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;
401 if (entry->msi_attrib.maskbit) {
402 entry->mask_base = (void __iomem *)(long)msi_mask_bits_reg(pos,
403 entry->msi_attrib.is_64);
406 if (entry->msi_attrib.maskbit) {
407 unsigned int maskbits, temp;
408 /* All MSIs are unmasked by default, Mask them all */
409 pci_read_config_dword(dev,
410 msi_mask_bits_reg(pos, entry->msi_attrib.is_64),
412 temp = (1 << multi_msi_capable(control));
413 temp = ((temp - 1) & ~temp);
415 pci_write_config_dword(dev, entry->msi_attrib.is_64, maskbits);
416 entry->msi_attrib.maskbits_mask = temp;
418 list_add_tail(&entry->list, &dev->msi_list);
420 /* Configure MSI capability structure */
421 ret = arch_setup_msi_irqs(dev, 1, PCI_CAP_ID_MSI);
427 /* Set MSI enabled bits */
428 pci_intx_for_msi(dev, 0);
429 msi_set_enable(dev, 1);
430 dev->msi_enabled = 1;
432 dev->irq = entry->irq;
437 * msix_capability_init - configure device's MSI-X capability
438 * @dev: pointer to the pci_dev data structure of MSI-X device function
439 * @entries: pointer to an array of struct msix_entry entries
440 * @nvec: number of @entries
442 * Setup the MSI-X capability structure of device function with a
443 * single MSI-X irq. A return of zero indicates the successful setup of
444 * requested MSI-X entries with allocated irqs or non-zero for otherwise.
446 static int msix_capability_init(struct pci_dev *dev,
447 struct msix_entry *entries, int nvec)
449 struct msi_desc *entry;
450 int pos, i, j, nr_entries, ret;
451 unsigned long phys_addr;
457 msix_set_enable(dev, 0);/* Ensure msix is disabled as I set it up */
459 pos = pci_find_capability(dev, PCI_CAP_ID_MSIX);
460 /* Request & Map MSI-X table region */
461 pci_read_config_word(dev, msi_control_reg(pos), &control);
462 nr_entries = multi_msix_capable(control);
464 pci_read_config_dword(dev, msix_table_offset_reg(pos), &table_offset);
465 bir = (u8)(table_offset & PCI_MSIX_FLAGS_BIRMASK);
466 table_offset &= ~PCI_MSIX_FLAGS_BIRMASK;
467 phys_addr = pci_resource_start (dev, bir) + table_offset;
468 base = ioremap_nocache(phys_addr, nr_entries * PCI_MSIX_ENTRY_SIZE);
472 /* MSI-X Table Initialization */
473 for (i = 0; i < nvec; i++) {
474 entry = alloc_msi_entry();
478 j = entries[i].entry;
479 entry->msi_attrib.type = PCI_CAP_ID_MSIX;
480 entry->msi_attrib.is_64 = 1;
481 entry->msi_attrib.entry_nr = j;
482 entry->msi_attrib.maskbit = 1;
483 entry->msi_attrib.masked = 1;
484 entry->msi_attrib.default_irq = dev->irq;
485 entry->msi_attrib.pos = pos;
487 entry->mask_base = base;
489 list_add_tail(&entry->list, &dev->msi_list);
492 ret = arch_setup_msi_irqs(dev, nvec, PCI_CAP_ID_MSIX);
495 list_for_each_entry(entry, &dev->msi_list, list) {
496 if (entry->irq != 0) {
503 /* If we had some success report the number of irqs
504 * we succeeded in setting up.
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_enable_msix - configure device's MSI-X capability structure
674 * @dev: pointer to the pci_dev data structure of MSI-X device function
675 * @entries: pointer to an array of MSI-X entries
676 * @nvec: number of MSI-X irqs requested for allocation by device driver
678 * Setup the MSI-X capability structure of device function with the number
679 * of requested irqs upon its software driver call to request for
680 * MSI-X mode enabled on its hardware device function. A return of zero
681 * indicates the successful configuration of MSI-X capability structure
682 * with new allocated MSI-X irqs. A return of < 0 indicates a failure.
683 * Or a return of > 0 indicates that driver request is exceeding the number
684 * of irqs available. Driver should use the returned value to re-send
687 int pci_enable_msix(struct pci_dev* dev, struct msix_entry *entries, int nvec)
689 int status, pos, nr_entries;
696 status = pci_msi_check_device(dev, nvec, PCI_CAP_ID_MSIX);
700 pos = pci_find_capability(dev, PCI_CAP_ID_MSIX);
701 pci_read_config_word(dev, msi_control_reg(pos), &control);
702 nr_entries = multi_msix_capable(control);
703 if (nvec > nr_entries)
706 /* Check for any invalid entries */
707 for (i = 0; i < nvec; i++) {
708 if (entries[i].entry >= nr_entries)
709 return -EINVAL; /* invalid entry */
710 for (j = i + 1; j < nvec; j++) {
711 if (entries[i].entry == entries[j].entry)
712 return -EINVAL; /* duplicate entry */
715 WARN_ON(!!dev->msix_enabled);
717 /* Check whether driver already requested for MSI irq */
718 if (dev->msi_enabled) {
719 dev_info(&dev->dev, "can't enable MSI-X "
720 "(MSI IRQ already assigned)\n");
723 status = msix_capability_init(dev, entries, nvec);
726 EXPORT_SYMBOL(pci_enable_msix);
728 static void msix_free_all_irqs(struct pci_dev *dev)
733 void pci_msix_shutdown(struct pci_dev* dev)
735 if (!pci_msi_enable || !dev || !dev->msix_enabled)
738 msix_set_enable(dev, 0);
739 pci_intx_for_msi(dev, 1);
740 dev->msix_enabled = 0;
742 void pci_disable_msix(struct pci_dev* dev)
744 if (!pci_msi_enable || !dev || !dev->msix_enabled)
747 pci_msix_shutdown(dev);
749 msix_free_all_irqs(dev);
751 EXPORT_SYMBOL(pci_disable_msix);
754 * msi_remove_pci_irq_vectors - reclaim MSI(X) irqs to unused state
755 * @dev: pointer to the pci_dev data structure of MSI(X) device function
757 * Being called during hotplug remove, from which the device function
758 * is hot-removed. All previous assigned MSI/MSI-X irqs, if
759 * allocated for this device function, are reclaimed to unused state,
760 * which may be used later on.
762 void msi_remove_pci_irq_vectors(struct pci_dev* dev)
764 if (!pci_msi_enable || !dev)
767 if (dev->msi_enabled)
770 if (dev->msix_enabled)
771 msix_free_all_irqs(dev);
774 void pci_no_msi(void)
780 * pci_msi_enabled - is MSI enabled?
782 * Returns true if MSI has not been disabled by the command-line option
785 int pci_msi_enabled(void)
787 return pci_msi_enable;
789 EXPORT_SYMBOL(pci_msi_enabled);
791 void pci_msi_init_pci_dev(struct pci_dev *dev)
793 INIT_LIST_HEAD(&dev->msi_list);