1 /*- 2 * SPDX-License-Identifier: BSD-2-Clause-FreeBSD 3 * 4 * Copyright (c) 2011 NetApp, Inc. 5 * All rights reserved. 6 * Copyright (c) 2018 Joyent, Inc. 7 * 8 * Redistribution and use in source and binary forms, with or without 9 * modification, are permitted provided that the following conditions 10 * are met: 11 * 1. Redistributions of source code must retain the above copyright 12 * notice, this list of conditions and the following disclaimer. 13 * 2. Redistributions in binary form must reproduce the above copyright 14 * notice, this list of conditions and the following disclaimer in the 15 * documentation and/or other materials provided with the distribution. 16 * 17 * THIS SOFTWARE IS PROVIDED BY NETAPP, INC ``AS IS'' AND 18 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 19 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 20 * ARE DISCLAIMED. IN NO EVENT SHALL NETAPP, INC OR CONTRIBUTORS BE LIABLE 21 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 22 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 23 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 24 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 25 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 26 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 27 * SUCH DAMAGE. 28 * 29 * $FreeBSD$ 30 */ 31 32 #include <sys/cdefs.h> 33 __FBSDID("$FreeBSD$"); 34 35 #include "opt_bhyve_snapshot.h" 36 37 #include <sys/param.h> 38 #include <sys/systm.h> 39 #include <sys/smp.h> 40 #include <sys/kernel.h> 41 #include <sys/malloc.h> 42 #include <sys/pcpu.h> 43 #include <sys/proc.h> 44 #include <sys/sysctl.h> 45 46 #include <vm/vm.h> 47 #include <vm/pmap.h> 48 49 #include <machine/psl.h> 50 #include <machine/cpufunc.h> 51 #include <machine/md_var.h> 52 #include <machine/reg.h> 53 #include <machine/segments.h> 54 #include <machine/smp.h> 55 #include <machine/specialreg.h> 56 #include <machine/vmparam.h> 57 58 #include <machine/vmm.h> 59 #include <machine/vmm_dev.h> 60 #include <machine/vmm_instruction_emul.h> 61 #include <machine/vmm_snapshot.h> 62 63 #include "vmm_lapic.h" 64 #include "vmm_host.h" 65 #include "vmm_ioport.h" 66 #include "vmm_ktr.h" 67 #include "vmm_stat.h" 68 #include "vatpic.h" 69 #include "vlapic.h" 70 #include "vlapic_priv.h" 71 72 #include "ept.h" 73 #include "vmx_cpufunc.h" 74 #include "vmx.h" 75 #include "vmx_msr.h" 76 #include "x86.h" 77 #include "vmx_controls.h" 78 79 #define PINBASED_CTLS_ONE_SETTING \ 80 (PINBASED_EXTINT_EXITING | \ 81 PINBASED_NMI_EXITING | \ 82 PINBASED_VIRTUAL_NMI) 83 #define PINBASED_CTLS_ZERO_SETTING 0 84 85 #define PROCBASED_CTLS_WINDOW_SETTING \ 86 (PROCBASED_INT_WINDOW_EXITING | \ 87 PROCBASED_NMI_WINDOW_EXITING) 88 89 #define PROCBASED_CTLS_ONE_SETTING \ 90 (PROCBASED_SECONDARY_CONTROLS | \ 91 PROCBASED_MWAIT_EXITING | \ 92 PROCBASED_MONITOR_EXITING | \ 93 PROCBASED_IO_EXITING | \ 94 PROCBASED_MSR_BITMAPS | \ 95 PROCBASED_CTLS_WINDOW_SETTING | \ 96 PROCBASED_CR8_LOAD_EXITING | \ 97 PROCBASED_CR8_STORE_EXITING) 98 #define PROCBASED_CTLS_ZERO_SETTING \ 99 (PROCBASED_CR3_LOAD_EXITING | \ 100 PROCBASED_CR3_STORE_EXITING | \ 101 PROCBASED_IO_BITMAPS) 102 103 #define PROCBASED_CTLS2_ONE_SETTING PROCBASED2_ENABLE_EPT 104 #define PROCBASED_CTLS2_ZERO_SETTING 0 105 106 #define VM_EXIT_CTLS_ONE_SETTING \ 107 (VM_EXIT_SAVE_DEBUG_CONTROLS | \ 108 VM_EXIT_HOST_LMA | \ 109 VM_EXIT_SAVE_EFER | \ 110 VM_EXIT_LOAD_EFER | \ 111 VM_EXIT_ACKNOWLEDGE_INTERRUPT) 112 113 #define VM_EXIT_CTLS_ZERO_SETTING 0 114 115 #define VM_ENTRY_CTLS_ONE_SETTING \ 116 (VM_ENTRY_LOAD_DEBUG_CONTROLS | \ 117 VM_ENTRY_LOAD_EFER) 118 119 #define VM_ENTRY_CTLS_ZERO_SETTING \ 120 (VM_ENTRY_INTO_SMM | \ 121 VM_ENTRY_DEACTIVATE_DUAL_MONITOR) 122 123 #define HANDLED 1 124 #define UNHANDLED 0 125 126 static MALLOC_DEFINE(M_VMX, "vmx", "vmx"); 127 static MALLOC_DEFINE(M_VLAPIC, "vlapic", "vlapic"); 128 129 SYSCTL_DECL(_hw_vmm); 130 SYSCTL_NODE(_hw_vmm, OID_AUTO, vmx, CTLFLAG_RW | CTLFLAG_MPSAFE, NULL, 131 NULL); 132 133 int vmxon_enabled[MAXCPU]; 134 static char vmxon_region[MAXCPU][PAGE_SIZE] __aligned(PAGE_SIZE); 135 136 static uint32_t pinbased_ctls, procbased_ctls, procbased_ctls2; 137 static uint32_t exit_ctls, entry_ctls; 138 139 static uint64_t cr0_ones_mask, cr0_zeros_mask; 140 SYSCTL_ULONG(_hw_vmm_vmx, OID_AUTO, cr0_ones_mask, CTLFLAG_RD, 141 &cr0_ones_mask, 0, NULL); 142 SYSCTL_ULONG(_hw_vmm_vmx, OID_AUTO, cr0_zeros_mask, CTLFLAG_RD, 143 &cr0_zeros_mask, 0, NULL); 144 145 static uint64_t cr4_ones_mask, cr4_zeros_mask; 146 SYSCTL_ULONG(_hw_vmm_vmx, OID_AUTO, cr4_ones_mask, CTLFLAG_RD, 147 &cr4_ones_mask, 0, NULL); 148 SYSCTL_ULONG(_hw_vmm_vmx, OID_AUTO, cr4_zeros_mask, CTLFLAG_RD, 149 &cr4_zeros_mask, 0, NULL); 150 151 static int vmx_initialized; 152 SYSCTL_INT(_hw_vmm_vmx, OID_AUTO, initialized, CTLFLAG_RD, 153 &vmx_initialized, 0, "Intel VMX initialized"); 154 155 /* 156 * Optional capabilities 157 */ 158 static SYSCTL_NODE(_hw_vmm_vmx, OID_AUTO, cap, 159 CTLFLAG_RW | CTLFLAG_MPSAFE, NULL, 160 NULL); 161 162 static int cap_halt_exit; 163 SYSCTL_INT(_hw_vmm_vmx_cap, OID_AUTO, halt_exit, CTLFLAG_RD, &cap_halt_exit, 0, 164 "HLT triggers a VM-exit"); 165 166 static int cap_pause_exit; 167 SYSCTL_INT(_hw_vmm_vmx_cap, OID_AUTO, pause_exit, CTLFLAG_RD, &cap_pause_exit, 168 0, "PAUSE triggers a VM-exit"); 169 170 static int cap_rdpid; 171 SYSCTL_INT(_hw_vmm_vmx_cap, OID_AUTO, rdpid, CTLFLAG_RD, &cap_rdpid, 0, 172 "Guests are allowed to use RDPID"); 173 174 static int cap_rdtscp; 175 SYSCTL_INT(_hw_vmm_vmx_cap, OID_AUTO, rdtscp, CTLFLAG_RD, &cap_rdtscp, 0, 176 "Guests are allowed to use RDTSCP"); 177 178 static int cap_unrestricted_guest; 179 SYSCTL_INT(_hw_vmm_vmx_cap, OID_AUTO, unrestricted_guest, CTLFLAG_RD, 180 &cap_unrestricted_guest, 0, "Unrestricted guests"); 181 182 static int cap_monitor_trap; 183 SYSCTL_INT(_hw_vmm_vmx_cap, OID_AUTO, monitor_trap, CTLFLAG_RD, 184 &cap_monitor_trap, 0, "Monitor trap flag"); 185 186 static int cap_invpcid; 187 SYSCTL_INT(_hw_vmm_vmx_cap, OID_AUTO, invpcid, CTLFLAG_RD, &cap_invpcid, 188 0, "Guests are allowed to use INVPCID"); 189 190 static int tpr_shadowing; 191 SYSCTL_INT(_hw_vmm_vmx_cap, OID_AUTO, tpr_shadowing, CTLFLAG_RD, 192 &tpr_shadowing, 0, "TPR shadowing support"); 193 194 static int virtual_interrupt_delivery; 195 SYSCTL_INT(_hw_vmm_vmx_cap, OID_AUTO, virtual_interrupt_delivery, CTLFLAG_RD, 196 &virtual_interrupt_delivery, 0, "APICv virtual interrupt delivery support"); 197 198 static int posted_interrupts; 199 SYSCTL_INT(_hw_vmm_vmx_cap, OID_AUTO, posted_interrupts, CTLFLAG_RD, 200 &posted_interrupts, 0, "APICv posted interrupt support"); 201 202 static int pirvec = -1; 203 SYSCTL_INT(_hw_vmm_vmx, OID_AUTO, posted_interrupt_vector, CTLFLAG_RD, 204 &pirvec, 0, "APICv posted interrupt vector"); 205 206 static struct unrhdr *vpid_unr; 207 static u_int vpid_alloc_failed; 208 SYSCTL_UINT(_hw_vmm_vmx, OID_AUTO, vpid_alloc_failed, CTLFLAG_RD, 209 &vpid_alloc_failed, 0, NULL); 210 211 int guest_l1d_flush; 212 SYSCTL_INT(_hw_vmm_vmx, OID_AUTO, l1d_flush, CTLFLAG_RD, 213 &guest_l1d_flush, 0, NULL); 214 int guest_l1d_flush_sw; 215 SYSCTL_INT(_hw_vmm_vmx, OID_AUTO, l1d_flush_sw, CTLFLAG_RD, 216 &guest_l1d_flush_sw, 0, NULL); 217 218 static struct msr_entry msr_load_list[1] __aligned(16); 219 220 /* 221 * The definitions of SDT probes for VMX. 222 */ 223 224 SDT_PROBE_DEFINE3(vmm, vmx, exit, entry, 225 "struct vmx *", "int", "struct vm_exit *"); 226 227 SDT_PROBE_DEFINE4(vmm, vmx, exit, taskswitch, 228 "struct vmx *", "int", "struct vm_exit *", "struct vm_task_switch *"); 229 230 SDT_PROBE_DEFINE4(vmm, vmx, exit, craccess, 231 "struct vmx *", "int", "struct vm_exit *", "uint64_t"); 232 233 SDT_PROBE_DEFINE4(vmm, vmx, exit, rdmsr, 234 "struct vmx *", "int", "struct vm_exit *", "uint32_t"); 235 236 SDT_PROBE_DEFINE5(vmm, vmx, exit, wrmsr, 237 "struct vmx *", "int", "struct vm_exit *", "uint32_t", "uint64_t"); 238 239 SDT_PROBE_DEFINE3(vmm, vmx, exit, halt, 240 "struct vmx *", "int", "struct vm_exit *"); 241 242 SDT_PROBE_DEFINE3(vmm, vmx, exit, mtrap, 243 "struct vmx *", "int", "struct vm_exit *"); 244 245 SDT_PROBE_DEFINE3(vmm, vmx, exit, pause, 246 "struct vmx *", "int", "struct vm_exit *"); 247 248 SDT_PROBE_DEFINE3(vmm, vmx, exit, intrwindow, 249 "struct vmx *", "int", "struct vm_exit *"); 250 251 SDT_PROBE_DEFINE4(vmm, vmx, exit, interrupt, 252 "struct vmx *", "int", "struct vm_exit *", "uint32_t"); 253 254 SDT_PROBE_DEFINE3(vmm, vmx, exit, nmiwindow, 255 "struct vmx *", "int", "struct vm_exit *"); 256 257 SDT_PROBE_DEFINE3(vmm, vmx, exit, inout, 258 "struct vmx *", "int", "struct vm_exit *"); 259 260 SDT_PROBE_DEFINE3(vmm, vmx, exit, cpuid, 261 "struct vmx *", "int", "struct vm_exit *"); 262 263 SDT_PROBE_DEFINE5(vmm, vmx, exit, exception, 264 "struct vmx *", "int", "struct vm_exit *", "uint32_t", "int"); 265 266 SDT_PROBE_DEFINE5(vmm, vmx, exit, nestedfault, 267 "struct vmx *", "int", "struct vm_exit *", "uint64_t", "uint64_t"); 268 269 SDT_PROBE_DEFINE4(vmm, vmx, exit, mmiofault, 270 "struct vmx *", "int", "struct vm_exit *", "uint64_t"); 271 272 SDT_PROBE_DEFINE3(vmm, vmx, exit, eoi, 273 "struct vmx *", "int", "struct vm_exit *"); 274 275 SDT_PROBE_DEFINE3(vmm, vmx, exit, apicaccess, 276 "struct vmx *", "int", "struct vm_exit *"); 277 278 SDT_PROBE_DEFINE4(vmm, vmx, exit, apicwrite, 279 "struct vmx *", "int", "struct vm_exit *", "struct vlapic *"); 280 281 SDT_PROBE_DEFINE3(vmm, vmx, exit, xsetbv, 282 "struct vmx *", "int", "struct vm_exit *"); 283 284 SDT_PROBE_DEFINE3(vmm, vmx, exit, monitor, 285 "struct vmx *", "int", "struct vm_exit *"); 286 287 SDT_PROBE_DEFINE3(vmm, vmx, exit, mwait, 288 "struct vmx *", "int", "struct vm_exit *"); 289 290 SDT_PROBE_DEFINE3(vmm, vmx, exit, vminsn, 291 "struct vmx *", "int", "struct vm_exit *"); 292 293 SDT_PROBE_DEFINE4(vmm, vmx, exit, unknown, 294 "struct vmx *", "int", "struct vm_exit *", "uint32_t"); 295 296 SDT_PROBE_DEFINE4(vmm, vmx, exit, return, 297 "struct vmx *", "int", "struct vm_exit *", "int"); 298 299 /* 300 * Use the last page below 4GB as the APIC access address. This address is 301 * occupied by the boot firmware so it is guaranteed that it will not conflict 302 * with a page in system memory. 303 */ 304 #define APIC_ACCESS_ADDRESS 0xFFFFF000 305 306 static int vmx_getdesc(void *arg, int vcpu, int reg, struct seg_desc *desc); 307 static int vmx_getreg(void *arg, int vcpu, int reg, uint64_t *retval); 308 static int vmxctx_setreg(struct vmxctx *vmxctx, int reg, uint64_t val); 309 static void vmx_inject_pir(struct vlapic *vlapic); 310 #ifdef BHYVE_SNAPSHOT 311 static int vmx_restore_tsc(void *arg, int vcpu, uint64_t now); 312 #endif 313 314 static inline bool 315 host_has_rdpid(void) 316 { 317 return ((cpu_stdext_feature2 & CPUID_STDEXT2_RDPID) != 0); 318 } 319 320 static inline bool 321 host_has_rdtscp(void) 322 { 323 return ((amd_feature & AMDID_RDTSCP) != 0); 324 } 325 326 #ifdef KTR 327 static const char * 328 exit_reason_to_str(int reason) 329 { 330 static char reasonbuf[32]; 331 332 switch (reason) { 333 case EXIT_REASON_EXCEPTION: 334 return "exception"; 335 case EXIT_REASON_EXT_INTR: 336 return "extint"; 337 case EXIT_REASON_TRIPLE_FAULT: 338 return "triplefault"; 339 case EXIT_REASON_INIT: 340 return "init"; 341 case EXIT_REASON_SIPI: 342 return "sipi"; 343 case EXIT_REASON_IO_SMI: 344 return "iosmi"; 345 case EXIT_REASON_SMI: 346 return "smi"; 347 case EXIT_REASON_INTR_WINDOW: 348 return "intrwindow"; 349 case EXIT_REASON_NMI_WINDOW: 350 return "nmiwindow"; 351 case EXIT_REASON_TASK_SWITCH: 352 return "taskswitch"; 353 case EXIT_REASON_CPUID: 354 return "cpuid"; 355 case EXIT_REASON_GETSEC: 356 return "getsec"; 357 case EXIT_REASON_HLT: 358 return "hlt"; 359 case EXIT_REASON_INVD: 360 return "invd"; 361 case EXIT_REASON_INVLPG: 362 return "invlpg"; 363 case EXIT_REASON_RDPMC: 364 return "rdpmc"; 365 case EXIT_REASON_RDTSC: 366 return "rdtsc"; 367 case EXIT_REASON_RSM: 368 return "rsm"; 369 case EXIT_REASON_VMCALL: 370 return "vmcall"; 371 case EXIT_REASON_VMCLEAR: 372 return "vmclear"; 373 case EXIT_REASON_VMLAUNCH: 374 return "vmlaunch"; 375 case EXIT_REASON_VMPTRLD: 376 return "vmptrld"; 377 case EXIT_REASON_VMPTRST: 378 return "vmptrst"; 379 case EXIT_REASON_VMREAD: 380 return "vmread"; 381 case EXIT_REASON_VMRESUME: 382 return "vmresume"; 383 case EXIT_REASON_VMWRITE: 384 return "vmwrite"; 385 case EXIT_REASON_VMXOFF: 386 return "vmxoff"; 387 case EXIT_REASON_VMXON: 388 return "vmxon"; 389 case EXIT_REASON_CR_ACCESS: 390 return "craccess"; 391 case EXIT_REASON_DR_ACCESS: 392 return "draccess"; 393 case EXIT_REASON_INOUT: 394 return "inout"; 395 case EXIT_REASON_RDMSR: 396 return "rdmsr"; 397 case EXIT_REASON_WRMSR: 398 return "wrmsr"; 399 case EXIT_REASON_INVAL_VMCS: 400 return "invalvmcs"; 401 case EXIT_REASON_INVAL_MSR: 402 return "invalmsr"; 403 case EXIT_REASON_MWAIT: 404 return "mwait"; 405 case EXIT_REASON_MTF: 406 return "mtf"; 407 case EXIT_REASON_MONITOR: 408 return "monitor"; 409 case EXIT_REASON_PAUSE: 410 return "pause"; 411 case EXIT_REASON_MCE_DURING_ENTRY: 412 return "mce-during-entry"; 413 case EXIT_REASON_TPR: 414 return "tpr"; 415 case EXIT_REASON_APIC_ACCESS: 416 return "apic-access"; 417 case EXIT_REASON_GDTR_IDTR: 418 return "gdtridtr"; 419 case EXIT_REASON_LDTR_TR: 420 return "ldtrtr"; 421 case EXIT_REASON_EPT_FAULT: 422 return "eptfault"; 423 case EXIT_REASON_EPT_MISCONFIG: 424 return "eptmisconfig"; 425 case EXIT_REASON_INVEPT: 426 return "invept"; 427 case EXIT_REASON_RDTSCP: 428 return "rdtscp"; 429 case EXIT_REASON_VMX_PREEMPT: 430 return "vmxpreempt"; 431 case EXIT_REASON_INVVPID: 432 return "invvpid"; 433 case EXIT_REASON_WBINVD: 434 return "wbinvd"; 435 case EXIT_REASON_XSETBV: 436 return "xsetbv"; 437 case EXIT_REASON_APIC_WRITE: 438 return "apic-write"; 439 default: 440 snprintf(reasonbuf, sizeof(reasonbuf), "%d", reason); 441 return (reasonbuf); 442 } 443 } 444 #endif /* KTR */ 445 446 static int 447 vmx_allow_x2apic_msrs(struct vmx *vmx) 448 { 449 int i, error; 450 451 error = 0; 452 453 /* 454 * Allow readonly access to the following x2APIC MSRs from the guest. 455 */ 456 error += guest_msr_ro(vmx, MSR_APIC_ID); 457 error += guest_msr_ro(vmx, MSR_APIC_VERSION); 458 error += guest_msr_ro(vmx, MSR_APIC_LDR); 459 error += guest_msr_ro(vmx, MSR_APIC_SVR); 460 461 for (i = 0; i < 8; i++) 462 error += guest_msr_ro(vmx, MSR_APIC_ISR0 + i); 463 464 for (i = 0; i < 8; i++) 465 error += guest_msr_ro(vmx, MSR_APIC_TMR0 + i); 466 467 for (i = 0; i < 8; i++) 468 error += guest_msr_ro(vmx, MSR_APIC_IRR0 + i); 469 470 error += guest_msr_ro(vmx, MSR_APIC_ESR); 471 error += guest_msr_ro(vmx, MSR_APIC_LVT_TIMER); 472 error += guest_msr_ro(vmx, MSR_APIC_LVT_THERMAL); 473 error += guest_msr_ro(vmx, MSR_APIC_LVT_PCINT); 474 error += guest_msr_ro(vmx, MSR_APIC_LVT_LINT0); 475 error += guest_msr_ro(vmx, MSR_APIC_LVT_LINT1); 476 error += guest_msr_ro(vmx, MSR_APIC_LVT_ERROR); 477 error += guest_msr_ro(vmx, MSR_APIC_ICR_TIMER); 478 error += guest_msr_ro(vmx, MSR_APIC_DCR_TIMER); 479 error += guest_msr_ro(vmx, MSR_APIC_ICR); 480 481 /* 482 * Allow TPR, EOI and SELF_IPI MSRs to be read and written by the guest. 483 * 484 * These registers get special treatment described in the section 485 * "Virtualizing MSR-Based APIC Accesses". 486 */ 487 error += guest_msr_rw(vmx, MSR_APIC_TPR); 488 error += guest_msr_rw(vmx, MSR_APIC_EOI); 489 error += guest_msr_rw(vmx, MSR_APIC_SELF_IPI); 490 491 return (error); 492 } 493 494 u_long 495 vmx_fix_cr0(u_long cr0) 496 { 497 498 return ((cr0 | cr0_ones_mask) & ~cr0_zeros_mask); 499 } 500 501 u_long 502 vmx_fix_cr4(u_long cr4) 503 { 504 505 return ((cr4 | cr4_ones_mask) & ~cr4_zeros_mask); 506 } 507 508 static void 509 vpid_free(int vpid) 510 { 511 if (vpid < 0 || vpid > 0xffff) 512 panic("vpid_free: invalid vpid %d", vpid); 513 514 /* 515 * VPIDs [0,VM_MAXCPU] are special and are not allocated from 516 * the unit number allocator. 