vmx.c revision 249351
1/*- 2 * Copyright (c) 2011 NetApp, Inc. 3 * All rights reserved. 4 * 5 * Redistribution and use in source and binary forms, with or without 6 * modification, are permitted provided that the following conditions 7 * are met: 8 * 1. Redistributions of source code must retain the above copyright 9 * notice, this list of conditions and the following disclaimer. 10 * 2. Redistributions in binary form must reproduce the above copyright 11 * notice, this list of conditions and the following disclaimer in the 12 * documentation and/or other materials provided with the distribution. 13 * 14 * THIS SOFTWARE IS PROVIDED BY NETAPP, INC ``AS IS'' AND 15 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 16 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 17 * ARE DISCLAIMED. IN NO EVENT SHALL NETAPP, INC OR CONTRIBUTORS BE LIABLE 18 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 19 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 20 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 21 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 22 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 23 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 24 * SUCH DAMAGE. 25 * 26 * $FreeBSD: head/sys/amd64/vmm/intel/vmx.c 249351 2013-04-11 04:29:45Z neel $ 27 */ 28 29#include <sys/cdefs.h> 30__FBSDID("$FreeBSD: head/sys/amd64/vmm/intel/vmx.c 249351 2013-04-11 04:29:45Z neel $"); 31 32#include <sys/param.h> 33#include <sys/systm.h> 34#include <sys/smp.h> 35#include <sys/kernel.h> 36#include <sys/malloc.h> 37#include <sys/pcpu.h> 38#include <sys/proc.h> 39 40#include <vm/vm.h> 41#include <vm/pmap.h> 42 43#include <machine/psl.h> 44#include <machine/cpufunc.h> 45#include <machine/md_var.h> 46#include <machine/pmap.h> 47#include <machine/segments.h> 48#include <machine/specialreg.h> 49#include <machine/vmparam.h> 50 51#include <x86/apicreg.h> 52 53#include <machine/vmm.h> 54#include "vmm_host.h" 55#include "vmm_lapic.h" 56#include "vmm_msr.h" 57#include "vmm_ktr.h" 58#include "vmm_stat.h" 59 60#include "vmx_msr.h" 61#include "ept.h" 62#include "vmx_cpufunc.h" 63#include "vmx.h" 64#include "x86.h" 65#include "vmx_controls.h" 66 67#define PINBASED_CTLS_ONE_SETTING \ 68 (PINBASED_EXTINT_EXITING | \ 69 PINBASED_NMI_EXITING | \ 70 PINBASED_VIRTUAL_NMI) 71#define PINBASED_CTLS_ZERO_SETTING 0 72 73#define PROCBASED_CTLS_WINDOW_SETTING \ 74 (PROCBASED_INT_WINDOW_EXITING | \ 75 PROCBASED_NMI_WINDOW_EXITING) 76 77#define PROCBASED_CTLS_ONE_SETTING \ 78 (PROCBASED_SECONDARY_CONTROLS | \ 79 PROCBASED_IO_EXITING | \ 80 PROCBASED_MSR_BITMAPS | \ 81 PROCBASED_CTLS_WINDOW_SETTING) 82#define PROCBASED_CTLS_ZERO_SETTING \ 83 (PROCBASED_CR3_LOAD_EXITING | \ 84 PROCBASED_CR3_STORE_EXITING | \ 85 PROCBASED_IO_BITMAPS) 86 87#define PROCBASED_CTLS2_ONE_SETTING PROCBASED2_ENABLE_EPT 88#define PROCBASED_CTLS2_ZERO_SETTING 0 89 90#define VM_EXIT_CTLS_ONE_SETTING_NO_PAT \ 91 (VM_EXIT_HOST_LMA | \ 92 VM_EXIT_SAVE_EFER | \ 93 VM_EXIT_LOAD_EFER) 94 95#define VM_EXIT_CTLS_ONE_SETTING \ 96 (VM_EXIT_CTLS_ONE_SETTING_NO_PAT | \ 97 VM_EXIT_SAVE_PAT | \ 98 VM_EXIT_LOAD_PAT) 99#define VM_EXIT_CTLS_ZERO_SETTING VM_EXIT_SAVE_DEBUG_CONTROLS 100 101#define VM_ENTRY_CTLS_ONE_SETTING_NO_PAT VM_ENTRY_LOAD_EFER 102 103#define VM_ENTRY_CTLS_ONE_SETTING \ 104 (VM_ENTRY_CTLS_ONE_SETTING_NO_PAT | \ 105 VM_ENTRY_LOAD_PAT) 106#define VM_ENTRY_CTLS_ZERO_SETTING \ 107 (VM_ENTRY_LOAD_DEBUG_CONTROLS | \ 108 VM_ENTRY_INTO_SMM | \ 109 VM_ENTRY_DEACTIVATE_DUAL_MONITOR) 110 111#define guest_msr_rw(vmx, msr) \ 112 msr_bitmap_change_access((vmx)->msr_bitmap, (msr), MSR_BITMAP_ACCESS_RW) 113 114#define HANDLED 1 115#define UNHANDLED 0 116 117MALLOC_DEFINE(M_VMX, "vmx", "vmx"); 118 119int vmxon_enabled[MAXCPU]; 120static char vmxon_region[MAXCPU][PAGE_SIZE] __aligned(PAGE_SIZE); 121 122static uint32_t pinbased_ctls, procbased_ctls, procbased_ctls2; 123static uint32_t exit_ctls, entry_ctls; 124 125static uint64_t cr0_ones_mask, cr0_zeros_mask; 126static uint64_t cr4_ones_mask, cr4_zeros_mask; 127 128static volatile u_int nextvpid; 129 130static int vmx_no_patmsr; 131 132/* 133 * Virtual NMI blocking conditions. 134 * 135 * Some processor implementations also require NMI to be blocked if 136 * the STI_BLOCKING bit is set. It is possible to detect this at runtime 137 * based on the (exit_reason,exit_qual) tuple being set to 138 * (EXIT_REASON_INVAL_VMCS, EXIT_QUAL_NMI_WHILE_STI_BLOCKING). 139 * 140 * We take the easy way out and also include STI_BLOCKING as one of the 141 * gating items for vNMI injection. 142 */ 143static uint64_t nmi_blocking_bits = VMCS_INTERRUPTIBILITY_MOVSS_BLOCKING | 144 VMCS_INTERRUPTIBILITY_NMI_BLOCKING | 145 VMCS_INTERRUPTIBILITY_STI_BLOCKING; 146 147/* 148 * Optional capabilities 149 */ 150static int cap_halt_exit; 151static int cap_pause_exit; 152static int cap_unrestricted_guest; 153static int cap_monitor_trap; 154 155/* statistics */ 156static VMM_STAT_INTEL(VMEXIT_HLT_IGNORED, "number of times hlt was ignored"); 157 158#ifdef KTR 159static const char * 160exit_reason_to_str(int reason) 161{ 162 static char reasonbuf[32]; 163 164 switch (reason) { 165 case EXIT_REASON_EXCEPTION: 166 return "exception"; 167 case EXIT_REASON_EXT_INTR: 168 return "extint"; 169 case EXIT_REASON_TRIPLE_FAULT: 170 return "triplefault"; 171 case EXIT_REASON_INIT: 172 return "init"; 173 case EXIT_REASON_SIPI: 174 return "sipi"; 175 case EXIT_REASON_IO_SMI: 176 return "iosmi"; 177 case EXIT_REASON_SMI: 178 return "smi"; 179 case EXIT_REASON_INTR_WINDOW: 180 return "intrwindow"; 181 case EXIT_REASON_NMI_WINDOW: 182 return "nmiwindow"; 183 case EXIT_REASON_TASK_SWITCH: 184 return "taskswitch"; 185 case EXIT_REASON_CPUID: 186 return "cpuid"; 187 case EXIT_REASON_GETSEC: 188 return "getsec"; 189 case EXIT_REASON_HLT: 190 return "hlt"; 191 case EXIT_REASON_INVD: 192 return "invd"; 193 case EXIT_REASON_INVLPG: 194 return "invlpg"; 195 case EXIT_REASON_RDPMC: 196 return "rdpmc"; 197 case EXIT_REASON_RDTSC: 198 return "rdtsc"; 199 case EXIT_REASON_RSM: 200 return "rsm"; 201 case EXIT_REASON_VMCALL: 202 return "vmcall"; 203 case EXIT_REASON_VMCLEAR: 204 return "vmclear"; 205 case EXIT_REASON_VMLAUNCH: 206 return "vmlaunch"; 207 case EXIT_REASON_VMPTRLD: 208 return "vmptrld"; 209 case EXIT_REASON_VMPTRST: 210 return "vmptrst"; 211 case EXIT_REASON_VMREAD: 212 return "vmread"; 213 case EXIT_REASON_VMRESUME: 214 return "vmresume"; 215 case EXIT_REASON_VMWRITE: 216 return "vmwrite"; 217 case EXIT_REASON_VMXOFF: 218 return "vmxoff"; 219 case EXIT_REASON_VMXON: 220 return "vmxon"; 221 case EXIT_REASON_CR_ACCESS: 222 return "craccess"; 223 case EXIT_REASON_DR_ACCESS: 224 return "draccess"; 225 case EXIT_REASON_INOUT: 226 return "inout"; 227 case EXIT_REASON_RDMSR: 228 return "rdmsr"; 229 case EXIT_REASON_WRMSR: 230 return "wrmsr"; 231 case EXIT_REASON_INVAL_VMCS: 232 return "invalvmcs"; 233 case EXIT_REASON_INVAL_MSR: 234 return "invalmsr"; 235 case EXIT_REASON_MWAIT: 236 return "mwait"; 237 case EXIT_REASON_MTF: 238 return "mtf"; 239 case EXIT_REASON_MONITOR: 240 return "monitor"; 241 case EXIT_REASON_PAUSE: 242 return "pause"; 243 case EXIT_REASON_MCE: 244 return "mce"; 245 case EXIT_REASON_TPR: 246 return "tpr"; 247 case EXIT_REASON_APIC: 248 return "apic"; 249 case EXIT_REASON_GDTR_IDTR: 250 return "gdtridtr"; 251 case EXIT_REASON_LDTR_TR: 252 return "ldtrtr"; 253 case EXIT_REASON_EPT_FAULT: 254 return "eptfault"; 255 case EXIT_REASON_EPT_MISCONFIG: 256 return "eptmisconfig"; 257 case EXIT_REASON_INVEPT: 258 return "invept"; 259 case EXIT_REASON_RDTSCP: 260 return "rdtscp"; 261 case EXIT_REASON_VMX_PREEMPT: 262 return "vmxpreempt"; 263 case EXIT_REASON_INVVPID: 264 return "invvpid"; 265 case EXIT_REASON_WBINVD: 266 return "wbinvd"; 267 case EXIT_REASON_XSETBV: 268 return "xsetbv"; 269 default: 270 snprintf(reasonbuf, sizeof(reasonbuf), "%d", reason); 271 return (reasonbuf); 272 } 273} 274 275#ifdef SETJMP_TRACE 276static const char * 277vmx_setjmp_rc2str(int rc) 278{ 279 switch (rc) { 280 case VMX_RETURN_DIRECT: 281 return "direct"; 282 case VMX_RETURN_LONGJMP: 283 return "longjmp"; 284 case VMX_RETURN_VMRESUME: 285 return "vmresume"; 286 case VMX_RETURN_VMLAUNCH: 287 return "vmlaunch"; 288 case VMX_RETURN_AST: 289 return "ast"; 290 default: 291 return "unknown"; 292 } 293} 294 295#define SETJMP_TRACE(vmx, vcpu, vmxctx, regname) \ 296 VMM_CTR1((vmx)->vm, (vcpu), "setjmp trace " #regname " 0x%016lx", \ 297 (vmxctx)->regname) 298 299static void 300vmx_setjmp_trace(struct vmx *vmx, int vcpu, struct vmxctx *vmxctx, int rc) 301{ 302 uint64_t host_rip, host_rsp; 303 304 if (vmxctx != &vmx->ctx[vcpu]) 305 panic("vmx_setjmp_trace: invalid vmxctx %p; should be %p", 306 vmxctx, &vmx->ctx[vcpu]); 307 308 VMM_CTR1((vmx)->vm, (vcpu), "vmxctx = %p", vmxctx); 309 VMM_CTR2((vmx)->vm, (vcpu), "setjmp return code %s(%d)", 310 vmx_setjmp_rc2str(rc), rc); 311 312 host_rsp = host_rip = ~0; 313 vmread(VMCS_HOST_RIP, &host_rip); 314 vmread(VMCS_HOST_RSP, &host_rsp); 315 VMM_CTR2((vmx)->vm, (vcpu), "vmcs host_rip 0x%016lx, host_rsp 0x%016lx", 316 host_rip, host_rsp); 317 318 SETJMP_TRACE(vmx, vcpu, vmxctx, host_r15); 319 SETJMP_TRACE(vmx, vcpu, vmxctx, host_r14); 320 SETJMP_TRACE(vmx, vcpu, vmxctx, host_r13); 321 SETJMP_TRACE(vmx, vcpu, vmxctx, host_r12); 322 SETJMP_TRACE(vmx, vcpu, vmxctx, host_rbp); 323 SETJMP_TRACE(vmx, vcpu, vmxctx, host_rsp); 324 SETJMP_TRACE(vmx, vcpu, vmxctx, host_rbx); 325 SETJMP_TRACE(vmx, vcpu, vmxctx, host_rip); 326 327 SETJMP_TRACE(vmx, vcpu, vmxctx, guest_rdi); 328 SETJMP_TRACE(vmx, vcpu, vmxctx, guest_rsi); 329 SETJMP_TRACE(vmx, vcpu, vmxctx, guest_rdx); 330 SETJMP_TRACE(vmx, vcpu, vmxctx, guest_rcx); 331 SETJMP_TRACE(vmx, vcpu, vmxctx, guest_r8); 332 SETJMP_TRACE(vmx, vcpu, vmxctx, guest_r9); 333 SETJMP_TRACE(vmx, vcpu, vmxctx, guest_rax); 334 SETJMP_TRACE(vmx, vcpu, vmxctx, guest_rbx); 335 SETJMP_TRACE(vmx, vcpu, vmxctx, guest_rbp); 336 SETJMP_TRACE(vmx, vcpu, vmxctx, guest_r10); 337 SETJMP_TRACE(vmx, vcpu, vmxctx, guest_r11); 338 SETJMP_TRACE(vmx, vcpu, vmxctx, guest_r12); 339 SETJMP_TRACE(vmx, vcpu, vmxctx, guest_r13); 340 SETJMP_TRACE(vmx, vcpu, vmxctx, guest_r14); 341 SETJMP_TRACE(vmx, vcpu, vmxctx, guest_r15); 342 SETJMP_TRACE(vmx, vcpu, vmxctx, guest_cr2); 343} 344#endif 345#else 346static void __inline 347vmx_setjmp_trace(struct vmx *vmx, int vcpu, struct vmxctx *vmxctx, int rc) 348{ 349 return; 350} 351#endif /* KTR */ 352 353u_long 354vmx_fix_cr0(u_long cr0) 355{ 356 357 return ((cr0 | cr0_ones_mask) & ~cr0_zeros_mask); 358} 359 360u_long 361vmx_fix_cr4(u_long cr4) 362{ 363 364 return ((cr4 | cr4_ones_mask) & ~cr4_zeros_mask); 365} 366 367static void 368msr_save_area_init(struct msr_entry *g_area, int *g_count) 369{ 370 int cnt; 371 372 static struct msr_entry guest_msrs[] = { 373 { MSR_KGSBASE, 0, 0 }, 374 }; 375 376 cnt = sizeof(guest_msrs) / sizeof(guest_msrs[0]); 377 if (cnt > GUEST_MSR_MAX_ENTRIES) 378 panic("guest msr save area overrun"); 379 bcopy(guest_msrs, g_area, sizeof(guest_msrs)); 380 *g_count = cnt; 381} 382 383static void 384vmx_disable(void *arg __unused) 385{ 386 struct invvpid_desc invvpid_desc = { 0 }; 387 struct invept_desc invept_desc = { 0 }; 388 389 if (vmxon_enabled[curcpu]) { 390 /* 391 * See sections 25.3.3.3 and 25.3.3.4 in Intel Vol 3b. 392 * 393 * VMXON or VMXOFF are not required to invalidate any TLB 394 * caching structures. This prevents potential retention of 395 * cached information in the TLB between distinct VMX episodes. 396 */ 397 invvpid(INVVPID_TYPE_ALL_CONTEXTS, invvpid_desc); 398 invept(INVEPT_TYPE_ALL_CONTEXTS, invept_desc); 399 vmxoff(); 400 } 401 load_cr4(rcr4() & ~CR4_VMXE); 402} 403 404static int 405vmx_cleanup(void) 406{ 407 408 smp_rendezvous(NULL, vmx_disable, NULL, NULL); 409 410 return (0); 411} 412 413static void 414vmx_enable(void *arg __unused) 415{ 416 int error; 417 418 load_cr4(rcr4() | CR4_VMXE); 419 420 *(uint32_t *)vmxon_region[curcpu] = vmx_revision(); 421 error = vmxon(vmxon_region[curcpu]); 422 if (error == 0) 423 vmxon_enabled[curcpu] = 1; 424} 425 426static int 427vmx_init(void) 428{ 429 int error; 430 uint64_t fixed0, fixed1, feature_control; 431 uint32_t tmp; 432 433 /* CPUID.1:ECX[bit 5] must be 1 for processor to support VMX */ 434 if (!(cpu_feature2 & CPUID2_VMX)) { 435 printf("vmx_init: processor does not support VMX operation\n"); 436 return (ENXIO); 437 } 438 439 /* 440 * Verify that MSR_IA32_FEATURE_CONTROL lock and VMXON enable bits 441 * are set (bits 0 and 2 respectively). 