1 /*- 2 * SPDX-License-Identifier: BSD-2-Clause-FreeBSD 3 * 4 * Copyright (c) 2011 NetApp, Inc. 5 * All rights reserved. 6 * 7 * Redistribution and use in source and binary forms, with or without 8 * modification, are permitted provided that the following conditions 9 * are met: 10 * 1. Redistributions of source code must retain the above copyright 11 * notice, this list of conditions and the following disclaimer. 12 * 2. Redistributions in binary form must reproduce the above copyright 13 * notice, this list of conditions and the following disclaimer in the 14 * documentation and/or other materials provided with the distribution. 15 * 16 * THIS SOFTWARE IS PROVIDED BY NETAPP, INC ``AS IS'' AND 17 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 18 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 19 * ARE DISCLAIMED. IN NO EVENT SHALL NETAPP, INC OR CONTRIBUTORS BE LIABLE 20 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 21 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 22 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 23 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 24 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 25 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 26 * SUCH DAMAGE. 27 * 28 * $FreeBSD$ 29 */ 30 31 #include <sys/cdefs.h> 32 __FBSDID("$FreeBSD$"); 33 34 #include <sys/param.h> 35 #include <sys/systm.h> 36 #include <sys/smp.h> 37 #include <sys/kernel.h> 38 #include <sys/malloc.h> 39 #include <sys/pcpu.h> 40 #include <sys/proc.h> 41 #include <sys/sysctl.h> 42 43 #include <vm/vm.h> 44 #include <vm/pmap.h> 45 46 #include <machine/psl.h> 47 #include <machine/cpufunc.h> 48 #include <machine/md_var.h> 49 #include <machine/segments.h> 50 #include <machine/smp.h> 51 #include <machine/specialreg.h> 52 #include <machine/vmparam.h> 53 54 #include <machine/vmm.h> 55 #include <machine/vmm_dev.h> 56 #include <machine/vmm_instruction_emul.h> 57 #include "vmm_lapic.h" 58 #include "vmm_host.h" 59 #include "vmm_ioport.h" 60 #include "vmm_ktr.h" 61 #include "vmm_stat.h" 62 #include "vatpic.h" 63 #include "vlapic.h" 64 #include "vlapic_priv.h" 65 66 #include "ept.h" 67 #include "vmx_cpufunc.h" 68 #include "vmx.h" 69 #include "vmx_msr.h" 70 #include "x86.h" 71 #include "vmx_controls.h" 72 73 #define PINBASED_CTLS_ONE_SETTING \ 74 (PINBASED_EXTINT_EXITING | \ 75 PINBASED_NMI_EXITING | \ 76 PINBASED_VIRTUAL_NMI) 77 #define PINBASED_CTLS_ZERO_SETTING 0 78 79 #define PROCBASED_CTLS_WINDOW_SETTING \ 80 (PROCBASED_INT_WINDOW_EXITING | \ 81 PROCBASED_NMI_WINDOW_EXITING) 82 83 #define PROCBASED_CTLS_ONE_SETTING \ 84 (PROCBASED_SECONDARY_CONTROLS | \ 85 PROCBASED_MWAIT_EXITING | \ 86 PROCBASED_MONITOR_EXITING | \ 87 PROCBASED_IO_EXITING | \ 88 PROCBASED_MSR_BITMAPS | \ 89 PROCBASED_CTLS_WINDOW_SETTING | \ 90 PROCBASED_CR8_LOAD_EXITING | \ 91 PROCBASED_CR8_STORE_EXITING) 92 #define PROCBASED_CTLS_ZERO_SETTING \ 93 (PROCBASED_CR3_LOAD_EXITING | \ 94 PROCBASED_CR3_STORE_EXITING | \ 95 PROCBASED_IO_BITMAPS) 96 97 #define PROCBASED_CTLS2_ONE_SETTING PROCBASED2_ENABLE_EPT 98 #define PROCBASED_CTLS2_ZERO_SETTING 0 99 100 #define VM_EXIT_CTLS_ONE_SETTING \ 101 (VM_EXIT_HOST_LMA | \ 102 VM_EXIT_SAVE_EFER | \ 103 VM_EXIT_LOAD_EFER | \ 104 VM_EXIT_ACKNOWLEDGE_INTERRUPT) 105 106 #define VM_EXIT_CTLS_ZERO_SETTING VM_EXIT_SAVE_DEBUG_CONTROLS 107 108 #define VM_ENTRY_CTLS_ONE_SETTING (VM_ENTRY_LOAD_EFER) 109 110 #define VM_ENTRY_CTLS_ZERO_SETTING \ 111 (VM_ENTRY_LOAD_DEBUG_CONTROLS | \ 112 VM_ENTRY_INTO_SMM | \ 113 VM_ENTRY_DEACTIVATE_DUAL_MONITOR) 114 115 #define HANDLED 1 116 #define UNHANDLED 0 117 118 static MALLOC_DEFINE(M_VMX, "vmx", "vmx"); 119 static MALLOC_DEFINE(M_VLAPIC, "vlapic", "vlapic"); 120 121 SYSCTL_DECL(_hw_vmm); 122 SYSCTL_NODE(_hw_vmm, OID_AUTO, vmx, CTLFLAG_RW, NULL, NULL); 123 124 int vmxon_enabled[MAXCPU]; 125 static char vmxon_region[MAXCPU][PAGE_SIZE] __aligned(PAGE_SIZE); 126 127 static uint32_t pinbased_ctls, procbased_ctls, procbased_ctls2; 128 static uint32_t exit_ctls, entry_ctls; 129 130 static uint64_t cr0_ones_mask, cr0_zeros_mask; 131 SYSCTL_ULONG(_hw_vmm_vmx, OID_AUTO, cr0_ones_mask, CTLFLAG_RD, 132 &cr0_ones_mask, 0, NULL); 133 SYSCTL_ULONG(_hw_vmm_vmx, OID_AUTO, cr0_zeros_mask, CTLFLAG_RD, 134 &cr0_zeros_mask, 0, NULL); 135 136 static uint64_t cr4_ones_mask, cr4_zeros_mask; 137 SYSCTL_ULONG(_hw_vmm_vmx, OID_AUTO, cr4_ones_mask, CTLFLAG_RD, 138 &cr4_ones_mask, 0, NULL); 139 SYSCTL_ULONG(_hw_vmm_vmx, OID_AUTO, cr4_zeros_mask, CTLFLAG_RD, 140 &cr4_zeros_mask, 0, NULL); 141 142 static int vmx_initialized; 143 SYSCTL_INT(_hw_vmm_vmx, OID_AUTO, initialized, CTLFLAG_RD, 144 &vmx_initialized, 0, "Intel VMX initialized"); 145 146 /* 147 * Optional capabilities 148 */ 149 static SYSCTL_NODE(_hw_vmm_vmx, OID_AUTO, cap, CTLFLAG_RW, NULL, NULL); 150 151 static int cap_halt_exit; 152 SYSCTL_INT(_hw_vmm_vmx_cap, OID_AUTO, halt_exit, CTLFLAG_RD, &cap_halt_exit, 0, 153 "HLT triggers a VM-exit"); 154 155 static int cap_pause_exit; 156 SYSCTL_INT(_hw_vmm_vmx_cap, OID_AUTO, pause_exit, CTLFLAG_RD, &cap_pause_exit, 157 0, "PAUSE triggers a VM-exit"); 158 159 static int cap_unrestricted_guest; 160 SYSCTL_INT(_hw_vmm_vmx_cap, OID_AUTO, unrestricted_guest, CTLFLAG_RD, 161 &cap_unrestricted_guest, 0, "Unrestricted guests"); 162 163 static int cap_monitor_trap; 164 SYSCTL_INT(_hw_vmm_vmx_cap, OID_AUTO, monitor_trap, CTLFLAG_RD, 165 &cap_monitor_trap, 0, "Monitor trap flag"); 166 167 static int cap_invpcid; 168 SYSCTL_INT(_hw_vmm_vmx_cap, OID_AUTO, invpcid, CTLFLAG_RD, &cap_invpcid, 169 0, "Guests are allowed to use INVPCID"); 170 171 static int virtual_interrupt_delivery; 172 SYSCTL_INT(_hw_vmm_vmx_cap, OID_AUTO, virtual_interrupt_delivery, CTLFLAG_RD, 173 &virtual_interrupt_delivery, 0, "APICv virtual interrupt delivery support"); 174 175 static int posted_interrupts; 176 SYSCTL_INT(_hw_vmm_vmx_cap, OID_AUTO, posted_interrupts, CTLFLAG_RD, 177 &posted_interrupts, 0, "APICv posted interrupt support"); 178 179 static int pirvec = -1; 180 SYSCTL_INT(_hw_vmm_vmx, OID_AUTO, posted_interrupt_vector, CTLFLAG_RD, 181 &pirvec, 0, "APICv posted interrupt vector"); 182 183 static struct unrhdr *vpid_unr; 184 static u_int vpid_alloc_failed; 185 SYSCTL_UINT(_hw_vmm_vmx, OID_AUTO, vpid_alloc_failed, CTLFLAG_RD, 186 &vpid_alloc_failed, 0, NULL); 187 188 /* 189 * Use the last page below 4GB as the APIC access address. This address is 190 * occupied by the boot firmware so it is guaranteed that it will not conflict 191 * with a page in system memory. 192 */ 193 #define APIC_ACCESS_ADDRESS 0xFFFFF000 194 195 static int vmx_getdesc(void *arg, int vcpu, int reg, struct seg_desc *desc); 196 static int vmx_getreg(void *arg, int vcpu, int reg, uint64_t *retval); 197 static int vmxctx_setreg(struct vmxctx *vmxctx, int reg, uint64_t val); 198 static void vmx_inject_pir(struct vlapic *vlapic); 199 200 #ifdef KTR 201 static const char * 202 exit_reason_to_str(int reason) 203 { 204 static char reasonbuf[32]; 205 206 switch (reason) { 207 case EXIT_REASON_EXCEPTION: 208 return "exception"; 209 case EXIT_REASON_EXT_INTR: 210 return "extint"; 211 case EXIT_REASON_TRIPLE_FAULT: 212 return "triplefault"; 213 case EXIT_REASON_INIT: 214 return "init"; 215 case EXIT_REASON_SIPI: 216 return "sipi"; 217 case EXIT_REASON_IO_SMI: 218 return "iosmi"; 219 case EXIT_REASON_SMI: 220 return "smi"; 221 case EXIT_REASON_INTR_WINDOW: 222 return "intrwindow"; 223 case EXIT_REASON_NMI_WINDOW: 224 return "nmiwindow"; 225 case EXIT_REASON_TASK_SWITCH: 226 return "taskswitch"; 227 case EXIT_REASON_CPUID: 228 return "cpuid"; 229 case EXIT_REASON_GETSEC: 230 return "getsec"; 231 case EXIT_REASON_HLT: 232 return "hlt"; 233 case EXIT_REASON_INVD: 234 return "invd"; 235 case EXIT_REASON_INVLPG: 236 return "invlpg"; 237 case EXIT_REASON_RDPMC: 238 return "rdpmc"; 239 case EXIT_REASON_RDTSC: 240 return "rdtsc"; 241 case EXIT_REASON_RSM: 242 return "rsm"; 243 case EXIT_REASON_VMCALL: 244 return "vmcall"; 245 case EXIT_REASON_VMCLEAR: 246 return "vmclear"; 247 case EXIT_REASON_VMLAUNCH: 248 return "vmlaunch"; 249 case EXIT_REASON_VMPTRLD: 250 return "vmptrld"; 251 case EXIT_REASON_VMPTRST: 252 return "vmptrst"; 253 case EXIT_REASON_VMREAD: 254 return "vmread"; 255 case EXIT_REASON_VMRESUME: 256 return "vmresume"; 257 case EXIT_REASON_VMWRITE: 258 return "vmwrite"; 259 case EXIT_REASON_VMXOFF: 260 return "vmxoff"; 261 case EXIT_REASON_VMXON: 262 return "vmxon"; 263 case EXIT_REASON_CR_ACCESS: 264 return "craccess"; 265 case EXIT_REASON_DR_ACCESS: 266 return "draccess"; 267 case EXIT_REASON_INOUT: 268 return "inout"; 269 case EXIT_REASON_RDMSR: 270 return "rdmsr"; 271 case EXIT_REASON_WRMSR: 272 return "wrmsr"; 273 case EXIT_REASON_INVAL_VMCS: 274 return "invalvmcs"; 275 case EXIT_REASON_INVAL_MSR: 276 return "invalmsr"; 277 case EXIT_REASON_MWAIT: 278 return "mwait"; 279 case EXIT_REASON_MTF: 280 return "mtf"; 281 case EXIT_REASON_MONITOR: 282 return "monitor"; 283 case EXIT_REASON_PAUSE: 284 return "pause"; 285 case EXIT_REASON_MCE_DURING_ENTRY: 286 return "mce-during-entry"; 287 case EXIT_REASON_TPR: 288 return "tpr"; 289 case EXIT_REASON_APIC_ACCESS: 290 return "apic-access"; 291 case EXIT_REASON_GDTR_IDTR: 292 return "gdtridtr"; 293 case EXIT_REASON_LDTR_TR: 294 return "ldtrtr"; 295 case EXIT_REASON_EPT_FAULT: 296 return "eptfault"; 297 case EXIT_REASON_EPT_MISCONFIG: 298 return "eptmisconfig"; 299 case EXIT_REASON_INVEPT: 300 return "invept"; 301 case EXIT_REASON_RDTSCP: 302 return "rdtscp"; 303 case EXIT_REASON_VMX_PREEMPT: 304 return "vmxpreempt"; 305 case EXIT_REASON_INVVPID: 306 return "invvpid"; 307 case EXIT_REASON_WBINVD: 308 return "wbinvd"; 309 case EXIT_REASON_XSETBV: 310 return "xsetbv"; 311 case EXIT_REASON_APIC_WRITE: 312 return "apic-write"; 313 default: 314 snprintf(reasonbuf, sizeof(reasonbuf), "%d", reason); 315 return (reasonbuf); 316 } 317 } 318 #endif /* KTR */ 319 320 static int 321 vmx_allow_x2apic_msrs(struct vmx *vmx) 322 { 323 int i, error; 324 325 error = 0; 326 327 /* 328 * Allow readonly access to the following x2APIC MSRs from the guest. 329 */ 330 error += guest_msr_ro(vmx, MSR_APIC_ID); 331 error += guest_msr_ro(vmx, MSR_APIC_VERSION); 332 error += guest_msr_ro(vmx, MSR_APIC_LDR); 333 error += guest_msr_ro(vmx, MSR_APIC_SVR); 334 335 for (i = 0; i < 8; i++) 336 error += guest_msr_ro(vmx, MSR_APIC_ISR0 + i); 337 338 for (i = 0; i < 8; i++) 339 error += guest_msr_ro(vmx, MSR_APIC_TMR0 + i); 340 341 for (i = 0; i < 8; i++) 342 error += guest_msr_ro(vmx, MSR_APIC_IRR0 + i); 343 344 error += guest_msr_ro(vmx, MSR_APIC_ESR); 345 error += guest_msr_ro(vmx, MSR_APIC_LVT_TIMER); 346 error += guest_msr_ro(vmx, MSR_APIC_LVT_THERMAL); 347 error += guest_msr_ro(vmx, MSR_APIC_LVT_PCINT); 348 error += guest_msr_ro(vmx, MSR_APIC_LVT_LINT0); 349 error += guest_msr_ro(vmx, MSR_APIC_LVT_LINT1); 350 error += guest_msr_ro(vmx, MSR_APIC_LVT_ERROR); 351 error += guest_msr_ro(vmx, MSR_APIC_ICR_TIMER); 352 error += guest_msr_ro(vmx, MSR_APIC_DCR_TIMER); 353 error += guest_msr_ro(vmx, MSR_APIC_ICR); 354 355 /* 356 * Allow TPR, EOI and SELF_IPI MSRs to be read and written by the guest. 357 * 358 * These registers get special treatment described in the section 359 * "Virtualizing MSR-Based APIC Accesses". 360 */ 361 error += guest_msr_rw(vmx, MSR_APIC_TPR); 362 error += guest_msr_rw(vmx, MSR_APIC_EOI); 363 error += guest_msr_rw(vmx, MSR_APIC_SELF_IPI); 364 365 return (error); 366 } 367 368 u_long 369 vmx_fix_cr0(u_long cr0) 370 { 371 372 return ((cr0 | cr0_ones_mask) & ~cr0_zeros_mask); 373 } 374 375 u_long 376 vmx_fix_cr4(u_long cr4) 377 { 378 379 return ((cr4 | cr4_ones_mask) & ~cr4_zeros_mask); 380 } 381 382 static void 383 vpid_free(int vpid) 384 { 385 if (vpid < 0 || vpid > 0xffff) 386 panic("vpid_free: invalid vpid %d", vpid); 387 388 /* 389 * VPIDs [0,VM_MAXCPU] are special and are not allocated from 390 * the unit number allocator. 391 */ 392 393 if (vpid > VM_MAXCPU) 394 free_unr(vpid_unr, vpid); 395 } 396 397 static void 398 vpid_alloc(uint16_t *vpid, int num) 399 { 400 int i, x; 401 402 if (num <= 0 || num > VM_MAXCPU) 403 panic("invalid number of vpids requested: %d", num); 404 405 /* 406 * If the "enable vpid" execution control is not enabled then the 407 * VPID is required to be 0 for all vcpus. 408 */ 409 if ((procbased_ctls2 & PROCBASED2_ENABLE_VPID) == 0) { 410 for (i = 0; i < num; i++) 411 vpid[i] = 0; 412 return; 413 } 414 415 /* 416 * Allocate a unique VPID for each vcpu from the unit number allocator. 417 */ 418 for (i = 0; i < num; i++) { 419 x = alloc_unr(vpid_unr); 420 if (x == -1) 421 break; 422 else 423 vpid[i] = x; 424 } 425 426 if (i < num) { 427 atomic_add_int(&vpid_alloc_failed, 1); 428 429 /* 430 * If the unit number allocator does not have enough unique 431 * VPIDs then we need to allocate from the [1,VM_MAXCPU] range. 