msi.c revision 88187dfa4d8bb565df762f272511d2c91e427e0d
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 DEFINE_SPINLOCK(msi_lock);
28static struct msi_desc* msi_desc[NR_IRQS] = { [0 ... NR_IRQS-1] = NULL };
29static struct kmem_cache* msi_cachep;
30
31static int pci_msi_enable = 1;
32
33static int msi_cache_init(void)
34{
35	msi_cachep = kmem_cache_create("msi_cache", sizeof(struct msi_desc),
36					0, SLAB_HWCACHE_ALIGN, NULL, NULL);
37	if (!msi_cachep)
38		return -ENOMEM;
39
40	return 0;
41}
42
43static void msi_set_mask_bit(unsigned int irq, int flag)
44{
45	struct msi_desc *entry;
46
47	entry = msi_desc[irq];
48	BUG_ON(!entry || !entry->dev);
49	switch (entry->msi_attrib.type) {
50	case PCI_CAP_ID_MSI:
51		if (entry->msi_attrib.maskbit) {
52			int pos;
53			u32 mask_bits;
54
55			pos = (long)entry->mask_base;
56			pci_read_config_dword(entry->dev, pos, &mask_bits);
57			mask_bits &= ~(1);
58			mask_bits |= flag;
59			pci_write_config_dword(entry->dev, pos, mask_bits);
60		}
61		break;
62	case PCI_CAP_ID_MSIX:
63	{
64		int offset = entry->msi_attrib.entry_nr * PCI_MSIX_ENTRY_SIZE +
65			PCI_MSIX_ENTRY_VECTOR_CTRL_OFFSET;
66		writel(flag, entry->mask_base + offset);
67		break;
68	}
69	default:
70		BUG();
71		break;
72	}
73}
74
75void read_msi_msg(unsigned int irq, struct msi_msg *msg)
76{
77	struct msi_desc *entry = get_irq_data(irq);
78	switch(entry->msi_attrib.type) {
79	case PCI_CAP_ID_MSI:
80	{
81		struct pci_dev *dev = entry->dev;
82		int pos = entry->msi_attrib.pos;
83		u16 data;
84
85		pci_read_config_dword(dev, msi_lower_address_reg(pos),
86					&msg->address_lo);
87		if (entry->msi_attrib.is_64) {
88			pci_read_config_dword(dev, msi_upper_address_reg(pos),
89						&msg->address_hi);
90			pci_read_config_word(dev, msi_data_reg(pos, 1), &data);
91		} else {
92			msg->address_hi = 0;
93			pci_read_config_word(dev, msi_data_reg(pos, 1), &data);
94		}
95		msg->data = data;
96		break;
97	}
98	case PCI_CAP_ID_MSIX:
99	{
100		void __iomem *base;
101		base = entry->mask_base +
102			entry->msi_attrib.entry_nr * PCI_MSIX_ENTRY_SIZE;
103
104		msg->address_lo = readl(base + PCI_MSIX_ENTRY_LOWER_ADDR_OFFSET);
105		msg->address_hi = readl(base + PCI_MSIX_ENTRY_UPPER_ADDR_OFFSET);
106		msg->data = readl(base + PCI_MSIX_ENTRY_DATA_OFFSET);
107 		break;
108 	}
109 	default:
110		BUG();
111	}
112}
113
114void write_msi_msg(unsigned int irq, struct msi_msg *msg)
115{
116	struct msi_desc *entry = get_irq_data(irq);
117	switch (entry->msi_attrib.type) {
118	case PCI_CAP_ID_MSI:
119	{
120		struct pci_dev *dev = entry->dev;
121		int pos = entry->msi_attrib.pos;
122
123		pci_write_config_dword(dev, msi_lower_address_reg(pos),
124					msg->address_lo);
125		if (entry->msi_attrib.is_64) {
126			pci_write_config_dword(dev, msi_upper_address_reg(pos),
127						msg->address_hi);
128			pci_write_config_word(dev, msi_data_reg(pos, 1),
129						msg->data);
130		} else {
131			pci_write_config_word(dev, msi_data_reg(pos, 0),
