msi.c revision ab7de999a2c771482698efa6fe7c7b7fcb1d482a
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/pci.h>
16#include <linux/proc_fs.h>
17#include <linux/msi.h>
18#include <linux/smp.h>
19
20#include <asm/errno.h>
21#include <asm/io.h>
22
23#include "pci.h"
24#include "msi.h"
25
26static int pci_msi_enable = 1;
27
28/* Arch hooks */
29
30#ifndef arch_msi_check_device
31int arch_msi_check_device(struct pci_dev *dev, int nvec, int type)
32{
33	return 0;
34}
35#endif
36
37#ifndef arch_setup_msi_irqs
38int arch_setup_msi_irqs(struct pci_dev *dev, int nvec, int type)
39{
40	struct msi_desc *entry;
41	int ret;
42
43	/*
44	 * If an architecture wants to support multiple MSI, it needs to
45	 * override arch_setup_msi_irqs()
46	 */
47	if (type == PCI_CAP_ID_MSI && nvec > 1)
48		return 1;
49
50	list_for_each_entry(entry, &dev->msi_list, list) {
51		ret = arch_setup_msi_irq(dev, entry);
52		if (ret < 0)
53			return ret;
54		if (ret > 0)
55			return -ENOSPC;
56	}
57
58	return 0;
59}
60#endif
61
62#ifndef arch_teardown_msi_irqs
63void arch_teardown_msi_irqs(struct pci_dev *dev)
64{
65	struct msi_desc *entry;
66
67	list_for_each_entry(entry, &dev->msi_list, list) {
68		int i, nvec;
69		if (entry->irq == 0)
70			continue;
71		nvec = 1 << entry->msi_attrib.multiple;
72		for (i = 0; i < nvec; i++)
73			arch_teardown_msi_irq(entry->irq + i);
74	}
75}
76#endif
77
78static void __msi_set_enable(struct pci_dev *dev, int pos, int enable)
79{
80	u16 control;
81
82	if (pos) {
83		pci_read_config_word(dev, pos + PCI_MSI_FLAGS, &control);
84		control &= ~PCI_MSI_FLAGS_ENABLE;
85		if (enable)
86			control |= PCI_MSI_FLAGS_ENABLE;
87		pci_write_config_word(dev, pos + PCI_MSI_FLAGS, control);
88	}
89}
90
91static void msi_set_enable(struct pci_dev *dev, int enable)
92{
93	__msi_set_enable(dev, pci_find_capability(dev, PCI_CAP_ID_MSI), enable);
94}
95
96static void msix_set_enable(struct pci_dev *dev, int enable)
97{
98	int pos;
99	u16 control;
100
101	pos = pci_find_capability(dev, PCI_CAP_ID_MSIX);
102	if (pos) {
103		pci_read_config_word(dev, pos + PCI_MSIX_FLAGS, &control);
104		control &= ~PCI_MSIX_FLAGS_ENABLE;
105		if (enable)
106			control |= PCI_MSIX_FLAGS_ENABLE;
107		pci_write_config_word(dev, pos + PCI_MSIX_FLAGS, control);
108	}
109}
110
111static inline __attribute_const__ u32 msi_mask(unsigned x)
112{
113	/* Don't shift by >= width of type */
114	if (x >= 5)
115		return 0xffffffff;
116	return (1 << (1 << x)) - 1;
117}
118
119static inline __attribute_const__ u32 msi_capable_mask(u16 control)
120{
121	return msi_mask((control >> 1) & 7);
122}
123
124static inline __attribute_const__ u32 msi_enabled_mask(u16 control)
125{
126	return msi_mask((control >> 4) & 7);
127}
128
129/*
130 * PCI 2.3 does not specify mask bits for each MSI interrupt.  Attempting to
131 * mask all MSI interrupts by clearing the MSI enable bit does not work
132 * reliably as devices without an INTx disable bit will then generate a
133 * level IRQ which will never be cleared.
