xref: /openbmc/linux/arch/powerpc/kernel/eeh_pe.c (revision 3e30a927)
1 // SPDX-License-Identifier: GPL-2.0-or-later
2 /*
3  * The file intends to implement PE based on the information from
4  * platforms. Basically, there have 3 types of PEs: PHB/Bus/Device.
5  * All the PEs should be organized as hierarchy tree. The first level
6  * of the tree will be associated to existing PHBs since the particular
7  * PE is only meaningful in one PHB domain.
8  *
9  * Copyright Benjamin Herrenschmidt & Gavin Shan, IBM Corporation 2012.
10  */
11 
12 #include <linux/delay.h>
13 #include <linux/export.h>
14 #include <linux/gfp.h>
15 #include <linux/kernel.h>
16 #include <linux/pci.h>
17 #include <linux/string.h>
18 
19 #include <asm/pci-bridge.h>
20 #include <asm/ppc-pci.h>
21 
22 static int eeh_pe_aux_size = 0;
23 static LIST_HEAD(eeh_phb_pe);
24 
25 /**
26  * eeh_set_pe_aux_size - Set PE auxillary data size
27  * @size: PE auxillary data size
28  *
29  * Set PE auxillary data size
30  */
31 void eeh_set_pe_aux_size(int size)
32 {
33 	if (size < 0)
34 		return;
35 
36 	eeh_pe_aux_size = size;
37 }
38 
39 /**
40  * eeh_pe_alloc - Allocate PE
41  * @phb: PCI controller
42  * @type: PE type
43  *
44  * Allocate PE instance dynamically.
45  */
46 static struct eeh_pe *eeh_pe_alloc(struct pci_controller *phb, int type)
47 {
48 	struct eeh_pe *pe;
49 	size_t alloc_size;
50 
51 	alloc_size = sizeof(struct eeh_pe);
52 	if (eeh_pe_aux_size) {
53 		alloc_size = ALIGN(alloc_size, cache_line_size());
54 		alloc_size += eeh_pe_aux_size;
55 	}
56 
57 	/* Allocate PHB PE */
58 	pe = kzalloc(alloc_size, GFP_KERNEL);
59 	if (!pe) return NULL;
60 
61 	/* Initialize PHB PE */
62 	pe->type = type;
63 	pe->phb = phb;
64 	INIT_LIST_HEAD(&pe->child_list);
65 	INIT_LIST_HEAD(&pe->edevs);
66 
67 	pe->data = (void *)pe + ALIGN(sizeof(struct eeh_pe),
68 				      cache_line_size());
69 	return pe;
70 }
71 
72 /**
73  * eeh_phb_pe_create - Create PHB PE
74  * @phb: PCI controller
75  *
76  * The function should be called while the PHB is detected during
77  * system boot or PCI hotplug in order to create PHB PE.
78  */
79 int eeh_phb_pe_create(struct pci_controller *phb)
80 {
81 	struct eeh_pe *pe;
82 
83 	/* Allocate PHB PE */
84 	pe = eeh_pe_alloc(phb, EEH_PE_PHB);
85 	if (!pe) {
86 		pr_err("%s: out of memory!\n", __func__);
87 		return -ENOMEM;
88 	}
89 
90 	/* Put it into the list */
91 	list_add_tail(&pe->child, &eeh_phb_pe);
92 
93 	pr_debug("EEH: Add PE for PHB#%x\n", phb->global_number);
94 
95 	return 0;
96 }
97 
98 /**
99  * eeh_wait_state - Wait for PE state
100  * @pe: EEH PE
101  * @max_wait: maximal period in millisecond
102  *
103  * Wait for the state of associated PE. It might take some time
104  * to retrieve the PE's state.
105  */
106 int eeh_wait_state(struct eeh_pe *pe, int max_wait)
107 {
108 	int ret;
109 	int mwait;
110 
111 	/*
112 	 * According to PAPR, the state of PE might be temporarily
113 	 * unavailable. Under the circumstance, we have to wait
114 	 * for indicated time determined by firmware. The maximal
115 	 * wait time is 5 minutes, which is acquired from the original
116 	 * EEH implementation. Also, the original implementation
117 	 * also defined the minimal wait time as 1 second.
