palacios/src/palacios/vmx.c

   1 /*
   2  * This file is part of the Palacios Virtual Machine Monitor developed
   3  * by the V3VEE Project with funding from the United States National
   4  * Science Foundation and the Department of Energy.
   5  *
   6  * The V3VEE Project is a joint project between Northwestern University
   7  * and the University of New Mexico.  You can find out more at
   8  * http://www.v3vee.org
   9  *
  10  * Copyright (c) 2011, Jack Lange <jarusl@cs.northwestern.edu>
  11  * Copyright (c) 2011, The V3VEE Project <http://www.v3vee.org>
  12  * All rights reserved.
  13  *
  14  * Author: Jack Lange <jarusl@cs.northwestern.edu>
  15  *
  16  * This is free software.  You are permitted to use,
  17  * redistribute, and modify it as specified in the file "V3VEE_LICENSE".
  18  */
  19
  20
  21 #include <palacios/vmx.h>
  22 #include <palacios/vmm.h>
  23 #include <palacios/vmx_handler.h>
  24 #include <palacios/vmcs.h>
  25 #include <palacios/vmx_lowlevel.h>
  26 #include <palacios/vmm_lowlevel.h>
  27 #include <palacios/vmm_ctrl_regs.h>
  28 #include <palacios/vmm_config.h>
  29 #include <palacios/vmm_time.h>
  30 #include <palacios/vm_guest_mem.h>
  31 #include <palacios/vmm_direct_paging.h>
  32 #include <palacios/vmx_io.h>
  33 #include <palacios/vmx_msr.h>
  34 #include <palacios/vmm_decoder.h>
  35 #include <palacios/vmm_barrier.h>
  36 #include <palacios/vmm_timeout.h>
  37
  38 #ifdef V3_CONFIG_CHECKPOINT
  39 #include <palacios/vmm_checkpoint.h>
  40 #endif
  41
  42 #include <palacios/vmx_ept.h>
  43 #include <palacios/vmx_assist.h>
  44 #include <palacios/vmx_hw_info.h>
  45
  46 #ifndef V3_CONFIG_DEBUG_VMX
  47 #undef PrintDebug
  48 #define PrintDebug(fmt, args...)
  49 #endif
  50
  51
  52 /* These fields contain the hardware feature sets supported by the local CPU */
  53 static struct vmx_hw_info hw_info;
  54
  55 extern v3_cpu_arch_t v3_cpu_types[];
  56
  57 static addr_t host_vmcs_ptrs[V3_CONFIG_MAX_CPUS] = { [0 ... V3_CONFIG_MAX_CPUS - 1] = 0};
  58
  59 extern int v3_vmx_launch(struct v3_gprs * vm_regs, struct guest_info * info, struct v3_ctrl_regs * ctrl_regs);
  60 extern int v3_vmx_resume(struct v3_gprs * vm_regs, struct guest_info * info, struct v3_ctrl_regs * ctrl_regs);
  61
  62 static int inline check_vmcs_write(vmcs_field_t field, addr_t val) {
  63     int ret = 0;
  64
  65     ret = vmcs_write(field, val);
  66
  67     if (ret != VMX_SUCCESS) {
  68         PrintError("VMWRITE error on %s!: %d\n", v3_vmcs_field_to_str(field), ret);
  69         return 1;
  70     }
  71
  72
  73
  74
  75     return 0;
  76 }
  77
  78 static int inline check_vmcs_read(vmcs_field_t field, void * val) {
  79     int ret = 0;
  80
  81     ret = vmcs_read(field, val);
  82
  83     if (ret != VMX_SUCCESS) {
  84         PrintError("VMREAD error on %s!: %d\n", v3_vmcs_field_to_str(field), ret);
  85     }
  86
  87     return ret;
  88 }
  89
  90
  91
  92
  93 static addr_t allocate_vmcs() {
  94     struct vmcs_data * vmcs_page = NULL;
  95
  96     PrintDebug("Allocating page\n");
  97
  98     vmcs_page = (struct vmcs_data *)V3_VAddr(V3_AllocPages(1));
  99     memset(vmcs_page, 0, 4096);
 100
 101     vmcs_page->revision = hw_info.basic_info.revision;
 102     PrintDebug("VMX Revision: 0x%x\n", vmcs_page->revision);
 103
 104     return (addr_t)V3_PAddr((void *)vmcs_page);
 105 }
 106
 107 /*
 108
 109 static int debug_efer_read(struct guest_info * core, uint_t msr, struct v3_msr * src, void * priv_data) {
 110     struct v3_msr * efer = (struct v3_msr *)&(core->ctrl_regs.efer);
 111     V3_Print("\n\nEFER READ\n");
 112
 113     v3_print_guest_state(core);
 114
 115     src->value = efer->value;
 116     return 0;
 117 }
 118
 119 static int debug_efer_write(struct guest_info * core, uint_t msr, struct v3_msr src, void * priv_data) {
 120     struct v3_msr * efer = (struct v3_msr *)&(core->ctrl_regs.efer);
 121     V3_Print("\n\nEFER WRITE\n");
 122
 123     v3_print_guest_state(core);
 124
 125     efer->value = src.value;
 126
 127     {
 128         struct vmx_data * vmx_state = core->vmm_data;
 129
 130         V3_Print("Trapping page faults and GPFs\n");
 131         vmx_state->excp_bmap.pf = 1;
 132         vmx_state->excp_bmap.gp = 1;
 133
 134          check_vmcs_write(VMCS_EXCP_BITMAP, vmx_state->excp_bmap.value);
 135     }
 136
 137     return 0;
 138 }
 139 */
 140
 141
 142 static int init_vmcs_bios(struct guest_info * core, struct vmx_data * vmx_state) {
 143     int vmx_ret = 0;
 144
 145     /* Get Available features */
 146     struct vmx_pin_ctrls avail_pin_ctrls;
 147     avail_pin_ctrls.value = v3_vmx_get_ctrl_features(&(hw_info.pin_ctrls));
 148     /* ** */
 149
 150
 151     // disable global interrupts for vm state initialization
 152     v3_disable_ints();
 153
 154     PrintDebug("Loading VMCS\n");
 155     vmx_ret = vmcs_load(vmx_state->vmcs_ptr_phys);
 156     vmx_state->state = VMX_UNLAUNCHED;
 157
 158     if (vmx_ret != VMX_SUCCESS) {
 159         PrintError("VMPTRLD failed\n");
 160         return -1;
 161     }
 162
 163
 164     /*** Setup default state from HW ***/
 165
 166     vmx_state->pin_ctrls.value = hw_info.pin_ctrls.def_val;
 167     vmx_state->pri_proc_ctrls.value = hw_info.proc_ctrls.def_val;
 168     vmx_state->exit_ctrls.value = hw_info.exit_ctrls.def_val;
 169     vmx_state->entry_ctrls.value = hw_info.entry_ctrls.def_val;
 170     vmx_state->sec_proc_ctrls.value = hw_info.sec_proc_ctrls.def_val;
 171
 172     /* Print Control MSRs */
 173     PrintDebug("CR0 MSR: %p\n", (void *)(addr_t)hw_info.cr0.value);
 174     PrintDebug("CR4 MSR: %p\n", (void *)(addr_t)hw_info.cr4.value);
 175
 176
 177
 178     /******* Setup Host State **********/
 179
 180     /* Cache GDTR, IDTR, and TR in host struct */
