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
  37 #ifdef V3_CONFIG_CHECKPOINT
  38 #include <palacios/vmm_checkpoint.h>
  39 #endif
  40
  41 #include <palacios/vmx_ept.h>
  42 #include <palacios/vmx_assist.h>
  43 #include <palacios/vmx_hw_info.h>
  44
  45 #ifndef V3_CONFIG_DEBUG_VMX
  46 #undef PrintDebug
  47 #define PrintDebug(fmt, args...)
  48 #endif
  49
  50
  51 /* These fields contain the hardware feature sets supported by the local CPU */
  52 static struct vmx_hw_info hw_info;
  53
  54 extern v3_cpu_arch_t v3_cpu_types[];
  55
  56 static addr_t host_vmcs_ptrs[V3_CONFIG_MAX_CPUS] = { [0 ... V3_CONFIG_MAX_CPUS - 1] = 0};
  57
  58 extern int v3_vmx_launch(struct v3_gprs * vm_regs, struct guest_info * info, struct v3_ctrl_regs * ctrl_regs);
  59 extern int v3_vmx_resume(struct v3_gprs * vm_regs, struct guest_info * info, struct v3_ctrl_regs * ctrl_regs);
  60
  61 static int inline check_vmcs_write(vmcs_field_t field, addr_t val) {
  62     int ret = 0;
  63
  64     ret = vmcs_write(field, val);
  65
  66     if (ret != VMX_SUCCESS) {
  67         PrintError("VMWRITE error on %s!: %d\n", v3_vmcs_field_to_str(field), ret);
  68         return 1;
  69     }
  70
  71     return 0;
  72 }
  73
  74 static int inline check_vmcs_read(vmcs_field_t field, void * val) {
  75     int ret = 0;
  76
  77     ret = vmcs_read(field, val);
  78
  79     if (ret != VMX_SUCCESS) {
  80         PrintError("VMREAD error on %s!: %d\n", v3_vmcs_field_to_str(field), ret);
  81     }
  82
  83     return ret;
  84 }
  85
  86
  87
  88
  89 static addr_t allocate_vmcs() {
  90     struct vmcs_data * vmcs_page = NULL;
  91
  92     PrintDebug("Allocating page\n");
  93
  94     vmcs_page = (struct vmcs_data *)V3_VAddr(V3_AllocPages(1));
  95     memset(vmcs_page, 0, 4096);
  96
  97     vmcs_page->revision = hw_info.basic_info.revision;
  98     PrintDebug("VMX Revision: 0x%x\n", vmcs_page->revision);
  99
 100     return (addr_t)V3_PAddr((void *)vmcs_page);
 101 }
 102
 103
 104
 105
 106 static int init_vmcs_bios(struct guest_info * core, struct vmx_data * vmx_state) {
 107     int vmx_ret = 0;
 108
 109     // disable global interrupts for vm state initialization
 110     v3_disable_ints();
 111
 112     PrintDebug("Loading VMCS\n");
 113     vmx_ret = vmcs_load(vmx_state->vmcs_ptr_phys);
 114     vmx_state->state = VMX_UNLAUNCHED;
 115
 116     if (vmx_ret != VMX_SUCCESS) {
 117         PrintError("VMPTRLD failed\n");
 118         return -1;
 119     }
 120
 121
 122     /*** Setup default state from HW ***/
 123
 124     vmx_state->pin_ctrls.value = hw_info.pin_ctrls.def_val;
 125     vmx_state->pri_proc_ctrls.value = hw_info.proc_ctrls.def_val;
 126     vmx_state->exit_ctrls.value = hw_info.exit_ctrls.def_val;
 127     vmx_state->entry_ctrls.value = hw_info.entry_ctrls.def_val;
 128     vmx_state->sec_proc_ctrls.value = hw_info.sec_proc_ctrls.def_val;
 129
 130     /* Print Control MSRs */
 131     PrintDebug("CR0 MSR: %p\n", (void *)(addr_t)hw_info.cr0.value);
 132     PrintDebug("CR4 MSR: %p\n", (void *)(addr_t)hw_info.cr4.value);
 133
 134
 135
 136     /******* Setup Host State **********/
 137
 138     /* Cache GDTR, IDTR, and TR in host struct */
 139
 140
 141     /********** Setup VMX Control Fields ***********/
 142
 143     /* Add external interrupts, NMI exiting, and virtual NMI */
 144     vmx_state->pin_ctrls.nmi_exit = 1;
 145     vmx_state->pin_ctrls.ext_int_exit = 1;
 146
 147
 148     vmx_state->pri_proc_ctrls.hlt_exit = 1;
 149
 150
 151     vmx_state->pri_proc_ctrls.pause_exit = 0;
 152     vmx_state->pri_proc_ctrls.tsc_offset = 1;
 153 #ifdef V3_CONFIG_TIME_VIRTUALIZE_TSC
 154     vmx_state->pri_proc_ctrls.rdtsc_exit = 1;
 155 #endif
 156
 157     /* Setup IO map */
 158     vmx_state->pri_proc_ctrls.use_io_bitmap = 1;
 159     vmx_ret |= check_vmcs_write(VMCS_IO_BITMAP_A_ADDR, (addr_t)V3_PAddr(core->vm_info->io_map.arch_data));
 160     vmx_ret |= check_vmcs_write(VMCS_IO_BITMAP_B_ADDR,
 161             (addr_t)V3_PAddr(core->vm_info->io_map.arch_data) + PAGE_SIZE_4KB);
 162
 163
 164     vmx_state->pri_proc_ctrls.use_msr_bitmap = 1;
 165     vmx_ret |= check_vmcs_write(VMCS_MSR_BITMAP, (addr_t)V3_PAddr(core->vm_info->msr_map.arch_data));
 166
 167
 168
 169 #ifdef __V3_64BIT__
 170     // Ensure host runs in 64-bit mode at each VM EXIT
 171     vmx_state->exit_ctrls.host_64_on = 1;
 172 #endif
 173
 174     // Hook all accesses to EFER register
 175     v3_hook_msr(core->vm_info, EFER_MSR,
 176                 &v3_handle_efer_read,
 177                 &v3_handle_efer_write,
 178                 core);
