2 * SPDX-License-Identifier: BSD-2-Clause-FreeBSD
4 * Copyright (c) 2011 NetApp, Inc.
7 * Redistribution and use in source and binary forms, with or without
8 * modification, are permitted provided that the following conditions
10 * 1. Redistributions of source code must retain the above copyright
11 * notice, this list of conditions and the following disclaimer.
12 * 2. Redistributions in binary form must reproduce the above copyright
13 * notice, this list of conditions and the following disclaimer in the
14 * documentation and/or other materials provided with the distribution.
16 * THIS SOFTWARE IS PROVIDED BY NETAPP, INC ``AS IS'' AND
17 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
18 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
19 * ARE DISCLAIMED. IN NO EVENT SHALL NETAPP, INC OR CONTRIBUTORS BE LIABLE
20 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
21 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
22 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
23 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
24 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
25 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
31 #include <sys/cdefs.h>
32 __FBSDID("$FreeBSD$");
34 #include "opt_bhyve_snapshot.h"
36 #include <sys/param.h>
37 #include <sys/systm.h>
38 #include <sys/kernel.h>
39 #include <sys/module.h>
40 #include <sys/sysctl.h>
41 #include <sys/malloc.h>
44 #include <sys/mutex.h>
46 #include <sys/rwlock.h>
47 #include <sys/sched.h>
49 #include <sys/vnode.h>
52 #include <vm/vm_param.h>
53 #include <vm/vm_extern.h>
54 #include <vm/vm_object.h>
55 #include <vm/vm_page.h>
57 #include <vm/vm_map.h>
58 #include <vm/vm_pager.h>
59 #include <vm/vm_kern.h>
60 #include <vm/vnode_pager.h>
61 #include <vm/swap_pager.h>
64 #include <machine/cpu.h>
65 #include <machine/pcb.h>
66 #include <machine/smp.h>
67 #include <machine/md_var.h>
69 #include <x86/apicreg.h>
70 #include <x86/ifunc.h>
72 #include <machine/vmm.h>
73 #include <machine/vmm_dev.h>
74 #include <machine/vmm_instruction_emul.h>
75 #include <machine/vmm_snapshot.h>
77 #include "vmm_ioport.h"
90 #include "vmm_lapic.h"
99 * (a) allocated when vcpu is created
100 * (i) initialized when vcpu is created and when it is reinitialized
101 * (o) initialized the first time the vcpu is created
102 * (x) initialized before use
105 struct mtx mtx; /* (o) protects 'state' and 'hostcpu' */
106 enum vcpu_state state; /* (o) vcpu state */
107 int hostcpu; /* (o) vcpu's host cpu */
108 int reqidle; /* (i) request vcpu to idle */
109 struct vlapic *vlapic; /* (i) APIC device model */
110 enum x2apic_state x2apic_state; /* (i) APIC mode */
111 uint64_t exitintinfo; /* (i) events pending at VM exit */
112 int nmi_pending; /* (i) NMI pending */
113 int extint_pending; /* (i) INTR pending */
114 int exception_pending; /* (i) exception pending */
115 int exc_vector; /* (x) exception collateral */
116 int exc_errcode_valid;
117 uint32_t exc_errcode;
118 struct savefpu *guestfpu; /* (a,i) guest fpu state */
119 uint64_t guest_xcr0; /* (i) guest %xcr0 register */
120 void *stats; /* (a,i) statistics */
121 struct vm_exit exitinfo; /* (x) exit reason and collateral */
122 uint64_t nextrip; /* (x) next instruction to execute */
123 uint64_t tsc_offset; /* (o) TSC offsetting */
126 #define vcpu_lock_initialized(v) mtx_initialized(&((v)->mtx))
127 #define vcpu_lock_init(v) mtx_init(&((v)->mtx), "vcpu lock", 0, MTX_SPIN)
128 #define vcpu_lock(v) mtx_lock_spin(&((v)->mtx))
129 #define vcpu_unlock(v) mtx_unlock_spin(&((v)->mtx))
130 #define vcpu_assert_locked(v) mtx_assert(&((v)->mtx), MA_OWNED)
135 struct vm_object *object;
137 #define VM_MAX_MEMSEGS 4
147 #define VM_MAX_MEMMAPS 8
151 * (o) initialized the first time the VM is created
152 * (i) initialized when VM is created and when it is reinitialized
153 * (x) initialized before use
156 void *cookie; /* (i) cpu-specific data */
157 void *iommu; /* (x) iommu-specific data */
158 struct vhpet *vhpet; /* (i) virtual HPET */
159 struct vioapic *vioapic; /* (i) virtual ioapic */
160 struct vatpic *vatpic; /* (i) virtual atpic */
161 struct vatpit *vatpit; /* (i) virtual atpit */
162 struct vpmtmr *vpmtmr; /* (i) virtual ACPI PM timer */
163 struct vrtc *vrtc; /* (o) virtual RTC */
164 volatile cpuset_t active_cpus; /* (i) active vcpus */
165 volatile cpuset_t debug_cpus; /* (i) vcpus stopped for debug */
166 int suspend; /* (i) stop VM execution */
167 volatile cpuset_t suspended_cpus; /* (i) suspended vcpus */
168 volatile cpuset_t halted_cpus; /* (x) cpus in a hard halt */
169 cpuset_t rendezvous_req_cpus; /* (x) rendezvous requested */
170 cpuset_t rendezvous_done_cpus; /* (x) rendezvous finished */
171 void *rendezvous_arg; /* (x) rendezvous func/arg */
172 vm_rendezvous_func_t rendezvous_func;
173 struct mtx rendezvous_mtx; /* (o) rendezvous lock */
174 struct mem_map mem_maps[VM_MAX_MEMMAPS]; /* (i) guest address space */
175 struct mem_seg mem_segs[VM_MAX_MEMSEGS]; /* (o) guest memory regions */
176 struct vmspace *vmspace; /* (o) guest's address space */
177 char name[VM_MAX_NAMELEN+1]; /* (o) virtual machine name */
178 struct vcpu vcpu[VM_MAXCPU]; /* (i) guest vcpus */
179 /* The following describe the vm cpu topology */
180 uint16_t sockets; /* (o) num of sockets */
181 uint16_t cores; /* (o) num of cores/socket */
182 uint16_t threads; /* (o) num of threads/core */
183 uint16_t maxcpus; /* (o) max pluggable cpus */
186 static int vmm_initialized;
188 static void vmmops_panic(void);
193 panic("vmm_ops func called when !vmm_is_intel() && !vmm_is_svm()");
196 #define DEFINE_VMMOPS_IFUNC(ret_type, opname, args) \
197 DEFINE_IFUNC(static, ret_type, vmmops_##opname, args) \
199 if (vmm_is_intel()) \
200 return (vmm_ops_intel.opname); \
201 else if (vmm_is_svm()) \
202 return (vmm_ops_amd.opname); \
204 return ((ret_type (*)args)vmmops_panic); \
207 DEFINE_VMMOPS_IFUNC(int, modinit, (int ipinum))
208 DEFINE_VMMOPS_IFUNC(int, modcleanup, (void))
209 DEFINE_VMMOPS_IFUNC(void, modresume, (void))
210 DEFINE_VMMOPS_IFUNC(void *, init, (struct vm *vm, struct pmap *pmap))
211 DEFINE_VMMOPS_IFUNC(int, run, (void *vmi, int vcpu, register_t rip,
212 struct pmap *pmap, struct vm_eventinfo *info))
213 DEFINE_VMMOPS_IFUNC(void, cleanup, (void *vmi))
214 DEFINE_VMMOPS_IFUNC(int, getreg, (void *vmi, int vcpu, int num,
216 DEFINE_VMMOPS_IFUNC(int, setreg, (void *vmi, int vcpu, int num,
218 DEFINE_VMMOPS_IFUNC(int, getdesc, (void *vmi, int vcpu, int num,
219 struct seg_desc *desc))
220 DEFINE_VMMOPS_IFUNC(int, setdesc, (void *vmi, int vcpu, int num,
221 struct seg_desc *desc))
222 DEFINE_VMMOPS_IFUNC(int, getcap, (void *vmi, int vcpu, int num, int *retval))
223 DEFINE_VMMOPS_IFUNC(int, setcap, (void *vmi, int vcpu, int num, int val))
224 DEFINE_VMMOPS_IFUNC(struct vmspace *, vmspace_alloc, (vm_offset_t min,
226 DEFINE_VMMOPS_IFUNC(void, vmspace_free, (struct vmspace *vmspace))
227 DEFINE_VMMOPS_IFUNC(struct vlapic *, vlapic_init, (void *vmi, int vcpu))
228 DEFINE_VMMOPS_IFUNC(void, vlapic_cleanup, (void *vmi, struct vlapic *vlapic))
229 #ifdef BHYVE_SNAPSHOT
230 DEFINE_VMMOPS_IFUNC(int, snapshot, (void *vmi, struct vm_snapshot_meta
232 DEFINE_VMMOPS_IFUNC(int, vmcx_snapshot, (void *vmi, struct vm_snapshot_meta
234 DEFINE_VMMOPS_IFUNC(int, restore_tsc, (void *vmi, int vcpuid, uint64_t now))
237 #define fpu_start_emulating() load_cr0(rcr0() | CR0_TS)
238 #define fpu_stop_emulating() clts()
240 SDT_PROVIDER_DEFINE(vmm);
242 static MALLOC_DEFINE(M_VM, "vm", "vm");
245 static VMM_STAT(VCPU_TOTAL_RUNTIME, "vcpu total runtime");
247 SYSCTL_NODE(_hw, OID_AUTO, vmm, CTLFLAG_RW | CTLFLAG_MPSAFE, NULL,
251 * Halt the guest if all vcpus are executing a HLT instruction with
252 * interrupts disabled.
