David Brazdil | 0f672f6 | 2019-12-10 10:32:29 +0000 | [diff] [blame^] | 1 | // SPDX-License-Identifier: GPL-2.0-or-later |
Andrew Scull | b4b6d4a | 2019-01-02 15:54:55 +0000 | [diff] [blame] | 2 | /* |
| 3 | * pci_dn.c |
| 4 | * |
| 5 | * Copyright (C) 2001 Todd Inglett, IBM Corporation |
| 6 | * |
| 7 | * PCI manipulation via device_nodes. |
Andrew Scull | b4b6d4a | 2019-01-02 15:54:55 +0000 | [diff] [blame] | 8 | */ |
| 9 | #include <linux/kernel.h> |
| 10 | #include <linux/pci.h> |
| 11 | #include <linux/string.h> |
| 12 | #include <linux/export.h> |
| 13 | #include <linux/init.h> |
| 14 | #include <linux/gfp.h> |
| 15 | |
| 16 | #include <asm/io.h> |
| 17 | #include <asm/prom.h> |
| 18 | #include <asm/pci-bridge.h> |
| 19 | #include <asm/ppc-pci.h> |
| 20 | #include <asm/firmware.h> |
| 21 | #include <asm/eeh.h> |
| 22 | |
| 23 | /* |
| 24 | * The function is used to find the firmware data of one |
| 25 | * specific PCI device, which is attached to the indicated |
| 26 | * PCI bus. For VFs, their firmware data is linked to that |
| 27 | * one of PF's bridge. For other devices, their firmware |
| 28 | * data is linked to that of their bridge. |
| 29 | */ |
| 30 | static struct pci_dn *pci_bus_to_pdn(struct pci_bus *bus) |
| 31 | { |
| 32 | struct pci_bus *pbus; |
| 33 | struct device_node *dn; |
| 34 | struct pci_dn *pdn; |
| 35 | |
| 36 | /* |
| 37 | * We probably have virtual bus which doesn't |
| 38 | * have associated bridge. |
| 39 | */ |
| 40 | pbus = bus; |
| 41 | while (pbus) { |
| 42 | if (pci_is_root_bus(pbus) || pbus->self) |
| 43 | break; |
| 44 | |
| 45 | pbus = pbus->parent; |
| 46 | } |
| 47 | |
| 48 | /* |
| 49 | * Except virtual bus, all PCI buses should |
| 50 | * have device nodes. |
| 51 | */ |
| 52 | dn = pci_bus_to_OF_node(pbus); |
| 53 | pdn = dn ? PCI_DN(dn) : NULL; |
| 54 | |
| 55 | return pdn; |
| 56 | } |
| 57 | |
| 58 | struct pci_dn *pci_get_pdn_by_devfn(struct pci_bus *bus, |
| 59 | int devfn) |
| 60 | { |
| 61 | struct device_node *dn = NULL; |
| 62 | struct pci_dn *parent, *pdn; |
| 63 | struct pci_dev *pdev = NULL; |
| 64 | |
| 65 | /* Fast path: fetch from PCI device */ |
| 66 | list_for_each_entry(pdev, &bus->devices, bus_list) { |
| 67 | if (pdev->devfn == devfn) { |
| 68 | if (pdev->dev.archdata.pci_data) |
| 69 | return pdev->dev.archdata.pci_data; |
| 70 | |
| 71 | dn = pci_device_to_OF_node(pdev); |
| 72 | break; |
| 73 | } |
| 74 | } |
| 75 | |
| 76 | /* Fast path: fetch from device node */ |
| 77 | pdn = dn ? PCI_DN(dn) : NULL; |
| 78 | if (pdn) |
| 79 | return pdn; |
| 80 | |
| 81 | /* Slow path: fetch from firmware data hierarchy */ |
| 82 | parent = pci_bus_to_pdn(bus); |
| 83 | if (!parent) |
| 84 | return NULL; |
| 85 | |
| 86 | list_for_each_entry(pdn, &parent->child_list, list) { |
| 87 | if (pdn->busno == bus->number && |
| 88 | pdn->devfn == devfn) |
| 89 | return pdn; |
| 90 | } |
| 91 | |
| 92 | return NULL; |
| 93 | } |
| 94 | |
| 95 | struct pci_dn *pci_get_pdn(struct pci_dev *pdev) |
