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 struct vlapic *vlapic; /* (i) APIC device model */
99 enum x2apic_state x2apic_state; /* (i) APIC mode */
100 uint64_t exitintinfo; /* (i) events pending at VM exit */
101 int nmi_pending; /* (i) NMI pending */
102 int extint_pending; /* (i) INTR pending */
103 struct vm_exception exception; /* (x) exception collateral */
104 int exception_pending; /* (i) exception pending */
105 struct savefpu *guestfpu; /* (a,i) guest fpu state */
106 uint64_t guest_xcr0; /* (i) guest %xcr0 register */
107 void *stats; /* (a,i) statistics */
108 uint64_t guest_msrs[VMM_MSR_NUM]; /* (i) emulated MSRs */
109 struct vm_exit exitinfo; /* (x) exit reason and collateral */
112 #define vcpu_lock_initialized(v) mtx_initialized(&((v)->mtx))
113 #define vcpu_lock_init(v) mtx_init(&((v)->mtx), "vcpu lock", 0, MTX_SPIN)
114 #define vcpu_lock(v) mtx_lock_spin(&((v)->mtx))
115 #define vcpu_unlock(v) mtx_unlock_spin(&((v)->mtx))
116 #define vcpu_assert_locked(v) mtx_assert(&((v)->mtx), MA_OWNED)
124 #define VM_MAX_MEMORY_SEGMENTS 2
128 * (o) initialized the first time the VM is created
129 * (i) initialized when VM is created and when it is reinitialized
130 * (x) initialized before use
133 void *cookie; /* (i) cpu-specific data */
134 void *iommu; /* (x) iommu-specific data */
135 struct vhpet *vhpet; /* (i) virtual HPET */
136 struct vioapic *vioapic; /* (i) virtual ioapic */
137 struct vatpic *vatpic; /* (i) virtual atpic */
138 struct vatpit *vatpit; /* (i) virtual atpit */
139 volatile cpuset_t active_cpus; /* (i) active vcpus */
140 int suspend; /* (i) stop VM execution */
141 volatile cpuset_t suspended_cpus; /* (i) suspended vcpus */
142 volatile cpuset_t halted_cpus; /* (x) cpus in a hard halt */
143 cpuset_t rendezvous_req_cpus; /* (x) rendezvous requested */
144 cpuset_t rendezvous_done_cpus; /* (x) rendezvous finished */
145 void *rendezvous_arg; /* (x) rendezvous func/arg */
146 vm_rendezvous_func_t rendezvous_func;
147 struct mtx rendezvous_mtx; /* (o) rendezvous lock */
148 int num_mem_segs; /* (o) guest memory segments */
149 struct mem_seg mem_segs[VM_MAX_MEMORY_SEGMENTS];
150 struct vmspace *vmspace; /* (o) guest's address space */
151 char name[VM_MAX_NAMELEN]; /* (o) virtual machine name */
152 struct vcpu vcpu[VM_MAXCPU]; /* (i) guest vcpus */
155 static int vmm_initialized;
157 static struct vmm_ops *ops;
158 #define VMM_INIT(num) (ops != NULL ? (*ops->init)(num) : 0)
159 #define VMM_CLEANUP() (ops != NULL ? (*ops->cleanup)() : 0)
160 #define VMM_RESUME() (ops != NULL ? (*ops->resume)() : 0)
162 #define VMINIT(vm, pmap) (ops != NULL ? (*ops->vminit)(vm, pmap): NULL)
163 #define VMRUN(vmi, vcpu, rip, pmap, rptr, sptr) \
164 (ops != NULL ? (*ops->vmrun)(vmi, vcpu, rip, pmap, rptr, sptr) : ENXIO)
165 #define VMCLEANUP(vmi) (ops != NULL ? (*ops->vmcleanup)(vmi) : NULL)
166 #define VMSPACE_ALLOC(min, max) \
167 (ops != NULL ? (*ops->vmspace_alloc)(min, max) : NULL)
168 #define VMSPACE_FREE(vmspace) \
169 (ops != NULL ? (*ops->vmspace_free)(vmspace) : ENXIO)
170 #define VMGETREG(vmi, vcpu, num, retval) \
171 (ops != NULL ? (*ops->vmgetreg)(vmi, vcpu, num, retval) : ENXIO)
172 #define VMSETREG(vmi, vcpu, num, val) \
173 (ops != NULL ? (*ops->vmsetreg)(vmi, vcpu, num, val) : ENXIO)
174 #define VMGETDESC(vmi, vcpu, num, desc) \
175 (ops != NULL ? (*ops->vmgetdesc)(vmi, vcpu, num, desc) : ENXIO)
176 #define VMSETDESC(vmi, vcpu, num, desc) \
177 (ops != NULL ? (*ops->vmsetdesc)(vmi, vcpu, num, desc) : ENXIO)
178 #define VMGETCAP(vmi, vcpu, num, retval) \
179 (ops != NULL ? (*ops->vmgetcap)(vmi, vcpu, num, retval) : ENXIO)
180 #define VMSETCAP(vmi, vcpu, num, val) \
181 (ops != NULL ? (*ops->vmsetcap)(vmi, vcpu, num, val) : ENXIO)
182 #define VLAPIC_INIT(vmi, vcpu) \
183 (ops != NULL ? (*ops->vlapic_init)(vmi, vcpu) : NULL)
184 #define VLAPIC_CLEANUP(vmi, vlapic) \
185 (ops != NULL ? (*ops->vlapic_cleanup)(vmi, vlapic) : NULL)
187 #define fpu_start_emulating() load_cr0(rcr0() | CR0_TS)
188 #define fpu_stop_emulating() clts()
190 static MALLOC_DEFINE(M_VM, "vm", "vm");
191 CTASSERT(VMM_MSR_NUM <= 64); /* msr_mask can keep track of up to 64 msrs */
194 static VMM_STAT(VCPU_TOTAL_RUNTIME, "vcpu total runtime");
196 SYSCTL_NODE(_hw, OID_AUTO, vmm, CTLFLAG_RW, NULL, NULL);
199 * Halt the guest if all vcpus are executing a HLT instruction with
200 * interrupts disabled.
