2 * Copyright (c) 2011 NetApp, Inc.
5 * Redistribution and use in source and binary forms, with or without
6 * modification, are permitted provided that the following conditions
8 * 1. Redistributions of source code must retain the above copyright
9 * notice, this list of conditions and the following disclaimer.
10 * 2. Redistributions in binary form must reproduce the above copyright
11 * notice, this list of conditions and the following disclaimer in the
12 * documentation and/or other materials provided with the distribution.
14 * THIS SOFTWARE IS PROVIDED BY NETAPP, INC ``AS IS'' AND
15 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
16 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
17 * ARE DISCLAIMED. IN NO EVENT SHALL NETAPP, INC OR CONTRIBUTORS BE LIABLE
18 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
19 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
20 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
21 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
22 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
23 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
29 #include <sys/cdefs.h>
30 __FBSDID("$FreeBSD$");
32 #include <sys/param.h>
33 #include <sys/systm.h>
34 #include <sys/kernel.h>
35 #include <sys/module.h>
36 #include <sys/sysctl.h>
37 #include <sys/malloc.h>
40 #include <sys/mutex.h>
42 #include <sys/rwlock.h>
43 #include <sys/sched.h>
45 #include <sys/systm.h>
48 #include <vm/vm_object.h>
49 #include <vm/vm_page.h>
51 #include <vm/vm_map.h>
52 #include <vm/vm_extern.h>
53 #include <vm/vm_param.h>
55 #include <machine/cpu.h>
56 #include <machine/vm.h>
57 #include <machine/pcb.h>
58 #include <machine/smp.h>
60 #include <x86/apicreg.h>
61 #include <machine/vmparam.h>
63 #include <machine/vmm.h>
64 #include <machine/vmm_dev.h>
65 #include <machine/vmm_instruction_emul.h>
67 #include "vmm_ioport.h"
80 #include "vmm_lapic.h"
89 * (a) allocated when vcpu is created
90 * (i) initialized when vcpu is created and when it is reinitialized
91 * (o) initialized the first time the vcpu is created
92 * (x) initialized before use
95 struct mtx mtx; /* (o) protects 'state' and 'hostcpu' */
96 enum vcpu_state state; /* (o) vcpu state */
97 int hostcpu; /* (o) vcpu's host cpu */
98 int reqidle; /* (i) request vcpu to idle */
99 struct vlapic *vlapic; /* (i) APIC device model */
100 enum x2apic_state x2apic_state; /* (i) APIC mode */
101 uint64_t exitintinfo; /* (i) events pending at VM exit */
102 int nmi_pending; /* (i) NMI pending */
103 int extint_pending; /* (i) INTR pending */
104 int exception_pending; /* (i) exception pending */
105 int exc_vector; /* (x) exception collateral */
106 int exc_errcode_valid;
107 uint32_t exc_errcode;
108 struct savefpu *guestfpu; /* (a,i) guest fpu state */
109 uint64_t guest_xcr0; /* (i) guest %xcr0 register */
110 void *stats; /* (a,i) statistics */
111 struct vm_exit exitinfo; /* (x) exit reason and collateral */
112 uint64_t nextrip; /* (x) next instruction to execute */
115 #define vcpu_lock_initialized(v) mtx_initialized(&((v)->mtx))
116 #define vcpu_lock_init(v) mtx_init(&((v)->mtx), "vcpu lock", 0, MTX_SPIN)
117 #define vcpu_lock(v) mtx_lock_spin(&((v)->mtx))
118 #define vcpu_unlock(v) mtx_unlock_spin(&((v)->mtx))
119 #define vcpu_assert_locked(v) mtx_assert(&((v)->mtx), MA_OWNED)
124 struct vm_object *object;
126 #define VM_MAX_MEMSEGS 2
136 #define VM_MAX_MEMMAPS 4
140 * (o) initialized the first time the VM is created
141 * (i) initialized when VM is created and when it is reinitialized
142 * (x) initialized before use
145 void *cookie; /* (i) cpu-specific data */
146 void *iommu; /* (x) iommu-specific data */
147 struct vhpet *vhpet; /* (i) virtual HPET */
148 struct vioapic *vioapic; /* (i) virtual ioapic */
149 struct vatpic *vatpic; /* (i) virtual atpic */
150 struct vatpit *vatpit; /* (i) virtual atpit */
151 struct vpmtmr *vpmtmr; /* (i) virtual ACPI PM timer */
152 struct vrtc *vrtc; /* (o) virtual RTC */
153 volatile cpuset_t active_cpus; /* (i) active vcpus */
154 int suspend; /* (i) stop VM execution */
155 volatile cpuset_t suspended_cpus; /* (i) suspended vcpus */
156 volatile cpuset_t halted_cpus; /* (x) cpus in a hard halt */
157 cpuset_t rendezvous_req_cpus; /* (x) rendezvous requested */
158 cpuset_t rendezvous_done_cpus; /* (x) rendezvous finished */
159 void *rendezvous_arg; /* (x) rendezvous func/arg */
160 vm_rendezvous_func_t rendezvous_func;
161 struct mtx rendezvous_mtx; /* (o) rendezvous lock */
162 struct mem_map mem_maps[VM_MAX_MEMMAPS]; /* (i) guest address space */
163 struct mem_seg mem_segs[VM_MAX_MEMSEGS]; /* (o) guest memory regions */
164 struct vmspace *vmspace; /* (o) guest's address space */
165 char name[VM_MAX_NAMELEN]; /* (o) virtual machine name */
166 struct vcpu vcpu[VM_MAXCPU]; /* (i) guest vcpus */
169 static int vmm_initialized;
171 static struct vmm_ops *ops;
172 #define VMM_INIT(num) (ops != NULL ? (*ops->init)(num) : 0)
173 #define VMM_CLEANUP() (ops != NULL ? (*ops->cleanup)() : 0)
174 #define VMM_RESUME() (ops != NULL ? (*ops->resume)() : 0)
176 #define VMINIT(vm, pmap) (ops != NULL ? (*ops->vminit)(vm, pmap): NULL)
177 #define VMRUN(vmi, vcpu, rip, pmap, evinfo) \
178 (ops != NULL ? (*ops->vmrun)(vmi, vcpu, rip, pmap, evinfo) : ENXIO)
179 #define VMCLEANUP(vmi) (ops != NULL ? (*ops->vmcleanup)(vmi) : NULL)
180 #define VMSPACE_ALLOC(min, max) \
181 (ops != NULL ? (*ops->vmspace_alloc)(min, max) : NULL)
182 #define VMSPACE_FREE(vmspace) \
183 (ops != NULL ? (*ops->vmspace_free)(vmspace) : ENXIO)
184 #define VMGETREG(vmi, vcpu, num, retval) \
185 (ops != NULL ? (*ops->vmgetreg)(vmi, vcpu, num, retval) : ENXIO)
186 #define VMSETREG(vmi, vcpu, num, val) \
187 (ops != NULL ? (*ops->vmsetreg)(vmi, vcpu, num, val) : ENXIO)
188 #define VMGETDESC(vmi, vcpu, num, desc) \
189 (ops != NULL ? (*ops->vmgetdesc)(vmi, vcpu, num, desc) : ENXIO)
190 #define VMSETDESC(vmi, vcpu, num, desc) \
191 (ops != NULL ? (*ops->vmsetdesc)(vmi, vcpu, num, desc) : ENXIO)
192 #define VMGETCAP(vmi, vcpu, num, retval) \
193 (ops != NULL ? (*ops->vmgetcap)(vmi, vcpu, num, retval) : ENXIO)
194 #define VMSETCAP(vmi, vcpu, num, val) \
195 (ops != NULL ? (*ops->vmsetcap)(vmi, vcpu, num, val) : ENXIO)
196 #define VLAPIC_INIT(vmi, vcpu) \
197 (ops != NULL ? (*ops->vlapic_init)(vmi, vcpu) : NULL)
198 #define VLAPIC_CLEANUP(vmi, vlapic) \
199 (ops != NULL ? (*ops->vlapic_cleanup)(vmi, vlapic) : NULL)
201 #define fpu_start_emulating() load_cr0(rcr0() | CR0_TS)
202 #define fpu_stop_emulating() clts()
204 static MALLOC_DEFINE(M_VM, "vm", "vm");
207 static VMM_STAT(VCPU_TOTAL_RUNTIME, "vcpu total runtime");
209 SYSCTL_NODE(_hw, OID_AUTO, vmm, CTLFLAG_RW, NULL, NULL);
212 * Halt the guest if all vcpus are executing a HLT instruction with
213 * interrupts disabled.