132						msg->data);
133		}
134		break;
135	}
136	case PCI_CAP_ID_MSIX:
137	{
138		void __iomem *base;
139		base = entry->mask_base +
140			entry->msi_attrib.entry_nr * PCI_MSIX_ENTRY_SIZE;
141
142		writel(msg->address_lo,
143			base + PCI_MSIX_ENTRY_LOWER_ADDR_OFFSET);
144		writel(msg->address_hi,
145			base + PCI_MSIX_ENTRY_UPPER_ADDR_OFFSET);
146		writel(msg->data, base + PCI_MSIX_ENTRY_DATA_OFFSET);
147		break;
148	}
149	default:
150		BUG();
151	}
152}
153
154void mask_msi_irq(unsigned int irq)
155{
156	msi_set_mask_bit(irq, 1);
157}
158
159void unmask_msi_irq(unsigned int irq)
160{
161	msi_set_mask_bit(irq, 0);
162}
163
164static int msi_free_irq(struct pci_dev* dev, int irq);
165
166static int msi_init(void)
167{
168	static int status = -ENOMEM;
169
170	if (!status)
171		return status;
172
173	status = msi_cache_init();
174	if (status < 0) {
175		pci_msi_enable = 0;
176		printk(KERN_WARNING "PCI: MSI cache init failed\n");
177		return status;
178	}
179
180	return status;
181}
182
183static struct msi_desc* alloc_msi_entry(void)
184{
185	struct msi_desc *entry;
186
187	entry = kmem_cache_zalloc(msi_cachep, GFP_KERNEL);
188	if (!entry)
189		return NULL;
190
191	entry->link.tail = entry->link.head = 0;	/* single message */
192	entry->dev = NULL;
193
194	return entry;
195}
196
197static void attach_msi_entry(struct msi_desc *entry, int irq)
198{
199	unsigned long flags;
200
201	spin_lock_irqsave(&msi_lock, flags);
202	msi_desc[irq] = entry;
203	spin_unlock_irqrestore(&msi_lock, flags);
204}
205
206static int create_msi_irq(void)
207{
208	struct msi_desc *entry;
209	int irq;
210
211	entry = alloc_msi_entry();
212	if (!entry)
213		return -ENOMEM;
214
215	irq = create_irq();
216	if (irq < 0) {
217		kmem_cache_free(msi_cachep, entry);
218		return -EBUSY;
219	}
220
221	set_irq_data(irq, entry);
222
223	return irq;
224}
225
226static void destroy_msi_irq(unsigned int irq)
227{
228	struct msi_desc *entry;
229
230	entry = get_irq_data(irq);
231	set_irq_chip(irq, NULL);
232	set_irq_data(irq, NULL);
233	destroy_irq(irq);
234	kmem_cache_free(msi_cachep, entry);
235}
236
237static void enable_msi_mode(struct pci_dev *dev, int pos, int type)
238{
239	u16 control;
240
241	pci_read_config_word(dev, msi_control_reg(pos), &control);
242	if (type == PCI_CAP_ID_MSI) {
243		/* Set enabled bits to single MSI & enable MSI_enable bit */
244		msi_enable(control, 1);
245		pci_write_config_word(dev, msi_control_reg(pos), control);
246		dev->msi_enabled = 1;
247	} else {
248		msix_enable(control);
249		pci_write_config_word(dev, msi_control_reg(pos), control);
250		dev->msix_enabled = 1;
251	}
252
253	pci_intx(dev, 0);  /* disable intx */
254}
255
256void disable_msi_mode(struct pci_dev *dev, int pos, int type)
257{
258	u16 control;
259
260	pci_read_config_word(dev, msi_control_reg(pos), &control);
261	if (type == PCI_CAP_ID_MSI) {
262		/* Set enabled bits to single MSI & enable MSI_enable bit */
