Merge branch 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/viro/vfs
[linux/fpc-iii.git] / drivers / pci / msi.c
blob7a0fec6ce5717baac5ef564c743d129cbc03a892
1 /*
2 * File: msi.c
3 * Purpose: PCI Message Signaled Interrupt (MSI)
5 * Copyright (C) 2003-2004 Intel
6 * Copyright (C) Tom Long Nguyen (tom.l.nguyen@intel.com)
7 */
9 #include <linux/err.h>
10 #include <linux/mm.h>
11 #include <linux/irq.h>
12 #include <linux/interrupt.h>
13 #include <linux/init.h>
14 #include <linux/export.h>
15 #include <linux/ioport.h>
16 #include <linux/pci.h>
17 #include <linux/proc_fs.h>
18 #include <linux/msi.h>
19 #include <linux/smp.h>
20 #include <linux/errno.h>
21 #include <linux/io.h>
22 #include <linux/slab.h>
24 #include "pci.h"
26 static int pci_msi_enable = 1;
28 #define msix_table_size(flags) ((flags & PCI_MSIX_FLAGS_QSIZE) + 1)
31 /* Arch hooks */
33 int __weak arch_setup_msi_irq(struct pci_dev *dev, struct msi_desc *desc)
35 struct msi_chip *chip = dev->bus->msi;
36 int err;
38 if (!chip || !chip->setup_irq)
39 return -EINVAL;
41 err = chip->setup_irq(chip, dev, desc);
42 if (err < 0)
43 return err;
45 irq_set_chip_data(desc->irq, chip);
47 return 0;
50 void __weak arch_teardown_msi_irq(unsigned int irq)
52 struct msi_chip *chip = irq_get_chip_data(irq);
54 if (!chip || !chip->teardown_irq)
55 return;
57 chip->teardown_irq(chip, irq);
60 int __weak arch_msi_check_device(struct pci_dev *dev, int nvec, int type)
62 struct msi_chip *chip = dev->bus->msi;
64 if (!chip || !chip->check_device)
65 return 0;
67 return chip->check_device(chip, dev, nvec, type);
70 int __weak arch_setup_msi_irqs(struct pci_dev *dev, int nvec, int type)
72 struct msi_desc *entry;
73 int ret;
76 * If an architecture wants to support multiple MSI, it needs to
77 * override arch_setup_msi_irqs()
79 if (type == PCI_CAP_ID_MSI && nvec > 1)
80 return 1;
82 list_for_each_entry(entry, &dev->msi_list, list) {
83 ret = arch_setup_msi_irq(dev, entry);
84 if (ret < 0)
85 return ret;
86 if (ret > 0)
87 return -ENOSPC;
90 return 0;
94 * We have a default implementation available as a separate non-weak
95 * function, as it is used by the Xen x86 PCI code
97 void default_teardown_msi_irqs(struct pci_dev *dev)
99 struct msi_desc *entry;
101 list_for_each_entry(entry, &dev->msi_list, list) {
102 int i, nvec;
103 if (entry->irq == 0)
104 continue;
105 if (entry->nvec_used)
106 nvec = entry->nvec_used;
107 else
108 nvec = 1 << entry->msi_attrib.multiple;
109 for (i = 0; i < nvec; i++)
110 arch_teardown_msi_irq(entry->irq + i);
114 void __weak arch_teardown_msi_irqs(struct pci_dev *dev)
116 return default_teardown_msi_irqs(dev);
119 static void default_restore_msi_irq(struct pci_dev *dev, int irq)
121 struct msi_desc *entry;
123 entry = NULL;
124 if (dev->msix_enabled) {
125 list_for_each_entry(entry, &dev->msi_list, list) {
126 if (irq == entry->irq)
127 break;
129 } else if (dev->msi_enabled) {
130 entry = irq_get_msi_desc(irq);
133 if (entry)
134 write_msi_msg(irq, &entry->msg);
137 void __weak arch_restore_msi_irqs(struct pci_dev *dev)
139 return default_restore_msi_irqs(dev);
142 static void msi_set_enable(struct pci_dev *dev, int enable)
144 u16 control;
146 pci_read_config_word(dev, dev->msi_cap + PCI_MSI_FLAGS, &control);
147 control &= ~PCI_MSI_FLAGS_ENABLE;
148 if (enable)
149 control |= PCI_MSI_FLAGS_ENABLE;
150 pci_write_config_word(dev, dev->msi_cap + PCI_MSI_FLAGS, control);
153 static void msix_set_enable(struct pci_dev *dev, int enable)
155 u16 control;
157 pci_read_config_word(dev, dev->msix_cap + PCI_MSIX_FLAGS, &control);
158 control &= ~PCI_MSIX_FLAGS_ENABLE;
159 if (enable)
160 control |= PCI_MSIX_FLAGS_ENABLE;
161 pci_write_config_word(dev, dev->msix_cap + PCI_MSIX_FLAGS, control);
164 static inline __attribute_const__ u32 msi_mask(unsigned x)
166 /* Don't shift by >= width of type */
167 if (x >= 5)
168 return 0xffffffff;
169 return (1 << (1 << x)) - 1;
172 static inline __attribute_const__ u32 msi_capable_mask(u16 control)
174 return msi_mask((control >> 1) & 7);
177 static inline __attribute_const__ u32 msi_enabled_mask(u16 control)
179 return msi_mask((control >> 4) & 7);
183 * PCI 2.3 does not specify mask bits for each MSI interrupt. Attempting to
184 * mask all MSI interrupts by clearing the MSI enable bit does not work
185 * reliably as devices without an INTx disable bit will then generate a
186 * level IRQ which will never be cleared.
