3 * Purpose: 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 = kzalloc(sizeof(void *) * (num_msi + 1), 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 = kzalloc(sizeof(void *) * 2, 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);
544 dev_err(&dev->dev, "can't allocate MSI affinity masks for %d vectors\n",
548 /* MSI Entry Initialization */
549 entry = alloc_msi_entry(&dev->dev, nvec, masks);
553 pci_read_config_word(dev, dev->msi_cap + PCI_MSI_FLAGS, &control);
555 entry->msi_attrib.is_msix = 0;
556 entry->msi_attrib.is_64 = !!(control & PCI_MSI_FLAGS_64BIT);
557 entry->msi_attrib.entry_nr = 0;
558 entry->msi_attrib.maskbit = !!(control & PCI_MSI_FLAGS_MASKBIT);
559 entry->msi_attrib.default_irq = dev->irq; /* Save IOAPIC IRQ */
560 entry->msi_attrib.multi_cap = (control & PCI_MSI_FLAGS_QMASK) >> 1;
561 entry->msi_attrib.multiple = ilog2(__roundup_pow_of_two(nvec));
563 if (control & PCI_MSI_FLAGS_64BIT)
564 entry->mask_pos = dev->msi_cap + PCI_MSI_MASK_64;
566 entry->mask_pos = dev->msi_cap + PCI_MSI_MASK_32;
568 /* Save the initial mask status */
569 if (entry->msi_attrib.maskbit)
570 pci_read_config_dword(dev, entry->mask_pos, &entry->masked);
577 static int msi_verify_entries(struct pci_dev *dev)
579 struct msi_desc *entry;
581 for_each_pci_msi_entry(entry, dev) {
582 if (!dev->no_64bit_msi || !entry->msg.address_hi)
584 dev_err(&dev->dev, "Device has broken 64-bit MSI but arch"
585 " tried to assign one above 4G\n");
592 * msi_capability_init - configure device's MSI capability structure
593 * @dev: pointer to the pci_dev data structure of MSI device function
594 * @nvec: number of interrupts to allocate
595 * @affd: description of automatic irq affinity assignments (may be %NULL)
597 * Setup the MSI capability structure of the device with the requested
598 * number of interrupts. A return value of zero indicates the successful
599 * setup of an entry with the new MSI irq. A negative return value indicates
600 * an error, and a positive return value indicates the number of interrupts
601 * which could have been allocated.
603 static int msi_capability_init(struct pci_dev *dev, int nvec,
604 const struct irq_affinity *affd)
606 struct msi_desc *entry;
610 pci_msi_set_enable(dev, 0); /* Disable MSI during set up */
612 entry = msi_setup_entry(dev, nvec, affd);
616 /* All MSIs are unmasked by default, Mask them all */
617 mask = msi_mask(entry->msi_attrib.multi_cap);
618 msi_mask_irq(entry, mask, mask);
620 list_add_tail(&entry->list, dev_to_msi_list(&dev->dev));
622 /* Configure MSI capability structure */
623 ret = pci_msi_setup_msi_irqs(dev, nvec, PCI_CAP_ID_MSI);
625 msi_mask_irq(entry, mask, ~mask);
630 ret = msi_verify_entries(dev);
632 msi_mask_irq(entry, mask, ~mask);
637 ret = populate_msi_sysfs(dev);
639 msi_mask_irq(entry, mask, ~mask);
644 /* Set MSI enabled bits */
645 pci_intx_for_msi(dev, 0);
646 pci_msi_set_enable(dev, 1);
647 dev->msi_enabled = 1;
649 pcibios_free_irq(dev);
650 dev->irq = entry->irq;
654 static void __iomem *msix_map_region(struct pci_dev *dev, unsigned nr_entries)
656 resource_size_t phys_addr;
661 pci_read_config_dword(dev, dev->msix_cap + PCI_MSIX_TABLE,
663 bir = (u8)(table_offset & PCI_MSIX_TABLE_BIR);
664 flags = pci_resource_flags(dev, bir);
665 if (!flags || (flags & IORESOURCE_UNSET))
668 table_offset &= PCI_MSIX_TABLE_OFFSET;