254 static int halt_detection_enabled = 1;
255 SYSCTL_INT(_hw_vmm, OID_AUTO, halt_detection, CTLFLAG_RDTUN,
256 &halt_detection_enabled, 0,
257 "Halt VM if all vcpus execute HLT with interrupts disabled");
259 static int vmm_ipinum;
260 SYSCTL_INT(_hw_vmm, OID_AUTO, ipinum, CTLFLAG_RD, &vmm_ipinum, 0,
261 "IPI vector used for vcpu notifications");
263 static int trace_guest_exceptions;
264 SYSCTL_INT(_hw_vmm, OID_AUTO, trace_guest_exceptions, CTLFLAG_RDTUN,
265 &trace_guest_exceptions, 0,
266 "Trap into hypervisor on all guest exceptions and reflect them back");
268 static int trap_wbinvd;
269 SYSCTL_INT(_hw_vmm, OID_AUTO, trap_wbinvd, CTLFLAG_RDTUN, &trap_wbinvd, 0,
270 "WBINVD triggers a VM-exit");
272 static void vm_free_memmap(struct vm *vm, int ident);
273 static bool sysmem_mapping(struct vm *vm, struct mem_map *mm);
274 static void vcpu_notify_event_locked(struct vcpu *vcpu, bool lapic_intr);
278 vcpu_state2str(enum vcpu_state state)
297 vcpu_cleanup(struct vm *vm, int i, bool destroy)
299 struct vcpu *vcpu = &vm->vcpu[i];
301 vmmops_vlapic_cleanup(vm->cookie, vcpu->vlapic);
303 vmm_stat_free(vcpu->stats);
304 fpu_save_area_free(vcpu->guestfpu);
309 vcpu_init(struct vm *vm, int vcpu_id, bool create)
313 KASSERT(vcpu_id >= 0 && vcpu_id < vm->maxcpus,
314 ("vcpu_init: invalid vcpu %d", vcpu_id));
316 vcpu = &vm->vcpu[vcpu_id];
319 KASSERT(!vcpu_lock_initialized(vcpu), ("vcpu %d already "
320 "initialized", vcpu_id));
321 vcpu_lock_init(vcpu);
322 vcpu->state = VCPU_IDLE;
323 vcpu->hostcpu = NOCPU;
324 vcpu->guestfpu = fpu_save_area_alloc();
325 vcpu->stats = vmm_stat_alloc();
326 vcpu->tsc_offset = 0;
329 vcpu->vlapic = vmmops_vlapic_init(vm->cookie, vcpu_id);
330 vm_set_x2apic_state(vm, vcpu_id, X2APIC_DISABLED);
332 vcpu->exitintinfo = 0;
333 vcpu->nmi_pending = 0;
334 vcpu->extint_pending = 0;
335 vcpu->exception_pending = 0;
336 vcpu->guest_xcr0 = XFEATURE_ENABLED_X87;
337 fpu_save_area_reset(vcpu->guestfpu);
338 vmm_stat_init(vcpu->stats);
342 vcpu_trace_exceptions(struct vm *vm, int vcpuid)
345 return (trace_guest_exceptions);
349 vcpu_trap_wbinvd(struct vm *vm, int vcpuid)
351 return (trap_wbinvd);
355 vm_exitinfo(struct vm *vm, int cpuid)
359 if (cpuid < 0 || cpuid >= vm->maxcpus)
360 panic("vm_exitinfo: invalid cpuid %d", cpuid);
362 vcpu = &vm->vcpu[cpuid];
364 return (&vcpu->exitinfo);
372 if (!vmm_is_hw_supported())
375 vmm_host_state_init();
377 vmm_ipinum = lapic_ipi_alloc(pti ? &IDTVEC(justreturn1_pti) :
378 &IDTVEC(justreturn));
380 vmm_ipinum = IPI_AST;
382 error = vmm_mem_init();
386 vmm_resume_p = vmmops_modresume;
388 return (vmmops_modinit(vmm_ipinum));
392 vmm_handler(module_t mod, int what, void *arg)
398 if (vmm_is_hw_supported()) {
408 if (vmm_is_hw_supported()) {
409 error = vmmdev_cleanup();
413 if (vmm_ipinum != IPI_AST)
414 lapic_ipi_free(vmm_ipinum);
415 error = vmmops_modcleanup();
417 * Something bad happened - prevent new
418 * VMs from being created
434 static moduledata_t vmm_kmod = {
441 * vmm initialization has the following dependencies:
443 * - VT-x initialization requires smp_rendezvous() and therefore must happen
444 * after SMP is fully functional (after SI_SUB_SMP).
446 DECLARE_MODULE(vmm, vmm_kmod, SI_SUB_SMP + 1, SI_ORDER_ANY);
447 MODULE_VERSION(vmm, 1);
450 vm_init(struct vm *vm, bool create)
454 vm->cookie = vmmops_init(vm, vmspace_pmap(vm->vmspace));
456 vm->vioapic = vioapic_init(vm);
457 vm->vhpet = vhpet_init(vm);
458 vm->vatpic = vatpic_init(vm);
459 vm->vatpit = vatpit_init(vm);
460 vm->vpmtmr = vpmtmr_init(vm);
462 vm->vrtc = vrtc_init(vm);
464 CPU_ZERO(&vm->active_cpus);
465 CPU_ZERO(&vm->debug_cpus);
468 CPU_ZERO(&vm->suspended_cpus);
470 for (i = 0; i < vm->maxcpus; i++)
471 vcpu_init(vm, i, create);
475 * The default CPU topology is a single thread per package.
477 u_int cores_per_package = 1;
478 u_int threads_per_core = 1;
481 vm_create(const char *name, struct vm **retvm)
484 struct vmspace *vmspace;
487 * If vmm.ko could not be successfully initialized then don't attempt
488 * to create the virtual machine.
490 if (!vmm_initialized)
493 if (name == NULL || strnlen(name, VM_MAX_NAMELEN + 1) ==
497 vmspace = vmmops_vmspace_alloc(0, VM_MAXUSER_ADDRESS_LA48);
501 vm = malloc(sizeof(struct vm), M_VM, M_WAITOK | M_ZERO);
502 strcpy(vm->name, name);
503 vm->vmspace = vmspace;
504 mtx_init(&vm->rendezvous_mtx, "vm rendezvous lock", 0, MTX_DEF);
507 vm->cores = cores_per_package; /* XXX backwards compatibility */
508 vm->threads = threads_per_core; /* XXX backwards compatibility */
509 vm->maxcpus = VM_MAXCPU; /* XXX temp to keep code working */
518 vm_get_topology(struct vm *vm, uint16_t *sockets, uint16_t *cores,
519 uint16_t *threads, uint16_t *maxcpus)
521 *sockets = vm->sockets;
523 *threads = vm->threads;
524 *maxcpus = vm->maxcpus;
528 vm_get_maxcpus(struct vm *vm)
530 return (vm->maxcpus);
534 vm_set_topology(struct vm *vm, uint16_t sockets, uint16_t cores,
535 uint16_t threads, uint16_t maxcpus)
538 return (EINVAL); /* XXX remove when supported */
539 if ((sockets * cores * threads) > vm->maxcpus)
541 /* XXX need to check sockets * cores * threads == vCPU, how? */
542 vm->sockets = sockets;
544 vm->threads = threads;
545 vm->maxcpus = VM_MAXCPU; /* XXX temp to keep code working */
550 vm_cleanup(struct vm *vm, bool destroy)
555 ppt_unassign_all(vm);
557 if (vm->iommu != NULL)
558 iommu_destroy_domain(vm->iommu);
561 vrtc_cleanup(vm->vrtc);
563 vrtc_reset(vm->vrtc);
564 vpmtmr_cleanup(vm->vpmtmr);
565 vatpit_cleanup(vm->vatpit);
566 vhpet_cleanup(vm->vhpet);
567 vatpic_cleanup(vm->vatpic);
568 vioapic_cleanup(vm->vioapic);
570 for (i = 0; i < vm->maxcpus; i++)
571 vcpu_cleanup(vm, i, destroy);
573 vmmops_cleanup(vm->cookie);
576 * System memory is removed from the guest address space only when
577 * the VM is destroyed. This is because the mapping remains the same
580 * Device memory can be relocated by the guest (e.g. using PCI BARs)
581 * so those mappings are removed on a VM reset.
583 for (i = 0; i < VM_MAX_MEMMAPS; i++) {
584 mm = &vm->mem_maps[i];
585 if (destroy || !sysmem_mapping(vm, mm))
586 vm_free_memmap(vm, i);
590 for (i = 0; i < VM_MAX_MEMSEGS; i++)
591 vm_free_memseg(vm, i);
593 vmmops_vmspace_free(vm->vmspace);
599 vm_destroy(struct vm *vm)
601 vm_cleanup(vm, true);
606 vm_reinit(struct vm *vm)
611 * A virtual machine can be reset only if all vcpus are suspended.
613 if (CPU_CMP(&vm->suspended_cpus, &vm->active_cpus) == 0) {
614 vm_cleanup(vm, false);
625 vm_name(struct vm *vm)
631 vm_map_mmio(struct vm *vm, vm_paddr_t gpa, size_t len, vm_paddr_t hpa)
635 if ((obj = vmm_mmio_alloc(vm->vmspace, gpa, len, hpa)) == NULL)
642 vm_unmap_mmio(struct vm *vm, vm_paddr_t gpa, size_t len)
645 vmm_mmio_free(vm->vmspace, gpa, len);
650 * Return 'true' if 'gpa' is allocated in the guest address space.
652 * This function is called in the context of a running vcpu which acts as
653 * an implicit lock on 'vm->mem_maps[]'.
