Linux kernel mirror (for testing)
git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git
kernel
os
linux
1/*
2 * File: msi.c
3 * Purpose: PCI Message Signaled Interrupt (MSI)
4 *
5 * Copyright (C) 2003-2004 Intel
6 * Copyright (C) Tom Long Nguyen (tom.l.nguyen@intel.com)
7 * Copyright (C) 2016 Christoph Hellwig.
8 */
9
10#include <linux/err.h>
11#include <linux/mm.h>
12#include <linux/irq.h>
13#include <linux/interrupt.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/acpi_iort.h>
23#include <linux/slab.h>
24#include <linux/irqdomain.h>
25#include <linux/of_irq.h>
26
27#include "pci.h"
28
29static int pci_msi_enable = 1;
30int pci_msi_ignore_mask;
31
32#define msix_table_size(flags) ((flags & PCI_MSIX_FLAGS_QSIZE) + 1)
33
34#ifdef CONFIG_PCI_MSI_IRQ_DOMAIN
35static int pci_msi_setup_msi_irqs(struct pci_dev *dev, int nvec, int type)
36{
37 struct irq_domain *domain;
38
39 domain = dev_get_msi_domain(&dev->dev);
40 if (domain && irq_domain_is_hierarchy(domain))
41 return msi_domain_alloc_irqs(domain, &dev->dev, nvec);
42
43 return arch_setup_msi_irqs(dev, nvec, type);
44}
45
46static void pci_msi_teardown_msi_irqs(struct pci_dev *dev)
47{
48 struct irq_domain *domain;
49
50 domain = dev_get_msi_domain(&dev->dev);
51 if (domain && irq_domain_is_hierarchy(domain))
52 msi_domain_free_irqs(domain, &dev->dev);
53 else
54 arch_teardown_msi_irqs(dev);
55}
56#else
57#define pci_msi_setup_msi_irqs arch_setup_msi_irqs
58#define pci_msi_teardown_msi_irqs arch_teardown_msi_irqs
59#endif
60
61/* Arch hooks */
62
63int __weak arch_setup_msi_irq(struct pci_dev *dev, struct msi_desc *desc)
64{
65 struct msi_controller *chip = dev->bus->msi;
66 int err;
67
68 if (!chip || !chip->setup_irq)
69 return -EINVAL;
70
71 err = chip->setup_irq(chip, dev, desc);
72 if (err < 0)
73 return err;
74
75 irq_set_chip_data(desc->irq, chip);
76
77 return 0;
78}
79
80void __weak arch_teardown_msi_irq(unsigned int irq)
81{
82 struct msi_controller *chip = irq_get_chip_data(irq);
83
84 if (!chip || !chip->teardown_irq)
85 return;
86
87 chip->teardown_irq(chip, irq);
88}
89
90int __weak arch_setup_msi_irqs(struct pci_dev *dev, int nvec, int type)
91{
92 struct msi_controller *chip = dev->bus->msi;
93 struct msi_desc *entry;
94 int ret;
95
96 if (chip && chip->setup_irqs)
97 return chip->setup_irqs(chip, dev, nvec, type);
98 /*
99 * If an architecture wants to support multiple MSI, it needs to
100 * override arch_setup_msi_irqs()
101 */
102 if (type == PCI_CAP_ID_MSI && nvec > 1)
103 return 1;
104
105 for_each_pci_msi_entry(entry, dev) {
106 ret = arch_setup_msi_irq(dev, entry);
107 if (ret < 0)
108 return ret;
109 if (ret > 0)
110 return -ENOSPC;
111 }
112
113 return 0;
114}
115
116/*
117 * We have a default implementation available as a separate non-weak
118 * function, as it is used by the Xen x86 PCI code
119 */
120void default_teardown_msi_irqs(struct pci_dev *dev)
121{
122 int i;
123 struct msi_desc *entry;
124
125 for_each_pci_msi_entry(entry, dev)
126 if (entry->irq)
127 for (i = 0; i < entry->nvec_used; i++)
128 arch_teardown_msi_irq(entry->irq + i);
129}
130
131void __weak arch_teardown_msi_irqs(struct pci_dev *dev)
132{
133 return default_teardown_msi_irqs(dev);
134}
135
136static void default_restore_msi_irq(struct pci_dev *dev, int irq)
137{
138 struct msi_desc *entry;
139
140 entry = NULL;
141 if (dev->msix_enabled) {
142 for_each_pci_msi_entry(entry, dev) {
143 if (irq == entry->irq)
144 break;
145 }
146 } else if (dev->msi_enabled) {
147 entry = irq_get_msi_desc(irq);
148 }
149
150 if (entry)
151 __pci_write_msi_msg(entry, &entry->msg);
152}
153
154void __weak arch_restore_msi_irqs(struct pci_dev *dev)
155{
156 return default_restore_msi_irqs(dev);
157}
158
159static inline __attribute_const__ u32 msi_mask(unsigned x)
160{
161 /* Don't shift by >= width of type */
162 if (x >= 5)
163 return 0xffffffff;
164 return (1 << (1 << x)) - 1;
165}
166
167/*
168 * PCI 2.3 does not specify mask bits for each MSI interrupt. Attempting to
169 * mask all MSI interrupts by clearing the MSI enable bit does not work
170 * reliably as devices without an INTx disable bit will then generate a
171 * level IRQ which will never be cleared.
172 */
173u32 __pci_msi_desc_mask_irq(struct msi_desc *desc, u32 mask, u32 flag)
174{
175 u32 mask_bits = desc->masked;
176
177 if (pci_msi_ignore_mask || !desc->msi_attrib.maskbit)
178 return 0;
179
180 mask_bits &= ~mask;
181 mask_bits |= flag;
182 pci_write_config_dword(msi_desc_to_pci_dev(desc), desc->mask_pos,
183 mask_bits);
184
185 return mask_bits;
186}
187
188static void msi_mask_irq(struct msi_desc *desc, u32 mask, u32 flag)
189{
190 desc->masked = __pci_msi_desc_mask_irq(desc, mask, flag);
191}
192
193static void __iomem *pci_msix_desc_addr(struct msi_desc *desc)
194{
195 return desc->mask_base +
196 desc->msi_attrib.entry_nr * PCI_MSIX_ENTRY_SIZE;
197}
198
199/*
200 * This internal function does not flush PCI writes to the device.
201 * All users must ensure that they read from the device before either
202 * assuming that the device state is up to date, or returning out of this
203 * file. This saves a few milliseconds when initialising devices with lots
204 * of MSI-X interrupts.
