palacios/src/palacios/svm.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) 2008, Jack Lange <jarusl@cs.northwestern.edu>
  11  * Copyright (c) 2008, 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/svm.h>
  22 #include <palacios/vmm.h>
  23
  24 #include <palacios/vmcb.h>
  25 #include <palacios/vmm_mem.h>
  26 #include <palacios/vmm_paging.h>
  27 #include <palacios/svm_handler.h>
  28
  29 #include <palacios/vmm_debug.h>
  30 #include <palacios/vm_guest_mem.h>
  31
  32 #include <palacios/vmm_decoder.h>
  33 #include <palacios/vmm_string.h>
  34 #include <palacios/vmm_lowlevel.h>
  35 #include <palacios/svm_msr.h>
  36
  37 #include <palacios/vmm_rbtree.h>
  38
  39 #include <palacios/vmm_direct_paging.h>
  40
  41 #include <palacios/vmm_ctrl_regs.h>
  42 #include <palacios/svm_io.h>
  43
  44 #include <palacios/vmm_sprintf.h>
  45
  46
  47 #ifndef V3_CONFIG_DEBUG_SVM
  48 #undef PrintDebug
  49 #define PrintDebug(fmt, args...)
  50 #endif
  51
  52
  53 uint32_t v3_last_exit;
  54
  55 // This is a global pointer to the host's VMCB
  56 static addr_t host_vmcbs[V3_CONFIG_MAX_CPUS] = { [0 ... V3_CONFIG_MAX_CPUS - 1] = 0};
  57
  58
  59
  60 extern void v3_stgi();
  61 extern void v3_clgi();
  62 //extern int v3_svm_launch(vmcb_t * vmcb, struct v3_gprs * vm_regs, uint64_t * fs, uint64_t * gs);
  63 extern int v3_svm_launch(vmcb_t * vmcb, struct v3_gprs * vm_regs, vmcb_t * host_vmcb);
  64
  65
  66 static vmcb_t * Allocate_VMCB() {
  67     vmcb_t * vmcb_page = NULL;
  68     addr_t vmcb_pa = (addr_t)V3_AllocPages(1);
  69
  70     if ((void *)vmcb_pa == NULL) {
  71         PrintError("Error allocating VMCB\n");
  72         return NULL;
  73     }
  74
  75     vmcb_page = (vmcb_t *)V3_VAddr((void *)vmcb_pa);
  76
  77     memset(vmcb_page, 0, 4096);
  78
  79     return vmcb_page;
  80 }
  81
  82
  83
  84 static void Init_VMCB_BIOS(vmcb_t * vmcb, struct guest_info * core) {
  85     vmcb_ctrl_t * ctrl_area = GET_VMCB_CTRL_AREA(vmcb);
  86     vmcb_saved_state_t * guest_state = GET_VMCB_SAVE_STATE_AREA(vmcb);
  87     uint_t i;
  88
  89
  90     //
  91     ctrl_area->svm_instrs.VMRUN = 1;
  92     ctrl_area->svm_instrs.VMMCALL = 1;
  93     ctrl_area->svm_instrs.VMLOAD = 1;
  94     ctrl_area->svm_instrs.VMSAVE = 1;
  95     ctrl_area->svm_instrs.STGI = 1;
  96     ctrl_area->svm_instrs.CLGI = 1;
  97     ctrl_area->svm_instrs.SKINIT = 1;
  98     ctrl_area->svm_instrs.ICEBP = 1;
  99     ctrl_area->svm_instrs.WBINVD = 1;
 100     ctrl_area->svm_instrs.MONITOR = 1;
 101     ctrl_area->svm_instrs.MWAIT_always = 1;
 102     ctrl_area->svm_instrs.MWAIT_if_armed = 1;
 103     ctrl_area->instrs.INVLPGA = 1;
 104     ctrl_area->instrs.CPUID = 1;
 105
 106     ctrl_area->instrs.HLT = 1;
 107
 108 #ifdef V3_CONFIG_TIME_VIRTUALIZE_TSC
 109     ctrl_area->instrs.RDTSC = 1;
 110     ctrl_area->svm_instrs.RDTSCP = 1;
 111 #endif
 112
 113     // guest_state->cr0 = 0x00000001;    // PE
 114
 115     /*
 116       ctrl_area->exceptions.de = 1;
 117       ctrl_area->exceptions.df = 1;
 118
 119       ctrl_area->exceptions.ts = 1;
 120       ctrl_area->exceptions.ss = 1;
 121       ctrl_area->exceptions.ac = 1;
 122       ctrl_area->exceptions.mc = 1;
 123       ctrl_area->exceptions.gp = 1;
 124       ctrl_area->exceptions.ud = 1;
 125       ctrl_area->exceptions.np = 1;
 126       ctrl_area->exceptions.of = 1;
 127
 128       ctrl_area->exceptions.nmi = 1;
 129     */
 130
 131
 132     ctrl_area->instrs.NMI = 1;
 133     ctrl_area->instrs.SMI = 0; // allow SMIs to run in guest
 134     ctrl_area->instrs.INIT = 1;
 135     ctrl_area->instrs.PAUSE = 1;
 136     ctrl_area->instrs.shutdown_evts = 1;
 137
 138
 139     /* DEBUG FOR RETURN CODE */
 140     ctrl_area->exit_code = 1;
 141
 142
 143     /* Setup Guest Machine state */
 144
 145     core->vm_regs.rsp = 0x00;
 146     core->rip = 0xfff0;
 147
 148     core->vm_regs.rdx = 0x00000f00;
 149
 150
 151     core->cpl = 0;
 152
 153     core->ctrl_regs.rflags = 0x00000002; // The reserved bit is always 1
 154     core->ctrl_regs.cr0 = 0x60010010; // Set the WP flag so the memory hooks work in real-mode
 155     core->ctrl_regs.efer |= EFER_MSR_svm_enable;
 156
 157
 158
 159
 160
 161     core->segments.cs.selector = 0xf000;
 162     core->segments.cs.limit = 0xffff;
 163     core->segments.cs.base = 0x0000000f0000LL;
 164
 165     // (raw attributes = 0xf3)
 166     core->segments.cs.type = 0x3;
 167     core->segments.cs.system = 0x1;
 168     core->segments.cs.dpl = 0x3;
 169     core->segments.cs.present = 1;
 170
 171
 172
 173     struct v3_segment * segregs [] = {&(core->segments.ss), &(core->segments.ds),
