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>
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>
66 #include <mach_apic.h>
67 #include <mach_apicdef.h>
69 #define __apicdebuginit(type) static type __init
72 * Is the SiS APIC rmw bug present ?
73 * -1 = don't know, 0 = no, 1 = yes
75 int sis_apic_bug = -1;
77 static DEFINE_SPINLOCK(ioapic_lock);
78 static DEFINE_SPINLOCK(vector_lock);
81 * # of IRQ routing registers
83 int nr_ioapic_registers[MAX_IO_APICS];
85 /* I/O APIC entries */
86 struct mp_config_ioapic mp_ioapics[MAX_IO_APICS];
89 /* MP IRQ source entries */
90 struct mp_config_intsrc mp_irqs[MAX_IRQ_SOURCES];
92 /* # of MP IRQ source entries */
95 #if defined (CONFIG_MCA) || defined (CONFIG_EISA)
96 int mp_bus_id_to_type[MAX_MP_BUSSES];
99 DECLARE_BITMAP(mp_bus_not_pci, MAX_MP_BUSSES);
101 int skip_ioapic_setup;
103 static int __init parse_noapic(char *str)
105 /* disable IO-APIC */
106 disable_ioapic_setup();
109 early_param("noapic", parse_noapic);
114 * This is performance-critical, we want to do it O(1)
116 * the indexing order of this array favors 1:1 mappings
117 * between pins and IRQs.
120 struct irq_pin_list {
122 struct irq_pin_list *next;
125 static struct irq_pin_list *get_one_free_irq_2_pin(int cpu)
127 struct irq_pin_list *pin;
130 node = cpu_to_node(cpu);
132 pin = kzalloc_node(sizeof(*pin), GFP_ATOMIC, node);
138 struct irq_pin_list *irq_2_pin;
139 cpumask_var_t domain;
140 cpumask_var_t old_domain;
141 unsigned move_cleanup_count;
143 u8 move_in_progress : 1;
144 #ifdef CONFIG_NUMA_MIGRATE_IRQ_DESC
145 u8 move_desc_pending : 1;
149 /* irq_cfg is indexed by the sum of all RTEs in all I/O APICs. */
150 #ifdef CONFIG_SPARSE_IRQ
151 static struct irq_cfg irq_cfgx[] = {
153 static struct irq_cfg irq_cfgx[NR_IRQS] = {
155 [0] = { .vector = IRQ0_VECTOR, },
156 [1] = { .vector = IRQ1_VECTOR, },
157 [2] = { .vector = IRQ2_VECTOR, },
158 [3] = { .vector = IRQ3_VECTOR, },
159 [4] = { .vector = IRQ4_VECTOR, },
160 [5] = { .vector = IRQ5_VECTOR, },
161 [6] = { .vector = IRQ6_VECTOR, },
162 [7] = { .vector = IRQ7_VECTOR, },
163 [8] = { .vector = IRQ8_VECTOR, },
164 [9] = { .vector = IRQ9_VECTOR, },
165 [10] = { .vector = IRQ10_VECTOR, },
166 [11] = { .vector = IRQ11_VECTOR, },
167 [12] = { .vector = IRQ12_VECTOR, },
168 [13] = { .vector = IRQ13_VECTOR, },
169 [14] = { .vector = IRQ14_VECTOR, },
170 [15] = { .vector = IRQ15_VECTOR, },
173 int __init arch_early_irq_init(void)
176 struct irq_desc *desc;
181 count = ARRAY_SIZE(irq_cfgx);
183 for (i = 0; i < count; i++) {
184 desc = irq_to_desc(i);
185 desc->chip_data = &cfg[i];
186 alloc_bootmem_cpumask_var(&cfg[i].domain);
187 alloc_bootmem_cpumask_var(&cfg[i].old_domain);
188 if (i < NR_IRQS_LEGACY)
189 cpumask_setall(cfg[i].domain);
195 #ifdef CONFIG_SPARSE_IRQ
196 static struct irq_cfg *irq_cfg(unsigned int irq)
198 struct irq_cfg *cfg = NULL;
199 struct irq_desc *desc;
201 desc = irq_to_desc(irq);
203 cfg = desc->chip_data;
208 static struct irq_cfg *get_one_free_irq_cfg(int cpu)
213 node = cpu_to_node(cpu);
215 cfg = kzalloc_node(sizeof(*cfg), GFP_ATOMIC, node);
217 if (!alloc_cpumask_var_node(&cfg->domain, GFP_ATOMIC, node)) {
220 } else if (!alloc_cpumask_var_node(&cfg->old_domain,
222 free_cpumask_var(cfg->domain);
226 cpumask_clear(cfg->domain);
227 cpumask_clear(cfg->old_domain);
234 int arch_init_chip_data(struct irq_desc *desc, int cpu)
238 cfg = desc->chip_data;
240 desc->chip_data = get_one_free_irq_cfg(cpu);
241 if (!desc->chip_data) {
242 printk(KERN_ERR "can not alloc irq_cfg\n");
250 #ifdef CONFIG_NUMA_MIGRATE_IRQ_DESC
253 init_copy_irq_2_pin(struct irq_cfg *old_cfg, struct irq_cfg *cfg, int cpu)
255 struct irq_pin_list *old_entry, *head, *tail, *entry;
257 cfg->irq_2_pin = NULL;
258 old_entry = old_cfg->irq_2_pin;
262 entry = get_one_free_irq_2_pin(cpu);
266 entry->apic = old_entry->apic;
267 entry->pin = old_entry->pin;
270 old_entry = old_entry->next;
272 entry = get_one_free_irq_2_pin(cpu);
280 /* still use the old one */
283 entry->apic = old_entry->apic;
284 entry->pin = old_entry->pin;
287 old_entry = old_entry->next;
291 cfg->irq_2_pin = head;
294 static void free_irq_2_pin(struct irq_cfg *old_cfg, struct irq_cfg *cfg)
296 struct irq_pin_list *entry, *next;
298 if (old_cfg->irq_2_pin == cfg->irq_2_pin)
301 entry = old_cfg->irq_2_pin;
308 old_cfg->irq_2_pin = NULL;
311 void arch_init_copy_chip_data(struct irq_desc *old_desc,
312 struct irq_desc *desc, int cpu)
315 struct irq_cfg *old_cfg;
317 cfg = get_one_free_irq_cfg(cpu);
322 desc->chip_data = cfg;
324 old_cfg = old_desc->chip_data;
326 memcpy(cfg, old_cfg, sizeof(struct irq_cfg));
328 init_copy_irq_2_pin(old_cfg, cfg, cpu);
331 static void free_irq_cfg(struct irq_cfg *old_cfg)
336 void arch_free_chip_data(struct irq_desc *old_desc, struct irq_desc *desc)
338 struct irq_cfg *old_cfg, *cfg;
340 old_cfg = old_desc->chip_data;
341 cfg = desc->chip_data;
347 free_irq_2_pin(old_cfg, cfg);
348 free_irq_cfg(old_cfg);
349 old_desc->chip_data = NULL;
354 set_extra_move_desc(struct irq_desc *desc, const struct cpumask *mask)
356 struct irq_cfg *cfg = desc->chip_data;
358 if (!cfg->move_in_progress) {
359 /* it means that domain is not changed */
360 if (!cpumask_intersects(&desc->affinity, mask))
361 cfg->move_desc_pending = 1;
367 static struct irq_cfg *irq_cfg(unsigned int irq)
369 return irq < nr_irqs ? irq_cfgx + irq : NULL;
374 #ifndef CONFIG_NUMA_MIGRATE_IRQ_DESC
376 set_extra_move_desc(struct irq_desc *desc, const struct cpumask *mask)
383 unsigned int unused[3];
387 static __attribute_const__ struct io_apic __iomem *io_apic_base(int idx)
389 return (void __iomem *) __fix_to_virt(FIX_IO_APIC_BASE_0 + idx)
390 + (mp_ioapics[idx].mp_apicaddr & ~PAGE_MASK);
393 static inline unsigned int io_apic_read(unsigned int apic, unsigned int reg)
395 struct io_apic __iomem *io_apic = io_apic_base(apic);
396 writel(reg, &io_apic->index);
397 return readl(&io_apic->data);
400 static inline void io_apic_write(unsigned int apic, unsigned int reg, unsigned int value)
402 struct io_apic __iomem *io_apic = io_apic_base(apic);
403 writel(reg, &io_apic->index);
404 writel(value, &io_apic->data);
408 * Re-write a value: to be used for read-modify-write
409 * cycles where the read already set up the index register.
411 * Older SiS APIC requires we rewrite the index register
413 static inline void io_apic_modify(unsigned int apic, unsigned int reg, unsigned int value)
415 struct io_apic __iomem *io_apic = io_apic_base(apic);
418 writel(reg, &io_apic->index);
419 writel(value, &io_apic->data);
422 static bool io_apic_level_ack_pending(struct irq_cfg *cfg)
424 struct irq_pin_list *entry;
427 spin_lock_irqsave(&ioapic_lock, flags);
428 entry = cfg->irq_2_pin;
436 reg = io_apic_read(entry->apic, 0x10 + pin*2);
437 /* Is the remote IRR bit set? */
438 if (reg & IO_APIC_REDIR_REMOTE_IRR) {
439 spin_unlock_irqrestore(&ioapic_lock, flags);
446 spin_unlock_irqrestore(&ioapic_lock, flags);
452 struct { u32 w1, w2; };
453 struct IO_APIC_route_entry entry;
456 static struct IO_APIC_route_entry ioapic_read_entry(int apic, int pin)
458 union entry_union eu;
460 spin_lock_irqsave(&ioapic_lock, flags);
461 eu.w1 = io_apic_read(apic, 0x10 + 2 * pin);
462 eu.w2 = io_apic_read(apic, 0x11 + 2 * pin);
463 spin_unlock_irqrestore(&ioapic_lock, flags);
468 * When we write a new IO APIC routing entry, we need to write the high
469 * word first! If the mask bit in the low word is clear, we will enable
470 * the interrupt, and we need to make sure the entry is fully populated
471 * before that happens.
