xref: /openbmc/linux/drivers/pci/hotplug/pnv_php.c (revision de528723)
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 			of_node_put(child);
225 			break;
226 		}
227 
228 		ret = pnv_php_populate_changeset(ocs, child);
229 		if (ret) {
230 			of_node_put(child);
231 			break;
232 		}
233 	}
234 
235 	return ret;
236 }
237 
238 static void *pnv_php_add_one_pdn(struct device_node *dn, void *data)
239 {
240 	struct pci_controller *hose = (struct pci_controller *)data;
241 	struct pci_dn *pdn;
242 
243 	pdn = pci_add_device_node_info(hose, dn);
244 	if (!pdn)
245 		return ERR_PTR(-ENOMEM);
246 
247 	return NULL;
248 }
249 
250 static void pnv_php_add_pdns(struct pnv_php_slot *slot)
251 {
252 	struct pci_controller *hose = pci_bus_to_host(slot->bus);
253 
254 	pci_traverse_device_nodes(slot->dn, pnv_php_add_one_pdn, hose);
255 }
256 
257 static int pnv_php_add_devtree(struct pnv_php_slot *php_slot)
258 {
259 	void *fdt, *fdt1, *dt;
260 	int ret;
261 
262 	/* We don't know the FDT blob size. We try to get it through
263 	 * maximal memory chunk and then copy it to another chunk that
264 	 * fits the real size.
265 	 */
266 	fdt1 = kzalloc(0x10000, GFP_KERNEL);
267 	if (!fdt1) {
268 		ret = -ENOMEM;
269 		goto out;
270 	}
271 
272 	ret = pnv_pci_get_device_tree(php_slot->dn->phandle, fdt1, 0x10000);
273 	if (ret) {
274 		pci_warn(php_slot->pdev, "Error %d getting FDT blob\n", ret);
275 		goto free_fdt1;
276 	}
277 
278 	fdt = kmemdup(fdt1, fdt_totalsize(fdt1), GFP_KERNEL);
279 	if (!fdt) {
280 		ret = -ENOMEM;
281 		goto free_fdt1;
282 	}
283 
284 	/* Unflatten device tree blob */
285 	dt = of_fdt_unflatten_tree(fdt, php_slot->dn, NULL);
286 	if (!dt) {
287 		ret = -EINVAL;
288 		pci_warn(php_slot->pdev, "Cannot unflatten FDT\n");
289 		goto free_fdt;
290 	}
291 
292 	/* Initialize and apply the changeset */
293 	of_changeset_init(&php_slot->ocs);
294 	pnv_php_reverse_nodes(php_slot->dn);
295 	ret = pnv_php_populate_changeset(&php_slot->ocs, php_slot->dn);
296 	if (ret) {
297 		pnv_php_reverse_nodes(php_slot->dn);
298 		pci_warn(php_slot->pdev, "Error %d populating changeset\n",
299 			 ret);
300 		goto free_dt;
301 	}
302 
303 	php_slot->dn->child = NULL;
304 	ret = of_changeset_apply(&php_slot->ocs);
305 	if (ret) {
306 		pci_warn(php_slot->pdev, "Error %d applying changeset\n", ret);
307 		goto destroy_changeset;
308 	}
309 
310 	/* Add device node firmware data */
311 	pnv_php_add_pdns(php_slot);
312 	php_slot->fdt = fdt;
