msi.c revision 5b912c108c8b1fcecbfe13d6d9a183db97b682d3
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)
7 */
8
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>
19
20#include <asm/errno.h>
21#include <asm/io.h>
22#include <asm/smp.h>
23
24#include "pci.h"
25#include "msi.h"
26
27static struct kmem_cache* msi_cachep;
28
29static int pci_msi_enable = 1;
30
31static int msi_cache_init(void)
32{
33	msi_cachep = kmem_cache_create("msi_cache", sizeof(struct msi_desc),
34					0, SLAB_HWCACHE_ALIGN, NULL, NULL);
35	if (!msi_cachep)
36		return -ENOMEM;
37
38	return 0;
39}
40
41static void msi_set_mask_bit(unsigned int irq, int flag)
42{
43	struct msi_desc *entry;
44
45	entry = get_irq_msi(irq);
46	BUG_ON(!entry || !entry->dev);
47	switch (entry->msi_attrib.type) {
48	case PCI_CAP_ID_MSI:
49		if (entry->msi_attrib.maskbit) {
50			int pos;
51			u32 mask_bits;
52
53			pos = (long)entry->mask_base;
54			pci_read_config_dword(entry->dev, pos, &mask_bits);
55			mask_bits &= ~(1);
56			mask_bits |= flag;
57			pci_write_config_dword(entry->dev, pos, mask_bits);
58		}
59		break;
60	case PCI_CAP_ID_MSIX:
61	{
62		int offset = entry->msi_attrib.entry_nr * PCI_MSIX_ENTRY_SIZE +
63			PCI_MSIX_ENTRY_VECTOR_CTRL_OFFSET;
64		writel(flag, entry->mask_base + offset);
65		break;
66	}
67	default:
68		BUG();
69		break;
70	}
71}
72
73void read_msi_msg(unsigned int irq, struct msi_msg *msg)
74{
75	struct msi_desc *entry = get_irq_msi(irq);
76	switch(entry->msi_attrib.type) {
77	case PCI_CAP_ID_MSI:
78	{
79		struct pci_dev *dev = entry->dev;
80		int pos = entry->msi_attrib.pos;
81		u16 data;
82
83		pci_read_config_dword(dev, msi_lower_address_reg(pos),
84					&msg->address_lo);
85		if (entry->msi_attrib.is_64) {
86			pci_read_config_dword(dev, msi_upper_address_reg(pos),
87						&msg->address_hi);
88			pci_read_config_word(dev, msi_data_reg(pos, 1), &data);
89		} else {
90			msg->address_hi = 0;
91			pci_read_config_word(dev, msi_data_reg(pos, 1), &data);
92		}
93		msg->data = data;
94		break;
95	}
96	case PCI_CAP_ID_MSIX:
97	{
98		void __iomem *base;
99		base = entry->mask_base +
100			entry->msi_attrib.entry_nr * PCI_MSIX_ENTRY_SIZE;
101
102		msg->address_lo = readl(base + PCI_MSIX_ENTRY_LOWER_ADDR_OFFSET);
103		msg->address_hi = readl(base + PCI_MSIX_ENTRY_UPPER_ADDR_OFFSET);
104		msg->data = readl(base + PCI_MSIX_ENTRY_DATA_OFFSET);
105 		break;
106 	}
107 	default:
108		BUG();
109	}
110}
111
112void write_msi_msg(unsigned int irq, struct msi_msg *msg)
113{
114	struct msi_desc *entry = get_irq_msi(irq);
115	switch (entry->msi_attrib.type) {
116	case PCI_CAP_ID_MSI:
117	{
118		struct pci_dev *dev = entry->dev;
119		int pos = entry->msi_attrib.pos;
120
121		pci_write_config_dword(dev, msi_lower_address_reg(pos),
122					msg->address_lo);
