1221828Sgrehan/*- 2221828Sgrehan * Copyright (c) 2011 NetApp, Inc. 3221828Sgrehan * All rights reserved. 4221828Sgrehan * 5221828Sgrehan * Redistribution and use in source and binary forms, with or without 6221828Sgrehan * modification, are permitted provided that the following conditions 7221828Sgrehan * are met: 8221828Sgrehan * 1. Redistributions of source code must retain the above copyright 9221828Sgrehan * notice, this list of conditions and the following disclaimer. 10221828Sgrehan * 2. Redistributions in binary form must reproduce the above copyright 11221828Sgrehan * notice, this list of conditions and the following disclaimer in the 12221828Sgrehan * documentation and/or other materials provided with the distribution. 13221828Sgrehan * 14221828Sgrehan * THIS SOFTWARE IS PROVIDED BY NETAPP, INC ``AS IS'' AND 15221828Sgrehan * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 16221828Sgrehan * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 17221828Sgrehan * ARE DISCLAIMED. IN NO EVENT SHALL NETAPP, INC OR CONTRIBUTORS BE LIABLE 18221828Sgrehan * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 19221828Sgrehan * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 20221828Sgrehan * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 21221828Sgrehan * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 22221828Sgrehan * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 23221828Sgrehan * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 24221828Sgrehan * SUCH DAMAGE. 25221828Sgrehan * 26221828Sgrehan * $FreeBSD: stable/10/sys/amd64/vmm/vmm.c 306472 2016-09-30 01:42:29Z jhb $ 27221828Sgrehan */ 28221828Sgrehan 29221828Sgrehan#include <sys/cdefs.h> 30221828Sgrehan__FBSDID("$FreeBSD: stable/10/sys/amd64/vmm/vmm.c 306472 2016-09-30 01:42:29Z jhb $"); 31221828Sgrehan 32221828Sgrehan#include <sys/param.h> 33234695Sgrehan#include <sys/systm.h> 34221828Sgrehan#include <sys/kernel.h> 35221828Sgrehan#include <sys/module.h> 36221828Sgrehan#include <sys/sysctl.h> 37221828Sgrehan#include <sys/malloc.h> 38221828Sgrehan#include <sys/pcpu.h> 39221828Sgrehan#include <sys/lock.h> 40221828Sgrehan#include <sys/mutex.h> 41221828Sgrehan#include <sys/proc.h> 42256072Sneel#include <sys/rwlock.h> 43221828Sgrehan#include <sys/sched.h> 44221828Sgrehan#include <sys/smp.h> 45221828Sgrehan#include <sys/systm.h> 46221828Sgrehan 47221828Sgrehan#include <vm/vm.h> 48256072Sneel#include <vm/vm_object.h> 49256072Sneel#include <vm/vm_page.h> 50256072Sneel#include <vm/pmap.h> 51256072Sneel#include <vm/vm_map.h> 52256072Sneel#include <vm/vm_extern.h> 53256072Sneel#include <vm/vm_param.h> 54221828Sgrehan 55261275Sjhb#include <machine/cpu.h> 56221828Sgrehan#include <machine/vm.h> 57221828Sgrehan#include <machine/pcb.h> 58241489Sneel#include <machine/smp.h> 59262350Sjhb#include <x86/psl.h> 60221914Sjhb#include <x86/apicreg.h> 61256072Sneel#include <machine/vmparam.h> 62221828Sgrehan 63221828Sgrehan#include <machine/vmm.h> 64261088Sjhb#include <machine/vmm_dev.h> 65268976Sjhb#include <machine/vmm_instruction_emul.h> 66261088Sjhb 67268976Sjhb#include "vmm_ioport.h" 68256072Sneel#include "vmm_ktr.h" 69242275Sneel#include "vmm_host.h" 70221828Sgrehan#include "vmm_mem.h" 71221828Sgrehan#include "vmm_util.h" 72268891Sjhb#include "vatpic.h" 73268891Sjhb#include "vatpit.h" 74261088Sjhb#include "vhpet.h" 75261088Sjhb#include "vioapic.h" 76221828Sgrehan#include "vlapic.h" 77276429Sneel#include "vpmtmr.h" 78284894Sneel#include "vrtc.h" 79221828Sgrehan#include "vmm_ipi.h" 80221828Sgrehan#include "vmm_stat.h" 81242065Sneel#include "vmm_lapic.h" 82221828Sgrehan 83221828Sgrehan#include "io/ppt.h" 84221828Sgrehan#include "io/iommu.h" 85221828Sgrehan 86221828Sgrehanstruct vlapic; 87221828Sgrehan 88270071Sgrehan/* 89270071Sgrehan * Initialization: 90270071Sgrehan * (a) allocated when vcpu is created 91270071Sgrehan * (i) initialized when vcpu is created and when it is reinitialized 92270071Sgrehan * (o) initialized the first time the vcpu is created 93270071Sgrehan * (x) initialized before use 94270071Sgrehan */ 95221828Sgrehanstruct vcpu { 96270071Sgrehan struct mtx mtx; /* (o) protects 'state' and 'hostcpu' */ 97270071Sgrehan enum vcpu_state state; /* (o) vcpu state */ 98270071Sgrehan int hostcpu; /* (o) vcpu's host cpu */ 99284900Sneel int reqidle; /* (i) request vcpu to idle */ 100270071Sgrehan struct vlapic *vlapic; /* (i) APIC device model */ 101270071Sgrehan enum x2apic_state x2apic_state; /* (i) APIC mode */ 102270159Sgrehan uint64_t exitintinfo; /* (i) events pending at VM exit */ 103270071Sgrehan int nmi_pending; /* (i) NMI pending */ 104270071Sgrehan int extint_pending; /* (i) INTR pending */ 105270071Sgrehan int exception_pending; /* (i) exception pending */ 106284894Sneel int exc_vector; /* (x) exception collateral */ 107284894Sneel int exc_errcode_valid; 108284894Sneel uint32_t exc_errcode; 109270071Sgrehan struct savefpu *guestfpu; /* (a,i) guest fpu state */ 110270071Sgrehan uint64_t guest_xcr0; /* (i) guest %xcr0 register */ 111270071Sgrehan void *stats; /* (a,i) statistics */ 112270071Sgrehan struct vm_exit exitinfo; /* (x) exit reason and collateral */ 113284894Sneel uint64_t nextrip; /* (x) next instruction to execute */ 114221828Sgrehan}; 115221828Sgrehan 116270071Sgrehan#define vcpu_lock_initialized(v) mtx_initialized(&((v)->mtx)) 117242065Sneel#define vcpu_lock_init(v) mtx_init(&((v)->mtx), "vcpu lock", 0, MTX_SPIN) 118242065Sneel#define vcpu_lock(v) mtx_lock_spin(&((v)->mtx)) 119242065Sneel#define vcpu_unlock(v) mtx_unlock_spin(&((v)->mtx)) 120256072Sneel#define vcpu_assert_locked(v) mtx_assert(&((v)->mtx), MA_OWNED) 121241489Sneel 122256072Sneelstruct mem_seg { 123295124Sgrehan size_t len; 124295124Sgrehan bool sysmem; 125295124Sgrehan struct vm_object *object; 126295124Sgrehan}; 127295124Sgrehan#define VM_MAX_MEMSEGS 2 128295124Sgrehan 129295124Sgrehanstruct mem_map { 130256072Sneel vm_paddr_t gpa; 131256072Sneel size_t len; 132295124Sgrehan vm_ooffset_t segoff; 133295124Sgrehan int segid; 134295124Sgrehan int prot; 135295124Sgrehan int flags; 136256072Sneel}; 137295124Sgrehan#define VM_MAX_MEMMAPS 4 138221828Sgrehan 139270071Sgrehan/* 140270071Sgrehan * Initialization: 141270071Sgrehan * (o) initialized the first time the VM is created 142270071Sgrehan * (i) initialized when VM is created and when it is reinitialized 143270071Sgrehan * (x) initialized before use 144270071Sgrehan */ 145221828Sgrehanstruct vm { 146270071Sgrehan void *cookie; /* (i) cpu-specific data */ 147270071Sgrehan void *iommu; /* (x) iommu-specific data */ 148270071Sgrehan struct vhpet *vhpet; /* (i) virtual HPET */ 149270071Sgrehan struct vioapic *vioapic; /* (i) virtual ioapic */ 150270071Sgrehan struct vatpic *vatpic; /* (i) virtual atpic */ 151270071Sgrehan struct vatpit *vatpit; /* (i) virtual atpit */ 152276429Sneel struct vpmtmr *vpmtmr; /* (i) virtual ACPI PM timer */ 153284894Sneel struct vrtc *vrtc; /* (o) virtual RTC */ 154270071Sgrehan volatile cpuset_t active_cpus; /* (i) active vcpus */ 155270071Sgrehan int suspend; /* (i) stop VM execution */ 156270071Sgrehan volatile cpuset_t suspended_cpus; /* (i) suspended vcpus */ 157270071Sgrehan volatile cpuset_t halted_cpus; /* (x) cpus in a hard halt */ 158270071Sgrehan cpuset_t rendezvous_req_cpus; /* (x) rendezvous requested */ 159270071Sgrehan cpuset_t rendezvous_done_cpus; /* (x) rendezvous finished */ 160270071Sgrehan void *rendezvous_arg; /* (x) rendezvous func/arg */ 161270071Sgrehan vm_rendezvous_func_t rendezvous_func; 162270071Sgrehan struct mtx rendezvous_mtx; /* (o) rendezvous lock */ 163295124Sgrehan struct mem_map mem_maps[VM_MAX_MEMMAPS]; /* (i) guest address space */ 164295124Sgrehan struct mem_seg mem_segs[VM_MAX_MEMSEGS]; /* (o) guest memory regions */ 165270071Sgrehan struct vmspace *vmspace; /* (o) guest's address space */ 166270071Sgrehan char name[VM_MAX_NAMELEN]; /* (o) virtual machine name */ 167270071Sgrehan struct vcpu vcpu[VM_MAXCPU]; /* (i) guest vcpus */ 168221828Sgrehan}; 169221828Sgrehan 170249396Sneelstatic int vmm_initialized; 171249396Sneel 172221828Sgrehanstatic struct vmm_ops *ops; 173266339Sjhb#define VMM_INIT(num) (ops != NULL ? (*ops->init)(num) : 0) 174221828Sgrehan#define VMM_CLEANUP() (ops != NULL ? (*ops->cleanup)() : 0) 175261275Sjhb#define VMM_RESUME() (ops != NULL ? (*ops->resume)() : 0) 176221828Sgrehan 177256072Sneel#define VMINIT(vm, pmap) (ops != NULL ? (*ops->vminit)(vm, pmap): NULL) 178284900Sneel#define VMRUN(vmi, vcpu, rip, pmap, evinfo) \ 179284900Sneel (ops != NULL ? (*ops->vmrun)(vmi, vcpu, rip, pmap, evinfo) : ENXIO) 180221828Sgrehan#define VMCLEANUP(vmi) (ops != NULL ? (*ops->vmcleanup)(vmi) : NULL) 181256072Sneel#define VMSPACE_ALLOC(min, max) \ 182256072Sneel (ops != NULL ? (*ops->vmspace_alloc)(min, max) : NULL) 183256072Sneel#define VMSPACE_FREE(vmspace) \ 184256072Sneel (ops != NULL ? (*ops->vmspace_free)(vmspace) : ENXIO) 185221828Sgrehan#define VMGETREG(vmi, vcpu, num, retval) \ 186221828Sgrehan (ops != NULL ? (*ops->vmgetreg)(vmi, vcpu, num, retval) : ENXIO) 187221828Sgrehan#define VMSETREG(vmi, vcpu, num, val) \ 188221828Sgrehan (ops != NULL ? (*ops->vmsetreg)(vmi, vcpu, num, val) : ENXIO) 189221828Sgrehan#define VMGETDESC(vmi, vcpu, num, desc) \ 190221828Sgrehan (ops != NULL ? (*ops->vmgetdesc)(vmi, vcpu, num, desc) : ENXIO) 191221828Sgrehan#define VMSETDESC(vmi, vcpu, num, desc) \ 192221828Sgrehan (ops != NULL ? (*ops->vmsetdesc)(vmi, vcpu, num, desc) : ENXIO) 193221828Sgrehan#define VMGETCAP(vmi, vcpu, num, retval) \ 194221828Sgrehan (ops != NULL ? (*ops->vmgetcap)(vmi, vcpu, num, retval) : ENXIO) 195221828Sgrehan#define VMSETCAP(vmi, vcpu, num, val) \ 196221828Sgrehan (ops != NULL ? (*ops->vmsetcap)(vmi, vcpu, num, val) : ENXIO) 197266339Sjhb#define VLAPIC_INIT(vmi, vcpu) \ 198266339Sjhb (ops != NULL ? (*ops->vlapic_init)(vmi, vcpu) : NULL) 199266339Sjhb#define VLAPIC_CLEANUP(vmi, vlapic) \ 200266339Sjhb (ops != NULL ? (*ops->vlapic_cleanup)(vmi, vlapic) : NULL) 201221828Sgrehan 202245021Sneel#define fpu_start_emulating() load_cr0(rcr0() | CR0_TS) 203245021Sneel#define fpu_stop_emulating() clts() 204221828Sgrehan 205221828Sgrehanstatic MALLOC_DEFINE(M_VM, "vm", "vm"); 206221828Sgrehan 207221828Sgrehan/* statistics */ 208248389Sneelstatic VMM_STAT(VCPU_TOTAL_RUNTIME, "vcpu total runtime"); 209221828Sgrehan 210266339SjhbSYSCTL_NODE(_hw, OID_AUTO, vmm, CTLFLAG_RW, NULL, NULL); 211266339Sjhb 212268935Sjhb/* 213268935Sjhb * Halt the guest if all vcpus are executing a HLT instruction with 214268935Sjhb * interrupts disabled. 