3 * Purpose: PCI Message Signaled Interrupt (MSI)
5 * Copyright (C) 2003-2004 Intel
6 * Copyright (C) Tom Long Nguyen (tom.l.nguyen@intel.com)
10 #include <linux/irq.h>
11 #include <linux/interrupt.h>
12 #include <linux/init.h>
13 #include <linux/config.h>
14 #include <linux/ioport.h>
15 #include <linux/smp_lock.h>
16 #include <linux/pci.h>
17 #include <linux/proc_fs.h>
19 #include <asm/errno.h>
26 static DEFINE_SPINLOCK(msi_lock
);
27 static struct msi_desc
* msi_desc
[NR_IRQS
] = { [0 ... NR_IRQS
-1] = NULL
};
28 static kmem_cache_t
* msi_cachep
;
30 static int pci_msi_enable
= 1;
31 static int last_alloc_vector
;
32 static int nr_released_vectors
;
33 static int nr_reserved_vectors
= NR_HP_RESERVED_VECTORS
;
34 static int nr_msix_devices
;
36 #ifndef CONFIG_X86_IO_APIC
37 int vector_irq
[NR_VECTORS
] = { [0 ... NR_VECTORS
- 1] = -1};
38 u8 irq_vector
[NR_IRQ_VECTORS
];
41 static struct msi_ops
*msi_ops
;
44 msi_register(struct msi_ops
*ops
)
50 static void msi_cache_ctor(void *p
, kmem_cache_t
*cache
, unsigned long flags
)
52 memset(p
, 0, NR_IRQS
* sizeof(struct msi_desc
));
55 static int msi_cache_init(void)
57 msi_cachep
= kmem_cache_create("msi_cache",
58 NR_IRQS
* sizeof(struct msi_desc
),
59 0, SLAB_HWCACHE_ALIGN
, msi_cache_ctor
, NULL
);
66 static void msi_set_mask_bit(unsigned int vector
, int flag
)
68 struct msi_desc
*entry
;
70 entry
= (struct msi_desc
*)msi_desc
[vector
];
71 if (!entry
|| !entry
->dev
|| !entry
->mask_base
)
73 switch (entry
->msi_attrib
.type
) {
79 pos
= (long)entry
->mask_base
;
80 pci_read_config_dword(entry
->dev
, pos
, &mask_bits
);
83 pci_write_config_dword(entry
->dev
, pos
, mask_bits
);
88 int offset
= entry
->msi_attrib
.entry_nr
* PCI_MSIX_ENTRY_SIZE
+
89 PCI_MSIX_ENTRY_VECTOR_CTRL_OFFSET
;
90 writel(flag
, entry
->mask_base
+ offset
);
99 static void set_msi_affinity(unsigned int vector
, cpumask_t cpu_mask
)
101 struct msi_desc
*entry
;
102 u32 address_hi
, address_lo
;
103 unsigned int irq
= vector
;
104 unsigned int dest_cpu
= first_cpu(cpu_mask
);
106 entry
= (struct msi_desc
*)msi_desc
[vector
];
107 if (!entry
|| !entry
->dev
)
110 switch (entry
->msi_attrib
.type
) {
113 int pos
= pci_find_capability(entry
->dev
, PCI_CAP_ID_MSI
);
118 pci_read_config_dword(entry
->dev
, msi_upper_address_reg(pos
),
120 pci_read_config_dword(entry
->dev
, msi_lower_address_reg(pos
),
123 msi_ops
->target(vector
, dest_cpu
, &address_hi
, &address_lo
);
125 pci_write_config_dword(entry
->dev
, msi_upper_address_reg(pos
),
127 pci_write_config_dword(entry
->dev
, msi_lower_address_reg(pos
),
129 set_native_irq_info(irq
, cpu_mask
);
132 case PCI_CAP_ID_MSIX
:
135 entry
->msi_attrib
.entry_nr
* PCI_MSIX_ENTRY_SIZE
+
136 PCI_MSIX_ENTRY_UPPER_ADDR_OFFSET
;
138 entry
->msi_attrib
.entry_nr
* PCI_MSIX_ENTRY_SIZE
+
139 PCI_MSIX_ENTRY_LOWER_ADDR_OFFSET
;
141 address_hi
= readl(entry
->mask_base
+ offset_hi
);
142 address_lo
= readl(entry
->mask_base
+ offset_lo
);
144 msi_ops
->target(vector
, dest_cpu
, &address_hi
, &address_lo
);
146 writel(address_hi
, entry
->mask_base
+ offset_hi
);
147 writel(address_lo
, entry
->mask_base
+ offset_lo
);
148 set_native_irq_info(irq
, cpu_mask
);
156 #define set_msi_affinity NULL
157 #endif /* CONFIG_SMP */
159 static void mask_MSI_irq(unsigned int vector
)
161 msi_set_mask_bit(vector
, 1);
164 static void unmask_MSI_irq(unsigned int vector
)
166 msi_set_mask_bit(vector
, 0);
169 static unsigned int startup_msi_irq_wo_maskbit(unsigned int vector
)
171 struct msi_desc
*entry
;
174 spin_lock_irqsave(&msi_lock
, flags
);
175 entry
= msi_desc
[vector
];
176 if (!entry
|| !entry
->dev
) {
177 spin_unlock_irqrestore(&msi_lock
, flags
);
180 entry
->msi_attrib
.state
= 1; /* Mark it active */
181 spin_unlock_irqrestore(&msi_lock
, flags
);
183 return 0; /* never anything pending */
186 static unsigned int startup_msi_irq_w_maskbit(unsigned int vector
)
188 startup_msi_irq_wo_maskbit(vector
);
189 unmask_MSI_irq(vector
);
190 return 0; /* never anything pending */
193 static void shutdown_msi_irq(unsigned int vector
)
195 struct msi_desc
*entry
;
198 spin_lock_irqsave(&msi_lock
, flags
);
199 entry
= msi_desc
[vector
];
200 if (entry
&& entry
->dev
)
201 entry
->msi_attrib
.state
= 0; /* Mark it not active */
202 spin_unlock_irqrestore(&msi_lock
, flags
);
205 static void end_msi_irq_wo_maskbit(unsigned int vector
)
207 move_native_irq(vector
);
211 static void end_msi_irq_w_maskbit(unsigned int vector
)
213 move_native_irq(vector
);
214 unmask_MSI_irq(vector
);
218 static void do_nothing(unsigned int vector
)
223 * Interrupt Type for MSI-X PCI/PCI-X/PCI-Express Devices,
224 * which implement the MSI-X Capability Structure.