474 __ioapic_write_entry(int apic, int pin, struct IO_APIC_route_entry e)
476 union entry_union eu;
478 io_apic_write(apic, 0x11 + 2*pin, eu.w2);
479 io_apic_write(apic, 0x10 + 2*pin, eu.w1);
482 static void ioapic_write_entry(int apic, int pin, struct IO_APIC_route_entry e)
485 spin_lock_irqsave(&ioapic_lock, flags);
486 __ioapic_write_entry(apic, pin, e);
487 spin_unlock_irqrestore(&ioapic_lock, flags);
491 * When we mask an IO APIC routing entry, we need to write the low
492 * word first, in order to set the mask bit before we change the
495 static void ioapic_mask_entry(int apic, int pin)
498 union entry_union eu = { .entry.mask = 1 };
500 spin_lock_irqsave(&ioapic_lock, flags);
501 io_apic_write(apic, 0x10 + 2*pin, eu.w1);
502 io_apic_write(apic, 0x11 + 2*pin, eu.w2);
503 spin_unlock_irqrestore(&ioapic_lock, flags);
507 static void send_cleanup_vector(struct irq_cfg *cfg)
509 cpumask_var_t cleanup_mask;
511 if (unlikely(!alloc_cpumask_var(&cleanup_mask, GFP_ATOMIC))) {
513 cfg->move_cleanup_count = 0;
514 for_each_cpu_and(i, cfg->old_domain, cpu_online_mask)
515 cfg->move_cleanup_count++;
516 for_each_cpu_and(i, cfg->old_domain, cpu_online_mask)
517 send_IPI_mask(cpumask_of(i), IRQ_MOVE_CLEANUP_VECTOR);
519 cpumask_and(cleanup_mask, cfg->old_domain, cpu_online_mask);
520 cfg->move_cleanup_count = cpumask_weight(cleanup_mask);
521 send_IPI_mask(cleanup_mask, IRQ_MOVE_CLEANUP_VECTOR);
522 free_cpumask_var(cleanup_mask);
524 cfg->move_in_progress = 0;
527 static void __target_IO_APIC_irq(unsigned int irq, unsigned int dest, struct irq_cfg *cfg)
530 struct irq_pin_list *entry;
531 u8 vector = cfg->vector;
533 entry = cfg->irq_2_pin;
542 #ifdef CONFIG_INTR_REMAP
544 * With interrupt-remapping, destination information comes
545 * from interrupt-remapping table entry.
547 if (!irq_remapped(irq))
548 io_apic_write(apic, 0x11 + pin*2, dest);
550 io_apic_write(apic, 0x11 + pin*2, dest);
552 reg = io_apic_read(apic, 0x10 + pin*2);
553 reg &= ~IO_APIC_REDIR_VECTOR_MASK;
555 io_apic_modify(apic, 0x10 + pin*2, reg);
563 assign_irq_vector(int irq, struct irq_cfg *cfg, const struct cpumask *mask);
566 * Either sets desc->affinity to a valid value, and returns cpu_mask_to_apicid
567 * of that, or returns BAD_APICID and leaves desc->affinity untouched.
570 set_desc_affinity(struct irq_desc *desc, const struct cpumask *mask)
575 if (!cpumask_intersects(mask, cpu_online_mask))
579 cfg = desc->chip_data;
580 if (assign_irq_vector(irq, cfg, mask))
583 cpumask_and(&desc->affinity, cfg->domain, mask);
584 set_extra_move_desc(desc, mask);
585 return cpu_mask_to_apicid_and(&desc->affinity, cpu_online_mask);
589 set_ioapic_affinity_irq_desc(struct irq_desc *desc, const struct cpumask *mask)
597 cfg = desc->chip_data;
599 spin_lock_irqsave(&ioapic_lock, flags);
600 dest = set_desc_affinity(desc, mask);
601 if (dest != BAD_APICID) {
602 /* Only the high 8 bits are valid. */
603 dest = SET_APIC_LOGICAL_ID(dest);
604 __target_IO_APIC_irq(irq, dest, cfg);
606 spin_unlock_irqrestore(&ioapic_lock, flags);
610 set_ioapic_affinity_irq(unsigned int irq, const struct cpumask *mask)
612 struct irq_desc *desc;
614 desc = irq_to_desc(irq);
616 set_ioapic_affinity_irq_desc(desc, mask);
618 #endif /* CONFIG_SMP */
621 * The common case is 1:1 IRQ<->pin mappings. Sometimes there are
622 * shared ISA-space IRQs, so we have to support them. We are super
623 * fast in the common case, and fast for shared ISA-space IRQs.
625 static void add_pin_to_irq_cpu(struct irq_cfg *cfg, int cpu, int apic, int pin)
627 struct irq_pin_list *entry;
629 entry = cfg->irq_2_pin;
631 entry = get_one_free_irq_2_pin(cpu);
633 printk(KERN_ERR "can not alloc irq_2_pin to add %d - %d\n",
637 cfg->irq_2_pin = entry;
643 while (entry->next) {
644 /* not again, please */
645 if (entry->apic == apic && entry->pin == pin)
651 entry->next = get_one_free_irq_2_pin(cpu);
658 * Reroute an IRQ to a different pin.
660 static void __init replace_pin_at_irq_cpu(struct irq_cfg *cfg, int cpu,
661 int oldapic, int oldpin,
662 int newapic, int newpin)
664 struct irq_pin_list *entry = cfg->irq_2_pin;
668 if (entry->apic == oldapic && entry->pin == oldpin) {
669 entry->apic = newapic;
672 /* every one is different, right? */
678 /* why? call replace before add? */
680 add_pin_to_irq_cpu(cfg, cpu, newapic, newpin);
683 static inline void io_apic_modify_irq(struct irq_cfg *cfg,
684 int mask_and, int mask_or,
685 void (*final)(struct irq_pin_list *entry))
688 struct irq_pin_list *entry;
690 for (entry = cfg->irq_2_pin; entry != NULL; entry = entry->next) {
693 reg = io_apic_read(entry->apic, 0x10 + pin * 2);
696 io_apic_modify(entry->apic, 0x10 + pin * 2, reg);
702 static void __unmask_IO_APIC_irq(struct irq_cfg *cfg)
704 io_apic_modify_irq(cfg, ~IO_APIC_REDIR_MASKED, 0, NULL);
708 static void io_apic_sync(struct irq_pin_list *entry)
711 * Synchronize the IO-APIC and the CPU by doing
712 * a dummy read from the IO-APIC
714 struct io_apic __iomem *io_apic;
715 io_apic = io_apic_base(entry->apic);
716 readl(&io_apic->data);
719 static void __mask_IO_APIC_irq(struct irq_cfg *cfg)
721 io_apic_modify_irq(cfg, ~0, IO_APIC_REDIR_MASKED, &io_apic_sync);
723 #else /* CONFIG_X86_32 */
724 static void __mask_IO_APIC_irq(struct irq_cfg *cfg)
726 io_apic_modify_irq(cfg, ~0, IO_APIC_REDIR_MASKED, NULL);
729 static void __mask_and_edge_IO_APIC_irq(struct irq_cfg *cfg)
731 io_apic_modify_irq(cfg, ~IO_APIC_REDIR_LEVEL_TRIGGER,
732 IO_APIC_REDIR_MASKED, NULL);
735 static void __unmask_and_level_IO_APIC_irq(struct irq_cfg *cfg)
737 io_apic_modify_irq(cfg, ~IO_APIC_REDIR_MASKED,
738 IO_APIC_REDIR_LEVEL_TRIGGER, NULL);
740 #endif /* CONFIG_X86_32 */
742 static void mask_IO_APIC_irq_desc(struct irq_desc *desc)
744 struct irq_cfg *cfg = desc->chip_data;
749 spin_lock_irqsave(&ioapic_lock, flags);
750 __mask_IO_APIC_irq(cfg);
751 spin_unlock_irqrestore(&ioapic_lock, flags);
754 static void unmask_IO_APIC_irq_desc(struct irq_desc *desc)
756 struct irq_cfg *cfg = desc->chip_data;
759 spin_lock_irqsave(&ioapic_lock, flags);
760 __unmask_IO_APIC_irq(cfg);
761 spin_unlock_irqrestore(&ioapic_lock, flags);
764 static void mask_IO_APIC_irq(unsigned int irq)
766 struct irq_desc *desc = irq_to_desc(irq);
768 mask_IO_APIC_irq_desc(desc);
770 static void unmask_IO_APIC_irq(unsigned int irq)
772 struct irq_desc *desc = irq_to_desc(irq);
774 unmask_IO_APIC_irq_desc(desc);
777 static void clear_IO_APIC_pin(unsigned int apic, unsigned int pin)
779 struct IO_APIC_route_entry entry;
781 /* Check delivery_mode to be sure we're not clearing an SMI pin */
782 entry = ioapic_read_entry(apic, pin);
783 if (entry.delivery_mode == dest_SMI)
786 * Disable it in the IO-APIC irq-routing table:
788 ioapic_mask_entry(apic, pin);
791 static void clear_IO_APIC (void)
795 for (apic = 0; apic < nr_ioapics; apic++)
796 for (pin = 0; pin < nr_ioapic_registers[apic]; pin++)
797 clear_IO_APIC_pin(apic, pin);
800 #if !defined(CONFIG_SMP) && defined(CONFIG_X86_32)
801 void send_IPI_self(int vector)
808 apic_wait_icr_idle();
809 cfg = APIC_DM_FIXED | APIC_DEST_SELF | vector | APIC_DEST_LOGICAL;
811 * Send the IPI. The write to APIC_ICR fires this off.
813 apic_write(APIC_ICR, cfg);
815 #endif /* !CONFIG_SMP && CONFIG_X86_32*/
819 * support for broken MP BIOSs, enables hand-redirection of PIRQ0-7 to
820 * specific CPU-side IRQs.
824 static int pirq_entries [MAX_PIRQS];
825 static int pirqs_enabled;
827 static int __init ioapic_pirq_setup(char *str)
830 int ints[MAX_PIRQS+1];
832 get_options(str, ARRAY_SIZE(ints), ints);
834 for (i = 0; i < MAX_PIRQS; i++)
835 pirq_entries[i] = -1;
838 apic_printk(APIC_VERBOSE, KERN_INFO
839 "PIRQ redirection, working around broken MP-BIOS.\n");
841 if (ints[0] < MAX_PIRQS)
844 for (i = 0; i < max; i++) {
845 apic_printk(APIC_VERBOSE, KERN_DEBUG
846 "... PIRQ%d -> IRQ %d\n", i, ints[i+1]);
848 * PIRQs are mapped upside down, usually.
