Line data Source code
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/types.h>
12 : #include <linux/device.h>
13 : #include <linux/irq.h>
14 : #include <linux/irqdomain.h>
15 : #include <linux/msi.h>
16 : #include <linux/slab.h>
17 : #include <linux/sysfs.h>
18 : #include <linux/pci.h>
19 :
20 : #include "internals.h"
21 :
22 : /**
23 : * struct msi_ctrl - MSI internal management control structure
24 : * @domid: ID of the domain on which management operations should be done
25 : * @first: First (hardware) slot index to operate on
26 : * @last: Last (hardware) slot index to operate on
27 : * @nirqs: The number of Linux interrupts to allocate. Can be larger
28 : * than the range due to PCI/multi-MSI.
29 : */
30 : struct msi_ctrl {
31 : unsigned int domid;
32 : unsigned int first;
33 : unsigned int last;
34 : unsigned int nirqs;
35 : };
36 :
37 : /* Invalid Xarray index which is outside of any searchable range */
38 : #define MSI_XA_MAX_INDEX (ULONG_MAX - 1)
39 : /* The maximum domain size */
40 : #define MSI_XA_DOMAIN_SIZE (MSI_MAX_INDEX + 1)
41 :
42 : static void msi_domain_free_locked(struct device *dev, struct msi_ctrl *ctrl);
43 : static unsigned int msi_domain_get_hwsize(struct device *dev, unsigned int domid);
44 : static inline int msi_sysfs_create_group(struct device *dev);
45 :
46 :
47 : /**
48 : * msi_alloc_desc - Allocate an initialized msi_desc
49 : * @dev: Pointer to the device for which this is allocated
50 : * @nvec: The number of vectors used in this entry
51 : * @affinity: Optional pointer to an affinity mask array size of @nvec
52 : *
53 : * If @affinity is not %NULL then an affinity array[@nvec] is allocated
54 : * and the affinity masks and flags from @affinity are copied.
55 : *
56 : * Return: pointer to allocated &msi_desc on success or %NULL on failure
57 : */
58 0 : static struct msi_desc *msi_alloc_desc(struct device *dev, int nvec,
59 : const struct irq_affinity_desc *affinity)
60 : {
61 0 : struct msi_desc *desc = kzalloc(sizeof(*desc), GFP_KERNEL);
62 :
63 0 : if (!desc)
64 : return NULL;
65 :
66 0 : desc->dev = dev;
67 0 : desc->nvec_used = nvec;
68 0 : if (affinity) {
69 0 : desc->affinity = kmemdup(affinity, nvec * sizeof(*desc->affinity), GFP_KERNEL);
70 0 : if (!desc->affinity) {
71 0 : kfree(desc);
72 0 : return NULL;
73 : }
74 : }
75 : return desc;
76 : }
77 :
78 : static void msi_free_desc(struct msi_desc *desc)
79 : {
80 0 : kfree(desc->affinity);
81 0 : kfree(desc);
82 : }
83 :
84 0 : static int msi_insert_desc(struct device *dev, struct msi_desc *desc,
85 : unsigned int domid, unsigned int index)
86 : {
87 0 : struct msi_device_data *md = dev->msi.data;
88 0 : struct xarray *xa = &md->__domains[domid].store;
89 : unsigned int hwsize;
90 : int ret;
91 :
92 0 : hwsize = msi_domain_get_hwsize(dev, domid);
93 :
94 0 : if (index == MSI_ANY_INDEX) {
95 0 : struct xa_limit limit = { .min = 0, .max = hwsize - 1 };
96 : unsigned int index;
97 :
98 : /* Let the xarray allocate a free index within the limit */
99 0 : ret = xa_alloc(xa, &index, desc, limit, GFP_KERNEL);
100 0 : if (ret)
101 : goto fail;
102 :
103 0 : desc->msi_index = index;
104 0 : return 0;
105 : } else {
106 0 : if (index >= hwsize) {
107 : ret = -ERANGE;
108 : goto fail;
109 : }
110 :
111 0 : desc->msi_index = index;
112 0 : ret = xa_insert(xa, index, desc, GFP_KERNEL);
113 0 : if (ret)
114 : goto fail;
115 : return 0;
116 : }
117 : fail:
118 0 : msi_free_desc(desc);
119 0 : return ret;
120 : }
121 :
122 : /**
123 : * msi_domain_insert_msi_desc - Allocate and initialize a MSI descriptor and
124 : * insert it at @init_desc->msi_index
125 : *
126 : * @dev: Pointer to the device for which the descriptor is allocated
127 : * @domid: The id of the interrupt domain to which the desriptor is added
128 : * @init_desc: Pointer to an MSI descriptor to initialize the new descriptor
129 : *
130 : * Return: 0 on success or an appropriate failure code.
131 : */
132 0 : int msi_domain_insert_msi_desc(struct device *dev, unsigned int domid,
133 : struct msi_desc *init_desc)
134 : {
135 : struct msi_desc *desc;
136 :
137 : lockdep_assert_held(&dev->msi.data->mutex);
138 :
139 0 : desc = msi_alloc_desc(dev, init_desc->nvec_used, init_desc->affinity);
140 0 : if (!desc)
141 : return -ENOMEM;
142 :
143 : /* Copy type specific data to the new descriptor. */
144 0 : desc->pci = init_desc->pci;
145 :
146 0 : return msi_insert_desc(dev, desc, domid, init_desc->msi_index);
147 : }
148 :
149 0 : static bool msi_desc_match(struct msi_desc *desc, enum msi_desc_filter filter)
150 : {
151 0 : switch (filter) {
152 : case MSI_DESC_ALL:
153 : return true;
154 : case MSI_DESC_NOTASSOCIATED:
155 0 : return !desc->irq;
156 : case MSI_DESC_ASSOCIATED:
157 0 : return !!desc->irq;
158 : }
159 0 : WARN_ON_ONCE(1);
160 : return false;
161 : }
162 :
163 0 : static bool msi_ctrl_valid(struct device *dev, struct msi_ctrl *ctrl)
164 : {
165 : unsigned int hwsize;
166 :
167 0 : if (WARN_ON_ONCE(ctrl->domid >= MSI_MAX_DEVICE_IRQDOMAINS ||
168 : (dev->msi.domain &&
169 : !dev->msi.data->__domains[ctrl->domid].domain)))
170 : return false;
171 :
172 0 : hwsize = msi_domain_get_hwsize(dev, ctrl->domid);
173 0 : if (WARN_ON_ONCE(ctrl->first > ctrl->last ||
174 : ctrl->first >= hwsize ||
175 : ctrl->last >= hwsize))
176 : return false;
177 0 : return true;
178 : }
179 :
180 0 : static void msi_domain_free_descs(struct device *dev, struct msi_ctrl *ctrl)
181 : {
182 : struct msi_desc *desc;
183 : struct xarray *xa;
184 : unsigned long idx;
185 :
186 : lockdep_assert_held(&dev->msi.data->mutex);
187 :
188 0 : if (!msi_ctrl_valid(dev, ctrl))
189 0 : return;
190 :
191 0 : xa = &dev->msi.data->__domains[ctrl->domid].store;
192 0 : xa_for_each_range(xa, idx, desc, ctrl->first, ctrl->last) {
193 0 : xa_erase(xa, idx);
194 :
195 : /* Leak the descriptor when it is still referenced */
196 0 : if (WARN_ON_ONCE(msi_desc_match(desc, MSI_DESC_ASSOCIATED)))
197 0 : continue;
198 : msi_free_desc(desc);
199 : }
200 : }
201 :
202 : /**
203 : * msi_domain_free_msi_descs_range - Free a range of MSI descriptors of a device in an irqdomain
204 : * @dev: Device for which to free the descriptors
205 : * @domid: Id of the domain to operate on
206 : * @first: Index to start freeing from (inclusive)
207 : * @last: Last index to be freed (inclusive)
208 : */
209 0 : void msi_domain_free_msi_descs_range(struct device *dev, unsigned int domid,
210 : unsigned int first, unsigned int last)
211 : {
212 0 : struct msi_ctrl ctrl = {
213 : .domid = domid,
214 : .first = first,
215 : .last = last,
216 : };
217 :
218 0 : msi_domain_free_descs(dev, &ctrl);
219 0 : }
220 :
221 : /**
222 : * msi_domain_add_simple_msi_descs - Allocate and initialize MSI descriptors
223 : * @dev: Pointer to the device for which the descriptors are allocated
224 : * @ctrl: Allocation control struct
225 : *
226 : * Return: 0 on success or an appropriate failure code.
