xref: /openbmc/linux/arch/powerpc/kernel/eeh_pe.c (revision 6dfcd296)
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#%d\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 static struct eeh_pe *eeh_pe_next(struct eeh_pe *pe,
146 				  struct eeh_pe *root)
147 {
148 	struct list_head *next = pe->child_list.next;
149 
150 	if (next == &pe->child_list) {
151 		while (1) {
152 			if (pe == root)
153 				return NULL;
154 			next = pe->child.next;
155 			if (next != &pe->parent->child_list)
156 				break;
157 			pe = pe->parent;
158 		}
159 	}
160 
161 	return list_entry(next, struct eeh_pe, child);
162 }
163 
164 /**
165  * eeh_pe_traverse - Traverse PEs in the specified PHB
166  * @root: root PE
167  * @fn: callback
168  * @flag: extra parameter to callback
169  *
170  * The function is used to traverse the specified PE and its
171  * child PEs. The traversing is to be terminated once the
172  * callback returns something other than NULL, or no more PEs
173  * to be traversed.
174  */
175 void *eeh_pe_traverse(struct eeh_pe *root,
176 		      eeh_traverse_func fn, void *flag)
177 {
178 	struct eeh_pe *pe;
179 	void *ret;
180 
181 	for (pe = root; pe; pe = eeh_pe_next(pe, root)) {
182 		ret = fn(pe, flag);
183 		if (ret) return ret;
184 	}
185 
186 	return NULL;
187 }
188 
189 /**
190  * eeh_pe_dev_traverse - Traverse the devices from the PE
191  * @root: EEH PE
192  * @fn: function callback
193  * @flag: extra parameter to callback
194  *
195  * The function is used to traverse the devices of the specified
196  * PE and its child PEs.
197  */
198 void *eeh_pe_dev_traverse(struct eeh_pe *root,
199 		eeh_traverse_func fn, void *flag)
200 {
201 	struct eeh_pe *pe;
202 	struct eeh_dev *edev, *tmp;
203 	void *ret;
204 
205 	if (!root) {
206 		pr_warn("%s: Invalid PE %p\n",
207 			__func__, root);
208 		return NULL;
209 	}
210 
211 	/* Traverse root PE */
212 	for (pe = root; pe; pe = eeh_pe_next(pe, root)) {
213 		eeh_pe_for_each_dev(pe, edev, tmp) {
214 			ret = fn(edev, flag);
215 			if (ret)
216 				return ret;
217 		}
218 	}
219 
220 	return NULL;
221 }
222 
223 /**
224  * __eeh_pe_get - Check the PE address
225  * @data: EEH PE
226  * @flag: EEH device
227  *
228  * For one particular PE, it can be identified by PE address
229  * or tranditional BDF address. BDF address is composed of
230  * Bus/Device/Function number. The extra data referred by flag
231  * indicates which type of address should be used.
232  */
233 static void *__eeh_pe_get(void *data, void *flag)
234 {
235 	struct eeh_pe *pe = (struct eeh_pe *)data;
236 	struct eeh_dev *edev = (struct eeh_dev *)flag;
237 
238 	/* Unexpected PHB PE */
239 	if (pe->type & EEH_PE_PHB)
240 		return NULL;
241 
242 	/*
243 	 * We prefer PE address. For most cases, we should
244 	 * have non-zero PE address
245 	 */
246 	if (eeh_has_flag(EEH_VALID_PE_ZERO)) {
247 		if (edev->pe_config_addr == pe->addr)
248 			return pe;
249 	} else {
250 		if (edev->pe_config_addr &&
251 		    (edev->pe_config_addr == pe->addr))
252 			return pe;
253 	}
254 
255 	/* Try BDF address */
256 	if (edev->config_addr &&
257 	   (edev->config_addr == pe->config_addr))
258 		return pe;
259 
260 	return NULL;
261 }
262 
263 /**
264  * eeh_pe_get - Search PE based on the given address
265  * @edev: EEH device
266  *
267  * Search the corresponding PE based on the specified address which
268  * is included in the eeh device. The function is used to check if
269  * the associated PE has been created against the PE address. It's
270  * notable that the PE address has 2 format: traditional PE address
271  * which is composed of PCI bus/device/function number, or unified
272  * PE address.
