MSI: Give archs the option to free all MSI/Xs at once.
[linux-2.6/openmoko-kernel/knife-kernel.git] / drivers / pci / msi.c
blob9e1321d0d5e683d281c9e10fc42c826c38e8f4cf
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_irqs(struct pci_dev* dev);
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 BUG_ON(list_empty(&dev->msi_list));
273 entry = list_entry(dev->msi_list.next, struct msi_desc, list);
274 pos = entry->msi_attrib.pos;
275 pci_read_config_word(dev, pos + PCI_MSIX_FLAGS, &control);
276 control &= ~PCI_MSIX_FLAGS_MASKALL;
277 control |= PCI_MSIX_FLAGS_ENABLE;
278 pci_write_config_word(dev, pos + PCI_MSIX_FLAGS, control);
281 void pci_restore_msi_state(struct pci_dev *dev)
283 __pci_restore_msi_state(dev);
284 __pci_restore_msix_state(dev);
286 #endif /* CONFIG_PM */
289 * msi_capability_init - configure device's MSI capability structure
290 * @dev: pointer to the pci_dev data structure of MSI device function
292 * Setup the MSI capability structure of device function with a single
293 * MSI irq, regardless of device function is capable of handling
294 * multiple messages. A return of zero indicates the successful setup
295 * of an entry zero with the new MSI irq or non-zero for otherwise.
297 static int msi_capability_init(struct pci_dev *dev)
299 struct msi_desc *entry;
300 int pos, ret;
301 u16 control;
303 msi_set_enable(dev, 0); /* Ensure msi is disabled as I set it up */
305 pos = pci_find_capability(dev, PCI_CAP_ID_MSI);
306 pci_read_config_word(dev, msi_control_reg(pos), &control);
307 /* MSI Entry Initialization */
308 entry = alloc_msi_entry();
309 if (!entry)
310 return -ENOMEM;
312 entry->msi_attrib.type = PCI_CAP_ID_MSI;
313 entry->msi_attrib.is_64 = is_64bit_address(control);
314 entry->msi_attrib.entry_nr = 0;
315 entry->msi_attrib.maskbit = is_mask_bit_support(control);
316 entry->msi_attrib.masked = 1;
317 entry->msi_attrib.default_irq = dev->irq; /* Save IOAPIC IRQ */
318 entry->msi_attrib.pos = pos;
319 if (is_mask_bit_support(control)) {
320 entry->mask_base = (void __iomem *)(long)msi_mask_bits_reg(pos,
321 is_64bit_address(control));
323 entry->dev = dev;
324 if (entry->msi_attrib.maskbit) {
325 unsigned int maskbits, temp;
326 /* All MSIs are unmasked by default, Mask them all */
327 pci_read_config_dword(dev,
328 msi_mask_bits_reg(pos, is_64bit_address(control)),
329 &maskbits);
330 temp = (1 << multi_msi_capable(control));
331 temp = ((temp - 1) & ~temp);
332 maskbits |= temp;
333 pci_write_config_dword(dev,
334 msi_mask_bits_reg(pos, is_64bit_address(control)),
335 maskbits);
337 list_add(&entry->list, &dev->msi_list);
339 /* Configure MSI capability structure */
340 ret = arch_setup_msi_irqs(dev, 1, PCI_CAP_ID_MSI);
341 if (ret) {
342 msi_free_irqs(dev);
343 return ret;
346 /* Set MSI enabled bits */
347 pci_intx(dev, 0); /* disable intx */
348 msi_set_enable(dev, 1);
349 dev->msi_enabled = 1;
351 dev->irq = entry->irq;
352 return 0;
356 * msix_capability_init - configure device's MSI-X capability
357 * @dev: pointer to the pci_dev data structure of MSI-X device function
358 * @entries: pointer to an array of struct msix_entry entries
359 * @nvec: number of @entries
361 * Setup the MSI-X capability structure of device function with a
362 * single MSI-X irq. A return of zero indicates the successful setup of
363 * requested MSI-X entries with allocated irqs or non-zero for otherwise.
