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
  35 #include <palacios/vmx_ept.h>
  36 #include <palacios/vmx_assist.h>
  37 #include <palacios/vmx_hw_info.h>
  38
  39 #ifndef V3_CONFIG_DEBUG_VMX
  40 #undef PrintDebug
  41 #define PrintDebug(fmt, args...)
  42 #endif
  43
  44
  45 /* These fields contain the hardware feature sets supported by the local CPU */
  46 static struct vmx_hw_info hw_info;
  47
  48 extern v3_cpu_arch_t v3_cpu_types[];
  49
  50 static addr_t host_vmcs_ptrs[V3_CONFIG_MAX_CPUS] = { [0 ... V3_CONFIG_MAX_CPUS - 1] = 0};
  51
  52 extern int v3_vmx_launch(struct v3_gprs * vm_regs, struct guest_info * info, struct v3_ctrl_regs * ctrl_regs);
  53 extern int v3_vmx_resume(struct v3_gprs * vm_regs, struct guest_info * info, struct v3_ctrl_regs * ctrl_regs);
  54
  55 static int inline check_vmcs_write(vmcs_field_t field, addr_t val) {
  56     int ret = 0;
  57
  58     ret = vmcs_write(field, val);
  59
  60     if (ret != VMX_SUCCESS) {
  61         PrintError("VMWRITE error on %s!: %d\n", v3_vmcs_field_to_str(field), ret);
  62         return 1;
  63     }
  64
  65     return 0;
  66 }
  67
  68 static int inline check_vmcs_read(vmcs_field_t field, void * val) {
  69     int ret = 0;
  70
  71     ret = vmcs_read(field, val);
  72
  73     if (ret != VMX_SUCCESS) {
  74         PrintError("VMREAD error on %s!: %d\n", v3_vmcs_field_to_str(field), ret);
  75     }
  76
  77     return ret;
  78 }
  79
  80
  81
  82
  83 static addr_t allocate_vmcs() {
  84     struct vmcs_data * vmcs_page = NULL;
  85
  86     PrintDebug("Allocating page\n");
  87
  88     vmcs_page = (struct vmcs_data *)V3_VAddr(V3_AllocPages(1));
  89     memset(vmcs_page, 0, 4096);
  90
  91     vmcs_page->revision = hw_info.basic_info.revision;
  92     PrintDebug("VMX Revision: 0x%x\n", vmcs_page->revision);
  93
  94     return (addr_t)V3_PAddr((void *)vmcs_page);
  95 }
  96
  97
  98
  99
 100 static int init_vmcs_bios(struct guest_info * core, struct vmx_data * vmx_state) {
 101     int vmx_ret = 0;
 102
 103     // disable global interrupts for vm state initialization
 104     v3_disable_ints();
 105
 106     PrintDebug("Loading VMCS\n");
 107     vmx_ret = vmcs_load(vmx_state->vmcs_ptr_phys);
 108     vmx_state->state = VMX_UNLAUNCHED;
 109
 110     if (vmx_ret != VMX_SUCCESS) {
 111         PrintError("VMPTRLD failed\n");
 112         return -1;
 113     }
 114
 115
 116     /*** Setup default state from HW ***/
 117
 118     vmx_state->pin_ctrls.value = hw_info.pin_ctrls.def_val;
 119     vmx_state->pri_proc_ctrls.value = hw_info.proc_ctrls.def_val;
 120     vmx_state->exit_ctrls.value = hw_info.exit_ctrls.def_val;
 121     vmx_state->entry_ctrls.value = hw_info.entry_ctrls.def_val;
 122     vmx_state->sec_proc_ctrls.value = hw_info.sec_proc_ctrls.def_val;
 123
 124     /* Print Control MSRs */
 125     PrintDebug("CR0 MSR: %p\n", (void *)(addr_t)hw_info.cr0.value);
 126     PrintDebug("CR4 MSR: %p\n", (void *)(addr_t)hw_info.cr4.value);
 127
 128
 129
 130     /******* Setup Host State **********/
 131
 132     /* Cache GDTR, IDTR, and TR in host struct */
 133     addr_t gdtr_base;
 134     struct {
 135         uint16_t selector;
 136         addr_t   base;
 137     } __attribute__((packed)) tmp_seg;
 138
 139
 140     __asm__ __volatile__(
 141                          "sgdt (%0);"
 142                          :
 143                          : "q"(&tmp_seg)
 144                          : "memory"
 145                          );
 146     gdtr_base = tmp_seg.base;
 147     vmx_state->host_state.gdtr.base = gdtr_base;
 148
 149     __asm__ __volatile__(
 150                          "sidt (%0);"
 151                          :
 152                          : "q"(&tmp_seg)
 153                          : "memory"
 154                          );
 155     vmx_state->host_state.idtr.base = tmp_seg.base;
 156
 157     __asm__ __volatile__(
 158                          "str (%0);"
 159                          :
 160                          : "q"(&tmp_seg)
 161                          : "memory"
 162                          );
 163     vmx_state->host_state.tr.selector = tmp_seg.selector;
