xref: /openbmc/linux/kernel/irq/msi.c (revision 2d972b6a)
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 Signalled Interrupt 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 
18 #include "internals.h"
19 
20 /**
21  * alloc_msi_entry - Allocate an initialize msi_entry
22  * @dev:	Pointer to the device for which this is allocated
23  * @nvec:	The number of vectors used in this entry
24  * @affinity:	Optional pointer to an affinity mask array size of @nvec
25  *
26  * If @affinity is not NULL then a an affinity array[@nvec] is allocated
27  * and the affinity masks from @affinity are copied.
28  */
29 struct msi_desc *
30 alloc_msi_entry(struct device *dev, int nvec, const struct cpumask *affinity)
31 {
32 	struct msi_desc *desc;
33 
34 	desc = kzalloc(sizeof(*desc), GFP_KERNEL);
35 	if (!desc)
36 		return NULL;
37 
38 	INIT_LIST_HEAD(&desc->list);
39 	desc->dev = dev;
40 	desc->nvec_used = nvec;
41 	if (affinity) {
42 		desc->affinity = kmemdup(affinity,
43 			nvec * sizeof(*desc->affinity), GFP_KERNEL);
44 		if (!desc->affinity) {
45 			kfree(desc);
46 			return NULL;
47 		}
48 	}
49 
50 	return desc;
51 }
52 
53 void free_msi_entry(struct msi_desc *entry)
54 {
55 	kfree(entry->affinity);
56 	kfree(entry);
57 }
58 
59 void __get_cached_msi_msg(struct msi_desc *entry, struct msi_msg *msg)
60 {
61 	*msg = entry->msg;
62 }
63 
64 void get_cached_msi_msg(unsigned int irq, struct msi_msg *msg)
65 {
66 	struct msi_desc *entry = irq_get_msi_desc(irq);
67 
68 	__get_cached_msi_msg(entry, msg);
69 }
70 EXPORT_SYMBOL_GPL(get_cached_msi_msg);
71 
72 #ifdef CONFIG_GENERIC_MSI_IRQ_DOMAIN
73 static inline void irq_chip_write_msi_msg(struct irq_data *data,
74 					  struct msi_msg *msg)
75 {
76 	data->chip->irq_write_msi_msg(data, msg);
77 }
78 
79 /**
80  * msi_domain_set_affinity - Generic affinity setter function for MSI domains
81  * @irq_data:	The irq data associated to the interrupt
82  * @mask:	The affinity mask to set
83  * @force:	Flag to enforce setting (disable online checks)
84  *
85  * Intended to be used by MSI interrupt controllers which are
86  * implemented with hierarchical domains.
87  */
88 int msi_domain_set_affinity(struct irq_data *irq_data,
89 			    const struct cpumask *mask, bool force)
90 {
91 	struct irq_data *parent = irq_data->parent_data;
92 	struct msi_msg msg;
93 	int ret;
94 
95 	ret = parent->chip->irq_set_affinity(parent, mask, force);
96 	if (ret >= 0 && ret != IRQ_SET_MASK_OK_DONE) {
97 		BUG_ON(irq_chip_compose_msi_msg(irq_data, &msg));
98 		irq_chip_write_msi_msg(irq_data, &msg);
99 	}
100 
101 	return ret;
102 }
103 
104 static int msi_domain_activate(struct irq_domain *domain,
105 			       struct irq_data *irq_data, bool early)
106 {
107 	struct msi_msg msg;
108 
109 	BUG_ON(irq_chip_compose_msi_msg(irq_data, &msg));
110 	irq_chip_write_msi_msg(irq_data, &msg);
111 	return 0;
112 }
113 
114 static void msi_domain_deactivate(struct irq_domain *domain,
115 				  struct irq_data *irq_data)
116 {
117 	struct msi_msg msg;
118 
119 	memset(&msg, 0, sizeof(msg));
120 	irq_chip_write_msi_msg(irq_data, &msg);
121 }
122 
123 static int msi_domain_alloc(struct irq_domain *domain, unsigned int virq,
124 			    unsigned int nr_irqs, void *arg)
125 {
126 	struct msi_domain_info *info = domain->host_data;
127 	struct msi_domain_ops *ops = info->ops;
128 	irq_hw_number_t hwirq = ops->get_hwirq(info, arg);
129 	int i, ret;
130 
131 	if (irq_find_mapping(domain, hwirq) > 0)
132 		return -EEXIST;
133 
134 	if (domain->parent) {
135 		ret = irq_domain_alloc_irqs_parent(domain, virq, nr_irqs, arg);
136 		if (ret < 0)
137 			return ret;
138 	}
139 
140 	for (i = 0; i < nr_irqs; i++) {
141 		ret = ops->msi_init(domain, info, virq + i, hwirq + i, arg);
142 		if (ret < 0) {