 174                                       &(core->segments.es), &(core->segments.fs),
 175                                       &(core->segments.gs), NULL};
 176
 177     for ( i = 0; segregs[i] != NULL; i++) {
 178         struct v3_segment * seg = segregs[i];
 179
 180         seg->selector = 0x0000;
 181         //    seg->base = seg->selector << 4;
 182         seg->base = 0x00000000;
 183         seg->limit = ~0u;
 184
 185         // (raw attributes = 0xf3)
 186         seg->type = 0x3;
 187         seg->system = 0x1;
 188         seg->dpl = 0x3;
 189         seg->present = 1;
 190     }
 191
 192     core->segments.gdtr.limit = 0x0000ffff;
 193     core->segments.gdtr.base = 0x0000000000000000LL;
 194     core->segments.idtr.limit = 0x0000ffff;
 195     core->segments.idtr.base = 0x0000000000000000LL;
 196
 197     core->segments.ldtr.selector = 0x0000;
 198     core->segments.ldtr.limit = 0x0000ffff;
 199     core->segments.ldtr.base = 0x0000000000000000LL;
 200     core->segments.tr.selector = 0x0000;
 201     core->segments.tr.limit = 0x0000ffff;
 202     core->segments.tr.base = 0x0000000000000000LL;
 203
 204
 205     core->dbg_regs.dr6 = 0x00000000ffff0ff0LL;
 206     core->dbg_regs.dr7 = 0x0000000000000400LL;
 207
 208
 209     ctrl_area->IOPM_BASE_PA = (addr_t)V3_PAddr(core->vm_info->io_map.arch_data);
 210     ctrl_area->instrs.IOIO_PROT = 1;
 211
 212     ctrl_area->MSRPM_BASE_PA = (addr_t)V3_PAddr(core->vm_info->msr_map.arch_data);
 213     ctrl_area->instrs.MSR_PROT = 1;
 214
 215
 216     PrintDebug("Exiting on interrupts\n");
 217     ctrl_area->guest_ctrl.V_INTR_MASKING = 1;
 218     ctrl_area->instrs.INTR = 1;
 219
 220
 221     v3_hook_msr(core->vm_info, EFER_MSR,
 222                 &v3_handle_efer_read,
 223                 &v3_handle_efer_write,
 224                 core);
 225
 226     if (core->shdw_pg_mode == SHADOW_PAGING) {
 227         PrintDebug("Creating initial shadow page table\n");
 228
 229         /* JRL: This is a performance killer, and a simplistic solution */
 230         /* We need to fix this */
 231         ctrl_area->TLB_CONTROL = 1;
 232         ctrl_area->guest_ASID = 1;
 233
 234
 235         if (v3_init_passthrough_pts(core) == -1) {
 236             PrintError("Could not initialize passthrough page tables\n");
 237             return ;
 238         }
 239
 240
 241         core->shdw_pg_state.guest_cr0 = 0x0000000000000010LL;
 242         PrintDebug("Created\n");
 243
 244         core->ctrl_regs.cr0 |= 0x80000000;
 245         core->ctrl_regs.cr3 = core->direct_map_pt;
 246
 247         ctrl_area->cr_reads.cr0 = 1;
 248         ctrl_area->cr_writes.cr0 = 1;
 249         //ctrl_area->cr_reads.cr4 = 1;
 250         ctrl_area->cr_writes.cr4 = 1;
 251         ctrl_area->cr_reads.cr3 = 1;
 252         ctrl_area->cr_writes.cr3 = 1;
 253
 254
 255
 256         ctrl_area->instrs.INVLPG = 1;
 257
 258         ctrl_area->exceptions.pf = 1;
 259
 260         guest_state->g_pat = 0x7040600070406ULL;
 261
 262
 263
 264     } else if (core->shdw_pg_mode == NESTED_PAGING) {
 265         // Flush the TLB on entries/exits
 266         ctrl_area->TLB_CONTROL = 1;
 267         ctrl_area->guest_ASID = 1;
 268
 269         // Enable Nested Paging
 270         ctrl_area->NP_ENABLE = 1;
 271
 272         PrintDebug("NP_Enable at 0x%p\n", (void *)&(ctrl_area->NP_ENABLE));
 273
 274         // Set the Nested Page Table pointer
 275         if (v3_init_passthrough_pts(core) == -1) {
 276             PrintError("Could not initialize Nested page tables\n");
 277             return ;
 278         }
 279
 280         ctrl_area->N_CR3 = core->direct_map_pt;
 281
 282         guest_state->g_pat = 0x7040600070406ULL;
 283     }
 284
 285     /* tell the guest that we don't support SVM */
 286     v3_hook_msr(core->vm_info, SVM_VM_CR_MSR,
 287         &v3_handle_vm_cr_read,
 288         &v3_handle_vm_cr_write,
 289         core);
 290 }
 291
 292
 293 int v3_init_svm_vmcb(struct guest_info * core, v3_vm_class_t vm_class) {
 294
 295     PrintDebug("Allocating VMCB\n");
 296     core->vmm_data = (void *)Allocate_VMCB();
 297
 298     if (core->vmm_data == NULL) {
 299         PrintError("Could not allocate VMCB, Exiting...\n");
 300         return -1;
 301     }
 302
 303     if (vm_class == V3_PC_VM) {
 304         PrintDebug("Initializing VMCB (addr=%p)\n", (void *)core->vmm_data);
 305         Init_VMCB_BIOS((vmcb_t*)(core->vmm_data), core);
 306     } else {
 307         PrintError("Invalid VM class\n");
 308         return -1;
 309     }
 310
 311     return 0;
 312 }
 313
 314
 315 int v3_deinit_svm_vmcb(struct guest_info * core) {
 316     V3_FreePages(V3_PAddr(core->vmm_data), 1);