850 pirq_entries[MAX_PIRQS-i-1] = ints[i+1];
855 __setup("pirq=", ioapic_pirq_setup);
856 #endif /* CONFIG_X86_32 */
858 #ifdef CONFIG_INTR_REMAP
859 /* I/O APIC RTE contents at the OS boot up */
860 static struct IO_APIC_route_entry *early_ioapic_entries[MAX_IO_APICS];
863 * Saves and masks all the unmasked IO-APIC RTE's
865 int save_mask_IO_APIC_setup(void)
867 union IO_APIC_reg_01 reg_01;
872 * The number of IO-APIC IRQ registers (== #pins):
874 for (apic = 0; apic < nr_ioapics; apic++) {
875 spin_lock_irqsave(&ioapic_lock, flags);
876 reg_01.raw = io_apic_read(apic, 1);
877 spin_unlock_irqrestore(&ioapic_lock, flags);
878 nr_ioapic_registers[apic] = reg_01.bits.entries+1;
881 for (apic = 0; apic < nr_ioapics; apic++) {
882 early_ioapic_entries[apic] =
883 kzalloc(sizeof(struct IO_APIC_route_entry) *
884 nr_ioapic_registers[apic], GFP_KERNEL);
885 if (!early_ioapic_entries[apic])
889 for (apic = 0; apic < nr_ioapics; apic++)
890 for (pin = 0; pin < nr_ioapic_registers[apic]; pin++) {
891 struct IO_APIC_route_entry entry;
893 entry = early_ioapic_entries[apic][pin] =
894 ioapic_read_entry(apic, pin);
897 ioapic_write_entry(apic, pin, entry);
905 kfree(early_ioapic_entries[apic--]);
906 memset(early_ioapic_entries, 0,
907 ARRAY_SIZE(early_ioapic_entries));
912 void restore_IO_APIC_setup(void)
916 for (apic = 0; apic < nr_ioapics; apic++) {
917 if (!early_ioapic_entries[apic])
919 for (pin = 0; pin < nr_ioapic_registers[apic]; pin++)
920 ioapic_write_entry(apic, pin,
921 early_ioapic_entries[apic][pin]);
922 kfree(early_ioapic_entries[apic]);
923 early_ioapic_entries[apic] = NULL;
927 void reinit_intr_remapped_IO_APIC(int intr_remapping)
930 * for now plain restore of previous settings.
931 * TBD: In the case of OS enabling interrupt-remapping,
932 * IO-APIC RTE's need to be setup to point to interrupt-remapping
933 * table entries. for now, do a plain restore, and wait for
934 * the setup_IO_APIC_irqs() to do proper initialization.
936 restore_IO_APIC_setup();
941 * Find the IRQ entry number of a certain pin.
943 static int find_irq_entry(int apic, int pin, int type)
947 for (i = 0; i < mp_irq_entries; i++)
948 if (mp_irqs[i].mp_irqtype == type &&
949 (mp_irqs[i].mp_dstapic == mp_ioapics[apic].mp_apicid ||
950 mp_irqs[i].mp_dstapic == MP_APIC_ALL) &&
951 mp_irqs[i].mp_dstirq == pin)
958 * Find the pin to which IRQ[irq] (ISA) is connected
960 static int __init find_isa_irq_pin(int irq, int type)
964 for (i = 0; i < mp_irq_entries; i++) {
965 int lbus = mp_irqs[i].mp_srcbus;
967 if (test_bit(lbus, mp_bus_not_pci) &&
968 (mp_irqs[i].mp_irqtype == type) &&
969 (mp_irqs[i].mp_srcbusirq == irq))
971 return mp_irqs[i].mp_dstirq;
976 static int __init find_isa_irq_apic(int irq, int type)
980 for (i = 0; i < mp_irq_entries; i++) {
981 int lbus = mp_irqs[i].mp_srcbus;
983 if (test_bit(lbus, mp_bus_not_pci) &&
984 (mp_irqs[i].mp_irqtype == type) &&
985 (mp_irqs[i].mp_srcbusirq == irq))
988 if (i < mp_irq_entries) {
990 for(apic = 0; apic < nr_ioapics; apic++) {
991 if (mp_ioapics[apic].mp_apicid == mp_irqs[i].mp_dstapic)
1000 * Find a specific PCI IRQ entry.
1001 * Not an __init, possibly needed by modules
1003 static int pin_2_irq(int idx, int apic, int pin);
1005 int IO_APIC_get_PCI_irq_vector(int bus, int slot, int pin)
1007 int apic, i, best_guess = -1;
1009 apic_printk(APIC_DEBUG, "querying PCI -> IRQ mapping bus:%d, slot:%d, pin:%d.\n",
1011 if (test_bit(bus, mp_bus_not_pci)) {
1012 apic_printk(APIC_VERBOSE, "PCI BIOS passed nonexistent PCI bus %d!\n", bus);
1015 for (i = 0; i < mp_irq_entries; i++) {
1016 int lbus = mp_irqs[i].mp_srcbus;
1018 for (apic = 0; apic < nr_ioapics; apic++)
1019 if (mp_ioapics[apic].mp_apicid == mp_irqs[i].mp_dstapic ||
1020 mp_irqs[i].mp_dstapic == MP_APIC_ALL)
1023 if (!test_bit(lbus, mp_bus_not_pci) &&
1024 !mp_irqs[i].mp_irqtype &&
1026 (slot == ((mp_irqs[i].mp_srcbusirq >> 2) & 0x1f))) {
1027 int irq = pin_2_irq(i,apic,mp_irqs[i].mp_dstirq);
1029 if (!(apic || IO_APIC_IRQ(irq)))
1032 if (pin == (mp_irqs[i].mp_srcbusirq & 3))
1035 * Use the first all-but-pin matching entry as a
1036 * best-guess fuzzy result for broken mptables.
1045 EXPORT_SYMBOL(IO_APIC_get_PCI_irq_vector);
1047 #if defined(CONFIG_EISA) || defined(CONFIG_MCA)
1049 * EISA Edge/Level control register, ELCR
1051 static int EISA_ELCR(unsigned int irq)
1053 if (irq < NR_IRQS_LEGACY) {
1054 unsigned int port = 0x4d0 + (irq >> 3);
1055 return (inb(port) >> (irq & 7)) & 1;
1057 apic_printk(APIC_VERBOSE, KERN_INFO
1058 "Broken MPtable reports ISA irq %d\n", irq);
1064 /* ISA interrupts are always polarity zero edge triggered,
1065 * when listed as conforming in the MP table. */
1067 #define default_ISA_trigger(idx) (0)
1068 #define default_ISA_polarity(idx) (0)
1070 /* EISA interrupts are always polarity zero and can be edge or level
1071 * trigger depending on the ELCR value. If an interrupt is listed as
1072 * EISA conforming in the MP table, that means its trigger type must
1073 * be read in from the ELCR */
1075 #define default_EISA_trigger(idx) (EISA_ELCR(mp_irqs[idx].mp_srcbusirq))
1076 #define default_EISA_polarity(idx) default_ISA_polarity(idx)
1078 /* PCI interrupts are always polarity one level triggered,
1079 * when listed as conforming in the MP table. */
1081 #define default_PCI_trigger(idx) (1)
1082 #define default_PCI_polarity(idx) (1)
1084 /* MCA interrupts are always polarity zero level triggered,
1085 * when listed as conforming in the MP table. */
1087 #define default_MCA_trigger(idx) (1)
1088 #define default_MCA_polarity(idx) default_ISA_polarity(idx)
1090 static int MPBIOS_polarity(int idx)
1092 int bus = mp_irqs[idx].mp_srcbus;
1096 * Determine IRQ line polarity (high active or low active):
1098 switch (mp_irqs[idx].mp_irqflag & 3)
1100 case 0: /* conforms, ie. bus-type dependent polarity */
1101 if (test_bit(bus, mp_bus_not_pci))
1102 polarity = default_ISA_polarity(idx);
1104 polarity = default_PCI_polarity(idx);
1106 case 1: /* high active */
1111 case 2: /* reserved */
1113 printk(KERN_WARNING "broken BIOS!!\n");
1117 case 3: /* low active */
1122 default: /* invalid */
1124 printk(KERN_WARNING "broken BIOS!!\n");
1132 static int MPBIOS_trigger(int idx)
1134 int bus = mp_irqs[idx].mp_srcbus;
1138 * Determine IRQ trigger mode (edge or level sensitive):
1140 switch ((mp_irqs[idx].mp_irqflag>>2) & 3)
1142 case 0: /* conforms, ie. bus-type dependent */
1143 if (test_bit(bus, mp_bus_not_pci))
1144 trigger = default_ISA_trigger(idx);
1146 trigger = default_PCI_trigger(idx);
1147 #if defined(CONFIG_EISA) || defined(CONFIG_MCA)
1148 switch (mp_bus_id_to_type[bus]) {
1149 case MP_BUS_ISA: /* ISA pin */
1151 /* set before the switch */
1154 case MP_BUS_EISA: /* EISA pin */
1156 trigger = default_EISA_trigger(idx);
1159 case MP_BUS_PCI: /* PCI pin */
1161 /* set before the switch */
1164 case MP_BUS_MCA: /* MCA pin */
1166 trigger = default_MCA_trigger(idx);
1171 printk(KERN_WARNING "broken BIOS!!\n");
1183 case 2: /* reserved */
1185 printk(KERN_WARNING "broken BIOS!!\n");
1194 default: /* invalid */
1196 printk(KERN_WARNING "broken BIOS!!\n");
1204 static inline int irq_polarity(int idx)
1206 return MPBIOS_polarity(idx);
1209 static inline int irq_trigger(int idx)
1211 return MPBIOS_trigger(idx);
1214 int (*ioapic_renumber_irq)(int ioapic, int irq);
1215 static int pin_2_irq(int idx, int apic, int pin)
1218 int bus = mp_irqs[idx].mp_srcbus;
1221 * Debugging check, we are in big trouble if this message pops up!