656 vm_mem_allocated(struct vm *vm, int vcpuid, vm_paddr_t gpa)
663 state = vcpu_get_state(vm, vcpuid, &hostcpu);
664 KASSERT(state == VCPU_RUNNING && hostcpu == curcpu,
665 ("%s: invalid vcpu state %d/%d", __func__, state, hostcpu));
668 for (i = 0; i < VM_MAX_MEMMAPS; i++) {
669 mm = &vm->mem_maps[i];
670 if (mm->len != 0 && gpa >= mm->gpa && gpa < mm->gpa + mm->len)
671 return (true); /* 'gpa' is sysmem or devmem */
674 if (ppt_is_mmio(vm, gpa))
675 return (true); /* 'gpa' is pci passthru mmio */
681 vm_alloc_memseg(struct vm *vm, int ident, size_t len, bool sysmem)
686 if (ident < 0 || ident >= VM_MAX_MEMSEGS)
689 if (len == 0 || (len & PAGE_MASK))
692 seg = &vm->mem_segs[ident];
693 if (seg->object != NULL) {
694 if (seg->len == len && seg->sysmem == sysmem)
700 obj = vm_object_allocate(OBJT_SWAP, len >> PAGE_SHIFT);
706 seg->sysmem = sysmem;
711 vm_get_memseg(struct vm *vm, int ident, size_t *len, bool *sysmem,
716 if (ident < 0 || ident >= VM_MAX_MEMSEGS)
719 seg = &vm->mem_segs[ident];
723 *sysmem = seg->sysmem;
725 *objptr = seg->object;
730 vm_free_memseg(struct vm *vm, int ident)
734 KASSERT(ident >= 0 && ident < VM_MAX_MEMSEGS,
735 ("%s: invalid memseg ident %d", __func__, ident));
737 seg = &vm->mem_segs[ident];
738 if (seg->object != NULL) {
739 vm_object_deallocate(seg->object);
740 bzero(seg, sizeof(struct mem_seg));
745 vm_mmap_memseg(struct vm *vm, vm_paddr_t gpa, int segid, vm_ooffset_t first,
746 size_t len, int prot, int flags)
749 struct mem_map *m, *map;
753 if (prot == 0 || (prot & ~(VM_PROT_ALL)) != 0)
756 if (flags & ~VM_MEMMAP_F_WIRED)
759 if (segid < 0 || segid >= VM_MAX_MEMSEGS)
762 seg = &vm->mem_segs[segid];
763 if (seg->object == NULL)
767 if (first < 0 || first >= last || last > seg->len)
770 if ((gpa | first | last) & PAGE_MASK)
774 for (i = 0; i < VM_MAX_MEMMAPS; i++) {
775 m = &vm->mem_maps[i];
785 error = vm_map_find(&vm->vmspace->vm_map, seg->object, first, &gpa,
786 len, 0, VMFS_NO_SPACE, prot, prot, 0);
787 if (error != KERN_SUCCESS)
790 vm_object_reference(seg->object);
792 if (flags & VM_MEMMAP_F_WIRED) {
793 error = vm_map_wire(&vm->vmspace->vm_map, gpa, gpa + len,
794 VM_MAP_WIRE_USER | VM_MAP_WIRE_NOHOLES);
795 if (error != KERN_SUCCESS) {
796 vm_map_remove(&vm->vmspace->vm_map, gpa, gpa + len);
797 return (error == KERN_RESOURCE_SHORTAGE ? ENOMEM :
812 vm_munmap_memseg(struct vm *vm, vm_paddr_t gpa, size_t len)
817 for (i = 0; i < VM_MAX_MEMMAPS; i++) {
818 m = &vm->mem_maps[i];
819 if (m->gpa == gpa && m->len == len &&
820 (m->flags & VM_MEMMAP_F_IOMMU) == 0) {
821 vm_free_memmap(vm, i);
830 vm_mmap_getnext(struct vm *vm, vm_paddr_t *gpa, int *segid,
831 vm_ooffset_t *segoff, size_t *len, int *prot, int *flags)
833 struct mem_map *mm, *mmnext;
837 for (i = 0; i < VM_MAX_MEMMAPS; i++) {
838 mm = &vm->mem_maps[i];
839 if (mm->len == 0 || mm->gpa < *gpa)
841 if (mmnext == NULL || mm->gpa < mmnext->gpa)
845 if (mmnext != NULL) {
848 *segid = mmnext->segid;
850 *segoff = mmnext->segoff;
854 *prot = mmnext->prot;
856 *flags = mmnext->flags;
864 vm_free_memmap(struct vm *vm, int ident)
869 mm = &vm->mem_maps[ident];
871 error = vm_map_remove(&vm->vmspace->vm_map, mm->gpa,
873 KASSERT(error == KERN_SUCCESS, ("%s: vm_map_remove error %d",
875 bzero(mm, sizeof(struct mem_map));
880 sysmem_mapping(struct vm *vm, struct mem_map *mm)
883 if (mm->len != 0 && vm->mem_segs[mm->segid].sysmem)
890 vmm_sysmem_maxaddr(struct vm *vm)
897 for (i = 0; i < VM_MAX_MEMMAPS; i++) {
898 mm = &vm->mem_maps[i];
899 if (sysmem_mapping(vm, mm)) {
900 if (maxaddr < mm->gpa + mm->len)
901 maxaddr = mm->gpa + mm->len;
908 vm_iommu_modify(struct vm *vm, bool map)
913 void *vp, *cookie, *host_domain;
916 host_domain = iommu_host_domain();
918 for (i = 0; i < VM_MAX_MEMMAPS; i++) {
919 mm = &vm->mem_maps[i];
920 if (!sysmem_mapping(vm, mm))
924 KASSERT((mm->flags & VM_MEMMAP_F_IOMMU) == 0,
925 ("iommu map found invalid memmap %#lx/%#lx/%#x",
926 mm->gpa, mm->len, mm->flags));
927 if ((mm->flags & VM_MEMMAP_F_WIRED) == 0)
929 mm->flags |= VM_MEMMAP_F_IOMMU;
931 if ((mm->flags & VM_MEMMAP_F_IOMMU) == 0)
933 mm->flags &= ~VM_MEMMAP_F_IOMMU;
934 KASSERT((mm->flags & VM_MEMMAP_F_WIRED) != 0,
935 ("iommu unmap found invalid memmap %#lx/%#lx/%#x",
936 mm->gpa, mm->len, mm->flags));
940 while (gpa < mm->gpa + mm->len) {
941 vp = vm_gpa_hold(vm, -1, gpa, PAGE_SIZE, VM_PROT_WRITE,
943 KASSERT(vp != NULL, ("vm(%s) could not map gpa %#lx",
946 vm_gpa_release(cookie);
948 hpa = DMAP_TO_PHYS((uintptr_t)vp);
950 iommu_create_mapping(vm->iommu, gpa, hpa, sz);
952 iommu_remove_mapping(vm->iommu, gpa, sz);
960 * Invalidate the cached translations associated with the domain
961 * from which pages were removed.
964 iommu_invalidate_tlb(host_domain);
966 iommu_invalidate_tlb(vm->iommu);
969 #define vm_iommu_unmap(vm) vm_iommu_modify((vm), false)
970 #define vm_iommu_map(vm) vm_iommu_modify((vm), true)
973 vm_unassign_pptdev(struct vm *vm, int bus, int slot, int func)
977 error = ppt_unassign_device(vm, bus, slot, func);
981 if (ppt_assigned_devices(vm) == 0)
988 vm_assign_pptdev(struct vm *vm, int bus, int slot, int func)
993 /* Set up the IOMMU to do the 'gpa' to 'hpa' translation */
994 if (ppt_assigned_devices(vm) == 0) {
995 KASSERT(vm->iommu == NULL,
996 ("vm_assign_pptdev: iommu must be NULL"));
997 maxaddr = vmm_sysmem_maxaddr(vm);
998 vm->iommu = iommu_create_domain(maxaddr);
999 if (vm->iommu == NULL)
1004 error = ppt_assign_device(vm, bus, slot, func);
1009 vm_gpa_hold(struct vm *vm, int vcpuid, vm_paddr_t gpa, size_t len, int reqprot,
1012 int i, count, pageoff;
1017 * All vcpus are frozen by ioctls that modify the memory map
1018 * (e.g. VM_MMAP_MEMSEG). Therefore 'vm->memmap[]' stability is
1019 * guaranteed if at least one vcpu is in the VCPU_FROZEN state.
1022 KASSERT(vcpuid >= -1 && vcpuid < vm->maxcpus, ("%s: invalid vcpuid %d",
1024 for (i = 0; i < vm->maxcpus; i++) {
1025 if (vcpuid != -1 && vcpuid != i)
1027 state = vcpu_get_state(vm, i, NULL);
1028 KASSERT(state == VCPU_FROZEN, ("%s: invalid vcpu state %d",
1032 pageoff = gpa & PAGE_MASK;
1033 if (len > PAGE_SIZE - pageoff)
1034 panic("vm_gpa_hold: invalid gpa/len: 0x%016lx/%lu", gpa, len);
1037 for (i = 0; i < VM_MAX_MEMMAPS; i++) {
1038 mm = &vm->mem_maps[i];
1039 if (gpa >= mm->gpa && gpa < mm->gpa + mm->len) {
1040 count = vm_fault_quick_hold_pages(&vm->vmspace->vm_map,
1041 trunc_page(gpa), PAGE_SIZE, reqprot, &m, 1);
1048 return ((void *)(PHYS_TO_DMAP(VM_PAGE_TO_PHYS(m)) + pageoff));
1056 vm_gpa_release(void *cookie)
1058 vm_page_t m = cookie;
1060 vm_page_unwire(m, PQ_ACTIVE);
1064 vm_get_register(struct vm *vm, int vcpu, int reg, uint64_t *retval)
1067 if (vcpu < 0 || vcpu >= vm->maxcpus)
1070 if (reg >= VM_REG_LAST)
1073 return (vmmops_getreg(vm->cookie, vcpu, reg, retval));
1077 vm_set_register(struct vm *vm, int vcpuid, int reg, uint64_t val)
1082 if (vcpuid < 0 || vcpuid >= vm->maxcpus)
1085 if (reg >= VM_REG_LAST)
1088 error = vmmops_setreg(vm->cookie, vcpuid, reg, val);
1089 if (error || reg != VM_REG_GUEST_RIP)
1092 /* Set 'nextrip' to match the value of %rip */
1093 VCPU_CTR1(vm, vcpuid, "Setting nextrip to %#lx", val);
1094 vcpu = &vm->vcpu[vcpuid];
1095 vcpu->nextrip = val;
1100 is_descriptor_table(int reg)
1104 case VM_REG_GUEST_IDTR:
1105 case VM_REG_GUEST_GDTR:
1113 is_segment_register(int reg)
1117 case VM_REG_GUEST_ES:
1118 case VM_REG_GUEST_CS:
1119 case VM_REG_GUEST_SS:
1120 case VM_REG_GUEST_DS:
1121 case VM_REG_GUEST_FS:
1122 case VM_REG_GUEST_GS:
1123 case VM_REG_GUEST_TR:
1124 case VM_REG_GUEST_LDTR:
1132 vm_get_seg_desc(struct vm *vm, int vcpu, int reg,
1133 struct seg_desc *desc)
1136 if (vcpu < 0 || vcpu >= vm->maxcpus)
1139 if (!is_segment_register(reg) && !is_descriptor_table(reg))
1142 return (vmmops_getdesc(vm->cookie, vcpu, reg, desc));
1146 vm_set_seg_desc(struct vm *vm, int vcpu, int reg,
1147 struct seg_desc *desc)
1149 if (vcpu < 0 || vcpu >= vm->maxcpus)
1152 if (!is_segment_register(reg) && !is_descriptor_table(reg))
1155 return (vmmops_setdesc(vm->cookie, vcpu, reg, desc));
1159 restore_guest_fpustate(struct vcpu *vcpu)
1162 /* flush host state to the pcb */
1165 /* restore guest FPU state */
1166 fpu_stop_emulating();
1167 fpurestore(vcpu->guestfpu);
1169 /* restore guest XCR0 if XSAVE is enabled in the host */
1170 if (rcr4() & CR4_XSAVE)
1171 load_xcr(0, vcpu->guest_xcr0);
1174 * The FPU is now "dirty" with the guest's state so turn on emulation
1175 * to trap any access to the FPU by the host.
