2 * Intel IO-APIC support for multi-Pentium hosts.
4 * Copyright (C) 1997, 1998, 1999, 2000 Ingo Molnar, Hajnalka Szabo
6 * Many thanks to Stig Venaas for trying out countless experimental
7 * patches and reporting/debugging problems patiently!
9 * (c) 1999, Multiple IO-APIC support, developed by
10 * Ken-ichi Yaku <yaku@css1.kbnes.nec.co.jp> and
11 * Hidemi Kishimoto <kisimoto@css1.kbnes.nec.co.jp>,
12 * further tested and cleaned up by Zach Brown <zab@redhat.com>
13 * and Ingo Molnar <mingo@redhat.com>
16 * Maciej W. Rozycki : Bits for genuine 82489DX APICs;
17 * thanks to Eric Gilmore
19 * for testing these extensively
20 * Paul Diefenbaugh : Added full ACPI support
24 #include <linux/interrupt.h>
25 #include <linux/init.h>
26 #include <linux/delay.h>
27 #include <linux/sched.h>
28 #include <linux/pci.h>
29 #include <linux/mc146818rtc.h>
30 #include <linux/compiler.h>
31 #include <linux/acpi.h>
32 #include <linux/module.h>
33 #include <linux/sysdev.h>
34 #include <linux/msi.h>
35 #include <linux/htirq.h>
36 #include <linux/freezer.h>
37 #include <linux/kthread.h>
38 #include <linux/jiffies.h> /* time_after() */
40 #include <acpi/acpi_bus.h>
42 #include <linux/bootmem.h>
43 #include <linux/dmar.h>
44 #include <linux/hpet.h>
50 #include <asm/proto.h>
53 #include <asm/timer.h>
54 #include <asm/i8259.h>
56 #include <asm/msidef.h>
57 #include <asm/hypertransport.h>
58 #include <asm/setup.h>
59 #include <asm/irq_remapping.h>
61 #include <asm/uv/uv_hub.h>
62 #include <asm/uv/uv_irq.h>
65 #include <mach_apic.h>
66 #include <mach_apicdef.h>
68 #define __apicdebuginit(type) static type __init
71 * Is the SiS APIC rmw bug present ?
72 * -1 = don't know, 0 = no, 1 = yes
74 int sis_apic_bug = -1;
76 static DEFINE_SPINLOCK(ioapic_lock);
77 static DEFINE_SPINLOCK(vector_lock);
80 * # of IRQ routing registers
82 int nr_ioapic_registers[MAX_IO_APICS];
84 /* I/O APIC entries */
85 struct mp_config_ioapic mp_ioapics[MAX_IO_APICS];
88 /* MP IRQ source entries */
89 struct mp_config_intsrc mp_irqs[MAX_IRQ_SOURCES];
91 /* # of MP IRQ source entries */
94 #if defined (CONFIG_MCA) || defined (CONFIG_EISA)
95 int mp_bus_id_to_type[MAX_MP_BUSSES];
98 DECLARE_BITMAP(mp_bus_not_pci, MAX_MP_BUSSES);
100 int skip_ioapic_setup;
102 static int __init parse_noapic(char *str)
104 /* disable IO-APIC */
105 disable_ioapic_setup();
108 early_param("noapic", parse_noapic);
113 * This is performance-critical, we want to do it O(1)
115 * the indexing order of this array favors 1:1 mappings
116 * between pins and IRQs.
119 struct irq_pin_list {
121 struct irq_pin_list *next;
124 static struct irq_pin_list *get_one_free_irq_2_pin(int cpu)
126 struct irq_pin_list *pin;
129 node = cpu_to_node(cpu);
131 pin = kzalloc_node(sizeof(*pin), GFP_ATOMIC, node);
137 struct irq_pin_list *irq_2_pin;
138 cpumask_var_t domain;
139 cpumask_var_t old_domain;
140 unsigned move_cleanup_count;
142 u8 move_in_progress : 1;
143 #ifdef CONFIG_NUMA_MIGRATE_IRQ_DESC
144 u8 move_desc_pending : 1;
148 /* irq_cfg is indexed by the sum of all RTEs in all I/O APICs. */
149 #ifdef CONFIG_SPARSE_IRQ
150 static struct irq_cfg irq_cfgx[] = {
152 static struct irq_cfg irq_cfgx[NR_IRQS] = {
154 [0] = { .vector = IRQ0_VECTOR, },
155 [1] = { .vector = IRQ1_VECTOR, },
156 [2] = { .vector = IRQ2_VECTOR, },
157 [3] = { .vector = IRQ3_VECTOR, },
158 [4] = { .vector = IRQ4_VECTOR, },
159 [5] = { .vector = IRQ5_VECTOR, },
160 [6] = { .vector = IRQ6_VECTOR, },
161 [7] = { .vector = IRQ7_VECTOR, },
162 [8] = { .vector = IRQ8_VECTOR, },
163 [9] = { .vector = IRQ9_VECTOR, },
164 [10] = { .vector = IRQ10_VECTOR, },
165 [11] = { .vector = IRQ11_VECTOR, },
166 [12] = { .vector = IRQ12_VECTOR, },
167 [13] = { .vector = IRQ13_VECTOR, },
168 [14] = { .vector = IRQ14_VECTOR, },
169 [15] = { .vector = IRQ15_VECTOR, },
172 int __init arch_early_irq_init(void)
175 struct irq_desc *desc;
180 count = ARRAY_SIZE(irq_cfgx);
182 for (i = 0; i < count; i++) {
183 desc = irq_to_desc(i);
184 desc->chip_data = &cfg[i];
185 alloc_bootmem_cpumask_var(&cfg[i].domain);
186 alloc_bootmem_cpumask_var(&cfg[i].old_domain);
187 if (i < NR_IRQS_LEGACY)
188 cpumask_setall(cfg[i].domain);
194 #ifdef CONFIG_SPARSE_IRQ
195 static struct irq_cfg *irq_cfg(unsigned int irq)
197 struct irq_cfg *cfg = NULL;
198 struct irq_desc *desc;
200 desc = irq_to_desc(irq);
202 cfg = desc->chip_data;
207 static struct irq_cfg *get_one_free_irq_cfg(int cpu)
212 node = cpu_to_node(cpu);
214 cfg = kzalloc_node(sizeof(*cfg), GFP_ATOMIC, node);
216 if (!alloc_cpumask_var_node(&cfg->domain, GFP_ATOMIC, node)) {
219 } else if (!alloc_cpumask_var_node(&cfg->old_domain,
221 free_cpumask_var(cfg->domain);
225 cpumask_clear(cfg->domain);
226 cpumask_clear(cfg->old_domain);
233 int arch_init_chip_data(struct irq_desc *desc, int cpu)
237 cfg = desc->chip_data;
239 desc->chip_data = get_one_free_irq_cfg(cpu);
240 if (!desc->chip_data) {
241 printk(KERN_ERR "can not alloc irq_cfg\n");
249 #ifdef CONFIG_NUMA_MIGRATE_IRQ_DESC
252 init_copy_irq_2_pin(struct irq_cfg *old_cfg, struct irq_cfg *cfg, int cpu)
254 struct irq_pin_list *old_entry, *head, *tail, *entry;
256 cfg->irq_2_pin = NULL;
257 old_entry = old_cfg->irq_2_pin;
261 entry = get_one_free_irq_2_pin(cpu);
265 entry->apic = old_entry->apic;
266 entry->pin = old_entry->pin;
269 old_entry = old_entry->next;
271 entry = get_one_free_irq_2_pin(cpu);
279 /* still use the old one */
282 entry->apic = old_entry->apic;
283 entry->pin = old_entry->pin;
286 old_entry = old_entry->next;
290 cfg->irq_2_pin = head;
293 static void free_irq_2_pin(struct irq_cfg *old_cfg, struct irq_cfg *cfg)
295 struct irq_pin_list *entry, *next;
297 if (old_cfg->irq_2_pin == cfg->irq_2_pin)
300 entry = old_cfg->irq_2_pin;
307 old_cfg->irq_2_pin = NULL;
310 void arch_init_copy_chip_data(struct irq_desc *old_desc,
311 struct irq_desc *desc, int cpu)
314 struct irq_cfg *old_cfg;
316 cfg = get_one_free_irq_cfg(cpu);
321 desc->chip_data = cfg;
323 old_cfg = old_desc->chip_data;
325 memcpy(cfg, old_cfg, sizeof(struct irq_cfg));
327 init_copy_irq_2_pin(old_cfg, cfg, cpu);
330 static void free_irq_cfg(struct irq_cfg *old_cfg)
335 void arch_free_chip_data(struct irq_desc *old_desc, struct irq_desc *desc)
337 struct irq_cfg *old_cfg, *cfg;
339 old_cfg = old_desc->chip_data;
340 cfg = desc->chip_data;
346 free_irq_2_pin(old_cfg, cfg);
347 free_irq_cfg(old_cfg);
348 old_desc->chip_data = NULL;
353 set_extra_move_desc(struct irq_desc *desc, const struct cpumask *mask)
355 struct irq_cfg *cfg = desc->chip_data;
357 if (!cfg->move_in_progress) {
358 /* it means that domain is not changed */
359 if (!cpumask_intersects(&desc->affinity, mask))
360 cfg->move_desc_pending = 1;
366 static struct irq_cfg *irq_cfg(unsigned int irq)
368 return irq < nr_irqs ? irq_cfgx + irq : NULL;
373 #ifndef CONFIG_NUMA_MIGRATE_IRQ_DESC
375 set_extra_move_desc(struct irq_desc *desc, const struct cpumask *mask)
382 unsigned int unused[3];
386 static __attribute_const__ struct io_apic __iomem *io_apic_base(int idx)
388 return (void __iomem *) __fix_to_virt(FIX_IO_APIC_BASE_0 + idx)
389 + (mp_ioapics[idx].mp_apicaddr & ~PAGE_MASK);
392 static inline unsigned int io_apic_read(unsigned int apic, unsigned int reg)
394 struct io_apic __iomem *io_apic = io_apic_base(apic);
395 writel(reg, &io_apic->index);
396 return readl(&io_apic->data);
399 static inline void io_apic_write(unsigned int apic, unsigned int reg, unsigned int value)
401 struct io_apic __iomem *io_apic = io_apic_base(apic);
402 writel(reg, &io_apic->index);
403 writel(value, &io_apic->data);
407 * Re-write a value: to be used for read-modify-write
408 * cycles where the read already set up the index register.
410 * Older SiS APIC requires we rewrite the index register
412 static inline void io_apic_modify(unsigned int apic, unsigned int reg, unsigned int value)
414 struct io_apic __iomem *io_apic = io_apic_base(apic);
417 writel(reg, &io_apic->index);
418 writel(value, &io_apic->data);
421 static bool io_apic_level_ack_pending(struct irq_cfg *cfg)
423 struct irq_pin_list *entry;
426 spin_lock_irqsave(&ioapic_lock, flags);
427 entry = cfg->irq_2_pin;
435 reg = io_apic_read(entry->apic, 0x10 + pin*2);
436 /* Is the remote IRR bit set? */
437 if (reg & IO_APIC_REDIR_REMOTE_IRR) {
438 spin_unlock_irqrestore(&ioapic_lock, flags);
445 spin_unlock_irqrestore(&ioapic_lock, flags);
451 struct { u32 w1, w2; };
452 struct IO_APIC_route_entry entry;
455 static struct IO_APIC_route_entry ioapic_read_entry(int apic, int pin)
457 union entry_union eu;
459 spin_lock_irqsave(&ioapic_lock, flags);
460 eu.w1 = io_apic_read(apic, 0x10 + 2 * pin);
461 eu.w2 = io_apic_read(apic, 0x11 + 2 * pin);
462 spin_unlock_irqrestore(&ioapic_lock, flags);
467 * When we write a new IO APIC routing entry, we need to write the high
468 * word first! If the mask bit in the low word is clear, we will enable
469 * the interrupt, and we need to make sure the entry is fully populated
470 * before that happens.
473 __ioapic_write_entry(int apic, int pin, struct IO_APIC_route_entry e)
475 union entry_union eu;
477 io_apic_write(apic, 0x11 + 2*pin, eu.w2);
478 io_apic_write(apic, 0x10 + 2*pin, eu.w1);
481 static void ioapic_write_entry(int apic, int pin, struct IO_APIC_route_entry e)
484 spin_lock_irqsave(&ioapic_lock, flags);
485 __ioapic_write_entry(apic, pin, e);
486 spin_unlock_irqrestore(&ioapic_lock, flags);
490 * When we mask an IO APIC routing entry, we need to write the low
491 * word first, in order to set the mask bit before we change the
494 static void ioapic_mask_entry(int apic, int pin)
497 union entry_union eu = { .entry.mask = 1 };
499 spin_lock_irqsave(&ioapic_lock, flags);
500 io_apic_write(apic, 0x10 + 2*pin, eu.w1);
501 io_apic_write(apic, 0x11 + 2*pin, eu.w2);
502 spin_unlock_irqrestore(&ioapic_lock, flags);
506 static void send_cleanup_vector(struct irq_cfg *cfg)
508 cpumask_var_t cleanup_mask;
510 if (unlikely(!alloc_cpumask_var(&cleanup_mask, GFP_ATOMIC))) {
512 cfg->move_cleanup_count = 0;
513 for_each_cpu_and(i, cfg->old_domain, cpu_online_mask)
514 cfg->move_cleanup_count++;
515 for_each_cpu_and(i, cfg->old_domain, cpu_online_mask)
516 send_IPI_mask(cpumask_of(i), IRQ_MOVE_CLEANUP_VECTOR);
518 cpumask_and(cleanup_mask, cfg->old_domain, cpu_online_mask);
519 cfg->move_cleanup_count = cpumask_weight(cleanup_mask);
520 send_IPI_mask(cleanup_mask, IRQ_MOVE_CLEANUP_VECTOR);
521 free_cpumask_var(cleanup_mask);
523 cfg->move_in_progress = 0;
526 static void __target_IO_APIC_irq(unsigned int irq, unsigned int dest, struct irq_cfg *cfg)
529 struct irq_pin_list *entry;
530 u8 vector = cfg->vector;
532 entry = cfg->irq_2_pin;
541 #ifdef CONFIG_INTR_REMAP
543 * With interrupt-remapping, destination information comes
544 * from interrupt-remapping table entry.
546 if (!irq_remapped(irq))
547 io_apic_write(apic, 0x11 + pin*2, dest);
549 io_apic_write(apic, 0x11 + pin*2, dest);
551 reg = io_apic_read(apic, 0x10 + pin*2);
552 reg &= ~IO_APIC_REDIR_VECTOR_MASK;
554 io_apic_modify(apic, 0x10 + pin*2, reg);
562 assign_irq_vector(int irq, struct irq_cfg *cfg, const struct cpumask *mask);
565 * Either sets desc->affinity to a valid value, and returns cpu_mask_to_apicid
566 * of that, or returns BAD_APICID and leaves desc->affinity untouched.
