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/bitfield.h>
9 #include <linux/libfdt.h>
10 #include <linux/module.h>
11 #include <linux/pci.h>
12 #include <linux/pci_hotplug.h>
13 #include <linux/of_fdt.h>
14
15 #include <asm/opal.h>
16 #include <asm/pnv-pci.h>
17 #include <asm/ppc-pci.h>
18
19 #define DRIVER_VERSION "0.1"
20 #define DRIVER_AUTHOR "Gavin Shan, IBM Corporation"
21 #define DRIVER_DESC "PowerPC PowerNV PCI Hotplug Driver"
22
23 #define SLOT_WARN(sl, x...) \
24 ((sl)->pdev ? pci_warn((sl)->pdev, x) : dev_warn(&(sl)->bus->dev, x))
25
26 struct pnv_php_event {
27 bool added;
28 struct pnv_php_slot *php_slot;
29 struct work_struct work;
30 };
31
32 static LIST_HEAD(pnv_php_slot_list);
33 static DEFINE_SPINLOCK(pnv_php_lock);
34
35 static void pnv_php_register(struct device_node *dn);
36 static void pnv_php_unregister_one(struct device_node *dn);
37 static void pnv_php_unregister(struct device_node *dn);
38
pnv_php_disable_irq(struct pnv_php_slot * php_slot,bool disable_device)39 static void pnv_php_disable_irq(struct pnv_php_slot *php_slot,
40 bool disable_device)
41 {
42 struct pci_dev *pdev = php_slot->pdev;
43 u16 ctrl;
44
45 if (php_slot->irq > 0) {
46 pcie_capability_read_word(pdev, PCI_EXP_SLTCTL, &ctrl);
47 ctrl &= ~(PCI_EXP_SLTCTL_HPIE |
48 PCI_EXP_SLTCTL_PDCE |
49 PCI_EXP_SLTCTL_DLLSCE);
50 pcie_capability_write_word(pdev, PCI_EXP_SLTCTL, ctrl);
51
52 free_irq(php_slot->irq, php_slot);
53 php_slot->irq = 0;
54 }
55
56 if (php_slot->wq) {
57 destroy_workqueue(php_slot->wq);
58 php_slot->wq = NULL;
59 }
60
61 if (disable_device) {
62 if (pdev->msix_enabled)
63 pci_disable_msix(pdev);
64 else if (pdev->msi_enabled)
65 pci_disable_msi(pdev);
66
67 pci_disable_device(pdev);
68 }
69 }
70
pnv_php_free_slot(struct kref * kref)71 static void pnv_php_free_slot(struct kref *kref)
72 {
73 struct pnv_php_slot *php_slot = container_of(kref,
74 struct pnv_php_slot, kref);
75
76 WARN_ON(!list_empty(&php_slot->children));
77 pnv_php_disable_irq(php_slot, false);
78 kfree(php_slot->name);
79 kfree(php_slot);
80 }
81
pnv_php_put_slot(struct pnv_php_slot * php_slot)82 static inline void pnv_php_put_slot(struct pnv_php_slot *php_slot)
83 {
84
85 if (!php_slot)
86 return;
87
88 kref_put(&php_slot->kref, pnv_php_free_slot);
89 }
90
pnv_php_match(struct device_node * dn,struct pnv_php_slot * php_slot)91 static struct pnv_php_slot *pnv_php_match(struct device_node *dn,
92 struct pnv_php_slot *php_slot)
93 {
94 struct pnv_php_slot *target, *tmp;
95
96 if (php_slot->dn == dn) {
97 kref_get(&php_slot->kref);
98 return php_slot;
99 }
100
101 list_for_each_entry(tmp, &php_slot->children, link) {
102 target = pnv_php_match(dn, tmp);
103 if (target)
104 return target;
105 }
106
107 return NULL;
108 }
109
pnv_php_find_slot(struct device_node * dn)110 struct pnv_php_slot *pnv_php_find_slot(struct device_node *dn)
111 {
112 struct pnv_php_slot *php_slot, *tmp;
113 unsigned long flags;
114
115 spin_lock_irqsave(&pnv_php_lock, flags);
116 list_for_each_entry(tmp, &pnv_php_slot_list, link) {
117 php_slot = pnv_php_match(dn, tmp);
118 if (php_slot) {
119 spin_unlock_irqrestore(&pnv_php_lock, flags);
120 return php_slot;
121 }
122 }
123 spin_unlock_irqrestore(&pnv_php_lock, flags);
124
125 return NULL;
126 }
127 EXPORT_SYMBOL_GPL(pnv_php_find_slot);
128
129 /*
130 * Remove pdn for all children of the indicated device node.
131 * The function should remove pdn in a depth-first manner.
132 */
pnv_php_rmv_pdns(struct device_node * dn)133 static void pnv_php_rmv_pdns(struct device_node *dn)
134 {
135 struct device_node *child;
136
137 for_each_child_of_node(dn, child) {
138 pnv_php_rmv_pdns(child);
139
140 pci_remove_device_node_info(child);
141 }
142 }
143
144 /*
145 * Detach all child nodes of the indicated device nodes. The
146 * function should handle device nodes in depth-first manner.