227 : */
228 0 : static int msi_domain_add_simple_msi_descs(struct device *dev, struct msi_ctrl *ctrl)
229 : {
230 : struct msi_desc *desc;
231 : unsigned int idx;
232 : int ret;
233 :
234 : lockdep_assert_held(&dev->msi.data->mutex);
235 :
236 0 : if (!msi_ctrl_valid(dev, ctrl))
237 : return -EINVAL;
238 :
239 0 : for (idx = ctrl->first; idx <= ctrl->last; idx++) {
240 0 : desc = msi_alloc_desc(dev, 1, NULL);
241 0 : if (!desc)
242 : goto fail_mem;
243 0 : ret = msi_insert_desc(dev, desc, ctrl->domid, idx);
244 0 : if (ret)
245 : goto fail;
246 : }
247 : return 0;
248 :
249 : fail_mem:
250 : ret = -ENOMEM;
251 : fail:
252 0 : msi_domain_free_descs(dev, ctrl);
253 0 : return ret;
254 : }
255 :
256 0 : void __get_cached_msi_msg(struct msi_desc *entry, struct msi_msg *msg)
257 : {
258 0 : *msg = entry->msg;
259 0 : }
260 :
261 0 : void get_cached_msi_msg(unsigned int irq, struct msi_msg *msg)
262 : {
263 0 : struct msi_desc *entry = irq_get_msi_desc(irq);
264 :
265 0 : __get_cached_msi_msg(entry, msg);
266 0 : }
267 : EXPORT_SYMBOL_GPL(get_cached_msi_msg);
268 :
269 0 : static void msi_device_data_release(struct device *dev, void *res)
270 : {
271 0 : struct msi_device_data *md = res;
272 : int i;
273 :
274 0 : for (i = 0; i < MSI_MAX_DEVICE_IRQDOMAINS; i++) {
275 0 : msi_remove_device_irq_domain(dev, i);
276 0 : WARN_ON_ONCE(!xa_empty(&md->__domains[i].store));
277 0 : xa_destroy(&md->__domains[i].store);
278 : }
279 0 : dev->msi.data = NULL;
280 0 : }
281 :
282 : /**
283 : * msi_setup_device_data - Setup MSI device data
284 : * @dev: Device for which MSI device data should be set up
285 : *
286 : * Return: 0 on success, appropriate error code otherwise
287 : *
288 : * This can be called more than once for @dev. If the MSI device data is
289 : * already allocated the call succeeds. The allocated memory is
290 : * automatically released when the device is destroyed.
291 : */
292 0 : int msi_setup_device_data(struct device *dev)
293 : {
294 : struct msi_device_data *md;
295 : int ret, i;
296 :
297 0 : if (dev->msi.data)
298 : return 0;
299 :
300 0 : md = devres_alloc(msi_device_data_release, sizeof(*md), GFP_KERNEL);
301 0 : if (!md)
302 : return -ENOMEM;
303 :
304 0 : ret = msi_sysfs_create_group(dev);
305 0 : if (ret) {
306 0 : devres_free(md);
307 0 : return ret;
308 : }
309 :
310 0 : for (i = 0; i < MSI_MAX_DEVICE_IRQDOMAINS; i++)
311 0 : xa_init_flags(&md->__domains[i].store, XA_FLAGS_ALLOC);
312 :
313 : /*
314 : * If @dev::msi::domain is set and is a global MSI domain, copy the
315 : * pointer into the domain array so all code can operate on domain
316 : * ids. The NULL pointer check is required to keep the legacy
317 : * architecture specific PCI/MSI support working.
318 : */
319 0 : if (dev->msi.domain && !irq_domain_is_msi_parent(dev->msi.domain))
320 0 : md->__domains[MSI_DEFAULT_DOMAIN].domain = dev->msi.domain;
321 :
322 0 : mutex_init(&md->mutex);
323 0 : dev->msi.data = md;
324 0 : devres_add(dev, md);
325 0 : return 0;
326 : }
327 :
328 : /**
329 : * msi_lock_descs - Lock the MSI descriptor storage of a device
330 : * @dev: Device to operate on
331 : */
332 0 : void msi_lock_descs(struct device *dev)
333 : {
334 0 : mutex_lock(&dev->msi.data->mutex);
335 0 : }
336 : EXPORT_SYMBOL_GPL(msi_lock_descs);
337 :
338 : /**
339 : * msi_unlock_descs - Unlock the MSI descriptor storage of a device
340 : * @dev: Device to operate on
341 : */
342 0 : void msi_unlock_descs(struct device *dev)
343 : {
344 : /* Invalidate the index which was cached by the iterator */
345 0 : dev->msi.data->__iter_idx = MSI_XA_MAX_INDEX;
346 0 : mutex_unlock(&dev->msi.data->mutex);
347 0 : }
348 : EXPORT_SYMBOL_GPL(msi_unlock_descs);
349 :
350 0 : static struct msi_desc *msi_find_desc(struct msi_device_data *md, unsigned int domid,
351 : enum msi_desc_filter filter)
352 : {
353 0 : struct xarray *xa = &md->__domains[domid].store;
354 : struct msi_desc *desc;
355 :
356 0 : xa_for_each_start(xa, md->__iter_idx, desc, md->__iter_idx) {
357 0 : if (msi_desc_match(desc, filter))
358 : return desc;
359 : }
360 0 : md->__iter_idx = MSI_XA_MAX_INDEX;
361 0 : return NULL;
362 : }
363 :
364 : /**
365 : * msi_domain_first_desc - Get the first MSI descriptor of an irqdomain associated to a device
366 : * @dev: Device to operate on
367 : * @domid: The id of the interrupt domain which should be walked.
368 : * @filter: Descriptor state filter
369 : *
370 : * Must be called with the MSI descriptor mutex held, i.e. msi_lock_descs()
371 : * must be invoked before the call.
372 : *
373 : * Return: Pointer to the first MSI descriptor matching the search
374 : * criteria, NULL if none found.
375 : */
376 0 : struct msi_desc *msi_domain_first_desc(struct device *dev, unsigned int domid,
377 : enum msi_desc_filter filter)
378 : {
379 0 : struct msi_device_data *md = dev->msi.data;
380 :
381 0 : if (WARN_ON_ONCE(!md || domid >= MSI_MAX_DEVICE_IRQDOMAINS))
382 : return NULL;
383 :
384 : lockdep_assert_held(&md->mutex);
385 :
386 0 : md->__iter_idx = 0;
387 0 : return msi_find_desc(md, domid, filter);
388 : }
389 : EXPORT_SYMBOL_GPL(msi_domain_first_desc);
390 :
391 : /**
392 : * msi_next_desc - Get the next MSI descriptor of a device
393 : * @dev: Device to operate on
394 : * @domid: The id of the interrupt domain which should be walked.
395 : * @filter: Descriptor state filter
396 : *
397 : * The first invocation of msi_next_desc() has to be preceeded by a
398 : * successful invocation of __msi_first_desc(). Consecutive invocations are
399 : * only valid if the previous one was successful. All these operations have
400 : * to be done within the same MSI mutex held region.
401 : *
402 : * Return: Pointer to the next MSI descriptor matching the search
403 : * criteria, NULL if none found.
404 : */
405 0 : struct msi_desc *msi_next_desc(struct device *dev, unsigned int domid,
406 : enum msi_desc_filter filter)
407 : {
408 0 : struct msi_device_data *md = dev->msi.data;
409 :
410 0 : if (WARN_ON_ONCE(!md || domid >= MSI_MAX_DEVICE_IRQDOMAINS))
411 : return NULL;
412 :
413 : lockdep_assert_held(&md->mutex);
414 :
415 0 : if (md->__iter_idx >= (unsigned long)MSI_MAX_INDEX)
416 : return NULL;
417 :
418 0 : md->__iter_idx++;
419 0 : return msi_find_desc(md, domid, filter);
420 : }
421 : EXPORT_SYMBOL_GPL(msi_next_desc);
422 :
423 : /**
424 : * msi_domain_get_virq - Lookup the Linux interrupt number for a MSI index on a interrupt domain
425 : * @dev: Device to operate on
426 : * @domid: Domain ID of the interrupt domain associated to the device
427 : * @index: MSI interrupt index to look for (0-based)
428 : *
429 : * Return: The Linux interrupt number on success (> 0), 0 if not found
430 : */
431 0 : unsigned int msi_domain_get_virq(struct device *dev, unsigned int domid, unsigned int index)
432 : {
433 : struct msi_desc *desc;
434 0 : unsigned int ret = 0;
435 0 : bool pcimsi = false;
436 : struct xarray *xa;
437 :
438 0 : if (!dev->msi.data)
439 : return 0;
440 :
441 0 : if (WARN_ON_ONCE(index > MSI_MAX_INDEX || domid >= MSI_MAX_DEVICE_IRQDOMAINS))
442 : return 0;
443 :
444 : /* This check is only valid for the PCI default MSI domain */
445 0 : if (dev_is_pci(dev) && domid == MSI_DEFAULT_DOMAIN)
446 0 : pcimsi = to_pci_dev(dev)->msi_enabled;
447 :
448 0 : msi_lock_descs(dev);
449 0 : xa = &dev->msi.data->__domains[domid].store;
450 0 : desc = xa_load(xa, pcimsi ? 0 : index);
451 0 : if (desc && desc->irq) {
452 : /*
453 : * PCI-MSI has only one descriptor for multiple interrupts.
