xref: /openbmc/linux/drivers/pci/hotplug/pnv_php.c (revision bc05aa6e)
1 // SPDX-License-Identifier: GPL-2.0+
2 /*
3  * PCI Hotplug Driver for PowerPC PowerNV platform.
4  *
5  * Copyright Gavin Shan, IBM Corporation 2016.
6  */
7 
8 #include <linux/libfdt.h>
9 #include <linux/module.h>
10 #include <linux/pci.h>
11 #include <linux/pci_hotplug.h>
12 
13 #include <asm/opal.h>
14 #include <asm/pnv-pci.h>
15 #include <asm/ppc-pci.h>
16 
17 #define DRIVER_VERSION	"0.1"
18 #define DRIVER_AUTHOR	"Gavin Shan, IBM Corporation"
19 #define DRIVER_DESC	"PowerPC PowerNV PCI Hotplug Driver"
20 
21 struct pnv_php_event {
22 	bool			added;
23 	struct pnv_php_slot	*php_slot;
24 	struct work_struct	work;
25 };
26 
27 static LIST_HEAD(pnv_php_slot_list);
28 static DEFINE_SPINLOCK(pnv_php_lock);
29 
30 static void pnv_php_register(struct device_node *dn);
31 static void pnv_php_unregister_one(struct device_node *dn);
32 static void pnv_php_unregister(struct device_node *dn);
33 
34 static void pnv_php_disable_irq(struct pnv_php_slot *php_slot,
35 				bool disable_device)
36 {
37 	struct pci_dev *pdev = php_slot->pdev;
38 	int irq = php_slot->irq;
39 	u16 ctrl;
40 
41 	if (php_slot->irq > 0) {
42 		pcie_capability_read_word(pdev, PCI_EXP_SLTCTL, &ctrl);
43 		ctrl &= ~(PCI_EXP_SLTCTL_HPIE |
44 			  PCI_EXP_SLTCTL_PDCE |
45 			  PCI_EXP_SLTCTL_DLLSCE);
46 		pcie_capability_write_word(pdev, PCI_EXP_SLTCTL, ctrl);
47 
48 		free_irq(php_slot->irq, php_slot);
49 		php_slot->irq = 0;
50 	}
51 
52 	if (php_slot->wq) {
53 		destroy_workqueue(php_slot->wq);
54 		php_slot->wq = NULL;
55 	}
56 
57 	if (disable_device || irq > 0) {
58 		if (pdev->msix_enabled)
59 			pci_disable_msix(pdev);
60 		else if (pdev->msi_enabled)
61 			pci_disable_msi(pdev);
62 
63 		pci_disable_device(pdev);
64 	}
65 }
66 
67 static void pnv_php_free_slot(struct kref *kref)
68 {
69 	struct pnv_php_slot *php_slot = container_of(kref,
70 					struct pnv_php_slot, kref);
71 
72 	WARN_ON(!list_empty(&php_slot->children));
73 	pnv_php_disable_irq(php_slot, false);
74 	kfree(php_slot->name);
75 	kfree(php_slot);
76 }
77 
78 static inline void pnv_php_put_slot(struct pnv_php_slot *php_slot)
79 {
80 
81 	if (!php_slot)
82 		return;
83 
84 	kref_put(&php_slot->kref, pnv_php_free_slot);
85 }
86 
87 static struct pnv_php_slot *pnv_php_match(struct device_node *dn,
88 					  struct pnv_php_slot *php_slot)
89 {
90 	struct pnv_php_slot *target, *tmp;
91 
92 	if (php_slot->dn == dn) {
93 		kref_get(&php_slot->kref);
94 		return php_slot;
95 	}
96 
97 	list_for_each_entry(tmp, &php_slot->children, link) {
98 		target = pnv_php_match(dn, tmp);
99 		if (target)
100 			return target;
101 	}
102 
103 	return NULL;
104 }
105 
106 struct pnv_php_slot *pnv_php_find_slot(struct device_node *dn)
107 {
108 	struct pnv_php_slot *php_slot, *tmp;
109 	unsigned long flags;
110 
111 	spin_lock_irqsave(&pnv_php_lock, flags);
112 	list_for_each_entry(tmp, &pnv_php_slot_list, link) {
113 		php_slot = pnv_php_match(dn, tmp);
114 		if (php_slot) {
115 			spin_unlock_irqrestore(&pnv_php_lock, flags);
116 			return php_slot;
117 		}
118 	}
119 	spin_unlock_irqrestore(&pnv_php_lock, flags);
120 
121 	return NULL;
122 }
123 EXPORT_SYMBOL_GPL(pnv_php_find_slot);
124 
125 /*
126  * Remove pdn for all children of the indicated device node.
