3 * Purpose: PCI Message Signaled Interrupt (MSI)
5 * Copyright (C) 2003-2004 Intel
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_desc *entry;
112 * If an architecture wants to support multiple MSI, it needs to
113 * override arch_setup_msi_irqs()
115 if (type == PCI_CAP_ID_MSI && nvec > 1)
118 for_each_pci_msi_entry(entry, dev) {
119 ret = arch_setup_msi_irq(dev, entry);
130 * We have a default implementation available as a separate non-weak
131 * function, as it is used by the Xen x86 PCI code
133 void default_teardown_msi_irqs(struct pci_dev *dev)
136 struct msi_desc *entry;
138 for_each_pci_msi_entry(entry, dev)
140 for (i = 0; i < entry->nvec_used; i++)
141 arch_teardown_msi_irq(entry->irq + i);
144 void __weak arch_teardown_msi_irqs(struct pci_dev *dev)
146 return default_teardown_msi_irqs(dev);
149 static void default_restore_msi_irq(struct pci_dev *dev, int irq)
151 struct msi_desc *entry;
154 if (dev->msix_enabled) {
155 for_each_pci_msi_entry(entry, dev) {
156 if (irq == entry->irq)
159 } else if (dev->msi_enabled) {
160 entry = irq_get_msi_desc(irq);
164 __pci_write_msi_msg(entry, &entry->msg);
167 void __weak arch_restore_msi_irqs(struct pci_dev *dev)
169 return default_restore_msi_irqs(dev);
172 static inline __attribute_const__ u32 msi_mask(unsigned x)
174 /* Don't shift by >= width of type */
177 return (1 << (1 << x)) - 1;
181 * PCI 2.3 does not specify mask bits for each MSI interrupt. Attempting to
182 * mask all MSI interrupts by clearing the MSI enable bit does not work
183 * reliably as devices without an INTx disable bit will then generate a
184 * level IRQ which will never be cleared.
186 u32 __pci_msi_desc_mask_irq(struct msi_desc *desc, u32 mask, u32 flag)
188 u32 mask_bits = desc->masked;
190 if (pci_msi_ignore_mask || !desc->msi_attrib.maskbit)
195 pci_write_config_dword(msi_desc_to_pci_dev(desc), desc->mask_pos,
201 static void msi_mask_irq(struct msi_desc *desc, u32 mask, u32 flag)
203 desc->masked = __pci_msi_desc_mask_irq(desc, mask, flag);
207 * This internal function does not flush PCI writes to the device.
208 * All users must ensure that they read from the device before either
209 * assuming that the device state is up to date, or returning out of this
210 * file. This saves a few milliseconds when initialising devices with lots
211 * of MSI-X interrupts.
213 u32 __pci_msix_desc_mask_irq(struct msi_desc *desc, u32 flag)
215 u32 mask_bits = desc->masked;
216 unsigned offset = desc->msi_attrib.entry_nr * PCI_MSIX_ENTRY_SIZE +
217 PCI_MSIX_ENTRY_VECTOR_CTRL;
219 if (pci_msi_ignore_mask)
222 mask_bits &= ~PCI_MSIX_ENTRY_CTRL_MASKBIT;
224 mask_bits |= PCI_MSIX_ENTRY_CTRL_MASKBIT;
225 writel(mask_bits, desc->mask_base + offset);
230 static void msix_mask_irq(struct msi_desc *desc, u32 flag)
232 desc->masked = __pci_msix_desc_mask_irq(desc, flag);
235 static void msi_set_mask_bit(struct irq_data *data, u32 flag)
237 struct msi_desc *desc = irq_data_get_msi_desc(data);
239 if (desc->msi_attrib.is_msix) {
240 msix_mask_irq(desc, flag);
241 readl(desc->mask_base); /* Flush write to device */
243 unsigned offset = data->irq - desc->irq;
244 msi_mask_irq(desc, 1 << offset, flag << offset);
249 * pci_msi_mask_irq - Generic irq chip callback to mask PCI/MSI interrupts
250 * @data: pointer to irqdata associated to that interrupt
252 void pci_msi_mask_irq(struct irq_data *data)
254 msi_set_mask_bit(data, 1);
258 * pci_msi_unmask_irq - Generic irq chip callback to unmask PCI/MSI interrupts
259 * @data: pointer to irqdata associated to that interrupt
261 void pci_msi_unmask_irq(struct irq_data *data)
263 msi_set_mask_bit(data, 0);
266 void default_restore_msi_irqs(struct pci_dev *dev)
268 struct msi_desc *entry;
270 for_each_pci_msi_entry(entry, dev)