215268935Sjhb */ 216268935Sjhbstatic int halt_detection_enabled = 1; 217268935SjhbTUNABLE_INT("hw.vmm.halt_detection", &halt_detection_enabled); 218268935SjhbSYSCTL_INT(_hw_vmm, OID_AUTO, halt_detection, CTLFLAG_RDTUN, 219268935Sjhb &halt_detection_enabled, 0, 220268935Sjhb "Halt VM if all vcpus execute HLT with interrupts disabled"); 221268935Sjhb 222266339Sjhbstatic int vmm_ipinum; 223266339SjhbSYSCTL_INT(_hw_vmm, OID_AUTO, ipinum, CTLFLAG_RD, &vmm_ipinum, 0, 224266339Sjhb "IPI vector used for vcpu notifications"); 225266339Sjhb 226276403Sneelstatic int trace_guest_exceptions; 227276403SneelSYSCTL_INT(_hw_vmm, OID_AUTO, trace_guest_exceptions, CTLFLAG_RDTUN, 228276403Sneel &trace_guest_exceptions, 0, 229276403Sneel "Trap into hypervisor on all guest exceptions and reflect them back"); 230276403Sneel 231295124Sgrehanstatic void vm_free_memmap(struct vm *vm, int ident); 232295124Sgrehanstatic bool sysmem_mapping(struct vm *vm, struct mem_map *mm); 233284900Sneelstatic void vcpu_notify_event_locked(struct vcpu *vcpu, bool lapic_intr); 234284900Sneel 235284900Sneel#ifdef KTR 236284900Sneelstatic const char * 237284900Sneelvcpu_state2str(enum vcpu_state state) 238284900Sneel{ 239284900Sneel 240284900Sneel switch (state) { 241284900Sneel case VCPU_IDLE: 242284900Sneel return ("idle"); 243284900Sneel case VCPU_FROZEN: 244284900Sneel return ("frozen"); 245284900Sneel case VCPU_RUNNING: 246284900Sneel return ("running"); 247284900Sneel case VCPU_SLEEPING: 248284900Sneel return ("sleeping"); 249284900Sneel default: 250284900Sneel return ("unknown"); 251284900Sneel } 252284900Sneel} 253284900Sneel#endif 254284900Sneel 255221828Sgrehanstatic void 256270071Sgrehanvcpu_cleanup(struct vm *vm, int i, bool destroy) 257221828Sgrehan{ 258266339Sjhb struct vcpu *vcpu = &vm->vcpu[i]; 259266339Sjhb 260266339Sjhb VLAPIC_CLEANUP(vm->cookie, vcpu->vlapic); 261270071Sgrehan if (destroy) { 262270071Sgrehan vmm_stat_free(vcpu->stats); 263270071Sgrehan fpu_save_area_free(vcpu->guestfpu); 264270071Sgrehan } 265221828Sgrehan} 266221828Sgrehan 267221828Sgrehanstatic void 268270071Sgrehanvcpu_init(struct vm *vm, int vcpu_id, bool create) 269221828Sgrehan{ 270221828Sgrehan struct vcpu *vcpu; 271270071Sgrehan 272270071Sgrehan KASSERT(vcpu_id >= 0 && vcpu_id < VM_MAXCPU, 273270071Sgrehan ("vcpu_init: invalid vcpu %d", vcpu_id)); 274270071Sgrehan 275221828Sgrehan vcpu = &vm->vcpu[vcpu_id]; 276221828Sgrehan 277270071Sgrehan if (create) { 278270071Sgrehan KASSERT(!vcpu_lock_initialized(vcpu), ("vcpu %d already " 279270071Sgrehan "initialized", vcpu_id)); 280270071Sgrehan vcpu_lock_init(vcpu); 281270071Sgrehan vcpu->state = VCPU_IDLE; 282270071Sgrehan vcpu->hostcpu = NOCPU; 283270071Sgrehan vcpu->guestfpu = fpu_save_area_alloc(); 284270071Sgrehan vcpu->stats = vmm_stat_alloc(); 285270071Sgrehan } 286270071Sgrehan 287266339Sjhb vcpu->vlapic = VLAPIC_INIT(vm->cookie, vcpu_id); 288267447Sjhb vm_set_x2apic_state(vm, vcpu_id, X2APIC_DISABLED); 289284900Sneel vcpu->reqidle = 0; 290270159Sgrehan vcpu->exitintinfo = 0; 291270071Sgrehan vcpu->nmi_pending = 0; 292270071Sgrehan vcpu->extint_pending = 0; 293270071Sgrehan vcpu->exception_pending = 0; 294267427Sjhb vcpu->guest_xcr0 = XFEATURE_ENABLED_X87; 295234695Sgrehan fpu_save_area_reset(vcpu->guestfpu); 296270071Sgrehan vmm_stat_init(vcpu->stats); 297221828Sgrehan} 298221828Sgrehan 299276403Sneelint 300276403Sneelvcpu_trace_exceptions(struct vm *vm, int vcpuid) 301276403Sneel{ 302276403Sneel 303276403Sneel return (trace_guest_exceptions); 304276403Sneel} 305276403Sneel 306240894Sneelstruct vm_exit * 307240894Sneelvm_exitinfo(struct vm *vm, int cpuid) 308240894Sneel{ 309240894Sneel struct vcpu *vcpu; 310240894Sneel 311240894Sneel if (cpuid < 0 || cpuid >= VM_MAXCPU) 312240894Sneel panic("vm_exitinfo: invalid cpuid %d", cpuid); 313240894Sneel 314240894Sneel vcpu = &vm->vcpu[cpuid]; 315240894Sneel 316240894Sneel return (&vcpu->exitinfo); 317240894Sneel} 318240894Sneel 319261275Sjhbstatic void 320261275Sjhbvmm_resume(void) 321261275Sjhb{ 322261275Sjhb VMM_RESUME(); 323261275Sjhb} 324261275Sjhb 325221828Sgrehanstatic int 326221828Sgrehanvmm_init(void) 327221828Sgrehan{ 328221828Sgrehan int error; 329221828Sgrehan 330242275Sneel vmm_host_state_init(); 331221828Sgrehan 332266339Sjhb vmm_ipinum = vmm_ipi_alloc(); 333266339Sjhb if (vmm_ipinum == 0) 334266339Sjhb vmm_ipinum = IPI_AST; 335266339Sjhb 336221828Sgrehan error = vmm_mem_init(); 337221828Sgrehan if (error) 338221828Sgrehan return (error); 339221828Sgrehan 340221828Sgrehan if (vmm_is_intel()) 341221828Sgrehan ops = &vmm_ops_intel; 342221828Sgrehan else if (vmm_is_amd()) 343221828Sgrehan ops = &vmm_ops_amd; 344221828Sgrehan else 345221828Sgrehan return (ENXIO); 346221828Sgrehan 347261275Sjhb vmm_resume_p = vmm_resume; 348221828Sgrehan 349266339Sjhb return (VMM_INIT(vmm_ipinum)); 350221828Sgrehan} 351221828Sgrehan 352221828Sgrehanstatic int 353221828Sgrehanvmm_handler(module_t mod, int what, void *arg) 354221828Sgrehan{ 355221828Sgrehan int error; 356221828Sgrehan 357221828Sgrehan switch (what) { 358221828Sgrehan case MOD_LOAD: 359221828Sgrehan vmmdev_init(); 360221828Sgrehan error = vmm_init(); 361249396Sneel if (error == 0) 362249396Sneel vmm_initialized = 1; 363221828Sgrehan break; 364221828Sgrehan case MOD_UNLOAD: 365241454Sneel error = vmmdev_cleanup(); 366241454Sneel if (error == 0) { 367261275Sjhb vmm_resume_p = NULL; 368241454Sneel iommu_cleanup(); 369266339Sjhb if (vmm_ipinum != IPI_AST) 370266339Sjhb vmm_ipi_free(vmm_ipinum); 371241454Sneel error = VMM_CLEANUP(); 372253854Sgrehan /* 373253854Sgrehan * Something bad happened - prevent new 374253854Sgrehan * VMs from being created 375253854Sgrehan */ 376253854Sgrehan if (error) 377253854Sgrehan vmm_initialized = 0; 378241454Sneel } 379221828Sgrehan break; 380221828Sgrehan default: 381221828Sgrehan error = 0; 382221828Sgrehan break; 383221828Sgrehan } 384221828Sgrehan return (error); 385221828Sgrehan} 386221828Sgrehan 387221828Sgrehanstatic moduledata_t vmm_kmod = { 388221828Sgrehan "vmm", 389221828Sgrehan vmm_handler, 390221828Sgrehan NULL 391221828Sgrehan}; 392221828Sgrehan 393221828Sgrehan/* 394245704Sneel * vmm initialization has the following dependencies: 395245704Sneel * 396245704Sneel * - VT-x initialization requires smp_rendezvous() and therefore must happen 397245704Sneel * after SMP is fully functional (after SI_SUB_SMP). 398221828Sgrehan */ 399245704SneelDECLARE_MODULE(vmm, vmm_kmod, SI_SUB_SMP + 1, SI_ORDER_ANY); 400221828SgrehanMODULE_VERSION(vmm, 1); 401221828Sgrehan 402270071Sgrehanstatic void 403270071Sgrehanvm_init(struct vm *vm, bool create) 404270071Sgrehan{ 405270071Sgrehan int i; 406270071Sgrehan 407270071Sgrehan vm->cookie = VMINIT(vm, vmspace_pmap(vm->vmspace)); 408270071Sgrehan vm->iommu = NULL; 409270071Sgrehan vm->vioapic = vioapic_init(vm); 410270071Sgrehan vm->vhpet = vhpet_init(vm); 411270071Sgrehan vm->vatpic = vatpic_init(vm); 412270071Sgrehan vm->vatpit = vatpit_init(vm); 413276429Sneel vm->vpmtmr = vpmtmr_init(vm); 414284894Sneel if (create) 415284894Sneel vm->vrtc = vrtc_init(vm); 416270071Sgrehan 417270071Sgrehan CPU_ZERO(&vm->active_cpus); 418270071Sgrehan 419270071Sgrehan vm->suspend = 0; 420270071Sgrehan CPU_ZERO(&vm->suspended_cpus); 421270071Sgrehan 422270071Sgrehan for (i = 0; i < VM_MAXCPU; i++) 423270071Sgrehan vcpu_init(vm, i, create); 424270071Sgrehan} 425270071Sgrehan 426249396Sneelint 427249396Sneelvm_create(const char *name, struct vm **retvm) 428221828Sgrehan{ 429221828Sgrehan struct vm *vm; 430256072Sneel struct vmspace *vmspace; 431221828Sgrehan 432249396Sneel /* 433249396Sneel * If vmm.ko could not be successfully initialized then don't attempt 434249396Sneel * to create the virtual machine. 435249396Sneel */ 436249396Sneel if (!vmm_initialized) 437249396Sneel return (ENXIO); 438249396Sneel 439221828Sgrehan if (name == NULL || strlen(name) >= VM_MAX_NAMELEN) 440249396Sneel return (EINVAL); 441221828Sgrehan 442276429Sneel vmspace = VMSPACE_ALLOC(0, VM_MAXUSER_ADDRESS); 443256072Sneel if (vmspace == NULL) 444256072Sneel return (ENOMEM); 445256072Sneel 446221828Sgrehan vm = malloc(sizeof(struct vm), M_VM, M_WAITOK | M_ZERO); 447221828Sgrehan strcpy(vm->name, name); 448266339Sjhb vm->vmspace = vmspace; 449266339Sjhb mtx_init(&vm->rendezvous_mtx, "vm rendezvous lock", 0, MTX_DEF); 450221828Sgrehan 451270071Sgrehan vm_init(vm, true); 452221828Sgrehan 453249396Sneel *retvm = vm; 454249396Sneel return (0); 455221828Sgrehan} 456221828Sgrehan 457241178Sneelstatic void 458270071Sgrehanvm_cleanup(struct vm *vm, bool destroy) 459221828Sgrehan{ 460295124Sgrehan struct mem_map *mm; 461221828Sgrehan int i; 462221828Sgrehan 463221828Sgrehan ppt_unassign_all(vm); 464221828Sgrehan 465256072Sneel if (vm->iommu != NULL) 466256072Sneel iommu_destroy_domain(vm->iommu); 467256072Sneel 468284894Sneel if (destroy) 469284894Sneel vrtc_cleanup(vm->vrtc); 470284894Sneel else 471284894Sneel vrtc_reset(vm->vrtc); 472276429Sneel vpmtmr_cleanup(vm->vpmtmr); 473268891Sjhb vatpit_cleanup(vm->vatpit); 474261088Sjhb vhpet_cleanup(vm->vhpet); 475268891Sjhb vatpic_cleanup(vm->vatpic); 476261088Sjhb vioapic_cleanup(vm->vioapic); 477261088Sjhb 478270071Sgrehan for (i = 0; i < VM_MAXCPU; i++) 479270071Sgrehan vcpu_cleanup(vm, i, destroy); 480221828Sgrehan 481270071Sgrehan VMCLEANUP(vm->cookie); 482241178Sneel 483295124Sgrehan /* 484295124Sgrehan * System memory is removed from the guest address space only when 485295124Sgrehan * the VM is destroyed. This is because the mapping remains the same 486295124Sgrehan * across VM reset. 487295124Sgrehan * 488295124Sgrehan * Device memory can be relocated by the guest (e.g. using PCI BARs) 489295124Sgrehan * so those mappings are removed on a VM reset. 490295124Sgrehan */ 491295124Sgrehan for (i = 0; i < VM_MAX_MEMMAPS; i++) { 492295124Sgrehan mm = &vm->mem_maps[i]; 493295124Sgrehan if (destroy || !sysmem_mapping(vm, mm)) 494295124Sgrehan vm_free_memmap(vm, i); 495295124Sgrehan } 496295124Sgrehan 497270071Sgrehan if (destroy) { 498295124Sgrehan for (i = 0; i < VM_MAX_MEMSEGS; i++) 499295124Sgrehan vm_free_memseg(vm, i); 500221828Sgrehan 501270071Sgrehan VMSPACE_FREE(vm->vmspace); 502270071Sgrehan vm->vmspace = NULL; 503270071Sgrehan } 504270071Sgrehan} 505221828Sgrehan 506270071Sgrehanvoid 507270071Sgrehanvm_destroy(struct vm *vm) 508270071Sgrehan{ 509270071Sgrehan vm_cleanup(vm, true); 510221828Sgrehan free(vm, M_VM); 511221828Sgrehan} 512221828Sgrehan 513270071Sgrehanint 514270071Sgrehanvm_reinit(struct vm *vm) 515270071Sgrehan{ 516270071Sgrehan int error; 517270071Sgrehan 518270071Sgrehan /* 519270071Sgrehan * A virtual machine can be reset only if all vcpus are suspended. 520270071Sgrehan */ 521270071Sgrehan if (CPU_CMP(&vm->suspended_cpus, &vm->active_cpus) == 0) { 522270071Sgrehan vm_cleanup(vm, false); 523270071Sgrehan vm_init(vm, false); 524270071Sgrehan error = 0; 525270071Sgrehan } else { 526270071Sgrehan error = EBUSY; 527270071Sgrehan } 528270071Sgrehan 529270071Sgrehan return (error); 530270071Sgrehan} 531270071Sgrehan 532221828Sgrehanconst char * 533221828Sgrehanvm_name(struct vm *vm) 534221828Sgrehan{ 535221828Sgrehan return (vm->name); 536221828Sgrehan} 537221828Sgrehan 538221828Sgrehanint 539221828Sgrehanvm_map_mmio(struct vm *vm, vm_paddr_t gpa, size_t len, vm_paddr_t hpa) 540221828Sgrehan{ 541256072Sneel vm_object_t obj; 542221828Sgrehan 543256072Sneel if ((obj = vmm_mmio_alloc(vm->vmspace, gpa, len, hpa)) == NULL) 544256072Sneel return (ENOMEM); 545256072Sneel else 546256072Sneel return (0); 547221828Sgrehan} 548221828Sgrehan 549221828Sgrehanint 550221828Sgrehanvm_unmap_mmio(struct vm *vm, vm_paddr_t gpa, size_t len) 551221828Sgrehan{ 552221828Sgrehan 553256072Sneel vmm_mmio_free(vm->vmspace, gpa, len); 554256072Sneel return (0); 555221828Sgrehan} 556221828Sgrehan 557295124Sgrehan/* 558295124Sgrehan * Return 'true' if 'gpa' is allocated in the guest address space. 559295124Sgrehan * 560295124Sgrehan * This function is called in the context of a running vcpu which acts as 561295124Sgrehan * an implicit lock on 'vm->mem_maps[]'. 562295124Sgrehan */ 563295124Sgrehanbool 564295124Sgrehanvm_mem_allocated(struct vm *vm, int vcpuid, vm_paddr_t gpa) 565241041Sneel{ 566295124Sgrehan struct mem_map *mm; 567241041Sneel int i; 568241041Sneel 569295124Sgrehan#ifdef INVARIANTS 570295124Sgrehan int hostcpu, state; 571295124Sgrehan state = vcpu_get_state(vm, vcpuid, &hostcpu); 572295124Sgrehan KASSERT(state == VCPU_RUNNING && hostcpu == curcpu, 573295124Sgrehan ("%s: invalid vcpu state %d/%d", __func__, state, hostcpu)); 574295124Sgrehan#endif 575295124Sgrehan 576295124Sgrehan for (i = 0; i < VM_MAX_MEMMAPS; i++) { 577295124Sgrehan mm = &vm->mem_maps[i]; 578295124Sgrehan if (mm->len != 0 && gpa >= mm->gpa && gpa < mm->gpa + mm->len) 579295124Sgrehan return (true); /* 'gpa' is sysmem or devmem */ 580241041Sneel } 581241041Sneel 582256072Sneel if (ppt_is_mmio(vm, gpa)) 583295124Sgrehan return (true); /* 'gpa' is pci passthru mmio */ 584256072Sneel 585295124Sgrehan return (false); 586241041Sneel} 587241041Sneel 588221828Sgrehanint 589295124Sgrehanvm_alloc_memseg(struct vm *vm, int ident, size_t len, bool sysmem) 590221828Sgrehan{ 591256072Sneel struct mem_seg *seg; 592295124Sgrehan vm_object_t obj; 593221828Sgrehan 594295124Sgrehan if (ident < 0 || ident >= VM_MAX_MEMSEGS) 595241041Sneel return (EINVAL); 596241041Sneel 597295124Sgrehan if (len == 0 || (len & PAGE_MASK)) 598241041Sneel return (EINVAL); 599221828Sgrehan 600295124Sgrehan seg = &vm->mem_segs[ident]; 601295124Sgrehan if (seg->object != NULL) { 602295124Sgrehan if (seg->len == len && seg->sysmem == sysmem) 603295124Sgrehan return (EEXIST); 604295124Sgrehan else 605295124Sgrehan return (EINVAL); 606295124Sgrehan } 607241041Sneel 608295124Sgrehan obj = vm_object_allocate(OBJT_DEFAULT, len >> PAGE_SHIFT); 609295124Sgrehan if (obj == NULL) 610256072Sneel return (ENOMEM); 611256072Sneel 612256072Sneel seg->len = len; 613295124Sgrehan seg->object = obj; 614295124Sgrehan seg->sysmem = sysmem; 615295124Sgrehan return (0); 616295124Sgrehan} 617241178Sneel 618295124Sgrehanint 619295124Sgrehanvm_get_memseg(struct vm *vm, int ident, size_t *len, bool *sysmem, 620295124Sgrehan vm_object_t *objptr) 621295124Sgrehan{ 