123		if (entry->msi_attrib.is_64) {
124			pci_write_config_dword(dev, msi_upper_address_reg(pos),
125						msg->address_hi);
126			pci_write_config_word(dev, msi_data_reg(pos, 1),
127						msg->data);
128		} else {
129			pci_write_config_word(dev, msi_data_reg(pos, 0),
130						msg->data);
131		}
132		break;
133	}
134	case PCI_CAP_ID_MSIX:
135	{
136		void __iomem *base;
137		base = entry->mask_base +
138			entry->msi_attrib.entry_nr * PCI_MSIX_ENTRY_SIZE;
139
140		writel(msg->address_lo,
141			base + PCI_MSIX_ENTRY_LOWER_ADDR_OFFSET);
142		writel(msg->address_hi,
143			base + PCI_MSIX_ENTRY_UPPER_ADDR_OFFSET);
144		writel(msg->data, base + PCI_MSIX_ENTRY_DATA_OFFSET);
145		break;
146	}
147	default:
148		BUG();
149	}
150}
151
152void mask_msi_irq(unsigned int irq)
153{
154	msi_set_mask_bit(irq, 1);
155}
156
157void unmask_msi_irq(unsigned int irq)
158{
159	msi_set_mask_bit(irq, 0);
160}
161
162static int msi_free_irq(struct pci_dev* dev, int irq);
163
164static int msi_init(void)
165{
166	static int status = -ENOMEM;
167
168	if (!status)
169		return status;
170
171	status = msi_cache_init();
172	if (status < 0) {
173		pci_msi_enable = 0;
174		printk(KERN_WARNING "PCI: MSI cache init failed\n");
175		return status;
176	}
177
178	return status;
179}
180
181static struct msi_desc* alloc_msi_entry(void)
182{
183	struct msi_desc *entry;
184
185	entry = kmem_cache_zalloc(msi_cachep, GFP_KERNEL);
186	if (!entry)
187		return NULL;
188
189	entry->link.tail = entry->link.head = 0;	/* single message */
190	entry->dev = NULL;
191
192	return entry;
193}
194
195static int create_msi_irq(void)
196{
197	struct msi_desc *entry;
198	int irq;
199
200	entry = alloc_msi_entry();
201	if (!entry)
202		return -ENOMEM;
203
204	irq = create_irq();
205	if (irq < 0) {
206		kmem_cache_free(msi_cachep, entry);
207		return -EBUSY;
208	}
209
210	set_irq_msi(irq, entry);
211
212	return irq;
213}
214
215static void destroy_msi_irq(unsigned int irq)
216{
217	struct msi_desc *entry;
218
219	entry = get_irq_msi(irq);
220	set_irq_chip(irq, NULL);
221	set_irq_msi(irq, NULL);
222	destroy_irq(irq);
223	kmem_cache_free(msi_cachep, entry);
224}
225
226static void enable_msi_mode(struct pci_dev *dev, int pos, int type)
227{
228	u16 control;
229
230	pci_read_config_word(dev, msi_control_reg(pos), &control);
231	if (type == PCI_CAP_ID_MSI) {
232		/* Set enabled bits to single MSI & enable MSI_enable bit */
233		msi_enable(control, 1);
234		pci_write_config_word(dev, msi_control_reg(pos), control);
235		dev->msi_enabled = 1;
236	} else {
237		msix_enable(control);
238		pci_write_config_word(dev, msi_control_reg(pos), control);
239		dev->msix_enabled = 1;
240	}
241
242	pci_intx(dev, 0);  /* disable intx */
243}
244
245void disable_msi_mode(struct pci_dev *dev, int pos, int type)