143 			if (ops->msi_free) {
144 				for (i--; i > 0; i--)
145 					ops->msi_free(domain, info, virq + i);
146 			}
147 			irq_domain_free_irqs_top(domain, virq, nr_irqs);
148 			return ret;
149 		}
150 	}
151 
152 	return 0;
153 }
154 
155 static void msi_domain_free(struct irq_domain *domain, unsigned int virq,
156 			    unsigned int nr_irqs)
157 {
158 	struct msi_domain_info *info = domain->host_data;
159 	int i;
160 
161 	if (info->ops->msi_free) {
162 		for (i = 0; i < nr_irqs; i++)
163 			info->ops->msi_free(domain, info, virq + i);
164 	}
165 	irq_domain_free_irqs_top(domain, virq, nr_irqs);
166 }
167 
168 static const struct irq_domain_ops msi_domain_ops = {
169 	.alloc		= msi_domain_alloc,
170 	.free		= msi_domain_free,
171 	.activate	= msi_domain_activate,
172 	.deactivate	= msi_domain_deactivate,
173 };
174 
175 #ifdef GENERIC_MSI_DOMAIN_OPS
176 static irq_hw_number_t msi_domain_ops_get_hwirq(struct msi_domain_info *info,
177 						msi_alloc_info_t *arg)
178 {
179 	return arg->hwirq;
180 }
181 
182 static int msi_domain_ops_prepare(struct irq_domain *domain, struct device *dev,
183 				  int nvec, msi_alloc_info_t *arg)
184 {
185 	memset(arg, 0, sizeof(*arg));
186 	return 0;
187 }
188 
189 static void msi_domain_ops_set_desc(msi_alloc_info_t *arg,
190 				    struct msi_desc *desc)
191 {
192 	arg->desc = desc;
193 }
194 #else
195 #define msi_domain_ops_get_hwirq	NULL
196 #define msi_domain_ops_prepare		NULL
197 #define msi_domain_ops_set_desc		NULL
198 #endif /* !GENERIC_MSI_DOMAIN_OPS */
199 
200 static int msi_domain_ops_init(struct irq_domain *domain,
201 			       struct msi_domain_info *info,
202 			       unsigned int virq, irq_hw_number_t hwirq,
203 			       msi_alloc_info_t *arg)
204 {
205 	irq_domain_set_hwirq_and_chip(domain, virq, hwirq, info->chip,
206 				      info->chip_data);
207 	if (info->handler && info->handler_name) {
208 		__irq_set_handler(virq, info->handler, 0, info->handler_name);
209 		if (info->handler_data)
210 			irq_set_handler_data(virq, info->handler_data);
211 	}
212 	return 0;
213 }
214 
215 static int msi_domain_ops_check(struct irq_domain *domain,
216 				struct msi_domain_info *info,
217 				struct device *dev)
218 {
219 	return 0;
220 }
221 
222 static struct msi_domain_ops msi_domain_ops_default = {
223 	.get_hwirq	= msi_domain_ops_get_hwirq,
224 	.msi_init	= msi_domain_ops_init,
225 	.msi_check	= msi_domain_ops_check,
226 	.msi_prepare	= msi_domain_ops_prepare,
227 	.set_desc	= msi_domain_ops_set_desc,
228 };
229 
230 static void msi_domain_update_dom_ops(struct msi_domain_info *info)
231 {
232 	struct msi_domain_ops *ops = info->ops;
233 
234 	if (ops == NULL) {
235 		info->ops = &msi_domain_ops_default;
236 		return;
237 	}
238 
239 	if (ops->get_hwirq == NULL)
240 		ops->get_hwirq = msi_domain_ops_default.get_hwirq;
241 	if (ops->msi_init == NULL)
242 		ops->msi_init = msi_domain_ops_default.msi_init;
243 	if (ops->msi_check == NULL)
244 		ops->msi_check = msi_domain_ops_default.msi_check;
245 	if (ops->msi_prepare == NULL)
246 		ops->msi_prepare = msi_domain_ops_default.msi_prepare;
247 	if (ops->set_desc == NULL)
248 		ops->set_desc = msi_domain_ops_default.set_desc;
249 }
250 
251 static void msi_domain_update_chip_ops(struct msi_domain_info *info)
252 {
253 	struct irq_chip *chip = info->chip;
254 
255 	BUG_ON(!chip || !chip->irq_mask || !chip->irq_unmask);
256 	if (!chip->irq_set_affinity)
257 		chip->irq_set_affinity = msi_domain_set_affinity;
258 }
259 
260 /**
261  * msi_create_irq_domain - Create a MSI interrupt domain
262  * @fwnode:	Optional fwnode of the interrupt controller
263  * @info:	MSI domain info
264  * @parent:	Parent irq domain
265  */
266 struct irq_domain *msi_create_irq_domain(struct fwnode_handle *fwnode,
