1 // SPDX-License-Identifier: GPL-2.0
3 * Copyright (C) 2014 Intel Corp.
6 * This file is licensed under GPLv2.
8 * This file contains common code to support Message Signaled Interrupts for
9 * PCI compatible and non PCI compatible devices.
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/pci.h>
19 #include "internals.h"
22 * alloc_msi_entry - Allocate an initialized msi_desc
23 * @dev: Pointer to the device for which this is allocated
24 * @nvec: The number of vectors used in this entry
25 * @affinity: Optional pointer to an affinity mask array size of @nvec
27 * If @affinity is not %NULL then an affinity array[@nvec] is allocated
28 * and the affinity masks and flags from @affinity are copied.
30 * Return: pointer to allocated &msi_desc on success or %NULL on failure
32 struct msi_desc *alloc_msi_entry(struct device *dev, int nvec,
33 const struct irq_affinity_desc *affinity)
35 struct msi_desc *desc;
37 desc = kzalloc(sizeof(*desc), GFP_KERNEL);
41 INIT_LIST_HEAD(&desc->list);
43 desc->nvec_used = nvec;
45 desc->affinity = kmemdup(affinity,
46 nvec * sizeof(*desc->affinity), GFP_KERNEL);
47 if (!desc->affinity) {
56 void free_msi_entry(struct msi_desc *entry)
58 kfree(entry->affinity);
62 void __get_cached_msi_msg(struct msi_desc *entry, struct msi_msg *msg)
67 void get_cached_msi_msg(unsigned int irq, struct msi_msg *msg)
69 struct msi_desc *entry = irq_get_msi_desc(irq);
71 __get_cached_msi_msg(entry, msg);
73 EXPORT_SYMBOL_GPL(get_cached_msi_msg);
75 static ssize_t msi_mode_show(struct device *dev, struct device_attribute *attr,
78 struct msi_desc *entry;
83 retval = kstrtoul(attr->attr.name, 10, &irq);
87 entry = irq_get_msi_desc(irq);
92 is_msix = entry->msi_attrib.is_msix;
94 return sysfs_emit(buf, "%s\n", is_msix ? "msix" : "msi");
98 * msi_populate_sysfs - Populate msi_irqs sysfs entries for devices
99 * @dev: The device(PCI, platform etc) who will get sysfs entries
101 * Return attribute_group ** so that specific bus MSI can save it to
102 * somewhere during initilizing msi irqs. If devices has no MSI irq,
103 * return NULL; if it fails to populate sysfs, return ERR_PTR
105 const struct attribute_group **msi_populate_sysfs(struct device *dev)
107 const struct attribute_group **msi_irq_groups;
108 struct attribute **msi_attrs, *msi_attr;
109 struct device_attribute *msi_dev_attr;
110 struct attribute_group *msi_irq_group;
111 struct msi_desc *entry;
117 /* Determine how many msi entries we have */
118 for_each_msi_entry(entry, dev)
119 num_msi += entry->nvec_used;
123 /* Dynamically create the MSI attributes for the device */
124 msi_attrs = kcalloc(num_msi + 1, sizeof(void *), GFP_KERNEL);
126 return ERR_PTR(-ENOMEM);
128 for_each_msi_entry(entry, dev) {
129 for (i = 0; i < entry->nvec_used; i++) {
130 msi_dev_attr = kzalloc(sizeof(*msi_dev_attr), GFP_KERNEL);
133 msi_attrs[count] = &msi_dev_attr->attr;
135 sysfs_attr_init(&msi_dev_attr->attr);
136 msi_dev_attr->attr.name = kasprintf(GFP_KERNEL, "%d",
138 if (!msi_dev_attr->attr.name)
140 msi_dev_attr->attr.mode = 0444;
141 msi_dev_attr->show = msi_mode_show;
146 msi_irq_group = kzalloc(sizeof(*msi_irq_group), GFP_KERNEL);
149 msi_irq_group->name = "msi_irqs";
150 msi_irq_group->attrs = msi_attrs;
