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/kernel.h>
39 #include <sys/queue.h>
41 #include <sys/mutex.h>
42 #include <sys/malloc.h>
44 #include <sys/sysctl.h>
45 #include <sys/libkern.h>
46 #include <sys/ioccom.h>
53 #include <vm/vm_map.h>
54 #include <vm/vm_object.h>
56 #include <machine/vmparam.h>
57 #include <machine/vmm.h>
58 #include <machine/vmm_dev.h>
59 #include <machine/vmm_instruction_emul.h>
60 #include <machine/vmm_snapshot.h>
61 #include <x86/apicreg.h>
63 #include "vmm_lapic.h"
67 #include "io/vatpic.h"
68 #include "io/vioapic.h"
76 struct vmmdev_softc *sc;
77 SLIST_ENTRY(devmem_softc) link;
81 struct vm *vm; /* vm instance cookie */
83 SLIST_ENTRY(vmmdev_softc) link;
84 SLIST_HEAD(, devmem_softc) devmem;
87 #define VSC_LINKED 0x01
89 static SLIST_HEAD(, vmmdev_softc) head;
91 static unsigned pr_allow_flag;
92 static struct mtx vmmdev_mtx;
94 static MALLOC_DEFINE(M_VMMDEV, "vmmdev", "vmmdev");
98 static int vmm_priv_check(struct ucred *ucred);
99 static int devmem_create_cdev(const char *vmname, int id, char *devmem);
100 static void devmem_destroy(void *arg);
103 vmm_priv_check(struct ucred *ucred)
107 !(ucred->cr_prison->pr_allow & pr_allow_flag))
114 vcpu_lock_one(struct vmmdev_softc *sc, int vcpu)
118 if (vcpu < 0 || vcpu >= vm_get_maxcpus(sc->vm))
121 error = vcpu_set_state(sc->vm, vcpu, VCPU_FROZEN, true);
126 vcpu_unlock_one(struct vmmdev_softc *sc, int vcpu)
128 enum vcpu_state state;
130 state = vcpu_get_state(sc->vm, vcpu, NULL);
131 if (state != VCPU_FROZEN) {
132 panic("vcpu %s(%d) has invalid state %d", vm_name(sc->vm),
136 vcpu_set_state(sc->vm, vcpu, VCPU_IDLE, false);
140 vcpu_lock_all(struct vmmdev_softc *sc)
145 maxcpus = vm_get_maxcpus(sc->vm);
146 for (vcpu = 0; vcpu < maxcpus; vcpu++) {
147 error = vcpu_lock_one(sc, vcpu);
154 vcpu_unlock_one(sc, vcpu);
161 vcpu_unlock_all(struct vmmdev_softc *sc)
166 maxcpus = vm_get_maxcpus(sc->vm);
167 for (vcpu = 0; vcpu < maxcpus; vcpu++)
168 vcpu_unlock_one(sc, vcpu);
171 static struct vmmdev_softc *
172 vmmdev_lookup(const char *name)
174 struct vmmdev_softc *sc;
176 #ifdef notyet /* XXX kernel is not compiled with invariants */
177 mtx_assert(&vmmdev_mtx, MA_OWNED);
180 SLIST_FOREACH(sc, &head, link) {
181 if (strcmp(name, vm_name(sc->vm)) == 0)
188 static struct vmmdev_softc *
189 vmmdev_lookup2(struct cdev *cdev)
192 return (cdev->si_drv1);
196 vmmdev_rw(struct cdev *cdev, struct uio *uio, int flags)
198 int error, off, c, prot;
199 vm_paddr_t gpa, maxaddr;
201 struct vmmdev_softc *sc;
204 error = vmm_priv_check(curthread->td_ucred);
208 sc = vmmdev_lookup2(cdev);
213 * Get a read lock on the guest memory map by freezing any vcpu.
215 lastcpu = vm_get_maxcpus(sc->vm) - 1;
216 error = vcpu_lock_one(sc, lastcpu);
220 prot = (uio->uio_rw == UIO_WRITE ? VM_PROT_WRITE : VM_PROT_READ);
221 maxaddr = vmm_sysmem_maxaddr(sc->vm);
222 while (uio->uio_resid > 0 && error == 0) {
223 gpa = uio->uio_offset;
224 off = gpa & PAGE_MASK;
225 c = min(uio->uio_resid, PAGE_SIZE - off);
228 * The VM has a hole in its physical memory map. If we want to
229 * use 'dd' to inspect memory beyond the hole we need to
230 * provide bogus data for memory that lies in the hole.