263		msi_disable(control);
264		pci_write_config_word(dev, msi_control_reg(pos), control);
265		dev->msi_enabled = 0;
266	} else {
267		msix_disable(control);
268		pci_write_config_word(dev, msi_control_reg(pos), control);
269		dev->msix_enabled = 0;
270	}
271
272	pci_intx(dev, 1);  /* enable intx */
273}
274
275static int msi_lookup_irq(struct pci_dev *dev, int type)
276{
277	int irq;
278	unsigned long flags;
279
280	spin_lock_irqsave(&msi_lock, flags);
281	for (irq = 0; irq < NR_IRQS; irq++) {
282		if (!msi_desc[irq] || msi_desc[irq]->dev != dev ||
283			msi_desc[irq]->msi_attrib.type != type ||
284			msi_desc[irq]->msi_attrib.default_irq != dev->irq)
285			continue;
286		spin_unlock_irqrestore(&msi_lock, flags);
287		/* This pre-assigned MSI irq for this device
288		   already exists. Override dev->irq with this irq */
289		dev->irq = irq;
290		return 0;
291	}
292	spin_unlock_irqrestore(&msi_lock, flags);
293
294	return -EACCES;
295}
296
297void pci_scan_msi_device(struct pci_dev *dev)
298{
299	if (!dev)
300		return;
301}
302
303#ifdef CONFIG_PM
304int pci_save_msi_state(struct pci_dev *dev)
305{
306	int pos, i = 0;
307	u16 control;
308	struct pci_cap_saved_state *save_state;
309	u32 *cap;
310
311	pos = pci_find_capability(dev, PCI_CAP_ID_MSI);
312	if (pos <= 0 || dev->no_msi)
313		return 0;
314
315	pci_read_config_word(dev, msi_control_reg(pos), &control);
316	if (!(control & PCI_MSI_FLAGS_ENABLE))
317		return 0;
318
319	save_state = kzalloc(sizeof(struct pci_cap_saved_state) + sizeof(u32) * 5,
320		GFP_KERNEL);
321	if (!save_state) {
322		printk(KERN_ERR "Out of memory in pci_save_msi_state\n");
323		return -ENOMEM;
324	}
325	cap = &save_state->data[0];
326
327	pci_read_config_dword(dev, pos, &cap[i++]);
328	control = cap[0] >> 16;
329	pci_read_config_dword(dev, pos + PCI_MSI_ADDRESS_LO, &cap[i++]);
330	if (control & PCI_MSI_FLAGS_64BIT) {
331		pci_read_config_dword(dev, pos + PCI_MSI_ADDRESS_HI, &cap[i++]);
332		pci_read_config_dword(dev, pos + PCI_MSI_DATA_64, &cap[i++]);
333	} else
334		pci_read_config_dword(dev, pos + PCI_MSI_DATA_32, &cap[i++]);
335	if (control & PCI_MSI_FLAGS_MASKBIT)
336		pci_read_config_dword(dev, pos + PCI_MSI_MASK_BIT, &cap[i++]);
337	save_state->cap_nr = PCI_CAP_ID_MSI;
338	pci_add_saved_cap(dev, save_state);
339	return 0;
340}
341
342void pci_restore_msi_state(struct pci_dev *dev)
343{
344	int i = 0, pos;
345	u16 control;
346	struct pci_cap_saved_state *save_state;
347	u32 *cap;
348
349	save_state = pci_find_saved_cap(dev, PCI_CAP_ID_MSI);
350	pos = pci_find_capability(dev, PCI_CAP_ID_MSI);
351	if (!save_state || pos <= 0)
352		return;
353	cap = &save_state->data[0];
354
355	control = cap[i++] >> 16;
356	pci_write_config_dword(dev, pos + PCI_MSI_ADDRESS_LO, cap[i++]);
357	if (control & PCI_MSI_FLAGS_64BIT) {
358		pci_write_config_dword(dev, pos + PCI_MSI_ADDRESS_HI, cap[i++]);
359		pci_write_config_dword(dev, pos + PCI_MSI_DATA_64, cap[i++]);
360	} else
