Loading...
1// SPDX-License-Identifier: GPL-2.0
2/*
3 * 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 if (desc->msi_attrib.is_virtual)
196 return NULL;
197
198 return desc->mask_base +
199 desc->msi_attrib.entry_nr * PCI_MSIX_ENTRY_SIZE;
200}
201
202/*
203 * This internal function does not flush PCI writes to the device.
204 * All users must ensure that they read from the device before either
205 * assuming that the device state is up to date, or returning out of this
206 * file. This saves a few milliseconds when initialising devices with lots
207 * of MSI-X interrupts.
208 */
209u32 __pci_msix_desc_mask_irq(struct msi_desc *desc, u32 flag)
210{
211 u32 mask_bits = desc->masked;
212 void __iomem *desc_addr;
213
214 if (pci_msi_ignore_mask)
215 return 0;
216 desc_addr = pci_msix_desc_addr(desc);
217 if (!desc_addr)
218 return 0;
219
220 mask_bits &= ~PCI_MSIX_ENTRY_CTRL_MASKBIT;
221 if (flag)
222 mask_bits |= PCI_MSIX_ENTRY_CTRL_MASKBIT;
223
224 writel(mask_bits, desc_addr + PCI_MSIX_ENTRY_VECTOR_CTRL);
225
226 return mask_bits;
227}
228
229static void msix_mask_irq(struct msi_desc *desc, u32 flag)
230{
231 desc->masked = __pci_msix_desc_mask_irq(desc, flag);
232}
233
234static void msi_set_mask_bit(struct irq_data *data, u32 flag)
235{
236 struct msi_desc *desc = irq_data_get_msi_desc(data);
237
238 if (desc->msi_attrib.is_msix) {
239 msix_mask_irq(desc, flag);
240 readl(desc->mask_base); /* Flush write to device */
241 } else {
242 unsigned offset = data->irq - desc->irq;
243 msi_mask_irq(desc, 1 << offset, flag << offset);
244 }
245}
246
247/**
248 * pci_msi_mask_irq - Generic IRQ chip callback to mask PCI/MSI interrupts
249 * @data: pointer to irqdata associated to that interrupt
250 */
251void pci_msi_mask_irq(struct irq_data *data)
252{
253 msi_set_mask_bit(data, 1);
254}
255EXPORT_SYMBOL_GPL(pci_msi_mask_irq);
256
257/**
258 * pci_msi_unmask_irq - Generic IRQ chip callback to unmask PCI/MSI interrupts
259 * @data: pointer to irqdata associated to that interrupt
260 */
261void pci_msi_unmask_irq(struct irq_data *data)
262{
263 msi_set_mask_bit(data, 0);
264}
265EXPORT_SYMBOL_GPL(pci_msi_unmask_irq);
266
267void default_restore_msi_irqs(struct pci_dev *dev)
268{
269 struct msi_desc *entry;
270
271 for_each_pci_msi_entry(entry, dev)
272 default_restore_msi_irq(dev, entry->irq);
273}
274
275void __pci_read_msi_msg(struct msi_desc *entry, struct msi_msg *msg)
276{
277 struct pci_dev *dev = msi_desc_to_pci_dev(entry);
278
279 BUG_ON(dev->current_state != PCI_D0);
280
281 if (entry->msi_attrib.is_msix) {
282 void __iomem *base = pci_msix_desc_addr(entry);
283
284 if (!base) {
285 WARN_ON(1);
286 return;
287 }
288
289 msg->address_lo = readl(base + PCI_MSIX_ENTRY_LOWER_ADDR);
290 msg->address_hi = readl(base + PCI_MSIX_ENTRY_UPPER_ADDR);
291 msg->data = readl(base + PCI_MSIX_ENTRY_DATA);
292 } else {
293 int pos = dev->msi_cap;
294 u16 data;
295
296 pci_read_config_dword(dev, pos + PCI_MSI_ADDRESS_LO,
297 &msg->address_lo);
298 if (entry->msi_attrib.is_64) {
299 pci_read_config_dword(dev, pos + PCI_MSI_ADDRESS_HI,
300 &msg->address_hi);
301 pci_read_config_word(dev, pos + PCI_MSI_DATA_64, &data);
302 } else {
303 msg->address_hi = 0;
304 pci_read_config_word(dev, pos + PCI_MSI_DATA_32, &data);
305 }
306 msg->data = data;
307 }
308}
309
310void __pci_write_msi_msg(struct msi_desc *entry, struct msi_msg *msg)
311{
312 struct pci_dev *dev = msi_desc_to_pci_dev(entry);
313
314 if (dev->current_state != PCI_D0 || pci_dev_is_disconnected(dev)) {
315 /* Don't touch the hardware now */
316 } else if (entry->msi_attrib.is_msix) {
317 void __iomem *base = pci_msix_desc_addr(entry);
318
319 if (!base)
320 goto skip;
321
322 writel(msg->address_lo, base + PCI_MSIX_ENTRY_LOWER_ADDR);
323 writel(msg->address_hi, base + PCI_MSIX_ENTRY_UPPER_ADDR);
324 writel(msg->data, base + PCI_MSIX_ENTRY_DATA);
325 } else {
326 int pos = dev->msi_cap;
327 u16 msgctl;
328
329 pci_read_config_word(dev, pos + PCI_MSI_FLAGS, &msgctl);
330 msgctl &= ~PCI_MSI_FLAGS_QSIZE;
331 msgctl |= entry->msi_attrib.multiple << 4;
332 pci_write_config_word(dev, pos + PCI_MSI_FLAGS, msgctl);
333
334 pci_write_config_dword(dev, pos + PCI_MSI_ADDRESS_LO,
335 msg->address_lo);
336 if (entry->msi_attrib.is_64) {
337 pci_write_config_dword(dev, pos + PCI_MSI_ADDRESS_HI,
338 msg->address_hi);
339 pci_write_config_word(dev, pos + PCI_MSI_DATA_64,
340 msg->data);
341 } else {
342 pci_write_config_word(dev, pos + PCI_MSI_DATA_32,
343 msg->data);
344 }
345 }
346
347skip:
348 entry->msg = *msg;
349
350 if (entry->write_msi_msg)
351 entry->write_msi_msg(entry, entry->write_msi_msg_data);
352
353}
354
355void pci_write_msi_msg(unsigned int irq, struct msi_msg *msg)
356{
357 struct msi_desc *entry = irq_get_msi_desc(irq);
358
359 __pci_write_msi_msg(entry, msg);
360}
361EXPORT_SYMBOL_GPL(pci_write_msi_msg);
362
363static void free_msi_irqs(struct pci_dev *dev)
364{
365 struct list_head *msi_list = dev_to_msi_list(&dev->dev);
366 struct msi_desc *entry, *tmp;
367 struct attribute **msi_attrs;
368 struct device_attribute *dev_attr;
369 int i, count = 0;
370
371 for_each_pci_msi_entry(entry, dev)
372 if (entry->irq)
373 for (i = 0; i < entry->nvec_used; i++)
374 BUG_ON(irq_has_action(entry->irq + i));
375
376 pci_msi_teardown_msi_irqs(dev);
377
378 list_for_each_entry_safe(entry, tmp, msi_list, list) {
379 if (entry->msi_attrib.is_msix) {
380 if (list_is_last(&entry->list, msi_list))
381 iounmap(entry->mask_base);
382 }
383
384 list_del(&entry->list);
385 free_msi_entry(entry);
386 }
387
388 if (dev->msi_irq_groups) {
389 sysfs_remove_groups(&dev->dev.kobj, dev->msi_irq_groups);
390 msi_attrs = dev->msi_irq_groups[0]->attrs;
391 while (msi_attrs[count]) {
392 dev_attr = container_of(msi_attrs[count],
393 struct device_attribute, attr);
394 kfree(dev_attr->attr.name);
395 kfree(dev_attr);
396 ++count;
397 }
398 kfree(msi_attrs);
399 kfree(dev->msi_irq_groups[0]);
400 kfree(dev->msi_irq_groups);
401 dev->msi_irq_groups = NULL;
402 }
403}
404
405static void pci_intx_for_msi(struct pci_dev *dev, int enable)
406{
407 if (!(dev->dev_flags & PCI_DEV_FLAGS_MSI_INTX_DISABLE_BUG))
408 pci_intx(dev, enable);
409}
410
411static void __pci_restore_msi_state(struct pci_dev *dev)
412{
413 u16 control;
414 struct msi_desc *entry;
415
416 if (!dev->msi_enabled)
417 return;
418
419 entry = irq_get_msi_desc(dev->irq);
420
421 pci_intx_for_msi(dev, 0);
422 pci_msi_set_enable(dev, 0);
423 arch_restore_msi_irqs(dev);
424
425 pci_read_config_word(dev, dev->msi_cap + PCI_MSI_FLAGS, &control);
426 msi_mask_irq(entry, msi_mask(entry->msi_attrib.multi_cap),
427 entry->masked);
428 control &= ~PCI_MSI_FLAGS_QSIZE;
429 control |= (entry->msi_attrib.multiple << 4) | PCI_MSI_FLAGS_ENABLE;
430 pci_write_config_word(dev, dev->msi_cap + PCI_MSI_FLAGS, control);
431}
432
433static void __pci_restore_msix_state(struct pci_dev *dev)
434{
435 struct msi_desc *entry;
436
437 if (!dev->msix_enabled)
438 return;
439 BUG_ON(list_empty(dev_to_msi_list(&dev->dev)));
440
441 /* route the table */
442 pci_intx_for_msi(dev, 0);
443 pci_msix_clear_and_set_ctrl(dev, 0,
444 PCI_MSIX_FLAGS_ENABLE | PCI_MSIX_FLAGS_MASKALL);
445
446 arch_restore_msi_irqs(dev);
447 for_each_pci_msi_entry(entry, dev)
448 msix_mask_irq(entry, entry->masked);
449
450 pci_msix_clear_and_set_ctrl(dev, PCI_MSIX_FLAGS_MASKALL, 0);
451}
452
453void pci_restore_msi_state(struct pci_dev *dev)
454{
455 __pci_restore_msi_state(dev);
456 __pci_restore_msix_state(dev);
457}
458EXPORT_SYMBOL_GPL(pci_restore_msi_state);
459
460static ssize_t msi_mode_show(struct device *dev, struct device_attribute *attr,
461 char *buf)
462{
463 struct msi_desc *entry;
464 unsigned long irq;
465 int retval;
466
467 retval = kstrtoul(attr->attr.name, 10, &irq);
468 if (retval)
469 return retval;
470
471 entry = irq_get_msi_desc(irq);
472 if (entry)
473 return sprintf(buf, "%s\n",
474 entry->msi_attrib.is_msix ? "msix" : "msi");
475
476 return -ENODEV;
477}
478
479static int populate_msi_sysfs(struct pci_dev *pdev)
480{
481 struct attribute **msi_attrs;
482 struct attribute *msi_attr;
483 struct device_attribute *msi_dev_attr;
484 struct attribute_group *msi_irq_group;
485 const struct attribute_group **msi_irq_groups;
486 struct msi_desc *entry;
487 int ret = -ENOMEM;
488 int num_msi = 0;
489 int count = 0;
490 int i;
491
492 /* Determine how many msi entries we have */
493 for_each_pci_msi_entry(entry, pdev)
494 num_msi += entry->nvec_used;
495 if (!num_msi)
496 return 0;
497
498 /* Dynamically create the MSI attributes for the PCI device */
499 msi_attrs = kcalloc(num_msi + 1, sizeof(void *), GFP_KERNEL);
500 if (!msi_attrs)
501 return -ENOMEM;
502 for_each_pci_msi_entry(entry, pdev) {
503 for (i = 0; i < entry->nvec_used; i++) {
504 msi_dev_attr = kzalloc(sizeof(*msi_dev_attr), GFP_KERNEL);
505 if (!msi_dev_attr)
506 goto error_attrs;
507 msi_attrs[count] = &msi_dev_attr->attr;
508
509 sysfs_attr_init(&msi_dev_attr->attr);
510 msi_dev_attr->attr.name = kasprintf(GFP_KERNEL, "%d",
511 entry->irq + i);
512 if (!msi_dev_attr->attr.name)
513 goto error_attrs;
514 msi_dev_attr->attr.mode = S_IRUGO;
515 msi_dev_attr->show = msi_mode_show;
516 ++count;
517 }
518 }
519
520 msi_irq_group = kzalloc(sizeof(*msi_irq_group), GFP_KERNEL);
521 if (!msi_irq_group)
522 goto error_attrs;
523 msi_irq_group->name = "msi_irqs";
524 msi_irq_group->attrs = msi_attrs;
525
526 msi_irq_groups = kcalloc(2, sizeof(void *), GFP_KERNEL);
527 if (!msi_irq_groups)
528 goto error_irq_group;
529 msi_irq_groups[0] = msi_irq_group;
530
531 ret = sysfs_create_groups(&pdev->dev.kobj, msi_irq_groups);
532 if (ret)
533 goto error_irq_groups;
534 pdev->msi_irq_groups = msi_irq_groups;
535
536 return 0;
537
538error_irq_groups:
539 kfree(msi_irq_groups);
540error_irq_group:
541 kfree(msi_irq_group);
542error_attrs:
543 count = 0;
544 msi_attr = msi_attrs[count];
545 while (msi_attr) {
546 msi_dev_attr = container_of(msi_attr, struct device_attribute, attr);
547 kfree(msi_attr->name);
548 kfree(msi_dev_attr);
549 ++count;
550 msi_attr = msi_attrs[count];
551 }
552 kfree(msi_attrs);
553 return ret;
554}
555
556static struct msi_desc *
557msi_setup_entry(struct pci_dev *dev, int nvec, struct irq_affinity *affd)
558{
559 struct irq_affinity_desc *masks = NULL;
560 struct msi_desc *entry;
561 u16 control;
562
563 if (affd)
564 masks = irq_create_affinity_masks(nvec, affd);
565
566 /* MSI Entry Initialization */
567 entry = alloc_msi_entry(&dev->dev, nvec, masks);
568 if (!entry)
569 goto out;
570
571 pci_read_config_word(dev, dev->msi_cap + PCI_MSI_FLAGS, &control);
572
573 entry->msi_attrib.is_msix = 0;
574 entry->msi_attrib.is_64 = !!(control & PCI_MSI_FLAGS_64BIT);
575 entry->msi_attrib.is_virtual = 0;
576 entry->msi_attrib.entry_nr = 0;
577 entry->msi_attrib.maskbit = !!(control & PCI_MSI_FLAGS_MASKBIT);
578 entry->msi_attrib.default_irq = dev->irq; /* Save IOAPIC IRQ */
579 entry->msi_attrib.multi_cap = (control & PCI_MSI_FLAGS_QMASK) >> 1;
580 entry->msi_attrib.multiple = ilog2(__roundup_pow_of_two(nvec));
581
582 if (control & PCI_MSI_FLAGS_64BIT)
583 entry->mask_pos = dev->msi_cap + PCI_MSI_MASK_64;
584 else
585 entry->mask_pos = dev->msi_cap + PCI_MSI_MASK_32;
586
587 /* Save the initial mask status */
588 if (entry->msi_attrib.maskbit)
589 pci_read_config_dword(dev, entry->mask_pos, &entry->masked);
590
591out:
592 kfree(masks);
593 return entry;
594}
595
596static int msi_verify_entries(struct pci_dev *dev)
597{
598 struct msi_desc *entry;
599
600 for_each_pci_msi_entry(entry, dev) {
601 if (!dev->no_64bit_msi || !entry->msg.address_hi)
602 continue;
603 pci_err(dev, "Device has broken 64-bit MSI but arch"
604 " tried to assign one above 4G\n");
605 return -EIO;
606 }
607 return 0;
608}
609
610/**
611 * msi_capability_init - configure device's MSI capability structure
612 * @dev: pointer to the pci_dev data structure of MSI device function
613 * @nvec: number of interrupts to allocate
614 * @affd: description of automatic IRQ affinity assignments (may be %NULL)
615 *
616 * Setup the MSI capability structure of the device with the requested
617 * number of interrupts. A return value of zero indicates the successful
618 * setup of an entry with the new MSI IRQ. A negative return value indicates
619 * an error, and a positive return value indicates the number of interrupts
620 * which could have been allocated.
