arch/x86/kvm/svm/nested.c

   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
  23 #include "kvm_emulate.h"
  24 #include "trace.h"
  25 #include "mmu.h"
  26 #include "x86.h"
  27 #include "svm.h"
  28
  29 static void nested_svm_inject_npf_exit(struct kvm_vcpu *vcpu,
  30                                        struct x86_exception *fault)
  31 {
  32         struct vcpu_svm *svm = to_svm(vcpu);
  33
  34         if (svm->vmcb->control.exit_code != SVM_EXIT_NPF) {
  35                 /*
  36                  * TODO: track the cause of the nested page fault, and
  37                  * correctly fill in the high bits of exit_info_1.
  38                  */
  39                 svm->vmcb->control.exit_code = SVM_EXIT_NPF;
  40                 svm->vmcb->control.exit_code_hi = 0;
  41                 svm->vmcb->control.exit_info_1 = (1ULL << 32);
  42                 svm->vmcb->control.exit_info_2 = fault->address;
  43         }
  44
  45         svm->vmcb->control.exit_info_1 &= ~0xffffffffULL;
  46         svm->vmcb->control.exit_info_1 |= fault->error_code;
  47
  48         /*
  49          * The present bit is always zero for page structure faults on real
  50          * hardware.
  51          */
  52         if (svm->vmcb->control.exit_info_1 & (2ULL << 32))
  53                 svm->vmcb->control.exit_info_1 &= ~1;
  54
  55         nested_svm_vmexit(svm);
  56 }
  57
  58 static u64 nested_svm_get_tdp_pdptr(struct kvm_vcpu *vcpu, int index)
  59 {
  60         struct vcpu_svm *svm = to_svm(vcpu);
  61         u64 cr3 = svm->nested.nested_cr3;
  62         u64 pdpte;
  63         int ret;
  64
  65         ret = kvm_vcpu_read_guest_page(vcpu, gpa_to_gfn(__sme_clr(cr3)), &pdpte,
  66                                        offset_in_page(cr3) + index * 8, 8);
  67         if (ret)
  68                 return 0;
  69         return pdpte;
  70 }
  71
  72 static unsigned long nested_svm_get_tdp_cr3(struct kvm_vcpu *vcpu)
  73 {
  74         struct vcpu_svm *svm = to_svm(vcpu);
  75
  76         return svm->nested.nested_cr3;
  77 }
  78
  79 static void nested_svm_init_mmu_context(struct kvm_vcpu *vcpu)
  80 {
  81         WARN_ON(mmu_is_nested(vcpu));
  82
  83         vcpu->arch.mmu = &vcpu->arch.guest_mmu;
  84         kvm_init_shadow_mmu(vcpu);
  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         vcpu->arch.mmu->shadow_root_level = kvm_x86_ops.get_tdp_level(vcpu);
  89         reset_shadow_zero_bits_mask(vcpu, vcpu->arch.mmu);
  90         vcpu->arch.walk_mmu              = &vcpu->arch.nested_mmu;
  91 }
  92
  93 static void nested_svm_uninit_mmu_context(struct kvm_vcpu *vcpu)
  94 {
  95         vcpu->arch.mmu = &vcpu->arch.root_mmu;
  96         vcpu->arch.walk_mmu = &vcpu->arch.root_mmu;
  97 }
  98
  99 void recalc_intercepts(struct vcpu_svm *svm)
 100 {
 101         struct vmcb_control_area *c, *h;
 102         struct nested_state *g;
 103
 104         mark_dirty(svm->vmcb, VMCB_INTERCEPTS);
 105
 106         if (!is_guest_mode(&svm->vcpu))
 107                 return;
 108
 109         c = &svm->vmcb->control;
 110         h = &svm->nested.hsave->control;
 111         g = &svm->nested;
 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 (svm->vcpu.arch.hflags & HF_VINTR_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
 140 static void copy_vmcb_control_area(struct vmcb *dst_vmcb, struct vmcb *from_vmcb)
 141 {
 142         struct vmcb_control_area *dst  = &dst_vmcb->control;
 143         struct vmcb_control_area *from = &from_vmcb->control;
 144
 145         dst->intercept_cr         = from->intercept_cr;
 146         dst->intercept_dr         = from->intercept_dr;
 147         dst->intercept_exceptions = from->intercept_exceptions;
 148         dst->intercept            = from->intercept;
 149         dst->iopm_base_pa         = from->iopm_base_pa;
 150         dst->msrpm_base_pa        = from->msrpm_base_pa;
 151         dst->tsc_offset           = from->tsc_offset;
