1 // SPDX-License-Identifier: GPL-2.0
3 * PCI Message Signaled Interrupt (MSI)
5 * Copyright (C) 2003-2004 Intel
6 * Copyright (C) Tom Long Nguyen (tom.l.nguyen@intel.com)
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>
29 static int pci_msi_enable = 1;
30 int pci_msi_ignore_mask;
32 #define msix_table_size(flags) ((flags & PCI_MSIX_FLAGS_QSIZE) + 1)
34 #ifdef CONFIG_PCI_MSI_IRQ_DOMAIN
35 static int pci_msi_setup_msi_irqs(struct pci_dev *dev, int nvec, int type)
37 struct irq_domain *domain;
39 domain = dev_get_msi_domain(&dev->dev);
40 if (domain && irq_domain_is_hierarchy(domain))
41 return msi_domain_alloc_irqs(domain, &dev->dev, nvec);
43 return arch_setup_msi_irqs(dev, nvec, type);
46 static void pci_msi_teardown_msi_irqs(struct pci_dev *dev)
48 struct irq_domain *domain;
50 domain = dev_get_msi_domain(&dev->dev);
51 if (domain && irq_domain_is_hierarchy(domain))
52 msi_domain_free_irqs(domain, &dev->dev);
54 arch_teardown_msi_irqs(dev);
57 #define pci_msi_setup_msi_irqs arch_setup_msi_irqs
58 #define pci_msi_teardown_msi_irqs arch_teardown_msi_irqs
61 #ifdef CONFIG_PCI_MSI_ARCH_FALLBACKS
63 int __weak arch_setup_msi_irq(struct pci_dev *dev, struct msi_desc *desc)
65 struct msi_controller *chip = dev->bus->msi;
68 if (!chip || !chip->setup_irq)
71 err = chip->setup_irq(chip, dev, desc);
75 irq_set_chip_data(desc->irq, chip);
80 void __weak arch_teardown_msi_irq(unsigned int irq)
82 struct msi_controller *chip = irq_get_chip_data(irq);
84 if (!chip || !chip->teardown_irq)
87 chip->teardown_irq(chip, irq);
90 int __weak arch_setup_msi_irqs(struct pci_dev *dev, int nvec, int type)
92 struct msi_controller *chip = dev->bus->msi;
93 struct msi_desc *entry;
96 if (chip && chip->setup_irqs)
97 return chip->setup_irqs(chip, dev, nvec, type);
99 * If an architecture wants to support multiple MSI, it needs to
100 * override arch_setup_msi_irqs()
102 if (type == PCI_CAP_ID_MSI && nvec > 1)
105 for_each_pci_msi_entry(entry, dev) {
106 ret = arch_setup_msi_irq(dev, entry);
117 * We have a default implementation available as a separate non-weak
118 * function, as it is used by the Xen x86 PCI code
120 void default_teardown_msi_irqs(struct pci_dev *dev)
123 struct msi_desc *entry;
125 for_each_pci_msi_entry(entry, dev)
127 for (i = 0; i < entry->nvec_used; i++)
128 arch_teardown_msi_irq(entry->irq + i);
131 void __weak arch_teardown_msi_irqs(struct pci_dev *dev)
133 return default_teardown_msi_irqs(dev);
135 #endif /* CONFIG_PCI_MSI_ARCH_FALLBACKS */
137 static void default_restore_msi_irq(struct pci_dev *dev, int irq)
139 struct msi_desc *entry;
142 if (dev->msix_enabled) {
143 for_each_pci_msi_entry(entry, dev) {
144 if (irq == entry->irq)
147 } else if (dev->msi_enabled) {
148 entry = irq_get_msi_desc(irq);
152 __pci_write_msi_msg(entry, &entry->msg);
155 void __weak arch_restore_msi_irqs(struct pci_dev *dev)
157 return default_restore_msi_irqs(dev);
160 static inline __attribute_const__ u32 msi_mask(unsigned x)
162 /* Don't shift by >= width of type */
165 return (1 << (1 << x)) - 1;
169 * PCI 2.3 does not specify mask bits for each MSI interrupt. Attempting to
170 * mask all MSI interrupts by clearing the MSI enable bit does not work
171 * reliably as devices without an INTx disable bit will then generate a
172 * level IRQ which will never be cleared.
174 u32 __pci_msi_desc_mask_irq(struct msi_desc *desc, u32 mask, u32 flag)
176 u32 mask_bits = desc->masked;
178 if (pci_msi_ignore_mask || !desc->msi_attrib.maskbit)
183 pci_write_config_dword(msi_desc_to_pci_dev(desc), desc->mask_pos,
189 static void msi_mask_irq(struct msi_desc *desc, u32 mask, u32 flag)
191 desc->masked = __pci_msi_desc_mask_irq(desc, mask, flag);
194 static void __iomem *pci_msix_desc_addr(struct msi_desc *desc)
196 if (desc->msi_attrib.is_virtual)
199 return desc->mask_base +
200 desc->msi_attrib.entry_nr * PCI_MSIX_ENTRY_SIZE;
204 * This internal function does not flush PCI writes to the device.
205 * All users must ensure that they read from the device before either
206 * assuming that the device state is up to date, or returning out of this
207 * file. This saves a few milliseconds when initialising devices with lots
208 * of MSI-X interrupts.