127  * The function should remove pdn in a depth-first manner.
128  */
129 static void pnv_php_rmv_pdns(struct device_node *dn)
130 {
131 	struct device_node *child;
132 
133 	for_each_child_of_node(dn, child) {
134 		pnv_php_rmv_pdns(child);
135 
136 		pci_remove_device_node_info(child);
137 	}
138 }
139 
140 /*
141  * Detach all child nodes of the indicated device nodes. The
142  * function should handle device nodes in depth-first manner.
143  *
144  * We should not invoke of_node_release() as the memory for
145  * individual device node is part of large memory block. The
146  * large block is allocated from memblock (system bootup) or
147  * kmalloc() when unflattening the device tree by OF changeset.
148  * We can not free the large block allocated from memblock. For
149  * later case, it should be released at once.
150  */
151 static void pnv_php_detach_device_nodes(struct device_node *parent)
152 {
153 	struct device_node *dn;
154 	int refcount;
155 
156 	for_each_child_of_node(parent, dn) {
157 		pnv_php_detach_device_nodes(dn);
158 
159 		of_node_put(dn);
160 		refcount = kref_read(&dn->kobj.kref);
161 		if (refcount != 1)
162 			pr_warn("Invalid refcount %d on <%pOF>\n",
163 				refcount, dn);
164 
165 		of_detach_node(dn);
166 	}
167 }
168 
169 static void pnv_php_rmv_devtree(struct pnv_php_slot *php_slot)
170 {
171 	pnv_php_rmv_pdns(php_slot->dn);
172 
173 	/*
174 	 * Decrease the refcount if the device nodes were created
175 	 * through OF changeset before detaching them.
176 	 */
177 	if (php_slot->fdt)
178 		of_changeset_destroy(&php_slot->ocs);
179 	pnv_php_detach_device_nodes(php_slot->dn);
180 
181 	if (php_slot->fdt) {
182 		kfree(php_slot->dt);
183 		kfree(php_slot->fdt);
184 		php_slot->dt        = NULL;
185 		php_slot->dn->child = NULL;
186 		php_slot->fdt       = NULL;
187 	}
188 }
189 
190 /*
191  * As the nodes in OF changeset are applied in reverse order, we
192  * need revert the nodes in advance so that we have correct node
193  * order after the changeset is applied.
194  */
195 static void pnv_php_reverse_nodes(struct device_node *parent)
196 {
197 	struct device_node *child, *next;
198 
199 	/* In-depth first */
200 	for_each_child_of_node(parent, child)
201 		pnv_php_reverse_nodes(child);
202 
203 	/* Reverse the nodes in the child list */
204 	child = parent->child;
205 	parent->child = NULL;
206 	while (child) {
207 		next = child->sibling;
208 
209 		child->sibling = parent->child;
210 		parent->child = child;
211 		child = next;
212 	}
213 }
214 
215 static int pnv_php_populate_changeset(struct of_changeset *ocs,
216 				      struct device_node *dn)
217 {
218 	struct device_node *child;
219 	int ret = 0;
220 
221 	for_each_child_of_node(dn, child) {
222 		ret = of_changeset_attach_node(ocs, child);
223 		if (ret)
224 			break;
225 
226 		ret = pnv_php_populate_changeset(ocs, child);
227 		if (ret)
228 			break;
229 	}
230 
231 	return ret;
232 }
233 
234 static void *pnv_php_add_one_pdn(struct device_node *dn, void *data)
235 {
236 	struct pci_controller *hose = (struct pci_controller *)data;
237 	struct pci_dn *pdn;
238 
239 	pdn = pci_add_device_node_info(hose, dn);
240 	if (!pdn)
241 		return ERR_PTR(-ENOMEM);
242 
243 	return NULL;
244 }
245 
246 static void pnv_php_add_pdns(struct pnv_php_slot *slot)
247 {
248 	struct pci_controller *hose = pci_bus_to_host(slot->bus);
249 
250 	pci_traverse_device_nodes(slot->dn, pnv_php_add_one_pdn, hose);
251 }
252 
253 static int pnv_php_add_devtree(struct pnv_php_slot *php_slot)
254 {
255 	void *fdt, *fdt1, *dt;
256 	int ret;
257 
258 	/* We don't know the FDT blob size. We try to get it through
259 	 * maximal memory chunk and then copy it to another chunk that
260 	 * fits the real size.
