xref: /linux/arch/x86/kvm/vmx/vmx.c (revision c909a49128a31bced8cfbd2dfb0a4fe56e01a6d0)
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/fred.h>
42 #include <asm/idtentry.h>
43 #include <asm/io.h>
44 #include <asm/irq_remapping.h>
45 #include <asm/reboot.h>
46 #include <asm/perf_event.h>
47 #include <asm/mmu_context.h>
48 #include <asm/mshyperv.h>
49 #include <asm/mwait.h>
50 #include <asm/spec-ctrl.h>
51 #include <asm/vmx.h>
52 
53 #include <trace/events/ipi.h>
54 
55 #include "capabilities.h"
56 #include "cpuid.h"
57 #include "hyperv.h"
58 #include "kvm_onhyperv.h"
59 #include "irq.h"
60 #include "kvm_cache_regs.h"
61 #include "lapic.h"
62 #include "mmu.h"
63 #include "nested.h"
64 #include "pmu.h"
65 #include "sgx.h"
66 #include "trace.h"
67 #include "vmcs.h"
68 #include "vmcs12.h"
69 #include "vmx.h"
70 #include "x86.h"
71 #include "x86_ops.h"
72 #include "smm.h"
73 #include "vmx_onhyperv.h"
74 #include "posted_intr.h"
75 
76 MODULE_AUTHOR("Qumranet");
77 MODULE_DESCRIPTION("KVM support for VMX (Intel VT-x) extensions");
78 MODULE_LICENSE("GPL");
79 
80 #ifdef MODULE
81 static const struct x86_cpu_id vmx_cpu_id[] = {
82 	X86_MATCH_FEATURE(X86_FEATURE_VMX, NULL),
83 	{}
84 };
85 MODULE_DEVICE_TABLE(x86cpu, vmx_cpu_id);
86 #endif
87 
88 bool __read_mostly enable_vpid = 1;
89 module_param_named(vpid, enable_vpid, bool, 0444);
90 
91 static bool __read_mostly enable_vnmi = 1;
92 module_param_named(vnmi, enable_vnmi, bool, 0444);
93 
94 bool __read_mostly flexpriority_enabled = 1;
95 module_param_named(flexpriority, flexpriority_enabled, bool, 0444);
96 
97 bool __read_mostly enable_ept = 1;
98 module_param_named(ept, enable_ept, bool, 0444);
99 
100 bool __read_mostly enable_unrestricted_guest = 1;
101 module_param_named(unrestricted_guest,
102 			enable_unrestricted_guest, bool, 0444);
103 
104 bool __read_mostly enable_ept_ad_bits = 1;
105 module_param_named(eptad, enable_ept_ad_bits, bool, 0444);
106 
107 static bool __read_mostly emulate_invalid_guest_state = true;
108 module_param(emulate_invalid_guest_state, bool, 0444);
109 
110 static bool __read_mostly fasteoi = 1;
111 module_param(fasteoi, bool, 0444);
112 
113 module_param(enable_apicv, bool, 0444);
114 
115 bool __read_mostly enable_ipiv = true;
116 module_param(enable_ipiv, bool, 0444);
117 
118 /*
119  * If nested=1, nested virtualization is supported, i.e., guests may use
120  * VMX and be a hypervisor for its own guests. If nested=0, guests may not
121  * use VMX instructions.
122  */
123 static bool __read_mostly nested = 1;
124 module_param(nested, bool, 0444);
125 
126 bool __read_mostly enable_pml = 1;
127 module_param_named(pml, enable_pml, bool, 0444);
128 
129 static bool __read_mostly error_on_inconsistent_vmcs_config = true;
130 module_param(error_on_inconsistent_vmcs_config, bool, 0444);
131 
132 static bool __read_mostly dump_invalid_vmcs = 0;
133 module_param(dump_invalid_vmcs, bool, 0644);
134 
135 #define MSR_BITMAP_MODE_X2APIC		1
136 #define MSR_BITMAP_MODE_X2APIC_APICV	2
137 
138 #define KVM_VMX_TSC_MULTIPLIER_MAX     0xffffffffffffffffULL
139 
140 /* Guest_tsc -> host_tsc conversion requires 64-bit division.  */
141 static int __read_mostly cpu_preemption_timer_multi;
142 static bool __read_mostly enable_preemption_timer = 1;
143 #ifdef CONFIG_X86_64
144 module_param_named(preemption_timer, enable_preemption_timer, bool, S_IRUGO);
145 #endif
146 
147 extern bool __read_mostly allow_smaller_maxphyaddr;
148 module_param(allow_smaller_maxphyaddr, bool, S_IRUGO);
149 
150 #define KVM_VM_CR0_ALWAYS_OFF (X86_CR0_NW | X86_CR0_CD)
151 #define KVM_VM_CR0_ALWAYS_ON_UNRESTRICTED_GUEST X86_CR0_NE
152 #define KVM_VM_CR0_ALWAYS_ON				\
153 	(KVM_VM_CR0_ALWAYS_ON_UNRESTRICTED_GUEST | X86_CR0_PG | X86_CR0_PE)
154 
155 #define KVM_VM_CR4_ALWAYS_ON_UNRESTRICTED_GUEST X86_CR4_VMXE
156 #define KVM_PMODE_VM_CR4_ALWAYS_ON (X86_CR4_PAE | X86_CR4_VMXE)
157 #define KVM_RMODE_VM_CR4_ALWAYS_ON (X86_CR4_VME | X86_CR4_PAE | X86_CR4_VMXE)
158 
159 #define RMODE_GUEST_OWNED_EFLAGS_BITS (~(X86_EFLAGS_IOPL | X86_EFLAGS_VM))
160 
161 #define MSR_IA32_RTIT_STATUS_MASK (~(RTIT_STATUS_FILTEREN | \
162 	RTIT_STATUS_CONTEXTEN | RTIT_STATUS_TRIGGEREN | \
163 	RTIT_STATUS_ERROR | RTIT_STATUS_STOPPED | \
164 	RTIT_STATUS_BYTECNT))
165 
166 /*
167  * List of MSRs that can be directly passed to the guest.
168  * In addition to these x2apic, PT and LBR MSRs are handled specially.
169  */
170 static u32 vmx_possible_passthrough_msrs[MAX_POSSIBLE_PASSTHROUGH_MSRS] = {
171 	MSR_IA32_SPEC_CTRL,
172 	MSR_IA32_PRED_CMD,
173 	MSR_IA32_FLUSH_CMD,
174 	MSR_IA32_TSC,
175 #ifdef CONFIG_X86_64
176 	MSR_FS_BASE,
177 	MSR_GS_BASE,
178 	MSR_KERNEL_GS_BASE,
179 	MSR_IA32_XFD,
180 	MSR_IA32_XFD_ERR,
181 #endif
182 	MSR_IA32_SYSENTER_CS,
183 	MSR_IA32_SYSENTER_ESP,
184 	MSR_IA32_SYSENTER_EIP,
185 	MSR_CORE_C1_RES,
186 	MSR_CORE_C3_RESIDENCY,
187 	MSR_CORE_C6_RESIDENCY,
188 	MSR_CORE_C7_RESIDENCY,
189 };
190 
191 /*
192  * These 2 parameters are used to config the controls for Pause-Loop Exiting:
193  * ple_gap:    upper bound on the amount of time between two successive
194  *             executions of PAUSE in a loop. Also indicate if ple enabled.
195  *             According to test, this time is usually smaller than 128 cycles.
196  * ple_window: upper bound on the amount of time a guest is allowed to execute
197  *             in a PAUSE loop. Tests indicate that most spinlocks are held for
198  *             less than 2^12 cycles
199  * Time is measured based on a counter that runs at the same rate as the TSC,
200  * refer SDM volume 3b section 21.6.13 & 22.1.3.
201  */
202 static unsigned int ple_gap = KVM_DEFAULT_PLE_GAP;
203 module_param(ple_gap, uint, 0444);
204 
205 static unsigned int ple_window = KVM_VMX_DEFAULT_PLE_WINDOW;
206 module_param(ple_window, uint, 0444);
207 
208 /* Default doubles per-vcpu window every exit. */
209 static unsigned int ple_window_grow = KVM_DEFAULT_PLE_WINDOW_GROW;
210 module_param(ple_window_grow, uint, 0444);
211 
212 /* Default resets per-vcpu window every exit to ple_window. */
213 static unsigned int ple_window_shrink = KVM_DEFAULT_PLE_WINDOW_SHRINK;
214 module_param(ple_window_shrink, uint, 0444);
215 
216 /* Default is to compute the maximum so we can never overflow. */
217 static unsigned int ple_window_max        = KVM_VMX_DEFAULT_PLE_WINDOW_MAX;
218 module_param(ple_window_max, uint, 0444);
219 
220 /* Default is SYSTEM mode, 1 for host-guest mode (which is BROKEN) */
221 int __read_mostly pt_mode = PT_MODE_SYSTEM;
222 #ifdef CONFIG_BROKEN
223 module_param(pt_mode, int, S_IRUGO);
224 #endif
225 
226 struct x86_pmu_lbr __ro_after_init vmx_lbr_caps;
227 
228 static DEFINE_STATIC_KEY_FALSE(vmx_l1d_should_flush);
229 static DEFINE_STATIC_KEY_FALSE(vmx_l1d_flush_cond);
230 static DEFINE_MUTEX(vmx_l1d_flush_mutex);
231 
232 /* Storage for pre module init parameter parsing */
233 static enum vmx_l1d_flush_state __read_mostly vmentry_l1d_flush_param = VMENTER_L1D_FLUSH_AUTO;
234 
235 static const struct {
236 	const char *option;
237 	bool for_parse;
238 } vmentry_l1d_param[] = {
239 	[VMENTER_L1D_FLUSH_AUTO]	 = {"auto", true},
240 	[VMENTER_L1D_FLUSH_NEVER]	 = {"never", true},
241 	[VMENTER_L1D_FLUSH_COND]	 = {"cond", true},
242 	[VMENTER_L1D_FLUSH_ALWAYS]	 = {"always", true},
243 	[VMENTER_L1D_FLUSH_EPT_DISABLED] = {"EPT disabled", false},
244 	[VMENTER_L1D_FLUSH_NOT_REQUIRED] = {"not required", false},
245 };
246 
247 #define L1D_CACHE_ORDER 4
248 static void *vmx_l1d_flush_pages;
249 
250 static int vmx_setup_l1d_flush(enum vmx_l1d_flush_state l1tf)
251 {
252 	struct page *page;
253 	unsigned int i;
254 
255 	if (!boot_cpu_has_bug(X86_BUG_L1TF)) {
256 		l1tf_vmx_mitigation = VMENTER_L1D_FLUSH_NOT_REQUIRED;
257 		return 0;
258 	}
259 
260 	if (!enable_ept) {
261 		l1tf_vmx_mitigation = VMENTER_L1D_FLUSH_EPT_DISABLED;
262 		return 0;
263 	}
264 
265 	if (kvm_host.arch_capabilities & ARCH_CAP_SKIP_VMENTRY_L1DFLUSH) {
266 		l1tf_vmx_mitigation = VMENTER_L1D_FLUSH_NOT_REQUIRED;
267 		return 0;
268 	}
269 
270 	/* If set to auto use the default l1tf mitigation method */
271 	if (l1tf == VMENTER_L1D_FLUSH_AUTO) {
272 		switch (l1tf_mitigation) {
273 		case L1TF_MITIGATION_OFF:
274 			l1tf = VMENTER_L1D_FLUSH_NEVER;
275 			break;
276 		case L1TF_MITIGATION_FLUSH_NOWARN:
277 		case L1TF_MITIGATION_FLUSH:
278 		case L1TF_MITIGATION_FLUSH_NOSMT:
279 			l1tf = VMENTER_L1D_FLUSH_COND;
280 			break;
281 		case L1TF_MITIGATION_FULL:
282 		case L1TF_MITIGATION_FULL_FORCE:
283 			l1tf = VMENTER_L1D_FLUSH_ALWAYS;
284 			break;
285 		}
286 	} else if (l1tf_mitigation == L1TF_MITIGATION_FULL_FORCE) {
287 		l1tf = VMENTER_L1D_FLUSH_ALWAYS;
288 	}
289 
290 	if (l1tf != VMENTER_L1D_FLUSH_NEVER && !vmx_l1d_flush_pages &&
291 	    !boot_cpu_has(X86_FEATURE_FLUSH_L1D)) {
292 		/*
293 		 * This allocation for vmx_l1d_flush_pages is not tied to a VM
294 		 * lifetime and so should not be charged to a memcg.
295 		 */
296 		page = alloc_pages(GFP_KERNEL, L1D_CACHE_ORDER);
297 		if (!page)
298 			return -ENOMEM;
299 		vmx_l1d_flush_pages = page_address(page);
300 
301 		/*
302 		 * Initialize each page with a different pattern in
303 		 * order to protect against KSM in the nested
304 		 * virtualization case.
305 		 */
306 		for (i = 0; i < 1u << L1D_CACHE_ORDER; ++i) {
307 			memset(vmx_l1d_flush_pages + i * PAGE_SIZE, i + 1,
308 			       PAGE_SIZE);
309 		}
310 	}
311 
312 	l1tf_vmx_mitigation = l1tf;
313 
314 	if (l1tf != VMENTER_L1D_FLUSH_NEVER)
315 		static_branch_enable(&vmx_l1d_should_flush);
316 	else
317 		static_branch_disable(&vmx_l1d_should_flush);
318 
319 	if (l1tf == VMENTER_L1D_FLUSH_COND)
320 		static_branch_enable(&vmx_l1d_flush_cond);
321 	else
322 		static_branch_disable(&vmx_l1d_flush_cond);
323 	return 0;
324 }
325 
326 static int vmentry_l1d_flush_parse(const char *s)
327 {
328 	unsigned int i;
329 
330 	if (s) {
331 		for (i = 0; i < ARRAY_SIZE(vmentry_l1d_param); i++) {
332 			if (vmentry_l1d_param[i].for_parse &&
333 			    sysfs_streq(s, vmentry_l1d_param[i].option))
334 				return i;
335 		}
336 	}
337 	return -EINVAL;
338 }
339 
340 static int vmentry_l1d_flush_set(const char *s, const struct kernel_param *kp)
341 {
342 	int l1tf, ret;
343 
344 	l1tf = vmentry_l1d_flush_parse(s);
345 	if (l1tf < 0)
346 		return l1tf;
347 
348 	if (!boot_cpu_has(X86_BUG_L1TF))
349 		return 0;
350 
351 	/*
352 	 * Has vmx_init() run already? If not then this is the pre init
353 	 * parameter parsing. In that case just store the value and let
354 	 * vmx_init() do the proper setup after enable_ept has been
355 	 * established.
356 	 */
357 	if (l1tf_vmx_mitigation == VMENTER_L1D_FLUSH_AUTO) {
358 		vmentry_l1d_flush_param = l1tf;
359 		return 0;
360 	}
361 
362 	mutex_lock(&vmx_l1d_flush_mutex);
363 	ret = vmx_setup_l1d_flush(l1tf);
364 	mutex_unlock(&vmx_l1d_flush_mutex);
365 	return ret;
366 }
367 
368 static int vmentry_l1d_flush_get(char *s, const struct kernel_param *kp)
369 {
370 	if (WARN_ON_ONCE(l1tf_vmx_mitigation >= ARRAY_SIZE(vmentry_l1d_param)))
371 		return sysfs_emit(s, "???\n");
372 
373 	return sysfs_emit(s, "%s\n", vmentry_l1d_param[l1tf_vmx_mitigation].option);
374 }
375 
376 static __always_inline void vmx_disable_fb_clear(struct vcpu_vmx *vmx)
377 {
378 	u64 msr;
379 
380 	if (!vmx->disable_fb_clear)
381 		return;
382 
383 	msr = __rdmsr(MSR_IA32_MCU_OPT_CTRL);
384 	msr |= FB_CLEAR_DIS;
385 	native_wrmsrl(MSR_IA32_MCU_OPT_CTRL, msr);
386 	/* Cache the MSR value to avoid reading it later */
387 	vmx->msr_ia32_mcu_opt_ctrl = msr;
388 }
389 
390 static __always_inline void vmx_enable_fb_clear(struct vcpu_vmx *vmx)
391 {
392 	if (!vmx->disable_fb_clear)
393 		return;
394 
395 	vmx->msr_ia32_mcu_opt_ctrl &= ~FB_CLEAR_DIS;
396 	native_wrmsrl(MSR_IA32_MCU_OPT_CTRL, vmx->msr_ia32_mcu_opt_ctrl);
397 }
398 
399 static void vmx_update_fb_clear_dis(struct kvm_vcpu *vcpu, struct vcpu_vmx *vmx)
400 {
401 	/*
402 	 * Disable VERW's behavior of clearing CPU buffers for the guest if the
403 	 * CPU isn't affected by MDS/TAA, and the host hasn't forcefully enabled
404 	 * the mitigation. Disabling the clearing behavior provides a
405 	 * performance boost for guests that aren't aware that manually clearing
406 	 * CPU buffers is unnecessary, at the cost of MSR accesses on VM-Entry
407 	 * and VM-Exit.
408 	 */
409 	vmx->disable_fb_clear = !cpu_feature_enabled(X86_FEATURE_CLEAR_CPU_BUF) &&
410 				(kvm_host.arch_capabilities & ARCH_CAP_FB_CLEAR_CTRL) &&
411 				!boot_cpu_has_bug(X86_BUG_MDS) &&
412 				!boot_cpu_has_bug(X86_BUG_TAA);
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)
487 {
488 	vmx_insn_failed("invept failed: ext=0x%lx eptp=%llx\n", ext, eptp);
489 }
490 
491 static DEFINE_PER_CPU(struct vmcs *, vmxarea);
492 DEFINE_PER_CPU(struct vmcs *, current_vmcs);
493 /*
494  * We maintain a per-CPU linked-list of VMCS loaded on that CPU. This is needed
495  * when a CPU is brought down, and we need to VMCLEAR all VMCSs loaded on it.
496  */
497 static DEFINE_PER_CPU(struct list_head, loaded_vmcss_on_cpu);
498 
499 static DECLARE_BITMAP(vmx_vpid_bitmap, VMX_NR_VPIDS);
500 static DEFINE_SPINLOCK(vmx_vpid_lock);
501 
502 struct vmcs_config vmcs_config __ro_after_init;
503 struct vmx_capability vmx_capability __ro_after_init;
504 
505 #define VMX_SEGMENT_FIELD(seg)					\
506 	[VCPU_SREG_##seg] = {                                   \
507 		.selector = GUEST_##seg##_SELECTOR,		\
508 		.base = GUEST_##seg##_BASE,		   	\
509 		.limit = GUEST_##seg##_LIMIT,		   	\
510 		.ar_bytes = GUEST_##seg##_AR_BYTES,	   	\
511 	}
512 
513 static const struct kvm_vmx_segment_field {
514 	unsigned selector;
515 	unsigned base;
516 	unsigned limit;
517 	unsigned ar_bytes;
518 } kvm_vmx_segment_fields[] = {
519 	VMX_SEGMENT_FIELD(CS),
520 	VMX_SEGMENT_FIELD(DS),
521 	VMX_SEGMENT_FIELD(ES),
522 	VMX_SEGMENT_FIELD(FS),
523 	VMX_SEGMENT_FIELD(GS),
524 	VMX_SEGMENT_FIELD(SS),
525 	VMX_SEGMENT_FIELD(TR),
526 	VMX_SEGMENT_FIELD(LDTR),
527 };
528 
529 
530 static unsigned long host_idt_base;
531 
532 #if IS_ENABLED(CONFIG_HYPERV)
533 static bool __read_mostly enlightened_vmcs = true;
534 module_param(enlightened_vmcs, bool, 0444);
535 
536 static int hv_enable_l2_tlb_flush(struct kvm_vcpu *vcpu)
537 {
538 	struct hv_enlightened_vmcs *evmcs;
539 	hpa_t partition_assist_page = hv_get_partition_assist_page(vcpu);
540 
541 	if (partition_assist_page == INVALID_PAGE)
542 		return -ENOMEM;
543 
544 	evmcs = (struct hv_enlightened_vmcs *)to_vmx(vcpu)->loaded_vmcs->vmcs;
545 
546 	evmcs->partition_assist_page = partition_assist_page;
547 	evmcs->hv_vm_id = (unsigned long)vcpu->kvm;
548 	evmcs->hv_enlightenments_control.nested_flush_hypercall = 1;
549 
550 	return 0;
551 }
552 
553 static __init void hv_init_evmcs(void)
554 {
555 	int cpu;
556 
557 	if (!enlightened_vmcs)
558 		return;
559 
560 	/*
561 	 * Enlightened VMCS usage should be recommended and the host needs
562 	 * to support eVMCS v1 or above.
563 	 */
564 	if (ms_hyperv.hints & HV_X64_ENLIGHTENED_VMCS_RECOMMENDED &&
565 	    (ms_hyperv.nested_features & HV_X64_ENLIGHTENED_VMCS_VERSION) >=
566 	     KVM_EVMCS_VERSION) {
567 
568 		/* Check that we have assist pages on all online CPUs */
569 		for_each_online_cpu(cpu) {
570 			if (!hv_get_vp_assist_page(cpu)) {
571 				enlightened_vmcs = false;
572 				break;
573 			}
574 		}
575 
576 		if (enlightened_vmcs) {
577 			pr_info("Using Hyper-V Enlightened VMCS\n");
578 			static_branch_enable(&__kvm_is_using_evmcs);
579 		}
580 
581 		if (ms_hyperv.nested_features & HV_X64_NESTED_DIRECT_FLUSH)
582 			vt_x86_ops.enable_l2_tlb_flush
583 				= hv_enable_l2_tlb_flush;
584 	} else {
585 		enlightened_vmcs = false;
586 	}
587 }
588 
589 static void hv_reset_evmcs(void)
590 {
591 	struct hv_vp_assist_page *vp_ap;
592 
593 	if (!kvm_is_using_evmcs())
594 		return;
595 
596 	/*
597 	 * KVM should enable eVMCS if and only if all CPUs have a VP assist
598 	 * page, and should reject CPU onlining if eVMCS is enabled the CPU
599 	 * doesn't have a VP assist page allocated.
600 	 */
601 	vp_ap = hv_get_vp_assist_page(smp_processor_id());
602 	if (WARN_ON_ONCE(!vp_ap))
603 		return;
604 
605 	/*
606 	 * Reset everything to support using non-enlightened VMCS access later
607 	 * (e.g. when we reload the module with enlightened_vmcs=0)
608 	 */
609 	vp_ap->nested_control.features.directhypercall = 0;
610 	vp_ap->current_nested_vmcs = 0;
611 	vp_ap->enlighten_vmentry = 0;
612 }
613 
614 #else /* IS_ENABLED(CONFIG_HYPERV) */
615 static void hv_init_evmcs(void) {}
616 static void hv_reset_evmcs(void) {}
617 #endif /* IS_ENABLED(CONFIG_HYPERV) */
618 
619 /*
620  * Comment's format: document - errata name - stepping - processor name.
621  * Refer from
622  * https://www.virtualbox.org/svn/vbox/trunk/src/VBox/VMM/VMMR0/HMR0.cpp
623  */
624 static u32 vmx_preemption_cpu_tfms[] = {
625 /* 323344.pdf - BA86   - D0 - Xeon 7500 Series */
626 0x000206E6,
627 /* 323056.pdf - AAX65  - C2 - Xeon L3406 */
628 /* 322814.pdf - AAT59  - C2 - i7-600, i5-500, i5-400 and i3-300 Mobile */
629 /* 322911.pdf - AAU65  - C2 - i5-600, i3-500 Desktop and Pentium G6950 */
630 0x00020652,
631 /* 322911.pdf - AAU65  - K0 - i5-600, i3-500 Desktop and Pentium G6950 */
632 0x00020655,
633 /* 322373.pdf - AAO95  - B1 - Xeon 3400 Series */
634 /* 322166.pdf - AAN92  - B1 - i7-800 and i5-700 Desktop */
635 /*
636  * 320767.pdf - AAP86  - B1 -
637  * i7-900 Mobile Extreme, i7-800 and i7-700 Mobile
638  */
639 0x000106E5,
640 /* 321333.pdf - AAM126 - C0 - Xeon 3500 */
641 0x000106A0,
642 /* 321333.pdf - AAM126 - C1 - Xeon 3500 */
643 0x000106A1,
644 /* 320836.pdf - AAJ124 - C0 - i7-900 Desktop Extreme and i7-900 Desktop */
645 0x000106A4,
646  /* 321333.pdf - AAM126 - D0 - Xeon 3500 */
647  /* 321324.pdf - AAK139 - D0 - Xeon 5500 */
648  /* 320836.pdf - AAJ124 - D0 - i7-900 Extreme and i7-900 Desktop */
649 0x000106A5,
650  /* Xeon E3-1220 V2 */
651 0x000306A8,
652 };
653 
654 static inline bool cpu_has_broken_vmx_preemption_timer(void)
655 {
656 	u32 eax = cpuid_eax(0x00000001), i;
657 
658 	/* Clear the reserved bits */
659 	eax &= ~(0x3U << 14 | 0xfU << 28);
660 	for (i = 0; i < ARRAY_SIZE(vmx_preemption_cpu_tfms); i++)
661 		if (eax == vmx_preemption_cpu_tfms[i])
662 			return true;
663 
664 	return false;
665 }
666 
667 static inline bool cpu_need_virtualize_apic_accesses(struct kvm_vcpu *vcpu)
668 {
669 	return flexpriority_enabled && lapic_in_kernel(vcpu);
670 }
671 
672 static int vmx_get_passthrough_msr_slot(u32 msr)
673 {
674 	int i;
675 
676 	switch (msr) {
677 	case 0x800 ... 0x8ff:
678 		/* x2APIC MSRs. These are handled in vmx_update_msr_bitmap_x2apic() */
679 		return -ENOENT;
680 	case MSR_IA32_RTIT_STATUS:
681 	case MSR_IA32_RTIT_OUTPUT_BASE:
682 	case MSR_IA32_RTIT_OUTPUT_MASK:
683 	case MSR_IA32_RTIT_CR3_MATCH:
684 	case MSR_IA32_RTIT_ADDR0_A ... MSR_IA32_RTIT_ADDR3_B:
685 		/* PT MSRs. These are handled in pt_update_intercept_for_msr() */
686 	case MSR_LBR_SELECT:
687 	case MSR_LBR_TOS:
688 	case MSR_LBR_INFO_0 ... MSR_LBR_INFO_0 + 31:
689 	case MSR_LBR_NHM_FROM ... MSR_LBR_NHM_FROM + 31:
690 	case MSR_LBR_NHM_TO ... MSR_LBR_NHM_TO + 31:
691 	case MSR_LBR_CORE_FROM ... MSR_LBR_CORE_FROM + 8:
692 	case MSR_LBR_CORE_TO ... MSR_LBR_CORE_TO + 8:
693 		/* LBR MSRs. These are handled in vmx_update_intercept_for_lbr_msrs() */
694 		return -ENOENT;
695 	}
696 
697 	for (i = 0; i < ARRAY_SIZE(vmx_possible_passthrough_msrs); i++) {
698 		if (vmx_possible_passthrough_msrs[i] == msr)
699 			return i;
700 	}
701 
702 	WARN(1, "Invalid MSR %x, please adapt vmx_possible_passthrough_msrs[]", msr);
703 	return -ENOENT;
704 }
705 
706 struct vmx_uret_msr *vmx_find_uret_msr(struct vcpu_vmx *vmx, u32 msr)
707 {
708 	int i;
709 
710 	i = kvm_find_user_return_msr(msr);
711 	if (i >= 0)
712 		return &vmx->guest_uret_msrs[i];
713 	return NULL;
714 }
715 
716 static int vmx_set_guest_uret_msr(struct vcpu_vmx *vmx,
717 				  struct vmx_uret_msr *msr, u64 data)
718 {
719 	unsigned int slot = msr - vmx->guest_uret_msrs;
720 	int ret = 0;
721 
722 	if (msr->load_into_hardware) {
723 		preempt_disable();
724 		ret = kvm_set_user_return_msr(slot, data, msr->mask);
725 		preempt_enable();
726 	}
727 	if (!ret)
728 		msr->data = data;
729 	return ret;
730 }
731 
732 /*
733  * Disable VMX and clear CR4.VMXE (even if VMXOFF faults)
734  *
735  * Note, VMXOFF causes a #UD if the CPU is !post-VMXON, but it's impossible to
736  * atomically track post-VMXON state, e.g. this may be called in NMI context.
737  * Eat all faults as all other faults on VMXOFF faults are mode related, i.e.
738  * faults are guaranteed to be due to the !post-VMXON check unless the CPU is
739  * magically in RM, VM86, compat mode, or at CPL>0.
740  */
741 static int kvm_cpu_vmxoff(void)
742 {
743 	asm goto("1: vmxoff\n\t"
744 			  _ASM_EXTABLE(1b, %l[fault])
745 			  ::: "cc", "memory" : fault);
746 
747 	cr4_clear_bits(X86_CR4_VMXE);
748 	return 0;
749 
750 fault:
751 	cr4_clear_bits(X86_CR4_VMXE);
752 	return -EIO;
753 }
754 
755 void vmx_emergency_disable_virtualization_cpu(void)
756 {
757 	int cpu = raw_smp_processor_id();
758 	struct loaded_vmcs *v;
759 
760 	kvm_rebooting = true;
761 
762 	/*
763 	 * Note, CR4.VMXE can be _cleared_ in NMI context, but it can only be
764 	 * set in task context.  If this races with VMX is disabled by an NMI,
765 	 * VMCLEAR and VMXOFF may #UD, but KVM will eat those faults due to
766 	 * kvm_rebooting set.
767 	 */
768 	if (!(__read_cr4() & X86_CR4_VMXE))
769 		return;
770 
771 	list_for_each_entry(v, &per_cpu(loaded_vmcss_on_cpu, cpu),
772 			    loaded_vmcss_on_cpu_link)
773 		vmcs_clear(v->vmcs);
774 
775 	kvm_cpu_vmxoff();
776 }
777 
778 static void __loaded_vmcs_clear(void *arg)
779 {
780 	struct loaded_vmcs *loaded_vmcs = arg;
781 	int cpu = raw_smp_processor_id();
782 
783 	if (loaded_vmcs->cpu != cpu)
784 		return; /* vcpu migration can race with cpu offline */
785 	if (per_cpu(current_vmcs, cpu) == loaded_vmcs->vmcs)
786 		per_cpu(current_vmcs, cpu) = NULL;
787 
788 	vmcs_clear(loaded_vmcs->vmcs);
789 	if (loaded_vmcs->shadow_vmcs && loaded_vmcs->launched)
790 		vmcs_clear(loaded_vmcs->shadow_vmcs);
791 
792 	list_del(&loaded_vmcs->loaded_vmcss_on_cpu_link);
793 
794 	/*
795 	 * Ensure all writes to loaded_vmcs, including deleting it from its
796 	 * current percpu list, complete before setting loaded_vmcs->cpu to
797 	 * -1, otherwise a different cpu can see loaded_vmcs->cpu == -1 first
798 	 * and add loaded_vmcs to its percpu list before it's deleted from this
799 	 * cpu's list. Pairs with the smp_rmb() in vmx_vcpu_load_vmcs().
800 	 */
801 	smp_wmb();
802 
803 	loaded_vmcs->cpu = -1;
804 	loaded_vmcs->launched = 0;
805 }
806 
807 void loaded_vmcs_clear(struct loaded_vmcs *loaded_vmcs)
808 {
809 	int cpu = loaded_vmcs->cpu;
810 
811 	if (cpu != -1)
812 		smp_call_function_single(cpu,
813 			 __loaded_vmcs_clear, loaded_vmcs, 1);
814 }
815 
816 static bool vmx_segment_cache_test_set(struct vcpu_vmx *vmx, unsigned seg,
817 				       unsigned field)
818 {
819 	bool ret;
820 	u32 mask = 1 << (seg * SEG_FIELD_NR + field);
821 
822 	if (!kvm_register_is_available(&vmx->vcpu, VCPU_EXREG_SEGMENTS)) {
823 		kvm_register_mark_available(&vmx->vcpu, VCPU_EXREG_SEGMENTS);
824 		vmx->segment_cache.bitmask = 0;
825 	}
826 	ret = vmx->segment_cache.bitmask & mask;
827 	vmx->segment_cache.bitmask |= mask;
828 	return ret;
829 }
830 
831 static u16 vmx_read_guest_seg_selector(struct vcpu_vmx *vmx, unsigned seg)
832 {
833 	u16 *p = &vmx->segment_cache.seg[seg].selector;
834 
835 	if (!vmx_segment_cache_test_set(vmx, seg, SEG_FIELD_SEL))
836 		*p = vmcs_read16(kvm_vmx_segment_fields[seg].selector);
837 	return *p;
838 }
839 
840 static ulong vmx_read_guest_seg_base(struct vcpu_vmx *vmx, unsigned seg)
841 {
842 	ulong *p = &vmx->segment_cache.seg[seg].base;
843 
844 	if (!vmx_segment_cache_test_set(vmx, seg, SEG_FIELD_BASE))
845 		*p = vmcs_readl(kvm_vmx_segment_fields[seg].base);
846 	return *p;
847 }
848 
849 static u32 vmx_read_guest_seg_limit(struct vcpu_vmx *vmx, unsigned seg)
850 {
851 	u32 *p = &vmx->segment_cache.seg[seg].limit;
852 
853 	if (!vmx_segment_cache_test_set(vmx, seg, SEG_FIELD_LIMIT))
854 		*p = vmcs_read32(kvm_vmx_segment_fields[seg].limit);
855 	return *p;
856 }
857 
858 static u32 vmx_read_guest_seg_ar(struct vcpu_vmx *vmx, unsigned seg)
859 {
860 	u32 *p = &vmx->segment_cache.seg[seg].ar;
861 
862 	if (!vmx_segment_cache_test_set(vmx, seg, SEG_FIELD_AR))
863 		*p = vmcs_read32(kvm_vmx_segment_fields[seg].ar_bytes);
864 	return *p;
865 }
866 
867 void vmx_update_exception_bitmap(struct kvm_vcpu *vcpu)
868 {
869 	u32 eb;
870 
871 	eb = (1u << PF_VECTOR) | (1u << UD_VECTOR) | (1u << MC_VECTOR) |
872 	     (1u << DB_VECTOR) | (1u << AC_VECTOR);
873 	/*
874 	 * #VE isn't used for VMX.  To test against unexpected changes
875 	 * related to #VE for VMX, intercept unexpected #VE and warn on it.
876 	 */
877 	if (IS_ENABLED(CONFIG_KVM_INTEL_PROVE_VE))
878 		eb |= 1u << VE_VECTOR;
879 	/*
880 	 * Guest access to VMware backdoor ports could legitimately
881 	 * trigger #GP because of TSS I/O permission bitmap.
882 	 * We intercept those #GP and allow access to them anyway
883 	 * as VMware does.
884 	 */
885 	if (enable_vmware_backdoor)
886 		eb |= (1u << GP_VECTOR);
887 	if ((vcpu->guest_debug &
888 	     (KVM_GUESTDBG_ENABLE | KVM_GUESTDBG_USE_SW_BP)) ==
889 	    (KVM_GUESTDBG_ENABLE | KVM_GUESTDBG_USE_SW_BP))
890 		eb |= 1u << BP_VECTOR;
891 	if (to_vmx(vcpu)->rmode.vm86_active)
892 		eb = ~0;
893 	if (!vmx_need_pf_intercept(vcpu))
894 		eb &= ~(1u << PF_VECTOR);
895 
896 	/* When we are running a nested L2 guest and L1 specified for it a
897 	 * certain exception bitmap, we must trap the same exceptions and pass
898 	 * them to L1. When running L2, we will only handle the exceptions
899 	 * specified above if L1 did not want them.
900 	 */
901 	if (is_guest_mode(vcpu))
902 		eb |= get_vmcs12(vcpu)->exception_bitmap;
903 	else {
904 		int mask = 0, match = 0;
905 
906 		if (enable_ept && (eb & (1u << PF_VECTOR))) {
907 			/*
908 			 * If EPT is enabled, #PF is currently only intercepted
909 			 * if MAXPHYADDR is smaller on the guest than on the
910 			 * host.  In that case we only care about present,
911 			 * non-reserved faults.  For vmcs02, however, PFEC_MASK
912 			 * and PFEC_MATCH are set in prepare_vmcs02_rare.
913 			 */
914 			mask = PFERR_PRESENT_MASK | PFERR_RSVD_MASK;
915 			match = PFERR_PRESENT_MASK;
916 		}
917 		vmcs_write32(PAGE_FAULT_ERROR_CODE_MASK, mask);
918 		vmcs_write32(PAGE_FAULT_ERROR_CODE_MATCH, match);
919 	}
920 
921 	/*
922 	 * Disabling xfd interception indicates that dynamic xfeatures
923 	 * might be used in the guest. Always trap #NM in this case
924 	 * to save guest xfd_err timely.
925 	 */
926 	if (vcpu->arch.xfd_no_write_intercept)
927 		eb |= (1u << NM_VECTOR);
928 
929 	vmcs_write32(EXCEPTION_BITMAP, eb);
930 }
931 
932 /*
933  * Check if MSR is intercepted for currently loaded MSR bitmap.
934  */
935 static bool msr_write_intercepted(struct vcpu_vmx *vmx, u32 msr)
936 {
937 	if (!(exec_controls_get(vmx) & CPU_BASED_USE_MSR_BITMAPS))
938 		return true;
939 
940 	return vmx_test_msr_bitmap_write(vmx->loaded_vmcs->msr_bitmap, msr);
941 }
942 
943 unsigned int __vmx_vcpu_run_flags(struct vcpu_vmx *vmx)
944 {
945 	unsigned int flags = 0;
946 
947 	if (vmx->loaded_vmcs->launched)
948 		flags |= VMX_RUN_VMRESUME;
949 
950 	/*
951 	 * If writes to the SPEC_CTRL MSR aren't intercepted, the guest is free
952 	 * to change it directly without causing a vmexit.  In that case read
953 	 * it after vmexit and store it in vmx->spec_ctrl.
954 	 */
955 	if (!msr_write_intercepted(vmx, MSR_IA32_SPEC_CTRL))
956 		flags |= VMX_RUN_SAVE_SPEC_CTRL;
957 
958 	return flags;
959 }
960 
961 static __always_inline void clear_atomic_switch_msr_special(struct vcpu_vmx *vmx,
962 		unsigned long entry, unsigned long exit)
963 {
964 	vm_entry_controls_clearbit(vmx, entry);
965 	vm_exit_controls_clearbit(vmx, exit);
966 }
967 
968 int vmx_find_loadstore_msr_slot(struct vmx_msrs *m, u32 msr)
969 {
970 	unsigned int i;
971 
972 	for (i = 0; i < m->nr; ++i) {
973 		if (m->val[i].index == msr)
974 			return i;
975 	}
976 	return -ENOENT;
977 }
978 
979 static void clear_atomic_switch_msr(struct vcpu_vmx *vmx, unsigned msr)
980 {
981 	int i;
982 	struct msr_autoload *m = &vmx->msr_autoload;
983 
984 	switch (msr) {
985 	case MSR_EFER:
986 		if (cpu_has_load_ia32_efer()) {
987 			clear_atomic_switch_msr_special(vmx,
988 					VM_ENTRY_LOAD_IA32_EFER,
989 					VM_EXIT_LOAD_IA32_EFER);
990 			return;
991 		}
992 		break;
993 	case MSR_CORE_PERF_GLOBAL_CTRL:
994 		if (cpu_has_load_perf_global_ctrl()) {
995 			clear_atomic_switch_msr_special(vmx,
996 					VM_ENTRY_LOAD_IA32_PERF_GLOBAL_CTRL,
997 					VM_EXIT_LOAD_IA32_PERF_GLOBAL_CTRL);
998 			return;
999 		}
1000 		break;
1001 	}
1002 	i = vmx_find_loadstore_msr_slot(&m->guest, msr);
1003 	if (i < 0)
1004 		goto skip_guest;
1005 	--m->guest.nr;
1006 	m->guest.val[i] = m->guest.val[m->guest.nr];
1007 	vmcs_write32(VM_ENTRY_MSR_LOAD_COUNT, m->guest.nr);
1008 
1009 skip_guest:
1010 	i = vmx_find_loadstore_msr_slot(&m->host, msr);
1011 	if (i < 0)
1012 		return;
1013 
1014 	--m->host.nr;
1015 	m->host.val[i] = m->host.val[m->host.nr];
1016 	vmcs_write32(VM_EXIT_MSR_LOAD_COUNT, m->host.nr);
1017 }
1018 
1019 static __always_inline void add_atomic_switch_msr_special(struct vcpu_vmx *vmx,
1020 		unsigned long entry, unsigned long exit,
1021 		unsigned long guest_val_vmcs, unsigned long host_val_vmcs,
1022 		u64 guest_val, u64 host_val)
1023 {
1024 	vmcs_write64(guest_val_vmcs, guest_val);
1025 	if (host_val_vmcs != HOST_IA32_EFER)
1026 		vmcs_write64(host_val_vmcs, host_val);
1027 	vm_entry_controls_setbit(vmx, entry);
1028 	vm_exit_controls_setbit(vmx, exit);
1029 }
1030 
1031 static void add_atomic_switch_msr(struct vcpu_vmx *vmx, unsigned msr,
1032 				  u64 guest_val, u64 host_val, bool entry_only)
1033 {
1034 	int i, j = 0;
1035 	struct msr_autoload *m = &vmx->msr_autoload;
1036 
1037 	switch (msr) {
1038 	case MSR_EFER:
1039 		if (cpu_has_load_ia32_efer()) {
1040 			add_atomic_switch_msr_special(vmx,
1041 					VM_ENTRY_LOAD_IA32_EFER,
1042 					VM_EXIT_LOAD_IA32_EFER,
1043 					GUEST_IA32_EFER,
1044 					HOST_IA32_EFER,
1045 					guest_val, host_val);
1046 			return;
1047 		}
1048 		break;
1049 	case MSR_CORE_PERF_GLOBAL_CTRL:
1050 		if (cpu_has_load_perf_global_ctrl()) {
1051 			add_atomic_switch_msr_special(vmx,
1052 					VM_ENTRY_LOAD_IA32_PERF_GLOBAL_CTRL,
1053 					VM_EXIT_LOAD_IA32_PERF_GLOBAL_CTRL,
1054 					GUEST_IA32_PERF_GLOBAL_CTRL,
1055 					HOST_IA32_PERF_GLOBAL_CTRL,
1056 					guest_val, host_val);
1057 			return;
1058 		}
1059 		break;
1060 	case MSR_IA32_PEBS_ENABLE:
1061 		/* PEBS needs a quiescent period after being disabled (to write
1062 		 * a record).  Disabling PEBS through VMX MSR swapping doesn't
1063 		 * provide that period, so a CPU could write host's record into
1064 		 * guest's memory.
1065 		 */
1066 		wrmsrl(MSR_IA32_PEBS_ENABLE, 0);
1067 	}
1068 
1069 	i = vmx_find_loadstore_msr_slot(&m->guest, msr);
1070 	if (!entry_only)
1071 		j = vmx_find_loadstore_msr_slot(&m->host, msr);
1072 
1073 	if ((i < 0 && m->guest.nr == MAX_NR_LOADSTORE_MSRS) ||
1074 	    (j < 0 &&  m->host.nr == MAX_NR_LOADSTORE_MSRS)) {
1075 		printk_once(KERN_WARNING "Not enough msr switch entries. "
1076 				"Can't add msr %x\n", msr);
1077 		return;
1078 	}
1079 	if (i < 0) {
1080 		i = m->guest.nr++;
1081 		vmcs_write32(VM_ENTRY_MSR_LOAD_COUNT, m->guest.nr);
1082 	}
1083 	m->guest.val[i].index = msr;
1084 	m->guest.val[i].value = guest_val;
1085 
1086 	if (entry_only)
1087 		return;
1088 
1089 	if (j < 0) {
1090 		j = m->host.nr++;
1091 		vmcs_write32(VM_EXIT_MSR_LOAD_COUNT, m->host.nr);
1092 	}
1093 	m->host.val[j].index = msr;
1094 	m->host.val[j].value = host_val;
1095 }
1096 
1097 static bool update_transition_efer(struct vcpu_vmx *vmx)
1098 {
1099 	u64 guest_efer = vmx->vcpu.arch.efer;
1100 	u64 ignore_bits = 0;
1101 	int i;
1102 
1103 	/* Shadow paging assumes NX to be available.  */
1104 	if (!enable_ept)
1105 		guest_efer |= EFER_NX;
1106 
1107 	/*
1108 	 * LMA and LME handled by hardware; SCE meaningless outside long mode.
1109 	 */
1110 	ignore_bits |= EFER_SCE;
1111 #ifdef CONFIG_X86_64
1112 	ignore_bits |= EFER_LMA | EFER_LME;
1113 	/* SCE is meaningful only in long mode on Intel */
1114 	if (guest_efer & EFER_LMA)
1115 		ignore_bits &= ~(u64)EFER_SCE;
1116 #endif
1117 
1118 	/*
1119 	 * On EPT, we can't emulate NX, so we must switch EFER atomically.
1120 	 * On CPUs that support "load IA32_EFER", always switch EFER
1121 	 * atomically, since it's faster than switching it manually.
1122 	 */
1123 	if (cpu_has_load_ia32_efer() ||
1124 	    (enable_ept && ((vmx->vcpu.arch.efer ^ kvm_host.efer) & EFER_NX))) {
1125 		if (!(guest_efer & EFER_LMA))
1126 			guest_efer &= ~EFER_LME;
1127 		if (guest_efer != kvm_host.efer)
1128 			add_atomic_switch_msr(vmx, MSR_EFER,
1129 					      guest_efer, kvm_host.efer, false);
1130 		else
1131 			clear_atomic_switch_msr(vmx, MSR_EFER);
1132 		return false;
1133 	}
1134 
1135 	i = kvm_find_user_return_msr(MSR_EFER);
1136 	if (i < 0)
1137 		return false;
1138 
1139 	clear_atomic_switch_msr(vmx, MSR_EFER);
1140 
1141 	guest_efer &= ~ignore_bits;
1142 	guest_efer |= kvm_host.efer & ignore_bits;
1143 
1144 	vmx->guest_uret_msrs[i].data = guest_efer;
1145 	vmx->guest_uret_msrs[i].mask = ~ignore_bits;
1146 
1147 	return true;
1148 }
1149 
1150 #ifdef CONFIG_X86_32
1151 /*
1152  * On 32-bit kernels, VM exits still load the FS and GS bases from the
1153  * VMCS rather than the segment table.  KVM uses this helper to figure
1154  * out the current bases to poke them into the VMCS before entry.
1155  */
1156 static unsigned long segment_base(u16 selector)
1157 {
1158 	struct desc_struct *table;
1159 	unsigned long v;
1160 
1161 	if (!(selector & ~SEGMENT_RPL_MASK))
1162 		return 0;
1163 
1164 	table = get_current_gdt_ro();
1165 
1166 	if ((selector & SEGMENT_TI_MASK) == SEGMENT_LDT) {
1167 		u16 ldt_selector = kvm_read_ldt();
1168 
1169 		if (!(ldt_selector & ~SEGMENT_RPL_MASK))
1170 			return 0;
1171 
1172 		table = (struct desc_struct *)segment_base(ldt_selector);
1173 	}
1174 	v = get_desc_base(&table[selector >> 3]);
1175 	return v;
1176 }
1177 #endif
1178 
1179 static inline bool pt_can_write_msr(struct vcpu_vmx *vmx)
1180 {
1181 	return vmx_pt_mode_is_host_guest() &&
1182 	       !(vmx->pt_desc.guest.ctl & RTIT_CTL_TRACEEN);
1183 }
1184 
1185 static inline bool pt_output_base_valid(struct kvm_vcpu *vcpu, u64 base)
1186 {
1187 	/* The base must be 128-byte aligned and a legal physical address. */
1188 	return kvm_vcpu_is_legal_aligned_gpa(vcpu, base, 128);
1189 }
1190 
1191 static inline void pt_load_msr(struct pt_ctx *ctx, u32 addr_range)
1192 {
1193 	u32 i;
1194 
1195 	wrmsrl(MSR_IA32_RTIT_STATUS, ctx->status);
1196 	wrmsrl(MSR_IA32_RTIT_OUTPUT_BASE, ctx->output_base);
1197 	wrmsrl(MSR_IA32_RTIT_OUTPUT_MASK, ctx->output_mask);
1198 	wrmsrl(MSR_IA32_RTIT_CR3_MATCH, ctx->cr3_match);
1199 	for (i = 0; i < addr_range; i++) {
1200 		wrmsrl(MSR_IA32_RTIT_ADDR0_A + i * 2, ctx->addr_a[i]);
1201 		wrmsrl(MSR_IA32_RTIT_ADDR0_B + i * 2, ctx->addr_b[i]);
1202 	}
1203 }
1204 
1205 static inline void pt_save_msr(struct pt_ctx *ctx, u32 addr_range)
1206 {
1207 	u32 i;
1208 
1209 	rdmsrl(MSR_IA32_RTIT_STATUS, ctx->status);
1210 	rdmsrl(MSR_IA32_RTIT_OUTPUT_BASE, ctx->output_base);
1211 	rdmsrl(MSR_IA32_RTIT_OUTPUT_MASK, ctx->output_mask);
1212 	rdmsrl(MSR_IA32_RTIT_CR3_MATCH, ctx->cr3_match);
1213 	for (i = 0; i < addr_range; i++) {
1214 		rdmsrl(MSR_IA32_RTIT_ADDR0_A + i * 2, ctx->addr_a[i]);
1215 		rdmsrl(MSR_IA32_RTIT_ADDR0_B + i * 2, ctx->addr_b[i]);
1216 	}
1217 }
1218 
1219 static void pt_guest_enter(struct vcpu_vmx *vmx)
1220 {
1221 	if (vmx_pt_mode_is_system())
1222 		return;
1223 
1224 	/*
1225 	 * GUEST_IA32_RTIT_CTL is already set in the VMCS.
1226 	 * Save host state before VM entry.
1227 	 */
1228 	rdmsrl(MSR_IA32_RTIT_CTL, vmx->pt_desc.host.ctl);
1229 	if (vmx->pt_desc.guest.ctl & RTIT_CTL_TRACEEN) {
1230 		wrmsrl(MSR_IA32_RTIT_CTL, 0);
1231 		pt_save_msr(&vmx->pt_desc.host, vmx->pt_desc.num_address_ranges);
1232 		pt_load_msr(&vmx->pt_desc.guest, vmx->pt_desc.num_address_ranges);
1233 	}
1234 }
1235 
1236 static void pt_guest_exit(struct vcpu_vmx *vmx)
1237 {
1238 	if (vmx_pt_mode_is_system())
1239 		return;
1240 
1241 	if (vmx->pt_desc.guest.ctl & RTIT_CTL_TRACEEN) {
1242 		pt_save_msr(&vmx->pt_desc.guest, vmx->pt_desc.num_address_ranges);
1243 		pt_load_msr(&vmx->pt_desc.host, vmx->pt_desc.num_address_ranges);
1244 	}
1245 
1246 	/*
1247 	 * KVM requires VM_EXIT_CLEAR_IA32_RTIT_CTL to expose PT to the guest,
1248 	 * i.e. RTIT_CTL is always cleared on VM-Exit.  Restore it if necessary.