273  */
274 struct eeh_pe *eeh_pe_get(struct eeh_dev *edev)
275 {
276 	struct eeh_pe *root = eeh_phb_pe_get(edev->phb);
277 	struct eeh_pe *pe;
278 
279 	pe = eeh_pe_traverse(root, __eeh_pe_get, edev);
280 
281 	return pe;
282 }
283 
284 /**
285  * eeh_pe_get_parent - Retrieve the parent PE
286  * @edev: EEH device
287  *
288  * The whole PEs existing in the system are organized as hierarchy
289  * tree. The function is used to retrieve the parent PE according
290  * to the parent EEH device.
291  */
292 static struct eeh_pe *eeh_pe_get_parent(struct eeh_dev *edev)
293 {
294 	struct eeh_dev *parent;
295 	struct pci_dn *pdn = eeh_dev_to_pdn(edev);
296 
297 	/*
298 	 * It might have the case for the indirect parent
299 	 * EEH device already having associated PE, but
300 	 * the direct parent EEH device doesn't have yet.
301 	 */
302 	if (edev->physfn)
303 		pdn = pci_get_pdn(edev->physfn);
304 	else
305 		pdn = pdn ? pdn->parent : NULL;
306 	while (pdn) {
307 		/* We're poking out of PCI territory */
308 		parent = pdn_to_eeh_dev(pdn);
309 		if (!parent)
310 			return NULL;
311 
312 		if (parent->pe)
313 			return parent->pe;
314 
315 		pdn = pdn->parent;
316 	}
317 
318 	return NULL;
319 }
320 
321 /**
322  * eeh_add_to_parent_pe - Add EEH device to parent PE
323  * @edev: EEH device
324  *
325  * Add EEH device to the parent PE. If the parent PE already
326  * exists, the PE type will be changed to EEH_PE_BUS. Otherwise,
327  * we have to create new PE to hold the EEH device and the new
328  * PE will be linked to its parent PE as well.
329  */
330 int eeh_add_to_parent_pe(struct eeh_dev *edev)
331 {
332 	struct eeh_pe *pe, *parent;
333 
334 	/* Check if the PE number is valid */
335 	if (!eeh_has_flag(EEH_VALID_PE_ZERO) && !edev->pe_config_addr) {
336 		pr_err("%s: Invalid PE#0 for edev 0x%x on PHB#%d\n",
337 		       __func__, edev->config_addr, edev->phb->global_number);
338 		return -EINVAL;
339 	}
340 
341 	/*
342 	 * Search the PE has been existing or not according
343 	 * to the PE address. If that has been existing, the
344 	 * PE should be composed of PCI bus and its subordinate
345 	 * components.
346 	 */
347 	pe = eeh_pe_get(edev);
348 	if (pe && !(pe->type & EEH_PE_INVALID)) {
349 		/* Mark the PE as type of PCI bus */
350 		pe->type = EEH_PE_BUS;
351 		edev->pe = pe;
352 
353 		/* Put the edev to PE */
354 		list_add_tail(&edev->list, &pe->edevs);
355 		pr_debug("EEH: Add %04x:%02x:%02x.%01x to Bus PE#%x\n",
356 			edev->phb->global_number,
357 			edev->config_addr >> 8,
358 			PCI_SLOT(edev->config_addr & 0xFF),
359 			PCI_FUNC(edev->config_addr & 0xFF),
360 			pe->addr);
361 		return 0;
362 	} else if (pe && (pe->type & EEH_PE_INVALID)) {
363 		list_add_tail(&edev->list, &pe->edevs);
364 		edev->pe = pe;
365 		/*
366 		 * We're running to here because of PCI hotplug caused by
367 		 * EEH recovery. We need clear EEH_PE_INVALID until the top.
368 		 */
369 		parent = pe;
370 		while (parent) {
371 			if (!(parent->type & EEH_PE_INVALID))
372 				break;
373 			parent->type &= ~(EEH_PE_INVALID | EEH_PE_KEEP);
374 			parent = parent->parent;
375 		}
376 
377 		pr_debug("EEH: Add %04x:%02x:%02x.%01x to Device "
378 			 "PE#%x, Parent PE#%x\n",
379 			edev->phb->global_number,
380 			edev->config_addr >> 8,
381                         PCI_SLOT(edev->config_addr & 0xFF),
382                         PCI_FUNC(edev->config_addr & 0xFF),
383 			pe->addr, pe->parent->addr);
384 		return 0;
385 	}
386 
387 	/* Create a new EEH PE */
388 	if (edev->physfn)
389 		pe = eeh_pe_alloc(edev->phb, EEH_PE_VF);
390 	else
391 		pe = eeh_pe_alloc(edev->phb, EEH_PE_DEVICE);
392 	if (!pe) {
393 		pr_err("%s: out of memory!\n", __func__);
394 		return -ENOMEM;
395 	}
396 	pe->addr	= edev->pe_config_addr;
397 	pe->config_addr	= edev->config_addr;
398 
399 	/*
400 	 * Put the new EEH PE into hierarchy tree. If the parent
401 	 * can't be found, the newly created PE will be attached
402 	 * to PHB directly. Otherwise, we have to associate the
403 	 * PE with its parent.
