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/init.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>
20 #include <asm/errno.h>
26 static int pci_msi_enable = 1;
30 #ifndef arch_msi_check_device
31 int arch_msi_check_device(struct pci_dev *dev, int nvec, int type)
37 #ifndef arch_setup_msi_irqs
38 int arch_setup_msi_irqs(struct pci_dev *dev, int nvec, int type)
40 struct msi_desc *entry;
43 list_for_each_entry(entry, &dev->msi_list, list) {
44 ret = arch_setup_msi_irq(dev, entry);
55 #ifndef arch_teardown_msi_irqs
56 void arch_teardown_msi_irqs(struct pci_dev *dev)
58 struct msi_desc *entry;
60 list_for_each_entry(entry, &dev->msi_list, list) {
62 arch_teardown_msi_irq(entry->irq);
67 static void __msi_set_enable(struct pci_dev *dev, int pos, int enable)
72 pci_read_config_word(dev, pos + PCI_MSI_FLAGS, &control);
73 control &= ~PCI_MSI_FLAGS_ENABLE;
75 control |= PCI_MSI_FLAGS_ENABLE;
76 pci_write_config_word(dev, pos + PCI_MSI_FLAGS, control);
80 static void msi_set_enable(struct pci_dev *dev, int enable)
82 __msi_set_enable(dev, pci_find_capability(dev, PCI_CAP_ID_MSI), enable);
85 static void msix_set_enable(struct pci_dev *dev, int enable)
90 pos = pci_find_capability(dev, PCI_CAP_ID_MSIX);
92 pci_read_config_word(dev, pos + PCI_MSIX_FLAGS, &control);
93 control &= ~PCI_MSIX_FLAGS_ENABLE;
95 control |= PCI_MSIX_FLAGS_ENABLE;
96 pci_write_config_word(dev, pos + PCI_MSIX_FLAGS, control);
100 static inline __attribute_const__ u32 msi_mask(unsigned x)
102 /* Don't shift by >= width of type */
105 return (1 << (1 << x)) - 1;
108 static void msix_flush_writes(struct irq_desc *desc)
110 struct msi_desc *entry;
112 entry = get_irq_desc_msi(desc);
113 BUG_ON(!entry || !entry->dev);
114 switch (entry->msi_attrib.type) {
118 case PCI_CAP_ID_MSIX:
120 int offset = entry->msi_attrib.entry_nr * PCI_MSIX_ENTRY_SIZE +
121 PCI_MSIX_ENTRY_VECTOR_CTRL_OFFSET;
122 readl(entry->mask_base + offset);
132 * PCI 2.3 does not specify mask bits for each MSI interrupt. Attempting to
133 * mask all MSI interrupts by clearing the MSI enable bit does not work
134 * reliably as devices without an INTx disable bit will then generate a
135 * level IRQ which will never be cleared.
137 * Returns 1 if it succeeded in masking the interrupt and 0 if the device
138 * doesn't support MSI masking.