232 * Since this device does not support lseek(2), dd(1) will
233 * read(2) blocks of data to simulate the lseek(2).
235 hpa = vm_gpa_hold(sc->vm, lastcpu, gpa, c,
238 if (uio->uio_rw == UIO_READ && gpa < maxaddr)
239 error = uiomove(__DECONST(void *, zero_region),
244 error = uiomove(hpa, c, uio);
245 vm_gpa_release(cookie);
248 vcpu_unlock_one(sc, lastcpu);
252 CTASSERT(sizeof(((struct vm_memseg *)0)->name) >= VM_MAX_SUFFIXLEN + 1);
255 get_memseg(struct vmmdev_softc *sc, struct vm_memseg *mseg, size_t len)
257 struct devmem_softc *dsc;
261 error = vm_get_memseg(sc->vm, mseg->segid, &mseg->len, &sysmem, NULL);
262 if (error || mseg->len == 0)
266 SLIST_FOREACH(dsc, &sc->devmem, link) {
267 if (dsc->segid == mseg->segid)
270 KASSERT(dsc != NULL, ("%s: devmem segment %d not found",
271 __func__, mseg->segid));
272 error = copystr(dsc->name, mseg->name, len, NULL);
274 bzero(mseg->name, len);
281 alloc_memseg(struct vmmdev_softc *sc, struct vm_memseg *mseg, size_t len)
292 * The allocation is lengthened by 1 to hold a terminating NUL. It'll
293 * by stripped off when devfs processes the full string.
295 if (VM_MEMSEG_NAME(mseg)) {
297 name = malloc(len, M_VMMDEV, M_WAITOK);
298 error = copystr(mseg->name, name, len, NULL);
303 error = vm_alloc_memseg(sc->vm, mseg->segid, mseg->len, sysmem);
307 if (VM_MEMSEG_NAME(mseg)) {
308 error = devmem_create_cdev(vm_name(sc->vm), mseg->segid, name);
310 vm_free_memseg(sc->vm, mseg->segid);
312 name = NULL; /* freed when 'cdev' is destroyed */
315 free(name, M_VMMDEV);
320 vm_get_register_set(struct vm *vm, int vcpu, unsigned int count, int *regnum,
326 for (i = 0; i < count; i++) {
327 error = vm_get_register(vm, vcpu, regnum[i], ®val[i]);
335 vm_set_register_set(struct vm *vm, int vcpu, unsigned int count, int *regnum,
341 for (i = 0; i < count; i++) {
342 error = vm_set_register(vm, vcpu, regnum[i], regval[i]);
350 vmmdev_ioctl(struct cdev *cdev, u_long cmd, caddr_t data, int fflag,
353 int error, vcpu, state_changed, size;
355 struct vmmdev_softc *sc;
356 struct vm_register *vmreg;
357 struct vm_seg_desc *vmsegdesc;
358 struct vm_register_set *vmregset;
359 struct vm_run *vmrun;
360 struct vm_exception *vmexc;
361 struct vm_lapic_irq *vmirq;
362 struct vm_lapic_msi *vmmsi;
363 struct vm_ioapic_irq *ioapic_irq;
364 struct vm_isa_irq *isa_irq;
365 struct vm_isa_irq_trigger *isa_irq_trigger;
366 struct vm_capability *vmcap;
367 struct vm_pptdev *pptdev;
368 struct vm_pptdev_mmio *pptmmio;
369 struct vm_pptdev_msi *pptmsi;
370 struct vm_pptdev_msix *pptmsix;
371 struct vm_nmi *vmnmi;
372 struct vm_stats *vmstats;
373 struct vm_stat_desc *statdesc;
374 struct vm_x2apic *x2apic;
375 struct vm_gpa_pte *gpapte;
376 struct vm_suspend *vmsuspend;
377 struct vm_gla2gpa *gg;
378 struct vm_activate_cpu *vac;
379 struct vm_cpuset *vm_cpuset;
380 struct vm_intinfo *vmii;
381 struct vm_rtc_time *rtctime;
382 struct vm_rtc_data *rtcdata;
383 struct vm_memmap *mm;
384 struct vm_cpu_topology *topology;
385 struct vm_readwrite_kernemu_device *kernemu;
388 #ifdef BHYVE_SNAPSHOT
389 struct vm_snapshot_meta *snapshot_meta;
392 error = vmm_priv_check(curthread->td_ucred);
396 sc = vmmdev_lookup2(cdev);
404 * Some VMM ioctls can operate only on vcpus that are not running.
