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
  36 #include <palacios/vmx_ept.h>
  37 #include <palacios/vmx_assist.h>
  38 #include <palacios/vmx_hw_info.h>
  39
  40 #ifndef V3_CONFIG_DEBUG_VMX
  41 #undef PrintDebug
  42 #define PrintDebug(fmt, args...)
  43 #endif
  44
  45
  46 /* These fields contain the hardware feature sets supported by the local CPU */
  47 static struct vmx_hw_info hw_info;
  48
  49 extern v3_cpu_arch_t v3_cpu_types[];
  50
  51 static addr_t host_vmcs_ptrs[V3_CONFIG_MAX_CPUS] = { [0 ... V3_CONFIG_MAX_CPUS - 1] = 0};
  52
  53 extern int v3_vmx_launch(struct v3_gprs * vm_regs, struct guest_info * info, struct v3_ctrl_regs * ctrl_regs);
  54 extern int v3_vmx_resume(struct v3_gprs * vm_regs, struct guest_info * info, struct v3_ctrl_regs * ctrl_regs);
  55
  56 static int inline check_vmcs_write(vmcs_field_t field, addr_t val) {
  57     int ret = 0;
  58
  59     ret = vmcs_write(field, val);
  60
  61     if (ret != VMX_SUCCESS) {
  62         PrintError("VMWRITE error on %s!: %d\n", v3_vmcs_field_to_str(field), ret);
  63         return 1;
  64     }
  65
  66     return 0;
  67 }
  68
  69 static int inline check_vmcs_read(vmcs_field_t field, void * val) {
  70     int ret = 0;
  71
  72     ret = vmcs_read(field, val);
  73
  74     if (ret != VMX_SUCCESS) {
  75         PrintError("VMREAD error on %s!: %d\n", v3_vmcs_field_to_str(field), ret);
  76     }
  77
  78     return ret;
  79 }
  80
  81
  82
  83
  84 static addr_t allocate_vmcs() {
  85     struct vmcs_data * vmcs_page = NULL;
  86
  87     PrintDebug("Allocating page\n");
  88
  89     vmcs_page = (struct vmcs_data *)V3_VAddr(V3_AllocPages(1));
  90     memset(vmcs_page, 0, 4096);
  91
  92     vmcs_page->revision = hw_info.basic_info.revision;
  93     PrintDebug("VMX Revision: 0x%x\n", vmcs_page->revision);
  94
  95     return (addr_t)V3_PAddr((void *)vmcs_page);
  96 }
  97
  98
  99
 100
 101 static int init_vmcs_bios(struct guest_info * core, struct vmx_data * vmx_state) {
 102     int vmx_ret = 0;
 103
 104     // disable global interrupts for vm state initialization
 105     v3_disable_ints();
 106
 107     PrintDebug("Loading VMCS\n");
 108     vmx_ret = vmcs_load(vmx_state->vmcs_ptr_phys);
 109     vmx_state->state = VMX_UNLAUNCHED;
 110
 111     if (vmx_ret != VMX_SUCCESS) {
 112         PrintError("VMPTRLD failed\n");
 113         return -1;
 114     }
 115
 116
 117     /*** Setup default state from HW ***/
 118
 119     vmx_state->pin_ctrls.value = hw_info.pin_ctrls.def_val;
 120     vmx_state->pri_proc_ctrls.value = hw_info.proc_ctrls.def_val;
 121     vmx_state->exit_ctrls.value = hw_info.exit_ctrls.def_val;
 122     vmx_state->entry_ctrls.value = hw_info.entry_ctrls.def_val;
 123     vmx_state->sec_proc_ctrls.value = hw_info.sec_proc_ctrls.def_val;
 124
 125     /* Print Control MSRs */
 126     PrintDebug("CR0 MSR: %p\n", (void *)(addr_t)hw_info.cr0.value);
 127     PrintDebug("CR4 MSR: %p\n", (void *)(addr_t)hw_info.cr4.value);
 128
 129
 130
 131     /******* Setup Host State **********/
 132
 133     /* Cache GDTR, IDTR, and TR in host struct */
 134     addr_t gdtr_base;
 135     struct {
 136         uint16_t selector;
 137         addr_t   base;
 138     } __attribute__((packed)) tmp_seg;
 139
 140
 141     __asm__ __volatile__(
 142                          "sgdt (%0);"
 143                          :
 144                          : "q"(&tmp_seg)
 145                          : "memory"
 146                          );
 147     gdtr_base = tmp_seg.base;
 148     vmx_state->host_state.gdtr.base = gdtr_base;
 149
 150     __asm__ __volatile__(
 151                          "sidt (%0);"
 152                          :
 153                          : "q"(&tmp_seg)
 154                          : "memory"
 155                          );
 156     vmx_state->host_state.idtr.base = tmp_seg.base;
 157
 158     __asm__ __volatile__(
 159                          "str (%0);"
 160                          :
 161                          : "q"(&tmp_seg)
 162                          : "memory"
 163                          );
 164     vmx_state->host_state.tr.selector = tmp_seg.selector;
 165