210 u32 __pci_msix_desc_mask_irq(struct msi_desc *desc, u32 flag)
212 u32 mask_bits = desc->masked;
213 void __iomem *desc_addr;
215 if (pci_msi_ignore_mask)
218 desc_addr = pci_msix_desc_addr(desc);
222 mask_bits &= ~PCI_MSIX_ENTRY_CTRL_MASKBIT;
223 if (flag & PCI_MSIX_ENTRY_CTRL_MASKBIT)
224 mask_bits |= PCI_MSIX_ENTRY_CTRL_MASKBIT;
226 writel(mask_bits, desc_addr + PCI_MSIX_ENTRY_VECTOR_CTRL);
231 static void msix_mask_irq(struct msi_desc *desc, u32 flag)
233 desc->masked = __pci_msix_desc_mask_irq(desc, flag);
236 static void msi_set_mask_bit(struct irq_data *data, u32 flag)
238 struct msi_desc *desc = irq_data_get_msi_desc(data);
240 if (desc->msi_attrib.is_msix) {
241 msix_mask_irq(desc, flag);
242 readl(desc->mask_base); /* Flush write to device */
244 unsigned offset = data->irq - desc->irq;
245 msi_mask_irq(desc, 1 << offset, flag << offset);
250 * pci_msi_mask_irq - Generic IRQ chip callback to mask PCI/MSI interrupts
251 * @data: pointer to irqdata associated to that interrupt
253 void pci_msi_mask_irq(struct irq_data *data)
255 msi_set_mask_bit(data, 1);
257 EXPORT_SYMBOL_GPL(pci_msi_mask_irq);
260 * pci_msi_unmask_irq - Generic IRQ chip callback to unmask PCI/MSI interrupts
261 * @data: pointer to irqdata associated to that interrupt
263 void pci_msi_unmask_irq(struct irq_data *data)
265 msi_set_mask_bit(data, 0);
267 EXPORT_SYMBOL_GPL(pci_msi_unmask_irq);
269 void default_restore_msi_irqs(struct pci_dev *dev)
271 struct msi_desc *entry;
273 for_each_pci_msi_entry(entry, dev)
274 default_restore_msi_irq(dev, entry->irq);
277 void __pci_read_msi_msg(struct msi_desc *entry, struct msi_msg *msg)
279 struct pci_dev *dev = msi_desc_to_pci_dev(entry);
281 BUG_ON(dev->current_state != PCI_D0);
283 if (entry->msi_attrib.is_msix) {
284 void __iomem *base = pci_msix_desc_addr(entry);
291 msg->address_lo = readl(base + PCI_MSIX_ENTRY_LOWER_ADDR);
292 msg->address_hi = readl(base + PCI_MSIX_ENTRY_UPPER_ADDR);
293 msg->data = readl(base + PCI_MSIX_ENTRY_DATA);
295 int pos = dev->msi_cap;
298 pci_read_config_dword(dev, pos + PCI_MSI_ADDRESS_LO,
300 if (entry->msi_attrib.is_64) {
301 pci_read_config_dword(dev, pos + PCI_MSI_ADDRESS_HI,
303 pci_read_config_word(dev, pos + PCI_MSI_DATA_64, &data);
306 pci_read_config_word(dev, pos + PCI_MSI_DATA_32, &data);
312 void __pci_write_msi_msg(struct msi_desc *entry, struct msi_msg *msg)
314 struct pci_dev *dev = msi_desc_to_pci_dev(entry);
316 if (dev->current_state != PCI_D0 || pci_dev_is_disconnected(dev)) {
317 /* Don't touch the hardware now */
318 } else if (entry->msi_attrib.is_msix) {
319 void __iomem *base = pci_msix_desc_addr(entry);
324 writel(msg->address_lo, base + PCI_MSIX_ENTRY_LOWER_ADDR);
325 writel(msg->address_hi, base + PCI_MSIX_ENTRY_UPPER_ADDR);
326 writel(msg->data, base + PCI_MSIX_ENTRY_DATA);
328 int pos = dev->msi_cap;
331 pci_read_config_word(dev, pos + PCI_MSI_FLAGS, &msgctl);
332 msgctl &= ~PCI_MSI_FLAGS_QSIZE;
333 msgctl |= entry->msi_attrib.multiple << 4;
334 pci_write_config_word(dev, pos + PCI_MSI_FLAGS, msgctl);
336 pci_write_config_dword(dev, pos + PCI_MSI_ADDRESS_LO,
338 if (entry->msi_attrib.is_64) {
339 pci_write_config_dword(dev, pos + PCI_MSI_ADDRESS_HI,
341 pci_write_config_word(dev, pos + PCI_MSI_DATA_64,
344 pci_write_config_word(dev, pos + PCI_MSI_DATA_32,
352 if (entry->write_msi_msg)
353 entry->write_msi_msg(entry, entry->write_msi_msg_data);
357 void pci_write_msi_msg(unsigned int irq, struct msi_msg *msg)
359 struct msi_desc *entry = irq_get_msi_desc(irq);
361 __pci_write_msi_msg(entry, msg);
363 EXPORT_SYMBOL_GPL(pci_write_msi_msg);
365 static void free_msi_irqs(struct pci_dev *dev)
367 struct list_head *msi_list = dev_to_msi_list(&dev->dev);
368 struct msi_desc *entry, *tmp;
369 struct attribute **msi_attrs;
370 struct device_attribute *dev_attr;
373 for_each_pci_msi_entry(entry, dev)
375 for (i = 0; i < entry->nvec_used; i++)
376 BUG_ON(irq_has_action(entry->irq + i));
378 pci_msi_teardown_msi_irqs(dev);
380 list_for_each_entry_safe(entry, tmp, msi_list, list) {
381 if (entry->msi_attrib.is_msix) {
382 if (list_is_last(&entry->list, msi_list))
383 iounmap(entry->mask_base);
386 list_del(&entry->list);
387 free_msi_entry(entry);
390 if (dev->msi_irq_groups) {
391 sysfs_remove_groups(&dev->dev.kobj, dev->msi_irq_groups);
392 msi_attrs = dev->msi_irq_groups[0]->attrs;
393 while (msi_attrs[count]) {
394 dev_attr = container_of(msi_attrs[count],
395 struct device_attribute, attr);
396 kfree(dev_attr->attr.name);
401 kfree(dev->msi_irq_groups[0]);
402 kfree(dev->msi_irq_groups);
403 dev->msi_irq_groups = NULL;
407 static void pci_intx_for_msi(struct pci_dev *dev, int enable)
409 if (!(dev->dev_flags & PCI_DEV_FLAGS_MSI_INTX_DISABLE_BUG))
410 pci_intx(dev, enable);
413 static void __pci_restore_msi_state(struct pci_dev *dev)
416 struct msi_desc *entry;
418 if (!dev->msi_enabled)
421 entry = irq_get_msi_desc(dev->irq);
423 pci_intx_for_msi(dev, 0);