517 */ 518 519 if (vpid > VM_MAXCPU) 520 free_unr(vpid_unr, vpid); 521 } 522 523 static void 524 vpid_alloc(uint16_t *vpid, int num) 525 { 526 int i, x; 527 528 if (num <= 0 || num > VM_MAXCPU) 529 panic("invalid number of vpids requested: %d", num); 530 531 /* 532 * If the "enable vpid" execution control is not enabled then the 533 * VPID is required to be 0 for all vcpus. 534 */ 535 if ((procbased_ctls2 & PROCBASED2_ENABLE_VPID) == 0) { 536 for (i = 0; i < num; i++) 537 vpid[i] = 0; 538 return; 539 } 540 541 /* 542 * Allocate a unique VPID for each vcpu from the unit number allocator. 543 */ 544 for (i = 0; i < num; i++) { 545 x = alloc_unr(vpid_unr); 546 if (x == -1) 547 break; 548 else 549 vpid[i] = x; 550 } 551 552 if (i < num) { 553 atomic_add_int(&vpid_alloc_failed, 1); 554 555 /* 556 * If the unit number allocator does not have enough unique 557 * VPIDs then we need to allocate from the [1,VM_MAXCPU] range. 558 * 559 * These VPIDs are not be unique across VMs but this does not 560 * affect correctness because the combined mappings are also 561 * tagged with the EP4TA which is unique for each VM. 562 * 563 * It is still sub-optimal because the invvpid will invalidate 564 * combined mappings for a particular VPID across all EP4TAs. 565 */ 566 while (i-- > 0) 567 vpid_free(vpid[i]); 568 569 for (i = 0; i < num; i++) 570 vpid[i] = i + 1; 571 } 572 } 573 574 static void 575 vpid_init(void) 576 { 577 /* 578 * VPID 0 is required when the "enable VPID" execution control is 579 * disabled. 580 * 581 * VPIDs [1,VM_MAXCPU] are used as the "overflow namespace" when the 582 * unit number allocator does not have sufficient unique VPIDs to 583 * satisfy the allocation. 584 * 585 * The remaining VPIDs are managed by the unit number allocator. 586 */ 587 vpid_unr = new_unrhdr(VM_MAXCPU + 1, 0xffff, NULL); 588 } 589 590 static void 591 vmx_disable(void *arg __unused) 592 { 593 struct invvpid_desc invvpid_desc = { 0 }; 594 struct invept_desc invept_desc = { 0 }; 595 596 if (vmxon_enabled[curcpu]) { 597 /* 598 * See sections 25.3.3.3 and 25.3.3.4 in Intel Vol 3b. 599 * 600 * VMXON or VMXOFF are not required to invalidate any TLB 601 * caching structures. This prevents potential retention of 602 * cached information in the TLB between distinct VMX episodes. 603 */ 604 invvpid(INVVPID_TYPE_ALL_CONTEXTS, invvpid_desc); 605 invept(INVEPT_TYPE_ALL_CONTEXTS, invept_desc); 606 vmxoff(); 607 } 608 load_cr4(rcr4() & ~CR4_VMXE); 609 } 610 611 static int 612 vmx_cleanup(void) 613 { 614 615 if (pirvec >= 0) 616 lapic_ipi_free(pirvec); 617 618 if (vpid_unr != NULL) { 619 delete_unrhdr(vpid_unr); 620 vpid_unr = NULL; 621 } 622 623 if (nmi_flush_l1d_sw == 1) 624 nmi_flush_l1d_sw = 0; 625 626 smp_rendezvous(NULL, vmx_disable, NULL, NULL); 627 628 return (0); 629 } 630 631 static void 632 vmx_enable(void *arg __unused) 633 { 634 int error; 635 uint64_t feature_control; 636 637 feature_control = rdmsr(MSR_IA32_FEATURE_CONTROL); 638 if ((feature_control & IA32_FEATURE_CONTROL_LOCK) == 0 || 639 (feature_control & IA32_FEATURE_CONTROL_VMX_EN) == 0) { 640 wrmsr(MSR_IA32_FEATURE_CONTROL, 641 feature_control | IA32_FEATURE_CONTROL_VMX_EN | 642 IA32_FEATURE_CONTROL_LOCK); 643 } 644 645 load_cr4(rcr4() | CR4_VMXE); 646 647 *(uint32_t *)vmxon_region[curcpu] = vmx_revision(); 648 error = vmxon(vmxon_region[curcpu]); 649 if (error == 0) 650 vmxon_enabled[curcpu] = 1; 651 } 652 653 static void 654 vmx_restore(void) 655 { 656 657 if (vmxon_enabled[curcpu]) 658 vmxon(vmxon_region[curcpu]); 659 } 660 661 static int 662 vmx_init(int ipinum) 663 { 664 int error; 665 uint64_t basic, fixed0, fixed1, feature_control; 666 uint32_t tmp, procbased2_vid_bits; 667 668 /* CPUID.1:ECX[bit 5] must be 1 for processor to support VMX */ 669 if (!(cpu_feature2 & CPUID2_VMX)) { 670 printf("vmx_init: processor does not support VMX operation\n"); 671 return (ENXIO); 672 } 673 674 /* 675 * Verify that MSR_IA32_FEATURE_CONTROL lock and VMXON enable bits 676 * are set (bits 0 and 2 respectively). 677 */ 678 feature_control = rdmsr(MSR_IA32_FEATURE_CONTROL); 679 if ((feature_control & IA32_FEATURE_CONTROL_LOCK) == 1 && 680 (feature_control & IA32_FEATURE_CONTROL_VMX_EN) == 0) { 681 printf("vmx_init: VMX operation disabled by BIOS\n"); 682 return (ENXIO); 683 } 684 685 /* 686 * Verify capabilities MSR_VMX_BASIC: 687 * - bit 54 indicates support for INS/OUTS decoding 688 */ 689 basic = rdmsr(MSR_VMX_BASIC); 690 if ((basic & (1UL << 54)) == 0) { 691 printf("vmx_init: processor does not support desired basic " 692 "capabilities\n"); 693 return (EINVAL); 694 } 695 696 /* Check support for primary processor-based VM-execution controls */ 697 error = vmx_set_ctlreg(MSR_VMX_PROCBASED_CTLS, 698 MSR_VMX_TRUE_PROCBASED_CTLS, 699 PROCBASED_CTLS_ONE_SETTING, 700 PROCBASED_CTLS_ZERO_SETTING, &procbased_ctls); 701 if (error) { 702 printf("vmx_init: processor does not support desired primary " 703 "processor-based controls\n"); 704 return (error); 705 } 706 707 /* Clear the processor-based ctl bits that are set on demand */ 708 procbased_ctls &= ~PROCBASED_CTLS_WINDOW_SETTING; 709 710 /* Check support for secondary processor-based VM-execution controls */ 711 error = vmx_set_ctlreg(MSR_VMX_PROCBASED_CTLS2, 712 MSR_VMX_PROCBASED_CTLS2, 713 PROCBASED_CTLS2_ONE_SETTING, 714 PROCBASED_CTLS2_ZERO_SETTING, &procbased_ctls2); 715 if (error) { 716 printf("vmx_init: processor does not support desired secondary " 717 "processor-based controls\n"); 718 return (error); 719 } 720 721 /* Check support for VPID */ 722 error = vmx_set_ctlreg(MSR_VMX_PROCBASED_CTLS2, MSR_VMX_PROCBASED_CTLS2, 723 PROCBASED2_ENABLE_VPID, 0, &tmp); 724 if (error == 0) 725 procbased_ctls2 |= PROCBASED2_ENABLE_VPID; 726 727 /* Check support for pin-based VM-execution controls */ 728 error = vmx_set_ctlreg(MSR_VMX_PINBASED_CTLS, 729 MSR_VMX_TRUE_PINBASED_CTLS, 730 PINBASED_CTLS_ONE_SETTING, 731 PINBASED_CTLS_ZERO_SETTING, &pinbased_ctls); 732 if (error) { 733 printf("vmx_init: processor does not support desired " 734 "pin-based controls\n"); 735 return (error); 736 } 737 738 /* Check support for VM-exit controls */ 739 error = vmx_set_ctlreg(MSR_VMX_EXIT_CTLS, MSR_VMX_TRUE_EXIT_CTLS, 740 VM_EXIT_CTLS_ONE_SETTING, 741 VM_EXIT_CTLS_ZERO_SETTING, 742 &exit_ctls); 743 if (error) { 744 printf("vmx_init: processor does not support desired " 745 "exit controls\n"); 746 return (error); 747 } 748 749 /* Check support for VM-entry controls */ 750 error = vmx_set_ctlreg(MSR_VMX_ENTRY_CTLS, MSR_VMX_TRUE_ENTRY_CTLS, 751 VM_ENTRY_CTLS_ONE_SETTING, VM_ENTRY_CTLS_ZERO_SETTING, 752 &entry_ctls); 753 if (error) { 754 printf("vmx_init: processor does not support desired " 755 "entry controls\n"); 756 return (error); 757 } 758 759 /* 760 * Check support for optional features by testing them 761 * as individual bits 762 */ 763 cap_halt_exit = (vmx_set_ctlreg(MSR_VMX_PROCBASED_CTLS, 764 MSR_VMX_TRUE_PROCBASED_CTLS, 765 PROCBASED_HLT_EXITING, 0, 766 &tmp) == 0); 767 768 cap_monitor_trap = (vmx_set_ctlreg(MSR_VMX_PROCBASED_CTLS, 769 MSR_VMX_PROCBASED_CTLS, 770 PROCBASED_MTF, 0, 771 &tmp) == 0); 772 773 cap_pause_exit = (vmx_set_ctlreg(MSR_VMX_PROCBASED_CTLS, 774 MSR_VMX_TRUE_PROCBASED_CTLS, 775 PROCBASED_PAUSE_EXITING, 0, 776 &tmp) == 0); 777 778 /* 779 * Check support for RDPID and/or RDTSCP. 780 * 781 * Support a pass-through-based implementation of these via the 782 * "enable RDTSCP" VM-execution control and the "RDTSC exiting" 783 * VM-execution control. 784 * 785 * The "enable RDTSCP" VM-execution control applies to both RDPID 786 * and RDTSCP (see SDM volume 3, section 25.3, "Changes to 787 * Instruction Behavior in VMX Non-root operation"); this is why 788 * only this VM-execution control needs to be enabled in order to 789 * enable passing through whichever of RDPID and/or RDTSCP are 790 * supported by the host. 791 * 792 * The "RDTSC exiting" VM-execution control applies to both RDTSC 793 * and RDTSCP (again, per SDM volume 3, section 25.3), and is 794 * already set up for RDTSC and RDTSCP pass-through by the current 795 * implementation of RDTSC. 796 * 797 * Although RDPID and RDTSCP are optional capabilities, since there 798 * does not currently seem to be a use case for enabling/disabling 799 * these via libvmmapi, choose not to support this and, instead, 800 * just statically always enable or always disable this support 801 * across all vCPUs on all VMs. (Note that there may be some 802 * complications to providing this functionality, e.g., the MSR 803 * bitmap is currently per-VM rather than per-vCPU while the 804 * capability API wants to be able to control capabilities on a 805 * per-vCPU basis). 806 */ 807 error = vmx_set_ctlreg(MSR_VMX_PROCBASED_CTLS2, 808 MSR_VMX_PROCBASED_CTLS2, 809 PROCBASED2_ENABLE_RDTSCP, 0, &tmp); 810 cap_rdpid = error == 0 && host_has_rdpid(); 811 cap_rdtscp = error == 0 && host_has_rdtscp(); 812 if (cap_rdpid || cap_rdtscp) 813 procbased_ctls2 |= PROCBASED2_ENABLE_RDTSCP; 814 815 cap_unrestricted_guest = (vmx_set_ctlreg(MSR_VMX_PROCBASED_CTLS2, 816 MSR_VMX_PROCBASED_CTLS2, 817 PROCBASED2_UNRESTRICTED_GUEST, 0, 818 &tmp) == 0); 819 820 cap_invpcid = (vmx_set_ctlreg(MSR_VMX_PROCBASED_CTLS2, 821 MSR_VMX_PROCBASED_CTLS2, PROCBASED2_ENABLE_INVPCID, 0, 822 &tmp) == 0); 823 824 /* 825 * Check support for TPR shadow. 826 */ 827 error = vmx_set_ctlreg(MSR_VMX_PROCBASED_CTLS, 828 MSR_VMX_TRUE_PROCBASED_CTLS, PROCBASED_USE_TPR_SHADOW, 0, 829 &tmp); 830 if (error == 0) { 831 tpr_shadowing = 1; 832 TUNABLE_INT_FETCH("hw.vmm.vmx.use_tpr_shadowing", 833 &tpr_shadowing); 834 } 835 836 if (tpr_shadowing) { 837 procbased_ctls |= PROCBASED_USE_TPR_SHADOW; 838 procbased_ctls &= ~PROCBASED_CR8_LOAD_EXITING; 839 procbased_ctls &= ~PROCBASED_CR8_STORE_EXITING; 840 } 841 842 /* 843 * Check support for virtual interrupt delivery. 844 */ 845 procbased2_vid_bits = (PROCBASED2_VIRTUALIZE_APIC_ACCESSES | 846 PROCBASED2_VIRTUALIZE_X2APIC_MODE | 847 PROCBASED2_APIC_REGISTER_VIRTUALIZATION | 848 PROCBASED2_VIRTUAL_INTERRUPT_DELIVERY); 849 850 error = vmx_set_ctlreg(MSR_VMX_PROCBASED_CTLS2, MSR_VMX_PROCBASED_CTLS2, 851 procbased2_vid_bits, 0, &tmp); 852 if (error == 0 && tpr_shadowing) { 853 virtual_interrupt_delivery = 1; 854 TUNABLE_INT_FETCH("hw.vmm.vmx.use_apic_vid", 855 &virtual_interrupt_delivery); 856 } 857 858 if (virtual_interrupt_delivery) { 859 procbased_ctls |= PROCBASED_USE_TPR_SHADOW; 860 procbased_ctls2 |= procbased2_vid_bits; 861 procbased_ctls2 &= ~PROCBASED2_VIRTUALIZE_X2APIC_MODE; 862 863 /* 864 * Check for Posted Interrupts only if Virtual Interrupt 865 * Delivery is enabled. 866 */ 867 error = vmx_set_ctlreg(MSR_VMX_PINBASED_CTLS, 868 MSR_VMX_TRUE_PINBASED_CTLS, PINBASED_POSTED_INTERRUPT, 0, 869 &tmp); 870 if (error == 0) { 871 pirvec = lapic_ipi_alloc(pti ? &IDTVEC(justreturn1_pti) : 872 &IDTVEC(justreturn)); 873 if (pirvec < 0) { 874 if (bootverbose) { 875 printf("vmx_init: unable to allocate " 876 "posted interrupt vector\n"); 877 } 878 } else { 879 posted_interrupts = 1; 880 TUNABLE_INT_FETCH("hw.vmm.vmx.use_apic_pir", 881 &posted_interrupts); 882 } 883 } 884 } 885 886 if (posted_interrupts) 887 pinbased_ctls |= PINBASED_POSTED_INTERRUPT; 888 889 /* Initialize EPT */ 890 error = ept_init(ipinum); 891 if (error) { 892 printf("vmx_init: ept initialization failed (%d)\n", error); 893 return (error); 894 } 895 896 guest_l1d_flush = (cpu_ia32_arch_caps & 897 IA32_ARCH_CAP_SKIP_L1DFL_VMENTRY) == 0; 898 TUNABLE_INT_FETCH("hw.vmm.l1d_flush", &guest_l1d_flush); 899 900 /* 901 * L1D cache flush is enabled. Use IA32_FLUSH_CMD MSR when 902 * available. Otherwise fall back to the software flush 903 * method which loads enough data from the kernel text to 904 * flush existing L1D content, both on VMX entry and on NMI 905 * return. 906 */ 907 if (guest_l1d_flush) { 908 if ((cpu_stdext_feature3 & CPUID_STDEXT3_L1D_FLUSH) == 0) { 909 guest_l1d_flush_sw = 1; 910 TUNABLE_INT_FETCH("hw.vmm.l1d_flush_sw", 911 &guest_l1d_flush_sw); 912 } 913 if (guest_l1d_flush_sw) { 914 if (nmi_flush_l1d_sw <= 1) 915 nmi_flush_l1d_sw = 1; 916 } else { 917 msr_load_list[0].index = MSR_IA32_FLUSH_CMD; 918 msr_load_list[0].val = IA32_FLUSH_CMD_L1D; 919 } 920 } 921 922 /* 923 * Stash the cr0 and cr4 bits that must be fixed to 0 or 1 924 */ 925 fixed0 = rdmsr(MSR_VMX_CR0_FIXED0); 926 fixed1 = rdmsr(MSR_VMX_CR0_FIXED1); 927 cr0_ones_mask = fixed0 & fixed1; 928 cr0_zeros_mask = ~fixed0 & ~fixed1; 929 930 /* 931 * CR0_PE and CR0_PG can be set to zero in VMX non-root operation 932 * if unrestricted guest execution is allowed. 933 */ 934 if (cap_unrestricted_guest) 935 cr0_ones_mask &= ~(CR0_PG | CR0_PE); 936 937 /* 938 * Do not allow the guest to set CR0_NW or CR0_CD. 939 */ 940 cr0_zeros_mask |= (CR0_NW | CR0_CD); 941 942 fixed0 = rdmsr(MSR_VMX_CR4_FIXED0); 943 fixed1 = rdmsr(MSR_VMX_CR4_FIXED1); 944 cr4_ones_mask = fixed0 & fixed1; 945 cr4_zeros_mask = ~fixed0 & ~fixed1; 946 947 vpid_init(); 948 949 vmx_msr_init(); 950 951 /* enable VMX operation */ 952 smp_rendezvous(NULL, vmx_enable, NULL, NULL); 953 954 vmx_initialized = 1; 955 956 return (0); 957 } 958 959 static void 960 vmx_trigger_hostintr(int vector) 961 { 962 uintptr_t func; 963 struct gate_descriptor *gd; 964 965 gd = &idt[vector]; 966 967 KASSERT(vector >= 32 && vector <= 255, ("vmx_trigger_hostintr: " 968 "invalid vector %d", vector)); 969 KASSERT(gd->gd_p == 1, ("gate descriptor for vector %d not present", 970 vector)); 971 KASSERT(gd->gd_type == SDT_SYSIGT, ("gate descriptor for vector %d " 972 "has invalid type %d", vector, gd->gd_type)); 973 KASSERT(gd->gd_dpl == SEL_KPL, ("gate descriptor for vector %d " 974 "has invalid dpl %d", vector, gd->gd_dpl)); 975 KASSERT(gd->gd_selector == GSEL(GCODE_SEL, SEL_KPL), ("gate descriptor " 976 "for vector %d has invalid selector %d", vector, gd->gd_selector)); 977 KASSERT(gd->gd_ist == 0, ("gate descriptor for vector %d has invalid " 978 "IST %d", vector, gd->gd_ist)); 979 980 func = ((long)gd->gd_hioffset << 16 | gd->gd_looffset); 981 vmx_call_isr(func); 982 } 983 984 static int 985 vmx_setup_cr_shadow(int which, struct vmcs *vmcs, uint32_t initial) 986 { 987 int error, mask_ident, shadow_ident; 988 uint64_t mask_value; 989 990 if (which != 0 && which != 4) 991 panic("vmx_setup_cr_shadow: unknown cr%d", which); 992 993 if (which == 0) { 994 mask_ident = VMCS_CR0_MASK; 995 mask_value = cr0_ones_mask | cr0_zeros_mask; 996 shadow_ident = VMCS_CR0_SHADOW; 997 } else { 998 mask_ident = VMCS_CR4_MASK; 999 mask_value = cr4_ones_mask | cr4_zeros_mask; 1000 shadow_ident = VMCS_CR4_SHADOW; 1001 } 1002 1003 error = vmcs_setreg(vmcs, 0, VMCS_IDENT(mask_ident), mask_value); 1004 if (error) 1005 return (error); 1006 1007 error = vmcs_setreg(vmcs, 0, VMCS_IDENT(shadow_ident), initial); 1008 if (error) 1009 return (error); 1010 1011 return (0); 1012 } 1013 #define vmx_setup_cr0_shadow(vmcs,init) vmx_setup_cr_shadow(0, (vmcs), (init)) 1014 #define vmx_setup_cr4_shadow(vmcs,init) vmx_setup_cr_shadow(4, (vmcs), (init)) 1015 1016 static void * 1017 vmx_vminit(struct vm *vm, pmap_t pmap) 1018 { 1019 uint16_t vpid[VM_MAXCPU]; 1020 int i, error; 1021 struct vmx *vmx; 1022 struct vmcs *vmcs; 1023 uint32_t exc_bitmap; 1024 uint16_t maxcpus; 1025 1026 vmx = malloc(sizeof(struct vmx), M_VMX, M_WAITOK | M_ZERO); 1027 if ((uintptr_t)vmx & PAGE_MASK) { 1028 panic("malloc of struct vmx not aligned on %d byte boundary", 1029 PAGE_SIZE); 1030 } 1031 vmx->vm = vm; 1032 1033 vmx->eptp = eptp(vtophys((vm_offset_t)pmap->pm_pmltop)); 1034 1035 /* 1036 * Clean up EPTP-tagged guest physical and combined mappings 1037 * 1038 * VMX transitions are not required to invalidate any guest physical 1039 * mappings. So, it may be possible for stale guest physical mappings 1040 * to be present in the processor TLBs. 1041 * 1042 * Combined mappings for this EP4TA are also invalidated for all VPIDs. 1043 */ 1044 ept_invalidate_mappings(vmx->eptp); 1045 1046 msr_bitmap_initialize(vmx->msr_bitmap); 1047 1048 /* 1049 * It is safe to allow direct access to MSR_GSBASE and MSR_FSBASE. 