432 * 433 * These VPIDs are not be unique across VMs but this does not 434 * affect correctness because the combined mappings are also 435 * tagged with the EP4TA which is unique for each VM. 436 * 437 * It is still sub-optimal because the invvpid will invalidate 438 * combined mappings for a particular VPID across all EP4TAs. 439 */ 440 while (i-- > 0) 441 vpid_free(vpid[i]); 442 443 for (i = 0; i < num; i++) 444 vpid[i] = i + 1; 445 } 446 } 447 448 static void 449 vpid_init(void) 450 { 451 /* 452 * VPID 0 is required when the "enable VPID" execution control is 453 * disabled. 454 * 455 * VPIDs [1,VM_MAXCPU] are used as the "overflow namespace" when the 456 * unit number allocator does not have sufficient unique VPIDs to 457 * satisfy the allocation. 458 * 459 * The remaining VPIDs are managed by the unit number allocator. 460 */ 461 vpid_unr = new_unrhdr(VM_MAXCPU + 1, 0xffff, NULL); 462 } 463 464 static void 465 vmx_disable(void *arg __unused) 466 { 467 struct invvpid_desc invvpid_desc = { 0 }; 468 struct invept_desc invept_desc = { 0 }; 469 470 if (vmxon_enabled[curcpu]) { 471 /* 472 * See sections 25.3.3.3 and 25.3.3.4 in Intel Vol 3b. 473 * 474 * VMXON or VMXOFF are not required to invalidate any TLB 475 * caching structures. This prevents potential retention of 476 * cached information in the TLB between distinct VMX episodes. 477 */ 478 invvpid(INVVPID_TYPE_ALL_CONTEXTS, invvpid_desc); 479 invept(INVEPT_TYPE_ALL_CONTEXTS, invept_desc); 480 vmxoff(); 481 } 482 load_cr4(rcr4() & ~CR4_VMXE); 483 } 484 485 static int 486 vmx_cleanup(void) 487 { 488 489 if (pirvec >= 0) 490 lapic_ipi_free(pirvec); 491 492 if (vpid_unr != NULL) { 493 delete_unrhdr(vpid_unr); 494 vpid_unr = NULL; 495 } 496 497 smp_rendezvous(NULL, vmx_disable, NULL, NULL); 498 499 return (0); 500 } 501 502 static void 503 vmx_enable(void *arg __unused) 504 { 505 int error; 506 uint64_t feature_control; 507 508 feature_control = rdmsr(MSR_IA32_FEATURE_CONTROL); 509 if ((feature_control & IA32_FEATURE_CONTROL_LOCK) == 0 || 510 (feature_control & IA32_FEATURE_CONTROL_VMX_EN) == 0) { 511 wrmsr(MSR_IA32_FEATURE_CONTROL, 512 feature_control | IA32_FEATURE_CONTROL_VMX_EN | 513 IA32_FEATURE_CONTROL_LOCK); 514 } 515 516 load_cr4(rcr4() | CR4_VMXE); 517 518 *(uint32_t *)vmxon_region[curcpu] = vmx_revision(); 519 error = vmxon(vmxon_region[curcpu]); 520 if (error == 0) 521 vmxon_enabled[curcpu] = 1; 522 } 523 524 static void 525 vmx_restore(void) 526 { 527 528 if (vmxon_enabled[curcpu]) 529 vmxon(vmxon_region[curcpu]); 530 } 531 532 static int 533 vmx_init(int ipinum) 534 { 535 int error, use_tpr_shadow; 536 uint64_t basic, fixed0, fixed1, feature_control; 537 uint32_t tmp, procbased2_vid_bits; 538 539 /* CPUID.1:ECX[bit 5] must be 1 for processor to support VMX */ 540 if (!(cpu_feature2 & CPUID2_VMX)) { 541 printf("vmx_init: processor does not support VMX operation\n"); 542 return (ENXIO); 543 } 544 545 /* 546 * Verify that MSR_IA32_FEATURE_CONTROL lock and VMXON enable bits 547 * are set (bits 0 and 2 respectively). 548 */ 549 feature_control = rdmsr(MSR_IA32_FEATURE_CONTROL); 550 if ((feature_control & IA32_FEATURE_CONTROL_LOCK) == 1 && 551 (feature_control & IA32_FEATURE_CONTROL_VMX_EN) == 0) { 552 printf("vmx_init: VMX operation disabled by BIOS\n"); 553 return (ENXIO); 554 } 555 556 /* 557 * Verify capabilities MSR_VMX_BASIC: 558 * - bit 54 indicates support for INS/OUTS decoding 559 */ 560 basic = rdmsr(MSR_VMX_BASIC); 561 if ((basic & (1UL << 54)) == 0) { 562 printf("vmx_init: processor does not support desired basic " 563 "capabilities\n"); 564 return (EINVAL); 565 } 566 567 /* Check support for primary processor-based VM-execution controls */ 568 error = vmx_set_ctlreg(MSR_VMX_PROCBASED_CTLS, 569 MSR_VMX_TRUE_PROCBASED_CTLS, 570 PROCBASED_CTLS_ONE_SETTING, 571 PROCBASED_CTLS_ZERO_SETTING, &procbased_ctls); 572 if (error) { 573 printf("vmx_init: processor does not support desired primary " 574 "processor-based controls\n"); 575 return (error); 576 } 577 578 /* Clear the processor-based ctl bits that are set on demand */ 579 procbased_ctls &= ~PROCBASED_CTLS_WINDOW_SETTING; 580 581 /* Check support for secondary processor-based VM-execution controls */ 582 error = vmx_set_ctlreg(MSR_VMX_PROCBASED_CTLS2, 583 MSR_VMX_PROCBASED_CTLS2, 584 PROCBASED_CTLS2_ONE_SETTING, 585 PROCBASED_CTLS2_ZERO_SETTING, &procbased_ctls2); 586 if (error) { 587 printf("vmx_init: processor does not support desired secondary " 588 "processor-based controls\n"); 589 return (error); 590 } 591 592 /* Check support for VPID */ 593 error = vmx_set_ctlreg(MSR_VMX_PROCBASED_CTLS2, MSR_VMX_PROCBASED_CTLS2, 594 PROCBASED2_ENABLE_VPID, 0, &tmp); 595 if (error == 0) 596 procbased_ctls2 |= PROCBASED2_ENABLE_VPID; 597 598 /* Check support for pin-based VM-execution controls */ 599 error = vmx_set_ctlreg(MSR_VMX_PINBASED_CTLS, 600 MSR_VMX_TRUE_PINBASED_CTLS, 601 PINBASED_CTLS_ONE_SETTING, 602 PINBASED_CTLS_ZERO_SETTING, &pinbased_ctls); 603 if (error) { 604 printf("vmx_init: processor does not support desired " 605 "pin-based controls\n"); 606 return (error); 607 } 608 609 /* Check support for VM-exit controls */ 610 error = vmx_set_ctlreg(MSR_VMX_EXIT_CTLS, MSR_VMX_TRUE_EXIT_CTLS, 611 VM_EXIT_CTLS_ONE_SETTING, 612 VM_EXIT_CTLS_ZERO_SETTING, 613 &exit_ctls); 614 if (error) { 615 printf("vmx_init: processor does not support desired " 616 "exit controls\n"); 617 return (error); 618 } 619 620 /* Check support for VM-entry controls */ 621 error = vmx_set_ctlreg(MSR_VMX_ENTRY_CTLS, MSR_VMX_TRUE_ENTRY_CTLS, 622 VM_ENTRY_CTLS_ONE_SETTING, VM_ENTRY_CTLS_ZERO_SETTING, 623 &entry_ctls); 624 if (error) { 625 printf("vmx_init: processor does not support desired " 626 "entry controls\n"); 627 return (error); 628 } 629 630 /* 631 * Check support for optional features by testing them 632 * as individual bits 633 */ 634 cap_halt_exit = (vmx_set_ctlreg(MSR_VMX_PROCBASED_CTLS, 635 MSR_VMX_TRUE_PROCBASED_CTLS, 636 PROCBASED_HLT_EXITING, 0, 637 &tmp) == 0); 638 639 cap_monitor_trap = (vmx_set_ctlreg(MSR_VMX_PROCBASED_CTLS, 640 MSR_VMX_PROCBASED_CTLS, 641 PROCBASED_MTF, 0, 642 &tmp) == 0); 643 644 cap_pause_exit = (vmx_set_ctlreg(MSR_VMX_PROCBASED_CTLS, 645 MSR_VMX_TRUE_PROCBASED_CTLS, 646 PROCBASED_PAUSE_EXITING, 0, 647 &tmp) == 0); 648 649 cap_unrestricted_guest = (vmx_set_ctlreg(MSR_VMX_PROCBASED_CTLS2, 650 MSR_VMX_PROCBASED_CTLS2, 651 PROCBASED2_UNRESTRICTED_GUEST, 0, 652 &tmp) == 0); 653 654 cap_invpcid = (vmx_set_ctlreg(MSR_VMX_PROCBASED_CTLS2, 655 MSR_VMX_PROCBASED_CTLS2, PROCBASED2_ENABLE_INVPCID, 0, 656 &tmp) == 0); 657 658 /* 659 * Check support for virtual interrupt delivery. 660 */ 661 procbased2_vid_bits = (PROCBASED2_VIRTUALIZE_APIC_ACCESSES | 662 PROCBASED2_VIRTUALIZE_X2APIC_MODE | 663 PROCBASED2_APIC_REGISTER_VIRTUALIZATION | 664 PROCBASED2_VIRTUAL_INTERRUPT_DELIVERY); 665 666 use_tpr_shadow = (vmx_set_ctlreg(MSR_VMX_PROCBASED_CTLS, 667 MSR_VMX_TRUE_PROCBASED_CTLS, PROCBASED_USE_TPR_SHADOW, 0, 668 &tmp) == 0); 669 670 error = vmx_set_ctlreg(MSR_VMX_PROCBASED_CTLS2, MSR_VMX_PROCBASED_CTLS2, 671 procbased2_vid_bits, 0, &tmp); 672 if (error == 0 && use_tpr_shadow) { 673 virtual_interrupt_delivery = 1; 674 TUNABLE_INT_FETCH("hw.vmm.vmx.use_apic_vid", 675 &virtual_interrupt_delivery); 676 } 677 678 if (virtual_interrupt_delivery) { 679 procbased_ctls |= PROCBASED_USE_TPR_SHADOW; 680 procbased_ctls2 |= procbased2_vid_bits; 681 procbased_ctls2 &= ~PROCBASED2_VIRTUALIZE_X2APIC_MODE; 682 683 /* 684 * No need to emulate accesses to %CR8 if virtual 685 * interrupt delivery is enabled. 686 */ 687 procbased_ctls &= ~PROCBASED_CR8_LOAD_EXITING; 688 procbased_ctls &= ~PROCBASED_CR8_STORE_EXITING; 689 690 /* 691 * Check for Posted Interrupts only if Virtual Interrupt 692 * Delivery is enabled. 693 */ 694 error = vmx_set_ctlreg(MSR_VMX_PINBASED_CTLS, 695 MSR_VMX_TRUE_PINBASED_CTLS, PINBASED_POSTED_INTERRUPT, 0, 696 &tmp); 697 if (error == 0) { 698 pirvec = lapic_ipi_alloc(&IDTVEC(justreturn)); 699 if (pirvec < 0) { 700 if (bootverbose) { 701 printf("vmx_init: unable to allocate " 702 "posted interrupt vector\n"); 703 } 704 } else { 705 posted_interrupts = 1; 706 TUNABLE_INT_FETCH("hw.vmm.vmx.use_apic_pir", 707 &posted_interrupts); 708 } 709 } 710 } 711 712 if (posted_interrupts) 713 pinbased_ctls |= PINBASED_POSTED_INTERRUPT; 714 715 /* Initialize EPT */ 716 error = ept_init(ipinum); 717 if (error) { 718 printf("vmx_init: ept initialization failed (%d)\n", error); 719 return (error); 720 } 721 722 /* 723 * Stash the cr0 and cr4 bits that must be fixed to 0 or 1 724 */ 725 fixed0 = rdmsr(MSR_VMX_CR0_FIXED0); 726 fixed1 = rdmsr(MSR_VMX_CR0_FIXED1); 727 cr0_ones_mask = fixed0 & fixed1; 728 cr0_zeros_mask = ~fixed0 & ~fixed1; 729 730 /* 731 * CR0_PE and CR0_PG can be set to zero in VMX non-root operation 732 * if unrestricted guest execution is allowed. 733 */ 734 if (cap_unrestricted_guest) 735 cr0_ones_mask &= ~(CR0_PG | CR0_PE); 736 737 /* 738 * Do not allow the guest to set CR0_NW or CR0_CD. 739 */ 740 cr0_zeros_mask |= (CR0_NW | CR0_CD); 741 742 fixed0 = rdmsr(MSR_VMX_CR4_FIXED0); 743 fixed1 = rdmsr(MSR_VMX_CR4_FIXED1); 744 cr4_ones_mask = fixed0 & fixed1; 745 cr4_zeros_mask = ~fixed0 & ~fixed1; 746 747 vpid_init(); 748 749 vmx_msr_init(); 750 751 /* enable VMX operation */ 752 smp_rendezvous(NULL, vmx_enable, NULL, NULL); 753 754 vmx_initialized = 1; 755 756 return (0); 757 } 758 759 static void 760 vmx_trigger_hostintr(int vector) 761 { 762 uintptr_t func; 763 struct gate_descriptor *gd; 764 765 gd = &idt[vector]; 766 767 KASSERT(vector >= 32 && vector <= 255, ("vmx_trigger_hostintr: " 768 "invalid vector %d", vector)); 769 KASSERT(gd->gd_p == 1, ("gate descriptor for vector %d not present", 770 vector)); 771 KASSERT(gd->gd_type == SDT_SYSIGT, ("gate descriptor for vector %d " 772 "has invalid type %d", vector, gd->gd_type)); 773 KASSERT(gd->gd_dpl == SEL_KPL, ("gate descriptor for vector %d " 774 "has invalid dpl %d", vector, gd->gd_dpl)); 775 KASSERT(gd->gd_selector == GSEL(GCODE_SEL, SEL_KPL), ("gate descriptor " 776 "for vector %d has invalid selector %d", vector, gd->gd_selector)); 777 KASSERT(gd->gd_ist == 0, ("gate descriptor for vector %d has invalid " 778 "IST %d", vector, gd->gd_ist)); 779 780 func = ((long)gd->gd_hioffset << 16 | gd->gd_looffset); 781 vmx_call_isr(func); 782 } 783 784 static int 785 vmx_setup_cr_shadow(int which, struct vmcs *vmcs, uint32_t initial) 786 { 787 int error, mask_ident, shadow_ident; 788 uint64_t mask_value; 789 790 if (which != 0 && which != 4) 791 panic("vmx_setup_cr_shadow: unknown cr%d", which); 792 793 if (which == 0) { 794 mask_ident = VMCS_CR0_MASK; 795 mask_value = cr0_ones_mask | cr0_zeros_mask; 796 shadow_ident = VMCS_CR0_SHADOW; 797 } else { 798 mask_ident = VMCS_CR4_MASK; 799 mask_value = cr4_ones_mask | cr4_zeros_mask; 800 shadow_ident = VMCS_CR4_SHADOW; 801 } 802 803 error = vmcs_setreg(vmcs, 0, VMCS_IDENT(mask_ident), mask_value); 804 if (error) 805 return (error); 806 807 error = vmcs_setreg(vmcs, 0, VMCS_IDENT(shadow_ident), initial); 808 if (error) 809 return (error); 810 811 return (0); 812 } 813 #define vmx_setup_cr0_shadow(vmcs,init) vmx_setup_cr_shadow(0, (vmcs), (init)) 814 #define vmx_setup_cr4_shadow(vmcs,init) vmx_setup_cr_shadow(4, (vmcs), (init)) 815 816 static void * 817 vmx_vminit(struct vm *vm, pmap_t pmap) 818 { 819 uint16_t vpid[VM_MAXCPU]; 820 int i, error; 821 struct vmx *vmx; 822 struct vmcs *vmcs; 823 uint32_t exc_bitmap; 824 825 vmx = malloc(sizeof(struct vmx), M_VMX, M_WAITOK | M_ZERO); 826 if ((uintptr_t)vmx & PAGE_MASK) { 827 panic("malloc of struct vmx not aligned on %d byte boundary", 828 PAGE_SIZE); 829 } 830 vmx->vm = vm; 831 832 vmx->eptp = eptp(vtophys((vm_offset_t)pmap->pm_pml4)); 833 834 /* 835 * Clean up EPTP-tagged guest physical and combined mappings 836 * 837 * VMX transitions are not required to invalidate any guest physical 838 * mappings. So, it may be possible for stale guest physical mappings 839 * to be present in the processor TLBs. 840 * 841 * Combined mappings for this EP4TA are also invalidated for all VPIDs. 842 */ 843 ept_invalidate_mappings(vmx->eptp); 844 845 msr_bitmap_initialize(vmx->msr_bitmap); 846 847 /* 848 * It is safe to allow direct access to MSR_GSBASE and MSR_FSBASE. 849 * The guest FSBASE and GSBASE are saved and restored during 850 * vm-exit and vm-entry respectively. The host FSBASE and GSBASE are 851 * always restored from the vmcs host state area on vm-exit. 852 * 853 * The SYSENTER_CS/ESP/EIP MSRs are identical to FS/GSBASE in 854 * how they are saved/restored so can be directly accessed by the 855 * guest. 