261 	 */
262 	fdt1 = kzalloc(0x10000, GFP_KERNEL);
263 	if (!fdt1) {
264 		ret = -ENOMEM;
265 		goto out;
266 	}
267 
268 	ret = pnv_pci_get_device_tree(php_slot->dn->phandle, fdt1, 0x10000);
269 	if (ret) {
270 		pci_warn(php_slot->pdev, "Error %d getting FDT blob\n", ret);
271 		goto free_fdt1;
272 	}
273 
274 	fdt = kzalloc(fdt_totalsize(fdt1), GFP_KERNEL);
275 	if (!fdt) {
276 		ret = -ENOMEM;
277 		goto free_fdt1;
278 	}
279 
280 	/* Unflatten device tree blob */
281 	memcpy(fdt, fdt1, fdt_totalsize(fdt1));
282 	dt = of_fdt_unflatten_tree(fdt, php_slot->dn, NULL);
283 	if (!dt) {
284 		ret = -EINVAL;
285 		pci_warn(php_slot->pdev, "Cannot unflatten FDT\n");
286 		goto free_fdt;
287 	}
288 
289 	/* Initialize and apply the changeset */
290 	of_changeset_init(&php_slot->ocs);
291 	pnv_php_reverse_nodes(php_slot->dn);
292 	ret = pnv_php_populate_changeset(&php_slot->ocs, php_slot->dn);
293 	if (ret) {
294 		pnv_php_reverse_nodes(php_slot->dn);
295 		pci_warn(php_slot->pdev, "Error %d populating changeset\n",
296 			 ret);
297 		goto free_dt;
298 	}
299 
300 	php_slot->dn->child = NULL;
301 	ret = of_changeset_apply(&php_slot->ocs);
302 	if (ret) {
303 		pci_warn(php_slot->pdev, "Error %d applying changeset\n", ret);
304 		goto destroy_changeset;
305 	}
306 
307 	/* Add device node firmware data */
308 	pnv_php_add_pdns(php_slot);
309 	php_slot->fdt = fdt;
310 	php_slot->dt  = dt;
311 	kfree(fdt1);
312 	goto out;
313 
314 destroy_changeset:
315 	of_changeset_destroy(&php_slot->ocs);
316 free_dt:
317 	kfree(dt);
318 	php_slot->dn->child = NULL;
319 free_fdt:
320 	kfree(fdt);
321 free_fdt1:
322 	kfree(fdt1);
323 out:
324 	return ret;
325 }
326 
327 int pnv_php_set_slot_power_state(struct hotplug_slot *slot,
328 				 uint8_t state)
329 {
330 	struct pnv_php_slot *php_slot = slot->private;
331 	struct opal_msg msg;
332 	int ret;
333 
334 	ret = pnv_pci_set_power_state(php_slot->id, state, &msg);
335 	if (ret > 0) {
336 		if (be64_to_cpu(msg.params[1]) != php_slot->dn->phandle	||
337 		    be64_to_cpu(msg.params[2]) != state			||
338 		    be64_to_cpu(msg.params[3]) != OPAL_SUCCESS) {
339 			pci_warn(php_slot->pdev, "Wrong msg (%lld, %lld, %lld)\n",
340 				 be64_to_cpu(msg.params[1]),
341 				 be64_to_cpu(msg.params[2]),
342 				 be64_to_cpu(msg.params[3]));
343 			return -ENOMSG;
344 		}
345 	} else if (ret < 0) {
346 		pci_warn(php_slot->pdev, "Error %d powering %s\n",
347 			 ret, (state == OPAL_PCI_SLOT_POWER_ON) ? "on" : "off");
348 		return ret;
349 	}
350 
351 	if (state == OPAL_PCI_SLOT_POWER_OFF || state == OPAL_PCI_SLOT_OFFLINE)
352 		pnv_php_rmv_devtree(php_slot);
353 	else
354 		ret = pnv_php_add_devtree(php_slot);
355 
356 	return ret;
357 }
358 EXPORT_SYMBOL_GPL(pnv_php_set_slot_power_state);
359 
360 static int pnv_php_get_power_state(struct hotplug_slot *slot, u8 *state)
361 {
362 	struct pnv_php_slot *php_slot = slot->private;
363 	uint8_t power_state = OPAL_PCI_SLOT_POWER_ON;
364 	int ret;
365 
366 	/*
367 	 * Retrieve power status from firmware. If we fail
368 	 * getting that, the power status fails back to
369 	 * be on.