188 u32 default_msi_mask_irq(struct msi_desc *desc, u32 mask, u32 flag)
190 u32 mask_bits = desc->masked;
192 if (!desc->msi_attrib.maskbit)
193 return 0;
195 mask_bits &= ~mask;
196 mask_bits |= flag;
197 pci_write_config_dword(desc->dev, desc->mask_pos, mask_bits);
199 return mask_bits;
202 __weak u32 arch_msi_mask_irq(struct msi_desc *desc, u32 mask, u32 flag)
204 return default_msi_mask_irq(desc, mask, flag);
207 static void msi_mask_irq(struct msi_desc *desc, u32 mask, u32 flag)
209 desc->masked = arch_msi_mask_irq(desc, mask, flag);
213 * This internal function does not flush PCI writes to the device.
214 * All users must ensure that they read from the device before either
215 * assuming that the device state is up to date, or returning out of this
216 * file. This saves a few milliseconds when initialising devices with lots
217 * of MSI-X interrupts.
219 u32 default_msix_mask_irq(struct msi_desc *desc, u32 flag)
221 u32 mask_bits = desc->masked;
222 unsigned offset = desc->msi_attrib.entry_nr * PCI_MSIX_ENTRY_SIZE +
223 PCI_MSIX_ENTRY_VECTOR_CTRL;
224 mask_bits &= ~PCI_MSIX_ENTRY_CTRL_MASKBIT;
225 if (flag)
226 mask_bits |= PCI_MSIX_ENTRY_CTRL_MASKBIT;
227 writel(mask_bits, desc->mask_base + offset);
229 return mask_bits;
232 __weak u32 arch_msix_mask_irq(struct msi_desc *desc, u32 flag)
234 return default_msix_mask_irq(desc, flag);
237 static void msix_mask_irq(struct msi_desc *desc, u32 flag)
239 desc->masked = arch_msix_mask_irq(desc, flag);
242 static void msi_set_mask_bit(struct irq_data *data, u32 flag)
244 struct msi_desc *desc = irq_data_get_msi(data);
246 if (desc->msi_attrib.is_msix) {
247 msix_mask_irq(desc, flag);
248 readl(desc->mask_base); /* Flush write to device */
249 } else {
250 unsigned offset = data->irq - desc->dev->irq;
251 msi_mask_irq(desc, 1 << offset, flag << offset);
255 void mask_msi_irq(struct irq_data *data)
257 msi_set_mask_bit(data, 1);
260 void unmask_msi_irq(struct irq_data *data)
262 msi_set_mask_bit(data, 0);
265 void default_restore_msi_irqs(struct pci_dev *dev)
267 struct msi_desc *entry;
269 list_for_each_entry(entry, &dev->msi_list, list) {
270 default_restore_msi_irq(dev, entry->irq);
274 void __read_msi_msg(struct msi_desc *entry, struct msi_msg *msg)
276 BUG_ON(entry->dev->current_state != PCI_D0);
278 if (entry->msi_attrib.is_msix) {
279 void __iomem *base = entry->mask_base +
280 entry->msi_attrib.entry_nr * PCI_MSIX_ENTRY_SIZE;
282 msg->address_lo = readl(base + PCI_MSIX_ENTRY_LOWER_ADDR);
283 msg->address_hi = readl(base + PCI_MSIX_ENTRY_UPPER_ADDR);
284 msg->data = readl(base + PCI_MSIX_ENTRY_DATA);
285 } else {
286 struct pci_dev *dev = entry->dev;
287 int pos = dev->msi_cap;
288 u16 data;
290 pci_read_config_dword(dev, pos + PCI_MSI_ADDRESS_LO,
291 &msg->address_lo);
292 if (entry->msi_attrib.is_64) {
293 pci_read_config_dword(dev, pos + PCI_MSI_ADDRESS_HI,
294 &msg->address_hi);
295 pci_read_config_word(dev, pos + PCI_MSI_DATA_64, &data);
296 } else {
297 msg->address_hi = 0;
298 pci_read_config_word(dev, pos + PCI_MSI_DATA_32, &data);
300 msg->data = data;
304 void read_msi_msg(unsigned int irq, struct msi_msg *msg)
306 struct msi_desc *entry = irq_get_msi_desc(irq);
308 __read_msi_msg(entry, msg);
311 void __get_cached_msi_msg(struct msi_desc *entry, struct msi_msg *msg)
313 /* Assert that the cache is valid, assuming that
314 * valid messages are not all-zeroes. */
315 BUG_ON(!(entry->msg.address_hi | entry->msg.address_lo |
316 entry->msg.data));
318 *msg = entry->msg;
321 void get_cached_msi_msg(unsigned int irq, struct msi_msg *msg)
323 struct msi_desc *entry = irq_get_msi_desc(irq);
325 __get_cached_msi_msg(entry, msg);
328 void __write_msi_msg(struct msi_desc *entry, struct msi_msg *msg)
330 if (entry->dev->current_state != PCI_D0) {
331 /* Don't touch the hardware now */