| 96 | { |
| 97 | struct device_node *dn; |
| 98 | struct pci_dn *parent, *pdn; |
| 99 | |
| 100 | /* Search device directly */ |
| 101 | if (pdev->dev.archdata.pci_data) |
| 102 | return pdev->dev.archdata.pci_data; |
| 103 | |
| 104 | /* Check device node */ |
| 105 | dn = pci_device_to_OF_node(pdev); |
| 106 | pdn = dn ? PCI_DN(dn) : NULL; |
| 107 | if (pdn) |
| 108 | return pdn; |
| 109 | |
| 110 | /* |
| 111 | * VFs don't have device nodes. We hook their |
| 112 | * firmware data to PF's bridge. |
| 113 | */ |
| 114 | parent = pci_bus_to_pdn(pdev->bus); |
| 115 | if (!parent) |
| 116 | return NULL; |
| 117 | |
| 118 | list_for_each_entry(pdn, &parent->child_list, list) { |
| 119 | if (pdn->busno == pdev->bus->number && |
| 120 | pdn->devfn == pdev->devfn) |
| 121 | return pdn; |
| 122 | } |
| 123 | |
| 124 | return NULL; |
| 125 | } |
| 126 | |
| 127 | #ifdef CONFIG_PCI_IOV |
| 128 | static struct pci_dn *add_one_dev_pci_data(struct pci_dn *parent, |
| 129 | int vf_index, |
| 130 | int busno, int devfn) |
| 131 | { |
| 132 | struct pci_dn *pdn; |
| 133 | |
| 134 | /* Except PHB, we always have the parent */ |
| 135 | if (!parent) |
| 136 | return NULL; |
| 137 | |
| 138 | pdn = kzalloc(sizeof(*pdn), GFP_KERNEL); |
| 139 | if (!pdn) |
| 140 | return NULL; |
| 141 | |
| 142 | pdn->phb = parent->phb; |
| 143 | pdn->parent = parent; |
| 144 | pdn->busno = busno; |
| 145 | pdn->devfn = devfn; |
| 146 | pdn->vf_index = vf_index; |
| 147 | pdn->pe_number = IODA_INVALID_PE; |
| 148 | INIT_LIST_HEAD(&pdn->child_list); |
| 149 | INIT_LIST_HEAD(&pdn->list); |
| 150 | list_add_tail(&pdn->list, &parent->child_list); |
| 151 | |
| 152 | return pdn; |
| 153 | } |
| 154 | #endif |
| 155 | |
| 156 | struct pci_dn *add_dev_pci_data(struct pci_dev *pdev) |
| 157 | { |
| 158 | #ifdef CONFIG_PCI_IOV |
| 159 | struct pci_dn *parent, *pdn; |
| 160 | int i; |
| 161 | |
| 162 | /* Only support IOV for now */ |
| 163 | if (!pdev->is_physfn) |
| 164 | return pci_get_pdn(pdev); |
| 165 | |
| 166 | /* Check if VFs have been populated */ |
| 167 | pdn = pci_get_pdn(pdev); |
| 168 | if (!pdn || (pdn->flags & PCI_DN_FLAG_IOV_VF)) |
| 169 | return NULL; |
| 170 | |
| 171 | pdn->flags |= PCI_DN_FLAG_IOV_VF; |
| 172 | parent = pci_bus_to_pdn(pdev->bus); |
| 173 | if (!parent) |
| 174 | return NULL; |
| 175 | |
| 176 | for (i = 0; i < pci_sriov_get_totalvfs(pdev); i++) { |
| 177 | struct eeh_dev *edev __maybe_unused; |
| 178 | |
| 179 | pdn = add_one_dev_pci_data(parent, i, |
| 180 | pci_iov_virtfn_bus(pdev, i), |
| 181 | pci_iov_virtfn_devfn(pdev, i)); |
| 182 | if (!pdn) { |
| 183 | dev_warn(&pdev->dev, "%s: Cannot create firmware data for VF#%d\n", |
| 184 | __func__, i); |
| 185 | return NULL; |
| 186 | } |
| 187 | |
| 188 | #ifdef CONFIG_EEH |
| 189 | /* Create the EEH device for the VF */ |
| 190 | edev = eeh_dev_init(pdn); |
| 191 | BUG_ON(!edev); |
| 192 | edev->physfn = pdev; |
| 193 | #endif /* CONFIG_EEH */ |