569 set_desc_affinity(struct irq_desc *desc, const struct cpumask *mask)
574 if (!cpumask_intersects(mask, cpu_online_mask))
578 cfg = desc->chip_data;
579 if (assign_irq_vector(irq, cfg, mask))
582 cpumask_and(&desc->affinity, cfg->domain, mask);
583 set_extra_move_desc(desc, mask);
584 return cpu_mask_to_apicid_and(&desc->affinity, cpu_online_mask);
588 set_ioapic_affinity_irq_desc(struct irq_desc *desc, const struct cpumask *mask)
596 cfg = desc->chip_data;
598 spin_lock_irqsave(&ioapic_lock, flags);
599 dest = set_desc_affinity(desc, mask);
600 if (dest != BAD_APICID) {
601 /* Only the high 8 bits are valid. */
602 dest = SET_APIC_LOGICAL_ID(dest);
603 __target_IO_APIC_irq(irq, dest, cfg);
605 spin_unlock_irqrestore(&ioapic_lock, flags);
609 set_ioapic_affinity_irq(unsigned int irq, const struct cpumask *mask)
611 struct irq_desc *desc;
613 desc = irq_to_desc(irq);
615 set_ioapic_affinity_irq_desc(desc, mask);
617 #endif /* CONFIG_SMP */
620 * The common case is 1:1 IRQ<->pin mappings. Sometimes there are
621 * shared ISA-space IRQs, so we have to support them. We are super
622 * fast in the common case, and fast for shared ISA-space IRQs.
624 static void add_pin_to_irq_cpu(struct irq_cfg *cfg, int cpu, int apic, int pin)
626 struct irq_pin_list *entry;
628 entry = cfg->irq_2_pin;
630 entry = get_one_free_irq_2_pin(cpu);
632 printk(KERN_ERR "can not alloc irq_2_pin to add %d - %d\n",
636 cfg->irq_2_pin = entry;
642 while (entry->next) {
643 /* not again, please */
644 if (entry->apic == apic && entry->pin == pin)
650 entry->next = get_one_free_irq_2_pin(cpu);
657 * Reroute an IRQ to a different pin.
659 static void __init replace_pin_at_irq_cpu(struct irq_cfg *cfg, int cpu,
660 int oldapic, int oldpin,
661 int newapic, int newpin)
663 struct irq_pin_list *entry = cfg->irq_2_pin;
667 if (entry->apic == oldapic && entry->pin == oldpin) {
668 entry->apic = newapic;
671 /* every one is different, right? */
677 /* why? call replace before add? */
679 add_pin_to_irq_cpu(cfg, cpu, newapic, newpin);
682 static inline void io_apic_modify_irq(struct irq_cfg *cfg,
683 int mask_and, int mask_or,
684 void (*final)(struct irq_pin_list *entry))
687 struct irq_pin_list *entry;
689 for (entry = cfg->irq_2_pin; entry != NULL; entry = entry->next) {
692 reg = io_apic_read(entry->apic, 0x10 + pin * 2);
695 io_apic_modify(entry->apic, 0x10 + pin * 2, reg);
701 static void __unmask_IO_APIC_irq(struct irq_cfg *cfg)
703 io_apic_modify_irq(cfg, ~IO_APIC_REDIR_MASKED, 0, NULL);
707 static void io_apic_sync(struct irq_pin_list *entry)
710 * Synchronize the IO-APIC and the CPU by doing
711 * a dummy read from the IO-APIC
713 struct io_apic __iomem *io_apic;
714 io_apic = io_apic_base(entry->apic);
715 readl(&io_apic->data);
718 static void __mask_IO_APIC_irq(struct irq_cfg *cfg)
720 io_apic_modify_irq(cfg, ~0, IO_APIC_REDIR_MASKED, &io_apic_sync);
722 #else /* CONFIG_X86_32 */
723 static void __mask_IO_APIC_irq(struct irq_cfg *cfg)
725 io_apic_modify_irq(cfg, ~0, IO_APIC_REDIR_MASKED, NULL);
728 static void __mask_and_edge_IO_APIC_irq(struct irq_cfg *cfg)
730 io_apic_modify_irq(cfg, ~IO_APIC_REDIR_LEVEL_TRIGGER,
731 IO_APIC_REDIR_MASKED, NULL);
734 static void __unmask_and_level_IO_APIC_irq(struct irq_cfg *cfg)
736 io_apic_modify_irq(cfg, ~IO_APIC_REDIR_MASKED,
737 IO_APIC_REDIR_LEVEL_TRIGGER, NULL);
739 #endif /* CONFIG_X86_32 */
741 static void mask_IO_APIC_irq_desc(struct irq_desc *desc)
743 struct irq_cfg *cfg = desc->chip_data;
748 spin_lock_irqsave(&ioapic_lock, flags);
749 __mask_IO_APIC_irq(cfg);
750 spin_unlock_irqrestore(&ioapic_lock, flags);
753 static void unmask_IO_APIC_irq_desc(struct irq_desc *desc)
755 struct irq_cfg *cfg = desc->chip_data;
758 spin_lock_irqsave(&ioapic_lock, flags);
759 __unmask_IO_APIC_irq(cfg);
760 spin_unlock_irqrestore(&ioapic_lock, flags);
763 static void mask_IO_APIC_irq(unsigned int irq)
765 struct irq_desc *desc = irq_to_desc(irq);
767 mask_IO_APIC_irq_desc(desc);
769 static void unmask_IO_APIC_irq(unsigned int irq)
771 struct irq_desc *desc = irq_to_desc(irq);
773 unmask_IO_APIC_irq_desc(desc);
776 static void clear_IO_APIC_pin(unsigned int apic, unsigned int pin)
778 struct IO_APIC_route_entry entry;
780 /* Check delivery_mode to be sure we're not clearing an SMI pin */
781 entry = ioapic_read_entry(apic, pin);
782 if (entry.delivery_mode == dest_SMI)
785 * Disable it in the IO-APIC irq-routing table:
787 ioapic_mask_entry(apic, pin);
790 static void clear_IO_APIC (void)
794 for (apic = 0; apic < nr_ioapics; apic++)
795 for (pin = 0; pin < nr_ioapic_registers[apic]; pin++)
796 clear_IO_APIC_pin(apic, pin);
799 #if !defined(CONFIG_SMP) && defined(CONFIG_X86_32)
800 void send_IPI_self(int vector)
807 apic_wait_icr_idle();
808 cfg = APIC_DM_FIXED | APIC_DEST_SELF | vector | APIC_DEST_LOGICAL;
810 * Send the IPI. The write to APIC_ICR fires this off.
812 apic_write(APIC_ICR, cfg);
814 #endif /* !CONFIG_SMP && CONFIG_X86_32*/
818 * support for broken MP BIOSs, enables hand-redirection of PIRQ0-7 to
819 * specific CPU-side IRQs.
823 static int pirq_entries [MAX_PIRQS];
824 static int pirqs_enabled;
826 static int __init ioapic_pirq_setup(char *str)
829 int ints[MAX_PIRQS+1];
831 get_options(str, ARRAY_SIZE(ints), ints);
833 for (i = 0; i < MAX_PIRQS; i++)
834 pirq_entries[i] = -1;
837 apic_printk(APIC_VERBOSE, KERN_INFO
838 "PIRQ redirection, working around broken MP-BIOS.\n");
840 if (ints[0] < MAX_PIRQS)
843 for (i = 0; i < max; i++) {
844 apic_printk(APIC_VERBOSE, KERN_DEBUG
845 "... PIRQ%d -> IRQ %d\n", i, ints[i+1]);
847 * PIRQs are mapped upside down, usually.
849 pirq_entries[MAX_PIRQS-i-1] = ints[i+1];
854 __setup("pirq=", ioapic_pirq_setup);
855 #endif /* CONFIG_X86_32 */
857 #ifdef CONFIG_INTR_REMAP
858 /* I/O APIC RTE contents at the OS boot up */
859 static struct IO_APIC_route_entry *early_ioapic_entries[MAX_IO_APICS];
862 * Saves and masks all the unmasked IO-APIC RTE's
864 int save_mask_IO_APIC_setup(void)
866 union IO_APIC_reg_01 reg_01;
871 * The number of IO-APIC IRQ registers (== #pins):
873 for (apic = 0; apic < nr_ioapics; apic++) {
874 spin_lock_irqsave(&ioapic_lock, flags);
875 reg_01.raw = io_apic_read(apic, 1);
876 spin_unlock_irqrestore(&ioapic_lock, flags);
877 nr_ioapic_registers[apic] = reg_01.bits.entries+1;
880 for (apic = 0; apic < nr_ioapics; apic++) {
881 early_ioapic_entries[apic] =
882 kzalloc(sizeof(struct IO_APIC_route_entry) *
883 nr_ioapic_registers[apic], GFP_KERNEL);
884 if (!early_ioapic_entries[apic])
888 for (apic = 0; apic < nr_ioapics; apic++)
889 for (pin = 0; pin < nr_ioapic_registers[apic]; pin++) {
890 struct IO_APIC_route_entry entry;
892 entry = early_ioapic_entries[apic][pin] =
893 ioapic_read_entry(apic, pin);
896 ioapic_write_entry(apic, pin, entry);
904 kfree(early_ioapic_entries[apic--]);
905 memset(early_ioapic_entries, 0,
906 ARRAY_SIZE(early_ioapic_entries));
911 void restore_IO_APIC_setup(void)
915 for (apic = 0; apic < nr_ioapics; apic++) {
916 if (!early_ioapic_entries[apic])
918 for (pin = 0; pin < nr_ioapic_registers[apic]; pin++)
919 ioapic_write_entry(apic, pin,
920 early_ioapic_entries[apic][pin]);
921 kfree(early_ioapic_entries[apic]);
922 early_ioapic_entries[apic] = NULL;
926 void reinit_intr_remapped_IO_APIC(int intr_remapping)
929 * for now plain restore of previous settings.
930 * TBD: In the case of OS enabling interrupt-remapping,
931 * IO-APIC RTE's need to be setup to point to interrupt-remapping
932 * table entries. for now, do a plain restore, and wait for
933 * the setup_IO_APIC_irqs() to do proper initialization.
935 restore_IO_APIC_setup();
940 * Find the IRQ entry number of a certain pin.
942 static int find_irq_entry(int apic, int pin, int type)
946 for (i = 0; i < mp_irq_entries; i++)
947 if (mp_irqs[i].mp_irqtype == type &&
948 (mp_irqs[i].mp_dstapic == mp_ioapics[apic].mp_apicid ||
949 mp_irqs[i].mp_dstapic == MP_APIC_ALL) &&
950 mp_irqs[i].mp_dstirq == pin)
957 * Find the pin to which IRQ[irq] (ISA) is connected
959 static int __init find_isa_irq_pin(int irq, int type)
963 for (i = 0; i < mp_irq_entries; i++) {
964 int lbus = mp_irqs[i].mp_srcbus;
966 if (test_bit(lbus, mp_bus_not_pci) &&
967 (mp_irqs[i].mp_irqtype == type) &&
968 (mp_irqs[i].mp_srcbusirq == irq))
970 return mp_irqs[i].mp_dstirq;
975 static int __init find_isa_irq_apic(int irq, int type)
979 for (i = 0; i < mp_irq_entries; i++) {
980 int lbus = mp_irqs[i].mp_srcbus;
982 if (test_bit(lbus, mp_bus_not_pci) &&
983 (mp_irqs[i].mp_irqtype == type) &&
984 (mp_irqs[i].mp_srcbusirq == irq))
987 if (i < mp_irq_entries) {
989 for(apic = 0; apic < nr_ioapics; apic++) {
990 if (mp_ioapics[apic].mp_apicid == mp_irqs[i].mp_dstapic)
999 * Find a specific PCI IRQ entry.
1000 * Not an __init, possibly needed by modules
1002 static int pin_2_irq(int idx, int apic, int pin);
1004 int IO_APIC_get_PCI_irq_vector(int bus, int slot, int pin)
1006 int apic, i, best_guess = -1;
1008 apic_printk(APIC_DEBUG, "querying PCI -> IRQ mapping bus:%d, slot:%d, pin:%d.\n",
1010 if (test_bit(bus, mp_bus_not_pci)) {
1011 apic_printk(APIC_VERBOSE, "PCI BIOS passed nonexistent PCI bus %d!\n", bus);
1014 for (i = 0; i < mp_irq_entries; i++) {
1015 int lbus = mp_irqs[i].mp_srcbus;
1017 for (apic = 0; apic < nr_ioapics; apic++)
1018 if (mp_ioapics[apic].mp_apicid == mp_irqs[i].mp_dstapic ||
1019 mp_irqs[i].mp_dstapic == MP_APIC_ALL)
1022 if (!test_bit(lbus, mp_bus_not_pci) &&
1023 !mp_irqs[i].mp_irqtype &&
1025 (slot == ((mp_irqs[i].mp_srcbusirq >> 2) & 0x1f))) {
1026 int irq = pin_2_irq(i,apic,mp_irqs[i].mp_dstirq);
1028 if (!(apic || IO_APIC_IRQ(irq)))
1031 if (pin == (mp_irqs[i].mp_srcbusirq & 3))
1034 * Use the first all-but-pin matching entry as a
1035 * best-guess fuzzy result for broken mptables.
1044 EXPORT_SYMBOL(IO_APIC_get_PCI_irq_vector);
1046 #if defined(CONFIG_EISA) || defined(CONFIG_MCA)
1048 * EISA Edge/Level control register, ELCR
1050 static int EISA_ELCR(unsigned int irq)
1052 if (irq < NR_IRQS_LEGACY) {
1053 unsigned int port = 0x4d0 + (irq >> 3);
1054 return (inb(port) >> (irq & 7)) & 1;
1056 apic_printk(APIC_VERBOSE, KERN_INFO
1057 "Broken MPtable reports ISA irq %d\n", irq);
1063 /* ISA interrupts are always polarity zero edge triggered,
1064 * when listed as conforming in the MP table. */
1066 #define default_ISA_trigger(idx) (0)
1067 #define default_ISA_polarity(idx) (0)
1069 /* EISA interrupts are always polarity zero and can be edge or level
1070 * trigger depending on the ELCR value. If an interrupt is listed as
1071 * EISA conforming in the MP table, that means its trigger type must
1072 * be read in from the ELCR */
1074 #define default_EISA_trigger(idx) (EISA_ELCR(mp_irqs[idx].mp_srcbusirq))
1075 #define default_EISA_polarity(idx) default_ISA_polarity(idx)
1077 /* PCI interrupts are always polarity one level triggered,
1078 * when listed as conforming in the MP table. */
1080 #define default_PCI_trigger(idx) (1)
1081 #define default_PCI_polarity(idx) (1)
1083 /* MCA interrupts are always polarity zero level triggered,
1084 * when listed as conforming in the MP table. */
1086 #define default_MCA_trigger(idx) (1)
1087 #define default_MCA_polarity(idx) default_ISA_polarity(idx)
1089 static int MPBIOS_polarity(int idx)
1091 int bus = mp_irqs[idx].mp_srcbus;
1095 * Determine IRQ line polarity (high active or low active):
1097 switch (mp_irqs[idx].mp_irqflag & 3)
1099 case 0: /* conforms, ie. bus-type dependent polarity */
1100 if (test_bit(bus, mp_bus_not_pci))
1101 polarity = default_ISA_polarity(idx);
1103 polarity = default_PCI_polarity(idx);
1105 case 1: /* high active */
1110 case 2: /* reserved */
1112 printk(KERN_WARNING "broken BIOS!!\n");
1116 case 3: /* low active */
1121 default: /* invalid */
1123 printk(KERN_WARNING "broken BIOS!!\n");
1131 static int MPBIOS_trigger(int idx)
1133 int bus = mp_irqs[idx].mp_srcbus;
1137 * Determine IRQ trigger mode (edge or level sensitive):
1139 switch ((mp_irqs[idx].mp_irqflag>>2) & 3)
1141 case 0: /* conforms, ie. bus-type dependent */
1142 if (test_bit(bus, mp_bus_not_pci))
1143 trigger = default_ISA_trigger(idx);
1145 trigger = default_PCI_trigger(idx);
1146 #if defined(CONFIG_EISA) || defined(CONFIG_MCA)
1147 switch (mp_bus_id_to_type[bus]) {
1148 case MP_BUS_ISA: /* ISA pin */
1150 /* set before the switch */
1153 case MP_BUS_EISA: /* EISA pin */
1155 trigger = default_EISA_trigger(idx);
1158 case MP_BUS_PCI: /* PCI pin */
1160 /* set before the switch */
1163 case MP_BUS_MCA: /* MCA pin */
1165 trigger = default_MCA_trigger(idx);
1170 printk(KERN_WARNING "broken BIOS!!\n");
1182 case 2: /* reserved */
1184 printk(KERN_WARNING "broken BIOS!!\n");
1193 default: /* invalid */
1195 printk(KERN_WARNING "broken BIOS!!\n");
1203 static inline int irq_polarity(int idx)
1205 return MPBIOS_polarity(idx);
1208 static inline int irq_trigger(int idx)
1210 return MPBIOS_trigger(idx);
1213 int (*ioapic_renumber_irq)(int ioapic, int irq);
1214 static int pin_2_irq(int idx, int apic, int pin)
1217 int bus = mp_irqs[idx].mp_srcbus;
1220 * Debugging check, we are in big trouble if this message pops up!
