]> Git Repo - linux.git/blame - drivers/pci/msi.c
Merge tag 'ext4_for_linus_stable' of git://git.kernel.org/pub/scm/linux/kernel/git...
[linux.git] / drivers / pci / msi.c
CommitLineData
7328c8f4 1// SPDX-License-Identifier: GPL-2.0
1da177e4 2/*
df62ab5e 3 * PCI Message Signaled Interrupt (MSI)
1da177e4
LT
4 *
5 * Copyright (C) 2003-2004 Intel
6 * Copyright (C) Tom Long Nguyen ([email protected])
aff17164 7 * Copyright (C) 2016 Christoph Hellwig.
1da177e4
LT
8 */
9
1ce03373 10#include <linux/err.h>
1da177e4
LT
11#include <linux/mm.h>
12#include <linux/irq.h>
13#include <linux/interrupt.h>
363c75db 14#include <linux/export.h>
1da177e4 15#include <linux/ioport.h>
1da177e4
LT
16#include <linux/pci.h>
17#include <linux/proc_fs.h>
3b7d1921 18#include <linux/msi.h>
4fdadebc 19#include <linux/smp.h>
500559a9
HS
20#include <linux/errno.h>
21#include <linux/io.h>
be2021ba 22#include <linux/acpi_iort.h>
5a0e3ad6 23#include <linux/slab.h>
3878eaef 24#include <linux/irqdomain.h>
b6eec9b7 25#include <linux/of_irq.h>
1da177e4
LT
26
27#include "pci.h"
1da177e4 28
1da177e4 29static int pci_msi_enable = 1;
38737d82 30int pci_msi_ignore_mask;
1da177e4 31
527eee29
BH
32#define msix_table_size(flags) ((flags & PCI_MSIX_FLAGS_QSIZE) + 1)
33
8e047ada 34#ifdef CONFIG_PCI_MSI_IRQ_DOMAIN
8e047ada
JL
35static int pci_msi_setup_msi_irqs(struct pci_dev *dev, int nvec, int type)
36{
37 struct irq_domain *domain;
38
47feb418 39 domain = dev_get_msi_domain(&dev->dev);
3845d295 40 if (domain && irq_domain_is_hierarchy(domain))
699c4cec 41 return msi_domain_alloc_irqs(domain, &dev->dev, nvec);
8e047ada
JL
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
47feb418 50 domain = dev_get_msi_domain(&dev->dev);
3845d295 51 if (domain && irq_domain_is_hierarchy(domain))
699c4cec 52 msi_domain_free_irqs(domain, &dev->dev);
8e047ada
JL
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
527eee29 60
6a9e7f20
AB
61/* Arch hooks */
62
4287d824
TP
63int __weak arch_setup_msi_irq(struct pci_dev *dev, struct msi_desc *desc)
64{
2291ec09 65 struct msi_controller *chip = dev->bus->msi;
0cbdcfcf
TR
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;
4287d824
TP
78}
79
80void __weak arch_teardown_msi_irq(unsigned int irq)
6a9e7f20 81{
c2791b80 82 struct msi_controller *chip = irq_get_chip_data(irq);
0cbdcfcf
TR
83
84 if (!chip || !chip->teardown_irq)
85 return;
86
87 chip->teardown_irq(chip, irq);
6a9e7f20
AB
88}
89
4287d824 90int __weak arch_setup_msi_irqs(struct pci_dev *dev, int nvec, int type)
6a9e7f20 91{
339e5b44 92 struct msi_controller *chip = dev->bus->msi;
6a9e7f20
AB
93 struct msi_desc *entry;
94 int ret;
95
339e5b44
LS
96 if (chip && chip->setup_irqs)
97 return chip->setup_irqs(chip, dev, nvec, type);
1c8d7b0a
MW
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
5004e98a 105 for_each_pci_msi_entry(entry, dev) {
6a9e7f20 106 ret = arch_setup_msi_irq(dev, entry);
b5fbf533 107 if (ret < 0)
6a9e7f20 108 return ret;
b5fbf533
ME
109 if (ret > 0)
110 return -ENOSPC;
6a9e7f20
AB
111 }
112
113 return 0;
114}
1525bf0d 115
4287d824
TP
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 */
1525bf0d 120void default_teardown_msi_irqs(struct pci_dev *dev)
6a9e7f20 121{
63a7b17e 122 int i;
6a9e7f20
AB
123 struct msi_desc *entry;
124
5004e98a 125 for_each_pci_msi_entry(entry, dev)
63a7b17e
JL
126 if (entry->irq)
127 for (i = 0; i < entry->nvec_used; i++)
128 arch_teardown_msi_irq(entry->irq + i);
6a9e7f20
AB
129}
130
4287d824
TP
131void __weak arch_teardown_msi_irqs(struct pci_dev *dev)
132{
133 return default_teardown_msi_irqs(dev);
134}
76ccc297 135
ac8344c4 136static void default_restore_msi_irq(struct pci_dev *dev, int irq)
76ccc297
KRW
137{
138 struct msi_desc *entry;
139
140 entry = NULL;
141 if (dev->msix_enabled) {
5004e98a 142 for_each_pci_msi_entry(entry, dev) {
76ccc297
KRW
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)
83a18912 151 __pci_write_msi_msg(entry, &entry->msg);
76ccc297 152}
4287d824 153
ac8344c4 154void __weak arch_restore_msi_irqs(struct pci_dev *dev)
4287d824 155{
ac8344c4 156 return default_restore_msi_irqs(dev);
4287d824 157}
76ccc297 158
bffac3c5
MW
159static inline __attribute_const__ u32 msi_mask(unsigned x)
160{
0b49ec37
MW
161 /* Don't shift by >= width of type */
162 if (x >= 5)
163 return 0xffffffff;
164 return (1 << (1 << x)) - 1;
bffac3c5
MW
165}
166
ce6fce42
MW
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.
ce6fce42 172 */
23ed8d57 173u32 __pci_msi_desc_mask_irq(struct msi_desc *desc, u32 mask, u32 flag)
1da177e4 174{
f2440d9a 175 u32 mask_bits = desc->masked;
1da177e4 176
38737d82 177 if (pci_msi_ignore_mask || !desc->msi_attrib.maskbit)
12abb8ba 178 return 0;
f2440d9a
MW
179
180 mask_bits &= ~mask;
181 mask_bits |= flag;
e39758e0
JL
182 pci_write_config_dword(msi_desc_to_pci_dev(desc), desc->mask_pos,
183 mask_bits);
12abb8ba
HS
184
185 return mask_bits;
186}
187
188static void msi_mask_irq(struct msi_desc *desc, u32 mask, u32 flag)
189{
23ed8d57 190 desc->masked = __pci_msi_desc_mask_irq(desc, mask, flag);
f2440d9a
MW
191}
192
5eb6d660
CH
193static void __iomem *pci_msix_desc_addr(struct msi_desc *desc)
194{
195 return desc->mask_base +
196 desc->msi_attrib.entry_nr * PCI_MSIX_ENTRY_SIZE;
197}
198
f2440d9a
MW
199/*
200 * This internal function does not flush PCI writes to the device.
201 * All users must ensure that they read from the device before either
202 * assuming that the device state is up to date, or returning out of this
203 * file. This saves a few milliseconds when initialising devices with lots
204 * of MSI-X interrupts.
