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 uint32_t v3_last_exit;
  48
  49 // This is a global pointer to the host's VMCB
  50 static addr_t host_vmcbs[CONFIG_MAX_CPUS] = { [0 ... CONFIG_MAX_CPUS - 1] = 0};
  51
  52
  53
  54 extern void v3_stgi();
  55 extern void v3_clgi();
  56 //extern int v3_svm_launch(vmcb_t * vmcb, struct v3_gprs * vm_regs, uint64_t * fs, uint64_t * gs);
  57 extern int v3_svm_launch(vmcb_t * vmcb, struct v3_gprs * vm_regs, vmcb_t * host_vmcb);
  58
  59
  60 static vmcb_t * Allocate_VMCB() {
  61     vmcb_t * vmcb_page = (vmcb_t *)V3_VAddr(V3_AllocPages(1));
  62
  63     memset(vmcb_page, 0, 4096);
  64
  65     return vmcb_page;
  66 }
  67
  68
  69
  70 static void Init_VMCB_BIOS(vmcb_t * vmcb, struct guest_info * core) {
  71     vmcb_ctrl_t * ctrl_area = GET_VMCB_CTRL_AREA(vmcb);
  72     vmcb_saved_state_t * guest_state = GET_VMCB_SAVE_STATE_AREA(vmcb);
  73     uint_t i;
  74
  75
  76     //
  77
  78
  79     ctrl_area->svm_instrs.VMRUN = 1;
  80     ctrl_area->svm_instrs.VMMCALL = 1;
  81     ctrl_area->svm_instrs.VMLOAD = 1;
  82     ctrl_area->svm_instrs.VMSAVE = 1;
  83     ctrl_area->svm_instrs.STGI = 1;
  84     ctrl_area->svm_instrs.CLGI = 1;
  85     ctrl_area->svm_instrs.SKINIT = 1;
  86     ctrl_area->svm_instrs.RDTSCP = 1;
  87     ctrl_area->svm_instrs.ICEBP = 1;
  88     ctrl_area->svm_instrs.WBINVD = 1;
  89     ctrl_area->svm_instrs.MONITOR = 1;
  90     ctrl_area->svm_instrs.MWAIT_always = 1;
  91     ctrl_area->svm_instrs.MWAIT_if_armed = 1;
  92     ctrl_area->instrs.INVLPGA = 1;
  93     ctrl_area->instrs.CPUID = 1;
  94
  95     ctrl_area->instrs.HLT = 1;
  96     // guest_state->cr0 = 0x00000001;    // PE
  97
  98     /*
  99       ctrl_area->exceptions.de = 1;
 100       ctrl_area->exceptions.df = 1;
 101
 102       ctrl_area->exceptions.ts = 1;
 103       ctrl_area->exceptions.ss = 1;
 104       ctrl_area->exceptions.ac = 1;
 105       ctrl_area->exceptions.mc = 1;
 106       ctrl_area->exceptions.gp = 1;
 107       ctrl_area->exceptions.ud = 1;
 108       ctrl_area->exceptions.np = 1;
 109       ctrl_area->exceptions.of = 1;
 110
 111       ctrl_area->exceptions.nmi = 1;
 112     */
 113
 114
 115     ctrl_area->instrs.NMI = 1;
 116     ctrl_area->instrs.SMI = 0; // allow SMIs to run in guest
 117     ctrl_area->instrs.INIT = 1;
 118     ctrl_area->instrs.PAUSE = 1;
 119     ctrl_area->instrs.shutdown_evts = 1;
 120
 121
 122     /* DEBUG FOR RETURN CODE */
 123     ctrl_area->exit_code = 1;
 124
 125
 126     /* Setup Guest Machine state */
 127
 128     core->vm_regs.rsp = 0x00;
 129     core->rip = 0xfff0;
 130
 131     core->vm_regs.rdx = 0x00000f00;
 132
 133
 134     core->cpl = 0;
 135
 136     core->ctrl_regs.rflags = 0x00000002; // The reserved bit is always 1
 137     core->ctrl_regs.cr0 = 0x60010010; // Set the WP flag so the memory hooks work in real-mode
 138     core->ctrl_regs.efer |= EFER_MSR_svm_enable;
 139
 140
 141
 142
 143
 144     core->segments.cs.selector = 0xf000;
 145     core->segments.cs.limit = 0xffff;
 146     core->segments.cs.base = 0x0000000f0000LL;
 147
 148     // (raw attributes = 0xf3)
 149     core->segments.cs.type = 0x3;
 150     core->segments.cs.system = 0x1;
 151     core->segments.cs.dpl = 0x3;
 152     core->segments.cs.present = 1;
 153
 154
 155
 156     struct v3_segment * segregs [] = {&(core->segments.ss), &(core->segments.ds),
 157                                       &(core->segments.es), &(core->segments.fs),
 158                                       &(core->segments.gs), NULL};
 159
 160     for ( i = 0; segregs[i] != NULL; i++) {
 161         struct v3_segment * seg = segregs[i];
 162
 163         seg->selector = 0x0000;
 164         //    seg->base = seg->selector << 4;
 165         seg->base = 0x00000000;
 166         seg->limit = ~0u;
 167
 168         // (raw attributes = 0xf3)
 169         seg->type = 0x3;
 170         seg->system = 0x1;
 171         seg->dpl = 0x3;
 172         seg->present = 1;
 173     }
 174
 175     core->segments.gdtr.limit = 0x0000ffff;