226 static struct hw_interrupt_type msix_irq_type
= {
227 .typename
= "PCI-MSI-X",
228 .startup
= startup_msi_irq_w_maskbit
,
229 .shutdown
= shutdown_msi_irq
,
230 .enable
= unmask_MSI_irq
,
231 .disable
= mask_MSI_irq
,
233 .end
= end_msi_irq_w_maskbit
,
234 .set_affinity
= set_msi_affinity
238 * Interrupt Type for MSI PCI/PCI-X/PCI-Express Devices,
239 * which implement the MSI Capability Structure with
240 * Mask-and-Pending Bits.
242 static struct hw_interrupt_type msi_irq_w_maskbit_type
= {
243 .typename
= "PCI-MSI",
244 .startup
= startup_msi_irq_w_maskbit
,
245 .shutdown
= shutdown_msi_irq
,
246 .enable
= unmask_MSI_irq
,
247 .disable
= mask_MSI_irq
,
249 .end
= end_msi_irq_w_maskbit
,
250 .set_affinity
= set_msi_affinity
254 * Interrupt Type for MSI PCI/PCI-X/PCI-Express Devices,
255 * which implement the MSI Capability Structure without
256 * Mask-and-Pending Bits.
258 static struct hw_interrupt_type msi_irq_wo_maskbit_type
= {
259 .typename
= "PCI-MSI",
260 .startup
= startup_msi_irq_wo_maskbit
,
261 .shutdown
= shutdown_msi_irq
,
262 .enable
= do_nothing
,
263 .disable
= do_nothing
,
265 .end
= end_msi_irq_wo_maskbit
,
266 .set_affinity
= set_msi_affinity
269 static int msi_free_vector(struct pci_dev
* dev
, int vector
, int reassign
);
270 static int assign_msi_vector(void)
272 static int new_vector_avail
= 1;
277 * msi_lock is provided to ensure that successful allocation of MSI
278 * vector is assigned unique among drivers.
280 spin_lock_irqsave(&msi_lock
, flags
);
282 if (!new_vector_avail
) {
286 * vector_irq[] = -1 indicates that this specific vector is:
287 * - assigned for MSI (since MSI have no associated IRQ) or
288 * - assigned for legacy if less than 16, or
289 * - having no corresponding 1:1 vector-to-IOxAPIC IRQ mapping
290 * vector_irq[] = 0 indicates that this vector, previously
291 * assigned for MSI, is freed by hotplug removed operations.
292 * This vector will be reused for any subsequent hotplug added
294 * vector_irq[] > 0 indicates that this vector is assigned for
295 * IOxAPIC IRQs. This vector and its value provides a 1-to-1
296 * vector-to-IOxAPIC IRQ mapping.
298 for (vector
= FIRST_DEVICE_VECTOR
; vector
< NR_IRQS
; vector
++) {
299 if (vector_irq
[vector
] != 0)
301 free_vector
= vector
;
302 if (!msi_desc
[vector
])
308 spin_unlock_irqrestore(&msi_lock
, flags
);
311 vector_irq
[free_vector
] = -1;
312 nr_released_vectors
--;
313 spin_unlock_irqrestore(&msi_lock
, flags
);
314 if (msi_desc
[free_vector
] != NULL
) {
318 /* free all linked vectors before re-assign */
320 spin_lock_irqsave(&msi_lock
, flags
);
321 dev
= msi_desc
[free_vector
]->dev
;
322 tail
= msi_desc
[free_vector
]->link
.tail
;
323 spin_unlock_irqrestore(&msi_lock
, flags
);
324 msi_free_vector(dev
, tail
, 1);
325 } while (free_vector
!= tail
);
330 vector
= assign_irq_vector(AUTO_ASSIGN
);
331 last_alloc_vector
= vector
;
332 if (vector
== LAST_DEVICE_VECTOR
)
333 new_vector_avail
= 0;
335 spin_unlock_irqrestore(&msi_lock
, flags
);
339 static int get_new_vector(void)
341 int vector
= assign_msi_vector();
344 set_intr_gate(vector
, interrupt
[vector
]);
349 static int msi_init(void)
351 static int status
= -ENOMEM
;
358 printk(KERN_WARNING
"PCI: MSI quirk detected. MSI disabled.\n");
363 status
= msi_arch_init();
367 "PCI: MSI arch init failed. MSI disabled.\n");
373 "PCI: MSI ops not registered. MSI disabled.\n");
378 last_alloc_vector
= assign_irq_vector(AUTO_ASSIGN
);
379 status
= msi_cache_init();
382 printk(KERN_WARNING
"PCI: MSI cache init failed\n");
386 #ifndef CONFIG_X86_IO_APIC
387 irq_vector
[0] = FIRST_DEVICE_VECTOR
;
390 if (last_alloc_vector
< 0) {
392 printk(KERN_WARNING
"PCI: No interrupt vectors available for MSI\n");
396 vector_irq
[last_alloc_vector
] = 0;
397 nr_released_vectors
++;
402 static int get_msi_vector(struct pci_dev
*dev
)
404 return get_new_vector();
407 static struct msi_desc
* alloc_msi_entry(void)
409 struct msi_desc
*entry
;
411 entry
= kmem_cache_alloc(msi_cachep
, SLAB_KERNEL
);
415 memset(entry
, 0, sizeof(struct msi_desc
));
416 entry
->link
.tail
= entry
->link
.head
= 0; /* single message */
422 static void attach_msi_entry(struct msi_desc
*entry