361		pci_write_config_dword(dev, pos + PCI_MSI_DATA_32, cap[i++]);
362	if (control & PCI_MSI_FLAGS_MASKBIT)
363		pci_write_config_dword(dev, pos + PCI_MSI_MASK_BIT, cap[i++]);
364	pci_write_config_word(dev, pos + PCI_MSI_FLAGS, control);
365	enable_msi_mode(dev, pos, PCI_CAP_ID_MSI);
366	pci_remove_saved_cap(save_state);
367	kfree(save_state);
368}
369
370int pci_save_msix_state(struct pci_dev *dev)
371{
372	int pos;
373	int temp;
374	int irq, head, tail = 0;
375	u16 control;
376	struct pci_cap_saved_state *save_state;
377
378	pos = pci_find_capability(dev, PCI_CAP_ID_MSIX);
379	if (pos <= 0 || dev->no_msi)
380		return 0;
381
382	/* save the capability */
383	pci_read_config_word(dev, msi_control_reg(pos), &control);
384	if (!(control & PCI_MSIX_FLAGS_ENABLE))
385		return 0;
386	save_state = kzalloc(sizeof(struct pci_cap_saved_state) + sizeof(u16),
387		GFP_KERNEL);
388	if (!save_state) {
389		printk(KERN_ERR "Out of memory in pci_save_msix_state\n");
390		return -ENOMEM;
391	}
392	*((u16 *)&save_state->data[0]) = control;
393
394	/* save the table */
395	temp = dev->irq;
396	if (msi_lookup_irq(dev, PCI_CAP_ID_MSIX)) {
397		kfree(save_state);
398		return -EINVAL;
399	}
400
401	irq = head = dev->irq;
402	while (head != tail) {
403		struct msi_desc *entry;
404
405		entry = msi_desc[irq];
406		read_msi_msg(irq, &entry->msg_save);
407
408		tail = msi_desc[irq]->link.tail;
409		irq = tail;
410	}
411	dev->irq = temp;
412
413	save_state->cap_nr = PCI_CAP_ID_MSIX;
414	pci_add_saved_cap(dev, save_state);
415	return 0;
416}
417
418void pci_restore_msix_state(struct pci_dev *dev)
419{
420	u16 save;
421	int pos;
422	int irq, head, tail = 0;
423	struct msi_desc *entry;
424	int temp;
425	struct pci_cap_saved_state *save_state;
426
427	save_state = pci_find_saved_cap(dev, PCI_CAP_ID_MSIX);
428	if (!save_state)
429		return;
430	save = *((u16 *)&save_state->data[0]);
431	pci_remove_saved_cap(save_state);
432	kfree(save_state);
433
434	pos = pci_find_capability(dev, PCI_CAP_ID_MSIX);
435	if (pos <= 0)
436		return;
437
438	/* route the table */
439	temp = dev->irq;
440	if (msi_lookup_irq(dev, PCI_CAP_ID_MSIX))
441		return;
442	irq = head = dev->irq;
443	while (head != tail) {
444		entry = msi_desc[irq];
445		write_msi_msg(irq, &entry->msg_save);
446
447		tail = msi_desc[irq]->link.tail;
448		irq = tail;
449	}
450	dev->irq = temp;
451
452	pci_write_config_word(dev, msi_control_reg(pos), save);
453	enable_msi_mode(dev, pos, PCI_CAP_ID_MSIX);
454}
455#endif	/* CONFIG_PM */
456
457/**
458 * msi_capability_init - configure device's MSI capability structure
459 * @dev: pointer to the pci_dev data structure of MSI device function
460 *
461 * Setup the MSI capability structure of device function with a single
462 * MSI irq, regardless of device function is capable of handling
463 * multiple messages. A return of zero indicates the successful setup
464 * of an entry zero with the new MSI irq or non-zero for otherwise.
