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  */
msi_alloc_desc(struct device * dev,int nvec,const struct irq_affinity_desc * affinity)58 static struct msi_desc *msi_alloc_desc(struct device *dev, int nvec,
59 				       const struct irq_affinity_desc *affinity)
60 {
61 	struct msi_desc *desc = kzalloc(sizeof(*desc), GFP_KERNEL);
62 
63 	if (!desc)
64 		return NULL;
65 
66 	desc->dev = dev;
67 	desc->nvec_used = nvec;
68 	if (affinity) {
69 		desc->affinity = kmemdup(affinity, nvec * sizeof(*desc->affinity), GFP_KERNEL);
70 		if (!desc->affinity) {
71 			kfree(desc);
72 			return NULL;
73 		}
74 	}
75 	return desc;
76 }
77 
msi_free_desc(struct msi_desc * desc)78 static void msi_free_desc(struct msi_desc *desc)
79 {
80 	kfree(desc->affinity);
81 	kfree(desc);
82 }
83 
msi_insert_desc(struct device * dev,struct msi_desc * desc,unsigned int domid,unsigned int index)84 static int msi_insert_desc(struct device *dev, struct msi_desc *desc,
85 			   unsigned int domid, unsigned int index)
86 {
87 	struct msi_device_data *md = dev->msi.data;
88 	struct xarray *xa = &md->__domains[domid].store;
89 	unsigned int hwsize;
90 	int ret;
91 
92 	hwsize = msi_domain_get_hwsize(dev, domid);
93 
94 	if (index == MSI_ANY_INDEX) {
95 		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 		ret = xa_alloc(xa, &index, desc, limit, GFP_KERNEL);
100 		if (ret)
101 			goto fail;
102 
103 		desc->msi_index = index;
104 		return 0;
105 	} else {
106 		if (index >= hwsize) {
107 			ret = -ERANGE;
108 			goto fail;
109 		}
110 
111 		desc->msi_index = index;
112 		ret = xa_insert(xa, index, desc, GFP_KERNEL);
113 		if (ret)
114 			goto fail;
115 		return 0;
116 	}
117 fail:
118 	msi_free_desc(desc);
119 	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  */
msi_domain_insert_msi_desc(struct device * dev,unsigned int domid,struct msi_desc * init_desc)132 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 	desc = msi_alloc_desc(dev, init_desc->nvec_used, init_desc->affinity);
140 	if (!desc)
141 		return -ENOMEM;
142 
143 	/* Copy type specific data to the new descriptor. */
144 	desc->pci = init_desc->pci;
145 
146 	return msi_insert_desc(dev, desc, domid, init_desc->msi_index);
147 }
148 
msi_desc_match(struct msi_desc * desc,enum msi_desc_filter filter)149 static bool msi_desc_match(struct msi_desc *desc, enum msi_desc_filter filter)
150 {
151 	switch (filter) {
152 	case MSI_DESC_ALL:
153 		return true;
154 	case MSI_DESC_NOTASSOCIATED:
155 		return !desc->irq;
156 	case MSI_DESC_ASSOCIATED:
157 		return !!desc->irq;
158 	}
159 	WARN_ON_ONCE(1);
160 	return false;
161 }
162 
msi_ctrl_valid(struct device * dev,struct msi_ctrl * ctrl)163 static bool msi_ctrl_valid(struct device *dev, struct msi_ctrl *ctrl)
164 {
165 	unsigned int hwsize;
166 
167 	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 	hwsize = msi_domain_get_hwsize(dev, ctrl->domid);
173 	if (WARN_ON_ONCE(ctrl->first > ctrl->last ||
174 			 ctrl->first >= hwsize ||
175 			 ctrl->last >= hwsize))
176 		return false;
177 	return true;
178 }
179 
msi_domain_free_descs(struct device * dev,struct msi_ctrl * ctrl)180 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 	if (!msi_ctrl_valid(dev, ctrl))
189 		return;
190 
191 	xa = &dev->msi.data->__domains[ctrl->domid].store;
192 	xa_for_each_range(xa, idx, desc, ctrl->first, ctrl->last) {
193 		xa_erase(xa, idx);
194 
195 		/* Leak the descriptor when it is still referenced */
196 		if (WARN_ON_ONCE(msi_desc_match(desc, MSI_DESC_ASSOCIATED)))
197 			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  */
msi_domain_free_msi_descs_range(struct device * dev,unsigned int domid,unsigned int first,unsigned int last)209 void msi_domain_free_msi_descs_range(struct device *dev, unsigned int domid,
210 				     unsigned int first, unsigned int last)
211 {
212 	struct msi_ctrl ctrl = {
213 		.domid	= domid,
214 		.first	= first,
215 		.last	= last,
216 	};
217 
218 	msi_domain_free_descs(dev, &ctrl);
219 }
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  */
msi_domain_add_simple_msi_descs(struct device * dev,struct msi_ctrl * ctrl)228 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 	if (!msi_ctrl_valid(dev, ctrl))
237 		return -EINVAL;
238 
239 	for (idx = ctrl->first; idx <= ctrl->last; idx++) {
240 		desc = msi_alloc_desc(dev, 1, NULL);
241 		if (!desc)
242 			goto fail_mem;
243 		ret = msi_insert_desc(dev, desc, ctrl->domid, idx);
244 		if (ret)
245 			goto fail;
246 	}
247 	return 0;
248 
249 fail_mem:
250 	ret = -ENOMEM;
251 fail:
252 	msi_domain_free_descs(dev, ctrl);
253 	return ret;
254 }
255 
__get_cached_msi_msg(struct msi_desc * entry,struct msi_msg * msg)256 void __get_cached_msi_msg(struct msi_desc *entry, struct msi_msg *msg)
257 {
258 	*msg = entry->msg;
259 }
260 
get_cached_msi_msg(unsigned int irq,struct msi_msg * msg)261 void get_cached_msi_msg(unsigned int irq, struct msi_msg *msg)
262 {
263 	struct msi_desc *entry = irq_get_msi_desc(irq);
264 
265 	__get_cached_msi_msg(entry, msg);
266 }
267 EXPORT_SYMBOL_GPL(get_cached_msi_msg);
268 
msi_device_data_release(struct device * dev,void * res)269 static void msi_device_data_release(struct device *dev, void *res)
270 {
271 	struct msi_device_data *md = res;
272 	int i;
273 
274 	for (i = 0; i < MSI_MAX_DEVICE_IRQDOMAINS; i++) {
275 		msi_remove_device_irq_domain(dev, i);
276 		WARN_ON_ONCE(!xa_empty(&md->__domains[i].store));
277 		xa_destroy(&md->__domains[i].store);
278 	}
279 	dev->msi.data = NULL;
280 }
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  */