 317     return 0;
 318 }
 319
 320
 321 static int update_irq_exit_state(struct guest_info * info) {
 322     vmcb_ctrl_t * guest_ctrl = GET_VMCB_CTRL_AREA((vmcb_t*)(info->vmm_data));
 323
 324     // Fix for QEMU bug using EVENTINJ as an internal cache
 325     guest_ctrl->EVENTINJ.valid = 0;
 326
 327     if ((info->intr_core_state.irq_pending == 1) && (guest_ctrl->guest_ctrl.V_IRQ == 0)) {
 328
 329 #ifdef V3_CONFIG_DEBUG_INTERRUPTS
 330         PrintDebug("INTAK cycle completed for irq %d\n", info->intr_core_state.irq_vector);
 331 #endif
 332
 333         info->intr_core_state.irq_started = 1;
 334         info->intr_core_state.irq_pending = 0;
 335
 336         v3_injecting_intr(info, info->intr_core_state.irq_vector, V3_EXTERNAL_IRQ);
 337     }
 338
 339     if ((info->intr_core_state.irq_started == 1) && (guest_ctrl->exit_int_info.valid == 0)) {
 340 #ifdef V3_CONFIG_DEBUG_INTERRUPTS
 341         PrintDebug("Interrupt %d taken by guest\n", info->intr_core_state.irq_vector);
 342 #endif
 343
 344         // Interrupt was taken fully vectored
 345         info->intr_core_state.irq_started = 0;
 346
 347     } else if ((info->intr_core_state.irq_started == 1) && (guest_ctrl->exit_int_info.valid == 1)) {
 348 #ifdef V3_CONFIG_DEBUG_INTERRUPTS
 349         PrintDebug("EXIT INT INFO is set (vec=%d)\n", guest_ctrl->exit_int_info.vector);
 350 #endif
 351     }
 352
 353     return 0;
 354 }
 355
 356
 357 static int update_irq_entry_state(struct guest_info * info) {
 358     vmcb_ctrl_t * guest_ctrl = GET_VMCB_CTRL_AREA((vmcb_t*)(info->vmm_data));
 359
 360
 361     if (info->intr_core_state.irq_pending == 0) {
 362         guest_ctrl->guest_ctrl.V_IRQ = 0;
 363         guest_ctrl->guest_ctrl.V_INTR_VECTOR = 0;
 364     }
 365
 366     if (v3_excp_pending(info)) {
 367         uint_t excp = v3_get_excp_number(info);
 368
 369         guest_ctrl->EVENTINJ.type = SVM_INJECTION_EXCEPTION;
 370
 371         if (info->excp_state.excp_error_code_valid) {
 372             guest_ctrl->EVENTINJ.error_code = info->excp_state.excp_error_code;
 373             guest_ctrl->EVENTINJ.ev = 1;
 374 #ifdef V3_CONFIG_DEBUG_INTERRUPTS
 375             PrintDebug("Injecting exception %d with error code %x\n", excp, guest_ctrl->EVENTINJ.error_code);
 376 #endif
 377         }
 378
 379         guest_ctrl->EVENTINJ.vector = excp;
 380
 381         guest_ctrl->EVENTINJ.valid = 1;
 382
 383 #ifdef V3_CONFIG_DEBUG_INTERRUPTS
 384         PrintDebug("<%d> Injecting Exception %d (CR2=%p) (EIP=%p)\n",
 385                    (int)info->num_exits,
 386                    guest_ctrl->EVENTINJ.vector,
 387                    (void *)(addr_t)info->ctrl_regs.cr2,
 388                    (void *)(addr_t)info->rip);
 389 #endif
 390
 391         v3_injecting_excp(info, excp);
 392     } else if (info->intr_core_state.irq_started == 1) {
 393 #ifdef V3_CONFIG_DEBUG_INTERRUPTS
 394         PrintDebug("IRQ pending from previous injection\n");
 395 #endif
 396         guest_ctrl->guest_ctrl.V_IRQ = 1;
 397         guest_ctrl->guest_ctrl.V_INTR_VECTOR = info->intr_core_state.irq_vector;
 398         guest_ctrl->guest_ctrl.V_IGN_TPR = 1;
 399         guest_ctrl->guest_ctrl.V_INTR_PRIO = 0xf;
 400
 401     } else {
 402         switch (v3_intr_pending(info)) {
 403             case V3_EXTERNAL_IRQ: {
 404                 uint32_t irq = v3_get_intr(info);
 405
 406                 guest_ctrl->guest_ctrl.V_IRQ = 1;
 407                 guest_ctrl->guest_ctrl.V_INTR_VECTOR = irq;
 408                 guest_ctrl->guest_ctrl.V_IGN_TPR = 1;
 409                 guest_ctrl->guest_ctrl.V_INTR_PRIO = 0xf;
 410
 411 #ifdef V3_CONFIG_DEBUG_INTERRUPTS
 412                 PrintDebug("Injecting Interrupt %d (EIP=%p)\n",
 413                            guest_ctrl->guest_ctrl.V_INTR_VECTOR,
 414                            (void *)(addr_t)info->rip);
 415 #endif
 416
 417                 info->intr_core_state.irq_pending = 1;
 418                 info->intr_core_state.irq_vector = irq;
 419
 420                 break;
 421             }
 422             case V3_NMI:
 423                 guest_ctrl->EVENTINJ.type = SVM_INJECTION_NMI;
 424                 break;
 425             case V3_SOFTWARE_INTR:
 426                 guest_ctrl->EVENTINJ.type = SVM_INJECTION_SOFT_INTR;
 427                 break;
 428             case V3_VIRTUAL_IRQ:
 429                 guest_ctrl->EVENTINJ.type = SVM_INJECTION_IRQ;
 430                 break;
 431
 432             case V3_INVALID_INTR:
 433             default:
 434                 break;
 435         }
 436
 437     }
 438
 439     return 0;
 440 }
 441
 442
 443 /*
 444  * CAUTION and DANGER!!!