313 	php_slot->dt  = dt;
314 	kfree(fdt1);
315 	goto out;
316 
317 destroy_changeset:
318 	of_changeset_destroy(&php_slot->ocs);
319 free_dt:
320 	kfree(dt);
321 	php_slot->dn->child = NULL;
322 free_fdt:
323 	kfree(fdt);
324 free_fdt1:
325 	kfree(fdt1);
326 out:
327 	return ret;
328 }
329 
330 static inline struct pnv_php_slot *to_pnv_php_slot(struct hotplug_slot *slot)
331 {
332 	return container_of(slot, struct pnv_php_slot, slot);
333 }
334 
335 int pnv_php_set_slot_power_state(struct hotplug_slot *slot,
336 				 uint8_t state)
337 {
338 	struct pnv_php_slot *php_slot = to_pnv_php_slot(slot);
339 	struct opal_msg msg;
340 	int ret;
341 
342 	ret = pnv_pci_set_power_state(php_slot->id, state, &msg);
343 	if (ret > 0) {
344 		if (be64_to_cpu(msg.params[1]) != php_slot->dn->phandle	||
345 		    be64_to_cpu(msg.params[2]) != state			||
346 		    be64_to_cpu(msg.params[3]) != OPAL_SUCCESS) {
347 			pci_warn(php_slot->pdev, "Wrong msg (%lld, %lld, %lld)\n",
348 				 be64_to_cpu(msg.params[1]),
349 				 be64_to_cpu(msg.params[2]),
350 				 be64_to_cpu(msg.params[3]));
351 			return -ENOMSG;
352 		}
353 	} else if (ret < 0) {
354 		pci_warn(php_slot->pdev, "Error %d powering %s\n",
355 			 ret, (state == OPAL_PCI_SLOT_POWER_ON) ? "on" : "off");
356 		return ret;
357 	}
358 
359 	if (state == OPAL_PCI_SLOT_POWER_OFF || state == OPAL_PCI_SLOT_OFFLINE)
360 		pnv_php_rmv_devtree(php_slot);
361 	else
362 		ret = pnv_php_add_devtree(php_slot);
363 
364 	return ret;
365 }
366 EXPORT_SYMBOL_GPL(pnv_php_set_slot_power_state);
367 
368 static int pnv_php_get_power_state(struct hotplug_slot *slot, u8 *state)
369 {
370 	struct pnv_php_slot *php_slot = to_pnv_php_slot(slot);
371 	uint8_t power_state = OPAL_PCI_SLOT_POWER_ON;
372 	int ret;
373 
374 	/*
375 	 * Retrieve power status from firmware. If we fail
376 	 * getting that, the power status fails back to
377 	 * be on.
378 	 */
379 	ret = pnv_pci_get_power_state(php_slot->id, &power_state);
380 	if (ret) {
381 		pci_warn(php_slot->pdev, "Error %d getting power status\n",
382 			 ret);
383 	} else {
384 		*state = power_state;
385 	}
386 
387 	return 0;
388 }
389 
390 static int pnv_php_get_adapter_state(struct hotplug_slot *slot, u8 *state)
391 {
392 	struct pnv_php_slot *php_slot = to_pnv_php_slot(slot);
393 	uint8_t presence = OPAL_PCI_SLOT_EMPTY;
394 	int ret;
395 
396 	/*
397 	 * Retrieve presence status from firmware. If we can't
398 	 * get that, it will fail back to be empty.
