Merge git://git.kernel.org/pub/scm/linux/kernel/git/gregkh/driver-2.6
[pv_ops_mirror.git] / drivers / pci / msi.c
blob26938da8f4380c5fd0bdb1786c71f8adb8418046
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/pci.h>
16 #include <linux/proc_fs.h>
17 #include <linux/msi.h>
18 #include <linux/smp.h>
20 #include <asm/errno.h>
21 #include <asm/io.h>
23 #include "pci.h"
24 #include "msi.h"
26 static int pci_msi_enable = 1;
28 /* Arch hooks */
30 int __attribute__ ((weak))
31 arch_msi_check_device(struct pci_dev *dev, int nvec, int type)
33 return 0;
36 int __attribute__ ((weak))
37 arch_setup_msi_irq(struct pci_dev *dev, struct msi_desc *entry)
39 return 0;
42 int __attribute__ ((weak))
43 arch_setup_msi_irqs(struct pci_dev *dev, int nvec, int type)
45 struct msi_desc *entry;
46 int ret;
48 list_for_each_entry(entry, &dev->msi_list, list) {
49 ret = arch_setup_msi_irq(dev, entry);
50 if (ret)
51 return ret;
54 return 0;
57 void __attribute__ ((weak)) arch_teardown_msi_irq(unsigned int irq)
59 return;
62 void __attribute__ ((weak))
63 arch_teardown_msi_irqs(struct pci_dev *dev)
65 struct msi_desc *entry;
67 list_for_each_entry(entry, &dev->msi_list, list) {
68 if (entry->irq != 0)
69 arch_teardown_msi_irq(entry->irq);
73 static void msi_set_enable(struct pci_dev *dev, int enable)
75 int pos;
76 u16 control;
78 pos = pci_find_capability(dev, PCI_CAP_ID_MSI);
79 if (pos) {
80 pci_read_config_word(dev, pos + PCI_MSI_FLAGS, &control);
81 control &= ~PCI_MSI_FLAGS_ENABLE;
82 if (enable)
83 control |= PCI_MSI_FLAGS_ENABLE;
84 pci_write_config_word(dev, pos + PCI_MSI_FLAGS, control);
88 static void msix_set_enable(struct pci_dev *dev, int enable)
90 int pos;
91 u16 control;
93 pos = pci_find_capability(dev, PCI_CAP_ID_MSIX);
94 if (pos) {
95 pci_read_config_word(dev, pos + PCI_MSIX_FLAGS, &control);
96 control &= ~PCI_MSIX_FLAGS_ENABLE;
97 if (enable)
98 control |= PCI_MSIX_FLAGS_ENABLE;
99 pci_write_config_word(dev, pos + PCI_MSIX_FLAGS, control);
103 static void msix_flush_writes(unsigned int irq)
105 struct msi_desc *entry;
107 entry = get_irq_msi(irq);
108 BUG_ON(!entry || !entry->dev);
109 switch (entry->msi_attrib.type) {
110 case PCI_CAP_ID_MSI:
111 /* nothing to do */
112 break;
113 case PCI_CAP_ID_MSIX:
115 int offset = entry->msi_attrib.entry_nr * PCI_MSIX_ENTRY_SIZE +
116 PCI_MSIX_ENTRY_VECTOR_CTRL_OFFSET;
117 readl(entry->mask_base + offset);
118 break;
120 default:
121 BUG();
122 break;
126 static void msi_set_mask_bit(unsigned int irq, int flag)
128 struct msi_desc *entry;
130 entry = get_irq_msi(irq);
131 BUG_ON(!entry || !entry->dev);
132 switch (entry->msi_attrib.type) {
133 case PCI_CAP_ID_MSI:
134 if (entry->msi_attrib.maskbit) {