152 msi_irq_groups = kcalloc(2, sizeof(void *), GFP_KERNEL);
154 goto error_irq_group;
155 msi_irq_groups[0] = msi_irq_group;
157 ret = sysfs_create_groups(&dev->kobj, msi_irq_groups);
159 goto error_irq_groups;
161 return msi_irq_groups;
164 kfree(msi_irq_groups);
166 kfree(msi_irq_group);
169 msi_attr = msi_attrs[count];
171 msi_dev_attr = container_of(msi_attr, struct device_attribute, attr);
172 kfree(msi_attr->name);
175 msi_attr = msi_attrs[count];
182 * msi_destroy_sysfs - Destroy msi_irqs sysfs entries for devices
183 * @dev: The device(PCI, platform etc) who will remove sysfs entries
184 * @msi_irq_groups: attribute_group for device msi_irqs entries
186 void msi_destroy_sysfs(struct device *dev, const struct attribute_group **msi_irq_groups)
188 struct device_attribute *dev_attr;
189 struct attribute **msi_attrs;
192 if (msi_irq_groups) {
193 sysfs_remove_groups(&dev->kobj, msi_irq_groups);
194 msi_attrs = msi_irq_groups[0]->attrs;
195 while (msi_attrs[count]) {
196 dev_attr = container_of(msi_attrs[count],
197 struct device_attribute, attr);
198 kfree(dev_attr->attr.name);
203 kfree(msi_irq_groups[0]);
204 kfree(msi_irq_groups);
208 #ifdef CONFIG_GENERIC_MSI_IRQ_DOMAIN
209 static inline void irq_chip_write_msi_msg(struct irq_data *data,
212 data->chip->irq_write_msi_msg(data, msg);
215 static void msi_check_level(struct irq_domain *domain, struct msi_msg *msg)
217 struct msi_domain_info *info = domain->host_data;
220 * If the MSI provider has messed with the second message and
221 * not advertized that it is level-capable, signal the breakage.
223 WARN_ON(!((info->flags & MSI_FLAG_LEVEL_CAPABLE) &&
224 (info->chip->flags & IRQCHIP_SUPPORTS_LEVEL_MSI)) &&
225 (msg[1].address_lo || msg[1].address_hi || msg[1].data));
229 * msi_domain_set_affinity - Generic affinity setter function for MSI domains
230 * @irq_data: The irq data associated to the interrupt
231 * @mask: The affinity mask to set
232 * @force: Flag to enforce setting (disable online checks)
234 * Intended to be used by MSI interrupt controllers which are
235 * implemented with hierarchical domains.
237 * Return: IRQ_SET_MASK_* result code
239 int msi_domain_set_affinity(struct irq_data *irq_data,
240 const struct cpumask *mask, bool force)
242 struct irq_data *parent = irq_data->parent_data;
243 struct msi_msg msg[2] = { [1] = { }, };
246 ret = parent->chip->irq_set_affinity(parent, mask, force);
247 if (ret >= 0 && ret != IRQ_SET_MASK_OK_DONE) {
248 BUG_ON(irq_chip_compose_msi_msg(irq_data, msg));
249 msi_check_level(irq_data->domain, msg);
250 irq_chip_write_msi_msg(irq_data, msg);
256 static int msi_domain_activate(struct irq_domain *domain,
257 struct irq_data *irq_data, bool early)
259 struct msi_msg msg[2] = { [1] = { }, };
261 BUG_ON(irq_chip_compose_msi_msg(irq_data, msg));
262 msi_check_level(irq_data->domain, msg);
263 irq_chip_write_msi_msg(irq_data, msg);
267 static void msi_domain_deactivate(struct irq_domain *domain,
268 struct irq_data *irq_data)
270 struct msi_msg msg[2];
272 memset(msg, 0, sizeof(msg));
273 irq_chip_write_msi_msg(irq_data, msg);
276 static int msi_domain_alloc(struct irq_domain *domain, unsigned int virq,
277 unsigned int nr_irqs, void *arg)