 445  *
 446  * The VMCB CANNOT(!!) be accessed outside of the clgi/stgi calls inside this function
 447  * When exectuing a symbiotic call, the VMCB WILL be overwritten, so any dependencies
 448  * on its contents will cause things to break. The contents at the time of the exit WILL
 449  * change before the exit handler is executed.
 450  */
 451 int v3_svm_enter(struct guest_info * info) {
 452     vmcb_ctrl_t * guest_ctrl = GET_VMCB_CTRL_AREA((vmcb_t*)(info->vmm_data));
 453     vmcb_saved_state_t * guest_state = GET_VMCB_SAVE_STATE_AREA((vmcb_t*)(info->vmm_data));
 454     addr_t exit_code = 0, exit_info1 = 0, exit_info2 = 0;
 455
 456     // Conditionally yield the CPU if the timeslice has expired
 457     v3_yield_cond(info);
 458
 459     // Perform any additional yielding needed for time adjustment
 460     v3_adjust_time(info);
 461
 462     // disable global interrupts for vm state transition
 463     v3_clgi();
 464
 465     // Update timer devices prior to entering VM.
 466     v3_update_timers(info);
 467
 468     // Synchronize the guest state to the VMCB
 469     guest_state->cr0 = info->ctrl_regs.cr0;
 470     guest_state->cr2 = info->ctrl_regs.cr2;
 471     guest_state->cr3 = info->ctrl_regs.cr3;
 472     guest_state->cr4 = info->ctrl_regs.cr4;
 473     guest_state->dr6 = info->dbg_regs.dr6;
 474     guest_state->dr7 = info->dbg_regs.dr7;
 475     guest_ctrl->guest_ctrl.V_TPR = info->ctrl_regs.cr8 & 0xff;
 476     guest_state->rflags = info->ctrl_regs.rflags;
 477     guest_state->efer = info->ctrl_regs.efer;
 478
 479     guest_state->cpl = info->cpl;
 480
 481     v3_set_vmcb_segments((vmcb_t*)(info->vmm_data), &(info->segments));
 482
 483     guest_state->rax = info->vm_regs.rax;
 484     guest_state->rip = info->rip;
 485     guest_state->rsp = info->vm_regs.rsp;
 486
 487 #ifdef V3_CONFIG_SYMCALL
 488     if (info->sym_core_state.symcall_state.sym_call_active == 0) {
 489         update_irq_entry_state(info);
 490     }
 491 #else
 492     update_irq_entry_state(info);
 493 #endif
 494
 495
 496     /* ** */
 497
 498     /*
 499       PrintDebug("SVM Entry to CS=%p  rip=%p...\n",
 500       (void *)(addr_t)info->segments.cs.base,
 501       (void *)(addr_t)info->rip);
 502     */
 503
 504 #ifdef V3_CONFIG_SYMCALL
 505     if (info->sym_core_state.symcall_state.sym_call_active == 1) {
 506         if (guest_ctrl->guest_ctrl.V_IRQ == 1) {
 507             V3_Print("!!! Injecting Interrupt during Sym call !!!\n");
 508         }
 509     }
 510 #endif
 511
 512     v3_time_enter_vm(info);
 513     guest_ctrl->TSC_OFFSET = v3_tsc_host_offset(&info->time_state);
 514
 515     //V3_Print("Calling v3_svm_launch\n");
 516
 517     v3_svm_launch((vmcb_t *)V3_PAddr(info->vmm_data), &(info->vm_regs), (vmcb_t *)host_vmcbs[V3_Get_CPU()]);
 518
 519     //V3_Print("SVM Returned: Exit Code: %x, guest_rip=%lx\n", (uint32_t)(guest_ctrl->exit_code), (unsigned long)guest_state->rip);
 520
 521     v3_last_exit = (uint32_t)(guest_ctrl->exit_code);
 522
 523     // Immediate exit from VM time bookkeeping
 524     v3_time_exit_vm(info);
 525
 526     info->num_exits++;
 527
 528     // Save Guest state from VMCB
 529     info->rip = guest_state->rip;
 530     info->vm_regs.rsp = guest_state->rsp;
 531     info->vm_regs.rax = guest_state->rax;
 532
 533     info->cpl = guest_state->cpl;
 534
 535     info->ctrl_regs.cr0 = guest_state->cr0;
 536     info->ctrl_regs.cr2 = guest_state->cr2;
 537     info->ctrl_regs.cr3 = guest_state->cr3;
 538     info->ctrl_regs.cr4 = guest_state->cr4;
 539     info->dbg_regs.dr6 = guest_state->dr6;
 540     info->dbg_regs.dr7 = guest_state->dr7;
 541     info->ctrl_regs.cr8 = guest_ctrl->guest_ctrl.V_TPR;
 542     info->ctrl_regs.rflags = guest_state->rflags;
 543     info->ctrl_regs.efer = guest_state->efer;
 544
 545     v3_get_vmcb_segments((vmcb_t*)(info->vmm_data), &(info->segments));
 546     info->cpu_mode = v3_get_vm_cpu_mode(info);
 547     info->mem_mode = v3_get_vm_mem_mode(info);
 548     /* ** */
 549
 550
 551     // save exit info here
 552     exit_code = guest_ctrl->exit_code;
 553     exit_info1 = guest_ctrl->exit_info1;
 554     exit_info2 = guest_ctrl->exit_info2;
 555
 556
 557 #ifdef V3_CONFIG_SYMCALL
 558     if (info->sym_core_state.symcall_state.sym_call_active == 0) {
 559         update_irq_exit_state(info);
 560     }
 561 #else
 562     update_irq_exit_state(info);
