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 enable)
78 pos = pci_find_capability(dev, PCI_CAP_ID_MSI);
80 pci_read_config_word(dev, pos + PCI_MSI_FLAGS, &control);
81 control &= ~PCI_MSI_FLAGS_ENABLE;
83 control |= PCI_MSI_FLAGS_ENABLE;
84 pci_write_config_word(dev, pos + PCI_MSI_FLAGS, control);
88 static void msix_set_enable(struct pci_dev *dev, int enable)
93 pos = pci_find_capability(dev, PCI_CAP_ID_MSIX);
95 pci_read_config_word(dev, pos + PCI_MSIX_FLAGS, &control);
96 control &= ~PCI_MSIX_FLAGS_ENABLE;
98 control |= PCI_MSIX_FLAGS_ENABLE;
99 pci_write_config_word(dev, pos + PCI_MSIX_FLAGS, control);
103 static void msix_flush_writes(unsigned int irq)
105 struct msi_desc *entry;
107 entry = get_irq_msi(irq);
108 BUG_ON(!entry || !entry->dev);
109 switch (entry->msi_attrib.type) {
113 case PCI_CAP_ID_MSIX:
115 int offset = entry->msi_attrib.entry_nr * PCI_MSIX_ENTRY_SIZE +
116 PCI_MSIX_ENTRY_VECTOR_CTRL_OFFSET;
117 readl(entry->mask_base + offset);
126 static void msi_set_mask_bits(unsigned int irq, u32 mask, u32 flag)
128 struct msi_desc *entry;
130 entry = get_irq_msi(irq);
131 BUG_ON(!entry || !entry->dev);
132 switch (entry->msi_attrib.type) {
134 if (entry->msi_attrib.maskbit) {
138 pos = (long)entry->mask_base;
139 pci_read_config_dword(entry->dev, pos, &mask_bits);
140 mask_bits &= ~(mask);
141 mask_bits |= flag & mask;
142 pci_write_config_dword(entry->dev, pos, mask_bits);
144 msi_set_enable(entry->dev, !flag);
147 case PCI_CAP_ID_MSIX:
149 int offset = entry->msi_attrib.entry_nr * PCI_MSIX_ENTRY_SIZE +
150 PCI_MSIX_ENTRY_VECTOR_CTRL_OFFSET;
151 writel(flag, entry->mask_base + offset);
152 readl(entry->mask_base + offset);
159 entry->msi_attrib.masked = !!flag;
162 void read_msi_msg(unsigned int irq, struct msi_msg *msg)
164 struct msi_desc *entry = get_irq_msi(irq);
165 switch(entry->msi_attrib.type) {
168 struct pci_dev *dev = entry->dev;
169 int pos = entry->msi_attrib.pos;
172 pci_read_config_dword(dev, msi_lower_address_reg(pos),
174 if (entry->msi_attrib.is_64) {
175 pci_read_config_dword(dev, msi_upper_address_reg(pos),
177 pci_read_config_word(dev, msi_data_reg(pos, 1), &data);
180 pci_read_config_word(dev, msi_data_reg(pos, 0), &data);
185 case PCI_CAP_ID_MSIX:
188 base = entry->mask_base +
189 entry->msi_attrib.entry_nr * PCI_MSIX_ENTRY_SIZE;
191 msg->address_lo = readl(base + PCI_MSIX_ENTRY_LOWER_ADDR_OFFSET);
192 msg->address_hi = readl(base + PCI_MSIX_ENTRY_UPPER_ADDR_OFFSET);
193 msg->data = readl(base + PCI_MSIX_ENTRY_DATA_OFFSET);
201 void write_msi_msg(unsigned int irq, struct msi_msg *msg)
203 struct msi_desc *entry = get_irq_msi(irq);
204 switch (entry->msi_attrib.type) {
207 struct pci_dev *dev = entry->dev;
208 int pos = entry->msi_attrib.pos;
210 pci_write_config_dword(dev, msi_lower_address_reg(pos),
212 if (entry->msi_attrib.is_64) {
213 pci_write_config_dword(dev, msi_upper_address_reg(pos),
215 pci_write_config_word(dev, msi_data_reg(pos, 1),