 176     core->segments.gdtr.base = 0x0000000000000000LL;
 177     core->segments.idtr.limit = 0x0000ffff;
 178     core->segments.idtr.base = 0x0000000000000000LL;
 179
 180     core->segments.ldtr.selector = 0x0000;
 181     core->segments.ldtr.limit = 0x0000ffff;
 182     core->segments.ldtr.base = 0x0000000000000000LL;
 183     core->segments.tr.selector = 0x0000;
 184     core->segments.tr.limit = 0x0000ffff;
 185     core->segments.tr.base = 0x0000000000000000LL;
 186
 187
 188     core->dbg_regs.dr6 = 0x00000000ffff0ff0LL;
 189     core->dbg_regs.dr7 = 0x0000000000000400LL;
 190
 191
 192     ctrl_area->IOPM_BASE_PA = (addr_t)V3_PAddr(core->vm_info->io_map.arch_data);
 193     ctrl_area->instrs.IOIO_PROT = 1;
 194
 195     ctrl_area->MSRPM_BASE_PA = (addr_t)V3_PAddr(core->vm_info->msr_map.arch_data);
 196     ctrl_area->instrs.MSR_PROT = 1;
 197
 198
 199     PrintDebug("Exiting on interrupts\n");
 200     ctrl_area->guest_ctrl.V_INTR_MASKING = 1;
 201     ctrl_area->instrs.INTR = 1;
 202
 203
 204     if (core->shdw_pg_mode == SHADOW_PAGING) {
 205         PrintDebug("Creating initial shadow page table\n");
 206
 207         /* JRL: This is a performance killer, and a simplistic solution */
 208         /* We need to fix this */
 209         ctrl_area->TLB_CONTROL = 1;
 210         ctrl_area->guest_ASID = 1;
 211
 212
 213         if (v3_init_passthrough_pts(core) == -1) {
 214             PrintError("Could not initialize passthrough page tables\n");
 215             return ;
 216         }
 217
 218
 219         core->shdw_pg_state.guest_cr0 = 0x0000000000000010LL;
 220         PrintDebug("Created\n");
 221
 222         core->ctrl_regs.cr0 |= 0x80000000;
 223         core->ctrl_regs.cr3 = core->direct_map_pt;
 224
 225         ctrl_area->cr_reads.cr0 = 1;
 226         ctrl_area->cr_writes.cr0 = 1;
 227         //ctrl_area->cr_reads.cr4 = 1;
 228         ctrl_area->cr_writes.cr4 = 1;
 229         ctrl_area->cr_reads.cr3 = 1;
 230         ctrl_area->cr_writes.cr3 = 1;
 231
 232         v3_hook_msr(core->vm_info, EFER_MSR,
 233                     &v3_handle_efer_read,
 234                     &v3_handle_efer_write,
 235                     core);
 236
 237         ctrl_area->instrs.INVLPG = 1;
 238
 239         ctrl_area->exceptions.pf = 1;
 240
 241         guest_state->g_pat = 0x7040600070406ULL;
 242
 243
 244
 245     } else if (core->shdw_pg_mode == NESTED_PAGING) {
 246         // Flush the TLB on entries/exits
 247         ctrl_area->TLB_CONTROL = 1;
 248         ctrl_area->guest_ASID = 1;
 249
 250         // Enable Nested Paging
 251         ctrl_area->NP_ENABLE = 1;
 252
 253         PrintDebug("NP_Enable at 0x%p\n", (void *)&(ctrl_area->NP_ENABLE));
 254
 255         // Set the Nested Page Table pointer
 256         if (v3_init_passthrough_pts(core) == -1) {
 257             PrintError("Could not initialize Nested page tables\n");
 258             return ;
 259         }
 260
 261         ctrl_area->N_CR3 = core->direct_map_pt;
 262
 263         guest_state->g_pat = 0x7040600070406ULL;
 264     }
 265 }
 266
 267
 268 int v3_init_svm_vmcb(struct guest_info * info, v3_vm_class_t vm_class) {
 269
 270     PrintDebug("Allocating VMCB\n");
 271     info->vmm_data = (void*)Allocate_VMCB();
 272
 273     if (vm_class == V3_PC_VM) {
 274         PrintDebug("Initializing VMCB (addr=%p)\n", (void *)info->vmm_data);
 275         Init_VMCB_BIOS((vmcb_t*)(info->vmm_data), info);
 276     } else {
 277         PrintError("Invalid VM class\n");
 278         return -1;
 279     }
 280
 281     return 0;
 282 }
 283
 284
 285
 286 static int update_irq_exit_state(struct guest_info * info) {
 287     vmcb_ctrl_t * guest_ctrl = GET_VMCB_CTRL_AREA((vmcb_t*)(info->vmm_data));
 288
 289     // Fix for QEMU bug using EVENTINJ as an internal cache
 290     guest_ctrl->EVENTINJ.valid = 0;
 291
 292     if ((info->intr_core_state.irq_pending == 1) && (guest_ctrl->guest_ctrl.V_IRQ == 0)) {
 293
 294 #ifdef CONFIG_DEBUG_INTERRUPTS
 295         PrintDebug("INTAK cycle completed for irq %d\n", info->intr_core_state.irq_vector);
 296 #endif
 297
