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 "svm.h"
  29
  30 static void nested_svm_inject_npf_exit(struct kvm_vcpu *vcpu,
  31                                        struct x86_exception *fault)
  32 {
  33         struct vcpu_svm *svm = to_svm(vcpu);
  34
  35         if (svm->vmcb->control.exit_code != SVM_EXIT_NPF) {
  36                 /*
  37                  * TODO: track the cause of the nested page fault, and
  38                  * correctly fill in the high bits of exit_info_1.
  39                  */
  40                 svm->vmcb->control.exit_code = SVM_EXIT_NPF;
  41                 svm->vmcb->control.exit_code_hi = 0;
  42                 svm->vmcb->control.exit_info_1 = (1ULL << 32);
  43                 svm->vmcb->control.exit_info_2 = fault->address;
  44         }
  45
  46         svm->vmcb->control.exit_info_1 &= ~0xffffffffULL;
  47         svm->vmcb->control.exit_info_1 |= fault->error_code;
  48
  49         /*
  50          * The present bit is always zero for page structure faults on real
  51          * hardware.
  52          */
  53         if (svm->vmcb->control.exit_info_1 & (2ULL << 32))
  54                 svm->vmcb->control.exit_info_1 &= ~1;
  55
  56         nested_svm_vmexit(svm);
  57 }
  58
  59 static u64 nested_svm_get_tdp_pdptr(struct kvm_vcpu *vcpu, int index)
  60 {
  61         struct vcpu_svm *svm = to_svm(vcpu);
  62         u64 cr3 = svm->nested.nested_cr3;
  63         u64 pdpte;
  64         int ret;
  65
  66         ret = kvm_vcpu_read_guest_page(vcpu, gpa_to_gfn(__sme_clr(cr3)), &pdpte,
  67                                        offset_in_page(cr3) + index * 8, 8);
  68         if (ret)
  69                 return 0;
  70         return pdpte;
  71 }
  72
  73 static unsigned long nested_svm_get_tdp_cr3(struct kvm_vcpu *vcpu)
  74 {
  75         struct vcpu_svm *svm = to_svm(vcpu);
  76
  77         return svm->nested.nested_cr3;
  78 }
  79
  80 static void nested_svm_init_mmu_context(struct kvm_vcpu *vcpu)
  81 {
  82         WARN_ON(mmu_is_nested(vcpu));
  83
  84         vcpu->arch.mmu = &vcpu->arch.guest_mmu;
  85         kvm_init_shadow_mmu(vcpu);
  86         vcpu->arch.mmu->get_guest_pgd     = nested_svm_get_tdp_cr3;
  87         vcpu->arch.mmu->get_pdptr         = nested_svm_get_tdp_pdptr;
  88         vcpu->arch.mmu->inject_page_fault = nested_svm_inject_npf_exit;
  89         vcpu->arch.mmu->shadow_root_level = kvm_x86_ops.get_tdp_level(vcpu);
  90         reset_shadow_zero_bits_mask(vcpu, vcpu->arch.mmu);
  91         vcpu->arch.walk_mmu              = &vcpu->arch.nested_mmu;
  92 }
  93
  94 static void nested_svm_uninit_mmu_context(struct kvm_vcpu *vcpu)
  95 {
  96         vcpu->arch.mmu = &vcpu->arch.root_mmu;
  97         vcpu->arch.walk_mmu = &vcpu->arch.root_mmu;
  98 }
  99
 100 void recalc_intercepts(struct vcpu_svm *svm)
 101 {
 102         struct vmcb_control_area *c, *h;
 103         struct nested_state *g;
 104
 105         mark_dirty(svm->vmcb, VMCB_INTERCEPTS);
 106
 107         if (!is_guest_mode(&svm->vcpu))
 108                 return;
 109
 110         c = &svm->vmcb->control;
 111         h = &svm->nested.hsave->control;
 112         g = &svm->nested;
 113
 114         c->intercept_cr = h->intercept_cr;
 115         c->intercept_dr = h->intercept_dr;
 116         c->intercept_exceptions = h->intercept_exceptions;
 117         c->intercept = h->intercept;
 118
 119         if (svm->vcpu.arch.hflags & HF_VINTR_MASK) {
 120                 /* We only want the cr8 intercept bits of L1 */
 121                 c->intercept_cr &= ~(1U << INTERCEPT_CR8_READ);
 122                 c->intercept_cr &= ~(1U << INTERCEPT_CR8_WRITE);
 123
 124                 /*
 125                  * Once running L2 with HF_VINTR_MASK, EFLAGS.IF does not
 126                  * affect any interrupt we may want to inject; therefore,
 127                  * interrupt window vmexits are irrelevant to L0.