140 static int msi_set_mask_bits(struct irq_desc *desc, u32 mask, u32 flag)
142 struct msi_desc *entry;
144 entry = get_irq_desc_msi(desc);
145 BUG_ON(!entry || !entry->dev);
146 switch (entry->msi_attrib.type) {
148 if (entry->msi_attrib.maskbit) {
152 pos = (long)entry->mask_base;
153 pci_read_config_dword(entry->dev, pos, &mask_bits);
154 mask_bits &= ~(mask);
155 mask_bits |= flag & mask;
156 pci_write_config_dword(entry->dev, pos, mask_bits);
161 case PCI_CAP_ID_MSIX:
163 int offset = entry->msi_attrib.entry_nr * PCI_MSIX_ENTRY_SIZE +
164 PCI_MSIX_ENTRY_VECTOR_CTRL_OFFSET;
165 writel(flag, entry->mask_base + offset);
166 readl(entry->mask_base + offset);
173 entry->msi_attrib.masked = !!flag;
177 void read_msi_msg_desc(struct irq_desc *desc, struct msi_msg *msg)
179 struct msi_desc *entry = get_irq_desc_msi(desc);
180 switch(entry->msi_attrib.type) {
183 struct pci_dev *dev = entry->dev;
184 int pos = entry->msi_attrib.pos;
187 pci_read_config_dword(dev, msi_lower_address_reg(pos),
189 if (entry->msi_attrib.is_64) {
190 pci_read_config_dword(dev, msi_upper_address_reg(pos),
192 pci_read_config_word(dev, msi_data_reg(pos, 1), &data);
195 pci_read_config_word(dev, msi_data_reg(pos, 0), &data);
200 case PCI_CAP_ID_MSIX:
203 base = entry->mask_base +
204 entry->msi_attrib.entry_nr * PCI_MSIX_ENTRY_SIZE;
206 msg->address_lo = readl(base + PCI_MSIX_ENTRY_LOWER_ADDR_OFFSET);
207 msg->address_hi = readl(base + PCI_MSIX_ENTRY_UPPER_ADDR_OFFSET);
208 msg->data = readl(base + PCI_MSIX_ENTRY_DATA_OFFSET);
216 void read_msi_msg(unsigned int irq, struct msi_msg *msg)
218 struct irq_desc *desc = irq_to_desc(irq);
220 read_msi_msg_desc(desc, msg);
223 void write_msi_msg_desc(struct irq_desc *desc, struct msi_msg *msg)
225 struct msi_desc *entry = get_irq_desc_msi(desc);
226 switch (entry->msi_attrib.type) {
229 struct pci_dev *dev = entry->dev;
230 int pos = entry->msi_attrib.pos;
232 pci_write_config_dword(dev, msi_lower_address_reg(pos),
234 if (entry->msi_attrib.is_64) {
235 pci_write_config_dword(dev, msi_upper_address_reg(pos),
237 pci_write_config_word(dev, msi_data_reg(pos, 1),
240 pci_write_config_word(dev, msi_data_reg(pos, 0),
245 case PCI_CAP_ID_MSIX:
248 base = entry->mask_base +
249 entry->msi_attrib.entry_nr * PCI_MSIX_ENTRY_SIZE;
251 writel(msg->address_lo,
252 base + PCI_MSIX_ENTRY_LOWER_ADDR_OFFSET);
253 writel(msg->address_hi,
254 base + PCI_MSIX_ENTRY_UPPER_ADDR_OFFSET);
255 writel(msg->data, base + PCI_MSIX_ENTRY_DATA_OFFSET);
264 void write_msi_msg(unsigned int irq, struct msi_msg *msg)
266 struct irq_desc *desc = irq_to_desc(irq);
268 write_msi_msg_desc(desc, msg);
271 void mask_msi_irq(unsigned int irq)
273 struct irq_desc *desc = irq_to_desc(irq);
275 msi_set_mask_bits(desc, 1, 1);
276 msix_flush_writes(desc);
279 void unmask_msi_irq(unsigned int irq)
281 struct irq_desc *desc = irq_to_desc(irq);
283 msi_set_mask_bits(desc, 1, 0);
284 msix_flush_writes(desc);
287 static int msi_free_irqs(struct pci_dev* dev);
289 static struct msi_desc* alloc_msi_entry(void)
291 struct msi_desc *entry;
293 entry = kzalloc(sizeof(struct msi_desc), GFP_KERNEL);
297 INIT_LIST_HEAD(&entry->list);
304 static void pci_intx_for_msi(struct pci_dev *dev, int enable)
306 if (!(dev->dev_flags & PCI_DEV_FLAGS_MSI_INTX_DISABLE_BUG))
307 pci_intx(dev, enable);
310 static void __pci_restore_msi_state(struct pci_dev *dev)
314 struct msi_desc *entry;
316 if (!dev->msi_enabled)
319 entry = get_irq_msi(dev->irq);
320 pos = entry->msi_attrib.pos;
322 pci_intx_for_msi(dev, 0);
323 msi_set_enable(dev, 0);
324 write_msi_msg(dev->irq, &entry->msg);
