xref: /openbmc/linux/drivers/pci/hotplug/pnv_php.c (revision 3a35093a)
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 #define SLOT_WARN(sl, x...) \
22 	((sl)->pdev ? pci_warn((sl)->pdev, x) : dev_warn(&(sl)->bus->dev, x))
23 
24 struct pnv_php_event {
25 	bool			added;
26 	struct pnv_php_slot	*php_slot;
27 	struct work_struct	work;
28 };
29 
30 static LIST_HEAD(pnv_php_slot_list);
31 static DEFINE_SPINLOCK(pnv_php_lock);
32 
33 static void pnv_php_register(struct device_node *dn);
34 static void pnv_php_unregister_one(struct device_node *dn);
35 static void pnv_php_unregister(struct device_node *dn);
36 
37 static void pnv_php_disable_irq(struct pnv_php_slot *php_slot,
38 				bool disable_device)
39 {
40 	struct pci_dev *pdev = php_slot->pdev;
41 	int irq = php_slot->irq;
42 	u16 ctrl;
43 
44 	if (php_slot->irq > 0) {
45 		pcie_capability_read_word(pdev, PCI_EXP_SLTCTL, &ctrl);
46 		ctrl &= ~(PCI_EXP_SLTCTL_HPIE |
47 			  PCI_EXP_SLTCTL_PDCE |
48 			  PCI_EXP_SLTCTL_DLLSCE);
49 		pcie_capability_write_word(pdev, PCI_EXP_SLTCTL, ctrl);
50 
51 		free_irq(php_slot->irq, php_slot);
52 		php_slot->irq = 0;
53 	}
54 
55 	if (php_slot->wq) {
56 		destroy_workqueue(php_slot->wq);
57 		php_slot->wq = NULL;
58 	}
59 
60 	if (disable_device || irq > 0) {
61 		if (pdev->msix_enabled)
62 			pci_disable_msix(pdev);
63 		else if (pdev->msi_enabled)
64 			pci_disable_msi(pdev);
65 
66 		pci_disable_device(pdev);
67 	}
68 }
69 
70 static void pnv_php_free_slot(struct kref *kref)
71 {
72 	struct pnv_php_slot *php_slot = container_of(kref,
73 					struct pnv_php_slot, kref);
74 
75 	WARN_ON(!list_empty(&php_slot->children));
76 	pnv_php_disable_irq(php_slot, false);
77 	kfree(php_slot->name);
78 	kfree(php_slot);
79 }
80 
81 static inline void pnv_php_put_slot(struct pnv_php_slot *php_slot)
82 {
83 
84 	if (!php_slot)
85 		return;
86 
87 	kref_put(&php_slot->kref, pnv_php_free_slot);
88 }
89 
90 static struct pnv_php_slot *pnv_php_match(struct device_node *dn,
91 					  struct pnv_php_slot *php_slot)
92 {
93 	struct pnv_php_slot *target, *tmp;
94 
95 	if (php_slot->dn == dn) {
96 		kref_get(&php_slot->kref);
97 		return php_slot;
98 	}
99 
100 	list_for_each_entry(tmp, &php_slot->children, link) {
101 		target = pnv_php_match(dn, tmp);
102 		if (target)
103 			return target;
104 	}
105 
106 	return NULL;
107 }
108 
109 struct pnv_php_slot *pnv_php_find_slot(struct device_node *dn)
110 {
111 	struct pnv_php_slot *php_slot, *tmp;
112 	unsigned long flags;
113 
114 	spin_lock_irqsave(&pnv_php_lock, flags);
115 	list_for_each_entry(tmp, &pnv_php_slot_list, link) {
116 		php_slot = pnv_php_match(dn, tmp);
117 		if (php_slot) {
118 			spin_unlock_irqrestore(&pnv_php_lock, flags);
119 			return php_slot;
120 		}
121 	}
122 	spin_unlock_irqrestore(&pnv_php_lock, flags);
123 
124 	return NULL;
125 }
126 EXPORT_SYMBOL_GPL(pnv_php_find_slot);
127 
128 /*
129  * Remove pdn for all children of the indicated device node.
130  * The function should remove pdn in a depth-first manner.
131  */
132 static void pnv_php_rmv_pdns(struct device_node *dn)
133 {
134 	struct device_node *child;
135 
136 	for_each_child_of_node(dn, child) {
137 		pnv_php_rmv_pdns(child);
138 
139 		pci_remove_device_node_info(child);
140 	}
141 }
142 
143 /*
144  * Detach all child nodes of the indicated device nodes. The
145  * function should handle device nodes in depth-first manner.
146  *
147  * We should not invoke of_node_release() as the memory for
148  * individual device node is part of large memory block. The
149  * large block is allocated from memblock (system bootup) or
150  * kmalloc() when unflattening the device tree by OF changeset.