147 *
148 * We should not invoke of_node_release() as the memory for
149 * individual device node is part of large memory block. The
150 * large block is allocated from memblock (system bootup) or
151 * kmalloc() when unflattening the device tree by OF changeset.
152 * We can not free the large block allocated from memblock. For
153 * later case, it should be released at once.
154 */
pnv_php_detach_device_nodes(struct device_node * parent)155 static void pnv_php_detach_device_nodes(struct device_node *parent)
156 {
157 struct device_node *dn;
158
159 for_each_child_of_node(parent, dn) {
160 pnv_php_detach_device_nodes(dn);
161
162 of_node_put(dn);
163 of_detach_node(dn);
164 }
165 }
166
pnv_php_rmv_devtree(struct pnv_php_slot * php_slot)167 static void pnv_php_rmv_devtree(struct pnv_php_slot *php_slot)
168 {
169 pnv_php_rmv_pdns(php_slot->dn);
170
171 /*
172 * Decrease the refcount if the device nodes were created
173 * through OF changeset before detaching them.
174 */
175 if (php_slot->fdt)
176 of_changeset_destroy(&php_slot->ocs);
177 pnv_php_detach_device_nodes(php_slot->dn);
178
179 if (php_slot->fdt) {
180 kfree(php_slot->dt);
181 kfree(php_slot->fdt);
182 php_slot->dt = NULL;
183 php_slot->dn->child = NULL;
184 php_slot->fdt = NULL;
185 }
186 }
187
188 /*
189 * As the nodes in OF changeset are applied in reverse order, we
190 * need revert the nodes in advance so that we have correct node
191 * order after the changeset is applied.
192 */
pnv_php_reverse_nodes(struct device_node * parent)193 static void pnv_php_reverse_nodes(struct device_node *parent)
194 {
195 struct device_node *child, *next;
196
197 /* In-depth first */
198 for_each_child_of_node(parent, child)
199 pnv_php_reverse_nodes(child);
200
201 /* Reverse the nodes in the child list */
202 child = parent->child;
203 parent->child = NULL;
204 while (child) {
205 next = child->sibling;
206
207 child->sibling = parent->child;
208 parent->child = child;
209 child = next;
210 }
211 }
212
pnv_php_populate_changeset(struct of_changeset * ocs,struct device_node * dn)213 static int pnv_php_populate_changeset(struct of_changeset *ocs,
214 struct device_node *dn)
215 {
216 struct device_node *child;
217 int ret = 0;
218
219 for_each_child_of_node(dn, child) {
220 ret = of_changeset_attach_node(ocs, child);
221 if (ret) {
222 of_node_put(child);
223 break;
224 }
225
226 ret = pnv_php_populate_changeset(ocs, child);
227 if (ret) {
228 of_node_put(child);
229 break;
230 }
231 }
232
233 return ret;
234 }
235
pnv_php_add_one_pdn(struct device_node * dn,void * data)236 static void *pnv_php_add_one_pdn(struct device_node *dn, void *data)
237 {
238 struct pci_controller *hose = (struct pci_controller *)data;
239 struct pci_dn *pdn;
240
241 pdn = pci_add_device_node_info(hose, dn);
242 if (!pdn)
243 return ERR_PTR(-ENOMEM);
244
245 return NULL;
246 }
247
pnv_php_add_pdns(struct pnv_php_slot * slot)248 static void pnv_php_add_pdns(struct pnv_php_slot *slot)
249 {
250 struct pci_controller *hose = pci_bus_to_host(slot->bus);
251
252 pci_traverse_device_nodes(slot->dn, pnv_php_add_one_pdn, hose);
253 }
254
pnv_php_add_devtree(struct pnv_php_slot * php_slot)255 static int pnv_php_add_devtree(struct pnv_php_slot *php_slot)
256 {
257 void *fdt, *fdt1, *dt;
258 int ret;
259
260 /* We don't know the FDT blob size. We try to get it through
261 * maximal memory chunk and then copy it to another chunk that
262 * fits the real size.
