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;
44 * If an architecture wants to support multiple MSI, it needs to
45 * override arch_setup_msi_irqs()
47 if (type == PCI_CAP_ID_MSI && nvec > 1)
50 list_for_each_entry(entry, &dev->msi_list, list) {
51 ret = arch_setup_msi_irq(dev, entry);
62 #ifndef arch_teardown_msi_irqs
63 void arch_teardown_msi_irqs(struct pci_dev *dev)
65 struct msi_desc *entry;
67 list_for_each_entry(entry, &dev->msi_list, list) {
71 nvec = 1 << entry->msi_attrib.multiple;
72 for (i = 0; i < nvec; i++)
73 arch_teardown_msi_irq(entry->irq + i);
78 static void __msi_set_enable(struct pci_dev *dev, int pos, int enable)
83 pci_read_config_word(dev, pos + PCI_MSI_FLAGS, &control);
84 control &= ~PCI_MSI_FLAGS_ENABLE;
86 control |= PCI_MSI_FLAGS_ENABLE;
87 pci_write_config_word(dev, pos + PCI_MSI_FLAGS, control);
91 static void msi_set_enable(struct pci_dev *dev, int enable)
93 __msi_set_enable(dev, pci_find_capability(dev, PCI_CAP_ID_MSI), enable);
96 static void msix_set_enable(struct pci_dev *dev, int enable)
101 pos = pci_find_capability(dev, PCI_CAP_ID_MSIX);
103 pci_read_config_word(dev, pos + PCI_MSIX_FLAGS, &control);
104 control &= ~PCI_MSIX_FLAGS_ENABLE;
106 control |= PCI_MSIX_FLAGS_ENABLE;
107 pci_write_config_word(dev, pos + PCI_MSIX_FLAGS, control);
111 static inline __attribute_const__ u32 msi_mask(unsigned x)
113 /* Don't shift by >= width of type */
116 return (1 << (1 << x)) - 1;
119 static inline __attribute_const__ u32 msi_capable_mask(u16 control)
121 return msi_mask((control >> 1) & 7);
124 static inline __attribute_const__ u32 msi_enabled_mask(u16 control)
126 return msi_mask((control >> 4) & 7);
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 void msi_mask_irq(struct msi_desc *desc, u32 mask, u32 flag)
140 u32 mask_bits = desc->masked;
142 if (!desc->msi_attrib.maskbit)
147 pci_write_config_dword(desc->dev, desc->mask_pos, mask_bits);
148 desc->masked = mask_bits;
152 * This internal function does not flush PCI writes to the device.
153 * All users must ensure that they read from the device before either
154 * assuming that the device state is up to date, or returning out of this
155 * file. This saves a few milliseconds when initialising devices with lots
156 * of MSI-X interrupts.
158 static void msix_mask_irq(struct msi_desc *desc, u32 flag)
160 u32 mask_bits = desc->masked;
161 unsigned offset = desc->msi_attrib.entry_nr * PCI_MSIX_ENTRY_SIZE +
162 PCI_MSIX_ENTRY_VECTOR_CTRL_OFFSET;
165 writel(mask_bits, desc->mask_base + offset);
166 desc->masked = mask_bits;
169 static void msi_set_mask_bit(unsigned irq, u32 flag)
171 struct msi_desc *desc = get_irq_msi(irq);
173 if (desc->msi_attrib.is_msix) {
174 msix_mask_irq(desc, flag);
175 readl(desc->mask_base); /* Flush write to device */
177 unsigned offset = irq - desc->dev->irq;
178 msi_mask_irq(desc, 1 << offset, flag << offset);
182 void mask_msi_irq(unsigned int irq)
184 msi_set_mask_bit(irq, 1);
187 void unmask_msi_irq(unsigned int irq)
189 msi_set_mask_bit(irq, 0);
192 void read_msi_msg_desc(struct irq_desc *desc, struct msi_msg *msg)
194 struct msi_desc *entry = get_irq_desc_msi(desc);
195 if (entry->msi_attrib.is_msix) {
196 void __iomem *base = entry->mask_base +
