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"
81 #include "vmm_lapic.h"
90 * (a) allocated when vcpu is created
91 * (i) initialized when vcpu is created and when it is reinitialized
92 * (o) initialized the first time the vcpu is created
93 * (x) initialized before use
96 struct mtx mtx; /* (o) protects 'state' and 'hostcpu' */
97 enum vcpu_state state; /* (o) vcpu state */
98 int hostcpu; /* (o) vcpu's host cpu */
99 struct vlapic *vlapic; /* (i) APIC device model */
100 enum x2apic_state x2apic_state; /* (i) APIC mode */
101 uint64_t exitintinfo; /* (i) events pending at VM exit */
102 int nmi_pending; /* (i) NMI pending */
103 int extint_pending; /* (i) INTR pending */
104 struct vm_exception exception; /* (x) exception collateral */
105 int exception_pending; /* (i) exception pending */
106 struct savefpu *guestfpu; /* (a,i) guest fpu state */
107 uint64_t guest_xcr0; /* (i) guest %xcr0 register */
108 void *stats; /* (a,i) statistics */
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 struct vpmtmr *vpmtmr; /* (i) virtual ACPI PM timer */
140 struct vrtc *vrtc; /* (o) virtual RTC */
141 volatile cpuset_t active_cpus; /* (i) active vcpus */
142 int suspend; /* (i) stop VM execution */
143 volatile cpuset_t suspended_cpus; /* (i) suspended vcpus */
144 volatile cpuset_t halted_cpus; /* (x) cpus in a hard halt */
145 cpuset_t rendezvous_req_cpus; /* (x) rendezvous requested */
146 cpuset_t rendezvous_done_cpus; /* (x) rendezvous finished */
147 void *rendezvous_arg; /* (x) rendezvous func/arg */
148 vm_rendezvous_func_t rendezvous_func;
149 struct mtx rendezvous_mtx; /* (o) rendezvous lock */
150 int num_mem_segs; /* (o) guest memory segments */
151 struct mem_seg mem_segs[VM_MAX_MEMORY_SEGMENTS];
152 struct vmspace *vmspace; /* (o) guest's address space */
153 char name[VM_MAX_NAMELEN]; /* (o) virtual machine name */
154 struct vcpu vcpu[VM_MAXCPU]; /* (i) guest vcpus */
157 static int vmm_initialized;
159 static struct vmm_ops *ops;
160 #define VMM_INIT(num) (ops != NULL ? (*ops->init)(num) : 0)
161 #define VMM_CLEANUP() (ops != NULL ? (*ops->cleanup)() : 0)
162 #define VMM_RESUME() (ops != NULL ? (*ops->resume)() : 0)
164 #define VMINIT(vm, pmap) (ops != NULL ? (*ops->vminit)(vm, pmap): NULL)
165 #define VMRUN(vmi, vcpu, rip, pmap, rptr, sptr) \
166 (ops != NULL ? (*ops->vmrun)(vmi, vcpu, rip, pmap, rptr, sptr) : ENXIO)
167 #define VMCLEANUP(vmi) (ops != NULL ? (*ops->vmcleanup)(vmi) : NULL)
168 #define VMSPACE_ALLOC(min, max) \
169 (ops != NULL ? (*ops->vmspace_alloc)(min, max) : NULL)
170 #define VMSPACE_FREE(vmspace) \
171 (ops != NULL ? (*ops->vmspace_free)(vmspace) : ENXIO)
172 #define VMGETREG(vmi, vcpu, num, retval) \
173 (ops != NULL ? (*ops->vmgetreg)(vmi, vcpu, num, retval) : ENXIO)
174 #define VMSETREG(vmi, vcpu, num, val) \
175 (ops != NULL ? (*ops->vmsetreg)(vmi, vcpu, num, val) : ENXIO)
176 #define VMGETDESC(vmi, vcpu, num, desc) \
177 (ops != NULL ? (*ops->vmgetdesc)(vmi, vcpu, num, desc) : ENXIO)
178 #define VMSETDESC(vmi, vcpu, num, desc) \
179 (ops != NULL ? (*ops->vmsetdesc)(vmi, vcpu, num, desc) : ENXIO)
180 #define VMGETCAP(vmi, vcpu, num, retval) \
181 (ops != NULL ? (*ops->vmgetcap)(vmi, vcpu, num, retval) : ENXIO)
182 #define VMSETCAP(vmi, vcpu, num, val) \
183 (ops != NULL ? (*ops->vmsetcap)(vmi, vcpu, num, val) : ENXIO)
184 #define VLAPIC_INIT(vmi, vcpu) \
185 (ops != NULL ? (*ops->vlapic_init)(vmi, vcpu) : NULL)
186 #define VLAPIC_CLEANUP(vmi, vlapic) \
187 (ops != NULL ? (*ops->vlapic_cleanup)(vmi, vlapic) : NULL)
189 #define fpu_start_emulating() load_cr0(rcr0() | CR0_TS)
190 #define fpu_stop_emulating() clts()
192 static MALLOC_DEFINE(M_VM, "vm", "vm");
195 static VMM_STAT(VCPU_TOTAL_RUNTIME, "vcpu total runtime");
197 SYSCTL_NODE(_hw, OID_AUTO, vmm, CTLFLAG_RW, NULL, NULL);
200 * Halt the guest if all vcpus are executing a HLT instruction with
201 * interrupts disabled.
203 static int halt_detection_enabled = 1;
204 SYSCTL_INT(_hw_vmm, OID_AUTO, halt_detection, CTLFLAG_RDTUN,
205 &halt_detection_enabled, 0,
206 "Halt VM if all vcpus execute HLT with interrupts disabled");
208 static int vmm_ipinum;
209 SYSCTL_INT(_hw_vmm, OID_AUTO, ipinum, CTLFLAG_RD, &vmm_ipinum, 0,
210 "IPI vector used for vcpu notifications");
212 static int trace_guest_exceptions;
213 SYSCTL_INT(_hw_vmm, OID_AUTO, trace_guest_exceptions, CTLFLAG_RDTUN,
214 &trace_guest_exceptions, 0,
215 "Trap into hypervisor on all guest exceptions and reflect them back");
218 vcpu_cleanup(struct vm *vm, int i, bool destroy)
220 struct vcpu *vcpu = &vm->vcpu[i];
222 VLAPIC_CLEANUP(vm->cookie, vcpu->vlapic);
224 vmm_stat_free(vcpu->stats);
225 fpu_save_area_free(vcpu->guestfpu);
230 vcpu_init(struct vm *vm, int vcpu_id, bool create)
234 KASSERT(vcpu_id >= 0 && vcpu_id < VM_MAXCPU,
235 ("vcpu_init: invalid vcpu %d", vcpu_id));
237 vcpu = &vm->vcpu[vcpu_id];
240 KASSERT(!vcpu_lock_initialized(vcpu), ("vcpu %d already "
241 "initialized", vcpu_id));
242 vcpu_lock_init(vcpu);
243 vcpu->state = VCPU_IDLE;
244 vcpu->hostcpu = NOCPU;
245 vcpu->guestfpu = fpu_save_area_alloc();
246 vcpu->stats = vmm_stat_alloc();
249 vcpu->vlapic = VLAPIC_INIT(vm->cookie, vcpu_id);
250 vm_set_x2apic_state(vm, vcpu_id, X2APIC_DISABLED);
251 vcpu->exitintinfo = 0;
252 vcpu->nmi_pending = 0;
253 vcpu->extint_pending = 0;
254 vcpu->exception_pending = 0;
255 vcpu->guest_xcr0 = XFEATURE_ENABLED_X87;
256 fpu_save_area_reset(vcpu->guestfpu);
257 vmm_stat_init(vcpu->stats);
261 vcpu_trace_exceptions(struct vm *vm, int vcpuid)
264 return (trace_guest_exceptions);
268 vm_exitinfo(struct vm *vm, int cpuid)
272 if (cpuid < 0 || cpuid >= VM_MAXCPU)
273 panic("vm_exitinfo: invalid cpuid %d", cpuid);
275 vcpu = &vm->vcpu[cpuid];
277 return (&vcpu->exitinfo);
291 vmm_host_state_init();
293 vmm_ipinum = vmm_ipi_alloc();
295 vmm_ipinum = IPI_AST;
297 error = vmm_mem_init();
302 ops = &vmm_ops_intel;
303 else if (vmm_is_amd())
308 vmm_resume_p = vmm_resume;
310 return (VMM_INIT(vmm_ipinum));
314 vmm_handler(module_t mod, int what, void *arg)
321 if (ppt_avail_devices() > 0)
328 error = vmmdev_cleanup();
332 if (vmm_ipinum != IPI_AST)
333 vmm_ipi_free(vmm_ipinum);
334 error = VMM_CLEANUP();
336 * Something bad happened - prevent new
337 * VMs from being created
350 static moduledata_t vmm_kmod = {
357 * vmm initialization has the following dependencies:
359 * - iommu initialization must happen after the pci passthru driver has had
360 * a chance to attach to any passthru devices (after SI_SUB_CONFIGURE).