1223 if (mp_irqs[idx].mp_dstirq != pin)
1224 printk(KERN_ERR "broken BIOS or MPTABLE parser, ayiee!!\n");
1226 if (test_bit(bus, mp_bus_not_pci)) {
1227 irq = mp_irqs[idx].mp_srcbusirq;
1230 * PCI IRQs are mapped in order
1234 irq += nr_ioapic_registers[i++];
1237 * For MPS mode, so far only needed by ES7000 platform
1239 if (ioapic_renumber_irq)
1240 irq = ioapic_renumber_irq(apic, irq);
1243 #ifdef CONFIG_X86_32
1245 * PCI IRQ command line redirection. Yes, limits are hardcoded.
1247 if ((pin >= 16) && (pin <= 23)) {
1248 if (pirq_entries[pin-16] != -1) {
1249 if (!pirq_entries[pin-16]) {
1250 apic_printk(APIC_VERBOSE, KERN_DEBUG
1251 "disabling PIRQ%d\n", pin-16);
1253 irq = pirq_entries[pin-16];
1254 apic_printk(APIC_VERBOSE, KERN_DEBUG
1255 "using PIRQ%d -> IRQ %d\n",
1265 void lock_vector_lock(void)
1267 /* Used to the online set of cpus does not change
1268 * during assign_irq_vector.
1270 spin_lock(&vector_lock);
1273 void unlock_vector_lock(void)
1275 spin_unlock(&vector_lock);
1279 __assign_irq_vector(int irq, struct irq_cfg *cfg, const struct cpumask *mask)
1282 * NOTE! The local APIC isn't very good at handling
1283 * multiple interrupts at the same interrupt level.
1284 * As the interrupt level is determined by taking the
1285 * vector number and shifting that right by 4, we
1286 * want to spread these out a bit so that they don't
1287 * all fall in the same interrupt level.
1289 * Also, we've got to be careful not to trash gate
1290 * 0x80, because int 0x80 is hm, kind of importantish. ;)
1292 static int current_vector = FIRST_DEVICE_VECTOR, current_offset = 0;
1293 unsigned int old_vector;
1295 cpumask_var_t tmp_mask;
1297 if ((cfg->move_in_progress) || cfg->move_cleanup_count)
1300 if (!alloc_cpumask_var(&tmp_mask, GFP_ATOMIC))
1303 old_vector = cfg->vector;
1305 cpumask_and(tmp_mask, mask, cpu_online_mask);
1306 cpumask_and(tmp_mask, cfg->domain, tmp_mask);
1307 if (!cpumask_empty(tmp_mask)) {
1308 free_cpumask_var(tmp_mask);
1313 /* Only try and allocate irqs on cpus that are present */
1315 for_each_cpu_and(cpu, mask, cpu_online_mask) {
1319 vector_allocation_domain(cpu, tmp_mask);
1321 vector = current_vector;
1322 offset = current_offset;
1325 if (vector >= first_system_vector) {
1326 /* If out of vectors on large boxen, must share them. */
1327 offset = (offset + 1) % 8;
1328 vector = FIRST_DEVICE_VECTOR + offset;
1330 if (unlikely(current_vector == vector))
1333 if (test_bit(vector, used_vectors))
1336 for_each_cpu_and(new_cpu, tmp_mask, cpu_online_mask)
1337 if (per_cpu(vector_irq, new_cpu)[vector] != -1)
1340 current_vector = vector;
1341 current_offset = offset;
1343 cfg->move_in_progress = 1;
1344 cpumask_copy(cfg->old_domain, cfg->domain);
1346 for_each_cpu_and(new_cpu, tmp_mask, cpu_online_mask)
1347 per_cpu(vector_irq, new_cpu)[vector] = irq;
1348 cfg->vector = vector;
1349 cpumask_copy(cfg->domain, tmp_mask);
1353 free_cpumask_var(tmp_mask);
1358 assign_irq_vector(int irq, struct irq_cfg *cfg, const struct cpumask *mask)
1361 unsigned long flags;
1363 spin_lock_irqsave(&vector_lock, flags);
1364 err = __assign_irq_vector(irq, cfg, mask);
1365 spin_unlock_irqrestore(&vector_lock, flags);
1369 static void __clear_irq_vector(int irq, struct irq_cfg *cfg)
1373 BUG_ON(!cfg->vector);
1375 vector = cfg->vector;
1376 for_each_cpu_and(cpu, cfg->domain, cpu_online_mask)
1377 per_cpu(vector_irq, cpu)[vector] = -1;
1380 cpumask_clear(cfg->domain);
1382 if (likely(!cfg->move_in_progress))
1384 for_each_cpu_and(cpu, cfg->old_domain, cpu_online_mask) {
1385 for (vector = FIRST_EXTERNAL_VECTOR; vector < NR_VECTORS;
1387 if (per_cpu(vector_irq, cpu)[vector] != irq)
1389 per_cpu(vector_irq, cpu)[vector] = -1;
1393 cfg->move_in_progress = 0;
1396 void __setup_vector_irq(int cpu)
1398 /* Initialize vector_irq on a new cpu */
1399 /* This function must be called with vector_lock held */
1401 struct irq_cfg *cfg;
1402 struct irq_desc *desc;
1404 /* Mark the inuse vectors */
1405 for_each_irq_desc(irq, desc) {
1406 cfg = desc->chip_data;
1407 if (!cpumask_test_cpu(cpu, cfg->domain))
1409 vector = cfg->vector;
1410 per_cpu(vector_irq, cpu)[vector] = irq;
1412 /* Mark the free vectors */
1413 for (vector = 0; vector < NR_VECTORS; ++vector) {
1414 irq = per_cpu(vector_irq, cpu)[vector];
1419 if (!cpumask_test_cpu(cpu, cfg->domain))
1420 per_cpu(vector_irq, cpu)[vector] = -1;
1424 static struct irq_chip ioapic_chip;
1425 #ifdef CONFIG_INTR_REMAP
1426 static struct irq_chip ir_ioapic_chip;
1429 #define IOAPIC_AUTO -1
1430 #define IOAPIC_EDGE 0
1431 #define IOAPIC_LEVEL 1
1433 #ifdef CONFIG_X86_32
1434 static inline int IO_APIC_irq_trigger(int irq)
1438 for (apic = 0; apic < nr_ioapics; apic++) {
1439 for (pin = 0; pin < nr_ioapic_registers[apic]; pin++) {
1440 idx = find_irq_entry(apic, pin, mp_INT);
1441 if ((idx != -1) && (irq == pin_2_irq(idx, apic, pin)))
1442 return irq_trigger(idx);
1446 * nonexistent IRQs are edge default
1451 static inline int IO_APIC_irq_trigger(int irq)
1457 static void ioapic_register_intr(int irq, struct irq_desc *desc, unsigned long trigger)
1460 if ((trigger == IOAPIC_AUTO && IO_APIC_irq_trigger(irq)) ||
1461 trigger == IOAPIC_LEVEL)
1462 desc->status |= IRQ_LEVEL;
1464 desc->status &= ~IRQ_LEVEL;
1466 #ifdef CONFIG_INTR_REMAP
1467 if (irq_remapped(irq)) {
1468 desc->status |= IRQ_MOVE_PCNTXT;
1470 set_irq_chip_and_handler_name(irq, &ir_ioapic_chip,
1474 set_irq_chip_and_handler_name(irq, &ir_ioapic_chip,
1475 handle_edge_irq, "edge");
1479 if ((trigger == IOAPIC_AUTO && IO_APIC_irq_trigger(irq)) ||
1480 trigger == IOAPIC_LEVEL)
1481 set_irq_chip_and_handler_name(irq, &ioapic_chip,
1485 set_irq_chip_and_handler_name(irq, &ioapic_chip,
1486 handle_edge_irq, "edge");
1489 static int setup_ioapic_entry(int apic, int irq,
1490 struct IO_APIC_route_entry *entry,
1491 unsigned int destination, int trigger,
1492 int polarity, int vector)
1495 * add it to the IO-APIC irq-routing table:
1497 memset(entry,0,sizeof(*entry));
1499 #ifdef CONFIG_INTR_REMAP
1500 if (intr_remapping_enabled) {
1501 struct intel_iommu *iommu = map_ioapic_to_ir(apic);
1503 struct IR_IO_APIC_route_entry *ir_entry =
1504 (struct IR_IO_APIC_route_entry *) entry;
1508 panic("No mapping iommu for ioapic %d\n", apic);
1510 index = alloc_irte(iommu, irq, 1);
1512 panic("Failed to allocate IRTE for ioapic %d\n", apic);
1514 memset(&irte, 0, sizeof(irte));
1517 irte.dst_mode = INT_DEST_MODE;
1518 irte.trigger_mode = trigger;
1519 irte.dlvry_mode = INT_DELIVERY_MODE;
1520 irte.vector = vector;
1521 irte.dest_id = IRTE_DEST(destination);
1523 modify_irte(irq, &irte);
1525 ir_entry->index2 = (index >> 15) & 0x1;
1527 ir_entry->format = 1;
1528 ir_entry->index = (index & 0x7fff);
1532 entry->delivery_mode = INT_DELIVERY_MODE;
1533 entry->dest_mode = INT_DEST_MODE;
1534 entry->dest = destination;
1537 entry->mask = 0; /* enable IRQ */
1538 entry->trigger = trigger;
1539 entry->polarity = polarity;
1540 entry->vector = vector;
1542 /* Mask level triggered irqs.
1543 * Use IRQ_DELAYED_DISABLE for edge triggered irqs.
