1 // SPDX-License-Identifier: GPL-2.0
3 * PCI Message Signaled Interrupt (MSI)
5 * Copyright (C) 2003-2004 Intel
7 * Copyright (C) 2016 Christoph Hellwig.
10 #include <linux/err.h>
12 #include <linux/irq.h>
13 #include <linux/interrupt.h>
14 #include <linux/export.h>
15 #include <linux/ioport.h>
16 #include <linux/pci.h>
17 #include <linux/proc_fs.h>
18 #include <linux/msi.h>
19 #include <linux/smp.h>
20 #include <linux/errno.h>
22 #include <linux/acpi_iort.h>
23 #include <linux/slab.h>
24 #include <linux/irqdomain.h>
25 #include <linux/of_irq.h>
31 static int pci_msi_enable = 1;
32 int pci_msi_ignore_mask;
34 #define msix_table_size(flags) ((flags & PCI_MSIX_FLAGS_QSIZE) + 1)
36 #ifdef CONFIG_PCI_MSI_IRQ_DOMAIN
37 static int pci_msi_setup_msi_irqs(struct pci_dev *dev, int nvec, int type)
39 struct irq_domain *domain;
41 domain = dev_get_msi_domain(&dev->dev);
42 if (domain && irq_domain_is_hierarchy(domain))
43 return msi_domain_alloc_irqs(domain, &dev->dev, nvec);
45 return arch_setup_msi_irqs(dev, nvec, type);
48 static void pci_msi_teardown_msi_irqs(struct pci_dev *dev)
50 struct irq_domain *domain;
52 domain = dev_get_msi_domain(&dev->dev);
53 if (domain && irq_domain_is_hierarchy(domain))
54 msi_domain_free_irqs(domain, &dev->dev);
56 arch_teardown_msi_irqs(dev);
59 #define pci_msi_setup_msi_irqs arch_setup_msi_irqs
60 #define pci_msi_teardown_msi_irqs arch_teardown_msi_irqs
63 #ifdef CONFIG_PCI_MSI_ARCH_FALLBACKS
65 int __weak arch_setup_msi_irq(struct pci_dev *dev, struct msi_desc *desc)
67 struct msi_controller *chip = dev->bus->msi;
70 if (!chip || !chip->setup_irq)
73 err = chip->setup_irq(chip, dev, desc);
77 irq_set_chip_data(desc->irq, chip);
82 void __weak arch_teardown_msi_irq(unsigned int irq)
84 struct msi_controller *chip = irq_get_chip_data(irq);
86 if (!chip || !chip->teardown_irq)
89 chip->teardown_irq(chip, irq);
92 int __weak arch_setup_msi_irqs(struct pci_dev *dev, int nvec, int type)
94 struct msi_controller *chip = dev->bus->msi;
95 struct msi_desc *entry;
98 if (chip && chip->setup_irqs)
99 return chip->setup_irqs(chip, dev, nvec, type);
101 * If an architecture wants to support multiple MSI, it needs to
102 * override arch_setup_msi_irqs()
104 if (type == PCI_CAP_ID_MSI && nvec > 1)
107 for_each_pci_msi_entry(entry, dev) {
108 ret = arch_setup_msi_irq(dev, entry);
119 * We have a default implementation available as a separate non-weak
120 * function, as it is used by the Xen x86 PCI code
122 void default_teardown_msi_irqs(struct pci_dev *dev)
125 struct msi_desc *entry;
127 for_each_pci_msi_entry(entry, dev)
129 for (i = 0; i < entry->nvec_used; i++)
130 arch_teardown_msi_irq(entry->irq + i);
133 void __weak arch_teardown_msi_irqs(struct pci_dev *dev)
135 return default_teardown_msi_irqs(dev);
137 #endif /* CONFIG_PCI_MSI_ARCH_FALLBACKS */
139 static void default_restore_msi_irq(struct pci_dev *dev, int irq)
141 struct msi_desc *entry;
144 if (dev->msix_enabled) {
145 for_each_pci_msi_entry(entry, dev) {
146 if (irq == entry->irq)
149 } else if (dev->msi_enabled) {
150 entry = irq_get_msi_desc(irq);
154 __pci_write_msi_msg(entry, &entry->msg);
157 void __weak arch_restore_msi_irqs(struct pci_dev *dev)
159 return default_restore_msi_irqs(dev);
162 static inline __attribute_const__ u32 msi_mask(unsigned x)
164 /* Don't shift by >= width of type */
167 return (1 << (1 << x)) - 1;
171 * PCI 2.3 does not specify mask bits for each MSI interrupt. Attempting to
172 * mask all MSI interrupts by clearing the MSI enable bit does not work
173 * reliably as devices without an INTx disable bit will then generate a
174 * level IRQ which will never be cleared.
176 u32 __pci_msi_desc_mask_irq(struct msi_desc *desc, u32 mask, u32 flag)
178 u32 mask_bits = desc->masked;
180 if (pci_msi_ignore_mask || !desc->msi_attrib.maskbit)
185 pci_write_config_dword(msi_desc_to_pci_dev(desc), desc->mask_pos,
191 static void msi_mask_irq(struct msi_desc *desc, u32 mask, u32 flag)
193 desc->masked = __pci_msi_desc_mask_irq(desc, mask, flag);
196 static void __iomem *pci_msix_desc_addr(struct msi_desc *desc)
198 if (desc->msi_attrib.is_virtual)
201 return desc->mask_base +
202 desc->msi_attrib.entry_nr * PCI_MSIX_ENTRY_SIZE;
206 * This internal function does not flush PCI writes to the device.
207 * All users must ensure that they read from the device before either
208 * assuming that the device state is up to date, or returning out of this
209 * file. This saves a few milliseconds when initialising devices with lots
210 * of MSI-X interrupts.