151  * We can not free the large block allocated from memblock. For
152  * later case, it should be released at once.
153  */
154 static void pnv_php_detach_device_nodes(struct device_node *parent)
155 {
156 	struct device_node *dn;
157 
158 	for_each_child_of_node(parent, dn) {
159 		pnv_php_detach_device_nodes(dn);
160 
161 		of_node_put(dn);
162 		of_detach_node(dn);
163 	}
164 }
165 
166 static void pnv_php_rmv_devtree(struct pnv_php_slot *php_slot)
167 {
168 	pnv_php_rmv_pdns(php_slot->dn);
169 
170 	/*
171 	 * Decrease the refcount if the device nodes were created
172 	 * through OF changeset before detaching them.
173 	 */
174 	if (php_slot->fdt)
175 		of_changeset_destroy(&php_slot->ocs);
176 	pnv_php_detach_device_nodes(php_slot->dn);
177 
178 	if (php_slot->fdt) {
179 		kfree(php_slot->dt);
180 		kfree(php_slot->fdt);
181 		php_slot->dt        = NULL;
182 		php_slot->dn->child = NULL;
183 		php_slot->fdt       = NULL;
184 	}
185 }
186 
187 /*
188  * As the nodes in OF changeset are applied in reverse order, we
189  * need revert the nodes in advance so that we have correct node
190  * order after the changeset is applied.
191  */
192 static void pnv_php_reverse_nodes(struct device_node *parent)
193 {
194 	struct device_node *child, *next;
195 
196 	/* In-depth first */
197 	for_each_child_of_node(parent, child)
198 		pnv_php_reverse_nodes(child);
199 
200 	/* Reverse the nodes in the child list */
201 	child = parent->child;
202 	parent->child = NULL;
203 	while (child) {
204 		next = child->sibling;
205 
206 		child->sibling = parent->child;
207 		parent->child = child;
208 		child = next;
209 	}
210 }
211 
212 static int pnv_php_populate_changeset(struct of_changeset *ocs,
213 				      struct device_node *dn)
214 {
215 	struct device_node *child;
216 	int ret = 0;
217 
218 	for_each_child_of_node(dn, child) {
219 		ret = of_changeset_attach_node(ocs, child);
220 		if (ret) {
221 			of_node_put(child);
222 			break;
223 		}
224 
225 		ret = pnv_php_populate_changeset(ocs, child);
226 		if (ret) {
227 			of_node_put(child);
228 			break;
229 		}
230 	}
231 
232 	return ret;
233 }
234 
235 static void *pnv_php_add_one_pdn(struct device_node *dn, void *data)
236 {
237 	struct pci_controller *hose = (struct pci_controller *)data;
238 	struct pci_dn *pdn;
239 
240 	pdn = pci_add_device_node_info(hose, dn);
241 	if (!pdn)
242 		return ERR_PTR(-ENOMEM);
243 
244 	return NULL;
245 }
246 
247 static void pnv_php_add_pdns(struct pnv_php_slot *slot)
248 {
249 	struct pci_controller *hose = pci_bus_to_host(slot->bus);
250 
251 	pci_traverse_device_nodes(slot->dn, pnv_php_add_one_pdn, hose);
252 }
253 
254 static int pnv_php_add_devtree(struct pnv_php_slot *php_slot)
255 {
256 	void *fdt, *fdt1, *dt;
257 	int ret;
258 
259 	/* We don't know the FDT blob size. We try to get it through
260 	 * maximal memory chunk and then copy it to another chunk that
261 	 * fits the real size.
