3 * Purpose: PCI Message Signaled Interrupt (MSI)
5 * Copyright (C) 2003-2004 Intel
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>
19 #include <linux/errno.h>
25 static int pci_msi_enable = 1;
29 #ifndef arch_msi_check_device
30 int arch_msi_check_device(struct pci_dev *dev, int nvec, int type)
36 #ifndef arch_setup_msi_irqs
37 int arch_setup_msi_irqs(struct pci_dev *dev, int nvec, int type)
39 struct msi_desc *entry;
43 * If an architecture wants to support multiple MSI, it needs to
44 * override arch_setup_msi_irqs()
46 if (type == PCI_CAP_ID_MSI && nvec > 1)
49 list_for_each_entry(entry, &dev->msi_list, list) {
50 ret = arch_setup_msi_irq(dev, entry);
61 #ifndef arch_teardown_msi_irqs
62 void arch_teardown_msi_irqs(struct pci_dev *dev)
64 struct msi_desc *entry;
66 list_for_each_entry(entry, &dev->msi_list, list) {
70 nvec = 1 << entry->msi_attrib.multiple;
71 for (i = 0; i < nvec; i++)
72 arch_teardown_msi_irq(entry->irq + i);
77 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);
90 static void msix_set_enable(struct pci_dev *dev, int enable)
95 pos = pci_find_capability(dev, PCI_CAP_ID_MSIX);
97 pci_read_config_word(dev, pos + PCI_MSIX_FLAGS, &control);
98 control &= ~PCI_MSIX_FLAGS_ENABLE;
100 control |= PCI_MSIX_FLAGS_ENABLE;
101 pci_write_config_word(dev, pos + PCI_MSIX_FLAGS, control);
105 static inline __attribute_const__ u32 msi_mask(unsigned x)
107 /* Don't shift by >= width of type */
110 return (1 << (1 << x)) - 1;
113 static inline __attribute_const__ u32 msi_capable_mask(u16 control)
115 return msi_mask((control >> 1) & 7);
118 static inline __attribute_const__ u32 msi_enabled_mask(u16 control)
120 return msi_mask((control >> 4) & 7);
124 * PCI 2.3 does not specify mask bits for each MSI interrupt. Attempting to
125 * mask all MSI interrupts by clearing the MSI enable bit does not work
126 * reliably as devices without an INTx disable bit will then generate a
127 * level IRQ which will never be cleared.
129 static u32 __msi_mask_irq(struct msi_desc *desc, u32 mask, u32 flag)
131 u32 mask_bits = desc->masked;
133 if (!desc->msi_attrib.maskbit)
138 pci_write_config_dword(desc->dev, desc->mask_pos, mask_bits);
143 static void msi_mask_irq(struct msi_desc *desc, u32 mask, u32 flag)
145 desc->masked = __msi_mask_irq(desc, mask, flag);
149 * This internal function does not flush PCI writes to the device.
150 * All users must ensure that they read from the device before either
151 * assuming that the device state is up to date, or returning out of this
152 * file. This saves a few milliseconds when initialising devices with lots
153 * of MSI-X interrupts.