202 static int halt_detection_enabled = 1;
203 SYSCTL_INT(_hw_vmm, OID_AUTO, halt_detection, CTLFLAG_RDTUN,
204 &halt_detection_enabled, 0,
205 "Halt VM if all vcpus execute HLT with interrupts disabled");
207 static int vmm_ipinum;
208 SYSCTL_INT(_hw_vmm, OID_AUTO, ipinum, CTLFLAG_RD, &vmm_ipinum, 0,
209 "IPI vector used for vcpu notifications");
212 vcpu_cleanup(struct vm *vm, int i, bool destroy)
214 struct vcpu *vcpu = &vm->vcpu[i];
216 VLAPIC_CLEANUP(vm->cookie, vcpu->vlapic);
218 vmm_stat_free(vcpu->stats);
219 fpu_save_area_free(vcpu->guestfpu);
224 vcpu_init(struct vm *vm, int vcpu_id, bool create)
228 KASSERT(vcpu_id >= 0 && vcpu_id < VM_MAXCPU,
229 ("vcpu_init: invalid vcpu %d", vcpu_id));
231 vcpu = &vm->vcpu[vcpu_id];
234 KASSERT(!vcpu_lock_initialized(vcpu), ("vcpu %d already "
235 "initialized", vcpu_id));
236 vcpu_lock_init(vcpu);
237 vcpu->state = VCPU_IDLE;
238 vcpu->hostcpu = NOCPU;
239 vcpu->guestfpu = fpu_save_area_alloc();
240 vcpu->stats = vmm_stat_alloc();
243 vcpu->vlapic = VLAPIC_INIT(vm->cookie, vcpu_id);
244 vm_set_x2apic_state(vm, vcpu_id, X2APIC_DISABLED);
245 vcpu->exitintinfo = 0;
246 vcpu->nmi_pending = 0;
247 vcpu->extint_pending = 0;
248 vcpu->exception_pending = 0;
249 vcpu->guest_xcr0 = XFEATURE_ENABLED_X87;
250 fpu_save_area_reset(vcpu->guestfpu);
251 vmm_stat_init(vcpu->stats);
252 guest_msrs_init(vm, vcpu_id);
256 vm_exitinfo(struct vm *vm, int cpuid)
260 if (cpuid < 0 || cpuid >= VM_MAXCPU)
261 panic("vm_exitinfo: invalid cpuid %d", cpuid);
263 vcpu = &vm->vcpu[cpuid];
265 return (&vcpu->exitinfo);
279 vmm_host_state_init();
281 vmm_ipinum = vmm_ipi_alloc();
283 vmm_ipinum = IPI_AST;
285 error = vmm_mem_init();
290 ops = &vmm_ops_intel;
291 else if (vmm_is_amd())
297 vmm_resume_p = vmm_resume;
299 return (VMM_INIT(vmm_ipinum));
303 vmm_handler(module_t mod, int what, void *arg)
310 if (ppt_avail_devices() > 0)
317 error = vmmdev_cleanup();
321 if (vmm_ipinum != IPI_AST)
322 vmm_ipi_free(vmm_ipinum);
323 error = VMM_CLEANUP();
325 * Something bad happened - prevent new
326 * VMs from being created
339 static moduledata_t vmm_kmod = {
346 * vmm initialization has the following dependencies:
348 * - iommu initialization must happen after the pci passthru driver has had
349 * a chance to attach to any passthru devices (after SI_SUB_CONFIGURE).
351 * - VT-x initialization requires smp_rendezvous() and therefore must happen
352 * after SMP is fully functional (after SI_SUB_SMP).
354 DECLARE_MODULE(vmm, vmm_kmod, SI_SUB_SMP + 1, SI_ORDER_ANY);
355 MODULE_VERSION(vmm, 1);
358 vm_init(struct vm *vm, bool create)
362 vm->cookie = VMINIT(vm, vmspace_pmap(vm->vmspace));
364 vm->vioapic = vioapic_init(vm);
365 vm->vhpet = vhpet_init(vm);
366 vm->vatpic = vatpic_init(vm);
367 vm->vatpit = vatpit_init(vm);
369 CPU_ZERO(&vm->active_cpus);
372 CPU_ZERO(&vm->suspended_cpus);
374 for (i = 0; i < VM_MAXCPU; i++)
375 vcpu_init(vm, i, create);
379 vm_create(const char *name, struct vm **retvm)
382 struct vmspace *vmspace;
385 * If vmm.ko could not be successfully initialized then don't attempt
386 * to create the virtual machine.
388 if (!vmm_initialized)
391 if (name == NULL || strlen(name) >= VM_MAX_NAMELEN)
394 vmspace = VMSPACE_ALLOC(VM_MIN_ADDRESS, VM_MAXUSER_ADDRESS);
398 vm = malloc(sizeof(struct vm), M_VM, M_WAITOK | M_ZERO);
399 strcpy(vm->name, name);
400 vm->num_mem_segs = 0;
401 vm->vmspace = vmspace;
402 mtx_init(&vm->rendezvous_mtx, "vm rendezvous lock", 0, MTX_DEF);
411 vm_free_mem_seg(struct vm *vm, struct mem_seg *seg)
414 if (seg->object != NULL)
415 vmm_mem_free(vm->vmspace, seg->gpa, seg->len);
417 bzero(seg, sizeof(*seg));
421 vm_cleanup(struct vm *vm, bool destroy)
425 ppt_unassign_all(vm);
427 if (vm->iommu != NULL)
428 iommu_destroy_domain(vm->iommu);
430 vatpit_cleanup(vm->vatpit);
431 vhpet_cleanup(vm->vhpet);
432 vatpic_cleanup(vm->vatpic);
433 vioapic_cleanup(vm->vioapic);
435 for (i = 0; i < VM_MAXCPU; i++)
436 vcpu_cleanup(vm, i, destroy);
438 VMCLEANUP(vm->cookie);
441 for (i = 0; i < vm->num_mem_segs; i++)
442 vm_free_mem_seg(vm, &vm->mem_segs[i]);
444 vm->num_mem_segs = 0;
446 VMSPACE_FREE(vm->vmspace);
452 vm_destroy(struct vm *vm)
454 vm_cleanup(vm, true);
459 vm_reinit(struct vm *vm)
464 * A virtual machine can be reset only if all vcpus are suspended.
466 if (CPU_CMP(&vm->suspended_cpus, &vm->active_cpus) == 0) {
467 vm_cleanup(vm, false);
478 vm_name(struct vm *vm)
484 vm_map_mmio(struct vm *vm, vm_paddr_t gpa, size_t len, vm_paddr_t hpa)
488 if ((obj = vmm_mmio_alloc(vm->vmspace, gpa, len, hpa)) == NULL)
495 vm_unmap_mmio(struct vm *vm, vm_paddr_t gpa, size_t len)
498 vmm_mmio_free(vm->vmspace, gpa, len);
503 vm_mem_allocated(struct vm *vm, vm_paddr_t gpa)
506 vm_paddr_t gpabase, gpalimit;
508 for (i = 0; i < vm->num_mem_segs; i++) {
509 gpabase = vm->mem_segs[i].gpa;
510 gpalimit = gpabase + vm->mem_segs[i].len;
511 if (gpa >= gpabase && gpa < gpalimit)
512 return (TRUE); /* 'gpa' is regular memory */
515 if (ppt_is_mmio(vm, gpa))
516 return (TRUE); /* 'gpa' is pci passthru mmio */
522 vm_malloc(struct vm *vm, vm_paddr_t gpa, size_t len)
524 int available, allocated;
529 if ((gpa & PAGE_MASK) || (len & PAGE_MASK) || len == 0)
532 available = allocated = 0;
534 while (g < gpa + len) {
535 if (vm_mem_allocated(vm, g))
544 * If there are some allocated and some available pages in the address
545 * range then it is an error.
547 if (allocated && available)
551 * If the entire address range being requested has already been
552 * allocated then there isn't anything more to do.