212 u32 __pci_msix_desc_mask_irq(struct msi_desc *desc, u32 flag)
214 u32 mask_bits = desc->masked;
215 void __iomem *desc_addr;
217 if (pci_msi_ignore_mask)
220 desc_addr = pci_msix_desc_addr(desc);
224 mask_bits &= ~PCI_MSIX_ENTRY_CTRL_MASKBIT;
225 if (flag & PCI_MSIX_ENTRY_CTRL_MASKBIT)
226 mask_bits |= PCI_MSIX_ENTRY_CTRL_MASKBIT;
228 writel(mask_bits, desc_addr + PCI_MSIX_ENTRY_VECTOR_CTRL);
233 static void msix_mask_irq(struct msi_desc *desc, u32 flag)
235 desc->masked = __pci_msix_desc_mask_irq(desc, flag);
238 static void msi_set_mask_bit(struct irq_data *data, u32 flag)
240 struct msi_desc *desc = irq_data_get_msi_desc(data);
242 if (desc->msi_attrib.is_msix) {
243 msix_mask_irq(desc, flag);
244 readl(desc->mask_base); /* Flush write to device */
246 unsigned offset = data->irq - desc->irq;
247 msi_mask_irq(desc, 1 << offset, flag << offset);
252 * pci_msi_mask_irq - Generic IRQ chip callback to mask PCI/MSI interrupts
253 * @data: pointer to irqdata associated to that interrupt
255 void pci_msi_mask_irq(struct irq_data *data)
257 msi_set_mask_bit(data, 1);
259 EXPORT_SYMBOL_GPL(pci_msi_mask_irq);
262 * pci_msi_unmask_irq - Generic IRQ chip callback to unmask PCI/MSI interrupts
263 * @data: pointer to irqdata associated to that interrupt
265 void pci_msi_unmask_irq(struct irq_data *data)
267 msi_set_mask_bit(data, 0);
269 EXPORT_SYMBOL_GPL(pci_msi_unmask_irq);
271 void default_restore_msi_irqs(struct pci_dev *dev)
273 struct msi_desc *entry;
275 for_each_pci_msi_entry(entry, dev)
276 default_restore_msi_irq(dev, entry->irq);
279 void __pci_read_msi_msg(struct msi_desc *entry, struct msi_msg *msg)
281 struct pci_dev *dev = msi_desc_to_pci_dev(entry);
283 BUG_ON(dev->current_state != PCI_D0);
285 if (entry->msi_attrib.is_msix) {
286 void __iomem *base = pci_msix_desc_addr(entry);
293 msg->address_lo = readl(base + PCI_MSIX_ENTRY_LOWER_ADDR);
294 msg->address_hi = readl(base + PCI_MSIX_ENTRY_UPPER_ADDR);
295 msg->data = readl(base + PCI_MSIX_ENTRY_DATA);
297 int pos = dev->msi_cap;
300 pci_read_config_dword(dev, pos + PCI_MSI_ADDRESS_LO,
302 if (entry->msi_attrib.is_64) {
303 pci_read_config_dword(dev, pos + PCI_MSI_ADDRESS_HI,
305 pci_read_config_word(dev, pos + PCI_MSI_DATA_64, &data);
308 pci_read_config_word(dev, pos + PCI_MSI_DATA_32, &data);
314 void __pci_write_msi_msg(struct msi_desc *entry, struct msi_msg *msg)
316 struct pci_dev *dev = msi_desc_to_pci_dev(entry);
318 if (dev->current_state != PCI_D0 || pci_dev_is_disconnected(dev)) {
319 /* Don't touch the hardware now */
320 } else if (entry->msi_attrib.is_msix) {
321 void __iomem *base = pci_msix_desc_addr(entry);
326 writel(msg->address_lo, base + PCI_MSIX_ENTRY_LOWER_ADDR);
327 writel(msg->address_hi, base + PCI_MSIX_ENTRY_UPPER_ADDR);
328 writel(msg->data, base + PCI_MSIX_ENTRY_DATA);
330 int pos = dev->msi_cap;
333 pci_read_config_word(dev, pos + PCI_MSI_FLAGS, &msgctl);
334 msgctl &= ~PCI_MSI_FLAGS_QSIZE;
335 msgctl |= entry->msi_attrib.multiple << 4;
336 pci_write_config_word(dev, pos + PCI_MSI_FLAGS, msgctl);
338 pci_write_config_dword(dev, pos + PCI_MSI_ADDRESS_LO,
340 if (entry->msi_attrib.is_64) {
341 pci_write_config_dword(dev, pos + PCI_MSI_ADDRESS_HI,
343 pci_write_config_word(dev, pos + PCI_MSI_DATA_64,
346 pci_write_config_word(dev, pos + PCI_MSI_DATA_32,
354 if (entry->write_msi_msg)
355 entry->write_msi_msg(entry, entry->write_msi_msg_data);
359 void pci_write_msi_msg(unsigned int irq, struct msi_msg *msg)
361 struct msi_desc *entry = irq_get_msi_desc(irq);
363 __pci_write_msi_msg(entry, msg);
365 EXPORT_SYMBOL_GPL(pci_write_msi_msg);
367 static void free_msi_irqs(struct pci_dev *dev)
369 struct list_head *msi_list = dev_to_msi_list(&dev->dev);
370 struct msi_desc *entry, *tmp;
371 struct attribute **msi_attrs;
372 struct device_attribute *dev_attr;
375 for_each_pci_msi_entry(entry, dev)
377 for (i = 0; i < entry->nvec_used; i++)
378 BUG_ON(irq_has_action(entry->irq + i));
380 pci_msi_teardown_msi_irqs(dev);
382 list_for_each_entry_safe(entry, tmp, msi_list, list) {
383 if (entry->msi_attrib.is_msix) {
384 if (list_is_last(&entry->list, msi_list))
385 iounmap(entry->mask_base);
388 list_del(&entry->list);
389 free_msi_entry(entry);
392 if (dev->msi_irq_groups) {
393 sysfs_remove_groups(&dev->dev.kobj, dev->msi_irq_groups);
394 msi_attrs = dev->msi_irq_groups[0]->attrs;
395 while (msi_attrs[count]) {
396 dev_attr = container_of(msi_attrs[count],
397 struct device_attribute, attr);
398 kfree(dev_attr->attr.name);
403 kfree(dev->msi_irq_groups[0]);
404 kfree(dev->msi_irq_groups);
405 dev->msi_irq_groups = NULL;
409 static void pci_intx_for_msi(struct pci_dev *dev, int enable)
411 if (!(dev->dev_flags & PCI_DEV_FLAGS_MSI_INTX_DISABLE_BUG))
412 pci_intx(dev, enable);
415 static void pci_msi_set_enable(struct pci_dev *dev, int enable)
419 pci_read_config_word(dev, dev->msi_cap + PCI_MSI_FLAGS, &control);
420 control &= ~PCI_MSI_FLAGS_ENABLE;
422 control |= PCI_MSI_FLAGS_ENABLE;
423 pci_write_config_word(dev, dev->msi_cap + PCI_MSI_FLAGS, control);
426 static void __pci_restore_msi_state(struct pci_dev *dev)
429 struct msi_desc *entry;
431 if (!dev->msi_enabled)
434 entry = irq_get_msi_desc(dev->irq);
436 pci_intx_for_msi(dev, 0);