669 phys_addr = pci_resource_start(dev, bir) + table_offset;
671 return ioremap_nocache(phys_addr, nr_entries * PCI_MSIX_ENTRY_SIZE);
674 static int msix_setup_entries(struct pci_dev *dev, void __iomem *base,
675 struct msix_entry *entries, int nvec,
676 const struct irq_affinity *affd)
678 struct cpumask *curmsk, *masks = NULL;
679 struct msi_desc *entry;
683 masks = irq_create_affinity_masks(nvec, affd);
685 dev_err(&dev->dev, "can't allocate MSI-X affinity masks for %d vectors\n",
689 for (i = 0, curmsk = masks; i < nvec; i++) {
690 entry = alloc_msi_entry(&dev->dev, 1, curmsk);
696 /* No enough memory. Don't try again */
701 entry->msi_attrib.is_msix = 1;
702 entry->msi_attrib.is_64 = 1;
704 entry->msi_attrib.entry_nr = entries[i].entry;
706 entry->msi_attrib.entry_nr = i;
707 entry->msi_attrib.default_irq = dev->irq;
708 entry->mask_base = base;
710 list_add_tail(&entry->list, dev_to_msi_list(&dev->dev));
720 static void msix_program_entries(struct pci_dev *dev,
721 struct msix_entry *entries)
723 struct msi_desc *entry;
726 for_each_pci_msi_entry(entry, dev) {
728 entries[i++].vector = entry->irq;
729 entry->masked = readl(pci_msix_desc_addr(entry) +
730 PCI_MSIX_ENTRY_VECTOR_CTRL);
731 msix_mask_irq(entry, 1);
736 * msix_capability_init - configure device's MSI-X capability
737 * @dev: pointer to the pci_dev data structure of MSI-X device function
738 * @entries: pointer to an array of struct msix_entry entries
739 * @nvec: number of @entries
740 * @affd: Optional pointer to enable automatic affinity assignement
742 * Setup the MSI-X capability structure of device function with a
743 * single MSI-X irq. A return of zero indicates the successful setup of
744 * requested MSI-X entries with allocated irqs or non-zero for otherwise.
746 static int msix_capability_init(struct pci_dev *dev, struct msix_entry *entries,
747 int nvec, const struct irq_affinity *affd)
753 /* Ensure MSI-X is disabled while it is set up */
754 pci_msix_clear_and_set_ctrl(dev, PCI_MSIX_FLAGS_ENABLE, 0);
756 pci_read_config_word(dev, dev->msix_cap + PCI_MSIX_FLAGS, &control);
757 /* Request & Map MSI-X table region */
758 base = msix_map_region(dev, msix_table_size(control));
762 ret = msix_setup_entries(dev, base, entries, nvec, affd);
766 ret = pci_msi_setup_msi_irqs(dev, nvec, PCI_CAP_ID_MSIX);
770 /* Check if all MSI entries honor device restrictions */
771 ret = msi_verify_entries(dev);
776 * Some devices require MSI-X to be enabled before we can touch the
777 * MSI-X registers. We need to mask all the vectors to prevent
778 * interrupts coming in before they're fully set up.
780 pci_msix_clear_and_set_ctrl(dev, 0,
781 PCI_MSIX_FLAGS_MASKALL | PCI_MSIX_FLAGS_ENABLE);
783 msix_program_entries(dev, entries);
785 ret = populate_msi_sysfs(dev);
789 /* Set MSI-X enabled bits and unmask the function */
790 pci_intx_for_msi(dev, 0);
791 dev->msix_enabled = 1;
792 pci_msix_clear_and_set_ctrl(dev, PCI_MSIX_FLAGS_MASKALL, 0);
794 pcibios_free_irq(dev);
800 * If we had some success, report the number of irqs
801 * we succeeded in setting up.
803 struct msi_desc *entry;
806 for_each_pci_msi_entry(entry, dev) {
821 * pci_msi_supported - check whether MSI may be enabled on a device
822 * @dev: pointer to the pci_dev data structure of MSI device function
823 * @nvec: how many MSIs have been requested ?