 166     /* The GDTR *index* is bits 3-15 of the selector. */
 167     struct tss_descriptor * desc = NULL;
 168     desc = (struct tss_descriptor *)(gdtr_base + (8 * (tmp_seg.selector >> 3)));
 169
 170     tmp_seg.base = ((desc->base1) |
 171                     (desc->base2 << 16) |
 172                     (desc->base3 << 24) |
 173 #ifdef __V3_64BIT__
 174                     ((uint64_t)desc->base4 << 32)
 175 #else
 176                     (0)
 177 #endif
 178                     );
 179
 180     vmx_state->host_state.tr.base = tmp_seg.base;
 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     vmx_state->pri_proc_ctrls.hlt_exit = 1;
 191
 192
 193     vmx_state->pri_proc_ctrls.pause_exit = 0;
 194     vmx_state->pri_proc_ctrls.tsc_offset = 1;
 195 #ifdef V3_CONFIG_TIME_VIRTUALIZE_TSC
 196     vmx_state->pri_proc_ctrls.rdtsc_exit = 1;
 197 #endif
 198
 199     /* Setup IO map */
 200     vmx_state->pri_proc_ctrls.use_io_bitmap = 1;
 201     vmx_ret |= check_vmcs_write(VMCS_IO_BITMAP_A_ADDR, (addr_t)V3_PAddr(core->vm_info->io_map.arch_data));
 202     vmx_ret |= check_vmcs_write(VMCS_IO_BITMAP_B_ADDR,
 203             (addr_t)V3_PAddr(core->vm_info->io_map.arch_data) + PAGE_SIZE_4KB);
 204
 205
 206     vmx_state->pri_proc_ctrls.use_msr_bitmap = 1;
 207     vmx_ret |= check_vmcs_write(VMCS_MSR_BITMAP, (addr_t)V3_PAddr(core->vm_info->msr_map.arch_data));
 208
 209
 210
 211
 212
 213
 214
 215 #ifdef __V3_64BIT__
 216     // Ensure host runs in 64-bit mode at each VM EXIT
 217     vmx_state->exit_ctrls.host_64_on = 1;
 218 #endif
 219
 220     // Hook all accesses to EFER register
 221     v3_hook_msr(core->vm_info, EFER_MSR,
 222                 &v3_handle_efer_read,
 223                 &v3_handle_efer_write,
 224                 core);
 225
 226     // Restore host's EFER register on each VM EXIT
 227     vmx_state->exit_ctrls.ld_efer = 1;
 228
 229     // Save/restore guest's EFER register to/from VMCS on VM EXIT/ENTRY
 230     vmx_state->exit_ctrls.save_efer = 1;
 231     vmx_state->entry_ctrls.ld_efer  = 1;
 232
 233     // Cause VM_EXIT whenever CR4.VMXE or CR4.PAE bits are written
 234     vmx_ret |= check_vmcs_write(VMCS_CR4_MASK, CR4_VMXE | CR4_PAE);
 235
 236
 237     /* Setup paging */
 238     if (core->shdw_pg_mode == SHADOW_PAGING) {
 239         PrintDebug("Creating initial shadow page table\n");
 240
 241         if (v3_init_passthrough_pts(core) == -1) {
 242             PrintError("Could not initialize passthrough page tables\n");
 243             return -1;
 244         }
 245
 246 #define CR0_PE 0x00000001
 247 #define CR0_PG 0x80000000
 248 #define CR0_WP 0x00010000 // To ensure mem hooks work
 249         vmx_ret |= check_vmcs_write(VMCS_CR0_MASK, (CR0_PE | CR0_PG | CR0_WP));
 250
 251         core->ctrl_regs.cr3 = core->direct_map_pt;
 252
 253         // vmx_state->pinbased_ctrls |= NMI_EXIT;
 254
 255         /* Add CR exits */
 256         vmx_state->pri_proc_ctrls.cr3_ld_exit = 1;
 257         vmx_state->pri_proc_ctrls.cr3_str_exit = 1;
 258
 259         vmx_state->pri_proc_ctrls.invlpg_exit = 1;
 260
 261         /* Add page fault exits */
 262         vmx_state->excp_bmap.pf = 1;
 263
 264         // Setup VMX Assist
 265         v3_vmxassist_init(core, vmx_state);
 266
 267     } else if ((core->shdw_pg_mode == NESTED_PAGING) &&
 268                (v3_cpu_types[core->pcpu_id] == V3_VMX_EPT_CPU)) {
 269
 270 #define CR0_PE 0x00000001
 271 #define CR0_PG 0x80000000
 272 #define CR0_WP 0x00010000 // To ensure mem hooks work
 273         vmx_ret |= check_vmcs_write(VMCS_CR0_MASK, (CR0_PE | CR0_PG | CR0_WP));
 274
 275         // vmx_state->pinbased_ctrls |= NMI_EXIT;
 276
 277         /* Disable CR exits */
 278         vmx_state->pri_proc_ctrls.cr3_ld_exit = 0;
 279         vmx_state->pri_proc_ctrls.cr3_str_exit = 0;
 280
 281         vmx_state->pri_proc_ctrls.invlpg_exit = 0;
 282
 283         /* Add page fault exits */
 284         //      vmx_state->excp_bmap.pf = 1; // This should never happen..., enabled to catch bugs
 285
 286         // Setup VMX Assist
 287         v3_vmxassist_init(core, vmx_state);
 288
 289         /* Enable EPT */
