vmm.c revision 270070
165310Sarchie/*- 265310Sarchie * Copyright (c) 2011 NetApp, Inc. 365310Sarchie * All rights reserved. 465310Sarchie * 565310Sarchie * Redistribution and use in source and binary forms, with or without 665310Sarchie * modification, are permitted provided that the following conditions 765310Sarchie * are met: 865310Sarchie * 1. Redistributions of source code must retain the above copyright 965310Sarchie * notice, this list of conditions and the following disclaimer. 1065310Sarchie * 2. Redistributions in binary form must reproduce the above copyright 1165310Sarchie * notice, this list of conditions and the following disclaimer in the 1265310Sarchie * documentation and/or other materials provided with the distribution. 1365310Sarchie * 1465310Sarchie * THIS SOFTWARE IS PROVIDED BY NETAPP, INC ``AS IS'' AND 1565310Sarchie * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 1665310Sarchie * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 1765310Sarchie * ARE DISCLAIMED. IN NO EVENT SHALL NETAPP, INC OR CONTRIBUTORS BE LIABLE 1865310Sarchie * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 1965310Sarchie * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 2065310Sarchie * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 2165310Sarchie * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 2265310Sarchie * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 2365310Sarchie * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 2465310Sarchie * SUCH DAMAGE. 2565310Sarchie * 2665310Sarchie * $FreeBSD: stable/10/sys/amd64/vmm/vmm.c 270070 2014-08-17 00:52:07Z grehan $ 2765310Sarchie */ 2865310Sarchie 2965310Sarchie#include <sys/cdefs.h> 3065310Sarchie__FBSDID("$FreeBSD: stable/10/sys/amd64/vmm/vmm.c 270070 2014-08-17 00:52:07Z grehan $"); 3165310Sarchie 3265310Sarchie#include <sys/param.h> 3365310Sarchie#include <sys/systm.h> 3465310Sarchie#include <sys/kernel.h> 3565310Sarchie#include <sys/module.h> 3665310Sarchie#include <sys/sysctl.h> 3765310Sarchie#include <sys/malloc.h> 3865310Sarchie#include <sys/pcpu.h> 3965310Sarchie#include <sys/lock.h> 4065310Sarchie#include <sys/mutex.h> 4165310Sarchie#include <sys/proc.h> 4265310Sarchie#include <sys/rwlock.h> 4365310Sarchie#include <sys/sched.h> 4465310Sarchie#include <sys/smp.h> 4565310Sarchie#include <sys/systm.h> 4665310Sarchie 4765310Sarchie#include <vm/vm.h> 4865310Sarchie#include <vm/vm_object.h> 4965310Sarchie#include <vm/vm_page.h> 5065310Sarchie#include <vm/pmap.h> 5165310Sarchie#include <vm/vm_map.h> 5265310Sarchie#include <vm/vm_extern.h> 5365310Sarchie#include <vm/vm_param.h> 5465310Sarchie 5565310Sarchie#include <machine/cpu.h> 5665310Sarchie#include <machine/vm.h> 5765310Sarchie#include <machine/pcb.h> 5865310Sarchie#include <machine/smp.h> 5965310Sarchie#include <x86/psl.h> 6065310Sarchie#include <x86/apicreg.h> 6165310Sarchie#include <machine/vmparam.h> 6265310Sarchie 6365310Sarchie#include <machine/vmm.h> 6465310Sarchie#include <machine/vmm_dev.h> 6565310Sarchie#include <machine/vmm_instruction_emul.h> 6665310Sarchie 6765310Sarchie#include "vmm_ioport.h" 6865310Sarchie#include "vmm_ktr.h" 6965310Sarchie#include "vmm_host.h" 7065310Sarchie#include "vmm_mem.h" 7165310Sarchie#include "vmm_util.h" 7265310Sarchie#include "vatpic.h" 7365310Sarchie#include "vatpit.h" 7465310Sarchie#include "vhpet.h" 7565310Sarchie#include "vioapic.h" 7665310Sarchie#include "vlapic.h" 7765310Sarchie#include "vmm_msr.h" 7865310Sarchie#include "vmm_ipi.h" 7965310Sarchie#include "vmm_stat.h" 8065310Sarchie#include "vmm_lapic.h" 8165310Sarchie 8265310Sarchie#include "io/ppt.h" 8365310Sarchie#include "io/iommu.h" 8465310Sarchie 8565310Sarchiestruct vlapic; 8665310Sarchie 8765310Sarchiestruct vcpu { 8865310Sarchie int flags; 8965310Sarchie enum vcpu_state state; 9065310Sarchie struct mtx mtx; 9165310Sarchie int hostcpu; /* host cpuid this vcpu last ran on */ 9265310Sarchie uint64_t guest_msrs[VMM_MSR_NUM]; 9365310Sarchie struct vlapic *vlapic; 9465310Sarchie int vcpuid; 9565310Sarchie struct savefpu *guestfpu; /* guest fpu state */ 9665310Sarchie uint64_t guest_xcr0; 9765310Sarchie void *stats; 9865310Sarchie struct vm_exit exitinfo; 9965310Sarchie enum x2apic_state x2apic_state; 10065310Sarchie int nmi_pending; 10165310Sarchie int extint_pending; 10265310Sarchie struct vm_exception exception; 10365310Sarchie int exception_pending; 10465310Sarchie}; 10565310Sarchie 10665310Sarchie#define vcpu_lock_init(v) mtx_init(&((v)->mtx), "vcpu lock", 0, MTX_SPIN) 10765310Sarchie#define vcpu_lock(v) mtx_lock_spin(&((v)->mtx)) 10865310Sarchie#define vcpu_unlock(v) mtx_unlock_spin(&((v)->mtx)) 10965310Sarchie#define vcpu_assert_locked(v) mtx_assert(&((v)->mtx), MA_OWNED) 11065310Sarchie 11165310Sarchiestruct mem_seg { 11265310Sarchie vm_paddr_t gpa; 11365310Sarchie size_t len; 11465310Sarchie boolean_t wired; 11565310Sarchie vm_object_t object; 11665310Sarchie}; 11765310Sarchie#define VM_MAX_MEMORY_SEGMENTS 2 11865310Sarchie 11965310Sarchiestruct vm { 12065310Sarchie void *cookie; /* processor-specific data */ 12165310Sarchie void *iommu; /* iommu-specific data */ 12265310Sarchie struct vhpet *vhpet; /* virtual HPET */ 12365310Sarchie struct vioapic *vioapic; /* virtual ioapic */ 12465310Sarchie struct vatpic *vatpic; /* virtual atpic */ 12565310Sarchie struct vatpit *vatpit; /* virtual atpit */ 12665310Sarchie struct vmspace *vmspace; /* guest's address space */ 12765310Sarchie struct vcpu vcpu[VM_MAXCPU]; 12865310Sarchie int num_mem_segs; 12965310Sarchie struct mem_seg mem_segs[VM_MAX_MEMORY_SEGMENTS]; 13065310Sarchie char name[VM_MAX_NAMELEN]; 13165310Sarchie 13265310Sarchie /* 13365310Sarchie * Set of active vcpus. 13465310Sarchie * An active vcpu is one that has been started implicitly (BSP) or 13565310Sarchie * explicitly (AP) by sending it a startup ipi. 13665310Sarchie */ 13765310Sarchie volatile cpuset_t active_cpus; 13865310Sarchie 13965310Sarchie struct mtx rendezvous_mtx; 14065310Sarchie cpuset_t rendezvous_req_cpus; 14165310Sarchie cpuset_t rendezvous_done_cpus; 14265310Sarchie void *rendezvous_arg; 14365310Sarchie vm_rendezvous_func_t rendezvous_func; 14465310Sarchie 14565310Sarchie int suspend; 14665310Sarchie volatile cpuset_t suspended_cpus; 14765310Sarchie 14865310Sarchie volatile cpuset_t halted_cpus; 14965310Sarchie}; 15065310Sarchie 15165310Sarchiestatic int vmm_initialized; 15265310Sarchie 15365310Sarchiestatic struct vmm_ops *ops; 15465310Sarchie#define VMM_INIT(num) (ops != NULL ? (*ops->init)(num) : 0) 15565310Sarchie#define VMM_CLEANUP() (ops != NULL ? (*ops->cleanup)() : 0) 15665310Sarchie#define VMM_RESUME() (ops != NULL ? (*ops->resume)() : 0) 15765310Sarchie 15865310Sarchie#define VMINIT(vm, pmap) (ops != NULL ? (*ops->vminit)(vm, pmap): NULL) 15965310Sarchie#define VMRUN(vmi, vcpu, rip, pmap, rptr, sptr) \ 16065310Sarchie (ops != NULL ? (*ops->vmrun)(vmi, vcpu, rip, pmap, rptr, sptr) : ENXIO) 16165310Sarchie#define VMCLEANUP(vmi) (ops != NULL ? (*ops->vmcleanup)(vmi) : NULL) 16265310Sarchie#define VMSPACE_ALLOC(min, max) \ 16365310Sarchie (ops != NULL ? (*ops->vmspace_alloc)(min, max) : NULL) 16465310Sarchie#define VMSPACE_FREE(vmspace) \ 16565310Sarchie (ops != NULL ? (*ops->vmspace_free)(vmspace) : ENXIO) 16665310Sarchie#define VMGETREG(vmi, vcpu, num, retval) \ 16765310Sarchie (ops != NULL ? (*ops->vmgetreg)(vmi, vcpu, num, retval) : ENXIO) 16865310Sarchie#define VMSETREG(vmi, vcpu, num, val) \ 16965310Sarchie (ops != NULL ? (*ops->vmsetreg)(vmi, vcpu, num, val) : ENXIO) 17065310Sarchie#define VMGETDESC(vmi, vcpu, num, desc) \ 17165310Sarchie (ops != NULL ? (*ops->vmgetdesc)(vmi, vcpu, num, desc) : ENXIO) 17265310Sarchie#define VMSETDESC(vmi, vcpu, num, desc) \ 17365310Sarchie (ops != NULL ? (*ops->vmsetdesc)(vmi, vcpu, num, desc) : ENXIO) 17465310Sarchie#define VMGETCAP(vmi, vcpu, num, retval) \ 17565310Sarchie (ops != NULL ? (*ops->vmgetcap)(vmi, vcpu, num, retval) : ENXIO) 17665310Sarchie#define VMSETCAP(vmi, vcpu, num, val) \ 17765310Sarchie (ops != NULL ? (*ops->vmsetcap)(vmi, vcpu, num, val) : ENXIO) 17865310Sarchie#define VLAPIC_INIT(vmi, vcpu) \ 17965310Sarchie (ops != NULL ? (*ops->vlapic_init)(vmi, vcpu) : NULL) 18065310Sarchie#define VLAPIC_CLEANUP(vmi, vlapic) \ 18165310Sarchie (ops != NULL ? (*ops->vlapic_cleanup)(vmi, vlapic) : NULL) 18265310Sarchie 18365310Sarchie#define fpu_start_emulating() load_cr0(rcr0() | CR0_TS) 18465310Sarchie#define fpu_stop_emulating() clts() 18565310Sarchie 18665310Sarchiestatic MALLOC_DEFINE(M_VM, "vm", "vm"); 18765310SarchieCTASSERT(VMM_MSR_NUM <= 64); /* msr_mask can keep track of up to 64 msrs */ 18865310Sarchie 18965310Sarchie/* statistics */ 19065310Sarchiestatic VMM_STAT(VCPU_TOTAL_RUNTIME, "vcpu total runtime"); 19165310Sarchie 19265310SarchieSYSCTL_NODE(_hw, OID_AUTO, vmm, CTLFLAG_RW, NULL, NULL); 19365310Sarchie 19465310Sarchie/* 19565310Sarchie * Halt the guest if all vcpus are executing a HLT instruction with 19665310Sarchie * interrupts disabled. 19765310Sarchie */ 19865310Sarchiestatic int halt_detection_enabled = 1; 19965310SarchieTUNABLE_INT("hw.vmm.halt_detection", &halt_detection_enabled); 20065310SarchieSYSCTL_INT(_hw_vmm, OID_AUTO, halt_detection, CTLFLAG_RDTUN, 20165310Sarchie &halt_detection_enabled, 0, 20265310Sarchie "Halt VM if all vcpus execute HLT with interrupts disabled"); 20365310Sarchie 20465310Sarchiestatic int vmm_ipinum; 20565310SarchieSYSCTL_INT(_hw_vmm, OID_AUTO, ipinum, CTLFLAG_RD, &vmm_ipinum, 0, 20665310Sarchie "IPI vector used for vcpu notifications"); 20765310Sarchie 20865310Sarchiestatic void 20965310Sarchievcpu_cleanup(struct vm *vm, int i) 21065310Sarchie{ 21165310Sarchie struct vcpu *vcpu = &vm->vcpu[i]; 21265310Sarchie 21365310Sarchie VLAPIC_CLEANUP(vm->cookie, vcpu->vlapic); 21465310Sarchie vmm_stat_free(vcpu->stats); 21565310Sarchie fpu_save_area_free(vcpu->guestfpu); 21665310Sarchie} 21765310Sarchie 21865310Sarchiestatic void 21965310Sarchievcpu_init(struct vm *vm, uint32_t vcpu_id) 22065310Sarchie{ 22165310Sarchie struct vcpu *vcpu; 22265310Sarchie 22365310Sarchie vcpu = &vm->vcpu[vcpu_id]; 22465310Sarchie 22565310Sarchie vcpu_lock_init(vcpu); 22665310Sarchie vcpu->hostcpu = NOCPU; 22765310Sarchie vcpu->vcpuid = vcpu_id; 22865310Sarchie vcpu->vlapic = VLAPIC_INIT(vm->cookie, vcpu_id); 22965310Sarchie vm_set_x2apic_state(vm, vcpu_id, X2APIC_DISABLED); 23065310Sarchie vcpu->guest_xcr0 = XFEATURE_ENABLED_X87; 23165310Sarchie vcpu->guestfpu = fpu_save_area_alloc(); 23265310Sarchie fpu_save_area_reset(vcpu->guestfpu); 23365310Sarchie vcpu->stats = vmm_stat_alloc(); 23465310Sarchie} 23565310Sarchie 23665310Sarchiestruct vm_exit * 23765310Sarchievm_exitinfo(struct vm *vm, int cpuid) 23865310Sarchie{ 23965310Sarchie struct vcpu *vcpu; 24065310Sarchie 24165310Sarchie if (cpuid < 0 || cpuid >= VM_MAXCPU) 24265310Sarchie panic("vm_exitinfo: invalid cpuid %d", cpuid); 24365310Sarchie 24465310Sarchie vcpu = &vm->vcpu[cpuid]; 24565310Sarchie 24665310Sarchie return (&vcpu->exitinfo); 24765310Sarchie} 24865310Sarchie 24965310Sarchiestatic void 25065310Sarchievmm_resume(void) 25165310Sarchie{ 25265310Sarchie VMM_RESUME(); 25365310Sarchie} 25465310Sarchie 25565310Sarchiestatic int 25665310Sarchievmm_init(void) 25765310Sarchie{ 25865310Sarchie int error; 25965310Sarchie 26065310Sarchie vmm_host_state_init(); 26165310Sarchie 26265310Sarchie vmm_ipinum = vmm_ipi_alloc(); 26365310Sarchie if (vmm_ipinum == 0) 26465310Sarchie vmm_ipinum = IPI_AST; 26565310Sarchie 26665310Sarchie error = vmm_mem_init(); 26765310Sarchie if (error) 26865310Sarchie return (error); 26965310Sarchie 27065310Sarchie if (vmm_is_intel()) 27165310Sarchie ops = &vmm_ops_intel; 27265310Sarchie else if (vmm_is_amd()) 27365310Sarchie ops = &vmm_ops_amd; 27465310Sarchie else 27565310Sarchie return (ENXIO); 27665310Sarchie 27765310Sarchie vmm_msr_init(); 27865310Sarchie vmm_resume_p = vmm_resume; 27965310Sarchie 28065310Sarchie return (VMM_INIT(vmm_ipinum)); 28165310Sarchie} 28265310Sarchie 28366887Sarchiestatic int 28465310Sarchievmm_handler(module_t mod, int what, void *arg) 28565310Sarchie{ 28665310Sarchie int error; 28765310Sarchie 28865310Sarchie switch (what) { 28965310Sarchie case MOD_LOAD: 29065310Sarchie vmmdev_init(); 29165310Sarchie if (ppt_avail_devices() > 0) 29265310Sarchie iommu_init(); 29365310Sarchie error = vmm_init(); 29465310Sarchie if (error == 0) 29565310Sarchie vmm_initialized = 1; 29665310Sarchie break; 29765310Sarchie case MOD_UNLOAD: 29865310Sarchie error = vmmdev_cleanup(); 29965310Sarchie if (error == 0) { 30065310Sarchie vmm_resume_p = NULL; 30165310Sarchie iommu_cleanup(); 30265310Sarchie if (vmm_ipinum != IPI_AST) 30365310Sarchie vmm_ipi_free(vmm_ipinum); 30465310Sarchie error = VMM_CLEANUP(); 30565310Sarchie /* 30665310Sarchie * Something bad happened - prevent new 30765310Sarchie * VMs from being created 30865310Sarchie */ 30965310Sarchie if (error) 31065310Sarchie vmm_initialized = 0; 31165310Sarchie } 31265310Sarchie break; 31365310Sarchie default: 31465310Sarchie error = 0; 31565310Sarchie break; 31665310Sarchie } 31765310Sarchie return (error); 31865310Sarchie} 31965310Sarchie 32065310Sarchiestatic moduledata_t vmm_kmod = { 32165310Sarchie "vmm", 32265310Sarchie vmm_handler, 32365310Sarchie NULL 32465310Sarchie}; 32565310Sarchie 32665310Sarchie/* 32765310Sarchie * vmm initialization has the following dependencies: 32865310Sarchie * 32965310Sarchie * - iommu initialization must happen after the pci passthru driver has had 33065310Sarchie * a chance to attach to any passthru devices (after SI_SUB_CONFIGURE). 33165310Sarchie * 33265310Sarchie * - VT-x initialization requires smp_rendezvous() and therefore must happen 33365310Sarchie * after SMP is fully functional (after SI_SUB_SMP). 33465310Sarchie */ 33565310SarchieDECLARE_MODULE(vmm, vmm_kmod, SI_SUB_SMP + 1, SI_ORDER_ANY); 33665310SarchieMODULE_VERSION(vmm, 1); 33765310Sarchie 33865310Sarchieint 33965310Sarchievm_create(const char *name, struct vm **retvm) 34065310Sarchie{ 34165310Sarchie int i; 34265310Sarchie struct vm *vm; 34365310Sarchie struct vmspace *vmspace; 34465310Sarchie 34565310Sarchie /* 34665310Sarchie * If vmm.ko could not be successfully initialized then don't attempt 34765310Sarchie * to create the virtual machine. 