205 */
206u32 __pci_msix_desc_mask_irq(struct msi_desc *desc, u32 flag)
207{
208 u32 mask_bits = desc->masked;
209
210 if (pci_msi_ignore_mask)
211 return 0;
212
213 mask_bits &= ~PCI_MSIX_ENTRY_CTRL_MASKBIT;
214 if (flag)
215 mask_bits |= PCI_MSIX_ENTRY_CTRL_MASKBIT;
216 writel(mask_bits, pci_msix_desc_addr(desc) + PCI_MSIX_ENTRY_VECTOR_CTRL);
217
218 return mask_bits;
219}
220
221static void msix_mask_irq(struct msi_desc *desc, u32 flag)
222{
223 desc->masked = __pci_msix_desc_mask_irq(desc, flag);
224}
225
226static void msi_set_mask_bit(struct irq_data *data, u32 flag)
227{
228 struct msi_desc *desc = irq_data_get_msi_desc(data);
229
230 if (desc->msi_attrib.is_msix) {
231 msix_mask_irq(desc, flag);
232 readl(desc->mask_base); /* Flush write to device */
233 } else {
234 unsigned offset = data->irq - desc->irq;
235 msi_mask_irq(desc, 1 << offset, flag << offset);
236 }
237}
238
239/**
240 * pci_msi_mask_irq - Generic irq chip callback to mask PCI/MSI interrupts
241 * @data: pointer to irqdata associated to that interrupt
242 */
243void pci_msi_mask_irq(struct irq_data *data)
244{
245 msi_set_mask_bit(data, 1);
246}
247EXPORT_SYMBOL_GPL(pci_msi_mask_irq);
248
249/**
250 * pci_msi_unmask_irq - Generic irq chip callback to unmask PCI/MSI interrupts
251 * @data: pointer to irqdata associated to that interrupt
252 */
253void pci_msi_unmask_irq(struct irq_data *data)
254{
255 msi_set_mask_bit(data, 0);
256}
257EXPORT_SYMBOL_GPL(pci_msi_unmask_irq);
258
259void default_restore_msi_irqs(struct pci_dev *dev)
260{
261 struct msi_desc *entry;
262
263 for_each_pci_msi_entry(entry, dev)
264 default_restore_msi_irq(dev, entry->irq);
265}
266
267void __pci_read_msi_msg(struct msi_desc *entry, struct msi_msg *msg)
268{
269 struct pci_dev *dev = msi_desc_to_pci_dev(entry);
270
271 BUG_ON(dev->current_state != PCI_D0);
272
273 if (entry->msi_attrib.is_msix) {
274 void __iomem *base = pci_msix_desc_addr(entry);
275
276 msg->address_lo = readl(base + PCI_MSIX_ENTRY_LOWER_ADDR);
277 msg->address_hi = readl(base + PCI_MSIX_ENTRY_UPPER_ADDR);
278 msg->data = readl(base + PCI_MSIX_ENTRY_DATA);
279 } else {
280 int pos = dev->msi_cap;
281 u16 data;
282
283 pci_read_config_dword(dev, pos + PCI_MSI_ADDRESS_LO,
284 &msg->address_lo);
285 if (entry->msi_attrib.is_64) {
286 pci_read_config_dword(dev, pos + PCI_MSI_ADDRESS_HI,
287 &msg->address_hi);
288 pci_read_config_word(dev, pos + PCI_MSI_DATA_64, &data);
289 } else {
290 msg->address_hi = 0;
291 pci_read_config_word(dev, pos + PCI_MSI_DATA_32, &data);
292 }
293 msg->data = data;
294 }
295}
296
297void __pci_write_msi_msg(struct msi_desc *entry, struct msi_msg *msg)
298{
299 struct pci_dev *dev = msi_desc_to_pci_dev(entry);
300
301 if (dev->current_state != PCI_D0) {
302 /* Don't touch the hardware now */
303 } else if (entry->msi_attrib.is_msix) {
304 void __iomem *base = pci_msix_desc_addr(entry);
305
306 writel(msg->address_lo, base + PCI_MSIX_ENTRY_LOWER_ADDR);
307 writel(msg->address_hi, base + PCI_MSIX_ENTRY_UPPER_ADDR);
308 writel(msg->data, base + PCI_MSIX_ENTRY_DATA);
309 } else {
310 int pos = dev->msi_cap;
311 u16 msgctl;
312
313 pci_read_config_word(dev, pos + PCI_MSI_FLAGS, &msgctl);
314 msgctl &= ~PCI_MSI_FLAGS_QSIZE;
315 msgctl |= entry->msi_attrib.multiple << 4;
316 pci_write_config_word(dev, pos + PCI_MSI_FLAGS, msgctl);
317
318 pci_write_config_dword(dev, pos + PCI_MSI_ADDRESS_LO,
319 msg->address_lo);
320 if (entry->msi_attrib.is_64) {
321 pci_write_config_dword(dev, pos + PCI_MSI_ADDRESS_HI,
322 msg->address_hi);
323 pci_write_config_word(dev, pos + PCI_MSI_DATA_64,
324 msg->data);
325 } else {
326 pci_write_config_word(dev, pos + PCI_MSI_DATA_32,
327 msg->data);
328 }
329 }
330 entry->msg = *msg;
331}
332
333void pci_write_msi_msg(unsigned int irq, struct msi_msg *msg)
334{
335 struct msi_desc *entry = irq_get_msi_desc(irq);
336
337 __pci_write_msi_msg(entry, msg);
338}
339EXPORT_SYMBOL_GPL(pci_write_msi_msg);
340
341static void free_msi_irqs(struct pci_dev *dev)
342{
343 struct list_head *msi_list = dev_to_msi_list(&dev->dev);
344 struct msi_desc *entry, *tmp;
345 struct attribute **msi_attrs;
346 struct device_attribute *dev_attr;
347 int i, count = 0;
348
349 for_each_pci_msi_entry(entry, dev)
350 if (entry->irq)
351 for (i = 0; i < entry->nvec_used; i++)
352 BUG_ON(irq_has_action(entry->irq + i));
353
354 pci_msi_teardown_msi_irqs(dev);
355
356 list_for_each_entry_safe(entry, tmp, msi_list, list) {
357 if (entry->msi_attrib.is_msix) {
358 if (list_is_last(&entry->list, msi_list))
359 iounmap(entry->mask_base);
360 }
361
362 list_del(&entry->list);
363 free_msi_entry(entry);
364 }
365
366 if (dev->msi_irq_groups) {
367 sysfs_remove_groups(&dev->dev.kobj, dev->msi_irq_groups);
368 msi_attrs = dev->msi_irq_groups[0]->attrs;
369 while (msi_attrs[count]) {
370 dev_attr = container_of(msi_attrs[count],
371 struct device_attribute, attr);
372 kfree(dev_attr->attr.name);
373 kfree(dev_attr);
374 ++count;
375 }
376 kfree(msi_attrs);
377 kfree(dev->msi_irq_groups[0]);
378 kfree(dev->msi_irq_groups);
379 dev->msi_irq_groups = NULL;
380 }
381}
382
383static void pci_intx_for_msi(struct pci_dev *dev, int enable)
384{
385 if (!(dev->dev_flags & PCI_DEV_FLAGS_MSI_INTX_DISABLE_BUG))
386 pci_intx(dev, enable);
387}
388
389static void __pci_restore_msi_state(struct pci_dev *dev)
390{
391 u16 control;
392 struct msi_desc *entry;
393
394 if (!dev->msi_enabled)
395 return;
396
397 entry = irq_get_msi_desc(dev->irq);
398
399 pci_intx_for_msi(dev, 0);
400 pci_msi_set_enable(dev, 0);
401 arch_restore_msi_irqs(dev);
402
403 pci_read_config_word(dev, dev->msi_cap + PCI_MSI_FLAGS, &control);
404 msi_mask_irq(entry, msi_mask(entry->msi_attrib.multi_cap),
405 entry->masked);
406 control &= ~PCI_MSI_FLAGS_QSIZE;