332 } else if (entry->msi_attrib.is_msix) {
333 void __iomem *base;
334 base = entry->mask_base +
335 entry->msi_attrib.entry_nr * PCI_MSIX_ENTRY_SIZE;
337 writel(msg->address_lo, base + PCI_MSIX_ENTRY_LOWER_ADDR);
338 writel(msg->address_hi, base + PCI_MSIX_ENTRY_UPPER_ADDR);
339 writel(msg->data, base + PCI_MSIX_ENTRY_DATA);
340 } else {
341 struct pci_dev *dev = entry->dev;
342 int pos = dev->msi_cap;
343 u16 msgctl;
345 pci_read_config_word(dev, pos + PCI_MSI_FLAGS, &msgctl);
346 msgctl &= ~PCI_MSI_FLAGS_QSIZE;
347 msgctl |= entry->msi_attrib.multiple << 4;
348 pci_write_config_word(dev, pos + PCI_MSI_FLAGS, msgctl);
350 pci_write_config_dword(dev, pos + PCI_MSI_ADDRESS_LO,
351 msg->address_lo);
352 if (entry->msi_attrib.is_64) {
353 pci_write_config_dword(dev, pos + PCI_MSI_ADDRESS_HI,
354 msg->address_hi);
355 pci_write_config_word(dev, pos + PCI_MSI_DATA_64,
356 msg->data);
357 } else {
358 pci_write_config_word(dev, pos + PCI_MSI_DATA_32,
359 msg->data);
362 entry->msg = *msg;
365 void write_msi_msg(unsigned int irq, struct msi_msg *msg)
367 struct msi_desc *entry = irq_get_msi_desc(irq);
369 __write_msi_msg(entry, msg);
372 static void free_msi_irqs(struct pci_dev *dev)
374 struct msi_desc *entry, *tmp;
375 struct attribute **msi_attrs;
376 struct device_attribute *dev_attr;
377 int count = 0;
379 list_for_each_entry(entry, &dev->msi_list, list) {
380 int i, nvec;
381 if (!entry->irq)
382 continue;
383 if (entry->nvec_used)
384 nvec = entry->nvec_used;
385 else
386 nvec = 1 << entry->msi_attrib.multiple;
387 for (i = 0; i < nvec; i++)
388 BUG_ON(irq_has_action(entry->irq + i));
391 arch_teardown_msi_irqs(dev);
393 list_for_each_entry_safe(entry, tmp, &dev->msi_list, list) {
394 if (entry->msi_attrib.is_msix) {
395 if (list_is_last(&entry->list, &dev->msi_list))
396 iounmap(entry->mask_base);
400 * Its possible that we get into this path
401 * When populate_msi_sysfs fails, which means the entries
402 * were not registered with sysfs. In that case don't
403 * unregister them.
405 if (entry->kobj.parent) {
406 kobject_del(&entry->kobj);
407 kobject_put(&entry->kobj);
410 list_del(&entry->list);
411 kfree(entry);
414 if (dev->msi_irq_groups) {
415 sysfs_remove_groups(&dev->dev.kobj, dev->msi_irq_groups);
416 msi_attrs = dev->msi_irq_groups[0]->attrs;
417 list_for_each_entry(entry, &dev->msi_list, list) {
418 dev_attr = container_of(msi_attrs[count],
419 struct device_attribute, attr);
420 kfree(dev_attr->attr.name);
421 kfree(dev_attr);
422 ++count;
424 kfree(msi_attrs);
425 kfree(dev->msi_irq_groups[0]);
426 kfree(dev->msi_irq_groups);
427 dev->msi_irq_groups = NULL;
431 static struct msi_desc *alloc_msi_entry(struct pci_dev *dev)
433 struct msi_desc *desc = kzalloc(sizeof(*desc), GFP_KERNEL);
434 if (!desc)
435 return NULL;
437 INIT_LIST_HEAD(&desc->list);
438 desc->dev = dev;
440 return desc;
443 static void pci_intx_for_msi(struct pci_dev *dev, int enable)
445 if (!(dev->dev_flags & PCI_DEV_FLAGS_MSI_INTX_DISABLE_BUG))
446 pci_intx(dev, enable);
449 static void __pci_restore_msi_state(struct pci_dev *dev)
451 u16 control;
452 struct msi_desc *entry;
454 if (!dev->msi_enabled)
455 return;
457 entry = irq_get_msi_desc(dev->irq);
459 pci_intx_for_msi(dev, 0);
460 msi_set_enable(dev, 0);
461 arch_restore_msi_irqs(dev);
463 pci_read_config_word(dev, dev->msi_cap + PCI_MSI_FLAGS, &control);
464 msi_mask_irq(entry, msi_capable_mask(control), entry->masked);
465 control &= ~PCI_MSI_FLAGS_QSIZE;
466 control |= (entry->msi_attrib.multiple << 4) | PCI_MSI_FLAGS_ENABLE;
467 pci_write_config_word(dev, dev->msi_cap + PCI_MSI_FLAGS, control);
470 static void __pci_restore_msix_state(struct pci_dev *dev)
472 struct msi_desc *entry;
473 u16 control;
475 if (!dev->msix_enabled)
476 return;
