Linux Audio

Check our new training course

Loading...
Note: File does not exist in v3.15.
   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
  32#define CC KVM_NESTED_VMENTER_CONSISTENCY_CHECK
  33
  34static void nested_svm_inject_npf_exit(struct kvm_vcpu *vcpu,
  35				       struct x86_exception *fault)
  36{
  37	struct vcpu_svm *svm = to_svm(vcpu);
  38
  39	if (svm->vmcb->control.exit_code != SVM_EXIT_NPF) {
  40		/*
  41		 * TODO: track the cause of the nested page fault, and
  42		 * correctly fill in the high bits of exit_info_1.
  43		 */
  44		svm->vmcb->control.exit_code = SVM_EXIT_NPF;
  45		svm->vmcb->control.exit_code_hi = 0;
  46		svm->vmcb->control.exit_info_1 = (1ULL << 32);
  47		svm->vmcb->control.exit_info_2 = fault->address;
  48	}
  49
  50	svm->vmcb->control.exit_info_1 &= ~0xffffffffULL;
  51	svm->vmcb->control.exit_info_1 |= fault->error_code;
  52
  53	nested_svm_vmexit(svm);
  54}
  55
  56static void svm_inject_page_fault_nested(struct kvm_vcpu *vcpu, struct x86_exception *fault)
  57{
  58       struct vcpu_svm *svm = to_svm(vcpu);
  59       WARN_ON(!is_guest_mode(vcpu));
  60
  61       if (vmcb_is_intercept(&svm->nested.ctl, INTERCEPT_EXCEPTION_OFFSET + PF_VECTOR) &&
  62	   !svm->nested.nested_run_pending) {
  63               svm->vmcb->control.exit_code = SVM_EXIT_EXCP_BASE + PF_VECTOR;
  64               svm->vmcb->control.exit_code_hi = 0;
  65               svm->vmcb->control.exit_info_1 = fault->error_code;
  66               svm->vmcb->control.exit_info_2 = fault->address;
  67               nested_svm_vmexit(svm);
  68       } else {
  69               kvm_inject_page_fault(vcpu, fault);
  70       }
  71}
  72
  73static u64 nested_svm_get_tdp_pdptr(struct kvm_vcpu *vcpu, int index)
  74{
  75	struct vcpu_svm *svm = to_svm(vcpu);
  76	u64 cr3 = svm->nested.ctl.nested_cr3;
  77	u64 pdpte;
  78	int ret;
  79
  80	ret = kvm_vcpu_read_guest_page(vcpu, gpa_to_gfn(cr3), &pdpte,
  81				       offset_in_page(cr3) + index * 8, 8);
  82	if (ret)
  83		return 0;
  84	return pdpte;
  85}
  86
  87static unsigned long nested_svm_get_tdp_cr3(struct kvm_vcpu *vcpu)
  88{
  89	struct vcpu_svm *svm = to_svm(vcpu);
  90
  91	return svm->nested.ctl.nested_cr3;
  92}
  93
  94static void nested_svm_init_mmu_context(struct kvm_vcpu *vcpu)
  95{
  96	struct vcpu_svm *svm = to_svm(vcpu);
  97
  98	WARN_ON(mmu_is_nested(vcpu));
  99
 100	vcpu->arch.mmu = &vcpu->arch.guest_mmu;
 101
 102	/*
 103	 * The NPT format depends on L1's CR4 and EFER, which is in vmcb01.  Note,
 104	 * when called via KVM_SET_NESTED_STATE, that state may _not_ match current
 105	 * vCPU state.  CR0.WP is explicitly ignored, while CR0.PG is required.
 106	 */
 107	kvm_init_shadow_npt_mmu(vcpu, X86_CR0_PG, svm->vmcb01.ptr->save.cr4,
 108				svm->vmcb01.ptr->save.efer,
 109				svm->nested.ctl.nested_cr3);
 110	vcpu->arch.mmu->get_guest_pgd     = nested_svm_get_tdp_cr3;
 111	vcpu->arch.mmu->get_pdptr         = nested_svm_get_tdp_pdptr;
 112	vcpu->arch.mmu->inject_page_fault = nested_svm_inject_npf_exit;
 113	vcpu->arch.walk_mmu              = &vcpu->arch.nested_mmu;
 114}
 115
 116static void nested_svm_uninit_mmu_context(struct kvm_vcpu *vcpu)
 117{
 118	vcpu->arch.mmu = &vcpu->arch.root_mmu;
 119	vcpu->arch.walk_mmu = &vcpu->arch.root_mmu;
 120}
 121
 122void recalc_intercepts(struct vcpu_svm *svm)
 123{
 124	struct vmcb_control_area *c, *h, *g;
 125	unsigned int i;
 126
 127	vmcb_mark_dirty(svm->vmcb, VMCB_INTERCEPTS);
 128
 129	if (!is_guest_mode(&svm->vcpu))
 130		return;
 131
 132	c = &svm->vmcb->control;
 133	h = &svm->vmcb01.ptr->control;
 134	g = &svm->nested.ctl;
 135
 136	for (i = 0; i < MAX_INTERCEPT; i++)
 137		c->intercepts[i] = h->intercepts[i];
 138
 139	if (g->int_ctl & V_INTR_MASKING_MASK) {
 140		/* We only want the cr8 intercept bits of L1 */
 141		vmcb_clr_intercept(c, INTERCEPT_CR8_READ);
 142		vmcb_clr_intercept(c, INTERCEPT_CR8_WRITE);
 143
 144		/*
 145		 * Once running L2 with HF_VINTR_MASK, EFLAGS.IF does not
 146		 * affect any interrupt we may want to inject; therefore,
 147		 * interrupt window vmexits are irrelevant to L0.
