Loading...
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#include "msi.h"
26
27static int pci_msi_enable = 1;
28
29/* Arch hooks */
30
31#ifndef arch_msi_check_device
32int arch_msi_check_device(struct pci_dev *dev, int nvec, int type)
33{
34 return 0;
35}
36#endif
37
38#ifndef arch_setup_msi_irqs
39# define arch_setup_msi_irqs default_setup_msi_irqs
40# define HAVE_DEFAULT_MSI_SETUP_IRQS
41#endif
42
43#ifdef HAVE_DEFAULT_MSI_SETUP_IRQS
44int default_setup_msi_irqs(struct pci_dev *dev, int nvec, int type)
45{
46 struct msi_desc *entry;
47 int ret;
48
49 /*
50 * If an architecture wants to support multiple MSI, it needs to
51 * override arch_setup_msi_irqs()
52 */
53 if (type == PCI_CAP_ID_MSI && nvec > 1)
54 return 1;
55
56 list_for_each_entry(entry, &dev->msi_list, list) {
57 ret = arch_setup_msi_irq(dev, entry);
58 if (ret < 0)
59 return ret;
60 if (ret > 0)
61 return -ENOSPC;
62 }
63
64 return 0;
65}
66#endif
67
68#ifndef arch_teardown_msi_irqs
69# define arch_teardown_msi_irqs default_teardown_msi_irqs
70# define HAVE_DEFAULT_MSI_TEARDOWN_IRQS
71#endif
72
73#ifdef HAVE_DEFAULT_MSI_TEARDOWN_IRQS
74void default_teardown_msi_irqs(struct pci_dev *dev)
75{
76 struct msi_desc *entry;
77
78 list_for_each_entry(entry, &dev->msi_list, list) {
79 int i, nvec;
80 if (entry->irq == 0)
81 continue;
82 nvec = 1 << entry->msi_attrib.multiple;
83 for (i = 0; i < nvec; i++)
84 arch_teardown_msi_irq(entry->irq + i);
85 }
86}
87#endif
88
89#ifndef arch_restore_msi_irqs
90# define arch_restore_msi_irqs default_restore_msi_irqs
91# define HAVE_DEFAULT_MSI_RESTORE_IRQS
92#endif
93
94#ifdef HAVE_DEFAULT_MSI_RESTORE_IRQS
95void default_restore_msi_irqs(struct pci_dev *dev, int irq)
96{
97 struct msi_desc *entry;
98
99 entry = NULL;
100 if (dev->msix_enabled) {
101 list_for_each_entry(entry, &dev->msi_list, list) {
102 if (irq == entry->irq)
103 break;
104 }
105 } else if (dev->msi_enabled) {
106 entry = irq_get_msi_desc(irq);
107 }
108
109 if (entry)
110 write_msi_msg(irq, &entry->msg);
111}
112#endif
113
114static void msi_set_enable(struct pci_dev *dev, int pos, int enable)
115{
116 u16 control;
117
118 BUG_ON(!pos);
119
120 pci_read_config_word(dev, pos + PCI_MSI_FLAGS, &control);
121 control &= ~PCI_MSI_FLAGS_ENABLE;
122 if (enable)
123 control |= PCI_MSI_FLAGS_ENABLE;
124 pci_write_config_word(dev, pos + PCI_MSI_FLAGS, control);
125}
126
127static void msix_set_enable(struct pci_dev *dev, int enable)
128{
129 int pos;
130 u16 control;
131
132 pos = pci_find_capability(dev, PCI_CAP_ID_MSIX);
133 if (pos) {
134 pci_read_config_word(dev, pos + PCI_MSIX_FLAGS, &control);
135 control &= ~PCI_MSIX_FLAGS_ENABLE;
136 if (enable)
137 control |= PCI_MSIX_FLAGS_ENABLE;
138 pci_write_config_word(dev, pos + PCI_MSIX_FLAGS, control);
139 }
140}
141
142static inline __attribute_const__ u32 msi_mask(unsigned x)
143{
144 /* Don't shift by >= width of type */
145 if (x >= 5)
146 return 0xffffffff;
147 return (1 << (1 << x)) - 1;
148}
149
150static inline __attribute_const__ u32 msi_capable_mask(u16 control)
151{
152 return msi_mask((control >> 1) & 7);
153}
154
155static inline __attribute_const__ u32 msi_enabled_mask(u16 control)
156{
157 return msi_mask((control >> 4) & 7);
158}
159
160/*
161 * PCI 2.3 does not specify mask bits for each MSI interrupt. Attempting to
162 * mask all MSI interrupts by clearing the MSI enable bit does not work
163 * reliably as devices without an INTx disable bit will then generate a
164 * level IRQ which will never be cleared.
165 */
166static u32 __msi_mask_irq(struct msi_desc *desc, u32 mask, u32 flag)
167{
168 u32 mask_bits = desc->masked;
169
170 if (!desc->msi_attrib.maskbit)
171 return 0;
172
173 mask_bits &= ~mask;
174 mask_bits |= flag;
175 pci_write_config_dword(desc->dev, desc->mask_pos, mask_bits);
176
177 return mask_bits;
178}
179
180static void msi_mask_irq(struct msi_desc *desc, u32 mask, u32 flag)
181{
182 desc->masked = __msi_mask_irq(desc, mask, flag);
183}
184
185/*
186 * This internal function does not flush PCI writes to the device.
187 * All users must ensure that they read from the device before either
188 * assuming that the device state is up to date, or returning out of this
189 * file. This saves a few milliseconds when initialising devices with lots
190 * of MSI-X interrupts.
191 */
192static u32 __msix_mask_irq(struct msi_desc *desc, u32 flag)
193{
194 u32 mask_bits = desc->masked;
195 unsigned offset = desc->msi_attrib.entry_nr * PCI_MSIX_ENTRY_SIZE +
196 PCI_MSIX_ENTRY_VECTOR_CTRL;
197 mask_bits &= ~PCI_MSIX_ENTRY_CTRL_MASKBIT;
198 if (flag)
199 mask_bits |= PCI_MSIX_ENTRY_CTRL_MASKBIT;
200 writel(mask_bits, desc->mask_base + offset);
201
202 return mask_bits;
203}
204
205static void msix_mask_irq(struct msi_desc *desc, u32 flag)
206{
207 desc->masked = __msix_mask_irq(desc, flag);
208}
209
210static void msi_set_mask_bit(struct irq_data *data, u32 flag)
211{
212 struct msi_desc *desc = irq_data_get_msi(data);
213
214 if (desc->msi_attrib.is_msix) {
215 msix_mask_irq(desc, flag);
216 readl(desc->mask_base); /* Flush write to device */
217 } else {
218 unsigned offset = data->irq - desc->dev->irq;
219 msi_mask_irq(desc, 1 << offset, flag << offset);
220 }
221}
222
223void mask_msi_irq(struct irq_data *data)
224{
225 msi_set_mask_bit(data, 1);
226}
227
228void unmask_msi_irq(struct irq_data *data)
229{
230 msi_set_mask_bit(data, 0);
231}
232
233void __read_msi_msg(struct msi_desc *entry, struct msi_msg *msg)
234{
235 BUG_ON(entry->dev->current_state != PCI_D0);
236
237 if (entry->msi_attrib.is_msix) {
238 void __iomem *base = entry->mask_base +
239 entry->msi_attrib.entry_nr * PCI_MSIX_ENTRY_SIZE;
240
241 msg->address_lo = readl(base + PCI_MSIX_ENTRY_LOWER_ADDR);
242 msg->address_hi = readl(base + PCI_MSIX_ENTRY_UPPER_ADDR);
243 msg->data = readl(base + PCI_MSIX_ENTRY_DATA);
244 } else {
245 struct pci_dev *dev = entry->dev;
246 int pos = entry->msi_attrib.pos;
247 u16 data;
248
249 pci_read_config_dword(dev, msi_lower_address_reg(pos),
250 &msg->address_lo);
251 if (entry->msi_attrib.is_64) {
252 pci_read_config_dword(dev, msi_upper_address_reg(pos),
253 &msg->address_hi);
254 pci_read_config_word(dev, msi_data_reg(pos, 1), &data);
255 } else {
256 msg->address_hi = 0;
257 pci_read_config_word(dev, msi_data_reg(pos, 0), &data);
258 }
259 msg->data = data;
260 }
261}
262
263void read_msi_msg(unsigned int irq, struct msi_msg *msg)
264{
265 struct msi_desc *entry = irq_get_msi_desc(irq);
266
267 __read_msi_msg(entry, msg);
268}
269
270void __get_cached_msi_msg(struct msi_desc *entry, struct msi_msg *msg)
271{
272 /* Assert that the cache is valid, assuming that
273 * valid messages are not all-zeroes. */
274 BUG_ON(!(entry->msg.address_hi | entry->msg.address_lo |
275 entry->msg.data));
276
277 *msg = entry->msg;
278}
279
280void get_cached_msi_msg(unsigned int irq, struct msi_msg *msg)
281{
282 struct msi_desc *entry = irq_get_msi_desc(irq);
283
284 __get_cached_msi_msg(entry, msg);
285}
286
287void __write_msi_msg(struct msi_desc *entry, struct msi_msg *msg)
288{
289 if (entry->dev->current_state != PCI_D0) {
290 /* Don't touch the hardware now */
291 } else if (entry->msi_attrib.is_msix) {
292 void __iomem *base;
293 base = entry->mask_base +
294 entry->msi_attrib.entry_nr * PCI_MSIX_ENTRY_SIZE;
295
296 writel(msg->address_lo, base + PCI_MSIX_ENTRY_LOWER_ADDR);
297 writel(msg->address_hi, base + PCI_MSIX_ENTRY_UPPER_ADDR);
298 writel(msg->data, base + PCI_MSIX_ENTRY_DATA);
299 } else {
300 struct pci_dev *dev = entry->dev;
301 int pos = entry->msi_attrib.pos;
302 u16 msgctl;
303
304 pci_read_config_word(dev, msi_control_reg(pos), &msgctl);
305 msgctl &= ~PCI_MSI_FLAGS_QSIZE;
306 msgctl |= entry->msi_attrib.multiple << 4;
307 pci_write_config_word(dev, msi_control_reg(pos), msgctl);
308
309 pci_write_config_dword(dev, msi_lower_address_reg(pos),
310 msg->address_lo);
311 if (entry->msi_attrib.is_64) {
312 pci_write_config_dword(dev, msi_upper_address_reg(pos),
313 msg->address_hi);
314 pci_write_config_word(dev, msi_data_reg(pos, 1),
315 msg->data);
316 } else {
317 pci_write_config_word(dev, msi_data_reg(pos, 0),
318 msg->data);
319 }
320 }
321 entry->msg = *msg;
322}
323
324void write_msi_msg(unsigned int irq, struct msi_msg *msg)
325{
326 struct msi_desc *entry = irq_get_msi_desc(irq);
327
328 __write_msi_msg(entry, msg);
329}
330
331static void free_msi_irqs(struct pci_dev *dev)
332{
333 struct msi_desc *entry, *tmp;
334
335 list_for_each_entry(entry, &dev->msi_list, list) {
336 int i, nvec;
337 if (!entry->irq)
338 continue;
339 nvec = 1 << entry->msi_attrib.multiple;
340 for (i = 0; i < nvec; i++)
341 BUG_ON(irq_has_action(entry->irq + i));
342 }
343
344 arch_teardown_msi_irqs(dev);
345
346 list_for_each_entry_safe(entry, tmp, &dev->msi_list, list) {
347 if (entry->msi_attrib.is_msix) {
348 if (list_is_last(&entry->list, &dev->msi_list))
349 iounmap(entry->mask_base);
350 }
351
352 /*
353 * Its possible that we get into this path
354 * When populate_msi_sysfs fails, which means the entries
355 * were not registered with sysfs. In that case don't
356 * unregister them.