246{
247	u16 control;
248
249	pci_read_config_word(dev, msi_control_reg(pos), &control);
250	if (type == PCI_CAP_ID_MSI) {
251		/* Set enabled bits to single MSI & enable MSI_enable bit */
252		msi_disable(control);
253		pci_write_config_word(dev, msi_control_reg(pos), control);
254		dev->msi_enabled = 0;
255	} else {
256		msix_disable(control);
257		pci_write_config_word(dev, msi_control_reg(pos), control);
258		dev->msix_enabled = 0;
259	}
260
261	pci_intx(dev, 1);  /* enable intx */
262}
263
264#ifdef CONFIG_PM
265static int __pci_save_msi_state(struct pci_dev *dev)
266{
267	int pos, i = 0;
268	u16 control;
269	struct pci_cap_saved_state *save_state;
270	u32 *cap;
271
272	pos = pci_find_capability(dev, PCI_CAP_ID_MSI);
273	if (pos <= 0 || dev->no_msi)
274		return 0;
275
276	pci_read_config_word(dev, msi_control_reg(pos), &control);
277	if (!(control & PCI_MSI_FLAGS_ENABLE))
278		return 0;
279
280	save_state = kzalloc(sizeof(struct pci_cap_saved_state) + sizeof(u32) * 5,
281		GFP_KERNEL);
282	if (!save_state) {
283		printk(KERN_ERR "Out of memory in pci_save_msi_state\n");
284		return -ENOMEM;
285	}
286	cap = &save_state->data[0];
287
288	pci_read_config_dword(dev, pos, &cap[i++]);
289	control = cap[0] >> 16;
290	pci_read_config_dword(dev, pos + PCI_MSI_ADDRESS_LO, &cap[i++]);
291	if (control & PCI_MSI_FLAGS_64BIT) {
292		pci_read_config_dword(dev, pos + PCI_MSI_ADDRESS_HI, &cap[i++]);
293		pci_read_config_dword(dev, pos + PCI_MSI_DATA_64, &cap[i++]);
294	} else
295		pci_read_config_dword(dev, pos + PCI_MSI_DATA_32, &cap[i++]);
296	if (control & PCI_MSI_FLAGS_MASKBIT)
297		pci_read_config_dword(dev, pos + PCI_MSI_MASK_BIT, &cap[i++]);
298	save_state->cap_nr = PCI_CAP_ID_MSI;
299	pci_add_saved_cap(dev, save_state);
300	return 0;
301}
302
303static void __pci_restore_msi_state(struct pci_dev *dev)
304{
305	int i = 0, pos;
306	u16 control;
307	struct pci_cap_saved_state *save_state;
308	u32 *cap;
309
310	save_state = pci_find_saved_cap(dev, PCI_CAP_ID_MSI);
311	pos = pci_find_capability(dev, PCI_CAP_ID_MSI);
312	if (!save_state || pos <= 0)
313		return;
314	cap = &save_state->data[0];
315
316	control = cap[i++] >> 16;
317	pci_write_config_dword(dev, pos + PCI_MSI_ADDRESS_LO, cap[i++]);
318	if (control & PCI_MSI_FLAGS_64BIT) {
319		pci_write_config_dword(dev, pos + PCI_MSI_ADDRESS_HI, cap[i++]);
320		pci_write_config_dword(dev, pos + PCI_MSI_DATA_64, cap[i++]);
321	} else
322		pci_write_config_dword(dev, pos + PCI_MSI_DATA_32, cap[i++]);
323	if (control & PCI_MSI_FLAGS_MASKBIT)
324		pci_write_config_dword(dev, pos + PCI_MSI_MASK_BIT, cap[i++]);
325	pci_write_config_word(dev, pos + PCI_MSI_FLAGS, control);
326	enable_msi_mode(dev, pos, PCI_CAP_ID_MSI);
327	pci_remove_saved_cap(save_state);
328	kfree(save_state);
329}
