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 #define MSI_TARGET_CPU first_cpu(cpu_online_map)
28 static DEFINE_SPINLOCK(msi_lock);
29 static struct msi_desc* msi_desc[NR_IRQS] = { [0 ... NR_IRQS-1] = NULL };
30 static kmem_cache_t* msi_cachep;
32 static int pci_msi_enable = 1;
33 static int last_alloc_vector;
34 static int nr_released_vectors;
35 static int nr_reserved_vectors = NR_HP_RESERVED_VECTORS;
36 static int nr_msix_devices;
38 #ifndef CONFIG_X86_IO_APIC
39 int vector_irq[NR_VECTORS] = { [0 ... NR_VECTORS - 1] = -1};
40 u8 irq_vector[NR_IRQ_VECTORS] = { FIRST_DEVICE_VECTOR , 0 };
43 static void msi_cache_ctor(void *p, kmem_cache_t *cache, unsigned long flags)
45 memset(p, 0, NR_IRQS * sizeof(struct msi_desc));
48 static int msi_cache_init(void)
50 msi_cachep = kmem_cache_create("msi_cache",
51 NR_IRQS * sizeof(struct msi_desc),
52 0, SLAB_HWCACHE_ALIGN, msi_cache_ctor, NULL);
59 static void msi_set_mask_bit(unsigned int vector, int flag)
61 struct msi_desc *entry;
63 entry = (struct msi_desc *)msi_desc[vector];
64 if (!entry || !entry->dev || !entry->mask_base)
66 switch (entry->msi_attrib.type) {
72 pos = (long)entry->mask_base;
73 pci_read_config_dword(entry->dev, pos, &mask_bits);
76 pci_write_config_dword(entry->dev, pos, mask_bits);
81 int offset = entry->msi_attrib.entry_nr * PCI_MSIX_ENTRY_SIZE +
82 PCI_MSIX_ENTRY_VECTOR_CTRL_OFFSET;
83 writel(flag, entry->mask_base + offset);
92 static void set_msi_affinity(unsigned int vector, cpumask_t cpu_mask)
94 struct msi_desc *entry;
95 struct msg_address address;
96 unsigned int irq = vector;
97 unsigned int dest_cpu = first_cpu(cpu_mask);
99 entry = (struct msi_desc *)msi_desc[vector];
100 if (!entry || !entry->dev)
103 switch (entry->msi_attrib.type) {
106 int pos = pci_find_capability(entry->dev, PCI_CAP_ID_MSI);
111 pci_read_config_dword(entry->dev, msi_lower_address_reg(pos),
112 &address.lo_address.value);
113 address.lo_address.value &= MSI_ADDRESS_DEST_ID_MASK;
114 address.lo_address.value |= (cpu_physical_id(dest_cpu) <<
115 MSI_TARGET_CPU_SHIFT);
116 entry->msi_attrib.current_cpu = cpu_physical_id(dest_cpu);
117 pci_write_config_dword(entry->dev, msi_lower_address_reg(pos),
118 address.lo_address.value);
119 set_native_irq_info(irq, cpu_mask);
122 case PCI_CAP_ID_MSIX:
124 int offset = entry->msi_attrib.entry_nr * PCI_MSIX_ENTRY_SIZE +
125 PCI_MSIX_ENTRY_LOWER_ADDR_OFFSET;
127 address.lo_address.value = readl(entry->mask_base + offset);
128 address.lo_address.value &= MSI_ADDRESS_DEST_ID_MASK;
129 address.lo_address.value |= (cpu_physical_id(dest_cpu) <<
130 MSI_TARGET_CPU_SHIFT);
131 entry->msi_attrib.current_cpu = cpu_physical_id(dest_cpu);
132 writel(address.lo_address.value, entry->mask_base + offset);
133 set_native_irq_info(irq, cpu_mask);
141 #define set_msi_affinity NULL
142 #endif /* CONFIG_SMP */
144 static void mask_MSI_irq(unsigned int vector)
146 msi_set_mask_bit(vector, 1);
149 static void unmask_MSI_irq(unsigned int vector)
151 msi_set_mask_bit(vector, 0);
154 static unsigned int startup_msi_irq_wo_maskbit(unsigned int vector)
156 struct msi_desc *entry;
159 spin_lock_irqsave(&msi_lock, flags);
160 entry = msi_desc[vector];
161 if (!entry || !entry->dev) {
162 spin_unlock_irqrestore(&msi_lock, flags);
165 entry->msi_attrib.state = 1; /* Mark it active */
166 spin_unlock_irqrestore(&msi_lock, flags);
168 return 0; /* never anything pending */
171 static unsigned int startup_msi_irq_w_maskbit(unsigned int vector)
173 startup_msi_irq_wo_maskbit(vector);
174 unmask_MSI_irq(vector);
175 return 0; /* never anything pending */
178 static void shutdown_msi_irq(unsigned int vector)
180 struct msi_desc *entry;
183 spin_lock_irqsave(&msi_lock, flags);
184 entry = msi_desc[vector];
185 if (entry && entry->dev)
186 entry->msi_attrib.state = 0; /* Mark it not active */
187 spin_unlock_irqrestore(&msi_lock, flags);
190 static void end_msi_irq_wo_maskbit(unsigned int vector)
192 move_native_irq(vector);
196 static void end_msi_irq_w_maskbit(unsigned int vector)
198 move_native_irq(vector);
199 unmask_MSI_irq(vector);
203 static void do_nothing(unsigned int vector)
208 * Interrupt Type for MSI-X PCI/PCI-X/PCI-Express Devices,
209 * which implement the MSI-X Capability Structure.
