vcpu->arch.l1tf_flush_l1d = false;
6629 
6630 		/*
6631 		 * Clear the per-cpu flush bit, it gets set again from
6632 		 * the interrupt handlers.
6633 		 */
6634 		flush_l1d |= kvm_get_cpu_l1tf_flush_l1d();
6635 		kvm_clear_cpu_l1tf_flush_l1d();
6636 
6637 		if (!flush_l1d)
6638 			return;
6639 	}
6640 
6641 	vcpu->stat.l1d_flush++;
6642 
6643 	if (static_cpu_has(X86_FEATURE_FLUSH_L1D)) {
6644 		native_wrmsrl(MSR_IA32_FLUSH_CMD, L1D_FLUSH);
6645 		return;
6646 	}
6647 
6648 	asm volatile(
6649 		/* First ensure the pages are in the TLB */
6650 		"xorl	%%eax, %%eax\n"
6651 		".Lpopulate_tlb:\n\t"
6652 		"movzbl	(%[flush_pages], %%" _ASM_AX "), %%ecx\n\t"
6653 		"addl	$4096, %%eax\n\t"
6654 		"cmpl	%%eax, %[size]\n\t"
6655 		"jne	.Lpopulate_tlb\n\t"
6656 		"xorl	%%eax, %%eax\n\t"
6657 		"cpuid\n\t"
6658 		/* Now fill the cache */
6659 		"xorl	%%eax, %%eax\n"
6660 		".Lfill_cache:\n"
6661 		"movzbl	(%[flush_pages], %%" _ASM_AX "), %%ecx\n\t"
6662 		"addl	$64, %%eax\n\t"
6663 		"cmpl	%%eax, %[size]\n\t"
6664 		"jne	.Lfill_cache\n\t"
6665 		"lfence\n"
6666 		:: [flush_pages] "r" (vmx_l1d_flush_pages),
6667 		    [size] "r" (size)
6668 		: "eax", "ebx", "ecx", "edx");
6669 }
6670 
6671 static void vmx_update_cr8_intercept(struct kvm_vcpu *vcpu, int tpr, int irr)
6672 {
6673 	struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
6674 	int tpr_threshold;
6675 
6676 	if (is_guest_mode(vcpu) &&
6677 		nested_cpu_has(vmcs12, CPU_BASED_TPR_SHADOW))
6678 		return;
6679 
6680 	tpr_threshold = (irr == -1 || tpr < irr) ? 0 : irr;
6681 	if (is_guest_mode(vcpu))
6682 		to_vmx(vcpu)->nested.l1_tpr_threshold = tpr_threshold;
6683 	else
6684 		vmcs_write32(TPR_THRESHOLD, tpr_threshold);
6685 }
6686 
6687 void vmx_set_virtual_apic_mode(struct kvm_vcpu *vcpu)
6688 {
6689 	struct vcpu_vmx *vmx = to_vmx(vcpu);
6690 	u32 sec_exec_control;
6691 
6692 	if (!lapic_in_kernel(vcpu))
6693 		return;
6694 
6695 	if (!flexpriority_enabled &&
6696 	    !cpu_has_vmx_virtualize_x2apic_mode())
6697 		return;
6698 
6699 	/* Postpone execution until vmcs01 is the current VMCS. */
6700 	if (is_guest_mode(vcpu)) {
6701 		vmx->nested.change_vmcs01_virtual_apic_mode = true;
6702 		return;
6703 	}
6704 
6705 	sec_exec_control = secondary_exec_controls_get(vmx);
6706 	sec_exec_control &= ~(SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES |
6707 			      SECONDARY_EXEC_VIRTUALIZE_X2APIC_MODE);
6708 
6709 	switch (kvm_get_apic_mode(vcpu)) {
6710 	case LAPIC_MODE_INVALID:
6711 		WARN_ONCE(true, "Invalid local APIC state");
6712 		break;
6713 	case LAPIC_MODE_DISABLED:
6714 		break;
6715 	case LAPIC_MODE_XAPIC:
6716 		if (flexpriority_enabled) {
6717 			sec_exec_control |=
6718 				SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES;
6719 			kvm_make_request(KVM_REQ_APIC_PAGE_RELOAD, vcpu);
6720 
6721 			/*
6722 			 * Flush the TLB, reloading the APIC access page will
6723 			 * only do so if its physical address has changed, but
6724 			 * the guest may have inserted a non-APIC mapping into
6725 			 * the TLB while the APIC access page was disabled.
6726 			 */
6727 			kvm_make_request(KVM_REQ_TLB_FLUSH_CURRENT, vcpu);
6728 		}
6729 		break;
6730 	case LAPIC_MODE_X2APIC:
6731 		if (cpu_has_vmx_virtualize_x2apic_mode())
6732 			sec_exec_control |=
6733 				SECONDARY_EXEC_VIRTUALIZE_X2APIC_MODE;
6734 		break;
6735 	}
6736 	secondary_exec_controls_set(vmx, sec_exec_control);
6737 
6738 	vmx_update_msr_bitmap_x2apic(vcpu);
6739 }
6740 
6741 static void vmx_set_apic_access_page_addr(struct kvm_vcpu *vcpu)
6742 {
6743 	const gfn_t gfn = APIC_DEFAULT_PHYS_BASE >> PAGE_SHIFT;
6744 	struct kvm *kvm = vcpu->kvm;
6745 	struct kvm_memslots *slots = kvm_memslots(kvm);
6746 	struct kvm_memory_slot *slot;
6747 	unsigned long mmu_seq;
6748 	kvm_pfn_t pfn;
6749 
6750 	/* Defer reload until vmcs01 is the current VMCS. */
6751 	if (is_guest_mode(vcpu)) {
6752 		to_vmx(vcpu)->nested.reload_vmcs01_apic_access_page = true;
6753 		return;
6754 	}
6755 
6756 	if (!(secondary_exec_controls_get(to_vmx(vcpu)) &
6757 	    SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES))
6758 		return;
6759 
6760 	/*
6761 	 * Grab the memslot so that the hva lookup for the mmu_notifier retry
6762 	 * is guaranteed to use the same memslot as the pfn lookup, i.e. rely
6763 	 * on the pfn lookup's validation of the memslot to ensure a valid hva
6764 	 * is used for the retry check.
6765 	 */
6766 	slot = id_to_memslot(slots, APIC_ACCESS_PAGE_PRIVATE_MEMSLOT);
6767 	if (!slot || slot->flags & KVM_MEMSLOT_INVALID)
6768 		return;
6769 
6770 	/*
6771 	 * Ensure that the mmu_notifier sequence count is read before KVM
6772 	 * retrieves the pfn from the primary MMU.  Note, the memslot is
6773 	 * protected by SRCU, not the mmu_notifier.  Pairs with the smp_wmb()
6774 	 * in kvm_mmu_invalidate_end().
6775 	 */
6776 	mmu_seq = kvm->mmu_invalidate_seq;
6777 	smp_rmb();
6778 
6779 	/*
6780 	 * No need to retry if the memslot does not exist or is invalid.  KVM
6781 	 * controls the APIC-access page memslot, and only deletes the memslot
6782 	 * if APICv is permanently inhibited, i.e. the memslot won't reappear.
6783 	 */
6784 	pfn = gfn_to_pfn_memslot(slot, gfn);
6785 	if (is_error_noslot_pfn(pfn))
6786 		return;
6787 
6788 	read_lock(&vcpu->kvm->mmu_lock);
6789 	if (mmu_invalidate_retry_hva(kvm, mmu_seq,
6790 				     gfn_to_hva_memslot(slot, gfn))) {
6791 		kvm_make_request(KVM_REQ_APIC_PAGE_RELOAD, vcpu);
6792 		read_unlock(&vcpu->kvm->mmu_lock);
6793 		goto out;
6794 	}
6795 
6796 	vmcs_write64(APIC_ACCESS_ADDR, pfn_to_hpa(pfn));
6797 	read_unlock(&vcpu->kvm->mmu_lock);
6798 
6799 	vmx_flush_tlb_current(vcpu);
6800 
6801 out:
6802 	/*
6803 	 * Do not pin apic access page in memory, the MMU notifier
6804 	 * will call us again if it is migrated or swapped out.
6805 	 */
6806 	kvm_release_pfn_clean(pfn);
6807 }
6808 
6809 static void vmx_hwapic_isr_update(int max_isr)
6810 {
6811 	u16 status;
6812 	u8 old;
6813 
6814 	if (max_isr == -1)
6815 		max_isr = 0;
6816 
6817 	status = vmcs_read16(GUEST_INTR_STATUS);
6818 	old = status >> 8;
6819 	if (max_isr != old) {
6820 		status &= 0xff;
6821 		status |= max_isr << 8;
6822 		vmcs_write16(GUEST_INTR_STATUS, status);
6823 	}
6824 }
6825 
6826 static void vmx_set_rvi(int vector)
6827 {
6828 	u16 status;
6829 	u8 old;
6830 
6831 	if (vector == -1)
6832 		vector = 0;
6833 
6834 	status = vmcs_read16(GUEST_INTR_STATUS);
6835 	old = (u8)status & 0xff;
6836 	if ((u8)vector != old) {
6837 		status &= ~0xff;
6838 		status |= (u8)vector;
6839 		vmcs_write16(GUEST_INTR_STATUS, status);
6840 	}
6841 }
6842 
6843 static void vmx_hwapic_irr_update(struct kvm_vcpu *vcpu, int max_irr)
6844 {
6845 	/*
6846 	 * When running L2, updating RVI is only relevant when
6847 	 * vmcs12 virtual-interrupt-delivery enabled.
6848 	 * However, it can be enabled only when L1 also
6849 	 * intercepts external-interrupts and in that case
6850 	 * we should not update vmcs02 RVI but instead intercept
6851 	 * interrupt. Therefore, do nothing when running L2.
6852 	 */
6853 	if (!is_guest_mode(vcpu))
6854 		vmx_set_rvi(max_irr);
6855 }
6856 
6857 static int vmx_sync_pir_to_irr(struct kvm_vcpu *vcpu)
6858 {
6859 	struct vcpu_vmx *vmx = to_vmx(vcpu);
6860 	int max_irr;
6861 	bool got_posted_interrupt;
6862 
6863 	if (KVM_BUG_ON(!enable_apicv, vcpu->kvm))
6864 		return -EIO;
6865 
6866 	if (pi_test_on(&vmx->pi_desc)) {
6867 		pi_clear_on(&vmx->pi_desc);
6868 		/*
6869 		 * IOMMU can write to PID.ON, so the barrier matters even on UP.
6870 		 * But on x86 this is just a compiler barrier anyway.
6871 		 */
6872 		smp_mb__after_atomic();
6873 		got_posted_interrupt =
6874 			kvm_apic_update_irr(vcpu, vmx->pi_desc.pir, &max_irr);
6875 	} else {
6876 		max_irr = kvm_lapic_find_highest_irr(vcpu);
6877 		got_posted_interrupt = false;
6878 	}
6879 
6880 	/*
6881 	 * Newly recognized interrupts are injected via either virtual interrupt
6882 	 * delivery (RVI) or KVM_REQ_EVENT.  Virtual interrupt delivery is
6883 	 * disabled in two cases:
6884 	 *
6885 	 * 1) If L2 is running and the vCPU has a new pending interrupt.  If L1
6886 	 * wants to exit on interrupts, KVM_REQ_EVENT is needed to synthesize a
6887 	 * VM-Exit to L1.  If L1 doesn't want to exit, the interrupt is injected
6888 	 * into L2, but KVM doesn't use virtual interrupt delivery to inject
6889 	 * interrupts into L2, and so KVM_REQ_EVENT is again needed.
6890 	 *
6891 	 * 2) If APICv is disabled for this vCPU, assigned devices may still
6892 	 * attempt to post interrupts.  The posted interrupt vector will cause
6893 	 * a VM-Exit and the subsequent entry will call sync_pir_to_irr.