 181
 182
 183     /********** Setup VMX Control Fields ***********/
 184
 185     /* Add external interrupts, NMI exiting, and virtual NMI */
 186     vmx_state->pin_ctrls.nmi_exit = 1;
 187     vmx_state->pin_ctrls.ext_int_exit = 1;
 188
 189
 190     /* We enable the preemption timer by default to measure accurate guest time */
 191     if (avail_pin_ctrls.active_preempt_timer) {
 192         V3_Print("VMX Preemption Timer is available\n");
 193         vmx_state->pin_ctrls.active_preempt_timer = 1;
 194         vmx_state->exit_ctrls.save_preempt_timer = 1;
 195     }
 196
 197     vmx_state->pri_proc_ctrls.hlt_exit = 1;
 198
 199
 200     vmx_state->pri_proc_ctrls.pause_exit = 0;
 201     vmx_state->pri_proc_ctrls.tsc_offset = 1;
 202 #ifdef V3_CONFIG_TIME_VIRTUALIZE_TSC
 203     vmx_state->pri_proc_ctrls.rdtsc_exit = 1;
 204 #endif
 205
 206     /* Setup IO map */
 207     vmx_state->pri_proc_ctrls.use_io_bitmap = 1;
 208     vmx_ret |= check_vmcs_write(VMCS_IO_BITMAP_A_ADDR, (addr_t)V3_PAddr(core->vm_info->io_map.arch_data));
 209     vmx_ret |= check_vmcs_write(VMCS_IO_BITMAP_B_ADDR,
 210             (addr_t)V3_PAddr(core->vm_info->io_map.arch_data) + PAGE_SIZE_4KB);
 211
 212
 213     vmx_state->pri_proc_ctrls.use_msr_bitmap = 1;
 214     vmx_ret |= check_vmcs_write(VMCS_MSR_BITMAP, (addr_t)V3_PAddr(core->vm_info->msr_map.arch_data));
 215
 216
 217
 218 #ifdef __V3_64BIT__
 219     // Ensure host runs in 64-bit mode at each VM EXIT
 220     vmx_state->exit_ctrls.host_64_on = 1;
 221 #endif
 222
 223
 224
 225     // Restore host's EFER register on each VM EXIT
 226     vmx_state->exit_ctrls.ld_efer = 1;
 227
 228     // Save/restore guest's EFER register to/from VMCS on VM EXIT/ENTRY
 229     vmx_state->exit_ctrls.save_efer = 1;
 230     vmx_state->entry_ctrls.ld_efer  = 1;
 231
 232     vmx_state->exit_ctrls.save_pat = 1;
 233     vmx_state->exit_ctrls.ld_pat = 1;
 234     vmx_state->entry_ctrls.ld_pat = 1;
 235
 236     /* Temporary GPF trap */
 237     //    vmx_state->excp_bmap.gp = 1;
 238
 239     // Setup Guests initial PAT field
 240     vmx_ret |= check_vmcs_write(VMCS_GUEST_PAT, 0x0007040600070406LL);
 241
 242     /* Setup paging */
 243     if (core->shdw_pg_mode == SHADOW_PAGING) {
 244         PrintDebug("Creating initial shadow page table\n");
 245
 246         if (v3_init_passthrough_pts(core) == -1) {
 247             PrintError("Could not initialize passthrough page tables\n");
 248             return -1;
 249         }
 250
 251 #define CR0_PE 0x00000001
 252 #define CR0_PG 0x80000000
 253 #define CR0_WP 0x00010000 // To ensure mem hooks work
 254         vmx_ret |= check_vmcs_write(VMCS_CR0_MASK, (CR0_PE | CR0_PG | CR0_WP));
 255
 256
 257         // Cause VM_EXIT whenever CR4.VMXE or CR4.PAE bits are written
 258         vmx_ret |= check_vmcs_write(VMCS_CR4_MASK, CR4_VMXE | CR4_PAE);
 259
 260         core->ctrl_regs.cr3 = core->direct_map_pt;
 261
 262         // vmx_state->pinbased_ctrls |= NMI_EXIT;
 263
 264         /* Add CR exits */
 265         vmx_state->pri_proc_ctrls.cr3_ld_exit = 1;
 266         vmx_state->pri_proc_ctrls.cr3_str_exit = 1;
 267
 268         vmx_state->pri_proc_ctrls.invlpg_exit = 1;
 269
 270         /* Add page fault exits */
 271         vmx_state->excp_bmap.pf = 1;
 272
 273         // Setup VMX Assist
 274         v3_vmxassist_init(core, vmx_state);
 275
 276         // Hook all accesses to EFER register
 277         v3_hook_msr(core->vm_info, EFER_MSR,
 278                     &v3_handle_efer_read,
 279                     &v3_handle_efer_write,
 280                     core);
 281
 282     } else if ((core->shdw_pg_mode == NESTED_PAGING) &&
 283                (v3_cpu_types[core->pcpu_id] == V3_VMX_EPT_CPU)) {
 284
 285 #define CR0_PE 0x00000001
 286 #define CR0_PG 0x80000000
 287 #define CR0_WP 0x00010000 // To ensure mem hooks work
 288         vmx_ret |= check_vmcs_write(VMCS_CR0_MASK, (CR0_PE | CR0_PG | CR0_WP));
 289
 290         // vmx_state->pinbased_ctrls |= NMI_EXIT;
 291
 292         // Cause VM_EXIT whenever CR4.VMXE or CR4.PAE bits are written
 293         vmx_ret |= check_vmcs_write(VMCS_CR4_MASK, CR4_VMXE | CR4_PAE);
 294
 295         /* Disable CR exits */
 296         vmx_state->pri_proc_ctrls.cr3_ld_exit = 0;
 297         vmx_state->pri_proc_ctrls.cr3_str_exit = 0;
 298
 299         vmx_state->pri_proc_ctrls.invlpg_exit = 0;
 300
 301         /* Add page fault exits */
 302         //      vmx_state->excp_bmap.pf = 1; // This should never happen..., enabled to catch bugs
 303
 304         // Setup VMX Assist
 305         v3_vmxassist_init(core, vmx_state);
 306
 307         /* Enable EPT */
 308         vmx_state->pri_proc_ctrls.sec_ctrls = 1; // Enable secondary proc controls
 309         vmx_state->sec_proc_ctrls.enable_ept = 1; // enable EPT paging
 310
 311
 312
 313         if (v3_init_ept(core, &hw_info) == -1) {
 314             PrintError("Error initializing EPT\n");
 315             return -1;
 316         }
 317
 318         // Hook all accesses to EFER register
 319         v3_hook_msr(core->vm_info, EFER_MSR, NULL, NULL, NULL);
 320
 321     } else if ((core->shdw_pg_mode == NESTED_PAGING) &&
 322                (v3_cpu_types[core->pcpu_id] == V3_VMX_EPT_UG_CPU)) {
 323         int i = 0;
 324         // For now we will assume that unrestricted guest mode is assured w/ EPT
 325
 326
 327         core->vm_regs.rsp = 0x00;
 328         core->rip = 0xfff0;
 329         core->vm_regs.rdx = 0x00000f00;
 330         core->ctrl_regs.rflags = 0x00000002; // The reserved bit is always 1
 331         core->ctrl_regs.cr0 = 0x00000030;
 332         core->ctrl_regs.cr4 = 0x00002010; // Enable VMX and PSE flag
 333
 334
 335         core->segments.cs.selector = 0xf000;