 290         vmx_state->pri_proc_ctrls.sec_ctrls = 1; // Enable secondary proc controls
 291         vmx_state->sec_proc_ctrls.enable_ept = 1; // enable EPT paging
 292
 293
 294
 295         if (v3_init_ept(core, &hw_info) == -1) {
 296             PrintError("Error initializing EPT\n");
 297             return -1;
 298         }
 299
 300     } else if ((core->shdw_pg_mode == NESTED_PAGING) &&
 301                (v3_cpu_types[core->pcpu_id] == V3_VMX_EPT_UG_CPU)) {
 302         int i = 0;
 303         // For now we will assume that unrestricted guest mode is assured w/ EPT
 304
 305
 306         core->vm_regs.rsp = 0x00;
 307         core->rip = 0xfff0;
 308         core->vm_regs.rdx = 0x00000f00;
 309         core->ctrl_regs.rflags = 0x00000002; // The reserved bit is always 1
 310         core->ctrl_regs.cr0 = 0x60010010; // Set the WP flag so the memory hooks work in real-mode
 311
 312
 313         core->segments.cs.selector = 0xf000;
 314         core->segments.cs.limit = 0xffff;
 315         core->segments.cs.base = 0x0000000f0000LL;
 316
 317         // (raw attributes = 0xf3)
 318         core->segments.cs.type = 0xb;
 319         core->segments.cs.system = 0x1;
 320         core->segments.cs.dpl = 0x0;
 321         core->segments.cs.present = 1;
 322
 323
 324
 325         struct v3_segment * segregs [] = {&(core->segments.ss), &(core->segments.ds),
 326                                           &(core->segments.es), &(core->segments.fs),
 327                                           &(core->segments.gs), NULL};
 328
 329         for ( i = 0; segregs[i] != NULL; i++) {
 330             struct v3_segment * seg = segregs[i];
 331
 332             seg->selector = 0x0000;
 333             //    seg->base = seg->selector << 4;
 334             seg->base = 0x00000000;
 335             seg->limit = 0xffff;
 336
 337
 338             seg->type = 0x3;
 339             seg->system = 0x1;
 340             seg->dpl = 0x0;
 341             seg->present = 1;
 342             //    seg->granularity = 1;
 343
 344         }
 345
 346
 347         core->segments.gdtr.limit = 0x0000ffff;
 348         core->segments.gdtr.base = 0x0000000000000000LL;
 349
 350         core->segments.idtr.limit = 0x0000ffff;
 351         core->segments.idtr.base = 0x0000000000000000LL;
 352
 353         core->segments.ldtr.selector = 0x0000;
 354         core->segments.ldtr.limit = 0x0000ffff;
 355         core->segments.ldtr.base = 0x0000000000000000LL;
 356         core->segments.ldtr.type = 2;
 357         core->segments.ldtr.present = 1;
 358
 359         core->segments.tr.selector = 0x0000;
 360         core->segments.tr.limit = 0x0000ffff;
 361         core->segments.tr.base = 0x0000000000000000LL;
 362         core->segments.tr.type = 0xb;
 363         core->segments.tr.present = 1;
 364
 365         //      core->dbg_regs.dr6 = 0x00000000ffff0ff0LL;
 366         core->dbg_regs.dr7 = 0x0000000000000400LL;
 367
 368         /* Enable EPT */
 369         vmx_state->pri_proc_ctrls.sec_ctrls = 1; // Enable secondary proc controls
 370         vmx_state->sec_proc_ctrls.enable_ept = 1; // enable EPT paging
 371         vmx_state->sec_proc_ctrls.unrstrct_guest = 1; // enable unrestricted guest operation
 372
 373
 374         /* Disable shadow paging stuff */
 375         vmx_state->pri_proc_ctrls.cr3_ld_exit = 0;
 376         vmx_state->pri_proc_ctrls.cr3_str_exit = 0;
 377
 378         vmx_state->pri_proc_ctrls.invlpg_exit = 0;
 379
 380
 381         if (v3_init_ept(core, &hw_info) == -1) {
 382             PrintError("Error initializing EPT\n");
 383             return -1;
 384         }
 385
 386     } else {
 387         PrintError("Invalid Virtual paging mode\n");
 388         return -1;
 389     }
 390
 391
 392     // hook vmx msrs
 393
 394     // Setup SYSCALL/SYSENTER MSRs in load/store area
 395
 396     // save STAR, LSTAR, FMASK, KERNEL_GS_BASE MSRs in MSR load/store area
 397     {
 398 #define IA32_STAR 0xc0000081
 399 #define IA32_LSTAR 0xc0000082
 400 #define IA32_FMASK 0xc0000084
 401 #define IA32_KERN_GS_BASE 0xc0000102
 402
 403 #define IA32_CSTAR 0xc0000083 // Compatibility mode STAR (ignored for now... hopefully its not that important...)