825 * Look at global flags, the device itself, and its parent buses
826 * to determine if MSI/-X are supported for the device. If MSI/-X is
827 * supported return 1, else return 0.
829 static int pci_msi_supported(struct pci_dev *dev, int nvec)
833 /* MSI must be globally enabled and supported by the device */
837 if (!dev || dev->no_msi || dev->current_state != PCI_D0)
841 * You can't ask to have 0 or less MSIs configured.
843 * b) the list manipulation code assumes nvec >= 1.
849 * Any bridge which does NOT route MSI transactions from its
850 * secondary bus to its primary bus must set NO_MSI flag on
851 * the secondary pci_bus.
852 * We expect only arch-specific PCI host bus controller driver
853 * or quirks for specific PCI bridges to be setting NO_MSI.
855 for (bus = dev->bus; bus; bus = bus->parent)
856 if (bus->bus_flags & PCI_BUS_FLAGS_NO_MSI)
863 * pci_msi_vec_count - Return the number of MSI vectors a device can send
864 * @dev: device to report about
866 * This function returns the number of MSI vectors a device requested via
867 * Multiple Message Capable register. It returns a negative errno if the
868 * device is not capable sending MSI interrupts. Otherwise, the call succeeds
869 * and returns a power of two, up to a maximum of 2^5 (32), according to the
872 int pci_msi_vec_count(struct pci_dev *dev)
880 pci_read_config_word(dev, dev->msi_cap + PCI_MSI_FLAGS, &msgctl);
881 ret = 1 << ((msgctl & PCI_MSI_FLAGS_QMASK) >> 1);
885 EXPORT_SYMBOL(pci_msi_vec_count);
887 static void pci_msi_shutdown(struct pci_dev *dev)
889 struct msi_desc *desc;
892 if (!pci_msi_enable || !dev || !dev->msi_enabled)
895 BUG_ON(list_empty(dev_to_msi_list(&dev->dev)));
896 desc = first_pci_msi_entry(dev);
898 pci_msi_set_enable(dev, 0);
899 pci_intx_for_msi(dev, 1);
900 dev->msi_enabled = 0;
902 /* Return the device with MSI unmasked as initial states */
903 mask = msi_mask(desc->msi_attrib.multi_cap);
904 /* Keep cached state to be restored */
905 __pci_msi_desc_mask_irq(desc, mask, ~mask);
907 /* Restore dev->irq to its default pin-assertion irq */
908 dev->irq = desc->msi_attrib.default_irq;
909 pcibios_alloc_irq(dev);
912 void pci_disable_msi(struct pci_dev *dev)
914 if (!pci_msi_enable || !dev || !dev->msi_enabled)
917 pci_msi_shutdown(dev);
920 EXPORT_SYMBOL(pci_disable_msi);
923 * pci_msix_vec_count - return the number of device's MSI-X table entries
924 * @dev: pointer to the pci_dev data structure of MSI-X device function
925 * This function returns the number of device's MSI-X table entries and
926 * therefore the number of MSI-X vectors device is capable of sending.
