palacios/src/palacios/vmx.c

   1 /*
   2  * This file is part of the Palacios Virtual Machine Monitor developed
   3  * by the V3VEE Project with funding from the United States National
   4  * Science Foundation and the Department of Energy.
   5  *
   6  * The V3VEE Project is a joint project between Northwestern University
   7  * and the University of New Mexico.  You can find out more at
   8  * http://www.v3vee.org
   9  *
  10  * Copyright (c) 2011, Jack Lange <jarusl@cs.northwestern.edu>
  11  * Copyright (c) 2011, The V3VEE Project <http://www.v3vee.org>
  12  * All rights reserved.
  13  *
  14  * Author: Jack Lange <jarusl@cs.northwestern.edu>
  15  *
  16  * This is free software.  You are permitted to use,
  17  * redistribute, and modify it as specified in the file "V3VEE_LICENSE".
  18  */
  19
  20
  21 #include <palacios/vmx.h>
  22 #include <palacios/vmm.h>
  23 #include <palacios/vmx_handler.h>
  24 #include <palacios/vmcs.h>
  25 #include <palacios/vmx_lowlevel.h>
  26 #include <palacios/vmm_lowlevel.h>
  27 #include <palacios/vmm_ctrl_regs.h>
  28 #include <palacios/vmm_config.h>
  29 #include <palacios/vmm_time.h>
  30 #include <palacios/vm_guest_mem.h>
  31 #include <palacios/vmm_direct_paging.h>
  32 #include <palacios/vmx_io.h>
  33 #include <palacios/vmx_msr.h>
  34 #include <palacios/vmm_decoder.h>
  35 #include <palacios/vmm_barrier.h>
  36 #include <palacios/vmm_timeout.h>
  37 #include <palacios/vmm_debug.h>
  38
  39 #ifdef V3_CONFIG_CHECKPOINT
  40 #include <palacios/vmm_checkpoint.h>
  41 #endif
  42
  43 #include <palacios/vmx_ept.h>
  44 #include <palacios/vmx_assist.h>
  45 #include <palacios/vmx_hw_info.h>
  46
  47 #ifdef V3_CONFIG_MEM_TRACK
  48 #include <palacios/vmm_mem_track.h>
  49 #endif
  50
  51 #ifndef V3_CONFIG_DEBUG_VMX
  52 #undef PrintDebug
  53 #define PrintDebug(fmt, args...)
  54 #endif
  55
  56
  57 /* These fields contain the hardware feature sets supported by the local CPU */
  58 static struct vmx_hw_info hw_info;
  59
  60 extern v3_cpu_arch_t v3_mach_type;
  61
  62 static addr_t host_vmcs_ptrs[V3_CONFIG_MAX_CPUS] = { [0 ... V3_CONFIG_MAX_CPUS - 1] = 0};
  63
  64 extern int v3_vmx_launch(struct v3_gprs * vm_regs, struct guest_info * info, struct v3_ctrl_regs * ctrl_regs);
  65 extern int v3_vmx_resume(struct v3_gprs * vm_regs, struct guest_info * info, struct v3_ctrl_regs * ctrl_regs);
  66
  67 static int inline check_vmcs_write(vmcs_field_t field, addr_t val) {
  68     int ret = 0;
  69
  70     ret = vmcs_write(field, val);
  71
  72     if (ret != VMX_SUCCESS) {
  73         PrintError(VM_NONE, VCORE_NONE, "VMWRITE error on %s!: %d\n", v3_vmcs_field_to_str(field), ret);
  74         return 1;
  75     }
  76
  77
  78
  79
  80     return 0;
  81 }
  82
  83 static int inline check_vmcs_read(vmcs_field_t field, void * val) {
  84     int ret = 0;
  85
  86     ret = vmcs_read(field, val);
  87
  88     if (ret != VMX_SUCCESS) {
  89         PrintError(VM_NONE, VCORE_NONE, "VMREAD error on %s!: %d\n", v3_vmcs_field_to_str(field), ret);
  90     }
  91
  92     return ret;
  93 }
  94
  95
  96
  97
  98 static addr_t allocate_vmcs() {
  99     void *temp;
 100     struct vmcs_data * vmcs_page = NULL;
 101
 102     PrintDebug(VM_NONE, VCORE_NONE, "Allocating page\n");
 103
 104     temp = V3_AllocPages(1); // need not be shadow-safe, not exposed to guest
 105     if (!temp) {
 106         PrintError(VM_NONE, VCORE_NONE, "Cannot allocate VMCS\n");
 107         return -1;
 108     }
 109     vmcs_page = (struct vmcs_data *)V3_VAddr(temp);
 110     memset(vmcs_page, 0, 4096);
 111
 112     vmcs_page->revision = hw_info.basic_info.revision;
 113     PrintDebug(VM_NONE, VCORE_NONE, "VMX Revision: 0x%x\n", vmcs_page->revision);
 114
 115     return (addr_t)V3_PAddr((void *)vmcs_page);
 116 }
 117
 118
 119 #if 0
 120 static int debug_efer_read(struct guest_info * core, uint_t msr, struct v3_msr * src, void * priv_data) {
 121     struct v3_msr * efer = (struct v3_msr *)&(core->ctrl_regs.efer);
 122     V3_Print(core->vm_info, core, "\n\nEFER READ (val = %p)\n", (void *)efer->value);
 123
 124     v3_print_guest_state(core);
 125     v3_print_vmcs();
 126
 127
 128     src->value = efer->value;
 129     return 0;
 130 }
 131
 132 static int debug_efer_write(struct guest_info * core, uint_t msr, struct v3_msr src, void * priv_data) {
 133     struct v3_msr * efer = (struct v3_msr *)&(core->ctrl_regs.efer);
 134     V3_Print(core->vm_info, core, "\n\nEFER WRITE (old_val = %p) (new_val = %p)\n", (void *)efer->value, (void *)src.value);
 135
 136     v3_print_guest_state(core);
 137     v3_print_vmcs();
 138
 139     efer->value = src.value;
 140
 141     return 0;
 142 }
 143 #endif
 144
 145
 146 static int init_vmcs_bios(struct guest_info * core, struct vmx_data * vmx_state) {
 147     int vmx_ret = 0;
 148
 149     /* Get Available features */
 150     struct vmx_pin_ctrls avail_pin_ctrls;
 151     avail_pin_ctrls.value = v3_vmx_get_ctrl_features(&(hw_info.pin_ctrls));
 152     /* ** */
 153
 154
 155     // disable global interrupts for vm state initialization
 156     v3_disable_ints();
 157
 158     PrintDebug(core->vm_info, core, "Loading VMCS\n");
 159     vmx_ret = vmcs_load(vmx_state->vmcs_ptr_phys);
 160     vmx_state->state = VMX_UNLAUNCHED;
 161
 162     if (vmx_ret != VMX_SUCCESS) {
 163         PrintError(core->vm_info, core, "VMPTRLD failed\n");
 164         return -1;
 165     }
 166
 167
 168     /*** Setup default state from HW ***/
 169
 170     vmx_state->pin_ctrls.value = hw_info.pin_ctrls.def_val;
 171     vmx_state->pri_proc_ctrls.value = hw_info.proc_ctrls.def_val;
 172     vmx_state->exit_ctrls.value = hw_info.exit_ctrls.def_val;
 173     vmx_state->entry_ctrls.value = hw_info.entry_ctrls.def_val;
 174     vmx_state->sec_proc_ctrls.value = hw_info.sec_proc_ctrls.def_val;
 175
 176     /* Print Control MSRs */
 177     V3_Print(core->vm_info, core, "CR0 MSR: req_val=%p, req_mask=%p\n", (void *)(addr_t)hw_info.cr0.req_val, (void *)(addr_t)hw_info.cr0.req_mask);
 178     V3_Print(core->vm_info, core, "CR4 MSR: req_val=%p, req_mask=%p\n", (void *)(addr_t)hw_info.cr4.req_val, (void *)(addr_t)hw_info.cr4.req_mask);
 179
 180
 181
 182     /******* Setup Host State **********/
 183
 184     /* Cache GDTR, IDTR, and TR in host struct */
 185
 186
 187     /********** Setup VMX Control Fields ***********/
 188