271 default_restore_msi_irq(dev, entry->irq);
274 void __pci_read_msi_msg(struct msi_desc *entry, struct msi_msg *msg)
276 struct pci_dev *dev = msi_desc_to_pci_dev(entry);
278 BUG_ON(dev->current_state != PCI_D0);
280 if (entry->msi_attrib.is_msix) {
281 void __iomem *base = entry->mask_base +
282 entry->msi_attrib.entry_nr * PCI_MSIX_ENTRY_SIZE;
284 msg->address_lo = readl(base + PCI_MSIX_ENTRY_LOWER_ADDR);
285 msg->address_hi = readl(base + PCI_MSIX_ENTRY_UPPER_ADDR);
286 msg->data = readl(base + PCI_MSIX_ENTRY_DATA);
288 int pos = dev->msi_cap;
291 pci_read_config_dword(dev, pos + PCI_MSI_ADDRESS_LO,
293 if (entry->msi_attrib.is_64) {
294 pci_read_config_dword(dev, pos + PCI_MSI_ADDRESS_HI,
296 pci_read_config_word(dev, pos + PCI_MSI_DATA_64, &data);
299 pci_read_config_word(dev, pos + PCI_MSI_DATA_32, &data);
305 void __pci_write_msi_msg(struct msi_desc *entry, struct msi_msg *msg)
307 struct pci_dev *dev = msi_desc_to_pci_dev(entry);
309 if (dev->current_state != PCI_D0) {
310 /* Don't touch the hardware now */
311 } else if (entry->msi_attrib.is_msix) {
313 base = entry->mask_base +
314 entry->msi_attrib.entry_nr * PCI_MSIX_ENTRY_SIZE;
316 writel(msg->address_lo, base + PCI_MSIX_ENTRY_LOWER_ADDR);
317 writel(msg->address_hi, base + PCI_MSIX_ENTRY_UPPER_ADDR);
318 writel(msg->data, base + PCI_MSIX_ENTRY_DATA);
320 int pos = dev->msi_cap;
323 pci_read_config_word(dev, pos + PCI_MSI_FLAGS, &msgctl);
324 msgctl &= ~PCI_MSI_FLAGS_QSIZE;
325 msgctl |= entry->msi_attrib.multiple << 4;
326 pci_write_config_word(dev, pos + PCI_MSI_FLAGS, msgctl);
328 pci_write_config_dword(dev, pos + PCI_MSI_ADDRESS_LO,
330 if (entry->msi_attrib.is_64) {
331 pci_write_config_dword(dev, pos + PCI_MSI_ADDRESS_HI,
333 pci_write_config_word(dev, pos + PCI_MSI_DATA_64,
336 pci_write_config_word(dev, pos + PCI_MSI_DATA_32,
343 void pci_write_msi_msg(unsigned int irq, struct msi_msg *msg)
345 struct msi_desc *entry = irq_get_msi_desc(irq);
347 __pci_write_msi_msg(entry, msg);
349 EXPORT_SYMBOL_GPL(pci_write_msi_msg);
351 static void free_msi_irqs(struct pci_dev *dev)
353 struct list_head *msi_list = dev_to_msi_list(&dev->dev);
354 struct msi_desc *entry, *tmp;
355 struct attribute **msi_attrs;
356 struct device_attribute *dev_attr;
359 for_each_pci_msi_entry(entry, dev)
361 for (i = 0; i < entry->nvec_used; i++)
362 BUG_ON(irq_has_action(entry->irq + i));
364 pci_msi_teardown_msi_irqs(dev);
366 list_for_each_entry_safe(entry, tmp, msi_list, list) {
367 if (entry->msi_attrib.is_msix) {
368 if (list_is_last(&entry->list, msi_list))
369 iounmap(entry->mask_base);
372 list_del(&entry->list);
376 if (dev->msi_irq_groups) {
377 sysfs_remove_groups(&dev->dev.kobj, dev->msi_irq_groups);
378 msi_attrs = dev->msi_irq_groups[0]->attrs;
379 while (msi_attrs[count]) {
380 dev_attr = container_of(msi_attrs[count],
381 struct device_attribute, attr);
382 kfree(dev_attr->attr.name);
387 kfree(dev->msi_irq_groups[0]);
388 kfree(dev->msi_irq_groups);
389 dev->msi_irq_groups = NULL;
393 static void pci_intx_for_msi(struct pci_dev *dev, int enable)
395 if (!(dev->dev_flags & PCI_DEV_FLAGS_MSI_INTX_DISABLE_BUG))
396 pci_intx(dev, enable);
399 static void __pci_restore_msi_state(struct pci_dev *dev)
402 struct msi_desc *entry;
404 if (!dev->msi_enabled)
407 entry = irq_get_msi_desc(dev->irq);
409 pci_intx_for_msi(dev, 0);
410 pci_msi_set_enable(dev, 0);
411 arch_restore_msi_irqs(dev);
413 pci_read_config_word(dev, dev->msi_cap + PCI_MSI_FLAGS, &control);
414 msi_mask_irq(entry, msi_mask(entry->msi_attrib.multi_cap),
416 control &= ~PCI_MSI_FLAGS_QSIZE;
417 control |= (entry->msi_attrib.multiple << 4) | PCI_MSI_FLAGS_ENABLE;
418 pci_write_config_word(dev, dev->msi_cap + PCI_MSI_FLAGS, control);
421 static void __pci_restore_msix_state(struct pci_dev *dev)