155 static u32 __msix_mask_irq(struct msi_desc *desc, u32 flag)
157 u32 mask_bits = desc->masked;
158 unsigned offset = desc->msi_attrib.entry_nr * PCI_MSIX_ENTRY_SIZE +
159 PCI_MSIX_ENTRY_VECTOR_CTRL;
162 writel(mask_bits, desc->mask_base + offset);
167 static void msix_mask_irq(struct msi_desc *desc, u32 flag)
169 desc->masked = __msix_mask_irq(desc, flag);
172 static void msi_set_mask_bit(unsigned irq, u32 flag)
174 struct msi_desc *desc = get_irq_msi(irq);
176 if (desc->msi_attrib.is_msix) {
177 msix_mask_irq(desc, flag);
178 readl(desc->mask_base); /* Flush write to device */
180 unsigned offset = irq - desc->dev->irq;
181 msi_mask_irq(desc, 1 << offset, flag << offset);
185 void mask_msi_irq(unsigned int irq)
187 msi_set_mask_bit(irq, 1);
190 void unmask_msi_irq(unsigned int irq)
192 msi_set_mask_bit(irq, 0);
195 void read_msi_msg_desc(struct irq_desc *desc, struct msi_msg *msg)
197 struct msi_desc *entry = get_irq_desc_msi(desc);
198 if (entry->msi_attrib.is_msix) {
199 void __iomem *base = entry->mask_base +
200 entry->msi_attrib.entry_nr * PCI_MSIX_ENTRY_SIZE;
202 msg->address_lo = readl(base + PCI_MSIX_ENTRY_LOWER_ADDR);
203 msg->address_hi = readl(base + PCI_MSIX_ENTRY_UPPER_ADDR);
204 msg->data = readl(base + PCI_MSIX_ENTRY_DATA);
206 struct pci_dev *dev = entry->dev;
207 int pos = entry->msi_attrib.pos;
210 pci_read_config_dword(dev, msi_lower_address_reg(pos),
212 if (entry->msi_attrib.is_64) {
213 pci_read_config_dword(dev, msi_upper_address_reg(pos),
215 pci_read_config_word(dev, msi_data_reg(pos, 1), &data);
218 pci_read_config_word(dev, msi_data_reg(pos, 0), &data);
224 void read_msi_msg(unsigned int irq, struct msi_msg *msg)
226 struct irq_desc *desc = irq_to_desc(irq);
228 read_msi_msg_desc(desc, msg);
231 void write_msi_msg_desc(struct irq_desc *desc, struct msi_msg *msg)
233 struct msi_desc *entry = get_irq_desc_msi(desc);
234 if (entry->msi_attrib.is_msix) {
236 base = entry->mask_base +
237 entry->msi_attrib.entry_nr * PCI_MSIX_ENTRY_SIZE;
239 writel(msg->address_lo, base + PCI_MSIX_ENTRY_LOWER_ADDR);
240 writel(msg->address_hi, base + PCI_MSIX_ENTRY_UPPER_ADDR);
241 writel(msg->data, base + PCI_MSIX_ENTRY_DATA);
243 struct pci_dev *dev = entry->dev;
244 int pos = entry->msi_attrib.pos;
247 pci_read_config_word(dev, msi_control_reg(pos), &msgctl);
248 msgctl &= ~PCI_MSI_FLAGS_QSIZE;
249 msgctl |= entry->msi_attrib.multiple << 4;
250 pci_write_config_word(dev, msi_control_reg(pos), msgctl);
252 pci_write_config_dword(dev, msi_lower_address_reg(pos),
254 if (entry->msi_attrib.is_64) {
255 pci_write_config_dword(dev, msi_upper_address_reg(pos),
257 pci_write_config_word(dev, msi_data_reg(pos, 1),
260 pci_write_config_word(dev, msi_data_reg(pos, 0),
267 void write_msi_msg(unsigned int irq, struct msi_msg *msg)
269 struct irq_desc *desc = irq_to_desc(irq);
271 write_msi_msg_desc(desc, msg);
274 static void free_msi_irqs(struct pci_dev *dev)
276 struct msi_desc *entry, *tmp;
278 list_for_each_entry(entry, &dev->msi_list, list) {
282 nvec = 1 << entry->msi_attrib.multiple;
283 for (i = 0; i < nvec; i++)
284 BUG_ON(irq_has_action(entry->irq + i));
287 arch_teardown_msi_irqs(dev);
289 list_for_each_entry_safe(entry, tmp, &dev->msi_list, list) {
290 if (entry->msi_attrib.is_msix) {
291 if (list_is_last(&entry->list, &dev->msi_list))
292 iounmap(entry->mask_base);
294 list_del(&entry->list);
