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 #include <palacios/vmm_barrier.h>
  39
  40 #ifdef V3_CONFIG_CHECKPOINT
  41 #include <palacios/vmm_checkpoint.h>
  42 #endif
  43
  44 #include <palacios/vmm_direct_paging.h>
  45
  46 #include <palacios/vmm_ctrl_regs.h>
  47 #include <palacios/svm_io.h>
  48
  49 #include <palacios/vmm_sprintf.h>
  50
  51
  52 #ifndef V3_CONFIG_DEBUG_SVM
  53 #undef PrintDebug
  54 #define PrintDebug(fmt, args...)
  55 #endif
  56
  57
  58 uint32_t v3_last_exit;
  59
  60 // This is a global pointer to the host's VMCB
  61 static addr_t host_vmcbs[V3_CONFIG_MAX_CPUS] = { [0 ... V3_CONFIG_MAX_CPUS - 1] = 0};
  62
  63
  64
  65 extern void v3_stgi();
  66 extern void v3_clgi();
  67 //extern int v3_svm_launch(vmcb_t * vmcb, struct v3_gprs * vm_regs, uint64_t * fs, uint64_t * gs);
  68 extern int v3_svm_launch(vmcb_t * vmcb, struct v3_gprs * vm_regs, vmcb_t * host_vmcb);
  69
  70
  71 static vmcb_t * Allocate_VMCB() {
  72     vmcb_t * vmcb_page = NULL;
  73     addr_t vmcb_pa = (addr_t)V3_AllocPages(1);
  74
  75     if ((void *)vmcb_pa == NULL) {
  76         PrintError("Error allocating VMCB\n");
  77         return NULL;
  78     }
  79
  80     vmcb_page = (vmcb_t *)V3_VAddr((void *)vmcb_pa);
  81
  82     memset(vmcb_page, 0, 4096);
  83
  84     return vmcb_page;
  85 }
  86
  87
  88 static int v3_svm_handle_efer_write(struct guest_info * core, uint_t msr, struct v3_msr src, void * priv_data)
  89 {
  90     int status;
  91
  92     // Call arch-independent handler
  93     if ((status = v3_handle_efer_write(core, msr, src, priv_data)) != 0) {
  94         return status;
  95     }
  96
  97     // SVM-specific code
  98     {
  99         // Ensure that hardware visible EFER.SVME bit is set (SVM Enable)
 100         struct efer_64 * hw_efer = (struct efer_64 *)&(core->ctrl_regs.efer);
 101         hw_efer->svme = 1;
 102     }
 103
 104     return 0;
 105 }
 106
 107
 108 static void Init_VMCB_BIOS(vmcb_t * vmcb, struct guest_info * core) {
 109     vmcb_ctrl_t * ctrl_area = GET_VMCB_CTRL_AREA(vmcb);
 110     vmcb_saved_state_t * guest_state = GET_VMCB_SAVE_STATE_AREA(vmcb);
 111     uint_t i;
 112
 113
 114     //
 115     ctrl_area->svm_instrs.VMRUN = 1;
 116     ctrl_area->svm_instrs.VMMCALL = 1;
 117     ctrl_area->svm_instrs.VMLOAD = 1;
 118     ctrl_area->svm_instrs.VMSAVE = 1;
 119     ctrl_area->svm_instrs.STGI = 1;
 120     ctrl_area->svm_instrs.CLGI = 1;
 121     ctrl_area->svm_instrs.SKINIT = 1;
 122     ctrl_area->svm_instrs.ICEBP = 1;
 123     ctrl_area->svm_instrs.WBINVD = 1;
 124     ctrl_area->svm_instrs.MONITOR = 1;
 125     ctrl_area->svm_instrs.MWAIT_always = 1;
 126     ctrl_area->svm_instrs.MWAIT_if_armed = 1;
 127     ctrl_area->instrs.INVLPGA = 1;
 128     ctrl_area->instrs.CPUID = 1;
 129
 130     ctrl_area->instrs.HLT = 1;
 131
 132 #ifdef V3_CONFIG_TIME_VIRTUALIZE_TSC
 133     ctrl_area->instrs.RDTSC = 1;
 134     ctrl_area->svm_instrs.RDTSCP = 1;
 135 #endif
 136
 137     // guest_state->cr0 = 0x00000001;    // PE
 138
 139     /*
 140       ctrl_area->exceptions.de = 1;
 141       ctrl_area->exceptions.df = 1;
 142
 143       ctrl_area->exceptions.ts = 1;
 144       ctrl_area->exceptions.ss = 1;
 145       ctrl_area->exceptions.ac = 1;
 146       ctrl_area->exceptions.mc = 1;
 147       ctrl_area->exceptions.gp = 1;
 148       ctrl_area->exceptions.ud = 1;
 149       ctrl_area->exceptions.np = 1;
 150       ctrl_area->exceptions.of = 1;
 151
 152       ctrl_area->exceptions.nmi = 1;
 153     */
 154
 155
 156     ctrl_area->instrs.NMI = 1;
 157     ctrl_area->instrs.SMI = 0; // allow SMIs to run in guest
 158     ctrl_area->instrs.INIT = 1;
 159     ctrl_area->instrs.PAUSE = 1;
 160     ctrl_area->instrs.shutdown_evts = 1;
 161
 162
 163     /* DEBUG FOR RETURN CODE */
 164     ctrl_area->exit_code = 1;
 165
 166
 167     /* Setup Guest Machine state */
 168
 169     core->vm_regs.rsp = 0x00;
 170     core->rip = 0xfff0;
 171
 172     core->vm_regs.rdx = 0x00000f00;
 173
 174
 175     core->cpl = 0;
 176
 177     core->ctrl_regs.rflags = 0x00000002; // The reserved bit is always 1
 178     core->ctrl_regs.cr0 = 0x60010010; // Set the WP flag so the memory hooks work in real-mode
 179     core->ctrl_regs.efer |= EFER_MSR_svm_enable;
 180
 181
 182
 183
 184
 185     core->segments.cs.selector = 0xf000;
 186     core->segments.cs.limit = 0xffff;
 187     core->segments.cs.base = 0x0000000f0000LL;
 188
 189     // (raw attributes = 0xf3)
 190     core->segments.cs.type = 0x3;
 191     core->segments.cs.system = 0x1;
 192     core->segments.cs.dpl = 0x3;
 193     core->segments.cs.present = 1;
 194
 195
 196
 197     struct v3_segment * segregs [] = {&(core->segments.ss), &(core->segments.ds),
