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(struct irq_desc *desc)
108 struct msi_desc *entry;
110 entry = get_irq_desc_msi(desc);
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(struct irq_desc *desc, u32 mask, u32 flag)
140 struct msi_desc *entry;
142 entry = get_irq_desc_msi(desc);
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_desc(struct irq_desc *desc, struct msi_msg *msg)
177 struct msi_desc *entry = get_irq_desc_msi(desc);
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 read_msi_msg(unsigned int irq, struct msi_msg *msg)
216 struct irq_desc *desc = irq_to_desc(irq);
218 read_msi_msg_desc(desc, msg);
221 void write_msi_msg_desc(struct irq_desc *desc, struct msi_msg *msg)
223 struct msi_desc *entry = get_irq_desc_msi(desc);
224 switch (entry->msi_attrib.type) {
227 struct pci_dev *dev = entry->dev;
228 int pos = entry->msi_attrib.pos;
230 pci_write_config_dword(dev, msi_lower_address_reg(pos),
232 if (entry->msi_attrib.is_64) {
233 pci_write_config_dword(dev, msi_upper_address_reg(pos),
235 pci_write_config_word(dev, msi_data_reg(pos, 1),
238 pci_write_config_word(dev, msi_data_reg(pos, 0),
243 case PCI_CAP_ID_MSIX:
246 base = entry->mask_base +
247 entry->msi_attrib.entry_nr * PCI_MSIX_ENTRY_SIZE;
249 writel(msg->address_lo,
250 base + PCI_MSIX_ENTRY_LOWER_ADDR_OFFSET);
251 writel(msg->address_hi,
252 base + PCI_MSIX_ENTRY_UPPER_ADDR_OFFSET);
253 writel(msg->data, base + PCI_MSIX_ENTRY_DATA_OFFSET);
262 void write_msi_msg(unsigned int irq, struct msi_msg *msg)
264 struct irq_desc *desc = irq_to_desc(irq);
266 write_msi_msg_desc(desc, msg);
269 void mask_msi_irq(unsigned int irq)
271 struct irq_desc *desc = irq_to_desc(irq);
273 msi_set_mask_bits(desc, 1, 1);
274 msix_flush_writes(desc);
277 void unmask_msi_irq(unsigned int irq)
279 struct irq_desc *desc = irq_to_desc(irq);
281 msi_set_mask_bits(desc, 1, 0);
282 msix_flush_writes(desc);
285 static int msi_free_irqs(struct pci_dev* dev);
287 static struct msi_desc* alloc_msi_entry(void)
289 struct msi_desc *entry;
291 entry = kzalloc(sizeof(struct msi_desc), GFP_KERNEL);
295 INIT_LIST_HEAD(&entry->list);
302 static void pci_intx_for_msi(struct pci_dev *dev, int enable)
304 if (!(dev->dev_flags & PCI_DEV_FLAGS_MSI_INTX_DISABLE_BUG))
305 pci_intx(dev, enable);
308 static void __pci_restore_msi_state(struct pci_dev *dev)
312 struct msi_desc *entry;
314 if (!dev->msi_enabled)
317 entry = get_irq_msi(dev->irq);
318 pos = entry->msi_attrib.pos;
320 pci_intx_for_msi(dev, 0);
321 msi_set_enable(dev, 0);
322 write_msi_msg(dev->irq, &entry->msg);
323 if (entry->msi_attrib.maskbit) {
324 struct irq_desc *desc = irq_to_desc(dev->irq);
325 msi_set_mask_bits(desc, entry->msi_attrib.maskbits_mask,
326 entry->msi_attrib.masked);
329 pci_read_config_word(dev, pos + PCI_MSI_FLAGS, &control);
330 control &= ~PCI_MSI_FLAGS_QSIZE;
331 control |= PCI_MSI_FLAGS_ENABLE;
332 pci_write_config_word(dev, pos + PCI_MSI_FLAGS, control);
335 static void __pci_restore_msix_state(struct pci_dev *dev)
338 struct msi_desc *entry;
341 if (!dev->msix_enabled)
344 /* route the table */
345 pci_intx_for_msi(dev, 0);
346 msix_set_enable(dev, 0);
348 list_for_each_entry(entry, &dev->msi_list, list) {
349 struct irq_desc *desc = irq_to_desc(entry->irq);
350 write_msi_msg(entry->irq, &entry->msg);
351 msi_set_mask_bits(desc, 1, entry->msi_attrib.masked);
354 BUG_ON(list_empty(&dev->msi_list));
355 entry = list_entry(dev->msi_list.next, struct msi_desc, list);
356 pos = entry->msi_attrib.pos;
357 pci_read_config_word(dev, pos + PCI_MSIX_FLAGS, &control);
358 control &= ~PCI_MSIX_FLAGS_MASKALL;
359 control |= PCI_MSIX_FLAGS_ENABLE;
360 pci_write_config_word(dev, pos + PCI_MSIX_FLAGS, control);
363 void pci_restore_msi_state(struct pci_dev *dev)
365 __pci_restore_msi_state(dev);
366 __pci_restore_msix_state(dev);
368 EXPORT_SYMBOL_GPL(pci_restore_msi_state);
371 * msi_capability_init - configure device's MSI capability structure
372 * @dev: pointer to the pci_dev data structure of MSI device function
374 * Setup the MSI capability structure of device function with a single
375 * MSI irq, regardless of device function is capable of handling
376 * multiple messages. A return of zero indicates the successful setup
377 * of an entry zero with the new MSI irq or non-zero for otherwise.