 404
 405         int msr_ret = 0;
 406
 407         struct vmcs_msr_entry * exit_store_msrs = NULL;
 408         struct vmcs_msr_entry * exit_load_msrs = NULL;
 409         struct vmcs_msr_entry * entry_load_msrs = NULL;;
 410         int max_msrs = (hw_info.misc_info.max_msr_cache_size + 1) * 4;
 411
 412         V3_Print("Setting up MSR load/store areas (max_msr_count=%d)\n", max_msrs);
 413
 414         if (max_msrs < 4) {
 415             PrintError("Max MSR cache size is too small (%d)\n", max_msrs);
 416             return -1;
 417         }
 418
 419         vmx_state->msr_area = V3_VAddr(V3_AllocPages(1));
 420
 421         if (vmx_state->msr_area == NULL) {
 422             PrintError("could not allocate msr load/store area\n");
 423             return -1;
 424         }
 425
 426         msr_ret |= check_vmcs_write(VMCS_EXIT_MSR_STORE_CNT, 4);
 427         msr_ret |= check_vmcs_write(VMCS_EXIT_MSR_LOAD_CNT, 4);
 428         msr_ret |= check_vmcs_write(VMCS_ENTRY_MSR_LOAD_CNT, 4);
 429
 430
 431         exit_store_msrs = (struct vmcs_msr_entry *)(vmx_state->msr_area);
 432         exit_load_msrs = (struct vmcs_msr_entry *)(vmx_state->msr_area + (sizeof(struct vmcs_msr_entry) * 4));
 433         entry_load_msrs = (struct vmcs_msr_entry *)(vmx_state->msr_area + (sizeof(struct vmcs_msr_entry) * 8));
 434
 435
 436         exit_store_msrs[0].index = IA32_STAR;
 437         exit_store_msrs[1].index = IA32_LSTAR;
 438         exit_store_msrs[2].index = IA32_FMASK;
 439         exit_store_msrs[3].index = IA32_KERN_GS_BASE;
 440
 441         memcpy(exit_store_msrs, exit_load_msrs, sizeof(struct vmcs_msr_entry) * 4);
 442         memcpy(exit_store_msrs, entry_load_msrs, sizeof(struct vmcs_msr_entry) * 4);
 443
 444
 445         v3_get_msr(IA32_STAR, &(exit_load_msrs[0].hi), &(exit_load_msrs[0].lo));
 446         v3_get_msr(IA32_LSTAR, &(exit_load_msrs[1].hi), &(exit_load_msrs[1].lo));
 447         v3_get_msr(IA32_FMASK, &(exit_load_msrs[2].hi), &(exit_load_msrs[2].lo));
 448         v3_get_msr(IA32_KERN_GS_BASE, &(exit_load_msrs[3].hi), &(exit_load_msrs[3].lo));
 449
 450         msr_ret |= check_vmcs_write(VMCS_EXIT_MSR_STORE_ADDR, (addr_t)V3_PAddr(exit_store_msrs));
 451         msr_ret |= check_vmcs_write(VMCS_EXIT_MSR_LOAD_ADDR, (addr_t)V3_PAddr(exit_load_msrs));
 452         msr_ret |= check_vmcs_write(VMCS_ENTRY_MSR_LOAD_ADDR, (addr_t)V3_PAddr(entry_load_msrs));
 453
 454     }
 455
 456     /* Sanity check ctrl/reg fields against hw_defaults */
 457
 458
 459
 460
 461     /*** Write all the info to the VMCS ***/
 462
 463     /*
 464     {
 465         // IS THIS NECESSARY???
 466 #define DEBUGCTL_MSR 0x1d9
 467         struct v3_msr tmp_msr;
 468         v3_get_msr(DEBUGCTL_MSR, &(tmp_msr.hi), &(tmp_msr.lo));
 469         vmx_ret |= check_vmcs_write(VMCS_GUEST_DBG_CTL, tmp_msr.value);
 470         core->dbg_regs.dr7 = 0x400;
 471     }
 472     */
 473
 474 #ifdef __V3_64BIT__
 475     vmx_ret |= check_vmcs_write(VMCS_LINK_PTR, (addr_t)0xffffffffffffffffULL);
 476 #else
 477     vmx_ret |= check_vmcs_write(VMCS_LINK_PTR, (addr_t)0xffffffffUL);
 478     vmx_ret |= check_vmcs_write(VMCS_LINK_PTR_HIGH, (addr_t)0xffffffffUL);
 479 #endif
 480
 481
 482
 483
 484     if (v3_update_vmcs_ctrl_fields(core)) {
 485         PrintError("Could not write control fields!\n");
 486         return -1;
 487     }
 488
 489     if (v3_update_vmcs_host_state(core)) {
 490         PrintError("Could not write host state\n");
 491         return -1;
 492     }
 493
 494     // reenable global interrupts for vm state initialization now
 495     // that the vm state is initialized. If another VM kicks us off,
 496     // it'll update our vmx state so that we know to reload ourself
 497     v3_enable_ints();
 498
 499     return 0;
 500 }
 501
 502 int v3_init_vmx_vmcs(struct guest_info * core, v3_vm_class_t vm_class) {
 503     struct vmx_data * vmx_state = NULL;
 504     int vmx_ret = 0;
 505
 506     vmx_state = (struct vmx_data *)V3_Malloc(sizeof(struct vmx_data));
 507     memset(vmx_state, 0, sizeof(struct vmx_data));
 508
 509     PrintDebug("vmx_data pointer: %p\n", (void *)vmx_state);
 510
 511     PrintDebug("Allocating VMCS\n");
 512     vmx_state->vmcs_ptr_phys = allocate_vmcs();
 513
 514     PrintDebug("VMCS pointer: %p\n", (void *)(vmx_state->vmcs_ptr_phys));
 515
 516     core->vmm_data = vmx_state;
 517     vmx_state->state = VMX_UNLAUNCHED;
 518
 519     PrintDebug("Initializing VMCS (addr=%p)\n", core->vmm_data);
 520
 521     // TODO: Fix vmcs fields so they're 32-bit
 522
 523     PrintDebug("Clearing VMCS: %p\n", (void *)vmx_state->vmcs_ptr_phys);
 524     vmx_ret = vmcs_clear(vmx_state->vmcs_ptr_phys);
 525
 526     if (vmx_ret != VMX_SUCCESS) {
 527         PrintError("VMCLEAR failed\n");
 528         return -1;
 529     }
 530