 336         core->segments.cs.limit = 0xffff;
 337         core->segments.cs.base = 0x0000000f0000LL;
 338
 339         // (raw attributes = 0xf3)
 340         core->segments.cs.type = 0xb;
 341         core->segments.cs.system = 0x1;
 342         core->segments.cs.dpl = 0x0;
 343         core->segments.cs.present = 1;
 344
 345
 346
 347         struct v3_segment * segregs [] = {&(core->segments.ss), &(core->segments.ds),
 348                                           &(core->segments.es), &(core->segments.fs),
 349                                           &(core->segments.gs), NULL};
 350
 351         for ( i = 0; segregs[i] != NULL; i++) {
 352             struct v3_segment * seg = segregs[i];
 353
 354             seg->selector = 0x0000;
 355             //    seg->base = seg->selector << 4;
 356             seg->base = 0x00000000;
 357             seg->limit = 0xffff;
 358
 359
 360             seg->type = 0x3;
 361             seg->system = 0x1;
 362             seg->dpl = 0x0;
 363             seg->present = 1;
 364             //    seg->granularity = 1;
 365
 366         }
 367
 368
 369         core->segments.gdtr.limit = 0x0000ffff;
 370         core->segments.gdtr.base = 0x0000000000000000LL;
 371
 372         core->segments.idtr.limit = 0x0000ffff;
 373         core->segments.idtr.base = 0x0000000000000000LL;
 374
 375         core->segments.ldtr.selector = 0x0000;
 376         core->segments.ldtr.limit = 0x0000ffff;
 377         core->segments.ldtr.base = 0x0000000000000000LL;
 378         core->segments.ldtr.type = 0x2;
 379         core->segments.ldtr.present = 1;
 380
 381         core->segments.tr.selector = 0x0000;
 382         core->segments.tr.limit = 0x0000ffff;
 383         core->segments.tr.base = 0x0000000000000000LL;
 384         core->segments.tr.type = 0xb;
 385         core->segments.tr.present = 1;
 386
 387         //      core->dbg_regs.dr6 = 0x00000000ffff0ff0LL;
 388         core->dbg_regs.dr7 = 0x0000000000000400LL;
 389
 390         /* Enable EPT */
 391         vmx_state->pri_proc_ctrls.sec_ctrls = 1; // Enable secondary proc controls
 392         vmx_state->sec_proc_ctrls.enable_ept = 1; // enable EPT paging
 393         vmx_state->sec_proc_ctrls.unrstrct_guest = 1; // enable unrestricted guest operation
 394
 395
 396         /* Disable shadow paging stuff */
 397         vmx_state->pri_proc_ctrls.cr3_ld_exit = 0;
 398         vmx_state->pri_proc_ctrls.cr3_str_exit = 0;
 399
 400         vmx_state->pri_proc_ctrls.invlpg_exit = 0;
 401
 402
 403         // Cause VM_EXIT whenever the CR4.VMXE bit is set
 404         vmx_ret |= check_vmcs_write(VMCS_CR4_MASK, CR4_VMXE);
 405
 406
 407         if (v3_init_ept(core, &hw_info) == -1) {
 408             PrintError("Error initializing EPT\n");
 409             return -1;
 410         }
 411
 412         // Hook all accesses to EFER register
 413         //v3_hook_msr(core->vm_info, EFER_MSR, &debug_efer_read, &debug_efer_write, core);
 414         v3_hook_msr(core->vm_info, EFER_MSR, NULL, NULL, NULL);
 415     } else {
 416         PrintError("Invalid Virtual paging mode\n");
 417         return -1;
 418     }
 419
 420
 421     // hook vmx msrs
 422
 423     // Setup SYSCALL/SYSENTER MSRs in load/store area
 424
 425     // save STAR, LSTAR, FMASK, KERNEL_GS_BASE MSRs in MSR load/store area
 426     {
 427
 428         struct vmcs_msr_save_area * msr_entries = NULL;
 429         int max_msrs = (hw_info.misc_info.max_msr_cache_size + 1) * 4;
 430         int msr_ret = 0;
 431
 432         V3_Print("Setting up MSR load/store areas (max_msr_count=%d)\n", max_msrs);
 433
 434         if (max_msrs < 4) {
 435             PrintError("Max MSR cache size is too small (%d)\n", max_msrs);
 436             return -1;
 437         }
 438
 439         vmx_state->msr_area_paddr = (addr_t)V3_AllocPages(1);
 440
 441         if (vmx_state->msr_area_paddr == (addr_t)NULL) {
 442             PrintError("could not allocate msr load/store area\n");
 443             return -1;
 444         }
 445
 446         msr_entries = (struct vmcs_msr_save_area *)V3_VAddr((void *)(vmx_state->msr_area_paddr));
 447         vmx_state->msr_area = msr_entries; // cache in vmx_info
 448
 449         memset(msr_entries, 0, PAGE_SIZE);
 450
 451         msr_entries->guest_star.index = IA32_STAR_MSR;
 452         msr_entries->guest_lstar.index = IA32_LSTAR_MSR;
 453         msr_entries->guest_fmask.index = IA32_FMASK_MSR;
 454         msr_entries->guest_kern_gs.index = IA32_KERN_GS_BASE_MSR;
 455
 456         msr_entries->host_star.index = IA32_STAR_MSR;
 457         msr_entries->host_lstar.index = IA32_LSTAR_MSR;
 458         msr_entries->host_fmask.index = IA32_FMASK_MSR;
 459         msr_entries->host_kern_gs.index = IA32_KERN_GS_BASE_MSR;
 460
 461         msr_ret |= check_vmcs_write(VMCS_EXIT_MSR_STORE_CNT, 4);
 462         msr_ret |= check_vmcs_write(VMCS_EXIT_MSR_LOAD_CNT, 4);
 463         msr_ret |= check_vmcs_write(VMCS_ENTRY_MSR_LOAD_CNT, 4);
 464
 465         msr_ret |= check_vmcs_write(VMCS_EXIT_MSR_STORE_ADDR, (addr_t)V3_PAddr(msr_entries->guest_msrs));
 466         msr_ret |= check_vmcs_write(VMCS_ENTRY_MSR_LOAD_ADDR, (addr_t)V3_PAddr(msr_entries->guest_msrs));
 467         msr_ret |= check_vmcs_write(VMCS_EXIT_MSR_LOAD_ADDR, (addr_t)V3_PAddr(msr_entries->host_msrs));
 468
 469
 470         msr_ret |= v3_hook_msr(core->vm_info, IA32_STAR_MSR, NULL, NULL, NULL);
 471         msr_ret |= v3_hook_msr(core->vm_info, IA32_LSTAR_MSR, NULL, NULL, NULL);
 472         msr_ret |= v3_hook_msr(core->vm_info, IA32_FMASK_MSR, NULL, NULL, NULL);
 473         msr_ret |= v3_hook_msr(core->vm_info, IA32_KERN_GS_BASE_MSR, NULL, NULL, NULL);
 474
 475
 476         // IMPORTANT: These MSRs appear to be cached by the hardware....