554 if (allocated && available == 0)
557 if (vm->num_mem_segs >= VM_MAX_MEMORY_SEGMENTS)
560 seg = &vm->mem_segs[vm->num_mem_segs];
562 if ((object = vmm_mem_alloc(vm->vmspace, gpa, len)) == NULL)
567 seg->object = object;
576 vm_gpa_unwire(struct vm *vm)
581 for (i = 0; i < vm->num_mem_segs; i++) {
582 seg = &vm->mem_segs[i];
586 rv = vm_map_unwire(&vm->vmspace->vm_map,
587 seg->gpa, seg->gpa + seg->len,
588 VM_MAP_WIRE_USER | VM_MAP_WIRE_NOHOLES);
589 KASSERT(rv == KERN_SUCCESS, ("vm(%s) memory segment "
590 "%#lx/%ld could not be unwired: %d",
591 vm_name(vm), seg->gpa, seg->len, rv));
598 vm_gpa_wire(struct vm *vm)
603 for (i = 0; i < vm->num_mem_segs; i++) {
604 seg = &vm->mem_segs[i];
609 rv = vm_map_wire(&vm->vmspace->vm_map,
610 seg->gpa, seg->gpa + seg->len,
611 VM_MAP_WIRE_USER | VM_MAP_WIRE_NOHOLES);
612 if (rv != KERN_SUCCESS)
618 if (i < vm->num_mem_segs) {
620 * Undo the wiring before returning an error.
630 vm_iommu_modify(struct vm *vm, boolean_t map)
635 void *vp, *cookie, *host_domain;
638 host_domain = iommu_host_domain();
640 for (i = 0; i < vm->num_mem_segs; i++) {
641 seg = &vm->mem_segs[i];
642 KASSERT(seg->wired, ("vm(%s) memory segment %#lx/%ld not wired",
643 vm_name(vm), seg->gpa, seg->len));
646 while (gpa < seg->gpa + seg->len) {
647 vp = vm_gpa_hold(vm, gpa, PAGE_SIZE, VM_PROT_WRITE,
649 KASSERT(vp != NULL, ("vm(%s) could not map gpa %#lx",
652 vm_gpa_release(cookie);
654 hpa = DMAP_TO_PHYS((uintptr_t)vp);
656 iommu_create_mapping(vm->iommu, gpa, hpa, sz);
657 iommu_remove_mapping(host_domain, hpa, sz);
659 iommu_remove_mapping(vm->iommu, gpa, sz);
660 iommu_create_mapping(host_domain, hpa, hpa, sz);
668 * Invalidate the cached translations associated with the domain
669 * from which pages were removed.
672 iommu_invalidate_tlb(host_domain);
674 iommu_invalidate_tlb(vm->iommu);
677 #define vm_iommu_unmap(vm) vm_iommu_modify((vm), FALSE)
678 #define vm_iommu_map(vm) vm_iommu_modify((vm), TRUE)
681 vm_unassign_pptdev(struct vm *vm, int bus, int slot, int func)
685 error = ppt_unassign_device(vm, bus, slot, func);
689 if (ppt_assigned_devices(vm) == 0) {
697 vm_assign_pptdev(struct vm *vm, int bus, int slot, int func)
703 * Virtual machines with pci passthru devices get special treatment:
704 * - the guest physical memory is wired
705 * - the iommu is programmed to do the 'gpa' to 'hpa' translation
707 * We need to do this before the first pci passthru device is attached.
709 if (ppt_assigned_devices(vm) == 0) {
710 KASSERT(vm->iommu == NULL,
711 ("vm_assign_pptdev: iommu must be NULL"));
712 maxaddr = vmm_mem_maxaddr();
713 vm->iommu = iommu_create_domain(maxaddr);
715 error = vm_gpa_wire(vm);
722 error = ppt_assign_device(vm, bus, slot, func);
727 vm_gpa_hold(struct vm *vm, vm_paddr_t gpa, size_t len, int reqprot,
733 pageoff = gpa & PAGE_MASK;
734 if (len > PAGE_SIZE - pageoff)
735 panic("vm_gpa_hold: invalid gpa/len: 0x%016lx/%lu", gpa, len);
737 count = vm_fault_quick_hold_pages(&vm->vmspace->vm_map,
738 trunc_page(gpa), PAGE_SIZE, reqprot, &m, 1);
742 return ((void *)(PHYS_TO_DMAP(VM_PAGE_TO_PHYS(m)) + pageoff));
750 vm_gpa_release(void *cookie)
752 vm_page_t m = cookie;
760 vm_gpabase2memseg(struct vm *vm, vm_paddr_t gpabase,
761 struct vm_memory_segment *seg)
765 for (i = 0; i < vm->num_mem_segs; i++) {
766 if (gpabase == vm->mem_segs[i].gpa) {
767 seg->gpa = vm->mem_segs[i].gpa;
768 seg->len = vm->mem_segs[i].len;
769 seg->wired = vm->mem_segs[i].wired;
777 vm_get_memobj(struct vm *vm, vm_paddr_t gpa, size_t len,
778 vm_offset_t *offset, struct vm_object **object)
785 for (i = 0; i < vm->num_mem_segs; i++) {
786 if ((seg_obj = vm->mem_segs[i].object) == NULL)
789 seg_gpa = vm->mem_segs[i].gpa;
790 seg_len = vm->mem_segs[i].len;
792 if (gpa >= seg_gpa && gpa < seg_gpa + seg_len) {
793 *offset = gpa - seg_gpa;
795 vm_object_reference(seg_obj);
804 vm_get_register(struct vm *vm, int vcpu, int reg, uint64_t *retval)
807 if (vcpu < 0 || vcpu >= VM_MAXCPU)
810 if (reg >= VM_REG_LAST)
813 return (VMGETREG(vm->cookie, vcpu, reg, retval));
817 vm_set_register(struct vm *vm, int vcpu, int reg, uint64_t val)
820 if (vcpu < 0 || vcpu >= VM_MAXCPU)
823 if (reg >= VM_REG_LAST)
826 return (VMSETREG(vm->cookie, vcpu, reg, val));
830 is_descriptor_table(int reg)
834 case VM_REG_GUEST_IDTR:
835 case VM_REG_GUEST_GDTR:
843 is_segment_register(int reg)
847 case VM_REG_GUEST_ES:
848 case VM_REG_GUEST_CS:
849 case VM_REG_GUEST_SS:
850 case VM_REG_GUEST_DS:
851 case VM_REG_GUEST_FS:
852 case VM_REG_GUEST_GS:
853 case VM_REG_GUEST_TR:
854 case VM_REG_GUEST_LDTR:
862 vm_get_seg_desc(struct vm *vm, int vcpu, int reg,
863 struct seg_desc *desc)
866 if (vcpu < 0 || vcpu >= VM_MAXCPU)
869 if (!is_segment_register(reg) && !is_descriptor_table(reg))
872 return (VMGETDESC(vm->cookie, vcpu, reg, desc));
876 vm_set_seg_desc(struct vm *vm, int vcpu, int reg,
877 struct seg_desc *desc)
879 if (vcpu < 0 || vcpu >= VM_MAXCPU)
882 if (!is_segment_register(reg) && !is_descriptor_table(reg))
885 return (VMSETDESC(vm->cookie, vcpu, reg, desc));
889 restore_guest_fpustate(struct vcpu *vcpu)
892 /* flush host state to the pcb */
895 /* restore guest FPU state */
896 fpu_stop_emulating();
897 fpurestore(vcpu->guestfpu);
899 /* restore guest XCR0 if XSAVE is enabled in the host */
900 if (rcr4() & CR4_XSAVE)
901 load_xcr(0, vcpu->guest_xcr0);
904 * The FPU is now "dirty" with the guest's state so turn on emulation
905 * to trap any access to the FPU by the host.