1222 if (mp_irqs[idx].mp_dstirq != pin)
1223 printk(KERN_ERR "broken BIOS or MPTABLE parser, ayiee!!\n");
1225 if (test_bit(bus, mp_bus_not_pci)) {
1226 irq = mp_irqs[idx].mp_srcbusirq;
1229 * PCI IRQs are mapped in order
1233 irq += nr_ioapic_registers[i++];
1236 * For MPS mode, so far only needed by ES7000 platform
1238 if (ioapic_renumber_irq)
1239 irq = ioapic_renumber_irq(apic, irq);
1242 #ifdef CONFIG_X86_32
1244 * PCI IRQ command line redirection. Yes, limits are hardcoded.
1246 if ((pin >= 16) && (pin <= 23)) {
1247 if (pirq_entries[pin-16] != -1) {
1248 if (!pirq_entries[pin-16]) {
1249 apic_printk(APIC_VERBOSE, KERN_DEBUG
1250 "disabling PIRQ%d\n", pin-16);
1252 irq = pirq_entries[pin-16];
1253 apic_printk(APIC_VERBOSE, KERN_DEBUG
1254 "using PIRQ%d -> IRQ %d\n",
1264 void lock_vector_lock(void)
1266 /* Used to the online set of cpus does not change
1267 * during assign_irq_vector.
1269 spin_lock(&vector_lock);
1272 void unlock_vector_lock(void)
1274 spin_unlock(&vector_lock);
1278 __assign_irq_vector(int irq, struct irq_cfg *cfg, const struct cpumask *mask)
1281 * NOTE! The local APIC isn't very good at handling
1282 * multiple interrupts at the same interrupt level.
1283 * As the interrupt level is determined by taking the
1284 * vector number and shifting that right by 4, we
1285 * want to spread these out a bit so that they don't
1286 * all fall in the same interrupt level.
1288 * Also, we've got to be careful not to trash gate
1289 * 0x80, because int 0x80 is hm, kind of importantish. ;)
1291 static int current_vector = FIRST_DEVICE_VECTOR, current_offset = 0;
1292 unsigned int old_vector;
1294 cpumask_var_t tmp_mask;
1296 if ((cfg->move_in_progress) || cfg->move_cleanup_count)
1299 if (!alloc_cpumask_var(&tmp_mask, GFP_ATOMIC))
1302 old_vector = cfg->vector;
1304 cpumask_and(tmp_mask, mask, cpu_online_mask);
1305 cpumask_and(tmp_mask, cfg->domain, tmp_mask);
1306 if (!cpumask_empty(tmp_mask)) {
1307 free_cpumask_var(tmp_mask);
1312 /* Only try and allocate irqs on cpus that are present */
1314 for_each_cpu_and(cpu, mask, cpu_online_mask) {
1318 vector_allocation_domain(cpu, tmp_mask);
1320 vector = current_vector;
1321 offset = current_offset;
1324 if (vector >= first_system_vector) {
1325 /* If out of vectors on large boxen, must share them. */
1326 offset = (offset + 1) % 8;
1327 vector = FIRST_DEVICE_VECTOR + offset;
1329 if (unlikely(current_vector == vector))
1332 if (test_bit(vector, used_vectors))
1335 for_each_cpu_and(new_cpu, tmp_mask, cpu_online_mask)
1336 if (per_cpu(vector_irq, new_cpu)[vector] != -1)
1339 current_vector = vector;
1340 current_offset = offset;
1342 cfg->move_in_progress = 1;
1343 cpumask_copy(cfg->old_domain, cfg->domain);
1345 for_each_cpu_and(new_cpu, tmp_mask, cpu_online_mask)
1346 per_cpu(vector_irq, new_cpu)[vector] = irq;
1347 cfg->vector = vector;
1348 cpumask_copy(cfg->domain, tmp_mask);
1352 free_cpumask_var(tmp_mask);
1357 assign_irq_vector(int irq, struct irq_cfg *cfg, const struct cpumask *mask)
1360 unsigned long flags;
1362 spin_lock_irqsave(&vector_lock, flags);
1363 err = __assign_irq_vector(irq, cfg, mask);
1364 spin_unlock_irqrestore(&vector_lock, flags);
1368 static void __clear_irq_vector(int irq, struct irq_cfg *cfg)
1372 BUG_ON(!cfg->vector);
1374 vector = cfg->vector;
1375 for_each_cpu_and(cpu, cfg->domain, cpu_online_mask)
1376 per_cpu(vector_irq, cpu)[vector] = -1;
1379 cpumask_clear(cfg->domain);
1381 if (likely(!cfg->move_in_progress))
1383 for_each_cpu_and(cpu, cfg->old_domain, cpu_online_mask) {
1384 for (vector = FIRST_EXTERNAL_VECTOR; vector < NR_VECTORS;
1386 if (per_cpu(vector_irq, cpu)[vector] != irq)
1388 per_cpu(vector_irq, cpu)[vector] = -1;
1392 cfg->move_in_progress = 0;
1395 void __setup_vector_irq(int cpu)
1397 /* Initialize vector_irq on a new cpu */
1398 /* This function must be called with vector_lock held */
1400 struct irq_cfg *cfg;
1401 struct irq_desc *desc;
1403 /* Mark the inuse vectors */
1404 for_each_irq_desc(irq, desc) {
1405 cfg = desc->chip_data;
1406 if (!cpumask_test_cpu(cpu, cfg->domain))
1408 vector = cfg->vector;
1409 per_cpu(vector_irq, cpu)[vector] = irq;
1411 /* Mark the free vectors */
1412 for (vector = 0; vector < NR_VECTORS; ++vector) {
1413 irq = per_cpu(vector_irq, cpu)[vector];
1418 if (!cpumask_test_cpu(cpu, cfg->domain))
1419 per_cpu(vector_irq, cpu)[vector] = -1;
1423 static struct irq_chip ioapic_chip;
1424 #ifdef CONFIG_INTR_REMAP
1425 static struct irq_chip ir_ioapic_chip;
1428 #define IOAPIC_AUTO -1
1429 #define IOAPIC_EDGE 0
1430 #define IOAPIC_LEVEL 1
1432 #ifdef CONFIG_X86_32
1433 static inline int IO_APIC_irq_trigger(int irq)
1437 for (apic = 0; apic < nr_ioapics; apic++) {
1438 for (pin = 0; pin < nr_ioapic_registers[apic]; pin++) {
1439 idx = find_irq_entry(apic, pin, mp_INT);
1440 if ((idx != -1) && (irq == pin_2_irq(idx, apic, pin)))
1441 return irq_trigger(idx);
1445 * nonexistent IRQs are edge default
1450 static inline int IO_APIC_irq_trigger(int irq)
1456 static void ioapic_register_intr(int irq, struct irq_desc *desc, unsigned long trigger)
1459 if ((trigger == IOAPIC_AUTO && IO_APIC_irq_trigger(irq)) ||
1460 trigger == IOAPIC_LEVEL)
1461 desc->status |= IRQ_LEVEL;
1463 desc->status &= ~IRQ_LEVEL;
1465 #ifdef CONFIG_INTR_REMAP
1466 if (irq_remapped(irq)) {
1467 desc->status |= IRQ_MOVE_PCNTXT;
1469 set_irq_chip_and_handler_name(irq, &ir_ioapic_chip,
1473 set_irq_chip_and_handler_name(irq, &ir_ioapic_chip,
1474 handle_edge_irq, "edge");
1478 if ((trigger == IOAPIC_AUTO && IO_APIC_irq_trigger(irq)) ||
1479 trigger == IOAPIC_LEVEL)
1480 set_irq_chip_and_handler_name(irq, &ioapic_chip,
1484 set_irq_chip_and_handler_name(irq, &ioapic_chip,
1485 handle_edge_irq, "edge");
1488 static int setup_ioapic_entry(int apic, int irq,
1489 struct IO_APIC_route_entry *entry,
1490 unsigned int destination, int trigger,
1491 int polarity, int vector)
1494 * add it to the IO-APIC irq-routing table:
1496 memset(entry,0,sizeof(*entry));
1498 #ifdef CONFIG_INTR_REMAP
1499 if (intr_remapping_enabled) {
1500 struct intel_iommu *iommu = map_ioapic_to_ir(apic);
1502 struct IR_IO_APIC_route_entry *ir_entry =
1503 (struct IR_IO_APIC_route_entry *) entry;
1507 panic("No mapping iommu for ioapic %d\n", apic);
1509 index = alloc_irte(iommu, irq, 1);
1511 panic("Failed to allocate IRTE for ioapic %d\n", apic);
1513 memset(&irte, 0, sizeof(irte));
1516 irte.dst_mode = INT_DEST_MODE;
1517 irte.trigger_mode = trigger;
1518 irte.dlvry_mode = INT_DELIVERY_MODE;
1519 irte.vector = vector;
1520 irte.dest_id = IRTE_DEST(destination);
1522 modify_irte(irq, &irte);
1524 ir_entry->index2 = (index >> 15) & 0x1;
1526 ir_entry->format = 1;
1527 ir_entry->index = (index & 0x7fff);
1531 entry->delivery_mode = INT_DELIVERY_MODE;
1532 entry->dest_mode = INT_DEST_MODE;
1533 entry->dest = destination;
1536 entry->mask = 0; /* enable IRQ */
1537 entry->trigger = trigger;
1538 entry->polarity = polarity;
1539 entry->vector = vector;
1541 /* Mask level triggered irqs.
1542 * Use IRQ_DELAYED_DISABLE for edge triggered irqs.
1549 static void setup_IO_APIC_irq(int apic, int pin, unsigned int irq, struct irq_desc *desc,
1550 int trigger, int polarity)
1552 struct irq_cfg *cfg;
1553 struct IO_APIC_route_entry entry;
1556 if (!IO_APIC_IRQ(irq))
1559 cfg = desc->chip_data;
1561 if (assign_irq_vector(irq, cfg, TARGET_CPUS))
1564 dest = cpu_mask_to_apicid_and(cfg->domain, TARGET_CPUS);
1566 apic_printk(APIC_VERBOSE,KERN_DEBUG
1567 "IOAPIC[%d]: Set routing entry (%d-%d -> 0x%x -> "
1568 "IRQ %d Mode:%i Active:%i)\n",
1569 apic, mp_ioapics[apic].mp_apicid, pin, cfg->vector,
1570 irq, trigger, polarity);
1573 if (setup_ioapic_entry(mp_ioapics[apic].mp_apicid, irq, &entry,
1574 dest, trigger, polarity, cfg->vector)) {
1575 printk("Failed to setup ioapic entry for ioapic %d, pin %d\n",
1576 mp_ioapics[apic].mp_apicid, pin);
1577 __clear_irq_vector(irq, cfg);
1581 ioapic_register_intr(irq, desc, trigger);
1582 if (irq < NR_IRQS_LEGACY)
1583 disable_8259A_irq(irq);
1585 ioapic_write_entry(apic, pin, entry);
1588 static void __init setup_IO_APIC_irqs(void)
1590 int apic, pin, idx, irq;
1592 struct irq_desc *desc;
1593 struct irq_cfg *cfg;
1594 int cpu = boot_cpu_id;
1596 apic_printk(APIC_VERBOSE, KERN_DEBUG "init IO_APIC IRQs\n");
1598 for (apic = 0; apic < nr_ioapics; apic++) {
1599 for (pin = 0; pin < nr_ioapic_registers[apic]; pin++) {
1601 idx = find_irq_entry(apic, pin, mp_INT);
1605 apic_printk(APIC_VERBOSE,
1606 KERN_DEBUG " %d-%d",
1607 mp_ioapics[apic].mp_apicid,
1610 apic_printk(APIC_VERBOSE, " %d-%d",
1611 mp_ioapics[apic].mp_apicid,
1616 apic_printk(APIC_VERBOSE,
1617 " (apicid-pin) not connected\n");
1621 irq = pin_2_irq(idx, apic, pin);
1622 #ifdef CONFIG_X86_32
1623 if (multi_timer_check(apic, irq))
1626 desc = irq_to_desc_alloc_cpu(irq, cpu);
1628 printk(KERN_INFO "can not get irq_desc for %d\n", irq);
1631 cfg = desc->chip_data;
1632 add_pin_to_irq_cpu(cfg, cpu, apic, pin);
1634 setup_IO_APIC_irq(apic, pin, irq, desc,
1635 irq_trigger(idx), irq_polarity(idx));
1640 apic_printk(APIC_VERBOSE,
1641 " (apicid-pin) not connected\n");
1645 * Set up the timer pin, possibly with the 8259A-master behind.
1647 static void __init setup_timer_IRQ0_pin(unsigned int apic, unsigned int pin,
1650 struct IO_APIC_route_entry entry;
1652 #ifdef CONFIG_INTR_REMAP
1653 if (intr_remapping_enabled)
1657 memset(&entry, 0, sizeof(entry));
1660 * We use logical delivery to get the timer IRQ
1663 entry.dest_mode = INT_DEST_MODE;
1664 entry.mask = 1; /* mask IRQ now */
1665 entry.dest = cpu_mask_to_apicid(TARGET_CPUS);
1666 entry.delivery_mode = INT_DELIVERY_MODE;
1669 entry.vector = vector;
1672 * The timer IRQ doesn't have to know that behind the
1673 * scene we may have a 8259A-master in AEOI mode ...