357 */
358 if (entry->kobj.parent) {
359 kobject_del(&entry->kobj);
360 kobject_put(&entry->kobj);
361 }
362
363 list_del(&entry->list);
364 kfree(entry);
365 }
366}
367
368static struct msi_desc *alloc_msi_entry(struct pci_dev *dev)
369{
370 struct msi_desc *desc = kzalloc(sizeof(*desc), GFP_KERNEL);
371 if (!desc)
372 return NULL;
373
374 INIT_LIST_HEAD(&desc->list);
375 desc->dev = dev;
376
377 return desc;
378}
379
380static void pci_intx_for_msi(struct pci_dev *dev, int enable)
381{
382 if (!(dev->dev_flags & PCI_DEV_FLAGS_MSI_INTX_DISABLE_BUG))
383 pci_intx(dev, enable);
384}
385
386static void __pci_restore_msi_state(struct pci_dev *dev)
387{
388 int pos;
389 u16 control;
390 struct msi_desc *entry;
391
392 if (!dev->msi_enabled)
393 return;
394
395 entry = irq_get_msi_desc(dev->irq);
396 pos = entry->msi_attrib.pos;
397
398 pci_intx_for_msi(dev, 0);
399 msi_set_enable(dev, pos, 0);
400 arch_restore_msi_irqs(dev, dev->irq);
401
402 pci_read_config_word(dev, pos + PCI_MSI_FLAGS, &control);
403 msi_mask_irq(entry, msi_capable_mask(control), entry->masked);
404 control &= ~PCI_MSI_FLAGS_QSIZE;
405 control |= (entry->msi_attrib.multiple << 4) | PCI_MSI_FLAGS_ENABLE;
406 pci_write_config_word(dev, pos + PCI_MSI_FLAGS, control);
407}
408
409static void __pci_restore_msix_state(struct pci_dev *dev)
410{
411 int pos;
412 struct msi_desc *entry;
413 u16 control;
414
415 if (!dev->msix_enabled)
416 return;
417 BUG_ON(list_empty(&dev->msi_list));
418 entry = list_first_entry(&dev->msi_list, struct msi_desc, list);
419 pos = entry->msi_attrib.pos;
420 pci_read_config_word(dev, pos + PCI_MSIX_FLAGS, &control);
421
422 /* route the table */
423 pci_intx_for_msi(dev, 0);
424 control |= PCI_MSIX_FLAGS_ENABLE | PCI_MSIX_FLAGS_MASKALL;
425 pci_write_config_word(dev, pos + PCI_MSIX_FLAGS, control);
426
427 list_for_each_entry(entry, &dev->msi_list, list) {
428 arch_restore_msi_irqs(dev, entry->irq);
429 msix_mask_irq(entry, entry->masked);
430 }
431
432 control &= ~PCI_MSIX_FLAGS_MASKALL;
433 pci_write_config_word(dev, pos + PCI_MSIX_FLAGS, control);
434}
435
436void pci_restore_msi_state(struct pci_dev *dev)
437{
438 __pci_restore_msi_state(dev);
439 __pci_restore_msix_state(dev);
440}
441EXPORT_SYMBOL_GPL(pci_restore_msi_state);
442
443
444#define to_msi_attr(obj) container_of(obj, struct msi_attribute, attr)
445#define to_msi_desc(obj) container_of(obj, struct msi_desc, kobj)
446
447struct msi_attribute {
448 struct attribute attr;
449 ssize_t (*show)(struct msi_desc *entry, struct msi_attribute *attr,
450 char *buf);
451 ssize_t (*store)(struct msi_desc *entry, struct msi_attribute *attr,
452 const char *buf, size_t count);
453};
454
455static ssize_t show_msi_mode(struct msi_desc *entry, struct msi_attribute *atr,
456 char *buf)
457{
458 return sprintf(buf, "%s\n", entry->msi_attrib.is_msix ? "msix" : "msi");
459}
460
461static ssize_t msi_irq_attr_show(struct kobject *kobj,
462 struct attribute *attr, char *buf)
463{
464 struct msi_attribute *attribute = to_msi_attr(attr);
465 struct msi_desc *entry = to_msi_desc(kobj);
466
467 if (!attribute->show)
468 return -EIO;
469
470 return attribute->show(entry, attribute, buf);
471}
472
473static const struct sysfs_ops msi_irq_sysfs_ops = {
474 .show = msi_irq_attr_show,
475};
476
477static struct msi_attribute mode_attribute =
478 __ATTR(mode, S_IRUGO, show_msi_mode, NULL);
479
480
481struct attribute *msi_irq_default_attrs[] = {
482 &mode_attribute.attr,
483 NULL
484};
485
486void msi_kobj_release(struct kobject *kobj)
487{
488 struct msi_desc *entry = to_msi_desc(kobj);
489
490 pci_dev_put(entry->dev);
491}
492
493static struct kobj_type msi_irq_ktype = {
494 .release = msi_kobj_release,
495 .sysfs_ops = &msi_irq_sysfs_ops,
496 .default_attrs = msi_irq_default_attrs,
497};
498
499static int populate_msi_sysfs(struct pci_dev *pdev)
500{
501 struct msi_desc *entry;
502 struct kobject *kobj;
503 int ret;
504 int count = 0;
505
506 pdev->msi_kset = kset_create_and_add("msi_irqs", NULL, &pdev->dev.kobj);
507 if (!pdev->msi_kset)
508 return -ENOMEM;
509
510 list_for_each_entry(entry, &pdev->msi_list, list) {
511 kobj = &entry->kobj;
512 kobj->kset = pdev->msi_kset;
513 pci_dev_get(pdev);
514 ret = kobject_init_and_add(kobj, &msi_irq_ktype, NULL,
515 "%u", entry->irq);
516 if (ret)
517 goto out_unroll;
518
519 count++;
520 }
521
522 return 0;
523
524out_unroll:
525 list_for_each_entry(entry, &pdev->msi_list, list) {
526 if (!count)
527 break;
528 kobject_del(&entry->kobj);
529 kobject_put(&entry->kobj);
530 count--;
531 }
532 return ret;
533}
534
535/**
536 * msi_capability_init - configure device's MSI capability structure
537 * @dev: pointer to the pci_dev data structure of MSI device function
538 * @nvec: number of interrupts to allocate
539 *
540 * Setup the MSI capability structure of the device with the requested
541 * number of interrupts. A return value of zero indicates the successful
542 * setup of an entry with the new MSI irq. A negative return value indicates
543 * an error, and a positive return value indicates the number of interrupts
544 * which could have been allocated.
545 */
546static int msi_capability_init(struct pci_dev *dev, int nvec)
547{
548 struct msi_desc *entry;
549 int pos, ret;
550 u16 control;
551 unsigned mask;
552
553 pos = pci_find_capability(dev, PCI_CAP_ID_MSI);
554 msi_set_enable(dev, pos, 0); /* Disable MSI during set up */
555
556 pci_read_config_word(dev, msi_control_reg(pos), &control);
557 /* MSI Entry Initialization */
558 entry = alloc_msi_entry(dev);
559 if (!entry)
560 return -ENOMEM;
561
562 entry->msi_attrib.is_msix = 0;
563 entry->msi_attrib.is_64 = is_64bit_address(control);
564 entry->msi_attrib.entry_nr = 0;
565 entry->msi_attrib.maskbit = is_mask_bit_support(control);
566 entry->msi_attrib.default_irq = dev->irq; /* Save IOAPIC IRQ */
567 entry->msi_attrib.pos = pos;
568
569 entry->mask_pos = msi_mask_reg(pos, entry->msi_attrib.is_64);
570 /* All MSIs are unmasked by default, Mask them all */
571 if (entry->msi_attrib.maskbit)
572 pci_read_config_dword(dev, entry->mask_pos, &entry->masked);
573 mask = msi_capable_mask(control);
574 msi_mask_irq(entry, mask, mask);
575
576 list_add_tail(&entry->list, &dev->msi_list);
577
578 /* Configure MSI capability structure */
579 ret = arch_setup_msi_irqs(dev, nvec, PCI_CAP_ID_MSI);
580 if (ret) {
581 msi_mask_irq(entry, mask, ~mask);
582 free_msi_irqs(dev);
583 return ret;
584 }
585
586 ret = populate_msi_sysfs(dev);
587 if (ret) {
588 msi_mask_irq(entry, mask, ~mask);
589 free_msi_irqs(dev);
590 return ret;
591 }
592
593 /* Set MSI enabled bits */
594 pci_intx_for_msi(dev, 0);
595 msi_set_enable(dev, pos, 1);
596 dev->msi_enabled = 1;
597
598 dev->irq = entry->irq;
599 return 0;
600}
601
602static void __iomem *msix_map_region(struct pci_dev *dev, unsigned pos,
603 unsigned nr_entries)
604{
605 resource_size_t phys_addr;
606 u32 table_offset;
607 u8 bir;
608
609 pci_read_config_dword(dev, msix_table_offset_reg(pos), &table_offset);
610 bir = (u8)(table_offset & PCI_MSIX_FLAGS_BIRMASK);
611 table_offset &= ~PCI_MSIX_FLAGS_BIRMASK;
612 phys_addr = pci_resource_start(dev, bir) + table_offset;
613
614 return ioremap_nocache(phys_addr, nr_entries * PCI_MSIX_ENTRY_SIZE);
615}
616
617static int msix_setup_entries(struct pci_dev *dev, unsigned pos,
618 void __iomem *base, struct msix_entry *entries,
619 int nvec)
620{
621 struct msi_desc *entry;
622 int i;
623
624 for (i = 0; i < nvec; i++) {
625 entry = alloc_msi_entry(dev);
626 if (!entry) {
627 if (!i)
628 iounmap(base);
629 else
630 free_msi_irqs(dev);
631 /* No enough memory. Don't try again */
632 return -ENOMEM;
633 }
634
635 entry->msi_attrib.is_msix = 1;
636 entry->msi_attrib.is_64 = 1;
637 entry->msi_attrib.entry_nr = entries[i].entry;
638 entry->msi_attrib.default_irq = dev->irq;
639 entry->msi_attrib.pos = pos;
640 entry->mask_base = base;
641
642 list_add_tail(&entry->list, &dev->msi_list);
643 }
644
645 return 0;
646}
647
648static void msix_program_entries(struct pci_dev *dev,
649 struct msix_entry *entries)
650{
651 struct msi_desc *entry;
652 int i = 0;
653
654 list_for_each_entry(entry, &dev->msi_list, list) {
655 int offset = entries[i].entry * PCI_MSIX_ENTRY_SIZE +
656 PCI_MSIX_ENTRY_VECTOR_CTRL;
657
658 entries[i].vector = entry->irq;
659 irq_set_msi_desc(entry->irq, entry);
660 entry->masked = readl(entry->mask_base + offset);
661 msix_mask_irq(entry, 1);
662 i++;
663 }
664}
665
666/**
667 * msix_capability_init - configure device's MSI-X capability
668 * @dev: pointer to the pci_dev data structure of MSI-X device function
669 * @entries: pointer to an array of struct msix_entry entries
670 * @nvec: number of @entries
671 *
672 * Setup the MSI-X capability structure of device function with a
673 * single MSI-X irq. A return of zero indicates the successful setup of
674 * requested MSI-X entries with allocated irqs or non-zero for otherwise.
