1 // SPDX-License-Identifier: GPL-2.0
2 /*
3 * Copyright (C) 2014 Intel Corp.
4 * Author: Jiang Liu <jiang.liu@linux.intel.com>
5 *
6 * This file is licensed under GPLv2.
7 *
8 * This file contains common code to support Message Signaled Interrupts for
9 * PCI compatible and non PCI compatible devices.
10 */
11 #include <linux/device.h>
12 #include <linux/irq.h>
13 #include <linux/irqdomain.h>
14 #include <linux/msi.h>
15 #include <linux/mutex.h>
16 #include <linux/pci.h>
17 #include <linux/slab.h>
18 #include <linux/sysfs.h>
19 #include <linux/types.h>
20 #include <linux/xarray.h>
21
22 #include "internals.h"
23
24 /**
25 * struct msi_device_data - MSI per device data
26 * @properties: MSI properties which are interesting to drivers
27 * @mutex: Mutex protecting the MSI descriptor store
28 * @__domains: Internal data for per device MSI domains
29 * @__iter_idx: Index to search the next entry for iterators
30 */
31 struct msi_device_data {
32 unsigned long properties;
33 struct mutex mutex;
34 struct msi_dev_domain __domains[MSI_MAX_DEVICE_IRQDOMAINS];
35 unsigned long __iter_idx;
36 };
37
38 /**
39 * struct msi_ctrl - MSI internal management control structure
40 * @domid: ID of the domain on which management operations should be done
41 * @first: First (hardware) slot index to operate on
42 * @last: Last (hardware) slot index to operate on
43 * @nirqs: The number of Linux interrupts to allocate. Can be larger
44 * than the range due to PCI/multi-MSI.
45 */
46 struct msi_ctrl {
47 unsigned int domid;
48 unsigned int first;
49 unsigned int last;
50 unsigned int nirqs;
51 };
52
53 /* Invalid Xarray index which is outside of any searchable range */
54 #define MSI_XA_MAX_INDEX (ULONG_MAX - 1)
55 /* The maximum domain size */
56 #define MSI_XA_DOMAIN_SIZE (MSI_MAX_INDEX + 1)
57
58 static void msi_domain_free_locked(struct device *dev, struct msi_ctrl *ctrl);
59 static unsigned int msi_domain_get_hwsize(struct device *dev, unsigned int domid);
60 static inline int msi_sysfs_create_group(struct device *dev);
61
62
63 /**
64 * msi_alloc_desc - Allocate an initialized msi_desc
65 * @dev: Pointer to the device for which this is allocated
66 * @nvec: The number of vectors used in this entry
67 * @affinity: Optional pointer to an affinity mask array size of @nvec
68 *
69 * If @affinity is not %NULL then an affinity array[@nvec] is allocated
70 * and the affinity masks and flags from @affinity are copied.
71 *
72 * Return: pointer to allocated &msi_desc on success or %NULL on failure
73 */
74 static struct msi_desc *msi_alloc_desc(struct device *dev, int nvec,
75 const struct irq_affinity_desc *affinity)
76 {
77 struct msi_desc *desc = kzalloc(sizeof(*desc), GFP_KERNEL);
78
79 if (!desc)
80 return NULL;
81
82 desc->dev = dev;
83 desc->nvec_used = nvec;
84 if (affinity) {
85 desc->affinity = kmemdup(affinity, nvec * sizeof(*desc->affinity), GFP_KERNEL);
86 if (!desc->affinity) {
87 kfree(desc);
88 return NULL;
89 }
90 }
91 return desc;
92 }
93
94 static void msi_free_desc(struct msi_desc *desc)
95 {
96 kfree(desc->affinity);
97 kfree(desc);
98 }
99
100 static int msi_insert_desc(struct device *dev, struct msi_desc *desc,
101 unsigned int domid, unsigned int index)
102 {
103 struct msi_device_data *md = dev->msi.data;
104 struct xarray *xa = &md->__domains[domid].store;
105 unsigned int hwsize;
106 int ret;
107
108 hwsize = msi_domain_get_hwsize(dev, domid);
109
110 if (index == MSI_ANY_INDEX) {
111 struct xa_limit limit = { .min = 0, .max = hwsize - 1 };
112 unsigned int index;
113
114 /* Let the xarray allocate a free index within the limit */
115 ret = xa_alloc(xa, &index, desc, limit, GFP_KERNEL);
116 if (ret)
117 goto fail;
118
119 desc->msi_index = index;
120 return 0;
121 } else {
122 if (index >= hwsize) {
123 ret = -ERANGE;
124 goto fail;
125 }
126
127 desc->msi_index = index;
128 ret = xa_insert(xa, index, desc, GFP_KERNEL);
129 if (ret)
130 goto fail;
131 return 0;
132 }
133 fail:
134 msi_free_desc(desc);
135 return ret;
136 }
137
138 /**
139 * msi_domain_insert_msi_desc - Allocate and initialize a MSI descriptor and
140 * insert it at @init_desc->msi_index
141 *
142 * @dev: Pointer to the device for which the descriptor is allocated
143 * @domid: The id of the interrupt domain to which the desriptor is added
144 * @init_desc: Pointer to an MSI descriptor to initialize the new descriptor
145 *
146 * Return: 0 on success or an appropriate failure code.
147 */
148 int msi_domain_insert_msi_desc(struct device *dev, unsigned int domid,
149 struct msi_desc *init_desc)
150 {
151 struct msi_desc *desc;
152
153 lockdep_assert_held(&dev->msi.data->mutex);
154
155 desc = msi_alloc_desc(dev, init_desc->nvec_used, init_desc->affinity);
156 if (!desc)
157 return -ENOMEM;
158
159 /* Copy type specific data to the new descriptor. */
160 desc->pci = init_desc->pci;
161
162 return msi_insert_desc(dev, desc, domid, init_desc->msi_index);
163 }
164
165 static bool msi_desc_match(struct msi_desc *desc, enum msi_desc_filter filter)
166 {
167 switch (filter) {
168 case MSI_DESC_ALL:
169 return true;
170 case MSI_DESC_NOTASSOCIATED:
171 return !desc->irq;
172 case MSI_DESC_ASSOCIATED:
173 return !!desc->irq;
174 }
175 WARN_ON_ONCE(1);
176 return false;
177 }
178
179 static bool msi_ctrl_valid(struct device *dev, struct msi_ctrl *ctrl)
180 {
181 unsigned int hwsize;
182
183 if (WARN_ON_ONCE(ctrl->domid >= MSI_MAX_DEVICE_IRQDOMAINS ||
184 (dev->msi.domain &&
185 !dev->msi.data->__domains[ctrl->domid].domain)))
186 return false;
187
188 hwsize = msi_domain_get_hwsize(dev, ctrl->domid);
189 if (WARN_ON_ONCE(ctrl->first > ctrl->last ||
190 ctrl->first >= hwsize ||
191 ctrl->last >= hwsize))
192 return false;
193 return true;
194 }
195
196 static void msi_domain_free_descs(struct device *dev, struct msi_ctrl *ctrl)
197 {
198 struct msi_desc *desc;
199 struct xarray *xa;
200 unsigned long idx;
201
202 lockdep_assert_held(&dev->msi.data->mutex);
203
204 if (!msi_ctrl_valid(dev, ctrl))
205 return;
206
207 xa = &dev->msi.data->__domains[ctrl->domid].store;
208 xa_for_each_range(xa, idx, desc, ctrl->first, ctrl->last) {
209 xa_erase(xa, idx);
210
211 /* Leak the descriptor when it is still referenced */
212 if (WARN_ON_ONCE(msi_desc_match(desc, MSI_DESC_ASSOCIATED)))
213 continue;
214 msi_free_desc(desc);
215 }
216 }
217
218 /**
219 * msi_domain_free_msi_descs_range - Free a range of MSI descriptors of a device in an irqdomain
220 * @dev: Device for which to free the descriptors
221 * @domid: Id of the domain to operate on
222 * @first: Index to start freeing from (inclusive)
223 * @last: Last index to be freed (inclusive)
224 */
225 void msi_domain_free_msi_descs_range(struct device *dev, unsigned int domid,
226 unsigned int first, unsigned int last)
227 {
228 struct msi_ctrl ctrl = {
229 .domid = domid,
230 .first = first,
231 .last = last,
232 };
233
234 msi_domain_free_descs(dev, &ctrl);
235 }
236
237 /**
238 * msi_domain_add_simple_msi_descs - Allocate and initialize MSI descriptors
239 * @dev: Pointer to the device for which the descriptors are allocated
240 * @ctrl: Allocation control struct
241 *
242 * Return: 0 on success or an appropriate failure code.
