Linux Audio

Check our new training course

Loading...
v6.13.7
   1// SPDX-License-Identifier: GPL-2.0-only
   2/*
   3 * Kernel-based Virtual Machine driver for Linux
   4 *
   5 * AMD SVM support
   6 *
   7 * Copyright (C) 2006 Qumranet, Inc.
   8 * Copyright 2010 Red Hat, Inc. and/or its affiliates.
   9 *
  10 * Authors:
  11 *   Yaniv Kamay  <yaniv@qumranet.com>
  12 *   Avi Kivity   <avi@qumranet.com>
  13 */
  14
  15#define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
  16
  17#include <linux/kvm_types.h>
  18#include <linux/kvm_host.h>
  19#include <linux/kernel.h>
  20
  21#include <asm/msr-index.h>
  22#include <asm/debugreg.h>
  23
  24#include "kvm_emulate.h"
  25#include "trace.h"
  26#include "mmu.h"
  27#include "x86.h"
  28#include "smm.h"
  29#include "cpuid.h"
  30#include "lapic.h"
  31#include "svm.h"
  32#include "hyperv.h"
  33
  34#define CC KVM_NESTED_VMENTER_CONSISTENCY_CHECK
  35
  36static void nested_svm_inject_npf_exit(struct kvm_vcpu *vcpu,
  37				       struct x86_exception *fault)
  38{
  39	struct vcpu_svm *svm = to_svm(vcpu);
  40	struct vmcb *vmcb = svm->vmcb;
  41
  42	if (vmcb->control.exit_code != SVM_EXIT_NPF) {
  43		/*
  44		 * TODO: track the cause of the nested page fault, and
  45		 * correctly fill in the high bits of exit_info_1.
  46		 */
  47		vmcb->control.exit_code = SVM_EXIT_NPF;
  48		vmcb->control.exit_code_hi = 0;
  49		vmcb->control.exit_info_1 = (1ULL << 32);
  50		vmcb->control.exit_info_2 = fault->address;
  51	}
  52
  53	vmcb->control.exit_info_1 &= ~0xffffffffULL;
  54	vmcb->control.exit_info_1 |= fault->error_code;
  55
  56	nested_svm_vmexit(svm);
  57}
  58
  59static u64 nested_svm_get_tdp_pdptr(struct kvm_vcpu *vcpu, int index)
  60{
  61	struct vcpu_svm *svm = to_svm(vcpu);
  62	u64 cr3 = svm->nested.ctl.nested_cr3;
  63	u64 pdpte;
  64	int ret;
  65
  66	/*
  67	 * Note, nCR3 is "assumed" to be 32-byte aligned, i.e. the CPU ignores
  68	 * nCR3[4:0] when loading PDPTEs from memory.
  69	 */
  70	ret = kvm_vcpu_read_guest_page(vcpu, gpa_to_gfn(cr3), &pdpte,
  71				       (cr3 & GENMASK(11, 5)) + index * 8, 8);
  72	if (ret)
  73		return 0;
  74	return pdpte;
  75}
  76
  77static unsigned long nested_svm_get_tdp_cr3(struct kvm_vcpu *vcpu)
  78{
  79	struct vcpu_svm *svm = to_svm(vcpu);
  80
  81	return svm->nested.ctl.nested_cr3;
  82}
  83
  84static void nested_svm_init_mmu_context(struct kvm_vcpu *vcpu)
  85{
  86	struct vcpu_svm *svm = to_svm(vcpu);
 
  87
  88	WARN_ON(mmu_is_nested(vcpu));
  89
  90	vcpu->arch.mmu = &vcpu->arch.guest_mmu;
  91
  92	/*
  93	 * The NPT format depends on L1's CR4 and EFER, which is in vmcb01.  Note,
  94	 * when called via KVM_SET_NESTED_STATE, that state may _not_ match current
  95	 * vCPU state.  CR0.WP is explicitly ignored, while CR0.PG is required.
  96	 */
  97	kvm_init_shadow_npt_mmu(vcpu, X86_CR0_PG, svm->vmcb01.ptr->save.cr4,
  98				svm->vmcb01.ptr->save.efer,
  99				svm->nested.ctl.nested_cr3);
 100	vcpu->arch.mmu->get_guest_pgd     = nested_svm_get_tdp_cr3;
 101	vcpu->arch.mmu->get_pdptr         = nested_svm_get_tdp_pdptr;
 102	vcpu->arch.mmu->inject_page_fault = nested_svm_inject_npf_exit;
 
 103	vcpu->arch.walk_mmu              = &vcpu->arch.nested_mmu;
 104}
 105
 106static void nested_svm_uninit_mmu_context(struct kvm_vcpu *vcpu)
 107{
 108	vcpu->arch.mmu = &vcpu->arch.root_mmu;
 109	vcpu->arch.walk_mmu = &vcpu->arch.root_mmu;
 110}
 111
 112static bool nested_vmcb_needs_vls_intercept(struct vcpu_svm *svm)
 113{
 114	if (!guest_can_use(&svm->vcpu, X86_FEATURE_V_VMSAVE_VMLOAD))
 115		return true;
 116
 117	if (!nested_npt_enabled(svm))
 118		return true;
 119
 120	if (!(svm->nested.ctl.virt_ext & VIRTUAL_VMLOAD_VMSAVE_ENABLE_MASK))
 121		return true;
 122
 123	return false;
 124}
 125
 126void recalc_intercepts(struct vcpu_svm *svm)
 127{
 128	struct vmcb_control_area *c, *h;
 129	struct vmcb_ctrl_area_cached *g;
 130	unsigned int i;
 131
 132	vmcb_mark_dirty(svm->vmcb, VMCB_INTERCEPTS);
 133
 134	if (!is_guest_mode(&svm->vcpu))
 135		return;
 136
 137	c = &svm->vmcb->control;
 138	h = &svm->vmcb01.ptr->control;
 139	g = &svm->nested.ctl;
 140
 141	for (i = 0; i < MAX_INTERCEPT; i++)
 142		c->intercepts[i] = h->intercepts[i];
 
 
 
 
 143
 144	if (g->int_ctl & V_INTR_MASKING_MASK) {
 
 
 
 
 145		/*
 146		 * If L2 is active and V_INTR_MASKING is enabled in vmcb12,
 147		 * disable intercept of CR8 writes as L2's CR8 does not affect
 148		 * any interrupt KVM may want to inject.
 149		 *
 150		 * Similarly, disable intercept of virtual interrupts (used to
 151		 * detect interrupt windows) if the saved RFLAGS.IF is '0', as
 152		 * the effective RFLAGS.IF for L1 interrupts will never be set
 153		 * while L2 is running (L2's RFLAGS.IF doesn't affect L1 IRQs).
 154		 */
 155		vmcb_clr_intercept(c, INTERCEPT_CR8_WRITE);
 156		if (!(svm->vmcb01.ptr->save.rflags & X86_EFLAGS_IF))
 157			vmcb_clr_intercept(c, INTERCEPT_VINTR);
 158	}
 159
 160	/*
 161	 * We want to see VMMCALLs from a nested guest only when Hyper-V L2 TLB
 162	 * flush feature is enabled.
 163	 */
 164	if (!nested_svm_l2_tlb_flush_enabled(&svm->vcpu))
 165		vmcb_clr_intercept(c, INTERCEPT_VMMCALL);
 166
 167	for (i = 0; i < MAX_INTERCEPT; i++)
 168		c->intercepts[i] |= g->intercepts[i];
 169
 170	/* If SMI is not intercepted, ignore guest SMI intercept as well  */
 171	if (!intercept_smi)
 172		vmcb_clr_intercept(c, INTERCEPT_SMI);
 
 
 173
 174	if (nested_vmcb_needs_vls_intercept(svm)) {
 175		/*
 176		 * If the virtual VMLOAD/VMSAVE is not enabled for the L2,
 177		 * we must intercept these instructions to correctly
 178		 * emulate them in case L1 doesn't intercept them.
 179		 */
 180		vmcb_set_intercept(c, INTERCEPT_VMLOAD);
 181		vmcb_set_intercept(c, INTERCEPT_VMSAVE);
 182	} else {
 183		WARN_ON(!(c->virt_ext & VIRTUAL_VMLOAD_VMSAVE_ENABLE_MASK));
 184	}
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 185}
 186
 187/*
 188 * Merge L0's (KVM) and L1's (Nested VMCB) MSR permission bitmaps. The function
 189 * is optimized in that it only merges the parts where KVM MSR permission bitmap
 190 * may contain zero bits.
 191 */
 192static bool nested_svm_vmrun_msrpm(struct vcpu_svm *svm)
 193{
 194	int i;
 195
 196	/*
 197	 * MSR bitmap update can be skipped when:
 198	 * - MSR bitmap for L1 hasn't changed.
 199	 * - Nested hypervisor (L1) is attempting to launch the same L2 as
 200	 *   before.
 201	 * - Nested hypervisor (L1) is using Hyper-V emulation interface and
 202	 * tells KVM (L0) there were no changes in MSR bitmap for L2.
 203	 */
 204#ifdef CONFIG_KVM_HYPERV
 205	if (!svm->nested.force_msr_bitmap_recalc) {
 206		struct hv_vmcb_enlightenments *hve = &svm->nested.ctl.hv_enlightenments;
 207
 208		if (kvm_hv_hypercall_enabled(&svm->vcpu) &&
 209		    hve->hv_enlightenments_control.msr_bitmap &&
 210		    (svm->nested.ctl.clean & BIT(HV_VMCB_NESTED_ENLIGHTENMENTS)))
 211			goto set_msrpm_base_pa;
 212	}
 213#endif
 214
 215	if (!(vmcb12_is_intercept(&svm->nested.ctl, INTERCEPT_MSR_PROT)))
 216		return true;
 217
 218	for (i = 0; i < MSRPM_OFFSETS; i++) {
 219		u32 value, p;
 220		u64 offset;
 221
 222		if (msrpm_offsets[i] == 0xffffffff)
 223			break;
 224
 225		p      = msrpm_offsets[i];
 226
 227		/* x2apic msrs are intercepted always for the nested guest */
 228		if (is_x2apic_msrpm_offset(p))
 229			continue;
 230
 231		offset = svm->nested.ctl.msrpm_base_pa + (p * 4);
 232
 233		if (kvm_vcpu_read_guest(&svm->vcpu, offset, &value, 4))
 234			return false;
 235
 236		svm->nested.msrpm[p] = svm->msrpm[p] | value;
 237	}
 238
 239	svm->nested.force_msr_bitmap_recalc = false;
 240
 241#ifdef CONFIG_KVM_HYPERV
 242set_msrpm_base_pa:
 243#endif
 244	svm->vmcb->control.msrpm_base_pa = __sme_set(__pa(svm->nested.msrpm));
 245
 246	return true;
 247}
 248
 249/*
 250 * Bits 11:0 of bitmap address are ignored by hardware
 251 */
 252static bool nested_svm_check_bitmap_pa(struct kvm_vcpu *vcpu, u64 pa, u32 size)
 253{
 254	u64 addr = PAGE_ALIGN(pa);
 255
 256	return kvm_vcpu_is_legal_gpa(vcpu, addr) &&
 257	    kvm_vcpu_is_legal_gpa(vcpu, addr + size - 1);
 258}
 259
 260static bool __nested_vmcb_check_controls(struct kvm_vcpu *vcpu,
 261					 struct vmcb_ctrl_area_cached *control)
 262{
 263	if (CC(!vmcb12_is_intercept(control, INTERCEPT_VMRUN)))
 264		return false;
 265
 266	if (CC(control->asid == 0))
 267		return false;
 268
 269	if (CC((control->nested_ctl & SVM_NESTED_CTL_NP_ENABLE) && !npt_enabled))
 
 270		return false;
 271
 272	if (CC(!nested_svm_check_bitmap_pa(vcpu, control->msrpm_base_pa,
 273					   MSRPM_SIZE)))
 274		return false;
 275	if (CC(!nested_svm_check_bitmap_pa(vcpu, control->iopm_base_pa,
 276					   IOPM_SIZE)))
 277		return false;
 278
 279	if (CC((control->int_ctl & V_NMI_ENABLE_MASK) &&
 280	       !vmcb12_is_intercept(control, INTERCEPT_NMI))) {
 281		return false;
 282	}
 283
 284	return true;
 285}
 286
 287/* Common checks that apply to both L1 and L2 state.  */
 288static bool __nested_vmcb_check_save(struct kvm_vcpu *vcpu,
 289				     struct vmcb_save_area_cached *save)
 290{
 291	if (CC(!(save->efer & EFER_SVME)))
 
 292		return false;
 293
 294	if (CC((save->cr0 & X86_CR0_CD) == 0 && (save->cr0 & X86_CR0_NW)) ||
 295	    CC(save->cr0 & ~0xffffffffULL))
 296		return false;
 297
 298	if (CC(!kvm_dr6_valid(save->dr6)) || CC(!kvm_dr7_valid(save->dr7)))
 299		return false;
 300
 301	/*
 302	 * These checks are also performed by KVM_SET_SREGS,
 303	 * except that EFER.LMA is not checked by SVM against
 304	 * CR0.PG && EFER.LME.
 305	 */
 306	if ((save->efer & EFER_LME) && (save->cr0 & X86_CR0_PG)) {
 307		if (CC(!(save->cr4 & X86_CR4_PAE)) ||
 308		    CC(!(save->cr0 & X86_CR0_PE)) ||
 309		    CC(!kvm_vcpu_is_legal_cr3(vcpu, save->cr3)))
 
 
 
 
 
 
 