362 * - VT-x initialization requires smp_rendezvous() and therefore must happen
363 * after SMP is fully functional (after SI_SUB_SMP).
365 DECLARE_MODULE(vmm, vmm_kmod, SI_SUB_SMP + 1, SI_ORDER_ANY);
366 MODULE_VERSION(vmm, 1);
369 vm_init(struct vm *vm, bool create)
373 vm->cookie = VMINIT(vm, vmspace_pmap(vm->vmspace));
375 vm->vioapic = vioapic_init(vm);
376 vm->vhpet = vhpet_init(vm);
377 vm->vatpic = vatpic_init(vm);
378 vm->vatpit = vatpit_init(vm);
379 vm->vpmtmr = vpmtmr_init(vm);
381 vm->vrtc = vrtc_init(vm);
383 CPU_ZERO(&vm->active_cpus);
386 CPU_ZERO(&vm->suspended_cpus);
388 for (i = 0; i < VM_MAXCPU; i++)
389 vcpu_init(vm, i, create);
393 vm_create(const char *name, struct vm **retvm)
396 struct vmspace *vmspace;
399 * If vmm.ko could not be successfully initialized then don't attempt
400 * to create the virtual machine.
402 if (!vmm_initialized)
405 if (name == NULL || strlen(name) >= VM_MAX_NAMELEN)
408 vmspace = VMSPACE_ALLOC(0, VM_MAXUSER_ADDRESS);
412 vm = malloc(sizeof(struct vm), M_VM, M_WAITOK | M_ZERO);
413 strcpy(vm->name, name);
414 vm->num_mem_segs = 0;
415 vm->vmspace = vmspace;
416 mtx_init(&vm->rendezvous_mtx, "vm rendezvous lock", 0, MTX_DEF);
425 vm_free_mem_seg(struct vm *vm, struct mem_seg *seg)
428 if (seg->object != NULL)
429 vmm_mem_free(vm->vmspace, seg->gpa, seg->len);
431 bzero(seg, sizeof(*seg));
435 vm_cleanup(struct vm *vm, bool destroy)
439 ppt_unassign_all(vm);
441 if (vm->iommu != NULL)
442 iommu_destroy_domain(vm->iommu);
445 vrtc_cleanup(vm->vrtc);
447 vrtc_reset(vm->vrtc);
448 vpmtmr_cleanup(vm->vpmtmr);
449 vatpit_cleanup(vm->vatpit);
450 vhpet_cleanup(vm->vhpet);
451 vatpic_cleanup(vm->vatpic);
452 vioapic_cleanup(vm->vioapic);
454 for (i = 0; i < VM_MAXCPU; i++)
455 vcpu_cleanup(vm, i, destroy);
457 VMCLEANUP(vm->cookie);
460 for (i = 0; i < vm->num_mem_segs; i++)
461 vm_free_mem_seg(vm, &vm->mem_segs[i]);
463 vm->num_mem_segs = 0;
465 VMSPACE_FREE(vm->vmspace);
471 vm_destroy(struct vm *vm)
473 vm_cleanup(vm, true);
478 vm_reinit(struct vm *vm)
483 * A virtual machine can be reset only if all vcpus are suspended.
485 if (CPU_CMP(&vm->suspended_cpus, &vm->active_cpus) == 0) {
486 vm_cleanup(vm, false);
497 vm_name(struct vm *vm)
503 vm_map_mmio(struct vm *vm, vm_paddr_t gpa, size_t len, vm_paddr_t hpa)
507 if ((obj = vmm_mmio_alloc(vm->vmspace, gpa, len, hpa)) == NULL)
514 vm_unmap_mmio(struct vm *vm, vm_paddr_t gpa, size_t len)
517 vmm_mmio_free(vm->vmspace, gpa, len);
522 vm_mem_allocated(struct vm *vm, vm_paddr_t gpa)
525 vm_paddr_t gpabase, gpalimit;
527 for (i = 0; i < vm->num_mem_segs; i++) {
528 gpabase = vm->mem_segs[i].gpa;
529 gpalimit = gpabase + vm->mem_segs[i].len;
530 if (gpa >= gpabase && gpa < gpalimit)
531 return (TRUE); /* 'gpa' is regular memory */
534 if (ppt_is_mmio(vm, gpa))
535 return (TRUE); /* 'gpa' is pci passthru mmio */
541 vm_malloc(struct vm *vm, vm_paddr_t gpa, size_t len)
543 int available, allocated;
548 if ((gpa & PAGE_MASK) || (len & PAGE_MASK) || len == 0)
551 available = allocated = 0;
553 while (g < gpa + len) {
554 if (vm_mem_allocated(vm, g))
563 * If there are some allocated and some available pages in the address
564 * range then it is an error.
566 if (allocated && available)
570 * If the entire address range being requested has already been
571 * allocated then there isn't anything more to do.
573 if (allocated && available == 0)
576 if (vm->num_mem_segs >= VM_MAX_MEMORY_SEGMENTS)
579 seg = &vm->mem_segs[vm->num_mem_segs];
581 if ((object = vmm_mem_alloc(vm->vmspace, gpa, len)) == NULL)
586 seg->object = object;
595 vm_maxmem(struct vm *vm)
598 vm_paddr_t gpa, maxmem;
601 for (i = 0; i < vm->num_mem_segs; i++) {
602 gpa = vm->mem_segs[i].gpa + vm->mem_segs[i].len;
610 vm_gpa_unwire(struct vm *vm)
615 for (i = 0; i < vm->num_mem_segs; i++) {
616 seg = &vm->mem_segs[i];
620 rv = vm_map_unwire(&vm->vmspace->vm_map,
621 seg->gpa, seg->gpa + seg->len,
622 VM_MAP_WIRE_USER | VM_MAP_WIRE_NOHOLES);
623 KASSERT(rv == KERN_SUCCESS, ("vm(%s) memory segment "
624 "%#lx/%ld could not be unwired: %d",
625 vm_name(vm), seg->gpa, seg->len, rv));
632 vm_gpa_wire(struct vm *vm)
637 for (i = 0; i < vm->num_mem_segs; i++) {
638 seg = &vm->mem_segs[i];
643 rv = vm_map_wire(&vm->vmspace->vm_map,
644 seg->gpa, seg->gpa + seg->len,
645 VM_MAP_WIRE_USER | VM_MAP_WIRE_NOHOLES);
646 if (rv != KERN_SUCCESS)
652 if (i < vm->num_mem_segs) {
654 * Undo the wiring before returning an error.