 477         msr_ret |= v3_hook_msr(core->vm_info, SYSENTER_CS_MSR, NULL, NULL, NULL);
 478         msr_ret |= v3_hook_msr(core->vm_info, SYSENTER_ESP_MSR, NULL, NULL, NULL);
 479         msr_ret |= v3_hook_msr(core->vm_info, SYSENTER_EIP_MSR, NULL, NULL, NULL);
 480
 481         msr_ret |= v3_hook_msr(core->vm_info, FS_BASE_MSR, NULL, NULL, NULL);
 482         msr_ret |= v3_hook_msr(core->vm_info, GS_BASE_MSR, NULL, NULL, NULL);
 483
 484         msr_ret |= v3_hook_msr(core->vm_info, IA32_PAT_MSR, NULL, NULL, NULL);
 485
 486         // Not sure what to do about this... Does not appear to be an explicit hardware cache version...
 487         msr_ret |= v3_hook_msr(core->vm_info, IA32_CSTAR_MSR, NULL, NULL, NULL);
 488
 489         if (msr_ret != 0) {
 490             PrintError("Error configuring MSR save/restore area\n");
 491             return -1;
 492         }
 493
 494
 495     }
 496
 497     /* Sanity check ctrl/reg fields against hw_defaults */
 498
 499
 500
 501
 502     /*** Write all the info to the VMCS ***/
 503
 504     /*
 505     {
 506         // IS THIS NECESSARY???
 507 #define DEBUGCTL_MSR 0x1d9
 508         struct v3_msr tmp_msr;
 509         v3_get_msr(DEBUGCTL_MSR, &(tmp_msr.hi), &(tmp_msr.lo));
 510         vmx_ret |= check_vmcs_write(VMCS_GUEST_DBG_CTL, tmp_msr.value);
 511         core->dbg_regs.dr7 = 0x400;
 512     }
 513     */
 514
 515 #ifdef __V3_64BIT__
 516     vmx_ret |= check_vmcs_write(VMCS_LINK_PTR, (addr_t)0xffffffffffffffffULL);
 517 #else
 518     vmx_ret |= check_vmcs_write(VMCS_LINK_PTR, (addr_t)0xffffffffUL);
 519     vmx_ret |= check_vmcs_write(VMCS_LINK_PTR_HIGH, (addr_t)0xffffffffUL);
 520 #endif
 521
 522
 523
 524
 525     if (v3_update_vmcs_ctrl_fields(core)) {
 526         PrintError("Could not write control fields!\n");
 527         return -1;
 528     }
 529
 530     /*
 531     if (v3_update_vmcs_host_state(core)) {
 532         PrintError("Could not write host state\n");
 533         return -1;
 534     }
 535     */
 536
 537     // reenable global interrupts for vm state initialization now
 538     // that the vm state is initialized. If another VM kicks us off,
 539     // it'll update our vmx state so that we know to reload ourself
 540     v3_enable_ints();
 541
 542     return 0;
 543 }
 544
 545 int v3_init_vmx_vmcs(struct guest_info * core, v3_vm_class_t vm_class) {
 546     struct vmx_data * vmx_state = NULL;
 547     int vmx_ret = 0;
 548
 549     vmx_state = (struct vmx_data *)V3_Malloc(sizeof(struct vmx_data));
 550     memset(vmx_state, 0, sizeof(struct vmx_data));
 551
 552     PrintDebug("vmx_data pointer: %p\n", (void *)vmx_state);
 553
 554     PrintDebug("Allocating VMCS\n");
 555     vmx_state->vmcs_ptr_phys = allocate_vmcs();
 556
 557     PrintDebug("VMCS pointer: %p\n", (void *)(vmx_state->vmcs_ptr_phys));
 558
 559     core->vmm_data = vmx_state;
 560     vmx_state->state = VMX_UNLAUNCHED;
 561
 562     PrintDebug("Initializing VMCS (addr=%p)\n", core->vmm_data);
 563
 564     // TODO: Fix vmcs fields so they're 32-bit
 565
 566     PrintDebug("Clearing VMCS: %p\n", (void *)vmx_state->vmcs_ptr_phys);
 567     vmx_ret = vmcs_clear(vmx_state->vmcs_ptr_phys);
 568
 569     if (vmx_ret != VMX_SUCCESS) {
 570         PrintError("VMCLEAR failed\n");
 571         return -1;
 572     }
 573
 574     if (vm_class == V3_PC_VM) {
 575         PrintDebug("Initializing VMCS\n");
 576         if (init_vmcs_bios(core, vmx_state) == -1) {
 577             PrintError("Error initializing VMCS to BIOS state\n");
 578             return -1;
 579         }
 580     } else {
 581         PrintError("Invalid VM Class\n");
 582         return -1;
 583     }
 584
 585     PrintDebug("Serializing VMCS: %p\n", (void *)vmx_state->vmcs_ptr_phys);
 586     vmx_ret = vmcs_clear(vmx_state->vmcs_ptr_phys);
 587
 588     return 0;
 589 }
 590
 591
 592 int v3_deinit_vmx_vmcs(struct guest_info * core) {
 593     struct vmx_data * vmx_state = core->vmm_data;
 594
 595     V3_FreePages((void *)(vmx_state->vmcs_ptr_phys), 1);
 596     V3_FreePages(V3_PAddr(vmx_state->msr_area), 1);
 597
 598     V3_Free(vmx_state);
 599
 600     return 0;
 601 }
 602
 603
 604
 605 #ifdef V3_CONFIG_CHECKPOINT
 606 /*
 607  * JRL: This is broken
 608  */
 609 int v3_vmx_save_core(struct guest_info * core, void * ctx){
 610     uint64_t vmcs_ptr = vmcs_store();
 611
 612     v3_chkpt_save(ctx, "vmcs_data", PAGE_SIZE, (void *)vmcs_ptr);
 613
 614     return 0;
 615 }
 616
 617 int v3_vmx_load_core(struct guest_info * core, void * ctx){
 618     struct vmx_data * vmx_info = (struct vmx_data *)(core->vmm_data);
 619     struct cr0_32 * shadow_cr0;
 620     char vmcs[PAGE_SIZE_4KB];
 621
 622     v3_chkpt_load(ctx, "vmcs_data", PAGE_SIZE_4KB, vmcs);
 623
 624     vmcs_clear(vmx_info->vmcs_ptr_phys);
 625     vmcs_load((addr_t)vmcs);
 626
 627     v3_vmx_save_vmcs(core);
 628
 629     shadow_cr0 = (struct cr0_32 *)&(core->ctrl_regs.cr0);
 630
 631
 632     /* Get the CPU mode to set the guest_ia32e entry ctrl */
 633