118 	 */
119 #define EEH_STATE_MIN_WAIT_TIME	(1000)
120 #define EEH_STATE_MAX_WAIT_TIME	(300 * 1000)
121 
122 	while (1) {
123 		ret = eeh_ops->get_state(pe, &mwait);
124 
125 		if (ret != EEH_STATE_UNAVAILABLE)
126 			return ret;
127 
128 		if (max_wait <= 0) {
129 			pr_warn("%s: Timeout when getting PE's state (%d)\n",
130 				__func__, max_wait);
131 			return EEH_STATE_NOT_SUPPORT;
132 		}
133 
134 		if (mwait < EEH_STATE_MIN_WAIT_TIME) {
135 			pr_warn("%s: Firmware returned bad wait value %d\n",
136 				__func__, mwait);
137 			mwait = EEH_STATE_MIN_WAIT_TIME;
138 		} else if (mwait > EEH_STATE_MAX_WAIT_TIME) {
139 			pr_warn("%s: Firmware returned too long wait value %d\n",
140 				__func__, mwait);
141 			mwait = EEH_STATE_MAX_WAIT_TIME;
142 		}
143 
144 		msleep(min(mwait, max_wait));
145 		max_wait -= mwait;
146 	}
147 }
148 
149 /**
150  * eeh_phb_pe_get - Retrieve PHB PE based on the given PHB
151  * @phb: PCI controller
152  *
153  * The overall PEs form hierarchy tree. The first layer of the
154  * hierarchy tree is composed of PHB PEs. The function is used
155  * to retrieve the corresponding PHB PE according to the given PHB.
156  */
157 struct eeh_pe *eeh_phb_pe_get(struct pci_controller *phb)
158 {
159 	struct eeh_pe *pe;
160 
161 	list_for_each_entry(pe, &eeh_phb_pe, child) {
162 		/*
163 		 * Actually, we needn't check the type since
164 		 * the PE for PHB has been determined when that
165 		 * was created.
166 		 */
167 		if ((pe->type & EEH_PE_PHB) && pe->phb == phb)
168 			return pe;
169 	}
170 
171 	return NULL;
172 }
173 
174 /**
175  * eeh_pe_next - Retrieve the next PE in the tree
176  * @pe: current PE
177  * @root: root PE
178  *
179  * The function is used to retrieve the next PE in the
180  * hierarchy PE tree.
181  */
182 struct eeh_pe *eeh_pe_next(struct eeh_pe *pe, struct eeh_pe *root)
183 {
184 	struct list_head *next = pe->child_list.next;
185 
186 	if (next == &pe->child_list) {
187 		while (1) {
188 			if (pe == root)
189 				return NULL;
190 			next = pe->child.next;
191 			if (next != &pe->parent->child_list)
192 				break;
193 			pe = pe->parent;
194 		}
195 	}
196 
197 	return list_entry(next, struct eeh_pe, child);
198 }
199 
200 /**
201  * eeh_pe_traverse - Traverse PEs in the specified PHB
202  * @root: root PE
203  * @fn: callback
204  * @flag: extra parameter to callback
205  *
206  * The function is used to traverse the specified PE and its
207  * child PEs. The traversing is to be terminated once the
208  * callback returns something other than NULL, or no more PEs
209  * to be traversed.
210  */
211 void *eeh_pe_traverse(struct eeh_pe *root,
212 		      eeh_pe_traverse_func fn, void *flag)
213 {
214 	struct eeh_pe *pe;
215 	void *ret;
216 
217 	eeh_for_each_pe(root, pe) {
218 		ret = fn(pe, flag);
219 		if (ret) return ret;
220 	}
221 
222 	return NULL;
223 }
224 
225 /**
226  * eeh_pe_dev_traverse - Traverse the devices from the PE
227  * @root: EEH PE
228  * @fn: function callback
229  * @flag: extra parameter to callback
230  *
231  * The function is used to traverse the devices of the specified
232  * PE and its child PEs.
233  */
234 void eeh_pe_dev_traverse(struct eeh_pe *root,
235 			  eeh_edev_traverse_func fn, void *flag)
236 {
237 	struct eeh_pe *pe;
238 	struct eeh_dev *edev, *tmp;
239 
240 	if (!root) {
241 		pr_warn("%s: Invalid PE %p\n",
242 			__func__, root);
243 		return;
244 	}
245 
246 	/* Traverse root PE */
247 	eeh_for_each_pe(root, pe)
248 		eeh_pe_for_each_dev(pe, edev, tmp)
249 			fn(edev, flag);
250 }
251 
252 /**
253  * __eeh_pe_get - Check the PE address
254  * @data: EEH PE
255  * @flag: EEH device
256  *
257  * For one particular PE, it can be identified by PE address
258  * or tranditional BDF address. BDF address is composed of
259  * Bus/Device/Function number. The extra data referred by flag
260  * indicates which type of address should be used.