 148		 */
 149		vmcb_clr_intercept(c, INTERCEPT_VINTR);
 150	}
 151
 152	/* We don't want to see VMMCALLs from a nested guest */
 153	vmcb_clr_intercept(c, INTERCEPT_VMMCALL);
 154
 155	for (i = 0; i < MAX_INTERCEPT; i++)
 156		c->intercepts[i] |= g->intercepts[i];
 157
 158	/* If SMI is not intercepted, ignore guest SMI intercept as well  */
 159	if (!intercept_smi)
 160		vmcb_clr_intercept(c, INTERCEPT_SMI);
 161
 162	vmcb_set_intercept(c, INTERCEPT_VMLOAD);
 163	vmcb_set_intercept(c, INTERCEPT_VMSAVE);
 164}
 165
 166static void copy_vmcb_control_area(struct vmcb_control_area *dst,
 167				   struct vmcb_control_area *from)
 168{
 169	unsigned int i;
 170
 171	for (i = 0; i < MAX_INTERCEPT; i++)
 172		dst->intercepts[i] = from->intercepts[i];
 173
 174	dst->iopm_base_pa         = from->iopm_base_pa;
 175	dst->msrpm_base_pa        = from->msrpm_base_pa;
 176	dst->tsc_offset           = from->tsc_offset;
 177	/* asid not copied, it is handled manually for svm->vmcb.  */
 178	dst->tlb_ctl              = from->tlb_ctl;
 179	dst->int_ctl              = from->int_ctl;
 180	dst->int_vector           = from->int_vector;
 181	dst->int_state            = from->int_state;
 182	dst->exit_code            = from->exit_code;
 183	dst->exit_code_hi         = from->exit_code_hi;
 184	dst->exit_info_1          = from->exit_info_1;
 185	dst->exit_info_2          = from->exit_info_2;
 186	dst->exit_int_info        = from->exit_int_info;
 187	dst->exit_int_info_err    = from->exit_int_info_err;
 188	dst->nested_ctl           = from->nested_ctl;
 189	dst->event_inj            = from->event_inj;
 190	dst->event_inj_err        = from->event_inj_err;
 191	dst->nested_cr3           = from->nested_cr3;
 192	dst->virt_ext              = from->virt_ext;
 193	dst->pause_filter_count   = from->pause_filter_count;
 194	dst->pause_filter_thresh  = from->pause_filter_thresh;
 195}
 196
 197static bool nested_svm_vmrun_msrpm(struct vcpu_svm *svm)
 198{
 199	/*
 200	 * This function merges the msr permission bitmaps of kvm and the
 201	 * nested vmcb. It is optimized in that it only merges the parts where
 202	 * the kvm msr permission bitmap may contain zero bits
 203	 */
 204	int i;
 205
 206	if (!(vmcb_is_intercept(&svm->nested.ctl, INTERCEPT_MSR_PROT)))
 207		return true;
 208
 209	for (i = 0; i < MSRPM_OFFSETS; i++) {
 210		u32 value, p;
 211		u64 offset;
 212
 213		if (msrpm_offsets[i] == 0xffffffff)
 214			break;
 215
 216		p      = msrpm_offsets[i];
 217		offset = svm->nested.ctl.msrpm_base_pa + (p * 4);
 218
 219		if (kvm_vcpu_read_guest(&svm->vcpu, offset, &value, 4))
 220			return false;
 221
 222		svm->nested.msrpm[p] = svm->msrpm[p] | value;
 223	}
 224
 225	svm->vmcb->control.msrpm_base_pa = __sme_set(__pa(svm->nested.msrpm));
 226
 227	return true;
 228}
 229
 230/*
 231 * Bits 11:0 of bitmap address are ignored by hardware
 232 */
 233static bool nested_svm_check_bitmap_pa(struct kvm_vcpu *vcpu, u64 pa, u32 size)
 234{
 235	u64 addr = PAGE_ALIGN(pa);
 236
 237	return kvm_vcpu_is_legal_gpa(vcpu, addr) &&
 238	    kvm_vcpu_is_legal_gpa(vcpu, addr + size - 1);
 239}
 240
 241static bool nested_vmcb_check_controls(struct kvm_vcpu *vcpu,
 242				       struct vmcb_control_area *control)
 243{
 244	if (CC(!vmcb_is_intercept(control, INTERCEPT_VMRUN)))
 245		return false;
 246
 247	if (CC(control->asid == 0))
 248		return false;
 249
 250	if (CC((control->nested_ctl & SVM_NESTED_CTL_NP_ENABLE) && !npt_enabled))
 251		return false;
 252
 253	if (CC(!nested_svm_check_bitmap_pa(vcpu, control->msrpm_base_pa,
 254					   MSRPM_SIZE)))
 255		return false;
 256	if (CC(!nested_svm_check_bitmap_pa(vcpu, control->iopm_base_pa,
 257					   IOPM_SIZE)))
 258		return false;
 259
 260	return true;
 261}
 262
 263static bool nested_vmcb_check_cr3_cr4(struct kvm_vcpu *vcpu,
 264				      struct vmcb_save_area *save)
 265{
 266	/*
 267	 * These checks are also performed by KVM_SET_SREGS,
 268	 * except that EFER.LMA is not checked by SVM against
 269	 * CR0.PG && EFER.LME.
 270	 */
 271	if ((save->efer & EFER_LME) && (save->cr0 & X86_CR0_PG)) {
 272		if (CC(!(save->cr4 & X86_CR4_PAE)) ||
 273		    CC(!(save->cr0 & X86_CR0_PE)) ||
 274		    CC(kvm_vcpu_is_illegal_gpa(vcpu, save->cr3)))
 275			return false;
 276	}
 277
 278	if (CC(!kvm_is_valid_cr4(vcpu, save->cr4)))
 279		return false;
 280
 281	return true;
 282}
 283
 284/* Common checks that apply to both L1 and L2 state.  */
 285static bool nested_vmcb_valid_sregs(struct kvm_vcpu *vcpu,
 286				    struct vmcb_save_area *save)
 287{
 288	/*
 289	 * FIXME: these should be done after copying the fields,
 290	 * to avoid TOC/TOU races.  For these save area checks
 291	 * the possible damage is limited since kvm_set_cr0 and
 292	 * kvm_set_cr4 handle failure; EFER_SVME is an exception
 293	 * so it is force-set later in nested_prepare_vmcb_save.
 294	 */
 295	if (CC(!(save->efer & EFER_SVME)))
 296		return false;
 297
 298	if (CC((save->cr0 & X86_CR0_CD) == 0 && (save->cr0 & X86_CR0_NW)) ||
 299	    CC(save->cr0 & ~0xffffffffULL))
 300		return false;
 301
 302	if (CC(!kvm_dr6_valid(save->dr6)) || CC(!kvm_dr7_valid(save->dr7)))
 303		return false;
 304
 305	if (!nested_vmcb_check_cr3_cr4(vcpu, save))
 306		return false;
 307
 308	if (CC(!kvm_valid_efer(vcpu, save->efer)))
 309		return false;
 310
 311	return true;
 312}
 313
 314void nested_load_control_from_vmcb12(struct vcpu_svm *svm,
 315				     struct vmcb_control_area *control)
 316{
 317	copy_vmcb_control_area(&svm->nested.ctl, control);
 318
 319	/* Copy it here because nested_svm_check_controls will check it.  */
 320	svm->nested.ctl.asid           = control->asid;
 321	svm->nested.ctl.msrpm_base_pa &= ~0x0fffULL;
 322	svm->nested.ctl.iopm_base_pa  &= ~0x0fffULL;
 323}
 324
 325/*
 326 * Synchronize fields that are written by the processor, so that
 327 * they can be copied back into the vmcb12.
 328 */
 329void nested_sync_control_from_vmcb02(struct vcpu_svm *svm)
 330{
 331	u32 mask;
 332	svm->nested.ctl.event_inj      = svm->vmcb->control.event_inj;
 333	svm->nested.ctl.event_inj_err  = svm->vmcb->control.event_inj_err;
 334
 335	/* Only a few fields of int_ctl are written by the processor.  */
 336	mask = V_IRQ_MASK | V_TPR_MASK;
 337	if (!(svm->nested.ctl.int_ctl & V_INTR_MASKING_MASK) &&
 338	    svm_is_intercept(svm, INTERCEPT_VINTR)) {
 339		/*
 340		 * In order to request an interrupt window, L0 is usurping
 341		 * svm->vmcb->control.int_ctl and possibly setting V_IRQ
 342		 * even if it was clear in L1's VMCB.  Restoring it would be
 343		 * wrong.  However, in this case V_IRQ will remain true until
 344		 * interrupt_window_interception calls svm_clear_vintr and
 345		 * restores int_ctl.  We can just leave it aside.
 346		 */
 347		mask &= ~V_IRQ_MASK;
 348	}
 349	svm->nested.ctl.int_ctl        &= ~mask;
 350	svm->nested.ctl.int_ctl        |= svm->vmcb->control.int_ctl & mask;
 351}
 352
 353/*
 354 * Transfer any event that L0 or L1 wanted to inject into L2 to
 355 * EXIT_INT_INFO.
 356 */
 357static void nested_save_pending_event_to_vmcb12(struct vcpu_svm *svm,
 358						struct vmcb *vmcb12)
 359{
 360	struct kvm_vcpu *vcpu = &svm->vcpu;
 361	u32 exit_int_info = 0;
 362	unsigned int nr;
 363
 364	if (vcpu->arch.exception.injected) {
 365		nr = vcpu->arch.exception.nr;
 366		exit_int_info = nr | SVM_EVTINJ_VALID | SVM_EVTINJ_TYPE_EXEPT;
 367
 368		if (vcpu->arch.exception.has_error_code) {
 369			exit_int_info |= SVM_EVTINJ_VALID_ERR;
 370			vmcb12->control.exit_int_info_err =
 371				vcpu->arch.exception.error_code;
 372		}
 373
 374	} else if (vcpu->arch.nmi_injected) {
 375		exit_int_info = SVM_EVTINJ_VALID | SVM_EVTINJ_TYPE_NMI;
 376
 377	} else if (vcpu->arch.interrupt.injected) {
 378		nr = vcpu->arch.interrupt.nr;
 379		exit_int_info = nr | SVM_EVTINJ_VALID;
 380
 381		if (vcpu->arch.interrupt.soft)
 382			exit_int_info |= SVM_EVTINJ_TYPE_SOFT;
 383		else
 384			exit_int_info |= SVM_EVTINJ_TYPE_INTR;
 385	}
 386
 387	vmcb12->control.exit_int_info = exit_int_info;
 388}
 389
 390static inline bool nested_npt_enabled(struct vcpu_svm *svm)
 391{
 392	return svm->nested.ctl.nested_ctl & SVM_NESTED_CTL_NP_ENABLE;
 393}
 394
 395static void nested_svm_transition_tlb_flush(struct kvm_vcpu *vcpu)
 396{
 397	/*
 398	 * TODO: optimize unconditional TLB flush/MMU sync.  A partial list of
 399	 * things to fix before this can be conditional:
 400	 *
 401	 *  - Flush TLBs for both L1 and L2 remote TLB flush
 402	 *  - Honor L1's request to flush an ASID on nested VMRUN
 403	 *  - Sync nested NPT MMU on VMRUN that flushes L2's ASID[*]
 404	 *  - Don't crush a pending TLB flush in vmcb02 on nested VMRUN
 405	 *  - Flush L1's ASID on KVM_REQ_TLB_FLUSH_GUEST
 406	 *
 407	 * [*] Unlike nested EPT, SVM's ASID management can invalidate nested
 408	 *     NPT guest-physical mappings on VMRUN.