407 control |= (entry->msi_attrib.multiple << 4) | PCI_MSI_FLAGS_ENABLE;
408 pci_write_config_word(dev, dev->msi_cap + PCI_MSI_FLAGS, control);
409}
410
411static void __pci_restore_msix_state(struct pci_dev *dev)
412{
413 struct msi_desc *entry;
414
415 if (!dev->msix_enabled)
416 return;
417 BUG_ON(list_empty(dev_to_msi_list(&dev->dev)));
418
419 /* route the table */
420 pci_intx_for_msi(dev, 0);
421 pci_msix_clear_and_set_ctrl(dev, 0,
422 PCI_MSIX_FLAGS_ENABLE | PCI_MSIX_FLAGS_MASKALL);
423
424 arch_restore_msi_irqs(dev);
425 for_each_pci_msi_entry(entry, dev)
426 msix_mask_irq(entry, entry->masked);
427
428 pci_msix_clear_and_set_ctrl(dev, PCI_MSIX_FLAGS_MASKALL, 0);
429}
430
431void pci_restore_msi_state(struct pci_dev *dev)
432{
433 __pci_restore_msi_state(dev);
434 __pci_restore_msix_state(dev);
435}
436EXPORT_SYMBOL_GPL(pci_restore_msi_state);
437
438static ssize_t msi_mode_show(struct device *dev, struct device_attribute *attr,
439 char *buf)
440{
441 struct msi_desc *entry;
442 unsigned long irq;
443 int retval;
444
445 retval = kstrtoul(attr->attr.name, 10, &irq);
446 if (retval)
447 return retval;
448
449 entry = irq_get_msi_desc(irq);
450 if (entry)
451 return sprintf(buf, "%s\n",
452 entry->msi_attrib.is_msix ? "msix" : "msi");
453
454 return -ENODEV;
455}
456
457static int populate_msi_sysfs(struct pci_dev *pdev)
458{
459 struct attribute **msi_attrs;
460 struct attribute *msi_attr;
461 struct device_attribute *msi_dev_attr;
462 struct attribute_group *msi_irq_group;
463 const struct attribute_group **msi_irq_groups;
464 struct msi_desc *entry;
465 int ret = -ENOMEM;
466 int num_msi = 0;
467 int count = 0;
468 int i;
469
470 /* Determine how many msi entries we have */
471 for_each_pci_msi_entry(entry, pdev)
472 num_msi += entry->nvec_used;
473 if (!num_msi)
474 return 0;
475
476 /* Dynamically create the MSI attributes for the PCI device */
477 msi_attrs = kzalloc(sizeof(void *) * (num_msi + 1), GFP_KERNEL);
478 if (!msi_attrs)
479 return -ENOMEM;
480 for_each_pci_msi_entry(entry, pdev) {
481 for (i = 0; i < entry->nvec_used; i++) {
482 msi_dev_attr = kzalloc(sizeof(*msi_dev_attr), GFP_KERNEL);
483 if (!msi_dev_attr)
484 goto error_attrs;
485 msi_attrs[count] = &msi_dev_attr->attr;
486
487 sysfs_attr_init(&msi_dev_attr->attr);
488 msi_dev_attr->attr.name = kasprintf(GFP_KERNEL, "%d",
489 entry->irq + i);
490 if (!msi_dev_attr->attr.name)
491 goto error_attrs;
492 msi_dev_attr->attr.mode = S_IRUGO;
493 msi_dev_attr->show = msi_mode_show;
494 ++count;
495 }
496 }
497
498 msi_irq_group = kzalloc(sizeof(*msi_irq_group), GFP_KERNEL);
499 if (!msi_irq_group)
500 goto error_attrs;
501 msi_irq_group->name = "msi_irqs";
502 msi_irq_group->attrs = msi_attrs;
503
504 msi_irq_groups = kzalloc(sizeof(void *) * 2, GFP_KERNEL);
505 if (!msi_irq_groups)
506 goto error_irq_group;
507 msi_irq_groups[0] = msi_irq_group;
508
509 ret = sysfs_create_groups(&pdev->dev.kobj, msi_irq_groups);
510 if (ret)
511 goto error_irq_groups;
512 pdev->msi_irq_groups = msi_irq_groups;
513
514 return 0;
515
516error_irq_groups:
517 kfree(msi_irq_groups);
518error_irq_group:
519 kfree(msi_irq_group);
520error_attrs:
521 count = 0;
522 msi_attr = msi_attrs[count];
523 while (msi_attr) {
524 msi_dev_attr = container_of(msi_attr, struct device_attribute, attr);
525 kfree(msi_attr->name);
526 kfree(msi_dev_attr);
527 ++count;
528 msi_attr = msi_attrs[count];
529 }
530 kfree(msi_attrs);
531 return ret;
532}
533
534static struct msi_desc *
535msi_setup_entry(struct pci_dev *dev, int nvec, const struct irq_affinity *affd)
536{
537 struct cpumask *masks = NULL;
538 struct msi_desc *entry;
539 u16 control;
540
541 if (affd) {
542 masks = irq_create_affinity_masks(nvec, affd);
543 if (!masks)
544 pr_err("Unable to allocate affinity masks, ignoring\n");
545 }
546
547 /* MSI Entry Initialization */
548 entry = alloc_msi_entry(&dev->dev, nvec, masks);
549 if (!entry)
550 goto out;
551
552 pci_read_config_word(dev, dev->msi_cap + PCI_MSI_FLAGS, &control);
553
554 entry->msi_attrib.is_msix = 0;
555 entry->msi_attrib.is_64 = !!(control & PCI_MSI_FLAGS_64BIT);
556 entry->msi_attrib.entry_nr = 0;
557 entry->msi_attrib.maskbit = !!(control & PCI_MSI_FLAGS_MASKBIT);
558 entry->msi_attrib.default_irq = dev->irq; /* Save IOAPIC IRQ */
559 entry->msi_attrib.multi_cap = (control & PCI_MSI_FLAGS_QMASK) >> 1;
560 entry->msi_attrib.multiple = ilog2(__roundup_pow_of_two(nvec));
561
562 if (control & PCI_MSI_FLAGS_64BIT)
563 entry->mask_pos = dev->msi_cap + PCI_MSI_MASK_64;
564 else
565 entry->mask_pos = dev->msi_cap + PCI_MSI_MASK_32;
566
567 /* Save the initial mask status */
568 if (entry->msi_attrib.maskbit)
569 pci_read_config_dword(dev, entry->mask_pos, &entry->masked);
570
571out:
572 kfree(masks);
573 return entry;
574}
575
576static int msi_verify_entries(struct pci_dev *dev)
577{
578 struct msi_desc *entry;
579
580 for_each_pci_msi_entry(entry, dev) {
581 if (!dev->no_64bit_msi || !entry->msg.address_hi)
582 continue;
583 dev_err(&dev->dev, "Device has broken 64-bit MSI but arch"
584 " tried to assign one above 4G\n");
585 return -EIO;
586 }
587 return 0;
588}
589
590/**
591 * msi_capability_init - configure device's MSI capability structure
592 * @dev: pointer to the pci_dev data structure of MSI device function
593 * @nvec: number of interrupts to allocate
594 * @affd: description of automatic irq affinity assignments (may be %NULL)
595 *
596 * Setup the MSI capability structure of the device with the requested
597 * number of interrupts. A return value of zero indicates the successful
598 * setup of an entry with the new MSI irq. A negative return value indicates
599 * an error, and a positive return value indicates the number of interrupts
600 * which could have been allocated.
