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
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);
136 static void default_restore_msi_irq(struct pci_dev *dev, int irq)
138 struct msi_desc *entry;
141 if (dev->msix_enabled) {
142 for_each_pci_msi_entry(entry, dev) {
143 if (irq == entry->irq)
146 } else if (dev->msi_enabled) {
147 entry = irq_get_msi_desc(irq);
151 __pci_write_msi_msg(entry, &entry->msg);
154 void __weak arch_restore_msi_irqs(struct pci_dev *dev)
156 return default_restore_msi_irqs(dev);
159 static inline __attribute_const__ u32 msi_mask(unsigned x)
161 /* Don't shift by >= width of type */
164 return (1 << (1 << x)) - 1;
168 * PCI 2.3 does not specify mask bits for each MSI interrupt. Attempting to
169 * mask all MSI interrupts by clearing the MSI enable bit does not work
170 * reliably as devices without an INTx disable bit will then generate a
171 * level IRQ which will never be cleared.
173 u32 __pci_msi_desc_mask_irq(struct msi_desc *desc, u32 mask, u32 flag)
175 u32 mask_bits = desc->masked;
177 if (pci_msi_ignore_mask || !desc->msi_attrib.maskbit)
182 pci_write_config_dword(msi_desc_to_pci_dev(desc), desc->mask_pos,
188 static void msi_mask_irq(struct msi_desc *desc, u32 mask, u32 flag)
190 desc->masked = __pci_msi_desc_mask_irq(desc, mask, flag);
193 static void __iomem *pci_msix_desc_addr(struct msi_desc *desc)
195 if (desc->msi_attrib.is_virtual)
198 return desc->mask_base +
199 desc->msi_attrib.entry_nr * PCI_MSIX_ENTRY_SIZE;
203 * This internal function does not flush PCI writes to the device.
204 * All users must ensure that they read from the device before either
205 * assuming that the device state is up to date, or returning out of this
206 * file. This saves a few milliseconds when initialising devices with lots
207 * of MSI-X interrupts.
209 u32 __pci_msix_desc_mask_irq(struct msi_desc *desc, u32 flag)
211 u32 mask_bits = desc->masked;
212 void __iomem *desc_addr;
214 if (pci_msi_ignore_mask)
217 desc_addr = pci_msix_desc_addr(desc);
221 mask_bits &= ~PCI_MSIX_ENTRY_CTRL_MASKBIT;
222 if (flag & PCI_MSIX_ENTRY_CTRL_MASKBIT)
223 mask_bits |= PCI_MSIX_ENTRY_CTRL_MASKBIT;
225 writel(mask_bits, desc_addr + PCI_MSIX_ENTRY_VECTOR_CTRL);
230 static void msix_mask_irq(struct msi_desc *desc, u32 flag)
232 desc->masked = __pci_msix_desc_mask_irq(desc, flag);
235 static void msi_set_mask_bit(struct irq_data *data, u32 flag)
237 struct msi_desc *desc = irq_data_get_msi_desc(data);
239 if (desc->msi_attrib.is_msix) {
240 msix_mask_irq(desc, flag);
241 readl(desc->mask_base); /* Flush write to device */
243 unsigned offset = data->irq - desc->irq;
244 msi_mask_irq(desc, 1 << offset, flag << offset);
249 * pci_msi_mask_irq - Generic IRQ chip callback to mask PCI/MSI interrupts
250 * @data: pointer to irqdata associated to that interrupt
252 void pci_msi_mask_irq(struct irq_data *data)
254 msi_set_mask_bit(data, 1);
256 EXPORT_SYMBOL_GPL(pci_msi_mask_irq);
259 * pci_msi_unmask_irq - Generic IRQ chip callback to unmask PCI/MSI interrupts
260 * @data: pointer to irqdata associated to that interrupt
262 void pci_msi_unmask_irq(struct irq_data *data)
264 msi_set_mask_bit(data, 0);
266 EXPORT_SYMBOL_GPL(pci_msi_unmask_irq);
268 void default_restore_msi_irqs(struct pci_dev *dev)
270 struct msi_desc *entry;
272 for_each_pci_msi_entry(entry, dev)
273 default_restore_msi_irq(dev, entry->irq);
276 void __pci_read_msi_msg(struct msi_desc *entry, struct msi_msg *msg)
278 struct pci_dev *dev = msi_desc_to_pci_dev(entry);
280 BUG_ON(dev->current_state != PCI_D0);
282 if (entry->msi_attrib.is_msix) {
283 void __iomem *base = pci_msix_desc_addr(entry);
290 msg->address_lo = readl(base + PCI_MSIX_ENTRY_LOWER_ADDR);
291 msg->address_hi = readl(base + PCI_MSIX_ENTRY_UPPER_ADDR);
292 msg->data = readl(base + PCI_MSIX_ENTRY_DATA);
294 int pos = dev->msi_cap;
297 pci_read_config_dword(dev, pos + PCI_MSI_ADDRESS_LO,