, int vector
)
426 spin_lock_irqsave(&msi_lock
, flags
);
427 msi_desc
[vector
] = entry
;
428 spin_unlock_irqrestore(&msi_lock
, flags
);
431 static void irq_handler_init(int cap_id
, int pos
, int mask
)
435 spin_lock_irqsave(&irq_desc
[pos
].lock
, flags
);
436 if (cap_id
== PCI_CAP_ID_MSIX
)
437 irq_desc
[pos
].handler
= &msix_irq_type
;
440 irq_desc
[pos
].handler
= &msi_irq_wo_maskbit_type
;
442 irq_desc
[pos
].handler
= &msi_irq_w_maskbit_type
;
444 spin_unlock_irqrestore(&irq_desc
[pos
].lock
, flags
);
447 static void enable_msi_mode(struct pci_dev
*dev
, int pos
, int type
)
451 pci_read_config_word(dev
, msi_control_reg(pos
), &control
);
452 if (type
== PCI_CAP_ID_MSI
) {
453 /* Set enabled bits to single MSI & enable MSI_enable bit */
454 msi_enable(control
, 1);
455 pci_write_config_word(dev
, msi_control_reg(pos
), control
);
456 dev
->msi_enabled
= 1;
458 msix_enable(control
);
459 pci_write_config_word(dev
, msi_control_reg(pos
), control
);
460 dev
->msix_enabled
= 1;
462 if (pci_find_capability(dev
, PCI_CAP_ID_EXP
)) {
463 /* PCI Express Endpoint device detected */
464 pci_intx(dev
, 0); /* disable intx */
468 void disable_msi_mode(struct pci_dev
*dev
, int pos
, int type
)
472 pci_read_config_word(dev
, msi_control_reg(pos
), &control
);
473 if (type
== PCI_CAP_ID_MSI
) {
474 /* Set enabled bits to single MSI & enable MSI_enable bit */
475 msi_disable(control
);
476 pci_write_config_word(dev
, msi_control_reg(pos
), control
);
477 dev
->msi_enabled
= 0;
479 msix_disable(control
);
480 pci_write_config_word(dev
, msi_control_reg(pos
), control
);
481 dev
->msix_enabled
= 0;
483 if (pci_find_capability(dev
, PCI_CAP_ID_EXP
)) {
484 /* PCI Express Endpoint device detected */
485 pci_intx(dev
, 1); /* enable intx */
489 static int msi_lookup_vector(struct pci_dev
*dev
, int type
)
494 spin_lock_irqsave(&msi_lock
, flags
);
495 for (vector
= FIRST_DEVICE_VECTOR
; vector
< NR_IRQS
; vector
++) {
496 if (!msi_desc
[vector
] || msi_desc
[vector
]->dev
!= dev
||
497 msi_desc
[vector
]->msi_attrib
.type
!= type
||
498 msi_desc
[vector
]->msi_attrib
.default_vector
!= dev
->irq
)
500 spin_unlock_irqrestore(&msi_lock
, flags
);
501 /* This pre-assigned MSI vector for this device
502 already exits. Override dev->irq with this vector */
506 spin_unlock_irqrestore(&msi_lock
, flags
);
511 void pci_scan_msi_device(struct pci_dev
*dev
)
516 if (pci_find_capability(dev
, PCI_CAP_ID_MSIX
) > 0)
518 else if (pci_find_capability(dev
, PCI_CAP_ID_MSI
) > 0)
519 nr_reserved_vectors
++;
523 int pci_save_msi_state(struct pci_dev
*dev
)
527 struct pci_cap_saved_state
*save_state
;
530 pos
= pci_find_capability(dev
, PCI_CAP_ID_MSI
);
531 if (pos
<= 0 || dev
->no_msi
)
534 pci_read_config_word(dev
, msi_control_reg(pos
), &control
);
535 if (!(control
& PCI_MSI_FLAGS_ENABLE
))
538 save_state
= kzalloc(sizeof(struct pci_cap_saved_state
) + sizeof(u32
) * 5,
541 printk(KERN_ERR
"Out of memory in pci_save_msi_state\n");
544 cap
= &save_state
->data
[0];
546 pci_read_config_dword(dev
, pos
, &cap
[i
++]);
547 control
= cap
[0] >> 16;
548 pci_read_config_dword(dev
, pos
+ PCI_MSI_ADDRESS_LO
, &cap
[i
++]);
549 if (control
& PCI_MSI_FLAGS_64BIT
) {
550 pci_read_config_dword(dev
, pos
+ PCI_MSI_ADDRESS_HI
, &cap
[i
++]);
551 pci_read_config_dword(dev
, pos
+ PCI_MSI_DATA_64
, &cap
[i
++]);
553 pci_read_config_dword(dev
, pos
+ PCI_MSI_DATA_32
, &cap
[i
++]);
554 if (control
& PCI_MSI_FLAGS_MASKBIT
)
555 pci_read_config_dword(dev
, pos
+ PCI_MSI_MASK_BIT
, &cap
[i
++]);
556 save_state
->cap_nr
= PCI_CAP_ID_MSI
;
557 pci_add_saved_cap(dev
, save_state
);
561 void pci_restore_msi_state(struct pci_dev
*dev
)
565 struct pci_cap_saved_state
*save_state
;
568 save_state
= pci_find_saved_cap(dev
, PCI_CAP_ID_MSI
);
569 pos
= pci_find_capability(dev
, PCI_CAP_ID_MSI
);
570 if (!save_state
|| pos
<= 0)
572 cap
= &save_state
->data
[0];
574 control
= cap
[i
++] >> 16;
575 pci_write_config_dword(dev
, pos
+ PCI_MSI_ADDRESS_LO
, cap
[i
++]);
576 if (control
& PCI_MSI_FLAGS_64BIT
) {
577 pci_write_config_dword(dev
, pos
+ PCI_MSI_ADDRESS_HI
, cap
[i
++]);
578 pci_write_config_dword(dev
, pos
+ PCI_MSI_DATA_64
, cap
[i