1249 	 */
1250 	if (vmx->pt_desc.host.ctl)
1251 		wrmsrl(MSR_IA32_RTIT_CTL, vmx->pt_desc.host.ctl);
1252 }
1253 
1254 void vmx_set_host_fs_gs(struct vmcs_host_state *host, u16 fs_sel, u16 gs_sel,
1255 			unsigned long fs_base, unsigned long gs_base)
1256 {
1257 	if (unlikely(fs_sel != host->fs_sel)) {
1258 		if (!(fs_sel & 7))
1259 			vmcs_write16(HOST_FS_SELECTOR, fs_sel);
1260 		else
1261 			vmcs_write16(HOST_FS_SELECTOR, 0);
1262 		host->fs_sel = fs_sel;
1263 	}
1264 	if (unlikely(gs_sel != host->gs_sel)) {
1265 		if (!(gs_sel & 7))
1266 			vmcs_write16(HOST_GS_SELECTOR, gs_sel);
1267 		else
1268 			vmcs_write16(HOST_GS_SELECTOR, 0);
1269 		host->gs_sel = gs_sel;
1270 	}
1271 	if (unlikely(fs_base != host->fs_base)) {
1272 		vmcs_writel(HOST_FS_BASE, fs_base);
1273 		host->fs_base = fs_base;
1274 	}
1275 	if (unlikely(gs_base != host->gs_base)) {
1276 		vmcs_writel(HOST_GS_BASE, gs_base);
1277 		host->gs_base = gs_base;
1278 	}
1279 }
1280 
1281 void vmx_prepare_switch_to_guest(struct kvm_vcpu *vcpu)
1282 {
1283 	struct vcpu_vmx *vmx = to_vmx(vcpu);
1284 	struct vmcs_host_state *host_state;
1285 #ifdef CONFIG_X86_64
1286 	int cpu = raw_smp_processor_id();
1287 #endif
1288 	unsigned long fs_base, gs_base;
1289 	u16 fs_sel, gs_sel;
1290 	int i;
1291 
1292 	/*
1293 	 * Note that guest MSRs to be saved/restored can also be changed
1294 	 * when guest state is loaded. This happens when guest transitions
1295 	 * to/from long-mode by setting MSR_EFER.LMA.
1296 	 */
1297 	if (!vmx->guest_uret_msrs_loaded) {
1298 		vmx->guest_uret_msrs_loaded = true;
1299 		for (i = 0; i < kvm_nr_uret_msrs; ++i) {
1300 			if (!vmx->guest_uret_msrs[i].load_into_hardware)
1301 				continue;
1302 
1303 			kvm_set_user_return_msr(i,
1304 						vmx->guest_uret_msrs[i].data,
1305 						vmx->guest_uret_msrs[i].mask);
1306 		}
1307 	}
1308 
1309 	if (vmx->nested.need_vmcs12_to_shadow_sync)
1310 		nested_sync_vmcs12_to_shadow(vcpu);
1311 
1312 	if (vmx->guest_state_loaded)
1313 		return;
1314 
1315 	host_state = &vmx->loaded_vmcs->host_state;
1316 
1317 	/*
1318 	 * Set host fs and gs selectors.  Unfortunately, 22.2.3 does not
1319 	 * allow segment selectors with cpl > 0 or ti == 1.
1320 	 */
1321 	host_state->ldt_sel = kvm_read_ldt();
1322 
1323 #ifdef CONFIG_X86_64
1324 	savesegment(ds, host_state->ds_sel);
1325 	savesegment(es, host_state->es_sel);
1326 
1327 	gs_base = cpu_kernelmode_gs_base(cpu);
1328 	if (likely(is_64bit_mm(current->mm))) {
1329 		current_save_fsgs();
1330 		fs_sel = current->thread.fsindex;
1331 		gs_sel = current->thread.gsindex;
1332 		fs_base = current->thread.fsbase;
1333 		vmx->msr_host_kernel_gs_base = current->thread.gsbase;
1334 	} else {
1335 		savesegment(fs, fs_sel);
1336 		savesegment(gs, gs_sel);
1337 		fs_base = read_msr(MSR_FS_BASE);
1338 		vmx->msr_host_kernel_gs_base = read_msr(MSR_KERNEL_GS_BASE);
1339 	}
1340 
1341 	wrmsrl(MSR_KERNEL_GS_BASE, vmx->msr_guest_kernel_gs_base);
1342 #else
1343 	savesegment(fs, fs_sel);
1344 	savesegment(gs, gs_sel);
1345 	fs_base = segment_base(fs_sel);
1346 	gs_base = segment_base(gs_sel);
1347 #endif
1348 
1349 	vmx_set_host_fs_gs(host_state, fs_sel, gs_sel, fs_base, gs_base);
1350 	vmx->guest_state_loaded = true;
1351 }
1352 
1353 static void vmx_prepare_switch_to_host(struct vcpu_vmx *vmx)
1354 {
1355 	struct vmcs_host_state *host_state;
1356 
1357 	if (!vmx->guest_state_loaded)
1358 		return;
1359 
1360 	host_state = &vmx->loaded_vmcs->host_state;
1361 
1362 	++vmx->vcpu.stat.host_state_reload;
1363 
1364 #ifdef CONFIG_X86_64
1365 	rdmsrl(MSR_KERNEL_GS_BASE, vmx->msr_guest_kernel_gs_base);
1366 #endif
1367 	if (host_state->ldt_sel || (host_state->gs_sel & 7)) {
1368 		kvm_load_ldt(host_state->ldt_sel);
1369 #ifdef CONFIG_X86_64
1370 		load_gs_index(host_state->gs_sel);
1371 #else
1372 		loadsegment(gs, host_state->gs_sel);
1373 #endif
1374 	}
1375 	if (host_state->fs_sel & 7)
1376 		loadsegment(fs, host_state->fs_sel);
1377 #ifdef CONFIG_X86_64
1378 	if (unlikely(host_state->ds_sel | host_state->es_sel)) {
1379 		loadsegment(ds, host_state->ds_sel);
1380 		loadsegment(es, host_state->es_sel);
1381 	}
1382 #endif
1383 	invalidate_tss_limit();
1384 #ifdef CONFIG_X86_64
1385 	wrmsrl(MSR_KERNEL_GS_BASE, vmx->msr_host_kernel_gs_base);
1386 #endif
1387 	load_fixmap_gdt(raw_smp_processor_id());
1388 	vmx->guest_state_loaded = false;
1389 	vmx->guest_uret_msrs_loaded = false;
1390 }
1391 
1392 #ifdef CONFIG_X86_64
1393 static u64 vmx_read_guest_kernel_gs_base(struct vcpu_vmx *vmx)
1394 {
1395 	preempt_disable();
1396 	if (vmx->guest_state_loaded)
1397 		rdmsrl(MSR_KERNEL_GS_BASE, vmx->msr_guest_kernel_gs_base);
1398 	preempt_enable();
1399 	return vmx->msr_guest_kernel_gs_base;
1400 }
1401 
1402 static void vmx_write_guest_kernel_gs_base(struct vcpu_vmx *vmx, u64 data)
1403 {
1404 	preempt_disable();
1405 	if (vmx->guest_state_loaded)
1406 		wrmsrl(MSR_KERNEL_GS_BASE, data);
1407 	preempt_enable();
1408 	vmx->msr_guest_kernel_gs_base = data;
1409 }
1410 #endif
1411 
1412 static void grow_ple_window(struct kvm_vcpu *vcpu)
1413 {
1414 	struct vcpu_vmx *vmx = to_vmx(vcpu);
1415 	unsigned int old = vmx->ple_window;
1416 
1417 	vmx->ple_window = __grow_ple_window(old, ple_window,
1418 					    ple_window_grow,
1419 					    ple_window_max);
1420 
1421 	if (vmx->ple_window != old) {
1422 		vmx->ple_window_dirty = true;
1423 		trace_kvm_ple_window_update(vcpu->vcpu_id,
1424 					    vmx->ple_window, old);
1425 	}
1426 }
1427 
1428 static void shrink_ple_window(struct kvm_vcpu *vcpu)
1429 {
1430 	struct vcpu_vmx *vmx = to_vmx(vcpu);
1431 	unsigned int old = vmx->ple_window;
1432 
1433 	vmx->ple_window = __shrink_ple_window(old, ple_window,
1434 					      ple_window_shrink,
1435 					      ple_window);
1436 
1437 	if (vmx->ple_window != old) {
1438 		vmx->ple_window_dirty = true;
1439 		trace_kvm_ple_window_update(vcpu->vcpu_id,
1440 					    vmx->ple_window, old);
1441 	}
1442 }
1443 
1444 void vmx_vcpu_load_vmcs(struct kvm_vcpu *vcpu, int cpu,
1445 			struct loaded_vmcs *buddy)
1446 {
1447 	struct vcpu_vmx *vmx = to_vmx(vcpu);
1448 	bool already_loaded = vmx->loaded_vmcs->cpu == cpu;
1449 	struct vmcs *prev;
1450 
1451 	if (!already_loaded) {
1452 		loaded_vmcs_clear(vmx->loaded_vmcs);
1453 		local_irq_disable();
1454 
1455 		/*
1456 		 * Ensure loaded_vmcs->cpu is read before adding loaded_vmcs to
1457 		 * this cpu's percpu list, otherwise it may not yet be deleted
1458 		 * from its previous cpu's percpu list.  Pairs with the
1459 		 * smb_wmb() in __loaded_vmcs_clear().
1460 		 */
1461 		smp_rmb();
1462 
1463 		list_add(&vmx->loaded_vmcs->loaded_vmcss_on_cpu_link,
1464 			 &per_cpu(loaded_vmcss_on_cpu, cpu));
1465 		local_irq_enable();
1466 	}
1467 
1468 	prev = per_cpu(current_vmcs, cpu);
1469 	if (prev != vmx->loaded_vmcs->vmcs) {
1470 		per_cpu(current_vmcs, cpu) = vmx->loaded_vmcs->vmcs;
1471 		vmcs_load(vmx->loaded_vmcs->vmcs);
1472 
1473 		/*
1474 		 * No indirect branch prediction barrier needed when switching
1475 		 * the active VMCS within a vCPU, unless IBRS is advertised to
1476 		 * the vCPU.  To minimize the number of IBPBs executed, KVM
1477 		 * performs IBPB on nested VM-Exit (a single nested transition
1478 		 * may switch the active VMCS multiple times).
1479 		 */
1480 		if (!buddy || WARN_ON_ONCE(buddy->vmcs != prev))
1481 			indirect_branch_prediction_barrier();
1482 	}
1483 
1484 	if (!already_loaded) {
1485 		void *gdt = get_current_gdt_ro();
1486 
1487 		/*
1488 		 * Flush all EPTP/VPID contexts, the new pCPU may have stale
1489 		 * TLB entries from its previous association with the vCPU.
1490 		 */
1491 		kvm_make_request(KVM_REQ_TLB_FLUSH, vcpu);
1492 
1493 		/*
1494 		 * Linux uses per-cpu TSS and GDT, so set these when switching
1495 		 * processors.  See 22.2.4.
1496 		 */
1497 		vmcs_writel(HOST_TR_BASE,
1498 			    (unsigned long)&get_cpu_entry_area(cpu)->tss.x86_tss);
1499 		vmcs_writel(HOST_GDTR_BASE, (unsigned long)gdt);   /* 22.2.4 */
1500 
1501 		if (IS_ENABLED(CONFIG_IA32_EMULATION) || IS_ENABLED(CONFIG_X86_32)) {
1502 			/* 22.2.3 */
1503 			vmcs_writel(HOST_IA32_SYSENTER_ESP,
1504 				    (unsigned long)(cpu_entry_stack(cpu) + 1));
1505 		}
1506 
1507 		vmx->loaded_vmcs->cpu = cpu;
1508 	}
1509 }
1510 
1511 /*
1512  * Switches to specified vcpu, until a matching vcpu_put(), but assumes
1513  * vcpu mutex is already taken.
1514  */
1515 void vmx_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
1516 {
1517 	struct vcpu_vmx *vmx = to_vmx(vcpu);
1518 
1519 	if (vcpu->scheduled_out && !kvm_pause_in_guest(vcpu->kvm))
1520 		shrink_ple_window(vcpu);
1521 
1522 	vmx_vcpu_load_vmcs(vcpu, cpu, NULL);
1523 
1524 	vmx_vcpu_pi_load(vcpu, cpu);
1525 
1526 	vmx->host_debugctlmsr = get_debugctlmsr();
1527 }
1528 
1529 void vmx_vcpu_put(struct kvm_vcpu *vcpu)
1530 {
1531 	vmx_vcpu_pi_put(vcpu);
1532 
1533 	vmx_prepare_switch_to_host(to_vmx(vcpu));
1534 }
1535 
1536 bool vmx_emulation_required(struct kvm_vcpu *vcpu)
1537 {
1538 	return emulate_invalid_guest_state && !vmx_guest_state_valid(vcpu);
1539 }
1540 
1541 unsigned long vmx_get_rflags(struct kvm_vcpu *vcpu)
1542 {
1543 	struct vcpu_vmx *vmx = to_vmx(vcpu);
1544 	unsigned long rflags, save_rflags;
1545 
1546 	if (!kvm_register_is_available(vcpu, VCPU_EXREG_RFLAGS)) {
1547 		kvm_register_mark_available(vcpu, VCPU_EXREG_RFLAGS);
1548 		rflags = vmcs_readl(GUEST_RFLAGS);
1549 		if (vmx->rmode.vm86_active) {
1550 			rflags &= RMODE_GUEST_OWNED_EFLAGS_BITS;
1551 			save_rflags = vmx->rmode.save_rflags;
1552 			rflags |= save_rflags & ~RMODE_GUEST_OWNED_EFLAGS_BITS;
1553 		}
1554 		vmx->rflags = rflags;
1555 	}
1556 	return vmx->rflags;
1557 }
1558 
1559 void vmx_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags)
1560 {
1561 	struct vcpu_vmx *vmx = to_vmx(vcpu);
1562 	unsigned long old_rflags;
1563 
1564 	/*
1565 	 * Unlike CR0 and CR4, RFLAGS handling requires checking if the vCPU
1566 	 * is an unrestricted guest in order to mark L2 as needing emulation
1567 	 * if L1 runs L2 as a restricted guest.
1568 	 */
1569 	if (is_unrestricted_guest(vcpu)) {
1570 		kvm_register_mark_available(vcpu, VCPU_EXREG_RFLAGS);
1571 		vmx->rflags = rflags;
1572 		vmcs_writel(GUEST_RFLAGS, rflags);
1573 		return;
1574 	}
1575 
1576 	old_rflags = vmx_get_rflags(vcpu);
1577 	vmx->rflags = rflags;
1578 	if (vmx->rmode.vm86_active) {
1579 		vmx->rmode.save_rflags = rflags;
1580 		rflags |= X86_EFLAGS_IOPL | X86_EFLAGS_VM;
1581 	}
1582 	vmcs_writel(GUEST_RFLAGS, rflags);
1583 
1584 	if ((old_rflags ^ vmx->rflags) & X86_EFLAGS_VM)
1585 		vmx->emulation_required = vmx_emulation_required(vcpu);
1586 }
1587 
1588 bool vmx_get_if_flag(struct kvm_vcpu *vcpu)
1589 {
1590 	return vmx_get_rflags(vcpu) & X86_EFLAGS_IF;
1591 }
1592 
1593 u32 vmx_get_interrupt_shadow(struct kvm_vcpu *vcpu)
1594 {
1595 	u32 interruptibility = vmcs_read32(GUEST_INTERRUPTIBILITY_INFO);
1596 	int ret = 0;
1597 
1598 	if (interruptibility & GUEST_INTR_STATE_STI)
1599 		ret |= KVM_X86_SHADOW_INT_STI;
1600 	if (interruptibility & GUEST_INTR_STATE_MOV_SS)
1601 		ret |= KVM_X86_SHADOW_INT_MOV_SS;
1602 
1603 	return ret;
1604 }
1605 
1606 void vmx_set_interrupt_shadow(struct kvm_vcpu *vcpu, int mask)
1607 {
1608 	u32 interruptibility_old = vmcs_read32(GUEST_INTERRUPTIBILITY_INFO);
1609 	u32 interruptibility = interruptibility_old;
1610 
1611 	interruptibility &= ~(GUEST_INTR_STATE_STI | GUEST_INTR_STATE_MOV_SS);
1612 
1613 	if (mask & KVM_X86_SHADOW_INT_MOV_SS)
1614 		interruptibility |= GUEST_INTR_STATE_MOV_SS;
1615 	else if (mask & KVM_X86_SHADOW_INT_STI)
1616 		interruptibility |= GUEST_INTR_STATE_STI;
1617 
1618 	if ((interruptibility != interruptibility_old))
1619 		vmcs_write32(GUEST_INTERRUPTIBILITY_INFO, interruptibility);
1620 }
1621 
1622 static int vmx_rtit_ctl_check(struct kvm_vcpu *vcpu, u64 data)
1623 {
1624 	struct vcpu_vmx *vmx = to_vmx(vcpu);
1625 	unsigned long value;
1626 
1627 	/*
1628 	 * Any MSR write that attempts to change bits marked reserved will
1629 	 * case a #GP fault.
1630 	 */
1631 	if (data & vmx->pt_desc.ctl_bitmask)
1632 		return 1;
1633 
1634 	/*
1635 	 * Any attempt to modify IA32_RTIT_CTL while TraceEn is set will
1636 	 * result in a #GP unless the same write also clears TraceEn.
1637 	 */
1638 	if ((vmx->pt_desc.guest.ctl & RTIT_CTL_TRACEEN) &&
1639 	    (data & RTIT_CTL_TRACEEN) &&
1640 	    data != vmx->pt_desc.guest.ctl)
1641 		return 1;
1642 
1643 	/*
1644 	 * WRMSR to IA32_RTIT_CTL that sets TraceEn but clears this bit
1645 	 * and FabricEn would cause #GP, if
1646 	 * CPUID.(EAX=14H, ECX=0):ECX.SNGLRGNOUT[bit 2] = 0
1647 	 */
1648 	if ((data & RTIT_CTL_TRACEEN) && !(data & RTIT_CTL_TOPA) &&
1649 		!(data & RTIT_CTL_FABRIC_EN) &&
1650 		!intel_pt_validate_cap(vmx->pt_desc.caps,
1651 					PT_CAP_single_range_output))
1652 		return 1;
1653 
1654 	/*
1655 	 * MTCFreq, CycThresh and PSBFreq encodings check, any MSR write that
1656 	 * utilize encodings marked reserved will cause a #GP fault.
1657 	 */
1658 	value = intel_pt_validate_cap(vmx->pt_desc.caps, PT_CAP_mtc_periods);
1659 	if (intel_pt_validate_cap(vmx->pt_desc.caps, PT_CAP_mtc) &&
1660 			!test_bit((data & RTIT_CTL_MTC_RANGE) >>
1661 			RTIT_CTL_MTC_RANGE_OFFSET, &value))
1662 		return 1;
1663 	value = intel_pt_validate_cap(vmx->pt_desc.caps,
1664 						PT_CAP_cycle_thresholds);
1665 	if (intel_pt_validate_cap(vmx->pt_desc.caps, PT_CAP_psb_cyc) &&
1666 			!test_bit((data & RTIT_CTL_CYC_THRESH) >>
1667 			RTIT_CTL_CYC_THRESH_OFFSET, &value))
1668 		return 1;
1669 	value = intel_pt_validate_cap(vmx->pt_desc.caps, PT_CAP_psb_periods);
1670 	if (intel_pt_validate_cap(vmx->pt_desc.caps, PT_CAP_psb_cyc) &&
1671 			!test_bit((data & RTIT_CTL_PSB_FREQ) >>
1672 			RTIT_CTL_PSB_FREQ_OFFSET, &value))
1673 		return 1;
1674 
1675 	/*
1676 	 * If ADDRx_CFG is reserved or the encodings is >2 will
1677 	 * cause a #GP fault.
1678 	 */
1679 	value = (data & RTIT_CTL_ADDR0) >> RTIT_CTL_ADDR0_OFFSET;
1680 	if ((value && (vmx->pt_desc.num_address_ranges < 1)) || (value > 2))
1681 		return 1;
1682 	value = (data & RTIT_CTL_ADDR1) >> RTIT_CTL_ADDR1_OFFSET;
1683 	if ((value && (vmx->pt_desc.num_address_ranges < 2)) || (value > 2))
1684 		return 1;
1685 	value = (data & RTIT_CTL_ADDR2) >> RTIT_CTL_ADDR2_OFFSET;
1686 	if ((value && (vmx->pt_desc.num_address_ranges < 3)) || (value > 2))
1687 		return 1;
1688 	value = (data & RTIT_CTL_ADDR3) >> RTIT_CTL_ADDR3_OFFSET;
1689 	if ((value && (vmx->pt_desc.num_address_ranges < 4)) || (value > 2))
1690 		return 1;
1691 
1692 	return 0;
1693 }
1694 
1695 int vmx_check_emulate_instruction(struct kvm_vcpu *vcpu, int emul_type,
1696 				  void *insn, int insn_len)
1697 {
1698 	/*
1699 	 * Emulation of instructions in SGX enclaves is impossible as RIP does
1700 	 * not point at the failing instruction, and even if it did, the code
1701 	 * stream is inaccessible.  Inject #UD instead of exiting to userspace
1702 	 * so that guest userspace can't DoS the guest simply by triggering
1703 	 * emulation (enclaves are CPL3 only).
1704 	 */
1705 	if (to_vmx(vcpu)->exit_reason.enclave_mode) {
1706 		kvm_queue_exception(vcpu, UD_VECTOR);
1707 		return X86EMUL_PROPAGATE_FAULT;
1708 	}
1709 
1710 	/* Check that emulation is possible during event vectoring */
1711 	if ((to_vmx(vcpu)->idt_vectoring_info & VECTORING_INFO_VALID_MASK) &&
1712 	    !kvm_can_emulate_event_vectoring(emul_type))
1713 		return X86EMUL_UNHANDLEABLE_VECTORING;
1714 
1715 	return X86EMUL_CONTINUE;
1716 }
1717 
1718 static int skip_emulated_instruction(struct kvm_vcpu *vcpu)
1719 {
1720 	union vmx_exit_reason exit_reason = to_vmx(vcpu)->exit_reason;
1721 	unsigned long rip, orig_rip;
1722 	u32 instr_len;
1723 
1724 	/*
1725 	 * Using VMCS.VM_EXIT_INSTRUCTION_LEN on EPT misconfig depends on
1726 	 * undefined behavior: Intel's SDM doesn't mandate the VMCS field be
1727 	 * set when EPT misconfig occurs.  In practice, real hardware updates
1728 	 * VM_EXIT_INSTRUCTION_LEN on EPT misconfig, but other hypervisors
1729 	 * (namely Hyper-V) don't set it due to it being undefined behavior,
1730 	 * i.e. we end up advancing IP with some random value.
1731 	 */
1732 	if (!static_cpu_has(X86_FEATURE_HYPERVISOR) ||
1733 	    exit_reason.basic != EXIT_REASON_EPT_MISCONFIG) {
1734 		instr_len = vmcs_read32(VM_EXIT_INSTRUCTION_LEN);
1735 
1736 		/*
1737 		 * Emulating an enclave's instructions isn't supported as KVM
1738 		 * cannot access the enclave's memory or its true RIP, e.g. the
1739 		 * vmcs.GUEST_RIP points at the exit point of the enclave, not
1740 		 * the RIP that actually triggered the VM-Exit.  But, because
1741 		 * most instructions that cause VM-Exit will #UD in an enclave,
1742 		 * most instruction-based VM-Exits simply do not occur.
1743 		 *
1744 		 * There are a few exceptions, notably the debug instructions
1745 		 * INT1ICEBRK and INT3, as they are allowed in debug enclaves
1746 		 * and generate #DB/#BP as expected, which KVM might intercept.
1747 		 * But again, the CPU does the dirty work and saves an instr
1748 		 * length of zero so VMMs don't shoot themselves in the foot.
1749 		 * WARN if KVM tries to skip a non-zero length instruction on
1750 		 * a VM-Exit from an enclave.
1751 		 */
1752 		if (!instr_len)
1753 			goto rip_updated;
1754 
1755 		WARN_ONCE(exit_reason.enclave_mode,
1756 			  "skipping instruction after SGX enclave VM-Exit");
1757 
1758 		orig_rip = kvm_rip_read(vcpu);
1759 		rip = orig_rip + instr_len;
1760 #ifdef CONFIG_X86_64
1761 		/*
1762 		 * We need to mask out the high 32 bits of RIP if not in 64-bit
1763 		 * mode, but just finding out that we are in 64-bit mode is
1764 		 * quite expensive.  Only do it if there was a carry.
1765 		 */
1766 		if (unlikely(((rip ^ orig_rip) >> 31) == 3) && !is_64_bit_mode(vcpu))
1767 			rip = (u32)rip;
1768 #endif
1769 		kvm_rip_write(vcpu, rip);
1770 	} else {
1771 		if (!kvm_emulate_instruction(vcpu, EMULTYPE_SKIP))
1772 			return 0;
1773 	}
1774 
1775 rip_updated:
1776 	/* skipping an emulated instruction also counts */
1777 	vmx_set_interrupt_shadow(vcpu, 0);
1778 
1779 	return 1;
1780 }
1781 
1782 /*
1783  * Recognizes a pending MTF VM-exit and records the nested state for later
1784  * delivery.
1785  */
1786 void vmx_update_emulated_instruction(struct kvm_vcpu *vcpu)
1787 {
1788 	struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
1789 	struct vcpu_vmx *vmx = to_vmx(vcpu);
1790 
1791 	if (!is_guest_mode(vcpu))
1792 		return;
1793 
1794 	/*
1795 	 * Per the SDM, MTF takes priority over debug-trap exceptions besides
1796 	 * TSS T-bit traps and ICEBP (INT1).  KVM doesn't emulate T-bit traps
1797 	 * or ICEBP (in the emulator proper), and skipping of ICEBP after an
1798 	 * intercepted #DB deliberately avoids single-step #DB and MTF updates
1799 	 * as ICEBP is higher priority than both.  As instruction emulation is
1800 	 * completed at this point (i.e. KVM is at the instruction boundary),
1801 	 * any #DB exception pending delivery must be a debug-trap of lower
1802 	 * priority than MTF.  Record the pending MTF state to be delivered in
1803 	 * vmx_check_nested_events().
1804 	 */
1805 	if (nested_cpu_has_mtf(vmcs12) &&
1806 	    (!vcpu->arch.exception.pending ||
1807 	     vcpu->arch.exception.vector == DB_VECTOR) &&
1808 	    (!vcpu->arch.exception_vmexit.pending ||
1809 	     vcpu->arch.exception_vmexit.vector == DB_VECTOR)) {
1810 		vmx->nested.mtf_pending = true;
1811 		kvm_make_request(KVM_REQ_EVENT, vcpu);
1812 	} else {
1813 		vmx->nested.mtf_pending = false;
1814 	}
1815 }
1816 
1817 int vmx_skip_emulated_instruction(struct kvm_vcpu *vcpu)
1818 {
1819 	vmx_update_emulated_instruction(vcpu);
1820 	return skip_emulated_instruction(vcpu);
1821 }
1822 
1823 static void vmx_clear_hlt(struct kvm_vcpu *vcpu)
1824 {
1825 	/*
1826 	 * Ensure that we clear the HLT state in the VMCS.  We don't need to
1827 	 * explicitly skip the instruction because if the HLT state is set,
1828 	 * then the instruction is already executing and RIP has already been
1829 	 * advanced.
1830 	 */
1831 	if (kvm_hlt_in_guest(vcpu->kvm) &&
1832 			vmcs_read32(GUEST_ACTIVITY_STATE) == GUEST_ACTIVITY_HLT)
1833 		vmcs_write32(GUEST_ACTIVITY_STATE, GUEST_ACTIVITY_ACTIVE);
1834 }
1835 
1836 void vmx_inject_exception(struct kvm_vcpu *vcpu)
1837 {
1838 	struct kvm_queued_exception *ex = &vcpu->arch.exception;
1839 	u32 intr_info = ex->vector | INTR_INFO_VALID_MASK;
1840 	struct vcpu_vmx *vmx = to_vmx(vcpu);
1841 
1842 	kvm_deliver_exception_payload(vcpu, ex);
1843 
1844 	if (ex->has_error_code) {
1845 		/*
1846 		 * Despite the error code being architecturally defined as 32
1847 		 * bits, and the VMCS field being 32 bits, Intel CPUs and thus
1848 		 * VMX don't actually supporting setting bits 31:16.  Hardware
1849 		 * will (should) never provide a bogus error code, but AMD CPUs
1850 		 * do generate error codes with bits 31:16 set, and so KVM's
1851 		 * ABI lets userspace shove in arbitrary 32-bit values.  Drop
1852 		 * the upper bits to avoid VM-Fail, losing information that
1853 		 * doesn't really exist is preferable to killing the VM.
1854 		 */
1855 		vmcs_write32(VM_ENTRY_EXCEPTION_ERROR_CODE, (u16)ex->error_code);
1856 		intr_info |= INTR_INFO_DELIVER_CODE_MASK;
1857 	}
1858 
1859 	if (vmx->rmode.vm86_active) {
1860 		int inc_eip = 0;
1861 		if (kvm_exception_is_soft(ex->vector))
1862 			inc_eip = vcpu->arch.event_exit_inst_len;
1863 		kvm_inject_realmode_interrupt(vcpu, ex->vector, inc_eip);
1864 		return;
1865 	}
1866 
1867 	WARN_ON_ONCE(vmx->emulation_required);
1868 
1869 	if (kvm_exception_is_soft(ex->vector)) {
1870 		vmcs_write32(VM_ENTRY_INSTRUCTION_LEN,
1871 			     vmx->vcpu.arch.event_exit_inst_len);
1872 		intr_info |= INTR_TYPE_SOFT_EXCEPTION;
1873 	} else
1874 		intr_info |= INTR_TYPE_HARD_EXCEPTION;
1875 
1876 	vmcs_write32(VM_ENTRY_INTR_INFO_FIELD, intr_info);
1877 
1878 	vmx_clear_hlt(vcpu);
1879 }
1880 
1881 static void vmx_setup_uret_msr(struct vcpu_vmx *vmx, unsigned int msr,
1882 			       bool load_into_hardware)
1883 {
1884 	struct vmx_uret_msr *uret_msr;
1885 
1886 	uret_msr = vmx_find_uret_msr(vmx, msr);
1887 	if (!uret_msr)
1888 		return;
1889 
1890 	uret_msr->load_into_hardware = load_into_hardware;
1891 }
1892 
1893 /*
1894  * Configuring user return MSRs to automatically save, load, and restore MSRs
1895  * that need to be shoved into hardware when running the guest.  Note, omitting
1896  * an MSR here does _NOT_ mean it's not emulated, only that it will not be
1897  * loaded into hardware when running the guest.
1898  */
1899 static void vmx_setup_uret_msrs(struct vcpu_vmx *vmx)
1900 {
1901 #ifdef CONFIG_X86_64
1902 	bool load_syscall_msrs;
1903 
1904 	/*
1905 	 * The SYSCALL MSRs are only needed on long mode guests, and only
1906 	 * when EFER.SCE is set.
1907 	 */
1908 	load_syscall_msrs = is_long_mode(&vmx->vcpu) &&
1909 			    (vmx->vcpu.arch.efer & EFER_SCE);
1910 
1911 	vmx_setup_uret_msr(vmx, MSR_STAR, load_syscall_msrs);
1912 	vmx_setup_uret_msr(vmx, MSR_LSTAR, load_syscall_msrs);
1913 	vmx_setup_uret_msr(vmx, MSR_SYSCALL_MASK, load_syscall_msrs);
1914 #endif
1915 	vmx_setup_uret_msr(vmx, MSR_EFER, update_transition_efer(vmx));
1916 
1917 	vmx_setup_uret_msr(vmx, MSR_TSC_AUX,
1918 			   guest_cpu_cap_has(&vmx->vcpu, X86_FEATURE_RDTSCP) ||
1919 			   guest_cpu_cap_has(&vmx->vcpu, X86_FEATURE_RDPID));
1920 
1921 	/*
1922 	 * hle=0, rtm=0, tsx_ctrl=1 can be found with some combinations of new
1923 	 * kernel and old userspace.  If those guests run on a tsx=off host, do
1924 	 * allow guests to use TSX_CTRL, but don't change the value in hardware
1925 	 * so that TSX remains always disabled.
1926 	 */
1927 	vmx_setup_uret_msr(vmx, MSR_IA32_TSX_CTRL, boot_cpu_has(X86_FEATURE_RTM));
1928 
1929 	/*
1930 	 * The set of MSRs to load may have changed, reload MSRs before the
1931 	 * next VM-Enter.
1932 	 */
1933 	vmx->guest_uret_msrs_loaded = false;
1934 }
1935 
1936 u64 vmx_get_l2_tsc_offset(struct kvm_vcpu *vcpu)
1937 {
1938 	struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
1939 
1940 	if (nested_cpu_has(vmcs12, CPU_BASED_USE_TSC_OFFSETTING))
1941 		return vmcs12->tsc_offset;
1942 
1943 	return 0;
1944 }
1945 
1946 u64 vmx_get_l2_tsc_multiplier(struct kvm_vcpu *vcpu)
1947 {
1948 	struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
1949 
1950 	if (nested_cpu_has(vmcs12, CPU_BASED_USE_TSC_OFFSETTING) &&
1951 	    nested_cpu_has2(vmcs12, SECONDARY_EXEC_TSC_SCALING))
1952 		return vmcs12->tsc_multiplier;
1953 
1954 	return kvm_caps.default_tsc_scaling_ratio;
1955 }
1956 
1957 void vmx_write_tsc_offset(struct kvm_vcpu *vcpu)
1958 {
1959 	vmcs_write64(TSC_OFFSET, vcpu->arch.tsc_offset);
1960 }
1961 
1962 void vmx_write_tsc_multiplier(struct kvm_vcpu *vcpu)
1963 {
1964 	vmcs_write64(TSC_MULTIPLIER, vcpu->arch.tsc_scaling_ratio);
1965 }
1966 
1967 /*
1968  * Userspace is allowed to set any supported IA32_FEATURE_CONTROL regardless of
1969  * guest CPUID.  Note, KVM allows userspace to set "VMX in SMX" to maintain
1970  * backwards compatibility even though KVM doesn't support emulating SMX.  And
1971  * because userspace set "VMX in SMX", the guest must also be allowed to set it,
1972  * e.g. if the MSR is left unlocked and the guest does a RMW operation.
1973  */
1974 #define KVM_SUPPORTED_FEATURE_CONTROL  (FEAT_CTL_LOCKED			 | \
1975 					FEAT_CTL_VMX_ENABLED_INSIDE_SMX	 | \
1976 					FEAT_CTL_VMX_ENABLED_OUTSIDE_SMX | \
1977 					FEAT_CTL_SGX_LC_ENABLED		 | \
1978 					FEAT_CTL_SGX_ENABLED		 | \
1979 					FEAT_CTL_LMCE_ENABLED)
1980 
1981 static inline bool is_vmx_feature_control_msr_valid(struct vcpu_vmx *vmx,
1982 						    struct msr_data *msr)
1983 {
1984 	uint64_t valid_bits;
1985 
1986 	/*
1987 	 * Ensure KVM_SUPPORTED_FEATURE_CONTROL is updated when new bits are
1988 	 * exposed to the guest.
1989 	 */
1990 	WARN_ON_ONCE(vmx->msr_ia32_feature_control_valid_bits &
1991 		     ~KVM_SUPPORTED_FEATURE_CONTROL);
1992 
1993 	if (!msr->host_initiated &&
1994 	    (vmx->msr_ia32_feature_control & FEAT_CTL_LOCKED))
1995 		return false;
1996 
1997 	if (msr->host_initiated)
1998 		valid_bits = KVM_SUPPORTED_FEATURE_CONTROL;
1999 	else
2000 		valid_bits = vmx->msr_ia32_feature_control_valid_bits;
2001 
2002 	return !(msr->data & ~valid_bits);
2003 }
2004 
2005 int vmx_get_feature_msr(u32 msr, u64 *data)
2006 {
2007 	switch (msr) {
2008 	case KVM_FIRST_EMULATED_VMX_MSR ... KVM_LAST_EMULATED_VMX_MSR:
2009 		if (!nested)
2010 			return 1;
2011 		return vmx_get_vmx_msr(&vmcs_config.nested, msr, data);
2012 	default:
2013 		return KVM_MSR_RET_UNSUPPORTED;
2014 	}
2015 }
2016 
2017 /*
2018  * Reads an msr value (of 'msr_info->index') into 'msr_info->data'.
2019  * Returns 0 on success, non-0 otherwise.
2020  * Assumes vcpu_load() was already called.
2021  */
2022 int vmx_get_msr(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
2023 {
2024 	struct vcpu_vmx *vmx = to_vmx(vcpu);
2025 	struct vmx_uret_msr *msr;
2026 	u32 index;
2027 
2028 	switch (msr_info->index) {
2029 #ifdef CONFIG_X86_64
2030 	case MSR_FS_BASE:
2031 		msr_info->data = vmcs_readl(GUEST_FS_BASE);
2032 		break;
2033 	case MSR_GS_BASE:
2034 		msr_info->data = vmcs_readl(GUEST_GS_BASE);
2035 		break;
2036 	case MSR_KERNEL_GS_BASE:
2037 		msr_info->data = vmx_read_guest_kernel_gs_base(vmx);
2038 		break;
2039 #endif
2040 	case MSR_EFER:
2041 		return kvm_get_msr_common(vcpu, msr_info);
2042 	case MSR_IA32_TSX_CTRL:
2043 		if (!msr_info->host_initiated &&
2044 		    !(vcpu->arch.arch_capabilities & ARCH_CAP_TSX_CTRL_MSR))
2045 			return 1;
2046 		goto find_uret_msr;
2047 	case MSR_IA32_UMWAIT_CONTROL:
2048 		if (!msr_info->host_initiated && !vmx_has_waitpkg(vmx))
2049 			return 1;
2050 
2051 		msr_info->data = vmx->msr_ia32_umwait_control;
2052 		break;
2053 	case MSR_IA32_SPEC_CTRL:
2054 		if (!msr_info->host_initiated &&
2055 		    !guest_has_spec_ctrl_msr(vcpu))
2056 			return 1;
2057 
2058 		msr_info->data = to_vmx(vcpu)->spec_ctrl;
2059 		break;
2060 	case MSR_IA32_SYSENTER_CS:
2061 		msr_info->data = vmcs_read32(GUEST_SYSENTER_CS);
2062 		break;
2063 	case MSR_IA32_SYSENTER_EIP:
2064 		msr_info->data = vmcs_readl(GUEST_SYSENTER_EIP);
2065 		break;
2066 	case MSR_IA32_SYSENTER_ESP:
2067 		msr_info->data = vmcs_readl(GUEST_SYSENTER_ESP);
2068 		break;
2069 	case MSR_IA32_BNDCFGS:
2070 		if (!kvm_mpx_supported() ||
2071 		    (!msr_info->host_initiated &&
2072 		     !guest_cpu_cap_has(vcpu, X86_FEATURE_MPX)))
2073 			return 1;
2074 		msr_info->data = vmcs_read64(GUEST_BNDCFGS);
2075 		break;
2076 	case MSR_IA32_MCG_EXT_CTL:
2077 		if (!msr_info->host_initiated &&
2078 		    !(vmx->msr_ia32_feature_control &
2079 		      FEAT_CTL_LMCE_ENABLED))
2080 			return 1;
2081 		msr_info->data = vcpu->arch.mcg_ext_ctl;
2082 		break;
2083 	case MSR_IA32_FEAT_CTL:
2084 		msr_info->data = vmx->msr_ia32_feature_control;
2085 		break;
2086 	case MSR_IA32_SGXLEPUBKEYHASH0 ... MSR_IA32_SGXLEPUBKEYHASH3:
2087 		if (!msr_info->host_initiated &&
2088 		    !guest_cpu_cap_has(vcpu, X86_FEATURE_SGX_LC))
2089 			return 1;
2090 		msr_info->data = to_vmx(vcpu)->msr_ia32_sgxlepubkeyhash
2091 			[msr_info->index - MSR_IA32_SGXLEPUBKEYHASH0];
2092 		break;
2093 	case KVM_FIRST_EMULATED_VMX_MSR ... KVM_LAST_EMULATED_VMX_MSR:
2094 		if (!guest_cpu_cap_has(vcpu, X86_FEATURE_VMX))
2095 			return 1;
2096 		if (vmx_get_vmx_msr(&vmx->nested.msrs, msr_info->index,
2097 				    &msr_info->data))
2098 			return 1;
2099 #ifdef CONFIG_KVM_HYPERV
2100 		/*
2101 		 * Enlightened VMCS v1 doesn't have certain VMCS fields but
2102 		 * instead of just ignoring the features, different Hyper-V
2103 		 * versions are either trying to use them and fail or do some
2104 		 * sanity checking and refuse to boot. Filter all unsupported
2105 		 * features out.
2106 		 */
2107 		if (!msr_info->host_initiated && guest_cpu_cap_has_evmcs(vcpu))
2108 			nested_evmcs_filter_control_msr(vcpu, msr_info->index,
2109 							&msr_info->data);
2110 #endif
2111 		break;
2112 	case MSR_IA32_RTIT_CTL:
2113 		if (!vmx_pt_mode_is_host_guest())
2114 			return 1;
2115 		msr_info->data = vmx->pt_desc.guest.ctl;
2116 		break;
2117 	case MSR_IA32_RTIT_STATUS:
2118 		if (!vmx_pt_mode_is_host_guest())
2119 			return 1;
2120 		msr_info->data = vmx->pt_desc.guest.status;
2121 		break;
2122 	case MSR_IA32_RTIT_CR3_MATCH:
2123 		if (!vmx_pt_mode_is_host_guest() ||
2124 			!intel_pt_validate_cap(vmx->pt_desc.caps,
2125 						PT_CAP_cr3_filtering))
2126 			return 1;
2127 		msr_info->data = vmx->pt_desc.guest.cr3_match;
2128 		break;
2129 	case MSR_IA32_RTIT_OUTPUT_BASE:
2130 		if (!vmx_pt_mode_is_host_guest() ||
2131 			(!intel_pt_validate_cap(vmx->pt_desc.caps,
2132 					PT_CAP_topa_output) &&
2133 			 !intel_pt_validate_cap(vmx->pt_desc.caps,
2134 					PT_CAP_single_range_output)))
2135 			return 1;
2136 		msr_info->data = vmx->pt_desc.guest.output_base;
2137 		break;
2138 	case MSR_IA32_RTIT_OUTPUT_MASK:
2139 		if (!vmx_pt_mode_is_host_guest() ||
2140 			(!intel_pt_validate_cap(vmx->pt_desc.caps,
2141 					PT_CAP_topa_output) &&
2142 			 !intel_pt_validate_cap(vmx->pt_desc.caps,
2143 					PT_CAP_single_range_output)))
2144 			return 1;
2145 		msr_info->data = vmx->pt_desc.guest.output_mask;
2146 		break;
2147 	case MSR_IA32_RTIT_ADDR0_A ... MSR_IA32_RTIT_ADDR3_B:
2148 		index = msr_info->index - MSR_IA32_RTIT_ADDR0_A;
2149 		if (!vmx_pt_mode_is_host_guest() ||
2150 		    (index >= 2 * vmx->pt_desc.num_address_ranges))
2151 			return 1;
2152 		if (index % 2)
2153 			msr_info->data = vmx->pt_desc.guest.addr_b[index / 2];
2154 		else
2155 			msr_info->data = vmx->pt_desc.guest.addr_a[index / 2];
2156 		break;
2157 	case MSR_IA32_DEBUGCTLMSR:
2158 		msr_info->data = vmcs_read64(GUEST_IA32_DEBUGCTL);
2159 		break;
2160 	default:
2161 	find_uret_msr:
2162 		msr = vmx_find_uret_msr(vmx, msr_info->index);
2163 		if (msr) {
2164 			msr_info->data = msr->data;
2165 			break;
2166 		}
2167 		return kvm_get_msr_common(vcpu, msr_info);
2168 	}
2169 
2170 	return 0;
2171 }
2172 
2173 static u64 nested_vmx_truncate_sysenter_addr(struct kvm_vcpu *vcpu,
2174 						    u64 data)
2175 {
2176 #ifdef CONFIG_X86_64
2177 	if (!guest_cpu_cap_has(vcpu, X86_FEATURE_LM))
2178 		return (u32)data;
2179 #endif
2180 	return (unsigned long)data;
2181 }
2182 
2183 static u64 vmx_get_supported_debugctl(struct kvm_vcpu *vcpu, bool host_initiated)
2184 {
2185 	u64 debugctl = 0;
2186 
2187 	if (boot_cpu_has(X86_FEATURE_BUS_LOCK_DETECT) &&
2188 	    (host_initiated || guest_cpu_cap_has(vcpu, X86_FEATURE_BUS_LOCK_DETECT)))
2189 		debugctl |= DEBUGCTLMSR_BUS_LOCK_DETECT;
2190 
2191 	if ((kvm_caps.supported_perf_cap & PMU_CAP_LBR_FMT) &&
2192 	    (host_initiated || intel_pmu_lbr_is_enabled(vcpu)))
2193 		debugctl |= DEBUGCTLMSR_LBR | DEBUGCTLMSR_FREEZE_LBRS_ON_PMI;
2194 
2195 	return debugctl;
2196 }
2197 
2198 /*
2199  * Writes msr value into the appropriate "register".
2200  * Returns 0 on success, non-0 otherwise.
2201  * Assumes vcpu_load() was already called.
2202  */
2203 int vmx_set_msr(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
2204 {
2205 	struct vcpu_vmx *vmx = to_vmx(vcpu);
2206 	struct vmx_uret_msr *msr;
2207 	int ret = 0;
2208 	u32 msr_index = msr_info->index;
2209 	u64 data = msr_info->data;
2210 	u32 index;
2211 
2212 	switch (msr_index) {
2213 	case MSR_EFER:
2214 		ret = kvm_set_msr_common(vcpu, msr_info);
2215 		break;
2216 #ifdef CONFIG_X86_64
2217 	case MSR_FS_BASE:
2218 		vmx_segment_cache_clear(vmx);
2219 		vmcs_writel(GUEST_FS_BASE, data);
2220 		break;
2221 	case MSR_GS_BASE:
2222 		vmx_segment_cache_clear(vmx);
2223 		vmcs_writel(GUEST_GS_BASE, data);
2224 		break;
2225 	case MSR_KERNEL_GS_BASE:
2226 		vmx_write_guest_kernel_gs_base(vmx, data);
2227 		break;
2228 	case MSR_IA32_XFD:
2229 		ret = kvm_set_msr_common(vcpu, msr_info);
2230 		/*
2231 		 * Always intercepting WRMSR could incur non-negligible
2232 		 * overhead given xfd might be changed frequently in
2233 		 * guest context switch. Disable write interception
2234 		 * upon the first write with a non-zero value (indicating
2235 		 * potential usage on dynamic xfeatures). Also update
2236 		 * exception bitmap to trap #NM for proper virtualization
2237 		 * of guest xfd_err.
2238 		 */
2239 		if (!ret && data) {
2240 			vmx_disable_intercept_for_msr(vcpu, MSR_IA32_XFD,
2241 						      MSR_TYPE_RW);
2242 			vcpu->arch.xfd_no_write_intercept = true;
2243 			vmx_update_exception_bitmap(vcpu);
2244 		}
2245 		break;
2246 #endif
2247 	case MSR_IA32_SYSENTER_CS:
2248 		if (is_guest_mode(vcpu))
2249 			get_vmcs12(vcpu)->guest_sysenter_cs = data;
2250 		vmcs_write32(GUEST_SYSENTER_CS, data);
2251 		break;
2252 	case MSR_IA32_SYSENTER_EIP:
2253 		if (is_guest_mode(vcpu)) {
2254 			data = nested_vmx_truncate_sysenter_addr(vcpu, data);
2255 			get_vmcs12(vcpu)->guest_sysenter_eip = data;
2256 		}
2257 		vmcs_writel(GUEST_SYSENTER_EIP, data);
2258 		break;
2259 	case MSR_IA32_SYSENTER_ESP:
2260 		if (is_guest_mode(vcpu)) {
2261 			data = nested_vmx_truncate_sysenter_addr(vcpu, data);
2262 			get_vmcs12(vcpu)->guest_sysenter_esp = data;
2263 		}
2264 		vmcs_writel(GUEST_SYSENTER_ESP, data);
2265 		break;
2266 	case MSR_IA32_DEBUGCTLMSR: {
2267 		u64 invalid;
2268 
2269 		invalid = data & ~vmx_get_supported_debugctl(vcpu, msr_info->host_initiated);
2270 		if (invalid & (DEBUGCTLMSR_BTF|DEBUGCTLMSR_LBR)) {
2271 			kvm_pr_unimpl_wrmsr(vcpu, msr_index, data);
2272 			data &= ~(DEBUGCTLMSR_BTF|DEBUGCTLMSR_LBR);
2273 			invalid &= ~(DEBUGCTLMSR_BTF|DEBUGCTLMSR_LBR);
2274 		}
2275 
2276 		if (invalid)
2277 			return 1;
2278 
2279 		if (is_guest_mode(vcpu) && get_vmcs12(vcpu)->vm_exit_controls &
2280 						VM_EXIT_SAVE_DEBUG_CONTROLS)
2281 			get_vmcs12(vcpu)->guest_ia32_debugctl = data;
2282 
2283 		vmcs_write64(GUEST_IA32_DEBUGCTL, data);
2284 		if (intel_pmu_lbr_is_enabled(vcpu) && !to_vmx(vcpu)->lbr_desc.event &&
2285 		    (data & DEBUGCTLMSR_LBR))
2286 			intel_pmu_create_guest_lbr_event(vcpu);
2287 		return 0;
2288 	}
2289 	case MSR_IA32_BNDCFGS:
2290 		if (!kvm_mpx_supported() ||
2291 		    (!msr_info->host_initiated &&
2292 		     !guest_cpu_cap_has(vcpu, X86_FEATURE_MPX)))
2293 			return 1;
2294 		if (is_noncanonical_msr_address(data & PAGE_MASK, vcpu) ||
2295 		    (data & MSR_IA32_BNDCFGS_RSVD))
2296 			return 1;
2297 
2298 		if (is_guest_mode(vcpu) &&
2299 		    ((vmx->nested.msrs.entry_ctls_high & VM_ENTRY_LOAD_BNDCFGS) ||
2300 		     (vmx->nested.msrs.exit_ctls_high & VM_EXIT_CLEAR_BNDCFGS)))
2301 			get_vmcs12(vcpu)->guest_bndcfgs = data;
2302 
2303 		vmcs_write64(GUEST_BNDCFGS, data);
2304 		break;
2305 	case MSR_IA32_UMWAIT_CONTROL:
2306 		if (!msr_info->host_initiated && !vmx_has_waitpkg(vmx))
2307 			return 1;
2308 
2309 		/* The reserved bit 1 and non-32 bit [63:32] should be zero */
2310 		if (data & (BIT_ULL(1) | GENMASK_ULL(63, 32)))
2311 			return 1;
2312 
2313 		vmx->msr_ia32_umwait_control = data;
2314 		break;
2315 	case MSR_IA32_SPEC_CTRL:
2316 		if (!msr_info->host_initiated &&
2317 		    !guest_has_spec_ctrl_msr(vcpu))
2318 			return 1;
2319 
2320 		if (kvm_spec_ctrl_test_value(data))
2321 			return 1;
2322 
2323 		vmx->spec_ctrl = data;
2324 		if (!data)
2325 			break;
2326 
2327 		/*
2328 		 * For non-nested:
2329 		 * When it's written (to non-zero) for the first time, pass
2330 		 * it through.
2331 		 *
2332 		 * For nested:
2333 		 * The handling of the MSR bitmap for L2 guests is done in
2334 		 * nested_vmx_prepare_msr_bitmap. We should not touch the
2335 		 * vmcs02.msr_bitmap here since it gets completely overwritten
2336 		 * in the merging. We update the vmcs01 here for L1 as well
2337 		 * since it will end up touching the MSR anyway now.