399 	 */
400 	ret = pnv_pci_get_presence_state(php_slot->id, &presence);
401 	if (ret >= 0) {
402 		*state = presence;
403 		ret = 0;
404 	} else {
405 		pci_warn(php_slot->pdev, "Error %d getting presence\n", ret);
406 	}
407 
408 	return ret;
409 }
410 
411 static int pnv_php_get_attention_state(struct hotplug_slot *slot, u8 *state)
412 {
413 	struct pnv_php_slot *php_slot = to_pnv_php_slot(slot);
414 
415 	*state = php_slot->attention_state;
416 	return 0;
417 }
418 
419 static int pnv_php_set_attention_state(struct hotplug_slot *slot, u8 state)
420 {
421 	struct pnv_php_slot *php_slot = to_pnv_php_slot(slot);
422 	struct pci_dev *bridge = php_slot->pdev;
423 	u16 new, mask;
424 
425 	php_slot->attention_state = state;
426 	if (!bridge)
427 		return 0;
428 
429 	mask = PCI_EXP_SLTCTL_AIC;
430 
431 	if (state)
432 		new = PCI_EXP_SLTCTL_ATTN_IND_ON;
433 	else
434 		new = PCI_EXP_SLTCTL_ATTN_IND_OFF;
435 
436 	pcie_capability_clear_and_set_word(bridge, PCI_EXP_SLTCTL, mask, new);
437 
438 	return 0;
439 }
440 
441 static int pnv_php_enable(struct pnv_php_slot *php_slot, bool rescan)
442 {
443 	struct hotplug_slot *slot = &php_slot->slot;
444 	uint8_t presence = OPAL_PCI_SLOT_EMPTY;
445 	uint8_t power_status = OPAL_PCI_SLOT_POWER_ON;
446 	int ret;
447 
448 	/* Check if the slot has been configured */
449 	if (php_slot->state != PNV_PHP_STATE_REGISTERED)
450 		return 0;
451 
452 	/* Retrieve slot presence status */
453 	ret = pnv_php_get_adapter_state(slot, &presence);
454 	if (ret)
455 		return ret;
456 
457 	/*
458 	 * Proceed if there have nothing behind the slot. However,
459 	 * we should leave the slot in registered state at the
460 	 * beginning. Otherwise, the PCI devices inserted afterwards
461 	 * won't be probed and populated.
462 	 */
463 	if (presence == OPAL_PCI_SLOT_EMPTY) {
464 		if (!php_slot->power_state_check) {
465 			php_slot->power_state_check = true;
466 
467 			return 0;
468 		}
469 
470 		goto scan;
471 	}
472 
473 	/*
474 	 * If the power supply to the slot is off, we can't detect
475 	 * adapter presence state. That means we have to turn the
476 	 * slot on before going to probe slot's presence state.
477 	 *
478 	 * On the first time, we don't change the power status to
479 	 * boost system boot with assumption that the firmware
480 	 * supplies consistent slot power status: empty slot always
481 	 * has its power off and non-empty slot has its power on.
482 	 */
483 	if (!php_slot->power_state_check) {
484 		php_slot->power_state_check = true;
485 
486 		ret = pnv_php_get_power_state(slot, &power_status);
487 		if (ret)
488 			return ret;
489 
490 		if (power_status != OPAL_PCI_SLOT_POWER_ON)
491 			return 0;
492 	}
493 
494 	/* Check the power status. Scan the slot if it is already on */
495 	ret = pnv_php_get_power_state(slot, &power_status);
496 	if (ret)
497 		return ret;
498 
499 	if (power_status == OPAL_PCI_SLOT_POWER_ON)
500 		goto scan;
501 
502 	/* Power is off, turn it on and then scan the slot */
503 	ret = pnv_php_set_slot_power_state(slot, OPAL_PCI_SLOT_POWER_ON);
504 	if (ret)
505 		return ret;
506 
507 scan:
508 	if (presence == OPAL_PCI_SLOT_PRESENT) {
509 		if (rescan) {
510 			pci_lock_rescan_remove();
511 			pci_hp_add_devices(php_slot->bus);
512 			pci_unlock_rescan_remove();
513 		}
514 
515 		/* Rescan for child hotpluggable slots */
516 		php_slot->state = PNV_PHP_STATE_POPULATED;
517 		if (rescan)
518 			pnv_php_register(php_slot->dn);
519 	} else {
520 		php_slot->state = PNV_PHP_STATE_POPULATED;
521 	}
522 
523 	return 0;
524 }
525 
526 static int pnv_php_reset_slot(struct hotplug_slot *slot, int probe)
527 {
528 	struct pnv_php_slot *php_slot = to_pnv_php_slot(slot);
529 	struct pci_dev *bridge = php_slot->pdev;
530 	uint16_t sts;
531 
532 	/*
533 	 * The CAPI folks want pnv_php to drive OpenCAPI slots
534 	 * which don't have a bridge. Only claim to support
535 	 * reset_slot() if we have a bridge device (for now...)