 189     /* Add external interrupts, NMI exiting, and virtual NMI */
 190     vmx_state->pin_ctrls.nmi_exit = 1;
 191     vmx_state->pin_ctrls.virt_nmi = 1;
 192     vmx_state->pin_ctrls.ext_int_exit = 1;
 193
 194
 195
 196     /* We enable the preemption timer by default to measure accurate guest time */
 197     if (avail_pin_ctrls.active_preempt_timer) {
 198         V3_Print(core->vm_info, core, "VMX Preemption Timer is available\n");
 199         vmx_state->pin_ctrls.active_preempt_timer = 1;
 200         vmx_state->exit_ctrls.save_preempt_timer = 1;
 201     }
 202
 203     // we want it to use this when halting
 204     vmx_state->pri_proc_ctrls.hlt_exit = 1;
 205
 206     // cpuid tells it that it does not have these instructions
 207     vmx_state->pri_proc_ctrls.monitor_exit = 1;
 208     vmx_state->pri_proc_ctrls.mwait_exit = 1;
 209
 210     // we don't need to handle a pause, although this is where
 211     // we could pull out of a spin lock acquire or schedule to find its partner
 212     vmx_state->pri_proc_ctrls.pause_exit = 0;
 213
 214     vmx_state->pri_proc_ctrls.tsc_offset = 1;
 215 #ifdef V3_CONFIG_TIME_VIRTUALIZE_TSC
 216     vmx_state->pri_proc_ctrls.rdtsc_exit = 1;
 217 #endif
 218
 219     /* Setup IO map */
 220     vmx_state->pri_proc_ctrls.use_io_bitmap = 1;
 221     vmx_ret |= check_vmcs_write(VMCS_IO_BITMAP_A_ADDR, (addr_t)V3_PAddr(core->vm_info->io_map.arch_data));
 222     vmx_ret |= check_vmcs_write(VMCS_IO_BITMAP_B_ADDR,
 223             (addr_t)V3_PAddr(core->vm_info->io_map.arch_data) + PAGE_SIZE_4KB);
 224
 225
 226     vmx_state->pri_proc_ctrls.use_msr_bitmap = 1;
 227     vmx_ret |= check_vmcs_write(VMCS_MSR_BITMAP, (addr_t)V3_PAddr(core->vm_info->msr_map.arch_data));
 228
 229
 230
 231 #ifdef __V3_64BIT__
 232     // Ensure host runs in 64-bit mode at each VM EXIT
 233     vmx_state->exit_ctrls.host_64_on = 1;
 234 #endif
 235
 236
 237
 238     // Restore host's EFER register on each VM EXIT
 239     vmx_state->exit_ctrls.ld_efer = 1;
 240
 241     // Save/restore guest's EFER register to/from VMCS on VM EXIT/ENTRY
 242     vmx_state->exit_ctrls.save_efer = 1;
 243     vmx_state->entry_ctrls.ld_efer  = 1;
 244
 245     vmx_state->exit_ctrls.save_pat = 1;
 246     vmx_state->exit_ctrls.ld_pat = 1;
 247     vmx_state->entry_ctrls.ld_pat = 1;
 248
 249     /* Temporary GPF trap */
 250     //  vmx_state->excp_bmap.gp = 1;
 251
 252     // Setup Guests initial PAT field
 253     vmx_ret |= check_vmcs_write(VMCS_GUEST_PAT, 0x0007040600070406LL);
 254
 255     // Capture CR8 mods so that we can keep the apic_tpr correct
 256     vmx_state->pri_proc_ctrls.cr8_ld_exit = 1;
 257     vmx_state->pri_proc_ctrls.cr8_str_exit = 1;
 258
 259
 260     /* Setup paging */
 261     if (core->shdw_pg_mode == SHADOW_PAGING) {
 262         PrintDebug(core->vm_info, core, "Creating initial shadow page table\n");
 263
 264         if (v3_init_passthrough_pts(core) == -1) {
 265             PrintError(core->vm_info, core, "Could not initialize passthrough page tables\n");
 266             return -1;
 267         }
 268
 269 #define CR0_PE 0x00000001
 270 #define CR0_PG 0x80000000
 271 #define CR0_WP 0x00010000 // To ensure mem hooks work
 272 #define CR0_NE 0x00000020
 273         vmx_ret |= check_vmcs_write(VMCS_CR0_MASK, (CR0_PE | CR0_PG | CR0_WP | CR0_NE));
 274
 275
 276         // Cause VM_EXIT whenever CR4.VMXE or CR4.PAE bits are written
 277         vmx_ret |= check_vmcs_write(VMCS_CR4_MASK, CR4_VMXE | CR4_PAE );
 278
 279         v3_activate_passthrough_pt(core);
 280
 281         // vmx_state->pinbased_ctrls |= NMI_EXIT;
 282
 283         /* Add CR exits */
 284         vmx_state->pri_proc_ctrls.cr3_ld_exit = 1;
 285         vmx_state->pri_proc_ctrls.cr3_str_exit = 1;
 286
 287         // Note that we intercept cr4.pae writes
 288         // and we have cr4 read-shadowed to the shadow pager's cr4
 289
 290         vmx_state->pri_proc_ctrls.invlpg_exit = 1;
 291
 292         /* Add page fault exits */
 293         vmx_state->excp_bmap.pf = 1;
 294
 295         // Setup VMX Assist
 296         v3_vmxassist_init(core, vmx_state);
 297
 298         // Hook all accesses to EFER register
 299         v3_hook_msr(core->vm_info, EFER_MSR,
 300                     &v3_handle_efer_read,
 301                     &v3_handle_efer_write,
 302                     core);
 303
 304     } else if ((core->shdw_pg_mode == NESTED_PAGING) &&
 305                (v3_mach_type == V3_VMX_EPT_CPU)) {
 306
 307 #define CR0_PE 0x00000001
 308 #define CR0_PG 0x80000000
 309 #define CR0_WP 0x00010000 // To ensure mem hooks work
 310 #define CR0_NE 0x00000020
 311         vmx_ret |= check_vmcs_write(VMCS_CR0_MASK, (CR0_PE | CR0_PG | CR0_WP | CR0_NE));
 312
 313         // vmx_state->pinbased_ctrls |= NMI_EXIT;
 314
 315         // Cause VM_EXIT whenever CR4.VMXE or CR4.PAE bits are written
 316         vmx_ret |= check_vmcs_write(VMCS_CR4_MASK, CR4_VMXE | CR4_PAE);
 317
 318         /* Disable CR exits */
 319         vmx_state->pri_proc_ctrls.cr3_ld_exit = 0;
 320         vmx_state->pri_proc_ctrls.cr3_str_exit = 0;
 321
 322         vmx_state->pri_proc_ctrls.invlpg_exit = 0;
 323
 324         /* Add page fault exits */
 325         //      vmx_state->excp_bmap.pf = 1; // This should never happen..., enabled to catch bugs
 326
 327         // Setup VMX Assist
 328         v3_vmxassist_init(core, vmx_state);
 329
 330         /* Enable EPT */
 331         vmx_state->pri_proc_ctrls.sec_ctrls = 1; // Enable secondary proc controls
 332         vmx_state->sec_proc_ctrls.enable_ept = 1; // enable EPT paging
 333
 334
 335
 336         if (v3_init_nested_paging_core(core, &hw_info) == -1) {
 337             PrintError(core->vm_info, core, "Error initializing EPT\n");
 338             return -1;
 339         }
 340
 341         // Hook all accesses to EFER register
 342         v3_hook_msr(core->vm_info, EFER_MSR, NULL, NULL, NULL);
 343
 344     } else if ((core->shdw_pg_mode == NESTED_PAGING) &&
 345                (v3_mach_type == V3_VMX_EPT_UG_CPU)) {
 346         int i = 0;
 347         // For now we will assume that unrestricted guest mode is assured w/ EPT
 348
 349
 350         core->vm_regs.rsp = 0x00;
 351         core->rip = 0xfff0;
 352         core->vm_regs.rdx = 0x00000f00;
 353         core->ctrl_regs.rflags = 0x00000002; // The reserved bit is always 1
 354         core->ctrl_regs.cr0 = 0x60010030;
