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 void __pci_msi_desc_mask_irq(struct msi_desc *desc, u32 mask, u32 flag)
176 raw_spinlock_t *lock = &desc->dev->msi_lock;
179 if (pci_msi_ignore_mask || !desc->msi_attrib.maskbit)
182 raw_spin_lock_irqsave(lock, flags);
183 desc->masked &= ~mask;
184 desc->masked |= flag;
185 pci_write_config_dword(msi_desc_to_pci_dev(desc), desc->mask_pos,
187 raw_spin_unlock_irqrestore(lock, flags);
190 static void msi_mask_irq(struct msi_desc *desc, u32 mask, u32 flag)
192 __pci_msi_desc_mask_irq(desc, mask, flag);
195 static void __iomem *pci_msix_desc_addr(struct msi_desc *desc)
197 if (desc->msi_attrib.is_virtual)
200 return desc->mask_base +
201 desc->msi_attrib.entry_nr * PCI_MSIX_ENTRY_SIZE;
205 * This internal function does not flush PCI writes to the device.
206 * All users must ensure that they read from the device before either
207 * assuming that the device state is up to date, or returning out of this
208 * file. This saves a few milliseconds when initialising devices with lots
209 * of MSI-X interrupts.
211 u32 __pci_msix_desc_mask_irq(struct msi_desc *desc, u32 flag)
213 u32 mask_bits = desc->masked;
214 void __iomem *desc_addr;
216 if (pci_msi_ignore_mask)
219 desc_addr = pci_msix_desc_addr(desc);
223 mask_bits &= ~PCI_MSIX_ENTRY_CTRL_MASKBIT;
224 if (flag & PCI_MSIX_ENTRY_CTRL_MASKBIT)
225 mask_bits |= PCI_MSIX_ENTRY_CTRL_MASKBIT;
227 writel(mask_bits, desc_addr + PCI_MSIX_ENTRY_VECTOR_CTRL);
232 static void msix_mask_irq(struct msi_desc *desc, u32 flag)
234 desc->masked = __pci_msix_desc_mask_irq(desc, flag);
237 static void msi_set_mask_bit(struct irq_data *data, u32 flag)
239 struct msi_desc *desc = irq_data_get_msi_desc(data);
241 if (desc->msi_attrib.is_msix) {
242 msix_mask_irq(desc, flag);
243 readl(desc->mask_base); /* Flush write to device */
245 unsigned offset = data->irq - desc->irq;
246 msi_mask_irq(desc, 1 << offset, flag << offset);
251 * pci_msi_mask_irq - Generic IRQ chip callback to mask PCI/MSI interrupts
252 * @data: pointer to irqdata associated to that interrupt
254 void pci_msi_mask_irq(struct irq_data *data)
256 msi_set_mask_bit(data, 1);
258 EXPORT_SYMBOL_GPL(pci_msi_mask_irq);
261 * pci_msi_unmask_irq - Generic IRQ chip callback to unmask PCI/MSI interrupts
262 * @data: pointer to irqdata associated to that interrupt
264 void pci_msi_unmask_irq(struct irq_data *data)
266 msi_set_mask_bit(data, 0);
268 EXPORT_SYMBOL_GPL(pci_msi_unmask_irq);
270 void default_restore_msi_irqs(struct pci_dev *dev)
272 struct msi_desc *entry;
274 for_each_pci_msi_entry(entry, dev)
275 default_restore_msi_irq(dev, entry->irq);
278 void __pci_read_msi_msg(struct msi_desc *entry, struct msi_msg *msg)
280 struct pci_dev *dev = msi_desc_to_pci_dev(entry);
282 BUG_ON(dev->current_state != PCI_D0);
284 if (entry->msi_attrib.is_msix) {
285 void __iomem *base = pci_msix_desc_addr(entry);
292 msg->address_lo = readl(base + PCI_MSIX_ENTRY_LOWER_ADDR);
293 msg->address_hi = readl(base + PCI_MSIX_ENTRY_UPPER_ADDR);
294 msg->data = readl(base + PCI_MSIX_ENTRY_DATA);
296 int pos = dev->msi_cap;
299 pci_read_config_dword(dev, pos + PCI_MSI_ADDRESS_LO,
301 if (entry->msi_attrib.is_64) {
302 pci_read_config_dword(dev, pos + PCI_MSI_ADDRESS_HI,
304 pci_read_config_word(dev, pos + PCI_MSI_DATA_64, &data);
307 pci_read_config_word(dev, pos + PCI_MSI_DATA_32, &data);
313 void __pci_write_msi_msg(struct msi_desc *entry, struct msi_msg *msg)
315 struct pci_dev *dev = msi_desc_to_pci_dev(entry);
317 if (dev->current_state != PCI_D0 || pci_dev_is_disconnected(dev)) {
318 /* Don't touch the hardware now */
319 } else if (entry->msi_attrib.is_msix) {
320 void __iomem *base = pci_msix_desc_addr(entry);
321 bool unmasked = !(entry->masked & PCI_MSIX_ENTRY_CTRL_MASKBIT);
327 * The specification mandates that the entry is masked
328 * when the message is modified:
330 * "If software changes the Address or Data value of an
331 * entry while the entry is unmasked, the result is
335 __pci_msix_desc_mask_irq(entry, PCI_MSIX_ENTRY_CTRL_MASKBIT);
337 writel(msg->address_lo, base + PCI_MSIX_ENTRY_LOWER_ADDR);
338 writel(msg->address_hi, base + PCI_MSIX_ENTRY_UPPER_ADDR);
339 writel(msg->data, base + PCI_MSIX_ENTRY_DATA);
342 __pci_msix_desc_mask_irq(entry, 0);
344 /* Ensure that the writes are visible in the device */
345 readl(base + PCI_MSIX_ENTRY_DATA);
347 int pos = dev->msi_cap;
350 pci_read_config_word(dev, pos + PCI_MSI_FLAGS, &msgctl);
351 msgctl &= ~PCI_MSI_FLAGS_QSIZE;
352 msgctl |= entry->msi_attrib.multiple << 4;
353 pci_write_config_word(dev, pos + PCI_MSI_FLAGS, msgctl);
355 pci_write_config_dword(dev, pos + PCI_MSI_ADDRESS_LO,
357 if (entry->msi_attrib.is_64) {
358 pci_write_config_dword(dev, pos + PCI_MSI_ADDRESS_HI,
360 pci_write_config_word(dev, pos + PCI_MSI_DATA_64,
363 pci_write_config_word(dev, pos + PCI_MSI_DATA_32,
366 /* Ensure that the writes are visible in the device */
367 pci_read_config_word(dev, pos + PCI_MSI_FLAGS, &msgctl);
373 if (entry->write_msi_msg)
374 entry->write_msi_msg(entry, entry->write_msi_msg_data);
378 void pci_write_msi_msg(unsigned int irq, struct msi_msg *msg)
380 struct msi_desc *entry = irq_get_msi_desc(irq);
382 __pci_write_msi_msg(entry, msg);
384 EXPORT_SYMBOL_GPL(pci_write_msi_msg);
386 static void free_msi_irqs(struct pci_dev *dev)
388 struct list_head *msi_list = dev_to_msi_list(&dev->dev);
389 struct msi_desc *entry, *tmp;
390 struct attribute **msi_attrs;
391 struct device_attribute *dev_attr;
394 for_each_pci_msi_entry(entry, dev)
396 for (i = 0; i < entry->nvec_used; i++)
397 BUG_ON(irq_has_action(entry->irq + i));
399 if (dev->msi_irq_groups) {
400 sysfs_remove_groups(&dev->dev.kobj, dev->msi_irq_groups);
401 msi_attrs = dev->msi_irq_groups[0]->attrs;
402 while (msi_attrs[count]) {
403 dev_attr = container_of(msi_attrs[count],
404 struct device_attribute, attr);
405 kfree(dev_attr->attr.name);
410 kfree(dev->msi_irq_groups[0]);
411 kfree(dev->msi_irq_groups);
412 dev->msi_irq_groups = NULL;
415 pci_msi_teardown_msi_irqs(dev);
417 list_for_each_entry_safe(entry, tmp, msi_list, list) {
418 if (entry->msi_attrib.is_msix) {
419 if (list_is_last(&entry->list, msi_list))
420 iounmap(entry->mask_base);
423 list_del(&entry->list);
424 free_msi_entry(entry);
428 static void pci_intx_for_msi(struct pci_dev *dev, int enable)
430 if (!(dev->dev_flags & PCI_DEV_FLAGS_MSI_INTX_DISABLE_BUG))
431 pci_intx(dev, enable);
434 static void __pci_restore_msi_state(struct pci_dev *dev)
437 struct msi_desc *entry;
439 if (!dev->msi_enabled)
442 entry = irq_get_msi_desc(dev->irq);
444 pci_intx_for_msi(dev, 0);
445 pci_msi_set_enable(dev, 0);
446 arch_restore_msi_irqs(dev);
448 pci_read_config_word(dev, dev->msi_cap + PCI_MSI_FLAGS, &control);
449 msi_mask_irq(entry, msi_mask(entry->msi_attrib.multi_cap),
451 control &= ~PCI_MSI_FLAGS_QSIZE;
452 control |= (entry->msi_attrib.multiple << 4) | PCI_MSI_FLAGS_ENABLE;
453 pci_write_config_word(dev, dev->msi_cap + PCI_MSI_FLAGS, control);
456 static void __pci_restore_msix_state(struct pci_dev *dev)
458 struct msi_desc *entry;
460 if (!dev->msix_enabled)
462 BUG_ON(list_empty(dev_to_msi_list(&dev->dev)));
464 /* route the table */
465 pci_intx_for_msi(dev, 0);
466 pci_msix_clear_and_set_ctrl(dev, 0,
467 PCI_MSIX_FLAGS_ENABLE | PCI_MSIX_FLAGS_MASKALL);
469 arch_restore_msi_irqs(dev);
470 for_each_pci_msi_entry(entry, dev)
471 msix_mask_irq(entry, entry->masked);
473 pci_msix_clear_and_set_ctrl(dev, PCI_MSIX_FLAGS_MASKALL, 0);
476 void pci_restore_msi_state(struct pci_dev *dev)
478 __pci_restore_msi_state(dev);
479 __pci_restore_msix_state(dev);
481 EXPORT_SYMBOL_GPL(pci_restore_msi_state);
483 static ssize_t msi_mode_show(struct device *dev, struct device_attribute *attr,
486 struct msi_desc *entry;