477 BUG_ON(list_empty(&dev->msi_list));
478 entry = list_first_entry(&dev->msi_list, struct msi_desc, list);
479 pci_read_config_word(dev, dev->msix_cap + PCI_MSIX_FLAGS, &control);
481 /* route the table */
482 pci_intx_for_msi(dev, 0);
483 control |= PCI_MSIX_FLAGS_ENABLE | PCI_MSIX_FLAGS_MASKALL;
484 pci_write_config_word(dev, dev->msix_cap + PCI_MSIX_FLAGS, control);
486 arch_restore_msi_irqs(dev);
487 list_for_each_entry(entry, &dev->msi_list, list) {
488 msix_mask_irq(entry, entry->masked);
491 control &= ~PCI_MSIX_FLAGS_MASKALL;
492 pci_write_config_word(dev, dev->msix_cap + PCI_MSIX_FLAGS, control);
495 void pci_restore_msi_state(struct pci_dev *dev)
497 __pci_restore_msi_state(dev);
498 __pci_restore_msix_state(dev);
500 EXPORT_SYMBOL_GPL(pci_restore_msi_state);
502 static ssize_t msi_mode_show(struct device *dev, struct device_attribute *attr,
503 char *buf)
505 struct pci_dev *pdev = to_pci_dev(dev);
506 struct msi_desc *entry;
507 unsigned long irq;
508 int retval;
510 retval = kstrtoul(attr->attr.name, 10, &irq);
511 if (retval)
512 return retval;
514 list_for_each_entry(entry, &pdev->msi_list, list) {
515 if (entry->irq == irq) {
516 return sprintf(buf, "%s\n",
517 entry->msi_attrib.is_msix ? "msix" : "msi");
520 return -ENODEV;
523 static int populate_msi_sysfs(struct pci_dev *pdev)
525 struct attribute **msi_attrs;
526 struct attribute *msi_attr;
527 struct device_attribute *msi_dev_attr;
528 struct attribute_group *msi_irq_group;
529 const struct attribute_group **msi_irq_groups;
530 struct msi_desc *entry;
531 int ret = -ENOMEM;
532 int num_msi = 0;
533 int count = 0;
535 /* Determine how many msi entries we have */
536 list_for_each_entry(entry, &pdev->msi_list, list) {
537 ++num_msi;
539 if (!num_msi)
540 return 0;
542 /* Dynamically create the MSI attributes for the PCI device */
543 msi_attrs = kzalloc(sizeof(void *) * (num_msi + 1), GFP_KERNEL);
544 if (!msi_attrs)
545 return -ENOMEM;
546 list_for_each_entry(entry, &pdev->msi_list, list) {
547 char *name = kmalloc(20, GFP_KERNEL);
548 msi_dev_attr = kzalloc(sizeof(*msi_dev_attr), GFP_KERNEL);
549 if (!msi_dev_attr)
550 goto error_attrs;
551 sprintf(name, "%d", entry->irq);
552 sysfs_attr_init(&msi_dev_attr->attr);
553 msi_dev_attr->attr.name = name;
554 msi_dev_attr->attr.mode = S_IRUGO;
555 msi_dev_attr->show = msi_mode_show;
556 msi_attrs[count] = &msi_dev_attr->attr;
557 ++count;
560 msi_irq_group = kzalloc(sizeof(*msi_irq_group), GFP_KERNEL);
561 if (!msi_irq_group)
562 goto error_attrs;
563 msi_irq_group->name = "msi_irqs";
564 msi_irq_group->attrs = msi_attrs;
566 msi_irq_groups = kzalloc(sizeof(void *) * 2, GFP_KERNEL);
567 if (!msi_irq_groups)
568 goto error_irq_group;
569 msi_irq_groups[0] = msi_irq_group;
571 ret = sysfs_create_groups(&pdev->dev.kobj, msi_irq_groups);
572 if (ret)
573 goto error_irq_groups;
574 pdev->msi_irq_groups = msi_irq_groups;
576 return 0;
578 error_irq_groups:
579 kfree(msi_irq_groups);
580 error_irq_group:
581 kfree(msi_irq_group);
582 error_attrs:
583 count = 0;
584 msi_attr = msi_attrs[count];
585 while (msi_attr) {
586 msi_dev_attr = container_of(msi_attr, struct device_attribute, attr);
587 kfree(msi_attr->name);
588 kfree(msi_dev_attr);
589 ++count;
590 msi_attr = msi_attrs[count];
592 return ret;
596 * msi_capability_init - configure device's MSI capability structure
597 * @dev: pointer to the pci_dev data structure of MSI device function
598 * @nvec: number of interrupts to allocate
600 * Setup the MSI capability structure of the device with the requested
601 * number of interrupts. A return value of zero indicates the successful
602 * setup of an entry with the new MSI irq. A negative return value indicates
603 * an error, and a positive return value indicates the number of interrupts
604 * which could have been allocated.