437 pci_msi_set_enable(dev, 0);
438 arch_restore_msi_irqs(dev);
440 pci_read_config_word(dev, dev->msi_cap + PCI_MSI_FLAGS, &control);
441 msi_mask_irq(entry, msi_mask(entry->msi_attrib.multi_cap),
443 control &= ~PCI_MSI_FLAGS_QSIZE;
444 control |= (entry->msi_attrib.multiple << 4) | PCI_MSI_FLAGS_ENABLE;
445 pci_write_config_word(dev, dev->msi_cap + PCI_MSI_FLAGS, control);
448 static void pci_msix_clear_and_set_ctrl(struct pci_dev *dev, u16 clear, u16 set)
452 pci_read_config_word(dev, dev->msix_cap + PCI_MSIX_FLAGS, &ctrl);
455 pci_write_config_word(dev, dev->msix_cap + PCI_MSIX_FLAGS, ctrl);
458 static void __pci_restore_msix_state(struct pci_dev *dev)
460 struct msi_desc *entry;
462 if (!dev->msix_enabled)
464 BUG_ON(list_empty(dev_to_msi_list(&dev->dev)));
466 /* route the table */
467 pci_intx_for_msi(dev, 0);
468 pci_msix_clear_and_set_ctrl(dev, 0,
469 PCI_MSIX_FLAGS_ENABLE | PCI_MSIX_FLAGS_MASKALL);
471 arch_restore_msi_irqs(dev);
472 for_each_pci_msi_entry(entry, dev)
473 msix_mask_irq(entry, entry->masked);
475 pci_msix_clear_and_set_ctrl(dev, PCI_MSIX_FLAGS_MASKALL, 0);
478 void pci_restore_msi_state(struct pci_dev *dev)
480 __pci_restore_msi_state(dev);
481 __pci_restore_msix_state(dev);
483 EXPORT_SYMBOL_GPL(pci_restore_msi_state);
485 static ssize_t msi_mode_show(struct device *dev, struct device_attribute *attr,
488 struct msi_desc *entry;
492 retval = kstrtoul(attr->attr.name, 10, &irq);
496 entry = irq_get_msi_desc(irq);
498 return sprintf(buf, "%s\n",
499 entry->msi_attrib.is_msix ? "msix" : "msi");
504 static int populate_msi_sysfs(struct pci_dev *pdev)
506 struct attribute **msi_attrs;
507 struct attribute *msi_attr;
508 struct device_attribute *msi_dev_attr;
509 struct attribute_group *msi_irq_group;
510 const struct attribute_group **msi_irq_groups;
511 struct msi_desc *entry;
517 /* Determine how many msi entries we have */
518 for_each_pci_msi_entry(entry, pdev)
519 num_msi += entry->nvec_used;
523 /* Dynamically create the MSI attributes for the PCI device */
524 msi_attrs = kcalloc(num_msi + 1, sizeof(void *), GFP_KERNEL);
527 for_each_pci_msi_entry(entry, pdev) {
528 for (i = 0; i < entry->nvec_used; i++) {
529 msi_dev_attr = kzalloc(sizeof(*msi_dev_attr), GFP_KERNEL);
532 msi_attrs[count] = &msi_dev_attr->attr;
534 sysfs_attr_init(&msi_dev_attr->attr);
535 msi_dev_attr->attr.name = kasprintf(GFP_KERNEL, "%d",
537 if (!msi_dev_attr->attr.name)
539 msi_dev_attr->attr.mode = S_IRUGO;
540 msi_dev_attr->show = msi_mode_show;
545 msi_irq_group = kzalloc(sizeof(*msi_irq_group), GFP_KERNEL);
548 msi_irq_group->name = "msi_irqs";
549 msi_irq_group->attrs = msi_attrs;
551 msi_irq_groups = kcalloc(2, sizeof(void *), GFP_KERNEL);
553 goto error_irq_group;
554 msi_irq_groups[0] = msi_irq_group;
556 ret = sysfs_create_groups(&pdev->dev.kobj, msi_irq_groups);
558 goto error_irq_groups;
559 pdev->msi_irq_groups = msi_irq_groups;
564 kfree(msi_irq_groups);
566 kfree(msi_irq_group);
569 msi_attr = msi_attrs[count];
571 msi_dev_attr = container_of(msi_attr, struct device_attribute, attr);
572 kfree(msi_attr->name);
575 msi_attr = msi_attrs[count];
581 static struct msi_desc *
582 msi_setup_entry(struct pci_dev *dev, int nvec, struct irq_affinity *affd)
584 struct irq_affinity_desc *masks = NULL;
585 struct msi_desc *entry;
589 masks = irq_create_affinity_masks(nvec, affd);
591 /* MSI Entry Initialization */
592 entry = alloc_msi_entry(&dev->dev, nvec, masks);
596 pci_read_config_word(dev, dev->msi_cap + PCI_MSI_FLAGS, &control);
598 entry->msi_attrib.is_msix = 0;
599 entry->msi_attrib.is_64 = !!(control & PCI_MSI_FLAGS_64BIT);
600 entry->msi_attrib.is_virtual = 0;
601 entry->msi_attrib.entry_nr = 0;
602 entry->msi_attrib.maskbit = !!(control & PCI_MSI_FLAGS_MASKBIT);
603 entry->msi_attrib.default_irq = dev->irq; /* Save IOAPIC IRQ */
604 entry->msi_attrib.multi_cap = (control & PCI_MSI_FLAGS_QMASK) >> 1;
605 entry->msi_attrib.multiple = ilog2(__roundup_pow_of_two(nvec));
607 if (control & PCI_MSI_FLAGS_64BIT)
608 entry->mask_pos = dev->msi_cap + PCI_MSI_MASK_64;
610 entry->mask_pos = dev->msi_cap + PCI_MSI_MASK_32;
612 /* Save the initial mask status */
613 if (entry->msi_attrib.maskbit)
614 pci_read_config_dword(dev, entry->mask_pos, &entry->masked);
621 static int msi_verify_entries(struct pci_dev *dev)
623 struct msi_desc *entry;
625 for_each_pci_msi_entry(entry, dev) {
626 if (entry->msg.address_hi && dev->no_64bit_msi) {
627 pci_err(dev, "arch assigned 64-bit MSI address %#x%08x but device only supports 32 bits\n",
628 entry->msg.address_hi, entry->msg.address_lo);
636 * msi_capability_init - configure device's MSI capability structure
637 * @dev: pointer to the pci_dev data structure of MSI device function
638 * @nvec: number of interrupts to allocate
639 * @affd: description of automatic IRQ affinity assignments (may be %NULL)
641 * Setup the MSI capability structure of the device with the requested
642 * number of interrupts. A return value of zero indicates the successful
643 * setup of an entry with the new MSI IRQ. A negative return value indicates
644 * an error, and a positive return value indicates the number of interrupts
645 * which could have been allocated.