 164
 165     /* The GDTR *index* is bits 3-15 of the selector. */
 166     struct tss_descriptor * desc = NULL;
 167     desc = (struct tss_descriptor *)(gdtr_base + (8 * (tmp_seg.selector >> 3)));
 168
 169     tmp_seg.base = ((desc->base1) |
 170                     (desc->base2 << 16) |
 171                     (desc->base3 << 24) |
 172 #ifdef __V3_64BIT__
 173                     ((uint64_t)desc->base4 << 32)
 174 #else
 175                     (0)
 176 #endif
 177                     );
 178
 179     vmx_state->host_state.tr.base = tmp_seg.base;
 180
 181
 182     /********** Setup VMX Control Fields ***********/
 183
 184     /* Add external interrupts, NMI exiting, and virtual NMI */
 185     vmx_state->pin_ctrls.nmi_exit = 1;
 186     vmx_state->pin_ctrls.ext_int_exit = 1;
 187
 188
 189     vmx_state->pri_proc_ctrls.hlt_exit = 1;
 190
 191
 192     vmx_state->pri_proc_ctrls.pause_exit = 0;
 193     vmx_state->pri_proc_ctrls.tsc_offset = 1;
 194 #ifdef V3_CONFIG_TIME_VIRTUALIZE_TSC
 195     vmx_state->pri_proc_ctrls.rdtsc_exit = 1;
 196 #endif
 197
 198     /* Setup IO map */
 199     vmx_state->pri_proc_ctrls.use_io_bitmap = 1;
 200     vmx_ret |= check_vmcs_write(VMCS_IO_BITMAP_A_ADDR, (addr_t)V3_PAddr(core->vm_info->io_map.arch_data));
 201     vmx_ret |= check_vmcs_write(VMCS_IO_BITMAP_B_ADDR,
 202             (addr_t)V3_PAddr(core->vm_info->io_map.arch_data) + PAGE_SIZE_4KB);
 203
 204
 205     vmx_state->pri_proc_ctrls.use_msr_bitmap = 1;
 206     vmx_ret |= check_vmcs_write(VMCS_MSR_BITMAP, (addr_t)V3_PAddr(core->vm_info->msr_map.arch_data));
 207
 208
 209
 210
 211
 212
 213
 214 #ifdef __V3_64BIT__
 215     // Ensure host runs in 64-bit mode at each VM EXIT
 216     vmx_state->exit_ctrls.host_64_on = 1;
 217 #endif
 218
 219     // Hook all accesses to EFER register
 220     v3_hook_msr(core->vm_info, EFER_MSR,
 221                 &v3_handle_efer_read,
 222                 &v3_handle_efer_write,
 223                 core);
 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     // Cause VM_EXIT whenever CR4.VMXE or CR4.PAE bits are written
 233     vmx_ret |= check_vmcs_write(VMCS_CR4_MASK, CR4_VMXE | CR4_PAE);
 234
 235
 236     /* Setup paging */
 237     if (core->shdw_pg_mode == SHADOW_PAGING) {
 238         PrintDebug("Creating initial shadow page table\n");
 239
 240         if (v3_init_passthrough_pts(core) == -1) {
 241             PrintError("Could not initialize passthrough page tables\n");
 242             return -1;
 243         }
 244
 245 #define CR0_PE 0x00000001
 246 #define CR0_PG 0x80000000
 247 #define CR0_WP 0x00010000 // To ensure mem hooks work
 248         vmx_ret |= check_vmcs_write(VMCS_CR0_MASK, (CR0_PE | CR0_PG | CR0_WP));
 249
 250         core->ctrl_regs.cr3 = core->direct_map_pt;
 251
 252         // vmx_state->pinbased_ctrls |= NMI_EXIT;
 253
 254         /* Add CR exits */
 255         vmx_state->pri_proc_ctrls.cr3_ld_exit = 1;
 256         vmx_state->pri_proc_ctrls.cr3_str_exit = 1;
 257
 258         vmx_state->pri_proc_ctrls.invlpg_exit = 1;
 259
 260         /* Add page fault exits */
 261         vmx_state->excp_bmap.pf = 1;
 262
 263         // Setup VMX Assist
 264         v3_vmxassist_init(core, vmx_state);
 265
 266     } else if ((core->shdw_pg_mode == NESTED_PAGING) &&
 267                (v3_cpu_types[core->pcpu_id] == V3_VMX_EPT_CPU)) {
 268
 269 #define CR0_PE 0x00000001
 270 #define CR0_PG 0x80000000
 271 #define CR0_WP 0x00010000 // To ensure mem hooks work
 272         vmx_ret |= check_vmcs_write(VMCS_CR0_MASK, (CR0_PE | CR0_PG | CR0_WP));
 273
 274         // vmx_state->pinbased_ctrls |= NMI_EXIT;
 275
 276         /* Disable CR exits */
 277         vmx_state->pri_proc_ctrls.cr3_ld_exit = 0;
 278         vmx_state->pri_proc_ctrls.cr3_str_exit = 0;
 279
 280         vmx_state->pri_proc_ctrls.invlpg_exit = 0;
 281
 282         /* Add page fault exits */
 283         //      vmx_state->excp_bmap.pf = 1; // This should never happen..., enabled to catch bugs
 284
 285         // Setup VMX Assist
 286         v3_vmxassist_init(core, vmx_state);
 287
 288         /* Enable EPT */