 152         dst->asid                 = from->asid;
 153         dst->tlb_ctl              = from->tlb_ctl;
 154         dst->int_ctl              = from->int_ctl;
 155         dst->int_vector           = from->int_vector;
 156         dst->int_state            = from->int_state;
 157         dst->exit_code            = from->exit_code;
 158         dst->exit_code_hi         = from->exit_code_hi;
 159         dst->exit_info_1          = from->exit_info_1;
 160         dst->exit_info_2          = from->exit_info_2;
 161         dst->exit_int_info        = from->exit_int_info;
 162         dst->exit_int_info_err    = from->exit_int_info_err;
 163         dst->nested_ctl           = from->nested_ctl;
 164         dst->event_inj            = from->event_inj;
 165         dst->event_inj_err        = from->event_inj_err;
 166         dst->nested_cr3           = from->nested_cr3;
 167         dst->virt_ext              = from->virt_ext;
 168         dst->pause_filter_count   = from->pause_filter_count;
 169         dst->pause_filter_thresh  = from->pause_filter_thresh;
 170 }
 171
 172 static bool nested_svm_vmrun_msrpm(struct vcpu_svm *svm)
 173 {
 174         /*
 175          * This function merges the msr permission bitmaps of kvm and the
 176          * nested vmcb. It is optimized in that it only merges the parts where
 177          * the kvm msr permission bitmap may contain zero bits
 178          */
 179         int i;
 180
 181         if (!(svm->nested.intercept & (1ULL << INTERCEPT_MSR_PROT)))
 182                 return true;
 183
 184         for (i = 0; i < MSRPM_OFFSETS; i++) {
 185                 u32 value, p;
 186                 u64 offset;
 187
 188                 if (msrpm_offsets[i] == 0xffffffff)
 189                         break;
 190
 191                 p      = msrpm_offsets[i];
 192                 offset = svm->nested.vmcb_msrpm + (p * 4);
 193
 194                 if (kvm_vcpu_read_guest(&svm->vcpu, offset, &value, 4))
 195                         return false;
 196
 197                 svm->nested.msrpm[p] = svm->msrpm[p] | value;
 198         }
 199
 200         svm->vmcb->control.msrpm_base_pa = __sme_set(__pa(svm->nested.msrpm));
 201
 202         return true;
 203 }
 204
 205 static bool nested_vmcb_checks(struct vmcb *vmcb)
 206 {
 207         if ((vmcb->save.efer & EFER_SVME) == 0)
 208                 return false;
 209
 210         if ((vmcb->control.intercept & (1ULL << INTERCEPT_VMRUN)) == 0)
 211                 return false;
 212
 213         if (vmcb->control.asid == 0)
 214                 return false;
 215
 216         if ((vmcb->control.nested_ctl & SVM_NESTED_CTL_NP_ENABLE) &&
 217             !npt_enabled)
 218                 return false;
 219
 220         return true;
 221 }
 222
 223 void enter_svm_guest_mode(struct vcpu_svm *svm, u64 vmcb_gpa,
 224                           struct vmcb *nested_vmcb, struct kvm_host_map *map)
 225 {
 226         bool evaluate_pending_interrupts =
 227                 is_intercept(svm, INTERCEPT_VINTR) ||
 228                 is_intercept(svm, INTERCEPT_IRET);
 229
 230         if (kvm_get_rflags(&svm->vcpu) & X86_EFLAGS_IF)
 231                 svm->vcpu.arch.hflags |= HF_HIF_MASK;
 232         else
 233                 svm->vcpu.arch.hflags &= ~HF_HIF_MASK;
 234
 235         if (nested_vmcb->control.nested_ctl & SVM_NESTED_CTL_NP_ENABLE) {
 236                 svm->nested.nested_cr3 = nested_vmcb->control.nested_cr3;
 237                 nested_svm_init_mmu_context(&svm->vcpu);
 238         }
 239
 240         /* Load the nested guest state */
 241         svm->vmcb->save.es = nested_vmcb->save.es;
 242         svm->vmcb->save.cs = nested_vmcb->save.cs;
 243         svm->vmcb->save.ss = nested_vmcb->save.ss;
 244         svm->vmcb->save.ds = nested_vmcb->save.ds;
 245         svm->vmcb->save.gdtr = nested_vmcb->save.gdtr;
 246         svm->vmcb->save.idtr = nested_vmcb->save.idtr;
 247         kvm_set_rflags(&svm->vcpu, nested_vmcb->save.rflags);