261  */
262 struct eeh_pe_get_flag {
263 	int pe_no;
264 	int config_addr;
265 };
266 
267 static void *__eeh_pe_get(struct eeh_pe *pe, void *flag)
268 {
269 	struct eeh_pe_get_flag *tmp = (struct eeh_pe_get_flag *) flag;
270 
271 	/* Unexpected PHB PE */
272 	if (pe->type & EEH_PE_PHB)
273 		return NULL;
274 
275 	/*
276 	 * We prefer PE address. For most cases, we should
277 	 * have non-zero PE address
278 	 */
279 	if (eeh_has_flag(EEH_VALID_PE_ZERO)) {
280 		if (tmp->pe_no == pe->addr)
281 			return pe;
282 	} else {
283 		if (tmp->pe_no &&
284 		    (tmp->pe_no == pe->addr))
285 			return pe;
286 	}
287 
288 	/* Try BDF address */
289 	if (tmp->config_addr &&
290 	   (tmp->config_addr == pe->config_addr))
291 		return pe;
292 
293 	return NULL;
294 }
295 
296 /**
297  * eeh_pe_get - Search PE based on the given address
298  * @phb: PCI controller
299  * @pe_no: PE number
300  * @config_addr: Config address
301  *
302  * Search the corresponding PE based on the specified address which
303  * is included in the eeh device. The function is used to check if
304  * the associated PE has been created against the PE address. It's
305  * notable that the PE address has 2 format: traditional PE address
306  * which is composed of PCI bus/device/function number, or unified
307  * PE address.
308  */
309 struct eeh_pe *eeh_pe_get(struct pci_controller *phb,
310 		int pe_no, int config_addr)
311 {
312 	struct eeh_pe *root = eeh_phb_pe_get(phb);
313 	struct eeh_pe_get_flag tmp = { pe_no, config_addr };
314 	struct eeh_pe *pe;
315 
316 	pe = eeh_pe_traverse(root, __eeh_pe_get, &tmp);
317 
318 	return pe;
319 }
320 
321 /**
322  * eeh_pe_get_parent - Retrieve the parent PE
323  * @edev: EEH device
324  *
325  * The whole PEs existing in the system are organized as hierarchy
326  * tree. The function is used to retrieve the parent PE according
327  * to the parent EEH device.
328  */
329 static struct eeh_pe *eeh_pe_get_parent(struct eeh_dev *edev)
330 {
331 	struct eeh_dev *parent;
332 	struct pci_dn *pdn = eeh_dev_to_pdn(edev);
333 
334 	/*
335 	 * It might have the case for the indirect parent
336 	 * EEH device already having associated PE, but
337 	 * the direct parent EEH device doesn't have yet.
338 	 */
339 	if (edev->physfn)
340 		pdn = pci_get_pdn(edev->physfn);
341 	else
342 		pdn = pdn ? pdn->parent : NULL;
343 	while (pdn) {
344 		/* We're poking out of PCI territory */
345 		parent = pdn_to_eeh_dev(pdn);
346 		if (!parent)
347 			return NULL;
348 
349 		if (parent->pe)
350 			return parent->pe;
351 
352 		pdn = pdn->parent;
353 	}
354 
355 	return NULL;
356 }
357 
358 /**
359  * eeh_add_to_parent_pe - Add EEH device to parent PE
360  * @edev: EEH device
361  *
362  * Add EEH device to the parent PE. If the parent PE already
363  * exists, the PE type will be changed to EEH_PE_BUS. Otherwise,
364  * we have to create new PE to hold the EEH device and the new
365  * PE will be linked to its parent PE as well.
366  */
367 int eeh_add_to_parent_pe(struct eeh_dev *edev)
368 {
369 	struct eeh_pe *pe, *parent;
370 	struct pci_dn *pdn = eeh_dev_to_pdn(edev);
371 	int config_addr = (pdn->busno << 8) | (pdn->devfn);
372 
373 	/* Check if the PE number is valid */
374 	if (!eeh_has_flag(EEH_VALID_PE_ZERO) && !edev->pe_config_addr) {
375 		eeh_edev_err(edev, "PE#0 is invalid for this PHB!\n");
376 		return -EINVAL;
377 	}
378 
379 	/*
380 	 * Search the PE has been existing or not according
381 	 * to the PE address. If that has been existing, the
382 	 * PE should be composed of PCI bus and its subordinate
383 	 * components.