205 */
23ed8d57 206u32 __pci_msix_desc_mask_irq(struct msi_desc *desc, u32 flag)
f2440d9a
MW
207{
208 u32 mask_bits = desc->masked;
38737d82
YW
209
210 if (pci_msi_ignore_mask)
211 return 0;
212
8d805286
SY
213 mask_bits &= ~PCI_MSIX_ENTRY_CTRL_MASKBIT;
214 if (flag)
215 mask_bits |= PCI_MSIX_ENTRY_CTRL_MASKBIT;
5eb6d660 216 writel(mask_bits, pci_msix_desc_addr(desc) + PCI_MSIX_ENTRY_VECTOR_CTRL);
12abb8ba
HS
217
218 return mask_bits;
219}
220
221static void msix_mask_irq(struct msi_desc *desc, u32 flag)
222{
23ed8d57 223 desc->masked = __pci_msix_desc_mask_irq(desc, flag);
f2440d9a 224}
24d27553 225
1c9db525 226static void msi_set_mask_bit(struct irq_data *data, u32 flag)
f2440d9a 227{
c391f262 228 struct msi_desc *desc = irq_data_get_msi_desc(data);
24d27553 229
f2440d9a
MW
230 if (desc->msi_attrib.is_msix) {
231 msix_mask_irq(desc, flag);
232 readl(desc->mask_base); /* Flush write to device */
233 } else {
a281b788 234 unsigned offset = data->irq - desc->irq;
1c8d7b0a 235 msi_mask_irq(desc, 1 << offset, flag << offset);
1da177e4 236 }
f2440d9a
MW
237}
238
23ed8d57
TG
239/**
240 * pci_msi_mask_irq - Generic irq chip callback to mask PCI/MSI interrupts
241 * @data: pointer to irqdata associated to that interrupt
242 */
243void pci_msi_mask_irq(struct irq_data *data)
f2440d9a 244{
1c9db525 245 msi_set_mask_bit(data, 1);
f2440d9a 246}
a4289dc2 247EXPORT_SYMBOL_GPL(pci_msi_mask_irq);
f2440d9a 248
23ed8d57
TG
249/**
250 * pci_msi_unmask_irq - Generic irq chip callback to unmask PCI/MSI interrupts
251 * @data: pointer to irqdata associated to that interrupt
252 */
253void pci_msi_unmask_irq(struct irq_data *data)
f2440d9a 254{
1c9db525 255 msi_set_mask_bit(data, 0);
1da177e4 256}
a4289dc2 257EXPORT_SYMBOL_GPL(pci_msi_unmask_irq);
1da177e4 258
ac8344c4
D
259void default_restore_msi_irqs(struct pci_dev *dev)
260{
261 struct msi_desc *entry;
262
5004e98a 263 for_each_pci_msi_entry(entry, dev)
ac8344c4 264 default_restore_msi_irq(dev, entry->irq);
ac8344c4
D
265}
266
891d4a48 267void __pci_read_msi_msg(struct msi_desc *entry, struct msi_msg *msg)
1da177e4 268{
e39758e0
JL
269 struct pci_dev *dev = msi_desc_to_pci_dev(entry);
270
271 BUG_ON(dev->current_state != PCI_D0);
30da5524
BH
272
273 if (entry->msi_attrib.is_msix) {
5eb6d660 274 void __iomem *base = pci_msix_desc_addr(entry);
30da5524
BH
275
276 msg->address_lo = readl(base + PCI_MSIX_ENTRY_LOWER_ADDR);
277 msg->address_hi = readl(base + PCI_MSIX_ENTRY_UPPER_ADDR);
278 msg->data = readl(base + PCI_MSIX_ENTRY_DATA);
279 } else {
f5322169 280 int pos = dev->msi_cap;
30da5524
BH
281 u16 data;
282
9925ad0c
BH
283 pci_read_config_dword(dev, pos + PCI_MSI_ADDRESS_LO,
284 &msg->address_lo);
30da5524 285 if (entry->msi_attrib.is_64) {
9925ad0c
BH
286 pci_read_config_dword(dev, pos + PCI_MSI_ADDRESS_HI,
287 &msg->address_hi);
2f221349 288 pci_read_config_word(dev, pos + PCI_MSI_DATA_64, &data);
30da5524
BH
289 } else {
290 msg->address_hi = 0;
2f221349 291 pci_read_config_word(dev, pos + PCI_MSI_DATA_32, &data);
30da5524
BH
292 }
293 msg->data = data;
294 }
295}
296
83a18912 297void __pci_write_msi_msg(struct msi_desc *entry, struct msi_msg *msg)
3145e941 298{
e39758e0
JL
299 struct pci_dev *dev = msi_desc_to_pci_dev(entry);
300
0170591b 301 if (dev->current_state != PCI_D0 || pci_dev_is_disconnected(dev)) {
fcd097f3
BH
302 /* Don't touch the hardware now */
303 } else if (entry->msi_attrib.is_msix) {
5eb6d660 304 void __iomem *base = pci_msix_desc_addr(entry);
24d27553 305
2c21fd4b
HS
306 writel(msg->address_lo, base + PCI_MSIX_ENTRY_LOWER_ADDR);
307 writel(msg->address_hi, base + PCI_MSIX_ENTRY_UPPER_ADDR);
308 writel(msg->data, base + PCI_MSIX_ENTRY_DATA);
24d27553 309 } else {
f5322169 310 int pos = dev->msi_cap;
1c8d7b0a
MW
311 u16 msgctl;
312
f84ecd28 313 pci_read_config_word(dev, pos + PCI_MSI_FLAGS, &msgctl);
1c8d7b0a
MW
314 msgctl &= ~PCI_MSI_FLAGS_QSIZE;
315 msgctl |= entry->msi_attrib.multiple << 4;
f84ecd28 316 pci_write_config_word(dev, pos + PCI_MSI_FLAGS, msgctl);
0366f8f7 317
9925ad0c
BH
318 pci_write_config_dword(dev, pos + PCI_MSI_ADDRESS_LO,
319 msg->address_lo);
0366f8f7 320 if (entry->msi_attrib.is_64) {
9925ad0c
BH
321 pci_write_config_dword(dev, pos + PCI_MSI_ADDRESS_HI,
322 msg->address_hi);
2f221349
BH
323 pci_write_config_word(dev, pos + PCI_MSI_DATA_64,
324 msg->data);
0366f8f7 325 } else {
2f221349
BH
326 pci_write_config_word(dev, pos + PCI_MSI_DATA_32,
327 msg->data);
0366f8f7 328 }
1da177e4 329 }
392ee1e6 330 entry->msg = *msg;
1da177e4 331}
0366f8f7 332
83a18912 333void pci_write_msi_msg(unsigned int irq, struct msi_msg *msg)
3145e941 334{
dced35ae 335 struct msi_desc *entry = irq_get_msi_desc(irq);
3145e941 336
83a18912 337 __pci_write_msi_msg(entry, msg);
3145e941 338}
83a18912 339EXPORT_SYMBOL_GPL(pci_write_msi_msg);
3145e941 340
f56e4481
HS
341static void free_msi_irqs(struct pci_dev *dev)
342{
5004e98a 343 struct list_head *msi_list = dev_to_msi_list(&dev->dev);
f56e4481 344 struct msi_desc *entry, *tmp;
1c51b50c
GKH
345 struct attribute **msi_attrs;
346 struct device_attribute *dev_attr;
63a7b17e 347 int i, count = 0;
f56e4481 348
5004e98a 349 for_each_pci_msi_entry(entry, dev)
63a7b17e
JL
350 if (entry->irq)
351 for (i = 0; i < entry->nvec_used; i++)
352 BUG_ON(irq_has_action(entry->irq + i));
f56e4481 353
8e047ada 354 pci_msi_teardown_msi_irqs(dev);
f56e4481 355
5004e98a 356 list_for_each_entry_safe(entry, tmp, msi_list, list) {
f56e4481 357 if (entry->msi_attrib.is_msix) {
5004e98a 358 if (list_is_last(&entry->list, msi_list))
f56e4481
HS
359 iounmap(entry->mask_base);
360 }
424eb391 361
f56e4481 362 list_del(&entry->list);
81efbadd 363 free_msi_entry(entry);
f56e4481 364 }
1c51b50c
GKH
365
366 if (dev->msi_irq_groups) {
367 sysfs_remove_groups(&dev->dev.kobj, dev->msi_irq_groups);
368 msi_attrs = dev->msi_irq_groups[0]->attrs;
b701c0b1 369 while (msi_attrs[count]) {
1c51b50c
GKH
370 dev_attr = container_of(msi_attrs[count],
371 struct device_attribute, attr);
372 kfree(dev_attr->attr.name);
373 kfree(dev_attr);
374 ++count;
375 }
376 kfree(msi_attrs);
377 kfree(dev->msi_irq_groups[0]);
378 kfree(dev->msi_irq_groups);
379 dev->msi_irq_groups = NULL;
380 }
f56e4481 381}
c54c1879 382
ba698ad4
DM
383static void pci_intx_for_msi(struct pci_dev *dev, int enable)
384{
385 if (!(dev->dev_flags & PCI_DEV_FLAGS_MSI_INTX_DISABLE_BUG))
386 pci_intx(dev, enable);
387}
388
8fed4b65 389static void __pci_restore_msi_state(struct pci_dev *dev)
41017f0c 390{
41017f0c 391 u16 control;
392ee1e6 392 struct msi_desc *entry;
41017f0c 393
b1cbf4e4
EB
394 if (!dev->msi_enabled)
395 return;
396
dced35ae 397 entry = irq_get_msi_desc(dev->irq);
41017f0c 398
ba698ad4 399 pci_intx_for_msi(dev, 0);
61b64abd 400 pci_msi_set_enable(dev, 0);
ac8344c4 401 arch_restore_msi_irqs(dev);
392ee1e6 402
