1 // SPDX-License-Identifier: GPL-2.0
3 * Copyright (C) 2014 Intel Corp.
4 * Author: Jiang Liu <jiang.liu@linux.intel.com>
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/sysfs.h>
18 #include <linux/pci.h>
20 #include "internals.h"
23 * alloc_msi_entry - Allocate an initialized msi_desc
24 * @dev: Pointer to the device for which this is allocated
25 * @nvec: The number of vectors used in this entry
26 * @affinity: Optional pointer to an affinity mask array size of @nvec
28 * If @affinity is not %NULL then an affinity array[@nvec] is allocated
29 * and the affinity masks and flags from @affinity are copied.
31 * Return: pointer to allocated &msi_desc on success or %NULL on failure
33 struct msi_desc *alloc_msi_entry(struct device *dev, int nvec,
34 const struct irq_affinity_desc *affinity)
36 struct msi_desc *desc;
38 desc = kzalloc(sizeof(*desc), GFP_KERNEL);
42 INIT_LIST_HEAD(&desc->list);
44 desc->nvec_used = nvec;
46 desc->affinity = kmemdup(affinity,
47 nvec * sizeof(*desc->affinity), GFP_KERNEL);
48 if (!desc->affinity) {
57 void free_msi_entry(struct msi_desc *entry)
59 kfree(entry->affinity);
63 void __get_cached_msi_msg(struct msi_desc *entry, struct msi_msg *msg)
68 void get_cached_msi_msg(unsigned int irq, struct msi_msg *msg)
70 struct msi_desc *entry = irq_get_msi_desc(irq);
72 __get_cached_msi_msg(entry, msg);
74 EXPORT_SYMBOL_GPL(get_cached_msi_msg);
76 static void msi_device_data_release(struct device *dev, void *res)
78 WARN_ON_ONCE(!list_empty(&dev->msi_list));
83 * msi_setup_device_data - Setup MSI device data
84 * @dev: Device for which MSI device data should be set up
86 * Return: 0 on success, appropriate error code otherwise
88 * This can be called more than once for @dev. If the MSI device data is
89 * already allocated the call succeeds. The allocated memory is
90 * automatically released when the device is destroyed.
92 int msi_setup_device_data(struct device *dev)
94 struct msi_device_data *md;
99 md = devres_alloc(msi_device_data_release, sizeof(*md), GFP_KERNEL);
109 * msi_get_virq - Return Linux interrupt number of a MSI interrupt
110 * @dev: Device to operate on
111 * @index: MSI interrupt index to look for (0-based)
113 * Return: The Linux interrupt number on success (> 0), 0 if not found
115 unsigned int msi_get_virq(struct device *dev, unsigned int index)
117 struct msi_desc *desc;
123 pcimsi = dev_is_pci(dev) ? to_pci_dev(dev)->msi_enabled : false;
125 for_each_msi_entry(desc, dev) {
126 /* PCI-MSI has only one descriptor for multiple interrupts. */
128 if (desc->irq && index < desc->nvec_used)
129 return desc->irq + index;
134 * PCI-MSIX and platform MSI use a descriptor per
137 if (desc->msi_index == index)
142 EXPORT_SYMBOL_GPL(msi_get_virq);
145 static ssize_t msi_mode_show(struct device *dev, struct device_attribute *attr,
148 /* MSI vs. MSIX is per device not per interrupt */
149 bool is_msix = dev_is_pci(dev) ? to_pci_dev(dev)->msix_enabled : false;
151 return sysfs_emit(buf, "%s\n", is_msix ? "msix" : "msi");
155 * msi_populate_sysfs - Populate msi_irqs sysfs entries for devices
156 * @dev: The device(PCI, platform etc) who will get sysfs entries
158 static const struct attribute_group **msi_populate_sysfs(struct device *dev)
160 const struct attribute_group **msi_irq_groups;
161 struct attribute **msi_attrs, *msi_attr;