215 static int halt_detection_enabled = 1;
216 SYSCTL_INT(_hw_vmm, OID_AUTO, halt_detection, CTLFLAG_RDTUN,
217 &halt_detection_enabled, 0,
218 "Halt VM if all vcpus execute HLT with interrupts disabled");
220 static int vmm_ipinum;
221 SYSCTL_INT(_hw_vmm, OID_AUTO, ipinum, CTLFLAG_RD, &vmm_ipinum, 0,
222 "IPI vector used for vcpu notifications");
224 static int trace_guest_exceptions;
225 SYSCTL_INT(_hw_vmm, OID_AUTO, trace_guest_exceptions, CTLFLAG_RDTUN,
226 &trace_guest_exceptions, 0,
227 "Trap into hypervisor on all guest exceptions and reflect them back");
229 static int vmm_force_iommu = 0;
230 TUNABLE_INT("hw.vmm.force_iommu", &vmm_force_iommu);
231 SYSCTL_INT(_hw_vmm, OID_AUTO, force_iommu, CTLFLAG_RDTUN, &vmm_force_iommu, 0,
232 "Force use of I/O MMU even if no passthrough devices were found.");
234 static void vm_free_memmap(struct vm *vm, int ident);
235 static bool sysmem_mapping(struct vm *vm, struct mem_map *mm);
236 static void vcpu_notify_event_locked(struct vcpu *vcpu, bool lapic_intr);
240 vcpu_state2str(enum vcpu_state state)
259 vcpu_cleanup(struct vm *vm, int i, bool destroy)
261 struct vcpu *vcpu = &vm->vcpu[i];
263 VLAPIC_CLEANUP(vm->cookie, vcpu->vlapic);
265 vmm_stat_free(vcpu->stats);
266 fpu_save_area_free(vcpu->guestfpu);
271 vcpu_init(struct vm *vm, int vcpu_id, bool create)
275 KASSERT(vcpu_id >= 0 && vcpu_id < VM_MAXCPU,
276 ("vcpu_init: invalid vcpu %d", vcpu_id));
278 vcpu = &vm->vcpu[vcpu_id];
281 KASSERT(!vcpu_lock_initialized(vcpu), ("vcpu %d already "
282 "initialized", vcpu_id));
283 vcpu_lock_init(vcpu);
284 vcpu->state = VCPU_IDLE;
285 vcpu->hostcpu = NOCPU;
286 vcpu->guestfpu = fpu_save_area_alloc();
287 vcpu->stats = vmm_stat_alloc();
290 vcpu->vlapic = VLAPIC_INIT(vm->cookie, vcpu_id);
291 vm_set_x2apic_state(vm, vcpu_id, X2APIC_DISABLED);
293 vcpu->exitintinfo = 0;
294 vcpu->nmi_pending = 0;
295 vcpu->extint_pending = 0;
296 vcpu->exception_pending = 0;
297 vcpu->guest_xcr0 = XFEATURE_ENABLED_X87;
298 fpu_save_area_reset(vcpu->guestfpu);
299 vmm_stat_init(vcpu->stats);
303 vcpu_trace_exceptions(struct vm *vm, int vcpuid)
306 return (trace_guest_exceptions);
310 vm_exitinfo(struct vm *vm, int cpuid)
314 if (cpuid < 0 || cpuid >= VM_MAXCPU)
315 panic("vm_exitinfo: invalid cpuid %d", cpuid);
317 vcpu = &vm->vcpu[cpuid];
319 return (&vcpu->exitinfo);
333 vmm_host_state_init();
335 vmm_ipinum = lapic_ipi_alloc(&IDTVEC(justreturn));
337 vmm_ipinum = IPI_AST;
339 error = vmm_mem_init();
344 ops = &vmm_ops_intel;
345 else if (vmm_is_amd())
350 vmm_resume_p = vmm_resume;
352 return (VMM_INIT(vmm_ipinum));
356 vmm_handler(module_t mod, int what, void *arg)
363 if (vmm_force_iommu || ppt_avail_devices() > 0)
370 error = vmmdev_cleanup();
374 if (vmm_ipinum != IPI_AST)
375 lapic_ipi_free(vmm_ipinum);
376 error = VMM_CLEANUP();
378 * Something bad happened - prevent new
379 * VMs from being created
392 static moduledata_t vmm_kmod = {
399 * vmm initialization has the following dependencies:
401 * - iommu initialization must happen after the pci passthru driver has had
402 * a chance to attach to any passthru devices (after SI_SUB_CONFIGURE).
404 * - VT-x initialization requires smp_rendezvous() and therefore must happen
405 * after SMP is fully functional (after SI_SUB_SMP).
407 DECLARE_MODULE(vmm, vmm_kmod, SI_SUB_SMP + 1, SI_ORDER_ANY);
408 MODULE_VERSION(vmm, 1);
411 vm_init(struct vm *vm, bool create)
415 vm->cookie = VMINIT(vm, vmspace_pmap(vm->vmspace));
417 vm->vioapic = vioapic_init(vm);
418 vm->vhpet = vhpet_init(vm);
419 vm->vatpic = vatpic_init(vm);
420 vm->vatpit = vatpit_init(vm);
421 vm->vpmtmr = vpmtmr_init(vm);
423 vm->vrtc = vrtc_init(vm);
425 CPU_ZERO(&vm->active_cpus);
428 CPU_ZERO(&vm->suspended_cpus);
430 for (i = 0; i < VM_MAXCPU; i++)
431 vcpu_init(vm, i, create);
435 vm_create(const char *name, struct vm **retvm)
438 struct vmspace *vmspace;
441 * If vmm.ko could not be successfully initialized then don't attempt
442 * to create the virtual machine.
444 if (!vmm_initialized)
447 if (name == NULL || strlen(name) >= VM_MAX_NAMELEN)
450 vmspace = VMSPACE_ALLOC(0, VM_MAXUSER_ADDRESS);
454 vm = malloc(sizeof(struct vm), M_VM, M_WAITOK | M_ZERO);
455 strcpy(vm->name, name);
456 vm->vmspace = vmspace;
457 mtx_init(&vm->rendezvous_mtx, "vm rendezvous lock", 0, MTX_DEF);
466 vm_cleanup(struct vm *vm, bool destroy)
471 ppt_unassign_all(vm);
473 if (vm->iommu != NULL)
474 iommu_destroy_domain(vm->iommu);
477 vrtc_cleanup(vm->vrtc);
479 vrtc_reset(vm->vrtc);
480 vpmtmr_cleanup(vm->vpmtmr);
481 vatpit_cleanup(vm->vatpit);
482 vhpet_cleanup(vm->vhpet);
483 vatpic_cleanup(vm->vatpic);
484 vioapic_cleanup(vm->vioapic);
486 for (i = 0; i < VM_MAXCPU; i++)
487 vcpu_cleanup(vm, i, destroy);
489 VMCLEANUP(vm->cookie);
492 * System memory is removed from the guest address space only when
493 * the VM is destroyed. This is because the mapping remains the same
496 * Device memory can be relocated by the guest (e.g. using PCI BARs)
497 * so those mappings are removed on a VM reset.
499 for (i = 0; i < VM_MAX_MEMMAPS; i++) {
500 mm = &vm->mem_maps[i];
501 if (destroy || !sysmem_mapping(vm, mm))
502 vm_free_memmap(vm, i);
506 for (i = 0; i < VM_MAX_MEMSEGS; i++)
507 vm_free_memseg(vm, i);
509 VMSPACE_FREE(vm->vmspace);
515 vm_destroy(struct vm *vm)
517 vm_cleanup(vm, true);
522 vm_reinit(struct vm *vm)
527 * A virtual machine can be reset only if all vcpus are suspended.
529 if (CPU_CMP(&vm->suspended_cpus, &vm->active_cpus) == 0) {
530 vm_cleanup(vm, false);
541 vm_name(struct vm *vm)
547 vm_map_mmio(struct vm *vm, vm_paddr_t gpa, size_t len, vm_paddr_t hpa)
551 if ((obj = vmm_mmio_alloc(vm->vmspace, gpa, len, hpa)) == NULL)
558 vm_unmap_mmio(struct vm *vm, vm_paddr_t gpa, size_t len)
561 vmm_mmio_free(vm->vmspace, gpa, len);
566 * Return 'true' if 'gpa' is allocated in the guest address space.
568 * This function is called in the context of a running vcpu which acts as
569 * an implicit lock on 'vm->mem_maps[]'.