211 static struct hw_interrupt_type msix_irq_type = {
212 .typename = "PCI-MSI-X",
213 .startup = startup_msi_irq_w_maskbit,
214 .shutdown = shutdown_msi_irq,
215 .enable = unmask_MSI_irq,
216 .disable = mask_MSI_irq,
218 .end = end_msi_irq_w_maskbit,
219 .set_affinity = set_msi_affinity
223 * Interrupt Type for MSI PCI/PCI-X/PCI-Express Devices,
224 * which implement the MSI Capability Structure with
225 * Mask-and-Pending Bits.
227 static struct hw_interrupt_type msi_irq_w_maskbit_type = {
228 .typename = "PCI-MSI",
229 .startup = startup_msi_irq_w_maskbit,
230 .shutdown = shutdown_msi_irq,
231 .enable = unmask_MSI_irq,
232 .disable = mask_MSI_irq,
234 .end = end_msi_irq_w_maskbit,
235 .set_affinity = set_msi_affinity
239 * Interrupt Type for MSI PCI/PCI-X/PCI-Express Devices,
240 * which implement the MSI Capability Structure without
241 * Mask-and-Pending Bits.
243 static struct hw_interrupt_type msi_irq_wo_maskbit_type = {
244 .typename = "PCI-MSI",
245 .startup = startup_msi_irq_wo_maskbit,
246 .shutdown = shutdown_msi_irq,
247 .enable = do_nothing,
248 .disable = do_nothing,
250 .end = end_msi_irq_wo_maskbit,
251 .set_affinity = set_msi_affinity
254 static void msi_data_init(struct msg_data *msi_data,
257 memset(msi_data, 0, sizeof(struct msg_data));
258 msi_data->vector = (u8)vector;
259 msi_data->delivery_mode = MSI_DELIVERY_MODE;
260 msi_data->level = MSI_LEVEL_MODE;
261 msi_data->trigger = MSI_TRIGGER_MODE;
264 static void msi_address_init(struct msg_address *msi_address)
266 unsigned int dest_id;
267 unsigned long dest_phys_id = cpu_physical_id(MSI_TARGET_CPU);
269 memset(msi_address, 0, sizeof(struct msg_address));
270 msi_address->hi_address = (u32)0;
271 dest_id = (MSI_ADDRESS_HEADER << MSI_ADDRESS_HEADER_SHIFT);
272 msi_address->lo_address.u.dest_mode = MSI_PHYSICAL_MODE;
273 msi_address->lo_address.u.redirection_hint = MSI_REDIRECTION_HINT_MODE;
274 msi_address->lo_address.u.dest_id = dest_id;
275 msi_address->lo_address.value |= (dest_phys_id << MSI_TARGET_CPU_SHIFT);
278 static int msi_free_vector(struct pci_dev* dev, int vector, int reassign);
279 static int assign_msi_vector(void)
281 static int new_vector_avail = 1;
286 * msi_lock is provided to ensure that successful allocation of MSI
287 * vector is assigned unique among drivers.
289 spin_lock_irqsave(&msi_lock, flags);
291 if (!new_vector_avail) {
295 * vector_irq[] = -1 indicates that this specific vector is:
296 * - assigned for MSI (since MSI have no associated IRQ) or
297 * - assigned for legacy if less than 16, or
298 * - having no corresponding 1:1 vector-to-IOxAPIC IRQ mapping
299 * vector_irq[] = 0 indicates that this vector, previously
300 * assigned for MSI, is freed by hotplug removed operations.
301 * This vector will be reused for any subsequent hotplug added
303 * vector_irq[] > 0 indicates that this vector is assigned for
304 * IOxAPIC IRQs. This vector and its value provides a 1-to-1
305 * vector-to-IOxAPIC IRQ mapping.