379 static int msi_capability_init(struct pci_dev *dev)
381 struct msi_desc *entry;
385 msi_set_enable(dev, 0); /* Ensure msi is disabled as I set it up */
387 pos = pci_find_capability(dev, PCI_CAP_ID_MSI);
388 pci_read_config_word(dev, msi_control_reg(pos), &control);
389 /* MSI Entry Initialization */
390 entry = alloc_msi_entry();
394 entry->msi_attrib.type = PCI_CAP_ID_MSI;
395 entry->msi_attrib.is_64 = is_64bit_address(control);
396 entry->msi_attrib.entry_nr = 0;
397 entry->msi_attrib.maskbit = is_mask_bit_support(control);
398 entry->msi_attrib.masked = 1;
399 entry->msi_attrib.default_irq = dev->irq; /* Save IOAPIC IRQ */
400 entry->msi_attrib.pos = pos;
402 if (entry->msi_attrib.maskbit) {
403 unsigned int base, maskbits, temp;
405 base = msi_mask_bits_reg(pos, entry->msi_attrib.is_64);
406 entry->mask_base = (void __iomem *)(long)base;
408 /* All MSIs are unmasked by default, Mask them all */
409 pci_read_config_dword(dev, base, &maskbits);
410 temp = (1 << multi_msi_capable(control));
411 temp = ((temp - 1) & ~temp);
413 pci_write_config_dword(dev, base, maskbits);
414 entry->msi_attrib.maskbits_mask = temp;
416 list_add_tail(&entry->list, &dev->msi_list);
418 /* Configure MSI capability structure */
419 ret = arch_setup_msi_irqs(dev, 1, PCI_CAP_ID_MSI);
425 /* Set MSI enabled bits */
426 pci_intx_for_msi(dev, 0);
427 msi_set_enable(dev, 1);
428 dev->msi_enabled = 1;
430 dev->irq = entry->irq;
435 * msix_capability_init - configure device's MSI-X capability
436 * @dev: pointer to the pci_dev data structure of MSI-X device function
437 * @entries: pointer to an array of struct msix_entry entries
438 * @nvec: number of @entries
440 * Setup the MSI-X capability structure of device function with a
441 * single MSI-X irq. A return of zero indicates the successful setup of
442 * requested MSI-X entries with allocated irqs or non-zero for otherwise.
444 static int msix_capability_init(struct pci_dev *dev,
445 struct msix_entry *entries, int nvec)
447 struct msi_desc *entry;
448 int pos, i, j, nr_entries, ret;
449 unsigned long phys_addr;
455 msix_set_enable(dev, 0);/* Ensure msix is disabled as I set it up */
457 pos = pci_find_capability(dev, PCI_CAP_ID_MSIX);
458 /* Request & Map MSI-X table region */
459 pci_read_config_word(dev, msi_control_reg(pos), &control);
460 nr_entries = multi_msix_capable(control);
462 pci_read_config_dword(dev, msix_table_offset_reg(pos), &table_offset);
463 bir = (u8)(table_offset & PCI_MSIX_FLAGS_BIRMASK);
464 table_offset &= ~PCI_MSIX_FLAGS_BIRMASK;
465 phys_addr = pci_resource_start (dev, bir) + table_offset;
466 base = ioremap_nocache(phys_addr, nr_entries * PCI_MSIX_ENTRY_SIZE);
470 /* MSI-X Table Initialization */
471 for (i = 0; i < nvec; i++) {
472 entry = alloc_msi_entry();
476 j = entries[i].entry;
477 entry->msi_attrib.type = PCI_CAP_ID_MSIX;
478 entry->msi_attrib.is_64 = 1;
479 entry->msi_attrib.entry_nr = j;
480 entry->msi_attrib.maskbit = 1;
481 entry->msi_attrib.masked = 1;
482 entry->msi_attrib.default_irq = dev->irq;
483 entry->msi_attrib.pos = pos;
485 entry->mask_base = base;
487 list_add_tail(&entry->list, &dev->msi_list);
490 ret = arch_setup_msi_irqs(dev, nvec, PCI_CAP_ID_MSIX);
493 list_for_each_entry(entry, &dev->msi_list, list) {
494 if (entry->irq != 0) {
501 /* If we had some success report the number of irqs
502 * we succeeded in setting up.