 179
 180     // Restore host's EFER register on each VM EXIT
 181     vmx_state->exit_ctrls.ld_efer = 1;
 182
 183     // Save/restore guest's EFER register to/from VMCS on VM EXIT/ENTRY
 184     vmx_state->exit_ctrls.save_efer = 1;
 185     vmx_state->entry_ctrls.ld_efer  = 1;
 186
 187     // Cause VM_EXIT whenever CR4.VMXE or CR4.PAE bits are written
 188     vmx_ret |= check_vmcs_write(VMCS_CR4_MASK, CR4_VMXE | CR4_PAE);
 189
 190
 191     /* Setup paging */
 192     if (core->shdw_pg_mode == SHADOW_PAGING) {
 193         PrintDebug("Creating initial shadow page table\n");
 194
 195         if (v3_init_passthrough_pts(core) == -1) {
 196             PrintError("Could not initialize passthrough page tables\n");
 197             return -1;
 198         }
 199
 200 #define CR0_PE 0x00000001
 201 #define CR0_PG 0x80000000
 202 #define CR0_WP 0x00010000 // To ensure mem hooks work
 203         vmx_ret |= check_vmcs_write(VMCS_CR0_MASK, (CR0_PE | CR0_PG | CR0_WP));
 204
 205         core->ctrl_regs.cr3 = core->direct_map_pt;
 206
 207         // vmx_state->pinbased_ctrls |= NMI_EXIT;
 208
 209         /* Add CR exits */
 210         vmx_state->pri_proc_ctrls.cr3_ld_exit = 1;
 211         vmx_state->pri_proc_ctrls.cr3_str_exit = 1;
 212
 213         vmx_state->pri_proc_ctrls.invlpg_exit = 1;
 214
 215         /* Add page fault exits */
 216         vmx_state->excp_bmap.pf = 1;
 217
 218         // Setup VMX Assist
 219         v3_vmxassist_init(core, vmx_state);
 220
 221     } else if ((core->shdw_pg_mode == NESTED_PAGING) &&
 222                (v3_cpu_types[core->pcpu_id] == V3_VMX_EPT_CPU)) {
 223
 224 #define CR0_PE 0x00000001
 225 #define CR0_PG 0x80000000
 226 #define CR0_WP 0x00010000 // To ensure mem hooks work
 227         vmx_ret |= check_vmcs_write(VMCS_CR0_MASK, (CR0_PE | CR0_PG | CR0_WP));
 228
 229         // vmx_state->pinbased_ctrls |= NMI_EXIT;
 230
 231         /* Disable CR exits */
 232         vmx_state->pri_proc_ctrls.cr3_ld_exit = 0;
 233         vmx_state->pri_proc_ctrls.cr3_str_exit = 0;
 234
 235         vmx_state->pri_proc_ctrls.invlpg_exit = 0;
 236
 237         /* Add page fault exits */
 238         //      vmx_state->excp_bmap.pf = 1; // This should never happen..., enabled to catch bugs
 239
 240         // Setup VMX Assist
 241         v3_vmxassist_init(core, vmx_state);
 242
 243         /* Enable EPT */
 244         vmx_state->pri_proc_ctrls.sec_ctrls = 1; // Enable secondary proc controls
 245         vmx_state->sec_proc_ctrls.enable_ept = 1; // enable EPT paging
 246
 247
 248
 249         if (v3_init_ept(core, &hw_info) == -1) {
 250             PrintError("Error initializing EPT\n");
 251             return -1;
 252         }
 253
 254     } else if ((core->shdw_pg_mode == NESTED_PAGING) &&
 255                (v3_cpu_types[core->pcpu_id] == V3_VMX_EPT_UG_CPU)) {
 256         int i = 0;
 257         // For now we will assume that unrestricted guest mode is assured w/ EPT
 258
 259
 260         core->vm_regs.rsp = 0x00;
 261         core->rip = 0xfff0;
 262         core->vm_regs.rdx = 0x00000f00;
 263         core->ctrl_regs.rflags = 0x00000002; // The reserved bit is always 1
 264         core->ctrl_regs.cr0 = 0x60010010; // Set the WP flag so the memory hooks work in real-mode
 265
 266
 267         core->segments.cs.selector = 0xf000;
 268         core->segments.cs.limit = 0xffff;
 269         core->segments.cs.base = 0x0000000f0000LL;
 270
 271         // (raw attributes = 0xf3)
 272         core->segments.cs.type = 0xb;
 273         core->segments.cs.system = 0x1;
 274         core->segments.cs.dpl = 0x0;
 275         core->segments.cs.present = 1;
 276
 277
 278
 279         struct v3_segment * segregs [] = {&(core->segments.ss), &(core->segments.ds),
 280                                           &(core->segments.es), &(core->segments.fs),
 281                                           &(core->segments.gs), NULL};
 282
 283         for ( i = 0; segregs[i] != NULL; i++) {
 284             struct v3_segment * seg = segregs[i];
 285
 286             seg->selector = 0x0000;
 287             //    seg->base = seg->selector << 4;
 288             seg->base = 0x00000000;
 289             seg->limit = 0xffff;
 290
 291
 292             seg->type = 0x3;
 293             seg->system = 0x1;
 294             seg->dpl = 0x0;
 295             seg->present = 1;
 296             //    seg->granularity = 1;
 297
 298         }
 299
 300
 301         core->segments.gdtr.limit = 0x0000ffff;
 302         core->segments.gdtr.base = 0x0000000000000000LL;
 303
 304         core->segments.idtr.limit = 0x0000ffff;