664 vm_iommu_modify(struct vm *vm, boolean_t map)
669 void *vp, *cookie, *host_domain;
672 host_domain = iommu_host_domain();
674 for (i = 0; i < vm->num_mem_segs; i++) {
675 seg = &vm->mem_segs[i];
676 KASSERT(seg->wired, ("vm(%s) memory segment %#lx/%ld not wired",
677 vm_name(vm), seg->gpa, seg->len));
680 while (gpa < seg->gpa + seg->len) {
681 vp = vm_gpa_hold(vm, gpa, PAGE_SIZE, VM_PROT_WRITE,
683 KASSERT(vp != NULL, ("vm(%s) could not map gpa %#lx",
686 vm_gpa_release(cookie);
688 hpa = DMAP_TO_PHYS((uintptr_t)vp);
690 iommu_create_mapping(vm->iommu, gpa, hpa, sz);
691 iommu_remove_mapping(host_domain, hpa, sz);
693 iommu_remove_mapping(vm->iommu, gpa, sz);
694 iommu_create_mapping(host_domain, hpa, hpa, sz);
702 * Invalidate the cached translations associated with the domain
703 * from which pages were removed.
706 iommu_invalidate_tlb(host_domain);
708 iommu_invalidate_tlb(vm->iommu);
711 #define vm_iommu_unmap(vm) vm_iommu_modify((vm), FALSE)
712 #define vm_iommu_map(vm) vm_iommu_modify((vm), TRUE)
715 vm_unassign_pptdev(struct vm *vm, int bus, int slot, int func)
719 error = ppt_unassign_device(vm, bus, slot, func);
723 if (ppt_assigned_devices(vm) == 0) {
731 vm_assign_pptdev(struct vm *vm, int bus, int slot, int func)
737 * Virtual machines with pci passthru devices get special treatment:
738 * - the guest physical memory is wired
739 * - the iommu is programmed to do the 'gpa' to 'hpa' translation
741 * We need to do this before the first pci passthru device is attached.
743 if (ppt_assigned_devices(vm) == 0) {
744 KASSERT(vm->iommu == NULL,
745 ("vm_assign_pptdev: iommu must be NULL"));
746 maxaddr = vm_maxmem(vm);
747 vm->iommu = iommu_create_domain(maxaddr);
749 error = vm_gpa_wire(vm);
756 error = ppt_assign_device(vm, bus, slot, func);
761 vm_gpa_hold(struct vm *vm, vm_paddr_t gpa, size_t len, int reqprot,
767 pageoff = gpa & PAGE_MASK;
768 if (len > PAGE_SIZE - pageoff)
769 panic("vm_gpa_hold: invalid gpa/len: 0x%016lx/%lu", gpa, len);
771 count = vm_fault_quick_hold_pages(&vm->vmspace->vm_map,
772 trunc_page(gpa), PAGE_SIZE, reqprot, &m, 1);
776 return ((void *)(PHYS_TO_DMAP(VM_PAGE_TO_PHYS(m)) + pageoff));
784 vm_gpa_release(void *cookie)
786 vm_page_t m = cookie;
794 vm_gpabase2memseg(struct vm *vm, vm_paddr_t gpabase,
795 struct vm_memory_segment *seg)
799 for (i = 0; i < vm->num_mem_segs; i++) {
800 if (gpabase == vm->mem_segs[i].gpa) {
801 seg->gpa = vm->mem_segs[i].gpa;
802 seg->len = vm->mem_segs[i].len;
803 seg->wired = vm->mem_segs[i].wired;
811 vm_get_memobj(struct vm *vm, vm_paddr_t gpa, size_t len,
812 vm_offset_t *offset, struct vm_object **object)
819 for (i = 0; i < vm->num_mem_segs; i++) {
820 if ((seg_obj = vm->mem_segs[i].object) == NULL)
823 seg_gpa = vm->mem_segs[i].gpa;
824 seg_len = vm->mem_segs[i].len;
826 if (gpa >= seg_gpa && gpa < seg_gpa + seg_len) {
827 *offset = gpa - seg_gpa;
829 vm_object_reference(seg_obj);
838 vm_get_register(struct vm *vm, int vcpu, int reg, uint64_t *retval)
841 if (vcpu < 0 || vcpu >= VM_MAXCPU)
844 if (reg >= VM_REG_LAST)
847 return (VMGETREG(vm->cookie, vcpu, reg, retval));
851 vm_set_register(struct vm *vm, int vcpu, int reg, uint64_t val)
854 if (vcpu < 0 || vcpu >= VM_MAXCPU)
857 if (reg >= VM_REG_LAST)
860 return (VMSETREG(vm->cookie, vcpu, reg, val));
864 is_descriptor_table(int reg)
868 case VM_REG_GUEST_IDTR:
869 case VM_REG_GUEST_GDTR:
877 is_segment_register(int reg)
881 case VM_REG_GUEST_ES:
882 case VM_REG_GUEST_CS:
883 case VM_REG_GUEST_SS:
884 case VM_REG_GUEST_DS:
885 case VM_REG_GUEST_FS:
886 case VM_REG_GUEST_GS:
887 case VM_REG_GUEST_TR:
888 case VM_REG_GUEST_LDTR:
896 vm_get_seg_desc(struct vm *vm, int vcpu, int reg,
897 struct seg_desc *desc)
900 if (vcpu < 0 || vcpu >= VM_MAXCPU)
903 if (!is_segment_register(reg) && !is_descriptor_table(reg))
906 return (VMGETDESC(vm->cookie, vcpu, reg, desc));
910 vm_set_seg_desc(struct vm *vm, int vcpu, int reg,
911 struct seg_desc *desc)
913 if (vcpu < 0 || vcpu >= VM_MAXCPU)
916 if (!is_segment_register(reg) && !is_descriptor_table(reg))
919 return (VMSETDESC(vm->cookie, vcpu, reg, desc));
923 restore_guest_fpustate(struct vcpu *vcpu)
926 /* flush host state to the pcb */
929 /* restore guest FPU state */
930 fpu_stop_emulating();
931 fpurestore(vcpu->guestfpu);
933 /* restore guest XCR0 if XSAVE is enabled in the host */
934 if (rcr4() & CR4_XSAVE)
935 load_xcr(0, vcpu->guest_xcr0);
938 * The FPU is now "dirty" with the guest's state so turn on emulation
939 * to trap any access to the FPU by the host.
941 fpu_start_emulating();
945 save_guest_fpustate(struct vcpu *vcpu)
948 if ((rcr0() & CR0_TS) == 0)
949 panic("fpu emulation not enabled in host!");
951 /* save guest XCR0 and restore host XCR0 */
952 if (rcr4() & CR4_XSAVE) {
953 vcpu->guest_xcr0 = rxcr(0);
954 load_xcr(0, vmm_get_host_xcr0());
957 /* save guest FPU state */
958 fpu_stop_emulating();
959 fpusave(vcpu->guestfpu);
960 fpu_start_emulating();
963 static VMM_STAT(VCPU_IDLE_TICKS, "number of ticks vcpu was idle");
966 vcpu_set_state_locked(struct vcpu *vcpu, enum vcpu_state newstate,
971 vcpu_assert_locked(vcpu);
974 * State transitions from the vmmdev_ioctl() must always begin from
975 * the VCPU_IDLE state. This guarantees that there is only a single
976 * ioctl() operating on a vcpu at any point.