907 fpu_start_emulating();
911 save_guest_fpustate(struct vcpu *vcpu)
914 if ((rcr0() & CR0_TS) == 0)
915 panic("fpu emulation not enabled in host!");
917 /* save guest XCR0 and restore host XCR0 */
918 if (rcr4() & CR4_XSAVE) {
919 vcpu->guest_xcr0 = rxcr(0);
920 load_xcr(0, vmm_get_host_xcr0());
923 /* save guest FPU state */
924 fpu_stop_emulating();
925 fpusave(vcpu->guestfpu);
926 fpu_start_emulating();
929 static VMM_STAT(VCPU_IDLE_TICKS, "number of ticks vcpu was idle");
932 vcpu_set_state_locked(struct vcpu *vcpu, enum vcpu_state newstate,
937 vcpu_assert_locked(vcpu);
940 * State transitions from the vmmdev_ioctl() must always begin from
941 * the VCPU_IDLE state. This guarantees that there is only a single
942 * ioctl() operating on a vcpu at any point.
945 while (vcpu->state != VCPU_IDLE)
946 msleep_spin(&vcpu->state, &vcpu->mtx, "vmstat", hz);
948 KASSERT(vcpu->state != VCPU_IDLE, ("invalid transition from "
952 if (vcpu->state == VCPU_RUNNING) {
953 KASSERT(vcpu->hostcpu == curcpu, ("curcpu %d and hostcpu %d "
954 "mismatch for running vcpu", curcpu, vcpu->hostcpu));
956 KASSERT(vcpu->hostcpu == NOCPU, ("Invalid hostcpu %d for a "
957 "vcpu that is not running", vcpu->hostcpu));
961 * The following state transitions are allowed:
962 * IDLE -> FROZEN -> IDLE
963 * FROZEN -> RUNNING -> FROZEN
964 * FROZEN -> SLEEPING -> FROZEN
966 switch (vcpu->state) {
970 error = (newstate != VCPU_FROZEN);
973 error = (newstate == VCPU_FROZEN);
983 vcpu->state = newstate;
984 if (newstate == VCPU_RUNNING)
985 vcpu->hostcpu = curcpu;
987 vcpu->hostcpu = NOCPU;
989 if (newstate == VCPU_IDLE)
990 wakeup(&vcpu->state);
996 vcpu_require_state(struct vm *vm, int vcpuid, enum vcpu_state newstate)
1000 if ((error = vcpu_set_state(vm, vcpuid, newstate, false)) != 0)
1001 panic("Error %d setting state to %d\n", error, newstate);
1005 vcpu_require_state_locked(struct vcpu *vcpu, enum vcpu_state newstate)
1009 if ((error = vcpu_set_state_locked(vcpu, newstate, false)) != 0)
1010 panic("Error %d setting state to %d", error, newstate);
1014 vm_set_rendezvous_func(struct vm *vm, vm_rendezvous_func_t func)
1017 KASSERT(mtx_owned(&vm->rendezvous_mtx), ("rendezvous_mtx not locked"));
1020 * Update 'rendezvous_func' and execute a write memory barrier to
1021 * ensure that it is visible across all host cpus. This is not needed
1022 * for correctness but it does ensure that all the vcpus will notice
1023 * that the rendezvous is requested immediately.
1025 vm->rendezvous_func = func;
1029 #define RENDEZVOUS_CTR0(vm, vcpuid, fmt) \
1032 VCPU_CTR0(vm, vcpuid, fmt); \
1038 vm_handle_rendezvous(struct vm *vm, int vcpuid)
1041 KASSERT(vcpuid == -1 || (vcpuid >= 0 && vcpuid < VM_MAXCPU),
1042 ("vm_handle_rendezvous: invalid vcpuid %d", vcpuid));
1044 mtx_lock(&vm->rendezvous_mtx);
1045 while (vm->rendezvous_func != NULL) {
1046 /* 'rendezvous_req_cpus' must be a subset of 'active_cpus' */
1047 CPU_AND(&vm->rendezvous_req_cpus, &vm->active_cpus);
1050 CPU_ISSET(vcpuid, &vm->rendezvous_req_cpus) &&
1051 !CPU_ISSET(vcpuid, &vm->rendezvous_done_cpus)) {
1052 VCPU_CTR0(vm, vcpuid, "Calling rendezvous func");
1053 (*vm->rendezvous_func)(vm, vcpuid, vm->rendezvous_arg);
1054 CPU_SET(vcpuid, &vm->rendezvous_done_cpus);
1056 if (CPU_CMP(&vm->rendezvous_req_cpus,
1057 &vm->rendezvous_done_cpus) == 0) {
1058 VCPU_CTR0(vm, vcpuid, "Rendezvous completed");
1059 vm_set_rendezvous_func(vm, NULL);
1060 wakeup(&vm->rendezvous_func);
1063 RENDEZVOUS_CTR0(vm, vcpuid, "Wait for rendezvous completion");
1064 mtx_sleep(&vm->rendezvous_func, &vm->rendezvous_mtx, 0,
1067 mtx_unlock(&vm->rendezvous_mtx);
1071 * Emulate a guest 'hlt' by sleeping until the vcpu is ready to run.
1074 vm_handle_hlt(struct vm *vm, int vcpuid, bool intr_disabled, bool *retu)
1078 int t, vcpu_halted, vm_halted;
1080 KASSERT(!CPU_ISSET(vcpuid, &vm->halted_cpus), ("vcpu already halted"));
1082 vcpu = &vm->vcpu[vcpuid];
1089 * Do a final check for pending NMI or interrupts before
1090 * really putting this thread to sleep. Also check for
1091 * software events that would cause this vcpu to wakeup.
1093 * These interrupts/events could have happened after the
1094 * vcpu returned from VMRUN() and before it acquired the
1097 if (vm->rendezvous_func != NULL || vm->suspend)
1099 if (vm_nmi_pending(vm, vcpuid))
1101 if (!intr_disabled) {
1102 if (vm_extint_pending(vm, vcpuid) ||
1103 vlapic_pending_intr(vcpu->vlapic, NULL)) {
1109 * Some Linux guests implement "halt" by having all vcpus
1110 * execute HLT with interrupts disabled. 'halted_cpus' keeps
1111 * track of the vcpus that have entered this state. When all
1112 * vcpus enter the halted state the virtual machine is halted.