299 static struct msi_desc *alloc_msi_entry(struct pci_dev *dev)
301 struct msi_desc *desc = kzalloc(sizeof(*desc), GFP_KERNEL);
305 INIT_LIST_HEAD(&desc->list);
311 static void pci_intx_for_msi(struct pci_dev *dev, int enable)
313 if (!(dev->dev_flags & PCI_DEV_FLAGS_MSI_INTX_DISABLE_BUG))
314 pci_intx(dev, enable);
317 static void __pci_restore_msi_state(struct pci_dev *dev)
321 struct msi_desc *entry;
323 if (!dev->msi_enabled)
326 entry = get_irq_msi(dev->irq);
327 pos = entry->msi_attrib.pos;
329 pci_intx_for_msi(dev, 0);
330 msi_set_enable(dev, pos, 0);
331 write_msi_msg(dev->irq, &entry->msg);
333 pci_read_config_word(dev, pos + PCI_MSI_FLAGS, &control);
334 msi_mask_irq(entry, msi_capable_mask(control), entry->masked);
335 control &= ~PCI_MSI_FLAGS_QSIZE;
336 control |= (entry->msi_attrib.multiple << 4) | PCI_MSI_FLAGS_ENABLE;
337 pci_write_config_word(dev, pos + PCI_MSI_FLAGS, control);
340 static void __pci_restore_msix_state(struct pci_dev *dev)
343 struct msi_desc *entry;
346 if (!dev->msix_enabled)
348 BUG_ON(list_empty(&dev->msi_list));
349 entry = list_first_entry(&dev->msi_list, struct msi_desc, list);
350 pos = entry->msi_attrib.pos;
351 pci_read_config_word(dev, pos + PCI_MSIX_FLAGS, &control);
353 /* route the table */
354 pci_intx_for_msi(dev, 0);
355 control |= PCI_MSIX_FLAGS_ENABLE | PCI_MSIX_FLAGS_MASKALL;
356 pci_write_config_word(dev, pos + PCI_MSIX_FLAGS, control);
358 list_for_each_entry(entry, &dev->msi_list, list) {
359 write_msi_msg(entry->irq, &entry->msg);
360 msix_mask_irq(entry, entry->masked);
363 control &= ~PCI_MSIX_FLAGS_MASKALL;
364 pci_write_config_word(dev, pos + PCI_MSIX_FLAGS, control);
367 void pci_restore_msi_state(struct pci_dev *dev)
369 __pci_restore_msi_state(dev);
370 __pci_restore_msix_state(dev);
372 EXPORT_SYMBOL_GPL(pci_restore_msi_state);
375 * msi_capability_init - configure device's MSI capability structure
376 * @dev: pointer to the pci_dev data structure of MSI device function
377 * @nvec: number of interrupts to allocate
379 * Setup the MSI capability structure of the device with the requested
380 * number of interrupts. A return value of zero indicates the successful
381 * setup of an entry with the new MSI irq. A negative return value indicates
382 * an error, and a positive return value indicates the number of interrupts
383 * which could have been allocated.
385 static int msi_capability_init(struct pci_dev *dev, int nvec)
387 struct msi_desc *entry;
392 pos = pci_find_capability(dev, PCI_CAP_ID_MSI);
393 msi_set_enable(dev, pos, 0); /* Disable MSI during set up */
395 pci_read_config_word(dev, msi_control_reg(pos), &control);
396 /* MSI Entry Initialization */
397 entry = alloc_msi_entry(dev);
401 entry->msi_attrib.is_msix = 0;
402 entry->msi_attrib.is_64 = is_64bit_address(control);
403 entry->msi_attrib.entry_nr = 0;
404 entry->msi_attrib.maskbit = is_mask_bit_support(control);
405 entry->msi_attrib.default_irq = dev->irq; /* Save IOAPIC IRQ */
406 entry->msi_attrib.pos = pos;
408 entry->mask_pos = msi_mask_reg(pos, entry->msi_attrib.is_64);
409 /* All MSIs are unmasked by default, Mask them all */
410 if (entry->msi_attrib.maskbit)
411 pci_read_config_dword(dev, entry->mask_pos, &entry->masked);
412 mask = msi_capable_mask(control);
413 msi_mask_irq(entry, mask, mask);
415 list_add_tail(&entry->list, &dev->msi_list);