135 int pos;
136 u32 mask_bits;
138 pos = (long)entry->mask_base;
139 pci_read_config_dword(entry->dev, pos, &mask_bits);
140 mask_bits &= ~(1);
141 mask_bits |= flag;
142 pci_write_config_dword(entry->dev, pos, mask_bits);
143 } else {
144 msi_set_enable(entry->dev, !flag);
146 break;
147 case PCI_CAP_ID_MSIX:
149 int offset = entry->msi_attrib.entry_nr * PCI_MSIX_ENTRY_SIZE +
150 PCI_MSIX_ENTRY_VECTOR_CTRL_OFFSET;
151 writel(flag, entry->mask_base + offset);
152 readl(entry->mask_base + offset);
153 break;
155 default:
156 BUG();
157 break;
159 entry->msi_attrib.masked = !!flag;
162 void read_msi_msg(unsigned int irq, struct msi_msg *msg)
164 struct msi_desc *entry = get_irq_msi(irq);
165 switch(entry->msi_attrib.type) {
166 case PCI_CAP_ID_MSI:
168 struct pci_dev *dev = entry->dev;
169 int pos = entry->msi_attrib.pos;
170 u16 data;
172 pci_read_config_dword(dev, msi_lower_address_reg(pos),
173 &msg->address_lo);
174 if (entry->msi_attrib.is_64) {
175 pci_read_config_dword(dev, msi_upper_address_reg(pos),
176 &msg->address_hi);
177 pci_read_config_word(dev, msi_data_reg(pos, 1), &data);
178 } else {
179 msg->address_hi = 0;
180 pci_read_config_word(dev, msi_data_reg(pos, 0), &data);
182 msg->data = data;
183 break;
185 case PCI_CAP_ID_MSIX:
187 void __iomem *base;
188 base = entry->mask_base +
189 entry->msi_attrib.entry_nr * PCI_MSIX_ENTRY_SIZE;
191 msg->address_lo = readl(base + PCI_MSIX_ENTRY_LOWER_ADDR_OFFSET);
192 msg->address_hi = readl(base + PCI_MSIX_ENTRY_UPPER_ADDR_OFFSET);
193 msg->data = readl(base + PCI_MSIX_ENTRY_DATA_OFFSET);
194 break;
196 default:
197 BUG();
201 void write_msi_msg(unsigned int irq, struct msi_msg *msg)
203 struct msi_desc *entry = get_irq_msi(irq);
204 switch (entry->msi_attrib.type) {
205 case PCI_CAP_ID_MSI:
207 struct pci_dev *dev = entry->dev;
208 int pos = entry->msi_attrib.pos;
210 pci_write_config_dword(dev, msi_lower_address_reg(pos),
211 msg->address_lo);
212 if (entry->msi_attrib.is_64) {
213 pci_write_config_dword(dev, msi_upper_address_reg(pos),
214 msg->address_hi);
215 pci_write_config_word(dev, msi_data_reg(pos, 1),
216 msg->data);
217 } else {
218 pci_write_config_word(dev, msi_data_reg(pos, 0),
219 msg->data);
221 break;
223 case PCI_CAP_ID_MSIX:
225 void __iomem *base;
226 base = entry->mask_base +
227 entry->msi_attrib.entry_nr * PCI_MSIX_ENTRY_SIZE;
229 writel(msg->address_lo,
230 base + PCI_MSIX_ENTRY_LOWER_ADDR_OFFSET);
231 writel(msg->address_hi,
232 base + PCI_MSIX_ENTRY_UPPER_ADDR_OFFSET);
233 writel(msg->data, base + PCI_MSIX_ENTRY_DATA_OFFSET);
234 break;
236 default:
237 BUG();
239 entry->msg = *msg;
242 void mask_msi_irq(unsigned int irq)