856 * 857 * MSR_EFER is saved and restored in the guest VMCS area on a 858 * VM exit and entry respectively. It is also restored from the 859 * host VMCS area on a VM exit. 860 * 861 * The TSC MSR is exposed read-only. Writes are disallowed as 862 * that will impact the host TSC. If the guest does a write 863 * the "use TSC offsetting" execution control is enabled and the 864 * difference between the host TSC and the guest TSC is written 865 * into the TSC offset in the VMCS. 866 */ 867 if (guest_msr_rw(vmx, MSR_GSBASE) || 868 guest_msr_rw(vmx, MSR_FSBASE) || 869 guest_msr_rw(vmx, MSR_SYSENTER_CS_MSR) || 870 guest_msr_rw(vmx, MSR_SYSENTER_ESP_MSR) || 871 guest_msr_rw(vmx, MSR_SYSENTER_EIP_MSR) || 872 guest_msr_rw(vmx, MSR_EFER) || 873 guest_msr_ro(vmx, MSR_TSC)) 874 panic("vmx_vminit: error setting guest msr access"); 875 876 vpid_alloc(vpid, VM_MAXCPU); 877 878 if (virtual_interrupt_delivery) { 879 error = vm_map_mmio(vm, DEFAULT_APIC_BASE, PAGE_SIZE, 880 APIC_ACCESS_ADDRESS); 881 /* XXX this should really return an error to the caller */ 882 KASSERT(error == 0, ("vm_map_mmio(apicbase) error %d", error)); 883 } 884 885 for (i = 0; i < VM_MAXCPU; i++) { 886 vmcs = &vmx->vmcs[i]; 887 vmcs->identifier = vmx_revision(); 888 error = vmclear(vmcs); 889 if (error != 0) { 890 panic("vmx_vminit: vmclear error %d on vcpu %d\n", 891 error, i); 892 } 893 894 vmx_msr_guest_init(vmx, i); 895 896 error = vmcs_init(vmcs); 897 KASSERT(error == 0, ("vmcs_init error %d", error)); 898 899 VMPTRLD(vmcs); 900 error = 0; 901 error += vmwrite(VMCS_HOST_RSP, (u_long)&vmx->ctx[i]); 902 error += vmwrite(VMCS_EPTP, vmx->eptp); 903 error += vmwrite(VMCS_PIN_BASED_CTLS, pinbased_ctls); 904 error += vmwrite(VMCS_PRI_PROC_BASED_CTLS, procbased_ctls); 905 error += vmwrite(VMCS_SEC_PROC_BASED_CTLS, procbased_ctls2); 906 error += vmwrite(VMCS_EXIT_CTLS, exit_ctls); 907 error += vmwrite(VMCS_ENTRY_CTLS, entry_ctls); 908 error += vmwrite(VMCS_MSR_BITMAP, vtophys(vmx->msr_bitmap)); 909 error += vmwrite(VMCS_VPID, vpid[i]); 910 911 /* exception bitmap */ 912 if (vcpu_trace_exceptions(vm, i)) 913 exc_bitmap = 0xffffffff; 914 else 915 exc_bitmap = 1 << IDT_MC; 916 error += vmwrite(VMCS_EXCEPTION_BITMAP, exc_bitmap); 917 918 if (virtual_interrupt_delivery) { 919 error += vmwrite(VMCS_APIC_ACCESS, APIC_ACCESS_ADDRESS); 920 error += vmwrite(VMCS_VIRTUAL_APIC, 921 vtophys(&vmx->apic_page[i])); 922 error += vmwrite(VMCS_EOI_EXIT0, 0); 923 error += vmwrite(VMCS_EOI_EXIT1, 0); 924 error += vmwrite(VMCS_EOI_EXIT2, 0); 925 error += vmwrite(VMCS_EOI_EXIT3, 0); 926 } 927 if (posted_interrupts) { 928 error += vmwrite(VMCS_PIR_VECTOR, pirvec); 929 error += vmwrite(VMCS_PIR_DESC, 930 vtophys(&vmx->pir_desc[i])); 931 } 932 VMCLEAR(vmcs); 933 KASSERT(error == 0, ("vmx_vminit: error customizing the vmcs")); 934 935 vmx->cap[i].set = 0; 936 vmx->cap[i].proc_ctls = procbased_ctls; 937 vmx->cap[i].proc_ctls2 = procbased_ctls2; 938 939 vmx->state[i].nextrip = ~0; 940 vmx->state[i].lastcpu = NOCPU; 941 vmx->state[i].vpid = vpid[i]; 942 943 /* 944 * Set up the CR0/4 shadows, and init the read shadow 945 * to the power-on register value from the Intel Sys Arch. 946 * CR0 - 0x60000010 947 * CR4 - 0 948 */ 949 error = vmx_setup_cr0_shadow(vmcs, 0x60000010); 950 if (error != 0) 951 panic("vmx_setup_cr0_shadow %d", error); 952 953 error = vmx_setup_cr4_shadow(vmcs, 0); 954 if (error != 0) 955 panic("vmx_setup_cr4_shadow %d", error); 956 957 vmx->ctx[i].pmap = pmap; 958 } 959 960 return (vmx); 961 } 962 963 static int 964 vmx_handle_cpuid(struct vm *vm, int vcpu, struct vmxctx *vmxctx) 965 { 966 int handled, func; 967 968 func = vmxctx->guest_rax; 969 970 handled = x86_emulate_cpuid(vm, vcpu, 971 (uint32_t*)(&vmxctx->guest_rax), 972 (uint32_t*)(&vmxctx->guest_rbx), 973 (uint32_t*)(&vmxctx->guest_rcx), 974 (uint32_t*)(&vmxctx->guest_rdx)); 975 return (handled); 976 } 977 978 static __inline void 979 vmx_run_trace(struct vmx *vmx, int vcpu) 980 { 981 #ifdef KTR 982 VCPU_CTR1(vmx->vm, vcpu, "Resume execution at %#lx", vmcs_guest_rip()); 983 #endif 984 } 985 986 static __inline void 987 vmx_exit_trace(struct vmx *vmx, int vcpu, uint64_t rip, uint32_t exit_reason, 988 int handled) 989 { 990 #ifdef KTR 991 VCPU_CTR3(vmx->vm, vcpu, "%s %s vmexit at 0x%0lx", 992 handled ? "handled" : "unhandled", 993 exit_reason_to_str(exit_reason), rip); 994 #endif 995 } 996 997 static __inline void 998 vmx_astpending_trace(struct vmx *vmx, int vcpu, uint64_t rip) 999 { 1000 #ifdef KTR 1001 VCPU_CTR1(vmx->vm, vcpu, "astpending vmexit at 0x%0lx", rip); 1002 #endif 1003 } 1004 1005 static VMM_STAT_INTEL(VCPU_INVVPID_SAVED, "Number of vpid invalidations saved"); 1006 static VMM_STAT_INTEL(VCPU_INVVPID_DONE, "Number of vpid invalidations done"); 1007 1008 /* 1009 * Invalidate guest mappings identified by its vpid from the TLB. 1010 */ 1011 static __inline void 1012 vmx_invvpid(struct vmx *vmx, int vcpu, pmap_t pmap, int running) 1013 { 1014 struct vmxstate *vmxstate; 1015 struct invvpid_desc invvpid_desc; 1016 1017 vmxstate = &vmx->state[vcpu]; 1018 if (vmxstate->vpid == 0) 1019 return; 1020 1021 if (!running) { 1022 /* 1023 * Set the 'lastcpu' to an invalid host cpu. 1024 * 1025 * This will invalidate TLB entries tagged with the vcpu's 1026 * vpid the next time it runs via vmx_set_pcpu_defaults(). 1027 */ 1028 vmxstate->lastcpu = NOCPU; 1029 return; 1030 } 1031 1032 KASSERT(curthread->td_critnest > 0, ("%s: vcpu %d running outside " 1033 "critical section", __func__, vcpu)); 1034 1035 /* 1036 * Invalidate all mappings tagged with 'vpid' 1037 * 1038 * We do this because this vcpu was executing on a different host 1039 * cpu when it last ran. We do not track whether it invalidated 1040 * mappings associated with its 'vpid' during that run. So we must 1041 * assume that the mappings associated with 'vpid' on 'curcpu' are 1042 * stale and invalidate them. 1043 * 1044 * Note that we incur this penalty only when the scheduler chooses to 1045 * move the thread associated with this vcpu between host cpus. 1046 * 1047 * Note also that this will invalidate mappings tagged with 'vpid' 1048 * for "all" EP4TAs. 1049 */ 1050 if (pmap->pm_eptgen == vmx->eptgen[curcpu]) { 1051 invvpid_desc._res1 = 0; 1052 invvpid_desc._res2 = 0; 1053 invvpid_desc.vpid = vmxstate->vpid; 1054 invvpid_desc.linear_addr = 0; 1055 invvpid(INVVPID_TYPE_SINGLE_CONTEXT, invvpid_desc); 1056 vmm_stat_incr(vmx->vm, vcpu, VCPU_INVVPID_DONE, 1); 1057 } else { 1058 /* 1059 * The invvpid can be skipped if an invept is going to 1060 * be performed before entering the guest. The invept 1061 * will invalidate combined mappings tagged with 1062 * 'vmx->eptp' for all vpids. 1063 */ 1064 vmm_stat_incr(vmx->vm, vcpu, VCPU_INVVPID_SAVED, 1); 1065 } 1066 } 1067 1068 static void 1069 vmx_set_pcpu_defaults(struct vmx *vmx, int vcpu, pmap_t pmap) 1070 { 1071 struct vmxstate *vmxstate; 1072 1073 vmxstate = &vmx->state[vcpu]; 1074 if (vmxstate->lastcpu == curcpu) 1075 return; 1076 1077 vmxstate->lastcpu = curcpu; 1078 1079 vmm_stat_incr(vmx->vm, vcpu, VCPU_MIGRATIONS, 1); 1080 1081 vmcs_write(VMCS_HOST_TR_BASE, vmm_get_host_trbase()); 1082 vmcs_write(VMCS_HOST_GDTR_BASE, vmm_get_host_gdtrbase()); 1083 vmcs_write(VMCS_HOST_GS_BASE, vmm_get_host_gsbase()); 1084 vmx_invvpid(vmx, vcpu, pmap, 1); 1085 } 1086 1087 /* 1088 * We depend on 'procbased_ctls' to have the Interrupt Window Exiting bit set. 1089 */ 1090 CTASSERT((PROCBASED_CTLS_ONE_SETTING & PROCBASED_INT_WINDOW_EXITING) != 0); 1091 1092 static void __inline 1093 vmx_set_int_window_exiting(struct vmx *vmx, int vcpu) 1094 { 1095 1096 if ((vmx->cap[vcpu].proc_ctls & PROCBASED_INT_WINDOW_EXITING) == 0) { 1097 vmx->cap[vcpu].proc_ctls |= PROCBASED_INT_WINDOW_EXITING; 1098 vmcs_write(VMCS_PRI_PROC_BASED_CTLS, vmx->cap[vcpu].proc_ctls); 1099 VCPU_CTR0(vmx->vm, vcpu, "Enabling interrupt window exiting"); 1100 } 1101 } 1102 1103 static void __inline 1104 vmx_clear_int_window_exiting(struct vmx *vmx, int vcpu) 1105 { 1106 1107 KASSERT((vmx->cap[vcpu].proc_ctls & PROCBASED_INT_WINDOW_EXITING) != 0, 1108 ("intr_window_exiting not set: %#x", vmx->cap[vcpu].proc_ctls)); 1109 vmx->cap[vcpu].proc_ctls &= ~PROCBASED_INT_WINDOW_EXITING; 1110 vmcs_write(VMCS_PRI_PROC_BASED_CTLS, vmx->cap[vcpu].proc_ctls); 1111 VCPU_CTR0(vmx->vm, vcpu, "Disabling interrupt window exiting"); 1112 } 1113 1114 static void __inline 1115 vmx_set_nmi_window_exiting(struct vmx *vmx, int vcpu) 1116 { 1117 1118 if ((vmx->cap[vcpu].proc_ctls & PROCBASED_NMI_WINDOW_EXITING) == 0) { 1119 vmx->cap[vcpu].proc_ctls |= PROCBASED_NMI_WINDOW_EXITING; 1120 vmcs_write(VMCS_PRI_PROC_BASED_CTLS, vmx->cap[vcpu].proc_ctls); 1121 VCPU_CTR0(vmx->vm, vcpu, "Enabling NMI window exiting"); 1122 } 1123 } 1124 1125 static void __inline 1126 vmx_clear_nmi_window_exiting(struct vmx *vmx, int vcpu) 1127 { 1128 1129 KASSERT((vmx->cap[vcpu].proc_ctls & PROCBASED_NMI_WINDOW_EXITING) != 0, 1130 ("nmi_window_exiting not set %#x", vmx->cap[vcpu].proc_ctls)); 1131 vmx->cap[vcpu].proc_ctls &= ~PROCBASED_NMI_WINDOW_EXITING; 1132 vmcs_write(VMCS_PRI_PROC_BASED_CTLS, vmx->cap[vcpu].proc_ctls); 1133 VCPU_CTR0(vmx->vm, vcpu, "Disabling NMI window exiting"); 1134 } 1135 1136 int 1137 vmx_set_tsc_offset(struct vmx *vmx, int vcpu, uint64_t offset) 1138 { 1139 int error; 1140 1141 if ((vmx->cap[vcpu].proc_ctls & PROCBASED_TSC_OFFSET) == 0) { 1142 vmx->cap[vcpu].proc_ctls |= PROCBASED_TSC_OFFSET; 1143 vmcs_write(VMCS_PRI_PROC_BASED_CTLS, vmx->cap[vcpu].proc_ctls); 1144 VCPU_CTR0(vmx->vm, vcpu, "Enabling TSC offsetting"); 1145 } 1146 1147 error = vmwrite(VMCS_TSC_OFFSET, offset); 1148 1149 return (error); 1150 } 1151 1152 #define NMI_BLOCKING (VMCS_INTERRUPTIBILITY_NMI_BLOCKING | \ 1153 VMCS_INTERRUPTIBILITY_MOVSS_BLOCKING) 1154 #define HWINTR_BLOCKING (VMCS_INTERRUPTIBILITY_STI_BLOCKING | \ 1155 VMCS_INTERRUPTIBILITY_MOVSS_BLOCKING) 1156 1157 static void 1158 vmx_inject_nmi(struct vmx *vmx, int vcpu) 1159 { 1160 uint32_t gi, info; 1161 1162 gi = vmcs_read(VMCS_GUEST_INTERRUPTIBILITY); 1163 KASSERT((gi & NMI_BLOCKING) == 0, ("vmx_inject_nmi: invalid guest " 1164 "interruptibility-state %#x", gi)); 1165 1166 info = vmcs_read(VMCS_ENTRY_INTR_INFO); 1167 KASSERT((info & VMCS_INTR_VALID) == 0, ("vmx_inject_nmi: invalid " 1168 "VM-entry interruption information %#x", info)); 1169 1170 /* 1171 * Inject the virtual NMI. The vector must be the NMI IDT entry 1172 * or the VMCS entry check will fail. 1173 */ 1174 info = IDT_NMI | VMCS_INTR_T_NMI | VMCS_INTR_VALID; 1175 vmcs_write(VMCS_ENTRY_INTR_INFO, info); 1176 1177 VCPU_CTR0(vmx->vm, vcpu, "Injecting vNMI"); 1178 1179 /* Clear the request */ 1180 vm_nmi_clear(vmx->vm, vcpu); 1181 } 1182 1183 static void 1184 vmx_inject_interrupts(struct vmx *vmx, int vcpu, struct vlapic *vlapic, 1185 uint64_t guestrip) 1186 { 1187 int vector, need_nmi_exiting, extint_pending; 1188 uint64_t rflags, entryinfo; 1189 uint32_t gi, info; 1190 1191 if (vmx->state[vcpu].nextrip != guestrip) { 1192 gi = vmcs_read(VMCS_GUEST_INTERRUPTIBILITY); 1193 if (gi & HWINTR_BLOCKING) { 1194 VCPU_CTR2(vmx->vm, vcpu, "Guest interrupt blocking " 1195 "cleared due to rip change: %#lx/%#lx", 1196 vmx->state[vcpu].nextrip, guestrip); 1197 gi &= ~HWINTR_BLOCKING; 1198 vmcs_write(VMCS_GUEST_INTERRUPTIBILITY, gi); 1199 } 1200 } 1201 1202 if (vm_entry_intinfo(vmx->vm, vcpu, &entryinfo)) { 1203 KASSERT((entryinfo & VMCS_INTR_VALID) != 0, ("%s: entry " 1204 "intinfo is not valid: %#lx", __func__, entryinfo)); 1205 1206 info = vmcs_read(VMCS_ENTRY_INTR_INFO); 1207 KASSERT((info & VMCS_INTR_VALID) == 0, ("%s: cannot inject " 1208 "pending exception: %#lx/%#x", __func__, entryinfo, info)); 1209 1210 info = entryinfo; 1211 vector = info & 0xff; 1212 if (vector == IDT_BP || vector == IDT_OF) { 1213 /* 1214 * VT-x requires #BP and #OF to be injected as software 1215 * exceptions. 1216 */ 1217 info &= ~VMCS_INTR_T_MASK; 1218 info |= VMCS_INTR_T_SWEXCEPTION; 1219 } 1220 1221 if (info & VMCS_INTR_DEL_ERRCODE) 1222 vmcs_write(VMCS_ENTRY_EXCEPTION_ERROR, entryinfo >> 32); 1223 1224 vmcs_write(VMCS_ENTRY_INTR_INFO, info); 1225 } 1226 1227 if (vm_nmi_pending(vmx->vm, vcpu)) { 1228 /* 1229 * If there are no conditions blocking NMI injection then 1230 * inject it directly here otherwise enable "NMI window 1231 * exiting" to inject it as soon as we can. 1232 * 1233 * We also check for STI_BLOCKING because some implementations 1234 * don't allow NMI injection in this case. If we are running 1235 * on a processor that doesn't have this restriction it will 1236 * immediately exit and the NMI will be injected in the 1237 * "NMI window exiting" handler. 