 305         core->segments.idtr.base = 0x0000000000000000LL;
 306
 307         core->segments.ldtr.selector = 0x0000;
 308         core->segments.ldtr.limit = 0x0000ffff;
 309         core->segments.ldtr.base = 0x0000000000000000LL;
 310         core->segments.ldtr.type = 2;
 311         core->segments.ldtr.present = 1;
 312
 313         core->segments.tr.selector = 0x0000;
 314         core->segments.tr.limit = 0x0000ffff;
 315         core->segments.tr.base = 0x0000000000000000LL;
 316         core->segments.tr.type = 0xb;
 317         core->segments.tr.present = 1;
 318
 319         //      core->dbg_regs.dr6 = 0x00000000ffff0ff0LL;
 320         core->dbg_regs.dr7 = 0x0000000000000400LL;
 321
 322         /* Enable EPT */
 323         vmx_state->pri_proc_ctrls.sec_ctrls = 1; // Enable secondary proc controls
 324         vmx_state->sec_proc_ctrls.enable_ept = 1; // enable EPT paging
 325         vmx_state->sec_proc_ctrls.unrstrct_guest = 1; // enable unrestricted guest operation
 326
 327
 328         /* Disable shadow paging stuff */
 329         vmx_state->pri_proc_ctrls.cr3_ld_exit = 0;
 330         vmx_state->pri_proc_ctrls.cr3_str_exit = 0;
 331
 332         vmx_state->pri_proc_ctrls.invlpg_exit = 0;
 333
 334
 335         if (v3_init_ept(core, &hw_info) == -1) {
 336             PrintError("Error initializing EPT\n");
 337             return -1;
 338         }
 339
 340     } else {
 341         PrintError("Invalid Virtual paging mode\n");
 342         return -1;
 343     }
 344
 345
 346     // hook vmx msrs
 347
 348     // Setup SYSCALL/SYSENTER MSRs in load/store area
 349
 350     // save STAR, LSTAR, FMASK, KERNEL_GS_BASE MSRs in MSR load/store area
 351     {
 352 #define IA32_STAR 0xc0000081
 353 #define IA32_LSTAR 0xc0000082
 354 #define IA32_FMASK 0xc0000084
 355 #define IA32_KERN_GS_BASE 0xc0000102
 356
 357 #define IA32_CSTAR 0xc0000083 // Compatibility mode STAR (ignored for now... hopefully its not that important...)
 358
 359         int msr_ret = 0;
 360
 361         struct vmcs_msr_entry * exit_store_msrs = NULL;
 362         struct vmcs_msr_entry * exit_load_msrs = NULL;
 363         struct vmcs_msr_entry * entry_load_msrs = NULL;;
 364         int max_msrs = (hw_info.misc_info.max_msr_cache_size + 1) * 4;
 365
 366         V3_Print("Setting up MSR load/store areas (max_msr_count=%d)\n", max_msrs);
 367
 368         if (max_msrs < 4) {
 369             PrintError("Max MSR cache size is too small (%d)\n", max_msrs);
 370             return -1;
 371         }
 372
 373         vmx_state->msr_area = V3_VAddr(V3_AllocPages(1));
 374
 375         if (vmx_state->msr_area == NULL) {
 376             PrintError("could not allocate msr load/store area\n");
 377             return -1;
 378         }
 379
 380         msr_ret |= check_vmcs_write(VMCS_EXIT_MSR_STORE_CNT, 4);
 381         msr_ret |= check_vmcs_write(VMCS_EXIT_MSR_LOAD_CNT, 4);
 382         msr_ret |= check_vmcs_write(VMCS_ENTRY_MSR_LOAD_CNT, 4);
 383
 384
 385         exit_store_msrs = (struct vmcs_msr_entry *)(vmx_state->msr_area);
 386         exit_load_msrs = (struct vmcs_msr_entry *)(vmx_state->msr_area + (sizeof(struct vmcs_msr_entry) * 4));
 387         entry_load_msrs = (struct vmcs_msr_entry *)(vmx_state->msr_area + (sizeof(struct vmcs_msr_entry) * 8));
 388
 389
 390         exit_store_msrs[0].index = IA32_STAR;
 391         exit_store_msrs[1].index = IA32_LSTAR;
 392         exit_store_msrs[2].index = IA32_FMASK;
 393         exit_store_msrs[3].index = IA32_KERN_GS_BASE;
 394
 395         memcpy(exit_store_msrs, exit_load_msrs, sizeof(struct vmcs_msr_entry) * 4);
 396         memcpy(exit_store_msrs, entry_load_msrs, sizeof(struct vmcs_msr_entry) * 4);
 397
 398
 399         v3_get_msr(IA32_STAR, &(exit_load_msrs[0].hi), &(exit_load_msrs[0].lo));
 400         v3_get_msr(IA32_LSTAR, &(exit_load_msrs[1].hi), &(exit_load_msrs[1].lo));
 401         v3_get_msr(IA32_FMASK, &(exit_load_msrs[2].hi), &(exit_load_msrs[2].lo));
 402         v3_get_msr(IA32_KERN_GS_BASE, &(exit_load_msrs[3].hi), &(exit_load_msrs[3].lo));
 403
 404         msr_ret |= check_vmcs_write(VMCS_EXIT_MSR_STORE_ADDR, (addr_t)V3_PAddr(exit_store_msrs));
 405         msr_ret |= check_vmcs_write(VMCS_EXIT_MSR_LOAD_ADDR, (addr_t)V3_PAddr(exit_load_msrs));
 406         msr_ret |= check_vmcs_write(VMCS_ENTRY_MSR_LOAD_ADDR, (addr_t)V3_PAddr(entry_load_msrs));
 407
 408     }
 409
 410     /* Sanity check ctrl/reg fields against hw_defaults */
 411
 412
 413
 414
 415     /*** Write all the info to the VMCS ***/
 416
 417     /*
 418     {
 419         // IS THIS NECESSARY???
