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() */
39 #include <linux/slab.h>
41 #include <acpi/acpi_bus.h>
43 #include <linux/bootmem.h>
44 #include <linux/dmar.h>
45 #include <linux/hpet.h>
52 #include <asm/proto.h>
55 #include <asm/timer.h>
56 #include <asm/i8259.h>
58 #include <asm/msidef.h>
59 #include <asm/hypertransport.h>
60 #include <asm/setup.h>
61 #include <asm/irq_remapping.h>
63 #include <asm/hw_irq.h>
67 #define __apicdebuginit(type) static type __init
68 #define for_each_irq_pin(entry, head) \
69 for (entry = head; entry; entry = entry->next)
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_RAW_SPINLOCK(ioapic_lock);
78 static DEFINE_RAW_SPINLOCK(vector_lock);
81 * # of IRQ routing registers
83 int nr_ioapic_registers[MAX_IO_APICS];
85 /* I/O APIC entries */
86 struct mpc_ioapic mp_ioapics[MAX_IO_APICS];
89 /* IO APIC gsi routing info */
90 struct mp_ioapic_gsi mp_gsi_routing[MAX_IO_APICS];
92 /* The one past the highest gsi number used */
95 /* MP IRQ source entries */
96 struct mpc_intsrc mp_irqs[MAX_IRQ_SOURCES];
98 /* # of MP IRQ source entries */
102 static int nr_irqs_gsi = NR_IRQS_LEGACY;
104 #if defined (CONFIG_MCA) || defined (CONFIG_EISA)
105 int mp_bus_id_to_type[MAX_MP_BUSSES];
108 DECLARE_BITMAP(mp_bus_not_pci, MAX_MP_BUSSES);
110 int skip_ioapic_setup;
112 void arch_disable_smp_support(void)
116 noioapicreroute = -1;
118 skip_ioapic_setup = 1;
121 static int __init parse_noapic(char *str)
123 /* disable IO-APIC */
124 arch_disable_smp_support();
127 early_param("noapic", parse_noapic);
129 struct irq_pin_list {
131 struct irq_pin_list *next;
134 static struct irq_pin_list *get_one_free_irq_2_pin(int node)
136 struct irq_pin_list *pin;
138 pin = kzalloc_node(sizeof(*pin), GFP_ATOMIC, node);
143 /* irq_cfg is indexed by the sum of all RTEs in all I/O APICs. */
144 #ifdef CONFIG_SPARSE_IRQ
145 static struct irq_cfg irq_cfgx[NR_IRQS_LEGACY];
147 static struct irq_cfg irq_cfgx[NR_IRQS];
150 int __init arch_early_irq_init(void)
153 struct irq_desc *desc;
158 if (!legacy_pic->nr_legacy_irqs) {
164 count = ARRAY_SIZE(irq_cfgx);
165 node = cpu_to_node(0);
167 for (i = 0; i < count; i++) {
168 desc = irq_to_desc(i);
169 desc->chip_data = &cfg[i];
170 zalloc_cpumask_var_node(&cfg[i].domain, GFP_NOWAIT, node);
171 zalloc_cpumask_var_node(&cfg[i].old_domain, GFP_NOWAIT, node);
173 * For legacy IRQ's, start with assigning irq0 to irq15 to
174 * IRQ0_VECTOR to IRQ15_VECTOR on cpu 0.
176 if (i < legacy_pic->nr_legacy_irqs) {
177 cfg[i].vector = IRQ0_VECTOR + i;
178 cpumask_set_cpu(0, cfg[i].domain);
185 #ifdef CONFIG_SPARSE_IRQ
186 struct irq_cfg *irq_cfg(unsigned int irq)
188 struct irq_cfg *cfg = NULL;
189 struct irq_desc *desc;
191 desc = irq_to_desc(irq);
193 cfg = desc->chip_data;
198 static struct irq_cfg *get_one_free_irq_cfg(int node)
202 cfg = kzalloc_node(sizeof(*cfg), GFP_ATOMIC, node);
204 if (!zalloc_cpumask_var_node(&cfg->domain, GFP_ATOMIC, node)) {
207 } else if (!zalloc_cpumask_var_node(&cfg->old_domain,
209 free_cpumask_var(cfg->domain);
218 int arch_init_chip_data(struct irq_desc *desc, int node)
222 cfg = desc->chip_data;
224 desc->chip_data = get_one_free_irq_cfg(node);
225 if (!desc->chip_data) {
226 printk(KERN_ERR "can not alloc irq_cfg\n");
234 /* for move_irq_desc */
236 init_copy_irq_2_pin(struct irq_cfg *old_cfg, struct irq_cfg *cfg, int node)
238 struct irq_pin_list *old_entry, *head, *tail, *entry;
240 cfg->irq_2_pin = NULL;
241 old_entry = old_cfg->irq_2_pin;
245 entry = get_one_free_irq_2_pin(node);
249 entry->apic = old_entry->apic;
250 entry->pin = old_entry->pin;
253 old_entry = old_entry->next;
255 entry = get_one_free_irq_2_pin(node);
263 /* still use the old one */
266 entry->apic = old_entry->apic;
267 entry->pin = old_entry->pin;
270 old_entry = old_entry->next;
274 cfg->irq_2_pin = head;
277 static void free_irq_2_pin(struct irq_cfg *old_cfg, struct irq_cfg *cfg)
279 struct irq_pin_list *entry, *next;
281 if (old_cfg->irq_2_pin == cfg->irq_2_pin)
284 entry = old_cfg->irq_2_pin;
291 old_cfg->irq_2_pin = NULL;
294 void arch_init_copy_chip_data(struct irq_desc *old_desc,
295 struct irq_desc *desc, int node)
298 struct irq_cfg *old_cfg;
300 cfg = get_one_free_irq_cfg(node);
305 desc->chip_data = cfg;
307 old_cfg = old_desc->chip_data;
309 cfg->vector = old_cfg->vector;
310 cfg->move_in_progress = old_cfg->move_in_progress;
311 cpumask_copy(cfg->domain, old_cfg->domain);
312 cpumask_copy(cfg->old_domain, old_cfg->old_domain);
314 init_copy_irq_2_pin(old_cfg, cfg, node);
317 static void free_irq_cfg(struct irq_cfg *cfg)
319 free_cpumask_var(cfg->domain);
320 free_cpumask_var(cfg->old_domain);
324 void arch_free_chip_data(struct irq_desc *old_desc, struct irq_desc *desc)
326 struct irq_cfg *old_cfg, *cfg;
328 old_cfg = old_desc->chip_data;
329 cfg = desc->chip_data;
335 free_irq_2_pin(old_cfg, cfg);
336 free_irq_cfg(old_cfg);
337 old_desc->chip_data = NULL;
340 /* end for move_irq_desc */
343 struct irq_cfg *irq_cfg(unsigned int irq)
345 return irq < nr_irqs ? irq_cfgx + irq : NULL;
352 unsigned int unused[3];
354 unsigned int unused2[11];
358 static __attribute_const__ struct io_apic __iomem *io_apic_base(int idx)
360 return (void __iomem *) __fix_to_virt(FIX_IO_APIC_BASE_0 + idx)
361 + (mp_ioapics[idx].apicaddr & ~PAGE_MASK);
364 static inline void io_apic_eoi(unsigned int apic, unsigned int vector)
366 struct io_apic __iomem *io_apic = io_apic_base(apic);
367 writel(vector, &io_apic->eoi);
370 static inline unsigned int io_apic_read(unsigned int apic, unsigned int reg)
372 struct io_apic __iomem *io_apic = io_apic_base(apic);
373 writel(reg, &io_apic->index);
374 return readl(&io_apic->data);
377 static inline void io_apic_write(unsigned int apic, unsigned int reg, unsigned int value)
379 struct io_apic __iomem *io_apic = io_apic_base(apic);
380 writel(reg, &io_apic->index);
381 writel(value, &io_apic->data);
385 * Re-write a value: to be used for read-modify-write
386 * cycles where the read already set up the index register.
388 * Older SiS APIC requires we rewrite the index register
390 static inline void io_apic_modify(unsigned int apic, unsigned int reg, unsigned int value)
392 struct io_apic __iomem *io_apic = io_apic_base(apic);
395 writel(reg, &io_apic->index);
396 writel(value, &io_apic->data);
399 static bool io_apic_level_ack_pending(struct irq_cfg *cfg)
401 struct irq_pin_list *entry;
404 raw_spin_lock_irqsave(&ioapic_lock, flags);
405 for_each_irq_pin(entry, cfg->irq_2_pin) {
410 reg = io_apic_read(entry->apic, 0x10 + pin*2);
411 /* Is the remote IRR bit set? */
412 if (reg & IO_APIC_REDIR_REMOTE_IRR) {
413 raw_spin_unlock_irqrestore(&ioapic_lock, flags);
417 raw_spin_unlock_irqrestore(&ioapic_lock, flags);
423 struct { u32 w1, w2; };
424 struct IO_APIC_route_entry entry;
427 static struct IO_APIC_route_entry ioapic_read_entry(int apic, int pin)
429 union entry_union eu;
431 raw_spin_lock_irqsave(&ioapic_lock, flags);
432 eu.w1 = io_apic_read(apic, 0x10 + 2 * pin);
433 eu.w2 = io_apic_read(apic, 0x11 + 2 * pin);
434 raw_spin_unlock_irqrestore(&ioapic_lock, flags);
439 * When we write a new IO APIC routing entry, we need to write the high
440 * word first! If the mask bit in the low word is clear, we will enable
441 * the interrupt, and we need to make sure the entry is fully populated
442 * before that happens.