197 entry->msi_attrib.entry_nr * PCI_MSIX_ENTRY_SIZE;
199 msg->address_lo = readl(base + PCI_MSIX_ENTRY_LOWER_ADDR_OFFSET);
200 msg->address_hi = readl(base + PCI_MSIX_ENTRY_UPPER_ADDR_OFFSET);
201 msg->data = readl(base + PCI_MSIX_ENTRY_DATA_OFFSET);
203 struct pci_dev *dev = entry->dev;
204 int pos = entry->msi_attrib.pos;
207 pci_read_config_dword(dev, msi_lower_address_reg(pos),
209 if (entry->msi_attrib.is_64) {
210 pci_read_config_dword(dev, msi_upper_address_reg(pos),
212 pci_read_config_word(dev, msi_data_reg(pos, 1), &data);
215 pci_read_config_word(dev, msi_data_reg(pos, 0), &data);
221 void read_msi_msg(unsigned int irq, struct msi_msg *msg)
223 struct irq_desc *desc = irq_to_desc(irq);
225 read_msi_msg_desc(desc, msg);
228 void write_msi_msg_desc(struct irq_desc *desc, struct msi_msg *msg)
230 struct msi_desc *entry = get_irq_desc_msi(desc);
231 if (entry->msi_attrib.is_msix) {
233 base = entry->mask_base +
234 entry->msi_attrib.entry_nr * PCI_MSIX_ENTRY_SIZE;
236 writel(msg->address_lo,
237 base + PCI_MSIX_ENTRY_LOWER_ADDR_OFFSET);
238 writel(msg->address_hi,
239 base + PCI_MSIX_ENTRY_UPPER_ADDR_OFFSET);
240 writel(msg->data, base + PCI_MSIX_ENTRY_DATA_OFFSET);
242 struct pci_dev *dev = entry->dev;
243 int pos = entry->msi_attrib.pos;
246 pci_read_config_word(dev, msi_control_reg(pos), &msgctl);
247 msgctl &= ~PCI_MSI_FLAGS_QSIZE;
248 msgctl |= entry->msi_attrib.multiple << 4;
249 pci_write_config_word(dev, msi_control_reg(pos), msgctl);
251 pci_write_config_dword(dev, msi_lower_address_reg(pos),
253 if (entry->msi_attrib.is_64) {
254 pci_write_config_dword(dev, msi_upper_address_reg(pos),
256 pci_write_config_word(dev, msi_data_reg(pos, 1),
259 pci_write_config_word(dev, msi_data_reg(pos, 0),
266 void write_msi_msg(unsigned int irq, struct msi_msg *msg)
268 struct irq_desc *desc = irq_to_desc(irq);
270 write_msi_msg_desc(desc, msg);
273 static int msi_free_irqs(struct pci_dev* dev);
275 static struct msi_desc *alloc_msi_entry(struct pci_dev *dev)
277 struct msi_desc *desc = kzalloc(sizeof(*desc), GFP_KERNEL);
281 INIT_LIST_HEAD(&desc->list);
287 static void pci_intx_for_msi(struct pci_dev *dev, int enable)
289 if (!(dev->dev_flags & PCI_DEV_FLAGS_MSI_INTX_DISABLE_BUG))
290 pci_intx(dev, enable);
293 static void __pci_restore_msi_state(struct pci_dev *dev)
297 struct msi_desc *entry;
299 if (!dev->msi_enabled)
302 entry = get_irq_msi(dev->irq);
303 pos = entry->msi_attrib.pos;
305 pci_intx_for_msi(dev, 0);
306 msi_set_enable(dev, 0);
307 write_msi_msg(dev->irq, &entry->msg);
309 pci_read_config_word(dev, pos + PCI_MSI_FLAGS, &control);
310 msi_mask_irq(entry, msi_capable_mask(control), entry->masked);
311 control &= ~PCI_MSI_FLAGS_QSIZE;
312 control |= (entry->msi_attrib.multiple << 4) | 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 msix_mask_irq(entry, entry->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
353 * @nvec: number of interrupts to allocate
355 * Setup the MSI capability structure of the device with the requested
356 * number of interrupts. A return value of zero indicates the successful
357 * setup of an entry with the new MSI irq. A negative return value indicates
358 * an error, and a positive return value indicates the number of interrupts
359 * which could have been allocated.