243 */
244 static int msi_domain_add_simple_msi_descs(struct device *dev, struct msi_ctrl *ctrl)
245 {
246 struct msi_desc *desc;
247 unsigned int idx;
248 int ret;
249
250 lockdep_assert_held(&dev->msi.data->mutex);
251
252 if (!msi_ctrl_valid(dev, ctrl))
253 return -EINVAL;
254
255 for (idx = ctrl->first; idx <= ctrl->last; idx++) {
256 desc = msi_alloc_desc(dev, 1, NULL);
257 if (!desc)
258 goto fail_mem;
259 ret = msi_insert_desc(dev, desc, ctrl->domid, idx);
260 if (ret)
261 goto fail;
262 }
263 return 0;
264
265 fail_mem:
266 ret = -ENOMEM;
267 fail:
268 msi_domain_free_descs(dev, ctrl);
269 return ret;
270 }
271
272 void __get_cached_msi_msg(struct msi_desc *entry, struct msi_msg *msg)
273 {
274 *msg = entry->msg;
275 }
276
277 void get_cached_msi_msg(unsigned int irq, struct msi_msg *msg)
278 {
279 struct msi_desc *entry = irq_get_msi_desc(irq);
280
281 __get_cached_msi_msg(entry, msg);
282 }
283 EXPORT_SYMBOL_GPL(get_cached_msi_msg);
284
285 static void msi_device_data_release(struct device *dev, void *res)
286 {
287 struct msi_device_data *md = res;
288 int i;
289
290 for (i = 0; i < MSI_MAX_DEVICE_IRQDOMAINS; i++) {
291 msi_remove_device_irq_domain(dev, i);
292 WARN_ON_ONCE(!xa_empty(&md->__domains[i].store));
293 xa_destroy(&md->__domains[i].store);
294 }
295 dev->msi.data = NULL;
296 }
297
298 /**
299 * msi_setup_device_data - Setup MSI device data
300 * @dev: Device for which MSI device data should be set up
301 *
302 * Return: 0 on success, appropriate error code otherwise
303 *
304 * This can be called more than once for @dev. If the MSI device data is
305 * already allocated the call succeeds. The allocated memory is
306 * automatically released when the device is destroyed.
307 */
308 int msi_setup_device_data(struct device *dev)
309 {
310 struct msi_device_data *md;
311 int ret, i;
312
313 if (dev->msi.data)
314 return 0;
315
316 md = devres_alloc(msi_device_data_release, sizeof(*md), GFP_KERNEL);
317 if (!md)
318 return -ENOMEM;
319
320 ret = msi_sysfs_create_group(dev);
321 if (ret) {
322 devres_free(md);
323 return ret;
324 }
325
326 for (i = 0; i < MSI_MAX_DEVICE_IRQDOMAINS; i++)
327 xa_init_flags(&md->__domains[i].store, XA_FLAGS_ALLOC);
328
329 /*
330 * If @dev::msi::domain is set and is a global MSI domain, copy the
331 * pointer into the domain array so all code can operate on domain
332 * ids. The NULL pointer check is required to keep the legacy
333 * architecture specific PCI/MSI support working.
334 */
335 if (dev->msi.domain && !irq_domain_is_msi_parent(dev->msi.domain))
336 md->__domains[MSI_DEFAULT_DOMAIN].domain = dev->msi.domain;
337
338 mutex_init(&md->mutex);
339 dev->msi.data = md;
340 devres_add(dev, md);
341 return 0;
342 }
343
344 /**
345 * msi_lock_descs - Lock the MSI descriptor storage of a device
346 * @dev: Device to operate on
347 */
348 void msi_lock_descs(struct device *dev)
349 {
350 mutex_lock(&dev->msi.data->mutex);
351 }
352 EXPORT_SYMBOL_GPL(msi_lock_descs);
353
354 /**
355 * msi_unlock_descs - Unlock the MSI descriptor storage of a device
356 * @dev: Device to operate on
357 */
358 void msi_unlock_descs(struct device *dev)
359 {
360 /* Invalidate the index which was cached by the iterator */
361 dev->msi.data->__iter_idx = MSI_XA_MAX_INDEX;
362 mutex_unlock(&dev->msi.data->mutex);
363 }
364 EXPORT_SYMBOL_GPL(msi_unlock_descs);
365
366 static struct msi_desc *msi_find_desc(struct msi_device_data *md, unsigned int domid,
367 enum msi_desc_filter filter)
368 {
369 struct xarray *xa = &md->__domains[domid].store;
370 struct msi_desc *desc;
371
372 xa_for_each_start(xa, md->__iter_idx, desc, md->__iter_idx) {
373 if (msi_desc_match(desc, filter))
374 return desc;
375 }
376 md->__iter_idx = MSI_XA_MAX_INDEX;
377 return NULL;
378 }
379
380 /**
381 * msi_domain_first_desc - Get the first MSI descriptor of an irqdomain associated to a device
382 * @dev: Device to operate on
383 * @domid: The id of the interrupt domain which should be walked.
384 * @filter: Descriptor state filter
385 *
386 * Must be called with the MSI descriptor mutex held, i.e. msi_lock_descs()
387 * must be invoked before the call.
388 *
389 * Return: Pointer to the first MSI descriptor matching the search
390 * criteria, NULL if none found.
391 */
392 struct msi_desc *msi_domain_first_desc(struct device *dev, unsigned int domid,
393 enum msi_desc_filter filter)
394 {
395 struct msi_device_data *md = dev->msi.data;
396
397 if (WARN_ON_ONCE(!md || domid >= MSI_MAX_DEVICE_IRQDOMAINS))
398 return NULL;
399
400 lockdep_assert_held(&md->mutex);
401
402 md->__iter_idx = 0;
403 return msi_find_desc(md, domid, filter);
404 }
405 EXPORT_SYMBOL_GPL(msi_domain_first_desc);
406
407 /**
408 * msi_next_desc - Get the next MSI descriptor of a device
409 * @dev: Device to operate on
410 * @domid: The id of the interrupt domain which should be walked.
411 * @filter: Descriptor state filter
412 *
413 * The first invocation of msi_next_desc() has to be preceeded by a
414 * successful invocation of __msi_first_desc(). Consecutive invocations are
415 * only valid if the previous one was successful. All these operations have
416 * to be done within the same MSI mutex held region.
417 *
418 * Return: Pointer to the next MSI descriptor matching the search
419 * criteria, NULL if none found.
420 */
421 struct msi_desc *msi_next_desc(struct device *dev, unsigned int domid,
422 enum msi_desc_filter filter)
423 {
424 struct msi_device_data *md = dev->msi.data;
425
426 if (WARN_ON_ONCE(!md || domid >= MSI_MAX_DEVICE_IRQDOMAINS))
427 return NULL;
428
429 lockdep_assert_held(&md->mutex);
430
431 if (md->__iter_idx >= (unsigned long)MSI_MAX_INDEX)
432 return NULL;
433
434 md->__iter_idx++;
435 return msi_find_desc(md, domid, filter);
436 }
437 EXPORT_SYMBOL_GPL(msi_next_desc);
438
439 /**
440 * msi_domain_get_virq - Lookup the Linux interrupt number for a MSI index on a interrupt domain
441 * @dev: Device to operate on
442 * @domid: Domain ID of the interrupt domain associated to the device
443 * @index: MSI interrupt index to look for (0-based)
444 *
445 * Return: The Linux interrupt number on success (> 0), 0 if not found
446 */
447 unsigned int msi_domain_get_virq(struct device *dev, unsigned int domid, unsigned int index)
448 {
449 struct msi_desc *desc;
450 unsigned int ret = 0;
451 bool pcimsi = false;
452 struct xarray *xa;
453
454 if (!dev->msi.data)
455 return 0;
456
457 if (WARN_ON_ONCE(index > MSI_MAX_INDEX || domid >= MSI_MAX_DEVICE_IRQDOMAINS))
458 return 0;
459
460 /* This check is only valid for the PCI default MSI domain */
461 if (dev_is_pci(dev) && domid == MSI_DEFAULT_DOMAIN)
462 pcimsi = to_pci_dev(dev)->msi_enabled;
463
464 msi_lock_descs(dev);
465 xa = &dev->msi.data->__domains[domid].store;
466 desc = xa_load(xa, pcimsi ? 0 : index);
467 if (desc && desc->irq) {
468 /*
469 * PCI-MSI has only one descriptor for multiple interrupts.