330
331static int __pci_save_msix_state(struct pci_dev *dev)
332{
333	int pos;
334	int irq, head, tail = 0;
335	u16 control;
336	struct pci_cap_saved_state *save_state;
337
338	if (!dev->msix_enabled)
339		return 0;
340
341	pos = pci_find_capability(dev, PCI_CAP_ID_MSIX);
342	if (pos <= 0 || dev->no_msi)
343		return 0;
344
345	/* save the capability */
346	pci_read_config_word(dev, msi_control_reg(pos), &control);
347	if (!(control & PCI_MSIX_FLAGS_ENABLE))
348		return 0;
349	save_state = kzalloc(sizeof(struct pci_cap_saved_state) + sizeof(u16),
350		GFP_KERNEL);
351	if (!save_state) {
352		printk(KERN_ERR "Out of memory in pci_save_msix_state\n");
353		return -ENOMEM;
354	}
355	*((u16 *)&save_state->data[0]) = control;
356
357	/* save the table */
358	irq = head = dev->first_msi_irq;
359	while (head != tail) {
360		struct msi_desc *entry;
361
362		entry = get_irq_msi(irq);
363		read_msi_msg(irq, &entry->msg_save);
364
365		tail = entry->link.tail;
366		irq = tail;
367	}
368
369	save_state->cap_nr = PCI_CAP_ID_MSIX;
370	pci_add_saved_cap(dev, save_state);
371	return 0;
372}
373
374int pci_save_msi_state(struct pci_dev *dev)
375{
376	int rc;
377
378	rc = __pci_save_msi_state(dev);
379	if (rc)
380		return rc;
381
382	rc = __pci_save_msix_state(dev);
383
384	return rc;
385}
386
387static void __pci_restore_msix_state(struct pci_dev *dev)
388{
389	u16 save;
390	int pos;
391	int irq, head, tail = 0;
392	struct msi_desc *entry;
393	struct pci_cap_saved_state *save_state;
394
395	if (!dev->msix_enabled)
396		return;
397
398	save_state = pci_find_saved_cap(dev, PCI_CAP_ID_MSIX);
399	if (!save_state)
400		return;
401	save = *((u16 *)&save_state->data[0]);
402	pci_remove_saved_cap(save_state);
403	kfree(save_state);
404
405	pos = pci_find_capability(dev, PCI_CAP_ID_MSIX);
406	if (pos <= 0)
407		return;
408
409	/* route the table */
410	irq = head = dev->first_msi_irq;
411	while (head != tail) {
412		entry = get_irq_msi(irq);
413		write_msi_msg(irq, &entry->msg_save);
414
415		tail = entry->link.tail;
416		irq = tail;
417	}
418
419	pci_write_config_word(dev, msi_control_reg(pos), save);
420	enable_msi_mode(dev, pos, PCI_CAP_ID_MSIX);
421}
422
423void pci_restore_msi_state(struct pci_dev *dev)
424{
425	__pci_restore_msi_state(dev);
426	__pci_restore_msix_state(dev);
427}
428#endif	/* CONFIG_PM */
429
430/**
431 * msi_capability_init - configure device's MSI capability structure
432 * @dev: pointer to the pci_dev data structure of MSI device function
433 *
434 * Setup the MSI capability structure of device function with a single
435 * MSI irq, regardless of device function is capable of handling
436 * multiple messages. A return of zero indicates the successful setup
437 * of an entry zero with the new MSI irq or non-zero for otherwise.
