| 1 | |
| 2 | |
| 3 | * Virtual I/O topology
|
| 4 | *
|
| 5 | * The Virtual I/O Translation Table (VIOT) describes the topology of
|
| 6 | * para-virtual IOMMUs and the endpoints they manage. The OS uses it to
|
| 7 | * initialize devices in the right order, preventing endpoints from issuing DMA
|
| 8 | * before their IOMMU is ready.
|
| 9 | *
|
| 10 | * When binding a driver to a device, before calling the device driver's probe()
|
| 11 | * method, the driver infrastructure calls dma_configure(). At that point the
|
| 12 | * VIOT driver looks for an IOMMU associated to the device in the VIOT table.
|
| 13 | * If an IOMMU exists and has been initialized, the VIOT driver initializes the
|
| 14 | * device's IOMMU fwspec, allowing the DMA infrastructure to invoke the IOMMU
|
| 15 | * ops when the device driver configures DMA mappings. If an IOMMU exists and
|
| 16 | * hasn't yet been initialized, VIOT returns -EPROBE_DEFER to postpone probing
|
| 17 | * the device until the IOMMU is available.
|
| 18 |
|
| 19 | #define pr_fmt(fmt) "ACPI: VIOT: " fmt
|
| 20 |
|
| 21 | #include <linux/acpi_viot.h>
|
| 22 | #include <linux/iommu.h>
|
| 23 | #include <linux/list.h>
|
| 24 | #include <linux/pci.h>
|
| 25 | #include <linux/platform_device.h>
|
| 26 | #include <linux/property.h>
|
| 27 |
|
| 28 | struct viot_iommu {
|
| 29 |
|
| 30 | unsigned int offset;
|
| 31 | struct fwnode_handle *fwnode;
|
| 32 | struct list_head list;
|
| 33 | };
|
| 34 |
|
| 35 | struct viot_endpoint {
|
| 36 | union {
|
| 37 |
|
| 38 | struct {
|
| 39 | u16 segment_start;
|
| 40 | u16 segment_end;
|
| 41 | u16 bdf_start;
|
| 42 | u16 bdf_end;
|
| 43 | };
|
| 44 |
|
| 45 | u64 address;
|
| 46 | };
|
| 47 | u32 endpoint_id;
|
| 48 | struct viot_iommu *viommu;
|
| 49 | struct list_head list;
|
| 50 | };
|
| 51 |
|
| 52 | static struct acpi_table_viot *viot;
|
| 53 | static LIST_HEAD(viot_iommus);
|
| 54 | static LIST_HEAD(viot_pci_ranges);
|
| 55 | static LIST_HEAD(viot_mmio_endpoints);
|
| 56 |
|
| 57 | static int __init viot_check_bounds(const struct acpi_viot_header *hdr)
|
| 58 | {
|
| 59 | struct acpi_viot_header *start, *end, *hdr_end;
|
| 60 |
|
| 61 | start = ACPI_ADD_PTR(struct acpi_viot_header, viot,
|
| 62 | max_t(size_t, sizeof(*viot), viot->node_offset));
|
| 63 | end = ACPI_ADD_PTR(struct acpi_viot_header, viot, viot->header.length);
|
| 64 | hdr_end = ACPI_ADD_PTR(struct acpi_viot_header, hdr, sizeof(*hdr));
|
| 65 |
|
| 66 | if (hdr < start || hdr_end > end) {
|
| 67 | pr_err(FW_BUG "Node pointer overflows\n");
|
| 68 | return -EOVERFLOW;
|
| 69 | }
|
| 70 | if (hdr->length < sizeof(*hdr)) {
|
| 71 | pr_err(FW_BUG "Empty node\n");
|
| 72 | return -EINVAL;
|
| 73 | }
|
| 74 | return 0;
|
| 75 | }
|
| 76 |
|
| 77 | static int __init viot_get_pci_iommu_fwnode(struct viot_iommu *viommu,
|
| 78 | u16 segment, u16 bdf)
|
| 79 | {
|
| 80 | struct pci_dev *pdev;
|
| 81 | struct fwnode_handle *fwnode;
|
| 82 |
|
| 83 | pdev = pci_get_domain_bus_and_slot(segment, PCI_BUS_NUM(bdf),
|
| 84 | bdf & 0xff);
|
| 85 | if (!pdev) {
|
| 86 | pr_err("Could not find PCI IOMMU\n");
|
| 87 | return -ENODEV;
|
| 88 | }
|
| 89 |
|
| 90 | fwnode = dev_fwnode(&pdev->dev);
|
| 91 | if (!fwnode) {
|
| 92 | |
| 93 | * PCI devices aren't necessarily described by ACPI. Create a
|
| 94 | * fwnode so the IOMMU subsystem can identify this device.