572 vm_mem_allocated(struct vm *vm, int vcpuid, vm_paddr_t gpa)
579 state = vcpu_get_state(vm, vcpuid, &hostcpu);
580 KASSERT(state == VCPU_RUNNING && hostcpu == curcpu,
581 ("%s: invalid vcpu state %d/%d", __func__, state, hostcpu));
584 for (i = 0; i < VM_MAX_MEMMAPS; i++) {
585 mm = &vm->mem_maps[i];
586 if (mm->len != 0 && gpa >= mm->gpa && gpa < mm->gpa + mm->len)
587 return (true); /* 'gpa' is sysmem or devmem */
590 if (ppt_is_mmio(vm, gpa))
591 return (true); /* 'gpa' is pci passthru mmio */
597 vm_alloc_memseg(struct vm *vm, int ident, size_t len, bool sysmem)
602 if (ident < 0 || ident >= VM_MAX_MEMSEGS)
605 if (len == 0 || (len & PAGE_MASK))
608 seg = &vm->mem_segs[ident];
609 if (seg->object != NULL) {
610 if (seg->len == len && seg->sysmem == sysmem)
616 obj = vm_object_allocate(OBJT_DEFAULT, len >> PAGE_SHIFT);
622 seg->sysmem = sysmem;
627 vm_get_memseg(struct vm *vm, int ident, size_t *len, bool *sysmem,
632 if (ident < 0 || ident >= VM_MAX_MEMSEGS)
635 seg = &vm->mem_segs[ident];
639 *sysmem = seg->sysmem;
641 *objptr = seg->object;
646 vm_free_memseg(struct vm *vm, int ident)
650 KASSERT(ident >= 0 && ident < VM_MAX_MEMSEGS,
651 ("%s: invalid memseg ident %d", __func__, ident));
653 seg = &vm->mem_segs[ident];
654 if (seg->object != NULL) {
655 vm_object_deallocate(seg->object);
656 bzero(seg, sizeof(struct mem_seg));
661 vm_mmap_memseg(struct vm *vm, vm_paddr_t gpa, int segid, vm_ooffset_t first,
662 size_t len, int prot, int flags)
665 struct mem_map *m, *map;
669 if (prot == 0 || (prot & ~(VM_PROT_ALL)) != 0)
672 if (flags & ~VM_MEMMAP_F_WIRED)
675 if (segid < 0 || segid >= VM_MAX_MEMSEGS)
678 seg = &vm->mem_segs[segid];
679 if (seg->object == NULL)
683 if (first < 0 || first >= last || last > seg->len)
686 if ((gpa | first | last) & PAGE_MASK)
690 for (i = 0; i < VM_MAX_MEMMAPS; i++) {
691 m = &vm->mem_maps[i];
701 error = vm_map_find(&vm->vmspace->vm_map, seg->object, first, &gpa,
702 len, 0, VMFS_NO_SPACE, prot, prot, 0);
703 if (error != KERN_SUCCESS)
706 vm_object_reference(seg->object);
708 if (flags & VM_MEMMAP_F_WIRED) {
709 error = vm_map_wire(&vm->vmspace->vm_map, gpa, gpa + len,
710 VM_MAP_WIRE_USER | VM_MAP_WIRE_NOHOLES);
711 if (error != KERN_SUCCESS) {
712 vm_map_remove(&vm->vmspace->vm_map, gpa, gpa + len);
727 vm_mmap_getnext(struct vm *vm, vm_paddr_t *gpa, int *segid,
728 vm_ooffset_t *segoff, size_t *len, int *prot, int *flags)
730 struct mem_map *mm, *mmnext;
734 for (i = 0; i < VM_MAX_MEMMAPS; i++) {
735 mm = &vm->mem_maps[i];
736 if (mm->len == 0 || mm->gpa < *gpa)
738 if (mmnext == NULL || mm->gpa < mmnext->gpa)
742 if (mmnext != NULL) {
745 *segid = mmnext->segid;
747 *segoff = mmnext->segoff;
751 *prot = mmnext->prot;
753 *flags = mmnext->flags;
761 vm_free_memmap(struct vm *vm, int ident)
766 mm = &vm->mem_maps[ident];
768 error = vm_map_remove(&vm->vmspace->vm_map, mm->gpa,
770 KASSERT(error == KERN_SUCCESS, ("%s: vm_map_remove error %d",
772 bzero(mm, sizeof(struct mem_map));
777 sysmem_mapping(struct vm *vm, struct mem_map *mm)
780 if (mm->len != 0 && vm->mem_segs[mm->segid].sysmem)
787 sysmem_maxaddr(struct vm *vm)
794 for (i = 0; i < VM_MAX_MEMMAPS; i++) {
795 mm = &vm->mem_maps[i];
796 if (sysmem_mapping(vm, mm)) {
797 if (maxaddr < mm->gpa + mm->len)
798 maxaddr = mm->gpa + mm->len;
805 vm_iommu_modify(struct vm *vm, boolean_t map)
810 void *vp, *cookie, *host_domain;
813 host_domain = iommu_host_domain();
815 for (i = 0; i < VM_MAX_MEMMAPS; i++) {
816 mm = &vm->mem_maps[i];
817 if (!sysmem_mapping(vm, mm))
821 KASSERT((mm->flags & VM_MEMMAP_F_IOMMU) == 0,
822 ("iommu map found invalid memmap %#lx/%#lx/%#x",
823 mm->gpa, mm->len, mm->flags));
824 if ((mm->flags & VM_MEMMAP_F_WIRED) == 0)
826 mm->flags |= VM_MEMMAP_F_IOMMU;
828 if ((mm->flags & VM_MEMMAP_F_IOMMU) == 0)
830 mm->flags &= ~VM_MEMMAP_F_IOMMU;
831 KASSERT((mm->flags & VM_MEMMAP_F_WIRED) != 0,
832 ("iommu unmap found invalid memmap %#lx/%#lx/%#x",
833 mm->gpa, mm->len, mm->flags));
837 while (gpa < mm->gpa + mm->len) {
838 vp = vm_gpa_hold(vm, -1, gpa, PAGE_SIZE, VM_PROT_WRITE,
840 KASSERT(vp != NULL, ("vm(%s) could not map gpa %#lx",
843 vm_gpa_release(cookie);
845 hpa = DMAP_TO_PHYS((uintptr_t)vp);
847 iommu_create_mapping(vm->iommu, gpa, hpa, sz);
848 iommu_remove_mapping(host_domain, hpa, sz);
850 iommu_remove_mapping(vm->iommu, gpa, sz);
851 iommu_create_mapping(host_domain, hpa, hpa, sz);
859 * Invalidate the cached translations associated with the domain
860 * from which pages were removed.
863 iommu_invalidate_tlb(host_domain);
865 iommu_invalidate_tlb(vm->iommu);
868 #define vm_iommu_unmap(vm) vm_iommu_modify((vm), FALSE)
869 #define vm_iommu_map(vm) vm_iommu_modify((vm), TRUE)
872 vm_unassign_pptdev(struct vm *vm, int bus, int slot, int func)
876 error = ppt_unassign_device(vm, bus, slot, func);
880 if (ppt_assigned_devices(vm) == 0)
887 vm_assign_pptdev(struct vm *vm, int bus, int slot, int func)
892 /* Set up the IOMMU to do the 'gpa' to 'hpa' translation */
893 if (ppt_assigned_devices(vm) == 0) {
894 KASSERT(vm->iommu == NULL,
895 ("vm_assign_pptdev: iommu must be NULL"));
896 maxaddr = sysmem_maxaddr(vm);
897 vm->iommu = iommu_create_domain(maxaddr);
901 error = ppt_assign_device(vm, bus, slot, func);
906 vm_gpa_hold(struct vm *vm, int vcpuid, vm_paddr_t gpa, size_t len, int reqprot,
909 int i, count, pageoff;
914 * All vcpus are frozen by ioctls that modify the memory map
915 * (e.g. VM_MMAP_MEMSEG). Therefore 'vm->memmap[]' stability is
916 * guaranteed if at least one vcpu is in the VCPU_FROZEN state.
919 KASSERT(vcpuid >= -1 || vcpuid < VM_MAXCPU, ("%s: invalid vcpuid %d",
921 for (i = 0; i < VM_MAXCPU; i++) {
922 if (vcpuid != -1 && vcpuid != i)
924 state = vcpu_get_state(vm, i, NULL);
925 KASSERT(state == VCPU_FROZEN, ("%s: invalid vcpu state %d",
929 pageoff = gpa & PAGE_MASK;
930 if (len > PAGE_SIZE - pageoff)
931 panic("vm_gpa_hold: invalid gpa/len: 0x%016lx/%lu", gpa, len);
934 for (i = 0; i < VM_MAX_MEMMAPS; i++) {
935 mm = &vm->mem_maps[i];
936 if (sysmem_mapping(vm, mm) && gpa >= mm->gpa &&
937 gpa < mm->gpa + mm->len) {
938 count = vm_fault_quick_hold_pages(&vm->vmspace->vm_map,
939 trunc_page(gpa), PAGE_SIZE, reqprot, &m, 1);
946 return ((void *)(PHYS_TO_DMAP(VM_PAGE_TO_PHYS(m)) + pageoff));
954 vm_gpa_release(void *cookie)
956 vm_page_t m = cookie;
964 vm_get_register(struct vm *vm, int vcpu, int reg, uint64_t *retval)
967 if (vcpu < 0 || vcpu >= VM_MAXCPU)
970 if (reg >= VM_REG_LAST)
973 return (VMGETREG(vm->cookie, vcpu, reg, retval));
977 vm_set_register(struct vm *vm, int vcpuid, int reg, uint64_t val)
982 if (vcpuid < 0 || vcpuid >= VM_MAXCPU)
985 if (reg >= VM_REG_LAST)
988 error = VMSETREG(vm->cookie, vcpuid, reg, val);
989 if (error || reg != VM_REG_GUEST_RIP)
992 /* Set 'nextrip' to match the value of %rip */
993 VCPU_CTR1(vm, vcpuid, "Setting nextrip to %#lx", val);
994 vcpu = &vm->vcpu[vcpuid];
1000 is_descriptor_table(int reg)
1004 case VM_REG_GUEST_IDTR:
1005 case VM_REG_GUEST_GDTR:
1013 is_segment_register(int reg)
1017 case VM_REG_GUEST_ES:
1018 case VM_REG_GUEST_CS:
1019 case VM_REG_GUEST_SS:
1020 case VM_REG_GUEST_DS:
1021 case VM_REG_GUEST_FS:
1022 case VM_REG_GUEST_GS:
1023 case VM_REG_GUEST_TR:
1024 case VM_REG_GUEST_LDTR:
1032 vm_get_seg_desc(struct vm *vm, int vcpu, int reg,
1033 struct seg_desc *desc)
1036 if (vcpu < 0 || vcpu >= VM_MAXCPU)
1039 if (!is_segment_register(reg) && !is_descriptor_table(reg))
1042 return (VMGETDESC(vm->cookie, vcpu, reg, desc));
1046 vm_set_seg_desc(struct vm *vm, int vcpu, int reg,
1047 struct seg_desc *desc)
1049 if (vcpu < 0 || vcpu >= VM_MAXCPU)
1052 if (!is_segment_register(reg) && !is_descriptor_table(reg))
1055 return (VMSETDESC(vm->cookie, vcpu, reg, desc));
1059 restore_guest_fpustate(struct vcpu *vcpu)
1062 /* flush host state to the pcb */
1065 /* restore guest FPU state */
1066 fpu_stop_emulating();
1067 fpurestore(vcpu->guestfpu);
1069 /* restore guest XCR0 if XSAVE is enabled in the host */
1070 if (rcr4() & CR4_XSAVE)
1071 load_xcr(0, vcpu->guest_xcr0);
1074 * The FPU is now "dirty" with the guest's state so turn on emulation
1075 * to trap any access to the FPU by the host.