msi_setup_device_data(struct device * dev)292 int msi_setup_device_data(struct device *dev)
293 {
294 	struct msi_device_data *md;
295 	int ret, i;
296 
297 	if (dev->msi.data)
298 		return 0;
299 
300 	md = devres_alloc(msi_device_data_release, sizeof(*md), GFP_KERNEL);
301 	if (!md)
302 		return -ENOMEM;
303 
304 	ret = msi_sysfs_create_group(dev);
305 	if (ret) {
306 		devres_free(md);
307 		return ret;
308 	}
309 
310 	for (i = 0; i < MSI_MAX_DEVICE_IRQDOMAINS; i++)
311 		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 	if (dev->msi.domain && !irq_domain_is_msi_parent(dev->msi.domain))
320 		md->__domains[MSI_DEFAULT_DOMAIN].domain = dev->msi.domain;
321 
322 	mutex_init(&md->mutex);
323 	dev->msi.data = md;
324 	devres_add(dev, md);
325 	return 0;
326 }
327 
328 /**
329  * msi_lock_descs - Lock the MSI descriptor storage of a device
330  * @dev:	Device to operate on
331  */
msi_lock_descs(struct device * dev)332 void msi_lock_descs(struct device *dev)
333 {
334 	mutex_lock(&dev->msi.data->mutex);
335 }
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  */
msi_unlock_descs(struct device * dev)342 void msi_unlock_descs(struct device *dev)
343 {
344 	/* Invalidate the index which was cached by the iterator */
345 	dev->msi.data->__iter_idx = MSI_XA_MAX_INDEX;
346 	mutex_unlock(&dev->msi.data->mutex);
347 }
348 EXPORT_SYMBOL_GPL(msi_unlock_descs);
349 
msi_find_desc(struct msi_device_data * md,unsigned int domid,enum msi_desc_filter filter)350 static struct msi_desc *msi_find_desc(struct msi_device_data *md, unsigned int domid,
351 				      enum msi_desc_filter filter)
352 {
353 	struct xarray *xa = &md->__domains[domid].store;
354 	struct msi_desc *desc;
355 
356 	xa_for_each_start(xa, md->__iter_idx, desc, md->__iter_idx) {
357 		if (msi_desc_match(desc, filter))
358 			return desc;
359 	}
360 	md->__iter_idx = MSI_XA_MAX_INDEX;
361 	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  */
msi_domain_first_desc(struct device * dev,unsigned int domid,enum msi_desc_filter filter)376 struct msi_desc *msi_domain_first_desc(struct device *dev, unsigned int domid,
377 				       enum msi_desc_filter filter)
378 {
379 	struct msi_device_data *md = dev->msi.data;
380 
381 	if (WARN_ON_ONCE(!md || domid >= MSI_MAX_DEVICE_IRQDOMAINS))
382 		return NULL;
383 
384 	lockdep_assert_held(&md->mutex);
385 
386 	md->__iter_idx = 0;
387 	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  */
msi_next_desc(struct device * dev,unsigned int domid,enum msi_desc_filter filter)405 struct msi_desc *msi_next_desc(struct device *dev, unsigned int domid,
406 			       enum msi_desc_filter filter)
407 {
408 	struct msi_device_data *md = dev->msi.data;
409 
410 	if (WARN_ON_ONCE(!md || domid >= MSI_MAX_DEVICE_IRQDOMAINS))
411 		return NULL;
412 
413 	lockdep_assert_held(&md->mutex);
414 
415 	if (md->__iter_idx >= (unsigned long)MSI_MAX_INDEX)
416 		return NULL;
417 
418 	md->__iter_idx++;
419 	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  */
msi_domain_get_virq(struct device * dev,unsigned int domid,unsigned int index)431 unsigned int msi_domain_get_virq(struct device *dev, unsigned int domid, unsigned int index)
432 {
433 	struct msi_desc *desc;
434 	unsigned int ret = 0;
435 	bool pcimsi = false;
436 	struct xarray *xa;
437 
438 	if (!dev->msi.data)
439 		return 0;
440 
441 	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 	if (dev_is_pci(dev) && domid == MSI_DEFAULT_DOMAIN)
446 		pcimsi = to_pci_dev(dev)->msi_enabled;
447 
448 	msi_lock_descs(dev);
449 	xa = &dev->msi.data->__domains[domid].store;
450 	desc = xa_load(xa, pcimsi ? 0 : index);
451 	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 		if (pcimsi) {
458 			if (index < desc->nvec_used)
459 				ret = desc->irq + index;
460 		} else {
461 			ret = desc->irq;
462 		}
463 	}
464 
465 	msi_unlock_descs(dev);
466 	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 
msi_sysfs_create_group(struct device * dev)480 static inline int msi_sysfs_create_group(struct device *dev)
481 {
482 	return devm_device_add_group(dev, &msi_irqs_group);
483 }
484 
msi_mode_show(struct device * dev,struct device_attribute * attr,char * buf)485 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 	bool is_msix = dev_is_pci(dev) ? to_pci_dev(dev)->msix_enabled : false;
490 
491 	return sysfs_emit(buf, "%s\n", is_msix ? "msix" : "msi");
492 }
493 
msi_sysfs_remove_desc(struct device * dev,struct msi_desc * desc)494 static void msi_sysfs_remove_desc(struct device *dev, struct msi_desc *desc)
495 {
496 	struct device_attribute *attrs = desc->sysfs_attrs;
497 	int i;
498 
499 	if (!attrs)
500 		return;
501 
502 	desc->sysfs_attrs = NULL;
503 	for (i = 0; i < desc->nvec_used; i++) {
504 		if (attrs[i].show)
505 			sysfs_remove_file_from_group(&dev->kobj, &attrs[i].attr, msi_irqs_group.name);
506 		kfree(attrs[i].attr.name);
507 	}
508 	kfree(attrs);
509 }
510 
msi_sysfs_populate_desc(struct device * dev,struct msi_desc * desc)511 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 	attrs = kcalloc(desc->nvec_used, sizeof(*attrs), GFP_KERNEL);
517 	if (!attrs)
518 		return -ENOMEM;
519 
520 	desc->sysfs_attrs = attrs;
521 	for (i = 0; i < desc->nvec_used; i++) {
522 		sysfs_attr_init(&attrs[i].attr);
523 		attrs[i].attr.name = kasprintf(GFP_KERNEL, "%d", desc->irq + i);
524 		if (!attrs[i].attr.name) {
525 			ret = -ENOMEM;
526 			goto fail;
527 		}
528 
529 		attrs[i].attr.mode = 0444;
530 		attrs[i].show = msi_mode_show;
531 
532 		ret = sysfs_add_file_to_group(&dev->kobj, &attrs[i].attr, msi_irqs_group.name);
533 		if (ret) {