 409	 */
 410	kvm_make_request(KVM_REQ_MMU_SYNC, vcpu);
 411	kvm_make_request(KVM_REQ_TLB_FLUSH_CURRENT, vcpu);
 412}
 413
 414/*
 415 * Load guest's/host's cr3 on nested vmentry or vmexit. @nested_npt is true
 416 * if we are emulating VM-Entry into a guest with NPT enabled.
 417 */
 418static int nested_svm_load_cr3(struct kvm_vcpu *vcpu, unsigned long cr3,
 419			       bool nested_npt, bool reload_pdptrs)
 420{
 421	if (CC(kvm_vcpu_is_illegal_gpa(vcpu, cr3)))
 422		return -EINVAL;
 423
 424	if (reload_pdptrs && !nested_npt && is_pae_paging(vcpu) &&
 425	    CC(!load_pdptrs(vcpu, vcpu->arch.walk_mmu, cr3)))
 426		return -EINVAL;
 427
 428	if (!nested_npt)
 429		kvm_mmu_new_pgd(vcpu, cr3);
 430
 431	vcpu->arch.cr3 = cr3;
 432	kvm_register_mark_available(vcpu, VCPU_EXREG_CR3);
 433
 434	/* Re-initialize the MMU, e.g. to pick up CR4 MMU role changes. */
 435	kvm_init_mmu(vcpu);
 436
 437	return 0;
 438}
 439
 440void nested_vmcb02_compute_g_pat(struct vcpu_svm *svm)
 441{
 442	if (!svm->nested.vmcb02.ptr)
 443		return;
 444
 445	/* FIXME: merge g_pat from vmcb01 and vmcb12.  */
 446	svm->nested.vmcb02.ptr->save.g_pat = svm->vmcb01.ptr->save.g_pat;
 447}
 448
 449static void nested_vmcb02_prepare_save(struct vcpu_svm *svm, struct vmcb *vmcb12)
 450{
 451	bool new_vmcb12 = false;
 452
 453	nested_vmcb02_compute_g_pat(svm);
 454
 455	/* Load the nested guest state */
 456	if (svm->nested.vmcb12_gpa != svm->nested.last_vmcb12_gpa) {
 457		new_vmcb12 = true;
 458		svm->nested.last_vmcb12_gpa = svm->nested.vmcb12_gpa;
 459	}
 460
 461	if (unlikely(new_vmcb12 || vmcb_is_dirty(vmcb12, VMCB_SEG))) {
 462		svm->vmcb->save.es = vmcb12->save.es;
 463		svm->vmcb->save.cs = vmcb12->save.cs;
 464		svm->vmcb->save.ss = vmcb12->save.ss;
 465		svm->vmcb->save.ds = vmcb12->save.ds;
 466		svm->vmcb->save.cpl = vmcb12->save.cpl;
 467		vmcb_mark_dirty(svm->vmcb, VMCB_SEG);
 468	}
 469
 470	if (unlikely(new_vmcb12 || vmcb_is_dirty(vmcb12, VMCB_DT))) {
 471		svm->vmcb->save.gdtr = vmcb12->save.gdtr;
 472		svm->vmcb->save.idtr = vmcb12->save.idtr;
 473		vmcb_mark_dirty(svm->vmcb, VMCB_DT);
 474	}
 475
 476	kvm_set_rflags(&svm->vcpu, vmcb12->save.rflags | X86_EFLAGS_FIXED);
 477
 478	/*
 479	 * Force-set EFER_SVME even though it is checked earlier on the
 480	 * VMCB12, because the guest can flip the bit between the check
 481	 * and now.  Clearing EFER_SVME would call svm_free_nested.
 482	 */
 483	svm_set_efer(&svm->vcpu, vmcb12->save.efer | EFER_SVME);
 484
 485	svm_set_cr0(&svm->vcpu, vmcb12->save.cr0);
 486	svm_set_cr4(&svm->vcpu, vmcb12->save.cr4);
 487
 488	svm->vcpu.arch.cr2 = vmcb12->save.cr2;
 489
 490	kvm_rax_write(&svm->vcpu, vmcb12->save.rax);
 491	kvm_rsp_write(&svm->vcpu, vmcb12->save.rsp);
 492	kvm_rip_write(&svm->vcpu, vmcb12->save.rip);
 493
 494	/* In case we don't even reach vcpu_run, the fields are not updated */
 495	svm->vmcb->save.rax = vmcb12->save.rax;
 496	svm->vmcb->save.rsp = vmcb12->save.rsp;
 497	svm->vmcb->save.rip = vmcb12->save.rip;
 498
 499	/* These bits will be set properly on the first execution when new_vmc12 is true */
 500	if (unlikely(new_vmcb12 || vmcb_is_dirty(vmcb12, VMCB_DR))) {
 501		svm->vmcb->save.dr7 = vmcb12->save.dr7 | DR7_FIXED_1;
 502		svm->vcpu.arch.dr6  = vmcb12->save.dr6 | DR6_ACTIVE_LOW;
 503		vmcb_mark_dirty(svm->vmcb, VMCB_DR);
 504	}
 505}
 506
 507static void nested_vmcb02_prepare_control(struct vcpu_svm *svm)
 508{
 509	const u32 int_ctl_vmcb01_bits =
 510		V_INTR_MASKING_MASK | V_GIF_MASK | V_GIF_ENABLE_MASK;
 511
 512	const u32 int_ctl_vmcb12_bits = V_TPR_MASK | V_IRQ_INJECTION_BITS_MASK;
 513
 514	struct kvm_vcpu *vcpu = &svm->vcpu;
 515
 516	/*
 517	 * Filled at exit: exit_code, exit_code_hi, exit_info_1, exit_info_2,
 518	 * exit_int_info, exit_int_info_err, next_rip, insn_len, insn_bytes.
 519	 */
 520
 521	/*
 522	 * Also covers avic_vapic_bar, avic_backing_page, avic_logical_id,
 523	 * avic_physical_id.
