1 /*- 2 * SPDX-License-Identifier: BSD-2-Clause 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 30 #include <sys/cdefs.h> 31 #include "opt_bhyve_snapshot.h" 32 33 #include <sys/param.h> 34 #include <sys/systm.h> 35 #include <sys/smp.h> 36 #include <sys/kernel.h> 37 #include <sys/malloc.h> 38 #include <sys/pcpu.h> 39 #include <sys/proc.h> 40 #include <sys/reg.h> 41 #include <sys/smr.h> 42 #include <sys/sysctl.h> 43 44 #include <vm/vm.h> 45 #include <vm/vm_extern.h> 46 #include <vm/pmap.h> 47 48 #include <machine/psl.h> 49 #include <machine/cpufunc.h> 50 #include <machine/md_var.h> 51 #include <machine/segments.h> 52 #include <machine/smp.h> 53 #include <machine/specialreg.h> 54 #include <machine/vmparam.h> 55 56 #include <machine/vmm.h> 57 #include <machine/vmm_dev.h> 58 #include <machine/vmm_instruction_emul.h> 59 #include <machine/vmm_snapshot.h> 60 61 #include <dev/vmm/vmm_ktr.h> 62 #include <dev/vmm/vmm_mem.h> 63 64 #include "vmm_lapic.h" 65 #include "vmm_host.h" 66 #include "vmm_ioport.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 #include "io/ppt.h" 79 80 #define PINBASED_CTLS_ONE_SETTING \ 81 (PINBASED_EXTINT_EXITING | \ 82 PINBASED_NMI_EXITING | \ 83 PINBASED_VIRTUAL_NMI) 84 #define PINBASED_CTLS_ZERO_SETTING 0 85 86 #define PROCBASED_CTLS_WINDOW_SETTING \ 87 (PROCBASED_INT_WINDOW_EXITING | \ 88 PROCBASED_NMI_WINDOW_EXITING) 89 90 #define PROCBASED_CTLS_ONE_SETTING \ 91 (PROCBASED_SECONDARY_CONTROLS | \ 92 PROCBASED_MWAIT_EXITING | \ 93 PROCBASED_MONITOR_EXITING | \ 94 PROCBASED_IO_EXITING | \ 95 PROCBASED_MSR_BITMAPS | \ 96 PROCBASED_CTLS_WINDOW_SETTING | \ 97 PROCBASED_CR8_LOAD_EXITING | \ 98 PROCBASED_CR8_STORE_EXITING) 99 #define PROCBASED_CTLS_ZERO_SETTING \ 100 (PROCBASED_CR3_LOAD_EXITING | \ 101 PROCBASED_CR3_STORE_EXITING | \ 102 PROCBASED_IO_BITMAPS) 103 104 #define PROCBASED_CTLS2_ONE_SETTING PROCBASED2_ENABLE_EPT 105 #define PROCBASED_CTLS2_ZERO_SETTING 0 106 107 #define VM_EXIT_CTLS_ONE_SETTING \ 108 (VM_EXIT_SAVE_DEBUG_CONTROLS | \ 109 VM_EXIT_HOST_LMA | \ 110 VM_EXIT_SAVE_EFER | \ 111 VM_EXIT_LOAD_EFER | \ 112 VM_EXIT_ACKNOWLEDGE_INTERRUPT) 113 114 #define VM_EXIT_CTLS_ZERO_SETTING 0 115 116 #define VM_ENTRY_CTLS_ONE_SETTING \ 117 (VM_ENTRY_LOAD_DEBUG_CONTROLS | \ 118 VM_ENTRY_LOAD_EFER) 119 120 #define VM_ENTRY_CTLS_ZERO_SETTING \ 121 (VM_ENTRY_INTO_SMM | \ 122 VM_ENTRY_DEACTIVATE_DUAL_MONITOR) 123 124 #define HANDLED 1 125 #define UNHANDLED 0 126 127 static MALLOC_DEFINE(M_VMX, "vmx", "vmx"); 128 static MALLOC_DEFINE(M_VLAPIC, "vlapic", "vlapic"); 129 130 bool vmx_have_msr_tsc_aux; 131 132 SYSCTL_DECL(_hw_vmm); 133 SYSCTL_NODE(_hw_vmm, OID_AUTO, vmx, CTLFLAG_RW | CTLFLAG_MPSAFE, NULL, 134 NULL); 135 136 int vmxon_enabled[MAXCPU]; 137 static uint8_t *vmxon_region; 138 139 static uint32_t pinbased_ctls, procbased_ctls, procbased_ctls2; 140 static uint32_t exit_ctls, entry_ctls; 141 142 static uint64_t cr0_ones_mask, cr0_zeros_mask; 143 SYSCTL_ULONG(_hw_vmm_vmx, OID_AUTO, cr0_ones_mask, CTLFLAG_RD, 144 &cr0_ones_mask, 0, NULL); 145 SYSCTL_ULONG(_hw_vmm_vmx, OID_AUTO, cr0_zeros_mask, CTLFLAG_RD, 146 &cr0_zeros_mask, 0, NULL); 147 148 static uint64_t cr4_ones_mask, cr4_zeros_mask; 149 SYSCTL_ULONG(_hw_vmm_vmx, OID_AUTO, cr4_ones_mask, CTLFLAG_RD, 150 &cr4_ones_mask, 0, NULL); 151 SYSCTL_ULONG(_hw_vmm_vmx, OID_AUTO, cr4_zeros_mask, CTLFLAG_RD, 152 &cr4_zeros_mask, 0, NULL); 153 154 static int vmx_initialized; 155 SYSCTL_INT(_hw_vmm_vmx, OID_AUTO, initialized, CTLFLAG_RD, 156 &vmx_initialized, 0, "Intel VMX initialized"); 157 158 /* 159 * Optional capabilities 160 */ 161 static SYSCTL_NODE(_hw_vmm_vmx, OID_AUTO, cap, 162 CTLFLAG_RW | CTLFLAG_MPSAFE, NULL, 163 NULL); 164 165 static int cap_halt_exit; 166 SYSCTL_INT(_hw_vmm_vmx_cap, OID_AUTO, halt_exit, CTLFLAG_RD, &cap_halt_exit, 0, 167 "HLT triggers a VM-exit"); 168 169 static int cap_pause_exit; 170 SYSCTL_INT(_hw_vmm_vmx_cap, OID_AUTO, pause_exit, CTLFLAG_RD, &cap_pause_exit, 171 0, "PAUSE triggers a VM-exit"); 172 173 static int cap_wbinvd_exit; 174 SYSCTL_INT(_hw_vmm_vmx_cap, OID_AUTO, wbinvd_exit, CTLFLAG_RD, &cap_wbinvd_exit, 175 0, "WBINVD triggers a VM-exit"); 176 177 static int cap_rdpid; 178 SYSCTL_INT(_hw_vmm_vmx_cap, OID_AUTO, rdpid, CTLFLAG_RD, &cap_rdpid, 0, 179 "Guests are allowed to use RDPID"); 180 181 static int cap_rdtscp; 182 SYSCTL_INT(_hw_vmm_vmx_cap, OID_AUTO, rdtscp, CTLFLAG_RD, &cap_rdtscp, 0, 183 "Guests are allowed to use RDTSCP"); 184 185 static int cap_unrestricted_guest; 186 SYSCTL_INT(_hw_vmm_vmx_cap, OID_AUTO, unrestricted_guest, CTLFLAG_RD, 187 &cap_unrestricted_guest, 0, "Unrestricted guests"); 188 189 static int cap_monitor_trap; 190 SYSCTL_INT(_hw_vmm_vmx_cap, OID_AUTO, monitor_trap, CTLFLAG_RD, 191 &cap_monitor_trap, 0, "Monitor trap flag"); 192 193 static int cap_invpcid; 194 SYSCTL_INT(_hw_vmm_vmx_cap, OID_AUTO, invpcid, CTLFLAG_RD, &cap_invpcid, 195 0, "Guests are allowed to use INVPCID"); 196 197 static int tpr_shadowing; 198 SYSCTL_INT(_hw_vmm_vmx_cap, OID_AUTO, tpr_shadowing, 199 CTLFLAG_RDTUN | CTLFLAG_NOFETCH, 200 &tpr_shadowing, 0, "TPR shadowing support"); 201 202 static int virtual_interrupt_delivery; 203 SYSCTL_INT(_hw_vmm_vmx_cap, OID_AUTO, virtual_interrupt_delivery, 204 CTLFLAG_RDTUN | CTLFLAG_NOFETCH, 205 &virtual_interrupt_delivery, 0, "APICv virtual interrupt delivery support"); 206 207 static int posted_interrupts; 208 SYSCTL_INT(_hw_vmm_vmx_cap, OID_AUTO, posted_interrupts, 209 CTLFLAG_RDTUN | CTLFLAG_NOFETCH, 210 &posted_interrupts, 0, "APICv posted interrupt support"); 211 212 static int pirvec = -1; 213 SYSCTL_INT(_hw_vmm_vmx, OID_AUTO, posted_interrupt_vector, CTLFLAG_RD, 214 &pirvec, 0, "APICv posted interrupt vector"); 215 216 static struct unrhdr *vpid_unr; 217 static u_int vpid_alloc_failed; 218 SYSCTL_UINT(_hw_vmm_vmx, OID_AUTO, vpid_alloc_failed, CTLFLAG_RD, 219 &vpid_alloc_failed, 0, NULL); 220 221 int guest_l1d_flush; 222 SYSCTL_INT(_hw_vmm_vmx, OID_AUTO, l1d_flush, CTLFLAG_RDTUN | CTLFLAG_NOFETCH, 223 &guest_l1d_flush, 0, NULL); 224 int guest_l1d_flush_sw; 225 SYSCTL_INT(_hw_vmm_vmx, OID_AUTO, l1d_flush_sw, CTLFLAG_RDTUN | CTLFLAG_NOFETCH, 226 &guest_l1d_flush_sw, 0, NULL); 227 228 static struct msr_entry msr_load_list[1] __aligned(16); 229 230 /* 231 * The definitions of SDT probes for VMX. 232 */ 233 234 SDT_PROBE_DEFINE3(vmm, vmx, exit, entry, 235 "struct vmx *", "int", "struct vm_exit *"); 236 237 SDT_PROBE_DEFINE4(vmm, vmx, exit, taskswitch, 238 "struct vmx *", "int", "struct vm_exit *", "struct vm_task_switch *"); 239 240 SDT_PROBE_DEFINE4(vmm, vmx, exit, craccess, 241 "struct vmx *", "int", "struct vm_exit *", "uint64_t"); 242 243 SDT_PROBE_DEFINE4(vmm, vmx, exit, rdmsr, 244 "struct vmx *", "int", "struct vm_exit *", "uint32_t"); 245 246 SDT_PROBE_DEFINE5(vmm, vmx, exit, wrmsr, 247 "struct vmx *", "int", "struct vm_exit *", "uint32_t", "uint64_t"); 248 249 SDT_PROBE_DEFINE3(vmm, vmx, exit, halt, 250 "struct vmx *", "int", "struct vm_exit *"); 251 252 SDT_PROBE_DEFINE3(vmm, vmx, exit, mtrap, 253 "struct vmx *", "int", "struct vm_exit *"); 254 255 SDT_PROBE_DEFINE3(vmm, vmx, exit, pause, 256 "struct vmx *", "int", "struct vm_exit *"); 257 258 SDT_PROBE_DEFINE3(vmm, vmx, exit, intrwindow, 259 "struct vmx *", "int", "struct vm_exit *"); 260 261 SDT_PROBE_DEFINE4(vmm, vmx, exit, interrupt, 262 "struct vmx *", "int", "struct vm_exit *", "uint32_t"); 263 264 SDT_PROBE_DEFINE3(vmm, vmx, exit, nmiwindow, 265 "struct vmx *", "int", "struct vm_exit *"); 266 267 SDT_PROBE_DEFINE3(vmm, vmx, exit, inout, 268 "struct vmx *", "int", "struct vm_exit *"); 269 270 SDT_PROBE_DEFINE3(vmm, vmx, exit, cpuid, 271 "struct vmx *", "int", "struct vm_exit *"); 272 273 SDT_PROBE_DEFINE5(vmm, vmx, exit, exception, 274 "struct vmx *", "int", "struct vm_exit *", "uint32_t", "int"); 275 276 SDT_PROBE_DEFINE5(vmm, vmx, exit, nestedfault, 277 "struct vmx *", "int", "struct vm_exit *", "uint64_t", "uint64_t"); 278 279 SDT_PROBE_DEFINE4(vmm, vmx, exit, mmiofault, 280 "struct vmx *", "int", "struct vm_exit *", "uint64_t"); 281 282 SDT_PROBE_DEFINE3(vmm, vmx, exit, eoi, 283 "struct vmx *", "int", "struct vm_exit *"); 284 285 SDT_PROBE_DEFINE3(vmm, vmx, exit, apicaccess, 286 "struct vmx *", "int", "struct vm_exit *"); 287 288 SDT_PROBE_DEFINE4(vmm, vmx, exit, apicwrite, 289 "struct vmx *", "int", "struct vm_exit *", "struct vlapic *"); 290 291 SDT_PROBE_DEFINE3(vmm, vmx, exit, xsetbv, 292 "struct vmx *", "int", "struct vm_exit *"); 293 294 SDT_PROBE_DEFINE3(vmm, vmx, exit, monitor, 295 "struct vmx *", "int", "struct vm_exit *"); 296 297 SDT_PROBE_DEFINE3(vmm, vmx, exit, mwait, 298 "struct vmx *", "int", "struct vm_exit *"); 299 300 SDT_PROBE_DEFINE3(vmm, vmx, exit, vminsn, 301 "struct vmx *", "int", "struct vm_exit *"); 302 303 SDT_PROBE_DEFINE4(vmm, vmx, exit, unknown, 304 "struct vmx *", "int", "struct vm_exit *", "uint32_t"); 305 306 SDT_PROBE_DEFINE4(vmm, vmx, exit, return, 307 "struct vmx *", "int", "struct vm_exit *", "int"); 308 309 /* 310 * Use the last page below 4GB as the APIC access address. This address is 311 * occupied by the boot firmware so it is guaranteed that it will not conflict 312 * with a page in system memory. 313 */ 314 #define APIC_ACCESS_ADDRESS 0xFFFFF000 315 316 static int vmx_getdesc(void *vcpui, int reg, struct seg_desc *desc); 317 static int vmx_getreg(void *vcpui, int reg, uint64_t *retval); 318 static int vmxctx_setreg(struct vmxctx *vmxctx, int reg, uint64_t val); 319 static void vmx_inject_pir(struct vlapic *vlapic); 320 #ifdef BHYVE_SNAPSHOT 321 static int vmx_restore_tsc(void *vcpui, uint64_t now); 322 #endif 323 324 static inline bool 325 host_has_rdpid(void) 326 { 327 return ((cpu_stdext_feature2 & CPUID_STDEXT2_RDPID) != 0); 328 } 329 330 static inline bool 331 host_has_rdtscp(void) 332 { 333 return ((amd_feature & AMDID_RDTSCP) != 0); 334 } 335 336 #ifdef KTR 337 static const char * 338 exit_reason_to_str(int reason) 339 { 340 static char reasonbuf[32]; 341 342 switch (reason) { 343 case EXIT_REASON_EXCEPTION: 344 return "exception"; 345 case EXIT_REASON_EXT_INTR: 346 return "extint"; 347 case EXIT_REASON_TRIPLE_FAULT: 348 return "triplefault"; 349 case EXIT_REASON_INIT: 350 return "init"; 351 case EXIT_REASON_SIPI: 352 return "sipi"; 353 case EXIT_REASON_IO_SMI: 354 return "iosmi"; 355 case EXIT_REASON_SMI: 356 return "smi"; 357 case EXIT_REASON_INTR_WINDOW: 358 return "intrwindow"; 359 case EXIT_REASON_NMI_WINDOW: 360 return "nmiwindow"; 361 case EXIT_REASON_TASK_SWITCH: 362 return "taskswitch"; 363 case EXIT_REASON_CPUID: 364 return "cpuid"; 365 case EXIT_REASON_GETSEC: 366 return "getsec"; 367 case EXIT_REASON_HLT: 368 return "hlt"; 369 case EXIT_REASON_INVD: 370 return "invd"; 371 case EXIT_REASON_INVLPG: 372 return "invlpg"; 373 case EXIT_REASON_RDPMC: 374 return "rdpmc"; 375 case EXIT_REASON_RDTSC: 376 return "rdtsc"; 377 case EXIT_REASON_RSM: 378 return "rsm"; 379 case EXIT_REASON_VMCALL: 380 return "vmcall"; 381 case EXIT_REASON_VMCLEAR: 382 return "vmclear"; 383 case EXIT_REASON_VMLAUNCH: 384 return "vmlaunch"; 385 case EXIT_REASON_VMPTRLD: 386 return "vmptrld"; 387 case EXIT_REASON_VMPTRST: 388 return "vmptrst"; 389 case EXIT_REASON_VMREAD: 390 return "vmread"; 391 case EXIT_REASON_VMRESUME: 392 return "vmresume"; 393 case EXIT_REASON_VMWRITE: 394 return "vmwrite"; 395 case EXIT_REASON_VMXOFF: 396 return "vmxoff"; 397 case EXIT_REASON_VMXON: 398 return "vmxon"; 399 case EXIT_REASON_CR_ACCESS: 400 return "craccess"; 401 case EXIT_REASON_DR_ACCESS: 402 return "draccess"; 403 case EXIT_REASON_INOUT: 404 return "inout"; 405 case EXIT_REASON_RDMSR: 406 return "rdmsr"; 407 case EXIT_REASON_WRMSR: 408 return "wrmsr"; 409 case EXIT_REASON_INVAL_VMCS: 410 return "invalvmcs"; 411 case EXIT_REASON_INVAL_MSR: 412 return "invalmsr"; 413 case EXIT_REASON_MWAIT: 414 return "mwait"; 415 case EXIT_REASON_MTF: 416 return "mtf"; 417 case EXIT_REASON_MONITOR: 418 return "monitor"; 419 case EXIT_REASON_PAUSE: 420 return "pause"; 421 case EXIT_REASON_MCE_DURING_ENTRY: 422 return "mce-during-entry"; 423 case EXIT_REASON_TPR: 424 return "tpr"; 425 case EXIT_REASON_APIC_ACCESS: 426 return "apic-access"; 427 case EXIT_REASON_GDTR_IDTR: 428 return "gdtridtr"; 429 case EXIT_REASON_LDTR_TR: 430 return "ldtrtr"; 431 case EXIT_REASON_EPT_FAULT: 432 return "eptfault"; 433 case EXIT_REASON_EPT_MISCONFIG: 434 return "eptmisconfig"; 435 case EXIT_REASON_INVEPT: 436 return "invept"; 437 case EXIT_REASON_RDTSCP: 438 return "rdtscp"; 439 case EXIT_REASON_VMX_PREEMPT: 440 return "vmxpreempt"; 441 case EXIT_REASON_INVVPID: 442 return "invvpid"; 443 case EXIT_REASON_WBINVD: 444 return "wbinvd"; 445 case EXIT_REASON_XSETBV: 446 return "xsetbv"; 447 case EXIT_REASON_APIC_WRITE: 448 return "apic-write"; 449 default: 450 snprintf(reasonbuf, sizeof(reasonbuf), "%d", reason); 451 return (reasonbuf); 452 } 453 } 454 #endif /* KTR */ 455 456 static int 457 vmx_allow_x2apic_msrs(struct vmx *vmx) 458 { 459 int i, error; 460 461 error = 0; 462 463 /* 464 * Allow readonly access to the following x2APIC MSRs from the guest. 465 */ 466 error += guest_msr_ro(vmx, MSR_APIC_ID); 467 error += guest_msr_ro(vmx, MSR_APIC_VERSION); 468 error += guest_msr_ro(vmx, MSR_APIC_LDR); 469 error += guest_msr_ro(vmx, MSR_APIC_SVR); 470 471 for (i = 0; i < 8; i++) 472 error += guest_msr_ro(vmx, MSR_APIC_ISR0 + i); 473 474 for (i = 0; i < 8; i++) 475 error += guest_msr_ro(vmx, MSR_APIC_TMR0 + i); 476 477 for (i = 0; i < 8; i++) 478 error += guest_msr_ro(vmx, MSR_APIC_IRR0 + i); 479 480 error += guest_msr_ro(vmx, MSR_APIC_ESR); 481 error += guest_msr_ro(vmx, MSR_APIC_LVT_TIMER); 482 error += guest_msr_ro(vmx, MSR_APIC_LVT_THERMAL); 483 error += guest_msr_ro(vmx, MSR_APIC_LVT_PCINT); 484 error += guest_msr_ro(vmx, MSR_APIC_LVT_LINT0); 485 error += guest_msr_ro(vmx, MSR_APIC_LVT_LINT1); 486 error += guest_msr_ro(vmx, MSR_APIC_LVT_ERROR); 487 error += guest_msr_ro(vmx, MSR_APIC_ICR_TIMER); 488 error += guest_msr_ro(vmx, MSR_APIC_DCR_TIMER); 489 error += guest_msr_ro(vmx, MSR_APIC_ICR); 490 491 /* 492 * Allow TPR, EOI and SELF_IPI MSRs to be read and written by the guest. 493 * 494 * These registers get special treatment described in the section 495 * "Virtualizing MSR-Based APIC Accesses". 496 */ 497 error += guest_msr_rw(vmx, MSR_APIC_TPR); 498 error += guest_msr_rw(vmx, MSR_APIC_EOI); 499 error += guest_msr_rw(vmx, MSR_APIC_SELF_IPI); 500 501 return (error); 502 } 503 504 u_long 505 vmx_fix_cr0(u_long cr0) 506 { 507 508 return ((cr0 | cr0_ones_mask) & ~cr0_zeros_mask); 509 } 510 511 u_long 512 vmx_fix_cr4(u_long cr4) 513 { 514 515 return ((cr4 | cr4_ones_mask) & ~cr4_zeros_mask); 516 } 517 518 static void 519 vpid_free(int vpid) 520 { 521 if (vpid < 0 || vpid > 0xffff) 522 panic("vpid_free: invalid vpid %d", vpid); 523 524 /* 525 * VPIDs [0,vm_maxcpu] are special and are not allocated from 526 * the unit number allocator. 527 */ 528 529 if (vpid > vm_maxcpu) 530 free_unr(vpid_unr, vpid); 531 } 532 533 static uint16_t 534 vpid_alloc(int vcpuid) 535 { 536 int x; 537 538 /* 539 * If the "enable vpid" execution control is not enabled then the 540 * VPID is required to be 0 for all vcpus. 541 */ 542 if ((procbased_ctls2 & PROCBASED2_ENABLE_VPID) == 0) 543 return (0); 544 545 /* 546 * Try to allocate a unique VPID for each from the unit number 547 * allocator. 548 */ 549 x = alloc_unr(vpid_unr); 550 551 if (x == -1) { 552 atomic_add_int(&vpid_alloc_failed, 1); 553 554 /* 555 * If the unit number allocator does not have enough unique 556 * VPIDs then we need to allocate from the [1,vm_maxcpu] range. 557 * 558 * These VPIDs are not be unique across VMs but this does not 559 * affect correctness because the combined mappings are also 560 * tagged with the EP4TA which is unique for each VM. 561 * 562 * It is still sub-optimal because the invvpid will invalidate 563 * combined mappings for a particular VPID across all EP4TAs. 564 */ 565 return (vcpuid + 1); 566 } 567 568 return (x); 569 } 570 571 static void 572 vpid_init(void) 573 { 574 /* 575 * VPID 0 is required when the "enable VPID" execution control is 576 * disabled. 