 634     if (core->shdw_pg_mode == SHADOW_PAGING) {
 635         if (v3_get_vm_mem_mode(core) == VIRTUAL_MEM) {
 636             if (v3_activate_shadow_pt(core) == -1) {
 637                 PrintError("Failed to activate shadow page tables\n");
 638                 return -1;
 639             }
 640         } else {
 641             if (v3_activate_passthrough_pt(core) == -1) {
 642                 PrintError("Failed to activate passthrough page tables\n");
 643                 return -1;
 644             }
 645         }
 646     }
 647
 648     return 0;
 649 }
 650 #endif
 651
 652
 653 void v3_flush_vmx_vm_core(struct guest_info * core) {
 654     struct vmx_data * vmx_info = (struct vmx_data *)(core->vmm_data);
 655     vmcs_clear(vmx_info->vmcs_ptr_phys);
 656     vmx_info->state = VMX_UNLAUNCHED;
 657 }
 658
 659
 660
 661 static int update_irq_exit_state(struct guest_info * info) {
 662     struct vmx_exit_idt_vec_info idt_vec_info;
 663
 664     check_vmcs_read(VMCS_IDT_VECTOR_INFO, &(idt_vec_info.value));
 665
 666     if ((info->intr_core_state.irq_started == 1) && (idt_vec_info.valid == 0)) {
 667 #ifdef V3_CONFIG_DEBUG_INTERRUPTS
 668         V3_Print("Calling v3_injecting_intr\n");
 669 #endif
 670         info->intr_core_state.irq_started = 0;
 671         v3_injecting_intr(info, info->intr_core_state.irq_vector, V3_EXTERNAL_IRQ);
 672     }
 673
 674     return 0;
 675 }
 676
 677 static int update_irq_entry_state(struct guest_info * info) {
 678     struct vmx_exit_idt_vec_info idt_vec_info;
 679     struct vmcs_interrupt_state intr_core_state;
 680     struct vmx_data * vmx_info = (struct vmx_data *)(info->vmm_data);
 681
 682     check_vmcs_read(VMCS_IDT_VECTOR_INFO, &(idt_vec_info.value));
 683     check_vmcs_read(VMCS_GUEST_INT_STATE, &(intr_core_state));
 684
 685     /* Check for pending exceptions to inject */
 686     if (v3_excp_pending(info)) {
 687         struct vmx_entry_int_info int_info;
 688         int_info.value = 0;
 689
 690         // In VMX, almost every exception is hardware
 691         // Software exceptions are pretty much only for breakpoint or overflow
 692         int_info.type = 3;
 693         int_info.vector = v3_get_excp_number(info);
 694
 695         if (info->excp_state.excp_error_code_valid) {
 696             check_vmcs_write(VMCS_ENTRY_EXCP_ERR, info->excp_state.excp_error_code);
 697             int_info.error_code = 1;
 698
 699 #ifdef V3_CONFIG_DEBUG_INTERRUPTS
 700             V3_Print("Injecting exception %d with error code %x\n",
 701                     int_info.vector, info->excp_state.excp_error_code);
 702 #endif
 703         }
 704
 705         int_info.valid = 1;
 706 #ifdef V3_CONFIG_DEBUG_INTERRUPTS
 707         V3_Print("Injecting exception %d (EIP=%p)\n", int_info.vector, (void *)(addr_t)info->rip);
 708 #endif
 709         check_vmcs_write(VMCS_ENTRY_INT_INFO, int_info.value);
 710
 711         v3_injecting_excp(info, int_info.vector);
 712
 713     } else if ((((struct rflags *)&(info->ctrl_regs.rflags))->intr == 1) &&
 714                (intr_core_state.val == 0)) {
 715
 716         if ((info->intr_core_state.irq_started == 1) && (idt_vec_info.valid == 1)) {
 717
 718 #ifdef V3_CONFIG_DEBUG_INTERRUPTS
 719             V3_Print("IRQ pending from previous injection\n");
 720 #endif
 721
 722             // Copy the IDT vectoring info over to reinject the old interrupt
 723             if (idt_vec_info.error_code == 1) {
 724                 uint32_t err_code = 0;
 725
 726                 check_vmcs_read(VMCS_IDT_VECTOR_ERR, &err_code);
 727                 check_vmcs_write(VMCS_ENTRY_EXCP_ERR, err_code);
 728             }
 729
 730             idt_vec_info.undef = 0;
 731             check_vmcs_write(VMCS_ENTRY_INT_INFO, idt_vec_info.value);
 732
 733         } else {
 734             struct vmx_entry_int_info ent_int;
 735             ent_int.value = 0;
 736
 737             switch (v3_intr_pending(info)) {
 738                 case V3_EXTERNAL_IRQ: {
 739                     info->intr_core_state.irq_vector = v3_get_intr(info);
 740                     ent_int.vector = info->intr_core_state.irq_vector;
 741                     ent_int.type = 0;
 742                     ent_int.error_code = 0;
 743                     ent_int.valid = 1;
 744
 745 #ifdef V3_CONFIG_DEBUG_INTERRUPTS
 746                     V3_Print("Injecting Interrupt %d at exit %u(EIP=%p)\n",
 747                                info->intr_core_state.irq_vector,
 748                                (uint32_t)info->num_exits,
 749                                (void *)(addr_t)info->rip);
 750 #endif
 751
 752                     check_vmcs_write(VMCS_ENTRY_INT_INFO, ent_int.value);
 753                     info->intr_core_state.irq_started = 1;
 754
 755                     break;
 756                 }
 757                 case V3_NMI:
 758                     PrintDebug("Injecting NMI\n");
 759
 760                     ent_int.type = 2;
 761                     ent_int.vector = 2;
 762                     ent_int.valid = 1;