470 * PCI-MSIX and platform MSI use a descriptor per
471 * interrupt.
472 */
473 if (pcimsi) {
474 if (index < desc->nvec_used)
475 ret = desc->irq + index;
476 } else {
477 ret = desc->irq;
478 }
479 }
480
481 msi_unlock_descs(dev);
482 return ret;
483 }
484 EXPORT_SYMBOL_GPL(msi_domain_get_virq);
485
486 #ifdef CONFIG_SYSFS
487 static struct attribute *msi_dev_attrs[] = {
488 NULL
489 };
490
491 static const struct attribute_group msi_irqs_group = {
492 .name = "msi_irqs",
493 .attrs = msi_dev_attrs,
494 };
495
496 static inline int msi_sysfs_create_group(struct device *dev)
497 {
498 return devm_device_add_group(dev, &msi_irqs_group);
499 }
500
501 static ssize_t msi_mode_show(struct device *dev, struct device_attribute *attr,
502 char *buf)
503 {
504 /* MSI vs. MSIX is per device not per interrupt */
505 bool is_msix = dev_is_pci(dev) ? to_pci_dev(dev)->msix_enabled : false;
506
507 return sysfs_emit(buf, "%s\n", is_msix ? "msix" : "msi");
508 }
509
510 static void msi_sysfs_remove_desc(struct device *dev, struct msi_desc *desc)
511 {
512 struct device_attribute *attrs = desc->sysfs_attrs;
513 int i;
514
515 if (!attrs)
516 return;
517
518 desc->sysfs_attrs = NULL;
519 for (i = 0; i < desc->nvec_used; i++) {
520 if (attrs[i].show)
521 sysfs_remove_file_from_group(&dev->kobj, &attrs[i].attr, msi_irqs_group.name);
522 kfree(attrs[i].attr.name);
523 }
524 kfree(attrs);
525 }
526
527 static int msi_sysfs_populate_desc(struct device *dev, struct msi_desc *desc)
528 {
529 struct device_attribute *attrs;
530 int ret, i;
531
532 attrs = kcalloc(desc->nvec_used, sizeof(*attrs), GFP_KERNEL);
533 if (!attrs)
534 return -ENOMEM;
535
536 desc->sysfs_attrs = attrs;
537 for (i = 0; i < desc->nvec_used; i++) {
538 sysfs_attr_init(&attrs[i].attr);
539 attrs[i].attr.name = kasprintf(GFP_KERNEL, "%d", desc->irq + i);
540 if (!attrs[i].attr.name) {
541 ret = -ENOMEM;
542 goto fail;
543 }
544
545 attrs[i].attr.mode = 0444;
546 attrs[i].show = msi_mode_show;
547
548 ret = sysfs_add_file_to_group(&dev->kobj, &attrs[i].attr, msi_irqs_group.name);
549 if (ret) {
550 attrs[i].show = NULL;
551 goto fail;
552 }
553 }
554 return 0;
555
556 fail:
557 msi_sysfs_remove_desc(dev, desc);
558 return ret;
559 }
560
561 #if defined(CONFIG_PCI_MSI_ARCH_FALLBACKS) || defined(CONFIG_PCI_XEN)
562 /**
563 * msi_device_populate_sysfs - Populate msi_irqs sysfs entries for a device
564 * @dev: The device (PCI, platform etc) which will get sysfs entries
565 */
566 int msi_device_populate_sysfs(struct device *dev)
567 {
568 struct msi_desc *desc;
569 int ret;
570
571 msi_for_each_desc(desc, dev, MSI_DESC_ASSOCIATED) {
572 if (desc->sysfs_attrs)
573 continue;
574 ret = msi_sysfs_populate_desc(dev, desc);
575 if (ret)
576 return ret;
577 }
578 return 0;
579 }
580
581 /**
582 * msi_device_destroy_sysfs - Destroy msi_irqs sysfs entries for a device
583 * @dev: The device (PCI, platform etc) for which to remove
584 * sysfs entries
585 */
586 void msi_device_destroy_sysfs(struct device *dev)
587 {
588 struct msi_desc *desc;
589
590 msi_for_each_desc(desc, dev, MSI_DESC_ALL)
591 msi_sysfs_remove_desc(dev, desc);
592 }
593 #endif /* CONFIG_PCI_MSI_ARCH_FALLBACK || CONFIG_PCI_XEN */
594 #else /* CONFIG_SYSFS */
595 static inline int msi_sysfs_create_group(struct device *dev) { return 0; }
596 static inline int msi_sysfs_populate_desc(struct device *dev, struct msi_desc *desc) { return 0; }
597 static inline void msi_sysfs_remove_desc(struct device *dev, struct msi_desc *desc) { }
598 #endif /* !CONFIG_SYSFS */
599
600 static struct irq_domain *msi_get_device_domain(struct device *dev, unsigned int domid)
601 {
602 struct irq_domain *domain;
603
604 lockdep_assert_held(&dev->msi.data->mutex);
605
606 if (WARN_ON_ONCE(domid >= MSI_MAX_DEVICE_IRQDOMAINS))
607 return NULL;
608
609 domain = dev->msi.data->__domains[domid].domain;
610 if (!domain)
611 return NULL;
612
613 if (WARN_ON_ONCE(irq_domain_is_msi_parent(domain)))
614 return NULL;
615
616 return domain;
617 }
618
619 static unsigned int msi_domain_get_hwsize(struct device *dev, unsigned int domid)
620 {
621 struct msi_domain_info *info;
622 struct irq_domain *domain;
623
624 domain = msi_get_device_domain(dev, domid);
625 if (domain) {
626 info = domain->host_data;
627 return info->hwsize;
628 }
629 /* No domain, default to MSI_XA_DOMAIN_SIZE */
630 return MSI_XA_DOMAIN_SIZE;
631 }
632
633 static inline void irq_chip_write_msi_msg(struct irq_data *data,
634 struct msi_msg *msg)
635 {
636 data->chip->irq_write_msi_msg(data, msg);
637 }
638
639 static void msi_check_level(struct irq_domain *domain, struct msi_msg *msg)
640 {
641 struct msi_domain_info *info = domain->host_data;
642
643 /*
644 * If the MSI provider has messed with the second message and
645 * not advertized that it is level-capable, signal the breakage.
646 */
647 WARN_ON(!((info->flags & MSI_FLAG_LEVEL_CAPABLE) &&
648 (info->chip->flags & IRQCHIP_SUPPORTS_LEVEL_MSI)) &&
649 (msg[1].address_lo || msg[1].address_hi || msg[1].data));
650 }
651
652 /**
653 * msi_domain_set_affinity - Generic affinity setter function for MSI domains
654 * @irq_data: The irq data associated to the interrupt
655 * @mask: The affinity mask to set
656 * @force: Flag to enforce setting (disable online checks)
657 *
658 * Intended to be used by MSI interrupt controllers which are
659 * implemented with hierarchical domains.