267 					 struct msi_domain_info *info,
268 					 struct irq_domain *parent)
269 {
270 	struct irq_domain *domain;
271 
272 	if (info->flags & MSI_FLAG_USE_DEF_DOM_OPS)
273 		msi_domain_update_dom_ops(info);
274 	if (info->flags & MSI_FLAG_USE_DEF_CHIP_OPS)
275 		msi_domain_update_chip_ops(info);
276 
277 	domain = irq_domain_create_hierarchy(parent, IRQ_DOMAIN_FLAG_MSI, 0,
278 					     fwnode, &msi_domain_ops, info);
279 
280 	if (domain && !domain->name && info->chip)
281 		domain->name = info->chip->name;
282 
283 	return domain;
284 }
285 
286 int msi_domain_prepare_irqs(struct irq_domain *domain, struct device *dev,
287 			    int nvec, msi_alloc_info_t *arg)
288 {
289 	struct msi_domain_info *info = domain->host_data;
290 	struct msi_domain_ops *ops = info->ops;
291 	int ret;
292 
293 	ret = ops->msi_check(domain, info, dev);
294 	if (ret == 0)
295 		ret = ops->msi_prepare(domain, dev, nvec, arg);
296 
297 	return ret;
298 }
299 
300 int msi_domain_populate_irqs(struct irq_domain *domain, struct device *dev,
301 			     int virq, int nvec, msi_alloc_info_t *arg)
302 {
303 	struct msi_domain_info *info = domain->host_data;
304 	struct msi_domain_ops *ops = info->ops;
305 	struct msi_desc *desc;
306 	int ret = 0;
307 
308 	for_each_msi_entry(desc, dev) {
309 		/* Don't even try the multi-MSI brain damage. */
310 		if (WARN_ON(!desc->irq || desc->nvec_used != 1)) {
311 			ret = -EINVAL;
312 			break;
313 		}
314 
315 		if (!(desc->irq >= virq && desc->irq < (virq + nvec)))
316 			continue;
317 
318 		ops->set_desc(arg, desc);
319 		/* Assumes the domain mutex is held! */
320 		ret = irq_domain_alloc_irqs_hierarchy(domain, desc->irq, 1,
321 						      arg);
322 		if (ret)
323 			break;
324 
325 		irq_set_msi_desc_off(desc->irq, 0, desc);
326 	}
327 
328 	if (ret) {
329 		/* Mop up the damage */
330 		for_each_msi_entry(desc, dev) {
331 			if (!(desc->irq >= virq && desc->irq < (virq + nvec)))
332 				continue;
333 
334 			irq_domain_free_irqs_common(domain, desc->irq, 1);
335 		}
336 	}
337 
338 	return ret;
339 }
340 
341 /*
342  * Carefully check whether the device can use reservation mode. If
343  * reservation mode is enabled then the early activation will assign a
344  * dummy vector to the device. If the PCI/MSI device does not support
345  * masking of the entry then this can result in spurious interrupts when
346  * the device driver is not absolutely careful. But even then a malfunction
347  * of the hardware could result in a spurious interrupt on the dummy vector
348  * and render the device unusable. If the entry can be masked then the core
349  * logic will prevent the spurious interrupt and reservation mode can be
350  * used. For now reservation mode is restricted to PCI/MSI.
351  */
352 static bool msi_check_reservation_mode(struct irq_domain *domain,
353 				       struct msi_domain_info *info,
354 				       struct device *dev)
355 {
356 	struct msi_desc *desc;
357 
358 	if (domain->bus_token != DOMAIN_BUS_PCI_MSI)
359 		return false;
360 
361 	if (!(info->flags & MSI_FLAG_MUST_REACTIVATE))
362 		return false;
363 
364 	if (IS_ENABLED(CONFIG_PCI_MSI) && pci_msi_ignore_mask)
365 		return false;
366 
367 	/*
368 	 * Checking the first MSI descriptor is sufficient. MSIX supports
369 	 * masking and MSI does so when the maskbit is set.
370 	 */
371 	desc = first_msi_entry(dev);
372 	return desc->msi_attrib.is_msix || desc->msi_attrib.maskbit;
373 }
374 
375 /**
376  * msi_domain_alloc_irqs - Allocate interrupts from a MSI interrupt domain
377  * @domain:	The domain to allocate from
378  * @dev:	Pointer to device struct of the device for which the interrupts
379  *		are allocated
380  * @nvec:	The number of interrupts to allocate
381  *
382  * Returns 0 on success or an error code.