536 	 */
537 	if (probe)
538 		return !bridge;
539 
540 	/* mask our interrupt while resetting the bridge */
541 	if (php_slot->irq > 0)
542 		disable_irq(php_slot->irq);
543 
544 	pci_bridge_secondary_bus_reset(bridge);
545 
546 	/* clear any state changes that happened due to the reset */
547 	pcie_capability_read_word(php_slot->pdev, PCI_EXP_SLTSTA, &sts);
548 	sts &= (PCI_EXP_SLTSTA_PDC | PCI_EXP_SLTSTA_DLLSC);
549 	pcie_capability_write_word(php_slot->pdev, PCI_EXP_SLTSTA, sts);
550 
551 	if (php_slot->irq > 0)
552 		enable_irq(php_slot->irq);
553 
554 	return 0;
555 }
556 
557 static int pnv_php_enable_slot(struct hotplug_slot *slot)
558 {
559 	struct pnv_php_slot *php_slot = to_pnv_php_slot(slot);
560 
561 	return pnv_php_enable(php_slot, true);
562 }
563 
564 static int pnv_php_disable_slot(struct hotplug_slot *slot)
565 {
566 	struct pnv_php_slot *php_slot = to_pnv_php_slot(slot);
567 	int ret;
568 
569 	if (php_slot->state != PNV_PHP_STATE_POPULATED)
570 		return 0;
571 
572 	/* Remove all devices behind the slot */
573 	pci_lock_rescan_remove();
574 	pci_hp_remove_devices(php_slot->bus);
575 	pci_unlock_rescan_remove();
576 
577 	/* Detach the child hotpluggable slots */
578 	pnv_php_unregister(php_slot->dn);
579 
580 	/* Notify firmware and remove device nodes */
581 	ret = pnv_php_set_slot_power_state(slot, OPAL_PCI_SLOT_POWER_OFF);
582 
583 	php_slot->state = PNV_PHP_STATE_REGISTERED;
584 	return ret;
585 }
586 
587 static const struct hotplug_slot_ops php_slot_ops = {
588 	.get_power_status	= pnv_php_get_power_state,
589 	.get_adapter_status	= pnv_php_get_adapter_state,
590 	.get_attention_status	= pnv_php_get_attention_state,
591 	.set_attention_status	= pnv_php_set_attention_state,
592 	.enable_slot		= pnv_php_enable_slot,
593 	.disable_slot		= pnv_php_disable_slot,
594 	.reset_slot		= pnv_php_reset_slot,
595 };
596 
597 static void pnv_php_release(struct pnv_php_slot *php_slot)
598 {
599 	unsigned long flags;
600 
601 	/* Remove from global or child list */
602 	spin_lock_irqsave(&pnv_php_lock, flags);
603 	list_del(&php_slot->link);
604 	spin_unlock_irqrestore(&pnv_php_lock, flags);
605 
606 	/* Detach from parent */
607 	pnv_php_put_slot(php_slot);
608 	pnv_php_put_slot(php_slot->parent);
609 }
610 
611 static struct pnv_php_slot *pnv_php_alloc_slot(struct device_node *dn)
612 {
613 	struct pnv_php_slot *php_slot;
614 	struct pci_bus *bus;
615 	const char *label;
616 	uint64_t id;
617 	int ret;
618 
619 	ret = of_property_read_string(dn, "ibm,slot-label", &label);
620 	if (ret)
621 		return NULL;
622 
623 	if (pnv_pci_get_slot_id(dn, &id))
624 		return NULL;
625 
626 	bus = pci_find_bus_by_node(dn);
627 	if (!bus)