442 */ 443 feature_control = rdmsr(MSR_IA32_FEATURE_CONTROL); 444 if ((feature_control & IA32_FEATURE_CONTROL_LOCK) == 0 || 445 (feature_control & IA32_FEATURE_CONTROL_VMX_EN) == 0) { 446 printf("vmx_init: VMX operation disabled by BIOS\n"); 447 return (ENXIO); 448 } 449 450 /* Check support for primary processor-based VM-execution controls */ 451 error = vmx_set_ctlreg(MSR_VMX_PROCBASED_CTLS, 452 MSR_VMX_TRUE_PROCBASED_CTLS, 453 PROCBASED_CTLS_ONE_SETTING, 454 PROCBASED_CTLS_ZERO_SETTING, &procbased_ctls); 455 if (error) { 456 printf("vmx_init: processor does not support desired primary " 457 "processor-based controls\n"); 458 return (error); 459 } 460 461 /* Clear the processor-based ctl bits that are set on demand */ 462 procbased_ctls &= ~PROCBASED_CTLS_WINDOW_SETTING; 463 464 /* Check support for secondary processor-based VM-execution controls */ 465 error = vmx_set_ctlreg(MSR_VMX_PROCBASED_CTLS2, 466 MSR_VMX_PROCBASED_CTLS2, 467 PROCBASED_CTLS2_ONE_SETTING, 468 PROCBASED_CTLS2_ZERO_SETTING, &procbased_ctls2); 469 if (error) { 470 printf("vmx_init: processor does not support desired secondary " 471 "processor-based controls\n"); 472 return (error); 473 } 474 475 /* Check support for VPID */ 476 error = vmx_set_ctlreg(MSR_VMX_PROCBASED_CTLS2, MSR_VMX_PROCBASED_CTLS2, 477 PROCBASED2_ENABLE_VPID, 0, &tmp); 478 if (error == 0) 479 procbased_ctls2 |= PROCBASED2_ENABLE_VPID; 480 481 /* Check support for pin-based VM-execution controls */ 482 error = vmx_set_ctlreg(MSR_VMX_PINBASED_CTLS, 483 MSR_VMX_TRUE_PINBASED_CTLS, 484 PINBASED_CTLS_ONE_SETTING, 485 PINBASED_CTLS_ZERO_SETTING, &pinbased_ctls); 486 if (error) { 487 printf("vmx_init: processor does not support desired " 488 "pin-based controls\n"); 489 return (error); 490 } 491 492 /* Check support for VM-exit controls */ 493 error = vmx_set_ctlreg(MSR_VMX_EXIT_CTLS, MSR_VMX_TRUE_EXIT_CTLS, 494 VM_EXIT_CTLS_ONE_SETTING, 495 VM_EXIT_CTLS_ZERO_SETTING, 496 &exit_ctls); 497 if (error) { 498 /* Try again without the PAT MSR bits */ 499 error = vmx_set_ctlreg(MSR_VMX_EXIT_CTLS, 500 MSR_VMX_TRUE_EXIT_CTLS, 501 VM_EXIT_CTLS_ONE_SETTING_NO_PAT, 502 VM_EXIT_CTLS_ZERO_SETTING, 503 &exit_ctls); 504 if (error) { 505 printf("vmx_init: processor does not support desired " 506 "exit controls\n"); 507 return (error); 508 } else { 509 if (bootverbose) 510 printf("vmm: PAT MSR access not supported\n"); 511 guest_msr_valid(MSR_PAT); 512 vmx_no_patmsr = 1; 513 } 514 } 515 516 /* Check support for VM-entry controls */ 517 if (!vmx_no_patmsr) { 518 error = vmx_set_ctlreg(MSR_VMX_ENTRY_CTLS, 519 MSR_VMX_TRUE_ENTRY_CTLS, 520 VM_ENTRY_CTLS_ONE_SETTING, 521 VM_ENTRY_CTLS_ZERO_SETTING, 522 &entry_ctls); 523 } else { 524 error = vmx_set_ctlreg(MSR_VMX_ENTRY_CTLS, 525 MSR_VMX_TRUE_ENTRY_CTLS, 526 VM_ENTRY_CTLS_ONE_SETTING_NO_PAT, 527 VM_ENTRY_CTLS_ZERO_SETTING, 528 &entry_ctls); 529 } 530 531 if (error) { 532 printf("vmx_init: processor does not support desired " 533 "entry controls\n"); 534 return (error); 535 } 536 537 /* 538 * Check support for optional features by testing them 539 * as individual bits 540 */ 541 cap_halt_exit = (vmx_set_ctlreg(MSR_VMX_PROCBASED_CTLS, 542 MSR_VMX_TRUE_PROCBASED_CTLS, 543 PROCBASED_HLT_EXITING, 0, 544 &tmp) == 0); 545 546 cap_monitor_trap = (vmx_set_ctlreg(MSR_VMX_PROCBASED_CTLS, 547 MSR_VMX_PROCBASED_CTLS, 548 PROCBASED_MTF, 0, 549 &tmp) == 0); 550 551 cap_pause_exit = (vmx_set_ctlreg(MSR_VMX_PROCBASED_CTLS, 552 MSR_VMX_TRUE_PROCBASED_CTLS, 553 PROCBASED_PAUSE_EXITING, 0, 554 &tmp) == 0); 555 556 cap_unrestricted_guest = (vmx_set_ctlreg(MSR_VMX_PROCBASED_CTLS2, 557 MSR_VMX_PROCBASED_CTLS2, 558 PROCBASED2_UNRESTRICTED_GUEST, 0, 559 &tmp) == 0); 560 561 /* Initialize EPT */ 562 error = ept_init(); 563 if (error) { 564 printf("vmx_init: ept initialization failed (%d)\n", error); 565 return (error); 566 } 567 568 /* 569 * Stash the cr0 and cr4 bits that must be fixed to 0 or 1 570 */ 571 fixed0 = rdmsr(MSR_VMX_CR0_FIXED0); 572 fixed1 = rdmsr(MSR_VMX_CR0_FIXED1); 573 cr0_ones_mask = fixed0 & fixed1; 574 cr0_zeros_mask = ~fixed0 & ~fixed1; 575 576 /* 577 * CR0_PE and CR0_PG can be set to zero in VMX non-root operation 578 * if unrestricted guest execution is allowed. 579 */ 580 if (cap_unrestricted_guest) 581 cr0_ones_mask &= ~(CR0_PG | CR0_PE); 582 583 /* 584 * Do not allow the guest to set CR0_NW or CR0_CD. 585 */ 586 cr0_zeros_mask |= (CR0_NW | CR0_CD); 587 588 fixed0 = rdmsr(MSR_VMX_CR4_FIXED0); 589 fixed1 = rdmsr(MSR_VMX_CR4_FIXED1); 590 cr4_ones_mask = fixed0 & fixed1; 591 cr4_zeros_mask = ~fixed0 & ~fixed1; 592 593 /* enable VMX operation */ 594 smp_rendezvous(NULL, vmx_enable, NULL, NULL); 595 596 return (0); 597} 598 599/* 600 * If this processor does not support VPIDs then simply return 0. 601 * 602 * Otherwise generate the next value of VPID to use. Any value is alright 603 * as long as it is non-zero. 604 * 605 * We always execute in VMX non-root context with EPT enabled. Thus all 606 * combined mappings are tagged with the (EP4TA, VPID, PCID) tuple. This 607 * in turn means that multiple VMs can share the same VPID as long as 608 * they have distinct EPT page tables. 609 * 610 * XXX 611 * We should optimize this so that it returns VPIDs that are not in 612 * use. Then we will not unnecessarily invalidate mappings in 613 * vmx_set_pcpu_defaults() just because two or more vcpus happen to 614 * use the same 'vpid'. 615 */ 616static uint16_t 617vmx_vpid(void) 618{ 619 uint16_t vpid = 0; 620 621 if ((procbased_ctls2 & PROCBASED2_ENABLE_VPID) != 0) { 622 do { 623 vpid = atomic_fetchadd_int(&nextvpid, 1); 624 } while (vpid == 0); 625 } 626 627 return (vpid); 628} 629 630static int 631vmx_setup_cr_shadow(int which, struct vmcs *vmcs) 632{ 633 int error, mask_ident, shadow_ident; 634 uint64_t mask_value, shadow_value; 635 636 if (which != 0 && which != 4) 637 panic("vmx_setup_cr_shadow: unknown cr%d", which); 638 639 if (which == 0) { 640 mask_ident = VMCS_CR0_MASK; 641 mask_value = cr0_ones_mask | cr0_zeros_mask; 642 shadow_ident = VMCS_CR0_SHADOW; 643 shadow_value = cr0_ones_mask; 644 } else { 645 mask_ident = VMCS_CR4_MASK; 646 mask_value = cr4_ones_mask | cr4_zeros_mask; 647 shadow_ident = VMCS_CR4_SHADOW; 648 shadow_value = cr4_ones_mask; 649 } 650 651 error = vmcs_setreg(vmcs, VMCS_IDENT(mask_ident), mask_value); 652 if (error) 653 return (error); 654 655 error = vmcs_setreg(vmcs, VMCS_IDENT(shadow_ident), shadow_value); 656 if (error) 657 return (error); 658 659 return (0); 660} 661#define vmx_setup_cr0_shadow(vmcs) vmx_setup_cr_shadow(0, (vmcs)) 662#define vmx_setup_cr4_shadow(vmcs) vmx_setup_cr_shadow(4, (vmcs)) 663 664static void * 665vmx_vminit(struct vm *vm) 666{ 667 uint16_t vpid; 668 int i, error, guest_msr_count; 669 struct vmx *vmx; 670 671 vmx = malloc(sizeof(struct vmx), M_VMX, M_WAITOK | M_ZERO); 672 if ((uintptr_t)vmx & PAGE_MASK) { 673 panic("malloc of struct vmx not aligned on %d byte boundary", 674 PAGE_SIZE); 675 } 676 vmx->vm = vm; 677 678 /* 679 * Clean up EPTP-tagged guest physical and combined mappings 680 * 681 * VMX transitions are not required to invalidate any guest physical 682 * mappings. So, it may be possible for stale guest physical mappings 683 * to be present in the processor TLBs. 684 * 685 * Combined mappings for this EP4TA are also invalidated for all VPIDs. 