msi.c revision d40f540ce6d992d4123827dbd62f68c4a39c53d0
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 msi_desc* msi_desc[NR_IRQS] = { [0 ... NR_IRQS-1] = NULL };
28static struct kmem_cache* msi_cachep;
29
30static int pci_msi_enable = 1;
31
32static int msi_cache_init(void)
33{
34	msi_cachep = kmem_cache_create("msi_cache", sizeof(struct msi_desc),
35					0, SLAB_HWCACHE_ALIGN, NULL, NULL);
36	if (!msi_cachep)
37		return -ENOMEM;
38
39	return 0;
40}
41
42static void msi_set_mask_bit(unsigned int irq, int flag)
43{
44	struct msi_desc *entry;
45
46	entry = msi_desc[irq];
47	BUG_ON(!entry || !entry->dev);
48	switch (entry->msi_attrib.type) {
49	case PCI_CAP_ID_MSI:
50		if (entry->msi_attrib.maskbit) {
51			int pos;
52			u32 mask_bits;
53
54			pos = (long)entry->mask_base;
55			pci_read_config_dword(entry->dev, pos, &mask_bits);
56			mask_bits &= ~(1);
57			mask_bits |= flag;
58			pci_write_config_dword(entry->dev, pos, mask_bits);
59		}
60		break;
61	case PCI_CAP_ID_MSIX:
62	{
63		int offset = entry->msi_attrib.entry_nr * PCI_MSIX_ENTRY_SIZE +
64			PCI_MSIX_ENTRY_VECTOR_CTRL_OFFSET;
65		writel(flag, entry->mask_base + offset);
66		break;
67	}
68	default:
69		BUG();
70		break;
71	}
72}
73
74void read_msi_msg(unsigned int irq, struct msi_msg *msg)
75{
76	struct msi_desc *entry = get_irq_data(irq);
77	switch(entry->msi_attrib.type) {
78	case PCI_CAP_ID_MSI:
79	{
80		struct pci_dev *dev = entry->dev;
81		int pos = entry->msi_attrib.pos;
82		u16 data;
83
84		pci_read_config_dword(dev, msi_lower_address_reg(pos),
85					&msg->address_lo);
86		if (entry->msi_attrib.is_64) {
87			pci_read_config_dword(dev, msi_upper_address_reg(pos),
88						&msg->address_hi);
89			pci_read_config_word(dev, msi_data_reg(pos, 1), &data);
90		} else {
91			msg->address_hi = 0;
92			pci_read_config_word(dev, msi_data_reg(pos, 1), &data);
93		}
94		msg->data = data;
95		break;
96	}
97	case PCI_CAP_ID_MSIX:
98	{
99		void __iomem *base;
100		base = entry->mask_base +
101			entry->msi_attrib.entry_nr * PCI_MSIX_ENTRY_SIZE;
102
103		msg->address_lo = readl(base + PCI_MSIX_ENTRY_LOWER_ADDR_OFFSET);
104		msg->address_hi = readl(base + PCI_MSIX_ENTRY_UPPER_ADDR_OFFSET);
105		msg->data = readl(base + PCI_MSIX_ENTRY_DATA_OFFSET);
106 		break;
107 	}
108 	default:
109		BUG();
110	}
111}
112
113void write_msi_msg(unsigned int irq, struct msi_msg *msg)
114{
115	struct msi_desc *entry = get_irq_data(irq);
116	switch (entry->msi_attrib.type) {
117	case PCI_CAP_ID_MSI:
118	{
119		struct pci_dev *dev = entry->dev;
120		int pos = entry->msi_attrib.pos;
121
122		pci_write_config_dword(dev, msi_lower_address_reg(pos),
123					msg->address_lo);
124		if (entry->msi_attrib.is_64) {
125			pci_write_config_dword(dev, msi_upper_address_reg(pos),
126						msg->address_hi);
