msi.c revision 12abb8ba8444f7c9b355bbdd44a6d0839f7a41b6
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	BUG_ON(!pos);
83
84	pci_read_config_word(dev, pos + PCI_MSI_FLAGS, &control);
85	control &= ~PCI_MSI_FLAGS_ENABLE;
86	if (enable)
87		control |= PCI_MSI_FLAGS_ENABLE;
88	pci_write_config_word(dev, pos + PCI_MSI_FLAGS, control);
89}
90
91static void msix_set_enable(struct pci_dev *dev, int enable)
92{
93	int pos;
94	u16 control;
95
96	pos = pci_find_capability(dev, PCI_CAP_ID_MSIX);
97	if (pos) {
98		pci_read_config_word(dev, pos + PCI_MSIX_FLAGS, &control);
99		control &= ~PCI_MSIX_FLAGS_ENABLE;
100		if (enable)
101			control |= PCI_MSIX_FLAGS_ENABLE;
102		pci_write_config_word(dev, pos + PCI_MSIX_FLAGS, control);
103	}
104}
105
106static inline __attribute_const__ u32 msi_mask(unsigned x)
107{
108	/* Don't shift by >= width of type */
109	if (x >= 5)
110		return 0xffffffff;
111	return (1 << (1 << x)) - 1;
112}
113
114static inline __attribute_const__ u32 msi_capable_mask(u16 control)
115{
116	return msi_mask((control >> 1) & 7);
117}
118
119static inline __attribute_const__ u32 msi_enabled_mask(u16 control)
120{
121	return msi_mask((control >> 4) & 7);
122}
123
124/*
125 * PCI 2.3 does not specify mask bits for each MSI interrupt.  Attempting to
126 * mask all MSI interrupts by clearing the MSI enable bit does not work
127 * reliably as devices without an INTx disable bit will then generate a
128 * level IRQ which will never be cleared.
129 */
130static u32 __msi_mask_irq(struct msi_desc *desc, u32 mask, u32 flag)
131{
132	u32 mask_bits = desc->masked;
133
134	if (!desc->msi_attrib.maskbit)
135		return 0;
136
137	mask_bits &= ~mask;
138	mask_bits |= flag;
139	pci_write_config_dword(desc->dev, desc->mask_pos, mask_bits);
140
141	return mask_bits;
142}
143
144static void msi_mask_irq(struct msi_desc *desc, u32 mask, u32 flag)
145{
146	desc->masked = __msi_mask_irq(desc, mask, flag);
147}
148
149/*
150 * This internal function does not flush PCI writes to the device.
151 * All users must ensure that they read from the device before either
152 * assuming that the device state is up to date, or returning out of this
153 * file.  This saves a few milliseconds when initialising devices with lots
154 * of MSI-X interrupts.
155 */
156static u32 __msix_mask_irq(struct msi_desc *desc, u32 flag)
157{
158	u32 mask_bits = desc->masked;
159	unsigned offset = desc->msi_attrib.entry_nr * PCI_MSIX_ENTRY_SIZE +
160						PCI_MSIX_ENTRY_VECTOR_CTRL;
161	mask_bits &= ~1;
162	mask_bits |= flag;
163	writel(mask_bits, desc->mask_base + offset);
164
165	return mask_bits;
166}
167
168static void msix_mask_irq(struct msi_desc *desc, u32 flag)
169{
170	desc->masked = __msix_mask_irq(desc, flag);
171}
172
173static void msi_set_mask_bit(unsigned irq, u32 flag)
174{
175	struct msi_desc *desc = get_irq_msi(irq);