577 * 578 * VPIDs [1,vm_maxcpu] are used as the "overflow namespace" when the 579 * unit number allocator does not have sufficient unique VPIDs to 580 * satisfy the allocation. 581 * 582 * The remaining VPIDs are managed by the unit number allocator. 583 */ 584 vpid_unr = new_unrhdr(vm_maxcpu + 1, 0xffff, NULL); 585 } 586 587 static void 588 vmx_disable(void *arg __unused) 589 { 590 struct invvpid_desc invvpid_desc = { 0 }; 591 struct invept_desc invept_desc = { 0 }; 592 593 if (vmxon_enabled[curcpu]) { 594 /* 595 * See sections 25.3.3.3 and 25.3.3.4 in Intel Vol 3b. 596 * 597 * VMXON or VMXOFF are not required to invalidate any TLB 598 * caching structures. This prevents potential retention of 599 * cached information in the TLB between distinct VMX episodes. 600 */ 601 invvpid(INVVPID_TYPE_ALL_CONTEXTS, invvpid_desc); 602 invept(INVEPT_TYPE_ALL_CONTEXTS, invept_desc); 603 vmxoff(); 604 } 605 load_cr4(rcr4() & ~CR4_VMXE); 606 } 607 608 static int 609 vmx_modcleanup(void) 610 { 611 612 if (pirvec >= 0) 613 lapic_ipi_free(pirvec); 614 615 if (vpid_unr != NULL) { 616 delete_unrhdr(vpid_unr); 617 vpid_unr = NULL; 618 } 619 620 if (nmi_flush_l1d_sw == 1) 621 nmi_flush_l1d_sw = 0; 622 623 smp_rendezvous(NULL, vmx_disable, NULL, NULL); 624 625 if (vmxon_region != NULL) 626 kmem_free(vmxon_region, (mp_maxid + 1) * PAGE_SIZE); 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 * PAGE_SIZE] = vmx_revision(); 648 error = vmxon(&vmxon_region[curcpu * PAGE_SIZE]); 649 if (error == 0) 650 vmxon_enabled[curcpu] = 1; 651 } 652 653 static void 654 vmx_modsuspend(void) 655 { 656 657 if (vmxon_enabled[curcpu]) 658 vmx_disable(NULL); 659 } 660 661 static void 662 vmx_modresume(void) 663 { 664 665 if (vmxon_enabled[curcpu]) 666 vmx_enable(NULL); 667 } 668 669 static int 670 vmx_modinit(int ipinum) 671 { 672 int error; 673 uint64_t basic, fixed0, fixed1, feature_control; 674 uint32_t tmp, procbased2_vid_bits; 675 676 /* CPUID.1:ECX[bit 5] must be 1 for processor to support VMX */ 677 if (!(cpu_feature2 & CPUID2_VMX)) { 678 printf("vmx_modinit: processor does not support VMX " 679 "operation\n"); 680 return (ENXIO); 681 } 682 683 /* 684 * Verify that MSR_IA32_FEATURE_CONTROL lock and VMXON enable bits 685 * are set (bits 0 and 2 respectively). 686 */ 687 feature_control = rdmsr(MSR_IA32_FEATURE_CONTROL); 688 if ((feature_control & IA32_FEATURE_CONTROL_LOCK) == 1 && 689 (feature_control & IA32_FEATURE_CONTROL_VMX_EN) == 0) { 690 printf("vmx_modinit: VMX operation disabled by BIOS\n"); 691 return (ENXIO); 692 } 693 694 /* 695 * Verify capabilities MSR_VMX_BASIC: 696 * - bit 54 indicates support for INS/OUTS decoding 697 */ 698 basic = rdmsr(MSR_VMX_BASIC); 699 if ((basic & (1UL << 54)) == 0) { 700 printf("vmx_modinit: processor does not support desired basic " 701 "capabilities\n"); 702 return (EINVAL); 703 } 704 705 /* Check support for primary processor-based VM-execution controls */ 706 error = vmx_set_ctlreg(MSR_VMX_PROCBASED_CTLS, 707 MSR_VMX_TRUE_PROCBASED_CTLS, 708 PROCBASED_CTLS_ONE_SETTING, 709 PROCBASED_CTLS_ZERO_SETTING, &procbased_ctls); 710 if (error) { 711 printf("vmx_modinit: processor does not support desired " 712 "primary processor-based controls\n"); 713 return (error); 714 } 715 716 /* Clear the processor-based ctl bits that are set on demand */ 717 procbased_ctls &= ~PROCBASED_CTLS_WINDOW_SETTING; 718 719 /* Check support for secondary processor-based VM-execution controls */ 720 error = vmx_set_ctlreg(MSR_VMX_PROCBASED_CTLS2, 721 MSR_VMX_PROCBASED_CTLS2, 722 PROCBASED_CTLS2_ONE_SETTING, 723 PROCBASED_CTLS2_ZERO_SETTING, &procbased_ctls2); 724 if (error) { 725 printf("vmx_modinit: processor does not support desired " 726 "secondary processor-based controls\n"); 727 return (error); 728 } 729 730 /* Check support for VPID */ 731 error = vmx_set_ctlreg(MSR_VMX_PROCBASED_CTLS2, MSR_VMX_PROCBASED_CTLS2, 732 PROCBASED2_ENABLE_VPID, 0, &tmp); 733 if (error == 0) 734 procbased_ctls2 |= PROCBASED2_ENABLE_VPID; 735 736 /* Check support for pin-based VM-execution controls */ 737 error = vmx_set_ctlreg(MSR_VMX_PINBASED_CTLS, 738 MSR_VMX_TRUE_PINBASED_CTLS, 739 PINBASED_CTLS_ONE_SETTING, 740 PINBASED_CTLS_ZERO_SETTING, &pinbased_ctls); 741 if (error) { 742 printf("vmx_modinit: processor does not support desired " 743 "pin-based controls\n"); 744 return (error); 745 } 746 747 /* Check support for VM-exit controls */ 748 error = vmx_set_ctlreg(MSR_VMX_EXIT_CTLS, MSR_VMX_TRUE_EXIT_CTLS, 749 VM_EXIT_CTLS_ONE_SETTING, 750 VM_EXIT_CTLS_ZERO_SETTING, 751 &exit_ctls); 752 if (error) { 753 printf("vmx_modinit: processor does not support desired " 754 "exit controls\n"); 755 return (error); 756 } 757 758 /* Check support for VM-entry controls */ 759 error = vmx_set_ctlreg(MSR_VMX_ENTRY_CTLS, MSR_VMX_TRUE_ENTRY_CTLS, 760 VM_ENTRY_CTLS_ONE_SETTING, VM_ENTRY_CTLS_ZERO_SETTING, 761 &entry_ctls); 762 if (error) { 763 printf("vmx_modinit: processor does not support desired " 764 "entry controls\n"); 765 return (error); 766 } 767 768 /* 769 * Check support for optional features by testing them 770 * as individual bits 771 */ 772 cap_halt_exit = (vmx_set_ctlreg(MSR_VMX_PROCBASED_CTLS, 773 MSR_VMX_TRUE_PROCBASED_CTLS, 774 PROCBASED_HLT_EXITING, 0, 775 &tmp) == 0); 776 777 cap_monitor_trap = (vmx_set_ctlreg(MSR_VMX_PROCBASED_CTLS, 778 MSR_VMX_PROCBASED_CTLS, 779 PROCBASED_MTF, 0, 780 &tmp) == 0); 781 782 cap_pause_exit = (vmx_set_ctlreg(MSR_VMX_PROCBASED_CTLS, 783 MSR_VMX_TRUE_PROCBASED_CTLS, 784 PROCBASED_PAUSE_EXITING, 0, 785 &tmp) == 0); 786 787 cap_wbinvd_exit = (vmx_set_ctlreg(MSR_VMX_PROCBASED_CTLS2, 788 MSR_VMX_PROCBASED_CTLS2, 789 PROCBASED2_WBINVD_EXITING, 790 0, 791 &tmp) == 0); 792 793 /* 794 * Check support for RDPID and/or RDTSCP. 795 * 796 * Support a pass-through-based implementation of these via the 797 * "enable RDTSCP" VM-execution control and the "RDTSC exiting" 798 * VM-execution control. 799 * 800 * The "enable RDTSCP" VM-execution control applies to both RDPID 801 * and RDTSCP (see SDM volume 3, section 25.3, "Changes to 802 * Instruction Behavior in VMX Non-root operation"); this is why 803 * only this VM-execution control needs to be enabled in order to 804 * enable passing through whichever of RDPID and/or RDTSCP are 805 * supported by the host. 806 * 807 * The "RDTSC exiting" VM-execution control applies to both RDTSC 808 * and RDTSCP (again, per SDM volume 3, section 25.3), and is 809 * already set up for RDTSC and RDTSCP pass-through by the current 810 * implementation of RDTSC. 811 * 812 * Although RDPID and RDTSCP are optional capabilities, since there 813 * does not currently seem to be a use case for enabling/disabling 814 * these via libvmmapi, choose not to support this and, instead, 815 * just statically always enable or always disable this support 816 * across all vCPUs on all VMs. (Note that there may be some 817 * complications to providing this functionality, e.g., the MSR 818 * bitmap is currently per-VM rather than per-vCPU while the 819 * capability API wants to be able to control capabilities on a 820 * per-vCPU basis). 821 */ 822 error = vmx_set_ctlreg(MSR_VMX_PROCBASED_CTLS2, 823 MSR_VMX_PROCBASED_CTLS2, 824 PROCBASED2_ENABLE_RDTSCP, 0, &tmp); 825 cap_rdpid = error == 0 && host_has_rdpid(); 826 cap_rdtscp = error == 0 && host_has_rdtscp(); 827 if (cap_rdpid || cap_rdtscp) { 828 procbased_ctls2 |= PROCBASED2_ENABLE_RDTSCP; 829 vmx_have_msr_tsc_aux = true; 830 } 831 832 cap_unrestricted_guest = (vmx_set_ctlreg(MSR_VMX_PROCBASED_CTLS2, 833 MSR_VMX_PROCBASED_CTLS2, 834 PROCBASED2_UNRESTRICTED_GUEST, 0, 835 &tmp) == 0); 836 837 cap_invpcid = (vmx_set_ctlreg(MSR_VMX_PROCBASED_CTLS2, 838 MSR_VMX_PROCBASED_CTLS2, PROCBASED2_ENABLE_INVPCID, 0, 839 &tmp) == 0); 840 841 /* 842 * Check support for TPR shadow. 843 */ 844 error = vmx_set_ctlreg(MSR_VMX_PROCBASED_CTLS, 845 MSR_VMX_TRUE_PROCBASED_CTLS, PROCBASED_USE_TPR_SHADOW, 0, 846 &tmp); 847 if (error == 0) { 848 tpr_shadowing = 1; 849 #ifndef BURN_BRIDGES 850 TUNABLE_INT_FETCH("hw.vmm.vmx.use_tpr_shadowing", 851 &tpr_shadowing); 852 #endif 853 TUNABLE_INT_FETCH("hw.vmm.vmx.cap.tpr_shadowing", 854 &tpr_shadowing); 855 } 856 857 if (tpr_shadowing) { 858 procbased_ctls |= PROCBASED_USE_TPR_SHADOW; 859 procbased_ctls &= ~PROCBASED_CR8_LOAD_EXITING; 860 procbased_ctls &= ~PROCBASED_CR8_STORE_EXITING; 861 } 862 863 /* 864 * Check support for virtual interrupt delivery. 865 */ 866 procbased2_vid_bits = (PROCBASED2_VIRTUALIZE_APIC_ACCESSES | 867 PROCBASED2_VIRTUALIZE_X2APIC_MODE | 868 PROCBASED2_APIC_REGISTER_VIRTUALIZATION | 869 PROCBASED2_VIRTUAL_INTERRUPT_DELIVERY); 870 871 error = vmx_set_ctlreg(MSR_VMX_PROCBASED_CTLS2, MSR_VMX_PROCBASED_CTLS2, 872 procbased2_vid_bits, 0, &tmp); 873 if (error == 0 && tpr_shadowing) { 874 virtual_interrupt_delivery = 1; 875 #ifndef BURN_BRIDGES 876 TUNABLE_INT_FETCH("hw.vmm.vmx.use_apic_vid", 877 &virtual_interrupt_delivery); 878 #endif 879 TUNABLE_INT_FETCH("hw.vmm.vmx.cap.virtual_interrupt_delivery", 880 &virtual_interrupt_delivery); 881 } 882 883 if (virtual_interrupt_delivery) { 884 procbased_ctls |= PROCBASED_USE_TPR_SHADOW; 885 procbased_ctls2 |= procbased2_vid_bits; 886 procbased_ctls2 &= ~PROCBASED2_VIRTUALIZE_X2APIC_MODE; 887 888 /* 889 * Check for Posted Interrupts only if Virtual Interrupt 890 * Delivery is enabled. 891 */ 892 error = vmx_set_ctlreg(MSR_VMX_PINBASED_CTLS, 893 MSR_VMX_TRUE_PINBASED_CTLS, PINBASED_POSTED_INTERRUPT, 0, 894 &tmp); 895 if (error == 0) { 896 pirvec = lapic_ipi_alloc(pti ? &IDTVEC(justreturn1_pti) : 897 &IDTVEC(justreturn)); 898 if (pirvec < 0) { 899 if (bootverbose) { 900 printf("vmx_modinit: unable to " 901 "allocate posted interrupt " 902 "vector\n"); 903 } 904 } else { 905 posted_interrupts = 1; 906 #ifndef BURN_BRIDGES 907 TUNABLE_INT_FETCH("hw.vmm.vmx.use_apic_pir", 908 &posted_interrupts); 909 #endif 910 TUNABLE_INT_FETCH("hw.vmm.vmx.cap.posted_interrupts", 911 &posted_interrupts); 912 } 913 } 914 } 915 916 if (posted_interrupts) 917 pinbased_ctls |= PINBASED_POSTED_INTERRUPT; 918 919 /* Initialize EPT */ 920 error = ept_init(ipinum); 921 if (error) { 922 printf("vmx_modinit: ept initialization failed (%d)\n", error); 923 return (error); 924 } 925 926 guest_l1d_flush = (cpu_ia32_arch_caps & 927 IA32_ARCH_CAP_SKIP_L1DFL_VMENTRY) == 0; 928 #ifndef BURN_BRIDGES 929 TUNABLE_INT_FETCH("hw.vmm.l1d_flush", &guest_l1d_flush); 930 #endif 931 TUNABLE_INT_FETCH("hw.vmm.vmx.l1d_flush", &guest_l1d_flush); 932 933 /* 934 * L1D cache flush is enabled. Use IA32_FLUSH_CMD MSR when 935 * available. Otherwise fall back to the software flush 936 * method which loads enough data from the kernel text to 937 * flush existing L1D content, both on VMX entry and on NMI 938 * return. 939 */ 940 if (guest_l1d_flush) { 941 if ((cpu_stdext_feature3 & CPUID_STDEXT3_L1D_FLUSH) == 0) { 942 guest_l1d_flush_sw = 1; 943 #ifndef BURN_BRIDGES 944 TUNABLE_INT_FETCH("hw.vmm.l1d_flush_sw", 945 &guest_l1d_flush_sw); 946 #endif 947 TUNABLE_INT_FETCH("hw.vmm.vmx.l1d_flush_sw", 948 &guest_l1d_flush_sw); 949 } 950 if (guest_l1d_flush_sw) { 951 if (nmi_flush_l1d_sw <= 1) 952 nmi_flush_l1d_sw = 1; 953 } else { 954 msr_load_list[0].index = MSR_IA32_FLUSH_CMD; 955 msr_load_list[0].val = IA32_FLUSH_CMD_L1D; 956 } 957 } 958 959 /* 960 * Stash the cr0 and cr4 bits that must be fixed to 0 or 1 961 */ 962 fixed0 = rdmsr(MSR_VMX_CR0_FIXED0); 963 fixed1 = rdmsr(MSR_VMX_CR0_FIXED1); 964 cr0_ones_mask = fixed0 & fixed1; 965 cr0_zeros_mask = ~fixed0 & ~fixed1; 966 967 /* 968 * CR0_PE and CR0_PG can be set to zero in VMX non-root operation 969 * if unrestricted guest execution is allowed. 970 */ 971 if (cap_unrestricted_guest) 972 cr0_ones_mask &= ~(CR0_PG | CR0_PE); 973 974 /* 975 * Do not allow the guest to set CR0_NW or CR0_CD. 976 */ 977 cr0_zeros_mask |= (CR0_NW | CR0_CD); 978 979 fixed0 = rdmsr(MSR_VMX_CR4_FIXED0); 980 fixed1 = rdmsr(MSR_VMX_CR4_FIXED1); 981 cr4_ones_mask = fixed0 & fixed1; 982 cr4_zeros_mask = ~fixed0 & ~fixed1; 983 984 vpid_init(); 985 986 vmx_msr_init(); 987 988 /* enable VMX operation */ 989 vmxon_region = kmem_malloc((mp_maxid + 1) * PAGE_SIZE, 990 M_WAITOK | M_ZERO); 991 smp_rendezvous(NULL, vmx_enable, NULL, NULL); 992 993 vmx_initialized = 1; 994 995 return (0); 996 } 997 998 static void 999 vmx_trigger_hostintr(int vector) 1000 { 1001 uintptr_t func; 1002 struct gate_descriptor *gd; 1003 1004 gd = &idt[vector]; 1005 1006 KASSERT(vector >= 32 && vector <= 255, ("vmx_trigger_hostintr: " 1007 "invalid vector %d", vector)); 1008 KASSERT(gd->gd_p == 1, ("gate descriptor for vector %d not present", 1009 vector)); 1010 KASSERT(gd->gd_type == SDT_SYSIGT, ("gate descriptor for vector %d " 1011 "has invalid type %d", vector, gd->gd_type)); 1012 KASSERT(gd->gd_dpl == SEL_KPL, ("gate descriptor for vector %d " 1013 "has invalid dpl %d", vector, gd->gd_dpl)); 1014 KASSERT(gd->gd_selector == GSEL(GCODE_SEL, SEL_KPL), ("gate descriptor " 1015 "for vector %d has invalid selector %d", vector, gd->gd_selector)); 1016 KASSERT(gd->gd_ist == 0, ("gate descriptor for vector %d has invalid " 1017 "IST %d", vector, gd->gd_ist)); 1018 1019 func = ((long)gd->gd_hioffset << 16 | gd->gd_looffset); 1020 vmx_call_isr(func); 1021 } 1022 1023 static int 1024 vmx_setup_cr_shadow(int which, struct vmcs *vmcs, uint32_t initial) 1025 { 1026 int error, mask_ident, shadow_ident; 1027 uint64_t mask_value; 1028 1029 if (which != 0 && which != 4) 1030 panic("vmx_setup_cr_shadow: unknown cr%d", which); 1031 1032 if (which == 0) { 1033 mask_ident = VMCS_CR0_MASK; 1034 mask_value = cr0_ones_mask | cr0_zeros_mask; 1035 shadow_ident = VMCS_CR0_SHADOW; 1036 } else { 1037 mask_ident = VMCS_CR4_MASK; 1038 mask_value = cr4_ones_mask | cr4_zeros_mask; 1039 shadow_ident = VMCS_CR4_SHADOW; 1040 } 1041 1042 error = vmcs_setreg(vmcs, 0, VMCS_IDENT(mask_ident), mask_value); 1043 if (error) 1044 return (error); 1045 1046 error = vmcs_setreg(vmcs, 0, VMCS_IDENT(shadow_ident), initial); 1047 if (error) 1048 return (error); 1049 1050 return (0); 1051 } 1052 #define vmx_setup_cr0_shadow(vmcs,init) vmx_setup_cr_shadow(0, (vmcs), (init)) 1053 #define vmx_setup_cr4_shadow(vmcs,init) vmx_setup_cr_shadow(4, (vmcs), (init)) 1054 1055 static void * 1056 vmx_init(struct vm *vm, pmap_t pmap) 1057 { 1058 int error __diagused; 1059 struct vmx *vmx; 1060 1061 vmx = malloc(sizeof(struct vmx), M_VMX, M_WAITOK | M_ZERO); 1062 vmx->vm = vm; 1063 1064 vmx->eptp = eptp(vtophys((vm_offset_t)pmap->pm_pmltop)); 1065 1066 /* 1067 * Clean up EPTP-tagged guest physical and combined mappings 1068 * 1069 * VMX transitions are not required to invalidate any guest physical 1070 * mappings. So, it may be possible for stale guest physical mappings 1071 * to be present in the processor TLBs. 1072 * 1073 * Combined mappings for this EP4TA are also invalidated for all VPIDs. 1074 */ 1075 ept_invalidate_mappings(vmx->eptp); 1076 1077 vmx->msr_bitmap = malloc_aligned(PAGE_SIZE, PAGE_SIZE, M_VMX, 1078 M_WAITOK | M_ZERO); 1079 msr_bitmap_initialize(vmx->msr_bitmap); 1080 1081 /* 1082 * It is safe to allow direct access to MSR_GSBASE and MSR_FSBASE. 1083 * The guest FSBASE and GSBASE are saved and restored during 1084 * vm-exit and vm-entry respectively. The host FSBASE and GSBASE are 1085 * always restored from the vmcs host state area on vm-exit. 1086 * 1087 * The SYSENTER_CS/ESP/EIP MSRs are identical to FS/GSBASE in 1088 * how they are saved/restored so can be directly accessed by the 1089 * guest. 1090 * 1091 * MSR_EFER is saved and restored in the guest VMCS area on a 1092 * VM exit and entry respectively. It is also restored from the 1093 * host VMCS area on a VM exit. 1094 * 1095 * The TSC MSR is exposed read-only. Writes are disallowed as 1096 * that will impact the host TSC. If the guest does a write 1097 * the "use TSC offsetting" execution control is enabled and the 1098 * difference between the host TSC and the guest TSC is written 1099 * into the TSC offset in the VMCS. 1100 * 1101 * Guest TSC_AUX support is enabled if any of guest RDPID and/or 1102 * guest RDTSCP support are enabled (since, as per Table 2-2 in SDM 1103 * volume 4, TSC_AUX is supported if any of RDPID and/or RDTSCP are 1104 * supported). If guest TSC_AUX support is enabled, TSC_AUX is 1105 * exposed read-only so that the VMM can do one fewer MSR read per 1106 * exit than if this register were exposed read-write; the guest 1107 * restore value can be updated during guest writes (expected to be 1108 * rare) instead of during all exits (common). 