384 	 */
385 	pe = eeh_pe_get(pdn->phb, edev->pe_config_addr, config_addr);
386 	if (pe) {
387 		if (pe->type & EEH_PE_INVALID) {
388 			list_add_tail(&edev->entry, &pe->edevs);
389 			edev->pe = pe;
390 			/*
391 			 * We're running to here because of PCI hotplug caused by
392 			 * EEH recovery. We need clear EEH_PE_INVALID until the top.
393 			 */
394 			parent = pe;
395 			while (parent) {
396 				if (!(parent->type & EEH_PE_INVALID))
397 					break;
398 				parent->type &= ~EEH_PE_INVALID;
399 				parent = parent->parent;
400 			}
401 
402 			eeh_edev_dbg(edev,
403 				     "Added to device PE (parent: PE#%x)\n",
404 				     pe->parent->addr);
405 		} else {
406 			/* Mark the PE as type of PCI bus */
407 			pe->type = EEH_PE_BUS;
408 			edev->pe = pe;
409 
410 			/* Put the edev to PE */
411 			list_add_tail(&edev->entry, &pe->edevs);
412 			eeh_edev_dbg(edev, "Added to bus PE\n");
413 		}
414 		return 0;
415 	}
416 
417 	/* Create a new EEH PE */
418 	if (edev->physfn)
419 		pe = eeh_pe_alloc(pdn->phb, EEH_PE_VF);
420 	else
421 		pe = eeh_pe_alloc(pdn->phb, EEH_PE_DEVICE);
422 	if (!pe) {
423 		pr_err("%s: out of memory!\n", __func__);
424 		return -ENOMEM;
425 	}
426 	pe->addr	= edev->pe_config_addr;
427 	pe->config_addr	= config_addr;
428 
429 	/*
430 	 * Put the new EEH PE into hierarchy tree. If the parent
431 	 * can't be found, the newly created PE will be attached
432 	 * to PHB directly. Otherwise, we have to associate the
433 	 * PE with its parent.
434 	 */
435 	parent = eeh_pe_get_parent(edev);
436 	if (!parent) {
437 		parent = eeh_phb_pe_get(pdn->phb);
438 		if (!parent) {
439 			pr_err("%s: No PHB PE is found (PHB Domain=%d)\n",
440 				__func__, pdn->phb->global_number);
441 			edev->pe = NULL;
442 			kfree(pe);
443 			return -EEXIST;
444 		}
445 	}
446 	pe->parent = parent;
447 
448 	/*
449 	 * Put the newly created PE into the child list and
450 	 * link the EEH device accordingly.
451 	 */
452 	list_add_tail(&pe->child, &parent->child_list);
453 	list_add_tail(&edev->entry, &pe->edevs);
454 	edev->pe = pe;
455 	eeh_edev_dbg(edev, "Added to device PE (parent: PE#%x)\n",
456 		     pe->parent->addr);
457 
458 	return 0;
459 }
460 
461 /**
462  * eeh_rmv_from_parent_pe - Remove one EEH device from the associated PE
463  * @edev: EEH device
464  *
465  * The PE hierarchy tree might be changed when doing PCI hotplug.
466  * Also, the PCI devices or buses could be removed from the system
467  * during EEH recovery. So we have to call the function remove the
468  * corresponding PE accordingly if necessary.
469  */
470 int eeh_rmv_from_parent_pe(struct eeh_dev *edev)
471 {
472 	struct eeh_pe *pe, *parent, *child;
473 	bool keep, recover;
474 	int cnt;
475 
476 	pe = eeh_dev_to_pe(edev);
477 	if (!pe) {
478 		eeh_edev_dbg(edev, "No PE found for device.\n");
479 		return -EEXIST;
480 	}
481 
482 	/* Remove the EEH device */
483 	edev->pe = NULL;
484 	list_del(&edev->entry);
485 
486 	/*
487 	 * Check if the parent PE includes any EEH devices.
488 	 * If not, we should delete that. Also, we should
489 	 * delete the parent PE if it doesn't have associated
490 	 * child PEs and EEH devices.
491 	 */
492 	while (1) {
493 		parent = pe->parent;
494 
495 		/* PHB PEs should never be removed */
496 		if (pe->type & EEH_PE_PHB)
497 			break;
498 
499 		/*
500 		 * XXX: KEEP is set while resetting a PE. I don't think it's
501 		 * ever set without RECOVERING also being set. I could
502 		 * be wrong though so catch that with a WARN.