660 *
661 * Return: IRQ_SET_MASK_* result code
662 */
663 int msi_domain_set_affinity(struct irq_data *irq_data,
664 const struct cpumask *mask, bool force)
665 {
666 struct irq_data *parent = irq_data->parent_data;
667 struct msi_msg msg[2] = { [1] = { }, };
668 int ret;
669
670 ret = parent->chip->irq_set_affinity(parent, mask, force);
671 if (ret >= 0 && ret != IRQ_SET_MASK_OK_DONE) {
672 BUG_ON(irq_chip_compose_msi_msg(irq_data, msg));
673 msi_check_level(irq_data->domain, msg);
674 irq_chip_write_msi_msg(irq_data, msg);
675 }
676
677 return ret;
678 }
679
680 static int msi_domain_activate(struct irq_domain *domain,
681 struct irq_data *irq_data, bool early)
682 {
683 struct msi_msg msg[2] = { [1] = { }, };
684
685 BUG_ON(irq_chip_compose_msi_msg(irq_data, msg));
686 msi_check_level(irq_data->domain, msg);
687 irq_chip_write_msi_msg(irq_data, msg);
688 return 0;
689 }
690
691 static void msi_domain_deactivate(struct irq_domain *domain,
692 struct irq_data *irq_data)
693 {
694 struct msi_msg msg[2];
695
696 memset(msg, 0, sizeof(msg));
697 irq_chip_write_msi_msg(irq_data, msg);
698 }
699
700 static int msi_domain_alloc(struct irq_domain *domain, unsigned int virq,
701 unsigned int nr_irqs, void *arg)
702 {
703 struct msi_domain_info *info = domain->host_data;
704 struct msi_domain_ops *ops = info->ops;
705 irq_hw_number_t hwirq = ops->get_hwirq(info, arg);
706 int i, ret;
707
708 if (irq_find_mapping(domain, hwirq) > 0)
709 return -EEXIST;
710
711 if (domain->parent) {
712 ret = irq_domain_alloc_irqs_parent(domain, virq, nr_irqs, arg);
713 if (ret < 0)
714 return ret;
715 }
716
717 for (i = 0; i < nr_irqs; i++) {
718 ret = ops->msi_init(domain, info, virq + i, hwirq + i, arg);
719 if (ret < 0) {
720 if (ops->msi_free) {
721 for (i--; i > 0; i--)
722 ops->msi_free(domain, info, virq + i);
723 }
724 irq_domain_free_irqs_top(domain, virq, nr_irqs);
725 return ret;
726 }
727 }
728
729 return 0;
730 }
731
732 static void msi_domain_free(struct irq_domain *domain, unsigned int virq,
733 unsigned int nr_irqs)
734 {
735 struct msi_domain_info *info = domain->host_data;
736 int i;
737
738 if (info->ops->msi_free) {
739 for (i = 0; i < nr_irqs; i++)
740 info->ops->msi_free(domain, info, virq + i);
741 }
742 irq_domain_free_irqs_top(domain, virq, nr_irqs);
743 }
744
745 static int msi_domain_translate(struct irq_domain *domain, struct irq_fwspec *fwspec,
746 irq_hw_number_t *hwirq, unsigned int *type)
747 {
748 struct msi_domain_info *info = domain->host_data;
749
750 /*
751 * This will catch allocations through the regular irqdomain path except
752 * for MSI domains which really support this, e.g. MBIGEN.
753 */
754 if (!info->ops->msi_translate)
755 return -ENOTSUPP;
756 return info->ops->msi_translate(domain, fwspec, hwirq, type);
757 }
758
759 static const struct irq_domain_ops msi_domain_ops = {
760 .alloc = msi_domain_alloc,
761 .free = msi_domain_free,
762 .activate = msi_domain_activate,
763 .deactivate = msi_domain_deactivate,
764 .translate = msi_domain_translate,
765 };
766
767 static irq_hw_number_t msi_domain_ops_get_hwirq(struct msi_domain_info *info,
768 msi_alloc_info_t *arg)
769 {
770 return arg->hwirq;
771 }
772
773 static int msi_domain_ops_prepare(struct irq_domain *domain, struct device *dev,
774 int nvec, msi_alloc_info_t *arg)
775 {
776 memset(arg, 0, sizeof(*arg));
777 return 0;
778 }
779
780 static void msi_domain_ops_set_desc(msi_alloc_info_t *arg,
781 struct msi_desc *desc)
782 {
783 arg->desc = desc;
784 }
785
786 static int msi_domain_ops_init(struct irq_domain *domain,
787 struct msi_domain_info *info,
788 unsigned int virq, irq_hw_number_t hwirq,
789 msi_alloc_info_t *arg)
790 {
791 irq_domain_set_hwirq_and_chip(domain, virq, hwirq, info->chip,
792 info->chip_data);
793 if (info->handler && info->handler_name) {
794 __irq_set_handler(virq, info->handler, 0, info->handler_name);
795 if (info->handler_data)
796 irq_set_handler_data(virq, info->handler_data);
797 }
798 return 0;
799 }
800
801 static struct msi_domain_ops msi_domain_ops_default = {
802 .get_hwirq = msi_domain_ops_get_hwirq,
803 .msi_init = msi_domain_ops_init,
804 .msi_prepare = msi_domain_ops_prepare,
805 .set_desc = msi_domain_ops_set_desc,
806 };
807
808 static void msi_domain_update_dom_ops(struct msi_domain_info *info)
809 {
810 struct msi_domain_ops *ops = info->ops;
811
812 if (ops == NULL) {
813 info->ops = &msi_domain_ops_default;
814 return;
815 }
816
817 if (!(info->flags & MSI_FLAG_USE_DEF_DOM_OPS))
818 return;
819
820 if (ops->get_hwirq == NULL)
821 ops->get_hwirq = msi_domain_ops_default.get_hwirq;
822 if (ops->msi_init == NULL)
823 ops->msi_init = msi_domain_ops_default.msi_init;
824 if (ops->msi_prepare == NULL)
825 ops->msi_prepare = msi_domain_ops_default.msi_prepare;
826 if (ops->set_desc == NULL)
827 ops->set_desc = msi_domain_ops_default.set_desc;
828 }
829
830 static void msi_domain_update_chip_ops(struct msi_domain_info *info)
831 {
832 struct irq_chip *chip = info->chip;
833
834 BUG_ON(!chip || !chip->irq_mask || !chip->irq_unmask);
835 if (!chip->irq_set_affinity)
836 chip->irq_set_affinity = msi_domain_set_affinity;
837 }
838
839 static struct irq_domain *__msi_create_irq_domain(struct fwnode_handle *fwnode,
840 struct msi_domain_info *info,
841 unsigned int flags,
842 struct irq_domain *parent)
843 {
844 struct irq_domain *domain;
845
846 if (info->hwsize > MSI_XA_DOMAIN_SIZE)
847 return NULL;
848
849 /*
850 * Hardware size 0 is valid for backwards compatibility and for
851 * domains which are not backed by a hardware table. Grant the
852 * maximum index space.
853 */
854 if (!info->hwsize)
855 info->hwsize = MSI_XA_DOMAIN_SIZE;
856
857 msi_domain_update_dom_ops(info);
858 if (info->flags & MSI_FLAG_USE_DEF_CHIP_OPS)
859 msi_domain_update_chip_ops(info);
860
861 domain = irq_domain_create_hierarchy(parent, flags | IRQ_DOMAIN_FLAG_MSI, 0,
862 fwnode, &msi_domain_ops, info);
863
864 if (domain) {
865 irq_domain_update_bus_token(domain, info->bus_token);
866 if (info->flags & MSI_FLAG_PARENT_PM_DEV)
867 domain->pm_dev = parent->pm_dev;
868 }
869
870 return domain;
871 }
872
873 /**
874 * msi_create_irq_domain - Create an MSI interrupt domain
875 * @fwnode: Optional fwnode of the interrupt controller
876 * @info: MSI domain info
877 * @parent: Parent irq domain
878 *
879 * Return: pointer to the created &struct irq_domain or %NULL on failure
880 */
881 struct irq_domain *msi_create_irq_domain(struct fwnode_handle *fwnode,
882 struct msi_domain_info *info,
883 struct irq_domain *parent)
884 {
885 return __msi_create_irq_domain(fwnode, info, 0, parent);
886 }
887
888 /**
889 * msi_parent_init_dev_msi_info - Delegate initialization of device MSI info down
890 * in the domain hierarchy
891 * @dev: The device for which the domain should be created
892 * @domain: The domain in the hierarchy this op is being called on
893 * @msi_parent_domain: The IRQ_DOMAIN_FLAG_MSI_PARENT domain for the child to
894 * be created
895 * @msi_child_info: The MSI domain info of the IRQ_DOMAIN_FLAG_MSI_DEVICE
896 * domain to be created
897 *
898 * Return: true on success, false otherwise
899 *
900 * This is the most complex problem of per device MSI domains and the
901 * underlying interrupt domain hierarchy:
902 *
903 * The device domain to be initialized requests the broadest feature set
904 * possible and the underlying domain hierarchy puts restrictions on it.
905 *
906 * That's trivial for a simple parent->child relationship, but it gets
907 * interesting with an intermediate domain: root->parent->child. The
908 * intermediate 'parent' can expand the capabilities which the 'root'
909 * domain is providing. So that creates a classic hen and egg problem:
910 * Which entity is doing the restrictions/expansions?