423 struct msi_desc *entry;
425 if (!dev->msix_enabled)
427 BUG_ON(list_empty(dev_to_msi_list(&dev->dev)));
429 /* route the table */
430 pci_intx_for_msi(dev, 0);
431 pci_msix_clear_and_set_ctrl(dev, 0,
432 PCI_MSIX_FLAGS_ENABLE | PCI_MSIX_FLAGS_MASKALL);
434 arch_restore_msi_irqs(dev);
435 for_each_pci_msi_entry(entry, dev)
436 msix_mask_irq(entry, entry->masked);
438 pci_msix_clear_and_set_ctrl(dev, PCI_MSIX_FLAGS_MASKALL, 0);
441 void pci_restore_msi_state(struct pci_dev *dev)
443 __pci_restore_msi_state(dev);
444 __pci_restore_msix_state(dev);
446 EXPORT_SYMBOL_GPL(pci_restore_msi_state);
448 static ssize_t msi_mode_show(struct device *dev, struct device_attribute *attr,
451 struct msi_desc *entry;
455 retval = kstrtoul(attr->attr.name, 10, &irq);
459 entry = irq_get_msi_desc(irq);
461 return sprintf(buf, "%s\n",
462 entry->msi_attrib.is_msix ? "msix" : "msi");
467 static int populate_msi_sysfs(struct pci_dev *pdev)
469 struct attribute **msi_attrs;
470 struct attribute *msi_attr;
471 struct device_attribute *msi_dev_attr;
472 struct attribute_group *msi_irq_group;
473 const struct attribute_group **msi_irq_groups;
474 struct msi_desc *entry;
479 /* Determine how many msi entries we have */
480 for_each_pci_msi_entry(entry, pdev)
485 /* Dynamically create the MSI attributes for the PCI device */
486 msi_attrs = kzalloc(sizeof(void *) * (num_msi + 1), GFP_KERNEL);
489 for_each_pci_msi_entry(entry, pdev) {
490 msi_dev_attr = kzalloc(sizeof(*msi_dev_attr), GFP_KERNEL);
493 msi_attrs[count] = &msi_dev_attr->attr;
495 sysfs_attr_init(&msi_dev_attr->attr);
496 msi_dev_attr->attr.name = kasprintf(GFP_KERNEL, "%d",
498 if (!msi_dev_attr->attr.name)
500 msi_dev_attr->attr.mode = S_IRUGO;
501 msi_dev_attr->show = msi_mode_show;
505 msi_irq_group = kzalloc(sizeof(*msi_irq_group), GFP_KERNEL);
508 msi_irq_group->name = "msi_irqs";
509 msi_irq_group->attrs = msi_attrs;
511 msi_irq_groups = kzalloc(sizeof(void *) * 2, GFP_KERNEL);
513 goto error_irq_group;
514 msi_irq_groups[0] = msi_irq_group;
516 ret = sysfs_create_groups(&pdev->dev.kobj, msi_irq_groups);
518 goto error_irq_groups;
519 pdev->msi_irq_groups = msi_irq_groups;
524 kfree(msi_irq_groups);
526 kfree(msi_irq_group);
529 msi_attr = msi_attrs[count];
531 msi_dev_attr = container_of(msi_attr, struct device_attribute, attr);
532 kfree(msi_attr->name);
535 msi_attr = msi_attrs[count];
541 static struct msi_desc *msi_setup_entry(struct pci_dev *dev, int nvec)
544 struct msi_desc *entry;
546 /* MSI Entry Initialization */
547 entry = alloc_msi_entry(&dev->dev);
551 pci_read_config_word(dev, dev->msi_cap + PCI_MSI_FLAGS, &control);
553 entry->msi_attrib.is_msix = 0;
554 entry->msi_attrib.is_64 = !!(control & PCI_MSI_FLAGS_64BIT);
555 entry->msi_attrib.entry_nr = 0;
556 entry->msi_attrib.maskbit = !!(control & PCI_MSI_FLAGS_MASKBIT);
557 entry->msi_attrib.default_irq = dev->irq; /* Save IOAPIC IRQ */
558 entry->msi_attrib.multi_cap = (control & PCI_MSI_FLAGS_QMASK) >> 1;
559 entry->msi_attrib.multiple = ilog2(__roundup_pow_of_two(nvec));
560 entry->nvec_used = nvec;
562 if (control & PCI_MSI_FLAGS_64BIT)
563 entry->mask_pos = dev->msi_cap + PCI_MSI_MASK_64;
565 entry->mask_pos = dev->msi_cap + PCI_MSI_MASK_32;
567 /* Save the initial mask status */
568 if (entry->msi_attrib.maskbit)
569 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 dev_err(&dev->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
593 * Setup the MSI capability structure of the device with the requested
594 * number of interrupts. A return value of zero indicates the successful
595 * setup of an entry with the new MSI irq. A negative return value indicates
596 * an error, and a positive return value indicates the number of interrupts
597 * which could have been allocated.