979 while (vcpu->state != VCPU_IDLE)
980 msleep_spin(&vcpu->state, &vcpu->mtx, "vmstat", hz);
982 KASSERT(vcpu->state != VCPU_IDLE, ("invalid transition from "
986 if (vcpu->state == VCPU_RUNNING) {
987 KASSERT(vcpu->hostcpu == curcpu, ("curcpu %d and hostcpu %d "
988 "mismatch for running vcpu", curcpu, vcpu->hostcpu));
990 KASSERT(vcpu->hostcpu == NOCPU, ("Invalid hostcpu %d for a "
991 "vcpu that is not running", vcpu->hostcpu));
995 * The following state transitions are allowed:
996 * IDLE -> FROZEN -> IDLE
997 * FROZEN -> RUNNING -> FROZEN
998 * FROZEN -> SLEEPING -> FROZEN
1000 switch (vcpu->state) {
1004 error = (newstate != VCPU_FROZEN);
1007 error = (newstate == VCPU_FROZEN);
1017 vcpu->state = newstate;
1018 if (newstate == VCPU_RUNNING)
1019 vcpu->hostcpu = curcpu;
1021 vcpu->hostcpu = NOCPU;
1023 if (newstate == VCPU_IDLE)
1024 wakeup(&vcpu->state);
1030 vcpu_require_state(struct vm *vm, int vcpuid, enum vcpu_state newstate)
1034 if ((error = vcpu_set_state(vm, vcpuid, newstate, false)) != 0)
1035 panic("Error %d setting state to %d\n", error, newstate);
1039 vcpu_require_state_locked(struct vcpu *vcpu, enum vcpu_state newstate)
1043 if ((error = vcpu_set_state_locked(vcpu, newstate, false)) != 0)
1044 panic("Error %d setting state to %d", error, newstate);
1048 vm_set_rendezvous_func(struct vm *vm, vm_rendezvous_func_t func)
1051 KASSERT(mtx_owned(&vm->rendezvous_mtx), ("rendezvous_mtx not locked"));
1054 * Update 'rendezvous_func' and execute a write memory barrier to
1055 * ensure that it is visible across all host cpus. This is not needed
1056 * for correctness but it does ensure that all the vcpus will notice
1057 * that the rendezvous is requested immediately.
1059 vm->rendezvous_func = func;
1063 #define RENDEZVOUS_CTR0(vm, vcpuid, fmt) \
1066 VCPU_CTR0(vm, vcpuid, fmt); \
1072 vm_handle_rendezvous(struct vm *vm, int vcpuid)
1075 KASSERT(vcpuid == -1 || (vcpuid >= 0 && vcpuid < VM_MAXCPU),
1076 ("vm_handle_rendezvous: invalid vcpuid %d", vcpuid));
1078 mtx_lock(&vm->rendezvous_mtx);
1079 while (vm->rendezvous_func != NULL) {
1080 /* 'rendezvous_req_cpus' must be a subset of 'active_cpus' */
1081 CPU_AND(&vm->rendezvous_req_cpus, &vm->active_cpus);
1084 CPU_ISSET(vcpuid, &vm->rendezvous_req_cpus) &&
1085 !CPU_ISSET(vcpuid, &vm->rendezvous_done_cpus)) {
1086 VCPU_CTR0(vm, vcpuid, "Calling rendezvous func");
1087 (*vm->rendezvous_func)(vm, vcpuid, vm->rendezvous_arg);
1088 CPU_SET(vcpuid, &vm->rendezvous_done_cpus);
1090 if (CPU_CMP(&vm->rendezvous_req_cpus,
1091 &vm->rendezvous_done_cpus) == 0) {
1092 VCPU_CTR0(vm, vcpuid, "Rendezvous completed");
1093 vm_set_rendezvous_func(vm, NULL);
1094 wakeup(&vm->rendezvous_func);
1097 RENDEZVOUS_CTR0(vm, vcpuid, "Wait for rendezvous completion");
1098 mtx_sleep(&vm->rendezvous_func, &vm->rendezvous_mtx, 0,
1101 mtx_unlock(&vm->rendezvous_mtx);
1105 * Emulate a guest 'hlt' by sleeping until the vcpu is ready to run.
1108 vm_handle_hlt(struct vm *vm, int vcpuid, bool intr_disabled, bool *retu)
1112 int t, vcpu_halted, vm_halted;
1114 KASSERT(!CPU_ISSET(vcpuid, &vm->halted_cpus), ("vcpu already halted"));
1116 vcpu = &vm->vcpu[vcpuid];
1123 * Do a final check for pending NMI or interrupts before
1124 * really putting this thread to sleep. Also check for
1125 * software events that would cause this vcpu to wakeup.
1127 * These interrupts/events could have happened after the
1128 * vcpu returned from VMRUN() and before it acquired the
1131 if (vm->rendezvous_func != NULL || vm->suspend)
1133 if (vm_nmi_pending(vm, vcpuid))
1135 if (!intr_disabled) {
1136 if (vm_extint_pending(vm, vcpuid) ||
1137 vlapic_pending_intr(vcpu->vlapic, NULL)) {
1142 /* Don't go to sleep if the vcpu thread needs to yield */
1143 if (vcpu_should_yield(vm, vcpuid))
1147 * Some Linux guests implement "halt" by having all vcpus
1148 * execute HLT with interrupts disabled. 'halted_cpus' keeps
1149 * track of the vcpus that have entered this state. When all
1150 * vcpus enter the halted state the virtual machine is halted.
1152 if (intr_disabled) {
1154 VCPU_CTR0(vm, vcpuid, "Halted");
1155 if (!vcpu_halted && halt_detection_enabled) {
1157 CPU_SET_ATOMIC(vcpuid, &vm->halted_cpus);
1159 if (CPU_CMP(&vm->halted_cpus, &vm->active_cpus) == 0) {
1168 vcpu_require_state_locked(vcpu, VCPU_SLEEPING);
1170 * XXX msleep_spin() cannot be interrupted by signals so
1171 * wake up periodically to check pending signals.
1173 msleep_spin(vcpu, &vcpu->mtx, wmesg, hz);
1174 vcpu_require_state_locked(vcpu, VCPU_FROZEN);
1175 vmm_stat_incr(vm, vcpuid, VCPU_IDLE_TICKS, ticks - t);
1179 CPU_CLR_ATOMIC(vcpuid, &vm->halted_cpus);
1184 vm_suspend(vm, VM_SUSPEND_HALT);
1190 vm_handle_paging(struct vm *vm, int vcpuid, bool *retu)
1195 struct vm_exit *vme;
1197 vcpu = &vm->vcpu[vcpuid];
1198 vme = &vcpu->exitinfo;
1200 ftype = vme->u.paging.fault_type;
1201 KASSERT(ftype == VM_PROT_READ ||
1202 ftype == VM_PROT_WRITE || ftype == VM_PROT_EXECUTE,
1203 ("vm_handle_paging: invalid fault_type %d", ftype));
1205 if (ftype == VM_PROT_READ || ftype == VM_PROT_WRITE) {
1206 rv = pmap_emulate_accessed_dirty(vmspace_pmap(vm->vmspace),
1207 vme->u.paging.gpa, ftype);
1209 VCPU_CTR2(vm, vcpuid, "%s bit emulation for gpa %#lx",
1210 ftype == VM_PROT_READ ? "accessed" : "dirty",
1216 map = &vm->vmspace->vm_map;
1217 rv = vm_fault(map, vme->u.paging.gpa, ftype, VM_FAULT_NORMAL);
1219 VCPU_CTR3(vm, vcpuid, "vm_handle_paging rv = %d, gpa = %#lx, "
1220 "ftype = %d", rv, vme->u.paging.gpa, ftype);
1222 if (rv != KERN_SUCCESS)
1225 /* restart execution at the faulting instruction */
1226 vme->inst_length = 0;
1232 vm_handle_inst_emul(struct vm *vm, int vcpuid, bool *retu)
1236 struct vm_exit *vme;
1238 struct vm_guest_paging *paging;
1239 mem_region_read_t mread;
1240 mem_region_write_t mwrite;
1241 enum vm_cpu_mode cpu_mode;
1242 int cs_d, error, length;
1244 vcpu = &vm->vcpu[vcpuid];
1245 vme = &vcpu->exitinfo;
1247 gla = vme->u.inst_emul.gla;
1248 gpa = vme->u.inst_emul.gpa;
1249 cs_d = vme->u.inst_emul.cs_d;
1250 vie = &vme->u.inst_emul.vie;
1251 paging = &vme->u.inst_emul.paging;
1252 cpu_mode = paging->cpu_mode;
1254 VCPU_CTR1(vm, vcpuid, "inst_emul fault accessing gpa %#lx", gpa);
1256 /* Fetch, decode and emulate the faulting instruction */
1257 if (vie->num_valid == 0) {
1259 * If the instruction length is not known then assume a
1260 * maximum size instruction.