263 */
264 fdt1 = kzalloc(0x10000, GFP_KERNEL);
265 if (!fdt1) {
266 ret = -ENOMEM;
267 goto out;
268 }
269
270 ret = pnv_pci_get_device_tree(php_slot->dn->phandle, fdt1, 0x10000);
271 if (ret) {
272 SLOT_WARN(php_slot, "Error %d getting FDT blob\n", ret);
273 goto free_fdt1;
274 }
275
276 fdt = kmemdup(fdt1, fdt_totalsize(fdt1), GFP_KERNEL);
277 if (!fdt) {
278 ret = -ENOMEM;
279 goto free_fdt1;
280 }
281
282 /* Unflatten device tree blob */
283 dt = of_fdt_unflatten_tree(fdt, php_slot->dn, NULL);
284 if (!dt) {
285 ret = -EINVAL;
286 SLOT_WARN(php_slot, "Cannot unflatten FDT\n");
287 goto free_fdt;
288 }
289
290 /* Initialize and apply the changeset */
291 of_changeset_init(&php_slot->ocs);
292 pnv_php_reverse_nodes(php_slot->dn);
293 ret = pnv_php_populate_changeset(&php_slot->ocs, php_slot->dn);
294 if (ret) {
295 pnv_php_reverse_nodes(php_slot->dn);
296 SLOT_WARN(php_slot, "Error %d populating changeset\n",
297 ret);
298 goto free_dt;
299 }
300
301 php_slot->dn->child = NULL;
302 ret = of_changeset_apply(&php_slot->ocs);
303 if (ret) {
304 SLOT_WARN(php_slot, "Error %d applying changeset\n", ret);
305 goto destroy_changeset;
306 }
307
308 /* Add device node firmware data */
309 pnv_php_add_pdns(php_slot);
310 php_slot->fdt = fdt;
311 php_slot->dt = dt;
312 kfree(fdt1);
313 goto out;
314
315 destroy_changeset:
316 of_changeset_destroy(&php_slot->ocs);
317 free_dt:
318 kfree(dt);
319 php_slot->dn->child = NULL;
320 free_fdt:
321 kfree(fdt);
322 free_fdt1:
323 kfree(fdt1);
324 out:
325 return ret;
326 }
327
to_pnv_php_slot(struct hotplug_slot * slot)328 static inline struct pnv_php_slot *to_pnv_php_slot(struct hotplug_slot *slot)
329 {
330 return container_of(slot, struct pnv_php_slot, slot);
331 }
332
pnv_php_set_slot_power_state(struct hotplug_slot * slot,uint8_t state)333 int pnv_php_set_slot_power_state(struct hotplug_slot *slot,
334 uint8_t state)
335 {
336 struct pnv_php_slot *php_slot = to_pnv_php_slot(slot);
337 struct opal_msg msg;
338 int ret;
339
340 ret = pnv_pci_set_power_state(php_slot->id, state, &msg);
341 if (ret > 0) {
342 if (be64_to_cpu(msg.params[1]) != php_slot->dn->phandle ||
343 be64_to_cpu(msg.params[2]) != state) {
344 SLOT_WARN(php_slot, "Wrong msg (%lld, %lld, %lld)\n",
345 be64_to_cpu(msg.params[1]),
346 be64_to_cpu(msg.params[2]),
347 be64_to_cpu(msg.params[3]));
348 return -ENOMSG;
349 }
350 if (be64_to_cpu(msg.params[3]) != OPAL_SUCCESS) {
351 ret = -ENODEV;
352 goto error;
353 }
354 } else if (ret < 0) {
355 goto error;
356 }
357
358 if (state == OPAL_PCI_SLOT_POWER_OFF || state == OPAL_PCI_SLOT_OFFLINE)
359 pnv_php_rmv_devtree(php_slot);
360 else
361 ret = pnv_php_add_devtree(php_slot);
362
363 return ret;
364
365 error:
366 SLOT_WARN(php_slot, "Error %d powering %s\n",
367 ret, (state == OPAL_PCI_SLOT_POWER_ON) ? "on" : "off");
368 return ret;
369 }
370 EXPORT_SYMBOL_GPL(pnv_php_set_slot_power_state);
371
pnv_php_get_power_state(struct hotplug_slot * slot,u8 * state)372 static int pnv_php_get_power_state(struct hotplug_slot *slot, u8 *state)
373 {
374 struct pnv_php_slot *php_slot = to_pnv_php_slot(slot);
375 uint8_t power_state = OPAL_PCI_SLOT_POWER_ON;
376 int ret;
377
378 /*
379 * Retrieve power status from firmware. If we fail
380 * getting that, the power status fails back to
381 * be on.
382 */
383 ret = pnv_pci_get_power_state(php_slot->id, &power_state);
384 if (ret) {
385 SLOT_WARN(php_slot, "Error %d getting power status\n",
386 ret);
387 } else {
388 *state = power_state;
389 }
390
391 return 0;
392 }
393
pnv_php_get_adapter_state(struct hotplug_slot * slot,u8 * state)394 static int pnv_php_get_adapter_state(struct hotplug_slot *slot, u8 *state)
395 {
396 struct pnv_php_slot *php_slot = to_pnv_php_slot(slot);
397 uint8_t presence = OPAL_PCI_SLOT_EMPTY;
398 int ret;
399
400 /*
401 * Retrieve presence status from firmware. If we can't
402 * get that, it will fail back to be empty.