404 	 */
405 	parent = eeh_pe_get_parent(edev);
406 	if (!parent) {
407 		parent = eeh_phb_pe_get(edev->phb);
408 		if (!parent) {
409 			pr_err("%s: No PHB PE is found (PHB Domain=%d)\n",
410 				__func__, edev->phb->global_number);
411 			edev->pe = NULL;
412 			kfree(pe);
413 			return -EEXIST;
414 		}
415 	}
416 	pe->parent = parent;
417 
418 	/*
419 	 * Put the newly created PE into the child list and
420 	 * link the EEH device accordingly.
421 	 */
422 	list_add_tail(&pe->child, &parent->child_list);
423 	list_add_tail(&edev->list, &pe->edevs);
424 	edev->pe = pe;
425 	pr_debug("EEH: Add %04x:%02x:%02x.%01x to "
426 		 "Device PE#%x, Parent PE#%x\n",
427 		 edev->phb->global_number,
428 		 edev->config_addr >> 8,
429 		 PCI_SLOT(edev->config_addr & 0xFF),
430 		 PCI_FUNC(edev->config_addr & 0xFF),
431 		 pe->addr, pe->parent->addr);
432 
433 	return 0;
434 }
435 
436 /**
437  * eeh_rmv_from_parent_pe - Remove one EEH device from the associated PE
438  * @edev: EEH device
439  *
440  * The PE hierarchy tree might be changed when doing PCI hotplug.
441  * Also, the PCI devices or buses could be removed from the system
442  * during EEH recovery. So we have to call the function remove the
443  * corresponding PE accordingly if necessary.
444  */
445 int eeh_rmv_from_parent_pe(struct eeh_dev *edev)
446 {
447 	struct eeh_pe *pe, *parent, *child;
448 	int cnt;
449 
450 	if (!edev->pe) {
451 		pr_debug("%s: No PE found for device %04x:%02x:%02x.%01x\n",
452 			 __func__,  edev->phb->global_number,
453 			 edev->config_addr >> 8,
454 			 PCI_SLOT(edev->config_addr & 0xFF),
455 			 PCI_FUNC(edev->config_addr & 0xFF));
456 		return -EEXIST;
457 	}
458 
459 	/* Remove the EEH device */
460 	pe = eeh_dev_to_pe(edev);
461 	edev->pe = NULL;
462 	list_del(&edev->list);
463 
464 	/*
465 	 * Check if the parent PE includes any EEH devices.
466 	 * If not, we should delete that. Also, we should
467 	 * delete the parent PE if it doesn't have associated
468 	 * child PEs and EEH devices.
469 	 */
470 	while (1) {
471 		parent = pe->parent;
472 		if (pe->type & EEH_PE_PHB)
473 			break;
474 
475 		if (!(pe->state & EEH_PE_KEEP)) {
476 			if (list_empty(&pe->edevs) &&
477 			    list_empty(&pe->child_list)) {
478 				list_del(&pe->child);
479 				kfree(pe);
480 			} else {
481 				break;
482 			}
483 		} else {
484 			if (list_empty(&pe->edevs)) {
485 				cnt = 0;
486 				list_for_each_entry(child, &pe->child_list, child) {
487 					if (!(child->type & EEH_PE_INVALID)) {
488 						cnt++;
489 						break;
490 					}
491 				}
492 
493 				if (!cnt)
494 					pe->type |= EEH_PE_INVALID;
495 				else
496 					break;
497 			}
498 		}
499 
500 		pe = parent;
501 	}
502 
503 	return 0;
504 }
505 
506 /**
507  * eeh_pe_update_time_stamp - Update PE's frozen time stamp
508  * @pe: EEH PE
509  *
510  * We have time stamp for each PE to trace its time of getting
511  * frozen in last hour. The function should be called to update
512  * the time stamp on first error of the specific PE. On the other
513  * handle, we needn't account for errors happened in last hour.
