xref: /openbmc/linux/arch/powerpc/kernel/eeh_pe.c (revision b830f94f)
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 	void *ret;
240 
241 	if (!root) {
242 		pr_warn("%s: Invalid PE %p\n",
243 			__func__, root);
244 		return NULL;
245 	}
246 
247 	/* Traverse root PE */
248 	eeh_for_each_pe(root, pe) {
249 		eeh_pe_for_each_dev(pe, edev, tmp) {
250 			ret = fn(edev, flag);
251 			if (ret)
252 				return ret;
253 		}
254 	}
255 
256 	return NULL;
257 }
258 
259 /**
260  * __eeh_pe_get - Check the PE address
261  * @data: EEH PE
262  * @flag: EEH device
263  *
264  * For one particular PE, it can be identified by PE address
265  * or tranditional BDF address. BDF address is composed of
266  * Bus/Device/Function number. The extra data referred by flag
267  * indicates which type of address should be used.
268  */
269 struct eeh_pe_get_flag {
270 	int pe_no;
271 	int config_addr;
272 };
273 
274 static void *__eeh_pe_get(struct eeh_pe *pe, void *flag)
275 {
276 	struct eeh_pe_get_flag *tmp = (struct eeh_pe_get_flag *) flag;
277 
278 	/* Unexpected PHB PE */
279 	if (pe->type & EEH_PE_PHB)
280 		return NULL;
281 
282 	/*
283 	 * We prefer PE address. For most cases, we should
284 	 * have non-zero PE address
285 	 */
286 	if (eeh_has_flag(EEH_VALID_PE_ZERO)) {
287 		if (tmp->pe_no == pe->addr)
288 			return pe;
289 	} else {
290 		if (tmp->pe_no &&
291 		    (tmp->pe_no == pe->addr))
292 			return pe;
293 	}
294 
295 	/* Try BDF address */
296 	if (tmp->config_addr &&
297 	   (tmp->config_addr == pe->config_addr))
298 		return pe;
299 
300 	return NULL;
301 }
302 
303 /**
304  * eeh_pe_get - Search PE based on the given address
305  * @phb: PCI controller
306  * @pe_no: PE number
307  * @config_addr: Config address
308  *
309  * Search the corresponding PE based on the specified address which
310  * is included in the eeh device. The function is used to check if
311  * the associated PE has been created against the PE address. It's
312  * notable that the PE address has 2 format: traditional PE address
313  * which is composed of PCI bus/device/function number, or unified
314  * PE address.
315  */
316 struct eeh_pe *eeh_pe_get(struct pci_controller *phb,
317 		int pe_no, int config_addr)
318 {
319 	struct eeh_pe *root = eeh_phb_pe_get(phb);
320 	struct eeh_pe_get_flag tmp = { pe_no, config_addr };
321 	struct eeh_pe *pe;
322 
323 	pe = eeh_pe_traverse(root, __eeh_pe_get, &tmp);
324 
325 	return pe;
326 }
327 
328 /**
329  * eeh_pe_get_parent - Retrieve the parent PE
330  * @edev: EEH device
331  *
332  * The whole PEs existing in the system are organized as hierarchy
333  * tree. The function is used to retrieve the parent PE according
334  * to the parent EEH device.
335  */
336 static struct eeh_pe *eeh_pe_get_parent(struct eeh_dev *edev)
337 {
338 	struct eeh_dev *parent;
339 	struct pci_dn *pdn = eeh_dev_to_pdn(edev);
340 
341 	/*
342 	 * It might have the case for the indirect parent
343 	 * EEH device already having associated PE, but
344 	 * the direct parent EEH device doesn't have yet.