383  */
384 int msi_domain_alloc_irqs(struct irq_domain *domain, struct device *dev,
385 			  int nvec)
386 {
387 	struct msi_domain_info *info = domain->host_data;
388 	struct msi_domain_ops *ops = info->ops;
389 	struct irq_data *irq_data;
390 	struct msi_desc *desc;
391 	msi_alloc_info_t arg;
392 	int i, ret, virq;
393 	bool can_reserve;
394 
395 	ret = msi_domain_prepare_irqs(domain, dev, nvec, &arg);
396 	if (ret)
397 		return ret;
398 
399 	for_each_msi_entry(desc, dev) {
400 		ops->set_desc(&arg, desc);
401 
402 		virq = __irq_domain_alloc_irqs(domain, -1, desc->nvec_used,
403 					       dev_to_node(dev), &arg, false,
404 					       desc->affinity);
405 		if (virq < 0) {
406 			ret = -ENOSPC;
407 			if (ops->handle_error)
408 				ret = ops->handle_error(domain, desc, ret);
409 			if (ops->msi_finish)
410 				ops->msi_finish(&arg, ret);
411 			return ret;
412 		}
413 
414 		for (i = 0; i < desc->nvec_used; i++) {
415 			irq_set_msi_desc_off(virq, i, desc);
416 			irq_debugfs_copy_devname(virq + i, dev);
417 		}
418 	}
419 
420 	if (ops->msi_finish)
421 		ops->msi_finish(&arg, 0);
422 
423 	can_reserve = msi_check_reservation_mode(domain, info, dev);
424 
425 	for_each_msi_entry(desc, dev) {
426 		virq = desc->irq;
427 		if (desc->nvec_used == 1)
428 			dev_dbg(dev, "irq %d for MSI\n", virq);
429 		else
430 			dev_dbg(dev, "irq [%d-%d] for MSI\n",
431 				virq, virq + desc->nvec_used - 1);
432 		/*
433 		 * This flag is set by the PCI layer as we need to activate
434 		 * the MSI entries before the PCI layer enables MSI in the
435 		 * card. Otherwise the card latches a random msi message.
436 		 */
437 		if (!(info->flags & MSI_FLAG_ACTIVATE_EARLY))
438 			continue;
439 
440 		irq_data = irq_domain_get_irq_data(domain, desc->irq);
441 		if (!can_reserve)
442 			irqd_clr_can_reserve(irq_data);
443 		ret = irq_domain_activate_irq(irq_data, can_reserve);
444 		if (ret)
445 			goto cleanup;
446 	}
447 
448 	/*
449 	 * If these interrupts use reservation mode, clear the activated bit
450 	 * so request_irq() will assign the final vector.
451 	 */
452 	if (can_reserve) {
453 		for_each_msi_entry(desc, dev) {
454 			irq_data = irq_domain_get_irq_data(domain, desc->irq);
455 			irqd_clr_activated(irq_data);
456 		}
457 	}
458 	return 0;
459 
460 cleanup:
461 	for_each_msi_entry(desc, dev) {
462 		struct irq_data *irqd;
463 
464 		if (desc->irq == virq)
465 			break;
466 
467 		irqd = irq_domain_get_irq_data(domain, desc->irq);
468 		if (irqd_is_activated(irqd))
469 			irq_domain_deactivate_irq(irqd);
470 	}
471 	msi_domain_free_irqs(domain, dev);
472 	return ret;
473 }
474 
475 /**
476  * msi_domain_free_irqs - Free interrupts from a MSI interrupt @domain associated tp @dev
477  * @domain:	The domain to managing the interrupts
478  * @dev:	Pointer to device struct of the device for which the interrupts
479  *		are free
480  */
481 void msi_domain_free_irqs(struct irq_domain *domain, struct device *dev)
482 {
483 	struct msi_desc *desc;
484 
485 	for_each_msi_entry(desc, dev) {
486 		/*
487 		 * We might have failed to allocate an MSI early
488 		 * enough that there is no IRQ associated to this
489 		 * entry. If that's the case, don't do anything.
490 		 */
491 		if (desc->irq) {
492 			irq_domain_free_irqs(desc->irq, desc->nvec_used);
493 			desc->irq = 0;
494 		}
495 	}
496 }
497 
498 /**
499  * msi_get_domain_info - Get the MSI interrupt domain info for @domain
500  * @domain:	The interrupt domain to retrieve data from
501  *
502  * Returns the pointer to the msi_domain_info stored in
503  * @domain->host_data.
504  */
505 struct msi_domain_info *msi_get_domain_info(struct irq_domain *domain)
506 {
507 	return (struct msi_domain_info *)domain->host_data;
508 }
509 
510 #endif /* CONFIG_GENERIC_MSI_IRQ_DOMAIN */
511