365 static int msix_capability_init(struct pci_dev *dev,
366 struct msix_entry *entries, int nvec)
368 struct msi_desc *entry;
369 int pos, i, j, nr_entries, ret;
370 unsigned long phys_addr;
371 u32 table_offset;
372 u16 control;
373 u8 bir;
374 void __iomem *base;
376 msix_set_enable(dev, 0);/* Ensure msix is disabled as I set it up */
378 pos = pci_find_capability(dev, PCI_CAP_ID_MSIX);
379 /* Request & Map MSI-X table region */
380 pci_read_config_word(dev, msi_control_reg(pos), &control);
381 nr_entries = multi_msix_capable(control);
383 pci_read_config_dword(dev, msix_table_offset_reg(pos), &table_offset);
384 bir = (u8)(table_offset & PCI_MSIX_FLAGS_BIRMASK);
385 table_offset &= ~PCI_MSIX_FLAGS_BIRMASK;
386 phys_addr = pci_resource_start (dev, bir) + table_offset;
387 base = ioremap_nocache(phys_addr, nr_entries * PCI_MSIX_ENTRY_SIZE);
388 if (base == NULL)
389 return -ENOMEM;
391 /* MSI-X Table Initialization */
392 for (i = 0; i < nvec; i++) {
393 entry = alloc_msi_entry();
394 if (!entry)
395 break;
397 j = entries[i].entry;
398 entry->msi_attrib.type = PCI_CAP_ID_MSIX;
399 entry->msi_attrib.is_64 = 1;
400 entry->msi_attrib.entry_nr = j;
401 entry->msi_attrib.maskbit = 1;
402 entry->msi_attrib.masked = 1;
403 entry->msi_attrib.default_irq = dev->irq;
404 entry->msi_attrib.pos = pos;
405 entry->dev = dev;
406 entry->mask_base = base;
408 list_add(&entry->list, &dev->msi_list);
411 ret = arch_setup_msi_irqs(dev, nvec, PCI_CAP_ID_MSIX);
412 if (ret) {
413 int avail = 0;
414 list_for_each_entry(entry, &dev->msi_list, list) {
415 if (entry->irq != 0) {
416 avail++;
420 msi_free_irqs(dev);
422 /* If we had some success report the number of irqs
423 * we succeeded in setting up.
425 if (avail == 0)
426 avail = ret;
427 return avail;
430 i = 0;
431 list_for_each_entry(entry, &dev->msi_list, list) {
432 entries[i].vector = entry->irq;
433 set_irq_msi(entry->irq, entry);
434 i++;
436 /* Set MSI-X enabled bits */
437 pci_intx(dev, 0); /* disable intx */
438 msix_set_enable(dev, 1);
439 dev->msix_enabled = 1;
441 return 0;
445 * pci_msi_check_device - check whether MSI may be enabled on a device
446 * @dev: pointer to the pci_dev data structure of MSI device function
447 * @nvec: how many MSIs have been requested ?
448 * @type: are we checking for MSI or MSI-X ?
450 * Look at global flags, the device itself, and its parent busses
451 * to determine if MSI/-X are supported for the device. If MSI/-X is
452 * supported return 0, else return an error code.
454 static int pci_msi_check_device(struct pci_dev* dev, int nvec, int type)
456 struct pci_bus *bus;
457 int ret;
459 /* MSI must be globally enabled and supported by the device */
460 if (!pci_msi_enable || !dev || dev->no_msi)
461 return -EINVAL;
464 * You can't ask to have 0 or less MSIs configured.
465 * a) it's stupid ..
466 * b) the list manipulation code assumes nvec >= 1.
468 if (nvec < 1)
469 return -ERANGE;
471 /* Any bridge which does NOT route MSI transactions from it's
472 * secondary bus to it's primary bus must set NO_MSI flag on
473 * the secondary pci_bus.
474 * We expect only arch-specific PCI host bus controller driver
475 * or quirks for specific PCI bridges to be setting NO_MSI.
477 for (bus = dev->bus; bus; bus = bus->parent)
478 if (bus->bus_flags & PCI_BUS_FLAGS_NO_MSI)
479 return -EINVAL;
481 ret = arch_msi_check_device(dev, nvec, type);
482 if (ret)
483 return ret;
485 if (!pci_find_capability(dev, type))
486 return -EINVAL;
488 return 0;
492 * pci_enable_msi - configure device's MSI capability structure
493 * @dev: pointer to the pci_dev data structure of MSI device function
495 * Setup the MSI capability structure of device function with
496 * a single MSI irq upon its software driver call to request for
497 * MSI mode enabled on its hardware device function. A return of zero
498 * indicates the successful setup of an entry zero with the new MSI
499 * irq or non-zero for otherwise.