279 struct msi_domain_info *info = domain->host_data;
280 struct msi_domain_ops *ops = info->ops;
281 irq_hw_number_t hwirq = ops->get_hwirq(info, arg);
284 if (irq_find_mapping(domain, hwirq) > 0)
287 if (domain->parent) {
288 ret = irq_domain_alloc_irqs_parent(domain, virq, nr_irqs, arg);
293 for (i = 0; i < nr_irqs; i++) {
294 ret = ops->msi_init(domain, info, virq + i, hwirq + i, arg);
297 for (i--; i > 0; i--)
298 ops->msi_free(domain, info, virq + i);
300 irq_domain_free_irqs_top(domain, virq, nr_irqs);
308 static void msi_domain_free(struct irq_domain *domain, unsigned int virq,
309 unsigned int nr_irqs)
311 struct msi_domain_info *info = domain->host_data;
314 if (info->ops->msi_free) {
315 for (i = 0; i < nr_irqs; i++)
316 info->ops->msi_free(domain, info, virq + i);
318 irq_domain_free_irqs_top(domain, virq, nr_irqs);
321 static const struct irq_domain_ops msi_domain_ops = {
322 .alloc = msi_domain_alloc,
323 .free = msi_domain_free,
324 .activate = msi_domain_activate,
325 .deactivate = msi_domain_deactivate,
328 static irq_hw_number_t msi_domain_ops_get_hwirq(struct msi_domain_info *info,
329 msi_alloc_info_t *arg)
334 static int msi_domain_ops_prepare(struct irq_domain *domain, struct device *dev,
335 int nvec, msi_alloc_info_t *arg)
337 memset(arg, 0, sizeof(*arg));
341 static void msi_domain_ops_set_desc(msi_alloc_info_t *arg,
342 struct msi_desc *desc)
347 static int msi_domain_ops_init(struct irq_domain *domain,
348 struct msi_domain_info *info,
349 unsigned int virq, irq_hw_number_t hwirq,
350 msi_alloc_info_t *arg)
352 irq_domain_set_hwirq_and_chip(domain, virq, hwirq, info->chip,
354 if (info->handler && info->handler_name) {
355 __irq_set_handler(virq, info->handler, 0, info->handler_name);
356 if (info->handler_data)
357 irq_set_handler_data(virq, info->handler_data);
362 static int msi_domain_ops_check(struct irq_domain *domain,
363 struct msi_domain_info *info,
369 static struct msi_domain_ops msi_domain_ops_default = {
370 .get_hwirq = msi_domain_ops_get_hwirq,
371 .msi_init = msi_domain_ops_init,
372 .msi_check = msi_domain_ops_check,
373 .msi_prepare = msi_domain_ops_prepare,
374 .set_desc = msi_domain_ops_set_desc,
375 .domain_alloc_irqs = __msi_domain_alloc_irqs,
376 .domain_free_irqs = __msi_domain_free_irqs,
379 static void msi_domain_update_dom_ops(struct msi_domain_info *info)
381 struct msi_domain_ops *ops = info->ops;
384 info->ops = &msi_domain_ops_default;
388 if (ops->domain_alloc_irqs == NULL)
389 ops->domain_alloc_irqs = msi_domain_ops_default.domain_alloc_irqs;
390 if (ops->domain_free_irqs == NULL)
391 ops->domain_free_irqs = msi_domain_ops_default.domain_free_irqs;
393 if (!(info->flags & MSI_FLAG_USE_DEF_DOM_OPS))
396 if (ops->get_hwirq == NULL)
397 ops->get_hwirq = msi_domain_ops_default.get_hwirq;
398 if (ops->msi_init == NULL)
399 ops->msi_init = msi_domain_ops_default.msi_init;
400 if (ops->msi_check == NULL)
401 ops->msi_check = msi_domain_ops_default.msi_check;
402 if (ops->msi_prepare == NULL)
403 ops->msi_prepare = msi_domain_ops_default.msi_prepare;
404 if (ops->set_desc == NULL)
405 ops->set_desc = msi_domain_ops_default.set_desc;
408 static void msi_domain_update_chip_ops(struct msi_domain_info *info)
