xref: /openbmc/linux/arch/powerpc/kernel/eeh_pe.c (revision cfdfc14e)
1 /*
2  * The file intends to implement PE based on the information from
3  * platforms. Basically, there have 3 types of PEs: PHB/Bus/Device.
4  * All the PEs should be organized as hierarchy tree. The first level
5  * of the tree will be associated to existing PHBs since the particular
6  * PE is only meaningful in one PHB domain.
7  *
8  * Copyright Benjamin Herrenschmidt & Gavin Shan, IBM Corporation 2012.
9  *
10  * This program is free software; you can redistribute it and/or modify
11  * it under the terms of the GNU General Public License as published by
12  * the Free Software Foundation; either version 2 of the License, or
13  * (at your option) any later version.
14  *
15  * This program is distributed in the hope that it will be useful,
16  * but WITHOUT ANY WARRANTY; without even the implied warranty of
17  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
18  * GNU General Public License for more details.
19  *
20  * You should have received a copy of the GNU General Public License
21  * along with this program; if not, write to the Free Software
22  * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA  02111-1307 USA
23  */
24 
25 #include <linux/delay.h>
26 #include <linux/export.h>
27 #include <linux/gfp.h>
28 #include <linux/kernel.h>
29 #include <linux/pci.h>
30 #include <linux/string.h>
31 
32 #include <asm/pci-bridge.h>
33 #include <asm/ppc-pci.h>
34 
35 static int eeh_pe_aux_size = 0;
36 static LIST_HEAD(eeh_phb_pe);
37 
38 /**
39  * eeh_set_pe_aux_size - Set PE auxillary data size
40  * @size: PE auxillary data size
41  *
42  * Set PE auxillary data size
43  */
44 void eeh_set_pe_aux_size(int size)
45 {
46 	if (size < 0)
47 		return;
48 
49 	eeh_pe_aux_size = size;
50 }
51 
52 /**
53  * eeh_pe_alloc - Allocate PE
54  * @phb: PCI controller
55  * @type: PE type
56  *
57  * Allocate PE instance dynamically.
58  */
59 static struct eeh_pe *eeh_pe_alloc(struct pci_controller *phb, int type)
60 {
61 	struct eeh_pe *pe;
62 	size_t alloc_size;
63 
64 	alloc_size = sizeof(struct eeh_pe);
65 	if (eeh_pe_aux_size) {
66 		alloc_size = ALIGN(alloc_size, cache_line_size());
67 		alloc_size += eeh_pe_aux_size;
68 	}
69 
70 	/* Allocate PHB PE */
71 	pe = kzalloc(alloc_size, GFP_KERNEL);
72 	if (!pe) return NULL;
73 
74 	/* Initialize PHB PE */
75 	pe->type = type;
76 	pe->phb = phb;
77 	INIT_LIST_HEAD(&pe->child_list);
78 	INIT_LIST_HEAD(&pe->child);
79 	INIT_LIST_HEAD(&pe->edevs);
80 
81 	pe->data = (void *)pe + ALIGN(sizeof(struct eeh_pe),
82 				      cache_line_size());
83 	return pe;
84 }
85 
86 /**
87  * eeh_phb_pe_create - Create PHB PE
88  * @phb: PCI controller
89  *
90  * The function should be called while the PHB is detected during
91  * system boot or PCI hotplug in order to create PHB PE.
92  */
93 int eeh_phb_pe_create(struct pci_controller *phb)
94 {
95 	struct eeh_pe *pe;
96 
97 	/* Allocate PHB PE */
98 	pe = eeh_pe_alloc(phb, EEH_PE_PHB);
99 	if (!pe) {
100 		pr_err("%s: out of memory!\n", __func__);
101 		return -ENOMEM;
102 	}
103 
104 	/* Put it into the list */
105 	list_add_tail(&pe->child, &eeh_phb_pe);
106 
107 	pr_debug("EEH: Add PE for PHB#%x\n", phb->global_number);
108 
109 	return 0;
110 }
111 
112 /**
113  * eeh_phb_pe_get - Retrieve PHB PE based on the given PHB
114  * @phb: PCI controller
115  *
116  * The overall PEs form hierarchy tree. The first layer of the
117  * hierarchy tree is composed of PHB PEs. The function is used
118  * to retrieve the corresponding PHB PE according to the given PHB.
119  */
120 struct eeh_pe *eeh_phb_pe_get(struct pci_controller *phb)
121 {
122 	struct eeh_pe *pe;
123 
124 	list_for_each_entry(pe, &eeh_phb_pe, child) {
125 		/*
126 		 * Actually, we needn't check the type since
127 		 * the PE for PHB has been determined when that
128 		 * was created.