606 static int msi_capability_init(struct pci_dev *dev, int nvec)
608 struct msi_desc *entry;
609 int ret;
610 u16 control;
611 unsigned mask;
613 msi_set_enable(dev, 0); /* Disable MSI during set up */
615 pci_read_config_word(dev, dev->msi_cap + PCI_MSI_FLAGS, &control);
616 /* MSI Entry Initialization */
617 entry = alloc_msi_entry(dev);
618 if (!entry)
619 return -ENOMEM;
621 entry->msi_attrib.is_msix = 0;
622 entry->msi_attrib.is_64 = !!(control & PCI_MSI_FLAGS_64BIT);
623 entry->msi_attrib.entry_nr = 0;
624 entry->msi_attrib.maskbit = !!(control & PCI_MSI_FLAGS_MASKBIT);
625 entry->msi_attrib.default_irq = dev->irq; /* Save IOAPIC IRQ */
626 entry->msi_attrib.pos = dev->msi_cap;
628 if (control & PCI_MSI_FLAGS_64BIT)
629 entry->mask_pos = dev->msi_cap + PCI_MSI_MASK_64;
630 else
631 entry->mask_pos = dev->msi_cap + PCI_MSI_MASK_32;
632 /* All MSIs are unmasked by default, Mask them all */
633 if (entry->msi_attrib.maskbit)
634 pci_read_config_dword(dev, entry->mask_pos, &entry->masked);
635 mask = msi_capable_mask(control);
636 msi_mask_irq(entry, mask, mask);
638 list_add_tail(&entry->list, &dev->msi_list);
640 /* Configure MSI capability structure */
641 ret = arch_setup_msi_irqs(dev, nvec, PCI_CAP_ID_MSI);
642 if (ret) {
643 msi_mask_irq(entry, mask, ~mask);
644 free_msi_irqs(dev);
645 return ret;
648 ret = populate_msi_sysfs(dev);
649 if (ret) {
650 msi_mask_irq(entry, mask, ~mask);
651 free_msi_irqs(dev);
652 return ret;
655 /* Set MSI enabled bits */
656 pci_intx_for_msi(dev, 0);
657 msi_set_enable(dev, 1);
658 dev->msi_enabled = 1;
660 dev->irq = entry->irq;
661 return 0;
664 static void __iomem *msix_map_region(struct pci_dev *dev, unsigned nr_entries)
666 resource_size_t phys_addr;
667 u32 table_offset;
668 u8 bir;
670 pci_read_config_dword(dev, dev->msix_cap + PCI_MSIX_TABLE,
671 &table_offset);
672 bir = (u8)(table_offset & PCI_MSIX_TABLE_BIR);
673 table_offset &= PCI_MSIX_TABLE_OFFSET;
674 phys_addr = pci_resource_start(dev, bir) + table_offset;
676 return ioremap_nocache(phys_addr, nr_entries * PCI_MSIX_ENTRY_SIZE);
679 static int msix_setup_entries(struct pci_dev *dev, void __iomem *base,
680 struct msix_entry *entries, int nvec)
682 struct msi_desc *entry;
683 int i;
685 for (i = 0; i < nvec; i++) {
686 entry = alloc_msi_entry(dev);
687 if (!entry) {
688 if (!i)
689 iounmap(base);
690 else
691 free_msi_irqs(dev);
692 /* No enough memory. Don't try again */
693 return -ENOMEM;
696 entry->msi_attrib.is_msix = 1;
697 entry->msi_attrib.is_64 = 1;
698 entry->msi_attrib.entry_nr = entries[i].entry;
699 entry->msi_attrib.default_irq = dev->irq;
700 entry->msi_attrib.pos = dev->msix_cap;
701 entry->mask_base = base;
703 list_add_tail(&entry->list, &dev->msi_list);
706 return 0;
709 static void msix_program_entries(struct pci_dev *dev,
710 struct msix_entry *entries)
712 struct msi_desc *entry;
713 int i = 0;
715 list_for_each_entry(entry, &dev->msi_list, list) {
716 int offset = entries[i].entry * PCI_MSIX_ENTRY_SIZE +
717 PCI_MSIX_ENTRY_VECTOR_CTRL;
719 entries[i].vector = entry->irq;
720 irq_set_msi_desc(entry->irq, entry);
721 entry->masked = readl(entry->mask_base + offset);
722 msix_mask_irq(entry, 1);
723 i++;
728 * msix_capability_init - configure device's MSI-X capability
729 * @dev: pointer to the pci_dev data structure of MSI-X device function
730 * @entries: pointer to an array of struct msix_entry entries
731 * @nvec: number of @entries
733 * Setup the MSI-X capability structure of device function with a
734 * single MSI-X irq. A return of zero indicates the successful setup of
735 * requested MSI-X entries with allocated irqs or non-zero for otherwise.