1550 static void setup_IO_APIC_irq(int apic, int pin, unsigned int irq, struct irq_desc *desc,
1551 int trigger, int polarity)
1553 struct irq_cfg *cfg;
1554 struct IO_APIC_route_entry entry;
1557 if (!IO_APIC_IRQ(irq))
1560 cfg = desc->chip_data;
1562 if (assign_irq_vector(irq, cfg, TARGET_CPUS))
1565 dest = cpu_mask_to_apicid_and(cfg->domain, TARGET_CPUS);
1567 apic_printk(APIC_VERBOSE,KERN_DEBUG
1568 "IOAPIC[%d]: Set routing entry (%d-%d -> 0x%x -> "
1569 "IRQ %d Mode:%i Active:%i)\n",
1570 apic, mp_ioapics[apic].mp_apicid, pin, cfg->vector,
1571 irq, trigger, polarity);
1574 if (setup_ioapic_entry(mp_ioapics[apic].mp_apicid, irq, &entry,
1575 dest, trigger, polarity, cfg->vector)) {
1576 printk("Failed to setup ioapic entry for ioapic %d, pin %d\n",
1577 mp_ioapics[apic].mp_apicid, pin);
1578 __clear_irq_vector(irq, cfg);
1582 ioapic_register_intr(irq, desc, trigger);
1583 if (irq < NR_IRQS_LEGACY)
1584 disable_8259A_irq(irq);
1586 ioapic_write_entry(apic, pin, entry);
1589 static void __init setup_IO_APIC_irqs(void)
1591 int apic, pin, idx, irq;
1593 struct irq_desc *desc;
1594 struct irq_cfg *cfg;
1595 int cpu = boot_cpu_id;
1597 apic_printk(APIC_VERBOSE, KERN_DEBUG "init IO_APIC IRQs\n");
1599 for (apic = 0; apic < nr_ioapics; apic++) {
1600 for (pin = 0; pin < nr_ioapic_registers[apic]; pin++) {
1602 idx = find_irq_entry(apic, pin, mp_INT);
1606 apic_printk(APIC_VERBOSE,
1607 KERN_DEBUG " %d-%d",
1608 mp_ioapics[apic].mp_apicid,
1611 apic_printk(APIC_VERBOSE, " %d-%d",
1612 mp_ioapics[apic].mp_apicid,
1617 apic_printk(APIC_VERBOSE,
1618 " (apicid-pin) not connected\n");
1622 irq = pin_2_irq(idx, apic, pin);
1623 #ifdef CONFIG_X86_32
1624 if (multi_timer_check(apic, irq))
1627 desc = irq_to_desc_alloc_cpu(irq, cpu);
1629 printk(KERN_INFO "can not get irq_desc for %d\n", irq);
1632 cfg = desc->chip_data;
1633 add_pin_to_irq_cpu(cfg, cpu, apic, pin);
1635 setup_IO_APIC_irq(apic, pin, irq, desc,
1636 irq_trigger(idx), irq_polarity(idx));
1641 apic_printk(APIC_VERBOSE,
1642 " (apicid-pin) not connected\n");
1646 * Set up the timer pin, possibly with the 8259A-master behind.
1648 static void __init setup_timer_IRQ0_pin(unsigned int apic, unsigned int pin,
1651 struct IO_APIC_route_entry entry;
1653 #ifdef CONFIG_INTR_REMAP
1654 if (intr_remapping_enabled)
1658 memset(&entry, 0, sizeof(entry));
1661 * We use logical delivery to get the timer IRQ
1664 entry.dest_mode = INT_DEST_MODE;
1665 entry.mask = 1; /* mask IRQ now */
1666 entry.dest = cpu_mask_to_apicid(TARGET_CPUS);
1667 entry.delivery_mode = INT_DELIVERY_MODE;
1670 entry.vector = vector;
1673 * The timer IRQ doesn't have to know that behind the
1674 * scene we may have a 8259A-master in AEOI mode ...
1676 set_irq_chip_and_handler_name(0, &ioapic_chip, handle_edge_irq, "edge");
1679 * Add it to the IO-APIC irq-routing table:
1681 ioapic_write_entry(apic, pin, entry);
1685 __apicdebuginit(void) print_IO_APIC(void)
1688 union IO_APIC_reg_00 reg_00;
1689 union IO_APIC_reg_01 reg_01;
1690 union IO_APIC_reg_02 reg_02;
1691 union IO_APIC_reg_03 reg_03;
1692 unsigned long flags;
1693 struct irq_cfg *cfg;
1694 struct irq_desc *desc;
1697 if (apic_verbosity == APIC_QUIET)
1700 printk(KERN_DEBUG "number of MP IRQ sources: %d.\n", mp_irq_entries);
1701 for (i = 0; i < nr_ioapics; i++)
1702 printk(KERN_DEBUG "number of IO-APIC #%d registers: %d.\n",
1703 mp_ioapics[i].mp_apicid, nr_ioapic_registers[i]);
1706 * We are a bit conservative about what we expect. We have to
1707 * know about every hardware change ASAP.
1709 printk(KERN_INFO "testing the IO APIC.......................\n");
1711 for (apic = 0; apic < nr_ioapics; apic++) {
1713 spin_lock_irqsave(&ioapic_lock, flags);
1714 reg_00.raw = io_apic_read(apic, 0);
1715 reg_01.raw = io_apic_read(apic, 1);
1716 if (reg_01.bits.version >= 0x10)
1717 reg_02.raw = io_apic_read(apic, 2);
1718 if (reg_01.bits.version >= 0x20)
1719 reg_03.raw = io_apic_read(apic, 3);
1720 spin_unlock_irqrestore(&ioapic_lock, flags);
1723 printk(KERN_DEBUG "IO APIC #%d......\n", mp_ioapics[apic].mp_apicid);
1724 printk(KERN_DEBUG ".... register #00: %08X\n", reg_00.raw);
1725 printk(KERN_DEBUG "....... : physical APIC id: %02X\n", reg_00.bits.ID);
1726 printk(KERN_DEBUG "....... : Delivery Type: %X\n", reg_00.bits.delivery_type);
1727 printk(KERN_DEBUG "....... : LTS : %X\n", reg_00.bits.LTS);
1729 printk(KERN_DEBUG ".... register #01: %08X\n", *(int *)®_01);
1730 printk(KERN_DEBUG "....... : max redirection entries: %04X\n", reg_01.bits.entries);
1732 printk(KERN_DEBUG "....... : PRQ implemented: %X\n", reg_01.bits.PRQ);
1733 printk(KERN_DEBUG "....... : IO APIC version: %04X\n", reg_01.bits.version);
1736 * Some Intel chipsets with IO APIC VERSION of 0x1? don't have reg_02,
1737 * but the value of reg_02 is read as the previous read register
1738 * value, so ignore it if reg_02 == reg_01.
1740 if (reg_01.bits.version >= 0x10 && reg_02.raw != reg_01.raw) {
1741 printk(KERN_DEBUG ".... register #02: %08X\n", reg_02.raw);
1742 printk(KERN_DEBUG "....... : arbitration: %02X\n", reg_02.bits.arbitration);
1746 * Some Intel chipsets with IO APIC VERSION of 0x2? don't have reg_02
1747 * or reg_03, but the value of reg_0[23] is read as the previous read
1748 * register value, so ignore it if reg_03 == reg_0[12].
1750 if (reg_01.bits.version >= 0x20 && reg_03.raw != reg_02.raw &&
1751 reg_03.raw != reg_01.raw) {
1752 printk(KERN_DEBUG ".... register #03: %08X\n", reg_03.raw);
1753 printk(KERN_DEBUG "....... : Boot DT : %X\n", reg_03.bits.boot_DT);
1756 printk(KERN_DEBUG ".... IRQ redirection table:\n");
1758 printk(KERN_DEBUG " NR Dst Mask Trig IRR Pol"
1759 " Stat Dmod Deli Vect: \n");
1761 for (i = 0; i <= reg_01.bits.entries; i++) {
1762 struct IO_APIC_route_entry entry;
1764 entry = ioapic_read_entry(apic, i);
1766 printk(KERN_DEBUG " %02x %03X ",
1771 printk("%1d %1d %1d %1d %1d %1d %1d %02X\n",
1776 entry.delivery_status,
1778 entry.delivery_mode,
1783 printk(KERN_DEBUG "IRQ to pin mappings:\n");
1784 for_each_irq_desc(irq, desc) {
1785 struct irq_pin_list *entry;
1787 cfg = desc->chip_data;
1788 entry = cfg->irq_2_pin;
1791 printk(KERN_DEBUG "IRQ%d ", irq);
1793 printk("-> %d:%d", entry->apic, entry->pin);
1796 entry = entry->next;
1801 printk(KERN_INFO ".................................... done.\n");
1806 __apicdebuginit(void) print_APIC_bitfield(int base)
1811 if (apic_verbosity == APIC_QUIET)
1814 printk(KERN_DEBUG "0123456789abcdef0123456789abcdef\n" KERN_DEBUG);
1815 for (i = 0; i < 8; i++) {
1816 v = apic_read(base + i*0x10);
1817 for (j = 0; j < 32; j++) {
1827 __apicdebuginit(void) print_local_APIC(void *dummy)
1829 unsigned int v, ver, maxlvt;
1832 if (apic_verbosity == APIC_QUIET)
1835 printk("\n" KERN_DEBUG "printing local APIC contents on CPU#%d/%d:\n",
1836 smp_processor_id(), hard_smp_processor_id());
1837 v = apic_read(APIC_ID);
1838 printk(KERN_INFO "... APIC ID: %08x (%01x)\n", v, read_apic_id());
1839 v = apic_read(APIC_LVR);
1840 printk(KERN_INFO "... APIC VERSION: %08x\n", v);
1841 ver = GET_APIC_VERSION(v);
1842 maxlvt = lapic_get_maxlvt();
1844 v = apic_read(APIC_TASKPRI);
1845 printk(KERN_DEBUG "... APIC TASKPRI: %08x (%02x)\n", v, v & APIC_TPRI_MASK);
1847 if (APIC_INTEGRATED(ver)) { /* !82489DX */
1848 if (!APIC_XAPIC(ver)) {
1849 v = apic_read(APIC_ARBPRI);
1850 printk(KERN_DEBUG "... APIC ARBPRI: %08x (%02x)\n", v,
1851 v & APIC_ARBPRI_MASK);
1853 v = apic_read(APIC_PROCPRI);
1854 printk(KERN_DEBUG "... APIC PROCPRI: %08x\n", v);
1858 * Remote read supported only in the 82489DX and local APIC for
1859 * Pentium processors.