417 /* Configure MSI capability structure */
418 ret = arch_setup_msi_irqs(dev, nvec, PCI_CAP_ID_MSI);
420 msi_mask_irq(entry, mask, ~mask);
425 /* Set MSI enabled bits */
426 pci_intx_for_msi(dev, 0);
427 msi_set_enable(dev, pos, 1);
428 dev->msi_enabled = 1;
430 dev->irq = entry->irq;
434 static void __iomem *msix_map_region(struct pci_dev *dev, unsigned pos,
437 unsigned long phys_addr;
441 pci_read_config_dword(dev, msix_table_offset_reg(pos), &table_offset);
442 bir = (u8)(table_offset & PCI_MSIX_FLAGS_BIRMASK);
443 table_offset &= ~PCI_MSIX_FLAGS_BIRMASK;
444 phys_addr = pci_resource_start(dev, bir) + table_offset;
446 return ioremap_nocache(phys_addr, nr_entries * PCI_MSIX_ENTRY_SIZE);
449 static int msix_setup_entries(struct pci_dev *dev, unsigned pos,
450 void __iomem *base, struct msix_entry *entries,
453 struct msi_desc *entry;
456 for (i = 0; i < nvec; i++) {
457 entry = alloc_msi_entry(dev);
463 /* No enough memory. Don't try again */
467 entry->msi_attrib.is_msix = 1;
468 entry->msi_attrib.is_64 = 1;
469 entry->msi_attrib.entry_nr = entries[i].entry;
470 entry->msi_attrib.default_irq = dev->irq;
471 entry->msi_attrib.pos = pos;
472 entry->mask_base = base;
474 list_add_tail(&entry->list, &dev->msi_list);
480 static void msix_program_entries(struct pci_dev *dev,
481 struct msix_entry *entries)
483 struct msi_desc *entry;
486 list_for_each_entry(entry, &dev->msi_list, list) {
487 int offset = entries[i].entry * PCI_MSIX_ENTRY_SIZE +
488 PCI_MSIX_ENTRY_VECTOR_CTRL;
490 entries[i].vector = entry->irq;
491 set_irq_msi(entry->irq, entry);
492 entry->masked = readl(entry->mask_base + offset);
493 msix_mask_irq(entry, 1);
499 * msix_capability_init - configure device's MSI-X capability
500 * @dev: pointer to the pci_dev data structure of MSI-X device function
501 * @entries: pointer to an array of struct msix_entry entries
502 * @nvec: number of @entries
504 * Setup the MSI-X capability structure of device function with a
505 * single MSI-X irq. A return of zero indicates the successful setup of
506 * requested MSI-X entries with allocated irqs or non-zero for otherwise.
508 static int msix_capability_init(struct pci_dev *dev,
509 struct msix_entry *entries, int nvec)
515 pos = pci_find_capability(dev, PCI_CAP_ID_MSIX);
516 pci_read_config_word(dev, pos + PCI_MSIX_FLAGS, &control);
518 /* Ensure MSI-X is disabled while it is set up */
519 control &= ~PCI_MSIX_FLAGS_ENABLE;
520 pci_write_config_word(dev, pos + PCI_MSIX_FLAGS, control);
522 /* Request & Map MSI-X table region */
523 base = msix_map_region(dev, pos, multi_msix_capable(control));
527 ret = msix_setup_entries(dev, pos, base, entries, nvec);
531 ret = arch_setup_msi_irqs(dev, nvec, PCI_CAP_ID_MSIX);
536 * Some devices require MSI-X to be enabled before we can touch the
537 * MSI-X registers. We need to mask all the vectors to prevent
538 * interrupts coming in before they're fully set up.
540 control |= PCI_MSIX_FLAGS_MASKALL | PCI_MSIX_FLAGS_ENABLE;
541 pci_write_config_word(dev, pos + PCI_MSIX_FLAGS, control);
543 msix_program_entries(dev, entries);
545 /* Set MSI-X enabled bits and unmask the function */
546 pci_intx_for_msi(dev, 0);
547 dev->msix_enabled = 1;
549 control &= ~PCI_MSIX_FLAGS_MASKALL;
550 pci_write_config_word(dev, pos + PCI_MSIX_FLAGS, control);
557 * If we had some success, report the number of irqs
558 * we succeeded in setting up.
