3 * Purpose: PCI Message Signaled Interrupt (MSI)
5 * Copyright (C) 2003-2004 Intel
6 * Copyright (C) Tom Long Nguyen (tom.l.nguyen@intel.com)
11 #include <linux/irq.h>
12 #include <linux/interrupt.h>
13 #include <linux/export.h>
14 #include <linux/ioport.h>
15 #include <linux/pci.h>
16 #include <linux/proc_fs.h>
17 #include <linux/msi.h>
18 #include <linux/smp.h>
19 #include <linux/errno.h>
21 #include <linux/slab.h>
22 #include <linux/irqdomain.h>
26 static int pci_msi_enable = 1;
27 int pci_msi_ignore_mask;
29 #define msix_table_size(flags) ((flags & PCI_MSIX_FLAGS_QSIZE) + 1)
31 #ifdef CONFIG_PCI_MSI_IRQ_DOMAIN
32 static struct irq_domain *pci_msi_default_domain;
33 static DEFINE_MUTEX(pci_msi_domain_lock);
35 struct irq_domain * __weak arch_get_pci_msi_domain(struct pci_dev *dev)
37 return pci_msi_default_domain;
40 static struct irq_domain *pci_msi_get_domain(struct pci_dev *dev)
42 struct irq_domain *domain;
44 domain = dev_get_msi_domain(&dev->dev);
48 return arch_get_pci_msi_domain(dev);
51 static int pci_msi_setup_msi_irqs(struct pci_dev *dev, int nvec, int type)
53 struct irq_domain *domain;
55 domain = pci_msi_get_domain(dev);
57 return pci_msi_domain_alloc_irqs(domain, dev, nvec, type);
59 return arch_setup_msi_irqs(dev, nvec, type);
62 static void pci_msi_teardown_msi_irqs(struct pci_dev *dev)
64 struct irq_domain *domain;
66 domain = pci_msi_get_domain(dev);
68 pci_msi_domain_free_irqs(domain, dev);
70 arch_teardown_msi_irqs(dev);
73 #define pci_msi_setup_msi_irqs arch_setup_msi_irqs
74 #define pci_msi_teardown_msi_irqs arch_teardown_msi_irqs
79 int __weak arch_setup_msi_irq(struct pci_dev *dev, struct msi_desc *desc)
81 struct msi_controller *chip = dev->bus->msi;
84 if (!chip || !chip->setup_irq)
87 err = chip->setup_irq(chip, dev, desc);
91 irq_set_chip_data(desc->irq, chip);
96 void __weak arch_teardown_msi_irq(unsigned int irq)
98 struct msi_controller *chip = irq_get_chip_data(irq);
100 if (!chip || !chip->teardown_irq)
103 chip->teardown_irq(chip, irq);
106 int __weak arch_setup_msi_irqs(struct pci_dev *dev, int nvec, int type)
108 struct msi_controller *chip = dev->bus->msi;
109 struct msi_desc *entry;
112 if (chip && chip->setup_irqs)
113 return chip->setup_irqs(chip, dev, nvec, type);
115 * If an architecture wants to support multiple MSI, it needs to
116 * override arch_setup_msi_irqs()
118 if (type == PCI_CAP_ID_MSI && nvec > 1)
121 for_each_pci_msi_entry(entry, dev) {
122 ret = arch_setup_msi_irq(dev, entry);
133 * We have a default implementation available as a separate non-weak
134 * function, as it is used by the Xen x86 PCI code
136 void default_teardown_msi_irqs(struct pci_dev *dev)
139 struct msi_desc *entry;
141 for_each_pci_msi_entry(entry, dev)
143 for (i = 0; i < entry->nvec_used; i++)
144 arch_teardown_msi_irq(entry->irq + i);
147 void __weak arch_teardown_msi_irqs(struct pci_dev *dev)
149 return default_teardown_msi_irqs(dev);
152 static void default_restore_msi_irq(struct pci_dev *dev, int irq)
154 struct msi_desc *entry;
157 if (dev->msix_enabled) {
158 for_each_pci_msi_entry(entry, dev) {
159 if (irq == entry->irq)
162 } else if (dev->msi_enabled) {
163 entry = irq_get_msi_desc(irq);
167 __pci_write_msi_msg(entry, &entry->msg);
170 void __weak arch_restore_msi_irqs(struct pci_dev *dev)
172 return default_restore_msi_irqs(dev);
175 static inline __attribute_const__ u32 msi_mask(unsigned x)
177 /* Don't shift by >= width of type */
180 return (1 << (1 << x)) - 1;
184 * PCI 2.3 does not specify mask bits for each MSI interrupt. Attempting to
185 * mask all MSI interrupts by clearing the MSI enable bit does not work
186 * reliably as devices without an INTx disable bit will then generate a
187 * level IRQ which will never be cleared.