601 */
602static int msi_capability_init(struct pci_dev *dev, int nvec,
603 const struct irq_affinity *affd)
604{
605 struct msi_desc *entry;
606 int ret;
607 unsigned mask;
608
609 pci_msi_set_enable(dev, 0); /* Disable MSI during set up */
610
611 entry = msi_setup_entry(dev, nvec, affd);
612 if (!entry)
613 return -ENOMEM;
614
615 /* All MSIs are unmasked by default, Mask them all */
616 mask = msi_mask(entry->msi_attrib.multi_cap);
617 msi_mask_irq(entry, mask, mask);
618
619 list_add_tail(&entry->list, dev_to_msi_list(&dev->dev));
620
621 /* Configure MSI capability structure */
622 ret = pci_msi_setup_msi_irqs(dev, nvec, PCI_CAP_ID_MSI);
623 if (ret) {
624 msi_mask_irq(entry, mask, ~mask);
625 free_msi_irqs(dev);
626 return ret;
627 }
628
629 ret = msi_verify_entries(dev);
630 if (ret) {
631 msi_mask_irq(entry, mask, ~mask);
632 free_msi_irqs(dev);
633 return ret;
634 }
635
636 ret = populate_msi_sysfs(dev);
637 if (ret) {
638 msi_mask_irq(entry, mask, ~mask);
639 free_msi_irqs(dev);
640 return ret;
641 }
642
643 /* Set MSI enabled bits */
644 pci_intx_for_msi(dev, 0);
645 pci_msi_set_enable(dev, 1);
646 dev->msi_enabled = 1;
647
648 pcibios_free_irq(dev);
649 dev->irq = entry->irq;
650 return 0;
651}
652
653static void __iomem *msix_map_region(struct pci_dev *dev, unsigned nr_entries)
654{
655 resource_size_t phys_addr;
656 u32 table_offset;
657 unsigned long flags;
658 u8 bir;
659
660 pci_read_config_dword(dev, dev->msix_cap + PCI_MSIX_TABLE,
661 &table_offset);
662 bir = (u8)(table_offset & PCI_MSIX_TABLE_BIR);
663 flags = pci_resource_flags(dev, bir);
664 if (!flags || (flags & IORESOURCE_UNSET))
665 return NULL;
666
667 table_offset &= PCI_MSIX_TABLE_OFFSET;
668 phys_addr = pci_resource_start(dev, bir) + table_offset;
669
670 return ioremap_nocache(phys_addr, nr_entries * PCI_MSIX_ENTRY_SIZE);
671}
672
673static int msix_setup_entries(struct pci_dev *dev, void __iomem *base,
674 struct msix_entry *entries, int nvec,
675 const struct irq_affinity *affd)
676{
677 struct cpumask *curmsk, *masks = NULL;
678 struct msi_desc *entry;
679 int ret, i;
680
681 if (affd) {
682 masks = irq_create_affinity_masks(nvec, affd);
683 if (!masks)
684 pr_err("Unable to allocate affinity masks, ignoring\n");
685 }
686
687 for (i = 0, curmsk = masks; i < nvec; i++) {
688 entry = alloc_msi_entry(&dev->dev, 1, curmsk);
689 if (!entry) {
690 if (!i)
691 iounmap(base);
692 else
693 free_msi_irqs(dev);
694 /* No enough memory. Don't try again */
695 ret = -ENOMEM;
696 goto out;
697 }
698
699 entry->msi_attrib.is_msix = 1;
700 entry->msi_attrib.is_64 = 1;
701 if (entries)
702 entry->msi_attrib.entry_nr = entries[i].entry;
703 else
704 entry->msi_attrib.entry_nr = i;
705 entry->msi_attrib.default_irq = dev->irq;
706 entry->mask_base = base;
707
708 list_add_tail(&entry->list, dev_to_msi_list(&dev->dev));
709 if (masks)
710 curmsk++;
711 }
712 ret = 0;
713out:
714 kfree(masks);
715 return ret;
716}
717
718static void msix_program_entries(struct pci_dev *dev,
719 struct msix_entry *entries)
720{
721 struct msi_desc *entry;
722 int i = 0;
723
724 for_each_pci_msi_entry(entry, dev) {
725 if (entries)
726 entries[i++].vector = entry->irq;
727 entry->masked = readl(pci_msix_desc_addr(entry) +
728 PCI_MSIX_ENTRY_VECTOR_CTRL);
729 msix_mask_irq(entry, 1);
730 }
731}
732
733/**
734 * msix_capability_init - configure device's MSI-X capability
735 * @dev: pointer to the pci_dev data structure of MSI-X device function
736 * @entries: pointer to an array of struct msix_entry entries
737 * @nvec: number of @entries
738 * @affd: Optional pointer to enable automatic affinity assignement
739 *
740 * Setup the MSI-X capability structure of device function with a
741 * single MSI-X irq. A return of zero indicates the successful setup of
742 * requested MSI-X entries with allocated irqs or non-zero for otherwise.
743 **/
744static int msix_capability_init(struct pci_dev *dev, struct msix_entry *entries,
745 int nvec, const struct irq_affinity *affd)
746{
747 int ret;
748 u16 control;
749 void __iomem *base;
750
751 /* Ensure MSI-X is disabled while it is set up */
752 pci_msix_clear_and_set_ctrl(dev, PCI_MSIX_FLAGS_ENABLE, 0);
753
754 pci_read_config_word(dev, dev->msix_cap + PCI_MSIX_FLAGS, &control);
755 /* Request & Map MSI-X table region */
756 base = msix_map_region(dev, msix_table_size(control));
757 if (!base)
758 return -ENOMEM;
759
760 ret = msix_setup_entries(dev, base, entries, nvec, affd);
761 if (ret)
762 return ret;
763
764 ret = pci_msi_setup_msi_irqs(dev, nvec, PCI_CAP_ID_MSIX);
765 if (ret)
766 goto out_avail;
767
768 /* Check if all MSI entries honor device restrictions */
769 ret = msi_verify_entries(dev);
770 if (ret)
771 goto out_free;
772
773 /*
774 * Some devices require MSI-X to be enabled before we can touch the
775 * MSI-X registers. We need to mask all the vectors to prevent
776 * interrupts coming in before they're fully set up.
