2 * Intel IO-APIC support for multi-Pentium hosts.
4 * Copyright (C) 1997, 1998, 1999, 2000, 2009 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>
51 #include <asm/proto.h>
54 #include <asm/timer.h>
55 #include <asm/i8259.h>
57 #include <asm/msidef.h>
58 #include <asm/hypertransport.h>
59 #include <asm/setup.h>
60 #include <asm/irq_remapping.h>
62 #include <asm/uv/uv_hub.h>
63 #include <asm/uv/uv_irq.h>
67 #define __apicdebuginit(type) static type __init
70 * Is the SiS APIC rmw bug present ?
71 * -1 = don't know, 0 = no, 1 = yes
73 int sis_apic_bug = -1;
75 static DEFINE_SPINLOCK(ioapic_lock);
76 static DEFINE_SPINLOCK(vector_lock);
79 * # of IRQ routing registers
81 int nr_ioapic_registers[MAX_IO_APICS];
83 /* I/O APIC entries */
84 struct mpc_ioapic mp_ioapics[MAX_IO_APICS];
87 /* MP IRQ source entries */
88 struct mpc_intsrc mp_irqs[MAX_IRQ_SOURCES];
90 /* # of MP IRQ source entries */
93 #if defined (CONFIG_MCA) || defined (CONFIG_EISA)
94 int mp_bus_id_to_type[MAX_MP_BUSSES];
97 DECLARE_BITMAP(mp_bus_not_pci, MAX_MP_BUSSES);
99 int skip_ioapic_setup;
101 void arch_disable_smp_support(void)
105 noioapicreroute = -1;
107 skip_ioapic_setup = 1;
110 static int __init parse_noapic(char *str)
112 /* disable IO-APIC */
113 arch_disable_smp_support();
116 early_param("noapic", parse_noapic);
121 * This is performance-critical, we want to do it O(1)
123 * the indexing order of this array favors 1:1 mappings
124 * between pins and IRQs.
127 struct irq_pin_list {
129 struct irq_pin_list *next;
132 static struct irq_pin_list *get_one_free_irq_2_pin(int cpu)
134 struct irq_pin_list *pin;
137 node = cpu_to_node(cpu);
139 pin = kzalloc_node(sizeof(*pin), GFP_ATOMIC, node);
145 struct irq_pin_list *irq_2_pin;
146 cpumask_var_t domain;
147 cpumask_var_t old_domain;
148 unsigned move_cleanup_count;
150 u8 move_in_progress : 1;
151 #ifdef CONFIG_NUMA_MIGRATE_IRQ_DESC
152 u8 move_desc_pending : 1;
156 /* irq_cfg is indexed by the sum of all RTEs in all I/O APICs. */
157 #ifdef CONFIG_SPARSE_IRQ
158 static struct irq_cfg irq_cfgx[] = {
160 static struct irq_cfg irq_cfgx[NR_IRQS] = {
162 [0] = { .vector = IRQ0_VECTOR, },
163 [1] = { .vector = IRQ1_VECTOR, },
164 [2] = { .vector = IRQ2_VECTOR, },
165 [3] = { .vector = IRQ3_VECTOR, },
166 [4] = { .vector = IRQ4_VECTOR, },
167 [5] = { .vector = IRQ5_VECTOR, },
168 [6] = { .vector = IRQ6_VECTOR, },
169 [7] = { .vector = IRQ7_VECTOR, },
170 [8] = { .vector = IRQ8_VECTOR, },
171 [9] = { .vector = IRQ9_VECTOR, },
172 [10] = { .vector = IRQ10_VECTOR, },
173 [11] = { .vector = IRQ11_VECTOR, },
174 [12] = { .vector = IRQ12_VECTOR, },
175 [13] = { .vector = IRQ13_VECTOR, },
176 [14] = { .vector = IRQ14_VECTOR, },
177 [15] = { .vector = IRQ15_VECTOR, },
180 int __init arch_early_irq_init(void)
183 struct irq_desc *desc;
188 count = ARRAY_SIZE(irq_cfgx);
190 for (i = 0; i < count; i++) {
191 desc = irq_to_desc(i);
192 desc->chip_data = &cfg[i];
193 alloc_bootmem_cpumask_var(&cfg[i].domain);
194 alloc_bootmem_cpumask_var(&cfg[i].old_domain);
195 if (i < NR_IRQS_LEGACY)
196 cpumask_setall(cfg[i].domain);
202 #ifdef CONFIG_SPARSE_IRQ
203 static struct irq_cfg *irq_cfg(unsigned int irq)
205 struct irq_cfg *cfg = NULL;
206 struct irq_desc *desc;
208 desc = irq_to_desc(irq);
210 cfg = desc->chip_data;
215 static struct irq_cfg *get_one_free_irq_cfg(int cpu)
220 node = cpu_to_node(cpu);
222 cfg = kzalloc_node(sizeof(*cfg), GFP_ATOMIC, node);
224 if (!alloc_cpumask_var_node(&cfg->domain, GFP_ATOMIC, node)) {
227 } else if (!alloc_cpumask_var_node(&cfg->old_domain,
229 free_cpumask_var(cfg->domain);
233 cpumask_clear(cfg->domain);
234 cpumask_clear(cfg->old_domain);
241 int arch_init_chip_data(struct irq_desc *desc, int cpu)
245 cfg = desc->chip_data;
247 desc->chip_data = get_one_free_irq_cfg(cpu);
248 if (!desc->chip_data) {
249 printk(KERN_ERR "can not alloc irq_cfg\n");
257 #ifdef CONFIG_NUMA_MIGRATE_IRQ_DESC
260 init_copy_irq_2_pin(struct irq_cfg *old_cfg, struct irq_cfg *cfg, int cpu)
262 struct irq_pin_list *old_entry, *head, *tail, *entry;
264 cfg->irq_2_pin = NULL;
265 old_entry = old_cfg->irq_2_pin;
269 entry = get_one_free_irq_2_pin(cpu);
273 entry->apic = old_entry->apic;
274 entry->pin = old_entry->pin;
277 old_entry = old_entry->next;
279 entry = get_one_free_irq_2_pin(cpu);
287 /* still use the old one */
290 entry->apic = old_entry->apic;
291 entry->pin = old_entry->pin;
294 old_entry = old_entry->next;
298 cfg->irq_2_pin = head;
301 static void free_irq_2_pin(struct irq_cfg *old_cfg, struct irq_cfg *cfg)
303 struct irq_pin_list *entry, *next;
305 if (old_cfg->irq_2_pin == cfg->irq_2_pin)
308 entry = old_cfg->irq_2_pin;
315 old_cfg->irq_2_pin = NULL;
318 void arch_init_copy_chip_data(struct irq_desc *old_desc,
319 struct irq_desc *desc, int cpu)
322 struct irq_cfg *old_cfg;
324 cfg = get_one_free_irq_cfg(cpu);
329 desc->chip_data = cfg;
331 old_cfg = old_desc->chip_data;
333 memcpy(cfg, old_cfg, sizeof(struct irq_cfg));
335 init_copy_irq_2_pin(old_cfg, cfg, cpu);
338 static void free_irq_cfg(struct irq_cfg *old_cfg)
343 void arch_free_chip_data(struct irq_desc *old_desc, struct irq_desc *desc)
345 struct irq_cfg *old_cfg, *cfg;
347 old_cfg = old_desc->chip_data;
348 cfg = desc->chip_data;
354 free_irq_2_pin(old_cfg, cfg);
355 free_irq_cfg(old_cfg);
356 old_desc->chip_data = NULL;
361 set_extra_move_desc(struct irq_desc *desc, const struct cpumask *mask)
363 struct irq_cfg *cfg = desc->chip_data;
365 if (!cfg->move_in_progress) {
366 /* it means that domain is not changed */
367 if (!cpumask_intersects(desc->affinity, mask))
368 cfg->move_desc_pending = 1;
374 static struct irq_cfg *irq_cfg(unsigned int irq)
376 return irq < nr_irqs ? irq_cfgx + irq : NULL;
381 #ifndef CONFIG_NUMA_MIGRATE_IRQ_DESC
383 set_extra_move_desc(struct irq_desc *desc, const struct cpumask *mask)
390 unsigned int unused[3];
392 unsigned int unused2[11];
396 static __attribute_const__ struct io_apic __iomem *io_apic_base(int idx)
398 return (void __iomem *) __fix_to_virt(FIX_IO_APIC_BASE_0 + idx)
399 + (mp_ioapics[idx].apicaddr & ~PAGE_MASK);
402 static inline void io_apic_eoi(unsigned int apic, unsigned int vector)
404 struct io_apic __iomem *io_apic = io_apic_base(apic);
405 writel(vector, &io_apic->eoi);
408 static inline unsigned int io_apic_read(unsigned int apic, unsigned int reg)
410 struct io_apic __iomem *io_apic = io_apic_base(apic);
411 writel(reg, &io_apic->index);
412 return readl(&io_apic->data);
415 static inline void io_apic_write(unsigned int apic, unsigned int reg, unsigned int value)
417 struct io_apic __iomem *io_apic = io_apic_base(apic);
418 writel(reg, &io_apic->index);
419 writel(value, &io_apic->data);
423 * Re-write a value: to be used for read-modify-write
424 * cycles where the read already set up the index register.
426 * Older SiS APIC requires we rewrite the index register
428 static inline void io_apic_modify(unsigned int apic, unsigned int reg, unsigned int value)
430 struct io_apic __iomem *io_apic = io_apic_base(apic);
433 writel(reg, &io_apic->index);
434 writel(value, &io_apic->data);
437 static bool io_apic_level_ack_pending(struct irq_cfg *cfg)
439 struct irq_pin_list *entry;
442 spin_lock_irqsave(&ioapic_lock, flags);
443 entry = cfg->irq_2_pin;
451 reg = io_apic_read(entry->apic, 0x10 + pin*2);
452 /* Is the remote IRR bit set? */
453 if (reg & IO_APIC_REDIR_REMOTE_IRR) {
454 spin_unlock_irqrestore(&ioapic_lock, flags);
461 spin_unlock_irqrestore(&ioapic_lock, flags);
467 struct { u32 w1, w2; };
468 struct IO_APIC_route_entry entry;
471 static struct IO_APIC_route_entry ioapic_read_entry(int apic, int pin)
473 union entry_union eu;
475 spin_lock_irqsave(&ioapic_lock, flags);
476 eu.w1 = io_apic_read(apic, 0x10 + 2 * pin);
477 eu.w2 = io_apic_read(apic, 0x11 + 2 * pin);
478 spin_unlock_irqrestore(&ioapic_lock, flags);
483 * When we write a new IO APIC routing entry, we need to write the high
484 * word first! If the mask bit in the low word is clear, we will enable
485 * the interrupt, and we need to make sure the entry is fully populated
486 * before that happens.
489 __ioapic_write_entry(int apic, int pin, struct IO_APIC_route_entry e)
491 union entry_union eu;
493 io_apic_write(apic, 0x11 + 2*pin, eu.w2);
494 io_apic_write(apic, 0x10 + 2*pin, eu.w1);
497 void ioapic_write_entry(int apic, int pin, struct IO_APIC_route_entry e)
500 spin_lock_irqsave(&ioapic_lock, flags);
501 __ioapic_write_entry(apic, pin, e);
502 spin_unlock_irqrestore(&ioapic_lock, flags);
506 * When we mask an IO APIC routing entry, we need to write the low
507 * word first, in order to set the mask bit before we change the
510 static void ioapic_mask_entry(int apic, int pin)
513 union entry_union eu = { .entry.mask = 1 };
515 spin_lock_irqsave(&ioapic_lock, flags);
516 io_apic_write(apic, 0x10 + 2*pin, eu.w1);
517 io_apic_write(apic, 0x11 + 2*pin, eu.w2);
518 spin_unlock_irqrestore(&ioapic_lock, flags);
522 static void send_cleanup_vector(struct irq_cfg *cfg)
524 cpumask_var_t cleanup_mask;
526 if (unlikely(!alloc_cpumask_var(&cleanup_mask, GFP_ATOMIC))) {
528 cfg->move_cleanup_count = 0;
529 for_each_cpu_and(i, cfg->old_domain, cpu_online_mask)
530 cfg->move_cleanup_count++;
531 for_each_cpu_and(i, cfg->old_domain, cpu_online_mask)
532 apic->send_IPI_mask(cpumask_of(i), IRQ_MOVE_CLEANUP_VECTOR);
534 cpumask_and(cleanup_mask, cfg->old_domain, cpu_online_mask);
535 cfg->move_cleanup_count = cpumask_weight(cleanup_mask);
536 apic->send_IPI_mask(cleanup_mask, IRQ_MOVE_CLEANUP_VECTOR);
537 free_cpumask_var(cleanup_mask);
539 cfg->move_in_progress = 0;
542 static void __target_IO_APIC_irq(unsigned int irq, unsigned int dest, struct irq_cfg *cfg)
545 struct irq_pin_list *entry;
546 u8 vector = cfg->vector;
548 entry = cfg->irq_2_pin;
558 * With interrupt-remapping, destination information comes
559 * from interrupt-remapping table entry.
561 if (!irq_remapped(irq))
562 io_apic_write(apic, 0x11 + pin*2, dest);
563 reg = io_apic_read(apic, 0x10 + pin*2);
564 reg &= ~IO_APIC_REDIR_VECTOR_MASK;
566 io_apic_modify(apic, 0x10 + pin*2, reg);
574 assign_irq_vector(int irq, struct irq_cfg *cfg, const struct cpumask *mask);
577 * Either sets desc->affinity to a valid value, and returns
578 * ->cpu_mask_to_apicid of that, or returns BAD_APICID and
579 * leaves desc->affinity untouched.
582 set_desc_affinity(struct irq_desc *desc, const struct cpumask *mask)
587 if (!cpumask_intersects(mask, cpu_online_mask))
591 cfg = desc->chip_data;
592 if (assign_irq_vector(irq, cfg, mask))
595 /* check that before desc->addinity get updated */
596 set_extra_move_desc(desc, mask);
598 cpumask_copy(desc->affinity, mask);
600 return apic->cpu_mask_to_apicid_and(desc->affinity, cfg->domain);
604 set_ioapic_affinity_irq_desc(struct irq_desc *desc, const struct cpumask *mask)
612 cfg = desc->chip_data;
614 spin_lock_irqsave(&ioapic_lock, flags);
615 dest = set_desc_affinity(desc, mask);
616 if (dest != BAD_APICID) {
617 /* Only the high 8 bits are valid. */
618 dest = SET_APIC_LOGICAL_ID(dest);
619 __target_IO_APIC_irq(irq, dest, cfg);
621 spin_unlock_irqrestore(&ioapic_lock, flags);
625 set_ioapic_affinity_irq(unsigned int irq, const struct cpumask *mask)
627 struct irq_desc *desc;
629 desc = irq_to_desc(irq);
631 set_ioapic_affinity_irq_desc(desc, mask);
633 #endif /* CONFIG_SMP */
636 * The common case is 1:1 IRQ<->pin mappings. Sometimes there are
637 * shared ISA-space IRQs, so we have to support them. We are super
638 * fast in the common case, and fast for shared ISA-space IRQs.