127			pci_write_config_word(dev, msi_data_reg(pos, 1),
128						msg->data);
129		} else {
130			pci_write_config_word(dev, msi_data_reg(pos, 0),
131						msg->data);
132		}
133		break;
134	}
135	case PCI_CAP_ID_MSIX:
136	{
137		void __iomem *base;
138		base = entry->mask_base +
139			entry->msi_attrib.entry_nr * PCI_MSIX_ENTRY_SIZE;
140
141		writel(msg->address_lo,
142			base + PCI_MSIX_ENTRY_LOWER_ADDR_OFFSET);
143		writel(msg->address_hi,
144			base + PCI_MSIX_ENTRY_UPPER_ADDR_OFFSET);
145		writel(msg->data, base + PCI_MSIX_ENTRY_DATA_OFFSET);
146		break;
147	}
148	default:
149		BUG();
150	}
151}
152
153void mask_msi_irq(unsigned int irq)
154{
155	msi_set_mask_bit(irq, 1);
156}
157
158void unmask_msi_irq(unsigned int irq)
159{
160	msi_set_mask_bit(irq, 0);
161}
162
163static int msi_free_irq(struct pci_dev* dev, int irq);
164
165static int msi_init(void)
166{
167	static int status = -ENOMEM;
168
169	if (!status)
170		return status;
171
172	status = msi_cache_init();
173	if (status < 0) {
174		pci_msi_enable = 0;
175		printk(KERN_WARNING "PCI: MSI cache init failed\n");
176		return status;
177	}
178
179	return status;
180}
181
182static struct msi_desc* alloc_msi_entry(void)
183{
184	struct msi_desc *entry;
185
186	entry = kmem_cache_zalloc(msi_cachep, GFP_KERNEL);
187	if (!entry)
188		return NULL;
189
190	entry->link.tail = entry->link.head = 0;	/* single message */
191	entry->dev = NULL;
192
193	return entry;
194}
195
196static void attach_msi_entry(struct msi_desc *entry, int irq)
197{
198	msi_desc[irq] = entry;
199}
200
201static int create_msi_irq(void)
202{
203	struct msi_desc *entry;
204	int irq;
205
206	entry = alloc_msi_entry();
207	if (!entry)
208		return -ENOMEM;
209
210	irq = create_irq();
211	if (irq < 0) {
212		kmem_cache_free(msi_cachep, entry);
213		return -EBUSY;
214	}
215
216	set_irq_data(irq, entry);
217
218	return irq;
219}
220
221static void destroy_msi_irq(unsigned int irq)
222{
223	struct msi_desc *entry;
224
225	entry = get_irq_data(irq);
226	set_irq_chip(irq, NULL);
227	set_irq_data(irq, NULL);
228	destroy_irq(irq);
229	kmem_cache_free(msi_cachep, entry);
230}
231
232static void enable_msi_mode(struct pci_dev *dev, int pos, int type)
233{
234	u16 control;
235
236	pci_read_config_word(dev, msi_control_reg(pos), &control);
237	if (type == PCI_CAP_ID_MSI) {
238		/* Set enabled bits to single MSI & enable MSI_enable bit */
239		msi_enable(control, 1);
240		pci_write_config_word(dev, msi_control_reg(pos), control);
241		dev->msi_enabled = 1;
242	} else {
243		msix_enable(control);
244		pci_write_config_word(dev, msi_control_reg(pos), control);
245		dev->msix_enabled = 1;
246	}
247
248	pci_intx(dev, 0);  /* disable intx */
249}
250
251void disable_msi_mode(struct pci_dev *dev, int pos, int type)
252{