647 static int msi_capability_init(struct pci_dev *dev, int nvec,
648 struct irq_affinity *affd)
650 struct msi_desc *entry;
654 pci_msi_set_enable(dev, 0); /* Disable MSI during set up */
656 entry = msi_setup_entry(dev, nvec, affd);
660 /* All MSIs are unmasked by default; mask them all */
661 mask = msi_mask(entry->msi_attrib.multi_cap);
662 msi_mask_irq(entry, mask, mask);
664 list_add_tail(&entry->list, dev_to_msi_list(&dev->dev));
666 /* Configure MSI capability structure */
667 ret = pci_msi_setup_msi_irqs(dev, nvec, PCI_CAP_ID_MSI);
669 msi_mask_irq(entry, mask, ~mask);
674 ret = msi_verify_entries(dev);
676 msi_mask_irq(entry, mask, ~mask);
681 ret = populate_msi_sysfs(dev);
683 msi_mask_irq(entry, mask, ~mask);
688 /* Set MSI enabled bits */
689 pci_intx_for_msi(dev, 0);
690 pci_msi_set_enable(dev, 1);
691 dev->msi_enabled = 1;
693 pcibios_free_irq(dev);
694 dev->irq = entry->irq;
698 static void __iomem *msix_map_region(struct pci_dev *dev, unsigned nr_entries)
700 resource_size_t phys_addr;
705 pci_read_config_dword(dev, dev->msix_cap + PCI_MSIX_TABLE,
707 bir = (u8)(table_offset & PCI_MSIX_TABLE_BIR);
708 flags = pci_resource_flags(dev, bir);
709 if (!flags || (flags & IORESOURCE_UNSET))
712 table_offset &= PCI_MSIX_TABLE_OFFSET;
713 phys_addr = pci_resource_start(dev, bir) + table_offset;
715 return ioremap(phys_addr, nr_entries * PCI_MSIX_ENTRY_SIZE);
718 static int msix_setup_entries(struct pci_dev *dev, void __iomem *base,
719 struct msix_entry *entries, int nvec,
720 struct irq_affinity *affd)
722 struct irq_affinity_desc *curmsk, *masks = NULL;
723 struct msi_desc *entry;
725 int vec_count = pci_msix_vec_count(dev);
728 masks = irq_create_affinity_masks(nvec, affd);
730 for (i = 0, curmsk = masks; i < nvec; i++) {
731 entry = alloc_msi_entry(&dev->dev, 1, curmsk);
737 /* No enough memory. Don't try again */
742 entry->msi_attrib.is_msix = 1;
743 entry->msi_attrib.is_64 = 1;
745 entry->msi_attrib.entry_nr = entries[i].entry;
747 entry->msi_attrib.entry_nr = i;
749 entry->msi_attrib.is_virtual =
750 entry->msi_attrib.entry_nr >= vec_count;
752 entry->msi_attrib.default_irq = dev->irq;
753 entry->mask_base = base;
755 list_add_tail(&entry->list, dev_to_msi_list(&dev->dev));
765 static void msix_program_entries(struct pci_dev *dev,
766 struct msix_entry *entries)
768 struct msi_desc *entry;
770 void __iomem *desc_addr;
772 for_each_pci_msi_entry(entry, dev) {
774 entries[i++].vector = entry->irq;
776 desc_addr = pci_msix_desc_addr(entry);
778 entry->masked = readl(desc_addr +
779 PCI_MSIX_ENTRY_VECTOR_CTRL);
783 msix_mask_irq(entry, 1);
788 * msix_capability_init - configure device's MSI-X capability
789 * @dev: pointer to the pci_dev data structure of MSI-X device function
790 * @entries: pointer to an array of struct msix_entry entries
791 * @nvec: number of @entries
792 * @affd: Optional pointer to enable automatic affinity assignment
794 * Setup the MSI-X capability structure of device function with a
795 * single MSI-X IRQ. A return of zero indicates the successful setup of
796 * requested MSI-X entries with allocated IRQs or non-zero for otherwise.
798 static int msix_capability_init(struct pci_dev *dev, struct msix_entry *entries,
799 int nvec, struct irq_affinity *affd)
805 /* Ensure MSI-X is disabled while it is set up */
806 pci_msix_clear_and_set_ctrl(dev, PCI_MSIX_FLAGS_ENABLE, 0);
808 pci_read_config_word(dev, dev->msix_cap + PCI_MSIX_FLAGS, &control);
809 /* Request & Map MSI-X table region */
810 base = msix_map_region(dev, msix_table_size(control));
814 ret = msix_setup_entries(dev, base, entries, nvec, affd);
818 ret = pci_msi_setup_msi_irqs(dev, nvec, PCI_CAP_ID_MSIX);
822 /* Check if all MSI entries honor device restrictions */
823 ret = msi_verify_entries(dev);
828 * Some devices require MSI-X to be enabled before we can touch the
829 * MSI-X registers. We need to mask all the vectors to prevent
830 * interrupts coming in before they're fully set up.