777 */
778 pci_msix_clear_and_set_ctrl(dev, 0,
779 PCI_MSIX_FLAGS_MASKALL | PCI_MSIX_FLAGS_ENABLE);
780
781 msix_program_entries(dev, entries);
782
783 ret = populate_msi_sysfs(dev);
784 if (ret)
785 goto out_free;
786
787 /* Set MSI-X enabled bits and unmask the function */
788 pci_intx_for_msi(dev, 0);
789 dev->msix_enabled = 1;
790 pci_msix_clear_and_set_ctrl(dev, PCI_MSIX_FLAGS_MASKALL, 0);
791
792 pcibios_free_irq(dev);
793 return 0;
794
795out_avail:
796 if (ret < 0) {
797 /*
798 * If we had some success, report the number of irqs
799 * we succeeded in setting up.
800 */
801 struct msi_desc *entry;
802 int avail = 0;
803
804 for_each_pci_msi_entry(entry, dev) {
805 if (entry->irq != 0)
806 avail++;
807 }
808 if (avail != 0)
809 ret = avail;
810 }
811
812out_free:
813 free_msi_irqs(dev);
814
815 return ret;
816}
817
818/**
819 * pci_msi_supported - check whether MSI may be enabled on a device
820 * @dev: pointer to the pci_dev data structure of MSI device function
821 * @nvec: how many MSIs have been requested ?
822 *
823 * Look at global flags, the device itself, and its parent buses
824 * to determine if MSI/-X are supported for the device. If MSI/-X is
825 * supported return 1, else return 0.
826 **/
827static int pci_msi_supported(struct pci_dev *dev, int nvec)
828{
829 struct pci_bus *bus;
830
831 /* MSI must be globally enabled and supported by the device */
832 if (!pci_msi_enable)
833 return 0;
834
835 if (!dev || dev->no_msi || dev->current_state != PCI_D0)
836 return 0;
837
838 /*
839 * You can't ask to have 0 or less MSIs configured.
840 * a) it's stupid ..
841 * b) the list manipulation code assumes nvec >= 1.
842 */
843 if (nvec < 1)
844 return 0;
845
846 /*
847 * Any bridge which does NOT route MSI transactions from its
848 * secondary bus to its primary bus must set NO_MSI flag on
849 * the secondary pci_bus.
850 * We expect only arch-specific PCI host bus controller driver
851 * or quirks for specific PCI bridges to be setting NO_MSI.
852 */
853 for (bus = dev->bus; bus; bus = bus->parent)
854 if (bus->bus_flags & PCI_BUS_FLAGS_NO_MSI)
855 return 0;
856
857 return 1;
858}
859
860/**
861 * pci_msi_vec_count - Return the number of MSI vectors a device can send
862 * @dev: device to report about
863 *
864 * This function returns the number of MSI vectors a device requested via
865 * Multiple Message Capable register. It returns a negative errno if the
866 * device is not capable sending MSI interrupts. Otherwise, the call succeeds
867 * and returns a power of two, up to a maximum of 2^5 (32), according to the
868 * MSI specification.
869 **/
870int pci_msi_vec_count(struct pci_dev *dev)
871{
872 int ret;
873 u16 msgctl;
874
875 if (!dev->msi_cap)
876 return -EINVAL;
877
878 pci_read_config_word(dev, dev->msi_cap + PCI_MSI_FLAGS, &msgctl);
879 ret = 1 << ((msgctl & PCI_MSI_FLAGS_QMASK) >> 1);
880
881 return ret;
882}
883EXPORT_SYMBOL(pci_msi_vec_count);
884
885void pci_msi_shutdown(struct pci_dev *dev)
886{
887 struct msi_desc *desc;
888 u32 mask;
889
890 if (!pci_msi_enable || !dev || !dev->msi_enabled)
891 return;
892
893 BUG_ON(list_empty(dev_to_msi_list(&dev->dev)));
894 desc = first_pci_msi_entry(dev);
895
896 pci_msi_set_enable(dev, 0);
897 pci_intx_for_msi(dev, 1);
898 dev->msi_enabled = 0;
899
900 /* Return the device with MSI unmasked as initial states */
901 mask = msi_mask(desc->msi_attrib.multi_cap);
902 /* Keep cached state to be restored */
903 __pci_msi_desc_mask_irq(desc, mask, ~mask);
904
905 /* Restore dev->irq to its default pin-assertion irq */
906 dev->irq = desc->msi_attrib.default_irq;
907 pcibios_alloc_irq(dev);
908}
909
910void pci_disable_msi(struct pci_dev *dev)
911{
912 if (!pci_msi_enable || !dev || !dev->msi_enabled)
913 return;
914
915 pci_msi_shutdown(dev);
916 free_msi_irqs(dev);
917}
918EXPORT_SYMBOL(pci_disable_msi);
919
920/**
921 * pci_msix_vec_count - return the number of device's MSI-X table entries
922 * @dev: pointer to the pci_dev data structure of MSI-X device function
923 * This function returns the number of device's MSI-X table entries and
924 * therefore the number of MSI-X vectors device is capable of sending.
925 * It returns a negative errno if the device is not capable of sending MSI-X
926 * interrupts.
927 **/
928int pci_msix_vec_count(struct pci_dev *dev)
929{
930 u16 control;
931
932 if (!dev->msix_cap)
933 return -EINVAL;
934
935 pci_read_config_word(dev, dev->msix_cap + PCI_MSIX_FLAGS, &control);
936 return msix_table_size(control);
937}
938EXPORT_SYMBOL(pci_msix_vec_count);
939
940static int __pci_enable_msix(struct pci_dev *dev, struct msix_entry *entries,
941 int nvec, const struct irq_affinity *affd)
942{
943 int nr_entries;
944 int i, j;
945
946 if (!pci_msi_supported(dev, nvec))
947 return -EINVAL;
948
949 nr_entries = pci_msix_vec_count(dev);
950 if (nr_entries < 0)
951 return nr_entries;
952 if (nvec > nr_entries)
953 return nr_entries;
954
955 if (entries) {
956 /* Check for any invalid entries */
957 for (i = 0; i < nvec; i++) {
958 if (entries[i].entry >= nr_entries)
959 return -EINVAL; /* invalid entry */
960 for (j = i + 1; j < nvec; j++) {
961 if (entries[i].entry == entries[j].entry)
962 return -EINVAL; /* duplicate entry */
963 }
964 }
965 }
966 WARN_ON(!!dev->msix_enabled);
967
968 /* Check whether driver already requested for MSI irq */
969 if (dev->msi_enabled) {
970 dev_info(&dev->dev, "can't enable MSI-X (MSI IRQ already assigned)\n");
971 return -EINVAL;
972 }
973 return msix_capability_init(dev, entries, nvec, affd);
974}
975
976/**
977 * pci_enable_msix - configure device's MSI-X capability structure
978 * @dev: pointer to the pci_dev data structure of MSI-X device function
979 * @entries: pointer to an array of MSI-X entries (optional)
980 * @nvec: number of MSI-X irqs requested for allocation by device driver
981 *
982 * Setup the MSI-X capability structure of device function with the number
983 * of requested irqs upon its software driver call to request for
984 * MSI-X mode enabled on its hardware device function. A return of zero
985 * indicates the successful configuration of MSI-X capability structure
986 * with new allocated MSI-X irqs. A return of < 0 indicates a failure.