1050 * The guest FSBASE and GSBASE are saved and restored during 1051 * vm-exit and vm-entry respectively. The host FSBASE and GSBASE are 1052 * always restored from the vmcs host state area on vm-exit. 1053 * 1054 * The SYSENTER_CS/ESP/EIP MSRs are identical to FS/GSBASE in 1055 * how they are saved/restored so can be directly accessed by the 1056 * guest. 1057 * 1058 * MSR_EFER is saved and restored in the guest VMCS area on a 1059 * VM exit and entry respectively. It is also restored from the 1060 * host VMCS area on a VM exit. 1061 * 1062 * The TSC MSR is exposed read-only. Writes are disallowed as 1063 * that will impact the host TSC. If the guest does a write 1064 * the "use TSC offsetting" execution control is enabled and the 1065 * difference between the host TSC and the guest TSC is written 1066 * into the TSC offset in the VMCS. 1067 * 1068 * Guest TSC_AUX support is enabled if any of guest RDPID and/or 1069 * guest RDTSCP support are enabled (since, as per Table 2-2 in SDM 1070 * volume 4, TSC_AUX is supported if any of RDPID and/or RDTSCP are 1071 * supported). If guest TSC_AUX support is enabled, TSC_AUX is 1072 * exposed read-only so that the VMM can do one fewer MSR read per 1073 * exit than if this register were exposed read-write; the guest 1074 * restore value can be updated during guest writes (expected to be 1075 * rare) instead of during all exits (common). 1076 */ 1077 if (guest_msr_rw(vmx, MSR_GSBASE) || 1078 guest_msr_rw(vmx, MSR_FSBASE) || 1079 guest_msr_rw(vmx, MSR_SYSENTER_CS_MSR) || 1080 guest_msr_rw(vmx, MSR_SYSENTER_ESP_MSR) || 1081 guest_msr_rw(vmx, MSR_SYSENTER_EIP_MSR) || 1082 guest_msr_rw(vmx, MSR_EFER) || 1083 guest_msr_ro(vmx, MSR_TSC) || 1084 ((cap_rdpid || cap_rdtscp) && guest_msr_ro(vmx, MSR_TSC_AUX))) 1085 panic("vmx_vminit: error setting guest msr access"); 1086 1087 vpid_alloc(vpid, VM_MAXCPU); 1088 1089 if (virtual_interrupt_delivery) { 1090 error = vm_map_mmio(vm, DEFAULT_APIC_BASE, PAGE_SIZE, 1091 APIC_ACCESS_ADDRESS); 1092 /* XXX this should really return an error to the caller */ 1093 KASSERT(error == 0, ("vm_map_mmio(apicbase) error %d", error)); 1094 } 1095 1096 maxcpus = vm_get_maxcpus(vm); 1097 for (i = 0; i < maxcpus; i++) { 1098 vmcs = &vmx->vmcs[i]; 1099 vmcs->identifier = vmx_revision(); 1100 error = vmclear(vmcs); 1101 if (error != 0) { 1102 panic("vmx_vminit: vmclear error %d on vcpu %d\n", 1103 error, i); 1104 } 1105 1106 vmx_msr_guest_init(vmx, i); 1107 1108 error = vmcs_init(vmcs); 1109 KASSERT(error == 0, ("vmcs_init error %d", error)); 1110 1111 VMPTRLD(vmcs); 1112 error = 0; 1113 error += vmwrite(VMCS_HOST_RSP, (u_long)&vmx->ctx[i]); 1114 error += vmwrite(VMCS_EPTP, vmx->eptp); 1115 error += vmwrite(VMCS_PIN_BASED_CTLS, pinbased_ctls); 1116 error += vmwrite(VMCS_PRI_PROC_BASED_CTLS, procbased_ctls); 1117 error += vmwrite(VMCS_SEC_PROC_BASED_CTLS, procbased_ctls2); 1118 error += vmwrite(VMCS_EXIT_CTLS, exit_ctls); 1119 error += vmwrite(VMCS_ENTRY_CTLS, entry_ctls); 1120 error += vmwrite(VMCS_MSR_BITMAP, vtophys(vmx->msr_bitmap)); 1121 error += vmwrite(VMCS_VPID, vpid[i]); 1122 1123 if (guest_l1d_flush && !guest_l1d_flush_sw) { 1124 vmcs_write(VMCS_ENTRY_MSR_LOAD, pmap_kextract( 1125 (vm_offset_t)&msr_load_list[0])); 1126 vmcs_write(VMCS_ENTRY_MSR_LOAD_COUNT, 1127 nitems(msr_load_list)); 1128 vmcs_write(VMCS_EXIT_MSR_STORE, 0); 1129 vmcs_write(VMCS_EXIT_MSR_STORE_COUNT, 0); 1130 } 1131 1132 /* exception bitmap */ 1133 if (vcpu_trace_exceptions(vm, i)) 1134 exc_bitmap = 0xffffffff; 1135 else 1136 exc_bitmap = 1 << IDT_MC; 1137 error += vmwrite(VMCS_EXCEPTION_BITMAP, exc_bitmap); 1138 1139 vmx->ctx[i].guest_dr6 = DBREG_DR6_RESERVED1; 1140 error += vmwrite(VMCS_GUEST_DR7, DBREG_DR7_RESERVED1); 1141 1142 if (tpr_shadowing) { 1143 error += vmwrite(VMCS_VIRTUAL_APIC, 1144 vtophys(&vmx->apic_page[i])); 1145 } 1146 1147 if (virtual_interrupt_delivery) { 1148 error += vmwrite(VMCS_APIC_ACCESS, APIC_ACCESS_ADDRESS); 1149 error += vmwrite(VMCS_EOI_EXIT0, 0); 1150 error += vmwrite(VMCS_EOI_EXIT1, 0); 1151 error += vmwrite(VMCS_EOI_EXIT2, 0); 1152 error += vmwrite(VMCS_EOI_EXIT3, 0); 1153 } 1154 if (posted_interrupts) { 1155 error += vmwrite(VMCS_PIR_VECTOR, pirvec); 1156 error += vmwrite(VMCS_PIR_DESC, 1157 vtophys(&vmx->pir_desc[i])); 1158 } 1159 VMCLEAR(vmcs); 1160 KASSERT(error == 0, ("vmx_vminit: error customizing the vmcs")); 1161 1162 vmx->cap[i].set = 0; 1163 vmx->cap[i].set |= cap_rdpid != 0 ? 1 << VM_CAP_RDPID : 0; 1164 vmx->cap[i].set |= cap_rdtscp != 0 ? 1 << VM_CAP_RDTSCP : 0; 1165 vmx->cap[i].proc_ctls = procbased_ctls; 1166 vmx->cap[i].proc_ctls2 = procbased_ctls2; 1167 vmx->cap[i].exc_bitmap = exc_bitmap; 1168 1169 vmx->state[i].nextrip = ~0; 1170 vmx->state[i].lastcpu = NOCPU; 1171 vmx->state[i].vpid = vpid[i]; 1172 1173 /* 1174 * Set up the CR0/4 shadows, and init the read shadow 1175 * to the power-on register value from the Intel Sys Arch. 1176 * CR0 - 0x60000010 1177 * CR4 - 0 1178 */ 1179 error = vmx_setup_cr0_shadow(vmcs, 0x60000010); 1180 if (error != 0) 1181 panic("vmx_setup_cr0_shadow %d", error); 1182 1183 error = vmx_setup_cr4_shadow(vmcs, 0); 1184 if (error != 0) 1185 panic("vmx_setup_cr4_shadow %d", error); 1186 1187 vmx->ctx[i].pmap = pmap; 1188 } 1189 1190 return (vmx); 1191 } 1192 1193 static int 1194 vmx_handle_cpuid(struct vm *vm, int vcpu, struct vmxctx *vmxctx) 1195 { 1196 int handled; 1197 1198 handled = x86_emulate_cpuid(vm, vcpu, (uint64_t *)&vmxctx->guest_rax, 1199 (uint64_t *)&vmxctx->guest_rbx, (uint64_t *)&vmxctx->guest_rcx, 1200 (uint64_t *)&vmxctx->guest_rdx); 1201 return (handled); 1202 } 1203 1204 static __inline void 1205 vmx_run_trace(struct vmx *vmx, int vcpu) 1206 { 1207 #ifdef KTR 1208 VCPU_CTR1(vmx->vm, vcpu, "Resume execution at %#lx", vmcs_guest_rip()); 1209 #endif 1210 } 1211 1212 static __inline void 1213 vmx_exit_trace(struct vmx *vmx, int vcpu, uint64_t rip, uint32_t exit_reason, 1214 int handled) 1215 { 1216 #ifdef KTR 1217 VCPU_CTR3(vmx->vm, vcpu, "%s %s vmexit at 0x%0lx", 1218 handled ? "handled" : "unhandled", 1219 exit_reason_to_str(exit_reason), rip); 1220 #endif 1221 } 1222 1223 static __inline void 1224 vmx_astpending_trace(struct vmx *vmx, int vcpu, uint64_t rip) 1225 { 1226 #ifdef KTR 1227 VCPU_CTR1(vmx->vm, vcpu, "astpending vmexit at 0x%0lx", rip); 1228 #endif 1229 } 1230 1231 static VMM_STAT_INTEL(VCPU_INVVPID_SAVED, "Number of vpid invalidations saved"); 1232 static VMM_STAT_INTEL(VCPU_INVVPID_DONE, "Number of vpid invalidations done"); 1233 1234 /* 1235 * Invalidate guest mappings identified by its vpid from the TLB. 1236 */ 1237 static __inline void 1238 vmx_invvpid(struct vmx *vmx, int vcpu, pmap_t pmap, int running) 1239 { 1240 struct vmxstate *vmxstate; 1241 struct invvpid_desc invvpid_desc; 1242 1243 vmxstate = &vmx->state[vcpu]; 1244 if (vmxstate->vpid == 0) 1245 return; 1246 1247 if (!running) { 1248 /* 1249 * Set the 'lastcpu' to an invalid host cpu. 1250 * 1251 * This will invalidate TLB entries tagged with the vcpu's 1252 * vpid the next time it runs via vmx_set_pcpu_defaults(). 1253 */ 1254 vmxstate->lastcpu = NOCPU; 1255 return; 1256 } 1257 1258 KASSERT(curthread->td_critnest > 0, ("%s: vcpu %d running outside " 1259 "critical section", __func__, vcpu)); 1260 1261 /* 1262 * Invalidate all mappings tagged with 'vpid' 1263 * 1264 * We do this because this vcpu was executing on a different host 1265 * cpu when it last ran. We do not track whether it invalidated 1266 * mappings associated with its 'vpid' during that run. So we must 1267 * assume that the mappings associated with 'vpid' on 'curcpu' are 1268 * stale and invalidate them. 1269 * 1270 * Note that we incur this penalty only when the scheduler chooses to 1271 * move the thread associated with this vcpu between host cpus. 1272 * 1273 * Note also that this will invalidate mappings tagged with 'vpid' 1274 * for "all" EP4TAs. 1275 */ 1276 if (pmap->pm_eptgen == vmx->eptgen[curcpu]) { 1277 invvpid_desc._res1 = 0; 1278 invvpid_desc._res2 = 0; 1279 invvpid_desc.vpid = vmxstate->vpid; 1280 invvpid_desc.linear_addr = 0; 1281 invvpid(INVVPID_TYPE_SINGLE_CONTEXT, invvpid_desc); 1282 vmm_stat_incr(vmx->vm, vcpu, VCPU_INVVPID_DONE, 1); 1283 } else { 1284 /* 1285 * The invvpid can be skipped if an invept is going to 1286 * be performed before entering the guest. The invept 1287 * will invalidate combined mappings tagged with 1288 * 'vmx->eptp' for all vpids. 1289 */ 1290 vmm_stat_incr(vmx->vm, vcpu, VCPU_INVVPID_SAVED, 1); 1291 } 1292 } 1293 1294 static void 1295 vmx_set_pcpu_defaults(struct vmx *vmx, int vcpu, pmap_t pmap) 1296 { 1297 struct vmxstate *vmxstate; 1298 1299 vmxstate = &vmx->state[vcpu]; 1300 if (vmxstate->lastcpu == curcpu) 1301 return; 1302 1303 vmxstate->lastcpu = curcpu; 1304 1305 vmm_stat_incr(vmx->vm, vcpu, VCPU_MIGRATIONS, 1); 1306 1307 vmcs_write(VMCS_HOST_TR_BASE, vmm_get_host_trbase()); 1308 vmcs_write(VMCS_HOST_GDTR_BASE, vmm_get_host_gdtrbase()); 1309 vmcs_write(VMCS_HOST_GS_BASE, vmm_get_host_gsbase()); 1310 vmx_invvpid(vmx, vcpu, pmap, 1); 1311 } 1312 1313 /* 1314 * We depend on 'procbased_ctls' to have the Interrupt Window Exiting bit set. 1315 */ 1316 CTASSERT((PROCBASED_CTLS_ONE_SETTING & PROCBASED_INT_WINDOW_EXITING) != 0); 1317 1318 static void __inline 1319 vmx_set_int_window_exiting(struct vmx *vmx, int vcpu) 1320 { 1321 1322 if ((vmx->cap[vcpu].proc_ctls & PROCBASED_INT_WINDOW_EXITING) == 0) { 1323 vmx->cap[vcpu].proc_ctls |= PROCBASED_INT_WINDOW_EXITING; 1324 vmcs_write(VMCS_PRI_PROC_BASED_CTLS, vmx->cap[vcpu].proc_ctls); 1325 VCPU_CTR0(vmx->vm, vcpu, "Enabling interrupt window exiting"); 1326 } 1327 } 1328 1329 static void __inline 1330 vmx_clear_int_window_exiting(struct vmx *vmx, int vcpu) 1331 { 1332 1333 KASSERT((vmx->cap[vcpu].proc_ctls & PROCBASED_INT_WINDOW_EXITING) != 0, 1334 ("intr_window_exiting not set: %#x", vmx->cap[vcpu].proc_ctls)); 1335 vmx->cap[vcpu].proc_ctls &= ~PROCBASED_INT_WINDOW_EXITING; 1336 vmcs_write(VMCS_PRI_PROC_BASED_CTLS, vmx->cap[vcpu].proc_ctls); 1337 VCPU_CTR0(vmx->vm, vcpu, "Disabling interrupt window exiting"); 1338 } 1339 1340 static void __inline 1341 vmx_set_nmi_window_exiting(struct vmx *vmx, int vcpu) 1342 { 1343 1344 if ((vmx->cap[vcpu].proc_ctls & PROCBASED_NMI_WINDOW_EXITING) == 0) { 1345 vmx->cap[vcpu].proc_ctls |= PROCBASED_NMI_WINDOW_EXITING; 1346 vmcs_write(VMCS_PRI_PROC_BASED_CTLS, vmx->cap[vcpu].proc_ctls); 1347 VCPU_CTR0(vmx->vm, vcpu, "Enabling NMI window exiting"); 1348 } 1349 } 1350 1351 static void __inline 1352 vmx_clear_nmi_window_exiting(struct vmx *vmx, int vcpu) 1353 { 1354 1355 KASSERT((vmx->cap[vcpu].proc_ctls & PROCBASED_NMI_WINDOW_EXITING) != 0, 1356 ("nmi_window_exiting not set %#x", vmx->cap[vcpu].proc_ctls)); 1357 vmx->cap[vcpu].proc_ctls &= ~PROCBASED_NMI_WINDOW_EXITING; 1358 vmcs_write(VMCS_PRI_PROC_BASED_CTLS, vmx->cap[vcpu].proc_ctls); 1359 VCPU_CTR0(vmx->vm, vcpu, "Disabling NMI window exiting"); 1360 } 1361 1362 int 1363 vmx_set_tsc_offset(struct vmx *vmx, int vcpu, uint64_t offset) 1364 { 1365 int error; 1366 1367 if ((vmx->cap[vcpu].proc_ctls & PROCBASED_TSC_OFFSET) == 0) { 1368 vmx->cap[vcpu].proc_ctls |= PROCBASED_TSC_OFFSET; 1369 vmcs_write(VMCS_PRI_PROC_BASED_CTLS, vmx->cap[vcpu].proc_ctls); 1370 VCPU_CTR0(vmx->vm, vcpu, "Enabling TSC offsetting"); 1371 } 1372 1373 error = vmwrite(VMCS_TSC_OFFSET, offset); 1374 #ifdef BHYVE_SNAPSHOT 1375 if (error == 0) 1376 error = vm_set_tsc_offset(vmx->vm, vcpu, offset); 1377 #endif 1378 return (error); 1379 } 1380 1381 #define NMI_BLOCKING (VMCS_INTERRUPTIBILITY_NMI_BLOCKING | \ 1382 VMCS_INTERRUPTIBILITY_MOVSS_BLOCKING) 1383 #define HWINTR_BLOCKING (VMCS_INTERRUPTIBILITY_STI_BLOCKING | \ 1384 VMCS_INTERRUPTIBILITY_MOVSS_BLOCKING) 1385 1386 static void 1387 vmx_inject_nmi(struct vmx *vmx, int vcpu) 1388 { 1389 uint32_t gi, info; 1390 1391 gi = vmcs_read(VMCS_GUEST_INTERRUPTIBILITY); 1392 KASSERT((gi & NMI_BLOCKING) == 0, ("vmx_inject_nmi: invalid guest " 1393 "interruptibility-state %#x", gi)); 1394 1395 info = vmcs_read(VMCS_ENTRY_INTR_INFO); 1396 KASSERT((info & VMCS_INTR_VALID) == 0, ("vmx_inject_nmi: invalid " 1397 "VM-entry interruption information %#x", info)); 1398 1399 /* 1400 * Inject the virtual NMI. The vector must be the NMI IDT entry 1401 * or the VMCS entry check will fail. 1402 */ 1403 info = IDT_NMI | VMCS_INTR_T_NMI | VMCS_INTR_VALID; 1404 vmcs_write(VMCS_ENTRY_INTR_INFO, info); 1405 1406 VCPU_CTR0(vmx->vm, vcpu, "Injecting vNMI"); 1407 1408 /* Clear the request */ 1409 vm_nmi_clear(vmx->vm, vcpu); 1410 } 1411 1412 static void 1413 vmx_inject_interrupts(struct vmx *vmx, int vcpu, struct vlapic *vlapic, 1414 uint64_t guestrip) 1415 { 1416 int vector, need_nmi_exiting, extint_pending; 1417 uint64_t rflags, entryinfo; 1418 uint32_t gi, info; 1419 1420 if (vmx->state[vcpu].nextrip != guestrip) { 1421 gi = vmcs_read(VMCS_GUEST_INTERRUPTIBILITY); 1422 if (gi & HWINTR_BLOCKING) { 1423 VCPU_CTR2(vmx->vm, vcpu, "Guest interrupt blocking " 1424 "cleared due to rip change: %#lx/%#lx", 1425 vmx->state[vcpu].nextrip, guestrip); 1426 gi &= ~HWINTR_BLOCKING; 1427 vmcs_write(VMCS_GUEST_INTERRUPTIBILITY, gi); 1428 } 1429 } 1430 1431 if (vm_entry_intinfo(vmx->vm, vcpu, &entryinfo)) { 1432 KASSERT((entryinfo & VMCS_INTR_VALID) != 0, ("%s: entry " 1433 "intinfo is not valid: %#lx", __func__, entryinfo)); 1434 1435 info = vmcs_read(VMCS_ENTRY_INTR_INFO); 1436 KASSERT((info & VMCS_INTR_VALID) == 0, ("%s: cannot inject " 1437 "pending exception: %#lx/%#x", __func__, entryinfo, info)); 1438 1439 info = entryinfo; 1440 vector = info & 0xff; 1441 if (vector == IDT_BP || vector == IDT_OF) { 1442 /* 1443 * VT-x requires #BP and #OF to be injected as software 1444 * exceptions. 1445 */ 1446 info &= ~VMCS_INTR_T_MASK; 1447 info |= VMCS_INTR_T_SWEXCEPTION; 1448 } 1449 1450 if (info & VMCS_INTR_DEL_ERRCODE) 1451 vmcs_write(VMCS_ENTRY_EXCEPTION_ERROR, entryinfo >> 32); 1452 1453 vmcs_write(VMCS_ENTRY_INTR_INFO, info); 1454 } 1455 1456 if (vm_nmi_pending(vmx->vm, vcpu)) { 1457 /* 1458 * If there are no conditions blocking NMI injection then 1459 * inject it directly here otherwise enable "NMI window 1460 * exiting" to inject it as soon as we can. 1461 * 1462 * We also check for STI_BLOCKING because some implementations 1463 * don't allow NMI injection in this case. If we are running 1464 * on a processor that doesn't have this restriction it will 1465 * immediately exit and the NMI will be injected in the 1466 * "NMI window exiting" handler. 1467 */ 1468 need_nmi_exiting = 1; 1469 gi = vmcs_read(VMCS_GUEST_INTERRUPTIBILITY); 1470 if ((gi & (HWINTR_BLOCKING | NMI_BLOCKING)) == 0) { 1471 info = vmcs_read(VMCS_ENTRY_INTR_INFO); 1472 if ((info & VMCS_INTR_VALID) == 0) { 1473 vmx_inject_nmi(vmx, vcpu); 1474 need_nmi_exiting = 0; 1475 } else { 1476 VCPU_CTR1(vmx->vm, vcpu, "Cannot inject NMI " 1477 "due to VM-entry intr info %#x", info); 1478 } 1479 } else { 1480 VCPU_CTR1(vmx->vm, vcpu, "Cannot inject NMI due to " 1481 "Guest Interruptibility-state %#x", gi); 1482 } 1483 1484 if (need_nmi_exiting) 1485 vmx_set_nmi_window_exiting(vmx, vcpu); 1486 } 1487 1488 extint_pending = vm_extint_pending(vmx->vm, vcpu); 1489 1490 if (!extint_pending && virtual_interrupt_delivery) { 1491 vmx_inject_pir(vlapic); 1492 return; 1493 } 1494 1495 /* 1496 * If interrupt-window exiting is already in effect then don't bother 1497 * checking for pending interrupts. This is just an optimization and 1498 * not needed for correctness. 