34865310Sarchie */ 34965310Sarchie if (!vmm_initialized) 35065310Sarchie return (ENXIO); 35165310Sarchie 35265310Sarchie if (name == NULL || strlen(name) >= VM_MAX_NAMELEN) 35365310Sarchie return (EINVAL); 35465310Sarchie 35565310Sarchie vmspace = VMSPACE_ALLOC(VM_MIN_ADDRESS, VM_MAXUSER_ADDRESS); 35665310Sarchie if (vmspace == NULL) 35765310Sarchie return (ENOMEM); 35865310Sarchie 35965310Sarchie vm = malloc(sizeof(struct vm), M_VM, M_WAITOK | M_ZERO); 36065310Sarchie strcpy(vm->name, name); 36165310Sarchie vm->vmspace = vmspace; 36265310Sarchie mtx_init(&vm->rendezvous_mtx, "vm rendezvous lock", 0, MTX_DEF); 36365310Sarchie vm->cookie = VMINIT(vm, vmspace_pmap(vmspace)); 36465310Sarchie vm->vioapic = vioapic_init(vm); 36565310Sarchie vm->vhpet = vhpet_init(vm); 36665310Sarchie vm->vatpic = vatpic_init(vm); 36765310Sarchie vm->vatpit = vatpit_init(vm); 36865310Sarchie 36965310Sarchie for (i = 0; i < VM_MAXCPU; i++) { 37065310Sarchie vcpu_init(vm, i); 37165310Sarchie guest_msrs_init(vm, i); 37265310Sarchie } 37365310Sarchie 37465310Sarchie *retvm = vm; 37565310Sarchie return (0); 37665310Sarchie} 37765310Sarchie 37865310Sarchiestatic void 37965310Sarchievm_free_mem_seg(struct vm *vm, struct mem_seg *seg) 38065310Sarchie{ 38165310Sarchie 38265310Sarchie if (seg->object != NULL) 38365310Sarchie vmm_mem_free(vm->vmspace, seg->gpa, seg->len); 38465310Sarchie 38565310Sarchie bzero(seg, sizeof(*seg)); 38665310Sarchie} 38765310Sarchie 38865310Sarchievoid 38965310Sarchievm_destroy(struct vm *vm) 39065310Sarchie{ 39165310Sarchie int i; 39265310Sarchie 39365310Sarchie ppt_unassign_all(vm); 39465310Sarchie 39565310Sarchie if (vm->iommu != NULL) 39665310Sarchie iommu_destroy_domain(vm->iommu); 39765310Sarchie 39865310Sarchie vatpit_cleanup(vm->vatpit); 39965310Sarchie vhpet_cleanup(vm->vhpet); 40065310Sarchie vatpic_cleanup(vm->vatpic); 40165310Sarchie vioapic_cleanup(vm->vioapic); 40265310Sarchie 40365310Sarchie for (i = 0; i < vm->num_mem_segs; i++) 40465310Sarchie vm_free_mem_seg(vm, &vm->mem_segs[i]); 40565310Sarchie 40665310Sarchie vm->num_mem_segs = 0; 40765310Sarchie 40865310Sarchie for (i = 0; i < VM_MAXCPU; i++) 40965310Sarchie vcpu_cleanup(vm, i); 41065310Sarchie 41165310Sarchie VMSPACE_FREE(vm->vmspace); 41265310Sarchie 41365310Sarchie VMCLEANUP(vm->cookie); 41465310Sarchie 41565310Sarchie free(vm, M_VM); 41665310Sarchie} 41765310Sarchie 41865310Sarchieconst char * 41965310Sarchievm_name(struct vm *vm) 42065310Sarchie{ 42165310Sarchie return (vm->name); 42265310Sarchie} 42365310Sarchie 42465310Sarchieint 42565310Sarchievm_map_mmio(struct vm *vm, vm_paddr_t gpa, size_t len, vm_paddr_t hpa) 42665310Sarchie{ 42765310Sarchie vm_object_t obj; 42865310Sarchie 42965310Sarchie if ((obj = vmm_mmio_alloc(vm->vmspace, gpa, len, hpa)) == NULL) 43065310Sarchie return (ENOMEM); 43165310Sarchie else 43265310Sarchie return (0); 43365310Sarchie} 43465310Sarchie 43565310Sarchieint 43665310Sarchievm_unmap_mmio(struct vm *vm, vm_paddr_t gpa, size_t len) 43765310Sarchie{ 43865310Sarchie 43965310Sarchie vmm_mmio_free(vm->vmspace, gpa, len); 44065310Sarchie return (0); 44165310Sarchie} 44265310Sarchie 44365310Sarchieboolean_t 44465310Sarchievm_mem_allocated(struct vm *vm, vm_paddr_t gpa) 44565310Sarchie{ 44665310Sarchie int i; 44765310Sarchie vm_paddr_t gpabase, gpalimit; 44865310Sarchie 44965310Sarchie for (i = 0; i < vm->num_mem_segs; i++) { 45065310Sarchie gpabase = vm->mem_segs[i].gpa; 45165310Sarchie gpalimit = gpabase + vm->mem_segs[i].len; 45265310Sarchie if (gpa >= gpabase && gpa < gpalimit) 45365310Sarchie return (TRUE); /* 'gpa' is regular memory */ 45465310Sarchie } 45565310Sarchie 45665310Sarchie if (ppt_is_mmio(vm, gpa)) 45765310Sarchie return (TRUE); /* 'gpa' is pci passthru mmio */ 45865310Sarchie 45965310Sarchie return (FALSE); 46065310Sarchie} 46165310Sarchie 46265310Sarchieint 46365310Sarchievm_malloc(struct vm *vm, vm_paddr_t gpa, size_t len) 46465310Sarchie{ 46565310Sarchie int available, allocated; 46665310Sarchie struct mem_seg *seg; 46765310Sarchie vm_object_t object; 46865310Sarchie vm_paddr_t g; 46965310Sarchie 47065310Sarchie if ((gpa & PAGE_MASK) || (len & PAGE_MASK) || len == 0) 47165310Sarchie return (EINVAL); 47265310Sarchie 47365310Sarchie available = allocated = 0; 47465310Sarchie g = gpa; 47565310Sarchie while (g < gpa + len) { 47665310Sarchie if (vm_mem_allocated(vm, g)) 47765310Sarchie allocated++; 47865310Sarchie else 47965310Sarchie available++; 48065310Sarchie 48165310Sarchie g += PAGE_SIZE; 48265310Sarchie } 48365310Sarchie 48465310Sarchie /* 48565310Sarchie * If there are some allocated and some available pages in the address 48665310Sarchie * range then it is an error. 48765310Sarchie */ 48865310Sarchie if (allocated && available) 48965310Sarchie return (EINVAL); 49065310Sarchie 49165310Sarchie /* 49265310Sarchie * If the entire address range being requested has already been 49365310Sarchie * allocated then there isn't anything more to do. 49465310Sarchie */ 49565310Sarchie if (allocated && available == 0) 49665310Sarchie return (0); 49765310Sarchie 49865310Sarchie if (vm->num_mem_segs >= VM_MAX_MEMORY_SEGMENTS) 49965310Sarchie return (E2BIG); 50065310Sarchie 50165310Sarchie seg = &vm->mem_segs[vm->num_mem_segs]; 50265310Sarchie 50365310Sarchie if ((object = vmm_mem_alloc(vm->vmspace, gpa, len)) == NULL) 50465310Sarchie return (ENOMEM); 50565310Sarchie 50665310Sarchie seg->gpa = gpa; 50765310Sarchie seg->len = len; 50865310Sarchie seg->object = object; 50965310Sarchie seg->wired = FALSE; 51065310Sarchie 51165310Sarchie vm->num_mem_segs++; 51265310Sarchie 51365310Sarchie return (0); 51465310Sarchie} 51565310Sarchie 51665310Sarchiestatic void 51765310Sarchievm_gpa_unwire(struct vm *vm) 51865310Sarchie{ 51965310Sarchie int i, rv; 52065310Sarchie struct mem_seg *seg; 52165310Sarchie 52265310Sarchie for (i = 0; i < vm->num_mem_segs; i++) { 52365310Sarchie seg = &vm->mem_segs[i]; 52465310Sarchie if (!seg->wired) 52565310Sarchie continue; 52665310Sarchie 52765310Sarchie rv = vm_map_unwire(&vm->vmspace->vm_map, 52865310Sarchie seg->gpa, seg->gpa + seg->len, 52965310Sarchie VM_MAP_WIRE_USER | VM_MAP_WIRE_NOHOLES); 53065310Sarchie KASSERT(rv == KERN_SUCCESS, ("vm(%s) memory segment " 53165310Sarchie "%#lx/%ld could not be unwired: %d", 53265310Sarchie vm_name(vm), seg->gpa, seg->len, rv)); 53365310Sarchie 53465310Sarchie seg->wired = FALSE; 53565310Sarchie } 53665310Sarchie} 53765310Sarchie 53865310Sarchiestatic int 53965310Sarchievm_gpa_wire(struct vm *vm) 54065310Sarchie{ 54165310Sarchie int i, rv; 54265310Sarchie struct mem_seg *seg; 54365310Sarchie 54465310Sarchie for (i = 0; i < vm->num_mem_segs; i++) { 54565310Sarchie seg = &vm->mem_segs[i]; 54665310Sarchie if (seg->wired) 54765310Sarchie continue; 54865310Sarchie 54965310Sarchie /* XXX rlimits? */ 55065310Sarchie rv = vm_map_wire(&vm->vmspace->vm_map, 55165310Sarchie seg->gpa, seg->gpa + seg->len, 55265310Sarchie VM_MAP_WIRE_USER | VM_MAP_WIRE_NOHOLES); 55365310Sarchie if (rv != KERN_SUCCESS) 55465310Sarchie break; 55565310Sarchie 55665310Sarchie seg->wired = TRUE; 55765310Sarchie } 55865310Sarchie 55965310Sarchie if (i < vm->num_mem_segs) { 56065310Sarchie /* 56165310Sarchie * Undo the wiring before returning an error. 56265310Sarchie */ 56365310Sarchie vm_gpa_unwire(vm); 56465310Sarchie return (EAGAIN); 56565310Sarchie } 56665310Sarchie 56765310Sarchie return (0); 56865310Sarchie} 56965310Sarchie 57065310Sarchiestatic void 57165310Sarchievm_iommu_modify(struct vm *vm, boolean_t map) 57265310Sarchie{ 57365310Sarchie int i, sz; 57465310Sarchie vm_paddr_t gpa, hpa; 57565310Sarchie struct mem_seg *seg; 57665310Sarchie void *vp, *cookie, *host_domain; 57765310Sarchie 57865310Sarchie sz = PAGE_SIZE; 57965310Sarchie host_domain = iommu_host_domain(); 58065310Sarchie 58165310Sarchie for (i = 0; i < vm->num_mem_segs; i++) { 58265310Sarchie seg = &vm->mem_segs[i]; 58365310Sarchie KASSERT(seg->wired, ("vm(%s) memory segment %#lx/%ld not wired", 58465310Sarchie vm_name(vm), seg->gpa, seg->len)); 58565310Sarchie 58665310Sarchie gpa = seg->gpa; 58765310Sarchie while (gpa < seg->gpa + seg->len) { 58865310Sarchie vp = vm_gpa_hold(vm, gpa, PAGE_SIZE, VM_PROT_WRITE, 58965310Sarchie &cookie); 59065310Sarchie KASSERT(vp != NULL, ("vm(%s) could not map gpa %#lx", 59165310Sarchie vm_name(vm), gpa)); 59265310Sarchie 59365310Sarchie vm_gpa_release(cookie); 59465310Sarchie 59565310Sarchie hpa = DMAP_TO_PHYS((uintptr_t)vp); 59665310Sarchie if (map) { 59765310Sarchie iommu_create_mapping(vm->iommu, gpa, hpa, sz); 59865310Sarchie iommu_remove_mapping(host_domain, hpa, sz); 59965310Sarchie } else { 60065310Sarchie iommu_remove_mapping(vm->iommu, gpa, sz); 60165310Sarchie iommu_create_mapping(host_domain, hpa, hpa, sz); 60265310Sarchie } 60365310Sarchie 60465310Sarchie gpa += PAGE_SIZE; 60565310Sarchie } 60665310Sarchie } 60765310Sarchie 60865310Sarchie /* 60965310Sarchie * Invalidate the cached translations associated with the domain 61065310Sarchie * from which pages were removed. 