510 list_for_each_entry(entry, &dev->msi_list, list) {
511 entries[i].vector = entry->irq;
512 set_irq_msi(entry->irq, entry);
515 /* Set MSI-X enabled bits */
516 pci_intx_for_msi(dev, 0);
517 msix_set_enable(dev, 1);
518 dev->msix_enabled = 1;
524 * pci_msi_check_device - check whether MSI may be enabled on a device
525 * @dev: pointer to the pci_dev data structure of MSI device function
526 * @nvec: how many MSIs have been requested ?
527 * @type: are we checking for MSI or MSI-X ?
529 * Look at global flags, the device itself, and its parent busses
530 * to determine if MSI/-X are supported for the device. If MSI/-X is
531 * supported return 0, else return an error code.
533 static int pci_msi_check_device(struct pci_dev* dev, int nvec, int type)
538 /* MSI must be globally enabled and supported by the device */
539 if (!pci_msi_enable || !dev || dev->no_msi)
543 * You can't ask to have 0 or less MSIs configured.
545 * b) the list manipulation code assumes nvec >= 1.
550 /* Any bridge which does NOT route MSI transactions from it's
551 * secondary bus to it's primary bus must set NO_MSI flag on
552 * the secondary pci_bus.
553 * We expect only arch-specific PCI host bus controller driver
554 * or quirks for specific PCI bridges to be setting NO_MSI.
556 for (bus = dev->bus; bus; bus = bus->parent)
557 if (bus->bus_flags & PCI_BUS_FLAGS_NO_MSI)
560 ret = arch_msi_check_device(dev, nvec, type);
564 if (!pci_find_capability(dev, type))
571 * pci_enable_msi - configure device's MSI capability structure
572 * @dev: pointer to the pci_dev data structure of MSI device function
574 * Setup the MSI capability structure of device function with
575 * a single MSI irq upon its software driver call to request for
576 * MSI mode enabled on its hardware device function. A return of zero
577 * indicates the successful setup of an entry zero with the new MSI
578 * irq or non-zero for otherwise.
580 int pci_enable_msi(struct pci_dev* dev)
584 status = pci_msi_check_device(dev, 1, PCI_CAP_ID_MSI);
588 WARN_ON(!!dev->msi_enabled);
590 /* Check whether driver already requested for MSI-X irqs */
591 if (dev->msix_enabled) {
592 dev_info(&dev->dev, "can't enable MSI "
593 "(MSI-X already enabled)\n");
596 status = msi_capability_init(dev);
599 EXPORT_SYMBOL(pci_enable_msi);
601 void pci_msi_shutdown(struct pci_dev* dev)
603 struct msi_desc *entry;
605 if (!pci_msi_enable || !dev || !dev->msi_enabled)
608 msi_set_enable(dev, 0);
609 pci_intx_for_msi(dev, 1);
610 dev->msi_enabled = 0;
612 BUG_ON(list_empty(&dev->msi_list));
613 entry = list_entry(dev->msi_list.next, struct msi_desc, list);
614 /* Return the the pci reset with msi irqs unmasked */
615 if (entry->msi_attrib.maskbit) {
616 u32 mask = entry->msi_attrib.maskbits_mask;
617 struct irq_desc *desc = irq_to_desc(dev->irq);
618 msi_set_mask_bits(desc, mask, ~mask);
620 if (!entry->dev || entry->msi_attrib.type != PCI_CAP_ID_MSI)
623 /* Restore dev->irq to its default pin-assertion irq */
624 dev->irq = entry->msi_attrib.default_irq;
626 void pci_disable_msi(struct pci_dev* dev)
628 struct msi_desc *entry;
630 if (!pci_msi_enable || !dev || !dev->msi_enabled)
633 pci_msi_shutdown(dev);
635 entry = list_entry(dev->msi_list.next, struct msi_desc, list);
636 if (!entry->dev || entry->msi_attrib.type != PCI_CAP_ID_MSI)
641 EXPORT_SYMBOL(pci_disable_msi);
643 static int msi_free_irqs(struct pci_dev* dev)
645 struct msi_desc *entry, *tmp;
647 list_for_each_entry(entry, &dev->msi_list, list) {