134 */
135static void msi_mask_irq(struct msi_desc *desc, u32 mask, u32 flag)
136{
137	u32 mask_bits = desc->masked;
138
139	if (!desc->msi_attrib.maskbit)
140		return;
141
142	mask_bits &= ~mask;
143	mask_bits |= flag;
144	pci_write_config_dword(desc->dev, desc->mask_pos, mask_bits);
145	desc->masked = mask_bits;
146}
147
148/*
149 * This internal function does not flush PCI writes to the device.
150 * All users must ensure that they read from the device before either
151 * assuming that the device state is up to date, or returning out of this
152 * file.  This saves a few milliseconds when initialising devices with lots
153 * of MSI-X interrupts.
154 */
155static void msix_mask_irq(struct msi_desc *desc, u32 flag)
156{
157	u32 mask_bits = desc->masked;
158	unsigned offset = desc->msi_attrib.entry_nr * PCI_MSIX_ENTRY_SIZE +
159					PCI_MSIX_ENTRY_VECTOR_CTRL_OFFSET;
160	mask_bits &= ~1;
161	mask_bits |= flag;
162	writel(mask_bits, desc->mask_base + offset);
163	desc->masked = mask_bits;
164}
165
166static void msi_set_mask_bit(unsigned irq, u32 flag)
167{
168	struct msi_desc *desc = get_irq_msi(irq);
169
170	if (desc->msi_attrib.is_msix) {
171		msix_mask_irq(desc, flag);
172		readl(desc->mask_base);		/* Flush write to device */
173	} else {
174		unsigned offset = irq - desc->dev->irq;
175		msi_mask_irq(desc, 1 << offset, flag << offset);
176	}
177}
178
179void mask_msi_irq(unsigned int irq)
180{
181	msi_set_mask_bit(irq, 1);
182}
183
184void unmask_msi_irq(unsigned int irq)
185{
186	msi_set_mask_bit(irq, 0);
187}
188
189void read_msi_msg_desc(struct irq_desc *desc, struct msi_msg *msg)
190{
191	struct msi_desc *entry = get_irq_desc_msi(desc);
192	if (entry->msi_attrib.is_msix) {
193		void __iomem *base = entry->mask_base +
194			entry->msi_attrib.entry_nr * PCI_MSIX_ENTRY_SIZE;
195
196		msg->address_lo = readl(base + PCI_MSIX_ENTRY_LOWER_ADDR_OFFSET);
197		msg->address_hi = readl(base + PCI_MSIX_ENTRY_UPPER_ADDR_OFFSET);
198		msg->data = readl(base + PCI_MSIX_ENTRY_DATA_OFFSET);
199	} else {
200		struct pci_dev *dev = entry->dev;
201		int pos = entry->msi_attrib.pos;
202		u16 data;
203
204		pci_read_config_dword(dev, msi_lower_address_reg(pos),
205					&msg->address_lo);
206		if (entry->msi_attrib.is_64) {
207			pci_read_config_dword(dev, msi_upper_address_reg(pos),
208						&msg->address_hi);
209			pci_read_config_word(dev, msi_data_reg(pos, 1), &data);
210		} else {
211			msg->address_hi = 0;
212			pci_read_config_word(dev, msi_data_reg(pos, 0), &data);
213		}
214		msg->data = data;
215	}
216}
217
218void read_msi_msg(unsigned int irq, struct msi_msg *msg)
219{
220	struct irq_desc *desc = irq_to_desc(irq);
221
222	read_msi_msg_desc(desc, msg);
223}
224
225void write_msi_msg_desc(struct irq_desc *desc, struct msi_msg *msg)
226{
227	struct msi_desc *entry = get_irq_desc_msi(desc);
228	if (entry->msi_attrib.is_msix) {
229		void __iomem *base;
230		base = entry->mask_base +
231			entry->msi_attrib.entry_nr * PCI_MSIX_ENTRY_SIZE;
232
233		writel(msg->address_lo,
234			base + PCI_MSIX_ENTRY_LOWER_ADDR_OFFSET);
235		writel(msg->address_hi,
236			base + PCI_MSIX_ENTRY_UPPER_ADDR_OFFSET);
237		writel(msg->data, base + PCI_MSIX_ENTRY_DATA_OFFSET);
238	} else {
239		struct pci_dev *dev = entry->dev;
240		int pos = entry->msi_attrib.pos;
241		u16 msgctl;
242
243		pci_read_config_word(dev, msi_control_reg(pos), &msgctl);
