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 <sys/param.h>
35 #include <sys/kernel.h>
37 #include <sys/queue.h>
39 #include <sys/mutex.h>
40 #include <sys/malloc.h>
42 #include <sys/sysctl.h>
43 #include <sys/libkern.h>
44 #include <sys/ioccom.h>
51 #include <vm/vm_map.h>
52 #include <vm/vm_object.h>
54 #include <machine/vmparam.h>
55 #include <machine/vmm.h>
56 #include <machine/vmm_instruction_emul.h>
57 #include <machine/vmm_dev.h>
59 #include "vmm_lapic.h"
63 #include "io/vatpic.h"
64 #include "io/vioapic.h"
72 struct vmmdev_softc *sc;
73 SLIST_ENTRY(devmem_softc) link;
77 struct vm *vm; /* vm instance cookie */
79 SLIST_ENTRY(vmmdev_softc) link;
80 SLIST_HEAD(, devmem_softc) devmem;
83 #define VSC_LINKED 0x01
85 static SLIST_HEAD(, vmmdev_softc) head;
87 static unsigned pr_allow_flag;
88 static struct mtx vmmdev_mtx;
90 static MALLOC_DEFINE(M_VMMDEV, "vmmdev", "vmmdev");
94 static int vmm_priv_check(struct ucred *ucred);
95 static int devmem_create_cdev(const char *vmname, int id, char *devmem);
96 static void devmem_destroy(void *arg);
99 vmm_priv_check(struct ucred *ucred)
103 !(ucred->cr_prison->pr_allow & pr_allow_flag))
110 vcpu_lock_one(struct vmmdev_softc *sc, int vcpu)
114 if (vcpu < 0 || vcpu >= VM_MAXCPU)
117 error = vcpu_set_state(sc->vm, vcpu, VCPU_FROZEN, true);
122 vcpu_unlock_one(struct vmmdev_softc *sc, int vcpu)
124 enum vcpu_state state;
126 state = vcpu_get_state(sc->vm, vcpu, NULL);
127 if (state != VCPU_FROZEN) {
128 panic("vcpu %s(%d) has invalid state %d", vm_name(sc->vm),
132 vcpu_set_state(sc->vm, vcpu, VCPU_IDLE, false);
136 vcpu_lock_all(struct vmmdev_softc *sc)
140 for (vcpu = 0; vcpu < VM_MAXCPU; vcpu++) {
141 error = vcpu_lock_one(sc, vcpu);
148 vcpu_unlock_one(sc, vcpu);
155 vcpu_unlock_all(struct vmmdev_softc *sc)
159 for (vcpu = 0; vcpu < VM_MAXCPU; vcpu++)
160 vcpu_unlock_one(sc, vcpu);
163 static struct vmmdev_softc *
164 vmmdev_lookup(const char *name)
166 struct vmmdev_softc *sc;
168 #ifdef notyet /* XXX kernel is not compiled with invariants */
169 mtx_assert(&vmmdev_mtx, MA_OWNED);
172 SLIST_FOREACH(sc, &head, link) {
173 if (strcmp(name, vm_name(sc->vm)) == 0)
180 static struct vmmdev_softc *
181 vmmdev_lookup2(struct cdev *cdev)
184 return (cdev->si_drv1);
188 vmmdev_rw(struct cdev *cdev, struct uio *uio, int flags)
190 int error, off, c, prot;
191 vm_paddr_t gpa, maxaddr;
193 struct vmmdev_softc *sc;
195 error = vmm_priv_check(curthread->td_ucred);
199 sc = vmmdev_lookup2(cdev);
204 * Get a read lock on the guest memory map by freezing any vcpu.
206 error = vcpu_lock_one(sc, VM_MAXCPU - 1);
210 prot = (uio->uio_rw == UIO_WRITE ? VM_PROT_WRITE : VM_PROT_READ);
211 maxaddr = vmm_sysmem_maxaddr(sc->vm);
212 while (uio->uio_resid > 0 && error == 0) {
213 gpa = uio->uio_offset;
214 off = gpa & PAGE_MASK;
215 c = min(uio->uio_resid, PAGE_SIZE - off);
218 * The VM has a hole in its physical memory map. If we want to
219 * use 'dd' to inspect memory beyond the hole we need to
220 * provide bogus data for memory that lies in the hole.