1499 */ 1500 if ((vmx->cap[vcpu].proc_ctls & PROCBASED_INT_WINDOW_EXITING) != 0) { 1501 VCPU_CTR0(vmx->vm, vcpu, "Skip interrupt injection due to " 1502 "pending int_window_exiting"); 1503 return; 1504 } 1505 1506 if (!extint_pending) { 1507 /* Ask the local apic for a vector to inject */ 1508 if (!vlapic_pending_intr(vlapic, &vector)) 1509 return; 1510 1511 /* 1512 * From the Intel SDM, Volume 3, Section "Maskable 1513 * Hardware Interrupts": 1514 * - maskable interrupt vectors [16,255] can be delivered 1515 * through the local APIC. 1516 */ 1517 KASSERT(vector >= 16 && vector <= 255, 1518 ("invalid vector %d from local APIC", vector)); 1519 } else { 1520 /* Ask the legacy pic for a vector to inject */ 1521 vatpic_pending_intr(vmx->vm, &vector); 1522 1523 /* 1524 * From the Intel SDM, Volume 3, Section "Maskable 1525 * Hardware Interrupts": 1526 * - maskable interrupt vectors [0,255] can be delivered 1527 * through the INTR pin. 1528 */ 1529 KASSERT(vector >= 0 && vector <= 255, 1530 ("invalid vector %d from INTR", vector)); 1531 } 1532 1533 /* Check RFLAGS.IF and the interruptibility state of the guest */ 1534 rflags = vmcs_read(VMCS_GUEST_RFLAGS); 1535 if ((rflags & PSL_I) == 0) { 1536 VCPU_CTR2(vmx->vm, vcpu, "Cannot inject vector %d due to " 1537 "rflags %#lx", vector, rflags); 1538 goto cantinject; 1539 } 1540 1541 gi = vmcs_read(VMCS_GUEST_INTERRUPTIBILITY); 1542 if (gi & HWINTR_BLOCKING) { 1543 VCPU_CTR2(vmx->vm, vcpu, "Cannot inject vector %d due to " 1544 "Guest Interruptibility-state %#x", vector, gi); 1545 goto cantinject; 1546 } 1547 1548 info = vmcs_read(VMCS_ENTRY_INTR_INFO); 1549 if (info & VMCS_INTR_VALID) { 1550 /* 1551 * This is expected and could happen for multiple reasons: 1552 * - A vectoring VM-entry was aborted due to astpending 1553 * - A VM-exit happened during event injection. 1554 * - An exception was injected above. 1555 * - An NMI was injected above or after "NMI window exiting" 1556 */ 1557 VCPU_CTR2(vmx->vm, vcpu, "Cannot inject vector %d due to " 1558 "VM-entry intr info %#x", vector, info); 1559 goto cantinject; 1560 } 1561 1562 /* Inject the interrupt */ 1563 info = VMCS_INTR_T_HWINTR | VMCS_INTR_VALID; 1564 info |= vector; 1565 vmcs_write(VMCS_ENTRY_INTR_INFO, info); 1566 1567 if (!extint_pending) { 1568 /* Update the Local APIC ISR */ 1569 vlapic_intr_accepted(vlapic, vector); 1570 } else { 1571 vm_extint_clear(vmx->vm, vcpu); 1572 vatpic_intr_accepted(vmx->vm, vector); 1573 1574 /* 1575 * After we accepted the current ExtINT the PIC may 1576 * have posted another one. If that is the case, set 1577 * the Interrupt Window Exiting execution control so 1578 * we can inject that one too. 1579 * 1580 * Also, interrupt window exiting allows us to inject any 1581 * pending APIC vector that was preempted by the ExtINT 1582 * as soon as possible. This applies both for the software 1583 * emulated vlapic and the hardware assisted virtual APIC. 1584 */ 1585 vmx_set_int_window_exiting(vmx, vcpu); 1586 } 1587 1588 VCPU_CTR1(vmx->vm, vcpu, "Injecting hwintr at vector %d", vector); 1589 1590 return; 1591 1592 cantinject: 1593 /* 1594 * Set the Interrupt Window Exiting execution control so we can inject 1595 * the interrupt as soon as blocking condition goes away. 1596 */ 1597 vmx_set_int_window_exiting(vmx, vcpu); 1598 } 1599 1600 /* 1601 * If the Virtual NMIs execution control is '1' then the logical processor 1602 * tracks virtual-NMI blocking in the Guest Interruptibility-state field of 1603 * the VMCS. An IRET instruction in VMX non-root operation will remove any 1604 * virtual-NMI blocking. 1605 * 1606 * This unblocking occurs even if the IRET causes a fault. In this case the 1607 * hypervisor needs to restore virtual-NMI blocking before resuming the guest. 1608 */ 1609 static void 1610 vmx_restore_nmi_blocking(struct vmx *vmx, int vcpuid) 1611 { 1612 uint32_t gi; 1613 1614 VCPU_CTR0(vmx->vm, vcpuid, "Restore Virtual-NMI blocking"); 1615 gi = vmcs_read(VMCS_GUEST_INTERRUPTIBILITY); 1616 gi |= VMCS_INTERRUPTIBILITY_NMI_BLOCKING; 1617 vmcs_write(VMCS_GUEST_INTERRUPTIBILITY, gi); 1618 } 1619 1620 static void 1621 vmx_clear_nmi_blocking(struct vmx *vmx, int vcpuid) 1622 { 1623 uint32_t gi; 1624 1625 VCPU_CTR0(vmx->vm, vcpuid, "Clear Virtual-NMI blocking"); 1626 gi = vmcs_read(VMCS_GUEST_INTERRUPTIBILITY); 1627 gi &= ~VMCS_INTERRUPTIBILITY_NMI_BLOCKING; 1628 vmcs_write(VMCS_GUEST_INTERRUPTIBILITY, gi); 1629 } 1630 1631 static void 1632 vmx_assert_nmi_blocking(struct vmx *vmx, int vcpuid) 1633 { 1634 uint32_t gi; 1635 1636 gi = vmcs_read(VMCS_GUEST_INTERRUPTIBILITY); 1637 KASSERT(gi & VMCS_INTERRUPTIBILITY_NMI_BLOCKING, 1638 ("NMI blocking is not in effect %#x", gi)); 1639 } 1640 1641 static int 1642 vmx_emulate_xsetbv(struct vmx *vmx, int vcpu, struct vm_exit *vmexit) 1643 { 1644 struct vmxctx *vmxctx; 1645 uint64_t xcrval; 1646 const struct xsave_limits *limits; 1647 1648 vmxctx = &vmx->ctx[vcpu]; 1649 limits = vmm_get_xsave_limits(); 1650 1651 /* 1652 * Note that the processor raises a GP# fault on its own if 1653 * xsetbv is executed for CPL != 0, so we do not have to 1654 * emulate that fault here. 1655 */ 1656 1657 /* Only xcr0 is supported. */ 1658 if (vmxctx->guest_rcx != 0) { 1659 vm_inject_gp(vmx->vm, vcpu); 1660 return (HANDLED); 1661 } 1662 1663 /* We only handle xcr0 if both the host and guest have XSAVE enabled. */ 1664 if (!limits->xsave_enabled || !(vmcs_read(VMCS_GUEST_CR4) & CR4_XSAVE)) { 1665 vm_inject_ud(vmx->vm, vcpu); 1666 return (HANDLED); 1667 } 1668 1669 xcrval = vmxctx->guest_rdx << 32 | (vmxctx->guest_rax & 0xffffffff); 1670 if ((xcrval & ~limits->xcr0_allowed) != 0) { 1671 vm_inject_gp(vmx->vm, vcpu); 1672 return (HANDLED); 1673 } 1674 1675 if (!(xcrval & XFEATURE_ENABLED_X87)) { 1676 vm_inject_gp(vmx->vm, vcpu); 1677 return (HANDLED); 1678 } 1679 1680 /* AVX (YMM_Hi128) requires SSE. */ 1681 if (xcrval & XFEATURE_ENABLED_AVX && 1682 (xcrval & XFEATURE_AVX) != XFEATURE_AVX) { 1683 vm_inject_gp(vmx->vm, vcpu); 1684 return (HANDLED); 1685 } 1686 1687 /* 1688 * AVX512 requires base AVX (YMM_Hi128) as well as OpMask, 1689 * ZMM_Hi256, and Hi16_ZMM. 1690 */ 1691 if (xcrval & XFEATURE_AVX512 && 1692 (xcrval & (XFEATURE_AVX512 | XFEATURE_AVX)) != 1693 (XFEATURE_AVX512 | XFEATURE_AVX)) { 1694 vm_inject_gp(vmx->vm, vcpu); 1695 return (HANDLED); 1696 } 1697 1698 /* 1699 * Intel MPX requires both bound register state flags to be 1700 * set. 1701 */ 1702 if (((xcrval & XFEATURE_ENABLED_BNDREGS) != 0) != 1703 ((xcrval & XFEATURE_ENABLED_BNDCSR) != 0)) { 1704 vm_inject_gp(vmx->vm, vcpu); 1705 return (HANDLED); 1706 } 1707 1708 /* 1709 * This runs "inside" vmrun() with the guest's FPU state, so 1710 * modifying xcr0 directly modifies the guest's xcr0, not the 1711 * host's. 1712 */ 1713 load_xcr(0, xcrval); 1714 return (HANDLED); 1715 } 1716 1717 static uint64_t 1718 vmx_get_guest_reg(struct vmx *vmx, int vcpu, int ident) 1719 { 1720 const struct vmxctx *vmxctx; 1721 1722 vmxctx = &vmx->ctx[vcpu]; 1723 1724 switch (ident) { 1725 case 0: 1726 return (vmxctx->guest_rax); 1727 case 1: 1728 return (vmxctx->guest_rcx); 1729 case 2: 1730 return (vmxctx->guest_rdx); 1731 case 3: 1732 return (vmxctx->guest_rbx); 1733 case 4: 1734 return (vmcs_read(VMCS_GUEST_RSP)); 1735 case 5: 1736 return (vmxctx->guest_rbp); 1737 case 6: 1738 return (vmxctx->guest_rsi); 1739 case 7: 1740 return (vmxctx->guest_rdi); 1741 case 8: 1742 return (vmxctx->guest_r8); 1743 case 9: 1744 return (vmxctx->guest_r9); 1745 case 10: 1746 return (vmxctx->guest_r10); 1747 case 11: 1748 return (vmxctx->guest_r11); 1749 case 12: 1750 return (vmxctx->guest_r12); 1751 case 13: 1752 return (vmxctx->guest_r13); 1753 case 14: 1754 return (vmxctx->guest_r14); 1755 case 15: 1756 return (vmxctx->guest_r15); 1757 default: 1758 panic("invalid vmx register %d", ident); 1759 } 1760 } 1761 1762 static void 1763 vmx_set_guest_reg(struct vmx *vmx, int vcpu, int ident, uint64_t regval) 1764 { 1765 struct vmxctx *vmxctx; 1766 1767 vmxctx = &vmx->ctx[vcpu]; 1768 1769 switch (ident) { 1770 case 0: 1771 vmxctx->guest_rax = regval; 1772 break; 1773 case 1: 1774 vmxctx->guest_rcx = regval; 1775 break; 1776 case 2: 1777 vmxctx->guest_rdx = regval; 1778 break; 1779 case 3: 1780 vmxctx->guest_rbx = regval; 1781 break; 1782 case 4: 1783 vmcs_write(VMCS_GUEST_RSP, regval); 1784 break; 1785 case 5: 1786 vmxctx->guest_rbp = regval; 1787 break; 1788 case 6: 1789 vmxctx->guest_rsi = regval; 1790 break; 1791 case 7: 1792 vmxctx->guest_rdi = regval; 1793 break; 1794 case 8: 1795 vmxctx->guest_r8 = regval; 1796 break; 1797 case 9: 1798 vmxctx->guest_r9 = regval; 1799 break; 1800 case 10: 1801 vmxctx->guest_r10 = regval; 1802 break; 1803 case 11: 1804 vmxctx->guest_r11 = regval; 1805 break; 1806 case 12: 1807 vmxctx->guest_r12 = regval; 1808 break; 1809 case 13: 1810 vmxctx->guest_r13 = regval; 1811 break; 1812 case 14: 1813 vmxctx->guest_r14 = regval; 1814 break; 1815 case 15: 1816 vmxctx->guest_r15 = regval; 1817 break; 1818 default: 1819 panic("invalid vmx register %d", ident); 1820 } 1821 } 1822 1823 static int 1824 vmx_emulate_cr0_access(struct vmx *vmx, int vcpu, uint64_t exitqual) 1825 { 1826 uint64_t crval, regval; 1827 1828 /* We only handle mov to %cr0 at this time */ 1829 if ((exitqual & 0xf0) != 0x00) 1830 return (UNHANDLED); 1831 1832 regval = vmx_get_guest_reg(vmx, vcpu, (exitqual >> 8) & 0xf); 1833 1834 vmcs_write(VMCS_CR0_SHADOW, regval); 1835 1836 crval = regval | cr0_ones_mask; 1837 crval &= ~cr0_zeros_mask; 1838 vmcs_write(VMCS_GUEST_CR0, crval); 1839 1840 if (regval & CR0_PG) { 1841 uint64_t efer, entry_ctls; 1842 1843 /* 1844 * If CR0.PG is 1 and EFER.LME is 1 then EFER.LMA and 1845 * the "IA-32e mode guest" bit in VM-entry control must be 1846 * equal. 1847 */ 1848 efer = vmcs_read(VMCS_GUEST_IA32_EFER); 1849 if (efer & EFER_LME) { 1850 efer |= EFER_LMA; 1851 vmcs_write(VMCS_GUEST_IA32_EFER, efer); 1852 entry_ctls = vmcs_read(VMCS_ENTRY_CTLS); 1853 entry_ctls |= VM_ENTRY_GUEST_LMA; 1854 vmcs_write(VMCS_ENTRY_CTLS, entry_ctls); 1855 } 1856 } 1857 1858 return (HANDLED); 1859 } 1860 1861 static int 1862 vmx_emulate_cr4_access(struct vmx *vmx, int vcpu, uint64_t exitqual) 1863 { 1864 uint64_t crval, regval; 1865 1866 /* We only handle mov to %cr4 at this time */ 1867 if ((exitqual & 0xf0) != 0x00) 1868 return (UNHANDLED); 1869 1870 regval = vmx_get_guest_reg(vmx, vcpu, (exitqual >> 8) & 0xf); 1871 1872 vmcs_write(VMCS_CR4_SHADOW, regval); 1873 1874 crval = regval | cr4_ones_mask; 1875 crval &= ~cr4_zeros_mask; 1876 vmcs_write(VMCS_GUEST_CR4, crval); 1877 1878 return (HANDLED); 1879 } 1880 1881 static int 1882 vmx_emulate_cr8_access(struct vmx *vmx, int vcpu, uint64_t exitqual) 1883 { 1884 struct vlapic *vlapic; 1885 uint64_t cr8; 1886 int regnum; 1887 1888 /* We only handle mov %cr8 to/from a register at this time. */ 1889 if ((exitqual & 0xe0) != 0x00) { 1890 return (UNHANDLED); 1891 } 1892 1893 vlapic = vm_lapic(vmx->vm, vcpu); 1894 regnum = (exitqual >> 8) & 0xf; 1895 if (exitqual & 0x10) { 1896 cr8 = vlapic_get_cr8(vlapic); 1897 vmx_set_guest_reg(vmx, vcpu, regnum, cr8); 1898 } else { 1899 cr8 = vmx_get_guest_reg(vmx, vcpu, regnum); 1900 vlapic_set_cr8(vlapic, cr8); 1901 } 1902 1903 return (HANDLED); 1904 } 1905 1906 /* 1907 * From section "Guest Register State" in the Intel SDM: CPL = SS.DPL 1908 */ 1909 static int 1910 vmx_cpl(void) 1911 { 1912 uint32_t ssar; 1913 1914 ssar = vmcs_read(VMCS_GUEST_SS_ACCESS_RIGHTS); 1915 return ((ssar >> 5) & 0x3); 1916 } 1917 1918 static enum vm_cpu_mode 1919 vmx_cpu_mode(void) 1920 { 1921 uint32_t csar; 1922 1923 if (vmcs_read(VMCS_GUEST_IA32_EFER) & EFER_LMA) { 1924 csar = vmcs_read(VMCS_GUEST_CS_ACCESS_RIGHTS); 1925 if (csar & 0x2000) 1926 return (CPU_MODE_64BIT); /* CS.L = 1 */ 1927 else 1928 return (CPU_MODE_COMPATIBILITY); 1929 } else if (vmcs_read(VMCS_GUEST_CR0) & CR0_PE) { 1930 return (CPU_MODE_PROTECTED); 1931 } else { 1932 return (CPU_MODE_REAL); 1933 } 1934 } 1935 1936 static enum vm_paging_mode 1937 vmx_paging_mode(void) 1938 { 1939 uint64_t cr4; 1940 1941 if (!(vmcs_read(VMCS_GUEST_CR0) & CR0_PG)) 1942 return (PAGING_MODE_FLAT); 1943 cr4 = vmcs_read(VMCS_GUEST_CR4); 1944 if (!(cr4 & CR4_PAE)) 1945 return (PAGING_MODE_32); 1946 if (vmcs_read(VMCS_GUEST_IA32_EFER) & EFER_LME) { 1947 if (!(cr4 & CR4_LA57)) 1948 return (PAGING_MODE_64); 1949 return (PAGING_MODE_64_LA57); 1950 } else 1951 return (PAGING_MODE_PAE); 1952 } 1953 1954 static uint64_t 1955 inout_str_index(struct vmx *vmx, int vcpuid, int in) 1956 { 1957 uint64_t val; 1958 int error; 1959 enum vm_reg_name reg; 1960 1961 reg = in ? VM_REG_GUEST_RDI : VM_REG_GUEST_RSI; 1962 error = vmx_getreg(vmx, vcpuid, reg, &val); 1963 KASSERT(error == 0, ("%s: vmx_getreg error %d", __func__, error)); 1964 return (val); 1965 } 1966 1967 static uint64_t 1968 inout_str_count(struct vmx *vmx, int vcpuid, int rep) 1969 { 1970 uint64_t val; 1971 int error; 1972 1973 if (rep) { 1974 error = vmx_getreg(vmx, vcpuid, VM_REG_GUEST_RCX, &val); 1975 KASSERT(!error, ("%s: vmx_getreg error %d", __func__, error)); 1976 } else { 1977 val = 1; 1978 } 1979 return (val); 1980 } 1981 1982 static int 1983 inout_str_addrsize(uint32_t inst_info) 1984 { 1985 uint32_t size; 1986 1987 size = (inst_info >> 7) & 0x7; 1988 switch (size) { 1989 case 0: 1990 return (2); /* 16 bit */ 1991 case 1: 1992 return (4); /* 32 bit */ 1993 case 2: 1994 return (8); /* 64 bit */ 1995 default: 1996 panic("%s: invalid size encoding %d", __func__, size); 1997 } 1998 } 1999 2000 static void 2001 inout_str_seginfo(struct vmx *vmx, int vcpuid, uint32_t inst_info, int in, 2002 struct vm_inout_str *vis) 2003 { 2004 int error, s; 2005 2006 if (in) { 2007 vis->seg_name = VM_REG_GUEST_ES; 2008 } else { 2009 s = (inst_info >> 15) & 0x7; 2010 vis->seg_name = vm_segment_name(s); 2011 } 2012 2013 error = vmx_getdesc(vmx, vcpuid, vis->seg_name, &vis->seg_desc); 2014 KASSERT(error == 0, ("%s: vmx_getdesc error %d", __func__, error)); 2015 } 2016 2017 static void 2018 vmx_paging_info(struct vm_guest_paging *paging) 2019 { 2020 paging->cr3 = vmcs_guest_cr3(); 2021 paging->cpl = vmx_cpl(); 2022 paging->cpu_mode = vmx_cpu_mode(); 2023 paging->paging_mode = vmx_paging_mode(); 2024 } 2025 2026 static void 2027 vmexit_inst_emul(struct vm_exit *vmexit, uint64_t gpa, uint64_t gla) 2028 { 2029 struct vm_guest_paging *paging; 2030 uint32_t csar; 2031 2032 paging = &vmexit->u.inst_emul.paging; 2033 2034 vmexit->exitcode = VM_EXITCODE_INST_EMUL; 2035 vmexit->inst_length = 0; 2036 vmexit->u.inst_emul.gpa = gpa; 2037 vmexit->u.inst_emul.gla = gla; 2038 vmx_paging_info(paging); 2039 switch (paging->cpu_mode) { 2040 case CPU_MODE_REAL: 2041 vmexit->u.inst_emul.cs_base = vmcs_read(VMCS_GUEST_CS_BASE); 2042 vmexit->u.inst_emul.cs_d = 0; 2043 break; 2044 case CPU_MODE_PROTECTED: 2045 case CPU_MODE_COMPATIBILITY: 2046 vmexit->u.inst_emul.cs_base = vmcs_read(VMCS_GUEST_CS_BASE); 2047 csar = vmcs_read(VMCS_GUEST_CS_ACCESS_RIGHTS); 2048 vmexit->u.inst_emul.cs_d = SEG_DESC_DEF32(csar); 2049 break; 2050 default: 2051 vmexit->u.inst_emul.cs_base = 0; 2052 vmexit->u.inst_emul.cs_d = 0; 2053 break; 2054 } 2055 vie_init(&vmexit->u.inst_emul.vie, NULL, 0); 2056 } 2057 2058 static int 2059 ept_fault_type(uint64_t ept_qual) 2060 { 2061 int fault_type; 2062 2063 if (ept_qual & EPT_VIOLATION_DATA_WRITE) 2064 fault_type = VM_PROT_WRITE; 2065 else if (ept_qual & EPT_VIOLATION_INST_FETCH) 2066 fault_type = VM_PROT_EXECUTE; 2067 else 2068 fault_type= VM_PROT_READ; 2069 2070 return (fault_type); 2071 } 2072 2073 static bool 2074 ept_emulation_fault(uint64_t ept_qual) 2075 { 2076 int read, write; 2077 2078 /* EPT fault on an instruction fetch doesn't make sense here */ 2079 if (ept_qual & EPT_VIOLATION_INST_FETCH) 2080 return (false); 2081 2082 /* EPT fault must be a read fault or a write fault */ 2083 read = ept_qual & EPT_VIOLATION_DATA_READ ? 1 : 0; 2084 write = ept_qual & EPT_VIOLATION_DATA_WRITE ? 1 : 0; 2085 if ((read | write) == 0) 2086 return (false); 2087 2088 /* 2089 * The EPT violation must have been caused by accessing a 2090 * guest-physical address that is a translation of a guest-linear 2091 * address. 2092 */ 2093 if ((ept_qual & EPT_VIOLATION_GLA_VALID) == 0 || 2094 (ept_qual & EPT_VIOLATION_XLAT_VALID) == 0) { 2095 return (false); 2096 } 2097 2098 return (true); 2099 } 2100 2101 static __inline int 2102 apic_access_virtualization(struct vmx *vmx, int vcpuid) 2103 { 2104 uint32_t proc_ctls2; 2105 2106 proc_ctls2 = vmx->cap[vcpuid].proc_ctls2; 2107 return ((proc_ctls2 & PROCBASED2_VIRTUALIZE_APIC_ACCESSES) ? 