diff options
Diffstat (limited to 'sys/amd64/vmm')
| -rw-r--r-- | sys/amd64/vmm/vmm.c | 187 | ||||
| -rw-r--r-- | sys/amd64/vmm/vmm_dev_machdep.c | 255 | ||||
| -rw-r--r-- | sys/amd64/vmm/vmm_mem.h | 5 | ||||
| -rw-r--r-- | sys/amd64/vmm/vmm_mem_machdep.c | 61 |
4 files changed, 350 insertions, 158 deletions
diff --git a/sys/amd64/vmm/vmm.c b/sys/amd64/vmm/vmm.c index 2ac076551165..f2bea0d82b5c 100644 --- a/sys/amd64/vmm/vmm.c +++ b/sys/amd64/vmm/vmm.c @@ -562,9 +562,9 @@ vm_alloc_vcpu(struct vm *vm, int vcpuid) } void -vm_slock_vcpus(struct vm *vm) +vm_lock_vcpus(struct vm *vm) { - sx_slock(&vm->vcpus_init_lock); + sx_xlock(&vm->vcpus_init_lock); } void @@ -724,12 +724,7 @@ vm_name(struct vm *vm) int vm_map_mmio(struct vm *vm, vm_paddr_t gpa, size_t len, vm_paddr_t hpa) { - vm_object_t obj; - - if ((obj = vmm_mmio_alloc(vm_vmspace(vm), gpa, len, hpa)) == NULL) - return (ENOMEM); - else - return (0); + return (vmm_mmio_alloc(vm_vmspace(vm), gpa, len, hpa)); } int @@ -870,7 +865,7 @@ vm_assign_pptdev(struct vm *vm, int bus, int slot, int func) int vm_get_register(struct vcpu *vcpu, int reg, uint64_t *retval) { - + /* Negative values represent VM control structure fields. */ if (reg >= VM_REG_LAST) return (EINVAL); @@ -882,6 +877,7 @@ vm_set_register(struct vcpu *vcpu, int reg, uint64_t val) { int error; + /* Negative values represent VM control structure fields. */ if (reg >= VM_REG_LAST) return (EINVAL); @@ -990,6 +986,54 @@ save_guest_fpustate(struct vcpu *vcpu) static VMM_STAT(VCPU_IDLE_TICKS, "number of ticks vcpu was idle"); +/* + * Invoke the rendezvous function on the specified vcpu if applicable. Return + * true if the rendezvous is finished, false otherwise. + */ +static bool +vm_rendezvous(struct vcpu *vcpu) +{ + struct vm *vm = vcpu->vm; + int vcpuid; + + mtx_assert(&vcpu->vm->rendezvous_mtx, MA_OWNED); + KASSERT(vcpu->vm->rendezvous_func != NULL, + ("vm_rendezvous: no rendezvous pending")); + + /* 'rendezvous_req_cpus' must be a subset of 'active_cpus' */ + CPU_AND(&vm->rendezvous_req_cpus, &vm->rendezvous_req_cpus, + &vm->active_cpus); + + vcpuid = vcpu->vcpuid; + if (CPU_ISSET(vcpuid, &vm->rendezvous_req_cpus) && + !CPU_ISSET(vcpuid, &vm->rendezvous_done_cpus)) { + VMM_CTR0(vcpu, "Calling rendezvous func"); + (*vm->rendezvous_func)(vcpu, vm->rendezvous_arg); + CPU_SET(vcpuid, &vm->rendezvous_done_cpus); + } + if (CPU_CMP(&vm->rendezvous_req_cpus, + &vm->rendezvous_done_cpus) == 0) { + VMM_CTR0(vcpu, "Rendezvous completed"); + CPU_ZERO(&vm->rendezvous_req_cpus); + vm->rendezvous_func = NULL; + wakeup(&vm->rendezvous_func); + return (true); + } + return (false); +} + +static void +vcpu_wait_idle(struct vcpu *vcpu) +{ + KASSERT(vcpu->state != VCPU_IDLE, ("vcpu already idle")); + + vcpu->reqidle = 1; + vcpu_notify_event_locked(vcpu, false); + VMM_CTR1(vcpu, "vcpu