1 // SPDX-License-Identifier: GPL-2.0
3 * PCI Message Signaled Interrupt (MSI)
5 * Copyright (C) 2003-2004 Intel
7 * Copyright (C) 2016 Christoph Hellwig.
10 #include <linux/err.h>
12 #include <linux/irq.h>
13 #include <linux/interrupt.h>
14 #include <linux/export.h>
15 #include <linux/ioport.h>
16 #include <linux/pci.h>
17 #include <linux/proc_fs.h>
18 #include <linux/msi.h>
19 #include <linux/smp.h>
20 #include <linux/errno.h>
22 #include <linux/acpi_iort.h>
23 #include <linux/slab.h>
24 #include <linux/irqdomain.h>
25 #include <linux/of_irq.h>
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 return desc->mask_base +
196 desc->msi_attrib.entry_nr * PCI_MSIX_ENTRY_SIZE;
200 * This internal function does not flush PCI writes to the device.
201 * All users must ensure that they read from the device before either
202 * assuming that the device state is up to date, or returning out of this
203 * file. This saves a few milliseconds when initialising devices with lots
204 * of MSI-X interrupts.
206 u32 __pci_msix_desc_mask_irq(struct msi_desc *desc, u32 flag)
208 u32 mask_bits = desc->masked;
210 if (pci_msi_ignore_mask)
213 mask_bits &= ~PCI_MSIX_ENTRY_CTRL_MASKBIT;
215 mask_bits |= PCI_MSIX_ENTRY_CTRL_MASKBIT;
216 writel(mask_bits, pci_msix_desc_addr(desc) + PCI_MSIX_ENTRY_VECTOR_CTRL);
221 static void msix_mask_irq(struct msi_desc *desc, u32 flag)
223 desc->masked = __pci_msix_desc_mask_irq(desc, flag);
226 static void msi_set_mask_bit(struct irq_data *data, u32 flag)
228 struct msi_desc *desc = irq_data_get_msi_desc(data);
230 if (desc->msi_attrib.is_msix) {
231 msix_mask_irq(desc, flag);
232 readl(desc->mask_base); /* Flush write to device */
234 unsigned offset = data->irq - desc->irq;
235 msi_mask_irq(desc, 1 << offset, flag << offset);
240 * pci_msi_mask_irq - Generic irq chip callback to mask PCI/MSI interrupts
241 * @data: pointer to irqdata associated to that interrupt
243 void pci_msi_mask_irq(struct irq_data *data)
245 msi_set_mask_bit(data, 1);
247 EXPORT_SYMBOL_GPL(pci_msi_mask_irq);
250 * pci_msi_unmask_irq - Generic irq chip callback to unmask PCI/MSI interrupts
251 * @data: pointer to irqdata associated to that interrupt
253 void pci_msi_unmask_irq(struct irq_data *data)
255 msi_set_mask_bit(data, 0);
257 EXPORT_SYMBOL_GPL(pci_msi_unmask_irq);
259 void default_restore_msi_irqs(struct pci_dev *dev)
261 struct msi_desc *entry;
263 for_each_pci_msi_entry(entry, dev)
264 default_restore_msi_irq(dev, entry->irq);
267 void __pci_read_msi_msg(struct msi_desc *entry, struct msi_msg *msg)
269 struct pci_dev *dev = msi_desc_to_pci_dev(entry);
271 BUG_ON(dev->current_state != PCI_D0);
273 if (entry->msi_attrib.is_msix) {
274 void __iomem *base = pci_msix_desc_addr(entry);
276 msg->address_lo = readl(base + PCI_MSIX_ENTRY_LOWER_ADDR);
277 msg->address_hi = readl(base + PCI_MSIX_ENTRY_UPPER_ADDR);
278 msg->data = readl(base + PCI_MSIX_ENTRY_DATA);
280 int pos = dev->msi_cap;
283 pci_read_config_dword(dev, pos + PCI_MSI_ADDRESS_LO,
285 if (entry->msi_attrib.is_64) {
286 pci_read_config_dword(dev, pos + PCI_MSI_ADDRESS_HI,
288 pci_read_config_word(dev, pos + PCI_MSI_DATA_64, &data);
291 pci_read_config_word(dev, pos + PCI_MSI_DATA_32, &data);
297 void __pci_write_msi_msg(struct msi_desc *entry, struct msi_msg *msg)
299 struct pci_dev *dev = msi_desc_to_pci_dev(entry);
301 if (dev->current_state != PCI_D0 || pci_dev_is_disconnected(dev)) {
302 /* Don't touch the hardware now */
303 } else if (entry->msi_attrib.is_msix) {
304 void __iomem *base = pci_msix_desc_addr(entry);