244 msi_set_mask_bit(irq, 1);
245 msix_flush_writes(irq);
248 void unmask_msi_irq(unsigned int irq)
250 msi_set_mask_bit(irq, 0);
251 msix_flush_writes(irq);
254 static int msi_free_irqs(struct pci_dev* dev);
257 static struct msi_desc* alloc_msi_entry(void)
259 struct msi_desc *entry;
261 entry = kzalloc(sizeof(struct msi_desc), GFP_KERNEL);
262 if (!entry)
263 return NULL;
265 INIT_LIST_HEAD(&entry->list);
266 entry->irq = 0;
267 entry->dev = NULL;
269 return entry;
272 static void pci_intx_for_msi(struct pci_dev *dev, int enable)
274 if (!(dev->dev_flags & PCI_DEV_FLAGS_MSI_INTX_DISABLE_BUG))
275 pci_intx(dev, enable);
278 static void __pci_restore_msi_state(struct pci_dev *dev)
280 int pos;
281 u16 control;
282 struct msi_desc *entry;
284 if (!dev->msi_enabled)
285 return;
287 entry = get_irq_msi(dev->irq);
288 pos = entry->msi_attrib.pos;
290 pci_intx_for_msi(dev, 0);
291 msi_set_enable(dev, 0);
292 write_msi_msg(dev->irq, &entry->msg);
293 if (entry->msi_attrib.maskbit)
294 msi_set_mask_bit(dev->irq, entry->msi_attrib.masked);
296 pci_read_config_word(dev, pos + PCI_MSI_FLAGS, &control);
297 control &= ~(PCI_MSI_FLAGS_QSIZE | PCI_MSI_FLAGS_ENABLE);
298 if (entry->msi_attrib.maskbit || !entry->msi_attrib.masked)
299 control |= PCI_MSI_FLAGS_ENABLE;
300 pci_write_config_word(dev, pos + PCI_MSI_FLAGS, control);
303 static void __pci_restore_msix_state(struct pci_dev *dev)
305 int pos;
306 struct msi_desc *entry;
307 u16 control;
309 if (!dev->msix_enabled)
310 return;
312 /* route the table */
313 pci_intx_for_msi(dev, 0);
314 msix_set_enable(dev, 0);
316 list_for_each_entry(entry, &dev->msi_list, list) {
317 write_msi_msg(entry->irq, &entry->msg);
318 msi_set_mask_bit(entry->irq, entry->msi_attrib.masked);
321 BUG_ON(list_empty(&dev->msi_list));
322 entry = list_entry(dev->msi_list.next, struct msi_desc, list);
323 pos = entry->msi_attrib.pos;
324 pci_read_config_word(dev, pos + PCI_MSIX_FLAGS, &control);
325 control &= ~PCI_MSIX_FLAGS_MASKALL;
326 control |= PCI_MSIX_FLAGS_ENABLE;
327 pci_write_config_word(dev, pos + PCI_MSIX_FLAGS, control);
330 void pci_restore_msi_state(struct pci_dev *dev)
332 __pci_restore_msi_state(dev);
333 __pci_restore_msix_state(dev);
335 EXPORT_SYMBOL_GPL(pci_restore_msi_state);
338 * msi_capability_init - configure device's MSI capability structure
339 * @dev: pointer to the pci_dev data structure of MSI device function
341 * Setup the MSI capability structure of device function with a single
342 * MSI irq, regardless of device function is capable of handling
343 * multiple messages. A return of zero indicates the successful setup
344 * of an entry zero with the new MSI irq or non-zero for otherwise.