2338 		 */
2339 		vmx_disable_intercept_for_msr(vcpu,
2340 					      MSR_IA32_SPEC_CTRL,
2341 					      MSR_TYPE_RW);
2342 		break;
2343 	case MSR_IA32_TSX_CTRL:
2344 		if (!msr_info->host_initiated &&
2345 		    !(vcpu->arch.arch_capabilities & ARCH_CAP_TSX_CTRL_MSR))
2346 			return 1;
2347 		if (data & ~(TSX_CTRL_RTM_DISABLE | TSX_CTRL_CPUID_CLEAR))
2348 			return 1;
2349 		goto find_uret_msr;
2350 	case MSR_IA32_CR_PAT:
2351 		ret = kvm_set_msr_common(vcpu, msr_info);
2352 		if (ret)
2353 			break;
2354 
2355 		if (is_guest_mode(vcpu) &&
2356 		    get_vmcs12(vcpu)->vm_exit_controls & VM_EXIT_SAVE_IA32_PAT)
2357 			get_vmcs12(vcpu)->guest_ia32_pat = data;
2358 
2359 		if (vmcs_config.vmentry_ctrl & VM_ENTRY_LOAD_IA32_PAT)
2360 			vmcs_write64(GUEST_IA32_PAT, data);
2361 		break;
2362 	case MSR_IA32_MCG_EXT_CTL:
2363 		if ((!msr_info->host_initiated &&
2364 		     !(to_vmx(vcpu)->msr_ia32_feature_control &
2365 		       FEAT_CTL_LMCE_ENABLED)) ||
2366 		    (data & ~MCG_EXT_CTL_LMCE_EN))
2367 			return 1;
2368 		vcpu->arch.mcg_ext_ctl = data;
2369 		break;
2370 	case MSR_IA32_FEAT_CTL:
2371 		if (!is_vmx_feature_control_msr_valid(vmx, msr_info))
2372 			return 1;
2373 
2374 		vmx->msr_ia32_feature_control = data;
2375 		if (msr_info->host_initiated && data == 0)
2376 			vmx_leave_nested(vcpu);
2377 
2378 		/* SGX may be enabled/disabled by guest's firmware */
2379 		vmx_write_encls_bitmap(vcpu, NULL);
2380 		break;
2381 	case MSR_IA32_SGXLEPUBKEYHASH0 ... MSR_IA32_SGXLEPUBKEYHASH3:
2382 		/*
2383 		 * On real hardware, the LE hash MSRs are writable before
2384 		 * the firmware sets bit 0 in MSR 0x7a ("activating" SGX),
2385 		 * at which point SGX related bits in IA32_FEATURE_CONTROL
2386 		 * become writable.
2387 		 *
2388 		 * KVM does not emulate SGX activation for simplicity, so
2389 		 * allow writes to the LE hash MSRs if IA32_FEATURE_CONTROL
2390 		 * is unlocked.  This is technically not architectural
2391 		 * behavior, but it's close enough.
2392 		 */
2393 		if (!msr_info->host_initiated &&
2394 		    (!guest_cpu_cap_has(vcpu, X86_FEATURE_SGX_LC) ||
2395 		    ((vmx->msr_ia32_feature_control & FEAT_CTL_LOCKED) &&
2396 		    !(vmx->msr_ia32_feature_control & FEAT_CTL_SGX_LC_ENABLED))))
2397 			return 1;
2398 		vmx->msr_ia32_sgxlepubkeyhash
2399 			[msr_index - MSR_IA32_SGXLEPUBKEYHASH0] = data;
2400 		break;
2401 	case KVM_FIRST_EMULATED_VMX_MSR ... KVM_LAST_EMULATED_VMX_MSR:
2402 		if (!msr_info->host_initiated)
2403 			return 1; /* they are read-only */
2404 		if (!guest_cpu_cap_has(vcpu, X86_FEATURE_VMX))
2405 			return 1;
2406 		return vmx_set_vmx_msr(vcpu, msr_index, data);
2407 	case MSR_IA32_RTIT_CTL:
2408 		if (!vmx_pt_mode_is_host_guest() ||
2409 			vmx_rtit_ctl_check(vcpu, data) ||
2410 			vmx->nested.vmxon)
2411 			return 1;
2412 		vmcs_write64(GUEST_IA32_RTIT_CTL, data);
2413 		vmx->pt_desc.guest.ctl = data;
2414 		pt_update_intercept_for_msr(vcpu);
2415 		break;
2416 	case MSR_IA32_RTIT_STATUS:
2417 		if (!pt_can_write_msr(vmx))
2418 			return 1;
2419 		if (data & MSR_IA32_RTIT_STATUS_MASK)
2420 			return 1;
2421 		vmx->pt_desc.guest.status = data;
2422 		break;
2423 	case MSR_IA32_RTIT_CR3_MATCH:
2424 		if (!pt_can_write_msr(vmx))
2425 			return 1;
2426 		if (!intel_pt_validate_cap(vmx->pt_desc.caps,
2427 					   PT_CAP_cr3_filtering))
2428 			return 1;
2429 		vmx->pt_desc.guest.cr3_match = data;
2430 		break;
2431 	case MSR_IA32_RTIT_OUTPUT_BASE:
2432 		if (!pt_can_write_msr(vmx))
2433 			return 1;
2434 		if (!intel_pt_validate_cap(vmx->pt_desc.caps,
2435 					   PT_CAP_topa_output) &&
2436 		    !intel_pt_validate_cap(vmx->pt_desc.caps,
2437 					   PT_CAP_single_range_output))
2438 			return 1;
2439 		if (!pt_output_base_valid(vcpu, data))
2440 			return 1;
2441 		vmx->pt_desc.guest.output_base = data;
2442 		break;
2443 	case MSR_IA32_RTIT_OUTPUT_MASK:
2444 		if (!pt_can_write_msr(vmx))
2445 			return 1;
2446 		if (!intel_pt_validate_cap(vmx->pt_desc.caps,
2447 					   PT_CAP_topa_output) &&
2448 		    !intel_pt_validate_cap(vmx->pt_desc.caps,
2449 					   PT_CAP_single_range_output))
2450 			return 1;
2451 		vmx->pt_desc.guest.output_mask = data;
2452 		break;
2453 	case MSR_IA32_RTIT_ADDR0_A ... MSR_IA32_RTIT_ADDR3_B:
2454 		if (!pt_can_write_msr(vmx))
2455 			return 1;
2456 		index = msr_info->index - MSR_IA32_RTIT_ADDR0_A;
2457 		if (index >= 2 * vmx->pt_desc.num_address_ranges)
2458 			return 1;
2459 		if (is_noncanonical_msr_address(data, vcpu))
2460 			return 1;
2461 		if (index % 2)
2462 			vmx->pt_desc.guest.addr_b[index / 2] = data;
2463 		else
2464 			vmx->pt_desc.guest.addr_a[index / 2] = data;
2465 		break;
2466 	case MSR_IA32_PERF_CAPABILITIES:
2467 		if (data & PMU_CAP_LBR_FMT) {
2468 			if ((data & PMU_CAP_LBR_FMT) !=
2469 			    (kvm_caps.supported_perf_cap & PMU_CAP_LBR_FMT))
2470 				return 1;
2471 			if (!cpuid_model_is_consistent(vcpu))
2472 				return 1;
2473 		}
2474 		if (data & PERF_CAP_PEBS_FORMAT) {
2475 			if ((data & PERF_CAP_PEBS_MASK) !=
2476 			    (kvm_caps.supported_perf_cap & PERF_CAP_PEBS_MASK))
2477 				return 1;
2478 			if (!guest_cpu_cap_has(vcpu, X86_FEATURE_DS))
2479 				return 1;
2480 			if (!guest_cpu_cap_has(vcpu, X86_FEATURE_DTES64))
2481 				return 1;
2482 			if (!cpuid_model_is_consistent(vcpu))
2483 				return 1;
2484 		}
2485 		ret = kvm_set_msr_common(vcpu, msr_info);
2486 		break;
2487 
2488 	default:
2489 	find_uret_msr:
2490 		msr = vmx_find_uret_msr(vmx, msr_index);
2491 		if (msr)
2492 			ret = vmx_set_guest_uret_msr(vmx, msr, data);
2493 		else
2494 			ret = kvm_set_msr_common(vcpu, msr_info);
2495 	}
2496 
2497 	/* FB_CLEAR may have changed, also update the FB_CLEAR_DIS behavior */
2498 	if (msr_index == MSR_IA32_ARCH_CAPABILITIES)
2499 		vmx_update_fb_clear_dis(vcpu, vmx);
2500 
2501 	return ret;
2502 }
2503 
2504 void vmx_cache_reg(struct kvm_vcpu *vcpu, enum kvm_reg reg)
2505 {
2506 	unsigned long guest_owned_bits;
2507 
2508 	kvm_register_mark_available(vcpu, reg);
2509 
2510 	switch (reg) {
2511 	case VCPU_REGS_RSP:
2512 		vcpu->arch.regs[VCPU_REGS_RSP] = vmcs_readl(GUEST_RSP);
2513 		break;
2514 	case VCPU_REGS_RIP:
2515 		vcpu->arch.regs[VCPU_REGS_RIP] = vmcs_readl(GUEST_RIP);
2516 		break;
2517 	case VCPU_EXREG_PDPTR:
2518 		if (enable_ept)
2519 			ept_save_pdptrs(vcpu);
2520 		break;
2521 	case VCPU_EXREG_CR0:
2522 		guest_owned_bits = vcpu->arch.cr0_guest_owned_bits;
2523 
2524 		vcpu->arch.cr0 &= ~guest_owned_bits;
2525 		vcpu->arch.cr0 |= vmcs_readl(GUEST_CR0) & guest_owned_bits;
2526 		break;
2527 	case VCPU_EXREG_CR3:
2528 		/*
2529 		 * When intercepting CR3 loads, e.g. for shadowing paging, KVM's
2530 		 * CR3 is loaded into hardware, not the guest's CR3.
2531 		 */
2532 		if (!(exec_controls_get(to_vmx(vcpu)) & CPU_BASED_CR3_LOAD_EXITING))
2533 			vcpu->arch.cr3 = vmcs_readl(GUEST_CR3);
2534 		break;
2535 	case VCPU_EXREG_CR4:
2536 		guest_owned_bits = vcpu->arch.cr4_guest_owned_bits;
2537 
2538 		vcpu->arch.cr4 &= ~guest_owned_bits;
2539 		vcpu->arch.cr4 |= vmcs_readl(GUEST_CR4) & guest_owned_bits;
2540 		break;
2541 	default:
2542 		KVM_BUG_ON(1, vcpu->kvm);
2543 		break;
2544 	}
2545 }
2546 
2547 /*
2548  * There is no X86_FEATURE for SGX yet, but anyway we need to query CPUID
2549  * directly instead of going through cpu_has(), to ensure KVM is trapping
2550  * ENCLS whenever it's supported in hardware.  It does not matter whether
2551  * the host OS supports or has enabled SGX.
2552  */
2553 static bool cpu_has_sgx(void)
2554 {
2555 	return cpuid_eax(0) >= 0x12 && (cpuid_eax(0x12) & BIT(0));
2556 }
2557 
2558 static int adjust_vmx_controls(u32 ctl_min, u32 ctl_opt, u32 msr, u32 *result)
2559 {
2560 	u32 vmx_msr_low, vmx_msr_high;
2561 	u32 ctl = ctl_min | ctl_opt;
2562 
2563 	rdmsr(msr, vmx_msr_low, vmx_msr_high);
2564 
2565 	ctl &= vmx_msr_high; /* bit == 0 in high word ==> must be zero */
2566 	ctl |= vmx_msr_low;  /* bit == 1 in low word  ==> must be one  */
2567 
2568 	/* Ensure minimum (required) set of control bits are supported. */
2569 	if (ctl_min & ~ctl)
2570 		return -EIO;
2571 
2572 	*result = ctl;
2573 	return 0;
2574 }
2575 
2576 static u64 adjust_vmx_controls64(u64 ctl_opt, u32 msr)
2577 {
2578 	u64 allowed;
2579 
2580 	rdmsrl(msr, allowed);
2581 
2582 	return  ctl_opt & allowed;
2583 }
2584 
2585 static int setup_vmcs_config(struct vmcs_config *vmcs_conf,
2586 			     struct vmx_capability *vmx_cap)
2587 {
2588 	u32 _pin_based_exec_control = 0;
2589 	u32 _cpu_based_exec_control = 0;
2590 	u32 _cpu_based_2nd_exec_control = 0;
2591 	u64 _cpu_based_3rd_exec_control = 0;
2592 	u32 _vmexit_control = 0;
2593 	u32 _vmentry_control = 0;
2594 	u64 basic_msr;
2595 	u64 misc_msr;
2596 	int i;
2597 
2598 	/*
2599 	 * LOAD/SAVE_DEBUG_CONTROLS are absent because both are mandatory.
2600 	 * SAVE_IA32_PAT and SAVE_IA32_EFER are absent because KVM always
2601 	 * intercepts writes to PAT and EFER, i.e. never enables those controls.
2602 	 */
2603 	struct {
2604 		u32 entry_control;
2605 		u32 exit_control;
2606 	} const vmcs_entry_exit_pairs[] = {
2607 		{ VM_ENTRY_LOAD_IA32_PERF_GLOBAL_CTRL,	VM_EXIT_LOAD_IA32_PERF_GLOBAL_CTRL },
2608 		{ VM_ENTRY_LOAD_IA32_PAT,		VM_EXIT_LOAD_IA32_PAT },
2609 		{ VM_ENTRY_LOAD_IA32_EFER,		VM_EXIT_LOAD_IA32_EFER },
2610 		{ VM_ENTRY_LOAD_BNDCFGS,		VM_EXIT_CLEAR_BNDCFGS },
2611 		{ VM_ENTRY_LOAD_IA32_RTIT_CTL,		VM_EXIT_CLEAR_IA32_RTIT_CTL },
2612 	};
2613 
2614 	memset(vmcs_conf, 0, sizeof(*vmcs_conf));
2615 
2616 	if (adjust_vmx_controls(KVM_REQUIRED_VMX_CPU_BASED_VM_EXEC_CONTROL,
2617 				KVM_OPTIONAL_VMX_CPU_BASED_VM_EXEC_CONTROL,
2618 				MSR_IA32_VMX_PROCBASED_CTLS,
2619 				&_cpu_based_exec_control))
2620 		return -EIO;
2621 	if (_cpu_based_exec_control & CPU_BASED_ACTIVATE_SECONDARY_CONTROLS) {
2622 		if (adjust_vmx_controls(KVM_REQUIRED_VMX_SECONDARY_VM_EXEC_CONTROL,
2623 					KVM_OPTIONAL_VMX_SECONDARY_VM_EXEC_CONTROL,
2624 					MSR_IA32_VMX_PROCBASED_CTLS2,
2625 					&_cpu_based_2nd_exec_control))
2626 			return -EIO;
2627 	}
2628 	if (!IS_ENABLED(CONFIG_KVM_INTEL_PROVE_VE))
2629 		_cpu_based_2nd_exec_control &= ~SECONDARY_EXEC_EPT_VIOLATION_VE;
2630 
2631 #ifndef CONFIG_X86_64
2632 	if (!(_cpu_based_2nd_exec_control &
2633 				SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES))
2634 		_cpu_based_exec_control &= ~CPU_BASED_TPR_SHADOW;
2635 #endif
2636 
2637 	if (!(_cpu_based_exec_control & CPU_BASED_TPR_SHADOW))
2638 		_cpu_based_2nd_exec_control &= ~(
2639 				SECONDARY_EXEC_APIC_REGISTER_VIRT |
2640 				SECONDARY_EXEC_VIRTUALIZE_X2APIC_MODE |
2641 				SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY);
2642 
2643 	rdmsr_safe(MSR_IA32_VMX_EPT_VPID_CAP,
2644 		&vmx_cap->ept, &vmx_cap->vpid);
2645 
2646 	if (!(_cpu_based_2nd_exec_control & SECONDARY_EXEC_ENABLE_EPT) &&
2647 	    vmx_cap->ept) {
2648 		pr_warn_once("EPT CAP should not exist if not support "
2649 				"1-setting enable EPT VM-execution control\n");
2650 
2651 		if (error_on_inconsistent_vmcs_config)
2652 			return -EIO;
2653 
2654 		vmx_cap->ept = 0;
2655 		_cpu_based_2nd_exec_control &= ~SECONDARY_EXEC_EPT_VIOLATION_VE;
2656 	}
2657 	if (!(_cpu_based_2nd_exec_control & SECONDARY_EXEC_ENABLE_VPID) &&
2658 	    vmx_cap->vpid) {
2659 		pr_warn_once("VPID CAP should not exist if not support "
2660 				"1-setting enable VPID VM-execution control\n");
2661 
2662 		if (error_on_inconsistent_vmcs_config)
2663 			return -EIO;
2664 
2665 		vmx_cap->vpid = 0;
2666 	}
2667 
2668 	if (!cpu_has_sgx())
2669 		_cpu_based_2nd_exec_control &= ~SECONDARY_EXEC_ENCLS_EXITING;
2670 
2671 	if (_cpu_based_exec_control & CPU_BASED_ACTIVATE_TERTIARY_CONTROLS)
2672 		_cpu_based_3rd_exec_control =
2673 			adjust_vmx_controls64(KVM_OPTIONAL_VMX_TERTIARY_VM_EXEC_CONTROL,
2674 					      MSR_IA32_VMX_PROCBASED_CTLS3);
2675 
2676 	if (adjust_vmx_controls(KVM_REQUIRED_VMX_VM_EXIT_CONTROLS,
2677 				KVM_OPTIONAL_VMX_VM_EXIT_CONTROLS,
2678 				MSR_IA32_VMX_EXIT_CTLS,
2679 				&_vmexit_control))
2680 		return -EIO;
2681 
2682 	if (adjust_vmx_controls(KVM_REQUIRED_VMX_PIN_BASED_VM_EXEC_CONTROL,
2683 				KVM_OPTIONAL_VMX_PIN_BASED_VM_EXEC_CONTROL,
2684 				MSR_IA32_VMX_PINBASED_CTLS,
2685 				&_pin_based_exec_control))
2686 		return -EIO;
2687 
2688 	if (cpu_has_broken_vmx_preemption_timer())
2689 		_pin_based_exec_control &= ~PIN_BASED_VMX_PREEMPTION_TIMER;
2690 	if (!(_cpu_based_2nd_exec_control &
2691 		SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY))
2692 		_pin_based_exec_control &= ~PIN_BASED_POSTED_INTR;
2693 
2694 	if (adjust_vmx_controls(KVM_REQUIRED_VMX_VM_ENTRY_CONTROLS,
2695 				KVM_OPTIONAL_VMX_VM_ENTRY_CONTROLS,
2696 				MSR_IA32_VMX_ENTRY_CTLS,
2697 				&_vmentry_control))
2698 		return -EIO;
2699 
2700 	for (i = 0; i < ARRAY_SIZE(vmcs_entry_exit_pairs); i++) {
2701 		u32 n_ctrl = vmcs_entry_exit_pairs[i].entry_control;
2702 		u32 x_ctrl = vmcs_entry_exit_pairs[i].exit_control;
2703 
2704 		if (!(_vmentry_control & n_ctrl) == !(_vmexit_control & x_ctrl))
2705 			continue;
2706 
2707 		pr_warn_once("Inconsistent VM-Entry/VM-Exit pair, entry = %x, exit = %x\n",
2708 			     _vmentry_control & n_ctrl, _vmexit_control & x_ctrl);
2709 
2710 		if (error_on_inconsistent_vmcs_config)
2711 			return -EIO;
2712 
2713 		_vmentry_control &= ~n_ctrl;
2714 		_vmexit_control &= ~x_ctrl;
2715 	}
2716 
2717 	/*
2718 	 * Some cpus support VM_{ENTRY,EXIT}_IA32_PERF_GLOBAL_CTRL but they
2719 	 * can't be used due to an errata where VM Exit may incorrectly clear
2720 	 * IA32_PERF_GLOBAL_CTRL[34:32].  Workaround the errata by using the
2721 	 * MSR load mechanism to switch IA32_PERF_GLOBAL_CTRL.
2722 	 */
2723 	switch (boot_cpu_data.x86_vfm) {
2724 	case INTEL_NEHALEM_EP:	/* AAK155 */
2725 	case INTEL_NEHALEM:	/* AAP115 */
2726 	case INTEL_WESTMERE:	/* AAT100 */
2727 	case INTEL_WESTMERE_EP:	/* BC86,AAY89,BD102 */
2728 	case INTEL_NEHALEM_EX:	/* BA97 */
2729 		_vmentry_control &= ~VM_ENTRY_LOAD_IA32_PERF_GLOBAL_CTRL;
2730 		_vmexit_control &= ~VM_EXIT_LOAD_IA32_PERF_GLOBAL_CTRL;
2731 		pr_warn_once("VM_EXIT_LOAD_IA32_PERF_GLOBAL_CTRL "
2732 			     "does not work properly. Using workaround\n");
2733 		break;
2734 	default:
2735 		break;
2736 	}
2737 
2738 	rdmsrl(MSR_IA32_VMX_BASIC, basic_msr);
2739 
2740 	/* IA-32 SDM Vol 3B: VMCS size is never greater than 4kB. */
2741 	if (vmx_basic_vmcs_size(basic_msr) > PAGE_SIZE)
2742 		return -EIO;
2743 
2744 #ifdef CONFIG_X86_64
2745 	/*
2746 	 * KVM expects to be able to shove all legal physical addresses into
2747 	 * VMCS fields for 64-bit kernels, and per the SDM, "This bit is always
2748 	 * 0 for processors that support Intel 64 architecture".
2749 	 */
2750 	if (basic_msr & VMX_BASIC_32BIT_PHYS_ADDR_ONLY)
2751 		return -EIO;
2752 #endif
2753 
2754 	/* Require Write-Back (WB) memory type for VMCS accesses. */
2755 	if (vmx_basic_vmcs_mem_type(basic_msr) != X86_MEMTYPE_WB)
2756 		return -EIO;
2757 
2758 	rdmsrl(MSR_IA32_VMX_MISC, misc_msr);
2759 
2760 	vmcs_conf->basic = basic_msr;
2761 	vmcs_conf->pin_based_exec_ctrl = _pin_based_exec_control;
2762 	vmcs_conf->cpu_based_exec_ctrl = _cpu_based_exec_control;
2763 	vmcs_conf->cpu_based_2nd_exec_ctrl = _cpu_based_2nd_exec_control;
2764 	vmcs_conf->cpu_based_3rd_exec_ctrl = _cpu_based_3rd_exec_control;
2765 	vmcs_conf->vmexit_ctrl         = _vmexit_control;
2766 	vmcs_conf->vmentry_ctrl        = _vmentry_control;
2767 	vmcs_conf->misc	= misc_msr;
2768 
2769 #if IS_ENABLED(CONFIG_HYPERV)
2770 	if (enlightened_vmcs)
2771 		evmcs_sanitize_exec_ctrls(vmcs_conf);
2772 #endif
2773 
2774 	return 0;
2775 }
2776 
2777 static bool __kvm_is_vmx_supported(void)
2778 {
2779 	int cpu = smp_processor_id();
2780 
2781 	if (!(cpuid_ecx(1) & feature_bit(VMX))) {
2782 		pr_err("VMX not supported by CPU %d\n", cpu);
2783 		return false;
2784 	}
2785 
2786 	if (!this_cpu_has(X86_FEATURE_MSR_IA32_FEAT_CTL) ||
2787 	    !this_cpu_has(X86_FEATURE_VMX)) {
2788 		pr_err("VMX not enabled (by BIOS) in MSR_IA32_FEAT_CTL on CPU %d\n", cpu);
2789 		return false;
2790 	}
2791 
2792 	return true;
2793 }
2794 
2795 static bool kvm_is_vmx_supported(void)
2796 {
2797 	bool supported;
2798 
2799 	migrate_disable();
2800 	supported = __kvm_is_vmx_supported();
2801 	migrate_enable();
2802 
2803 	return supported;
2804 }
2805 
2806 int vmx_check_processor_compat(void)
2807 {
2808 	int cpu = raw_smp_processor_id();
2809 	struct vmcs_config vmcs_conf;
2810 	struct vmx_capability vmx_cap;
2811 
2812 	if (!__kvm_is_vmx_supported())
2813 		return -EIO;
2814 
2815 	if (setup_vmcs_config(&vmcs_conf, &vmx_cap) < 0) {
2816 		pr_err("Failed to setup VMCS config on CPU %d\n", cpu);
2817 		return -EIO;
2818 	}
2819 	if (nested)
2820 		nested_vmx_setup_ctls_msrs(&vmcs_conf, vmx_cap.ept);
2821 	if (memcmp(&vmcs_config, &vmcs_conf, sizeof(struct vmcs_config))) {
2822 		pr_err("Inconsistent VMCS config on CPU %d\n", cpu);
2823 		return -EIO;
2824 	}
2825 	return 0;
2826 }
2827 
2828 static int kvm_cpu_vmxon(u64 vmxon_pointer)
2829 {
2830 	u64 msr;
2831 
2832 	cr4_set_bits(X86_CR4_VMXE);
2833 
2834 	asm goto("1: vmxon %[vmxon_pointer]\n\t"
2835 			  _ASM_EXTABLE(1b, %l[fault])
2836 			  : : [vmxon_pointer] "m"(vmxon_pointer)
2837 			  : : fault);
2838 	return 0;
2839 
2840 fault:
2841 	WARN_ONCE(1, "VMXON faulted, MSR_IA32_FEAT_CTL (0x3a) = 0x%llx\n",
2842 		  rdmsrl_safe(MSR_IA32_FEAT_CTL, &msr) ? 0xdeadbeef : msr);
2843 	cr4_clear_bits(X86_CR4_VMXE);
2844 
2845 	return -EFAULT;
2846 }
2847 
2848 int vmx_enable_virtualization_cpu(void)
2849 {
2850 	int cpu = raw_smp_processor_id();
2851 	u64 phys_addr = __pa(per_cpu(vmxarea, cpu));
2852 	int r;
2853 
2854 	if (cr4_read_shadow() & X86_CR4_VMXE)
2855 		return -EBUSY;
2856 
2857 	/*
2858 	 * This can happen if we hot-added a CPU but failed to allocate
2859 	 * VP assist page for it.
2860 	 */
2861 	if (kvm_is_using_evmcs() && !hv_get_vp_assist_page(cpu))
2862 		return -EFAULT;
2863 
2864 	intel_pt_handle_vmx(1);
2865 
2866 	r = kvm_cpu_vmxon(phys_addr);
2867 	if (r) {
2868 		intel_pt_handle_vmx(0);
2869 		return r;
2870 	}
2871 
2872 	return 0;
2873 }
2874 
2875 static void vmclear_local_loaded_vmcss(void)
2876 {
2877 	int cpu = raw_smp_processor_id();
2878 	struct loaded_vmcs *v, *n;
2879 
2880 	list_for_each_entry_safe(v, n, &per_cpu(loaded_vmcss_on_cpu, cpu),
2881 				 loaded_vmcss_on_cpu_link)
2882 		__loaded_vmcs_clear(v);
2883 }
2884 
2885 void vmx_disable_virtualization_cpu(void)
2886 {
2887 	vmclear_local_loaded_vmcss();
2888 
2889 	if (kvm_cpu_vmxoff())
2890 		kvm_spurious_fault();
2891 
2892 	hv_reset_evmcs();
2893 
2894 	intel_pt_handle_vmx(0);
2895 }
2896 
2897 struct vmcs *alloc_vmcs_cpu(bool shadow, int cpu, gfp_t flags)
2898 {
2899 	int node = cpu_to_node(cpu);
2900 	struct page *pages;
2901 	struct vmcs *vmcs;
2902 
2903 	pages = __alloc_pages_node(node, flags, 0);
2904 	if (!pages)
2905 		return NULL;
2906 	vmcs = page_address(pages);
2907 	memset(vmcs, 0, vmx_basic_vmcs_size(vmcs_config.basic));
2908 
2909 	/* KVM supports Enlightened VMCS v1 only */
2910 	if (kvm_is_using_evmcs())
2911 		vmcs->hdr.revision_id = KVM_EVMCS_VERSION;
2912 	else
2913 		vmcs->hdr.revision_id = vmx_basic_vmcs_revision_id(vmcs_config.basic);
2914 
2915 	if (shadow)
2916 		vmcs->hdr.shadow_vmcs = 1;
2917 	return vmcs;
2918 }
2919 
2920 void free_vmcs(struct vmcs *vmcs)
2921 {
2922 	free_page((unsigned long)vmcs);
2923 }
2924 
2925 /*
2926  * Free a VMCS, but before that VMCLEAR it on the CPU where it was last loaded
2927  */
2928 void free_loaded_vmcs(struct loaded_vmcs *loaded_vmcs)
2929 {
2930 	if (!loaded_vmcs->vmcs)
2931 		return;
2932 	loaded_vmcs_clear(loaded_vmcs);
2933 	free_vmcs(loaded_vmcs->vmcs);
2934 	loaded_vmcs->vmcs = NULL;
2935 	if (loaded_vmcs->msr_bitmap)
2936 		free_page((unsigned long)loaded_vmcs->msr_bitmap);
2937 	WARN_ON(loaded_vmcs->shadow_vmcs != NULL);
2938 }
2939 
2940 int alloc_loaded_vmcs(struct loaded_vmcs *loaded_vmcs)
2941 {
2942 	loaded_vmcs->vmcs = alloc_vmcs(false);
2943 	if (!loaded_vmcs->vmcs)
2944 		return -ENOMEM;
2945 
2946 	vmcs_clear(loaded_vmcs->vmcs);
2947 
2948 	loaded_vmcs->shadow_vmcs = NULL;
2949 	loaded_vmcs->hv_timer_soft_disabled = false;
2950 	loaded_vmcs->cpu = -1;
2951 	loaded_vmcs->launched = 0;
2952 
2953 	if (cpu_has_vmx_msr_bitmap()) {
2954 		loaded_vmcs->msr_bitmap = (unsigned long *)
2955 				__get_free_page(GFP_KERNEL_ACCOUNT);
2956 		if (!loaded_vmcs->msr_bitmap)
2957 			goto out_vmcs;
2958 		memset(loaded_vmcs->msr_bitmap, 0xff, PAGE_SIZE);
2959 	}
2960 
2961 	memset(&loaded_vmcs->host_state, 0, sizeof(struct vmcs_host_state));
2962 	memset(&loaded_vmcs->controls_shadow, 0,
2963 		sizeof(struct vmcs_controls_shadow));
2964 
2965 	return 0;
2966 
2967 out_vmcs:
2968 	free_loaded_vmcs(loaded_vmcs);
2969 	return -ENOMEM;
2970 }
2971 
2972 static void free_kvm_area(void)
2973 {
2974 	int cpu;
2975 
2976 	for_each_possible_cpu(cpu) {
2977 		free_vmcs(per_cpu(vmxarea, cpu));
2978 		per_cpu(vmxarea, cpu) = NULL;
2979 	}
2980 }
2981 
2982 static __init int alloc_kvm_area(void)
2983 {
2984 	int cpu;
2985 
2986 	for_each_possible_cpu(cpu) {
2987 		struct vmcs *vmcs;
2988 
2989 		vmcs = alloc_vmcs_cpu(false, cpu, GFP_KERNEL);
2990 		if (!vmcs) {
2991 			free_kvm_area();
2992 			return -ENOMEM;
2993 		}
2994 
2995 		/*
2996 		 * When eVMCS is enabled, alloc_vmcs_cpu() sets
2997 		 * vmcs->revision_id to KVM_EVMCS_VERSION instead of
2998 		 * revision_id reported by MSR_IA32_VMX_BASIC.
2999 		 *
3000 		 * However, even though not explicitly documented by
3001 		 * TLFS, VMXArea passed as VMXON argument should
3002 		 * still be marked with revision_id reported by
3003 		 * physical CPU.
3004 		 */
3005 		if (kvm_is_using_evmcs())
3006 			vmcs->hdr.revision_id = vmx_basic_vmcs_revision_id(vmcs_config.basic);
3007 
3008 		per_cpu(vmxarea, cpu) = vmcs;
3009 	}
3010 	return 0;
3011 }
3012 
3013 static void fix_pmode_seg(struct kvm_vcpu *vcpu, int seg,
3014 		struct kvm_segment *save)
3015 {
3016 	if (!emulate_invalid_guest_state) {
3017 		/*
3018 		 * CS and SS RPL should be equal during guest entry according
3019 		 * to VMX spec, but in reality it is not always so. Since vcpu
3020 		 * is in the middle of the transition from real mode to
3021 		 * protected mode it is safe to assume that RPL 0 is a good
3022 		 * default value.
3023 		 */
3024 		if (seg == VCPU_SREG_CS || seg == VCPU_SREG_SS)
3025 			save->selector &= ~SEGMENT_RPL_MASK;
3026 		save->dpl = save->selector & SEGMENT_RPL_MASK;
3027 		save->s = 1;
3028 	}
3029 	__vmx_set_segment(vcpu, save, seg);
3030 }
3031 
3032 static void enter_pmode(struct kvm_vcpu *vcpu)
3033 {
3034 	unsigned long flags;
3035 	struct vcpu_vmx *vmx = to_vmx(vcpu);
3036 
3037 	/*
3038 	 * Update real mode segment cache. It may be not up-to-date if segment
3039 	 * register was written while vcpu was in a guest mode.
3040 	 */
3041 	vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_ES], VCPU_SREG_ES);
3042 	vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_DS], VCPU_SREG_DS);
3043 	vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_FS], VCPU_SREG_FS);
3044 	vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_GS], VCPU_SREG_GS);
3045 	vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_SS], VCPU_SREG_SS);
3046 	vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_CS], VCPU_SREG_CS);
3047 
3048 	vmx->rmode.vm86_active = 0;
3049 
3050 	__vmx_set_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_TR], VCPU_SREG_TR);
3051 
3052 	flags = vmcs_readl(GUEST_RFLAGS);
3053 	flags &= RMODE_GUEST_OWNED_EFLAGS_BITS;
3054 	flags |= vmx->rmode.save_rflags & ~RMODE_GUEST_OWNED_EFLAGS_BITS;
3055 	vmcs_writel(GUEST_RFLAGS, flags);
3056 
3057 	vmcs_writel(GUEST_CR4, (vmcs_readl(GUEST_CR4) & ~X86_CR4_VME) |
3058 			(vmcs_readl(CR4_READ_SHADOW) & X86_CR4_VME));
3059 
3060 	vmx_update_exception_bitmap(vcpu);
3061 
3062 	fix_pmode_seg(vcpu, VCPU_SREG_CS, &vmx->rmode.segs[VCPU_SREG_CS]);
3063 	fix_pmode_seg(vcpu, VCPU_SREG_SS, &vmx->rmode.segs[VCPU_SREG_SS]);
3064 	fix_pmode_seg(vcpu, VCPU_SREG_ES, &vmx->rmode.segs[VCPU_SREG_ES]);
3065 	fix_pmode_seg(vcpu, VCPU_SREG_DS, &vmx->rmode.segs[VCPU_SREG_DS]);
3066 	fix_pmode_seg(vcpu, VCPU_SREG_FS, &vmx->rmode.segs[VCPU_SREG_FS]);
3067 	fix_pmode_seg(vcpu, VCPU_SREG_GS, &vmx->rmode.segs[VCPU_SREG_GS]);
3068 }
3069 
3070 static void fix_rmode_seg(int seg, struct kvm_segment *save)
3071 {
3072 	const struct kvm_vmx_segment_field *sf = &kvm_vmx_segment_fields[seg];
3073 	struct kvm_segment var = *save;
3074 
3075 	var.dpl = 0x3;
3076 	if (seg == VCPU_SREG_CS)
3077 		var.type = 0x3;
3078 
3079 	if (!emulate_invalid_guest_state) {
3080 		var.selector = var.base >> 4;
3081 		var.base = var.base & 0xffff0;
3082 		var.limit = 0xffff;
3083 		var.g = 0;
3084 		var.db = 0;
3085 		var.present = 1;
3086 		var.s = 1;
3087 		var.l = 0;
3088 		var.unusable = 0;
3089 		var.type = 0x3;
3090 		var.avl = 0;
3091 		if (save->base & 0xf)
3092 			pr_warn_once("segment base is not paragraph aligned "
3093 				     "when entering protected mode (seg=%d)", seg);
3094 	}
3095 
3096 	vmcs_write16(sf->selector, var.selector);
3097 	vmcs_writel(sf->base, var.base);
3098 	vmcs_write32(sf->limit, var.limit);
3099 	vmcs_write32(sf->ar_bytes, vmx_segment_access_rights(&var));
3100 }
3101 
3102 static void enter_rmode(struct kvm_vcpu *vcpu)
3103 {
3104 	unsigned long flags;
3105 	struct vcpu_vmx *vmx = to_vmx(vcpu);
3106 	struct kvm_vmx *kvm_vmx = to_kvm_vmx(vcpu->kvm);
3107 
3108 	/*
3109 	 * KVM should never use VM86 to virtualize Real Mode when L2 is active,
3110 	 * as using VM86 is unnecessary if unrestricted guest is enabled, and
3111 	 * if unrestricted guest is disabled, VM-Enter (from L1) with CR0.PG=0
3112 	 * should VM-Fail and KVM should reject userspace attempts to stuff
3113 	 * CR0.PG=0 when L2 is active.
3114 	 */
3115 	WARN_ON_ONCE(is_guest_mode(vcpu));
3116 
3117 	vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_TR], VCPU_SREG_TR);
3118 	vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_ES], VCPU_SREG_ES);
3119 	vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_DS], VCPU_SREG_DS);
3120 	vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_FS], VCPU_SREG_FS);
3121 	vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_GS], VCPU_SREG_GS);
3122 	vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_SS], VCPU_SREG_SS);
3123 	vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_CS], VCPU_SREG_CS);
3124 
3125 	vmx->rmode.vm86_active = 1;
3126 
3127 	vmx_segment_cache_clear(vmx);
3128 
3129 	vmcs_writel(GUEST_TR_BASE, kvm_vmx->tss_addr);
3130 	vmcs_write32(GUEST_TR_LIMIT, RMODE_TSS_SIZE - 1);
3131 	vmcs_write32(GUEST_TR_AR_BYTES, 0x008b);
3132 
3133 	flags = vmcs_readl(GUEST_RFLAGS);
3134 	vmx->rmode.save_rflags = flags;
3135 
3136 	flags |= X86_EFLAGS_IOPL | X86_EFLAGS_VM;
3137 
3138 	vmcs_writel(GUEST_RFLAGS, flags);
3139 	vmcs_writel(GUEST_CR4, vmcs_readl(GUEST_CR4) | X86_CR4_VME);
3140 	vmx_update_exception_bitmap(vcpu);
3141 
3142 	fix_rmode_seg(VCPU_SREG_SS, &vmx->rmode.segs[VCPU_SREG_SS]);
3143 	fix_rmode_seg(VCPU_SREG_CS, &vmx->rmode.segs[VCPU_SREG_CS]);
3144 	fix_rmode_seg(VCPU_SREG_ES, &vmx->rmode.segs[VCPU_SREG_ES]);
3145 	fix_rmode_seg(VCPU_SREG_DS, &vmx->rmode.segs[VCPU_SREG_DS]);
3146 	fix_rmode_seg(VCPU_SREG_GS, &vmx->rmode.segs[VCPU_SREG_GS]);
3147 	fix_rmode_seg(VCPU_SREG_FS, &vmx->rmode.segs[VCPU_SREG_FS]);
3148 }
3149 
3150 int vmx_set_efer(struct kvm_vcpu *vcpu, u64 efer)
3151 {
3152 	struct vcpu_vmx *vmx = to_vmx(vcpu);
3153 
3154 	/* Nothing to do if hardware doesn't support EFER. */
3155 	if (!vmx_find_uret_msr(vmx, MSR_EFER))
3156 		return 0;
3157 
3158 	vcpu->arch.efer = efer;
3159 #ifdef CONFIG_X86_64
3160 	if (efer & EFER_LMA)
3161 		vm_entry_controls_setbit(vmx, VM_ENTRY_IA32E_MODE);
3162 	else
3163 		vm_entry_controls_clearbit(vmx, VM_ENTRY_IA32E_MODE);
3164 #else
3165 	if (KVM_BUG_ON(efer & EFER_LMA, vcpu->kvm))
3166 		return 1;
3167 #endif
3168 
3169 	vmx_setup_uret_msrs(vmx);
3170 	return 0;
3171 }
3172 
3173 #ifdef CONFIG_X86_64
3174 
3175 static void enter_lmode(struct kvm_vcpu *vcpu)
3176 {
3177 	u32 guest_tr_ar;
3178 
3179 	vmx_segment_cache_clear(to_vmx(vcpu));
3180 
3181 	guest_tr_ar = vmcs_read32(GUEST_TR_AR_BYTES);
3182 	if ((guest_tr_ar & VMX_AR_TYPE_MASK) != VMX_AR_TYPE_BUSY_64_TSS) {
3183 		pr_debug_ratelimited("%s: tss fixup for long mode. \n",
3184 				     __func__);
3185 		vmcs_write32(GUEST_TR_AR_BYTES,
3186 			     (guest_tr_ar & ~VMX_AR_TYPE_MASK)
3187 			     | VMX_AR_TYPE_BUSY_64_TSS);
3188 	}
3189 	vmx_set_efer(vcpu, vcpu->arch.efer | EFER_LMA);
3190 }
3191 
3192 static void exit_lmode(struct kvm_vcpu *vcpu)
3193 {
3194 	vmx_set_efer(vcpu, vcpu->arch.efer & ~EFER_LMA);
3195 }
3196 
3197 #endif
3198 
3199 void vmx_flush_tlb_all(struct kvm_vcpu *vcpu)
3200 {
3201 	struct vcpu_vmx *vmx = to_vmx(vcpu);
3202 
3203 	/*
3204 	 * INVEPT must be issued when EPT is enabled, irrespective of VPID, as
3205 	 * the CPU is not required to invalidate guest-physical mappings on
3206 	 * VM-Entry, even if VPID is disabled.  Guest-physical mappings are
3207 	 * associated with the root EPT structure and not any particular VPID
3208 	 * (INVVPID also isn't required to invalidate guest-physical mappings).
3209 	 */
3210 	if (enable_ept) {
3211 		ept_sync_global();
3212 	} else if (enable_vpid) {
3213 		if (cpu_has_vmx_invvpid_global()) {
3214 			vpid_sync_vcpu_global();
3215 		} else {
3216 			vpid_sync_vcpu_single(vmx->vpid);
3217 			vpid_sync_vcpu_single(vmx->nested.vpid02);
3218 		}
3219 	}
3220 }
3221 
3222 static inline int vmx_get_current_vpid(struct kvm_vcpu *vcpu)
3223 {
3224 	if (is_guest_mode(vcpu) && nested_cpu_has_vpid(get_vmcs12(vcpu)))
3225 		return nested_get_vpid02(vcpu);
3226 	return to_vmx(vcpu)->vpid;
3227 }
3228 
3229 void vmx_flush_tlb_current(struct kvm_vcpu *vcpu)
3230 {
3231 	struct kvm_mmu *mmu = vcpu->arch.mmu;
3232 	u64 root_hpa = mmu->root.hpa;
3233 
3234 	/* No flush required if the current context is invalid. */
3235 	if (!VALID_PAGE(root_hpa))
3236 		return;
3237 
3238 	if (enable_ept)
3239 		ept_sync_context(construct_eptp(vcpu, root_hpa,
3240 						mmu->root_role.level));
3241 	else
3242 		vpid_sync_context(vmx_get_current_vpid(vcpu));
3243 }
3244 
3245 void vmx_flush_tlb_gva(struct kvm_vcpu *vcpu, gva_t addr)
3246 {
3247 	/*
3248 	 * vpid_sync_vcpu_addr() is a nop if vpid==0, see the comment in
3249 	 * vmx_flush_tlb_guest() for an explanation of why this is ok.
3250 	 */
3251 	vpid_sync_vcpu_addr(vmx_get_current_vpid(vcpu), addr);
3252 }
3253 
3254 void vmx_flush_tlb_guest(struct kvm_vcpu *vcpu)
3255 {
3256 	/*
3257 	 * vpid_sync_context() is a nop if vpid==0, e.g. if enable_vpid==0 or a
3258 	 * vpid couldn't be allocated for this vCPU.  VM-Enter and VM-Exit are
3259 	 * required to flush GVA->{G,H}PA mappings from the TLB if vpid is
3260 	 * disabled (VM-Enter with vpid enabled and vpid==0 is disallowed),
3261 	 * i.e. no explicit INVVPID is necessary.
3262 	 */
3263 	vpid_sync_context(vmx_get_current_vpid(vcpu));
3264 }
3265 
3266 void vmx_ept_load_pdptrs(struct kvm_vcpu *vcpu)
3267 {
3268 	struct kvm_mmu *mmu = vcpu->arch.walk_mmu;
3269 
3270 	if (!kvm_register_is_dirty(vcpu, VCPU_EXREG_PDPTR))
3271 		return;
3272 
3273 	if (is_pae_paging(vcpu)) {
3274 		vmcs_write64(GUEST_PDPTR0, mmu->pdptrs[0]);
3275 		vmcs_write64(GUEST_PDPTR1, mmu->pdptrs[1]);
3276 		vmcs_write64(GUEST_PDPTR2, mmu->pdptrs[2]);
3277 		vmcs_write64(GUEST_PDPTR3, mmu->pdptrs[3]);
3278 	}
3279 }
3280 
3281 void ept_save_pdptrs(struct kvm_vcpu *vcpu)
3282 {
3283 	struct kvm_mmu *mmu = vcpu->arch.walk_mmu;
3284 
3285 	if (WARN_ON_ONCE(!is_pae_paging(vcpu)))
3286 		return;
3287 
3288 	mmu->pdptrs[0] = vmcs_read64(GUEST_PDPTR0);
3289 	mmu->pdptrs[1] = vmcs_read64(GUEST_PDPTR1);
3290 	mmu->pdptrs[2] = vmcs_read64(GUEST_PDPTR2);
3291 	mmu->pdptrs[3] = vmcs_read64(GUEST_PDPTR3);
3292 
3293 	kvm_register_mark_available(vcpu, VCPU_EXREG_PDPTR);
3294 }
3295 
3296 #define CR3_EXITING_BITS (CPU_BASED_CR3_LOAD_EXITING | \
3297 			  CPU_BASED_CR3_STORE_EXITING)
3298 
3299 bool vmx_is_valid_cr0(struct kvm_vcpu *vcpu, unsigned long cr0)
3300 {
3301 	if (is_guest_mode(vcpu))
3302 		return nested_guest_cr0_valid(vcpu, cr0);
3303 
3304 	if (to_vmx(vcpu)->nested.vmxon)
3305 		return nested_host_cr0_valid(vcpu, cr0);
3306 
3307 	return true;
3308 }
3309 
3310 void vmx_set_cr0(struct kvm_vcpu *vcpu, unsigned long cr0)
3311 {
3312 	struct vcpu_vmx *vmx = to_vmx(vcpu);
3313 	unsigned long hw_cr0, old_cr0_pg;
3314 	u32 tmp;
3315 
3316 	old_cr0_pg = kvm_read_cr0_bits(vcpu, X86_CR0_PG);
3317 
3318 	hw_cr0 = (cr0 & ~KVM_VM_CR0_ALWAYS_OFF);
3319 	if (enable_unrestricted_guest)
3320 		hw_cr0 |= KVM_VM_CR0_ALWAYS_ON_UNRESTRICTED_GUEST;
3321 	else {
3322 		hw_cr0 |= KVM_VM_CR0_ALWAYS_ON;
3323 		if (!enable_ept)
3324 			hw_cr0 |= X86_CR0_WP;
3325 
3326 		if (vmx->rmode.vm86_active && (cr0 & X86_CR0_PE))
3327 			enter_pmode(vcpu);
3328 
3329 		if (!vmx->rmode.vm86_active && !(cr0 & X86_CR0_PE))
3330 			enter_rmode(vcpu);
3331 	}
3332 
3333 	vmcs_writel(CR0_READ_SHADOW, cr0);
3334 	vmcs_writel(GUEST_CR0, hw_cr0);
3335 	vcpu->arch.cr0 = cr0;
3336 	kvm_register_mark_available(vcpu, VCPU_EXREG_CR0);
3337 
3338 #ifdef CONFIG_X86_64
3339 	if (vcpu->arch.efer & EFER_LME) {
3340 		if (!old_cr0_pg && (cr0 & X86_CR0_PG))
3341 			enter_lmode(vcpu);
3342 		else if (old_cr0_pg && !(cr0 & X86_CR0_PG))
3343 			exit_lmode(vcpu);
3344 	}
3345 #endif
3346 
3347 	if (enable_ept && !enable_unrestricted_guest) {
3348 		/*
3349 		 * Ensure KVM has an up-to-date snapshot of the guest's CR3.  If
3350 		 * the below code _enables_ CR3 exiting, vmx_cache_reg() will
3351 		 * (correctly) stop reading vmcs.GUEST_CR3 because it thinks
3352 		 * KVM's CR3 is installed.
3353 		 */
3354 		if (!kvm_register_is_available(vcpu, VCPU_EXREG_CR3))
3355 			vmx_cache_reg(vcpu, VCPU_EXREG_CR3);
3356 
3357 		/*
3358 		 * When running with EPT but not unrestricted guest, KVM must
3359 		 * intercept CR3 accesses when paging is _disabled_.  This is
3360 		 * necessary because restricted guests can't actually run with
3361 		 * paging disabled, and so KVM stuffs its own CR3 in order to
3362 		 * run the guest when identity mapped page tables.
3363 		 *
3364 		 * Do _NOT_ check the old CR0.PG, e.g. to optimize away the
3365 		 * update, it may be stale with respect to CR3 interception,
3366 		 * e.g. after nested VM-Enter.
3367 		 *
3368 		 * Lastly, honor L1's desires, i.e. intercept CR3 loads and/or
3369 		 * stores to forward them to L1, even if KVM does not need to
3370 		 * intercept them to preserve its identity mapped page tables.
3371 		 */
3372 		if (!(cr0 & X86_CR0_PG)) {
3373 			exec_controls_setbit(vmx, CR3_EXITING_BITS);
3374 		} else if (!is_guest_mode(vcpu)) {
3375 			exec_controls_clearbit(vmx, CR3_EXITING_BITS);
3376 		} else {
3377 			tmp = exec_controls_get(vmx);
3378 			tmp &= ~CR3_EXITING_BITS;
3379 			tmp |= get_vmcs12(vcpu)->cpu_based_vm_exec_control & CR3_EXITING_BITS;
3380 			exec_controls_set(vmx, tmp);
3381 		}
3382 
3383 		/* Note, vmx_set_cr4() consumes the new vcpu->arch.cr0. */
3384 		if ((old_cr0_pg ^ cr0) & X86_CR0_PG)
3385 			vmx_set_cr4(vcpu, kvm_read_cr4(vcpu));
3386 
3387 		/*
3388 		 * When !CR0_PG -> CR0_PG, vcpu->arch.cr3 becomes active, but
3389 		 * GUEST_CR3 is still vmx->ept_identity_map_addr if EPT + !URG.