534 			attrs[i].show = NULL;
535 			goto fail;
536 		}
537 	}
538 	return 0;
539 
540 fail:
541 	msi_sysfs_remove_desc(dev, desc);
542 	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  */
msi_device_populate_sysfs(struct device * dev)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  */
msi_device_destroy_sysfs(struct device * dev)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 */
msi_sysfs_create_group(struct device * dev)579 static inline int msi_sysfs_create_group(struct device *dev) { return 0; }
msi_sysfs_populate_desc(struct device * dev,struct msi_desc * desc)580 static inline int msi_sysfs_populate_desc(struct device *dev, struct msi_desc *desc) { return 0; }
msi_sysfs_remove_desc(struct device * dev,struct msi_desc * desc)581 static inline void msi_sysfs_remove_desc(struct device *dev, struct msi_desc *desc) { }
582 #endif /* !CONFIG_SYSFS */
583 
msi_get_device_domain(struct device * dev,unsigned int domid)584 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 	if (WARN_ON_ONCE(domid >= MSI_MAX_DEVICE_IRQDOMAINS))
591 		return NULL;
592 
593 	domain = dev->msi.data->__domains[domid].domain;
594 	if (!domain)
595 		return NULL;
596 
597 	if (WARN_ON_ONCE(irq_domain_is_msi_parent(domain)))
598 		return NULL;
599 
600 	return domain;
601 }
602 
msi_domain_get_hwsize(struct device * dev,unsigned int domid)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 	domain = msi_get_device_domain(dev, domid);
609 	if (domain) {
610 		info = domain->host_data;
611 		return info->hwsize;
612 	}
613 	/* No domain, default to MSI_XA_DOMAIN_SIZE */
614 	return MSI_XA_DOMAIN_SIZE;
615 }
616 
irq_chip_write_msi_msg(struct irq_data * data,struct msi_msg * msg)617 static inline void irq_chip_write_msi_msg(struct irq_data *data,
618 					  struct msi_msg *msg)
619 {
620 	data->chip->irq_write_msi_msg(data, msg);
621 }
622 
msi_check_level(struct irq_domain * domain,struct msi_msg * msg)623 static void msi_check_level(struct irq_domain *domain, struct msi_msg *msg)
624 {
625 	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 	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 }
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  */
msi_domain_set_affinity(struct irq_data * irq_data,const struct cpumask * mask,bool force)647 int msi_domain_set_affinity(struct irq_data *irq_data,
648 			    const struct cpumask *mask, bool force)
649 {
650 	struct irq_data *parent = irq_data->parent_data;
651 	struct msi_msg msg[2] = { [1] = { }, };
652 	int ret;
653 
654 	ret = parent->chip->irq_set_affinity(parent, mask, force);
655 	if (ret >= 0 && ret != IRQ_SET_MASK_OK_DONE) {
656 		BUG_ON(irq_chip_compose_msi_msg(irq_data, msg));
657 		msi_check_level(irq_data->domain, msg);
658 		irq_chip_write_msi_msg(irq_data, msg);
659 	}
660 
661 	return ret;
662 }
663 
msi_domain_activate(struct irq_domain * domain,struct irq_data * irq_data,bool early)664 static int msi_domain_activate(struct irq_domain *domain,
665 			       struct irq_data *irq_data, bool early)
666 {
667 	struct msi_msg msg[2] = { [1] = { }, };
668 
669 	BUG_ON(irq_chip_compose_msi_msg(irq_data, msg));
670 	msi_check_level(irq_data->domain, msg);
671 	irq_chip_write_msi_msg(irq_data, msg);
672 	return 0;
673 }
674 
msi_domain_deactivate(struct irq_domain * domain,struct irq_data * irq_data)675 static void msi_domain_deactivate(struct irq_domain *domain,
676 				  struct irq_data *irq_data)
677 {
678 	struct msi_msg msg[2];
679 
680 	memset(msg, 0, sizeof(msg));
681 	irq_chip_write_msi_msg(irq_data, msg);
682 }
683 
msi_domain_alloc(struct irq_domain * domain,unsigned int virq,unsigned int nr_irqs,void * arg)684 static int msi_domain_alloc(struct irq_domain *domain, unsigned int virq,
685 			    unsigned int nr_irqs, void *arg)
686 {
687 	struct msi_domain_info *info = domain->host_data;
688 	struct msi_domain_ops *ops = info->ops;
689 	irq_hw_number_t hwirq = ops->get_hwirq(info, arg);
690 	int i, ret;
691 
692 	if (irq_find_mapping(domain, hwirq) > 0)
693 		return -EEXIST;
694 
695 	if (domain->parent) {
696 		ret = irq_domain_alloc_irqs_parent(domain, virq, nr_irqs, arg);
697 		if (ret < 0)
698 			return ret;
699 	}
700 
701 	for (i = 0; i < nr_irqs; i++) {
702 		ret = ops->msi_init(domain, info, virq + i, hwirq + i, arg);
703 		if (ret < 0) {
704 			if (ops->msi_free) {
705 				for (i--; i > 0; i--)
706 					ops->msi_free(domain, info, virq + i);
707 			}
708 			irq_domain_free_irqs_top(domain, virq, nr_irqs);
709 			return ret;
710 		}
711 	}
712 
713 	return 0;
714 }
715 
msi_domain_free(struct irq_domain * domain,unsigned int virq,unsigned int nr_irqs)716 static void msi_domain_free(struct irq_domain *domain, unsigned int virq,
717 			    unsigned int nr_irqs)
718 {
719 	struct msi_domain_info *info = domain->host_data;
720 	int i;
721 
722 	if (info->ops->msi_free) {
723 		for (i = 0; i < nr_irqs; i++)
724 			info->ops->msi_free(domain, info, virq + i);
725 	}
726 	irq_domain_free_irqs_top(domain, virq, nr_irqs);
727 }
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 
msi_domain_ops_get_hwirq(struct msi_domain_info * info,msi_alloc_info_t * arg)736 static irq_hw_number_t msi_domain_ops_get_hwirq(struct msi_domain_info *info,
737 						msi_alloc_info_t *arg)
738 {
739 	return arg->hwirq;
740 }
741 
msi_domain_ops_prepare(struct irq_domain * domain,struct device * dev,int nvec,msi_alloc_info_t * arg)742 static int msi_domain_ops_prepare(struct irq_domain *domain, struct device *dev,
743 				  int nvec, msi_alloc_info_t *arg)
744 {
745 	memset(arg, 0, sizeof(*arg));