 763                     check_vmcs_write(VMCS_ENTRY_INT_INFO, ent_int.value);
 764
 765                     break;
 766                 case V3_SOFTWARE_INTR:
 767                     PrintDebug("Injecting software interrupt\n");
 768                     ent_int.type = 4;
 769
 770                     ent_int.valid = 1;
 771                     check_vmcs_write(VMCS_ENTRY_INT_INFO, ent_int.value);
 772
 773                     break;
 774                 case V3_VIRTUAL_IRQ:
 775                     // Not sure what to do here, Intel doesn't have virtual IRQs
 776                     // May be the same as external interrupts/IRQs
 777
 778                     break;
 779                 case V3_INVALID_INTR:
 780                 default:
 781                     break;
 782             }
 783         }
 784     } else if ((v3_intr_pending(info)) && (vmx_info->pri_proc_ctrls.int_wndw_exit == 0)) {
 785         // Enable INTR window exiting so we know when IF=1
 786         uint32_t instr_len;
 787
 788         check_vmcs_read(VMCS_EXIT_INSTR_LEN, &instr_len);
 789
 790 #ifdef V3_CONFIG_DEBUG_INTERRUPTS
 791         V3_Print("Enabling Interrupt-Window exiting: %d\n", instr_len);
 792 #endif
 793
 794         vmx_info->pri_proc_ctrls.int_wndw_exit = 1;
 795         check_vmcs_write(VMCS_PROC_CTRLS, vmx_info->pri_proc_ctrls.value);
 796     }
 797
 798
 799     return 0;
 800 }
 801
 802
 803
 804 static struct vmx_exit_info exit_log[10];
 805 static uint64_t rip_log[10];
 806
 807
 808
 809 static void print_exit_log(struct guest_info * info) {
 810     int cnt = info->num_exits % 10;
 811     int i = 0;
 812
 813
 814     V3_Print("\nExit Log (%d total exits):\n", (uint32_t)info->num_exits);
 815
 816     for (i = 0; i < 10; i++) {
 817         struct vmx_exit_info * tmp = &exit_log[cnt];
 818
 819         V3_Print("%d:\texit_reason = %p\n", i, (void *)(addr_t)tmp->exit_reason);
 820         V3_Print("\texit_qual = %p\n", (void *)tmp->exit_qual);
 821         V3_Print("\tint_info = %p\n", (void *)(addr_t)tmp->int_info);
 822         V3_Print("\tint_err = %p\n", (void *)(addr_t)tmp->int_err);
 823         V3_Print("\tinstr_info = %p\n", (void *)(addr_t)tmp->instr_info);
 824         V3_Print("\tguest_linear_addr= %p\n", (void *)(addr_t)tmp->guest_linear_addr);
 825         V3_Print("\tRIP = %p\n", (void *)rip_log[cnt]);
 826
 827
 828         cnt--;
 829
 830         if (cnt == -1) {
 831             cnt = 9;
 832         }
 833
 834     }
 835
 836 }
 837
 838
 839
 840 /*
 841  * CAUTION and DANGER!!!
 842  *
 843  * The VMCS CANNOT(!!) be accessed outside of the cli/sti calls inside this function
 844  * When exectuing a symbiotic call, the VMCS WILL be overwritten, so any dependencies
 845  * on its contents will cause things to break. The contents at the time of the exit WILL
 846  * change before the exit handler is executed.
 847  */
 848 int v3_vmx_enter(struct guest_info * info) {
 849     int ret = 0;
 850     sint64_t tsc_offset;
 851     uint32_t tsc_offset_low, tsc_offset_high;
 852     struct vmx_exit_info exit_info;
 853     struct vmx_data * vmx_info = (struct vmx_data *)(info->vmm_data);
 854     uint64_t guest_cycles = 0;
 855
 856     // Conditionally yield the CPU if the timeslice has expired
 857     v3_yield_cond(info);
 858
 859     // disable global interrupts for vm state transition
 860     v3_disable_ints();
 861
 862     // Update timer devices late after being in the VM so that as much
 863     // of the time in the VM is accounted for as possible. Also do it before
 864     // updating IRQ entry state so that any interrupts the timers raise get
 865     // handled on the next VM entry. Must be done with interrupts disabled.
 866     v3_advance_time(info);
 867     v3_update_timers(info);
 868
 869     if (vmcs_store() != vmx_info->vmcs_ptr_phys) {
 870         vmcs_clear(vmx_info->vmcs_ptr_phys);
 871         vmcs_load(vmx_info->vmcs_ptr_phys);
 872         vmx_info->state = VMX_UNLAUNCHED;
 873     }
 874
 875     v3_vmx_restore_vmcs(info);
 876
 877
 878 #ifdef V3_CONFIG_SYMCALL
 879     if (info->sym_core_state.symcall_state.sym_call_active == 0) {
 880         update_irq_entry_state(info);
 881     }
 882 #else
 883     update_irq_entry_state(info);
 884 #endif
 885
 886     {
 887         addr_t guest_cr3;
 888         vmcs_read(VMCS_GUEST_CR3, &guest_cr3);
 889         vmcs_write(VMCS_GUEST_CR3, guest_cr3);
 890     }
 891
 892
 893     // Perform last-minute time bookkeeping prior to entering the VM
 894     v3_time_enter_vm(info);
 895
 896     tsc_offset = v3_tsc_host_offset(&info->time_state);
 897     tsc_offset_high = (uint32_t)(( tsc_offset >> 32) & 0xffffffff);
 898     tsc_offset_low = (uint32_t)(tsc_offset & 0xffffffff);
 899
 900     check_vmcs_write(VMCS_TSC_OFFSET_HIGH, tsc_offset_high);
 901     check_vmcs_write(VMCS_TSC_OFFSET, tsc_offset_low);
 902
 903
 904
 905     if (v3_update_vmcs_host_state(info)) {
 906         v3_enable_ints();
 907         PrintError("Could not write host state\n");
 908         return -1;
 909     }
