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 int __attribute__ ((weak))
31 arch_msi_check_device(struct pci_dev *dev, int nvec, int type)
36 int __attribute__ ((weak))
37 arch_setup_msi_irq(struct pci_dev *dev, struct msi_desc *entry)
42 int __attribute__ ((weak))
43 arch_setup_msi_irqs(struct pci_dev *dev, int nvec, int type)
45 struct msi_desc *entry;
48 list_for_each_entry(entry, &dev->msi_list, list) {
49 ret = arch_setup_msi_irq(dev, entry);
57 void __attribute__ ((weak)) arch_teardown_msi_irq(unsigned int irq)
62 void __attribute__ ((weak))
63 arch_teardown_msi_irqs(struct pci_dev *dev)
65 struct msi_desc *entry;
67 list_for_each_entry(entry, &dev->msi_list, list) {
69 arch_teardown_msi_irq(entry->irq);
73 static void __msi_set_enable(struct pci_dev *dev, int pos, int enable)
78 pci_read_config_word(dev, pos + PCI_MSI_FLAGS, &control);
79 control &= ~PCI_MSI_FLAGS_ENABLE;
81 control |= PCI_MSI_FLAGS_ENABLE;
82 pci_write_config_word(dev, pos + PCI_MSI_FLAGS, control);
86 static void msi_set_enable(struct pci_dev *dev, int enable)
88 __msi_set_enable(dev, pci_find_capability(dev, PCI_CAP_ID_MSI), enable);
91 static void msix_set_enable(struct pci_dev *dev, int enable)
96 pos = pci_find_capability(dev, PCI_CAP_ID_MSIX);
98 pci_read_config_word(dev, pos + PCI_MSIX_FLAGS, &control);
99 control &= ~PCI_MSIX_FLAGS_ENABLE;
101 control |= PCI_MSIX_FLAGS_ENABLE;
102 pci_write_config_word(dev, pos + PCI_MSIX_FLAGS, control);
106 static void msix_flush_writes(unsigned int irq)
108 struct msi_desc *entry;
110 entry = get_irq_msi(irq);
111 BUG_ON(!entry || !entry->dev);
112 switch (entry->msi_attrib.type) {
116 case PCI_CAP_ID_MSIX:
118 int offset = entry->msi_attrib.entry_nr * PCI_MSIX_ENTRY_SIZE +
119 PCI_MSIX_ENTRY_VECTOR_CTRL_OFFSET;
120 readl(entry->mask_base + offset);
130 * PCI 2.3 does not specify mask bits for each MSI interrupt. Attempting to
131 * mask all MSI interrupts by clearing the MSI enable bit does not work
132 * reliably as devices without an INTx disable bit will then generate a
133 * level IRQ which will never be cleared.
135 * Returns 1 if it succeeded in masking the interrupt and 0 if the device
136 * doesn't support MSI masking.
138 static int msi_set_mask_bits(unsigned int irq, u32 mask, u32 flag)
140 struct msi_desc *entry;
142 entry = get_irq_msi(irq);
143 BUG_ON(!entry || !entry->dev);
144 switch (entry->msi_attrib.type) {
146 if (entry->msi_attrib.maskbit) {
150 pos = (long)entry->mask_base;
151 pci_read_config_dword(entry->dev, pos, &mask_bits);
152 mask_bits &= ~(mask);
153 mask_bits |= flag & mask;
154 pci_write_config_dword(entry->dev, pos, mask_bits);
159 case PCI_CAP_ID_MSIX:
161 int offset = entry->msi_attrib.entry_nr * PCI_MSIX_ENTRY_SIZE +
162 PCI_MSIX_ENTRY_VECTOR_CTRL_OFFSET;
163 writel(flag, entry->mask_base + offset);
164 readl(entry->mask_base + offset);
171 entry->msi_attrib.masked = !!flag;
175 void read_msi_msg(unsigned int irq, struct msi_msg *msg)
177 struct msi_desc *entry = get_irq_msi(irq);
178 switch(entry->msi_attrib.type) {
181 struct pci_dev *dev = entry->dev;
182 int pos = entry->msi_attrib.pos;
185 pci_read_config_dword(dev, msi_lower_address_reg(pos),
187 if (entry->msi_attrib.is_64) {
188 pci_read_config_dword(dev, msi_upper_address_reg(pos),
190 pci_read_config_word(dev, msi_data_reg(pos, 1), &data);
193 pci_read_config_word(dev, msi_data_reg(pos, 0), &data);
198 case PCI_CAP_ID_MSIX:
201 base = entry->mask_base +
202 entry->msi_attrib.entry_nr * PCI_MSIX_ENTRY_SIZE;