262 	 */
263 	fdt1 = kzalloc(0x10000, GFP_KERNEL);
264 	if (!fdt1) {
265 		ret = -ENOMEM;
266 		goto out;
267 	}
268 
269 	ret = pnv_pci_get_device_tree(php_slot->dn->phandle, fdt1, 0x10000);
270 	if (ret) {
271 		SLOT_WARN(php_slot, "Error %d getting FDT blob\n", ret);
272 		goto free_fdt1;
273 	}
274 
275 	fdt = kmemdup(fdt1, fdt_totalsize(fdt1), GFP_KERNEL);
276 	if (!fdt) {
277 		ret = -ENOMEM;
278 		goto free_fdt1;
279 	}
280 
281 	/* Unflatten device tree blob */
282 	dt = of_fdt_unflatten_tree(fdt, php_slot->dn, NULL);
283 	if (!dt) {
284 		ret = -EINVAL;
285 		SLOT_WARN(php_slot, "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 		SLOT_WARN(php_slot, "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 		SLOT_WARN(php_slot, "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 static inline struct pnv_php_slot *to_pnv_php_slot(struct hotplug_slot *slot)
328 {
329 	return container_of(slot, struct pnv_php_slot, slot);
330 }
331 
332 int pnv_php_set_slot_power_state(struct hotplug_slot *slot,
333 				 uint8_t state)
334 {
335 	struct pnv_php_slot *php_slot = to_pnv_php_slot(slot);
336 	struct opal_msg msg;
337 	int ret;
338 
339 	ret = pnv_pci_set_power_state(php_slot->id, state, &msg);
340 	if (ret > 0) {
341 		if (be64_to_cpu(msg.params[1]) != php_slot->dn->phandle	||
342 		    be64_to_cpu(msg.params[2]) != state) {
343 			SLOT_WARN(php_slot, "Wrong msg (%lld, %lld, %lld)\n",
344 				  be64_to_cpu(msg.params[1]),
345 				  be64_to_cpu(msg.params[2]),
346 				  be64_to_cpu(msg.params[3]));
347 			return -ENOMSG;
348 		}
349 		if (be64_to_cpu(msg.params[3]) != OPAL_SUCCESS) {
350 			ret = -ENODEV;
351 			goto error;
352 		}
353 	} else if (ret < 0) {
354 		goto error;
355 	}
356 
357 	if (state == OPAL_PCI_SLOT_POWER_OFF || state == OPAL_PCI_SLOT_OFFLINE)
358 		pnv_php_rmv_devtree(php_slot);
359 	else
360 		ret = pnv_php_add_devtree(php_slot);
361 
362 	return ret;
363 
364 error:
365 	SLOT_WARN(php_slot, "Error %d powering %s\n",
366 		  ret, (state == OPAL_PCI_SLOT_POWER_ON) ? "on" : "off");
367 	return ret;
368 }
369 EXPORT_SYMBOL_GPL(pnv_php_set_slot_power_state);
370 
371 static int pnv_php_get_power_state(struct hotplug_slot *slot, u8 *state)
372 {
373 	struct pnv_php_slot *php_slot = to_pnv_php_slot(slot);
374 	uint8_t power_state = OPAL_PCI_SLOT_POWER_ON;
375 	int ret;
376 
377 	/*
378 	 * Retrieve power status from firmware. If we fail
379 	 * getting that, the power status fails back to
380 	 * be on.
381 	 */
382 	ret = pnv_pci_get_power_state(php_slot->id, &power_state);
383 	if (ret) {
384 		SLOT_WARN(php_slot, "Error %d getting power status\n",
385 			  ret);
386 	} else {
387 		*state = power_state;
388 	}
389 
390 	return 0;
391 }
392 
393 static int pnv_php_get_adapter_state(struct hotplug_slot *slot, u8 *state)
394 {
395 	struct pnv_php_slot *php_slot = to_pnv_php_slot(slot);
396 	uint8_t presence = OPAL_PCI_SLOT_EMPTY;
397 	int ret;
398 
399 	/*
400 	 * Retrieve presence status from firmware. If we can't
401 	 * get that, it will fail back to be empty.
402 	 */
403 	ret = pnv_pci_get_presence_state(php_slot->id, &presence);
404 	if (ret >= 0) {
405 		*state = presence;
406 		ret = 0;
407 	} else {
408 		SLOT_WARN(php_slot, "Error %d getting presence\n", ret);
409 	}
410 
411 	return ret;
412 }
413 
414 static int pnv_php_get_attention_state(struct hotplug_slot *slot, u8 *state)
415 {
416 	struct pnv_php_slot *php_slot = to_pnv_php_slot(slot);
417 
418 	*state = php_slot->attention_state;
419 	return 0;
420 }
421 
422 static int pnv_php_set_attention_state(struct hotplug_slot *slot, u8 state)
423 {
424 	struct pnv_php_slot *php_slot = to_pnv_php_slot(slot);
425 	struct pci_dev *bridge = php_slot->pdev;
426 	u16 new, mask;
427 
428 	php_slot->attention_state = state;
429 	if (!bridge)
430 		return 0;
431 
432 	mask = PCI_EXP_SLTCTL_AIC;
433 
434 	if (state)
435 		new = PCI_EXP_SLTCTL_ATTN_IND_ON;
436 	else
437 		new = PCI_EXP_SLTCTL_ATTN_IND_OFF;
438 
439 	pcie_capability_clear_and_set_word(bridge, PCI_EXP_SLTCTL, mask, new);
440 
441 	return 0;
442 }
443 
444 static int pnv_php_enable(struct pnv_php_slot *php_slot, bool rescan)
445 {
446 	struct hotplug_slot *slot = &php_slot->slot;
447 	uint8_t presence = OPAL_PCI_SLOT_EMPTY;
448 	uint8_t power_status = OPAL_PCI_SLOT_POWER_ON;
449 	int ret;
450 
451 	/* Check if the slot has been configured */
452 	if (php_slot->state != PNV_PHP_STATE_REGISTERED)
453 		return 0;
454 
455 	/* Retrieve slot presence status */
456 	ret = pnv_php_get_adapter_state(slot, &presence);
457 	if (ret)
458 		return ret;
459 
460 	/*
461 	 * Proceed if there have nothing behind the slot. However,
462 	 * we should leave the slot in registered state at the
463 	 * beginning. Otherwise, the PCI devices inserted afterwards
464 	 * won't be probed and populated.