++]);
580 pci_write_config_dword(dev
, pos
+ PCI_MSI_DATA_32
, cap
[i
++]);
581 if (control
& PCI_MSI_FLAGS_MASKBIT
)
582 pci_write_config_dword(dev
, pos
+ PCI_MSI_MASK_BIT
, cap
[i
++]);
583 pci_write_config_word(dev
, pos
+ PCI_MSI_FLAGS
, control
);
584 enable_msi_mode(dev
, pos
, PCI_CAP_ID_MSI
);
585 pci_remove_saved_cap(save_state
);
589 int pci_save_msix_state(struct pci_dev
*dev
)
593 int vector
, head
, tail
= 0;
595 struct pci_cap_saved_state
*save_state
;
597 pos
= pci_find_capability(dev
, PCI_CAP_ID_MSIX
);
598 if (pos
<= 0 || dev
->no_msi
)
601 /* save the capability */
602 pci_read_config_word(dev
, msi_control_reg(pos
), &control
);
603 if (!(control
& PCI_MSIX_FLAGS_ENABLE
))
605 save_state
= kzalloc(sizeof(struct pci_cap_saved_state
) + sizeof(u16
),
608 printk(KERN_ERR
"Out of memory in pci_save_msix_state\n");
611 *((u16
*)&save_state
->data
[0]) = control
;
615 if (msi_lookup_vector(dev
, PCI_CAP_ID_MSIX
)) {
620 vector
= head
= dev
->irq
;
621 while (head
!= tail
) {
624 struct msi_desc
*entry
;
626 entry
= msi_desc
[vector
];
627 base
= entry
->mask_base
;
628 j
= entry
->msi_attrib
.entry_nr
;
630 entry
->address_lo_save
=
631 readl(base
+ j
* PCI_MSIX_ENTRY_SIZE
+
632 PCI_MSIX_ENTRY_LOWER_ADDR_OFFSET
);
633 entry
->address_hi_save
=
634 readl(base
+ j
* PCI_MSIX_ENTRY_SIZE
+
635 PCI_MSIX_ENTRY_UPPER_ADDR_OFFSET
);
637 readl(base
+ j
* PCI_MSIX_ENTRY_SIZE
+
638 PCI_MSIX_ENTRY_DATA_OFFSET
);
640 tail
= msi_desc
[vector
]->link
.tail
;
645 save_state
->cap_nr
= PCI_CAP_ID_MSIX
;
646 pci_add_saved_cap(dev
, save_state
);
650 void pci_restore_msix_state(struct pci_dev
*dev
)
654 int vector
, head
, tail
= 0;
657 struct msi_desc
*entry
;
659 struct pci_cap_saved_state
*save_state
;
661 save_state
= pci_find_saved_cap(dev
, PCI_CAP_ID_MSIX
);
664 save
= *((u16
*)&save_state
->data
[0]);
665 pci_remove_saved_cap(save_state
);
668 pos
= pci_find_capability(dev
, PCI_CAP_ID_MSIX
);
672 /* route the table */
674 if (msi_lookup_vector(dev
, PCI_CAP_ID_MSIX
))
676 vector
= head
= dev
->irq
;
677 while (head
!= tail
) {
678 entry
= msi_desc
[vector
];
679 base
= entry
->mask_base
;
680 j
= entry
->msi_attrib
.entry_nr
;
682 writel(entry
->address_lo_save
,
683 base
+ j
* PCI_MSIX_ENTRY_SIZE
+
684 PCI_MSIX_ENTRY_LOWER_ADDR_OFFSET
);
685 writel(entry
->address_hi_save
,
686 base
+ j
* PCI_MSIX_ENTRY_SIZE
+
687 PCI_MSIX_ENTRY_UPPER_ADDR_OFFSET
);
688 writel(entry
->data_save
,
689 base
+ j
* PCI_MSIX_ENTRY_SIZE
+
690 PCI_MSIX_ENTRY_DATA_OFFSET
);
692 tail
= msi_desc
[vector
]->link
.tail
;
697 pci_write_config_word(dev
, msi_control_reg(pos
), save
);
698 enable_msi_mode(dev
, pos
, PCI_CAP_ID_MSIX
);
702 static int msi_register_init(struct pci_dev
*dev
, struct msi_desc
*entry
)
708 int pos
, vector
= dev
->irq
;
711 pos
= pci_find_capability(dev
, PCI_CAP_ID_MSI
);
712 pci_read_config_word(dev
, msi_control_reg(pos
), &control
);
714 /* Configure MSI capability structure */
715 status
= msi_ops
->setup(dev
, vector
, &address_hi
, &address_lo
, &data
);
719 pci_write_config_dword(dev
, msi_lower_address_reg(pos
), address_lo
);
720 if (is_64bit_address(control
)) {
721 pci_write_config_dword(dev
,
722 msi_upper_address_reg(pos
), address_hi
);
723 pci_write_config_word(dev
,
724 msi_data_reg(pos
, 1), data
);
726 pci_write_config_word(dev
,
727 msi_data_reg(pos
, 0), data
);
728 if (entry
->msi_attrib
.maskbit
) {
729 unsigned int maskbits
, temp
;
730 /* All MSIs are unmasked by default, Mask them all */
731 pci_read_config_dword(dev
,
732 msi_mask_bits_reg(pos
, is_64bit_address(control
)),
734 temp
= (1 << multi_msi_capable(control
));
735 temp
= ((temp
- 1) & ~temp
);
737 pci_write_config_dword(dev
,
738 msi_mask_bits_reg(pos
, is_64bit_address(control
)),
746 * msi_capability_init - configure device's MSI capability structure
747 * @dev: pointer to the pci_dev data structure of MSI device function
749 * Setup the MSI capability structure of device function with a single
750 * MSI vector, regardless of device function is capable of handling
751 * multiple messages. A return of zero indicates the successful setup
752 * of an entry zero with the new MSI vector or non-zero for otherwise.