1861 if (!APIC_INTEGRATED(ver) || maxlvt == 3) {
1862 v = apic_read(APIC_RRR);
1863 printk(KERN_DEBUG "... APIC RRR: %08x\n", v);
1866 v = apic_read(APIC_LDR);
1867 printk(KERN_DEBUG "... APIC LDR: %08x\n", v);
1868 if (!x2apic_enabled()) {
1869 v = apic_read(APIC_DFR);
1870 printk(KERN_DEBUG "... APIC DFR: %08x\n", v);
1872 v = apic_read(APIC_SPIV);
1873 printk(KERN_DEBUG "... APIC SPIV: %08x\n", v);
1875 printk(KERN_DEBUG "... APIC ISR field:\n");
1876 print_APIC_bitfield(APIC_ISR);
1877 printk(KERN_DEBUG "... APIC TMR field:\n");
1878 print_APIC_bitfield(APIC_TMR);
1879 printk(KERN_DEBUG "... APIC IRR field:\n");
1880 print_APIC_bitfield(APIC_IRR);
1882 if (APIC_INTEGRATED(ver)) { /* !82489DX */
1883 if (maxlvt > 3) /* Due to the Pentium erratum 3AP. */
1884 apic_write(APIC_ESR, 0);
1886 v = apic_read(APIC_ESR);
1887 printk(KERN_DEBUG "... APIC ESR: %08x\n", v);
1890 icr = apic_icr_read();
1891 printk(KERN_DEBUG "... APIC ICR: %08x\n", (u32)icr);
1892 printk(KERN_DEBUG "... APIC ICR2: %08x\n", (u32)(icr >> 32));
1894 v = apic_read(APIC_LVTT);
1895 printk(KERN_DEBUG "... APIC LVTT: %08x\n", v);
1897 if (maxlvt > 3) { /* PC is LVT#4. */
1898 v = apic_read(APIC_LVTPC);
1899 printk(KERN_DEBUG "... APIC LVTPC: %08x\n", v);
1901 v = apic_read(APIC_LVT0);
1902 printk(KERN_DEBUG "... APIC LVT0: %08x\n", v);
1903 v = apic_read(APIC_LVT1);
1904 printk(KERN_DEBUG "... APIC LVT1: %08x\n", v);
1906 if (maxlvt > 2) { /* ERR is LVT#3. */
1907 v = apic_read(APIC_LVTERR);
1908 printk(KERN_DEBUG "... APIC LVTERR: %08x\n", v);
1911 v = apic_read(APIC_TMICT);
1912 printk(KERN_DEBUG "... APIC TMICT: %08x\n", v);
1913 v = apic_read(APIC_TMCCT);
1914 printk(KERN_DEBUG "... APIC TMCCT: %08x\n", v);
1915 v = apic_read(APIC_TDCR);
1916 printk(KERN_DEBUG "... APIC TDCR: %08x\n", v);
1920 __apicdebuginit(void) print_all_local_APICs(void)
1925 for_each_online_cpu(cpu)
1926 smp_call_function_single(cpu, print_local_APIC, NULL, 1);
1930 __apicdebuginit(void) print_PIC(void)
1933 unsigned long flags;
1935 if (apic_verbosity == APIC_QUIET)
1938 printk(KERN_DEBUG "\nprinting PIC contents\n");
1940 spin_lock_irqsave(&i8259A_lock, flags);
1942 v = inb(0xa1) << 8 | inb(0x21);
1943 printk(KERN_DEBUG "... PIC IMR: %04x\n", v);
1945 v = inb(0xa0) << 8 | inb(0x20);
1946 printk(KERN_DEBUG "... PIC IRR: %04x\n", v);
1950 v = inb(0xa0) << 8 | inb(0x20);
1954 spin_unlock_irqrestore(&i8259A_lock, flags);
1956 printk(KERN_DEBUG "... PIC ISR: %04x\n", v);
1958 v = inb(0x4d1) << 8 | inb(0x4d0);
1959 printk(KERN_DEBUG "... PIC ELCR: %04x\n", v);
1962 __apicdebuginit(int) print_all_ICs(void)
1965 print_all_local_APICs();
1971 fs_initcall(print_all_ICs);
1974 /* Where if anywhere is the i8259 connect in external int mode */
1975 static struct { int pin, apic; } ioapic_i8259 = { -1, -1 };
1977 void __init enable_IO_APIC(void)
1979 union IO_APIC_reg_01 reg_01;
1980 int i8259_apic, i8259_pin;
1982 unsigned long flags;
1984 #ifdef CONFIG_X86_32
1987 for (i = 0; i < MAX_PIRQS; i++)
1988 pirq_entries[i] = -1;
1992 * The number of IO-APIC IRQ registers (== #pins):
1994 for (apic = 0; apic < nr_ioapics; apic++) {
1995 spin_lock_irqsave(&ioapic_lock, flags);
1996 reg_01.raw = io_apic_read(apic, 1);
1997 spin_unlock_irqrestore(&ioapic_lock, flags);
1998 nr_ioapic_registers[apic] = reg_01.bits.entries+1;
2000 for(apic = 0; apic < nr_ioapics; apic++) {
2002 /* See if any of the pins is in ExtINT mode */
2003 for (pin = 0; pin < nr_ioapic_registers[apic]; pin++) {
2004 struct IO_APIC_route_entry entry;
2005 entry = ioapic_read_entry(apic, pin);
2007 /* If the interrupt line is enabled and in ExtInt mode
2008 * I have found the pin where the i8259 is connected.
2010 if ((entry.mask == 0) && (entry.delivery_mode == dest_ExtINT)) {
2011 ioapic_i8259.apic = apic;
2012 ioapic_i8259.pin = pin;
2018 /* Look to see what if the MP table has reported the ExtINT */
2019 /* If we could not find the appropriate pin by looking at the ioapic
2020 * the i8259 probably is not connected the ioapic but give the
2021 * mptable a chance anyway.
2023 i8259_pin = find_isa_irq_pin(0, mp_ExtINT);
2024 i8259_apic = find_isa_irq_apic(0, mp_ExtINT);
2025 /* Trust the MP table if nothing is setup in the hardware */
2026 if ((ioapic_i8259.pin == -1) && (i8259_pin >= 0)) {
2027 printk(KERN_WARNING "ExtINT not setup in hardware but reported by MP table\n");
2028 ioapic_i8259.pin = i8259_pin;
2029 ioapic_i8259.apic = i8259_apic;
2031 /* Complain if the MP table and the hardware disagree */
2032 if (((ioapic_i8259.apic != i8259_apic) || (ioapic_i8259.pin != i8259_pin)) &&
2033 (i8259_pin >= 0) && (ioapic_i8259.pin >= 0))
2035 printk(KERN_WARNING "ExtINT in hardware and MP table differ\n");
2039 * Do not trust the IO-APIC being empty at bootup
2045 * Not an __init, needed by the reboot code
2047 void disable_IO_APIC(void)
2050 * Clear the IO-APIC before rebooting:
2055 * If the i8259 is routed through an IOAPIC
2056 * Put that IOAPIC in virtual wire mode
2057 * so legacy interrupts can be delivered.
2059 if (ioapic_i8259.pin != -1) {
2060 struct IO_APIC_route_entry entry;
2062 memset(&entry, 0, sizeof(entry));
2063 entry.mask = 0; /* Enabled */
2064 entry.trigger = 0; /* Edge */
2066 entry.polarity = 0; /* High */
2067 entry.delivery_status = 0;
2068 entry.dest_mode = 0; /* Physical */
2069 entry.delivery_mode = dest_ExtINT; /* ExtInt */
2071 entry.dest = read_apic_id();
2074 * Add it to the IO-APIC irq-routing table:
2076 ioapic_write_entry(ioapic_i8259.apic, ioapic_i8259.pin, entry);
2079 disconnect_bsp_APIC(ioapic_i8259.pin != -1);
2082 #ifdef CONFIG_X86_32
2084 * function to set the IO-APIC physical IDs based on the
2085 * values stored in the MPC table.
2087 * by Matt Domsch <Matt_Domsch@dell.com> Tue Dec 21 12:25:05 CST 1999
2090 static void __init setup_ioapic_ids_from_mpc(void)
2092 union IO_APIC_reg_00 reg_00;
2093 physid_mask_t phys_id_present_map;
2096 unsigned char old_id;
2097 unsigned long flags;
2099 if (x86_quirks->setup_ioapic_ids && x86_quirks->setup_ioapic_ids())
2103 * Don't check I/O APIC IDs for xAPIC systems. They have
2104 * no meaning without the serial APIC bus.
2106 if (!(boot_cpu_data.x86_vendor == X86_VENDOR_INTEL)
2107 || APIC_XAPIC(apic_version[boot_cpu_physical_apicid]))
2110 * This is broken; anything with a real cpu count has to
2111 * circumvent this idiocy regardless.
2113 phys_id_present_map = ioapic_phys_id_map(phys_cpu_present_map);
2116 * Set the IOAPIC ID to the value stored in the MPC table.
2118 for (apic = 0; apic < nr_ioapics; apic++) {
2120 /* Read the register 0 value */
2121 spin_lock_irqsave(&ioapic_lock, flags);
2122 reg_00.raw = io_apic_read(apic, 0);
2123 spin_unlock_irqrestore(&ioapic_lock, flags);
2125 old_id = mp_ioapics[apic].mp_apicid;
2127 if (mp_ioapics[apic].mp_apicid >= get_physical_broadcast()) {
2128 printk(KERN_ERR "BIOS bug, IO-APIC#%d ID is %d in the MPC table!...\n",
2129 apic, mp_ioapics[apic].mp_apicid);
2130 printk(KERN_ERR "... fixing up to %d. (tell your hw vendor)\n",
2132 mp_ioapics[apic].mp_apicid = reg_00.bits.ID;
2136 * Sanity check, is the ID really free? Every APIC in a
2137 * system must have a unique ID or we get lots of nice
2138 * 'stuck on smp_invalidate_needed IPI wait' messages.
2140 if (check_apicid_used(phys_id_present_map,
2141 mp_ioapics[apic].mp_apicid)) {
2142 printk(KERN_ERR "BIOS bug, IO-APIC#%d ID %d is already used!...\n",
2143 apic, mp_ioapics[apic].mp_apicid);
2144 for (i = 0; i < get_physical_broadcast(); i++)
2145 if (!physid_isset(i, phys_id_present_map))
2147 if (i >= get_physical_broadcast())
2148 panic("Max APIC ID exceeded!\n");
2149 printk(KERN_ERR "... fixing up to %d. (tell your hw vendor)\n",
2151 physid_set(i, phys_id_present_map);
2152 mp_ioapics[apic].mp_apicid = i;
2155 tmp = apicid_to_cpu_present(mp_ioapics[apic].mp_apicid);
2156 apic_printk(APIC_VERBOSE, "Setting %d in the "
2157 "phys_id_present_map\n",
2158 mp_ioapics[apic].mp_apicid);
2159 physids_or(phys_id_present_map, phys_id_present_map, tmp);
2164 * We need to adjust the IRQ routing table
2165 * if the ID changed.