746 	return 0;
747 }
748 
msi_domain_ops_set_desc(msi_alloc_info_t * arg,struct msi_desc * desc)749 static void msi_domain_ops_set_desc(msi_alloc_info_t *arg,
750 				    struct msi_desc *desc)
751 {
752 	arg->desc = desc;
753 }
754 
msi_domain_ops_init(struct irq_domain * domain,struct msi_domain_info * info,unsigned int virq,irq_hw_number_t hwirq,msi_alloc_info_t * arg)755 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 	irq_domain_set_hwirq_and_chip(domain, virq, hwirq, info->chip,
761 				      info->chip_data);
762 	if (info->handler && info->handler_name) {
763 		__irq_set_handler(virq, info->handler, 0, info->handler_name);
764 		if (info->handler_data)
765 			irq_set_handler_data(virq, info->handler_data);
766 	}
767 	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 
msi_domain_update_dom_ops(struct msi_domain_info * info)777 static void msi_domain_update_dom_ops(struct msi_domain_info *info)
778 {
779 	struct msi_domain_ops *ops = info->ops;
780 
781 	if (ops == NULL) {
782 		info->ops = &msi_domain_ops_default;
783 		return;
784 	}
785 
786 	if (!(info->flags & MSI_FLAG_USE_DEF_DOM_OPS))
787 		return;
788 
789 	if (ops->get_hwirq == NULL)
790 		ops->get_hwirq = msi_domain_ops_default.get_hwirq;
791 	if (ops->msi_init == NULL)
792 		ops->msi_init = msi_domain_ops_default.msi_init;
793 	if (ops->msi_prepare == NULL)
794 		ops->msi_prepare = msi_domain_ops_default.msi_prepare;
795 	if (ops->set_desc == NULL)
796 		ops->set_desc = msi_domain_ops_default.set_desc;
797 }
798 
msi_domain_update_chip_ops(struct msi_domain_info * info)799 static void msi_domain_update_chip_ops(struct msi_domain_info *info)
800 {
801 	struct irq_chip *chip = info->chip;
802 
803 	BUG_ON(!chip || !chip->irq_mask || !chip->irq_unmask);
804 	if (!chip->irq_set_affinity)
805 		chip->irq_set_affinity = msi_domain_set_affinity;
806 }
807 
__msi_create_irq_domain(struct fwnode_handle * fwnode,struct msi_domain_info * info,unsigned int flags,struct irq_domain * parent)808 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 	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 	if (!info->hwsize)
824 		info->hwsize = MSI_XA_DOMAIN_SIZE;
825 
826 	msi_domain_update_dom_ops(info);
827 	if (info->flags & MSI_FLAG_USE_DEF_CHIP_OPS)
828 		msi_domain_update_chip_ops(info);
829 
830 	domain = irq_domain_create_hierarchy(parent, flags | IRQ_DOMAIN_FLAG_MSI, 0,
831 					     fwnode, &msi_domain_ops, info);
832 
833 	if (domain)
834 		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  */
msi_create_irq_domain(struct fwnode_handle * fwnode,struct msi_domain_info * info,struct irq_domain * parent)847 struct irq_domain *msi_create_irq_domain(struct fwnode_handle *fwnode,
848 					 struct msi_domain_info *info,
849 					 struct irq_domain *parent)
850 {
851 	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  */
msi_parent_init_dev_msi_info(struct device * dev,struct irq_domain * domain,struct irq_domain * msi_parent_domain,struct msi_domain_info * msi_child_info)881 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 	struct irq_domain *parent = domain->parent;
886 
887 	if (WARN_ON_ONCE(!parent || !parent->msi_parent_ops ||
888 			 !parent->msi_parent_ops->init_dev_msi_info))
889 		return false;
890 
891 	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  */
msi_create_device_irq_domain(struct device * dev,unsigned int domid,const struct msi_domain_template * template,unsigned int hwsize,void * domain_data,void * chip_data)942 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 	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 	if (!irq_domain_is_msi_parent(parent))
953 		return false;
954 
955 	if (domid >= MSI_MAX_DEVICE_IRQDOMAINS)
956 		return false;
957 
958 	bundle = kmemdup(template, sizeof(*bundle), GFP_KERNEL);
959 	if (!bundle)
960 		return false;
961 
962 	bundle->info.hwsize = hwsize;
963 	bundle->info.chip = &bundle->chip;
964 	bundle->info.ops = &bundle->ops;
965 	bundle->info.data = domain_data;
966 	bundle->info.chip_data = chip_data;
967 
968 	pops = parent->msi_parent_ops;
969 	snprintf(bundle->name, sizeof(bundle->name), "%s%s-%s",
970 		 pops->prefix ? : "", bundle->chip.name, dev_name(dev));
971 	bundle->chip.name = bundle->name;
972 
973 	fwnode = irq_domain_alloc_named_fwnode(bundle->name);
974 	if (!fwnode)
975 		goto free_bundle;
976 
977 	if (msi_setup_device_data(dev))
978 		goto free_fwnode;
979 
980 	msi_lock_descs(dev);
981 
982 	if (WARN_ON_ONCE(msi_get_device_domain(dev, domid)))
983 		goto fail;
984 
985 	if (!pops->init_dev_msi_info(dev, parent, parent, &bundle->info))
986 		goto fail;
987 
988 	domain = __msi_create_irq_domain(fwnode, &bundle->info, IRQ_DOMAIN_FLAG_MSI_DEVICE, parent);
989 	if (!domain)
990 		goto fail;
991 
992 	domain->dev = dev;
993 	dev->msi.data->__domains[domid].domain = domain;
994 	msi_unlock_descs(dev);
995 	return true;
996 
997 fail:
998 	msi_unlock_descs(dev);
999 free_fwnode:
1000 	irq_domain_free_fwnode(fwnode);
1001 free_bundle:
1002 	kfree(bundle);
1003 	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  */
msi_remove_device_irq_domain(struct device * dev,unsigned int domid)1011 void msi_remove_device_irq_domain(struct device *dev, unsigned int domid)
1012 {
1013 	struct fwnode_handle *fwnode = NULL;
1014 	struct msi_domain_info *info;
1015 	struct irq_domain *domain;
1016 
1017 	msi_lock_descs(dev);
1018 
1019 	domain = msi_get_device_domain(dev, domid);
1020 
1021 	if (!domain || !irq_domain_is_msi_device(domain))
1022 		goto unlock;
1023 
1024 	dev->msi.data->__domains[domid].domain = NULL;