754 static int msi_capability_init(struct pci_dev
*dev
)
757 struct msi_desc
*entry
;
761 pos
= pci_find_capability(dev
, PCI_CAP_ID_MSI
);
762 pci_read_config_word(dev
, msi_control_reg(pos
), &control
);
763 /* MSI Entry Initialization */
764 entry
= alloc_msi_entry();
768 vector
= get_msi_vector(dev
);
770 kmem_cache_free(msi_cachep
, entry
);
773 entry
->link
.head
= vector
;
774 entry
->link
.tail
= vector
;
775 entry
->msi_attrib
.type
= PCI_CAP_ID_MSI
;
776 entry
->msi_attrib
.state
= 0; /* Mark it not active */
777 entry
->msi_attrib
.entry_nr
= 0;
778 entry
->msi_attrib
.maskbit
= is_mask_bit_support(control
);
779 entry
->msi_attrib
.default_vector
= dev
->irq
; /* Save IOAPIC IRQ */
782 if (is_mask_bit_support(control
)) {
783 entry
->mask_base
= (void __iomem
*)(long)msi_mask_bits_reg(pos
,
784 is_64bit_address(control
));
786 /* Replace with MSI handler */
787 irq_handler_init(PCI_CAP_ID_MSI
, vector
, entry
->msi_attrib
.maskbit
);
788 /* Configure MSI capability structure */
789 status
= msi_register_init(dev
, entry
);
791 dev
->irq
= entry
->msi_attrib
.default_vector
;
792 kmem_cache_free(msi_cachep
, entry
);
796 attach_msi_entry(entry
, vector
);
797 /* Set MSI enabled bits */
798 enable_msi_mode(dev
, pos
, PCI_CAP_ID_MSI
);
804 * msix_capability_init - configure device's MSI-X capability
805 * @dev: pointer to the pci_dev data structure of MSI-X device function
806 * @entries: pointer to an array of struct msix_entry entries
807 * @nvec: number of @entries
809 * Setup the MSI-X capability structure of device function with a
810 * single MSI-X vector. A return of zero indicates the successful setup of
811 * requested MSI-X entries with allocated vectors or non-zero for otherwise.
813 static int msix_capability_init(struct pci_dev
*dev
,
814 struct msix_entry
*entries
, int nvec
)
816 struct msi_desc
*head
= NULL
, *tail
= NULL
, *entry
= NULL
;
821 int vector
, pos
, i
, j
, nr_entries
, temp
= 0;
822 unsigned long phys_addr
;
828 pos
= pci_find_capability(dev
, PCI_CAP_ID_MSIX
);
829 /* Request & Map MSI-X table region */
830 pci_read_config_word(dev
, msi_control_reg(pos
), &control
);
831 nr_entries
= multi_msix_capable(control
);
833 pci_read_config_dword(dev
, msix_table_offset_reg(pos
), &table_offset
);
834 bir
= (u8
)(table_offset
& PCI_MSIX_FLAGS_BIRMASK
);
835 table_offset
&= ~PCI_MSIX_FLAGS_BIRMASK
;
836 phys_addr
= pci_resource_start (dev
, bir
) + table_offset
;
837 base
= ioremap_nocache(phys_addr
, nr_entries
* PCI_MSIX_ENTRY_SIZE
);
841 /* MSI-X Table Initialization */
842 for (i
= 0; i
< nvec
; i
++) {
843 entry
= alloc_msi_entry();
846 vector
= get_msi_vector(dev
);
848 kmem_cache_free(msi_cachep
, entry
);
852 j
= entries
[i
].entry
;
853 entries
[i
].vector
= vector
;
854 entry
->msi_attrib
.type
= PCI_CAP_ID_MSIX
;
855 entry
->msi_attrib
.state
= 0; /* Mark it not active */
856 entry
->msi_attrib
.entry_nr
= j
;
857 entry
->msi_attrib
.maskbit
= 1;
858 entry
->msi_attrib
.default_vector
= dev
->irq
;
860 entry
->mask_base
= base
;
862 entry
->link
.head
= vector
;
863 entry
->link
.tail
= vector
;
866 entry
->link
.head
= temp
;
867 entry
->link
.tail
= tail
->link
.tail
;
868 tail
->link
.tail
= vector
;
869 head
->link
.head
= vector
;
873 /* Replace with MSI-X handler */
874 irq_handler_init(PCI_CAP_ID_MSIX
, vector
, 1);
875 /* Configure MSI-X capability structure */
876 status
= msi_ops
->setup(dev
, vector
,
884 base
+ j
* PCI_MSIX_ENTRY_SIZE
+
885 PCI_MSIX_ENTRY_LOWER_ADDR_OFFSET
);
887 base
+ j
* PCI_MSIX_ENTRY_SIZE
+
888 PCI_MSIX_ENTRY_UPPER_ADDR_OFFSET
);
890 base
+ j
* PCI_MSIX_ENTRY_SIZE
+
891 PCI_MSIX_ENTRY_DATA_OFFSET
);
892 attach_msi_entry(entry
, vector
);
896 for (; i
>= 0; i
--) {
897 vector
= (entries
+ i
)->vector
;
898 msi_free_vector(dev
, vector
, 0);
899 (entries
+ i
)->vector
= 0;
903 /* Set MSI-X enabled bits */
904 enable_msi_mode(dev
, pos
, PCI_CAP_ID_MSIX
);
910 * pci_enable_msi - configure device's MSI capability structure
911 * @dev: pointer to the pci_dev data structure of MSI device function
913 * Setup the MSI capability structure of device function with
914 * a single MSI vector upon its software driver call to request for
915 * MSI mode enabled on its hardware device function. A return of zero
916 * indicates the successful setup of an entry zero with the new MSI
917 * vector or non-zero for otherwise.