3390 		 */
3391 		if (!(old_cr0_pg & X86_CR0_PG) && (cr0 & X86_CR0_PG))
3392 			kvm_register_mark_dirty(vcpu, VCPU_EXREG_CR3);
3393 	}
3394 
3395 	/* depends on vcpu->arch.cr0 to be set to a new value */
3396 	vmx->emulation_required = vmx_emulation_required(vcpu);
3397 }
3398 
3399 static int vmx_get_max_ept_level(void)
3400 {
3401 	if (cpu_has_vmx_ept_5levels())
3402 		return 5;
3403 	return 4;
3404 }
3405 
3406 u64 construct_eptp(struct kvm_vcpu *vcpu, hpa_t root_hpa, int root_level)
3407 {
3408 	u64 eptp = VMX_EPTP_MT_WB;
3409 
3410 	eptp |= (root_level == 5) ? VMX_EPTP_PWL_5 : VMX_EPTP_PWL_4;
3411 
3412 	if (enable_ept_ad_bits &&
3413 	    (!is_guest_mode(vcpu) || nested_ept_ad_enabled(vcpu)))
3414 		eptp |= VMX_EPTP_AD_ENABLE_BIT;
3415 	eptp |= root_hpa;
3416 
3417 	return eptp;
3418 }
3419 
3420 void vmx_load_mmu_pgd(struct kvm_vcpu *vcpu, hpa_t root_hpa, int root_level)
3421 {
3422 	struct kvm *kvm = vcpu->kvm;
3423 	bool update_guest_cr3 = true;
3424 	unsigned long guest_cr3;
3425 	u64 eptp;
3426 
3427 	if (enable_ept) {
3428 		eptp = construct_eptp(vcpu, root_hpa, root_level);
3429 		vmcs_write64(EPT_POINTER, eptp);
3430 
3431 		hv_track_root_tdp(vcpu, root_hpa);
3432 
3433 		if (!enable_unrestricted_guest && !is_paging(vcpu))
3434 			guest_cr3 = to_kvm_vmx(kvm)->ept_identity_map_addr;
3435 		else if (kvm_register_is_dirty(vcpu, VCPU_EXREG_CR3))
3436 			guest_cr3 = vcpu->arch.cr3;
3437 		else /* vmcs.GUEST_CR3 is already up-to-date. */
3438 			update_guest_cr3 = false;
3439 		vmx_ept_load_pdptrs(vcpu);
3440 	} else {
3441 		guest_cr3 = root_hpa | kvm_get_active_pcid(vcpu) |
3442 			    kvm_get_active_cr3_lam_bits(vcpu);
3443 	}
3444 
3445 	if (update_guest_cr3)
3446 		vmcs_writel(GUEST_CR3, guest_cr3);
3447 }
3448 
3449 bool vmx_is_valid_cr4(struct kvm_vcpu *vcpu, unsigned long cr4)
3450 {
3451 	/*
3452 	 * We operate under the default treatment of SMM, so VMX cannot be
3453 	 * enabled under SMM.  Note, whether or not VMXE is allowed at all,
3454 	 * i.e. is a reserved bit, is handled by common x86 code.
3455 	 */
3456 	if ((cr4 & X86_CR4_VMXE) && is_smm(vcpu))
3457 		return false;
3458 
3459 	if (to_vmx(vcpu)->nested.vmxon && !nested_cr4_valid(vcpu, cr4))
3460 		return false;
3461 
3462 	return true;
3463 }
3464 
3465 void vmx_set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4)
3466 {
3467 	unsigned long old_cr4 = kvm_read_cr4(vcpu);
3468 	struct vcpu_vmx *vmx = to_vmx(vcpu);
3469 	unsigned long hw_cr4;
3470 
3471 	/*
3472 	 * Pass through host's Machine Check Enable value to hw_cr4, which
3473 	 * is in force while we are in guest mode.  Do not let guests control
3474 	 * this bit, even if host CR4.MCE == 0.
3475 	 */
3476 	hw_cr4 = (cr4_read_shadow() & X86_CR4_MCE) | (cr4 & ~X86_CR4_MCE);
3477 	if (enable_unrestricted_guest)
3478 		hw_cr4 |= KVM_VM_CR4_ALWAYS_ON_UNRESTRICTED_GUEST;
3479 	else if (vmx->rmode.vm86_active)
3480 		hw_cr4 |= KVM_RMODE_VM_CR4_ALWAYS_ON;
3481 	else
3482 		hw_cr4 |= KVM_PMODE_VM_CR4_ALWAYS_ON;
3483 
3484 	if (vmx_umip_emulated()) {
3485 		if (cr4 & X86_CR4_UMIP) {
3486 			secondary_exec_controls_setbit(vmx, SECONDARY_EXEC_DESC);
3487 			hw_cr4 &= ~X86_CR4_UMIP;
3488 		} else if (!is_guest_mode(vcpu) ||
3489 			!nested_cpu_has2(get_vmcs12(vcpu), SECONDARY_EXEC_DESC)) {
3490 			secondary_exec_controls_clearbit(vmx, SECONDARY_EXEC_DESC);
3491 		}
3492 	}
3493 
3494 	vcpu->arch.cr4 = cr4;
3495 	kvm_register_mark_available(vcpu, VCPU_EXREG_CR4);
3496 
3497 	if (!enable_unrestricted_guest) {
3498 		if (enable_ept) {
3499 			if (!is_paging(vcpu)) {
3500 				hw_cr4 &= ~X86_CR4_PAE;
3501 				hw_cr4 |= X86_CR4_PSE;
3502 			} else if (!(cr4 & X86_CR4_PAE)) {
3503 				hw_cr4 &= ~X86_CR4_PAE;
3504 			}
3505 		}
3506 
3507 		/*
3508 		 * SMEP/SMAP/PKU is disabled if CPU is in non-paging mode in
3509 		 * hardware.  To emulate this behavior, SMEP/SMAP/PKU needs
3510 		 * to be manually disabled when guest switches to non-paging
3511 		 * mode.
3512 		 *
3513 		 * If !enable_unrestricted_guest, the CPU is always running
3514 		 * with CR0.PG=1 and CR4 needs to be modified.
3515 		 * If enable_unrestricted_guest, the CPU automatically
3516 		 * disables SMEP/SMAP/PKU when the guest sets CR0.PG=0.
3517 		 */
3518 		if (!is_paging(vcpu))
3519 			hw_cr4 &= ~(X86_CR4_SMEP | X86_CR4_SMAP | X86_CR4_PKE);
3520 	}
3521 
3522 	vmcs_writel(CR4_READ_SHADOW, cr4);
3523 	vmcs_writel(GUEST_CR4, hw_cr4);
3524 
3525 	if ((cr4 ^ old_cr4) & (X86_CR4_OSXSAVE | X86_CR4_PKE))
3526 		kvm_update_cpuid_runtime(vcpu);
3527 }
3528 
3529 void vmx_get_segment(struct kvm_vcpu *vcpu, struct kvm_segment *var, int seg)
3530 {
3531 	struct vcpu_vmx *vmx = to_vmx(vcpu);
3532 	u32 ar;
3533 
3534 	if (vmx->rmode.vm86_active && seg != VCPU_SREG_LDTR) {
3535 		*var = vmx->rmode.segs[seg];
3536 		if (seg == VCPU_SREG_TR
3537 		    || var->selector == vmx_read_guest_seg_selector(vmx, seg))
3538 			return;
3539 		var->base = vmx_read_guest_seg_base(vmx, seg);
3540 		var->selector = vmx_read_guest_seg_selector(vmx, seg);
3541 		return;
3542 	}
3543 	var->base = vmx_read_guest_seg_base(vmx, seg);
3544 	var->limit = vmx_read_guest_seg_limit(vmx, seg);
3545 	var->selector = vmx_read_guest_seg_selector(vmx, seg);
3546 	ar = vmx_read_guest_seg_ar(vmx, seg);
3547 	var->unusable = (ar >> 16) & 1;
3548 	var->type = ar & 15;
3549 	var->s = (ar >> 4) & 1;
3550 	var->dpl = (ar >> 5) & 3;
3551 	/*
3552 	 * Some userspaces do not preserve unusable property. Since usable
3553 	 * segment has to be present according to VMX spec we can use present
3554 	 * property to amend userspace bug by making unusable segment always
3555 	 * nonpresent. vmx_segment_access_rights() already marks nonpresent
3556 	 * segment as unusable.
3557 	 */
3558 	var->present = !var->unusable;
3559 	var->avl = (ar >> 12) & 1;
3560 	var->l = (ar >> 13) & 1;
3561 	var->db = (ar >> 14) & 1;
3562 	var->g = (ar >> 15) & 1;
3563 }
3564 
3565 u64 vmx_get_segment_base(struct kvm_vcpu *vcpu, int seg)
3566 {
3567 	struct kvm_segment s;
3568 
3569 	if (to_vmx(vcpu)->rmode.vm86_active) {
3570 		vmx_get_segment(vcpu, &s, seg);
3571 		return s.base;
3572 	}
3573 	return vmx_read_guest_seg_base(to_vmx(vcpu), seg);
3574 }
3575 
3576 static int __vmx_get_cpl(struct kvm_vcpu *vcpu, bool no_cache)
3577 {
3578 	struct vcpu_vmx *vmx = to_vmx(vcpu);
3579 	int ar;
3580 
3581 	if (unlikely(vmx->rmode.vm86_active))
3582 		return 0;
3583 
3584 	if (no_cache)
3585 		ar = vmcs_read32(GUEST_SS_AR_BYTES);
3586 	else
3587 		ar = vmx_read_guest_seg_ar(vmx, VCPU_SREG_SS);
3588 	return VMX_AR_DPL(ar);
3589 }
3590 
3591 int vmx_get_cpl(struct kvm_vcpu *vcpu)
3592 {
3593 	return __vmx_get_cpl(vcpu, false);
3594 }
3595 
3596 int vmx_get_cpl_no_cache(struct kvm_vcpu *vcpu)
3597 {
3598 	return __vmx_get_cpl(vcpu, true);
3599 }
3600 
3601 static u32 vmx_segment_access_rights(struct kvm_segment *var)
3602 {
3603 	u32 ar;
3604 
3605 	ar = var->type & 15;
3606 	ar |= (var->s & 1) << 4;
3607 	ar |= (var->dpl & 3) << 5;
3608 	ar |= (var->present & 1) << 7;
3609 	ar |= (var->avl & 1) << 12;
3610 	ar |= (var->l & 1) << 13;
3611 	ar |= (var->db & 1) << 14;
3612 	ar |= (var->g & 1) << 15;
3613 	ar |= (var->unusable || !var->present) << 16;
3614 
3615 	return ar;
3616 }
3617 
3618 void __vmx_set_segment(struct kvm_vcpu *vcpu, struct kvm_segment *var, int seg)
3619 {
3620 	struct vcpu_vmx *vmx = to_vmx(vcpu);
3621 	const struct kvm_vmx_segment_field *sf = &kvm_vmx_segment_fields[seg];
3622 
3623 	vmx_segment_cache_clear(vmx);
3624 
3625 	if (vmx->rmode.vm86_active && seg != VCPU_SREG_LDTR) {
3626 		vmx->rmode.segs[seg] = *var;
3627 		if (seg == VCPU_SREG_TR)
3628 			vmcs_write16(sf->selector, var->selector);
3629 		else if (var->s)
3630 			fix_rmode_seg(seg, &vmx->rmode.segs[seg]);
3631 		return;
3632 	}
3633 
3634 	vmcs_writel(sf->base, var->base);
3635 	vmcs_write32(sf->limit, var->limit);
3636 	vmcs_write16(sf->selector, var->selector);
3637 
3638 	/*
3639 	 *   Fix the "Accessed" bit in AR field of segment registers for older
3640 	 * qemu binaries.
3641 	 *   IA32 arch specifies that at the time of processor reset the
3642 	 * "Accessed" bit in the AR field of segment registers is 1. And qemu
3643 	 * is setting it to 0 in the userland code. This causes invalid guest
3644 	 * state vmexit when "unrestricted guest" mode is turned on.
3645 	 *    Fix for this setup issue in cpu_reset is being pushed in the qemu
3646 	 * tree. Newer qemu binaries with that qemu fix would not need this
3647 	 * kvm hack.
3648 	 */
3649 	if (is_unrestricted_guest(vcpu) && (seg != VCPU_SREG_LDTR))
3650 		var->type |= 0x1; /* Accessed */
3651 
3652 	vmcs_write32(sf->ar_bytes, vmx_segment_access_rights(var));
3653 }
3654 
3655 void vmx_set_segment(struct kvm_vcpu *vcpu, struct kvm_segment *var, int seg)
3656 {
3657 	__vmx_set_segment(vcpu, var, seg);
3658 
3659 	to_vmx(vcpu)->emulation_required = vmx_emulation_required(vcpu);
3660 }
3661 
3662 void vmx_get_cs_db_l_bits(struct kvm_vcpu *vcpu, int *db, int *l)
3663 {
3664 	u32 ar = vmx_read_guest_seg_ar(to_vmx(vcpu), VCPU_SREG_CS);
3665 
3666 	*db = (ar >> 14) & 1;
3667 	*l = (ar >> 13) & 1;
3668 }
3669 
3670 void vmx_get_idt(struct kvm_vcpu *vcpu, struct desc_ptr *dt)
3671 {
3672 	dt->size = vmcs_read32(GUEST_IDTR_LIMIT);
3673 	dt->address = vmcs_readl(GUEST_IDTR_BASE);
3674 }
3675 
3676 void vmx_set_idt(struct kvm_vcpu *vcpu, struct desc_ptr *dt)
3677 {
3678 	vmcs_write32(GUEST_IDTR_LIMIT, dt->size);
3679 	vmcs_writel(GUEST_IDTR_BASE, dt->address);
3680 }
3681 
3682 void vmx_get_gdt(struct kvm_vcpu *vcpu, struct desc_ptr *dt)
3683 {
3684 	dt->size = vmcs_read32(GUEST_GDTR_LIMIT);
3685 	dt->address = vmcs_readl(GUEST_GDTR_BASE);
3686 }
3687 
3688 void vmx_set_gdt(struct kvm_vcpu *vcpu, struct desc_ptr *dt)
3689 {
3690 	vmcs_write32(GUEST_GDTR_LIMIT, dt->size);
3691 	vmcs_writel(GUEST_GDTR_BASE, dt->address);
3692 }
3693 
3694 static bool rmode_segment_valid(struct kvm_vcpu *vcpu, int seg)
3695 {
3696 	struct kvm_segment var;
3697 	u32 ar;
3698 
3699 	vmx_get_segment(vcpu, &var, seg);
3700 	var.dpl = 0x3;
3701 	if (seg == VCPU_SREG_CS)
3702 		var.type = 0x3;
3703 	ar = vmx_segment_access_rights(&var);
3704 
3705 	if (var.base != (var.selector << 4))
3706 		return false;
3707 	if (var.limit != 0xffff)
3708 		return false;
3709 	if (ar != 0xf3)
3710 		return false;
3711 
3712 	return true;
3713 }
3714 
3715 static bool code_segment_valid(struct kvm_vcpu *vcpu)
3716 {
3717 	struct kvm_segment cs;
3718 	unsigned int cs_rpl;
3719 
3720 	vmx_get_segment(vcpu, &cs, VCPU_SREG_CS);
3721 	cs_rpl = cs.selector & SEGMENT_RPL_MASK;
3722 
3723 	if (cs.unusable)
3724 		return false;
3725 	if (~cs.type & (VMX_AR_TYPE_CODE_MASK|VMX_AR_TYPE_ACCESSES_MASK))
3726 		return false;
3727 	if (!cs.s)
3728 		return false;
3729 	if (cs.type & VMX_AR_TYPE_WRITEABLE_MASK) {
3730 		if (cs.dpl > cs_rpl)
3731 			return false;
3732 	} else {
3733 		if (cs.dpl != cs_rpl)
3734 			return false;
3735 	}
3736 	if (!cs.present)
3737 		return false;
3738 
3739 	/* TODO: Add Reserved field check, this'll require a new member in the kvm_segment_field structure */
3740 	return true;
3741 }
3742 
3743 static bool stack_segment_valid(struct kvm_vcpu *vcpu)
3744 {
3745 	struct kvm_segment ss;
3746 	unsigned int ss_rpl;
3747 
3748 	vmx_get_segment(vcpu, &ss, VCPU_SREG_SS);
3749 	ss_rpl = ss.selector & SEGMENT_RPL_MASK;
3750 
3751 	if (ss.unusable)
3752 		return true;
3753 	if (ss.type != 3 && ss.type != 7)
3754 		return false;
3755 	if (!ss.s)
3756 		return false;
3757 	if (ss.dpl != ss_rpl) /* DPL != RPL */
3758 		return false;
3759 	if (!ss.present)
3760 		return false;
3761 
3762 	return true;
3763 }
3764 
3765 static bool data_segment_valid(struct kvm_vcpu *vcpu, int seg)
3766 {
3767 	struct kvm_segment var;
3768 	unsigned int rpl;
3769 
3770 	vmx_get_segment(vcpu, &var, seg);
3771 	rpl = var.selector & SEGMENT_RPL_MASK;
3772 
3773 	if (var.unusable)
3774 		return true;
3775 	if (!var.s)
3776 		return false;
3777 	if (!var.present)
3778 		return false;
3779 	if (~var.type & (VMX_AR_TYPE_CODE_MASK|VMX_AR_TYPE_WRITEABLE_MASK)) {
3780 		if (var.dpl < rpl) /* DPL < RPL */
3781 			return false;
3782 	}
3783 
3784 	/* TODO: Add other members to kvm_segment_field to allow checking for other access
3785 	 * rights flags
3786 	 */
3787 	return true;
3788 }
3789 
3790 static bool tr_valid(struct kvm_vcpu *vcpu)
3791 {
3792 	struct kvm_segment tr;
3793 
3794 	vmx_get_segment(vcpu, &tr, VCPU_SREG_TR);
3795 
3796 	if (tr.unusable)
3797 		return false;
3798 	if (tr.selector & SEGMENT_TI_MASK)	/* TI = 1 */
3799 		return false;
3800 	if (tr.type != 3 && tr.type != 11) /* TODO: Check if guest is in IA32e mode */
3801 		return false;
3802 	if (!tr.present)
3803 		return false;
3804 
3805 	return true;
3806 }
3807 
3808 static bool ldtr_valid(struct kvm_vcpu *vcpu)
3809 {
3810 	struct kvm_segment ldtr;
3811 
3812 	vmx_get_segment(vcpu, &ldtr, VCPU_SREG_LDTR);
3813 
3814 	if (ldtr.unusable)
3815 		return true;
3816 	if (ldtr.selector & SEGMENT_TI_MASK)	/* TI = 1 */
3817 		return false;
3818 	if (ldtr.type != 2)
3819 		return false;
3820 	if (!ldtr.present)
3821 		return false;
3822 
3823 	return true;
3824 }
3825 
3826 static bool cs_ss_rpl_check(struct kvm_vcpu *vcpu)
3827 {
3828 	struct kvm_segment cs, ss;
3829 
3830 	vmx_get_segment(vcpu, &cs, VCPU_SREG_CS);
3831 	vmx_get_segment(vcpu, &ss, VCPU_SREG_SS);
3832 
3833 	return ((cs.selector & SEGMENT_RPL_MASK) ==
3834 		 (ss.selector & SEGMENT_RPL_MASK));
3835 }
3836 
3837 /*
3838  * Check if guest state is valid. Returns true if valid, false if
3839  * not.
3840  * We assume that registers are always usable
3841  */
3842 bool __vmx_guest_state_valid(struct kvm_vcpu *vcpu)
3843 {
3844 	/* real mode guest state checks */
3845 	if (!is_protmode(vcpu) || (vmx_get_rflags(vcpu) & X86_EFLAGS_VM)) {
3846 		if (!rmode_segment_valid(vcpu, VCPU_SREG_CS))
3847 			return false;
3848 		if (!rmode_segment_valid(vcpu, VCPU_SREG_SS))
3849 			return false;
3850 		if (!rmode_segment_valid(vcpu, VCPU_SREG_DS))
3851 			return false;
3852 		if (!rmode_segment_valid(vcpu, VCPU_SREG_ES))
3853 			return false;
3854 		if (!rmode_segment_valid(vcpu, VCPU_SREG_FS))
3855 			return false;
3856 		if (!rmode_segment_valid(vcpu, VCPU_SREG_GS))
3857 			return false;
3858 	} else {
3859 	/* protected mode guest state checks */
3860 		if (!cs_ss_rpl_check(vcpu))
3861 			return false;
3862 		if (!code_segment_valid(vcpu))
3863 			return false;
3864 		if (!stack_segment_valid(vcpu))
3865 			return false;
3866 		if (!data_segment_valid(vcpu, VCPU_SREG_DS))
3867 			return false;
3868 		if (!data_segment_valid(vcpu, VCPU_SREG_ES))
3869 			return false;
3870 		if (!data_segment_valid(vcpu, VCPU_SREG_FS))
3871 			return false;
3872 		if (!data_segment_valid(vcpu, VCPU_SREG_GS))
3873 			return false;
3874 		if (!tr_valid(vcpu))
3875 			return false;
3876 		if (!ldtr_valid(vcpu))
3877 			return false;
3878 	}
3879 	/* TODO:
3880 	 * - Add checks on RIP
3881 	 * - Add checks on RFLAGS
3882 	 */
3883 
3884 	return true;
3885 }
3886 
3887 static int init_rmode_tss(struct kvm *kvm, void __user *ua)
3888 {
3889 	const void *zero_page = (const void *) __va(page_to_phys(ZERO_PAGE(0)));
3890 	u16 data;
3891 	int i;
3892 
3893 	for (i = 0; i < 3; i++) {
3894 		if (__copy_to_user(ua + PAGE_SIZE * i, zero_page, PAGE_SIZE))
3895 			return -EFAULT;
3896 	}
3897 
3898 	data = TSS_BASE_SIZE + TSS_REDIRECTION_SIZE;
3899 	if (__copy_to_user(ua + TSS_IOPB_BASE_OFFSET, &data, sizeof(u16)))
3900 		return -EFAULT;
3901 
3902 	data = ~0;
3903 	if (__copy_to_user(ua + RMODE_TSS_SIZE - 1, &data, sizeof(u8)))
3904 		return -EFAULT;
3905 
3906 	return 0;
3907 }
3908 
3909 static int init_rmode_identity_map(struct kvm *kvm)
3910 {
3911 	struct kvm_vmx *kvm_vmx = to_kvm_vmx(kvm);
3912 	int i, r = 0;
3913 	void __user *uaddr;
3914 	u32 tmp;
3915 
3916 	/* Protect kvm_vmx->ept_identity_pagetable_done. */
3917 	mutex_lock(&kvm->slots_lock);
3918 
3919 	if (likely(kvm_vmx->ept_identity_pagetable_done))
3920 		goto out;
3921 
3922 	if (!kvm_vmx->ept_identity_map_addr)
3923 		kvm_vmx->ept_identity_map_addr = VMX_EPT_IDENTITY_PAGETABLE_ADDR;
3924 
3925 	uaddr = __x86_set_memory_region(kvm,
3926 					IDENTITY_PAGETABLE_PRIVATE_MEMSLOT,
3927 					kvm_vmx->ept_identity_map_addr,
3928 					PAGE_SIZE);
3929 	if (IS_ERR(uaddr)) {
3930 		r = PTR_ERR(uaddr);
3931 		goto out;
3932 	}
3933 
3934 	/* Set up identity-mapping pagetable for EPT in real mode */
3935 	for (i = 0; i < (PAGE_SIZE / sizeof(tmp)); i++) {
3936 		tmp = (i << 22) + (_PAGE_PRESENT | _PAGE_RW | _PAGE_USER |
3937 			_PAGE_ACCESSED | _PAGE_DIRTY | _PAGE_PSE);
3938 		if (__copy_to_user(uaddr + i * sizeof(tmp), &tmp, sizeof(tmp))) {
3939 			r = -EFAULT;
3940 			goto out;
3941 		}
3942 	}
3943 	kvm_vmx->ept_identity_pagetable_done = true;
3944 
3945 out:
3946 	mutex_unlock(&kvm->slots_lock);
3947 	return r;
3948 }
3949 
3950 static void seg_setup(int seg)
3951 {
3952 	const struct kvm_vmx_segment_field *sf = &kvm_vmx_segment_fields[seg];
3953 	unsigned int ar;
3954 
3955 	vmcs_write16(sf->selector, 0);
3956 	vmcs_writel(sf->base, 0);
3957 	vmcs_write32(sf->limit, 0xffff);
3958 	ar = 0x93;
3959 	if (seg == VCPU_SREG_CS)
3960 		ar |= 0x08; /* code segment */
3961 
3962 	vmcs_write32(sf->ar_bytes, ar);
3963 }
3964 
3965 int allocate_vpid(void)
3966 {
3967 	int vpid;
3968 
3969 	if (!enable_vpid)
3970 		return 0;
3971 	spin_lock(&vmx_vpid_lock);
3972 	vpid = find_first_zero_bit(vmx_vpid_bitmap, VMX_NR_VPIDS);
3973 	if (vpid < VMX_NR_VPIDS)
3974 		__set_bit(vpid, vmx_vpid_bitmap);
3975 	else
3976 		vpid = 0;
3977 	spin_unlock(&vmx_vpid_lock);
3978 	return vpid;
3979 }
3980 
3981 void free_vpid(int vpid)
3982 {
3983 	if (!enable_vpid || vpid == 0)
3984 		return;
3985 	spin_lock(&vmx_vpid_lock);
3986 	__clear_bit(vpid, vmx_vpid_bitmap);
3987 	spin_unlock(&vmx_vpid_lock);
3988 }
3989 
3990 static void vmx_msr_bitmap_l01_changed(struct vcpu_vmx *vmx)
3991 {
3992 	/*
3993 	 * When KVM is a nested hypervisor on top of Hyper-V and uses
3994 	 * 'Enlightened MSR Bitmap' feature L0 needs to know that MSR
3995 	 * bitmap has changed.
3996 	 */
3997 	if (kvm_is_using_evmcs()) {
3998 		struct hv_enlightened_vmcs *evmcs = (void *)vmx->vmcs01.vmcs;
3999 
4000 		if (evmcs->hv_enlightenments_control.msr_bitmap)
4001 			evmcs->hv_clean_fields &=
4002 				~HV_VMX_ENLIGHTENED_CLEAN_FIELD_MSR_BITMAP;
4003 	}
4004 
4005 	vmx->nested.force_msr_bitmap_recalc = true;
4006 }
4007 
4008 void vmx_disable_intercept_for_msr(struct kvm_vcpu *vcpu, u32 msr, int type)
4009 {
4010 	struct vcpu_vmx *vmx = to_vmx(vcpu);
4011 	unsigned long *msr_bitmap = vmx->vmcs01.msr_bitmap;
4012 	int idx;
4013 
4014 	if (!cpu_has_vmx_msr_bitmap())
4015 		return;
4016 
4017 	vmx_msr_bitmap_l01_changed(vmx);
4018 
4019 	/*
4020 	 * Mark the desired intercept state in shadow bitmap, this is needed
4021 	 * for resync when the MSR filters change.
4022 	 */
4023 	idx = vmx_get_passthrough_msr_slot(msr);
4024 	if (idx >= 0) {
4025 		if (type & MSR_TYPE_R)
4026 			clear_bit(idx, vmx->shadow_msr_intercept.read);
4027 		if (type & MSR_TYPE_W)
4028 			clear_bit(idx, vmx->shadow_msr_intercept.write);
4029 	}
4030 
4031 	if ((type & MSR_TYPE_R) &&
4032 	    !kvm_msr_allowed(vcpu, msr, KVM_MSR_FILTER_READ)) {
4033 		vmx_set_msr_bitmap_read(msr_bitmap, msr);
4034 		type &= ~MSR_TYPE_R;
4035 	}
4036 
4037 	if ((type & MSR_TYPE_W) &&
4038 	    !kvm_msr_allowed(vcpu, msr, KVM_MSR_FILTER_WRITE)) {
4039 		vmx_set_msr_bitmap_write(msr_bitmap, msr);
4040 		type &= ~MSR_TYPE_W;
4041 	}
4042 
4043 	if (type & MSR_TYPE_R)
4044 		vmx_clear_msr_bitmap_read(msr_bitmap, msr);
4045 
4046 	if (type & MSR_TYPE_W)
4047 		vmx_clear_msr_bitmap_write(msr_bitmap, msr);
4048 }
4049 
4050 void vmx_enable_intercept_for_msr(struct kvm_vcpu *vcpu, u32 msr, int type)
4051 {
4052 	struct vcpu_vmx *vmx = to_vmx(vcpu);
4053 	unsigned long *msr_bitmap = vmx->vmcs01.msr_bitmap;
4054 	int idx;
4055 
4056 	if (!cpu_has_vmx_msr_bitmap())
4057 		return;
4058 
4059 	vmx_msr_bitmap_l01_changed(vmx);
4060 
4061 	/*
4062 	 * Mark the desired intercept state in shadow bitmap, this is needed
4063 	 * for resync when the MSR filter changes.
4064 	 */
4065 	idx = vmx_get_passthrough_msr_slot(msr);
4066 	if (idx >= 0) {
4067 		if (type & MSR_TYPE_R)
4068 			set_bit(idx, vmx->shadow_msr_intercept.read);
4069 		if (type & MSR_TYPE_W)
4070 			set_bit(idx, vmx->shadow_msr_intercept.write);
4071 	}
4072 
4073 	if (type & MSR_TYPE_R)
4074 		vmx_set_msr_bitmap_read(msr_bitmap, msr);
4075 
4076 	if (type & MSR_TYPE_W)
4077 		vmx_set_msr_bitmap_write(msr_bitmap, msr);
4078 }
4079 
4080 static void vmx_update_msr_bitmap_x2apic(struct kvm_vcpu *vcpu)
4081 {
4082 	/*
4083 	 * x2APIC indices for 64-bit accesses into the RDMSR and WRMSR halves
4084 	 * of the MSR bitmap.  KVM emulates APIC registers up through 0x3f0,
4085 	 * i.e. MSR 0x83f, and so only needs to dynamically manipulate 64 bits.
4086 	 */
4087 	const int read_idx = APIC_BASE_MSR / BITS_PER_LONG_LONG;
4088 	const int write_idx = read_idx + (0x800 / sizeof(u64));
4089 	struct vcpu_vmx *vmx = to_vmx(vcpu);
4090 	u64 *msr_bitmap = (u64 *)vmx->vmcs01.msr_bitmap;
4091 	u8 mode;
4092 
4093 	if (!cpu_has_vmx_msr_bitmap() || WARN_ON_ONCE(!lapic_in_kernel(vcpu)))
4094 		return;
4095 
4096 	if (cpu_has_secondary_exec_ctrls() &&
4097 	    (secondary_exec_controls_get(vmx) &
4098 	     SECONDARY_EXEC_VIRTUALIZE_X2APIC_MODE)) {
4099 		mode = MSR_BITMAP_MODE_X2APIC;
4100 		if (enable_apicv && kvm_vcpu_apicv_active(vcpu))
4101 			mode |= MSR_BITMAP_MODE_X2APIC_APICV;
4102 	} else {
4103 		mode = 0;
4104 	}
4105 
4106 	if (mode == vmx->x2apic_msr_bitmap_mode)
4107 		return;
4108 
4109 	vmx->x2apic_msr_bitmap_mode = mode;
4110 
4111 	/*
4112 	 * Reset the bitmap for MSRs 0x800 - 0x83f.  Leave AMD's uber-extended
4113 	 * registers (0x840 and above) intercepted, KVM doesn't support them.
4114 	 * Intercept all writes by default and poke holes as needed.  Pass
4115 	 * through reads for all valid registers by default in x2APIC+APICv
4116 	 * mode, only the current timer count needs on-demand emulation by KVM.
4117 	 */
4118 	if (mode & MSR_BITMAP_MODE_X2APIC_APICV)
4119 		msr_bitmap[read_idx] = ~kvm_lapic_readable_reg_mask(vcpu->arch.apic);
4120 	else
4121 		msr_bitmap[read_idx] = ~0ull;
4122 	msr_bitmap[write_idx] = ~0ull;
4123 
4124 	/*
4125 	 * TPR reads and writes can be virtualized even if virtual interrupt
4126 	 * delivery is not in use.
4127 	 */
4128 	vmx_set_intercept_for_msr(vcpu, X2APIC_MSR(APIC_TASKPRI), MSR_TYPE_RW,
4129 				  !(mode & MSR_BITMAP_MODE_X2APIC));
4130 
4131 	if (mode & MSR_BITMAP_MODE_X2APIC_APICV) {
4132 		vmx_enable_intercept_for_msr(vcpu, X2APIC_MSR(APIC_TMCCT), MSR_TYPE_RW);
4133 		vmx_disable_intercept_for_msr(vcpu, X2APIC_MSR(APIC_EOI), MSR_TYPE_W);
4134 		vmx_disable_intercept_for_msr(vcpu, X2APIC_MSR(APIC_SELF_IPI), MSR_TYPE_W);
4135 		if (enable_ipiv)
4136 			vmx_disable_intercept_for_msr(vcpu, X2APIC_MSR(APIC_ICR), MSR_TYPE_RW);
4137 	}
4138 }
4139 
4140 void pt_update_intercept_for_msr(struct kvm_vcpu *vcpu)
4141 {
4142 	struct vcpu_vmx *vmx = to_vmx(vcpu);
4143 	bool flag = !(vmx->pt_desc.guest.ctl & RTIT_CTL_TRACEEN);
4144 	u32 i;
4145 
4146 	vmx_set_intercept_for_msr(vcpu, MSR_IA32_RTIT_STATUS, MSR_TYPE_RW, flag);
4147 	vmx_set_intercept_for_msr(vcpu, MSR_IA32_RTIT_OUTPUT_BASE, MSR_TYPE_RW, flag);
4148 	vmx_set_intercept_for_msr(vcpu, MSR_IA32_RTIT_OUTPUT_MASK, MSR_TYPE_RW, flag);
4149 	vmx_set_intercept_for_msr(vcpu, MSR_IA32_RTIT_CR3_MATCH, MSR_TYPE_RW, flag);
4150 	for (i = 0; i < vmx->pt_desc.num_address_ranges; i++) {
4151 		vmx_set_intercept_for_msr(vcpu, MSR_IA32_RTIT_ADDR0_A + i * 2, MSR_TYPE_RW, flag);
4152 		vmx_set_intercept_for_msr(vcpu, MSR_IA32_RTIT_ADDR0_B + i * 2, MSR_TYPE_RW, flag);
4153 	}
4154 }
4155 
4156 void vmx_msr_filter_changed(struct kvm_vcpu *vcpu)
4157 {
4158 	struct vcpu_vmx *vmx = to_vmx(vcpu);
4159 	u32 i;
4160 
4161 	if (!cpu_has_vmx_msr_bitmap())
4162 		return;
4163 
4164 	/*
4165 	 * Redo intercept permissions for MSRs that KVM is passing through to
4166 	 * the guest.  Disabling interception will check the new MSR filter and
4167 	 * ensure that KVM enables interception if usersepace wants to filter
4168 	 * the MSR.  MSRs that KVM is already intercepting don't need to be
4169 	 * refreshed since KVM is going to intercept them regardless of what
4170 	 * userspace wants.
4171 	 */
4172 	for (i = 0; i < ARRAY_SIZE(vmx_possible_passthrough_msrs); i++) {
4173 		u32 msr = vmx_possible_passthrough_msrs[i];
4174 
4175 		if (!test_bit(i, vmx->shadow_msr_intercept.read))
4176 			vmx_disable_intercept_for_msr(vcpu, msr, MSR_TYPE_R);
4177 
4178 		if (!test_bit(i, vmx->shadow_msr_intercept.write))
4179 			vmx_disable_intercept_for_msr(vcpu, msr, MSR_TYPE_W);
4180 	}
4181 
4182 	/* PT MSRs can be passed through iff PT is exposed to the guest. */
4183 	if (vmx_pt_mode_is_host_guest())
4184 		pt_update_intercept_for_msr(vcpu);
4185 }
4186 
4187 static inline void kvm_vcpu_trigger_posted_interrupt(struct kvm_vcpu *vcpu,
4188 						     int pi_vec)
4189 {
4190 #ifdef CONFIG_SMP
4191 	if (vcpu->mode == IN_GUEST_MODE) {
4192 		/*
4193 		 * The vector of the virtual has already been set in the PIR.
4194 		 * Send a notification event to deliver the virtual interrupt
4195 		 * unless the vCPU is the currently running vCPU, i.e. the
4196 		 * event is being sent from a fastpath VM-Exit handler, in
4197 		 * which case the PIR will be synced to the vIRR before
4198 		 * re-entering the guest.
4199 		 *
4200 		 * When the target is not the running vCPU, the following
4201 		 * possibilities emerge:
4202 		 *
4203 		 * Case 1: vCPU stays in non-root mode. Sending a notification
4204 		 * event posts the interrupt to the vCPU.
4205 		 *
4206 		 * Case 2: vCPU exits to root mode and is still runnable. The
4207 		 * PIR will be synced to the vIRR before re-entering the guest.
4208 		 * Sending a notification event is ok as the host IRQ handler
4209 		 * will ignore the spurious event.
4210 		 *
4211 		 * Case 3: vCPU exits to root mode and is blocked. vcpu_block()
4212 		 * has already synced PIR to vIRR and never blocks the vCPU if
4213 		 * the vIRR is not empty. Therefore, a blocked vCPU here does
4214 		 * not wait for any requested interrupts in PIR, and sending a
4215 		 * notification event also results in a benign, spurious event.
4216 		 */
4217 
4218 		if (vcpu != kvm_get_running_vcpu())
4219 			__apic_send_IPI_mask(get_cpu_mask(vcpu->cpu), pi_vec);
4220 		return;
4221 	}
4222 #endif
4223 	/*
4224 	 * The vCPU isn't in the guest; wake the vCPU in case it is blocking,
4225 	 * otherwise do nothing as KVM will grab the highest priority pending
4226 	 * IRQ via ->sync_pir_to_irr() in vcpu_enter_guest().
4227 	 */
4228 	kvm_vcpu_wake_up(vcpu);
4229 }
4230 
4231 static int vmx_deliver_nested_posted_interrupt(struct kvm_vcpu *vcpu,
4232 						int vector)
4233 {
4234 	struct vcpu_vmx *vmx = to_vmx(vcpu);
4235 
4236 	/*
4237 	 * DO NOT query the vCPU's vmcs12, as vmcs12 is dynamically allocated
4238 	 * and freed, and must not be accessed outside of vcpu->mutex.  The
4239 	 * vCPU's cached PI NV is valid if and only if posted interrupts
4240 	 * enabled in its vmcs12, i.e. checking the vector also checks that
4241 	 * L1 has enabled posted interrupts for L2.
4242 	 */
4243 	if (is_guest_mode(vcpu) &&
4244 	    vector == vmx->nested.posted_intr_nv) {
4245 		/*
4246 		 * If a posted intr is not recognized by hardware,
4247 		 * we will accomplish it in the next vmentry.
4248 		 */
4249 		vmx->nested.pi_pending = true;
4250 		kvm_make_request(KVM_REQ_EVENT, vcpu);
4251 
4252 		/*
4253 		 * This pairs with the smp_mb_*() after setting vcpu->mode in
4254 		 * vcpu_enter_guest() to guarantee the vCPU sees the event
4255 		 * request if triggering a posted interrupt "fails" because
4256 		 * vcpu->mode != IN_GUEST_MODE.  The extra barrier is needed as
4257 		 * the smb_wmb() in kvm_make_request() only ensures everything
4258 		 * done before making the request is visible when the request
4259 		 * is visible, it doesn't ensure ordering between the store to
4260 		 * vcpu->requests and the load from vcpu->mode.
4261 		 */
4262 		smp_mb__after_atomic();
4263 
4264 		/* the PIR and ON have been set by L1. */
4265 		kvm_vcpu_trigger_posted_interrupt(vcpu, POSTED_INTR_NESTED_VECTOR);
4266 		return 0;
4267 	}
4268 	return -1;
4269 }
4270 /*
4271  * Send interrupt to vcpu via posted interrupt way.
4272  * 1. If target vcpu is running(non-root mode), send posted interrupt
4273  * notification to vcpu and hardware will sync PIR to vIRR atomically.
4274  * 2. If target vcpu isn't running(root mode), kick it to pick up the
4275  * interrupt from PIR in next vmentry.
4276  */
4277 static int vmx_deliver_posted_interrupt(struct kvm_vcpu *vcpu, int vector)
4278 {
4279 	struct vcpu_vmx *vmx = to_vmx(vcpu);
4280 	int r;
4281 
4282 	r = vmx_deliver_nested_posted_interrupt(vcpu, vector);
4283 	if (!r)
4284 		return 0;
4285 
4286 	/* Note, this is called iff the local APIC is in-kernel. */
4287 	if (!vcpu->arch.apic->apicv_active)
4288 		return -1;
4289 
4290 	if (pi_test_and_set_pir(vector, &vmx->pi_desc))
4291 		return 0;
4292 
4293 	/* If a previous notification has sent the IPI, nothing to do.  */
4294 	if (pi_test_and_set_on(&vmx->pi_desc))
4295 		return 0;
4296 
4297 	/*
4298 	 * The implied barrier in pi_test_and_set_on() pairs with the smp_mb_*()
4299 	 * after setting vcpu->mode in vcpu_enter_guest(), thus the vCPU is
4300 	 * guaranteed to see PID.ON=1 and sync the PIR to IRR if triggering a
4301 	 * posted interrupt "fails" because vcpu->mode != IN_GUEST_MODE.
4302 	 */
4303 	kvm_vcpu_trigger_posted_interrupt(vcpu, POSTED_INTR_VECTOR);
4304 	return 0;
4305 }
4306 
4307 void vmx_deliver_interrupt(struct kvm_lapic *apic, int delivery_mode,
4308 			   int trig_mode, int vector)
4309 {
4310 	struct kvm_vcpu *vcpu = apic->vcpu;
4311 
4312 	if (vmx_deliver_posted_interrupt(vcpu, vector)) {
4313 		kvm_lapic_set_irr(vector, apic);
4314 		kvm_make_request(KVM_REQ_EVENT, vcpu);
4315 		kvm_vcpu_kick(vcpu);
4316 	} else {
4317 		trace_kvm_apicv_accept_irq(vcpu->vcpu_id, delivery_mode,
4318 					   trig_mode, vector);
4319 	}
4320 }
4321 
4322 /*
4323  * Set up the vmcs's constant host-state fields, i.e., host-state fields that
4324  * will not change in the lifetime of the guest.
4325  * Note that host-state that does change is set elsewhere. E.g., host-state
4326  * that is set differently for each CPU is set in vmx_vcpu_load(), not here.
4327  */
4328 void vmx_set_constant_host_state(struct vcpu_vmx *vmx)
4329 {
4330 	u32 low32, high32;
4331 	unsigned long tmpl;
4332 	unsigned long cr0, cr3, cr4;
4333 
4334 	cr0 = read_cr0();
4335 	WARN_ON(cr0 & X86_CR0_TS);
4336 	vmcs_writel(HOST_CR0, cr0);  /* 22.2.3 */
4337 
4338 	/*
4339 	 * Save the most likely value for this task's CR3 in the VMCS.
4340 	 * We can't use __get_current_cr3_fast() because we're not atomic.
4341 	 */
4342 	cr3 = __read_cr3();
4343 	vmcs_writel(HOST_CR3, cr3);		/* 22.2.3  FIXME: shadow tables */
4344 	vmx->loaded_vmcs->host_state.cr3 = cr3;
4345 
4346 	/* Save the most likely value for this task's CR4 in the VMCS. */
4347 	cr4 = cr4_read_shadow();
4348 	vmcs_writel(HOST_CR4, cr4);			/* 22.2.3, 22.2.5 */
4349 	vmx->loaded_vmcs->host_state.cr4 = cr4;
4350 
4351 	vmcs_write16(HOST_CS_SELECTOR, __KERNEL_CS);  /* 22.2.4 */
4352 #ifdef CONFIG_X86_64
4353 	/*
4354 	 * Load null selectors, so we can avoid reloading them in
4355 	 * vmx_prepare_switch_to_host(), in case userspace uses
4356 	 * the null selectors too (the expected case).
4357 	 */
4358 	vmcs_write16(HOST_DS_SELECTOR, 0);
4359 	vmcs_write16(HOST_ES_SELECTOR, 0);
4360 #else
4361 	vmcs_write16(HOST_DS_SELECTOR, __KERNEL_DS);  /* 22.2.4 */
4362 	vmcs_write16(HOST_ES_SELECTOR, __KERNEL_DS);  /* 22.2.4 */
4363 #endif
4364 	vmcs_write16(HOST_SS_SELECTOR, __KERNEL_DS);  /* 22.2.4 */
4365 	vmcs_write16(HOST_TR_SELECTOR, GDT_ENTRY_TSS*8);  /* 22.2.4 */
4366 
4367 	vmcs_writel(HOST_IDTR_BASE, host_idt_base);   /* 22.2.4 */
4368 
4369 	vmcs_writel(HOST_RIP, (unsigned long)vmx_vmexit); /* 22.2.5 */
4370 
4371 	rdmsr(MSR_IA32_SYSENTER_CS, low32, high32);
4372 	vmcs_write32(HOST_IA32_SYSENTER_CS, low32);
4373 
4374 	/*
4375 	 * SYSENTER is used for 32-bit system calls on either 32-bit or
4376 	 * 64-bit kernels.  It is always zero If neither is allowed, otherwise
4377 	 * vmx_vcpu_load_vmcs loads it with the per-CPU entry stack (and may
4378 	 * have already done so!).
4379 	 */
4380 	if (!IS_ENABLED(CONFIG_IA32_EMULATION) && !IS_ENABLED(CONFIG_X86_32))
4381 		vmcs_writel(HOST_IA32_SYSENTER_ESP, 0);
4382 
4383 	rdmsrl(MSR_IA32_SYSENTER_EIP, tmpl);
4384 	vmcs_writel(HOST_IA32_SYSENTER_EIP, tmpl);   /* 22.2.3 */
4385 
4386 	if (vmcs_config.vmexit_ctrl & VM_EXIT_LOAD_IA32_PAT) {
4387 		rdmsr(MSR_IA32_CR_PAT, low32, high32);
4388 		vmcs_write64(HOST_IA32_PAT, low32 | ((u64) high32 << 32));
4389 	}
4390 
4391 	if (cpu_has_load_ia32_efer())
4392 		vmcs_write64(HOST_IA32_EFER, kvm_host.efer);
4393 }
4394 
4395 void set_cr4_guest_host_mask(struct vcpu_vmx *vmx)
4396 {
4397 	struct kvm_vcpu *vcpu = &vmx->vcpu;
4398 
4399 	vcpu->arch.cr4_guest_owned_bits = KVM_POSSIBLE_CR4_GUEST_BITS &
4400 					  ~vcpu->arch.cr4_guest_rsvd_bits;
4401 	if (!enable_ept) {
4402 		vcpu->arch.cr4_guest_owned_bits &= ~X86_CR4_TLBFLUSH_BITS;
4403 		vcpu->arch.cr4_guest_owned_bits &= ~X86_CR4_PDPTR_BITS;
4404 	}
4405 	if (is_guest_mode(&vmx->vcpu))
4406 		vcpu->arch.cr4_guest_owned_bits &=
4407 			~get_vmcs12(vcpu)->cr4_guest_host_mask;
4408 	vmcs_writel(CR4_GUEST_HOST_MASK, ~vcpu->arch.cr4_guest_owned_bits);
4409 }
4410 
4411 static u32 vmx_pin_based_exec_ctrl(struct vcpu_vmx *vmx)
4412 {
4413 	u32 pin_based_exec_ctrl = vmcs_config.pin_based_exec_ctrl;
4414 
4415 	if (!kvm_vcpu_apicv_active(&vmx->vcpu))
4416 		pin_based_exec_ctrl &= ~PIN_BASED_POSTED_INTR;
4417 
4418 	if (!enable_vnmi)
4419 		pin_based_exec_ctrl &= ~PIN_BASED_VIRTUAL_NMIS;
4420 
4421 	if (!enable_preemption_timer)
4422 		pin_based_exec_ctrl &= ~PIN_BASED_VMX_PREEMPTION_TIMER;
4423 
4424 	return pin_based_exec_ctrl;
4425 }
4426 
4427 static u32 vmx_vmentry_ctrl(void)
4428 {
4429 	u32 vmentry_ctrl = vmcs_config.vmentry_ctrl;
4430 
4431 	if (vmx_pt_mode_is_system())
4432 		vmentry_ctrl &= ~(VM_ENTRY_PT_CONCEAL_PIP |
4433 				  VM_ENTRY_LOAD_IA32_RTIT_CTL);
4434 	/*
4435 	 * IA32e mode, and loading of EFER and PERF_GLOBAL_CTRL are toggled dynamically.
4436 	 */
4437 	vmentry_ctrl &= ~(VM_ENTRY_LOAD_IA32_PERF_GLOBAL_CTRL |
4438 			  VM_ENTRY_LOAD_IA32_EFER |
4439 			  VM_ENTRY_IA32E_MODE);
4440 
4441 	return vmentry_ctrl;
4442 }
4443 
4444 static u32 vmx_vmexit_ctrl(void)
4445 {
4446 	u32 vmexit_ctrl = vmcs_config.vmexit_ctrl;
4447 
4448 	/*
4449 	 * Not used by KVM and never set in vmcs01 or vmcs02, but emulated for
4450 	 * nested virtualization and thus allowed to be set in vmcs12.
4451 	 */
4452 	vmexit_ctrl &= ~(VM_EXIT_SAVE_IA32_PAT | VM_EXIT_SAVE_IA32_EFER |
4453 			 VM_EXIT_SAVE_VMX_PREEMPTION_TIMER);
4454 
4455 	if (vmx_pt_mode_is_system())
4456 		vmexit_ctrl &= ~(VM_EXIT_PT_CONCEAL_PIP |
4457 				 VM_EXIT_CLEAR_IA32_RTIT_CTL);
4458 	/* Loading of EFER and PERF_GLOBAL_CTRL are toggled dynamically */
4459 	return vmexit_ctrl &
4460 		~(VM_EXIT_LOAD_IA32_PERF_GLOBAL_CTRL | VM_EXIT_LOAD_IA32_EFER);
4461 }
4462 
4463 void vmx_refresh_apicv_exec_ctrl(struct kvm_vcpu *vcpu)
4464 {
4465 	struct vcpu_vmx *vmx = to_vmx(vcpu);
4466 
4467 	if (is_guest_mode(vcpu)) {
4468 		vmx->nested.update_vmcs01_apicv_status = true;
4469 		return;
4470 	}
4471 
4472 	pin_controls_set(vmx, vmx_pin_based_exec_ctrl(vmx));
4473 
4474 	if (kvm_vcpu_apicv_active(vcpu)) {
4475 		secondary_exec_controls_setbit(vmx,
4476 					       SECONDARY_EXEC_APIC_REGISTER_VIRT |
4477 					       SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY);
4478 		if (enable_ipiv)
4479 			tertiary_exec_controls_setbit(vmx, TERTIARY_EXEC_IPI_VIRT);
4480 	} else {
4481 		secondary_exec_controls_clearbit(vmx,
4482 						 SECONDARY_EXEC_APIC_REGISTER_VIRT |
4483 						 SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY);
4484 		if (enable_ipiv)
4485 			tertiary_exec_controls_clearbit(vmx, TERTIARY_EXEC_IPI_VIRT);
4486 	}
4487 
4488 	vmx_update_msr_bitmap_x2apic(vcpu);
4489 }
4490 
4491 static u32 vmx_exec_control(struct vcpu_vmx *vmx)
4492 {
4493 	u32 exec_control = vmcs_config.cpu_based_exec_ctrl;
4494 
4495 	/*
4496 	 * Not used by KVM, but fully supported for nesting, i.e. are allowed in
4497 	 * vmcs12 and propagated to vmcs02 when set in vmcs12.