306 writel(msg->address_lo, base + PCI_MSIX_ENTRY_LOWER_ADDR);
307 writel(msg->address_hi, base + PCI_MSIX_ENTRY_UPPER_ADDR);
308 writel(msg->data, base + PCI_MSIX_ENTRY_DATA);
310 int pos = dev->msi_cap;
313 pci_read_config_word(dev, pos + PCI_MSI_FLAGS, &msgctl);
314 msgctl &= ~PCI_MSI_FLAGS_QSIZE;
315 msgctl |= entry->msi_attrib.multiple << 4;
316 pci_write_config_word(dev, pos + PCI_MSI_FLAGS, msgctl);
318 pci_write_config_dword(dev, pos + PCI_MSI_ADDRESS_LO,
320 if (entry->msi_attrib.is_64) {
321 pci_write_config_dword(dev, pos + PCI_MSI_ADDRESS_HI,
323 pci_write_config_word(dev, pos + PCI_MSI_DATA_64,
326 pci_write_config_word(dev, pos + PCI_MSI_DATA_32,
333 void pci_write_msi_msg(unsigned int irq, struct msi_msg *msg)
335 struct msi_desc *entry = irq_get_msi_desc(irq);
337 __pci_write_msi_msg(entry, msg);
339 EXPORT_SYMBOL_GPL(pci_write_msi_msg);
341 static void free_msi_irqs(struct pci_dev *dev)
343 struct list_head *msi_list = dev_to_msi_list(&dev->dev);
344 struct msi_desc *entry, *tmp;
345 struct attribute **msi_attrs;
346 struct device_attribute *dev_attr;
349 for_each_pci_msi_entry(entry, dev)
351 for (i = 0; i < entry->nvec_used; i++)
352 BUG_ON(irq_has_action(entry->irq + i));
354 pci_msi_teardown_msi_irqs(dev);
356 list_for_each_entry_safe(entry, tmp, msi_list, list) {
357 if (entry->msi_attrib.is_msix) {
358 if (list_is_last(&entry->list, msi_list))
359 iounmap(entry->mask_base);
362 list_del(&entry->list);
363 free_msi_entry(entry);
366 if (dev->msi_irq_groups) {
367 sysfs_remove_groups(&dev->dev.kobj, dev->msi_irq_groups);
368 msi_attrs = dev->msi_irq_groups[0]->attrs;
369 while (msi_attrs[count]) {
370 dev_attr = container_of(msi_attrs[count],
371 struct device_attribute, attr);
372 kfree(dev_attr->attr.name);
377 kfree(dev->msi_irq_groups[0]);
378 kfree(dev->msi_irq_groups);
379 dev->msi_irq_groups = NULL;
383 static void pci_intx_for_msi(struct pci_dev *dev, int enable)
385 if (!(dev->dev_flags & PCI_DEV_FLAGS_MSI_INTX_DISABLE_BUG))
386 pci_intx(dev, enable);
389 static void __pci_restore_msi_state(struct pci_dev *dev)
392 struct msi_desc *entry;
394 if (!dev->msi_enabled)
397 entry = irq_get_msi_desc(dev->irq);
399 pci_intx_for_msi(dev, 0);
400 pci_msi_set_enable(dev, 0);
401 arch_restore_msi_irqs(dev);
403 pci_read_config_word(dev, dev->msi_cap + PCI_MSI_FLAGS, &control);
404 msi_mask_irq(entry, msi_mask(entry->msi_attrib.multi_cap),
406 control &= ~PCI_MSI_FLAGS_QSIZE;
407 control |= (entry->msi_attrib.multiple << 4) | PCI_MSI_FLAGS_ENABLE;
408 pci_write_config_word(dev, dev->msi_cap + PCI_MSI_FLAGS, control);
411 static void __pci_restore_msix_state(struct pci_dev *dev)
413 struct msi_desc *entry;
415 if (!dev->msix_enabled)
417 BUG_ON(list_empty(dev_to_msi_list(&dev->dev)));
419 /* route the table */
420 pci_intx_for_msi(dev, 0);
421 pci_msix_clear_and_set_ctrl(dev, 0,
422 PCI_MSIX_FLAGS_ENABLE | PCI_MSIX_FLAGS_MASKALL);
424 arch_restore_msi_irqs(dev);
425 for_each_pci_msi_entry(entry, dev)
426 msix_mask_irq(entry, entry->masked);
428 pci_msix_clear_and_set_ctrl(dev, PCI_MSIX_FLAGS_MASKALL, 0);
431 void pci_restore_msi_state(struct pci_dev *dev)
433 __pci_restore_msi_state(dev);
434 __pci_restore_msix_state(dev);
436 EXPORT_SYMBOL_GPL(pci_restore_msi_state);
438 static ssize_t msi_mode_show(struct device *dev, struct device_attribute *attr,
441 struct msi_desc *entry;
445 retval = kstrtoul(attr->attr.name, 10, &irq);
449 entry = irq_get_msi_desc(irq);
451 return sprintf(buf, "%s\n",
452 entry->msi_attrib.is_msix ? "msix" : "msi");
