]> git.proxmox.com Git - mirror_ubuntu-bionic-kernel.git/blame - drivers/pci/msi.c
Merge remote-tracking branches 'asoc/topic/sgtl5000', 'asoc/topic/simple', 'asoc...
[mirror_ubuntu-bionic-kernel.git] / drivers / pci / msi.c
CommitLineData
1da177e4
LT
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)
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>
5a0e3ad6 22#include <linux/slab.h>
3878eaef 23#include <linux/irqdomain.h>
b6eec9b7 24#include <linux/of_irq.h>
1da177e4
LT
25
26#include "pci.h"
1da177e4 27
1da177e4 28static int pci_msi_enable = 1;
38737d82 29int pci_msi_ignore_mask;
1da177e4 30
527eee29
BH
31#define msix_table_size(flags) ((flags & PCI_MSIX_FLAGS_QSIZE) + 1)
32
8e047ada
JL
33#ifdef CONFIG_PCI_MSI_IRQ_DOMAIN
34static struct irq_domain *pci_msi_default_domain;
35static DEFINE_MUTEX(pci_msi_domain_lock);
36
37struct irq_domain * __weak arch_get_pci_msi_domain(struct pci_dev *dev)
38{
39 return pci_msi_default_domain;
40}
41
020c3126
MZ
42static struct irq_domain *pci_msi_get_domain(struct pci_dev *dev)
43{
d8a1cb75 44 struct irq_domain *domain;
020c3126 45
d8a1cb75
MZ
46 domain = dev_get_msi_domain(&dev->dev);
47 if (domain)
48 return domain;
020c3126 49
d8a1cb75 50 return arch_get_pci_msi_domain(dev);
020c3126
MZ
51}
52
8e047ada
JL
53static int pci_msi_setup_msi_irqs(struct pci_dev *dev, int nvec, int type)
54{
55 struct irq_domain *domain;
56
020c3126 57 domain = pci_msi_get_domain(dev);
3845d295 58 if (domain && irq_domain_is_hierarchy(domain))
8e047ada
JL
59 return pci_msi_domain_alloc_irqs(domain, dev, nvec, type);
60
61 return arch_setup_msi_irqs(dev, nvec, type);
62}
63
64static void pci_msi_teardown_msi_irqs(struct pci_dev *dev)
65{
66 struct irq_domain *domain;
67
020c3126 68 domain = pci_msi_get_domain(dev);
3845d295 69 if (domain && irq_domain_is_hierarchy(domain))
8e047ada
JL
70 pci_msi_domain_free_irqs(domain, dev);
71 else
72 arch_teardown_msi_irqs(dev);
73}
74#else
75#define pci_msi_setup_msi_irqs arch_setup_msi_irqs
76#define pci_msi_teardown_msi_irqs arch_teardown_msi_irqs
77#endif
527eee29 78
6a9e7f20
AB
79/* Arch hooks */
80
4287d824
TP
81int __weak arch_setup_msi_irq(struct pci_dev *dev, struct msi_desc *desc)
82{
2291ec09 83 struct msi_controller *chip = dev->bus->msi;
0cbdcfcf
TR
84 int err;
85
86 if (!chip || !chip->setup_irq)
87 return -EINVAL;
88
89 err = chip->setup_irq(chip, dev, desc);
90 if (err < 0)
91 return err;
92
93 irq_set_chip_data(desc->irq, chip);
94
95 return 0;
4287d824
TP
96}
97
98void __weak arch_teardown_msi_irq(unsigned int irq)
6a9e7f20 99{
c2791b80 100 struct msi_controller *chip = irq_get_chip_data(irq);
0cbdcfcf
TR
101
102 if (!chip || !chip->teardown_irq)
103 return;
104
105 chip->teardown_irq(chip, irq);
6a9e7f20
AB
106}
107
4287d824 108int __weak arch_setup_msi_irqs(struct pci_dev *dev, int nvec, int type)
6a9e7f20 109{
339e5b44 110 struct msi_controller *chip = dev->bus->msi;
6a9e7f20
AB
111 struct msi_desc *entry;
112 int ret;
113
339e5b44
LS
114 if (chip && chip->setup_irqs)
115 return chip->setup_irqs(chip, dev, nvec, type);
1c8d7b0a
MW
116 /*
117 * If an architecture wants to support multiple MSI, it needs to
118 * override arch_setup_msi_irqs()
119 */
120 if (type == PCI_CAP_ID_MSI && nvec > 1)
121 return 1;
122
5004e98a 123 for_each_pci_msi_entry(entry, dev) {
6a9e7f20 124 ret = arch_setup_msi_irq(dev, entry);
b5fbf533 125 if (ret < 0)
6a9e7f20 126 return ret;
b5fbf533
ME
127 if (ret > 0)
128 return -ENOSPC;
6a9e7f20
AB
129 }
130
131 return 0;
132}
1525bf0d 133
4287d824
TP
134/*
135 * We have a default implementation available as a separate non-weak
136 * function, as it is used by the Xen x86 PCI code
137 */
1525bf0d 138void default_teardown_msi_irqs(struct pci_dev *dev)
6a9e7f20 139{
63a7b17e 140 int i;
6a9e7f20
AB
141 struct msi_desc *entry;
142
5004e98a 143 for_each_pci_msi_entry(entry, dev)
63a7b17e
JL
144 if (entry->irq)
145 for (i = 0; i < entry->nvec_used; i++)
146 arch_teardown_msi_irq(entry->irq + i);
6a9e7f20
AB
147}
148
4287d824
TP
149void __weak arch_teardown_msi_irqs(struct pci_dev *dev)
150{
151 return default_teardown_msi_irqs(dev);
152}
76ccc297 153
ac8344c4 154static void default_restore_msi_irq(struct pci_dev *dev, int irq)
76ccc297
KRW
155{
156 struct msi_desc *entry;
157
158 entry = NULL;
159 if (dev->msix_enabled) {
5004e98a 160 for_each_pci_msi_entry(entry, dev) {
76ccc297
KRW
161 if (irq == entry->irq)
162 break;
163 }
164 } else if (dev->msi_enabled) {
165 entry = irq_get_msi_desc(irq);
166 }
167
168 if (entry)
83a18912 169 __pci_write_msi_msg(entry, &entry->msg);
76ccc297 170}
4287d824 171
ac8344c4 172void __weak arch_restore_msi_irqs(struct pci_dev *dev)
4287d824 173{
ac8344c4 174 return default_restore_msi_irqs(dev);
4287d824 175}
76ccc297 176
bffac3c5
MW
177static inline __attribute_const__ u32 msi_mask(unsigned x)
178{
0b49ec37
MW
179 /* Don't shift by >= width of type */
180 if (x >= 5)
181 return 0xffffffff;
182 return (1 << (1 << x)) - 1;
bffac3c5
MW
183}
184
ce6fce42
MW
185/*
186 * PCI 2.3 does not specify mask bits for each MSI interrupt. Attempting to
187 * mask all MSI interrupts by clearing the MSI enable bit does not work
188 * reliably as devices without an INTx disable bit will then generate a
189 * level IRQ which will never be cleared.
ce6fce42 190 */
23ed8d57 191u32 __pci_msi_desc_mask_irq(struct msi_desc *desc, u32 mask, u32 flag)
1da177e4 192{
f2440d9a 193 u32 mask_bits = desc->masked;
1da177e4 194
38737d82 195 if (pci_msi_ignore_mask || !desc->msi_attrib.maskbit)
12abb8ba 196 return 0;
f2440d9a
MW
197
198 mask_bits &= ~mask;
199 mask_bits |= flag;
e39758e0
JL
200 pci_write_config_dword(msi_desc_to_pci_dev(desc), desc->mask_pos,
201 mask_bits);
12abb8ba
HS
202
203 return mask_bits;
204}
205
206static void msi_mask_irq(struct msi_desc *desc, u32 mask, u32 flag)
207{
23ed8d57 208 desc->masked = __pci_msi_desc_mask_irq(desc, mask, flag);
f2440d9a
MW
209}
210
5eb6d660
CH
211static void __iomem *pci_msix_desc_addr(struct msi_desc *desc)
212{
213 return desc->mask_base +
214 desc->msi_attrib.entry_nr * PCI_MSIX_ENTRY_SIZE;
215}
216
f2440d9a
MW
217/*
218 * This internal function does not flush PCI writes to the device.
