]> git.proxmox.com Git - mirror_ubuntu-jammy-kernel.git/blob - kernel/irq/msi.c
Merge tag 'for-linus-5.11-rc1' of git://git.kernel.org/pub/scm/linux/kernel/git/rw...
[mirror_ubuntu-jammy-kernel.git] / kernel / irq / msi.c
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 an affinity array[@nvec] is allocated
27 * and the affinity masks and flags from @affinity are copied.
28 */
29 struct msi_desc *alloc_msi_entry(struct device *dev, int nvec,
30 const struct irq_affinity_desc *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 static void msi_check_level(struct irq_domain *domain, struct msi_msg *msg)
80 {
81 struct msi_domain_info *info = domain->host_data;
82
83 /*
84 * If the MSI provider has messed with the second message and
85 * not advertized that it is level-capable, signal the breakage.
86 */
87 WARN_ON(!((info->flags & MSI_FLAG_LEVEL_CAPABLE) &&
88 (info->chip->flags & IRQCHIP_SUPPORTS_LEVEL_MSI)) &&
89 (msg[1].address_lo || msg[1].address_hi || msg[1].data));
90 }
91
92 /**
93 * msi_domain_set_affinity - Generic affinity setter function for MSI domains
94 * @irq_data: The irq data associated to the interrupt
95 * @mask: The affinity mask to set
96 * @force: Flag to enforce setting (disable online checks)
97 *
98 * Intended to be used by MSI interrupt controllers which are
99 * implemented with hierarchical domains.
100 */
101 int msi_domain_set_affinity(struct irq_data *irq_data,
102 const struct cpumask *mask, bool force)
103 {
104 struct irq_data *parent = irq_data->parent_data;
105 struct msi_msg msg[2] = { [1] = { }, };
106 int ret;
107
108 ret = parent->chip->irq_set_affinity(parent, mask, force);
109 if (ret >= 0 && ret != IRQ_SET_MASK_OK_DONE) {
110 BUG_ON(irq_chip_compose_msi_msg(irq_data, msg));
111 msi_check_level(irq_data->domain, msg);
112 irq_chip_write_msi_msg(irq_data, msg);
113 }
114
115 return ret;
116 }
117
118 static int msi_domain_activate(struct irq_domain *domain,
119 struct irq_data *irq_data, bool early)
120 {
121 struct msi_msg msg[2] = { [1] = { }, };
122
123 BUG_ON(irq_chip_compose_msi_msg(irq_data, msg));
124 msi_check_level(irq_data->domain, msg);
125 irq_chip_write_msi_msg(irq_data, msg);
126 return 0;
127 }
128
129 static void msi_domain_deactivate(struct irq_domain *domain,
130 struct irq_data *irq_data)
131 {
132 struct msi_msg msg[2];
133
134 memset(msg, 0, sizeof(msg));
135 irq_chip_write_msi_msg(irq_data, msg);
136 }
137
138 static int msi_domain_alloc(struct irq_domain *domain, unsigned int virq,
139 unsigned int nr_irqs, void *arg)
140 {
141 struct msi_domain_info *info = domain->host_data;
142 struct msi_domain_ops *ops = info->ops;
143 irq_hw_number_t hwirq = ops->get_hwirq(info, arg);
144 int i, ret;
145
146 if (irq_find_mapping(domain, hwirq) > 0)
147 return -EEXIST;
148
149 if (domain->parent) {
150 ret = irq_domain_alloc_irqs_parent(domain, virq, nr_irqs, arg);
151 if (ret < 0)
152 return ret;
153 }
154
155 for (i = 0; i < nr_irqs; i++) {
156 ret = ops->msi_init(domain, info, virq + i, hwirq + i, arg);
157 if (ret < 0) {
158 if (ops->msi_free) {
159 for (i--; i > 0; i--)
160 ops->msi_free(domain, info, virq + i);
161 }
162 irq_domain_free_irqs_top(domain, virq, nr_irqs);
163 return ret;
164 }
165 }
166
167 return 0;
168 }
169
170 static void msi_domain_free(struct irq_domain *domain, unsigned int virq,
171 unsigned int nr_irqs)
172 {
173 struct msi_domain_info *info = domain->host_data;
174 int i;
175
176 if (info->ops->msi_free) {
177 for (i = 0; i < nr_irqs; i++)
178 info->ops->msi_free(domain, info, virq + i);
179 }
180 irq_domain_free_irqs_top(domain, virq, nr_irqs);
181 }
182
183 static const struct irq_domain_ops msi_domain_ops = {
184 .alloc = msi_domain_alloc,
185 .free = msi_domain_free,
186 .activate = msi_domain_activate,
187 .deactivate = msi_domain_deactivate,
188 };
189