162 struct device_attribute *msi_dev_attr;
163 struct attribute_group *msi_irq_group;
164 struct msi_desc *entry;
170 /* Determine how many msi entries we have */
171 for_each_msi_entry(entry, dev)
172 num_msi += entry->nvec_used;
176 /* Dynamically create the MSI attributes for the device */
177 msi_attrs = kcalloc(num_msi + 1, sizeof(void *), GFP_KERNEL);
179 return ERR_PTR(-ENOMEM);
181 for_each_msi_entry(entry, dev) {
182 for (i = 0; i < entry->nvec_used; i++) {
183 msi_dev_attr = kzalloc(sizeof(*msi_dev_attr), GFP_KERNEL);
186 msi_attrs[count] = &msi_dev_attr->attr;
188 sysfs_attr_init(&msi_dev_attr->attr);
189 msi_dev_attr->attr.name = kasprintf(GFP_KERNEL, "%d",
191 if (!msi_dev_attr->attr.name)
193 msi_dev_attr->attr.mode = 0444;
194 msi_dev_attr->show = msi_mode_show;
199 msi_irq_group = kzalloc(sizeof(*msi_irq_group), GFP_KERNEL);
202 msi_irq_group->name = "msi_irqs";
203 msi_irq_group->attrs = msi_attrs;
205 msi_irq_groups = kcalloc(2, sizeof(void *), GFP_KERNEL);
207 goto error_irq_group;
208 msi_irq_groups[0] = msi_irq_group;
210 ret = sysfs_create_groups(&dev->kobj, msi_irq_groups);
212 goto error_irq_groups;
214 return msi_irq_groups;
217 kfree(msi_irq_groups);
219 kfree(msi_irq_group);
222 msi_attr = msi_attrs[count];
224 msi_dev_attr = container_of(msi_attr, struct device_attribute, attr);
225 kfree(msi_attr->name);
228 msi_attr = msi_attrs[count];
235 * msi_device_populate_sysfs - Populate msi_irqs sysfs entries for a device
236 * @dev: The device (PCI, platform etc) which will get sysfs entries
238 int msi_device_populate_sysfs(struct device *dev)
240 const struct attribute_group **group = msi_populate_sysfs(dev);
243 return PTR_ERR(group);
244 dev->msi.data->attrs = group;
249 * msi_device_destroy_sysfs - Destroy msi_irqs sysfs entries for a device
250 * @dev: The device (PCI, platform etc) for which to remove
253 void msi_device_destroy_sysfs(struct device *dev)
255 const struct attribute_group **msi_irq_groups = dev->msi.data->attrs;
256 struct device_attribute *dev_attr;
257 struct attribute **msi_attrs;
260 dev->msi.data->attrs = NULL;
264 sysfs_remove_groups(&dev->kobj, msi_irq_groups);
265 msi_attrs = msi_irq_groups[0]->attrs;
266 while (msi_attrs[count]) {
267 dev_attr = container_of(msi_attrs[count], struct device_attribute, attr);
268 kfree(dev_attr->attr.name);
273 kfree(msi_irq_groups[0]);
274 kfree(msi_irq_groups);
278 #ifdef CONFIG_GENERIC_MSI_IRQ_DOMAIN
279 static inline void irq_chip_write_msi_msg(struct irq_data *data,
282 data->chip->irq_write_msi_msg(data, msg);
285 static void msi_check_level(struct irq_domain *domain, struct msi_msg *msg)
287 struct msi_domain_info *info = domain->host_data;
290 * If the MSI provider has messed with the second message and
291 * not advertized that it is level-capable, signal the breakage.
293 WARN_ON(!((info->flags & MSI_FLAG_LEVEL_CAPABLE) &&
294 (info->chip->flags & IRQCHIP_SUPPORTS_LEVEL_MSI)) &&
295 (msg[1].address_lo || msg[1].address_hi || msg[1].data));
299 * msi_domain_set_affinity - Generic affinity setter function for MSI domains
300 * @irq_data: The irq data associated to the interrupt
301 * @mask: The affinity mask to set
302 * @force: Flag to enforce setting (disable online checks)
304 * Intended to be used by MSI interrupt controllers which are
305 * implemented with hierarchical domains.