6894 	 */
6895 	if (!is_guest_mode(vcpu) && kvm_vcpu_apicv_active(vcpu))
6896 		vmx_set_rvi(max_irr);
6897 	else if (got_posted_interrupt)
6898 		kvm_make_request(KVM_REQ_EVENT, vcpu);
6899 
6900 	return max_irr;
6901 }
6902 
6903 static void vmx_load_eoi_exitmap(struct kvm_vcpu *vcpu, u64 *eoi_exit_bitmap)
6904 {
6905 	if (!kvm_vcpu_apicv_active(vcpu))
6906 		return;
6907 
6908 	vmcs_write64(EOI_EXIT_BITMAP0, eoi_exit_bitmap[0]);
6909 	vmcs_write64(EOI_EXIT_BITMAP1, eoi_exit_bitmap[1]);
6910 	vmcs_write64(EOI_EXIT_BITMAP2, eoi_exit_bitmap[2]);
6911 	vmcs_write64(EOI_EXIT_BITMAP3, eoi_exit_bitmap[3]);
6912 }
6913 
6914 static void vmx_apicv_post_state_restore(struct kvm_vcpu *vcpu)
6915 {
6916 	struct vcpu_vmx *vmx = to_vmx(vcpu);
6917 
6918 	pi_clear_on(&vmx->pi_desc);
6919 	memset(vmx->pi_desc.pir, 0, sizeof(vmx->pi_desc.pir));
6920 }
6921 
6922 void vmx_do_interrupt_irqoff(unsigned long entry);
6923 void vmx_do_nmi_irqoff(void);
6924 
6925 static void handle_nm_fault_irqoff(struct kvm_vcpu *vcpu)
6926 {
6927 	/*
6928 	 * Save xfd_err to guest_fpu before interrupt is enabled, so the
6929 	 * MSR value is not clobbered by the host activity before the guest
6930 	 * has chance to consume it.
6931 	 *
6932 	 * Do not blindly read xfd_err here, since this exception might
6933 	 * be caused by L1 interception on a platform which doesn't
6934 	 * support xfd at all.
6935 	 *
6936 	 * Do it conditionally upon guest_fpu::xfd. xfd_err matters
6937 	 * only when xfd contains a non-zero value.
6938 	 *
6939 	 * Queuing exception is done in vmx_handle_exit. See comment there.
6940 	 */
6941 	if (vcpu->arch.guest_fpu.fpstate->xfd)
6942 		rdmsrl(MSR_IA32_XFD_ERR, vcpu->arch.guest_fpu.xfd_err);
6943 }
6944 
6945 static void handle_exception_irqoff(struct vcpu_vmx *vmx)
6946 {
6947 	u32 intr_info = vmx_get_intr_info(&vmx->vcpu);
6948 
6949 	/* if exit due to PF check for async PF */
6950 	if (is_page_fault(intr_info))
6951 		vmx->vcpu.arch.apf.host_apf_flags = kvm_read_and_reset_apf_flags();
6952 	/* if exit due to NM, handle before interrupts are enabled */
6953 	else if (is_nm_fault(intr_info))
6954 		handle_nm_fault_irqoff(&vmx->vcpu);
6955 	/* Handle machine checks before interrupts are enabled */
6956 	else if (is_machine_check(intr_info))
6957 		kvm_machine_check();
6958 }
6959 
6960 static void handle_external_interrupt_irqoff(struct kvm_vcpu *vcpu)
6961 {
6962 	u32 intr_info = vmx_get_intr_info(vcpu);
6963 	unsigned int vector = intr_info & INTR_INFO_VECTOR_MASK;
6964 	gate_desc *desc = (gate_desc *)host_idt_base + vector;
6965 
6966 	if (KVM_BUG(!is_external_intr(intr_info), vcpu->kvm,
6967 	    "unexpected VM-Exit interrupt info: 0x%x", intr_info))
6968 		return;
6969 
6970 	kvm_before_interrupt(vcpu, KVM_HANDLING_IRQ);
6971 	vmx_do_interrupt_irqoff(gate_offset(desc));
6972 	kvm_after_interrupt(vcpu);
6973 
6974 	vcpu->arch.at_instruction_boundary = true;
6975 }
6976 
6977 static void vmx_handle_exit_irqoff(struct kvm_vcpu *vcpu)
6978 {
6979 	struct vcpu_vmx *vmx = to_vmx(vcpu);
6980 
6981 	if (vmx->emulation_required)
6982 		return;
6983 
6984 	if (vmx->exit_reason.basic == EXIT_REASON_EXTERNAL_INTERRUPT)
6985 		handle_external_interrupt_irqoff(vcpu);
6986 	else if (vmx->exit_reason.basic == EXIT_REASON_EXCEPTION_NMI)
6987 		handle_exception_irqoff(vmx);
6988 }
6989 
6990 /*
6991  * The kvm parameter can be NULL (module initialization, or invocation before
6992  * VM creation). Be sure to check the kvm parameter before using it.
6993  */
6994 static bool vmx_has_emulated_msr(struct kvm *kvm, u32 index)
6995 {
6996 	switch (index) {
6997 	case MSR_IA32_SMBASE:
6998 		if (!IS_ENABLED(CONFIG_KVM_SMM))
6999 			return false;
7000 		/*
7001 		 * We cannot do SMM unless we can run the guest in big
7002 		 * real mode.
7003 		 */
7004 		return enable_unrestricted_guest || emulate_invalid_guest_state;
7005 	case KVM_FIRST_EMULATED_VMX_MSR ... KVM_LAST_EMULATED_VMX_MSR:
7006 		return nested;
7007 	case MSR_AMD64_VIRT_SPEC_CTRL:
7008 	case MSR_AMD64_TSC_RATIO:
7009 		/* This is AMD only.  */
7010 		return false;
7011 	default:
7012 		return true;
7013 	}
7014 }
7015 
7016 static void vmx_recover_nmi_blocking(struct vcpu_vmx *vmx)
7017 {
7018 	u32 exit_intr_info;
7019 	bool unblock_nmi;
7020 	u8 vector;
7021 	bool idtv_info_valid;
7022 
7023 	idtv_info_valid = vmx->idt_vectoring_info & VECTORING_INFO_VALID_MASK;
7024 
7025 	if (enable_vnmi) {
7026 		if (vmx->loaded_vmcs->nmi_known_unmasked)
7027 			return;
7028 
7029 		exit_intr_info = vmx_get_intr_info(&vmx->vcpu);
7030 		unblock_nmi = (exit_intr_info & INTR_INFO_UNBLOCK_NMI) != 0;
7031 		vector = exit_intr_info & INTR_INFO_VECTOR_MASK;
7032 		/*
7033 		 * SDM 3: 27.7.1.2 (September 2008)
7034 		 * Re-set bit "block by NMI" before VM entry if vmexit caused by
7035 		 * a guest IRET fault.
7036 		 * SDM 3: 23.2.2 (September 2008)
7037 		 * Bit 12 is undefined in any of the following cases:
7038 		 *  If the VM exit sets the valid bit in the IDT-vectoring
7039 		 *   information field.
7040 		 *  If the VM exit is due to a double fault.
7041 		 */
7042 		if ((exit_intr_info & INTR_INFO_VALID_MASK) && unblock_nmi &&
7043 		    vector != DF_VECTOR && !idtv_info_valid)
7044 			vmcs_set_bits(GUEST_INTERRUPTIBILITY_INFO,
7045 				      GUEST_INTR_STATE_NMI);
7046 		else
7047 			vmx->loaded_vmcs->nmi_known_unmasked =
7048 				!(vmcs_read32(GUEST_INTERRUPTIBILITY_INFO)
7049 				  & GUEST_INTR_STATE_NMI);
7050 	} else if (unlikely(vmx->loaded_vmcs->soft_vnmi_blocked))
7051 		vmx->loaded_vmcs->vnmi_blocked_time +=
7052 			ktime_to_ns(ktime_sub(ktime_get(),
7053 					      vmx->loaded_vmcs->entry_time));
7054 }
7055 
7056 static void __vmx_complete_interrupts(struct kvm_vcpu *vcpu,
7057 				      u32 idt_vectoring_info,
7058 				      int instr_len_field,
7059 				      int error_code_field)
7060 {
7061 	u8 vector;
7062 	int type;
7063 	bool idtv_info_valid;
7064 
7065 	idtv_info_valid = idt_vectoring_info & VECTORING_INFO_VALID_MASK;
7066 
7067 	vcpu->arch.nmi_injected = false;
7068 	kvm_clear_exception_queue(vcpu);
7069 	kvm_clear_interrupt_queue(vcpu);
7070 
7071 	if (!idtv_info_valid)
7072 		return;
7073 
7074 	kvm_make_request(KVM_REQ_EVENT, vcpu);
7075 
7076 	vector = idt_vectoring_info & VECTORING_INFO_VECTOR_MASK;
7077 	type = idt_vectoring_info & VECTORING_INFO_TYPE_MASK;
7078 
7079 	switch (type) {
7080 	case INTR_TYPE_NMI_INTR:
7081 		vcpu->arch.nmi_injected = true;
7082 		/*
7083 		 * SDM 3: 27.7.1.2 (September 2008)
7084 		 * Clear bit "block by NMI" before VM entry if a NMI
7085 		 * delivery faulted.
7086 		 */
7087 		vmx_set_nmi_mask(vcpu, false);
7088 		break;
7089 	case INTR_TYPE_SOFT_EXCEPTION:
7090 		vcpu->arch.event_exit_inst_len = vmcs_read32(instr_len_field);
7091 		fallthrough;
7092 	case INTR_TYPE_HARD_EXCEPTION:
7093 		if (idt_vectoring_info & VECTORING_INFO_DELIVER_CODE_MASK) {
7094 			u32 err = vmcs_read32(error_code_field);
7095 			kvm_requeue_exception_e(vcpu, vector, err);
7096 		} else
7097 			kvm_requeue_exception(vcpu, vector);
7098 		break;
7099 	case INTR_TYPE_SOFT_INTR:
7100 		vcpu->arch.event_exit_inst_len = vmcs_read32(instr_len_field);
7101 		fallthrough;
7102 	case INTR_TYPE_EXT_INTR:
7103 		kvm_queue_interrupt(vcpu, vector, type == INTR_TYPE_SOFT_INTR);
7104 		break;
7105 	default:
7106 		break;
7107 	}
7108 }
7109 
7110 static void vmx_complete_interrupts(struct vcpu_vmx *vmx)
7111 {
7112 	__vmx_complete_interrupts(&vmx->vcpu, vmx->idt_vectoring_info,
7113 				  VM_EXIT_INSTRUCTION_LEN,
7114 				  IDT_VECTORING_ERROR_CODE);
7115 }
7116 
7117 static void vmx_cancel_injection(struct kvm_vcpu *vcpu)
7118 {
7119 	__vmx_complete_interrupts(vcpu,
7120 				  vmcs_read32(VM_ENTRY_INTR_INFO_FIELD),
7121 				  VM_ENTRY_INSTRUCTION_LEN,
7122 				  VM_ENTRY_EXCEPTION_ERROR_CODE);
7123 
7124 	vmcs_write32(VM_ENTRY_INTR_INFO_FIELD, 0);
7125 }
7126 
7127 static void atomic_switch_perf_msrs(struct vcpu_vmx *vmx)
7128 {
7129 	int i, nr_msrs;
7130 	struct perf_guest_switch_msr *msrs;
7131 	struct kvm_pmu *pmu = vcpu_to_pmu(&vmx->vcpu);
7132 
7133 	pmu->host_cross_mapped_mask = 0;
7134 	if (pmu->pebs_enable & pmu->global_ctrl)
7135 		intel_pmu_cross_mapped_check(pmu);
7136 
7137 	/* Note, nr_msrs may be garbage if perf_guest_get_msrs() returns NULL. */
7138 	msrs = perf_guest_get_msrs(&nr_msrs, (void *)pmu);
7139 	if (!msrs)
7140 		return;
7141 
7142 	for (i = 0; i < nr_msrs; i++)
7143 		if (msrs[i].host == msrs[i].guest)
7144 			clear_atomic_switch_msr(vmx, msrs[i].msr);
7145 		else
7146 			add_atomic_switch_msr(vmx, msrs[i].msr, msrs[i].guest,
7147 					msrs[i].host, false);
7148 }
7149 
7150 static void vmx_update_hv_timer(struct kvm_vcpu *vcpu)
7151 {
7152 	struct vcpu_vmx *vmx = to_vmx(vcpu);
7153 	u64 tscl;
7154 	u32 delta_tsc;
7155 
7156 	if (vmx->req_immediate_exit) {
7157 		vmcs_write32(VMX_PREEMPTION_TIMER_VALUE, 0);
7158 		vmx->loaded_vmcs->hv_timer_soft_disabled = false;
7159 	} else if (vmx->hv_deadline_tsc != -1) {
7160 		tscl = rdtsc();
7161 		if (vmx->hv_deadline_tsc > tscl)
7162 			/* set_hv_timer ensures the delta fits in 32-bits */
7163 			delta_tsc = (u32)((vmx->hv_deadline_tsc - tscl) >>
7164 				cpu_preemption_timer_multi);
7165 		else
7166 			delta_tsc = 0;
7167 
7168 		vmcs_write32(VMX_PREEMPTION_TIMER_VALUE, delta_tsc);
7169 		vmx->loaded_vmcs->hv_timer_soft_disabled = false;
7170 	} else if (!vmx->loaded_vmcs->hv_timer_soft_disabled) {
7171 		vmcs_write32(VMX_PREEMPTION_TIMER_VALUE, -1);
7172 		vmx->loaded_vmcs->hv_timer_soft_disabled = true;
7173 	}
7174 }
7175 
7176 void noinstr vmx_update_host_rsp(struct vcpu_vmx *vmx, unsigned long host_rsp)
7177 {
7178 	if (unlikely(host_rsp != vmx->loaded_vmcs->host_state.rsp)) {
7179 		vmx->loaded_vmcs->host_state.rsp = host_rsp;
7180 		vmcs_writel(HOST_RSP, host_rsp);
7181 	}
7182 }
7183 
7184 void noinstr vmx_spec_ctrl_restore_host(struct vcpu_vmx *vmx,
7185 					unsigned int flags)
7186 {
7187 	u64 hostval = this_cpu_read(x86_spec_ctrl_current);
7188 
7189 	if (!cpu_feature_enabled(X86_FEATURE_MSR_SPEC_CTRL))
7190 		return;
7191 
7192 	if (flags & VMX_RUN_SAVE_SPEC_CTRL)
7193 		vmx->spec_ctrl = __rdmsr(MSR_IA32_SPEC_CTRL);
7194 
7195 	/*
7196 	 * If the guest/host SPEC_CTRL values differ, restore the host value.
