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 inline __attribute_const__ u32 msi_capable_mask(u16 control)
110 return msi_mask((control >> 1) & 7);
113 static inline __attribute_const__ u32 msi_enabled_mask(u16 control)
115 return msi_mask((control >> 4) & 7);
119 * PCI 2.3 does not specify mask bits for each MSI interrupt. Attempting to
120 * mask all MSI interrupts by clearing the MSI enable bit does not work
121 * reliably as devices without an INTx disable bit will then generate a
122 * level IRQ which will never be cleared.
124 * Returns 1 if it succeeded in masking the interrupt and 0 if the device
125 * doesn't support MSI masking.
127 static void msi_mask_irq(struct msi_desc *desc, u32 mask, u32 flag)
129 u32 mask_bits = desc->masked;
131 if (!desc->msi_attrib.maskbit)
136 pci_write_config_dword(desc->dev, desc->mask_pos, mask_bits);
137 desc->masked = mask_bits;
141 * This internal function does not flush PCI writes to the device.
142 * All users must ensure that they read from the device before either
143 * assuming that the device state is up to date, or returning out of this
144 * file. This saves a few milliseconds when initialising devices with lots
145 * of MSI-X interrupts.
147 static void msix_mask_irq(struct msi_desc *desc, u32 flag)
149 u32 mask_bits = desc->masked;
150 unsigned offset = desc->msi_attrib.entry_nr * PCI_MSIX_ENTRY_SIZE +
151 PCI_MSIX_ENTRY_VECTOR_CTRL_OFFSET;
154 writel(mask_bits, desc->mask_base + offset);
155 desc->masked = mask_bits;
158 static void msi_set_mask_bit(unsigned irq, u32 flag)
160 struct msi_desc *desc = get_irq_msi(irq);
162 if (desc->msi_attrib.is_msix) {
163 msix_mask_irq(desc, flag);
164 readl(desc->mask_base); /* Flush write to device */
166 msi_mask_irq(desc, 1, flag);
170 void mask_msi_irq(unsigned int irq)
172 msi_set_mask_bit(irq, 1);
175 void unmask_msi_irq(unsigned int irq)
177 msi_set_mask_bit(irq, 0);
180 void read_msi_msg_desc(struct irq_desc *desc, struct msi_msg *msg)
182 struct msi_desc *entry = get_irq_desc_msi(desc);
183 if (entry->msi_attrib.is_msix) {
184 void __iomem *base = entry->mask_base +
185 entry->msi_attrib.entry_nr * PCI_MSIX_ENTRY_SIZE;
187 msg->address_lo = readl(base + PCI_MSIX_ENTRY_LOWER_ADDR_OFFSET);
188 msg->address_hi = readl(base + PCI_MSIX_ENTRY_UPPER_ADDR_OFFSET);
189 msg->data = readl(base + PCI_MSIX_ENTRY_DATA_OFFSET);
191 struct pci_dev *dev = entry->dev;
192 int pos = entry->msi_attrib.pos;
195 pci_read_config_dword(dev, msi_lower_address_reg(pos),
197 if (entry->msi_attrib.is_64) {
198 pci_read_config_dword(dev, msi_upper_address_reg(pos),
200 pci_read_config_word(dev, msi_data_reg(pos, 1), &data);
203 pci_read_config_word(dev, msi_data_reg(pos, 0), &data);
209 void read_msi_msg(unsigned int irq, struct msi_msg *msg)
211 struct irq_desc *desc = irq_to_desc(irq);
213 read_msi_msg_desc(desc, msg);
216 void write_msi_msg_desc(struct irq_desc *desc, struct msi_msg *msg)
218 struct msi_desc *entry = get_irq_desc_msi(desc);