424 pci_msi_set_enable(dev, 0);
425 arch_restore_msi_irqs(dev);
427 pci_read_config_word(dev, dev->msi_cap + PCI_MSI_FLAGS, &control);
428 msi_mask_irq(entry, msi_mask(entry->msi_attrib.multi_cap),
430 control &= ~PCI_MSI_FLAGS_QSIZE;
431 control |= (entry->msi_attrib.multiple << 4) | PCI_MSI_FLAGS_ENABLE;
432 pci_write_config_word(dev, dev->msi_cap + PCI_MSI_FLAGS, control);
435 static void __pci_restore_msix_state(struct pci_dev *dev)
437 struct msi_desc *entry;
439 if (!dev->msix_enabled)
441 BUG_ON(list_empty(dev_to_msi_list(&dev->dev)));
443 /* route the table */
444 pci_intx_for_msi(dev, 0);
445 pci_msix_clear_and_set_ctrl(dev, 0,
446 PCI_MSIX_FLAGS_ENABLE | PCI_MSIX_FLAGS_MASKALL);
448 arch_restore_msi_irqs(dev);
449 for_each_pci_msi_entry(entry, dev)
450 msix_mask_irq(entry, entry->masked);
452 pci_msix_clear_and_set_ctrl(dev, PCI_MSIX_FLAGS_MASKALL, 0);
455 void pci_restore_msi_state(struct pci_dev *dev)
457 __pci_restore_msi_state(dev);
458 __pci_restore_msix_state(dev);
460 EXPORT_SYMBOL_GPL(pci_restore_msi_state);
462 static ssize_t msi_mode_show(struct device *dev, struct device_attribute *attr,
465 struct msi_desc *entry;
469 retval = kstrtoul(attr->attr.name, 10, &irq);
473 entry = irq_get_msi_desc(irq);
475 return sprintf(buf, "%s\n",
476 entry->msi_attrib.is_msix ? "msix" : "msi");
481 static int populate_msi_sysfs(struct pci_dev *pdev)
483 struct attribute **msi_attrs;
484 struct attribute *msi_attr;
485 struct device_attribute *msi_dev_attr;
486 struct attribute_group *msi_irq_group;
487 const struct attribute_group **msi_irq_groups;
488 struct msi_desc *entry;
494 /* Determine how many msi entries we have */
495 for_each_pci_msi_entry(entry, pdev)
496 num_msi += entry->nvec_used;
500 /* Dynamically create the MSI attributes for the PCI device */
501 msi_attrs = kcalloc(num_msi + 1, sizeof(void *), GFP_KERNEL);
504 for_each_pci_msi_entry(entry, pdev) {
505 for (i = 0; i < entry->nvec_used; i++) {
506 msi_dev_attr = kzalloc(sizeof(*msi_dev_attr), GFP_KERNEL);
509 msi_attrs[count] = &msi_dev_attr->attr;
511 sysfs_attr_init(&msi_dev_attr->attr);
512 msi_dev_attr->attr.name = kasprintf(GFP_KERNEL, "%d",
514 if (!msi_dev_attr->attr.name)
516 msi_dev_attr->attr.mode = S_IRUGO;
517 msi_dev_attr->show = msi_mode_show;
522 msi_irq_group = kzalloc(sizeof(*msi_irq_group), GFP_KERNEL);
525 msi_irq_group->name = "msi_irqs";
526 msi_irq_group->attrs = msi_attrs;
528 msi_irq_groups = kcalloc(2, sizeof(void *), GFP_KERNEL);
530 goto error_irq_group;
531 msi_irq_groups[0] = msi_irq_group;
533 ret = sysfs_create_groups(&pdev->dev.kobj, msi_irq_groups);
535 goto error_irq_groups;
536 pdev->msi_irq_groups = msi_irq_groups;
541 kfree(msi_irq_groups);
543 kfree(msi_irq_group);
546 msi_attr = msi_attrs[count];
548 msi_dev_attr = container_of(msi_attr, struct device_attribute, attr);
549 kfree(msi_attr->name);
552 msi_attr = msi_attrs[count];
558 static struct msi_desc *
559 msi_setup_entry(struct pci_dev *dev, int nvec, struct irq_affinity *affd)
561 struct irq_affinity_desc *masks = NULL;
562 struct msi_desc *entry;
566 masks = irq_create_affinity_masks(nvec, affd);
568 /* MSI Entry Initialization */
569 entry = alloc_msi_entry(&dev->dev, nvec, masks);
573 pci_read_config_word(dev, dev->msi_cap + PCI_MSI_FLAGS, &control);
575 entry->msi_attrib.is_msix = 0;
576 entry->msi_attrib.is_64 = !!(control & PCI_MSI_FLAGS_64BIT);
577 entry->msi_attrib.is_virtual = 0;
578 entry->msi_attrib.entry_nr = 0;
579 entry->msi_attrib.maskbit = !!(control & PCI_MSI_FLAGS_MASKBIT);
580 entry->msi_attrib.default_irq = dev->irq; /* Save IOAPIC IRQ */
581 entry->msi_attrib.multi_cap = (control & PCI_MSI_FLAGS_QMASK) >> 1;
582 entry->msi_attrib.multiple = ilog2(__roundup_pow_of_two(nvec));
584 if (control & PCI_MSI_FLAGS_64BIT)
585 entry->mask_pos = dev->msi_cap + PCI_MSI_MASK_64;
587 entry->mask_pos = dev->msi_cap + PCI_MSI_MASK_32;
589 /* Save the initial mask status */
590 if (entry->msi_attrib.maskbit)
591 pci_read_config_dword(dev, entry->mask_pos, &entry->masked);
598 static int msi_verify_entries(struct pci_dev *dev)
600 struct msi_desc *entry;
602 for_each_pci_msi_entry(entry, dev) {
603 if (!dev->no_64bit_msi || !entry->msg.address_hi)
605 pci_err(dev, "Device has broken 64-bit MSI but arch"
606 " tried to assign one above 4G\n");
613 * msi_capability_init - configure device's MSI capability structure
614 * @dev: pointer to the pci_dev data structure of MSI device function
615 * @nvec: number of interrupts to allocate
616 * @affd: description of automatic IRQ affinity assignments (may be %NULL)
618 * Setup the MSI capability structure of the device with the requested
619 * number of interrupts. A return value of zero indicates the successful
620 * setup of an entry with the new MSI IRQ. A negative return value indicates
621 * an error, and a positive return value indicates the number of interrupts
622 * which could have been allocated.