445 __ioapic_write_entry(int apic, int pin, struct IO_APIC_route_entry e)
447 union entry_union eu = {{0, 0}};
450 io_apic_write(apic, 0x11 + 2*pin, eu.w2);
451 io_apic_write(apic, 0x10 + 2*pin, eu.w1);
454 void ioapic_write_entry(int apic, int pin, struct IO_APIC_route_entry e)
457 raw_spin_lock_irqsave(&ioapic_lock, flags);
458 __ioapic_write_entry(apic, pin, e);
459 raw_spin_unlock_irqrestore(&ioapic_lock, flags);
463 * When we mask an IO APIC routing entry, we need to write the low
464 * word first, in order to set the mask bit before we change the
467 static void ioapic_mask_entry(int apic, int pin)
470 union entry_union eu = { .entry.mask = 1 };
472 raw_spin_lock_irqsave(&ioapic_lock, flags);
473 io_apic_write(apic, 0x10 + 2*pin, eu.w1);
474 io_apic_write(apic, 0x11 + 2*pin, eu.w2);
475 raw_spin_unlock_irqrestore(&ioapic_lock, flags);
479 * The common case is 1:1 IRQ<->pin mappings. Sometimes there are
480 * shared ISA-space IRQs, so we have to support them. We are super
481 * fast in the common case, and fast for shared ISA-space IRQs.
484 add_pin_to_irq_node_nopanic(struct irq_cfg *cfg, int node, int apic, int pin)
486 struct irq_pin_list **last, *entry;
488 /* don't allow duplicates */
489 last = &cfg->irq_2_pin;
490 for_each_irq_pin(entry, cfg->irq_2_pin) {
491 if (entry->apic == apic && entry->pin == pin)
496 entry = get_one_free_irq_2_pin(node);
498 printk(KERN_ERR "can not alloc irq_pin_list (%d,%d,%d)\n",
509 static void add_pin_to_irq_node(struct irq_cfg *cfg, int node, int apic, int pin)
511 if (add_pin_to_irq_node_nopanic(cfg, node, apic, pin))
512 panic("IO-APIC: failed to add irq-pin. Can not proceed\n");
516 * Reroute an IRQ to a different pin.
518 static void __init replace_pin_at_irq_node(struct irq_cfg *cfg, int node,
519 int oldapic, int oldpin,
520 int newapic, int newpin)
522 struct irq_pin_list *entry;
524 for_each_irq_pin(entry, cfg->irq_2_pin) {
525 if (entry->apic == oldapic && entry->pin == oldpin) {
526 entry->apic = newapic;
528 /* every one is different, right? */
533 /* old apic/pin didn't exist, so just add new ones */
534 add_pin_to_irq_node(cfg, node, newapic, newpin);
537 static void __io_apic_modify_irq(struct irq_pin_list *entry,
538 int mask_and, int mask_or,
539 void (*final)(struct irq_pin_list *entry))
541 unsigned int reg, pin;
544 reg = io_apic_read(entry->apic, 0x10 + pin * 2);
547 io_apic_modify(entry->apic, 0x10 + pin * 2, reg);
552 static void io_apic_modify_irq(struct irq_cfg *cfg,
553 int mask_and, int mask_or,
554 void (*final)(struct irq_pin_list *entry))
556 struct irq_pin_list *entry;
558 for_each_irq_pin(entry, cfg->irq_2_pin)
559 __io_apic_modify_irq(entry, mask_and, mask_or, final);
562 static void __mask_and_edge_IO_APIC_irq(struct irq_pin_list *entry)
564 __io_apic_modify_irq(entry, ~IO_APIC_REDIR_LEVEL_TRIGGER,
565 IO_APIC_REDIR_MASKED, NULL);
568 static void __unmask_and_level_IO_APIC_irq(struct irq_pin_list *entry)
570 __io_apic_modify_irq(entry, ~IO_APIC_REDIR_MASKED,
571 IO_APIC_REDIR_LEVEL_TRIGGER, NULL);
574 static void __unmask_IO_APIC_irq(struct irq_cfg *cfg)
576 io_apic_modify_irq(cfg, ~IO_APIC_REDIR_MASKED, 0, NULL);
579 static void io_apic_sync(struct irq_pin_list *entry)
582 * Synchronize the IO-APIC and the CPU by doing
583 * a dummy read from the IO-APIC
585 struct io_apic __iomem *io_apic;
586 io_apic = io_apic_base(entry->apic);
587 readl(&io_apic->data);
590 static void __mask_IO_APIC_irq(struct irq_cfg *cfg)
592 io_apic_modify_irq(cfg, ~0, IO_APIC_REDIR_MASKED, &io_apic_sync);
595 static void mask_IO_APIC_irq_desc(struct irq_desc *desc)
597 struct irq_cfg *cfg = desc->chip_data;
602 raw_spin_lock_irqsave(&ioapic_lock, flags);
603 __mask_IO_APIC_irq(cfg);
604 raw_spin_unlock_irqrestore(&ioapic_lock, flags);
607 static void unmask_IO_APIC_irq_desc(struct irq_desc *desc)
609 struct irq_cfg *cfg = desc->chip_data;
612 raw_spin_lock_irqsave(&ioapic_lock, flags);
613 __unmask_IO_APIC_irq(cfg);
614 raw_spin_unlock_irqrestore(&ioapic_lock, flags);
617 static void mask_IO_APIC_irq(unsigned int irq)
619 struct irq_desc *desc = irq_to_desc(irq);
621 mask_IO_APIC_irq_desc(desc);
623 static void unmask_IO_APIC_irq(unsigned int irq)
625 struct irq_desc *desc = irq_to_desc(irq);
627 unmask_IO_APIC_irq_desc(desc);
630 static void clear_IO_APIC_pin(unsigned int apic, unsigned int pin)
632 struct IO_APIC_route_entry entry;
634 /* Check delivery_mode to be sure we're not clearing an SMI pin */
635 entry = ioapic_read_entry(apic, pin);
636 if (entry.delivery_mode == dest_SMI)
639 * Disable it in the IO-APIC irq-routing table:
641 ioapic_mask_entry(apic, pin);
644 static void clear_IO_APIC (void)
648 for (apic = 0; apic < nr_ioapics; apic++)
649 for (pin = 0; pin < nr_ioapic_registers[apic]; pin++)
650 clear_IO_APIC_pin(apic, pin);
655 * support for broken MP BIOSs, enables hand-redirection of PIRQ0-7 to
656 * specific CPU-side IRQs.
660 static int pirq_entries[MAX_PIRQS] = {
661 [0 ... MAX_PIRQS - 1] = -1
664 static int __init ioapic_pirq_setup(char *str)
667 int ints[MAX_PIRQS+1];
669 get_options(str, ARRAY_SIZE(ints), ints);
671 apic_printk(APIC_VERBOSE, KERN_INFO
672 "PIRQ redirection, working around broken MP-BIOS.\n");
674 if (ints[0] < MAX_PIRQS)
677 for (i = 0; i < max; i++) {
678 apic_printk(APIC_VERBOSE, KERN_DEBUG
679 "... PIRQ%d -> IRQ %d\n", i, ints[i+1]);
681 * PIRQs are mapped upside down, usually.
683 pirq_entries[MAX_PIRQS-i-1] = ints[i+1];
688 __setup("pirq=", ioapic_pirq_setup);
689 #endif /* CONFIG_X86_32 */
691 struct IO_APIC_route_entry **alloc_ioapic_entries(void)
694 struct IO_APIC_route_entry **ioapic_entries;
696 ioapic_entries = kzalloc(sizeof(*ioapic_entries) * nr_ioapics,
701 for (apic = 0; apic < nr_ioapics; apic++) {
702 ioapic_entries[apic] =
703 kzalloc(sizeof(struct IO_APIC_route_entry) *
704 nr_ioapic_registers[apic], GFP_ATOMIC);
705 if (!ioapic_entries[apic])
709 return ioapic_entries;
713 kfree(ioapic_entries[apic]);
714 kfree(ioapic_entries);
720 * Saves all the IO-APIC RTE's
722 int save_IO_APIC_setup(struct IO_APIC_route_entry **ioapic_entries)
729 for (apic = 0; apic < nr_ioapics; apic++) {
730 if (!ioapic_entries[apic])
733 for (pin = 0; pin < nr_ioapic_registers[apic]; pin++)
734 ioapic_entries[apic][pin] =
735 ioapic_read_entry(apic, pin);
742 * Mask all IO APIC entries.
744 void mask_IO_APIC_setup(struct IO_APIC_route_entry **ioapic_entries)
751 for (apic = 0; apic < nr_ioapics; apic++) {
752 if (!ioapic_entries[apic])
755 for (pin = 0; pin < nr_ioapic_registers[apic]; pin++) {
756 struct IO_APIC_route_entry entry;
758 entry = ioapic_entries[apic][pin];
761 ioapic_write_entry(apic, pin, entry);
768 * Restore IO APIC entries which was saved in ioapic_entries.
770 int restore_IO_APIC_setup(struct IO_APIC_route_entry **ioapic_entries)
777 for (apic = 0; apic < nr_ioapics; apic++) {
778 if (!ioapic_entries[apic])
781 for (pin = 0; pin < nr_ioapic_registers[apic]; pin++)
782 ioapic_write_entry(apic, pin,
783 ioapic_entries[apic][pin]);
788 void free_ioapic_entries(struct IO_APIC_route_entry **ioapic_entries)
792 for (apic = 0; apic < nr_ioapics; apic++)
793 kfree(ioapic_entries[apic]);
795 kfree(ioapic_entries);
799 * Find the IRQ entry number of a certain pin.