919 int pci_enable_msi(struct pci_dev
* dev
)
922 int pos
, temp
, status
= -EINVAL
;
925 if (!pci_msi_enable
|| !dev
)
931 for (bus
= dev
->bus
; bus
; bus
= bus
->parent
)
932 if (bus
->bus_flags
& PCI_BUS_FLAGS_NO_MSI
)
941 pos
= pci_find_capability(dev
, PCI_CAP_ID_MSI
);
945 if (!msi_lookup_vector(dev
, PCI_CAP_ID_MSI
)) {
949 pci_read_config_word(dev
, msi_control_reg(pos
), &control
);
950 if (control
& PCI_MSI_FLAGS_ENABLE
)
951 return 0; /* Already in MSI mode */
952 spin_lock_irqsave(&msi_lock
, flags
);
953 if (!vector_irq
[dev
->irq
]) {
954 msi_desc
[dev
->irq
]->msi_attrib
.state
= 0;
955 vector_irq
[dev
->irq
] = -1;
956 nr_released_vectors
--;
957 spin_unlock_irqrestore(&msi_lock
, flags
);
958 status
= msi_register_init(dev
, msi_desc
[dev
->irq
]);
960 enable_msi_mode(dev
, pos
, PCI_CAP_ID_MSI
);
963 spin_unlock_irqrestore(&msi_lock
, flags
);
966 /* Check whether driver already requested for MSI-X vectors */
967 pos
= pci_find_capability(dev
, PCI_CAP_ID_MSIX
);
968 if (pos
> 0 && !msi_lookup_vector(dev
, PCI_CAP_ID_MSIX
)) {
969 printk(KERN_INFO
"PCI: %s: Can't enable MSI. "
970 "Device already has MSI-X vectors assigned\n",
975 status
= msi_capability_init(dev
);
978 nr_reserved_vectors
--; /* Only MSI capable */
979 else if (nr_msix_devices
> 0)
980 nr_msix_devices
--; /* Both MSI and MSI-X capable,
981 but choose enabling MSI */
987 void pci_disable_msi(struct pci_dev
* dev
)
989 struct msi_desc
*entry
;
990 int pos
, default_vector
;
999 pos
= pci_find_capability(dev
, PCI_CAP_ID_MSI
);
1003 pci_read_config_word(dev
, msi_control_reg(pos
), &control
);
1004 if (!(control
& PCI_MSI_FLAGS_ENABLE
))
1007 spin_lock_irqsave(&msi_lock
, flags
);
1008 entry
= msi_desc
[dev
->irq
];
1009 if (!entry
|| !entry
->dev
|| entry
->msi_attrib
.type
!= PCI_CAP_ID_MSI
) {
1010 spin_unlock_irqrestore(&msi_lock
, flags
);
1013 if (entry
->msi_attrib
.state
) {
1014 spin_unlock_irqrestore(&msi_lock
, flags
);
1015 printk(KERN_WARNING
"PCI: %s: pci_disable_msi() called without "
1016 "free_irq() on MSI vector %d\n",
1017 pci_name(dev
), dev
->irq
);
1018 BUG_ON(entry
->msi_attrib
.state
> 0);
1020 vector_irq
[dev
->irq
] = 0; /* free it */
1021 nr_released_vectors
++;
1022 default_vector
= entry
->msi_attrib
.default_vector
;
1023 spin_unlock_irqrestore(&msi_lock
, flags
);
1024 /* Restore dev->irq to its default pin-assertion vector */
1025 dev
->irq
= default_vector
;
1026 disable_msi_mode(dev
, pci_find_capability(dev
, PCI_CAP_ID_MSI
),
1031 static int msi_free_vector(struct pci_dev
* dev
, int vector
, int reassign
)
1033 struct msi_desc
*entry
;
1034 int head
, entry_nr
, type
;
1036 unsigned long flags
;
1038 msi_ops
->teardown(vector
);
1040 spin_lock_irqsave(&msi_lock
, flags
);
1041 entry
= msi_desc
[vector
];
1042 if (!entry
|| entry
->dev
!= dev
) {
1043 spin_unlock_irqrestore(&msi_lock
, flags
);
1046 type
= entry
->msi_attrib
.type
;
1047 entry_nr
= entry
->msi_attrib
.entry_nr
;
1048 head
= entry
->link
.head
;
1049 base
= entry
->mask_base
;
1050 msi_desc
[entry
->link
.head
]->link
.tail
= entry
->link
.tail
;
1051 msi_desc
[entry
->link
.tail
]->link
.head
= entry
->link
.head
;
1054 vector_irq
[vector
] = 0;
1055 nr_released_vectors
++;
1057 msi_desc
[vector
] = NULL
;
1058 spin_unlock_irqrestore(&msi_lock
, flags
);
1060 kmem_cache_free(msi_cachep
, entry
);
1062 if (type
== PCI_CAP_ID_MSIX
) {
1065 entry_nr
* PCI_MSIX_ENTRY_SIZE
+
1066 PCI_MSIX_ENTRY_VECTOR_CTRL_OFFSET
);
1068 if (head
== vector
) {
1070 * Detect last MSI-X vector to be released.
1071 * Release the MSI-X memory-mapped table.
1074 int pos
, nr_entries
;
1075 unsigned long phys_addr
;
1080 pos
= pci_find_capability(dev
, PCI_CAP_ID_MSIX
);
1081 pci_read_config_word(dev
, msi_control_reg(pos
),
1083 nr_entries
= multi_msix_capable(control
);
1084 pci_read_config_dword(dev
, msix_table_offset_reg(pos
),
1086 bir
= (u8
)(table_offset
& PCI_MSIX_FLAGS_BIRMASK
);
1087 table_offset
&= ~PCI_MSIX_FLAGS_BIRMASK
;
1088 phys_addr
= pci_resource_start(dev
, bir
) + table_offset
;
1090 * FIXME! and what did you want to do with phys_addr?