 531     if (vm_class == V3_PC_VM) {
 532         PrintDebug("Initializing VMCS\n");
 533         if (init_vmcs_bios(core, vmx_state) == -1) {
 534             PrintError("Error initializing VMCS to BIOS state\n");
 535             return -1;
 536         }
 537     } else {
 538         PrintError("Invalid VM Class\n");
 539         return -1;
 540     }
 541
 542     PrintDebug("Serializing VMCS: %p\n", (void *)vmx_state->vmcs_ptr_phys);
 543     vmx_ret = vmcs_clear(vmx_state->vmcs_ptr_phys);
 544
 545     return 0;
 546 }
 547
 548
 549 int v3_deinit_vmx_vmcs(struct guest_info * core) {
 550     struct vmx_data * vmx_state = core->vmm_data;
 551
 552     V3_FreePages((void *)(vmx_state->vmcs_ptr_phys), 1);
 553     V3_FreePages(V3_PAddr(vmx_state->msr_area), 1);
 554
 555     V3_Free(vmx_state);
 556
 557     return 0;
 558 }
 559
 560
 561 static int update_irq_exit_state(struct guest_info * info) {
 562     struct vmx_exit_idt_vec_info idt_vec_info;
 563
 564     check_vmcs_read(VMCS_IDT_VECTOR_INFO, &(idt_vec_info.value));
 565
 566     if ((info->intr_core_state.irq_started == 1) && (idt_vec_info.valid == 0)) {
 567 #ifdef V3_CONFIG_DEBUG_INTERRUPTS
 568         V3_Print("Calling v3_injecting_intr\n");
 569 #endif
 570         info->intr_core_state.irq_started = 0;
 571         v3_injecting_intr(info, info->intr_core_state.irq_vector, V3_EXTERNAL_IRQ);
 572     }
 573
 574     return 0;
 575 }
 576
 577 static int update_irq_entry_state(struct guest_info * info) {
 578     struct vmx_exit_idt_vec_info idt_vec_info;
 579     struct vmcs_interrupt_state intr_core_state;
 580     struct vmx_data * vmx_info = (struct vmx_data *)(info->vmm_data);
 581
 582     check_vmcs_read(VMCS_IDT_VECTOR_INFO, &(idt_vec_info.value));
 583     check_vmcs_read(VMCS_GUEST_INT_STATE, &(intr_core_state));
 584
 585     /* Check for pending exceptions to inject */
 586     if (v3_excp_pending(info)) {
 587         struct vmx_entry_int_info int_info;
 588         int_info.value = 0;
 589
 590         // In VMX, almost every exception is hardware
 591         // Software exceptions are pretty much only for breakpoint or overflow
 592         int_info.type = 3;
 593         int_info.vector = v3_get_excp_number(info);
 594
 595         if (info->excp_state.excp_error_code_valid) {
 596             check_vmcs_write(VMCS_ENTRY_EXCP_ERR, info->excp_state.excp_error_code);
 597             int_info.error_code = 1;
 598
 599 #ifdef V3_CONFIG_DEBUG_INTERRUPTS
 600             V3_Print("Injecting exception %d with error code %x\n",
 601                     int_info.vector, info->excp_state.excp_error_code);
 602 #endif
 603         }
 604
 605         int_info.valid = 1;
 606 #ifdef V3_CONFIG_DEBUG_INTERRUPTS
 607         V3_Print("Injecting exception %d (EIP=%p)\n", int_info.vector, (void *)(addr_t)info->rip);
 608 #endif
 609         check_vmcs_write(VMCS_ENTRY_INT_INFO, int_info.value);
 610
 611         v3_injecting_excp(info, int_info.vector);
 612
 613     } else if ((((struct rflags *)&(info->ctrl_regs.rflags))->intr == 1) &&
 614                (intr_core_state.val == 0)) {
 615
 616         if ((info->intr_core_state.irq_started == 1) && (idt_vec_info.valid == 1)) {
 617
 618 #ifdef V3_CONFIG_DEBUG_INTERRUPTS
 619             V3_Print("IRQ pending from previous injection\n");
 620 #endif
 621
 622             // Copy the IDT vectoring info over to reinject the old interrupt
 623             if (idt_vec_info.error_code == 1) {
 624                 uint32_t err_code = 0;
 625
 626                 check_vmcs_read(VMCS_IDT_VECTOR_ERR, &err_code);
 627                 check_vmcs_write(VMCS_ENTRY_EXCP_ERR, err_code);
 628             }
 629
 630             idt_vec_info.undef = 0;
 631             check_vmcs_write(VMCS_ENTRY_INT_INFO, idt_vec_info.value);
 632
 633         } else {
 634             struct vmx_entry_int_info ent_int;
 635             ent_int.value = 0;
 636
 637             switch (v3_intr_pending(info)) {
 638                 case V3_EXTERNAL_IRQ: {
 639                     info->intr_core_state.irq_vector = v3_get_intr(info);
 640                     ent_int.vector = info->intr_core_state.irq_vector;
 641                     ent_int.type = 0;
 642                     ent_int.error_code = 0;
 643                     ent_int.valid = 1;
 644
 645 #ifdef V3_CONFIG_DEBUG_INTERRUPTS
 646                     V3_Print("Injecting Interrupt %d at exit %u(EIP=%p)\n",
 647                                info->intr_core_state.irq_vector,
 648                                (uint32_t)info->num_exits,
 649                                (void *)(addr_t)info->rip);
 650 #endif
 651
 652                     check_vmcs_write(VMCS_ENTRY_INT_INFO, ent_int.value);
 653                     info->intr_core_state.irq_started = 1;
 654
 655                     break;
 656                 }