189 u32 __pci_msi_desc_mask_irq(struct msi_desc *desc, u32 mask, u32 flag)
191 u32 mask_bits = desc->masked;
193 if (pci_msi_ignore_mask || !desc->msi_attrib.maskbit)
198 pci_write_config_dword(msi_desc_to_pci_dev(desc), desc->mask_pos,
204 static void msi_mask_irq(struct msi_desc *desc, u32 mask, u32 flag)
206 desc->masked = __pci_msi_desc_mask_irq(desc, mask, flag);
210 * This internal function does not flush PCI writes to the device.
211 * All users must ensure that they read from the device before either
212 * assuming that the device state is up to date, or returning out of this
213 * file. This saves a few milliseconds when initialising devices with lots
214 * of MSI-X interrupts.
216 u32 __pci_msix_desc_mask_irq(struct msi_desc *desc, u32 flag)
218 u32 mask_bits = desc->masked;
219 unsigned offset = desc->msi_attrib.entry_nr * PCI_MSIX_ENTRY_SIZE +
220 PCI_MSIX_ENTRY_VECTOR_CTRL;
222 if (pci_msi_ignore_mask)
225 mask_bits &= ~PCI_MSIX_ENTRY_CTRL_MASKBIT;
227 mask_bits |= PCI_MSIX_ENTRY_CTRL_MASKBIT;
228 writel(mask_bits, desc->mask_base + offset);
233 static void msix_mask_irq(struct msi_desc *desc, u32 flag)
235 desc->masked = __pci_msix_desc_mask_irq(desc, flag);
238 static void msi_set_mask_bit(struct irq_data *data, u32 flag)
240 struct msi_desc *desc = irq_data_get_msi_desc(data);
242 if (desc->msi_attrib.is_msix) {
243 msix_mask_irq(desc, flag);
244 readl(desc->mask_base); /* Flush write to device */
246 unsigned offset = data->irq - desc->irq;
247 msi_mask_irq(desc, 1 << offset, flag << offset);
252 * pci_msi_mask_irq - Generic irq chip callback to mask PCI/MSI interrupts
253 * @data: pointer to irqdata associated to that interrupt
255 void pci_msi_mask_irq(struct irq_data *data)
257 msi_set_mask_bit(data, 1);
261 * pci_msi_unmask_irq - Generic irq chip callback to unmask PCI/MSI interrupts
262 * @data: pointer to irqdata associated to that interrupt
264 void pci_msi_unmask_irq(struct irq_data *data)
266 msi_set_mask_bit(data, 0);
269 void default_restore_msi_irqs(struct pci_dev *dev)
271 struct msi_desc *entry;
273 for_each_pci_msi_entry(entry, dev)
274 default_restore_msi_irq(dev, entry->irq);
277 void __pci_read_msi_msg(struct msi_desc *entry, struct msi_msg *msg)
279 struct pci_dev *dev = msi_desc_to_pci_dev(entry);
281 BUG_ON(dev->current_state != PCI_D0);
283 if (entry->msi_attrib.is_msix) {
284 void __iomem *base = entry->mask_base +
285 entry->msi_attrib.entry_nr * PCI_MSIX_ENTRY_SIZE;
287 msg->address_lo = readl(base + PCI_MSIX_ENTRY_LOWER_ADDR);
288 msg->address_hi = readl(base + PCI_MSIX_ENTRY_UPPER_ADDR);
289 msg->data = readl(base + PCI_MSIX_ENTRY_DATA);
291 int pos = dev->msi_cap;
294 pci_read_config_dword(dev, pos + PCI_MSI_ADDRESS_LO,
296 if (entry->msi_attrib.is_64) {
297 pci_read_config_dword(dev, pos + PCI_MSI_ADDRESS_HI,
299 pci_read_config_word(dev, pos + PCI_MSI_DATA_64, &data);
302 pci_read_config_word(dev, pos + PCI_MSI_DATA_32, &data);
308 void __pci_write_msi_msg(struct msi_desc *entry, struct msi_msg *msg)
310 struct pci_dev *dev = msi_desc_to_pci_dev(entry);
312 if (dev->current_state != PCI_D0) {
313 /* Don't touch the hardware now */
314 } else if (entry->msi_attrib.is_msix) {
316 base = entry->mask_base +
317 entry->msi_attrib.entry_nr * PCI_MSIX_ENTRY_SIZE;
319 writel(msg->address_lo, base + PCI_MSIX_ENTRY_LOWER_ADDR);
320 writel(msg->address_hi, base + PCI_MSIX_ENTRY_UPPER_ADDR);
321 writel(msg->data, base + PCI_MSIX_ENTRY_DATA);
323 int pos = dev->msi_cap;
326 pci_read_config_word(dev, pos + PCI_MSI_FLAGS, &msgctl);
327 msgctl &= ~PCI_MSI_FLAGS_QSIZE;
328 msgctl |= entry->msi_attrib.multiple << 4;
329 pci_write_config_word(dev, pos + PCI_MSI_FLAGS, msgctl);
331 pci_write_config_dword(dev, pos + PCI_MSI_ADDRESS_LO,
333 if (entry->msi_attrib.is_64) {