649 BUG_ON(irq_has_action(entry->irq));
652 arch_teardown_msi_irqs(dev);
654 list_for_each_entry_safe(entry, tmp, &dev->msi_list, list) {
655 if (entry->msi_attrib.type == PCI_CAP_ID_MSIX) {
656 writel(1, entry->mask_base + entry->msi_attrib.entry_nr
657 * PCI_MSIX_ENTRY_SIZE
658 + PCI_MSIX_ENTRY_VECTOR_CTRL_OFFSET);
660 if (list_is_last(&entry->list, &dev->msi_list))
661 iounmap(entry->mask_base);
663 list_del(&entry->list);
671 * pci_enable_msix - configure device's MSI-X capability structure
672 * @dev: pointer to the pci_dev data structure of MSI-X device function
673 * @entries: pointer to an array of MSI-X entries
674 * @nvec: number of MSI-X irqs requested for allocation by device driver
676 * Setup the MSI-X capability structure of device function with the number
677 * of requested irqs upon its software driver call to request for
678 * MSI-X mode enabled on its hardware device function. A return of zero
679 * indicates the successful configuration of MSI-X capability structure
680 * with new allocated MSI-X irqs. A return of < 0 indicates a failure.
681 * Or a return of > 0 indicates that driver request is exceeding the number
682 * of irqs available. Driver should use the returned value to re-send
685 int pci_enable_msix(struct pci_dev* dev, struct msix_entry *entries, int nvec)
687 int status, pos, nr_entries;
694 status = pci_msi_check_device(dev, nvec, PCI_CAP_ID_MSIX);
698 pos = pci_find_capability(dev, PCI_CAP_ID_MSIX);
699 pci_read_config_word(dev, msi_control_reg(pos), &control);
700 nr_entries = multi_msix_capable(control);
701 if (nvec > nr_entries)
704 /* Check for any invalid entries */
705 for (i = 0; i < nvec; i++) {
706 if (entries[i].entry >= nr_entries)
707 return -EINVAL; /* invalid entry */
708 for (j = i + 1; j < nvec; j++) {
709 if (entries[i].entry == entries[j].entry)
710 return -EINVAL; /* duplicate entry */
713 WARN_ON(!!dev->msix_enabled);
715 /* Check whether driver already requested for MSI irq */
716 if (dev->msi_enabled) {
717 dev_info(&dev->dev, "can't enable MSI-X "
718 "(MSI IRQ already assigned)\n");
721 status = msix_capability_init(dev, entries, nvec);
724 EXPORT_SYMBOL(pci_enable_msix);
726 static void msix_free_all_irqs(struct pci_dev *dev)
731 void pci_msix_shutdown(struct pci_dev* dev)
733 if (!pci_msi_enable || !dev || !dev->msix_enabled)
736 msix_set_enable(dev, 0);
737 pci_intx_for_msi(dev, 1);
738 dev->msix_enabled = 0;
740 void pci_disable_msix(struct pci_dev* dev)
742 if (!pci_msi_enable || !dev || !dev->msix_enabled)
745 pci_msix_shutdown(dev);
747 msix_free_all_irqs(dev);
749 EXPORT_SYMBOL(pci_disable_msix);
752 * msi_remove_pci_irq_vectors - reclaim MSI(X) irqs to unused state
753 * @dev: pointer to the pci_dev data structure of MSI(X) device function
755 * Being called during hotplug remove, from which the device function
756 * is hot-removed. All previous assigned MSI/MSI-X irqs, if
757 * allocated for this device function, are reclaimed to unused state,
758 * which may be used later on.
760 void msi_remove_pci_irq_vectors(struct pci_dev* dev)
762 if (!pci_msi_enable || !dev)
765 if (dev->msi_enabled)
768 if (dev->msix_enabled)
769 msix_free_all_irqs(dev);
772 void pci_no_msi(void)
778 * pci_msi_enabled - is MSI enabled?
780 * Returns true if MSI has not been disabled by the command-line option
783 int pci_msi_enabled(void)
785 return pci_msi_enable;
787 EXPORT_SYMBOL(pci_msi_enabled);
789 void pci_msi_init_pci_dev(struct pci_dev *dev)
791 INIT_LIST_HEAD(&dev->msi_list);