MSI: Use a list instead of the custom link structure
[linux-2.6.git] / drivers / pci / msi.c
blob434c7182d9261e5af0ec2086f8cebde17a74a174
1 /*
2 * File: msi.c
3 * Purpose: PCI Message Signaled Interrupt (MSI)
5 * Copyright (C) 2003-2004 Intel
6 * Copyright (C) Tom Long Nguyen (tom.l.nguyen@intel.com)
7 */
9 #include <linux/err.h>
10 #include <linux/mm.h>
11 #include <linux/irq.h>
12 #include <linux/interrupt.h>
13 #include <linux/init.h>
14 #include <linux/ioport.h>
15 #include <linux/smp_lock.h>
16 #include <linux/pci.h>
17 #include <linux/proc_fs.h>
18 #include <linux/msi.h>
20 #include <asm/errno.h>
21 #include <asm/io.h>
22 #include <asm/smp.h>
24 #include "pci.h"
25 #include "msi.h"
27 static int pci_msi_enable = 1;
29 static void msi_set_enable(struct pci_dev *dev, int enable)
31 int pos;
32 u16 control;
34 pos = pci_find_capability(dev, PCI_CAP_ID_MSI);
35 if (pos) {
36 pci_read_config_word(dev, pos + PCI_MSI_FLAGS, &control);
37 control &= ~PCI_MSI_FLAGS_ENABLE;
38 if (enable)
39 control |= PCI_MSI_FLAGS_ENABLE;
40 pci_write_config_word(dev, pos + PCI_MSI_FLAGS, control);
44 static void msix_set_enable(struct pci_dev *dev, int enable)
46 int pos;
47 u16 control;
49 pos = pci_find_capability(dev, PCI_CAP_ID_MSIX);
50 if (pos) {
51 pci_read_config_word(dev, pos + PCI_MSIX_FLAGS, &control);
52 control &= ~PCI_MSIX_FLAGS_ENABLE;
53 if (enable)
54 control |= PCI_MSIX_FLAGS_ENABLE;
55 pci_write_config_word(dev, pos + PCI_MSIX_FLAGS, control);
59 static void msix_flush_writes(unsigned int irq)
61 struct msi_desc *entry;
63 entry = get_irq_msi(irq);
64 BUG_ON(!entry || !entry->dev);
65 switch (entry->msi_attrib.type) {
66 case PCI_CAP_ID_MSI:
67 /* nothing to do */
68 break;
69 case PCI_CAP_ID_MSIX:
71 int offset = entry->msi_attrib.entry_nr * PCI_MSIX_ENTRY_SIZE +
72 PCI_MSIX_ENTRY_VECTOR_CTRL_OFFSET;
73 readl(entry->mask_base + offset);
74 break;
76 default:
77 BUG();
78 break;
82 static void msi_set_mask_bit(unsigned int irq, int flag)
84 struct msi_desc *entry;
86 entry = get_irq_msi(irq);
87 BUG_ON(!entry || !entry->dev);
88 switch (entry->msi_attrib.type) {
89 case PCI_CAP_ID_MSI:
90 if (entry->msi_attrib.maskbit) {
91 int pos;
92 u32 mask_bits;
94 pos = (long)entry->mask_base;
95 pci_read_config_dword(entry->dev, pos, &mask_bits);
96 mask_bits &= ~(1);
97 mask_bits |= flag;
98 pci_write_config_dword(entry->dev, pos, mask_bits);
99 } else {
100 msi_set_enable(entry->dev, !flag);
102 break;
103 case PCI_CAP_ID_MSIX:
105 int offset = entry->msi_attrib.entry_nr * PCI_MSIX_ENTRY_SIZE +
106 PCI_MSIX_ENTRY_VECTOR_CTRL_OFFSET;
107 writel(flag, entry->mask_base + offset);
108 readl(entry->mask_base + offset);
109 break;
111 default:
112 BUG();
113 break;
115 entry->msi_attrib.masked = !!flag;
118 void read_msi_msg(unsigned int irq, struct msi_msg *msg)
120 struct msi_desc *entry = get_irq_msi(irq);
121 switch(entry->msi_attrib.type) {
122 case PCI_CAP_ID_MSI:
124 struct pci_dev *dev = entry->dev;
125 int pos = entry->msi_attrib.pos;
126 u16 data;
128 pci_read_config_dword(dev, msi_lower_address_reg(pos),
129 &msg->address_lo);
130 if (entry->msi_attrib.is_64) {
131 pci_read_config_dword(dev, msi_upper_address_reg(pos),