490 retval = kstrtoul(attr->attr.name, 10, &irq);
494 entry = irq_get_msi_desc(irq);
496 return sprintf(buf, "%s\n",
497 entry->msi_attrib.is_msix ? "msix" : "msi");
502 static int populate_msi_sysfs(struct pci_dev *pdev)
504 struct attribute **msi_attrs;
505 struct attribute *msi_attr;
506 struct device_attribute *msi_dev_attr;
507 struct attribute_group *msi_irq_group;
508 const struct attribute_group **msi_irq_groups;
509 struct msi_desc *entry;
515 /* Determine how many msi entries we have */
516 for_each_pci_msi_entry(entry, pdev)
517 num_msi += entry->nvec_used;
521 /* Dynamically create the MSI attributes for the PCI device */
522 msi_attrs = kcalloc(num_msi + 1, sizeof(void *), GFP_KERNEL);
525 for_each_pci_msi_entry(entry, pdev) {
526 for (i = 0; i < entry->nvec_used; i++) {
527 msi_dev_attr = kzalloc(sizeof(*msi_dev_attr), GFP_KERNEL);
530 msi_attrs[count] = &msi_dev_attr->attr;
532 sysfs_attr_init(&msi_dev_attr->attr);
533 msi_dev_attr->attr.name = kasprintf(GFP_KERNEL, "%d",
535 if (!msi_dev_attr->attr.name)
537 msi_dev_attr->attr.mode = S_IRUGO;
538 msi_dev_attr->show = msi_mode_show;
543 msi_irq_group = kzalloc(sizeof(*msi_irq_group), GFP_KERNEL);
546 msi_irq_group->name = "msi_irqs";
547 msi_irq_group->attrs = msi_attrs;
549 msi_irq_groups = kcalloc(2, sizeof(void *), GFP_KERNEL);
551 goto error_irq_group;
552 msi_irq_groups[0] = msi_irq_group;
554 ret = sysfs_create_groups(&pdev->dev.kobj, msi_irq_groups);
556 goto error_irq_groups;
557 pdev->msi_irq_groups = msi_irq_groups;
562 kfree(msi_irq_groups);
564 kfree(msi_irq_group);
567 msi_attr = msi_attrs[count];
569 msi_dev_attr = container_of(msi_attr, struct device_attribute, attr);
570 kfree(msi_attr->name);
573 msi_attr = msi_attrs[count];
579 static struct msi_desc *
580 msi_setup_entry(struct pci_dev *dev, int nvec, struct irq_affinity *affd)
582 struct irq_affinity_desc *masks = NULL;
583 struct msi_desc *entry;
587 masks = irq_create_affinity_masks(nvec, affd);
589 /* MSI Entry Initialization */
590 entry = alloc_msi_entry(&dev->dev, nvec, masks);
594 pci_read_config_word(dev, dev->msi_cap + PCI_MSI_FLAGS, &control);
595 /* Lies, damned lies, and MSIs */
596 if (dev->dev_flags & PCI_DEV_FLAGS_HAS_MSI_MASKING)
597 control |= PCI_MSI_FLAGS_MASKBIT;
599 entry->msi_attrib.is_msix = 0;
600 entry->msi_attrib.is_64 = !!(control & PCI_MSI_FLAGS_64BIT);
601 entry->msi_attrib.is_virtual = 0;
602 entry->msi_attrib.entry_nr = 0;
603 entry->msi_attrib.maskbit = !!(control & PCI_MSI_FLAGS_MASKBIT);
604 entry->msi_attrib.default_irq = dev->irq; /* Save IOAPIC IRQ */
605 entry->msi_attrib.multi_cap = (control & PCI_MSI_FLAGS_QMASK) >> 1;
606 entry->msi_attrib.multiple = ilog2(__roundup_pow_of_two(nvec));
608 if (control & PCI_MSI_FLAGS_64BIT)
609 entry->mask_pos = dev->msi_cap + PCI_MSI_MASK_64;
611 entry->mask_pos = dev->msi_cap + PCI_MSI_MASK_32;
613 /* Save the initial mask status */
614 if (entry->msi_attrib.maskbit)
615 pci_read_config_dword(dev, entry->mask_pos, &entry->masked);
622 static int msi_verify_entries(struct pci_dev *dev)
624 struct msi_desc *entry;
626 for_each_pci_msi_entry(entry, dev) {
627 if (!dev->no_64bit_msi || !entry->msg.address_hi)
629 pci_err(dev, "Device has broken 64-bit MSI but arch"
630 " tried to assign one above 4G\n");
637 * msi_capability_init - configure device's MSI capability structure
638 * @dev: pointer to the pci_dev data structure of MSI device function
639 * @nvec: number of interrupts to allocate
640 * @affd: description of automatic IRQ affinity assignments (may be %NULL)
642 * Setup the MSI capability structure of the device with the requested
643 * number of interrupts. A return value of zero indicates the successful
644 * setup of an entry with the new MSI IRQ. A negative return value indicates
645 * an error, and a positive return value indicates the number of interrupts
646 * which could have been allocated.