832 pci_msix_clear_and_set_ctrl(dev, 0,
833 PCI_MSIX_FLAGS_MASKALL | PCI_MSIX_FLAGS_ENABLE);
835 msix_program_entries(dev, entries);
837 ret = populate_msi_sysfs(dev);
841 /* Set MSI-X enabled bits and unmask the function */
842 pci_intx_for_msi(dev, 0);
843 dev->msix_enabled = 1;
844 pci_msix_clear_and_set_ctrl(dev, PCI_MSIX_FLAGS_MASKALL, 0);
846 pcibios_free_irq(dev);
852 * If we had some success, report the number of IRQs
853 * we succeeded in setting up.
855 struct msi_desc *entry;
858 for_each_pci_msi_entry(entry, dev) {
873 * pci_msi_supported - check whether MSI may be enabled on a device
874 * @dev: pointer to the pci_dev data structure of MSI device function
875 * @nvec: how many MSIs have been requested?
877 * Look at global flags, the device itself, and its parent buses
878 * to determine if MSI/-X are supported for the device. If MSI/-X is
879 * supported return 1, else return 0.
881 static int pci_msi_supported(struct pci_dev *dev, int nvec)
885 /* MSI must be globally enabled and supported by the device */
889 if (!dev || dev->no_msi)
893 * You can't ask to have 0 or less MSIs configured.
895 * b) the list manipulation code assumes nvec >= 1.
901 * Any bridge which does NOT route MSI transactions from its
902 * secondary bus to its primary bus must set NO_MSI flag on
903 * the secondary pci_bus.
904 * We expect only arch-specific PCI host bus controller driver
905 * or quirks for specific PCI bridges to be setting NO_MSI.
907 for (bus = dev->bus; bus; bus = bus->parent)
908 if (bus->bus_flags & PCI_BUS_FLAGS_NO_MSI)
915 * pci_msi_vec_count - Return the number of MSI vectors a device can send
916 * @dev: device to report about
918 * This function returns the number of MSI vectors a device requested via
919 * Multiple Message Capable register. It returns a negative errno if the
920 * device is not capable sending MSI interrupts. Otherwise, the call succeeds
921 * and returns a power of two, up to a maximum of 2^5 (32), according to the
924 int pci_msi_vec_count(struct pci_dev *dev)
932 pci_read_config_word(dev, dev->msi_cap + PCI_MSI_FLAGS, &msgctl);
933 ret = 1 << ((msgctl & PCI_MSI_FLAGS_QMASK) >> 1);
937 EXPORT_SYMBOL(pci_msi_vec_count);
939 static void pci_msi_shutdown(struct pci_dev *dev)
941 struct msi_desc *desc;
944 if (!pci_msi_enable || !dev || !dev->msi_enabled)
947 BUG_ON(list_empty(dev_to_msi_list(&dev->dev)));
948 desc = first_pci_msi_entry(dev);
950 pci_msi_set_enable(dev, 0);
951 pci_intx_for_msi(dev, 1);
952 dev->msi_enabled = 0;
954 /* Return the device with MSI unmasked as initial states */
955 mask = msi_mask(desc->msi_attrib.multi_cap);
956 /* Keep cached state to be restored */
957 __pci_msi_desc_mask_irq(desc, mask, ~mask);
959 /* Restore dev->irq to its default pin-assertion IRQ */
960 dev->irq = desc->msi_attrib.default_irq;
961 pcibios_alloc_irq(dev);
964 void pci_disable_msi(struct pci_dev *dev)
966 if (!pci_msi_enable || !dev || !dev->msi_enabled)
969 pci_msi_shutdown(dev);
972 EXPORT_SYMBOL(pci_disable_msi);
975 * pci_msix_vec_count - return the number of device's MSI-X table entries
976 * @dev: pointer to the pci_dev data structure of MSI-X device function
977 * This function returns the number of device's MSI-X table entries and
978 * therefore the number of MSI-X vectors device is capable of sending.
979 * It returns a negative errno if the device is not capable of sending MSI-X
982 int pci_msix_vec_count(struct pci_dev *dev)
989 pci_read_config_word(dev, dev->msix_cap + PCI_MSIX_FLAGS, &control);
990 return msix_table_size(control);
992 EXPORT_SYMBOL(pci_msix_vec_count);
994 static int __pci_enable_msix(struct pci_dev *dev, struct msix_entry *entries,
995 int nvec, struct irq_affinity *affd, int flags)
1000 if (!pci_msi_supported(dev, nvec) || dev->current_state != PCI_D0)
1003 nr_entries = pci_msix_vec_count(dev);
1006 if (nvec > nr_entries && !(flags & PCI_IRQ_VIRTUAL))
1010 /* Check for any invalid entries */
1011 for (i = 0; i < nvec; i++) {
1012 if (entries[i].entry >= nr_entries)
1013 return -EINVAL; /* invalid entry */
1014 for (j = i + 1; j < nvec; j++) {
1015 if (entries[i].entry == entries[j].entry)
1016 return -EINVAL; /* duplicate entry */
1021 /* Check whether driver already requested for MSI IRQ */
1022 if (dev->msi_enabled) {
1023 pci_info(dev, "can't enable MSI-X (MSI IRQ already assigned)\n");
1026 return msix_capability_init(dev, entries, nvec, affd);
1029 static void pci_msix_shutdown(struct pci_dev *dev)
1031 struct msi_desc *entry;
1033 if (!pci_msi_enable || !dev || !dev->msix_enabled)
1036 if (pci_dev_is_disconnected(dev)) {
1037 dev->msix_enabled = 0;
1041 /* Return the device with MSI-X masked as initial states */
1042 for_each_pci_msi_entry(entry, dev) {
1043 /* Keep cached states to be restored */
1044 __pci_msix_desc_mask_irq(entry, 1);
1047 pci_msix_clear_and_set_ctrl(dev, PCI_MSIX_FLAGS_ENABLE, 0);
1048 pci_intx_for_msi(dev, 1);
1049 dev->msix_enabled = 0;
1050 pcibios_alloc_irq(dev);
1053 void pci_disable_msix(struct pci_dev *dev)
1055 if (!pci_msi_enable || !dev || !dev->msix_enabled)
1058 pci_msix_shutdown(dev);
1061 EXPORT_SYMBOL(pci_disable_msix);
1063 void pci_no_msi(void)
1069 * pci_msi_enabled - is MSI enabled?