370 	 */
371 	ret = pnv_pci_get_power_state(php_slot->id, &power_state);
372 	if (ret) {
373 		pci_warn(php_slot->pdev, "Error %d getting power status\n",
374 			 ret);
375 	} else {
376 		*state = power_state;
377 		slot->info->power_status = power_state;
378 	}
379 
380 	return 0;
381 }
382 
383 static int pnv_php_get_adapter_state(struct hotplug_slot *slot, u8 *state)
384 {
385 	struct pnv_php_slot *php_slot = slot->private;
386 	uint8_t presence = OPAL_PCI_SLOT_EMPTY;
387 	int ret;
388 
389 	/*
390 	 * Retrieve presence status from firmware. If we can't
391 	 * get that, it will fail back to be empty.
392 	 */
393 	ret = pnv_pci_get_presence_state(php_slot->id, &presence);
394 	if (ret >= 0) {
395 		*state = presence;
396 		slot->info->adapter_status = presence;
397 		ret = 0;
398 	} else {
399 		pci_warn(php_slot->pdev, "Error %d getting presence\n", ret);
400 	}
401 
402 	return ret;
403 }
404 
405 static int pnv_php_set_attention_state(struct hotplug_slot *slot, u8 state)
406 {
407 	/* FIXME: Make it real once firmware supports it */
408 	slot->info->attention_status = state;
409 
410 	return 0;
411 }
412 
413 static int pnv_php_enable(struct pnv_php_slot *php_slot, bool rescan)
414 {
415 	struct hotplug_slot *slot = &php_slot->slot;
416 	uint8_t presence = OPAL_PCI_SLOT_EMPTY;
417 	uint8_t power_status = OPAL_PCI_SLOT_POWER_ON;
418 	int ret;
419 
420 	/* Check if the slot has been configured */
421 	if (php_slot->state != PNV_PHP_STATE_REGISTERED)
422 		return 0;
423 
424 	/* Retrieve slot presence status */
425 	ret = pnv_php_get_adapter_state(slot, &presence);
426 	if (ret)
427 		return ret;
428 
429 	/*
430 	 * Proceed if there have nothing behind the slot. However,
431 	 * we should leave the slot in registered state at the
432 	 * beginning. Otherwise, the PCI devices inserted afterwards
433 	 * won't be probed and populated.
434 	 */
435 	if (presence == OPAL_PCI_SLOT_EMPTY) {
436 		if (!php_slot->power_state_check) {
437 			php_slot->power_state_check = true;
438 
439 			return 0;
440 		}
441 
442 		goto scan;
443 	}
444 
445 	/*
446 	 * If the power supply to the slot is off, we can't detect
447 	 * adapter presence state. That means we have to turn the
448 	 * slot on before going to probe slot's presence state.
449 	 *
450 	 * On the first time, we don't change the power status to
451 	 * boost system boot with assumption that the firmware
452 	 * supplies consistent slot power status: empty slot always
453 	 * has its power off and non-empty slot has its power on.
454 	 */
455 	if (!php_slot->power_state_check) {
456 		php_slot->power_state_check = true;
457 
458 		ret = pnv_php_get_power_state(slot, &power_status);
459 		if (ret)
460 			return ret;
461 
462 		if (power_status != OPAL_PCI_SLOT_POWER_ON)
463 			return 0;
464 	}
465 
466 	/* Check the power status. Scan the slot if it is already on */
467 	ret = pnv_php_get_power_state(slot, &power_status);
468 	if (ret)
469 		return ret;
470 
471 	if (power_status == OPAL_PCI_SLOT_POWER_ON)
472 		goto scan;
473 
474 	/* Power is off, turn it on and then scan the slot */
475 	ret = pnv_php_set_slot_power_state(slot, OPAL_PCI_SLOT_POWER_ON);
476 	if (ret)
477 		return ret;
478 
479 scan:
480 	if (presence == OPAL_PCI_SLOT_PRESENT) {
481 		if (rescan) {
482 			pci_lock_rescan_remove();
483 			pci_hp_add_devices(php_slot->bus);
484 			pci_unlock_rescan_remove();
485 		}
486 
487 		/* Rescan for child hotpluggable slots */
488 		php_slot->state = PNV_PHP_STATE_POPULATED;
489 		if (rescan)
490 			pnv_php_register(php_slot->dn);
491 	} else {
492 		php_slot->state = PNV_PHP_STATE_POPULATED;
493 	}
494 
495 	return 0;
496 }
497 
