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/*
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 */
8
9#include <linux/err.h>
10#include <linux/mm.h>
11#include <linux/irq.h>
12#include <linux/interrupt.h>
13#include <linux/init.h>
14#include <linux/export.h>
15#include <linux/ioport.h>
16#include <linux/pci.h>
17#include <linux/proc_fs.h>
18#include <linux/msi.h>
19#include <linux/smp.h>
20#include <linux/errno.h>
21#include <linux/io.h>
22#include <linux/slab.h>
23
24#include "pci.h"
25
26static int pci_msi_enable = 1;
27
28#define msix_table_size(flags) ((flags & PCI_MSIX_FLAGS_QSIZE) + 1)
29
30
31/* Arch hooks */
32
33int __weak arch_setup_msi_irq(struct pci_dev *dev, struct msi_desc *desc)
34{
35 struct msi_chip *chip = dev->bus->msi;
36 int err;
37
38 if (!chip || !chip->setup_irq)
39 return -EINVAL;
40
41 err = chip->setup_irq(chip, dev, desc);
42 if (err < 0)
43 return err;
44
45 irq_set_chip_data(desc->irq, chip);
46
47 return 0;
48}
49
50void __weak arch_teardown_msi_irq(unsigned int irq)
51{
52 struct msi_chip *chip = irq_get_chip_data(irq);
53
54 if (!chip || !chip->teardown_irq)
55 return;
56
57 chip->teardown_irq(chip, irq);
58}
59
60int __weak arch_msi_check_device(struct pci_dev *dev, int nvec, int type)
61{
62 struct msi_chip *chip = dev->bus->msi;
63
64 if (!chip || !chip->check_device)
65 return 0;
66
67 return chip->check_device(chip, dev, nvec, type);
68}
69
70int __weak arch_setup_msi_irqs(struct pci_dev *dev, int nvec, int type)
71{
72 struct msi_desc *entry;
73 int ret;
74
75 /*
76 * If an architecture wants to support multiple MSI, it needs to
77 * override arch_setup_msi_irqs()
78 */
79 if (type == PCI_CAP_ID_MSI && nvec > 1)
80 return 1;
81
82 list_for_each_entry(entry, &dev->msi_list, list) {
83 ret = arch_setup_msi_irq(dev, entry);
84 if (ret < 0)
85 return ret;
86 if (ret > 0)
87 return -ENOSPC;
88 }
89
90 return 0;
91}
92
93/*
94 * We have a default implementation available as a separate non-weak
95 * function, as it is used by the Xen x86 PCI code
96 */
97void default_teardown_msi_irqs(struct pci_dev *dev)
98{
99 struct msi_desc *entry;
100
101 list_for_each_entry(entry, &dev->msi_list, list) {
102 int i, nvec;
103 if (entry->irq == 0)
104 continue;
105 if (entry->nvec_used)
106 nvec = entry->nvec_used;
107 else
108 nvec = 1 << entry->msi_attrib.multiple;
109 for (i = 0; i < nvec; i++)
110 arch_teardown_msi_irq(entry->irq + i);
111 }
112}
113
114void __weak arch_teardown_msi_irqs(struct pci_dev *dev)
115{
116 return default_teardown_msi_irqs(dev);
117}
118
119static void default_restore_msi_irq(struct pci_dev *dev, int irq)
120{
121 struct msi_desc *entry;
122
123 entry = NULL;
124 if (dev->msix_enabled) {
125 list_for_each_entry(entry, &dev->msi_list, list) {
126 if (irq == entry->irq)
127 break;
128 }
129 } else if (dev->msi_enabled) {
130 entry = irq_get_msi_desc(irq);
131 }
132
133 if (entry)
134 write_msi_msg(irq, &entry->msg);
135}
136
137void __weak arch_restore_msi_irqs(struct pci_dev *dev)
138{
139 return default_restore_msi_irqs(dev);
140}
141
142static void msi_set_enable(struct pci_dev *dev, int enable)
143{
144 u16 control;
145
146 pci_read_config_word(dev, dev->msi_cap + PCI_MSI_FLAGS, &control);
147 control &= ~PCI_MSI_FLAGS_ENABLE;
148 if (enable)
149 control |= PCI_MSI_FLAGS_ENABLE;
150 pci_write_config_word(dev, dev->msi_cap + PCI_MSI_FLAGS, control);
151}
152
153static void msix_set_enable(struct pci_dev *dev, int enable)
154{
155 u16 control;
156
157 pci_read_config_word(dev, dev->msix_cap + PCI_MSIX_FLAGS, &control);
158 control &= ~PCI_MSIX_FLAGS_ENABLE;
159 if (enable)
160 control |= PCI_MSIX_FLAGS_ENABLE;
161 pci_write_config_word(dev, dev->msix_cap + PCI_MSIX_FLAGS, control);
162}
163
164static inline __attribute_const__ u32 msi_mask(unsigned x)
165{
166 /* Don't shift by >= width of type */
167 if (x >= 5)
168 return 0xffffffff;
169 return (1 << (1 << x)) - 1;
170}
171
172static inline __attribute_const__ u32 msi_capable_mask(u16 control)
173{
174 return msi_mask((control >> 1) & 7);
175}
176
177static inline __attribute_const__ u32 msi_enabled_mask(u16 control)
178{
179 return msi_mask((control >> 4) & 7);
180}
181
182/*
183 * PCI 2.3 does not specify mask bits for each MSI interrupt. Attempting to
184 * mask all MSI interrupts by clearing the MSI enable bit does not work
185 * reliably as devices without an INTx disable bit will then generate a
186 * level IRQ which will never be cleared.