1071 * Returns true if MSI has not been disabled by the command-line option
1074 int pci_msi_enabled(void)
1076 return pci_msi_enable;
1078 EXPORT_SYMBOL(pci_msi_enabled);
1080 static int __pci_enable_msi_range(struct pci_dev *dev, int minvec, int maxvec,
1081 struct irq_affinity *affd)
1086 if (!pci_msi_supported(dev, minvec) || dev->current_state != PCI_D0)
1089 /* Check whether driver already requested MSI-X IRQs */
1090 if (dev->msix_enabled) {
1091 pci_info(dev, "can't enable MSI (MSI-X already enabled)\n");
1095 if (maxvec < minvec)
1098 if (WARN_ON_ONCE(dev->msi_enabled))
1101 nvec = pci_msi_vec_count(dev);
1112 nvec = irq_calc_affinity_vectors(minvec, nvec, affd);
1117 rc = msi_capability_init(dev, nvec, affd);
1130 /* deprecated, don't use */
1131 int pci_enable_msi(struct pci_dev *dev)
1133 int rc = __pci_enable_msi_range(dev, 1, 1, NULL);
1138 EXPORT_SYMBOL(pci_enable_msi);
1140 static int __pci_enable_msix_range(struct pci_dev *dev,
1141 struct msix_entry *entries, int minvec,
1142 int maxvec, struct irq_affinity *affd,
1145 int rc, nvec = maxvec;
1147 if (maxvec < minvec)
1150 if (WARN_ON_ONCE(dev->msix_enabled))
1155 nvec = irq_calc_affinity_vectors(minvec, nvec, affd);
1160 rc = __pci_enable_msix(dev, entries, nvec, affd, flags);
1174 * pci_enable_msix_range - configure device's MSI-X capability structure
1175 * @dev: pointer to the pci_dev data structure of MSI-X device function
1176 * @entries: pointer to an array of MSI-X entries
1177 * @minvec: minimum number of MSI-X IRQs requested
1178 * @maxvec: maximum number of MSI-X IRQs requested
1180 * Setup the MSI-X capability structure of device function with a maximum
1181 * possible number of interrupts in the range between @minvec and @maxvec
1182 * upon its software driver call to request for MSI-X mode enabled on its
1183 * hardware device function. It returns a negative errno if an error occurs.
1184 * If it succeeds, it returns the actual number of interrupts allocated and
1185 * indicates the successful configuration of MSI-X capability structure
1186 * with new allocated MSI-X interrupts.
1188 int pci_enable_msix_range(struct pci_dev *dev, struct msix_entry *entries,
1189 int minvec, int maxvec)
1191 return __pci_enable_msix_range(dev, entries, minvec, maxvec, NULL, 0);
1193 EXPORT_SYMBOL(pci_enable_msix_range);
1196 * pci_alloc_irq_vectors_affinity - allocate multiple IRQs for a device
1197 * @dev: PCI device to operate on
1198 * @min_vecs: minimum number of vectors required (must be >= 1)
1199 * @max_vecs: maximum (desired) number of vectors
1200 * @flags: flags or quirks for the allocation
1201 * @affd: optional description of the affinity requirements
1203 * Allocate up to @max_vecs interrupt vectors for @dev, using MSI-X or MSI
1204 * vectors if available, and fall back to a single legacy vector
1205 * if neither is available. Return the number of vectors allocated,
1206 * (which might be smaller than @max_vecs) if successful, or a negative
1207 * error code on error. If less than @min_vecs interrupt vectors are
1208 * available for @dev the function will fail with -ENOSPC.
1210 * To get the Linux IRQ number used for a vector that can be passed to
1211 * request_irq() use the pci_irq_vector() helper.
1213 int pci_alloc_irq_vectors_affinity(struct pci_dev *dev, unsigned int min_vecs,
1214 unsigned int max_vecs, unsigned int flags,
1215 struct irq_affinity *affd)
1217 struct irq_affinity msi_default_affd = {0};
1218 int nvecs = -ENOSPC;
1220 if (flags & PCI_IRQ_AFFINITY) {
1222 affd = &msi_default_affd;
1228 if (flags & PCI_IRQ_MSIX) {
1229 nvecs = __pci_enable_msix_range(dev, NULL, min_vecs, max_vecs,
1235 if (flags & PCI_IRQ_MSI) {
1236 nvecs = __pci_enable_msi_range(dev, min_vecs, max_vecs, affd);
1241 /* use legacy IRQ if allowed */
1242 if (flags & PCI_IRQ_LEGACY) {
1243 if (min_vecs == 1 && dev->irq) {
1245 * Invoke the affinity spreading logic to ensure that
1246 * the device driver can adjust queue configuration
1247 * for the single interrupt case.
1250 irq_create_affinity_masks(1, affd);
1258 EXPORT_SYMBOL(pci_alloc_irq_vectors_affinity);
1261 * pci_free_irq_vectors - free previously allocated IRQs for a device
1262 * @dev: PCI device to operate on
1264 * Undoes the allocations and enabling in pci_alloc_irq_vectors().
1266 void pci_free_irq_vectors(struct pci_dev *dev)
1268 pci_disable_msix(dev);
1269 pci_disable_msi(dev);
1271 EXPORT_SYMBOL(pci_free_irq_vectors);
1274 * pci_irq_vector - return Linux IRQ number of a device vector
1275 * @dev: PCI device to operate on
1276 * @nr: device-relative interrupt vector index (0-based).