1109 */ 1110 if (guest_msr_rw(vmx, MSR_GSBASE) || 1111 guest_msr_rw(vmx, MSR_FSBASE) || 1112 guest_msr_rw(vmx, MSR_SYSENTER_CS_MSR) || 1113 guest_msr_rw(vmx, MSR_SYSENTER_ESP_MSR) || 1114 guest_msr_rw(vmx, MSR_SYSENTER_EIP_MSR) || 1115 guest_msr_rw(vmx, MSR_EFER) || 1116 guest_msr_ro(vmx, MSR_TSC) || 1117 ((cap_rdpid || cap_rdtscp) && guest_msr_ro(vmx, MSR_TSC_AUX))) 1118 panic("vmx_init: error setting guest msr access"); 1119 1120 if (virtual_interrupt_delivery) { 1121 error = vm_map_mmio(vm, DEFAULT_APIC_BASE, PAGE_SIZE, 1122 APIC_ACCESS_ADDRESS); 1123 /* XXX this should really return an error to the caller */ 1124 KASSERT(error == 0, ("vm_map_mmio(apicbase) error %d", error)); 1125 } 1126 1127 vmx->pmap = pmap; 1128 return (vmx); 1129 } 1130 1131 static void * 1132 vmx_vcpu_init(void *vmi, struct vcpu *vcpu1, int vcpuid) 1133 { 1134 struct vmx *vmx = vmi; 1135 struct vmcs *vmcs; 1136 struct vmx_vcpu *vcpu; 1137 uint32_t exc_bitmap; 1138 uint16_t vpid; 1139 int error; 1140 1141 vpid = vpid_alloc(vcpuid); 1142 1143 vcpu = malloc(sizeof(*vcpu), M_VMX, M_WAITOK | M_ZERO); 1144 vcpu->vmx = vmx; 1145 vcpu->vcpu = vcpu1; 1146 vcpu->vcpuid = vcpuid; 1147 vcpu->vmcs = malloc_aligned(sizeof(*vmcs), PAGE_SIZE, M_VMX, 1148 M_WAITOK | M_ZERO); 1149 vcpu->apic_page = malloc_aligned(PAGE_SIZE, PAGE_SIZE, M_VMX, 1150 M_WAITOK | M_ZERO); 1151 vcpu->pir_desc = malloc_aligned(sizeof(*vcpu->pir_desc), 64, M_VMX, 1152 M_WAITOK | M_ZERO); 1153 1154 vmcs = vcpu->vmcs; 1155 vmcs->identifier = vmx_revision(); 1156 error = vmclear(vmcs); 1157 if (error != 0) { 1158 panic("vmx_init: vmclear error %d on vcpu %d\n", 1159 error, vcpuid); 1160 } 1161 1162 vmx_msr_guest_init(vmx, vcpu); 1163 1164 error = vmcs_init(vmcs); 1165 KASSERT(error == 0, ("vmcs_init error %d", error)); 1166 1167 VMPTRLD(vmcs); 1168 error = 0; 1169 error += vmwrite(VMCS_HOST_RSP, (u_long)&vcpu->ctx); 1170 error += vmwrite(VMCS_EPTP, vmx->eptp); 1171 error += vmwrite(VMCS_PIN_BASED_CTLS, pinbased_ctls); 1172 error += vmwrite(VMCS_PRI_PROC_BASED_CTLS, procbased_ctls); 1173 if (vcpu_trap_wbinvd(vcpu->vcpu)) { 1174 KASSERT(cap_wbinvd_exit, ("WBINVD trap not available")); 1175 procbased_ctls2 |= PROCBASED2_WBINVD_EXITING; 1176 } 1177 error += vmwrite(VMCS_SEC_PROC_BASED_CTLS, procbased_ctls2); 1178 error += vmwrite(VMCS_EXIT_CTLS, exit_ctls); 1179 error += vmwrite(VMCS_ENTRY_CTLS, entry_ctls); 1180 error += vmwrite(VMCS_MSR_BITMAP, vtophys(vmx->msr_bitmap)); 1181 error += vmwrite(VMCS_VPID, vpid); 1182 1183 if (guest_l1d_flush && !guest_l1d_flush_sw) { 1184 vmcs_write(VMCS_ENTRY_MSR_LOAD, pmap_kextract( 1185 (vm_offset_t)&msr_load_list[0])); 1186 vmcs_write(VMCS_ENTRY_MSR_LOAD_COUNT, 1187 nitems(msr_load_list)); 1188 vmcs_write(VMCS_EXIT_MSR_STORE, 0); 1189 vmcs_write(VMCS_EXIT_MSR_STORE_COUNT, 0); 1190 } 1191 1192 /* exception bitmap */ 1193 if (vcpu_trace_exceptions(vcpu->vcpu)) 1194 exc_bitmap = 0xffffffff; 1195 else 1196 exc_bitmap = 1 << IDT_MC; 1197 error += vmwrite(VMCS_EXCEPTION_BITMAP, exc_bitmap); 1198 1199 vcpu->ctx.guest_dr6 = DBREG_DR6_RESERVED1; 1200 error += vmwrite(VMCS_GUEST_DR7, DBREG_DR7_RESERVED1); 1201 1202 if (tpr_shadowing) { 1203 error += vmwrite(VMCS_VIRTUAL_APIC, vtophys(vcpu->apic_page)); 1204 } 1205 1206 if (virtual_interrupt_delivery) { 1207 error += vmwrite(VMCS_APIC_ACCESS, APIC_ACCESS_ADDRESS); 1208 error += vmwrite(VMCS_EOI_EXIT0, 0); 1209 error += vmwrite(VMCS_EOI_EXIT1, 0); 1210 error += vmwrite(VMCS_EOI_EXIT2, 0); 1211 error += vmwrite(VMCS_EOI_EXIT3, 0); 1212 } 1213 if (posted_interrupts) { 1214 error += vmwrite(VMCS_PIR_VECTOR, pirvec); 1215 error += vmwrite(VMCS_PIR_DESC, vtophys(vcpu->pir_desc)); 1216 } 1217 VMCLEAR(vmcs); 1218 KASSERT(error == 0, ("vmx_init: error customizing the vmcs")); 1219 1220 vcpu->cap.set = 0; 1221 vcpu->cap.set |= cap_rdpid != 0 ? 1 << VM_CAP_RDPID : 0; 1222 vcpu->cap.set |= cap_rdtscp != 0 ? 1 << VM_CAP_RDTSCP : 0; 1223 vcpu->cap.proc_ctls = procbased_ctls; 1224 vcpu->cap.proc_ctls2 = procbased_ctls2; 1225 vcpu->cap.exc_bitmap = exc_bitmap; 1226 1227 vcpu->state.nextrip = ~0; 1228 vcpu->state.lastcpu = NOCPU; 1229 vcpu->state.vpid = vpid; 1230 1231 /* 1232 * Set up the CR0/4 shadows, and init the read shadow 1233 * to the power-on register value from the Intel Sys Arch. 1234 * CR0 - 0x60000010 1235 * CR4 - 0 1236 */ 1237 error = vmx_setup_cr0_shadow(vmcs, 0x60000010); 1238 if (error != 0) 1239 panic("vmx_setup_cr0_shadow %d", error); 1240 1241 error = vmx_setup_cr4_shadow(vmcs, 0); 1242 if (error != 0) 1243 panic("vmx_setup_cr4_shadow %d", error); 1244 1245 vcpu->ctx.pmap = vmx->pmap; 1246 1247 return (vcpu); 1248 } 1249 1250 static int 1251 vmx_handle_cpuid(struct vmx_vcpu *vcpu, struct vmxctx *vmxctx) 1252 { 1253 int handled; 1254 1255 handled = x86_emulate_cpuid(vcpu->vcpu, (uint64_t *)&vmxctx->guest_rax, 1256 (uint64_t *)&vmxctx->guest_rbx, (uint64_t *)&vmxctx->guest_rcx, 1257 (uint64_t *)&vmxctx->guest_rdx); 1258 return (handled); 1259 } 1260 1261 static __inline void 1262 vmx_run_trace(struct vmx_vcpu *vcpu) 1263 { 1264 VMX_CTR1(vcpu, "Resume execution at %#lx", vmcs_guest_rip()); 1265 } 1266 1267 static __inline void 1268 vmx_exit_trace(struct vmx_vcpu *vcpu, uint64_t rip, uint32_t exit_reason, 1269 int handled) 1270 { 1271 VMX_CTR3(vcpu, "%s %s vmexit at 0x%0lx", 1272 handled ? "handled" : "unhandled", 1273 exit_reason_to_str(exit_reason), rip); 1274 } 1275 1276 static __inline void 1277 vmx_astpending_trace(struct vmx_vcpu *vcpu, uint64_t rip) 1278 { 1279 VMX_CTR1(vcpu, "astpending vmexit at 0x%0lx", rip); 1280 } 1281 1282 static VMM_STAT_INTEL(VCPU_INVVPID_SAVED, "Number of vpid invalidations saved"); 1283 static VMM_STAT_INTEL(VCPU_INVVPID_DONE, "Number of vpid invalidations done"); 1284 1285 /* 1286 * Invalidate guest mappings identified by its vpid from the TLB. 1287 */ 1288 static __inline void 1289 vmx_invvpid(struct vmx *vmx, struct vmx_vcpu *vcpu, pmap_t pmap, int running) 1290 { 1291 struct vmxstate *vmxstate; 1292 struct invvpid_desc invvpid_desc; 1293 1294 vmxstate = &vcpu->state; 1295 if (vmxstate->vpid == 0) 1296 return; 1297 1298 if (!running) { 1299 /* 1300 * Set the 'lastcpu' to an invalid host cpu. 1301 * 1302 * This will invalidate TLB entries tagged with the vcpu's 1303 * vpid the next time it runs via vmx_set_pcpu_defaults(). 1304 */ 1305 vmxstate->lastcpu = NOCPU; 1306 return; 1307 } 1308 1309 KASSERT(curthread->td_critnest > 0, ("%s: vcpu %d running outside " 1310 "critical section", __func__, vcpu->vcpuid)); 1311 1312 /* 1313 * Invalidate all mappings tagged with 'vpid' 1314 * 1315 * We do this because this vcpu was executing on a different host 1316 * cpu when it last ran. We do not track whether it invalidated 1317 * mappings associated with its 'vpid' during that run. So we must 1318 * assume that the mappings associated with 'vpid' on 'curcpu' are 1319 * stale and invalidate them. 1320 * 1321 * Note that we incur this penalty only when the scheduler chooses to 1322 * move the thread associated with this vcpu between host cpus. 1323 * 1324 * Note also that this will invalidate mappings tagged with 'vpid' 1325 * for "all" EP4TAs. 1326 */ 1327 if (atomic_load_long(&pmap->pm_eptgen) == vmx->eptgen[curcpu]) { 1328 invvpid_desc._res1 = 0; 1329 invvpid_desc._res2 = 0; 1330 invvpid_desc.vpid = vmxstate->vpid; 1331 invvpid_desc.linear_addr = 0; 1332 invvpid(INVVPID_TYPE_SINGLE_CONTEXT, invvpid_desc); 1333 vmm_stat_incr(vcpu->vcpu, VCPU_INVVPID_DONE, 1); 1334 } else { 1335 /* 1336 * The invvpid can be skipped if an invept is going to 1337 * be performed before entering the guest. The invept 1338 * will invalidate combined mappings tagged with 1339 * 'vmx->eptp' for all vpids. 1340 */ 1341 vmm_stat_incr(vcpu->vcpu, VCPU_INVVPID_SAVED, 1); 1342 } 1343 } 1344 1345 static void 1346 vmx_set_pcpu_defaults(struct vmx *vmx, struct vmx_vcpu *vcpu, pmap_t pmap) 1347 { 1348 struct vmxstate *vmxstate; 1349 1350 vmxstate = &vcpu->state; 1351 if (vmxstate->lastcpu == curcpu) 1352 return; 1353 1354 vmxstate->lastcpu = curcpu; 1355 1356 vmm_stat_incr(vcpu->vcpu, VCPU_MIGRATIONS, 1); 1357 1358 vmcs_write(VMCS_HOST_TR_BASE, vmm_get_host_trbase()); 1359 vmcs_write(VMCS_HOST_GDTR_BASE, vmm_get_host_gdtrbase()); 1360 vmcs_write(VMCS_HOST_GS_BASE, vmm_get_host_gsbase()); 1361 vmx_invvpid(vmx, vcpu, pmap, 1); 1362 } 1363 1364 /* 1365 * We depend on 'procbased_ctls' to have the Interrupt Window Exiting bit set. 1366 */ 1367 CTASSERT((PROCBASED_CTLS_ONE_SETTING & PROCBASED_INT_WINDOW_EXITING) != 0); 1368 1369 static void __inline 1370 vmx_set_int_window_exiting(struct vmx_vcpu *vcpu) 1371 { 1372 1373 if ((vcpu->cap.proc_ctls & PROCBASED_INT_WINDOW_EXITING) == 0) { 1374 vcpu->cap.proc_ctls |= PROCBASED_INT_WINDOW_EXITING; 1375 vmcs_write(VMCS_PRI_PROC_BASED_CTLS, vcpu->cap.proc_ctls); 1376 VMX_CTR0(vcpu, "Enabling interrupt window exiting"); 1377 } 1378 } 1379 1380 static void __inline 1381 vmx_clear_int_window_exiting(struct vmx_vcpu *vcpu) 1382 { 1383 1384 KASSERT((vcpu->cap.proc_ctls & PROCBASED_INT_WINDOW_EXITING) != 0, 1385 ("intr_window_exiting not set: %#x", vcpu->cap.proc_ctls)); 1386 vcpu->cap.proc_ctls &= ~PROCBASED_INT_WINDOW_EXITING; 1387 vmcs_write(VMCS_PRI_PROC_BASED_CTLS, vcpu->cap.proc_ctls); 1388 VMX_CTR0(vcpu, "Disabling interrupt window exiting"); 1389 } 1390 1391 static void __inline 1392 vmx_set_nmi_window_exiting(struct vmx_vcpu *vcpu) 1393 { 1394 1395 if ((vcpu->cap.proc_ctls & PROCBASED_NMI_WINDOW_EXITING) == 0) { 1396 vcpu->cap.proc_ctls |= PROCBASED_NMI_WINDOW_EXITING; 1397 vmcs_write(VMCS_PRI_PROC_BASED_CTLS, vcpu->cap.proc_ctls); 1398 VMX_CTR0(vcpu, "Enabling NMI window exiting"); 1399 } 1400 } 1401 1402 static void __inline 1403 vmx_clear_nmi_window_exiting(struct vmx_vcpu *vcpu) 1404 { 1405 1406 KASSERT((vcpu->cap.proc_ctls & PROCBASED_NMI_WINDOW_EXITING) != 0, 1407 ("nmi_window_exiting not set %#x", vcpu->cap.proc_ctls)); 1408 vcpu->cap.proc_ctls &= ~PROCBASED_NMI_WINDOW_EXITING; 1409 vmcs_write(VMCS_PRI_PROC_BASED_CTLS, vcpu->cap.proc_ctls); 1410 VMX_CTR0(vcpu, "Disabling NMI window exiting"); 1411 } 1412 1413 int 1414 vmx_set_tsc_offset(struct vmx_vcpu *vcpu, uint64_t offset) 1415 { 1416 int error; 1417 1418 if ((vcpu->cap.proc_ctls & PROCBASED_TSC_OFFSET) == 0) { 1419 vcpu->cap.proc_ctls |= PROCBASED_TSC_OFFSET; 1420 vmcs_write(VMCS_PRI_PROC_BASED_CTLS, vcpu->cap.proc_ctls); 1421 VMX_CTR0(vcpu, "Enabling TSC offsetting"); 1422 } 1423 1424 error = vmwrite(VMCS_TSC_OFFSET, offset); 1425 #ifdef BHYVE_SNAPSHOT 1426 if (error == 0) 1427 vm_set_tsc_offset(vcpu->vcpu, offset); 1428 #endif 1429 return (error); 1430 } 1431 1432 #define NMI_BLOCKING (VMCS_INTERRUPTIBILITY_NMI_BLOCKING | \ 1433 VMCS_INTERRUPTIBILITY_MOVSS_BLOCKING) 1434 #define HWINTR_BLOCKING (VMCS_INTERRUPTIBILITY_STI_BLOCKING | \ 1435 VMCS_INTERRUPTIBILITY_MOVSS_BLOCKING) 1436 1437 static void 1438 vmx_inject_nmi(struct vmx_vcpu *vcpu) 1439 { 1440 uint32_t gi __diagused, info; 1441 1442 gi = vmcs_read(VMCS_GUEST_INTERRUPTIBILITY); 1443 KASSERT((gi & NMI_BLOCKING) == 0, ("vmx_inject_nmi: invalid guest " 1444 "interruptibility-state %#x", gi)); 1445 1446 info = vmcs_read(VMCS_ENTRY_INTR_INFO); 1447 KASSERT((info & VMCS_INTR_VALID) == 0, ("vmx_inject_nmi: invalid " 1448 "VM-entry interruption information %#x", info)); 1449 1450 /* 1451 * Inject the virtual NMI. The vector must be the NMI IDT entry 1452 * or the VMCS entry check will fail. 1453 */ 1454 info = IDT_NMI | VMCS_INTR_T_NMI | VMCS_INTR_VALID; 1455 vmcs_write(VMCS_ENTRY_INTR_INFO, info); 1456 1457 VMX_CTR0(vcpu, "Injecting vNMI"); 1458 1459 /* Clear the request */ 1460 vm_nmi_clear(vcpu->vcpu); 1461 } 1462 1463 static void 1464 vmx_inject_interrupts(struct vmx_vcpu *vcpu, struct vlapic *vlapic, 1465 uint64_t guestrip) 1466 { 1467 int vector, need_nmi_exiting, extint_pending; 1468 uint64_t rflags, entryinfo; 1469 uint32_t gi, info; 1470 1471 if (vcpu->cap.set & (1 << VM_CAP_MASK_HWINTR)) { 1472 return; 1473 } 1474 1475 if (vcpu->state.nextrip != guestrip) { 1476 gi = vmcs_read(VMCS_GUEST_INTERRUPTIBILITY); 1477 if (gi & HWINTR_BLOCKING) { 1478 VMX_CTR2(vcpu, "Guest interrupt blocking " 1479 "cleared due to rip change: %#lx/%#lx", 1480 vcpu->state.nextrip, guestrip); 1481 gi &= ~HWINTR_BLOCKING; 1482 vmcs_write(VMCS_GUEST_INTERRUPTIBILITY, gi); 1483 } 1484 } 1485 1486 if (vm_entry_intinfo(vcpu->vcpu, &entryinfo)) { 1487 KASSERT((entryinfo & VMCS_INTR_VALID) != 0, ("%s: entry " 1488 "intinfo is not valid: %#lx", __func__, entryinfo)); 1489 1490 info = vmcs_read(VMCS_ENTRY_INTR_INFO); 1491 KASSERT((info & VMCS_INTR_VALID) == 0, ("%s: cannot inject " 1492 "pending exception: %#lx/%#x", __func__, entryinfo, info)); 1493 1494 info = entryinfo; 1495 vector = info & 0xff; 1496 if (vector == IDT_BP || vector == IDT_OF) { 1497 /* 1498 * VT-x requires #BP and #OF to be injected as software 1499 * exceptions. 1500 */ 1501 info &= ~VMCS_INTR_T_MASK; 1502 info |= VMCS_INTR_T_SWEXCEPTION; 1503 } 1504 1505 if (info & VMCS_INTR_DEL_ERRCODE) 1506 vmcs_write(VMCS_ENTRY_EXCEPTION_ERROR, entryinfo >> 32); 1507 1508 vmcs_write(VMCS_ENTRY_INTR_INFO, info); 1509 } 1510 1511 if (vm_nmi_pending(vcpu->vcpu)) { 1512 /* 1513 * If there are no conditions blocking NMI injection then 1514 * inject it directly here otherwise enable "NMI window 1515 * exiting" to inject it as soon as we can. 1516 * 1517 * We also check for STI_BLOCKING because some implementations 1518 * don't allow NMI injection in this case. If we are running 1519 * on a processor that doesn't have this restriction it will 1520 * immediately exit and the NMI will be injected in the 1521 * "NMI window exiting" handler. 1522 */ 1523 need_nmi_exiting = 1; 1524 gi = vmcs_read(VMCS_GUEST_INTERRUPTIBILITY); 1525 if ((gi & (HWINTR_BLOCKING | NMI_BLOCKING)) == 0) { 1526 info = vmcs_read(VMCS_ENTRY_INTR_INFO); 1527 if ((info & VMCS_INTR_VALID) == 0) { 1528 vmx_inject_nmi(vcpu); 1529 need_nmi_exiting = 0; 1530 } else { 1531 VMX_CTR1(vcpu, "Cannot inject NMI " 1532 "due to VM-entry intr info %#x", info); 1533 } 1534 } else { 1535 VMX_CTR1(vcpu, "Cannot inject NMI due to " 1536 "Guest Interruptibility-state %#x", gi); 1537 } 1538 1539 if (need_nmi_exiting) 1540 vmx_set_nmi_window_exiting(vcpu); 1541 } 1542 1543 extint_pending = vm_extint_pending(vcpu->vcpu); 1544 1545 if (!extint_pending && virtual_interrupt_delivery) { 1546 vmx_inject_pir(vlapic); 1547 return; 1548 } 1549 1550 /* 1551 * If interrupt-window exiting is already in effect then don't bother 1552 * checking for pending interrupts. This is just an optimization and 1553 * not needed for correctness. 1554 */ 1555 if ((vcpu->cap.proc_ctls & PROCBASED_INT_WINDOW_EXITING) != 0) { 1556 VMX_CTR0(vcpu, "Skip interrupt injection due to " 1557 "pending int_window_exiting"); 1558 return; 1559 } 1560 1561 if (!extint_pending) { 1562 /* Ask the local apic for a vector to inject */ 1563 if (!vlapic_pending_intr(vlapic, &vector)) 1564 return; 1565 1566 /* 1567 * From the Intel SDM, Volume 3, Section "Maskable 1568 * Hardware Interrupts": 1569 * - maskable interrupt vectors [16,255] can be delivered 1570 * through the local APIC. 1571 */ 1572 KASSERT(vector >= 16 && vector <= 255, 1573 ("invalid vector %d from local APIC", vector)); 1574 } else { 1575 /* Ask the legacy pic for a vector to inject */ 1576 vatpic_pending_intr(vcpu->vmx->vm, &vector); 1577 1578 /* 1579 * From the Intel SDM, Volume 3, Section "Maskable 1580 * Hardware Interrupts": 1581 * - maskable interrupt vectors [0,255] can be delivered 1582 * through the INTR pin. 1583 */ 1584 KASSERT(vector >= 0 && vector <= 255, 1585 ("invalid vector %d from INTR", vector)); 1586 } 1587 1588 /* Check RFLAGS.IF and the interruptibility state of the guest */ 1589 rflags = vmcs_read(VMCS_GUEST_RFLAGS); 1590 if ((rflags & PSL_I) == 0) { 1591 VMX_CTR2(vcpu, "Cannot inject vector %d due to " 1592 "rflags %#lx", vector, rflags); 1593 goto cantinject; 1594 } 1595 1596 gi = vmcs_read(VMCS_GUEST_INTERRUPTIBILITY); 1597 if (gi & HWINTR_BLOCKING) { 1598 VMX_CTR2(vcpu, "Cannot inject vector %d due to " 1599 "Guest Interruptibility-state %#x", vector, gi); 1600 goto cantinject; 1601 } 1602 1603 info = vmcs_read(VMCS_ENTRY_INTR_INFO); 1604 if (info & VMCS_INTR_VALID) { 1605 /* 1606 * This is expected and could happen for multiple reasons: 1607 * - A vectoring VM-entry was aborted due to astpending 1608 * - A VM-exit happened during event injection. 