1 : 0); 2108 } 2109 2110 static __inline int 2111 x2apic_virtualization(struct vmx *vmx, int vcpuid) 2112 { 2113 uint32_t proc_ctls2; 2114 2115 proc_ctls2 = vmx->cap[vcpuid].proc_ctls2; 2116 return ((proc_ctls2 & PROCBASED2_VIRTUALIZE_X2APIC_MODE) ? 1 : 0); 2117 } 2118 2119 static int 2120 vmx_handle_apic_write(struct vmx *vmx, int vcpuid, struct vlapic *vlapic, 2121 uint64_t qual) 2122 { 2123 int error, handled, offset; 2124 uint32_t *apic_regs, vector; 2125 bool retu; 2126 2127 handled = HANDLED; 2128 offset = APIC_WRITE_OFFSET(qual); 2129 2130 if (!apic_access_virtualization(vmx, vcpuid)) { 2131 /* 2132 * In general there should not be any APIC write VM-exits 2133 * unless APIC-access virtualization is enabled. 2134 * 2135 * However self-IPI virtualization can legitimately trigger 2136 * an APIC-write VM-exit so treat it specially. 2137 */ 2138 if (x2apic_virtualization(vmx, vcpuid) && 2139 offset == APIC_OFFSET_SELF_IPI) { 2140 apic_regs = (uint32_t *)(vlapic->apic_page); 2141 vector = apic_regs[APIC_OFFSET_SELF_IPI / 4]; 2142 vlapic_self_ipi_handler(vlapic, vector); 2143 return (HANDLED); 2144 } else 2145 return (UNHANDLED); 2146 } 2147 2148 switch (offset) { 2149 case APIC_OFFSET_ID: 2150 vlapic_id_write_handler(vlapic); 2151 break; 2152 case APIC_OFFSET_LDR: 2153 vlapic_ldr_write_handler(vlapic); 2154 break; 2155 case APIC_OFFSET_DFR: 2156 vlapic_dfr_write_handler(vlapic); 2157 break; 2158 case APIC_OFFSET_SVR: 2159 vlapic_svr_write_handler(vlapic); 2160 break; 2161 case APIC_OFFSET_ESR: 2162 vlapic_esr_write_handler(vlapic); 2163 break; 2164 case APIC_OFFSET_ICR_LOW: 2165 retu = false; 2166 error = vlapic_icrlo_write_handler(vlapic, &retu); 2167 if (error != 0 || retu) 2168 handled = UNHANDLED; 2169 break; 2170 case APIC_OFFSET_CMCI_LVT: 2171 case APIC_OFFSET_TIMER_LVT ... APIC_OFFSET_ERROR_LVT: 2172 vlapic_lvt_write_handler(vlapic, offset); 2173 break; 2174 case APIC_OFFSET_TIMER_ICR: 2175 vlapic_icrtmr_write_handler(vlapic); 2176 break; 2177 case APIC_OFFSET_TIMER_DCR: 2178 vlapic_dcr_write_handler(vlapic); 2179 break; 2180 default: 2181 handled = UNHANDLED; 2182 break; 2183 } 2184 return (handled); 2185 } 2186 2187 static bool 2188 apic_access_fault(struct vmx *vmx, int vcpuid, uint64_t gpa) 2189 { 2190 2191 if (apic_access_virtualization(vmx, vcpuid) && 2192 (gpa >= DEFAULT_APIC_BASE && gpa < DEFAULT_APIC_BASE + PAGE_SIZE)) 2193 return (true); 2194 else 2195 return (false); 2196 } 2197 2198 static int 2199 vmx_handle_apic_access(struct vmx *vmx, int vcpuid, struct vm_exit *vmexit) 2200 { 2201 uint64_t qual; 2202 int access_type, offset, allowed; 2203 2204 if (!apic_access_virtualization(vmx, vcpuid)) 2205 return (UNHANDLED); 2206 2207 qual = vmexit->u.vmx.exit_qualification; 2208 access_type = APIC_ACCESS_TYPE(qual); 2209 offset = APIC_ACCESS_OFFSET(qual); 2210 2211 allowed = 0; 2212 if (access_type == 0) { 2213 /* 2214 * Read data access to the following registers is expected. 2215 */ 2216 switch (offset) { 2217 case APIC_OFFSET_APR: 2218 case APIC_OFFSET_PPR: 2219 case APIC_OFFSET_RRR: 2220 case APIC_OFFSET_CMCI_LVT: 2221 case APIC_OFFSET_TIMER_CCR: 2222 allowed = 1; 2223 break; 2224 default: 2225 break; 2226 } 2227 } else if (access_type == 1) { 2228 /* 2229 * Write data access to the following registers is expected. 2230 */ 2231 switch (offset) { 2232 case APIC_OFFSET_VER: 2233 case APIC_OFFSET_APR: 2234 case APIC_OFFSET_PPR: 2235 case APIC_OFFSET_RRR: 2236 case APIC_OFFSET_ISR0 ... APIC_OFFSET_ISR7: 2237 case APIC_OFFSET_TMR0 ... APIC_OFFSET_TMR7: 2238 case APIC_OFFSET_IRR0 ... APIC_OFFSET_IRR7: 2239 case APIC_OFFSET_CMCI_LVT: 2240 case APIC_OFFSET_TIMER_CCR: 2241 allowed = 1; 2242 break; 2243 default: 2244 break; 2245 } 2246 } 2247 2248 if (allowed) { 2249 vmexit_inst_emul(vmexit, DEFAULT_APIC_BASE + offset, 2250 VIE_INVALID_GLA); 2251 } 2252 2253 /* 2254 * Regardless of whether the APIC-access is allowed this handler 2255 * always returns UNHANDLED: 2256 * - if the access is allowed then it is handled by emulating the 2257 * instruction that caused the VM-exit (outside the critical section) 2258 * - if the access is not allowed then it will be converted to an 2259 * exitcode of VM_EXITCODE_VMX and will be dealt with in userland. 2260 */ 2261 return (UNHANDLED); 2262 } 2263 2264 static enum task_switch_reason 2265 vmx_task_switch_reason(uint64_t qual) 2266 { 2267 int reason; 2268 2269 reason = (qual >> 30) & 0x3; 2270 switch (reason) { 2271 case 0: 2272 return (TSR_CALL); 2273 case 1: 2274 return (TSR_IRET); 2275 case 2: 2276 return (TSR_JMP); 2277 case 3: 2278 return (TSR_IDT_GATE); 2279 default: 2280 panic("%s: invalid reason %d", __func__, reason); 2281 } 2282 } 2283 2284 static int 2285 emulate_wrmsr(struct vmx *vmx, int vcpuid, u_int num, uint64_t val, bool *retu) 2286 { 2287 int error; 2288 2289 if (lapic_msr(num)) 2290 error = lapic_wrmsr(vmx->vm, vcpuid, num, val, retu); 2291 else 2292 error = vmx_wrmsr(vmx, vcpuid, num, val, retu); 2293 2294 return (error); 2295 } 2296 2297 static int 2298 emulate_rdmsr(struct vmx *vmx, int vcpuid, u_int num, bool *retu) 2299 { 2300 struct vmxctx *vmxctx; 2301 uint64_t result; 2302 uint32_t eax, edx; 2303 int error; 2304 2305 if (lapic_msr(num)) 2306 error = lapic_rdmsr(vmx->vm, vcpuid, num, &result, retu); 2307 else 2308 error = vmx_rdmsr(vmx, vcpuid, num, &result, retu); 2309 2310 if (error == 0) { 2311 eax = result; 2312 vmxctx = &vmx->ctx[vcpuid]; 2313 error = vmxctx_setreg(vmxctx, VM_REG_GUEST_RAX, eax); 2314 KASSERT(error == 0, ("vmxctx_setreg(rax) error %d", error)); 2315 2316 edx = result >> 32; 2317 error = vmxctx_setreg(vmxctx, VM_REG_GUEST_RDX, edx); 2318 KASSERT(error == 0, ("vmxctx_setreg(rdx) error %d", error)); 2319 } 2320 2321 return (error); 2322 } 2323 2324 static int 2325 vmx_exit_process(struct vmx *vmx, int vcpu, struct vm_exit *vmexit) 2326 { 2327 int error, errcode, errcode_valid, handled, in; 2328 struct vmxctx *vmxctx; 2329 struct vlapic *vlapic; 2330 struct vm_inout_str *vis; 2331 struct vm_task_switch *ts; 2332 uint32_t eax, ecx, edx, idtvec_info, idtvec_err, intr_info, inst_info; 2333 uint32_t intr_type, intr_vec, reason; 2334 uint64_t exitintinfo, qual, gpa; 2335 bool retu; 2336 2337 CTASSERT((PINBASED_CTLS_ONE_SETTING & PINBASED_VIRTUAL_NMI) != 0); 2338 CTASSERT((PINBASED_CTLS_ONE_SETTING & PINBASED_NMI_EXITING) != 0); 2339 2340 handled = UNHANDLED; 2341 vmxctx = &vmx->ctx[vcpu]; 2342 2343 qual = vmexit->u.vmx.exit_qualification; 2344 reason = vmexit->u.vmx.exit_reason; 2345 vmexit->exitcode = VM_EXITCODE_BOGUS; 2346 2347 vmm_stat_incr(vmx->vm, vcpu, VMEXIT_COUNT, 1); 2348 SDT_PROBE3(vmm, vmx, exit, entry, vmx, vcpu, vmexit); 2349 2350 /* 2351 * VM-entry failures during or after loading guest state. 2352 * 2353 * These VM-exits are uncommon but must be handled specially 2354 * as most VM-exit fields are not populated as usual. 2355 */ 2356 if (__predict_false(reason == EXIT_REASON_MCE_DURING_ENTRY)) { 2357 VCPU_CTR0(vmx->vm, vcpu, "Handling MCE during VM-entry"); 2358 __asm __volatile("int $18"); 2359 return (1); 2360 } 2361 2362 /* 2363 * VM exits that can be triggered during event delivery need to 2364 * be handled specially by re-injecting the event if the IDT 2365 * vectoring information field's valid bit is set. 2366 * 2367 * See "Information for VM Exits During Event Delivery" in Intel SDM 2368 * for details. 2369 */ 2370 idtvec_info = vmcs_idt_vectoring_info(); 2371 if (idtvec_info & VMCS_IDT_VEC_VALID) { 2372 idtvec_info &= ~(1 << 12); /* clear undefined bit */ 2373 exitintinfo = idtvec_info; 2374 if (idtvec_info & VMCS_IDT_VEC_ERRCODE_VALID) { 2375 idtvec_err = vmcs_idt_vectoring_err(); 2376 exitintinfo |= (uint64_t)idtvec_err << 32; 2377 } 2378 error = vm_exit_intinfo(vmx->vm, vcpu, exitintinfo); 2379 KASSERT(error == 0, ("%s: vm_set_intinfo error %d", 2380 __func__, error)); 2381 2382 /* 2383 * If 'virtual NMIs' are being used and the VM-exit 2384 * happened while injecting an NMI during the previous 2385 * VM-entry, then clear "blocking by NMI" in the 2386 * Guest Interruptibility-State so the NMI can be 2387 * reinjected on the subsequent VM-entry. 2388 * 2389 * However, if the NMI was being delivered through a task 2390 * gate, then the new task must start execution with NMIs 2391 * blocked so don't clear NMI blocking in this case. 2392 */ 2393 intr_type = idtvec_info & VMCS_INTR_T_MASK; 2394 if (intr_type == VMCS_INTR_T_NMI) { 2395 if (reason != EXIT_REASON_TASK_SWITCH) 2396 vmx_clear_nmi_blocking(vmx, vcpu); 2397 else 2398 vmx_assert_nmi_blocking(vmx, vcpu); 2399 } 2400 2401 /* 2402 * Update VM-entry instruction length if the event being 2403 * delivered was a software interrupt or software exception. 2404 */ 2405 if (intr_type == VMCS_INTR_T_SWINTR || 2406 intr_type == VMCS_INTR_T_PRIV_SWEXCEPTION || 2407 intr_type == VMCS_INTR_T_SWEXCEPTION) { 2408 vmcs_write(VMCS_ENTRY_INST_LENGTH, vmexit->inst_length); 2409 } 2410 } 2411 2412 switch (reason) { 2413 case EXIT_REASON_TASK_SWITCH: 2414 ts = &vmexit->u.task_switch; 2415 ts->tsssel = qual & 0xffff; 2416 ts->reason = vmx_task_switch_reason(qual); 2417 ts->ext = 0; 2418 ts->errcode_valid = 0; 2419 vmx_paging_info(&ts->paging); 2420 /* 2421 * If the task switch was due to a CALL, JMP, IRET, software 2422 * interrupt (INT n) or software exception (INT3, INTO), 2423 * then the saved %rip references the instruction that caused 2424 * the task switch. The instruction length field in the VMCS 2425 * is valid in this case. 2426 * 2427 * In all other cases (e.g., NMI, hardware exception) the 2428 * saved %rip is one that would have been saved in the old TSS 2429 * had the task switch completed normally so the instruction 2430 * length field is not needed in this case and is explicitly 2431 * set to 0. 2432 */ 2433 if (ts->reason == TSR_IDT_GATE) { 2434 KASSERT(idtvec_info & VMCS_IDT_VEC_VALID, 2435 ("invalid idtvec_info %#x for IDT task switch", 2436 idtvec_info)); 2437 intr_type = idtvec_info & VMCS_INTR_T_MASK; 2438 if (intr_type != VMCS_INTR_T_SWINTR && 2439 intr_type != VMCS_INTR_T_SWEXCEPTION && 2440 intr_type != VMCS_INTR_T_PRIV_SWEXCEPTION) { 2441 /* Task switch triggered by external event */ 2442 ts->ext = 1; 2443 vmexit->inst_length = 0; 2444 if (idtvec_info & VMCS_IDT_VEC_ERRCODE_VALID) { 2445 ts->errcode_valid = 1; 2446 ts->errcode = vmcs_idt_vectoring_err(); 2447 } 2448 } 2449 } 2450 vmexit->exitcode = VM_EXITCODE_TASK_SWITCH; 2451 SDT_PROBE4(vmm, vmx, exit, taskswitch, vmx, vcpu, vmexit, ts); 2452 VCPU_CTR4(vmx->vm, vcpu, "task switch reason %d, tss 0x%04x, " 2453 "%s errcode 0x%016lx", ts->reason, ts->tsssel, 2454 ts->ext ? "external" : "internal", 2455 ((uint64_t)ts->errcode << 32) | ts->errcode_valid); 2456 break; 2457 case EXIT_REASON_CR_ACCESS: 2458 vmm_stat_incr(vmx->vm, vcpu, VMEXIT_CR_ACCESS, 1); 2459 SDT_PROBE4(vmm, vmx, exit, craccess, vmx, vcpu, vmexit, qual); 2460 switch (qual & 0xf) { 2461 case 0: 2462 handled = vmx_emulate_cr0_access(vmx, vcpu, qual); 2463 break; 2464 case 4: 2465 handled = vmx_emulate_cr4_access(vmx, vcpu, qual); 2466 break; 2467 case 8: 2468 handled = vmx_emulate_cr8_access(vmx, vcpu, qual); 2469 break; 2470 } 2471 break; 2472 case EXIT_REASON_RDMSR: 2473 vmm_stat_incr(vmx->vm, vcpu, VMEXIT_RDMSR, 1); 2474 retu = false; 2475 ecx = vmxctx->guest_rcx; 2476 VCPU_CTR1(vmx->vm, vcpu, "rdmsr 0x%08x", ecx); 2477 SDT_PROBE4(vmm, vmx, exit, rdmsr, vmx, vcpu, vmexit, ecx); 2478 error = emulate_rdmsr(vmx, vcpu, ecx, &retu); 2479 if (error) { 2480 vmexit->exitcode = VM_EXITCODE_RDMSR; 2481 vmexit->u.msr.code = ecx; 2482 } else if (!retu) { 2483 handled = HANDLED; 2484 } else { 2485 /* Return to userspace with a valid exitcode */ 2486 KASSERT(vmexit->exitcode != VM_EXITCODE_BOGUS, 2487 ("emulate_rdmsr retu with bogus exitcode")); 2488 } 2489 break; 2490 case EXIT_REASON_WRMSR: 2491 vmm_stat_incr(vmx->vm, vcpu, VMEXIT_WRMSR, 1); 2492 retu = false; 2493 eax = vmxctx->guest_rax; 2494 ecx = vmxctx->guest_rcx; 2495 edx = vmxctx->guest_rdx; 2496 VCPU_CTR2(vmx->vm, vcpu, "wrmsr 0x%08x value 0x%016lx", 2497 ecx, (uint64_t)edx << 32 | eax); 2498 SDT_PROBE5(vmm, vmx, exit, wrmsr, vmx, vmexit, vcpu, ecx, 2499 (uint64_t)edx << 32 | eax); 2500 error = emulate_wrmsr(vmx, vcpu, ecx, 2501 (uint64_t)edx << 32 | eax, &retu); 2502 if (error) { 2503 vmexit->exitcode = VM_EXITCODE_WRMSR; 2504 vmexit->u.msr.code = ecx; 2505 vmexit->u.msr.wval = (uint64_t)edx << 32 | eax; 2506 } else if (!retu) { 2507 handled = HANDLED; 2508 } else { 2509 /* Return to userspace with a valid exitcode */ 2510 KASSERT(vmexit->exitcode != VM_EXITCODE_BOGUS, 2511 ("emulate_wrmsr retu with bogus exitcode")); 2512 } 2513 break; 2514 case EXIT_REASON_HLT: 2515 vmm_stat_incr(vmx->vm, vcpu, VMEXIT_HLT, 1); 2516 SDT_PROBE3(vmm, vmx, exit, halt, vmx, vcpu, vmexit); 2517 vmexit->exitcode = VM_EXITCODE_HLT; 2518 vmexit->u.hlt.rflags = vmcs_read(VMCS_GUEST_RFLAGS); 2519 if (virtual_interrupt_delivery) 2520 vmexit->u.hlt.intr_status = 2521 vmcs_read(VMCS_GUEST_INTR_STATUS); 2522 else 2523 vmexit->u.hlt.intr_status = 0; 2524 break; 2525 case EXIT_REASON_MTF: 2526 vmm_stat_incr(vmx->vm, vcpu, VMEXIT_MTRAP, 1); 2527 SDT_PROBE3(vmm, vmx, exit, mtrap, vmx, vcpu, vmexit); 2528 vmexit->exitcode = VM_EXITCODE_MTRAP; 2529 vmexit->inst_length = 0; 2530 break; 2531 case EXIT_REASON_PAUSE: 2532 vmm_stat_incr(vmx->vm, vcpu, VMEXIT_PAUSE, 1); 2533 SDT_PROBE3(vmm, vmx, exit, pause, vmx, vcpu, vmexit); 2534 vmexit->exitcode = VM_EXITCODE_PAUSE; 2535 break; 2536 case EXIT_REASON_INTR_WINDOW: 2537 vmm_stat_incr(vmx->vm, vcpu, VMEXIT_INTR_WINDOW, 1); 2538 SDT_PROBE3(vmm, vmx, exit, intrwindow, vmx, vcpu, vmexit); 2539 vmx_clear_int_window_exiting(vmx, vcpu); 2540 return (1); 2541 case EXIT_REASON_EXT_INTR: 2542 /* 2543 * External interrupts serve only to cause VM exits and allow 2544 * the host interrupt handler to run. 2545 * 2546 * If this external interrupt triggers a virtual interrupt 2547 * to a VM, then that state will be recorded by the 2548 * host interrupt handler in the VM's softc. We will inject 2549 * this virtual interrupt during the subsequent VM enter. 2550 */ 2551 intr_info = vmcs_read(VMCS_EXIT_INTR_INFO); 2552 SDT_PROBE4(vmm, vmx, exit, interrupt, 2553 vmx, vcpu, vmexit, intr_info); 2554 2555 /* 2556 * XXX: Ignore this exit if VMCS_INTR_VALID is not set. 2557 * This appears to be a bug in VMware Fusion? 2558 */ 2559 if (!(intr_info & VMCS_INTR_VALID)) 2560 return (1); 2561 KASSERT((intr_info & VMCS_INTR_VALID) != 0 && 2562 (intr_info & VMCS_INTR_T_MASK) == VMCS_INTR_T_HWINTR, 2563 ("VM exit interruption info invalid: %#x", intr_info)); 2564 vmx_trigger_hostintr(intr_info & 0xff); 2565 2566 /* 2567 * This is special. We want to treat this as an 'handled' 2568 * VM-exit but not increment the instruction pointer. 2569 */ 2570 vmm_stat_incr(vmx->vm, vcpu, VMEXIT_EXTINT, 1); 2571 return (1); 2572 case EXIT_REASON_NMI_WINDOW: 2573 SDT_PROBE3(vmm, vmx, exit, nmiwindow, vmx, vcpu, vmexit); 2574 /* Exit to allow the pending virtual NMI to be injected */ 2575 if (vm_nmi_pending(vmx->vm, vcpu)) 2576 vmx_inject_nmi(vmx, vcpu); 2577 vmx_clear_nmi_window_exiting(vmx, vcpu); 2578 vmm_stat_incr(vmx->vm, vcpu, VMEXIT_NMI_WINDOW, 1); 2579 return (1); 2580 case EXIT_REASON_INOUT: 2581 vmm_stat_incr(vmx->vm, vcpu, VMEXIT_INOUT, 1); 2582 vmexit->exitcode = VM_EXITCODE_INOUT; 2583 vmexit->u.inout.bytes = (qual & 0x7) + 1; 2584 vmexit->u.inout.in = in = (qual & 0x8) ? 1 : 0; 2585 vmexit->u.inout.string = (qual & 0x10) ? 1 : 0; 2586 vmexit->u.inout.rep = (qual & 0x20) ? 