4498 	 */
4499 	exec_control &= ~(CPU_BASED_RDTSC_EXITING |
4500 			  CPU_BASED_USE_IO_BITMAPS |
4501 			  CPU_BASED_MONITOR_TRAP_FLAG |
4502 			  CPU_BASED_PAUSE_EXITING);
4503 
4504 	/* INTR_WINDOW_EXITING and NMI_WINDOW_EXITING are toggled dynamically */
4505 	exec_control &= ~(CPU_BASED_INTR_WINDOW_EXITING |
4506 			  CPU_BASED_NMI_WINDOW_EXITING);
4507 
4508 	if (vmx->vcpu.arch.switch_db_regs & KVM_DEBUGREG_WONT_EXIT)
4509 		exec_control &= ~CPU_BASED_MOV_DR_EXITING;
4510 
4511 	if (!cpu_need_tpr_shadow(&vmx->vcpu))
4512 		exec_control &= ~CPU_BASED_TPR_SHADOW;
4513 
4514 #ifdef CONFIG_X86_64
4515 	if (exec_control & CPU_BASED_TPR_SHADOW)
4516 		exec_control &= ~(CPU_BASED_CR8_LOAD_EXITING |
4517 				  CPU_BASED_CR8_STORE_EXITING);
4518 	else
4519 		exec_control |= CPU_BASED_CR8_STORE_EXITING |
4520 				CPU_BASED_CR8_LOAD_EXITING;
4521 #endif
4522 	/* No need to intercept CR3 access or INVPLG when using EPT. */
4523 	if (enable_ept)
4524 		exec_control &= ~(CPU_BASED_CR3_LOAD_EXITING |
4525 				  CPU_BASED_CR3_STORE_EXITING |
4526 				  CPU_BASED_INVLPG_EXITING);
4527 	if (kvm_mwait_in_guest(vmx->vcpu.kvm))
4528 		exec_control &= ~(CPU_BASED_MWAIT_EXITING |
4529 				CPU_BASED_MONITOR_EXITING);
4530 	if (kvm_hlt_in_guest(vmx->vcpu.kvm))
4531 		exec_control &= ~CPU_BASED_HLT_EXITING;
4532 	return exec_control;
4533 }
4534 
4535 static u64 vmx_tertiary_exec_control(struct vcpu_vmx *vmx)
4536 {
4537 	u64 exec_control = vmcs_config.cpu_based_3rd_exec_ctrl;
4538 
4539 	/*
4540 	 * IPI virtualization relies on APICv. Disable IPI virtualization if
4541 	 * APICv is inhibited.
4542 	 */
4543 	if (!enable_ipiv || !kvm_vcpu_apicv_active(&vmx->vcpu))
4544 		exec_control &= ~TERTIARY_EXEC_IPI_VIRT;
4545 
4546 	return exec_control;
4547 }
4548 
4549 /*
4550  * Adjust a single secondary execution control bit to intercept/allow an
4551  * instruction in the guest.  This is usually done based on whether or not a
4552  * feature has been exposed to the guest in order to correctly emulate faults.
4553  */
4554 static inline void
4555 vmx_adjust_secondary_exec_control(struct vcpu_vmx *vmx, u32 *exec_control,
4556 				  u32 control, bool enabled, bool exiting)
4557 {
4558 	/*
4559 	 * If the control is for an opt-in feature, clear the control if the
4560 	 * feature is not exposed to the guest, i.e. not enabled.  If the
4561 	 * control is opt-out, i.e. an exiting control, clear the control if
4562 	 * the feature _is_ exposed to the guest, i.e. exiting/interception is
4563 	 * disabled for the associated instruction.  Note, the caller is
4564 	 * responsible presetting exec_control to set all supported bits.
4565 	 */
4566 	if (enabled == exiting)
4567 		*exec_control &= ~control;
4568 
4569 	/*
4570 	 * Update the nested MSR settings so that a nested VMM can/can't set
4571 	 * controls for features that are/aren't exposed to the guest.
4572 	 */
4573 	if (nested &&
4574 	    kvm_check_has_quirk(vmx->vcpu.kvm, KVM_X86_QUIRK_STUFF_FEATURE_MSRS)) {
4575 		/*
4576 		 * All features that can be added or removed to VMX MSRs must
4577 		 * be supported in the first place for nested virtualization.
4578 		 */
4579 		if (WARN_ON_ONCE(!(vmcs_config.nested.secondary_ctls_high & control)))
4580 			enabled = false;
4581 
4582 		if (enabled)
4583 			vmx->nested.msrs.secondary_ctls_high |= control;
4584 		else
4585 			vmx->nested.msrs.secondary_ctls_high &= ~control;
4586 	}
4587 }
4588 
4589 /*
4590  * Wrapper macro for the common case of adjusting a secondary execution control
4591  * based on a single guest CPUID bit, with a dedicated feature bit.  This also
4592  * verifies that the control is actually supported by KVM and hardware.
4593  */
4594 #define vmx_adjust_sec_exec_control(vmx, exec_control, name, feat_name, ctrl_name, exiting)	\
4595 ({												\
4596 	struct kvm_vcpu *__vcpu = &(vmx)->vcpu;							\
4597 	bool __enabled;										\
4598 												\
4599 	if (cpu_has_vmx_##name()) {								\
4600 		__enabled = guest_cpu_cap_has(__vcpu, X86_FEATURE_##feat_name);			\
4601 		vmx_adjust_secondary_exec_control(vmx, exec_control, SECONDARY_EXEC_##ctrl_name,\
4602 						  __enabled, exiting);				\
4603 	}											\
4604 })
4605 
4606 /* More macro magic for ENABLE_/opt-in versus _EXITING/opt-out controls. */
4607 #define vmx_adjust_sec_exec_feature(vmx, exec_control, lname, uname) \
4608 	vmx_adjust_sec_exec_control(vmx, exec_control, lname, uname, ENABLE_##uname, false)
4609 
4610 #define vmx_adjust_sec_exec_exiting(vmx, exec_control, lname, uname) \
4611 	vmx_adjust_sec_exec_control(vmx, exec_control, lname, uname, uname##_EXITING, true)
4612 
4613 static u32 vmx_secondary_exec_control(struct vcpu_vmx *vmx)
4614 {
4615 	struct kvm_vcpu *vcpu = &vmx->vcpu;
4616 
4617 	u32 exec_control = vmcs_config.cpu_based_2nd_exec_ctrl;
4618 
4619 	if (vmx_pt_mode_is_system())
4620 		exec_control &= ~(SECONDARY_EXEC_PT_USE_GPA | SECONDARY_EXEC_PT_CONCEAL_VMX);
4621 	if (!cpu_need_virtualize_apic_accesses(vcpu))
4622 		exec_control &= ~SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES;
4623 	if (vmx->vpid == 0)
4624 		exec_control &= ~SECONDARY_EXEC_ENABLE_VPID;
4625 	if (!enable_ept) {
4626 		exec_control &= ~SECONDARY_EXEC_ENABLE_EPT;
4627 		exec_control &= ~SECONDARY_EXEC_EPT_VIOLATION_VE;
4628 		enable_unrestricted_guest = 0;
4629 	}
4630 	if (!enable_unrestricted_guest)
4631 		exec_control &= ~SECONDARY_EXEC_UNRESTRICTED_GUEST;
4632 	if (kvm_pause_in_guest(vmx->vcpu.kvm))
4633 		exec_control &= ~SECONDARY_EXEC_PAUSE_LOOP_EXITING;
4634 	if (!kvm_vcpu_apicv_active(vcpu))
4635 		exec_control &= ~(SECONDARY_EXEC_APIC_REGISTER_VIRT |
4636 				  SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY);
4637 	exec_control &= ~SECONDARY_EXEC_VIRTUALIZE_X2APIC_MODE;
4638 
4639 	/*
4640 	 * KVM doesn't support VMFUNC for L1, but the control is set in KVM's
4641 	 * base configuration as KVM emulates VMFUNC[EPTP_SWITCHING] for L2.
4642 	 */
4643 	exec_control &= ~SECONDARY_EXEC_ENABLE_VMFUNC;
4644 
4645 	/* SECONDARY_EXEC_DESC is enabled/disabled on writes to CR4.UMIP,
4646 	 * in vmx_set_cr4.  */
4647 	exec_control &= ~SECONDARY_EXEC_DESC;
4648 
4649 	/* SECONDARY_EXEC_SHADOW_VMCS is enabled when L1 executes VMPTRLD
4650 	   (handle_vmptrld).
4651 	   We can NOT enable shadow_vmcs here because we don't have yet
4652 	   a current VMCS12
4653 	*/
4654 	exec_control &= ~SECONDARY_EXEC_SHADOW_VMCS;
4655 
4656 	/*
4657 	 * PML is enabled/disabled when dirty logging of memsmlots changes, but
4658 	 * it needs to be set here when dirty logging is already active, e.g.
4659 	 * if this vCPU was created after dirty logging was enabled.
4660 	 */
4661 	if (!enable_pml || !atomic_read(&vcpu->kvm->nr_memslots_dirty_logging))
4662 		exec_control &= ~SECONDARY_EXEC_ENABLE_PML;
4663 
4664 	vmx_adjust_sec_exec_feature(vmx, &exec_control, xsaves, XSAVES);
4665 
4666 	/*
4667 	 * RDPID is also gated by ENABLE_RDTSCP, turn on the control if either
4668 	 * feature is exposed to the guest.  This creates a virtualization hole
4669 	 * if both are supported in hardware but only one is exposed to the
4670 	 * guest, but letting the guest execute RDTSCP or RDPID when either one
4671 	 * is advertised is preferable to emulating the advertised instruction
4672 	 * in KVM on #UD, and obviously better than incorrectly injecting #UD.
4673 	 */
4674 	if (cpu_has_vmx_rdtscp()) {
4675 		bool rdpid_or_rdtscp_enabled =
4676 			guest_cpu_cap_has(vcpu, X86_FEATURE_RDTSCP) ||
4677 			guest_cpu_cap_has(vcpu, X86_FEATURE_RDPID);
4678 
4679 		vmx_adjust_secondary_exec_control(vmx, &exec_control,
4680 						  SECONDARY_EXEC_ENABLE_RDTSCP,
4681 						  rdpid_or_rdtscp_enabled, false);
4682 	}
4683 
4684 	vmx_adjust_sec_exec_feature(vmx, &exec_control, invpcid, INVPCID);
4685 
4686 	vmx_adjust_sec_exec_exiting(vmx, &exec_control, rdrand, RDRAND);
4687 	vmx_adjust_sec_exec_exiting(vmx, &exec_control, rdseed, RDSEED);
4688 
4689 	vmx_adjust_sec_exec_control(vmx, &exec_control, waitpkg, WAITPKG,
4690 				    ENABLE_USR_WAIT_PAUSE, false);
4691 
4692 	if (!vcpu->kvm->arch.bus_lock_detection_enabled)
4693 		exec_control &= ~SECONDARY_EXEC_BUS_LOCK_DETECTION;
4694 
4695 	if (!kvm_notify_vmexit_enabled(vcpu->kvm))
4696 		exec_control &= ~SECONDARY_EXEC_NOTIFY_VM_EXITING;
4697 
4698 	return exec_control;
4699 }
4700 
4701 static inline int vmx_get_pid_table_order(struct kvm *kvm)
4702 {
4703 	return get_order(kvm->arch.max_vcpu_ids * sizeof(*to_kvm_vmx(kvm)->pid_table));
4704 }
4705 
4706 static int vmx_alloc_ipiv_pid_table(struct kvm *kvm)
4707 {
4708 	struct page *pages;
4709 	struct kvm_vmx *kvm_vmx = to_kvm_vmx(kvm);
4710 
4711 	if (!irqchip_in_kernel(kvm) || !enable_ipiv)
4712 		return 0;
4713 
4714 	if (kvm_vmx->pid_table)
4715 		return 0;
4716 
4717 	pages = alloc_pages(GFP_KERNEL_ACCOUNT | __GFP_ZERO,
4718 			    vmx_get_pid_table_order(kvm));
4719 	if (!pages)
4720 		return -ENOMEM;
4721 
4722 	kvm_vmx->pid_table = (void *)page_address(pages);
4723 	return 0;
4724 }
4725 
4726 int vmx_vcpu_precreate(struct kvm *kvm)
4727 {
4728 	return vmx_alloc_ipiv_pid_table(kvm);
4729 }
4730 
4731 #define VMX_XSS_EXIT_BITMAP 0
4732 
4733 static void init_vmcs(struct vcpu_vmx *vmx)
4734 {
4735 	struct kvm *kvm = vmx->vcpu.kvm;
4736 	struct kvm_vmx *kvm_vmx = to_kvm_vmx(kvm);
4737 
4738 	if (nested)
4739 		nested_vmx_set_vmcs_shadowing_bitmap();
4740 
4741 	if (cpu_has_vmx_msr_bitmap())
4742 		vmcs_write64(MSR_BITMAP, __pa(vmx->vmcs01.msr_bitmap));
4743 
4744 	vmcs_write64(VMCS_LINK_POINTER, INVALID_GPA); /* 22.3.1.5 */
4745 
4746 	/* Control */
4747 	pin_controls_set(vmx, vmx_pin_based_exec_ctrl(vmx));
4748 
4749 	exec_controls_set(vmx, vmx_exec_control(vmx));
4750 
4751 	if (cpu_has_secondary_exec_ctrls()) {
4752 		secondary_exec_controls_set(vmx, vmx_secondary_exec_control(vmx));
4753 		if (vmx->ve_info)
4754 			vmcs_write64(VE_INFORMATION_ADDRESS,
4755 				     __pa(vmx->ve_info));
4756 	}
4757 
4758 	if (cpu_has_tertiary_exec_ctrls())
4759 		tertiary_exec_controls_set(vmx, vmx_tertiary_exec_control(vmx));
4760 
4761 	if (enable_apicv && lapic_in_kernel(&vmx->vcpu)) {
4762 		vmcs_write64(EOI_EXIT_BITMAP0, 0);
4763 		vmcs_write64(EOI_EXIT_BITMAP1, 0);
4764 		vmcs_write64(EOI_EXIT_BITMAP2, 0);
4765 		vmcs_write64(EOI_EXIT_BITMAP3, 0);
4766 
4767 		vmcs_write16(GUEST_INTR_STATUS, 0);
4768 
4769 		vmcs_write16(POSTED_INTR_NV, POSTED_INTR_VECTOR);
4770 		vmcs_write64(POSTED_INTR_DESC_ADDR, __pa((&vmx->pi_desc)));
4771 	}
4772 
4773 	if (vmx_can_use_ipiv(&vmx->vcpu)) {
4774 		vmcs_write64(PID_POINTER_TABLE, __pa(kvm_vmx->pid_table));
4775 		vmcs_write16(LAST_PID_POINTER_INDEX, kvm->arch.max_vcpu_ids - 1);
4776 	}
4777 
4778 	if (!kvm_pause_in_guest(kvm)) {
4779 		vmcs_write32(PLE_GAP, ple_gap);
4780 		vmx->ple_window = ple_window;
4781 		vmx->ple_window_dirty = true;
4782 	}
4783 
4784 	if (kvm_notify_vmexit_enabled(kvm))
4785 		vmcs_write32(NOTIFY_WINDOW, kvm->arch.notify_window);
4786 
4787 	vmcs_write32(PAGE_FAULT_ERROR_CODE_MASK, 0);
4788 	vmcs_write32(PAGE_FAULT_ERROR_CODE_MATCH, 0);
4789 	vmcs_write32(CR3_TARGET_COUNT, 0);           /* 22.2.1 */
4790 
4791 	vmcs_write16(HOST_FS_SELECTOR, 0);            /* 22.2.4 */
4792 	vmcs_write16(HOST_GS_SELECTOR, 0);            /* 22.2.4 */
4793 	vmx_set_constant_host_state(vmx);
4794 	vmcs_writel(HOST_FS_BASE, 0); /* 22.2.4 */
4795 	vmcs_writel(HOST_GS_BASE, 0); /* 22.2.4 */
4796 
4797 	if (cpu_has_vmx_vmfunc())
4798 		vmcs_write64(VM_FUNCTION_CONTROL, 0);
4799 
4800 	vmcs_write32(VM_EXIT_MSR_STORE_COUNT, 0);
4801 	vmcs_write32(VM_EXIT_MSR_LOAD_COUNT, 0);
4802 	vmcs_write64(VM_EXIT_MSR_LOAD_ADDR, __pa(vmx->msr_autoload.host.val));
4803 	vmcs_write32(VM_ENTRY_MSR_LOAD_COUNT, 0);
4804 	vmcs_write64(VM_ENTRY_MSR_LOAD_ADDR, __pa(vmx->msr_autoload.guest.val));
4805 
4806 	if (vmcs_config.vmentry_ctrl & VM_ENTRY_LOAD_IA32_PAT)
4807 		vmcs_write64(GUEST_IA32_PAT, vmx->vcpu.arch.pat);
4808 
4809 	vm_exit_controls_set(vmx, vmx_vmexit_ctrl());
4810 
4811 	/* 22.2.1, 20.8.1 */
4812 	vm_entry_controls_set(vmx, vmx_vmentry_ctrl());
4813 
4814 	vmx->vcpu.arch.cr0_guest_owned_bits = vmx_l1_guest_owned_cr0_bits();
4815 	vmcs_writel(CR0_GUEST_HOST_MASK, ~vmx->vcpu.arch.cr0_guest_owned_bits);
4816 
4817 	set_cr4_guest_host_mask(vmx);
4818 
4819 	if (vmx->vpid != 0)
4820 		vmcs_write16(VIRTUAL_PROCESSOR_ID, vmx->vpid);
4821 
4822 	if (cpu_has_vmx_xsaves())
4823 		vmcs_write64(XSS_EXIT_BITMAP, VMX_XSS_EXIT_BITMAP);
4824 
4825 	if (enable_pml) {
4826 		vmcs_write64(PML_ADDRESS, page_to_phys(vmx->pml_pg));
4827 		vmcs_write16(GUEST_PML_INDEX, PML_HEAD_INDEX);
4828 	}
4829 
4830 	vmx_write_encls_bitmap(&vmx->vcpu, NULL);
4831 
4832 	if (vmx_pt_mode_is_host_guest()) {
4833 		memset(&vmx->pt_desc, 0, sizeof(vmx->pt_desc));
4834 		/* Bit[6~0] are forced to 1, writes are ignored. */
4835 		vmx->pt_desc.guest.output_mask = 0x7F;
4836 		vmcs_write64(GUEST_IA32_RTIT_CTL, 0);
4837 	}
4838 
4839 	vmcs_write32(GUEST_SYSENTER_CS, 0);
4840 	vmcs_writel(GUEST_SYSENTER_ESP, 0);
4841 	vmcs_writel(GUEST_SYSENTER_EIP, 0);
4842 	vmcs_write64(GUEST_IA32_DEBUGCTL, 0);
4843 
4844 	if (cpu_has_vmx_tpr_shadow()) {
4845 		vmcs_write64(VIRTUAL_APIC_PAGE_ADDR, 0);
4846 		if (cpu_need_tpr_shadow(&vmx->vcpu))
4847 			vmcs_write64(VIRTUAL_APIC_PAGE_ADDR,
4848 				     __pa(vmx->vcpu.arch.apic->regs));
4849 		vmcs_write32(TPR_THRESHOLD, 0);
4850 	}
4851 
4852 	vmx_setup_uret_msrs(vmx);
4853 }
4854 
4855 static void __vmx_vcpu_reset(struct kvm_vcpu *vcpu)
4856 {
4857 	struct vcpu_vmx *vmx = to_vmx(vcpu);
4858 
4859 	init_vmcs(vmx);
4860 
4861 	if (nested &&
4862 	    kvm_check_has_quirk(vcpu->kvm, KVM_X86_QUIRK_STUFF_FEATURE_MSRS))
4863 		memcpy(&vmx->nested.msrs, &vmcs_config.nested, sizeof(vmx->nested.msrs));
4864 
4865 	vcpu_setup_sgx_lepubkeyhash(vcpu);
4866 
4867 	vmx->nested.posted_intr_nv = -1;
4868 	vmx->nested.vmxon_ptr = INVALID_GPA;
4869 	vmx->nested.current_vmptr = INVALID_GPA;
4870 
4871 #ifdef CONFIG_KVM_HYPERV
4872 	vmx->nested.hv_evmcs_vmptr = EVMPTR_INVALID;
4873 #endif
4874 
4875 	if (kvm_check_has_quirk(vcpu->kvm, KVM_X86_QUIRK_STUFF_FEATURE_MSRS))
4876 		vcpu->arch.microcode_version = 0x100000000ULL;
4877 	vmx->msr_ia32_feature_control_valid_bits = FEAT_CTL_LOCKED;
4878 
4879 	/*
4880 	 * Enforce invariant: pi_desc.nv is always either POSTED_INTR_VECTOR
4881 	 * or POSTED_INTR_WAKEUP_VECTOR.
4882 	 */
4883 	vmx->pi_desc.nv = POSTED_INTR_VECTOR;
4884 	__pi_set_sn(&vmx->pi_desc);
4885 }
4886 
4887 void vmx_vcpu_reset(struct kvm_vcpu *vcpu, bool init_event)
4888 {
4889 	struct vcpu_vmx *vmx = to_vmx(vcpu);
4890 
4891 	if (!init_event)
4892 		__vmx_vcpu_reset(vcpu);
4893 
4894 	vmx->rmode.vm86_active = 0;
4895 	vmx->spec_ctrl = 0;
4896 
4897 	vmx->msr_ia32_umwait_control = 0;
4898 
4899 	vmx->hv_deadline_tsc = -1;
4900 	kvm_set_cr8(vcpu, 0);
4901 
4902 	seg_setup(VCPU_SREG_CS);
4903 	vmcs_write16(GUEST_CS_SELECTOR, 0xf000);
4904 	vmcs_writel(GUEST_CS_BASE, 0xffff0000ul);
4905 
4906 	seg_setup(VCPU_SREG_DS);
4907 	seg_setup(VCPU_SREG_ES);
4908 	seg_setup(VCPU_SREG_FS);
4909 	seg_setup(VCPU_SREG_GS);
4910 	seg_setup(VCPU_SREG_SS);
4911 
4912 	vmcs_write16(GUEST_TR_SELECTOR, 0);
4913 	vmcs_writel(GUEST_TR_BASE, 0);
4914 	vmcs_write32(GUEST_TR_LIMIT, 0xffff);
4915 	vmcs_write32(GUEST_TR_AR_BYTES, 0x008b);
4916 
4917 	vmcs_write16(GUEST_LDTR_SELECTOR, 0);
4918 	vmcs_writel(GUEST_LDTR_BASE, 0);
4919 	vmcs_write32(GUEST_LDTR_LIMIT, 0xffff);
4920 	vmcs_write32(GUEST_LDTR_AR_BYTES, 0x00082);
4921 
4922 	vmcs_writel(GUEST_GDTR_BASE, 0);
4923 	vmcs_write32(GUEST_GDTR_LIMIT, 0xffff);
4924 
4925 	vmcs_writel(GUEST_IDTR_BASE, 0);
4926 	vmcs_write32(GUEST_IDTR_LIMIT, 0xffff);
4927 
4928 	vmx_segment_cache_clear(vmx);
4929 	kvm_register_mark_available(vcpu, VCPU_EXREG_SEGMENTS);
4930 
4931 	vmcs_write32(GUEST_ACTIVITY_STATE, GUEST_ACTIVITY_ACTIVE);
4932 	vmcs_write32(GUEST_INTERRUPTIBILITY_INFO, 0);
4933 	vmcs_writel(GUEST_PENDING_DBG_EXCEPTIONS, 0);
4934 	if (kvm_mpx_supported())
4935 		vmcs_write64(GUEST_BNDCFGS, 0);
4936 
4937 	vmcs_write32(VM_ENTRY_INTR_INFO_FIELD, 0);  /* 22.2.1 */
4938 
4939 	kvm_make_request(KVM_REQ_APIC_PAGE_RELOAD, vcpu);
4940 
4941 	vpid_sync_context(vmx->vpid);
4942 
4943 	vmx_update_fb_clear_dis(vcpu, vmx);
4944 }
4945 
4946 void vmx_enable_irq_window(struct kvm_vcpu *vcpu)
4947 {
4948 	exec_controls_setbit(to_vmx(vcpu), CPU_BASED_INTR_WINDOW_EXITING);
4949 }
4950 
4951 void vmx_enable_nmi_window(struct kvm_vcpu *vcpu)
4952 {
4953 	if (!enable_vnmi ||
4954 	    vmcs_read32(GUEST_INTERRUPTIBILITY_INFO) & GUEST_INTR_STATE_STI) {
4955 		vmx_enable_irq_window(vcpu);
4956 		return;
4957 	}
4958 
4959 	exec_controls_setbit(to_vmx(vcpu), CPU_BASED_NMI_WINDOW_EXITING);
4960 }
4961 
4962 void vmx_inject_irq(struct kvm_vcpu *vcpu, bool reinjected)
4963 {
4964 	struct vcpu_vmx *vmx = to_vmx(vcpu);
4965 	uint32_t intr;
4966 	int irq = vcpu->arch.interrupt.nr;
4967 
4968 	trace_kvm_inj_virq(irq, vcpu->arch.interrupt.soft, reinjected);
4969 
4970 	++vcpu->stat.irq_injections;
4971 	if (vmx->rmode.vm86_active) {
4972 		int inc_eip = 0;
4973 		if (vcpu->arch.interrupt.soft)
4974 			inc_eip = vcpu->arch.event_exit_inst_len;
4975 		kvm_inject_realmode_interrupt(vcpu, irq, inc_eip);
4976 		return;
4977 	}
4978 	intr = irq | INTR_INFO_VALID_MASK;
4979 	if (vcpu->arch.interrupt.soft) {
4980 		intr |= INTR_TYPE_SOFT_INTR;
4981 		vmcs_write32(VM_ENTRY_INSTRUCTION_LEN,
4982 			     vmx->vcpu.arch.event_exit_inst_len);
4983 	} else
4984 		intr |= INTR_TYPE_EXT_INTR;
4985 	vmcs_write32(VM_ENTRY_INTR_INFO_FIELD, intr);
4986 
4987 	vmx_clear_hlt(vcpu);
4988 }
4989 
4990 void vmx_inject_nmi(struct kvm_vcpu *vcpu)
4991 {
4992 	struct vcpu_vmx *vmx = to_vmx(vcpu);
4993 
4994 	if (!enable_vnmi) {
4995 		/*
4996 		 * Tracking the NMI-blocked state in software is built upon
4997 		 * finding the next open IRQ window. This, in turn, depends on
4998 		 * well-behaving guests: They have to keep IRQs disabled at
4999 		 * least as long as the NMI handler runs. Otherwise we may
5000 		 * cause NMI nesting, maybe breaking the guest. But as this is
5001 		 * highly unlikely, we can live with the residual risk.
5002 		 */
5003 		vmx->loaded_vmcs->soft_vnmi_blocked = 1;
5004 		vmx->loaded_vmcs->vnmi_blocked_time = 0;
5005 	}
5006 
5007 	++vcpu->stat.nmi_injections;
5008 	vmx->loaded_vmcs->nmi_known_unmasked = false;
5009 
5010 	if (vmx->rmode.vm86_active) {
5011 		kvm_inject_realmode_interrupt(vcpu, NMI_VECTOR, 0);
5012 		return;
5013 	}
5014 
5015 	vmcs_write32(VM_ENTRY_INTR_INFO_FIELD,
5016 			INTR_TYPE_NMI_INTR | INTR_INFO_VALID_MASK | NMI_VECTOR);
5017 
5018 	vmx_clear_hlt(vcpu);
5019 }
5020 
5021 bool vmx_get_nmi_mask(struct kvm_vcpu *vcpu)
5022 {
5023 	struct vcpu_vmx *vmx = to_vmx(vcpu);
5024 	bool masked;
5025 
5026 	if (!enable_vnmi)
5027 		return vmx->loaded_vmcs->soft_vnmi_blocked;
5028 	if (vmx->loaded_vmcs->nmi_known_unmasked)
5029 		return false;
5030 	masked = vmcs_read32(GUEST_INTERRUPTIBILITY_INFO) & GUEST_INTR_STATE_NMI;
5031 	vmx->loaded_vmcs->nmi_known_unmasked = !masked;
5032 	return masked;
5033 }
5034 
5035 void vmx_set_nmi_mask(struct kvm_vcpu *vcpu, bool masked)
5036 {
5037 	struct vcpu_vmx *vmx = to_vmx(vcpu);
5038 
5039 	if (!enable_vnmi) {
5040 		if (vmx->loaded_vmcs->soft_vnmi_blocked != masked) {
5041 			vmx->loaded_vmcs->soft_vnmi_blocked = masked;
5042 			vmx->loaded_vmcs->vnmi_blocked_time = 0;
5043 		}
5044 	} else {
5045 		vmx->loaded_vmcs->nmi_known_unmasked = !masked;
5046 		if (masked)
5047 			vmcs_set_bits(GUEST_INTERRUPTIBILITY_INFO,
5048 				      GUEST_INTR_STATE_NMI);
5049 		else
5050 			vmcs_clear_bits(GUEST_INTERRUPTIBILITY_INFO,
5051 					GUEST_INTR_STATE_NMI);
5052 	}
5053 }
5054 
5055 bool vmx_nmi_blocked(struct kvm_vcpu *vcpu)
5056 {
5057 	if (is_guest_mode(vcpu) && nested_exit_on_nmi(vcpu))
5058 		return false;
5059 
5060 	if (!enable_vnmi && to_vmx(vcpu)->loaded_vmcs->soft_vnmi_blocked)
5061 		return true;
5062 
5063 	return (vmcs_read32(GUEST_INTERRUPTIBILITY_INFO) &
5064 		(GUEST_INTR_STATE_MOV_SS | GUEST_INTR_STATE_STI |
5065 		 GUEST_INTR_STATE_NMI));
5066 }
5067 
5068 int vmx_nmi_allowed(struct kvm_vcpu *vcpu, bool for_injection)
5069 {
5070 	if (to_vmx(vcpu)->nested.nested_run_pending)
5071 		return -EBUSY;
5072 
5073 	/* An NMI must not be injected into L2 if it's supposed to VM-Exit.  */
5074 	if (for_injection && is_guest_mode(vcpu) && nested_exit_on_nmi(vcpu))
5075 		return -EBUSY;
5076 
5077 	return !vmx_nmi_blocked(vcpu);
5078 }
5079 
5080 bool __vmx_interrupt_blocked(struct kvm_vcpu *vcpu)
5081 {
5082 	return !(vmx_get_rflags(vcpu) & X86_EFLAGS_IF) ||
5083 	       (vmcs_read32(GUEST_INTERRUPTIBILITY_INFO) &
5084 		(GUEST_INTR_STATE_STI | GUEST_INTR_STATE_MOV_SS));
5085 }
5086 
5087 bool vmx_interrupt_blocked(struct kvm_vcpu *vcpu)
5088 {
5089 	if (is_guest_mode(vcpu) && nested_exit_on_intr(vcpu))
5090 		return false;
5091 
5092 	return __vmx_interrupt_blocked(vcpu);
5093 }
5094 
5095 int vmx_interrupt_allowed(struct kvm_vcpu *vcpu, bool for_injection)
5096 {
5097 	if (to_vmx(vcpu)->nested.nested_run_pending)
5098 		return -EBUSY;
5099 
5100 	/*
5101 	 * An IRQ must not be injected into L2 if it's supposed to VM-Exit,
5102 	 * e.g. if the IRQ arrived asynchronously after checking nested events.
5103 	 */
5104 	if (for_injection && is_guest_mode(vcpu) && nested_exit_on_intr(vcpu))
5105 		return -EBUSY;
5106 
5107 	return !vmx_interrupt_blocked(vcpu);
5108 }
5109 
5110 int vmx_set_tss_addr(struct kvm *kvm, unsigned int addr)
5111 {
5112 	void __user *ret;
5113 
5114 	if (enable_unrestricted_guest)
5115 		return 0;
5116 
5117 	mutex_lock(&kvm->slots_lock);
5118 	ret = __x86_set_memory_region(kvm, TSS_PRIVATE_MEMSLOT, addr,
5119 				      PAGE_SIZE * 3);
5120 	mutex_unlock(&kvm->slots_lock);
5121 
5122 	if (IS_ERR(ret))
5123 		return PTR_ERR(ret);
5124 
5125 	to_kvm_vmx(kvm)->tss_addr = addr;
5126 
5127 	return init_rmode_tss(kvm, ret);
5128 }
5129 
5130 int vmx_set_identity_map_addr(struct kvm *kvm, u64 ident_addr)
5131 {
5132 	to_kvm_vmx(kvm)->ept_identity_map_addr = ident_addr;
5133 	return 0;
5134 }
5135 
5136 static bool rmode_exception(struct kvm_vcpu *vcpu, int vec)
5137 {
5138 	switch (vec) {
5139 	case BP_VECTOR:
5140 		/*
5141 		 * Update instruction length as we may reinject the exception
5142 		 * from user space while in guest debugging mode.
5143 		 */
5144 		to_vmx(vcpu)->vcpu.arch.event_exit_inst_len =
5145 			vmcs_read32(VM_EXIT_INSTRUCTION_LEN);
5146 		if (vcpu->guest_debug & KVM_GUESTDBG_USE_SW_BP)
5147 			return false;
5148 		fallthrough;
5149 	case DB_VECTOR:
5150 		return !(vcpu->guest_debug &
5151 			(KVM_GUESTDBG_SINGLESTEP | KVM_GUESTDBG_USE_HW_BP));
5152 	case DE_VECTOR:
5153 	case OF_VECTOR:
5154 	case BR_VECTOR:
5155 	case UD_VECTOR:
5156 	case DF_VECTOR:
5157 	case SS_VECTOR:
5158 	case GP_VECTOR:
5159 	case MF_VECTOR:
5160 		return true;
5161 	}
5162 	return false;
5163 }
5164 
5165 static int handle_rmode_exception(struct kvm_vcpu *vcpu,
5166 				  int vec, u32 err_code)
5167 {
5168 	/*
5169 	 * Instruction with address size override prefix opcode 0x67
5170 	 * Cause the #SS fault with 0 error code in VM86 mode.
5171 	 */
5172 	if (((vec == GP_VECTOR) || (vec == SS_VECTOR)) && err_code == 0) {
5173 		if (kvm_emulate_instruction(vcpu, 0)) {
5174 			if (vcpu->arch.halt_request) {
5175 				vcpu->arch.halt_request = 0;
5176 				return kvm_emulate_halt_noskip(vcpu);
5177 			}
5178 			return 1;
5179 		}
5180 		return 0;
5181 	}
5182 
5183 	/*
5184 	 * Forward all other exceptions that are valid in real mode.
5185 	 * FIXME: Breaks guest debugging in real mode, needs to be fixed with
5186 	 *        the required debugging infrastructure rework.
5187 	 */
5188 	kvm_queue_exception(vcpu, vec);
5189 	return 1;
5190 }
5191 
5192 static int handle_machine_check(struct kvm_vcpu *vcpu)
5193 {
5194 	/* handled by vmx_vcpu_run() */
5195 	return 1;
5196 }
5197 
5198 /*
5199  * If the host has split lock detection disabled, then #AC is
5200  * unconditionally injected into the guest, which is the pre split lock
5201  * detection behaviour.
5202  *
5203  * If the host has split lock detection enabled then #AC is
5204  * only injected into the guest when:
5205  *  - Guest CPL == 3 (user mode)
5206  *  - Guest has #AC detection enabled in CR0
5207  *  - Guest EFLAGS has AC bit set
5208  */
5209 bool vmx_guest_inject_ac(struct kvm_vcpu *vcpu)
5210 {
5211 	if (!boot_cpu_has(X86_FEATURE_SPLIT_LOCK_DETECT))
5212 		return true;
5213 
5214 	return vmx_get_cpl(vcpu) == 3 && kvm_is_cr0_bit_set(vcpu, X86_CR0_AM) &&
5215 	       (kvm_get_rflags(vcpu) & X86_EFLAGS_AC);
5216 }
5217 
5218 static int handle_exception_nmi(struct kvm_vcpu *vcpu)
5219 {
5220 	struct vcpu_vmx *vmx = to_vmx(vcpu);
5221 	struct kvm_run *kvm_run = vcpu->run;
5222 	u32 intr_info, ex_no, error_code;
5223 	unsigned long cr2, dr6;
5224 	u32 vect_info;
5225 
5226 	vect_info = vmx->idt_vectoring_info;
5227 	intr_info = vmx_get_intr_info(vcpu);
5228 
5229 	/*
5230 	 * Machine checks are handled by handle_exception_irqoff(), or by
5231 	 * vmx_vcpu_run() if a #MC occurs on VM-Entry.  NMIs are handled by
5232 	 * vmx_vcpu_enter_exit().
5233 	 */
5234 	if (is_machine_check(intr_info) || is_nmi(intr_info))
5235 		return 1;
5236 
5237 	/*
5238 	 * Queue the exception here instead of in handle_nm_fault_irqoff().
5239 	 * This ensures the nested_vmx check is not skipped so vmexit can
5240 	 * be reflected to L1 (when it intercepts #NM) before reaching this
5241 	 * point.
5242 	 */
5243 	if (is_nm_fault(intr_info)) {
5244 		kvm_queue_exception(vcpu, NM_VECTOR);
5245 		return 1;
5246 	}
5247 
5248 	if (is_invalid_opcode(intr_info))
5249 		return handle_ud(vcpu);
5250 
5251 	if (WARN_ON_ONCE(is_ve_fault(intr_info))) {
5252 		struct vmx_ve_information *ve_info = vmx->ve_info;
5253 
5254 		WARN_ONCE(ve_info->exit_reason != EXIT_REASON_EPT_VIOLATION,
5255 			  "Unexpected #VE on VM-Exit reason 0x%x", ve_info->exit_reason);
5256 		dump_vmcs(vcpu);
5257 		kvm_mmu_print_sptes(vcpu, ve_info->guest_physical_address, "#VE");
5258 		return 1;
5259 	}
5260 
5261 	error_code = 0;
5262 	if (intr_info & INTR_INFO_DELIVER_CODE_MASK)
5263 		error_code = vmcs_read32(VM_EXIT_INTR_ERROR_CODE);
5264 
5265 	if (!vmx->rmode.vm86_active && is_gp_fault(intr_info)) {
5266 		WARN_ON_ONCE(!enable_vmware_backdoor);
5267 
5268 		/*
5269 		 * VMware backdoor emulation on #GP interception only handles
5270 		 * IN{S}, OUT{S}, and RDPMC, none of which generate a non-zero
5271 		 * error code on #GP.
5272 		 */
5273 		if (error_code) {
5274 			kvm_queue_exception_e(vcpu, GP_VECTOR, error_code);
5275 			return 1;
5276 		}
5277 		return kvm_emulate_instruction(vcpu, EMULTYPE_VMWARE_GP);
5278 	}
5279 
5280 	/*
5281 	 * The #PF with PFEC.RSVD = 1 indicates the guest is accessing
5282 	 * MMIO, it is better to report an internal error.
5283 	 * See the comments in vmx_handle_exit.
5284 	 */
5285 	if ((vect_info & VECTORING_INFO_VALID_MASK) &&
5286 	    !(is_page_fault(intr_info) && !(error_code & PFERR_RSVD_MASK))) {
5287 		vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
5288 		vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_SIMUL_EX;
5289 		vcpu->run->internal.ndata = 4;
5290 		vcpu->run->internal.data[0] = vect_info;
5291 		vcpu->run->internal.data[1] = intr_info;
5292 		vcpu->run->internal.data[2] = error_code;
5293 		vcpu->run->internal.data[3] = vcpu->arch.last_vmentry_cpu;
5294 		return 0;
5295 	}
5296 
5297 	if (is_page_fault(intr_info)) {
5298 		cr2 = vmx_get_exit_qual(vcpu);
5299 		if (enable_ept && !vcpu->arch.apf.host_apf_flags) {
5300 			/*
5301 			 * EPT will cause page fault only if we need to
5302 			 * detect illegal GPAs.
5303 			 */
5304 			WARN_ON_ONCE(!allow_smaller_maxphyaddr);
5305 			kvm_fixup_and_inject_pf_error(vcpu, cr2, error_code);
5306 			return 1;
5307 		} else
5308 			return kvm_handle_page_fault(vcpu, error_code, cr2, NULL, 0);
5309 	}
5310 
5311 	ex_no = intr_info & INTR_INFO_VECTOR_MASK;
5312 
5313 	if (vmx->rmode.vm86_active && rmode_exception(vcpu, ex_no))
5314 		return handle_rmode_exception(vcpu, ex_no, error_code);
5315 
5316 	switch (ex_no) {
5317 	case DB_VECTOR:
5318 		dr6 = vmx_get_exit_qual(vcpu);
5319 		if (!(vcpu->guest_debug &
5320 		      (KVM_GUESTDBG_SINGLESTEP | KVM_GUESTDBG_USE_HW_BP))) {
5321 			/*
5322 			 * If the #DB was due to ICEBP, a.k.a. INT1, skip the
5323 			 * instruction.  ICEBP generates a trap-like #DB, but
5324 			 * despite its interception control being tied to #DB,
5325 			 * is an instruction intercept, i.e. the VM-Exit occurs
5326 			 * on the ICEBP itself.  Use the inner "skip" helper to
5327 			 * avoid single-step #DB and MTF updates, as ICEBP is
5328 			 * higher priority.  Note, skipping ICEBP still clears
5329 			 * STI and MOVSS blocking.
5330 			 *
5331 			 * For all other #DBs, set vmcs.PENDING_DBG_EXCEPTIONS.BS
5332 			 * if single-step is enabled in RFLAGS and STI or MOVSS
5333 			 * blocking is active, as the CPU doesn't set the bit
5334 			 * on VM-Exit due to #DB interception.  VM-Entry has a
5335 			 * consistency check that a single-step #DB is pending
5336 			 * in this scenario as the previous instruction cannot
5337 			 * have toggled RFLAGS.TF 0=>1 (because STI and POP/MOV
5338 			 * don't modify RFLAGS), therefore the one instruction
5339 			 * delay when activating single-step breakpoints must
5340 			 * have already expired.  Note, the CPU sets/clears BS
5341 			 * as appropriate for all other VM-Exits types.
5342 			 */
5343 			if (is_icebp(intr_info))
5344 				WARN_ON(!skip_emulated_instruction(vcpu));
5345 			else if ((vmx_get_rflags(vcpu) & X86_EFLAGS_TF) &&
5346 				 (vmcs_read32(GUEST_INTERRUPTIBILITY_INFO) &
5347 				  (GUEST_INTR_STATE_STI | GUEST_INTR_STATE_MOV_SS)))
5348 				vmcs_writel(GUEST_PENDING_DBG_EXCEPTIONS,
5349 					    vmcs_readl(GUEST_PENDING_DBG_EXCEPTIONS) | DR6_BS);
5350 
5351 			kvm_queue_exception_p(vcpu, DB_VECTOR, dr6);
5352 			return 1;
5353 		}
5354 		kvm_run->debug.arch.dr6 = dr6 | DR6_ACTIVE_LOW;
5355 		kvm_run->debug.arch.dr7 = vmcs_readl(GUEST_DR7);
5356 		fallthrough;
5357 	case BP_VECTOR:
5358 		/*
5359 		 * Update instruction length as we may reinject #BP from
5360 		 * user space while in guest debugging mode. Reading it for
5361 		 * #DB as well causes no harm, it is not used in that case.
5362 		 */
5363 		vmx->vcpu.arch.event_exit_inst_len =
5364 			vmcs_read32(VM_EXIT_INSTRUCTION_LEN);
5365 		kvm_run->exit_reason = KVM_EXIT_DEBUG;
5366 		kvm_run->debug.arch.pc = kvm_get_linear_rip(vcpu);
5367 		kvm_run->debug.arch.exception = ex_no;
5368 		break;
5369 	case AC_VECTOR:
5370 		if (vmx_guest_inject_ac(vcpu)) {
5371 			kvm_queue_exception_e(vcpu, AC_VECTOR, error_code);
5372 			return 1;
5373 		}
5374 
5375 		/*
5376 		 * Handle split lock. Depending on detection mode this will
5377 		 * either warn and disable split lock detection for this
5378 		 * task or force SIGBUS on it.
5379 		 */
5380 		if (handle_guest_split_lock(kvm_rip_read(vcpu)))
5381 			return 1;
5382 		fallthrough;
5383 	default:
5384 		kvm_run->exit_reason = KVM_EXIT_EXCEPTION;
5385 		kvm_run->ex.exception = ex_no;
5386 		kvm_run->ex.error_code = error_code;
5387 		break;
5388 	}
5389 	return 0;
5390 }
5391 
5392 static __always_inline int handle_external_interrupt(struct kvm_vcpu *vcpu)
5393 {
5394 	++vcpu->stat.irq_exits;
5395 	return 1;
5396 }
5397 
5398 static int handle_triple_fault(struct kvm_vcpu *vcpu)
5399 {
5400 	vcpu->run->exit_reason = KVM_EXIT_SHUTDOWN;
5401 	vcpu->mmio_needed = 0;
5402 	return 0;
5403 }
5404 
5405 static int handle_io(struct kvm_vcpu *vcpu)
5406 {
5407 	unsigned long exit_qualification;
5408 	int size, in, string;
5409 	unsigned port;
5410 
5411 	exit_qualification = vmx_get_exit_qual(vcpu);
5412 	string = (exit_qualification & 16) != 0;
5413 
5414 	++vcpu->stat.io_exits;
5415 
5416 	if (string)
5417 		return kvm_emulate_instruction(vcpu, 0);
5418 
5419 	port = exit_qualification >> 16;
5420 	size = (exit_qualification & 7) + 1;
5421 	in = (exit_qualification & 8) != 0;
5422 
5423 	return kvm_fast_pio(vcpu, size, port, in);
5424 }
5425 
5426 void vmx_patch_hypercall(struct kvm_vcpu *vcpu, unsigned char *hypercall)
5427 {
5428 	/*
5429 	 * Patch in the VMCALL instruction:
5430 	 */
5431 	hypercall[0] = 0x0f;
5432 	hypercall[1] = 0x01;
5433 	hypercall[2] = 0xc1;
5434 }
5435 
5436 /* called to set cr0 as appropriate for a mov-to-cr0 exit. */
5437 static int handle_set_cr0(struct kvm_vcpu *vcpu, unsigned long val)
5438 {
5439 	if (is_guest_mode(vcpu)) {
5440 		struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
5441 		unsigned long orig_val = val;
5442 
5443 		/*
5444 		 * We get here when L2 changed cr0 in a way that did not change
5445 		 * any of L1's shadowed bits (see nested_vmx_exit_handled_cr),
5446 		 * but did change L0 shadowed bits. So we first calculate the
5447 		 * effective cr0 value that L1 would like to write into the
5448 		 * hardware. It consists of the L2-owned bits from the new
5449 		 * value combined with the L1-owned bits from L1's guest_cr0.
5450 		 */
5451 		val = (val & ~vmcs12->cr0_guest_host_mask) |
5452 			(vmcs12->guest_cr0 & vmcs12->cr0_guest_host_mask);
5453 
5454 		if (kvm_set_cr0(vcpu, val))
5455 			return 1;
5456 		vmcs_writel(CR0_READ_SHADOW, orig_val);
5457 		return 0;
5458 	} else {
5459 		return kvm_set_cr0(vcpu, val);
5460 	}
5461 }
5462 
5463 static int handle_set_cr4(struct kvm_vcpu *vcpu, unsigned long val)
5464 {
5465 	if (is_guest_mode(vcpu)) {
5466 		struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
5467 		unsigned long orig_val = val;
5468 
5469 		/* analogously to handle_set_cr0 */
5470 		val = (val & ~vmcs12->cr4_guest_host_mask) |
5471 			(vmcs12->guest_cr4 & vmcs12->cr4_guest_host_mask);
5472 		if (kvm_set_cr4(vcpu, val))
5473 			return 1;
5474 		vmcs_writel(CR4_READ_SHADOW, orig_val);
5475 		return 0;
5476 	} else
5477 		return kvm_set_cr4(vcpu, val);
5478 }
5479 
5480 static int handle_desc(struct kvm_vcpu *vcpu)
5481 {
5482 	/*
5483 	 * UMIP emulation relies on intercepting writes to CR4.UMIP, i.e. this
5484 	 * and other code needs to be updated if UMIP can be guest owned.
5485 	 */
5486 	BUILD_BUG_ON(KVM_POSSIBLE_CR4_GUEST_BITS & X86_CR4_UMIP);
5487 
5488 	WARN_ON_ONCE(!kvm_is_cr4_bit_set(vcpu, X86_CR4_UMIP));
5489 	return kvm_emulate_instruction(vcpu, 0);
5490 }
5491 
5492 static int handle_cr(struct kvm_vcpu *vcpu)
5493 {
5494 	unsigned long exit_qualification, val;
5495 	int cr;
5496 	int reg;
5497 	int err;
5498 	int ret;
5499 
5500 	exit_qualification = vmx_get_exit_qual(vcpu);
5501 	cr = exit_qualification & 15;
5502 	reg = (exit_qualification >> 8) & 15;
5503 	switch ((exit_qualification >> 4) & 3) {
5504 	case 0: /* mov to cr */
5505 		val = kvm_register_read(vcpu, reg);
5506 		trace_kvm_cr_write(cr, val);
5507 		switch (cr) {
5508 		case 0:
5509 			err = handle_set_cr0(vcpu, val);
5510 			return kvm_complete_insn_gp(vcpu, err);
5511 		case 3:
5512 			WARN_ON_ONCE(enable_unrestricted_guest);
5513 
5514 			err = kvm_set_cr3(vcpu, val);
5515 			return kvm_complete_insn_gp(vcpu, err);
5516 		case 4:
5517 			err = handle_set_cr4(vcpu, val);
5518 			return kvm_complete_insn_gp(vcpu, err);
5519 		case 8: {
5520 				u8 cr8_prev = kvm_get_cr8(vcpu);
5521 				u8 cr8 = (u8)val;
5522 				err = kvm_set_cr8(vcpu, cr8);
5523 				ret = kvm_complete_insn_gp(vcpu, err);
5524 				if (lapic_in_kernel(vcpu))
5525 					return ret;
5526 				if (cr8_prev <= cr8)
5527 					return ret;
5528 				/*
5529 				 * TODO: we might be squashing a
5530 				 * KVM_GUESTDBG_SINGLESTEP-triggered
5531 				 * KVM_EXIT_DEBUG here.
5532 				 */
5533 				vcpu->run->exit_reason = KVM_EXIT_SET_TPR;
5534 				return 0;
5535 			}
5536 		}
5537 		break;
5538 	case 2: /* clts */
5539 		KVM_BUG(1, vcpu->kvm, "Guest always owns CR0.TS");
5540 		return -EIO;
5541 	case 1: /*mov from cr*/
5542 		switch (cr) {
5543 		case 3:
5544 			WARN_ON_ONCE(enable_unrestricted_guest);
5545 
5546 			val = kvm_read_cr3(vcpu);
5547 			kvm_register_write(vcpu, reg, val);
5548 			trace_kvm_cr_read(cr, val);
5549 			return kvm_skip_emulated_instruction(vcpu);
5550 		case 8:
5551 			val = kvm_get_cr8(vcpu);
5552 			kvm_register_write(vcpu, reg, val);
5553 			trace_kvm_cr_read(cr, val);
5554 			return kvm_skip_emulated_instruction(vcpu);
5555 		}
5556 		break;
5557 	case 3: /* lmsw */
5558 		val = (exit_qualification >> LMSW_SOURCE_DATA_SHIFT) & 0x0f;
5559 		trace_kvm_cr_write(0, (kvm_read_cr0_bits(vcpu, ~0xful) | val));
5560 		kvm_lmsw(vcpu, val);
5561 
5562 		return kvm_skip_emulated_instruction(vcpu);
5563 	default:
5564 		break;
5565 	}
5566 	vcpu->run->exit_reason = 0;
5567 	vcpu_unimpl(vcpu, "unhandled control register: op %d cr %d\n",
5568 	       (int)(exit_qualification >> 4) & 3, cr);
5569 	return 0;
5570 }
5571 
5572 static int handle_dr(struct kvm_vcpu *vcpu)
5573 {
5574 	unsigned long exit_qualification;
5575 	int dr, dr7, reg;
5576 	int err = 1;
5577 
5578 	exit_qualification = vmx_get_exit_qual(vcpu);
5579 	dr = exit_qualification & DEBUG_REG_ACCESS_NUM;
5580 
5581 	/* First, if DR does not exist, trigger UD */
5582 	if (!kvm_require_dr(vcpu, dr))
5583 		return 1;
5584 
5585 	if (vmx_get_cpl(vcpu) > 0)
5586 		goto out;
5587 
5588 	dr7 = vmcs_readl(GUEST_DR7);
5589 	if (dr7 & DR7_GD) {
5590 		/*
5591 		 * As the vm-exit takes precedence over the debug trap, we
5592 		 * need to emulate the latter, either for the host or the
5593 		 * guest debugging itself.
