| 1 | /* |
| 2 | * Copyright (c) 2024-2025 Oracle and/or its affiliates. |
| 3 | * |
| 4 | * SPDX-License-Identifier: GPL-2.0-or-later |
| 5 | */ |
| 6 | |
| 7 | #include "qemu/osdep.h" |
| 8 | #include "qemu/error-report.h" |
| 9 | #include "qapi/error.h" |
| 10 | #include "hw/vfio/vfio-cpr.h" |
| 11 | #include "hw/vfio/vfio-device.h" |
| 12 | #include "migration/blocker.h" |
| 13 | #include "migration/cpr.h" |
| 14 | #include "migration/migration.h" |
| 15 | #include "migration/vmstate.h" |
| 16 | #include "system/iommufd.h" |
| 17 | #include "vfio-iommufd.h" |
| 18 | #include "trace.h" |
| 19 | |
| 20 | typedef struct CprVFIODevice { |
| 21 | char *name; |
| 22 | unsigned int namelen; |
| 23 | uint32_t ioas_id; |
| 24 | int devid; |
| 25 | uint32_t hwpt_id; |
| 26 | QLIST_ENTRY(CprVFIODevice) next; |
| 27 | } CprVFIODevice; |
| 28 | |
| 29 | static const VMStateDescription vmstate_cpr_vfio_device = { |
| 30 | .name = "cpr vfio device", |
| 31 | .version_id = 1, |
| 32 | .minimum_version_id = 1, |
| 33 | .fields = (VMStateField[]) { |
| 34 | VMSTATE_UINT32(namelen, CprVFIODevice), |
| 35 | VMSTATE_VBUFFER_ALLOC_UINT32(name, CprVFIODevice, 0, NULL, namelen), |
| 36 | VMSTATE_INT32(devid, CprVFIODevice), |
| 37 | VMSTATE_UINT32(ioas_id, CprVFIODevice), |
| 38 | VMSTATE_UINT32(hwpt_id, CprVFIODevice), |
| 39 | VMSTATE_END_OF_LIST() |
| 40 | } |
| 41 | }; |
| 42 | |
| 43 | const VMStateDescription vmstate_cpr_vfio_devices = { |
| 44 | .name = CPR_STATE "/vfio devices", |
| 45 | .version_id = 1, |
| 46 | .minimum_version_id = 1, |
| 47 | .fields = (const VMStateField[]){ |
| 48 | VMSTATE_QLIST_V(vfio_devices, CprState, 1, vmstate_cpr_vfio_device, |
| 49 | CprVFIODevice, next), |
| 50 | VMSTATE_END_OF_LIST() |
| 51 | } |
| 52 | }; |
| 53 | |
| 54 | static void vfio_cpr_save_device(VFIODevice *vbasedev) |
| 55 | { |
| 56 | CprVFIODevice *elem = g_new0(CprVFIODevice, 1); |
| 57 | |
| 58 | elem->name = g_strdup(vbasedev->name); |
| 59 | elem->namelen = strlen(vbasedev->name) + 1; |
| 60 | elem->ioas_id = vbasedev->cpr.ioas_id; |
| 61 | elem->devid = vbasedev->devid; |
| 62 | elem->hwpt_id = vbasedev->cpr.hwpt_id; |
| 63 | QLIST_INSERT_HEAD(&cpr_state.vfio_devices, elem, next); |
| 64 | } |
| 65 | |
| 66 | static CprVFIODevice *find_device(const char *name) |
| 67 | { |
| 68 | CprVFIODeviceList *head = &cpr_state.vfio_devices; |
| 69 | CprVFIODevice *elem; |
| 70 | |
| 71 | QLIST_FOREACH(elem, head, next) { |
| 72 | if (!strcmp(elem->name, name)) { |
| 73 | return elem; |
| 74 | } |
| 75 | } |
| 76 | return NULL; |
| 77 | } |
| 78 | |
| 79 | static void vfio_cpr_delete_device(const char *name) |
| 80 | { |
| 81 | CprVFIODevice *elem = find_device(name); |
| 82 | |
| 83 | if (elem) { |
| 84 | QLIST_REMOVE(elem, next); |
| 85 | g_free(elem->name); |
| 86 | g_free(elem); |
| 87 | } |
| 88 | } |
| 89 | |
| 90 | static bool vfio_cpr_find_device(VFIODevice *vbasedev) |
| 91 | { |
| 92 | CprVFIODevice *elem = find_device(vbasedev->name); |
| 93 | |
| 94 | if (elem) { |
| 95 | vbasedev->cpr.ioas_id = elem->ioas_id; |
| 96 | vbasedev->devid = elem->devid; |
| 97 | vbasedev->cpr.hwpt_id = elem->hwpt_id; |
| 98 | trace_vfio_cpr_find_device(elem->ioas_id, elem->devid, elem->hwpt_id); |
| 99 | return true; |
| 100 | } |
| 101 | return false; |
| 102 | } |
| 103 | |
| 104 | static bool vfio_cpr_supported(IOMMUFDBackend *be, Error **errp) |
| 105 | { |
| 106 | if (!iommufd_change_process_capable(be)) { |
| 107 | if (errp) { |
| 108 | error_setg(errp, "vfio iommufd backend does not support " |
| 109 | "IOMMU_IOAS_CHANGE_PROCESS"); |
| 110 | } |
| 111 | return false; |
| 112 | } |