 289         vmx_state->pri_proc_ctrls.sec_ctrls = 1; // Enable secondary proc controls
 290         vmx_state->sec_proc_ctrls.enable_ept = 1; // enable EPT paging
 291
 292
 293
 294         if (v3_init_ept(core, &hw_info) == -1) {
 295             PrintError("Error initializing EPT\n");
 296             return -1;
 297         }
 298
 299     } else if ((core->shdw_pg_mode == NESTED_PAGING) &&
 300                (v3_cpu_types[core->pcpu_id] == V3_VMX_EPT_UG_CPU)) {
 301         int i = 0;
 302         // For now we will assume that unrestricted guest mode is assured w/ EPT
 303
 304
 305         core->vm_regs.rsp = 0x00;
 306         core->rip = 0xfff0;
 307         core->vm_regs.rdx = 0x00000f00;
 308         core->ctrl_regs.rflags = 0x00000002; // The reserved bit is always 1
 309         core->ctrl_regs.cr0 = 0x60010010; // Set the WP flag so the memory hooks work in real-mode
 310
 311
 312         core->segments.cs.selector = 0xf000;
 313         core->segments.cs.limit = 0xffff;
 314         core->segments.cs.base = 0x0000000f0000LL;
 315
 316         // (raw attributes = 0xf3)
 317         core->segments.cs.type = 0xb;
 318         core->segments.cs.system = 0x1;
 319         core->segments.cs.dpl = 0x0;
 320         core->segments.cs.present = 1;
 321
 322
 323
 324         struct v3_segment * segregs [] = {&(core->segments.ss), &(core->segments.ds),
 325                                           &(core->segments.es), &(core->segments.fs),
 326                                           &(core->segments.gs), NULL};
 327
 328         for ( i = 0; segregs[i] != NULL; i++) {
 329             struct v3_segment * seg = segregs[i];
 330
 331             seg->selector = 0x0000;
 332             //    seg->base = seg->selector << 4;
 333             seg->base = 0x00000000;
 334             seg->limit = 0xffff;
 335
 336
 337             seg->type = 0x3;
 338             seg->system = 0x1;
 339             seg->dpl = 0x0;
 340             seg->present = 1;
 341             //    seg->granularity = 1;
 342
 343         }
 344
 345
 346         core->segments.gdtr.limit = 0x0000ffff;
 347         core->segments.gdtr.base = 0x0000000000000000LL;
 348
 349         core->segments.idtr.limit = 0x0000ffff;
 350         core->segments.idtr.base = 0x0000000000000000LL;
 351
 352         core->segments.ldtr.selector = 0x0000;
 353         core->segments.ldtr.limit = 0x0000ffff;
 354         core->segments.ldtr.base = 0x0000000000000000LL;
 355         core->segments.ldtr.type = 2;
 356         core->segments.ldtr.present = 1;
 357
 358         core->segments.tr.selector = 0x0000;
 359         core->segments.tr.limit = 0x0000ffff;
 360         core->segments.tr.base = 0x0000000000000000LL;
 361         core->segments.tr.type = 0xb;
 362         core->segments.tr.present = 1;
 363
 364         //      core->dbg_regs.dr6 = 0x00000000ffff0ff0LL;
 365         core->dbg_regs.dr7 = 0x0000000000000400LL;
 366
 367         /* Enable EPT */
 368         vmx_state->pri_proc_ctrls.sec_ctrls = 1; // Enable secondary proc controls
 369         vmx_state->sec_proc_ctrls.enable_ept = 1; // enable EPT paging
 370         vmx_state->sec_proc_ctrls.unrstrct_guest = 1; // enable unrestricted guest operation
 371
 372
 373         /* Disable shadow paging stuff */
 374         vmx_state->pri_proc_ctrls.cr3_ld_exit = 0;
 375         vmx_state->pri_proc_ctrls.cr3_str_exit = 0;
 376
 377         vmx_state->pri_proc_ctrls.invlpg_exit = 0;
 378
 379
 380         if (v3_init_ept(core, &hw_info) == -1) {
 381             PrintError("Error initializing EPT\n");
 382             return -1;
 383         }
 384
 385     } else {
 386         PrintError("Invalid Virtual paging mode\n");
 387         return -1;
 388     }
 389
 390
 391     // hook vmx msrs
 392
 393     // Setup SYSCALL/SYSENTER MSRs in load/store area
 394
 395     // save STAR, LSTAR, FMASK, KERNEL_GS_BASE MSRs in MSR load/store area
 396     {
 397 #define IA32_STAR 0xc0000081
 398 #define IA32_LSTAR 0xc0000082
 399 #define IA32_FMASK 0xc0000084
 400 #define IA32_KERN_GS_BASE 0xc0000102
 401
 402 #define IA32_CSTAR 0xc0000083 // Compatibility mode STAR (ignored for now... hopefully its not that important...)