334 pci_write_config_dword(dev, pos + PCI_MSI_ADDRESS_HI,
336 pci_write_config_word(dev, pos + PCI_MSI_DATA_64,
339 pci_write_config_word(dev, pos + PCI_MSI_DATA_32,
346 void pci_write_msi_msg(unsigned int irq, struct msi_msg *msg)
348 struct msi_desc *entry = irq_get_msi_desc(irq);
350 __pci_write_msi_msg(entry, msg);
352 EXPORT_SYMBOL_GPL(pci_write_msi_msg);
354 static void free_msi_irqs(struct pci_dev *dev)
356 struct list_head *msi_list = dev_to_msi_list(&dev->dev);
357 struct msi_desc *entry, *tmp;
358 struct attribute **msi_attrs;
359 struct device_attribute *dev_attr;
362 for_each_pci_msi_entry(entry, dev)
364 for (i = 0; i < entry->nvec_used; i++)
365 BUG_ON(irq_has_action(entry->irq + i));
367 pci_msi_teardown_msi_irqs(dev);
369 list_for_each_entry_safe(entry, tmp, msi_list, list) {
370 if (entry->msi_attrib.is_msix) {
371 if (list_is_last(&entry->list, msi_list))
372 iounmap(entry->mask_base);
375 list_del(&entry->list);
379 if (dev->msi_irq_groups) {
380 sysfs_remove_groups(&dev->dev.kobj, dev->msi_irq_groups);
381 msi_attrs = dev->msi_irq_groups[0]->attrs;
382 while (msi_attrs[count]) {
383 dev_attr = container_of(msi_attrs[count],
384 struct device_attribute, attr);
385 kfree(dev_attr->attr.name);
390 kfree(dev->msi_irq_groups[0]);
391 kfree(dev->msi_irq_groups);
392 dev->msi_irq_groups = NULL;
396 static void pci_intx_for_msi(struct pci_dev *dev, int enable)
398 if (!(dev->dev_flags & PCI_DEV_FLAGS_MSI_INTX_DISABLE_BUG))
399 pci_intx(dev, enable);
402 static void __pci_restore_msi_state(struct pci_dev *dev)
405 struct msi_desc *entry;
407 if (!dev->msi_enabled)
410 entry = irq_get_msi_desc(dev->irq);
412 pci_intx_for_msi(dev, 0);
413 pci_msi_set_enable(dev, 0);
414 arch_restore_msi_irqs(dev);
416 pci_read_config_word(dev, dev->msi_cap + PCI_MSI_FLAGS, &control);
417 msi_mask_irq(entry, msi_mask(entry->msi_attrib.multi_cap),
419 control &= ~PCI_MSI_FLAGS_QSIZE;
420 control |= (entry->msi_attrib.multiple << 4) | PCI_MSI_FLAGS_ENABLE;
421 pci_write_config_word(dev, dev->msi_cap + PCI_MSI_FLAGS, control);
424 static void __pci_restore_msix_state(struct pci_dev *dev)
426 struct msi_desc *entry;
428 if (!dev->msix_enabled)
430 BUG_ON(list_empty(dev_to_msi_list(&dev->dev)));
432 /* route the table */
433 pci_intx_for_msi(dev, 0);
434 pci_msix_clear_and_set_ctrl(dev, 0,
435 PCI_MSIX_FLAGS_ENABLE | PCI_MSIX_FLAGS_MASKALL);
437 arch_restore_msi_irqs(dev);
438 for_each_pci_msi_entry(entry, dev)
439 msix_mask_irq(entry, entry->masked);
441 pci_msix_clear_and_set_ctrl(dev, PCI_MSIX_FLAGS_MASKALL, 0);
444 void pci_restore_msi_state(struct pci_dev *dev)
446 __pci_restore_msi_state(dev);
447 __pci_restore_msix_state(dev);
449 EXPORT_SYMBOL_GPL(pci_restore_msi_state);
451 static ssize_t msi_mode_show(struct device *dev, struct device_attribute *attr,
454 struct msi_desc *entry;
458 retval = kstrtoul(attr->attr.name, 10, &irq);
462 entry = irq_get_msi_desc(irq);
464 return sprintf(buf, "%s\n",
465 entry->msi_attrib.is_msix ? "msix" : "msi");
470 static int populate_msi_sysfs(struct pci_dev *pdev)
472 struct attribute **msi_attrs;
473 struct attribute *msi_attr;
474 struct device_attribute *msi_dev_attr;
475 struct attribute_group *msi_irq_group;
476 const struct attribute_group **msi_irq_groups;
477 struct msi_desc *entry;
483 /* Determine how many msi entries we have */
484 for_each_pci_msi_entry(entry, pdev)
485 num_msi += entry->nvec_used;
489 /* Dynamically create the MSI attributes for the PCI device */
490 msi_attrs = kzalloc(sizeof(void *) * (num_msi + 1), GFP_KERNEL);
493 for_each_pci_msi_entry(entry, pdev) {
494 for (i = 0; i < entry->nvec_used; i++) {
495 msi_dev_attr = kzalloc(sizeof(*msi_dev_attr), GFP_KERNEL);
498 msi_attrs[count] = &msi_dev_attr->attr;
500 sysfs_attr_init(&msi_dev_attr->attr);
501 msi_dev_attr->attr.name = kasprintf(GFP_KERNEL, "%d",