 310			return false;
 311	}
 312
 313	/* Note, SVM doesn't have any additional restrictions on CR4. */
 314	if (CC(!__kvm_is_valid_cr4(vcpu, save->cr4)))
 315		return false;
 316
 317	if (CC(!kvm_valid_efer(vcpu, save->efer)))
 318		return false;
 319
 320	return true;
 321}
 322
 323static bool nested_vmcb_check_save(struct kvm_vcpu *vcpu)
 324{
 325	struct vcpu_svm *svm = to_svm(vcpu);
 326	struct vmcb_save_area_cached *save = &svm->nested.save;
 327
 328	return __nested_vmcb_check_save(vcpu, save);
 329}
 330
 331static bool nested_vmcb_check_controls(struct kvm_vcpu *vcpu)
 332{
 333	struct vcpu_svm *svm = to_svm(vcpu);
 334	struct vmcb_ctrl_area_cached *ctl = &svm->nested.ctl;
 335
 336	return __nested_vmcb_check_controls(vcpu, ctl);
 337}
 338
 339static
 340void __nested_copy_vmcb_control_to_cache(struct kvm_vcpu *vcpu,
 341					 struct vmcb_ctrl_area_cached *to,
 342					 struct vmcb_control_area *from)
 343{
 344	unsigned int i;
 345
 346	for (i = 0; i < MAX_INTERCEPT; i++)
 347		to->intercepts[i] = from->intercepts[i];
 348
 349	to->iopm_base_pa        = from->iopm_base_pa;
 350	to->msrpm_base_pa       = from->msrpm_base_pa;
 351	to->tsc_offset          = from->tsc_offset;
 352	to->tlb_ctl             = from->tlb_ctl;
 353	to->int_ctl             = from->int_ctl;
 354	to->int_vector          = from->int_vector;
 355	to->int_state           = from->int_state;
 356	to->exit_code           = from->exit_code;
 357	to->exit_code_hi        = from->exit_code_hi;
 358	to->exit_info_1         = from->exit_info_1;
 359	to->exit_info_2         = from->exit_info_2;
 360	to->exit_int_info       = from->exit_int_info;
 361	to->exit_int_info_err   = from->exit_int_info_err;
 362	to->nested_ctl          = from->nested_ctl;
 363	to->event_inj           = from->event_inj;
 364	to->event_inj_err       = from->event_inj_err;
 365	to->next_rip            = from->next_rip;
 366	to->nested_cr3          = from->nested_cr3;
 367	to->virt_ext            = from->virt_ext;
 368	to->pause_filter_count  = from->pause_filter_count;
 369	to->pause_filter_thresh = from->pause_filter_thresh;
 370
 371	/* Copy asid here because nested_vmcb_check_controls will check it.  */
 372	to->asid           = from->asid;
 373	to->msrpm_base_pa &= ~0x0fffULL;
 374	to->iopm_base_pa  &= ~0x0fffULL;
 375
 376#ifdef CONFIG_KVM_HYPERV
 377	/* Hyper-V extensions (Enlightened VMCB) */
 378	if (kvm_hv_hypercall_enabled(vcpu)) {
 379		to->clean = from->clean;
 380		memcpy(&to->hv_enlightenments, &from->hv_enlightenments,
 381		       sizeof(to->hv_enlightenments));
 382	}
 383#endif
 384}
 385
 386void nested_copy_vmcb_control_to_cache(struct vcpu_svm *svm,
 387				       struct vmcb_control_area *control)
 388{
 389	__nested_copy_vmcb_control_to_cache(&svm->vcpu, &svm->nested.ctl, control);
 390}
 391
 392static void __nested_copy_vmcb_save_to_cache(struct vmcb_save_area_cached *to,
 393					     struct vmcb_save_area *from)
 394{
 395	/*
 396	 * Copy only fields that are validated, as we need them
 397	 * to avoid TOC/TOU races.
 398	 */
 399	to->efer = from->efer;
 400	to->cr0 = from->cr0;
 401	to->cr3 = from->cr3;
 402	to->cr4 = from->cr4;
 403
 404	to->dr6 = from->dr6;
 405	to->dr7 = from->dr7;
 406}
 407
 408void nested_copy_vmcb_save_to_cache(struct vcpu_svm *svm,
 409				    struct vmcb_save_area *save)
 410{
 411	__nested_copy_vmcb_save_to_cache(&svm->nested.save, save);
 412}
 413
 414/*
 415 * Synchronize fields that are written by the processor, so that
 416 * they can be copied back into the vmcb12.
 417 */
 418void nested_sync_control_from_vmcb02(struct vcpu_svm *svm)
 419{
 420	u32 mask;
 421	svm->nested.ctl.event_inj      = svm->vmcb->control.event_inj;
 422	svm->nested.ctl.event_inj_err  = svm->vmcb->control.event_inj_err;
 423
 424	/* Only a few fields of int_ctl are written by the processor.  */
 425	mask = V_IRQ_MASK | V_TPR_MASK;
 426	/*
 427	 * Don't sync vmcb02 V_IRQ back to vmcb12 if KVM (L0) is intercepting
 428	 * virtual interrupts in order to request an interrupt window, as KVM
 429	 * has usurped vmcb02's int_ctl.  If an interrupt window opens before
 430	 * the next VM-Exit, svm_clear_vintr() will restore vmcb12's int_ctl.
 431	 * If no window opens, V_IRQ will be correctly preserved in vmcb12's
 432	 * int_ctl (because it was never recognized while L2 was running).
 433	 */
 434	if (svm_is_intercept(svm, INTERCEPT_VINTR) &&
 435	    !test_bit(INTERCEPT_VINTR, (unsigned long *)svm->nested.ctl.intercepts))
 436		mask &= ~V_IRQ_MASK;
 437
 438	if (nested_vgif_enabled(svm))
 439		mask |= V_GIF_MASK;
 440
 441	if (nested_vnmi_enabled(svm))
 442		mask |= V_NMI_BLOCKING_MASK | V_NMI_PENDING_MASK;
 443
 444	svm->nested.ctl.int_ctl        &= ~mask;
 445	svm->nested.ctl.int_ctl        |= svm->vmcb->control.int_ctl & mask;
 446}
 447
 448/*
 449 * Transfer any event that L0 or L1 wanted to inject into L2 to
 450 * EXIT_INT_INFO.
 451 */
 452static void nested_save_pending_event_to_vmcb12(struct vcpu_svm *svm,
 453						struct vmcb *vmcb12)
 454{
 455	struct kvm_vcpu *vcpu = &svm->vcpu;
 456	u32 exit_int_info = 0;
 457	unsigned int nr;
 458
 459	if (vcpu->arch.exception.injected) {
 460		nr = vcpu->arch.exception.vector;
 461		exit_int_info = nr | SVM_EVTINJ_VALID | SVM_EVTINJ_TYPE_EXEPT;
 462
 463		if (vcpu->arch.exception.has_error_code) {
 464			exit_int_info |= SVM_EVTINJ_VALID_ERR;
 465			vmcb12->control.exit_int_info_err =
 466				vcpu->arch.exception.error_code;
 467		}
 468
 469	} else if (vcpu->arch.nmi_injected) {
 470		exit_int_info = SVM_EVTINJ_VALID | SVM_EVTINJ_TYPE_NMI;
 471
 472	} else if (vcpu->arch.interrupt.injected) {
 473		nr = vcpu->arch.interrupt.nr;
 474		exit_int_info = nr | SVM_EVTINJ_VALID;
 475
 476		if (vcpu->arch.interrupt.soft)
 477			exit_int_info |= SVM_EVTINJ_TYPE_SOFT;
 478		else
 479			exit_int_info |= SVM_EVTINJ_TYPE_INTR;
 480	}
 481
 482	vmcb12->control.exit_int_info = exit_int_info;
 483}
 484
 485static void nested_svm_transition_tlb_flush(struct kvm_vcpu *vcpu)
 486{
 487	/* Handle pending Hyper-V TLB flush requests */
 488	kvm_hv_nested_transtion_tlb_flush(vcpu, npt_enabled);
 489
 490	/*
 491	 * TODO: optimize unconditional TLB flush/MMU sync.  A partial list of
 492	 * things to fix before this can be conditional:
 493	 *
 494	 *  - Flush TLBs for both L1 and L2 remote TLB flush
 495	 *  - Honor L1's request to flush an ASID on nested VMRUN
 496	 *  - Sync nested NPT MMU on VMRUN that flushes L2's ASID[*]
 497	 *  - Don't crush a pending TLB flush in vmcb02 on nested VMRUN
 498	 *  - Flush L1's ASID on KVM_REQ_TLB_FLUSH_GUEST
 499	 *
 500	 * [*] Unlike nested EPT, SVM's ASID management can invalidate nested
 501	 *     NPT guest-physical mappings on VMRUN.
 502	 */
 503	kvm_make_request(KVM_REQ_MMU_SYNC, vcpu);
 504	kvm_make_request(KVM_REQ_TLB_FLUSH_CURRENT, vcpu);
 505}
 506
 507/*
 508 * Load guest's/host's cr3 on nested vmentry or vmexit. @nested_npt is true
 509 * if we are emulating VM-Entry into a guest with NPT enabled.
 510 */
 511static int nested_svm_load_cr3(struct kvm_vcpu *vcpu, unsigned long cr3,
 512			       bool nested_npt, bool reload_pdptrs)
 513{
 514	if (CC(!kvm_vcpu_is_legal_cr3(vcpu, cr3)))
 515		return -EINVAL;
 516
 517	if (reload_pdptrs && !nested_npt && is_pae_paging(vcpu) &&
 518	    CC(!load_pdptrs(vcpu, cr3)))
 519		return -EINVAL;
 520
 521	vcpu->arch.cr3 = cr3;
 522
 523	/* Re-initialize the MMU, e.g. to pick up CR4 MMU role changes. */
 524	kvm_init_mmu(vcpu);
 
 525
 
 
 
 
 526	if (!nested_npt)
 527		kvm_mmu_new_pgd(vcpu, cr3);
 528
 529	return 0;
 530}
 531
 532void nested_vmcb02_compute_g_pat(struct vcpu_svm *svm)
 533{
 534	if (!svm->nested.vmcb02.ptr)
 535		return;
 536
 537	/* FIXME: merge g_pat from vmcb01 and vmcb12.  */
 538	svm->nested.vmcb02.ptr->save.g_pat = svm->vmcb01.ptr->save.g_pat;
 539}
 540
 541static void nested_vmcb02_prepare_save(struct vcpu_svm *svm, struct vmcb *vmcb12)
 542{
 543	bool new_vmcb12 = false;
 544	struct vmcb *vmcb01 = svm->vmcb01.ptr;
 545	struct vmcb *vmcb02 = svm->nested.vmcb02.ptr;
 546	struct kvm_vcpu *vcpu = &svm->vcpu;
 547
 548	nested_vmcb02_compute_g_pat(svm);
 549
 550	/* Load the nested guest state */
 551	if (svm->nested.vmcb12_gpa != svm->nested.last_vmcb12_gpa) {
 552		new_vmcb12 = true;
 553		svm->nested.last_vmcb12_gpa = svm->nested.vmcb12_gpa;
 554		svm->nested.force_msr_bitmap_recalc = true;
 555	}
 556
 557	if (unlikely(new_vmcb12 || vmcb_is_dirty(vmcb12, VMCB_SEG))) {
 558		vmcb02->save.es = vmcb12->save.es;
 559		vmcb02->save.cs = vmcb12->save.cs;
 560		vmcb02->save.ss = vmcb12->save.ss;
 561		vmcb02->save.ds = vmcb12->save.ds;
 562		vmcb02->save.cpl = vmcb12->save.cpl;
 563		vmcb_mark_dirty(vmcb02, VMCB_SEG);
 564	}
 565
 566	if (unlikely(new_vmcb12 || vmcb_is_dirty(vmcb12, VMCB_DT))) {
 567		vmcb02->save.gdtr = vmcb12->save.gdtr;
 568		vmcb02->save.idtr = vmcb12->save.idtr;
 569		vmcb_mark_dirty(vmcb02, VMCB_DT);
 570	}
 571
 572	kvm_set_rflags(vcpu, vmcb12->save.rflags | X86_EFLAGS_FIXED);
 573
 574	svm_set_efer(vcpu, svm->nested.save.efer);
 575
 576	svm_set_cr0(vcpu, svm->nested.save.cr0);
 577	svm_set_cr4(vcpu, svm->nested.save.cr4);
 578
 579	svm->vcpu.arch.cr2 = vmcb12->save.cr2;
 580
 581	kvm_rax_write(vcpu, vmcb12->save.rax);
 582	kvm_rsp_write(vcpu, vmcb12->save.rsp);
 583	kvm_rip_write(vcpu, vmcb12->save.rip);
 584
 585	/* In case we don't even reach vcpu_run, the fields are not updated */
 586	vmcb02->save.rax = vmcb12->save.rax;
 587	vmcb02->save.rsp = vmcb12->save.rsp;
 588	vmcb02->save.rip = vmcb12->save.rip;
 589
 590	/* These bits will be set properly on the first execution when new_vmc12 is true */
 591	if (unlikely(new_vmcb12 || vmcb_is_dirty(vmcb12, VMCB_DR))) {
 592		vmcb02->save.dr7 = svm->nested.save.dr7 | DR7_FIXED_1;
 593		svm->vcpu.arch.dr6  = svm->nested.save.dr6 | DR6_ACTIVE_LOW;
 594		vmcb_mark_dirty(vmcb02, VMCB_DR);
 595	}
 596
 597	if (unlikely(guest_can_use(vcpu, X86_FEATURE_LBRV) &&
 598		     (svm->nested.ctl.virt_ext & LBR_CTL_ENABLE_MASK))) {
 599		/*
 600		 * Reserved bits of DEBUGCTL are ignored.  Be consistent with
 601		 * svm_set_msr's definition of reserved bits.
 602		 */
 603		svm_copy_lbrs(vmcb02, vmcb12);
 604		vmcb02->save.dbgctl &= ~DEBUGCTL_RESERVED_BITS;
 605		svm_update_lbrv(&svm->vcpu);
 606
 607	} else if (unlikely(vmcb01->control.virt_ext & LBR_CTL_ENABLE_MASK)) {
 608		svm_copy_lbrs(vmcb02, vmcb01);
 609	}
 610}
 611
 612static inline bool is_evtinj_soft(u32 evtinj)
 613{
 614	u32 type = evtinj & SVM_EVTINJ_TYPE_MASK;
 615	u8 vector = evtinj & SVM_EVTINJ_VEC_MASK;
 616
 617	if (!(evtinj & SVM_EVTINJ_VALID))
 618		return false;
 619
 620	if (type == SVM_EVTINJ_TYPE_SOFT)
 621		return true;
 622
 623	return type == SVM_EVTINJ_TYPE_EXEPT && kvm_exception_is_soft(vector);
 624}
 625
 626static bool is_evtinj_nmi(u32 evtinj)
 627{
 628	u32 type = evtinj & SVM_EVTINJ_TYPE_MASK;
 629
 630	if (!(evtinj & SVM_EVTINJ_VALID))
 631		return false;
 632
 633	return type == SVM_EVTINJ_TYPE_NMI;
 634}
 635
 636static void nested_vmcb02_prepare_control(struct vcpu_svm *svm,
 637					  unsigned long vmcb12_rip,
 638					  unsigned long vmcb12_csbase)
 639{
 640	u32 int_ctl_vmcb01_bits = V_INTR_MASKING_MASK;
 641	u32 int_ctl_vmcb12_bits = V_TPR_MASK | V_IRQ_INJECTION_BITS_MASK;
 642
 643	struct kvm_vcpu *vcpu = &svm->vcpu;
 644	struct vmcb *vmcb01 = svm->vmcb01.ptr;
 645	struct vmcb *vmcb02 = svm->nested.vmcb02.ptr;
 646	u32 pause_count12;
 647	u32 pause_thresh12;
 648
 649	nested_svm_transition_tlb_flush(vcpu);
 650
 651	/* Enter Guest-Mode */
 652	enter_guest_mode(vcpu);
 653
 654	/*
 655	 * Filled at exit: exit_code, exit_code_hi, exit_info_1, exit_info_2,
 656	 * exit_int_info, exit_int_info_err, next_rip, insn_len, insn_bytes.
 657	 */
 658
 659	if (guest_can_use(vcpu, X86_FEATURE_VGIF) &&
 660	    (svm->nested.ctl.int_ctl & V_GIF_ENABLE_MASK))
 661		int_ctl_vmcb12_bits |= (V_GIF_MASK | V_GIF_ENABLE_MASK);
 662	else
 663		int_ctl_vmcb01_bits |= (V_GIF_MASK | V_GIF_ENABLE_MASK);
 664
 665	if (vnmi) {
 666		if (vmcb01->control.int_ctl & V_NMI_PENDING_MASK) {
 667			svm->vcpu.arch.nmi_pending++;
 668			kvm_make_request(KVM_REQ_EVENT, &svm->vcpu);
 669		}
 670		if (nested_vnmi_enabled(svm))
 671			int_ctl_vmcb12_bits |= (V_NMI_PENDING_MASK |
 672						V_NMI_ENABLE_MASK |
 673						V_NMI_BLOCKING_MASK);
 674	}
 675
 676	/* Copied from vmcb01.  msrpm_base can be overwritten later.  */
 677	vmcb02->control.nested_ctl = vmcb01->control.nested_ctl;
 678	vmcb02->control.iopm_base_pa = vmcb01->control.iopm_base_pa;
 679	vmcb02->control.msrpm_base_pa = vmcb01->control.msrpm_base_pa;
 680
 681	/* Done at vmrun: asid.  */
 682
 683	/* Also overwritten later if necessary.  */
 684	vmcb02->control.tlb_ctl = TLB_CONTROL_DO_NOTHING;
 685
 686	/* nested_cr3.  */
 687	if (nested_npt_enabled(svm))
 688		nested_svm_init_mmu_context(vcpu);
 689
 690	vcpu->arch.tsc_offset = kvm_calc_nested_tsc_offset(
 691			vcpu->arch.l1_tsc_offset,
 692			svm->nested.ctl.tsc_offset,
 693			svm->tsc_ratio_msr);
 694
 695	vmcb02->control.tsc_offset = vcpu->arch.tsc_offset;
 696
 697	if (guest_can_use(vcpu, X86_FEATURE_TSCRATEMSR) &&
 698	    svm->tsc_ratio_msr != kvm_caps.default_tsc_scaling_ratio)
 699		nested_svm_update_tsc_ratio_msr(vcpu);
 700
 701	vmcb02->control.int_ctl             =
 702		(svm->nested.ctl.int_ctl & int_ctl_vmcb12_bits) |
 703		(vmcb01->control.int_ctl & int_ctl_vmcb01_bits);
 704
 705	vmcb02->control.int_vector          = svm->nested.ctl.int_vector;
 706	vmcb02->control.int_state           = svm->nested.ctl.int_state;
 707	vmcb02->control.event_inj           = svm->nested.ctl.event_inj;
 708	vmcb02->control.event_inj_err       = svm->nested.ctl.event_inj_err;
 709
 710	/*
 711	 * next_rip is consumed on VMRUN as the return address pushed on the
 712	 * stack for injected soft exceptions/interrupts.  If nrips is exposed
 713	 * to L1, take it verbatim from vmcb12.  If nrips is supported in
 714	 * hardware but not exposed to L1, stuff the actual L2 RIP to emulate
 715	 * what a nrips=0 CPU would do (L1 is responsible for advancing RIP
 716	 * prior to injecting the event).
 717	 */
 718	if (guest_can_use(vcpu, X86_FEATURE_NRIPS))
 719		vmcb02->control.next_rip    = svm->nested.ctl.next_rip;
 720	else if (boot_cpu_has(X86_FEATURE_NRIPS))
 721		vmcb02->control.next_rip    = vmcb12_rip;
 722
 723	svm->nmi_l1_to_l2 = is_evtinj_nmi(vmcb02->control.event_inj);
 724	if (is_evtinj_soft(vmcb02->control.event_inj)) {
 725		svm->soft_int_injected = true;
 726		svm->soft_int_csbase = vmcb12_csbase;
 727		svm->soft_int_old_rip = vmcb12_rip;
 728		if (guest_can_use(vcpu, X86_FEATURE_NRIPS))
 729			svm->soft_int_next_rip = svm->nested.ctl.next_rip;
 730		else
 731			svm->soft_int_next_rip = vmcb12_rip;
 732	}
 733
 734	vmcb02->control.virt_ext            = vmcb01->control.virt_ext &
 735					      LBR_CTL_ENABLE_MASK;
 736	if (guest_can_use(vcpu, X86_FEATURE_LBRV))
 737		vmcb02->control.virt_ext  |=
 738			(svm->nested.ctl.virt_ext & LBR_CTL_ENABLE_MASK);
 739
 740	if (!nested_vmcb_needs_vls_intercept(svm))
 741		vmcb02->control.virt_ext |= VIRTUAL_VMLOAD_VMSAVE_ENABLE_MASK;
 742
 743	if (guest_can_use(vcpu, X86_FEATURE_PAUSEFILTER))
 744		pause_count12 = svm->nested.ctl.pause_filter_count;
 745	else
 746		pause_count12 = 0;
 747	if (guest_can_use(vcpu, X86_FEATURE_PFTHRESHOLD))
 748		pause_thresh12 = svm->nested.ctl.pause_filter_thresh;
 749	else
 750		pause_thresh12 = 0;
 751	if (kvm_pause_in_guest(svm->vcpu.kvm)) {
 752		/* use guest values since host doesn't intercept PAUSE */
 753		vmcb02->control.pause_filter_count = pause_count12;
 754		vmcb02->control.pause_filter_thresh = pause_thresh12;
 755
 756	} else {
 757		/* start from host values otherwise */
 758		vmcb02->control.pause_filter_count = vmcb01->control.pause_filter_count;
 759		vmcb02->control.pause_filter_thresh = vmcb01->control.pause_filter_thresh;
 760
 761		/* ... but ensure filtering is disabled if so requested.  */
 762		if (vmcb12_is_intercept(&svm->nested.ctl, INTERCEPT_PAUSE)) {
 763			if (!pause_count12)
 764				vmcb02->control.pause_filter_count = 0;
 765			if (!pause_thresh12)
 766				vmcb02->control.pause_filter_thresh = 0;
 767		}
 768	}
 769
 770	/*
 771	 * Merge guest and host intercepts - must be called with vcpu in
 772	 * guest-mode to take effect.
 773	 */
 774	recalc_intercepts(svm);
 775}
 776
 777static void nested_svm_copy_common_state(struct vmcb *from_vmcb, struct vmcb *to_vmcb)
 778{
 779	/*
 780	 * Some VMCB state is shared between L1 and L2 and thus has to be
 781	 * moved at the time of nested vmrun and vmexit.
 782	 *
 783	 * VMLOAD/VMSAVE state would also belong in this category, but KVM
 784	 * always performs VMLOAD and VMSAVE from the VMCB01.
 785	 */
 786	to_vmcb->save.spec_ctrl = from_vmcb->save.spec_ctrl;
 787}
 788
 789int enter_svm_guest_mode(struct kvm_vcpu *vcpu, u64 vmcb12_gpa,
 790			 struct vmcb *vmcb12, bool from_vmrun)
 791{
 792	struct vcpu_svm *svm = to_svm(vcpu);
 793	int ret;
 794
 795	trace_kvm_nested_vmenter(svm->vmcb->save.rip,
 796				 vmcb12_gpa,
 797				 vmcb12->save.rip,
 798				 vmcb12->control.int_ctl,
 799				 vmcb12->control.event_inj,
 800				 vmcb12->control.nested_ctl,
 801				 vmcb12->control.nested_cr3,
 802				 vmcb12->save.cr3,
 803				 KVM_ISA_SVM);
 804
 805	trace_kvm_nested_intercepts(vmcb12->control.intercepts[INTERCEPT_CR] & 0xffff,
 806				    vmcb12->control.intercepts[INTERCEPT_CR] >> 16,
 807				    vmcb12->control.intercepts[INTERCEPT_EXCEPTION],
 808				    vmcb12->control.intercepts[INTERCEPT_WORD3],
 809				    vmcb12->control.intercepts[INTERCEPT_WORD4],
 810				    vmcb12->control.intercepts[INTERCEPT_WORD5]);
 811
 812
 813	svm->nested.vmcb12_gpa = vmcb12_gpa;
 814
 815	WARN_ON(svm->vmcb == svm->nested.vmcb02.ptr);
 816
 817	nested_svm_copy_common_state(svm->vmcb01.ptr, svm->nested.vmcb02.ptr);
 818
 819	svm_switch_vmcb(svm, &svm->nested.vmcb02);
 820	nested_vmcb02_prepare_control(svm, vmcb12->save.rip, vmcb12->save.cs.base);
 821	nested_vmcb02_prepare_save(svm, vmcb12);
 822
 823	ret = nested_svm_load_cr3(&svm->vcpu, svm->nested.save.cr3,
 824				  nested_npt_enabled(svm), from_vmrun);
 825	if (ret)
 826		return ret;
 827
 828	if (!from_vmrun)
 829		kvm_make_request(KVM_REQ_GET_NESTED_STATE_PAGES, vcpu);
 830
 831	svm_set_gif(svm, true);
 832
 833	if (kvm_vcpu_apicv_active(vcpu))
 834		kvm_make_request(KVM_REQ_APICV_UPDATE, vcpu);
 835
 836	nested_svm_hv_update_vm_vp_ids(vcpu);
 837
 838	return 0;
 839}
 840
 841int nested_svm_vmrun(struct kvm_vcpu *vcpu)
 842{
 843	struct vcpu_svm *svm = to_svm(vcpu);
 844	int ret;
 845	struct vmcb *vmcb12;
 