299 if (entry->msi_attrib.is_64) {
300 pci_read_config_dword(dev, pos + PCI_MSI_ADDRESS_HI,
302 pci_read_config_word(dev, pos + PCI_MSI_DATA_64, &data);
305 pci_read_config_word(dev, pos + PCI_MSI_DATA_32, &data);
311 void __pci_write_msi_msg(struct msi_desc *entry, struct msi_msg *msg)
313 struct pci_dev *dev = msi_desc_to_pci_dev(entry);
315 if (dev->current_state != PCI_D0 || pci_dev_is_disconnected(dev)) {
316 /* Don't touch the hardware now */
317 } else if (entry->msi_attrib.is_msix) {
318 void __iomem *base = pci_msix_desc_addr(entry);
323 writel(msg->address_lo, base + PCI_MSIX_ENTRY_LOWER_ADDR);
324 writel(msg->address_hi, base + PCI_MSIX_ENTRY_UPPER_ADDR);
325 writel(msg->data, base + PCI_MSIX_ENTRY_DATA);
327 int pos = dev->msi_cap;
330 pci_read_config_word(dev, pos + PCI_MSI_FLAGS, &msgctl);
331 msgctl &= ~PCI_MSI_FLAGS_QSIZE;
332 msgctl |= entry->msi_attrib.multiple << 4;
333 pci_write_config_word(dev, pos + PCI_MSI_FLAGS, msgctl);
335 pci_write_config_dword(dev, pos + PCI_MSI_ADDRESS_LO,
337 if (entry->msi_attrib.is_64) {
338 pci_write_config_dword(dev, pos + PCI_MSI_ADDRESS_HI,
340 pci_write_config_word(dev, pos + PCI_MSI_DATA_64,
343 pci_write_config_word(dev, pos + PCI_MSI_DATA_32,
351 if (entry->write_msi_msg)
352 entry->write_msi_msg(entry, entry->write_msi_msg_data);
356 void pci_write_msi_msg(unsigned int irq, struct msi_msg *msg)
358 struct msi_desc *entry = irq_get_msi_desc(irq);
360 __pci_write_msi_msg(entry, msg);
362 EXPORT_SYMBOL_GPL(pci_write_msi_msg);
364 static void free_msi_irqs(struct pci_dev *dev)
366 struct list_head *msi_list = dev_to_msi_list(&dev->dev);
367 struct msi_desc *entry, *tmp;
368 struct attribute **msi_attrs;
369 struct device_attribute *dev_attr;
372 for_each_pci_msi_entry(entry, dev)
374 for (i = 0; i < entry->nvec_used; i++)
375 BUG_ON(irq_has_action(entry->irq + i));
377 pci_msi_teardown_msi_irqs(dev);
379 list_for_each_entry_safe(entry, tmp, msi_list, list) {
380 if (entry->msi_attrib.is_msix) {
381 if (list_is_last(&entry->list, msi_list))
382 iounmap(entry->mask_base);
385 list_del(&entry->list);
386 free_msi_entry(entry);
389 if (dev->msi_irq_groups) {
390 sysfs_remove_groups(&dev->dev.kobj, dev->msi_irq_groups);
391 msi_attrs = dev->msi_irq_groups[0]->attrs;
392 while (msi_attrs[count]) {
393 dev_attr = container_of(msi_attrs[count],
394 struct device_attribute, attr);
395 kfree(dev_attr->attr.name);
400 kfree(dev->msi_irq_groups[0]);
401 kfree(dev->msi_irq_groups);
402 dev->msi_irq_groups = NULL;
406 static void pci_intx_for_msi(struct pci_dev *dev, int enable)
408 if (!(dev->dev_flags & PCI_DEV_FLAGS_MSI_INTX_DISABLE_BUG))
409 pci_intx(dev, enable);
412 static void __pci_restore_msi_state(struct pci_dev *dev)
415 struct msi_desc *entry;
417 if (!dev->msi_enabled)
420 entry = irq_get_msi_desc(dev->irq);
422 pci_intx_for_msi(dev, 0);
423 pci_msi_set_enable(dev, 0);
424 arch_restore_msi_irqs(dev);
426 pci_read_config_word(dev, dev->msi_cap + PCI_MSI_FLAGS, &control);
427 msi_mask_irq(entry, msi_mask(entry->msi_attrib.multi_cap),
429 control &= ~PCI_MSI_FLAGS_QSIZE;
430 control |= (entry->msi_attrib.multiple << 4) | PCI_MSI_FLAGS_ENABLE;
431 pci_write_config_word(dev, dev->msi_cap + PCI_MSI_FLAGS, control);
434 static void __pci_restore_msix_state(struct pci_dev *dev)
436 struct msi_desc *entry;
438 if (!dev->msix_enabled)
440 BUG_ON(list_empty(dev_to_msi_list(&dev->dev)));
442 /* route the table */
443 pci_intx_for_msi(dev, 0);
444 pci_msix_clear_and_set_ctrl(dev, 0,
445 PCI_MSIX_FLAGS_ENABLE | PCI_MSIX_FLAGS_MASKALL);
447 arch_restore_msi_irqs(dev);
448 for_each_pci_msi_entry(entry, dev)
449 msix_mask_irq(entry, entry->masked);
451 pci_msix_clear_and_set_ctrl(dev, PCI_MSIX_FLAGS_MASKALL, 0);
454 void pci_restore_msi_state(struct pci_dev *dev)
456 __pci_restore_msi_state(dev);
457 __pci_restore_msix_state(dev);
459 EXPORT_SYMBOL_GPL(pci_restore_msi_state);
