1 // SPDX-License-Identifier: GPL-2.0-or-later
2 /*
3 * pci_dn.c
4 *
5 * Copyright (C) 2001 Todd Inglett, IBM Corporation
6 *
7 * PCI manipulation via device_nodes.
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 */
pci_bus_to_pdn(struct pci_bus * bus)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
pci_get_pdn_by_devfn(struct pci_bus * bus,int devfn)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
pci_get_pdn(struct pci_dev * pdev)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
add_one_dev_pci_data(struct pci_dn * parent,int vf_index,int busno,int devfn)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
add_dev_pci_data(struct pci_dev * pdev)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
remove_dev_pci_data(struct pci_dev * pdev)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 /*
248 * Release EEH state for this VF. The PCI core
249 * has already torn down the pci_dev for this VF, but
250 * we're responsible to removing the eeh_dev since it
251 * has the same lifetime as the pci_dn that spawned it.
252 */
253 edev = pdn_to_eeh_dev(pdn);
254 if (edev) {
255 /*
256 * We allocate pci_dn's for the totalvfs count,
257 * but only only the vfs that were activated
258 * have a configured PE.
259 */
260 if (edev->pe)
261 eeh_rmv_from_parent_pe(edev);
262
263 pdn->edev = NULL;
264 kfree(edev);
265 }
266 #endif /* CONFIG_EEH */
267
268 if (!list_empty(&pdn->list))
269 list_del(&pdn->list);
270
271 kfree(pdn);
272 }
273 }
274 #endif /* CONFIG_PCI_IOV */
275 }
276
pci_add_device_node_info(struct pci_controller * hose,struct device_node * dn)277 struct pci_dn *pci_add_device_node_info(struct pci_controller *hose,
278 struct device_node *dn)
279 {
280 const __be32 *type = of_get_property(dn, "ibm,pci-config-space-type", NULL);
281 const __be32 *regs;
282 struct device_node *parent;
283 struct pci_dn *pdn;
284 #ifdef CONFIG_EEH
285 struct eeh_dev *edev;
286 #endif
287
288 pdn = kzalloc(sizeof(*pdn), GFP_KERNEL);
289 if (pdn == NULL)
290 return NULL;
291 dn->data = pdn;
292 pdn->phb = hose;
293 pdn->pe_number = IODA_INVALID_PE;
294 regs = of_get_property(dn, "reg", NULL);
295 if (regs) {
296 u32 addr = of_read_number(regs, 1);
297
298 /* First register entry is addr (00BBSS00) */
299 pdn->busno = (addr >> 16) & 0xff;
300 pdn->devfn = (addr >> 8) & 0xff;
301 }
302
303 /* vendor/device IDs and class code */
304 regs = of_get_property(dn, "vendor-id", NULL);
305 pdn->vendor_id = regs ? of_read_number(regs, 1) : 0;
306 regs = of_get_property(dn, "device-id", NULL);
307 pdn->device_id = regs ? of_read_number(regs, 1) : 0;
308 regs = of_get_property(dn, "class-code", NULL);
309 pdn->class_code = regs ? of_read_number(regs, 1) : 0;
310
311 /* Extended config space */
312 pdn->pci_ext_config_space = (type && of_read_number(type, 1) == 1);
313
314 /* Create EEH device */
315 #ifdef CONFIG_EEH
316 edev = eeh_dev_init(pdn);
317 if (!edev) {
318 kfree(pdn);
319 return NULL;
320 }
321 #endif
322
323 /* Attach to parent node */
324 INIT_LIST_HEAD(&pdn->child_list);
325 INIT_LIST_HEAD(&pdn->list);
326 parent = of_get_parent(dn);
327 pdn->parent = parent ? PCI_DN(parent) : NULL;
328 of_node_put(parent);
329 if (pdn->parent)
330 list_add_tail(&pdn->list, &pdn->parent->child_list);
331
332 return pdn;
333 }
334 EXPORT_SYMBOL_GPL(pci_add_device_node_info);
335
pci_remove_device_node_info(struct device_node * dn)336 void pci_remove_device_node_info(struct device_node *dn)
337 {
338 struct pci_dn *pdn = dn ? PCI_DN(dn) : NULL;
339 struct device_node *parent;
340 struct pci_dev *pdev;
341 #ifdef CONFIG_EEH
342 struct eeh_dev *edev = pdn_to_eeh_dev(pdn);
343
344 if (edev)
345 edev->pdn = NULL;
346 #endif
347
348 if (!pdn)
349 return;
350
351 WARN_ON(!list_empty(&pdn->child_list));
352 list_del(&pdn->list);
353
354 /* Drop the parent pci_dn's ref to our backing dt node */
355 parent = of_get_parent(dn);
356 if (parent)
357 of_node_put(parent);
358
359 /*
360 * At this point we *might* still have a pci_dev that was
361 * instantiated from this pci_dn. So defer free()ing it until
362 * the pci_dev's release function is called.