408 case VM_GET_REGISTER:
409 case VM_SET_REGISTER:
410 case VM_GET_SEGMENT_DESCRIPTOR:
411 case VM_SET_SEGMENT_DESCRIPTOR:
412 case VM_GET_REGISTER_SET:
413 case VM_SET_REGISTER_SET:
414 case VM_INJECT_EXCEPTION:
415 case VM_GET_CAPABILITY:
416 case VM_SET_CAPABILITY:
419 case VM_SET_X2APIC_STATE:
421 case VM_GLA2GPA_NOFAULT:
422 case VM_ACTIVATE_CPU:
425 case VM_RESTART_INSTRUCTION:
427 * XXX fragile, handle with care
428 * Assumes that the first field of the ioctl data is the vcpu.
431 error = vcpu_lock_one(sc, vcpu);
437 case VM_MAP_PPTDEV_MMIO:
439 case VM_UNBIND_PPTDEV:
440 #ifdef COMPAT_FREEBSD12
441 case VM_ALLOC_MEMSEG_FBSD12:
443 case VM_ALLOC_MEMSEG:
447 * ioctls that operate on the entire virtual machine must
448 * prevent all vcpus from running.
450 error = vcpu_lock_all(sc);
456 #ifdef COMPAT_FREEBSD12
457 case VM_GET_MEMSEG_FBSD12:
460 case VM_MMAP_GETNEXT:
462 * Lock a vcpu to make sure that the memory map cannot be
463 * modified while it is being inspected.
465 vcpu = vm_get_maxcpus(sc->vm) - 1;
466 error = vcpu_lock_one(sc, vcpu);
478 vmrun = (struct vm_run *)data;
479 error = vm_run(sc->vm, vmrun);
482 vmsuspend = (struct vm_suspend *)data;
483 error = vm_suspend(sc->vm, vmsuspend->how);
486 error = vm_reinit(sc->vm);
489 statdesc = (struct vm_stat_desc *)data;
490 error = vmm_stat_desc_copy(statdesc->index,
491 statdesc->desc, sizeof(statdesc->desc));
495 CTASSERT(MAX_VM_STATS >= MAX_VMM_STAT_ELEMS);
496 vmstats = (struct vm_stats *)data;
497 getmicrotime(&vmstats->tv);
498 error = vmm_stat_copy(sc->vm, vmstats->cpuid,
499 &vmstats->num_entries, vmstats->statbuf);
503 pptmsi = (struct vm_pptdev_msi *)data;
504 error = ppt_setup_msi(sc->vm, pptmsi->vcpu,
505 pptmsi->bus, pptmsi->slot, pptmsi->func,
506 pptmsi->addr, pptmsi->msg,
510 pptmsix = (struct vm_pptdev_msix *)data;
511 error = ppt_setup_msix(sc->vm, pptmsix->vcpu,
512 pptmsix->bus, pptmsix->slot,
513 pptmsix->func, pptmsix->idx,
514 pptmsix->addr, pptmsix->msg,
515 pptmsix->vector_control);
517 case VM_MAP_PPTDEV_MMIO:
518 pptmmio = (struct vm_pptdev_mmio *)data;
519 error = ppt_map_mmio(sc->vm, pptmmio->bus, pptmmio->slot,
520 pptmmio->func, pptmmio->gpa, pptmmio->len,
524 pptdev = (struct vm_pptdev *)data;
525 error = vm_assign_pptdev(sc->vm, pptdev->bus, pptdev->slot,
528 case VM_UNBIND_PPTDEV:
529 pptdev = (struct vm_pptdev *)data;
530 error = vm_unassign_pptdev(sc->vm, pptdev->bus, pptdev->slot,
533 case VM_INJECT_EXCEPTION:
534 vmexc = (struct vm_exception *)data;
535 error = vm_inject_exception(sc->vm, vmexc->cpuid,
536 vmexc->vector, vmexc->error_code_valid, vmexc->error_code,
537 vmexc->restart_instruction);
540 vmnmi = (struct vm_nmi *)data;
541 error = vm_inject_nmi(sc->vm, vmnmi->cpuid);
544 vmirq = (struct vm_lapic_irq *)data;
545 error = lapic_intr_edge(sc->vm, vmirq->cpuid, vmirq->vector);
547 case VM_LAPIC_LOCAL_IRQ:
548 vmirq = (struct vm_lapic_irq *)data;
549 error = lapic_set_local_intr(sc->vm, vmirq->cpuid,
553 vmmsi = (struct vm_lapic_msi *)data;
554 error = lapic_intr_msi(sc->vm, vmmsi->addr, vmmsi->msg);