|
| 95 |
|
| 96 | fwnode = acpi_alloc_fwnode_static();
|
| 97 | if (!fwnode) {
|
| 98 | pci_dev_put(pdev);
|
| 99 | return -ENOMEM;
|
| 100 | }
|
| 101 | set_primary_fwnode(&pdev->dev, fwnode);
|
| 102 | }
|
| 103 | viommu->fwnode = dev_fwnode(&pdev->dev);
|
| 104 | pci_dev_put(pdev);
|
| 105 | return 0;
|
| 106 | }
|
| 107 |
|
| 108 | static int __init viot_get_mmio_iommu_fwnode(struct viot_iommu *viommu,
|
| 109 | u64 address)
|
| 110 | {
|
| 111 | struct acpi_device *adev;
|
| 112 | struct resource res = {
|
| 113 | .start = address,
|
| 114 | .end = address,
|
| 115 | .flags = IORESOURCE_MEM,
|
| 116 | };
|
| 117 |
|
| 118 | adev = acpi_resource_consumer(&res);
|
| 119 | if (!adev) {
|
| 120 | pr_err("Could not find MMIO IOMMU\n");
|
| 121 | return -EINVAL;
|
| 122 | }
|
| 123 | viommu->fwnode = &adev->fwnode;
|
| 124 | return 0;
|
| 125 | }
|
| 126 |
|
| 127 | static struct viot_iommu * __init viot_get_iommu(unsigned int offset)
|
| 128 | {
|
| 129 | int ret;
|
| 130 | struct viot_iommu *viommu;
|
| 131 | struct acpi_viot_header *hdr = ACPI_ADD_PTR(struct acpi_viot_header,
|
| 132 | viot, offset);
|
| 133 | union {
|
| 134 | struct acpi_viot_virtio_iommu_pci pci;
|
| 135 | struct acpi_viot_virtio_iommu_mmio mmio;
|
| 136 | } *node = (void *)hdr;
|
| 137 |
|
| 138 | list_for_each_entry(viommu, &viot_iommus, list)
|
| 139 | if (viommu->offset == offset)
|
| 140 | return viommu;
|
| 141 |
|
| 142 | if (viot_check_bounds(hdr))
|
| 143 | return NULL;
|
| 144 |
|
| 145 | viommu = kzalloc_obj(*viommu);
|
| 146 | if (!viommu)
|
| 147 | return NULL;
|
| 148 |
|
| 149 | viommu->offset = offset;
|
| 150 | switch (hdr->type) {
|
| 151 | case ACPI_VIOT_NODE_VIRTIO_IOMMU_PCI:
|
| 152 | if (hdr->length < sizeof(node->pci))
|
| 153 | goto err_free;
|
| 154 |
|
| 155 | ret = viot_get_pci_iommu_fwnode(viommu, node->pci.segment,
|
| 156 | node->pci.bdf);
|
| 157 | break;
|
| 158 | case ACPI_VIOT_NODE_VIRTIO_IOMMU_MMIO:
|
| 159 | if (hdr->length < sizeof(node->mmio))
|
| 160 | goto err_free;
|
| 161 |
|
| 162 | ret = viot_get_mmio_iommu_fwnode(viommu,
|
| 163 | node->mmio.base_address);
|
| 164 | break;
|
| 165 | default:
|
| 166 | ret = -EINVAL;
|
| 167 | }
|
| 168 | if (ret)
|
| 169 | goto err_free;
|
| 170 |
|
| 171 | list_add(&viommu->list, &viot_iommus);
|
| 172 | return viommu;
|
| 173 |
|
| 174 | err_free:
|
| 175 | kfree(viommu);
|
| 176 | return NULL;
|
| 177 | }
|
| 178 |
|
| 179 | static int __init viot_parse_node(const struct acpi_viot_header *hdr)
|
| 180 | {
|
| 181 | int ret = -EINVAL;
|
| 182 | struct list_head *list;
|
| 183 | struct viot_endpoint *ep;
|
| 184 | union {
|
| 185 | struct acpi_viot_mmio mmio;
|
| 186 | struct acpi_viot_pci_range pci;
|