345 	 */
346 	if (edev->physfn)
347 		pdn = pci_get_pdn(edev->physfn);
348 	else
349 		pdn = pdn ? pdn->parent : NULL;
350 	while (pdn) {
351 		/* We're poking out of PCI territory */
352 		parent = pdn_to_eeh_dev(pdn);
353 		if (!parent)
354 			return NULL;
355 
356 		if (parent->pe)
357 			return parent->pe;
358 
359 		pdn = pdn->parent;
360 	}
361 
362 	return NULL;
363 }
364 
365 /**
366  * eeh_add_to_parent_pe - Add EEH device to parent PE
367  * @edev: EEH device
368  *
369  * Add EEH device to the parent PE. If the parent PE already
370  * exists, the PE type will be changed to EEH_PE_BUS. Otherwise,
371  * we have to create new PE to hold the EEH device and the new
372  * PE will be linked to its parent PE as well.
373  */
374 int eeh_add_to_parent_pe(struct eeh_dev *edev)
375 {
376 	struct eeh_pe *pe, *parent;
377 	struct pci_dn *pdn = eeh_dev_to_pdn(edev);
378 	int config_addr = (pdn->busno << 8) | (pdn->devfn);
379 
380 	/* Check if the PE number is valid */
381 	if (!eeh_has_flag(EEH_VALID_PE_ZERO) && !edev->pe_config_addr) {
382 		pr_err("%s: Invalid PE#0 for edev 0x%x on PHB#%x\n",
383 		       __func__, config_addr, pdn->phb->global_number);
384 		return -EINVAL;
385 	}
386 
387 	/*
388 	 * Search the PE has been existing or not according
389 	 * to the PE address. If that has been existing, the
390 	 * PE should be composed of PCI bus and its subordinate
391 	 * components.
392 	 */
393 	pe = eeh_pe_get(pdn->phb, edev->pe_config_addr, config_addr);
394 	if (pe && !(pe->type & EEH_PE_INVALID)) {
395 		/* Mark the PE as type of PCI bus */
396 		pe->type = EEH_PE_BUS;
397 		edev->pe = pe;
398 
399 		/* Put the edev to PE */
400 		list_add_tail(&edev->entry, &pe->edevs);
401 		pr_debug("EEH: Add %04x:%02x:%02x.%01x to Bus PE#%x\n",
402 			 pdn->phb->global_number,
403 			 pdn->busno,
404 			 PCI_SLOT(pdn->devfn),
405 			 PCI_FUNC(pdn->devfn),
406 			 pe->addr);
407 		return 0;
408 	} else if (pe && (pe->type & EEH_PE_INVALID)) {
409 		list_add_tail(&edev->entry, &pe->edevs);
410 		edev->pe = pe;
411 		/*
412 		 * We're running to here because of PCI hotplug caused by
413 		 * EEH recovery. We need clear EEH_PE_INVALID until the top.
414 		 */
415 		parent = pe;
416 		while (parent) {
417 			if (!(parent->type & EEH_PE_INVALID))
418 				break;
419 			parent->type &= ~EEH_PE_INVALID;
420 			parent = parent->parent;
421 		}
422 
423 		pr_debug("EEH: Add %04x:%02x:%02x.%01x to Device "
424 			 "PE#%x, Parent PE#%x\n",
425 			 pdn->phb->global_number,
426 			 pdn->busno,
427 			 PCI_SLOT(pdn->devfn),
428 			 PCI_FUNC(pdn->devfn),
429 			 pe->addr, pe->parent->addr);
430 		return 0;
431 	}
432 
433 	/* Create a new EEH PE */
434 	if (edev->physfn)
435 		pe = eeh_pe_alloc(pdn->phb, EEH_PE_VF);
436 	else
437 		pe = eeh_pe_alloc(pdn->phb, EEH_PE_DEVICE);
438 	if (!pe) {
439 		pr_err("%s: out of memory!\n", __func__);
440 		return -ENOMEM;
441 	}
442 	pe->addr	= edev->pe_config_addr;
443 	pe->config_addr	= config_addr;
444 
445 	/*
446 	 * Put the new EEH PE into hierarchy tree. If the parent
447 	 * can't be found, the newly created PE will be attached
448 	 * to PHB directly. Otherwise, we have to associate the
449 	 * PE with its parent.