307 * Return: IRQ_SET_MASK_* result code
309 int msi_domain_set_affinity(struct irq_data *irq_data,
310 const struct cpumask *mask, bool force)
312 struct irq_data *parent = irq_data->parent_data;
313 struct msi_msg msg[2] = { [1] = { }, };
316 ret = parent->chip->irq_set_affinity(parent, mask, force);
317 if (ret >= 0 && ret != IRQ_SET_MASK_OK_DONE) {
318 BUG_ON(irq_chip_compose_msi_msg(irq_data, msg));
319 msi_check_level(irq_data->domain, msg);
320 irq_chip_write_msi_msg(irq_data, msg);
326 static int msi_domain_activate(struct irq_domain *domain,
327 struct irq_data *irq_data, bool early)
329 struct msi_msg msg[2] = { [1] = { }, };
331 BUG_ON(irq_chip_compose_msi_msg(irq_data, msg));
332 msi_check_level(irq_data->domain, msg);
333 irq_chip_write_msi_msg(irq_data, msg);
337 static void msi_domain_deactivate(struct irq_domain *domain,
338 struct irq_data *irq_data)
340 struct msi_msg msg[2];
342 memset(msg, 0, sizeof(msg));
343 irq_chip_write_msi_msg(irq_data, msg);
346 static int msi_domain_alloc(struct irq_domain *domain, unsigned int virq,
347 unsigned int nr_irqs, void *arg)
349 struct msi_domain_info *info = domain->host_data;
350 struct msi_domain_ops *ops = info->ops;
351 irq_hw_number_t hwirq = ops->get_hwirq(info, arg);
354 if (irq_find_mapping(domain, hwirq) > 0)
357 if (domain->parent) {
358 ret = irq_domain_alloc_irqs_parent(domain, virq, nr_irqs, arg);
363 for (i = 0; i < nr_irqs; i++) {
364 ret = ops->msi_init(domain, info, virq + i, hwirq + i, arg);
367 for (i--; i > 0; i--)
368 ops->msi_free(domain, info, virq + i);
370 irq_domain_free_irqs_top(domain, virq, nr_irqs);
378 static void msi_domain_free(struct irq_domain *domain, unsigned int virq,
379 unsigned int nr_irqs)
381 struct msi_domain_info *info = domain->host_data;
384 if (info->ops->msi_free) {
385 for (i = 0; i < nr_irqs; i++)
386 info->ops->msi_free(domain, info, virq + i);
388 irq_domain_free_irqs_top(domain, virq, nr_irqs);
391 static const struct irq_domain_ops msi_domain_ops = {
392 .alloc = msi_domain_alloc,
393 .free = msi_domain_free,
394 .activate = msi_domain_activate,
395 .deactivate = msi_domain_deactivate,
398 static irq_hw_number_t msi_domain_ops_get_hwirq(struct msi_domain_info *info,
399 msi_alloc_info_t *arg)
404 static int msi_domain_ops_prepare(struct irq_domain *domain, struct device *dev,
405 int nvec, msi_alloc_info_t *arg)
407 memset(arg, 0, sizeof(*arg));
411 static void msi_domain_ops_set_desc(msi_alloc_info_t *arg,
412 struct msi_desc *desc)
417 static int msi_domain_ops_init(struct irq_domain *domain,
418 struct msi_domain_info *info,
419 unsigned int virq, irq_hw_number_t hwirq,
420 msi_alloc_info_t *arg)
422 irq_domain_set_hwirq_and_chip(domain, virq, hwirq, info->chip,
424 if (info->handler && info->handler_name) {
425 __irq_set_handler(virq, info->handler, 0, info->handler_name);
426 if (info->handler_data)
427 irq_set_handler_data(virq, info->handler_data);
432 static int msi_domain_ops_check(struct irq_domain *domain,
433 struct msi_domain_info *info,
439 static struct msi_domain_ops msi_domain_ops_default = {
440 .get_hwirq = msi_domain_ops_get_hwirq,
441 .msi_init = msi_domain_ops_init,
442 .msi_check = msi_domain_ops_check,
443 .msi_prepare = msi_domain_ops_prepare,
444 .set_desc = msi_domain_ops_set_desc,
445 .domain_alloc_irqs = __msi_domain_alloc_irqs,
446 .domain_free_irqs = __msi_domain_free_irqs,
449 static void msi_domain_update_dom_ops(struct msi_domain_info *info)
451 struct msi_domain_ops *ops = info->ops;
454 info->ops = &msi_domain_ops_default;
458 if (ops->domain_alloc_irqs == NULL)
459 ops->domain_alloc_irqs = msi_domain_ops_default.domain_alloc_irqs;
460 if (ops->domain_free_irqs == NULL)
461 ops->domain_free_irqs = msi_domain_ops_default.domain_free_irqs;