503 if (!msi_dev_attr->attr.name)
505 msi_dev_attr->attr.mode = S_IRUGO;
506 msi_dev_attr->show = msi_mode_show;
511 msi_irq_group = kzalloc(sizeof(*msi_irq_group), GFP_KERNEL);
514 msi_irq_group->name = "msi_irqs";
515 msi_irq_group->attrs = msi_attrs;
517 msi_irq_groups = kzalloc(sizeof(void *) * 2, GFP_KERNEL);
519 goto error_irq_group;
520 msi_irq_groups[0] = msi_irq_group;
522 ret = sysfs_create_groups(&pdev->dev.kobj, msi_irq_groups);
524 goto error_irq_groups;
525 pdev->msi_irq_groups = msi_irq_groups;
530 kfree(msi_irq_groups);
532 kfree(msi_irq_group);
535 msi_attr = msi_attrs[count];
537 msi_dev_attr = container_of(msi_attr, struct device_attribute, attr);
538 kfree(msi_attr->name);
541 msi_attr = msi_attrs[count];
547 static struct msi_desc *msi_setup_entry(struct pci_dev *dev, int nvec)
550 struct msi_desc *entry;
552 /* MSI Entry Initialization */
553 entry = alloc_msi_entry(&dev->dev);
557 pci_read_config_word(dev, dev->msi_cap + PCI_MSI_FLAGS, &control);
559 entry->msi_attrib.is_msix = 0;
560 entry->msi_attrib.is_64 = !!(control & PCI_MSI_FLAGS_64BIT);
561 entry->msi_attrib.entry_nr = 0;
562 entry->msi_attrib.maskbit = !!(control & PCI_MSI_FLAGS_MASKBIT);
563 entry->msi_attrib.default_irq = dev->irq; /* Save IOAPIC IRQ */
564 entry->msi_attrib.multi_cap = (control & PCI_MSI_FLAGS_QMASK) >> 1;
565 entry->msi_attrib.multiple = ilog2(__roundup_pow_of_two(nvec));
566 entry->nvec_used = nvec;
568 if (control & PCI_MSI_FLAGS_64BIT)
569 entry->mask_pos = dev->msi_cap + PCI_MSI_MASK_64;
571 entry->mask_pos = dev->msi_cap + PCI_MSI_MASK_32;
573 /* Save the initial mask status */
574 if (entry->msi_attrib.maskbit)
575 pci_read_config_dword(dev, entry->mask_pos, &entry->masked);
580 static int msi_verify_entries(struct pci_dev *dev)
582 struct msi_desc *entry;
584 for_each_pci_msi_entry(entry, dev) {
585 if (!dev->no_64bit_msi || !entry->msg.address_hi)
587 dev_err(&dev->dev, "Device has broken 64-bit MSI but arch"
588 " tried to assign one above 4G\n");
595 * msi_capability_init - configure device's MSI capability structure
596 * @dev: pointer to the pci_dev data structure of MSI device function
597 * @nvec: number of interrupts to allocate
599 * Setup the MSI capability structure of the device with the requested
600 * number of interrupts. A return value of zero indicates the successful
601 * setup of an entry with the new MSI irq. A negative return value indicates
602 * an error, and a positive return value indicates the number of interrupts
603 * which could have been allocated.
605 static int msi_capability_init(struct pci_dev *dev, int nvec)
607 struct msi_desc *entry;
611 pci_msi_set_enable(dev, 0); /* Disable MSI during set up */
613 entry = msi_setup_entry(dev, nvec);
617 /* All MSIs are unmasked by default, Mask them all */
618 mask = msi_mask(entry->msi_attrib.multi_cap);
619 msi_mask_irq(entry, mask, mask);
621 list_add_tail(&entry->list, dev_to_msi_list(&dev->dev));
623 /* Configure MSI capability structure */
624 ret = pci_msi_setup_msi_irqs(dev, nvec, PCI_CAP_ID_MSI);
626 msi_mask_irq(entry, mask, ~mask);
631 ret = msi_verify_entries(dev);
633 msi_mask_irq(entry, mask, ~mask);
638 ret = populate_msi_sysfs(dev);
640 msi_mask_irq(entry, mask, ~mask);
645 /* Set MSI enabled bits */
646 pci_intx_for_msi(dev, 0);
647 pci_msi_set_enable(dev, 1);
648 dev->msi_enabled = 1;
650 pcibios_free_irq(dev);
651 dev->irq = entry->irq;
655 static void __iomem *msix_map_region(struct pci_dev *dev, unsigned nr_entries)
657 resource_size_t phys_addr;
662 pci_read_config_dword(dev, dev->msix_cap + PCI_MSIX_TABLE,
664 bir = (u8)(table_offset & PCI_MSIX_TABLE_BIR);
665 flags = pci_resource_flags(dev, bir);
666 if (!flags || (flags & IORESOURCE_UNSET))
669 table_offset &= PCI_MSIX_TABLE_OFFSET;