675 **/
676static int msix_capability_init(struct pci_dev *dev,
677 struct msix_entry *entries, int nvec)
678{
679 int pos, ret;
680 u16 control;
681 void __iomem *base;
682
683 pos = pci_find_capability(dev, PCI_CAP_ID_MSIX);
684 pci_read_config_word(dev, pos + PCI_MSIX_FLAGS, &control);
685
686 /* Ensure MSI-X is disabled while it is set up */
687 control &= ~PCI_MSIX_FLAGS_ENABLE;
688 pci_write_config_word(dev, pos + PCI_MSIX_FLAGS, control);
689
690 /* Request & Map MSI-X table region */
691 base = msix_map_region(dev, pos, multi_msix_capable(control));
692 if (!base)
693 return -ENOMEM;
694
695 ret = msix_setup_entries(dev, pos, base, entries, nvec);
696 if (ret)
697 return ret;
698
699 ret = arch_setup_msi_irqs(dev, nvec, PCI_CAP_ID_MSIX);
700 if (ret)
701 goto error;
702
703 /*
704 * Some devices require MSI-X to be enabled before we can touch the
705 * MSI-X registers. We need to mask all the vectors to prevent
706 * interrupts coming in before they're fully set up.
707 */
708 control |= PCI_MSIX_FLAGS_MASKALL | PCI_MSIX_FLAGS_ENABLE;
709 pci_write_config_word(dev, pos + PCI_MSIX_FLAGS, control);
710
711 msix_program_entries(dev, entries);
712
713 ret = populate_msi_sysfs(dev);
714 if (ret) {
715 ret = 0;
716 goto error;
717 }
718
719 /* Set MSI-X enabled bits and unmask the function */
720 pci_intx_for_msi(dev, 0);
721 dev->msix_enabled = 1;
722
723 control &= ~PCI_MSIX_FLAGS_MASKALL;
724 pci_write_config_word(dev, pos + PCI_MSIX_FLAGS, control);
725
726 return 0;
727
728error:
729 if (ret < 0) {
730 /*
731 * If we had some success, report the number of irqs
732 * we succeeded in setting up.
733 */
734 struct msi_desc *entry;
735 int avail = 0;
736
737 list_for_each_entry(entry, &dev->msi_list, list) {
738 if (entry->irq != 0)
739 avail++;
740 }
741 if (avail != 0)
742 ret = avail;
743 }
744
745 free_msi_irqs(dev);
746
747 return ret;
748}
749
750/**
751 * pci_msi_check_device - check whether MSI may be enabled on a device
752 * @dev: pointer to the pci_dev data structure of MSI device function
753 * @nvec: how many MSIs have been requested ?
754 * @type: are we checking for MSI or MSI-X ?
755 *
756 * Look at global flags, the device itself, and its parent busses
757 * to determine if MSI/-X are supported for the device. If MSI/-X is
758 * supported return 0, else return an error code.
759 **/
760static int pci_msi_check_device(struct pci_dev *dev, int nvec, int type)
761{
762 struct pci_bus *bus;
763 int ret;
764
765 /* MSI must be globally enabled and supported by the device */
766 if (!pci_msi_enable || !dev || dev->no_msi)
767 return -EINVAL;
768
769 /*
770 * You can't ask to have 0 or less MSIs configured.
771 * a) it's stupid ..
772 * b) the list manipulation code assumes nvec >= 1.
773 */
774 if (nvec < 1)
775 return -ERANGE;
776
777 /*
778 * Any bridge which does NOT route MSI transactions from its
779 * secondary bus to its primary bus must set NO_MSI flag on
780 * the secondary pci_bus.
781 * We expect only arch-specific PCI host bus controller driver
782 * or quirks for specific PCI bridges to be setting NO_MSI.
783 */
784 for (bus = dev->bus; bus; bus = bus->parent)
785 if (bus->bus_flags & PCI_BUS_FLAGS_NO_MSI)
786 return -EINVAL;
787
788 ret = arch_msi_check_device(dev, nvec, type);
789 if (ret)
790 return ret;
791
792 if (!pci_find_capability(dev, type))
793 return -EINVAL;
794
795 return 0;
796}
797
798/**
799 * pci_enable_msi_block - configure device's MSI capability structure
800 * @dev: device to configure
801 * @nvec: number of interrupts to configure
802 *
803 * Allocate IRQs for a device with the MSI capability.
804 * This function returns a negative errno if an error occurs. If it
805 * is unable to allocate the number of interrupts requested, it returns
806 * the number of interrupts it might be able to allocate. If it successfully
807 * allocates at least the number of interrupts requested, it returns 0 and
808 * updates the @dev's irq member to the lowest new interrupt number; the
809 * other interrupt numbers allocated to this device are consecutive.
810 */
811int pci_enable_msi_block(struct pci_dev *dev, unsigned int nvec)
812{
813 int status, pos, maxvec;
814 u16 msgctl;
815
816 pos = pci_find_capability(dev, PCI_CAP_ID_MSI);
817 if (!pos)
818 return -EINVAL;
819 pci_read_config_word(dev, pos + PCI_MSI_FLAGS, &msgctl);
820 maxvec = 1 << ((msgctl & PCI_MSI_FLAGS_QMASK) >> 1);
821 if (nvec > maxvec)
822 return maxvec;
823
824 status = pci_msi_check_device(dev, nvec, PCI_CAP_ID_MSI);
825 if (status)
826 return status;
827
828 WARN_ON(!!dev->msi_enabled);
829
830 /* Check whether driver already requested MSI-X irqs */
831 if (dev->msix_enabled) {
832 dev_info(&dev->dev, "can't enable MSI "
833 "(MSI-X already enabled)\n");
834 return -EINVAL;
835 }
836
837 status = msi_capability_init(dev, nvec);
838 return status;
839}
840EXPORT_SYMBOL(pci_enable_msi_block);
841
842void pci_msi_shutdown(struct pci_dev *dev)
843{
844 struct msi_desc *desc;
845 u32 mask;
846 u16 ctrl;
847 unsigned pos;
848
849 if (!pci_msi_enable || !dev || !dev->msi_enabled)
850 return;
851
852 BUG_ON(list_empty(&dev->msi_list));
853 desc = list_first_entry(&dev->msi_list, struct msi_desc, list);
854 pos = desc->msi_attrib.pos;
855
856 msi_set_enable(dev, pos, 0);
857 pci_intx_for_msi(dev, 1);
858 dev->msi_enabled = 0;
859
860 /* Return the device with MSI unmasked as initial states */
861 pci_read_config_word(dev, pos + PCI_MSI_FLAGS, &ctrl);
862 mask = msi_capable_mask(ctrl);
863 /* Keep cached state to be restored */
864 __msi_mask_irq(desc, mask, ~mask);
865
866 /* Restore dev->irq to its default pin-assertion irq */
867 dev->irq = desc->msi_attrib.default_irq;
868}
869
870void pci_disable_msi(struct pci_dev *dev)
871{
872 if (!pci_msi_enable || !dev || !dev->msi_enabled)
873 return;
874
875 pci_msi_shutdown(dev);
876 free_msi_irqs(dev);
877 kset_unregister(dev->msi_kset);
878 dev->msi_kset = NULL;
879}
880EXPORT_SYMBOL(pci_disable_msi);
881
882/**
883 * pci_msix_table_size - return the number of device's MSI-X table entries
884 * @dev: pointer to the pci_dev data structure of MSI-X device function
885 */
886int pci_msix_table_size(struct pci_dev *dev)
887{
888 int pos;
889 u16 control;
890
891 pos = pci_find_capability(dev, PCI_CAP_ID_MSIX);
892 if (!pos)
893 return 0;
894
895 pci_read_config_word(dev, msi_control_reg(pos), &control);
896 return multi_msix_capable(control);
897}
898
899/**
900 * pci_enable_msix - configure device's MSI-X capability structure
901 * @dev: pointer to the pci_dev data structure of MSI-X device function
902 * @entries: pointer to an array of MSI-X entries
903 * @nvec: number of MSI-X irqs requested for allocation by device driver
904 *
905 * Setup the MSI-X capability structure of device function with the number
906 * of requested irqs upon its software driver call to request for
907 * MSI-X mode enabled on its hardware device function. A return of zero
908 * indicates the successful configuration of MSI-X capability structure
909 * with new allocated MSI-X irqs. A return of < 0 indicates a failure.
910 * Or a return of > 0 indicates that driver request is exceeding the number
911 * of irqs or MSI-X vectors available. Driver should use the returned value to
912 * re-send its request.
913 **/
914int pci_enable_msix(struct pci_dev *dev, struct msix_entry *entries, int nvec)
915{
916 int status, nr_entries;
917 int i, j;
918
919 if (!entries)
920 return -EINVAL;
921
922 status = pci_msi_check_device(dev, nvec, PCI_CAP_ID_MSIX);
923 if (status)
924 return status;
925
926 nr_entries = pci_msix_table_size(dev);
927 if (nvec > nr_entries)
928 return nr_entries;
929
930 /* Check for any invalid entries */
931 for (i = 0; i < nvec; i++) {
932 if (entries[i].entry >= nr_entries)
933 return -EINVAL; /* invalid entry */
934 for (j = i + 1; j < nvec; j++) {
935 if (entries[i].entry == entries[j].entry)
936 return -EINVAL; /* duplicate entry */
937 }
938 }
939 WARN_ON(!!dev->msix_enabled);
940
941 /* Check whether driver already requested for MSI irq */
942 if (dev->msi_enabled) {
943 dev_info(&dev->dev, "can't enable MSI-X "
944 "(MSI IRQ already assigned)\n");
945 return -EINVAL;
946 }
947 status = msix_capability_init(dev, entries, nvec);
948 return status;
949}
950EXPORT_SYMBOL(pci_enable_msix);
951
952void pci_msix_shutdown(struct pci_dev *dev)
953{
954 struct msi_desc *entry;
955
956 if (!pci_msi_enable || !dev || !dev->msix_enabled)
957 return;
958
959 /* Return the device with MSI-X masked as initial states */
960 list_for_each_entry(entry, &dev->msi_list, list) {
961 /* Keep cached states to be restored */
962 __msix_mask_irq(entry, 1);
963 }
964
965 msix_set_enable(dev, 0);
966 pci_intx_for_msi(dev, 1);
967 dev->msix_enabled = 0;
968}
969
970void pci_disable_msix(struct pci_dev *dev)
971{
972 if (!pci_msi_enable || !dev || !dev->msix_enabled)
973 return;
974
975 pci_msix_shutdown(dev);
976 free_msi_irqs(dev);
977 kset_unregister(dev->msi_kset);
978 dev->msi_kset = NULL;
979}
980EXPORT_SYMBOL(pci_disable_msix);
981
982/**
983 * msi_remove_pci_irq_vectors - reclaim MSI(X) irqs to unused state
984 * @dev: pointer to the pci_dev data structure of MSI(X) device function
985 *
986 * Being called during hotplug remove, from which the device function
987 * is hot-removed. All previous assigned MSI/MSI-X irqs, if
988 * allocated for this device function, are reclaimed to unused state,
989 * which may be used later on.
990 **/
991void msi_remove_pci_irq_vectors(struct pci_dev *dev)
992{
993 if (!pci_msi_enable || !dev)
994 return;
995
996 if (dev->msi_enabled || dev->msix_enabled)
997 free_msi_irqs(dev);
998}
999
1000void pci_no_msi(void)
1001{
1002 pci_msi_enable = 0;
1003}
1004
1005/**
1006 * pci_msi_enabled - is MSI enabled?
1007 *
1008 * Returns true if MSI has not been disabled by the command-line option
1009 * pci=nomsi.