801 static int find_irq_entry(int apic, int pin, int type)
805 for (i = 0; i < mp_irq_entries; i++)
806 if (mp_irqs[i].irqtype == type &&
807 (mp_irqs[i].dstapic == mp_ioapics[apic].apicid ||
808 mp_irqs[i].dstapic == MP_APIC_ALL) &&
809 mp_irqs[i].dstirq == pin)
816 * Find the pin to which IRQ[irq] (ISA) is connected
818 static int __init find_isa_irq_pin(int irq, int type)
822 for (i = 0; i < mp_irq_entries; i++) {
823 int lbus = mp_irqs[i].srcbus;
825 if (test_bit(lbus, mp_bus_not_pci) &&
826 (mp_irqs[i].irqtype == type) &&
827 (mp_irqs[i].srcbusirq == irq))
829 return mp_irqs[i].dstirq;
834 static int __init find_isa_irq_apic(int irq, int type)
838 for (i = 0; i < mp_irq_entries; i++) {
839 int lbus = mp_irqs[i].srcbus;
841 if (test_bit(lbus, mp_bus_not_pci) &&
842 (mp_irqs[i].irqtype == type) &&
843 (mp_irqs[i].srcbusirq == irq))
846 if (i < mp_irq_entries) {
848 for(apic = 0; apic < nr_ioapics; apic++) {
849 if (mp_ioapics[apic].apicid == mp_irqs[i].dstapic)
857 #if defined(CONFIG_EISA) || defined(CONFIG_MCA)
859 * EISA Edge/Level control register, ELCR
861 static int EISA_ELCR(unsigned int irq)
863 if (irq < legacy_pic->nr_legacy_irqs) {
864 unsigned int port = 0x4d0 + (irq >> 3);
865 return (inb(port) >> (irq & 7)) & 1;
867 apic_printk(APIC_VERBOSE, KERN_INFO
868 "Broken MPtable reports ISA irq %d\n", irq);
874 /* ISA interrupts are always polarity zero edge triggered,
875 * when listed as conforming in the MP table. */
877 #define default_ISA_trigger(idx) (0)
878 #define default_ISA_polarity(idx) (0)
880 /* EISA interrupts are always polarity zero and can be edge or level
881 * trigger depending on the ELCR value. If an interrupt is listed as
882 * EISA conforming in the MP table, that means its trigger type must
883 * be read in from the ELCR */
885 #define default_EISA_trigger(idx) (EISA_ELCR(mp_irqs[idx].srcbusirq))
886 #define default_EISA_polarity(idx) default_ISA_polarity(idx)
888 /* PCI interrupts are always polarity one level triggered,
889 * when listed as conforming in the MP table. */
891 #define default_PCI_trigger(idx) (1)
892 #define default_PCI_polarity(idx) (1)
894 /* MCA interrupts are always polarity zero level triggered,
895 * when listed as conforming in the MP table. */
897 #define default_MCA_trigger(idx) (1)
898 #define default_MCA_polarity(idx) default_ISA_polarity(idx)
900 static int MPBIOS_polarity(int idx)
902 int bus = mp_irqs[idx].srcbus;
906 * Determine IRQ line polarity (high active or low active):
908 switch (mp_irqs[idx].irqflag & 3)
910 case 0: /* conforms, ie. bus-type dependent polarity */
911 if (test_bit(bus, mp_bus_not_pci))
912 polarity = default_ISA_polarity(idx);
914 polarity = default_PCI_polarity(idx);
916 case 1: /* high active */
921 case 2: /* reserved */
923 printk(KERN_WARNING "broken BIOS!!\n");
927 case 3: /* low active */
932 default: /* invalid */
934 printk(KERN_WARNING "broken BIOS!!\n");
942 static int MPBIOS_trigger(int idx)
944 int bus = mp_irqs[idx].srcbus;
948 * Determine IRQ trigger mode (edge or level sensitive):
950 switch ((mp_irqs[idx].irqflag>>2) & 3)
952 case 0: /* conforms, ie. bus-type dependent */
953 if (test_bit(bus, mp_bus_not_pci))
954 trigger = default_ISA_trigger(idx);
956 trigger = default_PCI_trigger(idx);
957 #if defined(CONFIG_EISA) || defined(CONFIG_MCA)
958 switch (mp_bus_id_to_type[bus]) {
959 case MP_BUS_ISA: /* ISA pin */
961 /* set before the switch */
964 case MP_BUS_EISA: /* EISA pin */
966 trigger = default_EISA_trigger(idx);
969 case MP_BUS_PCI: /* PCI pin */
971 /* set before the switch */
974 case MP_BUS_MCA: /* MCA pin */
976 trigger = default_MCA_trigger(idx);
981 printk(KERN_WARNING "broken BIOS!!\n");
993 case 2: /* reserved */
995 printk(KERN_WARNING "broken BIOS!!\n");
1004 default: /* invalid */
1006 printk(KERN_WARNING "broken BIOS!!\n");
1014 static inline int irq_polarity(int idx)
1016 return MPBIOS_polarity(idx);
1019 static inline int irq_trigger(int idx)
1021 return MPBIOS_trigger(idx);
1024 static int pin_2_irq(int idx, int apic, int pin)
1027 int bus = mp_irqs[idx].srcbus;
1030 * Debugging check, we are in big trouble if this message pops up!
1032 if (mp_irqs[idx].dstirq != pin)
1033 printk(KERN_ERR "broken BIOS or MPTABLE parser, ayiee!!\n");
1035 if (test_bit(bus, mp_bus_not_pci)) {
1036 irq = mp_irqs[idx].srcbusirq;
1038 u32 gsi = mp_gsi_routing[apic].gsi_base + pin;
1040 if (gsi >= NR_IRQS_LEGACY)
1043 irq = gsi_top + gsi;
1046 #ifdef CONFIG_X86_32
1048 * PCI IRQ command line redirection. Yes, limits are hardcoded.
1050 if ((pin >= 16) && (pin <= 23)) {
1051 if (pirq_entries[pin-16] != -1) {
1052 if (!pirq_entries[pin-16]) {
1053 apic_printk(APIC_VERBOSE, KERN_DEBUG
1054 "disabling PIRQ%d\n", pin-16);
1056 irq = pirq_entries[pin-16];
1057 apic_printk(APIC_VERBOSE, KERN_DEBUG
1058 "using PIRQ%d -> IRQ %d\n",
1069 * Find a specific PCI IRQ entry.
1070 * Not an __init, possibly needed by modules
1072 int IO_APIC_get_PCI_irq_vector(int bus, int slot, int pin,
1073 struct io_apic_irq_attr *irq_attr)
1075 int apic, i, best_guess = -1;
1077 apic_printk(APIC_DEBUG,
1078 "querying PCI -> IRQ mapping bus:%d, slot:%d, pin:%d.\n",
1080 if (test_bit(bus, mp_bus_not_pci)) {
1081 apic_printk(APIC_VERBOSE,
1082 "PCI BIOS passed nonexistent PCI bus %d!\n", bus);
1085 for (i = 0; i < mp_irq_entries; i++) {
1086 int lbus = mp_irqs[i].srcbus;
1088 for (apic = 0; apic < nr_ioapics; apic++)
1089 if (mp_ioapics[apic].apicid == mp_irqs[i].dstapic ||
1090 mp_irqs[i].dstapic == MP_APIC_ALL)
1093 if (!test_bit(lbus, mp_bus_not_pci) &&
1094 !mp_irqs[i].irqtype &&
1096 (slot == ((mp_irqs[i].srcbusirq >> 2) & 0x1f))) {
1097 int irq = pin_2_irq(i, apic, mp_irqs[i].dstirq);
1099 if (!(apic || IO_APIC_IRQ(irq)))
1102 if (pin == (mp_irqs[i].srcbusirq & 3)) {
1103 set_io_apic_irq_attr(irq_attr, apic,
1110 * Use the first all-but-pin matching entry as a
1111 * best-guess fuzzy result for broken mptables.
1113 if (best_guess < 0) {
1114 set_io_apic_irq_attr(irq_attr, apic,
1124 EXPORT_SYMBOL(IO_APIC_get_PCI_irq_vector);
1126 void lock_vector_lock(void)
1128 /* Used to the online set of cpus does not change
1129 * during assign_irq_vector.
1131 raw_spin_lock(&vector_lock);
1134 void unlock_vector_lock(void)
1136 raw_spin_unlock(&vector_lock);
1140 __assign_irq_vector(int irq, struct irq_cfg *cfg, const struct cpumask *mask)
1143 * NOTE! The local APIC isn't very good at handling
1144 * multiple interrupts at the same interrupt level.
1145 * As the interrupt level is determined by taking the
1146 * vector number and shifting that right by 4, we
1147 * want to spread these out a bit so that they don't
1148 * all fall in the same interrupt level.
1150 * Also, we've got to be careful not to trash gate
1151 * 0x80, because int 0x80 is hm, kind of importantish. ;)
1153 static int current_vector = FIRST_EXTERNAL_VECTOR + VECTOR_OFFSET_START;
1154 static int current_offset = VECTOR_OFFSET_START % 8;
1155 unsigned int old_vector;
1157 cpumask_var_t tmp_mask;
1159 if (cfg->move_in_progress)
1162 if (!alloc_cpumask_var(&tmp_mask, GFP_ATOMIC))
1165 old_vector = cfg->vector;
1167 cpumask_and(tmp_mask, mask, cpu_online_mask);
1168 cpumask_and(tmp_mask, cfg->domain, tmp_mask);
1169 if (!cpumask_empty(tmp_mask)) {
1170 free_cpumask_var(tmp_mask);
1175 /* Only try and allocate irqs on cpus that are present */
1177 for_each_cpu_and(cpu, mask, cpu_online_mask) {
1181 apic->vector_allocation_domain(cpu, tmp_mask);
1183 vector = current_vector;
1184 offset = current_offset;
1187 if (vector >= first_system_vector) {
1188 /* If out of vectors on large boxen, must share them. */
1189 offset = (offset + 1) % 8;
1190 vector = FIRST_EXTERNAL_VECTOR + offset;
1192 if (unlikely(current_vector == vector))
1195 if (test_bit(vector, used_vectors))
1198 for_each_cpu_and(new_cpu, tmp_mask, cpu_online_mask)
1199 if (per_cpu(vector_irq, new_cpu)[vector] != -1)
1202 current_vector = vector;
1203 current_offset = offset;
1205 cfg->move_in_progress = 1;
1206 cpumask_copy(cfg->old_domain, cfg->domain);
1208 for_each_cpu_and(new_cpu, tmp_mask, cpu_online_mask)
1209 per_cpu(vector_irq, new_cpu)[vector] = irq;
1210 cfg->vector = vector;
1211 cpumask_copy(cfg->domain, tmp_mask);
1215 free_cpumask_var(tmp_mask);
1219 int assign_irq_vector(int irq, struct irq_cfg *cfg, const struct cpumask *mask)
1222 unsigned long flags;
1224 raw_spin_lock_irqsave(&vector_lock, flags);
1225 err = __assign_irq_vector(irq, cfg, mask);
1226 raw_spin_unlock_irqrestore(&vector_lock, flags);
1230 static void __clear_irq_vector(int irq, struct irq_cfg *cfg)
1234 BUG_ON(!cfg->vector);
1236 vector = cfg->vector;
1237 for_each_cpu_and(cpu, cfg->domain, cpu_online_mask)
1238 per_cpu(vector_irq, cpu)[vector] = -1;
1241 cpumask_clear(cfg->domain);
1243 if (likely(!cfg->move_in_progress))
1245 for_each_cpu_and(cpu, cfg->old_domain, cpu_online_mask) {
1246 for (vector = FIRST_EXTERNAL_VECTOR; vector < NR_VECTORS;
1248 if (per_cpu(vector_irq, cpu)[vector] != irq)
1250 per_cpu(vector_irq, cpu)[vector] = -1;
1254 cfg->move_in_progress = 0;
1257 void __setup_vector_irq(int cpu)
1259 /* Initialize vector_irq on a new cpu */
1261 struct irq_cfg *cfg;
1262 struct irq_desc *desc;
1265 * vector_lock will make sure that we don't run into irq vector
1266 * assignments that might be happening on another cpu in parallel,
1267 * while we setup our initial vector to irq mappings.