628 		return NULL;
629 
630 	php_slot = kzalloc(sizeof(*php_slot), GFP_KERNEL);
631 	if (!php_slot)
632 		return NULL;
633 
634 	php_slot->name = kstrdup(label, GFP_KERNEL);
635 	if (!php_slot->name) {
636 		kfree(php_slot);
637 		return NULL;
638 	}
639 
640 	if (dn->child && PCI_DN(dn->child))
641 		php_slot->slot_no = PCI_SLOT(PCI_DN(dn->child)->devfn);
642 	else
643 		php_slot->slot_no = -1;   /* Placeholder slot */
644 
645 	kref_init(&php_slot->kref);
646 	php_slot->state	                = PNV_PHP_STATE_INITIALIZED;
647 	php_slot->dn	                = dn;
648 	php_slot->pdev	                = bus->self;
649 	php_slot->bus	                = bus;
650 	php_slot->id	                = id;
651 	php_slot->power_state_check     = false;
652 	php_slot->slot.ops              = &php_slot_ops;
653 
654 	INIT_LIST_HEAD(&php_slot->children);
655 	INIT_LIST_HEAD(&php_slot->link);
656 
657 	return php_slot;
658 }
659 
660 static int pnv_php_register_slot(struct pnv_php_slot *php_slot)
661 {
662 	struct pnv_php_slot *parent;
663 	struct device_node *dn = php_slot->dn;
664 	unsigned long flags;
665 	int ret;
666 
667 	/* Check if the slot is registered or not */
668 	parent = pnv_php_find_slot(php_slot->dn);
669 	if (parent) {
670 		pnv_php_put_slot(parent);
671 		return -EEXIST;
672 	}
673 
674 	/* Register PCI slot */
675 	ret = pci_hp_register(&php_slot->slot, php_slot->bus,
676 			      php_slot->slot_no, php_slot->name);
677 	if (ret) {
678 		pci_warn(php_slot->pdev, "Error %d registering slot\n", ret);
679 		return ret;
680 	}
681 
682 	/* Attach to the parent's child list or global list */
683 	while ((dn = of_get_parent(dn))) {
684 		if (!PCI_DN(dn)) {
685 			of_node_put(dn);
686 			break;
687 		}
688 
689 		parent = pnv_php_find_slot(dn);
690 		if (parent) {
691 			of_node_put(dn);
692 			break;
693 		}
694 
695 		of_node_put(dn);
696 	}
697 
698 	spin_lock_irqsave(&pnv_php_lock, flags);
699 	php_slot->parent = parent;
700 	if (parent)
701 		list_add_tail(&php_slot->link, &parent->children);
702 	else
703 		list_add_tail(&php_slot->link, &pnv_php_slot_list);
704 	spin_unlock_irqrestore(&pnv_php_lock, flags);
705 
706 	php_slot->state = PNV_PHP_STATE_REGISTERED;
707 	return 0;
708 }
709 
710 static int pnv_php_enable_msix(struct pnv_php_slot *php_slot)
711 {
712 	struct pci_dev *pdev = php_slot->pdev;
713 	struct msix_entry entry;
714 	int nr_entries, ret;
715 	u16 pcie_flag;
716 
717 	/* Get total number of MSIx entries */
718 	nr_entries = pci_msix_vec_count(pdev);
719 	if (nr_entries < 0)
720 		return nr_entries;
721 
722 	/* Check hotplug MSIx entry is in range */
723 	pcie_capability_read_word(pdev, PCI_EXP_FLAGS, &pcie_flag);