1114 if (intr_disabled) {
1116 VCPU_CTR0(vm, vcpuid, "Halted");
1117 if (!vcpu_halted && halt_detection_enabled) {
1119 CPU_SET_ATOMIC(vcpuid, &vm->halted_cpus);
1121 if (CPU_CMP(&vm->halted_cpus, &vm->active_cpus) == 0) {
1130 vcpu_require_state_locked(vcpu, VCPU_SLEEPING);
1131 msleep_spin(vcpu, &vcpu->mtx, wmesg, 0);
1132 vcpu_require_state_locked(vcpu, VCPU_FROZEN);
1133 vmm_stat_incr(vm, vcpuid, VCPU_IDLE_TICKS, ticks - t);
1137 CPU_CLR_ATOMIC(vcpuid, &vm->halted_cpus);
1142 vm_suspend(vm, VM_SUSPEND_HALT);
1148 vm_handle_paging(struct vm *vm, int vcpuid, bool *retu)
1153 struct vm_exit *vme;
1155 vcpu = &vm->vcpu[vcpuid];
1156 vme = &vcpu->exitinfo;
1158 ftype = vme->u.paging.fault_type;
1159 KASSERT(ftype == VM_PROT_READ ||
1160 ftype == VM_PROT_WRITE || ftype == VM_PROT_EXECUTE,
1161 ("vm_handle_paging: invalid fault_type %d", ftype));
1163 if (ftype == VM_PROT_READ || ftype == VM_PROT_WRITE) {
1164 rv = pmap_emulate_accessed_dirty(vmspace_pmap(vm->vmspace),
1165 vme->u.paging.gpa, ftype);
1170 map = &vm->vmspace->vm_map;
1171 rv = vm_fault(map, vme->u.paging.gpa, ftype, VM_FAULT_NORMAL);
1173 VCPU_CTR3(vm, vcpuid, "vm_handle_paging rv = %d, gpa = %#lx, "
1174 "ftype = %d", rv, vme->u.paging.gpa, ftype);
1176 if (rv != KERN_SUCCESS)
1179 /* restart execution at the faulting instruction */
1180 vme->inst_length = 0;
1186 vm_handle_inst_emul(struct vm *vm, int vcpuid, bool *retu)
1190 struct vm_exit *vme;
1192 struct vm_guest_paging *paging;
1193 mem_region_read_t mread;
1194 mem_region_write_t mwrite;
1195 enum vm_cpu_mode cpu_mode;
1198 vcpu = &vm->vcpu[vcpuid];
1199 vme = &vcpu->exitinfo;
1201 gla = vme->u.inst_emul.gla;
1202 gpa = vme->u.inst_emul.gpa;
1203 cs_d = vme->u.inst_emul.cs_d;
1204 vie = &vme->u.inst_emul.vie;
1205 paging = &vme->u.inst_emul.paging;
1206 cpu_mode = paging->cpu_mode;
1210 /* Fetch, decode and emulate the faulting instruction */
1211 error = vmm_fetch_instruction(vm, vcpuid, paging, vme->rip,
1212 vme->inst_length, vie);
1214 return (0); /* Resume guest to handle page fault */
1215 else if (error == -1)
1217 else if (error != 0)
1218 panic("%s: vmm_fetch_instruction error %d", __func__, error);
1220 if (vmm_decode_instruction(vm, vcpuid, gla, cpu_mode, cs_d, vie) != 0)
1223 /* return to userland unless this is an in-kernel emulated device */
1224 if (gpa >= DEFAULT_APIC_BASE && gpa < DEFAULT_APIC_BASE + PAGE_SIZE) {
1225 mread = lapic_mmio_read;
1226 mwrite = lapic_mmio_write;
1227 } else if (gpa >= VIOAPIC_BASE && gpa < VIOAPIC_BASE + VIOAPIC_SIZE) {
1228 mread = vioapic_mmio_read;
1229 mwrite = vioapic_mmio_write;
1230 } else if (gpa >= VHPET_BASE && gpa < VHPET_BASE + VHPET_SIZE) {
1231 mread = vhpet_mmio_read;
1232 mwrite = vhpet_mmio_write;
1238 error = vmm_emulate_instruction(vm, vcpuid, gpa, vie, mread, mwrite,
1245 vm_handle_suspend(struct vm *vm, int vcpuid, bool *retu)
1251 vcpu = &vm->vcpu[vcpuid];
1253 CPU_SET_ATOMIC(vcpuid, &vm->suspended_cpus);
1256 * Wait until all 'active_cpus' have suspended themselves.
1258 * Since a VM may be suspended at any time including when one or
1259 * more vcpus are doing a rendezvous we need to call the rendezvous
1260 * handler while we are waiting to prevent a deadlock.
1264 if (CPU_CMP(&vm->suspended_cpus, &vm->active_cpus) == 0) {
1265 VCPU_CTR0(vm, vcpuid, "All vcpus suspended");
1269 if (vm->rendezvous_func == NULL) {
1270 VCPU_CTR0(vm, vcpuid, "Sleeping during suspend");
1271 vcpu_require_state_locked(vcpu, VCPU_SLEEPING);
1272 msleep_spin(vcpu, &vcpu->mtx, "vmsusp", hz);
1273 vcpu_require_state_locked(vcpu, VCPU_FROZEN);
1275 VCPU_CTR0(vm, vcpuid, "Rendezvous during suspend");
1277 vm_handle_rendezvous(vm, vcpuid);
1284 * Wakeup the other sleeping vcpus and return to userspace.
1286 for (i = 0; i < VM_MAXCPU; i++) {
1287 if (CPU_ISSET(i, &vm->suspended_cpus)) {
1288 vcpu_notify_event(vm, i, false);
1297 vm_suspend(struct vm *vm, enum vm_suspend_how how)
1301 if (how <= VM_SUSPEND_NONE || how >= VM_SUSPEND_LAST)
1304 if (atomic_cmpset_int(&vm->suspend, 0, how) == 0) {
1305 VM_CTR2(vm, "virtual machine already suspended %d/%d",
1310 VM_CTR1(vm, "virtual machine successfully suspended %d", how);
1313 * Notify all active vcpus that they are now suspended.
1315 for (i = 0; i < VM_MAXCPU; i++) {
1316 if (CPU_ISSET(i, &vm->active_cpus))
1317 vcpu_notify_event(vm, i, false);
1324 vm_exit_suspended(struct vm *vm, int vcpuid, uint64_t rip)
1326 struct vm_exit *vmexit;
1328 KASSERT(vm->suspend > VM_SUSPEND_NONE && vm->suspend < VM_SUSPEND_LAST,
1329 ("vm_exit_suspended: invalid suspend type %d", vm->suspend));
1331 vmexit = vm_exitinfo(vm, vcpuid);
1333 vmexit->inst_length = 0;
1334 vmexit->exitcode = VM_EXITCODE_SUSPENDED;
1335 vmexit->u.suspended.how = vm->suspend;
1339 vm_exit_rendezvous(struct vm *vm, int vcpuid, uint64_t rip)
1341 struct vm_exit *vmexit;
1343 KASSERT(vm->rendezvous_func != NULL, ("rendezvous not in progress"));
1345 vmexit = vm_exitinfo(vm, vcpuid);
1347 vmexit->inst_length = 0;
1348 vmexit->exitcode = VM_EXITCODE_RENDEZVOUS;
1349 vmm_stat_incr(vm, vcpuid, VMEXIT_RENDEZVOUS, 1);
1353 vm_exit_astpending(struct vm *vm, int vcpuid, uint64_t rip)
1355 struct vm_exit *vmexit;
1357 vmexit = vm_exitinfo(vm, vcpuid);
1359 vmexit->inst_length = 0;
1360 vmexit->exitcode = VM_EXITCODE_BOGUS;
1361 vmm_stat_incr(vm, vcpuid, VMEXIT_ASTPENDING, 1);
1365 vm_run(struct vm *vm, struct vm_run *vmrun)
1370 uint64_t tscval, rip;
1371 struct vm_exit *vme;
1372 bool retu, intr_disabled;
1376 vcpuid = vmrun->cpuid;
1378 if (vcpuid < 0 || vcpuid >= VM_MAXCPU)
1381 if (!CPU_ISSET(vcpuid, &vm->active_cpus))
1384 if (CPU_ISSET(vcpuid, &vm->suspended_cpus))
1387 rptr = &vm->rendezvous_func;
1388 sptr = &vm->suspend;
1389 pmap = vmspace_pmap(vm->vmspace);
1390 vcpu = &vm->vcpu[vcpuid];