498 static int pnv_php_enable_slot(struct hotplug_slot *slot)
499 {
500 	struct pnv_php_slot *php_slot = container_of(slot,
501 						     struct pnv_php_slot, slot);
502 
503 	return pnv_php_enable(php_slot, true);
504 }
505 
506 static int pnv_php_disable_slot(struct hotplug_slot *slot)
507 {
508 	struct pnv_php_slot *php_slot = slot->private;
509 	int ret;
510 
511 	if (php_slot->state != PNV_PHP_STATE_POPULATED)
512 		return 0;
513 
514 	/* Remove all devices behind the slot */
515 	pci_lock_rescan_remove();
516 	pci_hp_remove_devices(php_slot->bus);
517 	pci_unlock_rescan_remove();
518 
519 	/* Detach the child hotpluggable slots */
520 	pnv_php_unregister(php_slot->dn);
521 
522 	/* Notify firmware and remove device nodes */
523 	ret = pnv_php_set_slot_power_state(slot, OPAL_PCI_SLOT_POWER_OFF);
524 
525 	php_slot->state = PNV_PHP_STATE_REGISTERED;
526 	return ret;
527 }
528 
529 static struct hotplug_slot_ops php_slot_ops = {
530 	.get_power_status	= pnv_php_get_power_state,
531 	.get_adapter_status	= pnv_php_get_adapter_state,
532 	.set_attention_status	= pnv_php_set_attention_state,
533 	.enable_slot		= pnv_php_enable_slot,
534 	.disable_slot		= pnv_php_disable_slot,
535 };
536 
537 static void pnv_php_release(struct hotplug_slot *slot)
538 {
539 	struct pnv_php_slot *php_slot = slot->private;
540 	unsigned long flags;
541 
542 	/* Remove from global or child list */
543 	spin_lock_irqsave(&pnv_php_lock, flags);
544 	list_del(&php_slot->link);
545 	spin_unlock_irqrestore(&pnv_php_lock, flags);
546 
547 	/* Detach from parent */
548 	pnv_php_put_slot(php_slot);
549 	pnv_php_put_slot(php_slot->parent);
550 }
551 
552 static struct pnv_php_slot *pnv_php_alloc_slot(struct device_node *dn)
553 {
554 	struct pnv_php_slot *php_slot;
555 	struct pci_bus *bus;
556 	const char *label;
557 	uint64_t id;
558 	int ret;
559 
560 	ret = of_property_read_string(dn, "ibm,slot-label", &label);
561 	if (ret)
562 		return NULL;
563 
564 	if (pnv_pci_get_slot_id(dn, &id))
565 		return NULL;
566 
567 	bus = pci_find_bus_by_node(dn);
568 	if (!bus)
569 		return NULL;
570 
571 	php_slot = kzalloc(sizeof(*php_slot), GFP_KERNEL);
572 	if (!php_slot)
573 		return NULL;
574 
575 	php_slot->name = kstrdup(label, GFP_KERNEL);
576 	if (!php_slot->name) {
577 		kfree(php_slot);
578 		return NULL;
579 	}
580 
581 	if (dn->child && PCI_DN(dn->child))
582 		php_slot->slot_no = PCI_SLOT(PCI_DN(dn->child)->devfn);
583 	else
584 		php_slot->slot_no = -1;   /* Placeholder slot */
585 
586 	kref_init(&php_slot->kref);
587 	php_slot->state	                = PNV_PHP_STATE_INITIALIZED;
588 	php_slot->dn	                = dn;
589 	php_slot->pdev	                = bus->self;
590 	php_slot->bus	                = bus;
591 	php_slot->id	                = id;
592 	php_slot->power_state_check     = false;
593 	php_slot->slot.ops              = &php_slot_ops;
594 	php_slot->slot.info             = &php_slot->slot_info;
595 	php_slot->slot.release          = pnv_php_release;
596 	php_slot->slot.private          = php_slot;
597 
598 	INIT_LIST_HEAD(&php_slot->children);
599 	INIT_LIST_HEAD(&php_slot->link);
600 
601 	return php_slot;
602 }
603 
604 static int pnv_php_register_slot(struct pnv_php_slot *php_slot)
605 {
606 	struct pnv_php_slot *parent;
607 	struct device_node *dn = php_slot->dn;
608 	unsigned long flags;
609 	int ret;
610 
611 	/* Check if the slot is registered or not */
612 	parent = pnv_php_find_slot(php_slot->dn);
613 	if (parent) {
614 		pnv_php_put_slot(parent);
615 		return -EEXIST;
616 	}
617 
618 	/* Register PCI slot */
619 	ret = pci_hp_register(&php_slot->slot, php_slot->bus,
620 			      php_slot->slot_no, php_slot->name);
621 	if (ret) {