346 static int msi_capability_init(struct pci_dev *dev)
348 struct msi_desc *entry;
349 int pos, ret;
350 u16 control;
352 msi_set_enable(dev, 0); /* Ensure msi is disabled as I set it up */
354 pos = pci_find_capability(dev, PCI_CAP_ID_MSI);
355 pci_read_config_word(dev, msi_control_reg(pos), &control);
356 /* MSI Entry Initialization */
357 entry = alloc_msi_entry();
358 if (!entry)
359 return -ENOMEM;
361 entry->msi_attrib.type = PCI_CAP_ID_MSI;
362 entry->msi_attrib.is_64 = is_64bit_address(control);
363 entry->msi_attrib.entry_nr = 0;
364 entry->msi_attrib.maskbit = is_mask_bit_support(control);
365 entry->msi_attrib.masked = 1;
366 entry->msi_attrib.default_irq = dev->irq; /* Save IOAPIC IRQ */
367 entry->msi_attrib.pos = pos;
368 if (is_mask_bit_support(control)) {
369 entry->mask_base = (void __iomem *)(long)msi_mask_bits_reg(pos,
370 is_64bit_address(control));
372 entry->dev = dev;
373 if (entry->msi_attrib.maskbit) {
374 unsigned int maskbits, temp;
375 /* All MSIs are unmasked by default, Mask them all */
376 pci_read_config_dword(dev,
377 msi_mask_bits_reg(pos, is_64bit_address(control)),
378 &maskbits);
379 temp = (1 << multi_msi_capable(control));
380 temp = ((temp - 1) & ~temp);
381 maskbits |= temp;
382 pci_write_config_dword(dev,
383 msi_mask_bits_reg(pos, is_64bit_address(control)),
384 maskbits);
386 list_add_tail(&entry->list, &dev->msi_list);
388 /* Configure MSI capability structure */
389 ret = arch_setup_msi_irqs(dev, 1, PCI_CAP_ID_MSI);
390 if (ret) {
391 msi_free_irqs(dev);
392 return ret;
395 /* Set MSI enabled bits */
396 pci_intx_for_msi(dev, 0);
397 msi_set_enable(dev, 1);
398 dev->msi_enabled = 1;
400 dev->irq = entry->irq;
401 return 0;
405 * msix_capability_init - configure device's MSI-X capability
406 * @dev: pointer to the pci_dev data structure of MSI-X device function
407 * @entries: pointer to an array of struct msix_entry entries
408 * @nvec: number of @entries
410 * Setup the MSI-X capability structure of device function with a
411 * single MSI-X irq. A return of zero indicates the successful setup of
412 * requested MSI-X entries with allocated irqs or non-zero for otherwise.
414 static int msix_capability_init(struct pci_dev *dev,
415 struct msix_entry *entries, int nvec)
417 struct msi_desc *entry;
418 int pos, i, j, nr_entries, ret;
419 unsigned long phys_addr;
420 u32 table_offset;
421 u16 control;
422 u8 bir;
423 void __iomem *base;
425 msix_set_enable(dev, 0);/* Ensure msix is disabled as I set it up */
427 pos = pci_find_capability(dev, PCI_CAP_ID_MSIX);
428 /* Request & Map MSI-X table region */
429 pci_read_config_word(dev, msi_control_reg(pos), &control);
430 nr_entries = multi_msix_capable(control);
432 pci_read_config_dword(dev, msix_table_offset_reg(pos), &table_offset);
433 bir = (u8)(table_offset & PCI_MSIX_FLAGS_BIRMASK);
434 table_offset &= ~PCI_MSIX_FLAGS_BIRMASK;
435 phys_addr = pci_resource_start (dev, bir) + table_offset;
436 base = ioremap_nocache(phys_addr, nr_entries * PCI_MSIX_ENTRY_SIZE);
437 if (base == NULL)
438 return -ENOMEM;
440 /* MSI-X Table Initialization */
441 for (i = 0; i < nvec; i++) {
442 entry = alloc_msi_entry();
443 if (!entry)
444 break;
446 j = entries[i].entry;
447 entry->msi_attrib.type = PCI_CAP_ID_MSIX;
448 entry->msi_attrib.is_64 = 1;
449 entry->msi_attrib.entry_nr = j;
450 entry->msi_attrib.maskbit = 1;
451 entry->msi_attrib.masked = 1;
452 entry->msi_attrib.default_irq = dev->irq;
453 entry->msi_attrib.pos = pos;
454 entry->dev = dev;
455 entry->mask_base = base;
457 list_add_tail(&entry->list, &dev->msi_list);
460 ret = arch_setup_msi_irqs(dev, nvec, PCI_CAP_ID_MSIX);
461 if (ret) {
462 int avail = 0;
463 list_for_each_entry(entry, &dev->msi_list, list) {
464 if (entry->irq != 0) {
465 avail++;
469 msi_free_irqs(dev);
471 /* If we had some success report the number of irqs
472 * we succeeded in setting up.