640 static void add_pin_to_irq_cpu(struct irq_cfg *cfg, int cpu, int apic, int pin)
642 struct irq_pin_list *entry;
644 entry = cfg->irq_2_pin;
646 entry = get_one_free_irq_2_pin(cpu);
648 printk(KERN_ERR "can not alloc irq_2_pin to add %d - %d\n",
652 cfg->irq_2_pin = entry;
658 while (entry->next) {
659 /* not again, please */
660 if (entry->apic == apic && entry->pin == pin)
666 entry->next = get_one_free_irq_2_pin(cpu);
673 * Reroute an IRQ to a different pin.
675 static void __init replace_pin_at_irq_cpu(struct irq_cfg *cfg, int cpu,
676 int oldapic, int oldpin,
677 int newapic, int newpin)
679 struct irq_pin_list *entry = cfg->irq_2_pin;
683 if (entry->apic == oldapic && entry->pin == oldpin) {
684 entry->apic = newapic;
687 /* every one is different, right? */
693 /* why? call replace before add? */
695 add_pin_to_irq_cpu(cfg, cpu, newapic, newpin);
698 static inline void io_apic_modify_irq(struct irq_cfg *cfg,
699 int mask_and, int mask_or,
700 void (*final)(struct irq_pin_list *entry))
703 struct irq_pin_list *entry;
705 for (entry = cfg->irq_2_pin; entry != NULL; entry = entry->next) {
708 reg = io_apic_read(entry->apic, 0x10 + pin * 2);
711 io_apic_modify(entry->apic, 0x10 + pin * 2, reg);
717 static void __unmask_IO_APIC_irq(struct irq_cfg *cfg)
719 io_apic_modify_irq(cfg, ~IO_APIC_REDIR_MASKED, 0, NULL);
723 static void io_apic_sync(struct irq_pin_list *entry)
726 * Synchronize the IO-APIC and the CPU by doing
727 * a dummy read from the IO-APIC
729 struct io_apic __iomem *io_apic;
730 io_apic = io_apic_base(entry->apic);
731 readl(&io_apic->data);
734 static void __mask_IO_APIC_irq(struct irq_cfg *cfg)
736 io_apic_modify_irq(cfg, ~0, IO_APIC_REDIR_MASKED, &io_apic_sync);
738 #else /* CONFIG_X86_32 */
739 static void __mask_IO_APIC_irq(struct irq_cfg *cfg)
741 io_apic_modify_irq(cfg, ~0, IO_APIC_REDIR_MASKED, NULL);
744 static void __mask_and_edge_IO_APIC_irq(struct irq_cfg *cfg)
746 io_apic_modify_irq(cfg, ~IO_APIC_REDIR_LEVEL_TRIGGER,
747 IO_APIC_REDIR_MASKED, NULL);
750 static void __unmask_and_level_IO_APIC_irq(struct irq_cfg *cfg)
752 io_apic_modify_irq(cfg, ~IO_APIC_REDIR_MASKED,
753 IO_APIC_REDIR_LEVEL_TRIGGER, NULL);
755 #endif /* CONFIG_X86_32 */
757 static void mask_IO_APIC_irq_desc(struct irq_desc *desc)
759 struct irq_cfg *cfg = desc->chip_data;
764 spin_lock_irqsave(&ioapic_lock, flags);
765 __mask_IO_APIC_irq(cfg);
766 spin_unlock_irqrestore(&ioapic_lock, flags);
769 static void unmask_IO_APIC_irq_desc(struct irq_desc *desc)
771 struct irq_cfg *cfg = desc->chip_data;
774 spin_lock_irqsave(&ioapic_lock, flags);
775 __unmask_IO_APIC_irq(cfg);
776 spin_unlock_irqrestore(&ioapic_lock, flags);
779 static void mask_IO_APIC_irq(unsigned int irq)
781 struct irq_desc *desc = irq_to_desc(irq);
783 mask_IO_APIC_irq_desc(desc);
785 static void unmask_IO_APIC_irq(unsigned int irq)
787 struct irq_desc *desc = irq_to_desc(irq);
789 unmask_IO_APIC_irq_desc(desc);
792 static void clear_IO_APIC_pin(unsigned int apic, unsigned int pin)
794 struct IO_APIC_route_entry entry;
796 /* Check delivery_mode to be sure we're not clearing an SMI pin */
797 entry = ioapic_read_entry(apic, pin);
798 if (entry.delivery_mode == dest_SMI)
801 * Disable it in the IO-APIC irq-routing table:
803 ioapic_mask_entry(apic, pin);
806 static void clear_IO_APIC (void)
810 for (apic = 0; apic < nr_ioapics; apic++)
811 for (pin = 0; pin < nr_ioapic_registers[apic]; pin++)
812 clear_IO_APIC_pin(apic, pin);
817 * support for broken MP BIOSs, enables hand-redirection of PIRQ0-7 to
818 * specific CPU-side IRQs.
822 static int pirq_entries[MAX_PIRQS] = {
823 [0 ... MAX_PIRQS - 1] = -1
826 static int __init ioapic_pirq_setup(char *str)
829 int ints[MAX_PIRQS+1];
831 get_options(str, ARRAY_SIZE(ints), ints);
833 apic_printk(APIC_VERBOSE, KERN_INFO
834 "PIRQ redirection, working around broken MP-BIOS.\n");
836 if (ints[0] < MAX_PIRQS)
839 for (i = 0; i < max; i++) {
840 apic_printk(APIC_VERBOSE, KERN_DEBUG
841 "... PIRQ%d -> IRQ %d\n", i, ints[i+1]);
843 * PIRQs are mapped upside down, usually.
845 pirq_entries[MAX_PIRQS-i-1] = ints[i+1];
850 __setup("pirq=", ioapic_pirq_setup);
851 #endif /* CONFIG_X86_32 */
853 #ifdef CONFIG_INTR_REMAP
854 /* I/O APIC RTE contents at the OS boot up */
855 static struct IO_APIC_route_entry *early_ioapic_entries[MAX_IO_APICS];
858 * Saves all the IO-APIC RTE's
860 int save_IO_APIC_setup(void)
862 union IO_APIC_reg_01 reg_01;
867 * The number of IO-APIC IRQ registers (== #pins):
869 for (apic = 0; apic < nr_ioapics; apic++) {
870 spin_lock_irqsave(&ioapic_lock, flags);
871 reg_01.raw = io_apic_read(apic, 1);
872 spin_unlock_irqrestore(&ioapic_lock, flags);
873 nr_ioapic_registers[apic] = reg_01.bits.entries+1;
876 for (apic = 0; apic < nr_ioapics; apic++) {
877 early_ioapic_entries[apic] =
878 kzalloc(sizeof(struct IO_APIC_route_entry) *
879 nr_ioapic_registers[apic], GFP_KERNEL);
880 if (!early_ioapic_entries[apic])
884 for (apic = 0; apic < nr_ioapics; apic++)
885 for (pin = 0; pin < nr_ioapic_registers[apic]; pin++)
886 early_ioapic_entries[apic][pin] =
887 ioapic_read_entry(apic, pin);
893 kfree(early_ioapic_entries[apic--]);
894 memset(early_ioapic_entries, 0,
895 ARRAY_SIZE(early_ioapic_entries));
900 void mask_IO_APIC_setup(void)
904 for (apic = 0; apic < nr_ioapics; apic++) {
905 if (!early_ioapic_entries[apic])
907 for (pin = 0; pin < nr_ioapic_registers[apic]; pin++) {
908 struct IO_APIC_route_entry entry;
910 entry = early_ioapic_entries[apic][pin];
913 ioapic_write_entry(apic, pin, entry);
919 void restore_IO_APIC_setup(void)
923 for (apic = 0; apic < nr_ioapics; apic++) {
924 if (!early_ioapic_entries[apic])
926 for (pin = 0; pin < nr_ioapic_registers[apic]; pin++)
927 ioapic_write_entry(apic, pin,
928 early_ioapic_entries[apic][pin]);
929 kfree(early_ioapic_entries[apic]);
930 early_ioapic_entries[apic] = NULL;
934 void reinit_intr_remapped_IO_APIC(int intr_remapping)
937 * for now plain restore of previous settings.
938 * TBD: In the case of OS enabling interrupt-remapping,
939 * IO-APIC RTE's need to be setup to point to interrupt-remapping
940 * table entries. for now, do a plain restore, and wait for
941 * the setup_IO_APIC_irqs() to do proper initialization.
943 restore_IO_APIC_setup();
948 * Find the IRQ entry number of a certain pin.
950 static int find_irq_entry(int apic, int pin, int type)
954 for (i = 0; i < mp_irq_entries; i++)
955 if (mp_irqs[i].irqtype == type &&
956 (mp_irqs[i].dstapic == mp_ioapics[apic].apicid ||
957 mp_irqs[i].dstapic == MP_APIC_ALL) &&
958 mp_irqs[i].dstirq == pin)
965 * Find the pin to which IRQ[irq] (ISA) is connected
967 static int __init find_isa_irq_pin(int irq, int type)
971 for (i = 0; i < mp_irq_entries; i++) {
972 int lbus = mp_irqs[i].srcbus;
974 if (test_bit(lbus, mp_bus_not_pci) &&
975 (mp_irqs[i].irqtype == type) &&
976 (mp_irqs[i].srcbusirq == irq))
978 return mp_irqs[i].dstirq;
983 static int __init find_isa_irq_apic(int irq, int type)
987 for (i = 0; i < mp_irq_entries; i++) {
988 int lbus = mp_irqs[i].srcbus;
990 if (test_bit(lbus, mp_bus_not_pci) &&
991 (mp_irqs[i].irqtype == type) &&
992 (mp_irqs[i].srcbusirq == irq))
995 if (i < mp_irq_entries) {
997 for(apic = 0; apic < nr_ioapics; apic++) {
998 if (mp_ioapics[apic].apicid == mp_irqs[i].dstapic)
1007 * Find a specific PCI IRQ entry.
1008 * Not an __init, possibly needed by modules
1010 static int pin_2_irq(int idx, int apic, int pin);
1012 int IO_APIC_get_PCI_irq_vector(int bus, int slot, int pin)
1014 int apic, i, best_guess = -1;
1016 apic_printk(APIC_DEBUG, "querying PCI -> IRQ mapping bus:%d, slot:%d, pin:%d.\n",
1018 if (test_bit(bus, mp_bus_not_pci)) {
1019 apic_printk(APIC_VERBOSE, "PCI BIOS passed nonexistent PCI bus %d!\n", bus);
1022 for (i = 0; i < mp_irq_entries; i++) {
1023 int lbus = mp_irqs[i].srcbus;
1025 for (apic = 0; apic < nr_ioapics; apic++)
1026 if (mp_ioapics[apic].apicid == mp_irqs[i].dstapic ||
1027 mp_irqs[i].dstapic == MP_APIC_ALL)
1030 if (!test_bit(lbus, mp_bus_not_pci) &&
1031 !mp_irqs[i].irqtype &&
1033 (slot == ((mp_irqs[i].srcbusirq >> 2) & 0x1f))) {
1034 int irq = pin_2_irq(i, apic, mp_irqs[i].dstirq);
1036 if (!(apic || IO_APIC_IRQ(irq)))
1039 if (pin == (mp_irqs[i].srcbusirq & 3))
1042 * Use the first all-but-pin matching entry as a
1043 * best-guess fuzzy result for broken mptables.
1052 EXPORT_SYMBOL(IO_APIC_get_PCI_irq_vector);
1054 #if defined(CONFIG_EISA) || defined(CONFIG_MCA)
1056 * EISA Edge/Level control register, ELCR
1058 static int EISA_ELCR(unsigned int irq)
1060 if (irq < NR_IRQS_LEGACY) {
1061 unsigned int port = 0x4d0 + (irq >> 3);
1062 return (inb(port) >> (irq & 7)) & 1;
1064 apic_printk(APIC_VERBOSE, KERN_INFO
1065 "Broken MPtable reports ISA irq %d\n", irq);
1071 /* ISA interrupts are always polarity zero edge triggered,
1072 * when listed as conforming in the MP table. */
1074 #define default_ISA_trigger(idx) (0)
1075 #define default_ISA_polarity(idx) (0)
1077 /* EISA interrupts are always polarity zero and can be edge or level
1078 * trigger depending on the ELCR value. If an interrupt is listed as
1079 * EISA conforming in the MP table, that means its trigger type must
1080 * be read in from the ELCR */
1082 #define default_EISA_trigger(idx) (EISA_ELCR(mp_irqs[idx].srcbusirq))
1083 #define default_EISA_polarity(idx) default_ISA_polarity(idx)
1085 /* PCI interrupts are always polarity one level triggered,
1086 * when listed as conforming in the MP table. */
1088 #define default_PCI_trigger(idx) (1)
1089 #define default_PCI_polarity(idx) (1)
1091 /* MCA interrupts are always polarity zero level triggered,
1092 * when listed as conforming in the MP table. */
1094 #define default_MCA_trigger(idx) (1)
1095 #define default_MCA_polarity(idx) default_ISA_polarity(idx)
1097 static int MPBIOS_polarity(int idx)
1099 int bus = mp_irqs[idx].srcbus;
1103 * Determine IRQ line polarity (high active or low active):
1105 switch (mp_irqs[idx].irqflag & 3)
1107 case 0: /* conforms, ie. bus-type dependent polarity */
1108 if (test_bit(bus, mp_bus_not_pci))
1109 polarity = default_ISA_polarity(idx);
1111 polarity = default_PCI_polarity(idx);
1113 case 1: /* high active */
1118 case 2: /* reserved */
1120 printk(KERN_WARNING "broken BIOS!!\n");
1124 case 3: /* low active */
1129 default: /* invalid */
1131 printk(KERN_WARNING "broken BIOS!!\n");
1139 static int MPBIOS_trigger(int idx)
1141 int bus = mp_irqs[idx].srcbus;
1145 * Determine IRQ trigger mode (edge or level sensitive):
1147 switch ((mp_irqs[idx].irqflag>>2) & 3)
1149 case 0: /* conforms, ie. bus-type dependent */
1150 if (test_bit(bus, mp_bus_not_pci))
1151 trigger = default_ISA_trigger(idx);
1153 trigger = default_PCI_trigger(idx);
1154 #if defined(CONFIG_EISA) || defined(CONFIG_MCA)
1155 switch (mp_bus_id_to_type[bus]) {
1156 case MP_BUS_ISA: /* ISA pin */
1158 /* set before the switch */
1161 case MP_BUS_EISA: /* EISA pin */
1163 trigger = default_EISA_trigger(idx);
1166 case MP_BUS_PCI: /* PCI pin */
1168 /* set before the switch */
1171 case MP_BUS_MCA: /* MCA pin */
1173 trigger = default_MCA_trigger(idx);
1178 printk(KERN_WARNING "broken BIOS!!\n");
1190 case 2: /* reserved */
1192 printk(KERN_WARNING "broken BIOS!!\n");
1201 default: /* invalid */
1203 printk(KERN_WARNING "broken BIOS!!\n");
1211 static inline int irq_polarity(int idx)
1213 return MPBIOS_polarity(idx);
1216 static inline int irq_trigger(int idx)
1218 return MPBIOS_trigger(idx);
1221 int (*ioapic_renumber_irq)(int ioapic, int irq);
1222 static int pin_2_irq(int idx, int apic, int pin)
1225 int bus = mp_irqs[idx].srcbus;
1228 * Debugging check, we are in big trouble if this message pops up!