244		msgctl &= ~PCI_MSI_FLAGS_QSIZE;
245		msgctl |= entry->msi_attrib.multiple << 4;
246		pci_write_config_word(dev, msi_control_reg(pos), msgctl);
247
248		pci_write_config_dword(dev, msi_lower_address_reg(pos),
249					msg->address_lo);
250		if (entry->msi_attrib.is_64) {
251			pci_write_config_dword(dev, msi_upper_address_reg(pos),
252						msg->address_hi);
253			pci_write_config_word(dev, msi_data_reg(pos, 1),
254						msg->data);
255		} else {
256			pci_write_config_word(dev, msi_data_reg(pos, 0),
257						msg->data);
258		}
259	}
260	entry->msg = *msg;
261}
262
263void write_msi_msg(unsigned int irq, struct msi_msg *msg)
264{
265	struct irq_desc *desc = irq_to_desc(irq);
266
267	write_msi_msg_desc(desc, msg);
268}
269
270static int msi_free_irqs(struct pci_dev* dev);
271
272static struct msi_desc *alloc_msi_entry(struct pci_dev *dev)
273{
274	struct msi_desc *desc = kzalloc(sizeof(*desc), GFP_KERNEL);
275	if (!desc)
276		return NULL;
277
278	INIT_LIST_HEAD(&desc->list);
279	desc->dev = dev;
280
281	return desc;
282}
283
284static void pci_intx_for_msi(struct pci_dev *dev, int enable)
285{
286	if (!(dev->dev_flags & PCI_DEV_FLAGS_MSI_INTX_DISABLE_BUG))
287		pci_intx(dev, enable);
288}
289
290static void __pci_restore_msi_state(struct pci_dev *dev)
291{
292	int pos;
293	u16 control;
294	struct msi_desc *entry;
295
296	if (!dev->msi_enabled)
297		return;
298
299	entry = get_irq_msi(dev->irq);
300	pos = entry->msi_attrib.pos;
301
302	pci_intx_for_msi(dev, 0);
303	msi_set_enable(dev, 0);
304	write_msi_msg(dev->irq, &entry->msg);
305
306	pci_read_config_word(dev, pos + PCI_MSI_FLAGS, &control);
307	msi_mask_irq(entry, msi_capable_mask(control), entry->masked);
308	control &= ~PCI_MSI_FLAGS_QSIZE;
309	control |= (entry->msi_attrib.multiple << 4) | PCI_MSI_FLAGS_ENABLE;
310	pci_write_config_word(dev, pos + PCI_MSI_FLAGS, control);
311}
312
313static void __pci_restore_msix_state(struct pci_dev *dev)
314{
315	int pos;
316	struct msi_desc *entry;
317	u16 control;
318
319	if (!dev->msix_enabled)
320		return;
321
322	/* route the table */
323	pci_intx_for_msi(dev, 0);
324	msix_set_enable(dev, 0);
325
326	list_for_each_entry(entry, &dev->msi_list, list) {
327		write_msi_msg(entry->irq, &entry->msg);
328		msix_mask_irq(entry, entry->masked);
329	}
330
331	BUG_ON(list_empty(&dev->msi_list));
332	entry = list_entry(dev->msi_list.next, struct msi_desc, list);
333	pos = entry->msi_attrib.pos;
334	pci_read_config_word(dev, pos + PCI_MSIX_FLAGS, &control);
335	control &= ~PCI_MSIX_FLAGS_MASKALL;
336	control |= PCI_MSIX_FLAGS_ENABLE;
337	pci_write_config_word(dev, pos + PCI_MSIX_FLAGS, control);
338}
339
340void pci_restore_msi_state(struct pci_dev *dev)
341{
342	__pci_restore_msi_state(dev);
343	__pci_restore_msix_state(dev);
344}
345EXPORT_SYMBOL_GPL(pci_restore_msi_state);
346
347/**
348 * msi_capability_init - configure device's MSI capability structure
349 * @dev: pointer to the pci_dev data structure of MSI device function
350 * @nvec: number of interrupts to allocate
351 *
352 * Setup the MSI capability structure of the device with the requested
353 * number of interrupts.  A return value of zero indicates the successful
354 * setup of an entry with the new MSI irq.  A negative return value indicates
355 * an error, and a positive return value indicates the number of interrupts
356 * which could have been allocated.