f5322169 403 pci_read_config_word(dev, dev->msi_cap + PCI_MSI_FLAGS, &control);
31ea5d4d
YW
404 msi_mask_irq(entry, msi_mask(entry->msi_attrib.multi_cap),
405 entry->masked);
abad2ec9 406 control &= ~PCI_MSI_FLAGS_QSIZE;
1c8d7b0a 407 control |= (entry->msi_attrib.multiple << 4) | PCI_MSI_FLAGS_ENABLE;
f5322169 408 pci_write_config_word(dev, dev->msi_cap + PCI_MSI_FLAGS, control);
8fed4b65
ME
409}
410
411static void __pci_restore_msix_state(struct pci_dev *dev)
41017f0c 412{
41017f0c 413 struct msi_desc *entry;
41017f0c 414
ded86d8d
EB
415 if (!dev->msix_enabled)
416 return;
5004e98a 417 BUG_ON(list_empty(dev_to_msi_list(&dev->dev)));
ded86d8d 418
41017f0c 419 /* route the table */
ba698ad4 420 pci_intx_for_msi(dev, 0);
61b64abd 421 pci_msix_clear_and_set_ctrl(dev, 0,
66f0d0c4 422 PCI_MSIX_FLAGS_ENABLE | PCI_MSIX_FLAGS_MASKALL);
41017f0c 423
ac8344c4 424 arch_restore_msi_irqs(dev);
5004e98a 425 for_each_pci_msi_entry(entry, dev)
f2440d9a 426 msix_mask_irq(entry, entry->masked);
41017f0c 427
61b64abd 428 pci_msix_clear_and_set_ctrl(dev, PCI_MSIX_FLAGS_MASKALL, 0);
41017f0c 429}
8fed4b65
ME
430
431void pci_restore_msi_state(struct pci_dev *dev)
432{
433 __pci_restore_msi_state(dev);
434 __pci_restore_msix_state(dev);
435}
94688cf2 436EXPORT_SYMBOL_GPL(pci_restore_msi_state);
41017f0c 437
1c51b50c 438static ssize_t msi_mode_show(struct device *dev, struct device_attribute *attr,
da8d1c8b
NH
439 char *buf)
440{
1c51b50c
GKH
441 struct msi_desc *entry;
442 unsigned long irq;
443 int retval;
da8d1c8b 444
1c51b50c
GKH
445 retval = kstrtoul(attr->attr.name, 10, &irq);
446 if (retval)
447 return retval;
da8d1c8b 448
e11ece5a
YW
449 entry = irq_get_msi_desc(irq);
450 if (entry)
451 return sprintf(buf, "%s\n",
452 entry->msi_attrib.is_msix ? "msix" : "msi");
453
1c51b50c 454 return -ENODEV;
da8d1c8b
NH
455}
456
da8d1c8b
NH
457static int populate_msi_sysfs(struct pci_dev *pdev)
458{
1c51b50c
GKH
459 struct attribute **msi_attrs;
460 struct attribute *msi_attr;
461 struct device_attribute *msi_dev_attr;
462 struct attribute_group *msi_irq_group;
463 const struct attribute_group **msi_irq_groups;
da8d1c8b 464 struct msi_desc *entry;
1c51b50c
GKH
465 int ret = -ENOMEM;
466 int num_msi = 0;
da8d1c8b 467 int count = 0;
a8676066 468 int i;
da8d1c8b 469
1c51b50c 470 /* Determine how many msi entries we have */
5004e98a 471 for_each_pci_msi_entry(entry, pdev)
a8676066 472 num_msi += entry->nvec_used;
1c51b50c
GKH
473 if (!num_msi)
474 return 0;
da8d1c8b 475
1c51b50c 476 /* Dynamically create the MSI attributes for the PCI device */
6396bb22 477 msi_attrs = kcalloc(num_msi + 1, sizeof(void *), GFP_KERNEL);
1c51b50c
GKH
478 if (!msi_attrs)
479 return -ENOMEM;
5004e98a 480 for_each_pci_msi_entry(entry, pdev) {
a8676066
RB
481 for (i = 0; i < entry->nvec_used; i++) {
482 msi_dev_attr = kzalloc(sizeof(*msi_dev_attr), GFP_KERNEL);
483 if (!msi_dev_attr)
484 goto error_attrs;
485 msi_attrs[count] = &msi_dev_attr->attr;
486
487 sysfs_attr_init(&msi_dev_attr->attr);
488 msi_dev_attr->attr.name = kasprintf(GFP_KERNEL, "%d",
489 entry->irq + i);
490 if (!msi_dev_attr->attr.name)
491 goto error_attrs;
492 msi_dev_attr->attr.mode = S_IRUGO;
493 msi_dev_attr->show = msi_mode_show;
494 ++count;
495 }
da8d1c8b
NH
496 }
497
1c51b50c
GKH
498 msi_irq_group = kzalloc(sizeof(*msi_irq_group), GFP_KERNEL);
499 if (!msi_irq_group)
500 goto error_attrs;
501 msi_irq_group->name = "msi_irqs";
502 msi_irq_group->attrs = msi_attrs;
503
6396bb22 504 msi_irq_groups = kcalloc(2, sizeof(void *), GFP_KERNEL);
1c51b50c
GKH
505 if (!msi_irq_groups)
506 goto error_irq_group;
507 msi_irq_groups[0] = msi_irq_group;
508
509 ret = sysfs_create_groups(&pdev->dev.kobj, msi_irq_groups);
510 if (ret)
511 goto error_irq_groups;
512 pdev->msi_irq_groups = msi_irq_groups;
513
da8d1c8b
NH
514 return 0;
515
1c51b50c
GKH
516error_irq_groups:
517 kfree(msi_irq_groups);
518error_irq_group:
519 kfree(msi_irq_group);
520error_attrs:
521 count = 0;
522 msi_attr = msi_attrs[count];
523 while (msi_attr) {
524 msi_dev_attr = container_of(msi_attr, struct device_attribute, attr);
525 kfree(msi_attr->name);
526 kfree(msi_dev_attr);
527 ++count;
528 msi_attr = msi_attrs[count];
da8d1c8b 529 }
29237756 530 kfree(msi_attrs);
da8d1c8b
NH
531 return ret;
532}
533
e75eafb9 534static struct msi_desc *
61e1c590 535msi_setup_entry(struct pci_dev *dev, int nvec, const struct irq_affinity *affd)
d873b4d4 536{
e75eafb9 537 struct cpumask *masks = NULL;
d873b4d4 538 struct msi_desc *entry;
e75eafb9
TG
539 u16 control;
540
8e1101d2 541 if (affd)
61e1c590 542 masks = irq_create_affinity_masks(nvec, affd);
8e1101d2 543
d873b4d4
YW
544
545 /* MSI Entry Initialization */
e75eafb9 546 entry = alloc_msi_entry(&dev->dev, nvec, masks);
d873b4d4 547 if (!entry)
e75eafb9 548 goto out;
d873b4d4
YW
549
550 pci_read_config_word(dev, dev->msi_cap + PCI_MSI_FLAGS, &control);
551
552 entry->msi_attrib.is_msix = 0;
553 entry->msi_attrib.is_64 = !!(control & PCI_MSI_FLAGS_64BIT);
554 entry->msi_attrib.entry_nr = 0;
555 entry->msi_attrib.maskbit = !!(control & PCI_MSI_FLAGS_MASKBIT);
556 entry->msi_attrib.default_irq = dev->irq; /* Save IOAPIC IRQ */
d873b4d4 557 entry->msi_attrib.multi_cap = (control & PCI_MSI_FLAGS_QMASK) >> 1;
63a7b17e 558 entry->msi_attrib.multiple = ilog2(__roundup_pow_of_two(nvec));
d873b4d4
YW
559
560 if (control & PCI_MSI_FLAGS_64BIT)
561 entry->mask_pos = dev->msi_cap + PCI_MSI_MASK_64;
562 else
563 entry->mask_pos = dev->msi_cap + PCI_MSI_MASK_32;
564
565 /* Save the initial mask status */
566 if (entry->msi_attrib.maskbit)
567 pci_read_config_dword(dev, entry->mask_pos, &entry->masked);
568
e75eafb9
TG
569out:
570 kfree(masks);
d873b4d4
YW
571 return entry;
572}
573
f144d149
BH
574static int msi_verify_entries(struct pci_dev *dev)
575{
576 struct msi_desc *entry;
577
5004e98a 578 for_each_pci_msi_entry(entry, dev) {
f144d149
BH
579 if (!dev->no_64bit_msi || !entry->msg.address_hi)
580 continue;
7506dc79 581 pci_err(dev, "Device has broken 64-bit MSI but arch"
f144d149
BH
582 " tried to assign one above 4G\n");
583 return -EIO;
584 }
585 return 0;
586}
587
1da177e4
LT
588/**
589 * msi_capability_init - configure device's MSI capability structure
590 * @dev: pointer to the pci_dev data structure of MSI device function
1c8d7b0a 591 * @nvec: number of interrupts to allocate
dadf1733 592 * @affd: description of automatic irq affinity assignments (may be %NULL)
1da177e4 593 *
1c8d7b0a
MW
594 * Setup the MSI capability structure of the device with the requested
595 * number of interrupts. A return value of zero indicates the successful
596 * setup of an entry with the new MSI irq. A negative return value indicates
597 * an error, and a positive return value indicates the number of interrupts
598 * which could have been allocated.