461 static ssize_t msi_mode_show(struct device *dev, struct device_attribute *attr,
464 struct msi_desc *entry;
468 retval = kstrtoul(attr->attr.name, 10, &irq);
472 entry = irq_get_msi_desc(irq);
474 return sprintf(buf, "%s\n",
475 entry->msi_attrib.is_msix ? "msix" : "msi");
480 static int populate_msi_sysfs(struct pci_dev *pdev)
482 struct attribute **msi_attrs;
483 struct attribute *msi_attr;
484 struct device_attribute *msi_dev_attr;
485 struct attribute_group *msi_irq_group;
486 const struct attribute_group **msi_irq_groups;
487 struct msi_desc *entry;
493 /* Determine how many msi entries we have */
494 for_each_pci_msi_entry(entry, pdev)
495 num_msi += entry->nvec_used;
499 /* Dynamically create the MSI attributes for the PCI device */
500 msi_attrs = kcalloc(num_msi + 1, sizeof(void *), GFP_KERNEL);
503 for_each_pci_msi_entry(entry, pdev) {
504 for (i = 0; i < entry->nvec_used; i++) {
505 msi_dev_attr = kzalloc(sizeof(*msi_dev_attr), GFP_KERNEL);
508 msi_attrs[count] = &msi_dev_attr->attr;
510 sysfs_attr_init(&msi_dev_attr->attr);
511 msi_dev_attr->attr.name = kasprintf(GFP_KERNEL, "%d",
513 if (!msi_dev_attr->attr.name)
515 msi_dev_attr->attr.mode = S_IRUGO;
516 msi_dev_attr->show = msi_mode_show;
521 msi_irq_group = kzalloc(sizeof(*msi_irq_group), GFP_KERNEL);
524 msi_irq_group->name = "msi_irqs";
525 msi_irq_group->attrs = msi_attrs;
527 msi_irq_groups = kcalloc(2, sizeof(void *), GFP_KERNEL);
529 goto error_irq_group;
530 msi_irq_groups[0] = msi_irq_group;
532 ret = sysfs_create_groups(&pdev->dev.kobj, msi_irq_groups);
534 goto error_irq_groups;
535 pdev->msi_irq_groups = msi_irq_groups;
540 kfree(msi_irq_groups);
542 kfree(msi_irq_group);
545 msi_attr = msi_attrs[count];
547 msi_dev_attr = container_of(msi_attr, struct device_attribute, attr);
548 kfree(msi_attr->name);
551 msi_attr = msi_attrs[count];
557 static struct msi_desc *
558 msi_setup_entry(struct pci_dev *dev, int nvec, struct irq_affinity *affd)
560 struct irq_affinity_desc *masks = NULL;
561 struct msi_desc *entry;
565 masks = irq_create_affinity_masks(nvec, affd);
567 /* MSI Entry Initialization */
568 entry = alloc_msi_entry(&dev->dev, nvec, masks);
572 pci_read_config_word(dev, dev->msi_cap + PCI_MSI_FLAGS, &control);
574 entry->msi_attrib.is_msix = 0;
575 entry->msi_attrib.is_64 = !!(control & PCI_MSI_FLAGS_64BIT);
576 entry->msi_attrib.is_virtual = 0;
577 entry->msi_attrib.entry_nr = 0;
578 entry->msi_attrib.maskbit = !!(control & PCI_MSI_FLAGS_MASKBIT);
579 entry->msi_attrib.default_irq = dev->irq; /* Save IOAPIC IRQ */
580 entry->msi_attrib.multi_cap = (control & PCI_MSI_FLAGS_QMASK) >> 1;
581 entry->msi_attrib.multiple = ilog2(__roundup_pow_of_two(nvec));
583 if (control & PCI_MSI_FLAGS_64BIT)
584 entry->mask_pos = dev->msi_cap + PCI_MSI_MASK_64;
586 entry->mask_pos = dev->msi_cap + PCI_MSI_MASK_32;
588 /* Save the initial mask status */
589 if (entry->msi_attrib.maskbit)
590 pci_read_config_dword(dev, entry->mask_pos, &entry->masked);
597 static int msi_verify_entries(struct pci_dev *dev)
599 struct msi_desc *entry;
601 for_each_pci_msi_entry(entry, dev) {
602 if (!dev->no_64bit_msi || !entry->msg.address_hi)
604 pci_err(dev, "Device has broken 64-bit MSI but arch"
605 " tried to assign one above 4G\n");
612 * msi_capability_init - configure device's MSI capability structure
613 * @dev: pointer to the pci_dev data structure of MSI device function
614 * @nvec: number of interrupts to allocate
615 * @affd: description of automatic IRQ affinity assignments (may be %NULL)
617 * Setup the MSI capability structure of the device with the requested
618 * number of interrupts. A return value of zero indicates the successful
619 * setup of an entry with the new MSI IRQ. A negative return value indicates
620 * an error, and a positive return value indicates the number of interrupts
621 * which could have been allocated.