357 */
358static int msi_capability_init(struct pci_dev *dev, int nvec)
359{
360	struct msi_desc *entry;
361	int pos, ret;
362	u16 control;
363	unsigned mask;
364
365	msi_set_enable(dev, 0);	/* Ensure msi is disabled as I set it up */
366
367   	pos = pci_find_capability(dev, PCI_CAP_ID_MSI);
368	pci_read_config_word(dev, msi_control_reg(pos), &control);
369	/* MSI Entry Initialization */
370	entry = alloc_msi_entry(dev);
371	if (!entry)
372		return -ENOMEM;
373
374	entry->msi_attrib.is_msix = 0;
375	entry->msi_attrib.is_64 = is_64bit_address(control);
376	entry->msi_attrib.entry_nr = 0;
377	entry->msi_attrib.maskbit = is_mask_bit_support(control);
378	entry->msi_attrib.default_irq = dev->irq;	/* Save IOAPIC IRQ */
379	entry->msi_attrib.pos = pos;
380
381	entry->mask_pos = msi_mask_reg(pos, entry->msi_attrib.is_64);
382	/* All MSIs are unmasked by default, Mask them all */
383	if (entry->msi_attrib.maskbit)
384		pci_read_config_dword(dev, entry->mask_pos, &entry->masked);
385	mask = msi_capable_mask(control);
386	msi_mask_irq(entry, mask, mask);
387
388	list_add_tail(&entry->list, &dev->msi_list);
389
390	/* Configure MSI capability structure */
391	ret = arch_setup_msi_irqs(dev, nvec, PCI_CAP_ID_MSI);
392	if (ret) {
393		msi_free_irqs(dev);
394		return ret;
395	}
396
397	/* Set MSI enabled bits	 */
398	pci_intx_for_msi(dev, 0);
399	msi_set_enable(dev, 1);
400	dev->msi_enabled = 1;
401
402	dev->irq = entry->irq;
403	return 0;
404}
405
406/**
407 * msix_capability_init - configure device's MSI-X capability
408 * @dev: pointer to the pci_dev data structure of MSI-X device function
409 * @entries: pointer to an array of struct msix_entry entries
410 * @nvec: number of @entries
411 *
412 * Setup the MSI-X capability structure of device function with a
413 * single MSI-X irq. A return of zero indicates the successful setup of
414 * requested MSI-X entries with allocated irqs or non-zero for otherwise.
415 **/
416static int msix_capability_init(struct pci_dev *dev,
417				struct msix_entry *entries, int nvec)
418{
419	struct msi_desc *entry;
420	int pos, i, j, nr_entries, ret;
421	unsigned long phys_addr;
422	u32 table_offset;
423 	u16 control;
424	u8 bir;
425	void __iomem *base;
426
427	msix_set_enable(dev, 0);/* Ensure msix is disabled as I set it up */
428
429   	pos = pci_find_capability(dev, PCI_CAP_ID_MSIX);
430	/* Request & Map MSI-X table region */
431 	pci_read_config_word(dev, msi_control_reg(pos), &control);
432	nr_entries = multi_msix_capable(control);
433
434 	pci_read_config_dword(dev, msix_table_offset_reg(pos), &table_offset);
435	bir = (u8)(table_offset & PCI_MSIX_FLAGS_BIRMASK);
436	table_offset &= ~PCI_MSIX_FLAGS_BIRMASK;
437	phys_addr = pci_resource_start (dev, bir) + table_offset;
438	base = ioremap_nocache(phys_addr, nr_entries * PCI_MSIX_ENTRY_SIZE);
439	if (base == NULL)
440		return -ENOMEM;
441
442	/* MSI-X Table Initialization */
443	for (i = 0; i < nvec; i++) {
444		entry = alloc_msi_entry(dev);
445		if (!entry)
446			break;
447
448 		j = entries[i].entry;
449		entry->msi_attrib.is_msix = 1;
450		entry->msi_attrib.is_64 = 1;
451		entry->msi_attrib.entry_nr = j;
452		entry->msi_attrib.default_irq = dev->irq;
453		entry->msi_attrib.pos = pos;
454		entry->mask_base = base;