454 : * PCI-MSIX and platform MSI use a descriptor per
455 : * interrupt.
456 : */
457 0 : if (pcimsi) {
458 0 : if (index < desc->nvec_used)
459 0 : ret = desc->irq + index;
460 : } else {
461 : ret = desc->irq;
462 : }
463 : }
464 :
465 0 : msi_unlock_descs(dev);
466 0 : return ret;
467 : }
468 : EXPORT_SYMBOL_GPL(msi_domain_get_virq);
469 :
470 : #ifdef CONFIG_SYSFS
471 : static struct attribute *msi_dev_attrs[] = {
472 : NULL
473 : };
474 :
475 : static const struct attribute_group msi_irqs_group = {
476 : .name = "msi_irqs",
477 : .attrs = msi_dev_attrs,
478 : };
479 :
480 : static inline int msi_sysfs_create_group(struct device *dev)
481 : {
482 0 : return devm_device_add_group(dev, &msi_irqs_group);
483 : }
484 :
485 0 : static ssize_t msi_mode_show(struct device *dev, struct device_attribute *attr,
486 : char *buf)
487 : {
488 : /* MSI vs. MSIX is per device not per interrupt */
489 0 : bool is_msix = dev_is_pci(dev) ? to_pci_dev(dev)->msix_enabled : false;
490 :
491 0 : return sysfs_emit(buf, "%s\n", is_msix ? "msix" : "msi");
492 : }
493 :
494 0 : static void msi_sysfs_remove_desc(struct device *dev, struct msi_desc *desc)
495 : {
496 0 : struct device_attribute *attrs = desc->sysfs_attrs;
497 : int i;
498 :
499 0 : if (!attrs)
500 : return;
501 :
502 0 : desc->sysfs_attrs = NULL;
503 0 : for (i = 0; i < desc->nvec_used; i++) {
504 0 : if (attrs[i].show)
505 0 : sysfs_remove_file_from_group(&dev->kobj, &attrs[i].attr, msi_irqs_group.name);
506 0 : kfree(attrs[i].attr.name);
507 : }
508 0 : kfree(attrs);
509 : }
510 :
511 0 : static int msi_sysfs_populate_desc(struct device *dev, struct msi_desc *desc)
512 : {
513 : struct device_attribute *attrs;
514 : int ret, i;
515 :
516 0 : attrs = kcalloc(desc->nvec_used, sizeof(*attrs), GFP_KERNEL);
517 0 : if (!attrs)
518 : return -ENOMEM;
519 :
520 0 : desc->sysfs_attrs = attrs;
521 0 : for (i = 0; i < desc->nvec_used; i++) {
522 : sysfs_attr_init(&attrs[i].attr);
523 0 : attrs[i].attr.name = kasprintf(GFP_KERNEL, "%d", desc->irq + i);
524 0 : if (!attrs[i].attr.name) {
525 : ret = -ENOMEM;
526 : goto fail;
527 : }
528 :
529 0 : attrs[i].attr.mode = 0444;
530 0 : attrs[i].show = msi_mode_show;
531 :
532 0 : ret = sysfs_add_file_to_group(&dev->kobj, &attrs[i].attr, msi_irqs_group.name);
533 0 : if (ret) {
534 0 : attrs[i].show = NULL;
535 0 : goto fail;
536 : }
537 : }
538 : return 0;
539 :
540 : fail:
541 0 : msi_sysfs_remove_desc(dev, desc);
542 0 : return ret;
543 : }
544 :
545 : #if defined(CONFIG_PCI_MSI_ARCH_FALLBACKS) || defined(CONFIG_PCI_XEN)
546 : /**
547 : * msi_device_populate_sysfs - Populate msi_irqs sysfs entries for a device
548 : * @dev: The device (PCI, platform etc) which will get sysfs entries
549 : */
550 : int msi_device_populate_sysfs(struct device *dev)
551 : {
552 : struct msi_desc *desc;
553 : int ret;
554 :
555 : msi_for_each_desc(desc, dev, MSI_DESC_ASSOCIATED) {
556 : if (desc->sysfs_attrs)
557 : continue;
558 : ret = msi_sysfs_populate_desc(dev, desc);
559 : if (ret)
560 : return ret;
561 : }
562 : return 0;
563 : }
564 :
565 : /**
566 : * msi_device_destroy_sysfs - Destroy msi_irqs sysfs entries for a device
567 : * @dev: The device (PCI, platform etc) for which to remove
568 : * sysfs entries
569 : */
570 : void msi_device_destroy_sysfs(struct device *dev)
571 : {
572 : struct msi_desc *desc;
573 :
574 : msi_for_each_desc(desc, dev, MSI_DESC_ALL)
575 : msi_sysfs_remove_desc(dev, desc);
576 : }
577 : #endif /* CONFIG_PCI_MSI_ARCH_FALLBACK || CONFIG_PCI_XEN */
578 : #else /* CONFIG_SYSFS */
579 : static inline int msi_sysfs_create_group(struct device *dev) { return 0; }
580 : static inline int msi_sysfs_populate_desc(struct device *dev, struct msi_desc *desc) { return 0; }
581 : static inline void msi_sysfs_remove_desc(struct device *dev, struct msi_desc *desc) { }
582 : #endif /* !CONFIG_SYSFS */
583 :
584 0 : static struct irq_domain *msi_get_device_domain(struct device *dev, unsigned int domid)
585 : {
586 : struct irq_domain *domain;
587 :
588 : lockdep_assert_held(&dev->msi.data->mutex);
589 :
590 0 : if (WARN_ON_ONCE(domid >= MSI_MAX_DEVICE_IRQDOMAINS))
591 : return NULL;
592 :
593 0 : domain = dev->msi.data->__domains[domid].domain;
594 0 : if (!domain)
595 : return NULL;
596 :
597 0 : if (WARN_ON_ONCE(irq_domain_is_msi_parent(domain)))
598 : return NULL;
599 :
600 : return domain;
601 : }
602 :
603 : static unsigned int msi_domain_get_hwsize(struct device *dev, unsigned int domid)
604 : {
605 : struct msi_domain_info *info;
606 : struct irq_domain *domain;
607 :
608 0 : domain = msi_get_device_domain(dev, domid);
609 0 : if (domain) {
610 0 : info = domain->host_data;
611 0 : return info->hwsize;
612 : }
613 : /* No domain, default to MSI_XA_DOMAIN_SIZE */
614 : return MSI_XA_DOMAIN_SIZE;
615 : }
616 :
617 : static inline void irq_chip_write_msi_msg(struct irq_data *data,
618 : struct msi_msg *msg)
619 : {
620 0 : data->chip->irq_write_msi_msg(data, msg);
621 : }
622 :
623 0 : static void msi_check_level(struct irq_domain *domain, struct msi_msg *msg)
624 : {
625 0 : struct msi_domain_info *info = domain->host_data;
626 :
627 : /*
628 : * If the MSI provider has messed with the second message and
629 : * not advertized that it is level-capable, signal the breakage.
630 : */
631 0 : WARN_ON(!((info->flags & MSI_FLAG_LEVEL_CAPABLE) &&
632 : (info->chip->flags & IRQCHIP_SUPPORTS_LEVEL_MSI)) &&
633 : (msg[1].address_lo || msg[1].address_hi || msg[1].data));
634 0 : }
635 :
636 : /**
637 : * msi_domain_set_affinity - Generic affinity setter function for MSI domains
638 : * @irq_data: The irq data associated to the interrupt
639 : * @mask: The affinity mask to set
640 : * @force: Flag to enforce setting (disable online checks)
641 : *
642 : * Intended to be used by MSI interrupt controllers which are
643 : * implemented with hierarchical domains.