129 		 */
130 		if ((pe->type & EEH_PE_PHB) && pe->phb == phb)
131 			return pe;
132 	}
133 
134 	return NULL;
135 }
136 
137 /**
138  * eeh_pe_next - Retrieve the next PE in the tree
139  * @pe: current PE
140  * @root: root PE
141  *
142  * The function is used to retrieve the next PE in the
143  * hierarchy PE tree.
144  */
145 struct eeh_pe *eeh_pe_next(struct eeh_pe *pe, struct eeh_pe *root)
146 {
147 	struct list_head *next = pe->child_list.next;
148 
149 	if (next == &pe->child_list) {
150 		while (1) {
151 			if (pe == root)
152 				return NULL;
153 			next = pe->child.next;
154 			if (next != &pe->parent->child_list)
155 				break;
156 			pe = pe->parent;
157 		}
158 	}
159 
160 	return list_entry(next, struct eeh_pe, child);
161 }
162 
163 /**
164  * eeh_pe_traverse - Traverse PEs in the specified PHB
165  * @root: root PE
166  * @fn: callback
167  * @flag: extra parameter to callback
168  *
169  * The function is used to traverse the specified PE and its
170  * child PEs. The traversing is to be terminated once the
171  * callback returns something other than NULL, or no more PEs
172  * to be traversed.
173  */
174 void *eeh_pe_traverse(struct eeh_pe *root,
175 		      eeh_pe_traverse_func fn, void *flag)
176 {
177 	struct eeh_pe *pe;
178 	void *ret;
179 
180 	eeh_for_each_pe(root, pe) {
181 		ret = fn(pe, flag);
182 		if (ret) return ret;
183 	}
184 
185 	return NULL;
186 }
187 
188 /**
189  * eeh_pe_dev_traverse - Traverse the devices from the PE
190  * @root: EEH PE
191  * @fn: function callback
192  * @flag: extra parameter to callback
193  *
194  * The function is used to traverse the devices of the specified
195  * PE and its child PEs.
196  */
197 void *eeh_pe_dev_traverse(struct eeh_pe *root,
198 			  eeh_edev_traverse_func fn, void *flag)
199 {
200 	struct eeh_pe *pe;
201 	struct eeh_dev *edev, *tmp;
202 	void *ret;
203 
204 	if (!root) {
205 		pr_warn("%s: Invalid PE %p\n",
206 			__func__, root);
207 		return NULL;
208 	}
209 
210 	/* Traverse root PE */
211 	eeh_for_each_pe(root, pe) {
212 		eeh_pe_for_each_dev(pe, edev, tmp) {
213 			ret = fn(edev, flag);
214 			if (ret)
215 				return ret;
216 		}
217 	}
218 
219 	return NULL;
220 }
221 
222 /**
223  * __eeh_pe_get - Check the PE address
224  * @data: EEH PE
225  * @flag: EEH device
226  *
227  * For one particular PE, it can be identified by PE address
228  * or tranditional BDF address. BDF address is composed of
229  * Bus/Device/Function number. The extra data referred by flag
230  * indicates which type of address should be used.
231  */
232 struct eeh_pe_get_flag {
233 	int pe_no;
234 	int config_addr;
235 };
236 
237 static void *__eeh_pe_get(struct eeh_pe *pe, void *flag)
238 {
239 	struct eeh_pe_get_flag *tmp = (struct eeh_pe_get_flag *) flag;
240 
241 	/* Unexpected PHB PE */
242 	if (pe->type & EEH_PE_PHB)
243 		return NULL;
244 
245 	/*
246 	 * We prefer PE address. For most cases, we should
247 	 * have non-zero PE address
248 	 */
249 	if (eeh_has_flag(EEH_VALID_PE_ZERO)) {
250 		if (tmp->pe_no == pe->addr)
251 			return pe;
252 	} else {
253 		if (tmp->pe_no &&
254 		    (tmp->pe_no == pe->addr))
255 			return pe;
256 	}
257 
258 	/* Try BDF address */
259 	if (tmp->config_addr &&
260 	   (tmp->config_addr == pe->config_addr))
261 		return pe;
262 
263 	return NULL;
264 }
265 
266 /**
267  * eeh_pe_get - Search PE based on the given address
268  * @phb: PCI controller
269  * @pe_no: PE number
270  * @config_addr: Config address
271  *
272  * Search the corresponding PE based on the specified address which
273  * is included in the eeh device. The function is used to check if
274  * the associated PE has been created against the PE address. It's
275  * notable that the PE address has 2 format: traditional PE address
276  * which is composed of PCI bus/device/function number, or unified
277  * PE address.