1391 vme = &vcpu->exitinfo;
1396 KASSERT(!CPU_ISSET(curcpu, &pmap->pm_active),
1397 ("vm_run: absurd pm_active"));
1401 pcb = PCPU_GET(curpcb);
1402 set_pcb_flags(pcb, PCB_FULL_IRET);
1404 restore_guest_msrs(vm, vcpuid);
1405 restore_guest_fpustate(vcpu);
1407 vcpu_require_state(vm, vcpuid, VCPU_RUNNING);
1408 error = VMRUN(vm->cookie, vcpuid, rip, pmap, rptr, sptr);
1409 vcpu_require_state(vm, vcpuid, VCPU_FROZEN);
1411 save_guest_fpustate(vcpu);
1412 restore_host_msrs(vm, vcpuid);
1414 vmm_stat_incr(vm, vcpuid, VCPU_TOTAL_RUNTIME, rdtsc() - tscval);
1420 switch (vme->exitcode) {
1421 case VM_EXITCODE_SUSPENDED:
1422 error = vm_handle_suspend(vm, vcpuid, &retu);
1424 case VM_EXITCODE_IOAPIC_EOI:
1425 vioapic_process_eoi(vm, vcpuid,
1426 vme->u.ioapic_eoi.vector);
1428 case VM_EXITCODE_RENDEZVOUS:
1429 vm_handle_rendezvous(vm, vcpuid);
1432 case VM_EXITCODE_HLT:
1433 intr_disabled = ((vme->u.hlt.rflags & PSL_I) == 0);
1434 error = vm_handle_hlt(vm, vcpuid, intr_disabled, &retu);
1436 case VM_EXITCODE_PAGING:
1437 error = vm_handle_paging(vm, vcpuid, &retu);
1439 case VM_EXITCODE_INST_EMUL:
1440 error = vm_handle_inst_emul(vm, vcpuid, &retu);
1442 case VM_EXITCODE_INOUT:
1443 case VM_EXITCODE_INOUT_STR:
1444 error = vm_handle_inout(vm, vcpuid, vme, &retu);
1447 retu = true; /* handled in userland */
1452 if (error == 0 && retu == false) {
1453 rip = vme->rip + vme->inst_length;
1457 /* copy the exit information */
1458 bcopy(vme, &vmrun->vm_exit, sizeof(struct vm_exit));
1463 vm_exit_intinfo(struct vm *vm, int vcpuid, uint64_t info)
1468 if (vcpuid < 0 || vcpuid >= VM_MAXCPU)
1471 vcpu = &vm->vcpu[vcpuid];
1473 if (info & VM_INTINFO_VALID) {
1474 type = info & VM_INTINFO_TYPE;
1475 vector = info & 0xff;
1476 if (type == VM_INTINFO_NMI && vector != IDT_NMI)
1478 if (type == VM_INTINFO_HWEXCEPTION && vector >= 32)
1480 if (info & VM_INTINFO_RSVD)
1485 VCPU_CTR2(vm, vcpuid, "%s: info1(%#lx)", __func__, info);
1486 vcpu->exitintinfo = info;
1496 #define IDT_VE 20 /* Virtualization Exception (Intel specific) */
1498 static enum exc_class
1499 exception_class(uint64_t info)
1503 KASSERT(info & VM_INTINFO_VALID, ("intinfo must be valid: %#lx", info));
1504 type = info & VM_INTINFO_TYPE;
1505 vector = info & 0xff;
1507 /* Table 6-4, "Interrupt and Exception Classes", Intel SDM, Vol 3 */
1509 case VM_INTINFO_HWINTR:
1510 case VM_INTINFO_SWINTR:
1511 case VM_INTINFO_NMI:
1512 return (EXC_BENIGN);
1515 * Hardware exception.
1517 * SVM and VT-x use identical type values to represent NMI,
1518 * hardware interrupt and software interrupt.
1520 * SVM uses type '3' for all exceptions. VT-x uses type '3'
1521 * for exceptions except #BP and #OF. #BP and #OF use a type
1522 * value of '5' or '6'. Therefore we don't check for explicit
1523 * values of 'type' to classify 'intinfo' into a hardware
1532 return (EXC_PAGEFAULT);
1538 return (EXC_CONTRIBUTORY);
1540 return (EXC_BENIGN);
1545 nested_fault(struct vm *vm, int vcpuid, uint64_t info1, uint64_t info2,
1548 enum exc_class exc1, exc2;
1551 KASSERT(info1 & VM_INTINFO_VALID, ("info1 %#lx is not valid", info1));
1552 KASSERT(info2 & VM_INTINFO_VALID, ("info2 %#lx is not valid", info2));
1555 * If an exception occurs while attempting to call the double-fault
1556 * handler the processor enters shutdown mode (aka triple fault).
1558 type1 = info1 & VM_INTINFO_TYPE;
1559 vector1 = info1 & 0xff;
1560 if (type1 == VM_INTINFO_HWEXCEPTION && vector1 == IDT_DF) {
1561 VCPU_CTR2(vm, vcpuid, "triple fault: info1(%#lx), info2(%#lx)",
1563 vm_suspend(vm, VM_SUSPEND_TRIPLEFAULT);
1569 * Table 6-5 "Conditions for Generating a Double Fault", Intel SDM, Vol3
1571 exc1 = exception_class(info1);
1572 exc2 = exception_class(info2);
1573 if ((exc1 == EXC_CONTRIBUTORY && exc2 == EXC_CONTRIBUTORY) ||
1574 (exc1 == EXC_PAGEFAULT && exc2 != EXC_BENIGN)) {
1575 /* Convert nested fault into a double fault. */
1577 *retinfo |= VM_INTINFO_VALID | VM_INTINFO_HWEXCEPTION;
1578 *retinfo |= VM_INTINFO_DEL_ERRCODE;
1580 /* Handle exceptions serially */
1587 vcpu_exception_intinfo(struct vcpu *vcpu)
1591 if (vcpu->exception_pending) {
1592 info = vcpu->exception.vector & 0xff;
1593 info |= VM_INTINFO_VALID | VM_INTINFO_HWEXCEPTION;
1594 if (vcpu->exception.error_code_valid) {
1595 info |= VM_INTINFO_DEL_ERRCODE;
1596 info |= (uint64_t)vcpu->exception.error_code << 32;
1603 vm_entry_intinfo(struct vm *vm, int vcpuid, uint64_t *retinfo)
1606 uint64_t info1, info2;
1609 KASSERT(vcpuid >= 0 && vcpuid < VM_MAXCPU, ("invalid vcpu %d", vcpuid));
1611 vcpu = &vm->vcpu[vcpuid];
1613 info1 = vcpu->exitintinfo;
1614 vcpu->exitintinfo = 0;
1617 if (vcpu->exception_pending) {
1618 info2 = vcpu_exception_intinfo(vcpu);
1619 vcpu->exception_pending = 0;
1620 VCPU_CTR2(vm, vcpuid, "Exception %d delivered: %#lx",
1621 vcpu->exception.vector, info2);
1624 if ((info1 & VM_INTINFO_VALID) && (info2 & VM_INTINFO_VALID)) {
1625 valid = nested_fault(vm, vcpuid, info1, info2, retinfo);
1626 } else if (info1 & VM_INTINFO_VALID) {
1629 } else if (info2 & VM_INTINFO_VALID) {
1637 VCPU_CTR4(vm, vcpuid, "%s: info1(%#lx), info2(%#lx), "
1638 "retinfo(%#lx)", __func__, info1, info2, *retinfo);
1645 vm_get_intinfo(struct vm *vm, int vcpuid, uint64_t *info1, uint64_t *info2)
1649 if (vcpuid < 0 || vcpuid >= VM_MAXCPU)
1652 vcpu = &vm->vcpu[vcpuid];
1653 *info1 = vcpu->exitintinfo;
1654 *info2 = vcpu_exception_intinfo(vcpu);
1659 vm_inject_exception(struct vm *vm, int vcpuid, struct vm_exception *exception)
1663 if (vcpuid < 0 || vcpuid >= VM_MAXCPU)
1666 if (exception->vector < 0 || exception->vector >= 32)
1670 * A double fault exception should never be injected directly into
1671 * the guest. It is a derived exception that results from specific
1672 * combinations of nested faults.