222 * Since this device does not support lseek(2), dd(1) will
223 * read(2) blocks of data to simulate the lseek(2).
225 hpa = vm_gpa_hold(sc->vm, VM_MAXCPU - 1, gpa, c, prot, &cookie);
227 if (uio->uio_rw == UIO_READ && gpa < maxaddr)
228 error = uiomove(__DECONST(void *, zero_region),
233 error = uiomove(hpa, c, uio);
234 vm_gpa_release(cookie);
237 vcpu_unlock_one(sc, VM_MAXCPU - 1);
241 CTASSERT(sizeof(((struct vm_memseg *)0)->name) >= SPECNAMELEN + 1);
244 get_memseg(struct vmmdev_softc *sc, struct vm_memseg *mseg)
246 struct devmem_softc *dsc;
250 error = vm_get_memseg(sc->vm, mseg->segid, &mseg->len, &sysmem, NULL);
251 if (error || mseg->len == 0)
255 SLIST_FOREACH(dsc, &sc->devmem, link) {
256 if (dsc->segid == mseg->segid)
259 KASSERT(dsc != NULL, ("%s: devmem segment %d not found",
260 __func__, mseg->segid));
261 error = copystr(dsc->name, mseg->name, SPECNAMELEN + 1, NULL);
263 bzero(mseg->name, sizeof(mseg->name));
270 alloc_memseg(struct vmmdev_softc *sc, struct vm_memseg *mseg)
280 if (VM_MEMSEG_NAME(mseg)) {
282 name = malloc(SPECNAMELEN + 1, M_VMMDEV, M_WAITOK);
283 error = copystr(mseg->name, name, SPECNAMELEN + 1, 0);
288 error = vm_alloc_memseg(sc->vm, mseg->segid, mseg->len, sysmem);
292 if (VM_MEMSEG_NAME(mseg)) {
293 error = devmem_create_cdev(vm_name(sc->vm), mseg->segid, name);
295 vm_free_memseg(sc->vm, mseg->segid);
297 name = NULL; /* freed when 'cdev' is destroyed */
300 free(name, M_VMMDEV);
305 vm_get_register_set(struct vm *vm, int vcpu, unsigned int count, int *regnum,
311 for (i = 0; i < count; i++) {
312 error = vm_get_register(vm, vcpu, regnum[i], ®val[i]);
320 vm_set_register_set(struct vm *vm, int vcpu, unsigned int count, int *regnum,
326 for (i = 0; i < count; i++) {
327 error = vm_set_register(vm, vcpu, regnum[i], regval[i]);
335 vmmdev_ioctl(struct cdev *cdev, u_long cmd, caddr_t data, int fflag,
338 int error, vcpu, state_changed, size;
340 struct vmmdev_softc *sc;
341 struct vm_register *vmreg;
342 struct vm_seg_desc *vmsegdesc;
343 struct vm_register_set *vmregset;
344 struct vm_run *vmrun;
345 struct vm_exception *vmexc;
346 struct vm_lapic_irq *vmirq;
347 struct vm_lapic_msi *vmmsi;
348 struct vm_ioapic_irq *ioapic_irq;
349 struct vm_isa_irq *isa_irq;
350 struct vm_isa_irq_trigger *isa_irq_trigger;
351 struct vm_capability *vmcap;
352 struct vm_pptdev *pptdev;
353 struct vm_pptdev_mmio *pptmmio;
354 struct vm_pptdev_msi *pptmsi;
355 struct vm_pptdev_msix *pptmsix;
356 struct vm_nmi *vmnmi;
357 struct vm_stats *vmstats;
358 struct vm_stat_desc *statdesc;
359 struct vm_x2apic *x2apic;
360 struct vm_gpa_pte *gpapte;
361 struct vm_suspend *vmsuspend;
362 struct vm_gla2gpa *gg;
363 struct vm_activate_cpu *vac;
364 struct vm_cpuset *vm_cpuset;
365 struct vm_intinfo *vmii;
366 struct vm_rtc_time *rtctime;
367 struct vm_rtc_data *rtcdata;
368 struct vm_memmap *mm;
369 struct vm_cpu_topology *topology;
373 error = vmm_priv_check(curthread->td_ucred);
377 sc = vmmdev_lookup2(cdev);
385 * Some VMM ioctls can operate only on vcpus that are not running.