278  */
279 struct eeh_pe *eeh_pe_get(struct pci_controller *phb,
280 		int pe_no, int config_addr)
281 {
282 	struct eeh_pe *root = eeh_phb_pe_get(phb);
283 	struct eeh_pe_get_flag tmp = { pe_no, config_addr };
284 	struct eeh_pe *pe;
285 
286 	pe = eeh_pe_traverse(root, __eeh_pe_get, &tmp);
287 
288 	return pe;
289 }
290 
291 /**
292  * eeh_pe_get_parent - Retrieve the parent PE
293  * @edev: EEH device
294  *
295  * The whole PEs existing in the system are organized as hierarchy
296  * tree. The function is used to retrieve the parent PE according
297  * to the parent EEH device.
298  */
299 static struct eeh_pe *eeh_pe_get_parent(struct eeh_dev *edev)
300 {
301 	struct eeh_dev *parent;
302 	struct pci_dn *pdn = eeh_dev_to_pdn(edev);
303 
304 	/*
305 	 * It might have the case for the indirect parent
306 	 * EEH device already having associated PE, but
307 	 * the direct parent EEH device doesn't have yet.
308 	 */
309 	if (edev->physfn)
310 		pdn = pci_get_pdn(edev->physfn);
311 	else
312 		pdn = pdn ? pdn->parent : NULL;
313 	while (pdn) {
314 		/* We're poking out of PCI territory */
315 		parent = pdn_to_eeh_dev(pdn);
316 		if (!parent)
317 			return NULL;
318 
319 		if (parent->pe)
320 			return parent->pe;
321 
322 		pdn = pdn->parent;
323 	}
324 
325 	return NULL;
326 }
327 
328 /**
329  * eeh_add_to_parent_pe - Add EEH device to parent PE
330  * @edev: EEH device
331  *
332  * Add EEH device to the parent PE. If the parent PE already
333  * exists, the PE type will be changed to EEH_PE_BUS. Otherwise,
334  * we have to create new PE to hold the EEH device and the new
335  * PE will be linked to its parent PE as well.
336  */
337 int eeh_add_to_parent_pe(struct eeh_dev *edev)
338 {
339 	struct eeh_pe *pe, *parent;
340 	struct pci_dn *pdn = eeh_dev_to_pdn(edev);
341 	int config_addr = (pdn->busno << 8) | (pdn->devfn);
342 
343 	/* Check if the PE number is valid */
344 	if (!eeh_has_flag(EEH_VALID_PE_ZERO) && !edev->pe_config_addr) {
345 		pr_err("%s: Invalid PE#0 for edev 0x%x on PHB#%x\n",
346 		       __func__, config_addr, pdn->phb->global_number);
347 		return -EINVAL;
348 	}
349 
350 	/*
351 	 * Search the PE has been existing or not according
352 	 * to the PE address. If that has been existing, the
353 	 * PE should be composed of PCI bus and its subordinate
354 	 * components.
355 	 */
356 	pe = eeh_pe_get(pdn->phb, edev->pe_config_addr, config_addr);
357 	if (pe && !(pe->type & EEH_PE_INVALID)) {
358 		/* Mark the PE as type of PCI bus */
359 		pe->type = EEH_PE_BUS;
360 		edev->pe = pe;
361 
362 		/* Put the edev to PE */
363 		list_add_tail(&edev->list, &pe->edevs);
364 		pr_debug("EEH: Add %04x:%02x:%02x.%01x to Bus PE#%x\n",
365 			 pdn->phb->global_number,
366 			 pdn->busno,
367 			 PCI_SLOT(pdn->devfn),
368 			 PCI_FUNC(pdn->devfn),
369 			 pe->addr);
370 		return 0;
371 	} else if (pe && (pe->type & EEH_PE_INVALID)) {
372 		list_add_tail(&edev->list, &pe->edevs);
373 		edev->pe = pe;
374 		/*
375 		 * We're running to here because of PCI hotplug caused by
376 		 * EEH recovery. We need clear EEH_PE_INVALID until the top.