 128                  */
 129                 c->intercept &= ~(1ULL << INTERCEPT_VINTR);
 130         }
 131
 132         /* We don't want to see VMMCALLs from a nested guest */
 133         c->intercept &= ~(1ULL << INTERCEPT_VMMCALL);
 134
 135         c->intercept_cr |= g->intercept_cr;
 136         c->intercept_dr |= g->intercept_dr;
 137         c->intercept_exceptions |= g->intercept_exceptions;
 138         c->intercept |= g->intercept;
 139 }
 140
 141 static void copy_vmcb_control_area(struct vmcb *dst_vmcb, struct vmcb *from_vmcb)
 142 {
 143         struct vmcb_control_area *dst  = &dst_vmcb->control;
 144         struct vmcb_control_area *from = &from_vmcb->control;
 145
 146         dst->intercept_cr         = from->intercept_cr;
 147         dst->intercept_dr         = from->intercept_dr;
 148         dst->intercept_exceptions = from->intercept_exceptions;
 149         dst->intercept            = from->intercept;
 150         dst->iopm_base_pa         = from->iopm_base_pa;
 151         dst->msrpm_base_pa        = from->msrpm_base_pa;
 152         dst->tsc_offset           = from->tsc_offset;
 153         dst->asid                 = from->asid;
 154         dst->tlb_ctl              = from->tlb_ctl;
 155         dst->int_ctl              = from->int_ctl;
 156         dst->int_vector           = from->int_vector;
 157         dst->int_state            = from->int_state;
 158         dst->exit_code            = from->exit_code;
 159         dst->exit_code_hi         = from->exit_code_hi;
 160         dst->exit_info_1          = from->exit_info_1;
 161         dst->exit_info_2          = from->exit_info_2;
 162         dst->exit_int_info        = from->exit_int_info;
 163         dst->exit_int_info_err    = from->exit_int_info_err;
 164         dst->nested_ctl           = from->nested_ctl;
 165         dst->event_inj            = from->event_inj;
 166         dst->event_inj_err        = from->event_inj_err;
 167         dst->nested_cr3           = from->nested_cr3;
 168         dst->virt_ext              = from->virt_ext;
 169         dst->pause_filter_count   = from->pause_filter_count;
 170         dst->pause_filter_thresh  = from->pause_filter_thresh;
 171 }
 172
 173 static bool nested_svm_vmrun_msrpm(struct vcpu_svm *svm)
 174 {
 175         /*
 176          * This function merges the msr permission bitmaps of kvm and the
 177          * nested vmcb. It is optimized in that it only merges the parts where
 178          * the kvm msr permission bitmap may contain zero bits
 179          */
 180         int i;
 181
 182         if (!(svm->nested.intercept & (1ULL << INTERCEPT_MSR_PROT)))
 183                 return true;
 184
 185         for (i = 0; i < MSRPM_OFFSETS; i++) {
 186                 u32 value, p;
 187                 u64 offset;
 188
 189                 if (msrpm_offsets[i] == 0xffffffff)
 190                         break;
 191
 192                 p      = msrpm_offsets[i];
 193                 offset = svm->nested.vmcb_msrpm + (p * 4);
 194
 195                 if (kvm_vcpu_read_guest(&svm->vcpu, offset, &value, 4))
 196                         return false;
 197
 198                 svm->nested.msrpm[p] = svm->msrpm[p] | value;
 199         }
 200
 201         svm->vmcb->control.msrpm_base_pa = __sme_set(__pa(svm->nested.msrpm));
 202
 203         return true;
 204 }
 205
 206 static bool nested_vmcb_checks(struct vmcb *vmcb)
 207 {
 208         if ((vmcb->save.efer & EFER_SVME) == 0)
 209                 return false;
 210
 211         if (((vmcb->save.cr0 & X86_CR0_CD) == 0) &&
 212             (vmcb->save.cr0 & X86_CR0_NW))
 213                 return false;
 214
 215         if ((vmcb->control.intercept & (1ULL << INTERCEPT_VMRUN)) == 0)
 216                 return false;
 217
 218         if (vmcb->control.asid == 0)
 219                 return false;
 220
 221         if ((vmcb->control.nested_ctl & SVM_NESTED_CTL_NP_ENABLE) &&
 222             !npt_enabled)
 223                 return false;
 224
 225         return true;
 226 }
 227
 228 void enter_svm_guest_mode(struct vcpu_svm *svm, u64 vmcb_gpa,
 229                           struct vmcb *nested_vmcb, struct kvm_host_map *map)
 230 {
 231         bool evaluate_pending_interrupts =
 232                 is_intercept(svm, INTERCEPT_VINTR) ||
 233                 is_intercept(svm, INTERCEPT_IRET);
 234
 235         if (kvm_get_rflags(&svm->vcpu) & X86_EFLAGS_IF)
 236                 svm->vcpu.arch.hflags |= HF_HIF_MASK;
 237         else
 238                 svm->vcpu.arch.hflags &= ~HF_HIF_MASK;
 239
 240         if (nested_vmcb->control.nested_ctl & SVM_NESTED_CTL_NP_ENABLE) {