 298         info->intr_core_state.irq_started = 1;
 299         info->intr_core_state.irq_pending = 0;
 300
 301         v3_injecting_intr(info, info->intr_core_state.irq_vector, V3_EXTERNAL_IRQ);
 302     }
 303
 304     if ((info->intr_core_state.irq_started == 1) && (guest_ctrl->exit_int_info.valid == 0)) {
 305 #ifdef CONFIG_DEBUG_INTERRUPTS
 306         PrintDebug("Interrupt %d taken by guest\n", info->intr_core_state.irq_vector);
 307 #endif
 308
 309         // Interrupt was taken fully vectored
 310         info->intr_core_state.irq_started = 0;
 311
 312     } else if ((info->intr_core_state.irq_started == 1) && (guest_ctrl->exit_int_info.valid == 1)) {
 313 #ifdef CONFIG_DEBUG_INTERRUPTS
 314         PrintDebug("EXIT INT INFO is set (vec=%d)\n", guest_ctrl->exit_int_info.vector);
 315 #endif
 316     }
 317
 318     return 0;
 319 }
 320
 321
 322 static int update_irq_entry_state(struct guest_info * info) {
 323     vmcb_ctrl_t * guest_ctrl = GET_VMCB_CTRL_AREA((vmcb_t*)(info->vmm_data));
 324
 325
 326     if (info->intr_core_state.irq_pending == 0) {
 327         guest_ctrl->guest_ctrl.V_IRQ = 0;
 328         guest_ctrl->guest_ctrl.V_INTR_VECTOR = 0;
 329     }
 330
 331     if (v3_excp_pending(info)) {
 332         uint_t excp = v3_get_excp_number(info);
 333
 334         guest_ctrl->EVENTINJ.type = SVM_INJECTION_EXCEPTION;
 335
 336         if (info->excp_state.excp_error_code_valid) {
 337             guest_ctrl->EVENTINJ.error_code = info->excp_state.excp_error_code;
 338             guest_ctrl->EVENTINJ.ev = 1;
 339 #ifdef CONFIG_DEBUG_INTERRUPTS
 340             PrintDebug("Injecting exception %d with error code %x\n", excp, guest_ctrl->EVENTINJ.error_code);
 341 #endif
 342         }
 343
 344         guest_ctrl->EVENTINJ.vector = excp;
 345
 346         guest_ctrl->EVENTINJ.valid = 1;
 347
 348 #ifdef CONFIG_DEBUG_INTERRUPTS
 349         PrintDebug("<%d> Injecting Exception %d (CR2=%p) (EIP=%p)\n",
 350                    (int)info->num_exits,
 351                    guest_ctrl->EVENTINJ.vector,
 352                    (void *)(addr_t)info->ctrl_regs.cr2,
 353                    (void *)(addr_t)info->rip);
 354 #endif
 355
 356         v3_injecting_excp(info, excp);
 357     } else if (info->intr_core_state.irq_started == 1) {
 358 #ifdef CONFIG_DEBUG_INTERRUPTS
 359         PrintDebug("IRQ pending from previous injection\n");
 360 #endif
 361         guest_ctrl->guest_ctrl.V_IRQ = 1;
 362         guest_ctrl->guest_ctrl.V_INTR_VECTOR = info->intr_core_state.irq_vector;
 363         guest_ctrl->guest_ctrl.V_IGN_TPR = 1;
 364         guest_ctrl->guest_ctrl.V_INTR_PRIO = 0xf;
 365
 366     } else {
 367         switch (v3_intr_pending(info)) {
 368             case V3_EXTERNAL_IRQ: {
 369                 uint32_t irq = v3_get_intr(info);
 370
 371                 guest_ctrl->guest_ctrl.V_IRQ = 1;
 372                 guest_ctrl->guest_ctrl.V_INTR_VECTOR = irq;
 373                 guest_ctrl->guest_ctrl.V_IGN_TPR = 1;
 374                 guest_ctrl->guest_ctrl.V_INTR_PRIO = 0xf;
 375
 376 #ifdef CONFIG_DEBUG_INTERRUPTS
 377                 PrintDebug("Injecting Interrupt %d (EIP=%p)\n",
 378                            guest_ctrl->guest_ctrl.V_INTR_VECTOR,
 379                            (void *)(addr_t)info->rip);
 380 #endif
 381
 382                 info->intr_core_state.irq_pending = 1;
 383                 info->intr_core_state.irq_vector = irq;
 384
 385                 break;
 386             }
 387             case V3_NMI:
 388                 guest_ctrl->EVENTINJ.type = SVM_INJECTION_NMI;
 389                 break;
 390             case V3_SOFTWARE_INTR:
 391                 guest_ctrl->EVENTINJ.type = SVM_INJECTION_SOFT_INTR;
 392                 break;
 393             case V3_VIRTUAL_IRQ:
 394                 guest_ctrl->EVENTINJ.type = SVM_INJECTION_IRQ;
 395                 break;
 396
 397             case V3_INVALID_INTR:
 398             default:
 399                 break;
 400         }
 401
 402     }
 403
 404     return 0;
 405 }
 406
 407
 408 /*
 409  * CAUTION and DANGER!!!