 
 846	struct kvm_host_map map;
 847	u64 vmcb12_gpa;
 848	struct vmcb *vmcb01 = svm->vmcb01.ptr;
 849
 850	if (!svm->nested.hsave_msr) {
 851		kvm_inject_gp(vcpu, 0);
 852		return 1;
 853	}
 854
 855	if (is_smm(vcpu)) {
 856		kvm_queue_exception(vcpu, UD_VECTOR);
 857		return 1;
 858	}
 859
 860	/* This fails when VP assist page is enabled but the supplied GPA is bogus */
 861	ret = kvm_hv_verify_vp_assist(vcpu);
 862	if (ret) {
 863		kvm_inject_gp(vcpu, 0);
 864		return ret;
 865	}
 866
 867	vmcb12_gpa = svm->vmcb->save.rax;
 868	ret = kvm_vcpu_map(vcpu, gpa_to_gfn(vmcb12_gpa), &map);
 869	if (ret == -EINVAL) {
 870		kvm_inject_gp(vcpu, 0);
 871		return 1;
 872	} else if (ret) {
 873		return kvm_skip_emulated_instruction(vcpu);
 874	}
 875
 876	ret = kvm_skip_emulated_instruction(vcpu);
 877
 878	vmcb12 = map.hva;
 879
 880	if (WARN_ON_ONCE(!svm->nested.initialized))
 881		return -EINVAL;
 882
 883	nested_copy_vmcb_control_to_cache(svm, &vmcb12->control);
 884	nested_copy_vmcb_save_to_cache(svm, &vmcb12->save);
 885
 886	if (!nested_vmcb_check_save(vcpu) ||
 887	    !nested_vmcb_check_controls(vcpu)) {
 888		vmcb12->control.exit_code    = SVM_EXIT_ERR;
 889		vmcb12->control.exit_code_hi = 0;
 890		vmcb12->control.exit_info_1  = 0;
 891		vmcb12->control.exit_info_2  = 0;
 892		goto out;
 893	}
 894
 895	/*
 896	 * Since vmcb01 is not in use, we can use it to store some of the L1
 897	 * state.
 898	 */
 899	vmcb01->save.efer   = vcpu->arch.efer;
 900	vmcb01->save.cr0    = kvm_read_cr0(vcpu);
 901	vmcb01->save.cr4    = vcpu->arch.cr4;
 902	vmcb01->save.rflags = kvm_get_rflags(vcpu);
 903	vmcb01->save.rip    = kvm_rip_read(vcpu);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 904
 905	if (!npt_enabled)
 906		vmcb01->save.cr3 = kvm_read_cr3(vcpu);
 907
 908	svm->nested.nested_run_pending = 1;
 909
 910	if (enter_svm_guest_mode(vcpu, vmcb12_gpa, vmcb12, true))
 911		goto out_exit_err;
 912
 913	if (nested_svm_vmrun_msrpm(svm))
 914		goto out;
 915
 916out_exit_err:
 917	svm->nested.nested_run_pending = 0;
 918	svm->nmi_l1_to_l2 = false;
 919	svm->soft_int_injected = false;
 920
 921	svm->vmcb->control.exit_code    = SVM_EXIT_ERR;
 922	svm->vmcb->control.exit_code_hi = 0;
 923	svm->vmcb->control.exit_info_1  = 0;
 924	svm->vmcb->control.exit_info_2  = 0;
 925
 926	nested_svm_vmexit(svm);
 927
 928out:
 929	kvm_vcpu_unmap(vcpu, &map);
 930
 931	return ret;
 932}
 933
 934/* Copy state save area fields which are handled by VMRUN */
 935void svm_copy_vmrun_state(struct vmcb_save_area *to_save,
 936			  struct vmcb_save_area *from_save)
 937{
 938	to_save->es = from_save->es;
 939	to_save->cs = from_save->cs;
 940	to_save->ss = from_save->ss;
 941	to_save->ds = from_save->ds;
 942	to_save->gdtr = from_save->gdtr;
 943	to_save->idtr = from_save->idtr;
 944	to_save->rflags = from_save->rflags | X86_EFLAGS_FIXED;
 945	to_save->efer = from_save->efer;
 946	to_save->cr0 = from_save->cr0;
 947	to_save->cr3 = from_save->cr3;
 948	to_save->cr4 = from_save->cr4;
 949	to_save->rax = from_save->rax;
 950	to_save->rsp = from_save->rsp;
 951	to_save->rip = from_save->rip;
 952	to_save->cpl = 0;
 953}
 954
 955void svm_copy_vmloadsave_state(struct vmcb *to_vmcb, struct vmcb *from_vmcb)
 956{
 957	to_vmcb->save.fs = from_vmcb->save.fs;
 958	to_vmcb->save.gs = from_vmcb->save.gs;
 959	to_vmcb->save.tr = from_vmcb->save.tr;
 960	to_vmcb->save.ldtr = from_vmcb->save.ldtr;
 961	to_vmcb->save.kernel_gs_base = from_vmcb->save.kernel_gs_base;
 962	to_vmcb->save.star = from_vmcb->save.star;
 963	to_vmcb->save.lstar = from_vmcb->save.lstar;
 964	to_vmcb->save.cstar = from_vmcb->save.cstar;
 965	to_vmcb->save.sfmask = from_vmcb->save.sfmask;
 966	to_vmcb->save.sysenter_cs = from_vmcb->save.sysenter_cs;
 967	to_vmcb->save.sysenter_esp = from_vmcb->save.sysenter_esp;
 968	to_vmcb->save.sysenter_eip = from_vmcb->save.sysenter_eip;
 969}
 970
 971int nested_svm_vmexit(struct vcpu_svm *svm)
 972{
 973	struct kvm_vcpu *vcpu = &svm->vcpu;
 974	struct vmcb *vmcb01 = svm->vmcb01.ptr;
 975	struct vmcb *vmcb02 = svm->nested.vmcb02.ptr;
 976	struct vmcb *vmcb12;
 977	struct kvm_host_map map;
 978	int rc;
 
 
 
 
 979
 980	rc = kvm_vcpu_map(vcpu, gpa_to_gfn(svm->nested.vmcb12_gpa), &map);
 981	if (rc) {
 982		if (rc == -EINVAL)
 983			kvm_inject_gp(vcpu, 0);
 984		return 1;
 985	}
 986
 987	vmcb12 = map.hva;
 988
 989	/* Exit Guest-Mode */
 990	leave_guest_mode(vcpu);
 991	svm->nested.vmcb12_gpa = 0;
 992	WARN_ON_ONCE(svm->nested.nested_run_pending);
 993
 994	kvm_clear_request(KVM_REQ_GET_NESTED_STATE_PAGES, vcpu);
 995
 996	/* in case we halted in L2 */
 997	svm->vcpu.arch.mp_state = KVM_MP_STATE_RUNNABLE;
 998
 999	/* Give the current vmcb to the guest */
1000
1001	vmcb12->save.es     = vmcb02->save.es;
1002	vmcb12->save.cs     = vmcb02->save.cs;
1003	vmcb12->save.ss     = vmcb02->save.ss;
1004	vmcb12->save.ds     = vmcb02->save.ds;
1005	vmcb12->save.gdtr   = vmcb02->save.gdtr;
1006	vmcb12->save.idtr   = vmcb02->save.idtr;
1007	vmcb12->save.efer   = svm->vcpu.arch.efer;
1008	vmcb12->save.cr0    = kvm_read_cr0(vcpu);
1009	vmcb12->save.cr3    = kvm_read_cr3(vcpu);
1010	vmcb12->save.cr2    = vmcb02->save.cr2;
1011	vmcb12->save.cr4    = svm->vcpu.arch.cr4;
1012	vmcb12->save.rflags = kvm_get_rflags(vcpu);
1013	vmcb12->save.rip    = kvm_rip_read(vcpu);
1014	vmcb12->save.rsp    = kvm_rsp_read(vcpu);
1015	vmcb12->save.rax    = kvm_rax_read(vcpu);
1016	vmcb12->save.dr7    = vmcb02->save.dr7;
1017	vmcb12->save.dr6    = svm->vcpu.arch.dr6;
1018	vmcb12->save.cpl    = vmcb02->save.cpl;
1019
1020	vmcb12->control.int_state         = vmcb02->control.int_state;
1021	vmcb12->control.exit_code         = vmcb02->control.exit_code;
1022	vmcb12->control.exit_code_hi      = vmcb02->control.exit_code_hi;
1023	vmcb12->control.exit_info_1       = vmcb02->control.exit_info_1;
1024	vmcb12->control.exit_info_2       = vmcb02->control.exit_info_2;
1025
1026	if (vmcb12->control.exit_code != SVM_EXIT_ERR)
1027		nested_save_pending_event_to_vmcb12(svm, vmcb12);
1028
1029	if (guest_can_use(vcpu, X86_FEATURE_NRIPS))
1030		vmcb12->control.next_rip  = vmcb02->control.next_rip;
1031
1032	vmcb12->control.int_ctl           = svm->nested.ctl.int_ctl;
1033	vmcb12->control.event_inj         = svm->nested.ctl.event_inj;
1034	vmcb12->control.event_inj_err     = svm->nested.ctl.event_inj_err;
1035
1036	if (!kvm_pause_in_guest(vcpu->kvm)) {
1037		vmcb01->control.pause_filter_count = vmcb02->control.pause_filter_count;
1038		vmcb_mark_dirty(vmcb01, VMCB_INTERCEPTS);
1039
1040	}
1041
1042	nested_svm_copy_common_state(svm->nested.vmcb02.ptr, svm->vmcb01.ptr);
1043
1044	svm_switch_vmcb(svm, &svm->vmcb01);
1045
1046	/*
1047	 * Rules for synchronizing int_ctl bits from vmcb02 to vmcb01:
1048	 *
1049	 * V_IRQ, V_IRQ_VECTOR, V_INTR_PRIO_MASK, V_IGN_TPR:  If L1 doesn't
1050	 * intercept interrupts, then KVM will use vmcb02's V_IRQ (and related
1051	 * flags) to detect interrupt windows for L1 IRQs (even if L1 uses
1052	 * virtual interrupt masking).  Raise KVM_REQ_EVENT to ensure that
1053	 * KVM re-requests an interrupt window if necessary, which implicitly
1054	 * copies this bits from vmcb02 to vmcb01.
1055	 *
1056	 * V_TPR: If L1 doesn't use virtual interrupt masking, then L1's vTPR
1057	 * is stored in vmcb02, but its value doesn't need to be copied from/to
1058	 * vmcb01 because it is copied from/to the virtual APIC's TPR register
1059	 * on each VM entry/exit.
1060	 *
1061	 * V_GIF: If nested vGIF is not used, KVM uses vmcb02's V_GIF for L1's
1062	 * V_GIF.  However, GIF is architecturally clear on each VM exit, thus
1063	 * there is no need to copy V_GIF from vmcb02 to vmcb01.
1064	 */
1065	if (!nested_exit_on_intr(svm))
1066		kvm_make_request(KVM_REQ_EVENT, &svm->vcpu);
1067
1068	if (unlikely(guest_can_use(vcpu, X86_FEATURE_LBRV) &&
1069		     (svm->nested.ctl.virt_ext & LBR_CTL_ENABLE_MASK))) {
1070		svm_copy_lbrs(vmcb12, vmcb02);
1071		svm_update_lbrv(vcpu);
1072	} else if (unlikely(vmcb01->control.virt_ext & LBR_CTL_ENABLE_MASK)) {
1073		svm_copy_lbrs(vmcb01, vmcb02);
1074		svm_update_lbrv(vcpu);
1075	}
1076
1077	if (vnmi) {
1078		if (vmcb02->control.int_ctl & V_NMI_BLOCKING_MASK)
1079			vmcb01->control.int_ctl |= V_NMI_BLOCKING_MASK;
1080		else
1081			vmcb01->control.int_ctl &= ~V_NMI_BLOCKING_MASK;
1082
1083		if (vcpu->arch.nmi_pending) {
1084			vcpu->arch.nmi_pending--;
1085			vmcb01->control.int_ctl |= V_NMI_PENDING_MASK;
1086		} else {
1087			vmcb01->control.int_ctl &= ~V_NMI_PENDING_MASK;
1088		}
1089	}
1090
1091	/*
1092	 * On vmexit the  GIF is set to false and
1093	 * no event can be injected in L1.
1094	 */
1095	svm_set_gif(svm, false);
1096	vmcb01->control.exit_int_info = 0;
1097
1098	svm->vcpu.arch.tsc_offset = svm->vcpu.arch.l1_tsc_offset;
1099	if (vmcb01->control.tsc_offset != svm->vcpu.arch.tsc_offset) {
1100		vmcb01->control.tsc_offset = svm->vcpu.arch.tsc_offset;
1101		vmcb_mark_dirty(vmcb01, VMCB_INTERCEPTS);
1102	}
1103
1104	if (kvm_caps.has_tsc_control &&
1105	    vcpu->arch.tsc_scaling_ratio != vcpu->arch.l1_tsc_scaling_ratio) {
1106		vcpu->arch.tsc_scaling_ratio = vcpu->arch.l1_tsc_scaling_ratio;
1107		svm_write_tsc_multiplier(vcpu);
1108	}
1109
1110	svm->nested.ctl.nested_cr3 = 0;
1111
1112	/*
1113	 * Restore processor state that had been saved in vmcb01
1114	 */
1115	kvm_set_rflags(vcpu, vmcb01->save.rflags);
1116	svm_set_efer(vcpu, vmcb01->save.efer);
1117	svm_set_cr0(vcpu, vmcb01->save.cr0 | X86_CR0_PE);
1118	svm_set_cr4(vcpu, vmcb01->save.cr4);
1119	kvm_rax_write(vcpu, vmcb01->save.rax);
1120	kvm_rsp_write(vcpu, vmcb01->save.rsp);
1121	kvm_rip_write(vcpu, vmcb01->save.rip);
1122
1123	svm->vcpu.arch.dr7 = DR7_FIXED_1;
1124	kvm_update_dr7(&svm->vcpu);
1125
1126	trace_kvm_nested_vmexit_inject(vmcb12->control.exit_code,
1127				       vmcb12->control.exit_info_1,
1128				       vmcb12->control.exit_info_2,
1129				       vmcb12->control.exit_int_info,
1130				       vmcb12->control.exit_int_info_err,
 
