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