987 * Or a return of > 0 indicates that driver request is exceeding the number
988 * of irqs or MSI-X vectors available. Driver should use the returned value to
989 * re-send its request.
990 **/
991int pci_enable_msix(struct pci_dev *dev, struct msix_entry *entries, int nvec)
992{
993 return __pci_enable_msix(dev, entries, nvec, NULL);
994}
995EXPORT_SYMBOL(pci_enable_msix);
996
997void pci_msix_shutdown(struct pci_dev *dev)
998{
999 struct msi_desc *entry;
1000
1001 if (!pci_msi_enable || !dev || !dev->msix_enabled)
1002 return;
1003
1004 /* Return the device with MSI-X masked as initial states */
1005 for_each_pci_msi_entry(entry, dev) {
1006 /* Keep cached states to be restored */
1007 __pci_msix_desc_mask_irq(entry, 1);
1008 }
1009
1010 pci_msix_clear_and_set_ctrl(dev, PCI_MSIX_FLAGS_ENABLE, 0);
1011 pci_intx_for_msi(dev, 1);
1012 dev->msix_enabled = 0;
1013 pcibios_alloc_irq(dev);
1014}
1015
1016void pci_disable_msix(struct pci_dev *dev)
1017{
1018 if (!pci_msi_enable || !dev || !dev->msix_enabled)
1019 return;
1020
1021 pci_msix_shutdown(dev);
1022 free_msi_irqs(dev);
1023}
1024EXPORT_SYMBOL(pci_disable_msix);
1025
1026void pci_no_msi(void)
1027{
1028 pci_msi_enable = 0;
1029}
1030
1031/**
1032 * pci_msi_enabled - is MSI enabled?
1033 *
1034 * Returns true if MSI has not been disabled by the command-line option
1035 * pci=nomsi.
1036 **/
1037int pci_msi_enabled(void)
1038{
1039 return pci_msi_enable;
1040}
1041EXPORT_SYMBOL(pci_msi_enabled);
1042
1043static int __pci_enable_msi_range(struct pci_dev *dev, int minvec, int maxvec,
1044 const struct irq_affinity *affd)
1045{
1046 int nvec;
1047 int rc;
1048
1049 if (!pci_msi_supported(dev, minvec))
1050 return -EINVAL;
1051
1052 WARN_ON(!!dev->msi_enabled);
1053
1054 /* Check whether driver already requested MSI-X irqs */
1055 if (dev->msix_enabled) {
1056 dev_info(&dev->dev,
1057 "can't enable MSI (MSI-X already enabled)\n");
1058 return -EINVAL;
1059 }
1060
1061 if (maxvec < minvec)
1062 return -ERANGE;
1063
1064 nvec = pci_msi_vec_count(dev);
1065 if (nvec < 0)
1066 return nvec;
1067 if (nvec < minvec)
1068 return -ENOSPC;
1069
1070 if (nvec > maxvec)
1071 nvec = maxvec;
1072
1073 for (;;) {
1074 if (affd) {
1075 nvec = irq_calc_affinity_vectors(nvec, affd);
1076 if (nvec < minvec)
1077 return -ENOSPC;
1078 }
1079
1080 rc = msi_capability_init(dev, nvec, affd);
1081 if (rc == 0)
1082 return nvec;
1083
1084 if (rc < 0)
1085 return rc;
1086 if (rc < minvec)
1087 return -ENOSPC;
1088
1089 nvec = rc;
1090 }
1091}
1092
1093/* deprecated, don't use */
1094int pci_enable_msi(struct pci_dev *dev)
1095{
1096 int rc = __pci_enable_msi_range(dev, 1, 1, NULL);
1097 if (rc < 0)
1098 return rc;
1099 return 0;
1100}
1101EXPORT_SYMBOL(pci_enable_msi);
1102
1103static int __pci_enable_msix_range(struct pci_dev *dev,
1104 struct msix_entry *entries, int minvec,
1105 int maxvec, const struct irq_affinity *affd)
1106{
1107 int rc, nvec = maxvec;
1108
1109 if (maxvec < minvec)
1110 return -ERANGE;
1111
1112 for (;;) {
1113 if (affd) {
1114 nvec = irq_calc_affinity_vectors(nvec, affd);
1115 if (nvec < minvec)
1116 return -ENOSPC;
1117 }
1118
1119 rc = __pci_enable_msix(dev, entries, nvec, affd);
1120 if (rc == 0)
1121 return nvec;
1122
1123 if (rc < 0)
1124 return rc;
1125 if (rc < minvec)
1126 return -ENOSPC;
1127
1128 nvec = rc;
1129 }
1130}
1131
1132/**
1133 * pci_enable_msix_range - configure device's MSI-X capability structure
1134 * @dev: pointer to the pci_dev data structure of MSI-X device function
1135 * @entries: pointer to an array of MSI-X entries
1136 * @minvec: minimum number of MSI-X irqs requested
1137 * @maxvec: maximum number of MSI-X irqs requested
1138 *
1139 * Setup the MSI-X capability structure of device function with a maximum
1140 * possible number of interrupts in the range between @minvec and @maxvec
1141 * upon its software driver call to request for MSI-X mode enabled on its
1142 * hardware device function. It returns a negative errno if an error occurs.
1143 * If it succeeds, it returns the actual number of interrupts allocated and
1144 * indicates the successful configuration of MSI-X capability structure
1145 * with new allocated MSI-X interrupts.
1146 **/
1147int pci_enable_msix_range(struct pci_dev *dev, struct msix_entry *entries,
1148 int minvec, int maxvec)
1149{
1150 return __pci_enable_msix_range(dev, entries, minvec, maxvec, NULL);
1151}
1152EXPORT_SYMBOL(pci_enable_msix_range);
1153
1154/**
1155 * pci_alloc_irq_vectors_affinity - allocate multiple IRQs for a device
1156 * @dev: PCI device to operate on
1157 * @min_vecs: minimum number of vectors required (must be >= 1)
1158 * @max_vecs: maximum (desired) number of vectors
1159 * @flags: flags or quirks for the allocation
1160 * @affd: optional description of the affinity requirements
1161 *
1162 * Allocate up to @max_vecs interrupt vectors for @dev, using MSI-X or MSI
1163 * vectors if available, and fall back to a single legacy vector
1164 * if neither is available. Return the number of vectors allocated,
1165 * (which might be smaller than @max_vecs) if successful, or a negative
1166 * error code on error. If less than @min_vecs interrupt vectors are
1167 * available for @dev the function will fail with -ENOSPC.
1168 *
1169 * To get the Linux IRQ number used for a vector that can be passed to
1170 * request_irq() use the pci_irq_vector() helper.