 403
 404         int msr_ret = 0;
 405
 406         struct vmcs_msr_entry * exit_store_msrs = NULL;
 407         struct vmcs_msr_entry * exit_load_msrs = NULL;
 408         struct vmcs_msr_entry * entry_load_msrs = NULL;;
 409         int max_msrs = (hw_info.misc_info.max_msr_cache_size + 1) * 4;
 410
 411         V3_Print("Setting up MSR load/store areas (max_msr_count=%d)\n", max_msrs);
 412
 413         if (max_msrs < 4) {
 414             PrintError("Max MSR cache size is too small (%d)\n", max_msrs);
 415             return -1;
 416         }
 417
 418         vmx_state->msr_area = V3_VAddr(V3_AllocPages(1));
 419
 420         if (vmx_state->msr_area == NULL) {
 421             PrintError("could not allocate msr load/store area\n");
 422             return -1;
 423         }
 424
 425         msr_ret |= check_vmcs_write(VMCS_EXIT_MSR_STORE_CNT, 4);
 426         msr_ret |= check_vmcs_write(VMCS_EXIT_MSR_LOAD_CNT, 4);
 427         msr_ret |= check_vmcs_write(VMCS_ENTRY_MSR_LOAD_CNT, 4);
 428
 429
 430         exit_store_msrs = (struct vmcs_msr_entry *)(vmx_state->msr_area);
 431         exit_load_msrs = (struct vmcs_msr_entry *)(vmx_state->msr_area + (sizeof(struct vmcs_msr_entry) * 4));
 432         entry_load_msrs = (struct vmcs_msr_entry *)(vmx_state->msr_area + (sizeof(struct vmcs_msr_entry) * 8));
 433
 434
 435         exit_store_msrs[0].index = IA32_STAR;
 436         exit_store_msrs[1].index = IA32_LSTAR;
 437         exit_store_msrs[2].index = IA32_FMASK;
 438         exit_store_msrs[3].index = IA32_KERN_GS_BASE;
 439
 440         memcpy(exit_store_msrs, exit_load_msrs, sizeof(struct vmcs_msr_entry) * 4);
 441         memcpy(exit_store_msrs, entry_load_msrs, sizeof(struct vmcs_msr_entry) * 4);
 442
 443
 444         v3_get_msr(IA32_STAR, &(exit_load_msrs[0].hi), &(exit_load_msrs[0].lo));
 445         v3_get_msr(IA32_LSTAR, &(exit_load_msrs[1].hi), &(exit_load_msrs[1].lo));
 446         v3_get_msr(IA32_FMASK, &(exit_load_msrs[2].hi), &(exit_load_msrs[2].lo));
 447         v3_get_msr(IA32_KERN_GS_BASE, &(exit_load_msrs[3].hi), &(exit_load_msrs[3].lo));
 448
 449         msr_ret |= check_vmcs_write(VMCS_EXIT_MSR_STORE_ADDR, (addr_t)V3_PAddr(exit_store_msrs));
 450         msr_ret |= check_vmcs_write(VMCS_EXIT_MSR_LOAD_ADDR, (addr_t)V3_PAddr(exit_load_msrs));
 451         msr_ret |= check_vmcs_write(VMCS_ENTRY_MSR_LOAD_ADDR, (addr_t)V3_PAddr(entry_load_msrs));
 452
 453     }
 454
 455     /* Sanity check ctrl/reg fields against hw_defaults */
 456
 457
 458
 459
 460     /*** Write all the info to the VMCS ***/
 461
 462     /*
 463     {
 464         // IS THIS NECESSARY???
 465 #define DEBUGCTL_MSR 0x1d9
 466         struct v3_msr tmp_msr;
 467         v3_get_msr(DEBUGCTL_MSR, &(tmp_msr.hi), &(tmp_msr.lo));
 468         vmx_ret |= check_vmcs_write(VMCS_GUEST_DBG_CTL, tmp_msr.value);
 469         core->dbg_regs.dr7 = 0x400;
 470     }
 471     */
 472
 473 #ifdef __V3_64BIT__
 474     vmx_ret |= check_vmcs_write(VMCS_LINK_PTR, (addr_t)0xffffffffffffffffULL);
 475 #else
 476     vmx_ret |= check_vmcs_write(VMCS_LINK_PTR, (addr_t)0xffffffffUL);
 477     vmx_ret |= check_vmcs_write(VMCS_LINK_PTR_HIGH, (addr_t)0xffffffffUL);
 478 #endif
 479
 480
 481
 482
 483     if (v3_update_vmcs_ctrl_fields(core)) {
 484         PrintError("Could not write control fields!\n");
 485         return -1;
 486     }
 487
 488     if (v3_update_vmcs_host_state(core)) {
 489         PrintError("Could not write host state\n");
 490         return -1;
 491     }
 492
 493     // reenable global interrupts for vm state initialization now
 494     // that the vm state is initialized. If another VM kicks us off,
 495     // it'll update our vmx state so that we know to reload ourself
 496     v3_enable_ints();
 497
 498     return 0;
 499 }
 500
 501 int v3_init_vmx_vmcs(struct guest_info * core, v3_vm_class_t vm_class) {
 502     struct vmx_data * vmx_state = NULL;
 503     int vmx_ret = 0;
 504
 505     vmx_state = (struct vmx_data *)V3_Malloc(sizeof(struct vmx_data));
 506     memset(vmx_state, 0, sizeof(struct vmx_data));
 507
 508     PrintDebug("vmx_data pointer: %p\n", (void *)vmx_state);
 509
 510     PrintDebug("Allocating VMCS\n");
 511     vmx_state->vmcs_ptr_phys = allocate_vmcs();
 512
 513     PrintDebug("VMCS pointer: %p\n", (void *)(vmx_state->vmcs_ptr_phys));
 514
 515     core->vmm_data = vmx_state;
 516     vmx_state->state = VMX_UNLAUNCHED;
 517
 518     PrintDebug("Initializing VMCS (addr=%p)\n", core->vmm_data);