1177 fpu_start_emulating();
1181 save_guest_fpustate(struct vcpu *vcpu)
1184 if ((rcr0() & CR0_TS) == 0)
1185 panic("fpu emulation not enabled in host!");
1187 /* save guest XCR0 and restore host XCR0 */
1188 if (rcr4() & CR4_XSAVE) {
1189 vcpu->guest_xcr0 = rxcr(0);
1190 load_xcr(0, vmm_get_host_xcr0());
1193 /* save guest FPU state */
1194 fpu_stop_emulating();
1195 fpusave(vcpu->guestfpu);
1196 fpu_start_emulating();
1199 static VMM_STAT(VCPU_IDLE_TICKS, "number of ticks vcpu was idle");
1202 vcpu_set_state_locked(struct vm *vm, int vcpuid, enum vcpu_state newstate,
1208 vcpu = &vm->vcpu[vcpuid];
1209 vcpu_assert_locked(vcpu);
1212 * State transitions from the vmmdev_ioctl() must always begin from
1213 * the VCPU_IDLE state. This guarantees that there is only a single
1214 * ioctl() operating on a vcpu at any point.
1217 while (vcpu->state != VCPU_IDLE) {
1219 vcpu_notify_event_locked(vcpu, false);
1220 VCPU_CTR1(vm, vcpuid, "vcpu state change from %s to "
1221 "idle requested", vcpu_state2str(vcpu->state));
1222 msleep_spin(&vcpu->state, &vcpu->mtx, "vmstat", hz);
1225 KASSERT(vcpu->state != VCPU_IDLE, ("invalid transition from "
1226 "vcpu idle state"));
1229 if (vcpu->state == VCPU_RUNNING) {
1230 KASSERT(vcpu->hostcpu == curcpu, ("curcpu %d and hostcpu %d "
1231 "mismatch for running vcpu", curcpu, vcpu->hostcpu));
1233 KASSERT(vcpu->hostcpu == NOCPU, ("Invalid hostcpu %d for a "
1234 "vcpu that is not running", vcpu->hostcpu));
1238 * The following state transitions are allowed:
1239 * IDLE -> FROZEN -> IDLE
1240 * FROZEN -> RUNNING -> FROZEN
1241 * FROZEN -> SLEEPING -> FROZEN
1243 switch (vcpu->state) {
1247 error = (newstate != VCPU_FROZEN);
1250 error = (newstate == VCPU_FROZEN);
1260 VCPU_CTR2(vm, vcpuid, "vcpu state changed from %s to %s",
1261 vcpu_state2str(vcpu->state), vcpu_state2str(newstate));
1263 vcpu->state = newstate;
1264 if (newstate == VCPU_RUNNING)
1265 vcpu->hostcpu = curcpu;
1267 vcpu->hostcpu = NOCPU;
1269 if (newstate == VCPU_IDLE)
1270 wakeup(&vcpu->state);
1276 vcpu_require_state(struct vm *vm, int vcpuid, enum vcpu_state newstate)
1280 if ((error = vcpu_set_state(vm, vcpuid, newstate, false)) != 0)
1281 panic("Error %d setting state to %d\n", error, newstate);
1285 vcpu_require_state_locked(struct vm *vm, int vcpuid, enum vcpu_state newstate)
1289 if ((error = vcpu_set_state_locked(vm, vcpuid, newstate, false)) != 0)
1290 panic("Error %d setting state to %d", error, newstate);
1293 #define RENDEZVOUS_CTR0(vm, vcpuid, fmt) \
1296 VCPU_CTR0(vm, vcpuid, fmt); \
1302 vm_handle_rendezvous(struct vm *vm, int vcpuid)
1307 KASSERT(vcpuid == -1 || (vcpuid >= 0 && vcpuid < vm->maxcpus),
1308 ("vm_handle_rendezvous: invalid vcpuid %d", vcpuid));
1312 mtx_lock(&vm->rendezvous_mtx);
1313 while (vm->rendezvous_func != NULL) {
1314 /* 'rendezvous_req_cpus' must be a subset of 'active_cpus' */
1315 CPU_AND(&vm->rendezvous_req_cpus, &vm->rendezvous_req_cpus, &vm->active_cpus);
1318 CPU_ISSET(vcpuid, &vm->rendezvous_req_cpus) &&
1319 !CPU_ISSET(vcpuid, &vm->rendezvous_done_cpus)) {
1320 VCPU_CTR0(vm, vcpuid, "Calling rendezvous func");
1321 (*vm->rendezvous_func)(vm, vcpuid, vm->rendezvous_arg);
1322 CPU_SET(vcpuid, &vm->rendezvous_done_cpus);
1324 if (CPU_CMP(&vm->rendezvous_req_cpus,
1325 &vm->rendezvous_done_cpus) == 0) {
1326 VCPU_CTR0(vm, vcpuid, "Rendezvous completed");
1327 vm->rendezvous_func = NULL;
1328 wakeup(&vm->rendezvous_func);
1331 RENDEZVOUS_CTR0(vm, vcpuid, "Wait for rendezvous completion");
1332 mtx_sleep(&vm->rendezvous_func, &vm->rendezvous_mtx, 0,
1334 if ((td->td_flags & TDF_NEEDSUSPCHK) != 0) {
1335 mtx_unlock(&vm->rendezvous_mtx);
1336 error = thread_check_susp(td, true);
1339 mtx_lock(&vm->rendezvous_mtx);
1342 mtx_unlock(&vm->rendezvous_mtx);
1347 * Emulate a guest 'hlt' by sleeping until the vcpu is ready to run.
1350 vm_handle_hlt(struct vm *vm, int vcpuid, bool intr_disabled, bool *retu)
1355 int error, t, vcpu_halted, vm_halted;
1357 KASSERT(!CPU_ISSET(vcpuid, &vm->halted_cpus), ("vcpu already halted"));
1359 vcpu = &vm->vcpu[vcpuid];
1368 * Do a final check for pending NMI or interrupts before
1369 * really putting this thread to sleep. Also check for
1370 * software events that would cause this vcpu to wakeup.
1372 * These interrupts/events could have happened after the
1373 * vcpu returned from vmmops_run() and before it acquired the
1376 if (vm->rendezvous_func != NULL || vm->suspend || vcpu->reqidle)
1378 if (vm_nmi_pending(vm, vcpuid))
1380 if (!intr_disabled) {
1381 if (vm_extint_pending(vm, vcpuid) ||
1382 vlapic_pending_intr(vcpu->vlapic, NULL)) {
1387 /* Don't go to sleep if the vcpu thread needs to yield */
1388 if (vcpu_should_yield(vm, vcpuid))
1391 if (vcpu_debugged(vm, vcpuid))
1395 * Some Linux guests implement "halt" by having all vcpus
1396 * execute HLT with interrupts disabled. 'halted_cpus' keeps
1397 * track of the vcpus that have entered this state. When all
1398 * vcpus enter the halted state the virtual machine is halted.
1400 if (intr_disabled) {
1402 VCPU_CTR0(vm, vcpuid, "Halted");
1403 if (!vcpu_halted && halt_detection_enabled) {
1405 CPU_SET_ATOMIC(vcpuid, &vm->halted_cpus);
1407 if (CPU_CMP(&vm->halted_cpus, &vm->active_cpus) == 0) {
1416 vcpu_require_state_locked(vm, vcpuid, VCPU_SLEEPING);
1418 * XXX msleep_spin() cannot be interrupted by signals so
1419 * wake up periodically to check pending signals.