410 struct irq_chip *chip = info->chip;
412 BUG_ON(!chip || !chip->irq_mask || !chip->irq_unmask);
413 if (!chip->irq_set_affinity)
414 chip->irq_set_affinity = msi_domain_set_affinity;
418 * msi_create_irq_domain - Create an MSI interrupt domain
419 * @fwnode: Optional fwnode of the interrupt controller
420 * @info: MSI domain info
421 * @parent: Parent irq domain
423 * Return: pointer to the created &struct irq_domain or %NULL on failure
425 struct irq_domain *msi_create_irq_domain(struct fwnode_handle *fwnode,
426 struct msi_domain_info *info,
427 struct irq_domain *parent)
429 struct irq_domain *domain;
431 msi_domain_update_dom_ops(info);
432 if (info->flags & MSI_FLAG_USE_DEF_CHIP_OPS)
433 msi_domain_update_chip_ops(info);
435 domain = irq_domain_create_hierarchy(parent, IRQ_DOMAIN_FLAG_MSI, 0,
436 fwnode, &msi_domain_ops, info);
438 if (domain && !domain->name && info->chip)
439 domain->name = info->chip->name;
444 int msi_domain_prepare_irqs(struct irq_domain *domain, struct device *dev,
445 int nvec, msi_alloc_info_t *arg)
447 struct msi_domain_info *info = domain->host_data;
448 struct msi_domain_ops *ops = info->ops;
451 ret = ops->msi_check(domain, info, dev);
453 ret = ops->msi_prepare(domain, dev, nvec, arg);
458 int msi_domain_populate_irqs(struct irq_domain *domain, struct device *dev,
459 int virq, int nvec, msi_alloc_info_t *arg)
461 struct msi_domain_info *info = domain->host_data;
462 struct msi_domain_ops *ops = info->ops;
463 struct msi_desc *desc;
466 for_each_msi_entry(desc, dev) {
467 /* Don't even try the multi-MSI brain damage. */
468 if (WARN_ON(!desc->irq || desc->nvec_used != 1)) {
473 if (!(desc->irq >= virq && desc->irq < (virq + nvec)))
476 ops->set_desc(arg, desc);
477 /* Assumes the domain mutex is held! */
478 ret = irq_domain_alloc_irqs_hierarchy(domain, desc->irq, 1,
483 irq_set_msi_desc_off(desc->irq, 0, desc);
487 /* Mop up the damage */
488 for_each_msi_entry(desc, dev) {
489 if (!(desc->irq >= virq && desc->irq < (virq + nvec)))
492 irq_domain_free_irqs_common(domain, desc->irq, 1);
500 * Carefully check whether the device can use reservation mode. If
501 * reservation mode is enabled then the early activation will assign a
502 * dummy vector to the device. If the PCI/MSI device does not support
503 * masking of the entry then this can result in spurious interrupts when
504 * the device driver is not absolutely careful. But even then a malfunction
505 * of the hardware could result in a spurious interrupt on the dummy vector
506 * and render the device unusable. If the entry can be masked then the core
507 * logic will prevent the spurious interrupt and reservation mode can be
508 * used. For now reservation mode is restricted to PCI/MSI.
510 static bool msi_check_reservation_mode(struct irq_domain *domain,
511 struct msi_domain_info *info,
514 struct msi_desc *desc;
516 switch(domain->bus_token) {
517 case DOMAIN_BUS_PCI_MSI:
518 case DOMAIN_BUS_VMD_MSI:
524 if (!(info->flags & MSI_FLAG_MUST_REACTIVATE))
527 if (IS_ENABLED(CONFIG_PCI_MSI) && pci_msi_ignore_mask)
531 * Checking the first MSI descriptor is sufficient. MSIX supports
532 * masking and MSI does so when the can_mask attribute is set.