624 static int msi_capability_init(struct pci_dev *dev, int nvec,
625 struct irq_affinity *affd)
627 struct msi_desc *entry;
631 pci_msi_set_enable(dev, 0); /* Disable MSI during set up */
633 entry = msi_setup_entry(dev, nvec, affd);
637 /* All MSIs are unmasked by default; mask them all */
638 mask = msi_mask(entry->msi_attrib.multi_cap);
639 msi_mask_irq(entry, mask, mask);
641 list_add_tail(&entry->list, dev_to_msi_list(&dev->dev));
643 /* Configure MSI capability structure */
644 ret = pci_msi_setup_msi_irqs(dev, nvec, PCI_CAP_ID_MSI);
646 msi_mask_irq(entry, mask, ~mask);
651 ret = msi_verify_entries(dev);
653 msi_mask_irq(entry, mask, ~mask);
658 ret = populate_msi_sysfs(dev);
660 msi_mask_irq(entry, mask, ~mask);
665 /* Set MSI enabled bits */
666 pci_intx_for_msi(dev, 0);
667 pci_msi_set_enable(dev, 1);
668 dev->msi_enabled = 1;
670 pcibios_free_irq(dev);
671 dev->irq = entry->irq;
675 static void __iomem *msix_map_region(struct pci_dev *dev, unsigned nr_entries)
677 resource_size_t phys_addr;
682 pci_read_config_dword(dev, dev->msix_cap + PCI_MSIX_TABLE,
684 bir = (u8)(table_offset & PCI_MSIX_TABLE_BIR);
685 flags = pci_resource_flags(dev, bir);
686 if (!flags || (flags & IORESOURCE_UNSET))
689 table_offset &= PCI_MSIX_TABLE_OFFSET;
690 phys_addr = pci_resource_start(dev, bir) + table_offset;
692 return ioremap(phys_addr, nr_entries * PCI_MSIX_ENTRY_SIZE);
695 static int msix_setup_entries(struct pci_dev *dev, void __iomem *base,
696 struct msix_entry *entries, int nvec,
697 struct irq_affinity *affd)
699 struct irq_affinity_desc *curmsk, *masks = NULL;
700 struct msi_desc *entry;
702 int vec_count = pci_msix_vec_count(dev);
705 masks = irq_create_affinity_masks(nvec, affd);
707 for (i = 0, curmsk = masks; i < nvec; i++) {
708 entry = alloc_msi_entry(&dev->dev, 1, curmsk);
714 /* No enough memory. Don't try again */
719 entry->msi_attrib.is_msix = 1;
720 entry->msi_attrib.is_64 = 1;
722 entry->msi_attrib.entry_nr = entries[i].entry;
724 entry->msi_attrib.entry_nr = i;
726 entry->msi_attrib.is_virtual =
727 entry->msi_attrib.entry_nr >= vec_count;
729 entry->msi_attrib.default_irq = dev->irq;
730 entry->mask_base = base;
732 list_add_tail(&entry->list, dev_to_msi_list(&dev->dev));
742 static void msix_program_entries(struct pci_dev *dev,
743 struct msix_entry *entries)
745 struct msi_desc *entry;
747 void __iomem *desc_addr;
749 for_each_pci_msi_entry(entry, dev) {
751 entries[i++].vector = entry->irq;
753 desc_addr = pci_msix_desc_addr(entry);
755 entry->masked = readl(desc_addr +
756 PCI_MSIX_ENTRY_VECTOR_CTRL);
760 msix_mask_irq(entry, 1);
765 * msix_capability_init - configure device's MSI-X capability
766 * @dev: pointer to the pci_dev data structure of MSI-X device function
767 * @entries: pointer to an array of struct msix_entry entries
768 * @nvec: number of @entries
769 * @affd: Optional pointer to enable automatic affinity assignment
771 * Setup the MSI-X capability structure of device function with a
772 * single MSI-X IRQ. A return of zero indicates the successful setup of
773 * requested MSI-X entries with allocated IRQs or non-zero for otherwise.
775 static int msix_capability_init(struct pci_dev *dev, struct msix_entry *entries,
776 int nvec, struct irq_affinity *affd)
782 /* Ensure MSI-X is disabled while it is set up */
783 pci_msix_clear_and_set_ctrl(dev, PCI_MSIX_FLAGS_ENABLE, 0);
785 pci_read_config_word(dev, dev->msix_cap + PCI_MSIX_FLAGS, &control);
786 /* Request & Map MSI-X table region */
787 base = msix_map_region(dev, msix_table_size(control));
791 ret = msix_setup_entries(dev, base, entries, nvec, affd);
795 ret = pci_msi_setup_msi_irqs(dev, nvec, PCI_CAP_ID_MSIX);
799 /* Check if all MSI entries honor device restrictions */
800 ret = msi_verify_entries(dev);
805 * Some devices require MSI-X to be enabled before we can touch the
806 * MSI-X registers. We need to mask all the vectors to prevent
807 * interrupts coming in before they're fully set up.