622295124Sgrehan struct mem_seg *seg; 623256072Sneel 624295124Sgrehan if (ident < 0 || ident >= VM_MAX_MEMSEGS) 625295124Sgrehan return (EINVAL); 626295124Sgrehan 627295124Sgrehan seg = &vm->mem_segs[ident]; 628295124Sgrehan if (len) 629295124Sgrehan *len = seg->len; 630295124Sgrehan if (sysmem) 631295124Sgrehan *sysmem = seg->sysmem; 632295124Sgrehan if (objptr) 633295124Sgrehan *objptr = seg->object; 634256072Sneel return (0); 635256072Sneel} 636256072Sneel 637295124Sgrehanvoid 638295124Sgrehanvm_free_memseg(struct vm *vm, int ident) 639270159Sgrehan{ 640295124Sgrehan struct mem_seg *seg; 641270159Sgrehan 642295124Sgrehan KASSERT(ident >= 0 && ident < VM_MAX_MEMSEGS, 643295124Sgrehan ("%s: invalid memseg ident %d", __func__, ident)); 644295124Sgrehan 645295124Sgrehan seg = &vm->mem_segs[ident]; 646295124Sgrehan if (seg->object != NULL) { 647295124Sgrehan vm_object_deallocate(seg->object); 648295124Sgrehan bzero(seg, sizeof(struct mem_seg)); 649270159Sgrehan } 650270159Sgrehan} 651270159Sgrehan 652295124Sgrehanint 653295124Sgrehanvm_mmap_memseg(struct vm *vm, vm_paddr_t gpa, int segid, vm_ooffset_t first, 654295124Sgrehan size_t len, int prot, int flags) 655256072Sneel{ 656256072Sneel struct mem_seg *seg; 657295124Sgrehan struct mem_map *m, *map; 658295124Sgrehan vm_ooffset_t last; 659295124Sgrehan int i, error; 660256072Sneel 661295124Sgrehan if (prot == 0 || (prot & ~(VM_PROT_ALL)) != 0) 662295124Sgrehan return (EINVAL); 663256072Sneel 664295124Sgrehan if (flags & ~VM_MEMMAP_F_WIRED) 665295124Sgrehan return (EINVAL); 666256072Sneel 667295124Sgrehan if (segid < 0 || segid >= VM_MAX_MEMSEGS) 668295124Sgrehan return (EINVAL); 669295124Sgrehan 670295124Sgrehan seg = &vm->mem_segs[segid]; 671295124Sgrehan if (seg->object == NULL) 672295124Sgrehan return (EINVAL); 673295124Sgrehan 674295124Sgrehan last = first + len; 675295124Sgrehan if (first < 0 || first >= last || last > seg->len) 676295124Sgrehan return (EINVAL); 677295124Sgrehan 678295124Sgrehan if ((gpa | first | last) & PAGE_MASK) 679295124Sgrehan return (EINVAL); 680295124Sgrehan 681295124Sgrehan map = NULL; 682295124Sgrehan for (i = 0; i < VM_MAX_MEMMAPS; i++) { 683295124Sgrehan m = &vm->mem_maps[i]; 684295124Sgrehan if (m->len == 0) { 685295124Sgrehan map = m; 686295124Sgrehan break; 687295124Sgrehan } 688256072Sneel } 689295124Sgrehan 690295124Sgrehan if (map == NULL) 691295124Sgrehan return (ENOSPC); 692295124Sgrehan 693295124Sgrehan error = vm_map_find(&vm->vmspace->vm_map, seg->object, first, &gpa, 694295124Sgrehan len, 0, VMFS_NO_SPACE, prot, prot, 0); 695295124Sgrehan if (error != KERN_SUCCESS) 696295124Sgrehan return (EFAULT); 697295124Sgrehan 698295124Sgrehan vm_object_reference(seg->object); 699295124Sgrehan 700295124Sgrehan if (flags & VM_MEMMAP_F_WIRED) { 701295124Sgrehan error = vm_map_wire(&vm->vmspace->vm_map, gpa, gpa + len, 702295124Sgrehan VM_MAP_WIRE_USER | VM_MAP_WIRE_NOHOLES); 703295124Sgrehan if (error != KERN_SUCCESS) { 704295124Sgrehan vm_map_remove(&vm->vmspace->vm_map, gpa, gpa + len); 705295124Sgrehan return (EFAULT); 706295124Sgrehan } 707295124Sgrehan } 708295124Sgrehan 709295124Sgrehan map->gpa = gpa; 710295124Sgrehan map->len = len; 711295124Sgrehan map->segoff = first; 712295124Sgrehan map->segid = segid; 713295124Sgrehan map->prot = prot; 714295124Sgrehan map->flags = flags; 715295124Sgrehan return (0); 716256072Sneel} 717256072Sneel 718295124Sgrehanint 719295124Sgrehanvm_mmap_getnext(struct vm *vm, vm_paddr_t *gpa, int *segid, 720295124Sgrehan vm_ooffset_t *segoff, size_t *len, int *prot, int *flags) 721256072Sneel{ 722295124Sgrehan struct mem_map *mm, *mmnext; 723295124Sgrehan int i; 724256072Sneel 725295124Sgrehan mmnext = NULL; 726295124Sgrehan for (i = 0; i < VM_MAX_MEMMAPS; i++) { 727295124Sgrehan mm = &vm->mem_maps[i]; 728295124Sgrehan if (mm->len == 0 || mm->gpa < *gpa) 729256072Sneel continue; 730295124Sgrehan if (mmnext == NULL || mm->gpa < mmnext->gpa) 731295124Sgrehan mmnext = mm; 732295124Sgrehan } 733256072Sneel 734295124Sgrehan if (mmnext != NULL) { 735295124Sgrehan *gpa = mmnext->gpa; 736295124Sgrehan if (segid) 737295124Sgrehan *segid = mmnext->segid; 738295124Sgrehan if (segoff) 739295124Sgrehan *segoff = mmnext->segoff; 740295124Sgrehan if (len) 741295124Sgrehan *len = mmnext->len; 742295124Sgrehan if (prot) 743295124Sgrehan *prot = mmnext->prot; 744295124Sgrehan if (flags) 745295124Sgrehan *flags = mmnext->flags; 746295124Sgrehan return (0); 747295124Sgrehan } else { 748295124Sgrehan return (ENOENT); 749256072Sneel } 750295124Sgrehan} 751256072Sneel 752295124Sgrehanstatic void 753295124Sgrehanvm_free_memmap(struct vm *vm, int ident) 754295124Sgrehan{ 755295124Sgrehan struct mem_map *mm; 756295124Sgrehan int error; 757295124Sgrehan 758295124Sgrehan mm = &vm->mem_maps[ident]; 759295124Sgrehan if (mm->len) { 760295124Sgrehan error = vm_map_remove(&vm->vmspace->vm_map, mm->gpa, 761295124Sgrehan mm->gpa + mm->len); 762295124Sgrehan KASSERT(error == KERN_SUCCESS, ("%s: vm_map_remove error %d", 763295124Sgrehan __func__, error)); 764295124Sgrehan bzero(mm, sizeof(struct mem_map)); 765256072Sneel } 766295124Sgrehan} 767241178Sneel 768295124Sgrehanstatic __inline bool 769295124Sgrehansysmem_mapping(struct vm *vm, struct mem_map *mm) 770295124Sgrehan{ 771295124Sgrehan 772295124Sgrehan if (mm->len != 0 && vm->mem_segs[mm->segid].sysmem) 773295124Sgrehan return (true); 774295124Sgrehan else 775295124Sgrehan return (false); 776256072Sneel} 777256072Sneel 778295124Sgrehanstatic vm_paddr_t 779295124Sgrehansysmem_maxaddr(struct vm *vm) 780295124Sgrehan{ 781295124Sgrehan struct mem_map *mm; 782295124Sgrehan vm_paddr_t maxaddr; 783295124Sgrehan int i; 784295124Sgrehan 785295124Sgrehan maxaddr = 0; 786295124Sgrehan for (i = 0; i < VM_MAX_MEMMAPS; i++) { 787295124Sgrehan mm = &vm->mem_maps[i]; 788295124Sgrehan if (sysmem_mapping(vm, mm)) { 789295124Sgrehan if (maxaddr < mm->gpa + mm->len) 790295124Sgrehan maxaddr = mm->gpa + mm->len; 791295124Sgrehan } 792295124Sgrehan } 793295124Sgrehan return (maxaddr); 794295124Sgrehan} 795295124Sgrehan 796256072Sneelstatic void 797256072Sneelvm_iommu_modify(struct vm *vm, boolean_t map) 798256072Sneel{ 799256072Sneel int i, sz; 800256072Sneel vm_paddr_t gpa, hpa; 801295124Sgrehan struct mem_map *mm; 802256072Sneel void *vp, *cookie, *host_domain; 803256072Sneel 804256072Sneel sz = PAGE_SIZE; 805256072Sneel host_domain = iommu_host_domain(); 806256072Sneel 807295124Sgrehan for (i = 0; i < VM_MAX_MEMMAPS; i++) { 808295124Sgrehan mm = &vm->mem_maps[i]; 809295124Sgrehan if (!sysmem_mapping(vm, mm)) 810295124Sgrehan continue; 811256072Sneel 812295124Sgrehan if (map) { 813295124Sgrehan KASSERT((mm->flags & VM_MEMMAP_F_IOMMU) == 0, 814295124Sgrehan ("iommu map found invalid memmap %#lx/%#lx/%#x", 815295124Sgrehan mm->gpa, mm->len, mm->flags)); 816295124Sgrehan if ((mm->flags & VM_MEMMAP_F_WIRED) == 0) 817295124Sgrehan continue; 818295124Sgrehan mm->flags |= VM_MEMMAP_F_IOMMU; 819295124Sgrehan } else { 820295124Sgrehan if ((mm->flags & VM_MEMMAP_F_IOMMU) == 0) 821295124Sgrehan continue; 822295124Sgrehan mm->flags &= ~VM_MEMMAP_F_IOMMU; 823295124Sgrehan KASSERT((mm->flags & VM_MEMMAP_F_WIRED) != 0, 824295124Sgrehan ("iommu unmap found invalid memmap %#lx/%#lx/%#x", 825295124Sgrehan mm->gpa, mm->len, mm->flags)); 826295124Sgrehan } 827295124Sgrehan 828295124Sgrehan gpa = mm->gpa; 829295124Sgrehan while (gpa < mm->gpa + mm->len) { 830295124Sgrehan vp = vm_gpa_hold(vm, -1, gpa, PAGE_SIZE, VM_PROT_WRITE, 831256072Sneel &cookie); 832256072Sneel KASSERT(vp != NULL, ("vm(%s) could not map gpa %#lx", 833256072Sneel vm_name(vm), gpa)); 834256072Sneel 835256072Sneel vm_gpa_release(cookie); 836256072Sneel 837256072Sneel hpa = DMAP_TO_PHYS((uintptr_t)vp); 838256072Sneel if (map) { 839256072Sneel iommu_create_mapping(vm->iommu, gpa, hpa, sz); 840256072Sneel iommu_remove_mapping(host_domain, hpa, sz); 841256072Sneel } else { 842256072Sneel iommu_remove_mapping(vm->iommu, gpa, sz); 843256072Sneel iommu_create_mapping(host_domain, hpa, hpa, sz); 844256072Sneel } 845256072Sneel 846256072Sneel gpa += PAGE_SIZE; 847256072Sneel } 848241178Sneel } 849241178Sneel 850256072Sneel /* 851256072Sneel * Invalidate the cached translations associated with the domain 852256072Sneel * from which pages were removed. 853256072Sneel */ 854256072Sneel if (map) 855256072Sneel iommu_invalidate_tlb(host_domain); 856256072Sneel else 857256072Sneel iommu_invalidate_tlb(vm->iommu); 858256072Sneel} 859256072Sneel 860256072Sneel#define vm_iommu_unmap(vm) vm_iommu_modify((vm), FALSE) 861256072Sneel#define vm_iommu_map(vm) vm_iommu_modify((vm), TRUE) 862256072Sneel 863256072Sneelint 864256072Sneelvm_unassign_pptdev(struct vm *vm, int bus, int slot, int func) 865256072Sneel{ 866256072Sneel int error; 867256072Sneel 868256072Sneel error = ppt_unassign_device(vm, bus, slot, func); 869256072Sneel if (error) 870221828Sgrehan return (error); 871256072Sneel 872295124Sgrehan if (ppt_assigned_devices(vm) == 0) 873256072Sneel vm_iommu_unmap(vm); 874295124Sgrehan 875256072Sneel return (0); 876256072Sneel} 877221828Sgrehan 878256072Sneelint 879256072Sneelvm_assign_pptdev(struct vm *vm, int bus, int slot, int func) 880256072Sneel{ 881256072Sneel int error; 882256072Sneel vm_paddr_t maxaddr; 883256072Sneel 884295124Sgrehan /* Set up the IOMMU to do the 'gpa' to 'hpa' translation */ 885267070Sjhb if (ppt_assigned_devices(vm) == 0) { 886256072Sneel KASSERT(vm->iommu == NULL, 887256072Sneel ("vm_assign_pptdev: iommu must be NULL")); 888295124Sgrehan maxaddr = sysmem_maxaddr(vm); 889256072Sneel vm->iommu = iommu_create_domain(maxaddr); 890306472Sjhb if (vm->iommu == NULL) 891306472Sjhb return (ENXIO); 892256072Sneel vm_iommu_map(vm); 893256072Sneel } 894256072Sneel 895256072Sneel error = ppt_assign_device(vm, bus, slot, func); 896256072Sneel return (error); 897221828Sgrehan} 898221828Sgrehan 899256072Sneelvoid * 900295124Sgrehanvm_gpa_hold(struct vm *vm, int vcpuid, vm_paddr_t gpa, size_t len, int reqprot, 901256072Sneel void **cookie) 902221828Sgrehan{ 903295124Sgrehan int i, count, pageoff; 904295124Sgrehan struct mem_map *mm; 905256072Sneel vm_page_t m; 906295124Sgrehan#ifdef INVARIANTS 907295124Sgrehan /* 908295124Sgrehan * All vcpus are frozen by ioctls that modify the memory map 909295124Sgrehan * (e.g. VM_MMAP_MEMSEG). Therefore 'vm->memmap[]' stability is 910295124Sgrehan * guaranteed if at least one vcpu is in the VCPU_FROZEN state. 911295124Sgrehan */ 912295124Sgrehan int state; 913305673Sjhb KASSERT(vcpuid >= -1 && vcpuid < VM_MAXCPU, ("%s: invalid vcpuid %d", 914295124Sgrehan __func__, vcpuid)); 915295124Sgrehan for (i = 0; i < VM_MAXCPU; i++) { 916295124Sgrehan if (vcpuid != -1 && vcpuid != i) 917295124Sgrehan continue; 918295124Sgrehan state = vcpu_get_state(vm, i, NULL); 919295124Sgrehan KASSERT(state == VCPU_FROZEN, ("%s: invalid vcpu state %d", 920295124Sgrehan __func__, state)); 921295124Sgrehan } 922295124Sgrehan#endif 923256072Sneel pageoff = gpa & PAGE_MASK; 924256072Sneel if (len > PAGE_SIZE - pageoff) 925256072Sneel panic("vm_gpa_hold: invalid gpa/len: 0x%016lx/%lu", gpa, len); 926241148Sneel 927295124Sgrehan count = 0; 928295124Sgrehan for (i = 0; i < VM_MAX_MEMMAPS; i++) { 929295124Sgrehan mm = &vm->mem_maps[i]; 930295124Sgrehan if (sysmem_mapping(vm, mm) && gpa >= mm->gpa && 931295124Sgrehan gpa < mm->gpa + mm->len) { 932295124Sgrehan count = vm_fault_quick_hold_pages(&vm->vmspace->vm_map, 933295124Sgrehan trunc_page(gpa), PAGE_SIZE, reqprot, &m, 1); 934295124Sgrehan break; 935295124Sgrehan } 936295124Sgrehan } 937256072Sneel 938256072Sneel if (count == 1) { 939256072Sneel *cookie = m; 940256072Sneel return ((void *)(PHYS_TO_DMAP(VM_PAGE_TO_PHYS(m)) + pageoff)); 941256072Sneel } else { 942256072Sneel *cookie = NULL; 943256072Sneel return (NULL); 944256072Sneel } 945221828Sgrehan} 946221828Sgrehan 947256072Sneelvoid 948256072Sneelvm_gpa_release(void *cookie) 949256072Sneel{ 950256072Sneel vm_page_t m = cookie; 951256072Sneel 952256072Sneel vm_page_lock(m); 953256072Sneel vm_page_unhold(m); 954256072Sneel vm_page_unlock(m); 955256072Sneel} 956256072Sneel 957221828Sgrehanint 958221828Sgrehanvm_get_register(struct vm *vm, int vcpu, int reg, uint64_t *retval) 959221828Sgrehan{ 960221828Sgrehan 961221828Sgrehan if (vcpu < 0 || vcpu >= VM_MAXCPU) 962221828Sgrehan return (EINVAL); 963221828Sgrehan 964221828Sgrehan if (reg >= VM_REG_LAST) 