644 : *
645 : * Return: IRQ_SET_MASK_* result code
646 : */
647 0 : int msi_domain_set_affinity(struct irq_data *irq_data,
648 : const struct cpumask *mask, bool force)
649 : {
650 0 : struct irq_data *parent = irq_data->parent_data;
651 0 : struct msi_msg msg[2] = { [1] = { }, };
652 : int ret;
653 :
654 0 : ret = parent->chip->irq_set_affinity(parent, mask, force);
655 0 : if (ret >= 0 && ret != IRQ_SET_MASK_OK_DONE) {
656 0 : BUG_ON(irq_chip_compose_msi_msg(irq_data, msg));
657 0 : msi_check_level(irq_data->domain, msg);
658 : irq_chip_write_msi_msg(irq_data, msg);
659 : }
660 :
661 0 : return ret;
662 : }
663 :
664 0 : static int msi_domain_activate(struct irq_domain *domain,
665 : struct irq_data *irq_data, bool early)
666 : {
667 0 : struct msi_msg msg[2] = { [1] = { }, };
668 :
669 0 : BUG_ON(irq_chip_compose_msi_msg(irq_data, msg));
670 0 : msi_check_level(irq_data->domain, msg);
671 0 : irq_chip_write_msi_msg(irq_data, msg);
672 0 : return 0;
673 : }
674 :
675 0 : static void msi_domain_deactivate(struct irq_domain *domain,
676 : struct irq_data *irq_data)
677 : {
678 : struct msi_msg msg[2];
679 :
680 0 : memset(msg, 0, sizeof(msg));
681 0 : irq_chip_write_msi_msg(irq_data, msg);
682 0 : }
683 :
684 0 : static int msi_domain_alloc(struct irq_domain *domain, unsigned int virq,
685 : unsigned int nr_irqs, void *arg)
686 : {
687 0 : struct msi_domain_info *info = domain->host_data;
688 0 : struct msi_domain_ops *ops = info->ops;
689 0 : irq_hw_number_t hwirq = ops->get_hwirq(info, arg);
690 : int i, ret;
691 :
692 0 : if (irq_find_mapping(domain, hwirq) > 0)
693 : return -EEXIST;
694 :
695 0 : if (domain->parent) {
696 0 : ret = irq_domain_alloc_irqs_parent(domain, virq, nr_irqs, arg);
697 0 : if (ret < 0)
698 : return ret;
699 : }
700 :
701 0 : for (i = 0; i < nr_irqs; i++) {
702 0 : ret = ops->msi_init(domain, info, virq + i, hwirq + i, arg);
703 0 : if (ret < 0) {
704 0 : if (ops->msi_free) {
705 0 : for (i--; i > 0; i--)
706 0 : ops->msi_free(domain, info, virq + i);
707 : }
708 0 : irq_domain_free_irqs_top(domain, virq, nr_irqs);
709 0 : return ret;
710 : }
711 : }
712 :
713 : return 0;
714 : }
715 :
716 0 : static void msi_domain_free(struct irq_domain *domain, unsigned int virq,
717 : unsigned int nr_irqs)
718 : {
719 0 : struct msi_domain_info *info = domain->host_data;
720 : int i;
721 :
722 0 : if (info->ops->msi_free) {
723 0 : for (i = 0; i < nr_irqs; i++)
724 0 : info->ops->msi_free(domain, info, virq + i);
725 : }
726 0 : irq_domain_free_irqs_top(domain, virq, nr_irqs);
727 0 : }
728 :
729 : static const struct irq_domain_ops msi_domain_ops = {
730 : .alloc = msi_domain_alloc,
731 : .free = msi_domain_free,
732 : .activate = msi_domain_activate,
733 : .deactivate = msi_domain_deactivate,
734 : };
735 :
736 0 : static irq_hw_number_t msi_domain_ops_get_hwirq(struct msi_domain_info *info,
737 : msi_alloc_info_t *arg)
738 : {
739 0 : return arg->hwirq;
740 : }
741 :
742 0 : static int msi_domain_ops_prepare(struct irq_domain *domain, struct device *dev,
743 : int nvec, msi_alloc_info_t *arg)
744 : {
745 0 : memset(arg, 0, sizeof(*arg));
746 0 : return 0;
747 : }
748 :
749 0 : static void msi_domain_ops_set_desc(msi_alloc_info_t *arg,
750 : struct msi_desc *desc)
751 : {
752 0 : arg->desc = desc;
753 0 : }
754 :
755 0 : static int msi_domain_ops_init(struct irq_domain *domain,
756 : struct msi_domain_info *info,
757 : unsigned int virq, irq_hw_number_t hwirq,
758 : msi_alloc_info_t *arg)
759 : {
760 0 : irq_domain_set_hwirq_and_chip(domain, virq, hwirq, info->chip,
761 : info->chip_data);
762 0 : if (info->handler && info->handler_name) {
763 0 : __irq_set_handler(virq, info->handler, 0, info->handler_name);
764 0 : if (info->handler_data)
765 0 : irq_set_handler_data(virq, info->handler_data);
766 : }
767 0 : return 0;
768 : }
769 :
770 : static struct msi_domain_ops msi_domain_ops_default = {
771 : .get_hwirq = msi_domain_ops_get_hwirq,
772 : .msi_init = msi_domain_ops_init,
773 : .msi_prepare = msi_domain_ops_prepare,
774 : .set_desc = msi_domain_ops_set_desc,
775 : };
776 :
777 0 : static void msi_domain_update_dom_ops(struct msi_domain_info *info)
778 : {
779 0 : struct msi_domain_ops *ops = info->ops;
780 :
781 0 : if (ops == NULL) {
782 0 : info->ops = &msi_domain_ops_default;
783 : return;
784 : }
785 :
786 0 : if (!(info->flags & MSI_FLAG_USE_DEF_DOM_OPS))
787 : return;
788 :
789 0 : if (ops->get_hwirq == NULL)
790 0 : ops->get_hwirq = msi_domain_ops_default.get_hwirq;
791 0 : if (ops->msi_init == NULL)
792 0 : ops->msi_init = msi_domain_ops_default.msi_init;
793 0 : if (ops->msi_prepare == NULL)
794 0 : ops->msi_prepare = msi_domain_ops_default.msi_prepare;
795 0 : if (ops->set_desc == NULL)
796 0 : ops->set_desc = msi_domain_ops_default.set_desc;
797 : }
798 :
799 0 : static void msi_domain_update_chip_ops(struct msi_domain_info *info)
800 : {
801 0 : struct irq_chip *chip = info->chip;
802 :
803 0 : BUG_ON(!chip || !chip->irq_mask || !chip->irq_unmask);
804 0 : if (!chip->irq_set_affinity)
805 0 : chip->irq_set_affinity = msi_domain_set_affinity;
806 0 : }
807 :
808 0 : static struct irq_domain *__msi_create_irq_domain(struct fwnode_handle *fwnode,
809 : struct msi_domain_info *info,
810 : unsigned int flags,
811 : struct irq_domain *parent)
812 : {
813 : struct irq_domain *domain;
814 :
815 0 : if (info->hwsize > MSI_XA_DOMAIN_SIZE)
816 : return NULL;
817 :
818 : /*
819 : * Hardware size 0 is valid for backwards compatibility and for
820 : * domains which are not backed by a hardware table. Grant the
821 : * maximum index space.
822 : */
823 0 : if (!info->hwsize)
824 0 : info->hwsize = MSI_XA_DOMAIN_SIZE;
825 :
826 0 : msi_domain_update_dom_ops(info);
827 0 : if (info->flags & MSI_FLAG_USE_DEF_CHIP_OPS)
828 0 : msi_domain_update_chip_ops(info);
829 :
830 0 : domain = irq_domain_create_hierarchy(parent, flags | IRQ_DOMAIN_FLAG_MSI, 0,
831 : fwnode, &msi_domain_ops, info);
832 :
833 0 : if (domain)
834 0 : irq_domain_update_bus_token(domain, info->bus_token);
835 :
836 : return domain;
837 : }
838 :
839 : /**
840 : * msi_create_irq_domain - Create an MSI interrupt domain
841 : * @fwnode: Optional fwnode of the interrupt controller
842 : * @info: MSI domain info
843 : * @parent: Parent irq domain
844 : *
845 : * Return: pointer to the created &struct irq_domain or %NULL on failure
846 : */
847 0 : struct irq_domain *msi_create_irq_domain(struct fwnode_handle *fwnode,
848 : struct msi_domain_info *info,
849 : struct irq_domain *parent)
850 : {
851 0 : return __msi_create_irq_domain(fwnode, info, 0, parent);
852 : }
853 :
854 : /**
855 : * msi_parent_init_dev_msi_info - Delegate initialization of device MSI info down
856 : * in the domain hierarchy
857 : * @dev: The device for which the domain should be created
858 : * @domain: The domain in the hierarchy this op is being called on
859 : * @msi_parent_domain: The IRQ_DOMAIN_FLAG_MSI_PARENT domain for the child to
860 : * be created
861 : * @msi_child_info: The MSI domain info of the IRQ_DOMAIN_FLAG_MSI_DEVICE
862 : * domain to be created
863 : *
864 : * Return: true on success, false otherwise
865 : *
866 : * This is the most complex problem of per device MSI domains and the
867 : * underlying interrupt domain hierarchy:
868 : *
869 : * The device domain to be initialized requests the broadest feature set
870 : * possible and the underlying domain hierarchy puts restrictions on it.