 410  *
 411  * The VMCB CANNOT(!!) be accessed outside of the clgi/stgi calls inside this function
 412  * When exectuing a symbiotic call, the VMCB WILL be overwritten, so any dependencies
 413  * on its contents will cause things to break. The contents at the time of the exit WILL
 414  * change before the exit handler is executed.
 415  */
 416 int v3_svm_enter(struct guest_info * info) {
 417     vmcb_ctrl_t * guest_ctrl = GET_VMCB_CTRL_AREA((vmcb_t*)(info->vmm_data));
 418     vmcb_saved_state_t * guest_state = GET_VMCB_SAVE_STATE_AREA((vmcb_t*)(info->vmm_data));
 419     addr_t exit_code = 0, exit_info1 = 0, exit_info2 = 0;
 420
 421     // Conditionally yield the CPU if the timeslice has expired
 422     v3_yield_cond(info);
 423
 424     // disable global interrupts for vm state transition
 425     v3_clgi();
 426
 427     // Synchronize the guest state to the VMCB
 428     guest_state->cr0 = info->ctrl_regs.cr0;
 429     guest_state->cr2 = info->ctrl_regs.cr2;
 430     guest_state->cr3 = info->ctrl_regs.cr3;
 431     guest_state->cr4 = info->ctrl_regs.cr4;
 432     guest_state->dr6 = info->dbg_regs.dr6;
 433     guest_state->dr7 = info->dbg_regs.dr7;
 434     guest_ctrl->guest_ctrl.V_TPR = info->ctrl_regs.cr8 & 0xff;
 435     guest_state->rflags = info->ctrl_regs.rflags;
 436     guest_state->efer = info->ctrl_regs.efer;
 437
 438     guest_state->cpl = info->cpl;
 439
 440     v3_set_vmcb_segments((vmcb_t*)(info->vmm_data), &(info->segments));
 441
 442     guest_state->rax = info->vm_regs.rax;
 443     guest_state->rip = info->rip;
 444     guest_state->rsp = info->vm_regs.rsp;
 445
 446 #ifdef CONFIG_SYMCALL
 447     if (info->sym_core_state.symcall_state.sym_call_active == 0) {
 448         update_irq_entry_state(info);
 449     }
 450 #else
 451     update_irq_entry_state(info);
 452 #endif
 453
 454
 455     /* ** */
 456
 457     /*
 458       PrintDebug("SVM Entry to CS=%p  rip=%p...\n",
 459       (void *)(addr_t)info->segments.cs.base,
 460       (void *)(addr_t)info->rip);
 461     */
 462
 463 #ifdef CONFIG_SYMCALL
 464     if (info->sym_core_state.symcall_state.sym_call_active == 1) {
 465         if (guest_ctrl->guest_ctrl.V_IRQ == 1) {
 466             V3_Print("!!! Injecting Interrupt during Sym call !!!\n");
 467         }
 468     }
 469 #endif
 470
 471
 472     v3_update_timers(info);
 473     v3_resume_time(info);
 474
 475     guest_ctrl->TSC_OFFSET = info->time_state.time_offset
 476         + info->time_state.tsc_time_offset;
 477
 478     //V3_Print("Calling v3_svm_launch\n");
 479
 480     v3_svm_launch((vmcb_t *)V3_PAddr(info->vmm_data), &(info->vm_regs), (vmcb_t *)host_vmcbs[info->cpu_id]);
 481
 482     v3_pause_time(info);
 483 #ifdef CONFIG_TIME_MASK_OVERHEAD
 484     v3_offset_time(info, -SVM_ENTRY_OVERHEAD);
 485 #endif
 486
 487     //V3_Print("SVM Returned: Exit Code: %x, guest_rip=%lx\n", (uint32_t)(guest_ctrl->exit_code), (unsigned long)guest_state->rip);
 488
 489     v3_last_exit = (uint32_t)(guest_ctrl->exit_code);
 490
 491     //PrintDebug("SVM Returned\n");
 492
 493     info->num_exits++;
 494
 495     // Save Guest state from VMCB
 496     info->rip = guest_state->rip;
 497     info->vm_regs.rsp = guest_state->rsp;
 498     info->vm_regs.rax = guest_state->rax;
 499
 500     info->cpl = guest_state->cpl;
 501
 502     info->ctrl_regs.cr0 = guest_state->cr0;
 503     info->ctrl_regs.cr2 = guest_state->cr2;
 504     info->ctrl_regs.cr3 = guest_state->cr3;
 505     info->ctrl_regs.cr4 = guest_state->cr4;
 506     info->dbg_regs.dr6 = guest_state->dr6;
 507     info->dbg_regs.dr7 = guest_state->dr7;
 508     info->ctrl_regs.cr8 = guest_ctrl->guest_ctrl.V_TPR;
 509     info->ctrl_regs.rflags = guest_state->rflags;
 510     info->ctrl_regs.efer = guest_state->efer;
 511
 512     v3_get_vmcb_segments((vmcb_t*)(info->vmm_data), &(info->segments));
 513     info->cpu_mode = v3_get_vm_cpu_mode(info);
 514     info->mem_mode = v3_get_vm_mem_mode(info);
 515     /* ** */
 516
 517
 518     // save exit info here
 519     exit_code = guest_ctrl->exit_code;
 520     exit_info1 = guest_ctrl->exit_info1;
 521     exit_info2 = guest_ctrl->exit_info2;
 522
 523
 524 #ifdef CONFIG_SYMCALL
 525     if (info->sym_core_state.symcall_state.sym_call_active == 0) {
 526         update_irq_exit_state(info);
 527     }
 528 #else