 248         svm_set_efer(&svm->vcpu, nested_vmcb->save.efer);
 249         svm_set_cr0(&svm->vcpu, nested_vmcb->save.cr0);
 250         svm_set_cr4(&svm->vcpu, nested_vmcb->save.cr4);
 251         if (npt_enabled) {
 252                 svm->vmcb->save.cr3 = nested_vmcb->save.cr3;
 253                 svm->vcpu.arch.cr3 = nested_vmcb->save.cr3;
 254         } else
 255                 (void)kvm_set_cr3(&svm->vcpu, nested_vmcb->save.cr3);
 256
 257         /* Guest paging mode is active - reset mmu */
 258         kvm_mmu_reset_context(&svm->vcpu);
 259
 260         svm->vmcb->save.cr2 = svm->vcpu.arch.cr2 = nested_vmcb->save.cr2;
 261         kvm_rax_write(&svm->vcpu, nested_vmcb->save.rax);
 262         kvm_rsp_write(&svm->vcpu, nested_vmcb->save.rsp);
 263         kvm_rip_write(&svm->vcpu, nested_vmcb->save.rip);
 264
 265         /* In case we don't even reach vcpu_run, the fields are not updated */
 266         svm->vmcb->save.rax = nested_vmcb->save.rax;
 267         svm->vmcb->save.rsp = nested_vmcb->save.rsp;
 268         svm->vmcb->save.rip = nested_vmcb->save.rip;
 269         svm->vmcb->save.dr7 = nested_vmcb->save.dr7;
 270         svm->vmcb->save.dr6 = nested_vmcb->save.dr6;
 271         svm->vmcb->save.cpl = nested_vmcb->save.cpl;
 272
 273         svm->nested.vmcb_msrpm = nested_vmcb->control.msrpm_base_pa & ~0x0fffULL;
 274         svm->nested.vmcb_iopm  = nested_vmcb->control.iopm_base_pa  & ~0x0fffULL;
 275
 276         /* cache intercepts */
 277         svm->nested.intercept_cr         = nested_vmcb->control.intercept_cr;
 278         svm->nested.intercept_dr         = nested_vmcb->control.intercept_dr;
 279         svm->nested.intercept_exceptions = nested_vmcb->control.intercept_exceptions;
 280         svm->nested.intercept            = nested_vmcb->control.intercept;
 281
 282         svm_flush_tlb(&svm->vcpu, true);
 283         svm->vmcb->control.int_ctl = nested_vmcb->control.int_ctl | V_INTR_MASKING_MASK;
 284         if (nested_vmcb->control.int_ctl & V_INTR_MASKING_MASK)
 285                 svm->vcpu.arch.hflags |= HF_VINTR_MASK;
 286         else
 287                 svm->vcpu.arch.hflags &= ~HF_VINTR_MASK;
 288
 289         svm->vcpu.arch.tsc_offset += nested_vmcb->control.tsc_offset;
 290         svm->vmcb->control.tsc_offset = svm->vcpu.arch.tsc_offset;
 291
 292         svm->vmcb->control.virt_ext = nested_vmcb->control.virt_ext;
 293         svm->vmcb->control.int_vector = nested_vmcb->control.int_vector;
 294         svm->vmcb->control.int_state = nested_vmcb->control.int_state;
 295         svm->vmcb->control.event_inj = nested_vmcb->control.event_inj;
 296         svm->vmcb->control.event_inj_err = nested_vmcb->control.event_inj_err;
 297
 298         svm->vmcb->control.pause_filter_count =
 299                 nested_vmcb->control.pause_filter_count;
 300         svm->vmcb->control.pause_filter_thresh =
 301                 nested_vmcb->control.pause_filter_thresh;
 302
 303         kvm_vcpu_unmap(&svm->vcpu, map, true);
 304
 305         /* Enter Guest-Mode */
 306         enter_guest_mode(&svm->vcpu);
 307
 308         /*
 309          * Merge guest and host intercepts - must be called  with vcpu in
 310          * guest-mode to take affect here
 311          */
 312         recalc_intercepts(svm);
 313
 314         svm->nested.vmcb = vmcb_gpa;
 315
 316         /*
 317          * If L1 had a pending IRQ/NMI before executing VMRUN,
 318          * which wasn't delivered because it was disallowed (e.g.
 319          * interrupts disabled), L0 needs to evaluate if this pending
 320          * event should cause an exit from L2 to L1 or be delivered
 321          * directly to L2.
 322          *
 323          * Usually this would be handled by the processor noticing an
 324          * IRQ/NMI window request.  However, VMRUN can unblock interrupts
 325          * by implicitly setting GIF, so force L0 to perform pending event
 326          * evaluation by requesting a KVM_REQ_EVENT.