219 * All users must ensure that they read from the device before either
220 * assuming that the device state is up to date, or returning out of this
221 * file. This saves a few milliseconds when initialising devices with lots
222 * of MSI-X interrupts.
223 */
23ed8d57 224u32 __pci_msix_desc_mask_irq(struct msi_desc *desc, u32 flag)
f2440d9a
MW
225{
226 u32 mask_bits = desc->masked;
38737d82
YW
227
228 if (pci_msi_ignore_mask)
229 return 0;
230
8d805286
SY
231 mask_bits &= ~PCI_MSIX_ENTRY_CTRL_MASKBIT;
232 if (flag)
233 mask_bits |= PCI_MSIX_ENTRY_CTRL_MASKBIT;
5eb6d660 234 writel(mask_bits, pci_msix_desc_addr(desc) + PCI_MSIX_ENTRY_VECTOR_CTRL);
12abb8ba
HS
235
236 return mask_bits;
237}
238
239static void msix_mask_irq(struct msi_desc *desc, u32 flag)
240{
23ed8d57 241 desc->masked = __pci_msix_desc_mask_irq(desc, flag);
f2440d9a 242}
24d27553 243
1c9db525 244static void msi_set_mask_bit(struct irq_data *data, u32 flag)
f2440d9a 245{
c391f262 246 struct msi_desc *desc = irq_data_get_msi_desc(data);
24d27553 247
f2440d9a
MW
248 if (desc->msi_attrib.is_msix) {
249 msix_mask_irq(desc, flag);
250 readl(desc->mask_base); /* Flush write to device */
251 } else {
a281b788 252 unsigned offset = data->irq - desc->irq;
1c8d7b0a 253 msi_mask_irq(desc, 1 << offset, flag << offset);
1da177e4 254 }
f2440d9a
MW
255}
256
23ed8d57
TG
257/**
258 * pci_msi_mask_irq - Generic irq chip callback to mask PCI/MSI interrupts
259 * @data: pointer to irqdata associated to that interrupt
260 */
261void pci_msi_mask_irq(struct irq_data *data)
f2440d9a 262{
1c9db525 263 msi_set_mask_bit(data, 1);
f2440d9a 264}
a4289dc2 265EXPORT_SYMBOL_GPL(pci_msi_mask_irq);
f2440d9a 266
23ed8d57
TG
267/**
268 * pci_msi_unmask_irq - Generic irq chip callback to unmask PCI/MSI interrupts
269 * @data: pointer to irqdata associated to that interrupt
270 */
271void pci_msi_unmask_irq(struct irq_data *data)
f2440d9a 272{
1c9db525 273 msi_set_mask_bit(data, 0);
1da177e4 274}
a4289dc2 275EXPORT_SYMBOL_GPL(pci_msi_unmask_irq);
1da177e4 276
ac8344c4
D
277void default_restore_msi_irqs(struct pci_dev *dev)
278{
279 struct msi_desc *entry;
280
5004e98a 281 for_each_pci_msi_entry(entry, dev)
ac8344c4 282 default_restore_msi_irq(dev, entry->irq);
ac8344c4
D
283}
284
891d4a48 285void __pci_read_msi_msg(struct msi_desc *entry, struct msi_msg *msg)
1da177e4 286{
e39758e0
JL
287 struct pci_dev *dev = msi_desc_to_pci_dev(entry);
288
289 BUG_ON(dev->current_state != PCI_D0);
30da5524
BH
290
291 if (entry->msi_attrib.is_msix) {
5eb6d660 292 void __iomem *base = pci_msix_desc_addr(entry);
30da5524
BH
293
294 msg->address_lo = readl(base + PCI_MSIX_ENTRY_LOWER_ADDR);
295 msg->address_hi = readl(base + PCI_MSIX_ENTRY_UPPER_ADDR);
296 msg->data = readl(base + PCI_MSIX_ENTRY_DATA);
297 } else {
f5322169 298 int pos = dev->msi_cap;
30da5524
BH
299 u16 data;
300
9925ad0c
BH
301 pci_read_config_dword(dev, pos + PCI_MSI_ADDRESS_LO,
302 &msg->address_lo);
30da5524 303 if (entry->msi_attrib.is_64) {
9925ad0c
BH
304 pci_read_config_dword(dev, pos + PCI_MSI_ADDRESS_HI,
305 &msg->address_hi);
2f221349 306 pci_read_config_word(dev, pos + PCI_MSI_DATA_64, &data);
30da5524
BH
307 } else {
308 msg->address_hi = 0;
2f221349 309 pci_read_config_word(dev, pos + PCI_MSI_DATA_32, &data);
30da5524
BH
310 }
311 msg->data = data;
312 }
313}
314
83a18912 315void __pci_write_msi_msg(struct msi_desc *entry, struct msi_msg *msg)
3145e941 316{
e39758e0
JL
317 struct pci_dev *dev = msi_desc_to_pci_dev(entry);
318
319 if (dev->current_state != PCI_D0) {
fcd097f3
BH
320 /* Don't touch the hardware now */
321 } else if (entry->msi_attrib.is_msix) {
5eb6d660 322 void __iomem *base = pci_msix_desc_addr(entry);
24d27553 323
2c21fd4b
HS
324 writel(msg->address_lo, base + PCI_MSIX_ENTRY_LOWER_ADDR);
325 writel(msg->address_hi, base + PCI_MSIX_ENTRY_UPPER_ADDR);
326 writel(msg->data, base + PCI_MSIX_ENTRY_DATA);
24d27553 327 } else {
f5322169 328 int pos = dev->msi_cap;
1c8d7b0a
MW
329 u16 msgctl;
330
f84ecd28 331 pci_read_config_word(dev, pos + PCI_MSI_FLAGS, &msgctl);
1c8d7b0a
MW
332 msgctl &= ~PCI_MSI_FLAGS_QSIZE;
333 msgctl |= entry->msi_attrib.multiple << 4;
f84ecd28 334 pci_write_config_word(dev, pos + PCI_MSI_FLAGS, msgctl);
0366f8f7 335
9925ad0c
BH
336 pci_write_config_dword(dev, pos + PCI_MSI_ADDRESS_LO,
337 msg->address_lo);
0366f8f7 338 if (entry->msi_attrib.is_64) {
9925ad0c
BH
339 pci_write_config_dword(dev, pos + PCI_MSI_ADDRESS_HI,
340 msg->address_hi);
2f221349
BH
341 pci_write_config_word(dev, pos + PCI_MSI_DATA_64,
342 msg->data);
0366f8f7 343 } else {
2f221349
BH
344 pci_write_config_word(dev, pos + PCI_MSI_DATA_32,
345 msg->data);
0366f8f7 346 }
1da177e4 347 }
392ee1e6 348 entry->msg = *msg;
1da177e4 349}
0366f8f7 350
83a18912 351void pci_write_msi_msg(unsigned int irq, struct msi_msg *msg)
3145e941 352{
dced35ae 353 struct msi_desc *entry = irq_get_msi_desc(irq);
3145e941 354
83a18912 355 __pci_write_msi_msg(entry, msg);
3145e941 356}
83a18912 357EXPORT_SYMBOL_GPL(pci_write_msi_msg);
3145e941 358
f56e4481
HS
359static void free_msi_irqs(struct pci_dev *dev)
360{
5004e98a 361 struct list_head *msi_list = dev_to_msi_list(&dev->dev);
f56e4481 362 struct msi_desc *entry, *tmp;
1c51b50c
GKH
363 struct attribute **msi_attrs;
364 struct device_attribute *dev_attr;
63a7b17e 365 int i, count = 0;
f56e4481 366
5004e98a 367 for_each_pci_msi_entry(entry, dev)
63a7b17e
JL
368 if (entry->irq)
369 for (i = 0; i < entry->nvec_used; i++)
370 BUG_ON(irq_has_action(entry->irq + i));
f56e4481 371
8e047ada 372 pci_msi_teardown_msi_irqs(dev);
f56e4481 373
5004e98a 374 list_for_each_entry_safe(entry, tmp, msi_list, list) {
f56e4481 375 if (entry->msi_attrib.is_msix) {
5004e98a 376 if (list_is_last(&entry->list, msi_list))
f56e4481
HS
377 iounmap(entry->mask_base);
378 }
424eb391 379
f56e4481
HS
380 list_del(&entry->list);
381 kfree(entry);
382 }
1c51b50c
GKH
383
384 if (dev->msi_irq_groups) {
385 sysfs_remove_groups(&dev->dev.kobj, dev->msi_irq_groups);
386 msi_attrs = dev->msi_irq_groups[0]->attrs;
b701c0b1 387 while (msi_attrs[count]) {
1c51b50c
GKH
388 dev_attr = container_of(msi_attrs[count],
389 struct device_attribute, attr);
390 kfree(dev_attr->attr.name);
391 kfree(dev_attr);
392 ++count;
393 }
394 kfree(msi_attrs);
395 kfree(dev->msi_irq_groups[0]);
396 kfree(dev->msi_irq_groups);
397 dev->msi_irq_groups = NULL;
398 }
f56e4481 399}
c54c1879 400
ba698ad4
DM
401static void pci_intx_for_msi(struct pci_dev *dev, int enable)
402{
403 if (!(dev->dev_flags & PCI_DEV_FLAGS_MSI_INTX_DISABLE_BUG))
404 pci_intx(dev, enable);
405}
406
8fed4b65 407static void __pci_restore_msi_state(struct pci_dev *dev)
41017f0c 408{
41017f0c 409 u16 control;
392ee1e6 410 struct msi_desc *entry;