 420 #define DEBUGCTL_MSR 0x1d9
 421         struct v3_msr tmp_msr;
 422         v3_get_msr(DEBUGCTL_MSR, &(tmp_msr.hi), &(tmp_msr.lo));
 423         vmx_ret |= check_vmcs_write(VMCS_GUEST_DBG_CTL, tmp_msr.value);
 424         core->dbg_regs.dr7 = 0x400;
 425     }
 426     */
 427
 428 #ifdef __V3_64BIT__
 429     vmx_ret |= check_vmcs_write(VMCS_LINK_PTR, (addr_t)0xffffffffffffffffULL);
 430 #else
 431     vmx_ret |= check_vmcs_write(VMCS_LINK_PTR, (addr_t)0xffffffffUL);
 432     vmx_ret |= check_vmcs_write(VMCS_LINK_PTR_HIGH, (addr_t)0xffffffffUL);
 433 #endif
 434
 435
 436
 437
 438     if (v3_update_vmcs_ctrl_fields(core)) {
 439         PrintError("Could not write control fields!\n");
 440         return -1;
 441     }
 442
 443     if (v3_update_vmcs_host_state(core)) {
 444         PrintError("Could not write host state\n");
 445         return -1;
 446     }
 447
 448     // reenable global interrupts for vm state initialization now
 449     // that the vm state is initialized. If another VM kicks us off,
 450     // it'll update our vmx state so that we know to reload ourself
 451     v3_enable_ints();
 452
 453     return 0;
 454 }
 455
 456 int v3_init_vmx_vmcs(struct guest_info * core, v3_vm_class_t vm_class) {
 457     struct vmx_data * vmx_state = NULL;
 458     int vmx_ret = 0;
 459
 460     vmx_state = (struct vmx_data *)V3_Malloc(sizeof(struct vmx_data));
 461     memset(vmx_state, 0, sizeof(struct vmx_data));
 462
 463     PrintDebug("vmx_data pointer: %p\n", (void *)vmx_state);
 464
 465     PrintDebug("Allocating VMCS\n");
 466     vmx_state->vmcs_ptr_phys = allocate_vmcs();
 467
 468     PrintDebug("VMCS pointer: %p\n", (void *)(vmx_state->vmcs_ptr_phys));
 469
 470     core->vmm_data = vmx_state;
 471     vmx_state->state = VMX_UNLAUNCHED;
 472
 473     PrintDebug("Initializing VMCS (addr=%p)\n", core->vmm_data);
 474
 475     // TODO: Fix vmcs fields so they're 32-bit
 476
 477     PrintDebug("Clearing VMCS: %p\n", (void *)vmx_state->vmcs_ptr_phys);
 478     vmx_ret = vmcs_clear(vmx_state->vmcs_ptr_phys);
 479
 480     if (vmx_ret != VMX_SUCCESS) {
 481         PrintError("VMCLEAR failed\n");
 482         return -1;
 483     }
 484
 485     if (vm_class == V3_PC_VM) {
 486         PrintDebug("Initializing VMCS\n");
 487         if (init_vmcs_bios(core, vmx_state) == -1) {
 488             PrintError("Error initializing VMCS to BIOS state\n");
 489             return -1;
 490         }
 491     } else {
 492         PrintError("Invalid VM Class\n");
 493         return -1;
 494     }
 495
 496     PrintDebug("Serializing VMCS: %p\n", (void *)vmx_state->vmcs_ptr_phys);
 497     vmx_ret = vmcs_clear(vmx_state->vmcs_ptr_phys);
 498
 499     return 0;
 500 }
 501
 502
 503 int v3_deinit_vmx_vmcs(struct guest_info * core) {
 504     struct vmx_data * vmx_state = core->vmm_data;
 505
 506     V3_FreePages((void *)(vmx_state->vmcs_ptr_phys), 1);
 507     V3_FreePages(V3_PAddr(vmx_state->msr_area), 1);
 508
 509     V3_Free(vmx_state);
 510
 511     return 0;
 512 }
 513
 514
 515
 516 #ifdef V3_CONFIG_CHECKPOINT
 517 /*
 518  * JRL: This is broken
 519  */
 520 int v3_vmx_save_core(struct guest_info * core, void * ctx){
 521     uint64_t vmcs_ptr = vmcs_store();
 522
 523     v3_chkpt_save(ctx, "vmcs_data", PAGE_SIZE, (void *)vmcs_ptr);
 524
 525     return 0;
 526 }
 527
 528 int v3_vmx_load_core(struct guest_info * core, void * ctx){
 529     struct vmx_data * vmx_info = (struct vmx_data *)(core->vmm_data);
 530     struct cr0_32 * shadow_cr0;
 531     char vmcs[PAGE_SIZE_4KB];
 532
 533     v3_chkpt_load(ctx, "vmcs_data", PAGE_SIZE_4KB, vmcs);
 534
 535     vmcs_clear(vmx_info->vmcs_ptr_phys);
 536     vmcs_load((addr_t)vmcs);
 537
 538     v3_vmx_save_vmcs(core);
 539
 540     shadow_cr0 = (struct cr0_32 *)&(core->ctrl_regs.cr0);
 541
 542
 543     /* Get the CPU mode to set the guest_ia32e entry ctrl */
 544
 545     if (core->shdw_pg_mode == SHADOW_PAGING) {
 546         if (v3_get_vm_mem_mode(core) == VIRTUAL_MEM) {
 547             if (v3_activate_shadow_pt(core) == -1) {
 548                 PrintError("Failed to activate shadow page tables\n");
 549                 return -1;
 550             }
 551         } else {
 552             if (v3_activate_passthrough_pt(core) == -1) {
 553                 PrintError("Failed to activate passthrough page tables\n");
 554                 return -1;
 555             }
 556         }
 557     }
 558
 559     return 0;
 560 }
 561 #endif
 562
 563
 564 void v3_flush_vmx_vm_core(struct guest_info * core) {
 565     struct vmx_data * vmx_info = (struct vmx_data *)(core->vmm_data);
 566     vmcs_clear(vmx_info->vmcs_ptr_phys);
 567     vmx_info->state = VMX_UNLAUNCHED;