503 		 */
504 		keep = !!(pe->state & EEH_PE_KEEP);
505 		recover = !!(pe->state & EEH_PE_RECOVERING);
506 		WARN_ON(keep && !recover);
507 
508 		if (!keep && !recover) {
509 			if (list_empty(&pe->edevs) &&
510 			    list_empty(&pe->child_list)) {
511 				list_del(&pe->child);
512 				kfree(pe);
513 			} else {
514 				break;
515 			}
516 		} else {
517 			/*
518 			 * Mark the PE as invalid. At the end of the recovery
519 			 * process any invalid PEs will be garbage collected.
520 			 *
521 			 * We need to delay the free()ing of them since we can
522 			 * remove edev's while traversing the PE tree which
523 			 * might trigger the removal of a PE and we can't
524 			 * deal with that (yet).
525 			 */
526 			if (list_empty(&pe->edevs)) {
527 				cnt = 0;
528 				list_for_each_entry(child, &pe->child_list, child) {
529 					if (!(child->type & EEH_PE_INVALID)) {
530 						cnt++;
531 						break;
532 					}
533 				}
534 
535 				if (!cnt)
536 					pe->type |= EEH_PE_INVALID;
537 				else
538 					break;
539 			}
540 		}
541 
542 		pe = parent;
543 	}
544 
545 	return 0;
546 }
547 
548 /**
549  * eeh_pe_update_time_stamp - Update PE's frozen time stamp
550  * @pe: EEH PE
551  *
552  * We have time stamp for each PE to trace its time of getting
553  * frozen in last hour. The function should be called to update
554  * the time stamp on first error of the specific PE. On the other
555  * handle, we needn't account for errors happened in last hour.
556  */
557 void eeh_pe_update_time_stamp(struct eeh_pe *pe)
558 {
559 	time64_t tstamp;
560 
561 	if (!pe) return;
562 
563 	if (pe->freeze_count <= 0) {
564 		pe->freeze_count = 0;
565 		pe->tstamp = ktime_get_seconds();
566 	} else {
567 		tstamp = ktime_get_seconds();
568 		if (tstamp - pe->tstamp > 3600) {
569 			pe->tstamp = tstamp;
570 			pe->freeze_count = 0;
571 		}
572 	}
573 }
574 
575 /**
576  * eeh_pe_state_mark - Mark specified state for PE and its associated device
577  * @pe: EEH PE
578  *
579  * EEH error affects the current PE and its child PEs. The function
580  * is used to mark appropriate state for the affected PEs and the
581  * associated devices.
582  */
583 void eeh_pe_state_mark(struct eeh_pe *root, int state)
584 {
585 	struct eeh_pe *pe;
586 
587 	eeh_for_each_pe(root, pe)
588 		if (!(pe->state & EEH_PE_REMOVED))
589 			pe->state |= state;
590 }
591 EXPORT_SYMBOL_GPL(eeh_pe_state_mark);
592 
593 /**
594  * eeh_pe_mark_isolated
595  * @pe: EEH PE
596  *
597  * Record that a PE has been isolated by marking the PE and it's children as
598  * EEH_PE_ISOLATED (and EEH_PE_CFG_BLOCKED, if required) and their PCI devices
599  * as pci_channel_io_frozen.
600  */
601 void eeh_pe_mark_isolated(struct eeh_pe *root)
602 {
603 	struct eeh_pe *pe;
604 	struct eeh_dev *edev;
605 	struct pci_dev *pdev;
606 
607 	eeh_pe_state_mark(root, EEH_PE_ISOLATED);
608 	eeh_for_each_pe(root, pe) {
609 		list_for_each_entry(edev, &pe->edevs, entry) {
610 			pdev = eeh_dev_to_pci_dev(edev);
611 			if (pdev)
612 				pdev->error_state = pci_channel_io_frozen;
613 		}
614 		/* Block PCI config access if required */
615 		if (pe->state & EEH_PE_CFG_RESTRICTED)
616 			pe->state |= EEH_PE_CFG_BLOCKED;
617 	}
618 }
619 EXPORT_SYMBOL_GPL(eeh_pe_mark_isolated);
620 
621 static void __eeh_pe_dev_mode_mark(struct eeh_dev *edev, void *flag)
622 {
623 	int mode = *((int *)flag);
624 
625 	edev->mode |= mode;
626 }
627 
628 /**
629  * eeh_pe_dev_state_mark - Mark state for all device under the PE
630  * @pe: EEH PE
631  *
632  * Mark specific state for all child devices of the PE.