363 */
364 pdev = pci_get_domain_bus_and_slot(pdn->phb->global_number,
365 pdn->busno, pdn->devfn);
366 if (pdev) {
367 /* NB: pdev has a ref to dn */
368 pci_dbg(pdev, "marked pdn (from %pOF) as dead\n", dn);
369 pdn->flags |= PCI_DN_FLAG_DEAD;
370 } else {
371 dn->data = NULL;
372 kfree(pdn);
373 }
374
375 pci_dev_put(pdev);
376 }
377 EXPORT_SYMBOL_GPL(pci_remove_device_node_info);
378
379 /*
380 * Traverse a device tree stopping each PCI device in the tree.
381 * This is done depth first. As each node is processed, a "pre"
382 * function is called and the children are processed recursively.
383 *
384 * The "pre" func returns a value. If non-zero is returned from
385 * the "pre" func, the traversal stops and this value is returned.
386 * This return value is useful when using traverse as a method of
387 * finding a device.
388 *
389 * NOTE: we do not run the func for devices that do not appear to
390 * be PCI except for the start node which we assume (this is good
391 * because the start node is often a phb which may be missing PCI
392 * properties).
393 * We use the class-code as an indicator. If we run into
394 * one of these nodes we also assume its siblings are non-pci for
395 * performance.
396 */
pci_traverse_device_nodes(struct device_node * start,void * (* fn)(struct device_node *,void *),void * data)397 void *pci_traverse_device_nodes(struct device_node *start,
398 void *(*fn)(struct device_node *, void *),
399 void *data)
400 {
401 struct device_node *dn, *nextdn;
402 void *ret;
403
404 /* We started with a phb, iterate all childs */
405 for (dn = start->child; dn; dn = nextdn) {
406 const __be32 *classp;
407 u32 class = 0;
408
409 nextdn = NULL;
410 classp = of_get_property(dn, "class-code", NULL);
411 if (classp)
412 class = of_read_number(classp, 1);
413
414 if (fn) {
415 ret = fn(dn, data);
416 if (ret)
417 return ret;
418 }
419
420 /* If we are a PCI bridge, go down */
421 if (dn->child && ((class >> 8) == PCI_CLASS_BRIDGE_PCI ||
422 (class >> 8) == PCI_CLASS_BRIDGE_CARDBUS))
423 /* Depth first...do children */
424 nextdn = dn->child;
425 else if (dn->sibling)
426 /* ok, try next sibling instead. */
427 nextdn = dn->sibling;
428 if (!nextdn) {
429 /* Walk up to next valid sibling. */
430 do {
431 dn = dn->parent;
432 if (dn == start)
433 return NULL;
434 } while (dn->sibling == NULL);
435 nextdn = dn->sibling;
436 }
437 }
438 return NULL;
439 }
440 EXPORT_SYMBOL_GPL(pci_traverse_device_nodes);
441
pci_dn_next_one(struct pci_dn * root,struct pci_dn * pdn)442 static struct pci_dn *pci_dn_next_one(struct pci_dn *root,
443 struct pci_dn *pdn)
444 {
445 struct list_head *next = pdn->child_list.next;
446
447 if (next != &pdn->child_list)
448 return list_entry(next, struct pci_dn, list);
449
450 while (1) {
451 if (pdn == root)
452 return NULL;
453
454 next = pdn->list.next;
455 if (next != &pdn->parent->child_list)