648 static int msi_capability_init(struct pci_dev *dev, int nvec,
649 struct irq_affinity *affd)
651 struct msi_desc *entry;
655 pci_msi_set_enable(dev, 0); /* Disable MSI during set up */
657 entry = msi_setup_entry(dev, nvec, affd);
661 /* All MSIs are unmasked by default; mask them all */
662 mask = msi_mask(entry->msi_attrib.multi_cap);
663 msi_mask_irq(entry, mask, mask);
665 list_add_tail(&entry->list, dev_to_msi_list(&dev->dev));
667 /* Configure MSI capability structure */
668 ret = pci_msi_setup_msi_irqs(dev, nvec, PCI_CAP_ID_MSI);
670 msi_mask_irq(entry, mask, 0);
675 ret = msi_verify_entries(dev);
677 msi_mask_irq(entry, mask, 0);
682 ret = populate_msi_sysfs(dev);
684 msi_mask_irq(entry, mask, 0);
689 /* Set MSI enabled bits */
690 pci_intx_for_msi(dev, 0);
691 pci_msi_set_enable(dev, 1);
692 dev->msi_enabled = 1;
694 pcibios_free_irq(dev);
695 dev->irq = entry->irq;
699 static void __iomem *msix_map_region(struct pci_dev *dev, unsigned nr_entries)
701 resource_size_t phys_addr;
706 pci_read_config_dword(dev, dev->msix_cap + PCI_MSIX_TABLE,
708 bir = (u8)(table_offset & PCI_MSIX_TABLE_BIR);
709 flags = pci_resource_flags(dev, bir);
710 if (!flags || (flags & IORESOURCE_UNSET))
713 table_offset &= PCI_MSIX_TABLE_OFFSET;
714 phys_addr = pci_resource_start(dev, bir) + table_offset;
716 return ioremap(phys_addr, nr_entries * PCI_MSIX_ENTRY_SIZE);
719 static int msix_setup_entries(struct pci_dev *dev, void __iomem *base,
720 struct msix_entry *entries, int nvec,
721 struct irq_affinity *affd)
723 struct irq_affinity_desc *curmsk, *masks = NULL;
724 struct msi_desc *entry;
727 int vec_count = pci_msix_vec_count(dev);
730 masks = irq_create_affinity_masks(nvec, affd);
732 for (i = 0, curmsk = masks; i < nvec; i++) {
733 entry = alloc_msi_entry(&dev->dev, 1, curmsk);
739 /* No enough memory. Don't try again */
744 entry->msi_attrib.is_msix = 1;
745 entry->msi_attrib.is_64 = 1;
748 entry->msi_attrib.entry_nr = entries[i].entry;
750 entry->msi_attrib.entry_nr = i;
752 entry->msi_attrib.is_virtual =
753 entry->msi_attrib.entry_nr >= vec_count;
755 entry->msi_attrib.default_irq = dev->irq;
756 entry->mask_base = base;
758 addr = pci_msix_desc_addr(entry);
760 entry->masked = readl(addr + PCI_MSIX_ENTRY_VECTOR_CTRL);
762 list_add_tail(&entry->list, dev_to_msi_list(&dev->dev));
772 static void msix_update_entries(struct pci_dev *dev, struct msix_entry *entries)
774 struct msi_desc *entry;
776 for_each_pci_msi_entry(entry, dev) {
778 entries->vector = entry->irq;
784 static void msix_mask_all(void __iomem *base, int tsize)
786 u32 ctrl = PCI_MSIX_ENTRY_CTRL_MASKBIT;
789 if (pci_msi_ignore_mask)
792 for (i = 0; i < tsize; i++, base += PCI_MSIX_ENTRY_SIZE)
793 writel(ctrl, base + PCI_MSIX_ENTRY_VECTOR_CTRL);
797 * msix_capability_init - configure device's MSI-X capability
798 * @dev: pointer to the pci_dev data structure of MSI-X device function
799 * @entries: pointer to an array of struct msix_entry entries
800 * @nvec: number of @entries
801 * @affd: Optional pointer to enable automatic affinity assignment
803 * Setup the MSI-X capability structure of device function with a
804 * single MSI-X IRQ. A return of zero indicates the successful setup of
805 * requested MSI-X entries with allocated IRQs or non-zero for otherwise.
807 static int msix_capability_init(struct pci_dev *dev, struct msix_entry *entries,
808 int nvec, struct irq_affinity *affd)
815 * Some devices require MSI-X to be enabled before the MSI-X
816 * registers can be accessed. Mask all the vectors to prevent
817 * interrupts coming in before they're fully set up.
819 pci_msix_clear_and_set_ctrl(dev, 0, PCI_MSIX_FLAGS_MASKALL |
820 PCI_MSIX_FLAGS_ENABLE);
822 pci_read_config_word(dev, dev->msix_cap + PCI_MSIX_FLAGS, &control);
823 /* Request & Map MSI-X table region */
824 tsize = msix_table_size(control);
825 base = msix_map_region(dev, tsize);
831 ret = msix_setup_entries(dev, base, entries, nvec, affd);
835 ret = pci_msi_setup_msi_irqs(dev, nvec, PCI_CAP_ID_MSIX);
839 /* Check if all MSI entries honor device restrictions */
840 ret = msi_verify_entries(dev);
844 msix_update_entries(dev, entries);
846 ret = populate_msi_sysfs(dev);
850 /* Set MSI-X enabled bits and unmask the function */
851 pci_intx_for_msi(dev, 0);
852 dev->msix_enabled = 1;
855 * Ensure that all table entries are masked to prevent
856 * stale entries from firing in a crash kernel.
858 * Done late to deal with a broken Marvell NVME device
859 * which takes the MSI-X mask bits into account even
860 * when MSI-X is disabled, which prevents MSI delivery.
862 msix_mask_all(base, tsize);
863 pci_msix_clear_and_set_ctrl(dev, PCI_MSIX_FLAGS_MASKALL, 0);
865 pcibios_free_irq(dev);
871 * If we had some success, report the number of IRQs
872 * we succeeded in setting up.
874 struct msi_desc *entry;
877 for_each_pci_msi_entry(entry, dev) {
889 pci_msix_clear_and_set_ctrl(dev, PCI_MSIX_FLAGS_MASKALL | PCI_MSIX_FLAGS_ENABLE, 0);
895 * pci_msi_supported - check whether MSI may be enabled on a device
896 * @dev: pointer to the pci_dev data structure of MSI device function
897 * @nvec: how many MSIs have been requested?
899 * Look at global flags, the device itself, and its parent buses
900 * to determine if MSI/-X are supported for the device. If MSI/-X is
901 * supported return 1, else return 0.