253	u16 control;
254
255	pci_read_config_word(dev, msi_control_reg(pos), &control);
256	if (type == PCI_CAP_ID_MSI) {
257		/* Set enabled bits to single MSI & enable MSI_enable bit */
258		msi_disable(control);
259		pci_write_config_word(dev, msi_control_reg(pos), control);
260		dev->msi_enabled = 0;
261	} else {
262		msix_disable(control);
263		pci_write_config_word(dev, msi_control_reg(pos), control);
264		dev->msix_enabled = 0;
265	}
266
267	pci_intx(dev, 1);  /* enable intx */
268}
269
270#ifdef CONFIG_PM
271static int __pci_save_msi_state(struct pci_dev *dev)
272{
273	int pos, i = 0;
274	u16 control;
275	struct pci_cap_saved_state *save_state;
276	u32 *cap;
277
278	pos = pci_find_capability(dev, PCI_CAP_ID_MSI);
279	if (pos <= 0 || dev->no_msi)
280		return 0;
281
282	pci_read_config_word(dev, msi_control_reg(pos), &control);
283	if (!(control & PCI_MSI_FLAGS_ENABLE))
284		return 0;
285
286	save_state = kzalloc(sizeof(struct pci_cap_saved_state) + sizeof(u32) * 5,
287		GFP_KERNEL);
288	if (!save_state) {
289		printk(KERN_ERR "Out of memory in pci_save_msi_state\n");
290		return -ENOMEM;
291	}
292	cap = &save_state->data[0];
293
294	pci_read_config_dword(dev, pos, &cap[i++]);
295	control = cap[0] >> 16;
296	pci_read_config_dword(dev, pos + PCI_MSI_ADDRESS_LO, &cap[i++]);
297	if (control & PCI_MSI_FLAGS_64BIT) {
298		pci_read_config_dword(dev, pos + PCI_MSI_ADDRESS_HI, &cap[i++]);
299		pci_read_config_dword(dev, pos + PCI_MSI_DATA_64, &cap[i++]);
300	} else
301		pci_read_config_dword(dev, pos + PCI_MSI_DATA_32, &cap[i++]);
302	if (control & PCI_MSI_FLAGS_MASKBIT)
303		pci_read_config_dword(dev, pos + PCI_MSI_MASK_BIT, &cap[i++]);
304	save_state->cap_nr = PCI_CAP_ID_MSI;
305	pci_add_saved_cap(dev, save_state);
306	return 0;
307}
308
309static void __pci_restore_msi_state(struct pci_dev *dev)
310{
311	int i = 0, pos;
312	u16 control;
313	struct pci_cap_saved_state *save_state;
314	u32 *cap;
315
316	save_state = pci_find_saved_cap(dev, PCI_CAP_ID_MSI);
317	pos = pci_find_capability(dev, PCI_CAP_ID_MSI);
318	if (!save_state || pos <= 0)
319		return;
320	cap = &save_state->data[0];
321
322	control = cap[i++] >> 16;
323	pci_write_config_dword(dev, pos + PCI_MSI_ADDRESS_LO, cap[i++]);
324	if (control & PCI_MSI_FLAGS_64BIT) {
325		pci_write_config_dword(dev, pos + PCI_MSI_ADDRESS_HI, cap[i++]);
326		pci_write_config_dword(dev, pos + PCI_MSI_DATA_64, cap[i++]);
327	} else
328		pci_write_config_dword(dev, pos + PCI_MSI_DATA_32, cap[i++]);
329	if (control & PCI_MSI_FLAGS_MASKBIT)
330		pci_write_config_dword(dev, pos + PCI_MSI_MASK_BIT, cap[i++]);
331	pci_write_config_word(dev, pos + PCI_MSI_FLAGS, control);
332	enable_msi_mode(dev, pos, PCI_CAP_ID_MSI);
333	pci_remove_saved_cap(save_state);
334	kfree(save_state);
335}
336