41017f0c 411
b1cbf4e4
EB
412 if (!dev->msi_enabled)
413 return;
414
dced35ae 415 entry = irq_get_msi_desc(dev->irq);
41017f0c 416
ba698ad4 417 pci_intx_for_msi(dev, 0);
61b64abd 418 pci_msi_set_enable(dev, 0);
ac8344c4 419 arch_restore_msi_irqs(dev);
392ee1e6 420
f5322169 421 pci_read_config_word(dev, dev->msi_cap + PCI_MSI_FLAGS, &control);
31ea5d4d
YW
422 msi_mask_irq(entry, msi_mask(entry->msi_attrib.multi_cap),
423 entry->masked);
abad2ec9 424 control &= ~PCI_MSI_FLAGS_QSIZE;
1c8d7b0a 425 control |= (entry->msi_attrib.multiple << 4) | PCI_MSI_FLAGS_ENABLE;
f5322169 426 pci_write_config_word(dev, dev->msi_cap + PCI_MSI_FLAGS, control);
8fed4b65
ME
427}
428
429static void __pci_restore_msix_state(struct pci_dev *dev)
41017f0c 430{
41017f0c 431 struct msi_desc *entry;
41017f0c 432
ded86d8d
EB
433 if (!dev->msix_enabled)
434 return;
5004e98a 435 BUG_ON(list_empty(dev_to_msi_list(&dev->dev)));
ded86d8d 436
41017f0c 437 /* route the table */
ba698ad4 438 pci_intx_for_msi(dev, 0);
61b64abd 439 pci_msix_clear_and_set_ctrl(dev, 0,
66f0d0c4 440 PCI_MSIX_FLAGS_ENABLE | PCI_MSIX_FLAGS_MASKALL);
41017f0c 441
ac8344c4 442 arch_restore_msi_irqs(dev);
5004e98a 443 for_each_pci_msi_entry(entry, dev)
f2440d9a 444 msix_mask_irq(entry, entry->masked);
41017f0c 445
61b64abd 446 pci_msix_clear_and_set_ctrl(dev, PCI_MSIX_FLAGS_MASKALL, 0);
41017f0c 447}
8fed4b65
ME
448
449void pci_restore_msi_state(struct pci_dev *dev)
450{
451 __pci_restore_msi_state(dev);
452 __pci_restore_msix_state(dev);
453}
94688cf2 454EXPORT_SYMBOL_GPL(pci_restore_msi_state);
41017f0c 455
1c51b50c 456static ssize_t msi_mode_show(struct device *dev, struct device_attribute *attr,
da8d1c8b
NH
457 char *buf)
458{
1c51b50c
GKH
459 struct msi_desc *entry;
460 unsigned long irq;
461 int retval;
da8d1c8b 462
1c51b50c
GKH
463 retval = kstrtoul(attr->attr.name, 10, &irq);
464 if (retval)
465 return retval;
da8d1c8b 466
e11ece5a
YW
467 entry = irq_get_msi_desc(irq);
468 if (entry)
469 return sprintf(buf, "%s\n",
470 entry->msi_attrib.is_msix ? "msix" : "msi");
471
1c51b50c 472 return -ENODEV;
da8d1c8b
NH
473}
474
da8d1c8b
NH
475static int populate_msi_sysfs(struct pci_dev *pdev)
476{
1c51b50c
GKH
477 struct attribute **msi_attrs;
478 struct attribute *msi_attr;
479 struct device_attribute *msi_dev_attr;
480 struct attribute_group *msi_irq_group;
481 const struct attribute_group **msi_irq_groups;
da8d1c8b 482 struct msi_desc *entry;
1c51b50c
GKH
483 int ret = -ENOMEM;
484 int num_msi = 0;
da8d1c8b 485 int count = 0;
a8676066 486 int i;
da8d1c8b 487
1c51b50c 488 /* Determine how many msi entries we have */
5004e98a 489 for_each_pci_msi_entry(entry, pdev)
a8676066 490 num_msi += entry->nvec_used;
1c51b50c
GKH
491 if (!num_msi)
492 return 0;
da8d1c8b 493
1c51b50c
GKH
494 /* Dynamically create the MSI attributes for the PCI device */
495 msi_attrs = kzalloc(sizeof(void *) * (num_msi + 1), GFP_KERNEL);
496 if (!msi_attrs)
497 return -ENOMEM;
5004e98a 498 for_each_pci_msi_entry(entry, pdev) {
a8676066
RB
499 for (i = 0; i < entry->nvec_used; i++) {
500 msi_dev_attr = kzalloc(sizeof(*msi_dev_attr), GFP_KERNEL);
501 if (!msi_dev_attr)
502 goto error_attrs;
503 msi_attrs[count] = &msi_dev_attr->attr;
504
505 sysfs_attr_init(&msi_dev_attr->attr);
506 msi_dev_attr->attr.name = kasprintf(GFP_KERNEL, "%d",
507 entry->irq + i);
508 if (!msi_dev_attr->attr.name)
509 goto error_attrs;
510 msi_dev_attr->attr.mode = S_IRUGO;
511 msi_dev_attr->show = msi_mode_show;
512 ++count;
513 }
da8d1c8b
NH
514 }
515
1c51b50c
GKH
516 msi_irq_group = kzalloc(sizeof(*msi_irq_group), GFP_KERNEL);
517 if (!msi_irq_group)
518 goto error_attrs;
519 msi_irq_group->name = "msi_irqs";
520 msi_irq_group->attrs = msi_attrs;
521
522 msi_irq_groups = kzalloc(sizeof(void *) * 2, GFP_KERNEL);
523 if (!msi_irq_groups)
524 goto error_irq_group;
525 msi_irq_groups[0] = msi_irq_group;
526
527 ret = sysfs_create_groups(&pdev->dev.kobj, msi_irq_groups);
528 if (ret)
529 goto error_irq_groups;
530 pdev->msi_irq_groups = msi_irq_groups;
531
da8d1c8b
NH
532 return 0;
533
1c51b50c
GKH
534error_irq_groups:
535 kfree(msi_irq_groups);
536error_irq_group:
537 kfree(msi_irq_group);
538error_attrs:
539 count = 0;
540 msi_attr = msi_attrs[count];
541 while (msi_attr) {
542 msi_dev_attr = container_of(msi_attr, struct device_attribute, attr);
543 kfree(msi_attr->name);
544 kfree(msi_dev_attr);
545 ++count;
546 msi_attr = msi_attrs[count];
da8d1c8b 547 }
29237756 548 kfree(msi_attrs);
da8d1c8b
NH
549 return ret;
550}
551
63a7b17e 552static struct msi_desc *msi_setup_entry(struct pci_dev *dev, int nvec)
d873b4d4
YW
553{
554 u16 control;
555 struct msi_desc *entry;
556
557 /* MSI Entry Initialization */
aa48b6f7 558 entry = alloc_msi_entry(&dev->dev);
d873b4d4
YW
559 if (!entry)
560 return NULL;
561
562 pci_read_config_word(dev, dev->msi_cap + PCI_MSI_FLAGS, &control);
563
564 entry->msi_attrib.is_msix = 0;
565 entry->msi_attrib.is_64 = !!(control & PCI_MSI_FLAGS_64BIT);
566 entry->msi_attrib.entry_nr = 0;
567 entry->msi_attrib.maskbit = !!(control & PCI_MSI_FLAGS_MASKBIT);
568 entry->msi_attrib.default_irq = dev->irq; /* Save IOAPIC IRQ */
d873b4d4 569 entry->msi_attrib.multi_cap = (control & PCI_MSI_FLAGS_QMASK) >> 1;
63a7b17e
JL
570 entry->msi_attrib.multiple = ilog2(__roundup_pow_of_two(nvec));
571 entry->nvec_used = nvec;
4ef33685 572 entry->affinity = dev->irq_affinity;
d873b4d4
YW
573
574 if (control & PCI_MSI_FLAGS_64BIT)
575 entry->mask_pos = dev->msi_cap + PCI_MSI_MASK_64;
576 else
577 entry->mask_pos = dev->msi_cap + PCI_MSI_MASK_32;
578
579 /* Save the initial mask status */
580 if (entry->msi_attrib.maskbit)
581 pci_read_config_dword(dev, entry->mask_pos, &entry->masked);
582
583 return entry;
584}
585
f144d149
BH
586static int msi_verify_entries(struct pci_dev *dev)
587{
588 struct msi_desc *entry;
589
5004e98a 590 for_each_pci_msi_entry(entry, dev) {
f144d149
BH
591 if (!dev->no_64bit_msi || !entry->msg.address_hi)
592 continue;
593 dev_err(&dev->dev, "Device has broken 64-bit MSI but arch"
594 " tried to assign one above 4G\n");
595 return -EIO;
596 }
597 return 0;
598}
599
1da177e4
LT
600/**
601 * msi_capability_init - configure device's MSI capability structure
602 * @dev: pointer to the pci_dev data structure of MSI device function
1c8d7b0a 603 * @nvec: number of interrupts to allocate
1da177e4 604 *
1c8d7b0a
MW
605 * Setup the MSI capability structure of the device with the requested
606 * number of interrupts. A return value of zero indicates the successful
607 * setup of an entry with the new MSI irq. A negative return value indicates
608 * an error, and a positive return value indicates the number of interrupts
609 * which could have been allocated.