599 */
61e1c590
CH
600static int msi_capability_init(struct pci_dev *dev, int nvec,
601 const struct irq_affinity *affd)
1da177e4
LT
602{
603 struct msi_desc *entry;
f465136d 604 int ret;
f2440d9a 605 unsigned mask;
1da177e4 606
61b64abd 607 pci_msi_set_enable(dev, 0); /* Disable MSI during set up */
110828c9 608
61e1c590 609 entry = msi_setup_entry(dev, nvec, affd);
f7feaca7
EB
610 if (!entry)
611 return -ENOMEM;
1ce03373 612
f2440d9a 613 /* All MSIs are unmasked by default, Mask them all */
31ea5d4d 614 mask = msi_mask(entry->msi_attrib.multi_cap);
f2440d9a
MW
615 msi_mask_irq(entry, mask, mask);
616
5004e98a 617 list_add_tail(&entry->list, dev_to_msi_list(&dev->dev));
9c831334 618
1da177e4 619 /* Configure MSI capability structure */
8e047ada 620 ret = pci_msi_setup_msi_irqs(dev, nvec, PCI_CAP_ID_MSI);
7fe3730d 621 if (ret) {
7ba1930d 622 msi_mask_irq(entry, mask, ~mask);
f56e4481 623 free_msi_irqs(dev);
7fe3730d 624 return ret;
fd58e55f 625 }
f7feaca7 626
f144d149
BH
627 ret = msi_verify_entries(dev);
628 if (ret) {
629 msi_mask_irq(entry, mask, ~mask);
630 free_msi_irqs(dev);
631 return ret;
632 }
633
da8d1c8b
NH
634 ret = populate_msi_sysfs(dev);
635 if (ret) {
636 msi_mask_irq(entry, mask, ~mask);
637 free_msi_irqs(dev);
638 return ret;
639 }
640
1da177e4 641 /* Set MSI enabled bits */
ba698ad4 642 pci_intx_for_msi(dev, 0);
61b64abd 643 pci_msi_set_enable(dev, 1);
b1cbf4e4 644 dev->msi_enabled = 1;
1da177e4 645
5f226991 646 pcibios_free_irq(dev);
7fe3730d 647 dev->irq = entry->irq;
1da177e4
LT
648 return 0;
649}
650
520fe9dc 651static void __iomem *msix_map_region(struct pci_dev *dev, unsigned nr_entries)
5a05a9d8 652{
4302e0fb 653 resource_size_t phys_addr;
5a05a9d8 654 u32 table_offset;
6a878e50 655 unsigned long flags;
5a05a9d8
HS
656 u8 bir;
657
909094c6
BH
658 pci_read_config_dword(dev, dev->msix_cap + PCI_MSIX_TABLE,
659 &table_offset);
4d18760c 660 bir = (u8)(table_offset & PCI_MSIX_TABLE_BIR);
6a878e50
YW
661 flags = pci_resource_flags(dev, bir);
662 if (!flags || (flags & IORESOURCE_UNSET))
663 return NULL;
664
4d18760c 665 table_offset &= PCI_MSIX_TABLE_OFFSET;
5a05a9d8
HS
666 phys_addr = pci_resource_start(dev, bir) + table_offset;
667
668 return ioremap_nocache(phys_addr, nr_entries * PCI_MSIX_ENTRY_SIZE);
669}
670
520fe9dc 671static int msix_setup_entries(struct pci_dev *dev, void __iomem *base,
e75eafb9 672 struct msix_entry *entries, int nvec,
61e1c590 673 const struct irq_affinity *affd)
d9d7070e 674{
e75eafb9 675 struct cpumask *curmsk, *masks = NULL;
d9d7070e 676 struct msi_desc *entry;
e75eafb9 677 int ret, i;
4ef33685 678
8e1101d2 679 if (affd)
61e1c590 680 masks = irq_create_affinity_masks(nvec, affd);
4ef33685 681
e75eafb9
TG
682 for (i = 0, curmsk = masks; i < nvec; i++) {
683 entry = alloc_msi_entry(&dev->dev, 1, curmsk);
d9d7070e
HS
684 if (!entry) {
685 if (!i)
686 iounmap(base);
687 else
688 free_msi_irqs(dev);
689 /* No enough memory. Don't try again */
e75eafb9
TG
690 ret = -ENOMEM;
691 goto out;
d9d7070e
HS
692 }
693
694 entry->msi_attrib.is_msix = 1;
695 entry->msi_attrib.is_64 = 1;
3ac020e0
CH
696 if (entries)
697 entry->msi_attrib.entry_nr = entries[i].entry;
698 else
699 entry->msi_attrib.entry_nr = i;
d9d7070e 700 entry->msi_attrib.default_irq = dev->irq;
d9d7070e
HS
701 entry->mask_base = base;
702
5004e98a 703 list_add_tail(&entry->list, dev_to_msi_list(&dev->dev));
e75eafb9
TG
704 if (masks)
705 curmsk++;
d9d7070e 706 }
e75eafb9
TG
707 ret = 0;
708out:
709 kfree(masks);
3adfb572 710 return ret;
d9d7070e
HS
711}
712
75cb3426 713static void msix_program_entries(struct pci_dev *dev,
520fe9dc 714 struct msix_entry *entries)
75cb3426
HS
715{
716 struct msi_desc *entry;
717 int i = 0;
718
5004e98a 719 for_each_pci_msi_entry(entry, dev) {
3ac020e0
CH
720 if (entries)
721 entries[i++].vector = entry->irq;
12eb21de
CH
722 entry->masked = readl(pci_msix_desc_addr(entry) +
723 PCI_MSIX_ENTRY_VECTOR_CTRL);
75cb3426 724 msix_mask_irq(entry, 1);
75cb3426
HS
725 }
726}
727
1da177e4
LT
728/**
729 * msix_capability_init - configure device's MSI-X capability
730 * @dev: pointer to the pci_dev data structure of MSI-X device function
8f7020d3
RD
731 * @entries: pointer to an array of struct msix_entry entries
732 * @nvec: number of @entries
61e1c590 733 * @affd: Optional pointer to enable automatic affinity assignement
1da177e4 734 *
eaae4b3a 735 * Setup the MSI-X capability structure of device function with a
1ce03373
EB
736 * single MSI-X irq. A return of zero indicates the successful setup of
737 * requested MSI-X entries with allocated irqs or non-zero for otherwise.
1da177e4 738 **/
e75eafb9 739static int msix_capability_init(struct pci_dev *dev, struct msix_entry *entries,
61e1c590 740 int nvec, const struct irq_affinity *affd)
1da177e4 741{
520fe9dc 742 int ret;
5a05a9d8 743 u16 control;
1da177e4
LT
744 void __iomem *base;
745
f598282f 746 /* Ensure MSI-X is disabled while it is set up */
61b64abd 747 pci_msix_clear_and_set_ctrl(dev, PCI_MSIX_FLAGS_ENABLE, 0);
f598282f 748
66f0d0c4 749 pci_read_config_word(dev, dev->msix_cap + PCI_MSIX_FLAGS, &control);
1da177e4 750 /* Request & Map MSI-X table region */
527eee29 751 base = msix_map_region(dev, msix_table_size(control));
5a05a9d8 752 if (!base)
1da177e4
LT
753 return -ENOMEM;
754
61e1c590 755 ret = msix_setup_entries(dev, base, entries, nvec, affd);
d9d7070e
HS
756 if (ret)
757 return ret;
9c831334 758
8e047ada 759 ret = pci_msi_setup_msi_irqs(dev, nvec, PCI_CAP_ID_MSIX);
583871d4 760 if (ret)
2adc7907 761 goto out_avail;
9c831334 762
f144d149
BH
763 /* Check if all MSI entries honor device restrictions */
764 ret = msi_verify_entries(dev);
765 if (ret)
766 goto out_free;
767
f598282f
MW
768 /*
769 * Some devices require MSI-X to be enabled before we can touch the
770 * MSI-X registers. We need to mask all the vectors to prevent
771 * interrupts coming in before they're fully set up.