 519
 520     // TODO: Fix vmcs fields so they're 32-bit
 521
 522     PrintDebug("Clearing VMCS: %p\n", (void *)vmx_state->vmcs_ptr_phys);
 523     vmx_ret = vmcs_clear(vmx_state->vmcs_ptr_phys);
 524
 525     if (vmx_ret != VMX_SUCCESS) {
 526         PrintError("VMCLEAR failed\n");
 527         return -1;
 528     }
 529
 530     if (vm_class == V3_PC_VM) {
 531         PrintDebug("Initializing VMCS\n");
 532         if (init_vmcs_bios(core, vmx_state) == -1) {
 533             PrintError("Error initializing VMCS to BIOS state\n");
 534             return -1;
 535         }
 536     } else {
 537         PrintError("Invalid VM Class\n");
 538         return -1;
 539     }
 540
 541     PrintDebug("Serializing VMCS: %p\n", (void *)vmx_state->vmcs_ptr_phys);
 542     vmx_ret = vmcs_clear(vmx_state->vmcs_ptr_phys);
 543
 544     return 0;
 545 }
 546
 547
 548 int v3_deinit_vmx_vmcs(struct guest_info * core) {
 549     struct vmx_data * vmx_state = core->vmm_data;
 550
 551     V3_FreePages((void *)(vmx_state->vmcs_ptr_phys), 1);
 552     V3_FreePages(V3_PAddr(vmx_state->msr_area), 1);
 553
 554     V3_Free(vmx_state);
 555
 556     return 0;
 557 }
 558
 559
 560 static int update_irq_exit_state(struct guest_info * info) {
 561     struct vmx_exit_idt_vec_info idt_vec_info;
 562
 563     check_vmcs_read(VMCS_IDT_VECTOR_INFO, &(idt_vec_info.value));
 564
 565     if ((info->intr_core_state.irq_started == 1) && (idt_vec_info.valid == 0)) {
 566 #ifdef V3_CONFIG_DEBUG_INTERRUPTS
 567         V3_Print("Calling v3_injecting_intr\n");
 568 #endif
 569         info->intr_core_state.irq_started = 0;
 570         v3_injecting_intr(info, info->intr_core_state.irq_vector, V3_EXTERNAL_IRQ);
 571     }
 572
 573     return 0;
 574 }
 575
 576 static int update_irq_entry_state(struct guest_info * info) {
 577     struct vmx_exit_idt_vec_info idt_vec_info;
 578     struct vmcs_interrupt_state intr_core_state;
 579     struct vmx_data * vmx_info = (struct vmx_data *)(info->vmm_data);
 580
 581     check_vmcs_read(VMCS_IDT_VECTOR_INFO, &(idt_vec_info.value));
 582     check_vmcs_read(VMCS_GUEST_INT_STATE, &(intr_core_state));
 583
 584     /* Check for pending exceptions to inject */
 585     if (v3_excp_pending(info)) {
 586         struct vmx_entry_int_info int_info;
 587         int_info.value = 0;
 588
 589         // In VMX, almost every exception is hardware
 590         // Software exceptions are pretty much only for breakpoint or overflow
 591         int_info.type = 3;
 592         int_info.vector = v3_get_excp_number(info);
 593
 594         if (info->excp_state.excp_error_code_valid) {
 595             check_vmcs_write(VMCS_ENTRY_EXCP_ERR, info->excp_state.excp_error_code);
 596             int_info.error_code = 1;
 597
 598 #ifdef V3_CONFIG_DEBUG_INTERRUPTS
 599             V3_Print("Injecting exception %d with error code %x\n",
 600                     int_info.vector, info->excp_state.excp_error_code);
 601 #endif
 602         }
 603
 604         int_info.valid = 1;
 605 #ifdef V3_CONFIG_DEBUG_INTERRUPTS
 606         V3_Print("Injecting exception %d (EIP=%p)\n", int_info.vector, (void *)(addr_t)info->rip);
 607 #endif
 608         check_vmcs_write(VMCS_ENTRY_INT_INFO, int_info.value);
 609
 610         v3_injecting_excp(info, int_info.vector);
 611
 612     } else if ((((struct rflags *)&(info->ctrl_regs.rflags))->intr == 1) &&
 613                (intr_core_state.val == 0)) {
 614
 615         if ((info->intr_core_state.irq_started == 1) && (idt_vec_info.valid == 1)) {
 616
 617 #ifdef V3_CONFIG_DEBUG_INTERRUPTS
 618             V3_Print("IRQ pending from previous injection\n");
 619 #endif
 620
 621             // Copy the IDT vectoring info over to reinject the old interrupt
 622             if (idt_vec_info.error_code == 1) {
 623                 uint32_t err_code = 0;
 624
 625                 check_vmcs_read(VMCS_IDT_VECTOR_ERR, &err_code);
 626                 check_vmcs_write(VMCS_ENTRY_EXCP_ERR, err_code);
 627             }
 628
 629             idt_vec_info.undef = 0;
 630             check_vmcs_write(VMCS_ENTRY_INT_INFO, idt_vec_info.value);
 631
 632         } else {
 633             struct vmx_entry_int_info ent_int;
 634             ent_int.value = 0;
 635
 636             switch (v3_intr_pending(info)) {
 637                 case V3_EXTERNAL_IRQ: {
 638                     info->intr_core_state.irq_vector = v3_get_intr(info);
 639                     ent_int.vector = info->intr_core_state.irq_vector;