218 pci_write_config_word(dev, msi_data_reg(pos, 0),
223 case PCI_CAP_ID_MSIX:
226 base = entry->mask_base +
227 entry->msi_attrib.entry_nr * PCI_MSIX_ENTRY_SIZE;
229 writel(msg->address_lo,
230 base + PCI_MSIX_ENTRY_LOWER_ADDR_OFFSET);
231 writel(msg->address_hi,
232 base + PCI_MSIX_ENTRY_UPPER_ADDR_OFFSET);
233 writel(msg->data, base + PCI_MSIX_ENTRY_DATA_OFFSET);
242 void mask_msi_irq(unsigned int irq)
244 msi_set_mask_bits(irq, 1, 1);
245 msix_flush_writes(irq);
248 void unmask_msi_irq(unsigned int irq)
250 msi_set_mask_bits(irq, 1, 0);
251 msix_flush_writes(irq);
254 static int msi_free_irqs(struct pci_dev* dev);
257 static struct msi_desc* alloc_msi_entry(void)
259 struct msi_desc *entry;
261 entry = kzalloc(sizeof(struct msi_desc), GFP_KERNEL);
265 INIT_LIST_HEAD(&entry->list);
272 static void pci_intx_for_msi(struct pci_dev *dev, int enable)
274 if (!(dev->dev_flags & PCI_DEV_FLAGS_MSI_INTX_DISABLE_BUG))
275 pci_intx(dev, enable);
278 static void __pci_restore_msi_state(struct pci_dev *dev)
282 struct msi_desc *entry;
284 if (!dev->msi_enabled)
287 entry = get_irq_msi(dev->irq);
288 pos = entry->msi_attrib.pos;
290 pci_intx_for_msi(dev, 0);
291 msi_set_enable(dev, 0);
292 write_msi_msg(dev->irq, &entry->msg);
293 if (entry->msi_attrib.maskbit)
294 msi_set_mask_bits(dev->irq, entry->msi_attrib.maskbits_mask,
295 entry->msi_attrib.masked);
297 pci_read_config_word(dev, pos + PCI_MSI_FLAGS, &control);
298 control &= ~(PCI_MSI_FLAGS_QSIZE | PCI_MSI_FLAGS_ENABLE);
299 if (entry->msi_attrib.maskbit || !entry->msi_attrib.masked)
300 control |= PCI_MSI_FLAGS_ENABLE;
301 pci_write_config_word(dev, pos + PCI_MSI_FLAGS, control);
304 static void __pci_restore_msix_state(struct pci_dev *dev)
307 struct msi_desc *entry;
310 if (!dev->msix_enabled)
313 /* route the table */
314 pci_intx_for_msi(dev, 0);
315 msix_set_enable(dev, 0);
317 list_for_each_entry(entry, &dev->msi_list, list) {
318 write_msi_msg(entry->irq, &entry->msg);
319 msi_set_mask_bits(entry->irq, 1, entry->msi_attrib.masked);
322 BUG_ON(list_empty(&dev->msi_list));
323 entry = list_entry(dev->msi_list.next, struct msi_desc, list);
324 pos = entry->msi_attrib.pos;
325 pci_read_config_word(dev, pos + PCI_MSIX_FLAGS, &control);
326 control &= ~PCI_MSIX_FLAGS_MASKALL;
327 control |= PCI_MSIX_FLAGS_ENABLE;
328 pci_write_config_word(dev, pos + PCI_MSIX_FLAGS, control);
331 void pci_restore_msi_state(struct pci_dev *dev)
333 __pci_restore_msi_state(dev);
334 __pci_restore_msix_state(dev);
336 EXPORT_SYMBOL_GPL(pci_restore_msi_state);
339 * msi_capability_init - configure device's MSI capability structure
340 * @dev: pointer to the pci_dev data structure of MSI device function
342 * Setup the MSI capability structure of device function with a single
343 * MSI irq, regardless of device function is capable of handling
344 * multiple messages. A return of zero indicates the successful setup
345 * of an entry zero with the new MSI irq or non-zero for otherwise.