187 */
188u32 default_msi_mask_irq(struct msi_desc *desc, u32 mask, u32 flag)
189{
190 u32 mask_bits = desc->masked;
191
192 if (!desc->msi_attrib.maskbit)
193 return 0;
194
195 mask_bits &= ~mask;
196 mask_bits |= flag;
197 pci_write_config_dword(desc->dev, desc->mask_pos, mask_bits);
198
199 return mask_bits;
200}
201
202__weak u32 arch_msi_mask_irq(struct msi_desc *desc, u32 mask, u32 flag)
203{
204 return default_msi_mask_irq(desc, mask, flag);
205}
206
207static void msi_mask_irq(struct msi_desc *desc, u32 mask, u32 flag)
208{
209 desc->masked = arch_msi_mask_irq(desc, mask, flag);
210}
211
212/*
213 * This internal function does not flush PCI writes to the device.
214 * All users must ensure that they read from the device before either
215 * assuming that the device state is up to date, or returning out of this
216 * file. This saves a few milliseconds when initialising devices with lots
217 * of MSI-X interrupts.
218 */
219u32 default_msix_mask_irq(struct msi_desc *desc, u32 flag)
220{
221 u32 mask_bits = desc->masked;
222 unsigned offset = desc->msi_attrib.entry_nr * PCI_MSIX_ENTRY_SIZE +
223 PCI_MSIX_ENTRY_VECTOR_CTRL;
224 mask_bits &= ~PCI_MSIX_ENTRY_CTRL_MASKBIT;
225 if (flag)
226 mask_bits |= PCI_MSIX_ENTRY_CTRL_MASKBIT;
227 writel(mask_bits, desc->mask_base + offset);
228
229 return mask_bits;
230}
231
232__weak u32 arch_msix_mask_irq(struct msi_desc *desc, u32 flag)
233{
234 return default_msix_mask_irq(desc, flag);
235}
236
237static void msix_mask_irq(struct msi_desc *desc, u32 flag)
238{
239 desc->masked = arch_msix_mask_irq(desc, flag);
240}
241
242static void msi_set_mask_bit(struct irq_data *data, u32 flag)
243{
244 struct msi_desc *desc = irq_data_get_msi(data);
245
246 if (desc->msi_attrib.is_msix) {
247 msix_mask_irq(desc, flag);
248 readl(desc->mask_base); /* Flush write to device */
249 } else {
250 unsigned offset = data->irq - desc->dev->irq;
251 msi_mask_irq(desc, 1 << offset, flag << offset);
252 }
253}
254
255void mask_msi_irq(struct irq_data *data)
256{
257 msi_set_mask_bit(data, 1);
258}
259
260void unmask_msi_irq(struct irq_data *data)
261{
262 msi_set_mask_bit(data, 0);
263}
264
265void default_restore_msi_irqs(struct pci_dev *dev)
266{
267 struct msi_desc *entry;
268
269 list_for_each_entry(entry, &dev->msi_list, list) {
270 default_restore_msi_irq(dev, entry->irq);
271 }
272}
273
274void __read_msi_msg(struct msi_desc *entry, struct msi_msg *msg)
275{
276 BUG_ON(entry->dev->current_state != PCI_D0);
277
278 if (entry->msi_attrib.is_msix) {
279 void __iomem *base = entry->mask_base +
280 entry->msi_attrib.entry_nr * PCI_MSIX_ENTRY_SIZE;
281
282 msg->address_lo = readl(base + PCI_MSIX_ENTRY_LOWER_ADDR);
283 msg->address_hi = readl(base + PCI_MSIX_ENTRY_UPPER_ADDR);
284 msg->data = readl(base + PCI_MSIX_ENTRY_DATA);
285 } else {
286 struct pci_dev *dev = entry->dev;
287 int pos = dev->msi_cap;
288 u16 data;
289
290 pci_read_config_dword(dev, pos + PCI_MSI_ADDRESS_LO,
291 &msg->address_lo);
292 if (entry->msi_attrib.is_64) {
293 pci_read_config_dword(dev, pos + PCI_MSI_ADDRESS_HI,
294 &msg->address_hi);
295 pci_read_config_word(dev, pos + PCI_MSI_DATA_64, &data);
296 } else {
297 msg->address_hi = 0;
298 pci_read_config_word(dev, pos + PCI_MSI_DATA_32, &data);
299 }
300 msg->data = data;
301 }
302}
303
304void read_msi_msg(unsigned int irq, struct msi_msg *msg)
305{
306 struct msi_desc *entry = irq_get_msi_desc(irq);
307
308 __read_msi_msg(entry, msg);
309}
310
311void __get_cached_msi_msg(struct msi_desc *entry, struct msi_msg *msg)
312{
313 /* Assert that the cache is valid, assuming that
314 * valid messages are not all-zeroes. */
315 BUG_ON(!(entry->msg.address_hi | entry->msg.address_lo |
316 entry->msg.data));
317
318 *msg = entry->msg;
319}
320
321void get_cached_msi_msg(unsigned int irq, struct msi_msg *msg)
322{
323 struct msi_desc *entry = irq_get_msi_desc(irq);
324
325 __get_cached_msi_msg(entry, msg);
326}
327
328void __write_msi_msg(struct msi_desc *entry, struct msi_msg *msg)
329{
330 if (entry->dev->current_state != PCI_D0) {