 
 
 
 
 
1131				       KVM_ISA_SVM);
1132
1133	kvm_vcpu_unmap(vcpu, &map);
1134
1135	nested_svm_transition_tlb_flush(vcpu);
1136
1137	nested_svm_uninit_mmu_context(vcpu);
1138
1139	rc = nested_svm_load_cr3(vcpu, vmcb01->save.cr3, false, true);
1140	if (rc)
1141		return 1;
1142
 
 
 
1143	/*
1144	 * Drop what we picked up for L2 via svm_complete_interrupts() so it
1145	 * doesn't end up in L1.
1146	 */
1147	svm->vcpu.arch.nmi_injected = false;
1148	kvm_clear_exception_queue(vcpu);
1149	kvm_clear_interrupt_queue(vcpu);
1150
1151	/*
1152	 * If we are here following the completion of a VMRUN that
1153	 * is being single-stepped, queue the pending #DB intercept
1154	 * right now so that it an be accounted for before we execute
1155	 * L1's next instruction.
1156	 */
1157	if (unlikely(vmcb01->save.rflags & X86_EFLAGS_TF))
1158		kvm_queue_exception(&(svm->vcpu), DB_VECTOR);
1159
1160	/*
1161	 * Un-inhibit the AVIC right away, so that other vCPUs can start
1162	 * to benefit from it right away.
1163	 */
1164	if (kvm_apicv_activated(vcpu->kvm))
1165		__kvm_vcpu_update_apicv(vcpu);
1166
1167	return 0;
1168}
1169
1170static void nested_svm_triple_fault(struct kvm_vcpu *vcpu)
1171{
1172	struct vcpu_svm *svm = to_svm(vcpu);
1173
1174	if (!vmcb12_is_intercept(&svm->nested.ctl, INTERCEPT_SHUTDOWN))
1175		return;
1176
1177	kvm_clear_request(KVM_REQ_TRIPLE_FAULT, vcpu);
1178	nested_svm_simple_vmexit(to_svm(vcpu), SVM_EXIT_SHUTDOWN);
1179}
1180
1181int svm_allocate_nested(struct vcpu_svm *svm)
1182{
1183	struct page *vmcb02_page;
1184
1185	if (svm->nested.initialized)
1186		return 0;
1187
1188	vmcb02_page = snp_safe_alloc_page();
1189	if (!vmcb02_page)
1190		return -ENOMEM;
1191	svm->nested.vmcb02.ptr = page_address(vmcb02_page);
1192	svm->nested.vmcb02.pa = __sme_set(page_to_pfn(vmcb02_page) << PAGE_SHIFT);
1193
1194	svm->nested.msrpm = svm_vcpu_alloc_msrpm();
1195	if (!svm->nested.msrpm)
1196		goto err_free_vmcb02;
1197	svm_vcpu_init_msrpm(&svm->vcpu, svm->nested.msrpm);
1198
1199	svm->nested.initialized = true;
1200	return 0;
1201
1202err_free_vmcb02:
1203	__free_page(vmcb02_page);
1204	return -ENOMEM;
1205}
1206
1207void svm_free_nested(struct vcpu_svm *svm)
1208{
1209	if (!svm->nested.initialized)
1210		return;
1211
1212	if (WARN_ON_ONCE(svm->vmcb != svm->vmcb01.ptr))
1213		svm_switch_vmcb(svm, &svm->vmcb01);
1214
1215	svm_vcpu_free_msrpm(svm->nested.msrpm);
1216	svm->nested.msrpm = NULL;
1217
1218	__free_page(virt_to_page(svm->nested.vmcb02.ptr));
1219	svm->nested.vmcb02.ptr = NULL;
1220
1221	/*
1222	 * When last_vmcb12_gpa matches the current vmcb12 gpa,
1223	 * some vmcb12 fields are not loaded if they are marked clean
1224	 * in the vmcb12, since in this case they are up to date already.
1225	 *
1226	 * When the vmcb02 is freed, this optimization becomes invalid.
1227	 */
1228	svm->nested.last_vmcb12_gpa = INVALID_GPA;
1229
1230	svm->nested.initialized = false;
1231}
1232
1233void svm_leave_nested(struct kvm_vcpu *vcpu)
1234{
1235	struct vcpu_svm *svm = to_svm(vcpu);
 
 
1236
1237	if (is_guest_mode(vcpu)) {
1238		svm->nested.nested_run_pending = 0;
1239		svm->nested.vmcb12_gpa = INVALID_GPA;
1240
1241		leave_guest_mode(vcpu);
1242
1243		svm_switch_vmcb(svm, &svm->vmcb01);
1244
1245		nested_svm_uninit_mmu_context(vcpu);
1246		vmcb_mark_all_dirty(svm->vmcb);
1247
1248		if (kvm_apicv_activated(vcpu->kvm))
1249			kvm_make_request(KVM_REQ_APICV_UPDATE, vcpu);
1250	}
1251
1252	kvm_clear_request(KVM_REQ_GET_NESTED_STATE_PAGES, vcpu);
1253}
1254
1255static int nested_svm_exit_handled_msr(struct vcpu_svm *svm)
1256{
1257	u32 offset, msr, value;
1258	int write, mask;
1259
1260	if (!(vmcb12_is_intercept(&svm->nested.ctl, INTERCEPT_MSR_PROT)))
1261		return NESTED_EXIT_HOST;
1262
1263	msr    = svm->vcpu.arch.regs[VCPU_REGS_RCX];
1264	offset = svm_msrpm_offset(msr);
1265	write  = svm->vmcb->control.exit_info_1 & 1;
1266	mask   = 1 << ((2 * (msr & 0xf)) + write);
1267
1268	if (offset == MSR_INVALID)
1269		return NESTED_EXIT_DONE;
1270
1271	/* Offset is in 32 bit units but need in 8 bit units */
1272	offset *= 4;
1273
1274	if (kvm_vcpu_read_guest(&svm->vcpu, svm->nested.ctl.msrpm_base_pa + offset, &value, 4))
1275		return NESTED_EXIT_DONE;
1276
1277	return (value & mask) ? NESTED_EXIT_DONE : NESTED_EXIT_HOST;
1278}
1279
1280static int nested_svm_intercept_ioio(struct vcpu_svm *svm)
1281{
1282	unsigned port, size, iopm_len;
1283	u16 val, mask;
1284	u8 start_bit;
1285	u64 gpa;
1286
1287	if (!(vmcb12_is_intercept(&svm->nested.ctl, INTERCEPT_IOIO_PROT)))
1288		return NESTED_EXIT_HOST;
1289
1290	port = svm->vmcb->control.exit_info_1 >> 16;
1291	size = (svm->vmcb->control.exit_info_1 & SVM_IOIO_SIZE_MASK) >>
1292		SVM_IOIO_SIZE_SHIFT;
1293	gpa  = svm->nested.ctl.iopm_base_pa + (port / 8);
1294	start_bit = port % 8;
1295	iopm_len = (start_bit + size > 8) ? 2 : 1;
1296	mask = (0xf >> (4 - size)) << start_bit;
1297	val = 0;
1298
1299	if (kvm_vcpu_read_guest(&svm->vcpu, gpa, &val, iopm_len))
1300		return NESTED_EXIT_DONE;
1301
1302	return (val & mask) ? NESTED_EXIT_DONE : NESTED_EXIT_HOST;
1303}
1304
1305static int nested_svm_intercept(struct vcpu_svm *svm)
1306{
1307	u32 exit_code = svm->vmcb->control.exit_code;
1308	int vmexit = NESTED_EXIT_HOST;
1309
1310	switch (exit_code) {
1311	case SVM_EXIT_MSR:
1312		vmexit = nested_svm_exit_handled_msr(svm);
1313		break;
1314	case SVM_EXIT_IOIO:
1315		vmexit = nested_svm_intercept_ioio(svm);
1316		break;
1317	case SVM_EXIT_READ_CR0 ... SVM_EXIT_WRITE_CR8: {
1318		if (vmcb12_is_intercept(&svm->nested.ctl, exit_code))
 
1319			vmexit = NESTED_EXIT_DONE;
1320		break;
1321	}
1322	case SVM_EXIT_READ_DR0 ... SVM_EXIT_WRITE_DR7: {
1323		if (vmcb12_is_intercept(&svm->nested.ctl, exit_code))
 
1324			vmexit = NESTED_EXIT_DONE;
1325		break;
1326	}
1327	case SVM_EXIT_EXCP_BASE ... SVM_EXIT_EXCP_BASE + 0x1f: {
1328		/*
1329		 * Host-intercepted exceptions have been checked already in
1330		 * nested_svm_exit_special.  There is nothing to do here,
1331		 * the vmexit is injected by svm_check_nested_events.
1332		 */
1333		vmexit = NESTED_EXIT_DONE;
1334		break;
1335	}
1336	case SVM_EXIT_ERR: {
1337		vmexit = NESTED_EXIT_DONE;
1338		break;
1339	}
1340	default: {
1341		if (vmcb12_is_intercept(&svm->nested.ctl, exit_code))
 
1342			vmexit = NESTED_EXIT_DONE;
1343	}
1344	}
1345
1346	return vmexit;
1347}
1348
1349int nested_svm_exit_handled(struct vcpu_svm *svm)
1350{
1351	int vmexit;
1352
1353	vmexit = nested_svm_intercept(svm);
1354
1355	if (vmexit == NESTED_EXIT_DONE)
1356		nested_svm_vmexit(svm);
1357
1358	return vmexit;
1359}
1360
1361int nested_svm_check_permissions(struct kvm_vcpu *vcpu)
1362{
1363	if (!(vcpu->arch.efer & EFER_SVME) || !is_paging(vcpu)) {
1364		kvm_queue_exception(vcpu, UD_VECTOR);
 
1365		return 1;
1366	}
1367
1368	if (to_svm(vcpu)->vmcb->save.cpl) {
1369		kvm_inject_gp(vcpu, 0);
1370		return 1;
1371	}
1372
1373	return 0;
1374}
1375
1376static bool nested_svm_is_exception_vmexit(struct kvm_vcpu *vcpu, u8 vector,
1377					   u32 error_code)
1378{
1379	struct vcpu_svm *svm = to_svm(vcpu);
1380
1381	return (svm->nested.ctl.intercepts[INTERCEPT_EXCEPTION] & BIT(vector));
1382}
1383
1384static void nested_svm_inject_exception_vmexit(struct kvm_vcpu *vcpu)
1385{
1386	struct kvm_queued_exception *ex = &vcpu->arch.exception_vmexit;
1387	struct vcpu_svm *svm = to_svm(vcpu);
1388	struct vmcb *vmcb = svm->vmcb;
1389
1390	vmcb->control.exit_code = SVM_EXIT_EXCP_BASE + ex->vector;
1391	vmcb->control.exit_code_hi = 0;
1392
1393	if (ex->has_error_code)
1394		vmcb->control.exit_info_1 = ex->error_code;
1395
1396	/*
1397	 * EXITINFO2 is undefined for all exception intercepts other
1398	 * than #PF.
1399	 */
1400	if (ex->vector == PF_VECTOR) {
1401		if (ex->has_payload)
1402			vmcb->control.exit_info_2 = ex->payload;
 
 
1403		else
1404			vmcb->control.exit_info_2 = vcpu->arch.cr2;
1405	} else if (ex->vector == DB_VECTOR) {
1406		/* See kvm_check_and_inject_events().  */
1407		kvm_deliver_exception_payload(vcpu, ex);
1408
1409		if (vcpu->arch.dr7 & DR7_GD) {
1410			vcpu->arch.dr7 &= ~DR7_GD;
1411			kvm_update_dr7(vcpu);
1412		}
1413	} else {
1414		WARN_ON(ex->has_payload);
1415	}
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1416
1417	nested_svm_vmexit(svm);
1418}
1419
1420static inline bool nested_exit_on_init(struct vcpu_svm *svm)
1421{
1422	return vmcb12_is_intercept(&svm->nested.ctl, INTERCEPT_INIT);
 
 
 
 
 
 
 
 
 
1423}
1424
 
1425static int svm_check_nested_events(struct kvm_vcpu *vcpu)
1426{
1427	struct kvm_lapic *apic = vcpu->arch.apic;
1428	struct vcpu_svm *svm = to_svm(vcpu);
1429	/*
1430	 * Only a pending nested run blocks a pending exception.  If there is a
1431	 * previously injected event, the pending exception occurred while said
1432	 * event was being delivered and thus needs to be handled.
1433	 */
1434	bool block_nested_exceptions = svm->nested.nested_run_pending;
1435	/*
1436	 * New events (not exceptions) are only recognized at instruction
1437	 * boundaries.  If an event needs reinjection, then KVM is handling a
1438	 * VM-Exit that occurred _during_ instruction execution; new events are
1439	 * blocked until the instruction completes.
1440	 */
1441	bool block_nested_events = block_nested_exceptions ||
1442				   kvm_event_needs_reinjection(vcpu);
1443
1444	if (lapic_in_kernel(vcpu) &&
1445	    test_bit(KVM_APIC_INIT, &apic->pending_events)) {
1446		if (block_nested_events)
1447			return -EBUSY;
1448		if (!nested_exit_on_init(svm))
1449			return 0;
1450		nested_svm_simple_vmexit(svm, SVM_EXIT_INIT);
1451		return 0;
1452	}
1453
1454	if (vcpu->arch.exception_vmexit.pending) {
1455		if (block_nested_exceptions)
1456                        return -EBUSY;
1457		nested_svm_inject_exception_vmexit(vcpu);
1458		return 0;
1459	}
1460
1461	if (vcpu->arch.exception.pending) {
1462		if (block_nested_exceptions)
1463			return -EBUSY;
 
 
 