903 static int pci_msi_supported(struct pci_dev *dev, int nvec)
907 /* MSI must be globally enabled and supported by the device */
911 if (!dev || dev->no_msi)
915 * You can't ask to have 0 or less MSIs configured.
917 * b) the list manipulation code assumes nvec >= 1.
923 * Any bridge which does NOT route MSI transactions from its
924 * secondary bus to its primary bus must set NO_MSI flag on
925 * the secondary pci_bus.
926 * We expect only arch-specific PCI host bus controller driver
927 * or quirks for specific PCI bridges to be setting NO_MSI.
929 for (bus = dev->bus; bus; bus = bus->parent)
930 if (bus->bus_flags & PCI_BUS_FLAGS_NO_MSI)
937 * pci_msi_vec_count - Return the number of MSI vectors a device can send
938 * @dev: device to report about
940 * This function returns the number of MSI vectors a device requested via
941 * Multiple Message Capable register. It returns a negative errno if the
942 * device is not capable sending MSI interrupts. Otherwise, the call succeeds
943 * and returns a power of two, up to a maximum of 2^5 (32), according to the
946 int pci_msi_vec_count(struct pci_dev *dev)
954 pci_read_config_word(dev, dev->msi_cap + PCI_MSI_FLAGS, &msgctl);
955 ret = 1 << ((msgctl & PCI_MSI_FLAGS_QMASK) >> 1);
959 EXPORT_SYMBOL(pci_msi_vec_count);
961 static void pci_msi_shutdown(struct pci_dev *dev)
963 struct msi_desc *desc;
966 if (!pci_msi_enable || !dev || !dev->msi_enabled)
969 BUG_ON(list_empty(dev_to_msi_list(&dev->dev)));
970 desc = first_pci_msi_entry(dev);
972 pci_msi_set_enable(dev, 0);
973 pci_intx_for_msi(dev, 1);
974 dev->msi_enabled = 0;
976 /* Return the device with MSI unmasked as initial states */
977 mask = msi_mask(desc->msi_attrib.multi_cap);
978 msi_mask_irq(desc, mask, 0);
980 /* Restore dev->irq to its default pin-assertion IRQ */
981 dev->irq = desc->msi_attrib.default_irq;
982 pcibios_alloc_irq(dev);
985 void pci_disable_msi(struct pci_dev *dev)
987 if (!pci_msi_enable || !dev || !dev->msi_enabled)
990 pci_msi_shutdown(dev);
993 EXPORT_SYMBOL(pci_disable_msi);
996 * pci_msix_vec_count - return the number of device's MSI-X table entries
997 * @dev: pointer to the pci_dev data structure of MSI-X device function
998 * This function returns the number of device's MSI-X table entries and
999 * therefore the number of MSI-X vectors device is capable of sending.
1000 * It returns a negative errno if the device is not capable of sending MSI-X
1003 int pci_msix_vec_count(struct pci_dev *dev)
1010 pci_read_config_word(dev, dev->msix_cap + PCI_MSIX_FLAGS, &control);
1011 return msix_table_size(control);
1013 EXPORT_SYMBOL(pci_msix_vec_count);
1015 static int __pci_enable_msix(struct pci_dev *dev, struct msix_entry *entries,
1016 int nvec, struct irq_affinity *affd, int flags)
1021 if (!pci_msi_supported(dev, nvec) || dev->current_state != PCI_D0)
1024 nr_entries = pci_msix_vec_count(dev);
1027 if (nvec > nr_entries && !(flags & PCI_IRQ_VIRTUAL))
1031 /* Check for any invalid entries */
1032 for (i = 0; i < nvec; i++) {
1033 if (entries[i].entry >= nr_entries)
1034 return -EINVAL; /* invalid entry */
1035 for (j = i + 1; j < nvec; j++) {
1036 if (entries[i].entry == entries[j].entry)
1037 return -EINVAL; /* duplicate entry */
1042 /* Check whether driver already requested for MSI IRQ */
1043 if (dev->msi_enabled) {
1044 pci_info(dev, "can't enable MSI-X (MSI IRQ already assigned)\n");
1047 return msix_capability_init(dev, entries, nvec, affd);
1050 static void pci_msix_shutdown(struct pci_dev *dev)
1052 struct msi_desc *entry;
1054 if (!pci_msi_enable || !dev || !dev->msix_enabled)
1057 if (pci_dev_is_disconnected(dev)) {
1058 dev->msix_enabled = 0;
1062 /* Return the device with MSI-X masked as initial states */
1063 for_each_pci_msi_entry(entry, dev)
1064 __pci_msix_desc_mask_irq(entry, 1);
1066 pci_msix_clear_and_set_ctrl(dev, PCI_MSIX_FLAGS_ENABLE, 0);
1067 pci_intx_for_msi(dev, 1);
1068 dev->msix_enabled = 0;
1069 pcibios_alloc_irq(dev);
1072 void pci_disable_msix(struct pci_dev *dev)
1074 if (!pci_msi_enable || !dev || !dev->msix_enabled)
1077 pci_msix_shutdown(dev);
1080 EXPORT_SYMBOL(pci_disable_msix);
1082 void pci_no_msi(void)
1088 * pci_msi_enabled - is MSI enabled?