 529     update_irq_exit_state(info);
 530 #endif
 531
 532
 533     // reenable global interrupts after vm exit
 534     v3_stgi();
 535
 536
 537     // Conditionally yield the CPU if the timeslice has expired
 538     v3_yield_cond(info);
 539
 540
 541
 542     if (v3_handle_svm_exit(info, exit_code, exit_info1, exit_info2) != 0) {
 543         PrintError("Error in SVM exit handler\n");
 544         return -1;
 545     }
 546
 547
 548     return 0;
 549 }
 550
 551
 552 int v3_start_svm_guest(struct guest_info *info) {
 553     //    vmcb_saved_state_t * guest_state = GET_VMCB_SAVE_STATE_AREA((vmcb_t*)(info->vmm_data));
 554     //  vmcb_ctrl_t * guest_ctrl = GET_VMCB_CTRL_AREA((vmcb_t*)(info->vmm_data));
 555
 556
 557     PrintDebug("Starting SVM core %u\n",info->cpu_id);
 558     if (info->cpu_mode==INIT) {
 559         PrintDebug("SVM core %u: I am an AP in INIT mode, waiting for that to change\n",info->cpu_id);
 560         while (info->cpu_mode==INIT) {
 561             v3_yield(info);
 562             //PrintDebug("SVM core %u: still waiting for INIT\n",info->cpu_id);
 563         }
 564         PrintDebug("SVM core %u: I am out of INIT\n",info->cpu_id);
 565         if (info->cpu_mode==SIPI) {
 566             PrintDebug("SVM core %u: I am waiting on a SIPI to set my starting address\n",info->cpu_id);
 567             while (info->cpu_mode==SIPI) {
 568                 v3_yield(info);
 569                 //PrintDebug("SVM core %u: still waiting for SIPI\n",info->cpu_id);
 570             }
 571         }
 572         PrintDebug("SVM core %u: I have my SIPI\n", info->cpu_id);
 573     }
 574
 575     if (info->cpu_mode!=REAL) {
 576         PrintError("SVM core %u: I am not in REAL mode at launch!  Huh?!\n", info->cpu_id);
 577         return -1;
 578     }
 579
 580     PrintDebug("SVM core %u: I am starting at CS=0x%x (base=0x%p, limit=0x%x),  RIP=0x%p\n",
 581                info->cpu_id, info->segments.cs.selector, (void*)(info->segments.cs.base),
 582                info->segments.cs.limit,(void*)(info->rip));
 583
 584
 585
 586     PrintDebug("SVM core %u: Launching SVM VM (vmcb=%p)\n", info->cpu_id, (void *)info->vmm_data);
 587     //PrintDebugVMCB((vmcb_t*)(info->vmm_data));
 588
 589     info->vm_info->run_state = VM_RUNNING;
 590     v3_start_time(info);
 591
 592     while (1) {
 593         if (v3_svm_enter(info) == -1) {
 594             vmcb_ctrl_t * guest_ctrl = GET_VMCB_CTRL_AREA((vmcb_t*)(info->vmm_data));
 595             addr_t host_addr;
 596             addr_t linear_addr = 0;
 597
 598             info->vm_info->run_state = VM_ERROR;
 599
 600             V3_Print("SVM core %u: SVM ERROR!!\n", info->cpu_id);
 601
 602             v3_print_guest_state(info);
 603
 604             V3_Print("SVM core %u: SVM Exit Code: %p\n", info->cpu_id, (void *)(addr_t)guest_ctrl->exit_code);
 605
 606             V3_Print("SVM core %u: exit_info1 low = 0x%.8x\n", info->cpu_id, *(uint_t*)&(guest_ctrl->exit_info1));
 607             V3_Print("SVM core %u: exit_info1 high = 0x%.8x\n", info->cpu_id, *(uint_t *)(((uchar_t *)&(guest_ctrl->exit_info1)) + 4));
 608
 609             V3_Print("SVM core %u: exit_info2 low = 0x%.8x\n", info->cpu_id, *(uint_t*)&(guest_ctrl->exit_info2));
 610             V3_Print("SVM core %u: exit_info2 high = 0x%.8x\n", info->cpu_id, *(uint_t *)(((uchar_t *)&(guest_ctrl->exit_info2)) + 4));
 611
 612             linear_addr = get_addr_linear(info, info->rip, &(info->segments.cs));
 613
 614             if (info->mem_mode == PHYSICAL_MEM) {
 615                 v3_gpa_to_hva(info, linear_addr, &host_addr);
 616             } else if (info->mem_mode == VIRTUAL_MEM) {
 617                 v3_gva_to_hva(info, linear_addr, &host_addr);
 618             }
 619
 620             V3_Print("SVM core %u: Host Address of rip = 0x%p\n", info->cpu_id, (void *)host_addr);
 621
 622             V3_Print("SVM core %u: Instr (15 bytes) at %p:\n", info->cpu_id, (void *)host_addr);
 623             v3_dump_mem((uint8_t *)host_addr, 15);
 624
 625             v3_print_stack(info);
 626
 627             break;
 628         }
 629
 630 /*
 631         if ((info->num_exits % 5000) == 0) {
 632             V3_Print("SVM Exit number %d\n", (uint32_t)info->num_exits);
 633         }
 634 */
 635
 636     }
 637
 638     // Need to take down the other cores on error...