438 **/
439static int msi_capability_init(struct pci_dev *dev)
440{
441	int status;
442	struct msi_desc *entry;
443	int pos, irq;
444	u16 control;
445
446   	pos = pci_find_capability(dev, PCI_CAP_ID_MSI);
447	pci_read_config_word(dev, msi_control_reg(pos), &control);
448	/* MSI Entry Initialization */
449	irq = create_msi_irq();
450	if (irq < 0)
451		return irq;
452
453	entry = get_irq_msi(irq);
454	entry->link.head = irq;
455	entry->link.tail = irq;
456	entry->msi_attrib.type = PCI_CAP_ID_MSI;
457	entry->msi_attrib.is_64 = is_64bit_address(control);
458	entry->msi_attrib.entry_nr = 0;
459	entry->msi_attrib.maskbit = is_mask_bit_support(control);
460	entry->msi_attrib.default_irq = dev->irq;	/* Save IOAPIC IRQ */
461	entry->msi_attrib.pos = pos;
462	if (is_mask_bit_support(control)) {
463		entry->mask_base = (void __iomem *)(long)msi_mask_bits_reg(pos,
464				is_64bit_address(control));
465	}
466	entry->dev = dev;
467	if (entry->msi_attrib.maskbit) {
468		unsigned int maskbits, temp;
469		/* All MSIs are unmasked by default, Mask them all */
470		pci_read_config_dword(dev,
471			msi_mask_bits_reg(pos, is_64bit_address(control)),
472			&maskbits);
473		temp = (1 << multi_msi_capable(control));
474		temp = ((temp - 1) & ~temp);
475		maskbits |= temp;
476		pci_write_config_dword(dev,
477			msi_mask_bits_reg(pos, is_64bit_address(control)),
478			maskbits);
479	}
480	/* Configure MSI capability structure */
481	status = arch_setup_msi_irq(irq, dev);
482	if (status < 0) {
483		destroy_msi_irq(irq);
484		return status;
485	}
486
487	dev->first_msi_irq = irq;
488	set_irq_msi(irq, entry);
489	/* Set MSI enabled bits	 */
490	enable_msi_mode(dev, pos, PCI_CAP_ID_MSI);
491
492	dev->irq = irq;
493	return 0;
494}
495
496/**
497 * msix_capability_init - configure device's MSI-X capability
498 * @dev: pointer to the pci_dev data structure of MSI-X device function
499 * @entries: pointer to an array of struct msix_entry entries
500 * @nvec: number of @entries
501 *
502 * Setup the MSI-X capability structure of device function with a
503 * single MSI-X irq. A return of zero indicates the successful setup of
504 * requested MSI-X entries with allocated irqs or non-zero for otherwise.
505 **/
506static int msix_capability_init(struct pci_dev *dev,
507				struct msix_entry *entries, int nvec)
508{
509	struct msi_desc *head = NULL, *tail = NULL, *entry = NULL;
510	int status;
511	int irq, pos, i, j, nr_entries, temp = 0;
512	unsigned long phys_addr;
513	u32 table_offset;
514 	u16 control;
515	u8 bir;
516	void __iomem *base;
517
518   	pos = pci_find_capability(dev, PCI_CAP_ID_MSIX);
519	/* Request & Map MSI-X table region */
520 	pci_read_config_word(dev, msi_control_reg(pos), &control);
521	nr_entries = multi_msix_capable(control);
522
523 	pci_read_config_dword(dev, msix_table_offset_reg(pos), &table_offset);
524	bir = (u8)(table_offset & PCI_MSIX_FLAGS_BIRMASK);
525	table_offset &= ~PCI_MSIX_FLAGS_BIRMASK;
526	phys_addr = pci_resource_start (dev, bir) + table_offset;
527	base = ioremap_nocache(phys_addr, nr_entries * PCI_MSIX_ENTRY_SIZE);
528	if (base == NULL)
529		return -ENOMEM;
530
531	/* MSI-X Table Initialization */
532	for (i = 0; i < nvec; i++) {
533		irq = create_msi_irq();
534		if (irq < 0)
535			break;
536
537		entry = get_irq_msi(irq);
538 		j = entries[i].entry;
539 		entries[i].vector = irq;
540		entry->msi_attrib.type = PCI_CAP_ID_MSIX;
541		entry->msi_attrib.is_64 = 1;
542		entry->msi_attrib.entry_nr = j;
543		entry->msi_attrib.maskbit = 1;
544		entry->msi_attrib.default_irq = dev->irq;
545		entry->msi_attrib.pos = pos;
546		entry->dev = dev;
547		entry->mask_base = base;
548		if (!head) {
549			entry->link.head = irq;
550			entry->link.tail = irq;
551			head = entry;
552		} else {
553			entry->link.head = temp;
554			entry->link.tail = tail->link.tail;
555			tail->link.tail = irq;
556			head->link.head = irq;
557		}
558		temp = irq;
559		tail = entry;
560		/* Configure MSI-X capability structure */
561		status = arch_setup_msi_irq(irq, dev);
562		if (status < 0) {
563			destroy_msi_irq(irq);
564			break;
565		}
566
567		set_irq_msi(irq, entry);
568	}
569	if (i != nvec) {
570		int avail = i - 1;
571		i--;
572		for (; i >= 0; i--) {
573			irq = (entries + i)->vector;
574			msi_free_irq(dev, irq);
575			(entries + i)->vector = 0;
576		}
577		/* If we had some success report the number of irqs
578		 * we succeeded in setting up.