737 static int msix_capability_init(struct pci_dev *dev,
738 struct msix_entry *entries, int nvec)
740 int ret;
741 u16 control;
742 void __iomem *base;
744 pci_read_config_word(dev, dev->msix_cap + PCI_MSIX_FLAGS, &control);
746 /* Ensure MSI-X is disabled while it is set up */
747 control &= ~PCI_MSIX_FLAGS_ENABLE;
748 pci_write_config_word(dev, dev->msix_cap + PCI_MSIX_FLAGS, control);
750 /* Request & Map MSI-X table region */
751 base = msix_map_region(dev, msix_table_size(control));
752 if (!base)
753 return -ENOMEM;
755 ret = msix_setup_entries(dev, base, entries, nvec);
756 if (ret)
757 return ret;
759 ret = arch_setup_msi_irqs(dev, nvec, PCI_CAP_ID_MSIX);
760 if (ret)
761 goto out_avail;
764 * Some devices require MSI-X to be enabled before we can touch the
765 * MSI-X registers. We need to mask all the vectors to prevent
766 * interrupts coming in before they're fully set up.
768 control |= PCI_MSIX_FLAGS_MASKALL | PCI_MSIX_FLAGS_ENABLE;
769 pci_write_config_word(dev, dev->msix_cap + PCI_MSIX_FLAGS, control);
771 msix_program_entries(dev, entries);
773 ret = populate_msi_sysfs(dev);
774 if (ret)
775 goto out_free;
777 /* Set MSI-X enabled bits and unmask the function */
778 pci_intx_for_msi(dev, 0);
779 dev->msix_enabled = 1;
781 control &= ~PCI_MSIX_FLAGS_MASKALL;
782 pci_write_config_word(dev, dev->msix_cap + PCI_MSIX_FLAGS, control);
784 return 0;
786 out_avail:
787 if (ret < 0) {
789 * If we had some success, report the number of irqs
790 * we succeeded in setting up.
792 struct msi_desc *entry;
793 int avail = 0;
795 list_for_each_entry(entry, &dev->msi_list, list) {
796 if (entry->irq != 0)
797 avail++;
799 if (avail != 0)
800 ret = avail;
803 out_free:
804 free_msi_irqs(dev);
806 return ret;
810 * pci_msi_check_device - check whether MSI may be enabled on a device
811 * @dev: pointer to the pci_dev data structure of MSI device function
812 * @nvec: how many MSIs have been requested ?
813 * @type: are we checking for MSI or MSI-X ?
815 * Look at global flags, the device itself, and its parent buses
816 * to determine if MSI/-X are supported for the device. If MSI/-X is
817 * supported return 0, else return an error code.
819 static int pci_msi_check_device(struct pci_dev *dev, int nvec, int type)
821 struct pci_bus *bus;
822 int ret;
824 /* MSI must be globally enabled and supported by the device */
825 if (!pci_msi_enable || !dev || dev->no_msi)
826 return -EINVAL;
829 * You can't ask to have 0 or less MSIs configured.
830 * a) it's stupid ..
831 * b) the list manipulation code assumes nvec >= 1.
833 if (nvec < 1)
834 return -ERANGE;
837 * Any bridge which does NOT route MSI transactions from its
838 * secondary bus to its primary bus must set NO_MSI flag on
839 * the secondary pci_bus.
840 * We expect only arch-specific PCI host bus controller driver
841 * or quirks for specific PCI bridges to be setting NO_MSI.
843 for (bus = dev->bus; bus; bus = bus->parent)
844 if (bus->bus_flags & PCI_BUS_FLAGS_NO_MSI)
845 return -EINVAL;
847 ret = arch_msi_check_device(dev, nvec, type);
848 if (ret)
849 return ret;
851 return 0;
855 * pci_msi_vec_count - Return the number of MSI vectors a device can send
856 * @dev: device to report about
858 * This function returns the number of MSI vectors a device requested via
859 * Multiple Message Capable register. It returns a negative errno if the
860 * device is not capable sending MSI interrupts. Otherwise, the call succeeds
861 * and returns a power of two, up to a maximum of 2^5 (32), according to the
862 * MSI specification.
864 int pci_msi_vec_count(struct pci_dev *dev)
866 int ret;
867 u16 msgctl;
869 if (!dev->msi_cap)
870 return -EINVAL;
872 pci_read_config_word(dev, dev->msi_cap + PCI_MSI_FLAGS, &msgctl);
873 ret = 1 << ((msgctl & PCI_MSI_FLAGS_QMASK) >> 1);
875 return ret;
877 EXPORT_SYMBOL(pci_msi_vec_count);
880 * pci_enable_msi_block - configure device's MSI capability structure
881 * @dev: device to configure
882 * @nvec: number of interrupts to configure
884 * Allocate IRQs for a device with the MSI capability.
885 * This function returns a negative errno if an error occurs. If it
886 * is unable to allocate the number of interrupts requested, it returns
887 * the number of interrupts it might be able to allocate. If it successfully
888 * allocates at least the number of interrupts requested, it returns 0 and
889 * updates the @dev's irq member to the lowest new interrupt number; the
890 * other interrupt numbers allocated to this device are consecutive.