389 case VM_GET_REGISTER:
390 case VM_SET_REGISTER:
391 case VM_GET_SEGMENT_DESCRIPTOR:
392 case VM_SET_SEGMENT_DESCRIPTOR:
393 case VM_GET_REGISTER_SET:
394 case VM_SET_REGISTER_SET:
395 case VM_INJECT_EXCEPTION:
396 case VM_GET_CAPABILITY:
397 case VM_SET_CAPABILITY:
400 case VM_SET_X2APIC_STATE:
402 case VM_GLA2GPA_NOFAULT:
403 case VM_ACTIVATE_CPU:
406 case VM_RESTART_INSTRUCTION:
408 * XXX fragile, handle with care
409 * Assumes that the first field of the ioctl data is the vcpu.
412 error = vcpu_lock_one(sc, vcpu);
418 case VM_MAP_PPTDEV_MMIO:
420 case VM_UNBIND_PPTDEV:
421 case VM_ALLOC_MEMSEG:
425 * ioctls that operate on the entire virtual machine must
426 * prevent all vcpus from running.
428 error = vcpu_lock_all(sc);
435 case VM_MMAP_GETNEXT:
437 * Lock a vcpu to make sure that the memory map cannot be
438 * modified while it is being inspected.
440 vcpu = VM_MAXCPU - 1;
441 error = vcpu_lock_one(sc, vcpu);
453 vmrun = (struct vm_run *)data;
454 error = vm_run(sc->vm, vmrun);
457 vmsuspend = (struct vm_suspend *)data;
458 error = vm_suspend(sc->vm, vmsuspend->how);
461 error = vm_reinit(sc->vm);
464 statdesc = (struct vm_stat_desc *)data;
465 error = vmm_stat_desc_copy(statdesc->index,
466 statdesc->desc, sizeof(statdesc->desc));
470 CTASSERT(MAX_VM_STATS >= MAX_VMM_STAT_ELEMS);
471 vmstats = (struct vm_stats *)data;
472 getmicrotime(&vmstats->tv);
473 error = vmm_stat_copy(sc->vm, vmstats->cpuid,
474 &vmstats->num_entries, vmstats->statbuf);
478 pptmsi = (struct vm_pptdev_msi *)data;
479 error = ppt_setup_msi(sc->vm, pptmsi->vcpu,
480 pptmsi->bus, pptmsi->slot, pptmsi->func,
481 pptmsi->addr, pptmsi->msg,
485 pptmsix = (struct vm_pptdev_msix *)data;
486 error = ppt_setup_msix(sc->vm, pptmsix->vcpu,
487 pptmsix->bus, pptmsix->slot,
488 pptmsix->func, pptmsix->idx,
489 pptmsix->addr, pptmsix->msg,
490 pptmsix->vector_control);
492 case VM_MAP_PPTDEV_MMIO:
493 pptmmio = (struct vm_pptdev_mmio *)data;
494 error = ppt_map_mmio(sc->vm, pptmmio->bus, pptmmio->slot,
495 pptmmio->func, pptmmio->gpa, pptmmio->len,
499 pptdev = (struct vm_pptdev *)data;
500 error = vm_assign_pptdev(sc->vm, pptdev->bus, pptdev->slot,
503 case VM_UNBIND_PPTDEV:
504 pptdev = (struct vm_pptdev *)data;
505 error = vm_unassign_pptdev(sc->vm, pptdev->bus, pptdev->slot,
508 case VM_INJECT_EXCEPTION:
509 vmexc = (struct vm_exception *)data;
510 error = vm_inject_exception(sc->vm, vmexc->cpuid,
511 vmexc->vector, vmexc->error_code_valid, vmexc->error_code,
512 vmexc->restart_instruction);
515 vmnmi = (struct vm_nmi *)data;
516 error = vm_inject_nmi(sc->vm, vmnmi->cpuid);
519 vmirq = (struct vm_lapic_irq *)data;