965221828Sgrehan return (EINVAL); 966221828Sgrehan 967221828Sgrehan return (VMGETREG(vm->cookie, vcpu, reg, retval)); 968221828Sgrehan} 969221828Sgrehan 970221828Sgrehanint 971284894Sneelvm_set_register(struct vm *vm, int vcpuid, int reg, uint64_t val) 972221828Sgrehan{ 973284894Sneel struct vcpu *vcpu; 974284894Sneel int error; 975221828Sgrehan 976284894Sneel if (vcpuid < 0 || vcpuid >= VM_MAXCPU) 977221828Sgrehan return (EINVAL); 978221828Sgrehan 979221828Sgrehan if (reg >= VM_REG_LAST) 980221828Sgrehan return (EINVAL); 981221828Sgrehan 982284894Sneel error = VMSETREG(vm->cookie, vcpuid, reg, val); 983284894Sneel if (error || reg != VM_REG_GUEST_RIP) 984284894Sneel return (error); 985284894Sneel 986284894Sneel /* Set 'nextrip' to match the value of %rip */ 987284894Sneel VCPU_CTR1(vm, vcpuid, "Setting nextrip to %#lx", val); 988284894Sneel vcpu = &vm->vcpu[vcpuid]; 989284894Sneel vcpu->nextrip = val; 990284894Sneel return (0); 991221828Sgrehan} 992221828Sgrehan 993221828Sgrehanstatic boolean_t 994221828Sgrehanis_descriptor_table(int reg) 995221828Sgrehan{ 996221828Sgrehan 997221828Sgrehan switch (reg) { 998221828Sgrehan case VM_REG_GUEST_IDTR: 999221828Sgrehan case VM_REG_GUEST_GDTR: 1000221828Sgrehan return (TRUE); 1001221828Sgrehan default: 1002221828Sgrehan return (FALSE); 1003221828Sgrehan } 1004221828Sgrehan} 1005221828Sgrehan 1006221828Sgrehanstatic boolean_t 1007221828Sgrehanis_segment_register(int reg) 1008221828Sgrehan{ 1009221828Sgrehan 1010221828Sgrehan switch (reg) { 1011221828Sgrehan case VM_REG_GUEST_ES: 1012221828Sgrehan case VM_REG_GUEST_CS: 1013221828Sgrehan case VM_REG_GUEST_SS: 1014221828Sgrehan case VM_REG_GUEST_DS: 1015221828Sgrehan case VM_REG_GUEST_FS: 1016221828Sgrehan case VM_REG_GUEST_GS: 1017221828Sgrehan case VM_REG_GUEST_TR: 1018221828Sgrehan case VM_REG_GUEST_LDTR: 1019221828Sgrehan return (TRUE); 1020221828Sgrehan default: 1021221828Sgrehan return (FALSE); 1022221828Sgrehan } 1023221828Sgrehan} 1024221828Sgrehan 1025221828Sgrehanint 1026221828Sgrehanvm_get_seg_desc(struct vm *vm, int vcpu, int reg, 1027221828Sgrehan struct seg_desc *desc) 1028221828Sgrehan{ 1029221828Sgrehan 1030221828Sgrehan if (vcpu < 0 || vcpu >= VM_MAXCPU) 1031221828Sgrehan return (EINVAL); 1032221828Sgrehan 1033221828Sgrehan if (!is_segment_register(reg) && !is_descriptor_table(reg)) 1034221828Sgrehan return (EINVAL); 1035221828Sgrehan 1036221828Sgrehan return (VMGETDESC(vm->cookie, vcpu, reg, desc)); 1037221828Sgrehan} 1038221828Sgrehan 1039221828Sgrehanint 1040221828Sgrehanvm_set_seg_desc(struct vm *vm, int vcpu, int reg, 1041221828Sgrehan struct seg_desc *desc) 1042221828Sgrehan{ 1043221828Sgrehan if (vcpu < 0 || vcpu >= VM_MAXCPU) 1044221828Sgrehan return (EINVAL); 1045221828Sgrehan 1046221828Sgrehan if (!is_segment_register(reg) && !is_descriptor_table(reg)) 1047221828Sgrehan return (EINVAL); 1048221828Sgrehan 1049221828Sgrehan return (VMSETDESC(vm->cookie, vcpu, reg, desc)); 1050221828Sgrehan} 1051221828Sgrehan 1052221828Sgrehanstatic void 1053221828Sgrehanrestore_guest_fpustate(struct vcpu *vcpu) 1054221828Sgrehan{ 1055221828Sgrehan 1056234695Sgrehan /* flush host state to the pcb */ 1057234695Sgrehan fpuexit(curthread); 1058242122Sneel 1059242122Sneel /* restore guest FPU state */ 1060221828Sgrehan fpu_stop_emulating(); 1061234695Sgrehan fpurestore(vcpu->guestfpu); 1062242122Sneel 1063267427Sjhb /* restore guest XCR0 if XSAVE is enabled in the host */ 1064267427Sjhb if (rcr4() & CR4_XSAVE) 1065267427Sjhb load_xcr(0, vcpu->guest_xcr0); 1066267427Sjhb 1067242122Sneel /* 1068242122Sneel * The FPU is now "dirty" with the guest's state so turn on emulation 1069242122Sneel * to trap any access to the FPU by the host. 1070242122Sneel */ 1071242122Sneel fpu_start_emulating(); 1072221828Sgrehan} 1073221828Sgrehan 1074221828Sgrehanstatic void 1075221828Sgrehansave_guest_fpustate(struct vcpu *vcpu) 1076221828Sgrehan{ 1077221828Sgrehan 1078242122Sneel if ((rcr0() & CR0_TS) == 0) 1079242122Sneel panic("fpu emulation not enabled in host!"); 1080242122Sneel 1081267427Sjhb /* save guest XCR0 and restore host XCR0 */ 1082267427Sjhb if (rcr4() & CR4_XSAVE) { 1083267427Sjhb vcpu->guest_xcr0 = rxcr(0); 1084267427Sjhb load_xcr(0, vmm_get_host_xcr0()); 1085267427Sjhb } 1086267427Sjhb 1087242122Sneel /* save guest FPU state */ 1088242122Sneel fpu_stop_emulating(); 1089234695Sgrehan fpusave(vcpu->guestfpu); 1090221828Sgrehan fpu_start_emulating(); 1091221828Sgrehan} 1092221828Sgrehan 1093248389Sneelstatic VMM_STAT(VCPU_IDLE_TICKS, "number of ticks vcpu was idle"); 1094242065Sneel 1095256072Sneelstatic int 1096284900Sneelvcpu_set_state_locked(struct vm *vm, int vcpuid, enum vcpu_state newstate, 1097266393Sjhb bool from_idle) 1098256072Sneel{ 1099284900Sneel struct vcpu *vcpu; 1100256072Sneel int error; 1101256072Sneel 1102284900Sneel vcpu = &vm->vcpu[vcpuid]; 1103256072Sneel vcpu_assert_locked(vcpu); 1104256072Sneel 1105256072Sneel /* 1106266393Sjhb * State transitions from the vmmdev_ioctl() must always begin from 1107266393Sjhb * the VCPU_IDLE state. This guarantees that there is only a single 1108266393Sjhb * ioctl() operating on a vcpu at any point. 1109266393Sjhb */ 1110266393Sjhb if (from_idle) { 1111284900Sneel while (vcpu->state != VCPU_IDLE) { 1112284900Sneel vcpu->reqidle = 1; 1113284900Sneel vcpu_notify_event_locked(vcpu, false); 1114284900Sneel VCPU_CTR1(vm, vcpuid, "vcpu state change from %s to " 1115284900Sneel "idle requested", vcpu_state2str(vcpu->state)); 1116266393Sjhb msleep_spin(&vcpu->state, &vcpu->mtx, "vmstat", hz); 1117284900Sneel } 1118266393Sjhb } else { 1119266393Sjhb KASSERT(vcpu->state != VCPU_IDLE, ("invalid transition from " 1120266393Sjhb "vcpu idle state")); 1121266393Sjhb } 1122266393Sjhb 1123266393Sjhb if (vcpu->state == VCPU_RUNNING) { 1124266393Sjhb KASSERT(vcpu->hostcpu == curcpu, ("curcpu %d and hostcpu %d " 1125266393Sjhb "mismatch for running vcpu", curcpu, vcpu->hostcpu)); 1126266393Sjhb } else { 1127266393Sjhb KASSERT(vcpu->hostcpu == NOCPU, ("Invalid hostcpu %d for a " 1128266393Sjhb "vcpu that is not running", vcpu->hostcpu)); 1129266393Sjhb } 1130266393Sjhb 1131266393Sjhb /* 1132256072Sneel * The following state transitions are allowed: 1133256072Sneel * IDLE -> FROZEN -> IDLE 1134256072Sneel * FROZEN -> RUNNING -> FROZEN 1135256072Sneel * FROZEN -> SLEEPING -> FROZEN 1136256072Sneel */ 1137256072Sneel switch (vcpu->state) { 1138256072Sneel case VCPU_IDLE: 1139256072Sneel case VCPU_RUNNING: 1140256072Sneel case VCPU_SLEEPING: 1141256072Sneel error = (newstate != VCPU_FROZEN); 1142256072Sneel break; 1143256072Sneel case VCPU_FROZEN: 1144256072Sneel error = (newstate == VCPU_FROZEN); 1145256072Sneel break; 1146256072Sneel default: 1147256072Sneel error = 1; 1148256072Sneel break; 1149256072Sneel } 1150256072Sneel 1151266393Sjhb if (error) 1152266393Sjhb return (EBUSY); 1153266393Sjhb 1154284900Sneel VCPU_CTR2(vm, vcpuid, "vcpu state changed from %s to %s", 1155284900Sneel vcpu_state2str(vcpu->state), vcpu_state2str(newstate)); 1156284900Sneel 1157266393Sjhb vcpu->state = newstate; 1158266393Sjhb if (newstate == VCPU_RUNNING) 1159266393Sjhb vcpu->hostcpu = curcpu; 1160256072Sneel else 1161266393Sjhb vcpu->hostcpu = NOCPU; 1162256072Sneel 1163266393Sjhb if (newstate == VCPU_IDLE) 1164266393Sjhb wakeup(&vcpu->state); 1165266393Sjhb 1166266393Sjhb return (0); 1167256072Sneel} 1168256072Sneel 1169256072Sneelstatic void 1170256072Sneelvcpu_require_state(struct vm *vm, int vcpuid, enum vcpu_state newstate) 1171256072Sneel{ 1172256072Sneel int error; 1173256072Sneel 1174266393Sjhb if ((error = vcpu_set_state(vm, vcpuid, newstate, false)) != 0) 1175256072Sneel panic("Error %d setting state to %d\n", error, newstate); 1176256072Sneel} 1177256072Sneel 1178256072Sneelstatic void 1179284900Sneelvcpu_require_state_locked(struct vm *vm, int vcpuid, enum vcpu_state newstate) 1180256072Sneel{ 1181256072Sneel int error; 1182256072Sneel 1183284900Sneel if ((error = vcpu_set_state_locked(vm, vcpuid, newstate, false)) != 0) 1184256072Sneel panic("Error %d setting state to %d", error, newstate); 1185256072Sneel} 1186256072Sneel 1187266339Sjhbstatic void 1188266339Sjhbvm_set_rendezvous_func(struct vm *vm, vm_rendezvous_func_t func) 1189266339Sjhb{ 1190266339Sjhb 1191266339Sjhb KASSERT(mtx_owned(&vm->rendezvous_mtx), ("rendezvous_mtx not locked")); 1192266339Sjhb 1193266339Sjhb /* 1194266339Sjhb * Update 'rendezvous_func' and execute a write memory barrier to 1195266339Sjhb * ensure that it is visible across all host cpus. This is not needed 1196266339Sjhb * for correctness but it does ensure that all the vcpus will notice 1197266339Sjhb * that the rendezvous is requested immediately. 1198266339Sjhb */ 1199266339Sjhb vm->rendezvous_func = func; 1200266339Sjhb wmb(); 1201266339Sjhb} 1202266339Sjhb 1203266339Sjhb#define RENDEZVOUS_CTR0(vm, vcpuid, fmt) \ 1204266339Sjhb do { \ 1205266339Sjhb if (vcpuid >= 0) \ 1206266339Sjhb VCPU_CTR0(vm, vcpuid, fmt); \ 1207266339Sjhb else \ 1208266339Sjhb VM_CTR0(vm, fmt); \ 1209266339Sjhb } while (0) 1210266339Sjhb 1211266339Sjhbstatic void 1212266339Sjhbvm_handle_rendezvous(struct vm *vm, int vcpuid) 1213266339Sjhb{ 1214266339Sjhb 1215266339Sjhb KASSERT(vcpuid == -1 || (vcpuid >= 0 && vcpuid < VM_MAXCPU), 1216266339Sjhb ("vm_handle_rendezvous: invalid vcpuid %d", vcpuid)); 1217266339Sjhb 1218266339Sjhb mtx_lock(&vm->rendezvous_mtx); 1219266339Sjhb while (vm->rendezvous_func != NULL) { 1220266339Sjhb /* 'rendezvous_req_cpus' must be a subset of 'active_cpus' */ 1221266339Sjhb CPU_AND(&vm->rendezvous_req_cpus, &vm->active_cpus); 1222266339Sjhb 1223266339Sjhb if (vcpuid != -1 && 1224266339Sjhb CPU_ISSET(vcpuid, &vm->rendezvous_req_cpus) && 1225266339Sjhb !CPU_ISSET(vcpuid, &vm->rendezvous_done_cpus)) { 1226266339Sjhb VCPU_CTR0(vm, vcpuid, "Calling rendezvous func"); 1227266339Sjhb (*vm->rendezvous_func)(vm, vcpuid, vm->rendezvous_arg); 1228266339Sjhb CPU_SET(vcpuid, &vm->rendezvous_done_cpus); 1229266339Sjhb } 1230266339Sjhb if (CPU_CMP(&vm->rendezvous_req_cpus, 1231266339Sjhb &vm->rendezvous_done_cpus) == 0) { 1232266339Sjhb VCPU_CTR0(vm, vcpuid, "Rendezvous completed"); 1233266339Sjhb vm_set_rendezvous_func(vm, NULL); 1234266339Sjhb wakeup(&vm->rendezvous_func); 1235266339Sjhb break; 1236266339Sjhb } 1237266339Sjhb RENDEZVOUS_CTR0(vm, vcpuid, "Wait for rendezvous completion"); 1238266339Sjhb mtx_sleep(&vm->rendezvous_func, &vm->rendezvous_mtx, 0, 1239266339Sjhb "vmrndv", 0); 1240266339Sjhb } 1241266339Sjhb mtx_unlock(&vm->rendezvous_mtx); 1242266339Sjhb} 1243266339Sjhb 1244256072Sneel/* 1245256072Sneel * Emulate a guest 'hlt' by sleeping until the vcpu is ready to run. 1246256072Sneel */ 1247256072Sneelstatic int 1248262350Sjhbvm_handle_hlt(struct vm *vm, int vcpuid, bool intr_disabled, bool *retu) 1249256072Sneel{ 1250256072Sneel struct vcpu *vcpu; 1251268935Sjhb const char *wmesg; 1252284894Sneel int t, vcpu_halted, vm_halted; 1253256072Sneel 1254268935Sjhb KASSERT(!CPU_ISSET(vcpuid, &vm->halted_cpus), ("vcpu already halted")); 1255268935Sjhb 1256256072Sneel vcpu = &vm->vcpu[vcpuid]; 1257268935Sjhb vcpu_halted = 0; 1258268935Sjhb vm_halted = 0; 1259256072Sneel 1260256072Sneel vcpu_lock(vcpu); 1261268935Sjhb while (1) { 1262268935Sjhb /* 1263268935Sjhb * Do a final check for pending NMI or interrupts before 1264268935Sjhb * really putting this thread to sleep. Also check for 1265268935Sjhb * software events that would cause this vcpu to wakeup. 1266268935Sjhb * 1267268935Sjhb * These interrupts/events could have happened after the 1268268935Sjhb * vcpu returned from VMRUN() and before it acquired the 1269268935Sjhb * vcpu lock above. 1270268935Sjhb */ 1271284900Sneel if (vm->rendezvous_func != NULL || vm->suspend || vcpu->reqidle) 1272268935Sjhb break; 1273268935Sjhb if (vm_nmi_pending(vm, vcpuid)) 1274268935Sjhb break; 1275268935Sjhb if (!intr_disabled) { 1276268935Sjhb if (vm_extint_pending(vm, vcpuid) || 1277268935Sjhb vlapic_pending_intr(vcpu->vlapic, NULL)) { 1278268935Sjhb break; 1279268935Sjhb } 1280268935Sjhb } 1281256072Sneel 1282270159Sgrehan /* Don't go to sleep if the vcpu thread needs to yield */ 1283270159Sgrehan if (vcpu_should_yield(vm, vcpuid)) 1284270159Sgrehan break; 1285270159Sgrehan 1286268935Sjhb /* 1287268935Sjhb * Some Linux guests implement "halt" by having all vcpus 1288268935Sjhb * execute HLT with interrupts disabled. 