204 msg->address_lo = readl(base + PCI_MSIX_ENTRY_LOWER_ADDR_OFFSET);
205 msg->address_hi = readl(base + PCI_MSIX_ENTRY_UPPER_ADDR_OFFSET);
206 msg->data = readl(base + PCI_MSIX_ENTRY_DATA_OFFSET);
214 void write_msi_msg(unsigned int irq, struct msi_msg *msg)
216 struct msi_desc *entry = get_irq_msi(irq);
217 switch (entry->msi_attrib.type) {
220 struct pci_dev *dev = entry->dev;
221 int pos = entry->msi_attrib.pos;
223 pci_write_config_dword(dev, msi_lower_address_reg(pos),
225 if (entry->msi_attrib.is_64) {
226 pci_write_config_dword(dev, msi_upper_address_reg(pos),
228 pci_write_config_word(dev, msi_data_reg(pos, 1),
231 pci_write_config_word(dev, msi_data_reg(pos, 0),
236 case PCI_CAP_ID_MSIX:
239 base = entry->mask_base +
240 entry->msi_attrib.entry_nr * PCI_MSIX_ENTRY_SIZE;
242 writel(msg->address_lo,
243 base + PCI_MSIX_ENTRY_LOWER_ADDR_OFFSET);
244 writel(msg->address_hi,
245 base + PCI_MSIX_ENTRY_UPPER_ADDR_OFFSET);
246 writel(msg->data, base + PCI_MSIX_ENTRY_DATA_OFFSET);
255 void mask_msi_irq(unsigned int irq)
257 msi_set_mask_bits(irq, 1, 1);
258 msix_flush_writes(irq);
261 void unmask_msi_irq(unsigned int irq)
263 msi_set_mask_bits(irq, 1, 0);
264 msix_flush_writes(irq);
267 static int msi_free_irqs(struct pci_dev* dev);
270 static struct msi_desc* alloc_msi_entry(void)
272 struct msi_desc *entry;
274 entry = kzalloc(sizeof(struct msi_desc), GFP_KERNEL);
278 INIT_LIST_HEAD(&entry->list);
285 static void pci_intx_for_msi(struct pci_dev *dev, int enable)
287 if (!(dev->dev_flags & PCI_DEV_FLAGS_MSI_INTX_DISABLE_BUG))
288 pci_intx(dev, enable);
291 static void __pci_restore_msi_state(struct pci_dev *dev)
295 struct msi_desc *entry;
297 if (!dev->msi_enabled)
300 entry = get_irq_msi(dev->irq);
301 pos = entry->msi_attrib.pos;
303 pci_intx_for_msi(dev, 0);
304 msi_set_enable(dev, 0);
305 write_msi_msg(dev->irq, &entry->msg);
306 if (entry->msi_attrib.maskbit)
307 msi_set_mask_bits(dev->irq, entry->msi_attrib.maskbits_mask,
308 entry->msi_attrib.masked);
310 pci_read_config_word(dev, pos + PCI_MSI_FLAGS, &control);
311 control &= ~PCI_MSI_FLAGS_QSIZE;
312 control |= PCI_MSI_FLAGS_ENABLE;
313 pci_write_config_word(dev, pos + PCI_MSI_FLAGS, control);
316 static void __pci_restore_msix_state(struct pci_dev *dev)
319 struct msi_desc *entry;
322 if (!dev->msix_enabled)
325 /* route the table */
326 pci_intx_for_msi(dev, 0);
327 msix_set_enable(dev, 0);
329 list_for_each_entry(entry, &dev->msi_list, list) {
330 write_msi_msg(entry->irq, &entry->msg);
331 msi_set_mask_bits(entry->irq, 1, entry->msi_attrib.masked);
334 BUG_ON(list_empty(&dev->msi_list));
335 entry = list_entry(dev->msi_list.next, struct msi_desc, list);
336 pos = entry->msi_attrib.pos;
337 pci_read_config_word(dev, pos + PCI_MSIX_FLAGS, &control);
338 control &= ~PCI_MSIX_FLAGS_MASKALL;
339 control |= PCI_MSIX_FLAGS_ENABLE;
340 pci_write_config_word(dev, pos + PCI_MSIX_FLAGS, control);
343 void pci_restore_msi_state(struct pci_dev *dev)
345 __pci_restore_msi_state(dev);
346 __pci_restore_msix_state(dev);
348 EXPORT_SYMBOL_GPL(pci_restore_msi_state);
351 * msi_capability_init - configure device's MSI capability structure
352 * @dev: pointer to the pci_dev data structure of MSI device function
354 * Setup the MSI capability structure of device function with a single
355 * MSI irq, regardless of device function is capable of handling
356 * multiple messages. A return of zero indicates the successful setup
357 * of an entry zero with the new MSI irq or non-zero for otherwise.