772 */
61b64abd 773 pci_msix_clear_and_set_ctrl(dev, 0,
66f0d0c4 774 PCI_MSIX_FLAGS_MASKALL | PCI_MSIX_FLAGS_ENABLE);
f598282f 775
75cb3426 776 msix_program_entries(dev, entries);
f598282f 777
da8d1c8b 778 ret = populate_msi_sysfs(dev);
2adc7907
AG
779 if (ret)
780 goto out_free;
da8d1c8b 781
f598282f 782 /* Set MSI-X enabled bits and unmask the function */
ba698ad4 783 pci_intx_for_msi(dev, 0);
b1cbf4e4 784 dev->msix_enabled = 1;
61b64abd 785 pci_msix_clear_and_set_ctrl(dev, PCI_MSIX_FLAGS_MASKALL, 0);
8d181018 786
5f226991 787 pcibios_free_irq(dev);
1da177e4 788 return 0;
583871d4 789
2adc7907 790out_avail:
583871d4
HS
791 if (ret < 0) {
792 /*
793 * If we had some success, report the number of irqs
794 * we succeeded in setting up.
795 */
d9d7070e 796 struct msi_desc *entry;
583871d4
HS
797 int avail = 0;
798
5004e98a 799 for_each_pci_msi_entry(entry, dev) {
583871d4
HS
800 if (entry->irq != 0)
801 avail++;
802 }
803 if (avail != 0)
804 ret = avail;
805 }
806
2adc7907 807out_free:
583871d4
HS
808 free_msi_irqs(dev);
809
810 return ret;
1da177e4
LT
811}
812
24334a12 813/**
a06cd74c 814 * pci_msi_supported - check whether MSI may be enabled on a device
24334a12 815 * @dev: pointer to the pci_dev data structure of MSI device function
c9953a73 816 * @nvec: how many MSIs have been requested ?
24334a12 817 *
f7625980 818 * Look at global flags, the device itself, and its parent buses
17bbc12a 819 * to determine if MSI/-X are supported for the device. If MSI/-X is
a06cd74c 820 * supported return 1, else return 0.
24334a12 821 **/
a06cd74c 822static int pci_msi_supported(struct pci_dev *dev, int nvec)
24334a12
BG
823{
824 struct pci_bus *bus;
825
0306ebfa 826 /* MSI must be globally enabled and supported by the device */
27e20603 827 if (!pci_msi_enable)
a06cd74c 828 return 0;
27e20603
AG
829
830 if (!dev || dev->no_msi || dev->current_state != PCI_D0)
a06cd74c 831 return 0;
24334a12 832
314e77b3
ME
833 /*
834 * You can't ask to have 0 or less MSIs configured.
835 * a) it's stupid ..
836 * b) the list manipulation code assumes nvec >= 1.
837 */
838 if (nvec < 1)
a06cd74c 839 return 0;
314e77b3 840
500559a9
HS
841 /*
842 * Any bridge which does NOT route MSI transactions from its
843 * secondary bus to its primary bus must set NO_MSI flag on
0306ebfa
BG
844 * the secondary pci_bus.
845 * We expect only arch-specific PCI host bus controller driver
846 * or quirks for specific PCI bridges to be setting NO_MSI.
847 */
24334a12
BG
848 for (bus = dev->bus; bus; bus = bus->parent)
849 if (bus->bus_flags & PCI_BUS_FLAGS_NO_MSI)
a06cd74c 850 return 0;
24334a12 851
a06cd74c 852 return 1;
24334a12
BG
853}
854
d1ac1d26
AG
855/**
856 * pci_msi_vec_count - Return the number of MSI vectors a device can send
857 * @dev: device to report about
858 *
859 * This function returns the number of MSI vectors a device requested via
860 * Multiple Message Capable register. It returns a negative errno if the
861 * device is not capable sending MSI interrupts. Otherwise, the call succeeds
862 * and returns a power of two, up to a maximum of 2^5 (32), according to the
863 * MSI specification.
864 **/
865int pci_msi_vec_count(struct pci_dev *dev)
866{
867 int ret;
868 u16 msgctl;
869
870 if (!dev->msi_cap)
871 return -EINVAL;
872
873 pci_read_config_word(dev, dev->msi_cap + PCI_MSI_FLAGS, &msgctl);
874 ret = 1 << ((msgctl & PCI_MSI_FLAGS_QMASK) >> 1);
875
876 return ret;
877}
878EXPORT_SYMBOL(pci_msi_vec_count);
879
688769f6 880static void pci_msi_shutdown(struct pci_dev *dev)
1da177e4 881{
f2440d9a
MW
882 struct msi_desc *desc;
883 u32 mask;
1da177e4 884
128bc5fc 885 if (!pci_msi_enable || !dev || !dev->msi_enabled)
ded86d8d
EB
886 return;
887
5004e98a 888 BUG_ON(list_empty(dev_to_msi_list(&dev->dev)));
4a7cc831 889 desc = first_pci_msi_entry(dev);
110828c9 890
61b64abd 891 pci_msi_set_enable(dev, 0);
ba698ad4 892 pci_intx_for_msi(dev, 1);
b1cbf4e4 893 dev->msi_enabled = 0;
7bd007e4 894
12abb8ba 895 /* Return the device with MSI unmasked as initial states */
31ea5d4d 896 mask = msi_mask(desc->msi_attrib.multi_cap);
12abb8ba 897 /* Keep cached state to be restored */
23ed8d57 898 __pci_msi_desc_mask_irq(desc, mask, ~mask);
e387b9ee
ME
899
900 /* Restore dev->irq to its default pin-assertion irq */
f2440d9a 901 dev->irq = desc->msi_attrib.default_irq;
5f226991 902 pcibios_alloc_irq(dev);
d52877c7 903}
24d27553 904
500559a9 905void pci_disable_msi(struct pci_dev *dev)
d52877c7 906{
d52877c7
YL
907 if (!pci_msi_enable || !dev || !dev->msi_enabled)
908 return;
909
910 pci_msi_shutdown(dev);
f56e4481 911 free_msi_irqs(dev);
1da177e4 912}
4cc086fa 913EXPORT_SYMBOL(pci_disable_msi);
1da177e4 914
a52e2e35 915/**
ff1aa430 916 * pci_msix_vec_count - return the number of device's MSI-X table entries
a52e2e35 917 * @dev: pointer to the pci_dev data structure of MSI-X device function
ff1aa430
AG
918 * This function returns the number of device's MSI-X table entries and
919 * therefore the number of MSI-X vectors device is capable of sending.
920 * It returns a negative errno if the device is not capable of sending MSI-X
921 * interrupts.