623 static int msi_capability_init(struct pci_dev *dev, int nvec,
624 struct irq_affinity *affd)
626 struct msi_desc *entry;
630 pci_msi_set_enable(dev, 0); /* Disable MSI during set up */
632 entry = msi_setup_entry(dev, nvec, affd);
636 /* All MSIs are unmasked by default; mask them all */
637 mask = msi_mask(entry->msi_attrib.multi_cap);
638 msi_mask_irq(entry, mask, mask);
640 list_add_tail(&entry->list, dev_to_msi_list(&dev->dev));
642 /* Configure MSI capability structure */
643 ret = pci_msi_setup_msi_irqs(dev, nvec, PCI_CAP_ID_MSI);
645 msi_mask_irq(entry, mask, ~mask);
650 ret = msi_verify_entries(dev);
652 msi_mask_irq(entry, mask, ~mask);
657 ret = populate_msi_sysfs(dev);
659 msi_mask_irq(entry, mask, ~mask);
664 /* Set MSI enabled bits */
665 pci_intx_for_msi(dev, 0);
666 pci_msi_set_enable(dev, 1);
667 dev->msi_enabled = 1;
669 pcibios_free_irq(dev);
670 dev->irq = entry->irq;
674 static void __iomem *msix_map_region(struct pci_dev *dev, unsigned nr_entries)
676 resource_size_t phys_addr;
681 pci_read_config_dword(dev, dev->msix_cap + PCI_MSIX_TABLE,
683 bir = (u8)(table_offset & PCI_MSIX_TABLE_BIR);
684 flags = pci_resource_flags(dev, bir);
685 if (!flags || (flags & IORESOURCE_UNSET))
688 table_offset &= PCI_MSIX_TABLE_OFFSET;
689 phys_addr = pci_resource_start(dev, bir) + table_offset;
691 return ioremap_nocache(phys_addr, nr_entries * PCI_MSIX_ENTRY_SIZE);
694 static int msix_setup_entries(struct pci_dev *dev, void __iomem *base,
695 struct msix_entry *entries, int nvec,
696 struct irq_affinity *affd)
698 struct irq_affinity_desc *curmsk, *masks = NULL;
699 struct msi_desc *entry;
701 int vec_count = pci_msix_vec_count(dev);
704 masks = irq_create_affinity_masks(nvec, affd);
706 for (i = 0, curmsk = masks; i < nvec; i++) {
707 entry = alloc_msi_entry(&dev->dev, 1, curmsk);
713 /* No enough memory. Don't try again */
718 entry->msi_attrib.is_msix = 1;
719 entry->msi_attrib.is_64 = 1;
721 entry->msi_attrib.entry_nr = entries[i].entry;
723 entry->msi_attrib.entry_nr = i;
725 entry->msi_attrib.is_virtual =
726 entry->msi_attrib.entry_nr >= vec_count;
728 entry->msi_attrib.default_irq = dev->irq;
729 entry->mask_base = base;
731 list_add_tail(&entry->list, dev_to_msi_list(&dev->dev));
741 static void msix_program_entries(struct pci_dev *dev,
742 struct msix_entry *entries)
744 struct msi_desc *entry;
746 void __iomem *desc_addr;
748 for_each_pci_msi_entry(entry, dev) {
750 entries[i++].vector = entry->irq;
752 desc_addr = pci_msix_desc_addr(entry);
754 entry->masked = readl(desc_addr +
755 PCI_MSIX_ENTRY_VECTOR_CTRL);
759 msix_mask_irq(entry, 1);
764 * msix_capability_init - configure device's MSI-X capability
765 * @dev: pointer to the pci_dev data structure of MSI-X device function
766 * @entries: pointer to an array of struct msix_entry entries
767 * @nvec: number of @entries
768 * @affd: Optional pointer to enable automatic affinity assignment
770 * Setup the MSI-X capability structure of device function with a
771 * single MSI-X IRQ. A return of zero indicates the successful setup of
772 * requested MSI-X entries with allocated IRQs or non-zero for otherwise.
774 static int msix_capability_init(struct pci_dev *dev, struct msix_entry *entries,
775 int nvec, struct irq_affinity *affd)
781 /* Ensure MSI-X is disabled while it is set up */
782 pci_msix_clear_and_set_ctrl(dev, PCI_MSIX_FLAGS_ENABLE, 0);
784 pci_read_config_word(dev, dev->msix_cap + PCI_MSIX_FLAGS, &control);
785 /* Request & Map MSI-X table region */
786 base = msix_map_region(dev, msix_table_size(control));
790 ret = msix_setup_entries(dev, base, entries, nvec, affd);
794 ret = pci_msi_setup_msi_irqs(dev, nvec, PCI_CAP_ID_MSIX);
798 /* Check if all MSI entries honor device restrictions */
799 ret = msi_verify_entries(dev);
804 * Some devices require MSI-X to be enabled before we can touch the
805 * MSI-X registers. We need to mask all the vectors to prevent
806 * interrupts coming in before they're fully set up.
808 pci_msix_clear_and_set_ctrl(dev, 0,
809 PCI_MSIX_FLAGS_MASKALL | PCI_MSIX_FLAGS_ENABLE);
811 msix_program_entries(dev, entries);
813 ret = populate_msi_sysfs(dev);
817 /* Set MSI-X enabled bits and unmask the function */
818 pci_intx_for_msi(dev, 0);
819 dev->msix_enabled = 1;
820 pci_msix_clear_and_set_ctrl(dev, PCI_MSIX_FLAGS_MASKALL, 0);
822 pcibios_free_irq(dev);
828 * If we had some success, report the number of IRQs
829 * we succeeded in setting up.
831 struct msi_desc *entry;
834 for_each_pci_msi_entry(entry, dev) {
849 * pci_msi_supported - check whether MSI may be enabled on a device
850 * @dev: pointer to the pci_dev data structure of MSI device function
851 * @nvec: how many MSIs have been requested?
853 * Look at global flags, the device itself, and its parent buses
854 * to determine if MSI/-X are supported for the device. If MSI/-X is
855 * supported return 1, else return 0.
857 static int pci_msi_supported(struct pci_dev *dev, int nvec)
861 /* MSI must be globally enabled and supported by the device */
865 if (!dev || dev->no_msi || dev->current_state != PCI_D0)
869 * You can't ask to have 0 or less MSIs configured.