 327          */
 328         enable_gif(svm);
 329         if (unlikely(evaluate_pending_interrupts))
 330                 kvm_make_request(KVM_REQ_EVENT, &svm->vcpu);
 331
 332         mark_all_dirty(svm->vmcb);
 333 }
 334
 335 int nested_svm_vmrun(struct vcpu_svm *svm)
 336 {
 337         int ret;
 338         struct vmcb *nested_vmcb;
 339         struct vmcb *hsave = svm->nested.hsave;
 340         struct vmcb *vmcb = svm->vmcb;
 341         struct kvm_host_map map;
 342         u64 vmcb_gpa;
 343
 344         vmcb_gpa = svm->vmcb->save.rax;
 345
 346         ret = kvm_vcpu_map(&svm->vcpu, gpa_to_gfn(vmcb_gpa), &map);
 347         if (ret == -EINVAL) {
 348                 kvm_inject_gp(&svm->vcpu, 0);
 349                 return 1;
 350         } else if (ret) {
 351                 return kvm_skip_emulated_instruction(&svm->vcpu);
 352         }
 353
 354         ret = kvm_skip_emulated_instruction(&svm->vcpu);
 355
 356         nested_vmcb = map.hva;
 357
 358         if (!nested_vmcb_checks(nested_vmcb)) {
 359                 nested_vmcb->control.exit_code    = SVM_EXIT_ERR;
 360                 nested_vmcb->control.exit_code_hi = 0;
 361                 nested_vmcb->control.exit_info_1  = 0;
 362                 nested_vmcb->control.exit_info_2  = 0;
 363
 364                 kvm_vcpu_unmap(&svm->vcpu, &map, true);
 365
 366                 return ret;
 367         }
 368
 369         trace_kvm_nested_vmrun(svm->vmcb->save.rip, vmcb_gpa,
 370                                nested_vmcb->save.rip,
 371                                nested_vmcb->control.int_ctl,
 372                                nested_vmcb->control.event_inj,
 373                                nested_vmcb->control.nested_ctl);
 374
 375         trace_kvm_nested_intercepts(nested_vmcb->control.intercept_cr & 0xffff,
 376                                     nested_vmcb->control.intercept_cr >> 16,
 377                                     nested_vmcb->control.intercept_exceptions,
 378                                     nested_vmcb->control.intercept);
 379
 380         /* Clear internal status */
 381         kvm_clear_exception_queue(&svm->vcpu);
 382         kvm_clear_interrupt_queue(&svm->vcpu);
 383
 384         /*
 385          * Save the old vmcb, so we don't need to pick what we save, but can
 386          * restore everything when a VMEXIT occurs
 387          */
 388         hsave->save.es     = vmcb->save.es;
 389         hsave->save.cs     = vmcb->save.cs;
 390         hsave->save.ss     = vmcb->save.ss;
 391         hsave->save.ds     = vmcb->save.ds;
 392         hsave->save.gdtr   = vmcb->save.gdtr;
 393         hsave->save.idtr   = vmcb->save.idtr;
 394         hsave->save.efer   = svm->vcpu.arch.efer;
 395         hsave->save.cr0    = kvm_read_cr0(&svm->vcpu);
 396         hsave->save.cr4    = svm->vcpu.arch.cr4;
 397         hsave->save.rflags = kvm_get_rflags(&svm->vcpu);
 398         hsave->save.rip    = kvm_rip_read(&svm->vcpu);
 399         hsave->save.rsp    = vmcb->save.rsp;
 400         hsave->save.rax    = vmcb->save.rax;
 401         if (npt_enabled)
 402                 hsave->save.cr3    = vmcb->save.cr3;
 403         else
 404                 hsave->save.cr3    = kvm_read_cr3(&svm->vcpu);
 405
 406         copy_vmcb_control_area(hsave, vmcb);
 407
 408         enter_svm_guest_mode(svm, vmcb_gpa, nested_vmcb, &map);
 409
 410         if (!nested_svm_vmrun_msrpm(svm)) {
 411                 svm->vmcb->control.exit_code    = SVM_EXIT_ERR;
 412                 svm->vmcb->control.exit_code_hi = 0;
 413                 svm->vmcb->control.exit_info_1  = 0;
 414                 svm->vmcb->control.exit_info_2  = 0;
 415
 416                 nested_svm_vmexit(svm);
 417         }
 418
 419         return ret;
 420 }
 421
 422 void nested_svm_vmloadsave(struct vmcb *from_vmcb, struct vmcb *to_vmcb)
 423 {
 424         to_vmcb->save.fs = from_vmcb->save.fs;
 425         to_vmcb->save.gs = from_vmcb->save.gs;
 426         to_vmcb->save.tr = from_vmcb->save.tr;
 427         to_vmcb->save.ldtr = from_vmcb->save.ldtr;