 524	 */
 525	WARN_ON(kvm_apicv_activated(svm->vcpu.kvm));
 526
 527	/* Copied from vmcb01.  msrpm_base can be overwritten later.  */
 528	svm->vmcb->control.nested_ctl = svm->vmcb01.ptr->control.nested_ctl;
 529	svm->vmcb->control.iopm_base_pa = svm->vmcb01.ptr->control.iopm_base_pa;
 530	svm->vmcb->control.msrpm_base_pa = svm->vmcb01.ptr->control.msrpm_base_pa;
 531
 532	/* Done at vmrun: asid.  */
 533
 534	/* Also overwritten later if necessary.  */
 535	svm->vmcb->control.tlb_ctl = TLB_CONTROL_DO_NOTHING;
 536
 537	/* nested_cr3.  */
 538	if (nested_npt_enabled(svm))
 539		nested_svm_init_mmu_context(vcpu);
 540
 541	svm->vmcb->control.tsc_offset = vcpu->arch.tsc_offset =
 542		vcpu->arch.l1_tsc_offset + svm->nested.ctl.tsc_offset;
 543
 544	svm->vmcb->control.int_ctl             =
 545		(svm->nested.ctl.int_ctl & int_ctl_vmcb12_bits) |
 546		(svm->vmcb01.ptr->control.int_ctl & int_ctl_vmcb01_bits);
 547
 548	svm->vmcb->control.int_vector          = svm->nested.ctl.int_vector;
 549	svm->vmcb->control.int_state           = svm->nested.ctl.int_state;
 550	svm->vmcb->control.event_inj           = svm->nested.ctl.event_inj;
 551	svm->vmcb->control.event_inj_err       = svm->nested.ctl.event_inj_err;
 552
 553	svm->vmcb->control.pause_filter_count  = svm->nested.ctl.pause_filter_count;
 554	svm->vmcb->control.pause_filter_thresh = svm->nested.ctl.pause_filter_thresh;
 555
 556	nested_svm_transition_tlb_flush(vcpu);
 557
 558	/* Enter Guest-Mode */
 559	enter_guest_mode(vcpu);
 560
 561	/*
 562	 * Merge guest and host intercepts - must be called with vcpu in
 563	 * guest-mode to take effect.
 564	 */
 565	recalc_intercepts(svm);
 566}
 567
 568static void nested_svm_copy_common_state(struct vmcb *from_vmcb, struct vmcb *to_vmcb)
 569{
 570	/*
 571	 * Some VMCB state is shared between L1 and L2 and thus has to be
 572	 * moved at the time of nested vmrun and vmexit.
 573	 *
 574	 * VMLOAD/VMSAVE state would also belong in this category, but KVM
 575	 * always performs VMLOAD and VMSAVE from the VMCB01.
 576	 */
 577	to_vmcb->save.spec_ctrl = from_vmcb->save.spec_ctrl;
 578}
 579
 580int enter_svm_guest_mode(struct kvm_vcpu *vcpu, u64 vmcb12_gpa,
 581			 struct vmcb *vmcb12)
 582{
 583	struct vcpu_svm *svm = to_svm(vcpu);
 584	int ret;
 585
 586	trace_kvm_nested_vmrun(svm->vmcb->save.rip, vmcb12_gpa,
 587			       vmcb12->save.rip,
 588			       vmcb12->control.int_ctl,
 589			       vmcb12->control.event_inj,
 590			       vmcb12->control.nested_ctl);
 591
 592	trace_kvm_nested_intercepts(vmcb12->control.intercepts[INTERCEPT_CR] & 0xffff,
 593				    vmcb12->control.intercepts[INTERCEPT_CR] >> 16,
 594				    vmcb12->control.intercepts[INTERCEPT_EXCEPTION],
 595				    vmcb12->control.intercepts[INTERCEPT_WORD3],
 596				    vmcb12->control.intercepts[INTERCEPT_WORD4],
 597				    vmcb12->control.intercepts[INTERCEPT_WORD5]);
 598
 599
 600	svm->nested.vmcb12_gpa = vmcb12_gpa;
 601
 602	WARN_ON(svm->vmcb == svm->nested.vmcb02.ptr);
 603
 604	nested_svm_copy_common_state(svm->vmcb01.ptr, svm->nested.vmcb02.ptr);
 605
 606	svm_switch_vmcb(svm, &svm->nested.vmcb02);
 607	nested_vmcb02_prepare_control(svm);
 608	nested_vmcb02_prepare_save(svm, vmcb12);
 609
 610	ret = nested_svm_load_cr3(&svm->vcpu, vmcb12->save.cr3,
 611				  nested_npt_enabled(svm), true);
 612	if (ret)
 613		return ret;
 614
 615	if (!npt_enabled)
 616		vcpu->arch.mmu->inject_page_fault = svm_inject_page_fault_nested;
 617
 618	svm_set_gif(svm, true);
 619
 620	return 0;
 621}
 622
 623int nested_svm_vmrun(struct kvm_vcpu *vcpu)
 624{
 625	struct vcpu_svm *svm = to_svm(vcpu);
 626	int ret;
 627	struct vmcb *vmcb12;
 628	struct kvm_host_map map;
 629	u64 vmcb12_gpa;
 630
 631	if (!svm->nested.hsave_msr) {
 632		kvm_inject_gp(vcpu, 0);
 633		return 1;
 634	}
 635
 636	if (is_smm(vcpu)) {
 637		kvm_queue_exception(vcpu, UD_VECTOR);
 638		return 1;
 639	}
 640
 641	vmcb12_gpa = svm->vmcb->save.rax;
 642	ret = kvm_vcpu_map(vcpu, gpa_to_gfn(vmcb12_gpa), &map);
 643	if (ret == -EINVAL) {
 644		kvm_inject_gp(vcpu, 0);
 645		return 1;
 646	} else if (ret) {
 647		return kvm_skip_emulated_instruction(vcpu);
 648	}
 649
 650	ret = kvm_skip_emulated_instruction(vcpu);
 651
 652	vmcb12 = map.hva;
 653
 654	if (WARN_ON_ONCE(!svm->nested.initialized))
 655		return -EINVAL;
 656
 657	nested_load_control_from_vmcb12(svm, &vmcb12->control);
 658
 659	if (!nested_vmcb_valid_sregs(vcpu, &vmcb12->save) ||
 660	    !nested_vmcb_check_controls(vcpu, &svm->nested.ctl)) {
 661		vmcb12->control.exit_code    = SVM_EXIT_ERR;
 662		vmcb12->control.exit_code_hi = 0;
 663		vmcb12->control.exit_info_1  = 0;
 664		vmcb12->control.exit_info_2  = 0;
 665		goto out;
 666	}
 667
 668
 669	/* Clear internal status */
 670	kvm_clear_exception_queue(vcpu);
 671	kvm_clear_interrupt_queue(vcpu);
 672
 673	/*
 674	 * Since vmcb01 is not in use, we can use it to store some of the L1
 675	 * state.