1269 raw_spin_lock(&vector_lock);
1270 /* Mark the inuse vectors */
1271 for_each_irq_desc(irq, desc) {
1272 cfg = desc->chip_data;
1275 * If it is a legacy IRQ handled by the legacy PIC, this cpu
1276 * will be part of the irq_cfg's domain.
1278 if (irq < legacy_pic->nr_legacy_irqs && !IO_APIC_IRQ(irq))
1279 cpumask_set_cpu(cpu, cfg->domain);
1281 if (!cpumask_test_cpu(cpu, cfg->domain))
1283 vector = cfg->vector;
1284 per_cpu(vector_irq, cpu)[vector] = irq;
1286 /* Mark the free vectors */
1287 for (vector = 0; vector < NR_VECTORS; ++vector) {
1288 irq = per_cpu(vector_irq, cpu)[vector];
1293 if (!cpumask_test_cpu(cpu, cfg->domain))
1294 per_cpu(vector_irq, cpu)[vector] = -1;
1296 raw_spin_unlock(&vector_lock);
1299 static struct irq_chip ioapic_chip;
1300 static struct irq_chip ir_ioapic_chip;
1302 #define IOAPIC_AUTO -1
1303 #define IOAPIC_EDGE 0
1304 #define IOAPIC_LEVEL 1
1306 #ifdef CONFIG_X86_32
1307 static inline int IO_APIC_irq_trigger(int irq)
1311 for (apic = 0; apic < nr_ioapics; apic++) {
1312 for (pin = 0; pin < nr_ioapic_registers[apic]; pin++) {
1313 idx = find_irq_entry(apic, pin, mp_INT);
1314 if ((idx != -1) && (irq == pin_2_irq(idx, apic, pin)))
1315 return irq_trigger(idx);
1319 * nonexistent IRQs are edge default
1324 static inline int IO_APIC_irq_trigger(int irq)
1330 static void ioapic_register_intr(int irq, struct irq_desc *desc, unsigned long trigger)
1333 if ((trigger == IOAPIC_AUTO && IO_APIC_irq_trigger(irq)) ||
1334 trigger == IOAPIC_LEVEL)
1335 desc->status |= IRQ_LEVEL;
1337 desc->status &= ~IRQ_LEVEL;
1339 if (irq_remapped(irq)) {
1340 desc->status |= IRQ_MOVE_PCNTXT;
1342 set_irq_chip_and_handler_name(irq, &ir_ioapic_chip,
1346 set_irq_chip_and_handler_name(irq, &ir_ioapic_chip,
1347 handle_edge_irq, "edge");
1351 if ((trigger == IOAPIC_AUTO && IO_APIC_irq_trigger(irq)) ||
1352 trigger == IOAPIC_LEVEL)
1353 set_irq_chip_and_handler_name(irq, &ioapic_chip,
1357 set_irq_chip_and_handler_name(irq, &ioapic_chip,
1358 handle_edge_irq, "edge");
1361 int setup_ioapic_entry(int apic_id, int irq,
1362 struct IO_APIC_route_entry *entry,
1363 unsigned int destination, int trigger,
1364 int polarity, int vector, int pin)
1367 * add it to the IO-APIC irq-routing table:
1369 memset(entry,0,sizeof(*entry));
1371 if (intr_remapping_enabled) {
1372 struct intel_iommu *iommu = map_ioapic_to_ir(apic_id);
1374 struct IR_IO_APIC_route_entry *ir_entry =
1375 (struct IR_IO_APIC_route_entry *) entry;
1379 panic("No mapping iommu for ioapic %d\n", apic_id);
1381 index = alloc_irte(iommu, irq, 1);
1383 panic("Failed to allocate IRTE for ioapic %d\n", apic_id);
1385 prepare_irte(&irte, vector, destination);
1387 /* Set source-id of interrupt request */
1388 set_ioapic_sid(&irte, apic_id);
1390 modify_irte(irq, &irte);
1392 ir_entry->index2 = (index >> 15) & 0x1;
1394 ir_entry->format = 1;
1395 ir_entry->index = (index & 0x7fff);
1397 * IO-APIC RTE will be configured with virtual vector.
1398 * irq handler will do the explicit EOI to the io-apic.
1400 ir_entry->vector = pin;
1402 entry->delivery_mode = apic->irq_delivery_mode;
1403 entry->dest_mode = apic->irq_dest_mode;
1404 entry->dest = destination;
1405 entry->vector = vector;
1408 entry->mask = 0; /* enable IRQ */
1409 entry->trigger = trigger;
1410 entry->polarity = polarity;
1412 /* Mask level triggered irqs.
1413 * Use IRQ_DELAYED_DISABLE for edge triggered irqs.
1420 static void setup_IO_APIC_irq(int apic_id, int pin, unsigned int irq, struct irq_desc *desc,
1421 int trigger, int polarity)
1423 struct irq_cfg *cfg;
1424 struct IO_APIC_route_entry entry;
1427 if (!IO_APIC_IRQ(irq))
1430 cfg = desc->chip_data;
1433 * For legacy irqs, cfg->domain starts with cpu 0 for legacy
1434 * controllers like 8259. Now that IO-APIC can handle this irq, update
1437 if (irq < legacy_pic->nr_legacy_irqs && cpumask_test_cpu(0, cfg->domain))
1438 apic->vector_allocation_domain(0, cfg->domain);
1440 if (assign_irq_vector(irq, cfg, apic->target_cpus()))
1443 dest = apic->cpu_mask_to_apicid_and(cfg->domain, apic->target_cpus());
1445 apic_printk(APIC_VERBOSE,KERN_DEBUG
1446 "IOAPIC[%d]: Set routing entry (%d-%d -> 0x%x -> "
1447 "IRQ %d Mode:%i Active:%i)\n",
1448 apic_id, mp_ioapics[apic_id].apicid, pin, cfg->vector,
1449 irq, trigger, polarity);
1452 if (setup_ioapic_entry(mp_ioapics[apic_id].apicid, irq, &entry,
1453 dest, trigger, polarity, cfg->vector, pin)) {
1454 printk("Failed to setup ioapic entry for ioapic %d, pin %d\n",
1455 mp_ioapics[apic_id].apicid, pin);
1456 __clear_irq_vector(irq, cfg);
1460 ioapic_register_intr(irq, desc, trigger);
1461 if (irq < legacy_pic->nr_legacy_irqs)
1462 legacy_pic->chip->mask(irq);
1464 ioapic_write_entry(apic_id, pin, entry);
1468 DECLARE_BITMAP(pin_programmed, MP_MAX_IOAPIC_PIN + 1);
1469 } mp_ioapic_routing[MAX_IO_APICS];
1471 static void __init setup_IO_APIC_irqs(void)
1473 int apic_id, pin, idx, irq;
1475 struct irq_desc *desc;
1476 struct irq_cfg *cfg;
1477 int node = cpu_to_node(0);
1479 apic_printk(APIC_VERBOSE, KERN_DEBUG "init IO_APIC IRQs\n");
1481 for (apic_id = 0; apic_id < nr_ioapics; apic_id++)
1482 for (pin = 0; pin < nr_ioapic_registers[apic_id]; pin++) {
1483 idx = find_irq_entry(apic_id, pin, mp_INT);
1487 apic_printk(APIC_VERBOSE,
1488 KERN_DEBUG " %d-%d",
1489 mp_ioapics[apic_id].apicid, pin);
1491 apic_printk(APIC_VERBOSE, " %d-%d",
1492 mp_ioapics[apic_id].apicid, pin);
1496 apic_printk(APIC_VERBOSE,
1497 " (apicid-pin) not connected\n");
1501 irq = pin_2_irq(idx, apic_id, pin);
1503 if ((apic_id > 0) && (irq > 16))
1507 * Skip the timer IRQ if there's a quirk handler
1508 * installed and if it returns 1:
1510 if (apic->multi_timer_check &&
1511 apic->multi_timer_check(apic_id, irq))
1514 desc = irq_to_desc_alloc_node(irq, node);
1516 printk(KERN_INFO "can not get irq_desc for %d\n", irq);
1519 cfg = desc->chip_data;
1520 add_pin_to_irq_node(cfg, node, apic_id, pin);
1522 * don't mark it in pin_programmed, so later acpi could
1523 * set it correctly when irq < 16
1525 setup_IO_APIC_irq(apic_id, pin, irq, desc,
1526 irq_trigger(idx), irq_polarity(idx));
1530 apic_printk(APIC_VERBOSE,
1531 " (apicid-pin) not connected\n");
1535 * for the gsit that is not in first ioapic
1536 * but could not use acpi_register_gsi()
1537 * like some special sci in IBM x3330
1539 void setup_IO_APIC_irq_extra(u32 gsi)
1541 int apic_id = 0, pin, idx, irq;
1542 int node = cpu_to_node(0);
1543 struct irq_desc *desc;
1544 struct irq_cfg *cfg;
1547 * Convert 'gsi' to 'ioapic.pin'.