 428         to_vmcb->save.kernel_gs_base = from_vmcb->save.kernel_gs_base;
 429         to_vmcb->save.star = from_vmcb->save.star;
 430         to_vmcb->save.lstar = from_vmcb->save.lstar;
 431         to_vmcb->save.cstar = from_vmcb->save.cstar;
 432         to_vmcb->save.sfmask = from_vmcb->save.sfmask;
 433         to_vmcb->save.sysenter_cs = from_vmcb->save.sysenter_cs;
 434         to_vmcb->save.sysenter_esp = from_vmcb->save.sysenter_esp;
 435         to_vmcb->save.sysenter_eip = from_vmcb->save.sysenter_eip;
 436 }
 437
 438 int nested_svm_vmexit(struct vcpu_svm *svm)
 439 {
 440         int rc;
 441         struct vmcb *nested_vmcb;
 442         struct vmcb *hsave = svm->nested.hsave;
 443         struct vmcb *vmcb = svm->vmcb;
 444         struct kvm_host_map map;
 445
 446         trace_kvm_nested_vmexit_inject(vmcb->control.exit_code,
 447                                        vmcb->control.exit_info_1,
 448                                        vmcb->control.exit_info_2,
 449                                        vmcb->control.exit_int_info,
 450                                        vmcb->control.exit_int_info_err,
 451                                        KVM_ISA_SVM);
 452
 453         rc = kvm_vcpu_map(&svm->vcpu, gpa_to_gfn(svm->nested.vmcb), &map);
 454         if (rc) {
 455                 if (rc == -EINVAL)
 456                         kvm_inject_gp(&svm->vcpu, 0);
 457                 return 1;
 458         }
 459
 460         nested_vmcb = map.hva;
 461
 462         /* Exit Guest-Mode */
 463         leave_guest_mode(&svm->vcpu);
 464         svm->nested.vmcb = 0;
 465
 466         /* Give the current vmcb to the guest */
 467         disable_gif(svm);
 468
 469         nested_vmcb->save.es     = vmcb->save.es;
 470         nested_vmcb->save.cs     = vmcb->save.cs;
 471         nested_vmcb->save.ss     = vmcb->save.ss;
 472         nested_vmcb->save.ds     = vmcb->save.ds;
 473         nested_vmcb->save.gdtr   = vmcb->save.gdtr;
 474         nested_vmcb->save.idtr   = vmcb->save.idtr;
 475         nested_vmcb->save.efer   = svm->vcpu.arch.efer;
 476         nested_vmcb->save.cr0    = kvm_read_cr0(&svm->vcpu);
 477         nested_vmcb->save.cr3    = kvm_read_cr3(&svm->vcpu);
 478         nested_vmcb->save.cr2    = vmcb->save.cr2;
 479         nested_vmcb->save.cr4    = svm->vcpu.arch.cr4;
 480         nested_vmcb->save.rflags = kvm_get_rflags(&svm->vcpu);
 481         nested_vmcb->save.rip    = vmcb->save.rip;
 482         nested_vmcb->save.rsp    = vmcb->save.rsp;
 483         nested_vmcb->save.rax    = vmcb->save.rax;
 484         nested_vmcb->save.dr7    = vmcb->save.dr7;
 485         nested_vmcb->save.dr6    = vmcb->save.dr6;
 486         nested_vmcb->save.cpl    = vmcb->save.cpl;
 487
 488         nested_vmcb->control.int_ctl           = vmcb->control.int_ctl;
 489         nested_vmcb->control.int_vector        = vmcb->control.int_vector;
 490         nested_vmcb->control.int_state         = vmcb->control.int_state;
 491         nested_vmcb->control.exit_code         = vmcb->control.exit_code;
 492         nested_vmcb->control.exit_code_hi      = vmcb->control.exit_code_hi;
 493         nested_vmcb->control.exit_info_1       = vmcb->control.exit_info_1;
 494         nested_vmcb->control.exit_info_2       = vmcb->control.exit_info_2;
 495         nested_vmcb->control.exit_int_info     = vmcb->control.exit_int_info;
 496         nested_vmcb->control.exit_int_info_err = vmcb->control.exit_int_info_err;
 497
 498         if (svm->nrips_enabled)
 499                 nested_vmcb->control.next_rip  = vmcb->control.next_rip;
 500
 501         /*
 502          * If we emulate a VMRUN/#VMEXIT in the same host #vmexit cycle we have
 503          * to make sure that we do not lose injected events. So check event_inj
 504          * here and copy it to exit_int_info if it is valid.
 505          * Exit_int_info and event_inj can't be both valid because the case
 506          * below only happens on a VMRUN instruction intercept which has
 507          * no valid exit_int_info set.