377 		 */
378 		parent = pe;
379 		while (parent) {
380 			if (!(parent->type & EEH_PE_INVALID))
381 				break;
382 			parent->type &= ~(EEH_PE_INVALID | EEH_PE_KEEP);
383 			parent = parent->parent;
384 		}
385 
386 		pr_debug("EEH: Add %04x:%02x:%02x.%01x to Device "
387 			 "PE#%x, Parent PE#%x\n",
388 			 pdn->phb->global_number,
389 			 pdn->busno,
390 			 PCI_SLOT(pdn->devfn),
391 			 PCI_FUNC(pdn->devfn),
392 			 pe->addr, pe->parent->addr);
393 		return 0;
394 	}
395 
396 	/* Create a new EEH PE */
397 	if (edev->physfn)
398 		pe = eeh_pe_alloc(pdn->phb, EEH_PE_VF);
399 	else
400 		pe = eeh_pe_alloc(pdn->phb, EEH_PE_DEVICE);
401 	if (!pe) {
402 		pr_err("%s: out of memory!\n", __func__);
403 		return -ENOMEM;
404 	}
405 	pe->addr	= edev->pe_config_addr;
406 	pe->config_addr	= config_addr;
407 
408 	/*
409 	 * Put the new EEH PE into hierarchy tree. If the parent
410 	 * can't be found, the newly created PE will be attached
411 	 * to PHB directly. Otherwise, we have to associate the
412 	 * PE with its parent.
413 	 */
414 	parent = eeh_pe_get_parent(edev);
415 	if (!parent) {
416 		parent = eeh_phb_pe_get(pdn->phb);
417 		if (!parent) {
418 			pr_err("%s: No PHB PE is found (PHB Domain=%d)\n",
419 				__func__, pdn->phb->global_number);
420 			edev->pe = NULL;
421 			kfree(pe);
422 			return -EEXIST;
423 		}
424 	}
425 	pe->parent = parent;
426 
427 	/*
428 	 * Put the newly created PE into the child list and
429 	 * link the EEH device accordingly.
430 	 */
431 	list_add_tail(&pe->child, &parent->child_list);
432 	list_add_tail(&edev->list, &pe->edevs);
433 	edev->pe = pe;
434 	pr_debug("EEH: Add %04x:%02x:%02x.%01x to "
435 		 "Device PE#%x, Parent PE#%x\n",
436 		 pdn->phb->global_number,
437 		 pdn->busno,
438 		 PCI_SLOT(pdn->devfn),
439 		 PCI_FUNC(pdn->devfn),
440 		 pe->addr, pe->parent->addr);
441 
442 	return 0;
443 }
444 
445 /**
446  * eeh_rmv_from_parent_pe - Remove one EEH device from the associated PE
447  * @edev: EEH device
448  *
449  * The PE hierarchy tree might be changed when doing PCI hotplug.
450  * Also, the PCI devices or buses could be removed from the system
451  * during EEH recovery. So we have to call the function remove the
452  * corresponding PE accordingly if necessary.
453  */
454 int eeh_rmv_from_parent_pe(struct eeh_dev *edev)
455 {
456 	struct eeh_pe *pe, *parent, *child;
457 	int cnt;
458 	struct pci_dn *pdn = eeh_dev_to_pdn(edev);
459 
460 	if (!edev->pe) {
461 		pr_debug("%s: No PE found for device %04x:%02x:%02x.%01x\n",
462 			 __func__,  pdn->phb->global_number,
463 			 pdn->busno,
464 			 PCI_SLOT(pdn->devfn),
465 			 PCI_FUNC(pdn->devfn));
466 		return -EEXIST;
467 	}
468 
469 	/* Remove the EEH device */
470 	pe = eeh_dev_to_pe(edev);
471 	edev->pe = NULL;
472 	list_del(&edev->list);
473 
474 	/*
475 	 * Check if the parent PE includes any EEH devices.
476 	 * If not, we should delete that. Also, we should
477 	 * delete the parent PE if it doesn't have associated
478 	 * child PEs and EEH devices.