176
177	if (desc->msi_attrib.is_msix) {
178		msix_mask_irq(desc, flag);
179		readl(desc->mask_base);		/* Flush write to device */
180	} else {
181		unsigned offset = irq - desc->dev->irq;
182		msi_mask_irq(desc, 1 << offset, flag << offset);
183	}
184}
185
186void mask_msi_irq(unsigned int irq)
187{
188	msi_set_mask_bit(irq, 1);
189}
190
191void unmask_msi_irq(unsigned int irq)
192{
193	msi_set_mask_bit(irq, 0);
194}
195
196void read_msi_msg_desc(struct irq_desc *desc, struct msi_msg *msg)
197{
198	struct msi_desc *entry = get_irq_desc_msi(desc);
199	if (entry->msi_attrib.is_msix) {
200		void __iomem *base = entry->mask_base +
201			entry->msi_attrib.entry_nr * PCI_MSIX_ENTRY_SIZE;
202
203		msg->address_lo = readl(base + PCI_MSIX_ENTRY_LOWER_ADDR);
204		msg->address_hi = readl(base + PCI_MSIX_ENTRY_UPPER_ADDR);
205		msg->data = readl(base + PCI_MSIX_ENTRY_DATA);
206	} else {
207		struct pci_dev *dev = entry->dev;
208		int pos = entry->msi_attrib.pos;
209		u16 data;
210
211		pci_read_config_dword(dev, msi_lower_address_reg(pos),
212					&msg->address_lo);
213		if (entry->msi_attrib.is_64) {
214			pci_read_config_dword(dev, msi_upper_address_reg(pos),
215						&msg->address_hi);
216			pci_read_config_word(dev, msi_data_reg(pos, 1), &data);
217		} else {
218			msg->address_hi = 0;
219			pci_read_config_word(dev, msi_data_reg(pos, 0), &data);
220		}
221		msg->data = data;
222	}
223}
224
225void read_msi_msg(unsigned int irq, struct msi_msg *msg)
226{
227	struct irq_desc *desc = irq_to_desc(irq);
228
229	read_msi_msg_desc(desc, msg);
230}
231
232void write_msi_msg_desc(struct irq_desc *desc, struct msi_msg *msg)
233{
234	struct msi_desc *entry = get_irq_desc_msi(desc);
235	if (entry->msi_attrib.is_msix) {
236		void __iomem *base;
237		base = entry->mask_base +
238			entry->msi_attrib.entry_nr * PCI_MSIX_ENTRY_SIZE;
239
240		writel(msg->address_lo, base + PCI_MSIX_ENTRY_LOWER_ADDR);
241		writel(msg->address_hi, base + PCI_MSIX_ENTRY_UPPER_ADDR);
242		writel(msg->data, base + PCI_MSIX_ENTRY_DATA);
243	} else {
244		struct pci_dev *dev = entry->dev;
245		int pos = entry->msi_attrib.pos;
246		u16 msgctl;
247
248		pci_read_config_word(dev, msi_control_reg(pos), &msgctl);
249		msgctl &= ~PCI_MSI_FLAGS_QSIZE;
250		msgctl |= entry->msi_attrib.multiple << 4;
251		pci_write_config_word(dev, msi_control_reg(pos), msgctl);
252
253		pci_write_config_dword(dev, msi_lower_address_reg(pos),
254					msg->address_lo);
255		if (entry->msi_attrib.is_64) {
256			pci_write_config_dword(dev, msi_upper_address_reg(pos),
257						msg->address_hi);
258			pci_write_config_word(dev, msi_data_reg(pos, 1),
259						msg->data);
260		} else {
261			pci_write_config_word(dev, msi_data_reg(pos, 0),
262						msg->data);
263		}
264	}
265	entry->msg = *msg;
266}
267
268void write_msi_msg(unsigned int irq, struct msi_msg *msg)