465 **/
466static int msi_capability_init(struct pci_dev *dev)
467{
468	int status;
469	struct msi_desc *entry;
470	int pos, irq;
471	u16 control;
472
473   	pos = pci_find_capability(dev, PCI_CAP_ID_MSI);
474	pci_read_config_word(dev, msi_control_reg(pos), &control);
475	/* MSI Entry Initialization */
476	irq = create_msi_irq();
477	if (irq < 0)
478		return irq;
479
480	entry = get_irq_data(irq);
481	entry->link.head = irq;
482	entry->link.tail = irq;
483	entry->msi_attrib.type = PCI_CAP_ID_MSI;
484	entry->msi_attrib.is_64 = is_64bit_address(control);
485	entry->msi_attrib.entry_nr = 0;
486	entry->msi_attrib.maskbit = is_mask_bit_support(control);
487	entry->msi_attrib.default_irq = dev->irq;	/* Save IOAPIC IRQ */
488	entry->msi_attrib.pos = pos;
489	if (is_mask_bit_support(control)) {
490		entry->mask_base = (void __iomem *)(long)msi_mask_bits_reg(pos,
491				is_64bit_address(control));
492	}
493	entry->dev = dev;
494	if (entry->msi_attrib.maskbit) {
495		unsigned int maskbits, temp;
496		/* All MSIs are unmasked by default, Mask them all */
497		pci_read_config_dword(dev,
498			msi_mask_bits_reg(pos, is_64bit_address(control)),
499			&maskbits);
500		temp = (1 << multi_msi_capable(control));
501		temp = ((temp - 1) & ~temp);
502		maskbits |= temp;
503		pci_write_config_dword(dev,
504			msi_mask_bits_reg(pos, is_64bit_address(control)),
505			maskbits);
506	}
507	/* Configure MSI capability structure */
508	status = arch_setup_msi_irq(irq, dev);
509	if (status < 0) {
510		destroy_msi_irq(irq);
511		return status;
512	}
513
514	attach_msi_entry(entry, irq);
515	/* Set MSI enabled bits	 */
516	enable_msi_mode(dev, pos, PCI_CAP_ID_MSI);
517
518	dev->irq = irq;
519	return 0;
520}
521
522/**
523 * msix_capability_init - configure device's MSI-X capability
524 * @dev: pointer to the pci_dev data structure of MSI-X device function
525 * @entries: pointer to an array of struct msix_entry entries
526 * @nvec: number of @entries
527 *
528 * Setup the MSI-X capability structure of device function with a
529 * single MSI-X irq. A return of zero indicates the successful setup of
530 * requested MSI-X entries with allocated irqs or non-zero for otherwise.
531 **/
532static int msix_capability_init(struct pci_dev *dev,
533				struct msix_entry *entries, int nvec)
534{
535	struct msi_desc *head = NULL, *tail = NULL, *entry = NULL;
536	int status;
537	int irq, pos, i, j, nr_entries, temp = 0;
538	unsigned long phys_addr;
539	u32 table_offset;
540 	u16 control;
541	u8 bir;
542	void __iomem *base;
543
544   	pos = pci_find_capability(dev, PCI_CAP_ID_MSIX);
545	/* Request & Map MSI-X table region */
546 	pci_read_config_word(dev, msi_control_reg(pos), &control);
547	nr_entries = multi_msix_capable(control);
548
549 	pci_read_config_dword(dev, msix_table_offset_reg(pos), &table_offset);
550	bir = (u8)(table_offset & PCI_MSIX_FLAGS_BIRMASK);
551	table_offset &= ~PCI_MSIX_FLAGS_BIRMASK;
552	phys_addr = pci_resource_start (dev, bir) + table_offset;
553	base = ioremap_nocache(phys_addr, nr_entries * PCI_MSIX_ENTRY_SIZE);
554	if (base == NULL)
555		return -ENOMEM;
556
557	/* MSI-X Table Initialization */
558	for (i = 0; i < nvec; i++) {
559		irq = create_msi_irq();
560		if (irq < 0)
561			break;
562
563		entry = get_irq_data(irq);
564 		j = entries[i].entry;
565 		entries[i].vector = irq;
566		entry->msi_attrib.type = PCI_CAP_ID_MSIX;
567		entry->msi_attrib.is_64 = 1;
568		entry->msi_attrib.entry_nr = j;
569		entry->msi_attrib.maskbit = 1;
570		entry->msi_attrib.default_irq = dev->irq;
571		entry->msi_attrib.pos = pos;
572		entry->dev = dev;
573		entry->mask_base = base;
574		if (!head) {
575			entry->link.head = irq;
576			entry->link.tail = irq;
577			head = entry;
578		} else {
579			entry->link.head = temp;
580			entry->link.tail = tail->link.tail;
581			tail->link.tail = irq;
582			head->link.head = irq;
583		}
584		temp = irq;
585		tail = entry;
586		/* Configure MSI-X capability structure */
587		status = arch_setup_msi_irq(irq, dev);
588		if (status < 0) {
589			destroy_msi_irq(irq);
590			break;
591		}
592
593		attach_msi_entry(entry, irq);
594	}
595	if (i != nvec) {
596		int avail = i - 1;
597		i--;
598		for (; i >= 0; i--) {
599			irq = (entries + i)->vector;
600			msi_free_irq(dev, irq);
601			(entries + i)->vector = 0;
602		}
603		/* If we had some success report the number of irqs
604		 * we succeeded in setting up.