633  */
634 void eeh_pe_dev_mode_mark(struct eeh_pe *pe, int mode)
635 {
636 	eeh_pe_dev_traverse(pe, __eeh_pe_dev_mode_mark, &mode);
637 }
638 
639 /**
640  * eeh_pe_state_clear - Clear state for the PE
641  * @data: EEH PE
642  * @state: state
643  * @include_passed: include passed-through devices?
644  *
645  * The function is used to clear the indicated state from the
646  * given PE. Besides, we also clear the check count of the PE
647  * as well.
648  */
649 void eeh_pe_state_clear(struct eeh_pe *root, int state, bool include_passed)
650 {
651 	struct eeh_pe *pe;
652 	struct eeh_dev *edev, *tmp;
653 	struct pci_dev *pdev;
654 
655 	eeh_for_each_pe(root, pe) {
656 		/* Keep the state of permanently removed PE intact */
657 		if (pe->state & EEH_PE_REMOVED)
658 			continue;
659 
660 		if (!include_passed && eeh_pe_passed(pe))
661 			continue;
662 
663 		pe->state &= ~state;
664 
665 		/*
666 		 * Special treatment on clearing isolated state. Clear
667 		 * check count since last isolation and put all affected
668 		 * devices to normal state.
669 		 */
670 		if (!(state & EEH_PE_ISOLATED))
671 			continue;
672 
673 		pe->check_count = 0;
674 		eeh_pe_for_each_dev(pe, edev, tmp) {
675 			pdev = eeh_dev_to_pci_dev(edev);
676 			if (!pdev)
677 				continue;
678 
679 			pdev->error_state = pci_channel_io_normal;
680 		}
681 
682 		/* Unblock PCI config access if required */
683 		if (pe->state & EEH_PE_CFG_RESTRICTED)
684 			pe->state &= ~EEH_PE_CFG_BLOCKED;
685 	}
686 }
687 
688 /*
689  * Some PCI bridges (e.g. PLX bridges) have primary/secondary
690  * buses assigned explicitly by firmware, and we probably have
691  * lost that after reset. So we have to delay the check until
692  * the PCI-CFG registers have been restored for the parent
693  * bridge.
694  *
695  * Don't use normal PCI-CFG accessors, which probably has been
696  * blocked on normal path during the stage. So we need utilize
697  * eeh operations, which is always permitted.
698  */
699 static void eeh_bridge_check_link(struct eeh_dev *edev)
700 {
701 	struct pci_dn *pdn = eeh_dev_to_pdn(edev);
702 	int cap;
703 	uint32_t val;
704 	int timeout = 0;
705 
706 	/*
707 	 * We only check root port and downstream ports of
708 	 * PCIe switches
709 	 */
710 	if (!(edev->mode & (EEH_DEV_ROOT_PORT | EEH_DEV_DS_PORT)))
711 		return;
712 
713 	eeh_edev_dbg(edev, "Checking PCIe link...\n");
714 
715 	/* Check slot status */
716 	cap = edev->pcie_cap;
717 	eeh_ops->read_config(pdn, cap + PCI_EXP_SLTSTA, 2, &val);
718 	if (!(val & PCI_EXP_SLTSTA_PDS)) {
719 		eeh_edev_dbg(edev, "No card in the slot (0x%04x) !\n", val);
720 		return;
721 	}
722 
723 	/* Check power status if we have the capability */
724 	eeh_ops->read_config(pdn, cap + PCI_EXP_SLTCAP, 2, &val);
725 	if (val & PCI_EXP_SLTCAP_PCP) {
726 		eeh_ops->read_config(pdn, cap + PCI_EXP_SLTCTL, 2, &val);
727 		if (val & PCI_EXP_SLTCTL_PCC) {
728 			eeh_edev_dbg(edev, "In power-off state, power it on ...\n");
729 			val &= ~(PCI_EXP_SLTCTL_PCC | PCI_EXP_SLTCTL_PIC);
730 			val |= (0x0100 & PCI_EXP_SLTCTL_PIC);
731 			eeh_ops->write_config(pdn, cap + PCI_EXP_SLTCTL, 2, val);
732 			msleep(2 * 1000);
733 		}
734 	}
735 
736 	/* Enable link */
737 	eeh_ops->read_config(pdn, cap + PCI_EXP_LNKCTL, 2, &val);
738 	val &= ~PCI_EXP_LNKCTL_LD;
739 	eeh_ops->write_config(pdn, cap + PCI_EXP_LNKCTL, 2, val);
740 