269{
270	struct irq_desc *desc = irq_to_desc(irq);
271
272	write_msi_msg_desc(desc, msg);
273}
274
275static int msi_free_irqs(struct pci_dev* dev);
276
277static struct msi_desc *alloc_msi_entry(struct pci_dev *dev)
278{
279	struct msi_desc *desc = kzalloc(sizeof(*desc), GFP_KERNEL);
280	if (!desc)
281		return NULL;
282
283	INIT_LIST_HEAD(&desc->list);
284	desc->dev = dev;
285
286	return desc;
287}
288
289static void pci_intx_for_msi(struct pci_dev *dev, int enable)
290{
291	if (!(dev->dev_flags & PCI_DEV_FLAGS_MSI_INTX_DISABLE_BUG))
292		pci_intx(dev, enable);
293}
294
295static void __pci_restore_msi_state(struct pci_dev *dev)
296{
297	int pos;
298	u16 control;
299	struct msi_desc *entry;
300
301	if (!dev->msi_enabled)
302		return;
303
304	entry = get_irq_msi(dev->irq);
305	pos = entry->msi_attrib.pos;
306
307	pci_intx_for_msi(dev, 0);
308	msi_set_enable(dev, pos, 0);
309	write_msi_msg(dev->irq, &entry->msg);
310
311	pci_read_config_word(dev, pos + PCI_MSI_FLAGS, &control);
312	msi_mask_irq(entry, msi_capable_mask(control), entry->masked);
313	control &= ~PCI_MSI_FLAGS_QSIZE;
314	control |= (entry->msi_attrib.multiple << 4) | PCI_MSI_FLAGS_ENABLE;
315	pci_write_config_word(dev, pos + PCI_MSI_FLAGS, control);
316}
317
318static void __pci_restore_msix_state(struct pci_dev *dev)
319{
320	int pos;
321	struct msi_desc *entry;
322	u16 control;
323
324	if (!dev->msix_enabled)
325		return;
326	BUG_ON(list_empty(&dev->msi_list));
327	entry = list_entry(dev->msi_list.next, struct msi_desc, list);
328	pos = entry->msi_attrib.pos;
329	pci_read_config_word(dev, pos + PCI_MSIX_FLAGS, &control);
330
331	/* route the table */
332	pci_intx_for_msi(dev, 0);
333	control |= PCI_MSIX_FLAGS_ENABLE | PCI_MSIX_FLAGS_MASKALL;
334	pci_write_config_word(dev, pos + PCI_MSIX_FLAGS, control);
335
336	list_for_each_entry(entry, &dev->msi_list, list) {
337		write_msi_msg(entry->irq, &entry->msg);
338		msix_mask_irq(entry, entry->masked);
339	}
340
341	control &= ~PCI_MSIX_FLAGS_MASKALL;
342	pci_write_config_word(dev, pos + PCI_MSIX_FLAGS, control);
343}
344
345void pci_restore_msi_state(struct pci_dev *dev)
346{
347	__pci_restore_msi_state(dev);
348	__pci_restore_msix_state(dev);
349}
350EXPORT_SYMBOL_GPL(pci_restore_msi_state);
351
352/**
353 * msi_capability_init - configure device's MSI capability structure
354 * @dev: pointer to the pci_dev data structure of MSI device function
355 * @nvec: number of interrupts to allocate
356 *
357 * Setup the MSI capability structure of the device with the requested
358 * number of interrupts.  A return value of zero indicates the successful
359 * setup of an entry with the new MSI irq.  A negative return value indicates
360 * an error, and a positive return value indicates the number of interrupts
361 * which could have been allocated.