1609 * - An exception was injected above. 1610 * - An NMI was injected above or after "NMI window exiting" 1611 */ 1612 VMX_CTR2(vcpu, "Cannot inject vector %d due to " 1613 "VM-entry intr info %#x", vector, info); 1614 goto cantinject; 1615 } 1616 1617 /* Inject the interrupt */ 1618 info = VMCS_INTR_T_HWINTR | VMCS_INTR_VALID; 1619 info |= vector; 1620 vmcs_write(VMCS_ENTRY_INTR_INFO, info); 1621 1622 if (!extint_pending) { 1623 /* Update the Local APIC ISR */ 1624 vlapic_intr_accepted(vlapic, vector); 1625 } else { 1626 vm_extint_clear(vcpu->vcpu); 1627 vatpic_intr_accepted(vcpu->vmx->vm, vector); 1628 1629 /* 1630 * After we accepted the current ExtINT the PIC may 1631 * have posted another one. If that is the case, set 1632 * the Interrupt Window Exiting execution control so 1633 * we can inject that one too. 1634 * 1635 * Also, interrupt window exiting allows us to inject any 1636 * pending APIC vector that was preempted by the ExtINT 1637 * as soon as possible. This applies both for the software 1638 * emulated vlapic and the hardware assisted virtual APIC. 1639 */ 1640 vmx_set_int_window_exiting(vcpu); 1641 } 1642 1643 VMX_CTR1(vcpu, "Injecting hwintr at vector %d", vector); 1644 1645 return; 1646 1647 cantinject: 1648 /* 1649 * Set the Interrupt Window Exiting execution control so we can inject 1650 * the interrupt as soon as blocking condition goes away. 1651 */ 1652 vmx_set_int_window_exiting(vcpu); 1653 } 1654 1655 /* 1656 * If the Virtual NMIs execution control is '1' then the logical processor 1657 * tracks virtual-NMI blocking in the Guest Interruptibility-state field of 1658 * the VMCS. An IRET instruction in VMX non-root operation will remove any 1659 * virtual-NMI blocking. 1660 * 1661 * This unblocking occurs even if the IRET causes a fault. In this case the 1662 * hypervisor needs to restore virtual-NMI blocking before resuming the guest. 1663 */ 1664 static void 1665 vmx_restore_nmi_blocking(struct vmx_vcpu *vcpu) 1666 { 1667 uint32_t gi; 1668 1669 VMX_CTR0(vcpu, "Restore Virtual-NMI blocking"); 1670 gi = vmcs_read(VMCS_GUEST_INTERRUPTIBILITY); 1671 gi |= VMCS_INTERRUPTIBILITY_NMI_BLOCKING; 1672 vmcs_write(VMCS_GUEST_INTERRUPTIBILITY, gi); 1673 } 1674 1675 static void 1676 vmx_clear_nmi_blocking(struct vmx_vcpu *vcpu) 1677 { 1678 uint32_t gi; 1679 1680 VMX_CTR0(vcpu, "Clear Virtual-NMI blocking"); 1681 gi = vmcs_read(VMCS_GUEST_INTERRUPTIBILITY); 1682 gi &= ~VMCS_INTERRUPTIBILITY_NMI_BLOCKING; 1683 vmcs_write(VMCS_GUEST_INTERRUPTIBILITY, gi); 1684 } 1685 1686 static void 1687 vmx_assert_nmi_blocking(struct vmx_vcpu *vcpu) 1688 { 1689 uint32_t gi __diagused; 1690 1691 gi = vmcs_read(VMCS_GUEST_INTERRUPTIBILITY); 1692 KASSERT(gi & VMCS_INTERRUPTIBILITY_NMI_BLOCKING, 1693 ("NMI blocking is not in effect %#x", gi)); 1694 } 1695 1696 static int 1697 vmx_emulate_xsetbv(struct vmx *vmx, struct vmx_vcpu *vcpu, 1698 struct vm_exit *vmexit) 1699 { 1700 struct vmxctx *vmxctx; 1701 uint64_t xcrval; 1702 const struct xsave_limits *limits; 1703 1704 vmxctx = &vcpu->ctx; 1705 limits = vmm_get_xsave_limits(); 1706 1707 /* 1708 * Note that the processor raises a GP# fault on its own if 1709 * xsetbv is executed for CPL != 0, so we do not have to 1710 * emulate that fault here. 1711 */ 1712 1713 /* Only xcr0 is supported. */ 1714 if (vmxctx->guest_rcx != 0) { 1715 vm_inject_gp(vcpu->vcpu); 1716 return (HANDLED); 1717 } 1718 1719 /* We only handle xcr0 if both the host and guest have XSAVE enabled. */ 1720 if (!limits->xsave_enabled || !(vmcs_read(VMCS_GUEST_CR4) & CR4_XSAVE)) { 1721 vm_inject_ud(vcpu->vcpu); 1722 return (HANDLED); 1723 } 1724 1725 xcrval = vmxctx->guest_rdx << 32 | (vmxctx->guest_rax & 0xffffffff); 1726 if ((xcrval & ~limits->xcr0_allowed) != 0) { 1727 vm_inject_gp(vcpu->vcpu); 1728 return (HANDLED); 1729 } 1730 1731 if (!(xcrval & XFEATURE_ENABLED_X87)) { 1732 vm_inject_gp(vcpu->vcpu); 1733 return (HANDLED); 1734 } 1735 1736 /* AVX (YMM_Hi128) requires SSE. */ 1737 if (xcrval & XFEATURE_ENABLED_AVX && 1738 (xcrval & XFEATURE_AVX) != XFEATURE_AVX) { 1739 vm_inject_gp(vcpu->vcpu); 1740 return (HANDLED); 1741 } 1742 1743 /* 1744 * AVX512 requires base AVX (YMM_Hi128) as well as OpMask, 1745 * ZMM_Hi256, and Hi16_ZMM. 1746 */ 1747 if (xcrval & XFEATURE_AVX512 && 1748 (xcrval & (XFEATURE_AVX512 | XFEATURE_AVX)) != 1749 (XFEATURE_AVX512 | XFEATURE_AVX)) { 1750 vm_inject_gp(vcpu->vcpu); 1751 return (HANDLED); 1752 } 1753 1754 /* 1755 * Intel MPX requires both bound register state flags to be 1756 * set. 1757 */ 1758 if (((xcrval & XFEATURE_ENABLED_BNDREGS) != 0) != 1759 ((xcrval & XFEATURE_ENABLED_BNDCSR) != 0)) { 1760 vm_inject_gp(vcpu->vcpu); 1761 return (HANDLED); 1762 } 1763 1764 /* 1765 * This runs "inside" vmrun() with the guest's FPU state, so 1766 * modifying xcr0 directly modifies the guest's xcr0, not the 1767 * host's. 1768 */ 1769 load_xcr(0, xcrval); 1770 return (HANDLED); 1771 } 1772 1773 static uint64_t 1774 vmx_get_guest_reg(struct vmx_vcpu *vcpu, int ident) 1775 { 1776 const struct vmxctx *vmxctx; 1777 1778 vmxctx = &vcpu->ctx; 1779 1780 switch (ident) { 1781 case 0: 1782 return (vmxctx->guest_rax); 1783 case 1: 1784 return (vmxctx->guest_rcx); 1785 case 2: 1786 return (vmxctx->guest_rdx); 1787 case 3: 1788 return (vmxctx->guest_rbx); 1789 case 4: 1790 return (vmcs_read(VMCS_GUEST_RSP)); 1791 case 5: 1792 return (vmxctx->guest_rbp); 1793 case 6: 1794 return (vmxctx->guest_rsi); 1795 case 7: 1796 return (vmxctx->guest_rdi); 1797 case 8: 1798 return (vmxctx->guest_r8); 1799 case 9: 1800 return (vmxctx->guest_r9); 1801 case 10: 1802 return (vmxctx->guest_r10); 1803 case 11: 1804 return (vmxctx->guest_r11); 1805 case 12: 1806 return (vmxctx->guest_r12); 1807 case 13: 1808 return (vmxctx->guest_r13); 1809 case 14: 1810 return (vmxctx->guest_r14); 1811 case 15: 1812 return (vmxctx->guest_r15); 1813 default: 1814 panic("invalid vmx register %d", ident); 1815 } 1816 } 1817 1818 static void 1819 vmx_set_guest_reg(struct vmx_vcpu *vcpu, int ident, uint64_t regval) 1820 { 1821 struct vmxctx *vmxctx; 1822 1823 vmxctx = &vcpu->ctx; 1824 1825 switch (ident) { 1826 case 0: 1827 vmxctx->guest_rax = regval; 1828 break; 1829 case 1: 1830 vmxctx->guest_rcx = regval; 1831 break; 1832 case 2: 1833 vmxctx->guest_rdx = regval; 1834 break; 1835 case 3: 1836 vmxctx->guest_rbx = regval; 1837 break; 1838 case 4: 1839 vmcs_write(VMCS_GUEST_RSP, regval); 1840 break; 1841 case 5: 1842 vmxctx->guest_rbp = regval; 1843 break; 1844 case 6: 1845 vmxctx->guest_rsi = regval; 1846 break; 1847 case 7: 1848 vmxctx->guest_rdi = regval; 1849 break; 1850 case 8: 1851 vmxctx->guest_r8 = regval; 1852 break; 1853 case 9: 1854 vmxctx->guest_r9 = regval; 1855 break; 1856 case 10: 1857 vmxctx->guest_r10 = regval; 1858 break; 1859 case 11: 1860 vmxctx->guest_r11 = regval; 1861 break; 1862 case 12: 1863 vmxctx->guest_r12 = regval; 1864 break; 1865 case 13: 1866 vmxctx->guest_r13 = regval; 1867 break; 1868 case 14: 1869 vmxctx->guest_r14 = regval; 1870 break; 1871 case 15: 1872 vmxctx->guest_r15 = regval; 1873 break; 1874 default: 1875 panic("invalid vmx register %d", ident); 1876 } 1877 } 1878 1879 static int 1880 vmx_emulate_cr0_access(struct vmx_vcpu *vcpu, uint64_t exitqual) 1881 { 1882 uint64_t crval, regval; 1883 1884 /* We only handle mov to %cr0 at this time */ 1885 if ((exitqual & 0xf0) != 0x00) 1886 return (UNHANDLED); 1887 1888 regval = vmx_get_guest_reg(vcpu, (exitqual >> 8) & 0xf); 1889 1890 vmcs_write(VMCS_CR0_SHADOW, regval); 1891 1892 crval = regval | cr0_ones_mask; 1893 crval &= ~cr0_zeros_mask; 1894 vmcs_write(VMCS_GUEST_CR0, crval); 1895 1896 if (regval & CR0_PG) { 1897 uint64_t efer, entry_ctls; 1898 1899 /* 1900 * If CR0.PG is 1 and EFER.LME is 1 then EFER.LMA and 1901 * the "IA-32e mode guest" bit in VM-entry control must be 1902 * equal. 1903 */ 1904 efer = vmcs_read(VMCS_GUEST_IA32_EFER); 1905 if (efer & EFER_LME) { 1906 efer |= EFER_LMA; 1907 vmcs_write(VMCS_GUEST_IA32_EFER, efer); 1908 entry_ctls = vmcs_read(VMCS_ENTRY_CTLS); 1909 entry_ctls |= VM_ENTRY_GUEST_LMA; 1910 vmcs_write(VMCS_ENTRY_CTLS, entry_ctls); 1911 } 1912 } 1913 1914 return (HANDLED); 1915 } 1916 1917 static int 1918 vmx_emulate_cr4_access(struct vmx_vcpu *vcpu, uint64_t exitqual) 1919 { 1920 uint64_t crval, regval; 1921 1922 /* We only handle mov to %cr4 at this time */ 1923 if ((exitqual & 0xf0) != 0x00) 1924 return (UNHANDLED); 1925 1926 regval = vmx_get_guest_reg(vcpu, (exitqual >> 8) & 0xf); 1927 1928 vmcs_write(VMCS_CR4_SHADOW, regval); 1929 1930 crval = regval | cr4_ones_mask; 1931 crval &= ~cr4_zeros_mask; 1932 vmcs_write(VMCS_GUEST_CR4, crval); 1933 1934 return (HANDLED); 1935 } 1936 1937 static int 1938 vmx_emulate_cr8_access(struct vmx *vmx, struct vmx_vcpu *vcpu, 1939 uint64_t exitqual) 1940 { 1941 struct vlapic *vlapic; 1942 uint64_t cr8; 1943 int regnum; 1944 1945 /* We only handle mov %cr8 to/from a register at this time. */ 1946 if ((exitqual & 0xe0) != 0x00) { 1947 return (UNHANDLED); 1948 } 1949 1950 vlapic = vm_lapic(vcpu->vcpu); 1951 regnum = (exitqual >> 8) & 0xf; 1952 if (exitqual & 0x10) { 1953 cr8 = vlapic_get_cr8(vlapic); 1954 vmx_set_guest_reg(vcpu, regnum, cr8); 1955 } else { 1956 cr8 = vmx_get_guest_reg(vcpu, regnum); 1957 vlapic_set_cr8(vlapic, cr8); 1958 } 1959 1960 return (HANDLED); 1961 } 1962 1963 /* 1964 * From section "Guest Register State" in the Intel SDM: CPL = SS.DPL 1965 */ 1966 static int 1967 vmx_cpl(void) 1968 { 1969 uint32_t ssar; 1970 1971 ssar = vmcs_read(VMCS_GUEST_SS_ACCESS_RIGHTS); 1972 return ((ssar >> 5) & 0x3); 1973 } 1974 1975 static enum vm_cpu_mode 1976 vmx_cpu_mode(void) 1977 { 1978 uint32_t csar; 1979 1980 if (vmcs_read(VMCS_GUEST_IA32_EFER) & EFER_LMA) { 1981 csar = vmcs_read(VMCS_GUEST_CS_ACCESS_RIGHTS); 1982 if (csar & 0x2000) 1983 return (CPU_MODE_64BIT); /* CS.L = 1 */ 1984 else 1985 return (CPU_MODE_COMPATIBILITY); 1986 } else if (vmcs_read(VMCS_GUEST_CR0) & CR0_PE) { 1987 return (CPU_MODE_PROTECTED); 1988 } else { 1989 return (CPU_MODE_REAL); 1990 } 1991 } 1992 1993 static enum vm_paging_mode 1994 vmx_paging_mode(void) 1995 { 1996 uint64_t cr4; 1997 1998 if (!(vmcs_read(VMCS_GUEST_CR0) & CR0_PG)) 1999 return (PAGING_MODE_FLAT); 2000 cr4 = vmcs_read(VMCS_GUEST_CR4); 2001 if (!(cr4 & CR4_PAE)) 2002 return (PAGING_MODE_32); 2003 if (vmcs_read(VMCS_GUEST_IA32_EFER) & EFER_LME) { 2004 if (!(cr4 & CR4_LA57)) 2005 return (PAGING_MODE_64); 2006 return (PAGING_MODE_64_LA57); 2007 } else 2008 return (PAGING_MODE_PAE); 2009 } 2010 2011 static uint64_t 2012 inout_str_index(struct vmx_vcpu *vcpu, int in) 2013 { 2014 uint64_t val; 2015 int error __diagused; 2016 enum vm_reg_name reg; 2017 2018 reg = in ? VM_REG_GUEST_RDI : VM_REG_GUEST_RSI; 2019 error = vmx_getreg(vcpu, reg, &val); 2020 KASSERT(error == 0, ("%s: vmx_getreg error %d", __func__, error)); 2021 return (val); 2022 } 2023 2024 static uint64_t 2025 inout_str_count(struct vmx_vcpu *vcpu, int rep) 2026 { 2027 uint64_t val; 2028 int error __diagused; 2029 2030 if (rep) { 2031 error = vmx_getreg(vcpu, VM_REG_GUEST_RCX, &val); 2032 KASSERT(!error, ("%s: vmx_getreg error %d", __func__, error)); 2033 } else { 2034 val = 1; 2035 } 2036 return (val); 2037 } 2038 2039 static int 2040 inout_str_addrsize(uint32_t inst_info) 2041 { 2042 uint32_t size; 2043 2044 size = (inst_info >> 7) & 0x7; 2045 switch (size) { 2046 case 0: 2047 return (2); /* 16 bit */ 2048 case 1: 2049 return (4); /* 32 bit */ 2050 case 2: 2051 return (8); /* 64 bit */ 2052 default: 2053 panic("%s: invalid size encoding %d", __func__, size); 2054 } 2055 } 2056 2057 static void 2058 inout_str_seginfo(struct vmx_vcpu *vcpu, uint32_t inst_info, int in, 2059 struct vm_inout_str *vis) 2060 { 2061 int error __diagused, s; 2062 2063 if (in) { 2064 vis->seg_name = VM_REG_GUEST_ES; 2065 } else { 2066 s = (inst_info >> 15) & 0x7; 2067 vis->seg_name = vm_segment_name(s); 2068 } 2069 2070 error = vmx_getdesc(vcpu, vis->seg_name, &vis->seg_desc); 2071 KASSERT(error == 0, ("%s: vmx_getdesc error %d", __func__, error)); 2072 } 2073 2074 static void 2075 vmx_paging_info(struct vm_guest_paging *paging) 2076 { 2077 paging->cr3 = vmcs_guest_cr3(); 2078 paging->cpl = vmx_cpl(); 2079 paging->cpu_mode = vmx_cpu_mode(); 2080 paging->paging_mode = vmx_paging_mode(); 2081 } 2082 2083 static void 2084 vmexit_inst_emul(struct vm_exit *vmexit, uint64_t gpa, uint64_t gla) 2085 { 2086 struct vm_guest_paging *paging; 2087 uint32_t csar; 2088 2089 paging = &vmexit->u.inst_emul.paging; 2090 2091 vmexit->exitcode = VM_EXITCODE_INST_EMUL; 2092 vmexit->inst_length = 0; 2093 vmexit->u.inst_emul.gpa = gpa; 2094 vmexit->u.inst_emul.gla = gla; 2095 vmx_paging_info(paging); 2096 switch (paging->cpu_mode) { 2097 case CPU_MODE_REAL: 2098 vmexit->u.inst_emul.cs_base = vmcs_read(VMCS_GUEST_CS_BASE); 2099 vmexit->u.inst_emul.cs_d = 0; 2100 break; 2101 case CPU_MODE_PROTECTED: 2102 case CPU_MODE_COMPATIBILITY: 2103 vmexit->u.inst_emul.cs_base = vmcs_read(VMCS_GUEST_CS_BASE); 2104 csar = vmcs_read(VMCS_GUEST_CS_ACCESS_RIGHTS); 2105 vmexit->u.inst_emul.cs_d = SEG_DESC_DEF32(csar); 2106 break; 2107 default: 2108 vmexit->u.inst_emul.cs_base = 0; 2109 vmexit->u.inst_emul.cs_d = 0; 2110 break; 2111 } 2112 vie_init(&vmexit->u.inst_emul.vie, NULL, 0); 2113 } 2114 2115 static int 2116 ept_fault_type(uint64_t ept_qual) 2117 { 2118 int fault_type; 2119 2120 if (ept_qual & EPT_VIOLATION_DATA_WRITE) 2121 fault_type = VM_PROT_WRITE; 2122 else if (ept_qual & EPT_VIOLATION_INST_FETCH) 2123 fault_type = VM_PROT_EXECUTE; 2124 else 2125 fault_type= VM_PROT_READ; 2126 2127 return (fault_type); 2128 } 2129 2130 static bool 2131 ept_emulation_fault(uint64_t ept_qual) 2132 { 2133 int read, write; 2134 2135 /* EPT fault on an instruction fetch doesn't make sense here */ 2136 if (ept_qual & EPT_VIOLATION_INST_FETCH) 2137 return (false); 2138 2139 /* EPT fault must be a read fault or a write fault */ 2140 read = ept_qual & EPT_VIOLATION_DATA_READ ? 1 : 0; 2141 write = ept_qual & EPT_VIOLATION_DATA_WRITE ? 1 : 0; 2142 if ((read | write) == 0) 2143 return (false); 2144 2145 /* 2146 * The EPT violation must have been caused by accessing a 2147 * guest-physical address that is a translation of a guest-linear 2148 * address. 2149 */ 2150 if ((ept_qual & EPT_VIOLATION_GLA_VALID) == 0 || 2151 (ept_qual & EPT_VIOLATION_XLAT_VALID) == 0) { 2152 return (false); 2153 } 2154 2155 return (true); 2156 } 2157 2158 static __inline int 2159 apic_access_virtualization(struct vmx_vcpu *vcpu) 2160 { 2161 uint32_t proc_ctls2; 2162 2163 proc_ctls2 = vcpu->cap.proc_ctls2; 2164 return ((proc_ctls2 & PROCBASED2_VIRTUALIZE_APIC_ACCESSES) ? 1 : 0); 2165 } 2166 2167 static __inline int 2168 x2apic_virtualization(struct vmx_vcpu *vcpu) 2169 { 2170 uint32_t proc_ctls2; 2171 2172 proc_ctls2 = vcpu->cap.proc_ctls2; 2173 return ((proc_ctls2 & PROCBASED2_VIRTUALIZE_X2APIC_MODE) ? 1 : 0); 2174 } 2175 2176 static int 2177 vmx_handle_apic_write(struct vmx_vcpu *vcpu, struct vlapic *vlapic, 2178 uint64_t qual) 2179 { 2180 int error, handled, offset; 2181 uint32_t *apic_regs, vector; 2182 bool retu; 2183 2184 handled = HANDLED; 2185 offset = APIC_WRITE_OFFSET(qual); 2186 2187 if (!apic_access_virtualization(vcpu)) { 2188 /* 2189 * In general there should not be any APIC write VM-exits 2190 * unless APIC-access virtualization is enabled. 2191 * 2192 * However self-IPI virtualization can legitimately trigger 2193 * an APIC-write VM-exit so treat it specially. 