7197 	 *
7198 	 * For legacy IBRS, the IBRS bit always needs to be written after
7199 	 * transitioning from a less privileged predictor mode, regardless of
7200 	 * whether the guest/host values differ.
7201 	 */
7202 	if (cpu_feature_enabled(X86_FEATURE_KERNEL_IBRS) ||
7203 	    vmx->spec_ctrl != hostval)
7204 		native_wrmsrl(MSR_IA32_SPEC_CTRL, hostval);
7205 
7206 	barrier_nospec();
7207 }
7208 
7209 static fastpath_t vmx_exit_handlers_fastpath(struct kvm_vcpu *vcpu)
7210 {
7211 	switch (to_vmx(vcpu)->exit_reason.basic) {
7212 	case EXIT_REASON_MSR_WRITE:
7213 		return handle_fastpath_set_msr_irqoff(vcpu);
7214 	case EXIT_REASON_PREEMPTION_TIMER:
7215 		return handle_fastpath_preemption_timer(vcpu);
7216 	default:
7217 		return EXIT_FASTPATH_NONE;
7218 	}
7219 }
7220 
7221 static noinstr void vmx_vcpu_enter_exit(struct kvm_vcpu *vcpu,
7222 					unsigned int flags)
7223 {
7224 	struct vcpu_vmx *vmx = to_vmx(vcpu);
7225 
7226 	guest_state_enter_irqoff();
7227 
7228 	/* L1D Flush includes CPU buffer clear to mitigate MDS */
7229 	if (static_branch_unlikely(&vmx_l1d_should_flush))
7230 		vmx_l1d_flush(vcpu);
7231 	else if (static_branch_unlikely(&mds_user_clear))
7232 		mds_clear_cpu_buffers();
7233 	else if (static_branch_unlikely(&mmio_stale_data_clear) &&
7234 		 kvm_arch_has_assigned_device(vcpu->kvm))
7235 		mds_clear_cpu_buffers();
7236 
7237 	vmx_disable_fb_clear(vmx);
7238 
7239 	if (vcpu->arch.cr2 != native_read_cr2())
7240 		native_write_cr2(vcpu->arch.cr2);
7241 
7242 	vmx->fail = __vmx_vcpu_run(vmx, (unsigned long *)&vcpu->arch.regs,
7243 				   flags);
7244 
7245 	vcpu->arch.cr2 = native_read_cr2();
7246 
7247 	vmx_enable_fb_clear(vmx);
7248 
7249 	if (unlikely(vmx->fail))
7250 		vmx->exit_reason.full = 0xdead;
7251 	else
7252 		vmx->exit_reason.full = vmcs_read32(VM_EXIT_REASON);
7253 
7254 	if ((u16)vmx->exit_reason.basic == EXIT_REASON_EXCEPTION_NMI &&
7255 	    is_nmi(vmx_get_intr_info(vcpu))) {
7256 		kvm_before_interrupt(vcpu, KVM_HANDLING_NMI);
7257 		vmx_do_nmi_irqoff();
7258 		kvm_after_interrupt(vcpu);
7259 	}
7260 
7261 	guest_state_exit_irqoff();
7262 }
7263 
7264 static fastpath_t vmx_vcpu_run(struct kvm_vcpu *vcpu)
7265 {
7266 	struct vcpu_vmx *vmx = to_vmx(vcpu);
7267 	unsigned long cr3, cr4;
7268 
7269 	/* Record the guest's net vcpu time for enforced NMI injections. */
7270 	if (unlikely(!enable_vnmi &&
7271 		     vmx->loaded_vmcs->soft_vnmi_blocked))
7272 		vmx->loaded_vmcs->entry_time = ktime_get();
7273 
7274 	/*
7275 	 * Don't enter VMX if guest state is invalid, let the exit handler
7276 	 * start emulation until we arrive back to a valid state.  Synthesize a
7277 	 * consistency check VM-Exit due to invalid guest state and bail.
7278 	 */
7279 	if (unlikely(vmx->emulation_required)) {
7280 		vmx->fail = 0;
7281 
7282 		vmx->exit_reason.full = EXIT_REASON_INVALID_STATE;
7283 		vmx->exit_reason.failed_vmentry = 1;
7284 		kvm_register_mark_available(vcpu, VCPU_EXREG_EXIT_INFO_1);
7285 		vmx->exit_qualification = ENTRY_FAIL_DEFAULT;
7286 		kvm_register_mark_available(vcpu, VCPU_EXREG_EXIT_INFO_2);
7287 		vmx->exit_intr_info = 0;
7288 		return EXIT_FASTPATH_NONE;
7289 	}
7290 
7291 	trace_kvm_entry(vcpu);
7292 
7293 	if (vmx->ple_window_dirty) {
7294 		vmx->ple_window_dirty = false;
7295 		vmcs_write32(PLE_WINDOW, vmx->ple_window);
7296 	}
7297 
7298 	/*
7299 	 * We did this in prepare_switch_to_guest, because it needs to
7300 	 * be within srcu_read_lock.
7301 	 */
7302 	WARN_ON_ONCE(vmx->nested.need_vmcs12_to_shadow_sync);
7303 
7304 	if (kvm_register_is_dirty(vcpu, VCPU_REGS_RSP))
7305 		vmcs_writel(GUEST_RSP, vcpu->arch.regs[VCPU_REGS_RSP]);
7306 	if (kvm_register_is_dirty(vcpu, VCPU_REGS_RIP))
7307 		vmcs_writel(GUEST_RIP, vcpu->arch.regs[VCPU_REGS_RIP]);
7308 	vcpu->arch.regs_dirty = 0;
7309 
7310 	/*
7311 	 * Refresh vmcs.HOST_CR3 if necessary.  This must be done immediately
7312 	 * prior to VM-Enter, as the kernel may load a new ASID (PCID) any time
7313 	 * it switches back to the current->mm, which can occur in KVM context
7314 	 * when switching to a temporary mm to patch kernel code, e.g. if KVM
7315 	 * toggles a static key while handling a VM-Exit.
7316 	 */
7317 	cr3 = __get_current_cr3_fast();
7318 	if (unlikely(cr3 != vmx->loaded_vmcs->host_state.cr3)) {
7319 		vmcs_writel(HOST_CR3, cr3);
7320 		vmx->loaded_vmcs->host_state.cr3 = cr3;
7321 	}
7322 
7323 	cr4 = cr4_read_shadow();
7324 	if (unlikely(cr4 != vmx->loaded_vmcs->host_state.cr4)) {
7325 		vmcs_writel(HOST_CR4, cr4);
7326 		vmx->loaded_vmcs->host_state.cr4 = cr4;
7327 	}
7328 
7329 	/* When KVM_DEBUGREG_WONT_EXIT, dr6 is accessible in guest. */
7330 	if (unlikely(vcpu->arch.switch_db_regs & KVM_DEBUGREG_WONT_EXIT))
7331 		set_debugreg(vcpu->arch.dr6, 6);
7332 
7333 	/* When single-stepping over STI and MOV SS, we must clear the
7334 	 * corresponding interruptibility bits in the guest state. Otherwise
7335 	 * vmentry fails as it then expects bit 14 (BS) in pending debug
7336 	 * exceptions being set, but that's not correct for the guest debugging
7337 	 * case. */
7338 	if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)
7339 		vmx_set_interrupt_shadow(vcpu, 0);
7340 
7341 	kvm_load_guest_xsave_state(vcpu);
7342 
7343 	pt_guest_enter(vmx);
7344 
7345 	atomic_switch_perf_msrs(vmx);
7346 	if (intel_pmu_lbr_is_enabled(vcpu))
7347 		vmx_passthrough_lbr_msrs(vcpu);
7348 
7349 	if (enable_preemption_timer)
7350 		vmx_update_hv_timer(vcpu);
7351 
7352 	kvm_wait_lapic_expire(vcpu);
7353 
7354 	/* The actual VMENTER/EXIT is in the .noinstr.text section. */
7355 	vmx_vcpu_enter_exit(vcpu, __vmx_vcpu_run_flags(vmx));
7356 
7357 	/* All fields are clean at this point */
7358 	if (kvm_is_using_evmcs()) {
7359 		current_evmcs->hv_clean_fields |=
7360 			HV_VMX_ENLIGHTENED_CLEAN_FIELD_ALL;
7361 
7362 		current_evmcs->hv_vp_id = kvm_hv_get_vpindex(vcpu);
7363 	}
7364 
7365 	/* MSR_IA32_DEBUGCTLMSR is zeroed on vmexit. Restore it if needed */
7366 	if (vmx->host_debugctlmsr)
7367 		update_debugctlmsr(vmx->host_debugctlmsr);
7368 
7369 #ifndef CONFIG_X86_64
7370 	/*
7371 	 * The sysexit path does not restore ds/es, so we must set them to
7372 	 * a reasonable value ourselves.
7373 	 *
7374 	 * We can't defer this to vmx_prepare_switch_to_host() since that
7375 	 * function may be executed in interrupt context, which saves and
7376 	 * restore segments around it, nullifying its effect.
7377 	 */
7378 	loadsegment(ds, __USER_DS);
7379 	loadsegment(es, __USER_DS);
7380 #endif
7381 
7382 	vcpu->arch.regs_avail &= ~VMX_REGS_LAZY_LOAD_SET;
7383 
7384 	pt_guest_exit(vmx);
7385 
7386 	kvm_load_host_xsave_state(vcpu);
7387 
7388 	if (is_guest_mode(vcpu)) {
7389 		/*
7390 		 * Track VMLAUNCH/VMRESUME that have made past guest state
7391 		 * checking.
7392 		 */
7393 		if (vmx->nested.nested_run_pending &&
7394 		    !vmx->exit_reason.failed_vmentry)
7395 			++vcpu->stat.nested_run;
7396 
7397 		vmx->nested.nested_run_pending = 0;
7398 	}
7399 
7400 	vmx->idt_vectoring_info = 0;
7401 
7402 	if (unlikely(vmx->fail))
7403 		return EXIT_FASTPATH_NONE;
7404 
7405 	if (unlikely((u16)vmx->exit_reason.basic == EXIT_REASON_MCE_DURING_VMENTRY))
7406 		kvm_machine_check();
7407 
7408 	if (likely(!vmx->exit_reason.failed_vmentry))
7409 		vmx->idt_vectoring_info = vmcs_read32(IDT_VECTORING_INFO_FIELD);
7410 
7411 	trace_kvm_exit(vcpu, KVM_ISA_VMX);
7412 
7413 	if (unlikely(vmx->exit_reason.failed_vmentry))
7414 		return EXIT_FASTPATH_NONE;
7415 
7416 	vmx->loaded_vmcs->launched = 1;
7417 
7418 	vmx_recover_nmi_blocking(vmx);
7419 	vmx_complete_interrupts(vmx);
7420 
7421 	if (is_guest_mode(vcpu))
7422 		return EXIT_FASTPATH_NONE;
7423 
7424 	return vmx_exit_handlers_fastpath(vcpu);
7425 }
7426 
7427 static void vmx_vcpu_free(struct kvm_vcpu *vcpu)
7428 {
7429 	struct vcpu_vmx *vmx = to_vmx(vcpu);
7430 
7431 	if (enable_pml)
7432 		vmx_destroy_pml_buffer(vmx);
7433 	free_vpid(vmx->vpid);
7434 	nested_vmx_free_vcpu(vcpu);
7435 	free_loaded_vmcs(vmx->loaded_vmcs);
7436 }
7437 
7438 static int vmx_vcpu_create(struct kvm_vcpu *vcpu)
7439 {
7440 	struct vmx_uret_msr *tsx_ctrl;
7441 	struct vcpu_vmx *vmx;
7442 	int i, err;
7443 
7444 	BUILD_BUG_ON(offsetof(struct vcpu_vmx, vcpu) != 0);
7445 	vmx = to_vmx(vcpu);
7446 
7447 	INIT_LIST_HEAD(&vmx->pi_wakeup_list);
7448 
7449 	err = -ENOMEM;
7450 
7451 	vmx->vpid = allocate_vpid();
7452 
7453 	/*
7454 	 * If PML is turned on, failure on enabling PML just results in failure
7455 	 * of creating the vcpu, therefore we can simplify PML logic (by
7456 	 * avoiding dealing with cases, such as enabling PML partially on vcpus
7457 	 * for the guest), etc.