219 if (entry->msi_attrib.is_msix) {
221 base = entry->mask_base +
222 entry->msi_attrib.entry_nr * PCI_MSIX_ENTRY_SIZE;
224 writel(msg->address_lo,
225 base + PCI_MSIX_ENTRY_LOWER_ADDR_OFFSET);
226 writel(msg->address_hi,
227 base + PCI_MSIX_ENTRY_UPPER_ADDR_OFFSET);
228 writel(msg->data, base + PCI_MSIX_ENTRY_DATA_OFFSET);
230 struct pci_dev *dev = entry->dev;
231 int pos = entry->msi_attrib.pos;
233 pci_write_config_dword(dev, msi_lower_address_reg(pos),
235 if (entry->msi_attrib.is_64) {
236 pci_write_config_dword(dev, msi_upper_address_reg(pos),
238 pci_write_config_word(dev, msi_data_reg(pos, 1),
241 pci_write_config_word(dev, msi_data_reg(pos, 0),
248 void write_msi_msg(unsigned int irq, struct msi_msg *msg)
250 struct irq_desc *desc = irq_to_desc(irq);
252 write_msi_msg_desc(desc, msg);
255 static int msi_free_irqs(struct pci_dev* dev);
257 static struct msi_desc *alloc_msi_entry(struct pci_dev *dev)
259 struct msi_desc *desc = kzalloc(sizeof(*desc), GFP_KERNEL);
263 INIT_LIST_HEAD(&desc->list);
269 static void pci_intx_for_msi(struct pci_dev *dev, int enable)
271 if (!(dev->dev_flags & PCI_DEV_FLAGS_MSI_INTX_DISABLE_BUG))
272 pci_intx(dev, enable);
275 static void __pci_restore_msi_state(struct pci_dev *dev)
279 struct msi_desc *entry;
281 if (!dev->msi_enabled)
284 entry = get_irq_msi(dev->irq);
285 pos = entry->msi_attrib.pos;
287 pci_intx_for_msi(dev, 0);
288 msi_set_enable(dev, 0);
289 write_msi_msg(dev->irq, &entry->msg);
291 pci_read_config_word(dev, pos + PCI_MSI_FLAGS, &control);
292 msi_mask_irq(entry, msi_capable_mask(control), entry->masked);
293 control &= ~PCI_MSI_FLAGS_QSIZE;
294 control |= PCI_MSI_FLAGS_ENABLE;
295 pci_write_config_word(dev, pos + PCI_MSI_FLAGS, control);
298 static void __pci_restore_msix_state(struct pci_dev *dev)
301 struct msi_desc *entry;
304 if (!dev->msix_enabled)
307 /* route the table */
308 pci_intx_for_msi(dev, 0);
309 msix_set_enable(dev, 0);
311 list_for_each_entry(entry, &dev->msi_list, list) {
312 write_msi_msg(entry->irq, &entry->msg);
313 msix_mask_irq(entry, entry->masked);
316 BUG_ON(list_empty(&dev->msi_list));
317 entry = list_entry(dev->msi_list.next, struct msi_desc, list);
318 pos = entry->msi_attrib.pos;
319 pci_read_config_word(dev, pos + PCI_MSIX_FLAGS, &control);
320 control &= ~PCI_MSIX_FLAGS_MASKALL;
321 control |= PCI_MSIX_FLAGS_ENABLE;
322 pci_write_config_word(dev, pos + PCI_MSIX_FLAGS, control);
325 void pci_restore_msi_state(struct pci_dev *dev)
327 __pci_restore_msi_state(dev);
328 __pci_restore_msix_state(dev);
330 EXPORT_SYMBOL_GPL(pci_restore_msi_state);
333 * msi_capability_init - configure device's MSI capability structure
334 * @dev: pointer to the pci_dev data structure of MSI device function
336 * Setup the MSI capability structure of device function with a single
337 * MSI irq, regardless of device function is capable of handling
338 * multiple messages. A return of zero indicates the successful setup
339 * of an entry zero with the new MSI irq or non-zero for otherwise.