871 * b) the list manipulation code assumes nvec >= 1.
877 * Any bridge which does NOT route MSI transactions from its
878 * secondary bus to its primary bus must set NO_MSI flag on
879 * the secondary pci_bus.
880 * We expect only arch-specific PCI host bus controller driver
881 * or quirks for specific PCI bridges to be setting NO_MSI.
883 for (bus = dev->bus; bus; bus = bus->parent)
884 if (bus->bus_flags & PCI_BUS_FLAGS_NO_MSI)
891 * pci_msi_vec_count - Return the number of MSI vectors a device can send
892 * @dev: device to report about
894 * This function returns the number of MSI vectors a device requested via
895 * Multiple Message Capable register. It returns a negative errno if the
896 * device is not capable sending MSI interrupts. Otherwise, the call succeeds
897 * and returns a power of two, up to a maximum of 2^5 (32), according to the
900 int pci_msi_vec_count(struct pci_dev *dev)
908 pci_read_config_word(dev, dev->msi_cap + PCI_MSI_FLAGS, &msgctl);
909 ret = 1 << ((msgctl & PCI_MSI_FLAGS_QMASK) >> 1);
913 EXPORT_SYMBOL(pci_msi_vec_count);
915 static void pci_msi_shutdown(struct pci_dev *dev)
917 struct msi_desc *desc;
920 if (!pci_msi_enable || !dev || !dev->msi_enabled)
923 BUG_ON(list_empty(dev_to_msi_list(&dev->dev)));
924 desc = first_pci_msi_entry(dev);
926 pci_msi_set_enable(dev, 0);
927 pci_intx_for_msi(dev, 1);
928 dev->msi_enabled = 0;
930 /* Return the device with MSI unmasked as initial states */
931 mask = msi_mask(desc->msi_attrib.multi_cap);
932 /* Keep cached state to be restored */
933 __pci_msi_desc_mask_irq(desc, mask, ~mask);
935 /* Restore dev->irq to its default pin-assertion IRQ */
936 dev->irq = desc->msi_attrib.default_irq;
937 pcibios_alloc_irq(dev);
940 void pci_disable_msi(struct pci_dev *dev)
942 if (!pci_msi_enable || !dev || !dev->msi_enabled)
945 pci_msi_shutdown(dev);
948 EXPORT_SYMBOL(pci_disable_msi);
951 * pci_msix_vec_count - return the number of device's MSI-X table entries
952 * @dev: pointer to the pci_dev data structure of MSI-X device function
953 * This function returns the number of device's MSI-X table entries and
954 * therefore the number of MSI-X vectors device is capable of sending.
955 * It returns a negative errno if the device is not capable of sending MSI-X
958 int pci_msix_vec_count(struct pci_dev *dev)
965 pci_read_config_word(dev, dev->msix_cap + PCI_MSIX_FLAGS, &control);
966 return msix_table_size(control);
968 EXPORT_SYMBOL(pci_msix_vec_count);
970 static int __pci_enable_msix(struct pci_dev *dev, struct msix_entry *entries,
971 int nvec, struct irq_affinity *affd, int flags)
976 if (!pci_msi_supported(dev, nvec))
979 nr_entries = pci_msix_vec_count(dev);
982 if (nvec > nr_entries && !(flags & PCI_IRQ_VIRTUAL))
986 /* Check for any invalid entries */
987 for (i = 0; i < nvec; i++) {
988 if (entries[i].entry >= nr_entries)
989 return -EINVAL; /* invalid entry */
990 for (j = i + 1; j < nvec; j++) {
991 if (entries[i].entry == entries[j].entry)
992 return -EINVAL; /* duplicate entry */
997 /* Check whether driver already requested for MSI IRQ */
998 if (dev->msi_enabled) {
999 pci_info(dev, "can't enable MSI-X (MSI IRQ already assigned)\n");
1002 return msix_capability_init(dev, entries, nvec, affd);
1005 static void pci_msix_shutdown(struct pci_dev *dev)
1007 struct msi_desc *entry;
1009 if (!pci_msi_enable || !dev || !dev->msix_enabled)
1012 if (pci_dev_is_disconnected(dev)) {
1013 dev->msix_enabled = 0;
1017 /* Return the device with MSI-X masked as initial states */
1018 for_each_pci_msi_entry(entry, dev) {
1019 /* Keep cached states to be restored */
1020 __pci_msix_desc_mask_irq(entry, 1);
1023 pci_msix_clear_and_set_ctrl(dev, PCI_MSIX_FLAGS_ENABLE, 0);
1024 pci_intx_for_msi(dev, 1);
1025 dev->msix_enabled = 0;
1026 pcibios_alloc_irq(dev);
1029 void pci_disable_msix(struct pci_dev *dev)
1031 if (!pci_msi_enable || !dev || !dev->msix_enabled)
1034 pci_msix_shutdown(dev);
1037 EXPORT_SYMBOL(pci_disable_msix);
1039 void pci_no_msi(void)
1045 * pci_msi_enabled - is MSI enabled?