579		 */
580		if (avail <= 0)
581			avail = -EBUSY;
582		return avail;
583	}
584	dev->first_msi_irq = entries[0].vector;
585	/* Set MSI-X enabled bits */
586	enable_msi_mode(dev, pos, PCI_CAP_ID_MSIX);
587
588	return 0;
589}
590
591/**
592 * pci_msi_supported - check whether MSI may be enabled on device
593 * @dev: pointer to the pci_dev data structure of MSI device function
594 *
595 * Look at global flags, the device itself, and its parent busses
596 * to return 0 if MSI are supported for the device.
597 **/
598static
599int pci_msi_supported(struct pci_dev * dev)
600{
601	struct pci_bus *bus;
602
603	/* MSI must be globally enabled and supported by the device */
604	if (!pci_msi_enable || !dev || dev->no_msi)
605		return -EINVAL;
606
607	/* Any bridge which does NOT route MSI transactions from it's
608	 * secondary bus to it's primary bus must set NO_MSI flag on
609	 * the secondary pci_bus.
610	 * We expect only arch-specific PCI host bus controller driver
611	 * or quirks for specific PCI bridges to be setting NO_MSI.
612	 */
613	for (bus = dev->bus; bus; bus = bus->parent)
614		if (bus->bus_flags & PCI_BUS_FLAGS_NO_MSI)
615			return -EINVAL;
616
617	return 0;
618}
619
620/**
621 * pci_enable_msi - configure device's MSI capability structure
622 * @dev: pointer to the pci_dev data structure of MSI device function
623 *
624 * Setup the MSI capability structure of device function with
625 * a single MSI irq upon its software driver call to request for
626 * MSI mode enabled on its hardware device function. A return of zero
627 * indicates the successful setup of an entry zero with the new MSI
628 * irq or non-zero for otherwise.
629 **/
630int pci_enable_msi(struct pci_dev* dev)
631{
632	int pos, status;
633
634	if (pci_msi_supported(dev) < 0)
635		return -EINVAL;
636
637	status = msi_init();
638	if (status < 0)
639		return status;
640
641	pos = pci_find_capability(dev, PCI_CAP_ID_MSI);
642	if (!pos)
643		return -EINVAL;
644
645	WARN_ON(!!dev->msi_enabled);
646
647	/* Check whether driver already requested for MSI-X irqs */
648	pos = pci_find_capability(dev, PCI_CAP_ID_MSIX);
649	if (pos > 0 && dev->msix_enabled) {
650			printk(KERN_INFO "PCI: %s: Can't enable MSI.  "
651			       "Device already has MSI-X enabled\n",
652			       pci_name(dev));
653			return -EINVAL;
654	}
655	status = msi_capability_init(dev);
656	return status;
657}
658
659void pci_disable_msi(struct pci_dev* dev)
660{
661	struct msi_desc *entry;
662	int pos, default_irq;
663	u16 control;
664
665	if (!pci_msi_enable)
666		return;
667	if (!dev)
668		return;
669
670	if (!dev->msi_enabled)
671		return;
672
673	pos = pci_find_capability(dev, PCI_CAP_ID_MSI);
674	if (!pos)
675		return;
676
677	pci_read_config_word(dev, msi_control_reg(pos), &control);
678	if (!(control & PCI_MSI_FLAGS_ENABLE))
679		return;
680
681
682	disable_msi_mode(dev, pos, PCI_CAP_ID_MSI);
683
684	entry = get_irq_msi(dev->first_msi_irq);
685	if (!entry || !entry->dev || entry->msi_attrib.type != PCI_CAP_ID_MSI) {
686		return;
687	}
688	if (irq_has_action(dev->first_msi_irq)) {