1025 	info = domain->host_data;
1026 	if (irq_domain_is_msi_device(domain))
1027 		fwnode = domain->fwnode;
1028 	irq_domain_remove(domain);
1029 	irq_domain_free_fwnode(fwnode);
1030 	kfree(container_of(info, struct msi_domain_template, info));
1031 
1032 unlock:
1033 	msi_unlock_descs(dev);
1034 }
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  */
msi_match_device_irq_domain(struct device * dev,unsigned int domid,enum irq_domain_bus_token bus_token)1044 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 	bool ret = false;
1050 
1051 	msi_lock_descs(dev);
1052 	domain = msi_get_device_domain(dev, domid);
1053 	if (domain && irq_domain_is_msi_device(domain)) {
1054 		info = domain->host_data;
1055 		ret = info->bus_token == bus_token;
1056 	}
1057 	msi_unlock_descs(dev);
1058 	return ret;
1059 }
1060 
msi_domain_prepare_irqs(struct irq_domain * domain,struct device * dev,int nvec,msi_alloc_info_t * arg)1061 int msi_domain_prepare_irqs(struct irq_domain *domain, struct device *dev,
1062 			    int nvec, msi_alloc_info_t *arg)
1063 {
1064 	struct msi_domain_info *info = domain->host_data;
1065 	struct msi_domain_ops *ops = info->ops;
1066 
1067 	return ops->msi_prepare(domain, dev, nvec, arg);
1068 }
1069 
msi_domain_populate_irqs(struct irq_domain * domain,struct device * dev,int virq_base,int nvec,msi_alloc_info_t * arg)1070 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 	struct msi_domain_info *info = domain->host_data;
1074 	struct msi_domain_ops *ops = info->ops;
1075 	struct msi_ctrl ctrl = {
1076 		.domid	= MSI_DEFAULT_DOMAIN,
1077 		.first  = virq_base,
1078 		.last	= virq_base + nvec - 1,
1079 	};
1080 	struct msi_desc *desc;
1081 	struct xarray *xa;
1082 	int ret, virq;
1083 
1084 	msi_lock_descs(dev);
1085 
1086 	if (!msi_ctrl_valid(dev, &ctrl)) {
1087 		ret = -EINVAL;
1088 		goto unlock;
1089 	}
1090 
1091 	ret = msi_domain_add_simple_msi_descs(dev, &ctrl);
1092 	if (ret)
1093 		goto unlock;
1094 
1095 	xa = &dev->msi.data->__domains[ctrl.domid].store;
1096 
1097 	for (virq = virq_base; virq < virq_base + nvec; virq++) {
1098 		desc = xa_load(xa, virq);
1099 		desc->irq = virq;
1100 
1101 		ops->set_desc(arg, desc);
1102 		ret = irq_domain_alloc_irqs_hierarchy(domain, virq, 1, arg);
1103 		if (ret)
1104 			goto fail;
1105 
1106 		irq_set_msi_desc(virq, desc);
1107 	}
1108 	msi_unlock_descs(dev);
1109 	return 0;
1110 
1111 fail:
1112 	for (--virq; virq >= virq_base; virq--) {
1113 		msi_domain_depopulate_descs(dev, virq, 1);
1114 		irq_domain_free_irqs_common(domain, virq, 1);
1115 	}
1116 	msi_domain_free_descs(dev, &ctrl);
1117 unlock:
1118 	msi_unlock_descs(dev);
1119 	return ret;
1120 }
1121 
msi_domain_depopulate_descs(struct device * dev,int virq_base,int nvec)1122 void msi_domain_depopulate_descs(struct device *dev, int virq_base, int nvec)
1123 {
1124 	struct msi_ctrl ctrl = {
1125 		.domid	= MSI_DEFAULT_DOMAIN,
1126 		.first  = virq_base,
1127 		.last	= virq_base + nvec - 1,
1128 	};
1129 	struct msi_desc *desc;
1130 	struct xarray *xa;
1131 	unsigned long idx;
1132 
1133 	if (!msi_ctrl_valid(dev, &ctrl))
1134 		return;
1135 
1136 	xa = &dev->msi.data->__domains[ctrl.domid].store;
1137 	xa_for_each_range(xa, idx, desc, ctrl.first, ctrl.last)
1138 		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  */
msi_check_reservation_mode(struct irq_domain * domain,struct msi_domain_info * info,struct device * dev)1152 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 	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 	if (!(info->flags & MSI_FLAG_MUST_REACTIVATE))
1169 		return false;
1170 
1171 	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 	desc = msi_first_desc(dev, MSI_DESC_ALL);
1179 	return desc->pci.msi_attrib.is_msix || desc->pci.msi_attrib.can_mask;
1180 }
1181 
msi_handle_pci_fail(struct irq_domain * domain,struct msi_desc * desc,int allocated)1182 static int msi_handle_pci_fail(struct irq_domain *domain, struct msi_desc *desc,
1183 			       int allocated)
1184 {
1185 	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 	if (desc->nvec_used > 1)
1199 		return 1;
1200 
1201 	/* If there was a successful allocation let the caller know */
1202 	return allocated ? allocated : -ENOSPC;
1203 }
1204 
1205 #define VIRQ_CAN_RESERVE	0x01
1206 #define VIRQ_ACTIVATE		0x02
1207 
msi_init_virq(struct irq_domain * domain,int virq,unsigned int vflags)1208 static int msi_init_virq(struct irq_domain *domain, int virq, unsigned int vflags)
1209 {
1210 	struct irq_data *irqd = irq_domain_get_irq_data(domain, virq);
1211 	int ret;
1212 
1213 	if (!(vflags & VIRQ_CAN_RESERVE)) {
1214 		irqd_clr_can_reserve(irqd);
1215 
1216 		/*
1217 		 * If the interrupt is managed but no CPU is available to
1218 		 * service it, shut it down until better times. Note that
1219 		 * we only do this on the !RESERVE path as x86 (the only
1220 		 * architecture using this flag) deals with this in a
1221 		 * different way by using a catch-all vector.
1222 		 */
1223 		if ((vflags & VIRQ_ACTIVATE) &&
1224 		    irqd_affinity_is_managed(irqd) &&
1225 		    !cpumask_intersects(irq_data_get_affinity_mask(irqd),
1226 					cpu_online_mask)) {
1227 			    irqd_set_managed_shutdown(irqd);
1228 			    return 0;
1229 		    }
1230 	}
1231 
1232 	if (!(vflags & VIRQ_ACTIVATE))
1233 		return 0;
1234 
1235 	ret = irq_domain_activate_irq(irqd, vflags & VIRQ_CAN_RESERVE);
1236 	if (ret)
1237 		return ret;
1238 	/*
1239 	 * If the interrupt uses reservation mode, clear the activated bit
1240 	 * so request_irq() will assign the final vector.