 568 }
 569
 570
 571
 572 static int update_irq_exit_state(struct guest_info * info) {
 573     struct vmx_exit_idt_vec_info idt_vec_info;
 574
 575     check_vmcs_read(VMCS_IDT_VECTOR_INFO, &(idt_vec_info.value));
 576
 577     if ((info->intr_core_state.irq_started == 1) && (idt_vec_info.valid == 0)) {
 578 #ifdef V3_CONFIG_DEBUG_INTERRUPTS
 579         V3_Print("Calling v3_injecting_intr\n");
 580 #endif
 581         info->intr_core_state.irq_started = 0;
 582         v3_injecting_intr(info, info->intr_core_state.irq_vector, V3_EXTERNAL_IRQ);
 583     }
 584
 585     return 0;
 586 }
 587
 588 static int update_irq_entry_state(struct guest_info * info) {
 589     struct vmx_exit_idt_vec_info idt_vec_info;
 590     struct vmcs_interrupt_state intr_core_state;
 591     struct vmx_data * vmx_info = (struct vmx_data *)(info->vmm_data);
 592
 593     check_vmcs_read(VMCS_IDT_VECTOR_INFO, &(idt_vec_info.value));
 594     check_vmcs_read(VMCS_GUEST_INT_STATE, &(intr_core_state));
 595
 596     /* Check for pending exceptions to inject */
 597     if (v3_excp_pending(info)) {
 598         struct vmx_entry_int_info int_info;
 599         int_info.value = 0;
 600
 601         // In VMX, almost every exception is hardware
 602         // Software exceptions are pretty much only for breakpoint or overflow
 603         int_info.type = 3;
 604         int_info.vector = v3_get_excp_number(info);
 605
 606         if (info->excp_state.excp_error_code_valid) {
 607             check_vmcs_write(VMCS_ENTRY_EXCP_ERR, info->excp_state.excp_error_code);
 608             int_info.error_code = 1;
 609
 610 #ifdef V3_CONFIG_DEBUG_INTERRUPTS
 611             V3_Print("Injecting exception %d with error code %x\n",
 612                     int_info.vector, info->excp_state.excp_error_code);
 613 #endif
 614         }
 615
 616         int_info.valid = 1;
 617 #ifdef V3_CONFIG_DEBUG_INTERRUPTS
 618         V3_Print("Injecting exception %d (EIP=%p)\n", int_info.vector, (void *)(addr_t)info->rip);
 619 #endif
 620         check_vmcs_write(VMCS_ENTRY_INT_INFO, int_info.value);
 621
 622         v3_injecting_excp(info, int_info.vector);
 623
 624     } else if ((((struct rflags *)&(info->ctrl_regs.rflags))->intr == 1) &&
 625                (intr_core_state.val == 0)) {
 626
 627         if ((info->intr_core_state.irq_started == 1) && (idt_vec_info.valid == 1)) {
 628
 629 #ifdef V3_CONFIG_DEBUG_INTERRUPTS
 630             V3_Print("IRQ pending from previous injection\n");
 631 #endif
 632
 633             // Copy the IDT vectoring info over to reinject the old interrupt
 634             if (idt_vec_info.error_code == 1) {
 635                 uint32_t err_code = 0;
 636
 637                 check_vmcs_read(VMCS_IDT_VECTOR_ERR, &err_code);
 638                 check_vmcs_write(VMCS_ENTRY_EXCP_ERR, err_code);
 639             }
 640
 641             idt_vec_info.undef = 0;
 642             check_vmcs_write(VMCS_ENTRY_INT_INFO, idt_vec_info.value);
 643
 644         } else {
 645             struct vmx_entry_int_info ent_int;
 646             ent_int.value = 0;
 647
 648             switch (v3_intr_pending(info)) {
 649                 case V3_EXTERNAL_IRQ: {
 650                     info->intr_core_state.irq_vector = v3_get_intr(info);
 651                     ent_int.vector = info->intr_core_state.irq_vector;
 652                     ent_int.type = 0;
 653                     ent_int.error_code = 0;
 654                     ent_int.valid = 1;
 655
 656 #ifdef V3_CONFIG_DEBUG_INTERRUPTS
 657                     V3_Print("Injecting Interrupt %d at exit %u(EIP=%p)\n",
 658                                info->intr_core_state.irq_vector,
 659                                (uint32_t)info->num_exits,
 660                                (void *)(addr_t)info->rip);
 661 #endif
 662
 663                     check_vmcs_write(VMCS_ENTRY_INT_INFO, ent_int.value);
 664                     info->intr_core_state.irq_started = 1;
 665
 666                     break;
 667                 }
 668                 case V3_NMI:
 669                     PrintDebug("Injecting NMI\n");
 670
 671                     ent_int.type = 2;
 672                     ent_int.vector = 2;
 673                     ent_int.valid = 1;
 674                     check_vmcs_write(VMCS_ENTRY_INT_INFO, ent_int.value);
 675
 676                     break;
 677                 case V3_SOFTWARE_INTR:
 678                     PrintDebug("Injecting software interrupt\n");
 679                     ent_int.type = 4;
 680
 681                     ent_int.valid = 1;
 682                     check_vmcs_write(VMCS_ENTRY_INT_INFO, ent_int.value);
 683
 684                     break;