809 pci_msix_clear_and_set_ctrl(dev, 0,
810 PCI_MSIX_FLAGS_MASKALL | PCI_MSIX_FLAGS_ENABLE);
812 msix_program_entries(dev, entries);
814 ret = populate_msi_sysfs(dev);
818 /* Set MSI-X enabled bits and unmask the function */
819 pci_intx_for_msi(dev, 0);
820 dev->msix_enabled = 1;
821 pci_msix_clear_and_set_ctrl(dev, PCI_MSIX_FLAGS_MASKALL, 0);
823 pcibios_free_irq(dev);
829 * If we had some success, report the number of IRQs
830 * we succeeded in setting up.
832 struct msi_desc *entry;
835 for_each_pci_msi_entry(entry, dev) {
850 * pci_msi_supported - check whether MSI may be enabled on a device
851 * @dev: pointer to the pci_dev data structure of MSI device function
852 * @nvec: how many MSIs have been requested?
854 * Look at global flags, the device itself, and its parent buses
855 * to determine if MSI/-X are supported for the device. If MSI/-X is
856 * supported return 1, else return 0.
858 static int pci_msi_supported(struct pci_dev *dev, int nvec)
862 /* MSI must be globally enabled and supported by the device */
866 if (!dev || dev->no_msi)
870 * You can't ask to have 0 or less MSIs configured.
872 * b) the list manipulation code assumes nvec >= 1.
878 * Any bridge which does NOT route MSI transactions from its
879 * secondary bus to its primary bus must set NO_MSI flag on
880 * the secondary pci_bus.
881 * We expect only arch-specific PCI host bus controller driver
882 * or quirks for specific PCI bridges to be setting NO_MSI.
884 for (bus = dev->bus; bus; bus = bus->parent)
885 if (bus->bus_flags & PCI_BUS_FLAGS_NO_MSI)
892 * pci_msi_vec_count - Return the number of MSI vectors a device can send
893 * @dev: device to report about
895 * This function returns the number of MSI vectors a device requested via
896 * Multiple Message Capable register. It returns a negative errno if the
897 * device is not capable sending MSI interrupts. Otherwise, the call succeeds
898 * and returns a power of two, up to a maximum of 2^5 (32), according to the
901 int pci_msi_vec_count(struct pci_dev *dev)
909 pci_read_config_word(dev, dev->msi_cap + PCI_MSI_FLAGS, &msgctl);
910 ret = 1 << ((msgctl & PCI_MSI_FLAGS_QMASK) >> 1);
914 EXPORT_SYMBOL(pci_msi_vec_count);
916 static void pci_msi_shutdown(struct pci_dev *dev)
918 struct msi_desc *desc;
921 if (!pci_msi_enable || !dev || !dev->msi_enabled)
924 BUG_ON(list_empty(dev_to_msi_list(&dev->dev)));
925 desc = first_pci_msi_entry(dev);
927 pci_msi_set_enable(dev, 0);
928 pci_intx_for_msi(dev, 1);
929 dev->msi_enabled = 0;
931 /* Return the device with MSI unmasked as initial states */
932 mask = msi_mask(desc->msi_attrib.multi_cap);
933 /* Keep cached state to be restored */
934 __pci_msi_desc_mask_irq(desc, mask, ~mask);
936 /* Restore dev->irq to its default pin-assertion IRQ */
937 dev->irq = desc->msi_attrib.default_irq;
938 pcibios_alloc_irq(dev);
941 void pci_disable_msi(struct pci_dev *dev)
943 if (!pci_msi_enable || !dev || !dev->msi_enabled)
946 pci_msi_shutdown(dev);
949 EXPORT_SYMBOL(pci_disable_msi);
952 * pci_msix_vec_count - return the number of device's MSI-X table entries
953 * @dev: pointer to the pci_dev data structure of MSI-X device function
954 * This function returns the number of device's MSI-X table entries and
955 * therefore the number of MSI-X vectors device is capable of sending.
956 * It returns a negative errno if the device is not capable of sending MSI-X
959 int pci_msix_vec_count(struct pci_dev *dev)
966 pci_read_config_word(dev, dev->msix_cap + PCI_MSIX_FLAGS, &control);
967 return msix_table_size(control);
969 EXPORT_SYMBOL(pci_msix_vec_count);
971 static int __pci_enable_msix(struct pci_dev *dev, struct msix_entry *entries,
972 int nvec, struct irq_affinity *affd, int flags)
977 if (!pci_msi_supported(dev, nvec) || dev->current_state != PCI_D0)
980 nr_entries = pci_msix_vec_count(dev);
983 if (nvec > nr_entries && !(flags & PCI_IRQ_VIRTUAL))
987 /* Check for any invalid entries */
988 for (i = 0; i < nvec; i++) {
989 if (entries[i].entry >= nr_entries)
990 return -EINVAL; /* invalid entry */
991 for (j = i + 1; j < nvec; j++) {
992 if (entries[i].entry == entries[j].entry)
993 return -EINVAL; /* duplicate entry */
998 /* Check whether driver already requested for MSI IRQ */
999 if (dev->msi_enabled) {
1000 pci_info(dev, "can't enable MSI-X (MSI IRQ already assigned)\n");
1003 return msix_capability_init(dev, entries, nvec, affd);
1006 static void pci_msix_shutdown(struct pci_dev *dev)
1008 struct msi_desc *entry;
1010 if (!pci_msi_enable || !dev || !dev->msix_enabled)
1013 if (pci_dev_is_disconnected(dev)) {
1014 dev->msix_enabled = 0;
1018 /* Return the device with MSI-X masked as initial states */
1019 for_each_pci_msi_entry(entry, dev) {
1020 /* Keep cached states to be restored */
1021 __pci_msix_desc_mask_irq(entry, 1);
1024 pci_msix_clear_and_set_ctrl(dev, PCI_MSIX_FLAGS_ENABLE, 0);
1025 pci_intx_for_msi(dev, 1);
1026 dev->msix_enabled = 0;
1027 pcibios_alloc_irq(dev);
1030 void pci_disable_msix(struct pci_dev *dev)
1032 if (!pci_msi_enable || !dev || !dev->msix_enabled)
1035 pci_msix_shutdown(dev);
1038 EXPORT_SYMBOL(pci_disable_msix);
1040 void pci_no_msi(void)
1046 * pci_msi_enabled - is MSI enabled?