741 	/* Check link */
742 	eeh_ops->read_config(pdn, cap + PCI_EXP_LNKCAP, 4, &val);
743 	if (!(val & PCI_EXP_LNKCAP_DLLLARC)) {
744 		eeh_edev_dbg(edev, "No link reporting capability (0x%08x) \n", val);
745 		msleep(1000);
746 		return;
747 	}
748 
749 	/* Wait the link is up until timeout (5s) */
750 	timeout = 0;
751 	while (timeout < 5000) {
752 		msleep(20);
753 		timeout += 20;
754 
755 		eeh_ops->read_config(pdn, cap + PCI_EXP_LNKSTA, 2, &val);
756 		if (val & PCI_EXP_LNKSTA_DLLLA)
757 			break;
758 	}
759 
760 	if (val & PCI_EXP_LNKSTA_DLLLA)
761 		eeh_edev_dbg(edev, "Link up (%s)\n",
762 			 (val & PCI_EXP_LNKSTA_CLS_2_5GB) ? "2.5GB" : "5GB");
763 	else
764 		eeh_edev_dbg(edev, "Link not ready (0x%04x)\n", val);
765 }
766 
767 #define BYTE_SWAP(OFF)	(8*((OFF)/4)+3-(OFF))
768 #define SAVED_BYTE(OFF)	(((u8 *)(edev->config_space))[BYTE_SWAP(OFF)])
769 
770 static void eeh_restore_bridge_bars(struct eeh_dev *edev)
771 {
772 	struct pci_dn *pdn = eeh_dev_to_pdn(edev);
773 	int i;
774 
775 	/*
776 	 * Device BARs: 0x10 - 0x18
777 	 * Bus numbers and windows: 0x18 - 0x30
778 	 */
779 	for (i = 4; i < 13; i++)
780 		eeh_ops->write_config(pdn, i*4, 4, edev->config_space[i]);
781 	/* Rom: 0x38 */
782 	eeh_ops->write_config(pdn, 14*4, 4, edev->config_space[14]);
783 
784 	/* Cache line & Latency timer: 0xC 0xD */
785 	eeh_ops->write_config(pdn, PCI_CACHE_LINE_SIZE, 1,
786                 SAVED_BYTE(PCI_CACHE_LINE_SIZE));
787         eeh_ops->write_config(pdn, PCI_LATENCY_TIMER, 1,
788                 SAVED_BYTE(PCI_LATENCY_TIMER));
789 	/* Max latency, min grant, interrupt ping and line: 0x3C */
790 	eeh_ops->write_config(pdn, 15*4, 4, edev->config_space[15]);
791 
792 	/* PCI Command: 0x4 */
793 	eeh_ops->write_config(pdn, PCI_COMMAND, 4, edev->config_space[1] |
794 			      PCI_COMMAND_MEMORY | PCI_COMMAND_MASTER);
795 
796 	/* Check the PCIe link is ready */
797 	eeh_bridge_check_link(edev);
798 }
799 
800 static void eeh_restore_device_bars(struct eeh_dev *edev)
801 {
802 	struct pci_dn *pdn = eeh_dev_to_pdn(edev);
803 	int i;
804 	u32 cmd;
805 
806 	for (i = 4; i < 10; i++)
807 		eeh_ops->write_config(pdn, i*4, 4, edev->config_space[i]);
808 	/* 12 == Expansion ROM Address */
809 	eeh_ops->write_config(pdn, 12*4, 4, edev->config_space[12]);
810 
811 	eeh_ops->write_config(pdn, PCI_CACHE_LINE_SIZE, 1,
812 		SAVED_BYTE(PCI_CACHE_LINE_SIZE));
813 	eeh_ops->write_config(pdn, PCI_LATENCY_TIMER, 1,
814 		SAVED_BYTE(PCI_LATENCY_TIMER));
815 
816 	/* max latency, min grant, interrupt pin and line */
817 	eeh_ops->write_config(pdn, 15*4, 4, edev->config_space[15]);
818 
819 	/*
820 	 * Restore PERR & SERR bits, some devices require it,
821 	 * don't touch the other command bits
822 	 */
823 	eeh_ops->read_config(pdn, PCI_COMMAND, 4, &cmd);
824 	if (edev->config_space[1] & PCI_COMMAND_PARITY)
825 		cmd |= PCI_COMMAND_PARITY;
826 	else
827 		cmd &= ~PCI_COMMAND_PARITY;
828 	if (edev->config_space[1] & PCI_COMMAND_SERR)
829 		cmd |= PCI_COMMAND_SERR;
830 	else
831 		cmd &= ~PCI_COMMAND_SERR;
832 	eeh_ops->write_config(pdn, PCI_COMMAND, 4, cmd);
833 }
834 
835 /**
836  * eeh_restore_one_device_bars - Restore the Base Address Registers for one device
837  * @data: EEH device
838  * @flag: Unused
839  *
840  * Loads the PCI configuration space base address registers,
841  * the expansion ROM base address, the latency timer, and etc.