 676	 */
 677	svm->vmcb01.ptr->save.efer   = vcpu->arch.efer;
 678	svm->vmcb01.ptr->save.cr0    = kvm_read_cr0(vcpu);
 679	svm->vmcb01.ptr->save.cr4    = vcpu->arch.cr4;
 680	svm->vmcb01.ptr->save.rflags = kvm_get_rflags(vcpu);
 681	svm->vmcb01.ptr->save.rip    = kvm_rip_read(vcpu);
 682
 683	if (!npt_enabled)
 684		svm->vmcb01.ptr->save.cr3 = kvm_read_cr3(vcpu);
 685
 686	svm->nested.nested_run_pending = 1;
 687
 688	if (enter_svm_guest_mode(vcpu, vmcb12_gpa, vmcb12))
 689		goto out_exit_err;
 690
 691	if (nested_svm_vmrun_msrpm(svm))
 692		goto out;
 693
 694out_exit_err:
 695	svm->nested.nested_run_pending = 0;
 696
 697	svm->vmcb->control.exit_code    = SVM_EXIT_ERR;
 698	svm->vmcb->control.exit_code_hi = 0;
 699	svm->vmcb->control.exit_info_1  = 0;
 700	svm->vmcb->control.exit_info_2  = 0;
 701
 702	nested_svm_vmexit(svm);
 703
 704out:
 705	kvm_vcpu_unmap(vcpu, &map, true);
 706
 707	return ret;
 708}
 709
 710/* Copy state save area fields which are handled by VMRUN */
 711void svm_copy_vmrun_state(struct vmcb_save_area *to_save,
 712			  struct vmcb_save_area *from_save)
 713{
 714	to_save->es = from_save->es;
 715	to_save->cs = from_save->cs;
 716	to_save->ss = from_save->ss;
 717	to_save->ds = from_save->ds;
 718	to_save->gdtr = from_save->gdtr;
 719	to_save->idtr = from_save->idtr;
 720	to_save->rflags = from_save->rflags | X86_EFLAGS_FIXED;
 721	to_save->efer = from_save->efer;
 722	to_save->cr0 = from_save->cr0;
 723	to_save->cr3 = from_save->cr3;
 724	to_save->cr4 = from_save->cr4;
 725	to_save->rax = from_save->rax;
 726	to_save->rsp = from_save->rsp;
 727	to_save->rip = from_save->rip;
 728	to_save->cpl = 0;
 729}
 730
 731void svm_copy_vmloadsave_state(struct vmcb *to_vmcb, struct vmcb *from_vmcb)
 732{
 733	to_vmcb->save.fs = from_vmcb->save.fs;
 734	to_vmcb->save.gs = from_vmcb->save.gs;
 735	to_vmcb->save.tr = from_vmcb->save.tr;
 736	to_vmcb->save.ldtr = from_vmcb->save.ldtr;
 737	to_vmcb->save.kernel_gs_base = from_vmcb->save.kernel_gs_base;
 738	to_vmcb->save.star = from_vmcb->save.star;
 739	to_vmcb->save.lstar = from_vmcb->save.lstar;
 740	to_vmcb->save.cstar = from_vmcb->save.cstar;
 741	to_vmcb->save.sfmask = from_vmcb->save.sfmask;
 742	to_vmcb->save.sysenter_cs = from_vmcb->save.sysenter_cs;
 743	to_vmcb->save.sysenter_esp = from_vmcb->save.sysenter_esp;
 744	to_vmcb->save.sysenter_eip = from_vmcb->save.sysenter_eip;
 745}
 746
 747int nested_svm_vmexit(struct vcpu_svm *svm)
 748{
 749	struct kvm_vcpu *vcpu = &svm->vcpu;
 750	struct vmcb *vmcb12;
 751	struct vmcb *vmcb = svm->vmcb;
 752	struct kvm_host_map map;
 753	int rc;
 754
 755	/* Triple faults in L2 should never escape. */
 756	WARN_ON_ONCE(kvm_check_request(KVM_REQ_TRIPLE_FAULT, vcpu));
 757
 758	rc = kvm_vcpu_map(vcpu, gpa_to_gfn(svm->nested.vmcb12_gpa), &map);
 759	if (rc) {
 760		if (rc == -EINVAL)
 761			kvm_inject_gp(vcpu, 0);
 762		return 1;
 763	}
 764
 765	vmcb12 = map.hva;
 766
 767	/* Exit Guest-Mode */
 768	leave_guest_mode(vcpu);
 769	svm->nested.vmcb12_gpa = 0;
 770	WARN_ON_ONCE(svm->nested.nested_run_pending);
 771
 772	kvm_clear_request(KVM_REQ_GET_NESTED_STATE_PAGES, vcpu);
 773
 774	/* in case we halted in L2 */
 775	svm->vcpu.arch.mp_state = KVM_MP_STATE_RUNNABLE;
 776
 777	/* Give the current vmcb to the guest */
 778
 779	vmcb12->save.es     = vmcb->save.es;
 780	vmcb12->save.cs     = vmcb->save.cs;
 781	vmcb12->save.ss     = vmcb->save.ss;
 782	vmcb12->save.ds     = vmcb->save.ds;
 783	vmcb12->save.gdtr   = vmcb->save.gdtr;
 784	vmcb12->save.idtr   = vmcb->save.idtr;
 785	vmcb12->save.efer   = svm->vcpu.arch.efer;
 786	vmcb12->save.cr0    = kvm_read_cr0(vcpu);
 787	vmcb12->save.cr3    = kvm_read_cr3(vcpu);
 788	vmcb12->save.cr2    = vmcb->save.cr2;
 789	vmcb12->save.cr4    = svm->vcpu.arch.cr4;
 790	vmcb12->save.rflags = kvm_get_rflags(vcpu);
 791	vmcb12->save.rip    = kvm_rip_read(vcpu);
 792	vmcb12->save.rsp    = kvm_rsp_read(vcpu);
 793	vmcb12->save.rax    = kvm_rax_read(vcpu);
 794	vmcb12->save.dr7    = vmcb->save.dr7;
 795	vmcb12->save.dr6    = svm->vcpu.arch.dr6;
 796	vmcb12->save.cpl    = vmcb->save.cpl;
 797
 798	vmcb12->control.int_state         = vmcb->control.int_state;
 799	vmcb12->control.exit_code         = vmcb->control.exit_code;
 800	vmcb12->control.exit_code_hi      = vmcb->control.exit_code_hi;
 801	vmcb12->control.exit_info_1       = vmcb->control.exit_info_1;
 802	vmcb12->control.exit_info_2       = vmcb->control.exit_info_2;
 803
 804	if (vmcb12->control.exit_code != SVM_EXIT_ERR)
 805		nested_save_pending_event_to_vmcb12(svm, vmcb12);
 806
 807	if (svm->nrips_enabled)
 808		vmcb12->control.next_rip  = vmcb->control.next_rip;
 809
 810	vmcb12->control.int_ctl           = svm->nested.ctl.int_ctl;
 811	vmcb12->control.tlb_ctl           = svm->nested.ctl.tlb_ctl;
 812	vmcb12->control.event_inj         = svm->nested.ctl.event_inj;
 813	vmcb12->control.event_inj_err     = svm->nested.ctl.event_inj_err;
 814
 815	vmcb12->control.pause_filter_count =
 816		svm->vmcb->control.pause_filter_count;
 817	vmcb12->control.pause_filter_thresh =
 818		svm->vmcb->control.pause_filter_thresh;
 819
 820	nested_svm_copy_common_state(svm->nested.vmcb02.ptr, svm->vmcb01.ptr);
 821
 822	svm_switch_vmcb(svm, &svm->vmcb01);
 823
 824	/*
 825	 * On vmexit the  GIF is set to false and
 826	 * no event can be injected in L1.
 827	 */
 828	svm_set_gif(svm, false);
 829	svm->vmcb->control.exit_int_info = 0;
 830
 831	svm->vcpu.arch.tsc_offset = svm->vcpu.arch.l1_tsc_offset;
 832	if (svm->vmcb->control.tsc_offset != svm->vcpu.arch.tsc_offset) {
 833		svm->vmcb->control.tsc_offset = svm->vcpu.arch.tsc_offset;
 834		vmcb_mark_dirty(svm->vmcb, VMCB_INTERCEPTS);
 835	}
 836
 837	svm->nested.ctl.nested_cr3 = 0;
 838
 839	/*
 840	 * Restore processor state that had been saved in vmcb01
 841	 */
 842	kvm_set_rflags(vcpu, svm->vmcb->save.rflags);
 843	svm_set_efer(vcpu, svm->vmcb->save.efer);
 844	svm_set_cr0(vcpu, svm->vmcb->save.cr0 | X86_CR0_PE);
 845	svm_set_cr4(vcpu, svm->vmcb->save.cr4);
 846	kvm_rax_write(vcpu, svm->vmcb->save.rax);
 847	kvm_rsp_write(vcpu, svm->vmcb->save.rsp);
 848	kvm_rip_write(vcpu, svm->vmcb->save.rip);
 849
 850	svm->vcpu.arch.dr7 = DR7_FIXED_1;
 851	kvm_update_dr7(&svm->vcpu);
 852
 853	trace_kvm_nested_vmexit_inject(vmcb12->control.exit_code,
 854				       vmcb12->control.exit_info_1,
 855				       vmcb12->control.exit_info_2,
 856				       vmcb12->control.exit_int_info,
 857				       vmcb12->control.exit_int_info_err,
 858				       KVM_ISA_SVM);
 859
 860	kvm_vcpu_unmap(vcpu, &map, true);
 861
 862	nested_svm_transition_tlb_flush(vcpu);
 863
 864	nested_svm_uninit_mmu_context(vcpu);
 865
 866	rc = nested_svm_load_cr3(vcpu, svm->vmcb->save.cr3, false, true);
 867	if (rc)
 868		return 1;
 869
 870	/*
 871	 * Drop what we picked up for L2 via svm_complete_interrupts() so it
 872	 * doesn't end up in L1.