 910
 911     if (vmx_info->pin_ctrls.active_preempt_timer) {
 912         /* Preemption timer is active */
 913         uint32_t preempt_window = 0xffffffff;
 914
 915         if (info->timeouts.timeout_active) {
 916             preempt_window = info->timeouts.next_timeout;
 917         }
 918
 919         check_vmcs_write(VMCS_PREEMPT_TIMER, preempt_window);
 920     }
 921
 922
 923     {
 924         uint64_t entry_tsc = 0;
 925         uint64_t exit_tsc = 0;
 926
 927         if (vmx_info->state == VMX_UNLAUNCHED) {
 928             vmx_info->state = VMX_LAUNCHED;
 929             rdtscll(entry_tsc);
 930             ret = v3_vmx_launch(&(info->vm_regs), info, &(info->ctrl_regs));
 931             rdtscll(exit_tsc);
 932
 933         } else {
 934             V3_ASSERT(vmx_info->state != VMX_UNLAUNCHED);
 935             rdtscll(entry_tsc);
 936             ret = v3_vmx_resume(&(info->vm_regs), info, &(info->ctrl_regs));
 937             rdtscll(exit_tsc);
 938         }
 939
 940         guest_cycles = exit_tsc - entry_tsc;
 941     }
 942
 943     //  PrintDebug("VMX Exit: ret=%d\n", ret);
 944
 945     if (ret != VMX_SUCCESS) {
 946         uint32_t error = 0;
 947         vmcs_read(VMCS_INSTR_ERR, &error);
 948
 949         v3_enable_ints();
 950
 951         PrintError("VMENTRY Error: %d (launch_ret = %d)\n", error, ret);
 952         return -1;
 953     }
 954
 955
 956     info->num_exits++;
 957
 958     /* If we have the preemption time, then use it to get more accurate guest time */
 959     if (vmx_info->pin_ctrls.active_preempt_timer) {
 960         uint32_t cycles_left = 0;
 961         check_vmcs_read(VMCS_PREEMPT_TIMER, &(cycles_left));
 962
 963         if (info->timeouts.timeout_active) {
 964             guest_cycles = info->timeouts.next_timeout - cycles_left;
 965         } else {
 966             guest_cycles = 0xffffffff - cycles_left;
 967         }
 968     }
 969
 970     // Immediate exit from VM time bookkeeping
 971     v3_time_exit_vm(info, &guest_cycles);
 972
 973
 974     /* Update guest state */
 975     v3_vmx_save_vmcs(info);
 976
 977     // info->cpl = info->segments.cs.selector & 0x3;
 978
 979     info->mem_mode = v3_get_vm_mem_mode(info);
 980     info->cpu_mode = v3_get_vm_cpu_mode(info);
 981
 982
 983
 984     check_vmcs_read(VMCS_EXIT_INSTR_LEN, &(exit_info.instr_len));
 985     check_vmcs_read(VMCS_EXIT_INSTR_INFO, &(exit_info.instr_info));
 986     check_vmcs_read(VMCS_EXIT_REASON, &(exit_info.exit_reason));
 987     check_vmcs_read(VMCS_EXIT_QUAL, &(exit_info.exit_qual));
 988     check_vmcs_read(VMCS_EXIT_INT_INFO, &(exit_info.int_info));
 989     check_vmcs_read(VMCS_EXIT_INT_ERR, &(exit_info.int_err));
 990     check_vmcs_read(VMCS_GUEST_LINEAR_ADDR, &(exit_info.guest_linear_addr));
 991
 992     if (info->shdw_pg_mode == NESTED_PAGING) {
 993         check_vmcs_read(VMCS_GUEST_PHYS_ADDR, &(exit_info.ept_fault_addr));
 994     }
 995
 996     //PrintDebug("VMX Exit taken, id-qual: %u-%lu\n", exit_info.exit_reason, exit_info.exit_qual);
 997
 998     exit_log[info->num_exits % 10] = exit_info;
 999     rip_log[info->num_exits % 10] = get_addr_linear(info, info->rip, &(info->segments.cs));
1000
1001 #ifdef V3_CONFIG_SYMCALL
1002     if (info->sym_core_state.symcall_state.sym_call_active == 0) {
1003         update_irq_exit_state(info);
1004     }
1005 #else
1006     update_irq_exit_state(info);
1007 #endif
1008
1009     if (exit_info.exit_reason == VMEXIT_INTR_WINDOW) {
1010         // This is a special case whose only job is to inject an interrupt
1011         vmcs_read(VMCS_PROC_CTRLS, &(vmx_info->pri_proc_ctrls.value));
1012         vmx_info->pri_proc_ctrls.int_wndw_exit = 0;
1013         vmcs_write(VMCS_PROC_CTRLS, vmx_info->pri_proc_ctrls.value);
1014
1015 #ifdef V3_CONFIG_DEBUG_INTERRUPTS
1016        V3_Print("Interrupts available again! (RIP=%llx)\n", info->rip);
1017 #endif
1018     }
1019
1020     // reenable global interrupts after vm exit
1021     v3_enable_ints();
1022
1023     // Conditionally yield the CPU if the timeslice has expired
1024     v3_yield_cond(info);
1025
1026     if (v3_handle_vmx_exit(info, &exit_info) == -1) {
1027         PrintError("Error in VMX exit handler (Exit reason=%x)\n", exit_info.exit_reason);
1028         return -1;
1029     }
1030
1031     if (info->timeouts.timeout_active) {
1032         /* Check to see if any timeouts have expired */
1033         v3_handle_timeouts(info, guest_cycles);
1034     }
1035
1036     return 0;
1037 }
1038
1039
1040 int v3_start_vmx_guest(struct guest_info * info) {
1041
1042     PrintDebug("Starting VMX core %u\n", info->vcpu_id);
1043
1044     if (info->vcpu_id == 0) {
1045         info->core_run_state = CORE_RUNNING;
1046     } else {
1047
1048         PrintDebug("VMX core %u: Waiting for core initialization\n", info->vcpu_id);
1049
1050         while (info->core_run_state == CORE_STOPPED) {
1051
1052             if (info->vm_info->run_state == VM_STOPPED) {
1053                 // The VM was stopped before this core was initialized.