 563 #endif
 564
 565
 566     // reenable global interrupts after vm exit
 567     v3_stgi();
 568
 569
 570     // Conditionally yield the CPU if the timeslice has expired
 571     v3_yield_cond(info);
 572
 573
 574
 575     if (v3_handle_svm_exit(info, exit_code, exit_info1, exit_info2) != 0) {
 576         PrintError("Error in SVM exit handler\n");
 577         PrintError("  last exit was %d\n", v3_last_exit);
 578         return -1;
 579     }
 580
 581
 582     return 0;
 583 }
 584
 585
 586 int v3_start_svm_guest(struct guest_info * info) {
 587     //    vmcb_saved_state_t * guest_state = GET_VMCB_SAVE_STATE_AREA((vmcb_t*)(info->vmm_data));
 588     //  vmcb_ctrl_t * guest_ctrl = GET_VMCB_CTRL_AREA((vmcb_t*)(info->vmm_data));
 589
 590     PrintDebug("Starting SVM core %u (on logical core %u)\n", info->vcpu_id, info->pcpu_id);
 591
 592     if (info->vcpu_id == 0) {
 593         info->core_run_state = CORE_RUNNING;
 594         info->vm_info->run_state = VM_RUNNING;
 595     } else  {
 596         PrintDebug("SVM core %u (on %u): Waiting for core initialization\n", info->vcpu_id, info->pcpu_id);
 597
 598         while (info->core_run_state == CORE_STOPPED) {
 599             v3_yield(info);
 600             //PrintDebug("SVM core %u: still waiting for INIT\n", info->vcpu_id);
 601         }
 602
 603         PrintDebug("SVM core %u(on %u) initialized\n", info->vcpu_id, info->pcpu_id);
 604     }
 605
 606     PrintDebug("SVM core %u(on %u): I am starting at CS=0x%x (base=0x%p, limit=0x%x),  RIP=0x%p\n",
 607                info->vcpu_id, info->pcpu_id,
 608                info->segments.cs.selector, (void *)(info->segments.cs.base),
 609                info->segments.cs.limit, (void *)(info->rip));
 610
 611
 612
 613     PrintDebug("SVM core %u: Launching SVM VM (vmcb=%p) (on cpu %u)\n",
 614                info->vcpu_id, (void *)info->vmm_data, info->pcpu_id);
 615     //PrintDebugVMCB((vmcb_t*)(info->vmm_data));
 616
 617     v3_start_time(info);
 618
 619     while (1) {
 620
 621         if (info->vm_info->run_state == VM_STOPPED) {
 622             info->core_run_state = CORE_STOPPED;
 623             break;
 624         }
 625
 626         if (v3_svm_enter(info) == -1) {
 627             vmcb_ctrl_t * guest_ctrl = GET_VMCB_CTRL_AREA((vmcb_t*)(info->vmm_data));
 628             addr_t host_addr;
 629             addr_t linear_addr = 0;
 630
 631             info->vm_info->run_state = VM_ERROR;
 632
 633             V3_Print("SVM core %u: SVM ERROR!!\n", info->vcpu_id);
 634
 635             v3_print_guest_state(info);
 636
 637             V3_Print("SVM core %u: SVM Exit Code: %p\n", info->vcpu_id, (void *)(addr_t)guest_ctrl->exit_code);
 638
 639             V3_Print("SVM core %u: exit_info1 low = 0x%.8x\n", info->vcpu_id, *(uint_t*)&(guest_ctrl->exit_info1));
 640             V3_Print("SVM core %u: exit_info1 high = 0x%.8x\n", info->vcpu_id, *(uint_t *)(((uchar_t *)&(guest_ctrl->exit_info1)) + 4));
 641
 642             V3_Print("SVM core %u: exit_info2 low = 0x%.8x\n", info->vcpu_id, *(uint_t*)&(guest_ctrl->exit_info2));
 643             V3_Print("SVM core %u: exit_info2 high = 0x%.8x\n", info->vcpu_id, *(uint_t *)(((uchar_t *)&(guest_ctrl->exit_info2)) + 4));
 644
 645             linear_addr = get_addr_linear(info, info->rip, &(info->segments.cs));
 646
 647             if (info->mem_mode == PHYSICAL_MEM) {
 648                 v3_gpa_to_hva(info, linear_addr, &host_addr);
 649             } else if (info->mem_mode == VIRTUAL_MEM) {
 650                 v3_gva_to_hva(info, linear_addr, &host_addr);
 651             }
 652
 653             V3_Print("SVM core %u: Host Address of rip = 0x%p\n", info->vcpu_id, (void *)host_addr);
 654
 655             V3_Print("SVM core %u: Instr (15 bytes) at %p:\n", info->vcpu_id, (void *)host_addr);
 656             v3_dump_mem((uint8_t *)host_addr, 15);
 657
 658             v3_print_stack(info);
 659
 660             break;
 661         }
 662
 663
 664         if (info->vm_info->run_state == VM_STOPPED) {
 665             info->core_run_state = CORE_STOPPED;
 666             break;
 667         }
 668
 669
 670 /*
 671         if ((info->num_exits % 5000) == 0) {
 672             V3_Print("SVM Exit number %d\n", (uint32_t)info->num_exits);
 673         }
 674 */
 675
 676     }
 677
 678     // Need to take down the other cores on error...