556 case VM_IOAPIC_ASSERT_IRQ:
557 ioapic_irq = (struct vm_ioapic_irq *)data;
558 error = vioapic_assert_irq(sc->vm, ioapic_irq->irq);
560 case VM_IOAPIC_DEASSERT_IRQ:
561 ioapic_irq = (struct vm_ioapic_irq *)data;
562 error = vioapic_deassert_irq(sc->vm, ioapic_irq->irq);
564 case VM_IOAPIC_PULSE_IRQ:
565 ioapic_irq = (struct vm_ioapic_irq *)data;
566 error = vioapic_pulse_irq(sc->vm, ioapic_irq->irq);
568 case VM_IOAPIC_PINCOUNT:
569 *(int *)data = vioapic_pincount(sc->vm);
571 case VM_SET_KERNEMU_DEV:
572 case VM_GET_KERNEMU_DEV: {
573 mem_region_write_t mwrite;
574 mem_region_read_t mread;
577 kernemu = (void *)data;
579 if (kernemu->access_width > 0)
580 size = (1u << kernemu->access_width);
584 if (kernemu->gpa >= DEFAULT_APIC_BASE && kernemu->gpa < DEFAULT_APIC_BASE + PAGE_SIZE) {
585 mread = lapic_mmio_read;
586 mwrite = lapic_mmio_write;
587 } else if (kernemu->gpa >= VIOAPIC_BASE && kernemu->gpa < VIOAPIC_BASE + VIOAPIC_SIZE) {
588 mread = vioapic_mmio_read;
589 mwrite = vioapic_mmio_write;
590 } else if (kernemu->gpa >= VHPET_BASE && kernemu->gpa < VHPET_BASE + VHPET_SIZE) {
591 mread = vhpet_mmio_read;
592 mwrite = vhpet_mmio_write;
598 if (cmd == VM_SET_KERNEMU_DEV)
599 error = mwrite(sc->vm, kernemu->vcpuid, kernemu->gpa,
600 kernemu->value, size, &arg);
602 error = mread(sc->vm, kernemu->vcpuid, kernemu->gpa,
603 &kernemu->value, size, &arg);
606 case VM_ISA_ASSERT_IRQ:
607 isa_irq = (struct vm_isa_irq *)data;
608 error = vatpic_assert_irq(sc->vm, isa_irq->atpic_irq);
609 if (error == 0 && isa_irq->ioapic_irq != -1)
610 error = vioapic_assert_irq(sc->vm,
611 isa_irq->ioapic_irq);
613 case VM_ISA_DEASSERT_IRQ:
614 isa_irq = (struct vm_isa_irq *)data;
615 error = vatpic_deassert_irq(sc->vm, isa_irq->atpic_irq);
616 if (error == 0 && isa_irq->ioapic_irq != -1)
617 error = vioapic_deassert_irq(sc->vm,
618 isa_irq->ioapic_irq);
620 case VM_ISA_PULSE_IRQ:
621 isa_irq = (struct vm_isa_irq *)data;
622 error = vatpic_pulse_irq(sc->vm, isa_irq->atpic_irq);
623 if (error == 0 && isa_irq->ioapic_irq != -1)
624 error = vioapic_pulse_irq(sc->vm, isa_irq->ioapic_irq);
626 case VM_ISA_SET_IRQ_TRIGGER:
627 isa_irq_trigger = (struct vm_isa_irq_trigger *)data;
628 error = vatpic_set_irq_trigger(sc->vm,
629 isa_irq_trigger->atpic_irq, isa_irq_trigger->trigger);
631 case VM_MMAP_GETNEXT:
632 mm = (struct vm_memmap *)data;
633 error = vm_mmap_getnext(sc->vm, &mm->gpa, &mm->segid,
634 &mm->segoff, &mm->len, &mm->prot, &mm->flags);
637 mm = (struct vm_memmap *)data;
638 error = vm_mmap_memseg(sc->vm, mm->gpa, mm->segid, mm->segoff,
639 mm->len, mm->prot, mm->flags);
641 #ifdef COMPAT_FREEBSD12
642 case VM_ALLOC_MEMSEG_FBSD12:
643 error = alloc_memseg(sc, (struct vm_memseg *)data,
644 sizeof(((struct vm_memseg_fbsd12 *)0)->name));
647 case VM_ALLOC_MEMSEG:
648 error = alloc_memseg(sc, (struct vm_memseg *)data,
649 sizeof(((struct vm_memseg *)0)->name));
651 #ifdef COMPAT_FREEBSD12
652 case VM_GET_MEMSEG_FBSD12:
653 error = get_memseg(sc, (struct vm_memseg *)data,
654 sizeof(((struct vm_memseg_fbsd12 *)0)->name));