1549 apic_id = mp_find_ioapic(gsi);
1553 pin = mp_find_ioapic_pin(apic_id, gsi);
1554 idx = find_irq_entry(apic_id, pin, mp_INT);
1558 irq = pin_2_irq(idx, apic_id, pin);
1559 #ifdef CONFIG_SPARSE_IRQ
1560 desc = irq_to_desc(irq);
1564 desc = irq_to_desc_alloc_node(irq, node);
1566 printk(KERN_INFO "can not get irq_desc for %d\n", irq);
1570 cfg = desc->chip_data;
1571 add_pin_to_irq_node(cfg, node, apic_id, pin);
1573 if (test_bit(pin, mp_ioapic_routing[apic_id].pin_programmed)) {
1574 pr_debug("Pin %d-%d already programmed\n",
1575 mp_ioapics[apic_id].apicid, pin);
1578 set_bit(pin, mp_ioapic_routing[apic_id].pin_programmed);
1580 setup_IO_APIC_irq(apic_id, pin, irq, desc,
1581 irq_trigger(idx), irq_polarity(idx));
1585 * Set up the timer pin, possibly with the 8259A-master behind.
1587 static void __init setup_timer_IRQ0_pin(unsigned int apic_id, unsigned int pin,
1590 struct IO_APIC_route_entry entry;
1592 if (intr_remapping_enabled)
1595 memset(&entry, 0, sizeof(entry));
1598 * We use logical delivery to get the timer IRQ
1601 entry.dest_mode = apic->irq_dest_mode;
1602 entry.mask = 0; /* don't mask IRQ for edge */
1603 entry.dest = apic->cpu_mask_to_apicid(apic->target_cpus());
1604 entry.delivery_mode = apic->irq_delivery_mode;
1607 entry.vector = vector;
1610 * The timer IRQ doesn't have to know that behind the
1611 * scene we may have a 8259A-master in AEOI mode ...
1613 set_irq_chip_and_handler_name(0, &ioapic_chip, handle_edge_irq, "edge");
1616 * Add it to the IO-APIC irq-routing table:
1618 ioapic_write_entry(apic_id, pin, entry);
1622 __apicdebuginit(void) print_IO_APIC(void)
1625 union IO_APIC_reg_00 reg_00;
1626 union IO_APIC_reg_01 reg_01;
1627 union IO_APIC_reg_02 reg_02;
1628 union IO_APIC_reg_03 reg_03;
1629 unsigned long flags;
1630 struct irq_cfg *cfg;
1631 struct irq_desc *desc;
1634 printk(KERN_DEBUG "number of MP IRQ sources: %d.\n", mp_irq_entries);
1635 for (i = 0; i < nr_ioapics; i++)
1636 printk(KERN_DEBUG "number of IO-APIC #%d registers: %d.\n",
1637 mp_ioapics[i].apicid, nr_ioapic_registers[i]);
1640 * We are a bit conservative about what we expect. We have to
1641 * know about every hardware change ASAP.
1643 printk(KERN_INFO "testing the IO APIC.......................\n");
1645 for (apic = 0; apic < nr_ioapics; apic++) {
1647 raw_spin_lock_irqsave(&ioapic_lock, flags);
1648 reg_00.raw = io_apic_read(apic, 0);
1649 reg_01.raw = io_apic_read(apic, 1);
1650 if (reg_01.bits.version >= 0x10)
1651 reg_02.raw = io_apic_read(apic, 2);
1652 if (reg_01.bits.version >= 0x20)
1653 reg_03.raw = io_apic_read(apic, 3);
1654 raw_spin_unlock_irqrestore(&ioapic_lock, flags);
1657 printk(KERN_DEBUG "IO APIC #%d......\n", mp_ioapics[apic].apicid);
1658 printk(KERN_DEBUG ".... register #00: %08X\n", reg_00.raw);
1659 printk(KERN_DEBUG "....... : physical APIC id: %02X\n", reg_00.bits.ID);
1660 printk(KERN_DEBUG "....... : Delivery Type: %X\n", reg_00.bits.delivery_type);
1661 printk(KERN_DEBUG "....... : LTS : %X\n", reg_00.bits.LTS);
1663 printk(KERN_DEBUG ".... register #01: %08X\n", *(int *)®_01);
1664 printk(KERN_DEBUG "....... : max redirection entries: %04X\n", reg_01.bits.entries);
1666 printk(KERN_DEBUG "....... : PRQ implemented: %X\n", reg_01.bits.PRQ);
1667 printk(KERN_DEBUG "....... : IO APIC version: %04X\n", reg_01.bits.version);
1670 * Some Intel chipsets with IO APIC VERSION of 0x1? don't have reg_02,
1671 * but the value of reg_02 is read as the previous read register
1672 * value, so ignore it if reg_02 == reg_01.
1674 if (reg_01.bits.version >= 0x10 && reg_02.raw != reg_01.raw) {
1675 printk(KERN_DEBUG ".... register #02: %08X\n", reg_02.raw);
1676 printk(KERN_DEBUG "....... : arbitration: %02X\n", reg_02.bits.arbitration);
1680 * Some Intel chipsets with IO APIC VERSION of 0x2? don't have reg_02
1681 * or reg_03, but the value of reg_0[23] is read as the previous read
1682 * register value, so ignore it if reg_03 == reg_0[12].
1684 if (reg_01.bits.version >= 0x20 && reg_03.raw != reg_02.raw &&
1685 reg_03.raw != reg_01.raw) {
1686 printk(KERN_DEBUG ".... register #03: %08X\n", reg_03.raw);
1687 printk(KERN_DEBUG "....... : Boot DT : %X\n", reg_03.bits.boot_DT);
1690 printk(KERN_DEBUG ".... IRQ redirection table:\n");
1692 printk(KERN_DEBUG " NR Dst Mask Trig IRR Pol"
1693 " Stat Dmod Deli Vect:\n");
1695 for (i = 0; i <= reg_01.bits.entries; i++) {
1696 struct IO_APIC_route_entry entry;
1698 entry = ioapic_read_entry(apic, i);
1700 printk(KERN_DEBUG " %02x %03X ",
1705 printk("%1d %1d %1d %1d %1d %1d %1d %02X\n",
1710 entry.delivery_status,
1712 entry.delivery_mode,
1717 printk(KERN_DEBUG "IRQ to pin mappings:\n");
1718 for_each_irq_desc(irq, desc) {
1719 struct irq_pin_list *entry;
1721 cfg = desc->chip_data;
1724 entry = cfg->irq_2_pin;
1727 printk(KERN_DEBUG "IRQ%d ", irq);
1728 for_each_irq_pin(entry, cfg->irq_2_pin)
1729 printk("-> %d:%d", entry->apic, entry->pin);
1733 printk(KERN_INFO ".................................... done.\n");
1738 __apicdebuginit(void) print_APIC_field(int base)
1744 for (i = 0; i < 8; i++)
1745 printk(KERN_CONT "%08x", apic_read(base + i*0x10));
1747 printk(KERN_CONT "\n");
1750 __apicdebuginit(void) print_local_APIC(void *dummy)
1752 unsigned int i, v, ver, maxlvt;
1755 printk(KERN_DEBUG "printing local APIC contents on CPU#%d/%d:\n",
1756 smp_processor_id(), hard_smp_processor_id());
1757 v = apic_read(APIC_ID);
1758 printk(KERN_INFO "... APIC ID: %08x (%01x)\n", v, read_apic_id());
1759 v = apic_read(APIC_LVR);
1760 printk(KERN_INFO "... APIC VERSION: %08x\n", v);
1761 ver = GET_APIC_VERSION(v);
1762 maxlvt = lapic_get_maxlvt();
1764 v = apic_read(APIC_TASKPRI);
1765 printk(KERN_DEBUG "... APIC TASKPRI: %08x (%02x)\n", v, v & APIC_TPRI_MASK);
1767 if (APIC_INTEGRATED(ver)) { /* !82489DX */
1768 if (!APIC_XAPIC(ver)) {
1769 v = apic_read(APIC_ARBPRI);
1770 printk(KERN_DEBUG "... APIC ARBPRI: %08x (%02x)\n", v,
1771 v & APIC_ARBPRI_MASK);
1773 v = apic_read(APIC_PROCPRI);
1774 printk(KERN_DEBUG "... APIC PROCPRI: %08x\n", v);
1778 * Remote read supported only in the 82489DX and local APIC for
1779 * Pentium processors.