1 : 0; 2587 vmexit->u.inout.port = (uint16_t)(qual >> 16); 2588 vmexit->u.inout.eax = (uint32_t)(vmxctx->guest_rax); 2589 if (vmexit->u.inout.string) { 2590 inst_info = vmcs_read(VMCS_EXIT_INSTRUCTION_INFO); 2591 vmexit->exitcode = VM_EXITCODE_INOUT_STR; 2592 vis = &vmexit->u.inout_str; 2593 vmx_paging_info(&vis->paging); 2594 vis->rflags = vmcs_read(VMCS_GUEST_RFLAGS); 2595 vis->cr0 = vmcs_read(VMCS_GUEST_CR0); 2596 vis->index = inout_str_index(vmx, vcpu, in); 2597 vis->count = inout_str_count(vmx, vcpu, vis->inout.rep); 2598 vis->addrsize = inout_str_addrsize(inst_info); 2599 inout_str_seginfo(vmx, vcpu, inst_info, in, vis); 2600 } 2601 SDT_PROBE3(vmm, vmx, exit, inout, vmx, vcpu, vmexit); 2602 break; 2603 case EXIT_REASON_CPUID: 2604 vmm_stat_incr(vmx->vm, vcpu, VMEXIT_CPUID, 1); 2605 SDT_PROBE3(vmm, vmx, exit, cpuid, vmx, vcpu, vmexit); 2606 handled = vmx_handle_cpuid(vmx->vm, vcpu, vmxctx); 2607 break; 2608 case EXIT_REASON_EXCEPTION: 2609 vmm_stat_incr(vmx->vm, vcpu, VMEXIT_EXCEPTION, 1); 2610 intr_info = vmcs_read(VMCS_EXIT_INTR_INFO); 2611 KASSERT((intr_info & VMCS_INTR_VALID) != 0, 2612 ("VM exit interruption info invalid: %#x", intr_info)); 2613 2614 intr_vec = intr_info & 0xff; 2615 intr_type = intr_info & VMCS_INTR_T_MASK; 2616 2617 /* 2618 * If Virtual NMIs control is 1 and the VM-exit is due to a 2619 * fault encountered during the execution of IRET then we must 2620 * restore the state of "virtual-NMI blocking" before resuming 2621 * the guest. 2622 * 2623 * See "Resuming Guest Software after Handling an Exception". 2624 * See "Information for VM Exits Due to Vectored Events". 2625 */ 2626 if ((idtvec_info & VMCS_IDT_VEC_VALID) == 0 && 2627 (intr_vec != IDT_DF) && 2628 (intr_info & EXIT_QUAL_NMIUDTI) != 0) 2629 vmx_restore_nmi_blocking(vmx, vcpu); 2630 2631 /* 2632 * The NMI has already been handled in vmx_exit_handle_nmi(). 2633 */ 2634 if (intr_type == VMCS_INTR_T_NMI) 2635 return (1); 2636 2637 /* 2638 * Call the machine check handler by hand. Also don't reflect 2639 * the machine check back into the guest. 2640 */ 2641 if (intr_vec == IDT_MC) { 2642 VCPU_CTR0(vmx->vm, vcpu, "Vectoring to MCE handler"); 2643 __asm __volatile("int $18"); 2644 return (1); 2645 } 2646 2647 /* 2648 * If the hypervisor has requested user exits for 2649 * debug exceptions, bounce them out to userland. 2650 */ 2651 if (intr_type == VMCS_INTR_T_SWEXCEPTION && intr_vec == IDT_BP && 2652 (vmx->cap[vcpu].set & (1 << VM_CAP_BPT_EXIT))) { 2653 vmexit->exitcode = VM_EXITCODE_BPT; 2654 vmexit->u.bpt.inst_length = vmexit->inst_length; 2655 vmexit->inst_length = 0; 2656 break; 2657 } 2658 2659 if (intr_vec == IDT_PF) { 2660 error = vmxctx_setreg(vmxctx, VM_REG_GUEST_CR2, qual); 2661 KASSERT(error == 0, ("%s: vmxctx_setreg(cr2) error %d", 2662 __func__, error)); 2663 } 2664 2665 /* 2666 * Software exceptions exhibit trap-like behavior. This in 2667 * turn requires populating the VM-entry instruction length 2668 * so that the %rip in the trap frame is past the INT3/INTO 2669 * instruction. 2670 */ 2671 if (intr_type == VMCS_INTR_T_SWEXCEPTION) 2672 vmcs_write(VMCS_ENTRY_INST_LENGTH, vmexit->inst_length); 2673 2674 /* Reflect all other exceptions back into the guest */ 2675 errcode_valid = errcode = 0; 2676 if (intr_info & VMCS_INTR_DEL_ERRCODE) { 2677 errcode_valid = 1; 2678 errcode = vmcs_read(VMCS_EXIT_INTR_ERRCODE); 2679 } 2680 VCPU_CTR2(vmx->vm, vcpu, "Reflecting exception %d/%#x into " 2681 "the guest", intr_vec, errcode); 2682 SDT_PROBE5(vmm, vmx, exit, exception, 2683 vmx, vcpu, vmexit, intr_vec, errcode); 2684 error = vm_inject_exception(vmx->vm, vcpu, intr_vec, 2685 errcode_valid, errcode, 0); 2686 KASSERT(error == 0, ("%s: vm_inject_exception error %d", 2687 __func__, error)); 2688 return (1); 2689 2690 case EXIT_REASON_EPT_FAULT: 2691 /* 2692 * If 'gpa' lies within the address space allocated to 2693 * memory then this must be a nested page fault otherwise 2694 * this must be an instruction that accesses MMIO space. 2695 */ 2696 gpa = vmcs_gpa(); 2697 if (vm_mem_allocated(vmx->vm, vcpu, gpa) || 2698 apic_access_fault(vmx, vcpu, gpa)) { 2699 vmexit->exitcode = VM_EXITCODE_PAGING; 2700 vmexit->inst_length = 0; 2701 vmexit->u.paging.gpa = gpa; 2702 vmexit->u.paging.fault_type = ept_fault_type(qual); 2703 vmm_stat_incr(vmx->vm, vcpu, VMEXIT_NESTED_FAULT, 1); 2704 SDT_PROBE5(vmm, vmx, exit, nestedfault, 2705 vmx, vcpu, vmexit, gpa, qual); 2706 } else if (ept_emulation_fault(qual)) { 2707 vmexit_inst_emul(vmexit, gpa, vmcs_gla()); 2708 vmm_stat_incr(vmx->vm, vcpu, VMEXIT_INST_EMUL, 1); 2709 SDT_PROBE4(vmm, vmx, exit, mmiofault, 2710 vmx, vcpu, vmexit, gpa); 2711 } 2712 /* 2713 * If Virtual NMIs control is 1 and the VM-exit is due to an 2714 * EPT fault during the execution of IRET then we must restore 2715 * the state of "virtual-NMI blocking" before resuming. 2716 * 2717 * See description of "NMI unblocking due to IRET" in 2718 * "Exit Qualification for EPT Violations". 2719 */ 2720 if ((idtvec_info & VMCS_IDT_VEC_VALID) == 0 && 2721 (qual & EXIT_QUAL_NMIUDTI) != 0) 2722 vmx_restore_nmi_blocking(vmx, vcpu); 2723 break; 2724 case EXIT_REASON_VIRTUALIZED_EOI: 2725 vmexit->exitcode = VM_EXITCODE_IOAPIC_EOI; 2726 vmexit->u.ioapic_eoi.vector = qual & 0xFF; 2727 SDT_PROBE3(vmm, vmx, exit, eoi, vmx, vcpu, vmexit); 2728 vmexit->inst_length = 0; /* trap-like */ 2729 break; 2730 case EXIT_REASON_APIC_ACCESS: 2731 SDT_PROBE3(vmm, vmx, exit, apicaccess, vmx, vcpu, vmexit); 2732 handled = vmx_handle_apic_access(vmx, vcpu, vmexit); 2733 break; 2734 case EXIT_REASON_APIC_WRITE: 2735 /* 2736 * APIC-write VM exit is trap-like so the %rip is already 2737 * pointing to the next instruction. 2738 */ 2739 vmexit->inst_length = 0; 2740 vlapic = vm_lapic(vmx->vm, vcpu); 2741 SDT_PROBE4(vmm, vmx, exit, apicwrite, 2742 vmx, vcpu, vmexit, vlapic); 2743 handled = vmx_handle_apic_write(vmx, vcpu, vlapic, qual); 2744 break; 2745 case EXIT_REASON_XSETBV: 2746 SDT_PROBE3(vmm, vmx, exit, xsetbv, vmx, vcpu, vmexit); 2747 handled = vmx_emulate_xsetbv(vmx, vcpu, vmexit); 2748 break; 2749 case EXIT_REASON_MONITOR: 2750 SDT_PROBE3(vmm, vmx, exit, monitor, vmx, vcpu, vmexit); 2751 vmexit->exitcode = VM_EXITCODE_MONITOR; 2752 break; 2753 case EXIT_REASON_MWAIT: 2754 SDT_PROBE3(vmm, vmx, exit, mwait, vmx, vcpu, vmexit); 2755 vmexit->exitcode = VM_EXITCODE_MWAIT; 2756 break; 2757 case EXIT_REASON_TPR: 2758 vlapic = vm_lapic(vmx->vm, vcpu); 2759 vlapic_sync_tpr(vlapic); 2760 vmexit->inst_length = 0; 2761 handled = HANDLED; 2762 break; 2763 case EXIT_REASON_VMCALL: 2764 case EXIT_REASON_VMCLEAR: 2765 case EXIT_REASON_VMLAUNCH: 2766 case EXIT_REASON_VMPTRLD: 2767 case EXIT_REASON_VMPTRST: 2768 case EXIT_REASON_VMREAD: 2769 case EXIT_REASON_VMRESUME: 2770 case EXIT_REASON_VMWRITE: 2771 case EXIT_REASON_VMXOFF: 2772 case EXIT_REASON_VMXON: 2773 SDT_PROBE3(vmm, vmx, exit, vminsn, vmx, vcpu, vmexit); 2774 vmexit->exitcode = VM_EXITCODE_VMINSN; 2775 break; 2776 default: 2777 SDT_PROBE4(vmm, vmx, exit, unknown, 2778 vmx, vcpu, vmexit, reason); 2779 vmm_stat_incr(vmx->vm, vcpu, VMEXIT_UNKNOWN, 1); 2780 break; 2781 } 2782 2783 if (handled) { 2784 /* 2785 * It is possible that control is returned to userland 2786 * even though we were able to handle the VM exit in the 2787 * kernel. 2788 * 2789 * In such a case we want to make sure that the userland 2790 * restarts guest execution at the instruction *after* 2791 * the one we just processed. Therefore we update the 2792 * guest rip in the VMCS and in 'vmexit'. 2793 */ 2794 vmexit->rip += vmexit->inst_length; 2795 vmexit->inst_length = 0; 2796 vmcs_write(VMCS_GUEST_RIP, vmexit->rip); 2797 } else { 2798 if (vmexit->exitcode == VM_EXITCODE_BOGUS) { 2799 /* 2800 * If this VM exit was not claimed by anybody then 2801 * treat it as a generic VMX exit. 2802 */ 2803 vmexit->exitcode = VM_EXITCODE_VMX; 2804 vmexit->u.vmx.status = VM_SUCCESS; 2805 vmexit->u.vmx.inst_type = 0; 2806 vmexit->u.vmx.inst_error = 0; 2807 } else { 2808 /* 2809 * The exitcode and collateral have been populated. 2810 * The VM exit will be processed further in userland. 2811 */ 2812 } 2813 } 2814 2815 SDT_PROBE4(vmm, vmx, exit, return, 2816 vmx, vcpu, vmexit, handled); 2817 return (handled); 2818 } 2819 2820 static __inline void 2821 vmx_exit_inst_error(struct vmxctx *vmxctx, int rc, struct vm_exit *vmexit) 2822 { 2823 2824 KASSERT(vmxctx->inst_fail_status != VM_SUCCESS, 2825 ("vmx_exit_inst_error: invalid inst_fail_status %d", 2826 vmxctx->inst_fail_status)); 2827 2828 vmexit->inst_length = 0; 2829 vmexit->exitcode = VM_EXITCODE_VMX; 2830 vmexit->u.vmx.status = vmxctx->inst_fail_status; 2831 vmexit->u.vmx.inst_error = vmcs_instruction_error(); 2832 vmexit->u.vmx.exit_reason = ~0; 2833 vmexit->u.vmx.exit_qualification = ~0; 2834 2835 switch (rc) { 2836 case VMX_VMRESUME_ERROR: 2837 case VMX_VMLAUNCH_ERROR: 2838 case VMX_INVEPT_ERROR: 2839 vmexit->u.vmx.inst_type = rc; 2840 break; 2841 default: 2842 panic("vm_exit_inst_error: vmx_enter_guest returned %d", rc); 2843 } 2844 } 2845 2846 /* 2847 * If the NMI-exiting VM execution control is set to '1' then an NMI in 2848 * non-root operation causes a VM-exit. NMI blocking is in effect so it is 2849 * sufficient to simply vector to the NMI handler via a software interrupt. 2850 * However, this must be done before maskable interrupts are enabled 2851 * otherwise the "iret" issued by an interrupt handler will incorrectly 2852 * clear NMI blocking. 2853 */ 2854 static __inline void 2855 vmx_exit_handle_nmi(struct vmx *vmx, int vcpuid, struct vm_exit *vmexit) 2856 { 2857 uint32_t intr_info; 2858 2859 KASSERT((read_rflags() & PSL_I) == 0, ("interrupts enabled")); 2860 2861 if (vmexit->u.vmx.exit_reason != EXIT_REASON_EXCEPTION) 2862 return; 2863 2864 intr_info = vmcs_read(VMCS_EXIT_INTR_INFO); 2865 KASSERT((intr_info & VMCS_INTR_VALID) != 0, 2866 ("VM exit interruption info invalid: %#x", intr_info)); 2867 2868 if ((intr_info & VMCS_INTR_T_MASK) == VMCS_INTR_T_NMI) { 2869 KASSERT((intr_info & 0xff) == IDT_NMI, ("VM exit due " 2870 "to NMI has invalid vector: %#x", intr_info)); 2871 VCPU_CTR0(vmx->vm, vcpuid, "Vectoring to NMI handler"); 2872 __asm __volatile("int $2"); 2873 } 2874 } 2875 2876 static __inline void 2877 vmx_dr_enter_guest(struct vmxctx *vmxctx) 2878 { 2879 register_t rflags; 2880 2881 /* Save host control debug registers. */ 2882 vmxctx->host_dr7 = rdr7(); 2883 vmxctx->host_debugctl = rdmsr(MSR_DEBUGCTLMSR); 2884 2885 /* 2886 * Disable debugging in DR7 and DEBUGCTL to avoid triggering 2887 * exceptions in the host based on the guest DRx values. The 2888 * guest DR7 and DEBUGCTL are saved/restored in the VMCS. 2889 */ 2890 load_dr7(0); 2891 wrmsr(MSR_DEBUGCTLMSR, 0); 2892 2893 /* 2894 * Disable single stepping the kernel to avoid corrupting the 2895 * guest DR6. A debugger might still be able to corrupt the 2896 * guest DR6 by setting a breakpoint after this point and then 2897 * single stepping. 2898 */ 2899 rflags = read_rflags(); 2900 vmxctx->host_tf = rflags & PSL_T; 2901 write_rflags(rflags & ~PSL_T); 2902 2903 /* Save host debug registers. */ 2904 vmxctx->host_dr0 = rdr0(); 2905 vmxctx->host_dr1 = rdr1(); 2906 vmxctx->host_dr2 = rdr2(); 2907 vmxctx->host_dr3 = rdr3(); 2908 vmxctx->host_dr6 = rdr6(); 2909 2910 /* Restore guest debug registers. */ 2911 load_dr0(vmxctx->guest_dr0); 2912 load_dr1(vmxctx->guest_dr1); 2913 load_dr2(vmxctx->guest_dr2); 2914 load_dr3(vmxctx->guest_dr3); 2915 load_dr6(vmxctx->guest_dr6); 2916 } 2917 2918 static __inline void 2919 vmx_dr_leave_guest(struct vmxctx *vmxctx) 2920 { 2921 2922 /* Save guest debug registers. */ 2923 vmxctx->guest_dr0 = rdr0(); 2924 vmxctx->guest_dr1 = rdr1(); 2925 vmxctx->guest_dr2 = rdr2(); 2926 vmxctx->guest_dr3 = rdr3(); 2927 vmxctx->guest_dr6 = rdr6(); 2928 2929 /* 2930 * Restore host debug registers. Restore DR7, DEBUGCTL, and 2931 * PSL_T last. 2932 */ 2933 load_dr0(vmxctx->host_dr0); 2934 load_dr1(vmxctx->host_dr1); 2935 load_dr2(vmxctx->host_dr2); 2936 load_dr3(vmxctx->host_dr3); 2937 load_dr6(vmxctx->host_dr6); 2938 wrmsr(MSR_DEBUGCTLMSR, vmxctx->host_debugctl); 2939 load_dr7(vmxctx->host_dr7); 2940 write_rflags(read_rflags() | vmxctx->host_tf); 2941 } 2942 2943 static int 2944 vmx_run(void *arg, int vcpu, register_t rip, pmap_t pmap, 2945 struct vm_eventinfo *evinfo) 2946 { 2947 int rc, handled, launched; 2948 struct vmx *vmx; 2949 struct vm *vm; 2950 struct vmxctx *vmxctx; 2951 struct vmcs *vmcs; 2952 struct vm_exit *vmexit; 2953 struct vlapic *vlapic; 2954 uint32_t exit_reason; 2955 struct region_descriptor gdtr, idtr; 2956 uint16_t ldt_sel; 2957 2958 vmx = arg; 2959 vm = vmx->vm; 2960 vmcs = &vmx->vmcs[vcpu]; 2961 vmxctx = &vmx->ctx[vcpu]; 2962 vlapic = vm_lapic(vm, vcpu); 2963 vmexit = vm_exitinfo(vm, vcpu); 2964 launched = 0; 2965 2966 KASSERT(vmxctx->pmap == pmap, 2967 ("pmap %p different than ctx pmap %p", pmap, vmxctx->pmap)); 2968 2969 vmx_msr_guest_enter(vmx, vcpu); 2970 2971 VMPTRLD(vmcs); 2972 2973 /* 2974 * XXX 2975 * We do this every time because we may setup the virtual machine 2976 * from a different process than the one that actually runs it. 2977 * 2978 * If the life of a virtual machine was spent entirely in the context 2979 * of a single process we could do this once in vmx_vminit(). 2980 */ 2981 vmcs_write(VMCS_HOST_CR3, rcr3()); 2982 2983 vmcs_write(VMCS_GUEST_RIP, rip); 2984 vmx_set_pcpu_defaults(vmx, vcpu, pmap); 2985 do { 2986 KASSERT(vmcs_guest_rip() == rip, ("%s: vmcs guest rip mismatch " 2987 "%#lx/%#lx", __func__, vmcs_guest_rip(), rip)); 2988 2989 handled = UNHANDLED; 2990 /* 2991 * Interrupts are disabled from this point on until the 2992 * guest starts executing. This is done for the following 2993 * reasons: 2994 * 2995 * If an AST is asserted on this thread after the check below, 2996 * then the IPI_AST notification will not be lost, because it 2997 * will cause a VM exit due to external interrupt as soon as 2998 * the guest state is loaded. 2999 * 3000 * A posted interrupt after 'vmx_inject_interrupts()' will 3001 * not be "lost" because it will be held pending in the host 3002 * APIC because interrupts are disabled. The pending interrupt 3003 * will be recognized as soon as the guest state is loaded. 3004 * 3005 * The same reasoning applies to the IPI generated by 3006 * pmap_invalidate_ept(). 3007 */ 3008 disable_intr(); 3009 vmx_inject_interrupts(vmx, vcpu, vlapic, rip); 3010 3011 /* 3012 * Check for vcpu suspension after injecting events because 3013 * vmx_inject_interrupts() can suspend the vcpu due to a 3014 * triple fault. 3015 */ 3016 if (vcpu_suspended(evinfo)) { 3017 enable_intr(); 3018 vm_exit_suspended(vmx->vm, vcpu, rip); 3019 break; 3020 } 3021 3022 if (vcpu_rendezvous_pending(evinfo)) { 3023 enable_intr(); 3024 vm_exit_rendezvous(vmx->vm, vcpu, rip); 3025 break; 3026 } 3027 3028 if (vcpu_reqidle(evinfo)) { 3029 enable_intr(); 3030 vm_exit_reqidle(vmx->vm, vcpu, rip); 3031 break; 3032 } 3033 3034 if (vcpu_should_yield(vm, vcpu)) { 3035 enable_intr(); 3036 vm_exit_astpending(vmx->vm, vcpu, rip); 3037 vmx_astpending_trace(vmx, vcpu, rip); 3038 handled = HANDLED; 3039 break; 3040 } 3041 3042 if (vcpu_debugged(vm, vcpu)) { 3043 enable_intr(); 3044 vm_exit_debug(vmx->vm, vcpu, rip); 3045 break; 3046 } 3047 3048 /* 3049 * If TPR Shadowing is enabled, the TPR Threshold 3050 * must be updated right before entering the guest. 3051 */ 3052 if (tpr_shadowing && !virtual_interrupt_delivery) { 3053 if ((vmx->cap[vcpu].proc_ctls & PROCBASED_USE_TPR_SHADOW) != 0) { 3054 vmcs_write(VMCS_TPR_THRESHOLD, vlapic_get_cr8(vlapic)); 3055 } 3056 } 3057 3058 /* 3059 * VM exits restore the base address but not the 3060 * limits of GDTR and IDTR. The VMCS only stores the 3061 * base address, so VM exits set the limits to 0xffff. 3062 * Save and restore the full GDTR and IDTR to restore 3063 * the limits. 3064 * 3065 * The VMCS does not save the LDTR at all, and VM 3066 * exits clear LDTR as if a NULL selector were loaded. 3067 * The userspace hypervisor probably doesn't use a 3068 * LDT, but save and restore it to be safe. 3069 */ 3070 sgdt(&gdtr); 3071 sidt(&idtr); 3072 ldt_sel = sldt(); 3073 3074 /* 3075 * The TSC_AUX MSR must be saved/restored while interrupts 3076 * are disabled so that it is not possible for the guest 3077 * TSC_AUX MSR value to be overwritten by the resume 3078 * portion of the IPI_SUSPEND codepath. This is why the 3079 * transition of this MSR is handled separately from those 3080 * handled by vmx_msr_guest_{enter,exit}(), which are ok to 3081 * be transitioned with preemption disabled but interrupts 3082 * enabled. 