 241                 svm->nested.nested_cr3 = nested_vmcb->control.nested_cr3;
 242                 nested_svm_init_mmu_context(&svm->vcpu);
 243         }
 244
 245         /* Load the nested guest state */
 246         svm->vmcb->save.es = nested_vmcb->save.es;
 247         svm->vmcb->save.cs = nested_vmcb->save.cs;
 248         svm->vmcb->save.ss = nested_vmcb->save.ss;
 249         svm->vmcb->save.ds = nested_vmcb->save.ds;
 250         svm->vmcb->save.gdtr = nested_vmcb->save.gdtr;
 251         svm->vmcb->save.idtr = nested_vmcb->save.idtr;
 252         kvm_set_rflags(&svm->vcpu, nested_vmcb->save.rflags);
 253         svm_set_efer(&svm->vcpu, nested_vmcb->save.efer);
 254         svm_set_cr0(&svm->vcpu, nested_vmcb->save.cr0);
 255         svm_set_cr4(&svm->vcpu, nested_vmcb->save.cr4);
 256         if (npt_enabled) {
 257                 svm->vmcb->save.cr3 = nested_vmcb->save.cr3;
 258                 svm->vcpu.arch.cr3 = nested_vmcb->save.cr3;
 259         } else
 260                 (void)kvm_set_cr3(&svm->vcpu, nested_vmcb->save.cr3);
 261
 262         /* Guest paging mode is active - reset mmu */
 263         kvm_mmu_reset_context(&svm->vcpu);
 264
 265         svm->vmcb->save.cr2 = svm->vcpu.arch.cr2 = nested_vmcb->save.cr2;
 266         kvm_rax_write(&svm->vcpu, nested_vmcb->save.rax);
 267         kvm_rsp_write(&svm->vcpu, nested_vmcb->save.rsp);
 268         kvm_rip_write(&svm->vcpu, nested_vmcb->save.rip);
 269
 270         /* In case we don't even reach vcpu_run, the fields are not updated */
 271         svm->vmcb->save.rax = nested_vmcb->save.rax;
 272         svm->vmcb->save.rsp = nested_vmcb->save.rsp;
 273         svm->vmcb->save.rip = nested_vmcb->save.rip;
 274         svm->vmcb->save.dr7 = nested_vmcb->save.dr7;
 275         svm->vcpu.arch.dr6  = nested_vmcb->save.dr6;
 276         svm->vmcb->save.cpl = nested_vmcb->save.cpl;
 277
 278         svm->nested.vmcb_msrpm = nested_vmcb->control.msrpm_base_pa & ~0x0fffULL;
 279         svm->nested.vmcb_iopm  = nested_vmcb->control.iopm_base_pa  & ~0x0fffULL;
 280
 281         /* cache intercepts */
 282         svm->nested.intercept_cr         = nested_vmcb->control.intercept_cr;
 283         svm->nested.intercept_dr         = nested_vmcb->control.intercept_dr;
 284         svm->nested.intercept_exceptions = nested_vmcb->control.intercept_exceptions;
 285         svm->nested.intercept            = nested_vmcb->control.intercept;
 286
 287         svm_flush_tlb(&svm->vcpu);
 288         svm->vmcb->control.int_ctl = nested_vmcb->control.int_ctl | V_INTR_MASKING_MASK;
 289         if (nested_vmcb->control.int_ctl & V_INTR_MASKING_MASK)
 290                 svm->vcpu.arch.hflags |= HF_VINTR_MASK;
 291         else
 292                 svm->vcpu.arch.hflags &= ~HF_VINTR_MASK;
 293
 294         svm->vcpu.arch.tsc_offset += nested_vmcb->control.tsc_offset;
 295         svm->vmcb->control.tsc_offset = svm->vcpu.arch.tsc_offset;
 296
 297         svm->vmcb->control.virt_ext = nested_vmcb->control.virt_ext;
 298         svm->vmcb->control.int_vector = nested_vmcb->control.int_vector;
 299         svm->vmcb->control.int_state = nested_vmcb->control.int_state;
 300         svm->vmcb->control.event_inj = nested_vmcb->control.event_inj;
 301         svm->vmcb->control.event_inj_err = nested_vmcb->control.event_inj_err;
 302
 303         svm->vmcb->control.pause_filter_count =
 304                 nested_vmcb->control.pause_filter_count;
 305         svm->vmcb->control.pause_filter_thresh =
 306                 nested_vmcb->control.pause_filter_thresh;
 307
 308         kvm_vcpu_unmap(&svm->vcpu, map, true);
 309
 310         /* Enter Guest-Mode */
 311         enter_guest_mode(&svm->vcpu);
 312
 313         /*
 314          * Merge guest and host intercepts - must be called  with vcpu in
 315          * guest-mode to take affect here
 316          */
 317         recalc_intercepts(svm);
 318
 319         svm->nested.vmcb = vmcb_gpa;
 320
 321         /*
 322          * If L1 had a pending IRQ/NMI before executing VMRUN,
 323          * which wasn't delivered because it was disallowed (e.g.
 324          * interrupts disabled), L0 needs to evaluate if this pending
 325          * event should cause an exit from L2 to L1 or be delivered
 326          * directly to L2.
 327          *
 328          * Usually this would be handled by the processor noticing an
 329          * IRQ/NMI window request.  However, VMRUN can unblock interrupts
 330          * by implicitly setting GIF, so force L0 to perform pending event
 331          * evaluation by requesting a KVM_REQ_EVENT.