1047 * Returns true if MSI has not been disabled by the command-line option
1050 int pci_msi_enabled(void)
1052 return pci_msi_enable;
1054 EXPORT_SYMBOL(pci_msi_enabled);
1056 static int __pci_enable_msi_range(struct pci_dev *dev, int minvec, int maxvec,
1057 struct irq_affinity *affd)
1062 if (!pci_msi_supported(dev, minvec))
1065 /* Check whether driver already requested MSI-X IRQs */
1066 if (dev->msix_enabled) {
1067 pci_info(dev, "can't enable MSI (MSI-X already enabled)\n");
1071 if (maxvec < minvec)
1074 if (WARN_ON_ONCE(dev->msi_enabled))
1077 nvec = pci_msi_vec_count(dev);
1088 nvec = irq_calc_affinity_vectors(minvec, nvec, affd);
1093 rc = msi_capability_init(dev, nvec, affd);
1106 /* deprecated, don't use */
1107 int pci_enable_msi(struct pci_dev *dev)
1109 int rc = __pci_enable_msi_range(dev, 1, 1, NULL);
1114 EXPORT_SYMBOL(pci_enable_msi);
1116 static int __pci_enable_msix_range(struct pci_dev *dev,
1117 struct msix_entry *entries, int minvec,
1118 int maxvec, struct irq_affinity *affd,
1121 int rc, nvec = maxvec;
1123 if (maxvec < minvec)
1126 if (WARN_ON_ONCE(dev->msix_enabled))
1131 nvec = irq_calc_affinity_vectors(minvec, nvec, affd);
1136 rc = __pci_enable_msix(dev, entries, nvec, affd, flags);
1150 * pci_enable_msix_range - configure device's MSI-X capability structure
1151 * @dev: pointer to the pci_dev data structure of MSI-X device function
1152 * @entries: pointer to an array of MSI-X entries
1153 * @minvec: minimum number of MSI-X IRQs requested
1154 * @maxvec: maximum number of MSI-X IRQs requested
1156 * Setup the MSI-X capability structure of device function with a maximum
1157 * possible number of interrupts in the range between @minvec and @maxvec
1158 * upon its software driver call to request for MSI-X mode enabled on its
1159 * hardware device function. It returns a negative errno if an error occurs.
1160 * If it succeeds, it returns the actual number of interrupts allocated and
1161 * indicates the successful configuration of MSI-X capability structure
1162 * with new allocated MSI-X interrupts.
1164 int pci_enable_msix_range(struct pci_dev *dev, struct msix_entry *entries,
1165 int minvec, int maxvec)
1167 return __pci_enable_msix_range(dev, entries, minvec, maxvec, NULL, 0);
1169 EXPORT_SYMBOL(pci_enable_msix_range);
1172 * pci_alloc_irq_vectors_affinity - allocate multiple IRQs for a device
1173 * @dev: PCI device to operate on
1174 * @min_vecs: minimum number of vectors required (must be >= 1)
1175 * @max_vecs: maximum (desired) number of vectors
1176 * @flags: flags or quirks for the allocation
1177 * @affd: optional description of the affinity requirements
1179 * Allocate up to @max_vecs interrupt vectors for @dev, using MSI-X or MSI
1180 * vectors if available, and fall back to a single legacy vector
1181 * if neither is available. Return the number of vectors allocated,
1182 * (which might be smaller than @max_vecs) if successful, or a negative
1183 * error code on error. If less than @min_vecs interrupt vectors are
1184 * available for @dev the function will fail with -ENOSPC.
1186 * To get the Linux IRQ number used for a vector that can be passed to
1187 * request_irq() use the pci_irq_vector() helper.
1189 int pci_alloc_irq_vectors_affinity(struct pci_dev *dev, unsigned int min_vecs,
1190 unsigned int max_vecs, unsigned int flags,
1191 struct irq_affinity *affd)
1193 struct irq_affinity msi_default_affd = {0};
1194 int msix_vecs = -ENOSPC;
1195 int msi_vecs = -ENOSPC;
1197 if (flags & PCI_IRQ_AFFINITY) {
1199 affd = &msi_default_affd;
1205 if (flags & PCI_IRQ_MSIX) {
1206 msix_vecs = __pci_enable_msix_range(dev, NULL, min_vecs,
1207 max_vecs, affd, flags);
1212 if (flags & PCI_IRQ_MSI) {
1213 msi_vecs = __pci_enable_msi_range(dev, min_vecs, max_vecs,
1219 /* use legacy IRQ if allowed */
1220 if (flags & PCI_IRQ_LEGACY) {
1221 if (min_vecs == 1 && dev->irq) {
1223 * Invoke the affinity spreading logic to ensure that
1224 * the device driver can adjust queue configuration
1225 * for the single interrupt case.
1228 irq_create_affinity_masks(1, affd);
1234 if (msix_vecs == -ENOSPC)
1238 EXPORT_SYMBOL(pci_alloc_irq_vectors_affinity);
1241 * pci_free_irq_vectors - free previously allocated IRQs for a device
1242 * @dev: PCI device to operate on
1244 * Undoes the allocations and enabling in pci_alloc_irq_vectors().
1246 void pci_free_irq_vectors(struct pci_dev *dev)
1248 pci_disable_msix(dev);
1249 pci_disable_msi(dev);
1251 EXPORT_SYMBOL(pci_free_irq_vectors);
1254 * pci_irq_vector - return Linux IRQ number of a device vector
1255 * @dev: PCI device to operate on
1256 * @nr: device-relative interrupt vector index (0-based).