1421 msleep_spin(vcpu, &vcpu->mtx, wmesg, hz);
1422 vcpu_require_state_locked(vm, vcpuid, VCPU_FROZEN);
1423 vmm_stat_incr(vm, vcpuid, VCPU_IDLE_TICKS, ticks - t);
1424 if ((td->td_flags & TDF_NEEDSUSPCHK) != 0) {
1426 error = thread_check_susp(td, false);
1434 CPU_CLR_ATOMIC(vcpuid, &vm->halted_cpus);
1439 vm_suspend(vm, VM_SUSPEND_HALT);
1445 vm_handle_paging(struct vm *vm, int vcpuid, bool *retu)
1450 struct vm_exit *vme;
1452 vcpu = &vm->vcpu[vcpuid];
1453 vme = &vcpu->exitinfo;
1455 KASSERT(vme->inst_length == 0, ("%s: invalid inst_length %d",
1456 __func__, vme->inst_length));
1458 ftype = vme->u.paging.fault_type;
1459 KASSERT(ftype == VM_PROT_READ ||
1460 ftype == VM_PROT_WRITE || ftype == VM_PROT_EXECUTE,
1461 ("vm_handle_paging: invalid fault_type %d", ftype));
1463 if (ftype == VM_PROT_READ || ftype == VM_PROT_WRITE) {
1464 rv = pmap_emulate_accessed_dirty(vmspace_pmap(vm->vmspace),
1465 vme->u.paging.gpa, ftype);
1467 VCPU_CTR2(vm, vcpuid, "%s bit emulation for gpa %#lx",
1468 ftype == VM_PROT_READ ? "accessed" : "dirty",
1474 map = &vm->vmspace->vm_map;
1475 rv = vm_fault(map, vme->u.paging.gpa, ftype, VM_FAULT_NORMAL, NULL);
1477 VCPU_CTR3(vm, vcpuid, "vm_handle_paging rv = %d, gpa = %#lx, "
1478 "ftype = %d", rv, vme->u.paging.gpa, ftype);
1480 if (rv != KERN_SUCCESS)
1487 vm_handle_inst_emul(struct vm *vm, int vcpuid, bool *retu)
1491 struct vm_exit *vme;
1492 uint64_t gla, gpa, cs_base;
1493 struct vm_guest_paging *paging;
1494 mem_region_read_t mread;
1495 mem_region_write_t mwrite;
1496 enum vm_cpu_mode cpu_mode;
1497 int cs_d, error, fault;
1499 vcpu = &vm->vcpu[vcpuid];
1500 vme = &vcpu->exitinfo;
1502 KASSERT(vme->inst_length == 0, ("%s: invalid inst_length %d",
1503 __func__, vme->inst_length));
1505 gla = vme->u.inst_emul.gla;
1506 gpa = vme->u.inst_emul.gpa;
1507 cs_base = vme->u.inst_emul.cs_base;
1508 cs_d = vme->u.inst_emul.cs_d;
1509 vie = &vme->u.inst_emul.vie;
1510 paging = &vme->u.inst_emul.paging;
1511 cpu_mode = paging->cpu_mode;
1513 VCPU_CTR1(vm, vcpuid, "inst_emul fault accessing gpa %#lx", gpa);
1515 /* Fetch, decode and emulate the faulting instruction */
1516 if (vie->num_valid == 0) {
1517 error = vmm_fetch_instruction(vm, vcpuid, paging, vme->rip +
1518 cs_base, VIE_INST_SIZE, vie, &fault);
1521 * The instruction bytes have already been copied into 'vie'
1528 if (vmm_decode_instruction(vm, vcpuid, gla, cpu_mode, cs_d, vie) != 0) {
1529 VCPU_CTR1(vm, vcpuid, "Error decoding instruction at %#lx",
1530 vme->rip + cs_base);
1531 *retu = true; /* dump instruction bytes in userspace */
1536 * Update 'nextrip' based on the length of the emulated instruction.
1538 vme->inst_length = vie->num_processed;
1539 vcpu->nextrip += vie->num_processed;
1540 VCPU_CTR1(vm, vcpuid, "nextrip updated to %#lx after instruction "
1541 "decoding", vcpu->nextrip);
1543 /* return to userland unless this is an in-kernel emulated device */
1544 if (gpa >= DEFAULT_APIC_BASE && gpa < DEFAULT_APIC_BASE + PAGE_SIZE) {
1545 mread = lapic_mmio_read;
1546 mwrite = lapic_mmio_write;
1547 } else if (gpa >= VIOAPIC_BASE && gpa < VIOAPIC_BASE + VIOAPIC_SIZE) {
1548 mread = vioapic_mmio_read;
1549 mwrite = vioapic_mmio_write;
1550 } else if (gpa >= VHPET_BASE && gpa < VHPET_BASE + VHPET_SIZE) {
1551 mread = vhpet_mmio_read;
1552 mwrite = vhpet_mmio_write;
1558 error = vmm_emulate_instruction(vm, vcpuid, gpa, vie, paging,
1559 mread, mwrite, retu);
1565 vm_handle_suspend(struct vm *vm, int vcpuid, bool *retu)
1572 vcpu = &vm->vcpu[vcpuid];
1575 CPU_SET_ATOMIC(vcpuid, &vm->suspended_cpus);
1578 * Wait until all 'active_cpus' have suspended themselves.
1580 * Since a VM may be suspended at any time including when one or
1581 * more vcpus are doing a rendezvous we need to call the rendezvous
1582 * handler while we are waiting to prevent a deadlock.
1585 while (error == 0) {
1586 if (CPU_CMP(&vm->suspended_cpus, &vm->active_cpus) == 0) {
1587 VCPU_CTR0(vm, vcpuid, "All vcpus suspended");
1591 if (vm->rendezvous_func == NULL) {
1592 VCPU_CTR0(vm, vcpuid, "Sleeping during suspend");
1593 vcpu_require_state_locked(vm, vcpuid, VCPU_SLEEPING);
1594 msleep_spin(vcpu, &vcpu->mtx, "vmsusp", hz);
1595 vcpu_require_state_locked(vm, vcpuid, VCPU_FROZEN);
1596 if ((td->td_flags & TDF_NEEDSUSPCHK) != 0) {
1598 error = thread_check_susp(td, false);
1602 VCPU_CTR0(vm, vcpuid, "Rendezvous during suspend");
1604 error = vm_handle_rendezvous(vm, vcpuid);
1611 * Wakeup the other sleeping vcpus and return to userspace.
1613 for (i = 0; i < vm->maxcpus; i++) {
1614 if (CPU_ISSET(i, &vm->suspended_cpus)) {
1615 vcpu_notify_event(vm, i, false);
1624 vm_handle_reqidle(struct vm *vm, int vcpuid, bool *retu)
1626 struct vcpu *vcpu = &vm->vcpu[vcpuid];
1629 KASSERT(vcpu->reqidle, ("invalid vcpu reqidle %d", vcpu->reqidle));
1637 vm_suspend(struct vm *vm, enum vm_suspend_how how)
1641 if (how <= VM_SUSPEND_NONE || how >= VM_SUSPEND_LAST)
1644 if (atomic_cmpset_int(&vm->suspend, 0, how) == 0) {
1645 VM_CTR2(vm, "virtual machine already suspended %d/%d",
1650 VM_CTR1(vm, "virtual machine successfully suspended %d", how);
1653 * Notify all active vcpus that they are now suspended.
1655 for (i = 0; i < vm->maxcpus; i++) {
1656 if (CPU_ISSET(i, &vm->active_cpus))
1657 vcpu_notify_event(vm, i, false);
1664 vm_exit_suspended(struct vm *vm, int vcpuid, uint64_t rip)
1666 struct vm_exit *vmexit;
1668 KASSERT(vm->suspend > VM_SUSPEND_NONE && vm->suspend < VM_SUSPEND_LAST,
1669 ("vm_exit_suspended: invalid suspend type %d", vm->suspend));
1671 vmexit = vm_exitinfo(vm, vcpuid);
1673 vmexit->inst_length = 0;
1674 vmexit->exitcode = VM_EXITCODE_SUSPENDED;
1675 vmexit->u.suspended.how = vm->suspend;
1679 vm_exit_debug(struct vm *vm, int vcpuid, uint64_t rip)
1681 struct vm_exit *vmexit;
1683 vmexit = vm_exitinfo(vm, vcpuid);
1685 vmexit->inst_length = 0;
1686 vmexit->exitcode = VM_EXITCODE_DEBUG;
1690 vm_exit_rendezvous(struct vm *vm, int vcpuid, uint64_t rip)
1692 struct vm_exit *vmexit;
1694 KASSERT(vm->rendezvous_func != NULL, ("rendezvous not in progress"));
1696 vmexit = vm_exitinfo(vm, vcpuid);
1698 vmexit->inst_length = 0;
1699 vmexit->exitcode = VM_EXITCODE_RENDEZVOUS;
1700 vmm_stat_incr(vm, vcpuid, VMEXIT_RENDEZVOUS, 1);
1704 vm_exit_reqidle(struct vm *vm, int vcpuid, uint64_t rip)
1706 struct vm_exit *vmexit;
1708 vmexit = vm_exitinfo(vm, vcpuid);
1710 vmexit->inst_length = 0;
1711 vmexit->exitcode = VM_EXITCODE_REQIDLE;
1712 vmm_stat_incr(vm, vcpuid, VMEXIT_REQIDLE, 1);
1716 vm_exit_astpending(struct vm *vm, int vcpuid, uint64_t rip)
1718 struct vm_exit *vmexit;
1720 vmexit = vm_exitinfo(vm, vcpuid);
1722 vmexit->inst_length = 0;
1723 vmexit->exitcode = VM_EXITCODE_BOGUS;
1724 vmm_stat_incr(vm, vcpuid, VMEXIT_ASTPENDING, 1);
1728 vm_run(struct vm *vm, struct vm_run *vmrun)
1730 struct vm_eventinfo evinfo;
1735 struct vm_exit *vme;
1736 bool retu, intr_disabled;
1739 vcpuid = vmrun->cpuid;
1741 if (vcpuid < 0 || vcpuid >= vm->maxcpus)
1744 if (!CPU_ISSET(vcpuid, &vm->active_cpus))
1747 if (CPU_ISSET(vcpuid, &vm->suspended_cpus))
1750 pmap = vmspace_pmap(vm->vmspace);
1751 vcpu = &vm->vcpu[vcpuid];
1752 vme = &vcpu->exitinfo;
1753 evinfo.rptr = &vm->rendezvous_func;
1754 evinfo.sptr = &vm->suspend;
1755 evinfo.iptr = &vcpu->reqidle;
1759 KASSERT(!CPU_ISSET(curcpu, &pmap->pm_active),
1760 ("vm_run: absurd pm_active"));
1764 pcb = PCPU_GET(curpcb);
1765 set_pcb_flags(pcb, PCB_FULL_IRET);
1767 restore_guest_fpustate(vcpu);
1769 vcpu_require_state(vm, vcpuid, VCPU_RUNNING);
1770 error = vmmops_run(vm->cookie, vcpuid, vcpu->nextrip, pmap, &evinfo);
1771 vcpu_require_state(vm, vcpuid, VCPU_FROZEN);
1773 save_guest_fpustate(vcpu);
1775 vmm_stat_incr(vm, vcpuid, VCPU_TOTAL_RUNTIME, rdtsc() - tscval);
1781 vcpu->nextrip = vme->rip + vme->inst_length;
1782 switch (vme->exitcode) {
1783 case VM_EXITCODE_REQIDLE:
1784 error = vm_handle_reqidle(vm, vcpuid, &retu);
1786 case VM_EXITCODE_SUSPENDED:
1787 error = vm_handle_suspend(vm, vcpuid, &retu);
1789 case VM_EXITCODE_IOAPIC_EOI:
1790 vioapic_process_eoi(vm, vcpuid,
1791 vme->u.ioapic_eoi.vector);
1793 case VM_EXITCODE_RENDEZVOUS:
1794 error = vm_handle_rendezvous(vm, vcpuid);
1796 case VM_EXITCODE_HLT:
1797 intr_disabled = ((vme->u.hlt.rflags & PSL_I) == 0);
1798 error = vm_handle_hlt(vm, vcpuid, intr_disabled, &retu);
1800 case VM_EXITCODE_PAGING:
1801 error = vm_handle_paging(vm, vcpuid, &retu);
1803 case VM_EXITCODE_INST_EMUL:
1804 error = vm_handle_inst_emul(vm, vcpuid, &retu);
1806 case VM_EXITCODE_INOUT:
1807 case VM_EXITCODE_INOUT_STR:
1808 error = vm_handle_inout(vm, vcpuid, vme, &retu);
1810 case VM_EXITCODE_MONITOR:
1811 case VM_EXITCODE_MWAIT:
1812 case VM_EXITCODE_VMINSN:
1813 vm_inject_ud(vm, vcpuid);
1816 retu = true; /* handled in userland */
1821 if (error == 0 && retu == false)
1824 vmm_stat_incr(vm, vcpuid, VMEXIT_USERSPACE, 1);
1825 VCPU_CTR2(vm, vcpuid, "retu %d/%d", error, vme->exitcode);
1827 /* copy the exit information */
1828 bcopy(vme, &vmrun->vm_exit, sizeof(struct vm_exit));
1833 vm_restart_instruction(void *arg, int vcpuid)
1837 enum vcpu_state state;
1842 if (vcpuid < 0 || vcpuid >= vm->maxcpus)
1845 vcpu = &vm->vcpu[vcpuid];
1846 state = vcpu_get_state(vm, vcpuid, NULL);
1847 if (state == VCPU_RUNNING) {
1849 * When a vcpu is "running" the next instruction is determined
1850 * by adding 'rip' and 'inst_length' in the vcpu's 'exitinfo'.