7458 	 */
7459 	if (enable_pml) {
7460 		vmx->pml_pg = alloc_page(GFP_KERNEL_ACCOUNT | __GFP_ZERO);
7461 		if (!vmx->pml_pg)
7462 			goto free_vpid;
7463 	}
7464 
7465 	for (i = 0; i < kvm_nr_uret_msrs; ++i)
7466 		vmx->guest_uret_msrs[i].mask = -1ull;
7467 	if (boot_cpu_has(X86_FEATURE_RTM)) {
7468 		/*
7469 		 * TSX_CTRL_CPUID_CLEAR is handled in the CPUID interception.
7470 		 * Keep the host value unchanged to avoid changing CPUID bits
7471 		 * under the host kernel's feet.
7472 		 */
7473 		tsx_ctrl = vmx_find_uret_msr(vmx, MSR_IA32_TSX_CTRL);
7474 		if (tsx_ctrl)
7475 			tsx_ctrl->mask = ~(u64)TSX_CTRL_CPUID_CLEAR;
7476 	}
7477 
7478 	err = alloc_loaded_vmcs(&vmx->vmcs01);
7479 	if (err < 0)
7480 		goto free_pml;
7481 
7482 	/*
7483 	 * Use Hyper-V 'Enlightened MSR Bitmap' feature when KVM runs as a
7484 	 * nested (L1) hypervisor and Hyper-V in L0 supports it. Enable the
7485 	 * feature only for vmcs01, KVM currently isn't equipped to realize any
7486 	 * performance benefits from enabling it for vmcs02.
7487 	 */
7488 	if (kvm_is_using_evmcs() &&
7489 	    (ms_hyperv.nested_features & HV_X64_NESTED_MSR_BITMAP)) {
7490 		struct hv_enlightened_vmcs *evmcs = (void *)vmx->vmcs01.vmcs;
7491 
7492 		evmcs->hv_enlightenments_control.msr_bitmap = 1;
7493 	}
7494 
7495 	/* The MSR bitmap starts with all ones */
7496 	bitmap_fill(vmx->shadow_msr_intercept.read, MAX_POSSIBLE_PASSTHROUGH_MSRS);
7497 	bitmap_fill(vmx->shadow_msr_intercept.write, MAX_POSSIBLE_PASSTHROUGH_MSRS);
7498 
7499 	vmx_disable_intercept_for_msr(vcpu, MSR_IA32_TSC, MSR_TYPE_R);
7500 #ifdef CONFIG_X86_64
7501 	vmx_disable_intercept_for_msr(vcpu, MSR_FS_BASE, MSR_TYPE_RW);
7502 	vmx_disable_intercept_for_msr(vcpu, MSR_GS_BASE, MSR_TYPE_RW);
7503 	vmx_disable_intercept_for_msr(vcpu, MSR_KERNEL_GS_BASE, MSR_TYPE_RW);
7504 #endif
7505 	vmx_disable_intercept_for_msr(vcpu, MSR_IA32_SYSENTER_CS, MSR_TYPE_RW);
7506 	vmx_disable_intercept_for_msr(vcpu, MSR_IA32_SYSENTER_ESP, MSR_TYPE_RW);
7507 	vmx_disable_intercept_for_msr(vcpu, MSR_IA32_SYSENTER_EIP, MSR_TYPE_RW);
7508 	if (kvm_cstate_in_guest(vcpu->kvm)) {
7509 		vmx_disable_intercept_for_msr(vcpu, MSR_CORE_C1_RES, MSR_TYPE_R);
7510 		vmx_disable_intercept_for_msr(vcpu, MSR_CORE_C3_RESIDENCY, MSR_TYPE_R);
7511 		vmx_disable_intercept_for_msr(vcpu, MSR_CORE_C6_RESIDENCY, MSR_TYPE_R);
7512 		vmx_disable_intercept_for_msr(vcpu, MSR_CORE_C7_RESIDENCY, MSR_TYPE_R);
7513 	}
7514 
7515 	vmx->loaded_vmcs = &vmx->vmcs01;
7516 
7517 	if (cpu_need_virtualize_apic_accesses(vcpu)) {
7518 		err = kvm_alloc_apic_access_page(vcpu->kvm);
7519 		if (err)
7520 			goto free_vmcs;
7521 	}
7522 
7523 	if (enable_ept && !enable_unrestricted_guest) {
7524 		err = init_rmode_identity_map(vcpu->kvm);
7525 		if (err)
7526 			goto free_vmcs;
7527 	}
7528 
7529 	if (vmx_can_use_ipiv(vcpu))
7530 		WRITE_ONCE(to_kvm_vmx(vcpu->kvm)->pid_table[vcpu->vcpu_id],
7531 			   __pa(&vmx->pi_desc) | PID_TABLE_ENTRY_VALID);
7532 
7533 	return 0;
7534 
7535 free_vmcs:
7536 	free_loaded_vmcs(vmx->loaded_vmcs);
7537 free_pml:
7538 	vmx_destroy_pml_buffer(vmx);
7539 free_vpid:
7540 	free_vpid(vmx->vpid);
7541 	return err;
7542 }
7543 
7544 #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"
7545 #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"
7546 
7547 static int vmx_vm_init(struct kvm *kvm)
7548 {
7549 	if (!ple_gap)
7550 		kvm->arch.pause_in_guest = true;
7551 
7552 	if (boot_cpu_has(X86_BUG_L1TF) && enable_ept) {
7553 		switch (l1tf_mitigation) {
7554 		case L1TF_MITIGATION_OFF:
7555 		case L1TF_MITIGATION_FLUSH_NOWARN:
7556 			/* 'I explicitly don't care' is set */
7557 			break;
7558 		case L1TF_MITIGATION_FLUSH:
7559 		case L1TF_MITIGATION_FLUSH_NOSMT:
7560 		case L1TF_MITIGATION_FULL:
7561 			/*
7562 			 * Warn upon starting the first VM in a potentially
7563 			 * insecure environment.
7564 			 */
7565 			if (sched_smt_active())
7566 				pr_warn_once(L1TF_MSG_SMT);
7567 			if (l1tf_vmx_mitigation == VMENTER_L1D_FLUSH_NEVER)
7568 				pr_warn_once(L1TF_MSG_L1D);
7569 			break;
7570 		case L1TF_MITIGATION_FULL_FORCE:
7571 			/* Flush is enforced */
7572 			break;
7573 		}
7574 	}
7575 	return 0;
7576 }
7577 
7578 static u8 vmx_get_mt_mask(struct kvm_vcpu *vcpu, gfn_t gfn, bool is_mmio)
7579 {
7580 	u8 cache;
7581 
7582 	/* We wanted to honor guest CD/MTRR/PAT, but doing so could result in
7583 	 * memory aliases with conflicting memory types and sometimes MCEs.
7584 	 * We have to be careful as to what are honored and when.
7585 	 *
7586 	 * For MMIO, guest CD/MTRR are ignored.  The EPT memory type is set to
7587 	 * UC.  The effective memory type is UC or WC depending on guest PAT.
7588 	 * This was historically the source of MCEs and we want to be
7589 	 * conservative.
7590 	 *
7591 	 * When there is no need to deal with noncoherent DMA (e.g., no VT-d
7592 	 * or VT-d has snoop control), guest CD/MTRR/PAT are all ignored.  The
7593 	 * EPT memory type is set to WB.  The effective memory type is forced
7594 	 * WB.
7595 	 *
7596 	 * Otherwise, we trust guest.  Guest CD/MTRR/PAT are all honored.  The
7597 	 * EPT memory type is used to emulate guest CD/MTRR.
7598 	 */
7599 
7600 	if (is_mmio)
7601 		return MTRR_TYPE_UNCACHABLE << VMX_EPT_MT_EPTE_SHIFT;
7602 
7603 	if (!kvm_arch_has_noncoherent_dma(vcpu->kvm))
7604 		return (MTRR_TYPE_WRBACK << VMX_EPT_MT_EPTE_SHIFT) | VMX_EPT_IPAT_BIT;
7605 
7606 	if (kvm_read_cr0_bits(vcpu, X86_CR0_CD)) {
7607 		if (kvm_check_has_quirk(vcpu->kvm, KVM_X86_QUIRK_CD_NW_CLEARED))
7608 			cache = MTRR_TYPE_WRBACK;
7609 		else
7610 			cache = MTRR_TYPE_UNCACHABLE;
7611 
7612 		return (cache << VMX_EPT_MT_EPTE_SHIFT) | VMX_EPT_IPAT_BIT;
7613 	}
7614 
7615 	return kvm_mtrr_get_guest_memory_type(vcpu, gfn) << VMX_EPT_MT_EPTE_SHIFT;
7616 }
7617 
7618 static void vmcs_set_secondary_exec_control(struct vcpu_vmx *vmx, u32 new_ctl)
7619 {
7620 	/*
7621 	 * These bits in the secondary execution controls field
7622 	 * are dynamic, the others are mostly based on the hypervisor
7623 	 * architecture and the guest's CPUID.  Do not touch the
7624 	 * dynamic bits.
7625 	 */
7626 	u32 mask =
7627 		SECONDARY_EXEC_SHADOW_VMCS |
7628 		SECONDARY_EXEC_VIRTUALIZE_X2APIC_MODE |
7629 		SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES |
7630 		SECONDARY_EXEC_DESC;
7631 
7632 	u32 cur_ctl = secondary_exec_controls_get(vmx);
7633 
7634 	secondary_exec_controls_set(vmx, (new_ctl & ~mask) | (cur_ctl & mask));
7635 }
7636 
7637 /*
7638  * Generate MSR_IA32_VMX_CR{0,4}_FIXED1 according to CPUID. Only set bits
7639  * (indicating "allowed-1") if they are supported in the guest's CPUID.