| 194 | } |
| 195 | #endif /* CONFIG_PCI_IOV */ |
| 196 | |
| 197 | return pci_get_pdn(pdev); |
| 198 | } |
| 199 | |
| 200 | void remove_dev_pci_data(struct pci_dev *pdev) |
| 201 | { |
| 202 | #ifdef CONFIG_PCI_IOV |
| 203 | struct pci_dn *parent; |
| 204 | struct pci_dn *pdn, *tmp; |
| 205 | int i; |
| 206 | |
| 207 | /* |
| 208 | * VF and VF PE are created/released dynamically, so we need to |
| 209 | * bind/unbind them. Otherwise the VF and VF PE would be mismatched |
| 210 | * when re-enabling SR-IOV. |
| 211 | */ |
| 212 | if (pdev->is_virtfn) { |
| 213 | pdn = pci_get_pdn(pdev); |
| 214 | pdn->pe_number = IODA_INVALID_PE; |
| 215 | return; |
| 216 | } |
| 217 | |
| 218 | /* Only support IOV PF for now */ |
| 219 | if (!pdev->is_physfn) |
| 220 | return; |
| 221 | |
| 222 | /* Check if VFs have been populated */ |
| 223 | pdn = pci_get_pdn(pdev); |
| 224 | if (!pdn || !(pdn->flags & PCI_DN_FLAG_IOV_VF)) |
| 225 | return; |
| 226 | |
| 227 | pdn->flags &= ~PCI_DN_FLAG_IOV_VF; |
| 228 | parent = pci_bus_to_pdn(pdev->bus); |
| 229 | if (!parent) |
| 230 | return; |
| 231 | |
| 232 | /* |
| 233 | * We might introduce flag to pci_dn in future |
| 234 | * so that we can release VF's firmware data in |
| 235 | * a batch mode. |
| 236 | */ |
| 237 | for (i = 0; i < pci_sriov_get_totalvfs(pdev); i++) { |
| 238 | struct eeh_dev *edev __maybe_unused; |
| 239 | |
| 240 | list_for_each_entry_safe(pdn, tmp, |
| 241 | &parent->child_list, list) { |
| 242 | if (pdn->busno != pci_iov_virtfn_bus(pdev, i) || |
| 243 | pdn->devfn != pci_iov_virtfn_devfn(pdev, i)) |
| 244 | continue; |
| 245 | |
| 246 | #ifdef CONFIG_EEH |
| 247 | /* Release EEH device for the VF */ |
| 248 | edev = pdn_to_eeh_dev(pdn); |
| 249 | if (edev) { |
| 250 | pdn->edev = NULL; |
| 251 | kfree(edev); |
| 252 | } |
| 253 | #endif /* CONFIG_EEH */ |
| 254 | |
| 255 | if (!list_empty(&pdn->list)) |
| 256 | list_del(&pdn->list); |
| 257 | |
| 258 | kfree(pdn); |
| 259 | } |
| 260 | } |
| 261 | #endif /* CONFIG_PCI_IOV */ |
| 262 | } |
| 263 | |
| 264 | struct pci_dn *pci_add_device_node_info(struct pci_controller *hose, |
| 265 | struct device_node *dn) |
| 266 | { |
| 267 | const __be32 *type = of_get_property(dn, "ibm,pci-config-space-type", NULL); |
| 268 | const __be32 *regs; |
| 269 | struct device_node *parent; |
| 270 | struct pci_dn *pdn; |
| 271 | #ifdef CONFIG_EEH |
| 272 | struct eeh_dev *edev; |
| 273 | #endif |
| 274 | |
| 275 | pdn = kzalloc(sizeof(*pdn), GFP_KERNEL); |
| 276 | if (pdn == NULL) |
| 277 | return NULL; |
| 278 | dn->data = pdn; |
| 279 | pdn->phb = hose; |
| 280 | pdn->pe_number = IODA_INVALID_PE; |
| 281 | regs = of_get_property(dn, "reg", NULL); |
| 282 | if (regs) { |
| 283 | u32 addr = of_read_number(regs, 1); |
| 284 | |
| 285 | /* First register entry is addr (00BBSS00) */ |
| 286 | pdn->busno = (addr >> 16) & 0xff; |
| 287 | pdn->devfn = (addr >> 8) & 0xff; |
| 288 | } |
| 289 | |
| 290 | /* vendor/device IDs and class code */ |