457 static int populate_msi_sysfs(struct pci_dev *pdev)
459 struct attribute **msi_attrs;
460 struct attribute *msi_attr;
461 struct device_attribute *msi_dev_attr;
462 struct attribute_group *msi_irq_group;
463 const struct attribute_group **msi_irq_groups;
464 struct msi_desc *entry;
470 /* Determine how many msi entries we have */
471 for_each_pci_msi_entry(entry, pdev)
472 num_msi += entry->nvec_used;
476 /* Dynamically create the MSI attributes for the PCI device */
477 msi_attrs = kcalloc(num_msi + 1, sizeof(void *), GFP_KERNEL);
480 for_each_pci_msi_entry(entry, pdev) {
481 for (i = 0; i < entry->nvec_used; i++) {
482 msi_dev_attr = kzalloc(sizeof(*msi_dev_attr), GFP_KERNEL);
485 msi_attrs[count] = &msi_dev_attr->attr;
487 sysfs_attr_init(&msi_dev_attr->attr);
488 msi_dev_attr->attr.name = kasprintf(GFP_KERNEL, "%d",
490 if (!msi_dev_attr->attr.name)
492 msi_dev_attr->attr.mode = S_IRUGO;
493 msi_dev_attr->show = msi_mode_show;
498 msi_irq_group = kzalloc(sizeof(*msi_irq_group), GFP_KERNEL);
501 msi_irq_group->name = "msi_irqs";
502 msi_irq_group->attrs = msi_attrs;
504 msi_irq_groups = kcalloc(2, sizeof(void *), GFP_KERNEL);
506 goto error_irq_group;
507 msi_irq_groups[0] = msi_irq_group;
509 ret = sysfs_create_groups(&pdev->dev.kobj, msi_irq_groups);
511 goto error_irq_groups;
512 pdev->msi_irq_groups = msi_irq_groups;
517 kfree(msi_irq_groups);
519 kfree(msi_irq_group);
522 msi_attr = msi_attrs[count];
524 msi_dev_attr = container_of(msi_attr, struct device_attribute, attr);
525 kfree(msi_attr->name);
528 msi_attr = msi_attrs[count];
534 static struct msi_desc *
535 msi_setup_entry(struct pci_dev *dev, int nvec, const struct irq_affinity *affd)
537 struct cpumask *masks = NULL;
538 struct msi_desc *entry;
542 masks = irq_create_affinity_masks(nvec, affd);
545 /* MSI Entry Initialization */
546 entry = alloc_msi_entry(&dev->dev, nvec, masks);
550 pci_read_config_word(dev, dev->msi_cap + PCI_MSI_FLAGS, &control);
552 entry->msi_attrib.is_msix = 0;
553 entry->msi_attrib.is_64 = !!(control & PCI_MSI_FLAGS_64BIT);
554 entry->msi_attrib.entry_nr = 0;
555 entry->msi_attrib.maskbit = !!(control & PCI_MSI_FLAGS_MASKBIT);
556 entry->msi_attrib.default_irq = dev->irq; /* Save IOAPIC IRQ */
557 entry->msi_attrib.multi_cap = (control & PCI_MSI_FLAGS_QMASK) >> 1;
558 entry->msi_attrib.multiple = ilog2(__roundup_pow_of_two(nvec));
560 if (control & PCI_MSI_FLAGS_64BIT)
561 entry->mask_pos = dev->msi_cap + PCI_MSI_MASK_64;
563 entry->mask_pos = dev->msi_cap + PCI_MSI_MASK_32;
565 /* Save the initial mask status */
566 if (entry->msi_attrib.maskbit)
567 pci_read_config_dword(dev, entry->mask_pos, &entry->masked);
574 static int msi_verify_entries(struct pci_dev *dev)
576 struct msi_desc *entry;
578 for_each_pci_msi_entry(entry, dev) {
579 if (!dev->no_64bit_msi || !entry->msg.address_hi)
581 pci_err(dev, "Device has broken 64-bit MSI but arch"
582 " tried to assign one above 4G\n");
589 * msi_capability_init - configure device's MSI capability structure
590 * @dev: pointer to the pci_dev data structure of MSI device function
591 * @nvec: number of interrupts to allocate
592 * @affd: description of automatic irq affinity assignments (may be %NULL)
594 * Setup the MSI capability structure of the device with the requested
595 * number of interrupts. A return value of zero indicates the successful
596 * setup of an entry with the new MSI irq. A negative return value indicates
597 * an error, and a positive return value indicates the number of interrupts
598 * which could have been allocated.