455		msix_mask_irq(entry, 1);
456
457		list_add_tail(&entry->list, &dev->msi_list);
458	}
459
460	ret = arch_setup_msi_irqs(dev, nvec, PCI_CAP_ID_MSIX);
461	if (ret < 0) {
462		/* If we had some success report the number of irqs
463		 * we succeeded in setting up. */
464		int avail = 0;
465		list_for_each_entry(entry, &dev->msi_list, list) {
466			if (entry->irq != 0) {
467				avail++;
468			}
469		}
470
471		if (avail != 0)
472			ret = avail;
473	}
474
475	if (ret) {
476		msi_free_irqs(dev);
477		return ret;
478	}
479
480	i = 0;
481	list_for_each_entry(entry, &dev->msi_list, list) {
482		entries[i].vector = entry->irq;
483		set_irq_msi(entry->irq, entry);
484		i++;
485	}
486	/* Set MSI-X enabled bits */
487	pci_intx_for_msi(dev, 0);
488	msix_set_enable(dev, 1);
489	dev->msix_enabled = 1;
490
491	list_for_each_entry(entry, &dev->msi_list, list) {
492		int vector = entry->msi_attrib.entry_nr;
493		entry->masked = readl(base + vector * PCI_MSIX_ENTRY_SIZE +
494					PCI_MSIX_ENTRY_VECTOR_CTRL_OFFSET);
495	}
496
497	return 0;
498}
499
500/**
501 * pci_msi_check_device - check whether MSI may be enabled on a device
502 * @dev: pointer to the pci_dev data structure of MSI device function
503 * @nvec: how many MSIs have been requested ?
504 * @type: are we checking for MSI or MSI-X ?
505 *
506 * Look at global flags, the device itself, and its parent busses
507 * to determine if MSI/-X are supported for the device. If MSI/-X is
508 * supported return 0, else return an error code.
509 **/
510static int pci_msi_check_device(struct pci_dev* dev, int nvec, int type)
511{
512	struct pci_bus *bus;
513	int ret;
514
515	/* MSI must be globally enabled and supported by the device */
516	if (!pci_msi_enable || !dev || dev->no_msi)
517		return -EINVAL;
518
519	/*
520	 * You can't ask to have 0 or less MSIs configured.
521	 *  a) it's stupid ..
522	 *  b) the list manipulation code assumes nvec >= 1.
523	 */
524	if (nvec < 1)
525		return -ERANGE;
526
527	/* Any bridge which does NOT route MSI transactions from it's
528	 * secondary bus to it's primary bus must set NO_MSI flag on
529	 * the secondary pci_bus.
530	 * We expect only arch-specific PCI host bus controller driver
531	 * or quirks for specific PCI bridges to be setting NO_MSI.
532	 */
533	for (bus = dev->bus; bus; bus = bus->parent)
534		if (bus->bus_flags & PCI_BUS_FLAGS_NO_MSI)
535			return -EINVAL;
536
537	ret = arch_msi_check_device(dev, nvec, type);
538	if (ret)
539		return ret;
540
541	if (!pci_find_capability(dev, type))
542		return -EINVAL;
543
544	return 0;
545}
546
547/**
548 * pci_enable_msi_block - configure device's MSI capability structure
549 * @dev: device to configure
550 * @nvec: number of interrupts to configure
551 *
552 * Allocate IRQs for a device with the MSI capability.
553 * This function returns a negative errno if an error occurs.  If it
554 * is unable to allocate the number of interrupts requested, it returns
555 * the number of interrupts it might be able to allocate.  If it successfully
556 * allocates at least the number of interrupts requested, it returns 0 and
557 * updates the @dev's irq member to the lowest new interrupt number; the
558 * other interrupt numbers allocated to this device are consecutive.