621 */
622static int msi_capability_init(struct pci_dev *dev, int nvec,
623 struct irq_affinity *affd)
624{
625 struct msi_desc *entry;
626 int ret;
627 unsigned mask;
628
629 pci_msi_set_enable(dev, 0); /* Disable MSI during set up */
630
631 entry = msi_setup_entry(dev, nvec, affd);
632 if (!entry)
633 return -ENOMEM;
634
635 /* All MSIs are unmasked by default; mask them all */
636 mask = msi_mask(entry->msi_attrib.multi_cap);
637 msi_mask_irq(entry, mask, mask);
638
639 list_add_tail(&entry->list, dev_to_msi_list(&dev->dev));
640
641 /* Configure MSI capability structure */
642 ret = pci_msi_setup_msi_irqs(dev, nvec, PCI_CAP_ID_MSI);
643 if (ret) {
644 msi_mask_irq(entry, mask, ~mask);
645 free_msi_irqs(dev);
646 return ret;
647 }
648
649 ret = msi_verify_entries(dev);
650 if (ret) {
651 msi_mask_irq(entry, mask, ~mask);
652 free_msi_irqs(dev);
653 return ret;
654 }
655
656 ret = populate_msi_sysfs(dev);
657 if (ret) {
658 msi_mask_irq(entry, mask, ~mask);
659 free_msi_irqs(dev);
660 return ret;
661 }
662
663 /* Set MSI enabled bits */
664 pci_intx_for_msi(dev, 0);
665 pci_msi_set_enable(dev, 1);
666 dev->msi_enabled = 1;
667
668 pcibios_free_irq(dev);
669 dev->irq = entry->irq;
670 return 0;
671}
672
673static void __iomem *msix_map_region(struct pci_dev *dev, unsigned nr_entries)
674{
675 resource_size_t phys_addr;
676 u32 table_offset;
677 unsigned long flags;
678 u8 bir;
679
680 pci_read_config_dword(dev, dev->msix_cap + PCI_MSIX_TABLE,
681 &table_offset);
682 bir = (u8)(table_offset & PCI_MSIX_TABLE_BIR);
683 flags = pci_resource_flags(dev, bir);
684 if (!flags || (flags & IORESOURCE_UNSET))
685 return NULL;
686
687 table_offset &= PCI_MSIX_TABLE_OFFSET;
688 phys_addr = pci_resource_start(dev, bir) + table_offset;
689
690 return ioremap_nocache(phys_addr, nr_entries * PCI_MSIX_ENTRY_SIZE);
691}
692
693static int msix_setup_entries(struct pci_dev *dev, void __iomem *base,
694 struct msix_entry *entries, int nvec,
695 struct irq_affinity *affd)
696{
697 struct irq_affinity_desc *curmsk, *masks = NULL;
698 struct msi_desc *entry;
699 int ret, i;
700 int vec_count = pci_msix_vec_count(dev);
701
702 if (affd)
703 masks = irq_create_affinity_masks(nvec, affd);
704
705 for (i = 0, curmsk = masks; i < nvec; i++) {
706 entry = alloc_msi_entry(&dev->dev, 1, curmsk);
707 if (!entry) {
708 if (!i)
709 iounmap(base);
710 else
711 free_msi_irqs(dev);
712 /* No enough memory. Don't try again */
713 ret = -ENOMEM;
714 goto out;
715 }
716
717 entry->msi_attrib.is_msix = 1;
718 entry->msi_attrib.is_64 = 1;
719 if (entries)
720 entry->msi_attrib.entry_nr = entries[i].entry;
721 else
722 entry->msi_attrib.entry_nr = i;
723
724 entry->msi_attrib.is_virtual =
725 entry->msi_attrib.entry_nr >= vec_count;
726
727 entry->msi_attrib.default_irq = dev->irq;
728 entry->mask_base = base;
729
730 list_add_tail(&entry->list, dev_to_msi_list(&dev->dev));
731 if (masks)
732 curmsk++;
733 }
734 ret = 0;
735out:
736 kfree(masks);
737 return ret;
738}
739
740static void msix_program_entries(struct pci_dev *dev,
741 struct msix_entry *entries)
742{
743 struct msi_desc *entry;
744 int i = 0;
745 void __iomem *desc_addr;
746
747 for_each_pci_msi_entry(entry, dev) {
748 if (entries)
749 entries[i++].vector = entry->irq;
750
751 desc_addr = pci_msix_desc_addr(entry);
752 if (desc_addr)
753 entry->masked = readl(desc_addr +
754 PCI_MSIX_ENTRY_VECTOR_CTRL);
755 else
756 entry->masked = 0;
757
758 msix_mask_irq(entry, 1);
759 }
760}
761
762/**
763 * msix_capability_init - configure device's MSI-X capability
764 * @dev: pointer to the pci_dev data structure of MSI-X device function
765 * @entries: pointer to an array of struct msix_entry entries
766 * @nvec: number of @entries
767 * @affd: Optional pointer to enable automatic affinity assignment
768 *
769 * Setup the MSI-X capability structure of device function with a
770 * single MSI-X IRQ. A return of zero indicates the successful setup of
771 * requested MSI-X entries with allocated IRQs or non-zero for otherwise.
772 **/
773static int msix_capability_init(struct pci_dev *dev, struct msix_entry *entries,
774 int nvec, struct irq_affinity *affd)
775{
776 int ret;
777 u16 control;
778 void __iomem *base;
779
780 /* Ensure MSI-X is disabled while it is set up */
781 pci_msix_clear_and_set_ctrl(dev, PCI_MSIX_FLAGS_ENABLE, 0);
782
783 pci_read_config_word(dev, dev->msix_cap + PCI_MSIX_FLAGS, &control);
784 /* Request & Map MSI-X table region */
785 base = msix_map_region(dev, msix_table_size(control));
786 if (!base)
787 return -ENOMEM;
788
789 ret = msix_setup_entries(dev, base, entries, nvec, affd);
790 if (ret)
791 return ret;
792
793 ret = pci_msi_setup_msi_irqs(dev, nvec, PCI_CAP_ID_MSIX);
794 if (ret)
795 goto out_avail;
796
797 /* Check if all MSI entries honor device restrictions */
798 ret = msi_verify_entries(dev);
799 if (ret)
800 goto out_free;
801
802 /*
803 * Some devices require MSI-X to be enabled before we can touch the
804 * MSI-X registers. We need to mask all the vectors to prevent
805 * interrupts coming in before they're fully set up.
806 */
807 pci_msix_clear_and_set_ctrl(dev, 0,
808 PCI_MSIX_FLAGS_MASKALL | PCI_MSIX_FLAGS_ENABLE);
809
810 msix_program_entries(dev, entries);
811
812 ret = populate_msi_sysfs(dev);
813 if (ret)
814 goto out_free;
815
816 /* Set MSI-X enabled bits and unmask the function */
817 pci_intx_for_msi(dev, 0);
818 dev->msix_enabled = 1;
819 pci_msix_clear_and_set_ctrl(dev, PCI_MSIX_FLAGS_MASKALL, 0);
820
821 pcibios_free_irq(dev);
822 return 0;
823
824out_avail:
825 if (ret < 0) {
826 /*
827 * If we had some success, report the number of IRQs
828 * we succeeded in setting up.
829 */
830 struct msi_desc *entry;
831 int avail = 0;
832
833 for_each_pci_msi_entry(entry, dev) {
834 if (entry->irq != 0)
835 avail++;
836 }
837 if (avail != 0)
838 ret = avail;
839 }
840
841out_free:
842 free_msi_irqs(dev);
843
844 return ret;
845}
846
847/**
848 * pci_msi_supported - check whether MSI may be enabled on a device
849 * @dev: pointer to the pci_dev data structure of MSI device function
850 * @nvec: how many MSIs have been requested?
851 *
852 * Look at global flags, the device itself, and its parent buses
853 * to determine if MSI/-X are supported for the device. If MSI/-X is
854 * supported return 1, else return 0.
855 **/
856static int pci_msi_supported(struct pci_dev *dev, int nvec)
857{
858 struct pci_bus *bus;
859
860 /* MSI must be globally enabled and supported by the device */
861 if (!pci_msi_enable)
862 return 0;
863
864 if (!dev || dev->no_msi || dev->current_state != PCI_D0)
865 return 0;
866
867 /*
868 * You can't ask to have 0 or less MSIs configured.
869 * a) it's stupid ..
870 * b) the list manipulation code assumes nvec >= 1.
871 */
872 if (nvec < 1)
873 return 0;
874
875 /*
876 * Any bridge which does NOT route MSI transactions from its
877 * secondary bus to its primary bus must set NO_MSI flag on
878 * the secondary pci_bus.
879 * We expect only arch-specific PCI host bus controller driver
880 * or quirks for specific PCI bridges to be setting NO_MSI.
881 */
882 for (bus = dev->bus; bus; bus = bus->parent)
883 if (bus->bus_flags & PCI_BUS_FLAGS_NO_MSI)
884 return 0;
885
886 return 1;
887}
888
889/**
890 * pci_msi_vec_count - Return the number of MSI vectors a device can send
891 * @dev: device to report about
892 *
893 * This function returns the number of MSI vectors a device requested via
894 * Multiple Message Capable register. It returns a negative errno if the
895 * device is not capable sending MSI interrupts. Otherwise, the call succeeds
896 * and returns a power of two, up to a maximum of 2^5 (32), according to the
897 * MSI specification.
898 **/
899int pci_msi_vec_count(struct pci_dev *dev)
900{
901 int ret;
902 u16 msgctl;
903
904 if (!dev->msi_cap)
905 return -EINVAL;
906
907 pci_read_config_word(dev, dev->msi_cap + PCI_MSI_FLAGS, &msgctl);
908 ret = 1 << ((msgctl & PCI_MSI_FLAGS_QMASK) >> 1);
909
910 return ret;
911}
912EXPORT_SYMBOL(pci_msi_vec_count);
913
914static void pci_msi_shutdown(struct pci_dev *dev)
915{
916 struct msi_desc *desc;
917 u32 mask;
918
919 if (!pci_msi_enable || !dev || !dev->msi_enabled)
920 return;
921
922 BUG_ON(list_empty(dev_to_msi_list(&dev->dev)));
923 desc = first_pci_msi_entry(dev);
924
925 pci_msi_set_enable(dev, 0);
926 pci_intx_for_msi(dev, 1);
927 dev->msi_enabled = 0;
928
929 /* Return the device with MSI unmasked as initial states */
930 mask = msi_mask(desc->msi_attrib.multi_cap);
931 /* Keep cached state to be restored */
932 __pci_msi_desc_mask_irq(desc, mask, ~mask);
933
934 /* Restore dev->irq to its default pin-assertion IRQ */
935 dev->irq = desc->msi_attrib.default_irq;
936 pcibios_alloc_irq(dev);
937}
938
939void pci_disable_msi(struct pci_dev *dev)
940{
941 if (!pci_msi_enable || !dev || !dev->msi_enabled)
942 return;
943
944 pci_msi_shutdown(dev);
945 free_msi_irqs(dev);
946}
947EXPORT_SYMBOL(pci_disable_msi);
948
949/**
950 * pci_msix_vec_count - return the number of device's MSI-X table entries
951 * @dev: pointer to the pci_dev data structure of MSI-X device function
952 * This function returns the number of device's MSI-X table entries and
953 * therefore the number of MSI-X vectors device is capable of sending.
954 * It returns a negative errno if the device is not capable of sending MSI-X
955 * interrupts.