 508          */
 509         if (vmcb->control.event_inj & SVM_EVTINJ_VALID) {
 510                 struct vmcb_control_area *nc = &nested_vmcb->control;
 511
 512                 nc->exit_int_info     = vmcb->control.event_inj;
 513                 nc->exit_int_info_err = vmcb->control.event_inj_err;
 514         }
 515
 516         nested_vmcb->control.tlb_ctl           = 0;
 517         nested_vmcb->control.event_inj         = 0;
 518         nested_vmcb->control.event_inj_err     = 0;
 519
 520         nested_vmcb->control.pause_filter_count =
 521                 svm->vmcb->control.pause_filter_count;
 522         nested_vmcb->control.pause_filter_thresh =
 523                 svm->vmcb->control.pause_filter_thresh;
 524
 525         /* We always set V_INTR_MASKING and remember the old value in hflags */
 526         if (!(svm->vcpu.arch.hflags & HF_VINTR_MASK))
 527                 nested_vmcb->control.int_ctl &= ~V_INTR_MASKING_MASK;
 528
 529         /* Restore the original control entries */
 530         copy_vmcb_control_area(vmcb, hsave);
 531
 532         svm->vcpu.arch.tsc_offset = svm->vmcb->control.tsc_offset;
 533         kvm_clear_exception_queue(&svm->vcpu);
 534         kvm_clear_interrupt_queue(&svm->vcpu);
 535
 536         svm->nested.nested_cr3 = 0;
 537
 538         /* Restore selected save entries */
 539         svm->vmcb->save.es = hsave->save.es;
 540         svm->vmcb->save.cs = hsave->save.cs;
 541         svm->vmcb->save.ss = hsave->save.ss;
 542         svm->vmcb->save.ds = hsave->save.ds;
 543         svm->vmcb->save.gdtr = hsave->save.gdtr;
 544         svm->vmcb->save.idtr = hsave->save.idtr;
 545         kvm_set_rflags(&svm->vcpu, hsave->save.rflags);
 546         svm_set_efer(&svm->vcpu, hsave->save.efer);
 547         svm_set_cr0(&svm->vcpu, hsave->save.cr0 | X86_CR0_PE);
 548         svm_set_cr4(&svm->vcpu, hsave->save.cr4);
 549         if (npt_enabled) {
 550                 svm->vmcb->save.cr3 = hsave->save.cr3;
 551                 svm->vcpu.arch.cr3 = hsave->save.cr3;
 552         } else {
 553                 (void)kvm_set_cr3(&svm->vcpu, hsave->save.cr3);
 554         }
 555         kvm_rax_write(&svm->vcpu, hsave->save.rax);
 556         kvm_rsp_write(&svm->vcpu, hsave->save.rsp);
 557         kvm_rip_write(&svm->vcpu, hsave->save.rip);
 558         svm->vmcb->save.dr7 = 0;
 559         svm->vmcb->save.cpl = 0;
 560         svm->vmcb->control.exit_int_info = 0;
 561
 562         mark_all_dirty(svm->vmcb);
 563
 564         kvm_vcpu_unmap(&svm->vcpu, &map, true);
 565
 566         nested_svm_uninit_mmu_context(&svm->vcpu);
 567         kvm_mmu_reset_context(&svm->vcpu);
 568         kvm_mmu_load(&svm->vcpu);
 569
 570         /*
 571          * Drop what we picked up for L2 via svm_complete_interrupts() so it
 572          * doesn't end up in L1.