362 */
363static int msi_capability_init(struct pci_dev *dev, int nvec)
364{
365	struct msi_desc *entry;
366	int pos, ret;
367	u16 control;
368	unsigned mask;
369
370   	pos = pci_find_capability(dev, PCI_CAP_ID_MSI);
371	msi_set_enable(dev, pos, 0);	/* Disable MSI during set up */
372
373	pci_read_config_word(dev, msi_control_reg(pos), &control);
374	/* MSI Entry Initialization */
375	entry = alloc_msi_entry(dev);
376	if (!entry)
377		return -ENOMEM;
378
379	entry->msi_attrib.is_msix = 0;
380	entry->msi_attrib.is_64 = is_64bit_address(control);
381	entry->msi_attrib.entry_nr = 0;
382	entry->msi_attrib.maskbit = is_mask_bit_support(control);
383	entry->msi_attrib.default_irq = dev->irq;	/* Save IOAPIC IRQ */
384	entry->msi_attrib.pos = pos;
385
386	entry->mask_pos = msi_mask_reg(pos, entry->msi_attrib.is_64);
387	/* All MSIs are unmasked by default, Mask them all */
388	if (entry->msi_attrib.maskbit)
389		pci_read_config_dword(dev, entry->mask_pos, &entry->masked);
390	mask = msi_capable_mask(control);
391	msi_mask_irq(entry, mask, mask);
392
393	list_add_tail(&entry->list, &dev->msi_list);
394
395	/* Configure MSI capability structure */
396	ret = arch_setup_msi_irqs(dev, nvec, PCI_CAP_ID_MSI);
397	if (ret) {
398		msi_mask_irq(entry, mask, ~mask);
399		msi_free_irqs(dev);
400		return ret;
401	}
402
403	/* Set MSI enabled bits	 */
404	pci_intx_for_msi(dev, 0);
405	msi_set_enable(dev, pos, 1);
406	dev->msi_enabled = 1;
407
408	dev->irq = entry->irq;
409	return 0;
410}
411
412/**
413 * msix_capability_init - configure device's MSI-X capability
414 * @dev: pointer to the pci_dev data structure of MSI-X device function
415 * @entries: pointer to an array of struct msix_entry entries
416 * @nvec: number of @entries
417 *
418 * Setup the MSI-X capability structure of device function with a
419 * single MSI-X irq. A return of zero indicates the successful setup of
420 * requested MSI-X entries with allocated irqs or non-zero for otherwise.
421 **/
422static int msix_capability_init(struct pci_dev *dev,
423				struct msix_entry *entries, int nvec)
424{
425	struct msi_desc *entry;
426	int pos, i, j, nr_entries, ret;
427	unsigned long phys_addr;
428	u32 table_offset;
429 	u16 control;
430	u8 bir;
431	void __iomem *base;
432
433   	pos = pci_find_capability(dev, PCI_CAP_ID_MSIX);
434	pci_read_config_word(dev, pos + PCI_MSIX_FLAGS, &control);
435
436	/* Ensure MSI-X is disabled while it is set up */
437	control &= ~PCI_MSIX_FLAGS_ENABLE;
438	pci_write_config_word(dev, pos + PCI_MSIX_FLAGS, control);
439
440	/* Request & Map MSI-X table region */
441	nr_entries = multi_msix_capable(control);
442
443 	pci_read_config_dword(dev, msix_table_offset_reg(pos), &table_offset);
444	bir = (u8)(table_offset & PCI_MSIX_FLAGS_BIRMASK);
445	table_offset &= ~PCI_MSIX_FLAGS_BIRMASK;
446	phys_addr = pci_resource_start (dev, bir) + table_offset;
447	base = ioremap_nocache(phys_addr, nr_entries * PCI_MSIX_ENTRY_SIZE);
448	if (base == NULL)
449		return -ENOMEM;
450
451	for (i = 0; i < nvec; i++) {
452		entry = alloc_msi_entry(dev);
453		if (!entry) {
454			if (!i)
455				iounmap(base);
456			else
457				msi_free_irqs(dev);
458			/* No enough memory. Don't try again */
459			return -ENOMEM;
460		}
461
462 		j = entries[i].entry;
463		entry->msi_attrib.is_msix = 1;
464		entry->msi_attrib.is_64 = 1;
465		entry->msi_attrib.entry_nr = j;
466		entry->msi_attrib.default_irq = dev->irq;
467		entry->msi_attrib.pos = pos;
468		entry->mask_base = base;
469
470		list_add_tail(&entry->list, &dev->msi_list);
471	}
472
473	ret = arch_setup_msi_irqs(dev, nvec, PCI_CAP_ID_MSIX);
474	if (ret < 0) {
475		/* If we had some success report the number of irqs
476		 * we succeeded in setting up. */
477		int avail = 0;
478		list_for_each_entry(entry, &dev->msi_list, list) {
479			if (entry->irq != 0) {
480				avail++;
481			}
482		}
483
484		if (avail != 0)
485			ret = avail;
486	}
487
488	if (ret) {
489		msi_free_irqs(dev);
490		return ret;
491	}
492
493	/*
494	 * Some devices require MSI-X to be enabled before we can touch the
495	 * MSI-X registers.  We need to mask all the vectors to prevent
496	 * interrupts coming in before they're fully set up.