 198                                       &(core->segments.es), &(core->segments.fs),
 199                                       &(core->segments.gs), NULL};
 200
 201     for ( i = 0; segregs[i] != NULL; i++) {
 202         struct v3_segment * seg = segregs[i];
 203
 204         seg->selector = 0x0000;
 205         //    seg->base = seg->selector << 4;
 206         seg->base = 0x00000000;
 207         seg->limit = ~0u;
 208
 209         // (raw attributes = 0xf3)
 210         seg->type = 0x3;
 211         seg->system = 0x1;
 212         seg->dpl = 0x3;
 213         seg->present = 1;
 214     }
 215
 216     core->segments.gdtr.limit = 0x0000ffff;
 217     core->segments.gdtr.base = 0x0000000000000000LL;
 218     core->segments.idtr.limit = 0x0000ffff;
 219     core->segments.idtr.base = 0x0000000000000000LL;
 220
 221     core->segments.ldtr.selector = 0x0000;
 222     core->segments.ldtr.limit = 0x0000ffff;
 223     core->segments.ldtr.base = 0x0000000000000000LL;
 224     core->segments.tr.selector = 0x0000;
 225     core->segments.tr.limit = 0x0000ffff;
 226     core->segments.tr.base = 0x0000000000000000LL;
 227
 228
 229     core->dbg_regs.dr6 = 0x00000000ffff0ff0LL;
 230     core->dbg_regs.dr7 = 0x0000000000000400LL;
 231
 232
 233     ctrl_area->IOPM_BASE_PA = (addr_t)V3_PAddr(core->vm_info->io_map.arch_data);
 234     ctrl_area->instrs.IOIO_PROT = 1;
 235
 236     ctrl_area->MSRPM_BASE_PA = (addr_t)V3_PAddr(core->vm_info->msr_map.arch_data);
 237     ctrl_area->instrs.MSR_PROT = 1;
 238
 239
 240     PrintDebug("Exiting on interrupts\n");
 241     ctrl_area->guest_ctrl.V_INTR_MASKING = 1;
 242     ctrl_area->instrs.INTR = 1;
 243
 244
 245     v3_hook_msr(core->vm_info, EFER_MSR,
 246                 &v3_handle_efer_read,
 247                 &v3_svm_handle_efer_write,
 248                 core);
 249
 250     if (core->shdw_pg_mode == SHADOW_PAGING) {
 251         PrintDebug("Creating initial shadow page table\n");
 252
 253         /* JRL: This is a performance killer, and a simplistic solution */
 254         /* We need to fix this */
 255         ctrl_area->TLB_CONTROL = 1;
 256         ctrl_area->guest_ASID = 1;
 257
 258
 259         if (v3_init_passthrough_pts(core) == -1) {
 260             PrintError("Could not initialize passthrough page tables\n");
 261             return ;
 262         }
 263
 264
 265         core->shdw_pg_state.guest_cr0 = 0x0000000000000010LL;
 266         PrintDebug("Created\n");
 267
 268         core->ctrl_regs.cr0 |= 0x80000000;
 269         core->ctrl_regs.cr3 = core->direct_map_pt;
 270
 271         ctrl_area->cr_reads.cr0 = 1;
 272         ctrl_area->cr_writes.cr0 = 1;
 273         //ctrl_area->cr_reads.cr4 = 1;
 274         ctrl_area->cr_writes.cr4 = 1;
 275         ctrl_area->cr_reads.cr3 = 1;
 276         ctrl_area->cr_writes.cr3 = 1;
 277
 278
 279
 280         ctrl_area->instrs.INVLPG = 1;
 281
 282         ctrl_area->exceptions.pf = 1;
 283
 284         guest_state->g_pat = 0x7040600070406ULL;
 285
 286
 287
 288     } else if (core->shdw_pg_mode == NESTED_PAGING) {
 289         // Flush the TLB on entries/exits
 290         ctrl_area->TLB_CONTROL = 1;
 291         ctrl_area->guest_ASID = 1;
 292
 293         // Enable Nested Paging
 294         ctrl_area->NP_ENABLE = 1;
 295
 296         PrintDebug("NP_Enable at 0x%p\n", (void *)&(ctrl_area->NP_ENABLE));
 297
 298         // Set the Nested Page Table pointer
 299         if (v3_init_passthrough_pts(core) == -1) {
 300             PrintError("Could not initialize Nested page tables\n");
 301             return ;
 302         }
 303
 304         ctrl_area->N_CR3 = core->direct_map_pt;
 305
 306         guest_state->g_pat = 0x7040600070406ULL;
 307     }
 308
 309     /* tell the guest that we don't support SVM */
 310     v3_hook_msr(core->vm_info, SVM_VM_CR_MSR,
 311         &v3_handle_vm_cr_read,
 312         &v3_handle_vm_cr_write,
 313         core);
 314
 315
 316     {
 317 #define INT_PENDING_AMD_MSR             0xc0010055
 318
 319         v3_hook_msr(core->vm_info, IA32_STAR_MSR, NULL, NULL, NULL);
 320         v3_hook_msr(core->vm_info, IA32_LSTAR_MSR, NULL, NULL, NULL);
 321         v3_hook_msr(core->vm_info, IA32_FMASK_MSR, NULL, NULL, NULL);
 322         v3_hook_msr(core->vm_info, IA32_KERN_GS_BASE_MSR, NULL, NULL, NULL);
 323         v3_hook_msr(core->vm_info, IA32_CSTAR_MSR, NULL, NULL, NULL);
 324
 325         v3_hook_msr(core->vm_info, SYSENTER_CS_MSR, NULL, NULL, NULL);
 326         v3_hook_msr(core->vm_info, SYSENTER_ESP_MSR, NULL, NULL, NULL);
 327         v3_hook_msr(core->vm_info, SYSENTER_EIP_MSR, NULL, NULL, NULL);
 328
 329
 330         v3_hook_msr(core->vm_info, FS_BASE_MSR, NULL, NULL, NULL);
 331         v3_hook_msr(core->vm_info, GS_BASE_MSR, NULL, NULL, NULL);
 332
 333         // Passthrough read operations are ok.