479 	 */
480 	while (1) {
481 		parent = pe->parent;
482 		if (pe->type & EEH_PE_PHB)
483 			break;
484 
485 		if (!(pe->state & EEH_PE_KEEP)) {
486 			if (list_empty(&pe->edevs) &&
487 			    list_empty(&pe->child_list)) {
488 				list_del(&pe->child);
489 				kfree(pe);
490 			} else {
491 				break;
492 			}
493 		} else {
494 			if (list_empty(&pe->edevs)) {
495 				cnt = 0;
496 				list_for_each_entry(child, &pe->child_list, child) {
497 					if (!(child->type & EEH_PE_INVALID)) {
498 						cnt++;
499 						break;
500 					}
501 				}
502 
503 				if (!cnt)
504 					pe->type |= EEH_PE_INVALID;
505 				else
506 					break;
507 			}
508 		}
509 
510 		pe = parent;
511 	}
512 
513 	return 0;
514 }
515 
516 /**
517  * eeh_pe_update_time_stamp - Update PE's frozen time stamp
518  * @pe: EEH PE
519  *
520  * We have time stamp for each PE to trace its time of getting
521  * frozen in last hour. The function should be called to update
522  * the time stamp on first error of the specific PE. On the other
523  * handle, we needn't account for errors happened in last hour.
524  */
525 void eeh_pe_update_time_stamp(struct eeh_pe *pe)
526 {
527 	time64_t tstamp;
528 
529 	if (!pe) return;
530 
531 	if (pe->freeze_count <= 0) {
532 		pe->freeze_count = 0;
533 		pe->tstamp = ktime_get_seconds();
534 	} else {
535 		tstamp = ktime_get_seconds();
536 		if (tstamp - pe->tstamp > 3600) {
537 			pe->tstamp = tstamp;
538 			pe->freeze_count = 0;
539 		}
540 	}
541 }
542 
543 /**
544  * __eeh_pe_state_mark - Mark the state for the PE
545  * @data: EEH PE
546  * @flag: state
547  *
548  * The function is used to mark the indicated state for the given
549  * PE. Also, the associated PCI devices will be put into IO frozen
550  * state as well.
551  */
552 static void *__eeh_pe_state_mark(struct eeh_pe *pe, void *flag)
553 {
554 	int state = *((int *)flag);
555 	struct eeh_dev *edev, *tmp;
556 	struct pci_dev *pdev;
557 
558 	/* Keep the state of permanently removed PE intact */
559 	if (pe->state & EEH_PE_REMOVED)
560 		return NULL;
561 
562 	pe->state |= state;
563 
564 	/* Offline PCI devices if applicable */
565 	if (!(state & EEH_PE_ISOLATED))
566 		return NULL;
567 
568 	eeh_pe_for_each_dev(pe, edev, tmp) {
569 		pdev = eeh_dev_to_pci_dev(edev);
570 		if (pdev)
571 			pdev->error_state = pci_channel_io_frozen;
572 	}
573 
574 	/* Block PCI config access if required */
575 	if (pe->state & EEH_PE_CFG_RESTRICTED)
576 		pe->state |= EEH_PE_CFG_BLOCKED;
577 
578 	return NULL;
579 }
580 
581 /**
582  * eeh_pe_state_mark - Mark specified state for PE and its associated device
583  * @pe: EEH PE
584  *
585  * EEH error affects the current PE and its child PEs. The function
586  * is used to mark appropriate state for the affected PEs and the
587  * associated devices.
588  */
589 void eeh_pe_state_mark(struct eeh_pe *pe, int state)
590 {
591 	eeh_pe_traverse(pe, __eeh_pe_state_mark, &state);
592 }
593 EXPORT_SYMBOL_GPL(eeh_pe_state_mark);
594 
595 static void *__eeh_pe_dev_mode_mark(struct eeh_dev *edev, void *flag)
596 {
597 	int mode = *((int *)flag);
598 
599 	edev->mode |= mode;
600 
601 	return NULL;
602 }
603 
604 /**
605  * eeh_pe_dev_state_mark - Mark state for all device under the PE
606  * @pe: EEH PE
607  *
608  * Mark specific state for all child devices of the PE.
609  */
610 void eeh_pe_dev_mode_mark(struct eeh_pe *pe, int mode)
611 {
612 	eeh_pe_dev_traverse(pe, __eeh_pe_dev_mode_mark, &mode);
613 }
614 
615 /**
616  * __eeh_pe_state_clear - Clear state for the PE
617  * @data: EEH PE
618  * @flag: state
619  *
620  * The function is used to clear the indicated state from the
621  * given PE. Besides, we also clear the check count of the PE
622  * as well.
623  */
624 static void *__eeh_pe_state_clear(struct eeh_pe *pe, void *flag)
625 {
626 	int state = *((int *)flag);
627 	struct eeh_dev *edev, *tmp;
628 	struct pci_dev *pdev;
629 
630 	/* Keep the state of permanently removed PE intact */
631 	if (pe->state & EEH_PE_REMOVED)
632 		return NULL;
633 
634 	pe->state &= ~state;
635 
636 	/*
637 	 * Special treatment on clearing isolated state. Clear
638 	 * check count since last isolation and put all affected
639 	 * devices to normal state.