605		 */
606		if (avail <= 0)
607			avail = -EBUSY;
608		return avail;
609	}
610	/* Set MSI-X enabled bits */
611	enable_msi_mode(dev, pos, PCI_CAP_ID_MSIX);
612
613	return 0;
614}
615
616/**
617 * pci_msi_supported - check whether MSI may be enabled on device
618 * @dev: pointer to the pci_dev data structure of MSI device function
619 *
620 * Look at global flags, the device itself, and its parent busses
621 * to return 0 if MSI are supported for the device.
622 **/
623static
624int pci_msi_supported(struct pci_dev * dev)
625{
626	struct pci_bus *bus;
627
628	/* MSI must be globally enabled and supported by the device */
629	if (!pci_msi_enable || !dev || dev->no_msi)
630		return -EINVAL;
631
632	/* Any bridge which does NOT route MSI transactions from it's
633	 * secondary bus to it's primary bus must set NO_MSI flag on
634	 * the secondary pci_bus.
635	 * We expect only arch-specific PCI host bus controller driver
636	 * or quirks for specific PCI bridges to be setting NO_MSI.
637	 */
638	for (bus = dev->bus; bus; bus = bus->parent)
639		if (bus->bus_flags & PCI_BUS_FLAGS_NO_MSI)
640			return -EINVAL;
641
642	return 0;
643}
644
645/**
646 * pci_enable_msi - configure device's MSI capability structure
647 * @dev: pointer to the pci_dev data structure of MSI device function
648 *
649 * Setup the MSI capability structure of device function with
650 * a single MSI irq upon its software driver call to request for
651 * MSI mode enabled on its hardware device function. A return of zero
652 * indicates the successful setup of an entry zero with the new MSI
653 * irq or non-zero for otherwise.
654 **/
655int pci_enable_msi(struct pci_dev* dev)
656{
657	int pos, temp, status;
658
659	if (pci_msi_supported(dev) < 0)
660		return -EINVAL;
661
662	temp = dev->irq;
663
664	status = msi_init();
665	if (status < 0)
666		return status;
667
668	pos = pci_find_capability(dev, PCI_CAP_ID_MSI);
669	if (!pos)
670		return -EINVAL;
671
672	WARN_ON(!msi_lookup_irq(dev, PCI_CAP_ID_MSI));
673
674	/* Check whether driver already requested for MSI-X irqs */
675	pos = pci_find_capability(dev, PCI_CAP_ID_MSIX);
676	if (pos > 0 && !msi_lookup_irq(dev, PCI_CAP_ID_MSIX)) {
677			printk(KERN_INFO "PCI: %s: Can't enable MSI.  "
678			       "Device already has MSI-X irq assigned\n",
679			       pci_name(dev));
680			dev->irq = temp;
681			return -EINVAL;
682	}
683	status = msi_capability_init(dev);
684	return status;
685}
686
687void pci_disable_msi(struct pci_dev* dev)
688{
689	struct msi_desc *entry;
690	int pos, default_irq;
691	u16 control;
692	unsigned long flags;
693
694	if (!pci_msi_enable)
695		return;
696	if (!dev)
697		return;
698
699	pos = pci_find_capability(dev, PCI_CAP_ID_MSI);
700	if (!pos)
701		return;
702
703	pci_read_config_word(dev, msi_control_reg(pos), &control);
704	if (!(control & PCI_MSI_FLAGS_ENABLE))
705		return;
706
707	disable_msi_mode(dev, pos, PCI_CAP_ID_MSI);
708
709	spin_lock_irqsave(&msi_lock, flags);
710	entry = msi_desc[dev->irq];
711	if (!entry || !entry->dev || entry->msi_attrib.type != PCI_CAP_ID_MSI) {
712		spin_unlock_irqrestore(&msi_lock, flags);
713		return;
714	}
715	if (irq_has_action(dev->irq)) {
716		spin_unlock_irqrestore(&msi_lock, flags);
717		printk(KERN_WARNING "PCI: %s: pci_disable_msi() called without "
718		       "free_irq() on MSI irq %d\n",
719		       pci_name(dev), dev->irq);
720		BUG_ON(irq_has_action(dev->irq));
721	} else {