 332          */
 333         enable_gif(svm);
 334         if (unlikely(evaluate_pending_interrupts))
 335                 kvm_make_request(KVM_REQ_EVENT, &svm->vcpu);
 336
 337         mark_all_dirty(svm->vmcb);
 338 }
 339
 340 int nested_svm_vmrun(struct vcpu_svm *svm)
 341 {
 342         int ret;
 343         struct vmcb *nested_vmcb;
 344         struct vmcb *hsave = svm->nested.hsave;
 345         struct vmcb *vmcb = svm->vmcb;
 346         struct kvm_host_map map;
 347         u64 vmcb_gpa;
 348
 349         if (is_smm(&svm->vcpu)) {
 350                 kvm_queue_exception(&svm->vcpu, UD_VECTOR);
 351                 return 1;
 352         }
 353
 354         vmcb_gpa = svm->vmcb->save.rax;
 355         ret = kvm_vcpu_map(&svm->vcpu, gpa_to_gfn(vmcb_gpa), &map);
 356         if (ret == -EINVAL) {
 357                 kvm_inject_gp(&svm->vcpu, 0);
 358                 return 1;
 359         } else if (ret) {
 360                 return kvm_skip_emulated_instruction(&svm->vcpu);
 361         }
 362
 363         ret = kvm_skip_emulated_instruction(&svm->vcpu);
 364
 365         nested_vmcb = map.hva;
 366
 367         if (!nested_vmcb_checks(nested_vmcb)) {
 368                 nested_vmcb->control.exit_code    = SVM_EXIT_ERR;
 369                 nested_vmcb->control.exit_code_hi = 0;
 370                 nested_vmcb->control.exit_info_1  = 0;
 371                 nested_vmcb->control.exit_info_2  = 0;
 372
 373                 kvm_vcpu_unmap(&svm->vcpu, &map, true);
 374
 375                 return ret;
 376         }
 377
 378         trace_kvm_nested_vmrun(svm->vmcb->save.rip, vmcb_gpa,
 379                                nested_vmcb->save.rip,
 380                                nested_vmcb->control.int_ctl,
 381                                nested_vmcb->control.event_inj,
 382                                nested_vmcb->control.nested_ctl);
 383
 384         trace_kvm_nested_intercepts(nested_vmcb->control.intercept_cr & 0xffff,
 385                                     nested_vmcb->control.intercept_cr >> 16,
 386                                     nested_vmcb->control.intercept_exceptions,
 387                                     nested_vmcb->control.intercept);
 388
 389         /* Clear internal status */
 390         kvm_clear_exception_queue(&svm->vcpu);
 391         kvm_clear_interrupt_queue(&svm->vcpu);
 392
 393         /*
 394          * Save the old vmcb, so we don't need to pick what we save, but can
 395          * restore everything when a VMEXIT occurs
 396          */
 397         hsave->save.es     = vmcb->save.es;
 398         hsave->save.cs     = vmcb->save.cs;
 399         hsave->save.ss     = vmcb->save.ss;
 400         hsave->save.ds     = vmcb->save.ds;
 401         hsave->save.gdtr   = vmcb->save.gdtr;
 402         hsave->save.idtr   = vmcb->save.idtr;
 403         hsave->save.efer   = svm->vcpu.arch.efer;
 404         hsave->save.cr0    = kvm_read_cr0(&svm->vcpu);
 405         hsave->save.cr4    = svm->vcpu.arch.cr4;
 406         hsave->save.rflags = kvm_get_rflags(&svm->vcpu);
 407         hsave->save.rip    = kvm_rip_read(&svm->vcpu);
 408         hsave->save.rsp    = vmcb->save.rsp;
 409         hsave->save.rax    = vmcb->save.rax;
 410         if (npt_enabled)
 411                 hsave->save.cr3    = vmcb->save.cr3;
 412         else
 413                 hsave->save.cr3    = kvm_read_cr3(&svm->vcpu);
 414
 415         copy_vmcb_control_area(hsave, vmcb);
 416
 417         svm->nested.nested_run_pending = 1;
 418         enter_svm_guest_mode(svm, vmcb_gpa, nested_vmcb, &map);
 419
 420         if (!nested_svm_vmrun_msrpm(svm)) {
 421                 svm->vmcb->control.exit_code    = SVM_EXIT_ERR;
 422                 svm->vmcb->control.exit_code_hi = 0;
 423                 svm->vmcb->control.exit_info_1  = 0;
 424                 svm->vmcb->control.exit_info_2  = 0;
 425
 426                 nested_svm_vmexit(svm);
 427         }
 428
 429         return ret;
 430 }
 431
 432 void nested_svm_vmloadsave(struct vmcb *from_vmcb, struct vmcb *to_vmcb)
 433 {
 434         to_vmcb->save.fs = from_vmcb->save.fs;
 435         to_vmcb->save.gs = from_vmcb->save.gs;
 436         to_vmcb->save.tr = from_vmcb->save.tr;