724 	entry.entry = (pcie_flag & PCI_EXP_FLAGS_IRQ) >> 9;
725 	if (entry.entry >= nr_entries)
726 		return -ERANGE;
727 
728 	/* Enable MSIx */
729 	ret = pci_enable_msix_exact(pdev, &entry, 1);
730 	if (ret) {
731 		pci_warn(pdev, "Error %d enabling MSIx\n", ret);
732 		return ret;
733 	}
734 
735 	return entry.vector;
736 }
737 
738 static void pnv_php_event_handler(struct work_struct *work)
739 {
740 	struct pnv_php_event *event =
741 		container_of(work, struct pnv_php_event, work);
742 	struct pnv_php_slot *php_slot = event->php_slot;
743 
744 	if (event->added)
745 		pnv_php_enable_slot(&php_slot->slot);
746 	else
747 		pnv_php_disable_slot(&php_slot->slot);
748 
749 	kfree(event);
750 }
751 
752 static irqreturn_t pnv_php_interrupt(int irq, void *data)
753 {
754 	struct pnv_php_slot *php_slot = data;
755 	struct pci_dev *pchild, *pdev = php_slot->pdev;
756 	struct eeh_dev *edev;
757 	struct eeh_pe *pe;
758 	struct pnv_php_event *event;
759 	u16 sts, lsts;
760 	u8 presence;
761 	bool added;
762 	unsigned long flags;
763 	int ret;
764 
765 	pcie_capability_read_word(pdev, PCI_EXP_SLTSTA, &sts);
766 	sts &= (PCI_EXP_SLTSTA_PDC | PCI_EXP_SLTSTA_DLLSC);
767 	pcie_capability_write_word(pdev, PCI_EXP_SLTSTA, sts);
768 
769 	pci_dbg(pdev, "PCI slot [%s]: HP int! DLAct: %d, PresDet: %d\n",
770 			php_slot->name,
771 			!!(sts & PCI_EXP_SLTSTA_DLLSC),
772 			!!(sts & PCI_EXP_SLTSTA_PDC));
773 
774 	if (sts & PCI_EXP_SLTSTA_DLLSC) {
775 		pcie_capability_read_word(pdev, PCI_EXP_LNKSTA, &lsts);
776 		added = !!(lsts & PCI_EXP_LNKSTA_DLLLA);
777 	} else if (!(php_slot->flags & PNV_PHP_FLAG_BROKEN_PDC) &&
778 		   (sts & PCI_EXP_SLTSTA_PDC)) {
779 		ret = pnv_pci_get_presence_state(php_slot->id, &presence);
780 		if (ret) {
781 			pci_warn(pdev, "PCI slot [%s] error %d getting presence (0x%04x), to retry the operation.\n",
782 				 php_slot->name, ret, sts);
783 			return IRQ_HANDLED;
784 		}
785 
786 		added = !!(presence == OPAL_PCI_SLOT_PRESENT);
787 	} else {
788 		pci_dbg(pdev, "PCI slot [%s]: Spurious IRQ?\n", php_slot->name);
789 		return IRQ_NONE;
790 	}
791 
792 	/* Freeze the removed PE to avoid unexpected error reporting */
793 	if (!added) {
794 		pchild = list_first_entry_or_null(&php_slot->bus->devices,
795 						  struct pci_dev, bus_list);
796 		edev = pchild ? pci_dev_to_eeh_dev(pchild) : NULL;
797 		pe = edev ? edev->pe : NULL;
798 		if (pe) {
799 			eeh_serialize_lock(&flags);
800 			eeh_pe_mark_isolated(pe);
801 			eeh_serialize_unlock(flags);
802 			eeh_pe_set_option(pe, EEH_OPT_FREEZE_PE);
803 		}
804 	}
805 
806 	/*
807 	 * The PE is left in frozen state if the event is missed. It's
808 	 * fine as the PCI devices (PE) aren't functional any more.