658 error = get_memseg(sc, (struct vm_memseg *)data,
659 sizeof(((struct vm_memseg *)0)->name));
661 case VM_GET_REGISTER:
662 vmreg = (struct vm_register *)data;
663 error = vm_get_register(sc->vm, vmreg->cpuid, vmreg->regnum,
666 case VM_SET_REGISTER:
667 vmreg = (struct vm_register *)data;
668 error = vm_set_register(sc->vm, vmreg->cpuid, vmreg->regnum,
671 case VM_SET_SEGMENT_DESCRIPTOR:
672 vmsegdesc = (struct vm_seg_desc *)data;
673 error = vm_set_seg_desc(sc->vm, vmsegdesc->cpuid,
677 case VM_GET_SEGMENT_DESCRIPTOR:
678 vmsegdesc = (struct vm_seg_desc *)data;
679 error = vm_get_seg_desc(sc->vm, vmsegdesc->cpuid,
683 case VM_GET_REGISTER_SET:
684 vmregset = (struct vm_register_set *)data;
685 if (vmregset->count > VM_REG_LAST) {
689 regvals = malloc(sizeof(regvals[0]) * vmregset->count, M_VMMDEV,
691 regnums = malloc(sizeof(regnums[0]) * vmregset->count, M_VMMDEV,
693 error = copyin(vmregset->regnums, regnums, sizeof(regnums[0]) *
696 error = vm_get_register_set(sc->vm, vmregset->cpuid,
697 vmregset->count, regnums, regvals);
699 error = copyout(regvals, vmregset->regvals,
700 sizeof(regvals[0]) * vmregset->count);
701 free(regvals, M_VMMDEV);
702 free(regnums, M_VMMDEV);
704 case VM_SET_REGISTER_SET:
705 vmregset = (struct vm_register_set *)data;
706 if (vmregset->count > VM_REG_LAST) {
710 regvals = malloc(sizeof(regvals[0]) * vmregset->count, M_VMMDEV,
712 regnums = malloc(sizeof(regnums[0]) * vmregset->count, M_VMMDEV,
714 error = copyin(vmregset->regnums, regnums, sizeof(regnums[0]) *
717 error = copyin(vmregset->regvals, regvals,
718 sizeof(regvals[0]) * vmregset->count);
720 error = vm_set_register_set(sc->vm, vmregset->cpuid,
721 vmregset->count, regnums, regvals);
722 free(regvals, M_VMMDEV);
723 free(regnums, M_VMMDEV);
725 case VM_GET_CAPABILITY:
726 vmcap = (struct vm_capability *)data;
727 error = vm_get_capability(sc->vm, vmcap->cpuid,
731 case VM_SET_CAPABILITY:
732 vmcap = (struct vm_capability *)data;
733 error = vm_set_capability(sc->vm, vmcap->cpuid,
737 case VM_SET_X2APIC_STATE:
738 x2apic = (struct vm_x2apic *)data;
739 error = vm_set_x2apic_state(sc->vm,
740 x2apic->cpuid, x2apic->state);
742 case VM_GET_X2APIC_STATE:
743 x2apic = (struct vm_x2apic *)data;
744 error = vm_get_x2apic_state(sc->vm,
745 x2apic->cpuid, &x2apic->state);
747 case VM_GET_GPA_PMAP:
748 gpapte = (struct vm_gpa_pte *)data;
749 pmap_get_mapping(vmspace_pmap(vm_get_vmspace(sc->vm)),
750 gpapte->gpa, gpapte->pte, &gpapte->ptenum);
753 case VM_GET_HPET_CAPABILITIES:
754 error = vhpet_getcap((struct vm_hpet_cap *)data);
757 CTASSERT(PROT_READ == VM_PROT_READ);
758 CTASSERT(PROT_WRITE == VM_PROT_WRITE);
759 CTASSERT(PROT_EXEC == VM_PROT_EXECUTE);
760 gg = (struct vm_gla2gpa *)data;
761 error = vm_gla2gpa(sc->vm, gg->vcpuid, &gg->paging, gg->gla,
762 gg->prot, &gg->gpa, &gg->fault);
763 KASSERT(error == 0 || error == EFAULT,
764 ("%s: vm_gla2gpa unknown error %d", __func__, error));
767 case VM_GLA2GPA_NOFAULT:
768 gg = (struct vm_gla2gpa *)data;
769 error = vm_gla2gpa_nofault(sc->vm, gg->vcpuid, &gg->paging,