1262 length = vme->inst_length ? vme->inst_length : VIE_INST_SIZE;
1263 error = vmm_fetch_instruction(vm, vcpuid, paging, vme->rip,
1267 * The instruction bytes have already been copied into 'vie'
1272 return (0); /* Resume guest to handle page fault */
1273 else if (error == -1)
1275 else if (error != 0)
1276 panic("%s: vmm_fetch_instruction error %d", __func__, error);
1278 if (vmm_decode_instruction(vm, vcpuid, gla, cpu_mode, cs_d, vie) != 0)
1282 * If the instruction length is not specified the update it now.
1284 if (vme->inst_length == 0)
1285 vme->inst_length = vie->num_processed;
1287 /* return to userland unless this is an in-kernel emulated device */
1288 if (gpa >= DEFAULT_APIC_BASE && gpa < DEFAULT_APIC_BASE + PAGE_SIZE) {
1289 mread = lapic_mmio_read;
1290 mwrite = lapic_mmio_write;
1291 } else if (gpa >= VIOAPIC_BASE && gpa < VIOAPIC_BASE + VIOAPIC_SIZE) {
1292 mread = vioapic_mmio_read;
1293 mwrite = vioapic_mmio_write;
1294 } else if (gpa >= VHPET_BASE && gpa < VHPET_BASE + VHPET_SIZE) {
1295 mread = vhpet_mmio_read;
1296 mwrite = vhpet_mmio_write;
1302 error = vmm_emulate_instruction(vm, vcpuid, gpa, vie, paging,
1303 mread, mwrite, retu);
1309 vm_handle_suspend(struct vm *vm, int vcpuid, bool *retu)
1315 vcpu = &vm->vcpu[vcpuid];
1317 CPU_SET_ATOMIC(vcpuid, &vm->suspended_cpus);
1320 * Wait until all 'active_cpus' have suspended themselves.
1322 * Since a VM may be suspended at any time including when one or
1323 * more vcpus are doing a rendezvous we need to call the rendezvous
1324 * handler while we are waiting to prevent a deadlock.
1328 if (CPU_CMP(&vm->suspended_cpus, &vm->active_cpus) == 0) {
1329 VCPU_CTR0(vm, vcpuid, "All vcpus suspended");
1333 if (vm->rendezvous_func == NULL) {
1334 VCPU_CTR0(vm, vcpuid, "Sleeping during suspend");
1335 vcpu_require_state_locked(vcpu, VCPU_SLEEPING);
1336 msleep_spin(vcpu, &vcpu->mtx, "vmsusp", hz);
1337 vcpu_require_state_locked(vcpu, VCPU_FROZEN);
1339 VCPU_CTR0(vm, vcpuid, "Rendezvous during suspend");
1341 vm_handle_rendezvous(vm, vcpuid);
1348 * Wakeup the other sleeping vcpus and return to userspace.
1350 for (i = 0; i < VM_MAXCPU; i++) {
1351 if (CPU_ISSET(i, &vm->suspended_cpus)) {
1352 vcpu_notify_event(vm, i, false);
1361 vm_suspend(struct vm *vm, enum vm_suspend_how how)
1365 if (how <= VM_SUSPEND_NONE || how >= VM_SUSPEND_LAST)
1368 if (atomic_cmpset_int(&vm->suspend, 0, how) == 0) {
1369 VM_CTR2(vm, "virtual machine already suspended %d/%d",
1374 VM_CTR1(vm, "virtual machine successfully suspended %d", how);
1377 * Notify all active vcpus that they are now suspended.
1379 for (i = 0; i < VM_MAXCPU; i++) {
1380 if (CPU_ISSET(i, &vm->active_cpus))
1381 vcpu_notify_event(vm, i, false);
1388 vm_exit_suspended(struct vm *vm, int vcpuid, uint64_t rip)
1390 struct vm_exit *vmexit;
1392 KASSERT(vm->suspend > VM_SUSPEND_NONE && vm->suspend < VM_SUSPEND_LAST,
1393 ("vm_exit_suspended: invalid suspend type %d", vm->suspend));
1395 vmexit = vm_exitinfo(vm, vcpuid);
1397 vmexit->inst_length = 0;
1398 vmexit->exitcode = VM_EXITCODE_SUSPENDED;
1399 vmexit->u.suspended.how = vm->suspend;
1403 vm_exit_rendezvous(struct vm *vm, int vcpuid, uint64_t rip)
1405 struct vm_exit *vmexit;
1407 KASSERT(vm->rendezvous_func != NULL, ("rendezvous not in progress"));
1409 vmexit = vm_exitinfo(vm, vcpuid);
1411 vmexit->inst_length = 0;
1412 vmexit->exitcode = VM_EXITCODE_RENDEZVOUS;
1413 vmm_stat_incr(vm, vcpuid, VMEXIT_RENDEZVOUS, 1);
1417 vm_exit_astpending(struct vm *vm, int vcpuid, uint64_t rip)
1419 struct vm_exit *vmexit;
1421 vmexit = vm_exitinfo(vm, vcpuid);
1423 vmexit->inst_length = 0;
1424 vmexit->exitcode = VM_EXITCODE_BOGUS;
1425 vmm_stat_incr(vm, vcpuid, VMEXIT_ASTPENDING, 1);
1429 vm_run(struct vm *vm, struct vm_run *vmrun)
1434 uint64_t tscval, rip;
1435 struct vm_exit *vme;
1436 bool retu, intr_disabled;
1440 vcpuid = vmrun->cpuid;
1442 if (vcpuid < 0 || vcpuid >= VM_MAXCPU)
1445 if (!CPU_ISSET(vcpuid, &vm->active_cpus))
1448 if (CPU_ISSET(vcpuid, &vm->suspended_cpus))
1451 rptr = &vm->rendezvous_func;
1452 sptr = &vm->suspend;
1453 pmap = vmspace_pmap(vm->vmspace);
1454 vcpu = &vm->vcpu[vcpuid];
1455 vme = &vcpu->exitinfo;
1460 KASSERT(!CPU_ISSET(curcpu, &pmap->pm_active),
1461 ("vm_run: absurd pm_active"));
1465 pcb = PCPU_GET(curpcb);
1466 set_pcb_flags(pcb, PCB_FULL_IRET);
1468 restore_guest_fpustate(vcpu);
1470 vcpu_require_state(vm, vcpuid, VCPU_RUNNING);
1471 error = VMRUN(vm->cookie, vcpuid, rip, pmap, rptr, sptr);
1472 vcpu_require_state(vm, vcpuid, VCPU_FROZEN);
1474 save_guest_fpustate(vcpu);
1476 vmm_stat_incr(vm, vcpuid, VCPU_TOTAL_RUNTIME, rdtsc() - tscval);
1482 switch (vme->exitcode) {
1483 case VM_EXITCODE_SUSPENDED:
1484 error = vm_handle_suspend(vm, vcpuid, &retu);
1486 case VM_EXITCODE_IOAPIC_EOI:
1487 vioapic_process_eoi(vm, vcpuid,
1488 vme->u.ioapic_eoi.vector);
1490 case VM_EXITCODE_RENDEZVOUS:
1491 vm_handle_rendezvous(vm, vcpuid);
1494 case VM_EXITCODE_HLT:
1495 intr_disabled = ((vme->u.hlt.rflags & PSL_I) == 0);
1496 error = vm_handle_hlt(vm, vcpuid, intr_disabled, &retu);
1498 case VM_EXITCODE_PAGING:
1499 error = vm_handle_paging(vm, vcpuid, &retu);
1501 case VM_EXITCODE_INST_EMUL:
1502 error = vm_handle_inst_emul(vm, vcpuid, &retu);
1504 case VM_EXITCODE_INOUT:
1505 case VM_EXITCODE_INOUT_STR:
1506 error = vm_handle_inout(vm, vcpuid, vme, &retu);
1508 case VM_EXITCODE_MONITOR:
1509 case VM_EXITCODE_MWAIT:
1510 vm_inject_ud(vm, vcpuid);
1513 retu = true; /* handled in userland */
1518 if (error == 0 && retu == false) {
1519 rip = vme->rip + vme->inst_length;
1523 /* copy the exit information */
1524 bcopy(vme, &vmrun->vm_exit, sizeof(struct vm_exit));
1529 vm_exit_intinfo(struct vm *vm, int vcpuid, uint64_t info)
1534 if (vcpuid < 0 || vcpuid >= VM_MAXCPU)
1537 vcpu = &vm->vcpu[vcpuid];
1539 if (info & VM_INTINFO_VALID) {
1540 type = info & VM_INTINFO_TYPE;
1541 vector = info & 0xff;
1542 if (type == VM_INTINFO_NMI && vector != IDT_NMI)
1544 if (type == VM_INTINFO_HWEXCEPTION && vector >= 32)
1546 if (info & VM_INTINFO_RSVD)
1551 VCPU_CTR2(vm, vcpuid, "%s: info1(%#lx)", __func__, info);
1552 vcpu->exitintinfo = info;
1562 #define IDT_VE 20 /* Virtualization Exception (Intel specific) */
1564 static enum exc_class
1565 exception_class(uint64_t info)
1569 KASSERT(info & VM_INTINFO_VALID, ("intinfo must be valid: %#lx", info));
1570 type = info & VM_INTINFO_TYPE;
1571 vector = info & 0xff;
1573 /* Table 6-4, "Interrupt and Exception Classes", Intel SDM, Vol 3 */
1575 case VM_INTINFO_HWINTR:
1576 case VM_INTINFO_SWINTR:
1577 case VM_INTINFO_NMI:
1578 return (EXC_BENIGN);
1581 * Hardware exception.
1583 * SVM and VT-x use identical type values to represent NMI,
1584 * hardware interrupt and software interrupt.
1586 * SVM uses type '3' for all exceptions. VT-x uses type '3'
1587 * for exceptions except #BP and #OF. #BP and #OF use a type
1588 * value of '5' or '6'. Therefore we don't check for explicit
1589 * values of 'type' to classify 'intinfo' into a hardware
1598 return (EXC_PAGEFAULT);
1604 return (EXC_CONTRIBUTORY);
1606 return (EXC_BENIGN);
1611 nested_fault(struct vm *vm, int vcpuid, uint64_t info1, uint64_t info2,
1614 enum exc_class exc1, exc2;
1617 KASSERT(info1 & VM_INTINFO_VALID, ("info1 %#lx is not valid", info1));
1618 KASSERT(info2 & VM_INTINFO_VALID, ("info2 %#lx is not valid", info2));
1621 * If an exception occurs while attempting to call the double-fault
1622 * handler the processor enters shutdown mode (aka triple fault).
1624 type1 = info1 & VM_INTINFO_TYPE;
1625 vector1 = info1 & 0xff;
1626 if (type1 == VM_INTINFO_HWEXCEPTION && vector1 == IDT_DF) {
1627 VCPU_CTR2(vm, vcpuid, "triple fault: info1(%#lx), info2(%#lx)",
1629 vm_suspend(vm, VM_SUSPEND_TRIPLEFAULT);
1635 * Table 6-5 "Conditions for Generating a Double Fault", Intel SDM, Vol3
1637 exc1 = exception_class(info1);
1638 exc2 = exception_class(info2);
1639 if ((exc1 == EXC_CONTRIBUTORY && exc2 == EXC_CONTRIBUTORY) ||
1640 (exc1 == EXC_PAGEFAULT && exc2 != EXC_BENIGN)) {
1641 /* Convert nested fault into a double fault. */
1643 *retinfo |= VM_INTINFO_VALID | VM_INTINFO_HWEXCEPTION;
1644 *retinfo |= VM_INTINFO_DEL_ERRCODE;
1646 /* Handle exceptions serially */
1653 vcpu_exception_intinfo(struct vcpu *vcpu)
1657 if (vcpu->exception_pending) {
1658 info = vcpu->exception.vector & 0xff;
1659 info |= VM_INTINFO_VALID | VM_INTINFO_HWEXCEPTION;
1660 if (vcpu->exception.error_code_valid) {
1661 info |= VM_INTINFO_DEL_ERRCODE;
1662 info |= (uint64_t)vcpu->exception.error_code << 32;
1669 vm_entry_intinfo(struct vm *vm, int vcpuid, uint64_t *retinfo)
1672 uint64_t info1, info2;
1675 KASSERT(vcpuid >= 0 && vcpuid < VM_MAXCPU, ("invalid vcpu %d", vcpuid));
1677 vcpu = &vm->vcpu[vcpuid];
1679 info1 = vcpu->exitintinfo;
1680 vcpu->exitintinfo = 0;
1683 if (vcpu->exception_pending) {
1684 info2 = vcpu_exception_intinfo(vcpu);
1685 vcpu->exception_pending = 0;
1686 VCPU_CTR2(vm, vcpuid, "Exception %d delivered: %#lx",
1687 vcpu->exception.vector, info2);
1690 if ((info1 & VM_INTINFO_VALID) && (info2 & VM_INTINFO_VALID)) {
1691 valid = nested_fault(vm, vcpuid, info1, info2, retinfo);
1692 } else if (info1 & VM_INTINFO_VALID) {
1695 } else if (info2 & VM_INTINFO_VALID) {
1703 VCPU_CTR4(vm, vcpuid, "%s: info1(%#lx), info2(%#lx), "
1704 "retinfo(%#lx)", __func__, info1, info2, *retinfo);
1711 vm_get_intinfo(struct vm *vm, int vcpuid, uint64_t *info1, uint64_t *info2)
1715 if (vcpuid < 0 || vcpuid >= VM_MAXCPU)
1718 vcpu = &vm->vcpu[vcpuid];
1719 *info1 = vcpu->exitintinfo;
1720 *info2 = vcpu_exception_intinfo(vcpu);
1725 vm_inject_exception(struct vm *vm, int vcpuid, struct vm_exception *exception)
1730 if (vcpuid < 0 || vcpuid >= VM_MAXCPU)
1733 if (exception->vector < 0 || exception->vector >= 32)
1737 * A double fault exception should never be injected directly into
1738 * the guest. It is a derived exception that results from specific
1739 * combinations of nested faults.
1741 if (exception->vector == IDT_DF)
1744 vcpu = &vm->vcpu[vcpuid];
1746 if (vcpu->exception_pending) {
1747 VCPU_CTR2(vm, vcpuid, "Unable to inject exception %d due to "
1748 "pending exception %d", exception->vector,
1749 vcpu->exception.vector);
1754 * From section 26.6.1 "Interruptibility State" in Intel SDM:
1756 * Event blocking by "STI" or "MOV SS" is cleared after guest executes
1757 * one instruction or incurs an exception.