689		printk(KERN_WARNING "PCI: %s: pci_disable_msi() called without "
690		       "free_irq() on MSI irq %d\n",
691		       pci_name(dev), dev->first_msi_irq);
692		BUG_ON(irq_has_action(dev->first_msi_irq));
693	} else {
694		default_irq = entry->msi_attrib.default_irq;
695		msi_free_irq(dev, dev->first_msi_irq);
696
697		/* Restore dev->irq to its default pin-assertion irq */
698		dev->irq = default_irq;
699	}
700	dev->first_msi_irq = 0;
701}
702
703static int msi_free_irq(struct pci_dev* dev, int irq)
704{
705	struct msi_desc *entry;
706	int head, entry_nr, type;
707	void __iomem *base;
708
709	arch_teardown_msi_irq(irq);
710
711	entry = get_irq_msi(irq);
712	if (!entry || entry->dev != dev) {
713		return -EINVAL;
714	}
715	type = entry->msi_attrib.type;
716	entry_nr = entry->msi_attrib.entry_nr;
717	head = entry->link.head;
718	base = entry->mask_base;
719	get_irq_msi(entry->link.head)->link.tail = entry->link.tail;
720	get_irq_msi(entry->link.tail)->link.head = entry->link.head;
721	entry->dev = NULL;
722
723	destroy_msi_irq(irq);
724
725	if (type == PCI_CAP_ID_MSIX) {
726		writel(1, base + entry_nr * PCI_MSIX_ENTRY_SIZE +
727			PCI_MSIX_ENTRY_VECTOR_CTRL_OFFSET);
728
729		if (head == irq)
730			iounmap(base);
731	}
732
733	return 0;
734}
735
736/**
737 * pci_enable_msix - configure device's MSI-X capability structure
738 * @dev: pointer to the pci_dev data structure of MSI-X device function
739 * @entries: pointer to an array of MSI-X entries
740 * @nvec: number of MSI-X irqs requested for allocation by device driver
741 *
742 * Setup the MSI-X capability structure of device function with the number
743 * of requested irqs upon its software driver call to request for
744 * MSI-X mode enabled on its hardware device function. A return of zero
745 * indicates the successful configuration of MSI-X capability structure
746 * with new allocated MSI-X irqs. A return of < 0 indicates a failure.
747 * Or a return of > 0 indicates that driver request is exceeding the number
748 * of irqs available. Driver should use the returned value to re-send
749 * its request.
750 **/
751int pci_enable_msix(struct pci_dev* dev, struct msix_entry *entries, int nvec)
752{
753	int status, pos, nr_entries;
754	int i, j;
755	u16 control;
756
757	if (!entries || pci_msi_supported(dev) < 0)
758 		return -EINVAL;
759
760	status = msi_init();
761	if (status < 0)
762		return status;
763
764	pos = pci_find_capability(dev, PCI_CAP_ID_MSIX);
765	if (!pos)
766 		return -EINVAL;
767
768	pci_read_config_word(dev, msi_control_reg(pos), &control);
769	nr_entries = multi_msix_capable(control);
770	if (nvec > nr_entries)
771		return -EINVAL;
772
773	/* Check for any invalid entries */
774	for (i = 0; i < nvec; i++) {
775		if (entries[i].entry >= nr_entries)
776			return -EINVAL;		/* invalid entry */
777		for (j = i + 1; j < nvec; j++) {
778			if (entries[i].entry == entries[j].entry)
779				return -EINVAL;	/* duplicate entry */
780		}
781	}
782	WARN_ON(!!dev->msix_enabled);
783
784	/* Check whether driver already requested for MSI irq */
785   	if (pci_find_capability(dev, PCI_CAP_ID_MSI) > 0 &&
786		dev->msi_enabled) {