497	 */
498	control |= PCI_MSIX_FLAGS_MASKALL | PCI_MSIX_FLAGS_ENABLE;
499	pci_write_config_word(dev, pos + PCI_MSIX_FLAGS, control);
500
501	i = 0;
502	list_for_each_entry(entry, &dev->msi_list, list) {
503		entries[i].vector = entry->irq;
504		set_irq_msi(entry->irq, entry);
505		j = entries[i].entry;
506		entry->masked = readl(base + j * PCI_MSIX_ENTRY_SIZE +
507						PCI_MSIX_ENTRY_VECTOR_CTRL);
508		msix_mask_irq(entry, 1);
509		i++;
510	}
511
512	/* Set MSI-X enabled bits and unmask the function */
513	pci_intx_for_msi(dev, 0);
514	dev->msix_enabled = 1;
515
516	control &= ~PCI_MSIX_FLAGS_MASKALL;
517	pci_write_config_word(dev, pos + PCI_MSIX_FLAGS, control);
518
519	return 0;
520}
521
522/**
523 * pci_msi_check_device - check whether MSI may be enabled on a device
524 * @dev: pointer to the pci_dev data structure of MSI device function
525 * @nvec: how many MSIs have been requested ?
526 * @type: are we checking for MSI or MSI-X ?
527 *
528 * Look at global flags, the device itself, and its parent busses
529 * to determine if MSI/-X are supported for the device. If MSI/-X is
530 * supported return 0, else return an error code.
531 **/
532static int pci_msi_check_device(struct pci_dev* dev, int nvec, int type)
533{
534	struct pci_bus *bus;
535	int ret;
536
537	/* MSI must be globally enabled and supported by the device */
538	if (!pci_msi_enable || !dev || dev->no_msi)
539		return -EINVAL;
540
541	/*
542	 * You can't ask to have 0 or less MSIs configured.
543	 *  a) it's stupid ..
544	 *  b) the list manipulation code assumes nvec >= 1.
545	 */
546	if (nvec < 1)
547		return -ERANGE;
548
549	/* Any bridge which does NOT route MSI transactions from it's
550	 * secondary bus to it's primary bus must set NO_MSI flag on
551	 * the secondary pci_bus.
552	 * We expect only arch-specific PCI host bus controller driver
553	 * or quirks for specific PCI bridges to be setting NO_MSI.
554	 */
555	for (bus = dev->bus; bus; bus = bus->parent)
556		if (bus->bus_flags & PCI_BUS_FLAGS_NO_MSI)
557			return -EINVAL;
558
559	ret = arch_msi_check_device(dev, nvec, type);
560	if (ret)
561		return ret;
562
563	if (!pci_find_capability(dev, type))
564		return -EINVAL;
565
566	return 0;
567}
568
569/**
570 * pci_enable_msi_block - configure device's MSI capability structure
571 * @dev: device to configure
572 * @nvec: number of interrupts to configure
573 *
574 * Allocate IRQs for a device with the MSI capability.
575 * This function returns a negative errno if an error occurs.  If it
576 * is unable to allocate the number of interrupts requested, it returns
577 * the number of interrupts it might be able to allocate.  If it successfully
578 * allocates at least the number of interrupts requested, it returns 0 and
579 * updates the @dev's irq member to the lowest new interrupt number; the
580 * other interrupt numbers allocated to this device are consecutive.