871 : *
872 : * That's trivial for a simple parent->child relationship, but it gets
873 : * interesting with an intermediate domain: root->parent->child. The
874 : * intermediate 'parent' can expand the capabilities which the 'root'
875 : * domain is providing. So that creates a classic hen and egg problem:
876 : * Which entity is doing the restrictions/expansions?
877 : *
878 : * One solution is to let the root domain handle the initialization that's
879 : * why there is the @domain and the @msi_parent_domain pointer.
880 : */
881 0 : bool msi_parent_init_dev_msi_info(struct device *dev, struct irq_domain *domain,
882 : struct irq_domain *msi_parent_domain,
883 : struct msi_domain_info *msi_child_info)
884 : {
885 0 : struct irq_domain *parent = domain->parent;
886 :
887 0 : if (WARN_ON_ONCE(!parent || !parent->msi_parent_ops ||
888 : !parent->msi_parent_ops->init_dev_msi_info))
889 : return false;
890 :
891 0 : return parent->msi_parent_ops->init_dev_msi_info(dev, parent, msi_parent_domain,
892 : msi_child_info);
893 : }
894 :
895 : /**
896 : * msi_create_device_irq_domain - Create a device MSI interrupt domain
897 : * @dev: Pointer to the device
898 : * @domid: Domain id
899 : * @template: MSI domain info bundle used as template
900 : * @hwsize: Maximum number of MSI table entries (0 if unknown or unlimited)
901 : * @domain_data: Optional pointer to domain specific data which is set in
902 : * msi_domain_info::data
903 : * @chip_data: Optional pointer to chip specific data which is set in
904 : * msi_domain_info::chip_data
905 : *
906 : * Return: True on success, false otherwise
907 : *
908 : * There is no firmware node required for this interface because the per
909 : * device domains are software constructs which are actually closer to the
910 : * hardware reality than any firmware can describe them.
911 : *
912 : * The domain name and the irq chip name for a MSI device domain are
913 : * composed by: "$(PREFIX)$(CHIPNAME)-$(DEVNAME)"
914 : *
915 : * $PREFIX: Optional prefix provided by the underlying MSI parent domain
916 : * via msi_parent_ops::prefix. If that pointer is NULL the prefix
917 : * is empty.
918 : * $CHIPNAME: The name of the irq_chip in @template
919 : * $DEVNAME: The name of the device
920 : *
921 : * This results in understandable chip names and hardware interrupt numbers
922 : * in e.g. /proc/interrupts
923 : *
924 : * PCI-MSI-0000:00:1c.0 0-edge Parent domain has no prefix
925 : * IR-PCI-MSI-0000:00:1c.4 0-edge Same with interrupt remapping prefix 'IR-'
926 : *
927 : * IR-PCI-MSIX-0000:3d:00.0 0-edge Hardware interrupt numbers reflect
928 : * IR-PCI-MSIX-0000:3d:00.0 1-edge the real MSI-X index on that device
929 : * IR-PCI-MSIX-0000:3d:00.0 2-edge
930 : *
931 : * On IMS domains the hardware interrupt number is either a table entry
932 : * index or a purely software managed index but it is guaranteed to be
933 : * unique.
934 : *
935 : * The domain pointer is stored in @dev::msi::data::__irqdomains[]. All
936 : * subsequent operations on the domain depend on the domain id.
937 : *
938 : * The domain is automatically freed when the device is removed via devres
939 : * in the context of @dev::msi::data freeing, but it can also be
940 : * independently removed via @msi_remove_device_irq_domain().
941 : */
942 0 : bool msi_create_device_irq_domain(struct device *dev, unsigned int domid,
943 : const struct msi_domain_template *template,
944 : unsigned int hwsize, void *domain_data,
945 : void *chip_data)
946 : {
947 0 : struct irq_domain *domain, *parent = dev->msi.domain;
948 : const struct msi_parent_ops *pops;
949 : struct msi_domain_template *bundle;
950 : struct fwnode_handle *fwnode;
951 :
952 0 : if (!irq_domain_is_msi_parent(parent))
953 : return false;
954 :
955 0 : if (domid >= MSI_MAX_DEVICE_IRQDOMAINS)
956 : return false;
957 :
958 0 : bundle = kmemdup(template, sizeof(*bundle), GFP_KERNEL);
959 0 : if (!bundle)
960 : return false;
961 :
962 0 : bundle->info.hwsize = hwsize;
963 0 : bundle->info.chip = &bundle->chip;
964 0 : bundle->info.ops = &bundle->ops;
965 0 : bundle->info.data = domain_data;
966 0 : bundle->info.chip_data = chip_data;
967 :
968 0 : pops = parent->msi_parent_ops;
969 0 : snprintf(bundle->name, sizeof(bundle->name), "%s%s-%s",
970 0 : pops->prefix ? : "", bundle->chip.name, dev_name(dev));
971 0 : bundle->chip.name = bundle->name;
972 :
973 0 : fwnode = irq_domain_alloc_named_fwnode(bundle->name);
974 0 : if (!fwnode)
975 : goto free_bundle;
976 :
977 0 : if (msi_setup_device_data(dev))
978 : goto free_fwnode;
979 :
980 0 : msi_lock_descs(dev);
981 :
982 0 : if (WARN_ON_ONCE(msi_get_device_domain(dev, domid)))
983 : goto fail;
984 :
985 0 : if (!pops->init_dev_msi_info(dev, parent, parent, &bundle->info))
986 : goto fail;
987 :
988 0 : domain = __msi_create_irq_domain(fwnode, &bundle->info, IRQ_DOMAIN_FLAG_MSI_DEVICE, parent);
989 0 : if (!domain)
990 : goto fail;
991 :
992 0 : domain->dev = dev;
993 0 : dev->msi.data->__domains[domid].domain = domain;
994 0 : msi_unlock_descs(dev);
995 0 : return true;
996 :
997 : fail:
998 : msi_unlock_descs(dev);
999 : free_fwnode:
1000 0 : irq_domain_free_fwnode(fwnode);
1001 : free_bundle:
1002 0 : kfree(bundle);
1003 0 : return false;
1004 : }
1005 :
1006 : /**
1007 : * msi_remove_device_irq_domain - Free a device MSI interrupt domain
1008 : * @dev: Pointer to the device
1009 : * @domid: Domain id
1010 : */
1011 0 : void msi_remove_device_irq_domain(struct device *dev, unsigned int domid)
1012 : {
1013 0 : struct fwnode_handle *fwnode = NULL;
1014 : struct msi_domain_info *info;
1015 : struct irq_domain *domain;
1016 :
1017 0 : msi_lock_descs(dev);
1018 :
1019 0 : domain = msi_get_device_domain(dev, domid);
1020 :
1021 0 : if (!domain || !irq_domain_is_msi_device(domain))
1022 : goto unlock;
1023 :
1024 0 : dev->msi.data->__domains[domid].domain = NULL;
1025 0 : info = domain->host_data;
1026 0 : if (irq_domain_is_msi_device(domain))
1027 0 : fwnode = domain->fwnode;
1028 0 : irq_domain_remove(domain);
1029 0 : irq_domain_free_fwnode(fwnode);
1030 0 : kfree(container_of(info, struct msi_domain_template, info));
1031 :
1032 : unlock:
1033 0 : msi_unlock_descs(dev);
1034 0 : }
1035 :
1036 : /**
1037 : * msi_match_device_irq_domain - Match a device irq domain against a bus token
1038 : * @dev: Pointer to the device
1039 : * @domid: Domain id
1040 : * @bus_token: Bus token to match against the domain bus token
1041 : *
1042 : * Return: True if device domain exists and bus tokens match.