403 */
404 ret = pnv_pci_get_presence_state(php_slot->id, &presence);
405 if (ret >= 0) {
406 *state = presence;
407 ret = 0;
408 } else {
409 SLOT_WARN(php_slot, "Error %d getting presence\n", ret);
410 }
411
412 return ret;
413 }
414
pnv_php_get_attention_state(struct hotplug_slot * slot,u8 * state)415 static int pnv_php_get_attention_state(struct hotplug_slot *slot, u8 *state)
416 {
417 struct pnv_php_slot *php_slot = to_pnv_php_slot(slot);
418
419 *state = php_slot->attention_state;
420 return 0;
421 }
422
pnv_php_set_attention_state(struct hotplug_slot * slot,u8 state)423 static int pnv_php_set_attention_state(struct hotplug_slot *slot, u8 state)
424 {
425 struct pnv_php_slot *php_slot = to_pnv_php_slot(slot);
426 struct pci_dev *bridge = php_slot->pdev;
427 u16 new, mask;
428
429 php_slot->attention_state = state;
430 if (!bridge)
431 return 0;
432
433 mask = PCI_EXP_SLTCTL_AIC;
434
435 if (state)
436 new = PCI_EXP_SLTCTL_ATTN_IND_ON;
437 else
438 new = PCI_EXP_SLTCTL_ATTN_IND_OFF;
439
440 pcie_capability_clear_and_set_word(bridge, PCI_EXP_SLTCTL, mask, new);
441
442 return 0;
443 }
444
pnv_php_enable(struct pnv_php_slot * php_slot,bool rescan)445 static int pnv_php_enable(struct pnv_php_slot *php_slot, bool rescan)
446 {
447 struct hotplug_slot *slot = &php_slot->slot;
448 uint8_t presence = OPAL_PCI_SLOT_EMPTY;
449 uint8_t power_status = OPAL_PCI_SLOT_POWER_ON;
450 int ret;
451
452 /* Check if the slot has been configured */
453 if (php_slot->state != PNV_PHP_STATE_REGISTERED)
454 return 0;
455
456 /* Retrieve slot presence status */
457 ret = pnv_php_get_adapter_state(slot, &presence);
458 if (ret)
459 return ret;
460
461 /*
462 * Proceed if there have nothing behind the slot. However,
463 * we should leave the slot in registered state at the
464 * beginning. Otherwise, the PCI devices inserted afterwards
465 * won't be probed and populated.
466 */
467 if (presence == OPAL_PCI_SLOT_EMPTY) {
468 if (!php_slot->power_state_check) {
469 php_slot->power_state_check = true;
470
471 return 0;
472 }
473
474 goto scan;
475 }
476
477 /*
478 * If the power supply to the slot is off, we can't detect
479 * adapter presence state. That means we have to turn the
480 * slot on before going to probe slot's presence state.
481 *
482 * On the first time, we don't change the power status to
483 * boost system boot with assumption that the firmware
484 * supplies consistent slot power status: empty slot always
485 * has its power off and non-empty slot has its power on.
486 */
487 if (!php_slot->power_state_check) {
488 php_slot->power_state_check = true;
489
490 ret = pnv_php_get_power_state(slot, &power_status);
491 if (ret)
492 return ret;
493
494 if (power_status != OPAL_PCI_SLOT_POWER_ON)
495 return 0;
496 }
497
498 /* Check the power status. Scan the slot if it is already on */
499 ret = pnv_php_get_power_state(slot, &power_status);
500 if (ret)
501 return ret;
502
503 if (power_status == OPAL_PCI_SLOT_POWER_ON)
504 goto scan;
505
506 /* Power is off, turn it on and then scan the slot */
507 ret = pnv_php_set_slot_power_state(slot, OPAL_PCI_SLOT_POWER_ON);
508 if (ret)
509 return ret;
510
511 scan:
512 if (presence == OPAL_PCI_SLOT_PRESENT) {
513 if (rescan) {
514 pci_lock_rescan_remove();
515 pci_hp_add_devices(php_slot->bus);
516 pci_unlock_rescan_remove();
517 }
518
519 /* Rescan for child hotpluggable slots */
520 php_slot->state = PNV_PHP_STATE_POPULATED;
521 if (rescan)
522 pnv_php_register(php_slot->dn);
523 } else {
524 php_slot->state = PNV_PHP_STATE_POPULATED;
525 }
526
527 return 0;
528 }
529
pnv_php_reset_slot(struct hotplug_slot * slot,bool probe)530 static int pnv_php_reset_slot(struct hotplug_slot *slot, bool probe)
531 {
532 struct pnv_php_slot *php_slot = to_pnv_php_slot(slot);
533 struct pci_dev *bridge = php_slot->pdev;
534 uint16_t sts;
535
536 /*
537 * The CAPI folks want pnv_php to drive OpenCAPI slots
538 * which don't have a bridge. Only claim to support
539 * reset_slot() if we have a bridge device (for now...)