465 	 */
466 	if (presence == OPAL_PCI_SLOT_EMPTY) {
467 		if (!php_slot->power_state_check) {
468 			php_slot->power_state_check = true;
469 
470 			return 0;
471 		}
472 
473 		goto scan;
474 	}
475 
476 	/*
477 	 * If the power supply to the slot is off, we can't detect
478 	 * adapter presence state. That means we have to turn the
479 	 * slot on before going to probe slot's presence state.
480 	 *
481 	 * On the first time, we don't change the power status to
482 	 * boost system boot with assumption that the firmware
483 	 * supplies consistent slot power status: empty slot always
484 	 * has its power off and non-empty slot has its power on.
485 	 */
486 	if (!php_slot->power_state_check) {
487 		php_slot->power_state_check = true;
488 
489 		ret = pnv_php_get_power_state(slot, &power_status);
490 		if (ret)
491 			return ret;
492 
493 		if (power_status != OPAL_PCI_SLOT_POWER_ON)
494 			return 0;
495 	}
496 
497 	/* Check the power status. Scan the slot if it is already on */
498 	ret = pnv_php_get_power_state(slot, &power_status);
499 	if (ret)
500 		return ret;
501 
502 	if (power_status == OPAL_PCI_SLOT_POWER_ON)
503 		goto scan;
504 
505 	/* Power is off, turn it on and then scan the slot */
506 	ret = pnv_php_set_slot_power_state(slot, OPAL_PCI_SLOT_POWER_ON);
507 	if (ret)
508 		return ret;
509 
510 scan:
511 	if (presence == OPAL_PCI_SLOT_PRESENT) {
512 		if (rescan) {
513 			pci_lock_rescan_remove();
514 			pci_hp_add_devices(php_slot->bus);
515 			pci_unlock_rescan_remove();
516 		}
517 
518 		/* Rescan for child hotpluggable slots */
519 		php_slot->state = PNV_PHP_STATE_POPULATED;
520 		if (rescan)
521 			pnv_php_register(php_slot->dn);
522 	} else {
523 		php_slot->state = PNV_PHP_STATE_POPULATED;
524 	}
525 
526 	return 0;
527 }
528 
529 static int pnv_php_reset_slot(struct hotplug_slot *slot, int probe)
530 {
531 	struct pnv_php_slot *php_slot = to_pnv_php_slot(slot);
532 	struct pci_dev *bridge = php_slot->pdev;
533 	uint16_t sts;
534 
535 	/*
536 	 * The CAPI folks want pnv_php to drive OpenCAPI slots
537 	 * which don't have a bridge. Only claim to support
538 	 * reset_slot() if we have a bridge device (for now...)
539 	 */
540 	if (probe)
541 		return !bridge;
542 
543 	/* mask our interrupt while resetting the bridge */
544 	if (php_slot->irq > 0)
545 		disable_irq(php_slot->irq);
546 
547 	pci_bridge_secondary_bus_reset(bridge);
548 
549 	/* clear any state changes that happened due to the reset */
550 	pcie_capability_read_word(php_slot->pdev, PCI_EXP_SLTSTA, &sts);
551 	sts &= (PCI_EXP_SLTSTA_PDC | PCI_EXP_SLTSTA_DLLSC);
552 	pcie_capability_write_word(php_slot->pdev, PCI_EXP_SLTSTA, sts);
553 
554 	if (php_slot->irq > 0)
555 		enable_irq(php_slot->irq);
556 
557 	return 0;
558 }
559 
560 static int pnv_php_enable_slot(struct hotplug_slot *slot)
561 {
562 	struct pnv_php_slot *php_slot = to_pnv_php_slot(slot);
563 
564 	return pnv_php_enable(php_slot, true);
565 }
566 
567 static int pnv_php_disable_slot(struct hotplug_slot *slot)
568 {
569 	struct pnv_php_slot *php_slot = to_pnv_php_slot(slot);
570 	int ret;
571 
572 	/*
573 	 * Allow to disable a slot already in the registered state to
574 	 * cover cases where the slot couldn't be enabled and never
575 	 * reached the populated state
576 	 */
577 	if (php_slot->state != PNV_PHP_STATE_POPULATED &&
578 	    php_slot->state != PNV_PHP_STATE_REGISTERED)
579 		return 0;
580 
581 	/* Remove all devices behind the slot */
582 	pci_lock_rescan_remove();