 679
 680     return 0;
 681 }
 682
 683
 684
 685
 686
 687 /* Checks machine SVM capability */
 688 /* Implemented from: AMD Arch Manual 3, sect 15.4 */
 689 int v3_is_svm_capable() {
 690     uint_t vm_cr_low = 0, vm_cr_high = 0;
 691     uint32_t eax = 0, ebx = 0, ecx = 0, edx = 0;
 692
 693     v3_cpuid(CPUID_EXT_FEATURE_IDS, &eax, &ebx, &ecx, &edx);
 694
 695     PrintDebug("CPUID_EXT_FEATURE_IDS_ecx=0x%x\n", ecx);
 696
 697     if ((ecx & CPUID_EXT_FEATURE_IDS_ecx_svm_avail) == 0) {
 698       V3_Print("SVM Not Available\n");
 699       return 0;
 700     }  else {
 701         v3_get_msr(SVM_VM_CR_MSR, &vm_cr_high, &vm_cr_low);
 702
 703         PrintDebug("SVM_VM_CR_MSR = 0x%x 0x%x\n", vm_cr_high, vm_cr_low);
 704
 705         if ((vm_cr_low & SVM_VM_CR_MSR_svmdis) == 1) {
 706             V3_Print("SVM is available but is disabled.\n");
 707
 708             v3_cpuid(CPUID_SVM_REV_AND_FEATURE_IDS, &eax, &ebx, &ecx, &edx);
 709
 710             PrintDebug("CPUID_SVM_REV_AND_FEATURE_IDS_edx=0x%x\n", edx);
 711
 712             if ((edx & CPUID_SVM_REV_AND_FEATURE_IDS_edx_svml) == 0) {
 713                 V3_Print("SVM BIOS Disabled, not unlockable\n");
 714             } else {
 715                 V3_Print("SVM is locked with a key\n");
 716             }
 717             return 0;
 718
 719         } else {
 720             V3_Print("SVM is available and  enabled.\n");
 721
 722             v3_cpuid(CPUID_SVM_REV_AND_FEATURE_IDS, &eax, &ebx, &ecx, &edx);
 723             PrintDebug("CPUID_SVM_REV_AND_FEATURE_IDS_eax=0x%x\n", eax);
 724             PrintDebug("CPUID_SVM_REV_AND_FEATURE_IDS_ebx=0x%x\n", ebx);
 725             PrintDebug("CPUID_SVM_REV_AND_FEATURE_IDS_ecx=0x%x\n", ecx);
 726             PrintDebug("CPUID_SVM_REV_AND_FEATURE_IDS_edx=0x%x\n", edx);
 727
 728             return 1;
 729         }
 730     }
 731 }
 732
 733 static int has_svm_nested_paging() {
 734     uint32_t eax = 0, ebx = 0, ecx = 0, edx = 0;
 735
 736     v3_cpuid(CPUID_SVM_REV_AND_FEATURE_IDS, &eax, &ebx, &ecx, &edx);
 737
 738     //PrintDebug("CPUID_EXT_FEATURE_IDS_edx=0x%x\n", edx);
 739
 740     if ((edx & CPUID_SVM_REV_AND_FEATURE_IDS_edx_np) == 0) {
 741         V3_Print("SVM Nested Paging not supported\n");
 742         return 0;
 743     } else {
 744         V3_Print("SVM Nested Paging supported\n");
 745         return 1;
 746     }
 747 }
 748
 749
 750 void v3_init_svm_cpu(int cpu_id) {
 751     reg_ex_t msr;
 752     extern v3_cpu_arch_t v3_cpu_types[];
 753
 754     // Enable SVM on the CPU
 755     v3_get_msr(EFER_MSR, &(msr.e_reg.high), &(msr.e_reg.low));
 756     msr.e_reg.low |= EFER_MSR_svm_enable;
 757     v3_set_msr(EFER_MSR, 0, msr.e_reg.low);
 758
 759     V3_Print("SVM Enabled\n");
 760
 761     // Setup the host state save area
 762     host_vmcbs[cpu_id] = (addr_t)V3_AllocPages(4);
 763
 764     /* 64-BIT-ISSUE */
 765     //  msr.e_reg.high = 0;
 766     //msr.e_reg.low = (uint_t)host_vmcb;
 767     msr.r_reg = host_vmcbs[cpu_id];
 768
 769     PrintDebug("Host State being saved at %p\n", (void *)host_vmcbs[cpu_id]);
 770     v3_set_msr(SVM_VM_HSAVE_PA_MSR, msr.e_reg.high, msr.e_reg.low);
 771
 772
 773     if (has_svm_nested_paging() == 1) {
 774         v3_cpu_types[cpu_id] = V3_SVM_REV3_CPU;
 775     } else {
 776         v3_cpu_types[cpu_id] = V3_SVM_CPU;
 777     }
 778 }
 779
 780
 781
 782 void v3_deinit_svm_cpu(int cpu_id) {
 783     reg_ex_t msr;
 784     extern v3_cpu_arch_t v3_cpu_types[];
 785
 786     // reset SVM_VM_HSAVE_PA_MSR
 787     // Does setting it to NULL disable??