1241 	 */
1242 	if (vflags & VIRQ_CAN_RESERVE)
1243 		irqd_clr_activated(irqd);
1244 	return 0;
1245 }
1246 
__msi_domain_alloc_irqs(struct device * dev,struct irq_domain * domain,struct msi_ctrl * ctrl)1247 static int __msi_domain_alloc_irqs(struct device *dev, struct irq_domain *domain,
1248 				   struct msi_ctrl *ctrl)
1249 {
1250 	struct xarray *xa = &dev->msi.data->__domains[ctrl->domid].store;
1251 	struct msi_domain_info *info = domain->host_data;
1252 	struct msi_domain_ops *ops = info->ops;
1253 	unsigned int vflags = 0, allocated = 0;
1254 	msi_alloc_info_t arg = { };
1255 	struct msi_desc *desc;
1256 	unsigned long idx;
1257 	int i, ret, virq;
1258 
1259 	ret = msi_domain_prepare_irqs(domain, dev, ctrl->nirqs, &arg);
1260 	if (ret)
1261 		return ret;
1262 
1263 	/*
1264 	 * This flag is set by the PCI layer as we need to activate
1265 	 * the MSI entries before the PCI layer enables MSI in the
1266 	 * card. Otherwise the card latches a random msi message.
1267 	 */
1268 	if (info->flags & MSI_FLAG_ACTIVATE_EARLY)
1269 		vflags |= VIRQ_ACTIVATE;
1270 
1271 	/*
1272 	 * Interrupt can use a reserved vector and will not occupy
1273 	 * a real device vector until the interrupt is requested.
1274 	 */
1275 	if (msi_check_reservation_mode(domain, info, dev))
1276 		vflags |= VIRQ_CAN_RESERVE;
1277 
1278 	xa_for_each_range(xa, idx, desc, ctrl->first, ctrl->last) {
1279 		if (!msi_desc_match(desc, MSI_DESC_NOTASSOCIATED))
1280 			continue;
1281 
1282 		/* This should return -ECONFUSED... */
1283 		if (WARN_ON_ONCE(allocated >= ctrl->nirqs))
1284 			return -EINVAL;
1285 
1286 		if (ops->prepare_desc)
1287 			ops->prepare_desc(domain, &arg, desc);
1288 
1289 		ops->set_desc(&arg, desc);
1290 
1291 		virq = __irq_domain_alloc_irqs(domain, -1, desc->nvec_used,
1292 					       dev_to_node(dev), &arg, false,
1293 					       desc->affinity);
1294 		if (virq < 0)
1295 			return msi_handle_pci_fail(domain, desc, allocated);
1296 
1297 		for (i = 0; i < desc->nvec_used; i++) {
1298 			irq_set_msi_desc_off(virq, i, desc);
1299 			irq_debugfs_copy_devname(virq + i, dev);
1300 			ret = msi_init_virq(domain, virq + i, vflags);
1301 			if (ret)
1302 				return ret;
1303 		}
1304 		if (info->flags & MSI_FLAG_DEV_SYSFS) {
1305 			ret = msi_sysfs_populate_desc(dev, desc);
1306 			if (ret)
1307 				return ret;
1308 		}
1309 		allocated++;
1310 	}
1311 	return 0;
1312 }
1313 
msi_domain_alloc_simple_msi_descs(struct device * dev,struct msi_domain_info * info,struct msi_ctrl * ctrl)1314 static int msi_domain_alloc_simple_msi_descs(struct device *dev,
1315 					     struct msi_domain_info *info,
1316 					     struct msi_ctrl *ctrl)
1317 {
1318 	if (!(info->flags & MSI_FLAG_ALLOC_SIMPLE_MSI_DESCS))
1319 		return 0;
1320 
1321 	return msi_domain_add_simple_msi_descs(dev, ctrl);
1322 }
1323 
__msi_domain_alloc_locked(struct device * dev,struct msi_ctrl * ctrl)1324 static int __msi_domain_alloc_locked(struct device *dev, struct msi_ctrl *ctrl)
1325 {
1326 	struct msi_domain_info *info;
1327 	struct msi_domain_ops *ops;
1328 	struct irq_domain *domain;
1329 	int ret;
1330 
1331 	if (!msi_ctrl_valid(dev, ctrl))
1332 		return -EINVAL;
1333 
1334 	domain = msi_get_device_domain(dev, ctrl->domid);
1335 	if (!domain)
1336 		return -ENODEV;
1337 
1338 	info = domain->host_data;
1339 
1340 	ret = msi_domain_alloc_simple_msi_descs(dev, info, ctrl);
1341 	if (ret)
1342 		return ret;
1343 
1344 	ops = info->ops;
1345 	if (ops->domain_alloc_irqs)
1346 		return ops->domain_alloc_irqs(domain, dev, ctrl->nirqs);
1347 
1348 	return __msi_domain_alloc_irqs(dev, domain, ctrl);
1349 }
1350 
msi_domain_alloc_locked(struct device * dev,struct msi_ctrl * ctrl)1351 static int msi_domain_alloc_locked(struct device *dev, struct msi_ctrl *ctrl)
1352 {
1353 	int ret = __msi_domain_alloc_locked(dev, ctrl);
1354 
1355 	if (ret)
1356 		msi_domain_free_locked(dev, ctrl);
1357 	return ret;
1358 }
1359 
1360 /**
1361  * msi_domain_alloc_irqs_range_locked - Allocate interrupts from a MSI interrupt domain
1362  * @dev:	Pointer to device struct of the device for which the interrupts
1363  *		are allocated
1364  * @domid:	Id of the interrupt domain to operate on
1365  * @first:	First index to allocate (inclusive)
1366  * @last:	Last index to allocate (inclusive)
1367  *
1368  * Must be invoked from within a msi_lock_descs() / msi_unlock_descs()
1369  * pair. Use this for MSI irqdomains which implement their own descriptor
1370  * allocation/free.
1371  *
1372  * Return: %0 on success or an error code.