 639
 640     return 0;
 641 }
 642
 643
 644
 645
 646
 647 /* Checks machine SVM capability */
 648 /* Implemented from: AMD Arch Manual 3, sect 15.4 */
 649 int v3_is_svm_capable() {
 650     uint_t vm_cr_low = 0, vm_cr_high = 0;
 651     uint32_t eax = 0, ebx = 0, ecx = 0, edx = 0;
 652
 653     v3_cpuid(CPUID_EXT_FEATURE_IDS, &eax, &ebx, &ecx, &edx);
 654
 655     PrintDebug("CPUID_EXT_FEATURE_IDS_ecx=0x%x\n", ecx);
 656
 657     if ((ecx & CPUID_EXT_FEATURE_IDS_ecx_svm_avail) == 0) {
 658       V3_Print("SVM Not Available\n");
 659       return 0;
 660     }  else {
 661         v3_get_msr(SVM_VM_CR_MSR, &vm_cr_high, &vm_cr_low);
 662
 663         PrintDebug("SVM_VM_CR_MSR = 0x%x 0x%x\n", vm_cr_high, vm_cr_low);
 664
 665         if ((vm_cr_low & SVM_VM_CR_MSR_svmdis) == 1) {
 666             V3_Print("SVM is available but is disabled.\n");
 667
 668             v3_cpuid(CPUID_SVM_REV_AND_FEATURE_IDS, &eax, &ebx, &ecx, &edx);
 669
 670             PrintDebug("CPUID_SVM_REV_AND_FEATURE_IDS_edx=0x%x\n", edx);
 671
 672             if ((edx & CPUID_SVM_REV_AND_FEATURE_IDS_edx_svml) == 0) {
 673                 V3_Print("SVM BIOS Disabled, not unlockable\n");
 674             } else {
 675                 V3_Print("SVM is locked with a key\n");
 676             }
 677             return 0;
 678
 679         } else {
 680             V3_Print("SVM is available and  enabled.\n");
 681
 682             v3_cpuid(CPUID_SVM_REV_AND_FEATURE_IDS, &eax, &ebx, &ecx, &edx);
 683             PrintDebug("CPUID_SVM_REV_AND_FEATURE_IDS_eax=0x%x\n", eax);
 684             PrintDebug("CPUID_SVM_REV_AND_FEATURE_IDS_ebx=0x%x\n", ebx);
 685             PrintDebug("CPUID_SVM_REV_AND_FEATURE_IDS_ecx=0x%x\n", ecx);
 686             PrintDebug("CPUID_SVM_REV_AND_FEATURE_IDS_edx=0x%x\n", edx);
 687
 688             return 1;
 689         }
 690     }
 691 }
 692
 693 static int has_svm_nested_paging() {
 694     uint32_t eax = 0, ebx = 0, ecx = 0, edx = 0;
 695
 696     v3_cpuid(CPUID_SVM_REV_AND_FEATURE_IDS, &eax, &ebx, &ecx, &edx);
 697
 698     //PrintDebug("CPUID_EXT_FEATURE_IDS_edx=0x%x\n", edx);
 699
 700     if ((edx & CPUID_SVM_REV_AND_FEATURE_IDS_edx_np) == 0) {
 701         V3_Print("SVM Nested Paging not supported\n");
 702         return 0;
 703     } else {
 704         V3_Print("SVM Nested Paging supported\n");
 705         return 1;
 706     }
 707 }
 708
 709
 710 void v3_init_svm_cpu(int cpu_id) {
 711     reg_ex_t msr;
 712     extern v3_cpu_arch_t v3_cpu_types[];
 713
 714     // Enable SVM on the CPU
 715     v3_get_msr(EFER_MSR, &(msr.e_reg.high), &(msr.e_reg.low));
 716     msr.e_reg.low |= EFER_MSR_svm_enable;
 717     v3_set_msr(EFER_MSR, 0, msr.e_reg.low);
 718
 719     V3_Print("SVM Enabled\n");
 720
 721     // Setup the host state save area
 722     host_vmcbs[cpu_id] = (addr_t)V3_AllocPages(4);
 723
 724     /* 64-BIT-ISSUE */
 725     //  msr.e_reg.high = 0;
 726     //msr.e_reg.low = (uint_t)host_vmcb;
 727     msr.r_reg = host_vmcbs[cpu_id];
 728
 729     PrintDebug("Host State being saved at %p\n", (void *)host_vmcbs[cpu_id]);
 730     v3_set_msr(SVM_VM_HSAVE_PA_MSR, msr.e_reg.high, msr.e_reg.low);
 731
 732
 733     if (has_svm_nested_paging() == 1) {
 734         v3_cpu_types[cpu_id] = V3_SVM_REV3_CPU;
 735     } else {
 736         v3_cpu_types[cpu_id] = V3_SVM_CPU;
 737     }
 738 }
 739
 740
 741
 742
 743
 744
 745
 746
 747
 748
 749
 750
 751
 752
 753
 754
 755
 756
 757
 758
 759
 760
 761
 762
 763
 764
 765