514  */
515 void eeh_pe_update_time_stamp(struct eeh_pe *pe)
516 {
517 	struct timeval tstamp;
518 
519 	if (!pe) return;
520 
521 	if (pe->freeze_count <= 0) {
522 		pe->freeze_count = 0;
523 		do_gettimeofday(&pe->tstamp);
524 	} else {
525 		do_gettimeofday(&tstamp);
526 		if (tstamp.tv_sec - pe->tstamp.tv_sec > 3600) {
527 			pe->tstamp = tstamp;
528 			pe->freeze_count = 0;
529 		}
530 	}
531 }
532 
533 /**
534  * __eeh_pe_state_mark - Mark the state for the PE
535  * @data: EEH PE
536  * @flag: state
537  *
538  * The function is used to mark the indicated state for the given
539  * PE. Also, the associated PCI devices will be put into IO frozen
540  * state as well.
541  */
542 static void *__eeh_pe_state_mark(void *data, void *flag)
543 {
544 	struct eeh_pe *pe = (struct eeh_pe *)data;
545 	int state = *((int *)flag);
546 	struct eeh_dev *edev, *tmp;
547 	struct pci_dev *pdev;
548 
549 	/* Keep the state of permanently removed PE intact */
550 	if (pe->state & EEH_PE_REMOVED)
551 		return NULL;
552 
553 	pe->state |= state;
554 
555 	/* Offline PCI devices if applicable */
556 	if (!(state & EEH_PE_ISOLATED))
557 		return NULL;
558 
559 	eeh_pe_for_each_dev(pe, edev, tmp) {
560 		pdev = eeh_dev_to_pci_dev(edev);
561 		if (pdev)
562 			pdev->error_state = pci_channel_io_frozen;
563 	}
564 
565 	/* Block PCI config access if required */
566 	if (pe->state & EEH_PE_CFG_RESTRICTED)
567 		pe->state |= EEH_PE_CFG_BLOCKED;
568 
569 	return NULL;
570 }
571 
572 /**
573  * eeh_pe_state_mark - Mark specified state for PE and its associated device
574  * @pe: EEH PE
575  *
576  * EEH error affects the current PE and its child PEs. The function
577  * is used to mark appropriate state for the affected PEs and the
578  * associated devices.
579  */
580 void eeh_pe_state_mark(struct eeh_pe *pe, int state)
581 {
582 	eeh_pe_traverse(pe, __eeh_pe_state_mark, &state);
583 }
584 EXPORT_SYMBOL_GPL(eeh_pe_state_mark);
585 
586 static void *__eeh_pe_dev_mode_mark(void *data, void *flag)
587 {
588 	struct eeh_dev *edev = data;
589 	int mode = *((int *)flag);
590 
591 	edev->mode |= mode;
592 
593 	return NULL;
594 }
595 
596 /**
597  * eeh_pe_dev_state_mark - Mark state for all device under the PE
598  * @pe: EEH PE
599  *
600  * Mark specific state for all child devices of the PE.
601  */
602 void eeh_pe_dev_mode_mark(struct eeh_pe *pe, int mode)
603 {
604 	eeh_pe_dev_traverse(pe, __eeh_pe_dev_mode_mark, &mode);
605 }
606 
607 /**
608  * __eeh_pe_state_clear - Clear state for the PE
609  * @data: EEH PE
610  * @flag: state
611  *
612  * The function is used to clear the indicated state from the
613  * given PE. Besides, we also clear the check count of the PE
614  * as well.
615  */
616 static void *__eeh_pe_state_clear(void *data, void *flag)
617 {
618 	struct eeh_pe *pe = (struct eeh_pe *)data;
619 	int state = *((int *)flag);
620 	struct eeh_dev *edev, *tmp;
621 	struct pci_dev *pdev;
622 
623 	/* Keep the state of permanently removed PE intact */
624 	if (pe->state & EEH_PE_REMOVED)
625 		return NULL;
626 
627 	pe->state &= ~state;
628 
629 	/*
630 	 * Special treatment on clearing isolated state. Clear
631 	 * check count since last isolation and put all affected
632 	 * devices to normal state.