 437         to_vmcb->save.ldtr = from_vmcb->save.ldtr;
 438         to_vmcb->save.kernel_gs_base = from_vmcb->save.kernel_gs_base;
 439         to_vmcb->save.star = from_vmcb->save.star;
 440         to_vmcb->save.lstar = from_vmcb->save.lstar;
 441         to_vmcb->save.cstar = from_vmcb->save.cstar;
 442         to_vmcb->save.sfmask = from_vmcb->save.sfmask;
 443         to_vmcb->save.sysenter_cs = from_vmcb->save.sysenter_cs;
 444         to_vmcb->save.sysenter_esp = from_vmcb->save.sysenter_esp;
 445         to_vmcb->save.sysenter_eip = from_vmcb->save.sysenter_eip;
 446 }
 447
 448 int nested_svm_vmexit(struct vcpu_svm *svm)
 449 {
 450         int rc;
 451         struct vmcb *nested_vmcb;
 452         struct vmcb *hsave = svm->nested.hsave;
 453         struct vmcb *vmcb = svm->vmcb;
 454         struct kvm_host_map map;
 455
 456         trace_kvm_nested_vmexit_inject(vmcb->control.exit_code,
 457                                        vmcb->control.exit_info_1,
 458                                        vmcb->control.exit_info_2,
 459                                        vmcb->control.exit_int_info,
 460                                        vmcb->control.exit_int_info_err,
 461                                        KVM_ISA_SVM);
 462
 463         rc = kvm_vcpu_map(&svm->vcpu, gpa_to_gfn(svm->nested.vmcb), &map);
 464         if (rc) {
 465                 if (rc == -EINVAL)
 466                         kvm_inject_gp(&svm->vcpu, 0);
 467                 return 1;
 468         }
 469
 470         nested_vmcb = map.hva;
 471
 472         /* Exit Guest-Mode */
 473         leave_guest_mode(&svm->vcpu);
 474         svm->nested.vmcb = 0;
 475
 476         /* in case we halted in L2 */
 477         svm->vcpu.arch.mp_state = KVM_MP_STATE_RUNNABLE;
 478
 479         /* Give the current vmcb to the guest */
 480         disable_gif(svm);
 481
 482         nested_vmcb->save.es     = vmcb->save.es;
 483         nested_vmcb->save.cs     = vmcb->save.cs;
 484         nested_vmcb->save.ss     = vmcb->save.ss;
 485         nested_vmcb->save.ds     = vmcb->save.ds;
 486         nested_vmcb->save.gdtr   = vmcb->save.gdtr;
 487         nested_vmcb->save.idtr   = vmcb->save.idtr;
 488         nested_vmcb->save.efer   = svm->vcpu.arch.efer;
 489         nested_vmcb->save.cr0    = kvm_read_cr0(&svm->vcpu);
 490         nested_vmcb->save.cr3    = kvm_read_cr3(&svm->vcpu);
 491         nested_vmcb->save.cr2    = vmcb->save.cr2;
 492         nested_vmcb->save.cr4    = svm->vcpu.arch.cr4;
 493         nested_vmcb->save.rflags = kvm_get_rflags(&svm->vcpu);
 494         nested_vmcb->save.rip    = vmcb->save.rip;
 495         nested_vmcb->save.rsp    = vmcb->save.rsp;
 496         nested_vmcb->save.rax    = vmcb->save.rax;
 497         nested_vmcb->save.dr7    = vmcb->save.dr7;
 498         nested_vmcb->save.dr6    = svm->vcpu.arch.dr6;
 499         nested_vmcb->save.cpl    = vmcb->save.cpl;
 500
 501         nested_vmcb->control.int_ctl           = vmcb->control.int_ctl;
 502         nested_vmcb->control.int_vector        = vmcb->control.int_vector;
 503         nested_vmcb->control.int_state         = vmcb->control.int_state;
 504         nested_vmcb->control.exit_code         = vmcb->control.exit_code;
 505         nested_vmcb->control.exit_code_hi      = vmcb->control.exit_code_hi;
 506         nested_vmcb->control.exit_info_1       = vmcb->control.exit_info_1;
 507         nested_vmcb->control.exit_info_2       = vmcb->control.exit_info_2;
 508         nested_vmcb->control.exit_int_info     = vmcb->control.exit_int_info;
 509         nested_vmcb->control.exit_int_info_err = vmcb->control.exit_int_info_err;
 510
 511         if (svm->nrips_enabled)
 512                 nested_vmcb->control.next_rip  = vmcb->control.next_rip;
 513
 514         /*
 515          * If we emulate a VMRUN/#VMEXIT in the same host #vmexit cycle we have
 516          * to make sure that we do not lose injected events. So check event_inj
 517          * here and copy it to exit_int_info if it is valid.
 518          * Exit_int_info and event_inj can't be both valid because the case
 519          * below only happens on a VMRUN instruction intercept which has
 520          * no valid exit_int_info set.