337static int __pci_save_msix_state(struct pci_dev *dev)
338{
339	int pos;
340	int irq, head, tail = 0;
341	u16 control;
342	struct pci_cap_saved_state *save_state;
343
344	if (!dev->msix_enabled)
345		return 0;
346
347	pos = pci_find_capability(dev, PCI_CAP_ID_MSIX);
348	if (pos <= 0 || dev->no_msi)
349		return 0;
350
351	/* save the capability */
352	pci_read_config_word(dev, msi_control_reg(pos), &control);
353	if (!(control & PCI_MSIX_FLAGS_ENABLE))
354		return 0;
355	save_state = kzalloc(sizeof(struct pci_cap_saved_state) + sizeof(u16),
356		GFP_KERNEL);
357	if (!save_state) {
358		printk(KERN_ERR "Out of memory in pci_save_msix_state\n");
359		return -ENOMEM;
360	}
361	*((u16 *)&save_state->data[0]) = control;
362
363	/* save the table */
364	irq = head = dev->first_msi_irq;
365	while (head != tail) {
366		struct msi_desc *entry;
367
368		entry = msi_desc[irq];
369		read_msi_msg(irq, &entry->msg_save);
370
371		tail = msi_desc[irq]->link.tail;
372		irq = tail;
373	}
374
375	save_state->cap_nr = PCI_CAP_ID_MSIX;
376	pci_add_saved_cap(dev, save_state);
377	return 0;
378}
379
380int pci_save_msi_state(struct pci_dev *dev)
381{
382	int rc;
383
384	rc = __pci_save_msi_state(dev);
385	if (rc)
386		return rc;
387
388	rc = __pci_save_msix_state(dev);
389
390	return rc;
391}
392
393static void __pci_restore_msix_state(struct pci_dev *dev)
394{
395	u16 save;
396	int pos;
397	int irq, head, tail = 0;
398	struct msi_desc *entry;
399	struct pci_cap_saved_state *save_state;
400
401	if (!dev->msix_enabled)
402		return;
403
404	save_state = pci_find_saved_cap(dev, PCI_CAP_ID_MSIX);
405	if (!save_state)
406		return;
407	save = *((u16 *)&save_state->data[0]);
408	pci_remove_saved_cap(save_state);
409	kfree(save_state);
410
411	pos = pci_find_capability(dev, PCI_CAP_ID_MSIX);
412	if (pos <= 0)
413		return;
414
415	/* route the table */
416	irq = head = dev->first_msi_irq;
417	while (head != tail) {
418		entry = msi_desc[irq];
419		write_msi_msg(irq, &entry->msg_save);
420
421		tail = msi_desc[irq]->link.tail;
422		irq = tail;
423	}
424
425	pci_write_config_word(dev, msi_control_reg(pos), save);
426	enable_msi_mode(dev, pos, PCI_CAP_ID_MSIX);
427}
428
429void pci_restore_msi_state(struct pci_dev *dev)
430{
431	__pci_restore_msi_state(dev);
432	__pci_restore_msix_state(dev);
433}
434#endif	/* CONFIG_PM */
435
436/**
437 * msi_capability_init - configure device's MSI capability structure
438 * @dev: pointer to the pci_dev data structure of MSI device function
439 *
440 * Setup the MSI capability structure of device function with a single
441 * MSI irq, regardless of device function is capable of handling
442 * multiple messages. A return of zero indicates the successful setup
443 * of an entry zero with the new MSI irq or non-zero for otherwise.