1010 **/
1011int pci_msi_enabled(void)
1012{
1013 return pci_msi_enable;
1014}
1015EXPORT_SYMBOL(pci_msi_enabled);
1016
1017void pci_msi_init_pci_dev(struct pci_dev *dev)
1018{
1019 int pos;
1020 INIT_LIST_HEAD(&dev->msi_list);
1021
1022 /* Disable the msi hardware to avoid screaming interrupts
1023 * during boot. This is the power on reset default so
1024 * usually this should be a noop.
1025 */
1026 pos = pci_find_capability(dev, PCI_CAP_ID_MSI);
1027 if (pos)
1028 msi_set_enable(dev, pos, 0);
1029 msix_set_enable(dev, 0);
1030}
1/*
2 * File: msi.c
3 * Purpose: PCI Message Signaled Interrupt (MSI)
4 *
5 * Copyright (C) 2003-2004 Intel
6 * Copyright (C) Tom Long Nguyen (tom.l.nguyen@intel.com)
7 * Copyright (C) 2016 Christoph Hellwig.
8 */
9
10#include <linux/err.h>
11#include <linux/mm.h>
12#include <linux/irq.h>
13#include <linux/interrupt.h>
14#include <linux/export.h>
15#include <linux/ioport.h>
16#include <linux/pci.h>
17#include <linux/proc_fs.h>
18#include <linux/msi.h>
19#include <linux/smp.h>
20#include <linux/errno.h>
21#include <linux/io.h>
22#include <linux/acpi_iort.h>
23#include <linux/slab.h>
24#include <linux/irqdomain.h>
25#include <linux/of_irq.h>
26
27#include "pci.h"
28
29static int pci_msi_enable = 1;
30int pci_msi_ignore_mask;
31
32#define msix_table_size(flags) ((flags & PCI_MSIX_FLAGS_QSIZE) + 1)
33
34#ifdef CONFIG_PCI_MSI_IRQ_DOMAIN
35static struct irq_domain *pci_msi_default_domain;
36static DEFINE_MUTEX(pci_msi_domain_lock);
37
38struct irq_domain * __weak arch_get_pci_msi_domain(struct pci_dev *dev)
39{
40 return pci_msi_default_domain;
41}
42
43static struct irq_domain *pci_msi_get_domain(struct pci_dev *dev)
44{
45 struct irq_domain *domain;
46
47 domain = dev_get_msi_domain(&dev->dev);
48 if (domain)
49 return domain;
50
51 return arch_get_pci_msi_domain(dev);
52}
53
54static int pci_msi_setup_msi_irqs(struct pci_dev *dev, int nvec, int type)
55{
56 struct irq_domain *domain;
57
58 domain = pci_msi_get_domain(dev);
59 if (domain && irq_domain_is_hierarchy(domain))
60 return pci_msi_domain_alloc_irqs(domain, dev, nvec, type);
61
62 return arch_setup_msi_irqs(dev, nvec, type);
63}
64
65static void pci_msi_teardown_msi_irqs(struct pci_dev *dev)
66{
67 struct irq_domain *domain;
68
69 domain = pci_msi_get_domain(dev);
70 if (domain && irq_domain_is_hierarchy(domain))
71 pci_msi_domain_free_irqs(domain, dev);
72 else
73 arch_teardown_msi_irqs(dev);
74}
75#else
76#define pci_msi_setup_msi_irqs arch_setup_msi_irqs
77#define pci_msi_teardown_msi_irqs arch_teardown_msi_irqs
78#endif
79
80/* Arch hooks */
81
82int __weak arch_setup_msi_irq(struct pci_dev *dev, struct msi_desc *desc)
83{
84 struct msi_controller *chip = dev->bus->msi;
85 int err;
86
87 if (!chip || !chip->setup_irq)
88 return -EINVAL;
89
90 err = chip->setup_irq(chip, dev, desc);
91 if (err < 0)
92 return err;
93
94 irq_set_chip_data(desc->irq, chip);
95
96 return 0;
97}
98
99void __weak arch_teardown_msi_irq(unsigned int irq)
100{
101 struct msi_controller *chip = irq_get_chip_data(irq);
102
103 if (!chip || !chip->teardown_irq)
104 return;
105
106 chip->teardown_irq(chip, irq);
107}
108
109int __weak arch_setup_msi_irqs(struct pci_dev *dev, int nvec, int type)
110{
111 struct msi_controller *chip = dev->bus->msi;
112 struct msi_desc *entry;
113 int ret;
114
115 if (chip && chip->setup_irqs)
116 return chip->setup_irqs(chip, dev, nvec, type);
117 /*
118 * If an architecture wants to support multiple MSI, it needs to
119 * override arch_setup_msi_irqs()
120 */
121 if (type == PCI_CAP_ID_MSI && nvec > 1)
122 return 1;
123
124 for_each_pci_msi_entry(entry, dev) {
125 ret = arch_setup_msi_irq(dev, entry);
126 if (ret < 0)
127 return ret;
128 if (ret > 0)
129 return -ENOSPC;
130 }
131
132 return 0;
133}
134
135/*
136 * We have a default implementation available as a separate non-weak
137 * function, as it is used by the Xen x86 PCI code
138 */
139void default_teardown_msi_irqs(struct pci_dev *dev)
140{
141 int i;
142 struct msi_desc *entry;
143
144 for_each_pci_msi_entry(entry, dev)
145 if (entry->irq)
146 for (i = 0; i < entry->nvec_used; i++)
147 arch_teardown_msi_irq(entry->irq + i);
148}
149
150void __weak arch_teardown_msi_irqs(struct pci_dev *dev)
151{
152 return default_teardown_msi_irqs(dev);
153}
154
155static void default_restore_msi_irq(struct pci_dev *dev, int irq)
156{
157 struct msi_desc *entry;
158
159 entry = NULL;
160 if (dev->msix_enabled) {
161 for_each_pci_msi_entry(entry, dev) {
162 if (irq == entry->irq)
163 break;
164 }
165 } else if (dev->msi_enabled) {
166 entry = irq_get_msi_desc(irq);
167 }
168
169 if (entry)
170 __pci_write_msi_msg(entry, &entry->msg);
171}
172
173void __weak arch_restore_msi_irqs(struct pci_dev *dev)
174{
175 return default_restore_msi_irqs(dev);
176}
177
178static inline __attribute_const__ u32 msi_mask(unsigned x)
179{
180 /* Don't shift by >= width of type */
181 if (x >= 5)
182 return 0xffffffff;
183 return (1 << (1 << x)) - 1;
184}
185
186/*
187 * PCI 2.3 does not specify mask bits for each MSI interrupt. Attempting to
188 * mask all MSI interrupts by clearing the MSI enable bit does not work
189 * reliably as devices without an INTx disable bit will then generate a
190 * level IRQ which will never be cleared.
191 */
192u32 __pci_msi_desc_mask_irq(struct msi_desc *desc, u32 mask, u32 flag)
193{
194 u32 mask_bits = desc->masked;
195
196 if (pci_msi_ignore_mask || !desc->msi_attrib.maskbit)
197 return 0;
198
199 mask_bits &= ~mask;
200 mask_bits |= flag;
201 pci_write_config_dword(msi_desc_to_pci_dev(desc), desc->mask_pos,
202 mask_bits);
203
204 return mask_bits;
205}
206
207static void msi_mask_irq(struct msi_desc *desc, u32 mask, u32 flag)
208{
209 desc->masked = __pci_msi_desc_mask_irq(desc, mask, flag);
210}
211
212static void __iomem *pci_msix_desc_addr(struct msi_desc *desc)
213{
214 return desc->mask_base +
215 desc->msi_attrib.entry_nr * PCI_MSIX_ENTRY_SIZE;
216}
217
218/*
219 * This internal function does not flush PCI writes to the device.
220 * All users must ensure that they read from the device before either
221 * assuming that the device state is up to date, or returning out of this
222 * file. This saves a few milliseconds when initialising devices with lots
223 * of MSI-X interrupts.
224 */
225u32 __pci_msix_desc_mask_irq(struct msi_desc *desc, u32 flag)
226{
227 u32 mask_bits = desc->masked;
228
229 if (pci_msi_ignore_mask)
230 return 0;
231
232 mask_bits &= ~PCI_MSIX_ENTRY_CTRL_MASKBIT;
233 if (flag)
234 mask_bits |= PCI_MSIX_ENTRY_CTRL_MASKBIT;
235 writel(mask_bits, pci_msix_desc_addr(desc) + PCI_MSIX_ENTRY_VECTOR_CTRL);
236
237 return mask_bits;
238}
239
240static void msix_mask_irq(struct msi_desc *desc, u32 flag)
241{
242 desc->masked = __pci_msix_desc_mask_irq(desc, flag);
243}
244
245static void msi_set_mask_bit(struct irq_data *data, u32 flag)
246{
247 struct msi_desc *desc = irq_data_get_msi_desc(data);
248
249 if (desc->msi_attrib.is_msix) {
250 msix_mask_irq(desc, flag);
251 readl(desc->mask_base); /* Flush write to device */
252 } else {
253 unsigned offset = data->irq - desc->irq;
254 msi_mask_irq(desc, 1 << offset, flag << offset);
255 }
256}
257
258/**
259 * pci_msi_mask_irq - Generic irq chip callback to mask PCI/MSI interrupts
260 * @data: pointer to irqdata associated to that interrupt
261 */
262void pci_msi_mask_irq(struct irq_data *data)
263{
264 msi_set_mask_bit(data, 1);
265}
266EXPORT_SYMBOL_GPL(pci_msi_mask_irq);
267
268/**
269 * pci_msi_unmask_irq - Generic irq chip callback to unmask PCI/MSI interrupts
270 * @data: pointer to irqdata associated to that interrupt
271 */
272void pci_msi_unmask_irq(struct irq_data *data)
273{
274 msi_set_mask_bit(data, 0);
275}
276EXPORT_SYMBOL_GPL(pci_msi_unmask_irq);
277
278void default_restore_msi_irqs(struct pci_dev *dev)
279{
280 struct msi_desc *entry;
281
282 for_each_pci_msi_entry(entry, dev)
283 default_restore_msi_irq(dev, entry->irq);
284}
285
286void __pci_read_msi_msg(struct msi_desc *entry, struct msi_msg *msg)
287{
288 struct pci_dev *dev = msi_desc_to_pci_dev(entry);
289
290 BUG_ON(dev->current_state != PCI_D0);
291
292 if (entry->msi_attrib.is_msix) {
293 void __iomem *base = pci_msix_desc_addr(entry);
294
295 msg->address_lo = readl(base + PCI_MSIX_ENTRY_LOWER_ADDR);
296 msg->address_hi = readl(base + PCI_MSIX_ENTRY_UPPER_ADDR);
297 msg->data = readl(base + PCI_MSIX_ENTRY_DATA);
298 } else {
299 int pos = dev->msi_cap;
300 u16 data;
301
302 pci_read_config_dword(dev, pos + PCI_MSI_ADDRESS_LO,
303 &msg->address_lo);
304 if (entry->msi_attrib.is_64) {
305 pci_read_config_dword(dev, pos + PCI_MSI_ADDRESS_HI,
306 &msg->address_hi);
307 pci_read_config_word(dev, pos + PCI_MSI_DATA_64, &data);
308 } else {
309 msg->address_hi = 0;