state change from %s to " + "idle requested", vcpu_state2str(vcpu->state)); + msleep_spin(&vcpu->state, &vcpu->mtx, "vmstat", hz); +} + static int vcpu_set_state_locked(struct vcpu *vcpu, enum vcpu_state newstate, bool from_idle) @@ -1004,13 +1048,8 @@ vcpu_set_state_locked(struct vcpu *vcpu, enum vcpu_state newstate, * ioctl() operating on a vcpu at any point. */ if (from_idle) { - while (vcpu->state != VCPU_IDLE) { - vcpu->reqidle = 1; - vcpu_notify_event_locked(vcpu, false); - VMM_CTR1(vcpu, "vcpu state change from %s to " - "idle requested", vcpu_state2str(vcpu->state)); - msleep_spin(&vcpu->state, &vcpu->mtx, "vmstat", hz); - } + while (vcpu->state != VCPU_IDLE) + vcpu_wait_idle(vcpu); } else { KASSERT(vcpu->state != VCPU_IDLE, ("invalid transition from " "vcpu idle state")); @@ -1062,6 +1101,95 @@ vcpu_set_state_locked(struct vcpu *vcpu, enum vcpu_state newstate, return (0); } +/* + * Try to lock all of the vCPUs in the VM while taking care to avoid deadlocks + * with vm_smp_rendezvous(). + * + * The complexity here suggests that the rendezvous mechanism needs a rethink. + */ +int +vcpu_set_state_all(struct vm *vm, enum vcpu_state newstate) +{ + cpuset_t locked; + struct vcpu *vcpu; + int error, i; + uint16_t maxcpus; + + KASSERT(newstate != VCPU_IDLE, + ("vcpu_set_state_all: invalid target state %d", newstate)); + + error = 0; + CPU_ZERO(&locked); + maxcpus = vm->maxcpus; + + mtx_lock(&vm->rendezvous_mtx); +restart: + if (vm->rendezvous_func != NULL) { + /* + * If we have a pending rendezvous, then the initiator may be + * blocked waiting for other vCPUs to execute the callback. The + * current thread may be a vCPU thread so we must not block + * waiting for the initiator, otherwise we get a deadlock. + * Thus, execute the callback on behalf of any idle vCPUs. + */ + for (i = 0; i < maxcpus; i++) { + vcpu = vm_vcpu(vm, i); + if (vcpu == NULL) + continue; + vcpu_lock(vcpu); + if (vcpu->state == VCPU_IDLE) { + (void)vcpu_set_state_locked(vcpu, VCPU_FROZEN, + true); + CPU_SET(i, &locked); + } + if (CPU_ISSET(i, &locked)) { + /* + * We can safely execute the callback on this + * vCPU's behalf. + */ + vcpu_unlock(vcpu); + (void)vm_rendezvous(vcpu); + vcpu_lock(vcpu); + } + vcpu_unlock(vcpu); + } + } + + /* + * Now wait for remaining vCPUs to become idle. This may include the + * initiator of a rendezvous that is currently blocked on the rendezvous + * mutex. + */ + CPU_FOREACH_ISCLR(i, &locked) { + if (i >= maxcpus) + break; + vcpu = vm_vcpu(vm, i); + if (vcpu == NULL) + continue; + vcpu_lock(vcpu); + while (vcpu->state != VCPU_IDLE) { + mtx_unlock(&vm->rendezvous_mtx); + vcpu_wait_idle(vcpu); + vcpu_unlock(vcpu); + mtx_lock(&vm->rendezvous_mtx); + if (vm->rendezvous_func != NULL) + goto restart; + vcpu_lock(vcpu); + } + error = vcpu_set_state_locked(vcpu, newstate, true); + vcpu_unlock(vcpu); + if (error != 0) { + /* Roll back state changes. */ + CPU_FOREACH_ISSET(i, &locked) + (void)vcpu_set_state(vcpu, VCPU_IDLE, false); + break; + } + CPU_SET(i, &locked); + } + mtx_unlock(&vm->rendezvous_mtx); + return (error); +} + static void vcpu_require_state(struct vcpu *vcpu, enum vcpu_state newstate) { @@ -1083,36 +1211,23 @@ vcpu_require_state_locked(struct vcpu *vcpu, enum vcpu_state newstate) static int vm_handle_rendezvous(struct vcpu *vcpu) { - struct vm *vm = vcpu->vm; + struct vm *vm; struct thread *td; - int error, vcpuid; - error = 0; - vcpuid = vcpu->vcpuid; td = curthread; + vm = vcpu->vm; + mtx_lock(&vm->rendezvous_mtx); while (vm->rendezvous_func != NULL) { - /* 'rendezvous_req_cpus' must be a subset of 'active_cpus' */ - CPU_AND(&vm->rendezvous_req_cpus, &vm->rendezvous_req_cpus, &vm->active_cpus); - - if (CPU_ISSET(vcpuid, &vm->rendezvous_req_cpus) && - !CPU_ISSET(vcpuid, &vm->rendezvous_done_cpus)) { - VMM_CTR0(vcpu, "Calling rendezvous func"); - (*vm->rendezvous_func)(vcpu, vm->rendezvous_arg); - CPU_SET(vcpuid, &vm->rendezvous_done_cpus); - } - if (CPU_CMP(&vm->rendezvous_req_cpus, - &vm->rendezvous_done_cpus) == 0) { - VMM_CTR0(vcpu, "Rendezvous completed"); - CPU_ZERO(&vm->rendezvous_req_cpus); - vm->rendezvous_func = NULL; - wakeup(&vm->rendezvous_func); + if (vm_rendezvous(vcpu)) break; - } + VMM_CTR0(vcpu, "Wait for rendezvous completion"); mtx_sleep(&vm->rendezvous_func, &vm->rendezvous_mtx, 0, "vmrndv", hz); if (td_ast_pending(td, TDA_SUSPEND)) { + int error; + mtx_unlock(&vm->rendezvous_mtx); error = thread_check_susp(td, true); if (error != 0) diff --git a/sys/amd64/vmm/vmm_dev_machdep.c b/sys/amd64/vmm/vmm_dev_machdep.c index dfebc9dcadbf..b84be809ea24 100644 --- a/sys/amd64/vmm/vmm_dev_machdep.c +++ b/sys/amd64/vmm/vmm_dev_machdep.c @@ -124,12 +124,16 @@ const struct vmmdev_ioctl vmmdev_machdep_ioctls[] = { VMMDEV_IOCTL(VM_SET_KERNEMU_DEV, VMMDEV_IOCTL_LOCK_ONE_VCPU), VMMDEV_IOCTL(VM_BIND_PPTDEV, - VMMDEV_IOCTL_XLOCK_MEMSEGS | VMMDEV_IOCTL_LOCK_ALL_VCPUS), + VMMDEV_IOCTL_XLOCK_MEMSEGS | VMMDEV_IOCTL_LOCK_ALL_VCPUS | + VMMDEV_IOCTL_PRIV_CHECK_DRIVER), VMMDEV_IOCTL(VM_UNBIND_PPTDEV, - VMMDEV_IOCTL_XLOCK_MEMSEGS | VMMDEV_IOCTL_LOCK_ALL_VCPUS), + VMMDEV_IOCTL_XLOCK_MEMSEGS | VMMDEV_IOCTL_LOCK_ALL_VCPUS | + VMMDEV_IOCTL_PRIV_CHECK_DRIVER), - VMMDEV_IOCTL(VM_MAP_PPTDEV_MMIO, VMMDEV_IOCTL_LOCK_ALL_VCPUS), - VMMDEV_IOCTL(VM_UNMAP_PPTDEV_MMIO, VMMDEV_IOCTL_LOCK_ALL_VCPUS), + VMMDEV_IOCTL(VM_MAP_PPTDEV_MMIO, VMMDEV_IOCTL_LOCK_ALL_VCPUS | + VMMDEV_IOCTL_PRIV_CHECK_DRIVER), + VMMDEV_IOCTL(VM_UNMAP_PPTDEV_MMIO, VMMDEV_IOCTL_LOCK_ALL_VCPUS | + VMMDEV_IOCTL_PRIV_CHECK_DRIVER), #ifdef BHYVE_SNAPSHOT #ifdef COMPAT_FREEBSD13 VMMDEV_IOCTL(VM_SNAPSHOT_REQ_13, VMMDEV_IOCTL_LOCK_ALL_VCPUS), @@ -147,9 +151,9 @@ const struct vmmdev_ioctl vmmdev_machdep_ioctls[] = { VMMDEV_IOCTL(VM_LAPIC_LOCAL_IRQ, VMMDEV_IOCTL_MAYBE_ALLOC_VCPU), - VMMDEV_IOCTL(VM_PPTDEV_MSI, 0), - VMMDEV_IOCTL(VM_PPTDEV_MSIX, 0), - VMMDEV_IOCTL(VM_PPTDEV_DISABLE_MSIX, 0), + VMMDEV_IOCTL(VM_PPTDEV_MSI, VMMDEV_IOCTL_PRIV_CHECK_DRIVER), + VMMDEV_IOCTL(VM_PPTDEV_MSIX, VMMDEV_IOCTL_PRIV_CHECK_DRIVER), + VMMDEV_IOCTL(VM_PPTDEV_DISABLE_MSIX, VMMDEV_IOCTL_PRIV_CHECK_DRIVER), VMMDEV_IOCTL(VM_LAPIC_MSI, 0), VMMDEV_IOCTL(VM_IOAPIC_ASSERT_IRQ, 0), VMMDEV_IOCTL(VM_IOAPIC_DEASSERT_IRQ, 0), @@ -172,40 +176,13 @@ int vmmdev_machdep_ioctl(struct vm *vm, struct vcpu *vcpu, u_long cmd, caddr_t data, int fflag, struct thread *td) { - struct vm_seg_desc *vmsegdesc; - struct vm_run *vmrun; -#ifdef COMPAT_FREEBSD13 - struct vm_run_13 *vmrun_13; -#endif - struct vm_exception *vmexc; - struct vm_lapic_irq *vmirq; - struct vm_lapic_msi *vmmsi; - struct vm_ioapic_irq *ioapic_irq; - struct vm_isa_irq *isa_irq; - struct vm_isa_irq_trigger *isa_irq_trigger; - struct vm_pptdev *pptdev; - struct vm_pptdev_mmio *pptmmio; - struct vm_pptdev_msi *pptmsi; - struct vm_pptdev_msix *pptmsix; - struct vm_x2apic *x2apic; - struct vm_gpa_pte *gpapte; - struct vm_gla2gpa *gg; - struct vm_intinfo *vmii; - struct vm_rtc_time *rtctime; - struct vm_rtc_data *rtcdata; - struct vm_readwrite_kernemu_device *kernemu; -#ifdef BHYVE_SNAPSHOT - struct vm_snapshot_meta *snapshot_meta; -#ifdef COMPAT_FREEBSD13 - struct vm_snapshot_meta_13 *snapshot_13; -#endif -#endif int error; error = 0; switch (cmd) { case VM_RUN: { struct vm_exit *vme; + struct vm_run *vmrun; vmrun = (struct vm_run *)data; vme = vm_exitinfo(vcpu); @@ -243,6 +220,7 @@ vmmdev_machdep_ioctl(struct vm *vm, struct vcpu *vcpu, u_long cmd, caddr_t data, case VM_RUN_13: { struct vm_exit *vme; struct vm_exit_13 *vme_13; + struct vm_run_13 *vmrun_13; vmrun_13 = (struct vm_run_13 *)data; vme_13 = &vmrun_13->vm_exit; @@ -281,85 +259,123 @@ vmmdev_machdep_ioctl(struct vm *vm, struct vcpu *vcpu, u_long cmd, caddr_t data, break; } #endif - case VM_PPTDEV_MSI: + case VM_PPTDEV_MSI: { + struct vm_pptdev_msi *pptmsi; + pptmsi = (struct vm_pptdev_msi *)data; - error = ppt_setup_msi(vm, - pptmsi->bus, pptmsi->slot, pptmsi->func, - pptmsi->addr, pptmsi->msg, - pptmsi->numvec); + error = ppt_setup_msi(vm, pptmsi->bus, pptmsi->slot, + pptmsi->func, pptmsi->addr, pptmsi->msg, pptmsi->numvec); break; - case VM_PPTDEV_MSIX: + } + case VM_PPTDEV_MSIX: { + struct vm_pptdev_msix *pptmsix; + pptmsix = (struct vm_pptdev_msix *)data; - error = ppt_setup_msix(vm, - pptmsix->bus, pptmsix->slot, - pptmsix->func, pptmsix->idx, - pptmsix->addr, pptmsix->msg, - pptmsix->vector_control); + error = ppt_setup_msix(vm, pptmsix->bus, pptmsix->slot, + pptmsix->func, pptmsix->idx, pptmsix->addr, pptmsix->msg, + pptmsix->vector_control); break; - case