842  * from the saved values in the device node.
843  */
844 static void eeh_restore_one_device_bars(struct eeh_dev *edev, void *flag)
845 {
846 	struct pci_dn *pdn = eeh_dev_to_pdn(edev);
847 
848 	/* Do special restore for bridges */
849 	if (edev->mode & EEH_DEV_BRIDGE)
850 		eeh_restore_bridge_bars(edev);
851 	else
852 		eeh_restore_device_bars(edev);
853 
854 	if (eeh_ops->restore_config && pdn)
855 		eeh_ops->restore_config(pdn);
856 }
857 
858 /**
859  * eeh_pe_restore_bars - Restore the PCI config space info
860  * @pe: EEH PE
861  *
862  * This routine performs a recursive walk to the children
863  * of this device as well.
864  */
865 void eeh_pe_restore_bars(struct eeh_pe *pe)
866 {
867 	/*
868 	 * We needn't take the EEH lock since eeh_pe_dev_traverse()
869 	 * will take that.
870 	 */
871 	eeh_pe_dev_traverse(pe, eeh_restore_one_device_bars, NULL);
872 }
873 
874 /**
875  * eeh_pe_loc_get - Retrieve location code binding to the given PE
876  * @pe: EEH PE
877  *
878  * Retrieve the location code of the given PE. If the primary PE bus
879  * is root bus, we will grab location code from PHB device tree node
880  * or root port. Otherwise, the upstream bridge's device tree node
881  * of the primary PE bus will be checked for the location code.
882  */
883 const char *eeh_pe_loc_get(struct eeh_pe *pe)
884 {
885 	struct pci_bus *bus = eeh_pe_bus_get(pe);
886 	struct device_node *dn;
887 	const char *loc = NULL;
888 
889 	while (bus) {
890 		dn = pci_bus_to_OF_node(bus);
891 		if (!dn) {
892 			bus = bus->parent;
893 			continue;
894 		}
895 
896 		if (pci_is_root_bus(bus))
897 			loc = of_get_property(dn, "ibm,io-base-loc-code", NULL);
898 		else
899 			loc = of_get_property(dn, "ibm,slot-location-code",
900 					      NULL);
901 
902 		if (loc)
903 			return loc;
904 
905 		bus = bus->parent;
906 	}
907 
908 	return "N/A";
909 }
910 
911 /**
912  * eeh_pe_bus_get - Retrieve PCI bus according to the given PE
913  * @pe: EEH PE
914  *
915  * Retrieve the PCI bus according to the given PE. Basically,
916  * there're 3 types of PEs: PHB/Bus/Device. For PHB PE, the
917  * primary PCI bus will be retrieved. The parent bus will be
918  * returned for BUS PE. However, we don't have associated PCI
919  * bus for DEVICE PE.
920  */
921 struct pci_bus *eeh_pe_bus_get(struct eeh_pe *pe)
922 {
923 	struct eeh_dev *edev;
924 	struct pci_dev *pdev;
925 
926 	if (pe->type & EEH_PE_PHB)
927 		return pe->phb->bus;
928 
929 	/* The primary bus might be cached during probe time */
930 	if (pe->state & EEH_PE_PRI_BUS)
931 		return pe->bus;
932 
933 	/* Retrieve the parent PCI bus of first (top) PCI device */
934 	edev = list_first_entry_or_null(&pe->edevs, struct eeh_dev, entry);
935 	pdev = eeh_dev_to_pci_dev(edev);
936 	if (pdev)
937 		return pdev->bus;
938 
939 	return NULL;
940 }
941