600 static int msi_capability_init(struct pci_dev *dev, int nvec,
601 const struct irq_affinity *affd)
603 struct msi_desc *entry;
607 pci_msi_set_enable(dev, 0); /* Disable MSI during set up */
609 entry = msi_setup_entry(dev, nvec, affd);
613 /* All MSIs are unmasked by default, Mask them all */
614 mask = msi_mask(entry->msi_attrib.multi_cap);
615 msi_mask_irq(entry, mask, mask);
617 list_add_tail(&entry->list, dev_to_msi_list(&dev->dev));
619 /* Configure MSI capability structure */
620 ret = pci_msi_setup_msi_irqs(dev, nvec, PCI_CAP_ID_MSI);
622 msi_mask_irq(entry, mask, ~mask);
627 ret = msi_verify_entries(dev);
629 msi_mask_irq(entry, mask, ~mask);
634 ret = populate_msi_sysfs(dev);
636 msi_mask_irq(entry, mask, ~mask);
641 /* Set MSI enabled bits */
642 pci_intx_for_msi(dev, 0);
643 pci_msi_set_enable(dev, 1);
644 dev->msi_enabled = 1;
646 pcibios_free_irq(dev);
647 dev->irq = entry->irq;
651 static void __iomem *msix_map_region(struct pci_dev *dev, unsigned nr_entries)
653 resource_size_t phys_addr;
658 pci_read_config_dword(dev, dev->msix_cap + PCI_MSIX_TABLE,
660 bir = (u8)(table_offset & PCI_MSIX_TABLE_BIR);
661 flags = pci_resource_flags(dev, bir);
662 if (!flags || (flags & IORESOURCE_UNSET))
665 table_offset &= PCI_MSIX_TABLE_OFFSET;
666 phys_addr = pci_resource_start(dev, bir) + table_offset;
668 return ioremap_nocache(phys_addr, nr_entries * PCI_MSIX_ENTRY_SIZE);
671 static int msix_setup_entries(struct pci_dev *dev, void __iomem *base,
672 struct msix_entry *entries, int nvec,
673 const struct irq_affinity *affd)
675 struct cpumask *curmsk, *masks = NULL;
676 struct msi_desc *entry;
680 masks = irq_create_affinity_masks(nvec, affd);
682 for (i = 0, curmsk = masks; i < nvec; i++) {
683 entry = alloc_msi_entry(&dev->dev, 1, curmsk);
689 /* No enough memory. Don't try again */
694 entry->msi_attrib.is_msix = 1;
695 entry->msi_attrib.is_64 = 1;
697 entry->msi_attrib.entry_nr = entries[i].entry;
699 entry->msi_attrib.entry_nr = i;
700 entry->msi_attrib.default_irq = dev->irq;
701 entry->mask_base = base;
703 list_add_tail(&entry->list, dev_to_msi_list(&dev->dev));
713 static void msix_program_entries(struct pci_dev *dev,
714 struct msix_entry *entries)
716 struct msi_desc *entry;
719 for_each_pci_msi_entry(entry, dev) {
721 entries[i++].vector = entry->irq;
722 entry->masked = readl(pci_msix_desc_addr(entry) +
723 PCI_MSIX_ENTRY_VECTOR_CTRL);
724 msix_mask_irq(entry, 1);
729 * msix_capability_init - configure device's MSI-X capability
730 * @dev: pointer to the pci_dev data structure of MSI-X device function
731 * @entries: pointer to an array of struct msix_entry entries
732 * @nvec: number of @entries
733 * @affd: Optional pointer to enable automatic affinity assignement
735 * Setup the MSI-X capability structure of device function with a
736 * single MSI-X irq. A return of zero indicates the successful setup of
737 * requested MSI-X entries with allocated irqs or non-zero for otherwise.
739 static int msix_capability_init(struct pci_dev *dev, struct msix_entry *entries,
740 int nvec, const struct irq_affinity *affd)
746 /* Ensure MSI-X is disabled while it is set up */
747 pci_msix_clear_and_set_ctrl(dev, PCI_MSIX_FLAGS_ENABLE, 0);
749 pci_read_config_word(dev, dev->msix_cap + PCI_MSIX_FLAGS, &control);
750 /* Request & Map MSI-X table region */
751 base = msix_map_region(dev, msix_table_size(control));
755 ret = msix_setup_entries(dev, base, entries, nvec, affd);
759 ret = pci_msi_setup_msi_irqs(dev, nvec, PCI_CAP_ID_MSIX);
763 /* Check if all MSI entries honor device restrictions */
764 ret = msi_verify_entries(dev);
769 * Some devices require MSI-X to be enabled before we can touch the
770 * MSI-X registers. We need to mask all the vectors to prevent
771 * interrupts coming in before they're fully set up.
773 pci_msix_clear_and_set_ctrl(dev, 0,
774 PCI_MSIX_FLAGS_MASKALL | PCI_MSIX_FLAGS_ENABLE);
776 msix_program_entries(dev, entries);
778 ret = populate_msi_sysfs(dev);
782 /* Set MSI-X enabled bits and unmask the function */
783 pci_intx_for_msi(dev, 0);
784 dev->msix_enabled = 1;
785 pci_msix_clear_and_set_ctrl(dev, PCI_MSIX_FLAGS_MASKALL, 0);
787 pcibios_free_irq(dev);
793 * If we had some success, report the number of irqs
794 * we succeeded in setting up.
796 struct msi_desc *entry;
799 for_each_pci_msi_entry(entry, dev) {
814 * pci_msi_supported - check whether MSI may be enabled on a device
815 * @dev: pointer to the pci_dev data structure of MSI device function
816 * @nvec: how many MSIs have been requested ?
818 * Look at global flags, the device itself, and its parent buses
819 * to determine if MSI/-X are supported for the device. If MSI/-X is
820 * supported return 1, else return 0.
822 static int pci_msi_supported(struct pci_dev *dev, int nvec)
826 /* MSI must be globally enabled and supported by the device */
830 if (!dev || dev->no_msi || dev->current_state != PCI_D0)
834 * You can't ask to have 0 or less MSIs configured.
836 * b) the list manipulation code assumes nvec >= 1.
842 * Any bridge which does NOT route MSI transactions from its
843 * secondary bus to its primary bus must set NO_MSI flag on
844 * the secondary pci_bus.
845 * We expect only arch-specific PCI host bus controller driver
846 * or quirks for specific PCI bridges to be setting NO_MSI.