1043 : */
1044 0 : bool msi_match_device_irq_domain(struct device *dev, unsigned int domid,
1045 : enum irq_domain_bus_token bus_token)
1046 : {
1047 : struct msi_domain_info *info;
1048 : struct irq_domain *domain;
1049 0 : bool ret = false;
1050 :
1051 0 : msi_lock_descs(dev);
1052 0 : domain = msi_get_device_domain(dev, domid);
1053 0 : if (domain && irq_domain_is_msi_device(domain)) {
1054 0 : info = domain->host_data;
1055 0 : ret = info->bus_token == bus_token;
1056 : }
1057 0 : msi_unlock_descs(dev);
1058 0 : return ret;
1059 : }
1060 :
1061 0 : int msi_domain_prepare_irqs(struct irq_domain *domain, struct device *dev,
1062 : int nvec, msi_alloc_info_t *arg)
1063 : {
1064 0 : struct msi_domain_info *info = domain->host_data;
1065 0 : struct msi_domain_ops *ops = info->ops;
1066 :
1067 0 : return ops->msi_prepare(domain, dev, nvec, arg);
1068 : }
1069 :
1070 0 : int msi_domain_populate_irqs(struct irq_domain *domain, struct device *dev,
1071 : int virq_base, int nvec, msi_alloc_info_t *arg)
1072 : {
1073 0 : struct msi_domain_info *info = domain->host_data;
1074 0 : struct msi_domain_ops *ops = info->ops;
1075 0 : struct msi_ctrl ctrl = {
1076 : .domid = MSI_DEFAULT_DOMAIN,
1077 : .first = virq_base,
1078 0 : .last = virq_base + nvec - 1,
1079 : };
1080 : struct msi_desc *desc;
1081 : struct xarray *xa;
1082 : int ret, virq;
1083 :
1084 0 : msi_lock_descs(dev);
1085 :
1086 0 : if (!msi_ctrl_valid(dev, &ctrl)) {
1087 : ret = -EINVAL;
1088 : goto unlock;
1089 : }
1090 :
1091 0 : ret = msi_domain_add_simple_msi_descs(dev, &ctrl);
1092 0 : if (ret)
1093 : goto unlock;
1094 :
1095 0 : xa = &dev->msi.data->__domains[ctrl.domid].store;
1096 :
1097 0 : for (virq = virq_base; virq < virq_base + nvec; virq++) {
1098 0 : desc = xa_load(xa, virq);
1099 0 : desc->irq = virq;
1100 :
1101 0 : ops->set_desc(arg, desc);
1102 0 : ret = irq_domain_alloc_irqs_hierarchy(domain, virq, 1, arg);
1103 0 : if (ret)
1104 : goto fail;
1105 :
1106 0 : irq_set_msi_desc(virq, desc);
1107 : }
1108 0 : msi_unlock_descs(dev);
1109 0 : return 0;
1110 :
1111 : fail:
1112 0 : for (--virq; virq >= virq_base; virq--) {
1113 0 : msi_domain_depopulate_descs(dev, virq, 1);
1114 0 : irq_domain_free_irqs_common(domain, virq, 1);
1115 : }
1116 0 : msi_domain_free_descs(dev, &ctrl);
1117 : unlock:
1118 0 : msi_unlock_descs(dev);
1119 0 : return ret;
1120 : }
1121 :
1122 0 : void msi_domain_depopulate_descs(struct device *dev, int virq_base, int nvec)
1123 : {
1124 0 : struct msi_ctrl ctrl = {
1125 : .domid = MSI_DEFAULT_DOMAIN,
1126 : .first = virq_base,
1127 0 : .last = virq_base + nvec - 1,
1128 : };
1129 : struct msi_desc *desc;
1130 : struct xarray *xa;
1131 : unsigned long idx;
1132 :
1133 0 : if (!msi_ctrl_valid(dev, &ctrl))
1134 0 : return;
1135 :
1136 0 : xa = &dev->msi.data->__domains[ctrl.domid].store;
1137 0 : xa_for_each_range(xa, idx, desc, ctrl.first, ctrl.last)
1138 0 : desc->irq = 0;
1139 : }
1140 :
1141 : /*
1142 : * Carefully check whether the device can use reservation mode. If
1143 : * reservation mode is enabled then the early activation will assign a
1144 : * dummy vector to the device. If the PCI/MSI device does not support
1145 : * masking of the entry then this can result in spurious interrupts when
1146 : * the device driver is not absolutely careful. But even then a malfunction
1147 : * of the hardware could result in a spurious interrupt on the dummy vector
1148 : * and render the device unusable. If the entry can be masked then the core
1149 : * logic will prevent the spurious interrupt and reservation mode can be
1150 : * used. For now reservation mode is restricted to PCI/MSI.
1151 : */
1152 0 : static bool msi_check_reservation_mode(struct irq_domain *domain,
1153 : struct msi_domain_info *info,
1154 : struct device *dev)
1155 : {
1156 : struct msi_desc *desc;
1157 :
1158 0 : switch(domain->bus_token) {
1159 : case DOMAIN_BUS_PCI_MSI:
1160 : case DOMAIN_BUS_PCI_DEVICE_MSI:
1161 : case DOMAIN_BUS_PCI_DEVICE_MSIX:
1162 : case DOMAIN_BUS_VMD_MSI:
1163 : break;
1164 : default:
1165 : return false;
1166 : }
1167 :
1168 0 : if (!(info->flags & MSI_FLAG_MUST_REACTIVATE))
1169 : return false;
1170 :
1171 0 : if (IS_ENABLED(CONFIG_PCI_MSI) && pci_msi_ignore_mask)
1172 : return false;
1173 :
1174 : /*
1175 : * Checking the first MSI descriptor is sufficient. MSIX supports
1176 : * masking and MSI does so when the can_mask attribute is set.
1177 : */
1178 0 : desc = msi_first_desc(dev, MSI_DESC_ALL);
1179 0 : return desc->pci.msi_attrib.is_msix || desc->pci.msi_attrib.can_mask;
1180 : }
1181 :
1182 : static int msi_handle_pci_fail(struct irq_domain *domain, struct msi_desc *desc,
1183 : int allocated)
1184 : {
1185 0 : switch(domain->bus_token) {
1186 : case DOMAIN_BUS_PCI_MSI:
1187 : case DOMAIN_BUS_PCI_DEVICE_MSI:
1188 : case DOMAIN_BUS_PCI_DEVICE_MSIX:
1189 : case DOMAIN_BUS_VMD_MSI:
1190 : if (IS_ENABLED(CONFIG_PCI_MSI))
1191 : break;
1192 : fallthrough;
1193 : default:
1194 : return -ENOSPC;
1195 : }
1196 :
1197 : /* Let a failed PCI multi MSI allocation retry */
1198 0 : if (desc->nvec_used > 1)
1199 : return 1;
1200 :
1201 : /* If there was a successful allocation let the caller know */
1202 0 : return allocated ? allocated : -ENOSPC;
1203 : }
1204 :
1205 : #define VIRQ_CAN_RESERVE 0x01
1206 : #define VIRQ_ACTIVATE 0x02
1207 : #define VIRQ_NOMASK_QUIRK 0x04
1208 :
1209 0 : static int msi_init_virq(struct irq_domain *domain, int virq, unsigned int vflags)
1210 : {
1211 0 : struct irq_data *irqd = irq_domain_get_irq_data(domain, virq);
1212 : int ret;
1213 :
1214 0 : if (!(vflags & VIRQ_CAN_RESERVE)) {
1215 0 : irqd_clr_can_reserve(irqd);
1216 0 : if (vflags & VIRQ_NOMASK_QUIRK)
1217 0 : irqd_set_msi_nomask_quirk(irqd);
1218 :
1219 : /*
1220 : * If the interrupt is managed but no CPU is available to
1221 : * service it, shut it down until better times. Note that
1222 : * we only do this on the !RESERVE path as x86 (the only
1223 : * architecture using this flag) deals with this in a
1224 : * different way by using a catch-all vector.
1225 : */
1226 0 : if ((vflags & VIRQ_ACTIVATE) &&
1227 0 : irqd_affinity_is_managed(irqd) &&
1228 0 : !cpumask_intersects(irq_data_get_affinity_mask(irqd),
1229 : cpu_online_mask)) {
1230 0 : irqd_set_managed_shutdown(irqd);
1231 0 : return 0;
1232 : }
1233 : }
1234 :
1235 0 : if (!(vflags & VIRQ_ACTIVATE))
1236 : return 0;
1237 :
1238 0 : ret = irq_domain_activate_irq(irqd, vflags & VIRQ_CAN_RESERVE);
1239 0 : if (ret)
1240 : return ret;
1241 : /*
1242 : * If the interrupt uses reservation mode, clear the activated bit
1243 : * so request_irq() will assign the final vector.
1244 : */
1245 0 : if (vflags & VIRQ_CAN_RESERVE)
1246 0 : irqd_clr_activated(irqd);
1247 : return 0;
1248 : }
1249 :
1250 0 : static int __msi_domain_alloc_irqs(struct device *dev, struct irq_domain *domain,
1251 : struct msi_ctrl *ctrl)
1252 : {
1253 0 : struct xarray *xa = &dev->msi.data->__domains[ctrl->domid].store;
1254 0 : struct msi_domain_info *info = domain->host_data;
1255 0 : struct msi_domain_ops *ops = info->ops;
1256 0 : unsigned int vflags = 0, allocated = 0;
1257 0 : msi_alloc_info_t arg = { };
1258 : struct msi_desc *desc;
1259 : unsigned long idx;
1260 : int i, ret, virq;
1261 :
1262 0 : ret = msi_domain_prepare_irqs(domain, dev, ctrl->nirqs, &arg);
1263 0 : if (ret)
1264 : return ret;
1265 :
1266 : /*
1267 : * This flag is set by the PCI layer as we need to activate
1268 : * the MSI entries before the PCI layer enables MSI in the
1269 : * card. Otherwise the card latches a random msi message.