501 int pci_enable_msi(struct pci_dev* dev)
503 int status;
505 status = pci_msi_check_device(dev, 1, PCI_CAP_ID_MSI);
506 if (status)
507 return status;
509 WARN_ON(!!dev->msi_enabled);
511 /* Check whether driver already requested for MSI-X irqs */
512 if (dev->msix_enabled) {
513 printk(KERN_INFO "PCI: %s: Can't enable MSI. "
514 "Device already has MSI-X enabled\n",
515 pci_name(dev));
516 return -EINVAL;
518 status = msi_capability_init(dev);
519 return status;
521 EXPORT_SYMBOL(pci_enable_msi);
523 void pci_disable_msi(struct pci_dev* dev)
525 struct msi_desc *entry;
526 int default_irq;
528 if (!pci_msi_enable || !dev || !dev->msi_enabled)
529 return;
531 msi_set_enable(dev, 0);
532 pci_intx(dev, 1); /* enable intx */
533 dev->msi_enabled = 0;
535 BUG_ON(list_empty(&dev->msi_list));
536 entry = list_entry(dev->msi_list.next, struct msi_desc, list);
537 if (!entry->dev || entry->msi_attrib.type != PCI_CAP_ID_MSI) {
538 return;
541 default_irq = entry->msi_attrib.default_irq;
542 msi_free_irqs(dev);
544 /* Restore dev->irq to its default pin-assertion irq */
545 dev->irq = default_irq;
547 EXPORT_SYMBOL(pci_disable_msi);
549 static int msi_free_irqs(struct pci_dev* dev)
551 struct msi_desc *entry, *tmp;
553 list_for_each_entry(entry, &dev->msi_list, list)
554 BUG_ON(irq_has_action(entry->irq));
556 arch_teardown_msi_irqs(dev);
558 list_for_each_entry_safe(entry, tmp, &dev->msi_list, list) {
559 if (entry->msi_attrib.type == PCI_CAP_ID_MSIX) {
560 if (list_is_last(&entry->list, &dev->msi_list))
561 iounmap(entry->mask_base);
563 writel(1, entry->mask_base + entry->msi_attrib.entry_nr
564 * PCI_MSIX_ENTRY_SIZE
565 + PCI_MSIX_ENTRY_VECTOR_CTRL_OFFSET);
567 list_del(&entry->list);
568 kfree(entry);
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 msi_free_irqs(dev);
636 void pci_disable_msix(struct pci_dev* dev)
638 if (!pci_msi_enable || !dev || !dev->msix_enabled)
639 return;
641 msix_set_enable(dev, 0);
642 pci_intx(dev, 1); /* enable intx */
643 dev->msix_enabled = 0;
645 msix_free_all_irqs(dev);
647 EXPORT_SYMBOL(pci_disable_msix);
650 * msi_remove_pci_irq_vectors - reclaim MSI(X) irqs to unused state
651 * @dev: pointer to the pci_dev data structure of MSI(X) device function
653 * Being called during hotplug remove, from which the device function
654 * is hot-removed. All previous assigned MSI/MSI-X irqs, if
655 * allocated for this device function, are reclaimed to unused state,
656 * which may be used later on.
658 void msi_remove_pci_irq_vectors(struct pci_dev* dev)
660 if (!pci_msi_enable || !dev)
661 return;
663 if (dev->msi_enabled)
664 msi_free_irqs(dev);
666 if (dev->msix_enabled)
667 msix_free_all_irqs(dev);
670 void pci_no_msi(void)
672 pci_msi_enable = 0;
675 void pci_msi_init_pci_dev(struct pci_dev *dev)
677 INIT_LIST_HEAD(&dev->msi_list);
681 /* Arch hooks */
683 int __attribute__ ((weak))
684 arch_msi_check_device(struct pci_dev* dev, int nvec, int type)
686 return 0;
689 int __attribute__ ((weak))
690 arch_setup_msi_irq(struct pci_dev *dev, struct msi_desc *entry)
692 return 0;
695 int __attribute__ ((weak))
696 arch_setup_msi_irqs(struct pci_dev *dev, int nvec, int type)
698 struct msi_desc *entry;
699 int ret;
701 list_for_each_entry(entry, &dev->msi_list, list) {
702 ret = arch_setup_msi_irq(dev, entry);
703 if (ret)
704 return ret;
707 return 0;
710 void __attribute__ ((weak)) arch_teardown_msi_irq(unsigned int irq)
712 return;
715 void __attribute__ ((weak))
716 arch_teardown_msi_irqs(struct pci_dev *dev)
718 struct msi_desc *entry;
720 list_for_each_entry(entry, &dev->msi_list, list) {
721 if (entry->irq != 0)
722 arch_teardown_msi_irq(entry->irq);