583 	pci_hp_remove_devices(php_slot->bus);
584 	pci_unlock_rescan_remove();
585 
586 	/* Detach the child hotpluggable slots */
587 	pnv_php_unregister(php_slot->dn);
588 
589 	/* Notify firmware and remove device nodes */
590 	ret = pnv_php_set_slot_power_state(slot, OPAL_PCI_SLOT_POWER_OFF);
591 
592 	php_slot->state = PNV_PHP_STATE_REGISTERED;
593 	return ret;
594 }
595 
596 static const struct hotplug_slot_ops php_slot_ops = {
597 	.get_power_status	= pnv_php_get_power_state,
598 	.get_adapter_status	= pnv_php_get_adapter_state,
599 	.get_attention_status	= pnv_php_get_attention_state,
600 	.set_attention_status	= pnv_php_set_attention_state,
601 	.enable_slot		= pnv_php_enable_slot,
602 	.disable_slot		= pnv_php_disable_slot,
603 	.reset_slot		= pnv_php_reset_slot,
604 };
605 
606 static void pnv_php_release(struct pnv_php_slot *php_slot)
607 {
608 	unsigned long flags;
609 
610 	/* Remove from global or child list */
611 	spin_lock_irqsave(&pnv_php_lock, flags);
612 	list_del(&php_slot->link);
613 	spin_unlock_irqrestore(&pnv_php_lock, flags);
614 
615 	/* Detach from parent */
616 	pnv_php_put_slot(php_slot);
617 	pnv_php_put_slot(php_slot->parent);
618 }
619 
620 static struct pnv_php_slot *pnv_php_alloc_slot(struct device_node *dn)
621 {
622 	struct pnv_php_slot *php_slot;
623 	struct pci_bus *bus;
624 	const char *label;
625 	uint64_t id;
626 	int ret;
627 
628 	ret = of_property_read_string(dn, "ibm,slot-label", &label);
629 	if (ret)
630 		return NULL;
631 
632 	if (pnv_pci_get_slot_id(dn, &id))
633 		return NULL;
634 
635 	bus = pci_find_bus_by_node(dn);
636 	if (!bus)
637 		return NULL;
638 
639 	php_slot = kzalloc(sizeof(*php_slot), GFP_KERNEL);
640 	if (!php_slot)
641 		return NULL;
642 
643 	php_slot->name = kstrdup(label, GFP_KERNEL);
644 	if (!php_slot->name) {
645 		kfree(php_slot);
646 		return NULL;
647 	}
648 
649 	if (dn->child && PCI_DN(dn->child))
650 		php_slot->slot_no = PCI_SLOT(PCI_DN(dn->child)->devfn);
651 	else
652 		php_slot->slot_no = -1;   /* Placeholder slot */
653 
654 	kref_init(&php_slot->kref);
655 	php_slot->state	                = PNV_PHP_STATE_INITIALIZED;
656 	php_slot->dn	                = dn;
657 	php_slot->pdev	                = bus->self;
658 	php_slot->bus	                = bus;
659 	php_slot->id	                = id;
660 	php_slot->power_state_check     = false;
661 	php_slot->slot.ops              = &php_slot_ops;
662 
663 	INIT_LIST_HEAD(&php_slot->children);
664 	INIT_LIST_HEAD(&php_slot->link);
665 
666 	return php_slot;
667 }
668 
669 static int pnv_php_register_slot(struct pnv_php_slot *php_slot)
670 {
671 	struct pnv_php_slot *parent;
672 	struct device_node *dn = php_slot->dn;
673 	unsigned long flags;
674 	int ret;
675 
676 	/* Check if the slot is registered or not */
677 	parent = pnv_php_find_slot(php_slot->dn);
678 	if (parent) {
679 		pnv_php_put_slot(parent);
680 		return -EEXIST;
681 	}
682 
683 	/* Register PCI slot */
684 	ret = pci_hp_register(&php_slot->slot, php_slot->bus,
685 			      php_slot->slot_no, php_slot->name);
686 	if (ret) {
687 		SLOT_WARN(php_slot, "Error %d registering slot\n", ret);
688 		return ret;
689 	}
690 
691 	/* Attach to the parent's child list or global list */
692 	while ((dn = of_get_parent(dn))) {
693 		if (!PCI_DN(dn)) {
694 			of_node_put(dn);
695 			break;
696 		}
697 
698 		parent = pnv_php_find_slot(dn);
699 		if (parent) {
700 			of_node_put(dn);
701 			break;
702 		}
703 
704 		of_node_put(dn);
705 	}
706 
707 	spin_lock_irqsave(&pnv_php_lock, flags);