1077 fpu_start_emulating();
1081 save_guest_fpustate(struct vcpu *vcpu)
1084 if ((rcr0() & CR0_TS) == 0)
1085 panic("fpu emulation not enabled in host!");
1087 /* save guest XCR0 and restore host XCR0 */
1088 if (rcr4() & CR4_XSAVE) {
1089 vcpu->guest_xcr0 = rxcr(0);
1090 load_xcr(0, vmm_get_host_xcr0());
1093 /* save guest FPU state */
1094 fpu_stop_emulating();
1095 fpusave(vcpu->guestfpu);
1096 fpu_start_emulating();
1099 static VMM_STAT(VCPU_IDLE_TICKS, "number of ticks vcpu was idle");
1102 vcpu_set_state_locked(struct vm *vm, int vcpuid, enum vcpu_state newstate,
1108 vcpu = &vm->vcpu[vcpuid];
1109 vcpu_assert_locked(vcpu);
1112 * State transitions from the vmmdev_ioctl() must always begin from
1113 * the VCPU_IDLE state. This guarantees that there is only a single
1114 * ioctl() operating on a vcpu at any point.
1117 while (vcpu->state != VCPU_IDLE) {
1119 vcpu_notify_event_locked(vcpu, false);
1120 VCPU_CTR1(vm, vcpuid, "vcpu state change from %s to "
1121 "idle requested", vcpu_state2str(vcpu->state));
1122 msleep_spin(&vcpu->state, &vcpu->mtx, "vmstat", hz);
1125 KASSERT(vcpu->state != VCPU_IDLE, ("invalid transition from "
1126 "vcpu idle state"));
1129 if (vcpu->state == VCPU_RUNNING) {
1130 KASSERT(vcpu->hostcpu == curcpu, ("curcpu %d and hostcpu %d "
1131 "mismatch for running vcpu", curcpu, vcpu->hostcpu));
1133 KASSERT(vcpu->hostcpu == NOCPU, ("Invalid hostcpu %d for a "
1134 "vcpu that is not running", vcpu->hostcpu));
1138 * The following state transitions are allowed:
1139 * IDLE -> FROZEN -> IDLE
1140 * FROZEN -> RUNNING -> FROZEN
1141 * FROZEN -> SLEEPING -> FROZEN
1143 switch (vcpu->state) {
1147 error = (newstate != VCPU_FROZEN);
1150 error = (newstate == VCPU_FROZEN);
1160 VCPU_CTR2(vm, vcpuid, "vcpu state changed from %s to %s",
1161 vcpu_state2str(vcpu->state), vcpu_state2str(newstate));
1163 vcpu->state = newstate;
1164 if (newstate == VCPU_RUNNING)
1165 vcpu->hostcpu = curcpu;
1167 vcpu->hostcpu = NOCPU;
1169 if (newstate == VCPU_IDLE)
1170 wakeup(&vcpu->state);
1176 vcpu_require_state(struct vm *vm, int vcpuid, enum vcpu_state newstate)
1180 if ((error = vcpu_set_state(vm, vcpuid, newstate, false)) != 0)
1181 panic("Error %d setting state to %d\n", error, newstate);
1185 vcpu_require_state_locked(struct vm *vm, int vcpuid, enum vcpu_state newstate)
1189 if ((error = vcpu_set_state_locked(vm, vcpuid, newstate, false)) != 0)
1190 panic("Error %d setting state to %d", error, newstate);
1194 vm_set_rendezvous_func(struct vm *vm, vm_rendezvous_func_t func)
1197 KASSERT(mtx_owned(&vm->rendezvous_mtx), ("rendezvous_mtx not locked"));
1200 * Update 'rendezvous_func' and execute a write memory barrier to
1201 * ensure that it is visible across all host cpus. This is not needed
1202 * for correctness but it does ensure that all the vcpus will notice
1203 * that the rendezvous is requested immediately.
1205 vm->rendezvous_func = func;
1209 #define RENDEZVOUS_CTR0(vm, vcpuid, fmt) \
1212 VCPU_CTR0(vm, vcpuid, fmt); \
1218 vm_handle_rendezvous(struct vm *vm, int vcpuid)
1221 KASSERT(vcpuid == -1 || (vcpuid >= 0 && vcpuid < VM_MAXCPU),
1222 ("vm_handle_rendezvous: invalid vcpuid %d", vcpuid));
1224 mtx_lock(&vm->rendezvous_mtx);
1225 while (vm->rendezvous_func != NULL) {
1226 /* 'rendezvous_req_cpus' must be a subset of 'active_cpus' */
1227 CPU_AND(&vm->rendezvous_req_cpus, &vm->active_cpus);
1230 CPU_ISSET(vcpuid, &vm->rendezvous_req_cpus) &&
1231 !CPU_ISSET(vcpuid, &vm->rendezvous_done_cpus)) {
1232 VCPU_CTR0(vm, vcpuid, "Calling rendezvous func");
1233 (*vm->rendezvous_func)(vm, vcpuid, vm->rendezvous_arg);
1234 CPU_SET(vcpuid, &vm->rendezvous_done_cpus);
1236 if (CPU_CMP(&vm->rendezvous_req_cpus,
1237 &vm->rendezvous_done_cpus) == 0) {
1238 VCPU_CTR0(vm, vcpuid, "Rendezvous completed");
1239 vm_set_rendezvous_func(vm, NULL);
1240 wakeup(&vm->rendezvous_func);
1243 RENDEZVOUS_CTR0(vm, vcpuid, "Wait for rendezvous completion");
1244 mtx_sleep(&vm->rendezvous_func, &vm->rendezvous_mtx, 0,
1247 mtx_unlock(&vm->rendezvous_mtx);
1251 * Emulate a guest 'hlt' by sleeping until the vcpu is ready to run.
1254 vm_handle_hlt(struct vm *vm, int vcpuid, bool intr_disabled, bool *retu)
1258 int t, vcpu_halted, vm_halted;
1260 KASSERT(!CPU_ISSET(vcpuid, &vm->halted_cpus), ("vcpu already halted"));
1262 vcpu = &vm->vcpu[vcpuid];
1269 * Do a final check for pending NMI or interrupts before
1270 * really putting this thread to sleep. Also check for
1271 * software events that would cause this vcpu to wakeup.
1273 * These interrupts/events could have happened after the
1274 * vcpu returned from VMRUN() and before it acquired the
1277 if (vm->rendezvous_func != NULL || vm->suspend || vcpu->reqidle)
1279 if (vm_nmi_pending(vm, vcpuid))
1281 if (!intr_disabled) {
1282 if (vm_extint_pending(vm, vcpuid) ||
1283 vlapic_pending_intr(vcpu->vlapic, NULL)) {
1288 /* Don't go to sleep if the vcpu thread needs to yield */
1289 if (vcpu_should_yield(vm, vcpuid))
1293 * Some Linux guests implement "halt" by having all vcpus
1294 * execute HLT with interrupts disabled. 'halted_cpus' keeps
1295 * track of the vcpus that have entered this state. When all
1296 * vcpus enter the halted state the virtual machine is halted.
1298 if (intr_disabled) {
1300 VCPU_CTR0(vm, vcpuid, "Halted");
1301 if (!vcpu_halted && halt_detection_enabled) {
1303 CPU_SET_ATOMIC(vcpuid, &vm->halted_cpus);
1305 if (CPU_CMP(&vm->halted_cpus, &vm->active_cpus) == 0) {
1314 vcpu_require_state_locked(vm, vcpuid, VCPU_SLEEPING);
1316 * XXX msleep_spin() cannot be interrupted by signals so
1317 * wake up periodically to check pending signals.