599 static int msi_capability_init(struct pci_dev *dev, int nvec)
601 struct msi_desc *entry;
605 pci_msi_set_enable(dev, 0); /* Disable MSI during set up */
607 entry = msi_setup_entry(dev, nvec);
611 /* All MSIs are unmasked by default, Mask them all */
612 mask = msi_mask(entry->msi_attrib.multi_cap);
613 msi_mask_irq(entry, mask, mask);
615 list_add_tail(&entry->list, dev_to_msi_list(&dev->dev));
617 /* Configure MSI capability structure */
618 ret = pci_msi_setup_msi_irqs(dev, nvec, PCI_CAP_ID_MSI);
620 msi_mask_irq(entry, mask, ~mask);
625 ret = msi_verify_entries(dev);
627 msi_mask_irq(entry, mask, ~mask);
632 ret = populate_msi_sysfs(dev);
634 msi_mask_irq(entry, mask, ~mask);
639 /* Set MSI enabled bits */
640 pci_intx_for_msi(dev, 0);
641 pci_msi_set_enable(dev, 1);
642 dev->msi_enabled = 1;
644 pcibios_free_irq(dev);
645 dev->irq = entry->irq;
649 static void __iomem *msix_map_region(struct pci_dev *dev, unsigned nr_entries)
651 resource_size_t phys_addr;
656 pci_read_config_dword(dev, dev->msix_cap + PCI_MSIX_TABLE,
658 bir = (u8)(table_offset & PCI_MSIX_TABLE_BIR);
659 flags = pci_resource_flags(dev, bir);
660 if (!flags || (flags & IORESOURCE_UNSET))
663 table_offset &= PCI_MSIX_TABLE_OFFSET;
664 phys_addr = pci_resource_start(dev, bir) + table_offset;
666 return ioremap_nocache(phys_addr, nr_entries * PCI_MSIX_ENTRY_SIZE);
669 static int msix_setup_entries(struct pci_dev *dev, void __iomem *base,
670 struct msix_entry *entries, int nvec)
672 struct msi_desc *entry;
675 for (i = 0; i < nvec; i++) {
676 entry = alloc_msi_entry(&dev->dev);
682 /* No enough memory. Don't try again */
686 entry->msi_attrib.is_msix = 1;
687 entry->msi_attrib.is_64 = 1;
688 entry->msi_attrib.entry_nr = entries[i].entry;
689 entry->msi_attrib.default_irq = dev->irq;
690 entry->mask_base = base;
691 entry->nvec_used = 1;
693 list_add_tail(&entry->list, dev_to_msi_list(&dev->dev));
699 static void msix_program_entries(struct pci_dev *dev,
700 struct msix_entry *entries)
702 struct msi_desc *entry;
705 for_each_pci_msi_entry(entry, dev) {
706 int offset = entries[i].entry * PCI_MSIX_ENTRY_SIZE +
707 PCI_MSIX_ENTRY_VECTOR_CTRL;
709 entries[i].vector = entry->irq;
710 entry->masked = readl(entry->mask_base + offset);
711 msix_mask_irq(entry, 1);
717 * msix_capability_init - configure device's MSI-X capability
718 * @dev: pointer to the pci_dev data structure of MSI-X device function
719 * @entries: pointer to an array of struct msix_entry entries
720 * @nvec: number of @entries
722 * Setup the MSI-X capability structure of device function with a
723 * single MSI-X irq. A return of zero indicates the successful setup of
724 * requested MSI-X entries with allocated irqs or non-zero for otherwise.