911 *
912 * One solution is to let the root domain handle the initialization that's
913 * why there is the @domain and the @msi_parent_domain pointer.
914 */
915 bool msi_parent_init_dev_msi_info(struct device *dev, struct irq_domain *domain,
916 struct irq_domain *msi_parent_domain,
917 struct msi_domain_info *msi_child_info)
918 {
919 struct irq_domain *parent = domain->parent;
920
921 if (WARN_ON_ONCE(!parent || !parent->msi_parent_ops ||
922 !parent->msi_parent_ops->init_dev_msi_info))
923 return false;
924
925 return parent->msi_parent_ops->init_dev_msi_info(dev, parent, msi_parent_domain,
926 msi_child_info);
927 }
928
929 /**
930 * msi_create_device_irq_domain - Create a device MSI interrupt domain
931 * @dev: Pointer to the device
932 * @domid: Domain id
933 * @template: MSI domain info bundle used as template
934 * @hwsize: Maximum number of MSI table entries (0 if unknown or unlimited)
935 * @domain_data: Optional pointer to domain specific data which is set in
936 * msi_domain_info::data
937 * @chip_data: Optional pointer to chip specific data which is set in
938 * msi_domain_info::chip_data
939 *
940 * Return: True on success, false otherwise
941 *
942 * There is no firmware node required for this interface because the per
943 * device domains are software constructs which are actually closer to the
944 * hardware reality than any firmware can describe them.
945 *
946 * The domain name and the irq chip name for a MSI device domain are
947 * composed by: "$(PREFIX)$(CHIPNAME)-$(DEVNAME)"
948 *
949 * $PREFIX: Optional prefix provided by the underlying MSI parent domain
950 * via msi_parent_ops::prefix. If that pointer is NULL the prefix
951 * is empty.
952 * $CHIPNAME: The name of the irq_chip in @template
953 * $DEVNAME: The name of the device
954 *
955 * This results in understandable chip names and hardware interrupt numbers
956 * in e.g. /proc/interrupts
957 *
958 * PCI-MSI-0000:00:1c.0 0-edge Parent domain has no prefix
959 * IR-PCI-MSI-0000:00:1c.4 0-edge Same with interrupt remapping prefix 'IR-'
960 *
961 * IR-PCI-MSIX-0000:3d:00.0 0-edge Hardware interrupt numbers reflect
962 * IR-PCI-MSIX-0000:3d:00.0 1-edge the real MSI-X index on that device
963 * IR-PCI-MSIX-0000:3d:00.0 2-edge
964 *
965 * On IMS domains the hardware interrupt number is either a table entry
966 * index or a purely software managed index but it is guaranteed to be
967 * unique.
968 *
969 * The domain pointer is stored in @dev::msi::data::__irqdomains[]. All
970 * subsequent operations on the domain depend on the domain id.
971 *
972 * The domain is automatically freed when the device is removed via devres
973 * in the context of @dev::msi::data freeing, but it can also be
974 * independently removed via @msi_remove_device_irq_domain().
975 */
976 bool msi_create_device_irq_domain(struct device *dev, unsigned int domid,
977 const struct msi_domain_template *template,
978 unsigned int hwsize, void *domain_data,
979 void *chip_data)
980 {
981 struct irq_domain *domain, *parent = dev->msi.domain;
982 struct fwnode_handle *fwnode, *fwnalloced = NULL;
983 struct msi_domain_template *bundle;
984 const struct msi_parent_ops *pops;
985
986 if (!irq_domain_is_msi_parent(parent))
987 return false;
988
989 if (domid >= MSI_MAX_DEVICE_IRQDOMAINS)
990 return false;
991
992 bundle = kmemdup(template, sizeof(*bundle), GFP_KERNEL);
993 if (!bundle)
994 return false;
995
996 bundle->info.hwsize = hwsize;
997 bundle->info.chip = &bundle->chip;
998 bundle->info.ops = &bundle->ops;
999 bundle->info.data = domain_data;
1000 bundle->info.chip_data = chip_data;
1001
1002 pops = parent->msi_parent_ops;
1003 snprintf(bundle->name, sizeof(bundle->name), "%s%s-%s",
1004 pops->prefix ? : "", bundle->chip.name, dev_name(dev));
1005 bundle->chip.name = bundle->name;
1006
1007 /*
1008 * Using the device firmware node is required for wire to MSI
1009 * device domains so that the existing firmware results in a domain
1010 * match.
1011 * All other device domains like PCI/MSI use the named firmware
1012 * node as they are not guaranteed to have a fwnode. They are never
1013 * looked up and always handled in the context of the device.
1014 */
1015 if (bundle->info.flags & MSI_FLAG_USE_DEV_FWNODE)
1016 fwnode = dev->fwnode;
1017 else
1018 fwnode = fwnalloced = irq_domain_alloc_named_fwnode(bundle->name);
1019
1020 if (!fwnode)
1021 goto free_bundle;
1022
1023 if (msi_setup_device_data(dev))
1024 goto free_fwnode;
1025
1026 msi_lock_descs(dev);
1027
1028 if (WARN_ON_ONCE(msi_get_device_domain(dev, domid)))
1029 goto fail;
1030
1031 if (!pops->init_dev_msi_info(dev, parent, parent, &bundle->info))
1032 goto fail;
1033
1034 domain = __msi_create_irq_domain(fwnode, &bundle->info, IRQ_DOMAIN_FLAG_MSI_DEVICE, parent);
1035 if (!domain)
1036 goto fail;
1037
1038 domain->dev = dev;
1039 dev->msi.data->__domains[domid].domain = domain;
1040 msi_unlock_descs(dev);
1041 return true;
1042
1043 fail:
1044 msi_unlock_descs(dev);
1045 free_fwnode:
1046 irq_domain_free_fwnode(fwnalloced);
1047 free_bundle:
1048 kfree(bundle);
1049 return false;
1050 }
1051
1052 /**
1053 * msi_remove_device_irq_domain - Free a device MSI interrupt domain
1054 * @dev: Pointer to the device
1055 * @domid: Domain id
1056 */
1057 void msi_remove_device_irq_domain(struct device *dev, unsigned int domid)
1058 {
1059 struct fwnode_handle *fwnode = NULL;
1060 struct msi_domain_info *info;
1061 struct irq_domain *domain;
1062
1063 msi_lock_descs(dev);
1064
1065 domain = msi_get_device_domain(dev, domid);
1066
1067 if (!domain || !irq_domain_is_msi_device(domain))
1068 goto unlock;
1069
1070 dev->msi.data->__domains[domid].domain = NULL;
1071 info = domain->host_data;
1072 if (irq_domain_is_msi_device(domain))
1073 fwnode = domain->fwnode;
1074 irq_domain_remove(domain);
1075 irq_domain_free_fwnode(fwnode);
1076 kfree(container_of(info, struct msi_domain_template, info));
1077
1078 unlock:
1079 msi_unlock_descs(dev);
1080 }
1081
1082 /**
1083 * msi_match_device_irq_domain - Match a device irq domain against a bus token
1084 * @dev: Pointer to the device
1085 * @domid: Domain id
1086 * @bus_token: Bus token to match against the domain bus token
1087 *
1088 * Return: True if device domain exists and bus tokens match.
1089 */
1090 bool msi_match_device_irq_domain(struct device *dev, unsigned int domid,
1091 enum irq_domain_bus_token bus_token)
1092 {
1093 struct msi_domain_info *info;
1094 struct irq_domain *domain;
1095 bool ret = false;
1096
1097 msi_lock_descs(dev);
1098 domain = msi_get_device_domain(dev, domid);
1099 if (domain && irq_domain_is_msi_device(domain)) {
1100 info = domain->host_data;
1101 ret = info->bus_token == bus_token;
1102 }
1103 msi_unlock_descs(dev);
1104 return ret;
1105 }
1106
1107 static int msi_domain_prepare_irqs(struct irq_domain *domain, struct device *dev,
1108 int nvec, msi_alloc_info_t *arg)
1109 {
1110 struct msi_domain_info *info = domain->host_data;
1111 struct msi_domain_ops *ops = info->ops;
1112
1113 return ops->msi_prepare(domain, dev, nvec, arg);
1114 }
1115
1116 /*
1117 * Carefully check whether the device can use reservation mode. If
1118 * reservation mode is enabled then the early activation will assign a
1119 * dummy vector to the device. If the PCI/MSI device does not support
1120 * masking of the entry then this can result in spurious interrupts when
1121 * the device driver is not absolutely careful. But even then a malfunction
1122 * of the hardware could result in a spurious interrupt on the dummy vector
1123 * and render the device unusable. If the entry can be masked then the core
1124 * logic will prevent the spurious interrupt and reservation mode can be
1125 * used. For now reservation mode is restricted to PCI/MSI.