1100 static int reroute_msix_table(int head
, struct msix_entry
*entries
, int *nvec
)
1102 int vector
= head
, tail
= 0;
1103 int i
, j
= 0, nr_entries
= 0;
1105 unsigned long flags
;
1107 spin_lock_irqsave(&msi_lock
, flags
);
1108 while (head
!= tail
) {
1110 tail
= msi_desc
[vector
]->link
.tail
;
1111 if (entries
[0].entry
== msi_desc
[vector
]->msi_attrib
.entry_nr
)
1115 if (*nvec
> nr_entries
) {
1116 spin_unlock_irqrestore(&msi_lock
, flags
);
1120 vector
= ((j
> 0) ? j
: head
);
1121 for (i
= 0; i
< *nvec
; i
++) {
1122 j
= msi_desc
[vector
]->msi_attrib
.entry_nr
;
1123 msi_desc
[vector
]->msi_attrib
.state
= 0; /* Mark it not active */
1124 vector_irq
[vector
] = -1; /* Mark it busy */
1125 nr_released_vectors
--;
1126 entries
[i
].vector
= vector
;
1127 if (j
!= (entries
+ i
)->entry
) {
1128 base
= msi_desc
[vector
]->mask_base
;
1129 msi_desc
[vector
]->msi_attrib
.entry_nr
=
1130 (entries
+ i
)->entry
;
1131 writel( readl(base
+ j
* PCI_MSIX_ENTRY_SIZE
+
1132 PCI_MSIX_ENTRY_LOWER_ADDR_OFFSET
), base
+
1133 (entries
+ i
)->entry
* PCI_MSIX_ENTRY_SIZE
+
1134 PCI_MSIX_ENTRY_LOWER_ADDR_OFFSET
);
1135 writel( readl(base
+ j
* PCI_MSIX_ENTRY_SIZE
+
1136 PCI_MSIX_ENTRY_UPPER_ADDR_OFFSET
), base
+
1137 (entries
+ i
)->entry
* PCI_MSIX_ENTRY_SIZE
+
1138 PCI_MSIX_ENTRY_UPPER_ADDR_OFFSET
);
1139 writel( (readl(base
+ j
* PCI_MSIX_ENTRY_SIZE
+
1140 PCI_MSIX_ENTRY_DATA_OFFSET
) & 0xff00) | vector
,
1141 base
+ (entries
+i
)->entry
*PCI_MSIX_ENTRY_SIZE
+
1142 PCI_MSIX_ENTRY_DATA_OFFSET
);
1144 vector
= msi_desc
[vector
]->link
.tail
;
1146 spin_unlock_irqrestore(&msi_lock
, flags
);
1152 * pci_enable_msix - configure device's MSI-X capability structure
1153 * @dev: pointer to the pci_dev data structure of MSI-X device function
1154 * @entries: pointer to an array of MSI-X entries
1155 * @nvec: number of MSI-X vectors requested for allocation by device driver
1157 * Setup the MSI-X capability structure of device function with the number
1158 * of requested vectors upon its software driver call to request for
1159 * MSI-X mode enabled on its hardware device function. A return of zero
1160 * indicates the successful configuration of MSI-X capability structure
1161 * with new allocated MSI-X vectors. A return of < 0 indicates a failure.
1162 * Or a return of > 0 indicates that driver request is exceeding the number
1163 * of vectors available. Driver should use the returned value to re-send
1166 int pci_enable_msix(struct pci_dev
* dev
, struct msix_entry
*entries
, int nvec
)
1168 struct pci_bus
*bus
;
1169 int status
, pos
, nr_entries
, free_vectors
;
1172 unsigned long flags
;
1174 if (!pci_msi_enable
|| !dev
|| !entries
)
1180 for (bus
= dev
->bus
; bus
; bus
= bus
->parent
)
1181 if (bus
->bus_flags
& PCI_BUS_FLAGS_NO_MSI
)
1184 status
= msi_init();
1188 pos
= pci_find_capability(dev
, PCI_CAP_ID_MSIX
);
1192 pci_read_config_word(dev
, msi_control_reg(pos
), &control
);
1193 if (control
& PCI_MSIX_FLAGS_ENABLE
)
1194 return -EINVAL
; /* Already in MSI-X mode */
1196 nr_entries
= multi_msix_capable(control
);
1197 if (nvec
> nr_entries
)
1200 /* Check for any invalid entries */
1201 for (i
= 0; i
< nvec
; i
++) {
1202 if (entries
[i
].entry
>= nr_entries
)
1203 return -EINVAL
; /* invalid entry */
1204 for (j
= i
+ 1; j
< nvec
; j
++) {
1205 if (entries
[i
].entry
== entries
[j
].entry
)
1206 return -EINVAL
; /* duplicate entry */
1210 if (!msi_lookup_vector(dev
, PCI_CAP_ID_MSIX
)) {
1213 /* Reroute MSI-X table */
1214 if (reroute_msix_table(dev
->irq
, entries
, &nr_entries
)) {
1215 /* #requested > #previous-assigned */
1220 enable_msi_mode(dev
, pos
, PCI_CAP_ID_MSIX
);
1223 /* Check whether driver already requested for MSI vector */
1224 if (pci_find_capability(dev
, PCI_CAP_ID_MSI
) > 0 &&
1225 !msi_lookup_vector(dev
, PCI_CAP_ID_MSI
)) {
1226 printk(KERN_INFO
"PCI: %s: Can't enable MSI-X. "
1227 "Device already has an MSI vector assigned\n",
1233 spin_lock_irqsave(&msi_lock
, flags
);
1235 * msi_lock is provided to ensure that enough vectors resources are
1236 * available before granting.