622 		pci_warn(php_slot->pdev, "Error %d registering slot\n", ret);
623 		return ret;
624 	}
625 
626 	/* Attach to the parent's child list or global list */
627 	while ((dn = of_get_parent(dn))) {
628 		if (!PCI_DN(dn)) {
629 			of_node_put(dn);
630 			break;
631 		}
632 
633 		parent = pnv_php_find_slot(dn);
634 		if (parent) {
635 			of_node_put(dn);
636 			break;
637 		}
638 
639 		of_node_put(dn);
640 	}
641 
642 	spin_lock_irqsave(&pnv_php_lock, flags);
643 	php_slot->parent = parent;
644 	if (parent)
645 		list_add_tail(&php_slot->link, &parent->children);
646 	else
647 		list_add_tail(&php_slot->link, &pnv_php_slot_list);
648 	spin_unlock_irqrestore(&pnv_php_lock, flags);
649 
650 	php_slot->state = PNV_PHP_STATE_REGISTERED;
651 	return 0;
652 }
653 
654 static int pnv_php_enable_msix(struct pnv_php_slot *php_slot)
655 {
656 	struct pci_dev *pdev = php_slot->pdev;
657 	struct msix_entry entry;
658 	int nr_entries, ret;
659 	u16 pcie_flag;
660 
661 	/* Get total number of MSIx entries */
662 	nr_entries = pci_msix_vec_count(pdev);
663 	if (nr_entries < 0)
664 		return nr_entries;
665 
666 	/* Check hotplug MSIx entry is in range */
667 	pcie_capability_read_word(pdev, PCI_EXP_FLAGS, &pcie_flag);
668 	entry.entry = (pcie_flag & PCI_EXP_FLAGS_IRQ) >> 9;
669 	if (entry.entry >= nr_entries)
670 		return -ERANGE;
671 
672 	/* Enable MSIx */
673 	ret = pci_enable_msix_exact(pdev, &entry, 1);
674 	if (ret) {
675 		pci_warn(pdev, "Error %d enabling MSIx\n", ret);
676 		return ret;
677 	}
678 
679 	return entry.vector;
680 }
681 
682 static void pnv_php_event_handler(struct work_struct *work)
683 {
684 	struct pnv_php_event *event =
685 		container_of(work, struct pnv_php_event, work);
686 	struct pnv_php_slot *php_slot = event->php_slot;
687 
688 	if (event->added)
689 		pnv_php_enable_slot(&php_slot->slot);
690 	else
691 		pnv_php_disable_slot(&php_slot->slot);
692 
693 	kfree(event);
694 }
695 
696 static irqreturn_t pnv_php_interrupt(int irq, void *data)
697 {
698 	struct pnv_php_slot *php_slot = data;
699 	struct pci_dev *pchild, *pdev = php_slot->pdev;
700 	struct eeh_dev *edev;
701 	struct eeh_pe *pe;
702 	struct pnv_php_event *event;
703 	u16 sts, lsts;
704 	u8 presence;
705 	bool added;
706 	unsigned long flags;
707 	int ret;
708 
709 	pcie_capability_read_word(pdev, PCI_EXP_SLTSTA, &sts);
710 	sts &= (PCI_EXP_SLTSTA_PDC | PCI_EXP_SLTSTA_DLLSC);
711 	pcie_capability_write_word(pdev, PCI_EXP_SLTSTA, sts);
712 	if (sts & PCI_EXP_SLTSTA_DLLSC) {
713 		pcie_capability_read_word(pdev, PCI_EXP_LNKSTA, &lsts);
714 		added = !!(lsts & PCI_EXP_LNKSTA_DLLLA);
715 	} else if (!(php_slot->flags & PNV_PHP_FLAG_BROKEN_PDC) &&
716 		   (sts & PCI_EXP_SLTSTA_PDC)) {
717 		ret = pnv_pci_get_presence_state(php_slot->id, &presence);
718 		if (ret) {
719 			pci_warn(pdev, "PCI slot [%s] error %d getting presence (0x%04x), to retry the operation.\n",
720 				 php_slot->name, ret, sts);
721 			return IRQ_HANDLED;
722 		}
723 
724 		added = !!(presence == OPAL_PCI_SLOT_PRESENT);
725 	} else {
726 		return IRQ_NONE;
727 	}
728 
729 	/* Freeze the removed PE to avoid unexpected error reporting */
730 	if (!added) {
731 		pchild = list_first_entry_or_null(&php_slot->bus->devices,
732 						  struct pci_dev, bus_list);
733 		edev = pchild ? pci_dev_to_eeh_dev(pchild) : NULL;
734 		pe = edev ? edev->pe : NULL;
735 		if (pe) {
736 			eeh_serialize_lock(&flags);
737 			eeh_pe_state_mark(pe, EEH_PE_ISOLATED);
738 			eeh_serialize_unlock(flags);
739 			eeh_pe_set_option(pe, EEH_OPT_FREEZE_PE);
740 		}
741 	}
742 
743 	/*
744 	 * The PE is left in frozen state if the event is missed. It's
745 	 * fine as the PCI devices (PE) aren't functional any more.