1090 * Returns true if MSI has not been disabled by the command-line option
1093 int pci_msi_enabled(void)
1095 return pci_msi_enable;
1097 EXPORT_SYMBOL(pci_msi_enabled);
1099 static int __pci_enable_msi_range(struct pci_dev *dev, int minvec, int maxvec,
1100 struct irq_affinity *affd)
1105 if (!pci_msi_supported(dev, minvec) || dev->current_state != PCI_D0)
1108 /* Check whether driver already requested MSI-X IRQs */
1109 if (dev->msix_enabled) {
1110 pci_info(dev, "can't enable MSI (MSI-X already enabled)\n");
1114 if (maxvec < minvec)
1117 if (WARN_ON_ONCE(dev->msi_enabled))
1120 nvec = pci_msi_vec_count(dev);
1131 nvec = irq_calc_affinity_vectors(minvec, nvec, affd);
1136 rc = msi_capability_init(dev, nvec, affd);
1149 /* deprecated, don't use */
1150 int pci_enable_msi(struct pci_dev *dev)
1152 int rc = __pci_enable_msi_range(dev, 1, 1, NULL);
1157 EXPORT_SYMBOL(pci_enable_msi);
1159 static int __pci_enable_msix_range(struct pci_dev *dev,
1160 struct msix_entry *entries, int minvec,
1161 int maxvec, struct irq_affinity *affd,
1164 int rc, nvec = maxvec;
1166 if (maxvec < minvec)
1169 if (WARN_ON_ONCE(dev->msix_enabled))
1174 nvec = irq_calc_affinity_vectors(minvec, nvec, affd);
1179 rc = __pci_enable_msix(dev, entries, nvec, affd, flags);
1193 * pci_enable_msix_range - configure device's MSI-X capability structure
1194 * @dev: pointer to the pci_dev data structure of MSI-X device function
1195 * @entries: pointer to an array of MSI-X entries
1196 * @minvec: minimum number of MSI-X IRQs requested
1197 * @maxvec: maximum number of MSI-X IRQs requested
1199 * Setup the MSI-X capability structure of device function with a maximum
1200 * possible number of interrupts in the range between @minvec and @maxvec
1201 * upon its software driver call to request for MSI-X mode enabled on its
1202 * hardware device function. It returns a negative errno if an error occurs.
1203 * If it succeeds, it returns the actual number of interrupts allocated and
1204 * indicates the successful configuration of MSI-X capability structure
1205 * with new allocated MSI-X interrupts.
1207 int pci_enable_msix_range(struct pci_dev *dev, struct msix_entry *entries,
1208 int minvec, int maxvec)
1210 return __pci_enable_msix_range(dev, entries, minvec, maxvec, NULL, 0);
1212 EXPORT_SYMBOL(pci_enable_msix_range);
1215 * pci_alloc_irq_vectors_affinity - allocate multiple IRQs for a device
1216 * @dev: PCI device to operate on
1217 * @min_vecs: minimum number of vectors required (must be >= 1)
1218 * @max_vecs: maximum (desired) number of vectors
1219 * @flags: flags or quirks for the allocation
1220 * @affd: optional description of the affinity requirements
1222 * Allocate up to @max_vecs interrupt vectors for @dev, using MSI-X or MSI
1223 * vectors if available, and fall back to a single legacy vector
1224 * if neither is available. Return the number of vectors allocated,
1225 * (which might be smaller than @max_vecs) if successful, or a negative
1226 * error code on error. If less than @min_vecs interrupt vectors are
1227 * available for @dev the function will fail with -ENOSPC.
1229 * To get the Linux IRQ number used for a vector that can be passed to
1230 * request_irq() use the pci_irq_vector() helper.
1232 int pci_alloc_irq_vectors_affinity(struct pci_dev *dev, unsigned int min_vecs,
1233 unsigned int max_vecs, unsigned int flags,
1234 struct irq_affinity *affd)
1236 struct irq_affinity msi_default_affd = {0};
1237 int nvecs = -ENOSPC;
1239 if (flags & PCI_IRQ_AFFINITY) {
1241 affd = &msi_default_affd;
1247 if (flags & PCI_IRQ_MSIX) {
1248 nvecs = __pci_enable_msix_range(dev, NULL, min_vecs, max_vecs,
1254 if (flags & PCI_IRQ_MSI) {
1255 nvecs = __pci_enable_msi_range(dev, min_vecs, max_vecs, affd);
1260 /* use legacy IRQ if allowed */
1261 if (flags & PCI_IRQ_LEGACY) {
1262 if (min_vecs == 1 && dev->irq) {
1264 * Invoke the affinity spreading logic to ensure that
1265 * the device driver can adjust queue configuration
1266 * for the single interrupt case.
1269 irq_create_affinity_masks(1, affd);
1277 EXPORT_SYMBOL(pci_alloc_irq_vectors_affinity);
1280 * pci_free_irq_vectors - free previously allocated IRQs for a device
1281 * @dev: PCI device to operate on
1283 * Undoes the allocations and enabling in pci_alloc_irq_vectors().
1285 void pci_free_irq_vectors(struct pci_dev *dev)
1287 pci_disable_msix(dev);
1288 pci_disable_msi(dev);
1290 EXPORT_SYMBOL(pci_free_irq_vectors);
1293 * pci_irq_vector - return Linux IRQ number of a device vector
1294 * @dev: PCI device to operate on
1295 * @nr: Interrupt vector index (0-based)
1297 * @nr has the following meanings depending on the interrupt mode:
1298 * MSI-X: The index in the MSI-X vector table
1299 * MSI: The index of the enabled MSI vectors
1302 * Return: The Linux interrupt number or -EINVAl if @nr is out of range.