331 /* Don't touch the hardware now */
332 } else if (entry->msi_attrib.is_msix) {
333 void __iomem *base;
334 base = entry->mask_base +
335 entry->msi_attrib.entry_nr * PCI_MSIX_ENTRY_SIZE;
336
337 writel(msg->address_lo, base + PCI_MSIX_ENTRY_LOWER_ADDR);
338 writel(msg->address_hi, base + PCI_MSIX_ENTRY_UPPER_ADDR);
339 writel(msg->data, base + PCI_MSIX_ENTRY_DATA);
340 } else {
341 struct pci_dev *dev = entry->dev;
342 int pos = dev->msi_cap;
343 u16 msgctl;
344
345 pci_read_config_word(dev, pos + PCI_MSI_FLAGS, &msgctl);
346 msgctl &= ~PCI_MSI_FLAGS_QSIZE;
347 msgctl |= entry->msi_attrib.multiple << 4;
348 pci_write_config_word(dev, pos + PCI_MSI_FLAGS, msgctl);
349
350 pci_write_config_dword(dev, pos + PCI_MSI_ADDRESS_LO,
351 msg->address_lo);
352 if (entry->msi_attrib.is_64) {
353 pci_write_config_dword(dev, pos + PCI_MSI_ADDRESS_HI,
354 msg->address_hi);
355 pci_write_config_word(dev, pos + PCI_MSI_DATA_64,
356 msg->data);
357 } else {
358 pci_write_config_word(dev, pos + PCI_MSI_DATA_32,
359 msg->data);
360 }
361 }
362 entry->msg = *msg;
363}
364
365void write_msi_msg(unsigned int irq, struct msi_msg *msg)
366{
367 struct msi_desc *entry = irq_get_msi_desc(irq);
368
369 __write_msi_msg(entry, msg);
370}
371
372static void free_msi_irqs(struct pci_dev *dev)
373{
374 struct msi_desc *entry, *tmp;
375 struct attribute **msi_attrs;
376 struct device_attribute *dev_attr;
377 int count = 0;
378
379 list_for_each_entry(entry, &dev->msi_list, list) {
380 int i, nvec;
381 if (!entry->irq)
382 continue;
383 if (entry->nvec_used)
384 nvec = entry->nvec_used;
385 else
386 nvec = 1 << entry->msi_attrib.multiple;
387 for (i = 0; i < nvec; i++)
388 BUG_ON(irq_has_action(entry->irq + i));
389 }
390
391 arch_teardown_msi_irqs(dev);
392
393 list_for_each_entry_safe(entry, tmp, &dev->msi_list, list) {
394 if (entry->msi_attrib.is_msix) {
395 if (list_is_last(&entry->list, &dev->msi_list))
396 iounmap(entry->mask_base);
397 }
398
399 /*
400 * Its possible that we get into this path
401 * When populate_msi_sysfs fails, which means the entries
402 * were not registered with sysfs. In that case don't
403 * unregister them.
404 */
405 if (entry->kobj.parent) {
406 kobject_del(&entry->kobj);
407 kobject_put(&entry->kobj);
408 }
409
410 list_del(&entry->list);
411 kfree(entry);
412 }
413
414 if (dev->msi_irq_groups) {
415 sysfs_remove_groups(&dev->dev.kobj, dev->msi_irq_groups);
416 msi_attrs = dev->msi_irq_groups[0]->attrs;
417 list_for_each_entry(entry, &dev->msi_list, list) {
418 dev_attr = container_of(msi_attrs[count],
419 struct device_attribute, attr);
420 kfree(dev_attr->attr.name);
421 kfree(dev_attr);
422 ++count;
423 }
424 kfree(msi_attrs);
425 kfree(dev->msi_irq_groups[0]);
426 kfree(dev->msi_irq_groups);
427 dev->msi_irq_groups = NULL;
428 }
429}
430
431static struct msi_desc *alloc_msi_entry(struct pci_dev *dev)
432{
433 struct msi_desc *desc = kzalloc(sizeof(*desc), GFP_KERNEL);
434 if (!desc)
435 return NULL;
436
437 INIT_LIST_HEAD(&desc->list);
438 desc->dev = dev;
439
440 return desc;
441}
442
443static void pci_intx_for_msi(struct pci_dev *dev, int enable)
444{
445 if (!(dev->dev_flags & PCI_DEV_FLAGS_MSI_INTX_DISABLE_BUG))
446 pci_intx(dev, enable);
447}
448
449static void __pci_restore_msi_state(struct pci_dev *dev)
450{
451 u16 control;
452 struct msi_desc *entry;
453
454 if (!dev->msi_enabled)
455 return;
456
457 entry = irq_get_msi_desc(dev->irq);
458
459 pci_intx_for_msi(dev, 0);
460 msi_set_enable(dev, 0);
461 arch_restore_msi_irqs(dev);
462
463 pci_read_config_word(dev, dev->msi_cap + PCI_MSI_FLAGS, &control);
464 msi_mask_irq(entry, msi_capable_mask(control), entry->masked);
465 control &= ~PCI_MSI_FLAGS_QSIZE;
466 control |= (entry->msi_attrib.multiple << 4) | PCI_MSI_FLAGS_ENABLE;
467 pci_write_config_word(dev, dev->msi_cap + PCI_MSI_FLAGS, control);
468}
469
470static void __pci_restore_msix_state(struct pci_dev *dev)
471{
472 struct msi_desc *entry;
473 u16 control;
474
475 if (!dev->msix_enabled)
476 return;