 334         v3_hook_msr(core->vm_info, INT_PENDING_AMD_MSR, NULL, v3_msr_unhandled_write, NULL);
 335     }
 336 }
 337
 338
 339 int v3_init_svm_vmcb(struct guest_info * core, v3_vm_class_t vm_class) {
 340
 341     PrintDebug("Allocating VMCB\n");
 342     core->vmm_data = (void *)Allocate_VMCB();
 343
 344     if (core->vmm_data == NULL) {
 345         PrintError("Could not allocate VMCB, Exiting...\n");
 346         return -1;
 347     }
 348
 349     if (vm_class == V3_PC_VM) {
 350         PrintDebug("Initializing VMCB (addr=%p)\n", (void *)core->vmm_data);
 351         Init_VMCB_BIOS((vmcb_t*)(core->vmm_data), core);
 352     } else {
 353         PrintError("Invalid VM class\n");
 354         return -1;
 355     }
 356
 357     return 0;
 358 }
 359
 360
 361 int v3_deinit_svm_vmcb(struct guest_info * core) {
 362     V3_FreePages(V3_PAddr(core->vmm_data), 1);
 363     return 0;
 364 }
 365
 366
 367 #ifdef V3_CONFIG_CHECKPOINT
 368 int v3_svm_save_core(struct guest_info * core, void * ctx){
 369
 370     v3_chkpt_save_8(ctx, "cpl", &(core->cpl));
 371     v3_chkpt_save(ctx, "vmcb_data", PAGE_SIZE, core->vmm_data);
 372
 373     return 0;
 374 }
 375
 376 int v3_svm_load_core(struct guest_info * core, void * ctx){
 377
 378     v3_chkpt_load_8(ctx, "cpl", &(core->cpl));
 379
 380     if (v3_chkpt_load(ctx, "vmcb_data", PAGE_SIZE, core->vmm_data) == -1) {
 381         return -1;
 382     }
 383
 384     return 0;
 385 }
 386 #endif
 387
 388 static int update_irq_exit_state(struct guest_info * info) {
 389     vmcb_ctrl_t * guest_ctrl = GET_VMCB_CTRL_AREA((vmcb_t*)(info->vmm_data));
 390
 391     // Fix for QEMU bug using EVENTINJ as an internal cache
 392     guest_ctrl->EVENTINJ.valid = 0;
 393
 394     if ((info->intr_core_state.irq_pending == 1) && (guest_ctrl->guest_ctrl.V_IRQ == 0)) {
 395
 396 #ifdef V3_CONFIG_DEBUG_INTERRUPTS
 397         PrintDebug("INTAK cycle completed for irq %d\n", info->intr_core_state.irq_vector);
 398 #endif
 399
 400         info->intr_core_state.irq_started = 1;
 401         info->intr_core_state.irq_pending = 0;
 402
 403         v3_injecting_intr(info, info->intr_core_state.irq_vector, V3_EXTERNAL_IRQ);
 404     }
 405
 406     if ((info->intr_core_state.irq_started == 1) && (guest_ctrl->exit_int_info.valid == 0)) {
 407 #ifdef V3_CONFIG_DEBUG_INTERRUPTS
 408         PrintDebug("Interrupt %d taken by guest\n", info->intr_core_state.irq_vector);
 409 #endif
 410
 411         // Interrupt was taken fully vectored
 412         info->intr_core_state.irq_started = 0;
 413
 414     } else if ((info->intr_core_state.irq_started == 1) && (guest_ctrl->exit_int_info.valid == 1)) {
 415 #ifdef V3_CONFIG_DEBUG_INTERRUPTS
 416         PrintDebug("EXIT INT INFO is set (vec=%d)\n", guest_ctrl->exit_int_info.vector);
 417 #endif
 418     }
 419
 420     return 0;
 421 }
 422
 423
 424 static int update_irq_entry_state(struct guest_info * info) {
 425     vmcb_ctrl_t * guest_ctrl = GET_VMCB_CTRL_AREA((vmcb_t*)(info->vmm_data));
 426
 427
 428     if (info->intr_core_state.irq_pending == 0) {
 429         guest_ctrl->guest_ctrl.V_IRQ = 0;
 430         guest_ctrl->guest_ctrl.V_INTR_VECTOR = 0;
 431     }
 432
 433     if (v3_excp_pending(info)) {
 434         uint_t excp = v3_get_excp_number(info);
 435
 436         guest_ctrl->EVENTINJ.type = SVM_INJECTION_EXCEPTION;
 437
 438         if (info->excp_state.excp_error_code_valid) {
 439             guest_ctrl->EVENTINJ.error_code = info->excp_state.excp_error_code;
 440             guest_ctrl->EVENTINJ.ev = 1;
 441 #ifdef V3_CONFIG_DEBUG_INTERRUPTS
 442             PrintDebug("Injecting exception %d with error code %x\n", excp, guest_ctrl->EVENTINJ.error_code);
 443 #endif
 444         }
 445
 446         guest_ctrl->EVENTINJ.vector = excp;
 447
 448         guest_ctrl->EVENTINJ.valid = 1;
 449
 450 #ifdef V3_CONFIG_DEBUG_INTERRUPTS
 451         PrintDebug("<%d> Injecting Exception %d (CR2=%p) (EIP=%p)\n",
 452                    (int)info->num_exits,