1230 if (mp_irqs[idx].dstirq != pin)
1231 printk(KERN_ERR "broken BIOS or MPTABLE parser, ayiee!!\n");
1233 if (test_bit(bus, mp_bus_not_pci)) {
1234 irq = mp_irqs[idx].srcbusirq;
1237 * PCI IRQs are mapped in order
1241 irq += nr_ioapic_registers[i++];
1244 * For MPS mode, so far only needed by ES7000 platform
1246 if (ioapic_renumber_irq)
1247 irq = ioapic_renumber_irq(apic, irq);
1250 #ifdef CONFIG_X86_32
1252 * PCI IRQ command line redirection. Yes, limits are hardcoded.
1254 if ((pin >= 16) && (pin <= 23)) {
1255 if (pirq_entries[pin-16] != -1) {
1256 if (!pirq_entries[pin-16]) {
1257 apic_printk(APIC_VERBOSE, KERN_DEBUG
1258 "disabling PIRQ%d\n", pin-16);
1260 irq = pirq_entries[pin-16];
1261 apic_printk(APIC_VERBOSE, KERN_DEBUG
1262 "using PIRQ%d -> IRQ %d\n",
1272 void lock_vector_lock(void)
1274 /* Used to the online set of cpus does not change
1275 * during assign_irq_vector.
1277 spin_lock(&vector_lock);
1280 void unlock_vector_lock(void)
1282 spin_unlock(&vector_lock);
1286 __assign_irq_vector(int irq, struct irq_cfg *cfg, const struct cpumask *mask)
1289 * NOTE! The local APIC isn't very good at handling
1290 * multiple interrupts at the same interrupt level.
1291 * As the interrupt level is determined by taking the
1292 * vector number and shifting that right by 4, we
1293 * want to spread these out a bit so that they don't
1294 * all fall in the same interrupt level.
1296 * Also, we've got to be careful not to trash gate
1297 * 0x80, because int 0x80 is hm, kind of importantish. ;)
1299 static int current_vector = FIRST_DEVICE_VECTOR, current_offset = 0;
1300 unsigned int old_vector;
1302 cpumask_var_t tmp_mask;
1304 if ((cfg->move_in_progress) || cfg->move_cleanup_count)
1307 if (!alloc_cpumask_var(&tmp_mask, GFP_ATOMIC))
1310 old_vector = cfg->vector;
1312 cpumask_and(tmp_mask, mask, cpu_online_mask);
1313 cpumask_and(tmp_mask, cfg->domain, tmp_mask);
1314 if (!cpumask_empty(tmp_mask)) {
1315 free_cpumask_var(tmp_mask);
1320 /* Only try and allocate irqs on cpus that are present */
1322 for_each_cpu_and(cpu, mask, cpu_online_mask) {
1326 apic->vector_allocation_domain(cpu, tmp_mask);
1328 vector = current_vector;
1329 offset = current_offset;
1332 if (vector >= first_system_vector) {
1333 /* If out of vectors on large boxen, must share them. */
1334 offset = (offset + 1) % 8;
1335 vector = FIRST_DEVICE_VECTOR + offset;
1337 if (unlikely(current_vector == vector))
1340 if (test_bit(vector, used_vectors))
1343 for_each_cpu_and(new_cpu, tmp_mask, cpu_online_mask)
1344 if (per_cpu(vector_irq, new_cpu)[vector] != -1)
1347 current_vector = vector;
1348 current_offset = offset;
1350 cfg->move_in_progress = 1;
1351 cpumask_copy(cfg->old_domain, cfg->domain);
1353 for_each_cpu_and(new_cpu, tmp_mask, cpu_online_mask)
1354 per_cpu(vector_irq, new_cpu)[vector] = irq;
1355 cfg->vector = vector;
1356 cpumask_copy(cfg->domain, tmp_mask);
1360 free_cpumask_var(tmp_mask);
1365 assign_irq_vector(int irq, struct irq_cfg *cfg, const struct cpumask *mask)
1368 unsigned long flags;
1370 spin_lock_irqsave(&vector_lock, flags);
1371 err = __assign_irq_vector(irq, cfg, mask);
1372 spin_unlock_irqrestore(&vector_lock, flags);
1376 static void __clear_irq_vector(int irq, struct irq_cfg *cfg)
1380 BUG_ON(!cfg->vector);
1382 vector = cfg->vector;
1383 for_each_cpu_and(cpu, cfg->domain, cpu_online_mask)
1384 per_cpu(vector_irq, cpu)[vector] = -1;
1387 cpumask_clear(cfg->domain);
1389 if (likely(!cfg->move_in_progress))
1391 for_each_cpu_and(cpu, cfg->old_domain, cpu_online_mask) {
1392 for (vector = FIRST_EXTERNAL_VECTOR; vector < NR_VECTORS;
1394 if (per_cpu(vector_irq, cpu)[vector] != irq)
1396 per_cpu(vector_irq, cpu)[vector] = -1;
1400 cfg->move_in_progress = 0;
1403 void __setup_vector_irq(int cpu)
1405 /* Initialize vector_irq on a new cpu */
1406 /* This function must be called with vector_lock held */
1408 struct irq_cfg *cfg;
1409 struct irq_desc *desc;
1411 /* Mark the inuse vectors */
1412 for_each_irq_desc(irq, desc) {
1413 cfg = desc->chip_data;
1414 if (!cpumask_test_cpu(cpu, cfg->domain))
1416 vector = cfg->vector;
1417 per_cpu(vector_irq, cpu)[vector] = irq;
1419 /* Mark the free vectors */
1420 for (vector = 0; vector < NR_VECTORS; ++vector) {
1421 irq = per_cpu(vector_irq, cpu)[vector];
1426 if (!cpumask_test_cpu(cpu, cfg->domain))
1427 per_cpu(vector_irq, cpu)[vector] = -1;
1431 static struct irq_chip ioapic_chip;
1432 static struct irq_chip ir_ioapic_chip;
1433 static struct irq_chip msi_ir_chip;
1435 #define IOAPIC_AUTO -1
1436 #define IOAPIC_EDGE 0
1437 #define IOAPIC_LEVEL 1
1439 #ifdef CONFIG_X86_32
1440 static inline int IO_APIC_irq_trigger(int irq)
1444 for (apic = 0; apic < nr_ioapics; apic++) {
1445 for (pin = 0; pin < nr_ioapic_registers[apic]; pin++) {
1446 idx = find_irq_entry(apic, pin, mp_INT);
1447 if ((idx != -1) && (irq == pin_2_irq(idx, apic, pin)))
1448 return irq_trigger(idx);
1452 * nonexistent IRQs are edge default
1457 static inline int IO_APIC_irq_trigger(int irq)
1463 static void ioapic_register_intr(int irq, struct irq_desc *desc, unsigned long trigger)
1466 if ((trigger == IOAPIC_AUTO && IO_APIC_irq_trigger(irq)) ||
1467 trigger == IOAPIC_LEVEL)
1468 desc->status |= IRQ_LEVEL;
1470 desc->status &= ~IRQ_LEVEL;
1472 if (irq_remapped(irq)) {
1473 desc->status |= IRQ_MOVE_PCNTXT;
1475 set_irq_chip_and_handler_name(irq, &ir_ioapic_chip,
1479 set_irq_chip_and_handler_name(irq, &ir_ioapic_chip,
1480 handle_edge_irq, "edge");
1484 if ((trigger == IOAPIC_AUTO && IO_APIC_irq_trigger(irq)) ||
1485 trigger == IOAPIC_LEVEL)
1486 set_irq_chip_and_handler_name(irq, &ioapic_chip,
1490 set_irq_chip_and_handler_name(irq, &ioapic_chip,
1491 handle_edge_irq, "edge");
1494 int setup_ioapic_entry(int apic_id, int irq,
1495 struct IO_APIC_route_entry *entry,
1496 unsigned int destination, int trigger,
1497 int polarity, int vector, int pin)
1500 * add it to the IO-APIC irq-routing table:
1502 memset(entry,0,sizeof(*entry));
1504 if (intr_remapping_enabled) {
1505 struct intel_iommu *iommu = map_ioapic_to_ir(apic_id);
1507 struct IR_IO_APIC_route_entry *ir_entry =
1508 (struct IR_IO_APIC_route_entry *) entry;
1512 panic("No mapping iommu for ioapic %d\n", apic_id);
1514 index = alloc_irte(iommu, irq, 1);
1516 panic("Failed to allocate IRTE for ioapic %d\n", apic_id);
1518 memset(&irte, 0, sizeof(irte));
1521 irte.dst_mode = apic->irq_dest_mode;
1523 * Trigger mode in the IRTE will always be edge, and the
1524 * actual level or edge trigger will be setup in the IO-APIC
1525 * RTE. This will help simplify level triggered irq migration.
1526 * For more details, see the comments above explainig IO-APIC
1527 * irq migration in the presence of interrupt-remapping.
1529 irte.trigger_mode = 0;
1530 irte.dlvry_mode = apic->irq_delivery_mode;
1531 irte.vector = vector;
1532 irte.dest_id = IRTE_DEST(destination);
1534 modify_irte(irq, &irte);
1536 ir_entry->index2 = (index >> 15) & 0x1;
1538 ir_entry->format = 1;
1539 ir_entry->index = (index & 0x7fff);
1541 * IO-APIC RTE will be configured with virtual vector.
1542 * irq handler will do the explicit EOI to the io-apic.
1544 ir_entry->vector = pin;
1546 entry->delivery_mode = apic->irq_delivery_mode;
1547 entry->dest_mode = apic->irq_dest_mode;
1548 entry->dest = destination;
1549 entry->vector = vector;
1552 entry->mask = 0; /* enable IRQ */
1553 entry->trigger = trigger;
1554 entry->polarity = polarity;
1556 /* Mask level triggered irqs.
1557 * Use IRQ_DELAYED_DISABLE for edge triggered irqs.
1564 static void setup_IO_APIC_irq(int apic_id, int pin, unsigned int irq, struct irq_desc *desc,
1565 int trigger, int polarity)
1567 struct irq_cfg *cfg;
1568 struct IO_APIC_route_entry entry;
1571 if (!IO_APIC_IRQ(irq))
1574 cfg = desc->chip_data;
1576 if (assign_irq_vector(irq, cfg, apic->target_cpus()))
1579 dest = apic->cpu_mask_to_apicid_and(cfg->domain, apic->target_cpus());
1581 apic_printk(APIC_VERBOSE,KERN_DEBUG
1582 "IOAPIC[%d]: Set routing entry (%d-%d -> 0x%x -> "
1583 "IRQ %d Mode:%i Active:%i)\n",
1584 apic_id, mp_ioapics[apic_id].apicid, pin, cfg->vector,
1585 irq, trigger, polarity);
1588 if (setup_ioapic_entry(mp_ioapics[apic_id].apicid, irq, &entry,
1589 dest, trigger, polarity, cfg->vector, pin)) {
1590 printk("Failed to setup ioapic entry for ioapic %d, pin %d\n",
1591 mp_ioapics[apic_id].apicid, pin);
1592 __clear_irq_vector(irq, cfg);
1596 ioapic_register_intr(irq, desc, trigger);
1597 if (irq < NR_IRQS_LEGACY)
1598 disable_8259A_irq(irq);
1600 ioapic_write_entry(apic_id, pin, entry);
1603 static void __init setup_IO_APIC_irqs(void)
1605 int apic_id, pin, idx, irq;
1607 struct irq_desc *desc;
1608 struct irq_cfg *cfg;
1609 int cpu = boot_cpu_id;
1611 apic_printk(APIC_VERBOSE, KERN_DEBUG "init IO_APIC IRQs\n");
1613 for (apic_id = 0; apic_id < nr_ioapics; apic_id++) {
1614 for (pin = 0; pin < nr_ioapic_registers[apic_id]; pin++) {
1616 idx = find_irq_entry(apic_id, pin, mp_INT);
1620 apic_printk(APIC_VERBOSE,
1621 KERN_DEBUG " %d-%d",
1622 mp_ioapics[apic_id].apicid, pin);
1624 apic_printk(APIC_VERBOSE, " %d-%d",
1625 mp_ioapics[apic_id].apicid, pin);
1629 apic_printk(APIC_VERBOSE,
1630 " (apicid-pin) not connected\n");
1634 irq = pin_2_irq(idx, apic_id, pin);
1637 * Skip the timer IRQ if there's a quirk handler
1638 * installed and if it returns 1:
1640 if (apic->multi_timer_check &&
1641 apic->multi_timer_check(apic_id, irq))
1644 desc = irq_to_desc_alloc_cpu(irq, cpu);
1646 printk(KERN_INFO "can not get irq_desc for %d\n", irq);
1649 cfg = desc->chip_data;
1650 add_pin_to_irq_cpu(cfg, cpu, apic_id, pin);
1652 setup_IO_APIC_irq(apic_id, pin, irq, desc,
1653 irq_trigger(idx), irq_polarity(idx));
1658 apic_printk(APIC_VERBOSE,
1659 " (apicid-pin) not connected\n");
1663 * Set up the timer pin, possibly with the 8259A-master behind.
1665 static void __init setup_timer_IRQ0_pin(unsigned int apic_id, unsigned int pin,
1668 struct IO_APIC_route_entry entry;
1670 if (intr_remapping_enabled)
1673 memset(&entry, 0, sizeof(entry));
1676 * We use logical delivery to get the timer IRQ
1679 entry.dest_mode = apic->irq_dest_mode;
1680 entry.mask = 0; /* don't mask IRQ for edge */
1681 entry.dest = apic->cpu_mask_to_apicid(apic->target_cpus());
1682 entry.delivery_mode = apic->irq_delivery_mode;
1685 entry.vector = vector;
1688 * The timer IRQ doesn't have to know that behind the
1689 * scene we may have a 8259A-master in AEOI mode ...
1691 set_irq_chip_and_handler_name(0, &ioapic_chip, handle_edge_irq, "edge");
1694 * Add it to the IO-APIC irq-routing table:
1696 ioapic_write_entry(apic_id, pin, entry);
1700 __apicdebuginit(void) print_IO_APIC(void)
1703 union IO_APIC_reg_00 reg_00;
1704 union IO_APIC_reg_01 reg_01;
1705 union IO_APIC_reg_02 reg_02;
1706 union IO_APIC_reg_03 reg_03;
1707 unsigned long flags;
1708 struct irq_cfg *cfg;
1709 struct irq_desc *desc;
1712 if (apic_verbosity == APIC_QUIET)
1715 printk(KERN_DEBUG "number of MP IRQ sources: %d.\n", mp_irq_entries);
1716 for (i = 0; i < nr_ioapics; i++)
1717 printk(KERN_DEBUG "number of IO-APIC #%d registers: %d.\n",
1718 mp_ioapics[i].apicid, nr_ioapic_registers[i]);
1721 * We are a bit conservative about what we expect. We have to
1722 * know about every hardware change ASAP.