325 if (entry->msi_attrib.maskbit) {
326 struct irq_desc *desc = irq_to_desc(dev->irq);
327 msi_set_mask_bits(desc, entry->msi_attrib.maskbits_mask,
328 entry->msi_attrib.masked);
331 pci_read_config_word(dev, pos + PCI_MSI_FLAGS, &control);
332 control &= ~PCI_MSI_FLAGS_QSIZE;
333 control |= PCI_MSI_FLAGS_ENABLE;
334 pci_write_config_word(dev, pos + PCI_MSI_FLAGS, control);
337 static void __pci_restore_msix_state(struct pci_dev *dev)
340 struct msi_desc *entry;
343 if (!dev->msix_enabled)
346 /* route the table */
347 pci_intx_for_msi(dev, 0);
348 msix_set_enable(dev, 0);
350 list_for_each_entry(entry, &dev->msi_list, list) {
351 struct irq_desc *desc = irq_to_desc(entry->irq);
352 write_msi_msg(entry->irq, &entry->msg);
353 msi_set_mask_bits(desc, 1, entry->msi_attrib.masked);
356 BUG_ON(list_empty(&dev->msi_list));
357 entry = list_entry(dev->msi_list.next, struct msi_desc, list);
358 pos = entry->msi_attrib.pos;
359 pci_read_config_word(dev, pos + PCI_MSIX_FLAGS, &control);
360 control &= ~PCI_MSIX_FLAGS_MASKALL;
361 control |= PCI_MSIX_FLAGS_ENABLE;
362 pci_write_config_word(dev, pos + PCI_MSIX_FLAGS, control);
365 void pci_restore_msi_state(struct pci_dev *dev)
367 __pci_restore_msi_state(dev);
368 __pci_restore_msix_state(dev);
370 EXPORT_SYMBOL_GPL(pci_restore_msi_state);
373 * msi_capability_init - configure device's MSI capability structure
374 * @dev: pointer to the pci_dev data structure of MSI device function
376 * Setup the MSI capability structure of device function with a single
377 * MSI irq, regardless of device function is capable of handling
378 * multiple messages. A return of zero indicates the successful setup
379 * of an entry zero with the new MSI irq or non-zero for otherwise.
381 static int msi_capability_init(struct pci_dev *dev)
383 struct msi_desc *entry;
387 msi_set_enable(dev, 0); /* Ensure msi is disabled as I set it up */
389 pos = pci_find_capability(dev, PCI_CAP_ID_MSI);
390 pci_read_config_word(dev, msi_control_reg(pos), &control);
391 /* MSI Entry Initialization */
392 entry = alloc_msi_entry();
396 entry->msi_attrib.type = PCI_CAP_ID_MSI;
397 entry->msi_attrib.is_64 = is_64bit_address(control);
398 entry->msi_attrib.entry_nr = 0;
399 entry->msi_attrib.maskbit = is_mask_bit_support(control);
400 entry->msi_attrib.masked = 1;
401 entry->msi_attrib.default_irq = dev->irq; /* Save IOAPIC IRQ */
402 entry->msi_attrib.pos = pos;
404 if (entry->msi_attrib.maskbit) {
405 unsigned int base, maskbits, temp;
407 base = msi_mask_bits_reg(pos, entry->msi_attrib.is_64);
408 entry->mask_base = (void __iomem *)(long)base;
410 /* All MSIs are unmasked by default, Mask them all */
411 pci_read_config_dword(dev, base, &maskbits);
412 temp = msi_mask((control & PCI_MSI_FLAGS_QMASK) >> 1);
414 pci_write_config_dword(dev, base, maskbits);
415 entry->msi_attrib.maskbits_mask = temp;
417 list_add_tail(&entry->list, &dev->msi_list);
419 /* Configure MSI capability structure */
420 ret = arch_setup_msi_irqs(dev, 1, PCI_CAP_ID_MSI);
426 /* Set MSI enabled bits */
427 pci_intx_for_msi(dev, 0);
428 msi_set_enable(dev, 1);
429 dev->msi_enabled = 1;
431 dev->irq = entry->irq;
436 * msix_capability_init - configure device's MSI-X capability
437 * @dev: pointer to the pci_dev data structure of MSI-X device function
438 * @entries: pointer to an array of struct msix_entry entries
439 * @nvec: number of @entries
441 * Setup the MSI-X capability structure of device function with a
442 * single MSI-X irq. A return of zero indicates the successful setup of
443 * requested MSI-X entries with allocated irqs or non-zero for otherwise.