 355         core->ctrl_regs.cr4 = 0x00002010; // Enable VMX and PSE flag
 356
 357
 358         core->segments.cs.selector = 0xf000;
 359         core->segments.cs.limit = 0xffff;
 360         core->segments.cs.base = 0x0000000f0000LL;
 361
 362         // (raw attributes = 0xf3)
 363         core->segments.cs.type = 0xb;
 364         core->segments.cs.system = 0x1;
 365         core->segments.cs.dpl = 0x0;
 366         core->segments.cs.present = 1;
 367
 368
 369
 370         struct v3_segment * segregs [] = {&(core->segments.ss), &(core->segments.ds),
 371                                           &(core->segments.es), &(core->segments.fs),
 372                                           &(core->segments.gs), NULL};
 373
 374         for ( i = 0; segregs[i] != NULL; i++) {
 375             struct v3_segment * seg = segregs[i];
 376
 377             seg->selector = 0x0000;
 378             //    seg->base = seg->selector << 4;
 379             seg->base = 0x00000000;
 380             seg->limit = 0xffff;
 381
 382
 383             seg->type = 0x3;
 384             seg->system = 0x1;
 385             seg->dpl = 0x0;
 386             seg->present = 1;
 387             //    seg->granularity = 1;
 388
 389         }
 390
 391
 392         core->segments.gdtr.limit = 0x0000ffff;
 393         core->segments.gdtr.base = 0x0000000000000000LL;
 394
 395         core->segments.idtr.limit = 0x0000ffff;
 396         core->segments.idtr.base = 0x0000000000000000LL;
 397
 398         core->segments.ldtr.selector = 0x0000;
 399         core->segments.ldtr.limit = 0x0000ffff;
 400         core->segments.ldtr.base = 0x0000000000000000LL;
 401         core->segments.ldtr.type = 0x2;
 402         core->segments.ldtr.present = 1;
 403
 404         core->segments.tr.selector = 0x0000;
 405         core->segments.tr.limit = 0x0000ffff;
 406         core->segments.tr.base = 0x0000000000000000LL;
 407         core->segments.tr.type = 0xb;
 408         core->segments.tr.present = 1;
 409
 410         //      core->dbg_regs.dr6 = 0x00000000ffff0ff0LL;
 411         core->dbg_regs.dr7 = 0x0000000000000400LL;
 412
 413         /* Enable EPT */
 414         vmx_state->pri_proc_ctrls.sec_ctrls = 1; // Enable secondary proc controls
 415         vmx_state->sec_proc_ctrls.enable_ept = 1; // enable EPT paging
 416         vmx_state->sec_proc_ctrls.unrstrct_guest = 1; // enable unrestricted guest operation
 417
 418
 419         /* Disable shadow paging stuff */
 420         vmx_state->pri_proc_ctrls.cr3_ld_exit = 0;
 421         vmx_state->pri_proc_ctrls.cr3_str_exit = 0;
 422
 423         vmx_state->pri_proc_ctrls.invlpg_exit = 0;
 424
 425
 426         // Cause VM_EXIT whenever the CR4.VMXE bit is set
 427         vmx_ret |= check_vmcs_write(VMCS_CR4_MASK, CR4_VMXE);
 428 #define CR0_NE 0x00000020
 429 #define CR0_CD 0x40000000
 430         vmx_ret |= check_vmcs_write(VMCS_CR0_MASK, CR0_NE | CR0_CD);
 431         ((struct cr0_32 *)&(core->shdw_pg_state.guest_cr0))->ne = 1;
 432         ((struct cr0_32 *)&(core->shdw_pg_state.guest_cr0))->cd = 0;
 433
 434         if (v3_init_nested_paging_core(core, &hw_info) == -1) {
 435             PrintError(core->vm_info, core, "Error initializing EPT\n");
 436             return -1;
 437         }
 438
 439         // Hook all accesses to EFER register
 440         //      v3_hook_msr(core->vm_info, EFER_MSR, &debug_efer_read, &debug_efer_write, core);
 441         v3_hook_msr(core->vm_info, EFER_MSR, NULL, NULL, NULL);
 442     } else {
 443         PrintError(core->vm_info, core, "Invalid Virtual paging mode (pg_mode=%d) (mach_type=%d)\n", core->shdw_pg_mode, v3_mach_type);
 444         return -1;
 445     }
 446
 447
 448     // hook vmx msrs
 449
 450     // Setup SYSCALL/SYSENTER MSRs in load/store area
 451
 452     // save STAR, LSTAR, FMASK, KERNEL_GS_BASE MSRs in MSR load/store area
 453     {
 454
 455         struct vmcs_msr_save_area * msr_entries = NULL;
 456         int max_msrs = (hw_info.misc_info.max_msr_cache_size + 1) * 4;
 457         int msr_ret = 0;
 458
 459         V3_Print(core->vm_info, core, "Setting up MSR load/store areas (max_msr_count=%d)\n", max_msrs);
 460
 461         if (max_msrs < 4) {
 462             PrintError(core->vm_info, core, "Max MSR cache size is too small (%d)\n", max_msrs);
 463             return -1;
 464         }
 465
 466         vmx_state->msr_area_paddr = (addr_t)V3_AllocPages(1); // need not be shadow-safe, not exposed to guest
 467
 468         if (vmx_state->msr_area_paddr == (addr_t)NULL) {
 469             PrintError(core->vm_info, core, "could not allocate msr load/store area\n");
 470             return -1;
 471         }
 472
 473         msr_entries = (struct vmcs_msr_save_area *)V3_VAddr((void *)(vmx_state->msr_area_paddr));
 474         vmx_state->msr_area = msr_entries; // cache in vmx_info
 475
 476         memset(msr_entries, 0, PAGE_SIZE);
 477
 478         msr_entries->guest_star.index = IA32_STAR_MSR;
 479         msr_entries->guest_lstar.index = IA32_LSTAR_MSR;
 480         msr_entries->guest_fmask.index = IA32_FMASK_MSR;
 481         msr_entries->guest_kern_gs.index = IA32_KERN_GS_BASE_MSR;
 482
 483         msr_entries->host_star.index = IA32_STAR_MSR;
 484         msr_entries->host_lstar.index = IA32_LSTAR_MSR;
 485         msr_entries->host_fmask.index = IA32_FMASK_MSR;
 486         msr_entries->host_kern_gs.index = IA32_KERN_GS_BASE_MSR;
 487
 488         msr_ret |= check_vmcs_write(VMCS_EXIT_MSR_STORE_CNT, 4);
 489         msr_ret |= check_vmcs_write(VMCS_EXIT_MSR_LOAD_CNT, 4);
 490         msr_ret |= check_vmcs_write(VMCS_ENTRY_MSR_LOAD_CNT, 4);
 491
 492         msr_ret |= check_vmcs_write(VMCS_EXIT_MSR_STORE_ADDR, (addr_t)V3_PAddr(msr_entries->guest_msrs));
 493         msr_ret |= check_vmcs_write(VMCS_ENTRY_MSR_LOAD_ADDR, (addr_t)V3_PAddr(msr_entries->guest_msrs));
 494         msr_ret |= check_vmcs_write(VMCS_EXIT_MSR_LOAD_ADDR, (addr_t)V3_PAddr(msr_entries->host_msrs));
 495
 496
 497         msr_ret |= v3_hook_msr(core->vm_info, IA32_STAR_MSR, NULL, NULL, NULL);
 498         msr_ret |= v3_hook_msr(core->vm_info, IA32_LSTAR_MSR, NULL, NULL, NULL);
 499         msr_ret |= v3_hook_msr(core->vm_info, IA32_FMASK_MSR, NULL, NULL, NULL);
 500         msr_ret |= v3_hook_msr(core->vm_info, IA32_KERN_GS_BASE_MSR, NULL, NULL, NULL);
 501
 502
 503         // IMPORTANT: These MSRs appear to be cached by the hardware....