633 	 */
634 	if (!(state & EEH_PE_ISOLATED))
635 		return NULL;
636 
637 	pe->check_count = 0;
638 	eeh_pe_for_each_dev(pe, edev, tmp) {
639 		pdev = eeh_dev_to_pci_dev(edev);
640 		if (!pdev)
641 			continue;
642 
643 		pdev->error_state = pci_channel_io_normal;
644 	}
645 
646 	/* Unblock PCI config access if required */
647 	if (pe->state & EEH_PE_CFG_RESTRICTED)
648 		pe->state &= ~EEH_PE_CFG_BLOCKED;
649 
650 	return NULL;
651 }
652 
653 /**
654  * eeh_pe_state_clear - Clear state for the PE and its children
655  * @pe: PE
656  * @state: state to be cleared
657  *
658  * When the PE and its children has been recovered from error,
659  * we need clear the error state for that. The function is used
660  * for the purpose.
661  */
662 void eeh_pe_state_clear(struct eeh_pe *pe, int state)
663 {
664 	eeh_pe_traverse(pe, __eeh_pe_state_clear, &state);
665 }
666 
667 /**
668  * eeh_pe_state_mark_with_cfg - Mark PE state with unblocked config space
669  * @pe: PE
670  * @state: PE state to be set
671  *
672  * Set specified flag to PE and its child PEs. The PCI config space
673  * of some PEs is blocked automatically when EEH_PE_ISOLATED is set,
674  * which isn't needed in some situations. The function allows to set
675  * the specified flag to indicated PEs without blocking their PCI
676  * config space.
677  */
678 void eeh_pe_state_mark_with_cfg(struct eeh_pe *pe, int state)
679 {
680 	eeh_pe_traverse(pe, __eeh_pe_state_mark, &state);
681 	if (!(state & EEH_PE_ISOLATED))
682 		return;
683 
684 	/* Clear EEH_PE_CFG_BLOCKED, which might be set just now */
685 	state = EEH_PE_CFG_BLOCKED;
686 	eeh_pe_traverse(pe, __eeh_pe_state_clear, &state);
687 }
688 
689 /*
690  * Some PCI bridges (e.g. PLX bridges) have primary/secondary
691  * buses assigned explicitly by firmware, and we probably have
692  * lost that after reset. So we have to delay the check until
693  * the PCI-CFG registers have been restored for the parent
694  * bridge.
695  *
696  * Don't use normal PCI-CFG accessors, which probably has been
697  * blocked on normal path during the stage. So we need utilize
698  * eeh operations, which is always permitted.
699  */
700 static void eeh_bridge_check_link(struct eeh_dev *edev)
701 {
702 	struct pci_dn *pdn = eeh_dev_to_pdn(edev);
703 	int cap;
704 	uint32_t val;
705 	int timeout = 0;
706 
707 	/*
708 	 * We only check root port and downstream ports of
709 	 * PCIe switches
710 	 */
711 	if (!(edev->mode & (EEH_DEV_ROOT_PORT | EEH_DEV_DS_PORT)))
712 		return;
713 
714 	pr_debug("%s: Check PCIe link for %04x:%02x:%02x.%01x ...\n",
715 		 __func__, edev->phb->global_number,
716 		 edev->config_addr >> 8,
717 		 PCI_SLOT(edev->config_addr & 0xFF),
718 		 PCI_FUNC(edev->config_addr & 0xFF));
719 
720 	/* Check slot status */
721 	cap = edev->pcie_cap;
722 	eeh_ops->read_config(pdn, cap + PCI_EXP_SLTSTA, 2, &val);
723 	if (!(val & PCI_EXP_SLTSTA_PDS)) {
724 		pr_debug("  No card in the slot (0x%04x) !\n", val);
725 		return;
726 	}
727 
728 	/* Check power status if we have the capability */
729 	eeh_ops->read_config(pdn, cap + PCI_EXP_SLTCAP, 2, &val);
730 	if (val & PCI_EXP_SLTCAP_PCP) {
731 		eeh_ops->read_config(pdn, cap + PCI_EXP_SLTCTL, 2, &val);
732 		if (val & PCI_EXP_SLTCTL_PCC) {
733 			pr_debug("  In power-off state, power it on ...\n");
734 			val &= ~(PCI_EXP_SLTCTL_PCC | PCI_EXP_SLTCTL_PIC);
735 			val |= (0x0100 & PCI_EXP_SLTCTL_PIC);
736 			eeh_ops->write_config(pdn, cap + PCI_EXP_SLTCTL, 2, val);
737 			msleep(2 * 1000);