1851 * Thus setting 'inst_length' to zero will cause the current
1852 * instruction to be restarted.
1854 vcpu->exitinfo.inst_length = 0;
1855 VCPU_CTR1(vm, vcpuid, "restarting instruction at %#lx by "
1856 "setting inst_length to zero", vcpu->exitinfo.rip);
1857 } else if (state == VCPU_FROZEN) {
1859 * When a vcpu is "frozen" it is outside the critical section
1860 * around vmmops_run() and 'nextrip' points to the next
1861 * instruction. Thus instruction restart is achieved by setting
1862 * 'nextrip' to the vcpu's %rip.
1864 error = vm_get_register(vm, vcpuid, VM_REG_GUEST_RIP, &rip);
1865 KASSERT(!error, ("%s: error %d getting rip", __func__, error));
1866 VCPU_CTR2(vm, vcpuid, "restarting instruction by updating "
1867 "nextrip from %#lx to %#lx", vcpu->nextrip, rip);
1868 vcpu->nextrip = rip;
1870 panic("%s: invalid state %d", __func__, state);
1876 vm_exit_intinfo(struct vm *vm, int vcpuid, uint64_t info)
1881 if (vcpuid < 0 || vcpuid >= vm->maxcpus)
1884 vcpu = &vm->vcpu[vcpuid];
1886 if (info & VM_INTINFO_VALID) {
1887 type = info & VM_INTINFO_TYPE;
1888 vector = info & 0xff;
1889 if (type == VM_INTINFO_NMI && vector != IDT_NMI)
1891 if (type == VM_INTINFO_HWEXCEPTION && vector >= 32)
1893 if (info & VM_INTINFO_RSVD)
1898 VCPU_CTR2(vm, vcpuid, "%s: info1(%#lx)", __func__, info);
1899 vcpu->exitintinfo = info;
1909 #define IDT_VE 20 /* Virtualization Exception (Intel specific) */
1911 static enum exc_class
1912 exception_class(uint64_t info)
1916 KASSERT(info & VM_INTINFO_VALID, ("intinfo must be valid: %#lx", info));
1917 type = info & VM_INTINFO_TYPE;
1918 vector = info & 0xff;
1920 /* Table 6-4, "Interrupt and Exception Classes", Intel SDM, Vol 3 */
1922 case VM_INTINFO_HWINTR:
1923 case VM_INTINFO_SWINTR:
1924 case VM_INTINFO_NMI:
1925 return (EXC_BENIGN);
1928 * Hardware exception.
1930 * SVM and VT-x use identical type values to represent NMI,
1931 * hardware interrupt and software interrupt.
1933 * SVM uses type '3' for all exceptions. VT-x uses type '3'
1934 * for exceptions except #BP and #OF. #BP and #OF use a type
1935 * value of '5' or '6'. Therefore we don't check for explicit
1936 * values of 'type' to classify 'intinfo' into a hardware
1945 return (EXC_PAGEFAULT);
1951 return (EXC_CONTRIBUTORY);
1953 return (EXC_BENIGN);
1958 nested_fault(struct vm *vm, int vcpuid, uint64_t info1, uint64_t info2,
1961 enum exc_class exc1, exc2;
1964 KASSERT(info1 & VM_INTINFO_VALID, ("info1 %#lx is not valid", info1));
1965 KASSERT(info2 & VM_INTINFO_VALID, ("info2 %#lx is not valid", info2));
1968 * If an exception occurs while attempting to call the double-fault
1969 * handler the processor enters shutdown mode (aka triple fault).
1971 type1 = info1 & VM_INTINFO_TYPE;
1972 vector1 = info1 & 0xff;
1973 if (type1 == VM_INTINFO_HWEXCEPTION && vector1 == IDT_DF) {
1974 VCPU_CTR2(vm, vcpuid, "triple fault: info1(%#lx), info2(%#lx)",
1976 vm_suspend(vm, VM_SUSPEND_TRIPLEFAULT);
1982 * Table 6-5 "Conditions for Generating a Double Fault", Intel SDM, Vol3
1984 exc1 = exception_class(info1);
1985 exc2 = exception_class(info2);
1986 if ((exc1 == EXC_CONTRIBUTORY && exc2 == EXC_CONTRIBUTORY) ||
1987 (exc1 == EXC_PAGEFAULT && exc2 != EXC_BENIGN)) {
1988 /* Convert nested fault into a double fault. */
1990 *retinfo |= VM_INTINFO_VALID | VM_INTINFO_HWEXCEPTION;
1991 *retinfo |= VM_INTINFO_DEL_ERRCODE;
1993 /* Handle exceptions serially */
2000 vcpu_exception_intinfo(struct vcpu *vcpu)
2004 if (vcpu->exception_pending) {
2005 info = vcpu->exc_vector & 0xff;
2006 info |= VM_INTINFO_VALID | VM_INTINFO_HWEXCEPTION;
2007 if (vcpu->exc_errcode_valid) {
2008 info |= VM_INTINFO_DEL_ERRCODE;
2009 info |= (uint64_t)vcpu->exc_errcode << 32;
2016 vm_entry_intinfo(struct vm *vm, int vcpuid, uint64_t *retinfo)
2019 uint64_t info1, info2;
2022 KASSERT(vcpuid >= 0 &&
2023 vcpuid < vm->maxcpus, ("invalid vcpu %d", vcpuid));
2025 vcpu = &vm->vcpu[vcpuid];
2027 info1 = vcpu->exitintinfo;
2028 vcpu->exitintinfo = 0;
2031 if (vcpu->exception_pending) {
2032 info2 = vcpu_exception_intinfo(vcpu);
2033 vcpu->exception_pending = 0;
2034 VCPU_CTR2(vm, vcpuid, "Exception %d delivered: %#lx",
2035 vcpu->exc_vector, info2);
2038 if ((info1 & VM_INTINFO_VALID) && (info2 & VM_INTINFO_VALID)) {
2039 valid = nested_fault(vm, vcpuid, info1, info2, retinfo);
2040 } else if (info1 & VM_INTINFO_VALID) {
2043 } else if (info2 & VM_INTINFO_VALID) {
2051 VCPU_CTR4(vm, vcpuid, "%s: info1(%#lx), info2(%#lx), "
2052 "retinfo(%#lx)", __func__, info1, info2, *retinfo);
2059 vm_get_intinfo(struct vm *vm, int vcpuid, uint64_t *info1, uint64_t *info2)
2063 if (vcpuid < 0 || vcpuid >= vm->maxcpus)
2066 vcpu = &vm->vcpu[vcpuid];
2067 *info1 = vcpu->exitintinfo;
2068 *info2 = vcpu_exception_intinfo(vcpu);
2073 vm_inject_exception(struct vm *vm, int vcpuid, int vector, int errcode_valid,
2074 uint32_t errcode, int restart_instruction)
2080 if (vcpuid < 0 || vcpuid >= vm->maxcpus)
2083 if (vector < 0 || vector >= 32)
2087 * A double fault exception should never be injected directly into
2088 * the guest. It is a derived exception that results from specific
2089 * combinations of nested faults.
2091 if (vector == IDT_DF)
2094 vcpu = &vm->vcpu[vcpuid];
2096 if (vcpu->exception_pending) {
2097 VCPU_CTR2(vm, vcpuid, "Unable to inject exception %d due to "
2098 "pending exception %d", vector, vcpu->exc_vector);
2102 if (errcode_valid) {
2104 * Exceptions don't deliver an error code in real mode.
2106 error = vm_get_register(vm, vcpuid, VM_REG_GUEST_CR0, ®val);
2107 KASSERT(!error, ("%s: error %d getting CR0", __func__, error));
2108 if (!(regval & CR0_PE))
2113 * From section 26.6.1 "Interruptibility State" in Intel SDM:
2115 * Event blocking by "STI" or "MOV SS" is cleared after guest executes
2116 * one instruction or incurs an exception.