520 error = lapic_intr_edge(sc->vm, vmirq->cpuid, vmirq->vector);
522 case VM_LAPIC_LOCAL_IRQ:
523 vmirq = (struct vm_lapic_irq *)data;
524 error = lapic_set_local_intr(sc->vm, vmirq->cpuid,
528 vmmsi = (struct vm_lapic_msi *)data;
529 error = lapic_intr_msi(sc->vm, vmmsi->addr, vmmsi->msg);
531 case VM_IOAPIC_ASSERT_IRQ:
532 ioapic_irq = (struct vm_ioapic_irq *)data;
533 error = vioapic_assert_irq(sc->vm, ioapic_irq->irq);
535 case VM_IOAPIC_DEASSERT_IRQ:
536 ioapic_irq = (struct vm_ioapic_irq *)data;
537 error = vioapic_deassert_irq(sc->vm, ioapic_irq->irq);
539 case VM_IOAPIC_PULSE_IRQ:
540 ioapic_irq = (struct vm_ioapic_irq *)data;
541 error = vioapic_pulse_irq(sc->vm, ioapic_irq->irq);
543 case VM_IOAPIC_PINCOUNT:
544 *(int *)data = vioapic_pincount(sc->vm);
546 case VM_ISA_ASSERT_IRQ:
547 isa_irq = (struct vm_isa_irq *)data;
548 error = vatpic_assert_irq(sc->vm, isa_irq->atpic_irq);
549 if (error == 0 && isa_irq->ioapic_irq != -1)
550 error = vioapic_assert_irq(sc->vm,
551 isa_irq->ioapic_irq);
553 case VM_ISA_DEASSERT_IRQ:
554 isa_irq = (struct vm_isa_irq *)data;
555 error = vatpic_deassert_irq(sc->vm, isa_irq->atpic_irq);
556 if (error == 0 && isa_irq->ioapic_irq != -1)
557 error = vioapic_deassert_irq(sc->vm,
558 isa_irq->ioapic_irq);
560 case VM_ISA_PULSE_IRQ:
561 isa_irq = (struct vm_isa_irq *)data;
562 error = vatpic_pulse_irq(sc->vm, isa_irq->atpic_irq);
563 if (error == 0 && isa_irq->ioapic_irq != -1)
564 error = vioapic_pulse_irq(sc->vm, isa_irq->ioapic_irq);
566 case VM_ISA_SET_IRQ_TRIGGER:
567 isa_irq_trigger = (struct vm_isa_irq_trigger *)data;
568 error = vatpic_set_irq_trigger(sc->vm,
569 isa_irq_trigger->atpic_irq, isa_irq_trigger->trigger);
571 case VM_MMAP_GETNEXT:
572 mm = (struct vm_memmap *)data;
573 error = vm_mmap_getnext(sc->vm, &mm->gpa, &mm->segid,
574 &mm->segoff, &mm->len, &mm->prot, &mm->flags);
577 mm = (struct vm_memmap *)data;
578 error = vm_mmap_memseg(sc->vm, mm->gpa, mm->segid, mm->segoff,
579 mm->len, mm->prot, mm->flags);
581 case VM_ALLOC_MEMSEG:
582 error = alloc_memseg(sc, (struct vm_memseg *)data);
585 error = get_memseg(sc, (struct vm_memseg *)data);
587 case VM_GET_REGISTER:
588 vmreg = (struct vm_register *)data;
589 error = vm_get_register(sc->vm, vmreg->cpuid, vmreg->regnum,
592 case VM_SET_REGISTER:
593 vmreg = (struct vm_register *)data;
594 error = vm_set_register(sc->vm, vmreg->cpuid, vmreg->regnum,
597 case VM_SET_SEGMENT_DESCRIPTOR:
598 vmsegdesc = (struct vm_seg_desc *)data;
599 error = vm_set_seg_desc(sc->vm, vmsegdesc->cpuid,
603 case VM_GET_SEGMENT_DESCRIPTOR:
604 vmsegdesc = (struct vm_seg_desc *)data;