'halted_cpus' keeps 1289268935Sjhb * track of the vcpus that have entered this state. When all 1290268935Sjhb * vcpus enter the halted state the virtual machine is halted. 1291268935Sjhb */ 1292268935Sjhb if (intr_disabled) { 1293268935Sjhb wmesg = "vmhalt"; 1294268935Sjhb VCPU_CTR0(vm, vcpuid, "Halted"); 1295268935Sjhb if (!vcpu_halted && halt_detection_enabled) { 1296268935Sjhb vcpu_halted = 1; 1297268935Sjhb CPU_SET_ATOMIC(vcpuid, &vm->halted_cpus); 1298268935Sjhb } 1299268935Sjhb if (CPU_CMP(&vm->halted_cpus, &vm->active_cpus) == 0) { 1300268935Sjhb vm_halted = 1; 1301268935Sjhb break; 1302268935Sjhb } 1303268935Sjhb } else { 1304268935Sjhb wmesg = "vmidle"; 1305268935Sjhb } 1306268935Sjhb 1307256072Sneel t = ticks; 1308284900Sneel vcpu_require_state_locked(vm, vcpuid, VCPU_SLEEPING); 1309270159Sgrehan /* 1310270159Sgrehan * XXX msleep_spin() cannot be interrupted by signals so 1311270159Sgrehan * wake up periodically to check pending signals. 1312270159Sgrehan */ 1313270159Sgrehan msleep_spin(vcpu, &vcpu->mtx, wmesg, hz); 1314284900Sneel vcpu_require_state_locked(vm, vcpuid, VCPU_FROZEN); 1315256072Sneel vmm_stat_incr(vm, vcpuid, VCPU_IDLE_TICKS, ticks - t); 1316256072Sneel } 1317268935Sjhb 1318268935Sjhb if (vcpu_halted) 1319268935Sjhb CPU_CLR_ATOMIC(vcpuid, &vm->halted_cpus); 1320268935Sjhb 1321256072Sneel vcpu_unlock(vcpu); 1322256072Sneel 1323268935Sjhb if (vm_halted) 1324268935Sjhb vm_suspend(vm, VM_SUSPEND_HALT); 1325266339Sjhb 1326256072Sneel return (0); 1327256072Sneel} 1328256072Sneel 1329256072Sneelstatic int 1330262350Sjhbvm_handle_paging(struct vm *vm, int vcpuid, bool *retu) 1331256072Sneel{ 1332256072Sneel int rv, ftype; 1333256072Sneel struct vm_map *map; 1334256072Sneel struct vcpu *vcpu; 1335256072Sneel struct vm_exit *vme; 1336256072Sneel 1337256072Sneel vcpu = &vm->vcpu[vcpuid]; 1338256072Sneel vme = &vcpu->exitinfo; 1339256072Sneel 1340284894Sneel KASSERT(vme->inst_length == 0, ("%s: invalid inst_length %d", 1341284894Sneel __func__, vme->inst_length)); 1342284894Sneel 1343256072Sneel ftype = vme->u.paging.fault_type; 1344256072Sneel KASSERT(ftype == VM_PROT_READ || 1345256072Sneel ftype == VM_PROT_WRITE || ftype == VM_PROT_EXECUTE, 1346256072Sneel ("vm_handle_paging: invalid fault_type %d", ftype)); 1347256072Sneel 1348256072Sneel if (ftype == VM_PROT_READ || ftype == VM_PROT_WRITE) { 1349256072Sneel rv = pmap_emulate_accessed_dirty(vmspace_pmap(vm->vmspace), 1350256072Sneel vme->u.paging.gpa, ftype); 1351276349Sneel if (rv == 0) { 1352276349Sneel VCPU_CTR2(vm, vcpuid, "%s bit emulation for gpa %#lx", 1353276349Sneel ftype == VM_PROT_READ ? "accessed" : "dirty", 1354276349Sneel vme->u.paging.gpa); 1355256072Sneel goto done; 1356276349Sneel } 1357256072Sneel } 1358256072Sneel 1359256072Sneel map = &vm->vmspace->vm_map; 1360256072Sneel rv = vm_fault(map, vme->u.paging.gpa, ftype, VM_FAULT_NORMAL); 1361256072Sneel 1362261088Sjhb VCPU_CTR3(vm, vcpuid, "vm_handle_paging rv = %d, gpa = %#lx, " 1363261088Sjhb "ftype = %d", rv, vme->u.paging.gpa, ftype); 1364256072Sneel 1365256072Sneel if (rv != KERN_SUCCESS) 1366256072Sneel return (EFAULT); 1367256072Sneeldone: 1368256072Sneel return (0); 1369256072Sneel} 1370256072Sneel 1371256072Sneelstatic int 1372262350Sjhbvm_handle_inst_emul(struct vm *vm, int vcpuid, bool *retu) 1373256072Sneel{ 1374256072Sneel struct vie *vie; 1375256072Sneel struct vcpu *vcpu; 1376256072Sneel struct vm_exit *vme; 1377284899Sneel uint64_t gla, gpa, cs_base; 1378268976Sjhb struct vm_guest_paging *paging; 1379261088Sjhb mem_region_read_t mread; 1380261088Sjhb mem_region_write_t mwrite; 1381270159Sgrehan enum vm_cpu_mode cpu_mode; 1382284900Sneel int cs_d, error, fault; 1383256072Sneel 1384256072Sneel vcpu = &vm->vcpu[vcpuid]; 1385256072Sneel vme = &vcpu->exitinfo; 1386256072Sneel 1387284900Sneel KASSERT(vme->inst_length == 0, ("%s: invalid inst_length %d", 1388284900Sneel __func__, vme->inst_length)); 1389284900Sneel 1390256072Sneel gla = vme->u.inst_emul.gla; 1391256072Sneel gpa = vme->u.inst_emul.gpa; 1392284899Sneel cs_base = vme->u.inst_emul.cs_base; 1393270159Sgrehan cs_d = vme->u.inst_emul.cs_d; 1394256072Sneel vie = &vme->u.inst_emul.vie; 1395268976Sjhb paging = &vme->u.inst_emul.paging; 1396270159Sgrehan cpu_mode = paging->cpu_mode; 1397256072Sneel 1398276349Sneel VCPU_CTR1(vm, vcpuid, "inst_emul fault accessing gpa %#lx", gpa); 1399276349Sneel 1400256072Sneel /* Fetch, decode and emulate the faulting instruction */ 1401276403Sneel if (vie->num_valid == 0) { 1402284899Sneel error = vmm_fetch_instruction(vm, vcpuid, paging, vme->rip + 1403284900Sneel cs_base, VIE_INST_SIZE, vie, &fault); 1404276403Sneel } else { 1405276403Sneel /* 1406276403Sneel * The instruction bytes have already been copied into 'vie' 1407276403Sneel */ 1408284900Sneel error = fault = 0; 1409276403Sneel } 1410284900Sneel if (error || fault) 1411284900Sneel return (error); 1412256072Sneel 1413284900Sneel if (vmm_decode_instruction(vm, vcpuid, gla, cpu_mode, cs_d, vie) != 0) { 1414284900Sneel VCPU_CTR1(vm, vcpuid, "Error decoding instruction at %#lx", 1415284900Sneel vme->rip + cs_base); 1416284900Sneel *retu = true; /* dump instruction bytes in userspace */ 1417284900Sneel return (0); 1418284900Sneel } 1419256072Sneel 1420276403Sneel /* 1421284900Sneel * Update 'nextrip' based on the length of the emulated instruction. 1422276403Sneel */ 1423284900Sneel vme->inst_length = vie->num_processed; 1424284900Sneel vcpu->nextrip += vie->num_processed; 1425284900Sneel VCPU_CTR1(vm, vcpuid, "nextrip updated to %#lx after instruction " 1426284900Sneel "decoding", vcpu->nextrip); 1427276403Sneel 1428261088Sjhb /* return to userland unless this is an in-kernel emulated device */ 1429261088Sjhb if (gpa >= DEFAULT_APIC_BASE && gpa < DEFAULT_APIC_BASE + PAGE_SIZE) { 1430261088Sjhb mread = lapic_mmio_read; 1431261088Sjhb mwrite = lapic_mmio_write; 1432261088Sjhb } else if (gpa >= VIOAPIC_BASE && gpa < VIOAPIC_BASE + VIOAPIC_SIZE) { 1433261088Sjhb mread = vioapic_mmio_read; 1434261088Sjhb mwrite = vioapic_mmio_write; 1435261088Sjhb } else if (gpa >= VHPET_BASE && gpa < VHPET_BASE + VHPET_SIZE) { 1436261088Sjhb mread = vhpet_mmio_read; 1437261088Sjhb mwrite = vhpet_mmio_write; 1438261088Sjhb } else { 1439262350Sjhb *retu = true; 1440256072Sneel return (0); 1441256072Sneel } 1442256072Sneel 1443270159Sgrehan error = vmm_emulate_instruction(vm, vcpuid, gpa, vie, paging, 1444270159Sgrehan mread, mwrite, retu); 1445256072Sneel 1446256072Sneel return (error); 1447256072Sneel} 1448256072Sneel 1449268935Sjhbstatic int 1450268935Sjhbvm_handle_suspend(struct vm *vm, int vcpuid, bool *retu) 1451268935Sjhb{ 1452268935Sjhb int i, done; 1453268935Sjhb struct vcpu *vcpu; 1454268935Sjhb 1455268935Sjhb done = 0; 1456268935Sjhb vcpu = &vm->vcpu[vcpuid]; 1457268935Sjhb 1458268935Sjhb CPU_SET_ATOMIC(vcpuid, &vm->suspended_cpus); 1459268935Sjhb 1460268935Sjhb /* 1461268935Sjhb * Wait until all 'active_cpus' have suspended themselves. 1462268935Sjhb * 1463268935Sjhb * Since a VM may be suspended at any time including when one or 1464268935Sjhb * more vcpus are doing a rendezvous we need to call the rendezvous 1465268935Sjhb * handler while we are waiting to prevent a deadlock. 1466268935Sjhb */ 1467268935Sjhb vcpu_lock(vcpu); 1468268935Sjhb while (1) { 1469268935Sjhb if (CPU_CMP(&vm->suspended_cpus, &vm->active_cpus) == 0) { 1470268935Sjhb VCPU_CTR0(vm, vcpuid, "All vcpus suspended"); 1471268935Sjhb break; 1472268935Sjhb } 1473268935Sjhb 1474268935Sjhb if (vm->rendezvous_func == NULL) { 1475268935Sjhb VCPU_CTR0(vm, vcpuid, "Sleeping during suspend"); 1476284900Sneel vcpu_require_state_locked(vm, vcpuid, VCPU_SLEEPING); 1477268935Sjhb msleep_spin(vcpu, &vcpu->mtx, "vmsusp", hz); 1478284900Sneel vcpu_require_state_locked(vm, vcpuid, VCPU_FROZEN); 1479268935Sjhb } else { 1480268935Sjhb VCPU_CTR0(vm, vcpuid, "Rendezvous during suspend"); 1481268935Sjhb vcpu_unlock(vcpu); 1482268935Sjhb vm_handle_rendezvous(vm, vcpuid); 1483268935Sjhb vcpu_lock(vcpu); 1484268935Sjhb } 1485268935Sjhb } 1486268935Sjhb vcpu_unlock(vcpu); 1487268935Sjhb 1488268935Sjhb /* 1489268935Sjhb * Wakeup the other sleeping vcpus and return to userspace. 1490268935Sjhb */ 1491268935Sjhb for (i = 0; i < VM_MAXCPU; i++) { 1492268935Sjhb if (CPU_ISSET(i, &vm->suspended_cpus)) { 1493268935Sjhb vcpu_notify_event(vm, i, false); 1494268935Sjhb } 1495268935Sjhb } 1496268935Sjhb 1497268935Sjhb *retu = true; 1498268935Sjhb return (0); 1499268935Sjhb} 1500268935Sjhb 1501284900Sneelstatic int 1502284900Sneelvm_handle_reqidle(struct vm *vm, int vcpuid, bool *retu) 1503284900Sneel{ 1504284900Sneel struct vcpu *vcpu = &vm->vcpu[vcpuid]; 1505284900Sneel 1506284900Sneel vcpu_lock(vcpu); 1507284900Sneel KASSERT(vcpu->reqidle, ("invalid vcpu reqidle %d", vcpu->reqidle)); 1508284900Sneel vcpu->reqidle = 0; 1509284900Sneel vcpu_unlock(vcpu); 1510284900Sneel *retu = true; 1511284900Sneel return (0); 1512284900Sneel} 1513284900Sneel 1514221828Sgrehanint 1515268935Sjhbvm_suspend(struct vm *vm, enum vm_suspend_how how) 1516268935Sjhb{ 1517268935Sjhb int i; 1518268935Sjhb 1519268935Sjhb if (how <= VM_SUSPEND_NONE || how >= VM_SUSPEND_LAST) 1520268935Sjhb return (EINVAL); 1521268935Sjhb 1522268935Sjhb if (atomic_cmpset_int(&vm->suspend, 0, how) == 0) { 1523268935Sjhb VM_CTR2(vm, "virtual machine already suspended %d/%d", 1524268935Sjhb vm->suspend, how); 1525268935Sjhb return (EALREADY); 1526268935Sjhb } 1527268935Sjhb 1528268935Sjhb VM_CTR1(vm, "virtual machine successfully suspended %d", how); 1529268935Sjhb 1530268935Sjhb /* 1531268935Sjhb * Notify all active vcpus that they are now suspended. 1532268935Sjhb */ 1533268935Sjhb for (i = 0; i < VM_MAXCPU; i++) { 1534268935Sjhb if (CPU_ISSET(i, &vm->active_cpus)) 1535268935Sjhb vcpu_notify_event(vm, i, false); 1536268935Sjhb } 1537268935Sjhb 1538268935Sjhb return (0); 1539268935Sjhb} 1540268935Sjhb 1541268935Sjhbvoid 1542268935Sjhbvm_exit_suspended(struct vm *vm, int vcpuid, uint64_t rip) 1543268935Sjhb{ 1544268935Sjhb struct vm_exit *vmexit; 1545268935Sjhb 1546268935Sjhb KASSERT(vm->suspend > VM_SUSPEND_NONE && vm->suspend < VM_SUSPEND_LAST, 1547268935Sjhb ("vm_exit_suspended: invalid suspend type %d", vm->suspend)); 1548268935Sjhb 1549268935Sjhb vmexit = vm_exitinfo(vm, vcpuid); 1550268935Sjhb vmexit->rip = rip; 1551268935Sjhb vmexit->inst_length = 0; 1552268935Sjhb vmexit->exitcode = VM_EXITCODE_SUSPENDED; 1553268935Sjhb vmexit->u.suspended.how = vm->suspend; 1554268935Sjhb} 1555268935Sjhb 1556270074Sgrehanvoid 1557270074Sgrehanvm_exit_rendezvous(struct vm *vm, int vcpuid, uint64_t rip) 1558270074Sgrehan{ 1559270074Sgrehan struct vm_exit *vmexit; 1560270074Sgrehan 1561270074Sgrehan KASSERT(vm->rendezvous_func != NULL, ("rendezvous not in progress")); 1562270074Sgrehan 1563270074Sgrehan vmexit = vm_exitinfo(vm, vcpuid); 1564270074Sgrehan vmexit->rip = rip; 1565270074Sgrehan vmexit->inst_length = 0; 1566270074Sgrehan vmexit->exitcode = VM_EXITCODE_RENDEZVOUS; 1567270074Sgrehan vmm_stat_incr(vm, vcpuid, VMEXIT_RENDEZVOUS, 1); 1568270074Sgrehan} 1569270074Sgrehan 1570270074Sgrehanvoid 1571284900Sneelvm_exit_reqidle(struct vm *vm, int vcpuid, uint64_t rip) 1572284900Sneel{ 1573284900Sneel struct vm_exit *vmexit; 1574284900Sneel 1575284900Sneel vmexit = vm_exitinfo(vm, vcpuid); 1576284900Sneel vmexit->rip = rip; 1577284900Sneel vmexit->inst_length = 0; 1578284900Sneel vmexit->exitcode = VM_EXITCODE_REQIDLE; 1579284900Sneel vmm_stat_incr(vm, vcpuid, VMEXIT_REQIDLE, 1); 1580284900Sneel} 1581284900Sneel 1582284900Sneelvoid 1583270074Sgrehanvm_exit_astpending(struct vm *vm, int vcpuid, uint64_t rip) 1584270074Sgrehan{ 1585270074Sgrehan struct vm_exit *vmexit; 1586270074Sgrehan 1587270074Sgrehan vmexit = vm_exitinfo(vm, vcpuid); 1588270074Sgrehan vmexit->rip = rip; 1589270074Sgrehan vmexit->inst_length = 0; 1590270074Sgrehan vmexit->exitcode = VM_EXITCODE_BOGUS; 1591270074Sgrehan vmm_stat_incr(vm, vcpuid, VMEXIT_ASTPENDING, 1); 1592270074Sgrehan} 1593270074Sgrehan 1594268935Sjhbint 1595221828Sgrehanvm_run(struct vm *vm, struct vm_run *vmrun) 1596221828Sgrehan{ 1597284900Sneel struct vm_eventinfo evinfo; 1598256072Sneel int error, vcpuid; 1599221828Sgrehan struct vcpu *vcpu; 1600221828Sgrehan struct pcb *pcb; 1601284894Sneel uint64_t tscval; 1602242065Sneel struct vm_exit *vme; 1603262350Sjhb bool retu, intr_disabled; 1604256072Sneel pmap_t pmap; 1605221828Sgrehan 1606221828Sgrehan