310 pci_read_config_word(dev, pos + PCI_MSI_DATA_32, &data);
311 }
312 msg->data = data;
313 }
314}
315
316void __pci_write_msi_msg(struct msi_desc *entry, struct msi_msg *msg)
317{
318 struct pci_dev *dev = msi_desc_to_pci_dev(entry);
319
320 if (dev->current_state != PCI_D0) {
321 /* Don't touch the hardware now */
322 } else if (entry->msi_attrib.is_msix) {
323 void __iomem *base = pci_msix_desc_addr(entry);
324
325 writel(msg->address_lo, base + PCI_MSIX_ENTRY_LOWER_ADDR);
326 writel(msg->address_hi, base + PCI_MSIX_ENTRY_UPPER_ADDR);
327 writel(msg->data, base + PCI_MSIX_ENTRY_DATA);
328 } else {
329 int pos = dev->msi_cap;
330 u16 msgctl;
331
332 pci_read_config_word(dev, pos + PCI_MSI_FLAGS, &msgctl);
333 msgctl &= ~PCI_MSI_FLAGS_QSIZE;
334 msgctl |= entry->msi_attrib.multiple << 4;
335 pci_write_config_word(dev, pos + PCI_MSI_FLAGS, msgctl);
336
337 pci_write_config_dword(dev, pos + PCI_MSI_ADDRESS_LO,
338 msg->address_lo);
339 if (entry->msi_attrib.is_64) {
340 pci_write_config_dword(dev, pos + PCI_MSI_ADDRESS_HI,
341 msg->address_hi);
342 pci_write_config_word(dev, pos + PCI_MSI_DATA_64,
343 msg->data);
344 } else {
345 pci_write_config_word(dev, pos + PCI_MSI_DATA_32,
346 msg->data);
347 }
348 }
349 entry->msg = *msg;
350}
351
352void pci_write_msi_msg(unsigned int irq, struct msi_msg *msg)
353{
354 struct msi_desc *entry = irq_get_msi_desc(irq);
355
356 __pci_write_msi_msg(entry, msg);
357}
358EXPORT_SYMBOL_GPL(pci_write_msi_msg);
359
360static void free_msi_irqs(struct pci_dev *dev)
361{
362 struct list_head *msi_list = dev_to_msi_list(&dev->dev);
363 struct msi_desc *entry, *tmp;
364 struct attribute **msi_attrs;
365 struct device_attribute *dev_attr;
366 int i, count = 0;
367
368 for_each_pci_msi_entry(entry, dev)
369 if (entry->irq)
370 for (i = 0; i < entry->nvec_used; i++)
371 BUG_ON(irq_has_action(entry->irq + i));
372
373 pci_msi_teardown_msi_irqs(dev);
374
375 list_for_each_entry_safe(entry, tmp, msi_list, list) {
376 if (entry->msi_attrib.is_msix) {
377 if (list_is_last(&entry->list, msi_list))
378 iounmap(entry->mask_base);
379 }
380
381 list_del(&entry->list);
382 kfree(entry);
383 }
384
385 if (dev->msi_irq_groups) {
386 sysfs_remove_groups(&dev->dev.kobj, dev->msi_irq_groups);
387 msi_attrs = dev->msi_irq_groups[0]->attrs;
388 while (msi_attrs[count]) {
389 dev_attr = container_of(msi_attrs[count],
390 struct device_attribute, attr);
391 kfree(dev_attr->attr.name);
392 kfree(dev_attr);
393 ++count;
394 }
395 kfree(msi_attrs);
396 kfree(dev->msi_irq_groups[0]);
397 kfree(dev->msi_irq_groups);
398 dev->msi_irq_groups = NULL;
399 }
400}
401
402static void pci_intx_for_msi(struct pci_dev *dev, int enable)
403{
404 if (!(dev->dev_flags & PCI_DEV_FLAGS_MSI_INTX_DISABLE_BUG))
405 pci_intx(dev, enable);
406}
407
408static void __pci_restore_msi_state(struct pci_dev *dev)
409{
410 u16 control;
411 struct msi_desc *entry;
412
413 if (!dev->msi_enabled)
414 return;
415
416 entry = irq_get_msi_desc(dev->irq);
417
418 pci_intx_for_msi(dev, 0);
419 pci_msi_set_enable(dev, 0);
420 arch_restore_msi_irqs(dev);
421
422 pci_read_config_word(dev, dev->msi_cap + PCI_MSI_FLAGS, &control);
423 msi_mask_irq(entry, msi_mask(entry->msi_attrib.multi_cap),
424 entry->masked);
425 control &= ~PCI_MSI_FLAGS_QSIZE;
426 control |= (entry->msi_attrib.multiple << 4) | PCI_MSI_FLAGS_ENABLE;
427 pci_write_config_word(dev, dev->msi_cap + PCI_MSI_FLAGS, control);
428}
429
430static void __pci_restore_msix_state(struct pci_dev *dev)
431{
432 struct msi_desc *entry;
433
434 if (!dev->msix_enabled)
435 return;
436 BUG_ON(list_empty(dev_to_msi_list(&dev->dev)));
437
438 /* route the table */
439 pci_intx_for_msi(dev, 0);
440 pci_msix_clear_and_set_ctrl(dev, 0,
441 PCI_MSIX_FLAGS_ENABLE | PCI_MSIX_FLAGS_MASKALL);
442
443 arch_restore_msi_irqs(dev);
444 for_each_pci_msi_entry(entry, dev)
445 msix_mask_irq(entry, entry->masked);
446
447 pci_msix_clear_and_set_ctrl(dev, PCI_MSIX_FLAGS_MASKALL, 0);
448}
449
450void pci_restore_msi_state(struct pci_dev *dev)
451{
452 __pci_restore_msi_state(dev);
453 __pci_restore_msix_state(dev);
454}
455EXPORT_SYMBOL_GPL(pci_restore_msi_state);
456
457static ssize_t msi_mode_show(struct device *dev, struct device_attribute *attr,
458 char *buf)
459{
460 struct msi_desc *entry;
461 unsigned long irq;
462 int retval;
463
464 retval = kstrtoul(attr->attr.name, 10, &irq);
465 if (retval)
466 return retval;
467
468 entry = irq_get_msi_desc(irq);
469 if (entry)
470 return sprintf(buf, "%s\n",
471 entry->msi_attrib.is_msix ? "msix" : "msi");
472
473 return -ENODEV;
474}
475
476static int populate_msi_sysfs(struct pci_dev *pdev)
477{
478 struct attribute **msi_attrs;
479 struct attribute *msi_attr;
480 struct device_attribute *msi_dev_attr;
481 struct attribute_group *msi_irq_group;
482 const struct attribute_group **msi_irq_groups;
483 struct msi_desc *entry;
484 int ret = -ENOMEM;
485 int num_msi = 0;
486 int count = 0;
487 int i;
488
489 /* Determine how many msi entries we have */
490 for_each_pci_msi_entry(entry, pdev)
491 num_msi += entry->nvec_used;
492 if (!num_msi)
493 return 0;
494
495 /* Dynamically create the MSI attributes for the PCI device */
496 msi_attrs = kzalloc(sizeof(void *) * (num_msi + 1), GFP_KERNEL);
497 if (!msi_attrs)
498 return -ENOMEM;
499 for_each_pci_msi_entry(entry, pdev) {
500 for (i = 0; i < entry->nvec_used; i++) {
501 msi_dev_attr = kzalloc(sizeof(*msi_dev_attr), GFP_KERNEL);
502 if (!msi_dev_attr)
503 goto error_attrs;
504 msi_attrs[count] = &msi_dev_attr->attr;
505
506 sysfs_attr_init(&msi_dev_attr->attr);
507 msi_dev_attr->attr.name = kasprintf(GFP_KERNEL, "%d",
508 entry->irq + i);
509 if (!msi_dev_attr->attr.name)
510 goto error_attrs;
511 msi_dev_attr->attr.mode = S_IRUGO;
512 msi_dev_attr->show = msi_mode_show;
513 ++count;
514 }
515 }
516
517 msi_irq_group = kzalloc(sizeof(*msi_irq_group), GFP_KERNEL);
518 if (!msi_irq_group)
519 goto error_attrs;
520 msi_irq_group->name = "msi_irqs";
521 msi_irq_group->attrs = msi_attrs;
522
523 msi_irq_groups = kzalloc(sizeof(void *) * 2, GFP_KERNEL);
524 if (!msi_irq_groups)
525 goto error_irq_group;
526 msi_irq_groups[0] = msi_irq_group;
527
528 ret = sysfs_create_groups(&pdev->dev.kobj, msi_irq_groups);
529 if (ret)
530 goto error_irq_groups;
531 pdev->msi_irq_groups = msi_irq_groups;
532
533 return 0;
534
535error_irq_groups:
536 kfree(msi_irq_groups);
537error_irq_group:
538 kfree(msi_irq_group);
539error_attrs:
540 count = 0;
541 msi_attr = msi_attrs[count];
542 while (msi_attr) {
543 msi_dev_attr = container_of(msi_attr, struct device_attribute, attr);
544 kfree(msi_attr->name);
545 kfree(msi_dev_attr);
546 ++count;
547 msi_attr = msi_attrs[count];
548 }
549 kfree(msi_attrs);
550 return ret;
551}
552
553static struct msi_desc *
554msi_setup_entry(struct pci_dev *dev, int nvec, const struct irq_affinity *affd)
555{
556 struct cpumask *masks = NULL;
557 struct msi_desc *entry;
558 u16 control;
559
560 if (affd) {
561 masks = irq_create_affinity_masks(nvec, affd);
562 if (!masks)
563 pr_err("Unable to allocate affinity masks, ignoring\n");
564 }
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.entry_nr = 0;
576 entry->msi_attrib.maskbit = !!(control & PCI_MSI_FLAGS_MASKBIT);
577 entry->msi_attrib.default_irq = dev->irq; /* Save IOAPIC IRQ */
578 entry->msi_attrib.multi_cap = (control & PCI_MSI_FLAGS_QMASK) >> 1;
579 entry->msi_attrib.multiple = ilog2(__roundup_pow_of_two(nvec));
580
581 if (control & PCI_MSI_FLAGS_64BIT)
582 entry->mask_pos = dev->msi_cap + PCI_MSI_MASK_64;
583 else
584 entry->mask_pos = dev->msi_cap + PCI_MSI_MASK_32;
585
586 /* Save the initial mask status */
587 if (entry->msi_attrib.maskbit)
588 pci_read_config_dword(dev, entry->mask_pos, &entry->masked);
589
590out:
591 kfree(masks);
592 return entry;
593}
594
595static int msi_verify_entries(struct pci_dev *dev)
596{
597 struct msi_desc *entry;
598
599 for_each_pci_msi_entry(entry, dev) {
600 if (!dev->no_64bit_msi || !entry->msg.address_hi)
601 continue;
602 dev_err(&dev->dev, "Device has broken 64-bit MSI but arch"
603 " tried to assign one above 4G\n");
604 return -EIO;
605 }
606 return 0;
607}
608
609/**
610 * msi_capability_init - configure device's MSI capability structure
611 * @dev: pointer to the pci_dev data structure of MSI device function
612 * @nvec: number of interrupts to allocate
613 * @affinity: flag to indicate cpu irq affinity mask should be set
614 *
615 * Setup the MSI capability structure of the device with the requested
616 * number of interrupts. A return value of zero indicates the successful
617 * setup of an entry with the new MSI irq. A negative return value indicates
618 * an error, and a positive return value indicates the number of interrupts
619 * which could have been allocated.