922 **/
923int pci_msix_vec_count(struct pci_dev *dev)
a52e2e35 924{
a52e2e35
RW
925 u16 control;
926
520fe9dc 927 if (!dev->msix_cap)
ff1aa430 928 return -EINVAL;
a52e2e35 929
f84ecd28 930 pci_read_config_word(dev, dev->msix_cap + PCI_MSIX_FLAGS, &control);
527eee29 931 return msix_table_size(control);
a52e2e35 932}
ff1aa430 933EXPORT_SYMBOL(pci_msix_vec_count);
a52e2e35 934
e75eafb9 935static int __pci_enable_msix(struct pci_dev *dev, struct msix_entry *entries,
61e1c590 936 int nvec, const struct irq_affinity *affd)
1da177e4 937{
5ec09405 938 int nr_entries;
ded86d8d 939 int i, j;
1da177e4 940
a06cd74c
AG
941 if (!pci_msi_supported(dev, nvec))
942 return -EINVAL;
c9953a73 943
ff1aa430
AG
944 nr_entries = pci_msix_vec_count(dev);
945 if (nr_entries < 0)
946 return nr_entries;
1da177e4 947 if (nvec > nr_entries)
57fbf52c 948 return nr_entries;
1da177e4 949
3ac020e0
CH
950 if (entries) {
951 /* Check for any invalid entries */
952 for (i = 0; i < nvec; i++) {
953 if (entries[i].entry >= nr_entries)
954 return -EINVAL; /* invalid entry */
955 for (j = i + 1; j < nvec; j++) {
956 if (entries[i].entry == entries[j].entry)
957 return -EINVAL; /* duplicate entry */
958 }
1da177e4
LT
959 }
960 }
7bd007e4 961
1ce03373 962 /* Check whether driver already requested for MSI irq */
500559a9 963 if (dev->msi_enabled) {
7506dc79 964 pci_info(dev, "can't enable MSI-X (MSI IRQ already assigned)\n");
1da177e4
LT
965 return -EINVAL;
966 }
61e1c590 967 return msix_capability_init(dev, entries, nvec, affd);
e75eafb9
TG
968}
969
688769f6 970static void pci_msix_shutdown(struct pci_dev *dev)
fc4afc7b 971{
12abb8ba
HS
972 struct msi_desc *entry;
973
128bc5fc 974 if (!pci_msi_enable || !dev || !dev->msix_enabled)
ded86d8d
EB
975 return;
976
0170591b
KB
977 if (pci_dev_is_disconnected(dev)) {
978 dev->msix_enabled = 0;
979 return;
980 }
981
12abb8ba 982 /* Return the device with MSI-X masked as initial states */
5004e98a 983 for_each_pci_msi_entry(entry, dev) {
12abb8ba 984 /* Keep cached states to be restored */
23ed8d57 985 __pci_msix_desc_mask_irq(entry, 1);
12abb8ba
HS
986 }
987
61b64abd 988 pci_msix_clear_and_set_ctrl(dev, PCI_MSIX_FLAGS_ENABLE, 0);
ba698ad4 989 pci_intx_for_msi(dev, 1);
b1cbf4e4 990 dev->msix_enabled = 0;
5f226991 991 pcibios_alloc_irq(dev);
d52877c7 992}
c901851f 993
500559a9 994void pci_disable_msix(struct pci_dev *dev)
d52877c7
YL
995{
996 if (!pci_msi_enable || !dev || !dev->msix_enabled)
997 return;
998
999 pci_msix_shutdown(dev);
f56e4481 1000 free_msi_irqs(dev);
1da177e4 1001}
4cc086fa 1002EXPORT_SYMBOL(pci_disable_msix);
1da177e4 1003
309e57df
MW
1004void pci_no_msi(void)
1005{
1006 pci_msi_enable = 0;
1007}
c9953a73 1008
07ae95f9
AP
1009/**
1010 * pci_msi_enabled - is MSI enabled?
1011 *
1012 * Returns true if MSI has not been disabled by the command-line option
1013 * pci=nomsi.
1014 **/
1015int pci_msi_enabled(void)
d389fec6 1016{
07ae95f9 1017 return pci_msi_enable;
d389fec6 1018}
07ae95f9 1019EXPORT_SYMBOL(pci_msi_enabled);
d389fec6 1020
4ef33685 1021static int __pci_enable_msi_range(struct pci_dev *dev, int minvec, int maxvec,
61e1c590 1022 const struct irq_affinity *affd)
302a2523 1023{
034cd97e 1024 int nvec;
302a2523
AG
1025 int rc;
1026
a06cd74c
AG
1027 if (!pci_msi_supported(dev, minvec))
1028 return -EINVAL;
034cd97e 1029
034cd97e
AG
1030 /* Check whether driver already requested MSI-X irqs */
1031 if (dev->msix_enabled) {
7506dc79 1032 pci_info(dev, "can't enable MSI (MSI-X already enabled)\n");
034cd97e
AG
1033 return -EINVAL;
1034 }
1035
302a2523
AG
1036 if (maxvec < minvec)
1037 return -ERANGE;
1038
4c1ef72e
TZ
1039 if (WARN_ON_ONCE(dev->msi_enabled))
1040 return -EINVAL;
1041
034cd97e
AG
1042 nvec = pci_msi_vec_count(dev);
1043 if (nvec < 0)
1044 return nvec;
4ef33685 1045 if (nvec < minvec)
948b7620 1046 return -ENOSPC;
4ef33685
CH
1047
1048 if (nvec > maxvec)
034cd97e
AG
1049 nvec = maxvec;
1050
4ef33685 1051 for (;;) {
61e1c590 1052 if (affd) {
6f9a22bc 1053 nvec = irq_calc_affinity_vectors(minvec, nvec, affd);
4ef33685
CH
1054 if (nvec < minvec)
1055 return -ENOSPC;
1056 }
1057
61e1c590 1058 rc = msi_capability_init(dev, nvec, affd);
4ef33685
CH
1059 if (rc == 0)
1060 return nvec;
1061
4ef33685 1062 if (rc < 0)
302a2523 1063 return rc;
4ef33685
CH
1064 if (rc < minvec)
1065 return -ENOSPC;
1066
1067 nvec = rc;
1068 }
1069}
1070
4fe03955
CH
1071/* deprecated, don't use */
1072int pci_enable_msi(struct pci_dev *dev)
4ef33685 1073{
4fe03955
CH
1074 int rc = __pci_enable_msi_range(dev, 1, 1, NULL);
1075 if (rc < 0)
1076 return rc;
1077 return 0;
4ef33685 1078}
4fe03955 1079EXPORT_SYMBOL(pci_enable_msi);
4ef33685
CH
1080
1081static int __pci_enable_msix_range(struct pci_dev *dev,
61e1c590
CH
1082 struct msix_entry *entries, int minvec,
1083 int maxvec, const struct irq_affinity *affd)
4ef33685 1084{
e75eafb9 1085 int rc, nvec = maxvec;
4ef33685
CH
1086
1087 if (maxvec < minvec)
1088 return -ERANGE;
1089
4c1ef72e
TZ
1090 if (WARN_ON_ONCE(dev->msix_enabled))
1091 return -EINVAL;
1092
4ef33685 1093 for (;;) {
61e1c590 1094 if (affd) {
6f9a22bc 1095 nvec = irq_calc_affinity_vectors(minvec, nvec, affd);
4ef33685 1096 if (nvec < minvec)
302a2523 1097 return -ENOSPC;
302a2523 1098 }
302a2523 1099
61e1c590 1100 rc = __pci_enable_msix(dev, entries, nvec, affd);
4ef33685
CH
1101 if (rc == 0)
1102 return nvec;
1103
4ef33685
CH
1104 if (rc < 0)
1105 return rc;
1106 if (rc < minvec)
1107 return -ENOSPC;
1108
1109 nvec = rc;
1110 }
302a2523 1111}
302a2523
AG
1112
1113/**
1114 * pci_enable_msix_range - configure device's MSI-X capability structure
1115 * @dev: pointer to the pci_dev data structure of MSI-X device function
1116 * @entries: pointer to an array of MSI-X entries
1117 * @minvec: minimum number of MSI-X irqs requested
1118 * @maxvec: maximum number of MSI-X irqs requested
1119 *
1120 * Setup the MSI-X capability structure of device function with a maximum
1121 * possible number of interrupts in the range between @minvec and @maxvec
1122 * upon its software driver call to request for MSI-X mode enabled on its
1123 * hardware device function. It returns a negative errno if an error occurs.
1124 * If it succeeds, it returns the actual number of interrupts allocated and
1125 * indicates the successful configuration of MSI-X capability structure
1126 * with new allocated MSI-X interrupts.
1127 **/
1128int pci_enable_msix_range(struct pci_dev *dev, struct msix_entry *entries,
4ef33685 1129 int minvec, int maxvec)
302a2523 1130{
61e1c590 1131 return __pci_enable_msix_range(dev, entries, minvec, maxvec, NULL);
302a2523
AG
1132}
1133EXPORT_SYMBOL(pci_enable_msix_range);
3878eaef 1134
aff17164 1135/**
402723ad 1136 * pci_alloc_irq_vectors_affinity - allocate multiple IRQs for a device
aff17164
CH
1137 * @dev: PCI device to operate on
1138 * @min_vecs: minimum number of vectors required (must be >= 1)
1139 * @max_vecs: maximum (desired) number of vectors
1140 * @flags: flags or quirks for the allocation
402723ad 1141 * @affd: optional description of the affinity requirements
aff17164
CH
1142 *
1143 * Allocate up to @max_vecs interrupt vectors for @dev, using MSI-X or MSI
1144 * vectors if available, and fall back to a single legacy vector
1145 * if neither is available. Return the number of vectors allocated,
1146 * (which might be smaller than @max_vecs) if successful, or a negative
1147 * error code on error. If less than @min_vecs interrupt vectors are
1148 * available for @dev the function will fail with -ENOSPC.