686 */ 687 ept_invalidate_mappings(vtophys(vmx->pml4ept)); 688 689 msr_bitmap_initialize(vmx->msr_bitmap); 690 691 /* 692 * It is safe to allow direct access to MSR_GSBASE and MSR_FSBASE. 693 * The guest FSBASE and GSBASE are saved and restored during 694 * vm-exit and vm-entry respectively. The host FSBASE and GSBASE are 695 * always restored from the vmcs host state area on vm-exit. 696 * 697 * The SYSENTER_CS/ESP/EIP MSRs are identical to FS/GSBASE in 698 * how they are saved/restored so can be directly accessed by the 699 * guest. 700 * 701 * Guest KGSBASE is saved and restored in the guest MSR save area. 702 * Host KGSBASE is restored before returning to userland from the pcb. 703 * There will be a window of time when we are executing in the host 704 * kernel context with a value of KGSBASE from the guest. This is ok 705 * because the value of KGSBASE is inconsequential in kernel context. 706 * 707 * MSR_EFER is saved and restored in the guest VMCS area on a 708 * VM exit and entry respectively. It is also restored from the 709 * host VMCS area on a VM exit. 710 */ 711 if (guest_msr_rw(vmx, MSR_GSBASE) || 712 guest_msr_rw(vmx, MSR_FSBASE) || 713 guest_msr_rw(vmx, MSR_SYSENTER_CS_MSR) || 714 guest_msr_rw(vmx, MSR_SYSENTER_ESP_MSR) || 715 guest_msr_rw(vmx, MSR_SYSENTER_EIP_MSR) || 716 guest_msr_rw(vmx, MSR_KGSBASE) || 717 guest_msr_rw(vmx, MSR_EFER)) 718 panic("vmx_vminit: error setting guest msr access"); 719 720 /* 721 * MSR_PAT is saved and restored in the guest VMCS are on a VM exit 722 * and entry respectively. It is also restored from the host VMCS 723 * area on a VM exit. However, if running on a system with no 724 * MSR_PAT save/restore support, leave access disabled so accesses 725 * will be trapped. 726 */ 727 if (!vmx_no_patmsr && guest_msr_rw(vmx, MSR_PAT)) 728 panic("vmx_vminit: error setting guest pat msr access"); 729 730 for (i = 0; i < VM_MAXCPU; i++) { 731 vmx->vmcs[i].identifier = vmx_revision(); 732 error = vmclear(&vmx->vmcs[i]); 733 if (error != 0) { 734 panic("vmx_vminit: vmclear error %d on vcpu %d\n", 735 error, i); 736 } 737 738 vpid = vmx_vpid(); 739 740 error = vmcs_set_defaults(&vmx->vmcs[i], 741 (u_long)vmx_longjmp, 742 (u_long)&vmx->ctx[i], 743 vtophys(vmx->pml4ept), 744 pinbased_ctls, 745 procbased_ctls, 746 procbased_ctls2, 747 exit_ctls, entry_ctls, 748 vtophys(vmx->msr_bitmap), 749 vpid); 750 751 if (error != 0) 752 panic("vmx_vminit: vmcs_set_defaults error %d", error); 753 754 vmx->cap[i].set = 0; 755 vmx->cap[i].proc_ctls = procbased_ctls; 756 757 vmx->state[i].lastcpu = -1; 758 vmx->state[i].vpid = vpid; 759 760 msr_save_area_init(vmx->guest_msrs[i], &guest_msr_count); 761 762 error = vmcs_set_msr_save(&vmx->vmcs[i], 763 vtophys(vmx->guest_msrs[i]), 764 guest_msr_count); 765 if (error != 0) 766 panic("vmcs_set_msr_save error %d", error); 767 768 error = vmx_setup_cr0_shadow(&vmx->vmcs[i]); 769 if (error != 0) 770 panic("vmx_setup_cr0_shadow %d", error); 771 772 error = vmx_setup_cr4_shadow(&vmx->vmcs[i]); 773 if (error != 0) 774 panic("vmx_setup_cr4_shadow %d", error); 775 } 776 777 return (vmx); 778} 779 780static int 781vmx_handle_cpuid(struct vm *vm, int vcpu, struct vmxctx *vmxctx) 782{ 783 int handled, func; 784 785 func = vmxctx->guest_rax; 786 787 handled = x86_emulate_cpuid(vm, vcpu, 788 (uint32_t*)(&vmxctx->guest_rax), 789 (uint32_t*)(&vmxctx->guest_rbx), 790 (uint32_t*)(&vmxctx->guest_rcx), 791 (uint32_t*)(&vmxctx->guest_rdx)); 792 return (handled); 793} 794 795static __inline void 796vmx_run_trace(struct vmx *vmx, int vcpu) 797{ 798#ifdef KTR 799 VMM_CTR1(vmx->vm, vcpu, "Resume execution at 0x%0lx", vmcs_guest_rip()); 800#endif 801} 802 803static __inline void 804vmx_exit_trace(struct vmx *vmx, int vcpu, uint64_t rip, uint32_t exit_reason, 805 int handled) 806{ 807#ifdef KTR 808 VMM_CTR3(vmx->vm, vcpu, "%s %s vmexit at 0x%0lx", 809 handled ? "handled" : "unhandled", 810 exit_reason_to_str(exit_reason), rip); 811#endif 812} 813 814static __inline void 815vmx_astpending_trace(struct vmx *vmx, int vcpu, uint64_t rip) 816{ 817#ifdef KTR 818 VMM_CTR1(vmx->vm, vcpu, "astpending vmexit at 0x%0lx", rip); 819#endif 820} 821 822static int 823vmx_set_pcpu_defaults(struct vmx *vmx, int vcpu) 824{ 825 int error, lastcpu; 826 struct vmxstate *vmxstate; 827 struct invvpid_desc invvpid_desc = { 0 }; 828 829 vmxstate = &vmx->state[vcpu]; 830 lastcpu = vmxstate->lastcpu; 831 vmxstate->lastcpu = curcpu; 832 833 if (lastcpu == curcpu) { 834 error = 0; 835 goto done; 836 } 837 838 vmm_stat_incr(vmx->vm, vcpu, VCPU_MIGRATIONS, 1); 839 840 error = vmwrite(VMCS_HOST_TR_BASE, vmm_get_host_trbase()); 841 if (error != 0) 842 goto done; 843 844 error = vmwrite(VMCS_HOST_GDTR_BASE, vmm_get_host_gdtrbase()); 845 if (error != 0) 846 goto done; 847 848 error = vmwrite(VMCS_HOST_GS_BASE, vmm_get_host_gsbase()); 849 if (error != 0) 850 goto done; 851 852 /* 853 * If we are using VPIDs then invalidate all mappings tagged with 'vpid' 854 * 855 * We do this because this vcpu was executing on a different host 856 * cpu when it last ran. We do not track whether it invalidated 857 * mappings associated with its 'vpid' during that run. So we must 858 * assume that the mappings associated with 'vpid' on 'curcpu' are 859 * stale and invalidate them. 860 * 861 * Note that we incur this penalty only when the scheduler chooses to 862 * move the thread associated with this vcpu between host cpus. 863 * 864 * Note also that this will invalidate mappings tagged with 'vpid' 865 * for "all" EP4TAs. 866 */ 867 if (vmxstate->vpid != 0) { 868 invvpid_desc.vpid = vmxstate->vpid; 869 invvpid(INVVPID_TYPE_SINGLE_CONTEXT, invvpid_desc); 870 } 871done: 872 return (error); 873} 874 875static void 876vm_exit_update_rip(struct vm_exit *vmexit) 877{ 878 int error; 879 880 error = vmwrite(VMCS_GUEST_RIP, vmexit->rip + vmexit->inst_length); 881 if (error) 882 panic("vmx_run: error %d writing to VMCS_GUEST_RIP", error); 883} 884 885/* 886 * We depend on 'procbased_ctls' to have the Interrupt Window Exiting bit set. 887 */ 888CTASSERT((PROCBASED_CTLS_ONE_SETTING & PROCBASED_INT_WINDOW_EXITING) != 0); 889 890static void __inline 891vmx_set_int_window_exiting(struct vmx *vmx, int vcpu) 892{ 893 int error; 894 895 vmx->cap[vcpu].proc_ctls |= PROCBASED_INT_WINDOW_EXITING; 896 897 error = vmwrite(VMCS_PRI_PROC_BASED_CTLS, vmx->cap[vcpu].proc_ctls); 898 if (error) 899 panic("vmx_set_int_window_exiting: vmwrite error %d", error); 900} 901 902static void __inline 903vmx_clear_int_window_exiting(struct vmx *vmx, int vcpu) 904{ 905 int error; 906 907 vmx->cap[vcpu].proc_ctls &= ~PROCBASED_INT_WINDOW_EXITING; 908 909 error = vmwrite(VMCS_PRI_PROC_BASED_CTLS, vmx->cap[vcpu].proc_ctls); 910 if (error) 911 panic("vmx_clear_int_window_exiting: vmwrite error %d", error); 912} 913 914static void __inline 915vmx_set_nmi_window_exiting(struct vmx *vmx, int vcpu) 916{ 917 int error; 918 919 vmx->cap[vcpu].proc_ctls |= PROCBASED_NMI_WINDOW_EXITING; 920 921 error = vmwrite(VMCS_PRI_PROC_BASED_CTLS, vmx->cap[vcpu].proc_ctls); 922 if (error) 923 panic("vmx_set_nmi_window_exiting: vmwrite error %d", error); 924} 925 926static void __inline 927vmx_clear_nmi_window_exiting(struct vmx *vmx, int vcpu) 928{ 929 int error; 930 931 vmx->cap[vcpu].proc_ctls &= ~PROCBASED_NMI_WINDOW_EXITING; 932 933 error = vmwrite(VMCS_PRI_PROC_BASED_CTLS, vmx->cap[vcpu].proc_ctls); 934 if (error) 935 panic("vmx_clear_nmi_window_exiting: vmwrite error %d", error); 936} 937 938static int 939vmx_inject_nmi(struct vmx *vmx, int vcpu) 940{ 941 int error; 942 uint64_t info, interruptibility; 943 944 /* Bail out if no NMI requested */ 945 if (!vm_nmi_pending(vmx->vm, vcpu)) 946 return (0); 947 948 error = vmread(VMCS_GUEST_INTERRUPTIBILITY, &interruptibility); 949 if (error) { 950 panic("vmx_inject_nmi: vmread(interruptibility) %d", 951 error); 952 } 953 if (interruptibility & nmi_blocking_bits) 954 goto nmiblocked; 955 956 /* 957 * Inject the virtual NMI. The vector must be the NMI IDT entry 958 * or the VMCS entry check will fail. 959 */ 960 info = VMCS_INTERRUPTION_INFO_NMI | VMCS_INTERRUPTION_INFO_VALID; 961 info |= IDT_NMI; 962 963 error = vmwrite(VMCS_ENTRY_INTR_INFO, info); 964 if (error) 965 panic("vmx_inject_nmi: vmwrite(intrinfo) %d", error); 966 967 VMM_CTR0(vmx->vm, vcpu, "Injecting vNMI"); 968 969 /* Clear the request */ 970 vm_nmi_clear(vmx->vm, vcpu); 971 return (1); 972 973nmiblocked: 974 /* 975 * Set the NMI Window Exiting execution control so we can inject 976 * the virtual NMI as soon as blocking condition goes away. 977 */ 978 vmx_set_nmi_window_exiting(vmx, vcpu); 979 980 VMM_CTR0(vmx->vm, vcpu, "Enabling NMI window exiting"); 981 return (1); 982} 983 984static void 985vmx_inject_interrupts(struct vmx *vmx, int vcpu) 986{ 987 int error, vector; 988 uint64_t info, rflags, interruptibility; 989 990 const int HWINTR_BLOCKED = VMCS_INTERRUPTIBILITY_STI_BLOCKING | 991 VMCS_INTERRUPTIBILITY_MOVSS_BLOCKING; 992 993 /* 994 * If there is already an interrupt pending then just return. 995 * 996 * This could happen if an interrupt was injected on a prior 997 * VM entry but the actual entry into guest mode was aborted 998 * because of a pending AST. 999 */ 1000 error = vmread(VMCS_ENTRY_INTR_INFO, &info); 1001 if (error) 1002 panic("vmx_inject_interrupts: vmread(intrinfo) %d", error); 1003 if (info & VMCS_INTERRUPTION_INFO_VALID) 1004 return; 1005 1006 /* 1007 * NMI injection has priority so deal with those first 1008 */ 1009 if (vmx_inject_nmi(vmx, vcpu)) 1010 return; 1011 1012 /* Ask the local apic for a vector to inject */ 1013 vector = lapic_pending_intr(vmx->vm, vcpu); 1014 if (vector < 0) 1015 return; 1016 1017 if (vector < 32 || vector > 255) 1018 panic("vmx_inject_interrupts: invalid vector %d\n", vector); 1019 1020 /* Check RFLAGS.IF and the interruptibility state of the guest */ 1021 error = vmread(VMCS_GUEST_RFLAGS, &rflags); 1022 if (error) 1023 panic("vmx_inject_interrupts: vmread(rflags) %d", error); 1024 1025 if ((rflags & PSL_I) == 0) 1026 goto cantinject; 1027 1028 error = vmread(VMCS_GUEST_INTERRUPTIBILITY, &interruptibility); 1029 if (error) { 1030 panic("vmx_inject_interrupts: vmread(interruptibility) %d", 1031 error); 1032 } 1033 if (interruptibility & HWINTR_BLOCKED) 1034 goto cantinject; 1035 1036 /* Inject the interrupt */ 1037 info = VMCS_INTERRUPTION_INFO_HW_INTR | VMCS_INTERRUPTION_INFO_VALID; 1038 info |= vector; 1039 error = vmwrite(VMCS_ENTRY_INTR_INFO, info); 1040 if (error) 1041 panic("vmx_inject_interrupts: vmwrite(intrinfo) %d", error); 1042 1043 /* Update the Local APIC ISR */ 1044 lapic_intr_accepted(vmx->vm, vcpu, vector); 1045 1046 VMM_CTR1(vmx->vm, vcpu, "Injecting hwintr at vector %d", vector); 1047 1048 return; 1049 1050cantinject: 1051 /* 1052 * Set the Interrupt Window Exiting execution control so we can inject 1053 * the interrupt as soon as blocking condition goes away. 1054 */ 1055 vmx_set_int_window_exiting(vmx, vcpu); 1056 1057 VMM_CTR0(vmx->vm, vcpu, "Enabling interrupt window exiting"); 1058} 1059 1060static int 1061vmx_emulate_cr_access(struct vmx *vmx, int vcpu, uint64_t exitqual) 1062{ 1063 int error, cr, vmcs_guest_cr; 1064 uint64_t regval, ones_mask, zeros_mask; 1065 const struct vmxctx *vmxctx; 1066 1067 /* We only handle mov to %cr0 or %cr4 at this time */ 1068 if ((exitqual & 0xf0) != 0x00) 1069 return (UNHANDLED); 1070 1071 cr = exitqual & 0xf; 1072 if (cr != 0 && cr != 4) 1073 return (UNHANDLED); 1074 1075 vmxctx = &vmx->ctx[vcpu]; 1076 1077 /* 1078 * We must use vmwrite() directly here because vmcs_setreg() will 1079 * call vmclear(vmcs) as a side-effect which we certainly don't want. 1080 */ 1081 switch ((exitqual >> 8) & 0xf) { 1082 case 0: 1083 regval = vmxctx->guest_rax; 1084 break; 1085 case 1: 1086 regval = vmxctx->guest_rcx; 1087 break; 1088 case 2: 1089 regval = vmxctx->guest_rdx; 1090 break; 1091 case 3: 1092 regval = vmxctx->guest_rbx; 1093 break; 1094 case 4: 1095 error = vmread(VMCS_GUEST_RSP, ®val); 1096 if (error) { 1097 panic("vmx_emulate_cr_access: " 1098 "error %d reading guest rsp", error); 1099 } 1100 break; 1101 case 5: 1102 regval = vmxctx->guest_rbp; 1103 break; 1104 case 6: 1105 regval = vmxctx->guest_rsi; 1106 break; 1107 case 7: 1108 regval = vmxctx->guest_rdi; 1109 break; 1110 case 8: 1111 regval = vmxctx->guest_r8; 1112 break; 1113 case 9: 1114 regval = vmxctx->guest_r9; 1115 break; 1116 case 10: 1117 regval = vmxctx->guest_r10; 1118 break; 1119 case 11: 1120 regval = vmxctx->guest_r11; 1121 break; 1122 case 12: 1123 regval = vmxctx->guest_r12; 1124 break; 1125 case 13: 1126 regval = vmxctx->guest_r13; 1127 break; 1128 case 14: 1129 regval = vmxctx->guest_r14; 1130 break; 1131 case 15: 1132 regval = vmxctx->guest_r15; 1133 break; 1134 } 1135 1136 if (cr == 0) { 1137 ones_mask = cr0_ones_mask; 1138 zeros_mask = cr0_zeros_mask; 1139 vmcs_guest_cr = VMCS_GUEST_CR0; 1140 } else { 1141 ones_mask = cr4_ones_mask; 1142 zeros_mask = cr4_zeros_mask; 1143 vmcs_guest_cr = VMCS_GUEST_CR4; 1144 } 1145 regval |= ones_mask; 1146 regval &= ~zeros_mask; 1147 error = vmwrite(vmcs_guest_cr, regval); 1148 if (error) { 1149 panic("vmx_emulate_cr_access: error %d writing cr%d", 1150 error, cr); 1151 } 1152 1153 return (HANDLED); 1154} 1155 1156static int 1157vmx_ept_fault(struct vm *vm, int cpu, 1158 uint64_t gla, uint64_t gpa, uint64_t rip, int inst_length, 1159 uint64_t cr3, uint64_t ept_qual, struct vie *vie) 1160{ 1161 int read, write, error; 1162 1163 /* EPT violation on an instruction fetch doesn't make sense here */ 1164 if (ept_qual & EPT_VIOLATION_INST_FETCH) 1165 return (UNHANDLED); 1166 1167 /* EPT violation must be a read fault or a write fault */ 1168 read = ept_qual & EPT_VIOLATION_DATA_READ ? 