| 187 | } *node = (void *)hdr;
|
| 188 |
|
| 189 | if (viot_check_bounds(hdr))
|
| 190 | return -EINVAL;
|
| 191 |
|
| 192 | if (hdr->type == ACPI_VIOT_NODE_VIRTIO_IOMMU_PCI ||
|
| 193 | hdr->type == ACPI_VIOT_NODE_VIRTIO_IOMMU_MMIO)
|
| 194 | return 0;
|
| 195 |
|
| 196 | ep = kzalloc_obj(*ep);
|
| 197 | if (!ep)
|
| 198 | return -ENOMEM;
|
| 199 |
|
| 200 | switch (hdr->type) {
|
| 201 | case ACPI_VIOT_NODE_PCI_RANGE:
|
| 202 | if (hdr->length < sizeof(node->pci)) {
|
| 203 | pr_err(FW_BUG "Invalid PCI node size\n");
|
| 204 | goto err_free;
|
| 205 | }
|
| 206 |
|
| 207 | ep->segment_start = node->pci.segment_start;
|
| 208 | ep->segment_end = node->pci.segment_end;
|
| 209 | ep->bdf_start = node->pci.bdf_start;
|
| 210 | ep->bdf_end = node->pci.bdf_end;
|
| 211 | ep->endpoint_id = node->pci.endpoint_start;
|
| 212 | ep->viommu = viot_get_iommu(node->pci.output_node);
|
| 213 | list = &viot_pci_ranges;
|
| 214 | break;
|
| 215 | case ACPI_VIOT_NODE_MMIO:
|
| 216 | if (hdr->length < sizeof(node->mmio)) {
|
| 217 | pr_err(FW_BUG "Invalid MMIO node size\n");
|
| 218 | goto err_free;
|
| 219 | }
|
| 220 |
|
| 221 | ep->address = node->mmio.base_address;
|
| 222 | ep->endpoint_id = node->mmio.endpoint;
|
| 223 | ep->viommu = viot_get_iommu(node->mmio.output_node);
|
| 224 | list = &viot_mmio_endpoints;
|
| 225 | break;
|
| 226 | default:
|
| 227 | pr_warn("Unsupported node %x\n", hdr->type);
|
| 228 | ret = 0;
|
| 229 | goto err_free;
|
| 230 | }
|
| 231 |
|
| 232 | if (!ep->viommu) {
|
| 233 | pr_warn("No IOMMU node found\n");
|
| 234 | |
| 235 | * A future version of the table may use the node for other
|
| 236 | * purposes. Keep parsing.
|
| 237 |
|
| 238 | ret = 0;
|
| 239 | goto err_free;
|
| 240 | }
|
| 241 |
|
| 242 | list_add(&ep->list, list);
|
| 243 | return 0;
|
| 244 |
|
| 245 | err_free:
|
| 246 | kfree(ep);
|
| 247 | return ret;
|
| 248 | }
|
| 249 |
|
| 250 | |
| 251 | * acpi_viot_early_init - Test the presence of VIOT and enable ACS
|
| 252 | *
|
| 253 | * If the VIOT does exist, ACS must be enabled. This cannot be
|
| 254 | * done in acpi_viot_init() which is called after the bus scan
|
| 255 |
|
| 256 | void __init acpi_viot_early_init(void)
|
| 257 | {
|
| 258 | #ifdef CONFIG_PCI
|
| 259 | acpi_status status;
|
| 260 | struct acpi_table_header *hdr;
|
| 261 |
|
| 262 | status = acpi_get_table(ACPI_SIG_VIOT, 0, &hdr);
|
| 263 | if (ACPI_FAILURE(status))
|
| 264 | return;
|
| 265 | pci_request_acs();
|
| 266 | acpi_put_table(hdr);
|
| 267 | #endif
|
| 268 | }
|
| 269 |
|
| 270 | |
| 271 | * acpi_viot_init - Parse the VIOT table
|
| 272 | *
|
| 273 | * Parse the VIOT table, prepare the list of endpoints to be used during DMA
|
| 274 | * setup of devices.