477 BUG_ON(list_empty(&dev->msi_list));
478 entry = list_first_entry(&dev->msi_list, struct msi_desc, list);
479 pci_read_config_word(dev, dev->msix_cap + PCI_MSIX_FLAGS, &control);
480
481 /* route the table */
482 pci_intx_for_msi(dev, 0);
483 control |= PCI_MSIX_FLAGS_ENABLE | PCI_MSIX_FLAGS_MASKALL;
484 pci_write_config_word(dev, dev->msix_cap + PCI_MSIX_FLAGS, control);
485
486 arch_restore_msi_irqs(dev);
487 list_for_each_entry(entry, &dev->msi_list, list) {
488 msix_mask_irq(entry, entry->masked);
489 }
490
491 control &= ~PCI_MSIX_FLAGS_MASKALL;
492 pci_write_config_word(dev, dev->msix_cap + PCI_MSIX_FLAGS, control);
493}
494
495void pci_restore_msi_state(struct pci_dev *dev)
496{
497 __pci_restore_msi_state(dev);
498 __pci_restore_msix_state(dev);
499}
500EXPORT_SYMBOL_GPL(pci_restore_msi_state);
501
502static ssize_t msi_mode_show(struct device *dev, struct device_attribute *attr,
503 char *buf)
504{
505 struct pci_dev *pdev = to_pci_dev(dev);
506 struct msi_desc *entry;
507 unsigned long irq;
508 int retval;
509
510 retval = kstrtoul(attr->attr.name, 10, &irq);
511 if (retval)
512 return retval;
513
514 list_for_each_entry(entry, &pdev->msi_list, list) {
515 if (entry->irq == irq) {
516 return sprintf(buf, "%s\n",
517 entry->msi_attrib.is_msix ? "msix" : "msi");
518 }
519 }
520 return -ENODEV;
521}
522
523static int populate_msi_sysfs(struct pci_dev *pdev)
524{
525 struct attribute **msi_attrs;
526 struct attribute *msi_attr;
527 struct device_attribute *msi_dev_attr;
528 struct attribute_group *msi_irq_group;
529 const struct attribute_group **msi_irq_groups;
530 struct msi_desc *entry;
531 int ret = -ENOMEM;
532 int num_msi = 0;
533 int count = 0;
534
535 /* Determine how many msi entries we have */
536 list_for_each_entry(entry, &pdev->msi_list, list) {
537 ++num_msi;
538 }
539 if (!num_msi)
540 return 0;
541
542 /* Dynamically create the MSI attributes for the PCI device */
543 msi_attrs = kzalloc(sizeof(void *) * (num_msi + 1), GFP_KERNEL);
544 if (!msi_attrs)
545 return -ENOMEM;
546 list_for_each_entry(entry, &pdev->msi_list, list) {
547 char *name = kmalloc(20, GFP_KERNEL);
548 if (!name)
549 goto error_attrs;
550
551 msi_dev_attr = kzalloc(sizeof(*msi_dev_attr), GFP_KERNEL);
552 if (!msi_dev_attr) {
553 kfree(name);
554 goto error_attrs;
555 }
556
557 sprintf(name, "%d", entry->irq);
558 sysfs_attr_init(&msi_dev_attr->attr);
559 msi_dev_attr->attr.name = name;
560 msi_dev_attr->attr.mode = S_IRUGO;
561 msi_dev_attr->show = msi_mode_show;
562 msi_attrs[count] = &msi_dev_attr->attr;
563 ++count;
564 }
565
566 msi_irq_group = kzalloc(sizeof(*msi_irq_group), GFP_KERNEL);
567 if (!msi_irq_group)
568 goto error_attrs;
569 msi_irq_group->name = "msi_irqs";
570 msi_irq_group->attrs = msi_attrs;
571
572 msi_irq_groups = kzalloc(sizeof(void *) * 2, GFP_KERNEL);
573 if (!msi_irq_groups)
574 goto error_irq_group;
575 msi_irq_groups[0] = msi_irq_group;
576
577 ret = sysfs_create_groups(&pdev->dev.kobj, msi_irq_groups);
578 if (ret)
579 goto error_irq_groups;
580 pdev->msi_irq_groups = msi_irq_groups;
581
582 return 0;
583
584error_irq_groups:
585 kfree(msi_irq_groups);
586error_irq_group:
587 kfree(msi_irq_group);
588error_attrs:
589 count = 0;
590 msi_attr = msi_attrs[count];
591 while (msi_attr) {
592 msi_dev_attr = container_of(msi_attr, struct device_attribute, attr);
593 kfree(msi_attr->name);
594 kfree(msi_dev_attr);
595 ++count;
596 msi_attr = msi_attrs[count];
597 }
598 kfree(msi_attrs);
599 return ret;
600}
601
602/**
603 * msi_capability_init - configure device's MSI capability structure
604 * @dev: pointer to the pci_dev data structure of MSI device function
605 * @nvec: number of interrupts to allocate
606 *
607 * Setup the MSI capability structure of the device with the requested
608 * number of interrupts. A return value of zero indicates the successful
609 * setup of an entry with the new MSI irq. A negative return value indicates
610 * an error, and a positive return value indicates the number of interrupts
611 * which could have been allocated.