 788     msr.r_reg = 0;
 789     v3_set_msr(SVM_VM_HSAVE_PA_MSR, msr.e_reg.high, msr.e_reg.low);
 790
 791     // Disable SVM?
 792     v3_get_msr(EFER_MSR, &(msr.e_reg.high), &(msr.e_reg.low));
 793     msr.e_reg.low &= ~EFER_MSR_svm_enable;
 794     v3_set_msr(EFER_MSR, 0, msr.e_reg.low);
 795
 796     v3_cpu_types[cpu_id] = V3_INVALID_CPU;
 797
 798     V3_FreePages((void *)host_vmcbs[cpu_id], 4);
 799
 800     V3_Print("Host CPU %d host area freed, and SVM disabled\n", cpu_id);
 801     return;
 802 }
 803
 804
 805
 806
 807
 808
 809
 810
 811
 812
 813
 814
 815
 816
 817
 818
 819
 820
 821
 822
 823
 824
 825
 826
 827
 828
 829
 830
 831
 832
 833
 834
 835
 836
 837
 838
 839
 840
 841
 842
 843
 844
 845
 846
 847
 848
 849
 850
 851
 852
 853 #if 0
 854 /*
 855  * Test VMSAVE/VMLOAD Latency
 856  */
 857 #define vmsave ".byte 0x0F,0x01,0xDB ; "
 858 #define vmload ".byte 0x0F,0x01,0xDA ; "
 859 {
 860     uint32_t start_lo, start_hi;
 861     uint32_t end_lo, end_hi;
 862     uint64_t start, end;
 863
 864     __asm__ __volatile__ (
 865                           "rdtsc ; "
 866                           "movl %%eax, %%esi ; "
 867                           "movl %%edx, %%edi ; "
 868                           "movq  %%rcx, %%rax ; "
 869                           vmsave
 870                           "rdtsc ; "
 871                           : "=D"(start_hi), "=S"(start_lo), "=a"(end_lo),"=d"(end_hi)
 872                           : "c"(host_vmcb[cpu_id]), "0"(0), "1"(0), "2"(0), "3"(0)
 873                           );
 874
 875     start = start_hi;
 876     start <<= 32;
 877     start += start_lo;
 878
 879     end = end_hi;
 880     end <<= 32;
 881     end += end_lo;
 882
 883     PrintDebug("VMSave Cycle Latency: %d\n", (uint32_t)(end - start));
 884
 885     __asm__ __volatile__ (
 886                           "rdtsc ; "
 887                           "movl %%eax, %%esi ; "
 888                           "movl %%edx, %%edi ; "
 889                           "movq  %%rcx, %%rax ; "
 890                           vmload
 891                           "rdtsc ; "
 892                           : "=D"(start_hi), "=S"(start_lo), "=a"(end_lo),"=d"(end_hi)
 893                               : "c"(host_vmcb[cpu_id]), "0"(0), "1"(0), "2"(0), "3"(0)
 894                               );
 895
 896         start = start_hi;
 897         start <<= 32;
 898         start += start_lo;
 899
 900         end = end_hi;
 901         end <<= 32;
 902         end += end_lo;
 903
 904
 905         PrintDebug("VMLoad Cycle Latency: %d\n", (uint32_t)(end - start));
 906     }
 907     /* End Latency Test */
 908
 909 #endif
 910
 911
 912
 913
 914
 915
 916
 917 #if 0
 918 void Init_VMCB_pe(vmcb_t *vmcb, struct guest_info vm_info) {
 919   vmcb_ctrl_t * ctrl_area = GET_VMCB_CTRL_AREA(vmcb);
 920   vmcb_saved_state_t * guest_state = GET_VMCB_SAVE_STATE_AREA(vmcb);
 921   uint_t i = 0;
 922
 923
 924   guest_state->rsp = vm_info.vm_regs.rsp;
 925   guest_state->rip = vm_info.rip;
 926
 927
 928   /* I pretty much just gutted this from TVMM */
 929   /* Note: That means its probably wrong */
 930
 931   // set the segment registers to mirror ours
 932   guest_state->cs.selector = 1<<3;
 933   guest_state->cs.attrib.fields.type = 0xa; // Code segment+read
 934   guest_state->cs.attrib.fields.S = 1;
 935   guest_state->cs.attrib.fields.P = 1;
 936   guest_state->cs.attrib.fields.db = 1;
 937   guest_state->cs.attrib.fields.G = 1;
 938   guest_state->cs.limit = 0xfffff;
 939   guest_state->cs.base = 0;
 940
 941   struct vmcb_selector *segregs [] = {&(guest_state->ss), &(guest_state->ds), &(guest_state->es), &(guest_state->fs), &(guest_state->gs), NULL};
 942   for ( i = 0; segregs[i] != NULL; i++) {
 943     struct vmcb_selector * seg = segregs[i];
 944
 945     seg->selector = 2<<3;