770 gg->gla, gg->prot, &gg->gpa, &gg->fault);
771 KASSERT(error == 0 || error == EFAULT,
772 ("%s: vm_gla2gpa unknown error %d", __func__, error));
774 case VM_ACTIVATE_CPU:
775 vac = (struct vm_activate_cpu *)data;
776 error = vm_activate_cpu(sc->vm, vac->vcpuid);
780 vm_cpuset = (struct vm_cpuset *)data;
781 size = vm_cpuset->cpusetsize;
782 if (size < sizeof(cpuset_t) || size > CPU_MAXSIZE / NBBY) {
786 cpuset = malloc(size, M_TEMP, M_WAITOK | M_ZERO);
787 if (vm_cpuset->which == VM_ACTIVE_CPUS)
788 *cpuset = vm_active_cpus(sc->vm);
789 else if (vm_cpuset->which == VM_SUSPENDED_CPUS)
790 *cpuset = vm_suspended_cpus(sc->vm);
791 else if (vm_cpuset->which == VM_DEBUG_CPUS)
792 *cpuset = vm_debug_cpus(sc->vm);
796 error = copyout(cpuset, vm_cpuset->cpus, size);
797 free(cpuset, M_TEMP);
800 vac = (struct vm_activate_cpu *)data;
801 error = vm_suspend_cpu(sc->vm, vac->vcpuid);
804 vac = (struct vm_activate_cpu *)data;
805 error = vm_resume_cpu(sc->vm, vac->vcpuid);
808 vmii = (struct vm_intinfo *)data;
809 error = vm_exit_intinfo(sc->vm, vmii->vcpuid, vmii->info1);
812 vmii = (struct vm_intinfo *)data;
813 error = vm_get_intinfo(sc->vm, vmii->vcpuid, &vmii->info1,
817 rtcdata = (struct vm_rtc_data *)data;
818 error = vrtc_nvram_write(sc->vm, rtcdata->offset,
822 rtcdata = (struct vm_rtc_data *)data;
823 error = vrtc_nvram_read(sc->vm, rtcdata->offset,
827 rtctime = (struct vm_rtc_time *)data;
828 error = vrtc_set_time(sc->vm, rtctime->secs);
832 rtctime = (struct vm_rtc_time *)data;
833 rtctime->secs = vrtc_get_time(sc->vm);
835 case VM_RESTART_INSTRUCTION:
836 error = vm_restart_instruction(sc->vm, vcpu);
838 case VM_SET_TOPOLOGY:
839 topology = (struct vm_cpu_topology *)data;
840 error = vm_set_topology(sc->vm, topology->sockets,
841 topology->cores, topology->threads, topology->maxcpus);
843 case VM_GET_TOPOLOGY:
844 topology = (struct vm_cpu_topology *)data;
845 vm_get_topology(sc->vm, &topology->sockets, &topology->cores,
846 &topology->threads, &topology->maxcpus);
849 #ifdef BHYVE_SNAPSHOT
850 case VM_SNAPSHOT_REQ:
851 snapshot_meta = (struct vm_snapshot_meta *)data;
852 error = vm_snapshot_req(sc->vm, snapshot_meta);
854 case VM_RESTORE_TIME:
855 error = vm_restore_time(sc->vm);
863 if (state_changed == 1)
864 vcpu_unlock_one(sc, vcpu);
865 else if (state_changed == 2)
870 * Make sure that no handler returns a kernel-internal
871 * error value to userspace.
873 KASSERT(error == ERESTART || error >= 0,
874 ("vmmdev_ioctl: invalid error return %d", error));
879 vmmdev_mmap_single(struct cdev *cdev, vm_ooffset_t *offset, vm_size_t mapsize,
880 struct vm_object **objp, int nprot)
882 struct vmmdev_softc *sc;
885 vm_ooffset_t segoff, first, last;
886 int error, found, segid;
890 error = vmm_priv_check(curthread->td_ucred);
895 last = first + mapsize;
896 if ((nprot & PROT_EXEC) || first < 0 || first >= last)
899 sc = vmmdev_lookup2(cdev);
901 /* virtual machine is in the process of being created */
906 * Get a read lock on the guest memory map by freezing any vcpu.