1724 printk(KERN_INFO "testing the IO APIC.......................\n");
1726 for (apic = 0; apic < nr_ioapics; apic++) {
1728 spin_lock_irqsave(&ioapic_lock, flags);
1729 reg_00.raw = io_apic_read(apic, 0);
1730 reg_01.raw = io_apic_read(apic, 1);
1731 if (reg_01.bits.version >= 0x10)
1732 reg_02.raw = io_apic_read(apic, 2);
1733 if (reg_01.bits.version >= 0x20)
1734 reg_03.raw = io_apic_read(apic, 3);
1735 spin_unlock_irqrestore(&ioapic_lock, flags);
1738 printk(KERN_DEBUG "IO APIC #%d......\n", mp_ioapics[apic].apicid);
1739 printk(KERN_DEBUG ".... register #00: %08X\n", reg_00.raw);
1740 printk(KERN_DEBUG "....... : physical APIC id: %02X\n", reg_00.bits.ID);
1741 printk(KERN_DEBUG "....... : Delivery Type: %X\n", reg_00.bits.delivery_type);
1742 printk(KERN_DEBUG "....... : LTS : %X\n", reg_00.bits.LTS);
1744 printk(KERN_DEBUG ".... register #01: %08X\n", *(int *)®_01);
1745 printk(KERN_DEBUG "....... : max redirection entries: %04X\n", reg_01.bits.entries);
1747 printk(KERN_DEBUG "....... : PRQ implemented: %X\n", reg_01.bits.PRQ);
1748 printk(KERN_DEBUG "....... : IO APIC version: %04X\n", reg_01.bits.version);
1751 * Some Intel chipsets with IO APIC VERSION of 0x1? don't have reg_02,
1752 * but the value of reg_02 is read as the previous read register
1753 * value, so ignore it if reg_02 == reg_01.
1755 if (reg_01.bits.version >= 0x10 && reg_02.raw != reg_01.raw) {
1756 printk(KERN_DEBUG ".... register #02: %08X\n", reg_02.raw);
1757 printk(KERN_DEBUG "....... : arbitration: %02X\n", reg_02.bits.arbitration);
1761 * Some Intel chipsets with IO APIC VERSION of 0x2? don't have reg_02
1762 * or reg_03, but the value of reg_0[23] is read as the previous read
1763 * register value, so ignore it if reg_03 == reg_0[12].
1765 if (reg_01.bits.version >= 0x20 && reg_03.raw != reg_02.raw &&
1766 reg_03.raw != reg_01.raw) {
1767 printk(KERN_DEBUG ".... register #03: %08X\n", reg_03.raw);
1768 printk(KERN_DEBUG "....... : Boot DT : %X\n", reg_03.bits.boot_DT);
1771 printk(KERN_DEBUG ".... IRQ redirection table:\n");
1773 printk(KERN_DEBUG " NR Dst Mask Trig IRR Pol"
1774 " Stat Dmod Deli Vect: \n");
1776 for (i = 0; i <= reg_01.bits.entries; i++) {
1777 struct IO_APIC_route_entry entry;
1779 entry = ioapic_read_entry(apic, i);
1781 printk(KERN_DEBUG " %02x %03X ",
1786 printk("%1d %1d %1d %1d %1d %1d %1d %02X\n",
1791 entry.delivery_status,
1793 entry.delivery_mode,
1798 printk(KERN_DEBUG "IRQ to pin mappings:\n");
1799 for_each_irq_desc(irq, desc) {
1800 struct irq_pin_list *entry;
1802 cfg = desc->chip_data;
1803 entry = cfg->irq_2_pin;
1806 printk(KERN_DEBUG "IRQ%d ", irq);
1808 printk("-> %d:%d", entry->apic, entry->pin);
1811 entry = entry->next;
1816 printk(KERN_INFO ".................................... done.\n");
1821 __apicdebuginit(void) print_APIC_bitfield(int base)
1826 if (apic_verbosity == APIC_QUIET)
1829 printk(KERN_DEBUG "0123456789abcdef0123456789abcdef\n" KERN_DEBUG);
1830 for (i = 0; i < 8; i++) {
1831 v = apic_read(base + i*0x10);
1832 for (j = 0; j < 32; j++) {
1842 __apicdebuginit(void) print_local_APIC(void *dummy)
1844 unsigned int v, ver, maxlvt;
1847 if (apic_verbosity == APIC_QUIET)
1850 printk("\n" KERN_DEBUG "printing local APIC contents on CPU#%d/%d:\n",
1851 smp_processor_id(), hard_smp_processor_id());
1852 v = apic_read(APIC_ID);
1853 printk(KERN_INFO "... APIC ID: %08x (%01x)\n", v, read_apic_id());
1854 v = apic_read(APIC_LVR);
1855 printk(KERN_INFO "... APIC VERSION: %08x\n", v);
1856 ver = GET_APIC_VERSION(v);
1857 maxlvt = lapic_get_maxlvt();
1859 v = apic_read(APIC_TASKPRI);
1860 printk(KERN_DEBUG "... APIC TASKPRI: %08x (%02x)\n", v, v & APIC_TPRI_MASK);
1862 if (APIC_INTEGRATED(ver)) { /* !82489DX */
1863 if (!APIC_XAPIC(ver)) {
1864 v = apic_read(APIC_ARBPRI);
1865 printk(KERN_DEBUG "... APIC ARBPRI: %08x (%02x)\n", v,
1866 v & APIC_ARBPRI_MASK);
1868 v = apic_read(APIC_PROCPRI);
1869 printk(KERN_DEBUG "... APIC PROCPRI: %08x\n", v);
1873 * Remote read supported only in the 82489DX and local APIC for
1874 * Pentium processors.
1876 if (!APIC_INTEGRATED(ver) || maxlvt == 3) {
1877 v = apic_read(APIC_RRR);
1878 printk(KERN_DEBUG "... APIC RRR: %08x\n", v);
1881 v = apic_read(APIC_LDR);
1882 printk(KERN_DEBUG "... APIC LDR: %08x\n", v);
1883 if (!x2apic_enabled()) {
1884 v = apic_read(APIC_DFR);
1885 printk(KERN_DEBUG "... APIC DFR: %08x\n", v);
1887 v = apic_read(APIC_SPIV);
1888 printk(KERN_DEBUG "... APIC SPIV: %08x\n", v);
1890 printk(KERN_DEBUG "... APIC ISR field:\n");
1891 print_APIC_bitfield(APIC_ISR);
1892 printk(KERN_DEBUG "... APIC TMR field:\n");
1893 print_APIC_bitfield(APIC_TMR);
1894 printk(KERN_DEBUG "... APIC IRR field:\n");
1895 print_APIC_bitfield(APIC_IRR);
1897 if (APIC_INTEGRATED(ver)) { /* !82489DX */
1898 if (maxlvt > 3) /* Due to the Pentium erratum 3AP. */
1899 apic_write(APIC_ESR, 0);
1901 v = apic_read(APIC_ESR);
1902 printk(KERN_DEBUG "... APIC ESR: %08x\n", v);
1905 icr = apic_icr_read();
1906 printk(KERN_DEBUG "... APIC ICR: %08x\n", (u32)icr);
1907 printk(KERN_DEBUG "... APIC ICR2: %08x\n", (u32)(icr >> 32));
1909 v = apic_read(APIC_LVTT);
1910 printk(KERN_DEBUG "... APIC LVTT: %08x\n", v);
1912 if (maxlvt > 3) { /* PC is LVT#4. */
1913 v = apic_read(APIC_LVTPC);
1914 printk(KERN_DEBUG "... APIC LVTPC: %08x\n", v);
1916 v = apic_read(APIC_LVT0);
1917 printk(KERN_DEBUG "... APIC LVT0: %08x\n", v);
1918 v = apic_read(APIC_LVT1);
1919 printk(KERN_DEBUG "... APIC LVT1: %08x\n", v);
1921 if (maxlvt > 2) { /* ERR is LVT#3. */
1922 v = apic_read(APIC_LVTERR);
1923 printk(KERN_DEBUG "... APIC LVTERR: %08x\n", v);
1926 v = apic_read(APIC_TMICT);
1927 printk(KERN_DEBUG "... APIC TMICT: %08x\n", v);
1928 v = apic_read(APIC_TMCCT);
1929 printk(KERN_DEBUG "... APIC TMCCT: %08x\n", v);
1930 v = apic_read(APIC_TDCR);
1931 printk(KERN_DEBUG "... APIC TDCR: %08x\n", v);
1935 __apicdebuginit(void) print_all_local_APICs(void)
1940 for_each_online_cpu(cpu)
1941 smp_call_function_single(cpu, print_local_APIC, NULL, 1);
1945 __apicdebuginit(void) print_PIC(void)
1948 unsigned long flags;
1950 if (apic_verbosity == APIC_QUIET)
1953 printk(KERN_DEBUG "\nprinting PIC contents\n");
1955 spin_lock_irqsave(&i8259A_lock, flags);
1957 v = inb(0xa1) << 8 | inb(0x21);
1958 printk(KERN_DEBUG "... PIC IMR: %04x\n", v);
1960 v = inb(0xa0) << 8 | inb(0x20);
1961 printk(KERN_DEBUG "... PIC IRR: %04x\n", v);
1965 v = inb(0xa0) << 8 | inb(0x20);
1969 spin_unlock_irqrestore(&i8259A_lock, flags);
1971 printk(KERN_DEBUG "... PIC ISR: %04x\n", v);
1973 v = inb(0x4d1) << 8 | inb(0x4d0);
1974 printk(KERN_DEBUG "... PIC ELCR: %04x\n", v);
1977 __apicdebuginit(int) print_all_ICs(void)
1980 print_all_local_APICs();
1986 fs_initcall(print_all_ICs);
1989 /* Where if anywhere is the i8259 connect in external int mode */
1990 static struct { int pin, apic; } ioapic_i8259 = { -1, -1 };
1992 void __init enable_IO_APIC(void)
1994 union IO_APIC_reg_01 reg_01;
1995 int i8259_apic, i8259_pin;
1997 unsigned long flags;
2000 * The number of IO-APIC IRQ registers (== #pins):
2002 for (apic = 0; apic < nr_ioapics; apic++) {
2003 spin_lock_irqsave(&ioapic_lock, flags);
2004 reg_01.raw = io_apic_read(apic, 1);
2005 spin_unlock_irqrestore(&ioapic_lock, flags);
2006 nr_ioapic_registers[apic] = reg_01.bits.entries+1;
2008 for(apic = 0; apic < nr_ioapics; apic++) {
2010 /* See if any of the pins is in ExtINT mode */
2011 for (pin = 0; pin < nr_ioapic_registers[apic]; pin++) {
2012 struct IO_APIC_route_entry entry;
2013 entry = ioapic_read_entry(apic, pin);
2015 /* If the interrupt line is enabled and in ExtInt mode
2016 * I have found the pin where the i8259 is connected.
2018 if ((entry.mask == 0) && (entry.delivery_mode == dest_ExtINT)) {
2019 ioapic_i8259.apic = apic;
2020 ioapic_i8259.pin = pin;
2026 /* Look to see what if the MP table has reported the ExtINT */
2027 /* If we could not find the appropriate pin by looking at the ioapic
2028 * the i8259 probably is not connected the ioapic but give the
2029 * mptable a chance anyway.
2031 i8259_pin = find_isa_irq_pin(0, mp_ExtINT);
2032 i8259_apic = find_isa_irq_apic(0, mp_ExtINT);
2033 /* Trust the MP table if nothing is setup in the hardware */
2034 if ((ioapic_i8259.pin == -1) && (i8259_pin >= 0)) {
2035 printk(KERN_WARNING "ExtINT not setup in hardware but reported by MP table\n");
2036 ioapic_i8259.pin = i8259_pin;
2037 ioapic_i8259.apic = i8259_apic;
2039 /* Complain if the MP table and the hardware disagree */
2040 if (((ioapic_i8259.apic != i8259_apic) || (ioapic_i8259.pin != i8259_pin)) &&
2041 (i8259_pin >= 0) && (ioapic_i8259.pin >= 0))
2043 printk(KERN_WARNING "ExtINT in hardware and MP table differ\n");
2047 * Do not trust the IO-APIC being empty at bootup
2053 * Not an __init, needed by the reboot code
2055 void disable_IO_APIC(void)
2058 * Clear the IO-APIC before rebooting:
2063 * If the i8259 is routed through an IOAPIC
2064 * Put that IOAPIC in virtual wire mode
2065 * so legacy interrupts can be delivered.
2067 * With interrupt-remapping, for now we will use virtual wire A mode,
2068 * as virtual wire B is little complex (need to configure both
2069 * IOAPIC RTE aswell as interrupt-remapping table entry).
2070 * As this gets called during crash dump, keep this simple for now.
2072 if (ioapic_i8259.pin != -1 && !intr_remapping_enabled) {
2073 struct IO_APIC_route_entry entry;
2075 memset(&entry, 0, sizeof(entry));
2076 entry.mask = 0; /* Enabled */
2077 entry.trigger = 0; /* Edge */
2079 entry.polarity = 0; /* High */
2080 entry.delivery_status = 0;
2081 entry.dest_mode = 0; /* Physical */
2082 entry.delivery_mode = dest_ExtINT; /* ExtInt */
2084 entry.dest = read_apic_id();
2087 * Add it to the IO-APIC irq-routing table:
2089 ioapic_write_entry(ioapic_i8259.apic, ioapic_i8259.pin, entry);
2093 * Use virtual wire A mode when interrupt remapping is enabled.
2095 disconnect_bsp_APIC(!intr_remapping_enabled && ioapic_i8259.pin != -1);
2098 #ifdef CONFIG_X86_32
2100 * function to set the IO-APIC physical IDs based on the
2101 * values stored in the MPC table.
2103 * by Matt Domsch <Matt_Domsch@dell.com> Tue Dec 21 12:25:05 CST 1999
2106 static void __init setup_ioapic_ids_from_mpc(void)
2108 union IO_APIC_reg_00 reg_00;
2109 physid_mask_t phys_id_present_map;
2112 unsigned char old_id;
2113 unsigned long flags;
2115 if (x86_quirks->setup_ioapic_ids && x86_quirks->setup_ioapic_ids())
2119 * Don't check I/O APIC IDs for xAPIC systems. They have
2120 * no meaning without the serial APIC bus.
2122 if (!(boot_cpu_data.x86_vendor == X86_VENDOR_INTEL)
2123 || APIC_XAPIC(apic_version[boot_cpu_physical_apicid]))
2126 * This is broken; anything with a real cpu count has to
2127 * circumvent this idiocy regardless.
2129 phys_id_present_map = apic->ioapic_phys_id_map(phys_cpu_present_map);
2132 * Set the IOAPIC ID to the value stored in the MPC table.