 640                     ent_int.type = 0;
 641                     ent_int.error_code = 0;
 642                     ent_int.valid = 1;
 643
 644 #ifdef V3_CONFIG_DEBUG_INTERRUPTS
 645                     V3_Print("Injecting Interrupt %d at exit %u(EIP=%p)\n",
 646                                info->intr_core_state.irq_vector,
 647                                (uint32_t)info->num_exits,
 648                                (void *)(addr_t)info->rip);
 649 #endif
 650
 651                     check_vmcs_write(VMCS_ENTRY_INT_INFO, ent_int.value);
 652                     info->intr_core_state.irq_started = 1;
 653
 654                     break;
 655                 }
 656                 case V3_NMI:
 657                     PrintDebug("Injecting NMI\n");
 658
 659                     ent_int.type = 2;
 660                     ent_int.vector = 2;
 661                     ent_int.valid = 1;
 662                     check_vmcs_write(VMCS_ENTRY_INT_INFO, ent_int.value);
 663
 664                     break;
 665                 case V3_SOFTWARE_INTR:
 666                     PrintDebug("Injecting software interrupt\n");
 667                     ent_int.type = 4;
 668
 669                     ent_int.valid = 1;
 670                     check_vmcs_write(VMCS_ENTRY_INT_INFO, ent_int.value);
 671
 672                     break;
 673                 case V3_VIRTUAL_IRQ:
 674                     // Not sure what to do here, Intel doesn't have virtual IRQs
 675                     // May be the same as external interrupts/IRQs
 676
 677                     break;
 678                 case V3_INVALID_INTR:
 679                 default:
 680                     break;
 681             }
 682         }
 683     } else if ((v3_intr_pending(info)) && (vmx_info->pri_proc_ctrls.int_wndw_exit == 0)) {
 684         // Enable INTR window exiting so we know when IF=1
 685         uint32_t instr_len;
 686
 687         check_vmcs_read(VMCS_EXIT_INSTR_LEN, &instr_len);
 688
 689 #ifdef V3_CONFIG_DEBUG_INTERRUPTS
 690         V3_Print("Enabling Interrupt-Window exiting: %d\n", instr_len);
 691 #endif
 692
 693         vmx_info->pri_proc_ctrls.int_wndw_exit = 1;
 694         check_vmcs_write(VMCS_PROC_CTRLS, vmx_info->pri_proc_ctrls.value);
 695     }
 696
 697
 698     return 0;
 699 }
 700
 701
 702
 703 static struct vmx_exit_info exit_log[10];
 704
 705 static void print_exit_log(struct guest_info * info) {
 706     int cnt = info->num_exits % 10;
 707     int i = 0;
 708
 709
 710     V3_Print("\nExit Log (%d total exits):\n", (uint32_t)info->num_exits);
 711
 712     for (i = 0; i < 10; i++) {
 713         struct vmx_exit_info * tmp = &exit_log[cnt];
 714
 715         V3_Print("%d:\texit_reason = %p\n", i, (void *)(addr_t)tmp->exit_reason);
 716         V3_Print("\texit_qual = %p\n", (void *)tmp->exit_qual);
 717         V3_Print("\tint_info = %p\n", (void *)(addr_t)tmp->int_info);
 718         V3_Print("\tint_err = %p\n", (void *)(addr_t)tmp->int_err);
 719         V3_Print("\tinstr_info = %p\n", (void *)(addr_t)tmp->instr_info);
 720
 721         cnt--;
 722
 723         if (cnt == -1) {
 724             cnt = 9;
 725         }
 726
 727     }
 728
 729 }
 730
 731 /*
 732  * CAUTION and DANGER!!!
 733  *
 734  * The VMCS CANNOT(!!) be accessed outside of the cli/sti calls inside this function
 735  * When exectuing a symbiotic call, the VMCS WILL be overwritten, so any dependencies
 736  * on its contents will cause things to break. The contents at the time of the exit WILL
 737  * change before the exit handler is executed.
 738  */
 739 int v3_vmx_enter(struct guest_info * info) {
 740     int ret = 0;
 741     uint32_t tsc_offset_low, tsc_offset_high;
 742     struct vmx_exit_info exit_info;
 743     struct vmx_data * vmx_info = (struct vmx_data *)(info->vmm_data);
 744
 745     // Conditionally yield the CPU if the timeslice has expired
 746     v3_yield_cond(info);
 747
 748     // Perform any additional yielding needed for time adjustment
 749     v3_adjust_time(info);
 750
 751     // disable global interrupts for vm state transition
 752     v3_disable_ints();
 753
 754     // Update timer devices late after being in the VM so that as much
 755     // of hte time in the VM is accounted for as possible. Also do it before
 756     // updating IRQ entry state so that any interrupts the timers raise get
 757     // handled on the next VM entry. Must be done with interrupts disabled.