908 lastcpu = vm_get_maxcpus(sc->vm) - 1;
909 error = vcpu_lock_one(sc, lastcpu);
916 error = vm_mmap_getnext(sc->vm, &gpa, &segid, &segoff, &len,
921 if (first >= gpa && last <= gpa + len)
928 error = vm_get_memseg(sc->vm, segid, &len, &sysmem, objp);
929 KASSERT(error == 0 && *objp != NULL,
930 ("%s: invalid memory segment %d", __func__, segid));
932 vm_object_reference(*objp);
933 *offset = segoff + (first - gpa);
938 vcpu_unlock_one(sc, lastcpu);
943 vmmdev_destroy(void *arg)
945 struct vmmdev_softc *sc = arg;
946 struct devmem_softc *dsc;
949 error = vcpu_lock_all(sc);
950 KASSERT(error == 0, ("%s: error %d freezing vcpus", __func__, error));
952 while ((dsc = SLIST_FIRST(&sc->devmem)) != NULL) {
953 KASSERT(dsc->cdev == NULL, ("%s: devmem not free", __func__));
954 SLIST_REMOVE_HEAD(&sc->devmem, link);
955 free(dsc->name, M_VMMDEV);
959 if (sc->cdev != NULL)
960 destroy_dev(sc->cdev);
965 if ((sc->flags & VSC_LINKED) != 0) {
966 mtx_lock(&vmmdev_mtx);
967 SLIST_REMOVE(&head, sc, vmmdev_softc, link);
968 mtx_unlock(&vmmdev_mtx);
975 sysctl_vmm_destroy(SYSCTL_HANDLER_ARGS)
977 struct devmem_softc *dsc;
978 struct vmmdev_softc *sc;
983 error = vmm_priv_check(req->td->td_ucred);
987 buflen = VM_MAX_NAMELEN + 1;
988 buf = malloc(buflen, M_VMMDEV, M_WAITOK | M_ZERO);
989 strlcpy(buf, "beavis", buflen);
990 error = sysctl_handle_string(oidp, buf, buflen, req);
991 if (error != 0 || req->newptr == NULL)
994 mtx_lock(&vmmdev_mtx);
995 sc = vmmdev_lookup(buf);
996 if (sc == NULL || sc->cdev == NULL) {
997 mtx_unlock(&vmmdev_mtx);
1003 * The 'cdev' will be destroyed asynchronously when 'si_threadcount'
1004 * goes down to 0 so we should not do it again in the callback.
1006 * Setting 'sc->cdev' to NULL is also used to indicate that the VM
1007 * is scheduled for destruction.
1011 mtx_unlock(&vmmdev_mtx);
1014 * Schedule all cdevs to be destroyed:
1016 * - any new operations on the 'cdev' will return an error (ENXIO).
1018 * - when the 'si_threadcount' dwindles down to zero the 'cdev' will
1019 * be destroyed and the callback will be invoked in a taskqueue
1022 * - the 'devmem' cdevs are destroyed before the virtual machine 'cdev'
1024 SLIST_FOREACH(dsc, &sc->devmem, link) {
1025 KASSERT(dsc->cdev != NULL, ("devmem cdev already destroyed"));
1026 destroy_dev_sched_cb(dsc->cdev, devmem_destroy, dsc);
1028 destroy_dev_sched_cb(cdev, vmmdev_destroy, sc);
1032 free(buf, M_VMMDEV);
1035 SYSCTL_PROC(_hw_vmm, OID_AUTO, destroy,
1036 CTLTYPE_STRING | CTLFLAG_RW | CTLFLAG_PRISON | CTLFLAG_MPSAFE,
1037 NULL, 0, sysctl_vmm_destroy, "A",
1040 static struct cdevsw vmmdevsw = {
1042 .d_version = D_VERSION,
1043 .d_ioctl = vmmdev_ioctl,
1044 .d_mmap_single = vmmdev_mmap_single,
1045 .d_read = vmmdev_rw,
1046 .d_write = vmmdev_rw,
1050 sysctl_vmm_create(SYSCTL_HANDLER_ARGS)
1054 struct vmmdev_softc *sc, *sc2;
1058 error = vmm_priv_check(req->td->td_ucred);
1062 buflen = VM_MAX_NAMELEN + 1;
1063 buf = malloc(buflen, M_VMMDEV, M_WAITOK | M_ZERO);
1064 strlcpy(buf, "beavis", buflen);