1126 */
1127 static bool msi_check_reservation_mode(struct irq_domain *domain,
1128 struct msi_domain_info *info,
1129 struct device *dev)
1130 {
1131 struct msi_desc *desc;
1132
1133 switch(domain->bus_token) {
1134 case DOMAIN_BUS_PCI_MSI:
1135 case DOMAIN_BUS_PCI_DEVICE_MSI:
1136 case DOMAIN_BUS_PCI_DEVICE_MSIX:
1137 case DOMAIN_BUS_VMD_MSI:
1138 break;
1139 default:
1140 return false;
1141 }
1142
1143 if (!(info->flags & MSI_FLAG_MUST_REACTIVATE))
1144 return false;
1145
1146 if (IS_ENABLED(CONFIG_PCI_MSI) && pci_msi_ignore_mask)
1147 return false;
1148
1149 /*
1150 * Checking the first MSI descriptor is sufficient. MSIX supports
1151 * masking and MSI does so when the can_mask attribute is set.
1152 */
1153 desc = msi_first_desc(dev, MSI_DESC_ALL);
1154 return desc->pci.msi_attrib.is_msix || desc->pci.msi_attrib.can_mask;
1155 }
1156
1157 static int msi_handle_pci_fail(struct irq_domain *domain, struct msi_desc *desc,
1158 int allocated)
1159 {
1160 switch(domain->bus_token) {
1161 case DOMAIN_BUS_PCI_MSI:
1162 case DOMAIN_BUS_PCI_DEVICE_MSI:
1163 case DOMAIN_BUS_PCI_DEVICE_MSIX:
1164 case DOMAIN_BUS_VMD_MSI:
1165 if (IS_ENABLED(CONFIG_PCI_MSI))
1166 break;
1167 fallthrough;
1168 default:
1169 return -ENOSPC;
1170 }
1171
1172 /* Let a failed PCI multi MSI allocation retry */
1173 if (desc->nvec_used > 1)
1174 return 1;
1175
1176 /* If there was a successful allocation let the caller know */
1177 return allocated ? allocated : -ENOSPC;
1178 }
1179
1180 #define VIRQ_CAN_RESERVE 0x01
1181 #define VIRQ_ACTIVATE 0x02
1182
1183 static int msi_init_virq(struct irq_domain *domain, int virq, unsigned int vflags)
1184 {
1185 struct irq_data *irqd = irq_domain_get_irq_data(domain, virq);
1186 int ret;
1187
1188 if (!(vflags & VIRQ_CAN_RESERVE)) {
1189 irqd_clr_can_reserve(irqd);
1190
1191 /*
1192 * If the interrupt is managed but no CPU is available to
1193 * service it, shut it down until better times. Note that
1194 * we only do this on the !RESERVE path as x86 (the only
1195 * architecture using this flag) deals with this in a
1196 * different way by using a catch-all vector.
1197 */
1198 if ((vflags & VIRQ_ACTIVATE) &&
1199 irqd_affinity_is_managed(irqd) &&
1200 !cpumask_intersects(irq_data_get_affinity_mask(irqd),
1201 cpu_online_mask)) {
1202 irqd_set_managed_shutdown(irqd);
1203 return 0;
1204 }
1205 }
1206
1207 if (!(vflags & VIRQ_ACTIVATE))
1208 return 0;
1209
1210 ret = irq_domain_activate_irq(irqd, vflags & VIRQ_CAN_RESERVE);
1211 if (ret)
1212 return ret;
1213 /*
1214 * If the interrupt uses reservation mode, clear the activated bit
1215 * so request_irq() will assign the final vector.
1216 */
1217 if (vflags & VIRQ_CAN_RESERVE)
1218 irqd_clr_activated(irqd);
1219 return 0;
1220 }
1221
1222 static int __msi_domain_alloc_irqs(struct device *dev, struct irq_domain *domain,
1223 struct msi_ctrl *ctrl)
1224 {
1225 struct xarray *xa = &dev->msi.data->__domains[ctrl->domid].store;
1226 struct msi_domain_info *info = domain->host_data;
1227 struct msi_domain_ops *ops = info->ops;
1228 unsigned int vflags = 0, allocated = 0;
1229 msi_alloc_info_t arg = { };
1230 struct msi_desc *desc;
1231 unsigned long idx;
1232 int i, ret, virq;
1233
1234 ret = msi_domain_prepare_irqs(domain, dev, ctrl->nirqs, &arg);
1235 if (ret)
1236 return ret;
1237
1238 /*
1239 * This flag is set by the PCI layer as we need to activate
1240 * the MSI entries before the PCI layer enables MSI in the
1241 * card. Otherwise the card latches a random msi message.
1242 */
1243 if (info->flags & MSI_FLAG_ACTIVATE_EARLY)
1244 vflags |= VIRQ_ACTIVATE;
1245
1246 /*
1247 * Interrupt can use a reserved vector and will not occupy
1248 * a real device vector until the interrupt is requested.
1249 */
1250 if (msi_check_reservation_mode(domain, info, dev))
1251 vflags |= VIRQ_CAN_RESERVE;
1252
1253 xa_for_each_range(xa, idx, desc, ctrl->first, ctrl->last) {
1254 if (!msi_desc_match(desc, MSI_DESC_NOTASSOCIATED))
1255 continue;
1256
1257 /* This should return -ECONFUSED... */
1258 if (WARN_ON_ONCE(allocated >= ctrl->nirqs))
1259 return -EINVAL;
1260
1261 if (ops->prepare_desc)
1262 ops->prepare_desc(domain, &arg, desc);
1263
1264 ops->set_desc(&arg, desc);
1265
1266 virq = __irq_domain_alloc_irqs(domain, -1, desc->nvec_used,
1267 dev_to_node(dev), &arg, false,
1268 desc->affinity);
1269 if (virq < 0)
1270 return msi_handle_pci_fail(domain, desc, allocated);
1271
1272 for (i = 0; i < desc->nvec_used; i++) {
1273 irq_set_msi_desc_off(virq, i, desc);
1274 irq_debugfs_copy_devname(virq + i, dev);
1275 ret = msi_init_virq(domain, virq + i, vflags);
1276 if (ret)
1277 return ret;
1278 }
1279 if (info->flags & MSI_FLAG_DEV_SYSFS) {
1280 ret = msi_sysfs_populate_desc(dev, desc);
1281 if (ret)
1282 return ret;
1283 }
1284 allocated++;
1285 }
1286 return 0;
1287 }
1288
1289 static int msi_domain_alloc_simple_msi_descs(struct device *dev,
1290 struct msi_domain_info *info,
1291 struct msi_ctrl *ctrl)
1292 {
1293 if (!(info->flags & MSI_FLAG_ALLOC_SIMPLE_MSI_DESCS))
1294 return 0;
1295
1296 return msi_domain_add_simple_msi_descs(dev, ctrl);
1297 }
1298
1299 static int __msi_domain_alloc_locked(struct device *dev, struct msi_ctrl *ctrl)
1300 {
1301 struct msi_domain_info *info;
1302 struct msi_domain_ops *ops;
1303 struct irq_domain *domain;
1304 int ret;
1305
1306 if (!msi_ctrl_valid(dev, ctrl))
1307 return -EINVAL;
1308
1309 domain = msi_get_device_domain(dev, ctrl->domid);
1310 if (!domain)
1311 return -ENODEV;
1312
1313 info = domain->host_data;
1314
1315 ret = msi_domain_alloc_simple_msi_descs(dev, info, ctrl);
1316 if (ret)
1317 return ret;
1318
1319 ops = info->ops;
1320 if (ops->domain_alloc_irqs)
1321 return ops->domain_alloc_irqs(domain, dev, ctrl->nirqs);
1322
1323 return __msi_domain_alloc_irqs(dev, domain, ctrl);
1324 }
1325
1326 static int msi_domain_alloc_locked(struct device *dev, struct msi_ctrl *ctrl)
1327 {
1328 int ret = __msi_domain_alloc_locked(dev, ctrl);
1329
1330 if (ret)
1331 msi_domain_free_locked(dev, ctrl);
1332 return ret;
1333 }
1334
1335 /**
1336 * msi_domain_alloc_irqs_range_locked - Allocate interrupts from a MSI interrupt domain
1337 * @dev: Pointer to device struct of the device for which the interrupts
1338 * are allocated
1339 * @domid: Id of the interrupt domain to operate on
1340 * @first: First index to allocate (inclusive)
1341 * @last: Last index to allocate (inclusive)
1342 *
1343 * Must be invoked from within a msi_lock_descs() / msi_unlock_descs()
1344 * pair. Use this for MSI irqdomains which implement their own descriptor
1345 * allocation/free.