445 static int msix_capability_init(struct pci_dev *dev,
446 struct msix_entry *entries, int nvec)
448 struct msi_desc *entry;
449 int pos, i, j, nr_entries, ret;
450 unsigned long phys_addr;
456 msix_set_enable(dev, 0);/* Ensure msix is disabled as I set it up */
458 pos = pci_find_capability(dev, PCI_CAP_ID_MSIX);
459 /* Request & Map MSI-X table region */
460 pci_read_config_word(dev, msi_control_reg(pos), &control);
461 nr_entries = multi_msix_capable(control);
463 pci_read_config_dword(dev, msix_table_offset_reg(pos), &table_offset);
464 bir = (u8)(table_offset & PCI_MSIX_FLAGS_BIRMASK);
465 table_offset &= ~PCI_MSIX_FLAGS_BIRMASK;
466 phys_addr = pci_resource_start (dev, bir) + table_offset;
467 base = ioremap_nocache(phys_addr, nr_entries * PCI_MSIX_ENTRY_SIZE);
471 /* MSI-X Table Initialization */
472 for (i = 0; i < nvec; i++) {
473 entry = alloc_msi_entry();
477 j = entries[i].entry;
478 entry->msi_attrib.type = PCI_CAP_ID_MSIX;
479 entry->msi_attrib.is_64 = 1;
480 entry->msi_attrib.entry_nr = j;
481 entry->msi_attrib.maskbit = 1;
482 entry->msi_attrib.masked = 1;
483 entry->msi_attrib.default_irq = dev->irq;
484 entry->msi_attrib.pos = pos;
486 entry->mask_base = base;
488 list_add_tail(&entry->list, &dev->msi_list);
491 ret = arch_setup_msi_irqs(dev, nvec, PCI_CAP_ID_MSIX);
493 /* If we had some success report the number of irqs
494 * we succeeded in setting up. */
496 list_for_each_entry(entry, &dev->msi_list, list) {
497 if (entry->irq != 0) {
512 list_for_each_entry(entry, &dev->msi_list, list) {
513 entries[i].vector = entry->irq;
514 set_irq_msi(entry->irq, entry);
517 /* Set MSI-X enabled bits */
518 pci_intx_for_msi(dev, 0);
519 msix_set_enable(dev, 1);
520 dev->msix_enabled = 1;
526 * pci_msi_check_device - check whether MSI may be enabled on a device
527 * @dev: pointer to the pci_dev data structure of MSI device function
528 * @nvec: how many MSIs have been requested ?
529 * @type: are we checking for MSI or MSI-X ?
531 * Look at global flags, the device itself, and its parent busses
532 * to determine if MSI/-X are supported for the device. If MSI/-X is
533 * supported return 0, else return an error code.
535 static int pci_msi_check_device(struct pci_dev* dev, int nvec, int type)
540 /* MSI must be globally enabled and supported by the device */
541 if (!pci_msi_enable || !dev || dev->no_msi)
545 * You can't ask to have 0 or less MSIs configured.