| 113 | return true; |
| 114 | } |
| 115 | |
| 116 | static int iommufd_cpr_pre_save(void *opaque) |
| 117 | { |
| 118 | IOMMUFDBackend *be = opaque; |
| 119 | |
| 120 | /* |
| 121 | * The process has not changed yet, but proactively try the ioctl, |
| 122 | * and it will fail if any DMA mappings are not supported. |
| 123 | */ |
| 124 | if (!iommufd_change_process_capable(be)) { |
| 125 | error_report("some memory regions do not support " |
| 126 | "IOMMU_IOAS_CHANGE_PROCESS"); |
| 127 | return -1; |
| 128 | } |
| 129 | return 0; |
| 130 | } |
| 131 | |
| 132 | static int iommufd_cpr_post_load(void *opaque, int version_id) |
| 133 | { |
| 134 | IOMMUFDBackend *be = opaque; |
| 135 | Error *local_err = NULL; |
| 136 | |
| 137 | if (!iommufd_change_process(be, &local_err)) { |
| 138 | error_report_err(local_err); |
| 139 | return -1; |
| 140 | } |
| 141 | return 0; |
| 142 | } |
| 143 | |
| 144 | static const VMStateDescription iommufd_cpr_vmstate = { |
| 145 | .name = "iommufd", |
| 146 | .version_id = 0, |
| 147 | .minimum_version_id = 0, |
| 148 | .pre_save = iommufd_cpr_pre_save, |
| 149 | .post_load = iommufd_cpr_post_load, |
| 150 | .needed = cpr_incoming_needed, |
| 151 | .fields = (VMStateField[]) { |
| 152 | VMSTATE_END_OF_LIST() |
| 153 | } |
| 154 | }; |
| 155 | |
| 156 | bool vfio_iommufd_cpr_register_iommufd(IOMMUFDBackend *be, Error **errp) |
| 157 | { |
| 158 | Error **cpr_blocker = &be->cpr_blocker; |
| 159 | |
| 160 | if (!vfio_cpr_supported(be, cpr_blocker)) { |
| 161 | return migrate_add_blocker_modes(cpr_blocker, |
| 162 | BIT(MIG_MODE_CPR_TRANSFER) | BIT(MIG_MODE_CPR_EXEC), |
| 163 | errp) == 0; |
| 164 | } |
| 165 | |
| 166 | vmstate_register(NULL, -1, &iommufd_cpr_vmstate, be); |
| 167 | |
| 168 | return true; |
| 169 | } |
| 170 | |
| 171 | void vfio_iommufd_cpr_unregister_iommufd(IOMMUFDBackend *be) |
| 172 | { |
| 173 | vmstate_unregister(NULL, &iommufd_cpr_vmstate, be); |
| 174 | migrate_del_blocker(&be->cpr_blocker); |
| 175 | } |
| 176 | |
| 177 | bool vfio_iommufd_cpr_register_container(VFIOIOMMUFDContainer *container, |
| 178 | Error **errp) |
| 179 | { |
| 180 | VFIOContainer *bcontainer = VFIO_IOMMU(container); |
| 181 | |
| 182 | migration_add_notifier_mode(&bcontainer->cpr_reboot_notifier, |
| 183 | vfio_cpr_reboot_notifier, |
| 184 | MIG_MODE_CPR_REBOOT); |
| 185 | |
| 186 | vfio_cpr_add_kvm_notifier(); |
| 187 | |
| 188 | return true; |
| 189 | } |
| 190 | |
| 191 | void vfio_iommufd_cpr_unregister_container(VFIOIOMMUFDContainer *container) |
| 192 | { |
| 193 | VFIOContainer *bcontainer = VFIO_IOMMU(container); |
| 194 | |
| 195 | migration_remove_notifier(&bcontainer->cpr_reboot_notifier); |
| 196 | } |
| 197 | |
| 198 | void vfio_iommufd_cpr_register_device(VFIODevice *vbasedev) |
| 199 | { |
| 200 | if (!cpr_is_incoming()) { |
| 201 | /* |
| 202 | * Beware fd may have already been saved by vfio_device_set_fd, |
| 203 | * so call resave to avoid a duplicate entry. |
| 204 | */ |
| 205 | cpr_resave_fd(vbasedev->name, 0, vbasedev->fd); |
| 206 | vfio_cpr_save_device(vbasedev); |
| 207 | } |
| 208 | } |
| 209 | |
| 210 | void vfio_iommufd_cpr_unregister_device(VFIODevice *vbasedev) |
| 211 | { |
| 212 | cpr_delete_fd(vbasedev->name, 0); |
| 213 | vfio_cpr_delete_device(vbasedev->name); |
| 214 | } |
| 215 | |
| 216 | void vfio_cpr_load_device(VFIODevice *vbasedev) |
| 217 | { |
| 218 | if (cpr_is_incoming()) { |
| 219 | bool ret = vfio_cpr_find_device(vbasedev); |
| 220 | g_assert(ret); |
| 221 | |
| 222 | if (vbasedev->fd < 0) { |
| 223 | vbasedev->fd = cpr_find_fd(vbasedev->name, 0); |
| 224 | } |
| 225 | } |
| 226 | } |