670 phys_addr = pci_resource_start(dev, bir) + table_offset;
672 return ioremap_nocache(phys_addr, nr_entries * PCI_MSIX_ENTRY_SIZE);
675 static int msix_setup_entries(struct pci_dev *dev, void __iomem *base,
676 struct msix_entry *entries, int nvec)
678 struct msi_desc *entry;
681 for (i = 0; i < nvec; i++) {
682 entry = alloc_msi_entry(&dev->dev);
688 /* No enough memory. Don't try again */
692 entry->msi_attrib.is_msix = 1;
693 entry->msi_attrib.is_64 = 1;
694 entry->msi_attrib.entry_nr = entries[i].entry;
695 entry->msi_attrib.default_irq = dev->irq;
696 entry->mask_base = base;
697 entry->nvec_used = 1;
699 list_add_tail(&entry->list, dev_to_msi_list(&dev->dev));
705 static void msix_program_entries(struct pci_dev *dev,
706 struct msix_entry *entries)
708 struct msi_desc *entry;
711 for_each_pci_msi_entry(entry, dev) {
712 int offset = entries[i].entry * PCI_MSIX_ENTRY_SIZE +
713 PCI_MSIX_ENTRY_VECTOR_CTRL;
715 entries[i].vector = entry->irq;
716 entry->masked = readl(entry->mask_base + offset);
717 msix_mask_irq(entry, 1);
723 * msix_capability_init - configure device's MSI-X capability
724 * @dev: pointer to the pci_dev data structure of MSI-X device function
725 * @entries: pointer to an array of struct msix_entry entries
726 * @nvec: number of @entries
728 * Setup the MSI-X capability structure of device function with a
729 * single MSI-X irq. A return of zero indicates the successful setup of
730 * requested MSI-X entries with allocated irqs or non-zero for otherwise.
732 static int msix_capability_init(struct pci_dev *dev,
733 struct msix_entry *entries, int nvec)
739 /* Ensure MSI-X is disabled while it is set up */
740 pci_msix_clear_and_set_ctrl(dev, PCI_MSIX_FLAGS_ENABLE, 0);
742 pci_read_config_word(dev, dev->msix_cap + PCI_MSIX_FLAGS, &control);
743 /* Request & Map MSI-X table region */
744 base = msix_map_region(dev, msix_table_size(control));
748 ret = msix_setup_entries(dev, base, entries, nvec);
752 ret = pci_msi_setup_msi_irqs(dev, nvec, PCI_CAP_ID_MSIX);
756 /* Check if all MSI entries honor device restrictions */
757 ret = msi_verify_entries(dev);
762 * Some devices require MSI-X to be enabled before we can touch the
763 * MSI-X registers. We need to mask all the vectors to prevent
764 * interrupts coming in before they're fully set up.
766 pci_msix_clear_and_set_ctrl(dev, 0,
767 PCI_MSIX_FLAGS_MASKALL | PCI_MSIX_FLAGS_ENABLE);
769 msix_program_entries(dev, entries);
771 ret = populate_msi_sysfs(dev);
775 /* Set MSI-X enabled bits and unmask the function */
776 pci_intx_for_msi(dev, 0);
777 dev->msix_enabled = 1;
778 pci_msix_clear_and_set_ctrl(dev, PCI_MSIX_FLAGS_MASKALL, 0);
780 pcibios_free_irq(dev);
786 * If we had some success, report the number of irqs
787 * we succeeded in setting up.
789 struct msi_desc *entry;
792 for_each_pci_msi_entry(entry, dev) {
807 * pci_msi_supported - check whether MSI may be enabled on a device
808 * @dev: pointer to the pci_dev data structure of MSI device function
809 * @nvec: how many MSIs have been requested ?
811 * Look at global flags, the device itself, and its parent buses
812 * to determine if MSI/-X are supported for the device. If MSI/-X is
813 * supported return 1, else return 0.
815 static int pci_msi_supported(struct pci_dev *dev, int nvec)
819 /* MSI must be globally enabled and supported by the device */
823 if (!dev || dev->no_msi || dev->current_state != PCI_D0)
827 * You can't ask to have 0 or less MSIs configured.
829 * b) the list manipulation code assumes nvec >= 1.
835 * Any bridge which does NOT route MSI transactions from its
836 * secondary bus to its primary bus must set NO_MSI flag on
837 * the secondary pci_bus.
838 * We expect only arch-specific PCI host bus controller driver
839 * or quirks for specific PCI bridges to be setting NO_MSI.