956 **/
957int pci_msix_vec_count(struct pci_dev *dev)
958{
959 u16 control;
960
961 if (!dev->msix_cap)
962 return -EINVAL;
963
964 pci_read_config_word(dev, dev->msix_cap + PCI_MSIX_FLAGS, &control);
965 return msix_table_size(control);
966}
967EXPORT_SYMBOL(pci_msix_vec_count);
968
969static int __pci_enable_msix(struct pci_dev *dev, struct msix_entry *entries,
970 int nvec, struct irq_affinity *affd, int flags)
971{
972 int nr_entries;
973 int i, j;
974
975 if (!pci_msi_supported(dev, nvec))
976 return -EINVAL;
977
978 nr_entries = pci_msix_vec_count(dev);
979 if (nr_entries < 0)
980 return nr_entries;
981 if (nvec > nr_entries && !(flags & PCI_IRQ_VIRTUAL))
982 return nr_entries;
983
984 if (entries) {
985 /* Check for any invalid entries */
986 for (i = 0; i < nvec; i++) {
987 if (entries[i].entry >= nr_entries)
988 return -EINVAL; /* invalid entry */
989 for (j = i + 1; j < nvec; j++) {
990 if (entries[i].entry == entries[j].entry)
991 return -EINVAL; /* duplicate entry */
992 }
993 }
994 }
995
996 /* Check whether driver already requested for MSI IRQ */
997 if (dev->msi_enabled) {
998 pci_info(dev, "can't enable MSI-X (MSI IRQ already assigned)\n");
999 return -EINVAL;
1000 }
1001 return msix_capability_init(dev, entries, nvec, affd);
1002}
1003
1004static void pci_msix_shutdown(struct pci_dev *dev)
1005{
1006 struct msi_desc *entry;
1007
1008 if (!pci_msi_enable || !dev || !dev->msix_enabled)
1009 return;
1010
1011 if (pci_dev_is_disconnected(dev)) {
1012 dev->msix_enabled = 0;
1013 return;
1014 }
1015
1016 /* Return the device with MSI-X masked as initial states */
1017 for_each_pci_msi_entry(entry, dev) {
1018 /* Keep cached states to be restored */
1019 __pci_msix_desc_mask_irq(entry, 1);
1020 }
1021
1022 pci_msix_clear_and_set_ctrl(dev, PCI_MSIX_FLAGS_ENABLE, 0);
1023 pci_intx_for_msi(dev, 1);
1024 dev->msix_enabled = 0;
1025 pcibios_alloc_irq(dev);
1026}
1027
1028void pci_disable_msix(struct pci_dev *dev)
1029{
1030 if (!pci_msi_enable || !dev || !dev->msix_enabled)
1031 return;
1032
1033 pci_msix_shutdown(dev);
1034 free_msi_irqs(dev);
1035}
1036EXPORT_SYMBOL(pci_disable_msix);
1037
1038void pci_no_msi(void)
1039{
1040 pci_msi_enable = 0;
1041}
1042
1043/**
1044 * pci_msi_enabled - is MSI enabled?
1045 *
1046 * Returns true if MSI has not been disabled by the command-line option
1047 * pci=nomsi.
1048 **/
1049int pci_msi_enabled(void)
1050{
1051 return pci_msi_enable;
1052}
1053EXPORT_SYMBOL(pci_msi_enabled);
1054
1055static int __pci_enable_msi_range(struct pci_dev *dev, int minvec, int maxvec,
1056 struct irq_affinity *affd)
1057{
1058 int nvec;
1059 int rc;
1060
1061 if (!pci_msi_supported(dev, minvec))
1062 return -EINVAL;
1063
1064 /* Check whether driver already requested MSI-X IRQs */
1065 if (dev->msix_enabled) {
1066 pci_info(dev, "can't enable MSI (MSI-X already enabled)\n");
1067 return -EINVAL;
1068 }
1069
1070 if (maxvec < minvec)
1071 return -ERANGE;
1072
1073 if (WARN_ON_ONCE(dev->msi_enabled))
1074 return -EINVAL;
1075
1076 nvec = pci_msi_vec_count(dev);
1077 if (nvec < 0)
1078 return nvec;
1079 if (nvec < minvec)
1080 return -ENOSPC;
1081
1082 if (nvec > maxvec)
1083 nvec = maxvec;
1084
1085 for (;;) {
1086 if (affd) {
1087 nvec = irq_calc_affinity_vectors(minvec, nvec, affd);
1088 if (nvec < minvec)
1089 return -ENOSPC;
1090 }
1091
1092 rc = msi_capability_init(dev, nvec, affd);
1093 if (rc == 0)
1094 return nvec;
1095
1096 if (rc < 0)
1097 return rc;
1098 if (rc < minvec)
1099 return -ENOSPC;
1100
1101 nvec = rc;
1102 }
1103}
1104
1105/* deprecated, don't use */
1106int pci_enable_msi(struct pci_dev *dev)
1107{
1108 int rc = __pci_enable_msi_range(dev, 1, 1, NULL);
1109 if (rc < 0)
1110 return rc;
1111 return 0;
1112}
1113EXPORT_SYMBOL(pci_enable_msi);
1114
1115static int __pci_enable_msix_range(struct pci_dev *dev,
1116 struct msix_entry *entries, int minvec,
1117 int maxvec, struct irq_affinity *affd,
1118 int flags)
1119{
1120 int rc, nvec = maxvec;
1121
1122 if (maxvec < minvec)
1123 return -ERANGE;
1124
1125 if (WARN_ON_ONCE(dev->msix_enabled))
1126 return -EINVAL;
1127
1128 for (;;) {
1129 if (affd) {
1130 nvec = irq_calc_affinity_vectors(minvec, nvec, affd);
1131 if (nvec < minvec)
1132 return -ENOSPC;
1133 }
1134
1135 rc = __pci_enable_msix(dev, entries, nvec, affd, flags);
1136 if (rc == 0)
1137 return nvec;
1138
1139 if (rc < 0)
1140 return rc;
1141 if (rc < minvec)
1142 return -ENOSPC;
1143
1144 nvec = rc;
1145 }
1146}
1147
1148/**
1149 * pci_enable_msix_range - configure device's MSI-X capability structure
1150 * @dev: pointer to the pci_dev data structure of MSI-X device function
1151 * @entries: pointer to an array of MSI-X entries
1152 * @minvec: minimum number of MSI-X IRQs requested
1153 * @maxvec: maximum number of MSI-X IRQs requested
1154 *
1155 * Setup the MSI-X capability structure of device function with a maximum
1156 * possible number of interrupts in the range between @minvec and @maxvec
1157 * upon its software driver call to request for MSI-X mode enabled on its
1158 * hardware device function. It returns a negative errno if an error occurs.
1159 * If it succeeds, it returns the actual number of interrupts allocated and
1160 * indicates the successful configuration of MSI-X capability structure
1161 * with new allocated MSI-X interrupts.
1162 **/
1163int pci_enable_msix_range(struct pci_dev *dev, struct msix_entry *entries,
1164 int minvec, int maxvec)
1165{
1166 return __pci_enable_msix_range(dev, entries, minvec, maxvec, NULL, 0);
1167}
1168EXPORT_SYMBOL(pci_enable_msix_range);
1169
1170/**
1171 * pci_alloc_irq_vectors_affinity - allocate multiple IRQs for a device
1172 * @dev: PCI device to operate on
1173 * @min_vecs: minimum number of vectors required (must be >= 1)
1174 * @max_vecs: maximum (desired) number of vectors
1175 * @flags: flags or quirks for the allocation
1176 * @affd: optional description of the affinity requirements
1177 *
1178 * Allocate up to @max_vecs interrupt vectors for @dev, using MSI-X or MSI
1179 * vectors if available, and fall back to a single legacy vector
1180 * if neither is available. Return the number of vectors allocated,
1181 * (which might be smaller than @max_vecs) if successful, or a negative
1182 * error code on error. If less than @min_vecs interrupt vectors are
1183 * available for @dev the function will fail with -ENOSPC.
1184 *
1185 * To get the Linux IRQ number used for a vector that can be passed to
1186 * request_irq() use the pci_irq_vector() helper.
1187 */
1188int pci_alloc_irq_vectors_affinity(struct pci_dev *dev, unsigned int min_vecs,
1189 unsigned int max_vecs, unsigned int flags,
1190 struct irq_affinity *affd)
1191{
1192 struct irq_affinity msi_default_affd = {0};
1193 int msix_vecs = -ENOSPC;
1194 int msi_vecs = -ENOSPC;
1195
1196 if (flags & PCI_IRQ_AFFINITY) {
1197 if (!affd)
1198 affd = &msi_default_affd;
1199 } else {
1200 if (WARN_ON(affd))
1201 affd = NULL;
1202 }
1203
1204 if (flags & PCI_IRQ_MSIX) {
1205 msix_vecs = __pci_enable_msix_range(dev, NULL, min_vecs,
1206 max_vecs, affd, flags);
1207 if (msix_vecs > 0)
1208 return msix_vecs;
1209 }
1210
1211 if (flags & PCI_IRQ_MSI) {
1212 msi_vecs = __pci_enable_msi_range(dev, min_vecs, max_vecs,
1213 affd);
1214 if (msi_vecs > 0)
1215 return msi_vecs;
1216 }
1217
1218 /* use legacy IRQ if allowed */
1219 if (flags & PCI_IRQ_LEGACY) {
1220 if (min_vecs == 1 && dev->irq) {
1221 /*
1222 * Invoke the affinity spreading logic to ensure that
1223 * the device driver can adjust queue configuration
1224 * for the single interrupt case.
1225 */
1226 if (affd)
1227 irq_create_affinity_masks(1, affd);
1228 pci_intx(dev, 1);
1229 return 1;
1230 }
1231 }
1232
1233 if (msix_vecs == -ENOSPC)
1234 return -ENOSPC;
1235 return msi_vecs;
1236}
1237EXPORT_SYMBOL(pci_alloc_irq_vectors_affinity);
1238
1239/**
1240 * pci_free_irq_vectors - free previously allocated IRQs for a device
1241 * @dev: PCI device to operate on
1242 *
1243 * Undoes the allocations and enabling in pci_alloc_irq_vectors().
1244 */
1245void pci_free_irq_vectors(struct pci_dev *dev)
1246{
1247 pci_disable_msix(dev);
1248 pci_disable_msi(dev);
1249}
1250EXPORT_SYMBOL(pci_free_irq_vectors);
1251
1252/**
1253 * pci_irq_vector - return Linux IRQ number of a device vector
1254 * @dev: PCI device to operate on
1255 * @nr: device-relative interrupt vector index (0-based).
1256 */
1257int pci_irq_vector(struct pci_dev *dev, unsigned int nr)
1258{
1259 if (dev->msix_enabled) {
1260 struct msi_desc *entry;
1261 int i = 0;
1262
1263 for_each_pci_msi_entry(entry, dev) {
1264 if (i == nr)
1265 return entry->irq;
1266 i++;
1267 }
1268 WARN_ON_ONCE(1);
1269 return -EINVAL;
1270 }
1271
1272 if (dev->msi_enabled) {
1273 struct msi_desc *entry = first_pci_msi_entry(dev);
1274
1275 if (WARN_ON_ONCE(nr >= entry->nvec_used))
1276 return -EINVAL;
1277 } else {
1278 if (WARN_ON_ONCE(nr > 0))
1279 return -EINVAL;
1280 }
1281
1282 return dev->irq + nr;
1283}
1284EXPORT_SYMBOL(pci_irq_vector);
1285
1286/**
1287 * pci_irq_get_affinity - return the affinity of a particular MSI vector
1288 * @dev: PCI device to operate on
1289 * @nr: device-relative interrupt vector index (0-based).
1290 */
1291const struct cpumask *pci_irq_get_affinity(struct pci_dev *dev, int nr)
1292{
1293 if (dev->msix_enabled) {
1294 struct msi_desc *entry;
1295 int i = 0;
1296
1297 for_each_pci_msi_entry(entry, dev) {
1298 if (i == nr)
1299 return &entry->affinity->mask;
1300 i++;
1301 }
1302 WARN_ON_ONCE(1);
1303 return NULL;
1304 } else if (dev->msi_enabled) {
1305 struct msi_desc *entry = first_pci_msi_entry(dev);
1306
1307 if (WARN_ON_ONCE(!entry || !entry->affinity ||
1308 nr >= entry->nvec_used))
1309 return NULL;
1310
1311 return &entry->affinity[nr].mask;
1312 } else {
1313 return cpu_possible_mask;
1314 }
1315}
1316EXPORT_SYMBOL(pci_irq_get_affinity);
1317
1318/**
1319 * pci_irq_get_node - return the NUMA node of a particular MSI vector
1320 * @pdev: PCI device to operate on
1321 * @vec: device-relative interrupt vector index (0-based).
1322 */
1323int pci_irq_get_node(struct pci_dev *pdev, int vec)
1324{
1325 const struct cpumask *mask;
1326
1327 mask = pci_irq_get_affinity(pdev, vec);
1328 if (mask)
1329 return local_memory_node(cpu_to_node(cpumask_first(mask)));
1330 return dev_to_node(&pdev->dev);
1331}
1332EXPORT_SYMBOL(pci_irq_get_node);
1333
1334struct pci_dev *msi_desc_to_pci_dev(struct msi_desc *desc)
1335{
1336 return to_pci_dev(desc->dev);
1337}
1338EXPORT_SYMBOL(msi_desc_to_pci_dev);
1339
1340void *msi_desc_to_pci_sysdata(struct msi_desc *desc)
1341{
1342 struct pci_dev *dev = msi_desc_to_pci_dev(desc);
1343
1344 return dev->bus->sysdata;
1345}
1346EXPORT_SYMBOL_GPL(msi_desc_to_pci_sysdata);
1347
1348#ifdef CONFIG_PCI_MSI_IRQ_DOMAIN
1349/**
1350 * pci_msi_domain_write_msg - Helper to write MSI message to PCI config space
1351 * @irq_data: Pointer to interrupt data of the MSI interrupt
1352 * @msg: Pointer to the message
1353 */
1354void pci_msi_domain_write_msg(struct irq_data *irq_data, struct msi_msg *msg)
1355{
1356 struct msi_desc *desc = irq_data_get_msi_desc(irq_data);
1357
1358 /*
1359 * For MSI-X desc->irq is always equal to irq_data->irq. For
1360 * MSI only the first interrupt of MULTI MSI passes the test.