7640  */
7641 static void nested_vmx_cr_fixed1_bits_update(struct kvm_vcpu *vcpu)
7642 {
7643 	struct vcpu_vmx *vmx = to_vmx(vcpu);
7644 	struct kvm_cpuid_entry2 *entry;
7645 
7646 	vmx->nested.msrs.cr0_fixed1 = 0xffffffff;
7647 	vmx->nested.msrs.cr4_fixed1 = X86_CR4_PCE;
7648 
7649 #define cr4_fixed1_update(_cr4_mask, _reg, _cpuid_mask) do {		\
7650 	if (entry && (entry->_reg & (_cpuid_mask)))			\
7651 		vmx->nested.msrs.cr4_fixed1 |= (_cr4_mask);	\
7652 } while (0)
7653 
7654 	entry = kvm_find_cpuid_entry(vcpu, 0x1);
7655 	cr4_fixed1_update(X86_CR4_VME,        edx, feature_bit(VME));
7656 	cr4_fixed1_update(X86_CR4_PVI,        edx, feature_bit(VME));
7657 	cr4_fixed1_update(X86_CR4_TSD,        edx, feature_bit(TSC));
7658 	cr4_fixed1_update(X86_CR4_DE,         edx, feature_bit(DE));
7659 	cr4_fixed1_update(X86_CR4_PSE,        edx, feature_bit(PSE));
7660 	cr4_fixed1_update(X86_CR4_PAE,        edx, feature_bit(PAE));
7661 	cr4_fixed1_update(X86_CR4_MCE,        edx, feature_bit(MCE));
7662 	cr4_fixed1_update(X86_CR4_PGE,        edx, feature_bit(PGE));
7663 	cr4_fixed1_update(X86_CR4_OSFXSR,     edx, feature_bit(FXSR));
7664 	cr4_fixed1_update(X86_CR4_OSXMMEXCPT, edx, feature_bit(XMM));
7665 	cr4_fixed1_update(X86_CR4_VMXE,       ecx, feature_bit(VMX));
7666 	cr4_fixed1_update(X86_CR4_SMXE,       ecx, feature_bit(SMX));
7667 	cr4_fixed1_update(X86_CR4_PCIDE,      ecx, feature_bit(PCID));
7668 	cr4_fixed1_update(X86_CR4_OSXSAVE,    ecx, feature_bit(XSAVE));
7669 
7670 	entry = kvm_find_cpuid_entry_index(vcpu, 0x7, 0);
7671 	cr4_fixed1_update(X86_CR4_FSGSBASE,   ebx, feature_bit(FSGSBASE));
7672 	cr4_fixed1_update(X86_CR4_SMEP,       ebx, feature_bit(SMEP));
7673 	cr4_fixed1_update(X86_CR4_SMAP,       ebx, feature_bit(SMAP));
7674 	cr4_fixed1_update(X86_CR4_PKE,        ecx, feature_bit(PKU));
7675 	cr4_fixed1_update(X86_CR4_UMIP,       ecx, feature_bit(UMIP));
7676 	cr4_fixed1_update(X86_CR4_LA57,       ecx, feature_bit(LA57));
7677 
7678 #undef cr4_fixed1_update
7679 }
7680 
7681 static void update_intel_pt_cfg(struct kvm_vcpu *vcpu)
7682 {
7683 	struct vcpu_vmx *vmx = to_vmx(vcpu);
7684 	struct kvm_cpuid_entry2 *best = NULL;
7685 	int i;
7686 
7687 	for (i = 0; i < PT_CPUID_LEAVES; i++) {
7688 		best = kvm_find_cpuid_entry_index(vcpu, 0x14, i);
7689 		if (!best)
7690 			return;
7691 		vmx->pt_desc.caps[CPUID_EAX + i*PT_CPUID_REGS_NUM] = best->eax;
7692 		vmx->pt_desc.caps[CPUID_EBX + i*PT_CPUID_REGS_NUM] = best->ebx;
7693 		vmx->pt_desc.caps[CPUID_ECX + i*PT_CPUID_REGS_NUM] = best->ecx;
7694 		vmx->pt_desc.caps[CPUID_EDX + i*PT_CPUID_REGS_NUM] = best->edx;
7695 	}
7696 
7697 	/* Get the number of configurable Address Ranges for filtering */
7698 	vmx->pt_desc.num_address_ranges = intel_pt_validate_cap(vmx->pt_desc.caps,
7699 						PT_CAP_num_address_ranges);
7700 
7701 	/* Initialize and clear the no dependency bits */
7702 	vmx->pt_desc.ctl_bitmask = ~(RTIT_CTL_TRACEEN | RTIT_CTL_OS |
7703 			RTIT_CTL_USR | RTIT_CTL_TSC_EN | RTIT_CTL_DISRETC |
7704 			RTIT_CTL_BRANCH_EN);
7705 
7706 	/*
7707 	 * If CPUID.(EAX=14H,ECX=0):EBX[0]=1 CR3Filter can be set otherwise
7708 	 * will inject an #GP
7709 	 */
7710 	if (intel_pt_validate_cap(vmx->pt_desc.caps, PT_CAP_cr3_filtering))
7711 		vmx->pt_desc.ctl_bitmask &= ~RTIT_CTL_CR3EN;
7712 
7713 	/*
7714 	 * If CPUID.(EAX=14H,ECX=0):EBX[1]=1 CYCEn, CycThresh and
7715 	 * PSBFreq can be set
7716 	 */
7717 	if (intel_pt_validate_cap(vmx->pt_desc.caps, PT_CAP_psb_cyc))
7718 		vmx->pt_desc.ctl_bitmask &= ~(RTIT_CTL_CYCLEACC |
7719 				RTIT_CTL_CYC_THRESH | RTIT_CTL_PSB_FREQ);
7720 
7721 	/*
7722 	 * If CPUID.(EAX=14H,ECX=0):EBX[3]=1 MTCEn and MTCFreq can be set
7723 	 */
7724 	if (intel_pt_validate_cap(vmx->pt_desc.caps, PT_CAP_mtc))
7725 		vmx->pt_desc.ctl_bitmask &= ~(RTIT_CTL_MTC_EN |
7726 					      RTIT_CTL_MTC_RANGE);
7727 
7728 	/* If CPUID.(EAX=14H,ECX=0):EBX[4]=1 FUPonPTW and PTWEn can be set */
7729 	if (intel_pt_validate_cap(vmx->pt_desc.caps, PT_CAP_ptwrite))
7730 		vmx->pt_desc.ctl_bitmask &= ~(RTIT_CTL_FUP_ON_PTW |
7731 							RTIT_CTL_PTW_EN);
7732 
7733 	/* If CPUID.(EAX=14H,ECX=0):EBX[5]=1 PwrEvEn can be set */
7734 	if (intel_pt_validate_cap(vmx->pt_desc.caps, PT_CAP_power_event_trace))
7735 		vmx->pt_desc.ctl_bitmask &= ~RTIT_CTL_PWR_EVT_EN;
7736 
7737 	/* If CPUID.(EAX=14H,ECX=0):ECX[0]=1 ToPA can be set */
7738 	if (intel_pt_validate_cap(vmx->pt_desc.caps, PT_CAP_topa_output))
7739 		vmx->pt_desc.ctl_bitmask &= ~RTIT_CTL_TOPA;
7740 
7741 	/* If CPUID.(EAX=14H,ECX=0):ECX[3]=1 FabricEn can be set */
7742 	if (intel_pt_validate_cap(vmx->pt_desc.caps, PT_CAP_output_subsys))
7743 		vmx->pt_desc.ctl_bitmask &= ~RTIT_CTL_FABRIC_EN;
7744 
7745 	/* unmask address range configure area */
7746 	for (i = 0; i < vmx->pt_desc.num_address_ranges; i++)
7747 		vmx->pt_desc.ctl_bitmask &= ~(0xfULL << (32 + i * 4));
7748 }
7749 
7750 static void vmx_vcpu_after_set_cpuid(struct kvm_vcpu *vcpu)
7751 {
7752 	struct vcpu_vmx *vmx = to_vmx(vcpu);
7753 
7754 	/* xsaves_enabled is recomputed in vmx_compute_secondary_exec_control(). */
7755 	vcpu->arch.xsaves_enabled = false;
7756 
7757 	vmx_setup_uret_msrs(vmx);
7758 
7759 	if (cpu_has_secondary_exec_ctrls())
7760 		vmcs_set_secondary_exec_control(vmx,
7761 						vmx_secondary_exec_control(vmx));
7762 
7763 	if (nested_vmx_allowed(vcpu))
7764 		vmx->msr_ia32_feature_control_valid_bits |=
7765 			FEAT_CTL_VMX_ENABLED_INSIDE_SMX |
7766 			FEAT_CTL_VMX_ENABLED_OUTSIDE_SMX;
7767 	else
7768 		vmx->msr_ia32_feature_control_valid_bits &=
7769 			~(FEAT_CTL_VMX_ENABLED_INSIDE_SMX |
7770 			  FEAT_CTL_VMX_ENABLED_OUTSIDE_SMX);
7771 
7772 	if (nested_vmx_allowed(vcpu))
7773 		nested_vmx_cr_fixed1_bits_update(vcpu);
7774 
7775 	if (boot_cpu_has(X86_FEATURE_INTEL_PT) &&
7776 			guest_cpuid_has(vcpu, X86_FEATURE_INTEL_PT))
7777 		update_intel_pt_cfg(vcpu);
7778 
7779 	if (boot_cpu_has(X86_FEATURE_RTM)) {
7780 		struct vmx_uret_msr *msr;
7781 		msr = vmx_find_uret_msr(vmx, MSR_IA32_TSX_CTRL);
7782 		if (msr) {
7783 			bool enabled = guest_cpuid_has(vcpu, X86_FEATURE_RTM);
7784 			vmx_set_guest_uret_msr(vmx, msr, enabled ? 0 : TSX_CTRL_RTM_DISABLE);
7785 		}
7786 	}
7787 
7788 	if (kvm_cpu_cap_has(X86_FEATURE_XFD))
7789 		vmx_set_intercept_for_msr(vcpu, MSR_IA32_XFD_ERR, MSR_TYPE_R,
7790 					  !guest_cpuid_has(vcpu, X86_FEATURE_XFD));
7791 
7792 	if (boot_cpu_has(X86_FEATURE_IBPB))
7793 		vmx_set_intercept_for_msr(vcpu, MSR_IA32_PRED_CMD, MSR_TYPE_W,
7794 					  !guest_has_pred_cmd_msr(vcpu));
7795 
7796 	if (boot_cpu_has(X86_FEATURE_FLUSH_L1D))
7797 		vmx_set_intercept_for_msr(vcpu, MSR_IA32_FLUSH_CMD, MSR_TYPE_W,
7798 					  !guest_cpuid_has(vcpu, X86_FEATURE_FLUSH_L1D));
7799 
7800 	set_cr4_guest_host_mask(vmx);
7801 
7802 	vmx_write_encls_bitmap(vcpu, NULL);
7803 	if (guest_cpuid_has(vcpu, X86_FEATURE_SGX))
7804 		vmx->msr_ia32_feature_control_valid_bits |= FEAT_CTL_SGX_ENABLED;
7805 	else
7806 		vmx->msr_ia32_feature_control_valid_bits &= ~FEAT_CTL_SGX_ENABLED;
7807 
7808 	if (guest_cpuid_has(vcpu, X86_FEATURE_SGX_LC))
7809 		vmx->msr_ia32_feature_control_valid_bits |=
7810 			FEAT_CTL_SGX_LC_ENABLED;
7811 	else
7812 		vmx->msr_ia32_feature_control_valid_bits &=
7813 			~FEAT_CTL_SGX_LC_ENABLED;
7814 
7815 	/* Refresh #PF interception to account for MAXPHYADDR changes. */
7816 	vmx_update_exception_bitmap(vcpu);
7817 }
7818 
7819 static u64 vmx_get_perf_capabilities(void)
7820 {
7821 	u64 perf_cap = PMU_CAP_FW_WRITES;
7822 	struct x86_pmu_lbr lbr;
7823 	u64 host_perf_cap = 0;
7824 
7825 	if (!enable_pmu)
7826 		return 0;
7827 
7828 	if (boot_cpu_has(X86_FEATURE_PDCM))
7829 		rdmsrl(MSR_IA32_PERF_CAPABILITIES, host_perf_cap);
7830 
7831 	if (!cpu_feature_enabled(X86_FEATURE_ARCH_LBR)) {
7832 		x86_perf_get_lbr(&lbr);
7833 		if (lbr.nr)
7834 			perf_cap |= host_perf_cap & PMU_CAP_LBR_FMT;
7835 	}
7836 
7837 	if (vmx_pebs_supported()) {
7838 		perf_cap |= host_perf_cap & PERF_CAP_PEBS_MASK;
7839 		if ((perf_cap & PERF_CAP_PEBS_FORMAT) < 4)
7840 			perf_cap &= ~PERF_CAP_PEBS_BASELINE;
7841 	}
7842 
7843 	return perf_cap;
7844 }
7845 
7846 static __init void vmx_set_cpu_caps(void)
7847 {
7848 	kvm_set_cpu_caps();
7849 
7850 	/* CPUID 0x1 */
7851 	if (nested)
7852 		kvm_cpu_cap_set(X86_FEATURE_VMX);
7853 
7854 	/* CPUID 0x7 */
7855 	if (kvm_mpx_supported())
7856 		kvm_cpu_cap_check_and_set(X86_FEATURE_MPX);
7857 	if (!cpu_has_vmx_invpcid())
7858 		kvm_cpu_cap_clear(X86_FEATURE_INVPCID);
7859 	if (vmx_pt_mode_is_host_guest())
7860 		kvm_cpu_cap_check_and_set(X86_FEATURE_INTEL_PT);
7861 	if (vmx_pebs_supported()) {
7862 		kvm_cpu_cap_check_and_set(X86_FEATURE_DS);
7863 		kvm_cpu_cap_check_and_set(X86_FEATURE_DTES64);
7864 	}
7865 
7866 	if (!enable_pmu)
7867 		kvm_cpu_cap_clear(X86_FEATURE_PDCM);
7868 	kvm_caps.supported_perf_cap = vmx_get_perf_capabilities();
7869 
7870 	if (!enable_sgx) {
7871 		kvm_cpu_cap_clear(X86_FEATURE_SGX);
7872 		kvm_cpu_cap_clear(X86_FEATURE_SGX_LC);
7873 		kvm_cpu_cap_clear(X86_FEATURE_SGX1);
7874 		kvm_cpu_cap_clear(X86_FEATURE_SGX2);
7875 	}
7876 
7877 	if (vmx_umip_emulated())
7878 		kvm_cpu_cap_set(X86_FEATURE_UMIP);
7879 
7880 	/* CPUID 0xD.1 */
7881 	kvm_caps.supported_xss = 0;
7882 	if (!cpu_has_vmx_xsaves())
7883 		kvm_cpu_cap_clear(X86_FEATURE_XSAVES);
7884 
7885 	/* CPUID 0x80000001 and 0x7 (RDPID) */
7886 	if (!cpu_has_vmx_rdtscp()) {
7887 		kvm_cpu_cap_clear(X86_FEATURE_RDTSCP);
7888 		kvm_cpu_cap_clear(X86_FEATURE_RDPID);
7889 	}
7890 
7891 	if (cpu_has_vmx_waitpkg())
7892 		kvm_cpu_cap_check_and_set(X86_FEATURE_WAITPKG);
7893 }
7894 
7895 static void vmx_request_immediate_exit(struct kvm_vcpu *vcpu)
7896 {
7897 	to_vmx(vcpu)->req_immediate_exit = true;
7898 }
7899 
7900 static int vmx_check_intercept_io(struct kvm_vcpu *vcpu,
7901 				  struct x86_instruction_info *info)
7902 {
7903 	struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
7904 	unsigned short port;
7905 	bool intercept;
7906 	int size;
7907 
7908 	if (info->intercept == x86_intercept_in ||
7909 	    info->intercept == x86_intercept_ins) {
7910 		port = info->src_val;
7911 		size = info->dst_bytes;
7912 	} else {
7913 		port = info->dst_val;
7914 		size = info->src_bytes;
7915 	}
7916 
7917 	/*
7918 	 * If the 'use IO bitmaps' VM-execution control is 0, IO instruction
7919 	 * VM-exits depend on the 'unconditional IO exiting' VM-execution
7920 	 * control.