1759 error = vm_set_register(vm, vcpuid, VM_REG_GUEST_INTR_SHADOW, 0);
1760 KASSERT(error == 0, ("%s: error %d clearing interrupt shadow",
1763 vcpu->exception_pending = 1;
1764 vcpu->exception = *exception;
1765 VCPU_CTR1(vm, vcpuid, "Exception %d pending", exception->vector);
1770 vm_inject_fault(void *vmarg, int vcpuid, int vector, int errcode_valid,
1773 struct vm_exception exception;
1774 struct vm_exit *vmexit;
1780 exception.vector = vector;
1781 exception.error_code = errcode;
1782 exception.error_code_valid = errcode_valid;
1783 error = vm_inject_exception(vm, vcpuid, &exception);
1784 KASSERT(error == 0, ("vm_inject_exception error %d", error));
1787 * A fault-like exception allows the instruction to be restarted
1788 * after the exception handler returns.
1790 * By setting the inst_length to 0 we ensure that the instruction
1791 * pointer remains at the faulting instruction.
1793 vmexit = vm_exitinfo(vm, vcpuid);
1794 vmexit->inst_length = 0;
1798 vm_inject_pf(void *vmarg, int vcpuid, int error_code, uint64_t cr2)
1804 VCPU_CTR2(vm, vcpuid, "Injecting page fault: error_code %#x, cr2 %#lx",
1807 error = vm_set_register(vm, vcpuid, VM_REG_GUEST_CR2, cr2);
1808 KASSERT(error == 0, ("vm_set_register(cr2) error %d", error));
1810 vm_inject_fault(vm, vcpuid, IDT_PF, 1, error_code);
1813 static VMM_STAT(VCPU_NMI_COUNT, "number of NMIs delivered to vcpu");
1816 vm_inject_nmi(struct vm *vm, int vcpuid)
1820 if (vcpuid < 0 || vcpuid >= VM_MAXCPU)
1823 vcpu = &vm->vcpu[vcpuid];
1825 vcpu->nmi_pending = 1;
1826 vcpu_notify_event(vm, vcpuid, false);
1831 vm_nmi_pending(struct vm *vm, int vcpuid)
1835 if (vcpuid < 0 || vcpuid >= VM_MAXCPU)
1836 panic("vm_nmi_pending: invalid vcpuid %d", vcpuid);
1838 vcpu = &vm->vcpu[vcpuid];
1840 return (vcpu->nmi_pending);
1844 vm_nmi_clear(struct vm *vm, int vcpuid)
1848 if (vcpuid < 0 || vcpuid >= VM_MAXCPU)
1849 panic("vm_nmi_pending: invalid vcpuid %d", vcpuid);
1851 vcpu = &vm->vcpu[vcpuid];
1853 if (vcpu->nmi_pending == 0)
1854 panic("vm_nmi_clear: inconsistent nmi_pending state");
1856 vcpu->nmi_pending = 0;
1857 vmm_stat_incr(vm, vcpuid, VCPU_NMI_COUNT, 1);
1860 static VMM_STAT(VCPU_EXTINT_COUNT, "number of ExtINTs delivered to vcpu");
1863 vm_inject_extint(struct vm *vm, int vcpuid)
1867 if (vcpuid < 0 || vcpuid >= VM_MAXCPU)
1870 vcpu = &vm->vcpu[vcpuid];
1872 vcpu->extint_pending = 1;
1873 vcpu_notify_event(vm, vcpuid, false);
1878 vm_extint_pending(struct vm *vm, int vcpuid)
1882 if (vcpuid < 0 || vcpuid >= VM_MAXCPU)
1883 panic("vm_extint_pending: invalid vcpuid %d", vcpuid);
1885 vcpu = &vm->vcpu[vcpuid];
1887 return (vcpu->extint_pending);
1891 vm_extint_clear(struct vm *vm, int vcpuid)
1895 if (vcpuid < 0 || vcpuid >= VM_MAXCPU)
1896 panic("vm_extint_pending: invalid vcpuid %d", vcpuid);
1898 vcpu = &vm->vcpu[vcpuid];
1900 if (vcpu->extint_pending == 0)
1901 panic("vm_extint_clear: inconsistent extint_pending state");
1903 vcpu->extint_pending = 0;
1904 vmm_stat_incr(vm, vcpuid, VCPU_EXTINT_COUNT, 1);
1908 vm_get_capability(struct vm *vm, int vcpu, int type, int *retval)
1910 if (vcpu < 0 || vcpu >= VM_MAXCPU)
1913 if (type < 0 || type >= VM_CAP_MAX)
1916 return (VMGETCAP(vm->cookie, vcpu, type, retval));
1920 vm_set_capability(struct vm *vm, int vcpu, int type, int val)
1922 if (vcpu < 0 || vcpu >= VM_MAXCPU)
1925 if (type < 0 || type >= VM_CAP_MAX)
1928 return (VMSETCAP(vm->cookie, vcpu, type, val));
1932 vm_lapic(struct vm *vm, int cpu)
1934 return (vm->vcpu[cpu].vlapic);
1938 vm_ioapic(struct vm *vm)
1941 return (vm->vioapic);
1945 vm_hpet(struct vm *vm)
1952 vmm_is_pptdev(int bus, int slot, int func)
1956 char *val, *cp, *cp2;
1960 * The length of an environment variable is limited to 128 bytes which
1961 * puts an upper limit on the number of passthru devices that may be
1962 * specified using a single environment variable.
1964 * Work around this by scanning multiple environment variable
1965 * names instead of a single one - yuck!
1967 const char *names[] = { "pptdevs", "pptdevs2", "pptdevs3", NULL };
1969 /* set pptdevs="1/2/3 4/5/6 7/8/9 10/11/12" */
1971 for (i = 0; names[i] != NULL && !found; i++) {
1972 cp = val = kern_getenv(names[i]);
1973 while (cp != NULL && *cp != '\0') {
1974 if ((cp2 = strchr(cp, ' ')) != NULL)
1977 n = sscanf(cp, "%d/%d/%d", &b, &s, &f);
1978 if (n == 3 && bus == b && slot == s && func == f) {
1994 vm_iommu_domain(struct vm *vm)
2001 vcpu_set_state(struct vm *vm, int vcpuid, enum vcpu_state newstate,
2007 if (vcpuid < 0 || vcpuid >= VM_MAXCPU)
2008 panic("vm_set_run_state: invalid vcpuid %d", vcpuid);
2010 vcpu = &vm->vcpu[vcpuid];
2013 error = vcpu_set_state_locked(vcpu, newstate, from_idle);
2020 vcpu_get_state(struct vm *vm, int vcpuid, int *hostcpu)
2023 enum vcpu_state state;
2025 if (vcpuid < 0 || vcpuid >= VM_MAXCPU)
2026 panic("vm_get_run_state: invalid vcpuid %d", vcpuid);
2028 vcpu = &vm->vcpu[vcpuid];
2031 state = vcpu->state;
2032 if (hostcpu != NULL)
2033 *hostcpu = vcpu->hostcpu;
2040 vm_activate_cpu(struct vm *vm, int vcpuid)
2043 if (vcpuid < 0 || vcpuid >= VM_MAXCPU)
2046 if (CPU_ISSET(vcpuid, &vm->active_cpus))
2049 VCPU_CTR0(vm, vcpuid, "activated");
2050 CPU_SET_ATOMIC(vcpuid, &vm->active_cpus);
2055 vm_active_cpus(struct vm *vm)
2058 return (vm->active_cpus);
2062 vm_suspended_cpus(struct vm *vm)
2065 return (vm->suspended_cpus);
2069 vcpu_stats(struct vm *vm, int vcpuid)
2072 return (vm->vcpu[vcpuid].stats);
2076 vm_get_x2apic_state(struct vm *vm, int vcpuid, enum x2apic_state *state)
2078 if (vcpuid < 0 || vcpuid >= VM_MAXCPU)
2081 *state = vm->vcpu[vcpuid].x2apic_state;
2087 vm_set_x2apic_state(struct vm *vm, int vcpuid, enum x2apic_state state)
2089 if (vcpuid < 0 || vcpuid >= VM_MAXCPU)
2092 if (state >= X2APIC_STATE_LAST)
2095 vm->vcpu[vcpuid].x2apic_state = state;
2097 vlapic_set_x2apic_state(vm, vcpuid, state);
2103 * This function is called to ensure that a vcpu "sees" a pending event
2104 * as soon as possible:
2105 * - If the vcpu thread is sleeping then it is woken up.