640 	 */
641 	if (!(state & EEH_PE_ISOLATED))
642 		return NULL;
643 
644 	pe->check_count = 0;
645 	eeh_pe_for_each_dev(pe, edev, tmp) {
646 		pdev = eeh_dev_to_pci_dev(edev);
647 		if (!pdev)
648 			continue;
649 
650 		pdev->error_state = pci_channel_io_normal;
651 	}
652 
653 	/* Unblock PCI config access if required */
654 	if (pe->state & EEH_PE_CFG_RESTRICTED)
655 		pe->state &= ~EEH_PE_CFG_BLOCKED;
656 
657 	return NULL;
658 }
659 
660 /**
661  * eeh_pe_state_clear - Clear state for the PE and its children
662  * @pe: PE
663  * @state: state to be cleared
664  *
665  * When the PE and its children has been recovered from error,
666  * we need clear the error state for that. The function is used
667  * for the purpose.
668  */
669 void eeh_pe_state_clear(struct eeh_pe *pe, int state)
670 {
671 	eeh_pe_traverse(pe, __eeh_pe_state_clear, &state);
672 }
673 
674 /**
675  * eeh_pe_state_mark_with_cfg - Mark PE state with unblocked config space
676  * @pe: PE
677  * @state: PE state to be set
678  *
679  * Set specified flag to PE and its child PEs. The PCI config space
680  * of some PEs is blocked automatically when EEH_PE_ISOLATED is set,
681  * which isn't needed in some situations. The function allows to set
682  * the specified flag to indicated PEs without blocking their PCI
683  * config space.
684  */
685 void eeh_pe_state_mark_with_cfg(struct eeh_pe *pe, int state)
686 {
687 	eeh_pe_traverse(pe, __eeh_pe_state_mark, &state);
688 	if (!(state & EEH_PE_ISOLATED))
689 		return;
690 
691 	/* Clear EEH_PE_CFG_BLOCKED, which might be set just now */
692 	state = EEH_PE_CFG_BLOCKED;
693 	eeh_pe_traverse(pe, __eeh_pe_state_clear, &state);
694 }
695 
696 /*
697  * Some PCI bridges (e.g. PLX bridges) have primary/secondary
698  * buses assigned explicitly by firmware, and we probably have
699  * lost that after reset. So we have to delay the check until
700  * the PCI-CFG registers have been restored for the parent
701  * bridge.
702  *
703  * Don't use normal PCI-CFG accessors, which probably has been
704  * blocked on normal path during the stage. So we need utilize
705  * eeh operations, which is always permitted.
706  */
707 static void eeh_bridge_check_link(struct eeh_dev *edev)
708 {
709 	struct pci_dn *pdn = eeh_dev_to_pdn(edev);
710 	int cap;
711 	uint32_t val;
712 	int timeout = 0;
713 
714 	/*
715 	 * We only check root port and downstream ports of
716 	 * PCIe switches
717 	 */
718 	if (!(edev->mode & (EEH_DEV_ROOT_PORT | EEH_DEV_DS_PORT)))
719 		return;
720 
721 	pr_debug("%s: Check PCIe link for %04x:%02x:%02x.%01x ...\n",
722 		 __func__, pdn->phb->global_number,
723 		 pdn->busno,
724 		 PCI_SLOT(pdn->devfn),
725 		 PCI_FUNC(pdn->devfn));
726 
727 	/* Check slot status */
728 	cap = edev->pcie_cap;
729 	eeh_ops->read_config(pdn, cap + PCI_EXP_SLTSTA, 2, &val);
730 	if (!(val & PCI_EXP_SLTSTA_PDS)) {
731 		pr_debug("  No card in the slot (0x%04x) !\n", val);
732 		return;
733 	}
734 
735 	/* Check power status if we have the capability */
736 	eeh_ops->read_config(pdn, cap + PCI_EXP_SLTCAP, 2, &val);
737 	if (val & PCI_EXP_SLTCAP_PCP) {
738 		eeh_ops->read_config(pdn, cap + PCI_EXP_SLTCTL, 2, &val);
739 		if (val & PCI_EXP_SLTCTL_PCC) {
740 			pr_debug("  In power-off state, power it on ...\n");
741 			val &= ~(PCI_EXP_SLTCTL_PCC | PCI_EXP_SLTCTL_PIC);
742 			val |= (0x0100 & PCI_EXP_SLTCTL_PIC);
743 			eeh_ops->write_config(pdn, cap + PCI_EXP_SLTCTL, 2, val);
744 			msleep(2 * 1000);
745 		}
746 	}
747 
748 	/* Enable link */