1054                 return 0;
1055             }
1056
1057             v3_yield(info);
1058             //PrintDebug("VMX core %u: still waiting for INIT\n",info->vcpu_id);
1059         }
1060
1061         PrintDebug("VMX core %u initialized\n", info->vcpu_id);
1062
1063         // We'll be paranoid about race conditions here
1064         v3_wait_at_barrier(info);
1065     }
1066
1067
1068     PrintDebug("VMX core %u: I am starting at CS=0x%x (base=0x%p, limit=0x%x),  RIP=0x%p\n",
1069                info->vcpu_id, info->segments.cs.selector, (void *)(info->segments.cs.base),
1070                info->segments.cs.limit, (void *)(info->rip));
1071
1072
1073     PrintDebug("VMX core %u: Launching VMX VM on logical core %u\n", info->vcpu_id, info->pcpu_id);
1074
1075     v3_start_time(info);
1076
1077     while (1) {
1078
1079         if (info->vm_info->run_state == VM_STOPPED) {
1080             info->core_run_state = CORE_STOPPED;
1081             break;
1082         }
1083
1084         if (v3_vmx_enter(info) == -1) {
1085
1086             addr_t host_addr;
1087             addr_t linear_addr = 0;
1088
1089             info->vm_info->run_state = VM_ERROR;
1090
1091             V3_Print("VMX core %u: VMX ERROR!!\n", info->vcpu_id);
1092
1093             v3_print_guest_state(info);
1094
1095             V3_Print("VMX core %u\n", info->vcpu_id);
1096
1097             linear_addr = get_addr_linear(info, info->rip, &(info->segments.cs));
1098
1099             if (info->mem_mode == PHYSICAL_MEM) {
1100                 v3_gpa_to_hva(info, linear_addr, &host_addr);
1101             } else if (info->mem_mode == VIRTUAL_MEM) {
1102                 v3_gva_to_hva(info, linear_addr, &host_addr);
1103             }
1104
1105             V3_Print("VMX core %u: Host Address of rip = 0x%p\n", info->vcpu_id, (void *)host_addr);
1106
1107             V3_Print("VMX core %u: Instr (15 bytes) at %p:\n", info->vcpu_id, (void *)host_addr);
1108             v3_dump_mem((uint8_t *)host_addr, 15);
1109
1110             v3_print_stack(info);
1111
1112
1113             v3_print_vmcs();
1114             print_exit_log(info);
1115             return -1;
1116         }
1117
1118         v3_wait_at_barrier(info);
1119
1120
1121         if (info->vm_info->run_state == VM_STOPPED) {
1122             info->core_run_state = CORE_STOPPED;
1123             break;
1124         }
1125 /*
1126         if ((info->num_exits % 5000) == 0) {
1127             V3_Print("VMX Exit number %d\n", (uint32_t)info->num_exits);
1128         }
1129 */
1130
1131     }
1132
1133     return 0;
1134 }
1135
1136
1137
1138
1139 #define VMX_FEATURE_CONTROL_MSR     0x0000003a
1140 #define CPUID_VMX_FEATURES 0x00000005  /* LOCK and VMXON */
1141 #define CPUID_1_ECX_VTXFLAG 0x00000020
1142
1143 int v3_is_vmx_capable() {
1144     v3_msr_t feature_msr;
1145     uint32_t eax = 0, ebx = 0, ecx = 0, edx = 0;
1146
1147     v3_cpuid(0x1, &eax, &ebx, &ecx, &edx);
1148
1149     PrintDebug("ECX: 0x%x\n", ecx);
1150
1151     if (ecx & CPUID_1_ECX_VTXFLAG) {
1152         v3_get_msr(VMX_FEATURE_CONTROL_MSR, &(feature_msr.hi), &(feature_msr.lo));
1153
1154         PrintDebug("MSRREGlow: 0x%.8x\n", feature_msr.lo);
1155
1156         if ((feature_msr.lo & CPUID_VMX_FEATURES) != CPUID_VMX_FEATURES) {
1157             PrintDebug("VMX is locked -- enable in the BIOS\n");
1158             return 0;
1159         }
1160
1161     } else {
1162         PrintDebug("VMX not supported on this cpu\n");
1163         return 0;
1164     }
1165
1166     return 1;
1167 }
1168
1169
1170 int v3_reset_vmx_vm_core(struct guest_info * core, addr_t rip) {
1171     // init vmcs bios
1172
1173     if ((core->shdw_pg_mode == NESTED_PAGING) &&
1174         (v3_cpu_types[core->pcpu_id] == V3_VMX_EPT_UG_CPU)) {
1175         // easy
1176         core->rip = 0;
1177         core->segments.cs.selector = rip << 8;
1178         core->segments.cs.limit = 0xffff;
1179         core->segments.cs.base = rip << 12;
1180     } else {
1181         core->vm_regs.rdx = core->vcpu_id;
1182         core->vm_regs.rbx = rip;
1183     }
1184
1185     return 0;
1186 }
1187
1188
1189
1190 void v3_init_vmx_cpu(int cpu_id) {
1191     addr_t vmx_on_region = 0;
1192     extern v3_cpu_arch_t v3_mach_type;
1193
1194     if (v3_mach_type == V3_INVALID_CPU) {
1195         if (v3_init_vmx_hw(&hw_info) == -1) {
1196             PrintError("Could not initialize VMX hardware features on cpu %d\n", cpu_id);
1197             return;
1198         }
1199     }
1200
1201     enable_vmx();
1202
1203
1204     // Setup VMXON Region
1205     vmx_on_region = allocate_vmcs();
1206
1207
1208     if (vmx_on(vmx_on_region) == VMX_SUCCESS) {
1209         V3_Print("VMX Enabled\n");
1210         host_vmcs_ptrs[cpu_id] = vmx_on_region;
1211     } else {
1212         V3_Print("VMX already enabled\n");
1213         V3_FreePages((void *)vmx_on_region, 1);
1214     }
1215
1216     PrintDebug("VMXON pointer: 0x%p\n", (void *)host_vmcs_ptrs[cpu_id]);
1217
1218     {
1219         struct vmx_sec_proc_ctrls sec_proc_ctrls;
1220         sec_proc_ctrls.value = v3_vmx_get_ctrl_features(&(hw_info.sec_proc_ctrls));
1221
1222         if (sec_proc_ctrls.enable_ept == 0) {
1223             V3_Print("VMX EPT (Nested) Paging not supported\n");
1224             v3_cpu_types[cpu_id] = V3_VMX_CPU;
1225         } else if (sec_proc_ctrls.unrstrct_guest == 0) {
1226             V3_Print("VMX EPT (Nested) Paging supported\n");
1227             v3_cpu_types[cpu_id] = V3_VMX_EPT_CPU;
1228         } else {
1229             V3_Print("VMX EPT (Nested) Paging + Unrestricted guest supported\n");
1230             v3_cpu_types[cpu_id] = V3_VMX_EPT_UG_CPU;
1231         }
1232     }
1233
1234 }
1235
1236
1237 void v3_deinit_vmx_cpu(int cpu_id) {
1238     extern v3_cpu_arch_t v3_cpu_types[];
1239     v3_cpu_types[cpu_id] = V3_INVALID_CPU;
1240
1241     if (host_vmcs_ptrs[cpu_id] != 0) {
1242         V3_Print("Disabling VMX\n");
1243
1244         if (vmx_off() != VMX_SUCCESS) {
1245             PrintError("Error executing VMXOFF\n");
1246         }
1247
1248         V3_FreePages((void *)host_vmcs_ptrs[cpu_id], 1);
1249
1250         host_vmcs_ptrs[cpu_id] = 0;
1251     }
1252 }