444 **/
445static int msi_capability_init(struct pci_dev *dev)
446{
447	int status;
448	struct msi_desc *entry;
449	int pos, irq;
450	u16 control;
451
452   	pos = pci_find_capability(dev, PCI_CAP_ID_MSI);
453	pci_read_config_word(dev, msi_control_reg(pos), &control);
454	/* MSI Entry Initialization */
455	irq = create_msi_irq();
456	if (irq < 0)
457		return irq;
458
459	entry = get_irq_data(irq);
460	entry->link.head = irq;
461	entry->link.tail = irq;
462	entry->msi_attrib.type = PCI_CAP_ID_MSI;
463	entry->msi_attrib.is_64 = is_64bit_address(control);
464	entry->msi_attrib.entry_nr = 0;
465	entry->msi_attrib.maskbit = is_mask_bit_support(control);
466	entry->msi_attrib.default_irq = dev->irq;	/* Save IOAPIC IRQ */
467	entry->msi_attrib.pos = pos;
468	if (is_mask_bit_support(control)) {
469		entry->mask_base = (void __iomem *)(long)msi_mask_bits_reg(pos,
470				is_64bit_address(control));
471	}
472	entry->dev = dev;
473	if (entry->msi_attrib.maskbit) {
474		unsigned int maskbits, temp;
475		/* All MSIs are unmasked by default, Mask them all */
476		pci_read_config_dword(dev,
477			msi_mask_bits_reg(pos, is_64bit_address(control)),
478			&maskbits);
479		temp = (1 << multi_msi_capable(control));
480		temp = ((temp - 1) & ~temp);
481		maskbits |= temp;
482		pci_write_config_dword(dev,
483			msi_mask_bits_reg(pos, is_64bit_address(control)),
484			maskbits);
485	}
486	/* Configure MSI capability structure */
487	status = arch_setup_msi_irq(irq, dev);
488	if (status < 0) {
489		destroy_msi_irq(irq);
490		return status;
491	}
492
493	dev->first_msi_irq = irq;
494	attach_msi_entry(entry, irq);
495	/* Set MSI enabled bits	 */
496	enable_msi_mode(dev, pos, PCI_CAP_ID_MSI);
497
498	dev->irq = irq;
499	return 0;
500}
501
502/**
503 * msix_capability_init - configure device's MSI-X capability
504 * @dev: pointer to the pci_dev data structure of MSI-X device function
505 * @entries: pointer to an array of struct msix_entry entries
506 * @nvec: number of @entries
507 *
508 * Setup the MSI-X capability structure of device function with a
509 * single MSI-X irq. A return of zero indicates the successful setup of
510 * requested MSI-X entries with allocated irqs or non-zero for otherwise.
511 **/
512static int msix_capability_init(struct pci_dev *dev,
513				struct msix_entry *entries, int nvec)
514{
515	struct msi_desc *head = NULL, *tail = NULL, *entry = NULL;
516	int status;
517	int irq, pos, i, j, nr_entries, temp = 0;
518	unsigned long phys_addr;
519	u32 table_offset;
520 	u16 control;
521	u8 bir;
522	void __iomem *base;
523
524   	pos = pci_find_capability(dev, PCI_CAP_ID_MSIX);
525	/* Request & Map MSI-X table region */
526 	pci_read_config_word(dev, msi_control_reg(pos), &control);
527	nr_entries = multi_msix_capable(control);
528
529 	pci_read_config_dword(dev, msix_table_offset_reg(pos), &table_offset);
530	bir = (u8)(table_offset & PCI_MSIX_FLAGS_BIRMASK);
531	table_offset &= ~PCI_MSIX_FLAGS_BIRMASK;
532	phys_addr = pci_resource_start (dev, bir) + table_offset;
533	base = ioremap_nocache(phys_addr, nr_entries * PCI_MSIX_ENTRY_SIZE);
534	if (base == NULL)
535		return -ENOMEM;
536
537	/* MSI-X Table Initialization */
538	for (i = 0; i < nvec; i++) {
539		irq = create_msi_irq();
540		if (irq < 0)
541			break;
542
543		entry = get_irq_data(irq);
544 		j = entries[i].entry;
545 		entries[i].vector = irq;
546		entry->msi_attrib.type = PCI_CAP_ID_MSIX;
547		entry->msi_attrib.is_64 = 1;
548		entry->msi_attrib.entry_nr = j;
549		entry->msi_attrib.maskbit = 1;
550		entry->msi_attrib.default_irq = dev->irq;
551		entry->msi_attrib.pos = pos;
552		entry->dev = dev;
553		entry->mask_base = base;
554		if (!head) {
555			entry->link.head = irq;
556			entry->link.tail = irq;
557			head = entry;
558		} else {
559			entry->link.head = temp;
560			entry->link.tail = tail->link.tail;
561			tail->link.tail = irq;
562			head->link.head = irq;
563		}
564		temp = irq;
565		tail = entry;
566		/* Configure MSI-X capability structure */
567		status = arch_setup_msi_irq(irq, dev);
568		if (status < 0) {
569			destroy_msi_irq(irq);
570			break;
571		}
572
573		attach_msi_entry(entry, irq);
574	}
575	if (i != nvec) {
576		int avail = i - 1;
577		i--;
578		for (; i >= 0; i--) {
579			irq = (entries + i)->vector;
580			msi_free_irq(dev, irq);
581			(entries + i)->vector = 0;
582		}
583		/* If we had some success report the number of irqs
584		 * we succeeded in setting up.