1258 int pci_irq_vector(struct pci_dev *dev, unsigned int nr)
1260 if (dev->msix_enabled) {
1261 struct msi_desc *entry;
1264 for_each_pci_msi_entry(entry, dev) {
1273 if (dev->msi_enabled) {
1274 struct msi_desc *entry = first_pci_msi_entry(dev);
1276 if (WARN_ON_ONCE(nr >= entry->nvec_used))
1279 if (WARN_ON_ONCE(nr > 0))
1283 return dev->irq + nr;
1285 EXPORT_SYMBOL(pci_irq_vector);
1288 * pci_irq_get_affinity - return the affinity of a particular MSI vector
1289 * @dev: PCI device to operate on
1290 * @nr: device-relative interrupt vector index (0-based).
1292 const struct cpumask *pci_irq_get_affinity(struct pci_dev *dev, int nr)
1294 if (dev->msix_enabled) {
1295 struct msi_desc *entry;
1298 for_each_pci_msi_entry(entry, dev) {
1300 return &entry->affinity->mask;
1305 } else if (dev->msi_enabled) {
1306 struct msi_desc *entry = first_pci_msi_entry(dev);
1308 if (WARN_ON_ONCE(!entry || !entry->affinity ||
1309 nr >= entry->nvec_used))
1312 return &entry->affinity[nr].mask;
1314 return cpu_possible_mask;
1317 EXPORT_SYMBOL(pci_irq_get_affinity);
1320 * pci_irq_get_node - return the NUMA node of a particular MSI vector
1321 * @pdev: PCI device to operate on
1322 * @vec: device-relative interrupt vector index (0-based).
1324 int pci_irq_get_node(struct pci_dev *pdev, int vec)
1326 const struct cpumask *mask;
1328 mask = pci_irq_get_affinity(pdev, vec);
1330 return local_memory_node(cpu_to_node(cpumask_first(mask)));
1331 return dev_to_node(&pdev->dev);
1333 EXPORT_SYMBOL(pci_irq_get_node);
1335 struct pci_dev *msi_desc_to_pci_dev(struct msi_desc *desc)
1337 return to_pci_dev(desc->dev);
1339 EXPORT_SYMBOL(msi_desc_to_pci_dev);
1341 void *msi_desc_to_pci_sysdata(struct msi_desc *desc)
1343 struct pci_dev *dev = msi_desc_to_pci_dev(desc);
1345 return dev->bus->sysdata;
1347 EXPORT_SYMBOL_GPL(msi_desc_to_pci_sysdata);
1349 #ifdef CONFIG_PCI_MSI_IRQ_DOMAIN
1351 * pci_msi_domain_write_msg - Helper to write MSI message to PCI config space
1352 * @irq_data: Pointer to interrupt data of the MSI interrupt
1353 * @msg: Pointer to the message
1355 void pci_msi_domain_write_msg(struct irq_data *irq_data, struct msi_msg *msg)
1357 struct msi_desc *desc = irq_data_get_msi_desc(irq_data);
1360 * For MSI-X desc->irq is always equal to irq_data->irq. For
1361 * MSI only the first interrupt of MULTI MSI passes the test.
1363 if (desc->irq == irq_data->irq)
1364 __pci_write_msi_msg(desc, msg);
1368 * pci_msi_domain_calc_hwirq - Generate a unique ID for an MSI source
1369 * @dev: Pointer to the PCI device
1370 * @desc: Pointer to the MSI descriptor
1372 * The ID number is only used within the irqdomain.
1374 irq_hw_number_t pci_msi_domain_calc_hwirq(struct pci_dev *dev,
1375 struct msi_desc *desc)
1377 return (irq_hw_number_t)desc->msi_attrib.entry_nr |
1378 pci_dev_id(dev) << 11 |
1379 (pci_domain_nr(dev->bus) & 0xFFFFFFFF) << 27;
1382 static inline bool pci_msi_desc_is_multi_msi(struct msi_desc *desc)
1384 return !desc->msi_attrib.is_msix && desc->nvec_used > 1;
1388 * pci_msi_domain_check_cap - Verify that @domain supports the capabilities
1390 * @domain: The interrupt domain to check
1391 * @info: The domain info for verification
1392 * @dev: The device to check
1395 * 0 if the functionality is supported
1396 * 1 if Multi MSI is requested, but the domain does not support it
1397 * -ENOTSUPP otherwise
1399 int pci_msi_domain_check_cap(struct irq_domain *domain,
1400 struct msi_domain_info *info, struct device *dev)
1402 struct msi_desc *desc = first_pci_msi_entry(to_pci_dev(dev));
1404 /* Special handling to support __pci_enable_msi_range() */
1405 if (pci_msi_desc_is_multi_msi(desc) &&
1406 !(info->flags & MSI_FLAG_MULTI_PCI_MSI))
1408 else if (desc->msi_attrib.is_msix && !(info->flags & MSI_FLAG_PCI_MSIX))
1414 static int pci_msi_domain_handle_error(struct irq_domain *domain,
1415 struct msi_desc *desc, int error)
1417 /* Special handling to support __pci_enable_msi_range() */
1418 if (pci_msi_desc_is_multi_msi(desc) && error == -ENOSPC)
1424 #ifdef GENERIC_MSI_DOMAIN_OPS
1425 static void pci_msi_domain_set_desc(msi_alloc_info_t *arg,
1426 struct msi_desc *desc)
1429 arg->hwirq = pci_msi_domain_calc_hwirq(msi_desc_to_pci_dev(desc),
1433 #define pci_msi_domain_set_desc NULL
1436 static struct msi_domain_ops pci_msi_domain_ops_default = {
1437 .set_desc = pci_msi_domain_set_desc,
1438 .msi_check = pci_msi_domain_check_cap,
1439 .handle_error = pci_msi_domain_handle_error,
1442 static void pci_msi_domain_update_dom_ops(struct msi_domain_info *info)
1444 struct msi_domain_ops *ops = info->ops;
1447 info->ops = &pci_msi_domain_ops_default;
1449 if (ops->set_desc == NULL)
1450 ops->set_desc = pci_msi_domain_set_desc;
1451 if (ops->msi_check == NULL)
1452 ops->msi_check = pci_msi_domain_check_cap;
1453 if (ops->handle_error == NULL)
1454 ops->handle_error = pci_msi_domain_handle_error;
1458 static void pci_msi_domain_update_chip_ops(struct msi_domain_info *info)
1460 struct irq_chip *chip = info->chip;
1463 if (!chip->irq_write_msi_msg)
1464 chip->irq_write_msi_msg = pci_msi_domain_write_msg;
1465 if (!chip->irq_mask)
1466 chip->irq_mask = pci_msi_mask_irq;
1467 if (!chip->irq_unmask)
1468 chip->irq_unmask = pci_msi_unmask_irq;
1472 * pci_msi_create_irq_domain - Create a MSI interrupt domain
1473 * @fwnode: Optional fwnode of the interrupt controller
1474 * @info: MSI domain info
1475 * @parent: Parent irq domain
1477 * Updates the domain and chip ops and creates a MSI interrupt domain.