1361 */
1362 if (desc->irq == irq_data->irq)
1363 __pci_write_msi_msg(desc, msg);
1364}
1365
1366/**
1367 * pci_msi_domain_calc_hwirq - Generate a unique ID for an MSI source
1368 * @dev: Pointer to the PCI device
1369 * @desc: Pointer to the MSI descriptor
1370 *
1371 * The ID number is only used within the irqdomain.
1372 */
1373irq_hw_number_t pci_msi_domain_calc_hwirq(struct pci_dev *dev,
1374 struct msi_desc *desc)
1375{
1376 return (irq_hw_number_t)desc->msi_attrib.entry_nr |
1377 pci_dev_id(dev) << 11 |
1378 (pci_domain_nr(dev->bus) & 0xFFFFFFFF) << 27;
1379}
1380
1381static inline bool pci_msi_desc_is_multi_msi(struct msi_desc *desc)
1382{
1383 return !desc->msi_attrib.is_msix && desc->nvec_used > 1;
1384}
1385
1386/**
1387 * pci_msi_domain_check_cap - Verify that @domain supports the capabilities
1388 * for @dev
1389 * @domain: The interrupt domain to check
1390 * @info: The domain info for verification
1391 * @dev: The device to check
1392 *
1393 * Returns:
1394 * 0 if the functionality is supported
1395 * 1 if Multi MSI is requested, but the domain does not support it
1396 * -ENOTSUPP otherwise
1397 */
1398int pci_msi_domain_check_cap(struct irq_domain *domain,
1399 struct msi_domain_info *info, struct device *dev)
1400{
1401 struct msi_desc *desc = first_pci_msi_entry(to_pci_dev(dev));
1402
1403 /* Special handling to support __pci_enable_msi_range() */
1404 if (pci_msi_desc_is_multi_msi(desc) &&
1405 !(info->flags & MSI_FLAG_MULTI_PCI_MSI))
1406 return 1;
1407 else if (desc->msi_attrib.is_msix && !(info->flags & MSI_FLAG_PCI_MSIX))
1408 return -ENOTSUPP;
1409
1410 return 0;
1411}
1412
1413static int pci_msi_domain_handle_error(struct irq_domain *domain,
1414 struct msi_desc *desc, int error)
1415{
1416 /* Special handling to support __pci_enable_msi_range() */
1417 if (pci_msi_desc_is_multi_msi(desc) && error == -ENOSPC)
1418 return 1;
1419
1420 return error;
1421}
1422
1423#ifdef GENERIC_MSI_DOMAIN_OPS
1424static void pci_msi_domain_set_desc(msi_alloc_info_t *arg,
1425 struct msi_desc *desc)
1426{
1427 arg->desc = desc;
1428 arg->hwirq = pci_msi_domain_calc_hwirq(msi_desc_to_pci_dev(desc),
1429 desc);
1430}
1431#else
1432#define pci_msi_domain_set_desc NULL
1433#endif
1434
1435static struct msi_domain_ops pci_msi_domain_ops_default = {
1436 .set_desc = pci_msi_domain_set_desc,
1437 .msi_check = pci_msi_domain_check_cap,
1438 .handle_error = pci_msi_domain_handle_error,
1439};
1440
1441static void pci_msi_domain_update_dom_ops(struct msi_domain_info *info)
1442{
1443 struct msi_domain_ops *ops = info->ops;
1444
1445 if (ops == NULL) {
1446 info->ops = &pci_msi_domain_ops_default;
1447 } else {
1448 if (ops->set_desc == NULL)
1449 ops->set_desc = pci_msi_domain_set_desc;
1450 if (ops->msi_check == NULL)
1451 ops->msi_check = pci_msi_domain_check_cap;
1452 if (ops->handle_error == NULL)
1453 ops->handle_error = pci_msi_domain_handle_error;
1454 }
1455}
1456
1457static void pci_msi_domain_update_chip_ops(struct msi_domain_info *info)
1458{
1459 struct irq_chip *chip = info->chip;
1460
1461 BUG_ON(!chip);
1462 if (!chip->irq_write_msi_msg)
1463 chip->irq_write_msi_msg = pci_msi_domain_write_msg;
1464 if (!chip->irq_mask)
1465 chip->irq_mask = pci_msi_mask_irq;
1466 if (!chip->irq_unmask)
1467 chip->irq_unmask = pci_msi_unmask_irq;
1468}
1469
1470/**
1471 * pci_msi_create_irq_domain - Create a MSI interrupt domain
1472 * @fwnode: Optional fwnode of the interrupt controller
1473 * @info: MSI domain info
1474 * @parent: Parent irq domain
1475 *
1476 * Updates the domain and chip ops and creates a MSI interrupt domain.
1477 *
1478 * Returns:
1479 * A domain pointer or NULL in case of failure.
1480 */
1481struct irq_domain *pci_msi_create_irq_domain(struct fwnode_handle *fwnode,
1482 struct msi_domain_info *info,
1483 struct irq_domain *parent)
1484{
1485 struct irq_domain *domain;
1486
1487 if (WARN_ON(info->flags & MSI_FLAG_LEVEL_CAPABLE))
1488 info->flags &= ~MSI_FLAG_LEVEL_CAPABLE;
1489
1490 if (info->flags & MSI_FLAG_USE_DEF_DOM_OPS)
1491 pci_msi_domain_update_dom_ops(info);
1492 if (info->flags & MSI_FLAG_USE_DEF_CHIP_OPS)
1493 pci_msi_domain_update_chip_ops(info);
1494
1495 info->flags |= MSI_FLAG_ACTIVATE_EARLY;
1496 if (IS_ENABLED(CONFIG_GENERIC_IRQ_RESERVATION_MODE))
1497 info->flags |= MSI_FLAG_MUST_REACTIVATE;
1498
1499 /* PCI-MSI is oneshot-safe */
1500 info->chip->flags |= IRQCHIP_ONESHOT_SAFE;
1501
1502 domain = msi_create_irq_domain(fwnode, info, parent);
1503 if (!domain)
1504 return NULL;
1505
1506 irq_domain_update_bus_token(domain, DOMAIN_BUS_PCI_MSI);
1507 return domain;
1508}
1509EXPORT_SYMBOL_GPL(pci_msi_create_irq_domain);
1510
1511/*
1512 * Users of the generic MSI infrastructure expect a device to have a single ID,
1513 * so with DMA aliases we have to pick the least-worst compromise. Devices with
1514 * DMA phantom functions tend to still emit MSIs from the real function number,
1515 * so we ignore those and only consider topological aliases where either the
1516 * alias device or RID appears on a different bus number. We also make the
1517 * reasonable assumption that bridges are walked in an upstream direction (so
1518 * the last one seen wins), and the much braver assumption that the most likely
1519 * case is that of PCI->PCIe so we should always use the alias RID. This echoes
1520 * the logic from intel_irq_remapping's set_msi_sid(), which presumably works
1521 * well enough in practice; in the face of the horrible PCIe<->PCI-X conditions
1522 * for taking ownership all we can really do is close our eyes and hope...
1523 */
1524static int get_msi_id_cb(struct pci_dev *pdev, u16 alias, void *data)
1525{
1526 u32 *pa = data;
1527 u8 bus = PCI_BUS_NUM(*pa);
1528
1529 if (pdev->bus->number != bus || PCI_BUS_NUM(alias) != bus)
1530 *pa = alias;
1531
1532 return 0;
1533}
1534
1535/**
1536 * pci_msi_domain_get_msi_rid - Get the MSI requester id (RID)
1537 * @domain: The interrupt domain
1538 * @pdev: The PCI device.
1539 *
1540 * The RID for a device is formed from the alias, with a firmware
1541 * supplied mapping applied
1542 *
1543 * Returns: The RID.
1544 */
1545u32 pci_msi_domain_get_msi_rid(struct irq_domain *domain, struct pci_dev *pdev)
1546{
1547 struct device_node *of_node;
1548 u32 rid = pci_dev_id(pdev);
1549
1550 pci_for_each_dma_alias(pdev, get_msi_id_cb, &rid);
1551
1552 of_node = irq_domain_get_of_node(domain);
1553 rid = of_node ? of_msi_map_rid(&pdev->dev, of_node, rid) :
1554 iort_msi_map_rid(&pdev->dev, rid);
1555
1556 return rid;
1557}
1558
1559/**
1560 * pci_msi_get_device_domain - Get the MSI domain for a given PCI device
1561 * @pdev: The PCI device
1562 *
1563 * Use the firmware data to find a device-specific MSI domain
1564 * (i.e. not one that is set as a default).
1565 *
1566 * Returns: The corresponding MSI domain or NULL if none has been found.
1567 */
1568struct irq_domain *pci_msi_get_device_domain(struct pci_dev *pdev)
1569{
1570 struct irq_domain *dom;
1571 u32 rid = pci_dev_id(pdev);
1572
1573 pci_for_each_dma_alias(pdev, get_msi_id_cb, &rid);
1574 dom = of_msi_map_get_device_domain(&pdev->dev, rid);
1575 if (!dom)
1576 dom = iort_get_device_domain(&pdev->dev, rid);
1577 return dom;
1578}
1579#endif /* CONFIG_PCI_MSI_IRQ_DOMAIN */
1// SPDX-License-Identifier: GPL-2.0
2/*
3 * 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
29#ifdef CONFIG_PCI_MSI
30
31static int pci_msi_enable = 1;
32int pci_msi_ignore_mask;
33
34#define msix_table_size(flags) ((flags & PCI_MSIX_FLAGS_QSIZE) + 1)
35
36#ifdef CONFIG_PCI_MSI_IRQ_DOMAIN
37static int pci_msi_setup_msi_irqs(struct pci_dev *dev, int nvec, int type)
38{
39 struct irq_domain *domain;
40
41 domain = dev_get_msi_domain(&dev->dev);
42 if (domain && irq_domain_is_hierarchy(domain))
43 return msi_domain_alloc_irqs(domain, &dev->dev, nvec);
44
45 return arch_setup_msi_irqs(dev, nvec, type);
46}
47
48static void pci_msi_teardown_msi_irqs(struct pci_dev *dev)
49{
50 struct irq_domain *domain;
51
52 domain = dev_get_msi_domain(&dev->dev);
53 if (domain && irq_domain_is_hierarchy(domain))
54 msi_domain_free_irqs(domain, &dev->dev);
55 else
56 arch_teardown_msi_irqs(dev);
57}
58#else
59#define pci_msi_setup_msi_irqs arch_setup_msi_irqs
60#define pci_msi_teardown_msi_irqs arch_teardown_msi_irqs
61#endif
62
63#ifdef CONFIG_PCI_MSI_ARCH_FALLBACKS
64/* Arch hooks */
65int __weak arch_setup_msi_irq(struct pci_dev *dev, struct msi_desc *desc)
66{
67 return -EINVAL;
68}
69
70void __weak arch_teardown_msi_irq(unsigned int irq)
71{
72}
73
74int __weak arch_setup_msi_irqs(struct pci_dev *dev, int nvec, int type)
75{
76 struct msi_desc *entry;
77 int ret;
78
79 /*
80 * If an architecture wants to support multiple MSI, it needs to
81 * override arch_setup_msi_irqs()
82 */
83 if (type == PCI_CAP_ID_MSI && nvec > 1)
84 return 1;
85
86 for_each_pci_msi_entry(entry, dev) {
87 ret = arch_setup_msi_irq(dev, entry);
88 if (ret < 0)
89 return ret;
90 if (ret > 0)
91 return -ENOSPC;
92 }
93
94 return 0;
95}
96
97void __weak arch_teardown_msi_irqs(struct pci_dev *dev)
98{
99 int i;
100 struct msi_desc *entry;
101
102 for_each_pci_msi_entry(entry, dev)
103 if (entry->irq)
104 for (i = 0; i < entry->nvec_used; i++)
105 arch_teardown_msi_irq(entry->irq + i);
106}
107#endif /* CONFIG_PCI_MSI_ARCH_FALLBACKS */
108
109static void default_restore_msi_irq(struct pci_dev *dev, int irq)
110{
111 struct msi_desc *entry;
112
113 entry = NULL;
114 if (dev->msix_enabled) {
115 for_each_pci_msi_entry(entry, dev) {
116 if (irq == entry->irq)
117 break;
118 }
119 } else if (dev->msi_enabled) {
120 entry = irq_get_msi_desc(irq);
121 }
122
123 if (entry)
124 __pci_write_msi_msg(entry, &entry->msg);
125}
126
127void __weak arch_restore_msi_irqs(struct pci_dev *dev)
128{
129 return default_restore_msi_irqs(dev);
130}
131
132static inline __attribute_const__ u32 msi_mask(unsigned x)
133{
134 /* Don't shift by >= width of type */
135 if (x >= 5)
136 return 0xffffffff;
137 return (1 << (1 << x)) - 1;
138}
139
140/*
141 * PCI 2.3 does not specify mask bits for each MSI interrupt. Attempting to
142 * mask all MSI interrupts by clearing the MSI enable bit does not work
143 * reliably as devices without an INTx disable bit will then generate a
144 * level IRQ which will never be cleared.