3083 * 3084 * These vmx_msr_guest_{enter,exit}_tsc_aux() calls can be 3085 * anywhere in this loop so long as they happen with 3086 * interrupts disabled. This location is chosen for 3087 * simplicity. 3088 */ 3089 vmx_msr_guest_enter_tsc_aux(vmx, vcpu); 3090 3091 vmx_run_trace(vmx, vcpu); 3092 vmx_dr_enter_guest(vmxctx); 3093 rc = vmx_enter_guest(vmxctx, vmx, launched); 3094 vmx_dr_leave_guest(vmxctx); 3095 3096 vmx_msr_guest_exit_tsc_aux(vmx, vcpu); 3097 3098 bare_lgdt(&gdtr); 3099 lidt(&idtr); 3100 lldt(ldt_sel); 3101 3102 /* Collect some information for VM exit processing */ 3103 vmexit->rip = rip = vmcs_guest_rip(); 3104 vmexit->inst_length = vmexit_instruction_length(); 3105 vmexit->u.vmx.exit_reason = exit_reason = vmcs_exit_reason(); 3106 vmexit->u.vmx.exit_qualification = vmcs_exit_qualification(); 3107 3108 /* Update 'nextrip' */ 3109 vmx->state[vcpu].nextrip = rip; 3110 3111 if (rc == VMX_GUEST_VMEXIT) { 3112 vmx_exit_handle_nmi(vmx, vcpu, vmexit); 3113 enable_intr(); 3114 handled = vmx_exit_process(vmx, vcpu, vmexit); 3115 } else { 3116 enable_intr(); 3117 vmx_exit_inst_error(vmxctx, rc, vmexit); 3118 } 3119 launched = 1; 3120 vmx_exit_trace(vmx, vcpu, rip, exit_reason, handled); 3121 rip = vmexit->rip; 3122 } while (handled); 3123 3124 /* 3125 * If a VM exit has been handled then the exitcode must be BOGUS 3126 * If a VM exit is not handled then the exitcode must not be BOGUS 3127 */ 3128 if ((handled && vmexit->exitcode != VM_EXITCODE_BOGUS) || 3129 (!handled && vmexit->exitcode == VM_EXITCODE_BOGUS)) { 3130 panic("Mismatch between handled (%d) and exitcode (%d)", 3131 handled, vmexit->exitcode); 3132 } 3133 3134 if (!handled) 3135 vmm_stat_incr(vm, vcpu, VMEXIT_USERSPACE, 1); 3136 3137 VCPU_CTR1(vm, vcpu, "returning from vmx_run: exitcode %d", 3138 vmexit->exitcode); 3139 3140 VMCLEAR(vmcs); 3141 vmx_msr_guest_exit(vmx, vcpu); 3142 3143 return (0); 3144 } 3145 3146 static void 3147 vmx_vmcleanup(void *arg) 3148 { 3149 int i; 3150 struct vmx *vmx = arg; 3151 uint16_t maxcpus; 3152 3153 if (apic_access_virtualization(vmx, 0)) 3154 vm_unmap_mmio(vmx->vm, DEFAULT_APIC_BASE, PAGE_SIZE); 3155 3156 maxcpus = vm_get_maxcpus(vmx->vm); 3157 for (i = 0; i < maxcpus; i++) 3158 vpid_free(vmx->state[i].vpid); 3159 3160 free(vmx, M_VMX); 3161 3162 return; 3163 } 3164 3165 static register_t * 3166 vmxctx_regptr(struct vmxctx *vmxctx, int reg) 3167 { 3168 3169 switch (reg) { 3170 case VM_REG_GUEST_RAX: 3171 return (&vmxctx->guest_rax); 3172 case VM_REG_GUEST_RBX: 3173 return (&vmxctx->guest_rbx); 3174 case VM_REG_GUEST_RCX: 3175 return (&vmxctx->guest_rcx); 3176 case VM_REG_GUEST_RDX: 3177 return (&vmxctx->guest_rdx); 3178 case VM_REG_GUEST_RSI: 3179 return (&vmxctx->guest_rsi); 3180 case VM_REG_GUEST_RDI: 3181 return (&vmxctx->guest_rdi); 3182 case VM_REG_GUEST_RBP: 3183 return (&vmxctx->guest_rbp); 3184 case VM_REG_GUEST_R8: 3185 return (&vmxctx->guest_r8); 3186 case VM_REG_GUEST_R9: 3187 return (&vmxctx->guest_r9); 3188 case VM_REG_GUEST_R10: 3189 return (&vmxctx->guest_r10); 3190 case VM_REG_GUEST_R11: 3191 return (&vmxctx->guest_r11); 3192 case VM_REG_GUEST_R12: 3193 return (&vmxctx->guest_r12); 3194 case VM_REG_GUEST_R13: 3195 return (&vmxctx->guest_r13); 3196 case VM_REG_GUEST_R14: 3197 return (&vmxctx->guest_r14); 3198 case VM_REG_GUEST_R15: 3199 return (&vmxctx->guest_r15); 3200 case VM_REG_GUEST_CR2: 3201 return (&vmxctx->guest_cr2); 3202 case VM_REG_GUEST_DR0: 3203 return (&vmxctx->guest_dr0); 3204 case VM_REG_GUEST_DR1: 3205 return (&vmxctx->guest_dr1); 3206 case VM_REG_GUEST_DR2: 3207 return (&vmxctx->guest_dr2); 3208 case VM_REG_GUEST_DR3: 3209 return (&vmxctx->guest_dr3); 3210 case VM_REG_GUEST_DR6: 3211 return (&vmxctx->guest_dr6); 3212 default: 3213 break; 3214 } 3215 return (NULL); 3216 } 3217 3218 static int 3219 vmxctx_getreg(struct vmxctx *vmxctx, int reg, uint64_t *retval) 3220 { 3221 register_t *regp; 3222 3223 if ((regp = vmxctx_regptr(vmxctx, reg)) != NULL) { 3224 *retval = *regp; 3225 return (0); 3226 } else 3227 return (EINVAL); 3228 } 3229 3230 static int 3231 vmxctx_setreg(struct vmxctx *vmxctx, int reg, uint64_t val) 3232 { 3233 register_t *regp; 3234 3235 if ((regp = vmxctx_regptr(vmxctx, reg)) != NULL) { 3236 *regp = val; 3237 return (0); 3238 } else 3239 return (EINVAL); 3240 } 3241 3242 static int 3243 vmx_get_intr_shadow(struct vmx *vmx, int vcpu, int running, uint64_t *retval) 3244 { 3245 uint64_t gi; 3246 int error; 3247 3248 error = vmcs_getreg(&vmx->vmcs[vcpu], running, 3249 VMCS_IDENT(VMCS_GUEST_INTERRUPTIBILITY), &gi); 3250 *retval = (gi & HWINTR_BLOCKING) ? 1 : 0; 3251 return (error); 3252 } 3253 3254 static int 3255 vmx_modify_intr_shadow(struct vmx *vmx, int vcpu, int running, uint64_t val) 3256 { 3257 struct vmcs *vmcs; 3258 uint64_t gi; 3259 int error, ident; 3260 3261 /* 3262 * Forcing the vcpu into an interrupt shadow is not supported. 3263 */ 3264 if (val) { 3265 error = EINVAL; 3266 goto done; 3267 } 3268 3269 vmcs = &vmx->vmcs[vcpu]; 3270 ident = VMCS_IDENT(VMCS_GUEST_INTERRUPTIBILITY); 3271 error = vmcs_getreg(vmcs, running, ident, &gi); 3272 if (error == 0) { 3273 gi &= ~HWINTR_BLOCKING; 3274 error = vmcs_setreg(vmcs, running, ident, gi); 3275 } 3276 done: 3277 VCPU_CTR2(vmx->vm, vcpu, "Setting intr_shadow to %#lx %s", val, 3278 error ? "failed" : "succeeded"); 3279 return (error); 3280 } 3281 3282 static int 3283 vmx_shadow_reg(int reg) 3284 { 3285 int shreg; 3286 3287 shreg = -1; 3288 3289 switch (reg) { 3290 case VM_REG_GUEST_CR0: 3291 shreg = VMCS_CR0_SHADOW; 3292 break; 3293 case VM_REG_GUEST_CR4: 3294 shreg = VMCS_CR4_SHADOW; 3295 break; 3296 default: 3297 break; 3298 } 3299 3300 return (shreg); 3301 } 3302 3303 static int 3304 vmx_getreg(void *arg, int vcpu, int reg, uint64_t *retval) 3305 { 3306 int running, hostcpu; 3307 struct vmx *vmx = arg; 3308 3309 running = vcpu_is_running(vmx->vm, vcpu, &hostcpu); 3310 if (running && hostcpu != curcpu) 3311 panic("vmx_getreg: %s%d is running", vm_name(vmx->vm), vcpu); 3312 3313 if (reg == VM_REG_GUEST_INTR_SHADOW) 3314 return (vmx_get_intr_shadow(vmx, vcpu, running, retval)); 3315 3316 if (vmxctx_getreg(&vmx->ctx[vcpu], reg, retval) == 0) 3317 return (0); 3318 3319 return (vmcs_getreg(&vmx->vmcs[vcpu], running, reg, retval)); 3320 } 3321 3322 static int 3323 vmx_setreg(void *arg, int vcpu, int reg, uint64_t val) 3324 { 3325 int error, hostcpu, running, shadow; 3326 uint64_t ctls; 3327 pmap_t pmap; 3328 struct vmx *vmx = arg; 3329 3330 running = vcpu_is_running(vmx->vm, vcpu, &hostcpu); 3331 if (running && hostcpu != curcpu) 3332 panic("vmx_setreg: %s%d is running", vm_name(vmx->vm), vcpu); 3333 3334 if (reg == VM_REG_GUEST_INTR_SHADOW) 3335 return (vmx_modify_intr_shadow(vmx, vcpu, running, val)); 3336 3337 if (vmxctx_setreg(&vmx->ctx[vcpu], reg, val) == 0) 3338 return (0); 3339 3340 /* Do not permit user write access to VMCS fields by offset. */ 3341 if (reg < 0) 3342 return (EINVAL); 3343 3344 error = vmcs_setreg(&vmx->vmcs[vcpu], running, reg, val); 3345 3346 if (error == 0) { 3347 /* 3348 * If the "load EFER" VM-entry control is 1 then the 3349 * value of EFER.LMA must be identical to "IA-32e mode guest" 3350 * bit in the VM-entry control. 3351 */ 3352 if ((entry_ctls & VM_ENTRY_LOAD_EFER) != 0 && 3353 (reg == VM_REG_GUEST_EFER)) { 3354 vmcs_getreg(&vmx->vmcs[vcpu], running, 3355 VMCS_IDENT(VMCS_ENTRY_CTLS), &ctls); 3356 if (val & EFER_LMA) 3357 ctls |= VM_ENTRY_GUEST_LMA; 3358 else 3359 ctls &= ~VM_ENTRY_GUEST_LMA; 3360 vmcs_setreg(&vmx->vmcs[vcpu], running, 3361 VMCS_IDENT(VMCS_ENTRY_CTLS), ctls); 3362 } 3363 3364 shadow = vmx_shadow_reg(reg); 3365 if (shadow > 0) { 3366 /* 3367 * Store the unmodified value in the shadow 3368 */ 3369 error = vmcs_setreg(&vmx->vmcs[vcpu], running, 3370 VMCS_IDENT(shadow), val); 3371 } 3372 3373 if (reg == VM_REG_GUEST_CR3) { 3374 /* 3375 * Invalidate the guest vcpu's TLB mappings to emulate 3376 * the behavior of updating %cr3. 3377 * 3378 * XXX the processor retains global mappings when %cr3 3379 * is updated but vmx_invvpid() does not. 3380 */ 3381 pmap = vmx->ctx[vcpu].pmap; 3382 vmx_invvpid(vmx, vcpu, pmap, running); 3383 } 3384 } 3385 3386 return (error); 3387 } 3388 3389 static int 3390 vmx_getdesc(void *arg, int vcpu, int reg, struct seg_desc *desc) 3391 { 3392 int hostcpu, running; 3393 struct vmx *vmx = arg; 3394 3395 running = vcpu_is_running(vmx->vm, vcpu, &hostcpu); 3396 if (running && hostcpu != curcpu) 3397 panic("vmx_getdesc: %s%d is running", vm_name(vmx->vm), vcpu); 3398 3399 return (vmcs_getdesc(&vmx->vmcs[vcpu], running, reg, desc)); 3400 } 3401 3402 static int 3403 vmx_setdesc(void *arg, int vcpu, int reg, struct seg_desc *desc) 3404 { 3405 int hostcpu, running; 3406 struct vmx *vmx = arg; 3407 3408 running = vcpu_is_running(vmx->vm, vcpu, &hostcpu); 3409 if (running && hostcpu != curcpu) 3410 panic("vmx_setdesc: %s%d is running", vm_name(vmx->vm), vcpu); 3411 3412 return (vmcs_setdesc(&vmx->vmcs[vcpu], running, reg, desc)); 3413 } 3414 3415 static int 3416 vmx_getcap(void *arg, int vcpu, int type, int *retval) 3417 { 3418 struct vmx *vmx = arg; 3419 int vcap; 3420 int ret; 3421 3422 ret = ENOENT; 3423 3424 vcap = vmx->cap[vcpu].set; 3425 3426 switch (type) { 3427 case VM_CAP_HALT_EXIT: 3428 if (cap_halt_exit) 3429 ret = 0; 3430 break; 3431 case VM_CAP_PAUSE_EXIT: 3432 if (cap_pause_exit) 3433 ret = 0; 3434 break; 3435 case VM_CAP_MTRAP_EXIT: 3436 if (cap_monitor_trap) 3437 ret = 0; 3438 break; 3439 case VM_CAP_RDPID: 3440 if (cap_rdpid) 3441 ret = 0; 3442 break; 3443 case VM_CAP_RDTSCP: 3444 if (cap_rdtscp) 3445 ret = 0; 3446 break; 3447 case VM_CAP_UNRESTRICTED_GUEST: 3448 if (cap_unrestricted_guest) 3449 ret = 0; 3450 break; 3451 case VM_CAP_ENABLE_INVPCID: 3452 if (cap_invpcid) 3453 ret = 0; 3454 break; 3455 case VM_CAP_BPT_EXIT: 3456 ret = 0; 3457 break; 3458 default: 3459 break; 3460 } 3461 3462 if (ret == 0) 3463 *retval = (vcap & (1 << type)) ? 1 : 0; 3464 3465 return (ret); 3466 } 3467 3468 static int 3469 vmx_setcap(void *arg, int vcpu, int type, int val) 3470 { 3471 struct vmx *vmx = arg; 3472 struct vmcs *vmcs = &vmx->vmcs[vcpu]; 3473 uint32_t baseval; 3474 uint32_t *pptr; 3475 int error; 3476 int flag; 3477 int reg; 3478 int retval; 3479 3480 retval = ENOENT; 3481 pptr = NULL; 3482 3483 switch (type) { 3484 case VM_CAP_HALT_EXIT: 3485 if (cap_halt_exit) { 3486 retval = 0; 3487 pptr = &vmx->cap[vcpu].proc_ctls; 3488 baseval = *pptr; 3489 flag = PROCBASED_HLT_EXITING; 3490 reg = VMCS_PRI_PROC_BASED_CTLS; 3491 } 3492 break; 3493 case VM_CAP_MTRAP_EXIT: 3494 if (cap_monitor_trap) { 3495 retval = 0; 3496 pptr = &vmx->cap[vcpu].proc_ctls; 3497 baseval = *pptr; 3498 flag = PROCBASED_MTF; 3499 reg = VMCS_PRI_PROC_BASED_CTLS; 3500 } 3501 break; 3502 case VM_CAP_PAUSE_EXIT: 3503 if (cap_pause_exit) { 3504 retval = 0; 3505 pptr = &vmx->cap[vcpu].proc_ctls; 3506 baseval = *pptr; 3507 flag = PROCBASED_PAUSE_EXITING; 3508 reg = VMCS_PRI_PROC_BASED_CTLS; 3509 } 3510 break; 3511 case VM_CAP_RDPID: 3512 case VM_CAP_RDTSCP: 3513 if (cap_rdpid || cap_rdtscp) 3514 /* 3515 * Choose not to support enabling/disabling 3516 * RDPID/RDTSCP via libvmmapi since, as per the 3517 * discussion in vmx_init(), RDPID/RDTSCP are 3518 * either always enabled or always disabled. 3519 */ 3520 error = EOPNOTSUPP; 3521 break; 3522 case VM_CAP_UNRESTRICTED_GUEST: 3523 if (cap_unrestricted_guest) { 3524 retval = 0; 3525 pptr = &vmx->cap[vcpu].proc_ctls2; 3526 baseval = *pptr; 3527 flag = PROCBASED2_UNRESTRICTED_GUEST; 3528 reg = VMCS_SEC_PROC_BASED_CTLS; 3529 } 3530 break; 3531 case VM_CAP_ENABLE_INVPCID: 3532 if (cap_invpcid) { 3533 retval = 0; 3534 pptr = &vmx->cap[vcpu].proc_ctls2; 3535 baseval = *pptr; 3536 flag = PROCBASED2_ENABLE_INVPCID; 3537 reg = VMCS_SEC_PROC_BASED_CTLS; 3538 } 3539 break; 3540 case VM_CAP_BPT_EXIT: 3541 retval = 0; 3542 3543 /* Don't change the bitmap if we are tracing all exceptions. */ 3544 if (vmx->cap[vcpu].exc_bitmap != 0xffffffff) { 3545 pptr = &vmx->cap[vcpu].exc_bitmap; 3546 baseval = *pptr; 3547 flag = (1 << IDT_BP); 3548 reg = VMCS_EXCEPTION_BITMAP; 3549 } 3550 break; 3551 default: 3552 break; 3553 } 3554 3555 if (retval) 3556 return (retval); 3557 3558 if (pptr != NULL) { 3559 if (val) { 3560 baseval |= flag; 3561 } else { 3562 baseval &= ~flag; 3563 } 3564 VMPTRLD(vmcs); 3565 error = vmwrite(reg, baseval); 3566 VMCLEAR(vmcs); 3567 3568 if (error) 3569 return (error); 3570 3571 /* 3572 * Update optional stored flags, and record 3573 * setting 3574 */ 3575 *pptr = baseval; 3576 } 3577 3578 if (val) { 3579 vmx->cap[vcpu].set |= (1 << type); 3580 } else { 3581 vmx->cap[vcpu].set &= ~(1 << type); 3582 } 3583 3584 return (0); 3585 } 3586 3587 struct vlapic_vtx { 3588 struct vlapic vlapic; 3589 struct pir_desc *pir_desc; 3590 struct vmx *vmx; 3591 u_int pending_prio; 3592 }; 3593 3594 #define VPR_PRIO_BIT(vpr) (1 << ((vpr) >> 4)) 3595 3596 #define VMX_CTR_PIR(vm, vcpuid, pir_desc, notify, vector, level, msg) \ 3597 do { \ 3598 VCPU_CTR2(vm, vcpuid, msg " assert %s-triggered vector %d", \ 3599 level ? "level" : "edge", vector); \ 3600 VCPU_CTR1(vm, vcpuid, msg " pir0 0x%016lx", pir_desc->pir[0]); \ 3601 VCPU_CTR1(vm, vcpuid, msg " pir1 0x%016lx", pir_desc->pir[1]); \ 3602 VCPU_CTR1(vm, vcpuid, msg " pir2 0x%016lx", pir_desc->pir[2]); \ 3603 VCPU_CTR1(vm, vcpuid, msg " pir3 0x%016lx", pir_desc->pir[3]); \ 3604 VCPU_CTR1(vm, vcpuid, msg " notify: %s", notify ? "yes" : "no");\ 3605 } while (0) 3606 3607 /* 3608 * vlapic->ops handlers that utilize the APICv hardware assist described in 3609 * Chapter 29 of the Intel SDM. 3610 */ 3611 static int 3612 vmx_set_intr_ready(struct vlapic *vlapic, int vector, bool level) 3613 { 3614 struct vlapic_vtx *vlapic_vtx; 3615 struct pir_desc *pir_desc; 3616 uint64_t mask; 3617 int idx, notify = 0; 3618 3619 vlapic_vtx = (struct vlapic_vtx *)vlapic; 3620 pir_desc = vlapic_vtx->pir_desc; 3621 3622 /* 3623 * Keep track of interrupt requests in the PIR descriptor. This is 3624 * because the virtual APIC page pointed to by the VMCS cannot be 3625 * modified if the vcpu is running. 3626 */ 3627 idx = vector / 64; 3628 mask = 1UL << (vector % 64); 3629 atomic_set_long(&pir_desc->pir[idx], mask); 3630 3631 /* 3632 * A notification is required whenever the 'pending' bit makes a 3633 * transition from 0->1. 3634 * 3635 * Even if the 'pending' bit is already asserted, notification about 3636 * the incoming interrupt may still be necessary. For example, if a 3637 * vCPU is HLTed with a high PPR, a low priority interrupt would cause 3638 * the 0->1 'pending' transition with a notification, but the vCPU 3639 * would ignore the interrupt for the time being. The same vCPU would 3640 * need to then be notified if a high-priority interrupt arrived which 3641 * satisfied the PPR. 3642 * 3643 * The priorities of interrupts injected while 'pending' is asserted 3644 * are tracked in a custom bitfield 'pending_prio'. Should the 3645 * to-be-injected interrupt exceed the priorities already present, the 3646 * notification is sent. The priorities recorded in 'pending_prio' are 3647 * cleared whenever the 'pending' bit makes another 0->1 transition. 3648 */ 3649 if (atomic_cmpset_long(&pir_desc->pending, 0, 1) != 0) { 3650 notify = 1; 3651 vlapic_vtx->pending_prio = 0; 3652 } else { 3653 const u_int old_prio = vlapic_vtx->pending_prio; 3654 const u_int prio_bit = VPR_PRIO_BIT(vector & APIC_TPR_INT); 3655 3656 if ((old_prio & prio_bit) == 0 && prio_bit > old_prio) { 3657 atomic_set_int(&vlapic_vtx->pending_prio, prio_bit); 3658 notify = 1; 3659 } 3660 } 3661 3662 VMX_CTR_PIR(vlapic->vm, vlapic->vcpuid, pir_desc, notify, vector, 3663 level, "vmx_set_intr_ready"); 3664 return (notify); 3665 } 3666 3667 static int 3668 vmx_pending_intr(struct vlapic *vlapic, int *vecptr) 3669 { 3670 struct vlapic_vtx *vlapic_vtx; 3671 struct pir_desc *pir_desc; 3672 struct LAPIC *lapic; 3673 uint64_t pending, pirval; 3674 uint32_t ppr, vpr; 3675 int i; 3676 3677 /* 3678 * This function is only expected to be called from the 'HLT' exit 3679 * handler which does not care about the vector that is pending. 