927 * It returns a negative errno if the device is not capable of sending MSI-X
930 int pci_msix_vec_count(struct pci_dev *dev)
937 pci_read_config_word(dev, dev->msix_cap + PCI_MSIX_FLAGS, &control);
938 return msix_table_size(control);
940 EXPORT_SYMBOL(pci_msix_vec_count);
942 static int __pci_enable_msix(struct pci_dev *dev, struct msix_entry *entries,
943 int nvec, const struct irq_affinity *affd)
948 if (!pci_msi_supported(dev, nvec))
951 nr_entries = pci_msix_vec_count(dev);
954 if (nvec > nr_entries)
958 /* Check for any invalid entries */
959 for (i = 0; i < nvec; i++) {
960 if (entries[i].entry >= nr_entries)
961 return -EINVAL; /* invalid entry */
962 for (j = i + 1; j < nvec; j++) {
963 if (entries[i].entry == entries[j].entry)
964 return -EINVAL; /* duplicate entry */
968 WARN_ON(!!dev->msix_enabled);
970 /* Check whether driver already requested for MSI irq */
971 if (dev->msi_enabled) {
972 dev_info(&dev->dev, "can't enable MSI-X (MSI IRQ already assigned)\n");
975 return msix_capability_init(dev, entries, nvec, affd);
978 static void pci_msix_shutdown(struct pci_dev *dev)
980 struct msi_desc *entry;
982 if (!pci_msi_enable || !dev || !dev->msix_enabled)
985 if (pci_dev_is_disconnected(dev)) {
986 dev->msix_enabled = 0;
990 /* Return the device with MSI-X masked as initial states */
991 for_each_pci_msi_entry(entry, dev) {
992 /* Keep cached states to be restored */
993 __pci_msix_desc_mask_irq(entry, 1);
996 pci_msix_clear_and_set_ctrl(dev, PCI_MSIX_FLAGS_ENABLE, 0);
997 pci_intx_for_msi(dev, 1);
998 dev->msix_enabled = 0;
999 pcibios_alloc_irq(dev);
1002 void pci_disable_msix(struct pci_dev *dev)
1004 if (!pci_msi_enable || !dev || !dev->msix_enabled)
1007 pci_msix_shutdown(dev);
1010 EXPORT_SYMBOL(pci_disable_msix);
1012 void pci_no_msi(void)
1018 * pci_msi_enabled - is MSI enabled?
1020 * Returns true if MSI has not been disabled by the command-line option
1023 int pci_msi_enabled(void)
1025 return pci_msi_enable;
1027 EXPORT_SYMBOL(pci_msi_enabled);
1029 static int __pci_enable_msi_range(struct pci_dev *dev, int minvec, int maxvec,
1030 const struct irq_affinity *affd)
1035 if (!pci_msi_supported(dev, minvec))
1038 WARN_ON(!!dev->msi_enabled);
1040 /* Check whether driver already requested MSI-X irqs */
1041 if (dev->msix_enabled) {
1043 "can't enable MSI (MSI-X already enabled)\n");
1047 if (maxvec < minvec)
1050 nvec = pci_msi_vec_count(dev);
1061 nvec = irq_calc_affinity_vectors(nvec, affd);
1066 rc = msi_capability_init(dev, nvec, affd);
1079 /* deprecated, don't use */
1080 int pci_enable_msi(struct pci_dev *dev)
1082 int rc = __pci_enable_msi_range(dev, 1, 1, NULL);
1087 EXPORT_SYMBOL(pci_enable_msi);
1089 static int __pci_enable_msix_range(struct pci_dev *dev,
1090 struct msix_entry *entries, int minvec,
1091 int maxvec, const struct irq_affinity *affd)
1093 int rc, nvec = maxvec;
1095 if (maxvec < minvec)
1100 nvec = irq_calc_affinity_vectors(nvec, affd);
1105 rc = __pci_enable_msix(dev, entries, nvec, affd);
1119 * pci_enable_msix_range - configure device's MSI-X capability structure
1120 * @dev: pointer to the pci_dev data structure of MSI-X device function
1121 * @entries: pointer to an array of MSI-X entries
1122 * @minvec: minimum number of MSI-X irqs requested
1123 * @maxvec: maximum number of MSI-X irqs requested
1125 * Setup the MSI-X capability structure of device function with a maximum
1126 * possible number of interrupts in the range between @minvec and @maxvec
1127 * upon its software driver call to request for MSI-X mode enabled on its
1128 * hardware device function. It returns a negative errno if an error occurs.
1129 * If it succeeds, it returns the actual number of interrupts allocated and
1130 * indicates the successful configuration of MSI-X capability structure
1131 * with new allocated MSI-X interrupts.