307 for (vector = FIRST_DEVICE_VECTOR; vector < NR_IRQS; vector++) {
308 if (vector_irq[vector] != 0)
310 free_vector = vector;
311 if (!msi_desc[vector])
317 spin_unlock_irqrestore(&msi_lock, flags);
320 vector_irq[free_vector] = -1;
321 nr_released_vectors--;
322 spin_unlock_irqrestore(&msi_lock, flags);
323 if (msi_desc[free_vector] != NULL) {
327 /* free all linked vectors before re-assign */
329 spin_lock_irqsave(&msi_lock, flags);
330 dev = msi_desc[free_vector]->dev;
331 tail = msi_desc[free_vector]->link.tail;
332 spin_unlock_irqrestore(&msi_lock, flags);
333 msi_free_vector(dev, tail, 1);
334 } while (free_vector != tail);
339 vector = assign_irq_vector(AUTO_ASSIGN);
340 last_alloc_vector = vector;
341 if (vector == LAST_DEVICE_VECTOR)
342 new_vector_avail = 0;
344 spin_unlock_irqrestore(&msi_lock, flags);
348 static int get_new_vector(void)
350 int vector = assign_msi_vector();
353 set_intr_gate(vector, interrupt[vector]);
358 static int msi_init(void)
360 static int status = -ENOMEM;
367 printk(KERN_WARNING "PCI: MSI quirk detected. MSI disabled.\n");
372 status = msi_cache_init();
375 printk(KERN_WARNING "PCI: MSI cache init failed\n");
378 last_alloc_vector = assign_irq_vector(AUTO_ASSIGN);
379 if (last_alloc_vector < 0) {
381 printk(KERN_WARNING "PCI: No interrupt vectors available for MSI\n");
385 vector_irq[last_alloc_vector] = 0;
386 nr_released_vectors++;
391 static int get_msi_vector(struct pci_dev *dev)
393 return get_new_vector();
396 static struct msi_desc* alloc_msi_entry(void)
398 struct msi_desc *entry;
400 entry = kmem_cache_alloc(msi_cachep, SLAB_KERNEL);
404 memset(entry, 0, sizeof(struct msi_desc));
405 entry->link.tail = entry->link.head = 0; /* single message */
411 static void attach_msi_entry(struct msi_desc *entry, int vector)
415 spin_lock_irqsave(&msi_lock, flags);
416 msi_desc[vector] = entry;
417 spin_unlock_irqrestore(&msi_lock, flags);
420 static void irq_handler_init(int cap_id, int pos, int mask)
424 spin_lock_irqsave(&irq_desc[pos].lock, flags);
425 if (cap_id == PCI_CAP_ID_MSIX)
426 irq_desc[pos].handler = &msix_irq_type;
429 irq_desc[pos].handler = &msi_irq_wo_maskbit_type;
431 irq_desc[pos].handler = &msi_irq_w_maskbit_type;
433 spin_unlock_irqrestore(&irq_desc[pos].lock, flags);
436 static void enable_msi_mode(struct pci_dev *dev, int pos, int type)
440 pci_read_config_word(dev, msi_control_reg(pos), &control);
441 if (type == PCI_CAP_ID_MSI) {
442 /* Set enabled bits to single MSI & enable MSI_enable bit */
443 msi_enable(control, 1);
444 pci_write_config_word(dev, msi_control_reg(pos), control);
446 msix_enable(control);
447 pci_write_config_word(dev, msi_control_reg(pos), control);
449 if (pci_find_capability(dev, PCI_CAP_ID_EXP)) {
450 /* PCI Express Endpoint device detected */
451 pci_intx(dev, 0); /* disable intx */
455 void disable_msi_mode(struct pci_dev *dev, int pos, int type)
459 pci_read_config_word(dev, msi_control_reg(pos), &control);
460 if (type == PCI_CAP_ID_MSI) {
461 /* Set enabled bits to single MSI & enable MSI_enable bit */
462 msi_disable(control);
463 pci_write_config_word(dev, msi_control_reg(pos), control);
465 msix_disable(control);
466 pci_write_config_word(dev, msi_control_reg(pos), control);
468 if (pci_find_capability(dev, PCI_CAP_ID_EXP)) {
469 /* PCI Express Endpoint device detected */
470 pci_intx(dev, 1); /* enable intx */
474 static int msi_lookup_vector(struct pci_dev *dev, int type)
479 spin_lock_irqsave(&msi_lock, flags);
480 for (vector = FIRST_DEVICE_VECTOR; vector < NR_IRQS; vector++) {
481 if (!msi_desc[vector] || msi_desc[vector]->dev != dev ||
482 msi_desc[vector]->msi_attrib.type != type ||
483 msi_desc[vector]->msi_attrib.default_vector != dev->irq)
485 spin_unlock_irqrestore(&msi_lock, flags);
486 /* This pre-assigned MSI vector for this device
487 already exits. Override dev->irq with this vector */
491 spin_unlock_irqrestore(&msi_lock, flags);
496 void pci_scan_msi_device(struct pci_dev *dev)
501 if (pci_find_capability(dev, PCI_CAP_ID_MSIX) > 0)
503 else if (pci_find_capability(dev, PCI_CAP_ID_MSI) > 0)
504 nr_reserved_vectors++;
508 int pci_save_msi_state(struct pci_dev *dev)
512 struct pci_cap_saved_state *save_state;
515 pos = pci_find_capability(dev, PCI_CAP_ID_MSI);
516 if (pos <= 0 || dev->no_msi)
519 pci_read_config_word(dev, msi_control_reg(pos), &control);
520 if (!(control & PCI_MSI_FLAGS_ENABLE))
523 save_state = kzalloc(sizeof(struct pci_cap_saved_state) + sizeof(u32) * 5,
526 printk(KERN_ERR "Out of memory in pci_save_msi_state\n");
529 cap = &save_state->data[0];
531 pci_read_config_dword(dev, pos, &cap[i++]);
532 control = cap[0] >> 16;
533 pci_read_config_dword(dev, pos + PCI_MSI_ADDRESS_LO, &cap[i++]);
534 if (control & PCI_MSI_FLAGS_64BIT) {
535 pci_read_config_dword(dev, pos + PCI_MSI_ADDRESS_HI, &cap[i++]);