746 	 */
747 	event = kzalloc(sizeof(*event), GFP_ATOMIC);
748 	if (!event) {
749 		pci_warn(pdev, "PCI slot [%s] missed hotplug event 0x%04x\n",
750 			 php_slot->name, sts);
751 		return IRQ_HANDLED;
752 	}
753 
754 	pci_info(pdev, "PCI slot [%s] %s (IRQ: %d)\n",
755 		 php_slot->name, added ? "added" : "removed", irq);
756 	INIT_WORK(&event->work, pnv_php_event_handler);
757 	event->added = added;
758 	event->php_slot = php_slot;
759 	queue_work(php_slot->wq, &event->work);
760 
761 	return IRQ_HANDLED;
762 }
763 
764 static void pnv_php_init_irq(struct pnv_php_slot *php_slot, int irq)
765 {
766 	struct pci_dev *pdev = php_slot->pdev;
767 	u32 broken_pdc = 0;
768 	u16 sts, ctrl;
769 	int ret;
770 
771 	/* Allocate workqueue */
772 	php_slot->wq = alloc_workqueue("pciehp-%s", 0, 0, php_slot->name);
773 	if (!php_slot->wq) {
774 		pci_warn(pdev, "Cannot alloc workqueue\n");
775 		pnv_php_disable_irq(php_slot, true);
776 		return;
777 	}
778 
779 	/* Check PDC (Presence Detection Change) is broken or not */
780 	ret = of_property_read_u32(php_slot->dn, "ibm,slot-broken-pdc",
781 				   &broken_pdc);
782 	if (!ret && broken_pdc)
783 		php_slot->flags |= PNV_PHP_FLAG_BROKEN_PDC;
784 
785 	/* Clear pending interrupts */
786 	pcie_capability_read_word(pdev, PCI_EXP_SLTSTA, &sts);
787 	if (php_slot->flags & PNV_PHP_FLAG_BROKEN_PDC)
788 		sts |= PCI_EXP_SLTSTA_DLLSC;
789 	else
790 		sts |= (PCI_EXP_SLTSTA_PDC | PCI_EXP_SLTSTA_DLLSC);
791 	pcie_capability_write_word(pdev, PCI_EXP_SLTSTA, sts);
792 
793 	/* Request the interrupt */
794 	ret = request_irq(irq, pnv_php_interrupt, IRQF_SHARED,
795 			  php_slot->name, php_slot);
796 	if (ret) {
797 		pnv_php_disable_irq(php_slot, true);
798 		pci_warn(pdev, "Error %d enabling IRQ %d\n", ret, irq);
799 		return;
800 	}
801 
802 	/* Enable the interrupts */
803 	pcie_capability_read_word(pdev, PCI_EXP_SLTCTL, &ctrl);
804 	if (php_slot->flags & PNV_PHP_FLAG_BROKEN_PDC) {
805 		ctrl &= ~PCI_EXP_SLTCTL_PDCE;
806 		ctrl |= (PCI_EXP_SLTCTL_HPIE |
807 			 PCI_EXP_SLTCTL_DLLSCE);
808 	} else {
809 		ctrl |= (PCI_EXP_SLTCTL_HPIE |
810 			 PCI_EXP_SLTCTL_PDCE |
811 			 PCI_EXP_SLTCTL_DLLSCE);
812 	}
813 	pcie_capability_write_word(pdev, PCI_EXP_SLTCTL, ctrl);
814 
815 	/* The interrupt is initialized successfully when @irq is valid */
816 	php_slot->irq = irq;
817 }
818 
819 static void pnv_php_enable_irq(struct pnv_php_slot *php_slot)
820 {
821 	struct pci_dev *pdev = php_slot->pdev;
822 	int irq, ret;
823 
824 	/*
825 	 * The MSI/MSIx interrupt might have been occupied by other
826 	 * drivers. Don't populate the surprise hotplug capability
827 	 * in that case.