581 */
582int pci_enable_msi_block(struct pci_dev *dev, unsigned int nvec)
583{
584	int status, pos, maxvec;
585	u16 msgctl;
586
587	pos = pci_find_capability(dev, PCI_CAP_ID_MSI);
588	if (!pos)
589		return -EINVAL;
590	pci_read_config_word(dev, pos + PCI_MSI_FLAGS, &msgctl);
591	maxvec = 1 << ((msgctl & PCI_MSI_FLAGS_QMASK) >> 1);
592	if (nvec > maxvec)
593		return maxvec;
594
595	status = pci_msi_check_device(dev, nvec, PCI_CAP_ID_MSI);
596	if (status)
597		return status;
598
599	WARN_ON(!!dev->msi_enabled);
600
601	/* Check whether driver already requested MSI-X irqs */
602	if (dev->msix_enabled) {
603		dev_info(&dev->dev, "can't enable MSI "
604			 "(MSI-X already enabled)\n");
605		return -EINVAL;
606	}
607
608	status = msi_capability_init(dev, nvec);
609	return status;
610}
611EXPORT_SYMBOL(pci_enable_msi_block);
612
613void pci_msi_shutdown(struct pci_dev *dev)
614{
615	struct msi_desc *desc;
616	u32 mask;
617	u16 ctrl;
618	unsigned pos;
619
620	if (!pci_msi_enable || !dev || !dev->msi_enabled)
621		return;
622
623	BUG_ON(list_empty(&dev->msi_list));
624	desc = list_first_entry(&dev->msi_list, struct msi_desc, list);
625	pos = desc->msi_attrib.pos;
626
627	msi_set_enable(dev, pos, 0);
628	pci_intx_for_msi(dev, 1);
629	dev->msi_enabled = 0;
630
631	/* Return the device with MSI unmasked as initial states */
632	pci_read_config_word(dev, pos + PCI_MSI_FLAGS, &ctrl);
633	mask = msi_capable_mask(ctrl);
634	/* Keep cached state to be restored */
635	__msi_mask_irq(desc, mask, ~mask);
636
637	/* Restore dev->irq to its default pin-assertion irq */
638	dev->irq = desc->msi_attrib.default_irq;
639}
640
641void pci_disable_msi(struct pci_dev* dev)
642{
643	struct msi_desc *entry;
644
645	if (!pci_msi_enable || !dev || !dev->msi_enabled)
646		return;
647
648	pci_msi_shutdown(dev);
649
650	entry = list_entry(dev->msi_list.next, struct msi_desc, list);
651	if (entry->msi_attrib.is_msix)
652		return;
653
654	msi_free_irqs(dev);
655}
656EXPORT_SYMBOL(pci_disable_msi);
657
658static int msi_free_irqs(struct pci_dev* dev)
659{
660	struct msi_desc *entry, *tmp;
661
662	list_for_each_entry(entry, &dev->msi_list, list) {
663		int i, nvec;
664		if (!entry->irq)
665			continue;
666		nvec = 1 << entry->msi_attrib.multiple;
667		for (i = 0; i < nvec; i++)
668			BUG_ON(irq_has_action(entry->irq + i));
669	}
670
671	arch_teardown_msi_irqs(dev);
672
673	list_for_each_entry_safe(entry, tmp, &dev->msi_list, list) {
674		if (entry->msi_attrib.is_msix) {
675			if (list_is_last(&entry->list, &dev->msi_list))
676				iounmap(entry->mask_base);
677		}
678		list_del(&entry->list);
679		kfree(entry);
680	}
681
682	return 0;
683}
684
685/**
686 * pci_msix_table_size - return the number of device's MSI-X table entries
687 * @dev: pointer to the pci_dev data structure of MSI-X device function
688 */
689int pci_msix_table_size(struct pci_dev *dev)
690{
691	int pos;
692	u16 control;
693
694	pos = pci_find_capability(dev, PCI_CAP_ID_MSIX);
695	if (!pos)
696		return 0;
697
698	pci_read_config_word(dev, msi_control_reg(pos), &control);
699	return multi_msix_capable(control);
700}
701
702/**
703 * pci_enable_msix - configure device's MSI-X capability structure
704 * @dev: pointer to the pci_dev data structure of MSI-X device function
705 * @entries: pointer to an array of MSI-X entries
706 * @nvec: number of MSI-X irqs requested for allocation by device driver
707 *
708 * Setup the MSI-X capability structure of device function with the number
709 * of requested irqs upon its software driver call to request for
710 * MSI-X mode enabled on its hardware device function. A return of zero
711 * indicates the successful configuration of MSI-X capability structure
712 * with new allocated MSI-X irqs. A return of < 0 indicates a failure.