347 static int msi_capability_init(struct pci_dev *dev)
349 struct msi_desc *entry;
353 msi_set_enable(dev, 0); /* Ensure msi is disabled as I set it up */
355 pos = pci_find_capability(dev, PCI_CAP_ID_MSI);
356 pci_read_config_word(dev, msi_control_reg(pos), &control);
357 /* MSI Entry Initialization */
358 entry = alloc_msi_entry();
362 entry->msi_attrib.type = PCI_CAP_ID_MSI;
363 entry->msi_attrib.is_64 = is_64bit_address(control);
364 entry->msi_attrib.entry_nr = 0;
365 entry->msi_attrib.maskbit = is_mask_bit_support(control);
366 entry->msi_attrib.masked = 1;
367 entry->msi_attrib.default_irq = dev->irq; /* Save IOAPIC IRQ */
368 entry->msi_attrib.pos = pos;
369 if (is_mask_bit_support(control)) {
370 entry->mask_base = (void __iomem *)(long)msi_mask_bits_reg(pos,
371 is_64bit_address(control));
374 if (entry->msi_attrib.maskbit) {
375 unsigned int maskbits, temp;
376 /* All MSIs are unmasked by default, Mask them all */
377 pci_read_config_dword(dev,
378 msi_mask_bits_reg(pos, is_64bit_address(control)),
380 temp = (1 << multi_msi_capable(control));
381 temp = ((temp - 1) & ~temp);
383 pci_write_config_dword(dev,
384 msi_mask_bits_reg(pos, is_64bit_address(control)),
386 entry->msi_attrib.maskbits_mask = temp;
388 list_add_tail(&entry->list, &dev->msi_list);
390 /* Configure MSI capability structure */
391 ret = arch_setup_msi_irqs(dev, 1, PCI_CAP_ID_MSI);
397 /* Set MSI enabled bits */
398 pci_intx_for_msi(dev, 0);
399 msi_set_enable(dev, 1);
400 dev->msi_enabled = 1;
402 dev->irq = entry->irq;
407 * msix_capability_init - configure device's MSI-X capability
408 * @dev: pointer to the pci_dev data structure of MSI-X device function
409 * @entries: pointer to an array of struct msix_entry entries
410 * @nvec: number of @entries
412 * Setup the MSI-X capability structure of device function with a
413 * single MSI-X irq. A return of zero indicates the successful setup of
414 * requested MSI-X entries with allocated irqs or non-zero for otherwise.
416 static int msix_capability_init(struct pci_dev *dev,
417 struct msix_entry *entries, int nvec)
419 struct msi_desc *entry;
420 int pos, i, j, nr_entries, ret;
421 unsigned long phys_addr;
427 msix_set_enable(dev, 0);/* Ensure msix is disabled as I set it up */
429 pos = pci_find_capability(dev, PCI_CAP_ID_MSIX);
430 /* Request & Map MSI-X table region */
431 pci_read_config_word(dev, msi_control_reg(pos), &control);
432 nr_entries = multi_msix_capable(control);
434 pci_read_config_dword(dev, msix_table_offset_reg(pos), &table_offset);
435 bir = (u8)(table_offset & PCI_MSIX_FLAGS_BIRMASK);
436 table_offset &= ~PCI_MSIX_FLAGS_BIRMASK;
437 phys_addr = pci_resource_start (dev, bir) + table_offset;
438 base = ioremap_nocache(phys_addr, nr_entries * PCI_MSIX_ENTRY_SIZE);
442 /* MSI-X Table Initialization */
443 for (i = 0; i < nvec; i++) {
444 entry = alloc_msi_entry();
448 j = entries[i].entry;
449 entry->msi_attrib.type = PCI_CAP_ID_MSIX;
450 entry->msi_attrib.is_64 = 1;
451 entry->msi_attrib.entry_nr = j;
452 entry->msi_attrib.maskbit = 1;
453 entry->msi_attrib.masked = 1;
454 entry->msi_attrib.default_irq = dev->irq;
455 entry->msi_attrib.pos = pos;
457 entry->mask_base = base;
459 list_add_tail(&entry->list, &dev->msi_list);
462 ret = arch_setup_msi_irqs(dev, nvec, PCI_CAP_ID_MSIX);
465 list_for_each_entry(entry, &dev->msi_list, list) {
466 if (entry->irq != 0) {
473 /* If we had some success report the number of irqs
474 * we succeeded in setting up.