1480 * A domain pointer or NULL in case of failure.
1482 struct irq_domain *pci_msi_create_irq_domain(struct fwnode_handle *fwnode,
1483 struct msi_domain_info *info,
1484 struct irq_domain *parent)
1486 struct irq_domain *domain;
1488 if (WARN_ON(info->flags & MSI_FLAG_LEVEL_CAPABLE))
1489 info->flags &= ~MSI_FLAG_LEVEL_CAPABLE;
1491 if (info->flags & MSI_FLAG_USE_DEF_DOM_OPS)
1492 pci_msi_domain_update_dom_ops(info);
1493 if (info->flags & MSI_FLAG_USE_DEF_CHIP_OPS)
1494 pci_msi_domain_update_chip_ops(info);
1496 info->flags |= MSI_FLAG_ACTIVATE_EARLY;
1497 if (IS_ENABLED(CONFIG_GENERIC_IRQ_RESERVATION_MODE))
1498 info->flags |= MSI_FLAG_MUST_REACTIVATE;
1500 /* PCI-MSI is oneshot-safe */
1501 info->chip->flags |= IRQCHIP_ONESHOT_SAFE;
1503 domain = msi_create_irq_domain(fwnode, info, parent);
1507 irq_domain_update_bus_token(domain, DOMAIN_BUS_PCI_MSI);
1510 EXPORT_SYMBOL_GPL(pci_msi_create_irq_domain);
1513 * Users of the generic MSI infrastructure expect a device to have a single ID,
1514 * so with DMA aliases we have to pick the least-worst compromise. Devices with
1515 * DMA phantom functions tend to still emit MSIs from the real function number,
1516 * so we ignore those and only consider topological aliases where either the
1517 * alias device or RID appears on a different bus number. We also make the
1518 * reasonable assumption that bridges are walked in an upstream direction (so
1519 * the last one seen wins), and the much braver assumption that the most likely
1520 * case is that of PCI->PCIe so we should always use the alias RID. This echoes
1521 * the logic from intel_irq_remapping's set_msi_sid(), which presumably works
1522 * well enough in practice; in the face of the horrible PCIe<->PCI-X conditions
1523 * for taking ownership all we can really do is close our eyes and hope...
1525 static int get_msi_id_cb(struct pci_dev *pdev, u16 alias, void *data)
1528 u8 bus = PCI_BUS_NUM(*pa);
1530 if (pdev->bus->number != bus || PCI_BUS_NUM(alias) != bus)
1537 * pci_msi_domain_get_msi_rid - Get the MSI requester id (RID)
1538 * @domain: The interrupt domain
1539 * @pdev: The PCI device.
1541 * The RID for a device is formed from the alias, with a firmware
1542 * supplied mapping applied
1546 u32 pci_msi_domain_get_msi_rid(struct irq_domain *domain, struct pci_dev *pdev)
1548 struct device_node *of_node;
1549 u32 rid = pci_dev_id(pdev);
1551 pci_for_each_dma_alias(pdev, get_msi_id_cb, &rid);
1553 of_node = irq_domain_get_of_node(domain);
1554 rid = of_node ? of_msi_map_rid(&pdev->dev, of_node, rid) :
1555 iort_msi_map_rid(&pdev->dev, rid);
1561 * pci_msi_get_device_domain - Get the MSI domain for a given PCI device
1562 * @pdev: The PCI device
1564 * Use the firmware data to find a device-specific MSI domain
1565 * (i.e. not one that is set as a default).
1567 * Returns: The corresponding MSI domain or NULL if none has been found.
1569 struct irq_domain *pci_msi_get_device_domain(struct pci_dev *pdev)
1571 struct irq_domain *dom;
1572 u32 rid = pci_dev_id(pdev);
1574 pci_for_each_dma_alias(pdev, get_msi_id_cb, &rid);
1575 dom = of_msi_map_get_device_domain(&pdev->dev, rid);
1577 dom = iort_get_device_domain(&pdev->dev, rid);
1580 #endif /* CONFIG_PCI_MSI_IRQ_DOMAIN */