 946     seg->attrib.fields.type = 0x2; // Data Segment+read/write
 947     seg->attrib.fields.S = 1;
 948     seg->attrib.fields.P = 1;
 949     seg->attrib.fields.db = 1;
 950     seg->attrib.fields.G = 1;
 951     seg->limit = 0xfffff;
 952     seg->base = 0;
 953   }
 954
 955
 956   {
 957     /* JRL THIS HAS TO GO */
 958
 959     //    guest_state->tr.selector = GetTR_Selector();
 960     guest_state->tr.attrib.fields.type = 0x9;
 961     guest_state->tr.attrib.fields.P = 1;
 962     // guest_state->tr.limit = GetTR_Limit();
 963     //guest_state->tr.base = GetTR_Base();// - 0x2000;
 964     /* ** */
 965   }
 966
 967
 968   /* ** */
 969
 970
 971   guest_state->efer |= EFER_MSR_svm_enable;
 972   guest_state->rflags = 0x00000002; // The reserved bit is always 1
 973   ctrl_area->svm_instrs.VMRUN = 1;
 974   guest_state->cr0 = 0x00000001;    // PE
 975   ctrl_area->guest_ASID = 1;
 976
 977
 978   //  guest_state->cpl = 0;
 979
 980
 981
 982   // Setup exits
 983
 984   ctrl_area->cr_writes.cr4 = 1;
 985
 986   ctrl_area->exceptions.de = 1;
 987   ctrl_area->exceptions.df = 1;
 988   ctrl_area->exceptions.pf = 1;
 989   ctrl_area->exceptions.ts = 1;
 990   ctrl_area->exceptions.ss = 1;
 991   ctrl_area->exceptions.ac = 1;
 992   ctrl_area->exceptions.mc = 1;
 993   ctrl_area->exceptions.gp = 1;
 994   ctrl_area->exceptions.ud = 1;
 995   ctrl_area->exceptions.np = 1;
 996   ctrl_area->exceptions.of = 1;
 997   ctrl_area->exceptions.nmi = 1;
 998
 999
1000
1001   ctrl_area->instrs.IOIO_PROT = 1;
1002   ctrl_area->IOPM_BASE_PA = (uint_t)V3_AllocPages(3);
1003
1004   {
1005     reg_ex_t tmp_reg;
1006     tmp_reg.r_reg = ctrl_area->IOPM_BASE_PA;
1007     memset((void*)(tmp_reg.e_reg.low), 0xffffffff, PAGE_SIZE * 2);
1008   }
1009
1010   ctrl_area->instrs.INTR = 1;
1011
1012
1013   {
1014     char gdt_buf[6];
1015     char idt_buf[6];
1016
1017     memset(gdt_buf, 0, 6);
1018     memset(idt_buf, 0, 6);
1019
1020
1021     uint_t gdt_base, idt_base;
1022     ushort_t gdt_limit, idt_limit;
1023
1024     GetGDTR(gdt_buf);
1025     gdt_base = *(ulong_t*)((uchar_t*)gdt_buf + 2) & 0xffffffff;
1026     gdt_limit = *(ushort_t*)(gdt_buf) & 0xffff;
1027     PrintDebug("GDT: base: %x, limit: %x\n", gdt_base, gdt_limit);
1028
1029     GetIDTR(idt_buf);
1030     idt_base = *(ulong_t*)(idt_buf + 2) & 0xffffffff;
1031     idt_limit = *(ushort_t*)(idt_buf) & 0xffff;
1032     PrintDebug("IDT: base: %x, limit: %x\n",idt_base, idt_limit);
1033
1034
1035     // gdt_base -= 0x2000;
1036     //idt_base -= 0x2000;
1037
1038     guest_state->gdtr.base = gdt_base;
1039     guest_state->gdtr.limit = gdt_limit;
1040     guest_state->idtr.base = idt_base;
1041     guest_state->idtr.limit = idt_limit;
1042
1043
1044   }
1045
1046
1047   // also determine if CPU supports nested paging
1048   /*
1049   if (vm_info.page_tables) {
1050     //   if (0) {
1051     // Flush the TLB on entries/exits
1052     ctrl_area->TLB_CONTROL = 1;
1053
1054     // Enable Nested Paging
1055     ctrl_area->NP_ENABLE = 1;
1056
1057     PrintDebug("NP_Enable at 0x%x\n", &(ctrl_area->NP_ENABLE));
1058
1059         // Set the Nested Page Table pointer
1060     ctrl_area->N_CR3 |= ((addr_t)vm_info.page_tables & 0xfffff000);
1061
1062
1063     //   ctrl_area->N_CR3 = Get_CR3();
1064     // guest_state->cr3 |= (Get_CR3() & 0xfffff000);
1065
1066     guest_state->g_pat = 0x7040600070406ULL;
1067
1068     PrintDebug("Set Nested CR3: lo: 0x%x  hi: 0x%x\n", (uint_t)*(&(ctrl_area->N_CR3)), (uint_t)*((unsigned char *)&(ctrl_area->N_CR3) + 4));
1069     PrintDebug("Set Guest CR3: lo: 0x%x  hi: 0x%x\n", (uint_t)*(&(guest_state->cr3)), (uint_t)*((unsigned char *)&(guest_state->cr3) + 4));
1070     // Enable Paging
1071     //    guest_state->cr0 |= 0x80000000;
1072   }
1073   */
1074
1075 }
1076
1077
1078
1079
1080
1081 #endif
1082
1083