1346 *
1347 * Return: %0 on success or an error code.
1348 */
1349 int msi_domain_alloc_irqs_range_locked(struct device *dev, unsigned int domid,
1350 unsigned int first, unsigned int last)
1351 {
1352 struct msi_ctrl ctrl = {
1353 .domid = domid,
1354 .first = first,
1355 .last = last,
1356 .nirqs = last + 1 - first,
1357 };
1358
1359 return msi_domain_alloc_locked(dev, &ctrl);
1360 }
1361
1362 /**
1363 * msi_domain_alloc_irqs_range - Allocate interrupts from a MSI interrupt domain
1364 * @dev: Pointer to device struct of the device for which the interrupts
1365 * are allocated
1366 * @domid: Id of the interrupt domain to operate on
1367 * @first: First index to allocate (inclusive)
1368 * @last: Last index to allocate (inclusive)
1369 *
1370 * Return: %0 on success or an error code.
1371 */
1372 int msi_domain_alloc_irqs_range(struct device *dev, unsigned int domid,
1373 unsigned int first, unsigned int last)
1374 {
1375 int ret;
1376
1377 msi_lock_descs(dev);
1378 ret = msi_domain_alloc_irqs_range_locked(dev, domid, first, last);
1379 msi_unlock_descs(dev);
1380 return ret;
1381 }
1382 EXPORT_SYMBOL_GPL(msi_domain_alloc_irqs_range);
1383
1384 /**
1385 * msi_domain_alloc_irqs_all_locked - Allocate all interrupts from a MSI interrupt domain
1386 *
1387 * @dev: Pointer to device struct of the device for which the interrupts
1388 * are allocated
1389 * @domid: Id of the interrupt domain to operate on
1390 * @nirqs: The number of interrupts to allocate
1391 *
1392 * This function scans all MSI descriptors of the MSI domain and allocates interrupts
1393 * for all unassigned ones. That function is to be used for MSI domain usage where
1394 * the descriptor allocation is handled at the call site, e.g. PCI/MSI[X].
1395 *
1396 * Return: %0 on success or an error code.
1397 */
1398 int msi_domain_alloc_irqs_all_locked(struct device *dev, unsigned int domid, int nirqs)
1399 {
1400 struct msi_ctrl ctrl = {
1401 .domid = domid,
1402 .first = 0,
1403 .last = msi_domain_get_hwsize(dev, domid) - 1,
1404 .nirqs = nirqs,
1405 };
1406
1407 return msi_domain_alloc_locked(dev, &ctrl);
1408 }
1409
1410 static struct msi_map __msi_domain_alloc_irq_at(struct device *dev, unsigned int domid,
1411 unsigned int index,
1412 const struct irq_affinity_desc *affdesc,
1413 union msi_instance_cookie *icookie)
1414 {
1415 struct msi_ctrl ctrl = { .domid = domid, .nirqs = 1, };
1416 struct irq_domain *domain;
1417 struct msi_map map = { };
1418 struct msi_desc *desc;
1419 int ret;
1420
1421 domain = msi_get_device_domain(dev, domid);
1422 if (!domain) {
1423 map.index = -ENODEV;
1424 return map;
1425 }
1426
1427 desc = msi_alloc_desc(dev, 1, affdesc);
1428 if (!desc) {
1429 map.index = -ENOMEM;
1430 return map;
1431 }
1432
1433 if (icookie)
1434 desc->data.icookie = *icookie;
1435
1436 ret = msi_insert_desc(dev, desc, domid, index);
1437 if (ret) {
1438 map.index = ret;
1439 return map;
1440 }
1441
1442 ctrl.first = ctrl.last = desc->msi_index;
1443
1444 ret = __msi_domain_alloc_irqs(dev, domain, &ctrl);
1445 if (ret) {
1446 map.index = ret;
1447 msi_domain_free_locked(dev, &ctrl);
1448 } else {
1449 map.index = desc->msi_index;
1450 map.virq = desc->irq;
1451 }
1452 return map;
1453 }
1454
1455 /**
1456 * msi_domain_alloc_irq_at - Allocate an interrupt from a MSI interrupt domain at
1457 * a given index - or at the next free index
1458 *
1459 * @dev: Pointer to device struct of the device for which the interrupts
1460 * are allocated
1461 * @domid: Id of the interrupt domain to operate on
1462 * @index: Index for allocation. If @index == %MSI_ANY_INDEX the allocation
1463 * uses the next free index.
1464 * @affdesc: Optional pointer to an interrupt affinity descriptor structure
1465 * @icookie: Optional pointer to a domain specific per instance cookie. If
1466 * non-NULL the content of the cookie is stored in msi_desc::data.
1467 * Must be NULL for MSI-X allocations
1468 *
1469 * This requires a MSI interrupt domain which lets the core code manage the
1470 * MSI descriptors.
1471 *
1472 * Return: struct msi_map
1473 *
1474 * On success msi_map::index contains the allocated index number and
1475 * msi_map::virq the corresponding Linux interrupt number
1476 *
1477 * On failure msi_map::index contains the error code and msi_map::virq
1478 * is %0.
1479 */
1480 struct msi_map msi_domain_alloc_irq_at(struct device *dev, unsigned int domid, unsigned int index,
1481 const struct irq_affinity_desc *affdesc,
1482 union msi_instance_cookie *icookie)
1483 {
1484 struct msi_map map;
1485
1486 msi_lock_descs(dev);
1487 map = __msi_domain_alloc_irq_at(dev, domid, index, affdesc, icookie);
1488 msi_unlock_descs(dev);
1489 return map;
1490 }
1491
1492 /**
1493 * msi_device_domain_alloc_wired - Allocate a "wired" interrupt on @domain
1494 * @domain: The domain to allocate on
1495 * @hwirq: The hardware interrupt number to allocate for
1496 * @type: The interrupt type
1497 *
1498 * This weirdness supports wire to MSI controllers like MBIGEN.
1499 *
1500 * @hwirq is the hardware interrupt number which is handed in from
1501 * irq_create_fwspec_mapping(). As the wire to MSI domain is sparse, but
1502 * sized in firmware, the hardware interrupt number cannot be used as MSI
1503 * index. For the underlying irq chip the MSI index is irrelevant and
1504 * all it needs is the hardware interrupt number.
1505 *
1506 * To handle this the MSI index is allocated with MSI_ANY_INDEX and the
1507 * hardware interrupt number is stored along with the type information in
1508 * msi_desc::cookie so the underlying interrupt chip and domain code can
1509 * retrieve it.