359 static int msi_capability_init(struct pci_dev *dev)
361 struct msi_desc *entry;
365 msi_set_enable(dev, 0); /* Ensure msi is disabled as I set it up */
367 pos = pci_find_capability(dev, PCI_CAP_ID_MSI);
368 pci_read_config_word(dev, msi_control_reg(pos), &control);
369 /* MSI Entry Initialization */
370 entry = alloc_msi_entry();
374 entry->msi_attrib.type = PCI_CAP_ID_MSI;
375 entry->msi_attrib.is_64 = is_64bit_address(control);
376 entry->msi_attrib.entry_nr = 0;
377 entry->msi_attrib.maskbit = is_mask_bit_support(control);
378 entry->msi_attrib.masked = 1;
379 entry->msi_attrib.default_irq = dev->irq; /* Save IOAPIC IRQ */
380 entry->msi_attrib.pos = pos;
381 if (is_mask_bit_support(control)) {
382 entry->mask_base = (void __iomem *)(long)msi_mask_bits_reg(pos,
383 is_64bit_address(control));
386 if (entry->msi_attrib.maskbit) {
387 unsigned int maskbits, temp;
388 /* All MSIs are unmasked by default, Mask them all */
389 pci_read_config_dword(dev,
390 msi_mask_bits_reg(pos, is_64bit_address(control)),
392 temp = (1 << multi_msi_capable(control));
393 temp = ((temp - 1) & ~temp);
395 pci_write_config_dword(dev,
396 msi_mask_bits_reg(pos, is_64bit_address(control)),
398 entry->msi_attrib.maskbits_mask = temp;
400 list_add_tail(&entry->list, &dev->msi_list);
402 /* Configure MSI capability structure */
403 ret = arch_setup_msi_irqs(dev, 1, PCI_CAP_ID_MSI);
409 /* Set MSI enabled bits */
410 pci_intx_for_msi(dev, 0);
411 msi_set_enable(dev, 1);
412 dev->msi_enabled = 1;
414 dev->irq = entry->irq;
419 * msix_capability_init - configure device's MSI-X capability
420 * @dev: pointer to the pci_dev data structure of MSI-X device function
421 * @entries: pointer to an array of struct msix_entry entries
422 * @nvec: number of @entries
424 * Setup the MSI-X capability structure of device function with a
425 * single MSI-X irq. A return of zero indicates the successful setup of
426 * requested MSI-X entries with allocated irqs or non-zero for otherwise.
428 static int msix_capability_init(struct pci_dev *dev,
429 struct msix_entry *entries, int nvec)
431 struct msi_desc *entry;
432 int pos, i, j, nr_entries, ret;
433 unsigned long phys_addr;
439 msix_set_enable(dev, 0);/* Ensure msix is disabled as I set it up */
441 pos = pci_find_capability(dev, PCI_CAP_ID_MSIX);
442 /* Request & Map MSI-X table region */
443 pci_read_config_word(dev, msi_control_reg(pos), &control);
444 nr_entries = multi_msix_capable(control);
446 pci_read_config_dword(dev, msix_table_offset_reg(pos), &table_offset);
447 bir = (u8)(table_offset & PCI_MSIX_FLAGS_BIRMASK);
448 table_offset &= ~PCI_MSIX_FLAGS_BIRMASK;
449 phys_addr = pci_resource_start (dev, bir) + table_offset;
450 base = ioremap_nocache(phys_addr, nr_entries * PCI_MSIX_ENTRY_SIZE);
454 /* MSI-X Table Initialization */
455 for (i = 0; i < nvec; i++) {
456 entry = alloc_msi_entry();
460 j = entries[i].entry;
461 entry->msi_attrib.type = PCI_CAP_ID_MSIX;
462 entry->msi_attrib.is_64 = 1;
463 entry->msi_attrib.entry_nr = j;
464 entry->msi_attrib.maskbit = 1;
465 entry->msi_attrib.masked = 1;
466 entry->msi_attrib.default_irq = dev->irq;
467 entry->msi_attrib.pos = pos;
469 entry->mask_base = base;
471 list_add_tail(&entry->list, &dev->msi_list);
474 ret = arch_setup_msi_irqs(dev, nvec, PCI_CAP_ID_MSIX);
477 list_for_each_entry(entry, &dev->msi_list, list) {
478 if (entry->irq != 0) {
485 /* If we had some success report the number of irqs
486 * we succeeded in setting up.