450 	 */
451 	parent = eeh_pe_get_parent(edev);
452 	if (!parent) {
453 		parent = eeh_phb_pe_get(pdn->phb);
454 		if (!parent) {
455 			pr_err("%s: No PHB PE is found (PHB Domain=%d)\n",
456 				__func__, pdn->phb->global_number);
457 			edev->pe = NULL;
458 			kfree(pe);
459 			return -EEXIST;
460 		}
461 	}
462 	pe->parent = parent;
463 
464 	/*
465 	 * Put the newly created PE into the child list and
466 	 * link the EEH device accordingly.
467 	 */
468 	list_add_tail(&pe->child, &parent->child_list);
469 	list_add_tail(&edev->entry, &pe->edevs);
470 	edev->pe = pe;
471 	pr_debug("EEH: Add %04x:%02x:%02x.%01x to "
472 		 "Device PE#%x, Parent PE#%x\n",
473 		 pdn->phb->global_number,
474 		 pdn->busno,
475 		 PCI_SLOT(pdn->devfn),
476 		 PCI_FUNC(pdn->devfn),
477 		 pe->addr, pe->parent->addr);
478 
479 	return 0;
480 }
481 
482 /**
483  * eeh_rmv_from_parent_pe - Remove one EEH device from the associated PE
484  * @edev: EEH device
485  *
486  * The PE hierarchy tree might be changed when doing PCI hotplug.
487  * Also, the PCI devices or buses could be removed from the system
488  * during EEH recovery. So we have to call the function remove the
489  * corresponding PE accordingly if necessary.
490  */
491 int eeh_rmv_from_parent_pe(struct eeh_dev *edev)
492 {
493 	struct eeh_pe *pe, *parent, *child;
494 	int cnt;
495 	struct pci_dn *pdn = eeh_dev_to_pdn(edev);
496 
497 	pe = eeh_dev_to_pe(edev);
498 	if (!pe) {
499 		pr_debug("%s: No PE found for device %04x:%02x:%02x.%01x\n",
500 			 __func__,  pdn->phb->global_number,
501 			 pdn->busno,
502 			 PCI_SLOT(pdn->devfn),
503 			 PCI_FUNC(pdn->devfn));
504 		return -EEXIST;
505 	}
506 
507 	/* Remove the EEH device */
508 	edev->pe = NULL;
509 	list_del(&edev->entry);
510 
511 	/*
512 	 * Check if the parent PE includes any EEH devices.
513 	 * If not, we should delete that. Also, we should
514 	 * delete the parent PE if it doesn't have associated
515 	 * child PEs and EEH devices.
516 	 */
517 	while (1) {
518 		parent = pe->parent;
519 		if (pe->type & EEH_PE_PHB)
520 			break;
521 
522 		if (!(pe->state & EEH_PE_KEEP)) {
523 			if (list_empty(&pe->edevs) &&
524 			    list_empty(&pe->child_list)) {
525 				list_del(&pe->child);
526 				kfree(pe);
527 			} else {
528 				break;
529 			}
530 		} else {
531 			if (list_empty(&pe->edevs)) {
532 				cnt = 0;
533 				list_for_each_entry(child, &pe->child_list, child) {
534 					if (!(child->type & EEH_PE_INVALID)) {
535 						cnt++;
536 						break;
537 					}
538 				}
539 
540 				if (!cnt)
541 					pe->type |= EEH_PE_INVALID;
542 				else
543 					break;
544 			}
545 		}
546 
547 		pe = parent;
548 	}
549 
550 	return 0;
551 }
552 
553 /**
554  * eeh_pe_update_time_stamp - Update PE's frozen time stamp
555  * @pe: EEH PE
556  *
557  * We have time stamp for each PE to trace its time of getting
558  * frozen in last hour. The function should be called to update
559  * the time stamp on first error of the specific PE. On the other
560  * handle, we needn't account for errors happened in last hour.