612 */
613static int msi_capability_init(struct pci_dev *dev, int nvec)
614{
615 struct msi_desc *entry;
616 int ret;
617 u16 control;
618 unsigned mask;
619
620 msi_set_enable(dev, 0); /* Disable MSI during set up */
621
622 pci_read_config_word(dev, dev->msi_cap + PCI_MSI_FLAGS, &control);
623 /* MSI Entry Initialization */
624 entry = alloc_msi_entry(dev);
625 if (!entry)
626 return -ENOMEM;
627
628 entry->msi_attrib.is_msix = 0;
629 entry->msi_attrib.is_64 = !!(control & PCI_MSI_FLAGS_64BIT);
630 entry->msi_attrib.entry_nr = 0;
631 entry->msi_attrib.maskbit = !!(control & PCI_MSI_FLAGS_MASKBIT);
632 entry->msi_attrib.default_irq = dev->irq; /* Save IOAPIC IRQ */
633 entry->msi_attrib.pos = dev->msi_cap;
634
635 if (control & PCI_MSI_FLAGS_64BIT)
636 entry->mask_pos = dev->msi_cap + PCI_MSI_MASK_64;
637 else
638 entry->mask_pos = dev->msi_cap + PCI_MSI_MASK_32;
639 /* All MSIs are unmasked by default, Mask them all */
640 if (entry->msi_attrib.maskbit)
641 pci_read_config_dword(dev, entry->mask_pos, &entry->masked);
642 mask = msi_capable_mask(control);
643 msi_mask_irq(entry, mask, mask);
644
645 list_add_tail(&entry->list, &dev->msi_list);
646
647 /* Configure MSI capability structure */
648 ret = arch_setup_msi_irqs(dev, nvec, PCI_CAP_ID_MSI);
649 if (ret) {
650 msi_mask_irq(entry, mask, ~mask);
651 free_msi_irqs(dev);
652 return ret;
653 }
654
655 ret = populate_msi_sysfs(dev);
656 if (ret) {
657 msi_mask_irq(entry, mask, ~mask);
658 free_msi_irqs(dev);
659 return ret;
660 }
661
662 /* Set MSI enabled bits */
663 pci_intx_for_msi(dev, 0);
664 msi_set_enable(dev, 1);
665 dev->msi_enabled = 1;
666
667 dev->irq = entry->irq;
668 return 0;
669}
670
671static void __iomem *msix_map_region(struct pci_dev *dev, unsigned nr_entries)
672{
673 resource_size_t phys_addr;
674 u32 table_offset;
675 u8 bir;
676
677 pci_read_config_dword(dev, dev->msix_cap + PCI_MSIX_TABLE,
678 &table_offset);
679 bir = (u8)(table_offset & PCI_MSIX_TABLE_BIR);
680 table_offset &= PCI_MSIX_TABLE_OFFSET;
681 phys_addr = pci_resource_start(dev, bir) + table_offset;
682
683 return ioremap_nocache(phys_addr, nr_entries * PCI_MSIX_ENTRY_SIZE);
684}
685
686static int msix_setup_entries(struct pci_dev *dev, void __iomem *base,
687 struct msix_entry *entries, int nvec)
688{
689 struct msi_desc *entry;
690 int i;
691
692 for (i = 0; i < nvec; i++) {
693 entry = alloc_msi_entry(dev);
694 if (!entry) {
695 if (!i)
696 iounmap(base);
697 else
698 free_msi_irqs(dev);
699 /* No enough memory. Don't try again */
700 return -ENOMEM;
701 }
702
703 entry->msi_attrib.is_msix = 1;
704 entry->msi_attrib.is_64 = 1;
705 entry->msi_attrib.entry_nr = entries[i].entry;
706 entry->msi_attrib.default_irq = dev->irq;
707 entry->msi_attrib.pos = dev->msix_cap;
708 entry->mask_base = base;
709
710 list_add_tail(&entry->list, &dev->msi_list);
711 }
712
713 return 0;
714}
715
716static void msix_program_entries(struct pci_dev *dev,
717 struct msix_entry *entries)
718{
719 struct msi_desc *entry;
720 int i = 0;
721
722 list_for_each_entry(entry, &dev->msi_list, list) {
723 int offset = entries[i].entry * PCI_MSIX_ENTRY_SIZE +
724 PCI_MSIX_ENTRY_VECTOR_CTRL;
725
726 entries[i].vector = entry->irq;
727 irq_set_msi_desc(entry->irq, entry);
728 entry->masked = readl(entry->mask_base + offset);
729 msix_mask_irq(entry, 1);
730 i++;
731 }
732}
733
734/**
735 * msix_capability_init - configure device's MSI-X capability
736 * @dev: pointer to the pci_dev data structure of MSI-X device function
737 * @entries: pointer to an array of struct msix_entry entries
738 * @nvec: number of @entries
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,
745 struct msix_entry *entries, int nvec)
746{
747 int ret;
748 u16 control;
749 void __iomem *base;
750
751 pci_read_config_word(dev, dev->msix_cap + PCI_MSIX_FLAGS, &control);
752
753 /* Ensure MSI-X is disabled while it is set up */
754 control &= ~PCI_MSIX_FLAGS_ENABLE;
755 pci_write_config_word(dev, dev->msix_cap + PCI_MSIX_FLAGS, control);
756
757 /* Request & Map MSI-X table region */
758 base = msix_map_region(dev, msix_table_size(control));
759 if (!base)
760 return -ENOMEM;
761
762 ret = msix_setup_entries(dev, base, entries, nvec);
763 if (ret)
764 return ret;
765
766 ret = arch_setup_msi_irqs(dev, nvec, PCI_CAP_ID_MSIX);
767 if (ret)
768 goto out_avail;
769
770 /*
771 * Some devices require MSI-X to be enabled before we can touch the
772 * MSI-X registers. We need to mask all the vectors to prevent
773 * interrupts coming in before they're fully set up.