 685                 case V3_VIRTUAL_IRQ:
 686                     // Not sure what to do here, Intel doesn't have virtual IRQs
 687                     // May be the same as external interrupts/IRQs
 688
 689                     break;
 690                 case V3_INVALID_INTR:
 691                 default:
 692                     break;
 693             }
 694         }
 695     } else if ((v3_intr_pending(info)) && (vmx_info->pri_proc_ctrls.int_wndw_exit == 0)) {
 696         // Enable INTR window exiting so we know when IF=1
 697         uint32_t instr_len;
 698
 699         check_vmcs_read(VMCS_EXIT_INSTR_LEN, &instr_len);
 700
 701 #ifdef V3_CONFIG_DEBUG_INTERRUPTS
 702         V3_Print("Enabling Interrupt-Window exiting: %d\n", instr_len);
 703 #endif
 704
 705         vmx_info->pri_proc_ctrls.int_wndw_exit = 1;
 706         check_vmcs_write(VMCS_PROC_CTRLS, vmx_info->pri_proc_ctrls.value);
 707     }
 708
 709
 710     return 0;
 711 }
 712
 713
 714
 715 static struct vmx_exit_info exit_log[10];
 716
 717 static void print_exit_log(struct guest_info * info) {
 718     int cnt = info->num_exits % 10;
 719     int i = 0;
 720
 721
 722     V3_Print("\nExit Log (%d total exits):\n", (uint32_t)info->num_exits);
 723
 724     for (i = 0; i < 10; i++) {
 725         struct vmx_exit_info * tmp = &exit_log[cnt];
 726
 727         V3_Print("%d:\texit_reason = %p\n", i, (void *)(addr_t)tmp->exit_reason);
 728         V3_Print("\texit_qual = %p\n", (void *)tmp->exit_qual);
 729         V3_Print("\tint_info = %p\n", (void *)(addr_t)tmp->int_info);
 730         V3_Print("\tint_err = %p\n", (void *)(addr_t)tmp->int_err);
 731         V3_Print("\tinstr_info = %p\n", (void *)(addr_t)tmp->instr_info);
 732
 733         cnt--;
 734
 735         if (cnt == -1) {
 736             cnt = 9;
 737         }
 738
 739     }
 740
 741 }
 742
 743 /*
 744  * CAUTION and DANGER!!!
 745  *
 746  * The VMCS CANNOT(!!) be accessed outside of the cli/sti calls inside this function
 747  * When exectuing a symbiotic call, the VMCS WILL be overwritten, so any dependencies
 748  * on its contents will cause things to break. The contents at the time of the exit WILL
 749  * change before the exit handler is executed.
 750  */
 751 int v3_vmx_enter(struct guest_info * info) {
 752     int ret = 0;
 753     uint32_t tsc_offset_low, tsc_offset_high;
 754     struct vmx_exit_info exit_info;
 755     struct vmx_data * vmx_info = (struct vmx_data *)(info->vmm_data);
 756
 757     // Conditionally yield the CPU if the timeslice has expired
 758     v3_yield_cond(info);
 759
 760     // Perform any additional yielding needed for time adjustment
 761     v3_adjust_time(info);
 762
 763     // disable global interrupts for vm state transition
 764     v3_disable_ints();
 765
 766     // Update timer devices late after being in the VM so that as much
 767     // of hte time in the VM is accounted for as possible. Also do it before
 768     // updating IRQ entry state so that any interrupts the timers raise get
 769     // handled on the next VM entry. Must be done with interrupts disabled.
 770     v3_update_timers(info);
 771
 772     if (vmcs_store() != vmx_info->vmcs_ptr_phys) {
 773         vmcs_clear(vmx_info->vmcs_ptr_phys);
 774         vmcs_load(vmx_info->vmcs_ptr_phys);
 775         vmx_info->state = VMX_UNLAUNCHED;
 776     }
 777
 778     v3_vmx_restore_vmcs(info);
 779
 780
 781 #ifdef V3_CONFIG_SYMCALL
 782     if (info->sym_core_state.symcall_state.sym_call_active == 0) {
 783         update_irq_entry_state(info);
 784     }
 785 #else
 786     update_irq_entry_state(info);
 787 #endif
 788
 789     {
 790         addr_t guest_cr3;
 791         vmcs_read(VMCS_GUEST_CR3, &guest_cr3);
 792         vmcs_write(VMCS_GUEST_CR3, guest_cr3);
 793     }
 794
 795     // Perform last-minute time bookkeeping prior to entering the VM
 796     v3_time_enter_vm(info);
 797
 798     tsc_offset_high = (uint32_t)((v3_tsc_host_offset(&info->time_state) >> 32) & 0xffffffff);
 799     tsc_offset_low = (uint32_t)(v3_tsc_host_offset(&info->time_state) & 0xffffffff);
 800     check_vmcs_write(VMCS_TSC_OFFSET_HIGH, tsc_offset_high);
 801     check_vmcs_write(VMCS_TSC_OFFSET, tsc_offset_low);
 802
 803
 804     if (v3_update_vmcs_host_state(info)) {
 805         v3_enable_ints();
 806         PrintError("Could not write host state\n");
 807         return -1;
 808     }
 809
 810
 811     if (vmx_info->state == VMX_UNLAUNCHED) {
 812         vmx_info->state = VMX_LAUNCHED;
 813         ret = v3_vmx_launch(&(info->vm_regs), info, &(info->ctrl_regs));
 814     } else {
 815         V3_ASSERT(vmx_info->state != VMX_UNLAUNCHED);