VM_PPTDEV_DISABLE_MSIX: + } + case VM_PPTDEV_DISABLE_MSIX: { + struct vm_pptdev *pptdev; + pptdev = (struct vm_pptdev *)data; error = ppt_disable_msix(vm, pptdev->bus, pptdev->slot, - pptdev->func); + pptdev->func); break; - case VM_MAP_PPTDEV_MMIO: + } + case VM_MAP_PPTDEV_MMIO: { + struct vm_pptdev_mmio *pptmmio; + pptmmio = (struct vm_pptdev_mmio *)data; error = ppt_map_mmio(vm, pptmmio->bus, pptmmio->slot, - pptmmio->func, pptmmio->gpa, pptmmio->len, - pptmmio->hpa); + pptmmio->func, pptmmio->gpa, pptmmio->len, pptmmio->hpa); break; - case VM_UNMAP_PPTDEV_MMIO: + } + case VM_UNMAP_PPTDEV_MMIO: { + struct vm_pptdev_mmio *pptmmio; + pptmmio = (struct vm_pptdev_mmio *)data; error = ppt_unmap_mmio(vm, pptmmio->bus, pptmmio->slot, - pptmmio->func, pptmmio->gpa, pptmmio->len); + pptmmio->func, pptmmio->gpa, pptmmio->len); break; - case VM_BIND_PPTDEV: + } + case VM_BIND_PPTDEV: { + struct vm_pptdev *pptdev; + pptdev = (struct vm_pptdev *)data; error = vm_assign_pptdev(vm, pptdev->bus, pptdev->slot, - pptdev->func); + pptdev->func); break; - case VM_UNBIND_PPTDEV: + } + case VM_UNBIND_PPTDEV: { + struct vm_pptdev *pptdev; + pptdev = (struct vm_pptdev *)data; error = vm_unassign_pptdev(vm, pptdev->bus, pptdev->slot, - pptdev->func); + pptdev->func); break; - case VM_INJECT_EXCEPTION: + } + case VM_INJECT_EXCEPTION: { + struct vm_exception *vmexc; + vmexc = (struct vm_exception *)data; error = vm_inject_exception(vcpu, vmexc->vector, vmexc->error_code_valid, vmexc->error_code, vmexc->restart_instruction); break; + } case VM_INJECT_NMI: error = vm_inject_nmi(vcpu); break; - case VM_LAPIC_IRQ: + case VM_LAPIC_IRQ: { + struct vm_lapic_irq *vmirq; + vmirq = (struct vm_lapic_irq *)data; error = lapic_intr_edge(vcpu, vmirq->vector); break; - case VM_LAPIC_LOCAL_IRQ: + } + case VM_LAPIC_LOCAL_IRQ: { + struct vm_lapic_irq *vmirq; + vmirq = (struct vm_lapic_irq *)data; error = lapic_set_local_intr(vm, vcpu, vmirq->vector); break; - case VM_LAPIC_MSI: + } + case VM_LAPIC_MSI: { + struct vm_lapic_msi *vmmsi; + vmmsi = (struct vm_lapic_msi *)data; error = lapic_intr_msi(vm, vmmsi->addr, vmmsi->msg); break; - case VM_IOAPIC_ASSERT_IRQ: + } + case VM_IOAPIC_ASSERT_IRQ: { + struct vm_ioapic_irq *ioapic_irq; + ioapic_irq = (struct vm_ioapic_irq *)data; error = vioapic_assert_irq(vm, ioapic_irq->irq); break; - case VM_IOAPIC_DEASSERT_IRQ: + } + case VM_IOAPIC_DEASSERT_IRQ: { + struct vm_ioapic_irq *ioapic_irq; + ioapic_irq = (struct vm_ioapic_irq *)data; error = vioapic_deassert_irq(vm, ioapic_irq->irq); break; - case VM_IOAPIC_PULSE_IRQ: + } + case VM_IOAPIC_PULSE_IRQ: { + struct vm_ioapic_irq *ioapic_irq; + ioapic_irq = (struct vm_ioapic_irq *)data; error = vioapic_pulse_irq(vm, ioapic_irq->irq); break; + } case VM_IOAPIC_PINCOUNT: *(int *)data = vioapic_pincount(vm); break; case