841 for (bus = dev->bus; bus; bus = bus->parent)
842 if (bus->bus_flags & PCI_BUS_FLAGS_NO_MSI)
849 * pci_msi_vec_count - Return the number of MSI vectors a device can send
850 * @dev: device to report about
852 * This function returns the number of MSI vectors a device requested via
853 * Multiple Message Capable register. It returns a negative errno if the
854 * device is not capable sending MSI interrupts. Otherwise, the call succeeds
855 * and returns a power of two, up to a maximum of 2^5 (32), according to the
858 int pci_msi_vec_count(struct pci_dev *dev)
866 pci_read_config_word(dev, dev->msi_cap + PCI_MSI_FLAGS, &msgctl);
867 ret = 1 << ((msgctl & PCI_MSI_FLAGS_QMASK) >> 1);
871 EXPORT_SYMBOL(pci_msi_vec_count);
873 void pci_msi_shutdown(struct pci_dev *dev)
875 struct msi_desc *desc;
878 if (!pci_msi_enable || !dev || !dev->msi_enabled)
881 BUG_ON(list_empty(dev_to_msi_list(&dev->dev)));
882 desc = first_pci_msi_entry(dev);
884 pci_msi_set_enable(dev, 0);
885 pci_intx_for_msi(dev, 1);
886 dev->msi_enabled = 0;
888 /* Return the device with MSI unmasked as initial states */
889 mask = msi_mask(desc->msi_attrib.multi_cap);
890 /* Keep cached state to be restored */
891 __pci_msi_desc_mask_irq(desc, mask, ~mask);
893 /* Restore dev->irq to its default pin-assertion irq */
894 dev->irq = desc->msi_attrib.default_irq;
895 pcibios_alloc_irq(dev);
898 void pci_disable_msi(struct pci_dev *dev)
900 if (!pci_msi_enable || !dev || !dev->msi_enabled)
903 pci_msi_shutdown(dev);
906 EXPORT_SYMBOL(pci_disable_msi);
909 * pci_msix_vec_count - return the number of device's MSI-X table entries
910 * @dev: pointer to the pci_dev data structure of MSI-X device function
911 * This function returns the number of device's MSI-X table entries and
912 * therefore the number of MSI-X vectors device is capable of sending.
913 * It returns a negative errno if the device is not capable of sending MSI-X
916 int pci_msix_vec_count(struct pci_dev *dev)
923 pci_read_config_word(dev, dev->msix_cap + PCI_MSIX_FLAGS, &control);
924 return msix_table_size(control);
926 EXPORT_SYMBOL(pci_msix_vec_count);
929 * pci_enable_msix - configure device's MSI-X capability structure
930 * @dev: pointer to the pci_dev data structure of MSI-X device function
931 * @entries: pointer to an array of MSI-X entries
932 * @nvec: number of MSI-X irqs requested for allocation by device driver
934 * Setup the MSI-X capability structure of device function with the number
935 * of requested irqs upon its software driver call to request for
936 * MSI-X mode enabled on its hardware device function. A return of zero
937 * indicates the successful configuration of MSI-X capability structure
938 * with new allocated MSI-X irqs. A return of < 0 indicates a failure.
939 * Or a return of > 0 indicates that driver request is exceeding the number
940 * of irqs or MSI-X vectors available. Driver should use the returned value to
941 * re-send its request.
943 int pci_enable_msix(struct pci_dev *dev, struct msix_entry *entries, int nvec)
948 if (!pci_msi_supported(dev, nvec))
954 nr_entries = pci_msix_vec_count(dev);
957 if (nvec > nr_entries)
960 /* Check for any invalid entries */
961 for (i = 0; i < nvec; i++) {
962 if (entries[i].entry >= nr_entries)
963 return -EINVAL; /* invalid entry */
964 for (j = i + 1; j < nvec; j++) {
965 if (entries[i].entry == entries[j].entry)
966 return -EINVAL; /* duplicate entry */
969 WARN_ON(!!dev->msix_enabled);
971 /* Check whether driver already requested for MSI irq */
972 if (dev->msi_enabled) {
973 dev_info(&dev->dev, "can't enable MSI-X (MSI IRQ already assigned)\n");
976 return msix_capability_init(dev, entries, nvec);
978 EXPORT_SYMBOL(pci_enable_msix);
980 void pci_msix_shutdown(struct pci_dev *dev)
982 struct msi_desc *entry;
984 if (!pci_msi_enable || !dev || !dev->msix_enabled)
987 /* Return the device with MSI-X masked as initial states */
988 for_each_pci_msi_entry(entry, dev) {
989 /* Keep cached states to be restored */
990 __pci_msix_desc_mask_irq(entry, 1);
993 pci_msix_clear_and_set_ctrl(dev, PCI_MSIX_FLAGS_ENABLE, 0);
994 pci_intx_for_msi(dev, 1);
995 dev->msix_enabled = 0;
996 pcibios_alloc_irq(dev);
999 void pci_disable_msix(struct pci_dev *dev)
1001 if (!pci_msi_enable || !dev || !dev->msix_enabled)
1004 pci_msix_shutdown(dev);
1007 EXPORT_SYMBOL(pci_disable_msix);
1009 void pci_no_msi(void)
1015 * pci_msi_enabled - is MSI enabled?