463 if (!(info->flags & MSI_FLAG_USE_DEF_DOM_OPS))
466 if (ops->get_hwirq == NULL)
467 ops->get_hwirq = msi_domain_ops_default.get_hwirq;
468 if (ops->msi_init == NULL)
469 ops->msi_init = msi_domain_ops_default.msi_init;
470 if (ops->msi_check == NULL)
471 ops->msi_check = msi_domain_ops_default.msi_check;
472 if (ops->msi_prepare == NULL)
473 ops->msi_prepare = msi_domain_ops_default.msi_prepare;
474 if (ops->set_desc == NULL)
475 ops->set_desc = msi_domain_ops_default.set_desc;
478 static void msi_domain_update_chip_ops(struct msi_domain_info *info)
480 struct irq_chip *chip = info->chip;
482 BUG_ON(!chip || !chip->irq_mask || !chip->irq_unmask);
483 if (!chip->irq_set_affinity)
484 chip->irq_set_affinity = msi_domain_set_affinity;
488 * msi_create_irq_domain - Create an MSI interrupt domain
489 * @fwnode: Optional fwnode of the interrupt controller
490 * @info: MSI domain info
491 * @parent: Parent irq domain
493 * Return: pointer to the created &struct irq_domain or %NULL on failure
495 struct irq_domain *msi_create_irq_domain(struct fwnode_handle *fwnode,
496 struct msi_domain_info *info,
497 struct irq_domain *parent)
499 struct irq_domain *domain;
501 msi_domain_update_dom_ops(info);
502 if (info->flags & MSI_FLAG_USE_DEF_CHIP_OPS)
503 msi_domain_update_chip_ops(info);
505 domain = irq_domain_create_hierarchy(parent, IRQ_DOMAIN_FLAG_MSI, 0,
506 fwnode, &msi_domain_ops, info);
508 if (domain && !domain->name && info->chip)
509 domain->name = info->chip->name;
514 int msi_domain_prepare_irqs(struct irq_domain *domain, struct device *dev,
515 int nvec, msi_alloc_info_t *arg)
517 struct msi_domain_info *info = domain->host_data;
518 struct msi_domain_ops *ops = info->ops;
521 ret = ops->msi_check(domain, info, dev);
523 ret = ops->msi_prepare(domain, dev, nvec, arg);
528 int msi_domain_populate_irqs(struct irq_domain *domain, struct device *dev,
529 int virq, int nvec, msi_alloc_info_t *arg)
531 struct msi_domain_info *info = domain->host_data;
532 struct msi_domain_ops *ops = info->ops;
533 struct msi_desc *desc;
536 for_each_msi_entry(desc, dev) {
537 /* Don't even try the multi-MSI brain damage. */
538 if (WARN_ON(!desc->irq || desc->nvec_used != 1)) {
543 if (!(desc->irq >= virq && desc->irq < (virq + nvec)))
546 ops->set_desc(arg, desc);
547 /* Assumes the domain mutex is held! */
548 ret = irq_domain_alloc_irqs_hierarchy(domain, desc->irq, 1,
553 irq_set_msi_desc_off(desc->irq, 0, desc);
557 /* Mop up the damage */
558 for_each_msi_entry(desc, dev) {
559 if (!(desc->irq >= virq && desc->irq < (virq + nvec)))
562 irq_domain_free_irqs_common(domain, desc->irq, 1);
570 * Carefully check whether the device can use reservation mode. If
571 * reservation mode is enabled then the early activation will assign a
572 * dummy vector to the device. If the PCI/MSI device does not support
573 * masking of the entry then this can result in spurious interrupts when
574 * the device driver is not absolutely careful. But even then a malfunction
575 * of the hardware could result in a spurious interrupt on the dummy vector
576 * and render the device unusable. If the entry can be masked then the core
577 * logic will prevent the spurious interrupt and reservation mode can be
578 * used. For now reservation mode is restricted to PCI/MSI.
580 static bool msi_check_reservation_mode(struct irq_domain *domain,
581 struct msi_domain_info *info,
584 struct msi_desc *desc;
586 switch(domain->bus_token) {
587 case DOMAIN_BUS_PCI_MSI:
588 case DOMAIN_BUS_VMD_MSI:
594 if (!(info->flags & MSI_FLAG_MUST_REACTIVATE))
597 if (IS_ENABLED(CONFIG_PCI_MSI) && pci_msi_ignore_mask)
601 * Checking the first MSI descriptor is sufficient. MSIX supports
602 * masking and MSI does so when the can_mask attribute is set.