341 static int msi_capability_init(struct pci_dev *dev)
343 struct msi_desc *entry;
348 msi_set_enable(dev, 0); /* Ensure msi is disabled as I set it up */
350 pos = pci_find_capability(dev, PCI_CAP_ID_MSI);
351 pci_read_config_word(dev, msi_control_reg(pos), &control);
352 /* MSI Entry Initialization */
353 entry = alloc_msi_entry(dev);
357 entry->msi_attrib.is_msix = 0;
358 entry->msi_attrib.is_64 = is_64bit_address(control);
359 entry->msi_attrib.entry_nr = 0;
360 entry->msi_attrib.maskbit = is_mask_bit_support(control);
361 entry->msi_attrib.default_irq = dev->irq; /* Save IOAPIC IRQ */
362 entry->msi_attrib.pos = pos;
364 entry->mask_pos = msi_mask_bits_reg(pos, entry->msi_attrib.is_64);
365 /* All MSIs are unmasked by default, Mask them all */
366 if (entry->msi_attrib.maskbit)
367 pci_read_config_dword(dev, entry->mask_pos, &entry->masked);
368 mask = msi_capable_mask(control);
369 msi_mask_irq(entry, mask, mask);
371 list_add_tail(&entry->list, &dev->msi_list);
373 /* Configure MSI capability structure */
374 ret = arch_setup_msi_irqs(dev, 1, PCI_CAP_ID_MSI);
380 /* Set MSI enabled bits */
381 pci_intx_for_msi(dev, 0);
382 msi_set_enable(dev, 1);
383 dev->msi_enabled = 1;
385 dev->irq = entry->irq;
390 * msix_capability_init - configure device's MSI-X capability
391 * @dev: pointer to the pci_dev data structure of MSI-X device function
392 * @entries: pointer to an array of struct msix_entry entries
393 * @nvec: number of @entries
395 * Setup the MSI-X capability structure of device function with a
396 * single MSI-X irq. A return of zero indicates the successful setup of
397 * requested MSI-X entries with allocated irqs or non-zero for otherwise.
399 static int msix_capability_init(struct pci_dev *dev,
400 struct msix_entry *entries, int nvec)
402 struct msi_desc *entry;
403 int pos, i, j, nr_entries, ret;
404 unsigned long phys_addr;
410 msix_set_enable(dev, 0);/* Ensure msix is disabled as I set it up */
412 pos = pci_find_capability(dev, PCI_CAP_ID_MSIX);
413 /* Request & Map MSI-X table region */
414 pci_read_config_word(dev, msi_control_reg(pos), &control);
415 nr_entries = multi_msix_capable(control);
417 pci_read_config_dword(dev, msix_table_offset_reg(pos), &table_offset);
418 bir = (u8)(table_offset & PCI_MSIX_FLAGS_BIRMASK);
419 table_offset &= ~PCI_MSIX_FLAGS_BIRMASK;
420 phys_addr = pci_resource_start (dev, bir) + table_offset;
421 base = ioremap_nocache(phys_addr, nr_entries * PCI_MSIX_ENTRY_SIZE);
425 /* MSI-X Table Initialization */
426 for (i = 0; i < nvec; i++) {
427 entry = alloc_msi_entry(dev);
431 j = entries[i].entry;
432 entry->msi_attrib.is_msix = 1;
433 entry->msi_attrib.is_64 = 1;
434 entry->msi_attrib.entry_nr = j;
435 entry->msi_attrib.default_irq = dev->irq;
436 entry->msi_attrib.pos = pos;
437 entry->mask_base = base;
438 entry->masked = readl(base + j * PCI_MSIX_ENTRY_SIZE +
439 PCI_MSIX_ENTRY_VECTOR_CTRL_OFFSET);
440 msix_mask_irq(entry, 1);
442 list_add_tail(&entry->list, &dev->msi_list);
445 ret = arch_setup_msi_irqs(dev, nvec, PCI_CAP_ID_MSIX);
447 /* If we had some success report the number of irqs
448 * we succeeded in setting up. */
450 list_for_each_entry(entry, &dev->msi_list, list) {
451 if (entry->irq != 0) {
466 list_for_each_entry(entry, &dev->msi_list, list) {
467 entries[i].vector = entry->irq;
468 set_irq_msi(entry->irq, entry);
471 /* Set MSI-X enabled bits */
472 pci_intx_for_msi(dev, 0);
473 msix_set_enable(dev, 1);
474 dev->msix_enabled = 1;
480 * pci_msi_check_device - check whether MSI may be enabled on a device
481 * @dev: pointer to the pci_dev data structure of MSI device function
482 * @nvec: how many MSIs have been requested ?