610 */
611static int msi_capability_init(struct pci_dev *dev, int nvec)
1da177e4
LT
612{
613 struct msi_desc *entry;
f465136d 614 int ret;
f2440d9a 615 unsigned mask;
1da177e4 616
61b64abd 617 pci_msi_set_enable(dev, 0); /* Disable MSI during set up */
110828c9 618
63a7b17e 619 entry = msi_setup_entry(dev, nvec);
f7feaca7
EB
620 if (!entry)
621 return -ENOMEM;
1ce03373 622
f2440d9a 623 /* All MSIs are unmasked by default, Mask them all */
31ea5d4d 624 mask = msi_mask(entry->msi_attrib.multi_cap);
f2440d9a
MW
625 msi_mask_irq(entry, mask, mask);
626
5004e98a 627 list_add_tail(&entry->list, dev_to_msi_list(&dev->dev));
9c831334 628
1da177e4 629 /* Configure MSI capability structure */
8e047ada 630 ret = pci_msi_setup_msi_irqs(dev, nvec, PCI_CAP_ID_MSI);
7fe3730d 631 if (ret) {
7ba1930d 632 msi_mask_irq(entry, mask, ~mask);
f56e4481 633 free_msi_irqs(dev);
7fe3730d 634 return ret;
fd58e55f 635 }
f7feaca7 636
f144d149
BH
637 ret = msi_verify_entries(dev);
638 if (ret) {
639 msi_mask_irq(entry, mask, ~mask);
640 free_msi_irqs(dev);
641 return ret;
642 }
643
da8d1c8b
NH
644 ret = populate_msi_sysfs(dev);
645 if (ret) {
646 msi_mask_irq(entry, mask, ~mask);
647 free_msi_irqs(dev);
648 return ret;
649 }
650
1da177e4 651 /* Set MSI enabled bits */
ba698ad4 652 pci_intx_for_msi(dev, 0);
61b64abd 653 pci_msi_set_enable(dev, 1);
b1cbf4e4 654 dev->msi_enabled = 1;
1da177e4 655
5f226991 656 pcibios_free_irq(dev);
7fe3730d 657 dev->irq = entry->irq;
1da177e4
LT
658 return 0;
659}
660
520fe9dc 661static void __iomem *msix_map_region(struct pci_dev *dev, unsigned nr_entries)
5a05a9d8 662{
4302e0fb 663 resource_size_t phys_addr;
5a05a9d8 664 u32 table_offset;
6a878e50 665 unsigned long flags;
5a05a9d8
HS
666 u8 bir;
667
909094c6
BH
668 pci_read_config_dword(dev, dev->msix_cap + PCI_MSIX_TABLE,
669 &table_offset);
4d18760c 670 bir = (u8)(table_offset & PCI_MSIX_TABLE_BIR);
6a878e50
YW
671 flags = pci_resource_flags(dev, bir);
672 if (!flags || (flags & IORESOURCE_UNSET))
673 return NULL;
674
4d18760c 675 table_offset &= PCI_MSIX_TABLE_OFFSET;
5a05a9d8
HS
676 phys_addr = pci_resource_start(dev, bir) + table_offset;
677
678 return ioremap_nocache(phys_addr, nr_entries * PCI_MSIX_ENTRY_SIZE);
679}
680
520fe9dc
GS
681static int msix_setup_entries(struct pci_dev *dev, void __iomem *base,
682 struct msix_entry *entries, int nvec)
d9d7070e 683{
4ef33685 684 const struct cpumask *mask = NULL;
d9d7070e 685 struct msi_desc *entry;
4ef33685 686 int cpu = -1, i;
d9d7070e
HS
687
688 for (i = 0; i < nvec; i++) {
4ef33685
CH
689 if (dev->irq_affinity) {
690 cpu = cpumask_next(cpu, dev->irq_affinity);
691 if (cpu >= nr_cpu_ids)
692 cpu = cpumask_first(dev->irq_affinity);
693 mask = cpumask_of(cpu);
694 }
695
aa48b6f7 696 entry = alloc_msi_entry(&dev->dev);
d9d7070e
HS
697 if (!entry) {
698 if (!i)
699 iounmap(base);
700 else
701 free_msi_irqs(dev);
702 /* No enough memory. Don't try again */
703 return -ENOMEM;
704 }
705
706 entry->msi_attrib.is_msix = 1;
707 entry->msi_attrib.is_64 = 1;
3ac020e0
CH
708 if (entries)
709 entry->msi_attrib.entry_nr = entries[i].entry;
710 else
711 entry->msi_attrib.entry_nr = i;
d9d7070e 712 entry->msi_attrib.default_irq = dev->irq;
d9d7070e 713 entry->mask_base = base;
63a7b17e 714 entry->nvec_used = 1;
4ef33685 715 entry->affinity = mask;
d9d7070e 716
5004e98a 717 list_add_tail(&entry->list, dev_to_msi_list(&dev->dev));
d9d7070e
HS
718 }
719
720 return 0;
721}
722
75cb3426 723static void msix_program_entries(struct pci_dev *dev,
520fe9dc 724 struct msix_entry *entries)
75cb3426
HS
725{
726 struct msi_desc *entry;
727 int i = 0;
728
5004e98a 729 for_each_pci_msi_entry(entry, dev) {
3ac020e0
CH
730 if (entries)
731 entries[i++].vector = entry->irq;
12eb21de
CH
732 entry->masked = readl(pci_msix_desc_addr(entry) +
733 PCI_MSIX_ENTRY_VECTOR_CTRL);
75cb3426 734 msix_mask_irq(entry, 1);
75cb3426
HS
735 }
736}
737
1da177e4
LT
738/**
739 * msix_capability_init - configure device's MSI-X capability
740 * @dev: pointer to the pci_dev data structure of MSI-X device function
8f7020d3
RD
741 * @entries: pointer to an array of struct msix_entry entries
742 * @nvec: number of @entries
1da177e4 743 *
eaae4b3a 744 * Setup the MSI-X capability structure of device function with a
1ce03373
EB
745 * single MSI-X irq. A return of zero indicates the successful setup of
746 * requested MSI-X entries with allocated irqs or non-zero for otherwise.