604 desc = first_msi_entry(dev);
605 return desc->pci.msi_attrib.is_msix || desc->pci.msi_attrib.can_mask;
608 static int msi_handle_pci_fail(struct irq_domain *domain, struct msi_desc *desc,
611 switch(domain->bus_token) {
612 case DOMAIN_BUS_PCI_MSI:
613 case DOMAIN_BUS_VMD_MSI:
614 if (IS_ENABLED(CONFIG_PCI_MSI))
621 /* Let a failed PCI multi MSI allocation retry */
622 if (desc->nvec_used > 1)
625 /* If there was a successful allocation let the caller know */
626 return allocated ? allocated : -ENOSPC;
629 int __msi_domain_alloc_irqs(struct irq_domain *domain, struct device *dev,
632 struct msi_domain_info *info = domain->host_data;
633 struct msi_domain_ops *ops = info->ops;
634 struct irq_data *irq_data;
635 struct msi_desc *desc;
636 msi_alloc_info_t arg = { };
641 ret = msi_domain_prepare_irqs(domain, dev, nvec, &arg);
645 for_each_msi_entry(desc, dev) {
646 ops->set_desc(&arg, desc);
648 virq = __irq_domain_alloc_irqs(domain, -1, desc->nvec_used,
649 dev_to_node(dev), &arg, false,
652 ret = msi_handle_pci_fail(domain, desc, allocated);
656 for (i = 0; i < desc->nvec_used; i++) {
657 irq_set_msi_desc_off(virq, i, desc);
658 irq_debugfs_copy_devname(virq + i, dev);
663 can_reserve = msi_check_reservation_mode(domain, info, dev);
666 * This flag is set by the PCI layer as we need to activate
667 * the MSI entries before the PCI layer enables MSI in the
668 * card. Otherwise the card latches a random msi message.
670 if (!(info->flags & MSI_FLAG_ACTIVATE_EARLY))
673 for_each_msi_vector(desc, i, dev) {
674 if (desc->irq == i) {
676 dev_dbg(dev, "irq [%d-%d] for MSI\n",
677 virq, virq + desc->nvec_used - 1);
680 irq_data = irq_domain_get_irq_data(domain, i);
682 irqd_clr_can_reserve(irq_data);
683 if (domain->flags & IRQ_DOMAIN_MSI_NOMASK_QUIRK)
684 irqd_set_msi_nomask_quirk(irq_data);
686 ret = irq_domain_activate_irq(irq_data, can_reserve);
693 * If these interrupts use reservation mode, clear the activated bit
694 * so request_irq() will assign the final vector.
697 for_each_msi_vector(desc, i, dev) {
698 irq_data = irq_domain_get_irq_data(domain, i);
699 irqd_clr_activated(irq_data);
705 msi_domain_free_irqs(domain, dev);
710 * msi_domain_alloc_irqs - Allocate interrupts from a MSI interrupt domain
711 * @domain: The domain to allocate from
712 * @dev: Pointer to device struct of the device for which the interrupts
714 * @nvec: The number of interrupts to allocate
716 * Return: %0 on success or an error code.
718 int msi_domain_alloc_irqs(struct irq_domain *domain, struct device *dev,
721 struct msi_domain_info *info = domain->host_data;
722 struct msi_domain_ops *ops = info->ops;
725 ret = ops->domain_alloc_irqs(domain, dev, nvec);
729 if (!(info->flags & MSI_FLAG_DEV_SYSFS))
732 ret = msi_device_populate_sysfs(dev);
734 msi_domain_free_irqs(domain, dev);
738 void __msi_domain_free_irqs(struct irq_domain *domain, struct device *dev)
740 struct irq_data *irq_data;
741 struct msi_desc *desc;
744 for_each_msi_vector(desc, i, dev) {
745 irq_data = irq_domain_get_irq_data(domain, i);
746 if (irqd_is_activated(irq_data))
747 irq_domain_deactivate_irq(irq_data);
750 for_each_msi_entry(desc, dev) {
752 * We might have failed to allocate an MSI early
753 * enough that there is no IRQ associated to this
754 * entry. If that's the case, don't do anything.
757 irq_domain_free_irqs(desc->irq, desc->nvec_used);
764 * msi_domain_free_irqs - Free interrupts from a MSI interrupt @domain associated to @dev
765 * @domain: The domain to managing the interrupts
766 * @dev: Pointer to device struct of the device for which the interrupts
769 void msi_domain_free_irqs(struct irq_domain *domain, struct device *dev)
771 struct msi_domain_info *info = domain->host_data;
772 struct msi_domain_ops *ops = info->ops;
774 if (info->flags & MSI_FLAG_DEV_SYSFS)
775 msi_device_destroy_sysfs(dev);
776 ops->domain_free_irqs(domain, dev);
780 * msi_get_domain_info - Get the MSI interrupt domain info for @domain
781 * @domain: The interrupt domain to retrieve data from
783 * Return: the pointer to the msi_domain_info stored in @domain->host_data.
785 struct msi_domain_info *msi_get_domain_info(struct irq_domain *domain)
787 return (struct msi_domain_info *)domain->host_data;
790 #endif /* CONFIG_GENERIC_MSI_IRQ_DOMAIN */