494 list_for_each_entry(entry, &dev->msi_list, list) {
495 entries[i].vector = entry->irq;
496 set_irq_msi(entry->irq, entry);
499 /* Set MSI-X enabled bits */
500 pci_intx_for_msi(dev, 0);
501 msix_set_enable(dev, 1);
502 dev->msix_enabled = 1;
508 * pci_msi_check_device - check whether MSI may be enabled on a device
509 * @dev: pointer to the pci_dev data structure of MSI device function
510 * @nvec: how many MSIs have been requested ?
511 * @type: are we checking for MSI or MSI-X ?
513 * Look at global flags, the device itself, and its parent busses
514 * to determine if MSI/-X are supported for the device. If MSI/-X is
515 * supported return 0, else return an error code.
517 static int pci_msi_check_device(struct pci_dev* dev, int nvec, int type)
522 /* MSI must be globally enabled and supported by the device */
523 if (!pci_msi_enable || !dev || dev->no_msi)
527 * You can't ask to have 0 or less MSIs configured.
529 * b) the list manipulation code assumes nvec >= 1.
534 /* Any bridge which does NOT route MSI transactions from it's
535 * secondary bus to it's primary bus must set NO_MSI flag on
536 * the secondary pci_bus.
537 * We expect only arch-specific PCI host bus controller driver
538 * or quirks for specific PCI bridges to be setting NO_MSI.
540 for (bus = dev->bus; bus; bus = bus->parent)
541 if (bus->bus_flags & PCI_BUS_FLAGS_NO_MSI)
544 ret = arch_msi_check_device(dev, nvec, type);
548 if (!pci_find_capability(dev, type))
555 * pci_enable_msi - configure device's MSI capability structure
556 * @dev: pointer to the pci_dev data structure of MSI device function
558 * Setup the MSI capability structure of device function with
559 * a single MSI irq upon its software driver call to request for
560 * MSI mode enabled on its hardware device function. A return of zero
561 * indicates the successful setup of an entry zero with the new MSI
562 * irq or non-zero for otherwise.
564 int pci_enable_msi(struct pci_dev* dev)
568 status = pci_msi_check_device(dev, 1, PCI_CAP_ID_MSI);
572 WARN_ON(!!dev->msi_enabled);
574 /* Check whether driver already requested for MSI-X irqs */
575 if (dev->msix_enabled) {
576 dev_info(&dev->dev, "can't enable MSI "
577 "(MSI-X already enabled)\n");
580 status = msi_capability_init(dev);
583 EXPORT_SYMBOL(pci_enable_msi);
585 void pci_msi_shutdown(struct pci_dev* dev)
587 struct msi_desc *entry;
589 if (!pci_msi_enable || !dev || !dev->msi_enabled)
592 msi_set_enable(dev, 0);
593 pci_intx_for_msi(dev, 1);
594 dev->msi_enabled = 0;
596 BUG_ON(list_empty(&dev->msi_list));
597 entry = list_entry(dev->msi_list.next, struct msi_desc, list);
598 /* Return the the pci reset with msi irqs unmasked */
599 if (entry->msi_attrib.maskbit) {
600 u32 mask = entry->msi_attrib.maskbits_mask;
601 msi_set_mask_bits(dev->irq, mask, ~mask);
603 if (!entry->dev || entry->msi_attrib.type != PCI_CAP_ID_MSI)
606 /* Restore dev->irq to its default pin-assertion irq */
607 dev->irq = entry->msi_attrib.default_irq;
609 void pci_disable_msi(struct pci_dev* dev)
611 struct msi_desc *entry;
613 if (!pci_msi_enable || !dev || !dev->msi_enabled)
616 pci_msi_shutdown(dev);
618 entry = list_entry(dev->msi_list.next, struct msi_desc, list);
619 if (!entry->dev || entry->msi_attrib.type != PCI_CAP_ID_MSI)
624 EXPORT_SYMBOL(pci_disable_msi);
626 static int msi_free_irqs(struct pci_dev* dev)