 573          */
 574         svm->vcpu.arch.nmi_injected = false;
 575         kvm_clear_exception_queue(&svm->vcpu);
 576         kvm_clear_interrupt_queue(&svm->vcpu);
 577
 578         return 0;
 579 }
 580
 581 static int nested_svm_exit_handled_msr(struct vcpu_svm *svm)
 582 {
 583         u32 offset, msr, value;
 584         int write, mask;
 585
 586         if (!(svm->nested.intercept & (1ULL << INTERCEPT_MSR_PROT)))
 587                 return NESTED_EXIT_HOST;
 588
 589         msr    = svm->vcpu.arch.regs[VCPU_REGS_RCX];
 590         offset = svm_msrpm_offset(msr);
 591         write  = svm->vmcb->control.exit_info_1 & 1;
 592         mask   = 1 << ((2 * (msr & 0xf)) + write);
 593
 594         if (offset == MSR_INVALID)
 595                 return NESTED_EXIT_DONE;
 596
 597         /* Offset is in 32 bit units but need in 8 bit units */
 598         offset *= 4;
 599
 600         if (kvm_vcpu_read_guest(&svm->vcpu, svm->nested.vmcb_msrpm + offset, &value, 4))
 601                 return NESTED_EXIT_DONE;
 602
 603         return (value & mask) ? NESTED_EXIT_DONE : NESTED_EXIT_HOST;
 604 }
 605
 606 /* DB exceptions for our internal use must not cause vmexit */
 607 static int nested_svm_intercept_db(struct vcpu_svm *svm)
 608 {
 609         unsigned long dr6;
 610
 611         /* if we're not singlestepping, it's not ours */
 612         if (!svm->nmi_singlestep)
 613                 return NESTED_EXIT_DONE;
 614
 615         /* if it's not a singlestep exception, it's not ours */
 616         if (kvm_get_dr(&svm->vcpu, 6, &dr6))
 617                 return NESTED_EXIT_DONE;
 618         if (!(dr6 & DR6_BS))
 619                 return NESTED_EXIT_DONE;
 620
 621         /* if the guest is singlestepping, it should get the vmexit */
 622         if (svm->nmi_singlestep_guest_rflags & X86_EFLAGS_TF) {
 623                 disable_nmi_singlestep(svm);
 624                 return NESTED_EXIT_DONE;
 625         }
 626
 627         /* it's ours, the nested hypervisor must not see this one */
 628         return NESTED_EXIT_HOST;
 629 }
 630
 631 static int nested_svm_intercept_ioio(struct vcpu_svm *svm)
 632 {
 633         unsigned port, size, iopm_len;
 634         u16 val, mask;
 635         u8 start_bit;
 636         u64 gpa;
 637
 638         if (!(svm->nested.intercept & (1ULL << INTERCEPT_IOIO_PROT)))
 639                 return NESTED_EXIT_HOST;
 640
 641         port = svm->vmcb->control.exit_info_1 >> 16;
 642         size = (svm->vmcb->control.exit_info_1 & SVM_IOIO_SIZE_MASK) >>
 643                 SVM_IOIO_SIZE_SHIFT;
 644         gpa  = svm->nested.vmcb_iopm + (port / 8);
 645         start_bit = port % 8;
 646         iopm_len = (start_bit + size > 8) ? 2 : 1;
 647         mask = (0xf >> (4 - size)) << start_bit;
 648         val = 0;
 649
 650         if (kvm_vcpu_read_guest(&svm->vcpu, gpa, &val, iopm_len))
 651                 return NESTED_EXIT_DONE;
 652
 653         return (val & mask) ? NESTED_EXIT_DONE : NESTED_EXIT_HOST;
 654 }
 655
 656 static int nested_svm_intercept(struct vcpu_svm *svm)
 657 {
 658         u32 exit_code = svm->vmcb->control.exit_code;
 659         int vmexit = NESTED_EXIT_HOST;
 660
 661         switch (exit_code) {
 662         case SVM_EXIT_MSR:
 663                 vmexit = nested_svm_exit_handled_msr(svm);
 664                 break;
 665         case SVM_EXIT_IOIO:
 666                 vmexit = nested_svm_intercept_ioio(svm);
 667                 break;
 668         case SVM_EXIT_READ_CR0 ... SVM_EXIT_WRITE_CR8: {
 669                 u32 bit = 1U << (exit_code - SVM_EXIT_READ_CR0);
 670                 if (svm->nested.intercept_cr & bit)
 671                         vmexit = NESTED_EXIT_DONE;
 672                 break;
 673         }
 674         case SVM_EXIT_READ_DR0 ... SVM_EXIT_WRITE_DR7: {
 675                 u32 bit = 1U << (exit_code - SVM_EXIT_READ_DR0);
 676                 if (svm->nested.intercept_dr & bit)
 677                         vmexit = NESTED_EXIT_DONE;
 678                 break;
 679         }
 680         case SVM_EXIT_EXCP_BASE ... SVM_EXIT_EXCP_BASE + 0x1f: {
 681                 u32 excp_bits = 1 << (exit_code - SVM_EXIT_EXCP_BASE);
 682                 if (svm->nested.intercept_exceptions & excp_bits) {
 683                         if (exit_code == SVM_EXIT_EXCP_BASE + DB_VECTOR)
 684                                 vmexit = nested_svm_intercept_db(svm);
 685                         else
 686                                 vmexit = NESTED_EXIT_DONE;
 687                 }
 688                 /* async page fault always cause vmexit */