774 */
775 control |= PCI_MSIX_FLAGS_MASKALL | PCI_MSIX_FLAGS_ENABLE;
776 pci_write_config_word(dev, dev->msix_cap + PCI_MSIX_FLAGS, control);
777
778 msix_program_entries(dev, entries);
779
780 ret = populate_msi_sysfs(dev);
781 if (ret)
782 goto out_free;
783
784 /* Set MSI-X enabled bits and unmask the function */
785 pci_intx_for_msi(dev, 0);
786 dev->msix_enabled = 1;
787
788 control &= ~PCI_MSIX_FLAGS_MASKALL;
789 pci_write_config_word(dev, dev->msix_cap + PCI_MSIX_FLAGS, control);
790
791 return 0;
792
793out_avail:
794 if (ret < 0) {
795 /*
796 * If we had some success, report the number of irqs
797 * we succeeded in setting up.
798 */
799 struct msi_desc *entry;
800 int avail = 0;
801
802 list_for_each_entry(entry, &dev->msi_list, list) {
803 if (entry->irq != 0)
804 avail++;
805 }
806 if (avail != 0)
807 ret = avail;
808 }
809
810out_free:
811 free_msi_irqs(dev);
812
813 return ret;
814}
815
816/**
817 * pci_msi_check_device - check whether MSI may be enabled on a device
818 * @dev: pointer to the pci_dev data structure of MSI device function
819 * @nvec: how many MSIs have been requested ?
820 * @type: are we checking for MSI or MSI-X ?
821 *
822 * Look at global flags, the device itself, and its parent buses
823 * to determine if MSI/-X are supported for the device. If MSI/-X is
824 * supported return 0, else return an error code.
825 **/
826static int pci_msi_check_device(struct pci_dev *dev, int nvec, int type)
827{
828 struct pci_bus *bus;
829 int ret;
830
831 /* MSI must be globally enabled and supported by the device */
832 if (!pci_msi_enable || !dev || dev->no_msi)
833 return -EINVAL;
834
835 /*
836 * You can't ask to have 0 or less MSIs configured.
837 * a) it's stupid ..
838 * b) the list manipulation code assumes nvec >= 1.
839 */
840 if (nvec < 1)
841 return -ERANGE;
842
843 /*
844 * Any bridge which does NOT route MSI transactions from its
845 * secondary bus to its primary bus must set NO_MSI flag on
846 * the secondary pci_bus.
847 * We expect only arch-specific PCI host bus controller driver
848 * or quirks for specific PCI bridges to be setting NO_MSI.
849 */
850 for (bus = dev->bus; bus; bus = bus->parent)
851 if (bus->bus_flags & PCI_BUS_FLAGS_NO_MSI)
852 return -EINVAL;
853
854 ret = arch_msi_check_device(dev, nvec, type);
855 if (ret)
856 return ret;
857
858 return 0;
859}
860
861/**
862 * pci_msi_vec_count - Return the number of MSI vectors a device can send
863 * @dev: device to report about
864 *
865 * This function returns the number of MSI vectors a device requested via
866 * Multiple Message Capable register. It returns a negative errno if the
867 * device is not capable sending MSI interrupts. Otherwise, the call succeeds
868 * and returns a power of two, up to a maximum of 2^5 (32), according to the
869 * MSI specification.
870 **/
871int pci_msi_vec_count(struct pci_dev *dev)
872{
873 int ret;
874 u16 msgctl;
875
876 if (!dev->msi_cap)
877 return -EINVAL;
878
879 pci_read_config_word(dev, dev->msi_cap + PCI_MSI_FLAGS, &msgctl);
880 ret = 1 << ((msgctl & PCI_MSI_FLAGS_QMASK) >> 1);
881
882 return ret;
883}
884EXPORT_SYMBOL(pci_msi_vec_count);
885
886/**
887 * pci_enable_msi_block - configure device's MSI capability structure
888 * @dev: device to configure
889 * @nvec: number of interrupts to configure
890 *
891 * Allocate IRQs for a device with the MSI capability.
892 * This function returns a negative errno if an error occurs. If it
893 * is unable to allocate the number of interrupts requested, it returns
894 * the number of interrupts it might be able to allocate. If it successfully
895 * allocates at least the number of interrupts requested, it returns 0 and
896 * updates the @dev's irq member to the lowest new interrupt number; the
897 * other interrupt numbers allocated to this device are consecutive.