483 * @type: are we checking for MSI or MSI-X ?
485 * Look at global flags, the device itself, and its parent busses
486 * to determine if MSI/-X are supported for the device. If MSI/-X is
487 * supported return 0, else return an error code.
489 static int pci_msi_check_device(struct pci_dev* dev, int nvec, int type)
494 /* MSI must be globally enabled and supported by the device */
495 if (!pci_msi_enable || !dev || dev->no_msi)
499 * You can't ask to have 0 or less MSIs configured.
501 * b) the list manipulation code assumes nvec >= 1.
506 /* Any bridge which does NOT route MSI transactions from it's
507 * secondary bus to it's primary bus must set NO_MSI flag on
508 * the secondary pci_bus.
509 * We expect only arch-specific PCI host bus controller driver
510 * or quirks for specific PCI bridges to be setting NO_MSI.
512 for (bus = dev->bus; bus; bus = bus->parent)
513 if (bus->bus_flags & PCI_BUS_FLAGS_NO_MSI)
516 ret = arch_msi_check_device(dev, nvec, type);
520 if (!pci_find_capability(dev, type))
527 * pci_enable_msi - configure device's MSI capability structure
528 * @dev: pointer to the pci_dev data structure of MSI device function
530 * Setup the MSI capability structure of device function with
531 * a single MSI irq upon its software driver call to request for
532 * MSI mode enabled on its hardware device function. A return of zero
533 * indicates the successful setup of an entry zero with the new MSI
534 * irq or non-zero for otherwise.
536 int pci_enable_msi(struct pci_dev* dev)
540 status = pci_msi_check_device(dev, 1, PCI_CAP_ID_MSI);
544 WARN_ON(!!dev->msi_enabled);
546 /* Check whether driver already requested for MSI-X irqs */
547 if (dev->msix_enabled) {
548 dev_info(&dev->dev, "can't enable MSI "
549 "(MSI-X already enabled)\n");
552 status = msi_capability_init(dev);
555 EXPORT_SYMBOL(pci_enable_msi);
557 void pci_msi_shutdown(struct pci_dev *dev)
559 struct msi_desc *desc;
563 if (!pci_msi_enable || !dev || !dev->msi_enabled)
566 msi_set_enable(dev, 0);
567 pci_intx_for_msi(dev, 1);
568 dev->msi_enabled = 0;
570 BUG_ON(list_empty(&dev->msi_list));
571 desc = list_first_entry(&dev->msi_list, struct msi_desc, list);
572 pci_read_config_word(dev, desc->msi_attrib.pos + PCI_MSI_FLAGS, &ctrl);
573 mask = msi_capable_mask(ctrl);
574 msi_mask_irq(desc, mask, ~mask);
576 /* Restore dev->irq to its default pin-assertion irq */
577 dev->irq = desc->msi_attrib.default_irq;
580 void pci_disable_msi(struct pci_dev* dev)
582 struct msi_desc *entry;
584 if (!pci_msi_enable || !dev || !dev->msi_enabled)
587 pci_msi_shutdown(dev);
589 entry = list_entry(dev->msi_list.next, struct msi_desc, list);
590 if (entry->msi_attrib.is_msix)
595 EXPORT_SYMBOL(pci_disable_msi);
597 static int msi_free_irqs(struct pci_dev* dev)
599 struct msi_desc *entry, *tmp;
601 list_for_each_entry(entry, &dev->msi_list, list) {
603 BUG_ON(irq_has_action(entry->irq));
606 arch_teardown_msi_irqs(dev);
608 list_for_each_entry_safe(entry, tmp, &dev->msi_list, list) {
609 if (entry->msi_attrib.is_msix) {
610 writel(1, entry->mask_base + entry->msi_attrib.entry_nr
611 * PCI_MSIX_ENTRY_SIZE
612 + PCI_MSIX_ENTRY_VECTOR_CTRL_OFFSET);
614 if (list_is_last(&entry->list, &dev->msi_list))
615 iounmap(entry->mask_base);
617 list_del(&entry->list);
625 * pci_msix_table_size - return the number of device's MSI-X table entries
626 * @dev: pointer to the pci_dev data structure of MSI-X device function
628 int pci_msix_table_size(struct pci_dev *dev)
633 pos = pci_find_capability(dev, PCI_CAP_ID_MSIX);
637 pci_read_config_word(dev, msi_control_reg(pos), &control);
638 return multi_msix_capable(control);
642 * pci_enable_msix - configure device's MSI-X capability structure
643 * @dev: pointer to the pci_dev data structure of MSI-X device function
644 * @entries: pointer to an array of MSI-X entries
645 * @nvec: number of MSI-X irqs requested for allocation by device driver
647 * Setup the MSI-X capability structure of device function with the number
648 * of requested irqs upon its software driver call to request for
649 * MSI-X mode enabled on its hardware device function. A return of zero
650 * indicates the successful configuration of MSI-X capability structure
651 * with new allocated MSI-X irqs. A return of < 0 indicates a failure.