547 * b) the list manipulation code assumes nvec >= 1.
552 /* Any bridge which does NOT route MSI transactions from it's
553 * secondary bus to it's primary bus must set NO_MSI flag on
554 * the secondary pci_bus.
555 * We expect only arch-specific PCI host bus controller driver
556 * or quirks for specific PCI bridges to be setting NO_MSI.
558 for (bus = dev->bus; bus; bus = bus->parent)
559 if (bus->bus_flags & PCI_BUS_FLAGS_NO_MSI)
562 ret = arch_msi_check_device(dev, nvec, type);
566 if (!pci_find_capability(dev, type))
573 * pci_enable_msi - configure device's MSI capability structure
574 * @dev: pointer to the pci_dev data structure of MSI device function
576 * Setup the MSI capability structure of device function with
577 * a single MSI irq upon its software driver call to request for
578 * MSI mode enabled on its hardware device function. A return of zero
579 * indicates the successful setup of an entry zero with the new MSI
580 * irq or non-zero for otherwise.
582 int pci_enable_msi(struct pci_dev* dev)
586 status = pci_msi_check_device(dev, 1, PCI_CAP_ID_MSI);
590 WARN_ON(!!dev->msi_enabled);
592 /* Check whether driver already requested for MSI-X irqs */
593 if (dev->msix_enabled) {
594 dev_info(&dev->dev, "can't enable MSI "
595 "(MSI-X already enabled)\n");
598 status = msi_capability_init(dev);
601 EXPORT_SYMBOL(pci_enable_msi);
603 void pci_msi_shutdown(struct pci_dev* dev)
605 struct msi_desc *entry;
607 if (!pci_msi_enable || !dev || !dev->msi_enabled)
610 msi_set_enable(dev, 0);
611 pci_intx_for_msi(dev, 1);
612 dev->msi_enabled = 0;
614 BUG_ON(list_empty(&dev->msi_list));
615 entry = list_entry(dev->msi_list.next, struct msi_desc, list);
616 /* Return the the pci reset with msi irqs unmasked */
617 if (entry->msi_attrib.maskbit) {
618 u32 mask = entry->msi_attrib.maskbits_mask;
619 struct irq_desc *desc = irq_to_desc(dev->irq);
620 msi_set_mask_bits(desc, mask, ~mask);
622 if (!entry->dev || entry->msi_attrib.type != PCI_CAP_ID_MSI)
625 /* Restore dev->irq to its default pin-assertion irq */
626 dev->irq = entry->msi_attrib.default_irq;
628 void pci_disable_msi(struct pci_dev* dev)
630 struct msi_desc *entry;
632 if (!pci_msi_enable || !dev || !dev->msi_enabled)
635 pci_msi_shutdown(dev);
637 entry = list_entry(dev->msi_list.next, struct msi_desc, list);
638 if (!entry->dev || entry->msi_attrib.type != PCI_CAP_ID_MSI)
643 EXPORT_SYMBOL(pci_disable_msi);
645 static int msi_free_irqs(struct pci_dev* dev)
647 struct msi_desc *entry, *tmp;
649 list_for_each_entry(entry, &dev->msi_list, list) {
651 BUG_ON(irq_has_action(entry->irq));
654 arch_teardown_msi_irqs(dev);
656 list_for_each_entry_safe(entry, tmp, &dev->msi_list, list) {
657 if (entry->msi_attrib.type == PCI_CAP_ID_MSIX) {
658 writel(1, entry->mask_base + entry->msi_attrib.entry_nr
659 * PCI_MSIX_ENTRY_SIZE
660 + PCI_MSIX_ENTRY_VECTOR_CTRL_OFFSET);
662 if (list_is_last(&entry->list, &dev->msi_list))
663 iounmap(entry->mask_base);
665 list_del(&entry->list);
673 * pci_msix_table_size - return the number of device's MSI-X table entries
674 * @dev: pointer to the pci_dev data structure of MSI-X device function
676 int pci_msix_table_size(struct pci_dev *dev)
681 pos = pci_find_capability(dev, PCI_CAP_ID_MSIX);
685 pci_read_config_word(dev, msi_control_reg(pos), &control);
686 return multi_msix_capable(control);
690 * pci_enable_msix - configure device's MSI-X capability structure
691 * @dev: pointer to the pci_dev data structure of MSI-X device function
692 * @entries: pointer to an array of MSI-X entries
693 * @nvec: number of MSI-X irqs requested for allocation by device driver
695 * Setup the MSI-X capability structure of device function with the number
696 * of requested irqs upon its software driver call to request for
697 * MSI-X mode enabled on its hardware device function. A return of zero
698 * indicates the successful configuration of MSI-X capability structure
699 * with new allocated MSI-X irqs. A return of < 0 indicates a failure.