361 static int msi_capability_init(struct pci_dev *dev, int nvec)
363 struct msi_desc *entry;
368 msi_set_enable(dev, 0); /* Ensure msi is disabled as I set it up */
370 pos = pci_find_capability(dev, PCI_CAP_ID_MSI);
371 pci_read_config_word(dev, msi_control_reg(pos), &control);
372 /* MSI Entry Initialization */
373 entry = alloc_msi_entry(dev);
377 entry->msi_attrib.is_msix = 0;
378 entry->msi_attrib.is_64 = is_64bit_address(control);
379 entry->msi_attrib.entry_nr = 0;
380 entry->msi_attrib.maskbit = is_mask_bit_support(control);
381 entry->msi_attrib.default_irq = dev->irq; /* Save IOAPIC IRQ */
382 entry->msi_attrib.pos = pos;
384 entry->mask_pos = msi_mask_bits_reg(pos, entry->msi_attrib.is_64);
385 /* All MSIs are unmasked by default, Mask them all */
386 if (entry->msi_attrib.maskbit)
387 pci_read_config_dword(dev, entry->mask_pos, &entry->masked);
388 mask = msi_capable_mask(control);
389 msi_mask_irq(entry, mask, mask);
391 list_add_tail(&entry->list, &dev->msi_list);
393 /* Configure MSI capability structure */
394 ret = arch_setup_msi_irqs(dev, nvec, PCI_CAP_ID_MSI);
400 /* Set MSI enabled bits */
401 pci_intx_for_msi(dev, 0);
402 msi_set_enable(dev, 1);
403 dev->msi_enabled = 1;
405 dev->irq = entry->irq;
410 * msix_capability_init - configure device's MSI-X capability
411 * @dev: pointer to the pci_dev data structure of MSI-X device function
412 * @entries: pointer to an array of struct msix_entry entries
413 * @nvec: number of @entries
415 * Setup the MSI-X capability structure of device function with a
416 * single MSI-X irq. A return of zero indicates the successful setup of
417 * requested MSI-X entries with allocated irqs or non-zero for otherwise.
419 static int msix_capability_init(struct pci_dev *dev,
420 struct msix_entry *entries, int nvec)
422 struct msi_desc *entry;
423 int pos, i, j, nr_entries, ret;
424 unsigned long phys_addr;
430 msix_set_enable(dev, 0);/* Ensure msix is disabled as I set it up */
432 pos = pci_find_capability(dev, PCI_CAP_ID_MSIX);
433 /* Request & Map MSI-X table region */
434 pci_read_config_word(dev, msi_control_reg(pos), &control);
435 nr_entries = multi_msix_capable(control);
437 pci_read_config_dword(dev, msix_table_offset_reg(pos), &table_offset);
438 bir = (u8)(table_offset & PCI_MSIX_FLAGS_BIRMASK);
439 table_offset &= ~PCI_MSIX_FLAGS_BIRMASK;
440 phys_addr = pci_resource_start (dev, bir) + table_offset;
441 base = ioremap_nocache(phys_addr, nr_entries * PCI_MSIX_ENTRY_SIZE);
445 /* MSI-X Table Initialization */
446 for (i = 0; i < nvec; i++) {
447 entry = alloc_msi_entry(dev);
451 j = entries[i].entry;
452 entry->msi_attrib.is_msix = 1;
453 entry->msi_attrib.is_64 = 1;
454 entry->msi_attrib.entry_nr = j;
455 entry->msi_attrib.default_irq = dev->irq;
456 entry->msi_attrib.pos = pos;
457 entry->mask_base = base;
458 msix_mask_irq(entry, 1);
460 list_add_tail(&entry->list, &dev->msi_list);
463 ret = arch_setup_msi_irqs(dev, nvec, PCI_CAP_ID_MSIX);
465 /* If we had some success report the number of irqs
466 * we succeeded in setting up. */
468 list_for_each_entry(entry, &dev->msi_list, list) {
469 if (entry->irq != 0) {
484 list_for_each_entry(entry, &dev->msi_list, list) {
485 entries[i].vector = entry->irq;
486 set_irq_msi(entry->irq, entry);
489 /* Set MSI-X enabled bits */
490 pci_intx_for_msi(dev, 0);
491 msix_set_enable(dev, 1);
492 dev->msix_enabled = 1;
494 list_for_each_entry(entry, &dev->msi_list, list) {
495 int vector = entry->msi_attrib.entry_nr;
496 entry->masked = readl(base + vector * PCI_MSIX_ENTRY_SIZE +
497 PCI_MSIX_ENTRY_VECTOR_CTRL_OFFSET);
504 * pci_msi_check_device - check whether MSI may be enabled on a device
505 * @dev: pointer to the pci_dev data structure of MSI device function
506 * @nvec: how many MSIs have been requested ?