536 pci_read_config_dword(dev, pos + PCI_MSI_DATA_64, &cap[i++]);
538 pci_read_config_dword(dev, pos + PCI_MSI_DATA_32, &cap[i++]);
539 if (control & PCI_MSI_FLAGS_MASKBIT)
540 pci_read_config_dword(dev, pos + PCI_MSI_MASK_BIT, &cap[i++]);
541 disable_msi_mode(dev, pos, PCI_CAP_ID_MSI);
542 save_state->cap_nr = PCI_CAP_ID_MSI;
543 pci_add_saved_cap(dev, save_state);
547 void pci_restore_msi_state(struct pci_dev *dev)
551 struct pci_cap_saved_state *save_state;
554 save_state = pci_find_saved_cap(dev, PCI_CAP_ID_MSI);
555 pos = pci_find_capability(dev, PCI_CAP_ID_MSI);
556 if (!save_state || pos <= 0)
558 cap = &save_state->data[0];
560 control = cap[i++] >> 16;
561 pci_write_config_dword(dev, pos + PCI_MSI_ADDRESS_LO, cap[i++]);
562 if (control & PCI_MSI_FLAGS_64BIT) {
563 pci_write_config_dword(dev, pos + PCI_MSI_ADDRESS_HI, cap[i++]);
564 pci_write_config_dword(dev, pos + PCI_MSI_DATA_64, cap[i++]);
566 pci_write_config_dword(dev, pos + PCI_MSI_DATA_32, cap[i++]);
567 if (control & PCI_MSI_FLAGS_MASKBIT)
568 pci_write_config_dword(dev, pos + PCI_MSI_MASK_BIT, cap[i++]);
569 pci_write_config_word(dev, pos + PCI_MSI_FLAGS, control);
570 enable_msi_mode(dev, pos, PCI_CAP_ID_MSI);
571 pci_remove_saved_cap(save_state);
575 int pci_save_msix_state(struct pci_dev *dev)
579 struct pci_cap_saved_state *save_state;
581 pos = pci_find_capability(dev, PCI_CAP_ID_MSIX);
582 if (pos <= 0 || dev->no_msi)
585 pci_read_config_word(dev, msi_control_reg(pos), &control);
586 if (!(control & PCI_MSIX_FLAGS_ENABLE))
588 save_state = kzalloc(sizeof(struct pci_cap_saved_state) + sizeof(u16),
591 printk(KERN_ERR "Out of memory in pci_save_msix_state\n");
594 *((u16 *)&save_state->data[0]) = control;
596 disable_msi_mode(dev, pos, PCI_CAP_ID_MSIX);
597 save_state->cap_nr = PCI_CAP_ID_MSIX;
598 pci_add_saved_cap(dev, save_state);
602 void pci_restore_msix_state(struct pci_dev *dev)
606 int vector, head, tail = 0;
609 struct msg_address address;
610 struct msg_data data;
611 struct msi_desc *entry;
613 struct pci_cap_saved_state *save_state;
615 save_state = pci_find_saved_cap(dev, PCI_CAP_ID_MSIX);
618 save = *((u16 *)&save_state->data[0]);
619 pci_remove_saved_cap(save_state);
622 pos = pci_find_capability(dev, PCI_CAP_ID_MSIX);
626 /* route the table */
628 if (msi_lookup_vector(dev, PCI_CAP_ID_MSIX))
630 vector = head = dev->irq;
631 while (head != tail) {
632 entry = msi_desc[vector];
633 base = entry->mask_base;
634 j = entry->msi_attrib.entry_nr;
636 msi_address_init(&address);
637 msi_data_init(&data, vector);
639 address.lo_address.value &= MSI_ADDRESS_DEST_ID_MASK;
640 address.lo_address.value |= entry->msi_attrib.current_cpu <<
641 MSI_TARGET_CPU_SHIFT;
643 writel(address.lo_address.value,
644 base + j * PCI_MSIX_ENTRY_SIZE +
645 PCI_MSIX_ENTRY_LOWER_ADDR_OFFSET);
646 writel(address.hi_address,
647 base + j * PCI_MSIX_ENTRY_SIZE +
648 PCI_MSIX_ENTRY_UPPER_ADDR_OFFSET);
650 base + j * PCI_MSIX_ENTRY_SIZE +
651 PCI_MSIX_ENTRY_DATA_OFFSET);
653 tail = msi_desc[vector]->link.tail;
658 pci_write_config_word(dev, msi_control_reg(pos), save);
659 enable_msi_mode(dev, pos, PCI_CAP_ID_MSIX);
663 static void msi_register_init(struct pci_dev *dev, struct msi_desc *entry)
665 struct msg_address address;
666 struct msg_data data;
667 int pos, vector = dev->irq;
670 pos = pci_find_capability(dev, PCI_CAP_ID_MSI);
671 pci_read_config_word(dev, msi_control_reg(pos), &control);
672 /* Configure MSI capability structure */
673 msi_address_init(&address);
674 msi_data_init(&data, vector);
675 entry->msi_attrib.current_cpu = ((address.lo_address.u.dest_id >>
676 MSI_TARGET_CPU_SHIFT) & MSI_TARGET_CPU_MASK);
677 pci_write_config_dword(dev, msi_lower_address_reg(pos),
678 address.lo_address.value);
679 if (is_64bit_address(control)) {
680 pci_write_config_dword(dev,
681 msi_upper_address_reg(pos), address.hi_address);
682 pci_write_config_word(dev,
683 msi_data_reg(pos, 1), *((u32*)&data));
685 pci_write_config_word(dev,
686 msi_data_reg(pos, 0), *((u32*)&data));
687 if (entry->msi_attrib.maskbit) {
688 unsigned int maskbits, temp;
689 /* All MSIs are unmasked by default, Mask them all */
690 pci_read_config_dword(dev,
691 msi_mask_bits_reg(pos, is_64bit_address(control)),
693 temp = (1 << multi_msi_capable(control));
694 temp = ((temp - 1) & ~temp);
696 pci_write_config_dword(dev,
697 msi_mask_bits_reg(pos, is_64bit_address(control)),
703 * msi_capability_init - configure device's MSI capability structure
704 * @dev: pointer to the pci_dev data structure of MSI device function
706 * Setup the MSI capability structure of device function with a single
707 * MSI vector, regardless of device function is capable of handling
708 * multiple messages. A return of zero indicates the successful setup
709 * of an entry zero with the new MSI vector or non-zero for otherwise.