 504         msr_ret |= v3_hook_msr(core->vm_info, SYSENTER_CS_MSR, NULL, NULL, NULL);
 505         msr_ret |= v3_hook_msr(core->vm_info, SYSENTER_ESP_MSR, NULL, NULL, NULL);
 506         msr_ret |= v3_hook_msr(core->vm_info, SYSENTER_EIP_MSR, NULL, NULL, NULL);
 507
 508         msr_ret |= v3_hook_msr(core->vm_info, FS_BASE_MSR, NULL, NULL, NULL);
 509         msr_ret |= v3_hook_msr(core->vm_info, GS_BASE_MSR, NULL, NULL, NULL);
 510
 511         msr_ret |= v3_hook_msr(core->vm_info, IA32_PAT_MSR, NULL, NULL, NULL);
 512
 513         // Not sure what to do about this... Does not appear to be an explicit hardware cache version...
 514         msr_ret |= v3_hook_msr(core->vm_info, IA32_CSTAR_MSR, NULL, NULL, NULL);
 515
 516         if (msr_ret != 0) {
 517             PrintError(core->vm_info, core, "Error configuring MSR save/restore area\n");
 518             return -1;
 519         }
 520
 521
 522     }
 523
 524     /* Sanity check ctrl/reg fields against hw_defaults */
 525
 526
 527
 528
 529     /*** Write all the info to the VMCS ***/
 530
 531     /*
 532     {
 533         // IS THIS NECESSARY???
 534 #define DEBUGCTL_MSR 0x1d9
 535         struct v3_msr tmp_msr;
 536         v3_get_msr(DEBUGCTL_MSR, &(tmp_msr.hi), &(tmp_msr.lo));
 537         vmx_ret |= check_vmcs_write(VMCS_GUEST_DBG_CTL, tmp_msr.value);
 538         core->dbg_regs.dr7 = 0x400;
 539     }
 540     */
 541
 542 #ifdef __V3_64BIT__
 543     vmx_ret |= check_vmcs_write(VMCS_LINK_PTR, (addr_t)0xffffffffffffffffULL);
 544 #else
 545     vmx_ret |= check_vmcs_write(VMCS_LINK_PTR, (addr_t)0xffffffffUL);
 546     vmx_ret |= check_vmcs_write(VMCS_LINK_PTR_HIGH, (addr_t)0xffffffffUL);
 547 #endif
 548
 549
 550
 551
 552
 553     if (v3_update_vmcs_ctrl_fields(core)) {
 554         PrintError(core->vm_info, core, "Could not write control fields!\n");
 555         return -1;
 556     }
 557
 558     /*
 559     if (v3_update_vmcs_host_state(core)) {
 560         PrintError(core->vm_info, core, "Could not write host state\n");
 561         return -1;
 562     }
 563     */
 564
 565     // reenable global interrupts for vm state initialization now
 566     // that the vm state is initialized. If another VM kicks us off,
 567     // it'll update our vmx state so that we know to reload ourself
 568     v3_enable_ints();
 569
 570     return 0;
 571 }
 572
 573
 574 static void __init_vmx_vmcs(void * arg) {
 575     struct guest_info * core = arg;
 576     struct vmx_data * vmx_state = NULL;
 577     int vmx_ret = 0;
 578
 579     vmx_state = (struct vmx_data *)V3_Malloc(sizeof(struct vmx_data));
 580
 581     if (!vmx_state) {
 582         PrintError(core->vm_info, core,  "Unable to allocate in initializing vmx vmcs\n");
 583         return;
 584     }
 585
 586     memset(vmx_state, 0, sizeof(struct vmx_data));
 587
 588     PrintDebug(core->vm_info, core,  "vmx_data pointer: %p\n", (void *)vmx_state);
 589
 590     PrintDebug(core->vm_info, core, "Allocating VMCS\n");
 591     vmx_state->vmcs_ptr_phys = allocate_vmcs();
 592
 593     PrintDebug(core->vm_info, core, "VMCS pointer: %p\n", (void *)(vmx_state->vmcs_ptr_phys));
 594
 595     core->vmm_data = vmx_state;
 596     vmx_state->state = VMX_UNLAUNCHED;
 597
 598     PrintDebug(core->vm_info, core, "Initializing VMCS (addr=%p)\n", core->vmm_data);
 599
 600     // TODO: Fix vmcs fields so they're 32-bit
 601
 602     PrintDebug(core->vm_info, core, "Clearing VMCS: %p\n", (void *)vmx_state->vmcs_ptr_phys);
 603     vmx_ret = vmcs_clear(vmx_state->vmcs_ptr_phys);
 604
 605     if (vmx_ret != VMX_SUCCESS) {
 606         PrintError(core->vm_info, core, "VMCLEAR failed\n");
 607         return;
 608     }
 609
 610     if (core->vm_info->vm_class == V3_PC_VM) {
 611         PrintDebug(core->vm_info, core, "Initializing VMCS\n");
 612         if (init_vmcs_bios(core, vmx_state) == -1) {
 613             PrintError(core->vm_info, core, "Error initializing VMCS to BIOS state\n");
 614             return;
 615         }
 616     } else {
 617         PrintError(core->vm_info, core, "Invalid VM Class\n");
 618         return;
 619     }
 620
 621     PrintDebug(core->vm_info, core, "Serializing VMCS: %p\n", (void *)vmx_state->vmcs_ptr_phys);
 622     vmx_ret = vmcs_clear(vmx_state->vmcs_ptr_phys);
 623
 624     core->core_run_state = CORE_STOPPED;
 625     return;
 626 }
 627
 628
 629
 630 int v3_init_vmx_vmcs(struct guest_info * core, v3_vm_class_t vm_class) {
 631     extern v3_cpu_arch_t v3_cpu_types[];
 632
 633     if (v3_cpu_types[V3_Get_CPU()] == V3_INVALID_CPU) {
 634         int i = 0;
 635
 636         for (i = 0; i < V3_CONFIG_MAX_CPUS; i++) {
 637             if (v3_cpu_types[i] != V3_INVALID_CPU) {
 638                 break;
 639             }
 640         }
 641
 642         if (i == V3_CONFIG_MAX_CPUS) {
 643             PrintError(core->vm_info, core, "Could not find VALID CPU for VMX guest initialization\n");
 644             return -1;
 645         }
 646
 647         V3_Call_On_CPU(i, __init_vmx_vmcs, core);
 648
 649     } else {
 650         __init_vmx_vmcs(core);
 651     }
 652
 653     if (core->core_run_state != CORE_STOPPED) {
 654         PrintError(core->vm_info, core, "Error initializing VMX Core\n");
 655         return -1;
 656     }
 657
 658     return 0;
 659 }
 660
 661
 662 int v3_deinit_vmx_vmcs(struct guest_info * core) {
 663     struct vmx_data * vmx_state = core->vmm_data;
 664
 665     V3_FreePages((void *)(vmx_state->vmcs_ptr_phys), 1);
 666     V3_FreePages(V3_PAddr(vmx_state->msr_area), 1);
 667
 668     V3_Free(vmx_state);
 669
 670     return 0;
 671 }
 672
 673
 674
 675 #ifdef V3_CONFIG_CHECKPOINT
 676 /*
 677  * JRL: This is broken
 678  */
 679 int v3_vmx_save_core(struct guest_info * core, void * ctx){
 680   struct vmx_data * vmx_info = (struct vmx_data *)(core->vmm_data);
 681
 682   // note that the vmcs pointer is an HPA, but we need an HVA
 683   if (v3_chkpt_save(ctx, "vmcs_data", PAGE_SIZE_4KB,
 684                     V3_VAddr((void*) (vmx_info->vmcs_ptr_phys)))) {
 685     PrintError(core->vm_info, core, "Could not save vmcs data for VMX\n");
 686     return -1;
 687   }
 688
 689   return 0;
 690 }
 691
 692 int v3_vmx_load_core(struct guest_info * core, void * ctx){
 693   struct vmx_data * vmx_info = (struct vmx_data *)(core->vmm_data);
 694   struct cr0_32 * shadow_cr0;
 695   addr_t vmcs_page_paddr;  //HPA
 696
 697   vmcs_page_paddr = (addr_t) V3_AllocPages(1); // need not be shadow-safe, not exposed to guest
 698
 699   if (!vmcs_page_paddr) {
 700     PrintError(core->vm_info, core, "Could not allocate space for a vmcs in VMX\n");
 701     return -1;
 702   }
 703
 704   if (v3_chkpt_load(ctx, "vmcs_data", PAGE_SIZE_4KB,
 705                     V3_VAddr((void *)vmcs_page_paddr)) == -1) {
 706     PrintError(core->vm_info, core, "Could not load vmcs data for VMX\n");
 707     V3_FreePages((void*)vmcs_page_paddr,1);
 708     return -1;
 709   }
 710
 711   vmcs_clear(vmx_info->vmcs_ptr_phys);
 712
 713   // Probably need to delete the old one...