1510 *
1511 * Return: The Linux interrupt number (> 0) or an error code
1512 */
1513 int msi_device_domain_alloc_wired(struct irq_domain *domain, unsigned int hwirq,
1514 unsigned int type)
1515 {
1516 unsigned int domid = MSI_DEFAULT_DOMAIN;
1517 union msi_instance_cookie icookie = { };
1518 struct device *dev = domain->dev;
1519 struct msi_map map = { };
1520
1521 if (WARN_ON_ONCE(!dev || domain->bus_token != DOMAIN_BUS_WIRED_TO_MSI))
1522 return -EINVAL;
1523
1524 icookie.value = ((u64)type << 32) | hwirq;
1525
1526 msi_lock_descs(dev);
1527 if (WARN_ON_ONCE(msi_get_device_domain(dev, domid) != domain))
1528 map.index = -EINVAL;
1529 else
1530 map = __msi_domain_alloc_irq_at(dev, domid, MSI_ANY_INDEX, NULL, &icookie);
1531 msi_unlock_descs(dev);
1532
1533 return map.index >= 0 ? map.virq : map.index;
1534 }
1535
1536 static void __msi_domain_free_irqs(struct device *dev, struct irq_domain *domain,
1537 struct msi_ctrl *ctrl)
1538 {
1539 struct xarray *xa = &dev->msi.data->__domains[ctrl->domid].store;
1540 struct msi_domain_info *info = domain->host_data;
1541 struct irq_data *irqd;
1542 struct msi_desc *desc;
1543 unsigned long idx;
1544 int i;
1545
1546 xa_for_each_range(xa, idx, desc, ctrl->first, ctrl->last) {
1547 /* Only handle MSI entries which have an interrupt associated */
1548 if (!msi_desc_match(desc, MSI_DESC_ASSOCIATED))
1549 continue;
1550
1551 /* Make sure all interrupts are deactivated */
1552 for (i = 0; i < desc->nvec_used; i++) {
1553 irqd = irq_domain_get_irq_data(domain, desc->irq + i);
1554 if (irqd && irqd_is_activated(irqd))
1555 irq_domain_deactivate_irq(irqd);
1556 }
1557
1558 irq_domain_free_irqs(desc->irq, desc->nvec_used);
1559 if (info->flags & MSI_FLAG_DEV_SYSFS)
1560 msi_sysfs_remove_desc(dev, desc);
1561 desc->irq = 0;
1562 }
1563 }
1564
1565 static void msi_domain_free_locked(struct device *dev, struct msi_ctrl *ctrl)
1566 {
1567 struct msi_domain_info *info;
1568 struct msi_domain_ops *ops;
1569 struct irq_domain *domain;
1570
1571 if (!msi_ctrl_valid(dev, ctrl))
1572 return;
1573
1574 domain = msi_get_device_domain(dev, ctrl->domid);
1575 if (!domain)
1576 return;
1577
1578 info = domain->host_data;
1579 ops = info->ops;
1580
1581 if (ops->domain_free_irqs)
1582 ops->domain_free_irqs(domain, dev);
1583 else
1584 __msi_domain_free_irqs(dev, domain, ctrl);
1585
1586 if (ops->msi_post_free)
1587 ops->msi_post_free(domain, dev);
1588
1589 if (info->flags & MSI_FLAG_FREE_MSI_DESCS)
1590 msi_domain_free_descs(dev, ctrl);
1591 }
1592
1593 /**
1594 * msi_domain_free_irqs_range_locked - Free a range of interrupts from a MSI interrupt domain
1595 * associated to @dev with msi_lock held
1596 * @dev: Pointer to device struct of the device for which the interrupts
1597 * are freed
1598 * @domid: Id of the interrupt domain to operate on
1599 * @first: First index to free (inclusive)
1600 * @last: Last index to free (inclusive)
1601 */
1602 void msi_domain_free_irqs_range_locked(struct device *dev, unsigned int domid,
1603 unsigned int first, unsigned int last)
1604 {
1605 struct msi_ctrl ctrl = {
1606 .domid = domid,
1607 .first = first,
1608 .last = last,
1609 };
1610 msi_domain_free_locked(dev, &ctrl);
1611 }
1612
1613 /**
1614 * msi_domain_free_irqs_range - Free a range of interrupts from a MSI interrupt domain
1615 * associated to @dev
1616 * @dev: Pointer to device struct of the device for which the interrupts
1617 * are freed
1618 * @domid: Id of the interrupt domain to operate on
1619 * @first: First index to free (inclusive)
1620 * @last: Last index to free (inclusive)
1621 */
1622 void msi_domain_free_irqs_range(struct device *dev, unsigned int domid,
1623 unsigned int first, unsigned int last)
1624 {
1625 msi_lock_descs(dev);
1626 msi_domain_free_irqs_range_locked(dev, domid, first, last);
1627 msi_unlock_descs(dev);
1628 }
1629 EXPORT_SYMBOL_GPL(msi_domain_free_irqs_all);
1630
1631 /**
1632 * msi_domain_free_irqs_all_locked - Free all interrupts from a MSI interrupt domain
1633 * associated to a device
1634 * @dev: Pointer to device struct of the device for which the interrupts
1635 * are freed
1636 * @domid: The id of the domain to operate on
1637 *
1638 * Must be invoked from within a msi_lock_descs() / msi_unlock_descs()
1639 * pair. Use this for MSI irqdomains which implement their own vector
1640 * allocation.
1641 */
1642 void msi_domain_free_irqs_all_locked(struct device *dev, unsigned int domid)
1643 {
1644 msi_domain_free_irqs_range_locked(dev, domid, 0,
1645 msi_domain_get_hwsize(dev, domid) - 1);
1646 }
1647
1648 /**
1649 * msi_domain_free_irqs_all - Free all interrupts from a MSI interrupt domain
1650 * associated to a device
1651 * @dev: Pointer to device struct of the device for which the interrupts
1652 * are freed
1653 * @domid: The id of the domain to operate on
1654 */
1655 void msi_domain_free_irqs_all(struct device *dev, unsigned int domid)
1656 {
1657 msi_lock_descs(dev);
1658 msi_domain_free_irqs_all_locked(dev, domid);
1659 msi_unlock_descs(dev);
1660 }
1661
1662 /**
1663 * msi_device_domain_free_wired - Free a wired interrupt in @domain
1664 * @domain: The domain to free the interrupt on
1665 * @virq: The Linux interrupt number to free
1666 *
1667 * This is the counterpart of msi_device_domain_alloc_wired() for the
1668 * weird wired to MSI converting domains.
1669 */
1670 void msi_device_domain_free_wired(struct irq_domain *domain, unsigned int virq)
1671 {
1672 struct msi_desc *desc = irq_get_msi_desc(virq);
1673 struct device *dev = domain->dev;
1674
1675 if (WARN_ON_ONCE(!dev || !desc || domain->bus_token != DOMAIN_BUS_WIRED_TO_MSI))
1676 return;
1677
1678 msi_lock_descs(dev);
1679 if (!WARN_ON_ONCE(msi_get_device_domain(dev, MSI_DEFAULT_DOMAIN) != domain)) {
1680 msi_domain_free_irqs_range_locked(dev, MSI_DEFAULT_DOMAIN, desc->msi_index,
1681 desc->msi_index);
1682 }
1683 msi_unlock_descs(dev);
1684 }
1685
1686 /**
1687 * msi_get_domain_info - Get the MSI interrupt domain info for @domain
1688 * @domain: The interrupt domain to retrieve data from
1689 *
1690 * Return: the pointer to the msi_domain_info stored in @domain->host_data.
1691 */
1692 struct msi_domain_info *msi_get_domain_info(struct irq_domain *domain)
1693 {
1694 return (struct msi_domain_info *)domain->host_data;
1695 }
1696
1697 /**
1698 * msi_device_has_isolated_msi - True if the device has isolated MSI
1699 * @dev: The device to check
1700 *
1701 * Isolated MSI means that HW modeled by an irq_domain on the path from the
1702 * initiating device to the CPU will validate that the MSI message specifies an
1703 * interrupt number that the device is authorized to trigger. This must block
1704 * devices from triggering interrupts they are not authorized to trigger.
1705 * Currently authorization means the MSI vector is one assigned to the device.
1706 *
1707 * This is interesting for securing VFIO use cases where a rouge MSI (eg created
1708 * by abusing a normal PCI MemWr DMA) must not allow the VFIO userspace to
1709 * impact outside its security domain, eg userspace triggering interrupts on
1710 * kernel drivers, a VM triggering interrupts on the hypervisor, or a VM
1711 * triggering interrupts on another VM.
1712 */
1713 bool msi_device_has_isolated_msi(struct device *dev)
1714 {
1715 struct irq_domain *domain = dev_get_msi_domain(dev);
1716
1717 for (; domain; domain = domain->parent)
1718 if (domain->flags & IRQ_DOMAIN_FLAG_ISOLATED_MSI)
1719 return true;
1720 return arch_is_isolated_msi();
1721 }
1722 EXPORT_SYMBOL_GPL(msi_device_has_isolated_msi);
1723
Linux® is a registered trademark of Linus Torvalds in the United States and other countries.
TOMOYO® is a registered trademark of NTT DATA CORPORATION.