1373  */
msi_domain_alloc_irqs_range_locked(struct device * dev,unsigned int domid,unsigned int first,unsigned int last)1374 int msi_domain_alloc_irqs_range_locked(struct device *dev, unsigned int domid,
1375 				       unsigned int first, unsigned int last)
1376 {
1377 	struct msi_ctrl ctrl = {
1378 		.domid	= domid,
1379 		.first	= first,
1380 		.last	= last,
1381 		.nirqs	= last + 1 - first,
1382 	};
1383 
1384 	return msi_domain_alloc_locked(dev, &ctrl);
1385 }
1386 
1387 /**
1388  * msi_domain_alloc_irqs_range - Allocate interrupts from a MSI interrupt domain
1389  * @dev:	Pointer to device struct of the device for which the interrupts
1390  *		are allocated
1391  * @domid:	Id of the interrupt domain to operate on
1392  * @first:	First index to allocate (inclusive)
1393  * @last:	Last index to allocate (inclusive)
1394  *
1395  * Return: %0 on success or an error code.
1396  */
msi_domain_alloc_irqs_range(struct device * dev,unsigned int domid,unsigned int first,unsigned int last)1397 int msi_domain_alloc_irqs_range(struct device *dev, unsigned int domid,
1398 				unsigned int first, unsigned int last)
1399 {
1400 	int ret;
1401 
1402 	msi_lock_descs(dev);
1403 	ret = msi_domain_alloc_irqs_range_locked(dev, domid, first, last);
1404 	msi_unlock_descs(dev);
1405 	return ret;
1406 }
1407 
1408 /**
1409  * msi_domain_alloc_irqs_all_locked - Allocate all interrupts from a MSI interrupt domain
1410  *
1411  * @dev:	Pointer to device struct of the device for which the interrupts
1412  *		are allocated
1413  * @domid:	Id of the interrupt domain to operate on
1414  * @nirqs:	The number of interrupts to allocate
1415  *
1416  * This function scans all MSI descriptors of the MSI domain and allocates interrupts
1417  * for all unassigned ones. That function is to be used for MSI domain usage where
1418  * the descriptor allocation is handled at the call site, e.g. PCI/MSI[X].
1419  *
1420  * Return: %0 on success or an error code.
1421  */
msi_domain_alloc_irqs_all_locked(struct device * dev,unsigned int domid,int nirqs)1422 int msi_domain_alloc_irqs_all_locked(struct device *dev, unsigned int domid, int nirqs)
1423 {
1424 	struct msi_ctrl ctrl = {
1425 		.domid	= domid,
1426 		.first	= 0,
1427 		.last	= msi_domain_get_hwsize(dev, domid) - 1,
1428 		.nirqs	= nirqs,
1429 	};
1430 
1431 	return msi_domain_alloc_locked(dev, &ctrl);
1432 }
1433 
1434 /**
1435  * msi_domain_alloc_irq_at - Allocate an interrupt from a MSI interrupt domain at
1436  *			     a given index - or at the next free index
1437  *
1438  * @dev:	Pointer to device struct of the device for which the interrupts
1439  *		are allocated
1440  * @domid:	Id of the interrupt domain to operate on
1441  * @index:	Index for allocation. If @index == %MSI_ANY_INDEX the allocation
1442  *		uses the next free index.
1443  * @affdesc:	Optional pointer to an interrupt affinity descriptor structure
1444  * @icookie:	Optional pointer to a domain specific per instance cookie. If
1445  *		non-NULL the content of the cookie is stored in msi_desc::data.
1446  *		Must be NULL for MSI-X allocations
1447  *
1448  * This requires a MSI interrupt domain which lets the core code manage the
1449  * MSI descriptors.
1450  *
1451  * Return: struct msi_map
1452  *
1453  *	On success msi_map::index contains the allocated index number and
1454  *	msi_map::virq the corresponding Linux interrupt number
1455  *
1456  *	On failure msi_map::index contains the error code and msi_map::virq
1457  *	is %0.
1458  */
msi_domain_alloc_irq_at(struct device * dev,unsigned int domid,unsigned int index,const struct irq_affinity_desc * affdesc,union msi_instance_cookie * icookie)1459 struct msi_map msi_domain_alloc_irq_at(struct device *dev, unsigned int domid, unsigned int index,
1460 				       const struct irq_affinity_desc *affdesc,
1461 				       union msi_instance_cookie *icookie)
1462 {
1463 	struct msi_ctrl ctrl = { .domid	= domid, .nirqs = 1, };
1464 	struct irq_domain *domain;
1465 	struct msi_map map = { };
1466 	struct msi_desc *desc;
1467 	int ret;
1468 
1469 	msi_lock_descs(dev);
1470 	domain = msi_get_device_domain(dev, domid);
1471 	if (!domain) {
1472 		map.index = -ENODEV;
1473 		goto unlock;
1474 	}
1475 
1476 	desc = msi_alloc_desc(dev, 1, affdesc);
1477 	if (!desc) {
1478 		map.index = -ENOMEM;
1479 		goto unlock;
1480 	}
1481 
1482 	if (icookie)
1483 		desc->data.icookie = *icookie;
1484 
1485 	ret = msi_insert_desc(dev, desc, domid, index);
1486 	if (ret) {
1487 		map.index = ret;
1488 		goto unlock;
1489 	}
1490 
1491 	ctrl.first = ctrl.last = desc->msi_index;
1492 
1493 	ret = __msi_domain_alloc_irqs(dev, domain, &ctrl);
1494 	if (ret) {
1495 		map.index = ret;
1496 		msi_domain_free_locked(dev, &ctrl);
1497 	} else {
1498 		map.index = desc->msi_index;
1499 		map.virq = desc->irq;
1500 	}
1501 unlock:
1502 	msi_unlock_descs(dev);
1503 	return map;
1504 }
1505 
__msi_domain_free_irqs(struct device * dev,struct irq_domain * domain,struct msi_ctrl * ctrl)1506 static void __msi_domain_free_irqs(struct device *dev, struct irq_domain *domain,
1507 				   struct msi_ctrl *ctrl)
1508 {
1509 	struct xarray *xa = &dev->msi.data->__domains[ctrl->domid].store;
1510 	struct msi_domain_info *info = domain->host_data;
1511 	struct irq_data *irqd;
1512 	struct msi_desc *desc;
1513 	unsigned long idx;
1514 	int i;
1515 
1516 	xa_for_each_range(xa, idx, desc, ctrl->first, ctrl->last) {
1517 		/* Only handle MSI entries which have an interrupt associated */
1518 		if (!msi_desc_match(desc, MSI_DESC_ASSOCIATED))
1519 			continue;
1520 
1521 		/* Make sure all interrupts are deactivated */
1522 		for (i = 0; i < desc->nvec_used; i++) {
1523 			irqd = irq_domain_get_irq_data(domain, desc->irq + i);
1524 			if (irqd && irqd_is_activated(irqd))