1278 int pci_irq_vector(struct pci_dev *dev, unsigned int nr)
1280 if (dev->msix_enabled) {
1281 struct msi_desc *entry;
1284 for_each_pci_msi_entry(entry, dev) {
1293 if (dev->msi_enabled) {
1294 struct msi_desc *entry = first_pci_msi_entry(dev);
1296 if (WARN_ON_ONCE(nr >= entry->nvec_used))
1299 if (WARN_ON_ONCE(nr > 0))
1303 return dev->irq + nr;
1305 EXPORT_SYMBOL(pci_irq_vector);
1308 * pci_irq_get_affinity - return the affinity of a particular MSI vector
1309 * @dev: PCI device to operate on
1310 * @nr: device-relative interrupt vector index (0-based).
1312 const struct cpumask *pci_irq_get_affinity(struct pci_dev *dev, int nr)
1314 if (dev->msix_enabled) {
1315 struct msi_desc *entry;
1318 for_each_pci_msi_entry(entry, dev) {
1320 return &entry->affinity->mask;
1325 } else if (dev->msi_enabled) {
1326 struct msi_desc *entry = first_pci_msi_entry(dev);
1328 if (WARN_ON_ONCE(!entry || !entry->affinity ||
1329 nr >= entry->nvec_used))
1332 return &entry->affinity[nr].mask;
1334 return cpu_possible_mask;
1337 EXPORT_SYMBOL(pci_irq_get_affinity);
1339 struct pci_dev *msi_desc_to_pci_dev(struct msi_desc *desc)
1341 return to_pci_dev(desc->dev);
1343 EXPORT_SYMBOL(msi_desc_to_pci_dev);
1345 void *msi_desc_to_pci_sysdata(struct msi_desc *desc)
1347 struct pci_dev *dev = msi_desc_to_pci_dev(desc);
1349 return dev->bus->sysdata;
1351 EXPORT_SYMBOL_GPL(msi_desc_to_pci_sysdata);
1353 #ifdef CONFIG_PCI_MSI_IRQ_DOMAIN
1355 * pci_msi_domain_write_msg - Helper to write MSI message to PCI config space
1356 * @irq_data: Pointer to interrupt data of the MSI interrupt
1357 * @msg: Pointer to the message
1359 void pci_msi_domain_write_msg(struct irq_data *irq_data, struct msi_msg *msg)
1361 struct msi_desc *desc = irq_data_get_msi_desc(irq_data);
1364 * For MSI-X desc->irq is always equal to irq_data->irq. For
1365 * MSI only the first interrupt of MULTI MSI passes the test.
1367 if (desc->irq == irq_data->irq)
1368 __pci_write_msi_msg(desc, msg);
1372 * pci_msi_domain_calc_hwirq - Generate a unique ID for an MSI source
1373 * @desc: Pointer to the MSI descriptor
1375 * The ID number is only used within the irqdomain.
1377 static irq_hw_number_t pci_msi_domain_calc_hwirq(struct msi_desc *desc)
1379 struct pci_dev *dev = msi_desc_to_pci_dev(desc);
1381 return (irq_hw_number_t)desc->msi_attrib.entry_nr |
1382 pci_dev_id(dev) << 11 |
1383 (pci_domain_nr(dev->bus) & 0xFFFFFFFF) << 27;
1386 static inline bool pci_msi_desc_is_multi_msi(struct msi_desc *desc)
1388 return !desc->msi_attrib.is_msix && desc->nvec_used > 1;
1392 * pci_msi_domain_check_cap - Verify that @domain supports the capabilities
1394 * @domain: The interrupt domain to check
1395 * @info: The domain info for verification
1396 * @dev: The device to check
1399 * 0 if the functionality is supported
1400 * 1 if Multi MSI is requested, but the domain does not support it
1401 * -ENOTSUPP otherwise
1403 int pci_msi_domain_check_cap(struct irq_domain *domain,
1404 struct msi_domain_info *info, struct device *dev)
1406 struct msi_desc *desc = first_pci_msi_entry(to_pci_dev(dev));
1408 /* Special handling to support __pci_enable_msi_range() */
1409 if (pci_msi_desc_is_multi_msi(desc) &&
1410 !(info->flags & MSI_FLAG_MULTI_PCI_MSI))
1412 else if (desc->msi_attrib.is_msix && !(info->flags & MSI_FLAG_PCI_MSIX))
1418 static int pci_msi_domain_handle_error(struct irq_domain *domain,
1419 struct msi_desc *desc, int error)
1421 /* Special handling to support __pci_enable_msi_range() */
1422 if (pci_msi_desc_is_multi_msi(desc) && error == -ENOSPC)
1428 static void pci_msi_domain_set_desc(msi_alloc_info_t *arg,
1429 struct msi_desc *desc)
1432 arg->hwirq = pci_msi_domain_calc_hwirq(desc);
1435 static struct msi_domain_ops pci_msi_domain_ops_default = {
1436 .set_desc = pci_msi_domain_set_desc,
1437 .msi_check = pci_msi_domain_check_cap,
1438 .handle_error = pci_msi_domain_handle_error,
1441 static void pci_msi_domain_update_dom_ops(struct msi_domain_info *info)
1443 struct msi_domain_ops *ops = info->ops;
1446 info->ops = &pci_msi_domain_ops_default;
1448 if (ops->set_desc == NULL)
1449 ops->set_desc = pci_msi_domain_set_desc;
1450 if (ops->msi_check == NULL)
1451 ops->msi_check = pci_msi_domain_check_cap;
1452 if (ops->handle_error == NULL)
1453 ops->handle_error = pci_msi_domain_handle_error;
1457 static void pci_msi_domain_update_chip_ops(struct msi_domain_info *info)
1459 struct irq_chip *chip = info->chip;
1462 if (!chip->irq_write_msi_msg)
1463 chip->irq_write_msi_msg = pci_msi_domain_write_msg;
1464 if (!chip->irq_mask)
1465 chip->irq_mask = pci_msi_mask_irq;
1466 if (!chip->irq_unmask)
1467 chip->irq_unmask = pci_msi_unmask_irq;
1471 * pci_msi_create_irq_domain - Create a MSI interrupt domain
1472 * @fwnode: Optional fwnode of the interrupt controller
1473 * @info: MSI domain info
1474 * @parent: Parent irq domain
1476 * Updates the domain and chip ops and creates a MSI interrupt domain.