 714   V3_FreePages((void*)(vmx_info->vmcs_ptr_phys),1);
 715
 716   vmcs_load(vmcs_page_paddr);
 717
 718   v3_vmx_save_vmcs(core);
 719
 720   shadow_cr0 = (struct cr0_32 *)&(core->ctrl_regs.cr0);
 721
 722
 723   /* Get the CPU mode to set the guest_ia32e entry ctrl */
 724
 725   if (core->shdw_pg_mode == SHADOW_PAGING) {
 726     if (v3_get_vm_mem_mode(core) == VIRTUAL_MEM) {
 727       if (v3_activate_shadow_pt(core) == -1) {
 728         PrintError(core->vm_info, core, "Failed to activate shadow page tables\n");
 729         return -1;
 730       }
 731     } else {
 732       if (v3_activate_passthrough_pt(core) == -1) {
 733         PrintError(core->vm_info, core, "Failed to activate passthrough page tables\n");
 734         return -1;
 735       }
 736     }
 737   }
 738
 739   return 0;
 740 }
 741 #endif
 742
 743
 744 void v3_flush_vmx_vm_core(struct guest_info * core) {
 745     struct vmx_data * vmx_info = (struct vmx_data *)(core->vmm_data);
 746     vmcs_clear(vmx_info->vmcs_ptr_phys);
 747     vmx_info->state = VMX_UNLAUNCHED;
 748 }
 749
 750
 751
 752 static int update_irq_exit_state(struct guest_info * info) {
 753     struct vmx_exit_idt_vec_info idt_vec_info;
 754
 755     check_vmcs_read(VMCS_IDT_VECTOR_INFO, &(idt_vec_info.value));
 756
 757     if ((info->intr_core_state.irq_started == 1) && (idt_vec_info.valid == 0)) {
 758 #ifdef V3_CONFIG_DEBUG_INTERRUPTS
 759         V3_Print(info->vm_info, info, "Calling v3_injecting_intr\n");
 760 #endif
 761         info->intr_core_state.irq_started = 0;
 762         v3_injecting_intr(info, info->intr_core_state.irq_vector, V3_EXTERNAL_IRQ);
 763     }
 764
 765     return 0;
 766 }
 767
 768 static int update_irq_entry_state(struct guest_info * info) {
 769     struct vmx_exit_idt_vec_info idt_vec_info;
 770     struct vmcs_interrupt_state intr_core_state;
 771     struct vmx_data * vmx_info = (struct vmx_data *)(info->vmm_data);
 772
 773     check_vmcs_read(VMCS_IDT_VECTOR_INFO, &(idt_vec_info.value));
 774     check_vmcs_read(VMCS_GUEST_INT_STATE, &(intr_core_state));
 775
 776     /* Check for pending exceptions to inject */
 777     if (v3_excp_pending(info)) {
 778         struct vmx_entry_int_info int_info;
 779         int_info.value = 0;
 780
 781         // In VMX, almost every exception is hardware
 782         // Software exceptions are pretty much only for breakpoint or overflow
 783         int_info.type = 3;
 784         int_info.vector = v3_get_excp_number(info);
 785
 786         if (info->excp_state.excp_error_code_valid) {
 787             check_vmcs_write(VMCS_ENTRY_EXCP_ERR, info->excp_state.excp_error_code);
 788             int_info.error_code = 1;
 789
 790 #ifdef V3_CONFIG_DEBUG_INTERRUPTS
 791             V3_Print(info->vm_info, info, "Injecting exception %d with error code %x\n",
 792                     int_info.vector, info->excp_state.excp_error_code);
 793 #endif
 794         }
 795
 796         int_info.valid = 1;
 797 #ifdef V3_CONFIG_DEBUG_INTERRUPTS
 798         V3_Print(info->vm_info, info, "Injecting exception %d (EIP=%p)\n", int_info.vector, (void *)(addr_t)info->rip);
 799 #endif
 800         check_vmcs_write(VMCS_ENTRY_INT_INFO, int_info.value);
 801
 802         v3_injecting_excp(info, int_info.vector);
 803
 804     } else if ((((struct rflags *)&(info->ctrl_regs.rflags))->intr == 1) &&
 805                (intr_core_state.val == 0)) {
 806
 807         if ((info->intr_core_state.irq_started == 1) && (idt_vec_info.valid == 1)) {
 808
 809 #ifdef V3_CONFIG_DEBUG_INTERRUPTS
 810             V3_Print(info->vm_info, info, "IRQ pending from previous injection\n");
 811 #endif
 812
 813             // Copy the IDT vectoring info over to reinject the old interrupt
 814             if (idt_vec_info.error_code == 1) {
 815                 uint32_t err_code = 0;
 816
 817                 check_vmcs_read(VMCS_IDT_VECTOR_ERR, &err_code);
 818                 check_vmcs_write(VMCS_ENTRY_EXCP_ERR, err_code);
 819             }
 820
 821             idt_vec_info.undef = 0;
 822             check_vmcs_write(VMCS_ENTRY_INT_INFO, idt_vec_info.value);
 823
 824         } else {
 825             struct vmx_entry_int_info ent_int;
 826             ent_int.value = 0;
 827
 828             switch (v3_intr_pending(info)) {
 829                 case V3_EXTERNAL_IRQ: {
 830
 831                     int irq = v3_get_intr(info);
 832
 833                     if (irq<0) {
 834                       break;
 835                     }
 836
 837                     info->intr_core_state.irq_vector = irq;
 838                     ent_int.vector = info->intr_core_state.irq_vector;
 839                     ent_int.type = 0;
 840                     ent_int.error_code = 0;
 841                     ent_int.valid = 1;
 842
 843 #ifdef V3_CONFIG_DEBUG_INTERRUPTS
 844                     V3_Print(info->vm_info, info, "Injecting Interrupt %d at exit %u(EIP=%p)\n",
 845                                info->intr_core_state.irq_vector,
 846                                (uint32_t)info->num_exits,
 847                                (void *)(addr_t)info->rip);
 848 #endif
 849
 850                     check_vmcs_write(VMCS_ENTRY_INT_INFO, ent_int.value);
 851                     info->intr_core_state.irq_started = 1;
 852
 853                     break;
 854                 }
 855                 case V3_NMI:
 856                     PrintDebug(info->vm_info, info, "Injecting NMI\n");
 857
 858                     ent_int.type = 2;
 859                     ent_int.vector = 2;
 860                     ent_int.valid = 1;
 861                     check_vmcs_write(VMCS_ENTRY_INT_INFO, ent_int.value);
 862
 863                     break;
 864                 case V3_SOFTWARE_INTR:
 865                     PrintDebug(info->vm_info, info, "Injecting software interrupt\n");
 866                     ent_int.type = 4;
 867
 868                     ent_int.valid = 1;
 869                     check_vmcs_write(VMCS_ENTRY_INT_INFO, ent_int.value);
 870
 871                     break;
 872                 case V3_VIRTUAL_IRQ:
 873                     // Not sure what to do here, Intel doesn't have virtual IRQs