 453                    guest_ctrl->EVENTINJ.vector,
 454                    (void *)(addr_t)info->ctrl_regs.cr2,
 455                    (void *)(addr_t)info->rip);
 456 #endif
 457
 458         v3_injecting_excp(info, excp);
 459     } else if (info->intr_core_state.irq_started == 1) {
 460 #ifdef V3_CONFIG_DEBUG_INTERRUPTS
 461         PrintDebug("IRQ pending from previous injection\n");
 462 #endif
 463         guest_ctrl->guest_ctrl.V_IRQ = 1;
 464         guest_ctrl->guest_ctrl.V_INTR_VECTOR = info->intr_core_state.irq_vector;
 465         guest_ctrl->guest_ctrl.V_IGN_TPR = 1;
 466         guest_ctrl->guest_ctrl.V_INTR_PRIO = 0xf;
 467
 468     } else {
 469         switch (v3_intr_pending(info)) {
 470             case V3_EXTERNAL_IRQ: {
 471                 uint32_t irq = v3_get_intr(info);
 472
 473                 guest_ctrl->guest_ctrl.V_IRQ = 1;
 474                 guest_ctrl->guest_ctrl.V_INTR_VECTOR = irq;
 475                 guest_ctrl->guest_ctrl.V_IGN_TPR = 1;
 476                 guest_ctrl->guest_ctrl.V_INTR_PRIO = 0xf;
 477
 478 #ifdef V3_CONFIG_DEBUG_INTERRUPTS
 479                 PrintDebug("Injecting Interrupt %d (EIP=%p)\n",
 480                            guest_ctrl->guest_ctrl.V_INTR_VECTOR,
 481                            (void *)(addr_t)info->rip);
 482 #endif
 483
 484                 info->intr_core_state.irq_pending = 1;
 485                 info->intr_core_state.irq_vector = irq;
 486
 487                 break;
 488             }
 489             case V3_NMI:
 490                 guest_ctrl->EVENTINJ.type = SVM_INJECTION_NMI;
 491                 break;
 492             case V3_SOFTWARE_INTR:
 493                 guest_ctrl->EVENTINJ.type = SVM_INJECTION_SOFT_INTR;
 494
 495 #ifdef V3_CONFIG_DEBUG_INTERRUPTS
 496                 PrintDebug("Injecting software interrupt --  type: %d, vector: %d\n",
 497                            SVM_INJECTION_SOFT_INTR, info->intr_core_state.swintr_vector);
 498 #endif
 499                 guest_ctrl->EVENTINJ.vector = info->intr_core_state.swintr_vector;
 500                 guest_ctrl->EVENTINJ.valid = 1;
 501
 502                 /* reset swintr state */
 503                 info->intr_core_state.swintr_posted = 0;
 504                 info->intr_core_state.swintr_vector = 0;
 505
 506                 break;
 507             case V3_VIRTUAL_IRQ:
 508                 guest_ctrl->EVENTINJ.type = SVM_INJECTION_IRQ;
 509                 break;
 510
 511             case V3_INVALID_INTR:
 512             default:
 513                 break;
 514         }
 515
 516     }
 517
 518     return 0;
 519 }
 520
 521
 522 /*
 523  * CAUTION and DANGER!!!
 524  *
 525  * The VMCB CANNOT(!!) be accessed outside of the clgi/stgi calls inside this function
 526  * When exectuing a symbiotic call, the VMCB WILL be overwritten, so any dependencies
 527  * on its contents will cause things to break. The contents at the time of the exit WILL
 528  * change before the exit handler is executed.
 529  */
 530 int v3_svm_enter(struct guest_info * info) {
 531     vmcb_ctrl_t * guest_ctrl = GET_VMCB_CTRL_AREA((vmcb_t*)(info->vmm_data));
 532     vmcb_saved_state_t * guest_state = GET_VMCB_SAVE_STATE_AREA((vmcb_t*)(info->vmm_data));
 533     addr_t exit_code = 0, exit_info1 = 0, exit_info2 = 0;
 534     sint64_t tsc_offset;
 535
 536     // Conditionally yield the CPU if the timeslice has expired
 537     v3_yield_cond(info);
 538
 539     // Perform any additional yielding needed for time adjustment
 540     v3_adjust_time(info);
 541
 542     // Check for timeout - since this calls generic hooks in devices
 543     // that may do things like pause the VM, it cannot be with interrupts
 544     // disabled.
 545     v3_check_timeout(info);
 546
 547     // disable global interrupts for vm state transition
 548     v3_clgi();
 549
 550     // Update timer devices after being in the VM, with interupts
 551     // disabled, but before doing IRQ updates, so that any interrupts they
 552     //raise get seen immediately.