1479 * A domain pointer or NULL in case of failure.
1481 struct irq_domain *pci_msi_create_irq_domain(struct fwnode_handle *fwnode,
1482 struct msi_domain_info *info,
1483 struct irq_domain *parent)
1485 struct irq_domain *domain;
1487 if (WARN_ON(info->flags & MSI_FLAG_LEVEL_CAPABLE))
1488 info->flags &= ~MSI_FLAG_LEVEL_CAPABLE;
1490 if (info->flags & MSI_FLAG_USE_DEF_DOM_OPS)
1491 pci_msi_domain_update_dom_ops(info);
1492 if (info->flags & MSI_FLAG_USE_DEF_CHIP_OPS)
1493 pci_msi_domain_update_chip_ops(info);
1495 info->flags |= MSI_FLAG_ACTIVATE_EARLY;
1496 if (IS_ENABLED(CONFIG_GENERIC_IRQ_RESERVATION_MODE))
1497 info->flags |= MSI_FLAG_MUST_REACTIVATE;
1499 /* PCI-MSI is oneshot-safe */
1500 info->chip->flags |= IRQCHIP_ONESHOT_SAFE;
1502 domain = msi_create_irq_domain(fwnode, info, parent);
1506 irq_domain_update_bus_token(domain, DOMAIN_BUS_PCI_MSI);
1509 EXPORT_SYMBOL_GPL(pci_msi_create_irq_domain);
1512 * Users of the generic MSI infrastructure expect a device to have a single ID,
1513 * so with DMA aliases we have to pick the least-worst compromise. Devices with
1514 * DMA phantom functions tend to still emit MSIs from the real function number,
1515 * so we ignore those and only consider topological aliases where either the
1516 * alias device or RID appears on a different bus number. We also make the
1517 * reasonable assumption that bridges are walked in an upstream direction (so
1518 * the last one seen wins), and the much braver assumption that the most likely
1519 * case is that of PCI->PCIe so we should always use the alias RID. This echoes
1520 * the logic from intel_irq_remapping's set_msi_sid(), which presumably works
1521 * well enough in practice; in the face of the horrible PCIe<->PCI-X conditions
1522 * for taking ownership all we can really do is close our eyes and hope...
1524 static int get_msi_id_cb(struct pci_dev *pdev, u16 alias, void *data)
1527 u8 bus = PCI_BUS_NUM(*pa);
1529 if (pdev->bus->number != bus || PCI_BUS_NUM(alias) != bus)
1536 * pci_msi_domain_get_msi_rid - Get the MSI requester id (RID)
1537 * @domain: The interrupt domain
1538 * @pdev: The PCI device.
1540 * The RID for a device is formed from the alias, with a firmware
1541 * supplied mapping applied
1545 u32 pci_msi_domain_get_msi_rid(struct irq_domain *domain, struct pci_dev *pdev)
1547 struct device_node *of_node;
1548 u32 rid = pci_dev_id(pdev);
1550 pci_for_each_dma_alias(pdev, get_msi_id_cb, &rid);
1552 of_node = irq_domain_get_of_node(domain);
1553 rid = of_node ? of_msi_map_id(&pdev->dev, of_node, rid) :
1554 iort_msi_map_id(&pdev->dev, rid);
1560 * pci_msi_get_device_domain - Get the MSI domain for a given PCI device
1561 * @pdev: The PCI device
1563 * Use the firmware data to find a device-specific MSI domain
1564 * (i.e. not one that is set as a default).
1566 * Returns: The corresponding MSI domain or NULL if none has been found.
1568 struct irq_domain *pci_msi_get_device_domain(struct pci_dev *pdev)
1570 struct irq_domain *dom;
1571 u32 rid = pci_dev_id(pdev);
1573 pci_for_each_dma_alias(pdev, get_msi_id_cb, &rid);
1574 dom = of_msi_map_get_device_domain(&pdev->dev, rid, DOMAIN_BUS_PCI_MSI);
1576 dom = iort_get_device_domain(&pdev->dev, rid,
1577 DOMAIN_BUS_PCI_MSI);
1582 * pci_dev_has_special_msi_domain - Check whether the device is handled by
1583 * a non-standard PCI-MSI domain
1584 * @pdev: The PCI device to check.
1586 * Returns: True if the device irqdomain or the bus irqdomain is
1587 * non-standard PCI/MSI.
1589 bool pci_dev_has_special_msi_domain(struct pci_dev *pdev)
1591 struct irq_domain *dom = dev_get_msi_domain(&pdev->dev);
1594 dom = dev_get_msi_domain(&pdev->bus->dev);
1599 return dom->bus_token != DOMAIN_BUS_PCI_MSI;
1602 #endif /* CONFIG_PCI_MSI_IRQ_DOMAIN */
1603 #endif /* CONFIG_PCI_MSI */
1605 void pci_msi_init(struct pci_dev *dev)
1610 * Disable the MSI hardware to avoid screaming interrupts
1611 * during boot. This is the power on reset default so
1612 * usually this should be a noop.
1614 dev->msi_cap = pci_find_capability(dev, PCI_CAP_ID_MSI);
1618 pci_read_config_word(dev, dev->msi_cap + PCI_MSI_FLAGS, &ctrl);
1619 if (ctrl & PCI_MSI_FLAGS_ENABLE)
1620 pci_write_config_word(dev, dev->msi_cap + PCI_MSI_FLAGS,
1621 ctrl & ~PCI_MSI_FLAGS_ENABLE);
1623 if (!(ctrl & PCI_MSI_FLAGS_64BIT))
1624 dev->no_64bit_msi = 1;
1627 void pci_msix_init(struct pci_dev *dev)
1631 dev->msix_cap = pci_find_capability(dev, PCI_CAP_ID_MSIX);
1635 pci_read_config_word(dev, dev->msix_cap + PCI_MSIX_FLAGS, &ctrl);
1636 if (ctrl & PCI_MSIX_FLAGS_ENABLE)
1637 pci_write_config_word(dev, dev->msix_cap + PCI_MSIX_FLAGS,
1638 ctrl & ~PCI_MSIX_FLAGS_ENABLE);