 874                     // May be the same as external interrupts/IRQs
 875
 876                     break;
 877                 case V3_INVALID_INTR:
 878                 default:
 879                     break;
 880             }
 881         }
 882     } else if ((v3_intr_pending(info)) && (vmx_info->pri_proc_ctrls.int_wndw_exit == 0)) {
 883         // Enable INTR window exiting so we know when IF=1
 884         uint32_t instr_len;
 885
 886         check_vmcs_read(VMCS_EXIT_INSTR_LEN, &instr_len);
 887
 888 #ifdef V3_CONFIG_DEBUG_INTERRUPTS
 889         V3_Print(info->vm_info, info, "Enabling Interrupt-Window exiting: %d\n", instr_len);
 890 #endif
 891
 892         vmx_info->pri_proc_ctrls.int_wndw_exit = 1;
 893         check_vmcs_write(VMCS_PROC_CTRLS, vmx_info->pri_proc_ctrls.value);
 894     }
 895
 896
 897     return 0;
 898 }
 899
 900
 901
 902 static struct vmx_exit_info exit_log[10];
 903 static uint64_t rip_log[10];
 904
 905
 906
 907 static void print_exit_log(struct guest_info * info) {
 908     int cnt = info->num_exits % 10;
 909     int i = 0;
 910
 911
 912     V3_Print(info->vm_info, info, "\nExit Log (%d total exits):\n", (uint32_t)info->num_exits);
 913
 914     for (i = 0; i < 10; i++) {
 915         struct vmx_exit_info * tmp = &exit_log[cnt];
 916
 917         V3_Print(info->vm_info, info, "%d:\texit_reason = %p\n", i, (void *)(addr_t)tmp->exit_reason);
 918         V3_Print(info->vm_info, info, "\texit_qual = %p\n", (void *)tmp->exit_qual);
 919         V3_Print(info->vm_info, info, "\tint_info = %p\n", (void *)(addr_t)tmp->int_info);
 920         V3_Print(info->vm_info, info, "\tint_err = %p\n", (void *)(addr_t)tmp->int_err);
 921         V3_Print(info->vm_info, info, "\tinstr_info = %p\n", (void *)(addr_t)tmp->instr_info);
 922         V3_Print(info->vm_info, info, "\tguest_linear_addr= %p\n", (void *)(addr_t)tmp->guest_linear_addr);
 923         V3_Print(info->vm_info, info, "\tRIP = %p\n", (void *)rip_log[cnt]);
 924
 925
 926         cnt--;
 927
 928         if (cnt == -1) {
 929             cnt = 9;
 930         }
 931
 932     }
 933
 934 }
 935
 936 int
 937 v3_vmx_config_tsc_virtualization(struct guest_info * info) {
 938     struct vmx_data * vmx_info = (struct vmx_data *)(info->vmm_data);
 939
 940     if (info->time_state.flags & VM_TIME_TRAP_RDTSC) {
 941         if  (!vmx_info->pri_proc_ctrls.rdtsc_exit) {
 942             vmx_info->pri_proc_ctrls.rdtsc_exit = 1;
 943             check_vmcs_write(VMCS_PROC_CTRLS, vmx_info->pri_proc_ctrls.value);
 944         }
 945     } else {
 946         sint64_t tsc_offset;
 947         uint32_t tsc_offset_low, tsc_offset_high;
 948
 949         if  (vmx_info->pri_proc_ctrls.rdtsc_exit) {
 950             vmx_info->pri_proc_ctrls.rdtsc_exit = 0;
 951             check_vmcs_write(VMCS_PROC_CTRLS, vmx_info->pri_proc_ctrls.value);
 952         }
 953
 954         if (info->time_state.flags & VM_TIME_TSC_PASSTHROUGH) {
 955             tsc_offset = 0;
 956         } else {
 957             tsc_offset = v3_tsc_host_offset(&info->time_state);
 958         }
 959         tsc_offset_high = (uint32_t)(( tsc_offset >> 32) & 0xffffffff);
 960         tsc_offset_low = (uint32_t)(tsc_offset & 0xffffffff);
 961
 962         check_vmcs_write(VMCS_TSC_OFFSET_HIGH, tsc_offset_high);
 963         check_vmcs_write(VMCS_TSC_OFFSET, tsc_offset_low);
 964     }
 965     return 0;
 966 }
 967
 968 /*
 969  * CAUTION and DANGER!!!
 970  *
 971  * The VMCS CANNOT(!!) be accessed outside of the cli/sti calls inside this function
 972  * When exectuing a symbiotic call, the VMCS WILL be overwritten, so any dependencies
 973  * on its contents will cause things to break. The contents at the time of the exit WILL
 974  * change before the exit handler is executed.
 975  */
 976 int v3_vmx_enter(struct guest_info * info) {
 977     int ret = 0;
 978     struct vmx_exit_info exit_info;
 979     struct vmx_data * vmx_info = (struct vmx_data *)(info->vmm_data);
 980     uint64_t guest_cycles = 0;
 981
 982     // Conditionally yield the CPU if the timeslice has expired
 983     v3_schedule(info);
 984
 985 #ifdef V3_CONFIG_MEM_TRACK
 986     v3_mem_track_entry(info);
 987 #endif
 988
 989     // Update timer devices late after being in the VM so that as much
 990     // of the time in the VM is accounted for as possible. Also do it before
 991     // updating IRQ entry state so that any interrupts the timers raise get
 992     // handled on the next VM entry.
 993     v3_advance_time(info, NULL);
 994     v3_update_timers(info);
 995
 996     // disable global interrupts for vm state transition
 997     v3_disable_ints();
 998
 999     if (vmcs_store() != vmx_info->vmcs_ptr_phys) {
1000         vmcs_clear(vmx_info->vmcs_ptr_phys);
1001         vmcs_load(vmx_info->vmcs_ptr_phys);
1002         vmx_info->state = VMX_UNLAUNCHED;
1003     }
1004
1005     v3_vmx_restore_vmcs(info);
1006
1007
1008 #ifdef V3_CONFIG_SYMCALL
1009     if (info->sym_core_state.symcall_state.sym_call_active == 0) {
1010         update_irq_entry_state(info);
1011     }
1012 #else
1013     update_irq_entry_state(info);
1014 #endif
1015
1016     {
1017         addr_t guest_cr3;
1018         vmcs_read(VMCS_GUEST_CR3, &guest_cr3);
1019         vmcs_write(VMCS_GUEST_CR3, guest_cr3);
1020     }
1021
1022
1023     // Perform last-minute time setup prior to entering the VM
1024     v3_vmx_config_tsc_virtualization(info);
1025
1026     if (v3_update_vmcs_host_state(info)) {
1027         v3_enable_ints();
1028         PrintError(info->vm_info, info, "Could not write host state\n");
1029         return -1;
1030     }
1031
1032     if (vmx_info->pin_ctrls.active_preempt_timer) {
1033         /* Preemption timer is active */
1034         uint32_t preempt_window = 0xffffffff;
1035
1036         if (info->timeouts.timeout_active) {
1037             preempt_window = info->timeouts.next_timeout;
1038         }
1039
1040         check_vmcs_write(VMCS_PREEMPT_TIMER, preempt_window);
1041     }
1042
1043     V3_FP_ENTRY_RESTORE(info);
1044
1045     {
1046         uint64_t entry_tsc = 0;