652 * Or a return of > 0 indicates that driver request is exceeding the number
653 * of irqs available. Driver should use the returned value to re-send
656 int pci_enable_msix(struct pci_dev* dev, struct msix_entry *entries, int nvec)
658 int status, nr_entries;
664 status = pci_msi_check_device(dev, nvec, PCI_CAP_ID_MSIX);
668 nr_entries = pci_msix_table_size(dev);
669 if (nvec > nr_entries)
672 /* Check for any invalid entries */
673 for (i = 0; i < nvec; i++) {
674 if (entries[i].entry >= nr_entries)
675 return -EINVAL; /* invalid entry */
676 for (j = i + 1; j < nvec; j++) {
677 if (entries[i].entry == entries[j].entry)
678 return -EINVAL; /* duplicate entry */
681 WARN_ON(!!dev->msix_enabled);
683 /* Check whether driver already requested for MSI irq */
684 if (dev->msi_enabled) {
685 dev_info(&dev->dev, "can't enable MSI-X "
686 "(MSI IRQ already assigned)\n");
689 status = msix_capability_init(dev, entries, nvec);
692 EXPORT_SYMBOL(pci_enable_msix);
694 static void msix_free_all_irqs(struct pci_dev *dev)
699 void pci_msix_shutdown(struct pci_dev* dev)
701 if (!pci_msi_enable || !dev || !dev->msix_enabled)
704 msix_set_enable(dev, 0);
705 pci_intx_for_msi(dev, 1);
706 dev->msix_enabled = 0;
708 void pci_disable_msix(struct pci_dev* dev)
710 if (!pci_msi_enable || !dev || !dev->msix_enabled)
713 pci_msix_shutdown(dev);
715 msix_free_all_irqs(dev);
717 EXPORT_SYMBOL(pci_disable_msix);
720 * msi_remove_pci_irq_vectors - reclaim MSI(X) irqs to unused state
721 * @dev: pointer to the pci_dev data structure of MSI(X) device function
723 * Being called during hotplug remove, from which the device function
724 * is hot-removed. All previous assigned MSI/MSI-X irqs, if
725 * allocated for this device function, are reclaimed to unused state,
726 * which may be used later on.
728 void msi_remove_pci_irq_vectors(struct pci_dev* dev)
730 if (!pci_msi_enable || !dev)
733 if (dev->msi_enabled)
736 if (dev->msix_enabled)
737 msix_free_all_irqs(dev);
740 void pci_no_msi(void)
746 * pci_msi_enabled - is MSI enabled?
748 * Returns true if MSI has not been disabled by the command-line option
751 int pci_msi_enabled(void)
753 return pci_msi_enable;
755 EXPORT_SYMBOL(pci_msi_enabled);
757 void pci_msi_init_pci_dev(struct pci_dev *dev)
759 INIT_LIST_HEAD(&dev->msi_list);