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