1319 msleep_spin(vcpu, &vcpu->mtx, wmesg, hz);
1320 vcpu_require_state_locked(vm, vcpuid, VCPU_FROZEN);
1321 vmm_stat_incr(vm, vcpuid, VCPU_IDLE_TICKS, ticks - t);
1325 CPU_CLR_ATOMIC(vcpuid, &vm->halted_cpus);
1330 vm_suspend(vm, VM_SUSPEND_HALT);
1336 vm_handle_paging(struct vm *vm, int vcpuid, bool *retu)
1341 struct vm_exit *vme;
1343 vcpu = &vm->vcpu[vcpuid];
1344 vme = &vcpu->exitinfo;
1346 KASSERT(vme->inst_length == 0, ("%s: invalid inst_length %d",
1347 __func__, vme->inst_length));
1349 ftype = vme->u.paging.fault_type;
1350 KASSERT(ftype == VM_PROT_READ ||
1351 ftype == VM_PROT_WRITE || ftype == VM_PROT_EXECUTE,
1352 ("vm_handle_paging: invalid fault_type %d", ftype));
1354 if (ftype == VM_PROT_READ || ftype == VM_PROT_WRITE) {
1355 rv = pmap_emulate_accessed_dirty(vmspace_pmap(vm->vmspace),
1356 vme->u.paging.gpa, ftype);
1358 VCPU_CTR2(vm, vcpuid, "%s bit emulation for gpa %#lx",
1359 ftype == VM_PROT_READ ? "accessed" : "dirty",
1365 map = &vm->vmspace->vm_map;
1366 rv = vm_fault(map, vme->u.paging.gpa, ftype, VM_FAULT_NORMAL);
1368 VCPU_CTR3(vm, vcpuid, "vm_handle_paging rv = %d, gpa = %#lx, "
1369 "ftype = %d", rv, vme->u.paging.gpa, ftype);
1371 if (rv != KERN_SUCCESS)
1378 vm_handle_inst_emul(struct vm *vm, int vcpuid, bool *retu)
1382 struct vm_exit *vme;
1383 uint64_t gla, gpa, cs_base;
1384 struct vm_guest_paging *paging;
1385 mem_region_read_t mread;
1386 mem_region_write_t mwrite;
1387 enum vm_cpu_mode cpu_mode;
1388 int cs_d, error, fault;
1390 vcpu = &vm->vcpu[vcpuid];
1391 vme = &vcpu->exitinfo;
1393 KASSERT(vme->inst_length == 0, ("%s: invalid inst_length %d",
1394 __func__, vme->inst_length));
1396 gla = vme->u.inst_emul.gla;
1397 gpa = vme->u.inst_emul.gpa;
1398 cs_base = vme->u.inst_emul.cs_base;
1399 cs_d = vme->u.inst_emul.cs_d;
1400 vie = &vme->u.inst_emul.vie;
1401 paging = &vme->u.inst_emul.paging;
1402 cpu_mode = paging->cpu_mode;
1404 VCPU_CTR1(vm, vcpuid, "inst_emul fault accessing gpa %#lx", gpa);
1406 /* Fetch, decode and emulate the faulting instruction */
1407 if (vie->num_valid == 0) {
1408 error = vmm_fetch_instruction(vm, vcpuid, paging, vme->rip +
1409 cs_base, VIE_INST_SIZE, vie, &fault);
1412 * The instruction bytes have already been copied into 'vie'
1419 if (vmm_decode_instruction(vm, vcpuid, gla, cpu_mode, cs_d, vie) != 0) {
1420 VCPU_CTR1(vm, vcpuid, "Error decoding instruction at %#lx",
1421 vme->rip + cs_base);
1422 *retu = true; /* dump instruction bytes in userspace */
1427 * Update 'nextrip' based on the length of the emulated instruction.
1429 vme->inst_length = vie->num_processed;
1430 vcpu->nextrip += vie->num_processed;
1431 VCPU_CTR1(vm, vcpuid, "nextrip updated to %#lx after instruction "
1432 "decoding", vcpu->nextrip);
1434 /* return to userland unless this is an in-kernel emulated device */
1435 if (gpa >= DEFAULT_APIC_BASE && gpa < DEFAULT_APIC_BASE + PAGE_SIZE) {
1436 mread = lapic_mmio_read;
1437 mwrite = lapic_mmio_write;
1438 } else if (gpa >= VIOAPIC_BASE && gpa < VIOAPIC_BASE + VIOAPIC_SIZE) {
1439 mread = vioapic_mmio_read;
1440 mwrite = vioapic_mmio_write;
1441 } else if (gpa >= VHPET_BASE && gpa < VHPET_BASE + VHPET_SIZE) {
1442 mread = vhpet_mmio_read;
1443 mwrite = vhpet_mmio_write;
1449 error = vmm_emulate_instruction(vm, vcpuid, gpa, vie, paging,
1450 mread, mwrite, retu);
1456 vm_handle_suspend(struct vm *vm, int vcpuid, bool *retu)
1462 vcpu = &vm->vcpu[vcpuid];
1464 CPU_SET_ATOMIC(vcpuid, &vm->suspended_cpus);
1467 * Wait until all 'active_cpus' have suspended themselves.
1469 * Since a VM may be suspended at any time including when one or
1470 * more vcpus are doing a rendezvous we need to call the rendezvous
1471 * handler while we are waiting to prevent a deadlock.
1475 if (CPU_CMP(&vm->suspended_cpus, &vm->active_cpus) == 0) {
1476 VCPU_CTR0(vm, vcpuid, "All vcpus suspended");
1480 if (vm->rendezvous_func == NULL) {
1481 VCPU_CTR0(vm, vcpuid, "Sleeping during suspend");
1482 vcpu_require_state_locked(vm, vcpuid, VCPU_SLEEPING);
1483 msleep_spin(vcpu, &vcpu->mtx, "vmsusp", hz);
1484 vcpu_require_state_locked(vm, vcpuid, VCPU_FROZEN);
1486 VCPU_CTR0(vm, vcpuid, "Rendezvous during suspend");
1488 vm_handle_rendezvous(vm, vcpuid);
1495 * Wakeup the other sleeping vcpus and return to userspace.
1497 for (i = 0; i < VM_MAXCPU; i++) {
1498 if (CPU_ISSET(i, &vm->suspended_cpus)) {
1499 vcpu_notify_event(vm, i, false);
1508 vm_handle_reqidle(struct vm *vm, int vcpuid, bool *retu)
1510 struct vcpu *vcpu = &vm->vcpu[vcpuid];
1513 KASSERT(vcpu->reqidle, ("invalid vcpu reqidle %d", vcpu->reqidle));
1521 vm_suspend(struct vm *vm, enum vm_suspend_how how)
1525 if (how <= VM_SUSPEND_NONE || how >= VM_SUSPEND_LAST)
1528 if (atomic_cmpset_int(&vm->suspend, 0, how) == 0) {
1529 VM_CTR2(vm, "virtual machine already suspended %d/%d",
1534 VM_CTR1(vm, "virtual machine successfully suspended %d", how);
1537 * Notify all active vcpus that they are now suspended.