1048 * Returns true if MSI has not been disabled by the command-line option
1051 int pci_msi_enabled(void)
1053 return pci_msi_enable;
1055 EXPORT_SYMBOL(pci_msi_enabled);
1057 static int __pci_enable_msi_range(struct pci_dev *dev, int minvec, int maxvec,
1058 struct irq_affinity *affd)
1063 if (!pci_msi_supported(dev, minvec) || dev->current_state != PCI_D0)
1066 /* Check whether driver already requested MSI-X IRQs */
1067 if (dev->msix_enabled) {
1068 pci_info(dev, "can't enable MSI (MSI-X already enabled)\n");
1072 if (maxvec < minvec)
1075 if (WARN_ON_ONCE(dev->msi_enabled))
1078 nvec = pci_msi_vec_count(dev);
1089 nvec = irq_calc_affinity_vectors(minvec, nvec, affd);
1094 rc = msi_capability_init(dev, nvec, affd);
1107 /* deprecated, don't use */
1108 int pci_enable_msi(struct pci_dev *dev)
1110 int rc = __pci_enable_msi_range(dev, 1, 1, NULL);
1115 EXPORT_SYMBOL(pci_enable_msi);
1117 static int __pci_enable_msix_range(struct pci_dev *dev,
1118 struct msix_entry *entries, int minvec,
1119 int maxvec, struct irq_affinity *affd,
1122 int rc, nvec = maxvec;
1124 if (maxvec < minvec)
1127 if (WARN_ON_ONCE(dev->msix_enabled))
1132 nvec = irq_calc_affinity_vectors(minvec, nvec, affd);
1137 rc = __pci_enable_msix(dev, entries, nvec, affd, flags);
1151 * pci_enable_msix_range - configure device's MSI-X capability structure
1152 * @dev: pointer to the pci_dev data structure of MSI-X device function
1153 * @entries: pointer to an array of MSI-X entries
1154 * @minvec: minimum number of MSI-X IRQs requested
1155 * @maxvec: maximum number of MSI-X IRQs requested
1157 * Setup the MSI-X capability structure of device function with a maximum
1158 * possible number of interrupts in the range between @minvec and @maxvec
1159 * upon its software driver call to request for MSI-X mode enabled on its
1160 * hardware device function. It returns a negative errno if an error occurs.
1161 * If it succeeds, it returns the actual number of interrupts allocated and
1162 * indicates the successful configuration of MSI-X capability structure
1163 * with new allocated MSI-X interrupts.
1165 int pci_enable_msix_range(struct pci_dev *dev, struct msix_entry *entries,
1166 int minvec, int maxvec)
1168 return __pci_enable_msix_range(dev, entries, minvec, maxvec, NULL, 0);
1170 EXPORT_SYMBOL(pci_enable_msix_range);
1173 * pci_alloc_irq_vectors_affinity - allocate multiple IRQs for a device
1174 * @dev: PCI device to operate on
1175 * @min_vecs: minimum number of vectors required (must be >= 1)
1176 * @max_vecs: maximum (desired) number of vectors
1177 * @flags: flags or quirks for the allocation
1178 * @affd: optional description of the affinity requirements
1180 * Allocate up to @max_vecs interrupt vectors for @dev, using MSI-X or MSI
1181 * vectors if available, and fall back to a single legacy vector
1182 * if neither is available. Return the number of vectors allocated,
1183 * (which might be smaller than @max_vecs) if successful, or a negative
1184 * error code on error. If less than @min_vecs interrupt vectors are
1185 * available for @dev the function will fail with -ENOSPC.
1187 * To get the Linux IRQ number used for a vector that can be passed to
1188 * request_irq() use the pci_irq_vector() helper.
1190 int pci_alloc_irq_vectors_affinity(struct pci_dev *dev, unsigned int min_vecs,
1191 unsigned int max_vecs, unsigned int flags,
1192 struct irq_affinity *affd)
1194 struct irq_affinity msi_default_affd = {0};
1195 int nvecs = -ENOSPC;
1197 if (flags & PCI_IRQ_AFFINITY) {
1199 affd = &msi_default_affd;
1205 if (flags & PCI_IRQ_MSIX) {
1206 nvecs = __pci_enable_msix_range(dev, NULL, min_vecs, max_vecs,
1212 if (flags & PCI_IRQ_MSI) {
1213 nvecs = __pci_enable_msi_range(dev, min_vecs, max_vecs, affd);
1218 /* use legacy IRQ if allowed */
1219 if (flags & PCI_IRQ_LEGACY) {
1220 if (min_vecs == 1 && dev->irq) {
1222 * Invoke the affinity spreading logic to ensure that
1223 * the device driver can adjust queue configuration
1224 * for the single interrupt case.
1227 irq_create_affinity_masks(1, affd);
1235 EXPORT_SYMBOL(pci_alloc_irq_vectors_affinity);
1238 * pci_free_irq_vectors - free previously allocated IRQs for a device
1239 * @dev: PCI device to operate on
1241 * Undoes the allocations and enabling in pci_alloc_irq_vectors().