848 for (bus = dev->bus; bus; bus = bus->parent)
849 if (bus->bus_flags & PCI_BUS_FLAGS_NO_MSI)
856 * pci_msi_vec_count - Return the number of MSI vectors a device can send
857 * @dev: device to report about
859 * This function returns the number of MSI vectors a device requested via
860 * Multiple Message Capable register. It returns a negative errno if the
861 * device is not capable sending MSI interrupts. Otherwise, the call succeeds
862 * and returns a power of two, up to a maximum of 2^5 (32), according to the
865 int pci_msi_vec_count(struct pci_dev *dev)
873 pci_read_config_word(dev, dev->msi_cap + PCI_MSI_FLAGS, &msgctl);
874 ret = 1 << ((msgctl & PCI_MSI_FLAGS_QMASK) >> 1);
878 EXPORT_SYMBOL(pci_msi_vec_count);
880 static void pci_msi_shutdown(struct pci_dev *dev)
882 struct msi_desc *desc;
885 if (!pci_msi_enable || !dev || !dev->msi_enabled)
888 BUG_ON(list_empty(dev_to_msi_list(&dev->dev)));
889 desc = first_pci_msi_entry(dev);
891 pci_msi_set_enable(dev, 0);
892 pci_intx_for_msi(dev, 1);
893 dev->msi_enabled = 0;
895 /* Return the device with MSI unmasked as initial states */
896 mask = msi_mask(desc->msi_attrib.multi_cap);
897 /* Keep cached state to be restored */
898 __pci_msi_desc_mask_irq(desc, mask, ~mask);
900 /* Restore dev->irq to its default pin-assertion irq */
901 dev->irq = desc->msi_attrib.default_irq;
902 pcibios_alloc_irq(dev);
905 void pci_disable_msi(struct pci_dev *dev)
907 if (!pci_msi_enable || !dev || !dev->msi_enabled)
910 pci_msi_shutdown(dev);
913 EXPORT_SYMBOL(pci_disable_msi);
916 * pci_msix_vec_count - return the number of device's MSI-X table entries
917 * @dev: pointer to the pci_dev data structure of MSI-X device function
918 * This function returns the number of device's MSI-X table entries and
919 * therefore the number of MSI-X vectors device is capable of sending.
920 * It returns a negative errno if the device is not capable of sending MSI-X
923 int pci_msix_vec_count(struct pci_dev *dev)
930 pci_read_config_word(dev, dev->msix_cap + PCI_MSIX_FLAGS, &control);
931 return msix_table_size(control);
933 EXPORT_SYMBOL(pci_msix_vec_count);
935 static int __pci_enable_msix(struct pci_dev *dev, struct msix_entry *entries,
936 int nvec, const struct irq_affinity *affd)
941 if (!pci_msi_supported(dev, nvec))
944 nr_entries = pci_msix_vec_count(dev);
947 if (nvec > nr_entries)
951 /* Check for any invalid entries */
952 for (i = 0; i < nvec; i++) {
953 if (entries[i].entry >= nr_entries)
954 return -EINVAL; /* invalid entry */
955 for (j = i + 1; j < nvec; j++) {
956 if (entries[i].entry == entries[j].entry)
957 return -EINVAL; /* duplicate entry */
961 WARN_ON(!!dev->msix_enabled);
963 /* Check whether driver already requested for MSI irq */
964 if (dev->msi_enabled) {
965 pci_info(dev, "can't enable MSI-X (MSI IRQ already assigned)\n");
968 return msix_capability_init(dev, entries, nvec, affd);
971 static void pci_msix_shutdown(struct pci_dev *dev)
973 struct msi_desc *entry;
975 if (!pci_msi_enable || !dev || !dev->msix_enabled)
978 if (pci_dev_is_disconnected(dev)) {
979 dev->msix_enabled = 0;
983 /* Return the device with MSI-X masked as initial states */
984 for_each_pci_msi_entry(entry, dev) {
985 /* Keep cached states to be restored */
986 __pci_msix_desc_mask_irq(entry, 1);
989 pci_msix_clear_and_set_ctrl(dev, PCI_MSIX_FLAGS_ENABLE, 0);
990 pci_intx_for_msi(dev, 1);
991 dev->msix_enabled = 0;
992 pcibios_alloc_irq(dev);
995 void pci_disable_msix(struct pci_dev *dev)
997 if (!pci_msi_enable || !dev || !dev->msix_enabled)
1000 pci_msix_shutdown(dev);
1003 EXPORT_SYMBOL(pci_disable_msix);
1005 void pci_no_msi(void)
1011 * pci_msi_enabled - is MSI enabled?