726 static int msix_capability_init(struct pci_dev *dev,
727 struct msix_entry *entries, int nvec)
733 /* Ensure MSI-X is disabled while it is set up */
734 pci_msix_clear_and_set_ctrl(dev, PCI_MSIX_FLAGS_ENABLE, 0);
736 pci_read_config_word(dev, dev->msix_cap + PCI_MSIX_FLAGS, &control);
737 /* Request & Map MSI-X table region */
738 base = msix_map_region(dev, msix_table_size(control));
742 ret = msix_setup_entries(dev, base, entries, nvec);
746 ret = pci_msi_setup_msi_irqs(dev, nvec, PCI_CAP_ID_MSIX);
750 /* Check if all MSI entries honor device restrictions */
751 ret = msi_verify_entries(dev);
756 * Some devices require MSI-X to be enabled before we can touch the
757 * MSI-X registers. We need to mask all the vectors to prevent
758 * interrupts coming in before they're fully set up.
760 pci_msix_clear_and_set_ctrl(dev, 0,
761 PCI_MSIX_FLAGS_MASKALL | PCI_MSIX_FLAGS_ENABLE);
763 msix_program_entries(dev, entries);
765 ret = populate_msi_sysfs(dev);
769 /* Set MSI-X enabled bits and unmask the function */
770 pci_intx_for_msi(dev, 0);
771 dev->msix_enabled = 1;
772 pci_msix_clear_and_set_ctrl(dev, PCI_MSIX_FLAGS_MASKALL, 0);
774 pcibios_free_irq(dev);
780 * If we had some success, report the number of irqs
781 * we succeeded in setting up.
783 struct msi_desc *entry;
786 for_each_pci_msi_entry(entry, dev) {
801 * pci_msi_supported - check whether MSI may be enabled on a device
802 * @dev: pointer to the pci_dev data structure of MSI device function
803 * @nvec: how many MSIs have been requested ?
805 * Look at global flags, the device itself, and its parent buses
806 * to determine if MSI/-X are supported for the device. If MSI/-X is
807 * supported return 1, else return 0.
809 static int pci_msi_supported(struct pci_dev *dev, int nvec)
813 /* MSI must be globally enabled and supported by the device */
817 if (!dev || dev->no_msi || dev->current_state != PCI_D0)
821 * You can't ask to have 0 or less MSIs configured.
823 * b) the list manipulation code assumes nvec >= 1.
829 * Any bridge which does NOT route MSI transactions from its
830 * secondary bus to its primary bus must set NO_MSI flag on
831 * the secondary pci_bus.
832 * We expect only arch-specific PCI host bus controller driver
833 * or quirks for specific PCI bridges to be setting NO_MSI.
835 for (bus = dev->bus; bus; bus = bus->parent)
836 if (bus->bus_flags & PCI_BUS_FLAGS_NO_MSI)
843 * pci_msi_vec_count - Return the number of MSI vectors a device can send
844 * @dev: device to report about
846 * This function returns the number of MSI vectors a device requested via
847 * Multiple Message Capable register. It returns a negative errno if the
848 * device is not capable sending MSI interrupts. Otherwise, the call succeeds
849 * and returns a power of two, up to a maximum of 2^5 (32), according to the
852 int pci_msi_vec_count(struct pci_dev *dev)
860 pci_read_config_word(dev, dev->msi_cap + PCI_MSI_FLAGS, &msgctl);
861 ret = 1 << ((msgctl & PCI_MSI_FLAGS_QMASK) >> 1);
865 EXPORT_SYMBOL(pci_msi_vec_count);
867 void pci_msi_shutdown(struct pci_dev *dev)
869 struct msi_desc *desc;
872 if (!pci_msi_enable || !dev || !dev->msi_enabled)
875 BUG_ON(list_empty(dev_to_msi_list(&dev->dev)));
876 desc = first_pci_msi_entry(dev);
878 pci_msi_set_enable(dev, 0);
879 pci_intx_for_msi(dev, 1);
880 dev->msi_enabled = 0;
882 /* Return the device with MSI unmasked as initial states */
883 mask = msi_mask(desc->msi_attrib.multi_cap);
884 /* Keep cached state to be restored */
885 __pci_msi_desc_mask_irq(desc, mask, ~mask);
887 /* Restore dev->irq to its default pin-assertion irq */
888 dev->irq = desc->msi_attrib.default_irq;
889 pcibios_alloc_irq(dev);
892 void pci_disable_msi(struct pci_dev *dev)
894 if (!pci_msi_enable || !dev || !dev->msi_enabled)
897 pci_msi_shutdown(dev);
900 EXPORT_SYMBOL(pci_disable_msi);
903 * pci_msix_vec_count - return the number of device's MSI-X table entries
904 * @dev: pointer to the pci_dev data structure of MSI-X device function
905 * This function returns the number of device's MSI-X table entries and
906 * therefore the number of MSI-X vectors device is capable of sending.