190 static irq_hw_number_t msi_domain_ops_get_hwirq(struct msi_domain_info *info,
191 msi_alloc_info_t *arg)
192 {
193 return arg->hwirq;
194 }
195
196 static int msi_domain_ops_prepare(struct irq_domain *domain, struct device *dev,
197 int nvec, msi_alloc_info_t *arg)
198 {
199 memset(arg, 0, sizeof(*arg));
200 return 0;
201 }
202
203 static void msi_domain_ops_set_desc(msi_alloc_info_t *arg,
204 struct msi_desc *desc)
205 {
206 arg->desc = desc;
207 }
208
209 static int msi_domain_ops_init(struct irq_domain *domain,
210 struct msi_domain_info *info,
211 unsigned int virq, irq_hw_number_t hwirq,
212 msi_alloc_info_t *arg)
213 {
214 irq_domain_set_hwirq_and_chip(domain, virq, hwirq, info->chip,
215 info->chip_data);
216 if (info->handler && info->handler_name) {
217 __irq_set_handler(virq, info->handler, 0, info->handler_name);
218 if (info->handler_data)
219 irq_set_handler_data(virq, info->handler_data);
220 }
221 return 0;
222 }
223
224 static int msi_domain_ops_check(struct irq_domain *domain,
225 struct msi_domain_info *info,
226 struct device *dev)
227 {
228 return 0;
229 }
230
231 static struct msi_domain_ops msi_domain_ops_default = {
232 .get_hwirq = msi_domain_ops_get_hwirq,
233 .msi_init = msi_domain_ops_init,
234 .msi_check = msi_domain_ops_check,
235 .msi_prepare = msi_domain_ops_prepare,
236 .set_desc = msi_domain_ops_set_desc,
237 .domain_alloc_irqs = __msi_domain_alloc_irqs,
238 .domain_free_irqs = __msi_domain_free_irqs,
239 };
240
241 static void msi_domain_update_dom_ops(struct msi_domain_info *info)
242 {
243 struct msi_domain_ops *ops = info->ops;
244
245 if (ops == NULL) {
246 info->ops = &msi_domain_ops_default;
247 return;
248 }
249
250 if (ops->domain_alloc_irqs == NULL)
251 ops->domain_alloc_irqs = msi_domain_ops_default.domain_alloc_irqs;
252 if (ops->domain_free_irqs == NULL)
253 ops->domain_free_irqs = msi_domain_ops_default.domain_free_irqs;
254
255 if (!(info->flags & MSI_FLAG_USE_DEF_DOM_OPS))
256 return;
257
258 if (ops->get_hwirq == NULL)
259 ops->get_hwirq = msi_domain_ops_default.get_hwirq;
260 if (ops->msi_init == NULL)
261 ops->msi_init = msi_domain_ops_default.msi_init;
262 if (ops->msi_check == NULL)
263 ops->msi_check = msi_domain_ops_default.msi_check;
264 if (ops->msi_prepare == NULL)
265 ops->msi_prepare = msi_domain_ops_default.msi_prepare;
266 if (ops->set_desc == NULL)
267 ops->set_desc = msi_domain_ops_default.set_desc;
268 }
269
270 static void msi_domain_update_chip_ops(struct msi_domain_info *info)
271 {
272 struct irq_chip *chip = info->chip;
273
274 BUG_ON(!chip || !chip->irq_mask || !chip->irq_unmask);
275 if (!chip->irq_set_affinity)
276 chip->irq_set_affinity = msi_domain_set_affinity;
277 }
278
279 /**
280 * msi_create_irq_domain - Create a MSI interrupt domain
281 * @fwnode: Optional fwnode of the interrupt controller
282 * @info: MSI domain info
283 * @parent: Parent irq domain
284 */
285 struct irq_domain *msi_create_irq_domain(struct fwnode_handle *fwnode,
286 struct msi_domain_info *info,
287 struct irq_domain *parent)
288 {
289 struct irq_domain *domain;
290
291 msi_domain_update_dom_ops(info);
292 if (info->flags & MSI_FLAG_USE_DEF_CHIP_OPS)
293 msi_domain_update_chip_ops(info);
294
295 domain = irq_domain_create_hierarchy(parent, IRQ_DOMAIN_FLAG_MSI, 0,
296 fwnode, &msi_domain_ops, info);
297
298 if (domain && !domain->name && info->chip)
299 domain->name = info->chip->name;
300
301 return domain;
302 }
303
304 int msi_domain_prepare_irqs(struct irq_domain *domain, struct device *dev,
305 int nvec, msi_alloc_info_t *arg)
306 {
307 struct msi_domain_info *info = domain->host_data;
308 struct msi_domain_ops *ops = info->ops;
309 int ret;
310
311 ret = ops->msi_check(domain, info, dev);
312 if (ret == 0)
313 ret = ops->msi_prepare(domain, dev, nvec, arg);
314
315 return ret;
316 }
317
318 int msi_domain_populate_irqs(struct irq_domain *domain, struct device *dev,