892 int pci_enable_msi_block(struct pci_dev *dev, int nvec)
894 int status, maxvec;
896 if (dev->current_state != PCI_D0)
897 return -EINVAL;
899 maxvec = pci_msi_vec_count(dev);
900 if (maxvec < 0)
901 return maxvec;
902 if (nvec > maxvec)
903 return maxvec;
905 status = pci_msi_check_device(dev, nvec, PCI_CAP_ID_MSI);
906 if (status)
907 return status;
909 WARN_ON(!!dev->msi_enabled);
911 /* Check whether driver already requested MSI-X irqs */
912 if (dev->msix_enabled) {
913 dev_info(&dev->dev, "can't enable MSI "
914 "(MSI-X already enabled)\n");
915 return -EINVAL;
918 status = msi_capability_init(dev, nvec);
919 return status;
921 EXPORT_SYMBOL(pci_enable_msi_block);
923 void pci_msi_shutdown(struct pci_dev *dev)
925 struct msi_desc *desc;
926 u32 mask;
927 u16 ctrl;
929 if (!pci_msi_enable || !dev || !dev->msi_enabled)
930 return;
932 BUG_ON(list_empty(&dev->msi_list));
933 desc = list_first_entry(&dev->msi_list, struct msi_desc, list);
935 msi_set_enable(dev, 0);
936 pci_intx_for_msi(dev, 1);
937 dev->msi_enabled = 0;
939 /* Return the device with MSI unmasked as initial states */
940 pci_read_config_word(dev, dev->msi_cap + PCI_MSI_FLAGS, &ctrl);
941 mask = msi_capable_mask(ctrl);
942 /* Keep cached state to be restored */
943 arch_msi_mask_irq(desc, mask, ~mask);
945 /* Restore dev->irq to its default pin-assertion irq */
946 dev->irq = desc->msi_attrib.default_irq;
949 void pci_disable_msi(struct pci_dev *dev)
951 if (!pci_msi_enable || !dev || !dev->msi_enabled)
952 return;
954 pci_msi_shutdown(dev);
955 free_msi_irqs(dev);
957 EXPORT_SYMBOL(pci_disable_msi);
960 * pci_msix_vec_count - return the number of device's MSI-X table entries
961 * @dev: pointer to the pci_dev data structure of MSI-X device function
963 * This function returns the number of device's MSI-X table entries and
964 * therefore the number of MSI-X vectors device is capable of sending.
965 * It returns a negative errno if the device is not capable of sending MSI-X
966 * interrupts.
968 int pci_msix_vec_count(struct pci_dev *dev)
970 u16 control;
972 if (!dev->msix_cap)
973 return -EINVAL;
975 pci_read_config_word(dev, dev->msix_cap + PCI_MSIX_FLAGS, &control);
976 return msix_table_size(control);
978 EXPORT_SYMBOL(pci_msix_vec_count);
981 * pci_enable_msix - configure device's MSI-X capability structure
982 * @dev: pointer to the pci_dev data structure of MSI-X device function
983 * @entries: pointer to an array of MSI-X entries
984 * @nvec: number of MSI-X irqs requested for allocation by device driver
986 * Setup the MSI-X capability structure of device function with the number
987 * of requested irqs upon its software driver call to request for
988 * MSI-X mode enabled on its hardware device function. A return of zero
989 * indicates the successful configuration of MSI-X capability structure
990 * with new allocated MSI-X irqs. A return of < 0 indicates a failure.
991 * Or a return of > 0 indicates that driver request is exceeding the number
992 * of irqs or MSI-X vectors available. Driver should use the returned value to
993 * re-send its request.
995 int pci_enable_msix(struct pci_dev *dev, struct msix_entry *entries, int nvec)
997 int status, nr_entries;
998 int i, j;
1000 if (!entries || !dev->msix_cap || dev->current_state != PCI_D0)
1001 return -EINVAL;
1003 status = pci_msi_check_device(dev, nvec, PCI_CAP_ID_MSIX);
1004 if (status)
1005 return status;
1007 nr_entries = pci_msix_vec_count(dev);
1008 if (nr_entries < 0)
1009 return nr_entries;
1010 if (nvec > nr_entries)
1011 return nr_entries;
1013 /* Check for any invalid entries */
1014 for (i = 0; i < nvec; i++) {
1015 if (entries[i].entry >= nr_entries)
1016 return -EINVAL; /* invalid entry */
1017 for (j = i + 1; j < nvec; j++) {
1018 if (entries[i].entry == entries[j].entry)
1019 return -EINVAL; /* duplicate entry */
1022 WARN_ON(!!dev->msix_enabled);
1024 /* Check whether driver already requested for MSI irq */
1025 if (dev->msi_enabled) {
1026 dev_info(&dev->dev, "can't enable MSI-X "
1027 "(MSI IRQ already assigned)\n");
1028 return -EINVAL;
1030 status = msix_capability_init(dev, entries, nvec);
1031 return status;
1033 EXPORT_SYMBOL(pci_enable_msix);
1035 void pci_msix_shutdown(struct pci_dev *dev)
1037 struct msi_desc *entry;
1039 if (!pci_msi_enable || !dev || !dev->msix_enabled)
1040 return;
1042 /* Return the device with MSI-X masked as initial states */
1043 list_for_each_entry(entry, &dev->msi_list, list) {
1044 /* Keep cached states to be restored */
1045 arch_msix_mask_irq(entry, 1);
1048 msix_set_enable(dev, 0);
1049 pci_intx_for_msi(dev, 1);
1050 dev->msix_enabled = 0;
1053 void pci_disable_msix(struct pci_dev *dev)
1055 if (!pci_msi_enable || !dev || !dev->msix_enabled)
1056 return;
1058 pci_msix_shutdown(dev);
1059 free_msi_irqs(dev);
1061 EXPORT_SYMBOL(pci_disable_msix);
1064 * msi_remove_pci_irq_vectors - reclaim MSI(X) irqs to unused state
1065 * @dev: pointer to the pci_dev data structure of MSI(X) device function
1067 * Being called during hotplug remove, from which the device function
1068 * is hot-removed. All previous assigned MSI/MSI-X irqs, if
1069 * allocated for this device function, are reclaimed to unused state,
1070 * which may be used later on.