1781 if (!APIC_INTEGRATED(ver) || maxlvt == 3) {
1782 v = apic_read(APIC_RRR);
1783 printk(KERN_DEBUG "... APIC RRR: %08x\n", v);
1786 v = apic_read(APIC_LDR);
1787 printk(KERN_DEBUG "... APIC LDR: %08x\n", v);
1788 if (!x2apic_enabled()) {
1789 v = apic_read(APIC_DFR);
1790 printk(KERN_DEBUG "... APIC DFR: %08x\n", v);
1792 v = apic_read(APIC_SPIV);
1793 printk(KERN_DEBUG "... APIC SPIV: %08x\n", v);
1795 printk(KERN_DEBUG "... APIC ISR field:\n");
1796 print_APIC_field(APIC_ISR);
1797 printk(KERN_DEBUG "... APIC TMR field:\n");
1798 print_APIC_field(APIC_TMR);
1799 printk(KERN_DEBUG "... APIC IRR field:\n");
1800 print_APIC_field(APIC_IRR);
1802 if (APIC_INTEGRATED(ver)) { /* !82489DX */
1803 if (maxlvt > 3) /* Due to the Pentium erratum 3AP. */
1804 apic_write(APIC_ESR, 0);
1806 v = apic_read(APIC_ESR);
1807 printk(KERN_DEBUG "... APIC ESR: %08x\n", v);
1810 icr = apic_icr_read();
1811 printk(KERN_DEBUG "... APIC ICR: %08x\n", (u32)icr);
1812 printk(KERN_DEBUG "... APIC ICR2: %08x\n", (u32)(icr >> 32));
1814 v = apic_read(APIC_LVTT);
1815 printk(KERN_DEBUG "... APIC LVTT: %08x\n", v);
1817 if (maxlvt > 3) { /* PC is LVT#4. */
1818 v = apic_read(APIC_LVTPC);
1819 printk(KERN_DEBUG "... APIC LVTPC: %08x\n", v);
1821 v = apic_read(APIC_LVT0);
1822 printk(KERN_DEBUG "... APIC LVT0: %08x\n", v);
1823 v = apic_read(APIC_LVT1);
1824 printk(KERN_DEBUG "... APIC LVT1: %08x\n", v);
1826 if (maxlvt > 2) { /* ERR is LVT#3. */
1827 v = apic_read(APIC_LVTERR);
1828 printk(KERN_DEBUG "... APIC LVTERR: %08x\n", v);
1831 v = apic_read(APIC_TMICT);
1832 printk(KERN_DEBUG "... APIC TMICT: %08x\n", v);
1833 v = apic_read(APIC_TMCCT);
1834 printk(KERN_DEBUG "... APIC TMCCT: %08x\n", v);
1835 v = apic_read(APIC_TDCR);
1836 printk(KERN_DEBUG "... APIC TDCR: %08x\n", v);
1838 if (boot_cpu_has(X86_FEATURE_EXTAPIC)) {
1839 v = apic_read(APIC_EFEAT);
1840 maxlvt = (v >> 16) & 0xff;
1841 printk(KERN_DEBUG "... APIC EFEAT: %08x\n", v);
1842 v = apic_read(APIC_ECTRL);
1843 printk(KERN_DEBUG "... APIC ECTRL: %08x\n", v);
1844 for (i = 0; i < maxlvt; i++) {
1845 v = apic_read(APIC_EILVTn(i));
1846 printk(KERN_DEBUG "... APIC EILVT%d: %08x\n", i, v);
1852 __apicdebuginit(void) print_local_APICs(int maxcpu)
1860 for_each_online_cpu(cpu) {
1863 smp_call_function_single(cpu, print_local_APIC, NULL, 1);
1868 __apicdebuginit(void) print_PIC(void)
1871 unsigned long flags;
1873 if (!legacy_pic->nr_legacy_irqs)
1876 printk(KERN_DEBUG "\nprinting PIC contents\n");
1878 raw_spin_lock_irqsave(&i8259A_lock, flags);
1880 v = inb(0xa1) << 8 | inb(0x21);
1881 printk(KERN_DEBUG "... PIC IMR: %04x\n", v);
1883 v = inb(0xa0) << 8 | inb(0x20);
1884 printk(KERN_DEBUG "... PIC IRR: %04x\n", v);
1888 v = inb(0xa0) << 8 | inb(0x20);
1892 raw_spin_unlock_irqrestore(&i8259A_lock, flags);
1894 printk(KERN_DEBUG "... PIC ISR: %04x\n", v);
1896 v = inb(0x4d1) << 8 | inb(0x4d0);
1897 printk(KERN_DEBUG "... PIC ELCR: %04x\n", v);
1900 static int __initdata show_lapic = 1;
1901 static __init int setup_show_lapic(char *arg)
1905 if (strcmp(arg, "all") == 0) {
1906 show_lapic = CONFIG_NR_CPUS;
1908 get_option(&arg, &num);
1915 __setup("show_lapic=", setup_show_lapic);
1917 __apicdebuginit(int) print_ICs(void)
1919 if (apic_verbosity == APIC_QUIET)
1924 /* don't print out if apic is not there */
1925 if (!cpu_has_apic && !apic_from_smp_config())
1928 print_local_APICs(show_lapic);
1934 fs_initcall(print_ICs);
1937 /* Where if anywhere is the i8259 connect in external int mode */
1938 static struct { int pin, apic; } ioapic_i8259 = { -1, -1 };
1940 void __init enable_IO_APIC(void)
1942 int i8259_apic, i8259_pin;
1945 if (!legacy_pic->nr_legacy_irqs)
1948 for(apic = 0; apic < nr_ioapics; apic++) {
1950 /* See if any of the pins is in ExtINT mode */
1951 for (pin = 0; pin < nr_ioapic_registers[apic]; pin++) {
1952 struct IO_APIC_route_entry entry;
1953 entry = ioapic_read_entry(apic, pin);
1955 /* If the interrupt line is enabled and in ExtInt mode
1956 * I have found the pin where the i8259 is connected.
1958 if ((entry.mask == 0) && (entry.delivery_mode == dest_ExtINT)) {
1959 ioapic_i8259.apic = apic;
1960 ioapic_i8259.pin = pin;
1966 /* Look to see what if the MP table has reported the ExtINT */
1967 /* If we could not find the appropriate pin by looking at the ioapic
1968 * the i8259 probably is not connected the ioapic but give the
1969 * mptable a chance anyway.
1971 i8259_pin = find_isa_irq_pin(0, mp_ExtINT);
1972 i8259_apic = find_isa_irq_apic(0, mp_ExtINT);
1973 /* Trust the MP table if nothing is setup in the hardware */
1974 if ((ioapic_i8259.pin == -1) && (i8259_pin >= 0)) {
1975 printk(KERN_WARNING "ExtINT not setup in hardware but reported by MP table\n");
1976 ioapic_i8259.pin = i8259_pin;
1977 ioapic_i8259.apic = i8259_apic;
1979 /* Complain if the MP table and the hardware disagree */
1980 if (((ioapic_i8259.apic != i8259_apic) || (ioapic_i8259.pin != i8259_pin)) &&
1981 (i8259_pin >= 0) && (ioapic_i8259.pin >= 0))
1983 printk(KERN_WARNING "ExtINT in hardware and MP table differ\n");
1987 * Do not trust the IO-APIC being empty at bootup
1993 * Not an __init, needed by the reboot code
1995 void disable_IO_APIC(void)
1998 * Clear the IO-APIC before rebooting:
2002 if (!legacy_pic->nr_legacy_irqs)
2006 * If the i8259 is routed through an IOAPIC
2007 * Put that IOAPIC in virtual wire mode
2008 * so legacy interrupts can be delivered.
2010 * With interrupt-remapping, for now we will use virtual wire A mode,
2011 * as virtual wire B is little complex (need to configure both
2012 * IOAPIC RTE aswell as interrupt-remapping table entry).
2013 * As this gets called during crash dump, keep this simple for now.
2015 if (ioapic_i8259.pin != -1 && !intr_remapping_enabled) {
2016 struct IO_APIC_route_entry entry;
2018 memset(&entry, 0, sizeof(entry));
2019 entry.mask = 0; /* Enabled */
2020 entry.trigger = 0; /* Edge */
2022 entry.polarity = 0; /* High */
2023 entry.delivery_status = 0;
2024 entry.dest_mode = 0; /* Physical */
2025 entry.delivery_mode = dest_ExtINT; /* ExtInt */
2027 entry.dest = read_apic_id();
2030 * Add it to the IO-APIC irq-routing table:
2032 ioapic_write_entry(ioapic_i8259.apic, ioapic_i8259.pin, entry);
2036 * Use virtual wire A mode when interrupt remapping is enabled.
2038 if (cpu_has_apic || apic_from_smp_config())
2039 disconnect_bsp_APIC(!intr_remapping_enabled &&
2040 ioapic_i8259.pin != -1);
2043 #ifdef CONFIG_X86_32
2045 * function to set the IO-APIC physical IDs based on the
2046 * values stored in the MPC table.
2048 * by Matt Domsch <Matt_Domsch@dell.com> Tue Dec 21 12:25:05 CST 1999
2051 void __init setup_ioapic_ids_from_mpc(void)
2053 union IO_APIC_reg_00 reg_00;
2054 physid_mask_t phys_id_present_map;
2057 unsigned char old_id;
2058 unsigned long flags;
2063 * Don't check I/O APIC IDs for xAPIC systems. They have
2064 * no meaning without the serial APIC bus.
2066 if (!(boot_cpu_data.x86_vendor == X86_VENDOR_INTEL)
2067 || APIC_XAPIC(apic_version[boot_cpu_physical_apicid]))
2070 * This is broken; anything with a real cpu count has to
2071 * circumvent this idiocy regardless.
2073 apic->ioapic_phys_id_map(&phys_cpu_present_map, &phys_id_present_map);
2076 * Set the IOAPIC ID to the value stored in the MPC table.
2078 for (apic_id = 0; apic_id < nr_ioapics; apic_id++) {
2080 /* Read the register 0 value */
2081 raw_spin_lock_irqsave(&ioapic_lock, flags);
2082 reg_00.raw = io_apic_read(apic_id, 0);
2083 raw_spin_unlock_irqrestore(&ioapic_lock, flags);
2085 old_id = mp_ioapics[apic_id].apicid;
2087 if (mp_ioapics[apic_id].apicid >= get_physical_broadcast()) {
2088 printk(KERN_ERR "BIOS bug, IO-APIC#%d ID is %d in the MPC table!...\n",
2089 apic_id, mp_ioapics[apic_id].apicid);
2090 printk(KERN_ERR "... fixing up to %d. (tell your hw vendor)\n",
2092 mp_ioapics[apic_id].apicid = reg_00.bits.ID;
2096 * Sanity check, is the ID really free? Every APIC in a
2097 * system must have a unique ID or we get lots of nice
2098 * 'stuck on smp_invalidate_needed IPI wait' messages.
2100 if (apic->check_apicid_used(&phys_id_present_map,
2101 mp_ioapics[apic_id].apicid)) {
2102 printk(KERN_ERR "BIOS bug, IO-APIC#%d ID %d is already used!...\n",
2103 apic_id, mp_ioapics[apic_id].apicid);
2104 for (i = 0; i < get_physical_broadcast(); i++)
2105 if (!physid_isset(i, phys_id_present_map))
2107 if (i >= get_physical_broadcast())
2108 panic("Max APIC ID exceeded!\n");
2109 printk(KERN_ERR "... fixing up to %d. (tell your hw vendor)\n",
2111 physid_set(i, phys_id_present_map);
2112 mp_ioapics[apic_id].apicid = i;
2115 apic->apicid_to_cpu_present(mp_ioapics[apic_id].apicid, &tmp);
2116 apic_printk(APIC_VERBOSE, "Setting %d in the "
2117 "phys_id_present_map\n",
2118 mp_ioapics[apic_id].apicid);
2119 physids_or(phys_id_present_map, phys_id_present_map, tmp);
2124 * We need to adjust the IRQ routing table
2125 * if the ID changed.