61165310Sarchie */ 61265310Sarchie if (map) 61365310Sarchie iommu_invalidate_tlb(host_domain); 61465310Sarchie else 61565310Sarchie iommu_invalidate_tlb(vm->iommu); 61665310Sarchie} 61765310Sarchie 61865310Sarchie#define vm_iommu_unmap(vm) vm_iommu_modify((vm), FALSE) 61965310Sarchie#define vm_iommu_map(vm) vm_iommu_modify((vm), TRUE) 62065310Sarchie 62165310Sarchieint 62265310Sarchievm_unassign_pptdev(struct vm *vm, int bus, int slot, int func) 62365310Sarchie{ 62465310Sarchie int error; 62565310Sarchie 62665310Sarchie error = ppt_unassign_device(vm, bus, slot, func); 62765310Sarchie if (error) 62865310Sarchie return (error); 62965310Sarchie 63065310Sarchie if (ppt_assigned_devices(vm) == 0) { 63165310Sarchie vm_iommu_unmap(vm); 63265310Sarchie vm_gpa_unwire(vm); 63365310Sarchie } 63465310Sarchie return (0); 63565310Sarchie} 63665310Sarchie 63765310Sarchieint 63865310Sarchievm_assign_pptdev(struct vm *vm, int bus, int slot, int func) 63965310Sarchie{ 64065310Sarchie int error; 64165310Sarchie vm_paddr_t maxaddr; 64265310Sarchie 64365310Sarchie /* 64465310Sarchie * Virtual machines with pci passthru devices get special treatment: 64565310Sarchie * - the guest physical memory is wired 64665310Sarchie * - the iommu is programmed to do the 'gpa' to 'hpa' translation 64765310Sarchie * 64865310Sarchie * We need to do this before the first pci passthru device is attached. 64965310Sarchie */ 65065310Sarchie if (ppt_assigned_devices(vm) == 0) { 65165310Sarchie KASSERT(vm->iommu == NULL, 65265310Sarchie ("vm_assign_pptdev: iommu must be NULL")); 65365310Sarchie maxaddr = vmm_mem_maxaddr(); 65465310Sarchie vm->iommu = iommu_create_domain(maxaddr); 65565310Sarchie 65665310Sarchie error = vm_gpa_wire(vm); 65765310Sarchie if (error) 65865310Sarchie return (error); 65965310Sarchie 66065310Sarchie vm_iommu_map(vm); 66165310Sarchie } 66265310Sarchie 66365310Sarchie error = ppt_assign_device(vm, bus, slot, func); 66465310Sarchie return (error); 66565310Sarchie} 66665310Sarchie 66765310Sarchievoid * 66865310Sarchievm_gpa_hold(struct vm *vm, vm_paddr_t gpa, size_t len, int reqprot, 66965310Sarchie void **cookie) 67065310Sarchie{ 67165310Sarchie int count, pageoff; 67265310Sarchie vm_page_t m; 67365310Sarchie 67465310Sarchie pageoff = gpa & PAGE_MASK; 67565310Sarchie if (len > PAGE_SIZE - pageoff) 67665310Sarchie panic("vm_gpa_hold: invalid gpa/len: 0x%016lx/%lu", gpa, len); 67765310Sarchie 67866313Sarchie count = vm_fault_quick_hold_pages(&vm->vmspace->vm_map, 67965310Sarchie trunc_page(gpa), PAGE_SIZE, reqprot, &m, 1); 68065310Sarchie 68165310Sarchie if (count == 1) { 68265310Sarchie *cookie = m; 68365310Sarchie return ((void *)(PHYS_TO_DMAP(VM_PAGE_TO_PHYS(m)) + pageoff)); 68465310Sarchie } else { 68565310Sarchie *cookie = NULL; 68665310Sarchie return (NULL); 68765310Sarchie } 68865310Sarchie} 68965310Sarchie 69065310Sarchievoid 69165310Sarchievm_gpa_release(void *cookie) 69265310Sarchie{ 69365310Sarchie vm_page_t m = cookie; 69465310Sarchie 69565310Sarchie vm_page_lock(m); 69665310Sarchie vm_page_unhold(m); 69765310Sarchie vm_page_unlock(m); 69865310Sarchie} 69965310Sarchie 70065310Sarchieint 70165310Sarchievm_gpabase2memseg(struct vm *vm, vm_paddr_t gpabase, 70265310Sarchie struct vm_memory_segment *seg) 70365310Sarchie{ 70465310Sarchie int i; 70565310Sarchie 70665310Sarchie for (i = 0; i < vm->num_mem_segs; i++) { 70765310Sarchie if (gpabase == vm->mem_segs[i].gpa) { 70865310Sarchie seg->gpa = vm->mem_segs[i].gpa; 70965310Sarchie seg->len = vm->mem_segs[i].len; 71065310Sarchie seg->wired = vm->mem_segs[i].wired; 71165310Sarchie return (0); 71265310Sarchie } 71365310Sarchie } 71465310Sarchie return (-1); 71565310Sarchie} 71665310Sarchie 71765310Sarchieint 71865310Sarchievm_get_memobj(struct vm *vm, vm_paddr_t gpa, size_t len, 71965310Sarchie vm_offset_t *offset, struct vm_object **object) 72065310Sarchie{ 72165310Sarchie int i; 72265310Sarchie size_t seg_len; 72365310Sarchie vm_paddr_t seg_gpa; 72465310Sarchie vm_object_t seg_obj; 72565310Sarchie 72665310Sarchie for (i = 0; i < vm->num_mem_segs; i++) { 72765310Sarchie if ((seg_obj = vm->mem_segs[i].object) == NULL) 72865310Sarchie continue; 72965310Sarchie 73065310Sarchie seg_gpa = vm->mem_segs[i].gpa; 73165310Sarchie seg_len = vm->mem_segs[i].len; 73265310Sarchie 73365310Sarchie if (gpa >= seg_gpa && gpa < seg_gpa + seg_len) { 73465310Sarchie *offset = gpa - seg_gpa; 73565310Sarchie *object = seg_obj; 73665310Sarchie vm_object_reference(seg_obj); 73765310Sarchie return (0); 73865310Sarchie } 73965310Sarchie } 74065310Sarchie 74165310Sarchie return (EINVAL); 74265310Sarchie} 74365310Sarchie 74465310Sarchieint 74565310Sarchievm_get_register(struct vm *vm, int vcpu, int reg, uint64_t *retval) 74665310Sarchie{ 74765310Sarchie 74865310Sarchie if (vcpu < 0 || vcpu >= VM_MAXCPU) 74965310Sarchie return (EINVAL); 75065310Sarchie 75165310Sarchie if (reg >= VM_REG_LAST) 75265310Sarchie return (EINVAL); 75365310Sarchie 75465310Sarchie return (VMGETREG(vm->cookie, vcpu, reg, retval)); 75565310Sarchie} 75665310Sarchie 75765310Sarchieint 75865310Sarchievm_set_register(struct vm *vm, int vcpu, int reg, uint64_t val) 75965310Sarchie{ 76065310Sarchie 76165310Sarchie if (vcpu < 0 || vcpu >= VM_MAXCPU) 76265310Sarchie return (EINVAL); 76365310Sarchie 76465310Sarchie if (reg >= VM_REG_LAST) 76565310Sarchie return (EINVAL); 76665310Sarchie 76765310Sarchie return (VMSETREG(vm->cookie, vcpu, reg, val)); 76865310Sarchie} 76965310Sarchie 77065310Sarchiestatic boolean_t 77165310Sarchieis_descriptor_table(int reg) 77265310Sarchie{ 77365310Sarchie 77465310Sarchie switch (reg) { 77565310Sarchie case VM_REG_GUEST_IDTR: 77665310Sarchie case VM_REG_GUEST_GDTR: 77765310Sarchie return (TRUE); 77865310Sarchie default: 77965310Sarchie return (FALSE); 78065310Sarchie } 78165310Sarchie} 78265310Sarchie 78365310Sarchiestatic boolean_t 78465310Sarchieis_segment_register(int reg) 78565310Sarchie{ 78665310Sarchie 78765310Sarchie switch (reg) { 78865310Sarchie case VM_REG_GUEST_ES: 78965310Sarchie case VM_REG_GUEST_CS: 79065310Sarchie case VM_REG_GUEST_SS: 79165310Sarchie case VM_REG_GUEST_DS: 79265310Sarchie case VM_REG_GUEST_FS: 79365310Sarchie case VM_REG_GUEST_GS: 79465310Sarchie case VM_REG_GUEST_TR: 79565310Sarchie case VM_REG_GUEST_LDTR: 79665310Sarchie return (TRUE); 79765310Sarchie default: 79865310Sarchie return (FALSE); 79965310Sarchie } 80065310Sarchie} 80165310Sarchie 80265310Sarchieint 80365310Sarchievm_get_seg_desc(struct vm *vm, int vcpu, int reg, 80465310Sarchie struct seg_desc *desc) 80565310Sarchie{ 80665310Sarchie 80765310Sarchie if (vcpu < 0 || vcpu >= VM_MAXCPU) 80865310Sarchie return (EINVAL); 80965310Sarchie 81065310Sarchie if (!is_segment_register(reg) && !is_descriptor_table(reg)) 81165310Sarchie return (EINVAL); 81265310Sarchie 81365310Sarchie return (VMGETDESC(vm->cookie, vcpu, reg, desc)); 81465310Sarchie} 81565310Sarchie 81665310Sarchieint 81765310Sarchievm_set_seg_desc(struct vm *vm, int vcpu, int reg, 81865310Sarchie struct seg_desc *desc) 81965310Sarchie{ 82065310Sarchie if (vcpu < 0 || vcpu >= VM_MAXCPU) 82165310Sarchie return (EINVAL); 82265310Sarchie 