1243 void pci_free_irq_vectors(struct pci_dev *dev)
1245 pci_disable_msix(dev);
1246 pci_disable_msi(dev);
1248 EXPORT_SYMBOL(pci_free_irq_vectors);
1251 * pci_irq_vector - return Linux IRQ number of a device vector
1252 * @dev: PCI device to operate on
1253 * @nr: device-relative interrupt vector index (0-based).
1255 int pci_irq_vector(struct pci_dev *dev, unsigned int nr)
1257 if (dev->msix_enabled) {
1258 struct msi_desc *entry;
1261 for_each_pci_msi_entry(entry, dev) {
1270 if (dev->msi_enabled) {
1271 struct msi_desc *entry = first_pci_msi_entry(dev);
1273 if (WARN_ON_ONCE(nr >= entry->nvec_used))
1276 if (WARN_ON_ONCE(nr > 0))
1280 return dev->irq + nr;
1282 EXPORT_SYMBOL(pci_irq_vector);
1285 * pci_irq_get_affinity - return the affinity of a particular MSI vector
1286 * @dev: PCI device to operate on
1287 * @nr: device-relative interrupt vector index (0-based).
1289 const struct cpumask *pci_irq_get_affinity(struct pci_dev *dev, int nr)
1291 if (dev->msix_enabled) {
1292 struct msi_desc *entry;
1295 for_each_pci_msi_entry(entry, dev) {
1297 return &entry->affinity->mask;
1302 } else if (dev->msi_enabled) {
1303 struct msi_desc *entry = first_pci_msi_entry(dev);
1305 if (WARN_ON_ONCE(!entry || !entry->affinity ||
1306 nr >= entry->nvec_used))
1309 return &entry->affinity[nr].mask;
1311 return cpu_possible_mask;
1314 EXPORT_SYMBOL(pci_irq_get_affinity);
1316 struct pci_dev *msi_desc_to_pci_dev(struct msi_desc *desc)
1318 return to_pci_dev(desc->dev);
1320 EXPORT_SYMBOL(msi_desc_to_pci_dev);
1322 void *msi_desc_to_pci_sysdata(struct msi_desc *desc)
1324 struct pci_dev *dev = msi_desc_to_pci_dev(desc);
1326 return dev->bus->sysdata;
1328 EXPORT_SYMBOL_GPL(msi_desc_to_pci_sysdata);
1330 #ifdef CONFIG_PCI_MSI_IRQ_DOMAIN
1332 * pci_msi_domain_write_msg - Helper to write MSI message to PCI config space
1333 * @irq_data: Pointer to interrupt data of the MSI interrupt
1334 * @msg: Pointer to the message
1336 void pci_msi_domain_write_msg(struct irq_data *irq_data, struct msi_msg *msg)
1338 struct msi_desc *desc = irq_data_get_msi_desc(irq_data);
1341 * For MSI-X desc->irq is always equal to irq_data->irq. For
1342 * MSI only the first interrupt of MULTI MSI passes the test.
1344 if (desc->irq == irq_data->irq)
1345 __pci_write_msi_msg(desc, msg);
1349 * pci_msi_domain_calc_hwirq - Generate a unique ID for an MSI source
1350 * @desc: Pointer to the MSI descriptor
1352 * The ID number is only used within the irqdomain.
1354 static irq_hw_number_t pci_msi_domain_calc_hwirq(struct msi_desc *desc)
1356 struct pci_dev *dev = msi_desc_to_pci_dev(desc);
1358 return (irq_hw_number_t)desc->msi_attrib.entry_nr |
1359 pci_dev_id(dev) << 11 |
1360 (pci_domain_nr(dev->bus) & 0xFFFFFFFF) << 27;
1363 static inline bool pci_msi_desc_is_multi_msi(struct msi_desc *desc)
1365 return !desc->msi_attrib.is_msix && desc->nvec_used > 1;
1369 * pci_msi_domain_check_cap - Verify that @domain supports the capabilities
1371 * @domain: The interrupt domain to check
1372 * @info: The domain info for verification
1373 * @dev: The device to check
1376 * 0 if the functionality is supported
1377 * 1 if Multi MSI is requested, but the domain does not support it
1378 * -ENOTSUPP otherwise
1380 int pci_msi_domain_check_cap(struct irq_domain *domain,
1381 struct msi_domain_info *info, struct device *dev)
1383 struct msi_desc *desc = first_pci_msi_entry(to_pci_dev(dev));
1385 /* Special handling to support __pci_enable_msi_range() */
1386 if (pci_msi_desc_is_multi_msi(desc) &&
1387 !(info->flags & MSI_FLAG_MULTI_PCI_MSI))
1389 else if (desc->msi_attrib.is_msix && !(info->flags & MSI_FLAG_PCI_MSIX))
1395 static int pci_msi_domain_handle_error(struct irq_domain *domain,
1396 struct msi_desc *desc, int error)
1398 /* Special handling to support __pci_enable_msi_range() */
1399 if (pci_msi_desc_is_multi_msi(desc) && error == -ENOSPC)
1405 static void pci_msi_domain_set_desc(msi_alloc_info_t *arg,
1406 struct msi_desc *desc)
1409 arg->hwirq = pci_msi_domain_calc_hwirq(desc);
1412 static struct msi_domain_ops pci_msi_domain_ops_default = {
1413 .set_desc = pci_msi_domain_set_desc,
1414 .msi_check = pci_msi_domain_check_cap,
1415 .handle_error = pci_msi_domain_handle_error,
1418 static void pci_msi_domain_update_dom_ops(struct msi_domain_info *info)
1420 struct msi_domain_ops *ops = info->ops;
1423 info->ops = &pci_msi_domain_ops_default;
1425 if (ops->set_desc == NULL)
1426 ops->set_desc = pci_msi_domain_set_desc;
1427 if (ops->msi_check == NULL)
1428 ops->msi_check = pci_msi_domain_check_cap;
1429 if (ops->handle_error == NULL)
1430 ops->handle_error = pci_msi_domain_handle_error;
1434 static void pci_msi_domain_update_chip_ops(struct msi_domain_info *info)
1436 struct irq_chip *chip = info->chip;
1439 if (!chip->irq_write_msi_msg)
1440 chip->irq_write_msi_msg = pci_msi_domain_write_msg;
1441 if (!chip->irq_mask)
1442 chip->irq_mask = pci_msi_mask_irq;
1443 if (!chip->irq_unmask)
1444 chip->irq_unmask = pci_msi_unmask_irq;
1448 * pci_msi_create_irq_domain - Create a MSI interrupt domain
1449 * @fwnode: Optional fwnode of the interrupt controller
1450 * @info: MSI domain info
1451 * @parent: Parent irq domain
1453 * Updates the domain and chip ops and creates a MSI interrupt domain.