628 struct msi_desc *entry, *tmp;
630 list_for_each_entry(entry, &dev->msi_list, list) {
632 BUG_ON(irq_has_action(entry->irq));
635 arch_teardown_msi_irqs(dev);
637 list_for_each_entry_safe(entry, tmp, &dev->msi_list, list) {
638 if (entry->msi_attrib.type == PCI_CAP_ID_MSIX) {
639 writel(1, entry->mask_base + entry->msi_attrib.entry_nr
640 * PCI_MSIX_ENTRY_SIZE
641 + PCI_MSIX_ENTRY_VECTOR_CTRL_OFFSET);
643 if (list_is_last(&entry->list, &dev->msi_list))
644 iounmap(entry->mask_base);
646 list_del(&entry->list);
654 * pci_enable_msix - configure device's MSI-X capability structure
655 * @dev: pointer to the pci_dev data structure of MSI-X device function
656 * @entries: pointer to an array of MSI-X entries
657 * @nvec: number of MSI-X irqs requested for allocation by device driver
659 * Setup the MSI-X capability structure of device function with the number
660 * of requested irqs upon its software driver call to request for
661 * MSI-X mode enabled on its hardware device function. A return of zero
662 * indicates the successful configuration of MSI-X capability structure
663 * with new allocated MSI-X irqs. A return of < 0 indicates a failure.
664 * Or a return of > 0 indicates that driver request is exceeding the number
665 * of irqs available. Driver should use the returned value to re-send
668 int pci_enable_msix(struct pci_dev* dev, struct msix_entry *entries, int nvec)
670 int status, pos, nr_entries;
677 status = pci_msi_check_device(dev, nvec, PCI_CAP_ID_MSIX);
681 pos = pci_find_capability(dev, PCI_CAP_ID_MSIX);
682 pci_read_config_word(dev, msi_control_reg(pos), &control);
683 nr_entries = multi_msix_capable(control);
684 if (nvec > nr_entries)
687 /* Check for any invalid entries */
688 for (i = 0; i < nvec; i++) {
689 if (entries[i].entry >= nr_entries)
690 return -EINVAL; /* invalid entry */
691 for (j = i + 1; j < nvec; j++) {
692 if (entries[i].entry == entries[j].entry)
693 return -EINVAL; /* duplicate entry */
696 WARN_ON(!!dev->msix_enabled);
698 /* Check whether driver already requested for MSI irq */
699 if (dev->msi_enabled) {
700 dev_info(&dev->dev, "can't enable MSI-X "
701 "(MSI IRQ already assigned)\n");
704 status = msix_capability_init(dev, entries, nvec);
707 EXPORT_SYMBOL(pci_enable_msix);
709 static void msix_free_all_irqs(struct pci_dev *dev)
714 void pci_msix_shutdown(struct pci_dev* dev)
716 if (!pci_msi_enable || !dev || !dev->msix_enabled)
719 msix_set_enable(dev, 0);
720 pci_intx_for_msi(dev, 1);
721 dev->msix_enabled = 0;
723 void pci_disable_msix(struct pci_dev* dev)
725 if (!pci_msi_enable || !dev || !dev->msix_enabled)
728 pci_msix_shutdown(dev);
730 msix_free_all_irqs(dev);
732 EXPORT_SYMBOL(pci_disable_msix);
735 * msi_remove_pci_irq_vectors - reclaim MSI(X) irqs to unused state
736 * @dev: pointer to the pci_dev data structure of MSI(X) device function
738 * Being called during hotplug remove, from which the device function
739 * is hot-removed. All previous assigned MSI/MSI-X irqs, if
740 * allocated for this device function, are reclaimed to unused state,
741 * which may be used later on.
743 void msi_remove_pci_irq_vectors(struct pci_dev* dev)
745 if (!pci_msi_enable || !dev)
748 if (dev->msi_enabled)
751 if (dev->msix_enabled)
752 msix_free_all_irqs(dev);
755 void pci_no_msi(void)
760 void pci_msi_init_pci_dev(struct pci_dev *dev)
762 INIT_LIST_HEAD(&dev->msi_list);