1171 */
1172int pci_alloc_irq_vectors_affinity(struct pci_dev *dev, unsigned int min_vecs,
1173 unsigned int max_vecs, unsigned int flags,
1174 const struct irq_affinity *affd)
1175{
1176 static const struct irq_affinity msi_default_affd;
1177 int vecs = -ENOSPC;
1178
1179 if (flags & PCI_IRQ_AFFINITY) {
1180 if (!affd)
1181 affd = &msi_default_affd;
1182
1183 if (affd->pre_vectors + affd->post_vectors > min_vecs)
1184 return -EINVAL;
1185
1186 /*
1187 * If there aren't any vectors left after applying the pre/post
1188 * vectors don't bother with assigning affinity.
1189 */
1190 if (affd->pre_vectors + affd->post_vectors == min_vecs)
1191 affd = NULL;
1192 } else {
1193 if (WARN_ON(affd))
1194 affd = NULL;
1195 }
1196
1197 if (flags & PCI_IRQ_MSIX) {
1198 vecs = __pci_enable_msix_range(dev, NULL, min_vecs, max_vecs,
1199 affd);
1200 if (vecs > 0)
1201 return vecs;
1202 }
1203
1204 if (flags & PCI_IRQ_MSI) {
1205 vecs = __pci_enable_msi_range(dev, min_vecs, max_vecs, affd);
1206 if (vecs > 0)
1207 return vecs;
1208 }
1209
1210 /* use legacy irq if allowed */
1211 if (flags & PCI_IRQ_LEGACY) {
1212 if (min_vecs == 1 && dev->irq) {
1213 pci_intx(dev, 1);
1214 return 1;
1215 }
1216 }
1217
1218 return vecs;
1219}
1220EXPORT_SYMBOL(pci_alloc_irq_vectors_affinity);
1221
1222/**
1223 * pci_free_irq_vectors - free previously allocated IRQs for a device
1224 * @dev: PCI device to operate on
1225 *
1226 * Undoes the allocations and enabling in pci_alloc_irq_vectors().
1227 */
1228void pci_free_irq_vectors(struct pci_dev *dev)
1229{
1230 pci_disable_msix(dev);
1231 pci_disable_msi(dev);
1232}
1233EXPORT_SYMBOL(pci_free_irq_vectors);
1234
1235/**
1236 * pci_irq_vector - return Linux IRQ number of a device vector
1237 * @dev: PCI device to operate on
1238 * @nr: device-relative interrupt vector index (0-based).
1239 */
1240int pci_irq_vector(struct pci_dev *dev, unsigned int nr)
1241{
1242 if (dev->msix_enabled) {
1243 struct msi_desc *entry;
1244 int i = 0;
1245
1246 for_each_pci_msi_entry(entry, dev) {
1247 if (i == nr)
1248 return entry->irq;
1249 i++;
1250 }
1251 WARN_ON_ONCE(1);
1252 return -EINVAL;
1253 }
1254
1255 if (dev->msi_enabled) {
1256 struct msi_desc *entry = first_pci_msi_entry(dev);
1257
1258 if (WARN_ON_ONCE(nr >= entry->nvec_used))
1259 return -EINVAL;
1260 } else {
1261 if (WARN_ON_ONCE(nr > 0))
1262 return -EINVAL;
1263 }
1264
1265 return dev->irq + nr;
1266}
1267EXPORT_SYMBOL(pci_irq_vector);
1268
1269/**
1270 * pci_irq_get_affinity - return the affinity of a particular msi vector
1271 * @dev: PCI device to operate on
1272 * @nr: device-relative interrupt vector index (0-based).
1273 */
1274const struct cpumask *pci_irq_get_affinity(struct pci_dev *dev, int nr)
1275{
1276 if (dev->msix_enabled) {
1277 struct msi_desc *entry;
1278 int i = 0;
1279
1280 for_each_pci_msi_entry(entry, dev) {
1281 if (i == nr)
1282 return entry->affinity;
1283 i++;
1284 }
1285 WARN_ON_ONCE(1);
1286 return NULL;
1287 } else if (dev->msi_enabled) {
1288 struct msi_desc *entry = first_pci_msi_entry(dev);
1289
1290 if (WARN_ON_ONCE(!entry || !entry->affinity ||
1291 nr >= entry->nvec_used))
1292 return NULL;
1293
1294 return &entry->affinity[nr];
1295 } else {
1296 return cpu_possible_mask;
1297 }
1298}
1299EXPORT_SYMBOL(pci_irq_get_affinity);
1300
1301/**
1302 * pci_irq_get_node - return the numa node of a particular msi vector
1303 * @pdev: PCI device to operate on
1304 * @vec: device-relative interrupt vector index (0-based).
1305 */
1306int pci_irq_get_node(struct pci_dev *pdev, int vec)
1307{
1308 const struct cpumask *mask;
1309
1310 mask = pci_irq_get_affinity(pdev, vec);
1311 if (mask)
1312 return local_memory_node(cpu_to_node(cpumask_first(mask)));
1313 return dev_to_node(&pdev->dev);
1314}
1315EXPORT_SYMBOL(pci_irq_get_node);
1316
1317struct pci_dev *msi_desc_to_pci_dev(struct msi_desc *desc)
1318{
1319 return to_pci_dev(desc->dev);
1320}
1321EXPORT_SYMBOL(msi_desc_to_pci_dev);
1322
1323void *msi_desc_to_pci_sysdata(struct msi_desc *desc)
1324{
1325 struct pci_dev *dev = msi_desc_to_pci_dev(desc);
1326
1327 return dev->bus->sysdata;
1328}
1329EXPORT_SYMBOL_GPL(msi_desc_to_pci_sysdata);
1330
1331#ifdef CONFIG_PCI_MSI_IRQ_DOMAIN
1332/**
1333 * pci_msi_domain_write_msg - Helper to write MSI message to PCI config space
1334 * @irq_data: Pointer to interrupt data of the MSI interrupt
1335 * @msg: Pointer to the message
1336 */
1337void pci_msi_domain_write_msg(struct irq_data *irq_data, struct msi_msg *msg)
1338{
1339 struct msi_desc *desc = irq_data_get_msi_desc(irq_data);
1340
1341 /*
1342 * For MSI-X desc->irq is always equal to irq_data->irq. For
1343 * MSI only the first interrupt of MULTI MSI passes the test.
1344 */
1345 if (desc->irq == irq_data->irq)
1346 __pci_write_msi_msg(desc, msg);
1347}
1348
1349/**
1350 * pci_msi_domain_calc_hwirq - Generate a unique ID for an MSI source
1351 * @dev: Pointer to the PCI device
1352 * @desc: Pointer to the msi descriptor
1353 *
1354 * The ID number is only used within the irqdomain.