82365310Sarchie if (!is_segment_register(reg) && !is_descriptor_table(reg)) 82465310Sarchie return (EINVAL); 82565310Sarchie 82665310Sarchie return (VMSETDESC(vm->cookie, vcpu, reg, desc)); 82765310Sarchie} 82865310Sarchie 82965310Sarchiestatic void 83065310Sarchierestore_guest_fpustate(struct vcpu *vcpu) 83165310Sarchie{ 83265310Sarchie 83365310Sarchie /* flush host state to the pcb */ 83465310Sarchie fpuexit(curthread); 83565310Sarchie 83665310Sarchie /* restore guest FPU state */ 83765310Sarchie fpu_stop_emulating(); 83865310Sarchie fpurestore(vcpu->guestfpu); 83965310Sarchie 84065310Sarchie /* restore guest XCR0 if XSAVE is enabled in the host */ 84165310Sarchie if (rcr4() & CR4_XSAVE) 84265310Sarchie load_xcr(0, vcpu->guest_xcr0); 84365310Sarchie 84465310Sarchie /* 84565310Sarchie * The FPU is now "dirty" with the guest's state so turn on emulation 84665310Sarchie * to trap any access to the FPU by the host. 84765310Sarchie */ 84865310Sarchie fpu_start_emulating(); 84965310Sarchie} 85065310Sarchie 85165310Sarchiestatic void 85265310Sarchiesave_guest_fpustate(struct vcpu *vcpu) 85365310Sarchie{ 85465310Sarchie 85565310Sarchie if ((rcr0() & CR0_TS) == 0) 85665310Sarchie panic("fpu emulation not enabled in host!"); 85765310Sarchie 85865310Sarchie /* save guest XCR0 and restore host XCR0 */ 85965310Sarchie if (rcr4() & CR4_XSAVE) { 86065310Sarchie vcpu->guest_xcr0 = rxcr(0); 86165310Sarchie load_xcr(0, vmm_get_host_xcr0()); 86265310Sarchie } 86365310Sarchie 86465310Sarchie /* save guest FPU state */ 86565310Sarchie fpu_stop_emulating(); 86665310Sarchie fpusave(vcpu->guestfpu); 86765310Sarchie fpu_start_emulating(); 86865310Sarchie} 86965310Sarchie 87065310Sarchiestatic VMM_STAT(VCPU_IDLE_TICKS, "number of ticks vcpu was idle"); 87165310Sarchie 87265310Sarchiestatic int 87365310Sarchievcpu_set_state_locked(struct vcpu *vcpu, enum vcpu_state newstate, 87465310Sarchie bool from_idle) 87565310Sarchie{ 87665310Sarchie int error; 87765310Sarchie 87865310Sarchie vcpu_assert_locked(vcpu); 87965310Sarchie 88065310Sarchie /* 88165310Sarchie * State transitions from the vmmdev_ioctl() must always begin from 88265310Sarchie * the VCPU_IDLE state. This guarantees that there is only a single 88365310Sarchie * ioctl() operating on a vcpu at any point. 88465310Sarchie */ 88565310Sarchie if (from_idle) { 88665310Sarchie while (vcpu->state != VCPU_IDLE) 88765310Sarchie msleep_spin(&vcpu->state, &vcpu->mtx, "vmstat", hz); 88865310Sarchie } else { 88965310Sarchie KASSERT(vcpu->state != VCPU_IDLE, ("invalid transition from " 89065310Sarchie "vcpu idle state")); 89165310Sarchie } 89265310Sarchie 89365310Sarchie if (vcpu->state == VCPU_RUNNING) { 89465310Sarchie KASSERT(vcpu->hostcpu == curcpu, ("curcpu %d and hostcpu %d " 89565310Sarchie "mismatch for running vcpu", curcpu, vcpu->hostcpu)); 89665310Sarchie } else { 89765310Sarchie KASSERT(vcpu->hostcpu == NOCPU, ("Invalid hostcpu %d for a " 89865310Sarchie "vcpu that is not running", vcpu->hostcpu)); 89965310Sarchie } 90065310Sarchie 90165310Sarchie /* 90265310Sarchie * The following state transitions are allowed: 90365310Sarchie * IDLE -> FROZEN -> IDLE 90465310Sarchie * FROZEN -> RUNNING -> FROZEN 90565310Sarchie * FROZEN -> SLEEPING -> FROZEN 90665310Sarchie */ 90765310Sarchie switch (vcpu->state) { 90865310Sarchie case VCPU_IDLE: 90965310Sarchie case VCPU_RUNNING: 91065310Sarchie case VCPU_SLEEPING: 91165310Sarchie error = (newstate != VCPU_FROZEN); 91265310Sarchie break; 91365310Sarchie case VCPU_FROZEN: 91465310Sarchie error = (newstate == VCPU_FROZEN); 91565310Sarchie break; 91665310Sarchie default: 91765310Sarchie error = 1; 91865310Sarchie break; 91965310Sarchie } 92065310Sarchie 92165310Sarchie if (error) 92265310Sarchie return (EBUSY); 92365310Sarchie 92465310Sarchie vcpu->state = newstate; 92565310Sarchie if (newstate == VCPU_RUNNING) 92665310Sarchie vcpu->hostcpu = curcpu; 92765310Sarchie else 92865310Sarchie vcpu->hostcpu = NOCPU; 92965310Sarchie 93065310Sarchie if (newstate == VCPU_IDLE) 93165310Sarchie wakeup(&vcpu->state); 93265310Sarchie 93365310Sarchie return (0); 93465310Sarchie} 93565310Sarchie 93665310Sarchiestatic void 93765310Sarchievcpu_require_state(struct vm *vm, int vcpuid, enum vcpu_state newstate) 93865310Sarchie{ 93965310Sarchie int error; 94065310Sarchie 94165310Sarchie if ((error = vcpu_set_state(vm, vcpuid, newstate, false)) != 0) 94265310Sarchie panic("Error %d setting state to %d\n", error, newstate); 94365310Sarchie} 94465310Sarchie 94565310Sarchiestatic void 94665310Sarchievcpu_require_state_locked(struct vcpu *vcpu, enum vcpu_state newstate) 94765310Sarchie{ 94865310Sarchie int error; 94965310Sarchie 95065310Sarchie if ((error = vcpu_set_state_locked(vcpu, newstate, false)) != 0) 95165310Sarchie panic("Error %d setting state to %d", error, newstate); 95265310Sarchie} 95365310Sarchie 95465310Sarchiestatic void 95565310Sarchievm_set_rendezvous_func(struct vm *vm, vm_rendezvous_func_t func) 95665310Sarchie{ 95765310Sarchie 95865310Sarchie KASSERT(mtx_owned(&vm->rendezvous_mtx), ("rendezvous_mtx not locked")); 95965310Sarchie 96065310Sarchie /* 96165310Sarchie * Update 'rendezvous_func' and execute a write memory barrier to 96265310Sarchie * ensure that it is visible across all host cpus. This is not needed 96365310Sarchie * for correctness but it does ensure that all the vcpus will notice 96465310Sarchie * that the rendezvous is requested immediately. 96565310Sarchie */ 96665310Sarchie vm->rendezvous_func = func; 96765310Sarchie wmb(); 96865310Sarchie} 96965310Sarchie 97065310Sarchie#define RENDEZVOUS_CTR0(vm, vcpuid, fmt) \ 97165310Sarchie do { \ 97265310Sarchie if (vcpuid >= 0) \ 97365310Sarchie VCPU_CTR0(vm, vcpuid, fmt); \ 97465310Sarchie else \ 97565310Sarchie VM_CTR0(vm, fmt); \ 97665310Sarchie } while (0) 97765310Sarchie 97865310Sarchiestatic void 97965310Sarchievm_handle_rendezvous(struct vm *vm, int vcpuid) 98065310Sarchie{ 98165310Sarchie 98265310Sarchie KASSERT(vcpuid == -1 || (vcpuid >= 0 && vcpuid < VM_MAXCPU), 98365310Sarchie ("vm_handle_rendezvous: invalid vcpuid %d", vcpuid)); 98465310Sarchie 98565310Sarchie mtx_lock(&vm->rendezvous_mtx); 98665310Sarchie while (vm->rendezvous_func != NULL) { 98765310Sarchie /* 'rendezvous_req_cpus' must be a subset of 'active_cpus' */ 98865310Sarchie CPU_AND(&vm->rendezvous_req_cpus, &vm->active_cpus); 98965310Sarchie 99065310Sarchie if (vcpuid != -1 && 99165310Sarchie CPU_ISSET(vcpuid, &vm->rendezvous_req_cpus) && 99265310Sarchie !CPU_ISSET(vcpuid, &vm->rendezvous_done_cpus)) { 99365310Sarchie VCPU_CTR0(vm, vcpuid, "Calling rendezvous func"); 99465310Sarchie (*vm->rendezvous_func)(vm, vcpuid, vm->rendezvous_arg); 99565310Sarchie CPU_SET(vcpuid, &vm->rendezvous_done_cpus); 99665310Sarchie } 99765310Sarchie if (CPU_CMP(&vm->rendezvous_req_cpus, 99865310Sarchie &vm->rendezvous_done_cpus) == 0) { 99965310Sarchie VCPU_CTR0(vm, vcpuid, "Rendezvous completed"); 100065310Sarchie vm_set_rendezvous_func(vm, NULL); 100165310Sarchie wakeup(&vm->rendezvous_func); 100265310Sarchie break; 100365310Sarchie } 100465310Sarchie RENDEZVOUS_CTR0(vm, vcpuid, "Wait for rendezvous completion"); 100565310Sarchie mtx_sleep(&vm->rendezvous_func, &vm->rendezvous_mtx, 0, 100665310Sarchie "vmrndv", 0); 100765310Sarchie } 100865310Sarchie mtx_unlock(&vm->rendezvous_mtx); 100965310Sarchie} 101065310Sarchie 101165310Sarchie/* 101265310Sarchie * Emulate a guest 'hlt' by sleeping until the vcpu is ready to run. 101365310Sarchie */ 101465310Sarchiestatic int 1015vm_handle_hlt(struct vm *vm, int vcpuid, bool intr_disabled, bool *retu) 1016{ 1017 struct vcpu *vcpu; 1018 const char *wmesg; 1019 int t, vcpu_halted, vm_halted; 1020 1021 KASSERT(!CPU_ISSET(vcpuid, &vm->halted_cpus), ("vcpu already halted")); 1022 1023 vcpu = &vm->vcpu[vcpuid]; 1024 vcpu_halted = 0; 1025 vm_halted = 0; 1026 1027 vcpu_lock(vcpu); 1028 while (1) { 1029 /* 1030 * Do a final check for pending NMI or interrupts before 1031 * really putting this thread to sleep. Also check for 1032 * software events that would cause this vcpu to wakeup. 