3680 */ 3681 KASSERT(vecptr == NULL, ("vmx_pending_intr: vecptr must be NULL")); 3682 3683 vlapic_vtx = (struct vlapic_vtx *)vlapic; 3684 pir_desc = vlapic_vtx->pir_desc; 3685 3686 pending = atomic_load_acq_long(&pir_desc->pending); 3687 if (!pending) { 3688 /* 3689 * While a virtual interrupt may have already been 3690 * processed the actual delivery maybe pending the 3691 * interruptibility of the guest. Recognize a pending 3692 * interrupt by reevaluating virtual interrupts 3693 * following Section 29.2.1 in the Intel SDM Volume 3. 3694 */ 3695 struct vm_exit *vmexit; 3696 uint8_t rvi, ppr; 3697 3698 vmexit = vm_exitinfo(vlapic->vm, vlapic->vcpuid); 3699 KASSERT(vmexit->exitcode == VM_EXITCODE_HLT, 3700 ("vmx_pending_intr: exitcode not 'HLT'")); 3701 rvi = vmexit->u.hlt.intr_status & APIC_TPR_INT; 3702 lapic = vlapic->apic_page; 3703 ppr = lapic->ppr & APIC_TPR_INT; 3704 if (rvi > ppr) { 3705 return (1); 3706 } 3707 3708 return (0); 3709 } 3710 3711 /* 3712 * If there is an interrupt pending then it will be recognized only 3713 * if its priority is greater than the processor priority. 3714 * 3715 * Special case: if the processor priority is zero then any pending 3716 * interrupt will be recognized. 3717 */ 3718 lapic = vlapic->apic_page; 3719 ppr = lapic->ppr & APIC_TPR_INT; 3720 if (ppr == 0) 3721 return (1); 3722 3723 VCPU_CTR1(vlapic->vm, vlapic->vcpuid, "HLT with non-zero PPR %d", 3724 lapic->ppr); 3725 3726 vpr = 0; 3727 for (i = 3; i >= 0; i--) { 3728 pirval = pir_desc->pir[i]; 3729 if (pirval != 0) { 3730 vpr = (i * 64 + flsl(pirval) - 1) & APIC_TPR_INT; 3731 break; 3732 } 3733 } 3734 3735 /* 3736 * If the highest-priority pending interrupt falls short of the 3737 * processor priority of this vCPU, ensure that 'pending_prio' does not 3738 * have any stale bits which would preclude a higher-priority interrupt 3739 * from incurring a notification later. 3740 */ 3741 if (vpr <= ppr) { 3742 const u_int prio_bit = VPR_PRIO_BIT(vpr); 3743 const u_int old = vlapic_vtx->pending_prio; 3744 3745 if (old > prio_bit && (old & prio_bit) == 0) { 3746 vlapic_vtx->pending_prio = prio_bit; 3747 } 3748 return (0); 3749 } 3750 return (1); 3751 } 3752 3753 static void 3754 vmx_intr_accepted(struct vlapic *vlapic, int vector) 3755 { 3756 3757 panic("vmx_intr_accepted: not expected to be called"); 3758 } 3759 3760 static void 3761 vmx_set_tmr(struct vlapic *vlapic, int vector, bool level) 3762 { 3763 struct vlapic_vtx *vlapic_vtx; 3764 struct vmx *vmx; 3765 struct vmcs *vmcs; 3766 uint64_t mask, val; 3767 3768 KASSERT(vector >= 0 && vector <= 255, ("invalid vector %d", vector)); 3769 KASSERT(!vcpu_is_running(vlapic->vm, vlapic->vcpuid, NULL), 3770 ("vmx_set_tmr: vcpu cannot be running")); 3771 3772 vlapic_vtx = (struct vlapic_vtx *)vlapic; 3773 vmx = vlapic_vtx->vmx; 3774 vmcs = &vmx->vmcs[vlapic->vcpuid]; 3775 mask = 1UL << (vector % 64); 3776 3777 VMPTRLD(vmcs); 3778 val = vmcs_read(VMCS_EOI_EXIT(vector)); 3779 if (level) 3780 val |= mask; 3781 else 3782 val &= ~mask; 3783 vmcs_write(VMCS_EOI_EXIT(vector), val); 3784 VMCLEAR(vmcs); 3785 } 3786 3787 static void 3788 vmx_enable_x2apic_mode_ts(struct vlapic *vlapic) 3789 { 3790 struct vmx *vmx; 3791 struct vmcs *vmcs; 3792 uint32_t proc_ctls; 3793 int vcpuid; 3794 3795 vcpuid = vlapic->vcpuid; 3796 vmx = ((struct vlapic_vtx *)vlapic)->vmx; 3797 vmcs = &vmx->vmcs[vcpuid]; 3798 3799 proc_ctls = vmx->cap[vcpuid].proc_ctls; 3800 proc_ctls &= ~PROCBASED_USE_TPR_SHADOW; 3801 proc_ctls |= PROCBASED_CR8_LOAD_EXITING; 3802 proc_ctls |= PROCBASED_CR8_STORE_EXITING; 3803 vmx->cap[vcpuid].proc_ctls = proc_ctls; 3804 3805 VMPTRLD(vmcs); 3806 vmcs_write(VMCS_PRI_PROC_BASED_CTLS, proc_ctls); 3807 VMCLEAR(vmcs); 3808 } 3809 3810 static void 3811 vmx_enable_x2apic_mode_vid(struct vlapic *vlapic) 3812 { 3813 struct vmx *vmx; 3814 struct vmcs *vmcs; 3815 uint32_t proc_ctls2; 3816 int vcpuid, error; 3817 3818 vcpuid = vlapic->vcpuid; 3819 vmx = ((struct vlapic_vtx *)vlapic)->vmx; 3820 vmcs = &vmx->vmcs[vcpuid]; 3821 3822 proc_ctls2 = vmx->cap[vcpuid].proc_ctls2; 3823 KASSERT((proc_ctls2 & PROCBASED2_VIRTUALIZE_APIC_ACCESSES) != 0, 3824 ("%s: invalid proc_ctls2 %#x", __func__, proc_ctls2)); 3825 3826 proc_ctls2 &= ~PROCBASED2_VIRTUALIZE_APIC_ACCESSES; 3827 proc_ctls2 |= PROCBASED2_VIRTUALIZE_X2APIC_MODE; 3828 vmx->cap[vcpuid].proc_ctls2 = proc_ctls2; 3829 3830 VMPTRLD(vmcs); 3831 vmcs_write(VMCS_SEC_PROC_BASED_CTLS, proc_ctls2); 3832 VMCLEAR(vmcs); 3833 3834 if (vlapic->vcpuid == 0) { 3835 /* 3836 * The nested page table mappings are shared by all vcpus 3837 * so unmap the APIC access page just once. 3838 */ 3839 error = vm_unmap_mmio(vmx->vm, DEFAULT_APIC_BASE, PAGE_SIZE); 3840 KASSERT(error == 0, ("%s: vm_unmap_mmio error %d", 3841 __func__, error)); 3842 3843 /* 3844 * The MSR bitmap is shared by all vcpus so modify it only 3845 * once in the context of vcpu 0. 3846 */ 3847 error = vmx_allow_x2apic_msrs(vmx); 3848 KASSERT(error == 0, ("%s: vmx_allow_x2apic_msrs error %d", 3849 __func__, error)); 3850 } 3851 } 3852 3853 static void 3854 vmx_post_intr(struct vlapic *vlapic, int hostcpu) 3855 { 3856 3857 ipi_cpu(hostcpu, pirvec); 3858 } 3859 3860 /* 3861 * Transfer the pending interrupts in the PIR descriptor to the IRR 3862 * in the virtual APIC page. 3863 */ 3864 static void 3865 vmx_inject_pir(struct vlapic *vlapic) 3866 { 3867 struct vlapic_vtx *vlapic_vtx; 3868 struct pir_desc *pir_desc; 3869 struct LAPIC *lapic; 3870 uint64_t val, pirval; 3871 int rvi, pirbase = -1; 3872 uint16_t intr_status_old, intr_status_new; 3873 3874 vlapic_vtx = (struct vlapic_vtx *)vlapic; 3875 pir_desc = vlapic_vtx->pir_desc; 3876 if (atomic_cmpset_long(&pir_desc->pending, 1, 0) == 0) { 3877 VCPU_CTR0(vlapic->vm, vlapic->vcpuid, "vmx_inject_pir: " 3878 "no posted interrupt pending"); 3879 return; 3880 } 3881 3882 pirval = 0; 3883 pirbase = -1; 3884 lapic = vlapic->apic_page; 3885 3886 val = atomic_readandclear_long(&pir_desc->pir[0]); 3887 if (val != 0) { 3888 lapic->irr0 |= val; 3889 lapic->irr1 |= val >> 32; 3890 pirbase = 0; 3891 pirval = val; 3892 } 3893 3894 val = atomic_readandclear_long(&pir_desc->pir[1]); 3895 if (val != 0) { 3896 lapic->irr2 |= val; 3897 lapic->irr3 |= val >> 32; 3898 pirbase = 64; 3899 pirval = val; 3900 } 3901 3902 val = atomic_readandclear_long(&pir_desc->pir[2]); 3903 if (val != 0) { 3904 lapic->irr4 |= val; 3905 lapic->irr5 |= val >> 32; 3906 pirbase = 128; 3907 pirval = val; 3908 } 3909 3910 val = atomic_readandclear_long(&pir_desc->pir[3]); 3911 if (val != 0) { 3912 lapic->irr6 |= val; 3913 lapic->irr7 |= val >> 32; 3914 pirbase = 192; 3915 pirval = val; 3916 } 3917 3918 VLAPIC_CTR_IRR(vlapic, "vmx_inject_pir"); 3919 3920 /* 3921 * Update RVI so the processor can evaluate pending virtual 3922 * interrupts on VM-entry. 3923 * 3924 * It is possible for pirval to be 0 here, even though the 3925 * pending bit has been set. The scenario is: 3926 * CPU-Y is sending a posted interrupt to CPU-X, which 3927 * is running a guest and processing posted interrupts in h/w. 3928 * CPU-X will eventually exit and the state seen in s/w is 3929 * the pending bit set, but no PIR bits set. 3930 * 3931 * CPU-X CPU-Y 3932 * (vm running) (host running) 3933 * rx posted interrupt 3934 * CLEAR pending bit 3935 * SET PIR bit 3936 * READ/CLEAR PIR bits 3937 * SET pending bit 3938 * (vm exit) 3939 * pending bit set, PIR 0 3940 */ 3941 if (pirval != 0) { 3942 rvi = pirbase + flsl(pirval) - 1; 3943 intr_status_old = vmcs_read(VMCS_GUEST_INTR_STATUS); 3944 intr_status_new = (intr_status_old & 0xFF00) | rvi; 3945 if (intr_status_new > intr_status_old) { 3946 vmcs_write(VMCS_GUEST_INTR_STATUS, intr_status_new); 3947 VCPU_CTR2(vlapic->vm, vlapic->vcpuid, "vmx_inject_pir: " 3948 "guest_intr_status changed from 0x%04x to 0x%04x", 3949 intr_status_old, intr_status_new); 3950 } 3951 } 3952 } 3953 3954 static struct vlapic * 3955 vmx_vlapic_init(void *arg, int vcpuid) 3956 { 3957 struct vmx *vmx; 3958 struct vlapic *vlapic; 3959 struct vlapic_vtx *vlapic_vtx; 3960 3961 vmx = arg; 3962 3963 vlapic = malloc(sizeof(struct vlapic_vtx), M_VLAPIC, M_WAITOK | M_ZERO); 3964 vlapic->vm = vmx->vm; 3965 vlapic->vcpuid = vcpuid; 3966 vlapic->apic_page = (struct LAPIC *)&vmx->apic_page[vcpuid]; 3967 3968 vlapic_vtx = (struct vlapic_vtx *)vlapic; 3969 vlapic_vtx->pir_desc = &vmx->pir_desc[vcpuid]; 3970 vlapic_vtx->vmx = vmx; 3971 3972 if (tpr_shadowing) { 3973 vlapic->ops.enable_x2apic_mode = vmx_enable_x2apic_mode_ts; 3974 } 3975 3976 if (virtual_interrupt_delivery) { 3977 vlapic->ops.set_intr_ready = vmx_set_intr_ready; 3978 vlapic->ops.pending_intr = vmx_pending_intr; 3979 vlapic->ops.intr_accepted = vmx_intr_accepted; 3980 vlapic->ops.set_tmr = vmx_set_tmr; 3981 vlapic->ops.enable_x2apic_mode = vmx_enable_x2apic_mode_vid; 3982 } 3983 3984 if (posted_interrupts) 3985 vlapic->ops.post_intr = vmx_post_intr; 3986 3987 vlapic_init(vlapic); 3988 3989 return (vlapic); 3990 } 3991 3992 static void 3993 vmx_vlapic_cleanup(void *arg, struct vlapic *vlapic) 3994 { 3995 3996 vlapic_cleanup(vlapic); 3997 free(vlapic, M_VLAPIC); 3998 } 3999 4000 #ifdef BHYVE_SNAPSHOT 4001 static int 4002 vmx_snapshot_vmi(void *arg, struct vm_snapshot_meta *meta) 4003 { 4004 struct vmx *vmx; 4005 struct vmxctx *vmxctx; 4006 int i; 4007 int ret; 4008 4009 vmx = arg; 4010 4011 KASSERT(vmx != NULL, ("%s: arg was NULL", __func__)); 4012 4013 for (i = 0; i < VM_MAXCPU; i++) { 4014 SNAPSHOT_BUF_OR_LEAVE(vmx->guest_msrs[i], 4015 sizeof(vmx->guest_msrs[i]), meta, ret, done); 4016 4017 vmxctx = &vmx->ctx[i]; 4018 SNAPSHOT_VAR_OR_LEAVE(vmxctx->guest_rdi, meta, ret, done); 4019 SNAPSHOT_VAR_OR_LEAVE(vmxctx->guest_rsi, meta, ret, done); 4020 SNAPSHOT_VAR_OR_LEAVE(vmxctx->guest_rdx, meta, ret, done); 4021 SNAPSHOT_VAR_OR_LEAVE(vmxctx->guest_rcx, meta, ret, done); 4022 SNAPSHOT_VAR_OR_LEAVE(vmxctx->guest_r8, meta, ret, done); 4023 SNAPSHOT_VAR_OR_LEAVE(vmxctx->guest_r9, meta, ret, done); 4024 SNAPSHOT_VAR_OR_LEAVE(vmxctx->guest_rax, meta, ret, done); 4025 SNAPSHOT_VAR_OR_LEAVE(vmxctx->guest_rbx, meta, ret, done); 4026 SNAPSHOT_VAR_OR_LEAVE(vmxctx->guest_rbp, meta, ret, done); 4027 SNAPSHOT_VAR_OR_LEAVE(vmxctx->guest_r10, meta, ret, done); 4028 SNAPSHOT_VAR_OR_LEAVE(vmxctx->guest_r11, meta, ret, done); 4029 SNAPSHOT_VAR_OR_LEAVE(vmxctx->guest_r12, meta, ret, done); 4030 SNAPSHOT_VAR_OR_LEAVE(vmxctx->guest_r13, meta, ret, done); 4031 SNAPSHOT_VAR_OR_LEAVE(vmxctx->guest_r14, meta, ret, done); 4032 SNAPSHOT_VAR_OR_LEAVE(vmxctx->guest_r15, meta, ret, done); 4033 SNAPSHOT_VAR_OR_LEAVE(vmxctx->guest_cr2, meta, ret, done); 4034 SNAPSHOT_VAR_OR_LEAVE(vmxctx->guest_dr0, meta, ret, done); 4035 SNAPSHOT_VAR_OR_LEAVE(vmxctx->guest_dr1, meta, ret, done); 4036 SNAPSHOT_VAR_OR_LEAVE(vmxctx->guest_dr2, meta, ret, done); 4037 SNAPSHOT_VAR_OR_LEAVE(vmxctx->guest_dr3, meta, ret, done); 4038 SNAPSHOT_VAR_OR_LEAVE(vmxctx->guest_dr6, meta, ret, done); 4039 } 4040 4041 done: 4042 return (ret); 4043 } 4044 4045 static int 4046 vmx_snapshot_vmcx(void *arg, struct vm_snapshot_meta *meta, int vcpu) 4047 { 4048 struct vmcs *vmcs; 4049 struct vmx *vmx; 4050 int err, run, hostcpu; 4051 4052 vmx = (struct vmx *)arg; 4053 err = 0; 4054 4055 KASSERT(arg != NULL, ("%s: arg was NULL", __func__)); 4056 vmcs = &vmx->vmcs[vcpu]; 4057 4058 run = vcpu_is_running(vmx->vm, vcpu, &hostcpu); 4059 if (run && hostcpu != curcpu) { 4060 printf("%s: %s%d is running", __func__, vm_name(vmx->vm), vcpu); 4061 return (EINVAL); 4062 } 4063 4064 err += vmcs_snapshot_reg(vmcs, run, VM_REG_GUEST_CR0, meta); 4065 err += vmcs_snapshot_reg(vmcs, run, VM_REG_GUEST_CR3, meta); 4066 err += vmcs_snapshot_reg(vmcs, run, VM_REG_GUEST_CR4, meta); 4067 err += vmcs_snapshot_reg(vmcs, run, VM_REG_GUEST_DR7, meta); 4068 err += vmcs_snapshot_reg(vmcs, run, VM_REG_GUEST_RSP, meta); 4069 err += vmcs_snapshot_reg(vmcs, run, VM_REG_GUEST_RIP, meta); 4070 err += vmcs_snapshot_reg(vmcs, run, VM_REG_GUEST_RFLAGS, meta); 4071 4072 /* Guest segments */ 4073 err += vmcs_snapshot_reg(vmcs, run, VM_REG_GUEST_ES, meta); 4074 err += vmcs_snapshot_desc(vmcs, run, VM_REG_GUEST_ES, meta); 4075 4076 err += vmcs_snapshot_reg(vmcs, run, VM_REG_GUEST_CS, meta); 4077 err += vmcs_snapshot_desc(vmcs, run, VM_REG_GUEST_CS, meta); 4078 4079 err += vmcs_snapshot_reg(vmcs, run, VM_REG_GUEST_SS, meta); 4080 err += vmcs_snapshot_desc(vmcs, run, VM_REG_GUEST_SS, meta); 4081 4082 err += vmcs_snapshot_reg(vmcs, run, VM_REG_GUEST_DS, meta); 4083 err += vmcs_snapshot_desc(vmcs, run, VM_REG_GUEST_DS, meta); 4084 4085 err += vmcs_snapshot_reg(vmcs, run, VM_REG_GUEST_FS, meta); 4086 err += vmcs_snapshot_desc(vmcs, run, VM_REG_GUEST_FS, meta); 4087 4088 err += vmcs_snapshot_reg(vmcs, run, VM_REG_GUEST_GS, meta); 4089 err += vmcs_snapshot_desc(vmcs, run, VM_REG_GUEST_GS, meta); 4090 4091 err += vmcs_snapshot_reg(vmcs, run, VM_REG_GUEST_TR, meta); 4092 err += vmcs_snapshot_desc(vmcs, run, VM_REG_GUEST_TR, meta); 4093 4094 err += vmcs_snapshot_reg(vmcs, run, VM_REG_GUEST_LDTR, meta); 4095 err += vmcs_snapshot_desc(vmcs, run, VM_REG_GUEST_LDTR, meta); 4096 4097 err += vmcs_snapshot_reg(vmcs, run, VM_REG_GUEST_EFER, meta); 4098 4099 err += vmcs_snapshot_desc(vmcs, run, VM_REG_GUEST_IDTR, meta); 4100 err += vmcs_snapshot_desc(vmcs, run, VM_REG_GUEST_GDTR, meta); 4101 4102 /* Guest page tables */ 4103 err += vmcs_snapshot_reg(vmcs, run, VM_REG_GUEST_PDPTE0, meta); 4104 err += vmcs_snapshot_reg(vmcs, run, VM_REG_GUEST_PDPTE1, meta); 4105 err += vmcs_snapshot_reg(vmcs, run, VM_REG_GUEST_PDPTE2, meta); 4106 err += vmcs_snapshot_reg(vmcs, run, VM_REG_GUEST_PDPTE3, meta); 4107 4108 /* Other guest state */ 4109 err += vmcs_snapshot_any(vmcs, run, VMCS_GUEST_IA32_SYSENTER_CS, meta); 4110 err += vmcs_snapshot_any(vmcs, run, VMCS_GUEST_IA32_SYSENTER_ESP, meta); 4111 err += vmcs_snapshot_any(vmcs, run, VMCS_GUEST_IA32_SYSENTER_EIP, meta); 4112 err += vmcs_snapshot_any(vmcs, run, VMCS_GUEST_INTERRUPTIBILITY, meta); 4113 err += vmcs_snapshot_any(vmcs, run, VMCS_GUEST_ACTIVITY, meta); 4114 err += vmcs_snapshot_any(vmcs, run, VMCS_ENTRY_CTLS, meta); 4115 err += vmcs_snapshot_any(vmcs, run, VMCS_EXIT_CTLS, meta); 4116 4117 return (err); 4118 } 4119 4120 static int 4121 vmx_restore_tsc(void *arg, int vcpu, uint64_t offset) 4122 { 4123 struct vmcs *vmcs; 4124 struct vmx *vmx = (struct vmx *)arg; 4125 int error, running, hostcpu; 4126 4127 KASSERT(arg != NULL, ("%s: arg was NULL", __func__)); 4128 vmcs = &vmx->vmcs[vcpu]; 4129 4130 running = vcpu_is_running(vmx->vm, vcpu, &hostcpu); 4131 if (running && hostcpu != curcpu) { 4132 printf("%s: %s%d is running", __func__, vm_name(vmx->vm), vcpu); 4133 return (EINVAL); 4134 } 4135 4136 if (!running) 4137 VMPTRLD(vmcs); 4138 4139 error = vmx_set_tsc_offset(vmx, vcpu, offset); 4140 4141 if (!running) 4142 VMCLEAR(vmcs); 4143 return (error); 4144 } 4145 #endif 4146 4147 struct vmm_ops vmm_ops_intel = { 4148 .init = vmx_init, 4149 .cleanup = vmx_cleanup, 4150 .resume = vmx_restore, 4151 .vminit = vmx_vminit, 4152 .vmrun = vmx_run, 4153 .vmcleanup = vmx_vmcleanup, 4154 .vmgetreg = vmx_getreg, 4155 .vmsetreg = vmx_setreg, 4156 .vmgetdesc = vmx_getdesc, 4157 .vmsetdesc = vmx_setdesc, 4158 .vmgetcap = vmx_getcap, 4159 .vmsetcap = vmx_setcap, 4160 .vmspace_alloc = ept_vmspace_alloc, 4161 .vmspace_free = ept_vmspace_free, 4162 .vlapic_init = vmx_vlapic_init, 4163 .vlapic_cleanup = vmx_vlapic_cleanup, 4164 #ifdef BHYVE_SNAPSHOT 4165 .vmsnapshot = vmx_snapshot_vmi, 4166 .vmcx_snapshot = vmx_snapshot_vmcx, 4167 .vm_restore_tsc = vmx_restore_tsc, 4168 #endif 4169 }; 4170