1 : 0; 1169 write = ept_qual & EPT_VIOLATION_DATA_WRITE ? 1 : 0; 1170 if ((read | write) == 0) 1171 return (UNHANDLED); 1172 1173 /* 1174 * The EPT violation must have been caused by accessing a 1175 * guest-physical address that is a translation of a guest-linear 1176 * address. 1177 */ 1178 if ((ept_qual & EPT_VIOLATION_GLA_VALID) == 0 || 1179 (ept_qual & EPT_VIOLATION_XLAT_VALID) == 0) { 1180 return (UNHANDLED); 1181 } 1182 1183 /* Fetch, decode and emulate the faulting instruction */ 1184 if (vmm_fetch_instruction(vm, cpu, rip, inst_length, cr3, vie) != 0) 1185 return (UNHANDLED); 1186 1187 if (vmm_decode_instruction(vm, cpu, gla, vie) != 0) 1188 return (UNHANDLED); 1189 1190 /* 1191 * Check if this is a local apic access 1192 */ 1193 if (gpa < DEFAULT_APIC_BASE || gpa >= DEFAULT_APIC_BASE + PAGE_SIZE) 1194 return (UNHANDLED); 1195 1196 error = vmm_emulate_instruction(vm, cpu, gpa, vie, 1197 lapic_mmio_read, lapic_mmio_write, 0); 1198 1199 return (error ? UNHANDLED : HANDLED); 1200} 1201 1202static int 1203vmx_exit_process(struct vmx *vmx, int vcpu, struct vm_exit *vmexit) 1204{ 1205 int error, handled; 1206 struct vmcs *vmcs; 1207 struct vmxctx *vmxctx; 1208 uint32_t eax, ecx, edx; 1209 uint64_t qual, gla, gpa, cr3, intr_info; 1210 1211 handled = 0; 1212 vmcs = &vmx->vmcs[vcpu]; 1213 vmxctx = &vmx->ctx[vcpu]; 1214 qual = vmexit->u.vmx.exit_qualification; 1215 vmexit->exitcode = VM_EXITCODE_BOGUS; 1216 1217 vmm_stat_incr(vmx->vm, vcpu, VMEXIT_COUNT, 1); 1218 1219 switch (vmexit->u.vmx.exit_reason) { 1220 case EXIT_REASON_CR_ACCESS: 1221 vmm_stat_incr(vmx->vm, vcpu, VMEXIT_CR_ACCESS, 1); 1222 handled = vmx_emulate_cr_access(vmx, vcpu, qual); 1223 break; 1224 case EXIT_REASON_RDMSR: 1225 vmm_stat_incr(vmx->vm, vcpu, VMEXIT_RDMSR, 1); 1226 ecx = vmxctx->guest_rcx; 1227 error = emulate_rdmsr(vmx->vm, vcpu, ecx); 1228 if (error) { 1229 vmexit->exitcode = VM_EXITCODE_RDMSR; 1230 vmexit->u.msr.code = ecx; 1231 } else 1232 handled = 1; 1233 break; 1234 case EXIT_REASON_WRMSR: 1235 vmm_stat_incr(vmx->vm, vcpu, VMEXIT_WRMSR, 1); 1236 eax = vmxctx->guest_rax; 1237 ecx = vmxctx->guest_rcx; 1238 edx = vmxctx->guest_rdx; 1239 error = emulate_wrmsr(vmx->vm, vcpu, ecx, 1240 (uint64_t)edx << 32 | eax); 1241 if (error) { 1242 vmexit->exitcode = VM_EXITCODE_WRMSR; 1243 vmexit->u.msr.code = ecx; 1244 vmexit->u.msr.wval = (uint64_t)edx << 32 | eax; 1245 } else 1246 handled = 1; 1247 break; 1248 case EXIT_REASON_HLT: 1249 vmm_stat_incr(vmx->vm, vcpu, VMEXIT_HLT, 1); 1250 /* 1251 * If there is an event waiting to be injected then there is 1252 * no need to 'hlt'. 1253 */ 1254 error = vmread(VMCS_ENTRY_INTR_INFO, &intr_info); 1255 if (error) 1256 panic("vmx_exit_process: vmread(intrinfo) %d", error); 1257 1258 if (intr_info & VMCS_INTERRUPTION_INFO_VALID) { 1259 handled = 1; 1260 vmm_stat_incr(vmx->vm, vcpu, VMEXIT_HLT_IGNORED, 1); 1261 } else 1262 vmexit->exitcode = VM_EXITCODE_HLT; 1263 break; 1264 case EXIT_REASON_MTF: 1265 vmm_stat_incr(vmx->vm, vcpu, VMEXIT_MTRAP, 1); 1266 vmexit->exitcode = VM_EXITCODE_MTRAP; 1267 break; 1268 case EXIT_REASON_PAUSE: 1269 vmm_stat_incr(vmx->vm, vcpu, VMEXIT_PAUSE, 1); 1270 vmexit->exitcode = VM_EXITCODE_PAUSE; 1271 break; 1272 case EXIT_REASON_INTR_WINDOW: 1273 vmm_stat_incr(vmx->vm, vcpu, VMEXIT_INTR_WINDOW, 1); 1274 vmx_clear_int_window_exiting(vmx, vcpu); 1275 VMM_CTR0(vmx->vm, vcpu, "Disabling interrupt window exiting"); 1276 return (1); 1277 case EXIT_REASON_EXT_INTR: 1278 /* 1279 * External interrupts serve only to cause VM exits and allow 1280 * the host interrupt handler to run. 1281 * 1282 * If this external interrupt triggers a virtual interrupt 1283 * to a VM, then that state will be recorded by the 1284 * host interrupt handler in the VM's softc. We will inject 1285 * this virtual interrupt during the subsequent VM enter. 1286 */ 1287 1288 /* 1289 * This is special. We want to treat this as an 'handled' 1290 * VM-exit but not increment the instruction pointer. 1291 */ 1292 vmm_stat_incr(vmx->vm, vcpu, VMEXIT_EXTINT, 1); 1293 return (1); 1294 case EXIT_REASON_NMI_WINDOW: 1295 /* Exit to allow the pending virtual NMI to be injected */ 1296 vmm_stat_incr(vmx->vm, vcpu, VMEXIT_NMI_WINDOW, 1); 1297 vmx_clear_nmi_window_exiting(vmx, vcpu); 1298 VMM_CTR0(vmx->vm, vcpu, "Disabling NMI window exiting"); 1299 return (1); 1300 case EXIT_REASON_INOUT: 1301 vmm_stat_incr(vmx->vm, vcpu, VMEXIT_INOUT, 1); 1302 vmexit->exitcode = VM_EXITCODE_INOUT; 1303 vmexit->u.inout.bytes = (qual & 0x7) + 1; 1304 vmexit->u.inout.in = (qual & 0x8) ? 1 : 0; 1305 vmexit->u.inout.string = (qual & 0x10) ? 1 : 0; 1306 vmexit->u.inout.rep = (qual & 0x20) ? 1 : 0; 1307 vmexit->u.inout.port = (uint16_t)(qual >> 16); 1308 vmexit->u.inout.eax = (uint32_t)(vmxctx->guest_rax); 1309 break; 1310 case EXIT_REASON_CPUID: 1311 vmm_stat_incr(vmx->vm, vcpu, VMEXIT_CPUID, 1); 1312 handled = vmx_handle_cpuid(vmx->vm, vcpu, vmxctx); 1313 break; 1314 case EXIT_REASON_EPT_FAULT: 1315 vmm_stat_incr(vmx->vm, vcpu, VMEXIT_EPT_FAULT, 1); 1316 gla = vmcs_gla(); 1317 gpa = vmcs_gpa(); 1318 cr3 = vmcs_guest_cr3(); 1319 handled = vmx_ept_fault(vmx->vm, vcpu, gla, gpa, 1320 vmexit->rip, vmexit->inst_length, 1321 cr3, qual, &vmexit->u.paging.vie); 1322 if (!handled) { 1323 vmexit->exitcode = VM_EXITCODE_PAGING; 1324 vmexit->u.paging.gpa = gpa; 1325 } 1326 break; 1327 default: 1328 vmm_stat_incr(vmx->vm, vcpu, VMEXIT_UNKNOWN, 1); 1329 break; 1330 } 1331 1332 if (handled) { 1333 /* 1334 * It is possible that control is returned to userland 1335 * even though we were able to handle the VM exit in the 1336 * kernel. 1337 * 1338 * In such a case we want to make sure that the userland 1339 * restarts guest execution at the instruction *after* 1340 * the one we just processed. Therefore we update the 1341 * guest rip in the VMCS and in 'vmexit'. 