 758     v3_update_timers(info);
 759
 760     if (vmcs_store() != vmx_info->vmcs_ptr_phys) {
 761         vmcs_clear(vmx_info->vmcs_ptr_phys);
 762         vmcs_load(vmx_info->vmcs_ptr_phys);
 763         vmx_info->state = VMX_UNLAUNCHED;
 764     }
 765
 766     v3_vmx_restore_vmcs(info);
 767
 768
 769 #ifdef V3_CONFIG_SYMCALL
 770     if (info->sym_core_state.symcall_state.sym_call_active == 0) {
 771         update_irq_entry_state(info);
 772     }
 773 #else
 774     update_irq_entry_state(info);
 775 #endif
 776
 777     {
 778         addr_t guest_cr3;
 779         vmcs_read(VMCS_GUEST_CR3, &guest_cr3);
 780         vmcs_write(VMCS_GUEST_CR3, guest_cr3);
 781     }
 782
 783     // Perform last-minute time bookkeeping prior to entering the VM
 784     v3_time_enter_vm(info);
 785
 786     tsc_offset_high = (uint32_t)((v3_tsc_host_offset(&info->time_state) >> 32) & 0xffffffff);
 787     tsc_offset_low = (uint32_t)(v3_tsc_host_offset(&info->time_state) & 0xffffffff);
 788     check_vmcs_write(VMCS_TSC_OFFSET_HIGH, tsc_offset_high);
 789     check_vmcs_write(VMCS_TSC_OFFSET, tsc_offset_low);
 790
 791
 792     if (v3_update_vmcs_host_state(info)) {
 793         v3_enable_ints();
 794         PrintError("Could not write host state\n");
 795         return -1;
 796     }
 797
 798
 799     if (vmx_info->state == VMX_UNLAUNCHED) {
 800         vmx_info->state = VMX_LAUNCHED;
 801
 802         info->vm_info->run_state = VM_RUNNING;
 803         ret = v3_vmx_launch(&(info->vm_regs), info, &(info->ctrl_regs));
 804     } else {
 805         V3_ASSERT(vmx_info->state != VMX_UNLAUNCHED);
 806         ret = v3_vmx_resume(&(info->vm_regs), info, &(info->ctrl_regs));
 807     }
 808
 809     //  PrintDebug("VMX Exit: ret=%d\n", ret);
 810
 811     if (ret != VMX_SUCCESS) {
 812         uint32_t error = 0;
 813
 814         vmcs_read(VMCS_INSTR_ERR, &error);
 815
 816         v3_enable_ints();
 817
 818         PrintError("VMENTRY Error: %d\n", error);
 819         return -1;
 820     }
 821
 822     // Immediate exit from VM time bookkeeping
 823     v3_time_exit_vm(info);
 824
 825     info->num_exits++;
 826
 827     /* Update guest state */
 828     v3_vmx_save_vmcs(info);
 829
 830     // info->cpl = info->segments.cs.selector & 0x3;
 831
 832     info->mem_mode = v3_get_vm_mem_mode(info);
 833     info->cpu_mode = v3_get_vm_cpu_mode(info);
 834
 835
 836     check_vmcs_read(VMCS_EXIT_INSTR_LEN, &(exit_info.instr_len));
 837     check_vmcs_read(VMCS_EXIT_INSTR_INFO, &(exit_info.instr_info));
 838     check_vmcs_read(VMCS_EXIT_REASON, &(exit_info.exit_reason));
 839     check_vmcs_read(VMCS_EXIT_QUAL, &(exit_info.exit_qual));
 840     check_vmcs_read(VMCS_EXIT_INT_INFO, &(exit_info.int_info));
 841     check_vmcs_read(VMCS_EXIT_INT_ERR, &(exit_info.int_err));
 842     check_vmcs_read(VMCS_GUEST_LINEAR_ADDR, &(exit_info.guest_linear_addr));
 843
 844     if (info->shdw_pg_mode == NESTED_PAGING) {
 845         check_vmcs_read(VMCS_GUEST_PHYS_ADDR, &(exit_info.ept_fault_addr));
 846     }
 847
 848     //PrintDebug("VMX Exit taken, id-qual: %u-%lu\n", exit_info.exit_reason, exit_info.exit_qual);
 849
 850     exit_log[info->num_exits % 10] = exit_info;
 851
 852 #ifdef V3_CONFIG_SYMCALL
 853     if (info->sym_core_state.symcall_state.sym_call_active == 0) {
 854         update_irq_exit_state(info);
 855     }
 856 #else
 857     update_irq_exit_state(info);
 858 #endif
 859
 860     if (exit_info.exit_reason == VMEXIT_INTR_WINDOW) {
 861         // This is a special case whose only job is to inject an interrupt
 862         vmcs_read(VMCS_PROC_CTRLS, &(vmx_info->pri_proc_ctrls.value));
 863         vmx_info->pri_proc_ctrls.int_wndw_exit = 0;
 864         vmcs_write(VMCS_PROC_CTRLS, vmx_info->pri_proc_ctrls.value);
 865
 866 #ifdef V3_CONFIG_DEBUG_INTERRUPTS
 867        V3_Print("Interrupts available again! (RIP=%llx)\n", info->rip);
 868 #endif
 869     }
 870
 871     // reenable global interrupts after vm exit
 872     v3_enable_ints();
 873
 874     // Conditionally yield the CPU if the timeslice has expired
 875     v3_yield_cond(info);
 876
 877     if (v3_handle_vmx_exit(info, &exit_info) == -1) {
 878         PrintError("Error in VMX exit handler\n");
 879         return -1;
 880     }
 881
 882     return 0;
 883 }
 884
 885
 886 int v3_start_vmx_guest(struct guest_info * info) {
 887
 888     PrintDebug("Starting VMX core %u\n", info->vcpu_id);
 889
 890     if (info->vcpu_id == 0) {
 891         info->core_run_state = CORE_RUNNING;
 892         info->vm_info->run_state = VM_RUNNING;
 893     } else {
 894
 895         PrintDebug("VMX core %u: Waiting for core initialization\n", info->vcpu_id);
 896
 897         while (info->core_run_state == CORE_STOPPED) {
 898             v3_yield(info);