560 struct msi_desc *entry;
563 list_for_each_entry(entry, &dev->msi_list, list) {
577 * pci_msi_check_device - check whether MSI may be enabled on a device
578 * @dev: pointer to the pci_dev data structure of MSI device function
579 * @nvec: how many MSIs have been requested ?
580 * @type: are we checking for MSI or MSI-X ?
582 * Look at global flags, the device itself, and its parent busses
583 * to determine if MSI/-X are supported for the device. If MSI/-X is
584 * supported return 0, else return an error code.
586 static int pci_msi_check_device(struct pci_dev *dev, int nvec, int type)
591 /* MSI must be globally enabled and supported by the device */
592 if (!pci_msi_enable || !dev || dev->no_msi)
596 * You can't ask to have 0 or less MSIs configured.
598 * b) the list manipulation code assumes nvec >= 1.
604 * Any bridge which does NOT route MSI transactions from its
605 * secondary bus to its primary bus must set NO_MSI flag on
606 * the secondary pci_bus.
607 * We expect only arch-specific PCI host bus controller driver
608 * or quirks for specific PCI bridges to be setting NO_MSI.
610 for (bus = dev->bus; bus; bus = bus->parent)
611 if (bus->bus_flags & PCI_BUS_FLAGS_NO_MSI)
614 ret = arch_msi_check_device(dev, nvec, type);
618 if (!pci_find_capability(dev, type))
625 * pci_enable_msi_block - configure device's MSI capability structure
626 * @dev: device to configure
627 * @nvec: number of interrupts to configure
629 * Allocate IRQs for a device with the MSI capability.
630 * This function returns a negative errno if an error occurs. If it
631 * is unable to allocate the number of interrupts requested, it returns
632 * the number of interrupts it might be able to allocate. If it successfully
633 * allocates at least the number of interrupts requested, it returns 0 and
634 * updates the @dev's irq member to the lowest new interrupt number; the
635 * other interrupt numbers allocated to this device are consecutive.
637 int pci_enable_msi_block(struct pci_dev *dev, unsigned int nvec)
639 int status, pos, maxvec;
642 pos = pci_find_capability(dev, PCI_CAP_ID_MSI);
645 pci_read_config_word(dev, pos + PCI_MSI_FLAGS, &msgctl);
646 maxvec = 1 << ((msgctl & PCI_MSI_FLAGS_QMASK) >> 1);
650 status = pci_msi_check_device(dev, nvec, PCI_CAP_ID_MSI);
654 WARN_ON(!!dev->msi_enabled);
656 /* Check whether driver already requested MSI-X irqs */
657 if (dev->msix_enabled) {
658 dev_info(&dev->dev, "can't enable MSI "
659 "(MSI-X already enabled)\n");
663 status = msi_capability_init(dev, nvec);
666 EXPORT_SYMBOL(pci_enable_msi_block);
668 void pci_msi_shutdown(struct pci_dev *dev)
670 struct msi_desc *desc;
675 if (!pci_msi_enable || !dev || !dev->msi_enabled)
678 BUG_ON(list_empty(&dev->msi_list));
679 desc = list_first_entry(&dev->msi_list, struct msi_desc, list);
680 pos = desc->msi_attrib.pos;
682 msi_set_enable(dev, pos, 0);
683 pci_intx_for_msi(dev, 1);
684 dev->msi_enabled = 0;
686 /* Return the device with MSI unmasked as initial states */
687 pci_read_config_word(dev, pos + PCI_MSI_FLAGS, &ctrl);
688 mask = msi_capable_mask(ctrl);
689 /* Keep cached state to be restored */
690 __msi_mask_irq(desc, mask, ~mask);
692 /* Restore dev->irq to its default pin-assertion irq */
693 dev->irq = desc->msi_attrib.default_irq;
696 void pci_disable_msi(struct pci_dev *dev)
698 if (!pci_msi_enable || !dev || !dev->msi_enabled)
701 pci_msi_shutdown(dev);
704 EXPORT_SYMBOL(pci_disable_msi);
707 * pci_msix_table_size - return the number of device's MSI-X table entries
708 * @dev: pointer to the pci_dev data structure of MSI-X device function
710 int pci_msix_table_size(struct pci_dev *dev)
715 pos = pci_find_capability(dev, PCI_CAP_ID_MSIX);
719 pci_read_config_word(dev, msi_control_reg(pos), &control);
720 return multi_msix_capable(control);
724 * pci_enable_msix - configure device's MSI-X capability structure
725 * @dev: pointer to the pci_dev data structure of MSI-X device function
726 * @entries: pointer to an array of MSI-X entries
727 * @nvec: number of MSI-X irqs requested for allocation by device driver
729 * Setup the MSI-X capability structure of device function with the number
730 * of requested irqs upon its software driver call to request for
731 * MSI-X mode enabled on its hardware device function. A return of zero
732 * indicates the successful configuration of MSI-X capability structure
733 * with new allocated MSI-X irqs. A return of < 0 indicates a failure.