907 * It returns a negative errno if the device is not capable of sending MSI-X
910 int pci_msix_vec_count(struct pci_dev *dev)
917 pci_read_config_word(dev, dev->msix_cap + PCI_MSIX_FLAGS, &control);
918 return msix_table_size(control);
920 EXPORT_SYMBOL(pci_msix_vec_count);
923 * pci_enable_msix - configure device's MSI-X capability structure
924 * @dev: pointer to the pci_dev data structure of MSI-X device function
925 * @entries: pointer to an array of MSI-X entries
926 * @nvec: number of MSI-X irqs requested for allocation by device driver
928 * Setup the MSI-X capability structure of device function with the number
929 * of requested irqs upon its software driver call to request for
930 * MSI-X mode enabled on its hardware device function. A return of zero
931 * indicates the successful configuration of MSI-X capability structure
932 * with new allocated MSI-X irqs. A return of < 0 indicates a failure.
933 * Or a return of > 0 indicates that driver request is exceeding the number
934 * of irqs or MSI-X vectors available. Driver should use the returned value to
935 * re-send its request.
937 int pci_enable_msix(struct pci_dev *dev, struct msix_entry *entries, int nvec)
942 if (!pci_msi_supported(dev, nvec))
948 nr_entries = pci_msix_vec_count(dev);
951 if (nvec > nr_entries)
954 /* Check for any invalid entries */
955 for (i = 0; i < nvec; i++) {
956 if (entries[i].entry >= nr_entries)
957 return -EINVAL; /* invalid entry */
958 for (j = i + 1; j < nvec; j++) {
959 if (entries[i].entry == entries[j].entry)
960 return -EINVAL; /* duplicate entry */
963 WARN_ON(!!dev->msix_enabled);
965 /* Check whether driver already requested for MSI irq */
966 if (dev->msi_enabled) {
967 dev_info(&dev->dev, "can't enable MSI-X (MSI IRQ already assigned)\n");
970 return msix_capability_init(dev, entries, nvec);
972 EXPORT_SYMBOL(pci_enable_msix);
974 void pci_msix_shutdown(struct pci_dev *dev)
976 struct msi_desc *entry;
978 if (!pci_msi_enable || !dev || !dev->msix_enabled)
981 /* Return the device with MSI-X masked as initial states */
982 for_each_pci_msi_entry(entry, dev) {
983 /* Keep cached states to be restored */
984 __pci_msix_desc_mask_irq(entry, 1);
987 pci_msix_clear_and_set_ctrl(dev, PCI_MSIX_FLAGS_ENABLE, 0);
988 pci_intx_for_msi(dev, 1);
989 dev->msix_enabled = 0;
990 pcibios_alloc_irq(dev);
993 void pci_disable_msix(struct pci_dev *dev)
995 if (!pci_msi_enable || !dev || !dev->msix_enabled)
998 pci_msix_shutdown(dev);
1001 EXPORT_SYMBOL(pci_disable_msix);
1003 void pci_no_msi(void)
1009 * pci_msi_enabled - is MSI enabled?
1011 * Returns true if MSI has not been disabled by the command-line option
1014 int pci_msi_enabled(void)
1016 return pci_msi_enable;
1018 EXPORT_SYMBOL(pci_msi_enabled);
1020 void pci_msi_init_pci_dev(struct pci_dev *dev)
1025 * pci_enable_msi_range - configure device's MSI capability structure
1026 * @dev: device to configure
1027 * @minvec: minimal number of interrupts to configure
1028 * @maxvec: maximum number of interrupts to configure
1030 * This function tries to allocate a maximum possible number of interrupts in a
1031 * range between @minvec and @maxvec. It returns a negative errno if an error
1032 * occurs. If it succeeds, it returns the actual number of interrupts allocated
1033 * and updates the @dev's irq member to the lowest new interrupt number;
1034 * the other interrupt numbers allocated to this device are consecutive.