1da177e4
LT
747 **/
748static int msix_capability_init(struct pci_dev *dev,
749 struct msix_entry *entries, int nvec)
750{
520fe9dc 751 int ret;
5a05a9d8 752 u16 control;
1da177e4
LT
753 void __iomem *base;
754
f598282f 755 /* Ensure MSI-X is disabled while it is set up */
61b64abd 756 pci_msix_clear_and_set_ctrl(dev, PCI_MSIX_FLAGS_ENABLE, 0);
f598282f 757
66f0d0c4 758 pci_read_config_word(dev, dev->msix_cap + PCI_MSIX_FLAGS, &control);
1da177e4 759 /* Request & Map MSI-X table region */
527eee29 760 base = msix_map_region(dev, msix_table_size(control));
5a05a9d8 761 if (!base)
1da177e4
LT
762 return -ENOMEM;
763
520fe9dc 764 ret = msix_setup_entries(dev, base, entries, nvec);
d9d7070e
HS
765 if (ret)
766 return ret;
9c831334 767
8e047ada 768 ret = pci_msi_setup_msi_irqs(dev, nvec, PCI_CAP_ID_MSIX);
583871d4 769 if (ret)
2adc7907 770 goto out_avail;
9c831334 771
f144d149
BH
772 /* Check if all MSI entries honor device restrictions */
773 ret = msi_verify_entries(dev);
774 if (ret)
775 goto out_free;
776
f598282f
MW
777 /*
778 * Some devices require MSI-X to be enabled before we can touch the
779 * MSI-X registers. We need to mask all the vectors to prevent
780 * interrupts coming in before they're fully set up.
781 */
61b64abd 782 pci_msix_clear_and_set_ctrl(dev, 0,
66f0d0c4 783 PCI_MSIX_FLAGS_MASKALL | PCI_MSIX_FLAGS_ENABLE);
f598282f 784
75cb3426 785 msix_program_entries(dev, entries);
f598282f 786
da8d1c8b 787 ret = populate_msi_sysfs(dev);
2adc7907
AG
788 if (ret)
789 goto out_free;
da8d1c8b 790
f598282f 791 /* Set MSI-X enabled bits and unmask the function */
ba698ad4 792 pci_intx_for_msi(dev, 0);
b1cbf4e4 793 dev->msix_enabled = 1;
61b64abd 794 pci_msix_clear_and_set_ctrl(dev, PCI_MSIX_FLAGS_MASKALL, 0);
8d181018 795
5f226991 796 pcibios_free_irq(dev);
1da177e4 797 return 0;
583871d4 798
2adc7907 799out_avail:
583871d4
HS
800 if (ret < 0) {
801 /*
802 * If we had some success, report the number of irqs
803 * we succeeded in setting up.
804 */
d9d7070e 805 struct msi_desc *entry;
583871d4
HS
806 int avail = 0;
807
5004e98a 808 for_each_pci_msi_entry(entry, dev) {
583871d4
HS
809 if (entry->irq != 0)
810 avail++;
811 }
812 if (avail != 0)
813 ret = avail;
814 }
815
2adc7907 816out_free:
583871d4
HS
817 free_msi_irqs(dev);
818
819 return ret;
1da177e4
LT
820}
821
24334a12 822/**
a06cd74c 823 * pci_msi_supported - check whether MSI may be enabled on a device
24334a12 824 * @dev: pointer to the pci_dev data structure of MSI device function
c9953a73 825 * @nvec: how many MSIs have been requested ?
24334a12 826 *
f7625980 827 * Look at global flags, the device itself, and its parent buses
17bbc12a 828 * to determine if MSI/-X are supported for the device. If MSI/-X is
a06cd74c 829 * supported return 1, else return 0.
24334a12 830 **/
a06cd74c 831static int pci_msi_supported(struct pci_dev *dev, int nvec)
24334a12
BG
832{
833 struct pci_bus *bus;
834
0306ebfa 835 /* MSI must be globally enabled and supported by the device */
27e20603 836 if (!pci_msi_enable)
a06cd74c 837 return 0;
27e20603
AG
838
839 if (!dev || dev->no_msi || dev->current_state != PCI_D0)
a06cd74c 840 return 0;
24334a12 841
314e77b3
ME
842 /*
843 * You can't ask to have 0 or less MSIs configured.
844 * a) it's stupid ..
845 * b) the list manipulation code assumes nvec >= 1.
846 */
847 if (nvec < 1)
a06cd74c 848 return 0;
314e77b3 849
500559a9
HS
850 /*
851 * Any bridge which does NOT route MSI transactions from its
852 * secondary bus to its primary bus must set NO_MSI flag on
0306ebfa
BG
853 * the secondary pci_bus.
854 * We expect only arch-specific PCI host bus controller driver
855 * or quirks for specific PCI bridges to be setting NO_MSI.
856 */
24334a12
BG
857 for (bus = dev->bus; bus; bus = bus->parent)
858 if (bus->bus_flags & PCI_BUS_FLAGS_NO_MSI)
a06cd74c 859 return 0;
24334a12 860
a06cd74c 861 return 1;
24334a12
BG
862}
863
d1ac1d26
AG
864/**
865 * pci_msi_vec_count - Return the number of MSI vectors a device can send
866 * @dev: device to report about
867 *
868 * This function returns the number of MSI vectors a device requested via
869 * Multiple Message Capable register. It returns a negative errno if the
870 * device is not capable sending MSI interrupts. Otherwise, the call succeeds
871 * and returns a power of two, up to a maximum of 2^5 (32), according to the
872 * MSI specification.
873 **/
874int pci_msi_vec_count(struct pci_dev *dev)
875{
876 int ret;
877 u16 msgctl;
878
879 if (!dev->msi_cap)
880 return -EINVAL;
881
882 pci_read_config_word(dev, dev->msi_cap + PCI_MSI_FLAGS, &msgctl);
883 ret = 1 << ((msgctl & PCI_MSI_FLAGS_QMASK) >> 1);
884
885 return ret;
886}
887EXPORT_SYMBOL(pci_msi_vec_count);
888
f2440d9a 889void pci_msi_shutdown(struct pci_dev *dev)
1da177e4 890{
f2440d9a
MW
891 struct msi_desc *desc;
892 u32 mask;
1da177e4 893
128bc5fc 894 if (!pci_msi_enable || !dev || !dev->msi_enabled)
ded86d8d
EB
895 return;
896
5004e98a 897 BUG_ON(list_empty(dev_to_msi_list(&dev->dev)));
4a7cc831 898 desc = first_pci_msi_entry(dev);
110828c9 899
61b64abd 900 pci_msi_set_enable(dev, 0);
ba698ad4 901 pci_intx_for_msi(dev, 1);
b1cbf4e4 902 dev->msi_enabled = 0;
7bd007e4 903
12abb8ba 904 /* Return the device with MSI unmasked as initial states */
31ea5d4d 905 mask = msi_mask(desc->msi_attrib.multi_cap);
12abb8ba 906 /* Keep cached state to be restored */
23ed8d57 907 __pci_msi_desc_mask_irq(desc, mask, ~mask);
e387b9ee
ME
908
909 /* Restore dev->irq to its default pin-assertion irq */
f2440d9a 910 dev->irq = desc->msi_attrib.default_irq;
5f226991 911 pcibios_alloc_irq(dev);
d52877c7 912}
24d27553 913
500559a9 914void pci_disable_msi(struct pci_dev *dev)
d52877c7 915{
d52877c7
YL
916 if (!pci_msi_enable || !dev || !dev->msi_enabled)
917 return;
918
919 pci_msi_shutdown(dev);
f56e4481 920 free_msi_irqs(dev);
1da177e4 921}
4cc086fa 922EXPORT_SYMBOL(pci_disable_msi);
1da177e4 923
a52e2e35 924/**
ff1aa430 925 * pci_msix_vec_count - return the number of device's MSI-X table entries
a52e2e35 926 * @dev: pointer to the pci_dev data structure of MSI-X device function
ff1aa430
AG
927 * This function returns the number of device's MSI-X table entries and
928 * therefore the number of MSI-X vectors device is capable of sending.
929 * It returns a negative errno if the device is not capable of sending MSI-X
930 * interrupts.