734 * Or a return of > 0 indicates that driver request is exceeding the number
735 * of irqs or MSI-X vectors available. Driver should use the returned value to
736 * re-send its request.
738 int pci_enable_msix(struct pci_dev *dev, struct msix_entry *entries, int nvec)
740 int status, nr_entries;
746 status = pci_msi_check_device(dev, nvec, PCI_CAP_ID_MSIX);
750 nr_entries = pci_msix_table_size(dev);
751 if (nvec > nr_entries)
754 /* Check for any invalid entries */
755 for (i = 0; i < nvec; i++) {
756 if (entries[i].entry >= nr_entries)
757 return -EINVAL; /* invalid entry */
758 for (j = i + 1; j < nvec; j++) {
759 if (entries[i].entry == entries[j].entry)
760 return -EINVAL; /* duplicate entry */
763 WARN_ON(!!dev->msix_enabled);
765 /* Check whether driver already requested for MSI irq */
766 if (dev->msi_enabled) {
767 dev_info(&dev->dev, "can't enable MSI-X "
768 "(MSI IRQ already assigned)\n");
771 status = msix_capability_init(dev, entries, nvec);
774 EXPORT_SYMBOL(pci_enable_msix);
776 void pci_msix_shutdown(struct pci_dev *dev)
778 struct msi_desc *entry;
780 if (!pci_msi_enable || !dev || !dev->msix_enabled)
783 /* Return the device with MSI-X masked as initial states */
784 list_for_each_entry(entry, &dev->msi_list, list) {
785 /* Keep cached states to be restored */
786 __msix_mask_irq(entry, 1);
789 msix_set_enable(dev, 0);
790 pci_intx_for_msi(dev, 1);
791 dev->msix_enabled = 0;
794 void pci_disable_msix(struct pci_dev *dev)
796 if (!pci_msi_enable || !dev || !dev->msix_enabled)
799 pci_msix_shutdown(dev);
802 EXPORT_SYMBOL(pci_disable_msix);
805 * msi_remove_pci_irq_vectors - reclaim MSI(X) irqs to unused state
806 * @dev: pointer to the pci_dev data structure of MSI(X) device function
808 * Being called during hotplug remove, from which the device function
809 * is hot-removed. All previous assigned MSI/MSI-X irqs, if
810 * allocated for this device function, are reclaimed to unused state,
811 * which may be used later on.
813 void msi_remove_pci_irq_vectors(struct pci_dev *dev)
815 if (!pci_msi_enable || !dev)
818 if (dev->msi_enabled || dev->msix_enabled)
822 void pci_no_msi(void)
828 * pci_msi_enabled - is MSI enabled?
830 * Returns true if MSI has not been disabled by the command-line option
833 int pci_msi_enabled(void)
835 return pci_msi_enable;
837 EXPORT_SYMBOL(pci_msi_enabled);
839 void pci_msi_init_pci_dev(struct pci_dev *dev)
841 INIT_LIST_HEAD(&dev->msi_list);