2194 */ 2195 if (x2apic_virtualization(vcpu) && 2196 offset == APIC_OFFSET_SELF_IPI) { 2197 apic_regs = (uint32_t *)(vlapic->apic_page); 2198 vector = apic_regs[APIC_OFFSET_SELF_IPI / 4]; 2199 vlapic_self_ipi_handler(vlapic, vector); 2200 return (HANDLED); 2201 } else 2202 return (UNHANDLED); 2203 } 2204 2205 switch (offset) { 2206 case APIC_OFFSET_ID: 2207 vlapic_id_write_handler(vlapic); 2208 break; 2209 case APIC_OFFSET_LDR: 2210 vlapic_ldr_write_handler(vlapic); 2211 break; 2212 case APIC_OFFSET_DFR: 2213 vlapic_dfr_write_handler(vlapic); 2214 break; 2215 case APIC_OFFSET_SVR: 2216 vlapic_svr_write_handler(vlapic); 2217 break; 2218 case APIC_OFFSET_ESR: 2219 vlapic_esr_write_handler(vlapic); 2220 break; 2221 case APIC_OFFSET_ICR_LOW: 2222 retu = false; 2223 error = vlapic_icrlo_write_handler(vlapic, &retu); 2224 if (error != 0 || retu) 2225 handled = UNHANDLED; 2226 break; 2227 case APIC_OFFSET_CMCI_LVT: 2228 case APIC_OFFSET_TIMER_LVT ... APIC_OFFSET_ERROR_LVT: 2229 vlapic_lvt_write_handler(vlapic, offset); 2230 break; 2231 case APIC_OFFSET_TIMER_ICR: 2232 vlapic_icrtmr_write_handler(vlapic); 2233 break; 2234 case APIC_OFFSET_TIMER_DCR: 2235 vlapic_dcr_write_handler(vlapic); 2236 break; 2237 default: 2238 handled = UNHANDLED; 2239 break; 2240 } 2241 return (handled); 2242 } 2243 2244 static bool 2245 apic_access_fault(struct vmx_vcpu *vcpu, uint64_t gpa) 2246 { 2247 2248 if (apic_access_virtualization(vcpu) && 2249 (gpa >= DEFAULT_APIC_BASE && gpa < DEFAULT_APIC_BASE + PAGE_SIZE)) 2250 return (true); 2251 else 2252 return (false); 2253 } 2254 2255 static int 2256 vmx_handle_apic_access(struct vmx_vcpu *vcpu, struct vm_exit *vmexit) 2257 { 2258 uint64_t qual; 2259 int access_type, offset, allowed; 2260 2261 if (!apic_access_virtualization(vcpu)) 2262 return (UNHANDLED); 2263 2264 qual = vmexit->u.vmx.exit_qualification; 2265 access_type = APIC_ACCESS_TYPE(qual); 2266 offset = APIC_ACCESS_OFFSET(qual); 2267 2268 allowed = 0; 2269 if (access_type == 0) { 2270 /* 2271 * Read data access to the following registers is expected. 2272 */ 2273 switch (offset) { 2274 case APIC_OFFSET_APR: 2275 case APIC_OFFSET_PPR: 2276 case APIC_OFFSET_RRR: 2277 case APIC_OFFSET_CMCI_LVT: 2278 case APIC_OFFSET_TIMER_CCR: 2279 allowed = 1; 2280 break; 2281 default: 2282 break; 2283 } 2284 } else if (access_type == 1) { 2285 /* 2286 * Write data access to the following registers is expected. 2287 */ 2288 switch (offset) { 2289 case APIC_OFFSET_VER: 2290 case APIC_OFFSET_APR: 2291 case APIC_OFFSET_PPR: 2292 case APIC_OFFSET_RRR: 2293 case APIC_OFFSET_ISR0 ... APIC_OFFSET_ISR7: 2294 case APIC_OFFSET_TMR0 ... APIC_OFFSET_TMR7: 2295 case APIC_OFFSET_IRR0 ... APIC_OFFSET_IRR7: 2296 case APIC_OFFSET_CMCI_LVT: 2297 case APIC_OFFSET_TIMER_CCR: 2298 allowed = 1; 2299 break; 2300 default: 2301 break; 2302 } 2303 } 2304 2305 if (allowed) { 2306 vmexit_inst_emul(vmexit, DEFAULT_APIC_BASE + offset, 2307 VIE_INVALID_GLA); 2308 } 2309 2310 /* 2311 * Regardless of whether the APIC-access is allowed this handler 2312 * always returns UNHANDLED: 2313 * - if the access is allowed then it is handled by emulating the 2314 * instruction that caused the VM-exit (outside the critical section) 2315 * - if the access is not allowed then it will be converted to an 2316 * exitcode of VM_EXITCODE_VMX and will be dealt with in userland. 2317 */ 2318 return (UNHANDLED); 2319 } 2320 2321 static enum task_switch_reason 2322 vmx_task_switch_reason(uint64_t qual) 2323 { 2324 int reason; 2325 2326 reason = (qual >> 30) & 0x3; 2327 switch (reason) { 2328 case 0: 2329 return (TSR_CALL); 2330 case 1: 2331 return (TSR_IRET); 2332 case 2: 2333 return (TSR_JMP); 2334 case 3: 2335 return (TSR_IDT_GATE); 2336 default: 2337 panic("%s: invalid reason %d", __func__, reason); 2338 } 2339 } 2340 2341 static int 2342 emulate_wrmsr(struct vmx_vcpu *vcpu, u_int num, uint64_t val, bool *retu) 2343 { 2344 int error; 2345 2346 if (lapic_msr(num)) 2347 error = lapic_wrmsr(vcpu->vcpu, num, val, retu); 2348 else 2349 error = vmx_wrmsr(vcpu, num, val, retu); 2350 2351 return (error); 2352 } 2353 2354 static int 2355 emulate_rdmsr(struct vmx_vcpu *vcpu, u_int num, bool *retu) 2356 { 2357 struct vmxctx *vmxctx; 2358 uint64_t result; 2359 uint32_t eax, edx; 2360 int error; 2361 2362 if (lapic_msr(num)) 2363 error = lapic_rdmsr(vcpu->vcpu, num, &result, retu); 2364 else 2365 error = vmx_rdmsr(vcpu, num, &result, retu); 2366 2367 if (error == 0) { 2368 eax = result; 2369 vmxctx = &vcpu->ctx; 2370 error = vmxctx_setreg(vmxctx, VM_REG_GUEST_RAX, eax); 2371 KASSERT(error == 0, ("vmxctx_setreg(rax) error %d", error)); 2372 2373 edx = result >> 32; 2374 error = vmxctx_setreg(vmxctx, VM_REG_GUEST_RDX, edx); 2375 KASSERT(error == 0, ("vmxctx_setreg(rdx) error %d", error)); 2376 } 2377 2378 return (error); 2379 } 2380 2381 static int 2382 vmx_exit_process(struct vmx *vmx, struct vmx_vcpu *vcpu, struct vm_exit *vmexit) 2383 { 2384 int error, errcode, errcode_valid, handled, in; 2385 struct vmxctx *vmxctx; 2386 struct vlapic *vlapic; 2387 struct vm_inout_str *vis; 2388 struct vm_task_switch *ts; 2389 uint32_t eax, ecx, edx, idtvec_info, idtvec_err, intr_info, inst_info; 2390 uint32_t intr_type, intr_vec, reason; 2391 uint64_t exitintinfo, qual, gpa; 2392 #ifdef KDTRACE_HOOKS 2393 int vcpuid; 2394 #endif 2395 bool retu; 2396 2397 CTASSERT((PINBASED_CTLS_ONE_SETTING & PINBASED_VIRTUAL_NMI) != 0); 2398 CTASSERT((PINBASED_CTLS_ONE_SETTING & PINBASED_NMI_EXITING) != 0); 2399 2400 handled = UNHANDLED; 2401 vmxctx = &vcpu->ctx; 2402 #ifdef KDTRACE_HOOKS 2403 vcpuid = vcpu->vcpuid; 2404 #endif 2405 2406 qual = vmexit->u.vmx.exit_qualification; 2407 reason = vmexit->u.vmx.exit_reason; 2408 vmexit->exitcode = VM_EXITCODE_BOGUS; 2409 2410 vmm_stat_incr(vcpu->vcpu, VMEXIT_COUNT, 1); 2411 SDT_PROBE3(vmm, vmx, exit, entry, vmx, vcpuid, vmexit); 2412 2413 /* 2414 * VM-entry failures during or after loading guest state. 2415 * 2416 * These VM-exits are uncommon but must be handled specially 2417 * as most VM-exit fields are not populated as usual. 2418 */ 2419 if (__predict_false(reason == EXIT_REASON_MCE_DURING_ENTRY)) { 2420 VMX_CTR0(vcpu, "Handling MCE during VM-entry"); 2421 __asm __volatile("int $18"); 2422 return (1); 2423 } 2424 2425 /* 2426 * VM exits that can be triggered during event delivery need to 2427 * be handled specially by re-injecting the event if the IDT 2428 * vectoring information field's valid bit is set. 2429 * 2430 * See "Information for VM Exits During Event Delivery" in Intel SDM 2431 * for details. 2432 */ 2433 idtvec_info = vmcs_idt_vectoring_info(); 2434 if (idtvec_info & VMCS_IDT_VEC_VALID) { 2435 idtvec_info &= ~(1 << 12); /* clear undefined bit */ 2436 exitintinfo = idtvec_info; 2437 if (idtvec_info & VMCS_IDT_VEC_ERRCODE_VALID) { 2438 idtvec_err = vmcs_idt_vectoring_err(); 2439 exitintinfo |= (uint64_t)idtvec_err << 32; 2440 } 2441 error = vm_exit_intinfo(vcpu->vcpu, exitintinfo); 2442 KASSERT(error == 0, ("%s: vm_set_intinfo error %d", 2443 __func__, error)); 2444 2445 /* 2446 * If 'virtual NMIs' are being used and the VM-exit 2447 * happened while injecting an NMI during the previous 2448 * VM-entry, then clear "blocking by NMI" in the 2449 * Guest Interruptibility-State so the NMI can be 2450 * reinjected on the subsequent VM-entry. 2451 * 2452 * However, if the NMI was being delivered through a task 2453 * gate, then the new task must start execution with NMIs 2454 * blocked so don't clear NMI blocking in this case. 2455 */ 2456 intr_type = idtvec_info & VMCS_INTR_T_MASK; 2457 if (intr_type == VMCS_INTR_T_NMI) { 2458 if (reason != EXIT_REASON_TASK_SWITCH) 2459 vmx_clear_nmi_blocking(vcpu); 2460 else 2461 vmx_assert_nmi_blocking(vcpu); 2462 } 2463 2464 /* 2465 * Update VM-entry instruction length if the event being 2466 * delivered was a software interrupt or software exception. 2467 */ 2468 if (intr_type == VMCS_INTR_T_SWINTR || 2469 intr_type == VMCS_INTR_T_PRIV_SWEXCEPTION || 2470 intr_type == VMCS_INTR_T_SWEXCEPTION) { 2471 vmcs_write(VMCS_ENTRY_INST_LENGTH, vmexit->inst_length); 2472 } 2473 } 2474 2475 switch (reason) { 2476 case EXIT_REASON_TASK_SWITCH: 2477 ts = &vmexit->u.task_switch; 2478 ts->tsssel = qual & 0xffff; 2479 ts->reason = vmx_task_switch_reason(qual); 2480 ts->ext = 0; 2481 ts->errcode_valid = 0; 2482 vmx_paging_info(&ts->paging); 2483 /* 2484 * If the task switch was due to a CALL, JMP, IRET, software 2485 * interrupt (INT n) or software exception (INT3, INTO), 2486 * then the saved %rip references the instruction that caused 2487 * the task switch. The instruction length field in the VMCS 2488 * is valid in this case. 2489 * 2490 * In all other cases (e.g., NMI, hardware exception) the 2491 * saved %rip is one that would have been saved in the old TSS 2492 * had the task switch completed normally so the instruction 2493 * length field is not needed in this case and is explicitly 2494 * set to 0. 2495 */ 2496 if (ts->reason == TSR_IDT_GATE) { 2497 KASSERT(idtvec_info & VMCS_IDT_VEC_VALID, 2498 ("invalid idtvec_info %#x for IDT task switch", 2499 idtvec_info)); 2500 intr_type = idtvec_info & VMCS_INTR_T_MASK; 2501 if (intr_type != VMCS_INTR_T_SWINTR && 2502 intr_type != VMCS_INTR_T_SWEXCEPTION && 2503 intr_type != VMCS_INTR_T_PRIV_SWEXCEPTION) { 2504 /* Task switch triggered by external event */ 2505 ts->ext = 1; 2506 vmexit->inst_length = 0; 2507 if (idtvec_info & VMCS_IDT_VEC_ERRCODE_VALID) { 2508 ts->errcode_valid = 1; 2509 ts->errcode = vmcs_idt_vectoring_err(); 2510 } 2511 } 2512 } 2513 vmexit->exitcode = VM_EXITCODE_TASK_SWITCH; 2514 SDT_PROBE4(vmm, vmx, exit, taskswitch, vmx, vcpuid, vmexit, ts); 2515 VMX_CTR4(vcpu, "task switch reason %d, tss 0x%04x, " 2516 "%s errcode 0x%016lx", ts->reason, ts->tsssel, 2517 ts->ext ? "external" : "internal", 2518 ((uint64_t)ts->errcode << 32) | ts->errcode_valid); 2519 break; 2520 case EXIT_REASON_CR_ACCESS: 2521 vmm_stat_incr(vcpu->vcpu, VMEXIT_CR_ACCESS, 1); 2522 SDT_PROBE4(vmm, vmx, exit, craccess, vmx, vcpuid, vmexit, qual); 2523 switch (qual & 0xf) { 2524 case 0: 2525 handled = vmx_emulate_cr0_access(vcpu, qual); 2526 break; 2527 case 4: 2528 handled = vmx_emulate_cr4_access(vcpu, qual); 2529 break; 2530 case 8: 2531 handled = vmx_emulate_cr8_access(vmx, vcpu, qual); 2532 break; 2533 } 2534 break; 2535 case EXIT_REASON_RDMSR: 2536 vmm_stat_incr(vcpu->vcpu, VMEXIT_RDMSR, 1); 2537 retu = false; 2538 ecx = vmxctx->guest_rcx; 2539 VMX_CTR1(vcpu, "rdmsr 0x%08x", ecx); 2540 SDT_PROBE4(vmm, vmx, exit, rdmsr, vmx, vcpuid, vmexit, ecx); 2541 error = emulate_rdmsr(vcpu, ecx, &retu); 2542 if (error) { 2543 vmexit->exitcode = VM_EXITCODE_RDMSR; 2544 vmexit->u.msr.code = ecx; 2545 } else if (!retu) { 2546 handled = HANDLED; 2547 } else { 2548 /* Return to userspace with a valid exitcode */ 2549 KASSERT(vmexit->exitcode != VM_EXITCODE_BOGUS, 2550 ("emulate_rdmsr retu with bogus exitcode")); 2551 } 2552 break; 2553 case EXIT_REASON_WRMSR: 2554 vmm_stat_incr(vcpu->vcpu, VMEXIT_WRMSR, 1); 2555 retu = false; 2556 eax = vmxctx->guest_rax; 2557 ecx = vmxctx->guest_rcx; 2558 edx = vmxctx->guest_rdx; 2559 VMX_CTR2(vcpu, "wrmsr 0x%08x value 0x%016lx", 2560 ecx, (uint64_t)edx << 32 | eax); 2561 SDT_PROBE5(vmm, vmx, exit, wrmsr, vmx, vmexit, vcpuid, ecx, 2562 (uint64_t)edx << 32 | eax); 2563 error = emulate_wrmsr(vcpu, ecx, (uint64_t)edx << 32 | eax, 2564 &retu); 2565 if (error) { 2566 vmexit->exitcode = VM_EXITCODE_WRMSR; 2567 vmexit->u.msr.code = ecx; 2568 vmexit->u.msr.wval = (uint64_t)edx << 32 | eax; 2569 } else if (!retu) { 2570 handled = HANDLED; 2571 } else { 2572 /* Return to userspace with a valid exitcode */ 2573 KASSERT(vmexit->exitcode != VM_EXITCODE_BOGUS, 2574 ("emulate_wrmsr retu with bogus exitcode")); 2575 } 2576 break; 2577 case EXIT_REASON_HLT: 2578 vmm_stat_incr(vcpu->vcpu, VMEXIT_HLT, 1); 2579 SDT_PROBE3(vmm, vmx, exit, halt, vmx, vcpuid, vmexit); 2580 vmexit->exitcode = VM_EXITCODE_HLT; 2581 vmexit->u.hlt.rflags = vmcs_read(VMCS_GUEST_RFLAGS); 2582 if (virtual_interrupt_delivery) 2583 vmexit->u.hlt.intr_status = 2584 vmcs_read(VMCS_GUEST_INTR_STATUS); 2585 else 2586 vmexit->u.hlt.intr_status = 0; 2587 break; 2588 case EXIT_REASON_MTF: 2589 vmm_stat_incr(vcpu->vcpu, VMEXIT_MTRAP, 1); 2590 SDT_PROBE3(vmm, vmx, exit, mtrap, vmx, vcpuid, vmexit); 2591 vmexit->exitcode = VM_EXITCODE_MTRAP; 2592 vmexit->inst_length = 0; 2593 break; 2594 case EXIT_REASON_PAUSE: 2595 vmm_stat_incr(vcpu->vcpu, VMEXIT_PAUSE, 1); 2596 SDT_PROBE3(vmm, vmx, exit, pause, vmx, vcpuid, vmexit); 2597 vmexit->exitcode = VM_EXITCODE_PAUSE; 2598 break; 2599 case EXIT_REASON_INTR_WINDOW: 2600 vmm_stat_incr(vcpu->vcpu, VMEXIT_INTR_WINDOW, 1); 2601 SDT_PROBE3(vmm, vmx, exit, intrwindow, vmx, vcpuid, vmexit); 2602 vmx_clear_int_window_exiting(vcpu); 2603 return (1); 2604 case EXIT_REASON_EXT_INTR: 2605 /* 2606 * External interrupts serve only to cause VM exits and allow 2607 * the host interrupt handler to run. 2608 * 2609 * If this external interrupt triggers a virtual interrupt 2610 * to a VM, then that state will be recorded by the 2611 * host interrupt handler in the VM's softc. We will inject 2612 * this virtual interrupt during the subsequent VM enter. 2613 */ 2614 intr_info = vmcs_read(VMCS_EXIT_INTR_INFO); 2615 SDT_PROBE4(vmm, vmx, exit, interrupt, 2616 vmx, vcpuid, vmexit, intr_info); 2617 2618 /* 2619 * XXX: Ignore this exit if VMCS_INTR_VALID is not set. 2620 * This appears to be a bug in VMware Fusion? 2621 */ 2622 if (!(intr_info & VMCS_INTR_VALID)) 2623 return (1); 2624 KASSERT((intr_info & VMCS_INTR_VALID) != 0 && 2625 (intr_info & VMCS_INTR_T_MASK) == VMCS_INTR_T_HWINTR, 2626 ("VM exit interruption info invalid: %#x", intr_info)); 2627 vmx_trigger_hostintr(intr_info & 0xff); 2628 2629 /* 2630 * This is special. We want to treat this as an 'handled' 2631 * VM-exit but not increment the instruction pointer. 2632 */ 2633 vmm_stat_incr(vcpu->vcpu, VMEXIT_EXTINT, 1); 2634 return (1); 2635 case EXIT_REASON_NMI_WINDOW: 2636 SDT_PROBE3(vmm, vmx, exit, nmiwindow, vmx, vcpuid, vmexit); 2637 /* Exit to allow the pending virtual NMI to be injected */ 2638 if (vm_nmi_pending(vcpu->vcpu)) 2639 vmx_inject_nmi(vcpu); 2640 vmx_clear_nmi_window_exiting(vcpu); 2641 vmm_stat_incr(vcpu->vcpu, VMEXIT_NMI_WINDOW, 1); 2642 return (1); 2643 case EXIT_REASON_INOUT: 2644 vmm_stat_incr(vcpu->vcpu, VMEXIT_INOUT, 1); 2645 vmexit->exitcode = VM_EXITCODE_INOUT; 2646 vmexit->u.inout.bytes = (qual & 0x7) + 1; 2647 vmexit->u.inout.in = in = (qual & 0x8) ? 1 : 0; 2648 vmexit->u.inout.string = (qual & 0x10) ? 1 : 0; 2649 vmexit->u.inout.rep = (qual & 0x20) ? 1 : 0; 2650 vmexit->u.inout.port = (uint16_t)(qual >> 16); 2651 vmexit->u.inout.eax = (uint32_t)(vmxctx->guest_rax); 2652 if (vmexit->u.inout.string) { 2653 inst_info = vmcs_read(VMCS_EXIT_INSTRUCTION_INFO); 2654 vmexit->exitcode = VM_EXITCODE_INOUT_STR; 2655 vis = &vmexit->u.inout_str; 2656 vmx_paging_info(&vis->paging); 2657 vis->rflags = vmcs_read(VMCS_GUEST_RFLAGS); 2658 vis->cr0 = vmcs_read(VMCS_GUEST_CR0); 2659 vis->index = inout_str_index(vcpu, in); 2660 vis->count = inout_str_count(vcpu, vis->inout.rep); 2661 vis->addrsize = inout_str_addrsize(inst_info); 2662 inout_str_seginfo(vcpu, inst_info, in, vis); 2663 } 2664 SDT_PROBE3(vmm, vmx, exit, inout, vmx, vcpuid, vmexit); 2665 break; 2666 case EXIT_REASON_CPUID: 2667 vmm_stat_incr(vcpu->vcpu, VMEXIT_CPUID, 1); 2668 SDT_PROBE3(vmm, vmx, exit, cpuid, vmx, vcpuid, vmexit); 2669 handled = vmx_handle_cpuid(vcpu, vmxctx); 2670 break; 2671 case EXIT_REASON_EXCEPTION: 2672 vmm_stat_incr(vcpu->vcpu, VMEXIT_EXCEPTION, 1); 2673 intr_info = vmcs_read(VMCS_EXIT_INTR_INFO); 2674 KASSERT((intr_info & VMCS_INTR_VALID) != 0, 2675 ("VM exit interruption info invalid: %#x", intr_info)); 2676 2677 intr_vec = intr_info & 0xff; 2678 intr_type = intr_info & VMCS_INTR_T_MASK; 2679 2680 /* 2681 * If Virtual NMIs control is 1 and the VM-exit is due to a 2682 * fault encountered during the execution of IRET then we must 2683 * restore the state of "virtual-NMI blocking" before resuming 2684 * the guest. 