809 	 */
810 	event = kzalloc(sizeof(*event), GFP_ATOMIC);
811 	if (!event) {
812 		pci_warn(pdev, "PCI slot [%s] missed hotplug event 0x%04x\n",
813 			 php_slot->name, sts);
814 		return IRQ_HANDLED;
815 	}
816 
817 	pci_info(pdev, "PCI slot [%s] %s (IRQ: %d)\n",
818 		 php_slot->name, added ? "added" : "removed", irq);
819 	INIT_WORK(&event->work, pnv_php_event_handler);
820 	event->added = added;
821 	event->php_slot = php_slot;
822 	queue_work(php_slot->wq, &event->work);
823 
824 	return IRQ_HANDLED;
825 }
826 
827 static void pnv_php_init_irq(struct pnv_php_slot *php_slot, int irq)
828 {
829 	struct pci_dev *pdev = php_slot->pdev;
830 	u32 broken_pdc = 0;
831 	u16 sts, ctrl;
832 	int ret;
833 
834 	/* Allocate workqueue */
835 	php_slot->wq = alloc_workqueue("pciehp-%s", 0, 0, php_slot->name);
836 	if (!php_slot->wq) {
837 		pci_warn(pdev, "Cannot alloc workqueue\n");
838 		pnv_php_disable_irq(php_slot, true);
839 		return;
840 	}
841 
842 	/* Check PDC (Presence Detection Change) is broken or not */
843 	ret = of_property_read_u32(php_slot->dn, "ibm,slot-broken-pdc",
844 				   &broken_pdc);
845 	if (!ret && broken_pdc)
846 		php_slot->flags |= PNV_PHP_FLAG_BROKEN_PDC;
847 
848 	/* Clear pending interrupts */
849 	pcie_capability_read_word(pdev, PCI_EXP_SLTSTA, &sts);
850 	if (php_slot->flags & PNV_PHP_FLAG_BROKEN_PDC)
851 		sts |= PCI_EXP_SLTSTA_DLLSC;
852 	else
853 		sts |= (PCI_EXP_SLTSTA_PDC | PCI_EXP_SLTSTA_DLLSC);
854 	pcie_capability_write_word(pdev, PCI_EXP_SLTSTA, sts);
855 
856 	/* Request the interrupt */
857 	ret = request_irq(irq, pnv_php_interrupt, IRQF_SHARED,
858 			  php_slot->name, php_slot);
859 	if (ret) {
860 		pnv_php_disable_irq(php_slot, true);
861 		pci_warn(pdev, "Error %d enabling IRQ %d\n", ret, irq);
862 		return;
863 	}
864 
865 	/* Enable the interrupts */
866 	pcie_capability_read_word(pdev, PCI_EXP_SLTCTL, &ctrl);
867 	if (php_slot->flags & PNV_PHP_FLAG_BROKEN_PDC) {
868 		ctrl &= ~PCI_EXP_SLTCTL_PDCE;
869 		ctrl |= (PCI_EXP_SLTCTL_HPIE |
870 			 PCI_EXP_SLTCTL_DLLSCE);
871 	} else {
872 		ctrl |= (PCI_EXP_SLTCTL_HPIE |
873 			 PCI_EXP_SLTCTL_PDCE |
874 			 PCI_EXP_SLTCTL_DLLSCE);
875 	}
876 	pcie_capability_write_word(pdev, PCI_EXP_SLTCTL, ctrl);
877 
878 	/* The interrupt is initialized successfully when @irq is valid */
879 	php_slot->irq = irq;
880 }
881 
882 static void pnv_php_enable_irq(struct pnv_php_slot *php_slot)
883 {
884 	struct pci_dev *pdev = php_slot->pdev;
885 	int irq, ret;
886 
887 	/*
888 	 * The MSI/MSIx interrupt might have been occupied by other
889 	 * drivers. Don't populate the surprise hotplug capability
890 	 * in that case.