708 	php_slot->parent = parent;
709 	if (parent)
710 		list_add_tail(&php_slot->link, &parent->children);
711 	else
712 		list_add_tail(&php_slot->link, &pnv_php_slot_list);
713 	spin_unlock_irqrestore(&pnv_php_lock, flags);
714 
715 	php_slot->state = PNV_PHP_STATE_REGISTERED;
716 	return 0;
717 }
718 
719 static int pnv_php_enable_msix(struct pnv_php_slot *php_slot)
720 {
721 	struct pci_dev *pdev = php_slot->pdev;
722 	struct msix_entry entry;
723 	int nr_entries, ret;
724 	u16 pcie_flag;
725 
726 	/* Get total number of MSIx entries */
727 	nr_entries = pci_msix_vec_count(pdev);
728 	if (nr_entries < 0)
729 		return nr_entries;
730 
731 	/* Check hotplug MSIx entry is in range */
732 	pcie_capability_read_word(pdev, PCI_EXP_FLAGS, &pcie_flag);
733 	entry.entry = (pcie_flag & PCI_EXP_FLAGS_IRQ) >> 9;
734 	if (entry.entry >= nr_entries)
735 		return -ERANGE;
736 
737 	/* Enable MSIx */
738 	ret = pci_enable_msix_exact(pdev, &entry, 1);
739 	if (ret) {
740 		SLOT_WARN(php_slot, "Error %d enabling MSIx\n", ret);
741 		return ret;
742 	}
743 
744 	return entry.vector;
745 }
746 
747 static void pnv_php_event_handler(struct work_struct *work)
748 {
749 	struct pnv_php_event *event =
750 		container_of(work, struct pnv_php_event, work);
751 	struct pnv_php_slot *php_slot = event->php_slot;
752 
753 	if (event->added)
754 		pnv_php_enable_slot(&php_slot->slot);
755 	else
756 		pnv_php_disable_slot(&php_slot->slot);
757 
758 	kfree(event);
759 }
760 
761 static irqreturn_t pnv_php_interrupt(int irq, void *data)
762 {
763 	struct pnv_php_slot *php_slot = data;
764 	struct pci_dev *pchild, *pdev = php_slot->pdev;
765 	struct eeh_dev *edev;
766 	struct eeh_pe *pe;
767 	struct pnv_php_event *event;
768 	u16 sts, lsts;
769 	u8 presence;
770 	bool added;
771 	unsigned long flags;
772 	int ret;
773 
774 	pcie_capability_read_word(pdev, PCI_EXP_SLTSTA, &sts);
775 	sts &= (PCI_EXP_SLTSTA_PDC | PCI_EXP_SLTSTA_DLLSC);
776 	pcie_capability_write_word(pdev, PCI_EXP_SLTSTA, sts);
777 
778 	pci_dbg(pdev, "PCI slot [%s]: HP int! DLAct: %d, PresDet: %d\n",
779 			php_slot->name,
780 			!!(sts & PCI_EXP_SLTSTA_DLLSC),
781 			!!(sts & PCI_EXP_SLTSTA_PDC));
782 
783 	if (sts & PCI_EXP_SLTSTA_DLLSC) {
784 		pcie_capability_read_word(pdev, PCI_EXP_LNKSTA, &lsts);
785 		added = !!(lsts & PCI_EXP_LNKSTA_DLLLA);
786 	} else if (!(php_slot->flags & PNV_PHP_FLAG_BROKEN_PDC) &&
787 		   (sts & PCI_EXP_SLTSTA_PDC)) {
788 		ret = pnv_pci_get_presence_state(php_slot->id, &presence);
789 		if (ret) {
790 			SLOT_WARN(php_slot,
791 				  "PCI slot [%s] error %d getting presence (0x%04x), to retry the operation.\n",
792 				  php_slot->name, ret, sts);
793 			return IRQ_HANDLED;
794 		}
795 
796 		added = !!(presence == OPAL_PCI_SLOT_PRESENT);
797 	} else {
798 		pci_dbg(pdev, "PCI slot [%s]: Spurious IRQ?\n", php_slot->name);
799 		return IRQ_NONE;
800 	}
801 
802 	/* Freeze the removed PE to avoid unexpected error reporting */
803 	if (!added) {
804 		pchild = list_first_entry_or_null(&php_slot->bus->devices,
805 						  struct pci_dev, bus_list);
806 		edev = pchild ? pci_dev_to_eeh_dev(pchild) : NULL;
807 		pe = edev ? edev->pe : NULL;
808 		if (pe) {
809 			eeh_serialize_lock(&flags);
810 			eeh_pe_mark_isolated(pe);
811 			eeh_serialize_unlock(flags);
812 			eeh_pe_set_option(pe, EEH_OPT_FREEZE_PE);
813 		}
814 	}
815 
816 	/*
817 	 * The PE is left in frozen state if the event is missed. It's
818 	 * fine as the PCI devices (PE) aren't functional any more.