474 if (avail == 0)
475 avail = ret;
476 return avail;
479 i = 0;
480 list_for_each_entry(entry, &dev->msi_list, list) {
481 entries[i].vector = entry->irq;
482 set_irq_msi(entry->irq, entry);
483 i++;
485 /* Set MSI-X enabled bits */
486 pci_intx_for_msi(dev, 0);
487 msix_set_enable(dev, 1);
488 dev->msix_enabled = 1;
490 return 0;
494 * pci_msi_check_device - check whether MSI may be enabled on a device
495 * @dev: pointer to the pci_dev data structure of MSI device function
496 * @nvec: how many MSIs have been requested ?
497 * @type: are we checking for MSI or MSI-X ?
499 * Look at global flags, the device itself, and its parent busses
500 * to determine if MSI/-X are supported for the device. If MSI/-X is
501 * supported return 0, else return an error code.
503 static int pci_msi_check_device(struct pci_dev* dev, int nvec, int type)
505 struct pci_bus *bus;
506 int ret;
508 /* MSI must be globally enabled and supported by the device */
509 if (!pci_msi_enable || !dev || dev->no_msi)
510 return -EINVAL;
513 * You can't ask to have 0 or less MSIs configured.
514 * a) it's stupid ..
515 * b) the list manipulation code assumes nvec >= 1.
517 if (nvec < 1)
518 return -ERANGE;
520 /* Any bridge which does NOT route MSI transactions from it's
521 * secondary bus to it's primary bus must set NO_MSI flag on
522 * the secondary pci_bus.
523 * We expect only arch-specific PCI host bus controller driver
524 * or quirks for specific PCI bridges to be setting NO_MSI.
526 for (bus = dev->bus; bus; bus = bus->parent)
527 if (bus->bus_flags & PCI_BUS_FLAGS_NO_MSI)
528 return -EINVAL;
530 ret = arch_msi_check_device(dev, nvec, type);
531 if (ret)
532 return ret;
534 if (!pci_find_capability(dev, type))
535 return -EINVAL;
537 return 0;
541 * pci_enable_msi - configure device's MSI capability structure
542 * @dev: pointer to the pci_dev data structure of MSI device function
544 * Setup the MSI capability structure of device function with
545 * a single MSI irq upon its software driver call to request for
546 * MSI mode enabled on its hardware device function. A return of zero
547 * indicates the successful setup of an entry zero with the new MSI
548 * irq or non-zero for otherwise.
550 int pci_enable_msi(struct pci_dev* dev)
552 int status;
554 status = pci_msi_check_device(dev, 1, PCI_CAP_ID_MSI);
555 if (status)
556 return status;
558 WARN_ON(!!dev->msi_enabled);
560 /* Check whether driver already requested for MSI-X irqs */
561 if (dev->msix_enabled) {
562 printk(KERN_INFO "PCI: %s: Can't enable MSI. "
563 "Device already has MSI-X enabled\n",
564 pci_name(dev));
565 return -EINVAL;
567 status = msi_capability_init(dev);
568 return status;
570 EXPORT_SYMBOL(pci_enable_msi);
572 void pci_disable_msi(struct pci_dev* dev)
574 struct msi_desc *entry;
575 int default_irq;
577 if (!pci_msi_enable || !dev || !dev->msi_enabled)
578 return;
580 msi_set_enable(dev, 0);
581 pci_intx_for_msi(dev, 1);
582 dev->msi_enabled = 0;
584 BUG_ON(list_empty(&dev->msi_list));
585 entry = list_entry(dev->msi_list.next, struct msi_desc, list);
586 if (!entry->dev || entry->msi_attrib.type != PCI_CAP_ID_MSI) {
587 return;
590 default_irq = entry->msi_attrib.default_irq;
591 msi_free_irqs(dev);
593 /* Restore dev->irq to its default pin-assertion irq */
594 dev->irq = default_irq;
596 EXPORT_SYMBOL(pci_disable_msi);
598 static int msi_free_irqs(struct pci_dev* dev)
600 struct msi_desc *entry, *tmp;
602 list_for_each_entry(entry, &dev->msi_list, list) {
603 if (entry->irq)
604 BUG_ON(irq_has_action(entry->irq));
607 arch_teardown_msi_irqs(dev);
609 list_for_each_entry_safe(entry, tmp, &dev->msi_list, list) {
610 if (entry->msi_attrib.type == PCI_CAP_ID_MSIX) {
611 writel(1, entry->mask_base + entry->msi_attrib.entry_nr
612 * PCI_MSIX_ENTRY_SIZE
613 + PCI_MSIX_ENTRY_VECTOR_CTRL_OFFSET);
615 if (list_is_last(&entry->list, &dev->msi_list))
616 iounmap(entry->mask_base);
618 list_del(&entry->list);
619 kfree(entry);
622 return 0;
626 * pci_enable_msix - configure device's MSI-X capability structure
627 * @dev: pointer to the pci_dev data structure of MSI-X device function
628 * @entries: pointer to an array of MSI-X entries
629 * @nvec: number of MSI-X irqs requested for allocation by device driver
631 * Setup the MSI-X capability structure of device function with the number
632 * of requested irqs upon its software driver call to request for
633 * MSI-X mode enabled on its hardware device function. A return of zero
634 * indicates the successful configuration of MSI-X capability structure
635 * with new allocated MSI-X irqs. A return of < 0 indicates a failure.