1033 * 1034 * These interrupts/events could have happened after the 1035 * vcpu returned from VMRUN() and before it acquired the 1036 * vcpu lock above. 1037 */ 1038 if (vm->rendezvous_func != NULL || vm->suspend) 1039 break; 1040 if (vm_nmi_pending(vm, vcpuid)) 1041 break; 1042 if (!intr_disabled) { 1043 if (vm_extint_pending(vm, vcpuid) || 1044 vlapic_pending_intr(vcpu->vlapic, NULL)) { 1045 break; 1046 } 1047 } 1048 1049 /* 1050 * Some Linux guests implement "halt" by having all vcpus 1051 * execute HLT with interrupts disabled. 'halted_cpus' keeps 1052 * track of the vcpus that have entered this state. When all 1053 * vcpus enter the halted state the virtual machine is halted. 1054 */ 1055 if (intr_disabled) { 1056 wmesg = "vmhalt"; 1057 VCPU_CTR0(vm, vcpuid, "Halted"); 1058 if (!vcpu_halted && halt_detection_enabled) { 1059 vcpu_halted = 1; 1060 CPU_SET_ATOMIC(vcpuid, &vm->halted_cpus); 1061 } 1062 if (CPU_CMP(&vm->halted_cpus, &vm->active_cpus) == 0) { 1063 vm_halted = 1; 1064 break; 1065 } 1066 } else { 1067 wmesg = "vmidle"; 1068 } 1069 1070 t = ticks; 1071 vcpu_require_state_locked(vcpu, VCPU_SLEEPING); 1072 msleep_spin(vcpu, &vcpu->mtx, wmesg, 0); 1073 vcpu_require_state_locked(vcpu, VCPU_FROZEN); 1074 vmm_stat_incr(vm, vcpuid, VCPU_IDLE_TICKS, ticks - t); 1075 } 1076 1077 if (vcpu_halted) 1078 CPU_CLR_ATOMIC(vcpuid, &vm->halted_cpus); 1079 1080 vcpu_unlock(vcpu); 1081 1082 if (vm_halted) 1083 vm_suspend(vm, VM_SUSPEND_HALT); 1084 1085 return (0); 1086} 1087 1088static int 1089vm_handle_paging(struct vm *vm, int vcpuid, bool *retu) 1090{ 1091 int rv, ftype; 1092 struct vm_map *map; 1093 struct vcpu *vcpu; 1094 struct vm_exit *vme; 1095 1096 vcpu = &vm->vcpu[vcpuid]; 1097 vme = &vcpu->exitinfo; 1098 1099 ftype = vme->u.paging.fault_type; 1100 KASSERT(ftype == VM_PROT_READ || 1101 ftype == VM_PROT_WRITE || ftype == VM_PROT_EXECUTE, 1102 ("vm_handle_paging: invalid fault_type %d", ftype)); 1103 1104 if (ftype == VM_PROT_READ || ftype == VM_PROT_WRITE) { 1105 rv = pmap_emulate_accessed_dirty(vmspace_pmap(vm->vmspace), 1106 vme->u.paging.gpa, ftype); 1107 if (rv == 0) 1108 goto done; 1109 } 1110 1111 map = &vm->vmspace->vm_map; 1112 rv = vm_fault(map, vme->u.paging.gpa, ftype, VM_FAULT_NORMAL); 1113 1114 VCPU_CTR3(vm, vcpuid, "vm_handle_paging rv = %d, gpa = %#lx, " 1115 "ftype = %d", rv, vme->u.paging.gpa, ftype); 1116 1117 if (rv != KERN_SUCCESS) 1118 return (EFAULT); 1119done: 1120 /* restart execution at the faulting instruction */ 1121 vme->inst_length = 0; 1122 1123 return (0); 1124} 1125 1126static int 1127vm_handle_inst_emul(struct vm *vm, int vcpuid, bool *retu) 1128{ 1129 struct vie *vie; 1130 struct vcpu *vcpu; 1131 struct vm_exit *vme; 1132 uint64_t gla, gpa; 1133 struct vm_guest_paging *paging; 1134 mem_region_read_t mread; 1135 mem_region_write_t mwrite; 1136 int error; 1137 1138 vcpu = &vm->vcpu[vcpuid]; 1139 vme = &vcpu->exitinfo; 1140 1141 gla = vme->u.inst_emul.gla; 1142 gpa = vme->u.inst_emul.gpa; 1143 vie = &vme->u.inst_emul.vie; 1144 paging = &vme->u.inst_emul.paging; 1145 1146 vie_init(vie); 1147 1148 /* Fetch, decode and emulate the faulting instruction */ 1149 error = vmm_fetch_instruction(vm, vcpuid, paging, vme->rip, 1150 vme->inst_length, vie); 1151 if (error == 1) 1152 return (0); /* Resume guest to handle page fault */ 1153 else if (error == -1) 1154 return (EFAULT); 1155 else if (error != 0) 1156 panic("%s: vmm_fetch_instruction error %d", __func__, error); 1157 1158 if (vmm_decode_instruction(vm, vcpuid, gla, paging->cpu_mode, vie) != 0) 1159 return (EFAULT); 1160 1161 /* return to userland unless this is an in-kernel emulated device */ 1162 if (gpa >= DEFAULT_APIC_BASE && gpa < DEFAULT_APIC_BASE + PAGE_SIZE) { 1163 mread = lapic_mmio_read; 1164 mwrite = lapic_mmio_write; 1165 } else if (gpa >= VIOAPIC_BASE && gpa < VIOAPIC_BASE + VIOAPIC_SIZE) { 1166 mread = vioapic_mmio_read; 1167 mwrite = vioapic_mmio_write; 1168 } else if (gpa >= VHPET_BASE && gpa < VHPET_BASE + VHPET_SIZE) { 1169 mread = vhpet_mmio_read; 1170 mwrite = vhpet_mmio_write; 1171 } else { 1172 *retu = true; 1173 return (0); 1174 } 1175 1176 error = vmm_emulate_instruction(vm, vcpuid, gpa, vie, mread, mwrite, 1177 retu); 1178 1179 return (error); 1180} 1181 1182static int 1183vm_handle_suspend(struct vm *vm, int vcpuid, bool *retu) 1184{ 1185 int i, done; 1186 struct vcpu *vcpu; 1187 1188 done = 0; 1189 vcpu = &vm->vcpu[vcpuid]; 1190 1191 CPU_SET_ATOMIC(vcpuid, &vm->suspended_cpus); 1192 1193 /* 1194 * Wait until all 'active_cpus' have suspended themselves. 1195 * 1196 * Since a VM may be suspended at any time including when one or 1197 * more vcpus are doing a rendezvous we need to call the rendezvous 1198 * handler while we are waiting to prevent a deadlock. 1199 */ 1200 vcpu_lock(vcpu); 1201 while (1) { 1202 if (CPU_CMP(&vm->suspended_cpus, &vm->active_cpus) == 0) { 1203 VCPU_CTR0(vm, vcpuid, "All vcpus suspended"); 1204 break; 1205 } 1206 1207 if (vm->rendezvous_func == NULL) { 1208 VCPU_CTR0(vm, vcpuid, "Sleeping during suspend"); 1209 vcpu_require_state_locked(vcpu, VCPU_SLEEPING); 1210 msleep_spin(vcpu, &vcpu->mtx, "vmsusp", hz); 1211 vcpu_require_state_locked(vcpu, VCPU_FROZEN); 1212 } else { 1213 VCPU_CTR0(vm, vcpuid, "Rendezvous during suspend"); 1214 vcpu_unlock(vcpu); 1215 vm_handle_rendezvous(vm, vcpuid); 1216 vcpu_lock(vcpu); 1217 } 1218 } 1219 vcpu_unlock(vcpu); 1220 1221 /* 1222 * Wakeup the other sleeping vcpus and return to userspace. 1223 */ 1224 for (i = 0; i < VM_MAXCPU; i++) { 1225 if (CPU_ISSET(i, &vm->suspended_cpus)) { 1226 vcpu_notify_event(vm, i, false); 1227 } 1228 } 1229 1230 *retu = true; 1231 return (0); 1232} 1233 1234int 1235vm_suspend(struct vm *vm, enum vm_suspend_how how) 1236{ 1237 int i; 1238 1239 if (how <= VM_SUSPEND_NONE || how >= VM_SUSPEND_LAST) 1240 return (EINVAL); 1241 1242 if (atomic_cmpset_int(&vm->suspend, 0, how) == 0) { 1243 VM_CTR2(vm, "virtual machine already suspended %d/%d", 1244 vm->suspend, how); 1245 return (EALREADY); 1246 } 1247 1248 VM_CTR1(vm, "virtual machine successfully suspended %d", how); 1249 1250 /* 1251 * Notify all active vcpus that they are now suspended. 