132 &msg->address_hi);
133 pci_read_config_word(dev, msi_data_reg(pos, 1), &data);
134 } else {
135 msg->address_hi = 0;
136 pci_read_config_word(dev, msi_data_reg(pos, 1), &data);
138 msg->data = data;
139 break;
141 case PCI_CAP_ID_MSIX:
143 void __iomem *base;
144 base = entry->mask_base +
145 entry->msi_attrib.entry_nr * PCI_MSIX_ENTRY_SIZE;
147 msg->address_lo = readl(base + PCI_MSIX_ENTRY_LOWER_ADDR_OFFSET);
148 msg->address_hi = readl(base + PCI_MSIX_ENTRY_UPPER_ADDR_OFFSET);
149 msg->data = readl(base + PCI_MSIX_ENTRY_DATA_OFFSET);
150 break;
152 default:
153 BUG();
157 void write_msi_msg(unsigned int irq, struct msi_msg *msg)
159 struct msi_desc *entry = get_irq_msi(irq);
160 switch (entry->msi_attrib.type) {
161 case PCI_CAP_ID_MSI:
163 struct pci_dev *dev = entry->dev;
164 int pos = entry->msi_attrib.pos;
166 pci_write_config_dword(dev, msi_lower_address_reg(pos),
167 msg->address_lo);
168 if (entry->msi_attrib.is_64) {
169 pci_write_config_dword(dev, msi_upper_address_reg(pos),
170 msg->address_hi);
171 pci_write_config_word(dev, msi_data_reg(pos, 1),
172 msg->data);
173 } else {
174 pci_write_config_word(dev, msi_data_reg(pos, 0),
175 msg->data);
177 break;
179 case PCI_CAP_ID_MSIX:
181 void __iomem *base;
182 base = entry->mask_base +
183 entry->msi_attrib.entry_nr * PCI_MSIX_ENTRY_SIZE;
185 writel(msg->address_lo,
186 base + PCI_MSIX_ENTRY_LOWER_ADDR_OFFSET);
187 writel(msg->address_hi,
188 base + PCI_MSIX_ENTRY_UPPER_ADDR_OFFSET);
189 writel(msg->data, base + PCI_MSIX_ENTRY_DATA_OFFSET);
190 break;
192 default:
193 BUG();
195 entry->msg = *msg;
198 void mask_msi_irq(unsigned int irq)
200 msi_set_mask_bit(irq, 1);
201 msix_flush_writes(irq);
204 void unmask_msi_irq(unsigned int irq)
206 msi_set_mask_bit(irq, 0);
207 msix_flush_writes(irq);
210 static int msi_free_irq(struct pci_dev* dev, int irq);
213 static struct msi_desc* alloc_msi_entry(void)
215 struct msi_desc *entry;
217 entry = kzalloc(sizeof(struct msi_desc), GFP_KERNEL);
218 if (!entry)
219 return NULL;
221 INIT_LIST_HEAD(&entry->list);
222 entry->irq = 0;
223 entry->dev = NULL;
225 return entry;
228 #ifdef CONFIG_PM
229 static void __pci_restore_msi_state(struct pci_dev *dev)
231 int pos;
232 u16 control;
233 struct msi_desc *entry;
235 if (!dev->msi_enabled)
236 return;
238 entry = get_irq_msi(dev->irq);
239 pos = entry->msi_attrib.pos;
241 pci_intx(dev, 0); /* disable intx */
242 msi_set_enable(dev, 0);
243 write_msi_msg(dev->irq, &entry->msg);
244 if (entry->msi_attrib.maskbit)
245 msi_set_mask_bit(dev->irq, entry->msi_attrib.masked);
247 pci_read_config_word(dev, pos + PCI_MSI_FLAGS, &control);
248 control &= ~(PCI_MSI_FLAGS_QSIZE | PCI_MSI_FLAGS_ENABLE);
249 if (entry->msi_attrib.maskbit || !entry->msi_attrib.masked)
250 control |= PCI_MSI_FLAGS_ENABLE;
251 pci_write_config_word(dev, pos + PCI_MSI_FLAGS, control);
254 static void __pci_restore_msix_state(struct pci_dev *dev)
256 int pos;
257 struct msi_desc *entry;
258 u16 control;
260 if (!dev->msix_enabled)
261 return;
263 /* route the table */
264 pci_intx(dev, 0); /* disable intx */
265 msix_set_enable(dev, 0);
267 list_for_each_entry(entry, &dev->msi_list, list) {