585		 */
586		if (avail <= 0)
587			avail = -EBUSY;
588		return avail;
589	}
590	dev->first_msi_irq = entries[0].vector;
591	/* Set MSI-X enabled bits */
592	enable_msi_mode(dev, pos, PCI_CAP_ID_MSIX);
593
594	return 0;
595}
596
597/**
598 * pci_msi_supported - check whether MSI may be enabled on device
599 * @dev: pointer to the pci_dev data structure of MSI device function
600 *
601 * Look at global flags, the device itself, and its parent busses
602 * to return 0 if MSI are supported for the device.
603 **/
604static
605int pci_msi_supported(struct pci_dev * dev)
606{
607	struct pci_bus *bus;
608
609	/* MSI must be globally enabled and supported by the device */
610	if (!pci_msi_enable || !dev || dev->no_msi)
611		return -EINVAL;
612
613	/* Any bridge which does NOT route MSI transactions from it's
614	 * secondary bus to it's primary bus must set NO_MSI flag on
615	 * the secondary pci_bus.
616	 * We expect only arch-specific PCI host bus controller driver
617	 * or quirks for specific PCI bridges to be setting NO_MSI.
618	 */
619	for (bus = dev->bus; bus; bus = bus->parent)
620		if (bus->bus_flags & PCI_BUS_FLAGS_NO_MSI)
621			return -EINVAL;
622
623	return 0;
624}
625
626/**
627 * pci_enable_msi - configure device's MSI capability structure
628 * @dev: pointer to the pci_dev data structure of MSI device function
629 *
630 * Setup the MSI capability structure of device function with
631 * a single MSI irq upon its software driver call to request for
632 * MSI mode enabled on its hardware device function. A return of zero
633 * indicates the successful setup of an entry zero with the new MSI
634 * irq or non-zero for otherwise.
635 **/
636int pci_enable_msi(struct pci_dev* dev)
637{
638	int pos, status;
639
640	if (pci_msi_supported(dev) < 0)
641		return -EINVAL;
642
643	status = msi_init();
644	if (status < 0)
645		return status;
646
647	pos = pci_find_capability(dev, PCI_CAP_ID_MSI);
648	if (!pos)
649		return -EINVAL;
650
651	WARN_ON(!!dev->msi_enabled);
652
653	/* Check whether driver already requested for MSI-X irqs */
654	pos = pci_find_capability(dev, PCI_CAP_ID_MSIX);
655	if (pos > 0 && dev->msix_enabled) {
656			printk(KERN_INFO "PCI: %s: Can't enable MSI.  "
657			       "Device already has MSI-X enabled\n",
658			       pci_name(dev));
659			return -EINVAL;
660	}
661	status = msi_capability_init(dev);
662	return status;
663}
664
665void pci_disable_msi(struct pci_dev* dev)
666{
667	struct msi_desc *entry;
668	int pos, default_irq;
669	u16 control;
670
671	if (!pci_msi_enable)
672		return;
673	if (!dev)
674		return;
675
676	if (!dev->msi_enabled)
677		return;
678
679	pos = pci_find_capability(dev, PCI_CAP_ID_MSI);
680	if (!pos)
681		return;
682
683	pci_read_config_word(dev, msi_control_reg(pos), &control);
684	if (!(control & PCI_MSI_FLAGS_ENABLE))
685		return;
686
687
688	disable_msi_mode(dev, pos, PCI_CAP_ID_MSI);
689
690	entry = msi_desc[dev->first_msi_irq];
691	if (!entry || !entry->dev || entry->msi_attrib.type != PCI_CAP_ID_MSI) {
692		return;
693	}
694	if (irq_has_action(dev->first_msi_irq)) {
695		printk(KERN_WARNING "PCI: %s: pci_disable_msi() called without "