145 */
146void __pci_msi_desc_mask_irq(struct msi_desc *desc, u32 mask, u32 flag)
147{
148 raw_spinlock_t *lock = &desc->dev->msi_lock;
149 unsigned long flags;
150
151 if (pci_msi_ignore_mask || !desc->msi_attrib.maskbit)
152 return;
153
154 raw_spin_lock_irqsave(lock, flags);
155 desc->masked &= ~mask;
156 desc->masked |= flag;
157 pci_write_config_dword(msi_desc_to_pci_dev(desc), desc->mask_pos,
158 desc->masked);
159 raw_spin_unlock_irqrestore(lock, flags);
160}
161
162static void msi_mask_irq(struct msi_desc *desc, u32 mask, u32 flag)
163{
164 __pci_msi_desc_mask_irq(desc, mask, flag);
165}
166
167static void __iomem *pci_msix_desc_addr(struct msi_desc *desc)
168{
169 if (desc->msi_attrib.is_virtual)
170 return NULL;
171
172 return desc->mask_base +
173 desc->msi_attrib.entry_nr * PCI_MSIX_ENTRY_SIZE;
174}
175
176/*
177 * This internal function does not flush PCI writes to the device.
178 * All users must ensure that they read from the device before either
179 * assuming that the device state is up to date, or returning out of this
180 * file. This saves a few milliseconds when initialising devices with lots
181 * of MSI-X interrupts.
182 */
183u32 __pci_msix_desc_mask_irq(struct msi_desc *desc, u32 flag)
184{
185 u32 mask_bits = desc->masked;
186 void __iomem *desc_addr;
187
188 if (pci_msi_ignore_mask)
189 return 0;
190
191 desc_addr = pci_msix_desc_addr(desc);
192 if (!desc_addr)
193 return 0;
194
195 mask_bits &= ~PCI_MSIX_ENTRY_CTRL_MASKBIT;
196 if (flag & PCI_MSIX_ENTRY_CTRL_MASKBIT)
197 mask_bits |= PCI_MSIX_ENTRY_CTRL_MASKBIT;
198
199 writel(mask_bits, desc_addr + PCI_MSIX_ENTRY_VECTOR_CTRL);
200
201 return mask_bits;
202}
203
204static void msix_mask_irq(struct msi_desc *desc, u32 flag)
205{
206 desc->masked = __pci_msix_desc_mask_irq(desc, flag);
207}
208
209static void msi_set_mask_bit(struct irq_data *data, u32 flag)
210{
211 struct msi_desc *desc = irq_data_get_msi_desc(data);
212
213 if (desc->msi_attrib.is_msix) {
214 msix_mask_irq(desc, flag);
215 readl(desc->mask_base); /* Flush write to device */
216 } else {
217 unsigned offset = data->irq - desc->irq;
218 msi_mask_irq(desc, 1 << offset, flag << offset);
219 }
220}
221
222/**
223 * pci_msi_mask_irq - Generic IRQ chip callback to mask PCI/MSI interrupts
224 * @data: pointer to irqdata associated to that interrupt
225 */
226void pci_msi_mask_irq(struct irq_data *data)
227{
228 msi_set_mask_bit(data, 1);
229}
230EXPORT_SYMBOL_GPL(pci_msi_mask_irq);
231
232/**
233 * pci_msi_unmask_irq - Generic IRQ chip callback to unmask PCI/MSI interrupts
234 * @data: pointer to irqdata associated to that interrupt
235 */
236void pci_msi_unmask_irq(struct irq_data *data)
237{
238 msi_set_mask_bit(data, 0);
239}
240EXPORT_SYMBOL_GPL(pci_msi_unmask_irq);
241
242void default_restore_msi_irqs(struct pci_dev *dev)
243{
244 struct msi_desc *entry;
245
246 for_each_pci_msi_entry(entry, dev)
247 default_restore_msi_irq(dev, entry->irq);
248}
249
250void __pci_read_msi_msg(struct msi_desc *entry, struct msi_msg *msg)
251{
252 struct pci_dev *dev = msi_desc_to_pci_dev(entry);
253
254 BUG_ON(dev->current_state != PCI_D0);
255
256 if (entry->msi_attrib.is_msix) {
257 void __iomem *base = pci_msix_desc_addr(entry);
258
259 if (!base) {
260 WARN_ON(1);
261 return;
262 }
263
264 msg->address_lo = readl(base + PCI_MSIX_ENTRY_LOWER_ADDR);
265 msg->address_hi = readl(base + PCI_MSIX_ENTRY_UPPER_ADDR);
266 msg->data = readl(base + PCI_MSIX_ENTRY_DATA);
267 } else {
268 int pos = dev->msi_cap;
269 u16 data;
270
271 pci_read_config_dword(dev, pos + PCI_MSI_ADDRESS_LO,
272 &msg->address_lo);
273 if (entry->msi_attrib.is_64) {
274 pci_read_config_dword(dev, pos + PCI_MSI_ADDRESS_HI,
275 &msg->address_hi);
276 pci_read_config_word(dev, pos + PCI_MSI_DATA_64, &data);
277 } else {
278 msg->address_hi = 0;
279 pci_read_config_word(dev, pos + PCI_MSI_DATA_32, &data);
280 }
281 msg->data = data;
282 }
283}
284
285void __pci_write_msi_msg(struct msi_desc *entry, struct msi_msg *msg)
286{
287 struct pci_dev *dev = msi_desc_to_pci_dev(entry);
288
289 if (dev->current_state != PCI_D0 || pci_dev_is_disconnected(dev)) {
290 /* Don't touch the hardware now */
291 } else if (entry->msi_attrib.is_msix) {
292 void __iomem *base = pci_msix_desc_addr(entry);
293 bool unmasked = !(entry->masked & PCI_MSIX_ENTRY_CTRL_MASKBIT);
294
295 if (!base)
296 goto skip;
297
298 /*
299 * The specification mandates that the entry is masked
300 * when the message is modified:
301 *
302 * "If software changes the Address or Data value of an
303 * entry while the entry is unmasked, the result is
304 * undefined."
305 */
306 if (unmasked)
307 __pci_msix_desc_mask_irq(entry, PCI_MSIX_ENTRY_CTRL_MASKBIT);
308
309 writel(msg->address_lo, base + PCI_MSIX_ENTRY_LOWER_ADDR);
310 writel(msg->address_hi, base + PCI_MSIX_ENTRY_UPPER_ADDR);
311 writel(msg->data, base + PCI_MSIX_ENTRY_DATA);
312
313 if (unmasked)
314 __pci_msix_desc_mask_irq(entry, 0);
315
316 /* Ensure that the writes are visible in the device */
317 readl(base + PCI_MSIX_ENTRY_DATA);
318 } else {
319 int pos = dev->msi_cap;
320 u16 msgctl;
321
322 pci_read_config_word(dev, pos + PCI_MSI_FLAGS, &msgctl);
323 msgctl &= ~PCI_MSI_FLAGS_QSIZE;
324 msgctl |= entry->msi_attrib.multiple << 4;
325 pci_write_config_word(dev, pos + PCI_MSI_FLAGS, msgctl);
326
327 pci_write_config_dword(dev, pos + PCI_MSI_ADDRESS_LO,
328 msg->address_lo);
329 if (entry->msi_attrib.is_64) {
330 pci_write_config_dword(dev, pos + PCI_MSI_ADDRESS_HI,
331 msg->address_hi);
332 pci_write_config_word(dev, pos + PCI_MSI_DATA_64,
333 msg->data);
334 } else {
335 pci_write_config_word(dev, pos + PCI_MSI_DATA_32,
336 msg->data);
337 }
338 /* Ensure that the writes are visible in the device */
339 pci_read_config_word(dev, pos + PCI_MSI_FLAGS, &msgctl);
340 }
341
342skip:
343 entry->msg = *msg;
344
345 if (entry->write_msi_msg)
346 entry->write_msi_msg(entry, entry->write_msi_msg_data);
347
348}
349
350void pci_write_msi_msg(unsigned int irq, struct msi_msg *msg)
351{
352 struct msi_desc *entry = irq_get_msi_desc(irq);
353
354 __pci_write_msi_msg(entry, msg);
355}
356EXPORT_SYMBOL_GPL(pci_write_msi_msg);
357
358static void free_msi_irqs(struct pci_dev *dev)
359{
360 struct list_head *msi_list = dev_to_msi_list(&dev->dev);
361 struct msi_desc *entry, *tmp;
362 struct attribute **msi_attrs;
363 struct device_attribute *dev_attr;
364 int i, count = 0;
365
366 for_each_pci_msi_entry(entry, dev)
367 if (entry->irq)
368 for (i = 0; i < entry->nvec_used; i++)
369 BUG_ON(irq_has_action(entry->irq + i));
370
371 pci_msi_teardown_msi_irqs(dev);
372
373 list_for_each_entry_safe(entry, tmp, msi_list, list) {
374 if (entry->msi_attrib.is_msix) {
375 if (list_is_last(&entry->list, msi_list))
376 iounmap(entry->mask_base);
377 }
378
379 list_del(&entry->list);
380 free_msi_entry(entry);
381 }
382
383 if (dev->msi_irq_groups) {
384 sysfs_remove_groups(&dev->dev.kobj, dev->msi_irq_groups);
385 msi_attrs = dev->msi_irq_groups[0]->attrs;
386 while (msi_attrs[count]) {
387 dev_attr = container_of(msi_attrs[count],
388 struct device_attribute, attr);
389 kfree(dev_attr->attr.name);
390 kfree(dev_attr);
391 ++count;
392 }
393 kfree(msi_attrs);
394 kfree(dev->msi_irq_groups[0]);
395 kfree(dev->msi_irq_groups);
396 dev->msi_irq_groups = NULL;
397 }
398}
399
400static void pci_intx_for_msi(struct pci_dev *dev, int enable)
401{
402 if (!(dev->dev_flags & PCI_DEV_FLAGS_MSI_INTX_DISABLE_BUG))
403 pci_intx(dev, enable);
404}
405
406static void pci_msi_set_enable(struct pci_dev *dev, int enable)
407{
408 u16 control;
409
410 pci_read_config_word(dev, dev->msi_cap + PCI_MSI_FLAGS, &control);
411 control &= ~PCI_MSI_FLAGS_ENABLE;
412 if (enable)
413 control |= PCI_MSI_FLAGS_ENABLE;
414 pci_write_config_word(dev, dev->msi_cap + PCI_MSI_FLAGS, control);
415}
416
417static void __pci_restore_msi_state(struct pci_dev *dev)
418{
419 u16 control;
420 struct msi_desc *entry;
421
422 if (!dev->msi_enabled)
423 return;
424
425 entry = irq_get_msi_desc(dev->irq);
426
427 pci_intx_for_msi(dev, 0);
428 pci_msi_set_enable(dev, 0);
429 arch_restore_msi_irqs(dev);
430
431 pci_read_config_word(dev, dev->msi_cap + PCI_MSI_FLAGS, &control);
432 msi_mask_irq(entry, msi_mask(entry->msi_attrib.multi_cap),
433 entry->masked);
434 control &= ~PCI_MSI_FLAGS_QSIZE;
435 control |= (entry->msi_attrib.multiple << 4) | PCI_MSI_FLAGS_ENABLE;
436 pci_write_config_word(dev, dev->msi_cap + PCI_MSI_FLAGS, control);
437}
438
439static void pci_msix_clear_and_set_ctrl(struct pci_dev *dev, u16 clear, u16 set)
440{
441 u16 ctrl;
442
443 pci_read_config_word(dev, dev->msix_cap + PCI_MSIX_FLAGS, &ctrl);
444 ctrl &= ~clear;
445 ctrl |= set;
446 pci_write_config_word(dev, dev->msix_cap + PCI_MSIX_FLAGS, ctrl);
447}
448
449static void __pci_restore_msix_state(struct pci_dev *dev)
450{
451 struct msi_desc *entry;
452
453 if (!dev->msix_enabled)
454 return;
455 BUG_ON(list_empty(dev_to_msi_list(&dev->dev)));
456
457 /* route the table */
458 pci_intx_for_msi(dev, 0);
459 pci_msix_clear_and_set_ctrl(dev, 0,
460 PCI_MSIX_FLAGS_ENABLE | PCI_MSIX_FLAGS_MASKALL);
461
462 arch_restore_msi_irqs(dev);
463 for_each_pci_msi_entry(entry, dev)
464 msix_mask_irq(entry, entry->masked);
465
466 pci_msix_clear_and_set_ctrl(dev, PCI_MSIX_FLAGS_MASKALL, 0);
467}
468
469void pci_restore_msi_state(struct pci_dev *dev)
470{
471 __pci_restore_msi_state(dev);
472 __pci_restore_msix_state(dev);
473}
474EXPORT_SYMBOL_GPL(pci_restore_msi_state);
475
476static ssize_t msi_mode_show(struct device *dev, struct device_attribute *attr,
477 char *buf)
478{
479 struct msi_desc *entry;
480 unsigned long irq;
481 int retval;
482
483 retval = kstrtoul(attr->attr.name, 10, &irq);
484 if (retval)
485 return retval;
486
487 entry = irq_get_msi_desc(irq);
488 if (!entry)
489 return -ENODEV;
490
491 return sysfs_emit(buf, "%s\n",
492 entry->msi_attrib.is_msix ? "msix" : "msi");
493}
494
495static int populate_msi_sysfs(struct pci_dev *pdev)
496{
497 struct attribute **msi_attrs;
498 struct attribute *msi_attr;
499 struct device_attribute *msi_dev_attr;
500 struct attribute_group *msi_irq_group;
501 const struct attribute_group **msi_irq_groups;
502 struct msi_desc *entry;
503 int ret = -ENOMEM;
504 int num_msi = 0;
505 int count = 0;
506 int i;
507
508 /* Determine how many msi entries we have */
509 for_each_pci_msi_entry(entry, pdev)
510 num_msi += entry->nvec_used;
511 if (!num_msi)
512 return 0;
513
514 /* Dynamically create the MSI attributes for the PCI device */
515 msi_attrs = kcalloc(num_msi + 1, sizeof(void *), GFP_KERNEL);
516 if (!msi_attrs)
517 return -ENOMEM;
518 for_each_pci_msi_entry(entry, pdev) {
519 for (i = 0; i < entry->nvec_used; i++) {
520 msi_dev_attr = kzalloc(sizeof(*msi_dev_attr), GFP_KERNEL);
521 if (!msi_dev_attr)
522 goto error_attrs;
523 msi_attrs[count] = &msi_dev_attr->attr;
524
525 sysfs_attr_init(&msi_dev_attr->attr);
526 msi_dev_attr->attr.name = kasprintf(GFP_KERNEL, "%d",
527 entry->irq + i);