540 */
541 if (probe)
542 return !bridge;
543
544 /* mask our interrupt while resetting the bridge */
545 if (php_slot->irq > 0)
546 disable_irq(php_slot->irq);
547
548 pci_bridge_secondary_bus_reset(bridge);
549
550 /* clear any state changes that happened due to the reset */
551 pcie_capability_read_word(php_slot->pdev, PCI_EXP_SLTSTA, &sts);
552 sts &= (PCI_EXP_SLTSTA_PDC | PCI_EXP_SLTSTA_DLLSC);
553 pcie_capability_write_word(php_slot->pdev, PCI_EXP_SLTSTA, sts);
554
555 if (php_slot->irq > 0)
556 enable_irq(php_slot->irq);
557
558 return 0;
559 }
560
pnv_php_enable_slot(struct hotplug_slot * slot)561 static int pnv_php_enable_slot(struct hotplug_slot *slot)
562 {
563 struct pnv_php_slot *php_slot = to_pnv_php_slot(slot);
564
565 return pnv_php_enable(php_slot, true);
566 }
567
pnv_php_disable_slot(struct hotplug_slot * slot)568 static int pnv_php_disable_slot(struct hotplug_slot *slot)
569 {
570 struct pnv_php_slot *php_slot = to_pnv_php_slot(slot);
571 int ret;
572
573 /*
574 * Allow to disable a slot already in the registered state to
575 * cover cases where the slot couldn't be enabled and never
576 * reached the populated state
577 */
578 if (php_slot->state != PNV_PHP_STATE_POPULATED &&
579 php_slot->state != PNV_PHP_STATE_REGISTERED)
580 return 0;
581
582 /* Remove all devices behind the slot */
583 pci_lock_rescan_remove();
584 pci_hp_remove_devices(php_slot->bus);
585 pci_unlock_rescan_remove();
586
587 /* Detach the child hotpluggable slots */
588 pnv_php_unregister(php_slot->dn);
589
590 /* Notify firmware and remove device nodes */
591 ret = pnv_php_set_slot_power_state(slot, OPAL_PCI_SLOT_POWER_OFF);
592
593 php_slot->state = PNV_PHP_STATE_REGISTERED;
594 return ret;
595 }
596
597 static const struct hotplug_slot_ops php_slot_ops = {
598 .get_power_status = pnv_php_get_power_state,
599 .get_adapter_status = pnv_php_get_adapter_state,
600 .get_attention_status = pnv_php_get_attention_state,
601 .set_attention_status = pnv_php_set_attention_state,
602 .enable_slot = pnv_php_enable_slot,
603 .disable_slot = pnv_php_disable_slot,
604 .reset_slot = pnv_php_reset_slot,
605 };
606
pnv_php_release(struct pnv_php_slot * php_slot)607 static void pnv_php_release(struct pnv_php_slot *php_slot)
608 {
609 unsigned long flags;
610
611 /* Remove from global or child list */
612 spin_lock_irqsave(&pnv_php_lock, flags);
613 list_del(&php_slot->link);
614 spin_unlock_irqrestore(&pnv_php_lock, flags);
615
616 /* Detach from parent */
617 pnv_php_put_slot(php_slot);
618 pnv_php_put_slot(php_slot->parent);
619 }
620
pnv_php_alloc_slot(struct device_node * dn)621 static struct pnv_php_slot *pnv_php_alloc_slot(struct device_node *dn)
622 {
623 struct pnv_php_slot *php_slot;
624 struct pci_bus *bus;
625 const char *label;
626 uint64_t id;
627 int ret;
628
629 ret = of_property_read_string(dn, "ibm,slot-label", &label);
630 if (ret)
631 return NULL;
632
633 if (pnv_pci_get_slot_id(dn, &id))
634 return NULL;
635
636 bus = pci_find_bus_by_node(dn);
637 if (!bus)
638 return NULL;
639
640 php_slot = kzalloc(sizeof(*php_slot), GFP_KERNEL);
641 if (!php_slot)
642 return NULL;
643
644 php_slot->name = kstrdup(label, GFP_KERNEL);
645 if (!php_slot->name) {
646 kfree(php_slot);
647 return NULL;
648 }
649
650 if (dn->child && PCI_DN(dn->child))
651 php_slot->slot_no = PCI_SLOT(PCI_DN(dn->child)->devfn);
652 else
653 php_slot->slot_no = -1; /* Placeholder slot */
654
655 kref_init(&php_slot->kref);
656 php_slot->state = PNV_PHP_STATE_INITIALIZED;
657 php_slot->dn = dn;
658 php_slot->pdev = bus->self;
659 php_slot->bus = bus;
660 php_slot->id = id;
661 php_slot->power_state_check = false;
662 php_slot->slot.ops = &php_slot_ops;
663
664 INIT_LIST_HEAD(&php_slot->children);
665 INIT_LIST_HEAD(&php_slot->link);
666
667 return php_slot;
668 }
669
pnv_php_register_slot(struct pnv_php_slot * php_slot)670 static int pnv_php_register_slot(struct pnv_php_slot *php_slot)
671 {
672 struct pnv_php_slot *parent;
673 struct device_node *dn = php_slot->dn;
674 unsigned long flags;
675 int ret;
676
677 /* Check if the slot is registered or not */
678 parent = pnv_php_find_slot(php_slot->dn);