2134 for (apic_id = 0; apic_id < nr_ioapics; apic_id++) {
2136 /* Read the register 0 value */
2137 spin_lock_irqsave(&ioapic_lock, flags);
2138 reg_00.raw = io_apic_read(apic_id, 0);
2139 spin_unlock_irqrestore(&ioapic_lock, flags);
2141 old_id = mp_ioapics[apic_id].apicid;
2143 if (mp_ioapics[apic_id].apicid >= get_physical_broadcast()) {
2144 printk(KERN_ERR "BIOS bug, IO-APIC#%d ID is %d in the MPC table!...\n",
2145 apic_id, mp_ioapics[apic_id].apicid);
2146 printk(KERN_ERR "... fixing up to %d. (tell your hw vendor)\n",
2148 mp_ioapics[apic_id].apicid = reg_00.bits.ID;
2152 * Sanity check, is the ID really free? Every APIC in a
2153 * system must have a unique ID or we get lots of nice
2154 * 'stuck on smp_invalidate_needed IPI wait' messages.
2156 if (apic->check_apicid_used(phys_id_present_map,
2157 mp_ioapics[apic_id].apicid)) {
2158 printk(KERN_ERR "BIOS bug, IO-APIC#%d ID %d is already used!...\n",
2159 apic_id, mp_ioapics[apic_id].apicid);
2160 for (i = 0; i < get_physical_broadcast(); i++)
2161 if (!physid_isset(i, phys_id_present_map))
2163 if (i >= get_physical_broadcast())
2164 panic("Max APIC ID exceeded!\n");
2165 printk(KERN_ERR "... fixing up to %d. (tell your hw vendor)\n",
2167 physid_set(i, phys_id_present_map);
2168 mp_ioapics[apic_id].apicid = i;
2171 tmp = apic->apicid_to_cpu_present(mp_ioapics[apic_id].apicid);
2172 apic_printk(APIC_VERBOSE, "Setting %d in the "
2173 "phys_id_present_map\n",
2174 mp_ioapics[apic_id].apicid);
2175 physids_or(phys_id_present_map, phys_id_present_map, tmp);
2180 * We need to adjust the IRQ routing table
2181 * if the ID changed.
2183 if (old_id != mp_ioapics[apic_id].apicid)
2184 for (i = 0; i < mp_irq_entries; i++)
2185 if (mp_irqs[i].dstapic == old_id)
2187 = mp_ioapics[apic_id].apicid;
2190 * Read the right value from the MPC table and
2191 * write it into the ID register.
2193 apic_printk(APIC_VERBOSE, KERN_INFO
2194 "...changing IO-APIC physical APIC ID to %d ...",
2195 mp_ioapics[apic_id].apicid);
2197 reg_00.bits.ID = mp_ioapics[apic_id].apicid;
2198 spin_lock_irqsave(&ioapic_lock, flags);
2199 io_apic_write(apic_id, 0, reg_00.raw);
2200 spin_unlock_irqrestore(&ioapic_lock, flags);
2205 spin_lock_irqsave(&ioapic_lock, flags);
2206 reg_00.raw = io_apic_read(apic_id, 0);
2207 spin_unlock_irqrestore(&ioapic_lock, flags);
2208 if (reg_00.bits.ID != mp_ioapics[apic_id].apicid)
2209 printk("could not set ID!\n");
2211 apic_printk(APIC_VERBOSE, " ok.\n");
2216 int no_timer_check __initdata;
2218 static int __init notimercheck(char *s)
2223 __setup("no_timer_check", notimercheck);
2226 * There is a nasty bug in some older SMP boards, their mptable lies
2227 * about the timer IRQ. We do the following to work around the situation:
2229 * - timer IRQ defaults to IO-APIC IRQ
2230 * - if this function detects that timer IRQs are defunct, then we fall
2231 * back to ISA timer IRQs
2233 static int __init timer_irq_works(void)
2235 unsigned long t1 = jiffies;
2236 unsigned long flags;
2241 local_save_flags(flags);
2243 /* Let ten ticks pass... */
2244 mdelay((10 * 1000) / HZ);
2245 local_irq_restore(flags);
2248 * Expect a few ticks at least, to be sure some possible
2249 * glue logic does not lock up after one or two first
2250 * ticks in a non-ExtINT mode. Also the local APIC
2251 * might have cached one ExtINT interrupt. Finally, at
2252 * least one tick may be lost due to delays.
2256 if (time_after(jiffies, t1 + 4))
2262 * In the SMP+IOAPIC case it might happen that there are an unspecified
2263 * number of pending IRQ events unhandled. These cases are very rare,
2264 * so we 'resend' these IRQs via IPIs, to the same CPU. It's much
2265 * better to do it this way as thus we do not have to be aware of
2266 * 'pending' interrupts in the IRQ path, except at this point.
2269 * Edge triggered needs to resend any interrupt
2270 * that was delayed but this is now handled in the device
2275 * Starting up a edge-triggered IO-APIC interrupt is
2276 * nasty - we need to make sure that we get the edge.
2277 * If it is already asserted for some reason, we need
2278 * return 1 to indicate that is was pending.
2280 * This is not complete - we should be able to fake
2281 * an edge even if it isn't on the 8259A...
2284 static unsigned int startup_ioapic_irq(unsigned int irq)
2286 int was_pending = 0;
2287 unsigned long flags;
2288 struct irq_cfg *cfg;
2290 spin_lock_irqsave(&ioapic_lock, flags);
2291 if (irq < NR_IRQS_LEGACY) {
2292 disable_8259A_irq(irq);
2293 if (i8259A_irq_pending(irq))
2297 __unmask_IO_APIC_irq(cfg);
2298 spin_unlock_irqrestore(&ioapic_lock, flags);
2303 #ifdef CONFIG_X86_64
2304 static int ioapic_retrigger_irq(unsigned int irq)
2307 struct irq_cfg *cfg = irq_cfg(irq);
2308 unsigned long flags;
2310 spin_lock_irqsave(&vector_lock, flags);
2311 apic->send_IPI_mask(cpumask_of(cpumask_first(cfg->domain)), cfg->vector);
2312 spin_unlock_irqrestore(&vector_lock, flags);
2317 static int ioapic_retrigger_irq(unsigned int irq)
2319 apic->send_IPI_self(irq_cfg(irq)->vector);
2326 * Level and edge triggered IO-APIC interrupts need different handling,
2327 * so we use two separate IRQ descriptors. Edge triggered IRQs can be
2328 * handled with the level-triggered descriptor, but that one has slightly
2329 * more overhead. Level-triggered interrupts cannot be handled with the
2330 * edge-triggered handler, without risking IRQ storms and other ugly
2336 #ifdef CONFIG_INTR_REMAP
2339 * Migrate the IO-APIC irq in the presence of intr-remapping.
2341 * For both level and edge triggered, irq migration is a simple atomic
2342 * update(of vector and cpu destination) of IRTE and flush the hardware cache.
2344 * For level triggered, we eliminate the io-apic RTE modification (with the
2345 * updated vector information), by using a virtual vector (io-apic pin number).
2346 * Real vector that is used for interrupting cpu will be coming from
2347 * the interrupt-remapping table entry.
2350 migrate_ioapic_irq_desc(struct irq_desc *desc, const struct cpumask *mask)
2352 struct irq_cfg *cfg;
2357 if (!cpumask_intersects(mask, cpu_online_mask))
2361 if (get_irte(irq, &irte))
2364 cfg = desc->chip_data;
2365 if (assign_irq_vector(irq, cfg, mask))
2368 set_extra_move_desc(desc, mask);
2370 dest = apic->cpu_mask_to_apicid_and(cfg->domain, mask);
2372 irte.vector = cfg->vector;
2373 irte.dest_id = IRTE_DEST(dest);
2376 * Modified the IRTE and flushes the Interrupt entry cache.
2378 modify_irte(irq, &irte);
2380 if (cfg->move_in_progress)
2381 send_cleanup_vector(cfg);
2383 cpumask_copy(desc->affinity, mask);
2387 * Migrates the IRQ destination in the process context.
2389 static void set_ir_ioapic_affinity_irq_desc(struct irq_desc *desc,
2390 const struct cpumask *mask)
2392 migrate_ioapic_irq_desc(desc, mask);
2394 static void set_ir_ioapic_affinity_irq(unsigned int irq,
2395 const struct cpumask *mask)
2397 struct irq_desc *desc = irq_to_desc(irq);
2399 set_ir_ioapic_affinity_irq_desc(desc, mask);
2402 static inline void set_ir_ioapic_affinity_irq_desc(struct irq_desc *desc,
2403 const struct cpumask *mask)
2408 asmlinkage void smp_irq_move_cleanup_interrupt(void)
2410 unsigned vector, me;
2416 me = smp_processor_id();
2417 for (vector = FIRST_EXTERNAL_VECTOR; vector < NR_VECTORS; vector++) {
2420 struct irq_desc *desc;
2421 struct irq_cfg *cfg;
2422 irq = __get_cpu_var(vector_irq)[vector];
2427 desc = irq_to_desc(irq);
2432 spin_lock(&desc->lock);
2433 if (!cfg->move_cleanup_count)
2436 if (vector == cfg->vector && cpumask_test_cpu(me, cfg->domain))
2439 irr = apic_read(APIC_IRR + (vector / 32 * 0x10));
2441 * Check if the vector that needs to be cleanedup is
2442 * registered at the cpu's IRR. If so, then this is not
2443 * the best time to clean it up. Lets clean it up in the
2444 * next attempt by sending another IRQ_MOVE_CLEANUP_VECTOR
2447 if (irr & (1 << (vector % 32))) {
2448 apic->send_IPI_self(IRQ_MOVE_CLEANUP_VECTOR);
2451 __get_cpu_var(vector_irq)[vector] = -1;
2452 cfg->move_cleanup_count--;
2454 spin_unlock(&desc->lock);
2460 static void irq_complete_move(struct irq_desc **descp)
2462 struct irq_desc *desc = *descp;
2463 struct irq_cfg *cfg = desc->chip_data;
2464 unsigned vector, me;
2466 if (likely(!cfg->move_in_progress)) {
2467 #ifdef CONFIG_NUMA_MIGRATE_IRQ_DESC
2468 if (likely(!cfg->move_desc_pending))
2471 /* domain has not changed, but affinity did */
2472 me = smp_processor_id();
2473 if (cpumask_test_cpu(me, desc->affinity)) {
2474 *descp = desc = move_irq_desc(desc, me);
2475 /* get the new one */
2476 cfg = desc->chip_data;
2477 cfg->move_desc_pending = 0;
2483 vector = ~get_irq_regs()->orig_ax;
2484 me = smp_processor_id();
2486 if (vector == cfg->vector && cpumask_test_cpu(me, cfg->domain)) {
2487 #ifdef CONFIG_NUMA_MIGRATE_IRQ_DESC
2488 *descp = desc = move_irq_desc(desc, me);
2489 /* get the new one */
2490 cfg = desc->chip_data;
2492 send_cleanup_vector(cfg);
2496 static inline void irq_complete_move(struct irq_desc **descp) {}
2499 #ifdef CONFIG_INTR_REMAP
2500 static void __eoi_ioapic_irq(unsigned int irq, struct irq_cfg *cfg)
2503 struct irq_pin_list *entry;
2505 entry = cfg->irq_2_pin;
2513 io_apic_eoi(apic, pin);
2514 entry = entry->next;
2519 eoi_ioapic_irq(struct irq_desc *desc)
2521 struct irq_cfg *cfg;
2522 unsigned long flags;
2526 cfg = desc->chip_data;
2528 spin_lock_irqsave(&ioapic_lock, flags);
2529 __eoi_ioapic_irq(irq, cfg);
2530 spin_unlock_irqrestore(&ioapic_lock, flags);
2533 static void ack_x2apic_level(unsigned int irq)
2535 struct irq_desc *desc = irq_to_desc(irq);
2537 eoi_ioapic_irq(desc);
2540 static void ack_x2apic_edge(unsigned int irq)
2547 static void ack_apic_edge(unsigned int irq)
2549 struct irq_desc *desc = irq_to_desc(irq);
2551 irq_complete_move(&desc);
2552 move_native_irq(irq);
2556 atomic_t irq_mis_count;
2558 static void ack_apic_level(unsigned int irq)
2560 struct irq_desc *desc = irq_to_desc(irq);
2562 #ifdef CONFIG_X86_32
2566 struct irq_cfg *cfg;
2567 int do_unmask_irq = 0;
2569 irq_complete_move(&desc);
2570 #ifdef CONFIG_GENERIC_PENDING_IRQ
2571 /* If we are moving the irq we need to mask it */
2572 if (unlikely(desc->status & IRQ_MOVE_PENDING)) {
2574 mask_IO_APIC_irq_desc(desc);
2578 #ifdef CONFIG_X86_32
2580 * It appears there is an erratum which affects at least version 0x11
2581 * of I/O APIC (that's the 82093AA and cores integrated into various
2582 * chipsets). Under certain conditions a level-triggered interrupt is
2583 * erroneously delivered as edge-triggered one but the respective IRR
2584 * bit gets set nevertheless. As a result the I/O unit expects an EOI
2585 * message but it will never arrive and further interrupts are blocked
2586 * from the source. The exact reason is so far unknown, but the
2587 * phenomenon was observed when two consecutive interrupt requests
2588 * from a given source get delivered to the same CPU and the source is
2589 * temporarily disabled in between.
2591 * A workaround is to simulate an EOI message manually. We achieve it
2592 * by setting the trigger mode to edge and then to level when the edge
2593 * trigger mode gets detected in the TMR of a local APIC for a
2594 * level-triggered interrupt. We mask the source for the time of the
2595 * operation to prevent an edge-triggered interrupt escaping meanwhile.
2596 * The idea is from Manfred Spraul. --macro
2598 cfg = desc->chip_data;
2601 v = apic_read(APIC_TMR + ((i & ~0x1f) >> 1));
2605 * We must acknowledge the irq before we move it or the acknowledge will
2606 * not propagate properly.
2610 /* Now we can move and renable the irq */
2611 if (unlikely(do_unmask_irq)) {
2612 /* Only migrate the irq if the ack has been received.
2614 * On rare occasions the broadcast level triggered ack gets
2615 * delayed going to ioapics, and if we reprogram the
2616 * vector while Remote IRR is still set the irq will never
2619 * To prevent this scenario we read the Remote IRR bit
2620 * of the ioapic. This has two effects.
2621 * - On any sane system the read of the ioapic will
2622 * flush writes (and acks) going to the ioapic from
2624 * - We get to see if the ACK has actually been delivered.
2626 * Based on failed experiments of reprogramming the
2627 * ioapic entry from outside of irq context starting
2628 * with masking the ioapic entry and then polling until
2629 * Remote IRR was clear before reprogramming the
2630 * ioapic I don't trust the Remote IRR bit to be
2631 * completey accurate.
2633 * However there appears to be no other way to plug
2634 * this race, so if the Remote IRR bit is not
2635 * accurate and is causing problems then it is a hardware bug
2636 * and you can go talk to the chipset vendor about it.