2118 error = vm_set_register(vm, vcpuid, VM_REG_GUEST_INTR_SHADOW, 0);
2119 KASSERT(error == 0, ("%s: error %d clearing interrupt shadow",
2122 if (restart_instruction)
2123 vm_restart_instruction(vm, vcpuid);
2125 vcpu->exception_pending = 1;
2126 vcpu->exc_vector = vector;
2127 vcpu->exc_errcode = errcode;
2128 vcpu->exc_errcode_valid = errcode_valid;
2129 VCPU_CTR1(vm, vcpuid, "Exception %d pending", vector);
2134 vm_inject_fault(void *vmarg, int vcpuid, int vector, int errcode_valid,
2138 int error, restart_instruction;
2141 restart_instruction = 1;
2143 error = vm_inject_exception(vm, vcpuid, vector, errcode_valid,
2144 errcode, restart_instruction);
2145 KASSERT(error == 0, ("vm_inject_exception error %d", error));
2149 vm_inject_pf(void *vmarg, int vcpuid, int error_code, uint64_t cr2)
2155 VCPU_CTR2(vm, vcpuid, "Injecting page fault: error_code %#x, cr2 %#lx",
2158 error = vm_set_register(vm, vcpuid, VM_REG_GUEST_CR2, cr2);
2159 KASSERT(error == 0, ("vm_set_register(cr2) error %d", error));
2161 vm_inject_fault(vm, vcpuid, IDT_PF, 1, error_code);
2164 static VMM_STAT(VCPU_NMI_COUNT, "number of NMIs delivered to vcpu");
2167 vm_inject_nmi(struct vm *vm, int vcpuid)
2171 if (vcpuid < 0 || vcpuid >= vm->maxcpus)
2174 vcpu = &vm->vcpu[vcpuid];
2176 vcpu->nmi_pending = 1;
2177 vcpu_notify_event(vm, vcpuid, false);
2182 vm_nmi_pending(struct vm *vm, int vcpuid)
2186 if (vcpuid < 0 || vcpuid >= vm->maxcpus)
2187 panic("vm_nmi_pending: invalid vcpuid %d", vcpuid);
2189 vcpu = &vm->vcpu[vcpuid];
2191 return (vcpu->nmi_pending);
2195 vm_nmi_clear(struct vm *vm, int vcpuid)
2199 if (vcpuid < 0 || vcpuid >= vm->maxcpus)
2200 panic("vm_nmi_pending: invalid vcpuid %d", vcpuid);
2202 vcpu = &vm->vcpu[vcpuid];
2204 if (vcpu->nmi_pending == 0)
2205 panic("vm_nmi_clear: inconsistent nmi_pending state");
2207 vcpu->nmi_pending = 0;
2208 vmm_stat_incr(vm, vcpuid, VCPU_NMI_COUNT, 1);
2211 static VMM_STAT(VCPU_EXTINT_COUNT, "number of ExtINTs delivered to vcpu");
2214 vm_inject_extint(struct vm *vm, int vcpuid)
2218 if (vcpuid < 0 || vcpuid >= vm->maxcpus)
2221 vcpu = &vm->vcpu[vcpuid];
2223 vcpu->extint_pending = 1;
2224 vcpu_notify_event(vm, vcpuid, false);
2229 vm_extint_pending(struct vm *vm, int vcpuid)
2233 if (vcpuid < 0 || vcpuid >= vm->maxcpus)
2234 panic("vm_extint_pending: invalid vcpuid %d", vcpuid);
2236 vcpu = &vm->vcpu[vcpuid];
2238 return (vcpu->extint_pending);
2242 vm_extint_clear(struct vm *vm, int vcpuid)
2246 if (vcpuid < 0 || vcpuid >= vm->maxcpus)
2247 panic("vm_extint_pending: invalid vcpuid %d", vcpuid);
2249 vcpu = &vm->vcpu[vcpuid];
2251 if (vcpu->extint_pending == 0)
2252 panic("vm_extint_clear: inconsistent extint_pending state");
2254 vcpu->extint_pending = 0;
2255 vmm_stat_incr(vm, vcpuid, VCPU_EXTINT_COUNT, 1);
2259 vm_get_capability(struct vm *vm, int vcpu, int type, int *retval)
2261 if (vcpu < 0 || vcpu >= vm->maxcpus)
2264 if (type < 0 || type >= VM_CAP_MAX)
2267 return (vmmops_getcap(vm->cookie, vcpu, type, retval));
2271 vm_set_capability(struct vm *vm, int vcpu, int type, int val)
2273 if (vcpu < 0 || vcpu >= vm->maxcpus)
2276 if (type < 0 || type >= VM_CAP_MAX)
2279 return (vmmops_setcap(vm->cookie, vcpu, type, val));
2283 vm_lapic(struct vm *vm, int cpu)
2285 return (vm->vcpu[cpu].vlapic);
2289 vm_ioapic(struct vm *vm)
2292 return (vm->vioapic);
2296 vm_hpet(struct vm *vm)
2303 vmm_is_pptdev(int bus, int slot, int func)
2306 char *val, *cp, *cp2;
2311 * The length of an environment variable is limited to 128 bytes which
2312 * puts an upper limit on the number of passthru devices that may be
2313 * specified using a single environment variable.
2315 * Work around this by scanning multiple environment variable
2316 * names instead of a single one - yuck!
2318 const char *names[] = { "pptdevs", "pptdevs2", "pptdevs3", NULL };
2320 /* set pptdevs="1/2/3 4/5/6 7/8/9 10/11/12" */
2322 for (i = 0; names[i] != NULL && !found; i++) {
2323 cp = val = kern_getenv(names[i]);
2324 while (cp != NULL && *cp != '\0') {
2325 if ((cp2 = strchr(cp, ' ')) != NULL)
2328 n = sscanf(cp, "%d/%d/%d", &b, &s, &f);
2329 if (n == 3 && bus == b && slot == s && func == f) {
2345 vm_iommu_domain(struct vm *vm)
2352 vcpu_set_state(struct vm *vm, int vcpuid, enum vcpu_state newstate,
2358 if (vcpuid < 0 || vcpuid >= vm->maxcpus)
2359 panic("vm_set_run_state: invalid vcpuid %d", vcpuid);
2361 vcpu = &vm->vcpu[vcpuid];
2364 error = vcpu_set_state_locked(vm, vcpuid, newstate, from_idle);
2371 vcpu_get_state(struct vm *vm, int vcpuid, int *hostcpu)
2374 enum vcpu_state state;
2376 if (vcpuid < 0 || vcpuid >= vm->maxcpus)
2377 panic("vm_get_run_state: invalid vcpuid %d", vcpuid);
2379 vcpu = &vm->vcpu[vcpuid];
2382 state = vcpu->state;
2383 if (hostcpu != NULL)
2384 *hostcpu = vcpu->hostcpu;
2391 vm_activate_cpu(struct vm *vm, int vcpuid)
2394 if (vcpuid < 0 || vcpuid >= vm->maxcpus)
2397 if (CPU_ISSET(vcpuid, &vm->active_cpus))
2400 VCPU_CTR0(vm, vcpuid, "activated");
2401 CPU_SET_ATOMIC(vcpuid, &vm->active_cpus);
2406 vm_suspend_cpu(struct vm *vm, int vcpuid)
2410 if (vcpuid < -1 || vcpuid >= vm->maxcpus)
2414 vm->debug_cpus = vm->active_cpus;
2415 for (i = 0; i < vm->maxcpus; i++) {
2416 if (CPU_ISSET(i, &vm->active_cpus))
2417 vcpu_notify_event(vm, i, false);
2420 if (!CPU_ISSET(vcpuid, &vm->active_cpus))
2423 CPU_SET_ATOMIC(vcpuid, &vm->debug_cpus);
2424 vcpu_notify_event(vm, vcpuid, false);
2430 vm_resume_cpu(struct vm *vm, int vcpuid)
2433 if (vcpuid < -1 || vcpuid >= vm->maxcpus)
2437 CPU_ZERO(&vm->debug_cpus);
2439 if (!CPU_ISSET(vcpuid, &vm->debug_cpus))
2442 CPU_CLR_ATOMIC(vcpuid, &vm->debug_cpus);
2448 vcpu_debugged(struct vm *vm, int vcpuid)
2451 return (CPU_ISSET(vcpuid, &vm->debug_cpus));
2455 vm_active_cpus(struct vm *vm)
2458 return (vm->active_cpus);
2462 vm_debug_cpus(struct vm *vm)
2465 return (vm->debug_cpus);
2469 vm_suspended_cpus(struct vm *vm)
2472 return (vm->suspended_cpus);
2476 vcpu_stats(struct vm *vm, int vcpuid)
2479 return (vm->vcpu[vcpuid].stats);
2483 vm_get_x2apic_state(struct vm *vm, int vcpuid, enum x2apic_state *state)
2485 if (vcpuid < 0 || vcpuid >= vm->maxcpus)
2488 *state = vm->vcpu[vcpuid].x2apic_state;
2494 vm_set_x2apic_state(struct vm *vm, int vcpuid, enum x2apic_state state)
2496 if (vcpuid < 0 || vcpuid >= vm->maxcpus)
2499 if (state >= X2APIC_STATE_LAST)
2502 vm->vcpu[vcpuid].x2apic_state = state;
2504 vlapic_set_x2apic_state(vm, vcpuid, state);
2510 * This function is called to ensure that a vcpu "sees" a pending event
2511 * as soon as possible:
2512 * - If the vcpu thread is sleeping then it is woken up.
2513 * - If the vcpu is running on a different host_cpu then an IPI will be directed
2514 * to the host_cpu to cause the vcpu to trap into the hypervisor.
2517 vcpu_notify_event_locked(struct vcpu *vcpu, bool lapic_intr)
2521 hostcpu = vcpu->hostcpu;
2522 if (vcpu->state == VCPU_RUNNING) {
2523 KASSERT(hostcpu != NOCPU, ("vcpu running on invalid hostcpu"));
2524 if (hostcpu != curcpu) {
2526 vlapic_post_intr(vcpu->vlapic, hostcpu,
2529 ipi_cpu(hostcpu, vmm_ipinum);
2533 * If the 'vcpu' is running on 'curcpu' then it must
2534 * be sending a notification to itself (e.g. SELF_IPI).
2535 * The pending event will be picked up when the vcpu
2536 * transitions back to guest context.