711 static int msi_capability_init(struct pci_dev *dev)
713 struct msi_desc *entry;
717 pos = pci_find_capability(dev, PCI_CAP_ID_MSI);
718 pci_read_config_word(dev, msi_control_reg(pos), &control);
719 /* MSI Entry Initialization */
720 entry = alloc_msi_entry();
724 vector = get_msi_vector(dev);
726 kmem_cache_free(msi_cachep, entry);
729 entry->link.head = vector;
730 entry->link.tail = vector;
731 entry->msi_attrib.type = PCI_CAP_ID_MSI;
732 entry->msi_attrib.state = 0; /* Mark it not active */
733 entry->msi_attrib.entry_nr = 0;
734 entry->msi_attrib.maskbit = is_mask_bit_support(control);
735 entry->msi_attrib.default_vector = dev->irq; /* Save IOAPIC IRQ */
738 if (is_mask_bit_support(control)) {
739 entry->mask_base = (void __iomem *)(long)msi_mask_bits_reg(pos,
740 is_64bit_address(control));
742 /* Replace with MSI handler */
743 irq_handler_init(PCI_CAP_ID_MSI, vector, entry->msi_attrib.maskbit);
744 /* Configure MSI capability structure */
745 msi_register_init(dev, entry);
747 attach_msi_entry(entry, vector);
748 /* Set MSI enabled bits */
749 enable_msi_mode(dev, pos, PCI_CAP_ID_MSI);
755 * msix_capability_init - configure device's MSI-X capability
756 * @dev: pointer to the pci_dev data structure of MSI-X device function
757 * @entries: pointer to an array of struct msix_entry entries
758 * @nvec: number of @entries
760 * Setup the MSI-X capability structure of device function with a
761 * single MSI-X vector. A return of zero indicates the successful setup of
762 * requested MSI-X entries with allocated vectors or non-zero for otherwise.
764 static int msix_capability_init(struct pci_dev *dev,
765 struct msix_entry *entries, int nvec)
767 struct msi_desc *head = NULL, *tail = NULL, *entry = NULL;
768 struct msg_address address;
769 struct msg_data data;
770 int vector, pos, i, j, nr_entries, temp = 0;
771 unsigned long phys_addr;
777 pos = pci_find_capability(dev, PCI_CAP_ID_MSIX);
778 /* Request & Map MSI-X table region */
779 pci_read_config_word(dev, msi_control_reg(pos), &control);
780 nr_entries = multi_msix_capable(control);
782 pci_read_config_dword(dev, msix_table_offset_reg(pos), &table_offset);
783 bir = (u8)(table_offset & PCI_MSIX_FLAGS_BIRMASK);
784 table_offset &= ~PCI_MSIX_FLAGS_BIRMASK;
785 phys_addr = pci_resource_start (dev, bir) + table_offset;
786 base = ioremap_nocache(phys_addr, nr_entries * PCI_MSIX_ENTRY_SIZE);
790 /* MSI-X Table Initialization */
791 for (i = 0; i < nvec; i++) {
792 entry = alloc_msi_entry();
795 vector = get_msi_vector(dev);
797 kmem_cache_free(msi_cachep, entry);
801 j = entries[i].entry;
802 entries[i].vector = vector;
803 entry->msi_attrib.type = PCI_CAP_ID_MSIX;
804 entry->msi_attrib.state = 0; /* Mark it not active */
805 entry->msi_attrib.entry_nr = j;
806 entry->msi_attrib.maskbit = 1;
807 entry->msi_attrib.default_vector = dev->irq;
809 entry->mask_base = base;
811 entry->link.head = vector;
812 entry->link.tail = vector;
815 entry->link.head = temp;
816 entry->link.tail = tail->link.tail;
817 tail->link.tail = vector;
818 head->link.head = vector;
822 /* Replace with MSI-X handler */
823 irq_handler_init(PCI_CAP_ID_MSIX, vector, 1);
824 /* Configure MSI-X capability structure */
825 msi_address_init(&address);
826 msi_data_init(&data, vector);
827 entry->msi_attrib.current_cpu =
828 ((address.lo_address.u.dest_id >>
829 MSI_TARGET_CPU_SHIFT) & MSI_TARGET_CPU_MASK);
830 writel(address.lo_address.value,
831 base + j * PCI_MSIX_ENTRY_SIZE +
832 PCI_MSIX_ENTRY_LOWER_ADDR_OFFSET);
833 writel(address.hi_address,
834 base + j * PCI_MSIX_ENTRY_SIZE +
835 PCI_MSIX_ENTRY_UPPER_ADDR_OFFSET);
837 base + j * PCI_MSIX_ENTRY_SIZE +
838 PCI_MSIX_ENTRY_DATA_OFFSET);
839 attach_msi_entry(entry, vector);
843 for (; i >= 0; i--) {
844 vector = (entries + i)->vector;
845 msi_free_vector(dev, vector, 0);
846 (entries + i)->vector = 0;
850 /* Set MSI-X enabled bits */
851 enable_msi_mode(dev, pos, PCI_CAP_ID_MSIX);
857 * pci_enable_msi - configure device's MSI capability structure
858 * @dev: pointer to the pci_dev data structure of MSI device function
860 * Setup the MSI capability structure of device function with
861 * a single MSI vector upon its software driver call to request for
862 * MSI mode enabled on its hardware device function. A return of zero
863 * indicates the successful setup of an entry zero with the new MSI
864 * vector or non-zero for otherwise.