5594 		 */
5595 		if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP) {
5596 			vcpu->run->debug.arch.dr6 = DR6_BD | DR6_ACTIVE_LOW;
5597 			vcpu->run->debug.arch.dr7 = dr7;
5598 			vcpu->run->debug.arch.pc = kvm_get_linear_rip(vcpu);
5599 			vcpu->run->debug.arch.exception = DB_VECTOR;
5600 			vcpu->run->exit_reason = KVM_EXIT_DEBUG;
5601 			return 0;
5602 		} else {
5603 			kvm_queue_exception_p(vcpu, DB_VECTOR, DR6_BD);
5604 			return 1;
5605 		}
5606 	}
5607 
5608 	if (vcpu->guest_debug == 0) {
5609 		exec_controls_clearbit(to_vmx(vcpu), CPU_BASED_MOV_DR_EXITING);
5610 
5611 		/*
5612 		 * No more DR vmexits; force a reload of the debug registers
5613 		 * and reenter on this instruction.  The next vmexit will
5614 		 * retrieve the full state of the debug registers.
5615 		 */
5616 		vcpu->arch.switch_db_regs |= KVM_DEBUGREG_WONT_EXIT;
5617 		return 1;
5618 	}
5619 
5620 	reg = DEBUG_REG_ACCESS_REG(exit_qualification);
5621 	if (exit_qualification & TYPE_MOV_FROM_DR) {
5622 		kvm_register_write(vcpu, reg, kvm_get_dr(vcpu, dr));
5623 		err = 0;
5624 	} else {
5625 		err = kvm_set_dr(vcpu, dr, kvm_register_read(vcpu, reg));
5626 	}
5627 
5628 out:
5629 	return kvm_complete_insn_gp(vcpu, err);
5630 }
5631 
5632 void vmx_sync_dirty_debug_regs(struct kvm_vcpu *vcpu)
5633 {
5634 	get_debugreg(vcpu->arch.db[0], 0);
5635 	get_debugreg(vcpu->arch.db[1], 1);
5636 	get_debugreg(vcpu->arch.db[2], 2);
5637 	get_debugreg(vcpu->arch.db[3], 3);
5638 	get_debugreg(vcpu->arch.dr6, 6);
5639 	vcpu->arch.dr7 = vmcs_readl(GUEST_DR7);
5640 
5641 	vcpu->arch.switch_db_regs &= ~KVM_DEBUGREG_WONT_EXIT;
5642 	exec_controls_setbit(to_vmx(vcpu), CPU_BASED_MOV_DR_EXITING);
5643 
5644 	/*
5645 	 * exc_debug expects dr6 to be cleared after it runs, avoid that it sees
5646 	 * a stale dr6 from the guest.
5647 	 */
5648 	set_debugreg(DR6_RESERVED, 6);
5649 }
5650 
5651 void vmx_set_dr7(struct kvm_vcpu *vcpu, unsigned long val)
5652 {
5653 	vmcs_writel(GUEST_DR7, val);
5654 }
5655 
5656 static int handle_tpr_below_threshold(struct kvm_vcpu *vcpu)
5657 {
5658 	kvm_apic_update_ppr(vcpu);
5659 	return 1;
5660 }
5661 
5662 static int handle_interrupt_window(struct kvm_vcpu *vcpu)
5663 {
5664 	exec_controls_clearbit(to_vmx(vcpu), CPU_BASED_INTR_WINDOW_EXITING);
5665 
5666 	kvm_make_request(KVM_REQ_EVENT, vcpu);
5667 
5668 	++vcpu->stat.irq_window_exits;
5669 	return 1;
5670 }
5671 
5672 static int handle_invlpg(struct kvm_vcpu *vcpu)
5673 {
5674 	unsigned long exit_qualification = vmx_get_exit_qual(vcpu);
5675 
5676 	kvm_mmu_invlpg(vcpu, exit_qualification);
5677 	return kvm_skip_emulated_instruction(vcpu);
5678 }
5679 
5680 static int handle_apic_access(struct kvm_vcpu *vcpu)
5681 {
5682 	if (likely(fasteoi)) {
5683 		unsigned long exit_qualification = vmx_get_exit_qual(vcpu);
5684 		int access_type, offset;
5685 
5686 		access_type = exit_qualification & APIC_ACCESS_TYPE;
5687 		offset = exit_qualification & APIC_ACCESS_OFFSET;
5688 		/*
5689 		 * Sane guest uses MOV to write EOI, with written value
5690 		 * not cared. So make a short-circuit here by avoiding
5691 		 * heavy instruction emulation.
5692 		 */
5693 		if ((access_type == TYPE_LINEAR_APIC_INST_WRITE) &&
5694 		    (offset == APIC_EOI)) {
5695 			kvm_lapic_set_eoi(vcpu);
5696 			return kvm_skip_emulated_instruction(vcpu);
5697 		}
5698 	}
5699 	return kvm_emulate_instruction(vcpu, 0);
5700 }
5701 
5702 static int handle_apic_eoi_induced(struct kvm_vcpu *vcpu)
5703 {
5704 	unsigned long exit_qualification = vmx_get_exit_qual(vcpu);
5705 	int vector = exit_qualification & 0xff;
5706 
5707 	/* EOI-induced VM exit is trap-like and thus no need to adjust IP */
5708 	kvm_apic_set_eoi_accelerated(vcpu, vector);
5709 	return 1;
5710 }
5711 
5712 static int handle_apic_write(struct kvm_vcpu *vcpu)
5713 {
5714 	unsigned long exit_qualification = vmx_get_exit_qual(vcpu);
5715 
5716 	/*
5717 	 * APIC-write VM-Exit is trap-like, KVM doesn't need to advance RIP and
5718 	 * hardware has done any necessary aliasing, offset adjustments, etc...
5719 	 * for the access.  I.e. the correct value has already been  written to
5720 	 * the vAPIC page for the correct 16-byte chunk.  KVM needs only to
5721 	 * retrieve the register value and emulate the access.
5722 	 */
5723 	u32 offset = exit_qualification & 0xff0;
5724 
5725 	kvm_apic_write_nodecode(vcpu, offset);
5726 	return 1;
5727 }
5728 
5729 static int handle_task_switch(struct kvm_vcpu *vcpu)
5730 {
5731 	struct vcpu_vmx *vmx = to_vmx(vcpu);
5732 	unsigned long exit_qualification;
5733 	bool has_error_code = false;
5734 	u32 error_code = 0;
5735 	u16 tss_selector;
5736 	int reason, type, idt_v, idt_index;
5737 
5738 	idt_v = (vmx->idt_vectoring_info & VECTORING_INFO_VALID_MASK);
5739 	idt_index = (vmx->idt_vectoring_info & VECTORING_INFO_VECTOR_MASK);
5740 	type = (vmx->idt_vectoring_info & VECTORING_INFO_TYPE_MASK);
5741 
5742 	exit_qualification = vmx_get_exit_qual(vcpu);
5743 
5744 	reason = (u32)exit_qualification >> 30;
5745 	if (reason == TASK_SWITCH_GATE && idt_v) {
5746 		switch (type) {
5747 		case INTR_TYPE_NMI_INTR:
5748 			vcpu->arch.nmi_injected = false;
5749 			vmx_set_nmi_mask(vcpu, true);
5750 			break;
5751 		case INTR_TYPE_EXT_INTR:
5752 		case INTR_TYPE_SOFT_INTR:
5753 			kvm_clear_interrupt_queue(vcpu);
5754 			break;
5755 		case INTR_TYPE_HARD_EXCEPTION:
5756 			if (vmx->idt_vectoring_info &
5757 			    VECTORING_INFO_DELIVER_CODE_MASK) {
5758 				has_error_code = true;
5759 				error_code =
5760 					vmcs_read32(IDT_VECTORING_ERROR_CODE);
5761 			}
5762 			fallthrough;
5763 		case INTR_TYPE_SOFT_EXCEPTION:
5764 			kvm_clear_exception_queue(vcpu);
5765 			break;
5766 		default:
5767 			break;
5768 		}
5769 	}
5770 	tss_selector = exit_qualification;
5771 
5772 	if (!idt_v || (type != INTR_TYPE_HARD_EXCEPTION &&
5773 		       type != INTR_TYPE_EXT_INTR &&
5774 		       type != INTR_TYPE_NMI_INTR))
5775 		WARN_ON(!skip_emulated_instruction(vcpu));
5776 
5777 	/*
5778 	 * TODO: What about debug traps on tss switch?
5779 	 *       Are we supposed to inject them and update dr6?
5780 	 */
5781 	return kvm_task_switch(vcpu, tss_selector,
5782 			       type == INTR_TYPE_SOFT_INTR ? idt_index : -1,
5783 			       reason, has_error_code, error_code);
5784 }
5785 
5786 static int handle_ept_violation(struct kvm_vcpu *vcpu)
5787 {
5788 	unsigned long exit_qualification;
5789 	gpa_t gpa;
5790 	u64 error_code;
5791 
5792 	exit_qualification = vmx_get_exit_qual(vcpu);
5793 
5794 	/*
5795 	 * EPT violation happened while executing iret from NMI,
5796 	 * "blocked by NMI" bit has to be set before next VM entry.
5797 	 * There are errata that may cause this bit to not be set:
5798 	 * AAK134, BY25.
5799 	 */
5800 	if (!(to_vmx(vcpu)->idt_vectoring_info & VECTORING_INFO_VALID_MASK) &&
5801 			enable_vnmi &&
5802 			(exit_qualification & INTR_INFO_UNBLOCK_NMI))
5803 		vmcs_set_bits(GUEST_INTERRUPTIBILITY_INFO, GUEST_INTR_STATE_NMI);
5804 
5805 	gpa = vmcs_read64(GUEST_PHYSICAL_ADDRESS);
5806 	trace_kvm_page_fault(vcpu, gpa, exit_qualification);
5807 
5808 	/* Is it a read fault? */
5809 	error_code = (exit_qualification & EPT_VIOLATION_ACC_READ)
5810 		     ? PFERR_USER_MASK : 0;
5811 	/* Is it a write fault? */
5812 	error_code |= (exit_qualification & EPT_VIOLATION_ACC_WRITE)
5813 		      ? PFERR_WRITE_MASK : 0;
5814 	/* Is it a fetch fault? */
5815 	error_code |= (exit_qualification & EPT_VIOLATION_ACC_INSTR)
5816 		      ? PFERR_FETCH_MASK : 0;
5817 	/* ept page table entry is present? */
5818 	error_code |= (exit_qualification & EPT_VIOLATION_RWX_MASK)
5819 		      ? PFERR_PRESENT_MASK : 0;
5820 
5821 	if (error_code & EPT_VIOLATION_GVA_IS_VALID)
5822 		error_code |= (exit_qualification & EPT_VIOLATION_GVA_TRANSLATED) ?
5823 			      PFERR_GUEST_FINAL_MASK : PFERR_GUEST_PAGE_MASK;
5824 
5825 	/*
5826 	 * Check that the GPA doesn't exceed physical memory limits, as that is
5827 	 * a guest page fault.  We have to emulate the instruction here, because
5828 	 * if the illegal address is that of a paging structure, then
5829 	 * EPT_VIOLATION_ACC_WRITE bit is set.  Alternatively, if supported we
5830 	 * would also use advanced VM-exit information for EPT violations to
5831 	 * reconstruct the page fault error code.
5832 	 */
5833 	if (unlikely(allow_smaller_maxphyaddr && !kvm_vcpu_is_legal_gpa(vcpu, gpa)))
5834 		return kvm_emulate_instruction(vcpu, 0);
5835 
5836 	return kvm_mmu_page_fault(vcpu, gpa, error_code, NULL, 0);
5837 }
5838 
5839 static int handle_ept_misconfig(struct kvm_vcpu *vcpu)
5840 {
5841 	gpa_t gpa;
5842 
5843 	if (vmx_check_emulate_instruction(vcpu, EMULTYPE_PF, NULL, 0))
5844 		return 1;
5845 
5846 	/*
5847 	 * A nested guest cannot optimize MMIO vmexits, because we have an
5848 	 * nGPA here instead of the required GPA.
5849 	 */
5850 	gpa = vmcs_read64(GUEST_PHYSICAL_ADDRESS);
5851 	if (!is_guest_mode(vcpu) &&
5852 	    !kvm_io_bus_write(vcpu, KVM_FAST_MMIO_BUS, gpa, 0, NULL)) {
5853 		trace_kvm_fast_mmio(gpa);
5854 		return kvm_skip_emulated_instruction(vcpu);
5855 	}
5856 
5857 	return kvm_mmu_page_fault(vcpu, gpa, PFERR_RSVD_MASK, NULL, 0);
5858 }
5859 
5860 static int handle_nmi_window(struct kvm_vcpu *vcpu)
5861 {
5862 	if (KVM_BUG_ON(!enable_vnmi, vcpu->kvm))
5863 		return -EIO;
5864 
5865 	exec_controls_clearbit(to_vmx(vcpu), CPU_BASED_NMI_WINDOW_EXITING);
5866 	++vcpu->stat.nmi_window_exits;
5867 	kvm_make_request(KVM_REQ_EVENT, vcpu);
5868 
5869 	return 1;
5870 }
5871 
5872 static bool vmx_emulation_required_with_pending_exception(struct kvm_vcpu *vcpu)
5873 {
5874 	struct vcpu_vmx *vmx = to_vmx(vcpu);
5875 
5876 	return vmx->emulation_required && !vmx->rmode.vm86_active &&
5877 	       (kvm_is_exception_pending(vcpu) || vcpu->arch.exception.injected);
5878 }
5879 
5880 static int handle_invalid_guest_state(struct kvm_vcpu *vcpu)
5881 {
5882 	struct vcpu_vmx *vmx = to_vmx(vcpu);
5883 	bool intr_window_requested;
5884 	unsigned count = 130;
5885 
5886 	intr_window_requested = exec_controls_get(vmx) &
5887 				CPU_BASED_INTR_WINDOW_EXITING;
5888 
5889 	while (vmx->emulation_required && count-- != 0) {
5890 		if (intr_window_requested && !vmx_interrupt_blocked(vcpu))
5891 			return handle_interrupt_window(&vmx->vcpu);
5892 
5893 		if (kvm_test_request(KVM_REQ_EVENT, vcpu))
5894 			return 1;
5895 
5896 		if (!kvm_emulate_instruction(vcpu, 0))
5897 			return 0;
5898 
5899 		if (vmx_emulation_required_with_pending_exception(vcpu)) {
5900 			kvm_prepare_emulation_failure_exit(vcpu);
5901 			return 0;
5902 		}
5903 
5904 		if (vcpu->arch.halt_request) {
5905 			vcpu->arch.halt_request = 0;
5906 			return kvm_emulate_halt_noskip(vcpu);
5907 		}
5908 
5909 		/*
5910 		 * Note, return 1 and not 0, vcpu_run() will invoke
5911 		 * xfer_to_guest_mode() which will create a proper return
5912 		 * code.
5913 		 */
5914 		if (__xfer_to_guest_mode_work_pending())
5915 			return 1;
5916 	}
5917 
5918 	return 1;
5919 }
5920 
5921 int vmx_vcpu_pre_run(struct kvm_vcpu *vcpu)
5922 {
5923 	if (vmx_emulation_required_with_pending_exception(vcpu)) {
5924 		kvm_prepare_emulation_failure_exit(vcpu);
5925 		return 0;
5926 	}
5927 
5928 	return 1;
5929 }
5930 
5931 /*
5932  * Indicate a busy-waiting vcpu in spinlock. We do not enable the PAUSE
5933  * exiting, so only get here on cpu with PAUSE-Loop-Exiting.
5934  */
5935 static int handle_pause(struct kvm_vcpu *vcpu)
5936 {
5937 	if (!kvm_pause_in_guest(vcpu->kvm))
5938 		grow_ple_window(vcpu);
5939 
5940 	/*
5941 	 * Intel sdm vol3 ch-25.1.3 says: The "PAUSE-loop exiting"
5942 	 * VM-execution control is ignored if CPL > 0. OTOH, KVM
5943 	 * never set PAUSE_EXITING and just set PLE if supported,
5944 	 * so the vcpu must be CPL=0 if it gets a PAUSE exit.
5945 	 */
5946 	kvm_vcpu_on_spin(vcpu, true);
5947 	return kvm_skip_emulated_instruction(vcpu);
5948 }
5949 
5950 static int handle_monitor_trap(struct kvm_vcpu *vcpu)
5951 {
5952 	return 1;
5953 }
5954 
5955 static int handle_invpcid(struct kvm_vcpu *vcpu)
5956 {
5957 	u32 vmx_instruction_info;
5958 	unsigned long type;
5959 	gva_t gva;
5960 	struct {
5961 		u64 pcid;
5962 		u64 gla;
5963 	} operand;
5964 	int gpr_index;
5965 
5966 	if (!guest_cpu_cap_has(vcpu, X86_FEATURE_INVPCID)) {
5967 		kvm_queue_exception(vcpu, UD_VECTOR);
5968 		return 1;
5969 	}
5970 
5971 	vmx_instruction_info = vmcs_read32(VMX_INSTRUCTION_INFO);
5972 	gpr_index = vmx_get_instr_info_reg2(vmx_instruction_info);
5973 	type = kvm_register_read(vcpu, gpr_index);
5974 
5975 	/* According to the Intel instruction reference, the memory operand
5976 	 * is read even if it isn't needed (e.g., for type==all)
5977 	 */
5978 	if (get_vmx_mem_address(vcpu, vmx_get_exit_qual(vcpu),
5979 				vmx_instruction_info, false,
5980 				sizeof(operand), &gva))
5981 		return 1;
5982 
5983 	return kvm_handle_invpcid(vcpu, type, gva);
5984 }
5985 
5986 static int handle_pml_full(struct kvm_vcpu *vcpu)
5987 {
5988 	unsigned long exit_qualification;
5989 
5990 	trace_kvm_pml_full(vcpu->vcpu_id);
5991 
5992 	exit_qualification = vmx_get_exit_qual(vcpu);
5993 
5994 	/*
5995 	 * PML buffer FULL happened while executing iret from NMI,
5996 	 * "blocked by NMI" bit has to be set before next VM entry.
5997 	 */
5998 	if (!(to_vmx(vcpu)->idt_vectoring_info & VECTORING_INFO_VALID_MASK) &&
5999 			enable_vnmi &&
6000 			(exit_qualification & INTR_INFO_UNBLOCK_NMI))
6001 		vmcs_set_bits(GUEST_INTERRUPTIBILITY_INFO,
6002 				GUEST_INTR_STATE_NMI);
6003 
6004 	/*
6005 	 * PML buffer already flushed at beginning of VMEXIT. Nothing to do
6006 	 * here.., and there's no userspace involvement needed for PML.
6007 	 */
6008 	return 1;
6009 }
6010 
6011 static fastpath_t handle_fastpath_preemption_timer(struct kvm_vcpu *vcpu,
6012 						   bool force_immediate_exit)
6013 {
6014 	struct vcpu_vmx *vmx = to_vmx(vcpu);
6015 
6016 	/*
6017 	 * In the *extremely* unlikely scenario that this is a spurious VM-Exit
6018 	 * due to the timer expiring while it was "soft" disabled, just eat the
6019 	 * exit and re-enter the guest.
6020 	 */
6021 	if (unlikely(vmx->loaded_vmcs->hv_timer_soft_disabled))
6022 		return EXIT_FASTPATH_REENTER_GUEST;
6023 
6024 	/*
6025 	 * If the timer expired because KVM used it to force an immediate exit,
6026 	 * then mission accomplished.
6027 	 */
6028 	if (force_immediate_exit)
6029 		return EXIT_FASTPATH_EXIT_HANDLED;
6030 
6031 	/*
6032 	 * If L2 is active, go down the slow path as emulating the guest timer
6033 	 * expiration likely requires synthesizing a nested VM-Exit.
6034 	 */
6035 	if (is_guest_mode(vcpu))
6036 		return EXIT_FASTPATH_NONE;
6037 
6038 	kvm_lapic_expired_hv_timer(vcpu);
6039 	return EXIT_FASTPATH_REENTER_GUEST;
6040 }
6041 
6042 static int handle_preemption_timer(struct kvm_vcpu *vcpu)
6043 {
6044 	/*
6045 	 * This non-fastpath handler is reached if and only if the preemption
6046 	 * timer was being used to emulate a guest timer while L2 is active.
6047 	 * All other scenarios are supposed to be handled in the fastpath.
6048 	 */
6049 	WARN_ON_ONCE(!is_guest_mode(vcpu));
6050 	kvm_lapic_expired_hv_timer(vcpu);
6051 	return 1;
6052 }
6053 
6054 /*
6055  * When nested=0, all VMX instruction VM Exits filter here.  The handlers
6056  * are overwritten by nested_vmx_hardware_setup() when nested=1.
6057  */
6058 static int handle_vmx_instruction(struct kvm_vcpu *vcpu)
6059 {
6060 	kvm_queue_exception(vcpu, UD_VECTOR);
6061 	return 1;
6062 }
6063 
6064 #ifndef CONFIG_X86_SGX_KVM
6065 static int handle_encls(struct kvm_vcpu *vcpu)
6066 {
6067 	/*
6068 	 * SGX virtualization is disabled.  There is no software enable bit for
6069 	 * SGX, so KVM intercepts all ENCLS leafs and injects a #UD to prevent
6070 	 * the guest from executing ENCLS (when SGX is supported by hardware).
6071 	 */
6072 	kvm_queue_exception(vcpu, UD_VECTOR);
6073 	return 1;
6074 }
6075 #endif /* CONFIG_X86_SGX_KVM */
6076 
6077 static int handle_bus_lock_vmexit(struct kvm_vcpu *vcpu)
6078 {
6079 	/*
6080 	 * Hardware may or may not set the BUS_LOCK_DETECTED flag on BUS_LOCK
6081 	 * VM-Exits. Unconditionally set the flag here and leave the handling to
6082 	 * vmx_handle_exit().
6083 	 */
6084 	to_vmx(vcpu)->exit_reason.bus_lock_detected = true;
6085 	return 1;
6086 }
6087 
6088 static int handle_notify(struct kvm_vcpu *vcpu)
6089 {
6090 	unsigned long exit_qual = vmx_get_exit_qual(vcpu);
6091 	bool context_invalid = exit_qual & NOTIFY_VM_CONTEXT_INVALID;
6092 
6093 	++vcpu->stat.notify_window_exits;
6094 
6095 	/*
6096 	 * Notify VM exit happened while executing iret from NMI,
6097 	 * "blocked by NMI" bit has to be set before next VM entry.
6098 	 */
6099 	if (enable_vnmi && (exit_qual & INTR_INFO_UNBLOCK_NMI))
6100 		vmcs_set_bits(GUEST_INTERRUPTIBILITY_INFO,
6101 			      GUEST_INTR_STATE_NMI);
6102 
6103 	if (vcpu->kvm->arch.notify_vmexit_flags & KVM_X86_NOTIFY_VMEXIT_USER ||
6104 	    context_invalid) {
6105 		vcpu->run->exit_reason = KVM_EXIT_NOTIFY;
6106 		vcpu->run->notify.flags = context_invalid ?
6107 					  KVM_NOTIFY_CONTEXT_INVALID : 0;
6108 		return 0;
6109 	}
6110 
6111 	return 1;
6112 }
6113 
6114 /*
6115  * The exit handlers return 1 if the exit was handled fully and guest execution
6116  * may resume.  Otherwise they set the kvm_run parameter to indicate what needs
6117  * to be done to userspace and return 0.
6118  */
6119 static int (*kvm_vmx_exit_handlers[])(struct kvm_vcpu *vcpu) = {
6120 	[EXIT_REASON_EXCEPTION_NMI]           = handle_exception_nmi,
6121 	[EXIT_REASON_EXTERNAL_INTERRUPT]      = handle_external_interrupt,
6122 	[EXIT_REASON_TRIPLE_FAULT]            = handle_triple_fault,
6123 	[EXIT_REASON_NMI_WINDOW]	      = handle_nmi_window,
6124 	[EXIT_REASON_IO_INSTRUCTION]          = handle_io,
6125 	[EXIT_REASON_CR_ACCESS]               = handle_cr,
6126 	[EXIT_REASON_DR_ACCESS]               = handle_dr,
6127 	[EXIT_REASON_CPUID]                   = kvm_emulate_cpuid,
6128 	[EXIT_REASON_MSR_READ]                = kvm_emulate_rdmsr,
6129 	[EXIT_REASON_MSR_WRITE]               = kvm_emulate_wrmsr,
6130 	[EXIT_REASON_INTERRUPT_WINDOW]        = handle_interrupt_window,
6131 	[EXIT_REASON_HLT]                     = kvm_emulate_halt,
6132 	[EXIT_REASON_INVD]		      = kvm_emulate_invd,
6133 	[EXIT_REASON_INVLPG]		      = handle_invlpg,
6134 	[EXIT_REASON_RDPMC]                   = kvm_emulate_rdpmc,
6135 	[EXIT_REASON_VMCALL]                  = kvm_emulate_hypercall,
6136 	[EXIT_REASON_VMCLEAR]		      = handle_vmx_instruction,
6137 	[EXIT_REASON_VMLAUNCH]		      = handle_vmx_instruction,
6138 	[EXIT_REASON_VMPTRLD]		      = handle_vmx_instruction,
6139 	[EXIT_REASON_VMPTRST]		      = handle_vmx_instruction,
6140 	[EXIT_REASON_VMREAD]		      = handle_vmx_instruction,
6141 	[EXIT_REASON_VMRESUME]		      = handle_vmx_instruction,
6142 	[EXIT_REASON_VMWRITE]		      = handle_vmx_instruction,
6143 	[EXIT_REASON_VMOFF]		      = handle_vmx_instruction,
6144 	[EXIT_REASON_VMON]		      = handle_vmx_instruction,
6145 	[EXIT_REASON_TPR_BELOW_THRESHOLD]     = handle_tpr_below_threshold,
6146 	[EXIT_REASON_APIC_ACCESS]             = handle_apic_access,
6147 	[EXIT_REASON_APIC_WRITE]              = handle_apic_write,
6148 	[EXIT_REASON_EOI_INDUCED]             = handle_apic_eoi_induced,
6149 	[EXIT_REASON_WBINVD]                  = kvm_emulate_wbinvd,
6150 	[EXIT_REASON_XSETBV]                  = kvm_emulate_xsetbv,
6151 	[EXIT_REASON_TASK_SWITCH]             = handle_task_switch,
6152 	[EXIT_REASON_MCE_DURING_VMENTRY]      = handle_machine_check,
6153 	[EXIT_REASON_GDTR_IDTR]		      = handle_desc,
6154 	[EXIT_REASON_LDTR_TR]		      = handle_desc,
6155 	[EXIT_REASON_EPT_VIOLATION]	      = handle_ept_violation,
6156 	[EXIT_REASON_EPT_MISCONFIG]           = handle_ept_misconfig,
6157 	[EXIT_REASON_PAUSE_INSTRUCTION]       = handle_pause,
6158 	[EXIT_REASON_MWAIT_INSTRUCTION]	      = kvm_emulate_mwait,
6159 	[EXIT_REASON_MONITOR_TRAP_FLAG]       = handle_monitor_trap,
6160 	[EXIT_REASON_MONITOR_INSTRUCTION]     = kvm_emulate_monitor,
6161 	[EXIT_REASON_INVEPT]                  = handle_vmx_instruction,
6162 	[EXIT_REASON_INVVPID]                 = handle_vmx_instruction,
6163 	[EXIT_REASON_RDRAND]                  = kvm_handle_invalid_op,
6164 	[EXIT_REASON_RDSEED]                  = kvm_handle_invalid_op,
6165 	[EXIT_REASON_PML_FULL]		      = handle_pml_full,
6166 	[EXIT_REASON_INVPCID]                 = handle_invpcid,
6167 	[EXIT_REASON_VMFUNC]		      = handle_vmx_instruction,
6168 	[EXIT_REASON_PREEMPTION_TIMER]	      = handle_preemption_timer,
6169 	[EXIT_REASON_ENCLS]		      = handle_encls,
6170 	[EXIT_REASON_BUS_LOCK]                = handle_bus_lock_vmexit,
6171 	[EXIT_REASON_NOTIFY]		      = handle_notify,
6172 };
6173 
6174 static const int kvm_vmx_max_exit_handlers =
6175 	ARRAY_SIZE(kvm_vmx_exit_handlers);
6176 
6177 void vmx_get_exit_info(struct kvm_vcpu *vcpu, u32 *reason,
6178 		       u64 *info1, u64 *info2, u32 *intr_info, u32 *error_code)
6179 {
6180 	struct vcpu_vmx *vmx = to_vmx(vcpu);
6181 
6182 	*reason = vmx->exit_reason.full;
6183 	*info1 = vmx_get_exit_qual(vcpu);
6184 	if (!(vmx->exit_reason.failed_vmentry)) {
6185 		*info2 = vmx->idt_vectoring_info;
6186 		*intr_info = vmx_get_intr_info(vcpu);
6187 		if (is_exception_with_error_code(*intr_info))
6188 			*error_code = vmcs_read32(VM_EXIT_INTR_ERROR_CODE);
6189 		else
6190 			*error_code = 0;
6191 	} else {
6192 		*info2 = 0;
6193 		*intr_info = 0;
6194 		*error_code = 0;
6195 	}
6196 }
6197 
6198 void vmx_get_entry_info(struct kvm_vcpu *vcpu, u32 *intr_info, u32 *error_code)
6199 {
6200 	*intr_info = vmcs_read32(VM_ENTRY_INTR_INFO_FIELD);
6201 	if (is_exception_with_error_code(*intr_info))
6202 		*error_code = vmcs_read32(VM_ENTRY_EXCEPTION_ERROR_CODE);
6203 	else
6204 		*error_code = 0;
6205 }
6206 
6207 static void vmx_destroy_pml_buffer(struct vcpu_vmx *vmx)
6208 {
6209 	if (vmx->pml_pg) {
6210 		__free_page(vmx->pml_pg);
6211 		vmx->pml_pg = NULL;
6212 	}
6213 }
6214 
6215 static void vmx_flush_pml_buffer(struct kvm_vcpu *vcpu)
6216 {
6217 	struct vcpu_vmx *vmx = to_vmx(vcpu);
6218 	u16 pml_idx, pml_tail_index;
6219 	u64 *pml_buf;
6220 	int i;
6221 
6222 	pml_idx = vmcs_read16(GUEST_PML_INDEX);
6223 
6224 	/* Do nothing if PML buffer is empty */
6225 	if (pml_idx == PML_HEAD_INDEX)
6226 		return;
6227 	/*
6228 	 * PML index always points to the next available PML buffer entity
6229 	 * unless PML log has just overflowed.
6230 	 */
6231 	pml_tail_index = (pml_idx >= PML_LOG_NR_ENTRIES) ? 0 : pml_idx + 1;
6232 
6233 	/*
6234 	 * PML log is written backwards: the CPU first writes the entry 511
6235 	 * then the entry 510, and so on.
6236 	 *
6237 	 * Read the entries in the same order they were written, to ensure that
6238 	 * the dirty ring is filled in the same order the CPU wrote them.
6239 	 */
6240 	pml_buf = page_address(vmx->pml_pg);
6241 
6242 	for (i = PML_HEAD_INDEX; i >= pml_tail_index; i--) {
6243 		u64 gpa;
6244 
6245 		gpa = pml_buf[i];
6246 		WARN_ON(gpa & (PAGE_SIZE - 1));
6247 		kvm_vcpu_mark_page_dirty(vcpu, gpa >> PAGE_SHIFT);
6248 	}
6249 
6250 	/* reset PML index */
6251 	vmcs_write16(GUEST_PML_INDEX, PML_HEAD_INDEX);
6252 }
6253 
6254 static void vmx_dump_sel(char *name, uint32_t sel)
6255 {
6256 	pr_err("%s sel=0x%04x, attr=0x%05x, limit=0x%08x, base=0x%016lx\n",
6257 	       name, vmcs_read16(sel),
6258 	       vmcs_read32(sel + GUEST_ES_AR_BYTES - GUEST_ES_SELECTOR),
6259 	       vmcs_read32(sel + GUEST_ES_LIMIT - GUEST_ES_SELECTOR),
6260 	       vmcs_readl(sel + GUEST_ES_BASE - GUEST_ES_SELECTOR));
6261 }
6262 
6263 static void vmx_dump_dtsel(char *name, uint32_t limit)
6264 {
6265 	pr_err("%s                           limit=0x%08x, base=0x%016lx\n",
6266 	       name, vmcs_read32(limit),
6267 	       vmcs_readl(limit + GUEST_GDTR_BASE - GUEST_GDTR_LIMIT));
6268 }
6269 
6270 static void vmx_dump_msrs(char *name, struct vmx_msrs *m)
6271 {
6272 	unsigned int i;
6273 	struct vmx_msr_entry *e;
6274 
6275 	pr_err("MSR %s:\n", name);
6276 	for (i = 0, e = m->val; i < m->nr; ++i, ++e)
6277 		pr_err("  %2d: msr=0x%08x value=0x%016llx\n", i, e->index, e->value);
6278 }
6279 
6280 void dump_vmcs(struct kvm_vcpu *vcpu)
6281 {
6282 	struct vcpu_vmx *vmx = to_vmx(vcpu);
6283 	u32 vmentry_ctl, vmexit_ctl;
6284 	u32 cpu_based_exec_ctrl, pin_based_exec_ctrl, secondary_exec_control;
6285 	u64 tertiary_exec_control;
6286 	unsigned long cr4;
6287 	int efer_slot;
6288 
6289 	if (!dump_invalid_vmcs) {
6290 		pr_warn_ratelimited("set kvm_intel.dump_invalid_vmcs=1 to dump internal KVM state.\n");
6291 		return;
6292 	}
6293 
6294 	vmentry_ctl = vmcs_read32(VM_ENTRY_CONTROLS);
6295 	vmexit_ctl = vmcs_read32(VM_EXIT_CONTROLS);
6296 	cpu_based_exec_ctrl = vmcs_read32(CPU_BASED_VM_EXEC_CONTROL);
6297 	pin_based_exec_ctrl = vmcs_read32(PIN_BASED_VM_EXEC_CONTROL);
6298 	cr4 = vmcs_readl(GUEST_CR4);
6299 
6300 	if (cpu_has_secondary_exec_ctrls())
6301 		secondary_exec_control = vmcs_read32(SECONDARY_VM_EXEC_CONTROL);
6302 	else
6303 		secondary_exec_control = 0;
6304 
6305 	if (cpu_has_tertiary_exec_ctrls())
6306 		tertiary_exec_control = vmcs_read64(TERTIARY_VM_EXEC_CONTROL);
6307 	else
6308 		tertiary_exec_control = 0;
6309 
6310 	pr_err("VMCS %p, last attempted VM-entry on CPU %d\n",
6311 	       vmx->loaded_vmcs->vmcs, vcpu->arch.last_vmentry_cpu);
6312 	pr_err("*** Guest State ***\n");
6313 	pr_err("CR0: actual=0x%016lx, shadow=0x%016lx, gh_mask=%016lx\n",
6314 	       vmcs_readl(GUEST_CR0), vmcs_readl(CR0_READ_SHADOW),
6315 	       vmcs_readl(CR0_GUEST_HOST_MASK));
6316 	pr_err("CR4: actual=0x%016lx, shadow=0x%016lx, gh_mask=%016lx\n",
6317 	       cr4, vmcs_readl(CR4_READ_SHADOW), vmcs_readl(CR4_GUEST_HOST_MASK));
6318 	pr_err("CR3 = 0x%016lx\n", vmcs_readl(GUEST_CR3));
6319 	if (cpu_has_vmx_ept()) {
6320 		pr_err("PDPTR0 = 0x%016llx  PDPTR1 = 0x%016llx\n",
6321 		       vmcs_read64(GUEST_PDPTR0), vmcs_read64(GUEST_PDPTR1));
6322 		pr_err("PDPTR2 = 0x%016llx  PDPTR3 = 0x%016llx\n",
6323 		       vmcs_read64(GUEST_PDPTR2), vmcs_read64(GUEST_PDPTR3));
6324 	}
6325 	pr_err("RSP = 0x%016lx  RIP = 0x%016lx\n",
6326 	       vmcs_readl(GUEST_RSP), vmcs_readl(GUEST_RIP));
6327 	pr_err("RFLAGS=0x%08lx         DR7 = 0x%016lx\n",
6328 	       vmcs_readl(GUEST_RFLAGS), vmcs_readl(GUEST_DR7));
6329 	pr_err("Sysenter RSP=%016lx CS:RIP=%04x:%016lx\n",
6330 	       vmcs_readl(GUEST_SYSENTER_ESP),
6331 	       vmcs_read32(GUEST_SYSENTER_CS), vmcs_readl(GUEST_SYSENTER_EIP));
6332 	vmx_dump_sel("CS:  ", GUEST_CS_SELECTOR);
6333 	vmx_dump_sel("DS:  ", GUEST_DS_SELECTOR);
6334 	vmx_dump_sel("SS:  ", GUEST_SS_SELECTOR);
6335 	vmx_dump_sel("ES:  ", GUEST_ES_SELECTOR);
6336 	vmx_dump_sel("FS:  ", GUEST_FS_SELECTOR);
6337 	vmx_dump_sel("GS:  ", GUEST_GS_SELECTOR);
6338 	vmx_dump_dtsel("GDTR:", GUEST_GDTR_LIMIT);
6339 	vmx_dump_sel("LDTR:", GUEST_LDTR_SELECTOR);
6340 	vmx_dump_dtsel("IDTR:", GUEST_IDTR_LIMIT);
6341 	vmx_dump_sel("TR:  ", GUEST_TR_SELECTOR);
6342 	efer_slot = vmx_find_loadstore_msr_slot(&vmx->msr_autoload.guest, MSR_EFER);
6343 	if (vmentry_ctl & VM_ENTRY_LOAD_IA32_EFER)
6344 		pr_err("EFER= 0x%016llx\n", vmcs_read64(GUEST_IA32_EFER));
6345 	else if (efer_slot >= 0)
6346 		pr_err("EFER= 0x%016llx (autoload)\n",
6347 		       vmx->msr_autoload.guest.val[efer_slot].value);
6348 	else if (vmentry_ctl & VM_ENTRY_IA32E_MODE)
6349 		pr_err("EFER= 0x%016llx (effective)\n",
6350 		       vcpu->arch.efer | (EFER_LMA | EFER_LME));
6351 	else
6352 		pr_err("EFER= 0x%016llx (effective)\n",
6353 		       vcpu->arch.efer & ~(EFER_LMA | EFER_LME));
6354 	if (vmentry_ctl & VM_ENTRY_LOAD_IA32_PAT)
6355 		pr_err("PAT = 0x%016llx\n", vmcs_read64(GUEST_IA32_PAT));
6356 	pr_err("DebugCtl = 0x%016llx  DebugExceptions = 0x%016lx\n",
6357 	       vmcs_read64(GUEST_IA32_DEBUGCTL),
6358 	       vmcs_readl(GUEST_PENDING_DBG_EXCEPTIONS));
6359 	if (cpu_has_load_perf_global_ctrl() &&
6360 	    vmentry_ctl & VM_ENTRY_LOAD_IA32_PERF_GLOBAL_CTRL)
6361 		pr_err("PerfGlobCtl = 0x%016llx\n",
6362 		       vmcs_read64(GUEST_IA32_PERF_GLOBAL_CTRL));
6363 	if (vmentry_ctl & VM_ENTRY_LOAD_BNDCFGS)
6364 		pr_err("BndCfgS = 0x%016llx\n", vmcs_read64(GUEST_BNDCFGS));
6365 	pr_err("Interruptibility = %08x  ActivityState = %08x\n",
6366 	       vmcs_read32(GUEST_INTERRUPTIBILITY_INFO),
6367 	       vmcs_read32(GUEST_ACTIVITY_STATE));
6368 	if (secondary_exec_control & SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY)
6369 		pr_err("InterruptStatus = %04x\n",
6370 		       vmcs_read16(GUEST_INTR_STATUS));
6371 	if (vmcs_read32(VM_ENTRY_MSR_LOAD_COUNT) > 0)
6372 		vmx_dump_msrs("guest autoload", &vmx->msr_autoload.guest);
6373 	if (vmcs_read32(VM_EXIT_MSR_STORE_COUNT) > 0)
6374 		vmx_dump_msrs("guest autostore", &vmx->msr_autostore.guest);
6375 
6376 	pr_err("*** Host State ***\n");
6377 	pr_err("RIP = 0x%016lx  RSP = 0x%016lx\n",
6378 	       vmcs_readl(HOST_RIP), vmcs_readl(HOST_RSP));
6379 	pr_err("CS=%04x SS=%04x DS=%04x ES=%04x FS=%04x GS=%04x TR=%04x\n",
6380 	       vmcs_read16(HOST_CS_SELECTOR), vmcs_read16(HOST_SS_SELECTOR),
6381 	       vmcs_read16(HOST_DS_SELECTOR), vmcs_read16(HOST_ES_SELECTOR),
6382 	       vmcs_read16(HOST_FS_SELECTOR), vmcs_read16(HOST_GS_SELECTOR),
6383 	       vmcs_read16(HOST_TR_SELECTOR));
6384 	pr_err("FSBase=%016lx GSBase=%016lx TRBase=%016lx\n",
6385 	       vmcs_readl(HOST_FS_BASE), vmcs_readl(HOST_GS_BASE),
6386 	       vmcs_readl(HOST_TR_BASE));
6387 	pr_err("GDTBase=%016lx IDTBase=%016lx\n",
6388 	       vmcs_readl(HOST_GDTR_BASE), vmcs_readl(HOST_IDTR_BASE));
6389 	pr_err("CR0=%016lx CR3=%016lx CR4=%016lx\n",
6390 	       vmcs_readl(HOST_CR0), vmcs_readl(HOST_CR3),
6391 	       vmcs_readl(HOST_CR4));
6392 	pr_err("Sysenter RSP=%016lx CS:RIP=%04x:%016lx\n",
6393 	       vmcs_readl(HOST_IA32_SYSENTER_ESP),
6394 	       vmcs_read32(HOST_IA32_SYSENTER_CS),
6395 	       vmcs_readl(HOST_IA32_SYSENTER_EIP));
6396 	if (vmexit_ctl & VM_EXIT_LOAD_IA32_EFER)
6397 		pr_err("EFER= 0x%016llx\n", vmcs_read64(HOST_IA32_EFER));
6398 	if (vmexit_ctl & VM_EXIT_LOAD_IA32_PAT)
6399 		pr_err("PAT = 0x%016llx\n", vmcs_read64(HOST_IA32_PAT));
6400 	if (cpu_has_load_perf_global_ctrl() &&
6401 	    vmexit_ctl & VM_EXIT_LOAD_IA32_PERF_GLOBAL_CTRL)
6402 		pr_err("PerfGlobCtl = 0x%016llx\n",
6403 		       vmcs_read64(HOST_IA32_PERF_GLOBAL_CTRL));
6404 	if (vmcs_read32(VM_EXIT_MSR_LOAD_COUNT) > 0)
6405 		vmx_dump_msrs("host autoload", &vmx->msr_autoload.host);
6406 
6407 	pr_err("*** Control State ***\n");
6408 	pr_err("CPUBased=0x%08x SecondaryExec=0x%08x TertiaryExec=0x%016llx\n",
6409 	       cpu_based_exec_ctrl, secondary_exec_control, tertiary_exec_control);
6410 	pr_err("PinBased=0x%08x EntryControls=%08x ExitControls=%08x\n",
6411 	       pin_based_exec_ctrl, vmentry_ctl, vmexit_ctl);
6412 	pr_err("ExceptionBitmap=%08x PFECmask=%08x PFECmatch=%08x\n",
6413 	       vmcs_read32(EXCEPTION_BITMAP),
6414 	       vmcs_read32(PAGE_FAULT_ERROR_CODE_MASK),
6415 	       vmcs_read32(PAGE_FAULT_ERROR_CODE_MATCH));
6416 	pr_err("VMEntry: intr_info=%08x errcode=%08x ilen=%08x\n",
6417 	       vmcs_read32(VM_ENTRY_INTR_INFO_FIELD),
6418 	       vmcs_read32(VM_ENTRY_EXCEPTION_ERROR_CODE),
6419 	       vmcs_read32(VM_ENTRY_INSTRUCTION_LEN));
6420 	pr_err("VMExit: intr_info=%08x errcode=%08x ilen=%08x\n",
6421 	       vmcs_read32(VM_EXIT_INTR_INFO),
6422 	       vmcs_read32(VM_EXIT_INTR_ERROR_CODE),
6423 	       vmcs_read32(VM_EXIT_INSTRUCTION_LEN));
6424 	pr_err("        reason=%08x qualification=%016lx\n",
6425 	       vmcs_read32(VM_EXIT_REASON), vmcs_readl(EXIT_QUALIFICATION));
6426 	pr_err("IDTVectoring: info=%08x errcode=%08x\n",
6427 	       vmcs_read32(IDT_VECTORING_INFO_FIELD),
6428 	       vmcs_read32(IDT_VECTORING_ERROR_CODE));
6429 	pr_err("TSC Offset = 0x%016llx\n", vmcs_read64(TSC_OFFSET));
6430 	if (secondary_exec_control & SECONDARY_EXEC_TSC_SCALING)
6431 		pr_err("TSC Multiplier = 0x%016llx\n",
6432 		       vmcs_read64(TSC_MULTIPLIER));
6433 	if (cpu_based_exec_ctrl & CPU_BASED_TPR_SHADOW) {
6434 		if (secondary_exec_control & SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY) {
6435 			u16 status = vmcs_read16(GUEST_INTR_STATUS);
6436 			pr_err("SVI|RVI = %02x|%02x ", status >> 8, status & 0xff);
6437 		}
6438 		pr_cont("TPR Threshold = 0x%02x\n", vmcs_read32(TPR_THRESHOLD));
6439 		if (secondary_exec_control & SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES)
6440 			pr_err("APIC-access addr = 0x%016llx ", vmcs_read64(APIC_ACCESS_ADDR));
6441 		pr_cont("virt-APIC addr = 0x%016llx\n", vmcs_read64(VIRTUAL_APIC_PAGE_ADDR));
6442 	}
6443 	if (pin_based_exec_ctrl & PIN_BASED_POSTED_INTR)
6444 		pr_err("PostedIntrVec = 0x%02x\n", vmcs_read16(POSTED_INTR_NV));
6445 	if ((secondary_exec_control & SECONDARY_EXEC_ENABLE_EPT))
6446 		pr_err("EPT pointer = 0x%016llx\n", vmcs_read64(EPT_POINTER));
6447 	if (secondary_exec_control & SECONDARY_EXEC_PAUSE_LOOP_EXITING)
6448 		pr_err("PLE Gap=%08x Window=%08x\n",
6449 		       vmcs_read32(PLE_GAP), vmcs_read32(PLE_WINDOW));
6450 	if (secondary_exec_control & SECONDARY_EXEC_ENABLE_VPID)
6451 		pr_err("Virtual processor ID = 0x%04x\n",
6452 		       vmcs_read16(VIRTUAL_PROCESSOR_ID));
6453 	if (secondary_exec_control & SECONDARY_EXEC_EPT_VIOLATION_VE) {
6454 		struct vmx_ve_information *ve_info = vmx->ve_info;
6455 		u64 ve_info_pa = vmcs_read64(VE_INFORMATION_ADDRESS);
6456 
6457 		/*
6458 		 * If KVM is dumping the VMCS, then something has gone wrong
6459 		 * already.  Derefencing an address from the VMCS, which could
6460 		 * very well be corrupted, is a terrible idea.  The virtual
6461 		 * address is known so use it.
6462 		 */
6463 		pr_err("VE info address = 0x%016llx%s\n", ve_info_pa,
6464 		       ve_info_pa == __pa(ve_info) ? "" : "(corrupted!)");
6465 		pr_err("ve_info: 0x%08x 0x%08x 0x%016llx 0x%016llx 0x%016llx 0x%04x\n",
6466 		       ve_info->exit_reason, ve_info->delivery,
6467 		       ve_info->exit_qualification,
6468 		       ve_info->guest_linear_address,
6469 		       ve_info->guest_physical_address, ve_info->eptp_index);
6470 	}
6471 }
6472 
6473 /*
6474  * The guest has exited.  See if we can fix it or if we need userspace
6475  * assistance.
6476  */
6477 static int __vmx_handle_exit(struct kvm_vcpu *vcpu, fastpath_t exit_fastpath)
6478 {
6479 	struct vcpu_vmx *vmx = to_vmx(vcpu);
6480 	union vmx_exit_reason exit_reason = vmx->exit_reason;
6481 	u32 vectoring_info = vmx->idt_vectoring_info;
6482 	u16 exit_handler_index;
6483 
6484 	/*
6485 	 * Flush logged GPAs PML buffer, this will make dirty_bitmap more
6486 	 * updated. Another good is, in kvm_vm_ioctl_get_dirty_log, before
6487 	 * querying dirty_bitmap, we only need to kick all vcpus out of guest
6488 	 * mode as if vcpus is in root mode, the PML buffer must has been
6489 	 * flushed already.  Note, PML is never enabled in hardware while
6490 	 * running L2.
6491 	 */
6492 	if (enable_pml && !is_guest_mode(vcpu))
6493 		vmx_flush_pml_buffer(vcpu);
6494 
6495 	/*
6496 	 * KVM should never reach this point with a pending nested VM-Enter.
6497 	 * More specifically, short-circuiting VM-Entry to emulate L2 due to
6498 	 * invalid guest state should never happen as that means KVM knowingly
6499 	 * allowed a nested VM-Enter with an invalid vmcs12.  More below.
6500 	 */
6501 	if (KVM_BUG_ON(vmx->nested.nested_run_pending, vcpu->kvm))
6502 		return -EIO;
6503 
6504 	if (is_guest_mode(vcpu)) {
6505 		/*
6506 		 * PML is never enabled when running L2, bail immediately if a
6507 		 * PML full exit occurs as something is horribly wrong.
6508 		 */
6509 		if (exit_reason.basic == EXIT_REASON_PML_FULL)
6510 			goto unexpected_vmexit;
6511 
6512 		/*
6513 		 * The host physical addresses of some pages of guest memory
6514 		 * are loaded into the vmcs02 (e.g. vmcs12's Virtual APIC
6515 		 * Page). The CPU may write to these pages via their host
6516 		 * physical address while L2 is running, bypassing any
6517 		 * address-translation-based dirty tracking (e.g. EPT write
6518 		 * protection).
6519 		 *
6520 		 * Mark them dirty on every exit from L2 to prevent them from
6521 		 * getting out of sync with dirty tracking.
6522 		 */
6523 		nested_mark_vmcs12_pages_dirty(vcpu);
6524 
6525 		/*
6526 		 * Synthesize a triple fault if L2 state is invalid.  In normal
6527 		 * operation, nested VM-Enter rejects any attempt to enter L2
6528 		 * with invalid state.  However, those checks are skipped if
6529 		 * state is being stuffed via RSM or KVM_SET_NESTED_STATE.  If
6530 		 * L2 state is invalid, it means either L1 modified SMRAM state
6531 		 * or userspace provided bad state.  Synthesize TRIPLE_FAULT as
6532 		 * doing so is architecturally allowed in the RSM case, and is
6533 		 * the least awful solution for the userspace case without
6534 		 * risking false positives.