749 	eeh_ops->read_config(pdn, cap + PCI_EXP_LNKCTL, 2, &val);
750 	val &= ~PCI_EXP_LNKCTL_LD;
751 	eeh_ops->write_config(pdn, cap + PCI_EXP_LNKCTL, 2, val);
752 
753 	/* Check link */
754 	eeh_ops->read_config(pdn, cap + PCI_EXP_LNKCAP, 4, &val);
755 	if (!(val & PCI_EXP_LNKCAP_DLLLARC)) {
756 		pr_debug("  No link reporting capability (0x%08x) \n", val);
757 		msleep(1000);
758 		return;
759 	}
760 
761 	/* Wait the link is up until timeout (5s) */
762 	timeout = 0;
763 	while (timeout < 5000) {
764 		msleep(20);
765 		timeout += 20;
766 
767 		eeh_ops->read_config(pdn, cap + PCI_EXP_LNKSTA, 2, &val);
768 		if (val & PCI_EXP_LNKSTA_DLLLA)
769 			break;
770 	}
771 
772 	if (val & PCI_EXP_LNKSTA_DLLLA)
773 		pr_debug("  Link up (%s)\n",
774 			 (val & PCI_EXP_LNKSTA_CLS_2_5GB) ? "2.5GB" : "5GB");
775 	else
776 		pr_debug("  Link not ready (0x%04x)\n", val);
777 }
778 
779 #define BYTE_SWAP(OFF)	(8*((OFF)/4)+3-(OFF))
780 #define SAVED_BYTE(OFF)	(((u8 *)(edev->config_space))[BYTE_SWAP(OFF)])
781 
782 static void eeh_restore_bridge_bars(struct eeh_dev *edev)
783 {
784 	struct pci_dn *pdn = eeh_dev_to_pdn(edev);
785 	int i;
786 
787 	/*
788 	 * Device BARs: 0x10 - 0x18
789 	 * Bus numbers and windows: 0x18 - 0x30
790 	 */
791 	for (i = 4; i < 13; i++)
792 		eeh_ops->write_config(pdn, i*4, 4, edev->config_space[i]);
793 	/* Rom: 0x38 */
794 	eeh_ops->write_config(pdn, 14*4, 4, edev->config_space[14]);
795 
796 	/* Cache line & Latency timer: 0xC 0xD */
797 	eeh_ops->write_config(pdn, PCI_CACHE_LINE_SIZE, 1,
798                 SAVED_BYTE(PCI_CACHE_LINE_SIZE));
799         eeh_ops->write_config(pdn, PCI_LATENCY_TIMER, 1,
800                 SAVED_BYTE(PCI_LATENCY_TIMER));
801 	/* Max latency, min grant, interrupt ping and line: 0x3C */
802 	eeh_ops->write_config(pdn, 15*4, 4, edev->config_space[15]);
803 
804 	/* PCI Command: 0x4 */
805 	eeh_ops->write_config(pdn, PCI_COMMAND, 4, edev->config_space[1] |
806 			      PCI_COMMAND_MEMORY | PCI_COMMAND_MASTER);
807 
808 	/* Check the PCIe link is ready */
809 	eeh_bridge_check_link(edev);
810 }
811 
812 static void eeh_restore_device_bars(struct eeh_dev *edev)
813 {
814 	struct pci_dn *pdn = eeh_dev_to_pdn(edev);
815 	int i;
816 	u32 cmd;
817 
818 	for (i = 4; i < 10; i++)
819 		eeh_ops->write_config(pdn, i*4, 4, edev->config_space[i]);
820 	/* 12 == Expansion ROM Address */
821 	eeh_ops->write_config(pdn, 12*4, 4, edev->config_space[12]);
822 
823 	eeh_ops->write_config(pdn, PCI_CACHE_LINE_SIZE, 1,
824 		SAVED_BYTE(PCI_CACHE_LINE_SIZE));
825 	eeh_ops->write_config(pdn, PCI_LATENCY_TIMER, 1,
826 		SAVED_BYTE(PCI_LATENCY_TIMER));
827 
828 	/* max latency, min grant, interrupt pin and line */
829 	eeh_ops->write_config(pdn, 15*4, 4, edev->config_space[15]);
830 
831 	/*
832 	 * Restore PERR & SERR bits, some devices require it,
833 	 * don't touch the other command bits
834 	 */
835 	eeh_ops->read_config(pdn, PCI_COMMAND, 4, &cmd);
836 	if (edev->config_space[1] & PCI_COMMAND_PARITY)
837 		cmd |= PCI_COMMAND_PARITY;
838 	else
839 		cmd &= ~PCI_COMMAND_PARITY;
840 	if (edev->config_space[1] & PCI_COMMAND_SERR)
841 		cmd |= PCI_COMMAND_SERR;
842 	else
843 		cmd &= ~PCI_COMMAND_SERR;
844 	eeh_ops->write_config(pdn, PCI_COMMAND, 4, cmd);
845 }
846 
847 /**
848  * eeh_restore_one_device_bars - Restore the Base Address Registers for one device
849  * @data: EEH device
850  * @flag: Unused
851  *
852  * Loads the PCI configuration space base address registers,
853  * the expansion ROM base address, the latency timer, and etc.