1675 set_irq_chip_and_handler_name(0, &ioapic_chip, handle_edge_irq, "edge");
1678 * Add it to the IO-APIC irq-routing table:
1680 ioapic_write_entry(apic, pin, entry);
1684 __apicdebuginit(void) print_IO_APIC(void)
1687 union IO_APIC_reg_00 reg_00;
1688 union IO_APIC_reg_01 reg_01;
1689 union IO_APIC_reg_02 reg_02;
1690 union IO_APIC_reg_03 reg_03;
1691 unsigned long flags;
1692 struct irq_cfg *cfg;
1693 struct irq_desc *desc;
1696 if (apic_verbosity == APIC_QUIET)
1699 printk(KERN_DEBUG "number of MP IRQ sources: %d.\n", mp_irq_entries);
1700 for (i = 0; i < nr_ioapics; i++)
1701 printk(KERN_DEBUG "number of IO-APIC #%d registers: %d.\n",
1702 mp_ioapics[i].mp_apicid, nr_ioapic_registers[i]);
1705 * We are a bit conservative about what we expect. We have to
1706 * know about every hardware change ASAP.
1708 printk(KERN_INFO "testing the IO APIC.......................\n");
1710 for (apic = 0; apic < nr_ioapics; apic++) {
1712 spin_lock_irqsave(&ioapic_lock, flags);
1713 reg_00.raw = io_apic_read(apic, 0);
1714 reg_01.raw = io_apic_read(apic, 1);
1715 if (reg_01.bits.version >= 0x10)
1716 reg_02.raw = io_apic_read(apic, 2);
1717 if (reg_01.bits.version >= 0x20)
1718 reg_03.raw = io_apic_read(apic, 3);
1719 spin_unlock_irqrestore(&ioapic_lock, flags);
1722 printk(KERN_DEBUG "IO APIC #%d......\n", mp_ioapics[apic].mp_apicid);
1723 printk(KERN_DEBUG ".... register #00: %08X\n", reg_00.raw);
1724 printk(KERN_DEBUG "....... : physical APIC id: %02X\n", reg_00.bits.ID);
1725 printk(KERN_DEBUG "....... : Delivery Type: %X\n", reg_00.bits.delivery_type);
1726 printk(KERN_DEBUG "....... : LTS : %X\n", reg_00.bits.LTS);
1728 printk(KERN_DEBUG ".... register #01: %08X\n", *(int *)®_01);
1729 printk(KERN_DEBUG "....... : max redirection entries: %04X\n", reg_01.bits.entries);
1731 printk(KERN_DEBUG "....... : PRQ implemented: %X\n", reg_01.bits.PRQ);
1732 printk(KERN_DEBUG "....... : IO APIC version: %04X\n", reg_01.bits.version);
1735 * Some Intel chipsets with IO APIC VERSION of 0x1? don't have reg_02,
1736 * but the value of reg_02 is read as the previous read register
1737 * value, so ignore it if reg_02 == reg_01.
1739 if (reg_01.bits.version >= 0x10 && reg_02.raw != reg_01.raw) {
1740 printk(KERN_DEBUG ".... register #02: %08X\n", reg_02.raw);
1741 printk(KERN_DEBUG "....... : arbitration: %02X\n", reg_02.bits.arbitration);
1745 * Some Intel chipsets with IO APIC VERSION of 0x2? don't have reg_02
1746 * or reg_03, but the value of reg_0[23] is read as the previous read
1747 * register value, so ignore it if reg_03 == reg_0[12].
1749 if (reg_01.bits.version >= 0x20 && reg_03.raw != reg_02.raw &&
1750 reg_03.raw != reg_01.raw) {
1751 printk(KERN_DEBUG ".... register #03: %08X\n", reg_03.raw);
1752 printk(KERN_DEBUG "....... : Boot DT : %X\n", reg_03.bits.boot_DT);
1755 printk(KERN_DEBUG ".... IRQ redirection table:\n");
1757 printk(KERN_DEBUG " NR Dst Mask Trig IRR Pol"
1758 " Stat Dmod Deli Vect: \n");
1760 for (i = 0; i <= reg_01.bits.entries; i++) {
1761 struct IO_APIC_route_entry entry;
1763 entry = ioapic_read_entry(apic, i);
1765 printk(KERN_DEBUG " %02x %03X ",
1770 printk("%1d %1d %1d %1d %1d %1d %1d %02X\n",
1775 entry.delivery_status,
1777 entry.delivery_mode,
1782 printk(KERN_DEBUG "IRQ to pin mappings:\n");
1783 for_each_irq_desc(irq, desc) {
1784 struct irq_pin_list *entry;
1786 cfg = desc->chip_data;
1787 entry = cfg->irq_2_pin;
1790 printk(KERN_DEBUG "IRQ%d ", irq);
1792 printk("-> %d:%d", entry->apic, entry->pin);
1795 entry = entry->next;
1800 printk(KERN_INFO ".................................... done.\n");
1805 __apicdebuginit(void) print_APIC_bitfield(int base)
1810 if (apic_verbosity == APIC_QUIET)
1813 printk(KERN_DEBUG "0123456789abcdef0123456789abcdef\n" KERN_DEBUG);
1814 for (i = 0; i < 8; i++) {
1815 v = apic_read(base + i*0x10);
1816 for (j = 0; j < 32; j++) {
1826 __apicdebuginit(void) print_local_APIC(void *dummy)
1828 unsigned int v, ver, maxlvt;
1831 if (apic_verbosity == APIC_QUIET)
1834 printk("\n" KERN_DEBUG "printing local APIC contents on CPU#%d/%d:\n",
1835 smp_processor_id(), hard_smp_processor_id());
1836 v = apic_read(APIC_ID);
1837 printk(KERN_INFO "... APIC ID: %08x (%01x)\n", v, read_apic_id());
1838 v = apic_read(APIC_LVR);
1839 printk(KERN_INFO "... APIC VERSION: %08x\n", v);
1840 ver = GET_APIC_VERSION(v);
1841 maxlvt = lapic_get_maxlvt();
1843 v = apic_read(APIC_TASKPRI);
1844 printk(KERN_DEBUG "... APIC TASKPRI: %08x (%02x)\n", v, v & APIC_TPRI_MASK);
1846 if (APIC_INTEGRATED(ver)) { /* !82489DX */
1847 if (!APIC_XAPIC(ver)) {
1848 v = apic_read(APIC_ARBPRI);
1849 printk(KERN_DEBUG "... APIC ARBPRI: %08x (%02x)\n", v,
1850 v & APIC_ARBPRI_MASK);
1852 v = apic_read(APIC_PROCPRI);
1853 printk(KERN_DEBUG "... APIC PROCPRI: %08x\n", v);
1857 * Remote read supported only in the 82489DX and local APIC for
1858 * Pentium processors.
1860 if (!APIC_INTEGRATED(ver) || maxlvt == 3) {
1861 v = apic_read(APIC_RRR);
1862 printk(KERN_DEBUG "... APIC RRR: %08x\n", v);
1865 v = apic_read(APIC_LDR);
1866 printk(KERN_DEBUG "... APIC LDR: %08x\n", v);
1867 if (!x2apic_enabled()) {
1868 v = apic_read(APIC_DFR);
1869 printk(KERN_DEBUG "... APIC DFR: %08x\n", v);
1871 v = apic_read(APIC_SPIV);
1872 printk(KERN_DEBUG "... APIC SPIV: %08x\n", v);
1874 printk(KERN_DEBUG "... APIC ISR field:\n");
1875 print_APIC_bitfield(APIC_ISR);
1876 printk(KERN_DEBUG "... APIC TMR field:\n");
1877 print_APIC_bitfield(APIC_TMR);
1878 printk(KERN_DEBUG "... APIC IRR field:\n");
1879 print_APIC_bitfield(APIC_IRR);
1881 if (APIC_INTEGRATED(ver)) { /* !82489DX */
1882 if (maxlvt > 3) /* Due to the Pentium erratum 3AP. */
1883 apic_write(APIC_ESR, 0);
1885 v = apic_read(APIC_ESR);
1886 printk(KERN_DEBUG "... APIC ESR: %08x\n", v);
1889 icr = apic_icr_read();
1890 printk(KERN_DEBUG "... APIC ICR: %08x\n", (u32)icr);
1891 printk(KERN_DEBUG "... APIC ICR2: %08x\n", (u32)(icr >> 32));
1893 v = apic_read(APIC_LVTT);
1894 printk(KERN_DEBUG "... APIC LVTT: %08x\n", v);
1896 if (maxlvt > 3) { /* PC is LVT#4. */
1897 v = apic_read(APIC_LVTPC);
1898 printk(KERN_DEBUG "... APIC LVTPC: %08x\n", v);
1900 v = apic_read(APIC_LVT0);
1901 printk(KERN_DEBUG "... APIC LVT0: %08x\n", v);
1902 v = apic_read(APIC_LVT1);
1903 printk(KERN_DEBUG "... APIC LVT1: %08x\n", v);
1905 if (maxlvt > 2) { /* ERR is LVT#3. */
1906 v = apic_read(APIC_LVTERR);
1907 printk(KERN_DEBUG "... APIC LVTERR: %08x\n", v);
1910 v = apic_read(APIC_TMICT);
1911 printk(KERN_DEBUG "... APIC TMICT: %08x\n", v);
1912 v = apic_read(APIC_TMCCT);
1913 printk(KERN_DEBUG "... APIC TMCCT: %08x\n", v);
1914 v = apic_read(APIC_TDCR);
1915 printk(KERN_DEBUG "... APIC TDCR: %08x\n", v);
1919 __apicdebuginit(void) print_all_local_APICs(void)
1924 for_each_online_cpu(cpu)
1925 smp_call_function_single(cpu, print_local_APIC, NULL, 1);
1929 __apicdebuginit(void) print_PIC(void)
1932 unsigned long flags;
1934 if (apic_verbosity == APIC_QUIET)
1937 printk(KERN_DEBUG "\nprinting PIC contents\n");
1939 spin_lock_irqsave(&i8259A_lock, flags);
1941 v = inb(0xa1) << 8 | inb(0x21);
1942 printk(KERN_DEBUG "... PIC IMR: %04x\n", v);
1944 v = inb(0xa0) << 8 | inb(0x20);
1945 printk(KERN_DEBUG "... PIC IRR: %04x\n", v);
1949 v = inb(0xa0) << 8 | inb(0x20);
1953 spin_unlock_irqrestore(&i8259A_lock, flags);
1955 printk(KERN_DEBUG "... PIC ISR: %04x\n", v);
1957 v = inb(0x4d1) << 8 | inb(0x4d0);
1958 printk(KERN_DEBUG "... PIC ELCR: %04x\n", v);
1961 __apicdebuginit(int) print_all_ICs(void)
1964 print_all_local_APICs();
1970 fs_initcall(print_all_ICs);
1973 /* Where if anywhere is the i8259 connect in external int mode */
1974 static struct { int pin, apic; } ioapic_i8259 = { -1, -1 };
1976 void __init enable_IO_APIC(void)
1978 union IO_APIC_reg_01 reg_01;
1979 int i8259_apic, i8259_pin;
1981 unsigned long flags;
1983 #ifdef CONFIG_X86_32
1986 for (i = 0; i < MAX_PIRQS; i++)
1987 pirq_entries[i] = -1;
1991 * The number of IO-APIC IRQ registers (== #pins):
1993 for (apic = 0; apic < nr_ioapics; apic++) {
1994 spin_lock_irqsave(&ioapic_lock, flags);
1995 reg_01.raw = io_apic_read(apic, 1);
1996 spin_unlock_irqrestore(&ioapic_lock, flags);
1997 nr_ioapic_registers[apic] = reg_01.bits.entries+1;
1999 for(apic = 0; apic < nr_ioapics; apic++) {
2001 /* See if any of the pins is in ExtINT mode */
2002 for (pin = 0; pin < nr_ioapic_registers[apic]; pin++) {
2003 struct IO_APIC_route_entry entry;
2004 entry = ioapic_read_entry(apic, pin);
2006 /* If the interrupt line is enabled and in ExtInt mode
2007 * I have found the pin where the i8259 is connected.
2009 if ((entry.mask == 0) && (entry.delivery_mode == dest_ExtINT)) {
2010 ioapic_i8259.apic = apic;
2011 ioapic_i8259.pin = pin;
2017 /* Look to see what if the MP table has reported the ExtINT */
2018 /* If we could not find the appropriate pin by looking at the ioapic
2019 * the i8259 probably is not connected the ioapic but give the
2020 * mptable a chance anyway.
2022 i8259_pin = find_isa_irq_pin(0, mp_ExtINT);
2023 i8259_apic = find_isa_irq_apic(0, mp_ExtINT);
2024 /* Trust the MP table if nothing is setup in the hardware */
2025 if ((ioapic_i8259.pin == -1) && (i8259_pin >= 0)) {
2026 printk(KERN_WARNING "ExtINT not setup in hardware but reported by MP table\n");
2027 ioapic_i8259.pin = i8259_pin;
2028 ioapic_i8259.apic = i8259_apic;
2030 /* Complain if the MP table and the hardware disagree */
2031 if (((ioapic_i8259.apic != i8259_apic) || (ioapic_i8259.pin != i8259_pin)) &&
2032 (i8259_pin >= 0) && (ioapic_i8259.pin >= 0))
2034 printk(KERN_WARNING "ExtINT in hardware and MP table differ\n");
2038 * Do not trust the IO-APIC being empty at bootup
2044 * Not an __init, needed by the reboot code
2046 void disable_IO_APIC(void)
2049 * Clear the IO-APIC before rebooting:
2054 * If the i8259 is routed through an IOAPIC
2055 * Put that IOAPIC in virtual wire mode
2056 * so legacy interrupts can be delivered.
2058 if (ioapic_i8259.pin != -1) {
2059 struct IO_APIC_route_entry entry;
2061 memset(&entry, 0, sizeof(entry));
2062 entry.mask = 0; /* Enabled */
2063 entry.trigger = 0; /* Edge */
2065 entry.polarity = 0; /* High */
2066 entry.delivery_status = 0;
2067 entry.dest_mode = 0; /* Physical */
2068 entry.delivery_mode = dest_ExtINT; /* ExtInt */
2070 entry.dest = read_apic_id();
2073 * Add it to the IO-APIC irq-routing table:
2075 ioapic_write_entry(ioapic_i8259.apic, ioapic_i8259.pin, entry);
2078 disconnect_bsp_APIC(ioapic_i8259.pin != -1);
2081 #ifdef CONFIG_X86_32
2083 * function to set the IO-APIC physical IDs based on the
2084 * values stored in the MPC table.
2086 * by Matt Domsch <Matt_Domsch@dell.com> Tue Dec 21 12:25:05 CST 1999
2089 static void __init setup_ioapic_ids_from_mpc(void)
2091 union IO_APIC_reg_00 reg_00;
2092 physid_mask_t phys_id_present_map;
2095 unsigned char old_id;
2096 unsigned long flags;
2098 if (x86_quirks->setup_ioapic_ids && x86_quirks->setup_ioapic_ids())
2102 * Don't check I/O APIC IDs for xAPIC systems. They have
2103 * no meaning without the serial APIC bus.
2105 if (!(boot_cpu_data.x86_vendor == X86_VENDOR_INTEL)
2106 || APIC_XAPIC(apic_version[boot_cpu_physical_apicid]))
2109 * This is broken; anything with a real cpu count has to
2110 * circumvent this idiocy regardless.
2112 phys_id_present_map = ioapic_phys_id_map(phys_cpu_present_map);
2115 * Set the IOAPIC ID to the value stored in the MPC table.