620 */
621static int msi_capability_init(struct pci_dev *dev, int nvec,
622 const struct irq_affinity *affd)
623{
624 struct msi_desc *entry;
625 int ret;
626 unsigned mask;
627
628 pci_msi_set_enable(dev, 0); /* Disable MSI during set up */
629
630 entry = msi_setup_entry(dev, nvec, affd);
631 if (!entry)
632 return -ENOMEM;
633
634 /* All MSIs are unmasked by default, Mask them all */
635 mask = msi_mask(entry->msi_attrib.multi_cap);
636 msi_mask_irq(entry, mask, mask);
637
638 list_add_tail(&entry->list, dev_to_msi_list(&dev->dev));
639
640 /* Configure MSI capability structure */
641 ret = pci_msi_setup_msi_irqs(dev, nvec, PCI_CAP_ID_MSI);
642 if (ret) {
643 msi_mask_irq(entry, mask, ~mask);
644 free_msi_irqs(dev);
645 return ret;
646 }
647
648 ret = msi_verify_entries(dev);
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 pci_msi_set_enable(dev, 1);
665 dev->msi_enabled = 1;
666
667 pcibios_free_irq(dev);
668 dev->irq = entry->irq;
669 return 0;
670}
671
672static void __iomem *msix_map_region(struct pci_dev *dev, unsigned nr_entries)
673{
674 resource_size_t phys_addr;
675 u32 table_offset;
676 unsigned long flags;
677 u8 bir;
678
679 pci_read_config_dword(dev, dev->msix_cap + PCI_MSIX_TABLE,
680 &table_offset);
681 bir = (u8)(table_offset & PCI_MSIX_TABLE_BIR);
682 flags = pci_resource_flags(dev, bir);
683 if (!flags || (flags & IORESOURCE_UNSET))
684 return NULL;
685
686 table_offset &= PCI_MSIX_TABLE_OFFSET;
687 phys_addr = pci_resource_start(dev, bir) + table_offset;
688
689 return ioremap_nocache(phys_addr, nr_entries * PCI_MSIX_ENTRY_SIZE);
690}
691
692static int msix_setup_entries(struct pci_dev *dev, void __iomem *base,
693 struct msix_entry *entries, int nvec,
694 const struct irq_affinity *affd)
695{
696 struct cpumask *curmsk, *masks = NULL;
697 struct msi_desc *entry;
698 int ret, i;
699
700 if (affd) {
701 masks = irq_create_affinity_masks(nvec, affd);
702 if (!masks)
703 pr_err("Unable to allocate affinity masks, ignoring\n");
704 }
705
706 for (i = 0, curmsk = masks; i < nvec; i++) {
707 entry = alloc_msi_entry(&dev->dev, 1, curmsk);
708 if (!entry) {
709 if (!i)
710 iounmap(base);
711 else
712 free_msi_irqs(dev);
713 /* No enough memory. Don't try again */
714 ret = -ENOMEM;
715 goto out;
716 }
717
718 entry->msi_attrib.is_msix = 1;
719 entry->msi_attrib.is_64 = 1;
720 if (entries)
721 entry->msi_attrib.entry_nr = entries[i].entry;
722 else
723 entry->msi_attrib.entry_nr = i;
724 entry->msi_attrib.default_irq = dev->irq;
725 entry->mask_base = base;
726
727 list_add_tail(&entry->list, dev_to_msi_list(&dev->dev));
728 if (masks)
729 curmsk++;
730 }
731 ret = 0;
732out:
733 kfree(masks);
734 return 0;
735}
736
737static void msix_program_entries(struct pci_dev *dev,
738 struct msix_entry *entries)
739{
740 struct msi_desc *entry;
741 int i = 0;
742
743 for_each_pci_msi_entry(entry, dev) {
744 if (entries)
745 entries[i++].vector = entry->irq;
746 entry->masked = readl(pci_msix_desc_addr(entry) +
747 PCI_MSIX_ENTRY_VECTOR_CTRL);
748 msix_mask_irq(entry, 1);
749 }
750}
751
752/**
753 * msix_capability_init - configure device's MSI-X capability
754 * @dev: pointer to the pci_dev data structure of MSI-X device function
755 * @entries: pointer to an array of struct msix_entry entries
756 * @nvec: number of @entries
757 * @affd: Optional pointer to enable automatic affinity assignement
758 *
759 * Setup the MSI-X capability structure of device function with a
760 * single MSI-X irq. A return of zero indicates the successful setup of
761 * requested MSI-X entries with allocated irqs or non-zero for otherwise.
762 **/
763static int msix_capability_init(struct pci_dev *dev, struct msix_entry *entries,
764 int nvec, const struct irq_affinity *affd)
765{
766 int ret;
767 u16 control;
768 void __iomem *base;
769
770 /* Ensure MSI-X is disabled while it is set up */
771 pci_msix_clear_and_set_ctrl(dev, PCI_MSIX_FLAGS_ENABLE, 0);
772
773 pci_read_config_word(dev, dev->msix_cap + PCI_MSIX_FLAGS, &control);
774 /* Request & Map MSI-X table region */
775 base = msix_map_region(dev, msix_table_size(control));
776 if (!base)
777 return -ENOMEM;
778
779 ret = msix_setup_entries(dev, base, entries, nvec, affd);
780 if (ret)
781 return ret;
782
783 ret = pci_msi_setup_msi_irqs(dev, nvec, PCI_CAP_ID_MSIX);
784 if (ret)
785 goto out_avail;
786
787 /* Check if all MSI entries honor device restrictions */
788 ret = msi_verify_entries(dev);
789 if (ret)
790 goto out_free;
791
792 /*
793 * Some devices require MSI-X to be enabled before we can touch the
794 * MSI-X registers. We need to mask all the vectors to prevent
795 * interrupts coming in before they're fully set up.
796 */
797 pci_msix_clear_and_set_ctrl(dev, 0,
798 PCI_MSIX_FLAGS_MASKALL | PCI_MSIX_FLAGS_ENABLE);
799
800 msix_program_entries(dev, entries);
801
802 ret = populate_msi_sysfs(dev);
803 if (ret)
804 goto out_free;
805
806 /* Set MSI-X enabled bits and unmask the function */
807 pci_intx_for_msi(dev, 0);
808 dev->msix_enabled = 1;
809 pci_msix_clear_and_set_ctrl(dev, PCI_MSIX_FLAGS_MASKALL, 0);
810
811 pcibios_free_irq(dev);
812 return 0;
813
814out_avail:
815 if (ret < 0) {
816 /*
817 * If we had some success, report the number of irqs
818 * we succeeded in setting up.
819 */
820 struct msi_desc *entry;
821 int avail = 0;
822
823 for_each_pci_msi_entry(entry, dev) {
824 if (entry->irq != 0)
825 avail++;
826 }
827 if (avail != 0)
828 ret = avail;
829 }
830
831out_free:
832 free_msi_irqs(dev);
833
834 return ret;
835}
836
837/**
838 * pci_msi_supported - check whether MSI may be enabled on a device
839 * @dev: pointer to the pci_dev data structure of MSI device function
840 * @nvec: how many MSIs have been requested ?
841 *
842 * Look at global flags, the device itself, and its parent buses
843 * to determine if MSI/-X are supported for the device. If MSI/-X is
844 * supported return 1, else return 0.
845 **/
846static int pci_msi_supported(struct pci_dev *dev, int nvec)
847{
848 struct pci_bus *bus;
849
850 /* MSI must be globally enabled and supported by the device */
851 if (!pci_msi_enable)
852 return 0;
853
854 if (!dev || dev->no_msi || dev->current_state != PCI_D0)
855 return 0;
856
857 /*
858 * You can't ask to have 0 or less MSIs configured.
859 * a) it's stupid ..
860 * b) the list manipulation code assumes nvec >= 1.
861 */
862 if (nvec < 1)
863 return 0;
864
865 /*
866 * Any bridge which does NOT route MSI transactions from its
867 * secondary bus to its primary bus must set NO_MSI flag on
868 * the secondary pci_bus.
869 * We expect only arch-specific PCI host bus controller driver
870 * or quirks for specific PCI bridges to be setting NO_MSI.
871 */
872 for (bus = dev->bus; bus; bus = bus->parent)
873 if (bus->bus_flags & PCI_BUS_FLAGS_NO_MSI)
874 return 0;
875
876 return 1;
877}
878
879/**
880 * pci_msi_vec_count - Return the number of MSI vectors a device can send
881 * @dev: device to report about
882 *
883 * This function returns the number of MSI vectors a device requested via
884 * Multiple Message Capable register. It returns a negative errno if the
885 * device is not capable sending MSI interrupts. Otherwise, the call succeeds
886 * and returns a power of two, up to a maximum of 2^5 (32), according to the
887 * MSI specification.
888 **/
889int pci_msi_vec_count(struct pci_dev *dev)
890{
891 int ret;
892 u16 msgctl;
893
894 if (!dev->msi_cap)
895 return -EINVAL;
896
897 pci_read_config_word(dev, dev->msi_cap + PCI_MSI_FLAGS, &msgctl);
898 ret = 1 << ((msgctl & PCI_MSI_FLAGS_QMASK) >> 1);
899
900 return ret;
901}
902EXPORT_SYMBOL(pci_msi_vec_count);
903
904void pci_msi_shutdown(struct pci_dev *dev)
905{
906 struct msi_desc *desc;
907 u32 mask;
908
909 if (!pci_msi_enable || !dev || !dev->msi_enabled)
910 return;
911
912 BUG_ON(list_empty(dev_to_msi_list(&dev->dev)));
913 desc = first_pci_msi_entry(dev);
914
915 pci_msi_set_enable(dev, 0);
916 pci_intx_for_msi(dev, 1);
917 dev->msi_enabled = 0;
918
919 /* Return the device with MSI unmasked as initial states */
920 mask = msi_mask(desc->msi_attrib.multi_cap);
921 /* Keep cached state to be restored */
922 __pci_msi_desc_mask_irq(desc, mask, ~mask);
923
924 /* Restore dev->irq to its default pin-assertion irq */
925 dev->irq = desc->msi_attrib.default_irq;
926 pcibios_alloc_irq(dev);
927}
928
929void pci_disable_msi(struct pci_dev *dev)
930{
931 if (!pci_msi_enable || !dev || !dev->msi_enabled)
932 return;
933
934 pci_msi_shutdown(dev);
935 free_msi_irqs(dev);
936}
937EXPORT_SYMBOL(pci_disable_msi);
938
939/**
940 * pci_msix_vec_count - return the number of device's MSI-X table entries
941 * @dev: pointer to the pci_dev data structure of MSI-X device function
942 * This function returns the number of device's MSI-X table entries and
943 * therefore the number of MSI-X vectors device is capable of sending.
944 * It returns a negative errno if the device is not capable of sending MSI-X
945 * interrupts.
946 **/
947int pci_msix_vec_count(struct pci_dev *dev)
948{
949 u16 control;
950
951 if (!dev->msix_cap)
952 return -EINVAL;
953
954 pci_read_config_word(dev, dev->msix_cap + PCI_MSIX_FLAGS, &control);
955 return msix_table_size(control);
956}
957EXPORT_SYMBOL(pci_msix_vec_count);
958
959static int __pci_enable_msix(struct pci_dev *dev, struct msix_entry *entries,
960 int nvec, const struct irq_affinity *affd)
961{
962 int nr_entries;
963 int i, j;
964
965 if (!pci_msi_supported(dev, nvec))
966 return -EINVAL;
967
968 nr_entries = pci_msix_vec_count(dev);
969 if (nr_entries < 0)
970 return nr_entries;
971 if (nvec > nr_entries)
972 return nr_entries;
973
974 if (entries) {
975 /* Check for any invalid entries */
976 for (i = 0; i < nvec; i++) {
977 if (entries[i].entry >= nr_entries)
978 return -EINVAL; /* invalid entry */
979 for (j = i + 1; j < nvec; j++) {
980 if (entries[i].entry == entries[j].entry)
981 return -EINVAL; /* duplicate entry */
982 }
983 }
984 }
985 WARN_ON(!!dev->msix_enabled);
986
987 /* Check whether driver already requested for MSI irq */
988 if (dev->msi_enabled) {
989 dev_info(&dev->dev, "can't enable MSI-X (MSI IRQ already assigned)\n");
990 return -EINVAL;
991 }
992 return msix_capability_init(dev, entries, nvec, affd);
993}
994
995/**
996 * pci_enable_msix - configure device's MSI-X capability structure
997 * @dev: pointer to the pci_dev data structure of MSI-X device function
998 * @entries: pointer to an array of MSI-X entries (optional)
999 * @nvec: number of MSI-X irqs requested for allocation by device driver
1000 *
1001 * Setup the MSI-X capability structure of device function with the number
1002 * of requested irqs upon its software driver call to request for
1003 * MSI-X mode enabled on its hardware device function. A return of zero
1004 * indicates the successful configuration of MSI-X capability structure
1005 * with new allocated MSI-X irqs. A return of < 0 indicates a failure.