561  */
562 void eeh_pe_update_time_stamp(struct eeh_pe *pe)
563 {
564 	time64_t tstamp;
565 
566 	if (!pe) return;
567 
568 	if (pe->freeze_count <= 0) {
569 		pe->freeze_count = 0;
570 		pe->tstamp = ktime_get_seconds();
571 	} else {
572 		tstamp = ktime_get_seconds();
573 		if (tstamp - pe->tstamp > 3600) {
574 			pe->tstamp = tstamp;
575 			pe->freeze_count = 0;
576 		}
577 	}
578 }
579 
580 /**
581  * eeh_pe_state_mark - Mark specified state for PE and its associated device
582  * @pe: EEH PE
583  *
584  * EEH error affects the current PE and its child PEs. The function
585  * is used to mark appropriate state for the affected PEs and the
586  * associated devices.
587  */
588 void eeh_pe_state_mark(struct eeh_pe *root, int state)
589 {
590 	struct eeh_pe *pe;
591 
592 	eeh_for_each_pe(root, pe)
593 		if (!(pe->state & EEH_PE_REMOVED))
594 			pe->state |= state;
595 }
596 EXPORT_SYMBOL_GPL(eeh_pe_state_mark);
597 
598 /**
599  * eeh_pe_mark_isolated
600  * @pe: EEH PE
601  *
602  * Record that a PE has been isolated by marking the PE and it's children as
603  * EEH_PE_ISOLATED (and EEH_PE_CFG_BLOCKED, if required) and their PCI devices
604  * as pci_channel_io_frozen.
605  */
606 void eeh_pe_mark_isolated(struct eeh_pe *root)
607 {
608 	struct eeh_pe *pe;
609 	struct eeh_dev *edev;
610 	struct pci_dev *pdev;
611 
612 	eeh_pe_state_mark(root, EEH_PE_ISOLATED);
613 	eeh_for_each_pe(root, pe) {
614 		list_for_each_entry(edev, &pe->edevs, entry) {
615 			pdev = eeh_dev_to_pci_dev(edev);
616 			if (pdev)
617 				pdev->error_state = pci_channel_io_frozen;
618 		}
619 		/* Block PCI config access if required */
620 		if (pe->state & EEH_PE_CFG_RESTRICTED)
621 			pe->state |= EEH_PE_CFG_BLOCKED;
622 	}
623 }
624 EXPORT_SYMBOL_GPL(eeh_pe_mark_isolated);
625 
626 static void *__eeh_pe_dev_mode_mark(struct eeh_dev *edev, void *flag)
627 {
628 	int mode = *((int *)flag);
629 
630 	edev->mode |= mode;
631 
632 	return NULL;
633 }
634 
635 /**
636  * eeh_pe_dev_state_mark - Mark state for all device under the PE
637  * @pe: EEH PE
638  *
639  * Mark specific state for all child devices of the PE.
640  */
641 void eeh_pe_dev_mode_mark(struct eeh_pe *pe, int mode)
642 {
643 	eeh_pe_dev_traverse(pe, __eeh_pe_dev_mode_mark, &mode);
644 }
645 
646 /**
647  * eeh_pe_state_clear - Clear state for the PE
648  * @data: EEH PE
649  * @state: state
650  * @include_passed: include passed-through devices?
651  *
652  * The function is used to clear the indicated state from the
653  * given PE. Besides, we also clear the check count of the PE
654  * as well.
655  */
656 void eeh_pe_state_clear(struct eeh_pe *root, int state, bool include_passed)
657 {
658 	struct eeh_pe *pe;
659 	struct eeh_dev *edev, *tmp;
660 	struct pci_dev *pdev;
661 
662 	eeh_for_each_pe(root, pe) {
663 		/* Keep the state of permanently removed PE intact */
664 		if (pe->state & EEH_PE_REMOVED)
665 			continue;
666 
667 		if (!include_passed && eeh_pe_passed(pe))
668 			continue;
669 
670 		pe->state &= ~state;
671 
672 		/*
673 		 * Special treatment on clearing isolated state. Clear
674 		 * check count since last isolation and put all affected
675 		 * devices to normal state.