528 if (!msi_dev_attr->attr.name)
529 goto error_attrs;
530 msi_dev_attr->attr.mode = S_IRUGO;
531 msi_dev_attr->show = msi_mode_show;
532 ++count;
533 }
534 }
535
536 msi_irq_group = kzalloc(sizeof(*msi_irq_group), GFP_KERNEL);
537 if (!msi_irq_group)
538 goto error_attrs;
539 msi_irq_group->name = "msi_irqs";
540 msi_irq_group->attrs = msi_attrs;
541
542 msi_irq_groups = kcalloc(2, sizeof(void *), GFP_KERNEL);
543 if (!msi_irq_groups)
544 goto error_irq_group;
545 msi_irq_groups[0] = msi_irq_group;
546
547 ret = sysfs_create_groups(&pdev->dev.kobj, msi_irq_groups);
548 if (ret)
549 goto error_irq_groups;
550 pdev->msi_irq_groups = msi_irq_groups;
551
552 return 0;
553
554error_irq_groups:
555 kfree(msi_irq_groups);
556error_irq_group:
557 kfree(msi_irq_group);
558error_attrs:
559 count = 0;
560 msi_attr = msi_attrs[count];
561 while (msi_attr) {
562 msi_dev_attr = container_of(msi_attr, struct device_attribute, attr);
563 kfree(msi_attr->name);
564 kfree(msi_dev_attr);
565 ++count;
566 msi_attr = msi_attrs[count];
567 }
568 kfree(msi_attrs);
569 return ret;
570}
571
572static struct msi_desc *
573msi_setup_entry(struct pci_dev *dev, int nvec, struct irq_affinity *affd)
574{
575 struct irq_affinity_desc *masks = NULL;
576 struct msi_desc *entry;
577 u16 control;
578
579 if (affd)
580 masks = irq_create_affinity_masks(nvec, affd);
581
582 /* MSI Entry Initialization */
583 entry = alloc_msi_entry(&dev->dev, nvec, masks);
584 if (!entry)
585 goto out;
586
587 pci_read_config_word(dev, dev->msi_cap + PCI_MSI_FLAGS, &control);
588
589 entry->msi_attrib.is_msix = 0;
590 entry->msi_attrib.is_64 = !!(control & PCI_MSI_FLAGS_64BIT);
591 entry->msi_attrib.is_virtual = 0;
592 entry->msi_attrib.entry_nr = 0;
593 entry->msi_attrib.maskbit = !!(control & PCI_MSI_FLAGS_MASKBIT);
594 entry->msi_attrib.default_irq = dev->irq; /* Save IOAPIC IRQ */
595 entry->msi_attrib.multi_cap = (control & PCI_MSI_FLAGS_QMASK) >> 1;
596 entry->msi_attrib.multiple = ilog2(__roundup_pow_of_two(nvec));
597
598 if (control & PCI_MSI_FLAGS_64BIT)
599 entry->mask_pos = dev->msi_cap + PCI_MSI_MASK_64;
600 else
601 entry->mask_pos = dev->msi_cap + PCI_MSI_MASK_32;
602
603 /* Save the initial mask status */
604 if (entry->msi_attrib.maskbit)
605 pci_read_config_dword(dev, entry->mask_pos, &entry->masked);
606
607out:
608 kfree(masks);
609 return entry;
610}
611
612static int msi_verify_entries(struct pci_dev *dev)
613{
614 struct msi_desc *entry;
615
616 for_each_pci_msi_entry(entry, dev) {
617 if (entry->msg.address_hi && dev->no_64bit_msi) {
618 pci_err(dev, "arch assigned 64-bit MSI address %#x%08x but device only supports 32 bits\n",
619 entry->msg.address_hi, entry->msg.address_lo);
620 return -EIO;
621 }
622 }
623 return 0;
624}
625
626/**
627 * msi_capability_init - configure device's MSI capability structure
628 * @dev: pointer to the pci_dev data structure of MSI device function
629 * @nvec: number of interrupts to allocate
630 * @affd: description of automatic IRQ affinity assignments (may be %NULL)
631 *
632 * Setup the MSI capability structure of the device with the requested
633 * number of interrupts. A return value of zero indicates the successful
634 * setup of an entry with the new MSI IRQ. A negative return value indicates
635 * an error, and a positive return value indicates the number of interrupts
636 * which could have been allocated.
637 */
638static int msi_capability_init(struct pci_dev *dev, int nvec,
639 struct irq_affinity *affd)
640{
641 struct msi_desc *entry;
642 int ret;
643 unsigned mask;
644
645 pci_msi_set_enable(dev, 0); /* Disable MSI during set up */
646
647 entry = msi_setup_entry(dev, nvec, affd);
648 if (!entry)
649 return -ENOMEM;
650
651 /* All MSIs are unmasked by default; mask them all */
652 mask = msi_mask(entry->msi_attrib.multi_cap);
653 msi_mask_irq(entry, mask, mask);
654
655 list_add_tail(&entry->list, dev_to_msi_list(&dev->dev));
656
657 /* Configure MSI capability structure */
658 ret = pci_msi_setup_msi_irqs(dev, nvec, PCI_CAP_ID_MSI);
659 if (ret) {
660 msi_mask_irq(entry, mask, 0);
661 free_msi_irqs(dev);
662 return ret;
663 }
664
665 ret = msi_verify_entries(dev);
666 if (ret) {
667 msi_mask_irq(entry, mask, 0);
668 free_msi_irqs(dev);
669 return ret;
670 }
671
672 ret = populate_msi_sysfs(dev);
673 if (ret) {
674 msi_mask_irq(entry, mask, 0);
675 free_msi_irqs(dev);
676 return ret;
677 }
678
679 /* Set MSI enabled bits */
680 pci_intx_for_msi(dev, 0);
681 pci_msi_set_enable(dev, 1);
682 dev->msi_enabled = 1;
683
684 pcibios_free_irq(dev);
685 dev->irq = entry->irq;
686 return 0;
687}
688
689static void __iomem *msix_map_region(struct pci_dev *dev, unsigned nr_entries)
690{
691 resource_size_t phys_addr;
692 u32 table_offset;
693 unsigned long flags;
694 u8 bir;
695
696 pci_read_config_dword(dev, dev->msix_cap + PCI_MSIX_TABLE,
697 &table_offset);
698 bir = (u8)(table_offset & PCI_MSIX_TABLE_BIR);
699 flags = pci_resource_flags(dev, bir);
700 if (!flags || (flags & IORESOURCE_UNSET))
701 return NULL;
702
703 table_offset &= PCI_MSIX_TABLE_OFFSET;
704 phys_addr = pci_resource_start(dev, bir) + table_offset;
705
706 return ioremap(phys_addr, nr_entries * PCI_MSIX_ENTRY_SIZE);
707}
708
709static int msix_setup_entries(struct pci_dev *dev, void __iomem *base,
710 struct msix_entry *entries, int nvec,
711 struct irq_affinity *affd)
712{
713 struct irq_affinity_desc *curmsk, *masks = NULL;
714 struct msi_desc *entry;
715 void __iomem *addr;
716 int ret, i;
717 int vec_count = pci_msix_vec_count(dev);
718
719 if (affd)
720 masks = irq_create_affinity_masks(nvec, affd);
721
722 for (i = 0, curmsk = masks; i < nvec; i++) {
723 entry = alloc_msi_entry(&dev->dev, 1, curmsk);
724 if (!entry) {
725 if (!i)
726 iounmap(base);
727 else
728 free_msi_irqs(dev);
729 /* No enough memory. Don't try again */
730 ret = -ENOMEM;
731 goto out;
732 }
733
734 entry->msi_attrib.is_msix = 1;
735 entry->msi_attrib.is_64 = 1;
736
737 if (entries)
738 entry->msi_attrib.entry_nr = entries[i].entry;
739 else
740 entry->msi_attrib.entry_nr = i;
741
742 entry->msi_attrib.is_virtual =
743 entry->msi_attrib.entry_nr >= vec_count;
744
745 entry->msi_attrib.default_irq = dev->irq;
746 entry->mask_base = base;
747
748 addr = pci_msix_desc_addr(entry);
749 if (addr)
750 entry->masked = readl(addr + PCI_MSIX_ENTRY_VECTOR_CTRL);
751
752 list_add_tail(&entry->list, dev_to_msi_list(&dev->dev));
753 if (masks)
754 curmsk++;
755 }
756 ret = 0;
757out:
758 kfree(masks);
759 return ret;
760}
761
762static void msix_update_entries(struct pci_dev *dev, struct msix_entry *entries)
763{
764 struct msi_desc *entry;
765
766 for_each_pci_msi_entry(entry, dev) {
767 if (entries) {
768 entries->vector = entry->irq;
769 entries++;
770 }
771 }
772}
773
774static void msix_mask_all(void __iomem *base, int tsize)
775{
776 u32 ctrl = PCI_MSIX_ENTRY_CTRL_MASKBIT;
777 int i;
778
779 if (pci_msi_ignore_mask)
780 return;
781
782 for (i = 0; i < tsize; i++, base += PCI_MSIX_ENTRY_SIZE)
783 writel(ctrl, base + PCI_MSIX_ENTRY_VECTOR_CTRL);
784}
785
786/**
787 * msix_capability_init - configure device's MSI-X capability
788 * @dev: pointer to the pci_dev data structure of MSI-X device function
789 * @entries: pointer to an array of struct msix_entry entries
790 * @nvec: number of @entries
791 * @affd: Optional pointer to enable automatic affinity assignment
792 *
793 * Setup the MSI-X capability structure of device function with a
794 * single MSI-X IRQ. A return of zero indicates the successful setup of
795 * requested MSI-X entries with allocated IRQs or non-zero for otherwise.
796 **/
797static int msix_capability_init(struct pci_dev *dev, struct msix_entry *entries,
798 int nvec, struct irq_affinity *affd)
799{
800 void __iomem *base;
801 int ret, tsize;
802 u16 control;
803
804 /*
805 * Some devices require MSI-X to be enabled before the MSI-X
806 * registers can be accessed. Mask all the vectors to prevent
807 * interrupts coming in before they're fully set up.
808 */
809 pci_msix_clear_and_set_ctrl(dev, 0, PCI_MSIX_FLAGS_MASKALL |
810 PCI_MSIX_FLAGS_ENABLE);
811
812 pci_read_config_word(dev, dev->msix_cap + PCI_MSIX_FLAGS, &control);
813 /* Request & Map MSI-X table region */
814 tsize = msix_table_size(control);
815 base = msix_map_region(dev, tsize);
816 if (!base) {
817 ret = -ENOMEM;
818 goto out_disable;
819 }
820
821 /* Ensure that all table entries are masked. */
822 msix_mask_all(base, tsize);
823
824 ret = msix_setup_entries(dev, base, entries, nvec, affd);
825 if (ret)
826 goto out_disable;
827
828 ret = pci_msi_setup_msi_irqs(dev, nvec, PCI_CAP_ID_MSIX);
829 if (ret)
830 goto out_avail;
831
832 /* Check if all MSI entries honor device restrictions */
833 ret = msi_verify_entries(dev);
834 if (ret)
835 goto out_free;
836
837 msix_update_entries(dev, entries);
838
839 ret = populate_msi_sysfs(dev);
840 if (ret)
841 goto out_free;
842
843 /* Set MSI-X enabled bits and unmask the function */
844 pci_intx_for_msi(dev, 0);
845 dev->msix_enabled = 1;
846 pci_msix_clear_and_set_ctrl(dev, PCI_MSIX_FLAGS_MASKALL, 0);
847
848 pcibios_free_irq(dev);
849 return 0;
850
851out_avail:
852 if (ret < 0) {
853 /*
854 * If we had some success, report the number of IRQs
855 * we succeeded in setting up.
856 */
857 struct msi_desc *entry;
858 int avail = 0;
859
860 for_each_pci_msi_entry(entry, dev) {
861 if (entry->irq != 0)
862 avail++;
863 }
864 if (avail != 0)
865 ret = avail;
866 }
867
868out_free:
869 free_msi_irqs(dev);
870
871out_disable:
872 pci_msix_clear_and_set_ctrl(dev, PCI_MSIX_FLAGS_ENABLE, 0);
873
874 return ret;
875}
876
877/**
878 * pci_msi_supported - check whether MSI may be enabled on a device
879 * @dev: pointer to the pci_dev data structure of MSI device function
880 * @nvec: how many MSIs have been requested?
881 *
882 * Look at global flags, the device itself, and its parent buses
883 * to determine if MSI/-X are supported for the device. If MSI/-X is
884 * supported return 1, else return 0.
885 **/
886static int pci_msi_supported(struct pci_dev *dev, int nvec)
887{
888 struct pci_bus *bus;
889
890 /* MSI must be globally enabled and supported by the device */
891 if (!pci_msi_enable)
892 return 0;
893
894 if (!dev || dev->no_msi)
895 return 0;
896
897 /*
898 * You can't ask to have 0 or less MSIs configured.
899 * a) it's stupid ..
900 * b) the list manipulation code assumes nvec >= 1.
901 */
902 if (nvec < 1)
903 return 0;
904
905 /*
906 * Any bridge which does NOT route MSI transactions from its
907 * secondary bus to its primary bus must set NO_MSI flag on
908 * the secondary pci_bus.
909 *
910 * The NO_MSI flag can either be set directly by:
911 * - arch-specific PCI host bus controller drivers (deprecated)
912 * - quirks for specific PCI bridges
913 *
914 * or indirectly by platform-specific PCI host bridge drivers by
915 * advertising the 'msi_domain' property, which results in
916 * the NO_MSI flag when no MSI domain is found for this bridge
917 * at probe time.