891 	 */
892 	if (pci_dev_msi_enabled(pdev))
893 		return;
894 
895 	ret = pci_enable_device(pdev);
896 	if (ret) {
897 		pci_warn(pdev, "Error %d enabling device\n", ret);
898 		return;
899 	}
900 
901 	pci_set_master(pdev);
902 
903 	/* Enable MSIx interrupt */
904 	irq = pnv_php_enable_msix(php_slot);
905 	if (irq > 0) {
906 		pnv_php_init_irq(php_slot, irq);
907 		return;
908 	}
909 
910 	/*
911 	 * Use MSI if MSIx doesn't work. Fail back to legacy INTx
912 	 * if MSI doesn't work either
913 	 */
914 	ret = pci_enable_msi(pdev);
915 	if (!ret || pdev->irq) {
916 		irq = pdev->irq;
917 		pnv_php_init_irq(php_slot, irq);
918 	}
919 }
920 
921 static int pnv_php_register_one(struct device_node *dn)
922 {
923 	struct pnv_php_slot *php_slot;
924 	u32 prop32;
925 	int ret;
926 
927 	/* Check if it's hotpluggable slot */
928 	ret = of_property_read_u32(dn, "ibm,slot-pluggable", &prop32);
929 	if (ret || !prop32)
930 		return -ENXIO;
931 
932 	ret = of_property_read_u32(dn, "ibm,reset-by-firmware", &prop32);
933 	if (ret || !prop32)
934 		return -ENXIO;
935 
936 	php_slot = pnv_php_alloc_slot(dn);
937 	if (!php_slot)
938 		return -ENODEV;
939 
940 	ret = pnv_php_register_slot(php_slot);
941 	if (ret)
942 		goto free_slot;
943 
944 	ret = pnv_php_enable(php_slot, false);
945 	if (ret)
946 		goto unregister_slot;
947 
948 	/* Enable interrupt if the slot supports surprise hotplug */
949 	ret = of_property_read_u32(dn, "ibm,slot-surprise-pluggable", &prop32);
950 	if (!ret && prop32)
951 		pnv_php_enable_irq(php_slot);
952 
953 	return 0;
954 
955 unregister_slot:
956 	pnv_php_unregister_one(php_slot->dn);
957 free_slot:
958 	pnv_php_put_slot(php_slot);
959 	return ret;
960 }
961 
962 static void pnv_php_register(struct device_node *dn)
963 {
964 	struct device_node *child;
965 
966 	/*
967 	 * The parent slots should be registered before their
968 	 * child slots.
969 	 */
970 	for_each_child_of_node(dn, child) {
971 		pnv_php_register_one(child);
972 		pnv_php_register(child);
973 	}
974 }
975 
976 static void pnv_php_unregister_one(struct device_node *dn)
977 {
978 	struct pnv_php_slot *php_slot;
979 
980 	php_slot = pnv_php_find_slot(dn);
981 	if (!php_slot)
982 		return;
983 
984 	php_slot->state = PNV_PHP_STATE_OFFLINE;
985 	pci_hp_deregister(&php_slot->slot);
986 	pnv_php_release(php_slot);
987 	pnv_php_put_slot(php_slot);
988 }
989 
990 static void pnv_php_unregister(struct device_node *dn)
991 {
992 	struct device_node *child;
993 
994 	/* The child slots should go before their parent slots */
995 	for_each_child_of_node(dn, child) {
996 		pnv_php_unregister(child);
997 		pnv_php_unregister_one(child);
998 	}
999 }
1000 
1001 static int __init pnv_php_init(void)
1002 {
1003 	struct device_node *dn;
1004 
1005 	pr_info(DRIVER_DESC " version: " DRIVER_VERSION "\n");
1006 	for_each_compatible_node(dn, NULL, "ibm,ioda2-phb")
1007 		pnv_php_register(dn);
1008 
1009 	for_each_compatible_node(dn, NULL, "ibm,ioda3-phb")
1010 		pnv_php_register(dn);
1011 
1012 	return 0;
1013 }
1014 
1015 static void __exit pnv_php_exit(void)
1016 {
1017 	struct device_node *dn;
1018 
1019 	for_each_compatible_node(dn, NULL, "ibm,ioda2-phb")
1020 		pnv_php_unregister(dn);
1021 
1022 	for_each_compatible_node(dn, NULL, "ibm,ioda3-phb")
1023 		pnv_php_unregister(dn);
1024 }
1025 
1026 module_init(pnv_php_init);
1027 module_exit(pnv_php_exit);
1028 
1029 MODULE_VERSION(DRIVER_VERSION);
1030 MODULE_LICENSE("GPL v2");
1031 MODULE_AUTHOR(DRIVER_AUTHOR);
1032 MODULE_DESCRIPTION(DRIVER_DESC);
1033