1013 * Returns true if MSI has not been disabled by the command-line option
1016 int pci_msi_enabled(void)
1018 return pci_msi_enable;
1020 EXPORT_SYMBOL(pci_msi_enabled);
1022 static int __pci_enable_msi_range(struct pci_dev *dev, int minvec, int maxvec,
1023 const struct irq_affinity *affd)
1028 if (!pci_msi_supported(dev, minvec))
1031 WARN_ON(!!dev->msi_enabled);
1033 /* Check whether driver already requested MSI-X irqs */
1034 if (dev->msix_enabled) {
1035 pci_info(dev, "can't enable MSI (MSI-X already enabled)\n");
1039 if (maxvec < minvec)
1042 nvec = pci_msi_vec_count(dev);
1053 nvec = irq_calc_affinity_vectors(minvec, nvec, affd);
1058 rc = msi_capability_init(dev, nvec, affd);
1071 /* deprecated, don't use */
1072 int pci_enable_msi(struct pci_dev *dev)
1074 int rc = __pci_enable_msi_range(dev, 1, 1, NULL);
1079 EXPORT_SYMBOL(pci_enable_msi);
1081 static int __pci_enable_msix_range(struct pci_dev *dev,
1082 struct msix_entry *entries, int minvec,
1083 int maxvec, const struct irq_affinity *affd)
1085 int rc, nvec = maxvec;
1087 if (maxvec < minvec)
1092 nvec = irq_calc_affinity_vectors(minvec, nvec, affd);
1097 rc = __pci_enable_msix(dev, entries, nvec, affd);
1111 * pci_enable_msix_range - configure device's MSI-X capability structure
1112 * @dev: pointer to the pci_dev data structure of MSI-X device function
1113 * @entries: pointer to an array of MSI-X entries
1114 * @minvec: minimum number of MSI-X irqs requested
1115 * @maxvec: maximum number of MSI-X irqs requested
1117 * Setup the MSI-X capability structure of device function with a maximum
1118 * possible number of interrupts in the range between @minvec and @maxvec
1119 * upon its software driver call to request for MSI-X mode enabled on its
1120 * hardware device function. It returns a negative errno if an error occurs.
1121 * If it succeeds, it returns the actual number of interrupts allocated and
1122 * indicates the successful configuration of MSI-X capability structure
1123 * with new allocated MSI-X interrupts.
1125 int pci_enable_msix_range(struct pci_dev *dev, struct msix_entry *entries,
1126 int minvec, int maxvec)
1128 return __pci_enable_msix_range(dev, entries, minvec, maxvec, NULL);
1130 EXPORT_SYMBOL(pci_enable_msix_range);
1133 * pci_alloc_irq_vectors_affinity - allocate multiple IRQs for a device
1134 * @dev: PCI device to operate on
1135 * @min_vecs: minimum number of vectors required (must be >= 1)
1136 * @max_vecs: maximum (desired) number of vectors
1137 * @flags: flags or quirks for the allocation
1138 * @affd: optional description of the affinity requirements
1140 * Allocate up to @max_vecs interrupt vectors for @dev, using MSI-X or MSI
1141 * vectors if available, and fall back to a single legacy vector
1142 * if neither is available. Return the number of vectors allocated,
1143 * (which might be smaller than @max_vecs) if successful, or a negative
1144 * error code on error. If less than @min_vecs interrupt vectors are
1145 * available for @dev the function will fail with -ENOSPC.
1147 * To get the Linux IRQ number used for a vector that can be passed to
1148 * request_irq() use the pci_irq_vector() helper.
1150 int pci_alloc_irq_vectors_affinity(struct pci_dev *dev, unsigned int min_vecs,
1151 unsigned int max_vecs, unsigned int flags,
1152 const struct irq_affinity *affd)
1154 static const struct irq_affinity msi_default_affd;
1157 if (flags & PCI_IRQ_AFFINITY) {
1159 affd = &msi_default_affd;
1165 if (flags & PCI_IRQ_MSIX) {
1166 vecs = __pci_enable_msix_range(dev, NULL, min_vecs, max_vecs,
1172 if (flags & PCI_IRQ_MSI) {
1173 vecs = __pci_enable_msi_range(dev, min_vecs, max_vecs, affd);
1178 /* use legacy irq if allowed */
1179 if (flags & PCI_IRQ_LEGACY) {
1180 if (min_vecs == 1 && dev->irq) {
1188 EXPORT_SYMBOL(pci_alloc_irq_vectors_affinity);
1191 * pci_free_irq_vectors - free previously allocated IRQs for a device
1192 * @dev: PCI device to operate on
1194 * Undoes the allocations and enabling in pci_alloc_irq_vectors().
1196 void pci_free_irq_vectors(struct pci_dev *dev)
1198 pci_disable_msix(dev);
1199 pci_disable_msi(dev);
1201 EXPORT_SYMBOL(pci_free_irq_vectors);
1204 * pci_irq_vector - return Linux IRQ number of a device vector
1205 * @dev: PCI device to operate on
1206 * @nr: device-relative interrupt vector index (0-based).
1208 int pci_irq_vector(struct pci_dev *dev, unsigned int nr)
1210 if (dev->msix_enabled) {
1211 struct msi_desc *entry;
1214 for_each_pci_msi_entry(entry, dev) {
1223 if (dev->msi_enabled) {
1224 struct msi_desc *entry = first_pci_msi_entry(dev);
1226 if (WARN_ON_ONCE(nr >= entry->nvec_used))
1229 if (WARN_ON_ONCE(nr > 0))
1233 return dev->irq + nr;
1235 EXPORT_SYMBOL(pci_irq_vector);
1238 * pci_irq_get_affinity - return the affinity of a particular msi vector
1239 * @dev: PCI device to operate on
1240 * @nr: device-relative interrupt vector index (0-based).