636 * Or a return of > 0 indicates that driver request is exceeding the number
637 * of irqs available. Driver should use the returned value to re-send
638 * its request.
640 int pci_enable_msix(struct pci_dev* dev, struct msix_entry *entries, int nvec)
642 int status, pos, nr_entries;
643 int i, j;
644 u16 control;
646 if (!entries)
647 return -EINVAL;
649 status = pci_msi_check_device(dev, nvec, PCI_CAP_ID_MSIX);
650 if (status)
651 return status;
653 pos = pci_find_capability(dev, PCI_CAP_ID_MSIX);
654 pci_read_config_word(dev, msi_control_reg(pos), &control);
655 nr_entries = multi_msix_capable(control);
656 if (nvec > nr_entries)
657 return -EINVAL;
659 /* Check for any invalid entries */
660 for (i = 0; i < nvec; i++) {
661 if (entries[i].entry >= nr_entries)
662 return -EINVAL; /* invalid entry */
663 for (j = i + 1; j < nvec; j++) {
664 if (entries[i].entry == entries[j].entry)
665 return -EINVAL; /* duplicate entry */
668 WARN_ON(!!dev->msix_enabled);
670 /* Check whether driver already requested for MSI irq */
671 if (dev->msi_enabled) {
672 printk(KERN_INFO "PCI: %s: Can't enable MSI-X. "
673 "Device already has an MSI irq assigned\n",
674 pci_name(dev));
675 return -EINVAL;
677 status = msix_capability_init(dev, entries, nvec);
678 return status;
680 EXPORT_SYMBOL(pci_enable_msix);
682 static void msix_free_all_irqs(struct pci_dev *dev)
684 msi_free_irqs(dev);
687 void pci_disable_msix(struct pci_dev* dev)
689 if (!pci_msi_enable || !dev || !dev->msix_enabled)
690 return;
692 msix_set_enable(dev, 0);
693 pci_intx_for_msi(dev, 1);
694 dev->msix_enabled = 0;
696 msix_free_all_irqs(dev);
698 EXPORT_SYMBOL(pci_disable_msix);
701 * msi_remove_pci_irq_vectors - reclaim MSI(X) irqs to unused state
702 * @dev: pointer to the pci_dev data structure of MSI(X) device function
704 * Being called during hotplug remove, from which the device function
705 * is hot-removed. All previous assigned MSI/MSI-X irqs, if
706 * allocated for this device function, are reclaimed to unused state,
707 * which may be used later on.
709 void msi_remove_pci_irq_vectors(struct pci_dev* dev)
711 if (!pci_msi_enable || !dev)
712 return;
714 if (dev->msi_enabled)
715 msi_free_irqs(dev);
717 if (dev->msix_enabled)
718 msix_free_all_irqs(dev);
721 void pci_no_msi(void)
723 pci_msi_enable = 0;
726 void pci_msi_init_pci_dev(struct pci_dev *dev)
728 INIT_LIST_HEAD(&dev->msi_list);