 873	 */
 874	svm->vcpu.arch.nmi_injected = false;
 875	kvm_clear_exception_queue(vcpu);
 876	kvm_clear_interrupt_queue(vcpu);
 877
 878	/*
 879	 * If we are here following the completion of a VMRUN that
 880	 * is being single-stepped, queue the pending #DB intercept
 881	 * right now so that it an be accounted for before we execute
 882	 * L1's next instruction.
 883	 */
 884	if (unlikely(svm->vmcb->save.rflags & X86_EFLAGS_TF))
 885		kvm_queue_exception(&(svm->vcpu), DB_VECTOR);
 886
 887	return 0;
 888}
 889
 890static void nested_svm_triple_fault(struct kvm_vcpu *vcpu)
 891{
 892	nested_svm_simple_vmexit(to_svm(vcpu), SVM_EXIT_SHUTDOWN);
 893}
 894
 895int svm_allocate_nested(struct vcpu_svm *svm)
 896{
 897	struct page *vmcb02_page;
 898
 899	if (svm->nested.initialized)
 900		return 0;
 901
 902	vmcb02_page = alloc_page(GFP_KERNEL_ACCOUNT | __GFP_ZERO);
 903	if (!vmcb02_page)
 904		return -ENOMEM;
 905	svm->nested.vmcb02.ptr = page_address(vmcb02_page);
 906	svm->nested.vmcb02.pa = __sme_set(page_to_pfn(vmcb02_page) << PAGE_SHIFT);
 907
 908	svm->nested.msrpm = svm_vcpu_alloc_msrpm();
 909	if (!svm->nested.msrpm)
 910		goto err_free_vmcb02;
 911	svm_vcpu_init_msrpm(&svm->vcpu, svm->nested.msrpm);
 912
 913	svm->nested.initialized = true;
 914	return 0;
 915
 916err_free_vmcb02:
 917	__free_page(vmcb02_page);
 918	return -ENOMEM;
 919}
 920
 921void svm_free_nested(struct vcpu_svm *svm)
 922{
 923	if (!svm->nested.initialized)
 924		return;
 925
 926	svm_vcpu_free_msrpm(svm->nested.msrpm);
 927	svm->nested.msrpm = NULL;
 928
 929	__free_page(virt_to_page(svm->nested.vmcb02.ptr));
 930	svm->nested.vmcb02.ptr = NULL;
 931
 932	/*
 933	 * When last_vmcb12_gpa matches the current vmcb12 gpa,
 934	 * some vmcb12 fields are not loaded if they are marked clean
 935	 * in the vmcb12, since in this case they are up to date already.
 936	 *
 937	 * When the vmcb02 is freed, this optimization becomes invalid.
 938	 */
 939	svm->nested.last_vmcb12_gpa = INVALID_GPA;
 940
 941	svm->nested.initialized = false;
 942}
 943
 944/*
 945 * Forcibly leave nested mode in order to be able to reset the VCPU later on.
 946 */
 947void svm_leave_nested(struct vcpu_svm *svm)
 948{
 949	struct kvm_vcpu *vcpu = &svm->vcpu;
 950
 951	if (is_guest_mode(vcpu)) {
 952		svm->nested.nested_run_pending = 0;
 953		svm->nested.vmcb12_gpa = INVALID_GPA;
 954
 955		leave_guest_mode(vcpu);
 956
 957		svm_switch_vmcb(svm, &svm->vmcb01);
 958
 959		nested_svm_uninit_mmu_context(vcpu);
 960		vmcb_mark_all_dirty(svm->vmcb);
 961	}
 962
 963	kvm_clear_request(KVM_REQ_GET_NESTED_STATE_PAGES, vcpu);
 964}
 965
 966static int nested_svm_exit_handled_msr(struct vcpu_svm *svm)
 967{
 968	u32 offset, msr, value;
 969	int write, mask;
 970
 971	if (!(vmcb_is_intercept(&svm->nested.ctl, INTERCEPT_MSR_PROT)))
 972		return NESTED_EXIT_HOST;
 973
 974	msr    = svm->vcpu.arch.regs[VCPU_REGS_RCX];
 975	offset = svm_msrpm_offset(msr);
 976	write  = svm->vmcb->control.exit_info_1 & 1;
 977	mask   = 1 << ((2 * (msr & 0xf)) + write);
 978
 979	if (offset == MSR_INVALID)
 980		return NESTED_EXIT_DONE;
 981
 982	/* Offset is in 32 bit units but need in 8 bit units */
 983	offset *= 4;
 984
 985	if (kvm_vcpu_read_guest(&svm->vcpu, svm->nested.ctl.msrpm_base_pa + offset, &value, 4))
 986		return NESTED_EXIT_DONE;
 987
 988	return (value & mask) ? NESTED_EXIT_DONE : NESTED_EXIT_HOST;
 989}
 990
 991static int nested_svm_intercept_ioio(struct vcpu_svm *svm)
 992{
 993	unsigned port, size, iopm_len;
 994	u16 val, mask;
 995	u8 start_bit;
 996	u64 gpa;
 997
 998	if (!(vmcb_is_intercept(&svm->nested.ctl, INTERCEPT_IOIO_PROT)))
 999		return NESTED_EXIT_HOST;
1000
1001	port = svm->vmcb->control.exit_info_1 >> 16;
1002	size = (svm->vmcb->control.exit_info_1 & SVM_IOIO_SIZE_MASK) >>
1003		SVM_IOIO_SIZE_SHIFT;
1004	gpa  = svm->nested.ctl.iopm_base_pa + (port / 8);
1005	start_bit = port % 8;
1006	iopm_len = (start_bit + size > 8) ? 2 : 1;
1007	mask = (0xf >> (4 - size)) << start_bit;
1008	val = 0;
1009
1010	if (kvm_vcpu_read_guest(&svm->vcpu, gpa, &val, iopm_len))
1011		return NESTED_EXIT_DONE;
1012
1013	return (val & mask) ? NESTED_EXIT_DONE : NESTED_EXIT_HOST;
1014}
1015
1016static int nested_svm_intercept(struct vcpu_svm *svm)
1017{
1018	u32 exit_code = svm->vmcb->control.exit_code;
1019	int vmexit = NESTED_EXIT_HOST;
1020
1021	switch (exit_code) {
1022	case SVM_EXIT_MSR:
1023		vmexit = nested_svm_exit_handled_msr(svm);
1024		break;
1025	case SVM_EXIT_IOIO:
1026		vmexit = nested_svm_intercept_ioio(svm);
1027		break;
1028	case SVM_EXIT_READ_CR0 ... SVM_EXIT_WRITE_CR8: {
1029		if (vmcb_is_intercept(&svm->nested.ctl, exit_code))
1030			vmexit = NESTED_EXIT_DONE;
1031		break;
1032	}
1033	case SVM_EXIT_READ_DR0 ... SVM_EXIT_WRITE_DR7: {
1034		if (vmcb_is_intercept(&svm->nested.ctl, exit_code))
1035			vmexit = NESTED_EXIT_DONE;
1036		break;
1037	}
1038	case SVM_EXIT_EXCP_BASE ... SVM_EXIT_EXCP_BASE + 0x1f: {
1039		/*
1040		 * Host-intercepted exceptions have been checked already in
1041		 * nested_svm_exit_special.  There is nothing to do here,
1042		 * the vmexit is injected by svm_check_nested_events.