2127 if (old_id != mp_ioapics[apic_id].apicid)
2128 for (i = 0; i < mp_irq_entries; i++)
2129 if (mp_irqs[i].dstapic == old_id)
2131 = mp_ioapics[apic_id].apicid;
2134 * Read the right value from the MPC table and
2135 * write it into the ID register.
2137 apic_printk(APIC_VERBOSE, KERN_INFO
2138 "...changing IO-APIC physical APIC ID to %d ...",
2139 mp_ioapics[apic_id].apicid);
2141 reg_00.bits.ID = mp_ioapics[apic_id].apicid;
2142 raw_spin_lock_irqsave(&ioapic_lock, flags);
2143 io_apic_write(apic_id, 0, reg_00.raw);
2144 raw_spin_unlock_irqrestore(&ioapic_lock, flags);
2149 raw_spin_lock_irqsave(&ioapic_lock, flags);
2150 reg_00.raw = io_apic_read(apic_id, 0);
2151 raw_spin_unlock_irqrestore(&ioapic_lock, flags);
2152 if (reg_00.bits.ID != mp_ioapics[apic_id].apicid)
2153 printk("could not set ID!\n");
2155 apic_printk(APIC_VERBOSE, " ok.\n");
2160 int no_timer_check __initdata;
2162 static int __init notimercheck(char *s)
2167 __setup("no_timer_check", notimercheck);
2170 * There is a nasty bug in some older SMP boards, their mptable lies
2171 * about the timer IRQ. We do the following to work around the situation:
2173 * - timer IRQ defaults to IO-APIC IRQ
2174 * - if this function detects that timer IRQs are defunct, then we fall
2175 * back to ISA timer IRQs
2177 static int __init timer_irq_works(void)
2179 unsigned long t1 = jiffies;
2180 unsigned long flags;
2185 local_save_flags(flags);
2187 /* Let ten ticks pass... */
2188 mdelay((10 * 1000) / HZ);
2189 local_irq_restore(flags);
2192 * Expect a few ticks at least, to be sure some possible
2193 * glue logic does not lock up after one or two first
2194 * ticks in a non-ExtINT mode. Also the local APIC
2195 * might have cached one ExtINT interrupt. Finally, at
2196 * least one tick may be lost due to delays.
2200 if (time_after(jiffies, t1 + 4))
2206 * In the SMP+IOAPIC case it might happen that there are an unspecified
2207 * number of pending IRQ events unhandled. These cases are very rare,
2208 * so we 'resend' these IRQs via IPIs, to the same CPU. It's much
2209 * better to do it this way as thus we do not have to be aware of
2210 * 'pending' interrupts in the IRQ path, except at this point.
2213 * Edge triggered needs to resend any interrupt
2214 * that was delayed but this is now handled in the device
2219 * Starting up a edge-triggered IO-APIC interrupt is
2220 * nasty - we need to make sure that we get the edge.
2221 * If it is already asserted for some reason, we need
2222 * return 1 to indicate that is was pending.
2224 * This is not complete - we should be able to fake
2225 * an edge even if it isn't on the 8259A...
2228 static unsigned int startup_ioapic_irq(unsigned int irq)
2230 int was_pending = 0;
2231 unsigned long flags;
2232 struct irq_cfg *cfg;
2234 raw_spin_lock_irqsave(&ioapic_lock, flags);
2235 if (irq < legacy_pic->nr_legacy_irqs) {
2236 legacy_pic->chip->mask(irq);
2237 if (legacy_pic->irq_pending(irq))
2241 __unmask_IO_APIC_irq(cfg);
2242 raw_spin_unlock_irqrestore(&ioapic_lock, flags);
2247 static int ioapic_retrigger_irq(unsigned int irq)
2250 struct irq_cfg *cfg = irq_cfg(irq);
2251 unsigned long flags;
2253 raw_spin_lock_irqsave(&vector_lock, flags);
2254 apic->send_IPI_mask(cpumask_of(cpumask_first(cfg->domain)), cfg->vector);
2255 raw_spin_unlock_irqrestore(&vector_lock, flags);
2261 * Level and edge triggered IO-APIC interrupts need different handling,
2262 * so we use two separate IRQ descriptors. Edge triggered IRQs can be
2263 * handled with the level-triggered descriptor, but that one has slightly
2264 * more overhead. Level-triggered interrupts cannot be handled with the
2265 * edge-triggered handler, without risking IRQ storms and other ugly
2270 void send_cleanup_vector(struct irq_cfg *cfg)
2272 cpumask_var_t cleanup_mask;
2274 if (unlikely(!alloc_cpumask_var(&cleanup_mask, GFP_ATOMIC))) {
2276 for_each_cpu_and(i, cfg->old_domain, cpu_online_mask)
2277 apic->send_IPI_mask(cpumask_of(i), IRQ_MOVE_CLEANUP_VECTOR);
2279 cpumask_and(cleanup_mask, cfg->old_domain, cpu_online_mask);
2280 apic->send_IPI_mask(cleanup_mask, IRQ_MOVE_CLEANUP_VECTOR);
2281 free_cpumask_var(cleanup_mask);
2283 cfg->move_in_progress = 0;
2286 static void __target_IO_APIC_irq(unsigned int irq, unsigned int dest, struct irq_cfg *cfg)
2289 struct irq_pin_list *entry;
2290 u8 vector = cfg->vector;
2292 for_each_irq_pin(entry, cfg->irq_2_pin) {
2298 * With interrupt-remapping, destination information comes
2299 * from interrupt-remapping table entry.
2301 if (!irq_remapped(irq))
2302 io_apic_write(apic, 0x11 + pin*2, dest);
2303 reg = io_apic_read(apic, 0x10 + pin*2);
2304 reg &= ~IO_APIC_REDIR_VECTOR_MASK;
2306 io_apic_modify(apic, 0x10 + pin*2, reg);
2311 * Either sets desc->affinity to a valid value, and returns
2312 * ->cpu_mask_to_apicid of that in dest_id, or returns -1 and
2313 * leaves desc->affinity untouched.
2316 set_desc_affinity(struct irq_desc *desc, const struct cpumask *mask,
2317 unsigned int *dest_id)
2319 struct irq_cfg *cfg;
2322 if (!cpumask_intersects(mask, cpu_online_mask))
2326 cfg = desc->chip_data;
2327 if (assign_irq_vector(irq, cfg, mask))
2330 cpumask_copy(desc->affinity, mask);
2332 *dest_id = apic->cpu_mask_to_apicid_and(desc->affinity, cfg->domain);
2337 set_ioapic_affinity_irq_desc(struct irq_desc *desc, const struct cpumask *mask)
2339 struct irq_cfg *cfg;
2340 unsigned long flags;
2346 cfg = desc->chip_data;
2348 raw_spin_lock_irqsave(&ioapic_lock, flags);
2349 ret = set_desc_affinity(desc, mask, &dest);
2351 /* Only the high 8 bits are valid. */
2352 dest = SET_APIC_LOGICAL_ID(dest);
2353 __target_IO_APIC_irq(irq, dest, cfg);
2355 raw_spin_unlock_irqrestore(&ioapic_lock, flags);
2361 set_ioapic_affinity_irq(unsigned int irq, const struct cpumask *mask)
2363 struct irq_desc *desc;
2365 desc = irq_to_desc(irq);
2367 return set_ioapic_affinity_irq_desc(desc, mask);
2370 #ifdef CONFIG_INTR_REMAP
2373 * Migrate the IO-APIC irq in the presence of intr-remapping.
2375 * For both level and edge triggered, irq migration is a simple atomic
2376 * update(of vector and cpu destination) of IRTE and flush the hardware cache.
2378 * For level triggered, we eliminate the io-apic RTE modification (with the
2379 * updated vector information), by using a virtual vector (io-apic pin number).
2380 * Real vector that is used for interrupting cpu will be coming from
2381 * the interrupt-remapping table entry.
2384 migrate_ioapic_irq_desc(struct irq_desc *desc, const struct cpumask *mask)
2386 struct irq_cfg *cfg;
2392 if (!cpumask_intersects(mask, cpu_online_mask))
2396 if (get_irte(irq, &irte))
2399 cfg = desc->chip_data;
2400 if (assign_irq_vector(irq, cfg, mask))
2403 dest = apic->cpu_mask_to_apicid_and(cfg->domain, mask);
2405 irte.vector = cfg->vector;
2406 irte.dest_id = IRTE_DEST(dest);
2409 * Modified the IRTE and flushes the Interrupt entry cache.
2411 modify_irte(irq, &irte);
2413 if (cfg->move_in_progress)
2414 send_cleanup_vector(cfg);
2416 cpumask_copy(desc->affinity, mask);
2422 * Migrates the IRQ destination in the process context.
2424 static int set_ir_ioapic_affinity_irq_desc(struct irq_desc *desc,
2425 const struct cpumask *mask)
2427 return migrate_ioapic_irq_desc(desc, mask);
2429 static int set_ir_ioapic_affinity_irq(unsigned int irq,
2430 const struct cpumask *mask)
2432 struct irq_desc *desc = irq_to_desc(irq);
2434 return set_ir_ioapic_affinity_irq_desc(desc, mask);
2437 static inline int set_ir_ioapic_affinity_irq_desc(struct irq_desc *desc,
2438 const struct cpumask *mask)
2444 asmlinkage void smp_irq_move_cleanup_interrupt(void)
2446 unsigned vector, me;
2452 me = smp_processor_id();
2453 for (vector = FIRST_EXTERNAL_VECTOR; vector < NR_VECTORS; vector++) {
2456 struct irq_desc *desc;
2457 struct irq_cfg *cfg;
2458 irq = __get_cpu_var(vector_irq)[vector];
2463 desc = irq_to_desc(irq);
2468 raw_spin_lock(&desc->lock);
2471 * Check if the irq migration is in progress. If so, we
2472 * haven't received the cleanup request yet for this irq.