676 		 */
677 		if (!(state & EEH_PE_ISOLATED))
678 			continue;
679 
680 		pe->check_count = 0;
681 		eeh_pe_for_each_dev(pe, edev, tmp) {
682 			pdev = eeh_dev_to_pci_dev(edev);
683 			if (!pdev)
684 				continue;
685 
686 			pdev->error_state = pci_channel_io_normal;
687 		}
688 
689 		/* Unblock PCI config access if required */
690 		if (pe->state & EEH_PE_CFG_RESTRICTED)
691 			pe->state &= ~EEH_PE_CFG_BLOCKED;
692 	}
693 }
694 
695 /*
696  * Some PCI bridges (e.g. PLX bridges) have primary/secondary
697  * buses assigned explicitly by firmware, and we probably have
698  * lost that after reset. So we have to delay the check until
699  * the PCI-CFG registers have been restored for the parent
700  * bridge.
701  *
702  * Don't use normal PCI-CFG accessors, which probably has been
703  * blocked on normal path during the stage. So we need utilize
704  * eeh operations, which is always permitted.
705  */
706 static void eeh_bridge_check_link(struct eeh_dev *edev)
707 {
708 	struct pci_dn *pdn = eeh_dev_to_pdn(edev);
709 	int cap;
710 	uint32_t val;
711 	int timeout = 0;
712 
713 	/*
714 	 * We only check root port and downstream ports of
715 	 * PCIe switches
716 	 */
717 	if (!(edev->mode & (EEH_DEV_ROOT_PORT | EEH_DEV_DS_PORT)))
718 		return;
719 
720 	pr_debug("%s: Check PCIe link for %04x:%02x:%02x.%01x ...\n",
721 		 __func__, pdn->phb->global_number,
722 		 pdn->busno,
723 		 PCI_SLOT(pdn->devfn),
724 		 PCI_FUNC(pdn->devfn));
725 
726 	/* Check slot status */
727 	cap = edev->pcie_cap;
728 	eeh_ops->read_config(pdn, cap + PCI_EXP_SLTSTA, 2, &val);
729 	if (!(val & PCI_EXP_SLTSTA_PDS)) {
730 		pr_debug("  No card in the slot (0x%04x) !\n", val);
731 		return;
732 	}
733 
734 	/* Check power status if we have the capability */
735 	eeh_ops->read_config(pdn, cap + PCI_EXP_SLTCAP, 2, &val);
736 	if (val & PCI_EXP_SLTCAP_PCP) {
737 		eeh_ops->read_config(pdn, cap + PCI_EXP_SLTCTL, 2, &val);
738 		if (val & PCI_EXP_SLTCTL_PCC) {
739 			pr_debug("  In power-off state, power it on ...\n");
740 			val &= ~(PCI_EXP_SLTCTL_PCC | PCI_EXP_SLTCTL_PIC);
741 			val |= (0x0100 & PCI_EXP_SLTCTL_PIC);
742 			eeh_ops->write_config(pdn, cap + PCI_EXP_SLTCTL, 2, val);
743 			msleep(2 * 1000);
744 		}
745 	}
746 
747 	/* Enable link */
748 	eeh_ops->read_config(pdn, cap + PCI_EXP_LNKCTL, 2, &val);
749 	val &= ~PCI_EXP_LNKCTL_LD;
750 	eeh_ops->write_config(pdn, cap + PCI_EXP_LNKCTL, 2, val);
751 
752 	/* Check link */
753 	eeh_ops->read_config(pdn, cap + PCI_EXP_LNKCAP, 4, &val);
754 	if (!(val & PCI_EXP_LNKCAP_DLLLARC)) {
755 		pr_debug("  No link reporting capability (0x%08x) \n", val);
756 		msleep(1000);
757 		return;
758 	}
759 
760 	/* Wait the link is up until timeout (5s) */
761 	timeout = 0;
762 	while (timeout < 5000) {
763 		msleep(20);
764 		timeout += 20;
765 
766 		eeh_ops->read_config(pdn, cap + PCI_EXP_LNKSTA, 2, &val);