722		default_irq = entry->msi_attrib.default_irq;
723		spin_unlock_irqrestore(&msi_lock, flags);
724		msi_free_irq(dev, dev->irq);
725
726		/* Restore dev->irq to its default pin-assertion irq */
727		dev->irq = default_irq;
728	}
729}
730
731static int msi_free_irq(struct pci_dev* dev, int irq)
732{
733	struct msi_desc *entry;
734	int head, entry_nr, type;
735	void __iomem *base;
736	unsigned long flags;
737
738	arch_teardown_msi_irq(irq);
739
740	spin_lock_irqsave(&msi_lock, flags);
741	entry = msi_desc[irq];
742	if (!entry || entry->dev != dev) {
743		spin_unlock_irqrestore(&msi_lock, flags);
744		return -EINVAL;
745	}
746	type = entry->msi_attrib.type;
747	entry_nr = entry->msi_attrib.entry_nr;
748	head = entry->link.head;
749	base = entry->mask_base;
750	msi_desc[entry->link.head]->link.tail = entry->link.tail;
751	msi_desc[entry->link.tail]->link.head = entry->link.head;
752	entry->dev = NULL;
753	msi_desc[irq] = NULL;
754	spin_unlock_irqrestore(&msi_lock, flags);
755
756	destroy_msi_irq(irq);
757
758	if (type == PCI_CAP_ID_MSIX) {
759		writel(1, base + entry_nr * PCI_MSIX_ENTRY_SIZE +
760			PCI_MSIX_ENTRY_VECTOR_CTRL_OFFSET);
761
762		if (head == irq)
763			iounmap(base);
764	}
765
766	return 0;
767}
768
769/**
770 * pci_enable_msix - configure device's MSI-X capability structure
771 * @dev: pointer to the pci_dev data structure of MSI-X device function
772 * @entries: pointer to an array of MSI-X entries
773 * @nvec: number of MSI-X irqs requested for allocation by device driver
774 *
775 * Setup the MSI-X capability structure of device function with the number
776 * of requested irqs upon its software driver call to request for
777 * MSI-X mode enabled on its hardware device function. A return of zero
778 * indicates the successful configuration of MSI-X capability structure
779 * with new allocated MSI-X irqs. A return of < 0 indicates a failure.
780 * Or a return of > 0 indicates that driver request is exceeding the number
781 * of irqs available. Driver should use the returned value to re-send
782 * its request.
783 **/
784int pci_enable_msix(struct pci_dev* dev, struct msix_entry *entries, int nvec)
785{
786	int status, pos, nr_entries;
787	int i, j, temp;
788	u16 control;
789
790	if (!entries || pci_msi_supported(dev) < 0)
791 		return -EINVAL;
792
793	status = msi_init();
794	if (status < 0)
795		return status;
796
797	pos = pci_find_capability(dev, PCI_CAP_ID_MSIX);
798	if (!pos)
799 		return -EINVAL;
800
801	pci_read_config_word(dev, msi_control_reg(pos), &control);
802	nr_entries = multi_msix_capable(control);
803	if (nvec > nr_entries)
804		return -EINVAL;
805
806	/* Check for any invalid entries */
807	for (i = 0; i < nvec; i++) {
808		if (entries[i].entry >= nr_entries)
809			return -EINVAL;		/* invalid entry */
810		for (j = i + 1; j < nvec; j++) {
811			if (entries[i].entry == entries[j].entry)
812				return -EINVAL;	/* duplicate entry */
813		}
814	}
815	temp = dev->irq;
816	WARN_ON(!msi_lookup_irq(dev, PCI_CAP_ID_MSIX));
817
818	/* Check whether driver already requested for MSI irq */
819   	if (pci_find_capability(dev, PCI_CAP_ID_MSI) > 0 &&
820		!msi_lookup_irq(dev, PCI_CAP_ID_MSI)) {
821		printk(KERN_INFO "PCI: %s: Can't enable MSI-X.  "
822		       "Device already has an MSI irq assigned\n",
823		       pci_name(dev));
824		dev->irq = temp;
825		return -EINVAL;
826	}
827	status = msix_capability_init(dev, entries, nvec);
828	return status;
829}