819 	 */
820 	event = kzalloc(sizeof(*event), GFP_ATOMIC);
821 	if (!event) {
822 		SLOT_WARN(php_slot,
823 			  "PCI slot [%s] missed hotplug event 0x%04x\n",
824 			  php_slot->name, sts);
825 		return IRQ_HANDLED;
826 	}
827 
828 	pci_info(pdev, "PCI slot [%s] %s (IRQ: %d)\n",
829 		 php_slot->name, added ? "added" : "removed", irq);
830 	INIT_WORK(&event->work, pnv_php_event_handler);
831 	event->added = added;
832 	event->php_slot = php_slot;
833 	queue_work(php_slot->wq, &event->work);
834 
835 	return IRQ_HANDLED;
836 }
837 
838 static void pnv_php_init_irq(struct pnv_php_slot *php_slot, int irq)
839 {
840 	struct pci_dev *pdev = php_slot->pdev;
841 	u32 broken_pdc = 0;
842 	u16 sts, ctrl;
843 	int ret;
844 
845 	/* Allocate workqueue */
846 	php_slot->wq = alloc_workqueue("pciehp-%s", 0, 0, php_slot->name);
847 	if (!php_slot->wq) {
848 		SLOT_WARN(php_slot, "Cannot alloc workqueue\n");
849 		pnv_php_disable_irq(php_slot, true);
850 		return;
851 	}
852 
853 	/* Check PDC (Presence Detection Change) is broken or not */
854 	ret = of_property_read_u32(php_slot->dn, "ibm,slot-broken-pdc",
855 				   &broken_pdc);
856 	if (!ret && broken_pdc)
857 		php_slot->flags |= PNV_PHP_FLAG_BROKEN_PDC;
858 
859 	/* Clear pending interrupts */
860 	pcie_capability_read_word(pdev, PCI_EXP_SLTSTA, &sts);
861 	if (php_slot->flags & PNV_PHP_FLAG_BROKEN_PDC)
862 		sts |= PCI_EXP_SLTSTA_DLLSC;
863 	else
864 		sts |= (PCI_EXP_SLTSTA_PDC | PCI_EXP_SLTSTA_DLLSC);
865 	pcie_capability_write_word(pdev, PCI_EXP_SLTSTA, sts);
866 
867 	/* Request the interrupt */
868 	ret = request_irq(irq, pnv_php_interrupt, IRQF_SHARED,
869 			  php_slot->name, php_slot);
870 	if (ret) {
871 		pnv_php_disable_irq(php_slot, true);
872 		SLOT_WARN(php_slot, "Error %d enabling IRQ %d\n", ret, irq);
873 		return;
874 	}
875 
876 	/* Enable the interrupts */
877 	pcie_capability_read_word(pdev, PCI_EXP_SLTCTL, &ctrl);
878 	if (php_slot->flags & PNV_PHP_FLAG_BROKEN_PDC) {
879 		ctrl &= ~PCI_EXP_SLTCTL_PDCE;
880 		ctrl |= (PCI_EXP_SLTCTL_HPIE |
881 			 PCI_EXP_SLTCTL_DLLSCE);
882 	} else {
883 		ctrl |= (PCI_EXP_SLTCTL_HPIE |
884 			 PCI_EXP_SLTCTL_PDCE |
885 			 PCI_EXP_SLTCTL_DLLSCE);
886 	}
887 	pcie_capability_write_word(pdev, PCI_EXP_SLTCTL, ctrl);
888 
889 	/* The interrupt is initialized successfully when @irq is valid */
890 	php_slot->irq = irq;
891 }
892 
893 static void pnv_php_enable_irq(struct pnv_php_slot *php_slot)
894 {
895 	struct pci_dev *pdev = php_slot->pdev;
896 	int irq, ret;
897 
898 	/*
899 	 * The MSI/MSIx interrupt might have been occupied by other
900 	 * drivers. Don't populate the surprise hotplug capability
901 	 * in that case.