 553     v3_update_timers(info);
 554
 555     // Synchronize the guest state to the VMCB
 556     guest_state->cr0 = info->ctrl_regs.cr0;
 557     guest_state->cr2 = info->ctrl_regs.cr2;
 558     guest_state->cr3 = info->ctrl_regs.cr3;
 559     guest_state->cr4 = info->ctrl_regs.cr4;
 560     guest_state->dr6 = info->dbg_regs.dr6;
 561     guest_state->dr7 = info->dbg_regs.dr7;
 562     guest_ctrl->guest_ctrl.V_TPR = info->ctrl_regs.cr8 & 0xff;
 563     guest_state->rflags = info->ctrl_regs.rflags;
 564     guest_state->efer = info->ctrl_regs.efer;
 565
 566     guest_state->cpl = info->cpl;
 567
 568     v3_set_vmcb_segments((vmcb_t*)(info->vmm_data), &(info->segments));
 569
 570     guest_state->rax = info->vm_regs.rax;
 571     guest_state->rip = info->rip;
 572     guest_state->rsp = info->vm_regs.rsp;
 573
 574 #ifdef V3_CONFIG_SYMCALL
 575     if (info->sym_core_state.symcall_state.sym_call_active == 0) {
 576         update_irq_entry_state(info);
 577     }
 578 #else
 579     update_irq_entry_state(info);
 580 #endif
 581
 582
 583     /* ** */
 584
 585     /*
 586       PrintDebug("SVM Entry to CS=%p  rip=%p...\n",
 587       (void *)(addr_t)info->segments.cs.base,
 588       (void *)(addr_t)info->rip);
 589     */
 590
 591 #ifdef V3_CONFIG_SYMCALL
 592     if (info->sym_core_state.symcall_state.sym_call_active == 1) {
 593         if (guest_ctrl->guest_ctrl.V_IRQ == 1) {
 594             V3_Print("!!! Injecting Interrupt during Sym call !!!\n");
 595         }
 596     }
 597 #endif
 598
 599     v3_time_enter_vm(info);
 600     tsc_offset = v3_tsc_host_offset(&info->time_state);
 601     guest_ctrl->TSC_OFFSET = tsc_offset;
 602
 603
 604     //V3_Print("Calling v3_svm_launch\n");
 605
 606     v3_svm_launch((vmcb_t *)V3_PAddr(info->vmm_data), &(info->vm_regs), (vmcb_t *)host_vmcbs[V3_Get_CPU()]);
 607
 608     //V3_Print("SVM Returned: Exit Code: %x, guest_rip=%lx\n", (uint32_t)(guest_ctrl->exit_code), (unsigned long)guest_state->rip);
 609
 610     v3_last_exit = (uint32_t)(guest_ctrl->exit_code);
 611
 612     // Immediate exit from VM time bookkeeping
 613     v3_time_exit_vm(info);
 614
 615     info->num_exits++;
 616
 617     // Save Guest state from VMCB
 618     info->rip = guest_state->rip;
 619     info->vm_regs.rsp = guest_state->rsp;
 620     info->vm_regs.rax = guest_state->rax;
 621
 622     info->cpl = guest_state->cpl;
 623
 624     info->ctrl_regs.cr0 = guest_state->cr0;
 625     info->ctrl_regs.cr2 = guest_state->cr2;
 626     info->ctrl_regs.cr3 = guest_state->cr3;
 627     info->ctrl_regs.cr4 = guest_state->cr4;
 628     info->dbg_regs.dr6 = guest_state->dr6;
 629     info->dbg_regs.dr7 = guest_state->dr7;
 630     info->ctrl_regs.cr8 = guest_ctrl->guest_ctrl.V_TPR;
 631     info->ctrl_regs.rflags = guest_state->rflags;
 632     info->ctrl_regs.efer = guest_state->efer;
 633
 634     v3_get_vmcb_segments((vmcb_t*)(info->vmm_data), &(info->segments));
 635     info->cpu_mode = v3_get_vm_cpu_mode(info);
 636     info->mem_mode = v3_get_vm_mem_mode(info);
 637     /* ** */
 638
 639     // save exit info here
 640     exit_code = guest_ctrl->exit_code;
 641     exit_info1 = guest_ctrl->exit_info1;
 642     exit_info2 = guest_ctrl->exit_info2;
 643
 644 #ifdef V3_CONFIG_SYMCALL
 645     if (info->sym_core_state.symcall_state.sym_call_active == 0) {
 646         update_irq_exit_state(info);
 647     }
 648 #else
 649     update_irq_exit_state(info);
 650 #endif
 651
 652     // reenable global interrupts after vm exit
 653     v3_stgi();
 654
 655     // Conditionally yield the CPU if the timeslice has expired
 656     v3_yield_cond(info);
 657
 658     {
 659         int ret = v3_handle_svm_exit(info, exit_code, exit_info1, exit_info2);
 660
 661         if (ret != 0) {
 662             PrintError("Error in SVM exit handler (ret=%d)\n", ret);
 663             PrintError("  last Exit was %d (exit code=0x%llx)\n", v3_last_exit, (uint64_t) exit_code);
 664             return -1;
 665         }
 666     }
 667
 668
 669     return 0;
 670 }
 671
 672
 673 int v3_start_svm_guest(struct guest_info * info) {
 674     //    vmcb_saved_state_t * guest_state = GET_VMCB_SAVE_STATE_AREA((vmcb_t*)(info->vmm_data));
 675     //  vmcb_ctrl_t * guest_ctrl = GET_VMCB_CTRL_AREA((vmcb_t*)(info->vmm_data));
 676
 677     PrintDebug("Starting SVM core %u (on logical core %u)\n", info->vcpu_id, info->pcpu_id);
 678
 679     if (info->vcpu_id == 0) {
 680         info->core_run_state = CORE_RUNNING;
 681     } else  {
 682         PrintDebug("SVM core %u (on %u): Waiting for core initialization\n", info->vcpu_id, info->pcpu_id);
 683
 684         while (info->core_run_state == CORE_STOPPED) {
 685
 686             if (info->vm_info->run_state == VM_STOPPED) {
 687                 // The VM was stopped before this core was initialized.