 766
 767
 768
 769
 770
 771
 772
 773
 774
 775
 776
 777
 778
 779
 780
 781
 782
 783
 784
 785
 786
 787
 788
 789
 790
 791
 792
 793 #if 0
 794 /*
 795  * Test VMSAVE/VMLOAD Latency
 796  */
 797 #define vmsave ".byte 0x0F,0x01,0xDB ; "
 798 #define vmload ".byte 0x0F,0x01,0xDA ; "
 799 {
 800     uint32_t start_lo, start_hi;
 801     uint32_t end_lo, end_hi;
 802     uint64_t start, end;
 803
 804     __asm__ __volatile__ (
 805                           "rdtsc ; "
 806                           "movl %%eax, %%esi ; "
 807                           "movl %%edx, %%edi ; "
 808                           "movq  %%rcx, %%rax ; "
 809                           vmsave
 810                           "rdtsc ; "
 811                           : "=D"(start_hi), "=S"(start_lo), "=a"(end_lo),"=d"(end_hi)
 812                           : "c"(host_vmcb[cpu_id]), "0"(0), "1"(0), "2"(0), "3"(0)
 813                           );
 814
 815     start = start_hi;
 816     start <<= 32;
 817     start += start_lo;
 818
 819     end = end_hi;
 820     end <<= 32;
 821     end += end_lo;
 822
 823     PrintDebug("VMSave Cycle Latency: %d\n", (uint32_t)(end - start));
 824
 825     __asm__ __volatile__ (
 826                           "rdtsc ; "
 827                           "movl %%eax, %%esi ; "
 828                           "movl %%edx, %%edi ; "
 829                           "movq  %%rcx, %%rax ; "
 830                           vmload
 831                           "rdtsc ; "
 832                           : "=D"(start_hi), "=S"(start_lo), "=a"(end_lo),"=d"(end_hi)
 833                               : "c"(host_vmcb[cpu_id]), "0"(0), "1"(0), "2"(0), "3"(0)
 834                               );
 835
 836         start = start_hi;
 837         start <<= 32;
 838         start += start_lo;
 839
 840         end = end_hi;
 841         end <<= 32;
 842         end += end_lo;
 843
 844
 845         PrintDebug("VMLoad Cycle Latency: %d\n", (uint32_t)(end - start));
 846     }
 847     /* End Latency Test */
 848
 849 #endif
 850
 851
 852
 853
 854
 855
 856
 857 #if 0
 858 void Init_VMCB_pe(vmcb_t *vmcb, struct guest_info vm_info) {
 859   vmcb_ctrl_t * ctrl_area = GET_VMCB_CTRL_AREA(vmcb);
 860   vmcb_saved_state_t * guest_state = GET_VMCB_SAVE_STATE_AREA(vmcb);
 861   uint_t i = 0;
 862
 863
 864   guest_state->rsp = vm_info.vm_regs.rsp;
 865   guest_state->rip = vm_info.rip;
 866
 867
 868   /* I pretty much just gutted this from TVMM */
 869   /* Note: That means its probably wrong */
 870
 871   // set the segment registers to mirror ours
 872   guest_state->cs.selector = 1<<3;
 873   guest_state->cs.attrib.fields.type = 0xa; // Code segment+read
 874   guest_state->cs.attrib.fields.S = 1;
 875   guest_state->cs.attrib.fields.P = 1;
 876   guest_state->cs.attrib.fields.db = 1;
 877   guest_state->cs.attrib.fields.G = 1;
 878   guest_state->cs.limit = 0xfffff;
 879   guest_state->cs.base = 0;
 880
 881   struct vmcb_selector *segregs [] = {&(guest_state->ss), &(guest_state->ds), &(guest_state->es), &(guest_state->fs), &(guest_state->gs), NULL};
 882   for ( i = 0; segregs[i] != NULL; i++) {
 883     struct vmcb_selector * seg = segregs[i];
 884
 885     seg->selector = 2<<3;
 886     seg->attrib.fields.type = 0x2; // Data Segment+read/write
 887     seg->attrib.fields.S = 1;
 888     seg->attrib.fields.P = 1;
 889     seg->attrib.fields.db = 1;
 890     seg->attrib.fields.G = 1;
 891     seg->limit = 0xfffff;
 892     seg->base = 0;
 893   }