456 break;
457
458 pdn = pdn->parent;
459 }
460
461 return list_entry(next, struct pci_dn, list);
462 }
463
traverse_pci_dn(struct pci_dn * root,void * (* fn)(struct pci_dn *,void *),void * data)464 void *traverse_pci_dn(struct pci_dn *root,
465 void *(*fn)(struct pci_dn *, void *),
466 void *data)
467 {
468 struct pci_dn *pdn = root;
469 void *ret;
470
471 /* Only scan the child nodes */
472 for (pdn = pci_dn_next_one(root, pdn); pdn;
473 pdn = pci_dn_next_one(root, pdn)) {
474 ret = fn(pdn, data);
475 if (ret)
476 return ret;
477 }
478
479 return NULL;
480 }
481
add_pdn(struct device_node * dn,void * data)482 static void *add_pdn(struct device_node *dn, void *data)
483 {
484 struct pci_controller *hose = data;
485 struct pci_dn *pdn;
486
487 pdn = pci_add_device_node_info(hose, dn);
488 if (!pdn)
489 return ERR_PTR(-ENOMEM);
490
491 return NULL;
492 }
493
494 /**
495 * pci_devs_phb_init_dynamic - setup pci devices under this PHB
496 * phb: pci-to-host bridge (top-level bridge connecting to cpu)
497 *
498 * This routine is called both during boot, (before the memory
499 * subsystem is set up, before kmalloc is valid) and during the
500 * dynamic lpar operation of adding a PHB to a running system.
501 */
pci_devs_phb_init_dynamic(struct pci_controller * phb)502 void pci_devs_phb_init_dynamic(struct pci_controller *phb)
503 {
504 struct device_node *dn = phb->dn;
505 struct pci_dn *pdn;
506
507 /* PHB nodes themselves must not match */
508 pdn = pci_add_device_node_info(phb, dn);
509 if (pdn) {
510 pdn->devfn = pdn->busno = -1;
511 pdn->vendor_id = pdn->device_id = pdn->class_code = 0;
512 pdn->phb = phb;
513 phb->pci_data = pdn;
514 }
515
516 /* Update dn->phb ptrs for new phb and children devices */
517 pci_traverse_device_nodes(dn, add_pdn, phb);
518 }
519
520 /**
521 * pci_devs_phb_init - Initialize phbs and pci devs under them.
522 *
523 * This routine walks over all phb's (pci-host bridges) on the
524 * system, and sets up assorted pci-related structures
525 * (including pci info in the device node structs) for each
526 * pci device found underneath. This routine runs once,
527 * early in the boot sequence.
528 */
pci_devs_phb_init(void)529 static int __init pci_devs_phb_init(void)
530 {
531 struct pci_controller *phb, *tmp;
532
533 /* This must be done first so the device nodes have valid pci info! */
534 list_for_each_entry_safe(phb, tmp, &hose_list, list_node)
535 pci_devs_phb_init_dynamic(phb);
536
537 return 0;
538 }
539
540 core_initcall(pci_devs_phb_init);
541
pci_dev_pdn_setup(struct pci_dev * pdev)542 static void pci_dev_pdn_setup(struct pci_dev *pdev)
543 {
544 struct pci_dn *pdn;
545
546 if (pdev->dev.archdata.pci_data)
547 return;
548
549 /* Setup the fast path */
550 pdn = pci_get_pdn(pdev);
551 pdev->dev.archdata.pci_data = pdn;
552 }
553 DECLARE_PCI_FIXUP_EARLY(PCI_ANY_ID, PCI_ANY_ID, pci_dev_pdn_setup);
554