 688                 return 0;
 689             }
 690
 691             v3_yield(info);
 692             //PrintDebug("SVM core %u: still waiting for INIT\n", info->vcpu_id);
 693         }
 694
 695         PrintDebug("SVM core %u(on %u) initialized\n", info->vcpu_id, info->pcpu_id);
 696
 697         // We'll be paranoid about race conditions here
 698         v3_wait_at_barrier(info);
 699     }
 700
 701     PrintDebug("SVM core %u(on %u): I am starting at CS=0x%x (base=0x%p, limit=0x%x),  RIP=0x%p\n",
 702                info->vcpu_id, info->pcpu_id,
 703                info->segments.cs.selector, (void *)(info->segments.cs.base),
 704                info->segments.cs.limit, (void *)(info->rip));
 705
 706
 707
 708     PrintDebug("SVM core %u: Launching SVM VM (vmcb=%p) (on cpu %u)\n",
 709                info->vcpu_id, (void *)info->vmm_data, info->pcpu_id);
 710     //PrintDebugVMCB((vmcb_t*)(info->vmm_data));
 711
 712     v3_start_time(info);
 713
 714     while (1) {
 715
 716         if (info->vm_info->run_state == VM_STOPPED) {
 717             info->core_run_state = CORE_STOPPED;
 718             break;
 719         }
 720
 721         if (v3_svm_enter(info) == -1) {
 722             vmcb_ctrl_t * guest_ctrl = GET_VMCB_CTRL_AREA((vmcb_t*)(info->vmm_data));
 723             addr_t host_addr;
 724             addr_t linear_addr = 0;
 725
 726             info->vm_info->run_state = VM_ERROR;
 727
 728             V3_Print("SVM core %u: SVM ERROR!!\n", info->vcpu_id);
 729
 730             v3_print_guest_state(info);
 731
 732             V3_Print("SVM core %u: SVM Exit Code: %p\n", info->vcpu_id, (void *)(addr_t)guest_ctrl->exit_code);
 733
 734             V3_Print("SVM core %u: exit_info1 low = 0x%.8x\n", info->vcpu_id, *(uint_t*)&(guest_ctrl->exit_info1));
 735             V3_Print("SVM core %u: exit_info1 high = 0x%.8x\n", info->vcpu_id, *(uint_t *)(((uchar_t *)&(guest_ctrl->exit_info1)) + 4));
 736
 737             V3_Print("SVM core %u: exit_info2 low = 0x%.8x\n", info->vcpu_id, *(uint_t*)&(guest_ctrl->exit_info2));
 738             V3_Print("SVM core %u: exit_info2 high = 0x%.8x\n", info->vcpu_id, *(uint_t *)(((uchar_t *)&(guest_ctrl->exit_info2)) + 4));
 739
 740             linear_addr = get_addr_linear(info, info->rip, &(info->segments.cs));
 741
 742             if (info->mem_mode == PHYSICAL_MEM) {
 743                 v3_gpa_to_hva(info, linear_addr, &host_addr);
 744             } else if (info->mem_mode == VIRTUAL_MEM) {
 745                 v3_gva_to_hva(info, linear_addr, &host_addr);
 746             }
 747
 748             V3_Print("SVM core %u: Host Address of rip = 0x%p\n", info->vcpu_id, (void *)host_addr);
 749
 750             V3_Print("SVM core %u: Instr (15 bytes) at %p:\n", info->vcpu_id, (void *)host_addr);
 751             v3_dump_mem((uint8_t *)host_addr, 15);
 752
 753             v3_print_stack(info);
 754
 755             break;
 756         }
 757
 758         v3_wait_at_barrier(info);
 759
 760
 761         if (info->vm_info->run_state == VM_STOPPED) {
 762             info->core_run_state = CORE_STOPPED;
 763             break;
 764         }
 765
 766
 767
 768 /*
 769         if ((info->num_exits % 50000) == 0) {
 770             V3_Print("SVM Exit number %d\n", (uint32_t)info->num_exits);
 771             v3_print_guest_state(info);
 772         }
 773 */
 774
 775     }
 776
 777     // Need to take down the other cores on error...
 778
 779     return 0;
 780 }
 781
 782
 783
 784
 785 int v3_reset_svm_vm_core(struct guest_info * core, addr_t rip) {
 786     // init vmcb_bios
 787
 788     // Write the RIP, CS, and descriptor
 789     // assume the rest is already good to go
 790     //
 791     // vector VV -> rip at 0
 792     //              CS = VV00
 793     //  This means we start executing at linear address VV000
 794     //
 795     // So the selector needs to be VV00
 796     // and the base needs to be VV000
 797     //
 798     core->rip = 0;
 799     core->segments.cs.selector = rip << 8;
 800     core->segments.cs.limit = 0xffff;
 801     core->segments.cs.base = rip << 12;
 802
 803     return 0;
 804 }
 805
 806
 807
 808
 809
 810
 811 /* Checks machine SVM capability */
 812 /* Implemented from: AMD Arch Manual 3, sect 15.4 */
 813 int v3_is_svm_capable() {
 814     uint_t vm_cr_low = 0, vm_cr_high = 0;
 815     uint32_t eax = 0, ebx = 0, ecx = 0, edx = 0;
 816
 817     v3_cpuid(CPUID_EXT_FEATURE_IDS, &eax, &ebx, &ecx, &edx);
 818
 819     PrintDebug("CPUID_EXT_FEATURE_IDS_ecx=0x%x\n", ecx);
 820
 821     if ((ecx & CPUID_EXT_FEATURE_IDS_ecx_svm_avail) == 0) {
 822       V3_Print("SVM Not Available\n");
 823       return 0;
 824     }  else {
 825         v3_get_msr(SVM_VM_CR_MSR, &vm_cr_high, &vm_cr_low);
 826
 827         PrintDebug("SVM_VM_CR_MSR = 0x%x 0x%x\n", vm_cr_high, vm_cr_low);
 828
 829         if ((vm_cr_low & SVM_VM_CR_MSR_svmdis) == 1) {
 830             V3_Print("SVM is available but is disabled.\n");
 831
 832             v3_cpuid(CPUID_SVM_REV_AND_FEATURE_IDS, &eax, &ebx, &ecx, &edx);
 833