1065 error = sysctl_handle_string(oidp, buf, buflen, req);
1066 if (error != 0 || req->newptr == NULL)
1069 mtx_lock(&vmmdev_mtx);
1070 sc = vmmdev_lookup(buf);
1071 mtx_unlock(&vmmdev_mtx);
1077 error = vm_create(buf, &vm);
1081 sc = malloc(sizeof(struct vmmdev_softc), M_VMMDEV, M_WAITOK | M_ZERO);
1083 SLIST_INIT(&sc->devmem);
1086 * Lookup the name again just in case somebody sneaked in when we
1089 mtx_lock(&vmmdev_mtx);
1090 sc2 = vmmdev_lookup(buf);
1092 SLIST_INSERT_HEAD(&head, sc, link);
1093 sc->flags |= VSC_LINKED;
1095 mtx_unlock(&vmmdev_mtx);
1103 error = make_dev_p(MAKEDEV_CHECKNAME, &cdev, &vmmdevsw, NULL,
1104 UID_ROOT, GID_WHEEL, 0600, "vmm/%s", buf);
1110 mtx_lock(&vmmdev_mtx);
1112 sc->cdev->si_drv1 = sc;
1113 mtx_unlock(&vmmdev_mtx);
1116 free(buf, M_VMMDEV);
1119 SYSCTL_PROC(_hw_vmm, OID_AUTO, create,
1120 CTLTYPE_STRING | CTLFLAG_RW | CTLFLAG_PRISON | CTLFLAG_MPSAFE,
1121 NULL, 0, sysctl_vmm_create, "A",
1127 mtx_init(&vmmdev_mtx, "vmm device mutex", NULL, MTX_DEF);
1128 pr_allow_flag = prison_add_allow(NULL, "vmm", NULL,
1129 "Allow use of vmm in a jail.");
1133 vmmdev_cleanup(void)
1137 if (SLIST_EMPTY(&head))
1146 devmem_mmap_single(struct cdev *cdev, vm_ooffset_t *offset, vm_size_t len,
1147 struct vm_object **objp, int nprot)
1149 struct devmem_softc *dsc;
1150 vm_ooffset_t first, last;
1156 dsc = cdev->si_drv1;
1158 /* 'cdev' has been created but is not ready for use */
1163 last = *offset + len;
1164 if ((nprot & PROT_EXEC) || first < 0 || first >= last)
1167 lastcpu = vm_get_maxcpus(dsc->sc->vm) - 1;
1168 error = vcpu_lock_one(dsc->sc, lastcpu);
1172 error = vm_get_memseg(dsc->sc->vm, dsc->segid, &seglen, &sysmem, objp);
1173 KASSERT(error == 0 && !sysmem && *objp != NULL,
1174 ("%s: invalid devmem segment %d", __func__, dsc->segid));
1176 vcpu_unlock_one(dsc->sc, lastcpu);
1178 if (seglen >= last) {
1179 vm_object_reference(*objp);
1186 static struct cdevsw devmemsw = {
1188 .d_version = D_VERSION,
1189 .d_mmap_single = devmem_mmap_single,
1193 devmem_create_cdev(const char *vmname, int segid, char *devname)
1195 struct devmem_softc *dsc;
1196 struct vmmdev_softc *sc;
1200 error = make_dev_p(MAKEDEV_CHECKNAME, &cdev, &devmemsw, NULL,
1201 UID_ROOT, GID_WHEEL, 0600, "vmm.io/%s.%s", vmname, devname);
1205 dsc = malloc(sizeof(struct devmem_softc), M_VMMDEV, M_WAITOK | M_ZERO);
1207 mtx_lock(&vmmdev_mtx);
1208 sc = vmmdev_lookup(vmname);
1209 KASSERT(sc != NULL, ("%s: vm %s softc not found", __func__, vmname));
1210 if (sc->cdev == NULL) {
1211 /* virtual machine is being created or destroyed */
1212 mtx_unlock(&vmmdev_mtx);
1213 free(dsc, M_VMMDEV);
1214 destroy_dev_sched_cb(cdev, NULL, 0);
1219 dsc->name = devname;
1222 SLIST_INSERT_HEAD(&sc->devmem, dsc, link);
1223 mtx_unlock(&vmmdev_mtx);
1225 /* The 'cdev' is ready for use after 'si_drv1' is initialized */
1226 cdev->si_drv1 = dsc;
1231 devmem_destroy(void *arg)
1233 struct devmem_softc *dsc = arg;
1235 KASSERT(dsc->cdev, ("%s: devmem cdev already destroyed", __func__));