2638 cfg = desc->chip_data;
2639 if (!io_apic_level_ack_pending(cfg))
2640 move_masked_irq(irq);
2641 unmask_IO_APIC_irq_desc(desc);
2644 #ifdef CONFIG_X86_32
2645 if (!(v & (1 << (i & 0x1f)))) {
2646 atomic_inc(&irq_mis_count);
2647 spin_lock(&ioapic_lock);
2648 __mask_and_edge_IO_APIC_irq(cfg);
2649 __unmask_and_level_IO_APIC_irq(cfg);
2650 spin_unlock(&ioapic_lock);
2655 static struct irq_chip ioapic_chip __read_mostly = {
2657 .startup = startup_ioapic_irq,
2658 .mask = mask_IO_APIC_irq,
2659 .unmask = unmask_IO_APIC_irq,
2660 .ack = ack_apic_edge,
2661 .eoi = ack_apic_level,
2663 .set_affinity = set_ioapic_affinity_irq,
2665 .retrigger = ioapic_retrigger_irq,
2668 #ifdef CONFIG_INTR_REMAP
2669 static struct irq_chip ir_ioapic_chip __read_mostly = {
2670 .name = "IR-IO-APIC",
2671 .startup = startup_ioapic_irq,
2672 .mask = mask_IO_APIC_irq,
2673 .unmask = unmask_IO_APIC_irq,
2674 .ack = ack_x2apic_edge,
2675 .eoi = ack_x2apic_level,
2677 .set_affinity = set_ir_ioapic_affinity_irq,
2679 .retrigger = ioapic_retrigger_irq,
2683 static inline void init_IO_APIC_traps(void)
2686 struct irq_desc *desc;
2687 struct irq_cfg *cfg;
2690 * NOTE! The local APIC isn't very good at handling
2691 * multiple interrupts at the same interrupt level.
2692 * As the interrupt level is determined by taking the
2693 * vector number and shifting that right by 4, we
2694 * want to spread these out a bit so that they don't
2695 * all fall in the same interrupt level.
2697 * Also, we've got to be careful not to trash gate
2698 * 0x80, because int 0x80 is hm, kind of importantish. ;)
2700 for_each_irq_desc(irq, desc) {
2701 cfg = desc->chip_data;
2702 if (IO_APIC_IRQ(irq) && cfg && !cfg->vector) {
2704 * Hmm.. We don't have an entry for this,
2705 * so default to an old-fashioned 8259
2706 * interrupt if we can..
2708 if (irq < NR_IRQS_LEGACY)
2709 make_8259A_irq(irq);
2711 /* Strange. Oh, well.. */
2712 desc->chip = &no_irq_chip;
2718 * The local APIC irq-chip implementation:
2721 static void mask_lapic_irq(unsigned int irq)
2725 v = apic_read(APIC_LVT0);
2726 apic_write(APIC_LVT0, v | APIC_LVT_MASKED);
2729 static void unmask_lapic_irq(unsigned int irq)
2733 v = apic_read(APIC_LVT0);
2734 apic_write(APIC_LVT0, v & ~APIC_LVT_MASKED);
2737 static void ack_lapic_irq(unsigned int irq)
2742 static struct irq_chip lapic_chip __read_mostly = {
2743 .name = "local-APIC",
2744 .mask = mask_lapic_irq,
2745 .unmask = unmask_lapic_irq,
2746 .ack = ack_lapic_irq,
2749 static void lapic_register_intr(int irq, struct irq_desc *desc)
2751 desc->status &= ~IRQ_LEVEL;
2752 set_irq_chip_and_handler_name(irq, &lapic_chip, handle_edge_irq,
2756 static void __init setup_nmi(void)
2759 * Dirty trick to enable the NMI watchdog ...
2760 * We put the 8259A master into AEOI mode and
2761 * unmask on all local APICs LVT0 as NMI.
2763 * The idea to use the 8259A in AEOI mode ('8259A Virtual Wire')
2764 * is from Maciej W. Rozycki - so we do not have to EOI from
2765 * the NMI handler or the timer interrupt.
2767 apic_printk(APIC_VERBOSE, KERN_INFO "activating NMI Watchdog ...");
2769 enable_NMI_through_LVT0();
2771 apic_printk(APIC_VERBOSE, " done.\n");
2775 * This looks a bit hackish but it's about the only one way of sending
2776 * a few INTA cycles to 8259As and any associated glue logic. ICR does
2777 * not support the ExtINT mode, unfortunately. We need to send these
2778 * cycles as some i82489DX-based boards have glue logic that keeps the
2779 * 8259A interrupt line asserted until INTA. --macro
2781 static inline void __init unlock_ExtINT_logic(void)
2784 struct IO_APIC_route_entry entry0, entry1;
2785 unsigned char save_control, save_freq_select;
2787 pin = find_isa_irq_pin(8, mp_INT);
2792 apic = find_isa_irq_apic(8, mp_INT);
2798 entry0 = ioapic_read_entry(apic, pin);
2799 clear_IO_APIC_pin(apic, pin);
2801 memset(&entry1, 0, sizeof(entry1));
2803 entry1.dest_mode = 0; /* physical delivery */
2804 entry1.mask = 0; /* unmask IRQ now */
2805 entry1.dest = hard_smp_processor_id();
2806 entry1.delivery_mode = dest_ExtINT;
2807 entry1.polarity = entry0.polarity;
2811 ioapic_write_entry(apic, pin, entry1);
2813 save_control = CMOS_READ(RTC_CONTROL);
2814 save_freq_select = CMOS_READ(RTC_FREQ_SELECT);
2815 CMOS_WRITE((save_freq_select & ~RTC_RATE_SELECT) | 0x6,
2817 CMOS_WRITE(save_control | RTC_PIE, RTC_CONTROL);
2822 if ((CMOS_READ(RTC_INTR_FLAGS) & RTC_PF) == RTC_PF)
2826 CMOS_WRITE(save_control, RTC_CONTROL);
2827 CMOS_WRITE(save_freq_select, RTC_FREQ_SELECT);
2828 clear_IO_APIC_pin(apic, pin);
2830 ioapic_write_entry(apic, pin, entry0);
2833 static int disable_timer_pin_1 __initdata;
2834 /* Actually the next is obsolete, but keep it for paranoid reasons -AK */
2835 static int __init disable_timer_pin_setup(char *arg)
2837 disable_timer_pin_1 = 1;
2840 early_param("disable_timer_pin_1", disable_timer_pin_setup);
2842 int timer_through_8259 __initdata;
2845 * This code may look a bit paranoid, but it's supposed to cooperate with
2846 * a wide range of boards and BIOS bugs. Fortunately only the timer IRQ
2847 * is so screwy. Thanks to Brian Perkins for testing/hacking this beast
2848 * fanatically on his truly buggy board.
2850 * FIXME: really need to revamp this for all platforms.
2852 static inline void __init check_timer(void)
2854 struct irq_desc *desc = irq_to_desc(0);
2855 struct irq_cfg *cfg = desc->chip_data;
2856 int cpu = boot_cpu_id;
2857 int apic1, pin1, apic2, pin2;
2858 unsigned long flags;
2861 local_irq_save(flags);
2864 * get/set the timer IRQ vector:
2866 disable_8259A_irq(0);
2867 assign_irq_vector(0, cfg, apic->target_cpus());
2870 * As IRQ0 is to be enabled in the 8259A, the virtual
2871 * wire has to be disabled in the local APIC. Also
2872 * timer interrupts need to be acknowledged manually in
2873 * the 8259A for the i82489DX when using the NMI
2874 * watchdog as that APIC treats NMIs as level-triggered.
2875 * The AEOI mode will finish them in the 8259A
2878 apic_write(APIC_LVT0, APIC_LVT_MASKED | APIC_DM_EXTINT);
2880 #ifdef CONFIG_X86_32
2884 ver = apic_read(APIC_LVR);
2885 ver = GET_APIC_VERSION(ver);
2886 timer_ack = (nmi_watchdog == NMI_IO_APIC && !APIC_INTEGRATED(ver));
2890 pin1 = find_isa_irq_pin(0, mp_INT);
2891 apic1 = find_isa_irq_apic(0, mp_INT);
2892 pin2 = ioapic_i8259.pin;
2893 apic2 = ioapic_i8259.apic;
2895 apic_printk(APIC_QUIET, KERN_INFO "..TIMER: vector=0x%02X "
2896 "apic1=%d pin1=%d apic2=%d pin2=%d\n",
2897 cfg->vector, apic1, pin1, apic2, pin2);
2900 * Some BIOS writers are clueless and report the ExtINTA
2901 * I/O APIC input from the cascaded 8259A as the timer
2902 * interrupt input. So just in case, if only one pin
2903 * was found above, try it both directly and through the
2907 if (intr_remapping_enabled)
2908 panic("BIOS bug: timer not connected to IO-APIC");
2912 } else if (pin2 == -1) {
2919 * Ok, does IRQ0 through the IOAPIC work?
2922 add_pin_to_irq_cpu(cfg, cpu, apic1, pin1);
2923 setup_timer_IRQ0_pin(apic1, pin1, cfg->vector);
2925 /* for edge trigger, setup_IO_APIC_irq already
2926 * leave it unmasked.
2927 * so only need to unmask if it is level-trigger
2928 * do we really have level trigger timer?
2931 idx = find_irq_entry(apic1, pin1, mp_INT);
2932 if (idx != -1 && irq_trigger(idx))
2933 unmask_IO_APIC_irq_desc(desc);
2935 if (timer_irq_works()) {
2936 if (nmi_watchdog == NMI_IO_APIC) {
2938 enable_8259A_irq(0);
2940 if (disable_timer_pin_1 > 0)
2941 clear_IO_APIC_pin(0, pin1);
2944 if (intr_remapping_enabled)
2945 panic("timer doesn't work through Interrupt-remapped IO-APIC");
2946 local_irq_disable();
2947 clear_IO_APIC_pin(apic1, pin1);
2949 apic_printk(APIC_QUIET, KERN_ERR "..MP-BIOS bug: "
2950 "8254 timer not connected to IO-APIC\n");
2952 apic_printk(APIC_QUIET, KERN_INFO "...trying to set up timer "
2953 "(IRQ0) through the 8259A ...\n");
2954 apic_printk(APIC_QUIET, KERN_INFO
2955 "..... (found apic %d pin %d) ...\n", apic2, pin2);
2957 * legacy devices should be connected to IO APIC #0
2959 replace_pin_at_irq_cpu(cfg, cpu, apic1, pin1, apic2, pin2);
2960 setup_timer_IRQ0_pin(apic2, pin2, cfg->vector);
2961 enable_8259A_irq(0);
2962 if (timer_irq_works()) {
2963 apic_printk(APIC_QUIET, KERN_INFO "....... works.\n");
2964 timer_through_8259 = 1;
2965 if (nmi_watchdog == NMI_IO_APIC) {
2966 disable_8259A_irq(0);
2968 enable_8259A_irq(0);
2973 * Cleanup, just in case ...
2975 local_irq_disable();
2976 disable_8259A_irq(0);
2977 clear_IO_APIC_pin(apic2, pin2);
2978 apic_printk(APIC_QUIET, KERN_INFO "....... failed.\n");
2981 if (nmi_watchdog == NMI_IO_APIC) {
2982 apic_printk(APIC_QUIET, KERN_WARNING "timer doesn't work "
2983 "through the IO-APIC - disabling NMI Watchdog!\n");
2984 nmi_watchdog = NMI_NONE;
2986 #ifdef CONFIG_X86_32
2990 apic_printk(APIC_QUIET, KERN_INFO
2991 "...trying to set up timer as Virtual Wire IRQ...\n");
2993 lapic_register_intr(0, desc);
2994 apic_write(APIC_LVT0, APIC_DM_FIXED | cfg->vector); /* Fixed mode */
2995 enable_8259A_irq(0);
2997 if (timer_irq_works()) {
2998 apic_printk(APIC_QUIET, KERN_INFO "..... works.\n");
3001 local_irq_disable();
3002 disable_8259A_irq(0);
3003 apic_write(APIC_LVT0, APIC_LVT_MASKED | APIC_DM_FIXED | cfg->vector);
3004 apic_printk(APIC_QUIET, KERN_INFO "..... failed.\n");
3006 apic_printk(APIC_QUIET, KERN_INFO
3007 "...trying to set up timer as ExtINT IRQ...\n");
3011 apic_write(APIC_LVT0, APIC_DM_EXTINT);
3013 unlock_ExtINT_logic();
3015 if (timer_irq_works()) {
3016 apic_printk(APIC_QUIET, KERN_INFO "..... works.\n");
3019 local_irq_disable();
3020 apic_printk(APIC_QUIET, KERN_INFO "..... failed :(.\n");
3021 panic("IO-APIC + timer doesn't work! Boot with apic=debug and send a "
3022 "report. Then try booting with the 'noapic' option.\n");
3024 local_irq_restore(flags);
3028 * Traditionally ISA IRQ2 is the cascade IRQ, and is not available
3029 * to devices. However there may be an I/O APIC pin available for
3030 * this interrupt regardless. The pin may be left unconnected, but
3031 * typically it will be reused as an ExtINT cascade interrupt for
3032 * the master 8259A. In the MPS case such a pin will normally be
3033 * reported as an ExtINT interrupt in the MP table. With ACPI
3034 * there is no provision for ExtINT interrupts, and in the absence
3035 * of an override it would be treated as an ordinary ISA I/O APIC
3036 * interrupt, that is edge-triggered and unmasked by default. We
3037 * used to do this, but it caused problems on some systems because
3038 * of the NMI watchdog and sometimes IRQ0 of the 8254 timer using
3039 * the same ExtINT cascade interrupt to drive the local APIC of the
3040 * bootstrap processor. Therefore we refrain from routing IRQ2 to
3041 * the I/O APIC in all cases now. No actual device should request
3042 * it anyway. --macro
3044 #define PIC_IRQS (1 << PIC_CASCADE_IR)
3046 void __init setup_IO_APIC(void)
3050 * calling enable_IO_APIC() is moved to setup_local_APIC for BP
3053 io_apic_irqs = ~PIC_IRQS;
3055 apic_printk(APIC_VERBOSE, "ENABLING IO-APIC IRQs\n");
3057 * Set up IO-APIC IRQ routing.