 657                 case V3_NMI:
 658                     PrintDebug("Injecting NMI\n");
 659
 660                     ent_int.type = 2;
 661                     ent_int.vector = 2;
 662                     ent_int.valid = 1;
 663                     check_vmcs_write(VMCS_ENTRY_INT_INFO, ent_int.value);
 664
 665                     break;
 666                 case V3_SOFTWARE_INTR:
 667                     PrintDebug("Injecting software interrupt\n");
 668                     ent_int.type = 4;
 669
 670                     ent_int.valid = 1;
 671                     check_vmcs_write(VMCS_ENTRY_INT_INFO, ent_int.value);
 672
 673                     break;
 674                 case V3_VIRTUAL_IRQ:
 675                     // Not sure what to do here, Intel doesn't have virtual IRQs
 676                     // May be the same as external interrupts/IRQs
 677
 678                     break;
 679                 case V3_INVALID_INTR:
 680                 default:
 681                     break;
 682             }
 683         }
 684     } else if ((v3_intr_pending(info)) && (vmx_info->pri_proc_ctrls.int_wndw_exit == 0)) {
 685         // Enable INTR window exiting so we know when IF=1
 686         uint32_t instr_len;
 687
 688         check_vmcs_read(VMCS_EXIT_INSTR_LEN, &instr_len);
 689
 690 #ifdef V3_CONFIG_DEBUG_INTERRUPTS
 691         V3_Print("Enabling Interrupt-Window exiting: %d\n", instr_len);
 692 #endif
 693
 694         vmx_info->pri_proc_ctrls.int_wndw_exit = 1;
 695         check_vmcs_write(VMCS_PROC_CTRLS, vmx_info->pri_proc_ctrls.value);
 696     }
 697
 698
 699     return 0;
 700 }
 701
 702
 703
 704 static struct vmx_exit_info exit_log[10];
 705
 706 static void print_exit_log(struct guest_info * info) {
 707     int cnt = info->num_exits % 10;
 708     int i = 0;
 709
 710
 711     V3_Print("\nExit Log (%d total exits):\n", (uint32_t)info->num_exits);
 712
 713     for (i = 0; i < 10; i++) {
 714         struct vmx_exit_info * tmp = &exit_log[cnt];
 715
 716         V3_Print("%d:\texit_reason = %p\n", i, (void *)(addr_t)tmp->exit_reason);
 717         V3_Print("\texit_qual = %p\n", (void *)tmp->exit_qual);
 718         V3_Print("\tint_info = %p\n", (void *)(addr_t)tmp->int_info);
 719         V3_Print("\tint_err = %p\n", (void *)(addr_t)tmp->int_err);
 720         V3_Print("\tinstr_info = %p\n", (void *)(addr_t)tmp->instr_info);
 721
 722         cnt--;
 723
 724         if (cnt == -1) {
 725             cnt = 9;
 726         }
 727
 728     }
 729
 730 }
 731
 732 /*
 733  * CAUTION and DANGER!!!
 734  *
 735  * The VMCS CANNOT(!!) be accessed outside of the cli/sti calls inside this function
 736  * When exectuing a symbiotic call, the VMCS WILL be overwritten, so any dependencies
 737  * on its contents will cause things to break. The contents at the time of the exit WILL
 738  * change before the exit handler is executed.
 739  */
 740 int v3_vmx_enter(struct guest_info * info) {
 741     int ret = 0;
 742     uint32_t tsc_offset_low, tsc_offset_high;
 743     struct vmx_exit_info exit_info;
 744     struct vmx_data * vmx_info = (struct vmx_data *)(info->vmm_data);
 745
 746     // Conditionally yield the CPU if the timeslice has expired
 747     v3_yield_cond(info);
 748
 749     // Perform any additional yielding needed for time adjustment
 750     v3_adjust_time(info);
 751
 752     // disable global interrupts for vm state transition
 753     v3_disable_ints();
 754
 755     // Update timer devices late after being in the VM so that as much
 756     // of hte time in the VM is accounted for as possible. Also do it before
 757     // updating IRQ entry state so that any interrupts the timers raise get
 758     // handled on the next VM entry. Must be done with interrupts disabled.
 759     v3_update_timers(info);
 760
 761     if (vmcs_store() != vmx_info->vmcs_ptr_phys) {
 762         vmcs_clear(vmx_info->vmcs_ptr_phys);
 763         vmcs_load(vmx_info->vmcs_ptr_phys);
 764         vmx_info->state = VMX_UNLAUNCHED;
 765     }
 766
 767     v3_vmx_restore_vmcs(info);
 768
 769
 770 #ifdef V3_CONFIG_SYMCALL
 771     if (info->sym_core_state.symcall_state.sym_call_active == 0) {
 772         update_irq_entry_state(info);
 773     }
 774 #else
 775     update_irq_entry_state(info);
 776 #endif
 777
 778     {
 779         addr_t guest_cr3;
 780         vmcs_read(VMCS_GUEST_CR3, &guest_cr3);
 781         vmcs_write(VMCS_GUEST_CR3, guest_cr3);
 782     }
 783
 784     // Perform last-minute time bookkeeping prior to entering the VM
 785     v3_time_enter_vm(info);
 786
 787     tsc_offset_high = (uint32_t)((v3_tsc_host_offset(&info->time_state) >> 32) & 0xffffffff);
 788     tsc_offset_low = (uint32_t)(v3_tsc_host_offset(&info->time_state) & 0xffffffff);
 789     check_vmcs_write(VMCS_TSC_OFFSET_HIGH, tsc_offset_high);