507 * @type: are we checking for MSI or MSI-X ?
509 * Look at global flags, the device itself, and its parent busses
510 * to determine if MSI/-X are supported for the device. If MSI/-X is
511 * supported return 0, else return an error code.
513 static int pci_msi_check_device(struct pci_dev* dev, int nvec, int type)
518 /* MSI must be globally enabled and supported by the device */
519 if (!pci_msi_enable || !dev || dev->no_msi)
523 * You can't ask to have 0 or less MSIs configured.
525 * b) the list manipulation code assumes nvec >= 1.
530 /* Any bridge which does NOT route MSI transactions from it's
531 * secondary bus to it's primary bus must set NO_MSI flag on
532 * the secondary pci_bus.
533 * We expect only arch-specific PCI host bus controller driver
534 * or quirks for specific PCI bridges to be setting NO_MSI.
536 for (bus = dev->bus; bus; bus = bus->parent)
537 if (bus->bus_flags & PCI_BUS_FLAGS_NO_MSI)
540 ret = arch_msi_check_device(dev, nvec, type);
544 if (!pci_find_capability(dev, type))
551 * pci_enable_msi_block - configure device's MSI capability structure
552 * @dev: device to configure
553 * @nvec: number of interrupts to configure
555 * Allocate IRQs for a device with the MSI capability.
556 * This function returns a negative errno if an error occurs. If it
557 * is unable to allocate the number of interrupts requested, it returns
558 * the number of interrupts it might be able to allocate. If it successfully
559 * allocates at least the number of interrupts requested, it returns 0 and
560 * updates the @dev's irq member to the lowest new interrupt number; the
561 * other interrupt numbers allocated to this device are consecutive.
563 int pci_enable_msi_block(struct pci_dev *dev, unsigned int nvec)
565 int status, pos, maxvec;
568 pos = pci_find_capability(dev, PCI_CAP_ID_MSI);
571 pci_read_config_word(dev, pos + PCI_MSI_FLAGS, &msgctl);
572 maxvec = 1 << ((msgctl & PCI_MSI_FLAGS_QMASK) >> 1);
576 status = pci_msi_check_device(dev, nvec, PCI_CAP_ID_MSI);
580 WARN_ON(!!dev->msi_enabled);
582 /* Check whether driver already requested MSI-X irqs */
583 if (dev->msix_enabled) {
584 dev_info(&dev->dev, "can't enable MSI "
585 "(MSI-X already enabled)\n");
589 status = msi_capability_init(dev, nvec);
592 EXPORT_SYMBOL(pci_enable_msi_block);
594 void pci_msi_shutdown(struct pci_dev *dev)
596 struct msi_desc *desc;
600 if (!pci_msi_enable || !dev || !dev->msi_enabled)
603 msi_set_enable(dev, 0);
604 pci_intx_for_msi(dev, 1);
605 dev->msi_enabled = 0;
607 BUG_ON(list_empty(&dev->msi_list));
608 desc = list_first_entry(&dev->msi_list, struct msi_desc, list);
609 pci_read_config_word(dev, desc->msi_attrib.pos + PCI_MSI_FLAGS, &ctrl);
610 mask = msi_capable_mask(ctrl);
611 msi_mask_irq(desc, mask, ~mask);
613 /* Restore dev->irq to its default pin-assertion irq */
614 dev->irq = desc->msi_attrib.default_irq;
617 void pci_disable_msi(struct pci_dev* dev)
619 struct msi_desc *entry;
621 if (!pci_msi_enable || !dev || !dev->msi_enabled)
624 pci_msi_shutdown(dev);
626 entry = list_entry(dev->msi_list.next, struct msi_desc, list);
627 if (entry->msi_attrib.is_msix)
632 EXPORT_SYMBOL(pci_disable_msi);
634 static int msi_free_irqs(struct pci_dev* dev)
636 struct msi_desc *entry, *tmp;
638 list_for_each_entry(entry, &dev->msi_list, list) {
642 nvec = 1 << entry->msi_attrib.multiple;
643 for (i = 0; i < nvec; i++)
644 BUG_ON(irq_has_action(entry->irq + i));
647 arch_teardown_msi_irqs(dev);
649 list_for_each_entry_safe(entry, tmp, &dev->msi_list, list) {
650 if (entry->msi_attrib.is_msix) {