1464		return 0;
1465	}
1466
1467#ifdef CONFIG_KVM_SMM
1468	if (vcpu->arch.smi_pending && !svm_smi_blocked(vcpu)) {
1469		if (block_nested_events)
1470			return -EBUSY;
1471		if (!nested_exit_on_smi(svm))
1472			return 0;
1473		nested_svm_simple_vmexit(svm, SVM_EXIT_SMI);
1474		return 0;
1475	}
1476#endif
1477
1478	if (vcpu->arch.nmi_pending && !svm_nmi_blocked(vcpu)) {
1479		if (block_nested_events)
1480			return -EBUSY;
1481		if (!nested_exit_on_nmi(svm))
1482			return 0;
1483		nested_svm_simple_vmexit(svm, SVM_EXIT_NMI);
1484		return 0;
1485	}
1486
1487	if (kvm_cpu_has_interrupt(vcpu) && !svm_interrupt_blocked(vcpu)) {
1488		if (block_nested_events)
1489			return -EBUSY;
1490		if (!nested_exit_on_intr(svm))
1491			return 0;
1492		trace_kvm_nested_intr_vmexit(svm->vmcb->save.rip);
1493		nested_svm_simple_vmexit(svm, SVM_EXIT_INTR);
1494		return 0;
1495	}
1496
1497	return 0;
1498}
1499
1500int nested_svm_exit_special(struct vcpu_svm *svm)
1501{
1502	u32 exit_code = svm->vmcb->control.exit_code;
1503	struct kvm_vcpu *vcpu = &svm->vcpu;
1504
1505	switch (exit_code) {
1506	case SVM_EXIT_INTR:
1507	case SVM_EXIT_NMI:
1508	case SVM_EXIT_NPF:
1509		return NESTED_EXIT_HOST;
1510	case SVM_EXIT_EXCP_BASE ... SVM_EXIT_EXCP_BASE + 0x1f: {
1511		u32 excp_bits = 1 << (exit_code - SVM_EXIT_EXCP_BASE);
1512
1513		if (svm->vmcb01.ptr->control.intercepts[INTERCEPT_EXCEPTION] &
1514		    excp_bits)
1515			return NESTED_EXIT_HOST;
1516		else if (exit_code == SVM_EXIT_EXCP_BASE + PF_VECTOR &&
1517			 svm->vcpu.arch.apf.host_apf_flags)
1518			/* Trap async PF even if not shadowing */
1519			return NESTED_EXIT_HOST;
1520		break;
1521	}
1522	case SVM_EXIT_VMMCALL:
1523		/* Hyper-V L2 TLB flush hypercall is handled by L0 */
1524		if (guest_hv_cpuid_has_l2_tlb_flush(vcpu) &&
1525		    nested_svm_l2_tlb_flush_enabled(vcpu) &&
1526		    kvm_hv_is_tlb_flush_hcall(vcpu))
1527			return NESTED_EXIT_HOST;
1528		break;
1529	default:
1530		break;
1531	}
1532
1533	return NESTED_EXIT_CONTINUE;
1534}
1535
1536void nested_svm_update_tsc_ratio_msr(struct kvm_vcpu *vcpu)
1537{
1538	struct vcpu_svm *svm = to_svm(vcpu);
1539
1540	vcpu->arch.tsc_scaling_ratio =
1541		kvm_calc_nested_tsc_multiplier(vcpu->arch.l1_tsc_scaling_ratio,
1542					       svm->tsc_ratio_msr);
1543	svm_write_tsc_multiplier(vcpu);
1544}
1545
1546/* Inverse operation of nested_copy_vmcb_control_to_cache(). asid is copied too. */
1547static void nested_copy_vmcb_cache_to_control(struct vmcb_control_area *dst,
1548					      struct vmcb_ctrl_area_cached *from)
1549{
1550	unsigned int i;
1551
1552	memset(dst, 0, sizeof(struct vmcb_control_area));
1553
1554	for (i = 0; i < MAX_INTERCEPT; i++)
1555		dst->intercepts[i] = from->intercepts[i];
1556
1557	dst->iopm_base_pa         = from->iopm_base_pa;
1558	dst->msrpm_base_pa        = from->msrpm_base_pa;
1559	dst->tsc_offset           = from->tsc_offset;
1560	dst->asid                 = from->asid;
1561	dst->tlb_ctl              = from->tlb_ctl;
1562	dst->int_ctl              = from->int_ctl;
1563	dst->int_vector           = from->int_vector;
1564	dst->int_state            = from->int_state;
1565	dst->exit_code            = from->exit_code;
1566	dst->exit_code_hi         = from->exit_code_hi;
1567	dst->exit_info_1          = from->exit_info_1;
1568	dst->exit_info_2          = from->exit_info_2;
1569	dst->exit_int_info        = from->exit_int_info;
1570	dst->exit_int_info_err    = from->exit_int_info_err;
1571	dst->nested_ctl           = from->nested_ctl;
1572	dst->event_inj            = from->event_inj;
1573	dst->event_inj_err        = from->event_inj_err;
1574	dst->next_rip             = from->next_rip;
1575	dst->nested_cr3           = from->nested_cr3;
1576	dst->virt_ext              = from->virt_ext;
1577	dst->pause_filter_count   = from->pause_filter_count;
1578	dst->pause_filter_thresh  = from->pause_filter_thresh;
1579	/* 'clean' and 'hv_enlightenments' are not changed by KVM */
1580}
1581
1582static int svm_get_nested_state(struct kvm_vcpu *vcpu,
1583				struct kvm_nested_state __user *user_kvm_nested_state,
1584				u32 user_data_size)
1585{
1586	struct vcpu_svm *svm;
1587	struct vmcb_control_area *ctl;
1588	unsigned long r;
1589	struct kvm_nested_state kvm_state = {
1590		.flags = 0,
1591		.format = KVM_STATE_NESTED_FORMAT_SVM,
1592		.size = sizeof(kvm_state),
1593	};
1594	struct vmcb __user *user_vmcb = (struct vmcb __user *)
1595		&user_kvm_nested_state->data.svm[0];
1596
1597	if (!vcpu)
1598		return kvm_state.size + KVM_STATE_NESTED_SVM_VMCB_SIZE;
1599
1600	svm = to_svm(vcpu);
1601
1602	if (user_data_size < kvm_state.size)
1603		goto out;
1604
1605	/* First fill in the header and copy it out.  */
1606	if (is_guest_mode(vcpu)) {
1607		kvm_state.hdr.svm.vmcb_pa = svm->nested.vmcb12_gpa;
1608		kvm_state.size += KVM_STATE_NESTED_SVM_VMCB_SIZE;
1609		kvm_state.flags |= KVM_STATE_NESTED_GUEST_MODE;
1610
1611		if (svm->nested.nested_run_pending)
1612			kvm_state.flags |= KVM_STATE_NESTED_RUN_PENDING;
1613	}
1614
1615	if (gif_set(svm))
1616		kvm_state.flags |= KVM_STATE_NESTED_GIF_SET;
1617
1618	if (copy_to_user(user_kvm_nested_state, &kvm_state, sizeof(kvm_state)))
1619		return -EFAULT;
1620
1621	if (!is_guest_mode(vcpu))
1622		goto out;
1623
1624	/*
1625	 * Copy over the full size of the VMCB rather than just the size
1626	 * of the structs.
1627	 */
1628	if (clear_user(user_vmcb, KVM_STATE_NESTED_SVM_VMCB_SIZE))
1629		return -EFAULT;
1630
1631	ctl = kzalloc(sizeof(*ctl), GFP_KERNEL);
1632	if (!ctl)
1633		return -ENOMEM;
1634
1635	nested_copy_vmcb_cache_to_control(ctl, &svm->nested.ctl);
1636	r = copy_to_user(&user_vmcb->control, ctl,
1637			 sizeof(user_vmcb->control));
1638	kfree(ctl);
1639	if (r)
1640		return -EFAULT;
1641
1642	if (copy_to_user(&user_vmcb->save, &svm->vmcb01.ptr->save,
1643			 sizeof(user_vmcb->save)))
1644		return -EFAULT;
 
1645out:
1646	return kvm_state.size;
1647}
1648
1649static int svm_set_nested_state(struct kvm_vcpu *vcpu,
1650				struct kvm_nested_state __user *user_kvm_nested_state,
1651				struct kvm_nested_state *kvm_state)
1652{
1653	struct vcpu_svm *svm = to_svm(vcpu);
 
1654	struct vmcb __user *user_vmcb = (struct vmcb __user *)
1655		&user_kvm_nested_state->data.svm[0];
1656	struct vmcb_control_area *ctl;
1657	struct vmcb_save_area *save;
1658	struct vmcb_save_area_cached save_cached;
1659	struct vmcb_ctrl_area_cached ctl_cached;
1660	unsigned long cr0;
1661	int ret;
1662
1663	BUILD_BUG_ON(sizeof(struct vmcb_control_area) + sizeof(struct vmcb_save_area) >
1664		     KVM_STATE_NESTED_SVM_VMCB_SIZE);
1665
1666	if (kvm_state->format != KVM_STATE_NESTED_FORMAT_SVM)
1667		return -EINVAL;
1668
1669	if (kvm_state->flags & ~(KVM_STATE_NESTED_GUEST_MODE |
1670				 KVM_STATE_NESTED_RUN_PENDING |
1671				 KVM_STATE_NESTED_GIF_SET))
1672		return -EINVAL;
1673
1674	/*
1675	 * If in guest mode, vcpu->arch.efer actually refers to the L2 guest's
1676	 * EFER.SVME, but EFER.SVME still has to be 1 for VMRUN to succeed.
1677	 */
1678	if (!(vcpu->arch.efer & EFER_SVME)) {
1679		/* GIF=1 and no guest mode are required if SVME=0.  */
1680		if (kvm_state->flags != KVM_STATE_NESTED_GIF_SET)
1681			return -EINVAL;
1682	}
1683
1684	/* SMM temporarily disables SVM, so we cannot be in guest mode.  */
1685	if (is_smm(vcpu) && (kvm_state->flags & KVM_STATE_NESTED_GUEST_MODE))
1686		return -EINVAL;
1687
1688	if (!(kvm_state->flags & KVM_STATE_NESTED_GUEST_MODE)) {
1689		svm_leave_nested(vcpu);
1690		svm_set_gif(svm, !!(kvm_state->flags & KVM_STATE_NESTED_GIF_SET));
1691		return 0;
1692	}
1693
1694	if (!page_address_valid(vcpu, kvm_state->hdr.svm.vmcb_pa))
1695		return -EINVAL;
1696	if (kvm_state->size < sizeof(*kvm_state) + KVM_STATE_NESTED_SVM_VMCB_SIZE)
1697		return -EINVAL;
 
 
 
 
1698
1699	ret  = -ENOMEM;
1700	ctl  = kzalloc(sizeof(*ctl),  GFP_KERNEL);
1701	save = kzalloc(sizeof(*save), GFP_KERNEL);
1702	if (!ctl || !save)
1703		goto out_free;
1704
1705	ret = -EFAULT;
1706	if (copy_from_user(ctl, &user_vmcb->control, sizeof(*ctl)))
1707		goto out_free;
1708	if (copy_from_user(save, &user_vmcb->save, sizeof(*save)))
1709		goto out_free;
1710
1711	ret = -EINVAL;
1712	__nested_copy_vmcb_control_to_cache(vcpu, &ctl_cached, ctl);
1713	if (!__nested_vmcb_check_controls(vcpu, &ctl_cached))
1714		goto out_free;
1715
1716	/*
1717	 * Processor state contains L2 state.  Check that it is
1718	 * valid for guest mode (see nested_vmcb_check_save).
1719	 */
1720	cr0 = kvm_read_cr0(vcpu);
1721        if (((cr0 & X86_CR0_CD) == 0) && (cr0 & X86_CR0_NW))
1722		goto out_free;
1723
1724	/*
1725	 * Validate host state saved from before VMRUN (see
1726	 * nested_svm_check_permissions).
 
1727	 */
1728	__nested_copy_vmcb_save_to_cache(&save_cached, save);
1729	if (!(save->cr0 & X86_CR0_PG) ||
1730	    !(save->cr0 & X86_CR0_PE) ||
1731	    (save->rflags & X86_EFLAGS_VM) ||
1732	    !__nested_vmcb_check_save(vcpu, &save_cached))
1733		goto out_free;
1734
1735
1736	/*
1737	 * All checks done, we can enter guest mode. Userspace provides
1738	 * vmcb12.control, which will be combined with L1 and stored into
1739	 * vmcb02, and the L1 save state which we store in vmcb01.
1740	 * L2 registers if needed are moved from the current VMCB to VMCB02.
1741	 */
 
 
 
 
 
 
1742
1743	if (is_guest_mode(vcpu))
1744		svm_leave_nested(vcpu);
1745	else
1746		svm->nested.vmcb02.ptr->save = svm->vmcb01.ptr->save;
1747
 
1748	svm_set_gif(svm, !!(kvm_state->flags & KVM_STATE_NESTED_GIF_SET));
1749
1750	svm->nested.nested_run_pending =
1751		!!(kvm_state->flags & KVM_STATE_NESTED_RUN_PENDING);
1752
1753	svm->nested.vmcb12_gpa = kvm_state->hdr.svm.vmcb_pa;
1754
1755	svm_copy_vmrun_state(&svm->vmcb01.ptr->save, save);
1756	nested_copy_vmcb_control_to_cache(svm, ctl);
1757
1758	svm_switch_vmcb(svm, &svm->nested.vmcb02);
1759	nested_vmcb02_prepare_control(svm, svm->vmcb->save.rip, svm->vmcb->save.cs.base);
1760
1761	/*
1762	 * While the nested guest CR3 is already checked and set by
1763	 * KVM_SET_SREGS, it was set when nested state was yet loaded,
1764	 * thus MMU might not be initialized correctly.
1765	 * Set it again to fix this.
1766	 */
1767
1768	ret = nested_svm_load_cr3(&svm->vcpu, vcpu->arch.cr3,
1769				  nested_npt_enabled(svm), false);
1770	if (WARN_ON_ONCE(ret))
1771		goto out_free;
1772
1773	svm->nested.force_msr_bitmap_recalc = true;
1774
1775	kvm_make_request(KVM_REQ_GET_NESTED_STATE_PAGES, vcpu);
1776	ret = 0;
1777out_free:
1778	kfree(save);
1779	kfree(ctl);
1780
1781	return ret;
1782}
1783
1784static bool svm_get_nested_state_pages(struct kvm_vcpu *vcpu)
1785{
1786	struct vcpu_svm *svm = to_svm(vcpu);
1787
1788	if (WARN_ON(!is_guest_mode(vcpu)))
1789		return true;
1790
1791	if (!vcpu->arch.pdptrs_from_userspace &&
1792	    !nested_npt_enabled(svm) && is_pae_paging(vcpu))
1793		/*
1794		 * Reload the guest's PDPTRs since after a migration
1795		 * the guest CR3 might be restored prior to setting the nested
1796		 * state which can lead to a load of wrong PDPTRs.
1797		 */
1798		if (CC(!load_pdptrs(vcpu, vcpu->arch.cr3)))
1799			return false;
1800
1801	if (!nested_svm_vmrun_msrpm(svm)) {
1802		vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
1803		vcpu->run->internal.suberror =
1804			KVM_INTERNAL_ERROR_EMULATION;
1805		vcpu->run->internal.ndata = 0;
1806		return false;
1807	}
1808
1809	if (kvm_hv_verify_vp_assist(vcpu))
1810		return false;
1811
1812	return true;
1813}
1814
1815struct kvm_x86_nested_ops svm_nested_ops = {
1816	.leave_nested = svm_leave_nested,
1817	.is_exception_vmexit = nested_svm_is_exception_vmexit,
1818	.check_events = svm_check_nested_events,
1819	.triple_fault = nested_svm_triple_fault,
1820	.get_nested_state_pages = svm_get_nested_state_pages,
1821	.get_state = svm_get_nested_state,
1822	.set_state = svm_set_nested_state,
1823	.hv_inject_synthetic_vmexit_post_tlb_flush = svm_hv_inject_synthetic_vmexit_post_tlb_flush,
1824};
v5.9
   1// SPDX-License-Identifier: GPL-2.0-only
   2/*
   3 * Kernel-based Virtual Machine driver for Linux
   4 *
   5 * AMD SVM support
   6 *
   7 * Copyright (C) 2006 Qumranet, Inc.
   8 * Copyright 2010 Red Hat, Inc. and/or its affiliates.
   9 *
  10 * Authors:
  11 *   Yaniv Kamay  <yaniv@qumranet.com>
  12 *   Avi Kivity   <avi@qumranet.com>
  13 */
  14
  15#define pr_fmt(fmt) "SVM: " fmt
  16
  17#include <linux/kvm_types.h>
  18#include <linux/kvm_host.h>
  19#include <linux/kernel.h>
  20
  21#include <asm/msr-index.h>
  22#include <asm/debugreg.h>
  23
  24#include "kvm_emulate.h"
  25#include "trace.h"
  26#include "mmu.h"
  27#include "x86.h"
 
  28#include "cpuid.h"
  29#include "lapic.h"
  30#include "svm.h"
 
 
 
  31
  32static void nested_svm_inject_npf_exit(struct kvm_vcpu *vcpu,
  33				       struct x86_exception *fault)
  34{
  35	struct vcpu_svm *svm = to_svm(vcpu);
 