679 if (parent) {
680 pnv_php_put_slot(parent);
681 return -EEXIST;
682 }
683
684 /* Register PCI slot */
685 ret = pci_hp_register(&php_slot->slot, php_slot->bus,
686 php_slot->slot_no, php_slot->name);
687 if (ret) {
688 SLOT_WARN(php_slot, "Error %d registering slot\n", ret);
689 return ret;
690 }
691
692 /* Attach to the parent's child list or global list */
693 while ((dn = of_get_parent(dn))) {
694 if (!PCI_DN(dn)) {
695 of_node_put(dn);
696 break;
697 }
698
699 parent = pnv_php_find_slot(dn);
700 if (parent) {
701 of_node_put(dn);
702 break;
703 }
704
705 of_node_put(dn);
706 }
707
708 spin_lock_irqsave(&pnv_php_lock, flags);
709 php_slot->parent = parent;
710 if (parent)
711 list_add_tail(&php_slot->link, &parent->children);
712 else
713 list_add_tail(&php_slot->link, &pnv_php_slot_list);
714 spin_unlock_irqrestore(&pnv_php_lock, flags);
715
716 php_slot->state = PNV_PHP_STATE_REGISTERED;
717 return 0;
718 }
719
pnv_php_enable_msix(struct pnv_php_slot * php_slot)720 static int pnv_php_enable_msix(struct pnv_php_slot *php_slot)
721 {
722 struct pci_dev *pdev = php_slot->pdev;
723 struct msix_entry entry;
724 int nr_entries, ret;
725 u16 pcie_flag;
726
727 /* Get total number of MSIx entries */
728 nr_entries = pci_msix_vec_count(pdev);
729 if (nr_entries < 0)
730 return nr_entries;
731
732 /* Check hotplug MSIx entry is in range */
733 pcie_capability_read_word(pdev, PCI_EXP_FLAGS, &pcie_flag);
734 entry.entry = FIELD_GET(PCI_EXP_FLAGS_IRQ, pcie_flag);
735 if (entry.entry >= nr_entries)
736 return -ERANGE;
737
738 /* Enable MSIx */
739 ret = pci_enable_msix_exact(pdev, &entry, 1);
740 if (ret) {
741 SLOT_WARN(php_slot, "Error %d enabling MSIx\n", ret);
742 return ret;
743 }
744
745 return entry.vector;
746 }
747
pnv_php_event_handler(struct work_struct * work)748 static void pnv_php_event_handler(struct work_struct *work)
749 {
750 struct pnv_php_event *event =
751 container_of(work, struct pnv_php_event, work);
752 struct pnv_php_slot *php_slot = event->php_slot;
753
754 if (event->added)
755 pnv_php_enable_slot(&php_slot->slot);
756 else
757 pnv_php_disable_slot(&php_slot->slot);
758
759 kfree(event);
760 }
761
pnv_php_interrupt(int irq,void * data)762 static irqreturn_t pnv_php_interrupt(int irq, void *data)
763 {
764 struct pnv_php_slot *php_slot = data;
765 struct pci_dev *pchild, *pdev = php_slot->pdev;
766 struct eeh_dev *edev;
767 struct eeh_pe *pe;
768 struct pnv_php_event *event;
769 u16 sts, lsts;
770 u8 presence;
771 bool added;
772 unsigned long flags;
773 int ret;
774
775 pcie_capability_read_word(pdev, PCI_EXP_SLTSTA, &sts);
776 sts &= (PCI_EXP_SLTSTA_PDC | PCI_EXP_SLTSTA_DLLSC);
777 pcie_capability_write_word(pdev, PCI_EXP_SLTSTA, sts);
778
779 pci_dbg(pdev, "PCI slot [%s]: HP int! DLAct: %d, PresDet: %d\n",
780 php_slot->name,
781 !!(sts & PCI_EXP_SLTSTA_DLLSC),
782 !!(sts & PCI_EXP_SLTSTA_PDC));
783
784 if (sts & PCI_EXP_SLTSTA_DLLSC) {
785 pcie_capability_read_word(pdev, PCI_EXP_LNKSTA, &lsts);
786 added = !!(lsts & PCI_EXP_LNKSTA_DLLLA);
787 } else if (!(php_slot->flags & PNV_PHP_FLAG_BROKEN_PDC) &&
788 (sts & PCI_EXP_SLTSTA_PDC)) {
789 ret = pnv_pci_get_presence_state(php_slot->id, &presence);
790 if (ret) {
791 SLOT_WARN(php_slot,
792 "PCI slot [%s] error %d getting presence (0x%04x), to retry the operation.\n",
793 php_slot->name, ret, sts);
794 return IRQ_HANDLED;
795 }
796
797 added = !!(presence == OPAL_PCI_SLOT_PRESENT);
798 } else {
799 pci_dbg(pdev, "PCI slot [%s]: Spurious IRQ?\n", php_slot->name);
800 return IRQ_NONE;
801 }
802
803 /* Freeze the removed PE to avoid unexpected error reporting */
804 if (!added) {
805 pchild = list_first_entry_or_null(&php_slot->bus->devices,
806 struct pci_dev, bus_list);
807 edev = pchild ? pci_dev_to_eeh_dev(pchild) : NULL;
808 pe = edev ? edev->pe : NULL;
809 if (pe) {
810 eeh_serialize_lock(&flags);
811 eeh_pe_mark_isolated(pe);
812 eeh_serialize_unlock(flags);
813 eeh_pe_set_option(pe, EEH_OPT_FREEZE_PE);
814 }
815 }
816
817 /*
818 * The PE is left in frozen state if the event is missed. It's
819 * fine as the PCI devices (PE) aren't functional any more.