605 error = vm_get_seg_desc(sc->vm, vmsegdesc->cpuid,
609 case VM_GET_REGISTER_SET:
610 vmregset = (struct vm_register_set *)data;
611 if (vmregset->count > VM_REG_LAST) {
615 regvals = malloc(sizeof(regvals[0]) * vmregset->count, M_VMMDEV,
617 regnums = malloc(sizeof(regnums[0]) * vmregset->count, M_VMMDEV,
619 error = copyin(vmregset->regnums, regnums, sizeof(regnums[0]) *
622 error = vm_get_register_set(sc->vm, vmregset->cpuid,
623 vmregset->count, regnums, regvals);
625 error = copyout(regvals, vmregset->regvals,
626 sizeof(regvals[0]) * vmregset->count);
627 free(regvals, M_VMMDEV);
628 free(regnums, M_VMMDEV);
630 case VM_SET_REGISTER_SET:
631 vmregset = (struct vm_register_set *)data;
632 if (vmregset->count > VM_REG_LAST) {
636 regvals = malloc(sizeof(regvals[0]) * vmregset->count, M_VMMDEV,
638 regnums = malloc(sizeof(regnums[0]) * vmregset->count, M_VMMDEV,
640 error = copyin(vmregset->regnums, regnums, sizeof(regnums[0]) *
643 error = copyin(vmregset->regvals, regvals,
644 sizeof(regvals[0]) * vmregset->count);
646 error = vm_set_register_set(sc->vm, vmregset->cpuid,
647 vmregset->count, regnums, regvals);
648 free(regvals, M_VMMDEV);
649 free(regnums, M_VMMDEV);
651 case VM_GET_CAPABILITY:
652 vmcap = (struct vm_capability *)data;
653 error = vm_get_capability(sc->vm, vmcap->cpuid,
657 case VM_SET_CAPABILITY:
658 vmcap = (struct vm_capability *)data;
659 error = vm_set_capability(sc->vm, vmcap->cpuid,
663 case VM_SET_X2APIC_STATE:
664 x2apic = (struct vm_x2apic *)data;
665 error = vm_set_x2apic_state(sc->vm,
666 x2apic->cpuid, x2apic->state);
668 case VM_GET_X2APIC_STATE:
669 x2apic = (struct vm_x2apic *)data;
670 error = vm_get_x2apic_state(sc->vm,
671 x2apic->cpuid, &x2apic->state);
673 case VM_GET_GPA_PMAP:
674 gpapte = (struct vm_gpa_pte *)data;
675 pmap_get_mapping(vmspace_pmap(vm_get_vmspace(sc->vm)),
676 gpapte->gpa, gpapte->pte, &gpapte->ptenum);
679 case VM_GET_HPET_CAPABILITIES:
680 error = vhpet_getcap((struct vm_hpet_cap *)data);
683 CTASSERT(PROT_READ == VM_PROT_READ);
684 CTASSERT(PROT_WRITE == VM_PROT_WRITE);
685 CTASSERT(PROT_EXEC == VM_PROT_EXECUTE);
686 gg = (struct vm_gla2gpa *)data;
687 error = vm_gla2gpa(sc->vm, gg->vcpuid, &gg->paging, gg->gla,
688 gg->prot, &gg->gpa, &gg->fault);
689 KASSERT(error == 0 || error == EFAULT,
690 ("%s: vm_gla2gpa unknown error %d", __func__, error));
693 case VM_GLA2GPA_NOFAULT:
694 gg = (struct vm_gla2gpa *)data;
695 error = vm_gla2gpa_nofault(sc->vm, gg->vcpuid, &gg->paging,
696 gg->gla, gg->prot, &gg->gpa, &gg->fault);
697 KASSERT(error == 0 || error == EFAULT,
698 ("%s: vm_gla2gpa unknown error %d", __func__, error));