1149 *
1150 * To get the Linux IRQ number used for a vector that can be passed to
1151 * request_irq() use the pci_irq_vector() helper.
1152 */
402723ad
CH
1153int pci_alloc_irq_vectors_affinity(struct pci_dev *dev, unsigned int min_vecs,
1154 unsigned int max_vecs, unsigned int flags,
1155 const struct irq_affinity *affd)
aff17164 1156{
61e1c590 1157 static const struct irq_affinity msi_default_affd;
aff17164
CH
1158 int vecs = -ENOSPC;
1159
402723ad
CH
1160 if (flags & PCI_IRQ_AFFINITY) {
1161 if (!affd)
1162 affd = &msi_default_affd;
1163 } else {
1164 if (WARN_ON(affd))
1165 affd = NULL;
1166 }
61e1c590 1167
4fe0d154 1168 if (flags & PCI_IRQ_MSIX) {
4ef33685 1169 vecs = __pci_enable_msix_range(dev, NULL, min_vecs, max_vecs,
61e1c590 1170 affd);
aff17164
CH
1171 if (vecs > 0)
1172 return vecs;
1173 }
1174
4fe0d154 1175 if (flags & PCI_IRQ_MSI) {
61e1c590 1176 vecs = __pci_enable_msi_range(dev, min_vecs, max_vecs, affd);
aff17164
CH
1177 if (vecs > 0)
1178 return vecs;
1179 }
1180
1181 /* use legacy irq if allowed */
862290f9
CH
1182 if (flags & PCI_IRQ_LEGACY) {
1183 if (min_vecs == 1 && dev->irq) {
1184 pci_intx(dev, 1);
1185 return 1;
1186 }
5d0bdf28
CH
1187 }
1188
aff17164
CH
1189 return vecs;
1190}
402723ad 1191EXPORT_SYMBOL(pci_alloc_irq_vectors_affinity);
aff17164
CH
1192
1193/**
1194 * pci_free_irq_vectors - free previously allocated IRQs for a device
1195 * @dev: PCI device to operate on
1196 *
1197 * Undoes the allocations and enabling in pci_alloc_irq_vectors().
1198 */
1199void pci_free_irq_vectors(struct pci_dev *dev)
1200{
1201 pci_disable_msix(dev);
1202 pci_disable_msi(dev);
1203}
1204EXPORT_SYMBOL(pci_free_irq_vectors);
1205
1206/**
1207 * pci_irq_vector - return Linux IRQ number of a device vector
1208 * @dev: PCI device to operate on
1209 * @nr: device-relative interrupt vector index (0-based).
1210 */
1211int pci_irq_vector(struct pci_dev *dev, unsigned int nr)
1212{
1213 if (dev->msix_enabled) {
1214 struct msi_desc *entry;
1215 int i = 0;
1216
1217 for_each_pci_msi_entry(entry, dev) {
1218 if (i == nr)
1219 return entry->irq;
1220 i++;
1221 }
1222 WARN_ON_ONCE(1);
1223 return -EINVAL;
1224 }
1225
1226 if (dev->msi_enabled) {
1227 struct msi_desc *entry = first_pci_msi_entry(dev);
1228
1229 if (WARN_ON_ONCE(nr >= entry->nvec_used))
1230 return -EINVAL;
1231 } else {
1232 if (WARN_ON_ONCE(nr > 0))
1233 return -EINVAL;
1234 }
1235
1236 return dev->irq + nr;
1237}
1238EXPORT_SYMBOL(pci_irq_vector);
1239
ee8d41e5
TG
1240/**
1241 * pci_irq_get_affinity - return the affinity of a particular msi vector
1242 * @dev: PCI device to operate on
1243 * @nr: device-relative interrupt vector index (0-based).
1244 */
1245const struct cpumask *pci_irq_get_affinity(struct pci_dev *dev, int nr)
1246{
1247 if (dev->msix_enabled) {
1248 struct msi_desc *entry;
1249 int i = 0;
1250
1251 for_each_pci_msi_entry(entry, dev) {
1252 if (i == nr)
1253 return entry->affinity;
1254 i++;
1255 }
1256 WARN_ON_ONCE(1);
1257 return NULL;
1258 } else if (dev->msi_enabled) {
1259 struct msi_desc *entry = first_pci_msi_entry(dev);
1260
d1d111e0
JB
1261 if (WARN_ON_ONCE(!entry || !entry->affinity ||
1262 nr >= entry->nvec_used))
ee8d41e5
TG
1263 return NULL;
1264
1265 return &entry->affinity[nr];
1266 } else {
1267 return cpu_possible_mask;
1268 }
1269}
1270EXPORT_SYMBOL(pci_irq_get_affinity);
1271
27ddb689
SL
1272/**
1273 * pci_irq_get_node - return the numa node of a particular msi vector
1274 * @pdev: PCI device to operate on
1275 * @vec: device-relative interrupt vector index (0-based).
1276 */
1277int pci_irq_get_node(struct pci_dev *pdev, int vec)
1278{
1279 const struct cpumask *mask;
1280
1281 mask = pci_irq_get_affinity(pdev, vec);
1282 if (mask)
1283 return local_memory_node(cpu_to_node(cpumask_first(mask)));
1284 return dev_to_node(&pdev->dev);
1285}
1286EXPORT_SYMBOL(pci_irq_get_node);
1287
25a98bd4
JL
1288struct pci_dev *msi_desc_to_pci_dev(struct msi_desc *desc)
1289{
1290 return to_pci_dev(desc->dev);
1291}
a4289dc2 1292EXPORT_SYMBOL(msi_desc_to_pci_dev);
25a98bd4 1293
c179c9b9
JL
1294void *msi_desc_to_pci_sysdata(struct msi_desc *desc)
1295{
1296 struct pci_dev *dev = msi_desc_to_pci_dev(desc);
1297
1298 return dev->bus->sysdata;
1299}
1300EXPORT_SYMBOL_GPL(msi_desc_to_pci_sysdata);
1301
3878eaef
JL
1302#ifdef CONFIG_PCI_MSI_IRQ_DOMAIN
1303/**
1304 * pci_msi_domain_write_msg - Helper to write MSI message to PCI config space
1305 * @irq_data: Pointer to interrupt data of the MSI interrupt
1306 * @msg: Pointer to the message
1307 */
1308void pci_msi_domain_write_msg(struct irq_data *irq_data, struct msi_msg *msg)
1309{
507a883e 1310 struct msi_desc *desc = irq_data_get_msi_desc(irq_data);
3878eaef
JL
1311
1312 /*
1313 * For MSI-X desc->irq is always equal to irq_data->irq. For
1314 * MSI only the first interrupt of MULTI MSI passes the test.
1315 */
1316 if (desc->irq == irq_data->irq)
1317 __pci_write_msi_msg(desc, msg);
1318}
1319
1320/**
1321 * pci_msi_domain_calc_hwirq - Generate a unique ID for an MSI source
1322 * @dev: Pointer to the PCI device
1323 * @desc: Pointer to the msi descriptor
1324 *
1325 * The ID number is only used within the irqdomain.