1238 */ 1239 need_nmi_exiting = 1; 1240 gi = vmcs_read(VMCS_GUEST_INTERRUPTIBILITY); 1241 if ((gi & (HWINTR_BLOCKING | NMI_BLOCKING)) == 0) { 1242 info = vmcs_read(VMCS_ENTRY_INTR_INFO); 1243 if ((info & VMCS_INTR_VALID) == 0) { 1244 vmx_inject_nmi(vmx, vcpu); 1245 need_nmi_exiting = 0; 1246 } else { 1247 VCPU_CTR1(vmx->vm, vcpu, "Cannot inject NMI " 1248 "due to VM-entry intr info %#x", info); 1249 } 1250 } else { 1251 VCPU_CTR1(vmx->vm, vcpu, "Cannot inject NMI due to " 1252 "Guest Interruptibility-state %#x", gi); 1253 } 1254 1255 if (need_nmi_exiting) 1256 vmx_set_nmi_window_exiting(vmx, vcpu); 1257 } 1258 1259 extint_pending = vm_extint_pending(vmx->vm, vcpu); 1260 1261 if (!extint_pending && virtual_interrupt_delivery) { 1262 vmx_inject_pir(vlapic); 1263 return; 1264 } 1265 1266 /* 1267 * If interrupt-window exiting is already in effect then don't bother 1268 * checking for pending interrupts. This is just an optimization and 1269 * not needed for correctness. 1270 */ 1271 if ((vmx->cap[vcpu].proc_ctls & PROCBASED_INT_WINDOW_EXITING) != 0) { 1272 VCPU_CTR0(vmx->vm, vcpu, "Skip interrupt injection due to " 1273 "pending int_window_exiting"); 1274 return; 1275 } 1276 1277 if (!extint_pending) { 1278 /* Ask the local apic for a vector to inject */ 1279 if (!vlapic_pending_intr(vlapic, &vector)) 1280 return; 1281 1282 /* 1283 * From the Intel SDM, Volume 3, Section "Maskable 1284 * Hardware Interrupts": 1285 * - maskable interrupt vectors [16,255] can be delivered 1286 * through the local APIC. 1287 */ 1288 KASSERT(vector >= 16 && vector <= 255, 1289 ("invalid vector %d from local APIC", vector)); 1290 } else { 1291 /* Ask the legacy pic for a vector to inject */ 1292 vatpic_pending_intr(vmx->vm, &vector); 1293 1294 /* 1295 * From the Intel SDM, Volume 3, Section "Maskable 1296 * Hardware Interrupts": 1297 * - maskable interrupt vectors [0,255] can be delivered 1298 * through the INTR pin. 1299 */ 1300 KASSERT(vector >= 0 && vector <= 255, 1301 ("invalid vector %d from INTR", vector)); 1302 } 1303 1304 /* Check RFLAGS.IF and the interruptibility state of the guest */ 1305 rflags = vmcs_read(VMCS_GUEST_RFLAGS); 1306 if ((rflags & PSL_I) == 0) { 1307 VCPU_CTR2(vmx->vm, vcpu, "Cannot inject vector %d due to " 1308 "rflags %#lx", vector, rflags); 1309 goto cantinject; 1310 } 1311 1312 gi = vmcs_read(VMCS_GUEST_INTERRUPTIBILITY); 1313 if (gi & HWINTR_BLOCKING) { 1314 VCPU_CTR2(vmx->vm, vcpu, "Cannot inject vector %d due to " 1315 "Guest Interruptibility-state %#x", vector, gi); 1316 goto cantinject; 1317 } 1318 1319 info = vmcs_read(VMCS_ENTRY_INTR_INFO); 1320 if (info & VMCS_INTR_VALID) { 1321 /* 1322 * This is expected and could happen for multiple reasons: 1323 * - A vectoring VM-entry was aborted due to astpending 1324 * - A VM-exit happened during event injection. 1325 * - An exception was injected above. 1326 * - An NMI was injected above or after "NMI window exiting" 1327 */ 1328 VCPU_CTR2(vmx->vm, vcpu, "Cannot inject vector %d due to " 1329 "VM-entry intr info %#x", vector, info); 1330 goto cantinject; 1331 } 1332 1333 /* Inject the interrupt */ 1334 info = VMCS_INTR_T_HWINTR | VMCS_INTR_VALID; 1335 info |= vector; 1336 vmcs_write(VMCS_ENTRY_INTR_INFO, info); 1337 1338 if (!extint_pending) { 1339 /* Update the Local APIC ISR */ 1340 vlapic_intr_accepted(vlapic, vector); 1341 } else { 1342 vm_extint_clear(vmx->vm, vcpu); 1343 vatpic_intr_accepted(vmx->vm, vector); 1344 1345 /* 1346 * After we accepted the current ExtINT the PIC may 1347 * have posted another one. If that is the case, set 1348 * the Interrupt Window Exiting execution control so 1349 * we can inject that one too. 1350 * 1351 * Also, interrupt window exiting allows us to inject any 1352 * pending APIC vector that was preempted by the ExtINT 1353 * as soon as possible. This applies both for the software 1354 * emulated vlapic and the hardware assisted virtual APIC. 1355 */ 1356 vmx_set_int_window_exiting(vmx, vcpu); 1357 } 1358 1359 VCPU_CTR1(vmx->vm, vcpu, "Injecting hwintr at vector %d", vector); 1360 1361 return; 1362 1363 cantinject: 1364 /* 1365 * Set the Interrupt Window Exiting execution control so we can inject 1366 * the interrupt as soon as blocking condition goes away. 1367 */ 1368 vmx_set_int_window_exiting(vmx, vcpu); 1369 } 1370 1371 /* 1372 * If the Virtual NMIs execution control is '1' then the logical processor 1373 * tracks virtual-NMI blocking in the Guest Interruptibility-state field of 1374 * the VMCS. An IRET instruction in VMX non-root operation will remove any 1375 * virtual-NMI blocking. 1376 * 1377 * This unblocking occurs even if the IRET causes a fault. In this case the 1378 * hypervisor needs to restore virtual-NMI blocking before resuming the guest. 1379 */ 1380 static void 1381 vmx_restore_nmi_blocking(struct vmx *vmx, int vcpuid) 1382 { 1383 uint32_t gi; 1384 1385 VCPU_CTR0(vmx->vm, vcpuid, "Restore Virtual-NMI blocking"); 1386 gi = vmcs_read(VMCS_GUEST_INTERRUPTIBILITY); 1387 gi |= VMCS_INTERRUPTIBILITY_NMI_BLOCKING; 1388 vmcs_write(VMCS_GUEST_INTERRUPTIBILITY, gi); 1389 } 1390 1391 static void 1392 vmx_clear_nmi_blocking(struct vmx *vmx, int vcpuid) 1393 { 1394 uint32_t gi; 1395 1396 VCPU_CTR0(vmx->vm, vcpuid, "Clear Virtual-NMI blocking"); 1397 gi = vmcs_read(VMCS_GUEST_INTERRUPTIBILITY); 1398 gi &= ~VMCS_INTERRUPTIBILITY_NMI_BLOCKING; 1399 vmcs_write(VMCS_GUEST_INTERRUPTIBILITY, gi); 1400 } 1401 1402 static void 1403 vmx_assert_nmi_blocking(struct vmx *vmx, int vcpuid) 1404 { 1405 uint32_t gi; 1406 1407 gi = vmcs_read(VMCS_GUEST_INTERRUPTIBILITY); 1408 KASSERT(gi & VMCS_INTERRUPTIBILITY_NMI_BLOCKING, 1409 ("NMI blocking is not in effect %#x", gi)); 1410 } 1411 1412 static int 1413 vmx_emulate_xsetbv(struct vmx *vmx, int vcpu, struct vm_exit *vmexit) 1414 { 1415 struct vmxctx *vmxctx; 1416 uint64_t xcrval; 1417 const struct xsave_limits *limits; 1418 1419 vmxctx = &vmx->ctx[vcpu]; 1420 limits = vmm_get_xsave_limits(); 1421 1422 /* 1423 * Note that the processor raises a GP# fault on its own if 1424 * xsetbv is executed for CPL != 0, so we do not have to 1425 * emulate that fault here. 1426 */ 1427 1428 /* Only xcr0 is supported. */ 1429 if (vmxctx->guest_rcx != 0) { 1430 vm_inject_gp(vmx->vm, vcpu); 1431 return (HANDLED); 1432 } 1433 1434 /* We only handle xcr0 if both the host and guest have XSAVE enabled. */ 1435 if (!limits->xsave_enabled || !(vmcs_read(VMCS_GUEST_CR4) & CR4_XSAVE)) { 1436 vm_inject_ud(vmx->vm, vcpu); 1437 return (HANDLED); 1438 } 1439 1440 xcrval = vmxctx->guest_rdx << 32 | (vmxctx->guest_rax & 0xffffffff); 1441 if ((xcrval & ~limits->xcr0_allowed) != 0) { 1442 vm_inject_gp(vmx->vm, vcpu); 1443 return (HANDLED); 1444 } 1445 1446 if (!(xcrval & XFEATURE_ENABLED_X87)) { 1447 vm_inject_gp(vmx->vm, vcpu); 1448 return (HANDLED); 1449 } 1450 1451 /* AVX (YMM_Hi128) requires SSE. */ 1452 if (xcrval & XFEATURE_ENABLED_AVX && 1453 (xcrval & XFEATURE_AVX) != XFEATURE_AVX) { 1454 vm_inject_gp(vmx->vm, vcpu); 1455 return (HANDLED); 1456 } 1457 1458 /* 1459 * AVX512 requires base AVX (YMM_Hi128) as well as OpMask, 1460 * ZMM_Hi256, and Hi16_ZMM. 1461 */ 1462 if (xcrval & XFEATURE_AVX512 && 1463 (xcrval & (XFEATURE_AVX512 | XFEATURE_AVX)) != 1464 (XFEATURE_AVX512 | XFEATURE_AVX)) { 1465 vm_inject_gp(vmx->vm, vcpu); 1466 return (HANDLED); 1467 } 1468 1469 /* 1470 * Intel MPX requires both bound register state flags to be 1471 * set. 1472 */ 1473 if (((xcrval & XFEATURE_ENABLED_BNDREGS) != 0) != 1474 ((xcrval & XFEATURE_ENABLED_BNDCSR) != 0)) { 1475 vm_inject_gp(vmx->vm, vcpu); 1476 return (HANDLED); 1477 } 1478 1479 /* 1480 * This runs "inside" vmrun() with the guest's FPU state, so 1481 * modifying xcr0 directly modifies the guest's xcr0, not the 1482 * host's. 1483 */ 1484 load_xcr(0, xcrval); 1485 return (HANDLED); 1486 } 1487 1488 static uint64_t 1489 vmx_get_guest_reg(struct vmx *vmx, int vcpu, int ident) 1490 { 1491 const struct vmxctx *vmxctx; 1492 1493 vmxctx = &vmx->ctx[vcpu]; 1494 1495 switch (ident) { 1496 case 0: 1497 return (vmxctx->guest_rax); 1498 case 1: 1499 return (vmxctx->guest_rcx); 1500 case 2: 1501 return (vmxctx->guest_rdx); 1502 case 3: 1503 return (vmxctx->guest_rbx); 1504 case 4: 1505 return (vmcs_read(VMCS_GUEST_RSP)); 1506 case 5: 1507 return (vmxctx->guest_rbp); 1508 case 6: 1509 return (vmxctx->guest_rsi); 1510 case 7: 1511 return (vmxctx->guest_rdi); 1512 case 8: 1513 return (vmxctx->guest_r8); 1514 case 9: 1515 return (vmxctx->guest_r9); 1516 case 10: 1517 return (vmxctx->guest_r10); 1518 case 11: 1519 return (vmxctx->guest_r11); 1520 case 12: 1521 return (vmxctx->guest_r12); 1522 case 13: 1523 return (vmxctx->guest_r13); 1524 case 14: 1525 return (vmxctx->guest_r14); 1526 case 15: 1527 return (vmxctx->guest_r15); 1528 default: 1529 panic("invalid vmx register %d", ident); 1530 } 1531 } 1532 1533 static void 1534 vmx_set_guest_reg(struct vmx *vmx, int vcpu, int ident, uint64_t regval) 1535 { 1536 struct vmxctx *vmxctx; 1537 1538 vmxctx = &vmx->ctx[vcpu]; 1539 1540 switch (ident) { 1541 case 0: 1542 vmxctx->guest_rax = regval; 1543 break; 1544 case 1: 1545 vmxctx->guest_rcx = regval; 1546 break; 1547 case 2: 1548 vmxctx->guest_rdx = regval; 1549 break; 1550 case 3: 1551 vmxctx->guest_rbx = regval; 1552 break; 1553 case 4: 1554 vmcs_write(VMCS_GUEST_RSP, regval); 1555 break; 1556 case 5: 1557 vmxctx->guest_rbp = regval; 1558 break; 1559 case 6: 1560 vmxctx->guest_rsi = regval; 1561 break; 1562 case 7: 1563 vmxctx->guest_rdi = regval; 1564 break; 1565 case 8: 1566 vmxctx->guest_r8 = regval; 1567 break; 1568 case 9: 1569 vmxctx->guest_r9 = regval; 1570 break; 1571 case 10: 1572 vmxctx->guest_r10 = regval; 1573 break; 1574 case 11: 1575 vmxctx->guest_r11 = regval; 1576 break; 1577 case 12: 1578 vmxctx->guest_r12 = regval; 1579 break; 1580 case 13: 1581 vmxctx->guest_r13 = regval; 1582 break; 1583 case 14: 1584 vmxctx->guest_r14 = regval; 1585 break; 1586 case 15: 1587 vmxctx->guest_r15 = regval; 1588 break; 1589 default: 1590 panic("invalid vmx register %d", ident); 1591 } 1592 } 1593 1594 static int 1595 vmx_emulate_cr0_access(struct vmx *vmx, int vcpu, uint64_t exitqual) 1596 { 1597 uint64_t crval, regval; 1598 1599 /* We only handle mov to %cr0 at this time */ 1600 if ((exitqual & 0xf0) != 0x00) 1601 return (UNHANDLED); 1602 1603 regval = vmx_get_guest_reg(vmx, vcpu, (exitqual >> 8) & 0xf); 1604 1605 vmcs_write(VMCS_CR0_SHADOW, regval); 1606 1607 crval = regval | cr0_ones_mask; 1608 crval &= ~cr0_zeros_mask; 1609 vmcs_write(VMCS_GUEST_CR0, crval); 1610 1611 if (regval & CR0_PG) { 1612 uint64_t efer, entry_ctls; 1613 1614 /* 1615 * If CR0.PG is 1 and EFER.LME is 1 then EFER.LMA and 1616 * the "IA-32e mode guest" bit in VM-entry control must be 1617 * equal. 1618 */ 1619 efer = vmcs_read(VMCS_GUEST_IA32_EFER); 1620 if (efer & EFER_LME) { 1621 efer |= EFER_LMA; 1622 vmcs_write(VMCS_GUEST_IA32_EFER, efer); 1623 entry_ctls = vmcs_read(VMCS_ENTRY_CTLS); 1624 entry_ctls |= VM_ENTRY_GUEST_LMA; 1625 vmcs_write(VMCS_ENTRY_CTLS, entry_ctls); 1626 } 1627 } 1628 1629 return (HANDLED); 1630 } 1631 1632 static int 1633 vmx_emulate_cr4_access(struct vmx *vmx, int vcpu, uint64_t exitqual) 1634 { 1635 uint64_t crval, regval; 1636 1637 /* We only handle mov to %cr4 at this time */ 1638 if ((exitqual & 0xf0) != 0x00) 1639 return (UNHANDLED); 1640 1641 regval = vmx_get_guest_reg(vmx, vcpu, (exitqual >> 8) & 0xf); 1642 1643 vmcs_write(VMCS_CR4_SHADOW, regval); 1644 1645 crval = regval | cr4_ones_mask; 1646 crval &= ~cr4_zeros_mask; 1647 vmcs_write(VMCS_GUEST_CR4, crval); 1648 1649 return (HANDLED); 1650 } 1651 1652 static int 1653 vmx_emulate_cr8_access(struct vmx *vmx, int vcpu, uint64_t exitqual) 1654 { 1655 struct vlapic *vlapic; 1656 uint64_t cr8; 1657 int regnum; 1658 1659 /* We only handle mov %cr8 to/from a register at this time. */ 1660 if ((exitqual & 0xe0) != 0x00) { 1661 return (UNHANDLED); 1662 } 1663 1664 vlapic = vm_lapic(vmx->vm, vcpu); 1665 regnum = (exitqual >> 8) & 0xf; 1666 if (exitqual & 0x10) { 1667 cr8 = vlapic_get_cr8(vlapic); 1668 vmx_set_guest_reg(vmx, vcpu, regnum, cr8); 1669 } else { 1670 cr8 = vmx_get_guest_reg(vmx, vcpu, regnum); 1671 vlapic_set_cr8(vlapic, cr8); 1672 } 1673 1674 return (HANDLED); 1675 } 1676 1677 /* 1678 * From section "Guest Register State" in the Intel SDM: CPL = SS.DPL 1679 */ 1680 static int 1681 vmx_cpl(void) 1682 { 1683 uint32_t ssar; 1684 1685 ssar = vmcs_read(VMCS_GUEST_SS_ACCESS_RIGHTS); 1686 return ((ssar >> 5) & 0x3); 1687 } 1688 1689 static enum vm_cpu_mode 1690 vmx_cpu_mode(void) 1691 { 1692 uint32_t csar; 1693 1694 if (vmcs_read(VMCS_GUEST_IA32_EFER) & EFER_LMA) { 1695 csar = vmcs_read(VMCS_GUEST_CS_ACCESS_RIGHTS); 1696 if (csar & 0x2000) 1697 return (CPU_MODE_64BIT); /* CS.L = 1 */ 1698 else 1699 return (CPU_MODE_COMPATIBILITY); 1700 } else if (vmcs_read(VMCS_GUEST_CR0) & CR0_PE) { 1701 return (CPU_MODE_PROTECTED); 1702 } else { 1703 return (CPU_MODE_REAL); 1704 } 1705 } 1706 1707 static enum vm_paging_mode 1708 vmx_paging_mode(void) 1709 { 1710 1711 if (!