713 * Or a return of > 0 indicates that driver request is exceeding the number
714 * of irqs or MSI-X vectors available. Driver should use the returned value to
715 * re-send its request.
716 **/
717int pci_enable_msix(struct pci_dev* dev, struct msix_entry *entries, int nvec)
718{
719	int status, nr_entries;
720	int i, j;
721
722	if (!entries)
723 		return -EINVAL;
724
725	status = pci_msi_check_device(dev, nvec, PCI_CAP_ID_MSIX);
726	if (status)
727		return status;
728
729	nr_entries = pci_msix_table_size(dev);
730	if (nvec > nr_entries)
731		return nr_entries;
732
733	/* Check for any invalid entries */
734	for (i = 0; i < nvec; i++) {
735		if (entries[i].entry >= nr_entries)
736			return -EINVAL;		/* invalid entry */
737		for (j = i + 1; j < nvec; j++) {
738			if (entries[i].entry == entries[j].entry)
739				return -EINVAL;	/* duplicate entry */
740		}
741	}
742	WARN_ON(!!dev->msix_enabled);
743
744	/* Check whether driver already requested for MSI irq */
745   	if (dev->msi_enabled) {
746		dev_info(&dev->dev, "can't enable MSI-X "
747		       "(MSI IRQ already assigned)\n");
748		return -EINVAL;
749	}
750	status = msix_capability_init(dev, entries, nvec);
751	return status;
752}
753EXPORT_SYMBOL(pci_enable_msix);
754
755static void msix_free_all_irqs(struct pci_dev *dev)
756{
757	msi_free_irqs(dev);
758}
759
760void pci_msix_shutdown(struct pci_dev* dev)
761{
762	struct msi_desc *entry;
763
764	if (!pci_msi_enable || !dev || !dev->msix_enabled)
765		return;
766
767	/* Return the device with MSI-X masked as initial states */
768	list_for_each_entry(entry, &dev->msi_list, list) {
769		/* Keep cached states to be restored */
770		__msix_mask_irq(entry, 1);
771	}
772
773	msix_set_enable(dev, 0);
774	pci_intx_for_msi(dev, 1);
775	dev->msix_enabled = 0;
776}
777void pci_disable_msix(struct pci_dev* dev)
778{
779	if (!pci_msi_enable || !dev || !dev->msix_enabled)
780		return;
781
782	pci_msix_shutdown(dev);
783
784	msix_free_all_irqs(dev);
785}
786EXPORT_SYMBOL(pci_disable_msix);
787
788/**
789 * msi_remove_pci_irq_vectors - reclaim MSI(X) irqs to unused state
790 * @dev: pointer to the pci_dev data structure of MSI(X) device function
791 *
792 * Being called during hotplug remove, from which the device function
793 * is hot-removed. All previous assigned MSI/MSI-X irqs, if
794 * allocated for this device function, are reclaimed to unused state,
795 * which may be used later on.
796 **/
797void msi_remove_pci_irq_vectors(struct pci_dev* dev)
798{
799	if (!pci_msi_enable || !dev)
800 		return;
801
802	if (dev->msi_enabled)
803		msi_free_irqs(dev);
804
805	if (dev->msix_enabled)
806		msix_free_all_irqs(dev);
807}
808
809void pci_no_msi(void)
810{
811	pci_msi_enable = 0;
812}
813
814/**
815 * pci_msi_enabled - is MSI enabled?
816 *
817 * Returns true if MSI has not been disabled by the command-line option
818 * pci=nomsi.
819 **/
820int pci_msi_enabled(void)
821{
822	return pci_msi_enable;
823}
824EXPORT_SYMBOL(pci_msi_enabled);
825
826void pci_msi_init_pci_dev(struct pci_dev *dev)
827{
828	INIT_LIST_HEAD(&dev->msi_list);
829}
830