898 */
899int pci_enable_msi_block(struct pci_dev *dev, int nvec)
900{
901 int status, maxvec;
902
903 if (dev->current_state != PCI_D0)
904 return -EINVAL;
905
906 maxvec = pci_msi_vec_count(dev);
907 if (maxvec < 0)
908 return maxvec;
909 if (nvec > maxvec)
910 return maxvec;
911
912 status = pci_msi_check_device(dev, nvec, PCI_CAP_ID_MSI);
913 if (status)
914 return status;
915
916 WARN_ON(!!dev->msi_enabled);
917
918 /* Check whether driver already requested MSI-X irqs */
919 if (dev->msix_enabled) {
920 dev_info(&dev->dev, "can't enable MSI "
921 "(MSI-X already enabled)\n");
922 return -EINVAL;
923 }
924
925 status = msi_capability_init(dev, nvec);
926 return status;
927}
928EXPORT_SYMBOL(pci_enable_msi_block);
929
930void pci_msi_shutdown(struct pci_dev *dev)
931{
932 struct msi_desc *desc;
933 u32 mask;
934 u16 ctrl;
935
936 if (!pci_msi_enable || !dev || !dev->msi_enabled)
937 return;
938
939 BUG_ON(list_empty(&dev->msi_list));
940 desc = list_first_entry(&dev->msi_list, struct msi_desc, list);
941
942 msi_set_enable(dev, 0);
943 pci_intx_for_msi(dev, 1);
944 dev->msi_enabled = 0;
945
946 /* Return the device with MSI unmasked as initial states */
947 pci_read_config_word(dev, dev->msi_cap + PCI_MSI_FLAGS, &ctrl);
948 mask = msi_capable_mask(ctrl);
949 /* Keep cached state to be restored */
950 arch_msi_mask_irq(desc, mask, ~mask);
951
952 /* Restore dev->irq to its default pin-assertion irq */
953 dev->irq = desc->msi_attrib.default_irq;
954}
955
956void pci_disable_msi(struct pci_dev *dev)
957{
958 if (!pci_msi_enable || !dev || !dev->msi_enabled)
959 return;
960
961 pci_msi_shutdown(dev);
962 free_msi_irqs(dev);
963}
964EXPORT_SYMBOL(pci_disable_msi);
965
966/**
967 * pci_msix_vec_count - return the number of device's MSI-X table entries
968 * @dev: pointer to the pci_dev data structure of MSI-X device function
969 * This function returns the number of device's MSI-X table entries and
970 * therefore the number of MSI-X vectors device is capable of sending.
971 * It returns a negative errno if the device is not capable of sending MSI-X
972 * interrupts.
973 **/
974int pci_msix_vec_count(struct pci_dev *dev)
975{
976 u16 control;
977
978 if (!dev->msix_cap)
979 return -EINVAL;
980
981 pci_read_config_word(dev, dev->msix_cap + PCI_MSIX_FLAGS, &control);
982 return msix_table_size(control);
983}
984EXPORT_SYMBOL(pci_msix_vec_count);
985
986/**
987 * pci_enable_msix - configure device's MSI-X capability structure
988 * @dev: pointer to the pci_dev data structure of MSI-X device function
989 * @entries: pointer to an array of MSI-X entries
990 * @nvec: number of MSI-X irqs requested for allocation by device driver
991 *
992 * Setup the MSI-X capability structure of device function with the number
993 * of requested irqs upon its software driver call to request for
994 * MSI-X mode enabled on its hardware device function. A return of zero
995 * indicates the successful configuration of MSI-X capability structure
996 * with new allocated MSI-X irqs. A return of < 0 indicates a failure.
997 * Or a return of > 0 indicates that driver request is exceeding the number
998 * of irqs or MSI-X vectors available. Driver should use the returned value to
999 * re-send its request.
1000 **/
1001int pci_enable_msix(struct pci_dev *dev, struct msix_entry *entries, int nvec)
1002{
1003 int status, nr_entries;
1004 int i, j;
1005
1006 if (!entries || !dev->msix_cap || dev->current_state != PCI_D0)
1007 return -EINVAL;
1008
1009 status = pci_msi_check_device(dev, nvec, PCI_CAP_ID_MSIX);
1010 if (status)
1011 return status;
1012
1013 nr_entries = pci_msix_vec_count(dev);
1014 if (nr_entries < 0)
1015 return nr_entries;
1016 if (nvec > nr_entries)
1017 return nr_entries;
1018
1019 /* Check for any invalid entries */
1020 for (i = 0; i < nvec; i++) {
1021 if (entries[i].entry >= nr_entries)
1022 return -EINVAL; /* invalid entry */
1023 for (j = i + 1; j < nvec; j++) {
1024 if (entries[i].entry == entries[j].entry)
1025 return -EINVAL; /* duplicate entry */
1026 }
1027 }
1028 WARN_ON(!!dev->msix_enabled);
1029
1030 /* Check whether driver already requested for MSI irq */
1031 if (dev->msi_enabled) {
1032 dev_info(&dev->dev, "can't enable MSI-X "
1033 "(MSI IRQ already assigned)\n");
1034 return -EINVAL;
1035 }
1036 status = msix_capability_init(dev, entries, nvec);
1037 return status;
1038}
1039EXPORT_SYMBOL(pci_enable_msix);
1040
1041void pci_msix_shutdown(struct pci_dev *dev)
1042{
1043 struct msi_desc *entry;
1044
1045 if (!pci_msi_enable || !dev || !dev->msix_enabled)
1046 return;
1047
1048 /* Return the device with MSI-X masked as initial states */
1049 list_for_each_entry(entry, &dev->msi_list, list) {
1050 /* Keep cached states to be restored */
1051 arch_msix_mask_irq(entry, 1);
1052 }
1053
1054 msix_set_enable(dev, 0);
1055 pci_intx_for_msi(dev, 1);
1056 dev->msix_enabled = 0;
1057}
1058
1059void pci_disable_msix(struct pci_dev *dev)
1060{
1061 if (!pci_msi_enable || !dev || !dev->msix_enabled)
1062 return;
1063
1064 pci_msix_shutdown(dev);
1065 free_msi_irqs(dev);
1066}
1067EXPORT_SYMBOL(pci_disable_msix);
1068
1069/**
1070 * msi_remove_pci_irq_vectors - reclaim MSI(X) irqs to unused state
1071 * @dev: pointer to the pci_dev data structure of MSI(X) device function
1072 *
1073 * Being called during hotplug remove, from which the device function
1074 * is hot-removed. All previous assigned MSI/MSI-X irqs, if
1075 * allocated for this device function, are reclaimed to unused state,
1076 * which may be used later on.