268 write_msi_msg(entry->irq, &entry->msg);
269 msi_set_mask_bit(entry->irq, entry->msi_attrib.masked);
272 entry = get_irq_msi(dev->first_msi_irq);
273 pos = entry->msi_attrib.pos;
274 pci_read_config_word(dev, pos + PCI_MSIX_FLAGS, &control);
275 control &= ~PCI_MSIX_FLAGS_MASKALL;
276 control |= PCI_MSIX_FLAGS_ENABLE;
277 pci_write_config_word(dev, pos + PCI_MSIX_FLAGS, control);
280 void pci_restore_msi_state(struct pci_dev *dev)
282 __pci_restore_msi_state(dev);
283 __pci_restore_msix_state(dev);
285 #endif /* CONFIG_PM */
288 * msi_capability_init - configure device's MSI capability structure
289 * @dev: pointer to the pci_dev data structure of MSI device function
291 * Setup the MSI capability structure of device function with a single
292 * MSI irq, regardless of device function is capable of handling
293 * multiple messages. A return of zero indicates the successful setup
294 * of an entry zero with the new MSI irq or non-zero for otherwise.
296 static int msi_capability_init(struct pci_dev *dev)
298 struct msi_desc *entry;
299 int pos, irq;
300 u16 control;
302 msi_set_enable(dev, 0); /* Ensure msi is disabled as I set it up */
304 pos = pci_find_capability(dev, PCI_CAP_ID_MSI);
305 pci_read_config_word(dev, msi_control_reg(pos), &control);
306 /* MSI Entry Initialization */
307 entry = alloc_msi_entry();
308 if (!entry)
309 return -ENOMEM;
311 entry->msi_attrib.type = PCI_CAP_ID_MSI;
312 entry->msi_attrib.is_64 = is_64bit_address(control);
313 entry->msi_attrib.entry_nr = 0;
314 entry->msi_attrib.maskbit = is_mask_bit_support(control);
315 entry->msi_attrib.masked = 1;
316 entry->msi_attrib.default_irq = dev->irq; /* Save IOAPIC IRQ */
317 entry->msi_attrib.pos = pos;
318 if (is_mask_bit_support(control)) {
319 entry->mask_base = (void __iomem *)(long)msi_mask_bits_reg(pos,
320 is_64bit_address(control));
322 entry->dev = dev;
323 if (entry->msi_attrib.maskbit) {
324 unsigned int maskbits, temp;
325 /* All MSIs are unmasked by default, Mask them all */
326 pci_read_config_dword(dev,
327 msi_mask_bits_reg(pos, is_64bit_address(control)),
328 &maskbits);
329 temp = (1 << multi_msi_capable(control));
330 temp = ((temp - 1) & ~temp);
331 maskbits |= temp;
332 pci_write_config_dword(dev,
333 msi_mask_bits_reg(pos, is_64bit_address(control)),
334 maskbits);
336 /* Configure MSI capability structure */
337 irq = arch_setup_msi_irq(dev, entry);
338 if (irq < 0) {
339 kfree(entry);
340 return irq;
342 entry->irq = irq;
343 list_add(&entry->list, &dev->msi_list);
344 dev->first_msi_irq = irq;
345 set_irq_msi(irq, entry);
347 /* Set MSI enabled bits */
348 pci_intx(dev, 0); /* disable intx */
349 msi_set_enable(dev, 1);
350 dev->msi_enabled = 1;
352 dev->irq = irq;
353 return 0;
357 * msix_capability_init - configure device's MSI-X capability
358 * @dev: pointer to the pci_dev data structure of MSI-X device function
359 * @entries: pointer to an array of struct msix_entry entries
360 * @nvec: number of @entries
362 * Setup the MSI-X capability structure of device function with a
363 * single MSI-X irq. A return of zero indicates the successful setup of
364 * requested MSI-X entries with allocated irqs or non-zero for otherwise.