866 int pci_enable_msi(struct pci_dev* dev)
868 int pos, temp, status = -EINVAL;
871 if (!pci_msi_enable || !dev)
877 if (dev->bus->bus_flags & PCI_BUS_FLAGS_NO_MSI)
886 pos = pci_find_capability(dev, PCI_CAP_ID_MSI);
890 pci_read_config_word(dev, msi_control_reg(pos), &control);
891 if (control & PCI_MSI_FLAGS_ENABLE)
892 return 0; /* Already in MSI mode */
894 if (!msi_lookup_vector(dev, PCI_CAP_ID_MSI)) {
898 spin_lock_irqsave(&msi_lock, flags);
899 if (!vector_irq[dev->irq]) {
900 msi_desc[dev->irq]->msi_attrib.state = 0;
901 vector_irq[dev->irq] = -1;
902 nr_released_vectors--;
903 spin_unlock_irqrestore(&msi_lock, flags);
904 msi_register_init(dev, msi_desc[dev->irq]);
905 enable_msi_mode(dev, pos, PCI_CAP_ID_MSI);
908 spin_unlock_irqrestore(&msi_lock, flags);
911 /* Check whether driver already requested for MSI-X vectors */
912 pos = pci_find_capability(dev, PCI_CAP_ID_MSIX);
913 if (pos > 0 && !msi_lookup_vector(dev, PCI_CAP_ID_MSIX)) {
914 printk(KERN_INFO "PCI: %s: Can't enable MSI. "
915 "Device already has MSI-X vectors assigned\n",
920 status = msi_capability_init(dev);
923 nr_reserved_vectors--; /* Only MSI capable */
924 else if (nr_msix_devices > 0)
925 nr_msix_devices--; /* Both MSI and MSI-X capable,
926 but choose enabling MSI */
932 void pci_disable_msi(struct pci_dev* dev)
934 struct msi_desc *entry;
935 int pos, default_vector;
944 pos = pci_find_capability(dev, PCI_CAP_ID_MSI);
948 pci_read_config_word(dev, msi_control_reg(pos), &control);
949 if (!(control & PCI_MSI_FLAGS_ENABLE))
952 spin_lock_irqsave(&msi_lock, flags);
953 entry = msi_desc[dev->irq];
954 if (!entry || !entry->dev || entry->msi_attrib.type != PCI_CAP_ID_MSI) {
955 spin_unlock_irqrestore(&msi_lock, flags);
958 if (entry->msi_attrib.state) {
959 spin_unlock_irqrestore(&msi_lock, flags);
960 printk(KERN_WARNING "PCI: %s: pci_disable_msi() called without "
961 "free_irq() on MSI vector %d\n",
962 pci_name(dev), dev->irq);
963 BUG_ON(entry->msi_attrib.state > 0);
965 vector_irq[dev->irq] = 0; /* free it */
966 nr_released_vectors++;
967 default_vector = entry->msi_attrib.default_vector;
968 spin_unlock_irqrestore(&msi_lock, flags);
969 /* Restore dev->irq to its default pin-assertion vector */
970 dev->irq = default_vector;
971 disable_msi_mode(dev, pci_find_capability(dev, PCI_CAP_ID_MSI),
976 static int msi_free_vector(struct pci_dev* dev, int vector, int reassign)
978 struct msi_desc *entry;
979 int head, entry_nr, type;
983 spin_lock_irqsave(&msi_lock, flags);
984 entry = msi_desc[vector];
985 if (!entry || entry->dev != dev) {
986 spin_unlock_irqrestore(&msi_lock, flags);
989 type = entry->msi_attrib.type;
990 entry_nr = entry->msi_attrib.entry_nr;
991 head = entry->link.head;
992 base = entry->mask_base;
993 msi_desc[entry->link.head]->link.tail = entry->link.tail;
994 msi_desc[entry->link.tail]->link.head = entry->link.head;
997 vector_irq[vector] = 0;
998 nr_released_vectors++;
1000 msi_desc[vector] = NULL;
1001 spin_unlock_irqrestore(&msi_lock, flags);
1003 kmem_cache_free(msi_cachep, entry);
1005 if (type == PCI_CAP_ID_MSIX) {
1008 entry_nr * PCI_MSIX_ENTRY_SIZE +
1009 PCI_MSIX_ENTRY_VECTOR_CTRL_OFFSET);
1011 if (head == vector) {
1013 * Detect last MSI-X vector to be released.
1014 * Release the MSI-X memory-mapped table.