918 */
919 for (bus = dev->bus; bus; bus = bus->parent)
920 if (bus->bus_flags & PCI_BUS_FLAGS_NO_MSI)
921 return 0;
922
923 return 1;
924}
925
926/**
927 * pci_msi_vec_count - Return the number of MSI vectors a device can send
928 * @dev: device to report about
929 *
930 * This function returns the number of MSI vectors a device requested via
931 * Multiple Message Capable register. It returns a negative errno if the
932 * device is not capable sending MSI interrupts. Otherwise, the call succeeds
933 * and returns a power of two, up to a maximum of 2^5 (32), according to the
934 * MSI specification.
935 **/
936int pci_msi_vec_count(struct pci_dev *dev)
937{
938 int ret;
939 u16 msgctl;
940
941 if (!dev->msi_cap)
942 return -EINVAL;
943
944 pci_read_config_word(dev, dev->msi_cap + PCI_MSI_FLAGS, &msgctl);
945 ret = 1 << ((msgctl & PCI_MSI_FLAGS_QMASK) >> 1);
946
947 return ret;
948}
949EXPORT_SYMBOL(pci_msi_vec_count);
950
951static void pci_msi_shutdown(struct pci_dev *dev)
952{
953 struct msi_desc *desc;
954 u32 mask;
955
956 if (!pci_msi_enable || !dev || !dev->msi_enabled)
957 return;
958
959 BUG_ON(list_empty(dev_to_msi_list(&dev->dev)));
960 desc = first_pci_msi_entry(dev);
961
962 pci_msi_set_enable(dev, 0);
963 pci_intx_for_msi(dev, 1);
964 dev->msi_enabled = 0;
965
966 /* Return the device with MSI unmasked as initial states */
967 mask = msi_mask(desc->msi_attrib.multi_cap);
968 msi_mask_irq(desc, mask, 0);
969
970 /* Restore dev->irq to its default pin-assertion IRQ */
971 dev->irq = desc->msi_attrib.default_irq;
972 pcibios_alloc_irq(dev);
973}
974
975void pci_disable_msi(struct pci_dev *dev)
976{
977 if (!pci_msi_enable || !dev || !dev->msi_enabled)
978 return;
979
980 pci_msi_shutdown(dev);
981 free_msi_irqs(dev);
982}
983EXPORT_SYMBOL(pci_disable_msi);
984
985/**
986 * pci_msix_vec_count - return the number of device's MSI-X table entries
987 * @dev: pointer to the pci_dev data structure of MSI-X device function
988 * This function returns the number of device's MSI-X table entries and
989 * therefore the number of MSI-X vectors device is capable of sending.
990 * It returns a negative errno if the device is not capable of sending MSI-X
991 * interrupts.
992 **/
993int pci_msix_vec_count(struct pci_dev *dev)
994{
995 u16 control;
996
997 if (!dev->msix_cap)
998 return -EINVAL;
999
1000 pci_read_config_word(dev, dev->msix_cap + PCI_MSIX_FLAGS, &control);
1001 return msix_table_size(control);
1002}
1003EXPORT_SYMBOL(pci_msix_vec_count);
1004
1005static int __pci_enable_msix(struct pci_dev *dev, struct msix_entry *entries,
1006 int nvec, struct irq_affinity *affd, int flags)
1007{
1008 int nr_entries;
1009 int i, j;
1010
1011 if (!pci_msi_supported(dev, nvec) || dev->current_state != PCI_D0)
1012 return -EINVAL;
1013
1014 nr_entries = pci_msix_vec_count(dev);
1015 if (nr_entries < 0)
1016 return nr_entries;
1017 if (nvec > nr_entries && !(flags & PCI_IRQ_VIRTUAL))
1018 return nr_entries;
1019
1020 if (entries) {
1021 /* Check for any invalid entries */
1022 for (i = 0; i < nvec; i++) {
1023 if (entries[i].entry >= nr_entries)
1024 return -EINVAL; /* invalid entry */
1025 for (j = i + 1; j < nvec; j++) {
1026 if (entries[i].entry == entries[j].entry)
1027 return -EINVAL; /* duplicate entry */
1028 }
1029 }
1030 }
1031
1032 /* Check whether driver already requested for MSI IRQ */
1033 if (dev->msi_enabled) {
1034 pci_info(dev, "can't enable MSI-X (MSI IRQ already assigned)\n");
1035 return -EINVAL;
1036 }
1037 return msix_capability_init(dev, entries, nvec, affd);
1038}
1039
1040static void pci_msix_shutdown(struct pci_dev *dev)
1041{
1042 struct msi_desc *entry;
1043
1044 if (!pci_msi_enable || !dev || !dev->msix_enabled)
1045 return;
1046
1047 if (pci_dev_is_disconnected(dev)) {
1048 dev->msix_enabled = 0;
1049 return;
1050 }
1051
1052 /* Return the device with MSI-X masked as initial states */
1053 for_each_pci_msi_entry(entry, dev)
1054 __pci_msix_desc_mask_irq(entry, 1);
1055
1056 pci_msix_clear_and_set_ctrl(dev, PCI_MSIX_FLAGS_ENABLE, 0);
1057 pci_intx_for_msi(dev, 1);
1058 dev->msix_enabled = 0;
1059 pcibios_alloc_irq(dev);
1060}
1061
1062void pci_disable_msix(struct pci_dev *dev)
1063{
1064 if (!pci_msi_enable || !dev || !dev->msix_enabled)
1065 return;
1066
1067 pci_msix_shutdown(dev);
1068 free_msi_irqs(dev);
1069}
1070EXPORT_SYMBOL(pci_disable_msix);
1071
1072void pci_no_msi(void)
1073{
1074 pci_msi_enable = 0;
1075}
1076
1077/**
1078 * pci_msi_enabled - is MSI enabled?
1079 *
1080 * Returns true if MSI has not been disabled by the command-line option
1081 * pci=nomsi.
1082 **/
1083int pci_msi_enabled(void)
1084{
1085 return pci_msi_enable;
1086}
1087EXPORT_SYMBOL(pci_msi_enabled);
1088
1089static int __pci_enable_msi_range(struct pci_dev *dev, int minvec, int maxvec,
1090 struct irq_affinity *affd)
1091{
1092 int nvec;
1093 int rc;
1094
1095 if (!pci_msi_supported(dev, minvec) || dev->current_state != PCI_D0)
1096 return -EINVAL;
1097
1098 /* Check whether driver already requested MSI-X IRQs */
1099 if (dev->msix_enabled) {
1100 pci_info(dev, "can't enable MSI (MSI-X already enabled)\n");
1101 return -EINVAL;
1102 }
1103
1104 if (maxvec < minvec)
1105 return -ERANGE;
1106
1107 if (WARN_ON_ONCE(dev->msi_enabled))
1108 return -EINVAL;
1109
1110 nvec = pci_msi_vec_count(dev);
1111 if (nvec < 0)
1112 return nvec;
1113 if (nvec < minvec)
1114 return -ENOSPC;
1115
1116 if (nvec > maxvec)
1117 nvec = maxvec;
1118
1119 for (;;) {
1120 if (affd) {
1121 nvec = irq_calc_affinity_vectors(minvec, nvec, affd);
1122 if (nvec < minvec)
1123 return -ENOSPC;
1124 }
1125
1126 rc = msi_capability_init(dev, nvec, affd);
1127 if (rc == 0)
1128 return nvec;
1129
1130 if (rc < 0)
1131 return rc;
1132 if (rc < minvec)
1133 return -ENOSPC;
1134
1135 nvec = rc;
1136 }
1137}
1138
1139/* deprecated, don't use */
1140int pci_enable_msi(struct pci_dev *dev)
1141{
1142 int rc = __pci_enable_msi_range(dev, 1, 1, NULL);
1143 if (rc < 0)
1144 return rc;
1145 return 0;
1146}
1147EXPORT_SYMBOL(pci_enable_msi);
1148
1149static int __pci_enable_msix_range(struct pci_dev *dev,
1150 struct msix_entry *entries, int minvec,
1151 int maxvec, struct irq_affinity *affd,
1152 int flags)
1153{
1154 int rc, nvec = maxvec;
1155
1156 if (maxvec < minvec)
1157 return -ERANGE;
1158
1159 if (WARN_ON_ONCE(dev->msix_enabled))
1160 return -EINVAL;
1161
1162 for (;;) {
1163 if (affd) {
1164 nvec = irq_calc_affinity_vectors(minvec, nvec, affd);
1165 if (nvec < minvec)
1166 return -ENOSPC;
1167 }
1168
1169 rc = __pci_enable_msix(dev, entries, nvec, affd, flags);
1170 if (rc == 0)
1171 return nvec;
1172
1173 if (rc < 0)
1174 return rc;
1175 if (rc < minvec)
1176 return -ENOSPC;
1177
1178 nvec = rc;
1179 }
1180}
1181
1182/**
1183 * pci_enable_msix_range - configure device's MSI-X capability structure
1184 * @dev: pointer to the pci_dev data structure of MSI-X device function
1185 * @entries: pointer to an array of MSI-X entries
1186 * @minvec: minimum number of MSI-X IRQs requested
1187 * @maxvec: maximum number of MSI-X IRQs requested
1188 *
1189 * Setup the MSI-X capability structure of device function with a maximum
1190 * possible number of interrupts in the range between @minvec and @maxvec
1191 * upon its software driver call to request for MSI-X mode enabled on its
1192 * hardware device function. It returns a negative errno if an error occurs.
1193 * If it succeeds, it returns the actual number of interrupts allocated and
1194 * indicates the successful configuration of MSI-X capability structure
1195 * with new allocated MSI-X interrupts.
1196 **/
1197int pci_enable_msix_range(struct pci_dev *dev, struct msix_entry *entries,
1198 int minvec, int maxvec)
1199{
1200 return __pci_enable_msix_range(dev, entries, minvec, maxvec, NULL, 0);
1201}
1202EXPORT_SYMBOL(pci_enable_msix_range);
1203
1204/**
1205 * pci_alloc_irq_vectors_affinity - allocate multiple IRQs for a device
1206 * @dev: PCI device to operate on
1207 * @min_vecs: minimum number of vectors required (must be >= 1)
1208 * @max_vecs: maximum (desired) number of vectors
1209 * @flags: flags or quirks for the allocation
1210 * @affd: optional description of the affinity requirements
1211 *
1212 * Allocate up to @max_vecs interrupt vectors for @dev, using MSI-X or MSI
1213 * vectors if available, and fall back to a single legacy vector
1214 * if neither is available. Return the number of vectors allocated,
1215 * (which might be smaller than @max_vecs) if successful, or a negative
1216 * error code on error. If less than @min_vecs interrupt vectors are
1217 * available for @dev the function will fail with -ENOSPC.
1218 *
1219 * To get the Linux IRQ number used for a vector that can be passed to
1220 * request_irq() use the pci_irq_vector() helper.
1221 */
1222int pci_alloc_irq_vectors_affinity(struct pci_dev *dev, unsigned int min_vecs,
1223 unsigned int max_vecs, unsigned int flags,
1224 struct irq_affinity *affd)
1225{
1226 struct irq_affinity msi_default_affd = {0};
1227 int nvecs = -ENOSPC;
1228
1229 if (flags & PCI_IRQ_AFFINITY) {
1230 if (!affd)
1231 affd = &msi_default_affd;
1232 } else {
1233 if (WARN_ON(affd))
1234 affd = NULL;
1235 }
1236
1237 if (flags & PCI_IRQ_MSIX) {
1238 nvecs = __pci_enable_msix_range(dev, NULL, min_vecs, max_vecs,
1239 affd, flags);
1240 if (nvecs > 0)
1241 return nvecs;
1242 }
1243
1244 if (flags & PCI_IRQ_MSI) {
1245 nvecs = __pci_enable_msi_range(dev, min_vecs, max_vecs, affd);
1246 if (nvecs > 0)
1247 return nvecs;
1248 }
1249
1250 /* use legacy IRQ if allowed */
1251 if (flags & PCI_IRQ_LEGACY) {
1252 if (min_vecs == 1 && dev->irq) {
1253 /*
1254 * Invoke the affinity spreading logic to ensure that
1255 * the device driver can adjust queue configuration
1256 * for the single interrupt case.
1257 */
1258 if (affd)
1259 irq_create_affinity_masks(1, affd);
1260 pci_intx(dev, 1);
1261 return 1;
1262 }
1263 }
1264
1265 return nvecs;
1266}
1267EXPORT_SYMBOL(pci_alloc_irq_vectors_affinity);
1268
1269/**
1270 * pci_free_irq_vectors - free previously allocated IRQs for a device
1271 * @dev: PCI device to operate on
1272 *
1273 * Undoes the allocations and enabling in pci_alloc_irq_vectors().
1274 */
1275void pci_free_irq_vectors(struct pci_dev *dev)
1276{
1277 pci_disable_msix(dev);
1278 pci_disable_msi(dev);
1279}
1280EXPORT_SYMBOL(pci_free_irq_vectors);
1281
1282/**
1283 * pci_irq_vector - return Linux IRQ number of a device vector
1284 * @dev: PCI device to operate on
1285 * @nr: device-relative interrupt vector index (0-based).
1286 */
1287int pci_irq_vector(struct pci_dev *dev, unsigned int nr)
1288{
1289 if (dev->msix_enabled) {
1290 struct msi_desc *entry;
1291 int i = 0;
1292
1293 for_each_pci_msi_entry(entry, dev) {
1294 if (i == nr)
1295 return entry->irq;
1296 i++;
1297 }
1298 WARN_ON_ONCE(1);
1299 return -EINVAL;
1300 }
1301
1302 if (dev->msi_enabled) {
1303 struct msi_desc *entry = first_pci_msi_entry(dev);
1304
1305 if (WARN_ON_ONCE(nr >= entry->nvec_used))
1306 return -EINVAL;
1307 } else {
1308 if (WARN_ON_ONCE(nr > 0))
1309 return -EINVAL;
1310 }
1311
1312 return dev->irq + nr;
1313}
1314EXPORT_SYMBOL(pci_irq_vector);
1315
1316/**
1317 * pci_irq_get_affinity - return the affinity of a particular MSI vector
1318 * @dev: PCI device to operate on
1319 * @nr: device-relative interrupt vector index (0-based).