319 int virq, int nvec, msi_alloc_info_t *arg)
320 {
321 struct msi_domain_info *info = domain->host_data;
322 struct msi_domain_ops *ops = info->ops;
323 struct msi_desc *desc;
324 int ret = 0;
325
326 for_each_msi_entry(desc, dev) {
327 /* Don't even try the multi-MSI brain damage. */
328 if (WARN_ON(!desc->irq || desc->nvec_used != 1)) {
329 ret = -EINVAL;
330 break;
331 }
332
333 if (!(desc->irq >= virq && desc->irq < (virq + nvec)))
334 continue;
335
336 ops->set_desc(arg, desc);
337 /* Assumes the domain mutex is held! */
338 ret = irq_domain_alloc_irqs_hierarchy(domain, desc->irq, 1,
339 arg);
340 if (ret)
341 break;
342
343 irq_set_msi_desc_off(desc->irq, 0, desc);
344 }
345
346 if (ret) {
347 /* Mop up the damage */
348 for_each_msi_entry(desc, dev) {
349 if (!(desc->irq >= virq && desc->irq < (virq + nvec)))
350 continue;
351
352 irq_domain_free_irqs_common(domain, desc->irq, 1);
353 }
354 }
355
356 return ret;
357 }
358
359 /*
360 * Carefully check whether the device can use reservation mode. If
361 * reservation mode is enabled then the early activation will assign a
362 * dummy vector to the device. If the PCI/MSI device does not support
363 * masking of the entry then this can result in spurious interrupts when
364 * the device driver is not absolutely careful. But even then a malfunction
365 * of the hardware could result in a spurious interrupt on the dummy vector
366 * and render the device unusable. If the entry can be masked then the core
367 * logic will prevent the spurious interrupt and reservation mode can be
368 * used. For now reservation mode is restricted to PCI/MSI.
369 */
370 static bool msi_check_reservation_mode(struct irq_domain *domain,
371 struct msi_domain_info *info,
372 struct device *dev)
373 {
374 struct msi_desc *desc;
375
376 switch(domain->bus_token) {
377 case DOMAIN_BUS_PCI_MSI:
378 case DOMAIN_BUS_VMD_MSI:
379 break;
380 default:
381 return false;
382 }
383
384 if (!(info->flags & MSI_FLAG_MUST_REACTIVATE))
385 return false;
386
387 if (IS_ENABLED(CONFIG_PCI_MSI) && pci_msi_ignore_mask)
388 return false;
389
390 /*
391 * Checking the first MSI descriptor is sufficient. MSIX supports
392 * masking and MSI does so when the maskbit is set.
393 */
394 desc = first_msi_entry(dev);
395 return desc->msi_attrib.is_msix || desc->msi_attrib.maskbit;
396 }
397
398 int __msi_domain_alloc_irqs(struct irq_domain *domain, struct device *dev,
399 int nvec)
400 {
401 struct msi_domain_info *info = domain->host_data;
402 struct msi_domain_ops *ops = info->ops;
403 struct irq_data *irq_data;
404 struct msi_desc *desc;
405 msi_alloc_info_t arg;
406 int i, ret, virq;
407 bool can_reserve;
408
409 ret = msi_domain_prepare_irqs(domain, dev, nvec, &arg);
410 if (ret)
411 return ret;
412
413 for_each_msi_entry(desc, dev) {
414 ops->set_desc(&arg, desc);
415
416 virq = __irq_domain_alloc_irqs(domain, -1, desc->nvec_used,
417 dev_to_node(dev), &arg, false,
418 desc->affinity);
419 if (virq < 0) {
420 ret = -ENOSPC;
421 if (ops->handle_error)
422 ret = ops->handle_error(domain, desc, ret);
423 if (ops->msi_finish)
424 ops->msi_finish(&arg, ret);
425 return ret;
426 }
427
428 for (i = 0; i < desc->nvec_used; i++) {
429 irq_set_msi_desc_off(virq, i, desc);
430 irq_debugfs_copy_devname(virq + i, dev);
431 }
432 }
433
434 if (ops->msi_finish)
435 ops->msi_finish(&arg, 0);
436
437 can_reserve = msi_check_reservation_mode(domain, info, dev);
438
439 for_each_msi_entry(desc, dev) {
440 virq = desc->irq;
441 if (desc->nvec_used == 1)
442 dev_dbg(dev, "irq %d for MSI\n", virq);
443 else
444 dev_dbg(dev, "irq [%d-%d] for MSI\n",
445 virq, virq + desc->nvec_used - 1);
446 /*
447 * This flag is set by the PCI layer as we need to activate
448 * the MSI entries before the PCI layer enables MSI in the
449 * card. Otherwise the card latches a random msi message.