854  * from the saved values in the device node.
855  */
856 static void *eeh_restore_one_device_bars(struct eeh_dev *edev, void *flag)
857 {
858 	struct pci_dn *pdn = eeh_dev_to_pdn(edev);
859 
860 	/* Do special restore for bridges */
861 	if (edev->mode & EEH_DEV_BRIDGE)
862 		eeh_restore_bridge_bars(edev);
863 	else
864 		eeh_restore_device_bars(edev);
865 
866 	if (eeh_ops->restore_config && pdn)
867 		eeh_ops->restore_config(pdn);
868 
869 	return NULL;
870 }
871 
872 /**
873  * eeh_pe_restore_bars - Restore the PCI config space info
874  * @pe: EEH PE
875  *
876  * This routine performs a recursive walk to the children
877  * of this device as well.
878  */
879 void eeh_pe_restore_bars(struct eeh_pe *pe)
880 {
881 	/*
882 	 * We needn't take the EEH lock since eeh_pe_dev_traverse()
883 	 * will take that.
884 	 */
885 	eeh_pe_dev_traverse(pe, eeh_restore_one_device_bars, NULL);
886 }
887 
888 /**
889  * eeh_pe_loc_get - Retrieve location code binding to the given PE
890  * @pe: EEH PE
891  *
892  * Retrieve the location code of the given PE. If the primary PE bus
893  * is root bus, we will grab location code from PHB device tree node
894  * or root port. Otherwise, the upstream bridge's device tree node
895  * of the primary PE bus will be checked for the location code.
896  */
897 const char *eeh_pe_loc_get(struct eeh_pe *pe)
898 {
899 	struct pci_bus *bus = eeh_pe_bus_get(pe);
900 	struct device_node *dn;
901 	const char *loc = NULL;
902 
903 	while (bus) {
904 		dn = pci_bus_to_OF_node(bus);
905 		if (!dn) {
906 			bus = bus->parent;
907 			continue;
908 		}
909 
910 		if (pci_is_root_bus(bus))
911 			loc = of_get_property(dn, "ibm,io-base-loc-code", NULL);
912 		else
913 			loc = of_get_property(dn, "ibm,slot-location-code",
914 					      NULL);
915 
916 		if (loc)
917 			return loc;
918 
919 		bus = bus->parent;
920 	}
921 
922 	return "N/A";
923 }
924 
925 /**
926  * eeh_pe_bus_get - Retrieve PCI bus according to the given PE
927  * @pe: EEH PE
928  *
929  * Retrieve the PCI bus according to the given PE. Basically,
930  * there're 3 types of PEs: PHB/Bus/Device. For PHB PE, the
931  * primary PCI bus will be retrieved. The parent bus will be
932  * returned for BUS PE. However, we don't have associated PCI
933  * bus for DEVICE PE.
934  */
935 struct pci_bus *eeh_pe_bus_get(struct eeh_pe *pe)
936 {
937 	struct eeh_dev *edev;
938 	struct pci_dev *pdev;
939 
940 	if (pe->type & EEH_PE_PHB)
941 		return pe->phb->bus;
942 
943 	/* The primary bus might be cached during probe time */
944 	if (pe->state & EEH_PE_PRI_BUS)
945 		return pe->bus;
946 
947 	/* Retrieve the parent PCI bus of first (top) PCI device */
948 	edev = list_first_entry_or_null(&pe->edevs, struct eeh_dev, list);
949 	pdev = eeh_dev_to_pci_dev(edev);
950 	if (pdev)
951 		return pdev->bus;
952 
953 	return NULL;
954 }
955