1355 */
1356irq_hw_number_t pci_msi_domain_calc_hwirq(struct pci_dev *dev,
1357 struct msi_desc *desc)
1358{
1359 return (irq_hw_number_t)desc->msi_attrib.entry_nr |
1360 PCI_DEVID(dev->bus->number, dev->devfn) << 11 |
1361 (pci_domain_nr(dev->bus) & 0xFFFFFFFF) << 27;
1362}
1363
1364static inline bool pci_msi_desc_is_multi_msi(struct msi_desc *desc)
1365{
1366 return !desc->msi_attrib.is_msix && desc->nvec_used > 1;
1367}
1368
1369/**
1370 * pci_msi_domain_check_cap - Verify that @domain supports the capabilities for @dev
1371 * @domain: The interrupt domain to check
1372 * @info: The domain info for verification
1373 * @dev: The device to check
1374 *
1375 * Returns:
1376 * 0 if the functionality is supported
1377 * 1 if Multi MSI is requested, but the domain does not support it
1378 * -ENOTSUPP otherwise
1379 */
1380int pci_msi_domain_check_cap(struct irq_domain *domain,
1381 struct msi_domain_info *info, struct device *dev)
1382{
1383 struct msi_desc *desc = first_pci_msi_entry(to_pci_dev(dev));
1384
1385 /* Special handling to support __pci_enable_msi_range() */
1386 if (pci_msi_desc_is_multi_msi(desc) &&
1387 !(info->flags & MSI_FLAG_MULTI_PCI_MSI))
1388 return 1;
1389 else if (desc->msi_attrib.is_msix && !(info->flags & MSI_FLAG_PCI_MSIX))
1390 return -ENOTSUPP;
1391
1392 return 0;
1393}
1394
1395static int pci_msi_domain_handle_error(struct irq_domain *domain,
1396 struct msi_desc *desc, int error)
1397{
1398 /* Special handling to support __pci_enable_msi_range() */
1399 if (pci_msi_desc_is_multi_msi(desc) && error == -ENOSPC)
1400 return 1;
1401
1402 return error;
1403}
1404
1405#ifdef GENERIC_MSI_DOMAIN_OPS
1406static void pci_msi_domain_set_desc(msi_alloc_info_t *arg,
1407 struct msi_desc *desc)
1408{
1409 arg->desc = desc;
1410 arg->hwirq = pci_msi_domain_calc_hwirq(msi_desc_to_pci_dev(desc),
1411 desc);
1412}
1413#else
1414#define pci_msi_domain_set_desc NULL
1415#endif
1416
1417static struct msi_domain_ops pci_msi_domain_ops_default = {
1418 .set_desc = pci_msi_domain_set_desc,
1419 .msi_check = pci_msi_domain_check_cap,
1420 .handle_error = pci_msi_domain_handle_error,
1421};
1422
1423static void pci_msi_domain_update_dom_ops(struct msi_domain_info *info)
1424{
1425 struct msi_domain_ops *ops = info->ops;
1426
1427 if (ops == NULL) {
1428 info->ops = &pci_msi_domain_ops_default;
1429 } else {
1430 if (ops->set_desc == NULL)
1431 ops->set_desc = pci_msi_domain_set_desc;
1432 if (ops->msi_check == NULL)
1433 ops->msi_check = pci_msi_domain_check_cap;
1434 if (ops->handle_error == NULL)
1435 ops->handle_error = pci_msi_domain_handle_error;
1436 }
1437}
1438
1439static void pci_msi_domain_update_chip_ops(struct msi_domain_info *info)
1440{
1441 struct irq_chip *chip = info->chip;
1442
1443 BUG_ON(!chip);
1444 if (!chip->irq_write_msi_msg)
1445 chip->irq_write_msi_msg = pci_msi_domain_write_msg;
1446 if (!chip->irq_mask)
1447 chip->irq_mask = pci_msi_mask_irq;
1448 if (!chip->irq_unmask)
1449 chip->irq_unmask = pci_msi_unmask_irq;
1450}
1451
1452/**
1453 * pci_msi_create_irq_domain - Create a MSI interrupt domain
1454 * @fwnode: Optional fwnode of the interrupt controller
1455 * @info: MSI domain info
1456 * @parent: Parent irq domain
1457 *
1458 * Updates the domain and chip ops and creates a MSI interrupt domain.
1459 *
1460 * Returns:
1461 * A domain pointer or NULL in case of failure.
1462 */
1463struct irq_domain *pci_msi_create_irq_domain(struct fwnode_handle *fwnode,
1464 struct msi_domain_info *info,
1465 struct irq_domain *parent)
1466{
1467 struct irq_domain *domain;
1468
1469 if (info->flags & MSI_FLAG_USE_DEF_DOM_OPS)
1470 pci_msi_domain_update_dom_ops(info);
1471 if (info->flags & MSI_FLAG_USE_DEF_CHIP_OPS)
1472 pci_msi_domain_update_chip_ops(info);
1473
1474 info->flags |= MSI_FLAG_ACTIVATE_EARLY;
1475
1476 domain = msi_create_irq_domain(fwnode, info, parent);
1477 if (!domain)
1478 return NULL;
1479
1480 domain->bus_token = DOMAIN_BUS_PCI_MSI;
1481 return domain;
1482}
1483EXPORT_SYMBOL_GPL(pci_msi_create_irq_domain);
1484
1485static int get_msi_id_cb(struct pci_dev *pdev, u16 alias, void *data)
1486{
1487 u32 *pa = data;
1488
1489 *pa = alias;
1490 return 0;
1491}
1492/**
1493 * pci_msi_domain_get_msi_rid - Get the MSI requester id (RID)
1494 * @domain: The interrupt domain
1495 * @pdev: The PCI device.
1496 *
1497 * The RID for a device is formed from the alias, with a firmware
1498 * supplied mapping applied
1499 *
1500 * Returns: The RID.
1501 */
1502u32 pci_msi_domain_get_msi_rid(struct irq_domain *domain, struct pci_dev *pdev)
1503{
1504 struct device_node *of_node;
1505 u32 rid = 0;
1506
1507 pci_for_each_dma_alias(pdev, get_msi_id_cb, &rid);
1508
1509 of_node = irq_domain_get_of_node(domain);
1510 rid = of_node ? of_msi_map_rid(&pdev->dev, of_node, rid) :
1511 iort_msi_map_rid(&pdev->dev, rid);
1512
1513 return rid;
1514}
1515
1516/**
1517 * pci_msi_get_device_domain - Get the MSI domain for a given PCI device
1518 * @pdev: The PCI device
1519 *
1520 * Use the firmware data to find a device-specific MSI domain
1521 * (i.e. not one that is ste as a default).
1522 *
1523 * Returns: The coresponding MSI domain or NULL if none has been found.
1524 */
1525struct irq_domain *pci_msi_get_device_domain(struct pci_dev *pdev)
1526{
1527 struct irq_domain *dom;
1528 u32 rid = 0;
1529
1530 pci_for_each_dma_alias(pdev, get_msi_id_cb, &rid);
1531 dom = of_msi_map_get_device_domain(&pdev->dev, rid);
1532 if (!dom)
1533 dom = iort_get_device_domain(&pdev->dev, rid);
1534 return dom;
1535}
1536#endif /* CONFIG_PCI_MSI_IRQ_DOMAIN */