1342 */ 1343 vm_exit_update_rip(vmexit); 1344 vmexit->rip += vmexit->inst_length; 1345 vmexit->inst_length = 0; 1346 1347 /* 1348 * Special case for spinning up an AP - exit to userspace to 1349 * give the controlling process a chance to intercept and 1350 * spin up a thread for the AP. 1351 */ 1352 if (vmexit->exitcode == VM_EXITCODE_SPINUP_AP) 1353 handled = 0; 1354 } else { 1355 if (vmexit->exitcode == VM_EXITCODE_BOGUS) { 1356 /* 1357 * If this VM exit was not claimed by anybody then 1358 * treat it as a generic VMX exit. 1359 */ 1360 vmexit->exitcode = VM_EXITCODE_VMX; 1361 vmexit->u.vmx.error = 0; 1362 } else { 1363 /* 1364 * The exitcode and collateral have been populated. 1365 * The VM exit will be processed further in userland. 1366 */ 1367 } 1368 } 1369 return (handled); 1370} 1371 1372static int 1373vmx_run(void *arg, int vcpu, register_t rip) 1374{ 1375 int error, vie, rc, handled, astpending; 1376 uint32_t exit_reason; 1377 struct vmx *vmx; 1378 struct vmxctx *vmxctx; 1379 struct vmcs *vmcs; 1380 struct vm_exit *vmexit; 1381 1382 vmx = arg; 1383 vmcs = &vmx->vmcs[vcpu]; 1384 vmxctx = &vmx->ctx[vcpu]; 1385 vmxctx->launched = 0; 1386 1387 astpending = 0; 1388 vmexit = vm_exitinfo(vmx->vm, vcpu); 1389 1390 /* 1391 * XXX Can we avoid doing this every time we do a vm run? 1392 */ 1393 VMPTRLD(vmcs); 1394 1395 /* 1396 * XXX 1397 * We do this every time because we may setup the virtual machine 1398 * from a different process than the one that actually runs it. 1399 * 1400 * If the life of a virtual machine was spent entirely in the context 1401 * of a single process we could do this once in vmcs_set_defaults(). 1402 */ 1403 if ((error = vmwrite(VMCS_HOST_CR3, rcr3())) != 0) 1404 panic("vmx_run: error %d writing to VMCS_HOST_CR3", error); 1405 1406 if ((error = vmwrite(VMCS_GUEST_RIP, rip)) != 0) 1407 panic("vmx_run: error %d writing to VMCS_GUEST_RIP", error); 1408 1409 if ((error = vmx_set_pcpu_defaults(vmx, vcpu)) != 0) 1410 panic("vmx_run: error %d setting up pcpu defaults", error); 1411 1412 do { 1413 lapic_timer_tick(vmx->vm, vcpu); 1414 vmx_inject_interrupts(vmx, vcpu); 1415 vmx_run_trace(vmx, vcpu); 1416 rc = vmx_setjmp(vmxctx); 1417#ifdef SETJMP_TRACE 1418 vmx_setjmp_trace(vmx, vcpu, vmxctx, rc); 1419#endif 1420 switch (rc) { 1421 case VMX_RETURN_DIRECT: 1422 if (vmxctx->launched == 0) { 1423 vmxctx->launched = 1; 1424 vmx_launch(vmxctx); 1425 } else 1426 vmx_resume(vmxctx); 1427 panic("vmx_launch/resume should not return"); 1428 break; 1429 case VMX_RETURN_LONGJMP: 1430 break; /* vm exit */ 1431 case VMX_RETURN_AST: 1432 astpending = 1; 1433 break; 1434 case VMX_RETURN_VMRESUME: 1435 vie = vmcs_instruction_error(); 1436 if (vmxctx->launch_error == VM_FAIL_INVALID || 1437 vie != VMRESUME_WITH_NON_LAUNCHED_VMCS) { 1438 printf("vmresume error %d vmcs inst error %d\n", 1439 vmxctx->launch_error, vie); 1440 goto err_exit; 1441 } 1442 vmx_launch(vmxctx); /* try to launch the guest */ 1443 panic("vmx_launch should not return"); 1444 break; 1445 case VMX_RETURN_VMLAUNCH: 1446 vie = vmcs_instruction_error(); 1447#if 1 1448 printf("vmlaunch error %d vmcs inst error %d\n", 1449 vmxctx->launch_error, vie); 1450#endif 1451 goto err_exit; 1452 default: 1453 panic("vmx_setjmp returned %d", rc); 1454 } 1455 1456 /* enable interrupts */ 1457 enable_intr(); 1458 1459 /* collect some basic information for VM exit processing */ 1460 vmexit->rip = rip = vmcs_guest_rip(); 1461 vmexit->inst_length = vmexit_instruction_length(); 1462 vmexit->u.vmx.exit_reason = exit_reason = vmcs_exit_reason(); 1463 vmexit->u.vmx.exit_qualification = vmcs_exit_qualification(); 1464 1465 if (astpending) { 1466 handled = 1; 1467 vmexit->inst_length = 0; 1468 vmexit->exitcode = VM_EXITCODE_BOGUS; 1469 vmx_astpending_trace(vmx, vcpu, rip); 1470 vmm_stat_incr(vmx->vm, vcpu, VMEXIT_ASTPENDING, 1); 1471 break; 1472 } 1473 1474 handled = vmx_exit_process(vmx, vcpu, vmexit); 1475 vmx_exit_trace(vmx, vcpu, rip, exit_reason, handled); 1476 1477 } while (handled); 1478 1479 /* 1480 * If a VM exit has been handled then the exitcode must be BOGUS 1481 * If a VM exit is not handled then the exitcode must not be BOGUS 1482 */ 1483 if ((handled && vmexit->exitcode != VM_EXITCODE_BOGUS) || 1484 (!handled && vmexit->exitcode == VM_EXITCODE_BOGUS)) { 1485 panic("Mismatch between handled (%d) and exitcode (%d)", 1486 handled, vmexit->exitcode); 1487 } 1488 1489 if (!handled) 1490 vmm_stat_incr(vmx->vm, vcpu, VMEXIT_USERSPACE, 1); 1491 1492 VMM_CTR1(vmx->vm, vcpu, "goto userland: exitcode %d",vmexit->exitcode); 1493 1494 /* 1495 * XXX 1496 * We need to do this to ensure that any VMCS state cached by the 1497 * processor is flushed to memory. We need to do this in case the 1498 * VM moves to a different cpu the next time it runs. 1499 * 1500 * Can we avoid doing this? 1501 */ 1502 VMCLEAR(vmcs); 1503 return (0); 1504 1505err_exit: 1506 vmexit->exitcode = VM_EXITCODE_VMX; 1507 vmexit->u.vmx.exit_reason = (uint32_t)-1; 1508 vmexit->u.vmx.exit_qualification = (uint32_t)-1; 1509 vmexit->u.vmx.error = vie; 1510 VMCLEAR(vmcs); 1511 return (ENOEXEC); 1512} 1513 1514static void 1515vmx_vmcleanup(void *arg) 1516{ 1517 int error; 1518 struct vmx *vmx = arg; 1519 1520 /* 1521 * XXXSMP we also need to clear the VMCS active on the other vcpus. 1522 */ 1523 error = vmclear(&vmx->vmcs[0]); 1524 if (error != 0) 1525 panic("vmx_vmcleanup: vmclear error %d on vcpu 0", error); 1526 1527 ept_vmcleanup(vmx); 1528 free(vmx, M_VMX); 1529 1530 return; 1531} 1532 1533static register_t * 1534vmxctx_regptr(struct vmxctx *vmxctx, int reg) 1535{ 1536 1537 switch (reg) { 1538 case VM_REG_GUEST_RAX: 1539 return (&vmxctx->guest_rax); 1540 case VM_REG_GUEST_RBX: 1541 return (&vmxctx->guest_rbx); 1542 case VM_REG_GUEST_RCX: 1543 return (&vmxctx->guest_rcx); 1544 case VM_REG_GUEST_RDX: 1545 return (&vmxctx->guest_rdx); 1546 case VM_REG_GUEST_RSI: 1547 return (&vmxctx->guest_rsi); 1548 case VM_REG_GUEST_RDI: 1549 return (&vmxctx->guest_rdi); 1550 case VM_REG_GUEST_RBP: 1551 return (&vmxctx->guest_rbp); 1552 case VM_REG_GUEST_R8: 1553 return (&vmxctx->guest_r8); 1554 case VM_REG_GUEST_R9: 1555 return (&vmxctx->guest_r9); 1556 case VM_REG_GUEST_R10: 1557 return (&vmxctx->guest_r10); 1558 case VM_REG_GUEST_R11: 1559 return (&vmxctx->guest_r11); 1560 case VM_REG_GUEST_R12: 1561 return (&vmxctx->guest_r12); 1562 case VM_REG_GUEST_R13: 1563 return (&vmxctx->guest_r13); 1564 case VM_REG_GUEST_R14: 1565 return (&vmxctx->guest_r14); 1566 case VM_REG_GUEST_R15: 1567 return (&vmxctx->guest_r15); 1568 default: 1569 break; 1570 } 1571 return (NULL); 1572} 1573 1574static int 1575vmxctx_getreg(struct vmxctx *vmxctx, int reg, uint64_t *retval) 1576{ 1577 register_t *regp; 1578 1579 if ((regp = vmxctx_regptr(vmxctx, reg)) != NULL) { 1580 *retval = *regp; 1581 return (0); 1582 } else 1583 return (EINVAL); 1584} 1585 1586static int 1587vmxctx_setreg(struct vmxctx *vmxctx, int reg, uint64_t val) 