2106 * - If the vcpu is running on a different host_cpu then an IPI will be directed
2107 * to the host_cpu to cause the vcpu to trap into the hypervisor.
2110 vcpu_notify_event(struct vm *vm, int vcpuid, bool lapic_intr)
2115 vcpu = &vm->vcpu[vcpuid];
2118 hostcpu = vcpu->hostcpu;
2119 if (vcpu->state == VCPU_RUNNING) {
2120 KASSERT(hostcpu != NOCPU, ("vcpu running on invalid hostcpu"));
2121 if (hostcpu != curcpu) {
2123 vlapic_post_intr(vcpu->vlapic, hostcpu,
2126 ipi_cpu(hostcpu, vmm_ipinum);
2130 * If the 'vcpu' is running on 'curcpu' then it must
2131 * be sending a notification to itself (e.g. SELF_IPI).
2132 * The pending event will be picked up when the vcpu
2133 * transitions back to guest context.
2137 KASSERT(hostcpu == NOCPU, ("vcpu state %d not consistent "
2138 "with hostcpu %d", vcpu->state, hostcpu));
2139 if (vcpu->state == VCPU_SLEEPING)
2146 vm_get_vmspace(struct vm *vm)
2149 return (vm->vmspace);
2153 vm_apicid2vcpuid(struct vm *vm, int apicid)
2156 * XXX apic id is assumed to be numerically identical to vcpu id
2162 vm_smp_rendezvous(struct vm *vm, int vcpuid, cpuset_t dest,
2163 vm_rendezvous_func_t func, void *arg)
2168 * Enforce that this function is called without any locks
2170 WITNESS_WARN(WARN_PANIC, NULL, "vm_smp_rendezvous");
2171 KASSERT(vcpuid == -1 || (vcpuid >= 0 && vcpuid < VM_MAXCPU),
2172 ("vm_smp_rendezvous: invalid vcpuid %d", vcpuid));
2175 mtx_lock(&vm->rendezvous_mtx);
2176 if (vm->rendezvous_func != NULL) {
2178 * If a rendezvous is already in progress then we need to
2179 * call the rendezvous handler in case this 'vcpuid' is one
2180 * of the targets of the rendezvous.
2182 RENDEZVOUS_CTR0(vm, vcpuid, "Rendezvous already in progress");
2183 mtx_unlock(&vm->rendezvous_mtx);
2184 vm_handle_rendezvous(vm, vcpuid);
2187 KASSERT(vm->rendezvous_func == NULL, ("vm_smp_rendezvous: previous "
2188 "rendezvous is still in progress"));
2190 RENDEZVOUS_CTR0(vm, vcpuid, "Initiating rendezvous");
2191 vm->rendezvous_req_cpus = dest;
2192 CPU_ZERO(&vm->rendezvous_done_cpus);
2193 vm->rendezvous_arg = arg;
2194 vm_set_rendezvous_func(vm, func);
2195 mtx_unlock(&vm->rendezvous_mtx);
2198 * Wake up any sleeping vcpus and trigger a VM-exit in any running
2199 * vcpus so they handle the rendezvous as soon as possible.
2201 for (i = 0; i < VM_MAXCPU; i++) {
2202 if (CPU_ISSET(i, &dest))
2203 vcpu_notify_event(vm, i, false);
2206 vm_handle_rendezvous(vm, vcpuid);
2210 vm_atpic(struct vm *vm)
2212 return (vm->vatpic);
2216 vm_atpit(struct vm *vm)
2218 return (vm->vatpit);
2222 vm_pmtmr(struct vm *vm)
2225 return (vm->vpmtmr);
2229 vm_rtc(struct vm *vm)
2236 vm_segment_name(int seg)
2238 static enum vm_reg_name seg_names[] = {
2247 KASSERT(seg >= 0 && seg < nitems(seg_names),
2248 ("%s: invalid segment encoding %d", __func__, seg));
2249 return (seg_names[seg]);
2253 vm_copy_teardown(struct vm *vm, int vcpuid, struct vm_copyinfo *copyinfo,
2258 for (idx = 0; idx < num_copyinfo; idx++) {
2259 if (copyinfo[idx].cookie != NULL)
2260 vm_gpa_release(copyinfo[idx].cookie);
2262 bzero(copyinfo, num_copyinfo * sizeof(struct vm_copyinfo));
2266 vm_copy_setup(struct vm *vm, int vcpuid, struct vm_guest_paging *paging,
2267 uint64_t gla, size_t len, int prot, struct vm_copyinfo *copyinfo,
2270 int error, idx, nused;
2271 size_t n, off, remaining;
2275 bzero(copyinfo, sizeof(struct vm_copyinfo) * num_copyinfo);
2279 while (remaining > 0) {
2280 KASSERT(nused < num_copyinfo, ("insufficient vm_copyinfo"));
2281 error = vmm_gla2gpa(vm, vcpuid, paging, gla, prot, &gpa);
2284 off = gpa & PAGE_MASK;
2285 n = min(remaining, PAGE_SIZE - off);
2286 copyinfo[nused].gpa = gpa;
2287 copyinfo[nused].len = n;
2293 for (idx = 0; idx < nused; idx++) {
2294 hva = vm_gpa_hold(vm, copyinfo[idx].gpa, copyinfo[idx].len,
2298 copyinfo[idx].hva = hva;
2299 copyinfo[idx].cookie = cookie;
2303 vm_copy_teardown(vm, vcpuid, copyinfo, num_copyinfo);
2311 vm_copyin(struct vm *vm, int vcpuid, struct vm_copyinfo *copyinfo, void *kaddr,
2320 bcopy(copyinfo[idx].hva, dst, copyinfo[idx].len);
2321 len -= copyinfo[idx].len;
2322 dst += copyinfo[idx].len;
2328 vm_copyout(struct vm *vm, int vcpuid, const void *kaddr,
2329 struct vm_copyinfo *copyinfo, size_t len)
2337 bcopy(src, copyinfo[idx].hva, copyinfo[idx].len);
2338 len -= copyinfo[idx].len;
2339 src += copyinfo[idx].len;
2345 * Return the amount of in-use and wired memory for the VM. Since
2346 * these are global stats, only return the values with for vCPU 0
2348 VMM_STAT_DECLARE(VMM_MEM_RESIDENT);
2349 VMM_STAT_DECLARE(VMM_MEM_WIRED);
2352 vm_get_rescnt(struct vm *vm, int vcpu, struct vmm_stat_type *stat)
2356 vmm_stat_set(vm, vcpu, VMM_MEM_RESIDENT,
2357 PAGE_SIZE * vmspace_resident_count(vm->vmspace));
2362 vm_get_wiredcnt(struct vm *vm, int vcpu, struct vmm_stat_type *stat)
2366 vmm_stat_set(vm, vcpu, VMM_MEM_WIRED,
2367 PAGE_SIZE * pmap_wired_count(vmspace_pmap(vm->vmspace)));
2371 VMM_STAT_FUNC(VMM_MEM_RESIDENT, "Resident memory", vm_get_rescnt);
2372 VMM_STAT_FUNC(VMM_MEM_WIRED, "Wired memory", vm_get_wiredcnt);