1252 */ 1253 for (i = 0; i < VM_MAXCPU; i++) { 1254 if (CPU_ISSET(i, &vm->active_cpus)) 1255 vcpu_notify_event(vm, i, false); 1256 } 1257 1258 return (0); 1259} 1260 1261void 1262vm_exit_suspended(struct vm *vm, int vcpuid, uint64_t rip) 1263{ 1264 struct vm_exit *vmexit; 1265 1266 KASSERT(vm->suspend > VM_SUSPEND_NONE && vm->suspend < VM_SUSPEND_LAST, 1267 ("vm_exit_suspended: invalid suspend type %d", vm->suspend)); 1268 1269 vmexit = vm_exitinfo(vm, vcpuid); 1270 vmexit->rip = rip; 1271 vmexit->inst_length = 0; 1272 vmexit->exitcode = VM_EXITCODE_SUSPENDED; 1273 vmexit->u.suspended.how = vm->suspend; 1274} 1275 1276int 1277vm_run(struct vm *vm, struct vm_run *vmrun) 1278{ 1279 int error, vcpuid; 1280 struct vcpu *vcpu; 1281 struct pcb *pcb; 1282 uint64_t tscval, rip; 1283 struct vm_exit *vme; 1284 bool retu, intr_disabled; 1285 pmap_t pmap; 1286 void *rptr, *sptr; 1287 1288 vcpuid = vmrun->cpuid; 1289 1290 if (vcpuid < 0 || vcpuid >= VM_MAXCPU) 1291 return (EINVAL); 1292 1293 if (!CPU_ISSET(vcpuid, &vm->active_cpus)) 1294 return (EINVAL); 1295 1296 if (CPU_ISSET(vcpuid, &vm->suspended_cpus)) 1297 return (EINVAL); 1298 1299 rptr = &vm->rendezvous_func; 1300 sptr = &vm->suspend; 1301 pmap = vmspace_pmap(vm->vmspace); 1302 vcpu = &vm->vcpu[vcpuid]; 1303 vme = &vcpu->exitinfo; 1304 rip = vmrun->rip; 1305restart: 1306 critical_enter(); 1307 1308 KASSERT(!CPU_ISSET(curcpu, &pmap->pm_active), 1309 ("vm_run: absurd pm_active")); 1310 1311 tscval = rdtsc(); 1312 1313 pcb = PCPU_GET(curpcb); 1314 set_pcb_flags(pcb, PCB_FULL_IRET); 1315 1316 restore_guest_msrs(vm, vcpuid); 1317 restore_guest_fpustate(vcpu); 1318 1319 vcpu_require_state(vm, vcpuid, VCPU_RUNNING); 1320 error = VMRUN(vm->cookie, vcpuid, rip, pmap, rptr, sptr); 1321 vcpu_require_state(vm, vcpuid, VCPU_FROZEN); 1322 1323 save_guest_fpustate(vcpu); 1324 restore_host_msrs(vm, vcpuid); 1325 1326 vmm_stat_incr(vm, vcpuid, VCPU_TOTAL_RUNTIME, rdtsc() - tscval); 1327 1328 critical_exit(); 1329 1330 if (error == 0) { 1331 retu = false; 1332 switch (vme->exitcode) { 1333 case VM_EXITCODE_SUSPENDED: 1334 error = vm_handle_suspend(vm, vcpuid, &retu); 1335 break; 1336 case VM_EXITCODE_IOAPIC_EOI: 1337 vioapic_process_eoi(vm, vcpuid, 1338 vme->u.ioapic_eoi.vector); 1339 break; 1340 case VM_EXITCODE_RENDEZVOUS: 1341 vm_handle_rendezvous(vm, vcpuid); 1342 error = 0; 1343 break; 1344 case VM_EXITCODE_HLT: 1345 intr_disabled = ((vme->u.hlt.rflags & PSL_I) == 0); 1346 error = vm_handle_hlt(vm, vcpuid, intr_disabled, &retu); 1347 break; 1348 case VM_EXITCODE_PAGING: 1349 error = vm_handle_paging(vm, vcpuid, &retu); 1350 break; 1351 case VM_EXITCODE_INST_EMUL: 1352 error = vm_handle_inst_emul(vm, vcpuid, &retu); 1353 break; 1354 case VM_EXITCODE_INOUT: 1355 case VM_EXITCODE_INOUT_STR: 1356 error = vm_handle_inout(vm, vcpuid, vme, &retu); 1357 break; 1358 default: 1359 retu = true; /* handled in userland */ 1360 break; 1361 } 1362 } 1363 1364 if (error == 0 && retu == false) { 1365 rip = vme->rip + vme->inst_length; 1366 goto restart; 1367 } 1368 1369 /* copy the exit information */ 1370 bcopy(vme, &vmrun->vm_exit, sizeof(struct vm_exit)); 1371 return (error); 1372} 1373 1374int 1375vm_inject_exception(struct vm *vm, int vcpuid, struct vm_exception *exception) 1376{ 1377 struct vcpu *vcpu; 1378 1379 if (vcpuid < 0 || vcpuid >= VM_MAXCPU) 1380 return (EINVAL); 1381 1382 if (exception->vector < 0 || exception->vector >= 32) 1383 return (EINVAL); 1384 1385 vcpu = &vm->vcpu[vcpuid]; 1386 1387 if (vcpu->exception_pending) { 1388 VCPU_CTR2(vm, vcpuid, "Unable to inject exception %d due to " 1389 "pending exception %d", exception->vector, 1390 vcpu->exception.vector); 1391 return (EBUSY); 1392 } 1393 1394 vcpu->exception_pending = 1; 1395 vcpu->exception = *exception; 1396 VCPU_CTR1(vm, vcpuid, "Exception %d pending", exception->vector); 1397 return (0); 1398} 1399 1400int 1401vm_exception_pending(struct vm *vm, int vcpuid, struct vm_exception *exception) 1402{ 1403 struct vcpu *vcpu; 1404 int pending; 1405 1406 KASSERT(vcpuid >= 0 && vcpuid < VM_MAXCPU, ("invalid vcpu %d", vcpuid)); 1407 1408 vcpu = &vm->vcpu[vcpuid]; 1409 pending = vcpu->exception_pending; 1410 if (pending) { 1411 vcpu->exception_pending = 0; 1412 *exception = vcpu->exception; 1413 VCPU_CTR1(vm, vcpuid, "Exception %d delivered", 1414 exception->vector); 1415 } 1416 return (pending); 1417} 1418 1419static void 1420vm_inject_fault(struct vm *vm, int vcpuid, struct vm_exception *exception) 1421{ 1422 struct vm_exit *vmexit; 1423 int error; 1424 1425 error = vm_inject_exception(vm, vcpuid, exception); 1426 KASSERT(error == 0, ("vm_inject_exception error %d", error)); 1427 1428 /* 1429 * A fault-like exception allows the instruction to be restarted 1430 * after the exception handler returns. 1431 * 1432 * By setting the inst_length to 0 we ensure that the instruction 1433 * pointer remains at the faulting instruction. 1434 */ 1435 vmexit = vm_exitinfo(vm, vcpuid); 1436 vmexit->inst_length = 0; 1437} 1438 1439void 1440vm_inject_pf(struct vm *vm, int vcpuid, int error_code, uint64_t cr2) 1441{ 1442 struct vm_exception pf = { 1443 .vector = IDT_PF, 1444 .error_code_valid = 1, 1445 .error_code = error_code 1446 }; 1447 int error; 1448 1449 VCPU_CTR2(vm, vcpuid, "Injecting page fault: error_code %#x, cr2 %#lx", 1450 error_code, cr2); 1451 1452 error = vm_set_register(vm, vcpuid, VM_REG_GUEST_CR2, cr2); 1453 KASSERT(error == 0, ("vm_set_register(cr2) error %d", error)); 1454 1455 vm_inject_fault(vm, vcpuid, &pf); 1456} 1457 1458void 1459vm_inject_gp(struct vm *vm, int vcpuid) 1460{ 1461 struct vm_exception gpf = { 1462 .vector = IDT_GP, 1463 .error_code_valid = 1, 1464 .error_code = 0 1465 }; 1466 1467 vm_inject_fault(vm, vcpuid, &gpf); 1468} 1469 1470void 1471vm_inject_ud(struct vm *vm, int vcpuid) 1472{ 1473 struct vm_exception udf = { 1474 .vector = IDT_UD, 1475 .error_code_valid = 0 1476 }; 1477 1478 vm_inject_fault(vm, vcpuid, &udf); 1479} 1480 1481static VMM_STAT(VCPU_NMI_COUNT, "number of NMIs delivered to vcpu"); 1482 1483int 1484vm_inject_nmi(struct vm *vm, int vcpuid) 1485{ 1486 struct vcpu *vcpu; 1487 1488 if (vcpuid < 0 || vcpuid >= VM_MAXCPU) 1489 return (EINVAL); 1490 1491 vcpu = &vm->vcpu[vcpuid]; 1492 1493 vcpu->nmi_pending = 1; 1494 vcpu_notify_event(vm, vcpuid, false); 1495 return (0); 1496} 1497 1498int 1499vm_nmi_pending(struct vm *vm, int vcpuid) 1500{ 1501 struct vcpu *vcpu; 1502 1503 if (vcpuid < 0 || vcpuid >= VM_MAXCPU) 1504 panic("vm_nmi_pending: invalid vcpuid %d", vcpuid); 1505 1506 vcpu = &vm->vcpu[vcpuid]; 1507 1508 return (vcpu->nmi_pending); 1509} 1510 1511void 1512vm_nmi_clear(struct vm *vm, int vcpuid) 1513{ 1514 struct vcpu *vcpu; 1515 1516 if (vcpuid < 0 || vcpuid >= VM_MAXCPU) 1517 panic("vm_nmi_pending: invalid vcpuid %d", vcpuid); 1518 1519 vcpu = &vm->vcpu[vcpuid]; 1520 1521 if (vcpu->nmi_pending == 0) 1522 panic("vm_nmi_clear: inconsistent nmi_pending state"); 1523 1524 vcpu->nmi_pending = 0; 1525 vmm_stat_incr(vm, vcpuid, VCPU_NMI_COUNT, 1); 1526} 1527 1528static VMM_STAT(VCPU_EXTINT_COUNT, "number of ExtINTs delivered to vcpu"); 1529 1530int 1531vm_inject_extint(struct vm *vm, int vcpuid) 1532{ 1533 struct vcpu *vcpu; 1534 1535 if (vcpuid < 0 || vcpuid >= VM_MAXCPU) 1536 return (EINVAL); 1537 1538 vcpu = &vm->vcpu[vcpuid]; 1539 1540 vcpu->extint_pending = 1; 1541 vcpu_notify_event(vm, vcpuid, false); 1542 return (0); 1543} 1544 1545int 1546vm_extint_pending(struct vm *vm, int vcpuid) 1547{ 1548 struct vcpu *vcpu; 1549 1550 if (vcpuid < 0 || vcpuid >= VM_MAXCPU) 1551 panic("vm_extint_pending: invalid vcpuid %d", vcpuid); 1552 1553 vcpu = &vm->vcpu[vcpuid]; 1554 1555 return (vcpu->extint_pending); 1556} 1557 1558void 1559vm_extint_clear(struct vm *vm, int vcpuid) 1560{ 1561 struct vcpu *vcpu; 1562 1563 if (vcpuid < 0 || vcpuid >= VM_MAXCPU) 1564 panic("vm_extint_pending: invalid vcpuid %d", vcpuid); 1565 1566 vcpu = &vm->vcpu[vcpuid]; 1567 1568 if (vcpu->extint_pending == 0) 1569 panic("vm_extint_clear: inconsistent extint_pending state"); 1570 1571 vcpu->extint_pending = 0; 1572 vmm_stat_incr(vm, vcpuid, VCPU_EXTINT_COUNT, 1); 1573} 1574 1575int 1576vm_get_capability(struct vm *vm, int vcpu, int type, int *retval) 1577{ 1578 if (vcpu < 0 || vcpu >= VM_MAXCPU) 1579 return (EINVAL); 1580 1581 if (type < 0 || type >= VM_CAP_MAX) 1582 return (EINVAL); 1583 1584 return (VMGETCAP(vm->cookie, vcpu, type, retval)); 1585} 1586 1587int 1588vm_set_capability(struct vm *vm, int vcpu, int type, int val) 1589{ 1590 if (vcpu < 0 || vcpu >= VM_MAXCPU) 1591 return (EINVAL); 1592 1593 if (type < 0 || type >= VM_CAP_MAX) 1594 return (EINVAL); 1595 1596 return (VMSETCAP(vm->cookie, vcpu, type, val)); 1597} 1598 1599uint64_t * 1600vm_guest_msrs(struct vm *vm, int cpu) 1601{ 1602 return (vm->vcpu[cpu].guest_msrs); 1603} 1604 1605struct vlapic * 1606vm_lapic(struct vm *vm, int cpu) 1607{ 1608 return (vm->vcpu[cpu].vlapic); 1609} 1610 1611struct vioapic * 1612vm_ioapic(struct vm *vm) 1613{ 1614 1615 return (vm->vioapic); 1616} 1617 1618struct vhpet * 1619vm_hpet(struct vm *vm) 1620{ 1621 1622 return (vm->vhpet); 1623} 1624 1625boolean_t 1626vmm_is_pptdev(int bus, int slot, int func) 1627{ 1628 int found, i, n; 1629 int b, s, f; 1630 char *val, *cp, *cp2; 1631 1632 /* 1633 * XXX 1634 * The length of an environment variable is limited to 128 bytes which 1635 * puts an upper limit on the number of passthru devices that may be 1636 * specified using a single environment variable. 