 790     check_vmcs_write(VMCS_TSC_OFFSET, tsc_offset_low);
 791
 792
 793     if (v3_update_vmcs_host_state(info)) {
 794         v3_enable_ints();
 795         PrintError("Could not write host state\n");
 796         return -1;
 797     }
 798
 799
 800     if (vmx_info->state == VMX_UNLAUNCHED) {
 801         vmx_info->state = VMX_LAUNCHED;
 802
 803         info->vm_info->run_state = VM_RUNNING;
 804         ret = v3_vmx_launch(&(info->vm_regs), info, &(info->ctrl_regs));
 805     } else {
 806         V3_ASSERT(vmx_info->state != VMX_UNLAUNCHED);
 807         ret = v3_vmx_resume(&(info->vm_regs), info, &(info->ctrl_regs));
 808     }
 809
 810     //  PrintDebug("VMX Exit: ret=%d\n", ret);
 811
 812     if (ret != VMX_SUCCESS) {
 813         uint32_t error = 0;
 814
 815         vmcs_read(VMCS_INSTR_ERR, &error);
 816
 817         v3_enable_ints();
 818
 819         PrintError("VMENTRY Error: %d\n", error);
 820         return -1;
 821     }
 822
 823     // Immediate exit from VM time bookkeeping
 824     v3_time_exit_vm(info);
 825
 826     info->num_exits++;
 827
 828     /* Update guest state */
 829     v3_vmx_save_vmcs(info);
 830
 831     // info->cpl = info->segments.cs.selector & 0x3;
 832
 833     info->mem_mode = v3_get_vm_mem_mode(info);
 834     info->cpu_mode = v3_get_vm_cpu_mode(info);
 835
 836
 837     check_vmcs_read(VMCS_EXIT_INSTR_LEN, &(exit_info.instr_len));
 838     check_vmcs_read(VMCS_EXIT_INSTR_INFO, &(exit_info.instr_info));
 839     check_vmcs_read(VMCS_EXIT_REASON, &(exit_info.exit_reason));
 840     check_vmcs_read(VMCS_EXIT_QUAL, &(exit_info.exit_qual));
 841     check_vmcs_read(VMCS_EXIT_INT_INFO, &(exit_info.int_info));
 842     check_vmcs_read(VMCS_EXIT_INT_ERR, &(exit_info.int_err));
 843     check_vmcs_read(VMCS_GUEST_LINEAR_ADDR, &(exit_info.guest_linear_addr));
 844
 845     if (info->shdw_pg_mode == NESTED_PAGING) {
 846         check_vmcs_read(VMCS_GUEST_PHYS_ADDR, &(exit_info.ept_fault_addr));
 847     }
 848
 849     //PrintDebug("VMX Exit taken, id-qual: %u-%lu\n", exit_info.exit_reason, exit_info.exit_qual);
 850
 851     exit_log[info->num_exits % 10] = exit_info;
 852
 853 #ifdef V3_CONFIG_SYMCALL
 854     if (info->sym_core_state.symcall_state.sym_call_active == 0) {
 855         update_irq_exit_state(info);
 856     }
 857 #else
 858     update_irq_exit_state(info);
 859 #endif
 860
 861     if (exit_info.exit_reason == VMEXIT_INTR_WINDOW) {
 862         // This is a special case whose only job is to inject an interrupt
 863         vmcs_read(VMCS_PROC_CTRLS, &(vmx_info->pri_proc_ctrls.value));
 864         vmx_info->pri_proc_ctrls.int_wndw_exit = 0;
 865         vmcs_write(VMCS_PROC_CTRLS, vmx_info->pri_proc_ctrls.value);
 866
 867 #ifdef V3_CONFIG_DEBUG_INTERRUPTS
 868        V3_Print("Interrupts available again! (RIP=%llx)\n", info->rip);
 869 #endif
 870     }
 871
 872     // reenable global interrupts after vm exit
 873     v3_enable_ints();
 874
 875     // Conditionally yield the CPU if the timeslice has expired
 876     v3_yield_cond(info);
 877
 878     if (v3_handle_vmx_exit(info, &exit_info) == -1) {
 879         PrintError("Error in VMX exit handler (Exit reason=%x)\n", exit_info.exit_reason);
 880         return -1;
 881     }
 882
 883     return 0;
 884 }
 885
 886
 887 int v3_start_vmx_guest(struct guest_info * info) {
 888
 889     PrintDebug("Starting VMX core %u\n", info->vcpu_id);
 890
 891     if (info->vcpu_id == 0) {
 892         info->core_run_state = CORE_RUNNING;
 893         info->vm_info->run_state = VM_RUNNING;
 894     } else {
 895
 896         PrintDebug("VMX core %u: Waiting for core initialization\n", info->vcpu_id);
 897
 898         while (info->core_run_state == CORE_STOPPED) {
 899             v3_yield(info);
 900             //PrintDebug("VMX core %u: still waiting for INIT\n",info->vcpu_id);
 901         }
 902
 903         PrintDebug("VMX core %u initialized\n", info->vcpu_id);
 904     }
 905
 906
 907     PrintDebug("VMX core %u: I am starting at CS=0x%x (base=0x%p, limit=0x%x),  RIP=0x%p\n",
 908                info->vcpu_id, info->segments.cs.selector, (void *)(info->segments.cs.base),
 909                info->segments.cs.limit, (void *)(info->rip));
 910
 911
 912     PrintDebug("VMX core %u: Launching VMX VM on logical core %u\n", info->vcpu_id, info->pcpu_id);
 913
 914     v3_start_time(info);
 915
 916     while (1) {
 917
 918         if (info->vm_info->run_state == VM_STOPPED) {
 919             info->core_run_state = CORE_STOPPED;
 920             break;
 921         }
 922
 923         if (v3_vmx_enter(info) == -1) {
 924
 925             addr_t host_addr;
 926             addr_t linear_addr = 0;
 927
 928             info->vm_info->run_state = VM_ERROR;
 929
 930             V3_Print("VMX core %u: VMX ERROR!!\n", info->vcpu_id);