450 */
451 if (!(info->flags & MSI_FLAG_ACTIVATE_EARLY))
452 continue;
453
454 irq_data = irq_domain_get_irq_data(domain, desc->irq);
455 if (!can_reserve) {
456 irqd_clr_can_reserve(irq_data);
457 if (domain->flags & IRQ_DOMAIN_MSI_NOMASK_QUIRK)
458 irqd_set_msi_nomask_quirk(irq_data);
459 }
460 ret = irq_domain_activate_irq(irq_data, can_reserve);
461 if (ret)
462 goto cleanup;
463 }
464
465 /*
466 * If these interrupts use reservation mode, clear the activated bit
467 * so request_irq() will assign the final vector.
468 */
469 if (can_reserve) {
470 for_each_msi_entry(desc, dev) {
471 irq_data = irq_domain_get_irq_data(domain, desc->irq);
472 irqd_clr_activated(irq_data);
473 }
474 }
475 return 0;
476
477 cleanup:
478 for_each_msi_entry(desc, dev) {
479 struct irq_data *irqd;
480
481 if (desc->irq == virq)
482 break;
483
484 irqd = irq_domain_get_irq_data(domain, desc->irq);
485 if (irqd_is_activated(irqd))
486 irq_domain_deactivate_irq(irqd);
487 }
488 msi_domain_free_irqs(domain, dev);
489 return ret;
490 }
491
492 /**
493 * msi_domain_alloc_irqs - Allocate interrupts from a MSI interrupt domain
494 * @domain: The domain to allocate from
495 * @dev: Pointer to device struct of the device for which the interrupts
496 * are allocated
497 * @nvec: The number of interrupts to allocate
498 *
499 * Returns 0 on success or an error code.
500 */
501 int msi_domain_alloc_irqs(struct irq_domain *domain, struct device *dev,
502 int nvec)
503 {
504 struct msi_domain_info *info = domain->host_data;
505 struct msi_domain_ops *ops = info->ops;
506
507 return ops->domain_alloc_irqs(domain, dev, nvec);
508 }
509
510 void __msi_domain_free_irqs(struct irq_domain *domain, struct device *dev)
511 {
512 struct msi_desc *desc;
513
514 for_each_msi_entry(desc, dev) {
515 /*
516 * We might have failed to allocate an MSI early
517 * enough that there is no IRQ associated to this
518 * entry. If that's the case, don't do anything.
519 */
520 if (desc->irq) {
521 irq_domain_free_irqs(desc->irq, desc->nvec_used);
522 desc->irq = 0;
523 }
524 }
525 }
526
527 /**
528 * __msi_domain_free_irqs - Free interrupts from a MSI interrupt @domain associated tp @dev
529 * @domain: The domain to managing the interrupts
530 * @dev: Pointer to device struct of the device for which the interrupts
531 * are free
532 */
533 void msi_domain_free_irqs(struct irq_domain *domain, struct device *dev)
534 {
535 struct msi_domain_info *info = domain->host_data;
536 struct msi_domain_ops *ops = info->ops;
537
538 return ops->domain_free_irqs(domain, dev);
539 }
540
541 /**
542 * msi_get_domain_info - Get the MSI interrupt domain info for @domain
543 * @domain: The interrupt domain to retrieve data from
544 *
545 * Returns the pointer to the msi_domain_info stored in
546 * @domain->host_data.
547 */
548 struct msi_domain_info *msi_get_domain_info(struct irq_domain *domain)
549 {
550 return (struct msi_domain_info *)domain->host_data;
551 }
552
553 #endif /* CONFIG_GENERIC_MSI_IRQ_DOMAIN */