vcpuid = vmrun->cpuid; 1607221828Sgrehan 1608221828Sgrehan if (vcpuid < 0 || vcpuid >= VM_MAXCPU) 1609221828Sgrehan return (EINVAL); 1610221828Sgrehan 1611270070Sgrehan if (!CPU_ISSET(vcpuid, &vm->active_cpus)) 1612270070Sgrehan return (EINVAL); 1613270070Sgrehan 1614270070Sgrehan if (CPU_ISSET(vcpuid, &vm->suspended_cpus)) 1615270070Sgrehan return (EINVAL); 1616270070Sgrehan 1617256072Sneel pmap = vmspace_pmap(vm->vmspace); 1618221828Sgrehan vcpu = &vm->vcpu[vcpuid]; 1619256072Sneel vme = &vcpu->exitinfo; 1620284900Sneel evinfo.rptr = &vm->rendezvous_func; 1621284900Sneel evinfo.sptr = &vm->suspend; 1622284900Sneel evinfo.iptr = &vcpu->reqidle; 1623242065Sneelrestart: 1624221828Sgrehan critical_enter(); 1625221828Sgrehan 1626256072Sneel KASSERT(!CPU_ISSET(curcpu, &pmap->pm_active), 1627256072Sneel ("vm_run: absurd pm_active")); 1628256072Sneel 1629221828Sgrehan tscval = rdtsc(); 1630221828Sgrehan 1631221828Sgrehan pcb = PCPU_GET(curpcb); 1632221914Sjhb set_pcb_flags(pcb, PCB_FULL_IRET); 1633221828Sgrehan 1634221828Sgrehan restore_guest_fpustate(vcpu); 1635241489Sneel 1636256072Sneel vcpu_require_state(vm, vcpuid, VCPU_RUNNING); 1637284900Sneel error = VMRUN(vm->cookie, vcpuid, vcpu->nextrip, pmap, &evinfo); 1638256072Sneel vcpu_require_state(vm, vcpuid, VCPU_FROZEN); 1639241489Sneel 1640221828Sgrehan save_guest_fpustate(vcpu); 1641221828Sgrehan 1642221828Sgrehan vmm_stat_incr(vm, vcpuid, VCPU_TOTAL_RUNTIME, rdtsc() - tscval); 1643221828Sgrehan 1644221828Sgrehan critical_exit(); 1645221828Sgrehan 1646256072Sneel if (error == 0) { 1647262350Sjhb retu = false; 1648284894Sneel vcpu->nextrip = vme->rip + vme->inst_length; 1649256072Sneel switch (vme->exitcode) { 1650284900Sneel case VM_EXITCODE_REQIDLE: 1651284900Sneel error = vm_handle_reqidle(vm, vcpuid, &retu); 1652284900Sneel break; 1653268935Sjhb case VM_EXITCODE_SUSPENDED: 1654268935Sjhb error = vm_handle_suspend(vm, vcpuid, &retu); 1655268935Sjhb break; 1656266339Sjhb case VM_EXITCODE_IOAPIC_EOI: 1657266339Sjhb vioapic_process_eoi(vm, vcpuid, 1658266339Sjhb vme->u.ioapic_eoi.vector); 1659266339Sjhb break; 1660266339Sjhb case VM_EXITCODE_RENDEZVOUS: 1661266339Sjhb vm_handle_rendezvous(vm, vcpuid); 1662266339Sjhb error = 0; 1663266339Sjhb break; 1664256072Sneel case VM_EXITCODE_HLT: 1665262350Sjhb intr_disabled = ((vme->u.hlt.rflags & PSL_I) == 0); 1666262350Sjhb error = vm_handle_hlt(vm, vcpuid, intr_disabled, &retu); 1667256072Sneel break; 1668256072Sneel case VM_EXITCODE_PAGING: 1669256072Sneel error = vm_handle_paging(vm, vcpuid, &retu); 1670256072Sneel break; 1671256072Sneel case VM_EXITCODE_INST_EMUL: 1672256072Sneel error = vm_handle_inst_emul(vm, vcpuid, &retu); 1673256072Sneel break; 1674268976Sjhb case VM_EXITCODE_INOUT: 1675268976Sjhb case VM_EXITCODE_INOUT_STR: 1676268976Sjhb error = vm_handle_inout(vm, vcpuid, vme, &retu); 1677268976Sjhb break; 1678276349Sneel case VM_EXITCODE_MONITOR: 1679276349Sneel case VM_EXITCODE_MWAIT: 1680276349Sneel vm_inject_ud(vm, vcpuid); 1681276349Sneel break; 1682256072Sneel default: 1683262350Sjhb retu = true; /* handled in userland */ 1684256072Sneel break; 1685242065Sneel } 1686256072Sneel } 1687242065Sneel 1688284894Sneel if (error == 0 && retu == false) 1689242065Sneel goto restart; 1690242065Sneel 1691284900Sneel VCPU_CTR2(vm, vcpuid, "retu %d/%d", error, vme->exitcode); 1692284900Sneel 1693256072Sneel /* copy the exit information */ 1694256072Sneel bcopy(vme, &vmrun->vm_exit, sizeof(struct vm_exit)); 1695221828Sgrehan return (error); 1696221828Sgrehan} 1697221828Sgrehan 1698221828Sgrehanint 1699284894Sneelvm_restart_instruction(void *arg, int vcpuid) 1700284894Sneel{ 1701284894Sneel struct vm *vm; 1702284894Sneel struct vcpu *vcpu; 1703284894Sneel enum vcpu_state state; 1704284894Sneel uint64_t rip; 1705284894Sneel int error; 1706284894Sneel 1707284894Sneel vm = arg; 1708284894Sneel if (vcpuid < 0 || vcpuid >= VM_MAXCPU) 1709284894Sneel return (EINVAL); 1710284894Sneel 1711284894Sneel vcpu = &vm->vcpu[vcpuid]; 1712284894Sneel state = vcpu_get_state(vm, vcpuid, NULL); 1713284894Sneel if (state == VCPU_RUNNING) { 1714284894Sneel /* 1715284894Sneel * When a vcpu is "running" the next instruction is determined 1716284894Sneel * by adding 'rip' and 'inst_length' in the vcpu's 'exitinfo'. 1717284894Sneel * Thus setting 'inst_length' to zero will cause the current 1718284894Sneel * instruction to be restarted. 1719284894Sneel */ 1720284894Sneel vcpu->exitinfo.inst_length = 0; 1721284894Sneel VCPU_CTR1(vm, vcpuid, "restarting instruction at %#lx by " 1722284894Sneel "setting inst_length to zero", vcpu->exitinfo.rip); 1723284894Sneel } else if (state == VCPU_FROZEN) { 1724284894Sneel /* 1725284894Sneel * When a vcpu is "frozen" it is outside the critical section 1726284894Sneel * around VMRUN() and 'nextrip' points to the next instruction. 1727284894Sneel * Thus instruction restart is achieved by setting 'nextrip' 1728284894Sneel * to the vcpu's %rip. 1729284894Sneel */ 1730284894Sneel error = vm_get_register(vm, vcpuid, VM_REG_GUEST_RIP, &rip); 1731284894Sneel KASSERT(!error, ("%s: error %d getting rip", __func__, error)); 1732284894Sneel VCPU_CTR2(vm, vcpuid, "restarting instruction by updating " 1733284894Sneel "nextrip from %#lx to %#lx", vcpu->nextrip, rip); 1734284894Sneel vcpu->nextrip = rip; 1735284894Sneel } else { 1736284894Sneel panic("%s: invalid state %d", __func__, state); 1737284894Sneel } 1738284894Sneel return (0); 1739284894Sneel} 1740284894Sneel 1741284894Sneelint 1742270159Sgrehanvm_exit_intinfo(struct vm *vm, int vcpuid, uint64_t info) 1743270159Sgrehan{ 1744270159Sgrehan struct vcpu *vcpu; 1745270159Sgrehan int type, vector; 1746270159Sgrehan 1747270159Sgrehan if (vcpuid < 0 || vcpuid >= VM_MAXCPU) 1748270159Sgrehan return (EINVAL); 1749270159Sgrehan 1750270159Sgrehan vcpu = &vm->vcpu[vcpuid]; 1751270159Sgrehan 1752270159Sgrehan if (info & VM_INTINFO_VALID) { 1753270159Sgrehan type = info & VM_INTINFO_TYPE; 1754270159Sgrehan vector = info & 0xff; 1755270159Sgrehan if (type == VM_INTINFO_NMI && vector != IDT_NMI) 1756270159Sgrehan return (EINVAL); 1757270159Sgrehan if (type == VM_INTINFO_HWEXCEPTION && vector >= 32) 1758270159Sgrehan return (EINVAL); 1759270159Sgrehan if (info & VM_INTINFO_RSVD) 1760270159Sgrehan return (EINVAL); 1761270159Sgrehan } else { 1762270159Sgrehan info = 0; 1763270159Sgrehan } 1764270159Sgrehan VCPU_CTR2(vm, vcpuid, "%s: info1(%#lx)", __func__, info); 1765270159Sgrehan vcpu->exitintinfo = info; 1766270159Sgrehan return (0); 1767270159Sgrehan} 1768270159Sgrehan 1769270159Sgrehanenum exc_class { 1770270159Sgrehan EXC_BENIGN, 1771270159Sgrehan EXC_CONTRIBUTORY, 1772270159Sgrehan EXC_PAGEFAULT 1773270159Sgrehan}; 1774270159Sgrehan 1775270159Sgrehan#define IDT_VE 20 /* Virtualization Exception (Intel specific) */ 1776270159Sgrehan 1777270159Sgrehanstatic enum exc_class 1778270159Sgrehanexception_class(uint64_t info) 1779270159Sgrehan{ 1780270159Sgrehan int type, vector; 1781270159Sgrehan 1782270159Sgrehan KASSERT(info & VM_INTINFO_VALID, ("intinfo must be valid: %#lx", info)); 1783270159Sgrehan type = info & VM_INTINFO_TYPE; 1784270159Sgrehan vector = info & 0xff; 1785270159Sgrehan 1786270159Sgrehan /* Table 6-4, "Interrupt and Exception Classes", Intel SDM, Vol 3 */ 1787270159Sgrehan switch (type) { 1788270159Sgrehan case VM_INTINFO_HWINTR: 1789270159Sgrehan case VM_INTINFO_SWINTR: 1790270159Sgrehan case VM_INTINFO_NMI: 1791270159Sgrehan return (EXC_BENIGN); 1792270159Sgrehan default: 1793270159Sgrehan /* 1794270159Sgrehan * Hardware exception. 1795270159Sgrehan * 1796270159Sgrehan * SVM and VT-x use identical type values to represent NMI, 1797270159Sgrehan * hardware interrupt and software interrupt. 1798270159Sgrehan * 1799270159Sgrehan * SVM uses type '3' for all exceptions. VT-x uses type '3' 1800270159Sgrehan * for exceptions except #BP and #OF. #BP and #OF use a type 1801270159Sgrehan * value of '5' or '6'. Therefore we don't check for explicit 1802270159Sgrehan * values of 'type' to classify 'intinfo' into a hardware 1803270159Sgrehan * exception. 1804270159Sgrehan */ 1805270159Sgrehan break; 1806270159Sgrehan } 1807270159Sgrehan 1808270159Sgrehan switch (vector) { 1809270159Sgrehan case IDT_PF: 1810270159Sgrehan case IDT_VE: 1811270159Sgrehan return (EXC_PAGEFAULT); 1812270159Sgrehan case IDT_DE: 1813270159Sgrehan case IDT_TS: 1814270159Sgrehan case IDT_NP: 1815270159Sgrehan case IDT_SS: 1816270159Sgrehan case IDT_GP: 1817270159Sgrehan return (EXC_CONTRIBUTORY); 1818270159Sgrehan default: 1819270159Sgrehan return (EXC_BENIGN); 1820270159Sgrehan } 1821270159Sgrehan} 1822270159Sgrehan 1823270159Sgrehanstatic int 1824270159Sgrehannested_fault(struct vm *vm, int vcpuid, uint64_t info1, uint64_t info2, 1825270159Sgrehan uint64_t *retinfo) 1826270159Sgrehan{ 1827270159Sgrehan enum exc_class exc1, exc2; 1828270159Sgrehan int type1, vector1; 1829270159Sgrehan 1830270159Sgrehan KASSERT(info1 & VM_INTINFO_VALID, ("info1 %#lx is not valid", info1)); 1831270159Sgrehan KASSERT(info2 & VM_INTINFO_VALID, ("info2 %#lx is not valid", info2)); 1832270159Sgrehan 1833270159Sgrehan /* 1834270159Sgrehan * If an exception occurs while attempting to call the double-fault 1835270159Sgrehan * handler the processor enters shutdown mode (aka triple fault). 1836270159Sgrehan */ 1837270159Sgrehan type1 = info1 & VM_INTINFO_TYPE; 1838270159Sgrehan vector1 = info1 & 0xff; 1839270159Sgrehan if (type1 == VM_INTINFO_HWEXCEPTION && vector1 == IDT_DF) { 1840270159Sgrehan VCPU_CTR2(vm, vcpuid, "triple fault: info1(%#lx), info2(%#lx)", 1841270159Sgrehan info1, info2); 1842270159Sgrehan vm_suspend(vm, VM_SUSPEND_TRIPLEFAULT); 1843270159Sgrehan *retinfo = 0; 1844270159Sgrehan return (0); 1845270159Sgrehan } 1846270159Sgrehan 1847270159Sgrehan /* 1848270159Sgrehan * Table 6-5 "Conditions for Generating a Double Fault", Intel SDM, Vol3 1849270159Sgrehan */ 1850270159Sgrehan exc1 = exception_class(info1); 1851270159Sgrehan exc2 = exception_class(info2); 1852270159Sgrehan if ((exc1 == EXC_CONTRIBUTORY && exc2 == EXC_CONTRIBUTORY) || 1853270159Sgrehan (exc1 == EXC_PAGEFAULT && exc2 != EXC_BENIGN)) { 1854270159Sgrehan /* Convert nested fault into a double fault. */ 1855270159Sgrehan *retinfo = IDT_DF; 1856270159Sgrehan *retinfo |= VM_INTINFO_VALID | VM_INTINFO_HWEXCEPTION; 1857270159Sgrehan *retinfo |= VM_INTINFO_DEL_ERRCODE; 1858270159Sgrehan } else { 1859270159Sgrehan /* Handle exceptions serially */ 1860270159Sgrehan *retinfo = info2; 1861270159Sgrehan } 1862270159Sgrehan return (1); 1863270159Sgrehan} 1864270159Sgrehan 1865270159Sgrehanstatic uint64_t 1866270159Sgrehanvcpu_exception_intinfo(struct vcpu *vcpu) 1867270159Sgrehan{ 1868270159Sgrehan uint64_t info = 0; 1869270159Sgrehan 1870270159Sgrehan if (vcpu->exception_pending) { 1871284894Sneel info = vcpu->exc_vector & 0xff; 1872270159Sgrehan info |= VM_INTINFO_VALID | VM_INTINFO_HWEXCEPTION; 1873284894Sneel if (vcpu->exc_errcode_valid) { 1874270159Sgrehan info |= VM_INTINFO_DEL_ERRCODE; 1875284894Sneel info |= (uint64_t)vcpu->exc_errcode << 32; 1876270159Sgrehan } 1877270159Sgrehan } 1878270159Sgrehan return (info); 1879270159Sgrehan} 1880270159Sgrehan 1881270159Sgrehanint 1882270159Sgrehanvm_entry_intinfo(struct vm *vm, int vcpuid, uint64_t *retinfo) 1883270159Sgrehan{ 1884270159Sgrehan struct vcpu *vcpu; 1885270159Sgrehan uint64_t info1, info2; 1886270159Sgrehan int valid; 1887270159Sgrehan 1888270159Sgrehan KASSERT(vcpuid >= 0 && vcpuid < VM_MAXCPU, ("invalid vcpu %d", vcpuid)); 1889270159Sgrehan 1890270159Sgrehan vcpu = &vm->vcpu[vcpuid]; 1891270159Sgrehan 1892270159Sgrehan info1 = vcpu->exitintinfo; 1893270159Sgrehan vcpu->exitintinfo = 0; 1894270159Sgrehan 1895270159Sgrehan info2 = 0; 1896270159Sgrehan if (vcpu->exception_pending) { 1897270159Sgrehan info2 = vcpu_exception_intinfo(vcpu); 1898270159Sgrehan vcpu->exception_pending = 0; 1899270159Sgrehan VCPU_CTR2(vm, vcpuid, "Exception %d delivered: %#lx", 1900284894Sneel vcpu->exc_vector, info2); 1901270159Sgrehan } 1902270159Sgrehan 1903270159Sgrehan if ((info1 & VM_INTINFO_VALID) && (info2 & VM_INTINFO_VALID)) { 1904270159Sgrehan valid = nested_fault(vm, vcpuid, info1, info2, retinfo); 1905270159Sgrehan } else if (info1 & VM_INTINFO_VALID) { 1906270159Sgrehan *retinfo = info1; 1907270159Sgrehan valid = 1; 1908270159Sgrehan } else if (info2 & VM_INTINFO_VALID) { 1909270159Sgrehan *retinfo = info2; 1910270159Sgrehan valid = 1; 1911270159Sgrehan } else { 1912270159Sgrehan valid = 0; 1913270159Sgrehan } 1914270159Sgrehan 1915270159Sgrehan if (valid) { 1916270159Sgrehan VCPU_CTR4(vm, vcpuid, "%s: info1(%#lx), info2(%#lx), " 1917270159Sgrehan "retinfo(%#lx)", __func__, info1, info2, *retinfo); 1918270159Sgrehan } 1919270159Sgrehan 1920270159Sgrehan return (valid); 1921270159Sgrehan} 1922270159Sgrehan 1923270159Sgrehanint 1924270159Sgrehanvm_get_intinfo(struct vm *vm, int vcpuid, uint64_t *info1, uint64_t *info2) 1925270159Sgrehan{ 1926270159Sgrehan struct vcpu *vcpu; 1927270159Sgrehan 1928270159Sgrehan if (vcpuid < 0 || vcpuid >= VM_MAXCPU) 1929270159Sgrehan return (EINVAL); 1930270159Sgrehan 1931270159Sgrehan vcpu = &vm->vcpu[vcpuid]; 1932270159Sgrehan *info1 = vcpu->exitintinfo; 1933270159Sgrehan *info2 = vcpu_exception_intinfo(vcpu); 1934270159Sgrehan return (0); 1935270159Sgrehan} 1936270159Sgrehan 1937270159Sgrehanint 1938284894Sneelvm_inject_exception(struct vm *vm, int vcpuid, int vector, int errcode_valid, 1939284894Sneel uint32_t errcode, int restart_instruction) 1940221828Sgrehan{ 1941267427Sjhb struct vcpu *vcpu; 1942284900Sneel uint64_t regval; 1943284894Sneel int error; 1944267427Sjhb 1945221828Sgrehan if (vcpuid < 0 || vcpuid >= VM_MAXCPU) 1946221828Sgrehan return (EINVAL); 1947221828Sgrehan 1948284894Sneel if (vector < 0 || vector >= 32) 1949221828Sgrehan return (EINVAL); 1950221828Sgrehan 1951270159Sgrehan /* 1952270159Sgrehan * A double fault exception should never be injected directly into 1953270159Sgrehan * the guest. It is a derived exception that results from specific 1954270159Sgrehan * combinations of nested faults. 