  36
  37	if (svm->vmcb->control.exit_code != SVM_EXIT_NPF) {
  38		/*
  39		 * TODO: track the cause of the nested page fault, and
  40		 * correctly fill in the high bits of exit_info_1.
  41		 */
  42		svm->vmcb->control.exit_code = SVM_EXIT_NPF;
  43		svm->vmcb->control.exit_code_hi = 0;
  44		svm->vmcb->control.exit_info_1 = (1ULL << 32);
  45		svm->vmcb->control.exit_info_2 = fault->address;
  46	}
  47
  48	svm->vmcb->control.exit_info_1 &= ~0xffffffffULL;
  49	svm->vmcb->control.exit_info_1 |= fault->error_code;
  50
  51	nested_svm_vmexit(svm);
  52}
  53
  54static u64 nested_svm_get_tdp_pdptr(struct kvm_vcpu *vcpu, int index)
  55{
  56	struct vcpu_svm *svm = to_svm(vcpu);
  57	u64 cr3 = svm->nested.ctl.nested_cr3;
  58	u64 pdpte;
  59	int ret;
  60
  61	ret = kvm_vcpu_read_guest_page(vcpu, gpa_to_gfn(__sme_clr(cr3)), &pdpte,
  62				       offset_in_page(cr3) + index * 8, 8);
 
 
 
 
  63	if (ret)
  64		return 0;
  65	return pdpte;
  66}
  67
  68static unsigned long nested_svm_get_tdp_cr3(struct kvm_vcpu *vcpu)
  69{
  70	struct vcpu_svm *svm = to_svm(vcpu);
  71
  72	return svm->nested.ctl.nested_cr3;
  73}
  74
  75static void nested_svm_init_mmu_context(struct kvm_vcpu *vcpu)
  76{
  77	struct vcpu_svm *svm = to_svm(vcpu);
  78	struct vmcb *hsave = svm->nested.hsave;
  79
  80	WARN_ON(mmu_is_nested(vcpu));
  81
  82	vcpu->arch.mmu = &vcpu->arch.guest_mmu;
  83	kvm_init_shadow_npt_mmu(vcpu, X86_CR0_PG, hsave->save.cr4, hsave->save.efer,
 
 
 
 
 
 
 
  84				svm->nested.ctl.nested_cr3);
  85	vcpu->arch.mmu->get_guest_pgd     = nested_svm_get_tdp_cr3;
  86	vcpu->arch.mmu->get_pdptr         = nested_svm_get_tdp_pdptr;
  87	vcpu->arch.mmu->inject_page_fault = nested_svm_inject_npf_exit;
  88	reset_shadow_zero_bits_mask(vcpu, vcpu->arch.mmu);
  89	vcpu->arch.walk_mmu              = &vcpu->arch.nested_mmu;
  90}
  91
  92static void nested_svm_uninit_mmu_context(struct kvm_vcpu *vcpu)
  93{
  94	vcpu->arch.mmu = &vcpu->arch.root_mmu;
  95	vcpu->arch.walk_mmu = &vcpu->arch.root_mmu;
  96}
  97
 
 
 
 
 
 
 
 
 
 
 
 
 
 
  98void recalc_intercepts(struct vcpu_svm *svm)
  99{
 100	struct vmcb_control_area *c, *h, *g;
 
 
 101
 102	vmcb_mark_dirty(svm->vmcb, VMCB_INTERCEPTS);
 103
 104	if (!is_guest_mode(&svm->vcpu))
 105		return;
 106
 107	c = &svm->vmcb->control;
 108	h = &svm->nested.hsave->control;
 109	g = &svm->nested.ctl;
 110
 111	svm->nested.host_intercept_exceptions = h->intercept_exceptions;
 112
 113	c->intercept_cr = h->intercept_cr;
 114	c->intercept_dr = h->intercept_dr;
 115	c->intercept_exceptions = h->intercept_exceptions;
 116	c->intercept = h->intercept;
 117
 118	if (g->int_ctl & V_INTR_MASKING_MASK) {
 119		/* We only want the cr8 intercept bits of L1 */
 120		c->intercept_cr &= ~(1U << INTERCEPT_CR8_READ);
 121		c->intercept_cr &= ~(1U << INTERCEPT_CR8_WRITE);
 122
 123		/*
 124		 * Once running L2 with HF_VINTR_MASK, EFLAGS.IF does not
 125		 * affect any interrupt we may want to inject; therefore,
 126		 * interrupt window vmexits are irrelevant to L0.
 
 
 
 
 
 127		 */
 128		c->intercept &= ~(1ULL << INTERCEPT_VINTR);
 
 
 129	}
 130
 131	/* We don't want to see VMMCALLs from a nested guest */
 132	c->intercept &= ~(1ULL << INTERCEPT_VMMCALL);
 
 
 
 
 
 
 
 133
 134	c->intercept_cr |= g->intercept_cr;
 135	c->intercept_dr |= g->intercept_dr;
 136	c->intercept_exceptions |= g->intercept_exceptions;
 137	c->intercept |= g->intercept;
 138}
 139
 140static void copy_vmcb_control_area(struct vmcb_control_area *dst,
 141				   struct vmcb_control_area *from)
 142{
 143	dst->intercept_cr         = from->intercept_cr;
 144	dst->intercept_dr         = from->intercept_dr;
 145	dst->intercept_exceptions = from->intercept_exceptions;
 146	dst->intercept            = from->intercept;
 147	dst->iopm_base_pa         = from->iopm_base_pa;
 148	dst->msrpm_base_pa        = from->msrpm_base_pa;
 149	dst->tsc_offset           = from->tsc_offset;
 150	/* asid not copied, it is handled manually for svm->vmcb.  */
 151	dst->tlb_ctl              = from->tlb_ctl;
 152	dst->int_ctl              = from->int_ctl;
 153	dst->int_vector           = from->int_vector;
 154	dst->int_state            = from->int_state;
 155	dst->exit_code            = from->exit_code;
 156	dst->exit_code_hi         = from->exit_code_hi;
 157	dst->exit_info_1          = from->exit_info_1;
 158	dst->exit_info_2          = from->exit_info_2;
 159	dst->exit_int_info        = from->exit_int_info;
 160	dst->exit_int_info_err    = from->exit_int_info_err;
 161	dst->nested_ctl           = from->nested_ctl;
 162	dst->event_inj            = from->event_inj;
 163	dst->event_inj_err        = from->event_inj_err;
 164	dst->nested_cr3           = from->nested_cr3;
 165	dst->virt_ext              = from->virt_ext;
 166	dst->pause_filter_count   = from->pause_filter_count;
 167	dst->pause_filter_thresh  = from->pause_filter_thresh;
 168}
 169
 
 
 
 
 
 170static bool nested_svm_vmrun_msrpm(struct vcpu_svm *svm)
 171{
 
 
 172	/*
 173	 * This function merges the msr permission bitmaps of kvm and the
 174	 * nested vmcb. It is optimized in that it only merges the parts where
 175	 * the kvm msr permission bitmap may contain zero bits
 
 
 
 176	 */
 177	int i;
 
 
 
 
 
 
 
 
 
 178
 179	if (!(svm->nested.ctl.intercept & (1ULL << INTERCEPT_MSR_PROT)))
 180		return true;
 181
 182	for (i = 0; i < MSRPM_OFFSETS; i++) {
 183		u32 value, p;
 184		u64 offset;
 185
 186		if (msrpm_offsets[i] == 0xffffffff)
 187			break;
 188
 189		p      = msrpm_offsets[i];
 
 
 
 
 
 190		offset = svm->nested.ctl.msrpm_base_pa + (p * 4);
 191
 192		if (kvm_vcpu_read_guest(&svm->vcpu, offset, &value, 4))
 193			return false;
 194
 195		svm->nested.msrpm[p] = svm->msrpm[p] | value;
 196	}
 197
 
 
 
 
 
 198	svm->vmcb->control.msrpm_base_pa = __sme_set(__pa(svm->nested.msrpm));
 199
 200	return true;
 201}
 202
 203static bool nested_vmcb_check_controls(struct vmcb_control_area *control)
 
 
 
 
 
 
 
 
 
 
 
 
 204{
 205	if ((control->intercept & (1ULL << INTERCEPT_VMRUN)) == 0)
 206		return false;
 207
 208	if (control->asid == 0)
 209		return false;
 210
 211	if ((control->nested_ctl & SVM_NESTED_CTL_NP_ENABLE) &&
 212	    !npt_enabled)
 213		return false;
 214
 
 
 
 
 
 
 
 
 
 
 
 
 215	return true;
 216}
 217
 218static bool nested_vmcb_checks(struct vcpu_svm *svm, struct vmcb *vmcb)
 
 
 219{
 220	bool nested_vmcb_lma;
 221	if ((vmcb->save.efer & EFER_SVME) == 0)
 222		return false;
 223
 224	if (((vmcb->save.cr0 & X86_CR0_CD) == 0) &&
 225	    (vmcb->save.cr0 & X86_CR0_NW))
 226		return false;
 227
 228	if (!kvm_dr6_valid(vmcb->save.dr6) || !kvm_dr7_valid(vmcb->save.dr7))
 229		return false;
 230
 231	nested_vmcb_lma =
 232	        (vmcb->save.efer & EFER_LME) &&
 233		(vmcb->save.cr0 & X86_CR0_PG);
 234
 235	if (!nested_vmcb_lma) {
 236		if (vmcb->save.cr4 & X86_CR4_PAE) {
 237			if (vmcb->save.cr3 & MSR_CR3_LEGACY_PAE_RESERVED_MASK)
 238				return false;
 239		} else {
 240			if (vmcb->save.cr3 & MSR_CR3_LEGACY_RESERVED_MASK)
 241				return false;
 242		}
 243	} else {
 244		if (!(vmcb->save.cr4 & X86_CR4_PAE) ||
 245		    !(vmcb->save.cr0 & X86_CR0_PE) ||
 246		    (vmcb->save.cr3 & MSR_CR3_LONG_RESERVED_MASK))
 247			return false;
 248	}
 249	if (kvm_valid_cr4(&svm->vcpu, vmcb->save.cr4))
 
 
 250		return false;
 251
 252	return nested_vmcb_check_controls(&vmcb->control);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 253}
 254
 255static void load_nested_vmcb_control(struct vcpu_svm *svm,
 256				     struct vmcb_control_area *control)
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 257{
 258	copy_vmcb_control_area(&svm->nested.ctl, control);
 
 
 
 
 
 
 
 
 
 
 
 259
 260	/* Copy it here because nested_svm_check_controls will check it.  */
 261	svm->nested.ctl.asid           = control->asid;
 262	svm->nested.ctl.msrpm_base_pa &= ~0x0fffULL;
 263	svm->nested.ctl.iopm_base_pa  &= ~0x0fffULL;
 264}
 265
 266/*
 267 * Synchronize fields that are written by the processor, so that
 268 * they can be copied back into the nested_vmcb.
 269 */
 270void sync_nested_vmcb_control(struct vcpu_svm *svm)
 271{
 272	u32 mask;
 273	svm->nested.ctl.event_inj      = svm->vmcb->control.event_inj;
 274	svm->nested.ctl.event_inj_err  = svm->vmcb->control.event_inj_err;
 275
 276	/* Only a few fields of int_ctl are written by the processor.  */
 277	mask = V_IRQ_MASK | V_TPR_MASK;
 278	if (!(svm->nested.ctl.int_ctl & V_INTR_MASKING_MASK) &&
 279	    svm_is_intercept(svm, INTERCEPT_VINTR)) {
 280		/*
 281		 * In order to request an interrupt window, L0 is usurping
 282		 * svm->vmcb->control.int_ctl and possibly setting V_IRQ
 283		 * even if it was clear in L1's VMCB.  Restoring it would be
 284		 * wrong.  However, in this case V_IRQ will remain true until
 285		 * interrupt_window_interception calls svm_clear_vintr and
 286		 * restores int_ctl.  We can just leave it aside.
 287		 */
 288		mask &= ~V_IRQ_MASK;
 289	}
 
 
 
 
 
 
 290	svm->nested.ctl.int_ctl        &= ~mask;
 291	svm->nested.ctl.int_ctl        |= svm->vmcb->control.int_ctl & mask;
 292}
 293
 294/*
 295 * Transfer any event that L0 or L1 wanted to inject into L2 to
 296 * EXIT_INT_INFO.
 297 */
 298static void nested_vmcb_save_pending_event(struct vcpu_svm *svm,
 299					   struct vmcb *nested_vmcb)
 300{
 301	struct kvm_vcpu *vcpu = &svm->vcpu;
 302	u32 exit_int_info = 0;
 303	unsigned int nr;
 304
 305	if (vcpu->arch.exception.injected) {
 306		nr = vcpu->arch.exception.nr;
 307		exit_int_info = nr | SVM_EVTINJ_VALID | SVM_EVTINJ_TYPE_EXEPT;
 308
 309		if (vcpu->arch.exception.has_error_code) {
 310			exit_int_info |= SVM_EVTINJ_VALID_ERR;
 311			nested_vmcb->control.exit_int_info_err =
 312				vcpu->arch.exception.error_code;
 313		}
 314
 315	} else if (vcpu->arch.nmi_injected) {
 316		exit_int_info = SVM_EVTINJ_VALID | SVM_EVTINJ_TYPE_NMI;
 317
 318	} else if (vcpu->arch.interrupt.injected) {
 319		nr = vcpu->arch.interrupt.nr;
 320		exit_int_info = nr | SVM_EVTINJ_VALID;
 321
 322		if (vcpu->arch.interrupt.soft)
 323			exit_int_info |= SVM_EVTINJ_TYPE_SOFT;
 324		else
 325			exit_int_info |= SVM_EVTINJ_TYPE_INTR;
 326	}
 327
 328	nested_vmcb->control.exit_int_info = exit_int_info;
 329}
 330
 331static inline bool nested_npt_enabled(struct vcpu_svm *svm)
 332{
 333	return svm->nested.ctl.nested_ctl & SVM_NESTED_CTL_NP_ENABLE;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 334}
 335
 336/*
 337 * Load guest's/host's cr3 on nested vmentry or vmexit. @nested_npt is true
 338 * if we are emulating VM-Entry into a guest with NPT enabled.
 339 */
 340static int nested_svm_load_cr3(struct kvm_vcpu *vcpu, unsigned long cr3,
 341			       bool nested_npt)
 342{
 343	if (cr3 & rsvd_bits(cpuid_maxphyaddr(vcpu), 63))
 
 
 
 
 344		return -EINVAL;
 345
 346	if (!nested_npt && is_pae_paging(vcpu) &&
 347	    (cr3 != kvm_read_cr3(vcpu) || pdptrs_changed(vcpu))) {
 348		if (!load_pdptrs(vcpu, vcpu->arch.walk_mmu, cr3))
 349			return -EINVAL;
 350	}
 351
 352	/*
 353	 * TODO: optimize unconditional TLB flush/MMU sync here and in
 354	 * kvm_init_shadow_npt_mmu().
 355	 */
 356	if (!nested_npt)
 357		kvm_mmu_new_pgd(vcpu, cr3, false, false);
 358
 359	vcpu->arch.cr3 = cr3;
 360	kvm_register_mark_available(vcpu, VCPU_EXREG_CR3);
 361
 362	kvm_init_mmu(vcpu, false);
 
 
 
 363
 364	return 0;
 
 365}
 366
 367static void nested_prepare_vmcb_save(struct vcpu_svm *svm, struct vmcb *nested_vmcb)
 368{
 
 
 
 
 
 
 
 369	/* Load the nested guest state */
 370	svm->vmcb->save.es = nested_vmcb->save.es;
 371	svm->vmcb->save.cs = nested_vmcb->save.cs;
 372	svm->vmcb->save.ss = nested_vmcb->save.ss;
 373	svm->vmcb->save.ds = nested_vmcb->save.ds;
 374	svm->vmcb->save.gdtr = nested_vmcb->save.gdtr;
 375	svm->vmcb->save.idtr = nested_vmcb->save.idtr;
 376	kvm_set_rflags(&svm->vcpu, nested_vmcb->save.rflags);
 377	svm_set_efer(&svm->vcpu, nested_vmcb->save.efer);
 378	svm_set_cr0(&svm->vcpu, nested_vmcb->save.cr0);
 379	svm_set_cr4(&svm->vcpu, nested_vmcb->save.cr4);
 380	svm->vmcb->save.cr2 = svm->vcpu.arch.cr2 = nested_vmcb->save.cr2;
 381	kvm_rax_write(&svm->vcpu, nested_vmcb->save.rax);
 382	kvm_rsp_write(&svm->vcpu, nested_vmcb->save.rsp);
 383	kvm_rip_write(&svm->vcpu, nested_vmcb->save.rip);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 384
 385	/* In case we don't even reach vcpu_run, the fields are not updated */
 386	svm->vmcb->save.rax = nested_vmcb->save.rax;
 387	svm->vmcb->save.rsp = nested_vmcb->save.rsp;
 388	svm->vmcb->save.rip = nested_vmcb->save.rip;
 389	svm->vmcb->save.dr7 = nested_vmcb->save.dr7;
 390	svm->vcpu.arch.dr6  = nested_vmcb->save.dr6;
 391	svm->vmcb->save.cpl = nested_vmcb->save.cpl;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 392}
 393
 394static void nested_prepare_vmcb_control(struct vcpu_svm *svm)
 395{
 396	const u32 mask = V_INTR_MASKING_MASK | V_GIF_ENABLE_MASK | V_GIF_MASK;
 397
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 398	if (nested_npt_enabled(svm))
 399		nested_svm_init_mmu_context(&svm->vcpu);
 400
 401	svm->vmcb->control.tsc_offset = svm->vcpu.arch.tsc_offset =
 402		svm->vcpu.arch.l1_tsc_offset + svm->nested.ctl.tsc_offset;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 403
 404	svm->vmcb->control.int_ctl             =
 405		(svm->nested.ctl.int_ctl & ~mask) |
 406		(svm->nested.hsave->control.int_ctl & mask);
 407
 408	svm->vmcb->control.virt_ext            = svm->nested.ctl.virt_ext;
 409	svm->vmcb->control.int_vector          = svm->nested.ctl.int_vector;
 410	svm->vmcb->control.int_state           = svm->nested.ctl.int_state;
 411	svm->vmcb->control.event_inj           = svm->nested.ctl.event_inj;
 412	svm->vmcb->control.event_inj_err       = svm->nested.ctl.event_inj_err;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 413
 414	svm->vmcb->control.pause_filter_count  = svm->nested.ctl.pause_filter_count;
 415	svm->vmcb->control.pause_filter_thresh = svm->nested.ctl.pause_filter_thresh;
 