| 291 | regs = of_get_property(dn, "vendor-id", NULL); |
| 292 | pdn->vendor_id = regs ? of_read_number(regs, 1) : 0; |
| 293 | regs = of_get_property(dn, "device-id", NULL); |
| 294 | pdn->device_id = regs ? of_read_number(regs, 1) : 0; |
| 295 | regs = of_get_property(dn, "class-code", NULL); |
| 296 | pdn->class_code = regs ? of_read_number(regs, 1) : 0; |
| 297 | |
| 298 | /* Extended config space */ |
| 299 | pdn->pci_ext_config_space = (type && of_read_number(type, 1) == 1); |
| 300 | |
| 301 | /* Create EEH device */ |
| 302 | #ifdef CONFIG_EEH |
| 303 | edev = eeh_dev_init(pdn); |
| 304 | if (!edev) { |
| 305 | kfree(pdn); |
| 306 | return NULL; |
| 307 | } |
| 308 | #endif |
| 309 | |
| 310 | /* Attach to parent node */ |
| 311 | INIT_LIST_HEAD(&pdn->child_list); |
| 312 | INIT_LIST_HEAD(&pdn->list); |
| 313 | parent = of_get_parent(dn); |
| 314 | pdn->parent = parent ? PCI_DN(parent) : NULL; |
| 315 | if (pdn->parent) |
| 316 | list_add_tail(&pdn->list, &pdn->parent->child_list); |
| 317 | |
| 318 | return pdn; |
| 319 | } |
| 320 | EXPORT_SYMBOL_GPL(pci_add_device_node_info); |
| 321 | |
| 322 | void pci_remove_device_node_info(struct device_node *dn) |
| 323 | { |
| 324 | struct pci_dn *pdn = dn ? PCI_DN(dn) : NULL; |
| 325 | struct device_node *parent; |
David Brazdil | 0f672f6 | 2019-12-10 10:32:29 +0000 | [diff] [blame^] | 326 | struct pci_dev *pdev; |
Andrew Scull | b4b6d4a | 2019-01-02 15:54:55 +0000 | [diff] [blame] | 327 | #ifdef CONFIG_EEH |
| 328 | struct eeh_dev *edev = pdn_to_eeh_dev(pdn); |
| 329 | |
| 330 | if (edev) |
| 331 | edev->pdn = NULL; |
| 332 | #endif |
| 333 | |
| 334 | if (!pdn) |
| 335 | return; |
| 336 | |
| 337 | WARN_ON(!list_empty(&pdn->child_list)); |
| 338 | list_del(&pdn->list); |
| 339 | |
David Brazdil | 0f672f6 | 2019-12-10 10:32:29 +0000 | [diff] [blame^] | 340 | /* Drop the parent pci_dn's ref to our backing dt node */ |
Andrew Scull | b4b6d4a | 2019-01-02 15:54:55 +0000 | [diff] [blame] | 341 | parent = of_get_parent(dn); |
| 342 | if (parent) |
| 343 | of_node_put(parent); |
| 344 | |
David Brazdil | 0f672f6 | 2019-12-10 10:32:29 +0000 | [diff] [blame^] | 345 | /* |
| 346 | * At this point we *might* still have a pci_dev that was |
| 347 | * instantiated from this pci_dn. So defer free()ing it until |
| 348 | * the pci_dev's release function is called. |
| 349 | */ |
| 350 | pdev = pci_get_domain_bus_and_slot(pdn->phb->global_number, |
| 351 | pdn->busno, pdn->devfn); |
| 352 | if (pdev) { |
| 353 | /* NB: pdev has a ref to dn */ |
| 354 | pci_dbg(pdev, "marked pdn (from %pOF) as dead\n", dn); |
| 355 | pdn->flags |= PCI_DN_FLAG_DEAD; |
| 356 | } else { |
| 357 | dn->data = NULL; |
| 358 | kfree(pdn); |
| 359 | } |
| 360 | |
| 361 | pci_dev_put(pdev); |
Andrew Scull | b4b6d4a | 2019-01-02 15:54:55 +0000 | [diff] [blame] | 362 | } |
| 363 | EXPORT_SYMBOL_GPL(pci_remove_device_node_info); |
| 364 | |
| 365 | /* |
| 366 | * Traverse a device tree stopping each PCI device in the tree. |