366 static int msix_capability_init(struct pci_dev *dev,
367 struct msix_entry *entries, int nvec)
369 struct msi_desc *entry;
370 int irq, pos, i, j, nr_entries;
371 unsigned long phys_addr;
372 u32 table_offset;
373 u16 control;
374 u8 bir;
375 void __iomem *base;
377 msix_set_enable(dev, 0);/* Ensure msix is disabled as I set it up */
379 pos = pci_find_capability(dev, PCI_CAP_ID_MSIX);
380 /* Request & Map MSI-X table region */
381 pci_read_config_word(dev, msi_control_reg(pos), &control);
382 nr_entries = multi_msix_capable(control);
384 pci_read_config_dword(dev, msix_table_offset_reg(pos), &table_offset);
385 bir = (u8)(table_offset & PCI_MSIX_FLAGS_BIRMASK);
386 table_offset &= ~PCI_MSIX_FLAGS_BIRMASK;
387 phys_addr = pci_resource_start (dev, bir) + table_offset;
388 base = ioremap_nocache(phys_addr, nr_entries * PCI_MSIX_ENTRY_SIZE);
389 if (base == NULL)
390 return -ENOMEM;
392 /* MSI-X Table Initialization */
393 for (i = 0; i < nvec; i++) {
394 entry = alloc_msi_entry();
395 if (!entry)
396 break;
398 j = entries[i].entry;
399 entry->msi_attrib.type = PCI_CAP_ID_MSIX;
400 entry->msi_attrib.is_64 = 1;
401 entry->msi_attrib.entry_nr = j;
402 entry->msi_attrib.maskbit = 1;
403 entry->msi_attrib.masked = 1;
404 entry->msi_attrib.default_irq = dev->irq;
405 entry->msi_attrib.pos = pos;
406 entry->dev = dev;
407 entry->mask_base = base;
409 /* Configure MSI-X capability structure */
410 irq = arch_setup_msi_irq(dev, entry);
411 if (irq < 0) {
412 kfree(entry);
413 break;
415 entry->irq = irq;
416 entries[i].vector = irq;
417 list_add(&entry->list, &dev->msi_list);
419 set_irq_msi(irq, entry);
421 if (i != nvec) {
422 int avail = i - 1;
423 i--;
424 for (; i >= 0; i--) {
425 irq = (entries + i)->vector;
426 msi_free_irq(dev, irq);
427 (entries + i)->vector = 0;
429 /* If we had some success report the number of irqs
430 * we succeeded in setting up.
432 if (avail <= 0)
433 avail = -EBUSY;
434 return avail;
436 dev->first_msi_irq = entries[0].vector;
437 /* Set MSI-X enabled bits */
438 pci_intx(dev, 0); /* disable intx */
439 msix_set_enable(dev, 1);
440 dev->msix_enabled = 1;
442 return 0;
446 * pci_msi_check_device - check whether MSI may be enabled on a device
447 * @dev: pointer to the pci_dev data structure of MSI device function
448 * @nvec: how many MSIs have been requested ?