2117 for (apic = 0; apic < nr_ioapics; apic++) {
2119 /* Read the register 0 value */
2120 spin_lock_irqsave(&ioapic_lock, flags);
2121 reg_00.raw = io_apic_read(apic, 0);
2122 spin_unlock_irqrestore(&ioapic_lock, flags);
2124 old_id = mp_ioapics[apic].mp_apicid;
2126 if (mp_ioapics[apic].mp_apicid >= get_physical_broadcast()) {
2127 printk(KERN_ERR "BIOS bug, IO-APIC#%d ID is %d in the MPC table!...\n",
2128 apic, mp_ioapics[apic].mp_apicid);
2129 printk(KERN_ERR "... fixing up to %d. (tell your hw vendor)\n",
2131 mp_ioapics[apic].mp_apicid = reg_00.bits.ID;
2135 * Sanity check, is the ID really free? Every APIC in a
2136 * system must have a unique ID or we get lots of nice
2137 * 'stuck on smp_invalidate_needed IPI wait' messages.
2139 if (check_apicid_used(phys_id_present_map,
2140 mp_ioapics[apic].mp_apicid)) {
2141 printk(KERN_ERR "BIOS bug, IO-APIC#%d ID %d is already used!...\n",
2142 apic, mp_ioapics[apic].mp_apicid);
2143 for (i = 0; i < get_physical_broadcast(); i++)
2144 if (!physid_isset(i, phys_id_present_map))
2146 if (i >= get_physical_broadcast())
2147 panic("Max APIC ID exceeded!\n");
2148 printk(KERN_ERR "... fixing up to %d. (tell your hw vendor)\n",
2150 physid_set(i, phys_id_present_map);
2151 mp_ioapics[apic].mp_apicid = i;
2154 tmp = apicid_to_cpu_present(mp_ioapics[apic].mp_apicid);
2155 apic_printk(APIC_VERBOSE, "Setting %d in the "
2156 "phys_id_present_map\n",
2157 mp_ioapics[apic].mp_apicid);
2158 physids_or(phys_id_present_map, phys_id_present_map, tmp);
2163 * We need to adjust the IRQ routing table
2164 * if the ID changed.
2166 if (old_id != mp_ioapics[apic].mp_apicid)
2167 for (i = 0; i < mp_irq_entries; i++)
2168 if (mp_irqs[i].mp_dstapic == old_id)
2169 mp_irqs[i].mp_dstapic
2170 = mp_ioapics[apic].mp_apicid;
2173 * Read the right value from the MPC table and
2174 * write it into the ID register.
2176 apic_printk(APIC_VERBOSE, KERN_INFO
2177 "...changing IO-APIC physical APIC ID to %d ...",
2178 mp_ioapics[apic].mp_apicid);
2180 reg_00.bits.ID = mp_ioapics[apic].mp_apicid;
2181 spin_lock_irqsave(&ioapic_lock, flags);
2182 io_apic_write(apic, 0, reg_00.raw);
2183 spin_unlock_irqrestore(&ioapic_lock, flags);
2188 spin_lock_irqsave(&ioapic_lock, flags);
2189 reg_00.raw = io_apic_read(apic, 0);
2190 spin_unlock_irqrestore(&ioapic_lock, flags);
2191 if (reg_00.bits.ID != mp_ioapics[apic].mp_apicid)
2192 printk("could not set ID!\n");
2194 apic_printk(APIC_VERBOSE, " ok.\n");
2199 int no_timer_check __initdata;
2201 static int __init notimercheck(char *s)
2206 __setup("no_timer_check", notimercheck);
2209 * There is a nasty bug in some older SMP boards, their mptable lies
2210 * about the timer IRQ. We do the following to work around the situation:
2212 * - timer IRQ defaults to IO-APIC IRQ
2213 * - if this function detects that timer IRQs are defunct, then we fall
2214 * back to ISA timer IRQs
2216 static int __init timer_irq_works(void)
2218 unsigned long t1 = jiffies;
2219 unsigned long flags;
2224 local_save_flags(flags);
2226 /* Let ten ticks pass... */
2227 mdelay((10 * 1000) / HZ);
2228 local_irq_restore(flags);
2231 * Expect a few ticks at least, to be sure some possible
2232 * glue logic does not lock up after one or two first
2233 * ticks in a non-ExtINT mode. Also the local APIC
2234 * might have cached one ExtINT interrupt. Finally, at
2235 * least one tick may be lost due to delays.
2239 if (time_after(jiffies, t1 + 4))
2245 * In the SMP+IOAPIC case it might happen that there are an unspecified
2246 * number of pending IRQ events unhandled. These cases are very rare,
2247 * so we 'resend' these IRQs via IPIs, to the same CPU. It's much
2248 * better to do it this way as thus we do not have to be aware of
2249 * 'pending' interrupts in the IRQ path, except at this point.
2252 * Edge triggered needs to resend any interrupt
2253 * that was delayed but this is now handled in the device
2258 * Starting up a edge-triggered IO-APIC interrupt is
2259 * nasty - we need to make sure that we get the edge.
2260 * If it is already asserted for some reason, we need
2261 * return 1 to indicate that is was pending.
2263 * This is not complete - we should be able to fake
2264 * an edge even if it isn't on the 8259A...
2267 static unsigned int startup_ioapic_irq(unsigned int irq)
2269 int was_pending = 0;
2270 unsigned long flags;
2271 struct irq_cfg *cfg;
2273 spin_lock_irqsave(&ioapic_lock, flags);
2274 if (irq < NR_IRQS_LEGACY) {
2275 disable_8259A_irq(irq);
2276 if (i8259A_irq_pending(irq))
2280 __unmask_IO_APIC_irq(cfg);
2281 spin_unlock_irqrestore(&ioapic_lock, flags);
2286 #ifdef CONFIG_X86_64
2287 static int ioapic_retrigger_irq(unsigned int irq)
2290 struct irq_cfg *cfg = irq_cfg(irq);
2291 unsigned long flags;
2293 spin_lock_irqsave(&vector_lock, flags);
2294 send_IPI_mask(cpumask_of(cpumask_first(cfg->domain)), cfg->vector);
2295 spin_unlock_irqrestore(&vector_lock, flags);
2300 static int ioapic_retrigger_irq(unsigned int irq)
2302 send_IPI_self(irq_cfg(irq)->vector);
2309 * Level and edge triggered IO-APIC interrupts need different handling,
2310 * so we use two separate IRQ descriptors. Edge triggered IRQs can be
2311 * handled with the level-triggered descriptor, but that one has slightly
2312 * more overhead. Level-triggered interrupts cannot be handled with the
2313 * edge-triggered handler, without risking IRQ storms and other ugly
2319 #ifdef CONFIG_INTR_REMAP
2320 static void ir_irq_migration(struct work_struct *work);
2322 static DECLARE_DELAYED_WORK(ir_migration_work, ir_irq_migration);
2325 * Migrate the IO-APIC irq in the presence of intr-remapping.
2327 * For edge triggered, irq migration is a simple atomic update(of vector
2328 * and cpu destination) of IRTE and flush the hardware cache.
2330 * For level triggered, we need to modify the io-apic RTE aswell with the update
2331 * vector information, along with modifying IRTE with vector and destination.
2332 * So irq migration for level triggered is little bit more complex compared to
2333 * edge triggered migration. But the good news is, we use the same algorithm
2334 * for level triggered migration as we have today, only difference being,
2335 * we now initiate the irq migration from process context instead of the
2336 * interrupt context.
2338 * In future, when we do a directed EOI (combined with cpu EOI broadcast
2339 * suppression) to the IO-APIC, level triggered irq migration will also be
2340 * as simple as edge triggered migration and we can do the irq migration
2341 * with a simple atomic update to IO-APIC RTE.
2344 migrate_ioapic_irq_desc(struct irq_desc *desc, const struct cpumask *mask)
2346 struct irq_cfg *cfg;
2348 int modify_ioapic_rte;
2350 unsigned long flags;
2353 if (!cpumask_intersects(mask, cpu_online_mask))
2357 if (get_irte(irq, &irte))
2360 cfg = desc->chip_data;
2361 if (assign_irq_vector(irq, cfg, mask))
2364 set_extra_move_desc(desc, mask);
2366 dest = cpu_mask_to_apicid_and(cfg->domain, mask);
2368 modify_ioapic_rte = desc->status & IRQ_LEVEL;
2369 if (modify_ioapic_rte) {
2370 spin_lock_irqsave(&ioapic_lock, flags);
2371 __target_IO_APIC_irq(irq, dest, cfg);
2372 spin_unlock_irqrestore(&ioapic_lock, flags);
2375 irte.vector = cfg->vector;
2376 irte.dest_id = IRTE_DEST(dest);
2379 * Modified the IRTE and flushes the Interrupt entry cache.
2381 modify_irte(irq, &irte);
2383 if (cfg->move_in_progress)
2384 send_cleanup_vector(cfg);
2386 cpumask_copy(&desc->affinity, mask);
2389 static int migrate_irq_remapped_level_desc(struct irq_desc *desc)
2392 struct irq_cfg *cfg = desc->chip_data;
2394 mask_IO_APIC_irq_desc(desc);
2396 if (io_apic_level_ack_pending(cfg)) {
2398 * Interrupt in progress. Migrating irq now will change the
2399 * vector information in the IO-APIC RTE and that will confuse
2400 * the EOI broadcast performed by cpu.
2401 * So, delay the irq migration to the next instance.
2403 schedule_delayed_work(&ir_migration_work, 1);
2407 /* everthing is clear. we have right of way */
2408 migrate_ioapic_irq_desc(desc, &desc->pending_mask);
2411 desc->status &= ~IRQ_MOVE_PENDING;
2412 cpumask_clear(&desc->pending_mask);
2415 unmask_IO_APIC_irq_desc(desc);
2420 static void ir_irq_migration(struct work_struct *work)
2423 struct irq_desc *desc;
2425 for_each_irq_desc(irq, desc) {
2426 if (desc->status & IRQ_MOVE_PENDING) {
2427 unsigned long flags;
2429 spin_lock_irqsave(&desc->lock, flags);
2430 if (!desc->chip->set_affinity ||
2431 !(desc->status & IRQ_MOVE_PENDING)) {
2432 desc->status &= ~IRQ_MOVE_PENDING;
2433 spin_unlock_irqrestore(&desc->lock, flags);
2437 desc->chip->set_affinity(irq, &desc->pending_mask);
2438 spin_unlock_irqrestore(&desc->lock, flags);
2444 * Migrates the IRQ destination in the process context.
2446 static void set_ir_ioapic_affinity_irq_desc(struct irq_desc *desc,
2447 const struct cpumask *mask)
2449 if (desc->status & IRQ_LEVEL) {
2450 desc->status |= IRQ_MOVE_PENDING;
2451 cpumask_copy(&desc->pending_mask, mask);
2452 migrate_irq_remapped_level_desc(desc);
2456 migrate_ioapic_irq_desc(desc, mask);
2458 static void set_ir_ioapic_affinity_irq(unsigned int irq,
2459 const struct cpumask *mask)
2461 struct irq_desc *desc = irq_to_desc(irq);
2463 set_ir_ioapic_affinity_irq_desc(desc, mask);
2467 asmlinkage void smp_irq_move_cleanup_interrupt(void)
2469 unsigned vector, me;
2475 me = smp_processor_id();
2476 for (vector = FIRST_EXTERNAL_VECTOR; vector < NR_VECTORS; vector++) {
2478 struct irq_desc *desc;
2479 struct irq_cfg *cfg;
2480 irq = __get_cpu_var(vector_irq)[vector];
2485 desc = irq_to_desc(irq);
2490 spin_lock(&desc->lock);
2491 if (!cfg->move_cleanup_count)
2494 if (vector == cfg->vector && cpumask_test_cpu(me, cfg->domain))
2497 __get_cpu_var(vector_irq)[vector] = -1;
2498 cfg->move_cleanup_count--;
2500 spin_unlock(&desc->lock);
2506 static void irq_complete_move(struct irq_desc **descp)
2508 struct irq_desc *desc = *descp;
2509 struct irq_cfg *cfg = desc->chip_data;
2510 unsigned vector, me;
2512 if (likely(!cfg->move_in_progress)) {
2513 #ifdef CONFIG_NUMA_MIGRATE_IRQ_DESC
2514 if (likely(!cfg->move_desc_pending))
2517 /* domain has not changed, but affinity did */
2518 me = smp_processor_id();
2519 if (cpu_isset(me, desc->affinity)) {
2520 *descp = desc = move_irq_desc(desc, me);
2521 /* get the new one */
2522 cfg = desc->chip_data;
2523 cfg->move_desc_pending = 0;
2529 vector = ~get_irq_regs()->orig_ax;
2530 me = smp_processor_id();
2532 if (vector == cfg->vector && cpumask_test_cpu(me, cfg->domain)) {
2533 #ifdef CONFIG_NUMA_MIGRATE_IRQ_DESC
2534 *descp = desc = move_irq_desc(desc, me);
2535 /* get the new one */
2536 cfg = desc->chip_data;
2538 send_cleanup_vector(cfg);
2542 static inline void irq_complete_move(struct irq_desc **descp) {}
2545 #ifdef CONFIG_INTR_REMAP
2546 static void ack_x2apic_level(unsigned int irq)
2551 static void ack_x2apic_edge(unsigned int irq)
2558 static void ack_apic_edge(unsigned int irq)
2560 struct irq_desc *desc = irq_to_desc(irq);
2562 irq_complete_move(&desc);
2563 move_native_irq(irq);
2567 atomic_t irq_mis_count;
2569 static void ack_apic_level(unsigned int irq)
2571 struct irq_desc *desc = irq_to_desc(irq);
2573 #ifdef CONFIG_X86_32
2577 struct irq_cfg *cfg;
2578 int do_unmask_irq = 0;
2580 irq_complete_move(&desc);
2581 #ifdef CONFIG_GENERIC_PENDING_IRQ
2582 /* If we are moving the irq we need to mask it */
2583 if (unlikely(desc->status & IRQ_MOVE_PENDING)) {
2585 mask_IO_APIC_irq_desc(desc);
2589 #ifdef CONFIG_X86_32
2591 * It appears there is an erratum which affects at least version 0x11
2592 * of I/O APIC (that's the 82093AA and cores integrated into various
2593 * chipsets). Under certain conditions a level-triggered interrupt is
2594 * erroneously delivered as edge-triggered one but the respective IRR
2595 * bit gets set nevertheless. As a result the I/O unit expects an EOI
2596 * message but it will never arrive and further interrupts are blocked
2597 * from the source. The exact reason is so far unknown, but the
2598 * phenomenon was observed when two consecutive interrupt requests
2599 * from a given source get delivered to the same CPU and the source is
2600 * temporarily disabled in between.
2602 * A workaround is to simulate an EOI message manually. We achieve it
2603 * by setting the trigger mode to edge and then to level when the edge
2604 * trigger mode gets detected in the TMR of a local APIC for a
2605 * level-triggered interrupt. We mask the source for the time of the
2606 * operation to prevent an edge-triggered interrupt escaping meanwhile.
2607 * The idea is from Manfred Spraul. --macro
2609 cfg = desc->chip_data;
2612 v = apic_read(APIC_TMR + ((i & ~0x1f) >> 1));
2616 * We must acknowledge the irq before we move it or the acknowledge will
2617 * not propagate properly.