738 		}
739 	}
740 
741 	/* Enable link */
742 	eeh_ops->read_config(pdn, cap + PCI_EXP_LNKCTL, 2, &val);
743 	val &= ~PCI_EXP_LNKCTL_LD;
744 	eeh_ops->write_config(pdn, cap + PCI_EXP_LNKCTL, 2, val);
745 
746 	/* Check link */
747 	eeh_ops->read_config(pdn, cap + PCI_EXP_LNKCAP, 4, &val);
748 	if (!(val & PCI_EXP_LNKCAP_DLLLARC)) {
749 		pr_debug("  No link reporting capability (0x%08x) \n", val);
750 		msleep(1000);
751 		return;
752 	}
753 
754 	/* Wait the link is up until timeout (5s) */
755 	timeout = 0;
756 	while (timeout < 5000) {
757 		msleep(20);
758 		timeout += 20;
759 
760 		eeh_ops->read_config(pdn, cap + PCI_EXP_LNKSTA, 2, &val);
761 		if (val & PCI_EXP_LNKSTA_DLLLA)
762 			break;
763 	}
764 
765 	if (val & PCI_EXP_LNKSTA_DLLLA)
766 		pr_debug("  Link up (%s)\n",
767 			 (val & PCI_EXP_LNKSTA_CLS_2_5GB) ? "2.5GB" : "5GB");
768 	else
769 		pr_debug("  Link not ready (0x%04x)\n", val);
770 }
771 
772 #define BYTE_SWAP(OFF)	(8*((OFF)/4)+3-(OFF))
773 #define SAVED_BYTE(OFF)	(((u8 *)(edev->config_space))[BYTE_SWAP(OFF)])
774 
775 static void eeh_restore_bridge_bars(struct eeh_dev *edev)
776 {
777 	struct pci_dn *pdn = eeh_dev_to_pdn(edev);
778 	int i;
779 
780 	/*
781 	 * Device BARs: 0x10 - 0x18
782 	 * Bus numbers and windows: 0x18 - 0x30
783 	 */
784 	for (i = 4; i < 13; i++)
785 		eeh_ops->write_config(pdn, i*4, 4, edev->config_space[i]);
786 	/* Rom: 0x38 */
787 	eeh_ops->write_config(pdn, 14*4, 4, edev->config_space[14]);
788 
789 	/* Cache line & Latency timer: 0xC 0xD */
790 	eeh_ops->write_config(pdn, PCI_CACHE_LINE_SIZE, 1,
791                 SAVED_BYTE(PCI_CACHE_LINE_SIZE));
792         eeh_ops->write_config(pdn, PCI_LATENCY_TIMER, 1,
793                 SAVED_BYTE(PCI_LATENCY_TIMER));
794 	/* Max latency, min grant, interrupt ping and line: 0x3C */
795 	eeh_ops->write_config(pdn, 15*4, 4, edev->config_space[15]);
796 
797 	/* PCI Command: 0x4 */
798 	eeh_ops->write_config(pdn, PCI_COMMAND, 4, edev->config_space[1]);
799 
800 	/* Check the PCIe link is ready */
801 	eeh_bridge_check_link(edev);
802 }
803 
804 static void eeh_restore_device_bars(struct eeh_dev *edev)
805 {
806 	struct pci_dn *pdn = eeh_dev_to_pdn(edev);
807 	int i;
808 	u32 cmd;
809 
810 	for (i = 4; i < 10; i++)
811 		eeh_ops->write_config(pdn, i*4, 4, edev->config_space[i]);
812 	/* 12 == Expansion ROM Address */
813 	eeh_ops->write_config(pdn, 12*4, 4, edev->config_space[12]);
814 
815 	eeh_ops->write_config(pdn, PCI_CACHE_LINE_SIZE, 1,
816 		SAVED_BYTE(PCI_CACHE_LINE_SIZE));
817 	eeh_ops->write_config(pdn, PCI_LATENCY_TIMER, 1,
818 		SAVED_BYTE(PCI_LATENCY_TIMER));
819 
820 	/* max latency, min grant, interrupt pin and line */
821 	eeh_ops->write_config(pdn, 15*4, 4, edev->config_space[15]);
822 
823 	/*
824 	 * Restore PERR & SERR bits, some devices require it,
825 	 * don't touch the other command bits
826 	 */
827 	eeh_ops->read_config(pdn, PCI_COMMAND, 4, &cmd);
828 	if (edev->config_space[1] & PCI_COMMAND_PARITY)
829 		cmd |= PCI_COMMAND_PARITY;
830 	else
831 		cmd &= ~PCI_COMMAND_PARITY;
832 	if (edev->config_space[1] & PCI_COMMAND_SERR)
833 		cmd |= PCI_COMMAND_SERR;
834 	else
835 		cmd &= ~PCI_COMMAND_SERR;
836 	eeh_ops->write_config(pdn, PCI_COMMAND, 4, cmd);
837 }
838 
839 /**
840  * eeh_restore_one_device_bars - Restore the Base Address Registers for one device
841  * @data: EEH device
842  * @flag: Unused
843  *
844  * Loads the PCI configuration space base address registers,
845  * the expansion ROM base address, the latency timer, and etc.