1539 for (i = 0; i < VM_MAXCPU; i++) {
1540 if (CPU_ISSET(i, &vm->active_cpus))
1541 vcpu_notify_event(vm, i, false);
1548 vm_exit_suspended(struct vm *vm, int vcpuid, uint64_t rip)
1550 struct vm_exit *vmexit;
1552 KASSERT(vm->suspend > VM_SUSPEND_NONE && vm->suspend < VM_SUSPEND_LAST,
1553 ("vm_exit_suspended: invalid suspend type %d", vm->suspend));
1555 vmexit = vm_exitinfo(vm, vcpuid);
1557 vmexit->inst_length = 0;
1558 vmexit->exitcode = VM_EXITCODE_SUSPENDED;
1559 vmexit->u.suspended.how = vm->suspend;
1563 vm_exit_rendezvous(struct vm *vm, int vcpuid, uint64_t rip)
1565 struct vm_exit *vmexit;
1567 KASSERT(vm->rendezvous_func != NULL, ("rendezvous not in progress"));
1569 vmexit = vm_exitinfo(vm, vcpuid);
1571 vmexit->inst_length = 0;
1572 vmexit->exitcode = VM_EXITCODE_RENDEZVOUS;
1573 vmm_stat_incr(vm, vcpuid, VMEXIT_RENDEZVOUS, 1);
1577 vm_exit_reqidle(struct vm *vm, int vcpuid, uint64_t rip)
1579 struct vm_exit *vmexit;
1581 vmexit = vm_exitinfo(vm, vcpuid);
1583 vmexit->inst_length = 0;
1584 vmexit->exitcode = VM_EXITCODE_REQIDLE;
1585 vmm_stat_incr(vm, vcpuid, VMEXIT_REQIDLE, 1);
1589 vm_exit_astpending(struct vm *vm, int vcpuid, uint64_t rip)
1591 struct vm_exit *vmexit;
1593 vmexit = vm_exitinfo(vm, vcpuid);
1595 vmexit->inst_length = 0;
1596 vmexit->exitcode = VM_EXITCODE_BOGUS;
1597 vmm_stat_incr(vm, vcpuid, VMEXIT_ASTPENDING, 1);
1601 vm_run(struct vm *vm, struct vm_run *vmrun)
1603 struct vm_eventinfo evinfo;
1608 struct vm_exit *vme;
1609 bool retu, intr_disabled;
1612 vcpuid = vmrun->cpuid;
1614 if (vcpuid < 0 || vcpuid >= VM_MAXCPU)
1617 if (!CPU_ISSET(vcpuid, &vm->active_cpus))
1620 if (CPU_ISSET(vcpuid, &vm->suspended_cpus))
1623 pmap = vmspace_pmap(vm->vmspace);
1624 vcpu = &vm->vcpu[vcpuid];
1625 vme = &vcpu->exitinfo;
1626 evinfo.rptr = &vm->rendezvous_func;
1627 evinfo.sptr = &vm->suspend;
1628 evinfo.iptr = &vcpu->reqidle;
1632 KASSERT(!CPU_ISSET(curcpu, &pmap->pm_active),
1633 ("vm_run: absurd pm_active"));
1637 pcb = PCPU_GET(curpcb);
1638 set_pcb_flags(pcb, PCB_FULL_IRET);
1640 restore_guest_fpustate(vcpu);
1642 vcpu_require_state(vm, vcpuid, VCPU_RUNNING);
1643 error = VMRUN(vm->cookie, vcpuid, vcpu->nextrip, pmap, &evinfo);
1644 vcpu_require_state(vm, vcpuid, VCPU_FROZEN);
1646 save_guest_fpustate(vcpu);
1648 vmm_stat_incr(vm, vcpuid, VCPU_TOTAL_RUNTIME, rdtsc() - tscval);
1654 vcpu->nextrip = vme->rip + vme->inst_length;
1655 switch (vme->exitcode) {
1656 case VM_EXITCODE_REQIDLE:
1657 error = vm_handle_reqidle(vm, vcpuid, &retu);
1659 case VM_EXITCODE_SUSPENDED:
1660 error = vm_handle_suspend(vm, vcpuid, &retu);
1662 case VM_EXITCODE_IOAPIC_EOI:
1663 vioapic_process_eoi(vm, vcpuid,
1664 vme->u.ioapic_eoi.vector);
1666 case VM_EXITCODE_RENDEZVOUS:
1667 vm_handle_rendezvous(vm, vcpuid);
1670 case VM_EXITCODE_HLT:
1671 intr_disabled = ((vme->u.hlt.rflags & PSL_I) == 0);
1672 error = vm_handle_hlt(vm, vcpuid, intr_disabled, &retu);
1674 case VM_EXITCODE_PAGING:
1675 error = vm_handle_paging(vm, vcpuid, &retu);
1677 case VM_EXITCODE_INST_EMUL:
1678 error = vm_handle_inst_emul(vm, vcpuid, &retu);
1680 case VM_EXITCODE_INOUT:
1681 case VM_EXITCODE_INOUT_STR:
1682 error = vm_handle_inout(vm, vcpuid, vme, &retu);
1684 case VM_EXITCODE_MONITOR:
1685 case VM_EXITCODE_MWAIT:
1686 vm_inject_ud(vm, vcpuid);
1689 retu = true; /* handled in userland */
1694 if (error == 0 && retu == false)
1697 VCPU_CTR2(vm, vcpuid, "retu %d/%d", error, vme->exitcode);
1699 /* copy the exit information */
1700 bcopy(vme, &vmrun->vm_exit, sizeof(struct vm_exit));
1705 vm_restart_instruction(void *arg, int vcpuid)
1709 enum vcpu_state state;
1714 if (vcpuid < 0 || vcpuid >= VM_MAXCPU)
1717 vcpu = &vm->vcpu[vcpuid];
1718 state = vcpu_get_state(vm, vcpuid, NULL);
1719 if (state == VCPU_RUNNING) {
1721 * When a vcpu is "running" the next instruction is determined
1722 * by adding 'rip' and 'inst_length' in the vcpu's 'exitinfo'.
1723 * Thus setting 'inst_length' to zero will cause the current
1724 * instruction to be restarted.
1726 vcpu->exitinfo.inst_length = 0;
1727 VCPU_CTR1(vm, vcpuid, "restarting instruction at %#lx by "
1728 "setting inst_length to zero", vcpu->exitinfo.rip);
1729 } else if (state == VCPU_FROZEN) {
1731 * When a vcpu is "frozen" it is outside the critical section
1732 * around VMRUN() and 'nextrip' points to the next instruction.
1733 * Thus instruction restart is achieved by setting 'nextrip'
1734 * to the vcpu's %rip.
1736 error = vm_get_register(vm, vcpuid, VM_REG_GUEST_RIP, &rip);
1737 KASSERT(!error, ("%s: error %d getting rip", __func__, error));
1738 VCPU_CTR2(vm, vcpuid, "restarting instruction by updating "
1739 "nextrip from %#lx to %#lx", vcpu->nextrip, rip);
1740 vcpu->nextrip = rip;
1742 panic("%s: invalid state %d", __func__, state);
1748 vm_exit_intinfo(struct vm *vm, int vcpuid, uint64_t info)
1753 if (vcpuid < 0 || vcpuid >= VM_MAXCPU)
1756 vcpu = &vm->vcpu[vcpuid];
1758 if (info & VM_INTINFO_VALID) {
1759 type = info & VM_INTINFO_TYPE;
1760 vector = info & 0xff;
1761 if (type == VM_INTINFO_NMI && vector != IDT_NMI)
1763 if (type == VM_INTINFO_HWEXCEPTION && vector >= 32)
1765 if (info & VM_INTINFO_RSVD)
1770 VCPU_CTR2(vm, vcpuid, "%s: info1(%#lx)", __func__, info);
1771 vcpu->exitintinfo = info;
1781 #define IDT_VE 20 /* Virtualization Exception (Intel specific) */
1783 static enum exc_class
1784 exception_class(uint64_t info)
1788 KASSERT(info & VM_INTINFO_VALID, ("intinfo must be valid: %#lx", info));
1789 type = info & VM_INTINFO_TYPE;
1790 vector = info & 0xff;
1792 /* Table 6-4, "Interrupt and Exception Classes", Intel SDM, Vol 3 */
1794 case VM_INTINFO_HWINTR:
1795 case VM_INTINFO_SWINTR:
1796 case VM_INTINFO_NMI:
1797 return (EXC_BENIGN);
1800 * Hardware exception.
1802 * SVM and VT-x use identical type values to represent NMI,
1803 * hardware interrupt and software interrupt.
1805 * SVM uses type '3' for all exceptions. VT-x uses type '3'
1806 * for exceptions except #BP and #OF. #BP and #OF use a type
1807 * value of '5' or '6'. Therefore we don't check for explicit
1808 * values of 'type' to classify 'intinfo' into a hardware
1817 return (EXC_PAGEFAULT);
1823 return (EXC_CONTRIBUTORY);
1825 return (EXC_BENIGN);
1830 nested_fault(struct vm *vm, int vcpuid, uint64_t info1, uint64_t info2,
1833 enum exc_class exc1, exc2;
1836 KASSERT(info1 & VM_INTINFO_VALID, ("info1 %#lx is not valid", info1));
1837 KASSERT(info2 & VM_INTINFO_VALID, ("info2 %#lx is not valid", info2));
1840 * If an exception occurs while attempting to call the double-fault
1841 * handler the processor enters shutdown mode (aka triple fault).
1843 type1 = info1 & VM_INTINFO_TYPE;
1844 vector1 = info1 & 0xff;
1845 if (type1 == VM_INTINFO_HWEXCEPTION && vector1 == IDT_DF) {
1846 VCPU_CTR2(vm, vcpuid, "triple fault: info1(%#lx), info2(%#lx)",
1848 vm_suspend(vm, VM_SUSPEND_TRIPLEFAULT);
1854 * Table 6-5 "Conditions for Generating a Double Fault", Intel SDM, Vol3
1856 exc1 = exception_class(info1);
1857 exc2 = exception_class(info2);
1858 if ((exc1 == EXC_CONTRIBUTORY && exc2 == EXC_CONTRIBUTORY) ||
1859 (exc1 == EXC_PAGEFAULT && exc2 != EXC_BENIGN)) {
1860 /* Convert nested fault into a double fault. */
1862 *retinfo |= VM_INTINFO_VALID | VM_INTINFO_HWEXCEPTION;
1863 *retinfo |= VM_INTINFO_DEL_ERRCODE;
1865 /* Handle exceptions serially */
1872 vcpu_exception_intinfo(struct vcpu *vcpu)
1876 if (vcpu->exception_pending) {
1877 info = vcpu->exc_vector & 0xff;
1878 info |= VM_INTINFO_VALID | VM_INTINFO_HWEXCEPTION;
1879 if (vcpu->exc_errcode_valid) {
1880 info |= VM_INTINFO_DEL_ERRCODE;
1881 info |= (uint64_t)vcpu->exc_errcode << 32;
1888 vm_entry_intinfo(struct vm *vm, int vcpuid, uint64_t *retinfo)
1891 uint64_t info1, info2;
1894 KASSERT(vcpuid >= 0 && vcpuid < VM_MAXCPU, ("invalid vcpu %d", vcpuid));
1896 vcpu = &vm->vcpu[vcpuid];
1898 info1 = vcpu->exitintinfo;
1899 vcpu->exitintinfo = 0;
1902 if (vcpu->exception_pending) {
1903 info2 = vcpu_exception_intinfo(vcpu);
1904 vcpu->exception_pending = 0;
1905 VCPU_CTR2(vm, vcpuid, "Exception %d delivered: %#lx",
1906 vcpu->exc_vector, info2);
1909 if ((info1 & VM_INTINFO_VALID) && (info2 & VM_INTINFO_VALID)) {
1910 valid = nested_fault(vm, vcpuid, info1, info2, retinfo);
1911 } else if (info1 & VM_INTINFO_VALID) {
1914 } else if (info2 & VM_INTINFO_VALID) {
1922 VCPU_CTR4(vm, vcpuid, "%s: info1(%#lx), info2(%#lx), "
1923 "retinfo(%#lx)", __func__, info1, info2, *retinfo);
1930 vm_get_intinfo(struct vm *vm, int vcpuid, uint64_t *info1, uint64_t *info2)
1934 if (vcpuid < 0 || vcpuid >= VM_MAXCPU)
1937 vcpu = &vm->vcpu[vcpuid];
1938 *info1 = vcpu->exitintinfo;
1939 *info2 = vcpu_exception_intinfo(vcpu);
1944 vm_inject_exception(struct vm *vm, int vcpuid, int vector, int errcode_valid,
1945 uint32_t errcode, int restart_instruction)
1951 if (vcpuid < 0 || vcpuid >= VM_MAXCPU)
1954 if (vector < 0 || vector >= 32)
1958 * A double fault exception should never be injected directly into
1959 * the guest. It is a derived exception that results from specific
1960 * combinations of nested faults.