700 case VM_ACTIVATE_CPU:
701 vac = (struct vm_activate_cpu *)data;
702 error = vm_activate_cpu(sc->vm, vac->vcpuid);
706 vm_cpuset = (struct vm_cpuset *)data;
707 size = vm_cpuset->cpusetsize;
708 if (size < sizeof(cpuset_t) || size > CPU_MAXSIZE / NBBY) {
712 cpuset = malloc(size, M_TEMP, M_WAITOK | M_ZERO);
713 if (vm_cpuset->which == VM_ACTIVE_CPUS)
714 *cpuset = vm_active_cpus(sc->vm);
715 else if (vm_cpuset->which == VM_SUSPENDED_CPUS)
716 *cpuset = vm_suspended_cpus(sc->vm);
717 else if (vm_cpuset->which == VM_DEBUG_CPUS)
718 *cpuset = vm_debug_cpus(sc->vm);
722 error = copyout(cpuset, vm_cpuset->cpus, size);
723 free(cpuset, M_TEMP);
726 vac = (struct vm_activate_cpu *)data;
727 error = vm_suspend_cpu(sc->vm, vac->vcpuid);
730 vac = (struct vm_activate_cpu *)data;
731 error = vm_resume_cpu(sc->vm, vac->vcpuid);
734 vmii = (struct vm_intinfo *)data;
735 error = vm_exit_intinfo(sc->vm, vmii->vcpuid, vmii->info1);
738 vmii = (struct vm_intinfo *)data;
739 error = vm_get_intinfo(sc->vm, vmii->vcpuid, &vmii->info1,
743 rtcdata = (struct vm_rtc_data *)data;
744 error = vrtc_nvram_write(sc->vm, rtcdata->offset,
748 rtcdata = (struct vm_rtc_data *)data;
749 error = vrtc_nvram_read(sc->vm, rtcdata->offset,
753 rtctime = (struct vm_rtc_time *)data;
754 error = vrtc_set_time(sc->vm, rtctime->secs);
758 rtctime = (struct vm_rtc_time *)data;
759 rtctime->secs = vrtc_get_time(sc->vm);
761 case VM_RESTART_INSTRUCTION:
762 error = vm_restart_instruction(sc->vm, vcpu);
764 case VM_SET_TOPOLOGY:
765 topology = (struct vm_cpu_topology *)data;
766 error = vm_set_topology(sc->vm, topology->sockets,
767 topology->cores, topology->threads, topology->maxcpus);
769 case VM_GET_TOPOLOGY:
770 topology = (struct vm_cpu_topology *)data;
771 vm_get_topology(sc->vm, &topology->sockets, &topology->cores,
772 &topology->threads, &topology->maxcpus);
780 if (state_changed == 1)
781 vcpu_unlock_one(sc, vcpu);
782 else if (state_changed == 2)
786 /* Make sure that no handler returns a bogus value like ERESTART */
787 KASSERT(error >= 0, ("vmmdev_ioctl: invalid error return %d", error));
792 vmmdev_mmap_single(struct cdev *cdev, vm_ooffset_t *offset, vm_size_t mapsize,
793 struct vm_object **objp, int nprot)
795 struct vmmdev_softc *sc;
798 vm_ooffset_t segoff, first, last;
799 int error, found, segid;
802 error = vmm_priv_check(curthread->td_ucred);
807 last = first + mapsize;
808 if ((nprot & PROT_EXEC) || first < 0 || first >= last)
811 sc = vmmdev_lookup2(cdev);
813 /* virtual machine is in the process of being created */
818 * Get a read lock on the guest memory map by freezing any vcpu.