(vmcs_read(VMCS_GUEST_CR0) & CR0_PG)) 1712 return (PAGING_MODE_FLAT); 1713 if (!(vmcs_read(VMCS_GUEST_CR4) & CR4_PAE)) 1714 return (PAGING_MODE_32); 1715 if (vmcs_read(VMCS_GUEST_IA32_EFER) & EFER_LME) 1716 return (PAGING_MODE_64); 1717 else 1718 return (PAGING_MODE_PAE); 1719 } 1720 1721 static uint64_t 1722 inout_str_index(struct vmx *vmx, int vcpuid, int in) 1723 { 1724 uint64_t val; 1725 int error; 1726 enum vm_reg_name reg; 1727 1728 reg = in ? VM_REG_GUEST_RDI : VM_REG_GUEST_RSI; 1729 error = vmx_getreg(vmx, vcpuid, reg, &val); 1730 KASSERT(error == 0, ("%s: vmx_getreg error %d", __func__, error)); 1731 return (val); 1732 } 1733 1734 static uint64_t 1735 inout_str_count(struct vmx *vmx, int vcpuid, int rep) 1736 { 1737 uint64_t val; 1738 int error; 1739 1740 if (rep) { 1741 error = vmx_getreg(vmx, vcpuid, VM_REG_GUEST_RCX, &val); 1742 KASSERT(!error, ("%s: vmx_getreg error %d", __func__, error)); 1743 } else { 1744 val = 1; 1745 } 1746 return (val); 1747 } 1748 1749 static int 1750 inout_str_addrsize(uint32_t inst_info) 1751 { 1752 uint32_t size; 1753 1754 size = (inst_info >> 7) & 0x7; 1755 switch (size) { 1756 case 0: 1757 return (2); /* 16 bit */ 1758 case 1: 1759 return (4); /* 32 bit */ 1760 case 2: 1761 return (8); /* 64 bit */ 1762 default: 1763 panic("%s: invalid size encoding %d", __func__, size); 1764 } 1765 } 1766 1767 static void 1768 inout_str_seginfo(struct vmx *vmx, int vcpuid, uint32_t inst_info, int in, 1769 struct vm_inout_str *vis) 1770 { 1771 int error, s; 1772 1773 if (in) { 1774 vis->seg_name = VM_REG_GUEST_ES; 1775 } else { 1776 s = (inst_info >> 15) & 0x7; 1777 vis->seg_name = vm_segment_name(s); 1778 } 1779 1780 error = vmx_getdesc(vmx, vcpuid, vis->seg_name, &vis->seg_desc); 1781 KASSERT(error == 0, ("%s: vmx_getdesc error %d", __func__, error)); 1782 } 1783 1784 static void 1785 vmx_paging_info(struct vm_guest_paging *paging) 1786 { 1787 paging->cr3 = vmcs_guest_cr3(); 1788 paging->cpl = vmx_cpl(); 1789 paging->cpu_mode = vmx_cpu_mode(); 1790 paging->paging_mode = vmx_paging_mode(); 1791 } 1792 1793 static void 1794 vmexit_inst_emul(struct vm_exit *vmexit, uint64_t gpa, uint64_t gla) 1795 { 1796 struct vm_guest_paging *paging; 1797 uint32_t csar; 1798 1799 paging = &vmexit->u.inst_emul.paging; 1800 1801 vmexit->exitcode = VM_EXITCODE_INST_EMUL; 1802 vmexit->inst_length = 0; 1803 vmexit->u.inst_emul.gpa = gpa; 1804 vmexit->u.inst_emul.gla = gla; 1805 vmx_paging_info(paging); 1806 switch (paging->cpu_mode) { 1807 case CPU_MODE_REAL: 1808 vmexit->u.inst_emul.cs_base = vmcs_read(VMCS_GUEST_CS_BASE); 1809 vmexit->u.inst_emul.cs_d = 0; 1810 break; 1811 case CPU_MODE_PROTECTED: 1812 case CPU_MODE_COMPATIBILITY: 1813 vmexit->u.inst_emul.cs_base = vmcs_read(VMCS_GUEST_CS_BASE); 1814 csar = vmcs_read(VMCS_GUEST_CS_ACCESS_RIGHTS); 1815 vmexit->u.inst_emul.cs_d = SEG_DESC_DEF32(csar); 1816 break; 1817 default: 1818 vmexit->u.inst_emul.cs_base = 0; 1819 vmexit->u.inst_emul.cs_d = 0; 1820 break; 1821 } 1822 vie_init(&vmexit->u.inst_emul.vie, NULL, 0); 1823 } 1824 1825 static int 1826 ept_fault_type(uint64_t ept_qual) 1827 { 1828 int fault_type; 1829 1830 if (ept_qual & EPT_VIOLATION_DATA_WRITE) 1831 fault_type = VM_PROT_WRITE; 1832 else if (ept_qual & EPT_VIOLATION_INST_FETCH) 1833 fault_type = VM_PROT_EXECUTE; 1834 else 1835 fault_type= VM_PROT_READ; 1836 1837 return (fault_type); 1838 } 1839 1840 static boolean_t 1841 ept_emulation_fault(uint64_t ept_qual) 1842 { 1843 int read, write; 1844 1845 /* EPT fault on an instruction fetch doesn't make sense here */ 1846 if (ept_qual & EPT_VIOLATION_INST_FETCH) 1847 return (FALSE); 1848 1849 /* EPT fault must be a read fault or a write fault */ 1850 read = ept_qual & EPT_VIOLATION_DATA_READ ? 1 : 0; 1851 write = ept_qual & EPT_VIOLATION_DATA_WRITE ? 1 : 0; 1852 if ((read | write) == 0) 1853 return (FALSE); 1854 1855 /* 1856 * The EPT violation must have been caused by accessing a 1857 * guest-physical address that is a translation of a guest-linear 1858 * address. 1859 */ 1860 if ((ept_qual & EPT_VIOLATION_GLA_VALID) == 0 || 1861 (ept_qual & EPT_VIOLATION_XLAT_VALID) == 0) { 1862 return (FALSE); 1863 } 1864 1865 return (TRUE); 1866 } 1867 1868 static __inline int 1869 apic_access_virtualization(struct vmx *vmx, int vcpuid) 1870 { 1871 uint32_t proc_ctls2; 1872 1873 proc_ctls2 = vmx->cap[vcpuid].proc_ctls2; 1874 return ((proc_ctls2 & PROCBASED2_VIRTUALIZE_APIC_ACCESSES) ? 1 : 0); 1875 } 1876 1877 static __inline int 1878 x2apic_virtualization(struct vmx *vmx, int vcpuid) 1879 { 1880 uint32_t proc_ctls2; 1881 1882 proc_ctls2 = vmx->cap[vcpuid].proc_ctls2; 1883 return ((proc_ctls2 & PROCBASED2_VIRTUALIZE_X2APIC_MODE) ? 1 : 0); 1884 } 1885 1886 static int 1887 vmx_handle_apic_write(struct vmx *vmx, int vcpuid, struct vlapic *vlapic, 1888 uint64_t qual) 1889 { 1890 int error, handled, offset; 1891 uint32_t *apic_regs, vector; 1892 bool retu; 1893 1894 handled = HANDLED; 1895 offset = APIC_WRITE_OFFSET(qual); 1896 1897 if (!apic_access_virtualization(vmx, vcpuid)) { 1898 /* 1899 * In general there should not be any APIC write VM-exits 1900 * unless APIC-access virtualization is enabled. 1901 * 1902 * However self-IPI virtualization can legitimately trigger 1903 * an APIC-write VM-exit so treat it specially. 1904 */ 1905 if (x2apic_virtualization(vmx, vcpuid) && 1906 offset == APIC_OFFSET_SELF_IPI) { 1907 apic_regs = (uint32_t *)(vlapic->apic_page); 1908 vector = apic_regs[APIC_OFFSET_SELF_IPI / 4]; 1909 vlapic_self_ipi_handler(vlapic, vector); 1910 return (HANDLED); 1911 } else 1912 return (UNHANDLED); 1913 } 1914 1915 switch (offset) { 1916 case APIC_OFFSET_ID: 1917 vlapic_id_write_handler(vlapic); 1918 break; 1919 case APIC_OFFSET_LDR: 1920 vlapic_ldr_write_handler(vlapic); 1921 break; 1922 case APIC_OFFSET_DFR: 1923 vlapic_dfr_write_handler(vlapic); 1924 break; 1925 case APIC_OFFSET_SVR: 1926 vlapic_svr_write_handler(vlapic); 1927 break; 1928 case APIC_OFFSET_ESR: 1929 vlapic_esr_write_handler(vlapic); 1930 break; 1931 case APIC_OFFSET_ICR_LOW: 1932 retu = false; 1933 error = vlapic_icrlo_write_handler(vlapic, &retu); 1934 if (error != 0 || retu) 1935 handled = UNHANDLED; 1936 break; 1937 case APIC_OFFSET_CMCI_LVT: 1938 case APIC_OFFSET_TIMER_LVT ... APIC_OFFSET_ERROR_LVT: 1939 vlapic_lvt_write_handler(vlapic, offset); 1940 break; 1941 case APIC_OFFSET_TIMER_ICR: 1942 vlapic_icrtmr_write_handler(vlapic); 1943 break; 1944 case APIC_OFFSET_TIMER_DCR: 1945 vlapic_dcr_write_handler(vlapic); 1946 break; 1947 default: 1948 handled = UNHANDLED; 1949 break; 1950 } 1951 return (handled); 1952 } 1953 1954 static bool 1955 apic_access_fault(struct vmx *vmx, int vcpuid, uint64_t gpa) 1956 { 1957 1958 if (apic_access_virtualization(vmx, vcpuid) && 1959 (gpa >= DEFAULT_APIC_BASE && gpa < DEFAULT_APIC_BASE + PAGE_SIZE)) 1960 return (true); 1961 else 1962 return (false); 1963 } 1964 1965 static int 1966 vmx_handle_apic_access(struct vmx *vmx, int vcpuid, struct vm_exit *vmexit) 1967 { 1968 uint64_t qual; 1969 int access_type, offset, allowed; 1970 1971 if (!apic_access_virtualization(vmx, vcpuid)) 1972 return (UNHANDLED); 1973 1974 qual = vmexit->u.vmx.exit_qualification; 1975 access_type = APIC_ACCESS_TYPE(qual); 1976 offset = APIC_ACCESS_OFFSET(qual); 1977 1978 allowed = 0; 1979 if (access_type == 0) { 1980 /* 1981 * Read data access to the following registers is expected. 1982 */ 1983 switch (offset) { 1984 case APIC_OFFSET_APR: 1985 case APIC_OFFSET_PPR: 1986 case APIC_OFFSET_RRR: 1987 case APIC_OFFSET_CMCI_LVT: 1988 case APIC_OFFSET_TIMER_CCR: 1989 allowed = 1; 1990 break; 1991 default: 1992 break; 1993 } 1994 } else if (access_type == 1) { 1995 /* 1996 * Write data access to the following registers is expected. 1997 */ 1998 switch (offset) { 1999 case APIC_OFFSET_VER: 2000 case APIC_OFFSET_APR: 2001 case APIC_OFFSET_PPR: 2002 case APIC_OFFSET_RRR: 2003 case APIC_OFFSET_ISR0 ... APIC_OFFSET_ISR7: 2004 case APIC_OFFSET_TMR0 ... APIC_OFFSET_TMR7: 2005 case APIC_OFFSET_IRR0 ... APIC_OFFSET_IRR7: 2006 case APIC_OFFSET_CMCI_LVT: 2007 case APIC_OFFSET_TIMER_CCR: 2008 allowed = 1; 2009 break; 2010 default: 2011 break; 2012 } 2013 } 2014 2015 if (allowed) { 2016 vmexit_inst_emul(vmexit, DEFAULT_APIC_BASE + offset, 2017 VIE_INVALID_GLA); 2018 } 2019 2020 /* 2021 * Regardless of whether the APIC-access is allowed this handler 2022 * always returns UNHANDLED: 2023 * - if the access is allowed then it is handled by emulating the 2024 * instruction that caused the VM-exit (outside the critical section) 2025 * - if the access is not allowed then it will be converted to an 2026 * exitcode of VM_EXITCODE_VMX and will be dealt with in userland. 2027 */ 2028 return (UNHANDLED); 2029 } 2030 2031 static enum task_switch_reason 2032 vmx_task_switch_reason(uint64_t qual) 2033 { 2034 int reason; 2035 2036 reason = (qual >> 30) & 0x3; 2037 switch (reason) { 2038 case 0: 2039 return (TSR_CALL); 2040 case 1: 2041 return (TSR_IRET); 2042 case 2: 2043 return (TSR_JMP); 2044 case 3: 2045 return (TSR_IDT_GATE); 2046 default: 2047 panic("%s: invalid reason %d", __func__, reason); 2048 } 2049 } 2050 2051 static int 2052 emulate_wrmsr(struct vmx *vmx, int vcpuid, u_int num, uint64_t val, bool *retu) 2053 { 2054 int error; 2055 2056 if (lapic_msr(num)) 2057 error = lapic_wrmsr(vmx->vm, vcpuid, num, val, retu); 2058 else 2059 error = vmx_wrmsr(vmx, vcpuid, num, val, retu); 2060 2061 return (error); 2062 } 2063 2064 static int 2065 emulate_rdmsr(struct vmx *vmx, int vcpuid, u_int num, bool *retu) 2066 { 2067 struct vmxctx *vmxctx; 2068 uint64_t result; 2069 uint32_t eax, edx; 2070 int error; 2071 2072 if (lapic_msr(num)) 2073 error = lapic_rdmsr(vmx->vm, vcpuid, num, &result, retu); 2074 else 2075 error = vmx_rdmsr(vmx, vcpuid, num, &result, retu); 2076 2077 if (error == 0) { 2078 eax = result; 2079 vmxctx = &vmx->ctx[vcpuid]; 2080 error = vmxctx_setreg(vmxctx, VM_REG_GUEST_RAX, eax); 2081 KASSERT(error == 0, ("vmxctx_setreg(rax) error %d", error)); 2082 2083 edx = result >> 32; 2084 error = vmxctx_setreg(vmxctx, VM_REG_GUEST_RDX, edx); 2085 KASSERT(error == 0, ("vmxctx_setreg(rdx) error %d", error)); 2086 } 2087 2088 return (error); 2089 } 2090 2091 static int 2092 vmx_exit_process(struct vmx *vmx, int vcpu, struct vm_exit *vmexit) 2093 { 2094 int error, errcode, errcode_valid, handled, in; 2095 struct vmxctx *vmxctx; 2096 struct vlapic *vlapic; 2097 struct vm_inout_str *vis; 2098 struct vm_task_switch *ts; 2099 uint32_t eax, ecx, edx, idtvec_info, idtvec_err, intr_info, inst_info; 2100 uint32_t intr_type, intr_vec, reason; 2101 uint64_t exitintinfo, qual, gpa; 2102 bool retu; 2103 2104 CTASSERT((PINBASED_CTLS_ONE_SETTING & PINBASED_VIRTUAL_NMI) != 0); 2105 CTASSERT((PINBASED_CTLS_ONE_SETTING & PINBASED_NMI_EXITING) != 0); 2106 2107 handled = UNHANDLED; 2108 vmxctx = &vmx->ctx[vcpu]; 2109 2110 qual = vmexit->u.vmx.exit_qualification; 2111 reason = vmexit->u.vmx.exit_reason; 2112 vmexit->exitcode = VM_EXITCODE_BOGUS; 2113 2114 vmm_stat_incr(vmx->vm, vcpu, VMEXIT_COUNT, 1); 2115 2116 /* 2117 * VM-entry failures during or after loading guest state. 2118 * 2119 * These VM-exits are uncommon but must be handled specially 2120 * as most VM-exit fields are not populated as usual. 2121 */ 2122 if (__predict_false(reason == EXIT_REASON_MCE_DURING_ENTRY)) { 2123 VCPU_CTR0(vmx->vm, vcpu, "Handling MCE during VM-entry"); 2124 __asm __volatile("int $18"); 2125 return (1); 2126 } 2127 2128 /* 2129 * VM exits that can be triggered during event delivery need to 2130 * be handled specially by re-injecting the event if the IDT 2131 * vectoring information field's valid bit is set. 2132 * 2133 * See "Information for VM Exits During Event Delivery" in Intel SDM 2134 * for details. 2135 */ 2136 idtvec_info = vmcs_idt_vectoring_info(); 2137 if (idtvec_info & VMCS_IDT_VEC_VALID) { 2138 idtvec_info &= ~(1 << 12); /* clear undefined bit */ 2139 exitintinfo = idtvec_info; 2140 if (idtvec_info & VMCS_IDT_VEC_ERRCODE_VALID) { 2141 idtvec_err = vmcs_idt_vectoring_err(); 2142 exitintinfo |= (uint64_t)idtvec_err << 32; 2143 } 2144 error = vm_exit_intinfo(vmx->vm, vcpu, exitintinfo); 2145 KASSERT(error == 0, ("%s: vm_set_intinfo error %d", 2146 __func__, error)); 2147 2148 /* 2149 * If 'virtual NMIs' are being used and the VM-exit 2150 * happened while injecting an NMI during the previous 2151 * VM-entry, then clear "blocking by NMI" in the 2152 * Guest Interruptibility-State so the NMI can be 2153 * reinjected on the subsequent VM-entry. 