696		       "free_irq() on MSI irq %d\n",
697		       pci_name(dev), dev->first_msi_irq);
698		BUG_ON(irq_has_action(dev->first_msi_irq));
699	} else {
700		default_irq = entry->msi_attrib.default_irq;
701		msi_free_irq(dev, dev->first_msi_irq);
702
703		/* Restore dev->irq to its default pin-assertion irq */
704		dev->irq = default_irq;
705	}
706	dev->first_msi_irq = 0;
707}
708
709static int msi_free_irq(struct pci_dev* dev, int irq)
710{
711	struct msi_desc *entry;
712	int head, entry_nr, type;
713	void __iomem *base;
714
715	arch_teardown_msi_irq(irq);
716
717	entry = msi_desc[irq];
718	if (!entry || entry->dev != dev) {
719		return -EINVAL;
720	}
721	type = entry->msi_attrib.type;
722	entry_nr = entry->msi_attrib.entry_nr;
723	head = entry->link.head;
724	base = entry->mask_base;
725	msi_desc[entry->link.head]->link.tail = entry->link.tail;
726	msi_desc[entry->link.tail]->link.head = entry->link.head;
727	entry->dev = NULL;
728	msi_desc[irq] = NULL;
729
730	destroy_msi_irq(irq);
731
732	if (type == PCI_CAP_ID_MSIX) {
733		writel(1, base + entry_nr * PCI_MSIX_ENTRY_SIZE +
734			PCI_MSIX_ENTRY_VECTOR_CTRL_OFFSET);
735
736		if (head == irq)
737			iounmap(base);
738	}
739
740	return 0;
741}
742
743/**
744 * pci_enable_msix - configure device's MSI-X capability structure
745 * @dev: pointer to the pci_dev data structure of MSI-X device function
746 * @entries: pointer to an array of MSI-X entries
747 * @nvec: number of MSI-X irqs requested for allocation by device driver
748 *
749 * Setup the MSI-X capability structure of device function with the number
750 * of requested irqs upon its software driver call to request for
751 * MSI-X mode enabled on its hardware device function. A return of zero
752 * indicates the successful configuration of MSI-X capability structure
753 * with new allocated MSI-X irqs. A return of < 0 indicates a failure.
754 * Or a return of > 0 indicates that driver request is exceeding the number
755 * of irqs available. Driver should use the returned value to re-send
756 * its request.
757 **/
758int pci_enable_msix(struct pci_dev* dev, struct msix_entry *entries, int nvec)
759{
760	int status, pos, nr_entries;
761	int i, j;
762	u16 control;
763
764	if (!entries || pci_msi_supported(dev) < 0)
765 		return -EINVAL;
766
767	status = msi_init();
768	if (status < 0)
769		return status;
770
771	pos = pci_find_capability(dev, PCI_CAP_ID_MSIX);
772	if (!pos)
773 		return -EINVAL;
774
775	pci_read_config_word(dev, msi_control_reg(pos), &control);
776	nr_entries = multi_msix_capable(control);
777	if (nvec > nr_entries)
778		return -EINVAL;
779
780	/* Check for any invalid entries */
781	for (i = 0; i < nvec; i++) {
782		if (entries[i].entry >= nr_entries)
783			return -EINVAL;		/* invalid entry */
784		for (j = i + 1; j < nvec; j++) {
785			if (entries[i].entry == entries[j].entry)
786				return -EINVAL;	/* duplicate entry */
787		}
788	}
789	WARN_ON(!!dev->msix_enabled);
790
791	/* Check whether driver already requested for MSI irq */
792   	if (pci_find_capability(dev, PCI_CAP_ID_MSI) > 0 &&
793		dev->msi_enabled) {
794		printk(KERN_INFO "PCI: %s: Can't enable MSI-X.  "