1456 * A domain pointer or NULL in case of failure.
1458 struct irq_domain *pci_msi_create_irq_domain(struct fwnode_handle *fwnode,
1459 struct msi_domain_info *info,
1460 struct irq_domain *parent)
1462 struct irq_domain *domain;
1464 if (WARN_ON(info->flags & MSI_FLAG_LEVEL_CAPABLE))
1465 info->flags &= ~MSI_FLAG_LEVEL_CAPABLE;
1467 if (info->flags & MSI_FLAG_USE_DEF_DOM_OPS)
1468 pci_msi_domain_update_dom_ops(info);
1469 if (info->flags & MSI_FLAG_USE_DEF_CHIP_OPS)
1470 pci_msi_domain_update_chip_ops(info);
1472 info->flags |= MSI_FLAG_ACTIVATE_EARLY;
1473 if (IS_ENABLED(CONFIG_GENERIC_IRQ_RESERVATION_MODE))
1474 info->flags |= MSI_FLAG_MUST_REACTIVATE;
1476 /* PCI-MSI is oneshot-safe */
1477 info->chip->flags |= IRQCHIP_ONESHOT_SAFE;
1479 domain = msi_create_irq_domain(fwnode, info, parent);
1483 irq_domain_update_bus_token(domain, DOMAIN_BUS_PCI_MSI);
1486 EXPORT_SYMBOL_GPL(pci_msi_create_irq_domain);
1489 * Users of the generic MSI infrastructure expect a device to have a single ID,
1490 * so with DMA aliases we have to pick the least-worst compromise. Devices with
1491 * DMA phantom functions tend to still emit MSIs from the real function number,
1492 * so we ignore those and only consider topological aliases where either the
1493 * alias device or RID appears on a different bus number. We also make the
1494 * reasonable assumption that bridges are walked in an upstream direction (so
1495 * the last one seen wins), and the much braver assumption that the most likely
1496 * case is that of PCI->PCIe so we should always use the alias RID. This echoes
1497 * the logic from intel_irq_remapping's set_msi_sid(), which presumably works
1498 * well enough in practice; in the face of the horrible PCIe<->PCI-X conditions
1499 * for taking ownership all we can really do is close our eyes and hope...
1501 static int get_msi_id_cb(struct pci_dev *pdev, u16 alias, void *data)
1504 u8 bus = PCI_BUS_NUM(*pa);
1506 if (pdev->bus->number != bus || PCI_BUS_NUM(alias) != bus)
1513 * pci_msi_domain_get_msi_rid - Get the MSI requester id (RID)
1514 * @domain: The interrupt domain
1515 * @pdev: The PCI device.
1517 * The RID for a device is formed from the alias, with a firmware
1518 * supplied mapping applied
1522 u32 pci_msi_domain_get_msi_rid(struct irq_domain *domain, struct pci_dev *pdev)
1524 struct device_node *of_node;
1525 u32 rid = pci_dev_id(pdev);
1527 pci_for_each_dma_alias(pdev, get_msi_id_cb, &rid);
1529 of_node = irq_domain_get_of_node(domain);
1530 rid = of_node ? of_msi_map_id(&pdev->dev, of_node, rid) :
1531 iort_msi_map_id(&pdev->dev, rid);
1537 * pci_msi_get_device_domain - Get the MSI domain for a given PCI device
1538 * @pdev: The PCI device
1540 * Use the firmware data to find a device-specific MSI domain
1541 * (i.e. not one that is set as a default).
1543 * Returns: The corresponding MSI domain or NULL if none has been found.
1545 struct irq_domain *pci_msi_get_device_domain(struct pci_dev *pdev)
1547 struct irq_domain *dom;
1548 u32 rid = pci_dev_id(pdev);
1550 pci_for_each_dma_alias(pdev, get_msi_id_cb, &rid);
1551 dom = of_msi_map_get_device_domain(&pdev->dev, rid, DOMAIN_BUS_PCI_MSI);
1553 dom = iort_get_device_domain(&pdev->dev, rid,
1554 DOMAIN_BUS_PCI_MSI);
1559 * pci_dev_has_special_msi_domain - Check whether the device is handled by
1560 * a non-standard PCI-MSI domain
1561 * @pdev: The PCI device to check.
1563 * Returns: True if the device irqdomain or the bus irqdomain is
1564 * non-standard PCI/MSI.
1566 bool pci_dev_has_special_msi_domain(struct pci_dev *pdev)
1568 struct irq_domain *dom = dev_get_msi_domain(&pdev->dev);
1571 dom = dev_get_msi_domain(&pdev->bus->dev);
1576 return dom->bus_token != DOMAIN_BUS_PCI_MSI;
1579 #endif /* CONFIG_PCI_MSI_IRQ_DOMAIN */