 521          */
 522         if (vmcb->control.event_inj & SVM_EVTINJ_VALID) {
 523                 struct vmcb_control_area *nc = &nested_vmcb->control;
 524
 525                 nc->exit_int_info     = vmcb->control.event_inj;
 526                 nc->exit_int_info_err = vmcb->control.event_inj_err;
 527         }
 528
 529         nested_vmcb->control.tlb_ctl           = 0;
 530         nested_vmcb->control.event_inj         = 0;
 531         nested_vmcb->control.event_inj_err     = 0;
 532
 533         nested_vmcb->control.pause_filter_count =
 534                 svm->vmcb->control.pause_filter_count;
 535         nested_vmcb->control.pause_filter_thresh =
 536                 svm->vmcb->control.pause_filter_thresh;
 537
 538         /* We always set V_INTR_MASKING and remember the old value in hflags */
 539         if (!(svm->vcpu.arch.hflags & HF_VINTR_MASK))
 540                 nested_vmcb->control.int_ctl &= ~V_INTR_MASKING_MASK;
 541
 542         /* Restore the original control entries */
 543         copy_vmcb_control_area(vmcb, hsave);
 544
 545         svm->vcpu.arch.tsc_offset = svm->vmcb->control.tsc_offset;
 546         kvm_clear_exception_queue(&svm->vcpu);
 547         kvm_clear_interrupt_queue(&svm->vcpu);
 548
 549         svm->nested.nested_cr3 = 0;
 550
 551         /* Restore selected save entries */
 552         svm->vmcb->save.es = hsave->save.es;
 553         svm->vmcb->save.cs = hsave->save.cs;
 554         svm->vmcb->save.ss = hsave->save.ss;
 555         svm->vmcb->save.ds = hsave->save.ds;
 556         svm->vmcb->save.gdtr = hsave->save.gdtr;
 557         svm->vmcb->save.idtr = hsave->save.idtr;
 558         kvm_set_rflags(&svm->vcpu, hsave->save.rflags);
 559         svm_set_efer(&svm->vcpu, hsave->save.efer);
 560         svm_set_cr0(&svm->vcpu, hsave->save.cr0 | X86_CR0_PE);
 561         svm_set_cr4(&svm->vcpu, hsave->save.cr4);
 562         if (npt_enabled) {
 563                 svm->vmcb->save.cr3 = hsave->save.cr3;
 564                 svm->vcpu.arch.cr3 = hsave->save.cr3;
 565         } else {
 566                 (void)kvm_set_cr3(&svm->vcpu, hsave->save.cr3);
 567         }
 568         kvm_rax_write(&svm->vcpu, hsave->save.rax);
 569         kvm_rsp_write(&svm->vcpu, hsave->save.rsp);
 570         kvm_rip_write(&svm->vcpu, hsave->save.rip);
 571         svm->vmcb->save.dr7 = 0;
 572         svm->vmcb->save.cpl = 0;
 573         svm->vmcb->control.exit_int_info = 0;
 574
 575         mark_all_dirty(svm->vmcb);
 576
 577         kvm_vcpu_unmap(&svm->vcpu, &map, true);
 578
 579         nested_svm_uninit_mmu_context(&svm->vcpu);
 580         kvm_mmu_reset_context(&svm->vcpu);
 581         kvm_mmu_load(&svm->vcpu);
 582
 583         /*
 584          * Drop what we picked up for L2 via svm_complete_interrupts() so it
 585          * doesn't end up in L1.
 586          */
 587         svm->vcpu.arch.nmi_injected = false;
 588         kvm_clear_exception_queue(&svm->vcpu);
 589         kvm_clear_interrupt_queue(&svm->vcpu);
 590
 591         return 0;
 592 }
 593
 594 static int nested_svm_exit_handled_msr(struct vcpu_svm *svm)
 595 {
 596         u32 offset, msr, value;
 597         int write, mask;
 598
 599         if (!(svm->nested.intercept & (1ULL << INTERCEPT_MSR_PROT)))
 600                 return NESTED_EXIT_HOST;
 601
 602         msr    = svm->vcpu.arch.regs[VCPU_REGS_RCX];
 603         offset = svm_msrpm_offset(msr);
 604         write  = svm->vmcb->control.exit_info_1 & 1;
 605         mask   = 1 << ((2 * (msr & 0xf)) + write);
 606
 607         if (offset == MSR_INVALID)
 608                 return NESTED_EXIT_DONE;
 609
 610         /* Offset is in 32 bit units but need in 8 bit units */
 611         offset *= 4;
 612
 613         if (kvm_vcpu_read_guest(&svm->vcpu, svm->nested.vmcb_msrpm + offset, &value, 4))
 614                 return NESTED_EXIT_DONE;
 615
 616         return (value & mask) ? NESTED_EXIT_DONE : NESTED_EXIT_HOST;
 617 }
 618
 619 /* DB exceptions for our internal use must not cause vmexit */
 620 static int nested_svm_intercept_db(struct vcpu_svm *svm)
 621 {
 622         unsigned long dr6 = svm->vmcb->save.dr6;
 623
 624         /* Always catch it and pass it to userspace if debugging.  */
 625         if (svm->vcpu.guest_debug &
 626             (KVM_GUESTDBG_SINGLESTEP | KVM_GUESTDBG_USE_HW_BP))
 627                 return NESTED_EXIT_HOST;
 628
 629         /* if we're not singlestepping, it's not ours */
 630         if (!svm->nmi_singlestep)
 631                 goto reflected_db;
 632
 633         /* if it's not a singlestep exception, it's not ours */
 634         if (!(dr6 & DR6_BS))
 635                 goto reflected_db;
 636
 637         /* if the guest is singlestepping, it should get the vmexit */
 638         if (svm->nmi_singlestep_guest_rflags & X86_EFLAGS_TF) {
 639                 disable_nmi_singlestep(svm);
 640                 goto reflected_db;
 641         }
 642
 643         /* it's ours, the nested hypervisor must not see this one */
 644         return NESTED_EXIT_HOST;
 645
 646 reflected_db:
 647         /*
 648          * Synchronize guest DR6 here just like in kvm_deliver_exception_payload;
 649          * it will be moved into the nested VMCB by nested_svm_vmexit.  Once
 650          * exceptions will be moved to svm_check_nested_events, all this stuff
 651          * will just go away and we could just return NESTED_EXIT_HOST
 652          * unconditionally.  db_interception will queue the exception, which
 653          * will be processed by svm_check_nested_events if a nested vmexit is
 654          * required, and we will just use kvm_deliver_exception_payload to copy
 655          * the payload to DR6 before vmexit.