 899             //PrintDebug("VMX core %u: still waiting for INIT\n",info->vcpu_id);
 900         }
 901
 902         PrintDebug("VMX core %u initialized\n", info->vcpu_id);
 903     }
 904
 905
 906     PrintDebug("VMX core %u: I am starting at CS=0x%x (base=0x%p, limit=0x%x),  RIP=0x%p\n",
 907                info->vcpu_id, info->segments.cs.selector, (void *)(info->segments.cs.base),
 908                info->segments.cs.limit, (void *)(info->rip));
 909
 910
 911     PrintDebug("VMX core %u: Launching VMX VM on logical core %u\n", info->vcpu_id, info->pcpu_id);
 912
 913     v3_start_time(info);
 914
 915     while (1) {
 916
 917         if (info->vm_info->run_state == VM_STOPPED) {
 918             info->core_run_state = CORE_STOPPED;
 919             break;
 920         }
 921
 922         if (v3_vmx_enter(info) == -1) {
 923             v3_print_vmcs();
 924             print_exit_log(info);
 925             return -1;
 926         }
 927
 928
 929
 930         if (info->vm_info->run_state == VM_STOPPED) {
 931             info->core_run_state = CORE_STOPPED;
 932             break;
 933         }
 934 /*
 935         if ((info->num_exits % 5000) == 0) {
 936             V3_Print("VMX Exit number %d\n", (uint32_t)info->num_exits);
 937         }
 938 */
 939
 940     }
 941
 942     return 0;
 943 }
 944
 945
 946
 947
 948 #define VMX_FEATURE_CONTROL_MSR     0x0000003a
 949 #define CPUID_VMX_FEATURES 0x00000005  /* LOCK and VMXON */
 950 #define CPUID_1_ECX_VTXFLAG 0x00000020
 951
 952 int v3_is_vmx_capable() {
 953     v3_msr_t feature_msr;
 954     uint32_t eax = 0, ebx = 0, ecx = 0, edx = 0;
 955
 956     v3_cpuid(0x1, &eax, &ebx, &ecx, &edx);
 957
 958     PrintDebug("ECX: 0x%x\n", ecx);
 959
 960     if (ecx & CPUID_1_ECX_VTXFLAG) {
 961         v3_get_msr(VMX_FEATURE_CONTROL_MSR, &(feature_msr.hi), &(feature_msr.lo));
 962
 963         PrintDebug("MSRREGlow: 0x%.8x\n", feature_msr.lo);
 964
 965         if ((feature_msr.lo & CPUID_VMX_FEATURES) != CPUID_VMX_FEATURES) {
 966             PrintDebug("VMX is locked -- enable in the BIOS\n");
 967             return 0;
 968         }
 969
 970     } else {
 971         PrintDebug("VMX not supported on this cpu\n");
 972         return 0;
 973     }
 974
 975     return 1;
 976 }
 977
 978
 979 int v3_reset_vmx_vm_core(struct guest_info * core, addr_t rip) {
 980     // init vmcs bios
 981
 982     if ((core->shdw_pg_mode == NESTED_PAGING) &&
 983         (v3_cpu_types[core->pcpu_id] == V3_VMX_EPT_UG_CPU)) {
 984         // easy
 985         core->rip = 0;
 986         core->segments.cs.selector = rip << 8;
 987         core->segments.cs.limit = 0xffff;
 988         core->segments.cs.base = rip << 12;
 989     } else {
 990         core->vm_regs.rdx = core->vcpu_id;
 991         core->vm_regs.rbx = rip;
 992     }
 993
 994     return 0;
 995 }
 996
 997
 998
 999 void v3_init_vmx_cpu(int cpu_id) {
1000     addr_t vmx_on_region = 0;
1001
1002     if (cpu_id == 0) {
1003         if (v3_init_vmx_hw(&hw_info) == -1) {
1004             PrintError("Could not initialize VMX hardware features on cpu %d\n", cpu_id);
1005             return;
1006         }
1007     }
1008
1009     enable_vmx();
1010
1011
1012     // Setup VMXON Region
1013     vmx_on_region = allocate_vmcs();
1014
1015
1016     if (vmx_on(vmx_on_region) == VMX_SUCCESS) {
1017         V3_Print("VMX Enabled\n");
1018         host_vmcs_ptrs[cpu_id] = vmx_on_region;
1019     } else {
1020         V3_Print("VMX already enabled\n");
1021         V3_FreePages((void *)vmx_on_region, 1);
1022     }
1023
1024     PrintDebug("VMXON pointer: 0x%p\n", (void *)host_vmcs_ptrs[cpu_id]);
1025
1026     {
1027         struct vmx_sec_proc_ctrls sec_proc_ctrls;
1028         sec_proc_ctrls.value = v3_vmx_get_ctrl_features(&(hw_info.sec_proc_ctrls));
1029
1030         if (sec_proc_ctrls.enable_ept == 0) {
1031             V3_Print("VMX EPT (Nested) Paging not supported\n");
1032             v3_cpu_types[cpu_id] = V3_VMX_CPU;
1033         } else if (sec_proc_ctrls.unrstrct_guest == 0) {
1034             V3_Print("VMX EPT (Nested) Paging supported\n");
1035             v3_cpu_types[cpu_id] = V3_VMX_EPT_CPU;
1036         } else {
1037             V3_Print("VMX EPT (Nested) Paging + Unrestricted guest supported\n");
1038             v3_cpu_types[cpu_id] = V3_VMX_EPT_UG_CPU;
1039         }
1040     }
1041 }
1042
1043
1044 void v3_deinit_vmx_cpu(int cpu_id) {
1045     extern v3_cpu_arch_t v3_cpu_types[];
1046     v3_cpu_types[cpu_id] = V3_INVALID_CPU;
1047
1048     if (host_vmcs_ptrs[cpu_id] != 0) {
1049         V3_Print("Disabling VMX\n");
1050
1051         if (vmx_off() != VMX_SUCCESS) {
1052             PrintError("Error executing VMXOFF\n");
1053         }
1054
1055         V3_FreePages((void *)host_vmcs_ptrs[cpu_id], 1);
1056
1057         host_vmcs_ptrs[cpu_id] = 0;
1058     }
1059 }