1072 void msi_remove_pci_irq_vectors(struct pci_dev *dev)
1074 if (!pci_msi_enable || !dev)
1075 return;
1077 if (dev->msi_enabled || dev->msix_enabled)
1078 free_msi_irqs(dev);
1081 void pci_no_msi(void)
1083 pci_msi_enable = 0;
1087 * pci_msi_enabled - is MSI enabled?
1089 * Returns true if MSI has not been disabled by the command-line option
1090 * pci=nomsi.
1092 int pci_msi_enabled(void)
1094 return pci_msi_enable;
1096 EXPORT_SYMBOL(pci_msi_enabled);
1098 void pci_msi_init_pci_dev(struct pci_dev *dev)
1100 INIT_LIST_HEAD(&dev->msi_list);
1102 /* Disable the msi hardware to avoid screaming interrupts
1103 * during boot. This is the power on reset default so
1104 * usually this should be a noop.
1106 dev->msi_cap = pci_find_capability(dev, PCI_CAP_ID_MSI);
1107 if (dev->msi_cap)
1108 msi_set_enable(dev, 0);
1110 dev->msix_cap = pci_find_capability(dev, PCI_CAP_ID_MSIX);
1111 if (dev->msix_cap)
1112 msix_set_enable(dev, 0);
1116 * pci_enable_msi_range - configure device's MSI capability structure
1117 * @dev: device to configure
1118 * @minvec: minimal number of interrupts to configure
1119 * @maxvec: maximum number of interrupts to configure
1121 * This function tries to allocate a maximum possible number of interrupts in a
1122 * range between @minvec and @maxvec. It returns a negative errno if an error
1123 * occurs. If it succeeds, it returns the actual number of interrupts allocated
1124 * and updates the @dev's irq member to the lowest new interrupt number;
1125 * the other interrupt numbers allocated to this device are consecutive.
1127 int pci_enable_msi_range(struct pci_dev *dev, int minvec, int maxvec)
1129 int nvec = maxvec;
1130 int rc;
1132 if (maxvec < minvec)
1133 return -ERANGE;
1135 do {
1136 rc = pci_enable_msi_block(dev, nvec);
1137 if (rc < 0) {
1138 return rc;
1139 } else if (rc > 0) {
1140 if (rc < minvec)
1141 return -ENOSPC;
1142 nvec = rc;
1144 } while (rc);
1146 return nvec;
1148 EXPORT_SYMBOL(pci_enable_msi_range);
1151 * pci_enable_msix_range - configure device's MSI-X capability structure
1152 * @dev: pointer to the pci_dev data structure of MSI-X device function
1153 * @entries: pointer to an array of MSI-X entries
1154 * @minvec: minimum number of MSI-X irqs requested
1155 * @maxvec: maximum number of MSI-X irqs requested
1157 * Setup the MSI-X capability structure of device function with a maximum
1158 * possible number of interrupts in the range between @minvec and @maxvec
1159 * upon its software driver call to request for MSI-X mode enabled on its
1160 * hardware device function. It returns a negative errno if an error occurs.
1161 * If it succeeds, it returns the actual number of interrupts allocated and
1162 * indicates the successful configuration of MSI-X capability structure
1163 * with new allocated MSI-X interrupts.
1165 int pci_enable_msix_range(struct pci_dev *dev, struct msix_entry *entries,
1166 int minvec, int maxvec)
1168 int nvec = maxvec;
1169 int rc;
1171 if (maxvec < minvec)
1172 return -ERANGE;
1174 do {
1175 rc = pci_enable_msix(dev, entries, nvec);
1176 if (rc < 0) {
1177 return rc;
1178 } else if (rc > 0) {
1179 if (rc < minvec)
1180 return -ENOSPC;
1181 nvec = rc;
1183 } while (rc);
1185 return nvec;
1187 EXPORT_SYMBOL(pci_enable_msix_range);