820 */
821 event = kzalloc(sizeof(*event), GFP_ATOMIC);
822 if (!event) {
823 SLOT_WARN(php_slot,
824 "PCI slot [%s] missed hotplug event 0x%04x\n",
825 php_slot->name, sts);
826 return IRQ_HANDLED;
827 }
828
829 pci_info(pdev, "PCI slot [%s] %s (IRQ: %d)\n",
830 php_slot->name, added ? "added" : "removed", irq);
831 INIT_WORK(&event->work, pnv_php_event_handler);
832 event->added = added;
833 event->php_slot = php_slot;
834 queue_work(php_slot->wq, &event->work);
835
836 return IRQ_HANDLED;
837 }
838
pnv_php_init_irq(struct pnv_php_slot * php_slot,int irq)839 static void pnv_php_init_irq(struct pnv_php_slot *php_slot, int irq)
840 {
841 struct pci_dev *pdev = php_slot->pdev;
842 u32 broken_pdc = 0;
843 u16 sts, ctrl;
844 int ret;
845
846 /* Allocate workqueue */
847 php_slot->wq = alloc_workqueue("pciehp-%s", 0, 0, php_slot->name);
848 if (!php_slot->wq) {
849 SLOT_WARN(php_slot, "Cannot alloc workqueue\n");
850 pnv_php_disable_irq(php_slot, true);
851 return;
852 }
853
854 /* Check PDC (Presence Detection Change) is broken or not */
855 ret = of_property_read_u32(php_slot->dn, "ibm,slot-broken-pdc",
856 &broken_pdc);
857 if (!ret && broken_pdc)
858 php_slot->flags |= PNV_PHP_FLAG_BROKEN_PDC;
859
860 /* Clear pending interrupts */
861 pcie_capability_read_word(pdev, PCI_EXP_SLTSTA, &sts);
862 if (php_slot->flags & PNV_PHP_FLAG_BROKEN_PDC)
863 sts |= PCI_EXP_SLTSTA_DLLSC;
864 else
865 sts |= (PCI_EXP_SLTSTA_PDC | PCI_EXP_SLTSTA_DLLSC);
866 pcie_capability_write_word(pdev, PCI_EXP_SLTSTA, sts);
867
868 /* Request the interrupt */
869 ret = request_irq(irq, pnv_php_interrupt, IRQF_SHARED,
870 php_slot->name, php_slot);
871 if (ret) {
872 pnv_php_disable_irq(php_slot, true);
873 SLOT_WARN(php_slot, "Error %d enabling IRQ %d\n", ret, irq);
874 return;
875 }
876
877 /* Enable the interrupts */
878 pcie_capability_read_word(pdev, PCI_EXP_SLTCTL, &ctrl);
879 if (php_slot->flags & PNV_PHP_FLAG_BROKEN_PDC) {
880 ctrl &= ~PCI_EXP_SLTCTL_PDCE;
881 ctrl |= (PCI_EXP_SLTCTL_HPIE |
882 PCI_EXP_SLTCTL_DLLSCE);
883 } else {
884 ctrl |= (PCI_EXP_SLTCTL_HPIE |
885 PCI_EXP_SLTCTL_PDCE |
886 PCI_EXP_SLTCTL_DLLSCE);
887 }
888 pcie_capability_write_word(pdev, PCI_EXP_SLTCTL, ctrl);
889
890 /* The interrupt is initialized successfully when @irq is valid */
891 php_slot->irq = irq;
892 }
893
pnv_php_enable_irq(struct pnv_php_slot * php_slot)894 static void pnv_php_enable_irq(struct pnv_php_slot *php_slot)
895 {
896 struct pci_dev *pdev = php_slot->pdev;
897 int irq, ret;
898
899 /*
900 * The MSI/MSIx interrupt might have been occupied by other
901 * drivers. Don't populate the surprise hotplug capability
902 * in that case.