482 list_for_each_entry(entry, &dev->msi_list, list) {
483 entries[i].vector = entry->irq;
484 set_irq_msi(entry->irq, entry);
487 /* Set MSI-X enabled bits */
488 pci_intx_for_msi(dev, 0);
489 msix_set_enable(dev, 1);
490 dev->msix_enabled = 1;
496 * pci_msi_check_device - check whether MSI may be enabled on a device
497 * @dev: pointer to the pci_dev data structure of MSI device function
498 * @nvec: how many MSIs have been requested ?
499 * @type: are we checking for MSI or MSI-X ?
501 * Look at global flags, the device itself, and its parent busses
502 * to determine if MSI/-X are supported for the device. If MSI/-X is
503 * supported return 0, else return an error code.
505 static int pci_msi_check_device(struct pci_dev* dev, int nvec, int type)
510 /* MSI must be globally enabled and supported by the device */
511 if (!pci_msi_enable || !dev || dev->no_msi)
515 * You can't ask to have 0 or less MSIs configured.
517 * b) the list manipulation code assumes nvec >= 1.
522 /* Any bridge which does NOT route MSI transactions from it's
523 * secondary bus to it's primary bus must set NO_MSI flag on
524 * the secondary pci_bus.
525 * We expect only arch-specific PCI host bus controller driver
526 * or quirks for specific PCI bridges to be setting NO_MSI.
528 for (bus = dev->bus; bus; bus = bus->parent)
529 if (bus->bus_flags & PCI_BUS_FLAGS_NO_MSI)
532 ret = arch_msi_check_device(dev, nvec, type);
536 if (!pci_find_capability(dev, type))
543 * pci_enable_msi - configure device's MSI capability structure
544 * @dev: pointer to the pci_dev data structure of MSI device function
546 * Setup the MSI capability structure of device function with
547 * a single MSI irq upon its software driver call to request for
548 * MSI mode enabled on its hardware device function. A return of zero
549 * indicates the successful setup of an entry zero with the new MSI
550 * irq or non-zero for otherwise.
552 int pci_enable_msi(struct pci_dev* dev)
556 status = pci_msi_check_device(dev, 1, PCI_CAP_ID_MSI);
560 WARN_ON(!!dev->msi_enabled);
562 /* Check whether driver already requested for MSI-X irqs */
563 if (dev->msix_enabled) {
564 printk(KERN_INFO "PCI: %s: Can't enable MSI. "
565 "Device already has MSI-X enabled\n",
569 status = msi_capability_init(dev);
572 EXPORT_SYMBOL(pci_enable_msi);
574 void pci_disable_msi(struct pci_dev* dev)
576 struct msi_desc *entry;
579 if (!pci_msi_enable || !dev || !dev->msi_enabled)
582 msi_set_enable(dev, 0);
583 pci_intx_for_msi(dev, 1);
584 dev->msi_enabled = 0;
586 BUG_ON(list_empty(&dev->msi_list));
587 entry = list_entry(dev->msi_list.next, struct msi_desc, list);
588 /* Return the the pci reset with msi irqs unmasked */
589 if (entry->msi_attrib.maskbit) {
590 u32 mask = entry->msi_attrib.maskbits_mask;
591 msi_set_mask_bits(dev->irq, mask, ~mask);
593 if (!entry->dev || entry->msi_attrib.type != PCI_CAP_ID_MSI) {
597 default_irq = entry->msi_attrib.default_irq;
600 /* Restore dev->irq to its default pin-assertion irq */
601 dev->irq = default_irq;
603 EXPORT_SYMBOL(pci_disable_msi);
605 static int msi_free_irqs(struct pci_dev* dev)
607 struct msi_desc *entry, *tmp;
609 list_for_each_entry(entry, &dev->msi_list, list) {