1955270159Sgrehan */ 1956284894Sneel if (vector == IDT_DF) 1957270159Sgrehan return (EINVAL); 1958270159Sgrehan 1959267427Sjhb vcpu = &vm->vcpu[vcpuid]; 1960221828Sgrehan 1961267427Sjhb if (vcpu->exception_pending) { 1962267427Sjhb VCPU_CTR2(vm, vcpuid, "Unable to inject exception %d due to " 1963284894Sneel "pending exception %d", vector, vcpu->exc_vector); 1964267427Sjhb return (EBUSY); 1965267427Sjhb } 1966267427Sjhb 1967284900Sneel if (errcode_valid) { 1968284900Sneel /* 1969284900Sneel * Exceptions don't deliver an error code in real mode. 1970284900Sneel */ 1971284900Sneel error = vm_get_register(vm, vcpuid, VM_REG_GUEST_CR0, ®val); 1972284900Sneel KASSERT(!error, ("%s: error %d getting CR0", __func__, error)); 1973284900Sneel if (!(regval & CR0_PE)) 1974284900Sneel errcode_valid = 0; 1975284900Sneel } 1976284900Sneel 1977284894Sneel /* 1978284894Sneel * From section 26.6.1 "Interruptibility State" in Intel SDM: 1979284894Sneel * 1980284894Sneel * Event blocking by "STI" or "MOV SS" is cleared after guest executes 1981284894Sneel * one instruction or incurs an exception. 1982284894Sneel */ 1983284894Sneel error = vm_set_register(vm, vcpuid, VM_REG_GUEST_INTR_SHADOW, 0); 1984284894Sneel KASSERT(error == 0, ("%s: error %d clearing interrupt shadow", 1985284894Sneel __func__, error)); 1986284894Sneel 1987284894Sneel if (restart_instruction) 1988284894Sneel vm_restart_instruction(vm, vcpuid); 1989284894Sneel 1990267427Sjhb vcpu->exception_pending = 1; 1991284894Sneel vcpu->exc_vector = vector; 1992284894Sneel vcpu->exc_errcode = errcode; 1993284894Sneel vcpu->exc_errcode_valid = errcode_valid; 1994284894Sneel VCPU_CTR1(vm, vcpuid, "Exception %d pending", vector); 1995267427Sjhb return (0); 1996221828Sgrehan} 1997221828Sgrehan 1998270159Sgrehanvoid 1999270159Sgrehanvm_inject_fault(void *vmarg, int vcpuid, int vector, int errcode_valid, 2000270159Sgrehan int errcode) 2001267427Sjhb{ 2002270159Sgrehan struct vm *vm; 2003284894Sneel int error, restart_instruction; 2004267427Sjhb 2005270159Sgrehan vm = vmarg; 2006284894Sneel restart_instruction = 1; 2007270159Sgrehan 2008284894Sneel error = vm_inject_exception(vm, vcpuid, vector, errcode_valid, 2009284894Sneel errcode, restart_instruction); 2010267427Sjhb KASSERT(error == 0, ("vm_inject_exception error %d", error)); 2011267427Sjhb} 2012267427Sjhb 2013267427Sjhbvoid 2014270159Sgrehanvm_inject_pf(void *vmarg, int vcpuid, int error_code, uint64_t cr2) 2015268976Sjhb{ 2016270159Sgrehan struct vm *vm; 2017268976Sjhb int error; 2018268976Sjhb 2019270159Sgrehan vm = vmarg; 2020268976Sjhb VCPU_CTR2(vm, vcpuid, "Injecting page fault: error_code %#x, cr2 %#lx", 2021268976Sjhb error_code, cr2); 2022268976Sjhb 2023268976Sjhb error = vm_set_register(vm, vcpuid, VM_REG_GUEST_CR2, cr2); 2024268976Sjhb KASSERT(error == 0, ("vm_set_register(cr2) error %d", error)); 2025268976Sjhb 2026270159Sgrehan vm_inject_fault(vm, vcpuid, IDT_PF, 1, error_code); 2027268976Sjhb} 2028268976Sjhb 2029248389Sneelstatic VMM_STAT(VCPU_NMI_COUNT, "number of NMIs delivered to vcpu"); 2030241982Sneel 2031221828Sgrehanint 2032241982Sneelvm_inject_nmi(struct vm *vm, int vcpuid) 2033221828Sgrehan{ 2034241982Sneel struct vcpu *vcpu; 2035221828Sgrehan 2036241982Sneel if (vcpuid < 0 || vcpuid >= VM_MAXCPU) 2037221828Sgrehan return (EINVAL); 2038221828Sgrehan 2039241982Sneel vcpu = &vm->vcpu[vcpuid]; 2040241982Sneel 2041241982Sneel vcpu->nmi_pending = 1; 2042266339Sjhb vcpu_notify_event(vm, vcpuid, false); 2043241982Sneel return (0); 2044221828Sgrehan} 2045221828Sgrehan 2046221828Sgrehanint 2047241982Sneelvm_nmi_pending(struct vm *vm, int vcpuid) 2048241982Sneel{ 2049241982Sneel struct vcpu *vcpu; 2050241982Sneel 2051241982Sneel if (vcpuid < 0 || vcpuid >= VM_MAXCPU) 2052241982Sneel panic("vm_nmi_pending: invalid vcpuid %d", vcpuid); 2053241982Sneel 2054241982Sneel vcpu = &vm->vcpu[vcpuid]; 2055241982Sneel 2056241982Sneel return (vcpu->nmi_pending); 2057241982Sneel} 2058241982Sneel 2059241982Sneelvoid 2060241982Sneelvm_nmi_clear(struct vm *vm, int vcpuid) 2061241982Sneel{ 2062241982Sneel struct vcpu *vcpu; 2063241982Sneel 2064241982Sneel if (vcpuid < 0 || vcpuid >= VM_MAXCPU) 2065241982Sneel panic("vm_nmi_pending: invalid vcpuid %d", vcpuid); 2066241982Sneel 2067241982Sneel vcpu = &vm->vcpu[vcpuid]; 2068241982Sneel 2069241982Sneel if (vcpu->nmi_pending == 0) 2070241982Sneel panic("vm_nmi_clear: inconsistent nmi_pending state"); 2071241982Sneel 2072241982Sneel vcpu->nmi_pending = 0; 2073241982Sneel vmm_stat_incr(vm, vcpuid, VCPU_NMI_COUNT, 1); 2074241982Sneel} 2075241982Sneel 2076268891Sjhbstatic VMM_STAT(VCPU_EXTINT_COUNT, "number of ExtINTs delivered to vcpu"); 2077268891Sjhb 2078241982Sneelint 2079268891Sjhbvm_inject_extint(struct vm *vm, int vcpuid) 2080268891Sjhb{ 2081268891Sjhb struct vcpu *vcpu; 2082268891Sjhb 2083268891Sjhb if (vcpuid < 0 || vcpuid >= VM_MAXCPU) 2084268891Sjhb return (EINVAL); 2085268891Sjhb 2086268891Sjhb vcpu = &vm->vcpu[vcpuid]; 2087268891Sjhb 2088268891Sjhb vcpu->extint_pending = 1; 2089268891Sjhb vcpu_notify_event(vm, vcpuid, false); 2090268891Sjhb return (0); 2091268891Sjhb} 2092268891Sjhb 2093268891Sjhbint 2094268891Sjhbvm_extint_pending(struct vm *vm, int vcpuid) 2095268891Sjhb{ 2096268891Sjhb struct vcpu *vcpu; 2097268891Sjhb 2098268891Sjhb if (vcpuid < 0 || vcpuid >= VM_MAXCPU) 2099268891Sjhb panic("vm_extint_pending: invalid vcpuid %d", vcpuid); 2100268891Sjhb 2101268891Sjhb vcpu = &vm->vcpu[vcpuid]; 2102268891Sjhb 2103268891Sjhb return (vcpu->extint_pending); 2104268891Sjhb} 2105268891Sjhb 2106268891Sjhbvoid 2107268891Sjhbvm_extint_clear(struct vm *vm, int vcpuid) 2108268891Sjhb{ 2109268891Sjhb struct vcpu *vcpu; 2110268891Sjhb 2111268891Sjhb if (vcpuid < 0 || vcpuid >= VM_MAXCPU) 2112268891Sjhb panic("vm_extint_pending: invalid vcpuid %d", vcpuid); 2113268891Sjhb 2114268891Sjhb vcpu = &vm->vcpu[vcpuid]; 2115268891Sjhb 2116268891Sjhb if (vcpu->extint_pending == 0) 2117268891Sjhb panic("vm_extint_clear: inconsistent extint_pending state"); 2118268891Sjhb 2119268891Sjhb vcpu->extint_pending = 0; 2120268891Sjhb vmm_stat_incr(vm, vcpuid, VCPU_EXTINT_COUNT, 1); 2121268891Sjhb} 2122268891Sjhb 2123268891Sjhbint 2124221828Sgrehanvm_get_capability(struct vm *vm, int vcpu, int type, int *retval) 2125221828Sgrehan{ 2126221828Sgrehan if (vcpu < 0 || vcpu >= VM_MAXCPU) 2127221828Sgrehan return (EINVAL); 2128221828Sgrehan 2129221828Sgrehan if (type < 0 || type >= VM_CAP_MAX) 2130221828Sgrehan return (EINVAL); 2131221828Sgrehan 2132221828Sgrehan return (VMGETCAP(vm->cookie, vcpu, type, retval)); 2133221828Sgrehan} 2134221828Sgrehan 2135221828Sgrehanint 2136221828Sgrehanvm_set_capability(struct vm *vm, int vcpu, int type, int val) 2137221828Sgrehan{ 2138221828Sgrehan if (vcpu < 0 || vcpu >= VM_MAXCPU) 2139221828Sgrehan return (EINVAL); 2140221828Sgrehan 2141221828Sgrehan if (type < 0 || type >= VM_CAP_MAX) 2142221828Sgrehan return (EINVAL); 2143221828Sgrehan 2144221828Sgrehan return (VMSETCAP(vm->cookie, vcpu, type, val)); 2145221828Sgrehan} 2146221828Sgrehan 2147221828Sgrehanstruct vlapic * 2148221828Sgrehanvm_lapic(struct vm *vm, int cpu) 2149221828Sgrehan{ 2150221828Sgrehan return (vm->vcpu[cpu].vlapic); 2151221828Sgrehan} 2152221828Sgrehan 2153261088Sjhbstruct vioapic * 2154261088Sjhbvm_ioapic(struct vm *vm) 2155261088Sjhb{ 2156261088Sjhb 2157261088Sjhb return (vm->vioapic); 2158261088Sjhb} 2159261088Sjhb 2160261088Sjhbstruct vhpet * 2161261088Sjhbvm_hpet(struct vm *vm) 2162261088Sjhb{ 2163261088Sjhb 2164261088Sjhb return (vm->vhpet); 2165261088Sjhb} 2166261088Sjhb 2167221828Sgrehanboolean_t 2168221828Sgrehanvmm_is_pptdev(int bus, int slot, int func) 2169221828Sgrehan{ 2170246188Sneel int found, i, n; 2171246188Sneel int b, s, f; 2172221828Sgrehan char *val, *cp, *cp2; 2173221828Sgrehan 2174221828Sgrehan /* 2175246188Sneel * XXX 2176246188Sneel * The length of an environment variable is limited to 128 bytes which 2177246188Sneel * puts an upper limit on the number of passthru devices that may be 2178246188Sneel * specified using a single environment variable. 2179246188Sneel * 2180246188Sneel * Work around this by scanning multiple environment variable 2181246188Sneel * names instead of a single one - yuck! 2182221828Sgrehan */ 2183246188Sneel const char *names[] = { "pptdevs", "pptdevs2", "pptdevs3", NULL }; 2184246188Sneel 2185246188Sneel /* set pptdevs="1/2/3 4/5/6 7/8/9 10/11/12" */ 2186221828Sgrehan found = 0; 2187246188Sneel for (i = 0; names[i] != NULL && !found; i++) { 2188246188Sneel cp = val = getenv(names[i]); 2189246188Sneel while (cp != NULL && *cp != '\0') { 2190246188Sneel if ((cp2 = strchr(cp, ' ')) != NULL) 2191246188Sneel *cp2 = '\0'; 2192221828Sgrehan 2193246188Sneel n = sscanf(cp, "%d/%d/%d", &b, &s, &f); 2194246188Sneel if (n == 3 && bus == b && slot == s && func == f) { 2195246188Sneel found = 1; 2196246188Sneel break; 2197246188Sneel } 2198221828Sgrehan 2199246188Sneel if (cp2 != NULL) 2200246188Sneel *cp2++ = ' '; 2201221828Sgrehan 2202246188Sneel cp = cp2; 2203246188Sneel } 2204246188Sneel freeenv(val); 2205221828Sgrehan } 2206221828Sgrehan return (found); 2207221828Sgrehan} 2208221828Sgrehan 2209221828Sgrehanvoid * 2210221828Sgrehanvm_iommu_domain(struct vm *vm) 2211221828Sgrehan{ 2212221828Sgrehan 2213221828Sgrehan return (vm->iommu); 2214221828Sgrehan} 2215221828Sgrehan 2216241489Sneelint 2217266393Sjhbvcpu_set_state(struct vm *vm, int vcpuid, enum vcpu_state newstate, 2218266393Sjhb bool from_idle) 2219221828Sgrehan{ 2220241489Sneel int error; 2221221828Sgrehan struct vcpu *vcpu; 2222221828Sgrehan 2223221828Sgrehan if (vcpuid < 0 || vcpuid >= VM_MAXCPU) 2224221828Sgrehan panic("vm_set_run_state: invalid vcpuid %d", vcpuid); 2225221828Sgrehan 2226221828Sgrehan vcpu = &vm->vcpu[vcpuid]; 2227221828Sgrehan 2228241489Sneel vcpu_lock(vcpu); 2229284900Sneel error = vcpu_set_state_locked(vm, vcpuid, newstate, from_idle); 2230241489Sneel vcpu_unlock(vcpu); 2231241489Sneel 2232241489Sneel return (error); 2233221828Sgrehan} 2234221828Sgrehan 2235241489Sneelenum vcpu_state 2236249879Sgrehanvcpu_get_state(struct vm *vm, int vcpuid, int *hostcpu) 2237221828Sgrehan{ 2238221828Sgrehan struct vcpu *vcpu; 2239241489Sneel enum vcpu_state state; 2240221828Sgrehan 2241221828Sgrehan if (vcpuid < 0 || vcpuid >= VM_MAXCPU) 2242221828Sgrehan panic("vm_get_run_state: invalid vcpuid %d", vcpuid); 2243221828Sgrehan 2244221828Sgrehan vcpu = &vm->vcpu[vcpuid]; 2245221828Sgrehan 2246241489Sneel vcpu_lock(vcpu); 2247241489Sneel state = vcpu->state; 2248249879Sgrehan if (hostcpu != NULL) 2249249879Sgrehan *hostcpu = vcpu->hostcpu; 2250241489Sneel vcpu_unlock(vcpu); 2251221828Sgrehan 2252241489Sneel return (state); 2253221828Sgrehan} 2254221828Sgrehan 2255270070Sgrehanint 