 931
 932             v3_print_guest_state(info);
 933
 934             V3_Print("VMX core %u\n", info->vcpu_id);
 935
 936             linear_addr = get_addr_linear(info, info->rip, &(info->segments.cs));
 937
 938             if (info->mem_mode == PHYSICAL_MEM) {
 939                 v3_gpa_to_hva(info, linear_addr, &host_addr);
 940             } else if (info->mem_mode == VIRTUAL_MEM) {
 941                 v3_gva_to_hva(info, linear_addr, &host_addr);
 942             }
 943
 944             V3_Print("VMX core %u: Host Address of rip = 0x%p\n", info->vcpu_id, (void *)host_addr);
 945
 946             V3_Print("VMX core %u: Instr (15 bytes) at %p:\n", info->vcpu_id, (void *)host_addr);
 947             v3_dump_mem((uint8_t *)host_addr, 15);
 948
 949             v3_print_stack(info);
 950
 951
 952             v3_print_vmcs();
 953             print_exit_log(info);
 954             return -1;
 955         }
 956
 957
 958
 959         if (info->vm_info->run_state == VM_STOPPED) {
 960             info->core_run_state = CORE_STOPPED;
 961             break;
 962         }
 963 /*
 964         if ((info->num_exits % 5000) == 0) {
 965             V3_Print("VMX Exit number %d\n", (uint32_t)info->num_exits);
 966         }
 967 */
 968
 969     }
 970
 971     return 0;
 972 }
 973
 974
 975
 976
 977 #define VMX_FEATURE_CONTROL_MSR     0x0000003a
 978 #define CPUID_VMX_FEATURES 0x00000005  /* LOCK and VMXON */
 979 #define CPUID_1_ECX_VTXFLAG 0x00000020
 980
 981 int v3_is_vmx_capable() {
 982     v3_msr_t feature_msr;
 983     uint32_t eax = 0, ebx = 0, ecx = 0, edx = 0;
 984
 985     v3_cpuid(0x1, &eax, &ebx, &ecx, &edx);
 986
 987     PrintDebug("ECX: 0x%x\n", ecx);
 988
 989     if (ecx & CPUID_1_ECX_VTXFLAG) {
 990         v3_get_msr(VMX_FEATURE_CONTROL_MSR, &(feature_msr.hi), &(feature_msr.lo));
 991
 992         PrintDebug("MSRREGlow: 0x%.8x\n", feature_msr.lo);
 993
 994         if ((feature_msr.lo & CPUID_VMX_FEATURES) != CPUID_VMX_FEATURES) {
 995             PrintDebug("VMX is locked -- enable in the BIOS\n");
 996             return 0;
 997         }
 998
 999     } else {
1000         PrintDebug("VMX not supported on this cpu\n");
1001         return 0;
1002     }
1003
1004     return 1;
1005 }
1006
1007
1008 int v3_reset_vmx_vm_core(struct guest_info * core, addr_t rip) {
1009     // init vmcs bios
1010
1011     if ((core->shdw_pg_mode == NESTED_PAGING) &&
1012         (v3_cpu_types[core->pcpu_id] == V3_VMX_EPT_UG_CPU)) {
1013         // easy
1014         core->rip = 0;
1015         core->segments.cs.selector = rip << 8;
1016         core->segments.cs.limit = 0xffff;
1017         core->segments.cs.base = rip << 12;
1018     } else {
1019         core->vm_regs.rdx = core->vcpu_id;
1020         core->vm_regs.rbx = rip;
1021     }
1022
1023     return 0;
1024 }
1025
1026
1027
1028 void v3_init_vmx_cpu(int cpu_id) {
1029     addr_t vmx_on_region = 0;
1030
1031     if (cpu_id == 0) {
1032         if (v3_init_vmx_hw(&hw_info) == -1) {
1033             PrintError("Could not initialize VMX hardware features on cpu %d\n", cpu_id);
1034             return;
1035         }
1036     }
1037
1038     enable_vmx();
1039
1040
1041     // Setup VMXON Region
1042     vmx_on_region = allocate_vmcs();
1043
1044
1045     if (vmx_on(vmx_on_region) == VMX_SUCCESS) {
1046         V3_Print("VMX Enabled\n");
1047         host_vmcs_ptrs[cpu_id] = vmx_on_region;
1048     } else {
1049         V3_Print("VMX already enabled\n");
1050         V3_FreePages((void *)vmx_on_region, 1);
1051     }
1052
1053     PrintDebug("VMXON pointer: 0x%p\n", (void *)host_vmcs_ptrs[cpu_id]);
1054
1055     {
1056         struct vmx_sec_proc_ctrls sec_proc_ctrls;
1057         sec_proc_ctrls.value = v3_vmx_get_ctrl_features(&(hw_info.sec_proc_ctrls));
1058
1059         if (sec_proc_ctrls.enable_ept == 0) {
1060             V3_Print("VMX EPT (Nested) Paging not supported\n");
1061             v3_cpu_types[cpu_id] = V3_VMX_CPU;
1062         } else if (sec_proc_ctrls.unrstrct_guest == 0) {
1063             V3_Print("VMX EPT (Nested) Paging supported\n");
1064             v3_cpu_types[cpu_id] = V3_VMX_EPT_CPU;
1065         } else {
1066             V3_Print("VMX EPT (Nested) Paging + Unrestricted guest supported\n");
1067             v3_cpu_types[cpu_id] = V3_VMX_EPT_UG_CPU;
1068         }
1069     }
1070 }
1071
1072
1073 void v3_deinit_vmx_cpu(int cpu_id) {
1074     extern v3_cpu_arch_t v3_cpu_types[];
1075     v3_cpu_types[cpu_id] = V3_INVALID_CPU;
1076
1077     if (host_vmcs_ptrs[cpu_id] != 0) {
1078         V3_Print("Disabling VMX\n");
1079
1080         if (vmx_off() != VMX_SUCCESS) {
1081             PrintError("Error executing VMXOFF\n");
1082         }
1083
1084         V3_FreePages((void *)host_vmcs_ptrs[cpu_id], 1);
1085
1086         host_vmcs_ptrs[cpu_id] = 0;
1087     }
1088 }