1242 const struct cpumask *pci_irq_get_affinity(struct pci_dev *dev, int nr)
1244 if (dev->msix_enabled) {
1245 struct msi_desc *entry;
1248 for_each_pci_msi_entry(entry, dev) {
1250 return entry->affinity;
1255 } else if (dev->msi_enabled) {
1256 struct msi_desc *entry = first_pci_msi_entry(dev);
1258 if (WARN_ON_ONCE(!entry || !entry->affinity ||
1259 nr >= entry->nvec_used))
1262 return &entry->affinity[nr];
1264 return cpu_possible_mask;
1267 EXPORT_SYMBOL(pci_irq_get_affinity);
1270 * pci_irq_get_node - return the numa node of a particular msi vector
1271 * @pdev: PCI device to operate on
1272 * @vec: device-relative interrupt vector index (0-based).
1274 int pci_irq_get_node(struct pci_dev *pdev, int vec)
1276 const struct cpumask *mask;
1278 mask = pci_irq_get_affinity(pdev, vec);
1280 return local_memory_node(cpu_to_node(cpumask_first(mask)));
1281 return dev_to_node(&pdev->dev);
1283 EXPORT_SYMBOL(pci_irq_get_node);
1285 struct pci_dev *msi_desc_to_pci_dev(struct msi_desc *desc)
1287 return to_pci_dev(desc->dev);
1289 EXPORT_SYMBOL(msi_desc_to_pci_dev);
1291 void *msi_desc_to_pci_sysdata(struct msi_desc *desc)
1293 struct pci_dev *dev = msi_desc_to_pci_dev(desc);
1295 return dev->bus->sysdata;
1297 EXPORT_SYMBOL_GPL(msi_desc_to_pci_sysdata);
1299 #ifdef CONFIG_PCI_MSI_IRQ_DOMAIN
1301 * pci_msi_domain_write_msg - Helper to write MSI message to PCI config space
1302 * @irq_data: Pointer to interrupt data of the MSI interrupt
1303 * @msg: Pointer to the message
1305 void pci_msi_domain_write_msg(struct irq_data *irq_data, struct msi_msg *msg)
1307 struct msi_desc *desc = irq_data_get_msi_desc(irq_data);
1310 * For MSI-X desc->irq is always equal to irq_data->irq. For
1311 * MSI only the first interrupt of MULTI MSI passes the test.
1313 if (desc->irq == irq_data->irq)
1314 __pci_write_msi_msg(desc, msg);
1318 * pci_msi_domain_calc_hwirq - Generate a unique ID for an MSI source
1319 * @dev: Pointer to the PCI device
1320 * @desc: Pointer to the msi descriptor
1322 * The ID number is only used within the irqdomain.
1324 irq_hw_number_t pci_msi_domain_calc_hwirq(struct pci_dev *dev,
1325 struct msi_desc *desc)
1327 return (irq_hw_number_t)desc->msi_attrib.entry_nr |
1328 PCI_DEVID(dev->bus->number, dev->devfn) << 11 |
1329 (pci_domain_nr(dev->bus) & 0xFFFFFFFF) << 27;
1332 static inline bool pci_msi_desc_is_multi_msi(struct msi_desc *desc)
1334 return !desc->msi_attrib.is_msix && desc->nvec_used > 1;
1338 * pci_msi_domain_check_cap - Verify that @domain supports the capabilities for @dev
1339 * @domain: The interrupt domain to check
1340 * @info: The domain info for verification
1341 * @dev: The device to check
1344 * 0 if the functionality is supported
1345 * 1 if Multi MSI is requested, but the domain does not support it
1346 * -ENOTSUPP otherwise
1348 int pci_msi_domain_check_cap(struct irq_domain *domain,
1349 struct msi_domain_info *info, struct device *dev)
1351 struct msi_desc *desc = first_pci_msi_entry(to_pci_dev(dev));
1353 /* Special handling to support __pci_enable_msi_range() */
1354 if (pci_msi_desc_is_multi_msi(desc) &&
1355 !(info->flags & MSI_FLAG_MULTI_PCI_MSI))
1357 else if (desc->msi_attrib.is_msix && !(info->flags & MSI_FLAG_PCI_MSIX))
1363 static int pci_msi_domain_handle_error(struct irq_domain *domain,
1364 struct msi_desc *desc, int error)
1366 /* Special handling to support __pci_enable_msi_range() */
1367 if (pci_msi_desc_is_multi_msi(desc) && error == -ENOSPC)
1373 #ifdef GENERIC_MSI_DOMAIN_OPS
1374 static void pci_msi_domain_set_desc(msi_alloc_info_t *arg,
1375 struct msi_desc *desc)
1378 arg->hwirq = pci_msi_domain_calc_hwirq(msi_desc_to_pci_dev(desc),
1382 #define pci_msi_domain_set_desc NULL
1385 static struct msi_domain_ops pci_msi_domain_ops_default = {
1386 .set_desc = pci_msi_domain_set_desc,
1387 .msi_check = pci_msi_domain_check_cap,
1388 .handle_error = pci_msi_domain_handle_error,
1391 static void pci_msi_domain_update_dom_ops(struct msi_domain_info *info)
1393 struct msi_domain_ops *ops = info->ops;
1396 info->ops = &pci_msi_domain_ops_default;
1398 if (ops->set_desc == NULL)
1399 ops->set_desc = pci_msi_domain_set_desc;
1400 if (ops->msi_check == NULL)
1401 ops->msi_check = pci_msi_domain_check_cap;
1402 if (ops->handle_error == NULL)
1403 ops->handle_error = pci_msi_domain_handle_error;
1407 static void pci_msi_domain_update_chip_ops(struct msi_domain_info *info)
1409 struct irq_chip *chip = info->chip;
1412 if (!chip->irq_write_msi_msg)
1413 chip->irq_write_msi_msg = pci_msi_domain_write_msg;
1414 if (!chip->irq_mask)
1415 chip->irq_mask = pci_msi_mask_irq;
1416 if (!chip->irq_unmask)
1417 chip->irq_unmask = pci_msi_unmask_irq;
1421 * pci_msi_create_irq_domain - Create a MSI interrupt domain
1422 * @fwnode: Optional fwnode of the interrupt controller
1423 * @info: MSI domain info
1424 * @parent: Parent irq domain
1426 * Updates the domain and chip ops and creates a MSI interrupt domain.