1043		 */
1044		vmexit = NESTED_EXIT_DONE;
1045		break;
1046	}
1047	case SVM_EXIT_ERR: {
1048		vmexit = NESTED_EXIT_DONE;
1049		break;
1050	}
1051	default: {
1052		if (vmcb_is_intercept(&svm->nested.ctl, exit_code))
1053			vmexit = NESTED_EXIT_DONE;
1054	}
1055	}
1056
1057	return vmexit;
1058}
1059
1060int nested_svm_exit_handled(struct vcpu_svm *svm)
1061{
1062	int vmexit;
1063
1064	vmexit = nested_svm_intercept(svm);
1065
1066	if (vmexit == NESTED_EXIT_DONE)
1067		nested_svm_vmexit(svm);
1068
1069	return vmexit;
1070}
1071
1072int nested_svm_check_permissions(struct kvm_vcpu *vcpu)
1073{
1074	if (!(vcpu->arch.efer & EFER_SVME) || !is_paging(vcpu)) {
1075		kvm_queue_exception(vcpu, UD_VECTOR);
1076		return 1;
1077	}
1078
1079	if (to_svm(vcpu)->vmcb->save.cpl) {
1080		kvm_inject_gp(vcpu, 0);
1081		return 1;
1082	}
1083
1084	return 0;
1085}
1086
1087static bool nested_exit_on_exception(struct vcpu_svm *svm)
1088{
1089	unsigned int nr = svm->vcpu.arch.exception.nr;
1090
1091	return (svm->nested.ctl.intercepts[INTERCEPT_EXCEPTION] & BIT(nr));
1092}
1093
1094static void nested_svm_inject_exception_vmexit(struct vcpu_svm *svm)
1095{
1096	unsigned int nr = svm->vcpu.arch.exception.nr;
1097
1098	svm->vmcb->control.exit_code = SVM_EXIT_EXCP_BASE + nr;
1099	svm->vmcb->control.exit_code_hi = 0;
1100
1101	if (svm->vcpu.arch.exception.has_error_code)
1102		svm->vmcb->control.exit_info_1 = svm->vcpu.arch.exception.error_code;
1103
1104	/*
1105	 * EXITINFO2 is undefined for all exception intercepts other
1106	 * than #PF.
1107	 */
1108	if (nr == PF_VECTOR) {
1109		if (svm->vcpu.arch.exception.nested_apf)
1110			svm->vmcb->control.exit_info_2 = svm->vcpu.arch.apf.nested_apf_token;
1111		else if (svm->vcpu.arch.exception.has_payload)
1112			svm->vmcb->control.exit_info_2 = svm->vcpu.arch.exception.payload;
1113		else
1114			svm->vmcb->control.exit_info_2 = svm->vcpu.arch.cr2;
1115	} else if (nr == DB_VECTOR) {
1116		/* See inject_pending_event.  */
1117		kvm_deliver_exception_payload(&svm->vcpu);
1118		if (svm->vcpu.arch.dr7 & DR7_GD) {
1119			svm->vcpu.arch.dr7 &= ~DR7_GD;
1120			kvm_update_dr7(&svm->vcpu);
1121		}
1122	} else
1123		WARN_ON(svm->vcpu.arch.exception.has_payload);
1124
1125	nested_svm_vmexit(svm);
1126}
1127
1128static inline bool nested_exit_on_init(struct vcpu_svm *svm)
1129{
1130	return vmcb_is_intercept(&svm->nested.ctl, INTERCEPT_INIT);
1131}
1132
1133static int svm_check_nested_events(struct kvm_vcpu *vcpu)
1134{
1135	struct vcpu_svm *svm = to_svm(vcpu);
1136	bool block_nested_events =
1137		kvm_event_needs_reinjection(vcpu) || svm->nested.nested_run_pending;
1138	struct kvm_lapic *apic = vcpu->arch.apic;
1139
1140	if (lapic_in_kernel(vcpu) &&
1141	    test_bit(KVM_APIC_INIT, &apic->pending_events)) {
1142		if (block_nested_events)
1143			return -EBUSY;
1144		if (!nested_exit_on_init(svm))
1145			return 0;
1146		nested_svm_simple_vmexit(svm, SVM_EXIT_INIT);
1147		return 0;
1148	}
1149
1150	if (vcpu->arch.exception.pending) {
1151		/*
1152		 * Only a pending nested run can block a pending exception.
1153		 * Otherwise an injected NMI/interrupt should either be
1154		 * lost or delivered to the nested hypervisor in the EXITINTINFO
1155		 * vmcb field, while delivering the pending exception.
1156		 */
1157		if (svm->nested.nested_run_pending)
1158                        return -EBUSY;
1159		if (!nested_exit_on_exception(svm))
1160			return 0;
1161		nested_svm_inject_exception_vmexit(svm);
1162		return 0;
1163	}
1164
1165	if (vcpu->arch.smi_pending && !svm_smi_blocked(vcpu)) {
1166		if (block_nested_events)
1167			return -EBUSY;
1168		if (!nested_exit_on_smi(svm))
1169			return 0;
1170		nested_svm_simple_vmexit(svm, SVM_EXIT_SMI);
1171		return 0;
1172	}
1173
1174	if (vcpu->arch.nmi_pending && !svm_nmi_blocked(vcpu)) {
1175		if (block_nested_events)
1176			return -EBUSY;
1177		if (!nested_exit_on_nmi(svm))
1178			return 0;
1179		nested_svm_simple_vmexit(svm, SVM_EXIT_NMI);
1180		return 0;
1181	}
1182
1183	if (kvm_cpu_has_interrupt(vcpu) && !svm_interrupt_blocked(vcpu)) {
1184		if (block_nested_events)
1185			return -EBUSY;
1186		if (!nested_exit_on_intr(svm))
1187			return 0;
1188		trace_kvm_nested_intr_vmexit(svm->vmcb->save.rip);
1189		nested_svm_simple_vmexit(svm, SVM_EXIT_INTR);
1190		return 0;
1191	}
1192
1193	return 0;
1194}
1195
1196int nested_svm_exit_special(struct vcpu_svm *svm)
1197{
1198	u32 exit_code = svm->vmcb->control.exit_code;
1199
1200	switch (exit_code) {
1201	case SVM_EXIT_INTR:
1202	case SVM_EXIT_NMI:
1203	case SVM_EXIT_NPF:
1204		return NESTED_EXIT_HOST;
1205	case SVM_EXIT_EXCP_BASE ... SVM_EXIT_EXCP_BASE + 0x1f: {
1206		u32 excp_bits = 1 << (exit_code - SVM_EXIT_EXCP_BASE);
1207
1208		if (svm->vmcb01.ptr->control.intercepts[INTERCEPT_EXCEPTION] &
1209		    excp_bits)
1210			return NESTED_EXIT_HOST;
1211		else if (exit_code == SVM_EXIT_EXCP_BASE + PF_VECTOR &&
1212			 svm->vcpu.arch.apf.host_apf_flags)
1213			/* Trap async PF even if not shadowing */
1214			return NESTED_EXIT_HOST;
1215		break;
1216	}
1217	default:
1218		break;
1219	}
1220
1221	return NESTED_EXIT_CONTINUE;
1222}
1223
1224static int svm_get_nested_state(struct kvm_vcpu *vcpu,
1225				struct kvm_nested_state __user *user_kvm_nested_state,
1226				u32 user_data_size)
1227{
1228	struct vcpu_svm *svm;
1229	struct kvm_nested_state kvm_state = {
1230		.flags = 0,
1231		.format = KVM_STATE_NESTED_FORMAT_SVM,
1232		.size = sizeof(kvm_state),
1233	};
1234	struct vmcb __user *user_vmcb = (struct vmcb __user *)
1235		&user_kvm_nested_state->data.svm[0];
1236
1237	if (!vcpu)
1238		return kvm_state.size + KVM_STATE_NESTED_SVM_VMCB_SIZE;
1239
1240	svm = to_svm(vcpu);
1241
1242	if (user_data_size < kvm_state.size)
1243		goto out;
1244
1245	/* First fill in the header and copy it out.  */
1246	if (is_guest_mode(vcpu)) {
1247		kvm_state.hdr.svm.vmcb_pa = svm->nested.vmcb12_gpa;
1248		kvm_state.size += KVM_STATE_NESTED_SVM_VMCB_SIZE;
1249		kvm_state.flags |= KVM_STATE_NESTED_GUEST_MODE;
1250
1251		if (svm->nested.nested_run_pending)
1252			kvm_state.flags |= KVM_STATE_NESTED_RUN_PENDING;
1253	}
1254
1255	if (gif_set(svm))
1256		kvm_state.flags |= KVM_STATE_NESTED_GIF_SET;
1257
1258	if (copy_to_user(user_kvm_nested_state, &kvm_state, sizeof(kvm_state)))
1259		return -EFAULT;
1260
1261	if (!is_guest_mode(vcpu))
1262		goto out;
1263
1264	/*
1265	 * Copy over the full size of the VMCB rather than just the size
1266	 * of the structs.