931 **/
932int pci_msix_vec_count(struct pci_dev *dev)
a52e2e35 933{
a52e2e35
RW
934 u16 control;
935
520fe9dc 936 if (!dev->msix_cap)
ff1aa430 937 return -EINVAL;
a52e2e35 938
f84ecd28 939 pci_read_config_word(dev, dev->msix_cap + PCI_MSIX_FLAGS, &control);
527eee29 940 return msix_table_size(control);
a52e2e35 941}
ff1aa430 942EXPORT_SYMBOL(pci_msix_vec_count);
a52e2e35 943
1da177e4
LT
944/**
945 * pci_enable_msix - configure device's MSI-X capability structure
946 * @dev: pointer to the pci_dev data structure of MSI-X device function
3ac020e0 947 * @entries: pointer to an array of MSI-X entries (optional)
1ce03373 948 * @nvec: number of MSI-X irqs requested for allocation by device driver
1da177e4
LT
949 *
950 * Setup the MSI-X capability structure of device function with the number
1ce03373 951 * of requested irqs upon its software driver call to request for
1da177e4
LT
952 * MSI-X mode enabled on its hardware device function. A return of zero
953 * indicates the successful configuration of MSI-X capability structure
1ce03373 954 * with new allocated MSI-X irqs. A return of < 0 indicates a failure.
1da177e4 955 * Or a return of > 0 indicates that driver request is exceeding the number
57fbf52c
MT
956 * of irqs or MSI-X vectors available. Driver should use the returned value to
957 * re-send its request.
1da177e4 958 **/
500559a9 959int pci_enable_msix(struct pci_dev *dev, struct msix_entry *entries, int nvec)
1da177e4 960{
5ec09405 961 int nr_entries;
ded86d8d 962 int i, j;
1da177e4 963
a06cd74c
AG
964 if (!pci_msi_supported(dev, nvec))
965 return -EINVAL;
c9953a73 966
ff1aa430
AG
967 nr_entries = pci_msix_vec_count(dev);
968 if (nr_entries < 0)
969 return nr_entries;
1da177e4 970 if (nvec > nr_entries)
57fbf52c 971 return nr_entries;
1da177e4 972
3ac020e0
CH
973 if (entries) {
974 /* Check for any invalid entries */
975 for (i = 0; i < nvec; i++) {
976 if (entries[i].entry >= nr_entries)
977 return -EINVAL; /* invalid entry */
978 for (j = i + 1; j < nvec; j++) {
979 if (entries[i].entry == entries[j].entry)
980 return -EINVAL; /* duplicate entry */
981 }
1da177e4
LT
982 }
983 }
ded86d8d 984 WARN_ON(!!dev->msix_enabled);
7bd007e4 985
1ce03373 986 /* Check whether driver already requested for MSI irq */
500559a9 987 if (dev->msi_enabled) {
227f0647 988 dev_info(&dev->dev, "can't enable MSI-X (MSI IRQ already assigned)\n");
1da177e4
LT
989 return -EINVAL;
990 }
5ec09405 991 return msix_capability_init(dev, entries, nvec);
1da177e4 992}
4cc086fa 993EXPORT_SYMBOL(pci_enable_msix);
1da177e4 994
500559a9 995void pci_msix_shutdown(struct pci_dev *dev)
fc4afc7b 996{
12abb8ba
HS
997 struct msi_desc *entry;
998
128bc5fc 999 if (!pci_msi_enable || !dev || !dev->msix_enabled)
ded86d8d
EB
1000 return;
1001
12abb8ba 1002 /* Return the device with MSI-X masked as initial states */
5004e98a 1003 for_each_pci_msi_entry(entry, dev) {
12abb8ba 1004 /* Keep cached states to be restored */
23ed8d57 1005 __pci_msix_desc_mask_irq(entry, 1);
12abb8ba
HS
1006 }
1007
61b64abd 1008 pci_msix_clear_and_set_ctrl(dev, PCI_MSIX_FLAGS_ENABLE, 0);
ba698ad4 1009 pci_intx_for_msi(dev, 1);
b1cbf4e4 1010 dev->msix_enabled = 0;
5f226991 1011 pcibios_alloc_irq(dev);
d52877c7 1012}
c901851f 1013
500559a9 1014void pci_disable_msix(struct pci_dev *dev)
d52877c7
YL
1015{
1016 if (!pci_msi_enable || !dev || !dev->msix_enabled)
1017 return;
1018
1019 pci_msix_shutdown(dev);
f56e4481 1020 free_msi_irqs(dev);
1da177e4 1021}
4cc086fa 1022EXPORT_SYMBOL(pci_disable_msix);
1da177e4 1023
309e57df
MW
1024void pci_no_msi(void)
1025{
1026 pci_msi_enable = 0;
1027}
c9953a73 1028
07ae95f9
AP
1029/**
1030 * pci_msi_enabled - is MSI enabled?
1031 *
1032 * Returns true if MSI has not been disabled by the command-line option
1033 * pci=nomsi.
1034 **/
1035int pci_msi_enabled(void)
d389fec6 1036{
07ae95f9 1037 return pci_msi_enable;
d389fec6 1038}
07ae95f9 1039EXPORT_SYMBOL(pci_msi_enabled);
d389fec6 1040
4ef33685
CH
1041static int __pci_enable_msi_range(struct pci_dev *dev, int minvec, int maxvec,
1042 unsigned int flags)
302a2523 1043{
034cd97e 1044 int nvec;
302a2523
AG
1045 int rc;
1046
a06cd74c
AG
1047 if (!pci_msi_supported(dev, minvec))
1048 return -EINVAL;
034cd97e
AG
1049
1050 WARN_ON(!!dev->msi_enabled);
1051
1052 /* Check whether driver already requested MSI-X irqs */
1053 if (dev->msix_enabled) {
1054 dev_info(&dev->dev,
1055 "can't enable MSI (MSI-X already enabled)\n");
1056 return -EINVAL;
1057 }
1058
302a2523
AG
1059 if (maxvec < minvec)
1060 return -ERANGE;
1061
034cd97e
AG
1062 nvec = pci_msi_vec_count(dev);
1063 if (nvec < 0)
1064 return nvec;
4ef33685 1065 if (nvec < minvec)
034cd97e 1066 return -EINVAL;
4ef33685
CH
1067
1068 if (nvec > maxvec)
034cd97e
AG
1069 nvec = maxvec;
1070
4ef33685 1071 for (;;) {
4fe0d154 1072 if (flags & PCI_IRQ_AFFINITY) {
4ef33685
CH
1073 dev->irq_affinity = irq_create_affinity_mask(&nvec);
1074 if (nvec < minvec)
1075 return -ENOSPC;
1076 }
1077
034cd97e 1078 rc = msi_capability_init(dev, nvec);
4ef33685
CH
1079 if (rc == 0)
1080 return nvec;
1081
1082 kfree(dev->irq_affinity);
1083 dev->irq_affinity = NULL;
1084
1085 if (rc < 0)
302a2523 1086 return rc;
4ef33685
CH
1087 if (rc < minvec)
1088 return -ENOSPC;
1089
1090 nvec = rc;
1091 }
1092}
1093
1094/**
1095 * pci_enable_msi_range - configure device's MSI capability structure
1096 * @dev: device to configure
1097 * @minvec: minimal number of interrupts to configure
1098 * @maxvec: maximum number of interrupts to configure
1099 *
1100 * This function tries to allocate a maximum possible number of interrupts in a
1101 * range between @minvec and @maxvec. It returns a negative errno if an error
1102 * occurs. If it succeeds, it returns the actual number of interrupts allocated
1103 * and updates the @dev's irq member to the lowest new interrupt number;
1104 * the other interrupt numbers allocated to this device are consecutive.
1105 **/
1106int pci_enable_msi_range(struct pci_dev *dev, int minvec, int maxvec)
1107{
4fe0d154 1108 return __pci_enable_msi_range(dev, minvec, maxvec, 0);
4ef33685
CH
1109}
1110EXPORT_SYMBOL(pci_enable_msi_range);
1111
1112static int __pci_enable_msix_range(struct pci_dev *dev,
1113 struct msix_entry *entries, int minvec, int maxvec,
1114 unsigned int flags)
1115{
1116 int nvec = maxvec;
1117 int rc;
1118
1119 if (maxvec < minvec)
1120 return -ERANGE;
1121
1122 for (;;) {
4fe0d154 1123 if (flags & PCI_IRQ_AFFINITY) {
4ef33685
CH
1124 dev->irq_affinity = irq_create_affinity_mask(&nvec);
1125 if (nvec < minvec)
302a2523 1126 return -ENOSPC;
302a2523 1127 }
302a2523 1128
4ef33685
CH
1129 rc = pci_enable_msix(dev, entries, nvec);
1130 if (rc == 0)
1131 return nvec;
1132
1133 kfree(dev->irq_affinity);
1134 dev->irq_affinity = NULL;
1135
1136 if (rc < 0)
1137 return rc;
1138 if (rc < minvec)
1139 return -ENOSPC;
1140
1141 nvec = rc;
1142 }
302a2523 1143}
302a2523
AG
1144
1145/**
1146 * pci_enable_msix_range - configure device's MSI-X capability structure
1147 * @dev: pointer to the pci_dev data structure of MSI-X device function
1148 * @entries: pointer to an array of MSI-X entries
1149 * @minvec: minimum number of MSI-X irqs requested
1150 * @maxvec: maximum number of MSI-X irqs requested
1151 *
1152 * Setup the MSI-X capability structure of device function with a maximum
1153 * possible number of interrupts in the range between @minvec and @maxvec
1154 * upon its software driver call to request for MSI-X mode enabled on its
1155 * hardware device function. It returns a negative errno if an error occurs.