1238 free_vectors
= pci_vector_resources(last_alloc_vector
,
1239 nr_released_vectors
);
1240 /* Ensure that each MSI/MSI-X device has one vector reserved by
1241 default to avoid any MSI-X driver to take all available
1243 free_vectors
-= nr_reserved_vectors
;
1244 /* Find the average of free vectors among MSI-X devices */
1245 if (nr_msix_devices
> 0)
1246 free_vectors
/= nr_msix_devices
;
1247 spin_unlock_irqrestore(&msi_lock
, flags
);
1249 if (nvec
> free_vectors
) {
1250 if (free_vectors
> 0)
1251 return free_vectors
;
1256 status
= msix_capability_init(dev
, entries
, nvec
);
1257 if (!status
&& nr_msix_devices
> 0)
1263 void pci_disable_msix(struct pci_dev
* dev
)
1268 if (!pci_msi_enable
)
1273 pos
= pci_find_capability(dev
, PCI_CAP_ID_MSIX
);
1277 pci_read_config_word(dev
, msi_control_reg(pos
), &control
);
1278 if (!(control
& PCI_MSIX_FLAGS_ENABLE
))
1282 if (!msi_lookup_vector(dev
, PCI_CAP_ID_MSIX
)) {
1283 int state
, vector
, head
, tail
= 0, warning
= 0;
1284 unsigned long flags
;
1286 vector
= head
= dev
->irq
;
1287 spin_lock_irqsave(&msi_lock
, flags
);
1288 while (head
!= tail
) {
1289 state
= msi_desc
[vector
]->msi_attrib
.state
;
1293 vector_irq
[vector
] = 0; /* free it */
1294 nr_released_vectors
++;
1296 tail
= msi_desc
[vector
]->link
.tail
;
1299 spin_unlock_irqrestore(&msi_lock
, flags
);
1302 printk(KERN_WARNING
"PCI: %s: pci_disable_msix() called without "
1303 "free_irq() on all MSI-X vectors\n",
1305 BUG_ON(warning
> 0);
1308 disable_msi_mode(dev
,
1309 pci_find_capability(dev
, PCI_CAP_ID_MSIX
),
1317 * msi_remove_pci_irq_vectors - reclaim MSI(X) vectors to unused state
1318 * @dev: pointer to the pci_dev data structure of MSI(X) device function
1320 * Being called during hotplug remove, from which the device function
1321 * is hot-removed. All previous assigned MSI/MSI-X vectors, if
1322 * allocated for this device function, are reclaimed to unused state,
1323 * which may be used later on.
1325 void msi_remove_pci_irq_vectors(struct pci_dev
* dev
)
1327 int state
, pos
, temp
;
1328 unsigned long flags
;
1330 if (!pci_msi_enable
|| !dev
)
1333 temp
= dev
->irq
; /* Save IOAPIC IRQ */
1334 pos
= pci_find_capability(dev
, PCI_CAP_ID_MSI
);
1335 if (pos
> 0 && !msi_lookup_vector(dev
, PCI_CAP_ID_MSI
)) {
1336 spin_lock_irqsave(&msi_lock
, flags
);
1337 state
= msi_desc
[dev
->irq
]->msi_attrib
.state
;
1338 spin_unlock_irqrestore(&msi_lock
, flags
);
1340 printk(KERN_WARNING
"PCI: %s: msi_remove_pci_irq_vectors() "
1341 "called without free_irq() on MSI vector %d\n",
1342 pci_name(dev
), dev
->irq
);
1344 } else /* Release MSI vector assigned to this device */
1345 msi_free_vector(dev
, dev
->irq
, 0);
1346 dev
->irq
= temp
; /* Restore IOAPIC IRQ */
1348 pos
= pci_find_capability(dev
, PCI_CAP_ID_MSIX
);
1349 if (pos
> 0 && !msi_lookup_vector(dev
, PCI_CAP_ID_MSIX
)) {
1350 int vector
, head
, tail
= 0, warning
= 0;
1351 void __iomem
*base
= NULL
;
1353 vector
= head
= dev
->irq
;
1354 while (head
!= tail
) {
1355 spin_lock_irqsave(&msi_lock
, flags
);
1356 state
= msi_desc
[vector
]->msi_attrib
.state
;
1357 tail
= msi_desc
[vector
]->link
.tail
;
1358 base
= msi_desc
[vector
]->mask_base
;
1359 spin_unlock_irqrestore(&msi_lock
, flags
);
1362 else if (vector
!= head
) /* Release MSI-X vector */
1363 msi_free_vector(dev
, vector
, 0);
1366 msi_free_vector(dev
, vector
, 0);
1368 /* Force to release the MSI-X memory-mapped table */
1370 unsigned long phys_addr
;
1375 pci_read_config_word(dev
, msi_control_reg(pos
),
1377 pci_read_config_dword(dev
, msix_table_offset_reg(pos
),
1379 bir
= (u8
)(table_offset
& PCI_MSIX_FLAGS_BIRMASK
);
1380 table_offset
&= ~PCI_MSIX_FLAGS_BIRMASK
;
1381 phys_addr
= pci_resource_start(dev
, bir
) + table_offset
;
1383 * FIXME! and what did you want to do with phys_addr?
1387 printk(KERN_WARNING
"PCI: %s: msi_remove_pci_irq_vectors() "
1388 "called without free_irq() on all MSI-X vectors\n",
1390 BUG_ON(warning
> 0);
1392 dev
->irq
= temp
; /* Restore IOAPIC IRQ */
1396 void pci_no_msi(void)
1401 EXPORT_SYMBOL(pci_enable_msi
);
1402 EXPORT_SYMBOL(pci_disable_msi
);
1403 EXPORT_SYMBOL(pci_enable_msix
);
1404 EXPORT_SYMBOL(pci_disable_msix
);