1525 				irq_domain_deactivate_irq(irqd);
1526 		}
1527 
1528 		irq_domain_free_irqs(desc->irq, desc->nvec_used);
1529 		if (info->flags & MSI_FLAG_DEV_SYSFS)
1530 			msi_sysfs_remove_desc(dev, desc);
1531 		desc->irq = 0;
1532 	}
1533 }
1534 
msi_domain_free_locked(struct device * dev,struct msi_ctrl * ctrl)1535 static void msi_domain_free_locked(struct device *dev, struct msi_ctrl *ctrl)
1536 {
1537 	struct msi_domain_info *info;
1538 	struct msi_domain_ops *ops;
1539 	struct irq_domain *domain;
1540 
1541 	if (!msi_ctrl_valid(dev, ctrl))
1542 		return;
1543 
1544 	domain = msi_get_device_domain(dev, ctrl->domid);
1545 	if (!domain)
1546 		return;
1547 
1548 	info = domain->host_data;
1549 	ops = info->ops;
1550 
1551 	if (ops->domain_free_irqs)
1552 		ops->domain_free_irqs(domain, dev);
1553 	else
1554 		__msi_domain_free_irqs(dev, domain, ctrl);
1555 
1556 	if (ops->msi_post_free)
1557 		ops->msi_post_free(domain, dev);
1558 
1559 	if (info->flags & MSI_FLAG_FREE_MSI_DESCS)
1560 		msi_domain_free_descs(dev, ctrl);
1561 }
1562 
1563 /**
1564  * msi_domain_free_irqs_range_locked - Free a range of interrupts from a MSI interrupt domain
1565  *				       associated to @dev with msi_lock held
1566  * @dev:	Pointer to device struct of the device for which the interrupts
1567  *		are freed
1568  * @domid:	Id of the interrupt domain to operate on
1569  * @first:	First index to free (inclusive)
1570  * @last:	Last index to free (inclusive)
1571  */
msi_domain_free_irqs_range_locked(struct device * dev,unsigned int domid,unsigned int first,unsigned int last)1572 void msi_domain_free_irqs_range_locked(struct device *dev, unsigned int domid,
1573 				       unsigned int first, unsigned int last)
1574 {
1575 	struct msi_ctrl ctrl = {
1576 		.domid	= domid,
1577 		.first	= first,
1578 		.last	= last,
1579 	};
1580 	msi_domain_free_locked(dev, &ctrl);
1581 }
1582 
1583 /**
1584  * msi_domain_free_irqs_range - Free a range of interrupts from a MSI interrupt domain
1585  *				associated to @dev
1586  * @dev:	Pointer to device struct of the device for which the interrupts
1587  *		are freed
1588  * @domid:	Id of the interrupt domain to operate on
1589  * @first:	First index to free (inclusive)
1590  * @last:	Last index to free (inclusive)
1591  */
msi_domain_free_irqs_range(struct device * dev,unsigned int domid,unsigned int first,unsigned int last)1592 void msi_domain_free_irqs_range(struct device *dev, unsigned int domid,
1593 				unsigned int first, unsigned int last)
1594 {
1595 	msi_lock_descs(dev);
1596 	msi_domain_free_irqs_range_locked(dev, domid, first, last);
1597 	msi_unlock_descs(dev);
1598 }
1599 
1600 /**
1601  * msi_domain_free_irqs_all_locked - Free all interrupts from a MSI interrupt domain
1602  *				     associated to a device
1603  * @dev:	Pointer to device struct of the device for which the interrupts
1604  *		are freed
1605  * @domid:	The id of the domain to operate on
1606  *
1607  * Must be invoked from within a msi_lock_descs() / msi_unlock_descs()
1608  * pair. Use this for MSI irqdomains which implement their own vector
1609  * allocation.
1610  */
msi_domain_free_irqs_all_locked(struct device * dev,unsigned int domid)1611 void msi_domain_free_irqs_all_locked(struct device *dev, unsigned int domid)
1612 {
1613 	msi_domain_free_irqs_range_locked(dev, domid, 0,
1614 					  msi_domain_get_hwsize(dev, domid) - 1);
1615 }
1616 
1617 /**
1618  * msi_domain_free_irqs_all - Free all interrupts from a MSI interrupt domain
1619  *			      associated to a device
1620  * @dev:	Pointer to device struct of the device for which the interrupts
1621  *		are freed
1622  * @domid:	The id of the domain to operate on
1623  */
msi_domain_free_irqs_all(struct device * dev,unsigned int domid)1624 void msi_domain_free_irqs_all(struct device *dev, unsigned int domid)
1625 {
1626 	msi_lock_descs(dev);
1627 	msi_domain_free_irqs_all_locked(dev, domid);
1628 	msi_unlock_descs(dev);
1629 }
1630 
1631 /**
1632  * msi_get_domain_info - Get the MSI interrupt domain info for @domain
1633  * @domain:	The interrupt domain to retrieve data from
1634  *
1635  * Return: the pointer to the msi_domain_info stored in @domain->host_data.
1636  */
msi_get_domain_info(struct irq_domain * domain)1637 struct msi_domain_info *msi_get_domain_info(struct irq_domain *domain)
1638 {
1639 	return (struct msi_domain_info *)domain->host_data;
1640 }
1641 
1642 /**
1643  * msi_device_has_isolated_msi - True if the device has isolated MSI
1644  * @dev: The device to check
1645  *
1646  * Isolated MSI means that HW modeled by an irq_domain on the path from the
1647  * initiating device to the CPU will validate that the MSI message specifies an
1648  * interrupt number that the device is authorized to trigger. This must block
1649  * devices from triggering interrupts they are not authorized to trigger.
1650  * Currently authorization means the MSI vector is one assigned to the device.
1651  *
1652  * This is interesting for securing VFIO use cases where a rouge MSI (eg created
1653  * by abusing a normal PCI MemWr DMA) must not allow the VFIO userspace to
1654  * impact outside its security domain, eg userspace triggering interrupts on
1655  * kernel drivers, a VM triggering interrupts on the hypervisor, or a VM
1656  * triggering interrupts on another VM.
1657  */
msi_device_has_isolated_msi(struct device * dev)1658 bool msi_device_has_isolated_msi(struct device *dev)
1659 {
1660 	struct irq_domain *domain = dev_get_msi_domain(dev);
1661 
1662 	for (; domain; domain = domain->parent)
1663 		if (domain->flags & IRQ_DOMAIN_FLAG_ISOLATED_MSI)
1664 			return true;
1665 	return arch_is_isolated_msi();
1666 }
1667 EXPORT_SYMBOL_GPL(msi_device_has_isolated_msi);
1668