2154 * 2155 * However, if the NMI was being delivered through a task 2156 * gate, then the new task must start execution with NMIs 2157 * blocked so don't clear NMI blocking in this case. 2158 */ 2159 intr_type = idtvec_info & VMCS_INTR_T_MASK; 2160 if (intr_type == VMCS_INTR_T_NMI) { 2161 if (reason != EXIT_REASON_TASK_SWITCH) 2162 vmx_clear_nmi_blocking(vmx, vcpu); 2163 else 2164 vmx_assert_nmi_blocking(vmx, vcpu); 2165 } 2166 2167 /* 2168 * Update VM-entry instruction length if the event being 2169 * delivered was a software interrupt or software exception. 2170 */ 2171 if (intr_type == VMCS_INTR_T_SWINTR || 2172 intr_type == VMCS_INTR_T_PRIV_SWEXCEPTION || 2173 intr_type == VMCS_INTR_T_SWEXCEPTION) { 2174 vmcs_write(VMCS_ENTRY_INST_LENGTH, vmexit->inst_length); 2175 } 2176 } 2177 2178 switch (reason) { 2179 case EXIT_REASON_TASK_SWITCH: 2180 ts = &vmexit->u.task_switch; 2181 ts->tsssel = qual & 0xffff; 2182 ts->reason = vmx_task_switch_reason(qual); 2183 ts->ext = 0; 2184 ts->errcode_valid = 0; 2185 vmx_paging_info(&ts->paging); 2186 /* 2187 * If the task switch was due to a CALL, JMP, IRET, software 2188 * interrupt (INT n) or software exception (INT3, INTO), 2189 * then the saved %rip references the instruction that caused 2190 * the task switch. The instruction length field in the VMCS 2191 * is valid in this case. 2192 * 2193 * In all other cases (e.g., NMI, hardware exception) the 2194 * saved %rip is one that would have been saved in the old TSS 2195 * had the task switch completed normally so the instruction 2196 * length field is not needed in this case and is explicitly 2197 * set to 0. 2198 */ 2199 if (ts->reason == TSR_IDT_GATE) { 2200 KASSERT(idtvec_info & VMCS_IDT_VEC_VALID, 2201 ("invalid idtvec_info %#x for IDT task switch", 2202 idtvec_info)); 2203 intr_type = idtvec_info & VMCS_INTR_T_MASK; 2204 if (intr_type != VMCS_INTR_T_SWINTR && 2205 intr_type != VMCS_INTR_T_SWEXCEPTION && 2206 intr_type != VMCS_INTR_T_PRIV_SWEXCEPTION) { 2207 /* Task switch triggered by external event */ 2208 ts->ext = 1; 2209 vmexit->inst_length = 0; 2210 if (idtvec_info & VMCS_IDT_VEC_ERRCODE_VALID) { 2211 ts->errcode_valid = 1; 2212 ts->errcode = vmcs_idt_vectoring_err(); 2213 } 2214 } 2215 } 2216 vmexit->exitcode = VM_EXITCODE_TASK_SWITCH; 2217 VCPU_CTR4(vmx->vm, vcpu, "task switch reason %d, tss 0x%04x, " 2218 "%s errcode 0x%016lx", ts->reason, ts->tsssel, 2219 ts->ext ? "external" : "internal", 2220 ((uint64_t)ts->errcode << 32) | ts->errcode_valid); 2221 break; 2222 case EXIT_REASON_CR_ACCESS: 2223 vmm_stat_incr(vmx->vm, vcpu, VMEXIT_CR_ACCESS, 1); 2224 switch (qual & 0xf) { 2225 case 0: 2226 handled = vmx_emulate_cr0_access(vmx, vcpu, qual); 2227 break; 2228 case 4: 2229 handled = vmx_emulate_cr4_access(vmx, vcpu, qual); 2230 break; 2231 case 8: 2232 handled = vmx_emulate_cr8_access(vmx, vcpu, qual); 2233 break; 2234 } 2235 break; 2236 case EXIT_REASON_RDMSR: 2237 vmm_stat_incr(vmx->vm, vcpu, VMEXIT_RDMSR, 1); 2238 retu = false; 2239 ecx = vmxctx->guest_rcx; 2240 VCPU_CTR1(vmx->vm, vcpu, "rdmsr 0x%08x", ecx); 2241 error = emulate_rdmsr(vmx, vcpu, ecx, &retu); 2242 if (error) { 2243 vmexit->exitcode = VM_EXITCODE_RDMSR; 2244 vmexit->u.msr.code = ecx; 2245 } else if (!retu) { 2246 handled = HANDLED; 2247 } else { 2248 /* Return to userspace with a valid exitcode */ 2249 KASSERT(vmexit->exitcode != VM_EXITCODE_BOGUS, 2250 ("emulate_rdmsr retu with bogus exitcode")); 2251 } 2252 break; 2253 case EXIT_REASON_WRMSR: 2254 vmm_stat_incr(vmx->vm, vcpu, VMEXIT_WRMSR, 1); 2255 retu = false; 2256 eax = vmxctx->guest_rax; 2257 ecx = vmxctx->guest_rcx; 2258 edx = vmxctx->guest_rdx; 2259 VCPU_CTR2(vmx->vm, vcpu, "wrmsr 0x%08x value 0x%016lx", 2260 ecx, (uint64_t)edx << 32 | eax); 2261 error = emulate_wrmsr(vmx, vcpu, ecx, 2262 (uint64_t)edx << 32 | eax, &retu); 2263 if (error) { 2264 vmexit->exitcode = VM_EXITCODE_WRMSR; 2265 vmexit->u.msr.code = ecx; 2266 vmexit->u.msr.wval = (uint64_t)edx << 32 | eax; 2267 } else if (!retu) { 2268 handled = HANDLED; 2269 } else { 2270 /* Return to userspace with a valid exitcode */ 2271 KASSERT(vmexit->exitcode != VM_EXITCODE_BOGUS, 2272 ("emulate_wrmsr retu with bogus exitcode")); 2273 } 2274 break; 2275 case EXIT_REASON_HLT: 2276 vmm_stat_incr(vmx->vm, vcpu, VMEXIT_HLT, 1); 2277 vmexit->exitcode = VM_EXITCODE_HLT; 2278 vmexit->u.hlt.rflags = vmcs_read(VMCS_GUEST_RFLAGS); 2279 break; 2280 case EXIT_REASON_MTF: 2281 vmm_stat_incr(vmx->vm, vcpu, VMEXIT_MTRAP, 1); 2282 vmexit->exitcode = VM_EXITCODE_MTRAP; 2283 vmexit->inst_length = 0; 2284 break; 2285 case EXIT_REASON_PAUSE: 2286 vmm_stat_incr(vmx->vm, vcpu, VMEXIT_PAUSE, 1); 2287 vmexit->exitcode = VM_EXITCODE_PAUSE; 2288 break; 2289 case EXIT_REASON_INTR_WINDOW: 2290 vmm_stat_incr(vmx->vm, vcpu, VMEXIT_INTR_WINDOW, 1); 2291 vmx_clear_int_window_exiting(vmx, vcpu); 2292 return (1); 2293 case EXIT_REASON_EXT_INTR: 2294 /* 2295 * External interrupts serve only to cause VM exits and allow 2296 * the host interrupt handler to run. 2297 * 2298 * If this external interrupt triggers a virtual interrupt 2299 * to a VM, then that state will be recorded by the 2300 * host interrupt handler in the VM's softc. We will inject 2301 * this virtual interrupt during the subsequent VM enter. 2302 */ 2303 intr_info = vmcs_read(VMCS_EXIT_INTR_INFO); 2304 2305 /* 2306 * XXX: Ignore this exit if VMCS_INTR_VALID is not set. 2307 * This appears to be a bug in VMware Fusion? 2308 */ 2309 if (!(intr_info & VMCS_INTR_VALID)) 2310 return (1); 2311 KASSERT((intr_info & VMCS_INTR_VALID) != 0 && 2312 (intr_info & VMCS_INTR_T_MASK) == VMCS_INTR_T_HWINTR, 2313 ("VM exit interruption info invalid: %#x", intr_info)); 2314 vmx_trigger_hostintr(intr_info & 0xff); 2315 2316 /* 2317 * This is special. We want to treat this as an 'handled' 2318 * VM-exit but not increment the instruction pointer. 2319 */ 2320 vmm_stat_incr(vmx->vm, vcpu, VMEXIT_EXTINT, 1); 2321 return (1); 2322 case EXIT_REASON_NMI_WINDOW: 2323 /* Exit to allow the pending virtual NMI to be injected */ 2324 if (vm_nmi_pending(vmx->vm, vcpu)) 2325 vmx_inject_nmi(vmx, vcpu); 2326 vmx_clear_nmi_window_exiting(vmx, vcpu); 2327 vmm_stat_incr(vmx->vm, vcpu, VMEXIT_NMI_WINDOW, 1); 2328 return (1); 2329 case EXIT_REASON_INOUT: 2330 vmm_stat_incr(vmx->vm, vcpu, VMEXIT_INOUT, 1); 2331 vmexit->exitcode = VM_EXITCODE_INOUT; 2332 vmexit->u.inout.bytes = (qual & 0x7) + 1; 2333 vmexit->u.inout.in = in = (qual & 0x8) ? 1 : 0; 2334 vmexit->u.inout.string = (qual & 0x10) ? 1 : 0; 2335 vmexit->u.inout.rep = (qual & 0x20) ? 1 : 0; 2336 vmexit->u.inout.port = (uint16_t)(qual >> 16); 2337 vmexit->u.inout.eax = (uint32_t)(vmxctx->guest_rax); 2338 if (vmexit->u.inout.string) { 2339 inst_info = vmcs_read(VMCS_EXIT_INSTRUCTION_INFO); 2340 vmexit->exitcode = VM_EXITCODE_INOUT_STR; 2341 vis = &vmexit->u.inout_str; 2342 vmx_paging_info(&vis->paging); 2343 vis->rflags = vmcs_read(VMCS_GUEST_RFLAGS); 2344 vis->cr0 = vmcs_read(VMCS_GUEST_CR0); 2345 vis->index = inout_str_index(vmx, vcpu, in); 2346 vis->count = inout_str_count(vmx, vcpu, vis->inout.rep); 2347 vis->addrsize = inout_str_addrsize(inst_info); 2348 inout_str_seginfo(vmx, vcpu, inst_info, in, vis); 2349 } 2350 break; 2351 case EXIT_REASON_CPUID: 2352 vmm_stat_incr(vmx->vm, vcpu, VMEXIT_CPUID, 1); 2353 handled = vmx_handle_cpuid(vmx->vm, vcpu, vmxctx); 2354 break; 2355 case EXIT_REASON_EXCEPTION: 2356 vmm_stat_incr(vmx->vm, vcpu, VMEXIT_EXCEPTION, 1); 2357 intr_info = vmcs_read(VMCS_EXIT_INTR_INFO); 2358 KASSERT((intr_info & VMCS_INTR_VALID) != 0, 2359 ("VM exit interruption info invalid: %#x", intr_info)); 2360 2361 intr_vec = intr_info & 0xff; 2362 intr_type = intr_info & VMCS_INTR_T_MASK; 2363 2364 /* 2365 * If Virtual NMIs control is 1 and the VM-exit is due to a 2366 * fault encountered during the execution of IRET then we must 2367 * restore the state of "virtual-NMI blocking" before resuming 2368 * the guest. 2369 * 2370 * See "Resuming Guest Software after Handling an Exception". 2371 * See "Information for VM Exits Due to Vectored Events". 2372 */ 2373 if ((idtvec_info & VMCS_IDT_VEC_VALID) == 0 && 2374 (intr_vec != IDT_DF) && 2375 (intr_info & EXIT_QUAL_NMIUDTI) != 0) 2376 vmx_restore_nmi_blocking(vmx, vcpu); 2377 2378 /* 2379 * The NMI has already been handled in vmx_exit_handle_nmi(). 2380 */ 2381 if (intr_type == VMCS_INTR_T_NMI) 2382 return (1); 2383 2384 /* 2385 * Call the machine check handler by hand. Also don't reflect 2386 * the machine check back into the guest. 2387 */ 2388 if (intr_vec == IDT_MC) { 2389 VCPU_CTR0(vmx->vm, vcpu, "Vectoring to MCE handler"); 2390 __asm __volatile("int $18"); 2391 return (1); 2392 } 2393 2394 if (intr_vec == IDT_PF) { 2395 error = vmxctx_setreg(vmxctx, VM_REG_GUEST_CR2, qual); 2396 KASSERT(error == 0, ("%s: vmxctx_setreg(cr2) error %d", 2397 __func__, error)); 2398 } 2399 2400 /* 2401 * Software exceptions exhibit trap-like behavior. This in 2402 * turn requires populating the VM-entry instruction length 2403 * so that the %rip in the trap frame is past the INT3/INTO 2404 * instruction. 2405 */ 2406 if (intr_type == VMCS_INTR_T_SWEXCEPTION) 2407 vmcs_write(VMCS_ENTRY_INST_LENGTH, vmexit->inst_length); 2408 2409 /* Reflect all other exceptions back into the guest */ 2410 errcode_valid = errcode = 0; 2411 if (intr_info & VMCS_INTR_DEL_ERRCODE) { 2412 errcode_valid = 1; 2413 errcode = vmcs_read(VMCS_EXIT_INTR_ERRCODE); 2414 } 2415 VCPU_CTR2(vmx->vm, vcpu, "Reflecting exception %d/%#x into " 2416 "the guest", intr_vec, errcode); 2417 error = vm_inject_exception(vmx->vm, vcpu, intr_vec, 2418 errcode_valid, errcode, 0); 2419 KASSERT(error == 0, ("%s: vm_inject_exception error %d", 2420 __func__, error)); 2421 return (1); 2422 2423 case EXIT_REASON_EPT_FAULT: 2424 /* 2425 * If 'gpa' lies within the address space allocated to 2426 * memory then this must be a nested page fault otherwise 2427 * this must be an instruction that accesses MMIO space. 2428 */ 2429 gpa = vmcs_gpa(); 2430 if (vm_mem_allocated(vmx->vm, vcpu, gpa) || 2431 apic_access_fault(vmx, vcpu, gpa)) { 2432 vmexit->exitcode = VM_EXITCODE_PAGING; 2433 vmexit->inst_length = 0; 2434 vmexit->u.paging.gpa = gpa; 2435 vmexit->u.paging.fault_type = ept_fault_type(qual); 2436 vmm_stat_incr(vmx->vm, vcpu, VMEXIT_NESTED_FAULT, 1); 2437 } else if (ept_emulation_fault(qual)) { 2438 vmexit_inst_emul(vmexit, gpa, vmcs_gla()); 2439 vmm_stat_incr(vmx->vm, vcpu, VMEXIT_INST_EMUL, 1); 2440 } 2441 /* 2442 * If Virtual NMIs control is 1 and the VM-exit is due to an 2443 * EPT fault during the execution of IRET then we must restore 2444 * the state of "virtual-NMI blocking" before resuming. 2445 * 2446 * See description of "NMI unblocking due to IRET" in 2447 * "Exit Qualification for EPT Violations". 2448 */ 2449 if ((idtvec_info & VMCS_IDT_VEC_VALID) == 0 && 2450 (qual & EXIT_QUAL_NMIUDTI) != 0) 2451 vmx_restore_nmi_blocking(vmx, vcpu); 2452 break; 2453 case EXIT_REASON_VIRTUALIZED_EOI: 2454 vmexit->exitcode = VM_EXITCODE_IOAPIC_EOI; 2455 vmexit->u.ioapic_eoi.vector = qual & 0xFF; 2456 vmexit->inst_length = 0; /* trap-like */ 2457 break; 2458 case EXIT_REASON_APIC_ACCESS: 2459 handled = vmx_handle_apic_access(vmx, vcpu, vmexit); 2460 break; 2461 case EXIT_REASON_APIC_WRITE: 2462 /* 2463 * APIC-write VM exit is trap-like so the %rip is already 2464 * pointing to the next instruction. 2465 */ 2466 vmexit->inst_length = 0; 2467 vlapic = vm_lapic(vmx->vm, vcpu); 2468 handled = vmx_handle_apic_write(vmx, vcpu, vlapic, qual); 2469 break; 2470 case EXIT_REASON_XSETBV: 2471 handled = vmx_emulate_xsetbv(vmx, vcpu, vmexit); 2472 break; 2473 case EXIT_REASON_MONITOR: 2474 vmexit->exitcode = VM_EXITCODE_MONITOR; 2475 break; 2476 case EXIT_REASON_MWAIT: 2477 vmexit->exitcode = VM_EXITCODE_MWAIT; 2478 break; 2479 default: 2480 vmm_stat_incr(vmx->vm, vcpu, VMEXIT_UNKNOWN, 1); 2481 break; 2482 } 2483 2484 if (handled) { 2485 /* 2486 * It is possible that control is returned to userland 2487 * even though we were able to handle the VM exit in the 2488 * kernel. 2489 * 2490 * In such a case we want to make sure that the userland 2491 * restarts guest execution at the instruction *after* 2492 * the one we just processed. Therefore we update the 2493 * guest rip in the VMCS and in 'vmexit'. 2494 */ 2495 vmexit->rip += vmexit->inst_length; 2496 vmexit->inst_length = 0; 2497 vmcs_write(VMCS_GUEST_RIP, vmexit->rip); 2498 } else { 2499 if (vmexit->exitcode == VM_EXITCODE_BOGUS) { 2500 /* 2501 * If this VM exit was not claimed by anybody then 2502 * treat it as a generic VMX exit. 2503 */ 2504 vmexit->exitcode = VM_EXITCODE_VMX; 2505 vmexit->u.vmx.status = VM_SUCCESS; 2506 vmexit->u.vmx.inst_type = 0; 2507 vmexit->u.vmx.inst_error = 0; 2508 } else { 2509 /* 2510 * The exitcode and collateral have been populated. 