 816         ret = v3_vmx_resume(&(info->vm_regs), info, &(info->ctrl_regs));
 817     }
 818
 819
 820
 821     //  PrintDebug("VMX Exit: ret=%d\n", ret);
 822
 823     if (ret != VMX_SUCCESS) {
 824         uint32_t error = 0;
 825         vmcs_read(VMCS_INSTR_ERR, &error);
 826
 827         v3_enable_ints();
 828
 829         PrintError("VMENTRY Error: %d (launch_ret = %d)\n", error, ret);
 830         return -1;
 831     }
 832
 833
 834
 835     // Immediate exit from VM time bookkeeping
 836     v3_time_exit_vm(info);
 837
 838     info->num_exits++;
 839
 840     /* Update guest state */
 841     v3_vmx_save_vmcs(info);
 842
 843     // info->cpl = info->segments.cs.selector & 0x3;
 844
 845     info->mem_mode = v3_get_vm_mem_mode(info);
 846     info->cpu_mode = v3_get_vm_cpu_mode(info);
 847
 848
 849     check_vmcs_read(VMCS_EXIT_INSTR_LEN, &(exit_info.instr_len));
 850     check_vmcs_read(VMCS_EXIT_INSTR_INFO, &(exit_info.instr_info));
 851     check_vmcs_read(VMCS_EXIT_REASON, &(exit_info.exit_reason));
 852     check_vmcs_read(VMCS_EXIT_QUAL, &(exit_info.exit_qual));
 853     check_vmcs_read(VMCS_EXIT_INT_INFO, &(exit_info.int_info));
 854     check_vmcs_read(VMCS_EXIT_INT_ERR, &(exit_info.int_err));
 855     check_vmcs_read(VMCS_GUEST_LINEAR_ADDR, &(exit_info.guest_linear_addr));
 856
 857     if (info->shdw_pg_mode == NESTED_PAGING) {
 858         check_vmcs_read(VMCS_GUEST_PHYS_ADDR, &(exit_info.ept_fault_addr));
 859     }
 860
 861     //PrintDebug("VMX Exit taken, id-qual: %u-%lu\n", exit_info.exit_reason, exit_info.exit_qual);
 862
 863     exit_log[info->num_exits % 10] = exit_info;
 864
 865 #ifdef V3_CONFIG_SYMCALL
 866     if (info->sym_core_state.symcall_state.sym_call_active == 0) {
 867         update_irq_exit_state(info);
 868     }
 869 #else
 870     update_irq_exit_state(info);
 871 #endif
 872
 873     if (exit_info.exit_reason == VMEXIT_INTR_WINDOW) {
 874         // This is a special case whose only job is to inject an interrupt
 875         vmcs_read(VMCS_PROC_CTRLS, &(vmx_info->pri_proc_ctrls.value));
 876         vmx_info->pri_proc_ctrls.int_wndw_exit = 0;
 877         vmcs_write(VMCS_PROC_CTRLS, vmx_info->pri_proc_ctrls.value);
 878
 879 #ifdef V3_CONFIG_DEBUG_INTERRUPTS
 880        V3_Print("Interrupts available again! (RIP=%llx)\n", info->rip);
 881 #endif
 882     }
 883
 884     // reenable global interrupts after vm exit
 885     v3_enable_ints();
 886
 887     // Conditionally yield the CPU if the timeslice has expired
 888     v3_yield_cond(info);
 889
 890     if (v3_handle_vmx_exit(info, &exit_info) == -1) {
 891         PrintError("Error in VMX exit handler (Exit reason=%x)\n", exit_info.exit_reason);
 892         return -1;
 893     }
 894
 895     return 0;
 896 }
 897
 898
 899 int v3_start_vmx_guest(struct guest_info * info) {
 900
 901     PrintDebug("Starting VMX core %u\n", info->vcpu_id);
 902
 903     if (info->vcpu_id == 0) {
 904         info->core_run_state = CORE_RUNNING;
 905     } else {
 906
 907         PrintDebug("VMX core %u: Waiting for core initialization\n", info->vcpu_id);
 908
 909         while (info->core_run_state == CORE_STOPPED) {
 910
 911             if (info->vm_info->run_state == VM_STOPPED) {
 912                 // The VM was stopped before this core was initialized.
 913                 return 0;
 914             }
 915
 916             v3_yield(info);
 917             //PrintDebug("VMX core %u: still waiting for INIT\n",info->vcpu_id);
 918         }
 919
 920         PrintDebug("VMX core %u initialized\n", info->vcpu_id);
 921
 922         // We'll be paranoid about race conditions here
 923         v3_wait_at_barrier(info);
 924     }
 925
 926
 927     PrintDebug("VMX core %u: I am starting at CS=0x%x (base=0x%p, limit=0x%x),  RIP=0x%p\n",
 928                info->vcpu_id, info->segments.cs.selector, (void *)(info->segments.cs.base),
 929                info->segments.cs.limit, (void *)(info->rip));
 930
 931
 932     PrintDebug("VMX core %u: Launching VMX VM on logical core %u\n", info->vcpu_id, info->pcpu_id);
 933
 934     v3_start_time(info);
 935
 936     while (1) {
 937
 938         if (info->vm_info->run_state == VM_STOPPED) {
 939             info->core_run_state = CORE_STOPPED;
 940             break;
 941         }
 942
 943         if (v3_vmx_enter(info) == -1) {
 944
 945             addr_t host_addr;
 946             addr_t linear_addr = 0;
 947
 948             info->vm_info->run_state = VM_ERROR;
 949
 950             V3_Print("VMX core %u: VMX ERROR!!\n", info->vcpu_id);
 951
 952             v3_print_guest_state(info);
 953
 954             V3_Print("VMX core %u\n", info->vcpu_id);
 955
 956             linear_addr = get_addr_linear(info, info->rip, &(info->segments.cs));