1270 : */
1271 0 : if (info->flags & MSI_FLAG_ACTIVATE_EARLY)
1272 0 : vflags |= VIRQ_ACTIVATE;
1273 :
1274 : /*
1275 : * Interrupt can use a reserved vector and will not occupy
1276 : * a real device vector until the interrupt is requested.
1277 : */
1278 0 : if (msi_check_reservation_mode(domain, info, dev)) {
1279 0 : vflags |= VIRQ_CAN_RESERVE;
1280 : /*
1281 : * MSI affinity setting requires a special quirk (X86) when
1282 : * reservation mode is active.
1283 : */
1284 0 : if (info->flags & MSI_FLAG_NOMASK_QUIRK)
1285 0 : vflags |= VIRQ_NOMASK_QUIRK;
1286 : }
1287 :
1288 0 : xa_for_each_range(xa, idx, desc, ctrl->first, ctrl->last) {
1289 0 : if (!msi_desc_match(desc, MSI_DESC_NOTASSOCIATED))
1290 0 : continue;
1291 :
1292 : /* This should return -ECONFUSED... */
1293 0 : if (WARN_ON_ONCE(allocated >= ctrl->nirqs))
1294 : return -EINVAL;
1295 :
1296 0 : if (ops->prepare_desc)
1297 0 : ops->prepare_desc(domain, &arg, desc);
1298 :
1299 0 : ops->set_desc(&arg, desc);
1300 :
1301 0 : virq = __irq_domain_alloc_irqs(domain, -1, desc->nvec_used,
1302 : dev_to_node(dev), &arg, false,
1303 0 : desc->affinity);
1304 0 : if (virq < 0)
1305 0 : return msi_handle_pci_fail(domain, desc, allocated);
1306 :
1307 0 : for (i = 0; i < desc->nvec_used; i++) {
1308 0 : irq_set_msi_desc_off(virq, i, desc);
1309 0 : irq_debugfs_copy_devname(virq + i, dev);
1310 0 : ret = msi_init_virq(domain, virq + i, vflags);
1311 0 : if (ret)
1312 : return ret;
1313 : }
1314 0 : if (info->flags & MSI_FLAG_DEV_SYSFS) {
1315 0 : ret = msi_sysfs_populate_desc(dev, desc);
1316 0 : if (ret)
1317 : return ret;
1318 : }
1319 0 : allocated++;
1320 : }
1321 : return 0;
1322 : }
1323 :
1324 : static int msi_domain_alloc_simple_msi_descs(struct device *dev,
1325 : struct msi_domain_info *info,
1326 : struct msi_ctrl *ctrl)
1327 : {
1328 0 : if (!(info->flags & MSI_FLAG_ALLOC_SIMPLE_MSI_DESCS))
1329 : return 0;
1330 :
1331 0 : return msi_domain_add_simple_msi_descs(dev, ctrl);
1332 : }
1333 :
1334 0 : static int __msi_domain_alloc_locked(struct device *dev, struct msi_ctrl *ctrl)
1335 : {
1336 : struct msi_domain_info *info;
1337 : struct msi_domain_ops *ops;
1338 : struct irq_domain *domain;
1339 : int ret;
1340 :
1341 0 : if (!msi_ctrl_valid(dev, ctrl))
1342 : return -EINVAL;
1343 :
1344 0 : domain = msi_get_device_domain(dev, ctrl->domid);
1345 0 : if (!domain)
1346 : return -ENODEV;
1347 :
1348 0 : info = domain->host_data;
1349 :
1350 0 : ret = msi_domain_alloc_simple_msi_descs(dev, info, ctrl);
1351 0 : if (ret)
1352 : return ret;
1353 :
1354 0 : ops = info->ops;
1355 0 : if (ops->domain_alloc_irqs)
1356 0 : return ops->domain_alloc_irqs(domain, dev, ctrl->nirqs);
1357 :
1358 0 : return __msi_domain_alloc_irqs(dev, domain, ctrl);
1359 : }
1360 :
1361 0 : static int msi_domain_alloc_locked(struct device *dev, struct msi_ctrl *ctrl)
1362 : {
1363 0 : int ret = __msi_domain_alloc_locked(dev, ctrl);
1364 :
1365 0 : if (ret)
1366 0 : msi_domain_free_locked(dev, ctrl);
1367 0 : return ret;
1368 : }
1369 :
1370 : /**
1371 : * msi_domain_alloc_irqs_range_locked - Allocate interrupts from a MSI interrupt domain
1372 : * @dev: Pointer to device struct of the device for which the interrupts
1373 : * are allocated
1374 : * @domid: Id of the interrupt domain to operate on
1375 : * @first: First index to allocate (inclusive)
1376 : * @last: Last index to allocate (inclusive)
1377 : *
1378 : * Must be invoked from within a msi_lock_descs() / msi_unlock_descs()
1379 : * pair. Use this for MSI irqdomains which implement their own descriptor
1380 : * allocation/free.
1381 : *
1382 : * Return: %0 on success or an error code.
1383 : */
1384 0 : int msi_domain_alloc_irqs_range_locked(struct device *dev, unsigned int domid,
1385 : unsigned int first, unsigned int last)
1386 : {
1387 0 : struct msi_ctrl ctrl = {
1388 : .domid = domid,
1389 : .first = first,
1390 : .last = last,
1391 0 : .nirqs = last + 1 - first,
1392 : };
1393 :
1394 0 : return msi_domain_alloc_locked(dev, &ctrl);
1395 : }
1396 :
1397 : /**
1398 : * msi_domain_alloc_irqs_range - Allocate interrupts from a MSI interrupt domain
1399 : * @dev: Pointer to device struct of the device for which the interrupts
1400 : * are allocated
1401 : * @domid: Id of the interrupt domain to operate on
1402 : * @first: First index to allocate (inclusive)
1403 : * @last: Last index to allocate (inclusive)
1404 : *
1405 : * Return: %0 on success or an error code.
1406 : */
1407 0 : int msi_domain_alloc_irqs_range(struct device *dev, unsigned int domid,
1408 : unsigned int first, unsigned int last)
1409 : {
1410 : int ret;
1411 :
1412 0 : msi_lock_descs(dev);
1413 0 : ret = msi_domain_alloc_irqs_range_locked(dev, domid, first, last);
1414 0 : msi_unlock_descs(dev);
1415 0 : return ret;
1416 : }
1417 :
1418 : /**
1419 : * msi_domain_alloc_irqs_all_locked - Allocate all interrupts from a MSI interrupt domain
1420 : *
1421 : * @dev: Pointer to device struct of the device for which the interrupts
1422 : * are allocated
1423 : * @domid: Id of the interrupt domain to operate on
1424 : * @nirqs: The number of interrupts to allocate
1425 : *
1426 : * This function scans all MSI descriptors of the MSI domain and allocates interrupts
1427 : * for all unassigned ones. That function is to be used for MSI domain usage where
1428 : * the descriptor allocation is handled at the call site, e.g. PCI/MSI[X].
1429 : *
1430 : * Return: %0 on success or an error code.
1431 : */
1432 0 : int msi_domain_alloc_irqs_all_locked(struct device *dev, unsigned int domid, int nirqs)
1433 : {
1434 0 : struct msi_ctrl ctrl = {
1435 : .domid = domid,
1436 : .first = 0,
1437 0 : .last = msi_domain_get_hwsize(dev, domid) - 1,
1438 : .nirqs = nirqs,
1439 : };
1440 :
1441 0 : return msi_domain_alloc_locked(dev, &ctrl);
1442 : }
1443 :
1444 : /**
1445 : * msi_domain_alloc_irq_at - Allocate an interrupt from a MSI interrupt domain at
1446 : * a given index - or at the next free index
1447 : *
1448 : * @dev: Pointer to device struct of the device for which the interrupts
1449 : * are allocated
1450 : * @domid: Id of the interrupt domain to operate on
1451 : * @index: Index for allocation. If @index == %MSI_ANY_INDEX the allocation
1452 : * uses the next free index.
1453 : * @affdesc: Optional pointer to an interrupt affinity descriptor structure
1454 : * @icookie: Optional pointer to a domain specific per instance cookie. If
1455 : * non-NULL the content of the cookie is stored in msi_desc::data.
1456 : * Must be NULL for MSI-X allocations
1457 : *
1458 : * This requires a MSI interrupt domain which lets the core code manage the
1459 : * MSI descriptors.
1460 : *
1461 : * Return: struct msi_map
1462 : *
1463 : * On success msi_map::index contains the allocated index number and
1464 : * msi_map::virq the corresponding Linux interrupt number
1465 : *
1466 : * On failure msi_map::index contains the error code and msi_map::virq
1467 : * is %0.