2511 * The VM exit will be processed further in userland. 2512 */ 2513 } 2514 } 2515 return (handled); 2516 } 2517 2518 static __inline void 2519 vmx_exit_inst_error(struct vmxctx *vmxctx, int rc, struct vm_exit *vmexit) 2520 { 2521 2522 KASSERT(vmxctx->inst_fail_status != VM_SUCCESS, 2523 ("vmx_exit_inst_error: invalid inst_fail_status %d", 2524 vmxctx->inst_fail_status)); 2525 2526 vmexit->inst_length = 0; 2527 vmexit->exitcode = VM_EXITCODE_VMX; 2528 vmexit->u.vmx.status = vmxctx->inst_fail_status; 2529 vmexit->u.vmx.inst_error = vmcs_instruction_error(); 2530 vmexit->u.vmx.exit_reason = ~0; 2531 vmexit->u.vmx.exit_qualification = ~0; 2532 2533 switch (rc) { 2534 case VMX_VMRESUME_ERROR: 2535 case VMX_VMLAUNCH_ERROR: 2536 case VMX_INVEPT_ERROR: 2537 vmexit->u.vmx.inst_type = rc; 2538 break; 2539 default: 2540 panic("vm_exit_inst_error: vmx_enter_guest returned %d", rc); 2541 } 2542 } 2543 2544 /* 2545 * If the NMI-exiting VM execution control is set to '1' then an NMI in 2546 * non-root operation causes a VM-exit. NMI blocking is in effect so it is 2547 * sufficient to simply vector to the NMI handler via a software interrupt. 2548 * However, this must be done before maskable interrupts are enabled 2549 * otherwise the "iret" issued by an interrupt handler will incorrectly 2550 * clear NMI blocking. 2551 */ 2552 static __inline void 2553 vmx_exit_handle_nmi(struct vmx *vmx, int vcpuid, struct vm_exit *vmexit) 2554 { 2555 uint32_t intr_info; 2556 2557 KASSERT((read_rflags() & PSL_I) == 0, ("interrupts enabled")); 2558 2559 if (vmexit->u.vmx.exit_reason != EXIT_REASON_EXCEPTION) 2560 return; 2561 2562 intr_info = vmcs_read(VMCS_EXIT_INTR_INFO); 2563 KASSERT((intr_info & VMCS_INTR_VALID) != 0, 2564 ("VM exit interruption info invalid: %#x", intr_info)); 2565 2566 if ((intr_info & VMCS_INTR_T_MASK) == VMCS_INTR_T_NMI) { 2567 KASSERT((intr_info & 0xff) == IDT_NMI, ("VM exit due " 2568 "to NMI has invalid vector: %#x", intr_info)); 2569 VCPU_CTR0(vmx->vm, vcpuid, "Vectoring to NMI handler"); 2570 __asm __volatile("int $2"); 2571 } 2572 } 2573 2574 static int 2575 vmx_run(void *arg, int vcpu, register_t rip, pmap_t pmap, 2576 struct vm_eventinfo *evinfo) 2577 { 2578 int rc, handled, launched; 2579 struct vmx *vmx; 2580 struct vm *vm; 2581 struct vmxctx *vmxctx; 2582 struct vmcs *vmcs; 2583 struct vm_exit *vmexit; 2584 struct vlapic *vlapic; 2585 uint32_t exit_reason; 2586 2587 vmx = arg; 2588 vm = vmx->vm; 2589 vmcs = &vmx->vmcs[vcpu]; 2590 vmxctx = &vmx->ctx[vcpu]; 2591 vlapic = vm_lapic(vm, vcpu); 2592 vmexit = vm_exitinfo(vm, vcpu); 2593 launched = 0; 2594 2595 KASSERT(vmxctx->pmap == pmap, 2596 ("pmap %p different than ctx pmap %p", pmap, vmxctx->pmap)); 2597 2598 vmx_msr_guest_enter(vmx, vcpu); 2599 2600 VMPTRLD(vmcs); 2601 2602 /* 2603 * XXX 2604 * We do this every time because we may setup the virtual machine 2605 * from a different process than the one that actually runs it. 2606 * 2607 * If the life of a virtual machine was spent entirely in the context 2608 * of a single process we could do this once in vmx_vminit(). 2609 */ 2610 vmcs_write(VMCS_HOST_CR3, rcr3()); 2611 2612 vmcs_write(VMCS_GUEST_RIP, rip); 2613 vmx_set_pcpu_defaults(vmx, vcpu, pmap); 2614 do { 2615 KASSERT(vmcs_guest_rip() == rip, ("%s: vmcs guest rip mismatch " 2616 "%#lx/%#lx", __func__, vmcs_guest_rip(), rip)); 2617 2618 handled = UNHANDLED; 2619 /* 2620 * Interrupts are disabled from this point on until the 2621 * guest starts executing. This is done for the following 2622 * reasons: 2623 * 2624 * If an AST is asserted on this thread after the check below, 2625 * then the IPI_AST notification will not be lost, because it 2626 * will cause a VM exit due to external interrupt as soon as 2627 * the guest state is loaded. 2628 * 2629 * A posted interrupt after 'vmx_inject_interrupts()' will 2630 * not be "lost" because it will be held pending in the host 2631 * APIC because interrupts are disabled. The pending interrupt 2632 * will be recognized as soon as the guest state is loaded. 2633 * 2634 * The same reasoning applies to the IPI generated by 2635 * pmap_invalidate_ept(). 2636 */ 2637 disable_intr(); 2638 vmx_inject_interrupts(vmx, vcpu, vlapic, rip); 2639 2640 /* 2641 * Check for vcpu suspension after injecting events because 2642 * vmx_inject_interrupts() can suspend the vcpu due to a 2643 * triple fault. 2644 */ 2645 if (vcpu_suspended(evinfo)) { 2646 enable_intr(); 2647 vm_exit_suspended(vmx->vm, vcpu, rip); 2648 break; 2649 } 2650 2651 if (vcpu_rendezvous_pending(evinfo)) { 2652 enable_intr(); 2653 vm_exit_rendezvous(vmx->vm, vcpu, rip); 2654 break; 2655 } 2656 2657 if (vcpu_reqidle(evinfo)) { 2658 enable_intr(); 2659 vm_exit_reqidle(vmx->vm, vcpu, rip); 2660 break; 2661 } 2662 2663 if (vcpu_should_yield(vm, vcpu)) { 2664 enable_intr(); 2665 vm_exit_astpending(vmx->vm, vcpu, rip); 2666 vmx_astpending_trace(vmx, vcpu, rip); 2667 handled = HANDLED; 2668 break; 2669 } 2670 2671 vmx_run_trace(vmx, vcpu); 2672 rc = vmx_enter_guest(vmxctx, vmx, launched); 2673 2674 /* Collect some information for VM exit processing */ 2675 vmexit->rip = rip = vmcs_guest_rip(); 2676 vmexit->inst_length = vmexit_instruction_length(); 2677 vmexit->u.vmx.exit_reason = exit_reason = vmcs_exit_reason(); 2678 vmexit->u.vmx.exit_qualification = vmcs_exit_qualification(); 2679 2680 /* Update 'nextrip' */ 2681 vmx->state[vcpu].nextrip = rip; 2682 2683 if (rc == VMX_GUEST_VMEXIT) { 2684 vmx_exit_handle_nmi(vmx, vcpu, vmexit); 2685 enable_intr(); 2686 handled = vmx_exit_process(vmx, vcpu, vmexit); 2687 } else { 2688 enable_intr(); 2689 vmx_exit_inst_error(vmxctx, rc, vmexit); 2690 } 2691 launched = 1; 2692 vmx_exit_trace(vmx, vcpu, rip, exit_reason, handled); 2693 rip = vmexit->rip; 2694 } while (handled); 2695 2696 /* 2697 * If a VM exit has been handled then the exitcode must be BOGUS 2698 * If a VM exit is not handled then the exitcode must not be BOGUS 2699 */ 2700 if ((handled && vmexit->exitcode != VM_EXITCODE_BOGUS) || 2701 (!handled && vmexit->exitcode == VM_EXITCODE_BOGUS)) { 2702 panic("Mismatch between handled (%d) and exitcode (%d)", 2703 handled, vmexit->exitcode); 2704 } 2705 2706 if (!handled) 2707 vmm_stat_incr(vm, vcpu, VMEXIT_USERSPACE, 1); 2708 2709 VCPU_CTR1(vm, vcpu, "returning from vmx_run: exitcode %d", 2710 vmexit->exitcode); 2711 2712 VMCLEAR(vmcs); 2713 vmx_msr_guest_exit(vmx, vcpu); 2714 2715 return (0); 2716 } 2717 2718 static void 2719 vmx_vmcleanup(void *arg) 2720 { 2721 int i; 2722 struct vmx *vmx = arg; 2723 2724 if (apic_access_virtualization(vmx, 0)) 2725 vm_unmap_mmio(vmx->vm, DEFAULT_APIC_BASE, PAGE_SIZE); 2726 2727 for (i = 0; i < VM_MAXCPU; i++) 2728 vpid_free(vmx->state[i].vpid); 2729 2730 free(vmx, M_VMX); 2731 2732 return; 2733 } 2734 2735 static register_t * 2736 vmxctx_regptr(struct vmxctx *vmxctx, int reg) 2737 { 2738 2739 switch (reg) { 2740 case VM_REG_GUEST_RAX: 2741 return (&vmxctx->guest_rax); 2742 case VM_REG_GUEST_RBX: 2743 return (&vmxctx->guest_rbx); 2744 case VM_REG_GUEST_RCX: 2745 return (&vmxctx->guest_rcx); 2746 case VM_REG_GUEST_RDX: 2747 return (&vmxctx->guest_rdx); 2748 case VM_REG_GUEST_RSI: 2749 return (&vmxctx->guest_rsi); 2750 case VM_REG_GUEST_RDI: 2751 return (&vmxctx->guest_rdi); 2752 case VM_REG_GUEST_RBP: 2753 return (&vmxctx->guest_rbp); 2754 case VM_REG_GUEST_R8: 2755 return (&vmxctx->guest_r8); 2756 case VM_REG_GUEST_R9: 2757 return (&vmxctx->guest_r9); 2758 case VM_REG_GUEST_R10: 2759 return (&vmxctx->guest_r10); 2760 case VM_REG_GUEST_R11: 2761 return (&vmxctx->guest_r11); 2762 case VM_REG_GUEST_R12: 2763 return (&vmxctx->guest_r12); 2764 case VM_REG_GUEST_R13: 2765 return (&vmxctx->guest_r13); 2766 case VM_REG_GUEST_R14: 2767 return (&vmxctx->guest_r14); 2768 case VM_REG_GUEST_R15: 2769 return (&vmxctx->guest_r15); 2770 case VM_REG_GUEST_CR2: 2771 return (&vmxctx->guest_cr2); 2772 default: 2773 break; 2774 } 2775 return (NULL); 2776 } 2777 2778 static int 2779 vmxctx_getreg(struct vmxctx *vmxctx, int reg, uint64_t *retval) 2780 { 2781 register_t *regp; 2782 2783 if ((regp = vmxctx_regptr(vmxctx, reg)) != NULL) { 2784 *retval = *regp; 2785 return (0); 2786 } else 2787 return (EINVAL); 2788 } 2789 2790 static int 2791 vmxctx_setreg(struct vmxctx *vmxctx, int reg, uint64_t val) 2792 { 2793 register_t *regp; 2794 2795 if ((regp = vmxctx_regptr(vmxctx, reg)) != NULL) { 2796 *regp = val; 2797 return (0); 2798 } else 2799 return (EINVAL); 2800 } 2801 2802 static int 2803 vmx_get_intr_shadow(struct vmx *vmx, int vcpu, int running, uint64_t *retval) 2804 { 2805 uint64_t gi; 2806 int error; 2807 2808 error = vmcs_getreg(&vmx->vmcs[vcpu], running, 2809 VMCS_IDENT(VMCS_GUEST_INTERRUPTIBILITY), &gi); 2810 *retval = (gi & HWINTR_BLOCKING) ? 1 : 0; 2811 return (error); 2812 } 2813 2814 static int 2815 vmx_modify_intr_shadow(struct vmx *vmx, int vcpu, int running, uint64_t val) 2816 { 2817 struct vmcs *vmcs; 2818 uint64_t gi; 2819 int error, ident; 2820 2821 /* 2822 * Forcing the vcpu into an interrupt shadow is not supported. 2823 */ 2824 if (val) { 2825 error = EINVAL; 2826 goto done; 2827 } 2828 2829 vmcs = &vmx->vmcs[vcpu]; 2830 ident = VMCS_IDENT(VMCS_GUEST_INTERRUPTIBILITY); 2831 error = vmcs_getreg(vmcs, running, ident, &gi); 2832 if (error == 0) { 2833 gi &= ~HWINTR_BLOCKING; 2834 error = vmcs_setreg(vmcs, running, ident, gi); 2835 } 2836 done: 2837 VCPU_CTR2(vmx->vm, vcpu, "Setting intr_shadow to %#lx %s", val, 2838 error ? "failed" : "succeeded"); 2839 return (error); 2840 } 2841 2842 static int 2843 vmx_shadow_reg(int reg) 2844 { 2845 int shreg; 2846 2847 shreg = -1; 2848 2849 switch (reg) { 2850 case VM_REG_GUEST_CR0: 2851 shreg = VMCS_CR0_SHADOW; 2852 break; 2853 case VM_REG_GUEST_CR4: 2854 shreg = VMCS_CR4_SHADOW; 2855 break; 2856 default: 2857 break; 2858 } 2859 2860 return (shreg); 2861 } 2862 2863 static int 2864 vmx_getreg(void *arg, int vcpu, int reg, uint64_t *retval) 2865 { 2866 int running, hostcpu; 2867 struct vmx *vmx = arg; 2868 2869 running = vcpu_is_running(vmx->vm, vcpu, &hostcpu); 2870 if (running && hostcpu != curcpu) 2871 panic("vmx_getreg: %s%d is running", vm_name(vmx->vm), vcpu); 2872 2873 if (reg == VM_REG_GUEST_INTR_SHADOW) 2874 return (vmx_get_intr_shadow(vmx, vcpu, running, retval)); 2875 2876 if (vmxctx_getreg(&vmx->ctx[vcpu], reg, retval) == 0) 2877 return (0); 2878 2879 return (vmcs_getreg(&vmx->vmcs[vcpu], running, reg, retval)); 2880 } 2881 2882 static int 2883 vmx_setreg(void *arg, int vcpu, int reg, uint64_t val) 2884 { 2885 int error, hostcpu, running, shadow; 2886 uint64_t ctls; 2887 pmap_t pmap; 2888 struct vmx *vmx = arg; 2889 2890 running = vcpu_is_running(vmx->vm, vcpu, &hostcpu); 2891 if (running && hostcpu != curcpu) 2892 panic("vmx_setreg: %s%d is running", vm_name(vmx->vm), vcpu); 2893 2894 if (reg == VM_REG_GUEST_INTR_SHADOW) 2895 return (vmx_modify_intr_shadow(vmx, vcpu, running, val)); 2896 2897 if (vmxctx_setreg(&vmx->ctx[vcpu], reg, val) == 0) 2898 return (0); 2899 2900 error = vmcs_setreg(&vmx->vmcs[vcpu], running, reg, val); 2901 2902 if (error == 0) { 2903 /* 2904 * If the "load EFER" VM-entry control is 1 then the 2905 * value of EFER.LMA must be identical to "IA-32e mode guest" 2906 * bit in the VM-entry control. 2907 */ 2908 if ((entry_ctls & VM_ENTRY_LOAD_EFER) != 0 && 2909 (reg == VM_REG_GUEST_EFER)) { 2910 vmcs_getreg(&vmx->vmcs[vcpu], running, 2911 VMCS_IDENT(VMCS_ENTRY_CTLS), &ctls); 2912 if (val & EFER_LMA) 2913 ctls |= VM_ENTRY_GUEST_LMA; 2914 else 2915 ctls &= ~VM_ENTRY_GUEST_LMA; 2916 vmcs_setreg(&vmx->vmcs[vcpu], running, 2917 VMCS_IDENT(VMCS_ENTRY_CTLS), ctls); 2918 } 2919 2920 shadow = vmx_shadow_reg(reg); 2921 if (shadow > 0) { 2922 /* 2923 * Store the unmodified value in the shadow 2924 */ 2925 error = vmcs_setreg(&vmx->vmcs[vcpu], running, 2926 VMCS_IDENT(shadow), val); 2927 } 2928 2929 if (reg == VM_REG_GUEST_CR3) { 2930 /* 2931 * Invalidate the guest vcpu's TLB mappings to emulate 2932 * the behavior of updating %cr3. 