1006 * Or a return of > 0 indicates that driver request is exceeding the number
1007 * of irqs or MSI-X vectors available. Driver should use the returned value to
1008 * re-send its request.
1009 **/
1010int pci_enable_msix(struct pci_dev *dev, struct msix_entry *entries, int nvec)
1011{
1012 return __pci_enable_msix(dev, entries, nvec, NULL);
1013}
1014EXPORT_SYMBOL(pci_enable_msix);
1015
1016void pci_msix_shutdown(struct pci_dev *dev)
1017{
1018 struct msi_desc *entry;
1019
1020 if (!pci_msi_enable || !dev || !dev->msix_enabled)
1021 return;
1022
1023 /* Return the device with MSI-X masked as initial states */
1024 for_each_pci_msi_entry(entry, dev) {
1025 /* Keep cached states to be restored */
1026 __pci_msix_desc_mask_irq(entry, 1);
1027 }
1028
1029 pci_msix_clear_and_set_ctrl(dev, PCI_MSIX_FLAGS_ENABLE, 0);
1030 pci_intx_for_msi(dev, 1);
1031 dev->msix_enabled = 0;
1032 pcibios_alloc_irq(dev);
1033}
1034
1035void pci_disable_msix(struct pci_dev *dev)
1036{
1037 if (!pci_msi_enable || !dev || !dev->msix_enabled)
1038 return;
1039
1040 pci_msix_shutdown(dev);
1041 free_msi_irqs(dev);
1042}
1043EXPORT_SYMBOL(pci_disable_msix);
1044
1045void pci_no_msi(void)
1046{
1047 pci_msi_enable = 0;
1048}
1049
1050/**
1051 * pci_msi_enabled - is MSI enabled?
1052 *
1053 * Returns true if MSI has not been disabled by the command-line option
1054 * pci=nomsi.
1055 **/
1056int pci_msi_enabled(void)
1057{
1058 return pci_msi_enable;
1059}
1060EXPORT_SYMBOL(pci_msi_enabled);
1061
1062static int __pci_enable_msi_range(struct pci_dev *dev, int minvec, int maxvec,
1063 const struct irq_affinity *affd)
1064{
1065 int nvec;
1066 int rc;
1067
1068 if (!pci_msi_supported(dev, minvec))
1069 return -EINVAL;
1070
1071 WARN_ON(!!dev->msi_enabled);
1072
1073 /* Check whether driver already requested MSI-X irqs */
1074 if (dev->msix_enabled) {
1075 dev_info(&dev->dev,
1076 "can't enable MSI (MSI-X already enabled)\n");
1077 return -EINVAL;
1078 }
1079
1080 if (maxvec < minvec)
1081 return -ERANGE;
1082
1083 nvec = pci_msi_vec_count(dev);
1084 if (nvec < 0)
1085 return nvec;
1086 if (nvec < minvec)
1087 return -EINVAL;
1088
1089 if (nvec > maxvec)
1090 nvec = maxvec;
1091
1092 for (;;) {
1093 if (affd) {
1094 nvec = irq_calc_affinity_vectors(nvec, affd);
1095 if (nvec < minvec)
1096 return -ENOSPC;
1097 }
1098
1099 rc = msi_capability_init(dev, nvec, affd);
1100 if (rc == 0)
1101 return nvec;
1102
1103 if (rc < 0)
1104 return rc;
1105 if (rc < minvec)
1106 return -ENOSPC;
1107
1108 nvec = rc;
1109 }
1110}
1111
1112/**
1113 * pci_enable_msi_range - configure device's MSI capability structure
1114 * @dev: device to configure
1115 * @minvec: minimal number of interrupts to configure
1116 * @maxvec: maximum number of interrupts to configure
1117 *
1118 * This function tries to allocate a maximum possible number of interrupts in a
1119 * range between @minvec and @maxvec. It returns a negative errno if an error
1120 * occurs. If it succeeds, it returns the actual number of interrupts allocated
1121 * and updates the @dev's irq member to the lowest new interrupt number;
1122 * the other interrupt numbers allocated to this device are consecutive.
1123 **/
1124int pci_enable_msi_range(struct pci_dev *dev, int minvec, int maxvec)
1125{
1126 return __pci_enable_msi_range(dev, minvec, maxvec, NULL);
1127}
1128EXPORT_SYMBOL(pci_enable_msi_range);
1129
1130static int __pci_enable_msix_range(struct pci_dev *dev,
1131 struct msix_entry *entries, int minvec,
1132 int maxvec, const struct irq_affinity *affd)
1133{
1134 int rc, nvec = maxvec;
1135
1136 if (maxvec < minvec)
1137 return -ERANGE;
1138
1139 for (;;) {
1140 if (affd) {
1141 nvec = irq_calc_affinity_vectors(nvec, affd);
1142 if (nvec < minvec)
1143 return -ENOSPC;
1144 }
1145
1146 rc = __pci_enable_msix(dev, entries, nvec, affd);
1147 if (rc == 0)
1148 return nvec;
1149
1150 if (rc < 0)
1151 return rc;
1152 if (rc < minvec)
1153 return -ENOSPC;
1154
1155 nvec = rc;
1156 }
1157}
1158
1159/**
1160 * pci_enable_msix_range - configure device's MSI-X capability structure
1161 * @dev: pointer to the pci_dev data structure of MSI-X device function
1162 * @entries: pointer to an array of MSI-X entries
1163 * @minvec: minimum number of MSI-X irqs requested
1164 * @maxvec: maximum number of MSI-X irqs requested
1165 *
1166 * Setup the MSI-X capability structure of device function with a maximum
1167 * possible number of interrupts in the range between @minvec and @maxvec
1168 * upon its software driver call to request for MSI-X mode enabled on its
1169 * hardware device function. It returns a negative errno if an error occurs.
1170 * If it succeeds, it returns the actual number of interrupts allocated and
1171 * indicates the successful configuration of MSI-X capability structure
1172 * with new allocated MSI-X interrupts.
1173 **/
1174int pci_enable_msix_range(struct pci_dev *dev, struct msix_entry *entries,
1175 int minvec, int maxvec)
1176{
1177 return __pci_enable_msix_range(dev, entries, minvec, maxvec, NULL);
1178}
1179EXPORT_SYMBOL(pci_enable_msix_range);
1180
1181/**
1182 * pci_alloc_irq_vectors_affinity - allocate multiple IRQs for a device
1183 * @dev: PCI device to operate on
1184 * @min_vecs: minimum number of vectors required (must be >= 1)
1185 * @max_vecs: maximum (desired) number of vectors
1186 * @flags: flags or quirks for the allocation
1187 * @affd: optional description of the affinity requirements
1188 *
1189 * Allocate up to @max_vecs interrupt vectors for @dev, using MSI-X or MSI
1190 * vectors if available, and fall back to a single legacy vector
1191 * if neither is available. Return the number of vectors allocated,
1192 * (which might be smaller than @max_vecs) if successful, or a negative
1193 * error code on error. If less than @min_vecs interrupt vectors are
1194 * available for @dev the function will fail with -ENOSPC.
1195 *
1196 * To get the Linux IRQ number used for a vector that can be passed to
1197 * request_irq() use the pci_irq_vector() helper.
1198 */
1199int pci_alloc_irq_vectors_affinity(struct pci_dev *dev, unsigned int min_vecs,
1200 unsigned int max_vecs, unsigned int flags,
1201 const struct irq_affinity *affd)
1202{
1203 static const struct irq_affinity msi_default_affd;
1204 int vecs = -ENOSPC;
1205
1206 if (flags & PCI_IRQ_AFFINITY) {
1207 if (!affd)
1208 affd = &msi_default_affd;
1209
1210 if (affd->pre_vectors + affd->post_vectors > min_vecs)
1211 return -EINVAL;
1212
1213 /*
1214 * If there aren't any vectors left after applying the pre/post
1215 * vectors don't bother with assigning affinity.
1216 */
1217 if (affd->pre_vectors + affd->post_vectors == min_vecs)
1218 affd = NULL;
1219 } else {
1220 if (WARN_ON(affd))
1221 affd = NULL;
1222 }
1223
1224 if (flags & PCI_IRQ_MSIX) {
1225 vecs = __pci_enable_msix_range(dev, NULL, min_vecs, max_vecs,
1226 affd);
1227 if (vecs > 0)
1228 return vecs;
1229 }
1230
1231 if (flags & PCI_IRQ_MSI) {
1232 vecs = __pci_enable_msi_range(dev, min_vecs, max_vecs, affd);
1233 if (vecs > 0)
1234 return vecs;
1235 }
1236
1237 /* use legacy irq if allowed */
1238 if ((flags & PCI_IRQ_LEGACY) && min_vecs == 1) {
1239 pci_intx(dev, 1);
1240 return 1;
1241 }
1242
1243 return vecs;
1244}
1245EXPORT_SYMBOL(pci_alloc_irq_vectors_affinity);
1246
1247/**
1248 * pci_free_irq_vectors - free previously allocated IRQs for a device
1249 * @dev: PCI device to operate on
1250 *
1251 * Undoes the allocations and enabling in pci_alloc_irq_vectors().
1252 */
1253void pci_free_irq_vectors(struct pci_dev *dev)
1254{
1255 pci_disable_msix(dev);
1256 pci_disable_msi(dev);
1257}
1258EXPORT_SYMBOL(pci_free_irq_vectors);
1259
1260/**
1261 * pci_irq_vector - return Linux IRQ number of a device vector
1262 * @dev: PCI device to operate on
1263 * @nr: device-relative interrupt vector index (0-based).
1264 */
1265int pci_irq_vector(struct pci_dev *dev, unsigned int nr)
1266{
1267 if (dev->msix_enabled) {
1268 struct msi_desc *entry;
1269 int i = 0;
1270
1271 for_each_pci_msi_entry(entry, dev) {
1272 if (i == nr)
1273 return entry->irq;
1274 i++;
1275 }
1276 WARN_ON_ONCE(1);
1277 return -EINVAL;
1278 }
1279
1280 if (dev->msi_enabled) {
1281 struct msi_desc *entry = first_pci_msi_entry(dev);
1282
1283 if (WARN_ON_ONCE(nr >= entry->nvec_used))
1284 return -EINVAL;
1285 } else {
1286 if (WARN_ON_ONCE(nr > 0))
1287 return -EINVAL;
1288 }
1289
1290 return dev->irq + nr;
1291}
1292EXPORT_SYMBOL(pci_irq_vector);
1293
1294/**
1295 * pci_irq_get_affinity - return the affinity of a particular msi vector
1296 * @dev: PCI device to operate on
1297 * @nr: device-relative interrupt vector index (0-based).