 
 
 
 
 
 
 
 
 
 416
 417	/* Enter Guest-Mode */
 418	enter_guest_mode(&svm->vcpu);
 
 
 
 
 
 
 
 
 
 
 
 419
 420	/*
 421	 * Merge guest and host intercepts - must be called  with vcpu in
 422	 * guest-mode to take affect here
 423	 */
 424	recalc_intercepts(svm);
 
 425
 426	vmcb_mark_all_dirty(svm->vmcb);
 
 
 
 
 
 
 
 
 
 427}
 428
 429int enter_svm_guest_mode(struct vcpu_svm *svm, u64 vmcb_gpa,
 430			  struct vmcb *nested_vmcb)
 431{
 
 432	int ret;
 433
 434	svm->nested.vmcb = vmcb_gpa;
 435	load_nested_vmcb_control(svm, &nested_vmcb->control);
 436	nested_prepare_vmcb_save(svm, nested_vmcb);
 437	nested_prepare_vmcb_control(svm);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 438
 439	ret = nested_svm_load_cr3(&svm->vcpu, nested_vmcb->save.cr3,
 440				  nested_npt_enabled(svm));
 441	if (ret)
 442		return ret;
 443
 
 
 
 444	svm_set_gif(svm, true);
 445
 
 
 
 
 
 446	return 0;
 447}
 448
 449int nested_svm_vmrun(struct vcpu_svm *svm)
 450{
 
 451	int ret;
 452	struct vmcb *nested_vmcb;
 453	struct vmcb *hsave = svm->nested.hsave;
 454	struct vmcb *vmcb = svm->vmcb;
 455	struct kvm_host_map map;
 456	u64 vmcb_gpa;
 
 
 
 
 
 
 457
 458	if (is_smm(&svm->vcpu)) {
 459		kvm_queue_exception(&svm->vcpu, UD_VECTOR);
 460		return 1;
 461	}
 462
 463	vmcb_gpa = svm->vmcb->save.rax;
 464	ret = kvm_vcpu_map(&svm->vcpu, gpa_to_gfn(vmcb_gpa), &map);
 
 
 
 
 
 
 
 465	if (ret == -EINVAL) {
 466		kvm_inject_gp(&svm->vcpu, 0);
 467		return 1;
 468	} else if (ret) {
 469		return kvm_skip_emulated_instruction(&svm->vcpu);
 470	}
 471
 472	ret = kvm_skip_emulated_instruction(&svm->vcpu);
 
 
 
 
 
 473
 474	nested_vmcb = map.hva;
 
 475
 476	if (!nested_vmcb_checks(svm, nested_vmcb)) {
 477		nested_vmcb->control.exit_code    = SVM_EXIT_ERR;
 478		nested_vmcb->control.exit_code_hi = 0;
 479		nested_vmcb->control.exit_info_1  = 0;
 480		nested_vmcb->control.exit_info_2  = 0;
 
 481		goto out;
 482	}
 483
 484	trace_kvm_nested_vmrun(svm->vmcb->save.rip, vmcb_gpa,
 485			       nested_vmcb->save.rip,
 486			       nested_vmcb->control.int_ctl,
 487			       nested_vmcb->control.event_inj,
 488			       nested_vmcb->control.nested_ctl);
 489
 490	trace_kvm_nested_intercepts(nested_vmcb->control.intercept_cr & 0xffff,
 491				    nested_vmcb->control.intercept_cr >> 16,
 492				    nested_vmcb->control.intercept_exceptions,
 493				    nested_vmcb->control.intercept);
 494
 495	/* Clear internal status */
 496	kvm_clear_exception_queue(&svm->vcpu);
 497	kvm_clear_interrupt_queue(&svm->vcpu);
 498
 499	/*
 500	 * Save the old vmcb, so we don't need to pick what we save, but can
 501	 * restore everything when a VMEXIT occurs
 502	 */
 503	hsave->save.es     = vmcb->save.es;
 504	hsave->save.cs     = vmcb->save.cs;
 505	hsave->save.ss     = vmcb->save.ss;
 506	hsave->save.ds     = vmcb->save.ds;
 507	hsave->save.gdtr   = vmcb->save.gdtr;
 508	hsave->save.idtr   = vmcb->save.idtr;
 509	hsave->save.efer   = svm->vcpu.arch.efer;
 510	hsave->save.cr0    = kvm_read_cr0(&svm->vcpu);
 511	hsave->save.cr4    = svm->vcpu.arch.cr4;
 512	hsave->save.rflags = kvm_get_rflags(&svm->vcpu);
 513	hsave->save.rip    = kvm_rip_read(&svm->vcpu);
 514	hsave->save.rsp    = vmcb->save.rsp;
 515	hsave->save.rax    = vmcb->save.rax;
 516	if (npt_enabled)
 517		hsave->save.cr3    = vmcb->save.cr3;
 518	else
 519		hsave->save.cr3    = kvm_read_cr3(&svm->vcpu);
 520
 521	copy_vmcb_control_area(&hsave->control, &vmcb->control);
 
 522
 523	svm->nested.nested_run_pending = 1;
 524
 525	if (enter_svm_guest_mode(svm, vmcb_gpa, nested_vmcb))
 526		goto out_exit_err;
 527
 528	if (nested_svm_vmrun_msrpm(svm))
 529		goto out;
 530
 531out_exit_err:
 532	svm->nested.nested_run_pending = 0;
 
 
 533
 534	svm->vmcb->control.exit_code    = SVM_EXIT_ERR;
 535	svm->vmcb->control.exit_code_hi = 0;
 536	svm->vmcb->control.exit_info_1  = 0;
 537	svm->vmcb->control.exit_info_2  = 0;
 538
 539	nested_svm_vmexit(svm);
 540
 541out:
 542	kvm_vcpu_unmap(&svm->vcpu, &map, true);
 543
 544	return ret;
 545}
 546
 547void nested_svm_vmloadsave(struct vmcb *from_vmcb, struct vmcb *to_vmcb)
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 548{
 549	to_vmcb->save.fs = from_vmcb->save.fs;
 550	to_vmcb->save.gs = from_vmcb->save.gs;
 551	to_vmcb->save.tr = from_vmcb->save.tr;
 552	to_vmcb->save.ldtr = from_vmcb->save.ldtr;
 553	to_vmcb->save.kernel_gs_base = from_vmcb->save.kernel_gs_base;
 554	to_vmcb->save.star = from_vmcb->save.star;
 555	to_vmcb->save.lstar = from_vmcb->save.lstar;
 556	to_vmcb->save.cstar = from_vmcb->save.cstar;
 557	to_vmcb->save.sfmask = from_vmcb->save.sfmask;
 558	to_vmcb->save.sysenter_cs = from_vmcb->save.sysenter_cs;
 559	to_vmcb->save.sysenter_esp = from_vmcb->save.sysenter_esp;
 560	to_vmcb->save.sysenter_eip = from_vmcb->save.sysenter_eip;
 561}
 562
 563int nested_svm_vmexit(struct vcpu_svm *svm)
 564{
 
 
 
 
 
 565	int rc;
 566	struct vmcb *nested_vmcb;
 567	struct vmcb *hsave = svm->nested.hsave;
 568	struct vmcb *vmcb = svm->vmcb;
 569	struct kvm_host_map map;
 570
 571	rc = kvm_vcpu_map(&svm->vcpu, gpa_to_gfn(svm->nested.vmcb), &map);
 572	if (rc) {
 573		if (rc == -EINVAL)
 574			kvm_inject_gp(&svm->vcpu, 0);
 575		return 1;
 576	}
 577
 578	nested_vmcb = map.hva;
 579
 580	/* Exit Guest-Mode */
 581	leave_guest_mode(&svm->vcpu);
 582	svm->nested.vmcb = 0;
 583	WARN_ON_ONCE(svm->nested.nested_run_pending);
 584
 
 
 585	/* in case we halted in L2 */
 586	svm->vcpu.arch.mp_state = KVM_MP_STATE_RUNNABLE;
 587
 588	/* Give the current vmcb to the guest */
 589
 590	nested_vmcb->save.es     = vmcb->save.es;
 591	nested_vmcb->save.cs     = vmcb->save.cs;
 592	nested_vmcb->save.ss     = vmcb->save.ss;
 593	nested_vmcb->save.ds     = vmcb->save.ds;
 594	nested_vmcb->save.gdtr   = vmcb->save.gdtr;
 595	nested_vmcb->save.idtr   = vmcb->save.idtr;
 596	nested_vmcb->save.efer   = svm->vcpu.arch.efer;
 597	nested_vmcb->save.cr0    = kvm_read_cr0(&svm->vcpu);
 598	nested_vmcb->save.cr3    = kvm_read_cr3(&svm->vcpu);
 599	nested_vmcb->save.cr2    = vmcb->save.cr2;
 600	nested_vmcb->save.cr4    = svm->vcpu.arch.cr4;
 601	nested_vmcb->save.rflags = kvm_get_rflags(&svm->vcpu);
 602	nested_vmcb->save.rip    = kvm_rip_read(&svm->vcpu);
 603	nested_vmcb->save.rsp    = kvm_rsp_read(&svm->vcpu);
 604	nested_vmcb->save.rax    = kvm_rax_read(&svm->vcpu);
 605	nested_vmcb->save.dr7    = vmcb->save.dr7;
 606	nested_vmcb->save.dr6    = svm->vcpu.arch.dr6;
 607	nested_vmcb->save.cpl    = vmcb->save.cpl;
 608
 609	nested_vmcb->control.int_state         = vmcb->control.int_state;
 610	nested_vmcb->control.exit_code         = vmcb->control.exit_code;
 611	nested_vmcb->control.exit_code_hi      = vmcb->control.exit_code_hi;
 612	nested_vmcb->control.exit_info_1       = vmcb->control.exit_info_1;
 613	nested_vmcb->control.exit_info_2       = vmcb->control.exit_info_2;
 614
 615	if (nested_vmcb->control.exit_code != SVM_EXIT_ERR)
 616		nested_vmcb_save_pending_event(svm, nested_vmcb);
 617
 618	if (svm->nrips_enabled)
 619		nested_vmcb->control.next_rip  = vmcb->control.next_rip;
 620
 621	nested_vmcb->control.int_ctl           = svm->nested.ctl.int_ctl;
 622	nested_vmcb->control.tlb_ctl           = svm->nested.ctl.tlb_ctl;
 623	nested_vmcb->control.event_inj         = svm->nested.ctl.event_inj;
 624	nested_vmcb->control.event_inj_err     = svm->nested.ctl.event_inj_err;
 625
 626	nested_vmcb->control.pause_filter_count =
 627		svm->vmcb->control.pause_filter_count;
 628	nested_vmcb->control.pause_filter_thresh =
 629		svm->vmcb->control.pause_filter_thresh;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 630
 631	/* Restore the original control entries */
 632	copy_vmcb_control_area(&vmcb->control, &hsave->control);
 
 
 
 
 
 633
 634	/* On vmexit the  GIF is set to false */
 
 
 
 635	svm_set_gif(svm, false);
 
 636
 637	svm->vmcb->control.tsc_offset = svm->vcpu.arch.tsc_offset =
 638		svm->vcpu.arch.l1_tsc_offset;
 
 
 
 
 
 
 
 
 
 639
 640	svm->nested.ctl.nested_cr3 = 0;
 641
 642	/* Restore selected save entries */
 643	svm->vmcb->save.es = hsave->save.es;
 644	svm->vmcb->save.cs = hsave->save.cs;
 645	svm->vmcb->save.ss = hsave->save.ss;
 646	svm->vmcb->save.ds = hsave->save.ds;
 647	svm->vmcb->save.gdtr = hsave->save.gdtr;
 648	svm->vmcb->save.idtr = hsave->save.idtr;
 649	kvm_set_rflags(&svm->vcpu, hsave->save.rflags);
 650	svm_set_efer(&svm->vcpu, hsave->save.efer);
 651	svm_set_cr0(&svm->vcpu, hsave->save.cr0 | X86_CR0_PE);
 652	svm_set_cr4(&svm->vcpu, hsave->save.cr4);
 653	kvm_rax_write(&svm->vcpu, hsave->save.rax);
 654	kvm_rsp_write(&svm->vcpu, hsave->save.rsp);
 655	kvm_rip_write(&svm->vcpu, hsave->save.rip);
 656	svm->vmcb->save.dr7 = 0;
 657	svm->vmcb->save.cpl = 0;
 658	svm->vmcb->control.exit_int_info = 0;
 659
 660	vmcb_mark_all_dirty(svm->vmcb);
 661
 662	trace_kvm_nested_vmexit_inject(nested_vmcb->control.exit_code,
 663				       nested_vmcb->control.exit_info_1,
 664				       nested_vmcb->control.exit_info_2,
 665				       nested_vmcb->control.exit_int_info,
 666				       nested_vmcb->control.exit_int_info_err,
 667				       KVM_ISA_SVM);
 668
 669	kvm_vcpu_unmap(&svm->vcpu, &map, true);
 670
 671	nested_svm_uninit_mmu_context(&svm->vcpu);
 672
 673	rc = nested_svm_load_cr3(&svm->vcpu, hsave->save.cr3, false);
 
 
 674	if (rc)
 675		return 1;
 676
 677	if (npt_enabled)
 678		svm->vmcb->save.cr3 = hsave->save.cr3;
 679
 680	/*
 681	 * Drop what we picked up for L2 via svm_complete_interrupts() so it
 682	 * doesn't end up in L1.
 683	 */
 684	svm->vcpu.arch.nmi_injected = false;
 685	kvm_clear_exception_queue(&svm->vcpu);
 686	kvm_clear_interrupt_queue(&svm->vcpu);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 687
 688	return 0;
 689}
 690
 691/*
 692 * Forcibly leave nested mode in order to be able to reset the VCPU later on.
 693 */
 694void svm_leave_nested(struct vcpu_svm *svm)
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 695{
 696	if (is_guest_mode(&svm->vcpu)) {
 697		struct vmcb *hsave = svm->nested.hsave;
 698		struct vmcb *vmcb = svm->vmcb;
 699
 
 700		svm->nested.nested_run_pending = 0;
 701		leave_guest_mode(&svm->vcpu);
 702		copy_vmcb_control_area(&vmcb->control, &hsave->control);
 703		nested_svm_uninit_mmu_context(&svm->vcpu);
 
 
 
 
 
 
 