1962 if (vector == IDT_DF)
1965 vcpu = &vm->vcpu[vcpuid];
1967 if (vcpu->exception_pending) {
1968 VCPU_CTR2(vm, vcpuid, "Unable to inject exception %d due to "
1969 "pending exception %d", vector, vcpu->exc_vector);
1973 if (errcode_valid) {
1975 * Exceptions don't deliver an error code in real mode.
1977 error = vm_get_register(vm, vcpuid, VM_REG_GUEST_CR0, ®val);
1978 KASSERT(!error, ("%s: error %d getting CR0", __func__, error));
1979 if (!(regval & CR0_PE))
1984 * From section 26.6.1 "Interruptibility State" in Intel SDM:
1986 * Event blocking by "STI" or "MOV SS" is cleared after guest executes
1987 * one instruction or incurs an exception.
1989 error = vm_set_register(vm, vcpuid, VM_REG_GUEST_INTR_SHADOW, 0);
1990 KASSERT(error == 0, ("%s: error %d clearing interrupt shadow",
1993 if (restart_instruction)
1994 vm_restart_instruction(vm, vcpuid);
1996 vcpu->exception_pending = 1;
1997 vcpu->exc_vector = vector;
1998 vcpu->exc_errcode = errcode;
1999 vcpu->exc_errcode_valid = errcode_valid;
2000 VCPU_CTR1(vm, vcpuid, "Exception %d pending", vector);
2005 vm_inject_fault(void *vmarg, int vcpuid, int vector, int errcode_valid,
2009 int error, restart_instruction;
2012 restart_instruction = 1;
2014 error = vm_inject_exception(vm, vcpuid, vector, errcode_valid,
2015 errcode, restart_instruction);
2016 KASSERT(error == 0, ("vm_inject_exception error %d", error));
2020 vm_inject_pf(void *vmarg, int vcpuid, int error_code, uint64_t cr2)
2026 VCPU_CTR2(vm, vcpuid, "Injecting page fault: error_code %#x, cr2 %#lx",
2029 error = vm_set_register(vm, vcpuid, VM_REG_GUEST_CR2, cr2);
2030 KASSERT(error == 0, ("vm_set_register(cr2) error %d", error));
2032 vm_inject_fault(vm, vcpuid, IDT_PF, 1, error_code);
2035 static VMM_STAT(VCPU_NMI_COUNT, "number of NMIs delivered to vcpu");
2038 vm_inject_nmi(struct vm *vm, int vcpuid)
2042 if (vcpuid < 0 || vcpuid >= VM_MAXCPU)
2045 vcpu = &vm->vcpu[vcpuid];
2047 vcpu->nmi_pending = 1;
2048 vcpu_notify_event(vm, vcpuid, false);
2053 vm_nmi_pending(struct vm *vm, int vcpuid)
2057 if (vcpuid < 0 || vcpuid >= VM_MAXCPU)
2058 panic("vm_nmi_pending: invalid vcpuid %d", vcpuid);
2060 vcpu = &vm->vcpu[vcpuid];
2062 return (vcpu->nmi_pending);
2066 vm_nmi_clear(struct vm *vm, int vcpuid)
2070 if (vcpuid < 0 || vcpuid >= VM_MAXCPU)
2071 panic("vm_nmi_pending: invalid vcpuid %d", vcpuid);
2073 vcpu = &vm->vcpu[vcpuid];
2075 if (vcpu->nmi_pending == 0)
2076 panic("vm_nmi_clear: inconsistent nmi_pending state");
2078 vcpu->nmi_pending = 0;
2079 vmm_stat_incr(vm, vcpuid, VCPU_NMI_COUNT, 1);
2082 static VMM_STAT(VCPU_EXTINT_COUNT, "number of ExtINTs delivered to vcpu");
2085 vm_inject_extint(struct vm *vm, int vcpuid)
2089 if (vcpuid < 0 || vcpuid >= VM_MAXCPU)
2092 vcpu = &vm->vcpu[vcpuid];
2094 vcpu->extint_pending = 1;
2095 vcpu_notify_event(vm, vcpuid, false);
2100 vm_extint_pending(struct vm *vm, int vcpuid)
2104 if (vcpuid < 0 || vcpuid >= VM_MAXCPU)
2105 panic("vm_extint_pending: invalid vcpuid %d", vcpuid);
2107 vcpu = &vm->vcpu[vcpuid];
2109 return (vcpu->extint_pending);
2113 vm_extint_clear(struct vm *vm, int vcpuid)
2117 if (vcpuid < 0 || vcpuid >= VM_MAXCPU)
2118 panic("vm_extint_pending: invalid vcpuid %d", vcpuid);
2120 vcpu = &vm->vcpu[vcpuid];
2122 if (vcpu->extint_pending == 0)
2123 panic("vm_extint_clear: inconsistent extint_pending state");
2125 vcpu->extint_pending = 0;
2126 vmm_stat_incr(vm, vcpuid, VCPU_EXTINT_COUNT, 1);
2130 vm_get_capability(struct vm *vm, int vcpu, int type, int *retval)
2132 if (vcpu < 0 || vcpu >= VM_MAXCPU)
2135 if (type < 0 || type >= VM_CAP_MAX)
2138 return (VMGETCAP(vm->cookie, vcpu, type, retval));
2142 vm_set_capability(struct vm *vm, int vcpu, int type, int val)
2144 if (vcpu < 0 || vcpu >= VM_MAXCPU)
2147 if (type < 0 || type >= VM_CAP_MAX)
2150 return (VMSETCAP(vm->cookie, vcpu, type, val));
2154 vm_lapic(struct vm *vm, int cpu)
2156 return (vm->vcpu[cpu].vlapic);
2160 vm_ioapic(struct vm *vm)
2163 return (vm->vioapic);
2167 vm_hpet(struct vm *vm)
2174 vmm_is_pptdev(int bus, int slot, int func)
2178 char *val, *cp, *cp2;
2182 * The length of an environment variable is limited to 128 bytes which
2183 * puts an upper limit on the number of passthru devices that may be
2184 * specified using a single environment variable.
2186 * Work around this by scanning multiple environment variable
2187 * names instead of a single one - yuck!