7921 	 *
7922 	 * Otherwise, IO instruction VM-exits are controlled by the IO bitmaps.
7923 	 */
7924 	if (!nested_cpu_has(vmcs12, CPU_BASED_USE_IO_BITMAPS))
7925 		intercept = nested_cpu_has(vmcs12,
7926 					   CPU_BASED_UNCOND_IO_EXITING);
7927 	else
7928 		intercept = nested_vmx_check_io_bitmaps(vcpu, port, size);
7929 
7930 	/* FIXME: produce nested vmexit and return X86EMUL_INTERCEPTED.  */
7931 	return intercept ? X86EMUL_UNHANDLEABLE : X86EMUL_CONTINUE;
7932 }
7933 
7934 static int vmx_check_intercept(struct kvm_vcpu *vcpu,
7935 			       struct x86_instruction_info *info,
7936 			       enum x86_intercept_stage stage,
7937 			       struct x86_exception *exception)
7938 {
7939 	struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
7940 
7941 	switch (info->intercept) {
7942 	/*
7943 	 * RDPID causes #UD if disabled through secondary execution controls.
7944 	 * Because it is marked as EmulateOnUD, we need to intercept it here.
7945 	 * Note, RDPID is hidden behind ENABLE_RDTSCP.
7946 	 */
7947 	case x86_intercept_rdpid:
7948 		if (!nested_cpu_has2(vmcs12, SECONDARY_EXEC_ENABLE_RDTSCP)) {
7949 			exception->vector = UD_VECTOR;
7950 			exception->error_code_valid = false;
7951 			return X86EMUL_PROPAGATE_FAULT;
7952 		}
7953 		break;
7954 
7955 	case x86_intercept_in:
7956 	case x86_intercept_ins:
7957 	case x86_intercept_out:
7958 	case x86_intercept_outs:
7959 		return vmx_check_intercept_io(vcpu, info);
7960 
7961 	case x86_intercept_lgdt:
7962 	case x86_intercept_lidt:
7963 	case x86_intercept_lldt:
7964 	case x86_intercept_ltr:
7965 	case x86_intercept_sgdt:
7966 	case x86_intercept_sidt:
7967 	case x86_intercept_sldt:
7968 	case x86_intercept_str:
7969 		if (!nested_cpu_has2(vmcs12, SECONDARY_EXEC_DESC))
7970 			return X86EMUL_CONTINUE;
7971 
7972 		/* FIXME: produce nested vmexit and return X86EMUL_INTERCEPTED.  */
7973 		break;
7974 
7975 	case x86_intercept_pause:
7976 		/*
7977 		 * PAUSE is a single-byte NOP with a REPE prefix, i.e. collides
7978 		 * with vanilla NOPs in the emulator.  Apply the interception
7979 		 * check only to actual PAUSE instructions.  Don't check
7980 		 * PAUSE-loop-exiting, software can't expect a given PAUSE to
7981 		 * exit, i.e. KVM is within its rights to allow L2 to execute
7982 		 * the PAUSE.
7983 		 */
7984 		if ((info->rep_prefix != REPE_PREFIX) ||
7985 		    !nested_cpu_has2(vmcs12, CPU_BASED_PAUSE_EXITING))
7986 			return X86EMUL_CONTINUE;
7987 
7988 		break;
7989 
7990 	/* TODO: check more intercepts... */
7991 	default:
7992 		break;
7993 	}
7994 
7995 	return X86EMUL_UNHANDLEABLE;
7996 }
7997 
7998 #ifdef CONFIG_X86_64
7999 /* (a << shift) / divisor, return 1 if overflow otherwise 0 */
8000 static inline int u64_shl_div_u64(u64 a, unsigned int shift,
8001 				  u64 divisor, u64 *result)
8002 {
8003 	u64 low = a << shift, high = a >> (64 - shift);
8004 
8005 	/* To avoid the overflow on divq */
8006 	if (high >= divisor)
8007 		return 1;
8008 
8009 	/* Low hold the result, high hold rem which is discarded */
8010 	asm("divq %2\n\t" : "=a" (low), "=d" (high) :
8011 	    "rm" (divisor), "0" (low), "1" (high));
8012 	*result = low;
8013 
8014 	return 0;
8015 }
8016 
8017 static int vmx_set_hv_timer(struct kvm_vcpu *vcpu, u64 guest_deadline_tsc,
8018 			    bool *expired)
8019 {
8020 	struct vcpu_vmx *vmx;
8021 	u64 tscl, guest_tscl, delta_tsc, lapic_timer_advance_cycles;
8022 	struct kvm_timer *ktimer = &vcpu->arch.apic->lapic_timer;
8023 
8024 	vmx = to_vmx(vcpu);
8025 	tscl = rdtsc();
8026 	guest_tscl = kvm_read_l1_tsc(vcpu, tscl);
8027 	delta_tsc = max(guest_deadline_tsc, guest_tscl) - guest_tscl;
8028 	lapic_timer_advance_cycles = nsec_to_cycles(vcpu,
8029 						    ktimer->timer_advance_ns);
8030 
8031 	if (delta_tsc > lapic_timer_advance_cycles)
8032 		delta_tsc -= lapic_timer_advance_cycles;
8033 	else
8034 		delta_tsc = 0;
8035 
8036 	/* Convert to host delta tsc if tsc scaling is enabled */
8037 	if (vcpu->arch.l1_tsc_scaling_ratio != kvm_caps.default_tsc_scaling_ratio &&
8038 	    delta_tsc && u64_shl_div_u64(delta_tsc,
8039 				kvm_caps.tsc_scaling_ratio_frac_bits,
8040 				vcpu->arch.l1_tsc_scaling_ratio, &delta_tsc))
8041 		return -ERANGE;
8042 
8043 	/*
8044 	 * If the delta tsc can't fit in the 32 bit after the multi shift,
8045 	 * we can't use the preemption timer.
8046 	 * It's possible that it fits on later vmentries, but checking
8047 	 * on every vmentry is costly so we just use an hrtimer.
8048 	 */
8049 	if (delta_tsc >> (cpu_preemption_timer_multi + 32))
8050 		return -ERANGE;
8051 
8052 	vmx->hv_deadline_tsc = tscl + delta_tsc;
8053 	*expired = !delta_tsc;
8054 	return 0;
8055 }
8056 
8057 static void vmx_cancel_hv_timer(struct kvm_vcpu *vcpu)
8058 {
8059 	to_vmx(vcpu)->hv_deadline_tsc = -1;
8060 }
8061 #endif
8062 
8063 static void vmx_sched_in(struct kvm_vcpu *vcpu, int cpu)
8064 {
8065 	if (!kvm_pause_in_guest(vcpu->kvm))
8066 		shrink_ple_window(vcpu);
8067 }
8068 
8069 void vmx_update_cpu_dirty_logging(struct kvm_vcpu *vcpu)
8070 {
8071 	struct vcpu_vmx *vmx = to_vmx(vcpu);
8072 
8073 	if (WARN_ON_ONCE(!enable_pml))
8074 		return;
8075 
8076 	if (is_guest_mode(vcpu)) {
8077 		vmx->nested.update_vmcs01_cpu_dirty_logging = true;
8078 		return;
8079 	}
8080 
8081 	/*
8082 	 * Note, nr_memslots_dirty_logging can be changed concurrent with this
8083 	 * code, but in that case another update request will be made and so
8084 	 * the guest will never run with a stale PML value.
8085 	 */
8086 	if (atomic_read(&vcpu->kvm->nr_memslots_dirty_logging))
8087 		secondary_exec_controls_setbit(vmx, SECONDARY_EXEC_ENABLE_PML);
8088 	else
8089 		secondary_exec_controls_clearbit(vmx, SECONDARY_EXEC_ENABLE_PML);
8090 }
8091 
8092 static void vmx_setup_mce(struct kvm_vcpu *vcpu)
8093 {
8094 	if (vcpu->arch.mcg_cap & MCG_LMCE_P)
8095 		to_vmx(vcpu)->msr_ia32_feature_control_valid_bits |=
8096 			FEAT_CTL_LMCE_ENABLED;
8097 	else
8098 		to_vmx(vcpu)->msr_ia32_feature_control_valid_bits &=
8099 			~FEAT_CTL_LMCE_ENABLED;
8100 }
8101 
8102 #ifdef CONFIG_KVM_SMM
8103 static int vmx_smi_allowed(struct kvm_vcpu *vcpu, bool for_injection)
8104 {
8105 	/* we need a nested vmexit to enter SMM, postpone if run is pending */
8106 	if (to_vmx(vcpu)->nested.nested_run_pending)
8107 		return -EBUSY;
8108 	return !is_smm(vcpu);
8109 }
8110 
8111 static int vmx_enter_smm(struct kvm_vcpu *vcpu, union kvm_smram *smram)
8112 {
8113 	struct vcpu_vmx *vmx = to_vmx(vcpu);
8114 
8115 	/*
8116 	 * TODO: Implement custom flows for forcing the vCPU out/in of L2 on
8117 	 * SMI and RSM.  Using the common VM-Exit + VM-Enter routines is wrong
8118 	 * SMI and RSM only modify state that is saved and restored via SMRAM.
8119 	 * E.g. most MSRs are left untouched, but many are modified by VM-Exit
8120 	 * and VM-Enter, and thus L2's values may be corrupted on SMI+RSM.