 
 704	}
 
 
 705}
 706
 707static int nested_svm_exit_handled_msr(struct vcpu_svm *svm)
 708{
 709	u32 offset, msr, value;
 710	int write, mask;
 711
 712	if (!(svm->nested.ctl.intercept & (1ULL << INTERCEPT_MSR_PROT)))
 713		return NESTED_EXIT_HOST;
 714
 715	msr    = svm->vcpu.arch.regs[VCPU_REGS_RCX];
 716	offset = svm_msrpm_offset(msr);
 717	write  = svm->vmcb->control.exit_info_1 & 1;
 718	mask   = 1 << ((2 * (msr & 0xf)) + write);
 719
 720	if (offset == MSR_INVALID)
 721		return NESTED_EXIT_DONE;
 722
 723	/* Offset is in 32 bit units but need in 8 bit units */
 724	offset *= 4;
 725
 726	if (kvm_vcpu_read_guest(&svm->vcpu, svm->nested.ctl.msrpm_base_pa + offset, &value, 4))
 727		return NESTED_EXIT_DONE;
 728
 729	return (value & mask) ? NESTED_EXIT_DONE : NESTED_EXIT_HOST;
 730}
 731
 732static int nested_svm_intercept_ioio(struct vcpu_svm *svm)
 733{
 734	unsigned port, size, iopm_len;
 735	u16 val, mask;
 736	u8 start_bit;
 737	u64 gpa;
 738
 739	if (!(svm->nested.ctl.intercept & (1ULL << INTERCEPT_IOIO_PROT)))
 740		return NESTED_EXIT_HOST;
 741
 742	port = svm->vmcb->control.exit_info_1 >> 16;
 743	size = (svm->vmcb->control.exit_info_1 & SVM_IOIO_SIZE_MASK) >>
 744		SVM_IOIO_SIZE_SHIFT;
 745	gpa  = svm->nested.ctl.iopm_base_pa + (port / 8);
 746	start_bit = port % 8;
 747	iopm_len = (start_bit + size > 8) ? 2 : 1;
 748	mask = (0xf >> (4 - size)) << start_bit;
 749	val = 0;
 750
 751	if (kvm_vcpu_read_guest(&svm->vcpu, gpa, &val, iopm_len))
 752		return NESTED_EXIT_DONE;
 753
 754	return (val & mask) ? NESTED_EXIT_DONE : NESTED_EXIT_HOST;
 755}
 756
 757static int nested_svm_intercept(struct vcpu_svm *svm)
 758{
 759	u32 exit_code = svm->vmcb->control.exit_code;
 760	int vmexit = NESTED_EXIT_HOST;
 761
 762	switch (exit_code) {
 763	case SVM_EXIT_MSR:
 764		vmexit = nested_svm_exit_handled_msr(svm);
 765		break;
 766	case SVM_EXIT_IOIO:
 767		vmexit = nested_svm_intercept_ioio(svm);
 768		break;
 769	case SVM_EXIT_READ_CR0 ... SVM_EXIT_WRITE_CR8: {
 770		u32 bit = 1U << (exit_code - SVM_EXIT_READ_CR0);
 771		if (svm->nested.ctl.intercept_cr & bit)
 772			vmexit = NESTED_EXIT_DONE;
 773		break;
 774	}
 775	case SVM_EXIT_READ_DR0 ... SVM_EXIT_WRITE_DR7: {
 776		u32 bit = 1U << (exit_code - SVM_EXIT_READ_DR0);
 777		if (svm->nested.ctl.intercept_dr & bit)
 778			vmexit = NESTED_EXIT_DONE;
 779		break;
 780	}
 781	case SVM_EXIT_EXCP_BASE ... SVM_EXIT_EXCP_BASE + 0x1f: {
 782		/*
 783		 * Host-intercepted exceptions have been checked already in
 784		 * nested_svm_exit_special.  There is nothing to do here,
 785		 * the vmexit is injected by svm_check_nested_events.
 786		 */
 787		vmexit = NESTED_EXIT_DONE;
 788		break;
 789	}
 790	case SVM_EXIT_ERR: {
 791		vmexit = NESTED_EXIT_DONE;
 792		break;
 793	}
 794	default: {
 795		u64 exit_bits = 1ULL << (exit_code - SVM_EXIT_INTR);
 796		if (svm->nested.ctl.intercept & exit_bits)
 797			vmexit = NESTED_EXIT_DONE;
 798	}
 799	}
 800
 801	return vmexit;
 802}
 803
 804int nested_svm_exit_handled(struct vcpu_svm *svm)
 805{
 806	int vmexit;
 807
 808	vmexit = nested_svm_intercept(svm);
 809
 810	if (vmexit == NESTED_EXIT_DONE)
 811		nested_svm_vmexit(svm);
 812
 813	return vmexit;
 814}
 815
 816int nested_svm_check_permissions(struct vcpu_svm *svm)
 817{
 818	if (!(svm->vcpu.arch.efer & EFER_SVME) ||
 819	    !is_paging(&svm->vcpu)) {
 820		kvm_queue_exception(&svm->vcpu, UD_VECTOR);
 821		return 1;
 822	}
 823
 824	if (svm->vmcb->save.cpl) {
 825		kvm_inject_gp(&svm->vcpu, 0);
 826		return 1;
 827	}
 828
 829	return 0;
 830}
 831
 832static bool nested_exit_on_exception(struct vcpu_svm *svm)
 
 833{
 834	unsigned int nr = svm->vcpu.arch.exception.nr;
 835
 836	return (svm->nested.ctl.intercept_exceptions & (1 << nr));
 837}
 838
 839static void nested_svm_inject_exception_vmexit(struct vcpu_svm *svm)
 840{
 841	unsigned int nr = svm->vcpu.arch.exception.nr;
 
 
 842
 843	svm->vmcb->control.exit_code = SVM_EXIT_EXCP_BASE + nr;
 844	svm->vmcb->control.exit_code_hi = 0;
 845
 846	if (svm->vcpu.arch.exception.has_error_code)
 847		svm->vmcb->control.exit_info_1 = svm->vcpu.arch.exception.error_code;
 848
 849	/*
 850	 * EXITINFO2 is undefined for all exception intercepts other
 851	 * than #PF.
 852	 */
 853	if (nr == PF_VECTOR) {
 854		if (svm->vcpu.arch.exception.nested_apf)
 855			svm->vmcb->control.exit_info_2 = svm->vcpu.arch.apf.nested_apf_token;
 856		else if (svm->vcpu.arch.exception.has_payload)
 857			svm->vmcb->control.exit_info_2 = svm->vcpu.arch.exception.payload;
 858		else
 859			svm->vmcb->control.exit_info_2 = svm->vcpu.arch.cr2;
 860	} else if (nr == DB_VECTOR) {
 861		/* See inject_pending_event.  */
 862		kvm_deliver_exception_payload(&svm->vcpu);
 863		if (svm->vcpu.arch.dr7 & DR7_GD) {
 864			svm->vcpu.arch.dr7 &= ~DR7_GD;
 865			kvm_update_dr7(&svm->vcpu);
 
 866		}
 867	} else
 868		WARN_ON(svm->vcpu.arch.exception.has_payload);
 869
 870	nested_svm_vmexit(svm);
 871}
 872
 873static void nested_svm_smi(struct vcpu_svm *svm)
 874{
 875	svm->vmcb->control.exit_code = SVM_EXIT_SMI;
 876	svm->vmcb->control.exit_info_1 = 0;
 877	svm->vmcb->control.exit_info_2 = 0;
 878
 879	nested_svm_vmexit(svm);
 880}
 881
 882static void nested_svm_nmi(struct vcpu_svm *svm)
 883{
 884	svm->vmcb->control.exit_code = SVM_EXIT_NMI;
 885	svm->vmcb->control.exit_info_1 = 0;
 886	svm->vmcb->control.exit_info_2 = 0;
 887
 888	nested_svm_vmexit(svm);
 889}
 890
 891static void nested_svm_intr(struct vcpu_svm *svm)
 892{
 893	trace_kvm_nested_intr_vmexit(svm->vmcb->save.rip);
 894
 895	svm->vmcb->control.exit_code   = SVM_EXIT_INTR;
 896	svm->vmcb->control.exit_info_1 = 0;
 897	svm->vmcb->control.exit_info_2 = 0;
 898
 899	nested_svm_vmexit(svm);
 900}
 901
 902static inline bool nested_exit_on_init(struct vcpu_svm *svm)
 903{
 904	return (svm->nested.ctl.intercept & (1ULL << INTERCEPT_INIT));
 905}
 906
 907static void nested_svm_init(struct vcpu_svm *svm)
 908{
 909	svm->vmcb->control.exit_code   = SVM_EXIT_INIT;
 910	svm->vmcb->control.exit_info_1 = 0;
 911	svm->vmcb->control.exit_info_2 = 0;
 912
 913	nested_svm_vmexit(svm);
 914}
 915
 916
 917static int svm_check_nested_events(struct kvm_vcpu *vcpu)
 918{
 
 919	struct vcpu_svm *svm = to_svm(vcpu);
 920	bool block_nested_events =
 921		kvm_event_needs_reinjection(vcpu) || svm->nested.nested_run_pending;
 922	struct kvm_lapic *apic = vcpu->arch.apic;
 
 
 
 
 
 
 
 
 
 
 
 923
 924	if (lapic_in_kernel(vcpu) &&
 925	    test_bit(KVM_APIC_INIT, &apic->pending_events)) {
 926		if (block_nested_events)
 927			return -EBUSY;
 928		if (!nested_exit_on_init(svm))
 929			return 0;
 930		nested_svm_init(svm);
 
 
 
 
 
 
 
 931		return 0;
 932	}
 933
 934	if (vcpu->arch.exception.pending) {
 935		if (block_nested_events)
 936                        return -EBUSY;
 937		if (!nested_exit_on_exception(svm))
 938			return 0;
 939		nested_svm_inject_exception_vmexit(svm);
 940		return 0;
 941	}
 942
 
 943	if (vcpu->arch.smi_pending && !svm_smi_blocked(vcpu)) {
 944		if (block_nested_events)
 945			return -EBUSY;
 946		if (!nested_exit_on_smi(svm))
 947			return 0;
 948		nested_svm_smi(svm);
 949		return 0;
 950	}
 
 951
 952	if (vcpu->arch.nmi_pending && !svm_nmi_blocked(vcpu)) {
 953		if (block_nested_events)
 954			return -EBUSY;
 955		if (!nested_exit_on_nmi(svm))
 956			return 0;
 957		nested_svm_nmi(svm);
 958		return 0;
 959	}
 960
 961	if (kvm_cpu_has_interrupt(vcpu) && !svm_interrupt_blocked(vcpu)) {
 962		if (block_nested_events)
 963			return -EBUSY;
 964		if (!nested_exit_on_intr(svm))
 965			return 0;
 966		nested_svm_intr(svm);
 
 967		return 0;
 968	}
 969
 970	return 0;
 971}
 972
 973int nested_svm_exit_special(struct vcpu_svm *svm)
 974{
 975	u32 exit_code = svm->vmcb->control.exit_code;
 
 976
 977	switch (exit_code) {
 978	case SVM_EXIT_INTR:
 979	case SVM_EXIT_NMI:
 980	case SVM_EXIT_NPF:
 981		return NESTED_EXIT_HOST;
 982	case SVM_EXIT_EXCP_BASE ... SVM_EXIT_EXCP_BASE + 0x1f: {
 983		u32 excp_bits = 1 << (exit_code - SVM_EXIT_EXCP_BASE);
 984
 985		if (get_host_vmcb(svm)->control.intercept_exceptions & excp_bits)
 
 986			return NESTED_EXIT_HOST;
 987		else if (exit_code == SVM_EXIT_EXCP_BASE + PF_VECTOR &&
 988			 svm->vcpu.arch.apf.host_apf_flags)
 989			/* Trap async PF even if not shadowing */
 990			return NESTED_EXIT_HOST;
 991		break;
 992	}
 
 
 
 
 
 
 
 993	default:
 994		break;
 995	}
 996
 997	return NESTED_EXIT_CONTINUE;
 998}
 999
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1000static int svm_get_nested_state(struct kvm_vcpu *vcpu,
1001				struct kvm_nested_state __user *user_kvm_nested_state,
1002				u32 user_data_size)
1003{
1004	struct vcpu_svm *svm;
 
 
1005	struct kvm_nested_state kvm_state = {
1006		.flags = 0,
1007		.format = KVM_STATE_NESTED_FORMAT_SVM,
1008		.size = sizeof(kvm_state),
1009	};
1010	struct vmcb __user *user_vmcb = (struct vmcb __user *)
1011		&user_kvm_nested_state->data.svm[0];
1012
1013	if (!vcpu)
1014		return kvm_state.size + KVM_STATE_NESTED_SVM_VMCB_SIZE;
1015
1016	svm = to_svm(vcpu);
1017
1018	if (user_data_size < kvm_state.size)
1019		goto out;
1020
1021	/* First fill in the header and copy it out.  */
1022	if (is_guest_mode(vcpu)) {
1023		kvm_state.hdr.svm.vmcb_pa = svm->nested.vmcb;
1024		kvm_state.size += KVM_STATE_NESTED_SVM_VMCB_SIZE;
1025		kvm_state.flags |= KVM_STATE_NESTED_GUEST_MODE;
1026
1027		if (svm->nested.nested_run_pending)
1028			kvm_state.flags |= KVM_STATE_NESTED_RUN_PENDING;
1029	}
1030
1031	if (gif_set(svm))
1032		kvm_state.flags |= KVM_STATE_NESTED_GIF_SET;
1033
1034	if (copy_to_user(user_kvm_nested_state, &kvm_state, sizeof(kvm_state)))
1035		return -EFAULT;
1036
1037	if (!is_guest_mode(vcpu))
1038		goto out;
1039
1040	/*
1041	 * Copy over the full size of the VMCB rather than just the size
1042	 * of the structs.
1043	 */
1044	if (clear_user(user_vmcb, KVM_STATE_NESTED_SVM_VMCB_SIZE))
1045		return -EFAULT;
1046	if (copy_to_user(&user_vmcb->control, &svm->nested.ctl,
1047			 sizeof(user_vmcb->control)))
 
 
 
 
 
 
 
 
1048		return -EFAULT;
1049	if (copy_to_user(&user_vmcb->save, &svm->nested.hsave->save,
 
1050			 sizeof(user_vmcb->save)))
1051		return -EFAULT;
1052
1053out:
1054	return kvm_state.size;
1055}
1056
1057static int svm_set_nested_state(struct kvm_vcpu *vcpu,
1058				struct kvm_nested_state __user *user_kvm_nested_state,
1059				struct kvm_nested_state *kvm_state)
1060{
1061	struct vcpu_svm *svm = to_svm(vcpu);
1062	struct vmcb *hsave = svm->nested.hsave;
1063	struct vmcb __user *user_vmcb = (struct vmcb __user *)
1064		&user_kvm_nested_state->data.svm[0];
1065	struct vmcb_control_area ctl;
1066	struct vmcb_save_area save;
1067	u32 cr0;
 
 
 
 
 
 
1068
1069	if (kvm_state->format != KVM_STATE_NESTED_FORMAT_SVM)
1070		return -EINVAL;
1071
1072	if (kvm_state->flags & ~(KVM_STATE_NESTED_GUEST_MODE |
1073				 KVM_STATE_NESTED_RUN_PENDING |
1074				 KVM_STATE_NESTED_GIF_SET))
1075		return -EINVAL;
1076
1077	/*
1078	 * If in guest mode, vcpu->arch.efer actually refers to the L2 guest's
1079	 * EFER.SVME, but EFER.SVME still has to be 1 for VMRUN to succeed.
1080	 */
1081	if (!(vcpu->arch.efer & EFER_SVME)) {
1082		/* GIF=1 and no guest mode are required if SVME=0.  */
1083		if (kvm_state->flags != KVM_STATE_NESTED_GIF_SET)
1084			return -EINVAL;
1085	}
1086
1087	/* SMM temporarily disables SVM, so we cannot be in guest mode.  */
1088	if (is_smm(vcpu) && (kvm_state->flags & KVM_STATE_NESTED_GUEST_MODE))
1089		return -EINVAL;
1090
1091	if (!(kvm_state->flags & KVM_STATE_NESTED_GUEST_MODE)) {
1092		svm_leave_nested(svm);
1093		goto out_set_gif;
 
1094	}
1095
1096	if (!page_address_valid(vcpu, kvm_state->hdr.svm.vmcb_pa))
1097		return -EINVAL;
1098	if (kvm_state->size < sizeof(*kvm_state) + KVM_STATE_NESTED_SVM_VMCB_SIZE)
1099		return -EINVAL;
1100	if (copy_from_user(&ctl, &user_vmcb->control, sizeof(ctl)))
1101		return -EFAULT;
1102	if (copy_from_user(&save, &user_vmcb->save, sizeof(save)))
1103		return -EFAULT;
1104
1105	if (!nested_vmcb_check_controls(&ctl))
1106		return -EINVAL;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1107
1108	/*
1109	 * Processor state contains L2 state.  Check that it is
1110	 * valid for guest mode (see nested_vmcb_checks).
1111	 */
1112	cr0 = kvm_read_cr0(vcpu);
1113        if (((cr0 & X86_CR0_CD) == 0) && (cr0 & X86_CR0_NW))
1114                return -EINVAL;
1115
1116	/*
1117	 * Validate host state saved from before VMRUN (see
1118	 * nested_svm_check_permissions).
1119	 * TODO: validate reserved bits for all saved state.
1120	 */
1121	if (!(save.cr0 & X86_CR0_PG))
1122		return -EINVAL;
 
 
 
 
 
1123
1124	/*
1125	 * All checks done, we can enter guest mode.  L1 control fields
1126	 * come from the nested save state.  Guest state is already
1127	 * in the registers, the save area of the nested state instead
1128	 * contains saved L1 state.
1129	 */
1130	copy_vmcb_control_area(&hsave->control, &svm->vmcb->control);
1131	hsave->save = save;
1132
1133	svm->nested.vmcb = kvm_state->hdr.svm.vmcb_pa;
1134	load_nested_vmcb_control(svm, &ctl);
1135	nested_prepare_vmcb_control(svm);
1136
1137	if (!nested_svm_vmrun_msrpm(svm))
1138		return -EINVAL;
 
 
1139
1140out_set_gif:
1141	svm_set_gif(svm, !!(kvm_state->flags & KVM_STATE_NESTED_GIF_SET));
1142	return 0;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1143}
1144
1145struct kvm_x86_nested_ops svm_nested_ops = {
 
 
1146	.check_events = svm_check_nested_events,
 
 
1147	.get_state = svm_get_nested_state,
1148	.set_state = svm_set_nested_state,
 
1149};