1588{ 1589 register_t *regp; 1590 1591 if ((regp = vmxctx_regptr(vmxctx, reg)) != NULL) { 1592 *regp = val; 1593 return (0); 1594 } else 1595 return (EINVAL); 1596} 1597 1598static int 1599vmx_getreg(void *arg, int vcpu, int reg, uint64_t *retval) 1600{ 1601 struct vmx *vmx = arg; 1602 1603 if (vmxctx_getreg(&vmx->ctx[vcpu], reg, retval) == 0) 1604 return (0); 1605 1606 /* 1607 * If the vcpu is running then don't mess with the VMCS. 1608 * 1609 * vmcs_getreg will VMCLEAR the vmcs when it is done which will cause 1610 * the subsequent vmlaunch/vmresume to fail. 1611 */ 1612 if (vcpu_is_running(vmx->vm, vcpu)) 1613 panic("vmx_getreg: %s%d is running", vm_name(vmx->vm), vcpu); 1614 1615 return (vmcs_getreg(&vmx->vmcs[vcpu], reg, retval)); 1616} 1617 1618static int 1619vmx_setreg(void *arg, int vcpu, int reg, uint64_t val) 1620{ 1621 int error; 1622 uint64_t ctls; 1623 struct vmx *vmx = arg; 1624 1625 /* 1626 * XXX Allow caller to set contents of the guest registers saved in 1627 * the 'vmxctx' even though the vcpu might be running. We need this 1628 * specifically to support the rdmsr emulation that will set the 1629 * %eax and %edx registers during vm exit processing. 1630 */ 1631 if (vmxctx_setreg(&vmx->ctx[vcpu], reg, val) == 0) 1632 return (0); 1633 1634 /* 1635 * If the vcpu is running then don't mess with the VMCS. 1636 * 1637 * vmcs_setreg will VMCLEAR the vmcs when it is done which will cause 1638 * the subsequent vmlaunch/vmresume to fail. 1639 */ 1640 if (vcpu_is_running(vmx->vm, vcpu)) 1641 panic("vmx_setreg: %s%d is running", vm_name(vmx->vm), vcpu); 1642 1643 error = vmcs_setreg(&vmx->vmcs[vcpu], reg, val); 1644 1645 if (error == 0) { 1646 /* 1647 * If the "load EFER" VM-entry control is 1 then the 1648 * value of EFER.LMA must be identical to "IA-32e mode guest" 1649 * bit in the VM-entry control. 1650 */ 1651 if ((entry_ctls & VM_ENTRY_LOAD_EFER) != 0 && 1652 (reg == VM_REG_GUEST_EFER)) { 1653 vmcs_getreg(&vmx->vmcs[vcpu], 1654 VMCS_IDENT(VMCS_ENTRY_CTLS), &ctls); 1655 if (val & EFER_LMA) 1656 ctls |= VM_ENTRY_GUEST_LMA; 1657 else 1658 ctls &= ~VM_ENTRY_GUEST_LMA; 1659 vmcs_setreg(&vmx->vmcs[vcpu], 1660 VMCS_IDENT(VMCS_ENTRY_CTLS), ctls); 1661 } 1662 } 1663 1664 return (error); 1665} 1666 1667static int 1668vmx_getdesc(void *arg, int vcpu, int reg, struct seg_desc *desc) 1669{ 1670 struct vmx *vmx = arg; 1671 1672 return (vmcs_getdesc(&vmx->vmcs[vcpu], reg, desc)); 1673} 1674 1675static int 1676vmx_setdesc(void *arg, int vcpu, int reg, struct seg_desc *desc) 1677{ 1678 struct vmx *vmx = arg; 1679 1680 return (vmcs_setdesc(&vmx->vmcs[vcpu], reg, desc)); 1681} 1682 1683static int 1684vmx_inject(void *arg, int vcpu, int type, int vector, uint32_t code, 1685 int code_valid) 1686{ 1687 int error; 1688 uint64_t info; 1689 struct vmx *vmx = arg; 1690 struct vmcs *vmcs = &vmx->vmcs[vcpu]; 1691 1692 static uint32_t type_map[VM_EVENT_MAX] = { 1693 0x1, /* VM_EVENT_NONE */ 1694 0x0, /* VM_HW_INTR */ 1695 0x2, /* VM_NMI */ 1696 0x3, /* VM_HW_EXCEPTION */ 1697 0x4, /* VM_SW_INTR */ 1698 0x5, /* VM_PRIV_SW_EXCEPTION */ 1699 0x6, /* VM_SW_EXCEPTION */ 1700 }; 1701 1702 /* 1703 * If there is already an exception pending to be delivered to the 1704 * vcpu then just return. 1705 */ 1706 error = vmcs_getreg(vmcs, VMCS_IDENT(VMCS_ENTRY_INTR_INFO), &info); 1707 if (error) 1708 return (error); 1709 1710 if (info & VMCS_INTERRUPTION_INFO_VALID) 1711 return (EAGAIN); 1712 1713 info = vector | (type_map[type] << 8) | (code_valid ? 1 << 11 : 0); 1714 info |= VMCS_INTERRUPTION_INFO_VALID; 1715 error = vmcs_setreg(vmcs, VMCS_IDENT(VMCS_ENTRY_INTR_INFO), info); 1716 if (error != 0) 1717 return (error); 1718 1719 if (code_valid) { 1720 error = vmcs_setreg(vmcs, 1721 VMCS_IDENT(VMCS_ENTRY_EXCEPTION_ERROR), 1722 code); 1723 } 1724 return (error); 1725} 1726 1727static int 1728vmx_getcap(void *arg, int vcpu, int type, int *retval) 1729{ 1730 struct vmx *vmx = arg; 1731 int vcap; 1732 int ret; 1733 1734 ret = ENOENT; 1735 1736 vcap = vmx->cap[vcpu].set; 1737 1738 switch (type) { 1739 case VM_CAP_HALT_EXIT: 1740 if (cap_halt_exit) 1741 ret = 0; 1742 break; 1743 case VM_CAP_PAUSE_EXIT: 1744 if (cap_pause_exit) 1745 ret = 0; 1746 break; 1747 case VM_CAP_MTRAP_EXIT: 1748 if (cap_monitor_trap) 1749 ret = 0; 1750 break; 1751 case VM_CAP_UNRESTRICTED_GUEST: 1752 if (cap_unrestricted_guest) 1753 ret = 0; 1754 break; 1755 default: 1756 break; 1757 } 1758 1759 if (ret == 0) 1760 *retval = (vcap & (1 << type)) ? 1 : 0; 1761 1762 return (ret); 1763} 1764 1765static int 1766vmx_setcap(void *arg, int vcpu, int type, int val) 1767{ 1768 struct vmx *vmx = arg; 1769 struct vmcs *vmcs = &vmx->vmcs[vcpu]; 1770 uint32_t baseval; 1771 uint32_t *pptr; 1772 int error; 1773 int flag; 1774 int reg; 1775 int retval; 1776 1777 retval = ENOENT; 1778 pptr = NULL; 1779 1780 switch (type) { 1781 case VM_CAP_HALT_EXIT: 1782 if (cap_halt_exit) { 1783 retval = 0; 1784 pptr = &vmx->cap[vcpu].proc_ctls; 1785 baseval = *pptr; 1786 flag = PROCBASED_HLT_EXITING; 1787 reg = VMCS_PRI_PROC_BASED_CTLS; 1788 } 1789 break; 1790 case VM_CAP_MTRAP_EXIT: 1791 if (cap_monitor_trap) { 1792 retval = 0; 1793 pptr = &vmx->cap[vcpu].proc_ctls; 1794 baseval = *pptr; 1795 flag = PROCBASED_MTF; 1796 reg = VMCS_PRI_PROC_BASED_CTLS; 1797 } 1798 break; 1799 case VM_CAP_PAUSE_EXIT: 1800 if (cap_pause_exit) { 1801 retval = 0; 1802 pptr = &vmx->cap[vcpu].proc_ctls; 1803 baseval = *pptr; 1804 flag = PROCBASED_PAUSE_EXITING; 1805 reg = VMCS_PRI_PROC_BASED_CTLS; 1806 } 1807 break; 1808 case VM_CAP_UNRESTRICTED_GUEST: 1809 if (cap_unrestricted_guest) { 1810 retval = 0; 1811 baseval = procbased_ctls2; 1812 flag = PROCBASED2_UNRESTRICTED_GUEST; 1813 reg = VMCS_SEC_PROC_BASED_CTLS; 1814 } 1815 break; 1816 default: 1817 break; 1818 } 1819 1820 if (retval == 0) { 1821 if (val) { 1822 baseval |= flag; 1823 } else { 1824 baseval &= ~flag; 1825 } 1826 VMPTRLD(vmcs); 1827 error = vmwrite(reg, baseval); 1828 VMCLEAR(vmcs); 1829 1830 if (error) { 1831 retval = error; 1832 } else { 1833 /* 1834 * Update optional stored flags, and record 1835 * setting 1836 */ 1837 if (pptr != NULL) { 1838 *pptr = baseval; 1839 } 1840 1841 if (val) { 1842 vmx->cap[vcpu].set |= (1 << type); 1843 } else { 1844 vmx->cap[vcpu].set &= ~(1 << type); 1845 } 1846 } 1847 } 1848 1849 return (retval); 1850} 1851 1852struct vmm_ops vmm_ops_intel = { 1853 vmx_init, 1854 vmx_cleanup, 1855 vmx_vminit, 1856 vmx_run, 1857 vmx_vmcleanup, 1858 ept_vmmmap_set, 1859 ept_vmmmap_get, 1860 vmx_getreg, 1861 vmx_setreg, 1862 vmx_getdesc, 1863 vmx_setdesc, 1864 vmx_inject, 1865 vmx_getcap, 1866 vmx_setcap 1867}; 1868