795		       "Device already has an MSI irq assigned\n",
796		       pci_name(dev));
797		return -EINVAL;
798	}
799	status = msix_capability_init(dev, entries, nvec);
800	return status;
801}
802
803void pci_disable_msix(struct pci_dev* dev)
804{
805	int irq, head, tail = 0, warning = 0;
806	int pos;
807	u16 control;
808
809	if (!pci_msi_enable)
810		return;
811	if (!dev)
812		return;
813
814	if (!dev->msix_enabled)
815		return;
816
817	pos = pci_find_capability(dev, PCI_CAP_ID_MSIX);
818	if (!pos)
819		return;
820
821	pci_read_config_word(dev, msi_control_reg(pos), &control);
822	if (!(control & PCI_MSIX_FLAGS_ENABLE))
823		return;
824
825	disable_msi_mode(dev, pos, PCI_CAP_ID_MSIX);
826
827	irq = head = dev->first_msi_irq;
828	while (head != tail) {
829		tail = msi_desc[irq]->link.tail;
830		if (irq_has_action(irq))
831			warning = 1;
832		else if (irq != head)	/* Release MSI-X irq */
833			msi_free_irq(dev, irq);
834		irq = tail;
835	}
836	msi_free_irq(dev, irq);
837	if (warning) {
838		printk(KERN_WARNING "PCI: %s: pci_disable_msix() called without "
839			"free_irq() on all MSI-X irqs\n",
840			pci_name(dev));
841		BUG_ON(warning > 0);
842	}
843	dev->first_msi_irq = 0;
844}
845
846/**
847 * msi_remove_pci_irq_vectors - reclaim MSI(X) irqs to unused state
848 * @dev: pointer to the pci_dev data structure of MSI(X) device function
849 *
850 * Being called during hotplug remove, from which the device function
851 * is hot-removed. All previous assigned MSI/MSI-X irqs, if
852 * allocated for this device function, are reclaimed to unused state,
853 * which may be used later on.
854 **/
855void msi_remove_pci_irq_vectors(struct pci_dev* dev)
856{
857	int pos;
858
859	if (!pci_msi_enable || !dev)
860 		return;
861
862	pos = pci_find_capability(dev, PCI_CAP_ID_MSI);
863	if (pos > 0 && dev->msi_enabled) {
864		if (irq_has_action(dev->first_msi_irq)) {
865			printk(KERN_WARNING "PCI: %s: msi_remove_pci_irq_vectors() "
866			       "called without free_irq() on MSI irq %d\n",
867			       pci_name(dev), dev->first_msi_irq);
868			BUG_ON(irq_has_action(dev->first_msi_irq));
869		} else /* Release MSI irq assigned to this device */
870			msi_free_irq(dev, dev->first_msi_irq);
871	}
872	pos = pci_find_capability(dev, PCI_CAP_ID_MSIX);
873	if (pos > 0 && dev->msix_enabled) {
874		int irq, head, tail = 0, warning = 0;
875		void __iomem *base = NULL;
876
877		irq = head = dev->first_msi_irq;
878		while (head != tail) {
879			tail = msi_desc[irq]->link.tail;
880			base = msi_desc[irq]->mask_base;
881			if (irq_has_action(irq))
882				warning = 1;
883			else if (irq != head) /* Release MSI-X irq */
884				msi_free_irq(dev, irq);
885			irq = tail;
886		}
887		msi_free_irq(dev, irq);
888		if (warning) {
889			iounmap(base);
890			printk(KERN_WARNING "PCI: %s: msi_remove_pci_irq_vectors() "
891			       "called without free_irq() on all MSI-X irqs\n",
892			       pci_name(dev));
893			BUG_ON(warning > 0);
894		}
895	}
896}
897
898void pci_no_msi(void)
899{
900	pci_msi_enable = 0;
901}
902
903EXPORT_SYMBOL(pci_enable_msi);
904EXPORT_SYMBOL(pci_disable_msi);
905EXPORT_SYMBOL(pci_enable_msix);
906EXPORT_SYMBOL(pci_disable_msix);
907