534 desc = first_msi_entry(dev);
535 return desc->msi_attrib.is_msix || desc->msi_attrib.can_mask;
538 int __msi_domain_alloc_irqs(struct irq_domain *domain, struct device *dev,
541 struct msi_domain_info *info = domain->host_data;
542 struct msi_domain_ops *ops = info->ops;
543 struct irq_data *irq_data;
544 struct msi_desc *desc;
545 msi_alloc_info_t arg = { };
549 ret = msi_domain_prepare_irqs(domain, dev, nvec, &arg);
553 for_each_msi_entry(desc, dev) {
554 ops->set_desc(&arg, desc);
556 virq = __irq_domain_alloc_irqs(domain, -1, desc->nvec_used,
557 dev_to_node(dev), &arg, false,
561 if (ops->handle_error)
562 ret = ops->handle_error(domain, desc, ret);
564 ops->msi_finish(&arg, ret);
568 for (i = 0; i < desc->nvec_used; i++) {
569 irq_set_msi_desc_off(virq, i, desc);
570 irq_debugfs_copy_devname(virq + i, dev);
575 ops->msi_finish(&arg, 0);
577 can_reserve = msi_check_reservation_mode(domain, info, dev);
580 * This flag is set by the PCI layer as we need to activate
581 * the MSI entries before the PCI layer enables MSI in the
582 * card. Otherwise the card latches a random msi message.
584 if (!(info->flags & MSI_FLAG_ACTIVATE_EARLY))
587 for_each_msi_vector(desc, i, dev) {
588 if (desc->irq == i) {
590 dev_dbg(dev, "irq [%d-%d] for MSI\n",
591 virq, virq + desc->nvec_used - 1);
594 irq_data = irq_domain_get_irq_data(domain, i);
596 irqd_clr_can_reserve(irq_data);
597 if (domain->flags & IRQ_DOMAIN_MSI_NOMASK_QUIRK)
598 irqd_set_msi_nomask_quirk(irq_data);
600 ret = irq_domain_activate_irq(irq_data, can_reserve);
607 * If these interrupts use reservation mode, clear the activated bit
608 * so request_irq() will assign the final vector.
611 for_each_msi_vector(desc, i, dev) {
612 irq_data = irq_domain_get_irq_data(domain, i);
613 irqd_clr_activated(irq_data);
619 msi_domain_free_irqs(domain, dev);
624 * msi_domain_alloc_irqs - Allocate interrupts from a MSI interrupt domain
625 * @domain: The domain to allocate from
626 * @dev: Pointer to device struct of the device for which the interrupts
628 * @nvec: The number of interrupts to allocate
630 * Return: %0 on success or an error code.
632 int msi_domain_alloc_irqs(struct irq_domain *domain, struct device *dev,
635 struct msi_domain_info *info = domain->host_data;
636 struct msi_domain_ops *ops = info->ops;
638 return ops->domain_alloc_irqs(domain, dev, nvec);
641 void __msi_domain_free_irqs(struct irq_domain *domain, struct device *dev)
643 struct irq_data *irq_data;
644 struct msi_desc *desc;
647 for_each_msi_vector(desc, i, dev) {
648 irq_data = irq_domain_get_irq_data(domain, i);
649 if (irqd_is_activated(irq_data))
650 irq_domain_deactivate_irq(irq_data);
653 for_each_msi_entry(desc, dev) {
655 * We might have failed to allocate an MSI early
656 * enough that there is no IRQ associated to this
657 * entry. If that's the case, don't do anything.
660 irq_domain_free_irqs(desc->irq, desc->nvec_used);
667 * msi_domain_free_irqs - Free interrupts from a MSI interrupt @domain associated to @dev
668 * @domain: The domain to managing the interrupts
669 * @dev: Pointer to device struct of the device for which the interrupts
672 void msi_domain_free_irqs(struct irq_domain *domain, struct device *dev)
674 struct msi_domain_info *info = domain->host_data;
675 struct msi_domain_ops *ops = info->ops;
677 return ops->domain_free_irqs(domain, dev);
681 * msi_get_domain_info - Get the MSI interrupt domain info for @domain
682 * @domain: The interrupt domain to retrieve data from
684 * Return: the pointer to the msi_domain_info stored in @domain->host_data.
686 struct msi_domain_info *msi_get_domain_info(struct irq_domain *domain)
688 return (struct msi_domain_info *)domain->host_data;
691 #endif /* CONFIG_GENERIC_MSI_IRQ_DOMAIN */