VM_SET_KERNEMU_DEV: case VM_GET_KERNEMU_DEV: { + struct vm_readwrite_kernemu_device *kernemu; mem_region_write_t mwrite; mem_region_read_t mread; int size; @@ -396,60 +412,86 @@ vmmdev_machdep_ioctl(struct vm *vm, struct vcpu *vcpu, u_long cmd, caddr_t data, error = mread(vcpu, kernemu->gpa, &kernemu->value, size, &arg); break; - } - case VM_ISA_ASSERT_IRQ: + } + case VM_ISA_ASSERT_IRQ: { + struct vm_isa_irq *isa_irq; + isa_irq = (struct vm_isa_irq *)data; error = vatpic_assert_irq(vm, isa_irq->atpic_irq); if (error == 0 && isa_irq->ioapic_irq != -1) error = vioapic_assert_irq(vm, isa_irq->ioapic_irq); break; - case VM_ISA_DEASSERT_IRQ: + } + case VM_ISA_DEASSERT_IRQ: { + struct vm_isa_irq *isa_irq; + isa_irq = (struct vm_isa_irq *)data; error = vatpic_deassert_irq(vm, isa_irq->atpic_irq); if (error == 0 && isa_irq->ioapic_irq != -1) error = vioapic_deassert_irq(vm, isa_irq->ioapic_irq); break; - case VM_ISA_PULSE_IRQ: + } + case VM_ISA_PULSE_IRQ: { + struct vm_isa_irq *isa_irq; + isa_irq = (struct vm_isa_irq *)data; error = vatpic_pulse_irq(vm, isa_irq->atpic_irq); if (error == 0 && isa_irq->ioapic_irq != -1) error = vioapic_pulse_irq(vm, isa_irq->ioapic_irq); break; - case VM_ISA_SET_IRQ_TRIGGER: + } + case VM_ISA_SET_IRQ_TRIGGER: { + struct vm_isa_irq_trigger *isa_irq_trigger; + isa_irq_trigger = (struct vm_isa_irq_trigger *)data; error = vatpic_set_irq_trigger(vm, isa_irq_trigger->atpic_irq, isa_irq_trigger->trigger); break; - case VM_SET_SEGMENT_DESCRIPTOR: + } + case VM_SET_SEGMENT_DESCRIPTOR: { + struct vm_seg_desc *vmsegdesc; + vmsegdesc = (struct vm_seg_desc *)data; - error = vm_set_seg_desc(vcpu, - vmsegdesc->regnum, - &vmsegdesc->desc); + error = vm_set_seg_desc(vcpu, vmsegdesc->regnum, + &vmsegdesc->desc); break; - case VM_GET_SEGMENT_DESCRIPTOR: + } + case VM_GET_SEGMENT_DESCRIPTOR: { + struct vm_seg_desc *vmsegdesc; + vmsegdesc = (struct vm_seg_desc *)data; - error = vm_get_seg_desc(vcpu, - vmsegdesc->regnum, - &vmsegdesc->desc); + error = vm_get_seg_desc(vcpu, vmsegdesc->regnum, + &vmsegdesc->desc); break; - case VM_SET_X2APIC_STATE: + } + case VM_SET_X2APIC_STATE: { + struct vm_x2apic *x2apic; + x2apic = (struct vm_x2apic *)data; error = vm_set_x2apic_state(vcpu, x2apic->state); break; - case VM_GET_X2APIC_STATE: + } + case VM_GET_X2APIC_STATE: { + struct vm_x2apic *x2apic; + x2apic = (struct vm_x2apic *)data; error = vm_get_x2apic_state(vcpu, &x2apic->state); break; - case VM_GET_GPA_PMAP: + } + case VM_GET_GPA_PMAP: { + struct vm_gpa_pte *gpapte; + gpapte = (struct vm_gpa_pte *)data; - pmap_get_mapping(vmspace_pmap(vm_vmspace(vm)), - gpapte->gpa, gpapte->pte, &gpapte->ptenum); - error = 0; + pmap_get_mapping(vmspace_pmap(vm_vmspace(vm)), gpapte->gpa, + gpapte->pte, &gpapte->ptenum); break; + } case VM_GET_HPET_CAPABILITIES: error = vhpet_getcap((struct vm_hpet_cap *)data); break; case VM_GLA2GPA: { + struct vm_gla2gpa *gg; + CTASSERT(PROT_READ == VM_PROT_READ); CTASSERT(PROT_WRITE == VM_PROT_WRITE); CTASSERT(PROT_EXEC == VM_PROT_EXECUTE); @@ -460,50 +502,76 @@ vmmdev_machdep_ioctl(struct vm *vm, struct vcpu *vcpu, u_long cmd, caddr_t data, ("%s: vm_gla2gpa unknown error %d", __func__, error)); break; } - case VM_GLA2GPA_NOFAULT: + case VM_GLA2GPA_NOFAULT: { + struct vm_gla2gpa *gg; + gg = (struct vm_gla2gpa *)data; error = vm_gla2gpa_nofault(vcpu, &gg->paging, gg->gla, gg->prot, &gg->gpa, &gg->fault); KASSERT(error == 0 || error == EFAULT, ("%s: vm_gla2gpa unknown error %d", __func__, error)); break; - case VM_SET_INTINFO: + } + case VM_SET_INTINFO: { + struct vm_intinfo *vmii; + vmii = (struct vm_intinfo *)data; error = vm_exit_intinfo(vcpu, vmii->info1); break; - case VM_GET_INTINFO: + } + case VM_GET_INTINFO: { + struct vm_intinfo *vmii; + vmii = (struct vm_intinfo *)data; error = vm_get_intinfo(vcpu, &vmii->info1, &vmii->info2); break; - case VM_RTC_WRITE: + } + case VM_RTC_WRITE: { + struct vm_rtc_data *rtcdata; + rtcdata = (struct vm_rtc_data *)data; error = vrtc_nvram_write(vm, rtcdata->offset, rtcdata->value); break; - case VM_RTC_READ: + } + case VM_RTC_READ: { + struct vm_rtc_data *rtcdata; + rtcdata = (struct vm_rtc_data *)data; error = vrtc_nvram_read(vm, rtcdata->offset, &rtcdata->value); break; - case VM_RTC_SETTIME: + } + case VM_RTC_SETTIME: { + struct vm_rtc_time *rtctime; + rtctime = (struct vm_rtc_time *)data; error = vrtc_set_time(vm, rtctime->secs); break; - case VM_RTC_GETTIME: - error = 0; + } + case VM_RTC_GETTIME: { + struct vm_rtc_time *rtctime; + rtctime = (struct vm_rtc_time *)data; rtctime->secs = vrtc_get_time(vm); break; + } case VM_RESTART_INSTRUCTION: error = vm_restart_instruction(vcpu); break; #ifdef BHYVE_SNAPSHOT - case VM_SNAPSHOT_REQ: + case VM_SNAPSHOT_REQ: { + struct vm_snapshot_meta *snapshot_meta; + snapshot_meta = (struct vm_snapshot_meta *)data; error = vm_snapshot_req(vm, snapshot_meta); break; + } #ifdef COMPAT_FREEBSD13 - case VM_SNAPSHOT_REQ_13: + case VM_SNAPSHOT_REQ_13: { + struct vm_snapshot_meta *snapshot_meta; + struct vm_snapshot_meta_13 *snapshot_13; + /* * The old structure just has an additional pointer at * the start that is ignored. @@ -513,6 +581,7 @@ vmmdev_machdep_ioctl(struct vm *vm, struct vcpu *vcpu, u_long cmd, caddr_t data, (struct vm_snapshot_meta *)&snapshot_13->dev_data; error = vm_snapshot_req(vm, snapshot_meta); break; + } #endif case VM_RESTORE_TIME: error = vm_restore_time(vm); diff --git a/sys/amd64/vmm/vmm_mem.h b/sys/amd64/vmm/vmm_mem.h index 41b9bf07c4fc..d905fd37001d 100644 --- a/sys/amd64/vmm/vmm_mem.h +++ b/sys/amd64/vmm/vmm_mem.h @@ -30,10 +30,9 @@ #define _VMM_MEM_H_ struct vmspace; -struct vm_object; -struct