1036 int pci_enable_msi_range(struct pci_dev *dev, int minvec, int maxvec)
1041 if (!pci_msi_supported(dev, minvec))
1044 WARN_ON(!!dev->msi_enabled);
1046 /* Check whether driver already requested MSI-X irqs */
1047 if (dev->msix_enabled) {
1049 "can't enable MSI (MSI-X already enabled)\n");
1053 if (maxvec < minvec)
1056 nvec = pci_msi_vec_count(dev);
1059 else if (nvec < minvec)
1061 else if (nvec > maxvec)
1065 rc = msi_capability_init(dev, nvec);
1068 } else if (rc > 0) {
1077 EXPORT_SYMBOL(pci_enable_msi_range);
1080 * pci_enable_msix_range - configure device's MSI-X capability structure
1081 * @dev: pointer to the pci_dev data structure of MSI-X device function
1082 * @entries: pointer to an array of MSI-X entries
1083 * @minvec: minimum number of MSI-X irqs requested
1084 * @maxvec: maximum number of MSI-X irqs requested
1086 * Setup the MSI-X capability structure of device function with a maximum
1087 * possible number of interrupts in the range between @minvec and @maxvec
1088 * upon its software driver call to request for MSI-X mode enabled on its
1089 * hardware device function. It returns a negative errno if an error occurs.
1090 * If it succeeds, it returns the actual number of interrupts allocated and
1091 * indicates the successful configuration of MSI-X capability structure
1092 * with new allocated MSI-X interrupts.
1094 int pci_enable_msix_range(struct pci_dev *dev, struct msix_entry *entries,
1095 int minvec, int maxvec)
1100 if (maxvec < minvec)
1104 rc = pci_enable_msix(dev, entries, nvec);
1107 } else if (rc > 0) {
1116 EXPORT_SYMBOL(pci_enable_msix_range);
1118 struct pci_dev *msi_desc_to_pci_dev(struct msi_desc *desc)
1120 return to_pci_dev(desc->dev);
1123 void *msi_desc_to_pci_sysdata(struct msi_desc *desc)
1125 struct pci_dev *dev = msi_desc_to_pci_dev(desc);
1127 return dev->bus->sysdata;
1129 EXPORT_SYMBOL_GPL(msi_desc_to_pci_sysdata);
1131 #ifdef CONFIG_PCI_MSI_IRQ_DOMAIN
1133 * pci_msi_domain_write_msg - Helper to write MSI message to PCI config space
1134 * @irq_data: Pointer to interrupt data of the MSI interrupt
1135 * @msg: Pointer to the message
1137 void pci_msi_domain_write_msg(struct irq_data *irq_data, struct msi_msg *msg)
1139 struct msi_desc *desc = irq_data_get_msi_desc(irq_data);
1142 * For MSI-X desc->irq is always equal to irq_data->irq. For
1143 * MSI only the first interrupt of MULTI MSI passes the test.
1145 if (desc->irq == irq_data->irq)
1146 __pci_write_msi_msg(desc, msg);
1150 * pci_msi_domain_calc_hwirq - Generate a unique ID for an MSI source
1151 * @dev: Pointer to the PCI device
1152 * @desc: Pointer to the msi descriptor
1154 * The ID number is only used within the irqdomain.
1156 irq_hw_number_t pci_msi_domain_calc_hwirq(struct pci_dev *dev,
1157 struct msi_desc *desc)
1159 return (irq_hw_number_t)desc->msi_attrib.entry_nr |
1160 PCI_DEVID(dev->bus->number, dev->devfn) << 11 |
1161 (pci_domain_nr(dev->bus) & 0xFFFFFFFF) << 27;
1164 static inline bool pci_msi_desc_is_multi_msi(struct msi_desc *desc)
1166 return !desc->msi_attrib.is_msix && desc->nvec_used > 1;
1170 * pci_msi_domain_check_cap - Verify that @domain supports the capabilities for @dev
1171 * @domain: The interrupt domain to check
1172 * @info: The domain info for verification
1173 * @dev: The device to check
1176 * 0 if the functionality is supported
1177 * 1 if Multi MSI is requested, but the domain does not support it
1178 * -ENOTSUPP otherwise
1180 int pci_msi_domain_check_cap(struct irq_domain *domain,
1181 struct msi_domain_info *info, struct device *dev)
1183 struct msi_desc *desc = first_pci_msi_entry(to_pci_dev(dev));
1185 /* Special handling to support pci_enable_msi_range() */
1186 if (pci_msi_desc_is_multi_msi(desc) &&
1187 !(info->flags & MSI_FLAG_MULTI_PCI_MSI))
1189 else if (desc->msi_attrib.is_msix && !(info->flags & MSI_FLAG_PCI_MSIX))