2685 * 2686 * See "Resuming Guest Software after Handling an Exception". 2687 * See "Information for VM Exits Due to Vectored Events". 2688 */ 2689 if ((idtvec_info & VMCS_IDT_VEC_VALID) == 0 && 2690 (intr_vec != IDT_DF) && 2691 (intr_info & EXIT_QUAL_NMIUDTI) != 0) 2692 vmx_restore_nmi_blocking(vcpu); 2693 2694 /* 2695 * The NMI has already been handled in vmx_exit_handle_nmi(). 2696 */ 2697 if (intr_type == VMCS_INTR_T_NMI) 2698 return (1); 2699 2700 /* 2701 * Call the machine check handler by hand. Also don't reflect 2702 * the machine check back into the guest. 2703 */ 2704 if (intr_vec == IDT_MC) { 2705 VMX_CTR0(vcpu, "Vectoring to MCE handler"); 2706 __asm __volatile("int $18"); 2707 return (1); 2708 } 2709 2710 /* 2711 * If the hypervisor has requested user exits for 2712 * debug exceptions, bounce them out to userland. 2713 */ 2714 if (intr_type == VMCS_INTR_T_SWEXCEPTION && intr_vec == IDT_BP && 2715 (vcpu->cap.set & (1 << VM_CAP_BPT_EXIT))) { 2716 vmexit->exitcode = VM_EXITCODE_BPT; 2717 vmexit->u.bpt.inst_length = vmexit->inst_length; 2718 vmexit->inst_length = 0; 2719 break; 2720 } 2721 2722 if (intr_vec == IDT_PF) { 2723 error = vmxctx_setreg(vmxctx, VM_REG_GUEST_CR2, qual); 2724 KASSERT(error == 0, ("%s: vmxctx_setreg(cr2) error %d", 2725 __func__, error)); 2726 } 2727 2728 /* 2729 * Software exceptions exhibit trap-like behavior. This in 2730 * turn requires populating the VM-entry instruction length 2731 * so that the %rip in the trap frame is past the INT3/INTO 2732 * instruction. 2733 */ 2734 if (intr_type == VMCS_INTR_T_SWEXCEPTION) 2735 vmcs_write(VMCS_ENTRY_INST_LENGTH, vmexit->inst_length); 2736 2737 /* Reflect all other exceptions back into the guest */ 2738 errcode_valid = errcode = 0; 2739 if (intr_info & VMCS_INTR_DEL_ERRCODE) { 2740 errcode_valid = 1; 2741 errcode = vmcs_read(VMCS_EXIT_INTR_ERRCODE); 2742 } 2743 VMX_CTR2(vcpu, "Reflecting exception %d/%#x into " 2744 "the guest", intr_vec, errcode); 2745 SDT_PROBE5(vmm, vmx, exit, exception, 2746 vmx, vcpuid, vmexit, intr_vec, errcode); 2747 error = vm_inject_exception(vcpu->vcpu, intr_vec, 2748 errcode_valid, errcode, 0); 2749 KASSERT(error == 0, ("%s: vm_inject_exception error %d", 2750 __func__, error)); 2751 return (1); 2752 2753 case EXIT_REASON_EPT_FAULT: 2754 /* 2755 * If 'gpa' lies within the address space allocated to 2756 * memory then this must be a nested page fault otherwise 2757 * this must be an instruction that accesses MMIO space. 2758 */ 2759 gpa = vmcs_gpa(); 2760 if (vm_mem_allocated(vcpu->vcpu, gpa) || 2761 ppt_is_mmio(vmx->vm, gpa) || apic_access_fault(vcpu, gpa)) { 2762 vmexit->exitcode = VM_EXITCODE_PAGING; 2763 vmexit->inst_length = 0; 2764 vmexit->u.paging.gpa = gpa; 2765 vmexit->u.paging.fault_type = ept_fault_type(qual); 2766 vmm_stat_incr(vcpu->vcpu, VMEXIT_NESTED_FAULT, 1); 2767 SDT_PROBE5(vmm, vmx, exit, nestedfault, 2768 vmx, vcpuid, vmexit, gpa, qual); 2769 } else if (ept_emulation_fault(qual)) { 2770 vmexit_inst_emul(vmexit, gpa, vmcs_gla()); 2771 vmm_stat_incr(vcpu->vcpu, VMEXIT_INST_EMUL, 1); 2772 SDT_PROBE4(vmm, vmx, exit, mmiofault, 2773 vmx, vcpuid, vmexit, gpa); 2774 } 2775 /* 2776 * If Virtual NMIs control is 1 and the VM-exit is due to an 2777 * EPT fault during the execution of IRET then we must restore 2778 * the state of "virtual-NMI blocking" before resuming. 2779 * 2780 * See description of "NMI unblocking due to IRET" in 2781 * "Exit Qualification for EPT Violations". 2782 */ 2783 if ((idtvec_info & VMCS_IDT_VEC_VALID) == 0 && 2784 (qual & EXIT_QUAL_NMIUDTI) != 0) 2785 vmx_restore_nmi_blocking(vcpu); 2786 break; 2787 case EXIT_REASON_VIRTUALIZED_EOI: 2788 vmexit->exitcode = VM_EXITCODE_IOAPIC_EOI; 2789 vmexit->u.ioapic_eoi.vector = qual & 0xFF; 2790 SDT_PROBE3(vmm, vmx, exit, eoi, vmx, vcpuid, vmexit); 2791 vmexit->inst_length = 0; /* trap-like */ 2792 break; 2793 case EXIT_REASON_APIC_ACCESS: 2794 SDT_PROBE3(vmm, vmx, exit, apicaccess, vmx, vcpuid, vmexit); 2795 handled = vmx_handle_apic_access(vcpu, vmexit); 2796 break; 2797 case EXIT_REASON_APIC_WRITE: 2798 /* 2799 * APIC-write VM exit is trap-like so the %rip is already 2800 * pointing to the next instruction. 2801 */ 2802 vmexit->inst_length = 0; 2803 vlapic = vm_lapic(vcpu->vcpu); 2804 SDT_PROBE4(vmm, vmx, exit, apicwrite, 2805 vmx, vcpuid, vmexit, vlapic); 2806 handled = vmx_handle_apic_write(vcpu, vlapic, qual); 2807 break; 2808 case EXIT_REASON_XSETBV: 2809 SDT_PROBE3(vmm, vmx, exit, xsetbv, vmx, vcpuid, vmexit); 2810 handled = vmx_emulate_xsetbv(vmx, vcpu, vmexit); 2811 break; 2812 case EXIT_REASON_MONITOR: 2813 SDT_PROBE3(vmm, vmx, exit, monitor, vmx, vcpuid, vmexit); 2814 vmexit->exitcode = VM_EXITCODE_MONITOR; 2815 break; 2816 case EXIT_REASON_MWAIT: 2817 SDT_PROBE3(vmm, vmx, exit, mwait, vmx, vcpuid, vmexit); 2818 vmexit->exitcode = VM_EXITCODE_MWAIT; 2819 break; 2820 case EXIT_REASON_TPR: 2821 vlapic = vm_lapic(vcpu->vcpu); 2822 vlapic_sync_tpr(vlapic); 2823 vmexit->inst_length = 0; 2824 handled = HANDLED; 2825 break; 2826 case EXIT_REASON_VMCALL: 2827 case EXIT_REASON_VMCLEAR: 2828 case EXIT_REASON_VMLAUNCH: 2829 case EXIT_REASON_VMPTRLD: 2830 case EXIT_REASON_VMPTRST: 2831 case EXIT_REASON_VMREAD: 2832 case EXIT_REASON_VMRESUME: 2833 case EXIT_REASON_VMWRITE: 2834 case EXIT_REASON_VMXOFF: 2835 case EXIT_REASON_VMXON: 2836 SDT_PROBE3(vmm, vmx, exit, vminsn, vmx, vcpuid, vmexit); 2837 vmexit->exitcode = VM_EXITCODE_VMINSN; 2838 break; 2839 case EXIT_REASON_INVD: 2840 case EXIT_REASON_WBINVD: 2841 /* ignore exit */ 2842 handled = HANDLED; 2843 break; 2844 default: 2845 SDT_PROBE4(vmm, vmx, exit, unknown, 2846 vmx, vcpuid, vmexit, reason); 2847 vmm_stat_incr(vcpu->vcpu, VMEXIT_UNKNOWN, 1); 2848 break; 2849 } 2850 2851 if (handled) { 2852 /* 2853 * It is possible that control is returned to userland 2854 * even though we were able to handle the VM exit in the 2855 * kernel. 2856 * 2857 * In such a case we want to make sure that the userland 2858 * restarts guest execution at the instruction *after* 2859 * the one we just processed. Therefore we update the 2860 * guest rip in the VMCS and in 'vmexit'. 2861 */ 2862 vmexit->rip += vmexit->inst_length; 2863 vmexit->inst_length = 0; 2864 vmcs_write(VMCS_GUEST_RIP, vmexit->rip); 2865 } else { 2866 if (vmexit->exitcode == VM_EXITCODE_BOGUS) { 2867 /* 2868 * If this VM exit was not claimed by anybody then 2869 * treat it as a generic VMX exit. 2870 */ 2871 vmexit->exitcode = VM_EXITCODE_VMX; 2872 vmexit->u.vmx.status = VM_SUCCESS; 2873 vmexit->u.vmx.inst_type = 0; 2874 vmexit->u.vmx.inst_error = 0; 2875 } else { 2876 /* 2877 * The exitcode and collateral have been populated. 2878 * The VM exit will be processed further in userland. 2879 */ 2880 } 2881 } 2882 2883 SDT_PROBE4(vmm, vmx, exit, return, 2884 vmx, vcpuid, vmexit, handled); 2885 return (handled); 2886 } 2887 2888 static __inline void 2889 vmx_exit_inst_error(struct vmxctx *vmxctx, int rc, struct vm_exit *vmexit) 2890 { 2891 2892 KASSERT(vmxctx->inst_fail_status != VM_SUCCESS, 2893 ("vmx_exit_inst_error: invalid inst_fail_status %d", 2894 vmxctx->inst_fail_status)); 2895 2896 vmexit->inst_length = 0; 2897 vmexit->exitcode = VM_EXITCODE_VMX; 2898 vmexit->u.vmx.status = vmxctx->inst_fail_status; 2899 vmexit->u.vmx.inst_error = vmcs_instruction_error(); 2900 vmexit->u.vmx.exit_reason = ~0; 2901 vmexit->u.vmx.exit_qualification = ~0; 2902 2903 switch (rc) { 2904 case VMX_VMRESUME_ERROR: 2905 case VMX_VMLAUNCH_ERROR: 2906 vmexit->u.vmx.inst_type = rc; 2907 break; 2908 default: 2909 panic("vm_exit_inst_error: vmx_enter_guest returned %d", rc); 2910 } 2911 } 2912 2913 /* 2914 * If the NMI-exiting VM execution control is set to '1' then an NMI in 2915 * non-root operation causes a VM-exit. NMI blocking is in effect so it is 2916 * sufficient to simply vector to the NMI handler via a software interrupt. 2917 * However, this must be done before maskable interrupts are enabled 2918 * otherwise the "iret" issued by an interrupt handler will incorrectly 2919 * clear NMI blocking. 2920 */ 2921 static __inline void 2922 vmx_exit_handle_nmi(struct vmx_vcpu *vcpu, struct vm_exit *vmexit) 2923 { 2924 uint32_t intr_info; 2925 2926 KASSERT((read_rflags() & PSL_I) == 0, ("interrupts enabled")); 2927 2928 if (vmexit->u.vmx.exit_reason != EXIT_REASON_EXCEPTION) 2929 return; 2930 2931 intr_info = vmcs_read(VMCS_EXIT_INTR_INFO); 2932 KASSERT((intr_info & VMCS_INTR_VALID) != 0, 2933 ("VM exit interruption info invalid: %#x", intr_info)); 2934 2935 if ((intr_info & VMCS_INTR_T_MASK) == VMCS_INTR_T_NMI) { 2936 KASSERT((intr_info & 0xff) == IDT_NMI, ("VM exit due " 2937 "to NMI has invalid vector: %#x", intr_info)); 2938 VMX_CTR0(vcpu, "Vectoring to NMI handler"); 2939 __asm __volatile("int $2"); 2940 } 2941 } 2942 2943 static __inline void 2944 vmx_dr_enter_guest(struct vmxctx *vmxctx) 2945 { 2946 register_t rflags; 2947 2948 /* Save host control debug registers. */ 2949 vmxctx->host_dr7 = rdr7(); 2950 vmxctx->host_debugctl = rdmsr(MSR_DEBUGCTLMSR); 2951 2952 /* 2953 * Disable debugging in DR7 and DEBUGCTL to avoid triggering 2954 * exceptions in the host based on the guest DRx values. The 2955 * guest DR7 and DEBUGCTL are saved/restored in the VMCS. 2956 */ 2957 load_dr7(0); 2958 wrmsr(MSR_DEBUGCTLMSR, 0); 2959 2960 /* 2961 * Disable single stepping the kernel to avoid corrupting the 2962 * guest DR6. A debugger might still be able to corrupt the 2963 * guest DR6 by setting a breakpoint after this point and then 2964 * single stepping. 2965 */ 2966 rflags = read_rflags(); 2967 vmxctx->host_tf = rflags & PSL_T; 2968 write_rflags(rflags & ~PSL_T); 2969 2970 /* Save host debug registers. */ 2971 vmxctx->host_dr0 = rdr0(); 2972 vmxctx->host_dr1 = rdr1(); 2973 vmxctx->host_dr2 = rdr2(); 2974 vmxctx->host_dr3 = rdr3(); 2975 vmxctx->host_dr6 = rdr6(); 2976 2977 /* Restore guest debug registers. */ 2978 load_dr0(vmxctx->guest_dr0); 2979 load_dr1(vmxctx->guest_dr1); 2980 load_dr2(vmxctx->guest_dr2); 2981 load_dr3(vmxctx->guest_dr3); 2982 load_dr6(vmxctx->guest_dr6); 2983 } 2984 2985 static __inline void 2986 vmx_dr_leave_guest(struct vmxctx *vmxctx) 2987 { 2988 2989 /* Save guest debug registers. */ 2990 vmxctx->guest_dr0 = rdr0(); 2991 vmxctx->guest_dr1 = rdr1(); 2992 vmxctx->guest_dr2 = rdr2(); 2993 vmxctx->guest_dr3 = rdr3(); 2994 vmxctx->guest_dr6 = rdr6(); 2995 2996 /* 2997 * Restore host debug registers. Restore DR7, DEBUGCTL, and 2998 * PSL_T last. 2999 */ 3000 load_dr0(vmxctx->host_dr0); 3001 load_dr1(vmxctx->host_dr1); 3002 load_dr2(vmxctx->host_dr2); 3003 load_dr3(vmxctx->host_dr3); 3004 load_dr6(vmxctx->host_dr6); 3005 wrmsr(MSR_DEBUGCTLMSR, vmxctx->host_debugctl); 3006 load_dr7(vmxctx->host_dr7); 3007 write_rflags(read_rflags() | vmxctx->host_tf); 3008 } 3009 3010 static __inline void 3011 vmx_pmap_activate(struct vmx *vmx, pmap_t pmap) 3012 { 3013 long eptgen; 3014 int cpu; 3015 3016 cpu = curcpu; 3017 3018 CPU_SET_ATOMIC(cpu, &pmap->pm_active); 3019 smr_enter(pmap->pm_eptsmr); 3020 eptgen = atomic_load_long(&pmap->pm_eptgen); 3021 if (eptgen != vmx->eptgen[cpu]) { 3022 vmx->eptgen[cpu] = eptgen; 3023 invept(INVEPT_TYPE_SINGLE_CONTEXT, 3024 (struct invept_desc){ .eptp = vmx->eptp, ._res = 0 }); 3025 } 3026 } 3027 3028 static __inline void 3029 vmx_pmap_deactivate(struct vmx *vmx, pmap_t pmap) 3030 { 3031 smr_exit(pmap->pm_eptsmr); 3032 CPU_CLR_ATOMIC(curcpu, &pmap->pm_active); 3033 } 3034 3035 static int 3036 vmx_run(void *vcpui, register_t rip, pmap_t pmap, struct vm_eventinfo *evinfo) 3037 { 3038 int rc, handled, launched; 3039 struct vmx *vmx; 3040 struct vmx_vcpu *vcpu; 3041 struct vmxctx *vmxctx; 3042 struct vmcs *vmcs; 3043 struct vm_exit *vmexit; 3044 struct vlapic *vlapic; 3045 uint32_t exit_reason; 3046 struct region_descriptor gdtr, idtr; 3047 uint16_t ldt_sel; 3048 3049 vcpu = vcpui; 3050 vmx = vcpu->vmx; 3051 vmcs = vcpu->vmcs; 3052 vmxctx = &vcpu->ctx; 3053 vlapic = vm_lapic(vcpu->vcpu); 3054 vmexit = vm_exitinfo(vcpu->vcpu); 3055 launched = 0; 3056 3057 KASSERT(vmxctx->pmap == pmap, 3058 ("pmap %p different than ctx pmap %p", pmap, vmxctx->pmap)); 3059 3060 vmx_msr_guest_enter(vcpu); 3061 3062 VMPTRLD(vmcs); 3063 3064 /* 3065 * XXX 3066 * We do this every time because we may setup the virtual machine 3067 * from a different process than the one that actually runs it. 3068 * 3069 * If the life of a virtual machine was spent entirely in the context 3070 * of a single process we could do this once in vmx_init(). 3071 */ 3072 vmcs_write(VMCS_HOST_CR3, rcr3()); 3073 3074 vmcs_write(VMCS_GUEST_RIP, rip); 3075 vmx_set_pcpu_defaults(vmx, vcpu, pmap); 3076 do { 3077 KASSERT(vmcs_guest_rip() == rip, ("%s: vmcs guest rip mismatch " 3078 "%#lx/%#lx", __func__, vmcs_guest_rip(), rip)); 3079 3080 handled = UNHANDLED; 3081 /* 3082 * Interrupts are disabled from this point on until the 3083 * guest starts executing. This is done for the following 3084 * reasons: 3085 * 3086 * If an AST is asserted on this thread after the check below, 3087 * then the IPI_AST notification will not be lost, because it 3088 * will cause a VM exit due to external interrupt as soon as 3089 * the guest state is loaded. 3090 * 3091 * A posted interrupt after 'vmx_inject_interrupts()' will 3092 * not be "lost" because it will be held pending in the host 3093 * APIC because interrupts are disabled. The pending interrupt 3094 * will be recognized as soon as the guest state is loaded. 3095 * 3096 * The same reasoning applies to the IPI generated by 3097 * pmap_invalidate_ept(). 3098 */ 3099 disable_intr(); 3100 vmx_inject_interrupts(vcpu, vlapic, rip); 3101 3102 /* 3103 * Check for vcpu suspension after injecting events because 3104 * vmx_inject_interrupts() can suspend the vcpu due to a 3105 * triple fault. 3106 */ 3107 if (vcpu_suspended(evinfo)) { 3108 enable_intr(); 3109 vm_exit_suspended(vcpu->vcpu, rip); 3110 break; 3111 } 3112 3113 if (vcpu_rendezvous_pending(vcpu->vcpu, evinfo)) { 3114 enable_intr(); 3115 vm_exit_rendezvous(vcpu->vcpu, rip); 3116 break; 3117 } 3118 3119 if (vcpu_reqidle(evinfo)) { 3120 enable_intr(); 3121 vm_exit_reqidle(vcpu->vcpu, rip); 3122 break; 3123 } 3124 3125 if (vcpu_should_yield(vcpu->vcpu)) { 3126 enable_intr(); 3127 vm_exit_astpending(vcpu->vcpu, rip); 3128 vmx_astpending_trace(vcpu, rip); 3129 handled = HANDLED; 3130 break; 3131 } 3132 3133 if (vcpu_debugged(vcpu->vcpu)) { 3134 enable_intr(); 3135 vm_exit_debug(vcpu->vcpu, rip); 3136 break; 3137 } 3138 3139 /* 3140 * If TPR Shadowing is enabled, the TPR Threshold 3141 * must be updated right before entering the guest. 3142 */ 3143 if (tpr_shadowing && !virtual_interrupt_delivery) { 3144 if ((vcpu->cap.proc_ctls & PROCBASED_USE_TPR_SHADOW) != 0) { 3145 vmcs_write(VMCS_TPR_THRESHOLD, vlapic_get_cr8(vlapic)); 3146 } 3147 } 3148 3149 /* 3150 * VM exits restore the base address but not the 3151 * limits of GDTR and IDTR. The VMCS only stores the 3152 * base address, so VM exits set the limits to 0xffff. 3153 * Save and restore the full GDTR and IDTR to restore 3154 * the limits. 3155 * 3156 * The VMCS does not save the LDTR at all, and VM 3157 * exits clear LDTR as if a NULL selector were loaded. 3158 * The userspace hypervisor probably doesn't use a 3159 * LDT, but save and restore it to be safe. 3160 */ 3161 sgdt(&gdtr); 3162 sidt(&idtr); 3163 ldt_sel = sldt(); 3164 3165 /* 3166 * The TSC_AUX MSR must be saved/restored while interrupts 3167 * are disabled so that it is not possible for the guest 3168 * TSC_AUX MSR value to be overwritten by the resume 3169 * portion of the IPI_SUSPEND codepath. This is why the 3170 * transition of this MSR is handled separately from those 3171 * handled by vmx_msr_guest_{enter,exit}(), which are ok to 3172 * be transitioned with preemption disabled but interrupts 3173 * enabled. 3174 * 3175 * These vmx_msr_guest_{enter,exit}_tsc_aux() calls can be 3176 * anywhere in this loop so long as they happen with 3177 * interrupts disabled. This location is chosen for 3178 * simplicity. 