6535 		 */
6536 		if (vmx->emulation_required) {
6537 			nested_vmx_vmexit(vcpu, EXIT_REASON_TRIPLE_FAULT, 0, 0);
6538 			return 1;
6539 		}
6540 
6541 		if (nested_vmx_reflect_vmexit(vcpu))
6542 			return 1;
6543 	}
6544 
6545 	/* If guest state is invalid, start emulating.  L2 is handled above. */
6546 	if (vmx->emulation_required)
6547 		return handle_invalid_guest_state(vcpu);
6548 
6549 	if (exit_reason.failed_vmentry) {
6550 		dump_vmcs(vcpu);
6551 		vcpu->run->exit_reason = KVM_EXIT_FAIL_ENTRY;
6552 		vcpu->run->fail_entry.hardware_entry_failure_reason
6553 			= exit_reason.full;
6554 		vcpu->run->fail_entry.cpu = vcpu->arch.last_vmentry_cpu;
6555 		return 0;
6556 	}
6557 
6558 	if (unlikely(vmx->fail)) {
6559 		dump_vmcs(vcpu);
6560 		vcpu->run->exit_reason = KVM_EXIT_FAIL_ENTRY;
6561 		vcpu->run->fail_entry.hardware_entry_failure_reason
6562 			= vmcs_read32(VM_INSTRUCTION_ERROR);
6563 		vcpu->run->fail_entry.cpu = vcpu->arch.last_vmentry_cpu;
6564 		return 0;
6565 	}
6566 
6567 	if ((vectoring_info & VECTORING_INFO_VALID_MASK) &&
6568 	    (exit_reason.basic != EXIT_REASON_EXCEPTION_NMI &&
6569 	     exit_reason.basic != EXIT_REASON_EPT_VIOLATION &&
6570 	     exit_reason.basic != EXIT_REASON_PML_FULL &&
6571 	     exit_reason.basic != EXIT_REASON_APIC_ACCESS &&
6572 	     exit_reason.basic != EXIT_REASON_TASK_SWITCH &&
6573 	     exit_reason.basic != EXIT_REASON_NOTIFY &&
6574 	     exit_reason.basic != EXIT_REASON_EPT_MISCONFIG)) {
6575 		kvm_prepare_event_vectoring_exit(vcpu, INVALID_GPA);
6576 		return 0;
6577 	}
6578 
6579 	if (unlikely(!enable_vnmi &&
6580 		     vmx->loaded_vmcs->soft_vnmi_blocked)) {
6581 		if (!vmx_interrupt_blocked(vcpu)) {
6582 			vmx->loaded_vmcs->soft_vnmi_blocked = 0;
6583 		} else if (vmx->loaded_vmcs->vnmi_blocked_time > 1000000000LL &&
6584 			   vcpu->arch.nmi_pending) {
6585 			/*
6586 			 * This CPU don't support us in finding the end of an
6587 			 * NMI-blocked window if the guest runs with IRQs
6588 			 * disabled. So we pull the trigger after 1 s of
6589 			 * futile waiting, but inform the user about this.
6590 			 */
6591 			printk(KERN_WARNING "%s: Breaking out of NMI-blocked "
6592 			       "state on VCPU %d after 1 s timeout\n",
6593 			       __func__, vcpu->vcpu_id);
6594 			vmx->loaded_vmcs->soft_vnmi_blocked = 0;
6595 		}
6596 	}
6597 
6598 	if (exit_fastpath != EXIT_FASTPATH_NONE)
6599 		return 1;
6600 
6601 	if (exit_reason.basic >= kvm_vmx_max_exit_handlers)
6602 		goto unexpected_vmexit;
6603 #ifdef CONFIG_MITIGATION_RETPOLINE
6604 	if (exit_reason.basic == EXIT_REASON_MSR_WRITE)
6605 		return kvm_emulate_wrmsr(vcpu);
6606 	else if (exit_reason.basic == EXIT_REASON_PREEMPTION_TIMER)
6607 		return handle_preemption_timer(vcpu);
6608 	else if (exit_reason.basic == EXIT_REASON_INTERRUPT_WINDOW)
6609 		return handle_interrupt_window(vcpu);
6610 	else if (exit_reason.basic == EXIT_REASON_EXTERNAL_INTERRUPT)
6611 		return handle_external_interrupt(vcpu);
6612 	else if (exit_reason.basic == EXIT_REASON_HLT)
6613 		return kvm_emulate_halt(vcpu);
6614 	else if (exit_reason.basic == EXIT_REASON_EPT_MISCONFIG)
6615 		return handle_ept_misconfig(vcpu);
6616 #endif
6617 
6618 	exit_handler_index = array_index_nospec((u16)exit_reason.basic,
6619 						kvm_vmx_max_exit_handlers);
6620 	if (!kvm_vmx_exit_handlers[exit_handler_index])
6621 		goto unexpected_vmexit;
6622 
6623 	return kvm_vmx_exit_handlers[exit_handler_index](vcpu);
6624 
6625 unexpected_vmexit:
6626 	vcpu_unimpl(vcpu, "vmx: unexpected exit reason 0x%x\n",
6627 		    exit_reason.full);
6628 	dump_vmcs(vcpu);
6629 	vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
6630 	vcpu->run->internal.suberror =
6631 			KVM_INTERNAL_ERROR_UNEXPECTED_EXIT_REASON;
6632 	vcpu->run->internal.ndata = 2;
6633 	vcpu->run->internal.data[0] = exit_reason.full;
6634 	vcpu->run->internal.data[1] = vcpu->arch.last_vmentry_cpu;
6635 	return 0;
6636 }
6637 
6638 int vmx_handle_exit(struct kvm_vcpu *vcpu, fastpath_t exit_fastpath)
6639 {
6640 	int ret = __vmx_handle_exit(vcpu, exit_fastpath);
6641 
6642 	/*
6643 	 * Exit to user space when bus lock detected to inform that there is
6644 	 * a bus lock in guest.
6645 	 */
6646 	if (to_vmx(vcpu)->exit_reason.bus_lock_detected) {
6647 		if (ret > 0)
6648 			vcpu->run->exit_reason = KVM_EXIT_X86_BUS_LOCK;
6649 
6650 		vcpu->run->flags |= KVM_RUN_X86_BUS_LOCK;
6651 		return 0;
6652 	}
6653 	return ret;
6654 }
6655 
6656 /*
6657  * Software based L1D cache flush which is used when microcode providing
6658  * the cache control MSR is not loaded.
6659  *
6660  * The L1D cache is 32 KiB on Nehalem and later microarchitectures, but to
6661  * flush it is required to read in 64 KiB because the replacement algorithm
6662  * is not exactly LRU. This could be sized at runtime via topology
6663  * information but as all relevant affected CPUs have 32KiB L1D cache size
6664  * there is no point in doing so.
6665  */
6666 static noinstr void vmx_l1d_flush(struct kvm_vcpu *vcpu)
6667 {
6668 	int size = PAGE_SIZE << L1D_CACHE_ORDER;
6669 
6670 	/*
6671 	 * This code is only executed when the flush mode is 'cond' or
6672 	 * 'always'
6673 	 */
6674 	if (static_branch_likely(&vmx_l1d_flush_cond)) {
6675 		bool flush_l1d;
6676 
6677 		/*
6678 		 * Clear the per-vcpu flush bit, it gets set again if the vCPU
6679 		 * is reloaded, i.e. if the vCPU is scheduled out or if KVM
6680 		 * exits to userspace, or if KVM reaches one of the unsafe
6681 		 * VMEXIT handlers, e.g. if KVM calls into the emulator.
6682 		 */
6683 		flush_l1d = vcpu->arch.l1tf_flush_l1d;
6684 		vcpu->arch.l1tf_flush_l1d = false;
6685 
6686 		/*
6687 		 * Clear the per-cpu flush bit, it gets set again from
6688 		 * the interrupt handlers.
6689 		 */
6690 		flush_l1d |= kvm_get_cpu_l1tf_flush_l1d();
6691 		kvm_clear_cpu_l1tf_flush_l1d();
6692 
6693 		if (!flush_l1d)
6694 			return;
6695 	}
6696 
6697 	vcpu->stat.l1d_flush++;
6698 
6699 	if (static_cpu_has(X86_FEATURE_FLUSH_L1D)) {
6700 		native_wrmsrl(MSR_IA32_FLUSH_CMD, L1D_FLUSH);
6701 		return;
6702 	}
6703 
6704 	asm volatile(
6705 		/* First ensure the pages are in the TLB */
6706 		"xorl	%%eax, %%eax\n"
6707 		".Lpopulate_tlb:\n\t"
6708 		"movzbl	(%[flush_pages], %%" _ASM_AX "), %%ecx\n\t"
6709 		"addl	$4096, %%eax\n\t"
6710 		"cmpl	%%eax, %[size]\n\t"
6711 		"jne	.Lpopulate_tlb\n\t"
6712 		"xorl	%%eax, %%eax\n\t"
6713 		"cpuid\n\t"
6714 		/* Now fill the cache */
6715 		"xorl	%%eax, %%eax\n"
6716 		".Lfill_cache:\n"
6717 		"movzbl	(%[flush_pages], %%" _ASM_AX "), %%ecx\n\t"
6718 		"addl	$64, %%eax\n\t"
6719 		"cmpl	%%eax, %[size]\n\t"
6720 		"jne	.Lfill_cache\n\t"
6721 		"lfence\n"
6722 		:: [flush_pages] "r" (vmx_l1d_flush_pages),
6723 		    [size] "r" (size)
6724 		: "eax", "ebx", "ecx", "edx");
6725 }
6726 
6727 void vmx_update_cr8_intercept(struct kvm_vcpu *vcpu, int tpr, int irr)
6728 {
6729 	struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
6730 	int tpr_threshold;
6731 
6732 	if (is_guest_mode(vcpu) &&
6733 		nested_cpu_has(vmcs12, CPU_BASED_TPR_SHADOW))
6734 		return;
6735 
6736 	tpr_threshold = (irr == -1 || tpr < irr) ? 0 : irr;
6737 	if (is_guest_mode(vcpu))
6738 		to_vmx(vcpu)->nested.l1_tpr_threshold = tpr_threshold;
6739 	else
6740 		vmcs_write32(TPR_THRESHOLD, tpr_threshold);
6741 }
6742 
6743 void vmx_set_virtual_apic_mode(struct kvm_vcpu *vcpu)
6744 {
6745 	struct vcpu_vmx *vmx = to_vmx(vcpu);
6746 	u32 sec_exec_control;
6747 
6748 	if (!lapic_in_kernel(vcpu))
6749 		return;
6750 
6751 	if (!flexpriority_enabled &&
6752 	    !cpu_has_vmx_virtualize_x2apic_mode())
6753 		return;
6754 
6755 	/* Postpone execution until vmcs01 is the current VMCS. */
6756 	if (is_guest_mode(vcpu)) {
6757 		vmx->nested.change_vmcs01_virtual_apic_mode = true;
6758 		return;
6759 	}
6760 
6761 	sec_exec_control = secondary_exec_controls_get(vmx);
6762 	sec_exec_control &= ~(SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES |
6763 			      SECONDARY_EXEC_VIRTUALIZE_X2APIC_MODE);
6764 
6765 	switch (kvm_get_apic_mode(vcpu)) {
6766 	case LAPIC_MODE_INVALID:
6767 		WARN_ONCE(true, "Invalid local APIC state");
6768 		break;
6769 	case LAPIC_MODE_DISABLED:
6770 		break;
6771 	case LAPIC_MODE_XAPIC:
6772 		if (flexpriority_enabled) {
6773 			sec_exec_control |=
6774 				SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES;
6775 			kvm_make_request(KVM_REQ_APIC_PAGE_RELOAD, vcpu);
6776 
6777 			/*
6778 			 * Flush the TLB, reloading the APIC access page will
6779 			 * only do so if its physical address has changed, but
6780 			 * the guest may have inserted a non-APIC mapping into
6781 			 * the TLB while the APIC access page was disabled.
6782 			 */
6783 			kvm_make_request(KVM_REQ_TLB_FLUSH_CURRENT, vcpu);
6784 		}
6785 		break;
6786 	case LAPIC_MODE_X2APIC:
6787 		if (cpu_has_vmx_virtualize_x2apic_mode())
6788 			sec_exec_control |=
6789 				SECONDARY_EXEC_VIRTUALIZE_X2APIC_MODE;
6790 		break;
6791 	}
6792 	secondary_exec_controls_set(vmx, sec_exec_control);
6793 
6794 	vmx_update_msr_bitmap_x2apic(vcpu);
6795 }
6796 
6797 void vmx_set_apic_access_page_addr(struct kvm_vcpu *vcpu)
6798 {
6799 	const gfn_t gfn = APIC_DEFAULT_PHYS_BASE >> PAGE_SHIFT;
6800 	struct kvm *kvm = vcpu->kvm;
6801 	struct kvm_memslots *slots = kvm_memslots(kvm);
6802 	struct kvm_memory_slot *slot;
6803 	struct page *refcounted_page;
6804 	unsigned long mmu_seq;
6805 	kvm_pfn_t pfn;
6806 	bool writable;
6807 
6808 	/* Defer reload until vmcs01 is the current VMCS. */
6809 	if (is_guest_mode(vcpu)) {
6810 		to_vmx(vcpu)->nested.reload_vmcs01_apic_access_page = true;
6811 		return;
6812 	}
6813 
6814 	if (!(secondary_exec_controls_get(to_vmx(vcpu)) &
6815 	    SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES))
6816 		return;
6817 
6818 	/*
6819 	 * Explicitly grab the memslot using KVM's internal slot ID to ensure
6820 	 * KVM doesn't unintentionally grab a userspace memslot.  It _should_
6821 	 * be impossible for userspace to create a memslot for the APIC when
6822 	 * APICv is enabled, but paranoia won't hurt in this case.
6823 	 */
6824 	slot = id_to_memslot(slots, APIC_ACCESS_PAGE_PRIVATE_MEMSLOT);
6825 	if (!slot || slot->flags & KVM_MEMSLOT_INVALID)
6826 		return;
6827 
6828 	/*
6829 	 * Ensure that the mmu_notifier sequence count is read before KVM
6830 	 * retrieves the pfn from the primary MMU.  Note, the memslot is
6831 	 * protected by SRCU, not the mmu_notifier.  Pairs with the smp_wmb()
6832 	 * in kvm_mmu_invalidate_end().
6833 	 */
6834 	mmu_seq = kvm->mmu_invalidate_seq;
6835 	smp_rmb();
6836 
6837 	/*
6838 	 * No need to retry if the memslot does not exist or is invalid.  KVM
6839 	 * controls the APIC-access page memslot, and only deletes the memslot
6840 	 * if APICv is permanently inhibited, i.e. the memslot won't reappear.
6841 	 */
6842 	pfn = __kvm_faultin_pfn(slot, gfn, FOLL_WRITE, &writable, &refcounted_page);
6843 	if (is_error_noslot_pfn(pfn))
6844 		return;
6845 
6846 	read_lock(&vcpu->kvm->mmu_lock);
6847 	if (mmu_invalidate_retry_gfn(kvm, mmu_seq, gfn))
6848 		kvm_make_request(KVM_REQ_APIC_PAGE_RELOAD, vcpu);
6849 	else
6850 		vmcs_write64(APIC_ACCESS_ADDR, pfn_to_hpa(pfn));
6851 
6852 	/*
6853 	 * Do not pin the APIC access page in memory so that it can be freely
6854 	 * migrated, the MMU notifier will call us again if it is migrated or
6855 	 * swapped out.  KVM backs the memslot with anonymous memory, the pfn
6856 	 * should always point at a refcounted page (if the pfn is valid).
6857 	 */
6858 	if (!WARN_ON_ONCE(!refcounted_page))
6859 		kvm_release_page_clean(refcounted_page);
6860 
6861 	/*
6862 	 * No need for a manual TLB flush at this point, KVM has already done a
6863 	 * flush if there were SPTEs pointing at the previous page.
6864 	 */
6865 	read_unlock(&vcpu->kvm->mmu_lock);
6866 }
6867 
6868 void vmx_hwapic_isr_update(struct kvm_vcpu *vcpu, int max_isr)
6869 {
6870 	u16 status;
6871 	u8 old;
6872 
6873 	/*
6874 	 * If L2 is active, defer the SVI update until vmcs01 is loaded, as SVI
6875 	 * is only relevant for if and only if Virtual Interrupt Delivery is
6876 	 * enabled in vmcs12, and if VID is enabled then L2 EOIs affect L2's
6877 	 * vAPIC, not L1's vAPIC.  KVM must update vmcs01 on the next nested
6878 	 * VM-Exit, otherwise L1 with run with a stale SVI.
6879 	 */
6880 	if (is_guest_mode(vcpu)) {
6881 		/*
6882 		 * KVM is supposed to forward intercepted L2 EOIs to L1 if VID
6883 		 * is enabled in vmcs12; as above, the EOIs affect L2's vAPIC.
6884 		 * Note, userspace can stuff state while L2 is active; assert
6885 		 * that VID is disabled if and only if the vCPU is in KVM_RUN
6886 		 * to avoid false positives if userspace is setting APIC state.
6887 		 */
6888 		WARN_ON_ONCE(vcpu->wants_to_run &&
6889 			     nested_cpu_has_vid(get_vmcs12(vcpu)));
6890 		to_vmx(vcpu)->nested.update_vmcs01_hwapic_isr = true;
6891 		return;
6892 	}
6893 
6894 	if (max_isr == -1)
6895 		max_isr = 0;
6896 
6897 	status = vmcs_read16(GUEST_INTR_STATUS);
6898 	old = status >> 8;
6899 	if (max_isr != old) {
6900 		status &= 0xff;
6901 		status |= max_isr << 8;
6902 		vmcs_write16(GUEST_INTR_STATUS, status);
6903 	}
6904 }
6905 
6906 static void vmx_set_rvi(int vector)
6907 {
6908 	u16 status;
6909 	u8 old;
6910 
6911 	if (vector == -1)
6912 		vector = 0;
6913 
6914 	status = vmcs_read16(GUEST_INTR_STATUS);
6915 	old = (u8)status & 0xff;
6916 	if ((u8)vector != old) {
6917 		status &= ~0xff;
6918 		status |= (u8)vector;
6919 		vmcs_write16(GUEST_INTR_STATUS, status);
6920 	}
6921 }
6922 
6923 int vmx_sync_pir_to_irr(struct kvm_vcpu *vcpu)
6924 {
6925 	struct vcpu_vmx *vmx = to_vmx(vcpu);
6926 	int max_irr;
6927 	bool got_posted_interrupt;
6928 
6929 	if (KVM_BUG_ON(!enable_apicv, vcpu->kvm))
6930 		return -EIO;
6931 
6932 	if (pi_test_on(&vmx->pi_desc)) {
6933 		pi_clear_on(&vmx->pi_desc);
6934 		/*
6935 		 * IOMMU can write to PID.ON, so the barrier matters even on UP.
6936 		 * But on x86 this is just a compiler barrier anyway.
6937 		 */
6938 		smp_mb__after_atomic();
6939 		got_posted_interrupt =
6940 			kvm_apic_update_irr(vcpu, vmx->pi_desc.pir, &max_irr);
6941 	} else {
6942 		max_irr = kvm_lapic_find_highest_irr(vcpu);
6943 		got_posted_interrupt = false;
6944 	}
6945 
6946 	/*
6947 	 * Newly recognized interrupts are injected via either virtual interrupt
6948 	 * delivery (RVI) or KVM_REQ_EVENT.  Virtual interrupt delivery is
6949 	 * disabled in two cases:
6950 	 *
6951 	 * 1) If L2 is running and the vCPU has a new pending interrupt.  If L1
6952 	 * wants to exit on interrupts, KVM_REQ_EVENT is needed to synthesize a
6953 	 * VM-Exit to L1.  If L1 doesn't want to exit, the interrupt is injected
6954 	 * into L2, but KVM doesn't use virtual interrupt delivery to inject
6955 	 * interrupts into L2, and so KVM_REQ_EVENT is again needed.
6956 	 *
6957 	 * 2) If APICv is disabled for this vCPU, assigned devices may still
6958 	 * attempt to post interrupts.  The posted interrupt vector will cause
6959 	 * a VM-Exit and the subsequent entry will call sync_pir_to_irr.
6960 	 */
6961 	if (!is_guest_mode(vcpu) && kvm_vcpu_apicv_active(vcpu))
6962 		vmx_set_rvi(max_irr);
6963 	else if (got_posted_interrupt)
6964 		kvm_make_request(KVM_REQ_EVENT, vcpu);
6965 
6966 	return max_irr;
6967 }
6968 
6969 void vmx_load_eoi_exitmap(struct kvm_vcpu *vcpu, u64 *eoi_exit_bitmap)
6970 {
6971 	if (!kvm_vcpu_apicv_active(vcpu))
6972 		return;
6973 
6974 	vmcs_write64(EOI_EXIT_BITMAP0, eoi_exit_bitmap[0]);
6975 	vmcs_write64(EOI_EXIT_BITMAP1, eoi_exit_bitmap[1]);
6976 	vmcs_write64(EOI_EXIT_BITMAP2, eoi_exit_bitmap[2]);
6977 	vmcs_write64(EOI_EXIT_BITMAP3, eoi_exit_bitmap[3]);
6978 }
6979 
6980 void vmx_apicv_pre_state_restore(struct kvm_vcpu *vcpu)
6981 {
6982 	struct vcpu_vmx *vmx = to_vmx(vcpu);
6983 
6984 	pi_clear_on(&vmx->pi_desc);
6985 	memset(vmx->pi_desc.pir, 0, sizeof(vmx->pi_desc.pir));
6986 }
6987 
6988 void vmx_do_interrupt_irqoff(unsigned long entry);
6989 void vmx_do_nmi_irqoff(void);
6990 
6991 static void handle_nm_fault_irqoff(struct kvm_vcpu *vcpu)
6992 {
6993 	/*
6994 	 * Save xfd_err to guest_fpu before interrupt is enabled, so the
6995 	 * MSR value is not clobbered by the host activity before the guest
6996 	 * has chance to consume it.
6997 	 *
6998 	 * Do not blindly read xfd_err here, since this exception might
6999 	 * be caused by L1 interception on a platform which doesn't
7000 	 * support xfd at all.
7001 	 *
7002 	 * Do it conditionally upon guest_fpu::xfd. xfd_err matters
7003 	 * only when xfd contains a non-zero value.
7004 	 *
7005 	 * Queuing exception is done in vmx_handle_exit. See comment there.
7006 	 */
7007 	if (vcpu->arch.guest_fpu.fpstate->xfd)
7008 		rdmsrl(MSR_IA32_XFD_ERR, vcpu->arch.guest_fpu.xfd_err);
7009 }
7010 
7011 static void handle_exception_irqoff(struct kvm_vcpu *vcpu, u32 intr_info)
7012 {
7013 	/* if exit due to PF check for async PF */
7014 	if (is_page_fault(intr_info))
7015 		vcpu->arch.apf.host_apf_flags = kvm_read_and_reset_apf_flags();
7016 	/* if exit due to NM, handle before interrupts are enabled */
7017 	else if (is_nm_fault(intr_info))
7018 		handle_nm_fault_irqoff(vcpu);
7019 	/* Handle machine checks before interrupts are enabled */
7020 	else if (is_machine_check(intr_info))
7021 		kvm_machine_check();
7022 }
7023 
7024 static void handle_external_interrupt_irqoff(struct kvm_vcpu *vcpu,
7025 					     u32 intr_info)
7026 {
7027 	unsigned int vector = intr_info & INTR_INFO_VECTOR_MASK;
7028 
7029 	if (KVM_BUG(!is_external_intr(intr_info), vcpu->kvm,
7030 	    "unexpected VM-Exit interrupt info: 0x%x", intr_info))
7031 		return;
7032 
7033 	kvm_before_interrupt(vcpu, KVM_HANDLING_IRQ);
7034 	if (cpu_feature_enabled(X86_FEATURE_FRED))
7035 		fred_entry_from_kvm(EVENT_TYPE_EXTINT, vector);
7036 	else
7037 		vmx_do_interrupt_irqoff(gate_offset((gate_desc *)host_idt_base + vector));
7038 	kvm_after_interrupt(vcpu);
7039 
7040 	vcpu->arch.at_instruction_boundary = true;
7041 }
7042 
7043 void vmx_handle_exit_irqoff(struct kvm_vcpu *vcpu)
7044 {
7045 	struct vcpu_vmx *vmx = to_vmx(vcpu);
7046 
7047 	if (vmx->emulation_required)
7048 		return;
7049 
7050 	if (vmx->exit_reason.basic == EXIT_REASON_EXTERNAL_INTERRUPT)
7051 		handle_external_interrupt_irqoff(vcpu, vmx_get_intr_info(vcpu));
7052 	else if (vmx->exit_reason.basic == EXIT_REASON_EXCEPTION_NMI)
7053 		handle_exception_irqoff(vcpu, vmx_get_intr_info(vcpu));
7054 }
7055 
7056 /*
7057  * The kvm parameter can be NULL (module initialization, or invocation before
7058  * VM creation). Be sure to check the kvm parameter before using it.
7059  */
7060 bool vmx_has_emulated_msr(struct kvm *kvm, u32 index)
7061 {
7062 	switch (index) {
7063 	case MSR_IA32_SMBASE:
7064 		if (!IS_ENABLED(CONFIG_KVM_SMM))
7065 			return false;
7066 		/*
7067 		 * We cannot do SMM unless we can run the guest in big
7068 		 * real mode.
7069 		 */
7070 		return enable_unrestricted_guest || emulate_invalid_guest_state;
7071 	case KVM_FIRST_EMULATED_VMX_MSR ... KVM_LAST_EMULATED_VMX_MSR:
7072 		return nested;
7073 	case MSR_AMD64_VIRT_SPEC_CTRL:
7074 	case MSR_AMD64_TSC_RATIO:
7075 		/* This is AMD only.  */
7076 		return false;
7077 	default:
7078 		return true;
7079 	}
7080 }
7081 
7082 static void vmx_recover_nmi_blocking(struct vcpu_vmx *vmx)
7083 {
7084 	u32 exit_intr_info;
7085 	bool unblock_nmi;
7086 	u8 vector;
7087 	bool idtv_info_valid;
7088 
7089 	idtv_info_valid = vmx->idt_vectoring_info & VECTORING_INFO_VALID_MASK;
7090 
7091 	if (enable_vnmi) {
7092 		if (vmx->loaded_vmcs->nmi_known_unmasked)
7093 			return;
7094 
7095 		exit_intr_info = vmx_get_intr_info(&vmx->vcpu);
7096 		unblock_nmi = (exit_intr_info & INTR_INFO_UNBLOCK_NMI) != 0;
7097 		vector = exit_intr_info & INTR_INFO_VECTOR_MASK;
7098 		/*
7099 		 * SDM 3: 27.7.1.2 (September 2008)
7100 		 * Re-set bit "block by NMI" before VM entry if vmexit caused by
7101 		 * a guest IRET fault.
7102 		 * SDM 3: 23.2.2 (September 2008)
7103 		 * Bit 12 is undefined in any of the following cases:
7104 		 *  If the VM exit sets the valid bit in the IDT-vectoring
7105 		 *   information field.
7106 		 *  If the VM exit is due to a double fault.
7107 		 */
7108 		if ((exit_intr_info & INTR_INFO_VALID_MASK) && unblock_nmi &&
7109 		    vector != DF_VECTOR && !idtv_info_valid)
7110 			vmcs_set_bits(GUEST_INTERRUPTIBILITY_INFO,
7111 				      GUEST_INTR_STATE_NMI);
7112 		else
7113 			vmx->loaded_vmcs->nmi_known_unmasked =
7114 				!(vmcs_read32(GUEST_INTERRUPTIBILITY_INFO)
7115 				  & GUEST_INTR_STATE_NMI);
7116 	} else if (unlikely(vmx->loaded_vmcs->soft_vnmi_blocked))
7117 		vmx->loaded_vmcs->vnmi_blocked_time +=
7118 			ktime_to_ns(ktime_sub(ktime_get(),
7119 					      vmx->loaded_vmcs->entry_time));
7120 }
7121 
7122 static void __vmx_complete_interrupts(struct kvm_vcpu *vcpu,
7123 				      u32 idt_vectoring_info,
7124 				      int instr_len_field,
7125 				      int error_code_field)
7126 {
7127 	u8 vector;
7128 	int type;
7129 	bool idtv_info_valid;
7130 
7131 	idtv_info_valid = idt_vectoring_info & VECTORING_INFO_VALID_MASK;
7132 
7133 	vcpu->arch.nmi_injected = false;
7134 	kvm_clear_exception_queue(vcpu);
7135 	kvm_clear_interrupt_queue(vcpu);
7136 
7137 	if (!idtv_info_valid)
7138 		return;
7139 
7140 	kvm_make_request(KVM_REQ_EVENT, vcpu);
7141 
7142 	vector = idt_vectoring_info & VECTORING_INFO_VECTOR_MASK;
7143 	type = idt_vectoring_info & VECTORING_INFO_TYPE_MASK;
7144 
7145 	switch (type) {
7146 	case INTR_TYPE_NMI_INTR:
7147 		vcpu->arch.nmi_injected = true;
7148 		/*
7149 		 * SDM 3: 27.7.1.2 (September 2008)
7150 		 * Clear bit "block by NMI" before VM entry if a NMI
7151 		 * delivery faulted.
7152 		 */
7153 		vmx_set_nmi_mask(vcpu, false);
7154 		break;
7155 	case INTR_TYPE_SOFT_EXCEPTION:
7156 		vcpu->arch.event_exit_inst_len = vmcs_read32(instr_len_field);
7157 		fallthrough;
7158 	case INTR_TYPE_HARD_EXCEPTION:
7159 		if (idt_vectoring_info & VECTORING_INFO_DELIVER_CODE_MASK) {
7160 			u32 err = vmcs_read32(error_code_field);
7161 			kvm_requeue_exception_e(vcpu, vector, err);
7162 		} else
7163 			kvm_requeue_exception(vcpu, vector);
7164 		break;
7165 	case INTR_TYPE_SOFT_INTR:
7166 		vcpu->arch.event_exit_inst_len = vmcs_read32(instr_len_field);
7167 		fallthrough;
7168 	case INTR_TYPE_EXT_INTR:
7169 		kvm_queue_interrupt(vcpu, vector, type == INTR_TYPE_SOFT_INTR);
7170 		break;
7171 	default:
7172 		break;
7173 	}
7174 }
7175 
7176 static void vmx_complete_interrupts(struct vcpu_vmx *vmx)
7177 {
7178 	__vmx_complete_interrupts(&vmx->vcpu, vmx->idt_vectoring_info,
7179 				  VM_EXIT_INSTRUCTION_LEN,
7180 				  IDT_VECTORING_ERROR_CODE);
7181 }
7182 
7183 void vmx_cancel_injection(struct kvm_vcpu *vcpu)
7184 {
7185 	__vmx_complete_interrupts(vcpu,
7186 				  vmcs_read32(VM_ENTRY_INTR_INFO_FIELD),
7187 				  VM_ENTRY_INSTRUCTION_LEN,
7188 				  VM_ENTRY_EXCEPTION_ERROR_CODE);
7189 
7190 	vmcs_write32(VM_ENTRY_INTR_INFO_FIELD, 0);
7191 }
7192 
7193 static void atomic_switch_perf_msrs(struct vcpu_vmx *vmx)
7194 {
7195 	int i, nr_msrs;
7196 	struct perf_guest_switch_msr *msrs;
7197 	struct kvm_pmu *pmu = vcpu_to_pmu(&vmx->vcpu);
7198 
7199 	pmu->host_cross_mapped_mask = 0;
7200 	if (pmu->pebs_enable & pmu->global_ctrl)
7201 		intel_pmu_cross_mapped_check(pmu);
7202 
7203 	/* Note, nr_msrs may be garbage if perf_guest_get_msrs() returns NULL. */
7204 	msrs = perf_guest_get_msrs(&nr_msrs, (void *)pmu);
7205 	if (!msrs)
7206 		return;
7207 
7208 	for (i = 0; i < nr_msrs; i++)
7209 		if (msrs[i].host == msrs[i].guest)
7210 			clear_atomic_switch_msr(vmx, msrs[i].msr);
7211 		else
7212 			add_atomic_switch_msr(vmx, msrs[i].msr, msrs[i].guest,
7213 					msrs[i].host, false);
7214 }
7215 
7216 static void vmx_update_hv_timer(struct kvm_vcpu *vcpu, bool force_immediate_exit)
7217 {
7218 	struct vcpu_vmx *vmx = to_vmx(vcpu);
7219 	u64 tscl;
7220 	u32 delta_tsc;
7221 
7222 	if (force_immediate_exit) {
7223 		vmcs_write32(VMX_PREEMPTION_TIMER_VALUE, 0);
7224 		vmx->loaded_vmcs->hv_timer_soft_disabled = false;
7225 	} else if (vmx->hv_deadline_tsc != -1) {
7226 		tscl = rdtsc();
7227 		if (vmx->hv_deadline_tsc > tscl)
7228 			/* set_hv_timer ensures the delta fits in 32-bits */
7229 			delta_tsc = (u32)((vmx->hv_deadline_tsc - tscl) >>
7230 				cpu_preemption_timer_multi);
7231 		else
7232 			delta_tsc = 0;
7233 
7234 		vmcs_write32(VMX_PREEMPTION_TIMER_VALUE, delta_tsc);
7235 		vmx->loaded_vmcs->hv_timer_soft_disabled = false;
7236 	} else if (!vmx->loaded_vmcs->hv_timer_soft_disabled) {
7237 		vmcs_write32(VMX_PREEMPTION_TIMER_VALUE, -1);
7238 		vmx->loaded_vmcs->hv_timer_soft_disabled = true;
7239 	}
7240 }
7241 
7242 void noinstr vmx_update_host_rsp(struct vcpu_vmx *vmx, unsigned long host_rsp)
7243 {
7244 	if (unlikely(host_rsp != vmx->loaded_vmcs->host_state.rsp)) {
7245 		vmx->loaded_vmcs->host_state.rsp = host_rsp;
7246 		vmcs_writel(HOST_RSP, host_rsp);
7247 	}
7248 }
7249 
7250 void noinstr vmx_spec_ctrl_restore_host(struct vcpu_vmx *vmx,
7251 					unsigned int flags)
7252 {
7253 	u64 hostval = this_cpu_read(x86_spec_ctrl_current);
7254 
7255 	if (!cpu_feature_enabled(X86_FEATURE_MSR_SPEC_CTRL))
7256 		return;
7257 
7258 	if (flags & VMX_RUN_SAVE_SPEC_CTRL)
7259 		vmx->spec_ctrl = __rdmsr(MSR_IA32_SPEC_CTRL);
7260 
7261 	/*
7262 	 * If the guest/host SPEC_CTRL values differ, restore the host value.
7263 	 *
7264 	 * For legacy IBRS, the IBRS bit always needs to be written after
7265 	 * transitioning from a less privileged predictor mode, regardless of
7266 	 * whether the guest/host values differ.
7267 	 */
7268 	if (cpu_feature_enabled(X86_FEATURE_KERNEL_IBRS) ||
7269 	    vmx->spec_ctrl != hostval)
7270 		native_wrmsrl(MSR_IA32_SPEC_CTRL, hostval);
7271 
7272 	barrier_nospec();
7273 }
7274 
7275 static fastpath_t vmx_exit_handlers_fastpath(struct kvm_vcpu *vcpu,
7276 					     bool force_immediate_exit)
7277 {
7278 	/*
7279 	 * If L2 is active, some VMX preemption timer exits can be handled in
7280 	 * the fastpath even, all other exits must use the slow path.
7281 	 */
7282 	if (is_guest_mode(vcpu) &&
7283 	    to_vmx(vcpu)->exit_reason.basic != EXIT_REASON_PREEMPTION_TIMER)
7284 		return EXIT_FASTPATH_NONE;
7285 
7286 	switch (to_vmx(vcpu)->exit_reason.basic) {
7287 	case EXIT_REASON_MSR_WRITE:
7288 		return handle_fastpath_set_msr_irqoff(vcpu);
7289 	case EXIT_REASON_PREEMPTION_TIMER:
7290 		return handle_fastpath_preemption_timer(vcpu, force_immediate_exit);
7291 	case EXIT_REASON_HLT:
7292 		return handle_fastpath_hlt(vcpu);
7293 	default:
7294 		return EXIT_FASTPATH_NONE;
7295 	}
7296 }
7297 
7298 static noinstr void vmx_vcpu_enter_exit(struct kvm_vcpu *vcpu,
7299 					unsigned int flags)
7300 {
7301 	struct vcpu_vmx *vmx = to_vmx(vcpu);
7302 
7303 	guest_state_enter_irqoff();
7304 
7305 	/*
7306 	 * L1D Flush includes CPU buffer clear to mitigate MDS, but VERW
7307 	 * mitigation for MDS is done late in VMentry and is still
7308 	 * executed in spite of L1D Flush. This is because an extra VERW
7309 	 * should not matter much after the big hammer L1D Flush.
7310 	 */
7311 	if (static_branch_unlikely(&vmx_l1d_should_flush))
7312 		vmx_l1d_flush(vcpu);
7313 	else if (static_branch_unlikely(&mmio_stale_data_clear) &&
7314 		 kvm_arch_has_assigned_device(vcpu->kvm))
7315 		mds_clear_cpu_buffers();
7316 
7317 	vmx_disable_fb_clear(vmx);
7318 
7319 	if (vcpu->arch.cr2 != native_read_cr2())
7320 		native_write_cr2(vcpu->arch.cr2);
7321 
7322 	vmx->fail = __vmx_vcpu_run(vmx, (unsigned long *)&vcpu->arch.regs,
7323 				   flags);
7324 
7325 	vcpu->arch.cr2 = native_read_cr2();
7326 	vcpu->arch.regs_avail &= ~VMX_REGS_LAZY_LOAD_SET;
7327 
7328 	vmx->idt_vectoring_info = 0;
7329 
7330 	vmx_enable_fb_clear(vmx);
7331 
7332 	if (unlikely(vmx->fail)) {
7333 		vmx->exit_reason.full = 0xdead;
7334 		goto out;
7335 	}
7336 
7337 	vmx->exit_reason.full = vmcs_read32(VM_EXIT_REASON);
7338 	if (likely(!vmx->exit_reason.failed_vmentry))
7339 		vmx->idt_vectoring_info = vmcs_read32(IDT_VECTORING_INFO_FIELD);
7340 
7341 	if ((u16)vmx->exit_reason.basic == EXIT_REASON_EXCEPTION_NMI &&
7342 	    is_nmi(vmx_get_intr_info(vcpu))) {
7343 		kvm_before_interrupt(vcpu, KVM_HANDLING_NMI);
7344 		if (cpu_feature_enabled(X86_FEATURE_FRED))
7345 			fred_entry_from_kvm(EVENT_TYPE_NMI, NMI_VECTOR);
7346 		else
7347 			vmx_do_nmi_irqoff();
7348 		kvm_after_interrupt(vcpu);
7349 	}
7350 
7351 out:
7352 	guest_state_exit_irqoff();
7353 }
7354 
7355 fastpath_t vmx_vcpu_run(struct kvm_vcpu *vcpu, bool force_immediate_exit)
7356 {
7357 	struct vcpu_vmx *vmx = to_vmx(vcpu);
7358 	unsigned long cr3, cr4;
7359 
7360 	/* Record the guest's net vcpu time for enforced NMI injections. */
7361 	if (unlikely(!enable_vnmi &&
7362 		     vmx->loaded_vmcs->soft_vnmi_blocked))
7363 		vmx->loaded_vmcs->entry_time = ktime_get();
7364 
7365 	/*
7366 	 * Don't enter VMX if guest state is invalid, let the exit handler
7367 	 * start emulation until we arrive back to a valid state.  Synthesize a
7368 	 * consistency check VM-Exit due to invalid guest state and bail.
7369 	 */
7370 	if (unlikely(vmx->emulation_required)) {
7371 		vmx->fail = 0;
7372 
7373 		vmx->exit_reason.full = EXIT_REASON_INVALID_STATE;
7374 		vmx->exit_reason.failed_vmentry = 1;
7375 		kvm_register_mark_available(vcpu, VCPU_EXREG_EXIT_INFO_1);
7376 		vmx->exit_qualification = ENTRY_FAIL_DEFAULT;
7377 		kvm_register_mark_available(vcpu, VCPU_EXREG_EXIT_INFO_2);
7378 		vmx->exit_intr_info = 0;
7379 		return EXIT_FASTPATH_NONE;
7380 	}
7381 
7382 	trace_kvm_entry(vcpu, force_immediate_exit);
7383 
7384 	if (vmx->ple_window_dirty) {
7385 		vmx->ple_window_dirty = false;
7386 		vmcs_write32(PLE_WINDOW, vmx->ple_window);
7387 	}
7388 
7389 	/*
7390 	 * We did this in prepare_switch_to_guest, because it needs to
7391 	 * be within srcu_read_lock.
7392 	 */
7393 	WARN_ON_ONCE(vmx->nested.need_vmcs12_to_shadow_sync);
7394 
7395 	if (kvm_register_is_dirty(vcpu, VCPU_REGS_RSP))
7396 		vmcs_writel(GUEST_RSP, vcpu->arch.regs[VCPU_REGS_RSP]);
7397 	if (kvm_register_is_dirty(vcpu, VCPU_REGS_RIP))
7398 		vmcs_writel(GUEST_RIP, vcpu->arch.regs[VCPU_REGS_RIP]);
7399 	vcpu->arch.regs_dirty = 0;
7400 
7401 	/*
7402 	 * Refresh vmcs.HOST_CR3 if necessary.  This must be done immediately
7403 	 * prior to VM-Enter, as the kernel may load a new ASID (PCID) any time
7404 	 * it switches back to the current->mm, which can occur in KVM context
7405 	 * when switching to a temporary mm to patch kernel code, e.g. if KVM
7406 	 * toggles a static key while handling a VM-Exit.
7407 	 */
7408 	cr3 = __get_current_cr3_fast();
7409 	if (unlikely(cr3 != vmx->loaded_vmcs->host_state.cr3)) {
7410 		vmcs_writel(HOST_CR3, cr3);
7411 		vmx->loaded_vmcs->host_state.cr3 = cr3;
7412 	}
7413 
7414 	cr4 = cr4_read_shadow();
7415 	if (unlikely(cr4 != vmx->loaded_vmcs->host_state.cr4)) {
7416 		vmcs_writel(HOST_CR4, cr4);
7417 		vmx->loaded_vmcs->host_state.cr4 = cr4;
7418 	}
7419 
7420 	/* When KVM_DEBUGREG_WONT_EXIT, dr6 is accessible in guest. */
7421 	if (unlikely(vcpu->arch.switch_db_regs & KVM_DEBUGREG_WONT_EXIT))
7422 		set_debugreg(vcpu->arch.dr6, 6);
7423 
7424 	/* When single-stepping over STI and MOV SS, we must clear the
7425 	 * corresponding interruptibility bits in the guest state. Otherwise
7426 	 * vmentry fails as it then expects bit 14 (BS) in pending debug
7427 	 * exceptions being set, but that's not correct for the guest debugging
7428 	 * case. */
7429 	if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)
7430 		vmx_set_interrupt_shadow(vcpu, 0);
7431 
7432 	kvm_load_guest_xsave_state(vcpu);
7433 
7434 	pt_guest_enter(vmx);
7435 
7436 	atomic_switch_perf_msrs(vmx);
7437 	if (intel_pmu_lbr_is_enabled(vcpu))
7438 		vmx_passthrough_lbr_msrs(vcpu);
7439 
7440 	if (enable_preemption_timer)
7441 		vmx_update_hv_timer(vcpu, force_immediate_exit);
7442 	else if (force_immediate_exit)
7443 		smp_send_reschedule(vcpu->cpu);
7444 
7445 	kvm_wait_lapic_expire(vcpu);
7446 
7447 	/* The actual VMENTER/EXIT is in the .noinstr.text section. */
7448 	vmx_vcpu_enter_exit(vcpu, __vmx_vcpu_run_flags(vmx));
7449 
7450 	/* All fields are clean at this point */
7451 	if (kvm_is_using_evmcs()) {
7452 		current_evmcs->hv_clean_fields |=
7453 			HV_VMX_ENLIGHTENED_CLEAN_FIELD_ALL;
7454 
7455 		current_evmcs->hv_vp_id = kvm_hv_get_vpindex(vcpu);
7456 	}
7457 
7458 	/* MSR_IA32_DEBUGCTLMSR is zeroed on vmexit. Restore it if needed */
7459 	if (vmx->host_debugctlmsr)
7460 		update_debugctlmsr(vmx->host_debugctlmsr);
7461 
7462 #ifndef CONFIG_X86_64
7463 	/*
7464 	 * The sysexit path does not restore ds/es, so we must set them to
7465 	 * a reasonable value ourselves.
7466 	 *
7467 	 * We can't defer this to vmx_prepare_switch_to_host() since that
7468 	 * function may be executed in interrupt context, which saves and
7469 	 * restore segments around it, nullifying its effect.
7470 	 */
7471 	loadsegment(ds, __USER_DS);
7472 	loadsegment(es, __USER_DS);
7473 #endif
7474 
7475 	pt_guest_exit(vmx);
7476 
7477 	kvm_load_host_xsave_state(vcpu);
7478 
7479 	if (is_guest_mode(vcpu)) {
7480 		/*
7481 		 * Track VMLAUNCH/VMRESUME that have made past guest state
7482 		 * checking.
7483 		 */
7484 		if (vmx->nested.nested_run_pending &&
7485 		    !vmx->exit_reason.failed_vmentry)
7486 			++vcpu->stat.nested_run;
7487 
7488 		vmx->nested.nested_run_pending = 0;
7489 	}
7490 
7491 	if (unlikely(vmx->fail))
7492 		return EXIT_FASTPATH_NONE;
7493 
7494 	if (unlikely((u16)vmx->exit_reason.basic == EXIT_REASON_MCE_DURING_VMENTRY))
7495 		kvm_machine_check();
7496 
7497 	trace_kvm_exit(vcpu, KVM_ISA_VMX);
7498 
7499 	if (unlikely(vmx->exit_reason.failed_vmentry))
7500 		return EXIT_FASTPATH_NONE;
7501 
7502 	vmx->loaded_vmcs->launched = 1;
7503 
7504 	vmx_recover_nmi_blocking(vmx);
7505 	vmx_complete_interrupts(vmx);
7506 
7507 	return vmx_exit_handlers_fastpath(vcpu, force_immediate_exit);
7508 }
7509 
7510 void vmx_vcpu_free(struct kvm_vcpu *vcpu)
7511 {
7512 	struct vcpu_vmx *vmx = to_vmx(vcpu);
7513 
7514 	if (enable_pml)
7515 		vmx_destroy_pml_buffer(vmx);
7516 	free_vpid(vmx->vpid);
7517 	nested_vmx_free_vcpu(vcpu);
7518 	free_loaded_vmcs(vmx->loaded_vmcs);
7519 	free_page((unsigned long)vmx->ve_info);
7520 }
7521 
7522 int vmx_vcpu_create(struct kvm_vcpu *vcpu)
7523 {
7524 	struct vmx_uret_msr *tsx_ctrl;
7525 	struct vcpu_vmx *vmx;
7526 	int i, err;
7527 
7528 	BUILD_BUG_ON(offsetof(struct vcpu_vmx, vcpu) != 0);
7529 	vmx = to_vmx(vcpu);
7530 
7531 	INIT_LIST_HEAD(&vmx->pi_wakeup_list);
7532 
7533 	err = -ENOMEM;
7534 
7535 	vmx->vpid = allocate_vpid();
7536 
7537 	/*
7538 	 * If PML is turned on, failure on enabling PML just results in failure
7539 	 * of creating the vcpu, therefore we can simplify PML logic (by
7540 	 * avoiding dealing with cases, such as enabling PML partially on vcpus
7541 	 * for the guest), etc.
7542 	 */
7543 	if (enable_pml) {
7544 		vmx->pml_pg = alloc_page(GFP_KERNEL_ACCOUNT | __GFP_ZERO);
7545 		if (!vmx->pml_pg)
7546 			goto free_vpid;
7547 	}
7548 
7549 	for (i = 0; i < kvm_nr_uret_msrs; ++i)
7550 		vmx->guest_uret_msrs[i].mask = -1ull;
7551 	if (boot_cpu_has(X86_FEATURE_RTM)) {
7552 		/*
7553 		 * TSX_CTRL_CPUID_CLEAR is handled in the CPUID interception.
7554 		 * Keep the host value unchanged to avoid changing CPUID bits
7555 		 * under the host kernel's feet.
7556 		 */
7557 		tsx_ctrl = vmx_find_uret_msr(vmx, MSR_IA32_TSX_CTRL);
7558 		if (tsx_ctrl)
7559 			tsx_ctrl->mask = ~(u64)TSX_CTRL_CPUID_CLEAR;
7560 	}
7561 
7562 	err = alloc_loaded_vmcs(&vmx->vmcs01);
7563 	if (err < 0)
7564 		goto free_pml;
7565 
7566 	/*
7567 	 * Use Hyper-V 'Enlightened MSR Bitmap' feature when KVM runs as a
7568 	 * nested (L1) hypervisor and Hyper-V in L0 supports it. Enable the
7569 	 * feature only for vmcs01, KVM currently isn't equipped to realize any
7570 	 * performance benefits from enabling it for vmcs02.