vm_object *vmm_mmio_alloc(struct vmspace *, vm_paddr_t gpa, size_t len, - vm_paddr_t hpa); +int vmm_mmio_alloc(struct vmspace *, vm_paddr_t gpa, size_t len, + vm_paddr_t hpa); void vmm_mmio_free(struct vmspace *, vm_paddr_t gpa, size_t size); vm_paddr_t vmm_mem_maxaddr(void); diff --git a/sys/amd64/vmm/vmm_mem_machdep.c b/sys/amd64/vmm/vmm_mem_machdep.c index e96c9e4bdc66..afb3a0274e2a 100644 --- a/sys/amd64/vmm/vmm_mem_machdep.c +++ b/sys/amd64/vmm/vmm_mem_machdep.c @@ -36,6 +36,7 @@ #include <vm/vm.h> #include <vm/vm_param.h> #include <vm/pmap.h> +#include <vm/vm_extern.h> #include <vm/vm_map.h> #include <vm/vm_object.h> #include <vm/vm_page.h> @@ -45,40 +46,48 @@ #include "vmm_mem.h" -vm_object_t +int vmm_mmio_alloc(struct vmspace *vmspace, vm_paddr_t gpa, size_t len, - vm_paddr_t hpa) + vm_paddr_t hpa) { - int error; - vm_object_t obj; struct sglist *sg; + vm_object_t obj; + int error; + + if (gpa + len < gpa || hpa + len < hpa || (gpa & PAGE_MASK) != 0 || + (hpa & PAGE_MASK) != 0 || (len & PAGE_MASK) != 0) + return (EINVAL); sg = sglist_alloc(1, M_WAITOK); error = sglist_append_phys(sg, hpa, len); KASSERT(error == 0, ("error %d appending physaddr to sglist", error)); obj = vm_pager_allocate(OBJT_SG, sg, len, VM_PROT_RW, 0, NULL); - if (obj != NULL) { - /* - * VT-x ignores the MTRR settings when figuring out the - * memory type for translations obtained through EPT. - * - * Therefore we explicitly force the pages provided by - * this object to be mapped as uncacheable. - */ - VM_OBJECT_WLOCK(obj); - error = vm_object_set_memattr(obj, VM_MEMATTR_UNCACHEABLE); - VM_OBJECT_WUNLOCK(obj); - if (error != KERN_SUCCESS) { - panic("vmm_mmio_alloc: vm_object_set_memattr error %d", - error); - } - error = vm_map_find(&vmspace->vm_map, obj, 0, &gpa, len, 0, - VMFS_NO_SPACE, VM_PROT_RW, VM_PROT_RW, 0); - if (error != KERN_SUCCESS) { - vm_object_deallocate(obj); - obj = NULL; - } + if (obj == NULL) + return (ENOMEM); + + /* + * VT-x ignores the MTRR settings when figuring out the memory type for + * translations obtained through EPT. + * + * Therefore we explicitly force the pages provided by this object to be + * mapped as uncacheable. + */ + VM_OBJECT_WLOCK(obj); + error = vm_object_set_memattr(obj, VM_MEMATTR_UNCACHEABLE); + VM_OBJECT_WUNLOCK(obj); + if (error != KERN_SUCCESS) + panic("vmm_mmio_alloc: vm_object_set_memattr error %d", error); + + vm_map_lock(&vmspace->vm_map); + error = vm_map_insert(&vmspace->vm_map, obj, 0, gpa, gpa + len, + VM_PROT_RW, VM_PROT_RW, 0); + vm_map_unlock(&vmspace->vm_map); + if (error != KERN_SUCCESS) { + error = vm_mmap_to_errno(error); + vm_object_deallocate(obj); + } else { + error = 0; } /* @@ -94,7 +103,7 @@ vmm_mmio_alloc(struct vmspace *vmspace, vm_paddr_t gpa, size_t len, */ sglist_free(sg); - return (obj); + return (error); } void |