2621 /* Now we can move and renable the irq */
2622 if (unlikely(do_unmask_irq)) {
2623 /* Only migrate the irq if the ack has been received.
2625 * On rare occasions the broadcast level triggered ack gets
2626 * delayed going to ioapics, and if we reprogram the
2627 * vector while Remote IRR is still set the irq will never
2630 * To prevent this scenario we read the Remote IRR bit
2631 * of the ioapic. This has two effects.
2632 * - On any sane system the read of the ioapic will
2633 * flush writes (and acks) going to the ioapic from
2635 * - We get to see if the ACK has actually been delivered.
2637 * Based on failed experiments of reprogramming the
2638 * ioapic entry from outside of irq context starting
2639 * with masking the ioapic entry and then polling until
2640 * Remote IRR was clear before reprogramming the
2641 * ioapic I don't trust the Remote IRR bit to be
2642 * completey accurate.
2644 * However there appears to be no other way to plug
2645 * this race, so if the Remote IRR bit is not
2646 * accurate and is causing problems then it is a hardware bug
2647 * and you can go talk to the chipset vendor about it.
2649 cfg = desc->chip_data;
2650 if (!io_apic_level_ack_pending(cfg))
2651 move_masked_irq(irq);
2652 unmask_IO_APIC_irq_desc(desc);
2655 #ifdef CONFIG_X86_32
2656 if (!(v & (1 << (i & 0x1f)))) {
2657 atomic_inc(&irq_mis_count);
2658 spin_lock(&ioapic_lock);
2659 __mask_and_edge_IO_APIC_irq(cfg);
2660 __unmask_and_level_IO_APIC_irq(cfg);
2661 spin_unlock(&ioapic_lock);
2666 static struct irq_chip ioapic_chip __read_mostly = {
2668 .startup = startup_ioapic_irq,
2669 .mask = mask_IO_APIC_irq,
2670 .unmask = unmask_IO_APIC_irq,
2671 .ack = ack_apic_edge,
2672 .eoi = ack_apic_level,
2674 .set_affinity = set_ioapic_affinity_irq,
2676 .retrigger = ioapic_retrigger_irq,
2679 #ifdef CONFIG_INTR_REMAP
2680 static struct irq_chip ir_ioapic_chip __read_mostly = {
2681 .name = "IR-IO-APIC",
2682 .startup = startup_ioapic_irq,
2683 .mask = mask_IO_APIC_irq,
2684 .unmask = unmask_IO_APIC_irq,
2685 .ack = ack_x2apic_edge,
2686 .eoi = ack_x2apic_level,
2688 .set_affinity = set_ir_ioapic_affinity_irq,
2690 .retrigger = ioapic_retrigger_irq,
2694 static inline void init_IO_APIC_traps(void)
2697 struct irq_desc *desc;
2698 struct irq_cfg *cfg;
2701 * NOTE! The local APIC isn't very good at handling
2702 * multiple interrupts at the same interrupt level.
2703 * As the interrupt level is determined by taking the
2704 * vector number and shifting that right by 4, we
2705 * want to spread these out a bit so that they don't
2706 * all fall in the same interrupt level.
2708 * Also, we've got to be careful not to trash gate
2709 * 0x80, because int 0x80 is hm, kind of importantish. ;)
2711 for_each_irq_desc(irq, desc) {
2712 cfg = desc->chip_data;
2713 if (IO_APIC_IRQ(irq) && cfg && !cfg->vector) {
2715 * Hmm.. We don't have an entry for this,
2716 * so default to an old-fashioned 8259
2717 * interrupt if we can..
2719 if (irq < NR_IRQS_LEGACY)
2720 make_8259A_irq(irq);
2722 /* Strange. Oh, well.. */
2723 desc->chip = &no_irq_chip;
2729 * The local APIC irq-chip implementation:
2732 static void mask_lapic_irq(unsigned int irq)
2736 v = apic_read(APIC_LVT0);
2737 apic_write(APIC_LVT0, v | APIC_LVT_MASKED);
2740 static void unmask_lapic_irq(unsigned int irq)
2744 v = apic_read(APIC_LVT0);
2745 apic_write(APIC_LVT0, v & ~APIC_LVT_MASKED);
2748 static void ack_lapic_irq(unsigned int irq)
2753 static struct irq_chip lapic_chip __read_mostly = {
2754 .name = "local-APIC",
2755 .mask = mask_lapic_irq,
2756 .unmask = unmask_lapic_irq,
2757 .ack = ack_lapic_irq,
2760 static void lapic_register_intr(int irq, struct irq_desc *desc)
2762 desc->status &= ~IRQ_LEVEL;
2763 set_irq_chip_and_handler_name(irq, &lapic_chip, handle_edge_irq,
2767 static void __init setup_nmi(void)
2770 * Dirty trick to enable the NMI watchdog ...
2771 * We put the 8259A master into AEOI mode and
2772 * unmask on all local APICs LVT0 as NMI.
2774 * The idea to use the 8259A in AEOI mode ('8259A Virtual Wire')
2775 * is from Maciej W. Rozycki - so we do not have to EOI from
2776 * the NMI handler or the timer interrupt.
2778 apic_printk(APIC_VERBOSE, KERN_INFO "activating NMI Watchdog ...");
2780 enable_NMI_through_LVT0();
2782 apic_printk(APIC_VERBOSE, " done.\n");
2786 * This looks a bit hackish but it's about the only one way of sending
2787 * a few INTA cycles to 8259As and any associated glue logic. ICR does
2788 * not support the ExtINT mode, unfortunately. We need to send these
2789 * cycles as some i82489DX-based boards have glue logic that keeps the
2790 * 8259A interrupt line asserted until INTA. --macro
2792 static inline void __init unlock_ExtINT_logic(void)
2795 struct IO_APIC_route_entry entry0, entry1;
2796 unsigned char save_control, save_freq_select;
2798 pin = find_isa_irq_pin(8, mp_INT);
2803 apic = find_isa_irq_apic(8, mp_INT);
2809 entry0 = ioapic_read_entry(apic, pin);
2810 clear_IO_APIC_pin(apic, pin);
2812 memset(&entry1, 0, sizeof(entry1));
2814 entry1.dest_mode = 0; /* physical delivery */
2815 entry1.mask = 0; /* unmask IRQ now */
2816 entry1.dest = hard_smp_processor_id();
2817 entry1.delivery_mode = dest_ExtINT;
2818 entry1.polarity = entry0.polarity;
2822 ioapic_write_entry(apic, pin, entry1);
2824 save_control = CMOS_READ(RTC_CONTROL);
2825 save_freq_select = CMOS_READ(RTC_FREQ_SELECT);
2826 CMOS_WRITE((save_freq_select & ~RTC_RATE_SELECT) | 0x6,
2828 CMOS_WRITE(save_control | RTC_PIE, RTC_CONTROL);
2833 if ((CMOS_READ(RTC_INTR_FLAGS) & RTC_PF) == RTC_PF)
2837 CMOS_WRITE(save_control, RTC_CONTROL);
2838 CMOS_WRITE(save_freq_select, RTC_FREQ_SELECT);
2839 clear_IO_APIC_pin(apic, pin);
2841 ioapic_write_entry(apic, pin, entry0);
2844 static int disable_timer_pin_1 __initdata;
2845 /* Actually the next is obsolete, but keep it for paranoid reasons -AK */
2846 static int __init disable_timer_pin_setup(char *arg)
2848 disable_timer_pin_1 = 1;
2851 early_param("disable_timer_pin_1", disable_timer_pin_setup);
2853 int timer_through_8259 __initdata;
2856 * This code may look a bit paranoid, but it's supposed to cooperate with
2857 * a wide range of boards and BIOS bugs. Fortunately only the timer IRQ
2858 * is so screwy. Thanks to Brian Perkins for testing/hacking this beast
2859 * fanatically on his truly buggy board.
2861 * FIXME: really need to revamp this for all platforms.
2863 static inline void __init check_timer(void)
2865 struct irq_desc *desc = irq_to_desc(0);
2866 struct irq_cfg *cfg = desc->chip_data;
2867 int cpu = boot_cpu_id;
2868 int apic1, pin1, apic2, pin2;
2869 unsigned long flags;
2873 local_irq_save(flags);
2875 ver = apic_read(APIC_LVR);
2876 ver = GET_APIC_VERSION(ver);
2879 * get/set the timer IRQ vector:
2881 disable_8259A_irq(0);
2882 assign_irq_vector(0, cfg, TARGET_CPUS);
2885 * As IRQ0 is to be enabled in the 8259A, the virtual
2886 * wire has to be disabled in the local APIC. Also
2887 * timer interrupts need to be acknowledged manually in
2888 * the 8259A for the i82489DX when using the NMI
2889 * watchdog as that APIC treats NMIs as level-triggered.
2890 * The AEOI mode will finish them in the 8259A
2893 apic_write(APIC_LVT0, APIC_LVT_MASKED | APIC_DM_EXTINT);
2895 #ifdef CONFIG_X86_32
2896 timer_ack = (nmi_watchdog == NMI_IO_APIC && !APIC_INTEGRATED(ver));
2899 pin1 = find_isa_irq_pin(0, mp_INT);
2900 apic1 = find_isa_irq_apic(0, mp_INT);
2901 pin2 = ioapic_i8259.pin;
2902 apic2 = ioapic_i8259.apic;
2904 apic_printk(APIC_QUIET, KERN_INFO "..TIMER: vector=0x%02X "
2905 "apic1=%d pin1=%d apic2=%d pin2=%d\n",
2906 cfg->vector, apic1, pin1, apic2, pin2);
2909 * Some BIOS writers are clueless and report the ExtINTA
2910 * I/O APIC input from the cascaded 8259A as the timer
2911 * interrupt input. So just in case, if only one pin
2912 * was found above, try it both directly and through the
2916 #ifdef CONFIG_INTR_REMAP
2917 if (intr_remapping_enabled)
2918 panic("BIOS bug: timer not connected to IO-APIC");
2923 } else if (pin2 == -1) {
2930 * Ok, does IRQ0 through the IOAPIC work?
2933 add_pin_to_irq_cpu(cfg, cpu, apic1, pin1);
2934 setup_timer_IRQ0_pin(apic1, pin1, cfg->vector);
2936 unmask_IO_APIC_irq_desc(desc);
2937 if (timer_irq_works()) {
2938 if (nmi_watchdog == NMI_IO_APIC) {
2940 enable_8259A_irq(0);
2942 if (disable_timer_pin_1 > 0)
2943 clear_IO_APIC_pin(0, pin1);
2946 #ifdef CONFIG_INTR_REMAP
2947 if (intr_remapping_enabled)
2948 panic("timer doesn't work through Interrupt-remapped IO-APIC");
2950 clear_IO_APIC_pin(apic1, pin1);
2952 apic_printk(APIC_QUIET, KERN_ERR "..MP-BIOS bug: "
2953 "8254 timer not connected to IO-APIC\n");
2955 apic_printk(APIC_QUIET, KERN_INFO "...trying to set up timer "
2956 "(IRQ0) through the 8259A ...\n");
2957 apic_printk(APIC_QUIET, KERN_INFO
2958 "..... (found apic %d pin %d) ...\n", apic2, pin2);
2960 * legacy devices should be connected to IO APIC #0
2962 replace_pin_at_irq_cpu(cfg, cpu, apic1, pin1, apic2, pin2);
2963 setup_timer_IRQ0_pin(apic2, pin2, cfg->vector);
2964 unmask_IO_APIC_irq_desc(desc);
2965 enable_8259A_irq(0);
2966 if (timer_irq_works()) {
2967 apic_printk(APIC_QUIET, KERN_INFO "....... works.\n");
2968 timer_through_8259 = 1;
2969 if (nmi_watchdog == NMI_IO_APIC) {
2970 disable_8259A_irq(0);
2972 enable_8259A_irq(0);
2977 * Cleanup, just in case ...
2979 disable_8259A_irq(0);
2980 clear_IO_APIC_pin(apic2, pin2);
2981 apic_printk(APIC_QUIET, KERN_INFO "....... failed.\n");
2984 if (nmi_watchdog == NMI_IO_APIC) {
2985 apic_printk(APIC_QUIET, KERN_WARNING "timer doesn't work "
2986 "through the IO-APIC - disabling NMI Watchdog!\n");
2987 nmi_watchdog = NMI_NONE;
2989 #ifdef CONFIG_X86_32
2993 apic_printk(APIC_QUIET, KERN_INFO
2994 "...trying to set up timer as Virtual Wire IRQ...\n");
2996 lapic_register_intr(0, desc);
2997 apic_write(APIC_LVT0, APIC_DM_FIXED | cfg->vector); /* Fixed mode */
2998 enable_8259A_irq(0);
3000 if (timer_irq_works()) {
3001 apic_printk(APIC_QUIET, KERN_INFO "..... works.\n");
3004 disable_8259A_irq(0);
3005 apic_write(APIC_LVT0, APIC_LVT_MASKED | APIC_DM_FIXED | cfg->vector);
3006 apic_printk(APIC_QUIET, KERN_INFO "..... failed.\n");
3008 apic_printk(APIC_QUIET, KERN_INFO
3009 "...trying to set up timer as ExtINT IRQ...\n");
3013 apic_write(APIC_LVT0, APIC_DM_EXTINT);
3015 unlock_ExtINT_logic();
3017 if (timer_irq_works()) {
3018 apic_printk(APIC_QUIET, KERN_INFO "..... works.\n");
3021 apic_printk(APIC_QUIET, KERN_INFO "..... failed :(.\n");
3022 panic("IO-APIC + timer doesn't work! Boot with apic=debug and send a "
3023 "report. Then try booting with the 'noapic' option.\n");
3025 local_irq_restore(flags);
3029 * Traditionally ISA IRQ2 is the cascade IRQ, and is not available
3030 * to devices. However there may be an I/O APIC pin available for
3031 * this interrupt regardless. The pin may be left unconnected, but
3032 * typically it will be reused as an ExtINT cascade interrupt for
3033 * the master 8259A. In the MPS case such a pin will normally be
3034 * reported as an ExtINT interrupt in the MP table. With ACPI
3035 * there is no provision for ExtINT interrupts, and in the absence
3036 * of an override it would be treated as an ordinary ISA I/O APIC
3037 * interrupt, that is edge-triggered and unmasked by default. We
3038 * used to do this, but it caused problems on some systems because
3039 * of the NMI watchdog and sometimes IRQ0 of the 8254 timer using
3040 * the same ExtINT cascade interrupt to drive the local APIC of the
3041 * bootstrap processor. Therefore we refrain from routing IRQ2 to
3042 * the I/O APIC in all cases now. No actual device should request
3043 * it anyway. --macro
3045 #define PIC_IRQS (1 << PIC_CASCADE_IR)
3047 void __init setup_IO_APIC(void)
3050 #ifdef CONFIG_X86_32
3054 * calling enable_IO_APIC() is moved to setup_local_APIC for BP
3058 io_apic_irqs = ~PIC_IRQS;
3060 apic_printk(APIC_VERBOSE, "ENABLING IO-APIC IRQs\n");
3062 * Set up IO-APIC IRQ routing.