1133 int pci_enable_msix_range(struct pci_dev *dev, struct msix_entry *entries,
1134 int minvec, int maxvec)
1136 return __pci_enable_msix_range(dev, entries, minvec, maxvec, NULL);
1138 EXPORT_SYMBOL(pci_enable_msix_range);
1141 * pci_alloc_irq_vectors_affinity - allocate multiple IRQs for a device
1142 * @dev: PCI device to operate on
1143 * @min_vecs: minimum number of vectors required (must be >= 1)
1144 * @max_vecs: maximum (desired) number of vectors
1145 * @flags: flags or quirks for the allocation
1146 * @affd: optional description of the affinity requirements
1148 * Allocate up to @max_vecs interrupt vectors for @dev, using MSI-X or MSI
1149 * vectors if available, and fall back to a single legacy vector
1150 * if neither is available. Return the number of vectors allocated,
1151 * (which might be smaller than @max_vecs) if successful, or a negative
1152 * error code on error. If less than @min_vecs interrupt vectors are
1153 * available for @dev the function will fail with -ENOSPC.
1155 * To get the Linux IRQ number used for a vector that can be passed to
1156 * request_irq() use the pci_irq_vector() helper.
1158 int pci_alloc_irq_vectors_affinity(struct pci_dev *dev, unsigned int min_vecs,
1159 unsigned int max_vecs, unsigned int flags,
1160 const struct irq_affinity *affd)
1162 static const struct irq_affinity msi_default_affd;
1165 if (flags & PCI_IRQ_AFFINITY) {
1167 affd = &msi_default_affd;
1169 if (affd->pre_vectors + affd->post_vectors > min_vecs)
1173 * If there aren't any vectors left after applying the pre/post
1174 * vectors don't bother with assigning affinity.
1176 if (affd->pre_vectors + affd->post_vectors == min_vecs)
1183 if (flags & PCI_IRQ_MSIX) {
1184 vecs = __pci_enable_msix_range(dev, NULL, min_vecs, max_vecs,
1190 if (flags & PCI_IRQ_MSI) {
1191 vecs = __pci_enable_msi_range(dev, min_vecs, max_vecs, affd);
1196 /* use legacy irq if allowed */
1197 if (flags & PCI_IRQ_LEGACY) {
1198 if (min_vecs == 1 && dev->irq) {
1206 EXPORT_SYMBOL(pci_alloc_irq_vectors_affinity);
1209 * pci_free_irq_vectors - free previously allocated IRQs for a device
1210 * @dev: PCI device to operate on
1212 * Undoes the allocations and enabling in pci_alloc_irq_vectors().
1214 void pci_free_irq_vectors(struct pci_dev *dev)
1216 pci_disable_msix(dev);
1217 pci_disable_msi(dev);
1219 EXPORT_SYMBOL(pci_free_irq_vectors);
1222 * pci_irq_vector - return Linux IRQ number of a device vector
1223 * @dev: PCI device to operate on
1224 * @nr: device-relative interrupt vector index (0-based).
1226 int pci_irq_vector(struct pci_dev *dev, unsigned int nr)
1228 if (dev->msix_enabled) {
1229 struct msi_desc *entry;
1232 for_each_pci_msi_entry(entry, dev) {
1241 if (dev->msi_enabled) {
1242 struct msi_desc *entry = first_pci_msi_entry(dev);
1244 if (WARN_ON_ONCE(nr >= entry->nvec_used))
1247 if (WARN_ON_ONCE(nr > 0))
1251 return dev->irq + nr;
1253 EXPORT_SYMBOL(pci_irq_vector);
1256 * pci_irq_get_affinity - return the affinity of a particular msi vector
1257 * @dev: PCI device to operate on
1258 * @nr: device-relative interrupt vector index (0-based).
1260 const struct cpumask *pci_irq_get_affinity(struct pci_dev *dev, int nr)
1262 if (dev->msix_enabled) {
1263 struct msi_desc *entry;
1266 for_each_pci_msi_entry(entry, dev) {
1268 return entry->affinity;
1273 } else if (dev->msi_enabled) {
1274 struct msi_desc *entry = first_pci_msi_entry(dev);
1276 if (WARN_ON_ONCE(!entry || !entry->affinity ||
1277 nr >= entry->nvec_used))
1280 return &entry->affinity[nr];
1282 return cpu_possible_mask;
1285 EXPORT_SYMBOL(pci_irq_get_affinity);
1288 * pci_irq_get_node - return the numa node of a particular msi vector
1289 * @pdev: PCI device to operate on
1290 * @vec: device-relative interrupt vector index (0-based).