 656          */
 657         WARN_ON(svm->vcpu.arch.switch_db_regs & KVM_DEBUGREG_WONT_EXIT);
 658         svm->vcpu.arch.dr6 &= ~(DR_TRAP_BITS | DR6_RTM);
 659         svm->vcpu.arch.dr6 |= dr6 & ~DR6_FIXED_1;
 660         return NESTED_EXIT_DONE;
 661 }
 662
 663 static int nested_svm_intercept_ioio(struct vcpu_svm *svm)
 664 {
 665         unsigned port, size, iopm_len;
 666         u16 val, mask;
 667         u8 start_bit;
 668         u64 gpa;
 669
 670         if (!(svm->nested.intercept & (1ULL << INTERCEPT_IOIO_PROT)))
 671                 return NESTED_EXIT_HOST;
 672
 673         port = svm->vmcb->control.exit_info_1 >> 16;
 674         size = (svm->vmcb->control.exit_info_1 & SVM_IOIO_SIZE_MASK) >>
 675                 SVM_IOIO_SIZE_SHIFT;
 676         gpa  = svm->nested.vmcb_iopm + (port / 8);
 677         start_bit = port % 8;
 678         iopm_len = (start_bit + size > 8) ? 2 : 1;
 679         mask = (0xf >> (4 - size)) << start_bit;
 680         val = 0;
 681
 682         if (kvm_vcpu_read_guest(&svm->vcpu, gpa, &val, iopm_len))
 683                 return NESTED_EXIT_DONE;
 684
 685         return (val & mask) ? NESTED_EXIT_DONE : NESTED_EXIT_HOST;
 686 }
 687
 688 static int nested_svm_intercept(struct vcpu_svm *svm)
 689 {
 690         u32 exit_code = svm->vmcb->control.exit_code;
 691         int vmexit = NESTED_EXIT_HOST;
 692
 693         switch (exit_code) {
 694         case SVM_EXIT_MSR:
 695                 vmexit = nested_svm_exit_handled_msr(svm);
 696                 break;
 697         case SVM_EXIT_IOIO:
 698                 vmexit = nested_svm_intercept_ioio(svm);
 699                 break;
 700         case SVM_EXIT_READ_CR0 ... SVM_EXIT_WRITE_CR8: {
 701                 u32 bit = 1U << (exit_code - SVM_EXIT_READ_CR0);
 702                 if (svm->nested.intercept_cr & bit)
 703                         vmexit = NESTED_EXIT_DONE;
 704                 break;
 705         }
 706         case SVM_EXIT_READ_DR0 ... SVM_EXIT_WRITE_DR7: {
 707                 u32 bit = 1U << (exit_code - SVM_EXIT_READ_DR0);
 708                 if (svm->nested.intercept_dr & bit)
 709                         vmexit = NESTED_EXIT_DONE;
 710                 break;
 711         }
 712         case SVM_EXIT_EXCP_BASE ... SVM_EXIT_EXCP_BASE + 0x1f: {
 713                 u32 excp_bits = 1 << (exit_code - SVM_EXIT_EXCP_BASE);
 714                 if (svm->nested.intercept_exceptions & excp_bits) {
 715                         if (exit_code == SVM_EXIT_EXCP_BASE + DB_VECTOR)
 716                                 vmexit = nested_svm_intercept_db(svm);
 717                         else if (exit_code == SVM_EXIT_EXCP_BASE + BP_VECTOR &&
 718                                  svm->vcpu.guest_debug & KVM_GUESTDBG_USE_SW_BP)
 719                                 vmexit = NESTED_EXIT_HOST;
 720                         else
 721                                 vmexit = NESTED_EXIT_DONE;
 722                 }
 723                 /* async page fault always cause vmexit */
 724                 else if ((exit_code == SVM_EXIT_EXCP_BASE + PF_VECTOR) &&
 725                          svm->vcpu.arch.exception.nested_apf != 0)
 726                         vmexit = NESTED_EXIT_DONE;
 727                 break;
 728         }
 729         case SVM_EXIT_ERR: {
 730                 vmexit = NESTED_EXIT_DONE;
 731                 break;
 732         }
 733         default: {
 734                 u64 exit_bits = 1ULL << (exit_code - SVM_EXIT_INTR);
 735                 if (svm->nested.intercept & exit_bits)
 736                         vmexit = NESTED_EXIT_DONE;
 737         }
 738         }
 739
 740         return vmexit;
 741 }
 742
 743 int nested_svm_exit_handled(struct vcpu_svm *svm)
 744 {
 745         int vmexit;
 746
 747         vmexit = nested_svm_intercept(svm);
 748
 749         if (vmexit == NESTED_EXIT_DONE)
 750                 nested_svm_vmexit(svm);
 751
 752         return vmexit;
 753 }
 754
 755 int nested_svm_check_permissions(struct vcpu_svm *svm)
 756 {
 757         if (!(svm->vcpu.arch.efer & EFER_SVME) ||
 758             !is_paging(&svm->vcpu)) {
 759                 kvm_queue_exception(&svm->vcpu, UD_VECTOR);
 760                 return 1;
 761         }
 762
 763         if (svm->vmcb->save.cpl) {
 764                 kvm_inject_gp(&svm->vcpu, 0);
 765                 return 1;
 766         }
 767
 768         return 0;
 769 }
 770
 771 int nested_svm_check_exception(struct vcpu_svm *svm, unsigned nr,
 772                                bool has_error_code, u32 error_code)
 773 {
 774         int vmexit;
 775
 776         if (!is_guest_mode(&svm->vcpu))
 777                 return 0;
 778
 779         vmexit = nested_svm_intercept(svm);
 780         if (vmexit != NESTED_EXIT_DONE)
 781                 return 0;
 782
 783         svm->vmcb->control.exit_code = SVM_EXIT_EXCP_BASE + nr;
 784         svm->vmcb->control.exit_code_hi = 0;
 785         svm->vmcb->control.exit_info_1 = error_code;
 786
 787         /*
 788          * EXITINFO2 is undefined for all exception intercepts other
 789          * than #PF.