1017 int pos, nr_entries;
1018 unsigned long phys_addr;
1023 pos = pci_find_capability(dev, PCI_CAP_ID_MSIX);
1024 pci_read_config_word(dev, msi_control_reg(pos),
1026 nr_entries = multi_msix_capable(control);
1027 pci_read_config_dword(dev, msix_table_offset_reg(pos),
1029 bir = (u8)(table_offset & PCI_MSIX_FLAGS_BIRMASK);
1030 table_offset &= ~PCI_MSIX_FLAGS_BIRMASK;
1031 phys_addr = pci_resource_start(dev, bir) + table_offset;
1033 * FIXME! and what did you want to do with phys_addr?
1043 static int reroute_msix_table(int head, struct msix_entry *entries, int *nvec)
1045 int vector = head, tail = 0;
1046 int i, j = 0, nr_entries = 0;
1048 unsigned long flags;
1050 spin_lock_irqsave(&msi_lock, flags);
1051 while (head != tail) {
1053 tail = msi_desc[vector]->link.tail;
1054 if (entries[0].entry == msi_desc[vector]->msi_attrib.entry_nr)
1058 if (*nvec > nr_entries) {
1059 spin_unlock_irqrestore(&msi_lock, flags);
1063 vector = ((j > 0) ? j : head);
1064 for (i = 0; i < *nvec; i++) {
1065 j = msi_desc[vector]->msi_attrib.entry_nr;
1066 msi_desc[vector]->msi_attrib.state = 0; /* Mark it not active */
1067 vector_irq[vector] = -1; /* Mark it busy */
1068 nr_released_vectors--;
1069 entries[i].vector = vector;
1070 if (j != (entries + i)->entry) {
1071 base = msi_desc[vector]->mask_base;
1072 msi_desc[vector]->msi_attrib.entry_nr =
1073 (entries + i)->entry;
1074 writel( readl(base + j * PCI_MSIX_ENTRY_SIZE +
1075 PCI_MSIX_ENTRY_LOWER_ADDR_OFFSET), base +
1076 (entries + i)->entry * PCI_MSIX_ENTRY_SIZE +
1077 PCI_MSIX_ENTRY_LOWER_ADDR_OFFSET);
1078 writel( readl(base + j * PCI_MSIX_ENTRY_SIZE +
1079 PCI_MSIX_ENTRY_UPPER_ADDR_OFFSET), base +
1080 (entries + i)->entry * PCI_MSIX_ENTRY_SIZE +
1081 PCI_MSIX_ENTRY_UPPER_ADDR_OFFSET);
1082 writel( (readl(base + j * PCI_MSIX_ENTRY_SIZE +
1083 PCI_MSIX_ENTRY_DATA_OFFSET) & 0xff00) | vector,
1084 base + (entries+i)->entry*PCI_MSIX_ENTRY_SIZE +
1085 PCI_MSIX_ENTRY_DATA_OFFSET);
1087 vector = msi_desc[vector]->link.tail;
1089 spin_unlock_irqrestore(&msi_lock, flags);
1095 * pci_enable_msix - configure device's MSI-X capability structure
1096 * @dev: pointer to the pci_dev data structure of MSI-X device function
1097 * @entries: pointer to an array of MSI-X entries
1098 * @nvec: number of MSI-X vectors requested for allocation by device driver
1100 * Setup the MSI-X capability structure of device function with the number
1101 * of requested vectors upon its software driver call to request for
1102 * MSI-X mode enabled on its hardware device function. A return of zero
1103 * indicates the successful configuration of MSI-X capability structure
1104 * with new allocated MSI-X vectors. A return of < 0 indicates a failure.
1105 * Or a return of > 0 indicates that driver request is exceeding the number
1106 * of vectors available. Driver should use the returned value to re-send
1109 int pci_enable_msix(struct pci_dev* dev, struct msix_entry *entries, int nvec)
1111 int status, pos, nr_entries, free_vectors;
1114 unsigned long flags;
1116 if (!pci_msi_enable || !dev || !entries)
1119 status = msi_init();
1123 pos = pci_find_capability(dev, PCI_CAP_ID_MSIX);
1127 pci_read_config_word(dev, msi_control_reg(pos), &control);
1128 if (control & PCI_MSIX_FLAGS_ENABLE)
1129 return -EINVAL; /* Already in MSI-X mode */
1131 nr_entries = multi_msix_capable(control);
1132 if (nvec > nr_entries)
1135 /* Check for any invalid entries */
1136 for (i = 0; i < nvec; i++) {
1137 if (entries[i].entry >= nr_entries)
1138 return -EINVAL; /* invalid entry */
1139 for (j = i + 1; j < nvec; j++) {
1140 if (entries[i].entry == entries[j].entry)
1141 return -EINVAL; /* duplicate entry */
1145 if (!msi_lookup_vector(dev, PCI_CAP_ID_MSIX)) {
1148 /* Reroute MSI-X table */
1149 if (reroute_msix_table(dev->irq, entries, &nr_entries)) {
1150 /* #requested > #previous-assigned */
1155 enable_msi_mode(dev, pos, PCI_CAP_ID_MSIX);
1158 /* Check whether driver already requested for MSI vector */
1159 if (pci_find_capability(dev, PCI_CAP_ID_MSI) > 0 &&
1160 !msi_lookup_vector(dev, PCI_CAP_ID_MSI)) {
1161 printk(KERN_INFO "PCI: %s: Can't enable MSI-X. "
1162 "Device already has an MSI vector assigned\n",
1168 spin_lock_irqsave(&msi_lock, flags);
1170 * msi_lock is provided to ensure that enough vectors resources are
1171 * available before granting.