1267	 */
1268	if (clear_user(user_vmcb, KVM_STATE_NESTED_SVM_VMCB_SIZE))
1269		return -EFAULT;
1270	if (copy_to_user(&user_vmcb->control, &svm->nested.ctl,
1271			 sizeof(user_vmcb->control)))
1272		return -EFAULT;
1273	if (copy_to_user(&user_vmcb->save, &svm->vmcb01.ptr->save,
1274			 sizeof(user_vmcb->save)))
1275		return -EFAULT;
1276out:
1277	return kvm_state.size;
1278}
1279
1280static int svm_set_nested_state(struct kvm_vcpu *vcpu,
1281				struct kvm_nested_state __user *user_kvm_nested_state,
1282				struct kvm_nested_state *kvm_state)
1283{
1284	struct vcpu_svm *svm = to_svm(vcpu);
1285	struct vmcb __user *user_vmcb = (struct vmcb __user *)
1286		&user_kvm_nested_state->data.svm[0];
1287	struct vmcb_control_area *ctl;
1288	struct vmcb_save_area *save;
1289	unsigned long cr0;
1290	int ret;
1291
1292	BUILD_BUG_ON(sizeof(struct vmcb_control_area) + sizeof(struct vmcb_save_area) >
1293		     KVM_STATE_NESTED_SVM_VMCB_SIZE);
1294
1295	if (kvm_state->format != KVM_STATE_NESTED_FORMAT_SVM)
1296		return -EINVAL;
1297
1298	if (kvm_state->flags & ~(KVM_STATE_NESTED_GUEST_MODE |
1299				 KVM_STATE_NESTED_RUN_PENDING |
1300				 KVM_STATE_NESTED_GIF_SET))
1301		return -EINVAL;
1302
1303	/*
1304	 * If in guest mode, vcpu->arch.efer actually refers to the L2 guest's
1305	 * EFER.SVME, but EFER.SVME still has to be 1 for VMRUN to succeed.
1306	 */
1307	if (!(vcpu->arch.efer & EFER_SVME)) {
1308		/* GIF=1 and no guest mode are required if SVME=0.  */
1309		if (kvm_state->flags != KVM_STATE_NESTED_GIF_SET)
1310			return -EINVAL;
1311	}
1312
1313	/* SMM temporarily disables SVM, so we cannot be in guest mode.  */
1314	if (is_smm(vcpu) && (kvm_state->flags & KVM_STATE_NESTED_GUEST_MODE))
1315		return -EINVAL;
1316
1317	if (!(kvm_state->flags & KVM_STATE_NESTED_GUEST_MODE)) {
1318		svm_leave_nested(svm);
1319		svm_set_gif(svm, !!(kvm_state->flags & KVM_STATE_NESTED_GIF_SET));
1320		return 0;
1321	}
1322
1323	if (!page_address_valid(vcpu, kvm_state->hdr.svm.vmcb_pa))
1324		return -EINVAL;
1325	if (kvm_state->size < sizeof(*kvm_state) + KVM_STATE_NESTED_SVM_VMCB_SIZE)
1326		return -EINVAL;
1327
1328	ret  = -ENOMEM;
1329	ctl  = kzalloc(sizeof(*ctl),  GFP_KERNEL_ACCOUNT);
1330	save = kzalloc(sizeof(*save), GFP_KERNEL_ACCOUNT);
1331	if (!ctl || !save)
1332		goto out_free;
1333
1334	ret = -EFAULT;
1335	if (copy_from_user(ctl, &user_vmcb->control, sizeof(*ctl)))
1336		goto out_free;
1337	if (copy_from_user(save, &user_vmcb->save, sizeof(*save)))
1338		goto out_free;
1339
1340	ret = -EINVAL;
1341	if (!nested_vmcb_check_controls(vcpu, ctl))
1342		goto out_free;
1343
1344	/*
1345	 * Processor state contains L2 state.  Check that it is
1346	 * valid for guest mode (see nested_vmcb_check_save).
1347	 */
1348	cr0 = kvm_read_cr0(vcpu);
1349        if (((cr0 & X86_CR0_CD) == 0) && (cr0 & X86_CR0_NW))
1350		goto out_free;
1351
1352	/*
1353	 * Validate host state saved from before VMRUN (see
1354	 * nested_svm_check_permissions).
1355	 */
1356	if (!(save->cr0 & X86_CR0_PG) ||
1357	    !(save->cr0 & X86_CR0_PE) ||
1358	    (save->rflags & X86_EFLAGS_VM) ||
1359	    !nested_vmcb_valid_sregs(vcpu, save))
1360		goto out_free;
1361
1362	/*
1363	 * While the nested guest CR3 is already checked and set by
1364	 * KVM_SET_SREGS, it was set when nested state was yet loaded,
1365	 * thus MMU might not be initialized correctly.
1366	 * Set it again to fix this.
1367	 */
1368
1369	ret = nested_svm_load_cr3(&svm->vcpu, vcpu->arch.cr3,
1370				  nested_npt_enabled(svm), false);
1371	if (WARN_ON_ONCE(ret))
1372		goto out_free;
1373
1374
1375	/*
1376	 * All checks done, we can enter guest mode. Userspace provides
1377	 * vmcb12.control, which will be combined with L1 and stored into
1378	 * vmcb02, and the L1 save state which we store in vmcb01.
1379	 * L2 registers if needed are moved from the current VMCB to VMCB02.
1380	 */
1381
1382	if (is_guest_mode(vcpu))
1383		svm_leave_nested(svm);
1384	else
1385		svm->nested.vmcb02.ptr->save = svm->vmcb01.ptr->save;
1386
1387	svm_set_gif(svm, !!(kvm_state->flags & KVM_STATE_NESTED_GIF_SET));
1388
1389	svm->nested.nested_run_pending =
1390		!!(kvm_state->flags & KVM_STATE_NESTED_RUN_PENDING);
1391
1392	svm->nested.vmcb12_gpa = kvm_state->hdr.svm.vmcb_pa;
1393
1394	svm_copy_vmrun_state(&svm->vmcb01.ptr->save, save);
1395	nested_load_control_from_vmcb12(svm, ctl);
1396
1397	svm_switch_vmcb(svm, &svm->nested.vmcb02);
1398	nested_vmcb02_prepare_control(svm);
1399	kvm_make_request(KVM_REQ_GET_NESTED_STATE_PAGES, vcpu);
1400	ret = 0;
1401out_free:
1402	kfree(save);
1403	kfree(ctl);
1404
1405	return ret;
1406}
1407
1408static bool svm_get_nested_state_pages(struct kvm_vcpu *vcpu)
1409{
1410	struct vcpu_svm *svm = to_svm(vcpu);
1411
1412	if (WARN_ON(!is_guest_mode(vcpu)))
1413		return true;
1414
1415	if (!vcpu->arch.pdptrs_from_userspace &&
1416	    !nested_npt_enabled(svm) && is_pae_paging(vcpu))
1417		/*
1418		 * Reload the guest's PDPTRs since after a migration
1419		 * the guest CR3 might be restored prior to setting the nested
1420		 * state which can lead to a load of wrong PDPTRs.
1421		 */
1422		if (CC(!load_pdptrs(vcpu, vcpu->arch.walk_mmu, vcpu->arch.cr3)))
1423			return false;
1424
1425	if (!nested_svm_vmrun_msrpm(svm)) {
1426		vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
1427		vcpu->run->internal.suberror =
1428			KVM_INTERNAL_ERROR_EMULATION;
1429		vcpu->run->internal.ndata = 0;
1430		return false;
1431	}
1432
1433	return true;
1434}
1435
1436struct kvm_x86_nested_ops svm_nested_ops = {
1437	.check_events = svm_check_nested_events,
1438	.triple_fault = nested_svm_triple_fault,
1439	.get_nested_state_pages = svm_get_nested_state_pages,
1440	.get_state = svm_get_nested_state,
1441	.set_state = svm_set_nested_state,
1442};