3064 #ifdef CONFIG_X86_32
3066 setup_ioapic_ids_from_mpc();
3069 setup_IO_APIC_irqs();
3070 init_IO_APIC_traps();
3075 * Called after all the initialization is done. If we didnt find any
3076 * APIC bugs then we can allow the modify fast path
3079 static int __init io_apic_bug_finalize(void)
3081 if (sis_apic_bug == -1)
3086 late_initcall(io_apic_bug_finalize);
3088 struct sysfs_ioapic_data {
3089 struct sys_device dev;
3090 struct IO_APIC_route_entry entry[0];
3092 static struct sysfs_ioapic_data * mp_ioapic_data[MAX_IO_APICS];
3094 static int ioapic_suspend(struct sys_device *dev, pm_message_t state)
3096 struct IO_APIC_route_entry *entry;
3097 struct sysfs_ioapic_data *data;
3100 data = container_of(dev, struct sysfs_ioapic_data, dev);
3101 entry = data->entry;
3102 for (i = 0; i < nr_ioapic_registers[dev->id]; i ++, entry ++ )
3103 *entry = ioapic_read_entry(dev->id, i);
3108 static int ioapic_resume(struct sys_device *dev)
3110 struct IO_APIC_route_entry *entry;
3111 struct sysfs_ioapic_data *data;
3112 unsigned long flags;
3113 union IO_APIC_reg_00 reg_00;
3116 data = container_of(dev, struct sysfs_ioapic_data, dev);
3117 entry = data->entry;
3119 spin_lock_irqsave(&ioapic_lock, flags);
3120 reg_00.raw = io_apic_read(dev->id, 0);
3121 if (reg_00.bits.ID != mp_ioapics[dev->id].mp_apicid) {
3122 reg_00.bits.ID = mp_ioapics[dev->id].mp_apicid;
3123 io_apic_write(dev->id, 0, reg_00.raw);
3125 spin_unlock_irqrestore(&ioapic_lock, flags);
3126 for (i = 0; i < nr_ioapic_registers[dev->id]; i++)
3127 ioapic_write_entry(dev->id, i, entry[i]);
3132 static struct sysdev_class ioapic_sysdev_class = {
3134 .suspend = ioapic_suspend,
3135 .resume = ioapic_resume,
3138 static int __init ioapic_init_sysfs(void)
3140 struct sys_device * dev;
3143 error = sysdev_class_register(&ioapic_sysdev_class);
3147 for (i = 0; i < nr_ioapics; i++ ) {
3148 size = sizeof(struct sys_device) + nr_ioapic_registers[i]
3149 * sizeof(struct IO_APIC_route_entry);
3150 mp_ioapic_data[i] = kzalloc(size, GFP_KERNEL);
3151 if (!mp_ioapic_data[i]) {
3152 printk(KERN_ERR "Can't suspend/resume IOAPIC %d\n", i);
3155 dev = &mp_ioapic_data[i]->dev;
3157 dev->cls = &ioapic_sysdev_class;
3158 error = sysdev_register(dev);
3160 kfree(mp_ioapic_data[i]);
3161 mp_ioapic_data[i] = NULL;
3162 printk(KERN_ERR "Can't suspend/resume IOAPIC %d\n", i);
3170 device_initcall(ioapic_init_sysfs);
3173 * Dynamic irq allocate and deallocation
3175 unsigned int create_irq_nr(unsigned int irq_want)
3177 /* Allocate an unused irq */
3180 unsigned long flags;
3181 struct irq_cfg *cfg_new = NULL;
3182 int cpu = boot_cpu_id;
3183 struct irq_desc *desc_new = NULL;
3186 spin_lock_irqsave(&vector_lock, flags);
3187 for (new = irq_want; new < NR_IRQS; new++) {
3188 if (platform_legacy_irq(new))
3191 desc_new = irq_to_desc_alloc_cpu(new, cpu);
3193 printk(KERN_INFO "can not get irq_desc for %d\n", new);
3196 cfg_new = desc_new->chip_data;
3198 if (cfg_new->vector != 0)
3200 if (__assign_irq_vector(new, cfg_new, TARGET_CPUS) == 0)
3204 spin_unlock_irqrestore(&vector_lock, flags);
3207 dynamic_irq_init(irq);
3208 /* restore it, in case dynamic_irq_init clear it */
3210 desc_new->chip_data = cfg_new;
3215 static int nr_irqs_gsi = NR_IRQS_LEGACY;
3216 int create_irq(void)
3218 unsigned int irq_want;
3221 irq_want = nr_irqs_gsi;
3222 irq = create_irq_nr(irq_want);
3230 void destroy_irq(unsigned int irq)
3232 unsigned long flags;
3233 struct irq_cfg *cfg;
3234 struct irq_desc *desc;
3236 /* store it, in case dynamic_irq_cleanup clear it */
3237 desc = irq_to_desc(irq);
3238 cfg = desc->chip_data;
3239 dynamic_irq_cleanup(irq);
3240 /* connect back irq_cfg */
3242 desc->chip_data = cfg;
3244 #ifdef CONFIG_INTR_REMAP
3247 spin_lock_irqsave(&vector_lock, flags);
3248 __clear_irq_vector(irq, cfg);
3249 spin_unlock_irqrestore(&vector_lock, flags);
3253 * MSI message composition
3255 #ifdef CONFIG_PCI_MSI
3256 static int msi_compose_msg(struct pci_dev *pdev, unsigned int irq, struct msi_msg *msg)
3258 struct irq_cfg *cfg;
3263 err = assign_irq_vector(irq, cfg, TARGET_CPUS);
3267 dest = cpu_mask_to_apicid_and(cfg->domain, TARGET_CPUS);
3269 #ifdef CONFIG_INTR_REMAP
3270 if (irq_remapped(irq)) {
3275 ir_index = map_irq_to_irte_handle(irq, &sub_handle);
3276 BUG_ON(ir_index == -1);
3278 memset (&irte, 0, sizeof(irte));
3281 irte.dst_mode = INT_DEST_MODE;
3282 irte.trigger_mode = 0; /* edge */
3283 irte.dlvry_mode = INT_DELIVERY_MODE;
3284 irte.vector = cfg->vector;
3285 irte.dest_id = IRTE_DEST(dest);
3287 modify_irte(irq, &irte);
3289 msg->address_hi = MSI_ADDR_BASE_HI;
3290 msg->data = sub_handle;
3291 msg->address_lo = MSI_ADDR_BASE_LO | MSI_ADDR_IR_EXT_INT |
3293 MSI_ADDR_IR_INDEX1(ir_index) |
3294 MSI_ADDR_IR_INDEX2(ir_index);
3298 msg->address_hi = MSI_ADDR_BASE_HI;
3301 ((INT_DEST_MODE == 0) ?
3302 MSI_ADDR_DEST_MODE_PHYSICAL:
3303 MSI_ADDR_DEST_MODE_LOGICAL) |
3304 ((INT_DELIVERY_MODE != dest_LowestPrio) ?
3305 MSI_ADDR_REDIRECTION_CPU:
3306 MSI_ADDR_REDIRECTION_LOWPRI) |
3307 MSI_ADDR_DEST_ID(dest);
3310 MSI_DATA_TRIGGER_EDGE |
3311 MSI_DATA_LEVEL_ASSERT |
3312 ((INT_DELIVERY_MODE != dest_LowestPrio) ?
3313 MSI_DATA_DELIVERY_FIXED:
3314 MSI_DATA_DELIVERY_LOWPRI) |
3315 MSI_DATA_VECTOR(cfg->vector);
3321 static void set_msi_irq_affinity(unsigned int irq, const struct cpumask *mask)
3323 struct irq_desc *desc = irq_to_desc(irq);
3324 struct irq_cfg *cfg;
3328 dest = set_desc_affinity(desc, mask);
3329 if (dest == BAD_APICID)
3332 cfg = desc->chip_data;
3334 read_msi_msg_desc(desc, &msg);
3336 msg.data &= ~MSI_DATA_VECTOR_MASK;
3337 msg.data |= MSI_DATA_VECTOR(cfg->vector);
3338 msg.address_lo &= ~MSI_ADDR_DEST_ID_MASK;
3339 msg.address_lo |= MSI_ADDR_DEST_ID(dest);
3341 write_msi_msg_desc(desc, &msg);
3343 #ifdef CONFIG_INTR_REMAP
3345 * Migrate the MSI irq to another cpumask. This migration is
3346 * done in the process context using interrupt-remapping hardware.
3349 ir_set_msi_irq_affinity(unsigned int irq, const struct cpumask *mask)
3351 struct irq_desc *desc = irq_to_desc(irq);
3352 struct irq_cfg *cfg = desc->chip_data;
3356 if (get_irte(irq, &irte))
3359 dest = set_desc_affinity(desc, mask);
3360 if (dest == BAD_APICID)
3363 irte.vector = cfg->vector;
3364 irte.dest_id = IRTE_DEST(dest);
3367 * atomically update the IRTE with the new destination and vector.
3369 modify_irte(irq, &irte);
3372 * After this point, all the interrupts will start arriving
3373 * at the new destination. So, time to cleanup the previous
3374 * vector allocation.
3376 if (cfg->move_in_progress)
3377 send_cleanup_vector(cfg);
3381 #endif /* CONFIG_SMP */
3384 * IRQ Chip for MSI PCI/PCI-X/PCI-Express Devices,
3385 * which implement the MSI or MSI-X Capability Structure.
3387 static struct irq_chip msi_chip = {
3389 .unmask = unmask_msi_irq,
3390 .mask = mask_msi_irq,
3391 .ack = ack_apic_edge,
3393 .set_affinity = set_msi_irq_affinity,
3395 .retrigger = ioapic_retrigger_irq,
3398 #ifdef CONFIG_INTR_REMAP
3399 static struct irq_chip msi_ir_chip = {
3400 .name = "IR-PCI-MSI",
3401 .unmask = unmask_msi_irq,
3402 .mask = mask_msi_irq,
3403 .ack = ack_x2apic_edge,
3405 .set_affinity = ir_set_msi_irq_affinity,
3407 .retrigger = ioapic_retrigger_irq,
3411 * Map the PCI dev to the corresponding remapping hardware unit
3412 * and allocate 'nvec' consecutive interrupt-remapping table entries
3415 static int msi_alloc_irte(struct pci_dev *dev, int irq, int nvec)
3417 struct intel_iommu *iommu;
3420 iommu = map_dev_to_ir(dev);
3423 "Unable to map PCI %s to iommu\n", pci_name(dev));
3427 index = alloc_irte(iommu, irq, nvec);
3430 "Unable to allocate %d IRTE for PCI %s\n", nvec,
3438 static int setup_msi_irq(struct pci_dev *dev, struct msi_desc *msidesc, int irq)
3443 ret = msi_compose_msg(dev, irq, &msg);
3447 set_irq_msi(irq, msidesc);
3448 write_msi_msg(irq, &msg);
3450 #ifdef CONFIG_INTR_REMAP
3451 if (irq_remapped(irq)) {
3452 struct irq_desc *desc = irq_to_desc(irq);
3454 * irq migration in process context
3456 desc->status |= IRQ_MOVE_PCNTXT;
3457 set_irq_chip_and_handler_name(irq, &msi_ir_chip, handle_edge_irq, "edge");
3460 set_irq_chip_and_handler_name(irq, &msi_chip, handle_edge_irq, "edge");
3462 dev_printk(KERN_DEBUG, &dev->dev, "irq %d for MSI/MSI-X\n", irq);
3467 int arch_setup_msi_irq(struct pci_dev *dev, struct msi_desc *msidesc)
3471 unsigned int irq_want;
3473 irq_want = nr_irqs_gsi;
3474 irq = create_irq_nr(irq_want);
3478 #ifdef CONFIG_INTR_REMAP
3479 if (!intr_remapping_enabled)
3482 ret = msi_alloc_irte(dev, irq, 1);
3487 ret = setup_msi_irq(dev, msidesc, irq);
3494 #ifdef CONFIG_INTR_REMAP
3501 int arch_setup_msi_irqs(struct pci_dev *dev, int nvec, int type)
3504 int ret, sub_handle;
3505 struct msi_desc *msidesc;
3506 unsigned int irq_want;
3508 #ifdef CONFIG_INTR_REMAP
3509 struct intel_iommu *iommu = 0;
3513 irq_want = nr_irqs_gsi;
3515 list_for_each_entry(msidesc, &dev->msi_list, list) {
3516 irq = create_irq_nr(irq_want);
3520 #ifdef CONFIG_INTR_REMAP
3521 if (!intr_remapping_enabled)
3526 * allocate the consecutive block of IRTE's
3529 index = msi_alloc_irte(dev, irq, nvec);
3535 iommu = map_dev_to_ir(dev);
3541 * setup the mapping between the irq and the IRTE
3542 * base index, the sub_handle pointing to the
3543 * appropriate interrupt remap table entry.