|
| 275 |
|
| 276 | void __init acpi_viot_init(void)
|
| 277 | {
|
| 278 | int i;
|
| 279 | acpi_status status;
|
| 280 | struct acpi_table_header *hdr;
|
| 281 | struct acpi_viot_header *node;
|
| 282 |
|
| 283 | status = acpi_get_table(ACPI_SIG_VIOT, 0, &hdr);
|
| 284 | if (ACPI_FAILURE(status)) {
|
| 285 | if (status != AE_NOT_FOUND) {
|
| 286 | const char *msg = acpi_format_exception(status);
|
| 287 |
|
| 288 | pr_err("Failed to get table, %s\n", msg);
|
| 289 | }
|
| 290 | return;
|
| 291 | }
|
| 292 |
|
| 293 | viot = (void *)hdr;
|
| 294 |
|
| 295 | node = ACPI_ADD_PTR(struct acpi_viot_header, viot, viot->node_offset);
|
| 296 | for (i = 0; i < viot->node_count; i++) {
|
| 297 | if (viot_parse_node(node))
|
| 298 | return;
|
| 299 |
|
| 300 | node = ACPI_ADD_PTR(struct acpi_viot_header, node,
|
| 301 | node->length);
|
| 302 | }
|
| 303 |
|
| 304 | acpi_put_table(hdr);
|
| 305 | }
|
| 306 |
|
| 307 | static int viot_dev_iommu_init(struct device *dev, struct viot_iommu *viommu,
|
| 308 | u32 epid)
|
| 309 | {
|
| 310 | if (!viommu || !IS_ENABLED(CONFIG_VIRTIO_IOMMU))
|
| 311 | return -ENODEV;
|
| 312 |
|
| 313 |
|
| 314 | if (device_match_fwnode(dev, viommu->fwnode))
|
| 315 | return -EINVAL;
|
| 316 |
|
| 317 | return acpi_iommu_fwspec_init(dev, epid, viommu->fwnode);
|
| 318 | }
|
| 319 |
|
| 320 | static int viot_pci_dev_iommu_init(struct pci_dev *pdev, u16 dev_id, void *data)
|
| 321 | {
|
| 322 | u32 epid;
|
| 323 | struct viot_endpoint *ep;
|
| 324 | struct device *aliased_dev = data;
|
| 325 | u32 domain_nr = pci_domain_nr(pdev->bus);
|
| 326 |
|
| 327 | list_for_each_entry(ep, &viot_pci_ranges, list) {
|
| 328 | if (domain_nr >= ep->segment_start &&
|
| 329 | domain_nr <= ep->segment_end &&
|
| 330 | dev_id >= ep->bdf_start &&
|
| 331 | dev_id <= ep->bdf_end) {
|
| 332 | epid = ((domain_nr - ep->segment_start) << 16) +
|
| 333 | dev_id - ep->bdf_start + ep->endpoint_id;
|
| 334 |
|
| 335 | return viot_dev_iommu_init(aliased_dev, ep->viommu,
|
| 336 | epid);
|
| 337 | }
|
| 338 | }
|
| 339 | return -ENODEV;
|
| 340 | }
|
| 341 |
|
| 342 | static int viot_mmio_dev_iommu_init(struct platform_device *pdev)
|
| 343 | {
|
| 344 | struct resource *mem;
|
| 345 | struct viot_endpoint *ep;
|
| 346 |
|
| 347 | mem = platform_get_resource(pdev, IORESOURCE_MEM, 0);
|
| 348 | if (!mem)
|
| 349 | return -ENODEV;
|
| 350 |
|
| 351 | list_for_each_entry(ep, &viot_mmio_endpoints, list) {
|
| 352 | if (ep->address == mem->start)
|
| 353 | return viot_dev_iommu_init(&pdev->dev, ep->viommu,
|
| 354 | ep->endpoint_id);
|
| 355 | }
|
| 356 | return -ENODEV;
|
| 357 | }
|
| 358 |
|
| 359 | |
| 360 | * viot_iommu_configure - Setup IOMMU ops for an endpoint described by VIOT
|
| 361 | * @dev: the endpoint
|
| 362 | *
|
| 363 | * Return: 0 on success, <0 on failure
|
| 364 |
|
| 365 | int viot_iommu_configure(struct device *dev)
|
| 366 | {
|
| 367 | if (dev_is_pci(dev))
|
| 368 | return pci_for_each_dma_alias(to_pci_dev(dev),
|
| 369 | viot_pci_dev_iommu_init, dev);
|
| 370 | else if (dev_is_platform(dev))
|
| 371 | return viot_mmio_dev_iommu_init(to_platform_device(dev));
|
| 372 | return -ENODEV;
|
| 373 | }
|