3059 #ifdef CONFIG_X86_32
3061 setup_ioapic_ids_from_mpc();
3064 setup_IO_APIC_irqs();
3065 init_IO_APIC_traps();
3070 * Called after all the initialization is done. If we didnt find any
3071 * APIC bugs then we can allow the modify fast path
3074 static int __init io_apic_bug_finalize(void)
3076 if (sis_apic_bug == -1)
3081 late_initcall(io_apic_bug_finalize);
3083 struct sysfs_ioapic_data {
3084 struct sys_device dev;
3085 struct IO_APIC_route_entry entry[0];
3087 static struct sysfs_ioapic_data * mp_ioapic_data[MAX_IO_APICS];
3089 static int ioapic_suspend(struct sys_device *dev, pm_message_t state)
3091 struct IO_APIC_route_entry *entry;
3092 struct sysfs_ioapic_data *data;
3095 data = container_of(dev, struct sysfs_ioapic_data, dev);
3096 entry = data->entry;
3097 for (i = 0; i < nr_ioapic_registers[dev->id]; i ++, entry ++ )
3098 *entry = ioapic_read_entry(dev->id, i);
3103 static int ioapic_resume(struct sys_device *dev)
3105 struct IO_APIC_route_entry *entry;
3106 struct sysfs_ioapic_data *data;
3107 unsigned long flags;
3108 union IO_APIC_reg_00 reg_00;
3111 data = container_of(dev, struct sysfs_ioapic_data, dev);
3112 entry = data->entry;
3114 spin_lock_irqsave(&ioapic_lock, flags);
3115 reg_00.raw = io_apic_read(dev->id, 0);
3116 if (reg_00.bits.ID != mp_ioapics[dev->id].apicid) {
3117 reg_00.bits.ID = mp_ioapics[dev->id].apicid;
3118 io_apic_write(dev->id, 0, reg_00.raw);
3120 spin_unlock_irqrestore(&ioapic_lock, flags);
3121 for (i = 0; i < nr_ioapic_registers[dev->id]; i++)
3122 ioapic_write_entry(dev->id, i, entry[i]);
3127 static struct sysdev_class ioapic_sysdev_class = {
3129 .suspend = ioapic_suspend,
3130 .resume = ioapic_resume,
3133 static int __init ioapic_init_sysfs(void)
3135 struct sys_device * dev;
3138 error = sysdev_class_register(&ioapic_sysdev_class);
3142 for (i = 0; i < nr_ioapics; i++ ) {
3143 size = sizeof(struct sys_device) + nr_ioapic_registers[i]
3144 * sizeof(struct IO_APIC_route_entry);
3145 mp_ioapic_data[i] = kzalloc(size, GFP_KERNEL);
3146 if (!mp_ioapic_data[i]) {
3147 printk(KERN_ERR "Can't suspend/resume IOAPIC %d\n", i);
3150 dev = &mp_ioapic_data[i]->dev;
3152 dev->cls = &ioapic_sysdev_class;
3153 error = sysdev_register(dev);
3155 kfree(mp_ioapic_data[i]);
3156 mp_ioapic_data[i] = NULL;
3157 printk(KERN_ERR "Can't suspend/resume IOAPIC %d\n", i);
3165 device_initcall(ioapic_init_sysfs);
3167 static int nr_irqs_gsi = NR_IRQS_LEGACY;
3169 * Dynamic irq allocate and deallocation
3171 unsigned int create_irq_nr(unsigned int irq_want)
3173 /* Allocate an unused irq */
3176 unsigned long flags;
3177 struct irq_cfg *cfg_new = NULL;
3178 int cpu = boot_cpu_id;
3179 struct irq_desc *desc_new = NULL;
3182 if (irq_want < nr_irqs_gsi)
3183 irq_want = nr_irqs_gsi;
3185 spin_lock_irqsave(&vector_lock, flags);
3186 for (new = irq_want; new < nr_irqs; new++) {
3187 desc_new = irq_to_desc_alloc_cpu(new, cpu);
3189 printk(KERN_INFO "can not get irq_desc for %d\n", new);
3192 cfg_new = desc_new->chip_data;
3194 if (cfg_new->vector != 0)
3196 if (__assign_irq_vector(new, cfg_new, apic->target_cpus()) == 0)
3200 spin_unlock_irqrestore(&vector_lock, flags);
3203 dynamic_irq_init(irq);
3204 /* restore it, in case dynamic_irq_init clear it */
3206 desc_new->chip_data = cfg_new;
3211 int create_irq(void)
3213 unsigned int irq_want;
3216 irq_want = nr_irqs_gsi;
3217 irq = create_irq_nr(irq_want);
3225 void destroy_irq(unsigned int irq)
3227 unsigned long flags;
3228 struct irq_cfg *cfg;
3229 struct irq_desc *desc;
3231 /* store it, in case dynamic_irq_cleanup clear it */
3232 desc = irq_to_desc(irq);
3233 cfg = desc->chip_data;
3234 dynamic_irq_cleanup(irq);
3235 /* connect back irq_cfg */
3237 desc->chip_data = cfg;
3240 spin_lock_irqsave(&vector_lock, flags);
3241 __clear_irq_vector(irq, cfg);
3242 spin_unlock_irqrestore(&vector_lock, flags);
3246 * MSI message composition
3248 #ifdef CONFIG_PCI_MSI
3249 static int msi_compose_msg(struct pci_dev *pdev, unsigned int irq, struct msi_msg *msg)
3251 struct irq_cfg *cfg;
3259 err = assign_irq_vector(irq, cfg, apic->target_cpus());
3263 dest = apic->cpu_mask_to_apicid_and(cfg->domain, apic->target_cpus());
3265 if (irq_remapped(irq)) {
3270 ir_index = map_irq_to_irte_handle(irq, &sub_handle);
3271 BUG_ON(ir_index == -1);
3273 memset (&irte, 0, sizeof(irte));
3276 irte.dst_mode = apic->irq_dest_mode;
3277 irte.trigger_mode = 0; /* edge */
3278 irte.dlvry_mode = apic->irq_delivery_mode;
3279 irte.vector = cfg->vector;
3280 irte.dest_id = IRTE_DEST(dest);
3282 modify_irte(irq, &irte);
3284 msg->address_hi = MSI_ADDR_BASE_HI;
3285 msg->data = sub_handle;
3286 msg->address_lo = MSI_ADDR_BASE_LO | MSI_ADDR_IR_EXT_INT |
3288 MSI_ADDR_IR_INDEX1(ir_index) |
3289 MSI_ADDR_IR_INDEX2(ir_index);
3291 if (x2apic_enabled())
3292 msg->address_hi = MSI_ADDR_BASE_HI |
3293 MSI_ADDR_EXT_DEST_ID(dest);
3295 msg->address_hi = MSI_ADDR_BASE_HI;
3299 ((apic->irq_dest_mode == 0) ?
3300 MSI_ADDR_DEST_MODE_PHYSICAL:
3301 MSI_ADDR_DEST_MODE_LOGICAL) |
3302 ((apic->irq_delivery_mode != dest_LowestPrio) ?
3303 MSI_ADDR_REDIRECTION_CPU:
3304 MSI_ADDR_REDIRECTION_LOWPRI) |
3305 MSI_ADDR_DEST_ID(dest);
3308 MSI_DATA_TRIGGER_EDGE |
3309 MSI_DATA_LEVEL_ASSERT |
3310 ((apic->irq_delivery_mode != dest_LowestPrio) ?
3311 MSI_DATA_DELIVERY_FIXED:
3312 MSI_DATA_DELIVERY_LOWPRI) |
3313 MSI_DATA_VECTOR(cfg->vector);
3319 static void set_msi_irq_affinity(unsigned int irq, const struct cpumask *mask)
3321 struct irq_desc *desc = irq_to_desc(irq);
3322 struct irq_cfg *cfg;
3326 dest = set_desc_affinity(desc, mask);
3327 if (dest == BAD_APICID)
3330 cfg = desc->chip_data;
3332 read_msi_msg_desc(desc, &msg);
3334 msg.data &= ~MSI_DATA_VECTOR_MASK;
3335 msg.data |= MSI_DATA_VECTOR(cfg->vector);
3336 msg.address_lo &= ~MSI_ADDR_DEST_ID_MASK;
3337 msg.address_lo |= MSI_ADDR_DEST_ID(dest);
3339 write_msi_msg_desc(desc, &msg);
3341 #ifdef CONFIG_INTR_REMAP
3343 * Migrate the MSI irq to another cpumask. This migration is
3344 * done in the process context using interrupt-remapping hardware.
3347 ir_set_msi_irq_affinity(unsigned int irq, const struct cpumask *mask)
3349 struct irq_desc *desc = irq_to_desc(irq);
3350 struct irq_cfg *cfg = desc->chip_data;
3354 if (get_irte(irq, &irte))
3357 dest = set_desc_affinity(desc, mask);
3358 if (dest == BAD_APICID)
3361 irte.vector = cfg->vector;
3362 irte.dest_id = IRTE_DEST(dest);
3365 * atomically update the IRTE with the new destination and vector.
3367 modify_irte(irq, &irte);
3370 * After this point, all the interrupts will start arriving
3371 * at the new destination. So, time to cleanup the previous
3372 * vector allocation.
3374 if (cfg->move_in_progress)
3375 send_cleanup_vector(cfg);
3379 #endif /* CONFIG_SMP */
3382 * IRQ Chip for MSI PCI/PCI-X/PCI-Express Devices,
3383 * which implement the MSI or MSI-X Capability Structure.
3385 static struct irq_chip msi_chip = {
3387 .unmask = unmask_msi_irq,
3388 .mask = mask_msi_irq,
3389 .ack = ack_apic_edge,
3391 .set_affinity = set_msi_irq_affinity,
3393 .retrigger = ioapic_retrigger_irq,
3396 #ifdef CONFIG_INTR_REMAP
3397 static struct irq_chip msi_ir_chip = {
3398 .name = "IR-PCI-MSI",
3399 .unmask = unmask_msi_irq,
3400 .mask = mask_msi_irq,
3401 .ack = ack_x2apic_edge,
3403 .set_affinity = ir_set_msi_irq_affinity,
3405 .retrigger = ioapic_retrigger_irq,
3410 * Map the PCI dev to the corresponding remapping hardware unit
3411 * and allocate 'nvec' consecutive interrupt-remapping table entries
3414 static int msi_alloc_irte(struct pci_dev *dev, int irq, int nvec)
3416 struct intel_iommu *iommu;
3419 iommu = map_dev_to_ir(dev);
3422 "Unable to map PCI %s to iommu\n", pci_name(dev));
3426 index = alloc_irte(iommu, irq, nvec);
3429 "Unable to allocate %d IRTE for PCI %s\n", nvec,
3436 static int setup_msi_irq(struct pci_dev *dev, struct msi_desc *msidesc, int irq)
3441 ret = msi_compose_msg(dev, irq, &msg);
3445 set_irq_msi(irq, msidesc);
3446 write_msi_msg(irq, &msg);
3448 if (irq_remapped(irq)) {
3449 struct irq_desc *desc = irq_to_desc(irq);
3451 * irq migration in process context
3453 desc->status |= IRQ_MOVE_PCNTXT;
3454 set_irq_chip_and_handler_name(irq, &msi_ir_chip, handle_edge_irq, "edge");
3456 set_irq_chip_and_handler_name(irq, &msi_chip, handle_edge_irq, "edge");
3458 dev_printk(KERN_DEBUG, &dev->dev, "irq %d for MSI/MSI-X\n", irq);
3463 int arch_setup_msi_irqs(struct pci_dev *dev, int nvec, int type)
3466 int ret, sub_handle;
3467 struct msi_desc *msidesc;
3468 unsigned int irq_want;
3469 struct intel_iommu *iommu = NULL;
3472 irq_want = nr_irqs_gsi;
3474 list_for_each_entry(msidesc, &dev->msi_list, list) {
3475 irq = create_irq_nr(irq_want);
3479 if (!intr_remapping_enabled)
3484 * allocate the consecutive block of IRTE's
3487 index = msi_alloc_irte(dev, irq, nvec);
3493 iommu = map_dev_to_ir(dev);
3499 * setup the mapping between the irq and the IRTE
3500 * base index, the sub_handle pointing to the
3501 * appropriate interrupt remap table entry.
3503 set_irte_irq(irq, iommu, index, sub_handle);
3506 ret = setup_msi_irq(dev, msidesc, irq);
3518 void arch_teardown_msi_irq(unsigned int irq)
3523 #if defined (CONFIG_DMAR) || defined (CONFIG_INTR_REMAP)
3525 static void dmar_msi_set_affinity(unsigned int irq, const struct cpumask *mask)
3527 struct irq_desc *desc = irq_to_desc(irq);
3528 struct irq_cfg *cfg;
3532 dest = set_desc_affinity(desc, mask);
3533 if (dest == BAD_APICID)
3536 cfg = desc->chip_data;
3538 dmar_msi_read(irq, &msg);
3540 msg.data &= ~MSI_DATA_VECTOR_MASK;
3541 msg.data |= MSI_DATA_VECTOR(cfg->vector);
3542 msg.address_lo &= ~MSI_ADDR_DEST_ID_MASK;
3543 msg.address_lo |= MSI_ADDR_DEST_ID(dest);
3545 dmar_msi_write(irq, &msg);
3548 #endif /* CONFIG_SMP */
3550 struct irq_chip dmar_msi_type = {
3552 .unmask = dmar_msi_unmask,
3553 .mask = dmar_msi_mask,
3554 .ack = ack_apic_edge,
3556 .set_affinity = dmar_msi_set_affinity,
3558 .retrigger = ioapic_retrigger_irq,
3561 int arch_setup_dmar_msi(unsigned int irq)
3566 ret = msi_compose_msg(NULL, irq, &msg);
3569 dmar_msi_write(irq, &msg);
3570 set_irq_chip_and_handler_name(irq, &dmar_msi_type, handle_edge_irq,
3576 #ifdef CONFIG_HPET_TIMER
3579 static void hpet_msi_set_affinity(unsigned int irq, const struct cpumask *mask)
3581 struct irq_desc *desc = irq_to_desc(irq);
3582 struct irq_cfg *cfg;
3586 dest = set_desc_affinity(desc, mask);
3587 if (dest == BAD_APICID)
3590 cfg = desc->chip_data;
3592 hpet_msi_read(irq, &msg);
3594 msg.data &= ~MSI_DATA_VECTOR_MASK;
3595 msg.data |= MSI_DATA_VECTOR(cfg->vector);
3596 msg.address_lo &= ~MSI_ADDR_DEST_ID_MASK;
3597 msg.address_lo |= MSI_ADDR_DEST_ID(dest);
3599 hpet_msi_write(irq, &msg);
3602 #endif /* CONFIG_SMP */
3604 static struct irq_chip hpet_msi_type = {
3606 .unmask = hpet_msi_unmask,
3607 .mask = hpet_msi_mask,
3608 .ack = ack_apic_edge,
3610 .set_affinity = hpet_msi_set_affinity,
3612 .retrigger = ioapic_retrigger_irq,
3615 int arch_setup_hpet_msi(unsigned int irq)
3620 ret = msi_compose_msg(NULL, irq, &msg);
3624 hpet_msi_write(irq, &msg);
3625 set_irq_chip_and_handler_name(irq, &hpet_msi_type, handle_edge_irq,
3632 #endif /* CONFIG_PCI_MSI */
3634 * Hypertransport interrupt support
3636 #ifdef CONFIG_HT_IRQ
3640 static void target_ht_irq(unsigned int irq, unsigned int dest, u8 vector)
3642 struct ht_irq_msg msg;
3643 fetch_ht_irq_msg(irq, &msg);
3645 msg.address_lo &= ~(HT_IRQ_LOW_VECTOR_MASK | HT_IRQ_LOW_DEST_ID_MASK);
3646 msg.address_hi &= ~(HT_IRQ_HIGH_DEST_ID_MASK);
3648 msg.address_lo |= HT_IRQ_LOW_VECTOR(vector) | HT_IRQ_LOW_DEST_ID(dest);
3649 msg.address_hi |= HT_IRQ_HIGH_DEST_ID(dest);
3651 write_ht_irq_msg(irq, &msg);
3654 static void set_ht_irq_affinity(unsigned int irq, const struct cpumask *mask)
3656 struct irq_desc *desc = irq_to_desc(irq);
3657 struct irq_cfg *cfg;
3660 dest = set_desc_affinity(desc, mask);
3661 if (dest == BAD_APICID)
3664 cfg = desc->chip_data;
3666 target_ht_irq(irq, dest, cfg->vector);
3671 static struct irq_chip ht_irq_chip = {
3673 .mask = mask_ht_irq,
3674 .unmask = unmask_ht_irq,
3675 .ack = ack_apic_edge,
3677 .set_affinity = set_ht_irq_affinity,
3679 .retrigger = ioapic_retrigger_irq,
3682 int arch_setup_ht_irq(unsigned int irq, struct pci_dev *dev)
3684 struct irq_cfg *cfg;
3691 err = assign_irq_vector(irq, cfg, apic->target_cpus());
3693 struct ht_irq_msg msg;
3696 dest = apic->cpu_mask_to_apicid_and(cfg->domain,
3697 apic->target_cpus());
3699 msg.address_hi = HT_IRQ_HIGH_DEST_ID(dest);
3703 HT_IRQ_LOW_DEST_ID(dest) |
3704 HT_IRQ_LOW_VECTOR(cfg->vector) |
3705 ((apic->irq_dest_mode == 0) ?