559 */
560int pci_enable_msi_block(struct pci_dev *dev, unsigned int nvec)
561{
562	int status, pos, maxvec;
563	u16 msgctl;
564
565	pos = pci_find_capability(dev, PCI_CAP_ID_MSI);
566	if (!pos)
567		return -EINVAL;
568	pci_read_config_word(dev, pos + PCI_MSI_FLAGS, &msgctl);
569	maxvec = 1 << ((msgctl & PCI_MSI_FLAGS_QMASK) >> 1);
570	if (nvec > maxvec)
571		return maxvec;
572
573	status = pci_msi_check_device(dev, nvec, PCI_CAP_ID_MSI);
574	if (status)
575		return status;
576
577	WARN_ON(!!dev->msi_enabled);
578
579	/* Check whether driver already requested MSI-X irqs */
580	if (dev->msix_enabled) {
581		dev_info(&dev->dev, "can't enable MSI "
582			 "(MSI-X already enabled)\n");
583		return -EINVAL;
584	}
585
586	status = msi_capability_init(dev, nvec);
587	return status;
588}
589EXPORT_SYMBOL(pci_enable_msi_block);
590
591void pci_msi_shutdown(struct pci_dev *dev)
592{
593	struct msi_desc *desc;
594	u32 mask;
595	u16 ctrl;
596
597	if (!pci_msi_enable || !dev || !dev->msi_enabled)
598		return;
599
600	msi_set_enable(dev, 0);
601	pci_intx_for_msi(dev, 1);
602	dev->msi_enabled = 0;
603
604	BUG_ON(list_empty(&dev->msi_list));
605	desc = list_first_entry(&dev->msi_list, struct msi_desc, list);
606	pci_read_config_word(dev, desc->msi_attrib.pos + PCI_MSI_FLAGS, &ctrl);
607	mask = msi_capable_mask(ctrl);
608	msi_mask_irq(desc, mask, ~mask);
609
610	/* Restore dev->irq to its default pin-assertion irq */
611	dev->irq = desc->msi_attrib.default_irq;
612}
613
614void pci_disable_msi(struct pci_dev* dev)
615{
616	struct msi_desc *entry;
617
618	if (!pci_msi_enable || !dev || !dev->msi_enabled)
619		return;
620
621	pci_msi_shutdown(dev);
622
623	entry = list_entry(dev->msi_list.next, struct msi_desc, list);
624	if (entry->msi_attrib.is_msix)
625		return;
626
627	msi_free_irqs(dev);
628}
629EXPORT_SYMBOL(pci_disable_msi);
630
631static int msi_free_irqs(struct pci_dev* dev)
632{
633	struct msi_desc *entry, *tmp;
634
635	list_for_each_entry(entry, &dev->msi_list, list) {
636		int i, nvec;
637		if (!entry->irq)
638			continue;
639		nvec = 1 << entry->msi_attrib.multiple;
640		for (i = 0; i < nvec; i++)
641			BUG_ON(irq_has_action(entry->irq + i));
642	}
643
644	arch_teardown_msi_irqs(dev);
645
646	list_for_each_entry_safe(entry, tmp, &dev->msi_list, list) {
647		if (entry->msi_attrib.is_msix) {
648			writel(1, entry->mask_base + entry->msi_attrib.entry_nr
649				  * PCI_MSIX_ENTRY_SIZE
650				  + PCI_MSIX_ENTRY_VECTOR_CTRL_OFFSET);
651
652			if (list_is_last(&entry->list, &dev->msi_list))
653				iounmap(entry->mask_base);
654		}
655		list_del(&entry->list);
656		kfree(entry);
657	}
658
659	return 0;
660}
661
662/**
663 * pci_msix_table_size - return the number of device's MSI-X table entries
664 * @dev: pointer to the pci_dev data structure of MSI-X device function
665 */
666int pci_msix_table_size(struct pci_dev *dev)
667{
668	int pos;
669	u16 control;
670
671	pos = pci_find_capability(dev, PCI_CAP_ID_MSIX);
672	if (!pos)
673		return 0;
674
675	pci_read_config_word(dev, msi_control_reg(pos), &control);
676	return multi_msix_capable(control);
677}
678
679/**
680 * pci_enable_msix - configure device's MSI-X capability structure
681 * @dev: pointer to the pci_dev data structure of MSI-X device function
682 * @entries: pointer to an array of MSI-X entries
683 * @nvec: number of MSI-X irqs requested for allocation by device driver
684 *
685 * Setup the MSI-X capability structure of device function with the number
686 * of requested irqs upon its software driver call to request for
687 * MSI-X mode enabled on its hardware device function. A return of zero
688 * indicates the successful configuration of MSI-X capability structure
689 * with new allocated MSI-X irqs. A return of < 0 indicates a failure.