| 367 | * This is done depth first. As each node is processed, a "pre" |
| 368 | * function is called and the children are processed recursively. |
| 369 | * |
| 370 | * The "pre" func returns a value. If non-zero is returned from |
| 371 | * the "pre" func, the traversal stops and this value is returned. |
| 372 | * This return value is useful when using traverse as a method of |
| 373 | * finding a device. |
| 374 | * |
| 375 | * NOTE: we do not run the func for devices that do not appear to |
| 376 | * be PCI except for the start node which we assume (this is good |
| 377 | * because the start node is often a phb which may be missing PCI |
| 378 | * properties). |
| 379 | * We use the class-code as an indicator. If we run into |
| 380 | * one of these nodes we also assume its siblings are non-pci for |
| 381 | * performance. |
| 382 | */ |
| 383 | void *pci_traverse_device_nodes(struct device_node *start, |
| 384 | void *(*fn)(struct device_node *, void *), |
| 385 | void *data) |
| 386 | { |
| 387 | struct device_node *dn, *nextdn; |
| 388 | void *ret; |
| 389 | |
| 390 | /* We started with a phb, iterate all childs */ |
| 391 | for (dn = start->child; dn; dn = nextdn) { |
| 392 | const __be32 *classp; |
| 393 | u32 class = 0; |
| 394 | |
| 395 | nextdn = NULL; |
| 396 | classp = of_get_property(dn, "class-code", NULL); |
| 397 | if (classp) |
| 398 | class = of_read_number(classp, 1); |
| 399 | |
| 400 | if (fn) { |
| 401 | ret = fn(dn, data); |
| 402 | if (ret) |
| 403 | return ret; |
| 404 | } |
| 405 | |
| 406 | /* If we are a PCI bridge, go down */ |
| 407 | if (dn->child && ((class >> 8) == PCI_CLASS_BRIDGE_PCI || |
| 408 | (class >> 8) == PCI_CLASS_BRIDGE_CARDBUS)) |
| 409 | /* Depth first...do children */ |
| 410 | nextdn = dn->child; |
| 411 | else if (dn->sibling) |
| 412 | /* ok, try next sibling instead. */ |
| 413 | nextdn = dn->sibling; |
| 414 | if (!nextdn) { |
| 415 | /* Walk up to next valid sibling. */ |
| 416 | do { |
| 417 | dn = dn->parent; |
| 418 | if (dn == start) |
| 419 | return NULL; |
| 420 | } while (dn->sibling == NULL); |
| 421 | nextdn = dn->sibling; |
| 422 | } |
| 423 | } |
| 424 | return NULL; |
| 425 | } |
| 426 | EXPORT_SYMBOL_GPL(pci_traverse_device_nodes); |
| 427 | |
| 428 | static struct pci_dn *pci_dn_next_one(struct pci_dn *root, |
| 429 | struct pci_dn *pdn) |
| 430 | { |
| 431 | struct list_head *next = pdn->child_list.next; |
| 432 | |
| 433 | if (next != &pdn->child_list) |
| 434 | return list_entry(next, struct pci_dn, list); |
| 435 | |
| 436 | while (1) { |
| 437 | if (pdn == root) |
| 438 | return NULL; |
| 439 | |
| 440 | next = pdn->list.next; |
| 441 | if (next != &pdn->parent->child_list) |
| 442 | break; |
| 443 | |
| 444 | pdn = pdn->parent; |
| 445 | } |
| 446 | |
| 447 | return list_entry(next, struct pci_dn, list); |
| 448 | } |
| 449 | |
| 450 | void *traverse_pci_dn(struct pci_dn *root, |
| 451 | void *(*fn)(struct pci_dn *, void *), |