902 	 */
903 	if (pci_dev_msi_enabled(pdev))
904 		return;
905 
906 	ret = pci_enable_device(pdev);
907 	if (ret) {
908 		SLOT_WARN(php_slot, "Error %d enabling device\n", ret);
909 		return;
910 	}
911 
912 	pci_set_master(pdev);
913 
914 	/* Enable MSIx interrupt */
915 	irq = pnv_php_enable_msix(php_slot);
916 	if (irq > 0) {
917 		pnv_php_init_irq(php_slot, irq);
918 		return;
919 	}
920 
921 	/*
922 	 * Use MSI if MSIx doesn't work. Fail back to legacy INTx
923 	 * if MSI doesn't work either
924 	 */
925 	ret = pci_enable_msi(pdev);
926 	if (!ret || pdev->irq) {
927 		irq = pdev->irq;
928 		pnv_php_init_irq(php_slot, irq);
929 	}
930 }
931 
932 static int pnv_php_register_one(struct device_node *dn)
933 {
934 	struct pnv_php_slot *php_slot;
935 	u32 prop32;
936 	int ret;
937 
938 	/* Check if it's hotpluggable slot */
939 	ret = of_property_read_u32(dn, "ibm,slot-pluggable", &prop32);
940 	if (ret || !prop32)
941 		return -ENXIO;
942 
943 	ret = of_property_read_u32(dn, "ibm,reset-by-firmware", &prop32);
944 	if (ret || !prop32)
945 		return -ENXIO;
946 
947 	php_slot = pnv_php_alloc_slot(dn);
948 	if (!php_slot)
949 		return -ENODEV;
950 
951 	ret = pnv_php_register_slot(php_slot);
952 	if (ret)
953 		goto free_slot;
954 
955 	ret = pnv_php_enable(php_slot, false);
956 	if (ret)
957 		goto unregister_slot;
958 
959 	/* Enable interrupt if the slot supports surprise hotplug */
960 	ret = of_property_read_u32(dn, "ibm,slot-surprise-pluggable", &prop32);
961 	if (!ret && prop32)
962 		pnv_php_enable_irq(php_slot);
963 
964 	return 0;
965 
966 unregister_slot:
967 	pnv_php_unregister_one(php_slot->dn);
968 free_slot:
969 	pnv_php_put_slot(php_slot);
970 	return ret;
971 }
972 
973 static void pnv_php_register(struct device_node *dn)
974 {
975 	struct device_node *child;
976 
977 	/*
978 	 * The parent slots should be registered before their
979 	 * child slots.
980 	 */
981 	for_each_child_of_node(dn, child) {
982 		pnv_php_register_one(child);
983 		pnv_php_register(child);
984 	}
985 }
986 
987 static void pnv_php_unregister_one(struct device_node *dn)
988 {
989 	struct pnv_php_slot *php_slot;
990 
991 	php_slot = pnv_php_find_slot(dn);
992 	if (!php_slot)
993 		return;
994 
995 	php_slot->state = PNV_PHP_STATE_OFFLINE;
996 	pci_hp_deregister(&php_slot->slot);
997 	pnv_php_release(php_slot);
998 	pnv_php_put_slot(php_slot);
999 }
1000 
1001 static void pnv_php_unregister(struct device_node *dn)
1002 {
1003 	struct device_node *child;
1004 
1005 	/* The child slots should go before their parent slots */
1006 	for_each_child_of_node(dn, child) {
1007 		pnv_php_unregister(child);
1008 		pnv_php_unregister_one(child);
1009 	}
1010 }
1011 
1012 static int __init pnv_php_init(void)
1013 {
1014 	struct device_node *dn;
1015 
1016 	pr_info(DRIVER_DESC " version: " DRIVER_VERSION "\n");
1017 	for_each_compatible_node(dn, NULL, "ibm,ioda2-phb")
1018 		pnv_php_register(dn);
1019 
1020 	for_each_compatible_node(dn, NULL, "ibm,ioda3-phb")
1021 		pnv_php_register(dn);
1022 
1023 	for_each_compatible_node(dn, NULL, "ibm,ioda2-npu2-opencapi-phb")
1024 		pnv_php_register_one(dn); /* slot directly under the PHB */
1025 	return 0;
1026 }
1027 
1028 static void __exit pnv_php_exit(void)
1029 {
1030 	struct device_node *dn;
1031 
1032 	for_each_compatible_node(dn, NULL, "ibm,ioda2-phb")
1033 		pnv_php_unregister(dn);
1034 
1035 	for_each_compatible_node(dn, NULL, "ibm,ioda3-phb")
1036 		pnv_php_unregister(dn);
1037 
1038 	for_each_compatible_node(dn, NULL, "ibm,ioda2-npu2-opencapi-phb")
1039 		pnv_php_unregister_one(dn); /* slot directly under the PHB */
1040 }
1041 
1042 module_init(pnv_php_init);
1043 module_exit(pnv_php_exit);
1044 
1045 MODULE_VERSION(DRIVER_VERSION);
1046 MODULE_LICENSE("GPL v2");
1047 MODULE_AUTHOR(DRIVER_AUTHOR);
1048 MODULE_DESCRIPTION(DRIVER_DESC);
1049