1304 int pci_irq_vector(struct pci_dev *dev, unsigned int nr)
1306 if (dev->msix_enabled) {
1307 struct msi_desc *entry;
1309 for_each_pci_msi_entry(entry, dev) {
1310 if (entry->msi_attrib.entry_nr == nr)
1317 if (dev->msi_enabled) {
1318 struct msi_desc *entry = first_pci_msi_entry(dev);
1320 if (WARN_ON_ONCE(nr >= entry->nvec_used))
1323 if (WARN_ON_ONCE(nr > 0))
1327 return dev->irq + nr;
1329 EXPORT_SYMBOL(pci_irq_vector);
1332 * pci_irq_get_affinity - return the affinity of a particular MSI vector
1333 * @dev: PCI device to operate on
1334 * @nr: device-relative interrupt vector index (0-based).
1336 * @nr has the following meanings depending on the interrupt mode:
1337 * MSI-X: The index in the MSI-X vector table
1338 * MSI: The index of the enabled MSI vectors
1341 * Return: A cpumask pointer or NULL if @nr is out of range
1343 const struct cpumask *pci_irq_get_affinity(struct pci_dev *dev, int nr)
1345 if (dev->msix_enabled) {
1346 struct msi_desc *entry;
1348 for_each_pci_msi_entry(entry, dev) {
1349 if (entry->msi_attrib.entry_nr == nr)
1350 return &entry->affinity->mask;
1354 } else if (dev->msi_enabled) {
1355 struct msi_desc *entry = first_pci_msi_entry(dev);
1357 if (WARN_ON_ONCE(!entry || !entry->affinity ||
1358 nr >= entry->nvec_used))
1361 return &entry->affinity[nr].mask;
1363 return cpu_possible_mask;
1366 EXPORT_SYMBOL(pci_irq_get_affinity);
1368 struct pci_dev *msi_desc_to_pci_dev(struct msi_desc *desc)
1370 return to_pci_dev(desc->dev);
1372 EXPORT_SYMBOL(msi_desc_to_pci_dev);
1374 void *msi_desc_to_pci_sysdata(struct msi_desc *desc)
1376 struct pci_dev *dev = msi_desc_to_pci_dev(desc);
1378 return dev->bus->sysdata;
1380 EXPORT_SYMBOL_GPL(msi_desc_to_pci_sysdata);
1382 #ifdef CONFIG_PCI_MSI_IRQ_DOMAIN
1384 * pci_msi_domain_write_msg - Helper to write MSI message to PCI config space
1385 * @irq_data: Pointer to interrupt data of the MSI interrupt
1386 * @msg: Pointer to the message
1388 void pci_msi_domain_write_msg(struct irq_data *irq_data, struct msi_msg *msg)
1390 struct msi_desc *desc = irq_data_get_msi_desc(irq_data);
1393 * For MSI-X desc->irq is always equal to irq_data->irq. For
1394 * MSI only the first interrupt of MULTI MSI passes the test.
1396 if (desc->irq == irq_data->irq)
1397 __pci_write_msi_msg(desc, msg);
1401 * pci_msi_domain_calc_hwirq - Generate a unique ID for an MSI source
1402 * @desc: Pointer to the MSI descriptor
1404 * The ID number is only used within the irqdomain.
1406 static irq_hw_number_t pci_msi_domain_calc_hwirq(struct msi_desc *desc)
1408 struct pci_dev *dev = msi_desc_to_pci_dev(desc);
1410 return (irq_hw_number_t)desc->msi_attrib.entry_nr |
1411 pci_dev_id(dev) << 11 |
1412 (pci_domain_nr(dev->bus) & 0xFFFFFFFF) << 27;
1415 static inline bool pci_msi_desc_is_multi_msi(struct msi_desc *desc)
1417 return !desc->msi_attrib.is_msix && desc->nvec_used > 1;
1421 * pci_msi_domain_check_cap - Verify that @domain supports the capabilities
1423 * @domain: The interrupt domain to check
1424 * @info: The domain info for verification
1425 * @dev: The device to check
1428 * 0 if the functionality is supported
1429 * 1 if Multi MSI is requested, but the domain does not support it
1430 * -ENOTSUPP otherwise
1432 int pci_msi_domain_check_cap(struct irq_domain *domain,
1433 struct msi_domain_info *info, struct device *dev)
1435 struct msi_desc *desc = first_pci_msi_entry(to_pci_dev(dev));
1437 /* Special handling to support __pci_enable_msi_range() */
1438 if (pci_msi_desc_is_multi_msi(desc) &&
1439 !(info->flags & MSI_FLAG_MULTI_PCI_MSI))
1441 else if (desc->msi_attrib.is_msix && !(info->flags & MSI_FLAG_PCI_MSIX))
1447 static int pci_msi_domain_handle_error(struct irq_domain *domain,
1448 struct msi_desc *desc, int error)
1450 /* Special handling to support __pci_enable_msi_range() */
1451 if (pci_msi_desc_is_multi_msi(desc) && error == -ENOSPC)
1457 static void pci_msi_domain_set_desc(msi_alloc_info_t *arg,
1458 struct msi_desc *desc)
1461 arg->hwirq = pci_msi_domain_calc_hwirq(desc);
1464 static struct msi_domain_ops pci_msi_domain_ops_default = {
1465 .set_desc = pci_msi_domain_set_desc,
1466 .msi_check = pci_msi_domain_check_cap,
1467 .handle_error = pci_msi_domain_handle_error,
1470 static void pci_msi_domain_update_dom_ops(struct msi_domain_info *info)
1472 struct msi_domain_ops *ops = info->ops;
1475 info->ops = &pci_msi_domain_ops_default;
1477 if (ops->set_desc == NULL)
1478 ops->set_desc = pci_msi_domain_set_desc;
1479 if (ops->msi_check == NULL)
1480 ops->msi_check = pci_msi_domain_check_cap;
1481 if (ops->handle_error == NULL)
1482 ops->handle_error = pci_msi_domain_handle_error;
1486 static void pci_msi_domain_update_chip_ops(struct msi_domain_info *info)
1488 struct irq_chip *chip = info->chip;
1491 if (!chip->irq_write_msi_msg)
1492 chip->irq_write_msi_msg = pci_msi_domain_write_msg;
1493 if (!chip->irq_mask)
1494 chip->irq_mask = pci_msi_mask_irq;
1495 if (!chip->irq_unmask)
1496 chip->irq_unmask = pci_msi_unmask_irq;
1500 * pci_msi_create_irq_domain - Create a MSI interrupt domain
1501 * @fwnode: Optional fwnode of the interrupt controller
1502 * @info: MSI domain info
1503 * @parent: Parent irq domain
1505 * Updates the domain and chip ops and creates a MSI interrupt domain.