1017 * Returns true if MSI has not been disabled by the command-line option
1020 int pci_msi_enabled(void)
1022 return pci_msi_enable;
1024 EXPORT_SYMBOL(pci_msi_enabled);
1026 void pci_msi_init_pci_dev(struct pci_dev *dev)
1031 * pci_enable_msi_range - configure device's MSI capability structure
1032 * @dev: device to configure
1033 * @minvec: minimal number of interrupts to configure
1034 * @maxvec: maximum number of interrupts to configure
1036 * This function tries to allocate a maximum possible number of interrupts in a
1037 * range between @minvec and @maxvec. It returns a negative errno if an error
1038 * occurs. If it succeeds, it returns the actual number of interrupts allocated
1039 * and updates the @dev's irq member to the lowest new interrupt number;
1040 * the other interrupt numbers allocated to this device are consecutive.
1042 int pci_enable_msi_range(struct pci_dev *dev, int minvec, int maxvec)
1047 if (!pci_msi_supported(dev, minvec))
1050 WARN_ON(!!dev->msi_enabled);
1052 /* Check whether driver already requested MSI-X irqs */
1053 if (dev->msix_enabled) {
1055 "can't enable MSI (MSI-X already enabled)\n");
1059 if (maxvec < minvec)
1062 nvec = pci_msi_vec_count(dev);
1065 else if (nvec < minvec)
1067 else if (nvec > maxvec)
1071 rc = msi_capability_init(dev, nvec);
1074 } else if (rc > 0) {
1083 EXPORT_SYMBOL(pci_enable_msi_range);
1086 * pci_enable_msix_range - configure device's MSI-X capability structure
1087 * @dev: pointer to the pci_dev data structure of MSI-X device function
1088 * @entries: pointer to an array of MSI-X entries
1089 * @minvec: minimum number of MSI-X irqs requested
1090 * @maxvec: maximum number of MSI-X irqs requested
1092 * Setup the MSI-X capability structure of device function with a maximum
1093 * possible number of interrupts in the range between @minvec and @maxvec
1094 * upon its software driver call to request for MSI-X mode enabled on its
1095 * hardware device function. It returns a negative errno if an error occurs.
1096 * If it succeeds, it returns the actual number of interrupts allocated and
1097 * indicates the successful configuration of MSI-X capability structure
1098 * with new allocated MSI-X interrupts.
1100 int pci_enable_msix_range(struct pci_dev *dev, struct msix_entry *entries,
1101 int minvec, int maxvec)
1106 if (maxvec < minvec)
1110 rc = pci_enable_msix(dev, entries, nvec);
1113 } else if (rc > 0) {
1122 EXPORT_SYMBOL(pci_enable_msix_range);
1124 struct pci_dev *msi_desc_to_pci_dev(struct msi_desc *desc)
1126 return to_pci_dev(desc->dev);
1129 void *msi_desc_to_pci_sysdata(struct msi_desc *desc)
1131 struct pci_dev *dev = msi_desc_to_pci_dev(desc);
1133 return dev->bus->sysdata;
1135 EXPORT_SYMBOL_GPL(msi_desc_to_pci_sysdata);
1137 #ifdef CONFIG_PCI_MSI_IRQ_DOMAIN
1139 * pci_msi_domain_write_msg - Helper to write MSI message to PCI config space
1140 * @irq_data: Pointer to interrupt data of the MSI interrupt
1141 * @msg: Pointer to the message
1143 void pci_msi_domain_write_msg(struct irq_data *irq_data, struct msi_msg *msg)
1145 struct msi_desc *desc = irq_data_get_msi_desc(irq_data);
1148 * For MSI-X desc->irq is always equal to irq_data->irq. For
1149 * MSI only the first interrupt of MULTI MSI passes the test.
1151 if (desc->irq == irq_data->irq)
1152 __pci_write_msi_msg(desc, msg);
1156 * pci_msi_domain_calc_hwirq - Generate a unique ID for an MSI source
1157 * @dev: Pointer to the PCI device
1158 * @desc: Pointer to the msi descriptor
1160 * The ID number is only used within the irqdomain.