449 * @type: are we checking for MSI or MSI-X ?
451 * Look at global flags, the device itself, and its parent busses
452 * to determine if MSI/-X are supported for the device. If MSI/-X is
453 * supported return 0, else return an error code.
455 static int pci_msi_check_device(struct pci_dev* dev, int nvec, int type)
457 struct pci_bus *bus;
458 int ret;
460 /* MSI must be globally enabled and supported by the device */
461 if (!pci_msi_enable || !dev || dev->no_msi)
462 return -EINVAL;
464 /* Any bridge which does NOT route MSI transactions from it's
465 * secondary bus to it's primary bus must set NO_MSI flag on
466 * the secondary pci_bus.
467 * We expect only arch-specific PCI host bus controller driver
468 * or quirks for specific PCI bridges to be setting NO_MSI.
470 for (bus = dev->bus; bus; bus = bus->parent)
471 if (bus->bus_flags & PCI_BUS_FLAGS_NO_MSI)
472 return -EINVAL;
474 ret = arch_msi_check_device(dev, nvec, type);
475 if (ret)
476 return ret;
478 if (!pci_find_capability(dev, type))
479 return -EINVAL;
481 return 0;
485 * pci_enable_msi - configure device's MSI capability structure
486 * @dev: pointer to the pci_dev data structure of MSI device function
488 * Setup the MSI capability structure of device function with
489 * a single MSI irq upon its software driver call to request for
490 * MSI mode enabled on its hardware device function. A return of zero
491 * indicates the successful setup of an entry zero with the new MSI
492 * irq or non-zero for otherwise.
494 int pci_enable_msi(struct pci_dev* dev)
496 int status;
498 status = pci_msi_check_device(dev, 1, PCI_CAP_ID_MSI);
499 if (status)
500 return status;
502 WARN_ON(!!dev->msi_enabled);
504 /* Check whether driver already requested for MSI-X irqs */
505 if (dev->msix_enabled) {
506 printk(KERN_INFO "PCI: %s: Can't enable MSI. "
507 "Device already has MSI-X enabled\n",
508 pci_name(dev));
509 return -EINVAL;
511 status = msi_capability_init(dev);
512 return status;
514 EXPORT_SYMBOL(pci_enable_msi);
516 void pci_disable_msi(struct pci_dev* dev)
518 struct msi_desc *entry;
519 int default_irq;
521 if (!pci_msi_enable || !dev || !dev->msi_enabled)
522 return;
524 msi_set_enable(dev, 0);
525 pci_intx(dev, 1); /* enable intx */
526 dev->msi_enabled = 0;
528 entry = get_irq_msi(dev->first_msi_irq);
529 if (!entry || !entry->dev || entry->msi_attrib.type != PCI_CAP_ID_MSI) {
530 return;
533 default_irq = entry->msi_attrib.default_irq;
534 msi_free_irq(dev, dev->first_msi_irq);
536 /* Restore dev->irq to its default pin-assertion irq */
537 dev->irq = default_irq;
539 dev->first_msi_irq = 0;
541 EXPORT_SYMBOL(pci_disable_msi);
543 static int msi_free_irq(struct pci_dev* dev, int irq)
545 struct msi_desc *entry;
546 int entry_nr, type;
547 void __iomem *base;
549 BUG_ON(irq_has_action(irq));
551 entry = get_irq_msi(irq);
552 if (!entry || entry->dev != dev) {
553 return -EINVAL;
555 type = entry->msi_attrib.type;
556 entry_nr = entry->msi_attrib.entry_nr;
557 base = entry->mask_base;
558 list_del(&entry->list);
560 arch_teardown_msi_irq(irq);
561 kfree(entry);
563 if (type == PCI_CAP_ID_MSIX) {
564 writel(1, base + entry_nr * PCI_MSIX_ENTRY_SIZE +
565 PCI_MSIX_ENTRY_VECTOR_CTRL_OFFSET);
567 if (list_empty(&dev->msi_list))
568 iounmap(base);
571 return 0;
575 * pci_enable_msix - configure device's MSI-X capability structure
576 * @dev: pointer to the pci_dev data structure of MSI-X device function
577 * @entries: pointer to an array of MSI-X entries
578 * @nvec: number of MSI-X irqs requested for allocation by device driver
580 * Setup the MSI-X capability structure of device function with the number
581 * of requested irqs upon its software driver call to request for
582 * MSI-X mode enabled on its hardware device function. A return of zero
583 * indicates the successful configuration of MSI-X capability structure
584 * with new allocated MSI-X irqs. A return of < 0 indicates a failure.