2256221828Sgrehanvm_activate_cpu(struct vm *vm, int vcpuid) 2257221828Sgrehan{ 2258221828Sgrehan 2259270070Sgrehan if (vcpuid < 0 || vcpuid >= VM_MAXCPU) 2260270070Sgrehan return (EINVAL); 2261266339Sjhb 2262270070Sgrehan if (CPU_ISSET(vcpuid, &vm->active_cpus)) 2263270070Sgrehan return (EBUSY); 2264270070Sgrehan 2265266339Sjhb VCPU_CTR0(vm, vcpuid, "activated"); 2266266339Sjhb CPU_SET_ATOMIC(vcpuid, &vm->active_cpus); 2267270070Sgrehan return (0); 2268221828Sgrehan} 2269221828Sgrehan 2270223621Sgrehancpuset_t 2271221828Sgrehanvm_active_cpus(struct vm *vm) 2272221828Sgrehan{ 2273221828Sgrehan 2274221828Sgrehan return (vm->active_cpus); 2275221828Sgrehan} 2276221828Sgrehan 2277270070Sgrehancpuset_t 2278270070Sgrehanvm_suspended_cpus(struct vm *vm) 2279270070Sgrehan{ 2280270070Sgrehan 2281270070Sgrehan return (vm->suspended_cpus); 2282270070Sgrehan} 2283270070Sgrehan 2284221828Sgrehanvoid * 2285221828Sgrehanvcpu_stats(struct vm *vm, int vcpuid) 2286221828Sgrehan{ 2287221828Sgrehan 2288221828Sgrehan return (vm->vcpu[vcpuid].stats); 2289221828Sgrehan} 2290240922Sneel 2291240922Sneelint 2292240922Sneelvm_get_x2apic_state(struct vm *vm, int vcpuid, enum x2apic_state *state) 2293240922Sneel{ 2294240922Sneel if (vcpuid < 0 || vcpuid >= VM_MAXCPU) 2295240922Sneel return (EINVAL); 2296240922Sneel 2297240922Sneel *state = vm->vcpu[vcpuid].x2apic_state; 2298240922Sneel 2299240922Sneel return (0); 2300240922Sneel} 2301240922Sneel 2302240922Sneelint 2303240922Sneelvm_set_x2apic_state(struct vm *vm, int vcpuid, enum x2apic_state state) 2304240922Sneel{ 2305240922Sneel if (vcpuid < 0 || vcpuid >= VM_MAXCPU) 2306240922Sneel return (EINVAL); 2307240922Sneel 2308248392Sneel if (state >= X2APIC_STATE_LAST) 2309240922Sneel return (EINVAL); 2310240922Sneel 2311240922Sneel vm->vcpu[vcpuid].x2apic_state = state; 2312240922Sneel 2313240943Sneel vlapic_set_x2apic_state(vm, vcpuid, state); 2314240943Sneel 2315240922Sneel return (0); 2316240922Sneel} 2317241489Sneel 2318262350Sjhb/* 2319262350Sjhb * This function is called to ensure that a vcpu "sees" a pending event 2320262350Sjhb * as soon as possible: 2321262350Sjhb * - If the vcpu thread is sleeping then it is woken up. 2322262350Sjhb * - If the vcpu is running on a different host_cpu then an IPI will be directed 2323262350Sjhb * to the host_cpu to cause the vcpu to trap into the hypervisor. 2324262350Sjhb */ 2325284900Sneelstatic void 2326284900Sneelvcpu_notify_event_locked(struct vcpu *vcpu, bool lapic_intr) 2327241489Sneel{ 2328241489Sneel int hostcpu; 2329241489Sneel 2330241489Sneel hostcpu = vcpu->hostcpu; 2331266393Sjhb if (vcpu->state == VCPU_RUNNING) { 2332266393Sjhb KASSERT(hostcpu != NOCPU, ("vcpu running on invalid hostcpu")); 2333266339Sjhb if (hostcpu != curcpu) { 2334266393Sjhb if (lapic_intr) { 2335266339Sjhb vlapic_post_intr(vcpu->vlapic, hostcpu, 2336266339Sjhb vmm_ipinum); 2337266393Sjhb } else { 2338266339Sjhb ipi_cpu(hostcpu, vmm_ipinum); 2339266393Sjhb } 2340266393Sjhb } else { 2341266393Sjhb /* 2342266393Sjhb * If the 'vcpu' is running on 'curcpu' then it must 2343266393Sjhb * be sending a notification to itself (e.g. SELF_IPI). 2344266393Sjhb * The pending event will be picked up when the vcpu 2345266393Sjhb * transitions back to guest context. 2346266393Sjhb */ 2347266339Sjhb } 2348266393Sjhb } else { 2349266393Sjhb KASSERT(hostcpu == NOCPU, ("vcpu state %d not consistent " 2350266393Sjhb "with hostcpu %d", vcpu->state, hostcpu)); 2351266393Sjhb if (vcpu->state == VCPU_SLEEPING) 2352266393Sjhb wakeup_one(vcpu); 2353242065Sneel } 2354284900Sneel} 2355284900Sneel 2356284900Sneelvoid 2357284900Sneelvcpu_notify_event(struct vm *vm, int vcpuid, bool lapic_intr) 2358284900Sneel{ 2359284900Sneel struct vcpu *vcpu = &vm->vcpu[vcpuid]; 2360284900Sneel 2361284900Sneel vcpu_lock(vcpu); 2362284900Sneel vcpu_notify_event_locked(vcpu, lapic_intr); 2363242065Sneel vcpu_unlock(vcpu); 2364241489Sneel} 2365256072Sneel 2366256072Sneelstruct vmspace * 2367256072Sneelvm_get_vmspace(struct vm *vm) 2368256072Sneel{ 2369256072Sneel 2370256072Sneel return (vm->vmspace); 2371256072Sneel} 2372261088Sjhb 2373261088Sjhbint 2374261088Sjhbvm_apicid2vcpuid(struct vm *vm, int apicid) 2375261088Sjhb{ 2376261088Sjhb /* 2377261088Sjhb * XXX apic id is assumed to be numerically identical to vcpu id 2378261088Sjhb */ 2379261088Sjhb return (apicid); 2380261088Sjhb} 2381266339Sjhb 2382266339Sjhbvoid 2383266339Sjhbvm_smp_rendezvous(struct vm *vm, int vcpuid, cpuset_t dest, 2384266339Sjhb vm_rendezvous_func_t func, void *arg) 2385266339Sjhb{ 2386266339Sjhb int i; 2387266339Sjhb 2388266339Sjhb /* 2389266339Sjhb * Enforce that this function is called without any locks 2390266339Sjhb */ 2391266339Sjhb WITNESS_WARN(WARN_PANIC, NULL, "vm_smp_rendezvous"); 2392266339Sjhb KASSERT(vcpuid == -1 || (vcpuid >= 0 && vcpuid < VM_MAXCPU), 2393266339Sjhb ("vm_smp_rendezvous: invalid vcpuid %d", vcpuid)); 2394266339Sjhb 2395266339Sjhbrestart: 2396266339Sjhb mtx_lock(&vm->rendezvous_mtx); 2397266339Sjhb if (vm->rendezvous_func != NULL) { 2398266339Sjhb /* 2399266339Sjhb * If a rendezvous is already in progress then we need to 2400266339Sjhb * call the rendezvous handler in case this 'vcpuid' is one 2401266339Sjhb * of the targets of the rendezvous. 2402266339Sjhb */ 2403266339Sjhb RENDEZVOUS_CTR0(vm, vcpuid, "Rendezvous already in progress"); 2404266339Sjhb mtx_unlock(&vm->rendezvous_mtx); 2405266339Sjhb vm_handle_rendezvous(vm, vcpuid); 2406266339Sjhb goto restart; 2407266339Sjhb } 2408266339Sjhb KASSERT(vm->rendezvous_func == NULL, ("vm_smp_rendezvous: previous " 2409266339Sjhb "rendezvous is still in progress")); 2410266339Sjhb 2411266339Sjhb RENDEZVOUS_CTR0(vm, vcpuid, "Initiating rendezvous"); 2412266339Sjhb vm->rendezvous_req_cpus = dest; 2413266339Sjhb CPU_ZERO(&vm->rendezvous_done_cpus); 2414266339Sjhb vm->rendezvous_arg = arg; 2415266339Sjhb vm_set_rendezvous_func(vm, func); 2416266339Sjhb mtx_unlock(&vm->rendezvous_mtx); 2417266339Sjhb 2418266339Sjhb /* 2419266339Sjhb * Wake up any sleeping vcpus and trigger a VM-exit in any running 2420266339Sjhb * vcpus so they handle the rendezvous as soon as possible. 2421266339Sjhb */ 2422266339Sjhb for (i = 0; i < VM_MAXCPU; i++) { 2423266339Sjhb if (CPU_ISSET(i, &dest)) 2424266339Sjhb vcpu_notify_event(vm, i, false); 2425266339Sjhb } 2426266339Sjhb 2427266339Sjhb vm_handle_rendezvous(vm, vcpuid); 2428266339Sjhb} 2429268891Sjhb 2430268891Sjhbstruct vatpic * 2431268891Sjhbvm_atpic(struct vm *vm) 2432268891Sjhb{ 2433268891Sjhb return (vm->vatpic); 2434268891Sjhb} 2435268891Sjhb 2436268891Sjhbstruct vatpit * 2437268891Sjhbvm_atpit(struct vm *vm) 2438268891Sjhb{ 2439268891Sjhb return (vm->vatpit); 2440268891Sjhb} 2441268976Sjhb 2442276429Sneelstruct vpmtmr * 2443276429Sneelvm_pmtmr(struct vm *vm) 2444276429Sneel{ 2445276429Sneel 2446276429Sneel return (vm->vpmtmr); 2447276429Sneel} 2448276429Sneel 2449284894Sneelstruct vrtc * 2450284894Sneelvm_rtc(struct vm *vm) 2451284894Sneel{ 2452284894Sneel 2453284894Sneel return (vm->vrtc); 2454284894Sneel} 2455284894Sneel 2456268976Sjhbenum vm_reg_name 2457268976Sjhbvm_segment_name(int seg) 2458268976Sjhb{ 2459268976Sjhb static enum vm_reg_name seg_names[] = { 2460268976Sjhb VM_REG_GUEST_ES, 2461268976Sjhb VM_REG_GUEST_CS, 2462268976Sjhb VM_REG_GUEST_SS, 2463268976Sjhb VM_REG_GUEST_DS, 2464268976Sjhb VM_REG_GUEST_FS, 2465268976Sjhb VM_REG_GUEST_GS 2466268976Sjhb }; 2467268976Sjhb 2468268976Sjhb KASSERT(seg >= 0 && seg < nitems(seg_names), 2469268976Sjhb ("%s: invalid segment encoding %d", __func__, seg)); 2470268976Sjhb return (seg_names[seg]); 2471268976Sjhb} 2472270074Sgrehan 2473270159Sgrehanvoid 2474270159Sgrehanvm_copy_teardown(struct vm *vm, int vcpuid, struct vm_copyinfo *copyinfo, 2475270159Sgrehan int num_copyinfo) 2476270159Sgrehan{ 2477270159Sgrehan int idx; 2478270074Sgrehan 2479270159Sgrehan for (idx = 0; idx < num_copyinfo; idx++) { 2480270159Sgrehan if (copyinfo[idx].cookie != NULL) 2481270159Sgrehan vm_gpa_release(copyinfo[idx].cookie); 2482270159Sgrehan } 2483270159Sgrehan bzero(copyinfo, num_copyinfo * sizeof(struct vm_copyinfo)); 2484270159Sgrehan} 2485270159Sgrehan 2486270159Sgrehanint 2487270159Sgrehanvm_copy_setup(struct vm *vm, int vcpuid, struct vm_guest_paging *paging, 2488270159Sgrehan uint64_t gla, size_t len, int prot, struct vm_copyinfo *copyinfo, 2489284900Sneel int num_copyinfo, int *fault) 2490270159Sgrehan{ 2491270159Sgrehan int error, idx, nused; 2492270159Sgrehan size_t n, off, remaining; 2493270159Sgrehan void *hva, *cookie; 2494270159Sgrehan uint64_t gpa; 2495270159Sgrehan 2496270159Sgrehan bzero(copyinfo, sizeof(struct vm_copyinfo) * num_copyinfo); 2497270159Sgrehan 2498270159Sgrehan nused = 0; 2499270159Sgrehan remaining = len; 2500270159Sgrehan while (remaining > 0) { 2501270159Sgrehan KASSERT(nused < num_copyinfo, ("insufficient vm_copyinfo")); 2502284900Sneel error = vm_gla2gpa(vm, vcpuid, paging, gla, prot, &gpa, fault); 2503284900Sneel if (error || *fault) 2504270159Sgrehan return (error); 2505270159Sgrehan off = gpa & PAGE_MASK; 2506270159Sgrehan n = min(remaining, PAGE_SIZE - off); 2507270159Sgrehan copyinfo[nused].gpa = gpa; 2508270159Sgrehan copyinfo[nused].len = n; 2509270159Sgrehan remaining -= n; 2510270159Sgrehan gla += n; 2511270159Sgrehan nused++; 2512270159Sgrehan } 2513270159Sgrehan 2514270159Sgrehan for (idx = 0; idx < nused; idx++) { 2515295124Sgrehan hva = vm_gpa_hold(vm, vcpuid, copyinfo[idx].gpa, 2516295124Sgrehan copyinfo[idx].len, prot, &cookie); 2517270159Sgrehan if (hva == NULL) 2518270159Sgrehan break; 2519270159Sgrehan copyinfo[idx].hva = hva; 2520270159Sgrehan copyinfo[idx].cookie = cookie; 2521270159Sgrehan } 2522270159Sgrehan 2523270159Sgrehan if (idx != nused) { 2524270159Sgrehan vm_copy_teardown(vm, vcpuid, copyinfo, num_copyinfo); 2525284900Sneel return (EFAULT); 2526270159Sgrehan } else { 2527284900Sneel *fault = 0; 2528270159Sgrehan return (0); 2529270159Sgrehan } 2530270159Sgrehan} 2531270159Sgrehan 2532270159Sgrehanvoid 2533270159Sgrehanvm_copyin(struct vm *vm, int vcpuid, struct vm_copyinfo *copyinfo, void *kaddr, 2534270159Sgrehan size_t len) 2535270159Sgrehan{ 2536270159Sgrehan char *dst; 2537270159Sgrehan int idx; 2538270159Sgrehan 2539270159Sgrehan dst = kaddr; 2540270159Sgrehan idx = 0; 2541270159Sgrehan while (len > 0) { 2542270159Sgrehan bcopy(copyinfo[idx].hva, dst, copyinfo[idx].len); 2543270159Sgrehan len -= copyinfo[idx].len; 2544270159Sgrehan dst += copyinfo[idx].len; 2545270159Sgrehan idx++; 2546270159Sgrehan } 2547270159Sgrehan} 2548270159Sgrehan 2549270159Sgrehanvoid 2550270159Sgrehanvm_copyout(struct vm *vm, int vcpuid, const void *kaddr, 2551270159Sgrehan struct vm_copyinfo *copyinfo, size_t len) 2552270159Sgrehan{ 2553270159Sgrehan const char *src; 2554270159Sgrehan int idx; 2555270159Sgrehan 2556270159Sgrehan src = kaddr; 2557270159Sgrehan idx = 0; 2558270159Sgrehan while (len > 0) { 2559270159Sgrehan bcopy(src, copyinfo[idx].hva, copyinfo[idx].len); 2560270159Sgrehan len -= copyinfo[idx].len; 2561270159Sgrehan src += copyinfo[idx].len; 2562270159Sgrehan idx++; 2563270159Sgrehan } 2564270159Sgrehan} 2565270159Sgrehan 2566270074Sgrehan/* 2567270074Sgrehan * Return the amount of in-use and wired memory for the VM. Since 2568270074Sgrehan * these are global stats, only return the values with for vCPU 0 2569270074Sgrehan */ 2570270074SgrehanVMM_STAT_DECLARE(VMM_MEM_RESIDENT); 2571270074SgrehanVMM_STAT_DECLARE(VMM_MEM_WIRED); 2572270074Sgrehan 2573270074Sgrehanstatic void 2574270074Sgrehanvm_get_rescnt(struct vm *vm, int vcpu, struct vmm_stat_type *stat) 2575270074Sgrehan{ 2576270074Sgrehan 2577270074Sgrehan if (vcpu == 0) { 2578270074Sgrehan vmm_stat_set(vm, vcpu, VMM_MEM_RESIDENT, 2579270074Sgrehan PAGE_SIZE * vmspace_resident_count(vm->vmspace)); 2580270074Sgrehan } 2581270074Sgrehan} 2582270074Sgrehan 2583270074Sgrehanstatic void 2584270074Sgrehanvm_get_wiredcnt(struct vm *vm, int vcpu, struct vmm_stat_type *stat) 2585270074Sgrehan{ 2586270074Sgrehan 2587270074Sgrehan if (vcpu == 0) { 2588270074Sgrehan vmm_stat_set(vm, vcpu, VMM_MEM_WIRED, 2589270074Sgrehan PAGE_SIZE * pmap_wired_count(vmspace_pmap(vm->vmspace))); 2590270074Sgrehan } 2591270074Sgrehan} 2592270074Sgrehan 2593270074SgrehanVMM_STAT_FUNC(VMM_MEM_RESIDENT, "Resident memory", vm_get_rescnt); 2594270074SgrehanVMM_STAT_FUNC(VMM_MEM_WIRED, "Wired memory", vm_get_wiredcnt); 2595