 834             PrintDebug("CPUID_SVM_REV_AND_FEATURE_IDS_edx=0x%x\n", edx);
 835
 836             if ((edx & CPUID_SVM_REV_AND_FEATURE_IDS_edx_svml) == 0) {
 837                 V3_Print("SVM BIOS Disabled, not unlockable\n");
 838             } else {
 839                 V3_Print("SVM is locked with a key\n");
 840             }
 841             return 0;
 842
 843         } else {
 844             V3_Print("SVM is available and  enabled.\n");
 845
 846             v3_cpuid(CPUID_SVM_REV_AND_FEATURE_IDS, &eax, &ebx, &ecx, &edx);
 847             PrintDebug("CPUID_SVM_REV_AND_FEATURE_IDS_eax=0x%x\n", eax);
 848             PrintDebug("CPUID_SVM_REV_AND_FEATURE_IDS_ebx=0x%x\n", ebx);
 849             PrintDebug("CPUID_SVM_REV_AND_FEATURE_IDS_ecx=0x%x\n", ecx);
 850             PrintDebug("CPUID_SVM_REV_AND_FEATURE_IDS_edx=0x%x\n", edx);
 851
 852             return 1;
 853         }
 854     }
 855 }
 856
 857 static int has_svm_nested_paging() {
 858     uint32_t eax = 0, ebx = 0, ecx = 0, edx = 0;
 859
 860     v3_cpuid(CPUID_SVM_REV_AND_FEATURE_IDS, &eax, &ebx, &ecx, &edx);
 861
 862     //PrintDebug("CPUID_EXT_FEATURE_IDS_edx=0x%x\n", edx);
 863
 864     if ((edx & CPUID_SVM_REV_AND_FEATURE_IDS_edx_np) == 0) {
 865         V3_Print("SVM Nested Paging not supported\n");
 866         return 0;
 867     } else {
 868         V3_Print("SVM Nested Paging supported\n");
 869         return 1;
 870     }
 871  }
 872
 873
 874
 875 void v3_init_svm_cpu(int cpu_id) {
 876     reg_ex_t msr;
 877     extern v3_cpu_arch_t v3_cpu_types[];
 878
 879     // Enable SVM on the CPU
 880     v3_get_msr(EFER_MSR, &(msr.e_reg.high), &(msr.e_reg.low));
 881     msr.e_reg.low |= EFER_MSR_svm_enable;
 882     v3_set_msr(EFER_MSR, 0, msr.e_reg.low);
 883
 884     V3_Print("SVM Enabled\n");
 885
 886     // Setup the host state save area
 887     host_vmcbs[cpu_id] = (addr_t)V3_AllocPages(4);
 888
 889     /* 64-BIT-ISSUE */
 890     //  msr.e_reg.high = 0;
 891     //msr.e_reg.low = (uint_t)host_vmcb;
 892     msr.r_reg = host_vmcbs[cpu_id];
 893
 894     PrintDebug("Host State being saved at %p\n", (void *)host_vmcbs[cpu_id]);
 895     v3_set_msr(SVM_VM_HSAVE_PA_MSR, msr.e_reg.high, msr.e_reg.low);
 896
 897
 898     if (has_svm_nested_paging() == 1) {
 899         v3_cpu_types[cpu_id] = V3_SVM_REV3_CPU;
 900     } else {
 901         v3_cpu_types[cpu_id] = V3_SVM_CPU;
 902     }
 903 }
 904
 905
 906
 907 void v3_deinit_svm_cpu(int cpu_id) {
 908     reg_ex_t msr;
 909     extern v3_cpu_arch_t v3_cpu_types[];
 910
 911     // reset SVM_VM_HSAVE_PA_MSR
 912     // Does setting it to NULL disable??
 913     msr.r_reg = 0;
 914     v3_set_msr(SVM_VM_HSAVE_PA_MSR, msr.e_reg.high, msr.e_reg.low);
 915
 916     // Disable SVM?
 917     v3_get_msr(EFER_MSR, &(msr.e_reg.high), &(msr.e_reg.low));
 918     msr.e_reg.low &= ~EFER_MSR_svm_enable;
 919     v3_set_msr(EFER_MSR, 0, msr.e_reg.low);
 920
 921     v3_cpu_types[cpu_id] = V3_INVALID_CPU;
 922
 923     V3_FreePages((void *)host_vmcbs[cpu_id], 4);
 924
 925     V3_Print("Host CPU %d host area freed, and SVM disabled\n", cpu_id);
 926     return;
 927 }
 928
 929
 930
 931
 932
 933
 934
 935
 936
 937
 938
 939
 940
 941
 942
 943
 944
 945
 946
 947
 948
 949
 950
 951
 952
 953
 954
 955
 956
 957
 958
 959
 960
 961
 962
 963
 964
 965
 966
 967
 968
 969
 970
 971
 972
 973
 974
 975
 976
 977
 978 #if 0
 979 /*
 980  * Test VMSAVE/VMLOAD Latency
 981  */
 982 #define vmsave ".byte 0x0F,0x01,0xDB ; "
 983 #define vmload ".byte 0x0F,0x01,0xDA ; "
 984 {
 985     uint32_t start_lo, start_hi;
 986     uint32_t end_lo, end_hi;
 987     uint64_t start, end;
 988
 989     __asm__ __volatile__ (
 990                           "rdtsc ; "
 991                           "movl %%eax, %%esi ; "
 992                           "movl %%edx, %%edi ; "
 993                           "movq  %%rcx, %%rax ; "
 994                           vmsave
 995                           "rdtsc ; "
 996                           : "=D"(start_hi), "=S"(start_lo), "=a"(end_lo),"=d"(end_hi)
 997                           : "c"(host_vmcb[cpu_id]), "0"(0), "1"(0), "2"(0), "3"(0)
 998                           );
 999
1000     start = start_hi;
1001     start <<= 32;
1002     start += start_lo;
1003
1004     end = end_hi;
1005     end <<= 32;
1006     end += end_lo;
1007
1008     PrintDebug("VMSave Cycle Latency: %d\n", (uint32_t)(end - start));
1009
1010     __asm__ __volatile__ (
1011                           "rdtsc ; "
1012                           "movl %%eax, %%esi ; "
1013                           "movl %%edx, %%edi ; "
1014                           "movq  %%rcx, %%rax ; "
1015                           vmload
1016                           "rdtsc ; "
1017                           : "=D"(start_hi), "=S"(start_lo), "=a"(end_lo),"=d"(end_hi)
1018                               : "c"(host_vmcb[cpu_id]), "0"(0), "1"(0), "2"(0), "3"(0)
1019                               );
1020
1021         start = start_hi;
1022         start <<= 32;
1023         start += start_lo;
1024
1025         end = end_hi;
1026         end <<= 32;
1027         end += end_lo;