| 452 | void *data) |
| 453 | { |
| 454 | struct pci_dn *pdn = root; |
| 455 | void *ret; |
| 456 | |
| 457 | /* Only scan the child nodes */ |
| 458 | for (pdn = pci_dn_next_one(root, pdn); pdn; |
| 459 | pdn = pci_dn_next_one(root, pdn)) { |
| 460 | ret = fn(pdn, data); |
| 461 | if (ret) |
| 462 | return ret; |
| 463 | } |
| 464 | |
| 465 | return NULL; |
| 466 | } |
| 467 | |
| 468 | static void *add_pdn(struct device_node *dn, void *data) |
| 469 | { |
| 470 | struct pci_controller *hose = data; |
| 471 | struct pci_dn *pdn; |
| 472 | |
| 473 | pdn = pci_add_device_node_info(hose, dn); |
| 474 | if (!pdn) |
| 475 | return ERR_PTR(-ENOMEM); |
| 476 | |
| 477 | return NULL; |
| 478 | } |
| 479 | |
| 480 | /** |
| 481 | * pci_devs_phb_init_dynamic - setup pci devices under this PHB |
| 482 | * phb: pci-to-host bridge (top-level bridge connecting to cpu) |
| 483 | * |
| 484 | * This routine is called both during boot, (before the memory |
| 485 | * subsystem is set up, before kmalloc is valid) and during the |
| 486 | * dynamic lpar operation of adding a PHB to a running system. |
| 487 | */ |
| 488 | void pci_devs_phb_init_dynamic(struct pci_controller *phb) |
| 489 | { |
| 490 | struct device_node *dn = phb->dn; |
| 491 | struct pci_dn *pdn; |
| 492 | |
| 493 | /* PHB nodes themselves must not match */ |
| 494 | pdn = pci_add_device_node_info(phb, dn); |
| 495 | if (pdn) { |
| 496 | pdn->devfn = pdn->busno = -1; |
| 497 | pdn->vendor_id = pdn->device_id = pdn->class_code = 0; |
| 498 | pdn->phb = phb; |
| 499 | phb->pci_data = pdn; |
| 500 | } |
| 501 | |
| 502 | /* Update dn->phb ptrs for new phb and children devices */ |
| 503 | pci_traverse_device_nodes(dn, add_pdn, phb); |
| 504 | } |
| 505 | |
| 506 | /** |
| 507 | * pci_devs_phb_init - Initialize phbs and pci devs under them. |
| 508 | * |
| 509 | * This routine walks over all phb's (pci-host bridges) on the |
| 510 | * system, and sets up assorted pci-related structures |
| 511 | * (including pci info in the device node structs) for each |
| 512 | * pci device found underneath. This routine runs once, |
| 513 | * early in the boot sequence. |
| 514 | */ |
| 515 | static int __init pci_devs_phb_init(void) |
| 516 | { |
| 517 | struct pci_controller *phb, *tmp; |
| 518 | |
| 519 | /* This must be done first so the device nodes have valid pci info! */ |
| 520 | list_for_each_entry_safe(phb, tmp, &hose_list, list_node) |
| 521 | pci_devs_phb_init_dynamic(phb); |
| 522 | |
| 523 | return 0; |
| 524 | } |
| 525 | |
| 526 | core_initcall(pci_devs_phb_init); |
| 527 | |
| 528 | static void pci_dev_pdn_setup(struct pci_dev *pdev) |
| 529 | { |
| 530 | struct pci_dn *pdn; |
| 531 | |
| 532 | if (pdev->dev.archdata.pci_data) |
| 533 | return; |
| 534 | |
| 535 | /* Setup the fast path */ |
| 536 | pdn = pci_get_pdn(pdev); |
| 537 | pdev->dev.archdata.pci_data = pdn; |
| 538 | } |
| 539 | DECLARE_PCI_FIXUP_EARLY(PCI_ANY_ID, PCI_ANY_ID, pci_dev_pdn_setup); |