1298 */
1299const struct cpumask *pci_irq_get_affinity(struct pci_dev *dev, int nr)
1300{
1301 if (dev->msix_enabled) {
1302 struct msi_desc *entry;
1303 int i = 0;
1304
1305 for_each_pci_msi_entry(entry, dev) {
1306 if (i == nr)
1307 return entry->affinity;
1308 i++;
1309 }
1310 WARN_ON_ONCE(1);
1311 return NULL;
1312 } else if (dev->msi_enabled) {
1313 struct msi_desc *entry = first_pci_msi_entry(dev);
1314
1315 if (WARN_ON_ONCE(!entry || !entry->affinity ||
1316 nr >= entry->nvec_used))
1317 return NULL;
1318
1319 return &entry->affinity[nr];
1320 } else {
1321 return cpu_possible_mask;
1322 }
1323}
1324EXPORT_SYMBOL(pci_irq_get_affinity);
1325
1326struct pci_dev *msi_desc_to_pci_dev(struct msi_desc *desc)
1327{
1328 return to_pci_dev(desc->dev);
1329}
1330EXPORT_SYMBOL(msi_desc_to_pci_dev);
1331
1332void *msi_desc_to_pci_sysdata(struct msi_desc *desc)
1333{
1334 struct pci_dev *dev = msi_desc_to_pci_dev(desc);
1335
1336 return dev->bus->sysdata;
1337}
1338EXPORT_SYMBOL_GPL(msi_desc_to_pci_sysdata);
1339
1340#ifdef CONFIG_PCI_MSI_IRQ_DOMAIN
1341/**
1342 * pci_msi_domain_write_msg - Helper to write MSI message to PCI config space
1343 * @irq_data: Pointer to interrupt data of the MSI interrupt
1344 * @msg: Pointer to the message
1345 */
1346void pci_msi_domain_write_msg(struct irq_data *irq_data, struct msi_msg *msg)
1347{
1348 struct msi_desc *desc = irq_data_get_msi_desc(irq_data);
1349
1350 /*
1351 * For MSI-X desc->irq is always equal to irq_data->irq. For
1352 * MSI only the first interrupt of MULTI MSI passes the test.
1353 */
1354 if (desc->irq == irq_data->irq)
1355 __pci_write_msi_msg(desc, msg);
1356}
1357
1358/**
1359 * pci_msi_domain_calc_hwirq - Generate a unique ID for an MSI source
1360 * @dev: Pointer to the PCI device
1361 * @desc: Pointer to the msi descriptor
1362 *
1363 * The ID number is only used within the irqdomain.
1364 */
1365irq_hw_number_t pci_msi_domain_calc_hwirq(struct pci_dev *dev,
1366 struct msi_desc *desc)
1367{
1368 return (irq_hw_number_t)desc->msi_attrib.entry_nr |
1369 PCI_DEVID(dev->bus->number, dev->devfn) << 11 |
1370 (pci_domain_nr(dev->bus) & 0xFFFFFFFF) << 27;
1371}
1372
1373static inline bool pci_msi_desc_is_multi_msi(struct msi_desc *desc)
1374{
1375 return !desc->msi_attrib.is_msix && desc->nvec_used > 1;
1376}
1377
1378/**
1379 * pci_msi_domain_check_cap - Verify that @domain supports the capabilities for @dev
1380 * @domain: The interrupt domain to check
1381 * @info: The domain info for verification
1382 * @dev: The device to check
1383 *
1384 * Returns:
1385 * 0 if the functionality is supported
1386 * 1 if Multi MSI is requested, but the domain does not support it
1387 * -ENOTSUPP otherwise
1388 */
1389int pci_msi_domain_check_cap(struct irq_domain *domain,
1390 struct msi_domain_info *info, struct device *dev)
1391{
1392 struct msi_desc *desc = first_pci_msi_entry(to_pci_dev(dev));
1393
1394 /* Special handling to support pci_enable_msi_range() */
1395 if (pci_msi_desc_is_multi_msi(desc) &&
1396 !(info->flags & MSI_FLAG_MULTI_PCI_MSI))
1397 return 1;
1398 else if (desc->msi_attrib.is_msix && !(info->flags & MSI_FLAG_PCI_MSIX))
1399 return -ENOTSUPP;
1400
1401 return 0;
1402}
1403
1404static int pci_msi_domain_handle_error(struct irq_domain *domain,
1405 struct msi_desc *desc, int error)
1406{
1407 /* Special handling to support pci_enable_msi_range() */
1408 if (pci_msi_desc_is_multi_msi(desc) && error == -ENOSPC)
1409 return 1;
1410
1411 return error;
1412}
1413
1414#ifdef GENERIC_MSI_DOMAIN_OPS
1415static void pci_msi_domain_set_desc(msi_alloc_info_t *arg,
1416 struct msi_desc *desc)
1417{
1418 arg->desc = desc;
1419 arg->hwirq = pci_msi_domain_calc_hwirq(msi_desc_to_pci_dev(desc),
1420 desc);
1421}
1422#else
1423#define pci_msi_domain_set_desc NULL
1424#endif
1425
1426static struct msi_domain_ops pci_msi_domain_ops_default = {
1427 .set_desc = pci_msi_domain_set_desc,
1428 .msi_check = pci_msi_domain_check_cap,
1429 .handle_error = pci_msi_domain_handle_error,
1430};
1431
1432static void pci_msi_domain_update_dom_ops(struct msi_domain_info *info)
1433{
1434 struct msi_domain_ops *ops = info->ops;
1435
1436 if (ops == NULL) {
1437 info->ops = &pci_msi_domain_ops_default;
1438 } else {
1439 if (ops->set_desc == NULL)
1440 ops->set_desc = pci_msi_domain_set_desc;
1441 if (ops->msi_check == NULL)
1442 ops->msi_check = pci_msi_domain_check_cap;
1443 if (ops->handle_error == NULL)
1444 ops->handle_error = pci_msi_domain_handle_error;
1445 }
1446}
1447
1448static void pci_msi_domain_update_chip_ops(struct msi_domain_info *info)
1449{
1450 struct irq_chip *chip = info->chip;
1451
1452 BUG_ON(!chip);
1453 if (!chip->irq_write_msi_msg)
1454 chip->irq_write_msi_msg = pci_msi_domain_write_msg;
1455 if (!chip->irq_mask)
1456 chip->irq_mask = pci_msi_mask_irq;
1457 if (!chip->irq_unmask)
1458 chip->irq_unmask = pci_msi_unmask_irq;
1459}
1460
1461/**
1462 * pci_msi_create_irq_domain - Create a MSI interrupt domain
1463 * @fwnode: Optional fwnode of the interrupt controller
1464 * @info: MSI domain info
1465 * @parent: Parent irq domain
1466 *
1467 * Updates the domain and chip ops and creates a MSI interrupt domain.
1468 *
1469 * Returns:
1470 * A domain pointer or NULL in case of failure.
1471 */
1472struct irq_domain *pci_msi_create_irq_domain(struct fwnode_handle *fwnode,
1473 struct msi_domain_info *info,
1474 struct irq_domain *parent)
1475{
1476 struct irq_domain *domain;
1477
1478 if (info->flags & MSI_FLAG_USE_DEF_DOM_OPS)
1479 pci_msi_domain_update_dom_ops(info);
1480 if (info->flags & MSI_FLAG_USE_DEF_CHIP_OPS)
1481 pci_msi_domain_update_chip_ops(info);
1482
1483 info->flags |= MSI_FLAG_ACTIVATE_EARLY;
1484
1485 domain = msi_create_irq_domain(fwnode, info, parent);
1486 if (!domain)
1487 return NULL;
1488
1489 domain->bus_token = DOMAIN_BUS_PCI_MSI;
1490 return domain;
1491}
1492EXPORT_SYMBOL_GPL(pci_msi_create_irq_domain);
1493
1494/**
1495 * pci_msi_domain_alloc_irqs - Allocate interrupts for @dev in @domain
1496 * @domain: The interrupt domain to allocate from
1497 * @dev: The device for which to allocate
1498 * @nvec: The number of interrupts to allocate
1499 * @type: Unused to allow simpler migration from the arch_XXX interfaces
1500 *
1501 * Returns:
1502 * A virtual interrupt number or an error code in case of failure
1503 */
1504int pci_msi_domain_alloc_irqs(struct irq_domain *domain, struct pci_dev *dev,
1505 int nvec, int type)
1506{
1507 return msi_domain_alloc_irqs(domain, &dev->dev, nvec);
1508}
1509
1510/**
1511 * pci_msi_domain_free_irqs - Free interrupts for @dev in @domain
1512 * @domain: The interrupt domain
1513 * @dev: The device for which to free interrupts
1514 */
1515void pci_msi_domain_free_irqs(struct irq_domain *domain, struct pci_dev *dev)
1516{
1517 msi_domain_free_irqs(domain, &dev->dev);
1518}
1519
1520/**
1521 * pci_msi_create_default_irq_domain - Create a default MSI interrupt domain
1522 * @fwnode: Optional fwnode of the interrupt controller
1523 * @info: MSI domain info
1524 * @parent: Parent irq domain
1525 *
1526 * Returns: A domain pointer or NULL in case of failure. If successful
1527 * the default PCI/MSI irqdomain pointer is updated.
1528 */
1529struct irq_domain *pci_msi_create_default_irq_domain(struct fwnode_handle *fwnode,
1530 struct msi_domain_info *info, struct irq_domain *parent)
1531{
1532 struct irq_domain *domain;
1533
1534 mutex_lock(&pci_msi_domain_lock);
1535 if (pci_msi_default_domain) {
1536 pr_err("PCI: default irq domain for PCI MSI has already been created.\n");
1537 domain = NULL;
1538 } else {
1539 domain = pci_msi_create_irq_domain(fwnode, info, parent);
1540 pci_msi_default_domain = domain;
1541 }
1542 mutex_unlock(&pci_msi_domain_lock);
1543
1544 return domain;
1545}
1546
1547static int get_msi_id_cb(struct pci_dev *pdev, u16 alias, void *data)
1548{
1549 u32 *pa = data;
1550
1551 *pa = alias;
1552 return 0;
1553}
1554/**
1555 * pci_msi_domain_get_msi_rid - Get the MSI requester id (RID)
1556 * @domain: The interrupt domain
1557 * @pdev: The PCI device.
1558 *
1559 * The RID for a device is formed from the alias, with a firmware
1560 * supplied mapping applied
1561 *
1562 * Returns: The RID.
1563 */
1564u32 pci_msi_domain_get_msi_rid(struct irq_domain *domain, struct pci_dev *pdev)
1565{
1566 struct device_node *of_node;
1567 u32 rid = 0;
1568
1569 pci_for_each_dma_alias(pdev, get_msi_id_cb, &rid);
1570
1571 of_node = irq_domain_get_of_node(domain);
1572 rid = of_node ? of_msi_map_rid(&pdev->dev, of_node, rid) :
1573 iort_msi_map_rid(&pdev->dev, rid);
1574
1575 return rid;
1576}
1577
1578/**
1579 * pci_msi_get_device_domain - Get the MSI domain for a given PCI device
1580 * @pdev: The PCI device
1581 *
1582 * Use the firmware data to find a device-specific MSI domain
1583 * (i.e. not one that is ste as a default).
1584 *
1585 * Returns: The coresponding MSI domain or NULL if none has been found.
1586 */
1587struct irq_domain *pci_msi_get_device_domain(struct pci_dev *pdev)
1588{
1589 struct irq_domain *dom;
1590 u32 rid = 0;
1591
1592 pci_for_each_dma_alias(pdev, get_msi_id_cb, &rid);
1593 dom = of_msi_map_get_device_domain(&pdev->dev, rid);
1594 if (!dom)
1595 dom = iort_get_device_domain(&pdev->dev, rid);
1596 return dom;
1597}
1598#endif /* CONFIG_PCI_MSI_IRQ_DOMAIN */