1162 irq_hw_number_t pci_msi_domain_calc_hwirq(struct pci_dev *dev,
1163 struct msi_desc *desc)
1165 return (irq_hw_number_t)desc->msi_attrib.entry_nr |
1166 PCI_DEVID(dev->bus->number, dev->devfn) << 11 |
1167 (pci_domain_nr(dev->bus) & 0xFFFFFFFF) << 27;
1170 static inline bool pci_msi_desc_is_multi_msi(struct msi_desc *desc)
1172 return !desc->msi_attrib.is_msix && desc->nvec_used > 1;
1176 * pci_msi_domain_check_cap - Verify that @domain supports the capabilities for @dev
1177 * @domain: The interrupt domain to check
1178 * @info: The domain info for verification
1179 * @dev: The device to check
1182 * 0 if the functionality is supported
1183 * 1 if Multi MSI is requested, but the domain does not support it
1184 * -ENOTSUPP otherwise
1186 int pci_msi_domain_check_cap(struct irq_domain *domain,
1187 struct msi_domain_info *info, struct device *dev)
1189 struct msi_desc *desc = first_pci_msi_entry(to_pci_dev(dev));
1191 /* Special handling to support pci_enable_msi_range() */
1192 if (pci_msi_desc_is_multi_msi(desc) &&
1193 !(info->flags & MSI_FLAG_MULTI_PCI_MSI))
1195 else if (desc->msi_attrib.is_msix && !(info->flags & MSI_FLAG_PCI_MSIX))
1201 static int pci_msi_domain_handle_error(struct irq_domain *domain,
1202 struct msi_desc *desc, int error)
1204 /* Special handling to support pci_enable_msi_range() */
1205 if (pci_msi_desc_is_multi_msi(desc) && error == -ENOSPC)
1211 #ifdef GENERIC_MSI_DOMAIN_OPS
1212 static void pci_msi_domain_set_desc(msi_alloc_info_t *arg,
1213 struct msi_desc *desc)
1216 arg->hwirq = pci_msi_domain_calc_hwirq(msi_desc_to_pci_dev(desc),
1220 #define pci_msi_domain_set_desc NULL
1223 static struct msi_domain_ops pci_msi_domain_ops_default = {
1224 .set_desc = pci_msi_domain_set_desc,
1225 .msi_check = pci_msi_domain_check_cap,
1226 .handle_error = pci_msi_domain_handle_error,
1229 static void pci_msi_domain_update_dom_ops(struct msi_domain_info *info)
1231 struct msi_domain_ops *ops = info->ops;
1234 info->ops = &pci_msi_domain_ops_default;
1236 if (ops->set_desc == NULL)
1237 ops->set_desc = pci_msi_domain_set_desc;
1238 if (ops->msi_check == NULL)
1239 ops->msi_check = pci_msi_domain_check_cap;
1240 if (ops->handle_error == NULL)
1241 ops->handle_error = pci_msi_domain_handle_error;
1245 static void pci_msi_domain_update_chip_ops(struct msi_domain_info *info)
1247 struct irq_chip *chip = info->chip;
1250 if (!chip->irq_write_msi_msg)
1251 chip->irq_write_msi_msg = pci_msi_domain_write_msg;
1252 if (!chip->irq_mask)
1253 chip->irq_mask = pci_msi_mask_irq;
1254 if (!chip->irq_unmask)
1255 chip->irq_unmask = pci_msi_unmask_irq;
1259 * pci_msi_create_irq_domain - Creat a MSI interrupt domain
1260 * @node: Optional device-tree node of the interrupt controller
1261 * @info: MSI domain info
1262 * @parent: Parent irq domain
1264 * Updates the domain and chip ops and creates a MSI interrupt domain.
1267 * A domain pointer or NULL in case of failure.
1269 struct irq_domain *pci_msi_create_irq_domain(struct device_node *node,
1270 struct msi_domain_info *info,
1271 struct irq_domain *parent)
1273 struct irq_domain *domain;
1275 if (info->flags & MSI_FLAG_USE_DEF_DOM_OPS)
1276 pci_msi_domain_update_dom_ops(info);
1277 if (info->flags & MSI_FLAG_USE_DEF_CHIP_OPS)
1278 pci_msi_domain_update_chip_ops(info);
1280 domain = msi_create_irq_domain(node, info, parent);
1284 domain->bus_token = DOMAIN_BUS_PCI_MSI;
1289 * pci_msi_domain_alloc_irqs - Allocate interrupts for @dev in @domain
1290 * @domain: The interrupt domain to allocate from
1291 * @dev: The device for which to allocate
1292 * @nvec: The number of interrupts to allocate
1293 * @type: Unused to allow simpler migration from the arch_XXX interfaces
1296 * A virtual interrupt number or an error code in case of failure
1298 int pci_msi_domain_alloc_irqs(struct irq_domain *domain, struct pci_dev *dev,
1301 return msi_domain_alloc_irqs(domain, &dev->dev, nvec);
1305 * pci_msi_domain_free_irqs - Free interrupts for @dev in @domain
1306 * @domain: The interrupt domain
1307 * @dev: The device for which to free interrupts
1309 void pci_msi_domain_free_irqs(struct irq_domain *domain, struct pci_dev *dev)
1311 msi_domain_free_irqs(domain, &dev->dev);
1315 * pci_msi_create_default_irq_domain - Create a default MSI interrupt domain
1316 * @node: Optional device-tree node of the interrupt controller
1317 * @info: MSI domain info
1318 * @parent: Parent irq domain
1320 * Returns: A domain pointer or NULL in case of failure. If successful
1321 * the default PCI/MSI irqdomain pointer is updated.
1323 struct irq_domain *pci_msi_create_default_irq_domain(struct device_node *node,
1324 struct msi_domain_info *info, struct irq_domain *parent)
1326 struct irq_domain *domain;
1328 mutex_lock(&pci_msi_domain_lock);
1329 if (pci_msi_default_domain) {
1330 pr_err("PCI: default irq domain for PCI MSI has already been created.\n");
1333 domain = pci_msi_create_irq_domain(node, info, parent);
1334 pci_msi_default_domain = domain;
1336 mutex_unlock(&pci_msi_domain_lock);
1340 #endif /* CONFIG_PCI_MSI_IRQ_DOMAIN */