2167 if (old_id != mp_ioapics[apic].mp_apicid)
2168 for (i = 0; i < mp_irq_entries; i++)
2169 if (mp_irqs[i].mp_dstapic == old_id)
2170 mp_irqs[i].mp_dstapic
2171 = mp_ioapics[apic].mp_apicid;
2174 * Read the right value from the MPC table and
2175 * write it into the ID register.
2177 apic_printk(APIC_VERBOSE, KERN_INFO
2178 "...changing IO-APIC physical APIC ID to %d ...",
2179 mp_ioapics[apic].mp_apicid);
2181 reg_00.bits.ID = mp_ioapics[apic].mp_apicid;
2182 spin_lock_irqsave(&ioapic_lock, flags);
2183 io_apic_write(apic, 0, reg_00.raw);
2184 spin_unlock_irqrestore(&ioapic_lock, flags);
2189 spin_lock_irqsave(&ioapic_lock, flags);
2190 reg_00.raw = io_apic_read(apic, 0);
2191 spin_unlock_irqrestore(&ioapic_lock, flags);
2192 if (reg_00.bits.ID != mp_ioapics[apic].mp_apicid)
2193 printk("could not set ID!\n");
2195 apic_printk(APIC_VERBOSE, " ok.\n");
2200 int no_timer_check __initdata;
2202 static int __init notimercheck(char *s)
2207 __setup("no_timer_check", notimercheck);
2210 * There is a nasty bug in some older SMP boards, their mptable lies
2211 * about the timer IRQ. We do the following to work around the situation:
2213 * - timer IRQ defaults to IO-APIC IRQ
2214 * - if this function detects that timer IRQs are defunct, then we fall
2215 * back to ISA timer IRQs
2217 static int __init timer_irq_works(void)
2219 unsigned long t1 = jiffies;
2220 unsigned long flags;
2225 local_save_flags(flags);
2227 /* Let ten ticks pass... */
2228 mdelay((10 * 1000) / HZ);
2229 local_irq_restore(flags);
2232 * Expect a few ticks at least, to be sure some possible
2233 * glue logic does not lock up after one or two first
2234 * ticks in a non-ExtINT mode. Also the local APIC
2235 * might have cached one ExtINT interrupt. Finally, at
2236 * least one tick may be lost due to delays.
2240 if (time_after(jiffies, t1 + 4))
2246 * In the SMP+IOAPIC case it might happen that there are an unspecified
2247 * number of pending IRQ events unhandled. These cases are very rare,
2248 * so we 'resend' these IRQs via IPIs, to the same CPU. It's much
2249 * better to do it this way as thus we do not have to be aware of
2250 * 'pending' interrupts in the IRQ path, except at this point.
2253 * Edge triggered needs to resend any interrupt
2254 * that was delayed but this is now handled in the device
2259 * Starting up a edge-triggered IO-APIC interrupt is
2260 * nasty - we need to make sure that we get the edge.
2261 * If it is already asserted for some reason, we need
2262 * return 1 to indicate that is was pending.
2264 * This is not complete - we should be able to fake
2265 * an edge even if it isn't on the 8259A...
2268 static unsigned int startup_ioapic_irq(unsigned int irq)
2270 int was_pending = 0;
2271 unsigned long flags;
2272 struct irq_cfg *cfg;
2274 spin_lock_irqsave(&ioapic_lock, flags);
2275 if (irq < NR_IRQS_LEGACY) {
2276 disable_8259A_irq(irq);
2277 if (i8259A_irq_pending(irq))
2281 __unmask_IO_APIC_irq(cfg);
2282 spin_unlock_irqrestore(&ioapic_lock, flags);
2287 #ifdef CONFIG_X86_64
2288 static int ioapic_retrigger_irq(unsigned int irq)
2291 struct irq_cfg *cfg = irq_cfg(irq);
2292 unsigned long flags;
2294 spin_lock_irqsave(&vector_lock, flags);
2295 send_IPI_mask(cpumask_of(cpumask_first(cfg->domain)), cfg->vector);
2296 spin_unlock_irqrestore(&vector_lock, flags);
2301 static int ioapic_retrigger_irq(unsigned int irq)
2303 send_IPI_self(irq_cfg(irq)->vector);
2310 * Level and edge triggered IO-APIC interrupts need different handling,
2311 * so we use two separate IRQ descriptors. Edge triggered IRQs can be
2312 * handled with the level-triggered descriptor, but that one has slightly
2313 * more overhead. Level-triggered interrupts cannot be handled with the
2314 * edge-triggered handler, without risking IRQ storms and other ugly
2320 #ifdef CONFIG_INTR_REMAP
2321 static void ir_irq_migration(struct work_struct *work);
2323 static DECLARE_DELAYED_WORK(ir_migration_work, ir_irq_migration);
2326 * Migrate the IO-APIC irq in the presence of intr-remapping.
2328 * For edge triggered, irq migration is a simple atomic update(of vector
2329 * and cpu destination) of IRTE and flush the hardware cache.
2331 * For level triggered, we need to modify the io-apic RTE aswell with the update
2332 * vector information, along with modifying IRTE with vector and destination.
2333 * So irq migration for level triggered is little bit more complex compared to
2334 * edge triggered migration. But the good news is, we use the same algorithm
2335 * for level triggered migration as we have today, only difference being,
2336 * we now initiate the irq migration from process context instead of the
2337 * interrupt context.
2339 * In future, when we do a directed EOI (combined with cpu EOI broadcast
2340 * suppression) to the IO-APIC, level triggered irq migration will also be
2341 * as simple as edge triggered migration and we can do the irq migration
2342 * with a simple atomic update to IO-APIC RTE.
2345 migrate_ioapic_irq_desc(struct irq_desc *desc, const struct cpumask *mask)
2347 struct irq_cfg *cfg;
2349 int modify_ioapic_rte;
2351 unsigned long flags;
2354 if (!cpumask_intersects(mask, cpu_online_mask))
2358 if (get_irte(irq, &irte))
2361 cfg = desc->chip_data;
2362 if (assign_irq_vector(irq, cfg, mask))
2365 set_extra_move_desc(desc, mask);
2367 dest = cpu_mask_to_apicid_and(cfg->domain, mask);
2369 modify_ioapic_rte = desc->status & IRQ_LEVEL;
2370 if (modify_ioapic_rte) {
2371 spin_lock_irqsave(&ioapic_lock, flags);
2372 __target_IO_APIC_irq(irq, dest, cfg);
2373 spin_unlock_irqrestore(&ioapic_lock, flags);
2376 irte.vector = cfg->vector;
2377 irte.dest_id = IRTE_DEST(dest);
2380 * Modified the IRTE and flushes the Interrupt entry cache.
2382 modify_irte(irq, &irte);
2384 if (cfg->move_in_progress)
2385 send_cleanup_vector(cfg);
2387 cpumask_copy(&desc->affinity, mask);
2390 static int migrate_irq_remapped_level_desc(struct irq_desc *desc)
2393 struct irq_cfg *cfg = desc->chip_data;
2395 mask_IO_APIC_irq_desc(desc);
2397 if (io_apic_level_ack_pending(cfg)) {
2399 * Interrupt in progress. Migrating irq now will change the
2400 * vector information in the IO-APIC RTE and that will confuse
2401 * the EOI broadcast performed by cpu.
2402 * So, delay the irq migration to the next instance.
2404 schedule_delayed_work(&ir_migration_work, 1);
2408 /* everthing is clear. we have right of way */
2409 migrate_ioapic_irq_desc(desc, &desc->pending_mask);
2412 desc->status &= ~IRQ_MOVE_PENDING;
2413 cpumask_clear(&desc->pending_mask);
2416 unmask_IO_APIC_irq_desc(desc);
2421 static void ir_irq_migration(struct work_struct *work)
2424 struct irq_desc *desc;
2426 for_each_irq_desc(irq, desc) {
2427 if (desc->status & IRQ_MOVE_PENDING) {
2428 unsigned long flags;
2430 spin_lock_irqsave(&desc->lock, flags);
2431 if (!desc->chip->set_affinity ||
2432 !(desc->status & IRQ_MOVE_PENDING)) {
2433 desc->status &= ~IRQ_MOVE_PENDING;
2434 spin_unlock_irqrestore(&desc->lock, flags);
2438 desc->chip->set_affinity(irq, &desc->pending_mask);
2439 spin_unlock_irqrestore(&desc->lock, flags);
2445 * Migrates the IRQ destination in the process context.