1077 **/
1078void msi_remove_pci_irq_vectors(struct pci_dev *dev)
1079{
1080 if (!pci_msi_enable || !dev)
1081 return;
1082
1083 if (dev->msi_enabled || dev->msix_enabled)
1084 free_msi_irqs(dev);
1085}
1086
1087void pci_no_msi(void)
1088{
1089 pci_msi_enable = 0;
1090}
1091
1092/**
1093 * pci_msi_enabled - is MSI enabled?
1094 *
1095 * Returns true if MSI has not been disabled by the command-line option
1096 * pci=nomsi.
1097 **/
1098int pci_msi_enabled(void)
1099{
1100 return pci_msi_enable;
1101}
1102EXPORT_SYMBOL(pci_msi_enabled);
1103
1104void pci_msi_init_pci_dev(struct pci_dev *dev)
1105{
1106 INIT_LIST_HEAD(&dev->msi_list);
1107
1108 /* Disable the msi hardware to avoid screaming interrupts
1109 * during boot. This is the power on reset default so
1110 * usually this should be a noop.
1111 */
1112 dev->msi_cap = pci_find_capability(dev, PCI_CAP_ID_MSI);
1113 if (dev->msi_cap)
1114 msi_set_enable(dev, 0);
1115
1116 dev->msix_cap = pci_find_capability(dev, PCI_CAP_ID_MSIX);
1117 if (dev->msix_cap)
1118 msix_set_enable(dev, 0);
1119}
1120
1121/**
1122 * pci_enable_msi_range - configure device's MSI capability structure
1123 * @dev: device to configure
1124 * @minvec: minimal number of interrupts to configure
1125 * @maxvec: maximum number of interrupts to configure
1126 *
1127 * This function tries to allocate a maximum possible number of interrupts in a
1128 * range between @minvec and @maxvec. It returns a negative errno if an error
1129 * occurs. If it succeeds, it returns the actual number of interrupts allocated
1130 * and updates the @dev's irq member to the lowest new interrupt number;
1131 * the other interrupt numbers allocated to this device are consecutive.
1132 **/
1133int pci_enable_msi_range(struct pci_dev *dev, int minvec, int maxvec)
1134{
1135 int nvec = maxvec;
1136 int rc;
1137
1138 if (maxvec < minvec)
1139 return -ERANGE;
1140
1141 do {
1142 rc = pci_enable_msi_block(dev, nvec);
1143 if (rc < 0) {
1144 return rc;
1145 } else if (rc > 0) {
1146 if (rc < minvec)
1147 return -ENOSPC;
1148 nvec = rc;
1149 }
1150 } while (rc);
1151
1152 return nvec;
1153}
1154EXPORT_SYMBOL(pci_enable_msi_range);
1155
1156/**
1157 * pci_enable_msix_range - configure device's MSI-X capability structure
1158 * @dev: pointer to the pci_dev data structure of MSI-X device function
1159 * @entries: pointer to an array of MSI-X entries
1160 * @minvec: minimum number of MSI-X irqs requested
1161 * @maxvec: maximum number of MSI-X irqs requested
1162 *
1163 * Setup the MSI-X capability structure of device function with a maximum
1164 * possible number of interrupts in the range between @minvec and @maxvec
1165 * upon its software driver call to request for MSI-X mode enabled on its
1166 * hardware device function. It returns a negative errno if an error occurs.
1167 * If it succeeds, it returns the actual number of interrupts allocated and
1168 * indicates the successful configuration of MSI-X capability structure
1169 * with new allocated MSI-X interrupts.
1170 **/
1171int pci_enable_msix_range(struct pci_dev *dev, struct msix_entry *entries,
1172 int minvec, int maxvec)
1173{
1174 int nvec = maxvec;
1175 int rc;
1176
1177 if (maxvec < minvec)
1178 return -ERANGE;
1179
1180 do {
1181 rc = pci_enable_msix(dev, entries, nvec);
1182 if (rc < 0) {
1183 return rc;
1184 } else if (rc > 0) {
1185 if (rc < minvec)
1186 return -ENOSPC;
1187 nvec = rc;
1188 }
1189 } while (rc);
1190
1191 return nvec;
1192}
1193EXPORT_SYMBOL(pci_enable_msix_range);