3179 */ 3180 vmx_msr_guest_enter_tsc_aux(vmx, vcpu); 3181 3182 vmx_dr_enter_guest(vmxctx); 3183 3184 /* 3185 * Mark the EPT as active on this host CPU and invalidate 3186 * EPTP-tagged TLB entries if required. 3187 */ 3188 vmx_pmap_activate(vmx, pmap); 3189 3190 vmx_run_trace(vcpu); 3191 rc = vmx_enter_guest(vmxctx, vmx, launched); 3192 3193 vmx_pmap_deactivate(vmx, pmap); 3194 vmx_dr_leave_guest(vmxctx); 3195 vmx_msr_guest_exit_tsc_aux(vmx, vcpu); 3196 3197 bare_lgdt(&gdtr); 3198 lidt(&idtr); 3199 lldt(ldt_sel); 3200 3201 /* Collect some information for VM exit processing */ 3202 vmexit->rip = rip = vmcs_guest_rip(); 3203 vmexit->inst_length = vmexit_instruction_length(); 3204 vmexit->u.vmx.exit_reason = exit_reason = vmcs_exit_reason(); 3205 vmexit->u.vmx.exit_qualification = vmcs_exit_qualification(); 3206 3207 /* Update 'nextrip' */ 3208 vcpu->state.nextrip = rip; 3209 3210 if (rc == VMX_GUEST_VMEXIT) { 3211 vmx_exit_handle_nmi(vcpu, vmexit); 3212 enable_intr(); 3213 handled = vmx_exit_process(vmx, vcpu, vmexit); 3214 } else { 3215 enable_intr(); 3216 vmx_exit_inst_error(vmxctx, rc, vmexit); 3217 } 3218 launched = 1; 3219 vmx_exit_trace(vcpu, rip, exit_reason, handled); 3220 rip = vmexit->rip; 3221 } while (handled); 3222 3223 /* 3224 * If a VM exit has been handled then the exitcode must be BOGUS 3225 * If a VM exit is not handled then the exitcode must not be BOGUS 3226 */ 3227 if ((handled && vmexit->exitcode != VM_EXITCODE_BOGUS) || 3228 (!handled && vmexit->exitcode == VM_EXITCODE_BOGUS)) { 3229 panic("Mismatch between handled (%d) and exitcode (%d)", 3230 handled, vmexit->exitcode); 3231 } 3232 3233 VMX_CTR1(vcpu, "returning from vmx_run: exitcode %d", 3234 vmexit->exitcode); 3235 3236 VMCLEAR(vmcs); 3237 vmx_msr_guest_exit(vcpu); 3238 3239 return (0); 3240 } 3241 3242 static void 3243 vmx_vcpu_cleanup(void *vcpui) 3244 { 3245 struct vmx_vcpu *vcpu = vcpui; 3246 3247 vpid_free(vcpu->state.vpid); 3248 free(vcpu->pir_desc, M_VMX); 3249 free(vcpu->apic_page, M_VMX); 3250 free(vcpu->vmcs, M_VMX); 3251 free(vcpu, M_VMX); 3252 } 3253 3254 static void 3255 vmx_cleanup(void *vmi) 3256 { 3257 struct vmx *vmx = vmi; 3258 3259 if (virtual_interrupt_delivery) 3260 vm_unmap_mmio(vmx->vm, DEFAULT_APIC_BASE, PAGE_SIZE); 3261 3262 free(vmx->msr_bitmap, M_VMX); 3263 free(vmx, M_VMX); 3264 3265 return; 3266 } 3267 3268 static register_t * 3269 vmxctx_regptr(struct vmxctx *vmxctx, int reg) 3270 { 3271 3272 switch (reg) { 3273 case VM_REG_GUEST_RAX: 3274 return (&vmxctx->guest_rax); 3275 case VM_REG_GUEST_RBX: 3276 return (&vmxctx->guest_rbx); 3277 case VM_REG_GUEST_RCX: 3278 return (&vmxctx->guest_rcx); 3279 case VM_REG_GUEST_RDX: 3280 return (&vmxctx->guest_rdx); 3281 case VM_REG_GUEST_RSI: 3282 return (&vmxctx->guest_rsi); 3283 case VM_REG_GUEST_RDI: 3284 return (&vmxctx->guest_rdi); 3285 case VM_REG_GUEST_RBP: 3286 return (&vmxctx->guest_rbp); 3287 case VM_REG_GUEST_R8: 3288 return (&vmxctx->guest_r8); 3289 case VM_REG_GUEST_R9: 3290 return (&vmxctx->guest_r9); 3291 case VM_REG_GUEST_R10: 3292 return (&vmxctx->guest_r10); 3293 case VM_REG_GUEST_R11: 3294 return (&vmxctx->guest_r11); 3295 case VM_REG_GUEST_R12: 3296 return (&vmxctx->guest_r12); 3297 case VM_REG_GUEST_R13: 3298 return (&vmxctx->guest_r13); 3299 case VM_REG_GUEST_R14: 3300 return (&vmxctx->guest_r14); 3301 case VM_REG_GUEST_R15: 3302 return (&vmxctx->guest_r15); 3303 case VM_REG_GUEST_CR2: 3304 return (&vmxctx->guest_cr2); 3305 case VM_REG_GUEST_DR0: 3306 return (&vmxctx->guest_dr0); 3307 case VM_REG_GUEST_DR1: 3308 return (&vmxctx->guest_dr1); 3309 case VM_REG_GUEST_DR2: 3310 return (&vmxctx->guest_dr2); 3311 case VM_REG_GUEST_DR3: 3312 return (&vmxctx->guest_dr3); 3313 case VM_REG_GUEST_DR6: 3314 return (&vmxctx->guest_dr6); 3315 default: 3316 break; 3317 } 3318 return (NULL); 3319 } 3320 3321 static int 3322 vmxctx_getreg(struct vmxctx *vmxctx, int reg, uint64_t *retval) 3323 { 3324 register_t *regp; 3325 3326 if ((regp = vmxctx_regptr(vmxctx, reg)) != NULL) { 3327 *retval = *regp; 3328 return (0); 3329 } else 3330 return (EINVAL); 3331 } 3332 3333 static int 3334 vmxctx_setreg(struct vmxctx *vmxctx, int reg, uint64_t val) 3335 { 3336 register_t *regp; 3337 3338 if ((regp = vmxctx_regptr(vmxctx, reg)) != NULL) { 3339 *regp = val; 3340 return (0); 3341 } else 3342 return (EINVAL); 3343 } 3344 3345 static int 3346 vmx_get_intr_shadow(struct vmx_vcpu *vcpu, int running, uint64_t *retval) 3347 { 3348 uint64_t gi; 3349 int error; 3350 3351 error = vmcs_getreg(vcpu->vmcs, running, 3352 VMCS_IDENT(VMCS_GUEST_INTERRUPTIBILITY), &gi); 3353 *retval = (gi & HWINTR_BLOCKING) ? 1 : 0; 3354 return (error); 3355 } 3356 3357 static int 3358 vmx_modify_intr_shadow(struct vmx_vcpu *vcpu, int running, uint64_t val) 3359 { 3360 struct vmcs *vmcs; 3361 uint64_t gi; 3362 int error, ident; 3363 3364 /* 3365 * Forcing the vcpu into an interrupt shadow is not supported. 3366 */ 3367 if (val) { 3368 error = EINVAL; 3369 goto done; 3370 } 3371 3372 vmcs = vcpu->vmcs; 3373 ident = VMCS_IDENT(VMCS_GUEST_INTERRUPTIBILITY); 3374 error = vmcs_getreg(vmcs, running, ident, &gi); 3375 if (error == 0) { 3376 gi &= ~HWINTR_BLOCKING; 3377 error = vmcs_setreg(vmcs, running, ident, gi); 3378 } 3379 done: 3380 VMX_CTR2(vcpu, "Setting intr_shadow to %#lx %s", val, 3381 error ? "failed" : "succeeded"); 3382 return (error); 3383 } 3384 3385 static int 3386 vmx_shadow_reg(int reg) 3387 { 3388 int shreg; 3389 3390 shreg = -1; 3391 3392 switch (reg) { 3393 case VM_REG_GUEST_CR0: 3394 shreg = VMCS_CR0_SHADOW; 3395 break; 3396 case VM_REG_GUEST_CR4: 3397 shreg = VMCS_CR4_SHADOW; 3398 break; 3399 default: 3400 break; 3401 } 3402 3403 return (shreg); 3404 } 3405 3406 static int 3407 vmx_getreg(void *vcpui, int reg, uint64_t *retval) 3408 { 3409 int running, hostcpu; 3410 struct vmx_vcpu *vcpu = vcpui; 3411 struct vmx *vmx = vcpu->vmx; 3412 3413 running = vcpu_is_running(vcpu->vcpu, &hostcpu); 3414 if (running && hostcpu != curcpu) 3415 panic("vmx_getreg: %s%d is running", vm_name(vmx->vm), 3416 vcpu->vcpuid); 3417 3418 switch (reg) { 3419 case VM_REG_GUEST_INTR_SHADOW: 3420 return (vmx_get_intr_shadow(vcpu, running, retval)); 3421 case VM_REG_GUEST_KGS_BASE: 3422 *retval = vcpu->guest_msrs[IDX_MSR_KGSBASE]; 3423 return (0); 3424 case VM_REG_GUEST_TPR: 3425 *retval = vlapic_get_cr8(vm_lapic(vcpu->vcpu)); 3426 return (0); 3427 } 3428 3429 if (vmxctx_getreg(&vcpu->ctx, reg, retval) == 0) 3430 return (0); 3431 3432 return (vmcs_getreg(vcpu->vmcs, running, reg, retval)); 3433 } 3434 3435 static int 3436 vmx_setreg(void *vcpui, int reg, uint64_t val) 3437 { 3438 int error, hostcpu, running, shadow; 3439 uint64_t ctls; 3440 pmap_t pmap; 3441 struct vmx_vcpu *vcpu = vcpui; 3442 struct vmx *vmx = vcpu->vmx; 3443 3444 running = vcpu_is_running(vcpu->vcpu, &hostcpu); 3445 if (running && hostcpu != curcpu) 3446 panic("vmx_setreg: %s%d is running", vm_name(vmx->vm), 3447 vcpu->vcpuid); 3448 3449 if (reg == VM_REG_GUEST_INTR_SHADOW) 3450 return (vmx_modify_intr_shadow(vcpu, running, val)); 3451 3452 if (vmxctx_setreg(&vcpu->ctx, reg, val) == 0) 3453 return (0); 3454 3455 /* Do not permit user write access to VMCS fields by offset. */ 3456 if (reg < 0) 3457 return (EINVAL); 3458 3459 error = vmcs_setreg(vcpu->vmcs, running, reg, val); 3460 3461 if (error == 0) { 3462 /* 3463 * If the "load EFER" VM-entry control is 1 then the 3464 * value of EFER.LMA must be identical to "IA-32e mode guest" 3465 * bit in the VM-entry control. 3466 */ 3467 if ((entry_ctls & VM_ENTRY_LOAD_EFER) != 0 && 3468 (reg == VM_REG_GUEST_EFER)) { 3469 vmcs_getreg(vcpu->vmcs, running, 3470 VMCS_IDENT(VMCS_ENTRY_CTLS), &ctls); 3471 if (val & EFER_LMA) 3472 ctls |= VM_ENTRY_GUEST_LMA; 3473 else 3474 ctls &= ~VM_ENTRY_GUEST_LMA; 3475 vmcs_setreg(vcpu->vmcs, running, 3476 VMCS_IDENT(VMCS_ENTRY_CTLS), ctls); 3477 } 3478 3479 shadow = vmx_shadow_reg(reg); 3480 if (shadow > 0) { 3481 /* 3482 * Store the unmodified value in the shadow 3483 */ 3484 error = vmcs_setreg(vcpu->vmcs, running, 3485 VMCS_IDENT(shadow), val); 3486 } 3487 3488 if (reg == VM_REG_GUEST_CR3) { 3489 /* 3490 * Invalidate the guest vcpu's TLB mappings to emulate 3491 * the behavior of updating %cr3. 3492 * 3493 * XXX the processor retains global mappings when %cr3 3494 * is updated but vmx_invvpid() does not. 3495 */ 3496 pmap = vcpu->ctx.pmap; 3497 vmx_invvpid(vmx, vcpu, pmap, running); 3498 } 3499 } 3500 3501 return (error); 3502 } 3503 3504 static int 3505 vmx_getdesc(void *vcpui, int reg, struct seg_desc *desc) 3506 { 3507 int hostcpu, running; 3508 struct vmx_vcpu *vcpu = vcpui; 3509 struct vmx *vmx = vcpu->vmx; 3510 3511 running = vcpu_is_running(vcpu->vcpu, &hostcpu); 3512 if (running && hostcpu != curcpu) 3513 panic("vmx_getdesc: %s%d is running", vm_name(vmx->vm), 3514 vcpu->vcpuid); 3515 3516 return (vmcs_getdesc(vcpu->vmcs, running, reg, desc)); 3517 } 3518 3519 static int 3520 vmx_setdesc(void *vcpui, int reg, struct seg_desc *desc) 3521 { 3522 int hostcpu, running; 3523 struct vmx_vcpu *vcpu = vcpui; 3524 struct vmx *vmx = vcpu->vmx; 3525 3526 running = vcpu_is_running(vcpu->vcpu, &hostcpu); 3527 if (running && hostcpu != curcpu) 3528 panic("vmx_setdesc: %s%d is running", vm_name(vmx->vm), 3529 vcpu->vcpuid); 3530 3531 return (vmcs_setdesc(vcpu->vmcs, running, reg, desc)); 3532 } 3533 3534 static int 3535 vmx_getcap(void *vcpui, int type, int *retval) 3536 { 3537 struct vmx_vcpu *vcpu = vcpui; 3538 int vcap; 3539 int ret; 3540 3541 ret = ENOENT; 3542 3543 vcap = vcpu->cap.set; 3544 3545 switch (type) { 3546 case VM_CAP_HALT_EXIT: 3547 if (cap_halt_exit) 3548 ret = 0; 3549 break; 3550 case VM_CAP_PAUSE_EXIT: 3551 if (cap_pause_exit) 3552 ret = 0; 3553 break; 3554 case VM_CAP_MTRAP_EXIT: 3555 if (cap_monitor_trap) 3556 ret = 0; 3557 break; 3558 case VM_CAP_RDPID: 3559 if (cap_rdpid) 3560 ret = 0; 3561 break; 3562 case VM_CAP_RDTSCP: 3563 if (cap_rdtscp) 3564 ret = 0; 3565 break; 3566 case VM_CAP_UNRESTRICTED_GUEST: 3567 if (cap_unrestricted_guest) 3568 ret = 0; 3569 break; 3570 case VM_CAP_ENABLE_INVPCID: 3571 if (cap_invpcid) 3572 ret = 0; 3573 break; 3574 case VM_CAP_BPT_EXIT: 3575 case VM_CAP_IPI_EXIT: 3576 ret = 0; 3577 break; 3578 default: 3579 break; 3580 } 3581 3582 if (ret == 0) 3583 *retval = (vcap & (1 << type)) ? 1 : 0; 3584 3585 return (ret); 3586 } 3587 3588 static int 3589 vmx_setcap(void *vcpui, int type, int val) 3590 { 3591 struct vmx_vcpu *vcpu = vcpui; 3592 struct vmcs *vmcs = vcpu->vmcs; 3593 struct vlapic *vlapic; 3594 uint32_t baseval; 3595 uint32_t *pptr; 3596 int error; 3597 int flag; 3598 int reg; 3599 int retval; 3600 3601 retval = ENOENT; 3602 pptr = NULL; 3603 3604 switch (type) { 3605 case VM_CAP_HALT_EXIT: 3606 if (cap_halt_exit) { 3607 retval = 0; 3608 pptr = &vcpu->cap.proc_ctls; 3609 baseval = *pptr; 3610 flag = PROCBASED_HLT_EXITING; 3611 reg = VMCS_PRI_PROC_BASED_CTLS; 3612 } 3613 break; 3614 case VM_CAP_MTRAP_EXIT: 3615 if (cap_monitor_trap) { 3616 retval = 0; 3617 pptr = &vcpu->cap.proc_ctls; 3618 baseval = *pptr; 3619 flag = PROCBASED_MTF; 3620 reg = VMCS_PRI_PROC_BASED_CTLS; 3621 } 3622 break; 3623 case VM_CAP_PAUSE_EXIT: 3624 if (cap_pause_exit) { 3625 retval = 0; 3626 pptr = &vcpu->cap.proc_ctls; 3627 baseval = *pptr; 3628 flag = PROCBASED_PAUSE_EXITING; 3629 reg = VMCS_PRI_PROC_BASED_CTLS; 3630 } 3631 break; 3632 case VM_CAP_RDPID: 3633 case VM_CAP_RDTSCP: 3634 if (cap_rdpid || cap_rdtscp) 3635 /* 3636 * Choose not to support enabling/disabling 3637 * RDPID/RDTSCP via libvmmapi since, as per the 3638 * discussion in vmx_modinit(), RDPID/RDTSCP are 3639 * either always enabled or always disabled. 3640 */ 3641 error = EOPNOTSUPP; 3642 break; 3643 case VM_CAP_UNRESTRICTED_GUEST: 3644 if (cap_unrestricted_guest) { 3645 retval = 0; 3646 pptr = &vcpu->cap.proc_ctls2; 3647 baseval = *pptr; 3648 flag = PROCBASED2_UNRESTRICTED_GUEST; 3649 reg = VMCS_SEC_PROC_BASED_CTLS; 3650 } 3651 break; 3652 case VM_CAP_ENABLE_INVPCID: 3653 if (cap_invpcid) { 3654 retval = 0; 3655 pptr = &vcpu->cap.proc_ctls2; 3656 baseval = *pptr; 3657 flag = PROCBASED2_ENABLE_INVPCID; 3658 reg = VMCS_SEC_PROC_BASED_CTLS; 3659 } 3660 break; 3661 case VM_CAP_BPT_EXIT: 3662 retval = 0; 3663 3664 /* Don't change the bitmap if we are tracing all exceptions. */ 3665 if (vcpu->cap.exc_bitmap != 0xffffffff) { 3666 pptr = &vcpu->cap.exc_bitmap; 3667 baseval = *pptr; 3668 flag = (1 << IDT_BP); 3669 reg = VMCS_EXCEPTION_BITMAP; 3670 } 3671 break; 3672 case VM_CAP_IPI_EXIT: 3673 retval = 0; 3674 3675 vlapic = vm_lapic(vcpu->vcpu); 3676 vlapic->ipi_exit = val; 3677 break; 3678 case VM_CAP_MASK_HWINTR: 3679 retval = 0; 3680 break; 3681 default: 3682 break; 3683 } 3684 3685 if (retval) 3686 return (retval); 3687 3688 if (pptr != NULL) { 3689 if (val) { 3690 baseval |= flag; 3691 } else { 3692 baseval &= ~flag; 3693 } 3694 VMPTRLD(vmcs); 3695 error = vmwrite(reg, baseval); 3696 VMCLEAR(vmcs); 3697 3698 if (error) 3699 return (error); 3700 3701 /* 3702 * Update optional stored flags, and record 3703 * setting 3704 */ 3705 *pptr = baseval; 3706 } 3707 3708 if (val) { 3709 vcpu->cap.set |= (1 << type); 3710 } else { 3711 vcpu->cap.set &= ~(1 << type); 3712 } 3713 3714 return (0); 3715 } 3716 3717 static struct vmspace * 3718 vmx_vmspace_alloc(vm_offset_t min, vm_offset_t max) 3719 { 3720 return (ept_vmspace_alloc(min, max)); 3721 } 3722 3723 static void 3724 vmx_vmspace_free(struct vmspace *vmspace) 3725 { 3726 ept_vmspace_free(vmspace); 3727 } 3728 3729 struct vlapic_vtx { 3730 struct vlapic vlapic; 3731 struct pir_desc *pir_desc; 3732 struct vmx_vcpu *vcpu; 3733 u_int pending_prio; 3734 }; 3735 3736 #define VPR_PRIO_BIT(vpr) (1 << ((vpr) >> 4)) 3737 3738 #define VMX_CTR_PIR(vlapic, pir_desc, notify, vector, level, msg) \ 3739 do { \ 3740 VLAPIC_CTR2(vlapic, msg " assert %s-triggered vector %d", \ 3741 level ? "level" : "edge", vector); \ 3742 VLAPIC_CTR1(vlapic, msg " pir0 0x%016lx", pir_desc->pir[0]); \ 3743 VLAPIC_CTR1(vlapic, msg " pir1 0x%016lx", pir_desc->pir[1]); \ 3744 VLAPIC_CTR1(vlapic, msg " pir2 0x%016lx", pir_desc->pir[2]); \ 3745 VLAPIC_CTR1(vlapic, msg " pir3 0x%016lx", pir_desc->pir[3]); \ 3746 VLAPIC_CTR1(vlapic, msg " notify: %s", notify ? "yes" : "no"); \ 3747 } while (0) 3748 3749 /* 3750 * vlapic->ops handlers that utilize the APICv hardware assist described in 3751 * Chapter 29 of the Intel SDM. 3752 */ 3753 static int 3754 vmx_set_intr_ready(struct vlapic *vlapic, int vector, bool level) 3755 { 3756 struct vlapic_vtx *vlapic_vtx; 3757 struct pir_desc *pir_desc; 3758 uint64_t mask; 3759 int idx, notify = 0; 3760 3761 vlapic_vtx = (struct vlapic_vtx *)vlapic; 3762 pir_desc = vlapic_vtx->pir_desc; 3763 3764 /* 3765 * Keep track of interrupt requests in the PIR descriptor. This is 3766 * because the virtual APIC page pointed to by the VMCS cannot be 3767 * modified if the vcpu is running. 3768 */ 3769 idx = vector / 64; 3770 mask = 1UL << (vector % 64); 3771 atomic_set_long(&pir_desc->pir[idx], mask); 3772 3773 /* 3774 * A notification is required whenever the 'pending' bit makes a 3775 * transition from 0->1. 3776 * 3777 * Even if the 'pending' bit is already asserted, notification about 3778 * the incoming interrupt may still be necessary. For example, if a 3779 * vCPU is HLTed with a high PPR, a low priority interrupt would cause 3780 * the 0->1 'pending' transition with a notification, but the vCPU 3781 * would ignore the interrupt for the time being. The same vCPU would 3782 * need to then be notified if a high-priority interrupt arrived which 3783 * satisfied the PPR. 3784 * 3785 * The priorities of interrupts injected while 'pending' is asserted 3786 * are tracked in a custom bitfield 'pending_prio'. Should the 3787 * to-be-injected interrupt exceed the priorities already present, the 3788 * notification is sent. The priorities recorded in 'pending_prio' are 3789 * cleared whenever the 'pending' bit makes another 0->1 transition. 