690 * Or a return of > 0 indicates that driver request is exceeding the number
691 * of irqs or MSI-X vectors available. Driver should use the returned value to
692 * re-send its request.
693 **/
694int pci_enable_msix(struct pci_dev* dev, struct msix_entry *entries, int nvec)
695{
696	int status, nr_entries;
697	int i, j;
698
699	if (!entries)
700 		return -EINVAL;
701
702	status = pci_msi_check_device(dev, nvec, PCI_CAP_ID_MSIX);
703	if (status)
704		return status;
705
706	nr_entries = pci_msix_table_size(dev);
707	if (nvec > nr_entries)
708		return nr_entries;
709
710	/* Check for any invalid entries */
711	for (i = 0; i < nvec; i++) {
712		if (entries[i].entry >= nr_entries)
713			return -EINVAL;		/* invalid entry */
714		for (j = i + 1; j < nvec; j++) {
715			if (entries[i].entry == entries[j].entry)
716				return -EINVAL;	/* duplicate entry */
717		}
718	}
719	WARN_ON(!!dev->msix_enabled);
720
721	/* Check whether driver already requested for MSI irq */
722   	if (dev->msi_enabled) {
723		dev_info(&dev->dev, "can't enable MSI-X "
724		       "(MSI IRQ already assigned)\n");
725		return -EINVAL;
726	}
727	status = msix_capability_init(dev, entries, nvec);
728	return status;
729}
730EXPORT_SYMBOL(pci_enable_msix);
731
732static void msix_free_all_irqs(struct pci_dev *dev)
733{
734	msi_free_irqs(dev);
735}
736
737void pci_msix_shutdown(struct pci_dev* dev)
738{
739	if (!pci_msi_enable || !dev || !dev->msix_enabled)
740		return;
741
742	msix_set_enable(dev, 0);
743	pci_intx_for_msi(dev, 1);
744	dev->msix_enabled = 0;
745}
746void pci_disable_msix(struct pci_dev* dev)
747{
748	if (!pci_msi_enable || !dev || !dev->msix_enabled)
749		return;
750
751	pci_msix_shutdown(dev);
752
753	msix_free_all_irqs(dev);
754}
755EXPORT_SYMBOL(pci_disable_msix);
756
757/**
758 * msi_remove_pci_irq_vectors - reclaim MSI(X) irqs to unused state
759 * @dev: pointer to the pci_dev data structure of MSI(X) device function
760 *
761 * Being called during hotplug remove, from which the device function
762 * is hot-removed. All previous assigned MSI/MSI-X irqs, if
763 * allocated for this device function, are reclaimed to unused state,
764 * which may be used later on.
765 **/
766void msi_remove_pci_irq_vectors(struct pci_dev* dev)
767{
768	if (!pci_msi_enable || !dev)
769 		return;
770
771	if (dev->msi_enabled)
772		msi_free_irqs(dev);
773
774	if (dev->msix_enabled)
775		msix_free_all_irqs(dev);
776}
777
778void pci_no_msi(void)
779{
780	pci_msi_enable = 0;
781}
782
783/**
784 * pci_msi_enabled - is MSI enabled?
785 *
786 * Returns true if MSI has not been disabled by the command-line option
787 * pci=nomsi.
788 **/
789int pci_msi_enabled(void)
790{
791	return pci_msi_enable;
792}
793EXPORT_SYMBOL(pci_msi_enabled);
794
795void pci_msi_init_pci_dev(struct pci_dev *dev)
796{
797	INIT_LIST_HEAD(&dev->msi_list);
798}
799