2933 * 2934 * XXX the processor retains global mappings when %cr3 2935 * is updated but vmx_invvpid() does not. 2936 */ 2937 pmap = vmx->ctx[vcpu].pmap; 2938 vmx_invvpid(vmx, vcpu, pmap, running); 2939 } 2940 } 2941 2942 return (error); 2943 } 2944 2945 static int 2946 vmx_getdesc(void *arg, int vcpu, int reg, struct seg_desc *desc) 2947 { 2948 int hostcpu, running; 2949 struct vmx *vmx = arg; 2950 2951 running = vcpu_is_running(vmx->vm, vcpu, &hostcpu); 2952 if (running && hostcpu != curcpu) 2953 panic("vmx_getdesc: %s%d is running", vm_name(vmx->vm), vcpu); 2954 2955 return (vmcs_getdesc(&vmx->vmcs[vcpu], running, reg, desc)); 2956 } 2957 2958 static int 2959 vmx_setdesc(void *arg, int vcpu, int reg, struct seg_desc *desc) 2960 { 2961 int hostcpu, running; 2962 struct vmx *vmx = arg; 2963 2964 running = vcpu_is_running(vmx->vm, vcpu, &hostcpu); 2965 if (running && hostcpu != curcpu) 2966 panic("vmx_setdesc: %s%d is running", vm_name(vmx->vm), vcpu); 2967 2968 return (vmcs_setdesc(&vmx->vmcs[vcpu], running, reg, desc)); 2969 } 2970 2971 static int 2972 vmx_getcap(void *arg, int vcpu, int type, int *retval) 2973 { 2974 struct vmx *vmx = arg; 2975 int vcap; 2976 int ret; 2977 2978 ret = ENOENT; 2979 2980 vcap = vmx->cap[vcpu].set; 2981 2982 switch (type) { 2983 case VM_CAP_HALT_EXIT: 2984 if (cap_halt_exit) 2985 ret = 0; 2986 break; 2987 case VM_CAP_PAUSE_EXIT: 2988 if (cap_pause_exit) 2989 ret = 0; 2990 break; 2991 case VM_CAP_MTRAP_EXIT: 2992 if (cap_monitor_trap) 2993 ret = 0; 2994 break; 2995 case VM_CAP_UNRESTRICTED_GUEST: 2996 if (cap_unrestricted_guest) 2997 ret = 0; 2998 break; 2999 case VM_CAP_ENABLE_INVPCID: 3000 if (cap_invpcid) 3001 ret = 0; 3002 break; 3003 default: 3004 break; 3005 } 3006 3007 if (ret == 0) 3008 *retval = (vcap & (1 << type)) ? 1 : 0; 3009 3010 return (ret); 3011 } 3012 3013 static int 3014 vmx_setcap(void *arg, int vcpu, int type, int val) 3015 { 3016 struct vmx *vmx = arg; 3017 struct vmcs *vmcs = &vmx->vmcs[vcpu]; 3018 uint32_t baseval; 3019 uint32_t *pptr; 3020 int error; 3021 int flag; 3022 int reg; 3023 int retval; 3024 3025 retval = ENOENT; 3026 pptr = NULL; 3027 3028 switch (type) { 3029 case VM_CAP_HALT_EXIT: 3030 if (cap_halt_exit) { 3031 retval = 0; 3032 pptr = &vmx->cap[vcpu].proc_ctls; 3033 baseval = *pptr; 3034 flag = PROCBASED_HLT_EXITING; 3035 reg = VMCS_PRI_PROC_BASED_CTLS; 3036 } 3037 break; 3038 case VM_CAP_MTRAP_EXIT: 3039 if (cap_monitor_trap) { 3040 retval = 0; 3041 pptr = &vmx->cap[vcpu].proc_ctls; 3042 baseval = *pptr; 3043 flag = PROCBASED_MTF; 3044 reg = VMCS_PRI_PROC_BASED_CTLS; 3045 } 3046 break; 3047 case VM_CAP_PAUSE_EXIT: 3048 if (cap_pause_exit) { 3049 retval = 0; 3050 pptr = &vmx->cap[vcpu].proc_ctls; 3051 baseval = *pptr; 3052 flag = PROCBASED_PAUSE_EXITING; 3053 reg = VMCS_PRI_PROC_BASED_CTLS; 3054 } 3055 break; 3056 case VM_CAP_UNRESTRICTED_GUEST: 3057 if (cap_unrestricted_guest) { 3058 retval = 0; 3059 pptr = &vmx->cap[vcpu].proc_ctls2; 3060 baseval = *pptr; 3061 flag = PROCBASED2_UNRESTRICTED_GUEST; 3062 reg = VMCS_SEC_PROC_BASED_CTLS; 3063 } 3064 break; 3065 case VM_CAP_ENABLE_INVPCID: 3066 if (cap_invpcid) { 3067 retval = 0; 3068 pptr = &vmx->cap[vcpu].proc_ctls2; 3069 baseval = *pptr; 3070 flag = PROCBASED2_ENABLE_INVPCID; 3071 reg = VMCS_SEC_PROC_BASED_CTLS; 3072 } 3073 break; 3074 default: 3075 break; 3076 } 3077 3078 if (retval == 0) { 3079 if (val) { 3080 baseval |= flag; 3081 } else { 3082 baseval &= ~flag; 3083 } 3084 VMPTRLD(vmcs); 3085 error = vmwrite(reg, baseval); 3086 VMCLEAR(vmcs); 3087 3088 if (error) { 3089 retval = error; 3090 } else { 3091 /* 3092 * Update optional stored flags, and record 3093 * setting 3094 */ 3095 if (pptr != NULL) { 3096 *pptr = baseval; 3097 } 3098 3099 if (val) { 3100 vmx->cap[vcpu].set |= (1 << type); 3101 } else { 3102 vmx->cap[vcpu].set &= ~(1 << type); 3103 } 3104 } 3105 } 3106 3107 return (retval); 3108 } 3109 3110 struct vlapic_vtx { 3111 struct vlapic vlapic; 3112 struct pir_desc *pir_desc; 3113 struct vmx *vmx; 3114 }; 3115 3116 #define VMX_CTR_PIR(vm, vcpuid, pir_desc, notify, vector, level, msg) \ 3117 do { \ 3118 VCPU_CTR2(vm, vcpuid, msg " assert %s-triggered vector %d", \ 3119 level ? "level" : "edge", vector); \ 3120 VCPU_CTR1(vm, vcpuid, msg " pir0 0x%016lx", pir_desc->pir[0]); \ 3121 VCPU_CTR1(vm, vcpuid, msg " pir1 0x%016lx", pir_desc->pir[1]); \ 3122 VCPU_CTR1(vm, vcpuid, msg " pir2 0x%016lx", pir_desc->pir[2]); \ 3123 VCPU_CTR1(vm, vcpuid, msg " pir3 0x%016lx", pir_desc->pir[3]); \ 3124 VCPU_CTR1(vm, vcpuid, msg " notify: %s", notify ? "yes" : "no");\ 3125 } while (0) 3126 3127 /* 3128 * vlapic->ops handlers that utilize the APICv hardware assist described in 3129 * Chapter 29 of the Intel SDM. 3130 */ 3131 static int 3132 vmx_set_intr_ready(struct vlapic *vlapic, int vector, bool level) 3133 { 3134 struct vlapic_vtx *vlapic_vtx; 3135 struct pir_desc *pir_desc; 3136 uint64_t mask; 3137 int idx, notify; 3138 3139 vlapic_vtx = (struct vlapic_vtx *)vlapic; 3140 pir_desc = vlapic_vtx->pir_desc; 3141 3142 /* 3143 * Keep track of interrupt requests in the PIR descriptor. This is 3144 * because the virtual APIC page pointed to by the VMCS cannot be 3145 * modified if the vcpu is running. 3146 */ 3147 idx = vector / 64; 3148 mask = 1UL << (vector % 64); 3149 atomic_set_long(&pir_desc->pir[idx], mask); 3150 notify = atomic_cmpset_long(&pir_desc->pending, 0, 1); 3151 3152 VMX_CTR_PIR(vlapic->vm, vlapic->vcpuid, pir_desc, notify, vector, 3153 level, "vmx_set_intr_ready"); 3154 return (notify); 3155 } 3156 3157 static int 3158 vmx_pending_intr(struct vlapic *vlapic, int *vecptr) 3159 { 3160 struct vlapic_vtx *vlapic_vtx; 3161 struct pir_desc *pir_desc; 3162 struct LAPIC *lapic; 3163 uint64_t pending, pirval; 3164 uint32_t ppr, vpr; 3165 int i; 3166 3167 /* 3168 * This function is only expected to be called from the 'HLT' exit 3169 * handler which does not care about the vector that is pending. 3170 */ 3171 KASSERT(vecptr == NULL, ("vmx_pending_intr: vecptr must be NULL")); 3172 3173 vlapic_vtx = (struct vlapic_vtx *)vlapic; 3174 pir_desc = vlapic_vtx->pir_desc; 3175 3176 pending = atomic_load_acq_long(&pir_desc->pending); 3177 if (!pending) 3178 return (0); /* common case */ 3179 3180 /* 3181 * If there is an interrupt pending then it will be recognized only 3182 * if its priority is greater than the processor priority. 3183 * 3184 * Special case: if the processor priority is zero then any pending 3185 * interrupt will be recognized. 3186 */ 3187 lapic = vlapic->apic_page; 3188 ppr = lapic->ppr & 0xf0; 3189 if (ppr == 0) 3190 return (1); 3191 3192 VCPU_CTR1(vlapic->vm, vlapic->vcpuid, "HLT with non-zero PPR %d", 3193 lapic->ppr); 3194 3195 for (i = 3; i >= 0; i--) { 3196 pirval = pir_desc->pir[i]; 3197 if (pirval != 0) { 3198 vpr = (i * 64 + flsl(pirval) - 1) & 0xf0; 3199 return (vpr > ppr); 3200 } 3201 } 3202 return (0); 3203 } 3204 3205 static void 3206 vmx_intr_accepted(struct vlapic *vlapic, int vector) 3207 { 3208 3209 panic("vmx_intr_accepted: not expected to be called"); 3210 } 3211 3212 static void 3213 vmx_set_tmr(struct vlapic *vlapic, int vector, bool level) 3214 { 3215 struct vlapic_vtx *vlapic_vtx; 3216 struct vmx *vmx; 3217 struct vmcs *vmcs; 3218 uint64_t mask, val; 3219 3220 KASSERT(vector >= 0 && vector <= 255, ("invalid vector %d", vector)); 3221 KASSERT(!vcpu_is_running(vlapic->vm, vlapic->vcpuid, NULL), 3222 ("vmx_set_tmr: vcpu cannot be running")); 3223 3224 vlapic_vtx = (struct vlapic_vtx *)vlapic; 3225 vmx = vlapic_vtx->vmx; 3226 vmcs = &vmx->vmcs[vlapic->vcpuid]; 3227 mask = 1UL << (vector % 64); 3228 3229 VMPTRLD(vmcs); 3230 val = vmcs_read(VMCS_EOI_EXIT(vector)); 3231 if (level) 3232 val |= mask; 3233 else 3234 val &= ~mask; 3235 vmcs_write(VMCS_EOI_EXIT(vector), val); 3236 VMCLEAR(vmcs); 3237 } 3238 3239 static void 3240 vmx_enable_x2apic_mode(struct vlapic *vlapic) 3241 { 3242 struct vmx *vmx; 3243 struct vmcs *vmcs; 3244 uint32_t proc_ctls2; 3245 int vcpuid, error; 3246 3247 vcpuid = vlapic->vcpuid; 3248 vmx = ((struct vlapic_vtx *)vlapic)->vmx; 3249 vmcs = &vmx->vmcs[vcpuid]; 3250 3251 proc_ctls2 = vmx->cap[vcpuid].proc_ctls2; 3252 KASSERT((proc_ctls2 & PROCBASED2_VIRTUALIZE_APIC_ACCESSES) != 0, 3253 ("%s: invalid proc_ctls2 %#x", __func__, proc_ctls2)); 3254 3255 proc_ctls2 &= ~PROCBASED2_VIRTUALIZE_APIC_ACCESSES; 3256 proc_ctls2 |= PROCBASED2_VIRTUALIZE_X2APIC_MODE; 3257 vmx->cap[vcpuid].proc_ctls2 = proc_ctls2; 3258 3259 VMPTRLD(vmcs); 3260 vmcs_write(VMCS_SEC_PROC_BASED_CTLS, proc_ctls2); 3261 VMCLEAR(vmcs); 3262 3263 if (vlapic->vcpuid == 0) { 3264 /* 3265 * The nested page table mappings are shared by all vcpus 3266 * so unmap the APIC access page just once. 3267 */ 3268 error = vm_unmap_mmio(vmx->vm, DEFAULT_APIC_BASE, PAGE_SIZE); 3269 KASSERT(error == 0, ("%s: vm_unmap_mmio error %d", 3270 __func__, error)); 3271 3272 /* 3273 * The MSR bitmap is shared by all vcpus so modify it only 3274 * once in the context of vcpu 0. 3275 */ 3276 error = vmx_allow_x2apic_msrs(vmx); 3277 KASSERT(error == 0, ("%s: vmx_allow_x2apic_msrs error %d", 3278 __func__, error)); 3279 } 3280 } 3281 3282 static void 3283 vmx_post_intr(struct vlapic *vlapic, int hostcpu) 3284 { 3285 3286 ipi_cpu(hostcpu, pirvec); 3287 } 3288 3289 /* 3290 * Transfer the pending interrupts in the PIR descriptor to the IRR 3291 * in the virtual APIC page. 3292 */ 3293 static void 3294 vmx_inject_pir(struct vlapic *vlapic) 3295 { 3296 struct vlapic_vtx *vlapic_vtx; 3297 struct pir_desc *pir_desc; 3298 struct LAPIC *lapic; 3299 uint64_t val, pirval; 3300 int rvi, pirbase = -1; 3301 uint16_t intr_status_old, intr_status_new; 3302 3303 vlapic_vtx = (struct vlapic_vtx *)vlapic; 3304 pir_desc = vlapic_vtx->pir_desc; 3305 if (atomic_cmpset_long(&pir_desc->pending, 1, 0) == 0) { 3306 VCPU_CTR0(vlapic->vm, vlapic->vcpuid, "vmx_inject_pir: " 3307 "no posted interrupt pending"); 3308 return; 3309 } 3310 3311 pirval = 0; 3312 pirbase = -1; 3313 lapic = vlapic->apic_page; 3314 3315 val = atomic_readandclear_long(&pir_desc->pir[0]); 3316 if (val != 0) { 3317 lapic->irr0 |= val; 3318 lapic->irr1 |= val >> 32; 3319 pirbase = 0; 3320 pirval = val; 3321 } 3322 3323 val = atomic_readandclear_long(&pir_desc->pir[1]); 3324 if (val != 0) { 3325 lapic->irr2 |= val; 3326 lapic->irr3 |= val >> 32; 3327 pirbase = 64; 3328 pirval = val; 3329 } 3330 3331 val = atomic_readandclear_long(&pir_desc->pir[2]); 3332 if (val != 0) { 3333 lapic->irr4 |= val; 3334 lapic->irr5 |= val >> 32; 3335 pirbase = 128; 3336 pirval = val; 3337 } 3338 3339 val = atomic_readandclear_long(&pir_desc->pir[3]); 3340 if (val != 0) { 3341 lapic->irr6 |= val; 3342 lapic->irr7 |= val >> 32; 3343 pirbase = 192; 3344 pirval = val; 3345 } 3346 3347 VLAPIC_CTR_IRR(vlapic, "vmx_inject_pir"); 3348 3349 /* 3350 * Update RVI so the processor can evaluate pending virtual 3351 * interrupts on VM-entry. 3352 * 3353 * It is possible for pirval to be 0 here, even though the 3354 * pending bit has been set. The scenario is: 3355 * CPU-Y is sending a posted interrupt to CPU-X, which 3356 * is running a guest and processing posted interrupts in h/w. 3357 * CPU-X will eventually exit and the state seen in s/w is 3358 * the pending bit set, but no PIR bits set. 3359 * 3360 * CPU-X CPU-Y 3361 * (vm running) (host running) 3362 * rx posted interrupt 3363 * CLEAR pending bit 3364 * SET PIR bit 3365 * READ/CLEAR PIR bits 3366 * SET pending bit 3367 * (vm exit) 3368 * pending bit set, PIR 0 3369 */ 3370 if (pirval != 0) { 3371 rvi = pirbase + flsl(pirval) - 1; 3372 intr_status_old = vmcs_read(VMCS_GUEST_INTR_STATUS); 3373 intr_status_new = (intr_status_old & 0xFF00) | rvi; 3374 if (intr_status_new > intr_status_old) { 3375 vmcs_write(VMCS_GUEST_INTR_STATUS, intr_status_new); 3376 VCPU_CTR2(vlapic->vm, vlapic->vcpuid, "vmx_inject_pir: " 3377 "guest_intr_status changed from 0x%04x to 0x%04x", 3378 intr_status_old, intr_status_new); 3379 } 3380 } 3381 } 3382 3383 static struct vlapic * 3384 vmx_vlapic_init(void *arg, int vcpuid) 3385 { 3386 struct vmx *vmx; 3387 struct vlapic *vlapic; 3388 struct vlapic_vtx *vlapic_vtx; 3389 3390 vmx = arg; 3391 3392 vlapic = malloc(sizeof(struct vlapic_vtx), M_VLAPIC, M_WAITOK | M_ZERO); 3393 vlapic->vm = vmx->vm; 3394 vlapic->vcpuid = vcpuid; 3395 vlapic->apic_page = (struct LAPIC *)&vmx->apic_page[vcpuid]; 3396 3397 vlapic_vtx = (struct vlapic_vtx *)vlapic; 3398 vlapic_vtx->pir_desc = &vmx->pir_desc[vcpuid]; 3399 vlapic_vtx->vmx = vmx; 3400 3401 if (virtual_interrupt_delivery) { 3402 vlapic->ops.set_intr_ready = vmx_set_intr_ready; 3403 vlapic->ops.pending_intr = vmx_pending_intr; 3404 vlapic->ops.intr_accepted = vmx_intr_accepted; 3405 vlapic->ops.set_tmr = vmx_set_tmr; 3406 vlapic->ops.enable_x2apic_mode = vmx_enable_x2apic_mode; 3407 } 3408 3409 if (posted_interrupts) 3410 vlapic->ops.post_intr = vmx_post_intr; 3411 3412 vlapic_init(vlapic); 3413 3414 return (vlapic); 3415 } 3416 3417 static void 3418 vmx_vlapic_cleanup(void *arg, struct vlapic *vlapic) 3419 { 3420 3421 vlapic_cleanup(vlapic); 3422 free(vlapic, M_VLAPIC); 3423 } 3424 3425 struct vmm_ops vmm_ops_intel = { 3426 vmx_init, 3427 vmx_cleanup, 3428 vmx_restore, 3429 vmx_vminit, 3430 vmx_run, 3431 vmx_vmcleanup, 3432 vmx_getreg, 3433 vmx_setreg, 3434 vmx_getdesc, 3435 vmx_setdesc, 3436 vmx_getcap, 3437 vmx_setcap, 3438 ept_vmspace_alloc, 3439 ept_vmspace_free, 3440 vmx_vlapic_init, 3441 vmx_vlapic_cleanup, 3442 }; 3443