2540 KASSERT(hostcpu == NOCPU, ("vcpu state %d not consistent "
2541 "with hostcpu %d", vcpu->state, hostcpu));
2542 if (vcpu->state == VCPU_SLEEPING)
2548 vcpu_notify_event(struct vm *vm, int vcpuid, bool lapic_intr)
2550 struct vcpu *vcpu = &vm->vcpu[vcpuid];
2553 vcpu_notify_event_locked(vcpu, lapic_intr);
2558 vm_get_vmspace(struct vm *vm)
2561 return (vm->vmspace);
2565 vm_apicid2vcpuid(struct vm *vm, int apicid)
2568 * XXX apic id is assumed to be numerically identical to vcpu id
2574 vm_smp_rendezvous(struct vm *vm, int vcpuid, cpuset_t dest,
2575 vm_rendezvous_func_t func, void *arg)
2580 * Enforce that this function is called without any locks
2582 WITNESS_WARN(WARN_PANIC, NULL, "vm_smp_rendezvous");
2583 KASSERT(vcpuid == -1 || (vcpuid >= 0 && vcpuid < vm->maxcpus),
2584 ("vm_smp_rendezvous: invalid vcpuid %d", vcpuid));
2587 mtx_lock(&vm->rendezvous_mtx);
2588 if (vm->rendezvous_func != NULL) {
2590 * If a rendezvous is already in progress then we need to
2591 * call the rendezvous handler in case this 'vcpuid' is one
2592 * of the targets of the rendezvous.
2594 RENDEZVOUS_CTR0(vm, vcpuid, "Rendezvous already in progress");
2595 mtx_unlock(&vm->rendezvous_mtx);
2596 error = vm_handle_rendezvous(vm, vcpuid);
2601 KASSERT(vm->rendezvous_func == NULL, ("vm_smp_rendezvous: previous "
2602 "rendezvous is still in progress"));
2604 RENDEZVOUS_CTR0(vm, vcpuid, "Initiating rendezvous");
2605 vm->rendezvous_req_cpus = dest;
2606 CPU_ZERO(&vm->rendezvous_done_cpus);
2607 vm->rendezvous_arg = arg;
2608 vm->rendezvous_func = func;
2609 mtx_unlock(&vm->rendezvous_mtx);
2612 * Wake up any sleeping vcpus and trigger a VM-exit in any running
2613 * vcpus so they handle the rendezvous as soon as possible.
2615 for (i = 0; i < vm->maxcpus; i++) {
2616 if (CPU_ISSET(i, &dest))
2617 vcpu_notify_event(vm, i, false);
2620 return (vm_handle_rendezvous(vm, vcpuid));
2624 vm_atpic(struct vm *vm)
2626 return (vm->vatpic);
2630 vm_atpit(struct vm *vm)
2632 return (vm->vatpit);
2636 vm_pmtmr(struct vm *vm)
2639 return (vm->vpmtmr);
2643 vm_rtc(struct vm *vm)
2650 vm_segment_name(int seg)
2652 static enum vm_reg_name seg_names[] = {
2661 KASSERT(seg >= 0 && seg < nitems(seg_names),
2662 ("%s: invalid segment encoding %d", __func__, seg));
2663 return (seg_names[seg]);
2667 vm_copy_teardown(struct vm *vm, int vcpuid, struct vm_copyinfo *copyinfo,
2672 for (idx = 0; idx < num_copyinfo; idx++) {
2673 if (copyinfo[idx].cookie != NULL)
2674 vm_gpa_release(copyinfo[idx].cookie);
2676 bzero(copyinfo, num_copyinfo * sizeof(struct vm_copyinfo));
2680 vm_copy_setup(struct vm *vm, int vcpuid, struct vm_guest_paging *paging,
2681 uint64_t gla, size_t len, int prot, struct vm_copyinfo *copyinfo,
2682 int num_copyinfo, int *fault)
2684 int error, idx, nused;
2685 size_t n, off, remaining;
2689 bzero(copyinfo, sizeof(struct vm_copyinfo) * num_copyinfo);
2693 while (remaining > 0) {
2694 KASSERT(nused < num_copyinfo, ("insufficient vm_copyinfo"));
2695 error = vm_gla2gpa(vm, vcpuid, paging, gla, prot, &gpa, fault);
2696 if (error || *fault)
2698 off = gpa & PAGE_MASK;
2699 n = min(remaining, PAGE_SIZE - off);
2700 copyinfo[nused].gpa = gpa;
2701 copyinfo[nused].len = n;
2707 for (idx = 0; idx < nused; idx++) {
2708 hva = vm_gpa_hold(vm, vcpuid, copyinfo[idx].gpa,
2709 copyinfo[idx].len, prot, &cookie);
2712 copyinfo[idx].hva = hva;
2713 copyinfo[idx].cookie = cookie;
2717 vm_copy_teardown(vm, vcpuid, copyinfo, num_copyinfo);
2726 vm_copyin(struct vm *vm, int vcpuid, struct vm_copyinfo *copyinfo, void *kaddr,
2735 bcopy(copyinfo[idx].hva, dst, copyinfo[idx].len);
2736 len -= copyinfo[idx].len;
2737 dst += copyinfo[idx].len;
2743 vm_copyout(struct vm *vm, int vcpuid, const void *kaddr,
2744 struct vm_copyinfo *copyinfo, size_t len)
2752 bcopy(src, copyinfo[idx].hva, copyinfo[idx].len);
2753 len -= copyinfo[idx].len;
2754 src += copyinfo[idx].len;
2760 * Return the amount of in-use and wired memory for the VM. Since
2761 * these are global stats, only return the values with for vCPU 0
2763 VMM_STAT_DECLARE(VMM_MEM_RESIDENT);
2764 VMM_STAT_DECLARE(VMM_MEM_WIRED);
2767 vm_get_rescnt(struct vm *vm, int vcpu, struct vmm_stat_type *stat)
2771 vmm_stat_set(vm, vcpu, VMM_MEM_RESIDENT,
2772 PAGE_SIZE * vmspace_resident_count(vm->vmspace));
2777 vm_get_wiredcnt(struct vm *vm, int vcpu, struct vmm_stat_type *stat)
2781 vmm_stat_set(vm, vcpu, VMM_MEM_WIRED,
2782 PAGE_SIZE * pmap_wired_count(vmspace_pmap(vm->vmspace)));
2786 VMM_STAT_FUNC(VMM_MEM_RESIDENT, "Resident memory", vm_get_rescnt);
2787 VMM_STAT_FUNC(VMM_MEM_WIRED, "Wired memory", vm_get_wiredcnt);
2789 #ifdef BHYVE_SNAPSHOT
2791 vm_snapshot_vcpus(struct vm *vm, struct vm_snapshot_meta *meta)
2797 for (i = 0; i < VM_MAXCPU; i++) {
2798 vcpu = &vm->vcpu[i];
2800 SNAPSHOT_VAR_OR_LEAVE(vcpu->x2apic_state, meta, ret, done);
2801 SNAPSHOT_VAR_OR_LEAVE(vcpu->exitintinfo, meta, ret, done);
2802 SNAPSHOT_VAR_OR_LEAVE(vcpu->exc_vector, meta, ret, done);
2803 SNAPSHOT_VAR_OR_LEAVE(vcpu->exc_errcode_valid, meta, ret, done);
2804 SNAPSHOT_VAR_OR_LEAVE(vcpu->exc_errcode, meta, ret, done);
2805 SNAPSHOT_VAR_OR_LEAVE(vcpu->guest_xcr0, meta, ret, done);
2806 SNAPSHOT_VAR_OR_LEAVE(vcpu->exitinfo, meta, ret, done);
2807 SNAPSHOT_VAR_OR_LEAVE(vcpu->nextrip, meta, ret, done);
2808 /* XXX we're cheating here, since the value of tsc_offset as
2809 * saved here is actually the value of the guest's TSC value.
2811 * It will be turned turned back into an actual offset when the
2812 * TSC restore function is called
2814 SNAPSHOT_VAR_OR_LEAVE(vcpu->tsc_offset, meta, ret, done);
2822 vm_snapshot_vm(struct vm *vm, struct vm_snapshot_meta *meta)
2831 if (meta->op == VM_SNAPSHOT_SAVE) {
2832 /* XXX make tsc_offset take the value TSC proper as seen by the
2835 for (i = 0; i < VM_MAXCPU; i++)
2836 vm->vcpu[i].tsc_offset += now;
2839 ret = vm_snapshot_vcpus(vm, meta);
2841 printf("%s: failed to copy vm data to user buffer", __func__);
2845 if (meta->op == VM_SNAPSHOT_SAVE) {
2846 /* XXX turn tsc_offset back into an offset; actual value is only
2847 * required for restore; using it otherwise would be wrong
2849 for (i = 0; i < VM_MAXCPU; i++)
2850 vm->vcpu[i].tsc_offset -= now;
2858 vm_snapshot_vmcx(struct vm *vm, struct vm_snapshot_meta *meta)
2864 for (i = 0; i < VM_MAXCPU; i++) {
2865 error = vmmops_vmcx_snapshot(vm->cookie, meta, i);
2867 printf("%s: failed to snapshot vmcs/vmcb data for "
2868 "vCPU: %d; error: %d\n", __func__, i, error);
2878 * Save kernel-side structures to user-space for snapshotting.
2881 vm_snapshot_req(struct vm *vm, struct vm_snapshot_meta *meta)
2885 switch (meta->dev_req) {
2887 ret = vmmops_snapshot(vm->cookie, meta);
2890 ret = vm_snapshot_vmcx(vm, meta);
2893 ret = vm_snapshot_vm(vm, meta);
2895 case STRUCT_VIOAPIC:
2896 ret = vioapic_snapshot(vm_ioapic(vm), meta);
2899 ret = vlapic_snapshot(vm, meta);
2902 ret = vhpet_snapshot(vm_hpet(vm), meta);
2905 ret = vatpic_snapshot(vm_atpic(vm), meta);
2908 ret = vatpit_snapshot(vm_atpit(vm), meta);
2911 ret = vpmtmr_snapshot(vm_pmtmr(vm), meta);
2914 ret = vrtc_snapshot(vm_rtc(vm), meta);
2917 printf("%s: failed to find the requested type %#x\n",
2918 __func__, meta->dev_req);
2925 vm_set_tsc_offset(struct vm *vm, int vcpuid, uint64_t offset)
2929 if (vcpuid < 0 || vcpuid >= VM_MAXCPU)
2932 vcpu = &vm->vcpu[vcpuid];
2933 vcpu->tsc_offset = offset;
2939 vm_restore_time(struct vm *vm)
2947 error = vhpet_restore_time(vm_hpet(vm));
2951 for (i = 0; i < nitems(vm->vcpu); i++) {
2952 vcpu = &vm->vcpu[i];
2954 error = vmmops_restore_tsc(vm->cookie, i, vcpu->tsc_offset -