 689                 else if ((exit_code == SVM_EXIT_EXCP_BASE + PF_VECTOR) &&
 690                          svm->vcpu.arch.exception.nested_apf != 0)
 691                         vmexit = NESTED_EXIT_DONE;
 692                 break;
 693         }
 694         case SVM_EXIT_ERR: {
 695                 vmexit = NESTED_EXIT_DONE;
 696                 break;
 697         }
 698         default: {
 699                 u64 exit_bits = 1ULL << (exit_code - SVM_EXIT_INTR);
 700                 if (svm->nested.intercept & exit_bits)
 701                         vmexit = NESTED_EXIT_DONE;
 702         }
 703         }
 704
 705         return vmexit;
 706 }
 707
 708 int nested_svm_exit_handled(struct vcpu_svm *svm)
 709 {
 710         int vmexit;
 711
 712         vmexit = nested_svm_intercept(svm);
 713
 714         if (vmexit == NESTED_EXIT_DONE)
 715                 nested_svm_vmexit(svm);
 716
 717         return vmexit;
 718 }
 719
 720 int nested_svm_check_permissions(struct vcpu_svm *svm)
 721 {
 722         if (!(svm->vcpu.arch.efer & EFER_SVME) ||
 723             !is_paging(&svm->vcpu)) {
 724                 kvm_queue_exception(&svm->vcpu, UD_VECTOR);
 725                 return 1;
 726         }
 727
 728         if (svm->vmcb->save.cpl) {
 729                 kvm_inject_gp(&svm->vcpu, 0);
 730                 return 1;
 731         }
 732
 733         return 0;
 734 }
 735
 736 int nested_svm_check_exception(struct vcpu_svm *svm, unsigned nr,
 737                                bool has_error_code, u32 error_code)
 738 {
 739         int vmexit;
 740
 741         if (!is_guest_mode(&svm->vcpu))
 742                 return 0;
 743
 744         vmexit = nested_svm_intercept(svm);
 745         if (vmexit != NESTED_EXIT_DONE)
 746                 return 0;
 747
 748         svm->vmcb->control.exit_code = SVM_EXIT_EXCP_BASE + nr;
 749         svm->vmcb->control.exit_code_hi = 0;
 750         svm->vmcb->control.exit_info_1 = error_code;
 751
 752         /*
 753          * EXITINFO2 is undefined for all exception intercepts other
 754          * than #PF.
 755          */
 756         if (svm->vcpu.arch.exception.nested_apf)
 757                 svm->vmcb->control.exit_info_2 = svm->vcpu.arch.apf.nested_apf_token;
 758         else if (svm->vcpu.arch.exception.has_payload)
 759                 svm->vmcb->control.exit_info_2 = svm->vcpu.arch.exception.payload;
 760         else
 761                 svm->vmcb->control.exit_info_2 = svm->vcpu.arch.cr2;
 762
 763         svm->nested.exit_required = true;
 764         return vmexit;
 765 }
 766
 767 static void nested_svm_intr(struct vcpu_svm *svm)
 768 {
 769         svm->vmcb->control.exit_code   = SVM_EXIT_INTR;
 770         svm->vmcb->control.exit_info_1 = 0;
 771         svm->vmcb->control.exit_info_2 = 0;
 772
 773         /* nested_svm_vmexit this gets called afterwards from handle_exit */
 774         svm->nested.exit_required = true;
 775         trace_kvm_nested_intr_vmexit(svm->vmcb->save.rip);
 776 }
 777
 778 static bool nested_exit_on_intr(struct vcpu_svm *svm)
 779 {
 780         return (svm->nested.intercept & 1ULL);
 781 }
 782
 783 int svm_check_nested_events(struct kvm_vcpu *vcpu)
 784 {
 785         struct vcpu_svm *svm = to_svm(vcpu);
 786         bool block_nested_events =
 787                 kvm_event_needs_reinjection(vcpu) || svm->nested.exit_required;
 788
 789         if (kvm_cpu_has_interrupt(vcpu) && nested_exit_on_intr(svm)) {
 790                 if (block_nested_events)
 791                         return -EBUSY;
 792                 nested_svm_intr(svm);
 793                 return 0;
 794         }
 795
 796         return 0;
 797 }
 798
 799 int nested_svm_exit_special(struct vcpu_svm *svm)
 800 {
 801         u32 exit_code = svm->vmcb->control.exit_code;
 802
 803         switch (exit_code) {
 804         case SVM_EXIT_INTR:
 805         case SVM_EXIT_NMI:
 806         case SVM_EXIT_EXCP_BASE + MC_VECTOR:
 807                 return NESTED_EXIT_HOST;
 808         case SVM_EXIT_NPF:
 809                 /* For now we are always handling NPFs when using them */
 810                 if (npt_enabled)
 811                         return NESTED_EXIT_HOST;
 812                 break;
 813         case SVM_EXIT_EXCP_BASE + PF_VECTOR:
 814                 /* When we're shadowing, trap PFs, but not async PF */
 815                 if (!npt_enabled && svm->vcpu.arch.apf.host_apf_reason == 0)
 816                         return NESTED_EXIT_HOST;
 817                 break;
 818         default:
 819                 break;
 820         }
 821
 822         return NESTED_EXIT_CONTINUE;
 823 }