585 * Or a return of > 0 indicates that driver request is exceeding the number
586 * of irqs available. Driver should use the returned value to re-send
587 * its request.
589 int pci_enable_msix(struct pci_dev* dev, struct msix_entry *entries, int nvec)
591 int status, pos, nr_entries;
592 int i, j;
593 u16 control;
595 if (!entries)
596 return -EINVAL;
598 status = pci_msi_check_device(dev, nvec, PCI_CAP_ID_MSIX);
599 if (status)
600 return status;
602 pos = pci_find_capability(dev, PCI_CAP_ID_MSIX);
603 pci_read_config_word(dev, msi_control_reg(pos), &control);
604 nr_entries = multi_msix_capable(control);
605 if (nvec > nr_entries)
606 return -EINVAL;
608 /* Check for any invalid entries */
609 for (i = 0; i < nvec; i++) {
610 if (entries[i].entry >= nr_entries)
611 return -EINVAL; /* invalid entry */
612 for (j = i + 1; j < nvec; j++) {
613 if (entries[i].entry == entries[j].entry)
614 return -EINVAL; /* duplicate entry */
617 WARN_ON(!!dev->msix_enabled);
619 /* Check whether driver already requested for MSI irq */
620 if (dev->msi_enabled) {
621 printk(KERN_INFO "PCI: %s: Can't enable MSI-X. "
622 "Device already has an MSI irq assigned\n",
623 pci_name(dev));
624 return -EINVAL;
626 status = msix_capability_init(dev, entries, nvec);
627 return status;
629 EXPORT_SYMBOL(pci_enable_msix);
631 static void msix_free_all_irqs(struct pci_dev *dev)
633 struct msi_desc *entry;
635 list_for_each_entry(entry, &dev->msi_list, list)
636 msi_free_irq(dev, entry->irq);
637 dev->first_msi_irq = 0;
640 void pci_disable_msix(struct pci_dev* dev)
642 if (!pci_msi_enable || !dev || !dev->msix_enabled)
643 return;
645 msix_set_enable(dev, 0);
646 pci_intx(dev, 1); /* enable intx */
647 dev->msix_enabled = 0;
649 msix_free_all_irqs(dev);
651 EXPORT_SYMBOL(pci_disable_msix);
654 * msi_remove_pci_irq_vectors - reclaim MSI(X) irqs to unused state
655 * @dev: pointer to the pci_dev data structure of MSI(X) device function
657 * Being called during hotplug remove, from which the device function
658 * is hot-removed. All previous assigned MSI/MSI-X irqs, if
659 * allocated for this device function, are reclaimed to unused state,
660 * which may be used later on.
662 void msi_remove_pci_irq_vectors(struct pci_dev* dev)
664 if (!pci_msi_enable || !dev)
665 return;
667 if (dev->msi_enabled)
668 msi_free_irq(dev, dev->first_msi_irq);
670 if (dev->msix_enabled)
671 msix_free_all_irqs(dev);
674 void pci_no_msi(void)
676 pci_msi_enable = 0;
679 void pci_msi_init_pci_dev(struct pci_dev *dev)
681 INIT_LIST_HEAD(&dev->msi_list);
685 /* Arch hooks */
687 int __attribute__ ((weak))
688 arch_msi_check_device(struct pci_dev* dev, int nvec, int type)
690 return 0;