767 		if (val & PCI_EXP_LNKSTA_DLLLA)
768 			break;
769 	}
770 
771 	if (val & PCI_EXP_LNKSTA_DLLLA)
772 		pr_debug("  Link up (%s)\n",
773 			 (val & PCI_EXP_LNKSTA_CLS_2_5GB) ? "2.5GB" : "5GB");
774 	else
775 		pr_debug("  Link not ready (0x%04x)\n", val);
776 }
777 
778 #define BYTE_SWAP(OFF)	(8*((OFF)/4)+3-(OFF))
779 #define SAVED_BYTE(OFF)	(((u8 *)(edev->config_space))[BYTE_SWAP(OFF)])
780 
781 static void eeh_restore_bridge_bars(struct eeh_dev *edev)
782 {
783 	struct pci_dn *pdn = eeh_dev_to_pdn(edev);
784 	int i;
785 
786 	/*
787 	 * Device BARs: 0x10 - 0x18
788 	 * Bus numbers and windows: 0x18 - 0x30
789 	 */
790 	for (i = 4; i < 13; i++)
791 		eeh_ops->write_config(pdn, i*4, 4, edev->config_space[i]);
792 	/* Rom: 0x38 */
793 	eeh_ops->write_config(pdn, 14*4, 4, edev->config_space[14]);
794 
795 	/* Cache line & Latency timer: 0xC 0xD */
796 	eeh_ops->write_config(pdn, PCI_CACHE_LINE_SIZE, 1,
797                 SAVED_BYTE(PCI_CACHE_LINE_SIZE));
798         eeh_ops->write_config(pdn, PCI_LATENCY_TIMER, 1,
799                 SAVED_BYTE(PCI_LATENCY_TIMER));
800 	/* Max latency, min grant, interrupt ping and line: 0x3C */
801 	eeh_ops->write_config(pdn, 15*4, 4, edev->config_space[15]);
802 
803 	/* PCI Command: 0x4 */
804 	eeh_ops->write_config(pdn, PCI_COMMAND, 4, edev->config_space[1] |
805 			      PCI_COMMAND_MEMORY | PCI_COMMAND_MASTER);
806 
807 	/* Check the PCIe link is ready */
808 	eeh_bridge_check_link(edev);
809 }
810 
811 static void eeh_restore_device_bars(struct eeh_dev *edev)
812 {
813 	struct pci_dn *pdn = eeh_dev_to_pdn(edev);
814 	int i;
815 	u32 cmd;
816 
817 	for (i = 4; i < 10; i++)
818 		eeh_ops->write_config(pdn, i*4, 4, edev->config_space[i]);
819 	/* 12 == Expansion ROM Address */
820 	eeh_ops->write_config(pdn, 12*4, 4, edev->config_space[12]);
821 
822 	eeh_ops->write_config(pdn, PCI_CACHE_LINE_SIZE, 1,
823 		SAVED_BYTE(PCI_CACHE_LINE_SIZE));
824 	eeh_ops->write_config(pdn, PCI_LATENCY_TIMER, 1,
825 		SAVED_BYTE(PCI_LATENCY_TIMER));
826 
827 	/* max latency, min grant, interrupt pin and line */
828 	eeh_ops->write_config(pdn, 15*4, 4, edev->config_space[15]);
829 
830 	/*
831 	 * Restore PERR & SERR bits, some devices require it,
832 	 * don't touch the other command bits
833 	 */
834 	eeh_ops->read_config(pdn, PCI_COMMAND, 4, &cmd);
835 	if (edev->config_space[1] & PCI_COMMAND_PARITY)
836 		cmd |= PCI_COMMAND_PARITY;
837 	else
838 		cmd &= ~PCI_COMMAND_PARITY;
839 	if (edev->config_space[1] & PCI_COMMAND_SERR)
840 		cmd |= PCI_COMMAND_SERR;
841 	else
842 		cmd &= ~PCI_COMMAND_SERR;
843 	eeh_ops->write_config(pdn, PCI_COMMAND, 4, cmd);
844 }
845 
846 /**
847  * eeh_restore_one_device_bars - Restore the Base Address Registers for one device
848  * @data: EEH device
849  * @flag: Unused
850  *
851  * Loads the PCI configuration space base address registers,
852  * the expansion ROM base address, the latency timer, and etc.