3545 set_irte_irq(irq, iommu, index, sub_handle);
3549 ret = setup_msi_irq(dev, msidesc, irq);
3561 void arch_teardown_msi_irq(unsigned int irq)
3568 static void dmar_msi_set_affinity(unsigned int irq, const struct cpumask *mask)
3570 struct irq_desc *desc = irq_to_desc(irq);
3571 struct irq_cfg *cfg;
3575 dest = set_desc_affinity(desc, mask);
3576 if (dest == BAD_APICID)
3579 cfg = desc->chip_data;
3581 dmar_msi_read(irq, &msg);
3583 msg.data &= ~MSI_DATA_VECTOR_MASK;
3584 msg.data |= MSI_DATA_VECTOR(cfg->vector);
3585 msg.address_lo &= ~MSI_ADDR_DEST_ID_MASK;
3586 msg.address_lo |= MSI_ADDR_DEST_ID(dest);
3588 dmar_msi_write(irq, &msg);
3591 #endif /* CONFIG_SMP */
3593 struct irq_chip dmar_msi_type = {
3595 .unmask = dmar_msi_unmask,
3596 .mask = dmar_msi_mask,
3597 .ack = ack_apic_edge,
3599 .set_affinity = dmar_msi_set_affinity,
3601 .retrigger = ioapic_retrigger_irq,
3604 int arch_setup_dmar_msi(unsigned int irq)
3609 ret = msi_compose_msg(NULL, irq, &msg);
3612 dmar_msi_write(irq, &msg);
3613 set_irq_chip_and_handler_name(irq, &dmar_msi_type, handle_edge_irq,
3619 #ifdef CONFIG_HPET_TIMER
3622 static void hpet_msi_set_affinity(unsigned int irq, const struct cpumask *mask)
3624 struct irq_desc *desc = irq_to_desc(irq);
3625 struct irq_cfg *cfg;
3629 dest = set_desc_affinity(desc, mask);
3630 if (dest == BAD_APICID)
3633 cfg = desc->chip_data;
3635 hpet_msi_read(irq, &msg);
3637 msg.data &= ~MSI_DATA_VECTOR_MASK;
3638 msg.data |= MSI_DATA_VECTOR(cfg->vector);
3639 msg.address_lo &= ~MSI_ADDR_DEST_ID_MASK;
3640 msg.address_lo |= MSI_ADDR_DEST_ID(dest);
3642 hpet_msi_write(irq, &msg);
3645 #endif /* CONFIG_SMP */
3647 struct irq_chip hpet_msi_type = {
3649 .unmask = hpet_msi_unmask,
3650 .mask = hpet_msi_mask,
3651 .ack = ack_apic_edge,
3653 .set_affinity = hpet_msi_set_affinity,
3655 .retrigger = ioapic_retrigger_irq,
3658 int arch_setup_hpet_msi(unsigned int irq)
3663 ret = msi_compose_msg(NULL, irq, &msg);
3667 hpet_msi_write(irq, &msg);
3668 set_irq_chip_and_handler_name(irq, &hpet_msi_type, handle_edge_irq,
3675 #endif /* CONFIG_PCI_MSI */
3677 * Hypertransport interrupt support
3679 #ifdef CONFIG_HT_IRQ
3683 static void target_ht_irq(unsigned int irq, unsigned int dest, u8 vector)
3685 struct ht_irq_msg msg;
3686 fetch_ht_irq_msg(irq, &msg);
3688 msg.address_lo &= ~(HT_IRQ_LOW_VECTOR_MASK | HT_IRQ_LOW_DEST_ID_MASK);
3689 msg.address_hi &= ~(HT_IRQ_HIGH_DEST_ID_MASK);
3691 msg.address_lo |= HT_IRQ_LOW_VECTOR(vector) | HT_IRQ_LOW_DEST_ID(dest);
3692 msg.address_hi |= HT_IRQ_HIGH_DEST_ID(dest);
3694 write_ht_irq_msg(irq, &msg);
3697 static void set_ht_irq_affinity(unsigned int irq, const struct cpumask *mask)
3699 struct irq_desc *desc = irq_to_desc(irq);
3700 struct irq_cfg *cfg;
3703 dest = set_desc_affinity(desc, mask);
3704 if (dest == BAD_APICID)
3707 cfg = desc->chip_data;
3709 target_ht_irq(irq, dest, cfg->vector);
3714 static struct irq_chip ht_irq_chip = {
3716 .mask = mask_ht_irq,
3717 .unmask = unmask_ht_irq,
3718 .ack = ack_apic_edge,
3720 .set_affinity = set_ht_irq_affinity,
3722 .retrigger = ioapic_retrigger_irq,
3725 int arch_setup_ht_irq(unsigned int irq, struct pci_dev *dev)
3727 struct irq_cfg *cfg;
3731 err = assign_irq_vector(irq, cfg, TARGET_CPUS);
3733 struct ht_irq_msg msg;
3736 dest = cpu_mask_to_apicid_and(cfg->domain, TARGET_CPUS);
3738 msg.address_hi = HT_IRQ_HIGH_DEST_ID(dest);
3742 HT_IRQ_LOW_DEST_ID(dest) |
3743 HT_IRQ_LOW_VECTOR(cfg->vector) |
3744 ((INT_DEST_MODE == 0) ?
3745 HT_IRQ_LOW_DM_PHYSICAL :
3746 HT_IRQ_LOW_DM_LOGICAL) |
3747 HT_IRQ_LOW_RQEOI_EDGE |
3748 ((INT_DELIVERY_MODE != dest_LowestPrio) ?
3749 HT_IRQ_LOW_MT_FIXED :
3750 HT_IRQ_LOW_MT_ARBITRATED) |
3751 HT_IRQ_LOW_IRQ_MASKED;
3753 write_ht_irq_msg(irq, &msg);
3755 set_irq_chip_and_handler_name(irq, &ht_irq_chip,
3756 handle_edge_irq, "edge");
3758 dev_printk(KERN_DEBUG, &dev->dev, "irq %d for HT\n", irq);
3762 #endif /* CONFIG_HT_IRQ */
3764 #ifdef CONFIG_X86_64
3766 * Re-target the irq to the specified CPU and enable the specified MMR located
3767 * on the specified blade to allow the sending of MSIs to the specified CPU.
3769 int arch_enable_uv_irq(char *irq_name, unsigned int irq, int cpu, int mmr_blade,
3770 unsigned long mmr_offset)
3772 const struct cpumask *eligible_cpu = cpumask_of(cpu);
3773 struct irq_cfg *cfg;
3775 unsigned long mmr_value;
3776 struct uv_IO_APIC_route_entry *entry;
3777 unsigned long flags;
3782 err = assign_irq_vector(irq, cfg, eligible_cpu);
3786 spin_lock_irqsave(&vector_lock, flags);
3787 set_irq_chip_and_handler_name(irq, &uv_irq_chip, handle_percpu_irq,
3789 spin_unlock_irqrestore(&vector_lock, flags);
3792 entry = (struct uv_IO_APIC_route_entry *)&mmr_value;
3793 BUG_ON(sizeof(struct uv_IO_APIC_route_entry) != sizeof(unsigned long));
3795 entry->vector = cfg->vector;
3796 entry->delivery_mode = INT_DELIVERY_MODE;
3797 entry->dest_mode = INT_DEST_MODE;
3798 entry->polarity = 0;
3801 entry->dest = cpu_mask_to_apicid(eligible_cpu);
3803 mmr_pnode = uv_blade_to_pnode(mmr_blade);
3804 uv_write_global_mmr64(mmr_pnode, mmr_offset, mmr_value);
3810 * Disable the specified MMR located on the specified blade so that MSIs are
3811 * longer allowed to be sent.
3813 void arch_disable_uv_irq(int mmr_blade, unsigned long mmr_offset)
3815 unsigned long mmr_value;
3816 struct uv_IO_APIC_route_entry *entry;
3820 entry = (struct uv_IO_APIC_route_entry *)&mmr_value;
3821 BUG_ON(sizeof(struct uv_IO_APIC_route_entry) != sizeof(unsigned long));
3825 mmr_pnode = uv_blade_to_pnode(mmr_blade);
3826 uv_write_global_mmr64(mmr_pnode, mmr_offset, mmr_value);
3828 #endif /* CONFIG_X86_64 */
3830 int __init io_apic_get_redir_entries (int ioapic)
3832 union IO_APIC_reg_01 reg_01;
3833 unsigned long flags;
3835 spin_lock_irqsave(&ioapic_lock, flags);
3836 reg_01.raw = io_apic_read(ioapic, 1);
3837 spin_unlock_irqrestore(&ioapic_lock, flags);
3839 return reg_01.bits.entries;
3842 void __init probe_nr_irqs_gsi(void)
3846 nr = acpi_probe_gsi();
3847 if (nr > nr_irqs_gsi) {
3850 /* for acpi=off or acpi is not compiled in */
3854 for (idx = 0; idx < nr_ioapics; idx++)
3855 nr += io_apic_get_redir_entries(idx) + 1;
3857 if (nr > nr_irqs_gsi)
3861 printk(KERN_DEBUG "nr_irqs_gsi: %d\n", nr_irqs_gsi);
3864 /* --------------------------------------------------------------------------
3865 ACPI-based IOAPIC Configuration
3866 -------------------------------------------------------------------------- */
3870 #ifdef CONFIG_X86_32
3871 int __init io_apic_get_unique_id(int ioapic, int apic_id)
3873 union IO_APIC_reg_00 reg_00;
3874 static physid_mask_t apic_id_map = PHYSID_MASK_NONE;
3876 unsigned long flags;
3880 * The P4 platform supports up to 256 APIC IDs on two separate APIC
3881 * buses (one for LAPICs, one for IOAPICs), where predecessors only
3882 * supports up to 16 on one shared APIC bus.
3884 * TBD: Expand LAPIC/IOAPIC support on P4-class systems to take full
3885 * advantage of new APIC bus architecture.
3888 if (physids_empty(apic_id_map))
3889 apic_id_map = ioapic_phys_id_map(phys_cpu_present_map);
3891 spin_lock_irqsave(&ioapic_lock, flags);
3892 reg_00.raw = io_apic_read(ioapic, 0);
3893 spin_unlock_irqrestore(&ioapic_lock, flags);
3895 if (apic_id >= get_physical_broadcast()) {
3896 printk(KERN_WARNING "IOAPIC[%d]: Invalid apic_id %d, trying "
3897 "%d\n", ioapic, apic_id, reg_00.bits.ID);
3898 apic_id = reg_00.bits.ID;
3902 * Every APIC in a system must have a unique ID or we get lots of nice
3903 * 'stuck on smp_invalidate_needed IPI wait' messages.
3905 if (check_apicid_used(apic_id_map, apic_id)) {
3907 for (i = 0; i < get_physical_broadcast(); i++) {
3908 if (!check_apicid_used(apic_id_map, i))
3912 if (i == get_physical_broadcast())
3913 panic("Max apic_id exceeded!\n");
3915 printk(KERN_WARNING "IOAPIC[%d]: apic_id %d already used, "
3916 "trying %d\n", ioapic, apic_id, i);
3921 tmp = apicid_to_cpu_present(apic_id);
3922 physids_or(apic_id_map, apic_id_map, tmp);
3924 if (reg_00.bits.ID != apic_id) {
3925 reg_00.bits.ID = apic_id;
3927 spin_lock_irqsave(&ioapic_lock, flags);
3928 io_apic_write(ioapic, 0, reg_00.raw);
3929 reg_00.raw = io_apic_read(ioapic, 0);
3930 spin_unlock_irqrestore(&ioapic_lock, flags);
3933 if (reg_00.bits.ID != apic_id) {
3934 printk("IOAPIC[%d]: Unable to change apic_id!\n", ioapic);
3939 apic_printk(APIC_VERBOSE, KERN_INFO
3940 "IOAPIC[%d]: Assigned apic_id %d\n", ioapic, apic_id);
3945 int __init io_apic_get_version(int ioapic)
3947 union IO_APIC_reg_01 reg_01;
3948 unsigned long flags;
3950 spin_lock_irqsave(&ioapic_lock, flags);
3951 reg_01.raw = io_apic_read(ioapic, 1);
3952 spin_unlock_irqrestore(&ioapic_lock, flags);
3954 return reg_01.bits.version;
3958 int io_apic_set_pci_routing (int ioapic, int pin, int irq, int triggering, int polarity)
3960 struct irq_desc *desc;
3961 struct irq_cfg *cfg;
3962 int cpu = boot_cpu_id;
3964 if (!IO_APIC_IRQ(irq)) {
3965 apic_printk(APIC_QUIET,KERN_ERR "IOAPIC[%d]: Invalid reference to IRQ 0\n",
3970 desc = irq_to_desc_alloc_cpu(irq, cpu);
3972 printk(KERN_INFO "can not get irq_desc %d\n", irq);
3977 * IRQs < 16 are already in the irq_2_pin[] map
3979 if (irq >= NR_IRQS_LEGACY) {
3980 cfg = desc->chip_data;
3981 add_pin_to_irq_cpu(cfg, cpu, ioapic, pin);
3984 setup_IO_APIC_irq(ioapic, pin, irq, desc, triggering, polarity);
3990 int acpi_get_override_irq(int bus_irq, int *trigger, int *polarity)
3994 if (skip_ioapic_setup)
3997 for (i = 0; i < mp_irq_entries; i++)
3998 if (mp_irqs[i].mp_irqtype == mp_INT &&
3999 mp_irqs[i].mp_srcbusirq == bus_irq)
4001 if (i >= mp_irq_entries)
4004 *trigger = irq_trigger(i);
4005 *polarity = irq_polarity(i);
4009 #endif /* CONFIG_ACPI */
4012 * This function currently is only a helper for the i386 smp boot process where
4013 * we need to reprogram the ioredtbls to cater for the cpus which have come online
4014 * so mask in all cases should simply be TARGET_CPUS
4017 void __init setup_ioapic_dest(void)
4019 int pin, ioapic, irq, irq_entry;
4020 struct irq_desc *desc;
4021 struct irq_cfg *cfg;
4022 const struct cpumask *mask;
4024 if (skip_ioapic_setup == 1)
4027 for (ioapic = 0; ioapic < nr_ioapics; ioapic++) {
4028 for (pin = 0; pin < nr_ioapic_registers[ioapic]; pin++) {
4029 irq_entry = find_irq_entry(ioapic, pin, mp_INT);
4030 if (irq_entry == -1)
4032 irq = pin_2_irq(irq_entry, ioapic, pin);
4034 /* setup_IO_APIC_irqs could fail to get vector for some device
4035 * when you have too many devices, because at that time only boot
4038 desc = irq_to_desc(irq);
4039 cfg = desc->chip_data;
4041 setup_IO_APIC_irq(ioapic, pin, irq, desc,
4042 irq_trigger(irq_entry),
4043 irq_polarity(irq_entry));
4049 * Honour affinities which have been set in early boot
4052 (IRQ_NO_BALANCING | IRQ_AFFINITY_SET))
4053 mask = &desc->affinity;
4057 #ifdef CONFIG_INTR_REMAP
4058 if (intr_remapping_enabled)
4059 set_ir_ioapic_affinity_irq_desc(desc, mask);
4062 set_ioapic_affinity_irq_desc(desc, mask);
4069 #define IOAPIC_RESOURCE_NAME_SIZE 11
4071 static struct resource *ioapic_resources;
4073 static struct resource * __init ioapic_setup_resources(void)
4076 struct resource *res;
4080 if (nr_ioapics <= 0)
4083 n = IOAPIC_RESOURCE_NAME_SIZE + sizeof(struct resource);
4086 mem = alloc_bootmem(n);
4090 mem += sizeof(struct resource) * nr_ioapics;
4092 for (i = 0; i < nr_ioapics; i++) {
4094 res[i].flags = IORESOURCE_MEM | IORESOURCE_BUSY;
4095 sprintf(mem, "IOAPIC %u", i);
4096 mem += IOAPIC_RESOURCE_NAME_SIZE;
4100 ioapic_resources = res;
4105 void __init ioapic_init_mappings(void)
4107 unsigned long ioapic_phys, idx = FIX_IO_APIC_BASE_0;
4108 struct resource *ioapic_res;
4111 ioapic_res = ioapic_setup_resources();
4112 for (i = 0; i < nr_ioapics; i++) {
4113 if (smp_found_config) {
4114 ioapic_phys = mp_ioapics[i].mp_apicaddr;
4115 #ifdef CONFIG_X86_32
4118 "WARNING: bogus zero IO-APIC "
4119 "address found in MPTABLE, "
4120 "disabling IO/APIC support!\n");
4121 smp_found_config = 0;
4122 skip_ioapic_setup = 1;
4123 goto fake_ioapic_page;
4127 #ifdef CONFIG_X86_32
4130 ioapic_phys = (unsigned long)
4131 alloc_bootmem_pages(PAGE_SIZE);
4132 ioapic_phys = __pa(ioapic_phys);
4134 set_fixmap_nocache(idx, ioapic_phys);
4135 apic_printk(APIC_VERBOSE,
4136 "mapped IOAPIC to %08lx (%08lx)\n",
4137 __fix_to_virt(idx), ioapic_phys);
4140 if (ioapic_res != NULL) {
4141 ioapic_res->start = ioapic_phys;
4142 ioapic_res->end = ioapic_phys + (4 * 1024) - 1;
4148 static int __init ioapic_insert_resources(void)
4151 struct resource *r = ioapic_resources;
4155 "IO APIC resources could be not be allocated.\n");
4159 for (i = 0; i < nr_ioapics; i++) {
4160 insert_resource(&iomem_resource, r);
4167 /* Insert the IO APIC resources after PCI initialization has occured to handle
4168 * IO APICS that are mapped in on a BAR in PCI space. */
4169 late_initcall(ioapic_insert_resources);