1156 * If it succeeds, it returns the actual number of interrupts allocated and
1157 * indicates the successful configuration of MSI-X capability structure
1158 * with new allocated MSI-X interrupts.
1159 **/
1160int pci_enable_msix_range(struct pci_dev *dev, struct msix_entry *entries,
4ef33685 1161 int minvec, int maxvec)
302a2523 1162{
4fe0d154 1163 return __pci_enable_msix_range(dev, entries, minvec, maxvec, 0);
302a2523
AG
1164}
1165EXPORT_SYMBOL(pci_enable_msix_range);
3878eaef 1166
aff17164
CH
1167/**
1168 * pci_alloc_irq_vectors - allocate multiple IRQs for a device
1169 * @dev: PCI device to operate on
1170 * @min_vecs: minimum number of vectors required (must be >= 1)
1171 * @max_vecs: maximum (desired) number of vectors
1172 * @flags: flags or quirks for the allocation
1173 *
1174 * Allocate up to @max_vecs interrupt vectors for @dev, using MSI-X or MSI
1175 * vectors if available, and fall back to a single legacy vector
1176 * if neither is available. Return the number of vectors allocated,
1177 * (which might be smaller than @max_vecs) if successful, or a negative
1178 * error code on error. If less than @min_vecs interrupt vectors are
1179 * available for @dev the function will fail with -ENOSPC.
1180 *
1181 * To get the Linux IRQ number used for a vector that can be passed to
1182 * request_irq() use the pci_irq_vector() helper.
1183 */
1184int pci_alloc_irq_vectors(struct pci_dev *dev, unsigned int min_vecs,
1185 unsigned int max_vecs, unsigned int flags)
1186{
1187 int vecs = -ENOSPC;
1188
4fe0d154 1189 if (flags & PCI_IRQ_MSIX) {
4ef33685
CH
1190 vecs = __pci_enable_msix_range(dev, NULL, min_vecs, max_vecs,
1191 flags);
aff17164
CH
1192 if (vecs > 0)
1193 return vecs;
1194 }
1195
4fe0d154 1196 if (flags & PCI_IRQ_MSI) {
4ef33685 1197 vecs = __pci_enable_msi_range(dev, min_vecs, max_vecs, flags);
aff17164
CH
1198 if (vecs > 0)
1199 return vecs;
1200 }
1201
1202 /* use legacy irq if allowed */
5d0bdf28
CH
1203 if ((flags & PCI_IRQ_LEGACY) && min_vecs == 1) {
1204 pci_intx(dev, 1);
aff17164 1205 return 1;
5d0bdf28
CH
1206 }
1207
aff17164
CH
1208 return vecs;
1209}
1210EXPORT_SYMBOL(pci_alloc_irq_vectors);
1211
1212/**
1213 * pci_free_irq_vectors - free previously allocated IRQs for a device
1214 * @dev: PCI device to operate on
1215 *
1216 * Undoes the allocations and enabling in pci_alloc_irq_vectors().
1217 */
1218void pci_free_irq_vectors(struct pci_dev *dev)
1219{
1220 pci_disable_msix(dev);
1221 pci_disable_msi(dev);
1222}
1223EXPORT_SYMBOL(pci_free_irq_vectors);
1224
1225/**
1226 * pci_irq_vector - return Linux IRQ number of a device vector
1227 * @dev: PCI device to operate on
1228 * @nr: device-relative interrupt vector index (0-based).
1229 */
1230int pci_irq_vector(struct pci_dev *dev, unsigned int nr)
1231{
1232 if (dev->msix_enabled) {
1233 struct msi_desc *entry;
1234 int i = 0;
1235
1236 for_each_pci_msi_entry(entry, dev) {
1237 if (i == nr)
1238 return entry->irq;
1239 i++;
1240 }
1241 WARN_ON_ONCE(1);
1242 return -EINVAL;
1243 }
1244
1245 if (dev->msi_enabled) {
1246 struct msi_desc *entry = first_pci_msi_entry(dev);
1247
1248 if (WARN_ON_ONCE(nr >= entry->nvec_used))
1249 return -EINVAL;
1250 } else {
1251 if (WARN_ON_ONCE(nr > 0))
1252 return -EINVAL;
1253 }
1254
1255 return dev->irq + nr;
1256}
1257EXPORT_SYMBOL(pci_irq_vector);
1258
25a98bd4
JL
1259struct pci_dev *msi_desc_to_pci_dev(struct msi_desc *desc)
1260{
1261 return to_pci_dev(desc->dev);
1262}
a4289dc2 1263EXPORT_SYMBOL(msi_desc_to_pci_dev);
25a98bd4 1264
c179c9b9
JL
1265void *msi_desc_to_pci_sysdata(struct msi_desc *desc)
1266{
1267 struct pci_dev *dev = msi_desc_to_pci_dev(desc);
1268
1269 return dev->bus->sysdata;
1270}
1271EXPORT_SYMBOL_GPL(msi_desc_to_pci_sysdata);
1272
3878eaef
JL
1273#ifdef CONFIG_PCI_MSI_IRQ_DOMAIN
1274/**
1275 * pci_msi_domain_write_msg - Helper to write MSI message to PCI config space
1276 * @irq_data: Pointer to interrupt data of the MSI interrupt
1277 * @msg: Pointer to the message
1278 */
1279void pci_msi_domain_write_msg(struct irq_data *irq_data, struct msi_msg *msg)
1280{
507a883e 1281 struct msi_desc *desc = irq_data_get_msi_desc(irq_data);
3878eaef
JL
1282
1283 /*
1284 * For MSI-X desc->irq is always equal to irq_data->irq. For
1285 * MSI only the first interrupt of MULTI MSI passes the test.
1286 */
1287 if (desc->irq == irq_data->irq)
1288 __pci_write_msi_msg(desc, msg);
1289}
1290
1291/**
1292 * pci_msi_domain_calc_hwirq - Generate a unique ID for an MSI source
1293 * @dev: Pointer to the PCI device
1294 * @desc: Pointer to the msi descriptor
1295 *
1296 * The ID number is only used within the irqdomain.