8121 	 */
8122 	vmx->nested.smm.guest_mode = is_guest_mode(vcpu);
8123 	if (vmx->nested.smm.guest_mode)
8124 		nested_vmx_vmexit(vcpu, -1, 0, 0);
8125 
8126 	vmx->nested.smm.vmxon = vmx->nested.vmxon;
8127 	vmx->nested.vmxon = false;
8128 	vmx_clear_hlt(vcpu);
8129 	return 0;
8130 }
8131 
8132 static int vmx_leave_smm(struct kvm_vcpu *vcpu, const union kvm_smram *smram)
8133 {
8134 	struct vcpu_vmx *vmx = to_vmx(vcpu);
8135 	int ret;
8136 
8137 	if (vmx->nested.smm.vmxon) {
8138 		vmx->nested.vmxon = true;
8139 		vmx->nested.smm.vmxon = false;
8140 	}
8141 
8142 	if (vmx->nested.smm.guest_mode) {
8143 		ret = nested_vmx_enter_non_root_mode(vcpu, false);
8144 		if (ret)
8145 			return ret;
8146 
8147 		vmx->nested.nested_run_pending = 1;
8148 		vmx->nested.smm.guest_mode = false;
8149 	}
8150 	return 0;
8151 }
8152 
8153 static void vmx_enable_smi_window(struct kvm_vcpu *vcpu)
8154 {
8155 	/* RSM will cause a vmexit anyway.  */
8156 }
8157 #endif
8158 
8159 static bool vmx_apic_init_signal_blocked(struct kvm_vcpu *vcpu)
8160 {
8161 	return to_vmx(vcpu)->nested.vmxon && !is_guest_mode(vcpu);
8162 }
8163 
8164 static void vmx_migrate_timers(struct kvm_vcpu *vcpu)
8165 {
8166 	if (is_guest_mode(vcpu)) {
8167 		struct hrtimer *timer = &to_vmx(vcpu)->nested.preemption_timer;
8168 
8169 		if (hrtimer_try_to_cancel(timer) == 1)
8170 			hrtimer_start_expires(timer, HRTIMER_MODE_ABS_PINNED);
8171 	}
8172 }
8173 
8174 static void vmx_hardware_unsetup(void)
8175 {
8176 	kvm_set_posted_intr_wakeup_handler(NULL);
8177 
8178 	if (nested)
8179 		nested_vmx_hardware_unsetup();
8180 
8181 	free_kvm_area();
8182 }
8183 
8184 #define VMX_REQUIRED_APICV_INHIBITS			\
8185 (							\
8186 	BIT(APICV_INHIBIT_REASON_DISABLE)|		\
8187 	BIT(APICV_INHIBIT_REASON_ABSENT) |		\
8188 	BIT(APICV_INHIBIT_REASON_HYPERV) |		\
8189 	BIT(APICV_INHIBIT_REASON_BLOCKIRQ) |		\
8190 	BIT(APICV_INHIBIT_REASON_PHYSICAL_ID_ALIASED) |	\
8191 	BIT(APICV_INHIBIT_REASON_APIC_ID_MODIFIED) |	\
8192 	BIT(APICV_INHIBIT_REASON_APIC_BASE_MODIFIED)	\
8193 )
8194 
8195 static void vmx_vm_destroy(struct kvm *kvm)
8196 {
8197 	struct kvm_vmx *kvm_vmx = to_kvm_vmx(kvm);
8198 
8199 	free_pages((unsigned long)kvm_vmx->pid_table, vmx_get_pid_table_order(kvm));
8200 }
8201 
8202 static struct kvm_x86_ops vmx_x86_ops __initdata = {
8203 	.name = KBUILD_MODNAME,
8204 
8205 	.check_processor_compatibility = vmx_check_processor_compat,
8206 
8207 	.hardware_unsetup = vmx_hardware_unsetup,
8208 
8209 	.hardware_enable = vmx_hardware_enable,
8210 	.hardware_disable = vmx_hardware_disable,
8211 	.has_emulated_msr = vmx_has_emulated_msr,
8212 
8213 	.vm_size = sizeof(struct kvm_vmx),
8214 	.vm_init = vmx_vm_init,
8215 	.vm_destroy = vmx_vm_destroy,
8216 
8217 	.vcpu_precreate = vmx_vcpu_precreate,
8218 	.vcpu_create = vmx_vcpu_create,
8219 	.vcpu_free = vmx_vcpu_free,
8220 	.vcpu_reset = vmx_vcpu_reset,
8221 
8222 	.prepare_switch_to_guest = vmx_prepare_switch_to_guest,
8223 	.vcpu_load = vmx_vcpu_load,
8224 	.vcpu_put = vmx_vcpu_put,
8225 
8226 	.update_exception_bitmap = vmx_update_exception_bitmap,
8227 	.get_msr_feature = vmx_get_msr_feature,
8228 	.get_msr = vmx_get_msr,
8229 	.set_msr = vmx_set_msr,
8230 	.get_segment_base = vmx_get_segment_base,
8231 	.get_segment = vmx_get_segment,
8232 	.set_segment = vmx_set_segment,
8233 	.get_cpl = vmx_get_cpl,
8234 	.get_cs_db_l_bits = vmx_get_cs_db_l_bits,
8235 	.is_valid_cr0 = vmx_is_valid_cr0,
8236 	.set_cr0 = vmx_set_cr0,
8237 	.is_valid_cr4 = vmx_is_valid_cr4,
8238 	.set_cr4 = vmx_set_cr4,
8239 	.set_efer = vmx_set_efer,
8240 	.get_idt = vmx_get_idt,
8241 	.set_idt = vmx_set_idt,
8242 	.get_gdt = vmx_get_gdt,
8243 	.set_gdt = vmx_set_gdt,
8244 	.set_dr7 = vmx_set_dr7,
8245 	.sync_dirty_debug_regs = vmx_sync_dirty_debug_regs,
8246 	.cache_reg = vmx_cache_reg,
8247 	.get_rflags = vmx_get_rflags,
8248 	.set_rflags = vmx_set_rflags,
8249 	.get_if_flag = vmx_get_if_flag,
8250 
8251 	.flush_tlb_all = vmx_flush_tlb_all,
8252 	.flush_tlb_current = vmx_flush_tlb_current,
8253 	.flush_tlb_gva = vmx_flush_tlb_gva,
8254 	.flush_tlb_guest = vmx_flush_tlb_guest,
8255 
8256 	.vcpu_pre_run = vmx_vcpu_pre_run,
8257 	.vcpu_run = vmx_vcpu_run,
8258 	.handle_exit = vmx_handle_exit,
8259 	.skip_emulated_instruction = vmx_skip_emulated_instruction,
8260 	.update_emulated_instruction = vmx_update_emulated_instruction,
8261 	.set_interrupt_shadow = vmx_set_interrupt_shadow,
8262 	.get_interrupt_shadow = vmx_get_interrupt_shadow,
8263 	.patch_hypercall = vmx_patch_hypercall,
8264 	.inject_irq = vmx_inject_irq,
8265 	.inject_nmi = vmx_inject_nmi,
8266 	.inject_exception = vmx_inject_exception,
8267 	.cancel_injection = vmx_cancel_injection,
8268 	.interrupt_allowed = vmx_interrupt_allowed,
8269 	.nmi_allowed = vmx_nmi_allowed,
8270 	.get_nmi_mask = vmx_get_nmi_mask,
8271 	.set_nmi_mask = vmx_set_nmi_mask,
8272 	.enable_nmi_window = vmx_enable_nmi_window,
8273 	.enable_irq_window = vmx_enable_irq_window,
8274 	.update_cr8_intercept = vmx_update_cr8_intercept,
8275 	.set_virtual_apic_mode = vmx_set_virtual_apic_mode,
8276 	.set_apic_access_page_addr = vmx_set_apic_access_page_addr,
8277 	.refresh_apicv_exec_ctrl = vmx_refresh_apicv_exec_ctrl,
8278 	.load_eoi_exitmap = vmx_load_eoi_exitmap,
8279 	.apicv_post_state_restore = vmx_apicv_post_state_restore,
8280 	.required_apicv_inhibits = VMX_REQUIRED_APICV_INHIBITS,
8281 	.hwapic_irr_update = vmx_hwapic_irr_update,
8282 	.hwapic_isr_update = vmx_hwapic_isr_update,
8283 	.guest_apic_has_interrupt = vmx_guest_apic_has_interrupt,
8284 	.sync_pir_to_irr = vmx_sync_pir_to_irr,
8285 	.deliver_interrupt = vmx_deliver_interrupt,
8286 	.dy_apicv_has_pending_interrupt = pi_has_pending_interrupt,
8287 
8288 	.set_tss_addr = vmx_set_tss_addr,
8289 	.set_identity_map_addr = vmx_set_identity_map_addr,
8290 	.get_mt_mask = vmx_get_mt_mask,
8291 
8292 	.get_exit_info = vmx_get_exit_info,
8293 
8294 	.vcpu_after_set_cpuid = vmx_vcpu_after_set_cpuid,
8295 
8296 	.has_wbinvd_exit = cpu_has_vmx_wbinvd_exit,
8297 
8298 	.get_l2_tsc_offset = vmx_get_l2_tsc_offset,
8299 	.get_l2_tsc_multiplier = vmx_get_l2_tsc_multiplier,
8300 	.write_tsc_offset = vmx_write_tsc_offset,
8301 	.write_tsc_multiplier = vmx_write_tsc_multiplier,
8302 
8303 	.load_mmu_pgd = vmx_load_mmu_pgd,
8304 
8305 	.check_intercept = vmx_check_intercept,
8306 	.handle_exit_irqoff = vmx_handle_exit_irqoff,
8307 
8308 	.request_immediate_exit = vmx_request_immediate_exit,
8309 
8310 	.sched_in = vmx_sched_in,
8311 
8312 	.cpu_dirty_log_size = PML_ENTITY_NUM,
8313 	.update_cpu_dirty_logging = vmx_update_cpu_dirty_logging,
8314 
8315 	.nested_ops = &vmx_nested_ops,
8316 
8317 	.pi_update_irte = vmx_pi_update_irte,
8318 	.pi_start_assignment = vmx_pi_start_assignment,
8319 
8320 #ifdef CONFIG_X86_64
8321 	.set_hv_timer = vmx_set_hv_timer,
8322 	.cancel_hv_timer = vmx_cancel_hv_timer,
8323 #endif
8324 
8325 	.setup_mce = vmx_setup_mce,
8326 
8327 #ifdef CONFIG_KVM_SMM
8328 	.smi_allowed = vmx_smi_allowed,
8329 	.enter_smm = vmx_enter_smm,
8330 	.leave_smm = vmx_leave_smm,
8331 	.enable_smi_window = vmx_enable_smi_window,
8332 #endif
8333 
8334 	.can_emulate_instruction = vmx_can_emulate_instruction,
8335 	.apic_init_signal_blocked = vmx_apic_init_signal_blocked,
8336 	.migrate_timers = vmx_migrate_timers,
8337 
8338 	.msr_filter_changed = vmx_msr_filter_changed,
8339 	.complete_emulated_msr = kvm_complete_insn_gp,
8340 
8341 	.vcpu_deliver_sipi_vector = kvm_vcpu_deliver_sipi_vector,
8342 };
8343 
8344 static unsigned int vmx_handle_intel_pt_intr(void)
8345 {
8346 	struct kvm_vcpu *vcpu = kvm_get_running_vcpu();
8347 
8348 	/* '0' on failure so that the !PT case can use a RET0 static call. */
8349 	if (!vcpu || !kvm_handling_nmi_from_guest(vcpu))
8350 		return 0;
8351 
8352 	kvm_make_request(KVM_REQ_PMI, vcpu);
8353 	__set_bit(MSR_CORE_PERF_GLOBAL_OVF_CTRL_TRACE_TOPA_PMI_BIT,
8354 		  (unsigned long *)&vcpu->arch.pmu.global_status);
8355 	return 1;
8356 }
8357 
8358 static __init void vmx_setup_user_return_msrs(void)
8359 {
8360 
8361 	/*
8362 	 * Though SYSCALL is only supported in 64-bit mode on Intel CPUs, kvm
8363 	 * will emulate SYSCALL in legacy mode if the vendor string in guest
8364 	 * CPUID.0:{EBX,ECX,EDX} is "AuthenticAMD" or "AMDisbetter!" To
8365 	 * support this emulation, MSR_STAR is included in the list for i386,
8366 	 * but is never loaded into hardware.  MSR_CSTAR is also never loaded
8367 	 * into hardware and is here purely for emulation purposes.
8368 	 */
8369 	const u32 vmx_uret_msrs_list[] = {
8370 	#ifdef CONFIG_X86_64
8371 		MSR_SYSCALL_MASK, MSR_LSTAR, MSR_CSTAR,
8372 	#endif
8373 		MSR_EFER, MSR_TSC_AUX, MSR_STAR,
8374 		MSR_IA32_TSX_CTRL,
8375 	};
8376 	int i;
8377 
8378 	BUILD_BUG_ON(ARRAY_SIZE(vmx_uret_msrs_list) != MAX_NR_USER_RETURN_MSRS);
8379 
8380 	for (i = 0; i < ARRAY_SIZE(vmx_uret_msrs_list); ++i)
8381 		kvm_add_user_return_msr(vmx_uret_msrs_list[i]);
8382 }
8383 
8384 static void __init vmx_setup_me_spte_mask(void)
8385 {
8386 	u64 me_mask = 0;
8387 
8388 	/*
8389 	 * kvm_get_shadow_phys_bits() returns shadow_phys_bits.  Use
8390 	 * the former to avoid exposing shadow_phys_bits.
8391 	 *
8392 	 * On pre-MKTME system, boot_cpu_data.x86_phys_bits equals to
8393 	 * shadow_phys_bits.  On MKTME and/or TDX capable systems,
8394 	 * boot_cpu_data.x86_phys_bits holds the actual physical address
8395 	 * w/o the KeyID bits, and shadow_phys_bits equals to MAXPHYADDR
8396 	 * reported by CPUID.  Those bits between are KeyID bits.
8397 	 */
8398 	if (boot_cpu_data.x86_phys_bits != kvm_get_shadow_phys_bits())
8399 		me_mask = rsvd_bits(boot_cpu_data.x86_phys_bits,
8400 			kvm_get_shadow_phys_bits() - 1);
8401 	/*
8402 	 * Unlike SME, host kernel doesn't support setting up any
8403 	 * MKTME KeyID on Intel platforms.  No memory encryption
8404 	 * bits should be included into the SPTE.