1292 int pci_irq_get_node(struct pci_dev *pdev, int vec)
1294 const struct cpumask *mask;
1296 mask = pci_irq_get_affinity(pdev, vec);
1298 return local_memory_node(cpu_to_node(cpumask_first(mask)));
1299 return dev_to_node(&pdev->dev);
1301 EXPORT_SYMBOL(pci_irq_get_node);
1303 struct pci_dev *msi_desc_to_pci_dev(struct msi_desc *desc)
1305 return to_pci_dev(desc->dev);
1307 EXPORT_SYMBOL(msi_desc_to_pci_dev);
1309 void *msi_desc_to_pci_sysdata(struct msi_desc *desc)
1311 struct pci_dev *dev = msi_desc_to_pci_dev(desc);
1313 return dev->bus->sysdata;
1315 EXPORT_SYMBOL_GPL(msi_desc_to_pci_sysdata);
1317 #ifdef CONFIG_PCI_MSI_IRQ_DOMAIN
1319 * pci_msi_domain_write_msg - Helper to write MSI message to PCI config space
1320 * @irq_data: Pointer to interrupt data of the MSI interrupt
1321 * @msg: Pointer to the message
1323 void pci_msi_domain_write_msg(struct irq_data *irq_data, struct msi_msg *msg)
1325 struct msi_desc *desc = irq_data_get_msi_desc(irq_data);
1328 * For MSI-X desc->irq is always equal to irq_data->irq. For
1329 * MSI only the first interrupt of MULTI MSI passes the test.
1331 if (desc->irq == irq_data->irq)
1332 __pci_write_msi_msg(desc, msg);
1336 * pci_msi_domain_calc_hwirq - Generate a unique ID for an MSI source
1337 * @dev: Pointer to the PCI device
1338 * @desc: Pointer to the msi descriptor
1340 * The ID number is only used within the irqdomain.
1342 irq_hw_number_t pci_msi_domain_calc_hwirq(struct pci_dev *dev,
1343 struct msi_desc *desc)
1345 return (irq_hw_number_t)desc->msi_attrib.entry_nr |
1346 PCI_DEVID(dev->bus->number, dev->devfn) << 11 |
1347 (pci_domain_nr(dev->bus) & 0xFFFFFFFF) << 27;
1350 static inline bool pci_msi_desc_is_multi_msi(struct msi_desc *desc)
1352 return !desc->msi_attrib.is_msix && desc->nvec_used > 1;
1356 * pci_msi_domain_check_cap - Verify that @domain supports the capabilities for @dev
1357 * @domain: The interrupt domain to check
1358 * @info: The domain info for verification
1359 * @dev: The device to check
1362 * 0 if the functionality is supported
1363 * 1 if Multi MSI is requested, but the domain does not support it
1364 * -ENOTSUPP otherwise
1366 int pci_msi_domain_check_cap(struct irq_domain *domain,
1367 struct msi_domain_info *info, struct device *dev)
1369 struct msi_desc *desc = first_pci_msi_entry(to_pci_dev(dev));
1371 /* Special handling to support __pci_enable_msi_range() */
1372 if (pci_msi_desc_is_multi_msi(desc) &&
1373 !(info->flags & MSI_FLAG_MULTI_PCI_MSI))
1375 else if (desc->msi_attrib.is_msix && !(info->flags & MSI_FLAG_PCI_MSIX))
1381 static int pci_msi_domain_handle_error(struct irq_domain *domain,
1382 struct msi_desc *desc, int error)
1384 /* Special handling to support __pci_enable_msi_range() */
1385 if (pci_msi_desc_is_multi_msi(desc) && error == -ENOSPC)
1391 #ifdef GENERIC_MSI_DOMAIN_OPS
1392 static void pci_msi_domain_set_desc(msi_alloc_info_t *arg,
1393 struct msi_desc *desc)
1396 arg->hwirq = pci_msi_domain_calc_hwirq(msi_desc_to_pci_dev(desc),
1400 #define pci_msi_domain_set_desc NULL
1403 static struct msi_domain_ops pci_msi_domain_ops_default = {