1297 */
1298irq_hw_number_t pci_msi_domain_calc_hwirq(struct pci_dev *dev,
1299 struct msi_desc *desc)
1300{
1301 return (irq_hw_number_t)desc->msi_attrib.entry_nr |
1302 PCI_DEVID(dev->bus->number, dev->devfn) << 11 |
1303 (pci_domain_nr(dev->bus) & 0xFFFFFFFF) << 27;
1304}
1305
1306static inline bool pci_msi_desc_is_multi_msi(struct msi_desc *desc)
1307{
1308 return !desc->msi_attrib.is_msix && desc->nvec_used > 1;
1309}
1310
1311/**
1312 * pci_msi_domain_check_cap - Verify that @domain supports the capabilities for @dev
1313 * @domain: The interrupt domain to check
1314 * @info: The domain info for verification
1315 * @dev: The device to check
1316 *
1317 * Returns:
1318 * 0 if the functionality is supported
1319 * 1 if Multi MSI is requested, but the domain does not support it
1320 * -ENOTSUPP otherwise
1321 */
1322int pci_msi_domain_check_cap(struct irq_domain *domain,
1323 struct msi_domain_info *info, struct device *dev)
1324{
1325 struct msi_desc *desc = first_pci_msi_entry(to_pci_dev(dev));
1326
1327 /* Special handling to support pci_enable_msi_range() */
1328 if (pci_msi_desc_is_multi_msi(desc) &&
1329 !(info->flags & MSI_FLAG_MULTI_PCI_MSI))
1330 return 1;
1331 else if (desc->msi_attrib.is_msix && !(info->flags & MSI_FLAG_PCI_MSIX))
1332 return -ENOTSUPP;
1333
1334 return 0;
1335}
1336
1337static int pci_msi_domain_handle_error(struct irq_domain *domain,
1338 struct msi_desc *desc, int error)
1339{
1340 /* Special handling to support pci_enable_msi_range() */
1341 if (pci_msi_desc_is_multi_msi(desc) && error == -ENOSPC)
1342 return 1;
1343
1344 return error;
1345}
1346
1347#ifdef GENERIC_MSI_DOMAIN_OPS
1348static void pci_msi_domain_set_desc(msi_alloc_info_t *arg,
1349 struct msi_desc *desc)
1350{
1351 arg->desc = desc;
1352 arg->hwirq = pci_msi_domain_calc_hwirq(msi_desc_to_pci_dev(desc),
1353 desc);
1354}
1355#else
1356#define pci_msi_domain_set_desc NULL
1357#endif
1358
1359static struct msi_domain_ops pci_msi_domain_ops_default = {
1360 .set_desc = pci_msi_domain_set_desc,
1361 .msi_check = pci_msi_domain_check_cap,
1362 .handle_error = pci_msi_domain_handle_error,
1363};
1364
1365static void pci_msi_domain_update_dom_ops(struct msi_domain_info *info)
1366{
1367 struct msi_domain_ops *ops = info->ops;
1368
1369 if (ops == NULL) {
1370 info->ops = &pci_msi_domain_ops_default;
1371 } else {
1372 if (ops->set_desc == NULL)
1373 ops->set_desc = pci_msi_domain_set_desc;
1374 if (ops->msi_check == NULL)
1375 ops->msi_check = pci_msi_domain_check_cap;
1376 if (ops->handle_error == NULL)
1377 ops->handle_error = pci_msi_domain_handle_error;
1378 }
1379}
1380
1381static void pci_msi_domain_update_chip_ops(struct msi_domain_info *info)
1382{
1383 struct irq_chip *chip = info->chip;
1384
1385 BUG_ON(!chip);
1386 if (!chip->irq_write_msi_msg)
1387 chip->irq_write_msi_msg = pci_msi_domain_write_msg;
0701c53e
MZ
1388 if (!chip->irq_mask)
1389 chip->irq_mask = pci_msi_mask_irq;
1390 if (!chip->irq_unmask)
1391 chip->irq_unmask = pci_msi_unmask_irq;
3878eaef
JL
1392}
1393
1394/**
be5436c8
MZ
1395 * pci_msi_create_irq_domain - Create a MSI interrupt domain
1396 * @fwnode: Optional fwnode of the interrupt controller
3878eaef
JL
1397 * @info: MSI domain info
1398 * @parent: Parent irq domain
1399 *
1400 * Updates the domain and chip ops and creates a MSI interrupt domain.
1401 *
1402 * Returns:
1403 * A domain pointer or NULL in case of failure.
1404 */
be5436c8 1405struct irq_domain *pci_msi_create_irq_domain(struct fwnode_handle *fwnode,
3878eaef
JL
1406 struct msi_domain_info *info,
1407 struct irq_domain *parent)
1408{
0380839d
MZ
1409 struct irq_domain *domain;
1410
3878eaef
JL
1411 if (info->flags & MSI_FLAG_USE_DEF_DOM_OPS)
1412 pci_msi_domain_update_dom_ops(info);
1413 if (info->flags & MSI_FLAG_USE_DEF_CHIP_OPS)
1414 pci_msi_domain_update_chip_ops(info);
1415
f3b0946d
MZ
1416 info->flags |= MSI_FLAG_ACTIVATE_EARLY;
1417
be5436c8 1418 domain = msi_create_irq_domain(fwnode, info, parent);
0380839d
MZ
1419 if (!domain)
1420 return NULL;
1421
1422 domain->bus_token = DOMAIN_BUS_PCI_MSI;
1423 return domain;
3878eaef 1424}
a4289dc2 1425EXPORT_SYMBOL_GPL(pci_msi_create_irq_domain);
3878eaef
JL
1426
1427/**
1428 * pci_msi_domain_alloc_irqs - Allocate interrupts for @dev in @domain
1429 * @domain: The interrupt domain to allocate from
1430 * @dev: The device for which to allocate
1431 * @nvec: The number of interrupts to allocate
1432 * @type: Unused to allow simpler migration from the arch_XXX interfaces
1433 *
1434 * Returns:
1435 * A virtual interrupt number or an error code in case of failure
1436 */
1437int pci_msi_domain_alloc_irqs(struct irq_domain *domain, struct pci_dev *dev,
1438 int nvec, int type)
1439{
1440 return msi_domain_alloc_irqs(domain, &dev->dev, nvec);
1441}
1442
1443/**
1444 * pci_msi_domain_free_irqs - Free interrupts for @dev in @domain
1445 * @domain: The interrupt domain
1446 * @dev: The device for which to free interrupts
1447 */
1448void pci_msi_domain_free_irqs(struct irq_domain *domain, struct pci_dev *dev)
1449{
1450 msi_domain_free_irqs(domain, &dev->dev);
1451}
8e047ada
JL
1452
1453/**
1454 * pci_msi_create_default_irq_domain - Create a default MSI interrupt domain
be5436c8 1455 * @fwnode: Optional fwnode of the interrupt controller
8e047ada
JL
1456 * @info: MSI domain info
1457 * @parent: Parent irq domain
1458 *
1459 * Returns: A domain pointer or NULL in case of failure. If successful
1460 * the default PCI/MSI irqdomain pointer is updated.
1461 */
be5436c8 1462struct irq_domain *pci_msi_create_default_irq_domain(struct fwnode_handle *fwnode,
8e047ada
JL
1463 struct msi_domain_info *info, struct irq_domain *parent)
1464{
1465 struct irq_domain *domain;
1466
1467 mutex_lock(&pci_msi_domain_lock);
1468 if (pci_msi_default_domain) {
1469 pr_err("PCI: default irq domain for PCI MSI has already been created.\n");
1470 domain = NULL;
1471 } else {
be5436c8 1472 domain = pci_msi_create_irq_domain(fwnode, info, parent);
8e047ada
JL
1473 pci_msi_default_domain = domain;
1474 }
1475 mutex_unlock(&pci_msi_domain_lock);
1476
1477 return domain;
1478}
b6eec9b7
DD
1479
1480static int get_msi_id_cb(struct pci_dev *pdev, u16 alias, void *data)
1481{
1482 u32 *pa = data;
1483
1484 *pa = alias;
1485 return 0;
1486}
1487/**
1488 * pci_msi_domain_get_msi_rid - Get the MSI requester id (RID)
1489 * @domain: The interrupt domain
1490 * @pdev: The PCI device.
1491 *
1492 * The RID for a device is formed from the alias, with a firmware
1493 * supplied mapping applied
1494 *
1495 * Returns: The RID.
1496 */
1497u32 pci_msi_domain_get_msi_rid(struct irq_domain *domain, struct pci_dev *pdev)
1498{
1499 struct device_node *of_node;
1500 u32 rid = 0;
1501
1502 pci_for_each_dma_alias(pdev, get_msi_id_cb, &rid);
1503
1504 of_node = irq_domain_get_of_node(domain);
1505 if (of_node)
1506 rid = of_msi_map_rid(&pdev->dev, of_node, rid);
1507
1508 return rid;
1509}
54fa97ee
MZ
1510
1511/**
1512 * pci_msi_get_device_domain - Get the MSI domain for a given PCI device
1513 * @pdev: The PCI device
1514 *
1515 * Use the firmware data to find a device-specific MSI domain
1516 * (i.e. not one that is ste as a default).
1517 *
1518 * Returns: The coresponding MSI domain or NULL if none has been found.
1519 */
1520struct irq_domain *pci_msi_get_device_domain(struct pci_dev *pdev)
1521{
1522 u32 rid = 0;
1523
1524 pci_for_each_dma_alias(pdev, get_msi_id_cb, &rid);
1525 return of_msi_map_get_device_domain(&pdev->dev, rid);
1526}
3878eaef 1527#endif /* CONFIG_PCI_MSI_IRQ_DOMAIN */