787		printk(KERN_INFO "PCI: %s: Can't enable MSI-X.  "
788		       "Device already has an MSI irq assigned\n",
789		       pci_name(dev));
790		return -EINVAL;
791	}
792	status = msix_capability_init(dev, entries, nvec);
793	return status;
794}
795
796void pci_disable_msix(struct pci_dev* dev)
797{
798	int irq, head, tail = 0, warning = 0;
799	int pos;
800	u16 control;
801
802	if (!pci_msi_enable)
803		return;
804	if (!dev)
805		return;
806
807	if (!dev->msix_enabled)
808		return;
809
810	pos = pci_find_capability(dev, PCI_CAP_ID_MSIX);
811	if (!pos)
812		return;
813
814	pci_read_config_word(dev, msi_control_reg(pos), &control);
815	if (!(control & PCI_MSIX_FLAGS_ENABLE))
816		return;
817
818	disable_msi_mode(dev, pos, PCI_CAP_ID_MSIX);
819
820	irq = head = dev->first_msi_irq;
821	while (head != tail) {
822		tail = get_irq_msi(irq)->link.tail;
823		if (irq_has_action(irq))
824			warning = 1;
825		else if (irq != head)	/* Release MSI-X irq */
826			msi_free_irq(dev, irq);
827		irq = tail;
828	}
829	msi_free_irq(dev, irq);
830	if (warning) {
831		printk(KERN_WARNING "PCI: %s: pci_disable_msix() called without "
832			"free_irq() on all MSI-X irqs\n",
833			pci_name(dev));
834		BUG_ON(warning > 0);
835	}
836	dev->first_msi_irq = 0;
837}
838
839/**
840 * msi_remove_pci_irq_vectors - reclaim MSI(X) irqs to unused state
841 * @dev: pointer to the pci_dev data structure of MSI(X) device function
842 *
843 * Being called during hotplug remove, from which the device function
844 * is hot-removed. All previous assigned MSI/MSI-X irqs, if
845 * allocated for this device function, are reclaimed to unused state,
846 * which may be used later on.
847 **/
848void msi_remove_pci_irq_vectors(struct pci_dev* dev)
849{
850	if (!pci_msi_enable || !dev)
851 		return;
852
853	if (dev->msi_enabled) {
854		if (irq_has_action(dev->first_msi_irq)) {
855			printk(KERN_WARNING "PCI: %s: msi_remove_pci_irq_vectors() "
856			       "called without free_irq() on MSI irq %d\n",
857			       pci_name(dev), dev->first_msi_irq);
858			BUG_ON(irq_has_action(dev->first_msi_irq));
859		} else /* Release MSI irq assigned to this device */
860			msi_free_irq(dev, dev->first_msi_irq);
861	}
862	if (dev->msix_enabled) {
863		int irq, head, tail = 0, warning = 0;
864		void __iomem *base = NULL;
865
866		irq = head = dev->first_msi_irq;
867		while (head != tail) {
868			tail = get_irq_msi(irq)->link.tail;
869			base = get_irq_msi(irq)->mask_base;
870			if (irq_has_action(irq))
871				warning = 1;
872			else if (irq != head) /* Release MSI-X irq */
873				msi_free_irq(dev, irq);
874			irq = tail;
875		}
876		msi_free_irq(dev, irq);
877		if (warning) {
878			iounmap(base);
879			printk(KERN_WARNING "PCI: %s: msi_remove_pci_irq_vectors() "
880			       "called without free_irq() on all MSI-X irqs\n",
881			       pci_name(dev));
882			BUG_ON(warning > 0);
883		}
884	}
885}
886
887void pci_no_msi(void)
888{
889	pci_msi_enable = 0;
890}
891
892EXPORT_SYMBOL(pci_enable_msi);
893EXPORT_SYMBOL(pci_disable_msi);
894EXPORT_SYMBOL(pci_enable_msix);
895EXPORT_SYMBOL(pci_disable_msix);
896