820 error = vcpu_lock_one(sc, VM_MAXCPU - 1);
827 error = vm_mmap_getnext(sc->vm, &gpa, &segid, &segoff, &len,
832 if (first >= gpa && last <= gpa + len)
839 error = vm_get_memseg(sc->vm, segid, &len, &sysmem, objp);
840 KASSERT(error == 0 && *objp != NULL,
841 ("%s: invalid memory segment %d", __func__, segid));
843 vm_object_reference(*objp);
844 *offset = segoff + (first - gpa);
849 vcpu_unlock_one(sc, VM_MAXCPU - 1);
854 vmmdev_destroy(void *arg)
856 struct vmmdev_softc *sc = arg;
857 struct devmem_softc *dsc;
860 error = vcpu_lock_all(sc);
861 KASSERT(error == 0, ("%s: error %d freezing vcpus", __func__, error));
863 while ((dsc = SLIST_FIRST(&sc->devmem)) != NULL) {
864 KASSERT(dsc->cdev == NULL, ("%s: devmem not free", __func__));
865 SLIST_REMOVE_HEAD(&sc->devmem, link);
866 free(dsc->name, M_VMMDEV);
870 if (sc->cdev != NULL)
871 destroy_dev(sc->cdev);
876 if ((sc->flags & VSC_LINKED) != 0) {
877 mtx_lock(&vmmdev_mtx);
878 SLIST_REMOVE(&head, sc, vmmdev_softc, link);
879 mtx_unlock(&vmmdev_mtx);
886 sysctl_vmm_destroy(SYSCTL_HANDLER_ARGS)
889 char buf[VM_MAX_NAMELEN];
890 struct devmem_softc *dsc;
891 struct vmmdev_softc *sc;
894 error = vmm_priv_check(req->td->td_ucred);
898 strlcpy(buf, "beavis", sizeof(buf));
899 error = sysctl_handle_string(oidp, buf, sizeof(buf), req);
900 if (error != 0 || req->newptr == NULL)
903 mtx_lock(&vmmdev_mtx);
904 sc = vmmdev_lookup(buf);
905 if (sc == NULL || sc->cdev == NULL) {
906 mtx_unlock(&vmmdev_mtx);
911 * The 'cdev' will be destroyed asynchronously when 'si_threadcount'
912 * goes down to 0 so we should not do it again in the callback.
914 * Setting 'sc->cdev' to NULL is also used to indicate that the VM
915 * is scheduled for destruction.
919 mtx_unlock(&vmmdev_mtx);
922 * Schedule all cdevs to be destroyed:
924 * - any new operations on the 'cdev' will return an error (ENXIO).
926 * - when the 'si_threadcount' dwindles down to zero the 'cdev' will
927 * be destroyed and the callback will be invoked in a taskqueue
930 * - the 'devmem' cdevs are destroyed before the virtual machine 'cdev'
932 SLIST_FOREACH(dsc, &sc->devmem, link) {
933 KASSERT(dsc->cdev != NULL, ("devmem cdev already destroyed"));
934 destroy_dev_sched_cb(dsc->cdev, devmem_destroy, dsc);
936 destroy_dev_sched_cb(cdev, vmmdev_destroy, sc);
939 SYSCTL_PROC(_hw_vmm, OID_AUTO, destroy,
940 CTLTYPE_STRING | CTLFLAG_RW | CTLFLAG_PRISON,
941 NULL, 0, sysctl_vmm_destroy, "A", NULL);
943 static struct cdevsw vmmdevsw = {
945 .d_version = D_VERSION,
946 .d_ioctl = vmmdev_ioctl,
947 .d_mmap_single = vmmdev_mmap_single,
949 .d_write = vmmdev_rw,
953 sysctl_vmm_create(SYSCTL_HANDLER_ARGS)
958 struct vmmdev_softc *sc, *sc2;
959 char buf[VM_MAX_NAMELEN];
961 error = vmm_priv_check(req->td->td_ucred);
965 strlcpy(buf, "beavis", sizeof(buf));
966 error = sysctl_handle_string(oidp, buf, sizeof(buf), req);