2447 static void set_ir_ioapic_affinity_irq_desc(struct irq_desc *desc,
2448 const struct cpumask *mask)
2450 if (desc->status & IRQ_LEVEL) {
2451 desc->status |= IRQ_MOVE_PENDING;
2452 cpumask_copy(&desc->pending_mask, mask);
2453 migrate_irq_remapped_level_desc(desc);
2457 migrate_ioapic_irq_desc(desc, mask);
2459 static void set_ir_ioapic_affinity_irq(unsigned int irq,
2460 const struct cpumask *mask)
2462 struct irq_desc *desc = irq_to_desc(irq);
2464 set_ir_ioapic_affinity_irq_desc(desc, mask);
2468 asmlinkage void smp_irq_move_cleanup_interrupt(void)
2470 unsigned vector, me;
2476 me = smp_processor_id();
2477 for (vector = FIRST_EXTERNAL_VECTOR; vector < NR_VECTORS; vector++) {
2479 struct irq_desc *desc;
2480 struct irq_cfg *cfg;
2481 irq = __get_cpu_var(vector_irq)[vector];
2486 desc = irq_to_desc(irq);
2491 spin_lock(&desc->lock);
2492 if (!cfg->move_cleanup_count)
2495 if (vector == cfg->vector && cpumask_test_cpu(me, cfg->domain))
2498 __get_cpu_var(vector_irq)[vector] = -1;
2499 cfg->move_cleanup_count--;
2501 spin_unlock(&desc->lock);
2507 static void irq_complete_move(struct irq_desc **descp)
2509 struct irq_desc *desc = *descp;
2510 struct irq_cfg *cfg = desc->chip_data;
2511 unsigned vector, me;
2513 if (likely(!cfg->move_in_progress)) {
2514 #ifdef CONFIG_NUMA_MIGRATE_IRQ_DESC
2515 if (likely(!cfg->move_desc_pending))
2518 /* domain has not changed, but affinity did */
2519 me = smp_processor_id();
2520 if (cpu_isset(me, desc->affinity)) {
2521 *descp = desc = move_irq_desc(desc, me);
2522 /* get the new one */
2523 cfg = desc->chip_data;
2524 cfg->move_desc_pending = 0;
2530 vector = ~get_irq_regs()->orig_ax;
2531 me = smp_processor_id();
2532 #ifdef CONFIG_NUMA_MIGRATE_IRQ_DESC
2533 *descp = desc = move_irq_desc(desc, me);
2534 /* get the new one */
2535 cfg = desc->chip_data;
2538 if (vector == cfg->vector && cpumask_test_cpu(me, cfg->domain))
2539 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)
3847 for (idx = 0; idx < nr_ioapics; idx++)
3848 nr += io_apic_get_redir_entries(idx) + 1;
3850 if (nr > nr_irqs_gsi)
3854 /* --------------------------------------------------------------------------
3855 ACPI-based IOAPIC Configuration
3856 -------------------------------------------------------------------------- */
3860 #ifdef CONFIG_X86_32
3861 int __init io_apic_get_unique_id(int ioapic, int apic_id)
3863 union IO_APIC_reg_00 reg_00;
3864 static physid_mask_t apic_id_map = PHYSID_MASK_NONE;
3866 unsigned long flags;
3870 * The P4 platform supports up to 256 APIC IDs on two separate APIC
3871 * buses (one for LAPICs, one for IOAPICs), where predecessors only
3872 * supports up to 16 on one shared APIC bus.
3874 * TBD: Expand LAPIC/IOAPIC support on P4-class systems to take full
3875 * advantage of new APIC bus architecture.
3878 if (physids_empty(apic_id_map))
3879 apic_id_map = ioapic_phys_id_map(phys_cpu_present_map);
3881 spin_lock_irqsave(&ioapic_lock, flags);
3882 reg_00.raw = io_apic_read(ioapic, 0);
3883 spin_unlock_irqrestore(&ioapic_lock, flags);
3885 if (apic_id >= get_physical_broadcast()) {
3886 printk(KERN_WARNING "IOAPIC[%d]: Invalid apic_id %d, trying "
3887 "%d\n", ioapic, apic_id, reg_00.bits.ID);
3888 apic_id = reg_00.bits.ID;
3892 * Every APIC in a system must have a unique ID or we get lots of nice
3893 * 'stuck on smp_invalidate_needed IPI wait' messages.
3895 if (check_apicid_used(apic_id_map, apic_id)) {
3897 for (i = 0; i < get_physical_broadcast(); i++) {
3898 if (!check_apicid_used(apic_id_map, i))
3902 if (i == get_physical_broadcast())
3903 panic("Max apic_id exceeded!\n");
3905 printk(KERN_WARNING "IOAPIC[%d]: apic_id %d already used, "
3906 "trying %d\n", ioapic, apic_id, i);
3911 tmp = apicid_to_cpu_present(apic_id);
3912 physids_or(apic_id_map, apic_id_map, tmp);
3914 if (reg_00.bits.ID != apic_id) {
3915 reg_00.bits.ID = apic_id;
3917 spin_lock_irqsave(&ioapic_lock, flags);
3918 io_apic_write(ioapic, 0, reg_00.raw);
3919 reg_00.raw = io_apic_read(ioapic, 0);
3920 spin_unlock_irqrestore(&ioapic_lock, flags);
3923 if (reg_00.bits.ID != apic_id) {
3924 printk("IOAPIC[%d]: Unable to change apic_id!\n", ioapic);
3929 apic_printk(APIC_VERBOSE, KERN_INFO
3930 "IOAPIC[%d]: Assigned apic_id %d\n", ioapic, apic_id);
3935 int __init io_apic_get_version(int ioapic)
3937 union IO_APIC_reg_01 reg_01;
3938 unsigned long flags;
3940 spin_lock_irqsave(&ioapic_lock, flags);
3941 reg_01.raw = io_apic_read(ioapic, 1);
3942 spin_unlock_irqrestore(&ioapic_lock, flags);
3944 return reg_01.bits.version;
3948 int io_apic_set_pci_routing (int ioapic, int pin, int irq, int triggering, int polarity)
3950 struct irq_desc *desc;
3951 struct irq_cfg *cfg;
3952 int cpu = boot_cpu_id;
3954 if (!IO_APIC_IRQ(irq)) {
3955 apic_printk(APIC_QUIET,KERN_ERR "IOAPIC[%d]: Invalid reference to IRQ 0\n",
3960 desc = irq_to_desc_alloc_cpu(irq, cpu);
3962 printk(KERN_INFO "can not get irq_desc %d\n", irq);
3967 * IRQs < 16 are already in the irq_2_pin[] map
3969 if (irq >= NR_IRQS_LEGACY) {
3970 cfg = desc->chip_data;
3971 add_pin_to_irq_cpu(cfg, cpu, ioapic, pin);
3974 setup_IO_APIC_irq(ioapic, pin, irq, desc, triggering, polarity);
3980 int acpi_get_override_irq(int bus_irq, int *trigger, int *polarity)
3984 if (skip_ioapic_setup)
3987 for (i = 0; i < mp_irq_entries; i++)
3988 if (mp_irqs[i].mp_irqtype == mp_INT &&
3989 mp_irqs[i].mp_srcbusirq == bus_irq)
3991 if (i >= mp_irq_entries)
3994 *trigger = irq_trigger(i);
3995 *polarity = irq_polarity(i);
3999 #endif /* CONFIG_ACPI */
4002 * This function currently is only a helper for the i386 smp boot process where
4003 * we need to reprogram the ioredtbls to cater for the cpus which have come online
4004 * so mask in all cases should simply be TARGET_CPUS
4007 void __init setup_ioapic_dest(void)
4009 int pin, ioapic, irq, irq_entry;
4010 struct irq_desc *desc;
4011 struct irq_cfg *cfg;
4012 const struct cpumask *mask;
4014 if (skip_ioapic_setup == 1)
4017 for (ioapic = 0; ioapic < nr_ioapics; ioapic++) {
4018 for (pin = 0; pin < nr_ioapic_registers[ioapic]; pin++) {
4019 irq_entry = find_irq_entry(ioapic, pin, mp_INT);
4020 if (irq_entry == -1)
4022 irq = pin_2_irq(irq_entry, ioapic, pin);
4024 /* setup_IO_APIC_irqs could fail to get vector for some device
4025 * when you have too many devices, because at that time only boot
4028 desc = irq_to_desc(irq);
4029 cfg = desc->chip_data;
4031 setup_IO_APIC_irq(ioapic, pin, irq, desc,
4032 irq_trigger(irq_entry),
4033 irq_polarity(irq_entry));
4039 * Honour affinities which have been set in early boot
4042 (IRQ_NO_BALANCING | IRQ_AFFINITY_SET))
4043 mask = &desc->affinity;
4047 #ifdef CONFIG_INTR_REMAP
4048 if (intr_remapping_enabled)
4049 set_ir_ioapic_affinity_irq_desc(desc, mask);
4052 set_ioapic_affinity_irq_desc(desc, mask);
4059 #define IOAPIC_RESOURCE_NAME_SIZE 11
4061 static struct resource *ioapic_resources;
4063 static struct resource * __init ioapic_setup_resources(void)
4066 struct resource *res;
4070 if (nr_ioapics <= 0)
4073 n = IOAPIC_RESOURCE_NAME_SIZE + sizeof(struct resource);
4076 mem = alloc_bootmem(n);
4080 mem += sizeof(struct resource) * nr_ioapics;
4082 for (i = 0; i < nr_ioapics; i++) {
4084 res[i].flags = IORESOURCE_MEM | IORESOURCE_BUSY;
4085 sprintf(mem, "IOAPIC %u", i);
4086 mem += IOAPIC_RESOURCE_NAME_SIZE;
4090 ioapic_resources = res;
4095 void __init ioapic_init_mappings(void)
4097 unsigned long ioapic_phys, idx = FIX_IO_APIC_BASE_0;
4098 struct resource *ioapic_res;
4101 ioapic_res = ioapic_setup_resources();
4102 for (i = 0; i < nr_ioapics; i++) {
4103 if (smp_found_config) {
4104 ioapic_phys = mp_ioapics[i].mp_apicaddr;
4105 #ifdef CONFIG_X86_32
4108 "WARNING: bogus zero IO-APIC "
4109 "address found in MPTABLE, "
4110 "disabling IO/APIC support!\n");
4111 smp_found_config = 0;
4112 skip_ioapic_setup = 1;
4113 goto fake_ioapic_page;
4117 #ifdef CONFIG_X86_32
4120 ioapic_phys = (unsigned long)
4121 alloc_bootmem_pages(PAGE_SIZE);
4122 ioapic_phys = __pa(ioapic_phys);
4124 set_fixmap_nocache(idx, ioapic_phys);
4125 apic_printk(APIC_VERBOSE,
4126 "mapped IOAPIC to %08lx (%08lx)\n",
4127 __fix_to_virt(idx), ioapic_phys);
4130 if (ioapic_res != NULL) {
4131 ioapic_res->start = ioapic_phys;
4132 ioapic_res->end = ioapic_phys + (4 * 1024) - 1;
4138 static int __init ioapic_insert_resources(void)
4141 struct resource *r = ioapic_resources;
4145 "IO APIC resources could be not be allocated.\n");
4149 for (i = 0; i < nr_ioapics; i++) {
4150 insert_resource(&iomem_resource, r);
4157 /* Insert the IO APIC resources after PCI initialization has occured to handle
4158 * IO APICS that are mapped in on a BAR in PCI space. */
4159 late_initcall(ioapic_insert_resources);