1429 * A domain pointer or NULL in case of failure.
1431 struct irq_domain *pci_msi_create_irq_domain(struct fwnode_handle *fwnode,
1432 struct msi_domain_info *info,
1433 struct irq_domain *parent)
1435 struct irq_domain *domain;
1437 if (WARN_ON(info->flags & MSI_FLAG_LEVEL_CAPABLE))
1438 info->flags &= ~MSI_FLAG_LEVEL_CAPABLE;
1440 if (info->flags & MSI_FLAG_USE_DEF_DOM_OPS)
1441 pci_msi_domain_update_dom_ops(info);
1442 if (info->flags & MSI_FLAG_USE_DEF_CHIP_OPS)
1443 pci_msi_domain_update_chip_ops(info);
1445 info->flags |= MSI_FLAG_ACTIVATE_EARLY;
1446 if (IS_ENABLED(CONFIG_GENERIC_IRQ_RESERVATION_MODE))
1447 info->flags |= MSI_FLAG_MUST_REACTIVATE;
1449 /* PCI-MSI is oneshot-safe */
1450 info->chip->flags |= IRQCHIP_ONESHOT_SAFE;
1452 domain = msi_create_irq_domain(fwnode, info, parent);
1456 irq_domain_update_bus_token(domain, DOMAIN_BUS_PCI_MSI);
1459 EXPORT_SYMBOL_GPL(pci_msi_create_irq_domain);
1462 * Users of the generic MSI infrastructure expect a device to have a single ID,
1463 * so with DMA aliases we have to pick the least-worst compromise. Devices with
1464 * DMA phantom functions tend to still emit MSIs from the real function number,
1465 * so we ignore those and only consider topological aliases where either the
1466 * alias device or RID appears on a different bus number. We also make the
1467 * reasonable assumption that bridges are walked in an upstream direction (so
1468 * the last one seen wins), and the much braver assumption that the most likely
1469 * case is that of PCI->PCIe so we should always use the alias RID. This echoes
1470 * the logic from intel_irq_remapping's set_msi_sid(), which presumably works
1471 * well enough in practice; in the face of the horrible PCIe<->PCI-X conditions
1472 * for taking ownership all we can really do is close our eyes and hope...
1474 static int get_msi_id_cb(struct pci_dev *pdev, u16 alias, void *data)
1477 u8 bus = PCI_BUS_NUM(*pa);
1479 if (pdev->bus->number != bus || PCI_BUS_NUM(alias) != bus)
1486 * pci_msi_domain_get_msi_rid - Get the MSI requester id (RID)
1487 * @domain: The interrupt domain
1488 * @pdev: The PCI device.
1490 * The RID for a device is formed from the alias, with a firmware
1491 * supplied mapping applied
1495 u32 pci_msi_domain_get_msi_rid(struct irq_domain *domain, struct pci_dev *pdev)
1497 struct device_node *of_node;
1498 u32 rid = PCI_DEVID(pdev->bus->number, pdev->devfn);
1500 pci_for_each_dma_alias(pdev, get_msi_id_cb, &rid);
1502 of_node = irq_domain_get_of_node(domain);
1503 rid = of_node ? of_msi_map_rid(&pdev->dev, of_node, rid) :
1504 iort_msi_map_rid(&pdev->dev, rid);
1510 * pci_msi_get_device_domain - Get the MSI domain for a given PCI device
1511 * @pdev: The PCI device
1513 * Use the firmware data to find a device-specific MSI domain
1514 * (i.e. not one that is set as a default).
1516 * Returns: The corresponding MSI domain or NULL if none has been found.
1518 struct irq_domain *pci_msi_get_device_domain(struct pci_dev *pdev)
1520 struct irq_domain *dom;
1521 u32 rid = PCI_DEVID(pdev->bus->number, pdev->devfn);
1523 pci_for_each_dma_alias(pdev, get_msi_id_cb, &rid);
1524 dom = of_msi_map_get_device_domain(&pdev->dev, rid);
1526 dom = iort_get_device_domain(&pdev->dev, rid);
1529 #endif /* CONFIG_PCI_MSI_IRQ_DOMAIN */