1508 * A domain pointer or NULL in case of failure.
1510 struct irq_domain *pci_msi_create_irq_domain(struct fwnode_handle *fwnode,
1511 struct msi_domain_info *info,
1512 struct irq_domain *parent)
1514 struct irq_domain *domain;
1516 if (WARN_ON(info->flags & MSI_FLAG_LEVEL_CAPABLE))
1517 info->flags &= ~MSI_FLAG_LEVEL_CAPABLE;
1519 if (info->flags & MSI_FLAG_USE_DEF_DOM_OPS)
1520 pci_msi_domain_update_dom_ops(info);
1521 if (info->flags & MSI_FLAG_USE_DEF_CHIP_OPS)
1522 pci_msi_domain_update_chip_ops(info);
1524 info->flags |= MSI_FLAG_ACTIVATE_EARLY;
1525 if (IS_ENABLED(CONFIG_GENERIC_IRQ_RESERVATION_MODE))
1526 info->flags |= MSI_FLAG_MUST_REACTIVATE;
1528 /* PCI-MSI is oneshot-safe */
1529 info->chip->flags |= IRQCHIP_ONESHOT_SAFE;
1531 domain = msi_create_irq_domain(fwnode, info, parent);
1535 irq_domain_update_bus_token(domain, DOMAIN_BUS_PCI_MSI);
1538 EXPORT_SYMBOL_GPL(pci_msi_create_irq_domain);
1541 * Users of the generic MSI infrastructure expect a device to have a single ID,
1542 * so with DMA aliases we have to pick the least-worst compromise. Devices with
1543 * DMA phantom functions tend to still emit MSIs from the real function number,
1544 * so we ignore those and only consider topological aliases where either the
1545 * alias device or RID appears on a different bus number. We also make the
1546 * reasonable assumption that bridges are walked in an upstream direction (so
1547 * the last one seen wins), and the much braver assumption that the most likely
1548 * case is that of PCI->PCIe so we should always use the alias RID. This echoes
1549 * the logic from intel_irq_remapping's set_msi_sid(), which presumably works
1550 * well enough in practice; in the face of the horrible PCIe<->PCI-X conditions
1551 * for taking ownership all we can really do is close our eyes and hope...
1553 static int get_msi_id_cb(struct pci_dev *pdev, u16 alias, void *data)
1556 u8 bus = PCI_BUS_NUM(*pa);
1558 if (pdev->bus->number != bus || PCI_BUS_NUM(alias) != bus)
1565 * pci_msi_domain_get_msi_rid - Get the MSI requester id (RID)
1566 * @domain: The interrupt domain
1567 * @pdev: The PCI device.
1569 * The RID for a device is formed from the alias, with a firmware
1570 * supplied mapping applied
1574 u32 pci_msi_domain_get_msi_rid(struct irq_domain *domain, struct pci_dev *pdev)
1576 struct device_node *of_node;
1577 u32 rid = pci_dev_id(pdev);
1579 pci_for_each_dma_alias(pdev, get_msi_id_cb, &rid);
1581 of_node = irq_domain_get_of_node(domain);
1582 rid = of_node ? of_msi_map_id(&pdev->dev, of_node, rid) :
1583 iort_msi_map_id(&pdev->dev, rid);
1589 * pci_msi_get_device_domain - Get the MSI domain for a given PCI device
1590 * @pdev: The PCI device
1592 * Use the firmware data to find a device-specific MSI domain
1593 * (i.e. not one that is set as a default).
1595 * Returns: The corresponding MSI domain or NULL if none has been found.
1597 struct irq_domain *pci_msi_get_device_domain(struct pci_dev *pdev)
1599 struct irq_domain *dom;
1600 u32 rid = pci_dev_id(pdev);
1602 pci_for_each_dma_alias(pdev, get_msi_id_cb, &rid);
1603 dom = of_msi_map_get_device_domain(&pdev->dev, rid, DOMAIN_BUS_PCI_MSI);
1605 dom = iort_get_device_domain(&pdev->dev, rid,
1606 DOMAIN_BUS_PCI_MSI);
1611 * pci_dev_has_special_msi_domain - Check whether the device is handled by
1612 * a non-standard PCI-MSI domain
1613 * @pdev: The PCI device to check.
1615 * Returns: True if the device irqdomain or the bus irqdomain is
1616 * non-standard PCI/MSI.
1618 bool pci_dev_has_special_msi_domain(struct pci_dev *pdev)
1620 struct irq_domain *dom = dev_get_msi_domain(&pdev->dev);
1623 dom = dev_get_msi_domain(&pdev->bus->dev);
1628 return dom->bus_token != DOMAIN_BUS_PCI_MSI;
1631 #endif /* CONFIG_PCI_MSI_IRQ_DOMAIN */