846  * from the saved values in the device node.
847  */
848 static void *eeh_restore_one_device_bars(void *data, void *flag)
849 {
850 	struct eeh_dev *edev = (struct eeh_dev *)data;
851 	struct pci_dn *pdn = eeh_dev_to_pdn(edev);
852 
853 	/* Do special restore for bridges */
854 	if (edev->mode & EEH_DEV_BRIDGE)
855 		eeh_restore_bridge_bars(edev);
856 	else
857 		eeh_restore_device_bars(edev);
858 
859 	if (eeh_ops->restore_config && pdn)
860 		eeh_ops->restore_config(pdn);
861 
862 	return NULL;
863 }
864 
865 /**
866  * eeh_pe_restore_bars - Restore the PCI config space info
867  * @pe: EEH PE
868  *
869  * This routine performs a recursive walk to the children
870  * of this device as well.
871  */
872 void eeh_pe_restore_bars(struct eeh_pe *pe)
873 {
874 	/*
875 	 * We needn't take the EEH lock since eeh_pe_dev_traverse()
876 	 * will take that.
877 	 */
878 	eeh_pe_dev_traverse(pe, eeh_restore_one_device_bars, NULL);
879 }
880 
881 /**
882  * eeh_pe_loc_get - Retrieve location code binding to the given PE
883  * @pe: EEH PE
884  *
885  * Retrieve the location code of the given PE. If the primary PE bus
886  * is root bus, we will grab location code from PHB device tree node
887  * or root port. Otherwise, the upstream bridge's device tree node
888  * of the primary PE bus will be checked for the location code.
889  */
890 const char *eeh_pe_loc_get(struct eeh_pe *pe)
891 {
892 	struct pci_bus *bus = eeh_pe_bus_get(pe);
893 	struct device_node *dn;
894 	const char *loc = NULL;
895 
896 	while (bus) {
897 		dn = pci_bus_to_OF_node(bus);
898 		if (!dn) {
899 			bus = bus->parent;
900 			continue;
901 		}
902 
903 		if (pci_is_root_bus(bus))
904 			loc = of_get_property(dn, "ibm,io-base-loc-code", NULL);
905 		else
906 			loc = of_get_property(dn, "ibm,slot-location-code",
907 					      NULL);
908 
909 		if (loc)
910 			return loc;
911 
912 		bus = bus->parent;
913 	}
914 
915 	return "N/A";
916 }
917 
918 /**
919  * eeh_pe_bus_get - Retrieve PCI bus according to the given PE
920  * @pe: EEH PE
921  *
922  * Retrieve the PCI bus according to the given PE. Basically,
923  * there're 3 types of PEs: PHB/Bus/Device. For PHB PE, the
924  * primary PCI bus will be retrieved. The parent bus will be
925  * returned for BUS PE. However, we don't have associated PCI
926  * bus for DEVICE PE.
927  */
928 struct pci_bus *eeh_pe_bus_get(struct eeh_pe *pe)
929 {
930 	struct eeh_dev *edev;
931 	struct pci_dev *pdev;
932 
933 	if (pe->type & EEH_PE_PHB)
934 		return pe->phb->bus;
935 
936 	/* The primary bus might be cached during probe time */
937 	if (pe->state & EEH_PE_PRI_BUS)
938 		return pe->bus;
939 
940 	/* Retrieve the parent PCI bus of first (top) PCI device */
941 	edev = list_first_entry_or_null(&pe->edevs, struct eeh_dev, list);
942 	pdev = eeh_dev_to_pci_dev(edev);
943 	if (pdev)
944 		return pdev->bus;
945 
946 	return NULL;
947 }
948