967 if (error != 0 || req->newptr == NULL)
970 mtx_lock(&vmmdev_mtx);
971 sc = vmmdev_lookup(buf);
972 mtx_unlock(&vmmdev_mtx);
976 error = vm_create(buf, &vm);
980 sc = malloc(sizeof(struct vmmdev_softc), M_VMMDEV, M_WAITOK | M_ZERO);
982 SLIST_INIT(&sc->devmem);
985 * Lookup the name again just in case somebody sneaked in when we
988 mtx_lock(&vmmdev_mtx);
989 sc2 = vmmdev_lookup(buf);
991 SLIST_INSERT_HEAD(&head, sc, link);
992 sc->flags |= VSC_LINKED;
994 mtx_unlock(&vmmdev_mtx);
1001 error = make_dev_p(MAKEDEV_CHECKNAME, &cdev, &vmmdevsw, NULL,
1002 UID_ROOT, GID_WHEEL, 0600, "vmm/%s", buf);
1008 mtx_lock(&vmmdev_mtx);
1010 sc->cdev->si_drv1 = sc;
1011 mtx_unlock(&vmmdev_mtx);
1015 SYSCTL_PROC(_hw_vmm, OID_AUTO, create,
1016 CTLTYPE_STRING | CTLFLAG_RW | CTLFLAG_PRISON,
1017 NULL, 0, sysctl_vmm_create, "A", NULL);
1022 mtx_init(&vmmdev_mtx, "vmm device mutex", NULL, MTX_DEF);
1023 pr_allow_flag = prison_add_allow(NULL, "vmm", NULL,
1024 "Allow use of vmm in a jail.");
1028 vmmdev_cleanup(void)
1032 if (SLIST_EMPTY(&head))
1041 devmem_mmap_single(struct cdev *cdev, vm_ooffset_t *offset, vm_size_t len,
1042 struct vm_object **objp, int nprot)
1044 struct devmem_softc *dsc;
1045 vm_ooffset_t first, last;
1050 dsc = cdev->si_drv1;
1052 /* 'cdev' has been created but is not ready for use */
1057 last = *offset + len;
1058 if ((nprot & PROT_EXEC) || first < 0 || first >= last)
1061 error = vcpu_lock_one(dsc->sc, VM_MAXCPU - 1);
1065 error = vm_get_memseg(dsc->sc->vm, dsc->segid, &seglen, &sysmem, objp);
1066 KASSERT(error == 0 && !sysmem && *objp != NULL,
1067 ("%s: invalid devmem segment %d", __func__, dsc->segid));
1069 vcpu_unlock_one(dsc->sc, VM_MAXCPU - 1);
1071 if (seglen >= last) {
1072 vm_object_reference(*objp);
1079 static struct cdevsw devmemsw = {
1081 .d_version = D_VERSION,
1082 .d_mmap_single = devmem_mmap_single,
1086 devmem_create_cdev(const char *vmname, int segid, char *devname)
1088 struct devmem_softc *dsc;
1089 struct vmmdev_softc *sc;
1093 error = make_dev_p(MAKEDEV_CHECKNAME, &cdev, &devmemsw, NULL,
1094 UID_ROOT, GID_WHEEL, 0600, "vmm.io/%s.%s", vmname, devname);
1098 dsc = malloc(sizeof(struct devmem_softc), M_VMMDEV, M_WAITOK | M_ZERO);
1100 mtx_lock(&vmmdev_mtx);
1101 sc = vmmdev_lookup(vmname);
1102 KASSERT(sc != NULL, ("%s: vm %s softc not found", __func__, vmname));
1103 if (sc->cdev == NULL) {
1104 /* virtual machine is being created or destroyed */
1105 mtx_unlock(&vmmdev_mtx);
1106 free(dsc, M_VMMDEV);
1107 destroy_dev_sched_cb(cdev, NULL, 0);
1112 dsc->name = devname;
1115 SLIST_INSERT_HEAD(&sc->devmem, dsc, link);
1116 mtx_unlock(&vmmdev_mtx);
1118 /* The 'cdev' is ready for use after 'si_drv1' is initialized */
1119 cdev->si_drv1 = dsc;
1124 devmem_destroy(void *arg)
1126 struct devmem_softc *dsc = arg;
1128 KASSERT(dsc->cdev, ("%s: devmem cdev already destroyed", __func__));