3706 HT_IRQ_LOW_DM_PHYSICAL :
3707 HT_IRQ_LOW_DM_LOGICAL) |
3708 HT_IRQ_LOW_RQEOI_EDGE |
3709 ((apic->irq_delivery_mode != dest_LowestPrio) ?
3710 HT_IRQ_LOW_MT_FIXED :
3711 HT_IRQ_LOW_MT_ARBITRATED) |
3712 HT_IRQ_LOW_IRQ_MASKED;
3714 write_ht_irq_msg(irq, &msg);
3716 set_irq_chip_and_handler_name(irq, &ht_irq_chip,
3717 handle_edge_irq, "edge");
3719 dev_printk(KERN_DEBUG, &dev->dev, "irq %d for HT\n", irq);
3723 #endif /* CONFIG_HT_IRQ */
3725 #ifdef CONFIG_X86_UV
3727 * Re-target the irq to the specified CPU and enable the specified MMR located
3728 * on the specified blade to allow the sending of MSIs to the specified CPU.
3730 int arch_enable_uv_irq(char *irq_name, unsigned int irq, int cpu, int mmr_blade,
3731 unsigned long mmr_offset)
3733 const struct cpumask *eligible_cpu = cpumask_of(cpu);
3734 struct irq_cfg *cfg;
3736 unsigned long mmr_value;
3737 struct uv_IO_APIC_route_entry *entry;
3738 unsigned long flags;
3743 err = assign_irq_vector(irq, cfg, eligible_cpu);
3747 spin_lock_irqsave(&vector_lock, flags);
3748 set_irq_chip_and_handler_name(irq, &uv_irq_chip, handle_percpu_irq,
3750 spin_unlock_irqrestore(&vector_lock, flags);
3753 entry = (struct uv_IO_APIC_route_entry *)&mmr_value;
3754 BUG_ON(sizeof(struct uv_IO_APIC_route_entry) != sizeof(unsigned long));
3756 entry->vector = cfg->vector;
3757 entry->delivery_mode = apic->irq_delivery_mode;
3758 entry->dest_mode = apic->irq_dest_mode;
3759 entry->polarity = 0;
3762 entry->dest = apic->cpu_mask_to_apicid(eligible_cpu);
3764 mmr_pnode = uv_blade_to_pnode(mmr_blade);
3765 uv_write_global_mmr64(mmr_pnode, mmr_offset, mmr_value);
3771 * Disable the specified MMR located on the specified blade so that MSIs are
3772 * longer allowed to be sent.
3774 void arch_disable_uv_irq(int mmr_blade, unsigned long mmr_offset)
3776 unsigned long mmr_value;
3777 struct uv_IO_APIC_route_entry *entry;
3781 entry = (struct uv_IO_APIC_route_entry *)&mmr_value;
3782 BUG_ON(sizeof(struct uv_IO_APIC_route_entry) != sizeof(unsigned long));
3786 mmr_pnode = uv_blade_to_pnode(mmr_blade);
3787 uv_write_global_mmr64(mmr_pnode, mmr_offset, mmr_value);
3789 #endif /* CONFIG_X86_64 */
3791 int __init io_apic_get_redir_entries (int ioapic)
3793 union IO_APIC_reg_01 reg_01;
3794 unsigned long flags;
3796 spin_lock_irqsave(&ioapic_lock, flags);
3797 reg_01.raw = io_apic_read(ioapic, 1);
3798 spin_unlock_irqrestore(&ioapic_lock, flags);
3800 return reg_01.bits.entries;
3803 void __init probe_nr_irqs_gsi(void)
3807 nr = acpi_probe_gsi();
3808 if (nr > nr_irqs_gsi) {
3811 /* for acpi=off or acpi is not compiled in */
3815 for (idx = 0; idx < nr_ioapics; idx++)
3816 nr += io_apic_get_redir_entries(idx) + 1;
3818 if (nr > nr_irqs_gsi)
3822 printk(KERN_DEBUG "nr_irqs_gsi: %d\n", nr_irqs_gsi);
3825 #ifdef CONFIG_SPARSE_IRQ
3826 int __init arch_probe_nr_irqs(void)
3830 if (nr_irqs > (NR_VECTORS * nr_cpu_ids))
3831 nr_irqs = NR_VECTORS * nr_cpu_ids;
3833 nr = nr_irqs_gsi + 8 * nr_cpu_ids;
3834 #if defined(CONFIG_PCI_MSI) || defined(CONFIG_HT_IRQ)
3836 * for MSI and HT dyn irq
3838 nr += nr_irqs_gsi * 16;
3847 /* --------------------------------------------------------------------------
3848 ACPI-based IOAPIC Configuration
3849 -------------------------------------------------------------------------- */
3853 #ifdef CONFIG_X86_32
3854 int __init io_apic_get_unique_id(int ioapic, int apic_id)
3856 union IO_APIC_reg_00 reg_00;
3857 static physid_mask_t apic_id_map = PHYSID_MASK_NONE;
3859 unsigned long flags;
3863 * The P4 platform supports up to 256 APIC IDs on two separate APIC
3864 * buses (one for LAPICs, one for IOAPICs), where predecessors only
3865 * supports up to 16 on one shared APIC bus.
3867 * TBD: Expand LAPIC/IOAPIC support on P4-class systems to take full
3868 * advantage of new APIC bus architecture.
3871 if (physids_empty(apic_id_map))
3872 apic_id_map = apic->ioapic_phys_id_map(phys_cpu_present_map);
3874 spin_lock_irqsave(&ioapic_lock, flags);
3875 reg_00.raw = io_apic_read(ioapic, 0);
3876 spin_unlock_irqrestore(&ioapic_lock, flags);
3878 if (apic_id >= get_physical_broadcast()) {
3879 printk(KERN_WARNING "IOAPIC[%d]: Invalid apic_id %d, trying "
3880 "%d\n", ioapic, apic_id, reg_00.bits.ID);
3881 apic_id = reg_00.bits.ID;
3885 * Every APIC in a system must have a unique ID or we get lots of nice
3886 * 'stuck on smp_invalidate_needed IPI wait' messages.
3888 if (apic->check_apicid_used(apic_id_map, apic_id)) {
3890 for (i = 0; i < get_physical_broadcast(); i++) {
3891 if (!apic->check_apicid_used(apic_id_map, i))
3895 if (i == get_physical_broadcast())
3896 panic("Max apic_id exceeded!\n");
3898 printk(KERN_WARNING "IOAPIC[%d]: apic_id %d already used, "
3899 "trying %d\n", ioapic, apic_id, i);
3904 tmp = apic->apicid_to_cpu_present(apic_id);
3905 physids_or(apic_id_map, apic_id_map, tmp);
3907 if (reg_00.bits.ID != apic_id) {
3908 reg_00.bits.ID = apic_id;
3910 spin_lock_irqsave(&ioapic_lock, flags);
3911 io_apic_write(ioapic, 0, reg_00.raw);
3912 reg_00.raw = io_apic_read(ioapic, 0);
3913 spin_unlock_irqrestore(&ioapic_lock, flags);
3916 if (reg_00.bits.ID != apic_id) {
3917 printk("IOAPIC[%d]: Unable to change apic_id!\n", ioapic);
3922 apic_printk(APIC_VERBOSE, KERN_INFO
3923 "IOAPIC[%d]: Assigned apic_id %d\n", ioapic, apic_id);
3928 int __init io_apic_get_version(int ioapic)
3930 union IO_APIC_reg_01 reg_01;
3931 unsigned long flags;
3933 spin_lock_irqsave(&ioapic_lock, flags);
3934 reg_01.raw = io_apic_read(ioapic, 1);
3935 spin_unlock_irqrestore(&ioapic_lock, flags);
3937 return reg_01.bits.version;
3941 int io_apic_set_pci_routing (int ioapic, int pin, int irq, int triggering, int polarity)
3943 struct irq_desc *desc;
3944 struct irq_cfg *cfg;
3945 int cpu = boot_cpu_id;
3947 if (!IO_APIC_IRQ(irq)) {
3948 apic_printk(APIC_QUIET,KERN_ERR "IOAPIC[%d]: Invalid reference to IRQ 0\n",
3953 desc = irq_to_desc_alloc_cpu(irq, cpu);
3955 printk(KERN_INFO "can not get irq_desc %d\n", irq);
3960 * IRQs < 16 are already in the irq_2_pin[] map
3962 if (irq >= NR_IRQS_LEGACY) {
3963 cfg = desc->chip_data;
3964 add_pin_to_irq_cpu(cfg, cpu, ioapic, pin);
3967 setup_IO_APIC_irq(ioapic, pin, irq, desc, triggering, polarity);
3973 int acpi_get_override_irq(int bus_irq, int *trigger, int *polarity)
3977 if (skip_ioapic_setup)
3980 for (i = 0; i < mp_irq_entries; i++)
3981 if (mp_irqs[i].irqtype == mp_INT &&
3982 mp_irqs[i].srcbusirq == bus_irq)
3984 if (i >= mp_irq_entries)
3987 *trigger = irq_trigger(i);
3988 *polarity = irq_polarity(i);
3992 #endif /* CONFIG_ACPI */
3995 * This function currently is only a helper for the i386 smp boot process where
3996 * we need to reprogram the ioredtbls to cater for the cpus which have come online
3997 * so mask in all cases should simply be apic->target_cpus()
4000 void __init setup_ioapic_dest(void)
4002 int pin, ioapic, irq, irq_entry;
4003 struct irq_desc *desc;
4004 struct irq_cfg *cfg;
4005 const struct cpumask *mask;
4007 if (skip_ioapic_setup == 1)
4010 for (ioapic = 0; ioapic < nr_ioapics; ioapic++) {
4011 for (pin = 0; pin < nr_ioapic_registers[ioapic]; pin++) {
4012 irq_entry = find_irq_entry(ioapic, pin, mp_INT);
4013 if (irq_entry == -1)
4015 irq = pin_2_irq(irq_entry, ioapic, pin);
4017 /* setup_IO_APIC_irqs could fail to get vector for some device
4018 * when you have too many devices, because at that time only boot
4021 desc = irq_to_desc(irq);
4022 cfg = desc->chip_data;
4024 setup_IO_APIC_irq(ioapic, pin, irq, desc,
4025 irq_trigger(irq_entry),
4026 irq_polarity(irq_entry));
4032 * Honour affinities which have been set in early boot
4035 (IRQ_NO_BALANCING | IRQ_AFFINITY_SET))
4036 mask = desc->affinity;
4038 mask = apic->target_cpus();
4040 if (intr_remapping_enabled)
4041 set_ir_ioapic_affinity_irq_desc(desc, mask);
4043 set_ioapic_affinity_irq_desc(desc, mask);
4050 #define IOAPIC_RESOURCE_NAME_SIZE 11
4052 static struct resource *ioapic_resources;
4054 static struct resource * __init ioapic_setup_resources(void)
4057 struct resource *res;
4061 if (nr_ioapics <= 0)
4064 n = IOAPIC_RESOURCE_NAME_SIZE + sizeof(struct resource);
4067 mem = alloc_bootmem(n);
4071 mem += sizeof(struct resource) * nr_ioapics;
4073 for (i = 0; i < nr_ioapics; i++) {
4075 res[i].flags = IORESOURCE_MEM | IORESOURCE_BUSY;
4076 sprintf(mem, "IOAPIC %u", i);
4077 mem += IOAPIC_RESOURCE_NAME_SIZE;
4081 ioapic_resources = res;
4086 void __init ioapic_init_mappings(void)
4088 unsigned long ioapic_phys, idx = FIX_IO_APIC_BASE_0;
4089 struct resource *ioapic_res;
4092 ioapic_res = ioapic_setup_resources();
4093 for (i = 0; i < nr_ioapics; i++) {
4094 if (smp_found_config) {
4095 ioapic_phys = mp_ioapics[i].apicaddr;
4096 #ifdef CONFIG_X86_32
4099 "WARNING: bogus zero IO-APIC "
4100 "address found in MPTABLE, "
4101 "disabling IO/APIC support!\n");
4102 smp_found_config = 0;
4103 skip_ioapic_setup = 1;
4104 goto fake_ioapic_page;
4108 #ifdef CONFIG_X86_32
4111 ioapic_phys = (unsigned long)
4112 alloc_bootmem_pages(PAGE_SIZE);
4113 ioapic_phys = __pa(ioapic_phys);
4115 set_fixmap_nocache(idx, ioapic_phys);
4116 apic_printk(APIC_VERBOSE,
4117 "mapped IOAPIC to %08lx (%08lx)\n",
4118 __fix_to_virt(idx), ioapic_phys);
4121 if (ioapic_res != NULL) {
4122 ioapic_res->start = ioapic_phys;
4123 ioapic_res->end = ioapic_phys + (4 * 1024) - 1;
4129 static int __init ioapic_insert_resources(void)
4132 struct resource *r = ioapic_resources;
4135 if (nr_ioapics > 0) {
4137 "IO APIC resources couldn't be allocated.\n");
4143 for (i = 0; i < nr_ioapics; i++) {
4144 insert_resource(&iomem_resource, r);
4151 /* Insert the IO APIC resources after PCI initialization has occured to handle
4152 * IO APICS that are mapped in on a BAR in PCI space. */
4153 late_initcall(ioapic_insert_resources);