1195 static int pci_msi_domain_handle_error(struct irq_domain *domain,
1196 struct msi_desc *desc, int error)
1198 /* Special handling to support pci_enable_msi_range() */
1199 if (pci_msi_desc_is_multi_msi(desc) && error == -ENOSPC)
1205 #ifdef GENERIC_MSI_DOMAIN_OPS
1206 static void pci_msi_domain_set_desc(msi_alloc_info_t *arg,
1207 struct msi_desc *desc)
1210 arg->hwirq = pci_msi_domain_calc_hwirq(msi_desc_to_pci_dev(desc),
1214 #define pci_msi_domain_set_desc NULL
1217 static struct msi_domain_ops pci_msi_domain_ops_default = {
1218 .set_desc = pci_msi_domain_set_desc,
1219 .msi_check = pci_msi_domain_check_cap,
1220 .handle_error = pci_msi_domain_handle_error,
1223 static void pci_msi_domain_update_dom_ops(struct msi_domain_info *info)
1225 struct msi_domain_ops *ops = info->ops;
1228 info->ops = &pci_msi_domain_ops_default;
1230 if (ops->set_desc == NULL)
1231 ops->set_desc = pci_msi_domain_set_desc;
1232 if (ops->msi_check == NULL)
1233 ops->msi_check = pci_msi_domain_check_cap;
1234 if (ops->handle_error == NULL)
1235 ops->handle_error = pci_msi_domain_handle_error;
1239 static void pci_msi_domain_update_chip_ops(struct msi_domain_info *info)
1241 struct irq_chip *chip = info->chip;
1244 if (!chip->irq_write_msi_msg)
1245 chip->irq_write_msi_msg = pci_msi_domain_write_msg;
1246 if (!chip->irq_mask)
1247 chip->irq_mask = pci_msi_mask_irq;
1248 if (!chip->irq_unmask)
1249 chip->irq_unmask = pci_msi_unmask_irq;
1253 * pci_msi_create_irq_domain - Creat a MSI interrupt domain
1254 * @node: Optional device-tree node of the interrupt controller
1255 * @info: MSI domain info
1256 * @parent: Parent irq domain
1258 * Updates the domain and chip ops and creates a MSI interrupt domain.
1261 * A domain pointer or NULL in case of failure.
1263 struct irq_domain *pci_msi_create_irq_domain(struct device_node *node,
1264 struct msi_domain_info *info,
1265 struct irq_domain *parent)
1267 struct irq_domain *domain;
1269 if (info->flags & MSI_FLAG_USE_DEF_DOM_OPS)
1270 pci_msi_domain_update_dom_ops(info);
1271 if (info->flags & MSI_FLAG_USE_DEF_CHIP_OPS)
1272 pci_msi_domain_update_chip_ops(info);
1274 domain = msi_create_irq_domain(node, info, parent);
1278 domain->bus_token = DOMAIN_BUS_PCI_MSI;
1283 * pci_msi_domain_alloc_irqs - Allocate interrupts for @dev in @domain
1284 * @domain: The interrupt domain to allocate from
1285 * @dev: The device for which to allocate
1286 * @nvec: The number of interrupts to allocate
1287 * @type: Unused to allow simpler migration from the arch_XXX interfaces
1290 * A virtual interrupt number or an error code in case of failure
1292 int pci_msi_domain_alloc_irqs(struct irq_domain *domain, struct pci_dev *dev,
1295 return msi_domain_alloc_irqs(domain, &dev->dev, nvec);
1299 * pci_msi_domain_free_irqs - Free interrupts for @dev in @domain
1300 * @domain: The interrupt domain
1301 * @dev: The device for which to free interrupts
1303 void pci_msi_domain_free_irqs(struct irq_domain *domain, struct pci_dev *dev)
1305 msi_domain_free_irqs(domain, &dev->dev);
1309 * pci_msi_create_default_irq_domain - Create a default MSI interrupt domain
1310 * @node: Optional device-tree node of the interrupt controller
1311 * @info: MSI domain info
1312 * @parent: Parent irq domain
1314 * Returns: A domain pointer or NULL in case of failure. If successful
1315 * the default PCI/MSI irqdomain pointer is updated.
1317 struct irq_domain *pci_msi_create_default_irq_domain(struct device_node *node,
1318 struct msi_domain_info *info, struct irq_domain *parent)
1320 struct irq_domain *domain;
1322 mutex_lock(&pci_msi_domain_lock);
1323 if (pci_msi_default_domain) {
1324 pr_err("PCI: default irq domain for PCI MSI has already been created.\n");
1327 domain = pci_msi_create_irq_domain(node, info, parent);
1328 pci_msi_default_domain = domain;
1330 mutex_unlock(&pci_msi_domain_lock);
1334 #endif /* CONFIG_PCI_MSI_IRQ_DOMAIN */