1404 .set_desc = pci_msi_domain_set_desc,
1405 .msi_check = pci_msi_domain_check_cap,
1406 .handle_error = pci_msi_domain_handle_error,
1409 static void pci_msi_domain_update_dom_ops(struct msi_domain_info *info)
1411 struct msi_domain_ops *ops = info->ops;
1414 info->ops = &pci_msi_domain_ops_default;
1416 if (ops->set_desc == NULL)
1417 ops->set_desc = pci_msi_domain_set_desc;
1418 if (ops->msi_check == NULL)
1419 ops->msi_check = pci_msi_domain_check_cap;
1420 if (ops->handle_error == NULL)
1421 ops->handle_error = pci_msi_domain_handle_error;
1425 static void pci_msi_domain_update_chip_ops(struct msi_domain_info *info)
1427 struct irq_chip *chip = info->chip;
1430 if (!chip->irq_write_msi_msg)
1431 chip->irq_write_msi_msg = pci_msi_domain_write_msg;
1432 if (!chip->irq_mask)
1433 chip->irq_mask = pci_msi_mask_irq;
1434 if (!chip->irq_unmask)
1435 chip->irq_unmask = pci_msi_unmask_irq;
1439 * pci_msi_create_irq_domain - Create a MSI interrupt domain
1440 * @fwnode: Optional fwnode of the interrupt controller
1441 * @info: MSI domain info
1442 * @parent: Parent irq domain
1444 * Updates the domain and chip ops and creates a MSI interrupt domain.
1447 * A domain pointer or NULL in case of failure.
1449 struct irq_domain *pci_msi_create_irq_domain(struct fwnode_handle *fwnode,
1450 struct msi_domain_info *info,
1451 struct irq_domain *parent)
1453 struct irq_domain *domain;
1455 if (info->flags & MSI_FLAG_USE_DEF_DOM_OPS)
1456 pci_msi_domain_update_dom_ops(info);
1457 if (info->flags & MSI_FLAG_USE_DEF_CHIP_OPS)
1458 pci_msi_domain_update_chip_ops(info);
1460 info->flags |= MSI_FLAG_ACTIVATE_EARLY;
1462 domain = msi_create_irq_domain(fwnode, info, parent);
1466 irq_domain_update_bus_token(domain, DOMAIN_BUS_PCI_MSI);
1469 EXPORT_SYMBOL_GPL(pci_msi_create_irq_domain);
1471 static int get_msi_id_cb(struct pci_dev *pdev, u16 alias, void *data)
1479 * pci_msi_domain_get_msi_rid - Get the MSI requester id (RID)
1480 * @domain: The interrupt domain
1481 * @pdev: The PCI device.
1483 * The RID for a device is formed from the alias, with a firmware
1484 * supplied mapping applied
1488 u32 pci_msi_domain_get_msi_rid(struct irq_domain *domain, struct pci_dev *pdev)
1490 struct device_node *of_node;
1493 pci_for_each_dma_alias(pdev, get_msi_id_cb, &rid);
1495 of_node = irq_domain_get_of_node(domain);
1496 rid = of_node ? of_msi_map_rid(&pdev->dev, of_node, rid) :
1497 iort_msi_map_rid(&pdev->dev, rid);
1503 * pci_msi_get_device_domain - Get the MSI domain for a given PCI device
1504 * @pdev: The PCI device
1506 * Use the firmware data to find a device-specific MSI domain
1507 * (i.e. not one that is ste as a default).
1509 * Returns: The coresponding MSI domain or NULL if none has been found.
1511 struct irq_domain *pci_msi_get_device_domain(struct pci_dev *pdev)
1513 struct irq_domain *dom;
1516 pci_for_each_dma_alias(pdev, get_msi_id_cb, &rid);
1517 dom = of_msi_map_get_device_domain(&pdev->dev, rid);
1519 dom = iort_get_device_domain(&pdev->dev, rid);
1522 #endif /* CONFIG_PCI_MSI_IRQ_DOMAIN */