828 	 */
829 	if (pci_dev_msi_enabled(pdev))
830 		return;
831 
832 	ret = pci_enable_device(pdev);
833 	if (ret) {
834 		pci_warn(pdev, "Error %d enabling device\n", ret);
835 		return;
836 	}
837 
838 	pci_set_master(pdev);
839 
840 	/* Enable MSIx interrupt */
841 	irq = pnv_php_enable_msix(php_slot);
842 	if (irq > 0) {
843 		pnv_php_init_irq(php_slot, irq);
844 		return;
845 	}
846 
847 	/*
848 	 * Use MSI if MSIx doesn't work. Fail back to legacy INTx
849 	 * if MSI doesn't work either
850 	 */
851 	ret = pci_enable_msi(pdev);
852 	if (!ret || pdev->irq) {
853 		irq = pdev->irq;
854 		pnv_php_init_irq(php_slot, irq);
855 	}
856 }
857 
858 static int pnv_php_register_one(struct device_node *dn)
859 {
860 	struct pnv_php_slot *php_slot;
861 	u32 prop32;
862 	int ret;
863 
864 	/* Check if it's hotpluggable slot */
865 	ret = of_property_read_u32(dn, "ibm,slot-pluggable", &prop32);
866 	if (ret || !prop32)
867 		return -ENXIO;
868 
869 	ret = of_property_read_u32(dn, "ibm,reset-by-firmware", &prop32);
870 	if (ret || !prop32)
871 		return -ENXIO;
872 
873 	php_slot = pnv_php_alloc_slot(dn);
874 	if (!php_slot)
875 		return -ENODEV;
876 
877 	ret = pnv_php_register_slot(php_slot);
878 	if (ret)
879 		goto free_slot;
880 
881 	ret = pnv_php_enable(php_slot, false);
882 	if (ret)
883 		goto unregister_slot;
884 
885 	/* Enable interrupt if the slot supports surprise hotplug */
886 	ret = of_property_read_u32(dn, "ibm,slot-surprise-pluggable", &prop32);
887 	if (!ret && prop32)
888 		pnv_php_enable_irq(php_slot);
889 
890 	return 0;
891 
892 unregister_slot:
893 	pnv_php_unregister_one(php_slot->dn);
894 free_slot:
895 	pnv_php_put_slot(php_slot);
896 	return ret;
897 }
898 
899 static void pnv_php_register(struct device_node *dn)
900 {
901 	struct device_node *child;
902 
903 	/*
904 	 * The parent slots should be registered before their
905 	 * child slots.
906 	 */
907 	for_each_child_of_node(dn, child) {
908 		pnv_php_register_one(child);
909 		pnv_php_register(child);
910 	}
911 }
912 
913 static void pnv_php_unregister_one(struct device_node *dn)
914 {
915 	struct pnv_php_slot *php_slot;
916 
917 	php_slot = pnv_php_find_slot(dn);
918 	if (!php_slot)
919 		return;
920 
921 	php_slot->state = PNV_PHP_STATE_OFFLINE;
922 	pnv_php_put_slot(php_slot);
923 	pci_hp_deregister(&php_slot->slot);
924 }
925 
926 static void pnv_php_unregister(struct device_node *dn)
927 {
928 	struct device_node *child;
929 
930 	/* The child slots should go before their parent slots */
931 	for_each_child_of_node(dn, child) {
932 		pnv_php_unregister(child);
933 		pnv_php_unregister_one(child);
934 	}
935 }
936 
937 static int __init pnv_php_init(void)
938 {
939 	struct device_node *dn;
940 
941 	pr_info(DRIVER_DESC " version: " DRIVER_VERSION "\n");
942 	for_each_compatible_node(dn, NULL, "ibm,ioda2-phb")
943 		pnv_php_register(dn);
944 
945 	return 0;
946 }
947 
948 static void __exit pnv_php_exit(void)
949 {
950 	struct device_node *dn;
951 
952 	for_each_compatible_node(dn, NULL, "ibm,ioda2-phb")
953 		pnv_php_unregister(dn);
954 }
955 
956 module_init(pnv_php_init);
957 module_exit(pnv_php_exit);
958 
959 MODULE_VERSION(DRIVER_VERSION);
960 MODULE_LICENSE("GPL v2");
961 MODULE_AUTHOR(DRIVER_AUTHOR);
962 MODULE_DESCRIPTION(DRIVER_DESC);
963