 790          */
 791         if (svm->vcpu.arch.exception.nested_apf)
 792                 svm->vmcb->control.exit_info_2 = svm->vcpu.arch.apf.nested_apf_token;
 793         else if (svm->vcpu.arch.exception.has_payload)
 794                 svm->vmcb->control.exit_info_2 = svm->vcpu.arch.exception.payload;
 795         else
 796                 svm->vmcb->control.exit_info_2 = svm->vcpu.arch.cr2;
 797
 798         svm->nested.exit_required = true;
 799         return vmexit;
 800 }
 801
 802 static bool nested_exit_on_nmi(struct vcpu_svm *svm)
 803 {
 804         return (svm->nested.intercept & (1ULL << INTERCEPT_NMI));
 805 }
 806
 807 static void nested_svm_nmi(struct vcpu_svm *svm)
 808 {
 809         svm->vmcb->control.exit_code = SVM_EXIT_NMI;
 810         svm->vmcb->control.exit_info_1 = 0;
 811         svm->vmcb->control.exit_info_2 = 0;
 812
 813         nested_svm_vmexit(svm);
 814 }
 815
 816 static void nested_svm_intr(struct vcpu_svm *svm)
 817 {
 818         trace_kvm_nested_intr_vmexit(svm->vmcb->save.rip);
 819
 820         svm->vmcb->control.exit_code   = SVM_EXIT_INTR;
 821         svm->vmcb->control.exit_info_1 = 0;
 822         svm->vmcb->control.exit_info_2 = 0;
 823
 824         nested_svm_vmexit(svm);
 825 }
 826
 827 static bool nested_exit_on_intr(struct vcpu_svm *svm)
 828 {
 829         return (svm->nested.intercept & 1ULL);
 830 }
 831
 832 static int svm_check_nested_events(struct kvm_vcpu *vcpu)
 833 {
 834         struct vcpu_svm *svm = to_svm(vcpu);
 835         bool block_nested_events =
 836                 kvm_event_needs_reinjection(vcpu) || svm->nested.exit_required ||
 837                 svm->nested.nested_run_pending;
 838
 839         if (vcpu->arch.nmi_pending && nested_exit_on_nmi(svm)) {
 840                 if (block_nested_events)
 841                         return -EBUSY;
 842                 nested_svm_nmi(svm);
 843                 return 0;
 844         }
 845
 846         if (kvm_cpu_has_interrupt(vcpu) && nested_exit_on_intr(svm)) {
 847                 if (block_nested_events)
 848                         return -EBUSY;
 849                 nested_svm_intr(svm);
 850                 return 0;
 851         }
 852
 853         return 0;
 854 }
 855
 856 int nested_svm_exit_special(struct vcpu_svm *svm)
 857 {
 858         u32 exit_code = svm->vmcb->control.exit_code;
 859
 860         switch (exit_code) {
 861         case SVM_EXIT_INTR:
 862         case SVM_EXIT_NMI:
 863         case SVM_EXIT_EXCP_BASE + MC_VECTOR:
 864                 return NESTED_EXIT_HOST;
 865         case SVM_EXIT_NPF:
 866                 /* For now we are always handling NPFs when using them */
 867                 if (npt_enabled)
 868                         return NESTED_EXIT_HOST;
 869                 break;
 870         case SVM_EXIT_EXCP_BASE + PF_VECTOR:
 871                 /* When we're shadowing, trap PFs, but not async PF */
 872                 if (!npt_enabled && svm->vcpu.arch.apf.host_apf_reason == 0)
 873                         return NESTED_EXIT_HOST;
 874                 break;
 875         default:
 876                 break;
 877         }
 878
 879         return NESTED_EXIT_CONTINUE;
 880 }
 881
 882 struct kvm_x86_nested_ops svm_nested_ops = {
 883         .check_events = svm_check_nested_events,
 884 };