903 */
904 if (pci_dev_msi_enabled(pdev))
905 return;
906
907 ret = pci_enable_device(pdev);
908 if (ret) {
909 SLOT_WARN(php_slot, "Error %d enabling device\n", ret);
910 return;
911 }
912
913 pci_set_master(pdev);
914
915 /* Enable MSIx interrupt */
916 irq = pnv_php_enable_msix(php_slot);
917 if (irq > 0) {
918 pnv_php_init_irq(php_slot, irq);
919 return;
920 }
921
922 /*
923 * Use MSI if MSIx doesn't work. Fail back to legacy INTx
924 * if MSI doesn't work either
925 */
926 ret = pci_enable_msi(pdev);
927 if (!ret || pdev->irq) {
928 irq = pdev->irq;
929 pnv_php_init_irq(php_slot, irq);
930 }
931 }
932
pnv_php_register_one(struct device_node * dn)933 static int pnv_php_register_one(struct device_node *dn)
934 {
935 struct pnv_php_slot *php_slot;
936 u32 prop32;
937 int ret;
938
939 /* Check if it's hotpluggable slot */
940 ret = of_property_read_u32(dn, "ibm,slot-pluggable", &prop32);
941 if (ret || !prop32)
942 return -ENXIO;
943
944 ret = of_property_read_u32(dn, "ibm,reset-by-firmware", &prop32);
945 if (ret || !prop32)
946 return -ENXIO;
947
948 php_slot = pnv_php_alloc_slot(dn);
949 if (!php_slot)
950 return -ENODEV;
951
952 ret = pnv_php_register_slot(php_slot);
953 if (ret)
954 goto free_slot;
955
956 ret = pnv_php_enable(php_slot, false);
957 if (ret)
958 goto unregister_slot;
959
960 /* Enable interrupt if the slot supports surprise hotplug */
961 ret = of_property_read_u32(dn, "ibm,slot-surprise-pluggable", &prop32);
962 if (!ret && prop32)
963 pnv_php_enable_irq(php_slot);
964
965 return 0;
966
967 unregister_slot:
968 pnv_php_unregister_one(php_slot->dn);
969 free_slot:
970 pnv_php_put_slot(php_slot);
971 return ret;
972 }
973
pnv_php_register(struct device_node * dn)974 static void pnv_php_register(struct device_node *dn)
975 {
976 struct device_node *child;
977
978 /*
979 * The parent slots should be registered before their
980 * child slots.
981 */
982 for_each_child_of_node(dn, child) {
983 pnv_php_register_one(child);
984 pnv_php_register(child);
985 }
986 }
987
pnv_php_unregister_one(struct device_node * dn)988 static void pnv_php_unregister_one(struct device_node *dn)
989 {
990 struct pnv_php_slot *php_slot;
991
992 php_slot = pnv_php_find_slot(dn);
993 if (!php_slot)
994 return;
995
996 php_slot->state = PNV_PHP_STATE_OFFLINE;
997 pci_hp_deregister(&php_slot->slot);
998 pnv_php_release(php_slot);
999 pnv_php_put_slot(php_slot);
1000 }
1001
pnv_php_unregister(struct device_node * dn)1002 static void pnv_php_unregister(struct device_node *dn)
1003 {
1004 struct device_node *child;
1005
1006 /* The child slots should go before their parent slots */
1007 for_each_child_of_node(dn, child) {
1008 pnv_php_unregister(child);
1009 pnv_php_unregister_one(child);
1010 }
1011 }
1012
pnv_php_init(void)1013 static int __init pnv_php_init(void)
1014 {
1015 struct device_node *dn;
1016
1017 pr_info(DRIVER_DESC " version: " DRIVER_VERSION "\n");
1018 for_each_compatible_node(dn, NULL, "ibm,ioda2-phb")
1019 pnv_php_register(dn);
1020
1021 for_each_compatible_node(dn, NULL, "ibm,ioda3-phb")
1022 pnv_php_register(dn);
1023
1024 for_each_compatible_node(dn, NULL, "ibm,ioda2-npu2-opencapi-phb")
1025 pnv_php_register_one(dn); /* slot directly under the PHB */
1026 return 0;
1027 }
1028
pnv_php_exit(void)1029 static void __exit pnv_php_exit(void)
1030 {
1031 struct device_node *dn;
1032
1033 for_each_compatible_node(dn, NULL, "ibm,ioda2-phb")
1034 pnv_php_unregister(dn);
1035
1036 for_each_compatible_node(dn, NULL, "ibm,ioda3-phb")
1037 pnv_php_unregister(dn);
1038
1039 for_each_compatible_node(dn, NULL, "ibm,ioda2-npu2-opencapi-phb")
1040 pnv_php_unregister_one(dn); /* slot directly under the PHB */
1041 }
1042
1043 module_init(pnv_php_init);
1044 module_exit(pnv_php_exit);
1045
1046 MODULE_VERSION(DRIVER_VERSION);
1047 MODULE_LICENSE("GPL v2");
1048 MODULE_AUTHOR(DRIVER_AUTHOR);
1049 MODULE_DESCRIPTION(DRIVER_DESC);
1050