1326 */
1327irq_hw_number_t pci_msi_domain_calc_hwirq(struct pci_dev *dev,
1328 struct msi_desc *desc)
1329{
1330 return (irq_hw_number_t)desc->msi_attrib.entry_nr |
1331 PCI_DEVID(dev->bus->number, dev->devfn) << 11 |
1332 (pci_domain_nr(dev->bus) & 0xFFFFFFFF) << 27;
1333}
1334
1335static inline bool pci_msi_desc_is_multi_msi(struct msi_desc *desc)
1336{
1337 return !desc->msi_attrib.is_msix && desc->nvec_used > 1;
1338}
1339
1340/**
1341 * pci_msi_domain_check_cap - Verify that @domain supports the capabilities for @dev
1342 * @domain: The interrupt domain to check
1343 * @info: The domain info for verification
1344 * @dev: The device to check
1345 *
1346 * Returns:
1347 * 0 if the functionality is supported
1348 * 1 if Multi MSI is requested, but the domain does not support it
1349 * -ENOTSUPP otherwise
1350 */
1351int pci_msi_domain_check_cap(struct irq_domain *domain,
1352 struct msi_domain_info *info, struct device *dev)
1353{
1354 struct msi_desc *desc = first_pci_msi_entry(to_pci_dev(dev));
1355
4fe03955 1356 /* Special handling to support __pci_enable_msi_range() */
3878eaef
JL
1357 if (pci_msi_desc_is_multi_msi(desc) &&
1358 !(info->flags & MSI_FLAG_MULTI_PCI_MSI))
1359 return 1;
1360 else if (desc->msi_attrib.is_msix && !(info->flags & MSI_FLAG_PCI_MSIX))
1361 return -ENOTSUPP;
1362
1363 return 0;
1364}
1365
1366static int pci_msi_domain_handle_error(struct irq_domain *domain,
1367 struct msi_desc *desc, int error)
1368{
4fe03955 1369 /* Special handling to support __pci_enable_msi_range() */
3878eaef
JL
1370 if (pci_msi_desc_is_multi_msi(desc) && error == -ENOSPC)
1371 return 1;
1372
1373 return error;
1374}
1375
1376#ifdef GENERIC_MSI_DOMAIN_OPS
1377static void pci_msi_domain_set_desc(msi_alloc_info_t *arg,
1378 struct msi_desc *desc)
1379{
1380 arg->desc = desc;
1381 arg->hwirq = pci_msi_domain_calc_hwirq(msi_desc_to_pci_dev(desc),
1382 desc);
1383}
1384#else
1385#define pci_msi_domain_set_desc NULL
1386#endif
1387
1388static struct msi_domain_ops pci_msi_domain_ops_default = {
1389 .set_desc = pci_msi_domain_set_desc,
1390 .msi_check = pci_msi_domain_check_cap,
1391 .handle_error = pci_msi_domain_handle_error,
1392};
1393
1394static void pci_msi_domain_update_dom_ops(struct msi_domain_info *info)
1395{
1396 struct msi_domain_ops *ops = info->ops;
1397
1398 if (ops == NULL) {
1399 info->ops = &pci_msi_domain_ops_default;
1400 } else {
1401 if (ops->set_desc == NULL)
1402 ops->set_desc = pci_msi_domain_set_desc;
1403 if (ops->msi_check == NULL)
1404 ops->msi_check = pci_msi_domain_check_cap;
1405 if (ops->handle_error == NULL)
1406 ops->handle_error = pci_msi_domain_handle_error;
1407 }
1408}
1409
1410static void pci_msi_domain_update_chip_ops(struct msi_domain_info *info)
1411{
1412 struct irq_chip *chip = info->chip;
1413
1414 BUG_ON(!chip);
1415 if (!chip->irq_write_msi_msg)
1416 chip->irq_write_msi_msg = pci_msi_domain_write_msg;
0701c53e
MZ
1417 if (!chip->irq_mask)
1418 chip->irq_mask = pci_msi_mask_irq;
1419 if (!chip->irq_unmask)
1420 chip->irq_unmask = pci_msi_unmask_irq;
3878eaef
JL
1421}
1422
1423/**
be5436c8
MZ
1424 * pci_msi_create_irq_domain - Create a MSI interrupt domain
1425 * @fwnode: Optional fwnode of the interrupt controller
3878eaef
JL
1426 * @info: MSI domain info
1427 * @parent: Parent irq domain
1428 *
1429 * Updates the domain and chip ops and creates a MSI interrupt domain.
1430 *
1431 * Returns:
1432 * A domain pointer or NULL in case of failure.
1433 */
be5436c8 1434struct irq_domain *pci_msi_create_irq_domain(struct fwnode_handle *fwnode,
3878eaef
JL
1435 struct msi_domain_info *info,
1436 struct irq_domain *parent)
1437{
0380839d
MZ
1438 struct irq_domain *domain;
1439
6988e0e0
MZ
1440 if (WARN_ON(info->flags & MSI_FLAG_LEVEL_CAPABLE))
1441 info->flags &= ~MSI_FLAG_LEVEL_CAPABLE;
1442
3878eaef
JL
1443 if (info->flags & MSI_FLAG_USE_DEF_DOM_OPS)
1444 pci_msi_domain_update_dom_ops(info);
1445 if (info->flags & MSI_FLAG_USE_DEF_CHIP_OPS)
1446 pci_msi_domain_update_chip_ops(info);
1447
f3b0946d 1448 info->flags |= MSI_FLAG_ACTIVATE_EARLY;
25e960ef
TG
1449 if (IS_ENABLED(CONFIG_GENERIC_IRQ_RESERVATION_MODE))
1450 info->flags |= MSI_FLAG_MUST_REACTIVATE;
f3b0946d 1451
923aa4c3
HK
1452 /* PCI-MSI is oneshot-safe */
1453 info->chip->flags |= IRQCHIP_ONESHOT_SAFE;
1454
be5436c8 1455 domain = msi_create_irq_domain(fwnode, info, parent);
0380839d
MZ
1456 if (!domain)
1457 return NULL;
1458
96f0d93a 1459 irq_domain_update_bus_token(domain, DOMAIN_BUS_PCI_MSI);
0380839d 1460 return domain;
3878eaef 1461}
a4289dc2 1462EXPORT_SYMBOL_GPL(pci_msi_create_irq_domain);
3878eaef 1463
235b2c77
RM
1464/*
1465 * Users of the generic MSI infrastructure expect a device to have a single ID,
1466 * so with DMA aliases we have to pick the least-worst compromise. Devices with
1467 * DMA phantom functions tend to still emit MSIs from the real function number,
1468 * so we ignore those and only consider topological aliases where either the
1469 * alias device or RID appears on a different bus number. We also make the
1470 * reasonable assumption that bridges are walked in an upstream direction (so
1471 * the last one seen wins), and the much braver assumption that the most likely
1472 * case is that of PCI->PCIe so we should always use the alias RID. This echoes
1473 * the logic from intel_irq_remapping's set_msi_sid(), which presumably works
1474 * well enough in practice; in the face of the horrible PCIe<->PCI-X conditions
1475 * for taking ownership all we can really do is close our eyes and hope...
1476 */
b6eec9b7
DD
1477static int get_msi_id_cb(struct pci_dev *pdev, u16 alias, void *data)
1478{
1479 u32 *pa = data;
235b2c77
RM
1480 u8 bus = PCI_BUS_NUM(*pa);
1481
1482 if (pdev->bus->number != bus || PCI_BUS_NUM(alias) != bus)
1483 *pa = alias;
b6eec9b7 1484
b6eec9b7
DD
1485 return 0;
1486}
235b2c77 1487
b6eec9b7
DD
1488/**
1489 * pci_msi_domain_get_msi_rid - Get the MSI requester id (RID)
1490 * @domain: The interrupt domain
1491 * @pdev: The PCI device.
1492 *
1493 * The RID for a device is formed from the alias, with a firmware
1494 * supplied mapping applied
1495 *
1496 * Returns: The RID.
1497 */
1498u32 pci_msi_domain_get_msi_rid(struct irq_domain *domain, struct pci_dev *pdev)
1499{
1500 struct device_node *of_node;
235b2c77 1501 u32 rid = PCI_DEVID(pdev->bus->number, pdev->devfn);
b6eec9b7
DD
1502
1503 pci_for_each_dma_alias(pdev, get_msi_id_cb, &rid);
1504
1505 of_node = irq_domain_get_of_node(domain);
be2021ba
TN
1506 rid = of_node ? of_msi_map_rid(&pdev->dev, of_node, rid) :
1507 iort_msi_map_rid(&pdev->dev, rid);
b6eec9b7
DD
1508
1509 return rid;
1510}
54fa97ee
MZ
1511
1512/**
1513 * pci_msi_get_device_domain - Get the MSI domain for a given PCI device
1514 * @pdev: The PCI device
1515 *
1516 * Use the firmware data to find a device-specific MSI domain
235b2c77 1517 * (i.e. not one that is set as a default).
54fa97ee 1518 *
235b2c77 1519 * Returns: The corresponding MSI domain or NULL if none has been found.
54fa97ee
MZ
1520 */
1521struct irq_domain *pci_msi_get_device_domain(struct pci_dev *pdev)
1522{
be2021ba 1523 struct irq_domain *dom;
235b2c77 1524 u32 rid = PCI_DEVID(pdev->bus->number, pdev->devfn);
54fa97ee
MZ
1525
1526 pci_for_each_dma_alias(pdev, get_msi_id_cb, &rid);
be2021ba
TN
1527 dom = of_msi_map_get_device_domain(&pdev->dev, rid);
1528 if (!dom)
1529 dom = iort_get_device_domain(&pdev->dev, rid);
1530 return dom;
54fa97ee 1531}
3878eaef 1532#endif /* CONFIG_PCI_MSI_IRQ_DOMAIN */
This page took 1.321259 seconds and 4 git commands to generate.