 957
 958             if (info->mem_mode == PHYSICAL_MEM) {
 959                 v3_gpa_to_hva(info, linear_addr, &host_addr);
 960             } else if (info->mem_mode == VIRTUAL_MEM) {
 961                 v3_gva_to_hva(info, linear_addr, &host_addr);
 962             }
 963
 964             V3_Print("VMX core %u: Host Address of rip = 0x%p\n", info->vcpu_id, (void *)host_addr);
 965
 966             V3_Print("VMX core %u: Instr (15 bytes) at %p:\n", info->vcpu_id, (void *)host_addr);
 967             v3_dump_mem((uint8_t *)host_addr, 15);
 968
 969             v3_print_stack(info);
 970
 971
 972             v3_print_vmcs();
 973             print_exit_log(info);
 974             return -1;
 975         }
 976
 977         v3_wait_at_barrier(info);
 978
 979
 980         if (info->vm_info->run_state == VM_STOPPED) {
 981             info->core_run_state = CORE_STOPPED;
 982             break;
 983         }
 984 /*
 985         if ((info->num_exits % 5000) == 0) {
 986             V3_Print("VMX Exit number %d\n", (uint32_t)info->num_exits);
 987         }
 988 */
 989
 990     }
 991
 992     return 0;
 993 }
 994
 995
 996
 997
 998 #define VMX_FEATURE_CONTROL_MSR     0x0000003a
 999 #define CPUID_VMX_FEATURES 0x00000005  /* LOCK and VMXON */
1000 #define CPUID_1_ECX_VTXFLAG 0x00000020
1001
1002 int v3_is_vmx_capable() {
1003     v3_msr_t feature_msr;
1004     uint32_t eax = 0, ebx = 0, ecx = 0, edx = 0;
1005
1006     v3_cpuid(0x1, &eax, &ebx, &ecx, &edx);
1007
1008     PrintDebug("ECX: 0x%x\n", ecx);
1009
1010     if (ecx & CPUID_1_ECX_VTXFLAG) {
1011         v3_get_msr(VMX_FEATURE_CONTROL_MSR, &(feature_msr.hi), &(feature_msr.lo));
1012
1013         PrintDebug("MSRREGlow: 0x%.8x\n", feature_msr.lo);
1014
1015         if ((feature_msr.lo & CPUID_VMX_FEATURES) != CPUID_VMX_FEATURES) {
1016             PrintDebug("VMX is locked -- enable in the BIOS\n");
1017             return 0;
1018         }
1019
1020     } else {
1021         PrintDebug("VMX not supported on this cpu\n");
1022         return 0;
1023     }
1024
1025     return 1;
1026 }
1027
1028
1029 int v3_reset_vmx_vm_core(struct guest_info * core, addr_t rip) {
1030     // init vmcs bios
1031
1032     if ((core->shdw_pg_mode == NESTED_PAGING) &&
1033         (v3_cpu_types[core->pcpu_id] == V3_VMX_EPT_UG_CPU)) {
1034         // easy
1035         core->rip = 0;
1036         core->segments.cs.selector = rip << 8;
1037         core->segments.cs.limit = 0xffff;
1038         core->segments.cs.base = rip << 12;
1039     } else {
1040         core->vm_regs.rdx = core->vcpu_id;
1041         core->vm_regs.rbx = rip;
1042     }
1043
1044     return 0;
1045 }
1046
1047
1048
1049 void v3_init_vmx_cpu(int cpu_id) {
1050     addr_t vmx_on_region = 0;
1051
1052     if (cpu_id == 0) {
1053         if (v3_init_vmx_hw(&hw_info) == -1) {
1054             PrintError("Could not initialize VMX hardware features on cpu %d\n", cpu_id);
1055             return;
1056         }
1057     }
1058
1059     enable_vmx();
1060
1061
1062     // Setup VMXON Region
1063     vmx_on_region = allocate_vmcs();
1064
1065
1066     if (vmx_on(vmx_on_region) == VMX_SUCCESS) {
1067         V3_Print("VMX Enabled\n");
1068         host_vmcs_ptrs[cpu_id] = vmx_on_region;
1069     } else {
1070         V3_Print("VMX already enabled\n");
1071         V3_FreePages((void *)vmx_on_region, 1);
1072     }
1073
1074     PrintDebug("VMXON pointer: 0x%p\n", (void *)host_vmcs_ptrs[cpu_id]);
1075
1076     {
1077         struct vmx_sec_proc_ctrls sec_proc_ctrls;
1078         sec_proc_ctrls.value = v3_vmx_get_ctrl_features(&(hw_info.sec_proc_ctrls));
1079
1080         if (sec_proc_ctrls.enable_ept == 0) {
1081             V3_Print("VMX EPT (Nested) Paging not supported\n");
1082             v3_cpu_types[cpu_id] = V3_VMX_CPU;
1083         } else if (sec_proc_ctrls.unrstrct_guest == 0) {
1084             V3_Print("VMX EPT (Nested) Paging supported\n");
1085             v3_cpu_types[cpu_id] = V3_VMX_EPT_CPU;
1086         } else {
1087             V3_Print("VMX EPT (Nested) Paging + Unrestricted guest supported\n");
1088             v3_cpu_types[cpu_id] = V3_VMX_EPT_UG_CPU;
1089         }
1090     }
1091 }
1092
1093
1094 void v3_deinit_vmx_cpu(int cpu_id) {
1095     extern v3_cpu_arch_t v3_cpu_types[];
1096     v3_cpu_types[cpu_id] = V3_INVALID_CPU;
1097
1098     if (host_vmcs_ptrs[cpu_id] != 0) {
1099         V3_Print("Disabling VMX\n");
1100
1101         if (vmx_off() != VMX_SUCCESS) {
1102             PrintError("Error executing VMXOFF\n");
1103         }
1104
1105         V3_FreePages((void *)host_vmcs_ptrs[cpu_id], 1);
1106
1107         host_vmcs_ptrs[cpu_id] = 0;
1108     }
1109 }