2189 const char *names[] = { "pptdevs", "pptdevs2", "pptdevs3", NULL };
2191 /* set pptdevs="1/2/3 4/5/6 7/8/9 10/11/12" */
2193 for (i = 0; names[i] != NULL && !found; i++) {
2194 cp = val = kern_getenv(names[i]);
2195 while (cp != NULL && *cp != '\0') {
2196 if ((cp2 = strchr(cp, ' ')) != NULL)
2199 n = sscanf(cp, "%d/%d/%d", &b, &s, &f);
2200 if (n == 3 && bus == b && slot == s && func == f) {
2216 vm_iommu_domain(struct vm *vm)
2223 vcpu_set_state(struct vm *vm, int vcpuid, enum vcpu_state newstate,
2229 if (vcpuid < 0 || vcpuid >= VM_MAXCPU)
2230 panic("vm_set_run_state: invalid vcpuid %d", vcpuid);
2232 vcpu = &vm->vcpu[vcpuid];
2235 error = vcpu_set_state_locked(vm, vcpuid, newstate, from_idle);
2242 vcpu_get_state(struct vm *vm, int vcpuid, int *hostcpu)
2245 enum vcpu_state state;
2247 if (vcpuid < 0 || vcpuid >= VM_MAXCPU)
2248 panic("vm_get_run_state: invalid vcpuid %d", vcpuid);
2250 vcpu = &vm->vcpu[vcpuid];
2253 state = vcpu->state;
2254 if (hostcpu != NULL)
2255 *hostcpu = vcpu->hostcpu;
2262 vm_activate_cpu(struct vm *vm, int vcpuid)
2265 if (vcpuid < 0 || vcpuid >= VM_MAXCPU)
2268 if (CPU_ISSET(vcpuid, &vm->active_cpus))
2271 VCPU_CTR0(vm, vcpuid, "activated");
2272 CPU_SET_ATOMIC(vcpuid, &vm->active_cpus);
2277 vm_active_cpus(struct vm *vm)
2280 return (vm->active_cpus);
2284 vm_suspended_cpus(struct vm *vm)
2287 return (vm->suspended_cpus);
2291 vcpu_stats(struct vm *vm, int vcpuid)
2294 return (vm->vcpu[vcpuid].stats);
2298 vm_get_x2apic_state(struct vm *vm, int vcpuid, enum x2apic_state *state)
2300 if (vcpuid < 0 || vcpuid >= VM_MAXCPU)
2303 *state = vm->vcpu[vcpuid].x2apic_state;
2309 vm_set_x2apic_state(struct vm *vm, int vcpuid, enum x2apic_state state)
2311 if (vcpuid < 0 || vcpuid >= VM_MAXCPU)
2314 if (state >= X2APIC_STATE_LAST)
2317 vm->vcpu[vcpuid].x2apic_state = state;
2319 vlapic_set_x2apic_state(vm, vcpuid, state);
2325 * This function is called to ensure that a vcpu "sees" a pending event
2326 * as soon as possible:
2327 * - If the vcpu thread is sleeping then it is woken up.
2328 * - If the vcpu is running on a different host_cpu then an IPI will be directed
2329 * to the host_cpu to cause the vcpu to trap into the hypervisor.
2332 vcpu_notify_event_locked(struct vcpu *vcpu, bool lapic_intr)
2336 hostcpu = vcpu->hostcpu;
2337 if (vcpu->state == VCPU_RUNNING) {
2338 KASSERT(hostcpu != NOCPU, ("vcpu running on invalid hostcpu"));
2339 if (hostcpu != curcpu) {
2341 vlapic_post_intr(vcpu->vlapic, hostcpu,
2344 ipi_cpu(hostcpu, vmm_ipinum);
2348 * If the 'vcpu' is running on 'curcpu' then it must
2349 * be sending a notification to itself (e.g. SELF_IPI).
2350 * The pending event will be picked up when the vcpu
2351 * transitions back to guest context.
2355 KASSERT(hostcpu == NOCPU, ("vcpu state %d not consistent "
2356 "with hostcpu %d", vcpu->state, hostcpu));
2357 if (vcpu->state == VCPU_SLEEPING)
2363 vcpu_notify_event(struct vm *vm, int vcpuid, bool lapic_intr)
2365 struct vcpu *vcpu = &vm->vcpu[vcpuid];
2368 vcpu_notify_event_locked(vcpu, lapic_intr);
2373 vm_get_vmspace(struct vm *vm)
2376 return (vm->vmspace);
2380 vm_apicid2vcpuid(struct vm *vm, int apicid)
2383 * XXX apic id is assumed to be numerically identical to vcpu id
2389 vm_smp_rendezvous(struct vm *vm, int vcpuid, cpuset_t dest,
2390 vm_rendezvous_func_t func, void *arg)
2395 * Enforce that this function is called without any locks
2397 WITNESS_WARN(WARN_PANIC, NULL, "vm_smp_rendezvous");
2398 KASSERT(vcpuid == -1 || (vcpuid >= 0 && vcpuid < VM_MAXCPU),
2399 ("vm_smp_rendezvous: invalid vcpuid %d", vcpuid));
2402 mtx_lock(&vm->rendezvous_mtx);
2403 if (vm->rendezvous_func != NULL) {
2405 * If a rendezvous is already in progress then we need to
2406 * call the rendezvous handler in case this 'vcpuid' is one
2407 * of the targets of the rendezvous.
2409 RENDEZVOUS_CTR0(vm, vcpuid, "Rendezvous already in progress");
2410 mtx_unlock(&vm->rendezvous_mtx);
2411 vm_handle_rendezvous(vm, vcpuid);
2414 KASSERT(vm->rendezvous_func == NULL, ("vm_smp_rendezvous: previous "
2415 "rendezvous is still in progress"));
2417 RENDEZVOUS_CTR0(vm, vcpuid, "Initiating rendezvous");
2418 vm->rendezvous_req_cpus = dest;
2419 CPU_ZERO(&vm->rendezvous_done_cpus);
2420 vm->rendezvous_arg = arg;
2421 vm_set_rendezvous_func(vm, func);
2422 mtx_unlock(&vm->rendezvous_mtx);
2425 * Wake up any sleeping vcpus and trigger a VM-exit in any running
2426 * vcpus so they handle the rendezvous as soon as possible.
2428 for (i = 0; i < VM_MAXCPU; i++) {
2429 if (CPU_ISSET(i, &dest))
2430 vcpu_notify_event(vm, i, false);
2433 vm_handle_rendezvous(vm, vcpuid);
2437 vm_atpic(struct vm *vm)
2439 return (vm->vatpic);
2443 vm_atpit(struct vm *vm)
2445 return (vm->vatpit);
2449 vm_pmtmr(struct vm *vm)
2452 return (vm->vpmtmr);
2456 vm_rtc(struct vm *vm)
2463 vm_segment_name(int seg)
2465 static enum vm_reg_name seg_names[] = {
2474 KASSERT(seg >= 0 && seg < nitems(seg_names),
2475 ("%s: invalid segment encoding %d", __func__, seg));
2476 return (seg_names[seg]);
2480 vm_copy_teardown(struct vm *vm, int vcpuid, struct vm_copyinfo *copyinfo,
2485 for (idx = 0; idx < num_copyinfo; idx++) {
2486 if (copyinfo[idx].cookie != NULL)
2487 vm_gpa_release(copyinfo[idx].cookie);
2489 bzero(copyinfo, num_copyinfo * sizeof(struct vm_copyinfo));
2493 vm_copy_setup(struct vm *vm, int vcpuid, struct vm_guest_paging *paging,
2494 uint64_t gla, size_t len, int prot, struct vm_copyinfo *copyinfo,
2495 int num_copyinfo, int *fault)
2497 int error, idx, nused;
2498 size_t n, off, remaining;
2502 bzero(copyinfo, sizeof(struct vm_copyinfo) * num_copyinfo);
2506 while (remaining > 0) {
2507 KASSERT(nused < num_copyinfo, ("insufficient vm_copyinfo"));
2508 error = vm_gla2gpa(vm, vcpuid, paging, gla, prot, &gpa, fault);
2509 if (error || *fault)
2511 off = gpa & PAGE_MASK;
2512 n = min(remaining, PAGE_SIZE - off);
2513 copyinfo[nused].gpa = gpa;
2514 copyinfo[nused].len = n;
2520 for (idx = 0; idx < nused; idx++) {
2521 hva = vm_gpa_hold(vm, vcpuid, copyinfo[idx].gpa,
2522 copyinfo[idx].len, prot, &cookie);
2525 copyinfo[idx].hva = hva;
2526 copyinfo[idx].cookie = cookie;
2530 vm_copy_teardown(vm, vcpuid, copyinfo, num_copyinfo);
2539 vm_copyin(struct vm *vm, int vcpuid, struct vm_copyinfo *copyinfo, void *kaddr,
2548 bcopy(copyinfo[idx].hva, dst, copyinfo[idx].len);
2549 len -= copyinfo[idx].len;
2550 dst += copyinfo[idx].len;
2556 vm_copyout(struct vm *vm, int vcpuid, const void *kaddr,
2557 struct vm_copyinfo *copyinfo, size_t len)
2565 bcopy(src, copyinfo[idx].hva, copyinfo[idx].len);
2566 len -= copyinfo[idx].len;
2567 src += copyinfo[idx].len;
2573 * Return the amount of in-use and wired memory for the VM. Since
2574 * these are global stats, only return the values with for vCPU 0
2576 VMM_STAT_DECLARE(VMM_MEM_RESIDENT);
2577 VMM_STAT_DECLARE(VMM_MEM_WIRED);
2580 vm_get_rescnt(struct vm *vm, int vcpu, struct vmm_stat_type *stat)
2584 vmm_stat_set(vm, vcpu, VMM_MEM_RESIDENT,
2585 PAGE_SIZE * vmspace_resident_count(vm->vmspace));
2590 vm_get_wiredcnt(struct vm *vm, int vcpu, struct vmm_stat_type *stat)
2594 vmm_stat_set(vm, vcpu, VMM_MEM_WIRED,
2595 PAGE_SIZE * pmap_wired_count(vmspace_pmap(vm->vmspace)));
2599 VMM_STAT_FUNC(VMM_MEM_RESIDENT, "Resident memory", vm_get_rescnt);
2600 VMM_STAT_FUNC(VMM_MEM_WIRED, "Wired memory", vm_get_wiredcnt);