1674 if (exception->vector == IDT_DF)
1677 vcpu = &vm->vcpu[vcpuid];
1679 if (vcpu->exception_pending) {
1680 VCPU_CTR2(vm, vcpuid, "Unable to inject exception %d due to "
1681 "pending exception %d", exception->vector,
1682 vcpu->exception.vector);
1686 vcpu->exception_pending = 1;
1687 vcpu->exception = *exception;
1688 VCPU_CTR1(vm, vcpuid, "Exception %d pending", exception->vector);
1693 vm_inject_fault(struct vm *vm, int vcpuid, struct vm_exception *exception)
1695 struct vm_exit *vmexit;
1698 error = vm_inject_exception(vm, vcpuid, exception);
1699 KASSERT(error == 0, ("vm_inject_exception error %d", error));
1702 * A fault-like exception allows the instruction to be restarted
1703 * after the exception handler returns.
1705 * By setting the inst_length to 0 we ensure that the instruction
1706 * pointer remains at the faulting instruction.
1708 vmexit = vm_exitinfo(vm, vcpuid);
1709 vmexit->inst_length = 0;
1713 vm_inject_pf(struct vm *vm, int vcpuid, int error_code, uint64_t cr2)
1715 struct vm_exception pf = {
1717 .error_code_valid = 1,
1718 .error_code = error_code
1722 VCPU_CTR2(vm, vcpuid, "Injecting page fault: error_code %#x, cr2 %#lx",
1725 error = vm_set_register(vm, vcpuid, VM_REG_GUEST_CR2, cr2);
1726 KASSERT(error == 0, ("vm_set_register(cr2) error %d", error));
1728 vm_inject_fault(vm, vcpuid, &pf);
1732 vm_inject_gp(struct vm *vm, int vcpuid)
1734 struct vm_exception gpf = {
1736 .error_code_valid = 1,
1740 vm_inject_fault(vm, vcpuid, &gpf);
1744 vm_inject_ud(struct vm *vm, int vcpuid)
1746 struct vm_exception udf = {
1748 .error_code_valid = 0
1751 vm_inject_fault(vm, vcpuid, &udf);
1754 static VMM_STAT(VCPU_NMI_COUNT, "number of NMIs delivered to vcpu");
1757 vm_inject_nmi(struct vm *vm, int vcpuid)
1761 if (vcpuid < 0 || vcpuid >= VM_MAXCPU)
1764 vcpu = &vm->vcpu[vcpuid];
1766 vcpu->nmi_pending = 1;
1767 vcpu_notify_event(vm, vcpuid, false);
1772 vm_nmi_pending(struct vm *vm, int vcpuid)
1776 if (vcpuid < 0 || vcpuid >= VM_MAXCPU)
1777 panic("vm_nmi_pending: invalid vcpuid %d", vcpuid);
1779 vcpu = &vm->vcpu[vcpuid];
1781 return (vcpu->nmi_pending);
1785 vm_nmi_clear(struct vm *vm, int vcpuid)
1789 if (vcpuid < 0 || vcpuid >= VM_MAXCPU)
1790 panic("vm_nmi_pending: invalid vcpuid %d", vcpuid);
1792 vcpu = &vm->vcpu[vcpuid];
1794 if (vcpu->nmi_pending == 0)
1795 panic("vm_nmi_clear: inconsistent nmi_pending state");
1797 vcpu->nmi_pending = 0;
1798 vmm_stat_incr(vm, vcpuid, VCPU_NMI_COUNT, 1);
1801 static VMM_STAT(VCPU_EXTINT_COUNT, "number of ExtINTs delivered to vcpu");
1804 vm_inject_extint(struct vm *vm, int vcpuid)
1808 if (vcpuid < 0 || vcpuid >= VM_MAXCPU)
1811 vcpu = &vm->vcpu[vcpuid];
1813 vcpu->extint_pending = 1;
1814 vcpu_notify_event(vm, vcpuid, false);
1819 vm_extint_pending(struct vm *vm, int vcpuid)
1823 if (vcpuid < 0 || vcpuid >= VM_MAXCPU)
1824 panic("vm_extint_pending: invalid vcpuid %d", vcpuid);
1826 vcpu = &vm->vcpu[vcpuid];
1828 return (vcpu->extint_pending);
1832 vm_extint_clear(struct vm *vm, int vcpuid)
1836 if (vcpuid < 0 || vcpuid >= VM_MAXCPU)
1837 panic("vm_extint_pending: invalid vcpuid %d", vcpuid);
1839 vcpu = &vm->vcpu[vcpuid];
1841 if (vcpu->extint_pending == 0)
1842 panic("vm_extint_clear: inconsistent extint_pending state");
1844 vcpu->extint_pending = 0;
1845 vmm_stat_incr(vm, vcpuid, VCPU_EXTINT_COUNT, 1);
1849 vm_get_capability(struct vm *vm, int vcpu, int type, int *retval)
1851 if (vcpu < 0 || vcpu >= VM_MAXCPU)
1854 if (type < 0 || type >= VM_CAP_MAX)
1857 return (VMGETCAP(vm->cookie, vcpu, type, retval));
1861 vm_set_capability(struct vm *vm, int vcpu, int type, int val)
1863 if (vcpu < 0 || vcpu >= VM_MAXCPU)
1866 if (type < 0 || type >= VM_CAP_MAX)
1869 return (VMSETCAP(vm->cookie, vcpu, type, val));
1873 vm_guest_msrs(struct vm *vm, int cpu)
1875 return (vm->vcpu[cpu].guest_msrs);
1879 vm_lapic(struct vm *vm, int cpu)
1881 return (vm->vcpu[cpu].vlapic);
1885 vm_ioapic(struct vm *vm)
1888 return (vm->vioapic);
1892 vm_hpet(struct vm *vm)
1899 vmm_is_pptdev(int bus, int slot, int func)
1903 char *val, *cp, *cp2;
1907 * The length of an environment variable is limited to 128 bytes which
1908 * puts an upper limit on the number of passthru devices that may be
1909 * specified using a single environment variable.
1911 * Work around this by scanning multiple environment variable
1912 * names instead of a single one - yuck!