8405 	 */
8406 	kvm_mmu_set_me_spte_mask(0, me_mask);
8407 }
8408 
8409 static struct kvm_x86_init_ops vmx_init_ops __initdata;
8410 
8411 static __init int hardware_setup(void)
8412 {
8413 	unsigned long host_bndcfgs;
8414 	struct desc_ptr dt;
8415 	int r;
8416 
8417 	store_idt(&dt);
8418 	host_idt_base = dt.address;
8419 
8420 	vmx_setup_user_return_msrs();
8421 
8422 	if (setup_vmcs_config(&vmcs_config, &vmx_capability) < 0)
8423 		return -EIO;
8424 
8425 	if (cpu_has_perf_global_ctrl_bug())
8426 		pr_warn_once("VM_EXIT_LOAD_IA32_PERF_GLOBAL_CTRL "
8427 			     "does not work properly. Using workaround\n");
8428 
8429 	if (boot_cpu_has(X86_FEATURE_NX))
8430 		kvm_enable_efer_bits(EFER_NX);
8431 
8432 	if (boot_cpu_has(X86_FEATURE_MPX)) {
8433 		rdmsrl(MSR_IA32_BNDCFGS, host_bndcfgs);
8434 		WARN_ONCE(host_bndcfgs, "BNDCFGS in host will be lost");
8435 	}
8436 
8437 	if (!cpu_has_vmx_mpx())
8438 		kvm_caps.supported_xcr0 &= ~(XFEATURE_MASK_BNDREGS |
8439 					     XFEATURE_MASK_BNDCSR);
8440 
8441 	if (!cpu_has_vmx_vpid() || !cpu_has_vmx_invvpid() ||
8442 	    !(cpu_has_vmx_invvpid_single() || cpu_has_vmx_invvpid_global()))
8443 		enable_vpid = 0;
8444 
8445 	if (!cpu_has_vmx_ept() ||
8446 	    !cpu_has_vmx_ept_4levels() ||
8447 	    !cpu_has_vmx_ept_mt_wb() ||
8448 	    !cpu_has_vmx_invept_global())
8449 		enable_ept = 0;
8450 
8451 	/* NX support is required for shadow paging. */
8452 	if (!enable_ept && !boot_cpu_has(X86_FEATURE_NX)) {
8453 		pr_err_ratelimited("NX (Execute Disable) not supported\n");
8454 		return -EOPNOTSUPP;
8455 	}
8456 
8457 	if (!cpu_has_vmx_ept_ad_bits() || !enable_ept)
8458 		enable_ept_ad_bits = 0;
8459 
8460 	if (!cpu_has_vmx_unrestricted_guest() || !enable_ept)
8461 		enable_unrestricted_guest = 0;
8462 
8463 	if (!cpu_has_vmx_flexpriority())
8464 		flexpriority_enabled = 0;
8465 
8466 	if (!cpu_has_virtual_nmis())
8467 		enable_vnmi = 0;
8468 
8469 #ifdef CONFIG_X86_SGX_KVM
8470 	if (!cpu_has_vmx_encls_vmexit())
8471 		enable_sgx = false;
8472 #endif
8473 
8474 	/*
8475 	 * set_apic_access_page_addr() is used to reload apic access
8476 	 * page upon invalidation.  No need to do anything if not
8477 	 * using the APIC_ACCESS_ADDR VMCS field.
8478 	 */
8479 	if (!flexpriority_enabled)
8480 		vmx_x86_ops.set_apic_access_page_addr = NULL;
8481 
8482 	if (!cpu_has_vmx_tpr_shadow())
8483 		vmx_x86_ops.update_cr8_intercept = NULL;
8484 
8485 #if IS_ENABLED(CONFIG_HYPERV)
8486 	if (ms_hyperv.nested_features & HV_X64_NESTED_GUEST_MAPPING_FLUSH
8487 	    && enable_ept) {
8488 		vmx_x86_ops.flush_remote_tlbs = hv_flush_remote_tlbs;
8489 		vmx_x86_ops.flush_remote_tlbs_range = hv_flush_remote_tlbs_range;
8490 	}
8491 #endif
8492 
8493 	if (!cpu_has_vmx_ple()) {
8494 		ple_gap = 0;
8495 		ple_window = 0;
8496 		ple_window_grow = 0;
8497 		ple_window_max = 0;
8498 		ple_window_shrink = 0;
8499 	}
8500 
8501 	if (!cpu_has_vmx_apicv())
8502 		enable_apicv = 0;
8503 	if (!enable_apicv)
8504 		vmx_x86_ops.sync_pir_to_irr = NULL;
8505 
8506 	if (!enable_apicv || !cpu_has_vmx_ipiv())
8507 		enable_ipiv = false;
8508 
8509 	if (cpu_has_vmx_tsc_scaling())
8510 		kvm_caps.has_tsc_control = true;
8511 
8512 	kvm_caps.max_tsc_scaling_ratio = KVM_VMX_TSC_MULTIPLIER_MAX;
8513 	kvm_caps.tsc_scaling_ratio_frac_bits = 48;
8514 	kvm_caps.has_bus_lock_exit = cpu_has_vmx_bus_lock_detection();
8515 	kvm_caps.has_notify_vmexit = cpu_has_notify_vmexit();
8516 
8517 	set_bit(0, vmx_vpid_bitmap); /* 0 is reserved for host */
8518 
8519 	if (enable_ept)
8520 		kvm_mmu_set_ept_masks(enable_ept_ad_bits,
8521 				      cpu_has_vmx_ept_execute_only());
8522 
8523 	/*
8524 	 * Setup shadow_me_value/shadow_me_mask to include MKTME KeyID
8525 	 * bits to shadow_zero_check.
8526 	 */
8527 	vmx_setup_me_spte_mask();
8528 
8529 	kvm_configure_mmu(enable_ept, 0, vmx_get_max_tdp_level(),
8530 			  ept_caps_to_lpage_level(vmx_capability.ept));
8531 
8532 	/*
8533 	 * Only enable PML when hardware supports PML feature, and both EPT
8534 	 * and EPT A/D bit features are enabled -- PML depends on them to work.
8535 	 */
8536 	if (!enable_ept || !enable_ept_ad_bits || !cpu_has_vmx_pml())
8537 		enable_pml = 0;
8538 
8539 	if (!enable_pml)
8540 		vmx_x86_ops.cpu_dirty_log_size = 0;
8541 
8542 	if (!cpu_has_vmx_preemption_timer())
8543 		enable_preemption_timer = false;
8544 
8545 	if (enable_preemption_timer) {
8546 		u64 use_timer_freq = 5000ULL * 1000 * 1000;
8547 
8548 		cpu_preemption_timer_multi =
8549 			vmcs_config.misc & VMX_MISC_PREEMPTION_TIMER_RATE_MASK;
8550 
8551 		if (tsc_khz)
8552 			use_timer_freq = (u64)tsc_khz * 1000;
8553 		use_timer_freq >>= cpu_preemption_timer_multi;
8554 
8555 		/*
8556 		 * KVM "disables" the preemption timer by setting it to its max
8557 		 * value.  Don't use the timer if it might cause spurious exits
8558 		 * at a rate faster than 0.1 Hz (of uninterrupted guest time).
8559 		 */
8560 		if (use_timer_freq > 0xffffffffu / 10)
8561 			enable_preemption_timer = false;
8562 	}
8563 
8564 	if (!enable_preemption_timer) {
8565 		vmx_x86_ops.set_hv_timer = NULL;
8566 		vmx_x86_ops.cancel_hv_timer = NULL;
8567 		vmx_x86_ops.request_immediate_exit = __kvm_request_immediate_exit;
8568 	}
8569 
8570 	kvm_caps.supported_mce_cap |= MCG_LMCE_P;
8571 	kvm_caps.supported_mce_cap |= MCG_CMCI_P;
8572 
8573 	if (pt_mode != PT_MODE_SYSTEM && pt_mode != PT_MODE_HOST_GUEST)
8574 		return -EINVAL;
8575 	if (!enable_ept || !enable_pmu || !cpu_has_vmx_intel_pt())
8576 		pt_mode = PT_MODE_SYSTEM;
8577 	if (pt_mode == PT_MODE_HOST_GUEST)
8578 		vmx_init_ops.handle_intel_pt_intr = vmx_handle_intel_pt_intr;
8579 	else
8580 		vmx_init_ops.handle_intel_pt_intr = NULL;
8581 
8582 	setup_default_sgx_lepubkeyhash();
8583 
8584 	if (nested) {
8585 		nested_vmx_setup_ctls_msrs(&vmcs_config, vmx_capability.ept);
8586 
8587 		r = nested_vmx_hardware_setup(kvm_vmx_exit_handlers);
8588 		if (r)
8589 			return r;
8590 	}
8591 
8592 	vmx_set_cpu_caps();
8593 
8594 	r = alloc_kvm_area();
8595 	if (r && nested)
8596 		nested_vmx_hardware_unsetup();
8597 
8598 	kvm_set_posted_intr_wakeup_handler(pi_wakeup_handler);
8599 
8600 	return r;
8601 }
8602 
8603 static struct kvm_x86_init_ops vmx_init_ops __initdata = {
8604 	.hardware_setup = hardware_setup,
8605 	.handle_intel_pt_intr = NULL,
8606 
8607 	.runtime_ops = &vmx_x86_ops,
8608 	.pmu_ops = &intel_pmu_ops,
8609 };
8610 
8611 static void vmx_cleanup_l1d_flush(void)
8612 {
8613 	if (vmx_l1d_flush_pages) {
8614 		free_pages((unsigned long)vmx_l1d_flush_pages, L1D_CACHE_ORDER);
8615 		vmx_l1d_flush_pages = NULL;
8616 	}
8617 	/* Restore state so sysfs ignores VMX */
8618 	l1tf_vmx_mitigation = VMENTER_L1D_FLUSH_AUTO;
8619 }
8620 
8621 static void __vmx_exit(void)
8622 {
8623 	allow_smaller_maxphyaddr = false;
8624 
8625 #ifdef CONFIG_KEXEC_CORE
8626 	RCU_INIT_POINTER(crash_vmclear_loaded_vmcss, NULL);
8627 	synchronize_rcu();
8628 #endif
8629 	vmx_cleanup_l1d_flush();
8630 }
8631 
8632 static void vmx_exit(void)
8633 {
8634 	kvm_exit();
8635 	kvm_x86_vendor_exit();
8636 
8637 	__vmx_exit();
8638 }
8639 module_exit(vmx_exit);
8640 
8641 static int __init vmx_init(void)
8642 {
8643 	int r, cpu;
8644 
8645 	if (!kvm_is_vmx_supported())
8646 		return -EOPNOTSUPP;
8647 
8648 	/*
8649 	 * Note, hv_init_evmcs() touches only VMX knobs, i.e. there's nothing
8650 	 * to unwind if a later step fails.
8651 	 */
8652 	hv_init_evmcs();
8653 
8654 	r = kvm_x86_vendor_init(&vmx_init_ops);
8655 	if (r)
8656 		return r;
8657 
8658 	/*
8659 	 * Must be called after common x86 init so enable_ept is properly set
8660 	 * up. Hand the parameter mitigation value in which was stored in
8661 	 * the pre module init parser. If no parameter was given, it will
8662 	 * contain 'auto' which will be turned into the default 'cond'
8663 	 * mitigation mode.
8664 	 */
8665 	r = vmx_setup_l1d_flush(vmentry_l1d_flush_param);
8666 	if (r)
8667 		goto err_l1d_flush;
8668 
8669 	vmx_setup_fb_clear_ctrl();
8670 
8671 	for_each_possible_cpu(cpu) {
8672 		INIT_LIST_HEAD(&per_cpu(loaded_vmcss_on_cpu, cpu));
8673 
8674 		pi_init_cpu(cpu);
8675 	}
8676 
8677 #ifdef CONFIG_KEXEC_CORE
8678 	rcu_assign_pointer(crash_vmclear_loaded_vmcss,
8679 			   crash_vmclear_local_loaded_vmcss);
8680 #endif
8681 	vmx_check_vmcs12_offsets();
8682 
8683 	/*
8684 	 * Shadow paging doesn't have a (further) performance penalty
8685 	 * from GUEST_MAXPHYADDR < HOST_MAXPHYADDR so enable it
8686 	 * by default
8687 	 */
8688 	if (!enable_ept)
8689 		allow_smaller_maxphyaddr = true;
8690 
8691 	/*
8692 	 * Common KVM initialization _must_ come last, after this, /dev/kvm is
8693 	 * exposed to userspace!
8694 	 */
8695 	r = kvm_init(sizeof(struct vcpu_vmx), __alignof__(struct vcpu_vmx),
8696 		     THIS_MODULE);
8697 	if (r)
8698 		goto err_kvm_init;
8699 
8700 	return 0;
8701 
8702 err_kvm_init:
8703 	__vmx_exit();
8704 err_l1d_flush:
8705 	kvm_x86_vendor_exit();
8706 	return r;
8707 }
8708 module_init(vmx_init);
8709