700 * Or a return of > 0 indicates that driver request is exceeding the number
701 * of irqs available. Driver should use the returned value to re-send
704 int pci_enable_msix(struct pci_dev* dev, struct msix_entry *entries, int nvec)
706 int status, nr_entries;
712 status = pci_msi_check_device(dev, nvec, PCI_CAP_ID_MSIX);
716 nr_entries = pci_msix_table_size(dev);
717 if (nvec > nr_entries)
720 /* Check for any invalid entries */
721 for (i = 0; i < nvec; i++) {
722 if (entries[i].entry >= nr_entries)
723 return -EINVAL; /* invalid entry */
724 for (j = i + 1; j < nvec; j++) {
725 if (entries[i].entry == entries[j].entry)
726 return -EINVAL; /* duplicate entry */
729 WARN_ON(!!dev->msix_enabled);
731 /* Check whether driver already requested for MSI irq */
732 if (dev->msi_enabled) {
733 dev_info(&dev->dev, "can't enable MSI-X "
734 "(MSI IRQ already assigned)\n");
737 status = msix_capability_init(dev, entries, nvec);
740 EXPORT_SYMBOL(pci_enable_msix);
742 static void msix_free_all_irqs(struct pci_dev *dev)
747 void pci_msix_shutdown(struct pci_dev* dev)
749 if (!pci_msi_enable || !dev || !dev->msix_enabled)
752 msix_set_enable(dev, 0);
753 pci_intx_for_msi(dev, 1);
754 dev->msix_enabled = 0;
756 void pci_disable_msix(struct pci_dev* dev)
758 if (!pci_msi_enable || !dev || !dev->msix_enabled)
761 pci_msix_shutdown(dev);
763 msix_free_all_irqs(dev);
765 EXPORT_SYMBOL(pci_disable_msix);
768 * msi_remove_pci_irq_vectors - reclaim MSI(X) irqs to unused state
769 * @dev: pointer to the pci_dev data structure of MSI(X) device function
771 * Being called during hotplug remove, from which the device function
772 * is hot-removed. All previous assigned MSI/MSI-X irqs, if
773 * allocated for this device function, are reclaimed to unused state,
774 * which may be used later on.
776 void msi_remove_pci_irq_vectors(struct pci_dev* dev)
778 if (!pci_msi_enable || !dev)
781 if (dev->msi_enabled)
784 if (dev->msix_enabled)
785 msix_free_all_irqs(dev);
788 void pci_no_msi(void)
794 * pci_msi_enabled - is MSI enabled?
796 * Returns true if MSI has not been disabled by the command-line option
799 int pci_msi_enabled(void)
801 return pci_msi_enable;
803 EXPORT_SYMBOL(pci_msi_enabled);
805 void pci_msi_init_pci_dev(struct pci_dev *dev)
807 INIT_LIST_HEAD(&dev->msi_list);