611 BUG_ON(irq_has_action(entry->irq));
614 arch_teardown_msi_irqs(dev);
616 list_for_each_entry_safe(entry, tmp, &dev->msi_list, list) {
617 if (entry->msi_attrib.type == PCI_CAP_ID_MSIX) {
618 writel(1, entry->mask_base + entry->msi_attrib.entry_nr
619 * PCI_MSIX_ENTRY_SIZE
620 + PCI_MSIX_ENTRY_VECTOR_CTRL_OFFSET);
622 if (list_is_last(&entry->list, &dev->msi_list))
623 iounmap(entry->mask_base);
625 list_del(&entry->list);
633 * pci_enable_msix - configure device's MSI-X capability structure
634 * @dev: pointer to the pci_dev data structure of MSI-X device function
635 * @entries: pointer to an array of MSI-X entries
636 * @nvec: number of MSI-X irqs requested for allocation by device driver
638 * Setup the MSI-X capability structure of device function with the number
639 * of requested irqs upon its software driver call to request for
640 * MSI-X mode enabled on its hardware device function. A return of zero
641 * indicates the successful configuration of MSI-X capability structure
642 * with new allocated MSI-X irqs. A return of < 0 indicates a failure.
643 * Or a return of > 0 indicates that driver request is exceeding the number
644 * of irqs available. Driver should use the returned value to re-send
647 int pci_enable_msix(struct pci_dev* dev, struct msix_entry *entries, int nvec)
649 int status, pos, nr_entries;
656 status = pci_msi_check_device(dev, nvec, PCI_CAP_ID_MSIX);
660 pos = pci_find_capability(dev, PCI_CAP_ID_MSIX);
661 pci_read_config_word(dev, msi_control_reg(pos), &control);
662 nr_entries = multi_msix_capable(control);
663 if (nvec > nr_entries)
666 /* Check for any invalid entries */
667 for (i = 0; i < nvec; i++) {
668 if (entries[i].entry >= nr_entries)
669 return -EINVAL; /* invalid entry */
670 for (j = i + 1; j < nvec; j++) {
671 if (entries[i].entry == entries[j].entry)
672 return -EINVAL; /* duplicate entry */
675 WARN_ON(!!dev->msix_enabled);
677 /* Check whether driver already requested for MSI irq */
678 if (dev->msi_enabled) {
679 printk(KERN_INFO "PCI: %s: Can't enable MSI-X. "
680 "Device already has an MSI irq assigned\n",
684 status = msix_capability_init(dev, entries, nvec);
687 EXPORT_SYMBOL(pci_enable_msix);
689 static void msix_free_all_irqs(struct pci_dev *dev)
694 void pci_disable_msix(struct pci_dev* dev)
696 if (!pci_msi_enable || !dev || !dev->msix_enabled)
699 msix_set_enable(dev, 0);
700 pci_intx_for_msi(dev, 1);
701 dev->msix_enabled = 0;
703 msix_free_all_irqs(dev);
705 EXPORT_SYMBOL(pci_disable_msix);
708 * msi_remove_pci_irq_vectors - reclaim MSI(X) irqs to unused state
709 * @dev: pointer to the pci_dev data structure of MSI(X) device function
711 * Being called during hotplug remove, from which the device function
712 * is hot-removed. All previous assigned MSI/MSI-X irqs, if
713 * allocated for this device function, are reclaimed to unused state,
714 * which may be used later on.
716 void msi_remove_pci_irq_vectors(struct pci_dev* dev)
718 if (!pci_msi_enable || !dev)
721 if (dev->msi_enabled)
724 if (dev->msix_enabled)
725 msix_free_all_irqs(dev);
728 void pci_no_msi(void)
733 void pci_msi_init_pci_dev(struct pci_dev *dev)
735 INIT_LIST_HEAD(&dev->msi_list);