651 writel(1, entry->mask_base + entry->msi_attrib.entry_nr
652 * PCI_MSIX_ENTRY_SIZE
653 + PCI_MSIX_ENTRY_VECTOR_CTRL_OFFSET);
655 if (list_is_last(&entry->list, &dev->msi_list))
656 iounmap(entry->mask_base);
658 list_del(&entry->list);
666 * pci_msix_table_size - return the number of device's MSI-X table entries
667 * @dev: pointer to the pci_dev data structure of MSI-X device function
669 int pci_msix_table_size(struct pci_dev *dev)
674 pos = pci_find_capability(dev, PCI_CAP_ID_MSIX);
678 pci_read_config_word(dev, msi_control_reg(pos), &control);
679 return multi_msix_capable(control);
683 * pci_enable_msix - configure device's MSI-X capability structure
684 * @dev: pointer to the pci_dev data structure of MSI-X device function
685 * @entries: pointer to an array of MSI-X entries
686 * @nvec: number of MSI-X irqs requested for allocation by device driver
688 * Setup the MSI-X capability structure of device function with the number
689 * of requested irqs upon its software driver call to request for
690 * MSI-X mode enabled on its hardware device function. A return of zero
691 * indicates the successful configuration of MSI-X capability structure
692 * with new allocated MSI-X irqs. A return of < 0 indicates a failure.
693 * Or a return of > 0 indicates that driver request is exceeding the number
694 * of irqs available. Driver should use the returned value to re-send
697 int pci_enable_msix(struct pci_dev* dev, struct msix_entry *entries, int nvec)
699 int status, nr_entries;
705 status = pci_msi_check_device(dev, nvec, PCI_CAP_ID_MSIX);
709 nr_entries = pci_msix_table_size(dev);
710 if (nvec > nr_entries)
713 /* Check for any invalid entries */
714 for (i = 0; i < nvec; i++) {
715 if (entries[i].entry >= nr_entries)
716 return -EINVAL; /* invalid entry */
717 for (j = i + 1; j < nvec; j++) {
718 if (entries[i].entry == entries[j].entry)
719 return -EINVAL; /* duplicate entry */
722 WARN_ON(!!dev->msix_enabled);
724 /* Check whether driver already requested for MSI irq */
725 if (dev->msi_enabled) {
726 dev_info(&dev->dev, "can't enable MSI-X "
727 "(MSI IRQ already assigned)\n");
730 status = msix_capability_init(dev, entries, nvec);
733 EXPORT_SYMBOL(pci_enable_msix);
735 static void msix_free_all_irqs(struct pci_dev *dev)
740 void pci_msix_shutdown(struct pci_dev* dev)
742 if (!pci_msi_enable || !dev || !dev->msix_enabled)
745 msix_set_enable(dev, 0);
746 pci_intx_for_msi(dev, 1);
747 dev->msix_enabled = 0;
749 void pci_disable_msix(struct pci_dev* dev)
751 if (!pci_msi_enable || !dev || !dev->msix_enabled)
754 pci_msix_shutdown(dev);
756 msix_free_all_irqs(dev);
758 EXPORT_SYMBOL(pci_disable_msix);
761 * msi_remove_pci_irq_vectors - reclaim MSI(X) irqs to unused state
762 * @dev: pointer to the pci_dev data structure of MSI(X) device function
764 * Being called during hotplug remove, from which the device function
765 * is hot-removed. All previous assigned MSI/MSI-X irqs, if
766 * allocated for this device function, are reclaimed to unused state,
767 * which may be used later on.
769 void msi_remove_pci_irq_vectors(struct pci_dev* dev)
771 if (!pci_msi_enable || !dev)
774 if (dev->msi_enabled)
777 if (dev->msix_enabled)
778 msix_free_all_irqs(dev);
781 void pci_no_msi(void)
787 * pci_msi_enabled - is MSI enabled?
789 * Returns true if MSI has not been disabled by the command-line option
792 int pci_msi_enabled(void)
794 return pci_msi_enable;
796 EXPORT_SYMBOL(pci_msi_enabled);
798 void pci_msi_init_pci_dev(struct pci_dev *dev)
800 INIT_LIST_HEAD(&dev->msi_list);