1047         uint64_t exit_tsc = 0;
1048
1049 #ifdef V3_CONFIG_PWRSTAT_TELEMETRY
1050         v3_pwrstat_telemetry_enter(info);
1051 #endif
1052
1053 #ifdef V3_CONFIG_PMU_TELEMETRY
1054         v3_pmu_telemetry_enter(info);
1055 #endif
1056
1057         if (vmx_info->state == VMX_UNLAUNCHED) {
1058             vmx_info->state = VMX_LAUNCHED;
1059             rdtscll(entry_tsc);
1060             ret = v3_vmx_launch(&(info->vm_regs), info, &(info->ctrl_regs));
1061             rdtscll(exit_tsc);
1062
1063         } else {
1064             V3_ASSERT(info->vm_info, info,vmx_info->state != VMX_UNLAUNCHED);
1065             rdtscll(entry_tsc);
1066             ret = v3_vmx_resume(&(info->vm_regs), info, &(info->ctrl_regs));
1067             rdtscll(exit_tsc);
1068         }
1069
1070         guest_cycles = exit_tsc - entry_tsc;
1071
1072 #ifdef V3_CONFIG_PMU_TELEMETRY
1073         v3_pmu_telemetry_exit(info);
1074 #endif
1075
1076 #ifdef V3_CONFIG_PWRSTAT_TELEMETRY
1077         v3_pwrstat_telemetry_exit(info);
1078 #endif
1079     }
1080
1081     //  PrintDebug(info->vm_info, info, "VMX Exit: ret=%d\n", ret);
1082
1083     if (ret != VMX_SUCCESS) {
1084         uint32_t error = 0;
1085         vmcs_read(VMCS_INSTR_ERR, &error);
1086
1087         v3_enable_ints();
1088
1089         PrintError(info->vm_info, info, "VMENTRY Error: %d (launch_ret = %d)\n", error, ret);
1090         return -1;
1091     }
1092
1093
1094     info->num_exits++;
1095
1096     V3_FP_EXIT_SAVE(info);
1097
1098     /* If we have the preemption time, then use it to get more accurate guest time */
1099     if (vmx_info->pin_ctrls.active_preempt_timer) {
1100         uint32_t cycles_left = 0;
1101         check_vmcs_read(VMCS_PREEMPT_TIMER, &(cycles_left));
1102
1103         if (info->timeouts.timeout_active) {
1104             guest_cycles = info->timeouts.next_timeout - cycles_left;
1105         } else {
1106             guest_cycles = 0xffffffff - cycles_left;
1107         }
1108     }
1109
1110     // Immediate exit from VM time bookkeeping
1111     v3_advance_time(info, &guest_cycles);
1112
1113     /* Update guest state */
1114     v3_vmx_save_vmcs(info);
1115
1116     // info->cpl = info->segments.cs.selector & 0x3;
1117
1118     info->mem_mode = v3_get_vm_mem_mode(info);
1119     info->cpu_mode = v3_get_vm_cpu_mode(info);
1120
1121
1122
1123     check_vmcs_read(VMCS_EXIT_INSTR_LEN, &(exit_info.instr_len));
1124     check_vmcs_read(VMCS_EXIT_INSTR_INFO, &(exit_info.instr_info));
1125     check_vmcs_read(VMCS_EXIT_REASON, &(exit_info.exit_reason));
1126     check_vmcs_read(VMCS_EXIT_QUAL, &(exit_info.exit_qual));
1127     check_vmcs_read(VMCS_EXIT_INT_INFO, &(exit_info.int_info));
1128     check_vmcs_read(VMCS_EXIT_INT_ERR, &(exit_info.int_err));
1129     check_vmcs_read(VMCS_GUEST_LINEAR_ADDR, &(exit_info.guest_linear_addr));
1130
1131     if (info->shdw_pg_mode == NESTED_PAGING) {
1132         check_vmcs_read(VMCS_GUEST_PHYS_ADDR, &(exit_info.ept_fault_addr));
1133     }
1134
1135     //PrintDebug(info->vm_info, info, "VMX Exit taken, id-qual: %u-%lu\n", exit_info.exit_reason, exit_info.exit_qual);
1136
1137     exit_log[info->num_exits % 10] = exit_info;
1138     rip_log[info->num_exits % 10] = get_addr_linear(info, info->rip, &(info->segments.cs));
1139
1140 #ifdef V3_CONFIG_SYMCALL
1141     if (info->sym_core_state.symcall_state.sym_call_active == 0) {
1142         update_irq_exit_state(info);
1143     }
1144 #else
1145     update_irq_exit_state(info);
1146 #endif
1147
1148     if (exit_info.exit_reason == VMX_EXIT_INTR_WINDOW) {
1149         // This is a special case whose only job is to inject an interrupt
1150         vmcs_read(VMCS_PROC_CTRLS, &(vmx_info->pri_proc_ctrls.value));
1151         vmx_info->pri_proc_ctrls.int_wndw_exit = 0;
1152         vmcs_write(VMCS_PROC_CTRLS, vmx_info->pri_proc_ctrls.value);
1153
1154 #ifdef V3_CONFIG_DEBUG_INTERRUPTS
1155        V3_Print(info->vm_info, info, "Interrupts available again! (RIP=%llx)\n", info->rip);
1156 #endif
1157     }
1158
1159
1160     // Lastly we check for an NMI exit, and reinject if so
1161     {
1162         struct vmx_basic_exit_info * basic_info = (struct vmx_basic_exit_info *)&(exit_info.exit_reason);
1163
1164         if (basic_info->reason == VMX_EXIT_INFO_EXCEPTION_OR_NMI) {
1165             if ((uint8_t)exit_info.int_info == 2) {
1166                 asm("int $2");
1167             }
1168         }
1169     }
1170
1171     // reenable global interrupts after vm exit
1172     v3_enable_ints();
1173
1174     // Conditionally yield the CPU if the timeslice has expired
1175     v3_schedule(info);
1176     v3_advance_time(info, NULL);
1177     v3_update_timers(info);
1178
1179     if (v3_handle_vmx_exit(info, &exit_info) == -1) {
1180         PrintError(info->vm_info, info, "Error in VMX exit handler (Exit reason=%x)\n", exit_info.exit_reason);
1181         return -1;
1182     }
1183
1184     if (info->timeouts.timeout_active) {
1185         /* Check to see if any timeouts have expired */
1186         v3_handle_timeouts(info, guest_cycles);
1187     }
1188
1189 #ifdef V3_CONFIG_MEM_TRACK
1190     v3_mem_track_exit(info);
1191 #endif
1192
1193     return 0;
1194 }
1195
1196
1197 int v3_start_vmx_guest(struct guest_info * info) {
1198
1199     PrintDebug(info->vm_info, info, "Starting VMX core %u\n", info->vcpu_id);
1200
1201 #if V3_CONFIG_HVM
1202     if (v3_setup_hvm_vm_for_boot(vm)) {
1203         PrintError(vm, VCORE_NONE, "HVM setup for boot failed\n");
1204         return -1;
1205     }
1206 #endif
1207
1208     while (1) {
1209         if (info->core_run_state == CORE_STOPPED) {
1210             if (info->vcpu_id == 0) {
1211                 info->core_run_state = CORE_RUNNING;
1212             } else {
1213
1214                 PrintDebug(info->vm_info, info, "VMX core %u: Waiting for core initialization\n", info->vcpu_id);
1215
1216                 V3_NO_WORK(info);
1217
1218                 while (info->core_run_state == CORE_STOPPED) {
1219
1220                     if (info->vm_info->run_state == VM_STOPPED) {
1221                         // The VM was stopped before this core was initialized.