 894
 895
 896   {
 897     /* JRL THIS HAS TO GO */
 898
 899     //    guest_state->tr.selector = GetTR_Selector();
 900     guest_state->tr.attrib.fields.type = 0x9;
 901     guest_state->tr.attrib.fields.P = 1;
 902     // guest_state->tr.limit = GetTR_Limit();
 903     //guest_state->tr.base = GetTR_Base();// - 0x2000;
 904     /* ** */
 905   }
 906
 907
 908   /* ** */
 909
 910
 911   guest_state->efer |= EFER_MSR_svm_enable;
 912   guest_state->rflags = 0x00000002; // The reserved bit is always 1
 913   ctrl_area->svm_instrs.VMRUN = 1;
 914   guest_state->cr0 = 0x00000001;    // PE
 915   ctrl_area->guest_ASID = 1;
 916
 917
 918   //  guest_state->cpl = 0;
 919
 920
 921
 922   // Setup exits
 923
 924   ctrl_area->cr_writes.cr4 = 1;
 925
 926   ctrl_area->exceptions.de = 1;
 927   ctrl_area->exceptions.df = 1;
 928   ctrl_area->exceptions.pf = 1;
 929   ctrl_area->exceptions.ts = 1;
 930   ctrl_area->exceptions.ss = 1;
 931   ctrl_area->exceptions.ac = 1;
 932   ctrl_area->exceptions.mc = 1;
 933   ctrl_area->exceptions.gp = 1;
 934   ctrl_area->exceptions.ud = 1;
 935   ctrl_area->exceptions.np = 1;
 936   ctrl_area->exceptions.of = 1;
 937   ctrl_area->exceptions.nmi = 1;
 938
 939
 940
 941   ctrl_area->instrs.IOIO_PROT = 1;
 942   ctrl_area->IOPM_BASE_PA = (uint_t)V3_AllocPages(3);
 943
 944   {
 945     reg_ex_t tmp_reg;
 946     tmp_reg.r_reg = ctrl_area->IOPM_BASE_PA;
 947     memset((void*)(tmp_reg.e_reg.low), 0xffffffff, PAGE_SIZE * 2);
 948   }
 949
 950   ctrl_area->instrs.INTR = 1;
 951
 952
 953   {
 954     char gdt_buf[6];
 955     char idt_buf[6];
 956
 957     memset(gdt_buf, 0, 6);
 958     memset(idt_buf, 0, 6);
 959
 960
 961     uint_t gdt_base, idt_base;
 962     ushort_t gdt_limit, idt_limit;
 963
 964     GetGDTR(gdt_buf);
 965     gdt_base = *(ulong_t*)((uchar_t*)gdt_buf + 2) & 0xffffffff;
 966     gdt_limit = *(ushort_t*)(gdt_buf) & 0xffff;
 967     PrintDebug("GDT: base: %x, limit: %x\n", gdt_base, gdt_limit);
 968
 969     GetIDTR(idt_buf);
 970     idt_base = *(ulong_t*)(idt_buf + 2) & 0xffffffff;
 971     idt_limit = *(ushort_t*)(idt_buf) & 0xffff;
 972     PrintDebug("IDT: base: %x, limit: %x\n",idt_base, idt_limit);
 973
 974
 975     // gdt_base -= 0x2000;
 976     //idt_base -= 0x2000;
 977
 978     guest_state->gdtr.base = gdt_base;
 979     guest_state->gdtr.limit = gdt_limit;
 980     guest_state->idtr.base = idt_base;
 981     guest_state->idtr.limit = idt_limit;
 982
 983
 984   }
 985
 986
 987   // also determine if CPU supports nested paging
 988   /*
 989   if (vm_info.page_tables) {
 990     //   if (0) {
 991     // Flush the TLB on entries/exits
 992     ctrl_area->TLB_CONTROL = 1;
 993
 994     // Enable Nested Paging
 995     ctrl_area->NP_ENABLE = 1;
 996
 997     PrintDebug("NP_Enable at 0x%x\n", &(ctrl_area->NP_ENABLE));
 998
 999         // Set the Nested Page Table pointer
1000     ctrl_area->N_CR3 |= ((addr_t)vm_info.page_tables & 0xfffff000);
1001
1002
1003     //   ctrl_area->N_CR3 = Get_CR3();
1004     // guest_state->cr3 |= (Get_CR3() & 0xfffff000);
1005
1006     guest_state->g_pat = 0x7040600070406ULL;
1007
1008     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));
1009     PrintDebug("Set Guest CR3: lo: 0x%x  hi: 0x%x\n", (uint_t)*(&(guest_state->cr3)), (uint_t)*((unsigned char *)&(guest_state->cr3) + 4));
1010     // Enable Paging
1011     //    guest_state->cr0 |= 0x80000000;
1012   }
1013   */
1014
1015 }
1016
1017
1018
1019
1020
1021 #endif
1022
1023