7571 	 */
7572 	if (kvm_is_using_evmcs() &&
7573 	    (ms_hyperv.nested_features & HV_X64_NESTED_MSR_BITMAP)) {
7574 		struct hv_enlightened_vmcs *evmcs = (void *)vmx->vmcs01.vmcs;
7575 
7576 		evmcs->hv_enlightenments_control.msr_bitmap = 1;
7577 	}
7578 
7579 	/* The MSR bitmap starts with all ones */
7580 	bitmap_fill(vmx->shadow_msr_intercept.read, MAX_POSSIBLE_PASSTHROUGH_MSRS);
7581 	bitmap_fill(vmx->shadow_msr_intercept.write, MAX_POSSIBLE_PASSTHROUGH_MSRS);
7582 
7583 	vmx_disable_intercept_for_msr(vcpu, MSR_IA32_TSC, MSR_TYPE_R);
7584 #ifdef CONFIG_X86_64
7585 	vmx_disable_intercept_for_msr(vcpu, MSR_FS_BASE, MSR_TYPE_RW);
7586 	vmx_disable_intercept_for_msr(vcpu, MSR_GS_BASE, MSR_TYPE_RW);
7587 	vmx_disable_intercept_for_msr(vcpu, MSR_KERNEL_GS_BASE, MSR_TYPE_RW);
7588 #endif
7589 	vmx_disable_intercept_for_msr(vcpu, MSR_IA32_SYSENTER_CS, MSR_TYPE_RW);
7590 	vmx_disable_intercept_for_msr(vcpu, MSR_IA32_SYSENTER_ESP, MSR_TYPE_RW);
7591 	vmx_disable_intercept_for_msr(vcpu, MSR_IA32_SYSENTER_EIP, MSR_TYPE_RW);
7592 	if (kvm_cstate_in_guest(vcpu->kvm)) {
7593 		vmx_disable_intercept_for_msr(vcpu, MSR_CORE_C1_RES, MSR_TYPE_R);
7594 		vmx_disable_intercept_for_msr(vcpu, MSR_CORE_C3_RESIDENCY, MSR_TYPE_R);
7595 		vmx_disable_intercept_for_msr(vcpu, MSR_CORE_C6_RESIDENCY, MSR_TYPE_R);
7596 		vmx_disable_intercept_for_msr(vcpu, MSR_CORE_C7_RESIDENCY, MSR_TYPE_R);
7597 	}
7598 
7599 	vmx->loaded_vmcs = &vmx->vmcs01;
7600 
7601 	if (cpu_need_virtualize_apic_accesses(vcpu)) {
7602 		err = kvm_alloc_apic_access_page(vcpu->kvm);
7603 		if (err)
7604 			goto free_vmcs;
7605 	}
7606 
7607 	if (enable_ept && !enable_unrestricted_guest) {
7608 		err = init_rmode_identity_map(vcpu->kvm);
7609 		if (err)
7610 			goto free_vmcs;
7611 	}
7612 
7613 	err = -ENOMEM;
7614 	if (vmcs_config.cpu_based_2nd_exec_ctrl & SECONDARY_EXEC_EPT_VIOLATION_VE) {
7615 		struct page *page;
7616 
7617 		BUILD_BUG_ON(sizeof(*vmx->ve_info) > PAGE_SIZE);
7618 
7619 		/* ve_info must be page aligned. */
7620 		page = alloc_page(GFP_KERNEL_ACCOUNT | __GFP_ZERO);
7621 		if (!page)
7622 			goto free_vmcs;
7623 
7624 		vmx->ve_info = page_to_virt(page);
7625 	}
7626 
7627 	if (vmx_can_use_ipiv(vcpu))
7628 		WRITE_ONCE(to_kvm_vmx(vcpu->kvm)->pid_table[vcpu->vcpu_id],
7629 			   __pa(&vmx->pi_desc) | PID_TABLE_ENTRY_VALID);
7630 
7631 	return 0;
7632 
7633 free_vmcs:
7634 	free_loaded_vmcs(vmx->loaded_vmcs);
7635 free_pml:
7636 	vmx_destroy_pml_buffer(vmx);
7637 free_vpid:
7638 	free_vpid(vmx->vpid);
7639 	return err;
7640 }
7641 
7642 #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"
7643 #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"
7644 
7645 int vmx_vm_init(struct kvm *kvm)
7646 {
7647 	if (!ple_gap)
7648 		kvm->arch.pause_in_guest = true;
7649 
7650 	if (boot_cpu_has(X86_BUG_L1TF) && enable_ept) {
7651 		switch (l1tf_mitigation) {
7652 		case L1TF_MITIGATION_OFF:
7653 		case L1TF_MITIGATION_FLUSH_NOWARN:
7654 			/* 'I explicitly don't care' is set */
7655 			break;
7656 		case L1TF_MITIGATION_FLUSH:
7657 		case L1TF_MITIGATION_FLUSH_NOSMT:
7658 		case L1TF_MITIGATION_FULL:
7659 			/*
7660 			 * Warn upon starting the first VM in a potentially
7661 			 * insecure environment.
7662 			 */
7663 			if (sched_smt_active())
7664 				pr_warn_once(L1TF_MSG_SMT);
7665 			if (l1tf_vmx_mitigation == VMENTER_L1D_FLUSH_NEVER)
7666 				pr_warn_once(L1TF_MSG_L1D);
7667 			break;
7668 		case L1TF_MITIGATION_FULL_FORCE:
7669 			/* Flush is enforced */
7670 			break;
7671 		}
7672 	}
7673 	return 0;
7674 }
7675 
7676 u8 vmx_get_mt_mask(struct kvm_vcpu *vcpu, gfn_t gfn, bool is_mmio)
7677 {
7678 	/*
7679 	 * Force UC for host MMIO regions, as allowing the guest to access MMIO
7680 	 * with cacheable accesses will result in Machine Checks.
7681 	 */
7682 	if (is_mmio)
7683 		return MTRR_TYPE_UNCACHABLE << VMX_EPT_MT_EPTE_SHIFT;
7684 
7685 	/*
7686 	 * Force WB and ignore guest PAT if the VM does NOT have a non-coherent
7687 	 * device attached.  Letting the guest control memory types on Intel
7688 	 * CPUs may result in unexpected behavior, and so KVM's ABI is to trust
7689 	 * the guest to behave only as a last resort.
7690 	 */
7691 	if (!kvm_arch_has_noncoherent_dma(vcpu->kvm))
7692 		return (MTRR_TYPE_WRBACK << VMX_EPT_MT_EPTE_SHIFT) | VMX_EPT_IPAT_BIT;
7693 
7694 	return (MTRR_TYPE_WRBACK << VMX_EPT_MT_EPTE_SHIFT);
7695 }
7696 
7697 static void vmcs_set_secondary_exec_control(struct vcpu_vmx *vmx, u32 new_ctl)
7698 {
7699 	/*
7700 	 * These bits in the secondary execution controls field
7701 	 * are dynamic, the others are mostly based on the hypervisor
7702 	 * architecture and the guest's CPUID.  Do not touch the
7703 	 * dynamic bits.
7704 	 */
7705 	u32 mask =
7706 		SECONDARY_EXEC_SHADOW_VMCS |
7707 		SECONDARY_EXEC_VIRTUALIZE_X2APIC_MODE |
7708 		SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES |
7709 		SECONDARY_EXEC_DESC;
7710 
7711 	u32 cur_ctl = secondary_exec_controls_get(vmx);
7712 
7713 	secondary_exec_controls_set(vmx, (new_ctl & ~mask) | (cur_ctl & mask));
7714 }
7715 
7716 /*
7717  * Generate MSR_IA32_VMX_CR{0,4}_FIXED1 according to CPUID. Only set bits
7718  * (indicating "allowed-1") if they are supported in the guest's CPUID.
7719  */
7720 static void nested_vmx_cr_fixed1_bits_update(struct kvm_vcpu *vcpu)
7721 {
7722 	struct vcpu_vmx *vmx = to_vmx(vcpu);
7723 	struct kvm_cpuid_entry2 *entry;
7724 
7725 	vmx->nested.msrs.cr0_fixed1 = 0xffffffff;
7726 	vmx->nested.msrs.cr4_fixed1 = X86_CR4_PCE;
7727 
7728 #define cr4_fixed1_update(_cr4_mask, _reg, _cpuid_mask) do {		\
7729 	if (entry && (entry->_reg & (_cpuid_mask)))			\
7730 		vmx->nested.msrs.cr4_fixed1 |= (_cr4_mask);	\
7731 } while (0)
7732 
7733 	entry = kvm_find_cpuid_entry(vcpu, 0x1);
7734 	cr4_fixed1_update(X86_CR4_VME,        edx, feature_bit(VME));
7735 	cr4_fixed1_update(X86_CR4_PVI,        edx, feature_bit(VME));
7736 	cr4_fixed1_update(X86_CR4_TSD,        edx, feature_bit(TSC));
7737 	cr4_fixed1_update(X86_CR4_DE,         edx, feature_bit(DE));
7738 	cr4_fixed1_update(X86_CR4_PSE,        edx, feature_bit(PSE));
7739 	cr4_fixed1_update(X86_CR4_PAE,        edx, feature_bit(PAE));
7740 	cr4_fixed1_update(X86_CR4_MCE,        edx, feature_bit(MCE));
7741 	cr4_fixed1_update(X86_CR4_PGE,        edx, feature_bit(PGE));
7742 	cr4_fixed1_update(X86_CR4_OSFXSR,     edx, feature_bit(FXSR));
7743 	cr4_fixed1_update(X86_CR4_OSXMMEXCPT, edx, feature_bit(XMM));
7744 	cr4_fixed1_update(X86_CR4_VMXE,       ecx, feature_bit(VMX));
7745 	cr4_fixed1_update(X86_CR4_SMXE,       ecx, feature_bit(SMX));
7746 	cr4_fixed1_update(X86_CR4_PCIDE,      ecx, feature_bit(PCID));
7747 	cr4_fixed1_update(X86_CR4_OSXSAVE,    ecx, feature_bit(XSAVE));
7748 
7749 	entry = kvm_find_cpuid_entry_index(vcpu, 0x7, 0);
7750 	cr4_fixed1_update(X86_CR4_FSGSBASE,   ebx, feature_bit(FSGSBASE));
7751 	cr4_fixed1_update(X86_CR4_SMEP,       ebx, feature_bit(SMEP));
7752 	cr4_fixed1_update(X86_CR4_SMAP,       ebx, feature_bit(SMAP));
7753 	cr4_fixed1_update(X86_CR4_PKE,        ecx, feature_bit(PKU));
7754 	cr4_fixed1_update(X86_CR4_UMIP,       ecx, feature_bit(UMIP));
7755 	cr4_fixed1_update(X86_CR4_LA57,       ecx, feature_bit(LA57));
7756 
7757 	entry = kvm_find_cpuid_entry_index(vcpu, 0x7, 1);
7758 	cr4_fixed1_update(X86_CR4_LAM_SUP,    eax, feature_bit(LAM));
7759 
7760 #undef cr4_fixed1_update
7761 }
7762 
7763 static void update_intel_pt_cfg(struct kvm_vcpu *vcpu)
7764 {
7765 	struct vcpu_vmx *vmx = to_vmx(vcpu);
7766 	struct kvm_cpuid_entry2 *best = NULL;
7767 	int i;
7768 
7769 	for (i = 0; i < PT_CPUID_LEAVES; i++) {
7770 		best = kvm_find_cpuid_entry_index(vcpu, 0x14, i);
7771 		if (!best)
7772 			return;
7773 		vmx->pt_desc.caps[CPUID_EAX + i*PT_CPUID_REGS_NUM] = best->eax;
7774 		vmx->pt_desc.caps[CPUID_EBX + i*PT_CPUID_REGS_NUM] = best->ebx;
7775 		vmx->pt_desc.caps[CPUID_ECX + i*PT_CPUID_REGS_NUM] = best->ecx;
7776 		vmx->pt_desc.caps[CPUID_EDX + i*PT_CPUID_REGS_NUM] = best->edx;
7777 	}
7778 
7779 	/* Get the number of configurable Address Ranges for filtering */
7780 	vmx->pt_desc.num_address_ranges = intel_pt_validate_cap(vmx->pt_desc.caps,
7781 						PT_CAP_num_address_ranges);
7782 
7783 	/* Initialize and clear the no dependency bits */
7784 	vmx->pt_desc.ctl_bitmask = ~(RTIT_CTL_TRACEEN | RTIT_CTL_OS |
7785 			RTIT_CTL_USR | RTIT_CTL_TSC_EN | RTIT_CTL_DISRETC |
7786 			RTIT_CTL_BRANCH_EN);
7787 
7788 	/*
7789 	 * If CPUID.(EAX=14H,ECX=0):EBX[0]=1 CR3Filter can be set otherwise
7790 	 * will inject an #GP
7791 	 */
7792 	if (intel_pt_validate_cap(vmx->pt_desc.caps, PT_CAP_cr3_filtering))
7793 		vmx->pt_desc.ctl_bitmask &= ~RTIT_CTL_CR3EN;
7794 
7795 	/*
7796 	 * If CPUID.(EAX=14H,ECX=0):EBX[1]=1 CYCEn, CycThresh and
7797 	 * PSBFreq can be set
7798 	 */
7799 	if (intel_pt_validate_cap(vmx->pt_desc.caps, PT_CAP_psb_cyc))
7800 		vmx->pt_desc.ctl_bitmask &= ~(RTIT_CTL_CYCLEACC |
7801 				RTIT_CTL_CYC_THRESH | RTIT_CTL_PSB_FREQ);
7802 
7803 	/*
7804 	 * If CPUID.(EAX=14H,ECX=0):EBX[3]=1 MTCEn and MTCFreq can be set
7805 	 */
7806 	if (intel_pt_validate_cap(vmx->pt_desc.caps, PT_CAP_mtc))
7807 		vmx->pt_desc.ctl_bitmask &= ~(RTIT_CTL_MTC_EN |
7808 					      RTIT_CTL_MTC_RANGE);
7809 
7810 	/* If CPUID.(EAX=14H,ECX=0):EBX[4]=1 FUPonPTW and PTWEn can be set */
7811 	if (intel_pt_validate_cap(vmx->pt_desc.caps, PT_CAP_ptwrite))
7812 		vmx->pt_desc.ctl_bitmask &= ~(RTIT_CTL_FUP_ON_PTW |
7813 							RTIT_CTL_PTW_EN);
7814 
7815 	/* If CPUID.(EAX=14H,ECX=0):EBX[5]=1 PwrEvEn can be set */
7816 	if (intel_pt_validate_cap(vmx->pt_desc.caps, PT_CAP_power_event_trace))
7817 		vmx->pt_desc.ctl_bitmask &= ~RTIT_CTL_PWR_EVT_EN;
7818 
7819 	/* If CPUID.(EAX=14H,ECX=0):ECX[0]=1 ToPA can be set */
7820 	if (intel_pt_validate_cap(vmx->pt_desc.caps, PT_CAP_topa_output))
7821 		vmx->pt_desc.ctl_bitmask &= ~RTIT_CTL_TOPA;
7822 
7823 	/* If CPUID.(EAX=14H,ECX=0):ECX[3]=1 FabricEn can be set */
7824 	if (intel_pt_validate_cap(vmx->pt_desc.caps, PT_CAP_output_subsys))
7825 		vmx->pt_desc.ctl_bitmask &= ~RTIT_CTL_FABRIC_EN;
7826 
7827 	/* unmask address range configure area */
7828 	for (i = 0; i < vmx->pt_desc.num_address_ranges; i++)
7829 		vmx->pt_desc.ctl_bitmask &= ~(0xfULL << (32 + i * 4));
7830 }
7831 
7832 void vmx_vcpu_after_set_cpuid(struct kvm_vcpu *vcpu)
7833 {
7834 	struct vcpu_vmx *vmx = to_vmx(vcpu);
7835 
7836 	/*
7837 	 * XSAVES is effectively enabled if and only if XSAVE is also exposed
7838 	 * to the guest.  XSAVES depends on CR4.OSXSAVE, and CR4.OSXSAVE can be
7839 	 * set if and only if XSAVE is supported.
7840 	 */
7841 	if (!guest_cpu_cap_has(vcpu, X86_FEATURE_XSAVE))
7842 		guest_cpu_cap_clear(vcpu, X86_FEATURE_XSAVES);
7843 
7844 	vmx_setup_uret_msrs(vmx);
7845 
7846 	if (cpu_has_secondary_exec_ctrls())
7847 		vmcs_set_secondary_exec_control(vmx,
7848 						vmx_secondary_exec_control(vmx));
7849 
7850 	if (guest_cpu_cap_has(vcpu, X86_FEATURE_VMX))
7851 		vmx->msr_ia32_feature_control_valid_bits |=
7852 			FEAT_CTL_VMX_ENABLED_INSIDE_SMX |
7853 			FEAT_CTL_VMX_ENABLED_OUTSIDE_SMX;
7854 	else
7855 		vmx->msr_ia32_feature_control_valid_bits &=
7856 			~(FEAT_CTL_VMX_ENABLED_INSIDE_SMX |
7857 			  FEAT_CTL_VMX_ENABLED_OUTSIDE_SMX);
7858 
7859 	if (guest_cpu_cap_has(vcpu, X86_FEATURE_VMX))
7860 		nested_vmx_cr_fixed1_bits_update(vcpu);
7861 
7862 	if (boot_cpu_has(X86_FEATURE_INTEL_PT) &&
7863 			guest_cpu_cap_has(vcpu, X86_FEATURE_INTEL_PT))
7864 		update_intel_pt_cfg(vcpu);
7865 
7866 	if (boot_cpu_has(X86_FEATURE_RTM)) {
7867 		struct vmx_uret_msr *msr;
7868 		msr = vmx_find_uret_msr(vmx, MSR_IA32_TSX_CTRL);
7869 		if (msr) {
7870 			bool enabled = guest_cpu_cap_has(vcpu, X86_FEATURE_RTM);
7871 			vmx_set_guest_uret_msr(vmx, msr, enabled ? 0 : TSX_CTRL_RTM_DISABLE);
7872 		}
7873 	}
7874 
7875 	if (kvm_cpu_cap_has(X86_FEATURE_XFD))
7876 		vmx_set_intercept_for_msr(vcpu, MSR_IA32_XFD_ERR, MSR_TYPE_R,
7877 					  !guest_cpu_cap_has(vcpu, X86_FEATURE_XFD));
7878 
7879 	if (boot_cpu_has(X86_FEATURE_IBPB))
7880 		vmx_set_intercept_for_msr(vcpu, MSR_IA32_PRED_CMD, MSR_TYPE_W,
7881 					  !guest_has_pred_cmd_msr(vcpu));
7882 
7883 	if (boot_cpu_has(X86_FEATURE_FLUSH_L1D))
7884 		vmx_set_intercept_for_msr(vcpu, MSR_IA32_FLUSH_CMD, MSR_TYPE_W,
7885 					  !guest_cpu_cap_has(vcpu, X86_FEATURE_FLUSH_L1D));
7886 
7887 	set_cr4_guest_host_mask(vmx);
7888 
7889 	vmx_write_encls_bitmap(vcpu, NULL);
7890 	if (guest_cpu_cap_has(vcpu, X86_FEATURE_SGX))
7891 		vmx->msr_ia32_feature_control_valid_bits |= FEAT_CTL_SGX_ENABLED;
7892 	else
7893 		vmx->msr_ia32_feature_control_valid_bits &= ~FEAT_CTL_SGX_ENABLED;
7894 
7895 	if (guest_cpu_cap_has(vcpu, X86_FEATURE_SGX_LC))
7896 		vmx->msr_ia32_feature_control_valid_bits |=
7897 			FEAT_CTL_SGX_LC_ENABLED;
7898 	else
7899 		vmx->msr_ia32_feature_control_valid_bits &=
7900 			~FEAT_CTL_SGX_LC_ENABLED;
7901 
7902 	/* Refresh #PF interception to account for MAXPHYADDR changes. */
7903 	vmx_update_exception_bitmap(vcpu);
7904 }
7905 
7906 static __init u64 vmx_get_perf_capabilities(void)
7907 {
7908 	u64 perf_cap = PMU_CAP_FW_WRITES;
7909 	u64 host_perf_cap = 0;
7910 
7911 	if (!enable_pmu)
7912 		return 0;
7913 
7914 	if (boot_cpu_has(X86_FEATURE_PDCM))
7915 		rdmsrl(MSR_IA32_PERF_CAPABILITIES, host_perf_cap);
7916 
7917 	if (!cpu_feature_enabled(X86_FEATURE_ARCH_LBR)) {
7918 		x86_perf_get_lbr(&vmx_lbr_caps);
7919 
7920 		/*
7921 		 * KVM requires LBR callstack support, as the overhead due to
7922 		 * context switching LBRs without said support is too high.
7923 		 * See intel_pmu_create_guest_lbr_event() for more info.
7924 		 */
7925 		if (!vmx_lbr_caps.has_callstack)
7926 			memset(&vmx_lbr_caps, 0, sizeof(vmx_lbr_caps));
7927 		else if (vmx_lbr_caps.nr)
7928 			perf_cap |= host_perf_cap & PMU_CAP_LBR_FMT;
7929 	}
7930 
7931 	if (vmx_pebs_supported()) {
7932 		perf_cap |= host_perf_cap & PERF_CAP_PEBS_MASK;
7933 
7934 		/*
7935 		 * Disallow adaptive PEBS as it is functionally broken, can be
7936 		 * used by the guest to read *host* LBRs, and can be used to
7937 		 * bypass userspace event filters.  To correctly and safely
7938 		 * support adaptive PEBS, KVM needs to:
7939 		 *
7940 		 * 1. Account for the ADAPTIVE flag when (re)programming fixed
7941 		 *    counters.
7942 		 *
7943 		 * 2. Gain support from perf (or take direct control of counter
7944 		 *    programming) to support events without adaptive PEBS
7945 		 *    enabled for the hardware counter.
7946 		 *
7947 		 * 3. Ensure LBR MSRs cannot hold host data on VM-Entry with
7948 		 *    adaptive PEBS enabled and MSR_PEBS_DATA_CFG.LBRS=1.
7949 		 *
7950 		 * 4. Document which PMU events are effectively exposed to the
7951 		 *    guest via adaptive PEBS, and make adaptive PEBS mutually
7952 		 *    exclusive with KVM_SET_PMU_EVENT_FILTER if necessary.
7953 		 */
7954 		perf_cap &= ~PERF_CAP_PEBS_BASELINE;
7955 	}
7956 
7957 	return perf_cap;
7958 }
7959 
7960 static __init void vmx_set_cpu_caps(void)
7961 {
7962 	kvm_set_cpu_caps();
7963 
7964 	/* CPUID 0x1 */
7965 	if (nested)
7966 		kvm_cpu_cap_set(X86_FEATURE_VMX);
7967 
7968 	/* CPUID 0x7 */
7969 	if (kvm_mpx_supported())
7970 		kvm_cpu_cap_check_and_set(X86_FEATURE_MPX);
7971 	if (!cpu_has_vmx_invpcid())
7972 		kvm_cpu_cap_clear(X86_FEATURE_INVPCID);
7973 	if (vmx_pt_mode_is_host_guest())
7974 		kvm_cpu_cap_check_and_set(X86_FEATURE_INTEL_PT);
7975 	if (vmx_pebs_supported()) {
7976 		kvm_cpu_cap_check_and_set(X86_FEATURE_DS);
7977 		kvm_cpu_cap_check_and_set(X86_FEATURE_DTES64);
7978 	}
7979 
7980 	if (!enable_pmu)
7981 		kvm_cpu_cap_clear(X86_FEATURE_PDCM);
7982 	kvm_caps.supported_perf_cap = vmx_get_perf_capabilities();
7983 
7984 	if (!enable_sgx) {
7985 		kvm_cpu_cap_clear(X86_FEATURE_SGX);
7986 		kvm_cpu_cap_clear(X86_FEATURE_SGX_LC);
7987 		kvm_cpu_cap_clear(X86_FEATURE_SGX1);
7988 		kvm_cpu_cap_clear(X86_FEATURE_SGX2);
7989 		kvm_cpu_cap_clear(X86_FEATURE_SGX_EDECCSSA);
7990 	}
7991 
7992 	if (vmx_umip_emulated())
7993 		kvm_cpu_cap_set(X86_FEATURE_UMIP);
7994 
7995 	/* CPUID 0xD.1 */
7996 	kvm_caps.supported_xss = 0;
7997 	if (!cpu_has_vmx_xsaves())
7998 		kvm_cpu_cap_clear(X86_FEATURE_XSAVES);
7999 
8000 	/* CPUID 0x80000001 and 0x7 (RDPID) */
8001 	if (!cpu_has_vmx_rdtscp()) {
8002 		kvm_cpu_cap_clear(X86_FEATURE_RDTSCP);
8003 		kvm_cpu_cap_clear(X86_FEATURE_RDPID);
8004 	}
8005 
8006 	if (cpu_has_vmx_waitpkg())
8007 		kvm_cpu_cap_check_and_set(X86_FEATURE_WAITPKG);
8008 }
8009 
8010 static int vmx_check_intercept_io(struct kvm_vcpu *vcpu,
8011 				  struct x86_instruction_info *info)
8012 {
8013 	struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
8014 	unsigned short port;
8015 	bool intercept;
8016 	int size;
8017 
8018 	if (info->intercept == x86_intercept_in ||
8019 	    info->intercept == x86_intercept_ins) {
8020 		port = info->src_val;
8021 		size = info->dst_bytes;
8022 	} else {
8023 		port = info->dst_val;
8024 		size = info->src_bytes;
8025 	}
8026 
8027 	/*
8028 	 * If the 'use IO bitmaps' VM-execution control is 0, IO instruction
8029 	 * VM-exits depend on the 'unconditional IO exiting' VM-execution
8030 	 * control.
8031 	 *
8032 	 * Otherwise, IO instruction VM-exits are controlled by the IO bitmaps.
8033 	 */
8034 	if (!nested_cpu_has(vmcs12, CPU_BASED_USE_IO_BITMAPS))
8035 		intercept = nested_cpu_has(vmcs12,
8036 					   CPU_BASED_UNCOND_IO_EXITING);
8037 	else
8038 		intercept = nested_vmx_check_io_bitmaps(vcpu, port, size);
8039 
8040 	/* FIXME: produce nested vmexit and return X86EMUL_INTERCEPTED.  */
8041 	return intercept ? X86EMUL_UNHANDLEABLE : X86EMUL_CONTINUE;
8042 }
8043 
8044 int vmx_check_intercept(struct kvm_vcpu *vcpu,
8045 			struct x86_instruction_info *info,
8046 			enum x86_intercept_stage stage,
8047 			struct x86_exception *exception)
8048 {
8049 	struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
8050 
8051 	switch (info->intercept) {
8052 	/*
8053 	 * RDPID causes #UD if disabled through secondary execution controls.
8054 	 * Because it is marked as EmulateOnUD, we need to intercept it here.
8055 	 * Note, RDPID is hidden behind ENABLE_RDTSCP.
8056 	 */
8057 	case x86_intercept_rdpid:
8058 		if (!nested_cpu_has2(vmcs12, SECONDARY_EXEC_ENABLE_RDTSCP)) {
8059 			exception->vector = UD_VECTOR;
8060 			exception->error_code_valid = false;
8061 			return X86EMUL_PROPAGATE_FAULT;
8062 		}
8063 		break;
8064 
8065 	case x86_intercept_in:
8066 	case x86_intercept_ins:
8067 	case x86_intercept_out:
8068 	case x86_intercept_outs:
8069 		return vmx_check_intercept_io(vcpu, info);
8070 
8071 	case x86_intercept_lgdt:
8072 	case x86_intercept_lidt:
8073 	case x86_intercept_lldt:
8074 	case x86_intercept_ltr:
8075 	case x86_intercept_sgdt:
8076 	case x86_intercept_sidt:
8077 	case x86_intercept_sldt:
8078 	case x86_intercept_str:
8079 		if (!nested_cpu_has2(vmcs12, SECONDARY_EXEC_DESC))
8080 			return X86EMUL_CONTINUE;
8081 
8082 		/* FIXME: produce nested vmexit and return X86EMUL_INTERCEPTED.  */
8083 		break;
8084 
8085 	case x86_intercept_pause:
8086 		/*
8087 		 * PAUSE is a single-byte NOP with a REPE prefix, i.e. collides
8088 		 * with vanilla NOPs in the emulator.  Apply the interception
8089 		 * check only to actual PAUSE instructions.  Don't check
8090 		 * PAUSE-loop-exiting, software can't expect a given PAUSE to
8091 		 * exit, i.e. KVM is within its rights to allow L2 to execute
8092 		 * the PAUSE.
8093 		 */
8094 		if ((info->rep_prefix != REPE_PREFIX) ||
8095 		    !nested_cpu_has2(vmcs12, CPU_BASED_PAUSE_EXITING))
8096 			return X86EMUL_CONTINUE;
8097 
8098 		break;
8099 
8100 	/* TODO: check more intercepts... */
8101 	default:
8102 		break;
8103 	}
8104 
8105 	return X86EMUL_UNHANDLEABLE;
8106 }
8107 
8108 #ifdef CONFIG_X86_64
8109 /* (a << shift) / divisor, return 1 if overflow otherwise 0 */
8110 static inline int u64_shl_div_u64(u64 a, unsigned int shift,
8111 				  u64 divisor, u64 *result)
8112 {
8113 	u64 low = a << shift, high = a >> (64 - shift);
8114 
8115 	/* To avoid the overflow on divq */
8116 	if (high >= divisor)
8117 		return 1;
8118 
8119 	/* Low hold the result, high hold rem which is discarded */
8120 	asm("divq %2\n\t" : "=a" (low), "=d" (high) :
8121 	    "rm" (divisor), "0" (low), "1" (high));
8122 	*result = low;
8123 
8124 	return 0;
8125 }
8126 
8127 int vmx_set_hv_timer(struct kvm_vcpu *vcpu, u64 guest_deadline_tsc,
8128 		     bool *expired)
8129 {
8130 	struct vcpu_vmx *vmx;
8131 	u64 tscl, guest_tscl, delta_tsc, lapic_timer_advance_cycles;
8132 	struct kvm_timer *ktimer = &vcpu->arch.apic->lapic_timer;
8133 
8134 	vmx = to_vmx(vcpu);
8135 	tscl = rdtsc();
8136 	guest_tscl = kvm_read_l1_tsc(vcpu, tscl);
8137 	delta_tsc = max(guest_deadline_tsc, guest_tscl) - guest_tscl;
8138 	lapic_timer_advance_cycles = nsec_to_cycles(vcpu,
8139 						    ktimer->timer_advance_ns);
8140 
8141 	if (delta_tsc > lapic_timer_advance_cycles)
8142 		delta_tsc -= lapic_timer_advance_cycles;
8143 	else
8144 		delta_tsc = 0;
8145 
8146 	/* Convert to host delta tsc if tsc scaling is enabled */
8147 	if (vcpu->arch.l1_tsc_scaling_ratio != kvm_caps.default_tsc_scaling_ratio &&
8148 	    delta_tsc && u64_shl_div_u64(delta_tsc,
8149 				kvm_caps.tsc_scaling_ratio_frac_bits,
8150 				vcpu->arch.l1_tsc_scaling_ratio, &delta_tsc))
8151 		return -ERANGE;
8152 
8153 	/*
8154 	 * If the delta tsc can't fit in the 32 bit after the multi shift,
8155 	 * we can't use the preemption timer.
8156 	 * It's possible that it fits on later vmentries, but checking
8157 	 * on every vmentry is costly so we just use an hrtimer.
8158 	 */
8159 	if (delta_tsc >> (cpu_preemption_timer_multi + 32))
8160 		return -ERANGE;
8161 
8162 	vmx->hv_deadline_tsc = tscl + delta_tsc;
8163 	*expired = !delta_tsc;
8164 	return 0;
8165 }
8166 
8167 void vmx_cancel_hv_timer(struct kvm_vcpu *vcpu)
8168 {
8169 	to_vmx(vcpu)->hv_deadline_tsc = -1;
8170 }
8171 #endif
8172 
8173 void vmx_update_cpu_dirty_logging(struct kvm_vcpu *vcpu)
8174 {
8175 	struct vcpu_vmx *vmx = to_vmx(vcpu);
8176 
8177 	if (WARN_ON_ONCE(!enable_pml))
8178 		return;
8179 
8180 	if (is_guest_mode(vcpu)) {
8181 		vmx->nested.update_vmcs01_cpu_dirty_logging = true;
8182 		return;
8183 	}
8184 
8185 	/*
8186 	 * Note, nr_memslots_dirty_logging can be changed concurrent with this
8187 	 * code, but in that case another update request will be made and so
8188 	 * the guest will never run with a stale PML value.
8189 	 */
8190 	if (atomic_read(&vcpu->kvm->nr_memslots_dirty_logging))
8191 		secondary_exec_controls_setbit(vmx, SECONDARY_EXEC_ENABLE_PML);
8192 	else
8193 		secondary_exec_controls_clearbit(vmx, SECONDARY_EXEC_ENABLE_PML);
8194 }
8195 
8196 void vmx_setup_mce(struct kvm_vcpu *vcpu)
8197 {
8198 	if (vcpu->arch.mcg_cap & MCG_LMCE_P)
8199 		to_vmx(vcpu)->msr_ia32_feature_control_valid_bits |=
8200 			FEAT_CTL_LMCE_ENABLED;
8201 	else
8202 		to_vmx(vcpu)->msr_ia32_feature_control_valid_bits &=
8203 			~FEAT_CTL_LMCE_ENABLED;
8204 }
8205 
8206 #ifdef CONFIG_KVM_SMM
8207 int vmx_smi_allowed(struct kvm_vcpu *vcpu, bool for_injection)
8208 {
8209 	/* we need a nested vmexit to enter SMM, postpone if run is pending */
8210 	if (to_vmx(vcpu)->nested.nested_run_pending)
8211 		return -EBUSY;
8212 	return !is_smm(vcpu);
8213 }
8214 
8215 int vmx_enter_smm(struct kvm_vcpu *vcpu, union kvm_smram *smram)
8216 {
8217 	struct vcpu_vmx *vmx = to_vmx(vcpu);
8218 
8219 	/*
8220 	 * TODO: Implement custom flows for forcing the vCPU out/in of L2 on
8221 	 * SMI and RSM.  Using the common VM-Exit + VM-Enter routines is wrong
8222 	 * SMI and RSM only modify state that is saved and restored via SMRAM.
8223 	 * E.g. most MSRs are left untouched, but many are modified by VM-Exit
8224 	 * and VM-Enter, and thus L2's values may be corrupted on SMI+RSM.
8225 	 */
8226 	vmx->nested.smm.guest_mode = is_guest_mode(vcpu);
8227 	if (vmx->nested.smm.guest_mode)
8228 		nested_vmx_vmexit(vcpu, -1, 0, 0);
8229 
8230 	vmx->nested.smm.vmxon = vmx->nested.vmxon;
8231 	vmx->nested.vmxon = false;
8232 	vmx_clear_hlt(vcpu);
8233 	return 0;
8234 }
8235 
8236 int vmx_leave_smm(struct kvm_vcpu *vcpu, const union kvm_smram *smram)
8237 {
8238 	struct vcpu_vmx *vmx = to_vmx(vcpu);
8239 	int ret;
8240 
8241 	if (vmx->nested.smm.vmxon) {
8242 		vmx->nested.vmxon = true;
8243 		vmx->nested.smm.vmxon = false;
8244 	}
8245 
8246 	if (vmx->nested.smm.guest_mode) {
8247 		ret = nested_vmx_enter_non_root_mode(vcpu, false);
8248 		if (ret)
8249 			return ret;
8250 
8251 		vmx->nested.nested_run_pending = 1;
8252 		vmx->nested.smm.guest_mode = false;
8253 	}
8254 	return 0;
8255 }
8256 
8257 void vmx_enable_smi_window(struct kvm_vcpu *vcpu)
8258 {
8259 	/* RSM will cause a vmexit anyway.  */
8260 }
8261 #endif
8262 
8263 bool vmx_apic_init_signal_blocked(struct kvm_vcpu *vcpu)
8264 {
8265 	return to_vmx(vcpu)->nested.vmxon && !is_guest_mode(vcpu);
8266 }
8267 
8268 void vmx_migrate_timers(struct kvm_vcpu *vcpu)
8269 {
8270 	if (is_guest_mode(vcpu)) {
8271 		struct hrtimer *timer = &to_vmx(vcpu)->nested.preemption_timer;
8272 
8273 		if (hrtimer_try_to_cancel(timer) == 1)
8274 			hrtimer_start_expires(timer, HRTIMER_MODE_ABS_PINNED);
8275 	}
8276 }
8277 
8278 void vmx_hardware_unsetup(void)
8279 {
8280 	kvm_set_posted_intr_wakeup_handler(NULL);
8281 
8282 	if (nested)
8283 		nested_vmx_hardware_unsetup();
8284 
8285 	free_kvm_area();
8286 }
8287 
8288 void vmx_vm_destroy(struct kvm *kvm)
8289 {
8290 	struct kvm_vmx *kvm_vmx = to_kvm_vmx(kvm);
8291 
8292 	free_pages((unsigned long)kvm_vmx->pid_table, vmx_get_pid_table_order(kvm));
8293 }
8294 
8295 /*
8296  * Note, the SDM states that the linear address is masked *after* the modified
8297  * canonicality check, whereas KVM masks (untags) the address and then performs
8298  * a "normal" canonicality check.  Functionally, the two methods are identical,
8299  * and when the masking occurs relative to the canonicality check isn't visible
8300  * to software, i.e. KVM's behavior doesn't violate the SDM.
8301  */
8302 gva_t vmx_get_untagged_addr(struct kvm_vcpu *vcpu, gva_t gva, unsigned int flags)
8303 {
8304 	int lam_bit;
8305 	unsigned long cr3_bits;
8306 
8307 	if (flags & (X86EMUL_F_FETCH | X86EMUL_F_IMPLICIT | X86EMUL_F_INVLPG))
8308 		return gva;
8309 
8310 	if (!is_64_bit_mode(vcpu))
8311 		return gva;
8312 
8313 	/*
8314 	 * Bit 63 determines if the address should be treated as user address
8315 	 * or a supervisor address.
8316 	 */
8317 	if (!(gva & BIT_ULL(63))) {
8318 		cr3_bits = kvm_get_active_cr3_lam_bits(vcpu);
8319 		if (!(cr3_bits & (X86_CR3_LAM_U57 | X86_CR3_LAM_U48)))
8320 			return gva;
8321 
8322 		/* LAM_U48 is ignored if LAM_U57 is set. */
8323 		lam_bit = cr3_bits & X86_CR3_LAM_U57 ? 56 : 47;
8324 	} else {
8325 		if (!kvm_is_cr4_bit_set(vcpu, X86_CR4_LAM_SUP))
8326 			return gva;
8327 
8328 		lam_bit = kvm_is_cr4_bit_set(vcpu, X86_CR4_LA57) ? 56 : 47;
8329 	}
8330 
8331 	/*
8332 	 * Untag the address by sign-extending the lam_bit, but NOT to bit 63.
8333 	 * Bit 63 is retained from the raw virtual address so that untagging
8334 	 * doesn't change a user access to a supervisor access, and vice versa.
8335 	 */
8336 	return (sign_extend64(gva, lam_bit) & ~BIT_ULL(63)) | (gva & BIT_ULL(63));
8337 }
8338 
8339 static unsigned int vmx_handle_intel_pt_intr(void)
8340 {
8341 	struct kvm_vcpu *vcpu = kvm_get_running_vcpu();
8342 
8343 	/* '0' on failure so that the !PT case can use a RET0 static call. */
8344 	if (!vcpu || !kvm_handling_nmi_from_guest(vcpu))
8345 		return 0;
8346 
8347 	kvm_make_request(KVM_REQ_PMI, vcpu);
8348 	__set_bit(MSR_CORE_PERF_GLOBAL_OVF_CTRL_TRACE_TOPA_PMI_BIT,
8349 		  (unsigned long *)&vcpu->arch.pmu.global_status);
8350 	return 1;
8351 }
8352 
8353 static __init void vmx_setup_user_return_msrs(void)
8354 {
8355 
8356 	/*
8357 	 * Though SYSCALL is only supported in 64-bit mode on Intel CPUs, kvm
8358 	 * will emulate SYSCALL in legacy mode if the vendor string in guest
8359 	 * CPUID.0:{EBX,ECX,EDX} is "AuthenticAMD" or "AMDisbetter!" To
8360 	 * support this emulation, MSR_STAR is included in the list for i386,
8361 	 * but is never loaded into hardware.  MSR_CSTAR is also never loaded
8362 	 * into hardware and is here purely for emulation purposes.
8363 	 */
8364 	const u32 vmx_uret_msrs_list[] = {
8365 	#ifdef CONFIG_X86_64
8366 		MSR_SYSCALL_MASK, MSR_LSTAR, MSR_CSTAR,
8367 	#endif
8368 		MSR_EFER, MSR_TSC_AUX, MSR_STAR,
8369 		MSR_IA32_TSX_CTRL,
8370 	};
8371 	int i;
8372 
8373 	BUILD_BUG_ON(ARRAY_SIZE(vmx_uret_msrs_list) != MAX_NR_USER_RETURN_MSRS);
8374 
8375 	for (i = 0; i < ARRAY_SIZE(vmx_uret_msrs_list); ++i)
8376 		kvm_add_user_return_msr(vmx_uret_msrs_list[i]);
8377 }
8378 
8379 static void __init vmx_setup_me_spte_mask(void)
8380 {
8381 	u64 me_mask = 0;
8382 
8383 	/*
8384 	 * On pre-MKTME system, boot_cpu_data.x86_phys_bits equals to
8385 	 * kvm_host.maxphyaddr.  On MKTME and/or TDX capable systems,
8386 	 * boot_cpu_data.x86_phys_bits holds the actual physical address
8387 	 * w/o the KeyID bits, and kvm_host.maxphyaddr equals to
8388 	 * MAXPHYADDR reported by CPUID.  Those bits between are KeyID bits.
8389 	 */
8390 	if (boot_cpu_data.x86_phys_bits != kvm_host.maxphyaddr)
8391 		me_mask = rsvd_bits(boot_cpu_data.x86_phys_bits,
8392 				    kvm_host.maxphyaddr - 1);
8393 
8394 	/*
8395 	 * Unlike SME, host kernel doesn't support setting up any
8396 	 * MKTME KeyID on Intel platforms.  No memory encryption
8397 	 * bits should be included into the SPTE.
8398 	 */
8399 	kvm_mmu_set_me_spte_mask(0, me_mask);
8400 }
8401 
8402 __init int vmx_hardware_setup(void)
8403 {
8404 	unsigned long host_bndcfgs;
8405 	struct desc_ptr dt;
8406 	int r;
8407 
8408 	store_idt(&dt);
8409 	host_idt_base = dt.address;
8410 
8411 	vmx_setup_user_return_msrs();
8412 
8413 	if (setup_vmcs_config(&vmcs_config, &vmx_capability) < 0)
8414 		return -EIO;
8415 
8416 	if (boot_cpu_has(X86_FEATURE_NX))
8417 		kvm_enable_efer_bits(EFER_NX);
8418 
8419 	if (boot_cpu_has(X86_FEATURE_MPX)) {
8420 		rdmsrl(MSR_IA32_BNDCFGS, host_bndcfgs);
8421 		WARN_ONCE(host_bndcfgs, "BNDCFGS in host will be lost");
8422 	}
8423 
8424 	if (!cpu_has_vmx_mpx())
8425 		kvm_caps.supported_xcr0 &= ~(XFEATURE_MASK_BNDREGS |
8426 					     XFEATURE_MASK_BNDCSR);
8427 
8428 	if (!cpu_has_vmx_vpid() || !cpu_has_vmx_invvpid() ||
8429 	    !(cpu_has_vmx_invvpid_single() || cpu_has_vmx_invvpid_global()))
8430 		enable_vpid = 0;
8431 
8432 	if (!cpu_has_vmx_ept() ||
8433 	    !cpu_has_vmx_ept_4levels() ||
8434 	    !cpu_has_vmx_ept_mt_wb() ||
8435 	    !cpu_has_vmx_invept_global())
8436 		enable_ept = 0;
8437 
8438 	/* NX support is required for shadow paging. */
8439 	if (!enable_ept && !boot_cpu_has(X86_FEATURE_NX)) {
8440 		pr_err_ratelimited("NX (Execute Disable) not supported\n");
8441 		return -EOPNOTSUPP;
8442 	}
8443 
8444 	if (!cpu_has_vmx_ept_ad_bits() || !enable_ept)
8445 		enable_ept_ad_bits = 0;
8446 
8447 	if (!cpu_has_vmx_unrestricted_guest() || !enable_ept)
8448 		enable_unrestricted_guest = 0;
8449 
8450 	if (!cpu_has_vmx_flexpriority())
8451 		flexpriority_enabled = 0;
8452 
8453 	if (!cpu_has_virtual_nmis())
8454 		enable_vnmi = 0;
8455 
8456 #ifdef CONFIG_X86_SGX_KVM
8457 	if (!cpu_has_vmx_encls_vmexit())
8458 		enable_sgx = false;
8459 #endif
8460 
8461 	/*
8462 	 * set_apic_access_page_addr() is used to reload apic access
8463 	 * page upon invalidation.  No need to do anything if not
8464 	 * using the APIC_ACCESS_ADDR VMCS field.
8465 	 */
8466 	if (!flexpriority_enabled)
8467 		vt_x86_ops.set_apic_access_page_addr = NULL;
8468 
8469 	if (!cpu_has_vmx_tpr_shadow())
8470 		vt_x86_ops.update_cr8_intercept = NULL;
8471 
8472 #if IS_ENABLED(CONFIG_HYPERV)
8473 	if (ms_hyperv.nested_features & HV_X64_NESTED_GUEST_MAPPING_FLUSH
8474 	    && enable_ept) {
8475 		vt_x86_ops.flush_remote_tlbs = hv_flush_remote_tlbs;
8476 		vt_x86_ops.flush_remote_tlbs_range = hv_flush_remote_tlbs_range;
8477 	}
8478 #endif
8479 
8480 	if (!cpu_has_vmx_ple()) {
8481 		ple_gap = 0;
8482 		ple_window = 0;
8483 		ple_window_grow = 0;
8484 		ple_window_max = 0;
8485 		ple_window_shrink = 0;
8486 	}
8487 
8488 	if (!cpu_has_vmx_apicv())
8489 		enable_apicv = 0;
8490 	if (!enable_apicv)
8491 		vt_x86_ops.sync_pir_to_irr = NULL;
8492 
8493 	if (!enable_apicv || !cpu_has_vmx_ipiv())
8494 		enable_ipiv = false;
8495 
8496 	if (cpu_has_vmx_tsc_scaling())
8497 		kvm_caps.has_tsc_control = true;
8498 
8499 	kvm_caps.max_tsc_scaling_ratio = KVM_VMX_TSC_MULTIPLIER_MAX;
8500 	kvm_caps.tsc_scaling_ratio_frac_bits = 48;
8501 	kvm_caps.has_bus_lock_exit = cpu_has_vmx_bus_lock_detection();
8502 	kvm_caps.has_notify_vmexit = cpu_has_notify_vmexit();
8503 
8504 	set_bit(0, vmx_vpid_bitmap); /* 0 is reserved for host */
8505 
8506 	if (enable_ept)
8507 		kvm_mmu_set_ept_masks(enable_ept_ad_bits,
8508 				      cpu_has_vmx_ept_execute_only());
8509 
8510 	/*
8511 	 * Setup shadow_me_value/shadow_me_mask to include MKTME KeyID
8512 	 * bits to shadow_zero_check.
8513 	 */
8514 	vmx_setup_me_spte_mask();
8515 
8516 	kvm_configure_mmu(enable_ept, 0, vmx_get_max_ept_level(),
8517 			  ept_caps_to_lpage_level(vmx_capability.ept));
8518 
8519 	/*
8520 	 * Only enable PML when hardware supports PML feature, and both EPT
8521 	 * and EPT A/D bit features are enabled -- PML depends on them to work.
8522 	 */
8523 	if (!enable_ept || !enable_ept_ad_bits || !cpu_has_vmx_pml())
8524 		enable_pml = 0;
8525 
8526 	if (!enable_pml)
8527 		vt_x86_ops.cpu_dirty_log_size = 0;
8528 
8529 	if (!cpu_has_vmx_preemption_timer())
8530 		enable_preemption_timer = false;
8531 
8532 	if (enable_preemption_timer) {
8533 		u64 use_timer_freq = 5000ULL * 1000 * 1000;
8534 
8535 		cpu_preemption_timer_multi =
8536 			vmx_misc_preemption_timer_rate(vmcs_config.misc);
8537 
8538 		if (tsc_khz)
8539 			use_timer_freq = (u64)tsc_khz * 1000;
8540 		use_timer_freq >>= cpu_preemption_timer_multi;
8541 
8542 		/*
8543 		 * KVM "disables" the preemption timer by setting it to its max
8544 		 * value.  Don't use the timer if it might cause spurious exits
8545 		 * at a rate faster than 0.1 Hz (of uninterrupted guest time).
8546 		 */
8547 		if (use_timer_freq > 0xffffffffu / 10)
8548 			enable_preemption_timer = false;
8549 	}
8550 
8551 	if (!enable_preemption_timer) {
8552 		vt_x86_ops.set_hv_timer = NULL;
8553 		vt_x86_ops.cancel_hv_timer = NULL;
8554 	}
8555 
8556 	kvm_caps.supported_mce_cap |= MCG_LMCE_P;
8557 	kvm_caps.supported_mce_cap |= MCG_CMCI_P;
8558 
8559 	if (pt_mode != PT_MODE_SYSTEM && pt_mode != PT_MODE_HOST_GUEST)
8560 		return -EINVAL;
8561 	if (!enable_ept || !enable_pmu || !cpu_has_vmx_intel_pt())
8562 		pt_mode = PT_MODE_SYSTEM;
8563 	if (pt_mode == PT_MODE_HOST_GUEST)
8564 		vt_init_ops.handle_intel_pt_intr = vmx_handle_intel_pt_intr;
8565 	else
8566 		vt_init_ops.handle_intel_pt_intr = NULL;
8567 
8568 	setup_default_sgx_lepubkeyhash();
8569 
8570 	if (nested) {
8571 		nested_vmx_setup_ctls_msrs(&vmcs_config, vmx_capability.ept);
8572 
8573 		r = nested_vmx_hardware_setup(kvm_vmx_exit_handlers);
8574 		if (r)
8575 			return r;
8576 	}
8577 
8578 	vmx_set_cpu_caps();
8579 
8580 	r = alloc_kvm_area();
8581 	if (r && nested)
8582 		nested_vmx_hardware_unsetup();
8583 
8584 	kvm_set_posted_intr_wakeup_handler(pi_wakeup_handler);
8585 
8586 	return r;
8587 }
8588 
8589 static void vmx_cleanup_l1d_flush(void)
8590 {
8591 	if (vmx_l1d_flush_pages) {
8592 		free_pages((unsigned long)vmx_l1d_flush_pages, L1D_CACHE_ORDER);
8593 		vmx_l1d_flush_pages = NULL;
8594 	}
8595 	/* Restore state so sysfs ignores VMX */
8596 	l1tf_vmx_mitigation = VMENTER_L1D_FLUSH_AUTO;
8597 }
8598 
8599 static void __vmx_exit(void)
8600 {
8601 	allow_smaller_maxphyaddr = false;
8602 
8603 	vmx_cleanup_l1d_flush();
8604 }
8605 
8606 static void __exit vmx_exit(void)
8607 {
8608 	kvm_exit();
8609 	__vmx_exit();
8610 	kvm_x86_vendor_exit();
8611 
8612 }
8613 module_exit(vmx_exit);
8614 
8615 static int __init vmx_init(void)
8616 {
8617 	int r, cpu;
8618 
8619 	if (!kvm_is_vmx_supported())
8620 		return -EOPNOTSUPP;
8621 
8622 	/*
8623 	 * Note, hv_init_evmcs() touches only VMX knobs, i.e. there's nothing
8624 	 * to unwind if a later step fails.
8625 	 */
8626 	hv_init_evmcs();
8627 
8628 	r = kvm_x86_vendor_init(&vt_init_ops);
8629 	if (r)
8630 		return r;
8631 
8632 	/*
8633 	 * Must be called after common x86 init so enable_ept is properly set
8634 	 * up. Hand the parameter mitigation value in which was stored in
8635 	 * the pre module init parser. If no parameter was given, it will
8636 	 * contain 'auto' which will be turned into the default 'cond'
8637 	 * mitigation mode.
8638 	 */
8639 	r = vmx_setup_l1d_flush(vmentry_l1d_flush_param);
8640 	if (r)
8641 		goto err_l1d_flush;
8642 
8643 	for_each_possible_cpu(cpu) {
8644 		INIT_LIST_HEAD(&per_cpu(loaded_vmcss_on_cpu, cpu));
8645 
8646 		pi_init_cpu(cpu);
8647 	}
8648 
8649 	vmx_check_vmcs12_offsets();
8650 
8651 	/*
8652 	 * Shadow paging doesn't have a (further) performance penalty
8653 	 * from GUEST_MAXPHYADDR < HOST_MAXPHYADDR so enable it
8654 	 * by default
8655 	 */
8656 	if (!enable_ept)
8657 		allow_smaller_maxphyaddr = true;
8658 
8659 	/*
8660 	 * Common KVM initialization _must_ come last, after this, /dev/kvm is
8661 	 * exposed to userspace!
8662 	 */
8663 	r = kvm_init(sizeof(struct vcpu_vmx), __alignof__(struct vcpu_vmx),
8664 		     THIS_MODULE);
8665 	if (r)
8666 		goto err_kvm_init;
8667 
8668 	return 0;
8669 
8670 err_kvm_init:
8671 	__vmx_exit();
8672 err_l1d_flush:
8673 	kvm_x86_vendor_exit();
8674 	return r;
8675 }
8676 module_init(vmx_init);
8677