1320 */
1321const struct cpumask *pci_irq_get_affinity(struct pci_dev *dev, int nr)
1322{
1323 if (dev->msix_enabled) {
1324 struct msi_desc *entry;
1325 int i = 0;
1326
1327 for_each_pci_msi_entry(entry, dev) {
1328 if (i == nr)
1329 return &entry->affinity->mask;
1330 i++;
1331 }
1332 WARN_ON_ONCE(1);
1333 return NULL;
1334 } else if (dev->msi_enabled) {
1335 struct msi_desc *entry = first_pci_msi_entry(dev);
1336
1337 if (WARN_ON_ONCE(!entry || !entry->affinity ||
1338 nr >= entry->nvec_used))
1339 return NULL;
1340
1341 return &entry->affinity[nr].mask;
1342 } else {
1343 return cpu_possible_mask;
1344 }
1345}
1346EXPORT_SYMBOL(pci_irq_get_affinity);
1347
1348struct pci_dev *msi_desc_to_pci_dev(struct msi_desc *desc)
1349{
1350 return to_pci_dev(desc->dev);
1351}
1352EXPORT_SYMBOL(msi_desc_to_pci_dev);
1353
1354void *msi_desc_to_pci_sysdata(struct msi_desc *desc)
1355{
1356 struct pci_dev *dev = msi_desc_to_pci_dev(desc);
1357
1358 return dev->bus->sysdata;
1359}
1360EXPORT_SYMBOL_GPL(msi_desc_to_pci_sysdata);
1361
1362#ifdef CONFIG_PCI_MSI_IRQ_DOMAIN
1363/**
1364 * pci_msi_domain_write_msg - Helper to write MSI message to PCI config space
1365 * @irq_data: Pointer to interrupt data of the MSI interrupt
1366 * @msg: Pointer to the message
1367 */
1368void pci_msi_domain_write_msg(struct irq_data *irq_data, struct msi_msg *msg)
1369{
1370 struct msi_desc *desc = irq_data_get_msi_desc(irq_data);
1371
1372 /*
1373 * For MSI-X desc->irq is always equal to irq_data->irq. For
1374 * MSI only the first interrupt of MULTI MSI passes the test.
1375 */
1376 if (desc->irq == irq_data->irq)
1377 __pci_write_msi_msg(desc, msg);
1378}
1379
1380/**
1381 * pci_msi_domain_calc_hwirq - Generate a unique ID for an MSI source
1382 * @desc: Pointer to the MSI descriptor
1383 *
1384 * The ID number is only used within the irqdomain.
1385 */
1386static irq_hw_number_t pci_msi_domain_calc_hwirq(struct msi_desc *desc)
1387{
1388 struct pci_dev *dev = msi_desc_to_pci_dev(desc);
1389
1390 return (irq_hw_number_t)desc->msi_attrib.entry_nr |
1391 pci_dev_id(dev) << 11 |
1392 (pci_domain_nr(dev->bus) & 0xFFFFFFFF) << 27;
1393}
1394
1395static inline bool pci_msi_desc_is_multi_msi(struct msi_desc *desc)
1396{
1397 return !desc->msi_attrib.is_msix && desc->nvec_used > 1;
1398}
1399
1400/**
1401 * pci_msi_domain_check_cap - Verify that @domain supports the capabilities
1402 * for @dev
1403 * @domain: The interrupt domain to check
1404 * @info: The domain info for verification
1405 * @dev: The device to check
1406 *
1407 * Returns:
1408 * 0 if the functionality is supported
1409 * 1 if Multi MSI is requested, but the domain does not support it
1410 * -ENOTSUPP otherwise
1411 */
1412int pci_msi_domain_check_cap(struct irq_domain *domain,
1413 struct msi_domain_info *info, struct device *dev)
1414{
1415 struct msi_desc *desc = first_pci_msi_entry(to_pci_dev(dev));
1416
1417 /* Special handling to support __pci_enable_msi_range() */
1418 if (pci_msi_desc_is_multi_msi(desc) &&
1419 !(info->flags & MSI_FLAG_MULTI_PCI_MSI))
1420 return 1;
1421 else if (desc->msi_attrib.is_msix && !(info->flags & MSI_FLAG_PCI_MSIX))
1422 return -ENOTSUPP;
1423
1424 return 0;
1425}
1426
1427static int pci_msi_domain_handle_error(struct irq_domain *domain,
1428 struct msi_desc *desc, int error)
1429{
1430 /* Special handling to support __pci_enable_msi_range() */
1431 if (pci_msi_desc_is_multi_msi(desc) && error == -ENOSPC)
1432 return 1;
1433
1434 return error;
1435}
1436
1437static void pci_msi_domain_set_desc(msi_alloc_info_t *arg,
1438 struct msi_desc *desc)
1439{
1440 arg->desc = desc;
1441 arg->hwirq = pci_msi_domain_calc_hwirq(desc);
1442}
1443
1444static struct msi_domain_ops pci_msi_domain_ops_default = {
1445 .set_desc = pci_msi_domain_set_desc,
1446 .msi_check = pci_msi_domain_check_cap,
1447 .handle_error = pci_msi_domain_handle_error,
1448};
1449
1450static void pci_msi_domain_update_dom_ops(struct msi_domain_info *info)
1451{
1452 struct msi_domain_ops *ops = info->ops;
1453
1454 if (ops == NULL) {
1455 info->ops = &pci_msi_domain_ops_default;
1456 } else {
1457 if (ops->set_desc == NULL)
1458 ops->set_desc = pci_msi_domain_set_desc;
1459 if (ops->msi_check == NULL)
1460 ops->msi_check = pci_msi_domain_check_cap;
1461 if (ops->handle_error == NULL)
1462 ops->handle_error = pci_msi_domain_handle_error;
1463 }
1464}
1465
1466static void pci_msi_domain_update_chip_ops(struct msi_domain_info *info)
1467{
1468 struct irq_chip *chip = info->chip;
1469
1470 BUG_ON(!chip);
1471 if (!chip->irq_write_msi_msg)
1472 chip->irq_write_msi_msg = pci_msi_domain_write_msg;
1473 if (!chip->irq_mask)
1474 chip->irq_mask = pci_msi_mask_irq;
1475 if (!chip->irq_unmask)
1476 chip->irq_unmask = pci_msi_unmask_irq;
1477}
1478
1479/**
1480 * pci_msi_create_irq_domain - Create a MSI interrupt domain
1481 * @fwnode: Optional fwnode of the interrupt controller
1482 * @info: MSI domain info
1483 * @parent: Parent irq domain
1484 *
1485 * Updates the domain and chip ops and creates a MSI interrupt domain.
1486 *
1487 * Returns:
1488 * A domain pointer or NULL in case of failure.
1489 */
1490struct irq_domain *pci_msi_create_irq_domain(struct fwnode_handle *fwnode,
1491 struct msi_domain_info *info,
1492 struct irq_domain *parent)
1493{
1494 struct irq_domain *domain;
1495
1496 if (WARN_ON(info->flags & MSI_FLAG_LEVEL_CAPABLE))
1497 info->flags &= ~MSI_FLAG_LEVEL_CAPABLE;
1498
1499 if (info->flags & MSI_FLAG_USE_DEF_DOM_OPS)
1500 pci_msi_domain_update_dom_ops(info);
1501 if (info->flags & MSI_FLAG_USE_DEF_CHIP_OPS)
1502 pci_msi_domain_update_chip_ops(info);
1503
1504 info->flags |= MSI_FLAG_ACTIVATE_EARLY;
1505 if (IS_ENABLED(CONFIG_GENERIC_IRQ_RESERVATION_MODE))
1506 info->flags |= MSI_FLAG_MUST_REACTIVATE;
1507
1508 /* PCI-MSI is oneshot-safe */
1509 info->chip->flags |= IRQCHIP_ONESHOT_SAFE;
1510
1511 domain = msi_create_irq_domain(fwnode, info, parent);
1512 if (!domain)
1513 return NULL;
1514
1515 irq_domain_update_bus_token(domain, DOMAIN_BUS_PCI_MSI);
1516 return domain;
1517}
1518EXPORT_SYMBOL_GPL(pci_msi_create_irq_domain);
1519
1520/*
1521 * Users of the generic MSI infrastructure expect a device to have a single ID,
1522 * so with DMA aliases we have to pick the least-worst compromise. Devices with
1523 * DMA phantom functions tend to still emit MSIs from the real function number,
1524 * so we ignore those and only consider topological aliases where either the
1525 * alias device or RID appears on a different bus number. We also make the
1526 * reasonable assumption that bridges are walked in an upstream direction (so
1527 * the last one seen wins), and the much braver assumption that the most likely
1528 * case is that of PCI->PCIe so we should always use the alias RID. This echoes
1529 * the logic from intel_irq_remapping's set_msi_sid(), which presumably works
1530 * well enough in practice; in the face of the horrible PCIe<->PCI-X conditions
1531 * for taking ownership all we can really do is close our eyes and hope...
1532 */
1533static int get_msi_id_cb(struct pci_dev *pdev, u16 alias, void *data)
1534{
1535 u32 *pa = data;
1536 u8 bus = PCI_BUS_NUM(*pa);
1537
1538 if (pdev->bus->number != bus || PCI_BUS_NUM(alias) != bus)
1539 *pa = alias;
1540
1541 return 0;
1542}
1543
1544/**
1545 * pci_msi_domain_get_msi_rid - Get the MSI requester id (RID)
1546 * @domain: The interrupt domain
1547 * @pdev: The PCI device.
1548 *
1549 * The RID for a device is formed from the alias, with a firmware
1550 * supplied mapping applied
1551 *
1552 * Returns: The RID.
1553 */
1554u32 pci_msi_domain_get_msi_rid(struct irq_domain *domain, struct pci_dev *pdev)
1555{
1556 struct device_node *of_node;
1557 u32 rid = pci_dev_id(pdev);
1558
1559 pci_for_each_dma_alias(pdev, get_msi_id_cb, &rid);
1560
1561 of_node = irq_domain_get_of_node(domain);
1562 rid = of_node ? of_msi_map_id(&pdev->dev, of_node, rid) :
1563 iort_msi_map_id(&pdev->dev, rid);
1564
1565 return rid;
1566}
1567
1568/**
1569 * pci_msi_get_device_domain - Get the MSI domain for a given PCI device
1570 * @pdev: The PCI device
1571 *
1572 * Use the firmware data to find a device-specific MSI domain
1573 * (i.e. not one that is set as a default).
1574 *
1575 * Returns: The corresponding MSI domain or NULL if none has been found.
1576 */
1577struct irq_domain *pci_msi_get_device_domain(struct pci_dev *pdev)
1578{
1579 struct irq_domain *dom;
1580 u32 rid = pci_dev_id(pdev);
1581
1582 pci_for_each_dma_alias(pdev, get_msi_id_cb, &rid);
1583 dom = of_msi_map_get_device_domain(&pdev->dev, rid, DOMAIN_BUS_PCI_MSI);
1584 if (!dom)
1585 dom = iort_get_device_domain(&pdev->dev, rid,
1586 DOMAIN_BUS_PCI_MSI);
1587 return dom;
1588}
1589
1590/**
1591 * pci_dev_has_special_msi_domain - Check whether the device is handled by
1592 * a non-standard PCI-MSI domain
1593 * @pdev: The PCI device to check.
1594 *
1595 * Returns: True if the device irqdomain or the bus irqdomain is
1596 * non-standard PCI/MSI.
1597 */
1598bool pci_dev_has_special_msi_domain(struct pci_dev *pdev)
1599{
1600 struct irq_domain *dom = dev_get_msi_domain(&pdev->dev);
1601
1602 if (!dom)
1603 dom = dev_get_msi_domain(&pdev->bus->dev);
1604
1605 if (!dom)
1606 return true;
1607
1608 return dom->bus_token != DOMAIN_BUS_PCI_MSI;
1609}
1610
1611#endif /* CONFIG_PCI_MSI_IRQ_DOMAIN */
1612#endif /* CONFIG_PCI_MSI */
1613
1614void pci_msi_init(struct pci_dev *dev)
1615{
1616 u16 ctrl;
1617
1618 /*
1619 * Disable the MSI hardware to avoid screaming interrupts
1620 * during boot. This is the power on reset default so
1621 * usually this should be a noop.
1622 */
1623 dev->msi_cap = pci_find_capability(dev, PCI_CAP_ID_MSI);
1624 if (!dev->msi_cap)
1625 return;
1626
1627 pci_read_config_word(dev, dev->msi_cap + PCI_MSI_FLAGS, &ctrl);
1628 if (ctrl & PCI_MSI_FLAGS_ENABLE)
1629 pci_write_config_word(dev, dev->msi_cap + PCI_MSI_FLAGS,
1630 ctrl & ~PCI_MSI_FLAGS_ENABLE);
1631
1632 if (!(ctrl & PCI_MSI_FLAGS_64BIT))
1633 dev->no_64bit_msi = 1;
1634}
1635
1636void pci_msix_init(struct pci_dev *dev)
1637{
1638 u16 ctrl;
1639
1640 dev->msix_cap = pci_find_capability(dev, PCI_CAP_ID_MSIX);
1641 if (!dev->msix_cap)
1642 return;
1643
1644 pci_read_config_word(dev, dev->msix_cap + PCI_MSIX_FLAGS, &ctrl);
1645 if (ctrl & PCI_MSIX_FLAGS_ENABLE)
1646 pci_write_config_word(dev, dev->msix_cap + PCI_MSIX_FLAGS,
1647 ctrl & ~PCI_MSIX_FLAGS_ENABLE);
1648}