1914 const char *names[] = { "pptdevs", "pptdevs2", "pptdevs3", NULL };
1916 /* set pptdevs="1/2/3 4/5/6 7/8/9 10/11/12" */
1918 for (i = 0; names[i] != NULL && !found; i++) {
1919 cp = val = getenv(names[i]);
1920 while (cp != NULL && *cp != '\0') {
1921 if ((cp2 = strchr(cp, ' ')) != NULL)
1924 n = sscanf(cp, "%d/%d/%d", &b, &s, &f);
1925 if (n == 3 && bus == b && slot == s && func == f) {
1941 vm_iommu_domain(struct vm *vm)
1948 vcpu_set_state(struct vm *vm, int vcpuid, enum vcpu_state newstate,
1954 if (vcpuid < 0 || vcpuid >= VM_MAXCPU)
1955 panic("vm_set_run_state: invalid vcpuid %d", vcpuid);
1957 vcpu = &vm->vcpu[vcpuid];
1960 error = vcpu_set_state_locked(vcpu, newstate, from_idle);
1967 vcpu_get_state(struct vm *vm, int vcpuid, int *hostcpu)
1970 enum vcpu_state state;
1972 if (vcpuid < 0 || vcpuid >= VM_MAXCPU)
1973 panic("vm_get_run_state: invalid vcpuid %d", vcpuid);
1975 vcpu = &vm->vcpu[vcpuid];
1978 state = vcpu->state;
1979 if (hostcpu != NULL)
1980 *hostcpu = vcpu->hostcpu;
1987 vm_activate_cpu(struct vm *vm, int vcpuid)
1990 if (vcpuid < 0 || vcpuid >= VM_MAXCPU)
1993 if (CPU_ISSET(vcpuid, &vm->active_cpus))
1996 VCPU_CTR0(vm, vcpuid, "activated");
1997 CPU_SET_ATOMIC(vcpuid, &vm->active_cpus);
2002 vm_active_cpus(struct vm *vm)
2005 return (vm->active_cpus);
2009 vm_suspended_cpus(struct vm *vm)
2012 return (vm->suspended_cpus);
2016 vcpu_stats(struct vm *vm, int vcpuid)
2019 return (vm->vcpu[vcpuid].stats);
2023 vm_get_x2apic_state(struct vm *vm, int vcpuid, enum x2apic_state *state)
2025 if (vcpuid < 0 || vcpuid >= VM_MAXCPU)
2028 *state = vm->vcpu[vcpuid].x2apic_state;
2034 vm_set_x2apic_state(struct vm *vm, int vcpuid, enum x2apic_state state)
2036 if (vcpuid < 0 || vcpuid >= VM_MAXCPU)
2039 if (state >= X2APIC_STATE_LAST)
2042 vm->vcpu[vcpuid].x2apic_state = state;
2044 vlapic_set_x2apic_state(vm, vcpuid, state);
2050 * This function is called to ensure that a vcpu "sees" a pending event
2051 * as soon as possible:
2052 * - If the vcpu thread is sleeping then it is woken up.
2053 * - If the vcpu is running on a different host_cpu then an IPI will be directed
2054 * to the host_cpu to cause the vcpu to trap into the hypervisor.
2057 vcpu_notify_event(struct vm *vm, int vcpuid, bool lapic_intr)
2062 vcpu = &vm->vcpu[vcpuid];
2065 hostcpu = vcpu->hostcpu;
2066 if (vcpu->state == VCPU_RUNNING) {
2067 KASSERT(hostcpu != NOCPU, ("vcpu running on invalid hostcpu"));
2068 if (hostcpu != curcpu) {
2070 vlapic_post_intr(vcpu->vlapic, hostcpu,
2073 ipi_cpu(hostcpu, vmm_ipinum);
2077 * If the 'vcpu' is running on 'curcpu' then it must
2078 * be sending a notification to itself (e.g. SELF_IPI).
2079 * The pending event will be picked up when the vcpu
2080 * transitions back to guest context.
2084 KASSERT(hostcpu == NOCPU, ("vcpu state %d not consistent "
2085 "with hostcpu %d", vcpu->state, hostcpu));
2086 if (vcpu->state == VCPU_SLEEPING)
2093 vm_get_vmspace(struct vm *vm)
2096 return (vm->vmspace);
2100 vm_apicid2vcpuid(struct vm *vm, int apicid)
2103 * XXX apic id is assumed to be numerically identical to vcpu id
2109 vm_smp_rendezvous(struct vm *vm, int vcpuid, cpuset_t dest,
2110 vm_rendezvous_func_t func, void *arg)
2115 * Enforce that this function is called without any locks
2117 WITNESS_WARN(WARN_PANIC, NULL, "vm_smp_rendezvous");
2118 KASSERT(vcpuid == -1 || (vcpuid >= 0 && vcpuid < VM_MAXCPU),
2119 ("vm_smp_rendezvous: invalid vcpuid %d", vcpuid));
2122 mtx_lock(&vm->rendezvous_mtx);
2123 if (vm->rendezvous_func != NULL) {
2125 * If a rendezvous is already in progress then we need to
2126 * call the rendezvous handler in case this 'vcpuid' is one
2127 * of the targets of the rendezvous.
2129 RENDEZVOUS_CTR0(vm, vcpuid, "Rendezvous already in progress");
2130 mtx_unlock(&vm->rendezvous_mtx);
2131 vm_handle_rendezvous(vm, vcpuid);
2134 KASSERT(vm->rendezvous_func == NULL, ("vm_smp_rendezvous: previous "
2135 "rendezvous is still in progress"));
2137 RENDEZVOUS_CTR0(vm, vcpuid, "Initiating rendezvous");
2138 vm->rendezvous_req_cpus = dest;
2139 CPU_ZERO(&vm->rendezvous_done_cpus);
2140 vm->rendezvous_arg = arg;
2141 vm_set_rendezvous_func(vm, func);
2142 mtx_unlock(&vm->rendezvous_mtx);
2145 * Wake up any sleeping vcpus and trigger a VM-exit in any running
2146 * vcpus so they handle the rendezvous as soon as possible.
2148 for (i = 0; i < VM_MAXCPU; i++) {
2149 if (CPU_ISSET(i, &dest))
2150 vcpu_notify_event(vm, i, false);
2153 vm_handle_rendezvous(vm, vcpuid);
2157 vm_atpic(struct vm *vm)
2159 return (vm->vatpic);
2163 vm_atpit(struct vm *vm)
2165 return (vm->vatpit);
2169 vm_segment_name(int seg)
2171 static enum vm_reg_name seg_names[] = {
2180 KASSERT(seg >= 0 && seg < nitems(seg_names),
2181 ("%s: invalid segment encoding %d", __func__, seg));
2182 return (seg_names[seg]);
2187 * Return the amount of in-use and wired memory for the VM. Since
2188 * these are global stats, only return the values with for vCPU 0
2190 VMM_STAT_DECLARE(VMM_MEM_RESIDENT);
2191 VMM_STAT_DECLARE(VMM_MEM_WIRED);
2194 vm_get_rescnt(struct vm *vm, int vcpu, struct vmm_stat_type *stat)
2198 vmm_stat_set(vm, vcpu, VMM_MEM_RESIDENT,
2199 PAGE_SIZE * vmspace_resident_count(vm->vmspace));
2204 vm_get_wiredcnt(struct vm *vm, int vcpu, struct vmm_stat_type *stat)
2208 vmm_stat_set(vm, vcpu, VMM_MEM_WIRED,
2209 PAGE_SIZE * pmap_wired_count(vmspace_pmap(vm->vmspace)));
2213 VMM_STAT_FUNC(VMM_MEM_RESIDENT, "Resident memory", vm_get_rescnt);
2214 VMM_STAT_FUNC(VMM_MEM_WIRED, "Wired memory", vm_get_wiredcnt);