1028
1029
1030         PrintDebug("VMLoad Cycle Latency: %d\n", (uint32_t)(end - start));
1031     }
1032     /* End Latency Test */
1033
1034 #endif
1035
1036
1037
1038
1039
1040
1041
1042 #if 0
1043 void Init_VMCB_pe(vmcb_t *vmcb, struct guest_info vm_info) {
1044   vmcb_ctrl_t * ctrl_area = GET_VMCB_CTRL_AREA(vmcb);
1045   vmcb_saved_state_t * guest_state = GET_VMCB_SAVE_STATE_AREA(vmcb);
1046   uint_t i = 0;
1047
1048
1049   guest_state->rsp = vm_info.vm_regs.rsp;
1050   guest_state->rip = vm_info.rip;
1051
1052
1053   /* I pretty much just gutted this from TVMM */
1054   /* Note: That means its probably wrong */
1055
1056   // set the segment registers to mirror ours
1057   guest_state->cs.selector = 1<<3;
1058   guest_state->cs.attrib.fields.type = 0xa; // Code segment+read
1059   guest_state->cs.attrib.fields.S = 1;
1060   guest_state->cs.attrib.fields.P = 1;
1061   guest_state->cs.attrib.fields.db = 1;
1062   guest_state->cs.attrib.fields.G = 1;
1063   guest_state->cs.limit = 0xfffff;
1064   guest_state->cs.base = 0;
1065
1066   struct vmcb_selector *segregs [] = {&(guest_state->ss), &(guest_state->ds), &(guest_state->es), &(guest_state->fs), &(guest_state->gs), NULL};
1067   for ( i = 0; segregs[i] != NULL; i++) {
1068     struct vmcb_selector * seg = segregs[i];
1069
1070     seg->selector = 2<<3;
1071     seg->attrib.fields.type = 0x2; // Data Segment+read/write
1072     seg->attrib.fields.S = 1;
1073     seg->attrib.fields.P = 1;
1074     seg->attrib.fields.db = 1;
1075     seg->attrib.fields.G = 1;
1076     seg->limit = 0xfffff;
1077     seg->base = 0;
1078   }
1079
1080
1081   {
1082     /* JRL THIS HAS TO GO */
1083
1084     //    guest_state->tr.selector = GetTR_Selector();
1085     guest_state->tr.attrib.fields.type = 0x9;
1086     guest_state->tr.attrib.fields.P = 1;
1087     // guest_state->tr.limit = GetTR_Limit();
1088     //guest_state->tr.base = GetTR_Base();// - 0x2000;
1089     /* ** */
1090   }
1091
1092
1093   /* ** */
1094
1095
1096   guest_state->efer |= EFER_MSR_svm_enable;
1097   guest_state->rflags = 0x00000002; // The reserved bit is always 1
1098   ctrl_area->svm_instrs.VMRUN = 1;
1099   guest_state->cr0 = 0x00000001;    // PE
1100   ctrl_area->guest_ASID = 1;
1101
1102
1103   //  guest_state->cpl = 0;
1104
1105
1106
1107   // Setup exits
1108
1109   ctrl_area->cr_writes.cr4 = 1;
1110
1111   ctrl_area->exceptions.de = 1;
1112   ctrl_area->exceptions.df = 1;
1113   ctrl_area->exceptions.pf = 1;
1114   ctrl_area->exceptions.ts = 1;
1115   ctrl_area->exceptions.ss = 1;
1116   ctrl_area->exceptions.ac = 1;
1117   ctrl_area->exceptions.mc = 1;
1118   ctrl_area->exceptions.gp = 1;
1119   ctrl_area->exceptions.ud = 1;
1120   ctrl_area->exceptions.np = 1;
1121   ctrl_area->exceptions.of = 1;
1122   ctrl_area->exceptions.nmi = 1;
1123
1124
1125
1126   ctrl_area->instrs.IOIO_PROT = 1;
1127   ctrl_area->IOPM_BASE_PA = (uint_t)V3_AllocPages(3);
1128
1129   {
1130     reg_ex_t tmp_reg;
1131     tmp_reg.r_reg = ctrl_area->IOPM_BASE_PA;
1132     memset((void*)(tmp_reg.e_reg.low), 0xffffffff, PAGE_SIZE * 2);
1133   }
1134
1135   ctrl_area->instrs.INTR = 1;
1136
1137
1138   {
1139     char gdt_buf[6];
1140     char idt_buf[6];
1141
1142     memset(gdt_buf, 0, 6);
1143     memset(idt_buf, 0, 6);
1144
1145
1146     uint_t gdt_base, idt_base;
1147     ushort_t gdt_limit, idt_limit;
1148
1149     GetGDTR(gdt_buf);
1150     gdt_base = *(ulong_t*)((uchar_t*)gdt_buf + 2) & 0xffffffff;
1151     gdt_limit = *(ushort_t*)(gdt_buf) & 0xffff;
1152     PrintDebug("GDT: base: %x, limit: %x\n", gdt_base, gdt_limit);
1153
1154     GetIDTR(idt_buf);
1155     idt_base = *(ulong_t*)(idt_buf + 2) & 0xffffffff;
1156     idt_limit = *(ushort_t*)(idt_buf) & 0xffff;
1157     PrintDebug("IDT: base: %x, limit: %x\n",idt_base, idt_limit);
1158
1159
1160     // gdt_base -= 0x2000;
1161     //idt_base -= 0x2000;
1162
1163     guest_state->gdtr.base = gdt_base;
1164     guest_state->gdtr.limit = gdt_limit;
1165     guest_state->idtr.base = idt_base;
1166     guest_state->idtr.limit = idt_limit;
1167
1168
1169   }
1170
1171
1172   // also determine if CPU supports nested paging
1173   /*
1174   if (vm_info.page_tables) {
1175     //   if (0) {
1176     // Flush the TLB on entries/exits
1177     ctrl_area->TLB_CONTROL = 1;
1178
1179     // Enable Nested Paging
1180     ctrl_area->NP_ENABLE = 1;
1181
1182     PrintDebug("NP_Enable at 0x%x\n", &(ctrl_area->NP_ENABLE));
1183
1184         // Set the Nested Page Table pointer
1185     ctrl_area->N_CR3 |= ((addr_t)vm_info.page_tables & 0xfffff000);
1186
1187
1188     //   ctrl_area->N_CR3 = Get_CR3();
1189     // guest_state->cr3 |= (Get_CR3() & 0xfffff000);
1190
1191     guest_state->g_pat = 0x7040600070406ULL;
1192
1193     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));
1194     PrintDebug("Set Guest CR3: lo: 0x%x  hi: 0x%x\n", (uint_t)*(&(guest_state->cr3)), (uint_t)*((unsigned char *)&(guest_state->cr3) + 4));
1195     // Enable Paging
1196     //    guest_state->cr0 |= 0x80000000;
1197   }
1198   */
1199
1200 }
1201
1202
1203
1204
1205
1206 #endif
1207
1208