1222                         return 0;
1223                     }
1224
1225                     V3_STILL_NO_WORK(info);
1226                     //PrintDebug(info->vm_info, info, "VMX core %u: still waiting for INIT\n",info->vcpu_id);
1227                 }
1228
1229                 V3_HAVE_WORK_AGAIN(info);
1230
1231                 PrintDebug(info->vm_info, info, "VMX core %u initialized\n", info->vcpu_id);
1232
1233                 // We'll be paranoid about race conditions here
1234                 v3_wait_at_barrier(info);
1235             }
1236
1237
1238             PrintDebug(info->vm_info, info, "VMX core %u: I am starting at CS=0x%x (base=0x%p, limit=0x%x),  RIP=0x%p\n",
1239                        info->vcpu_id, info->segments.cs.selector, (void *)(info->segments.cs.base),
1240                        info->segments.cs.limit, (void *)(info->rip));
1241
1242
1243             PrintDebug(info->vm_info, info, "VMX core %u: Launching VMX VM on logical core %u\n", info->vcpu_id, info->pcpu_id);
1244
1245             v3_start_time(info);
1246
1247
1248             if (info->vm_info->run_state == VM_STOPPED) {
1249                 info->core_run_state = CORE_STOPPED;
1250                 break;
1251             }
1252         }
1253
1254
1255 #ifdef V3_CONFIG_PMU_TELEMETRY
1256         v3_pmu_telemetry_start(info);
1257 #endif
1258
1259 #ifdef V3_CONFIG_PWRSTAT_TELEMETRY
1260         v3_pwrstat_telemetry_start(info);
1261 #endif
1262
1263
1264         if (v3_vmx_enter(info) == -1) {
1265
1266             addr_t host_addr;
1267             addr_t linear_addr = 0;
1268
1269             info->vm_info->run_state = VM_ERROR;
1270
1271             V3_Print(info->vm_info, info, "VMX core %u: VMX ERROR!!\n", info->vcpu_id);
1272
1273             v3_print_guest_state(info);
1274
1275             V3_Print(info->vm_info, info, "VMX core %u\n", info->vcpu_id);
1276
1277             linear_addr = get_addr_linear(info, info->rip, &(info->segments.cs));
1278
1279             if (info->mem_mode == PHYSICAL_MEM) {
1280                 v3_gpa_to_hva(info, linear_addr, &host_addr);
1281             } else if (info->mem_mode == VIRTUAL_MEM) {
1282                 v3_gva_to_hva(info, linear_addr, &host_addr);
1283             }
1284
1285             V3_Print(info->vm_info, info, "VMX core %u: Host Address of rip = 0x%p\n", info->vcpu_id, (void *)host_addr);
1286
1287             V3_Print(info->vm_info, info, "VMX core %u: Instr (15 bytes) at %p:\n", info->vcpu_id, (void *)host_addr);
1288             v3_dump_mem((uint8_t *)host_addr, 15);
1289
1290             v3_print_stack(info);
1291
1292
1293             v3_print_vmcs();
1294             print_exit_log(info);
1295             return -1;
1296         }
1297
1298         v3_wait_at_barrier(info);
1299
1300
1301         if (info->vm_info->run_state == VM_STOPPED) {
1302             info->core_run_state = CORE_STOPPED;
1303             break;
1304         }
1305 /*
1306         if ((info->num_exits % 5000) == 0) {
1307             V3_Print(info->vm_info, info, "VMX Exit number %d\n", (uint32_t)info->num_exits);
1308         }
1309 */
1310
1311     }
1312
1313 #ifdef V3_CONFIG_PMU_TELEMETRY
1314     v3_pmu_telemetry_end(info);
1315 #endif
1316
1317 #ifdef V3_CONFIG_PWRSTAT_TELEMETRY
1318     v3_pwrstat_telemetry_end(info);
1319 #endif
1320
1321     return 0;
1322 }
1323
1324
1325
1326
1327 #define VMX_FEATURE_CONTROL_MSR     0x0000003a
1328 #define CPUID_VMX_FEATURES 0x00000005  /* LOCK and VMXON */
1329 #define CPUID_1_ECX_VTXFLAG 0x00000020
1330
1331 int v3_is_vmx_capable() {
1332     v3_msr_t feature_msr;
1333     uint32_t eax = 0, ebx = 0, ecx = 0, edx = 0;
1334
1335     v3_cpuid(0x1, &eax, &ebx, &ecx, &edx);
1336
1337     PrintDebug(VM_NONE, VCORE_NONE, "ECX: 0x%x\n", ecx);
1338
1339     if (ecx & CPUID_1_ECX_VTXFLAG) {
1340         v3_get_msr(VMX_FEATURE_CONTROL_MSR, &(feature_msr.hi), &(feature_msr.lo));
1341
1342         PrintDebug(VM_NONE, VCORE_NONE,  "MSRREGlow: 0x%.8x\n", feature_msr.lo);
1343
1344         if ((feature_msr.lo & CPUID_VMX_FEATURES) != CPUID_VMX_FEATURES) {
1345             PrintDebug(VM_NONE, VCORE_NONE,  "VMX is locked -- enable in the BIOS\n");
1346             return 0;
1347         }
1348
1349     } else {
1350         PrintDebug(VM_NONE, VCORE_NONE,  "VMX not supported on this cpu\n");
1351         return 0;
1352     }
1353
1354     return 1;
1355 }
1356
1357
1358 int v3_reset_vmx_vm_core(struct guest_info * core, addr_t rip) {
1359     // init vmcs bios
1360
1361     if ((core->shdw_pg_mode == NESTED_PAGING) &&
1362         (v3_mach_type == V3_VMX_EPT_UG_CPU)) {
1363         // easy
1364         core->rip = 0;
1365         core->segments.cs.selector = rip << 8;
1366         core->segments.cs.limit = 0xffff;
1367         core->segments.cs.base = rip << 12;
1368     } else {
1369         core->vm_regs.rdx = core->vcpu_id;
1370         core->vm_regs.rbx = rip;
1371     }
1372
1373     return 0;
1374 }
1375
1376
1377
1378 void v3_init_vmx_cpu(int cpu_id) {
1379     addr_t vmx_on_region = 0;
1380     extern v3_cpu_arch_t v3_mach_type;
1381     extern v3_cpu_arch_t v3_cpu_types[];
1382
1383     if (v3_mach_type == V3_INVALID_CPU) {
1384         if (v3_init_vmx_hw(&hw_info) == -1) {
1385             PrintError(VM_NONE, VCORE_NONE, "Could not initialize VMX hardware features on cpu %d\n", cpu_id);
1386             return;
1387         }
1388     }
1389
1390     enable_vmx();
1391
1392
1393     // Setup VMXON Region
1394     vmx_on_region = allocate_vmcs();
1395
1396
1397     if (vmx_on(vmx_on_region) == VMX_SUCCESS) {
1398         V3_Print(VM_NONE, VCORE_NONE,  "VMX Enabled\n");
1399         host_vmcs_ptrs[cpu_id] = vmx_on_region;
1400     } else {
1401         V3_Print(VM_NONE, VCORE_NONE,  "VMX already enabled\n");
1402         V3_FreePages((void *)vmx_on_region, 1);
1403     }
1404
1405     PrintDebug(VM_NONE, VCORE_NONE,  "VMXON pointer: 0x%p\n", (void *)host_vmcs_ptrs[cpu_id]);
1406
1407     {
1408         struct vmx_sec_proc_ctrls sec_proc_ctrls;
1409         sec_proc_ctrls.value = v3_vmx_get_ctrl_features(&(hw_info.sec_proc_ctrls));
1410
1411         if (sec_proc_ctrls.enable_ept == 0) {
1412             V3_Print(VM_NONE, VCORE_NONE, "VMX EPT (Nested) Paging not supported\n");
1413             v3_cpu_types[cpu_id] = V3_VMX_CPU;
1414         } else if (sec_proc_ctrls.unrstrct_guest == 0) {
1415             V3_Print(VM_NONE, VCORE_NONE, "VMX EPT (Nested) Paging supported\n");
1416             v3_cpu_types[cpu_id] = V3_VMX_EPT_CPU;
1417         } else {
1418             V3_Print(VM_NONE, VCORE_NONE, "VMX EPT (Nested) Paging + Unrestricted guest supported\n");
1419             v3_cpu_types[cpu_id] = V3_VMX_EPT_UG_CPU;
1420         }
1421     }
1422
1423 }
1424
1425
1426 void v3_deinit_vmx_cpu(int cpu_id) {
1427     extern v3_cpu_arch_t v3_cpu_types[];
1428     v3_cpu_types[cpu_id] = V3_INVALID_CPU;
1429
1430     if (host_vmcs_ptrs[cpu_id] != 0) {
1431         V3_Print(VM_NONE, VCORE_NONE, "Disabling VMX\n");
1432
1433         if (vmx_off() != VMX_SUCCESS) {
1434             PrintError(VM_NONE, VCORE_NONE, "Error executing VMXOFF\n");
1435         }
1436
1437         V3_FreePages((void *)host_vmcs_ptrs[cpu_id], 1);
1438
1439         host_vmcs_ptrs[cpu_id] = 0;
1440     }
1441 }