3790 */ 3791 if (atomic_cmpset_long(&pir_desc->pending, 0, 1) != 0) { 3792 notify = 1; 3793 vlapic_vtx->pending_prio = 0; 3794 } else { 3795 const u_int old_prio = vlapic_vtx->pending_prio; 3796 const u_int prio_bit = VPR_PRIO_BIT(vector & APIC_TPR_INT); 3797 3798 if ((old_prio & prio_bit) == 0 && prio_bit > old_prio) { 3799 atomic_set_int(&vlapic_vtx->pending_prio, prio_bit); 3800 notify = 1; 3801 } 3802 } 3803 3804 VMX_CTR_PIR(vlapic, pir_desc, notify, vector, level, 3805 "vmx_set_intr_ready"); 3806 return (notify); 3807 } 3808 3809 static int 3810 vmx_pending_intr(struct vlapic *vlapic, int *vecptr) 3811 { 3812 struct vlapic_vtx *vlapic_vtx; 3813 struct pir_desc *pir_desc; 3814 struct LAPIC *lapic; 3815 uint64_t pending, pirval; 3816 uint8_t ppr, vpr, rvi; 3817 struct vm_exit *vmexit; 3818 int i; 3819 3820 /* 3821 * This function is only expected to be called from the 'HLT' exit 3822 * handler which does not care about the vector that is pending. 3823 */ 3824 KASSERT(vecptr == NULL, ("vmx_pending_intr: vecptr must be NULL")); 3825 3826 vlapic_vtx = (struct vlapic_vtx *)vlapic; 3827 pir_desc = vlapic_vtx->pir_desc; 3828 lapic = vlapic->apic_page; 3829 3830 /* 3831 * While a virtual interrupt may have already been 3832 * processed the actual delivery maybe pending the 3833 * interruptibility of the guest. Recognize a pending 3834 * interrupt by reevaluating virtual interrupts 3835 * following Section 30.2.1 in the Intel SDM Volume 3. 3836 */ 3837 vmexit = vm_exitinfo(vlapic->vcpu); 3838 KASSERT(vmexit->exitcode == VM_EXITCODE_HLT, 3839 ("vmx_pending_intr: exitcode not 'HLT'")); 3840 rvi = vmexit->u.hlt.intr_status & APIC_TPR_INT; 3841 ppr = lapic->ppr & APIC_TPR_INT; 3842 if (rvi > ppr) 3843 return (1); 3844 3845 pending = atomic_load_acq_long(&pir_desc->pending); 3846 if (!pending) 3847 return (0); 3848 3849 /* 3850 * If there is an interrupt pending then it will be recognized only 3851 * if its priority is greater than the processor priority. 3852 * 3853 * Special case: if the processor priority is zero then any pending 3854 * interrupt will be recognized. 3855 */ 3856 if (ppr == 0) 3857 return (1); 3858 3859 VLAPIC_CTR1(vlapic, "HLT with non-zero PPR %d", lapic->ppr); 3860 3861 vpr = 0; 3862 for (i = 3; i >= 0; i--) { 3863 pirval = pir_desc->pir[i]; 3864 if (pirval != 0) { 3865 vpr = (i * 64 + flsl(pirval) - 1) & APIC_TPR_INT; 3866 break; 3867 } 3868 } 3869 3870 /* 3871 * If the highest-priority pending interrupt falls short of the 3872 * processor priority of this vCPU, ensure that 'pending_prio' does not 3873 * have any stale bits which would preclude a higher-priority interrupt 3874 * from incurring a notification later. 3875 */ 3876 if (vpr <= ppr) { 3877 const u_int prio_bit = VPR_PRIO_BIT(vpr); 3878 const u_int old = vlapic_vtx->pending_prio; 3879 3880 if (old > prio_bit && (old & prio_bit) == 0) { 3881 vlapic_vtx->pending_prio = prio_bit; 3882 } 3883 return (0); 3884 } 3885 return (1); 3886 } 3887 3888 static void 3889 vmx_intr_accepted(struct vlapic *vlapic, int vector) 3890 { 3891 3892 panic("vmx_intr_accepted: not expected to be called"); 3893 } 3894 3895 static void 3896 vmx_set_tmr(struct vlapic *vlapic, int vector, bool level) 3897 { 3898 struct vlapic_vtx *vlapic_vtx; 3899 struct vmcs *vmcs; 3900 uint64_t mask, val; 3901 3902 KASSERT(vector >= 0 && vector <= 255, ("invalid vector %d", vector)); 3903 KASSERT(!vcpu_is_running(vlapic->vcpu, NULL), 3904 ("vmx_set_tmr: vcpu cannot be running")); 3905 3906 vlapic_vtx = (struct vlapic_vtx *)vlapic; 3907 vmcs = vlapic_vtx->vcpu->vmcs; 3908 mask = 1UL << (vector % 64); 3909 3910 VMPTRLD(vmcs); 3911 val = vmcs_read(VMCS_EOI_EXIT(vector)); 3912 if (level) 3913 val |= mask; 3914 else 3915 val &= ~mask; 3916 vmcs_write(VMCS_EOI_EXIT(vector), val); 3917 VMCLEAR(vmcs); 3918 } 3919 3920 static void 3921 vmx_enable_x2apic_mode_ts(struct vlapic *vlapic) 3922 { 3923 struct vlapic_vtx *vlapic_vtx; 3924 struct vmx_vcpu *vcpu; 3925 struct vmcs *vmcs; 3926 uint32_t proc_ctls; 3927 3928 vlapic_vtx = (struct vlapic_vtx *)vlapic; 3929 vcpu = vlapic_vtx->vcpu; 3930 vmcs = vcpu->vmcs; 3931 3932 proc_ctls = vcpu->cap.proc_ctls; 3933 proc_ctls &= ~PROCBASED_USE_TPR_SHADOW; 3934 proc_ctls |= PROCBASED_CR8_LOAD_EXITING; 3935 proc_ctls |= PROCBASED_CR8_STORE_EXITING; 3936 vcpu->cap.proc_ctls = proc_ctls; 3937 3938 VMPTRLD(vmcs); 3939 vmcs_write(VMCS_PRI_PROC_BASED_CTLS, proc_ctls); 3940 VMCLEAR(vmcs); 3941 } 3942 3943 static void 3944 vmx_enable_x2apic_mode_vid(struct vlapic *vlapic) 3945 { 3946 struct vlapic_vtx *vlapic_vtx; 3947 struct vmx *vmx; 3948 struct vmx_vcpu *vcpu; 3949 struct vmcs *vmcs; 3950 uint32_t proc_ctls2; 3951 int error __diagused; 3952 3953 vlapic_vtx = (struct vlapic_vtx *)vlapic; 3954 vcpu = vlapic_vtx->vcpu; 3955 vmx = vcpu->vmx; 3956 vmcs = vcpu->vmcs; 3957 3958 proc_ctls2 = vcpu->cap.proc_ctls2; 3959 KASSERT((proc_ctls2 & PROCBASED2_VIRTUALIZE_APIC_ACCESSES) != 0, 3960 ("%s: invalid proc_ctls2 %#x", __func__, proc_ctls2)); 3961 3962 proc_ctls2 &= ~PROCBASED2_VIRTUALIZE_APIC_ACCESSES; 3963 proc_ctls2 |= PROCBASED2_VIRTUALIZE_X2APIC_MODE; 3964 vcpu->cap.proc_ctls2 = proc_ctls2; 3965 3966 VMPTRLD(vmcs); 3967 vmcs_write(VMCS_SEC_PROC_BASED_CTLS, proc_ctls2); 3968 VMCLEAR(vmcs); 3969 3970 if (vlapic->vcpuid == 0) { 3971 /* 3972 * The nested page table mappings are shared by all vcpus 3973 * so unmap the APIC access page just once. 3974 */ 3975 error = vm_unmap_mmio(vmx->vm, DEFAULT_APIC_BASE, PAGE_SIZE); 3976 KASSERT(error == 0, ("%s: vm_unmap_mmio error %d", 3977 __func__, error)); 3978 3979 /* 3980 * The MSR bitmap is shared by all vcpus so modify it only 3981 * once in the context of vcpu 0. 3982 */ 3983 error = vmx_allow_x2apic_msrs(vmx); 3984 KASSERT(error == 0, ("%s: vmx_allow_x2apic_msrs error %d", 3985 __func__, error)); 3986 } 3987 } 3988 3989 static void 3990 vmx_post_intr(struct vlapic *vlapic, int hostcpu) 3991 { 3992 3993 ipi_cpu(hostcpu, pirvec); 3994 } 3995 3996 /* 3997 * Transfer the pending interrupts in the PIR descriptor to the IRR 3998 * in the virtual APIC page. 3999 */ 4000 static void 4001 vmx_inject_pir(struct vlapic *vlapic) 4002 { 4003 struct vlapic_vtx *vlapic_vtx; 4004 struct pir_desc *pir_desc; 4005 struct LAPIC *lapic; 4006 uint64_t val, pirval; 4007 int rvi, pirbase = -1; 4008 uint16_t intr_status_old, intr_status_new; 4009 4010 vlapic_vtx = (struct vlapic_vtx *)vlapic; 4011 pir_desc = vlapic_vtx->pir_desc; 4012 if (atomic_cmpset_long(&pir_desc->pending, 1, 0) == 0) { 4013 VLAPIC_CTR0(vlapic, "vmx_inject_pir: " 4014 "no posted interrupt pending"); 4015 return; 4016 } 4017 4018 pirval = 0; 4019 pirbase = -1; 4020 lapic = vlapic->apic_page; 4021 4022 val = atomic_readandclear_long(&pir_desc->pir[0]); 4023 if (val != 0) { 4024 lapic->irr0 |= val; 4025 lapic->irr1 |= val >> 32; 4026 pirbase = 0; 4027 pirval = val; 4028 } 4029 4030 val = atomic_readandclear_long(&pir_desc->pir[1]); 4031 if (val != 0) { 4032 lapic->irr2 |= val; 4033 lapic->irr3 |= val >> 32; 4034 pirbase = 64; 4035 pirval = val; 4036 } 4037 4038 val = atomic_readandclear_long(&pir_desc->pir[2]); 4039 if (val != 0) { 4040 lapic->irr4 |= val; 4041 lapic->irr5 |= val >> 32; 4042 pirbase = 128; 4043 pirval = val; 4044 } 4045 4046 val = atomic_readandclear_long(&pir_desc->pir[3]); 4047 if (val != 0) { 4048 lapic->irr6 |= val; 4049 lapic->irr7 |= val >> 32; 4050 pirbase = 192; 4051 pirval = val; 4052 } 4053 4054 VLAPIC_CTR_IRR(vlapic, "vmx_inject_pir"); 4055 4056 /* 4057 * Update RVI so the processor can evaluate pending virtual 4058 * interrupts on VM-entry. 4059 * 4060 * It is possible for pirval to be 0 here, even though the 4061 * pending bit has been set. The scenario is: 4062 * CPU-Y is sending a posted interrupt to CPU-X, which 4063 * is running a guest and processing posted interrupts in h/w. 4064 * CPU-X will eventually exit and the state seen in s/w is 4065 * the pending bit set, but no PIR bits set. 4066 * 4067 * CPU-X CPU-Y 4068 * (vm running) (host running) 4069 * rx posted interrupt 4070 * CLEAR pending bit 4071 * SET PIR bit 4072 * READ/CLEAR PIR bits 4073 * SET pending bit 4074 * (vm exit) 4075 * pending bit set, PIR 0 4076 */ 4077 if (pirval != 0) { 4078 rvi = pirbase + flsl(pirval) - 1; 4079 intr_status_old = vmcs_read(VMCS_GUEST_INTR_STATUS); 4080 intr_status_new = (intr_status_old & 0xFF00) | rvi; 4081 if (intr_status_new > intr_status_old) { 4082 vmcs_write(VMCS_GUEST_INTR_STATUS, intr_status_new); 4083 VLAPIC_CTR2(vlapic, "vmx_inject_pir: " 4084 "guest_intr_status changed from 0x%04x to 0x%04x", 4085 intr_status_old, intr_status_new); 4086 } 4087 } 4088 } 4089 4090 static struct vlapic * 4091 vmx_vlapic_init(void *vcpui) 4092 { 4093 struct vmx *vmx; 4094 struct vmx_vcpu *vcpu; 4095 struct vlapic *vlapic; 4096 struct vlapic_vtx *vlapic_vtx; 4097 4098 vcpu = vcpui; 4099 vmx = vcpu->vmx; 4100 4101 vlapic = malloc(sizeof(struct vlapic_vtx), M_VLAPIC, M_WAITOK | M_ZERO); 4102 vlapic->vm = vmx->vm; 4103 vlapic->vcpu = vcpu->vcpu; 4104 vlapic->vcpuid = vcpu->vcpuid; 4105 vlapic->apic_page = (struct LAPIC *)vcpu->apic_page; 4106 4107 vlapic_vtx = (struct vlapic_vtx *)vlapic; 4108 vlapic_vtx->pir_desc = vcpu->pir_desc; 4109 vlapic_vtx->vcpu = vcpu; 4110 4111 if (tpr_shadowing) { 4112 vlapic->ops.enable_x2apic_mode = vmx_enable_x2apic_mode_ts; 4113 } 4114 4115 if (virtual_interrupt_delivery) { 4116 vlapic->ops.set_intr_ready = vmx_set_intr_ready; 4117 vlapic->ops.pending_intr = vmx_pending_intr; 4118 vlapic->ops.intr_accepted = vmx_intr_accepted; 4119 vlapic->ops.set_tmr = vmx_set_tmr; 4120 vlapic->ops.enable_x2apic_mode = vmx_enable_x2apic_mode_vid; 4121 } 4122 4123 if (posted_interrupts) 4124 vlapic->ops.post_intr = vmx_post_intr; 4125 4126 vlapic_init(vlapic); 4127 4128 return (vlapic); 4129 } 4130 4131 static void 4132 vmx_vlapic_cleanup(struct vlapic *vlapic) 4133 { 4134 4135 vlapic_cleanup(vlapic); 4136 free(vlapic, M_VLAPIC); 4137 } 4138 4139 #ifdef BHYVE_SNAPSHOT 4140 static int 4141 vmx_vcpu_snapshot(void *vcpui, struct vm_snapshot_meta *meta) 4142 { 4143 struct vmcs *vmcs; 4144 struct vmx *vmx; 4145 struct vmx_vcpu *vcpu; 4146 struct vmxctx *vmxctx; 4147 int err, run, hostcpu; 4148 4149 err = 0; 4150 vcpu = vcpui; 4151 vmx = vcpu->vmx; 4152 vmcs = vcpu->vmcs; 4153 4154 run = vcpu_is_running(vcpu->vcpu, &hostcpu); 4155 if (run && hostcpu != curcpu) { 4156 printf("%s: %s%d is running", __func__, vm_name(vmx->vm), 4157 vcpu->vcpuid); 4158 return (EINVAL); 4159 } 4160 4161 err += vmcs_snapshot_reg(vmcs, run, VM_REG_GUEST_CR0, meta); 4162 err += vmcs_snapshot_reg(vmcs, run, VM_REG_GUEST_CR3, meta); 4163 err += vmcs_snapshot_reg(vmcs, run, VM_REG_GUEST_CR4, meta); 4164 err += vmcs_snapshot_reg(vmcs, run, VM_REG_GUEST_DR7, meta); 4165 err += vmcs_snapshot_reg(vmcs, run, VM_REG_GUEST_RSP, meta); 4166 err += vmcs_snapshot_reg(vmcs, run, VM_REG_GUEST_RIP, meta); 4167 err += vmcs_snapshot_reg(vmcs, run, VM_REG_GUEST_RFLAGS, meta); 4168 4169 /* Guest segments */ 4170 err += vmcs_snapshot_reg(vmcs, run, VM_REG_GUEST_ES, meta); 4171 err += vmcs_snapshot_desc(vmcs, run, VM_REG_GUEST_ES, meta); 4172 4173 err += vmcs_snapshot_reg(vmcs, run, VM_REG_GUEST_CS, meta); 4174 err += vmcs_snapshot_desc(vmcs, run, VM_REG_GUEST_CS, meta); 4175 4176 err += vmcs_snapshot_reg(vmcs, run, VM_REG_GUEST_SS, meta); 4177 err += vmcs_snapshot_desc(vmcs, run, VM_REG_GUEST_SS, meta); 4178 4179 err += vmcs_snapshot_reg(vmcs, run, VM_REG_GUEST_DS, meta); 4180 err += vmcs_snapshot_desc(vmcs, run, VM_REG_GUEST_DS, meta); 4181 4182 err += vmcs_snapshot_reg(vmcs, run, VM_REG_GUEST_FS, meta); 4183 err += vmcs_snapshot_desc(vmcs, run, VM_REG_GUEST_FS, meta); 4184 4185 err += vmcs_snapshot_reg(vmcs, run, VM_REG_GUEST_GS, meta); 4186 err += vmcs_snapshot_desc(vmcs, run, VM_REG_GUEST_GS, meta); 4187 4188 err += vmcs_snapshot_reg(vmcs, run, VM_REG_GUEST_TR, meta); 4189 err += vmcs_snapshot_desc(vmcs, run, VM_REG_GUEST_TR, meta); 4190 4191 err += vmcs_snapshot_reg(vmcs, run, VM_REG_GUEST_LDTR, meta); 4192 err += vmcs_snapshot_desc(vmcs, run, VM_REG_GUEST_LDTR, meta); 4193 4194 err += vmcs_snapshot_reg(vmcs, run, VM_REG_GUEST_EFER, meta); 4195 4196 err += vmcs_snapshot_desc(vmcs, run, VM_REG_GUEST_IDTR, meta); 4197 err += vmcs_snapshot_desc(vmcs, run, VM_REG_GUEST_GDTR, meta); 4198 4199 /* Guest page tables */ 4200 err += vmcs_snapshot_reg(vmcs, run, VM_REG_GUEST_PDPTE0, meta); 4201 err += vmcs_snapshot_reg(vmcs, run, VM_REG_GUEST_PDPTE1, meta); 4202 err += vmcs_snapshot_reg(vmcs, run, VM_REG_GUEST_PDPTE2, meta); 4203 err += vmcs_snapshot_reg(vmcs, run, VM_REG_GUEST_PDPTE3, meta); 4204 4205 /* Other guest state */ 4206 err += vmcs_snapshot_any(vmcs, run, VMCS_GUEST_IA32_SYSENTER_CS, meta); 4207 err += vmcs_snapshot_any(vmcs, run, VMCS_GUEST_IA32_SYSENTER_ESP, meta); 4208 err += vmcs_snapshot_any(vmcs, run, VMCS_GUEST_IA32_SYSENTER_EIP, meta); 4209 err += vmcs_snapshot_any(vmcs, run, VMCS_GUEST_INTERRUPTIBILITY, meta); 4210 err += vmcs_snapshot_any(vmcs, run, VMCS_GUEST_ACTIVITY, meta); 4211 err += vmcs_snapshot_any(vmcs, run, VMCS_ENTRY_CTLS, meta); 4212 err += vmcs_snapshot_any(vmcs, run, VMCS_EXIT_CTLS, meta); 4213 if (err != 0) 4214 goto done; 4215 4216 SNAPSHOT_BUF_OR_LEAVE(vcpu->guest_msrs, 4217 sizeof(vcpu->guest_msrs), meta, err, done); 4218 4219 SNAPSHOT_BUF_OR_LEAVE(vcpu->pir_desc, 4220 sizeof(*vcpu->pir_desc), meta, err, done); 4221 4222 SNAPSHOT_BUF_OR_LEAVE(&vcpu->mtrr, 4223 sizeof(vcpu->mtrr), meta, err, done); 4224 4225 vmxctx = &vcpu->ctx; 4226 SNAPSHOT_VAR_OR_LEAVE(vmxctx->guest_rdi, meta, err, done); 4227 SNAPSHOT_VAR_OR_LEAVE(vmxctx->guest_rsi, meta, err, done); 4228 SNAPSHOT_VAR_OR_LEAVE(vmxctx->guest_rdx, meta, err, done); 4229 SNAPSHOT_VAR_OR_LEAVE(vmxctx->guest_rcx, meta, err, done); 4230 SNAPSHOT_VAR_OR_LEAVE(vmxctx->guest_r8, meta, err, done); 4231 SNAPSHOT_VAR_OR_LEAVE(vmxctx->guest_r9, meta, err, done); 4232 SNAPSHOT_VAR_OR_LEAVE(vmxctx->guest_rax, meta, err, done); 4233 SNAPSHOT_VAR_OR_LEAVE(vmxctx->guest_rbx, meta, err, done); 4234 SNAPSHOT_VAR_OR_LEAVE(vmxctx->guest_rbp, meta, err, done); 4235 SNAPSHOT_VAR_OR_LEAVE(vmxctx->guest_r10, meta, err, done); 4236 SNAPSHOT_VAR_OR_LEAVE(vmxctx->guest_r11, meta, err, done); 4237 SNAPSHOT_VAR_OR_LEAVE(vmxctx->guest_r12, meta, err, done); 4238 SNAPSHOT_VAR_OR_LEAVE(vmxctx->guest_r13, meta, err, done); 4239 SNAPSHOT_VAR_OR_LEAVE(vmxctx->guest_r14, meta, err, done); 4240 SNAPSHOT_VAR_OR_LEAVE(vmxctx->guest_r15, meta, err, done); 4241 SNAPSHOT_VAR_OR_LEAVE(vmxctx->guest_cr2, meta, err, done); 4242 SNAPSHOT_VAR_OR_LEAVE(vmxctx->guest_dr0, meta, err, done); 4243 SNAPSHOT_VAR_OR_LEAVE(vmxctx->guest_dr1, meta, err, done); 4244 SNAPSHOT_VAR_OR_LEAVE(vmxctx->guest_dr2, meta, err, done); 4245 SNAPSHOT_VAR_OR_LEAVE(vmxctx->guest_dr3, meta, err, done); 4246 SNAPSHOT_VAR_OR_LEAVE(vmxctx->guest_dr6, meta, err, done); 4247 4248 done: 4249 return (err); 4250 } 4251 4252 static int 4253 vmx_restore_tsc(void *vcpui, uint64_t offset) 4254 { 4255 struct vmx_vcpu *vcpu = vcpui; 4256 struct vmcs *vmcs; 4257 struct vmx *vmx; 4258 int error, running, hostcpu; 4259 4260 vmx = vcpu->vmx; 4261 vmcs = vcpu->vmcs; 4262 4263 running = vcpu_is_running(vcpu->vcpu, &hostcpu); 4264 if (running && hostcpu != curcpu) { 4265 printf("%s: %s%d is running", __func__, vm_name(vmx->vm), 4266 vcpu->vcpuid); 4267 return (EINVAL); 4268 } 4269 4270 if (!running) 4271 VMPTRLD(vmcs); 4272 4273 error = vmx_set_tsc_offset(vcpu, offset); 4274 4275 if (!running) 4276 VMCLEAR(vmcs); 4277 return (error); 4278 } 4279 #endif 4280 4281 const struct vmm_ops vmm_ops_intel = { 4282 .modinit = vmx_modinit, 4283 .modcleanup = vmx_modcleanup, 4284 .modsuspend = vmx_modsuspend, 4285 .modresume = vmx_modresume, 4286 .init = vmx_init, 4287 .run = vmx_run, 4288 .cleanup = vmx_cleanup, 4289 .vcpu_init = vmx_vcpu_init, 4290 .vcpu_cleanup = vmx_vcpu_cleanup, 4291 .getreg = vmx_getreg, 4292 .setreg = vmx_setreg, 4293 .getdesc = vmx_getdesc, 4294 .setdesc = vmx_setdesc, 4295 .getcap = vmx_getcap, 4296 .setcap = vmx_setcap, 4297 .vmspace_alloc = vmx_vmspace_alloc, 4298 .vmspace_free = vmx_vmspace_free, 4299 .vlapic_init = vmx_vlapic_init, 4300 .vlapic_cleanup = vmx_vlapic_cleanup, 4301 #ifdef BHYVE_SNAPSHOT 4302 .vcpu_snapshot = vmx_vcpu_snapshot, 4303 .restore_tsc = vmx_restore_tsc, 4304 #endif 4305 }; 4306