1468 : */
1469 0 : struct msi_map msi_domain_alloc_irq_at(struct device *dev, unsigned int domid, unsigned int index,
1470 : const struct irq_affinity_desc *affdesc,
1471 : union msi_instance_cookie *icookie)
1472 : {
1473 0 : struct msi_ctrl ctrl = { .domid = domid, .nirqs = 1, };
1474 : struct irq_domain *domain;
1475 0 : struct msi_map map = { };
1476 : struct msi_desc *desc;
1477 : int ret;
1478 :
1479 0 : msi_lock_descs(dev);
1480 0 : domain = msi_get_device_domain(dev, domid);
1481 0 : if (!domain) {
1482 : map.index = -ENODEV;
1483 : goto unlock;
1484 : }
1485 :
1486 0 : desc = msi_alloc_desc(dev, 1, affdesc);
1487 0 : if (!desc) {
1488 : map.index = -ENOMEM;
1489 : goto unlock;
1490 : }
1491 :
1492 0 : if (icookie)
1493 0 : desc->data.icookie = *icookie;
1494 :
1495 0 : ret = msi_insert_desc(dev, desc, domid, index);
1496 0 : if (ret) {
1497 : map.index = ret;
1498 : goto unlock;
1499 : }
1500 :
1501 0 : ctrl.first = ctrl.last = desc->msi_index;
1502 :
1503 0 : ret = __msi_domain_alloc_irqs(dev, domain, &ctrl);
1504 0 : if (ret) {
1505 0 : map.index = ret;
1506 0 : msi_domain_free_locked(dev, &ctrl);
1507 : } else {
1508 0 : map.index = desc->msi_index;
1509 0 : map.virq = desc->irq;
1510 : }
1511 : unlock:
1512 0 : msi_unlock_descs(dev);
1513 0 : return map;
1514 : }
1515 :
1516 0 : static void __msi_domain_free_irqs(struct device *dev, struct irq_domain *domain,
1517 : struct msi_ctrl *ctrl)
1518 : {
1519 0 : struct xarray *xa = &dev->msi.data->__domains[ctrl->domid].store;
1520 0 : struct msi_domain_info *info = domain->host_data;
1521 : struct irq_data *irqd;
1522 : struct msi_desc *desc;
1523 : unsigned long idx;
1524 : int i;
1525 :
1526 0 : xa_for_each_range(xa, idx, desc, ctrl->first, ctrl->last) {
1527 : /* Only handle MSI entries which have an interrupt associated */
1528 0 : if (!msi_desc_match(desc, MSI_DESC_ASSOCIATED))
1529 0 : continue;
1530 :
1531 : /* Make sure all interrupts are deactivated */
1532 0 : for (i = 0; i < desc->nvec_used; i++) {
1533 0 : irqd = irq_domain_get_irq_data(domain, desc->irq + i);
1534 0 : if (irqd && irqd_is_activated(irqd))
1535 0 : irq_domain_deactivate_irq(irqd);
1536 : }
1537 :
1538 0 : irq_domain_free_irqs(desc->irq, desc->nvec_used);
1539 0 : if (info->flags & MSI_FLAG_DEV_SYSFS)
1540 0 : msi_sysfs_remove_desc(dev, desc);
1541 0 : desc->irq = 0;
1542 : }
1543 0 : }
1544 :
1545 0 : static void msi_domain_free_locked(struct device *dev, struct msi_ctrl *ctrl)
1546 : {
1547 : struct msi_domain_info *info;
1548 : struct msi_domain_ops *ops;
1549 : struct irq_domain *domain;
1550 :
1551 0 : if (!msi_ctrl_valid(dev, ctrl))
1552 : return;
1553 :
1554 0 : domain = msi_get_device_domain(dev, ctrl->domid);
1555 0 : if (!domain)
1556 : return;
1557 :
1558 0 : info = domain->host_data;
1559 0 : ops = info->ops;
1560 :
1561 0 : if (ops->domain_free_irqs)
1562 0 : ops->domain_free_irqs(domain, dev);
1563 : else
1564 0 : __msi_domain_free_irqs(dev, domain, ctrl);
1565 :
1566 0 : if (ops->msi_post_free)
1567 0 : ops->msi_post_free(domain, dev);
1568 :
1569 0 : if (info->flags & MSI_FLAG_FREE_MSI_DESCS)
1570 0 : msi_domain_free_descs(dev, ctrl);
1571 : }
1572 :
1573 : /**
1574 : * msi_domain_free_irqs_range_locked - Free a range of interrupts from a MSI interrupt domain
1575 : * associated to @dev with msi_lock held
1576 : * @dev: Pointer to device struct of the device for which the interrupts
1577 : * are freed
1578 : * @domid: Id of the interrupt domain to operate on
1579 : * @first: First index to free (inclusive)
1580 : * @last: Last index to free (inclusive)
1581 : */
1582 0 : void msi_domain_free_irqs_range_locked(struct device *dev, unsigned int domid,
1583 : unsigned int first, unsigned int last)
1584 : {
1585 0 : struct msi_ctrl ctrl = {
1586 : .domid = domid,
1587 : .first = first,
1588 : .last = last,
1589 : };
1590 0 : msi_domain_free_locked(dev, &ctrl);
1591 0 : }
1592 :
1593 : /**
1594 : * msi_domain_free_irqs_range - Free a range of interrupts from a MSI interrupt domain
1595 : * associated to @dev
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 0 : void msi_domain_free_irqs_range(struct device *dev, unsigned int domid,
1603 : unsigned int first, unsigned int last)
1604 : {
1605 0 : msi_lock_descs(dev);
1606 0 : msi_domain_free_irqs_range_locked(dev, domid, first, last);
1607 0 : msi_unlock_descs(dev);
1608 0 : }
1609 :
1610 : /**
1611 : * msi_domain_free_irqs_all_locked - Free all interrupts from a MSI interrupt domain
1612 : * associated to a device
1613 : * @dev: Pointer to device struct of the device for which the interrupts
1614 : * are freed
1615 : * @domid: The id of the domain to operate on
1616 : *
1617 : * Must be invoked from within a msi_lock_descs() / msi_unlock_descs()
1618 : * pair. Use this for MSI irqdomains which implement their own vector
1619 : * allocation.
1620 : */
1621 0 : void msi_domain_free_irqs_all_locked(struct device *dev, unsigned int domid)
1622 : {
1623 0 : msi_domain_free_irqs_range_locked(dev, domid, 0,
1624 0 : msi_domain_get_hwsize(dev, domid) - 1);
1625 0 : }
1626 :
1627 : /**
1628 : * msi_domain_free_irqs_all - Free all interrupts from a MSI interrupt domain
1629 : * associated to a device
1630 : * @dev: Pointer to device struct of the device for which the interrupts
1631 : * are freed
1632 : * @domid: The id of the domain to operate on
1633 : */
1634 0 : void msi_domain_free_irqs_all(struct device *dev, unsigned int domid)
1635 : {
1636 0 : msi_lock_descs(dev);
1637 0 : msi_domain_free_irqs_all_locked(dev, domid);
1638 0 : msi_unlock_descs(dev);
1639 0 : }
1640 :
1641 : /**
1642 : * msi_get_domain_info - Get the MSI interrupt domain info for @domain
1643 : * @domain: The interrupt domain to retrieve data from
1644 : *
1645 : * Return: the pointer to the msi_domain_info stored in @domain->host_data.
1646 : */
1647 0 : struct msi_domain_info *msi_get_domain_info(struct irq_domain *domain)
1648 : {
1649 0 : return (struct msi_domain_info *)domain->host_data;
1650 : }
1651 :
1652 : /**
1653 : * msi_device_has_isolated_msi - True if the device has isolated MSI
1654 : * @dev: The device to check
1655 : *
1656 : * Isolated MSI means that HW modeled by an irq_domain on the path from the
1657 : * initiating device to the CPU will validate that the MSI message specifies an
1658 : * interrupt number that the device is authorized to trigger. This must block
1659 : * devices from triggering interrupts they are not authorized to trigger.
1660 : * Currently authorization means the MSI vector is one assigned to the device.
1661 : *
1662 : * This is interesting for securing VFIO use cases where a rouge MSI (eg created
1663 : * by abusing a normal PCI MemWr DMA) must not allow the VFIO userspace to
1664 : * impact outside its security domain, eg userspace triggering interrupts on
1665 : * kernel drivers, a VM triggering interrupts on the hypervisor, or a VM
1666 : * triggering interrupts on another VM.
1667 : */
1668 0 : bool msi_device_has_isolated_msi(struct device *dev)
1669 : {
1670 0 : struct irq_domain *domain = dev_get_msi_domain(dev);
1671 :
1672 0 : for (; domain; domain = domain->parent)
1673 0 : if (domain->flags & IRQ_DOMAIN_FLAG_ISOLATED_MSI)
1674 : return true;
1675 : return arch_is_isolated_msi();
1676 : }
1677 : EXPORT_SYMBOL_GPL(msi_device_has_isolated_msi);
|