1637 * 1638 * Work around this by scanning multiple environment variable 1639 * names instead of a single one - yuck! 1640 */ 1641 const char *names[] = { "pptdevs", "pptdevs2", "pptdevs3", NULL }; 1642 1643 /* set pptdevs="1/2/3 4/5/6 7/8/9 10/11/12" */ 1644 found = 0; 1645 for (i = 0; names[i] != NULL && !found; i++) { 1646 cp = val = getenv(names[i]); 1647 while (cp != NULL && *cp != '\0') { 1648 if ((cp2 = strchr(cp, ' ')) != NULL) 1649 *cp2 = '\0'; 1650 1651 n = sscanf(cp, "%d/%d/%d", &b, &s, &f); 1652 if (n == 3 && bus == b && slot == s && func == f) { 1653 found = 1; 1654 break; 1655 } 1656 1657 if (cp2 != NULL) 1658 *cp2++ = ' '; 1659 1660 cp = cp2; 1661 } 1662 freeenv(val); 1663 } 1664 return (found); 1665} 1666 1667void * 1668vm_iommu_domain(struct vm *vm) 1669{ 1670 1671 return (vm->iommu); 1672} 1673 1674int 1675vcpu_set_state(struct vm *vm, int vcpuid, enum vcpu_state newstate, 1676 bool from_idle) 1677{ 1678 int error; 1679 struct vcpu *vcpu; 1680 1681 if (vcpuid < 0 || vcpuid >= VM_MAXCPU) 1682 panic("vm_set_run_state: invalid vcpuid %d", vcpuid); 1683 1684 vcpu = &vm->vcpu[vcpuid]; 1685 1686 vcpu_lock(vcpu); 1687 error = vcpu_set_state_locked(vcpu, newstate, from_idle); 1688 vcpu_unlock(vcpu); 1689 1690 return (error); 1691} 1692 1693enum vcpu_state 1694vcpu_get_state(struct vm *vm, int vcpuid, int *hostcpu) 1695{ 1696 struct vcpu *vcpu; 1697 enum vcpu_state state; 1698 1699 if (vcpuid < 0 || vcpuid >= VM_MAXCPU) 1700 panic("vm_get_run_state: invalid vcpuid %d", vcpuid); 1701 1702 vcpu = &vm->vcpu[vcpuid]; 1703 1704 vcpu_lock(vcpu); 1705 state = vcpu->state; 1706 if (hostcpu != NULL) 1707 *hostcpu = vcpu->hostcpu; 1708 vcpu_unlock(vcpu); 1709 1710 return (state); 1711} 1712 1713int 1714vm_activate_cpu(struct vm *vm, int vcpuid) 1715{ 1716 1717 if (vcpuid < 0 || vcpuid >= VM_MAXCPU) 1718 return (EINVAL); 1719 1720 if (CPU_ISSET(vcpuid, &vm->active_cpus)) 1721 return (EBUSY); 1722 1723 VCPU_CTR0(vm, vcpuid, "activated"); 1724 CPU_SET_ATOMIC(vcpuid, &vm->active_cpus); 1725 return (0); 1726} 1727 1728cpuset_t 1729vm_active_cpus(struct vm *vm) 1730{ 1731 1732 return (vm->active_cpus); 1733} 1734 1735cpuset_t 1736vm_suspended_cpus(struct vm *vm) 1737{ 1738 1739 return (vm->suspended_cpus); 1740} 1741 1742void * 1743vcpu_stats(struct vm *vm, int vcpuid) 1744{ 1745 1746 return (vm->vcpu[vcpuid].stats); 1747} 1748 1749int 1750vm_get_x2apic_state(struct vm *vm, int vcpuid, enum x2apic_state *state) 1751{ 1752 if (vcpuid < 0 || vcpuid >= VM_MAXCPU) 1753 return (EINVAL); 1754 1755 *state = vm->vcpu[vcpuid].x2apic_state; 1756 1757 return (0); 1758} 1759 1760int 1761vm_set_x2apic_state(struct vm *vm, int vcpuid, enum x2apic_state state) 1762{ 1763 if (vcpuid < 0 || vcpuid >= VM_MAXCPU) 1764 return (EINVAL); 1765 1766 if (state >= X2APIC_STATE_LAST) 1767 return (EINVAL); 1768 1769 vm->vcpu[vcpuid].x2apic_state = state; 1770 1771 vlapic_set_x2apic_state(vm, vcpuid, state); 1772 1773 return (0); 1774} 1775 1776/* 1777 * This function is called to ensure that a vcpu "sees" a pending event 1778 * as soon as possible: 1779 * - If the vcpu thread is sleeping then it is woken up. 1780 * - If the vcpu is running on a different host_cpu then an IPI will be directed 1781 * to the host_cpu to cause the vcpu to trap into the hypervisor. 1782 */ 1783void 1784vcpu_notify_event(struct vm *vm, int vcpuid, bool lapic_intr) 1785{ 1786 int hostcpu; 1787 struct vcpu *vcpu; 1788 1789 vcpu = &vm->vcpu[vcpuid]; 1790 1791 vcpu_lock(vcpu); 1792 hostcpu = vcpu->hostcpu; 1793 if (vcpu->state == VCPU_RUNNING) { 1794 KASSERT(hostcpu != NOCPU, ("vcpu running on invalid hostcpu")); 1795 if (hostcpu != curcpu) { 1796 if (lapic_intr) { 1797 vlapic_post_intr(vcpu->vlapic, hostcpu, 1798 vmm_ipinum); 1799 } else { 1800 ipi_cpu(hostcpu, vmm_ipinum); 1801 } 1802 } else { 1803 /* 1804 * If the 'vcpu' is running on 'curcpu' then it must 1805 * be sending a notification to itself (e.g. SELF_IPI). 1806 * The pending event will be picked up when the vcpu 1807 * transitions back to guest context. 1808 */ 1809 } 1810 } else { 1811 KASSERT(hostcpu == NOCPU, ("vcpu state %d not consistent " 1812 "with hostcpu %d", vcpu->state, hostcpu)); 1813 if (vcpu->state == VCPU_SLEEPING) 1814 wakeup_one(vcpu); 1815 } 1816 vcpu_unlock(vcpu); 1817} 1818 1819struct vmspace * 1820vm_get_vmspace(struct vm *vm) 1821{ 1822 1823 return (vm->vmspace); 1824} 1825 1826int 1827vm_apicid2vcpuid(struct vm *vm, int apicid) 1828{ 1829 /* 1830 * XXX apic id is assumed to be numerically identical to vcpu id 1831 */ 1832 return (apicid); 1833} 1834 1835void 1836vm_smp_rendezvous(struct vm *vm, int vcpuid, cpuset_t dest, 1837 vm_rendezvous_func_t func, void *arg) 1838{ 1839 int i; 1840 1841 /* 1842 * Enforce that this function is called without any locks 1843 */ 1844 WITNESS_WARN(WARN_PANIC, NULL, "vm_smp_rendezvous"); 1845 KASSERT(vcpuid == -1 || (vcpuid >= 0 && vcpuid < VM_MAXCPU), 1846 ("vm_smp_rendezvous: invalid vcpuid %d", vcpuid)); 1847 1848restart: 1849 mtx_lock(&vm->rendezvous_mtx); 1850 if (vm->rendezvous_func != NULL) { 1851 /* 1852 * If a rendezvous is already in progress then we need to 1853 * call the rendezvous handler in case this 'vcpuid' is one 1854 * of the targets of the rendezvous. 1855 */ 1856 RENDEZVOUS_CTR0(vm, vcpuid, "Rendezvous already in progress"); 1857 mtx_unlock(&vm->rendezvous_mtx); 1858 vm_handle_rendezvous(vm, vcpuid); 1859 goto restart; 1860 } 1861 KASSERT(vm->rendezvous_func == NULL, ("vm_smp_rendezvous: previous " 1862 "rendezvous is still in progress")); 1863 1864 RENDEZVOUS_CTR0(vm, vcpuid, "Initiating rendezvous"); 1865 vm->rendezvous_req_cpus = dest; 1866 CPU_ZERO(&vm->rendezvous_done_cpus); 1867 vm->rendezvous_arg = arg; 1868 vm_set_rendezvous_func(vm, func); 1869 mtx_unlock(&vm->rendezvous_mtx); 1870 1871 /* 1872 * Wake up any sleeping vcpus and trigger a VM-exit in any running 1873 * vcpus so they handle the rendezvous as soon as possible. 1874 */ 1875 for (i = 0; i < VM_MAXCPU; i++) { 1876 if (CPU_ISSET(i, &dest)) 1877 vcpu_notify_event(vm, i, false); 1878 } 1879 1880 vm_handle_rendezvous(vm, vcpuid); 1881} 1882 1883struct vatpic * 1884vm_atpic(struct vm *vm) 1885{ 1886 return (vm->vatpic); 1887} 1888 1889struct vatpit * 1890vm_atpit(struct vm *vm) 1891{ 1892 return (vm->vatpit); 1893} 1894 1895enum vm_reg_name 1896vm_segment_name(int seg) 1897{ 1898 static enum vm_reg_name seg_names[] = { 1899 VM_REG_GUEST_ES, 1900 VM_REG_GUEST_CS, 1901 VM_REG_GUEST_SS, 1902 VM_REG_GUEST_DS, 1903 VM_REG_GUEST_FS, 1904 VM_REG_GUEST_GS 1905 }; 1906 1907 KASSERT(seg >= 0 && seg < nitems(seg_names), 1908 ("%s: invalid segment encoding %d", __func__, seg)); 1909 return (seg_names[seg]); 1910} 1911