830
831void pci_disable_msix(struct pci_dev* dev)
832{
833	int pos, temp;
834	u16 control;
835
836	if (!pci_msi_enable)
837		return;
838	if (!dev)
839		return;
840
841	pos = pci_find_capability(dev, PCI_CAP_ID_MSIX);
842	if (!pos)
843		return;
844
845	pci_read_config_word(dev, msi_control_reg(pos), &control);
846	if (!(control & PCI_MSIX_FLAGS_ENABLE))
847		return;
848
849	disable_msi_mode(dev, pos, PCI_CAP_ID_MSIX);
850
851	temp = dev->irq;
852	if (!msi_lookup_irq(dev, PCI_CAP_ID_MSIX)) {
853		int irq, head, tail = 0, warning = 0;
854		unsigned long flags;
855
856		irq = head = dev->irq;
857		dev->irq = temp;			/* Restore pin IRQ */
858		while (head != tail) {
859			spin_lock_irqsave(&msi_lock, flags);
860			tail = msi_desc[irq]->link.tail;
861			spin_unlock_irqrestore(&msi_lock, flags);
862			if (irq_has_action(irq))
863				warning = 1;
864			else if (irq != head)	/* Release MSI-X irq */
865				msi_free_irq(dev, irq);
866			irq = tail;
867		}
868		msi_free_irq(dev, irq);
869		if (warning) {
870			printk(KERN_WARNING "PCI: %s: pci_disable_msix() called without "
871			       "free_irq() on all MSI-X irqs\n",
872			       pci_name(dev));
873			BUG_ON(warning > 0);
874		}
875	}
876}
877
878/**
879 * msi_remove_pci_irq_vectors - reclaim MSI(X) irqs to unused state
880 * @dev: pointer to the pci_dev data structure of MSI(X) device function
881 *
882 * Being called during hotplug remove, from which the device function
883 * is hot-removed. All previous assigned MSI/MSI-X irqs, if
884 * allocated for this device function, are reclaimed to unused state,
885 * which may be used later on.
886 **/
887void msi_remove_pci_irq_vectors(struct pci_dev* dev)
888{
889	int pos, temp;
890	unsigned long flags;
891
892	if (!pci_msi_enable || !dev)
893 		return;
894
895	temp = dev->irq;		/* Save IOAPIC IRQ */
896	pos = pci_find_capability(dev, PCI_CAP_ID_MSI);
897	if (pos > 0 && !msi_lookup_irq(dev, PCI_CAP_ID_MSI)) {
898		if (irq_has_action(dev->irq)) {
899			printk(KERN_WARNING "PCI: %s: msi_remove_pci_irq_vectors() "
900			       "called without free_irq() on MSI irq %d\n",
901			       pci_name(dev), dev->irq);
902			BUG_ON(irq_has_action(dev->irq));
903		} else /* Release MSI irq assigned to this device */
904			msi_free_irq(dev, dev->irq);
905		dev->irq = temp;		/* Restore IOAPIC IRQ */
906	}
907	pos = pci_find_capability(dev, PCI_CAP_ID_MSIX);
908	if (pos > 0 && !msi_lookup_irq(dev, PCI_CAP_ID_MSIX)) {
909		int irq, head, tail = 0, warning = 0;
910		void __iomem *base = NULL;
911
912		irq = head = dev->irq;
913		while (head != tail) {
914			spin_lock_irqsave(&msi_lock, flags);
915			tail = msi_desc[irq]->link.tail;
916			base = msi_desc[irq]->mask_base;
917			spin_unlock_irqrestore(&msi_lock, flags);
918			if (irq_has_action(irq))
919				warning = 1;
920			else if (irq != head) /* Release MSI-X irq */
921				msi_free_irq(dev, irq);
922			irq = tail;
923		}
924		msi_free_irq(dev, irq);
925		if (warning) {
926			iounmap(base);
927			printk(KERN_WARNING "PCI: %s: msi_remove_pci_irq_vectors() "
928			       "called without free_irq() on all MSI-X irqs\n",
929			       pci_name(dev));
930			BUG_ON(warning > 0);
931		}
932		dev->irq = temp;		/* Restore IOAPIC IRQ */
933	}
934}
935
936void pci_no_msi(void)
937{
938	pci_msi_enable = 0;
939}
940
941EXPORT_SYMBOL(pci_enable_msi);
942EXPORT_SYMBOL(pci_disable_msi);
943EXPORT_SYMBOL(pci_enable_msix);
944EXPORT_SYMBOL(pci_disable_msix);
945