853  * from the saved values in the device node.
854  */
855 static void *eeh_restore_one_device_bars(struct eeh_dev *edev, void *flag)
856 {
857 	struct pci_dn *pdn = eeh_dev_to_pdn(edev);
858 
859 	/* Do special restore for bridges */
860 	if (edev->mode & EEH_DEV_BRIDGE)
861 		eeh_restore_bridge_bars(edev);
862 	else
863 		eeh_restore_device_bars(edev);
864 
865 	if (eeh_ops->restore_config && pdn)
866 		eeh_ops->restore_config(pdn);
867 
868 	return NULL;
869 }
870 
871 /**
872  * eeh_pe_restore_bars - Restore the PCI config space info
873  * @pe: EEH PE
874  *
875  * This routine performs a recursive walk to the children
876  * of this device as well.
877  */
878 void eeh_pe_restore_bars(struct eeh_pe *pe)
879 {
880 	/*
881 	 * We needn't take the EEH lock since eeh_pe_dev_traverse()
882 	 * will take that.
883 	 */
884 	eeh_pe_dev_traverse(pe, eeh_restore_one_device_bars, NULL);
885 }
886 
887 /**
888  * eeh_pe_loc_get - Retrieve location code binding to the given PE
889  * @pe: EEH PE
890  *
891  * Retrieve the location code of the given PE. If the primary PE bus
892  * is root bus, we will grab location code from PHB device tree node
893  * or root port. Otherwise, the upstream bridge's device tree node
894  * of the primary PE bus will be checked for the location code.
895  */
896 const char *eeh_pe_loc_get(struct eeh_pe *pe)
897 {
898 	struct pci_bus *bus = eeh_pe_bus_get(pe);
899 	struct device_node *dn;
900 	const char *loc = NULL;
901 
902 	while (bus) {
903 		dn = pci_bus_to_OF_node(bus);
904 		if (!dn) {
905 			bus = bus->parent;
906 			continue;
907 		}
908 
909 		if (pci_is_root_bus(bus))
910 			loc = of_get_property(dn, "ibm,io-base-loc-code", NULL);
911 		else
912 			loc = of_get_property(dn, "ibm,slot-location-code",
913 					      NULL);
914 
915 		if (loc)
916 			return loc;
917 
918 		bus = bus->parent;
919 	}
920 
921 	return "N/A";
922 }
923 
924 /**
925  * eeh_pe_bus_get - Retrieve PCI bus according to the given PE
926  * @pe: EEH PE
927  *
928  * Retrieve the PCI bus according to the given PE. Basically,
929  * there're 3 types of PEs: PHB/Bus/Device. For PHB PE, the
930  * primary PCI bus will be retrieved. The parent bus will be
931  * returned for BUS PE. However, we don't have associated PCI
932  * bus for DEVICE PE.
933  */
934 struct pci_bus *eeh_pe_bus_get(struct eeh_pe *pe)
935 {
936 	struct eeh_dev *edev;
937 	struct pci_dev *pdev;
938 
939 	if (pe->type & EEH_PE_PHB)
940 		return pe->phb->bus;
941 
942 	/* The primary bus might be cached during probe time */
943 	if (pe->state & EEH_PE_PRI_BUS)
944 		return pe->bus;
945 
946 	/* Retrieve the parent PCI bus of first (top) PCI device */
947 	edev = list_first_entry_or_null(&pe->edevs, struct eeh_dev, entry);
948 	pdev = eeh_dev_to_pci_dev(edev);
949 	if (pdev)
950 		return pdev->bus;
951 
952 	return NULL;
953 }
954