1173 free_vectors = pci_vector_resources(last_alloc_vector,
1174 nr_released_vectors);
1175 /* Ensure that each MSI/MSI-X device has one vector reserved by
1176 default to avoid any MSI-X driver to take all available
1178 free_vectors -= nr_reserved_vectors;
1179 /* Find the average of free vectors among MSI-X devices */
1180 if (nr_msix_devices > 0)
1181 free_vectors /= nr_msix_devices;
1182 spin_unlock_irqrestore(&msi_lock, flags);
1184 if (nvec > free_vectors) {
1185 if (free_vectors > 0)
1186 return free_vectors;
1191 status = msix_capability_init(dev, entries, nvec);
1192 if (!status && nr_msix_devices > 0)
1198 void pci_disable_msix(struct pci_dev* dev)
1203 if (!pci_msi_enable)
1208 pos = pci_find_capability(dev, PCI_CAP_ID_MSIX);
1212 pci_read_config_word(dev, msi_control_reg(pos), &control);
1213 if (!(control & PCI_MSIX_FLAGS_ENABLE))
1217 if (!msi_lookup_vector(dev, PCI_CAP_ID_MSIX)) {
1218 int state, vector, head, tail = 0, warning = 0;
1219 unsigned long flags;
1221 vector = head = dev->irq;
1222 spin_lock_irqsave(&msi_lock, flags);
1223 while (head != tail) {
1224 state = msi_desc[vector]->msi_attrib.state;
1228 vector_irq[vector] = 0; /* free it */
1229 nr_released_vectors++;
1231 tail = msi_desc[vector]->link.tail;
1234 spin_unlock_irqrestore(&msi_lock, flags);
1237 printk(KERN_WARNING "PCI: %s: pci_disable_msix() called without "
1238 "free_irq() on all MSI-X vectors\n",
1240 BUG_ON(warning > 0);
1243 disable_msi_mode(dev,
1244 pci_find_capability(dev, PCI_CAP_ID_MSIX),
1252 * msi_remove_pci_irq_vectors - reclaim MSI(X) vectors to unused state
1253 * @dev: pointer to the pci_dev data structure of MSI(X) device function
1255 * Being called during hotplug remove, from which the device function
1256 * is hot-removed. All previous assigned MSI/MSI-X vectors, if
1257 * allocated for this device function, are reclaimed to unused state,
1258 * which may be used later on.
1260 void msi_remove_pci_irq_vectors(struct pci_dev* dev)
1262 int state, pos, temp;
1263 unsigned long flags;
1265 if (!pci_msi_enable || !dev)
1268 temp = dev->irq; /* Save IOAPIC IRQ */
1269 pos = pci_find_capability(dev, PCI_CAP_ID_MSI);
1270 if (pos > 0 && !msi_lookup_vector(dev, PCI_CAP_ID_MSI)) {
1271 spin_lock_irqsave(&msi_lock, flags);
1272 state = msi_desc[dev->irq]->msi_attrib.state;
1273 spin_unlock_irqrestore(&msi_lock, flags);
1275 printk(KERN_WARNING "PCI: %s: msi_remove_pci_irq_vectors() "
1276 "called without free_irq() on MSI vector %d\n",
1277 pci_name(dev), dev->irq);
1279 } else /* Release MSI vector assigned to this device */
1280 msi_free_vector(dev, dev->irq, 0);
1281 dev->irq = temp; /* Restore IOAPIC IRQ */
1283 pos = pci_find_capability(dev, PCI_CAP_ID_MSIX);
1284 if (pos > 0 && !msi_lookup_vector(dev, PCI_CAP_ID_MSIX)) {
1285 int vector, head, tail = 0, warning = 0;
1286 void __iomem *base = NULL;
1288 vector = head = dev->irq;
1289 while (head != tail) {
1290 spin_lock_irqsave(&msi_lock, flags);
1291 state = msi_desc[vector]->msi_attrib.state;
1292 tail = msi_desc[vector]->link.tail;
1293 base = msi_desc[vector]->mask_base;
1294 spin_unlock_irqrestore(&msi_lock, flags);
1297 else if (vector != head) /* Release MSI-X vector */
1298 msi_free_vector(dev, vector, 0);
1301 msi_free_vector(dev, vector, 0);
1303 /* Force to release the MSI-X memory-mapped table */
1305 unsigned long phys_addr;
1310 pci_read_config_word(dev, msi_control_reg(pos),
1312 pci_read_config_dword(dev, msix_table_offset_reg(pos),
1314 bir = (u8)(table_offset & PCI_MSIX_FLAGS_BIRMASK);
1315 table_offset &= ~PCI_MSIX_FLAGS_BIRMASK;
1316 phys_addr = pci_resource_start(dev, bir) + table_offset;
1318 * FIXME! and what did you want to do with phys_addr?
1322 printk(KERN_WARNING "PCI: %s: msi_remove_pci_irq_vectors() "
1323 "called without free_irq() on all MSI-X vectors\n",
1325 BUG_ON(warning > 0);
1327 dev->irq = temp; /* Restore IOAPIC IRQ */
1331 void pci_no_msi(void)
1336 EXPORT_SYMBOL(pci_enable_msi);
1337 EXPORT_SYMBOL(pci_disable_msi);
1338 EXPORT_SYMBOL(pci_enable_msix);
1339 EXPORT_SYMBOL(pci_disable_msix);