2474 if (cfg->move_in_progress)
2477 if (vector == cfg->vector && cpumask_test_cpu(me, cfg->domain))
2480 irr = apic_read(APIC_IRR + (vector / 32 * 0x10));
2482 * Check if the vector that needs to be cleanedup is
2483 * registered at the cpu's IRR. If so, then this is not
2484 * the best time to clean it up. Lets clean it up in the
2485 * next attempt by sending another IRQ_MOVE_CLEANUP_VECTOR
2488 if (irr & (1 << (vector % 32))) {
2489 apic->send_IPI_self(IRQ_MOVE_CLEANUP_VECTOR);
2492 __get_cpu_var(vector_irq)[vector] = -1;
2494 raw_spin_unlock(&desc->lock);
2500 static void __irq_complete_move(struct irq_desc **descp, unsigned vector)
2502 struct irq_desc *desc = *descp;
2503 struct irq_cfg *cfg = desc->chip_data;
2506 if (likely(!cfg->move_in_progress))
2509 me = smp_processor_id();
2511 if (vector == cfg->vector && cpumask_test_cpu(me, cfg->domain))
2512 send_cleanup_vector(cfg);
2515 static void irq_complete_move(struct irq_desc **descp)
2517 __irq_complete_move(descp, ~get_irq_regs()->orig_ax);
2520 void irq_force_complete_move(int irq)
2522 struct irq_desc *desc = irq_to_desc(irq);
2523 struct irq_cfg *cfg = desc->chip_data;
2528 __irq_complete_move(&desc, cfg->vector);
2531 static inline void irq_complete_move(struct irq_desc **descp) {}
2534 static void ack_apic_edge(unsigned int irq)
2536 struct irq_desc *desc = irq_to_desc(irq);
2538 irq_complete_move(&desc);
2539 move_native_irq(irq);
2543 atomic_t irq_mis_count;
2546 * IO-APIC versions below 0x20 don't support EOI register.
2547 * For the record, here is the information about various versions:
2549 * 1Xh I/OAPIC or I/O(x)APIC which are not PCI 2.2 Compliant
2550 * 2Xh I/O(x)APIC which is PCI 2.2 Compliant
2553 * Some of the Intel ICH Specs (ICH2 to ICH5) documents the io-apic
2554 * version as 0x2. This is an error with documentation and these ICH chips
2555 * use io-apic's of version 0x20.
2557 * For IO-APIC's with EOI register, we use that to do an explicit EOI.
2558 * Otherwise, we simulate the EOI message manually by changing the trigger
2559 * mode to edge and then back to level, with RTE being masked during this.
2561 static void __eoi_ioapic_irq(unsigned int irq, struct irq_cfg *cfg)
2563 struct irq_pin_list *entry;
2565 for_each_irq_pin(entry, cfg->irq_2_pin) {
2566 if (mp_ioapics[entry->apic].apicver >= 0x20) {
2568 * Intr-remapping uses pin number as the virtual vector
2569 * in the RTE. Actual vector is programmed in
2570 * intr-remapping table entry. Hence for the io-apic
2571 * EOI we use the pin number.
2573 if (irq_remapped(irq))
2574 io_apic_eoi(entry->apic, entry->pin);
2576 io_apic_eoi(entry->apic, cfg->vector);
2578 __mask_and_edge_IO_APIC_irq(entry);
2579 __unmask_and_level_IO_APIC_irq(entry);
2584 static void eoi_ioapic_irq(struct irq_desc *desc)
2586 struct irq_cfg *cfg;
2587 unsigned long flags;
2591 cfg = desc->chip_data;
2593 raw_spin_lock_irqsave(&ioapic_lock, flags);
2594 __eoi_ioapic_irq(irq, cfg);
2595 raw_spin_unlock_irqrestore(&ioapic_lock, flags);
2598 static void ack_apic_level(unsigned int irq)
2600 struct irq_desc *desc = irq_to_desc(irq);
2603 struct irq_cfg *cfg;
2604 int do_unmask_irq = 0;
2606 irq_complete_move(&desc);
2607 #ifdef CONFIG_GENERIC_PENDING_IRQ
2608 /* If we are moving the irq we need to mask it */
2609 if (unlikely(desc->status & IRQ_MOVE_PENDING)) {
2611 mask_IO_APIC_irq_desc(desc);
2616 * It appears there is an erratum which affects at least version 0x11
2617 * of I/O APIC (that's the 82093AA and cores integrated into various
2618 * chipsets). Under certain conditions a level-triggered interrupt is
2619 * erroneously delivered as edge-triggered one but the respective IRR
2620 * bit gets set nevertheless. As a result the I/O unit expects an EOI
2621 * message but it will never arrive and further interrupts are blocked
2622 * from the source. The exact reason is so far unknown, but the
2623 * phenomenon was observed when two consecutive interrupt requests
2624 * from a given source get delivered to the same CPU and the source is
2625 * temporarily disabled in between.
2627 * A workaround is to simulate an EOI message manually. We achieve it
2628 * by setting the trigger mode to edge and then to level when the edge
2629 * trigger mode gets detected in the TMR of a local APIC for a
2630 * level-triggered interrupt. We mask the source for the time of the
2631 * operation to prevent an edge-triggered interrupt escaping meanwhile.
2632 * The idea is from Manfred Spraul. --macro
2634 * Also in the case when cpu goes offline, fixup_irqs() will forward
2635 * any unhandled interrupt on the offlined cpu to the new cpu
2636 * destination that is handling the corresponding interrupt. This
2637 * interrupt forwarding is done via IPI's. Hence, in this case also
2638 * level-triggered io-apic interrupt will be seen as an edge
2639 * interrupt in the IRR. And we can't rely on the cpu's EOI
2640 * to be broadcasted to the IO-APIC's which will clear the remoteIRR
2641 * corresponding to the level-triggered interrupt. Hence on IO-APIC's
2642 * supporting EOI register, we do an explicit EOI to clear the
2643 * remote IRR and on IO-APIC's which don't have an EOI register,
2644 * we use the above logic (mask+edge followed by unmask+level) from
2645 * Manfred Spraul to clear the remote IRR.
2647 cfg = desc->chip_data;
2649 v = apic_read(APIC_TMR + ((i & ~0x1f) >> 1));
2652 * We must acknowledge the irq before we move it or the acknowledge will
2653 * not propagate properly.
2658 * Tail end of clearing remote IRR bit (either by delivering the EOI
2659 * message via io-apic EOI register write or simulating it using
2660 * mask+edge followed by unnask+level logic) manually when the
2661 * level triggered interrupt is seen as the edge triggered interrupt
2664 if (!(v & (1 << (i & 0x1f)))) {
2665 atomic_inc(&irq_mis_count);
2667 eoi_ioapic_irq(desc);
2670 /* Now we can move and renable the irq */
2671 if (unlikely(do_unmask_irq)) {
2672 /* Only migrate the irq if the ack has been received.
2674 * On rare occasions the broadcast level triggered ack gets
2675 * delayed going to ioapics, and if we reprogram the
2676 * vector while Remote IRR is still set the irq will never
2679 * To prevent this scenario we read the Remote IRR bit
2680 * of the ioapic. This has two effects.
2681 * - On any sane system the read of the ioapic will
2682 * flush writes (and acks) going to the ioapic from
2684 * - We get to see if the ACK has actually been delivered.
2686 * Based on failed experiments of reprogramming the
2687 * ioapic entry from outside of irq context starting
2688 * with masking the ioapic entry and then polling until
2689 * Remote IRR was clear before reprogramming the
2690 * ioapic I don't trust the Remote IRR bit to be
2691 * completey accurate.
2693 * However there appears to be no other way to plug
2694 * this race, so if the Remote IRR bit is not
2695 * accurate and is causing problems then it is a hardware bug
2696 * and you can go talk to the chipset vendor about it.
2698 cfg = desc->chip_data;
2699 if (!io_apic_level_ack_pending(cfg))
2700 move_masked_irq(irq);
2701 unmask_IO_APIC_irq_desc(desc);
2705 #ifdef CONFIG_INTR_REMAP
2706 static void ir_ack_apic_edge(unsigned int irq)
2711 static void ir_ack_apic_level(unsigned int irq)
2713 struct irq_desc *desc = irq_to_desc(irq);
2716 eoi_ioapic_irq(desc);
2718 #endif /* CONFIG_INTR_REMAP */
2720 static struct irq_chip ioapic_chip __read_mostly = {
2722 .startup = startup_ioapic_irq,
2723 .mask = mask_IO_APIC_irq,
2724 .unmask = unmask_IO_APIC_irq,
2725 .ack = ack_apic_edge,
2726 .eoi = ack_apic_level,
2728 .set_affinity = set_ioapic_affinity_irq,
2730 .retrigger = ioapic_retrigger_irq,
2733 static struct irq_chip ir_ioapic_chip __read_mostly = {
2734 .name = "IR-IO-APIC",
2735 .startup = startup_ioapic_irq,
2736 .mask = mask_IO_APIC_irq,
2737 .unmask = unmask_IO_APIC_irq,
2738 #ifdef CONFIG_INTR_REMAP
2739 .ack = ir_ack_apic_edge,
2740 .eoi = ir_ack_apic_level,
2742 .set_affinity = set_ir_ioapic_affinity_irq,
2745 .retrigger = ioapic_retrigger_irq,
2748 static inline void init_IO_APIC_traps(void)
2751 struct irq_desc *desc;
2752 struct irq_cfg *cfg;
2755 * NOTE! The local APIC isn't very good at handling
2756 * multiple interrupts at the same interrupt level.
2757 * As the interrupt level is determined by taking the
2758 * vector number and shifting that right by 4, we
2759 * want to spread these out a bit so that they don't
2760 * all fall in the same interrupt level.
2762 * Also, we've got to be careful not to trash gate
2763 * 0x80, because int 0x80 is hm, kind of importantish. ;)
2765 for_each_irq_desc(irq, desc) {
2766 cfg = desc->chip_data;
2767 if (IO_APIC_IRQ(irq) && cfg && !cfg->vector) {
2769 * Hmm.. We don't have an entry for this,
2770 * so default to an old-fashioned 8259
2771 * interrupt if we can..
2773 if (irq < legacy_pic->nr_legacy_irqs)
2774 legacy_pic->make_irq(irq);
2776 /* Strange. Oh, well.. */
2777 desc->chip = &no_irq_chip;
2783 * The local APIC irq-chip implementation:
2786 static void mask_lapic_irq(unsigned int irq)
2790 v = apic_read(APIC_LVT0);
2791 apic_write(APIC_LVT0, v | APIC_LVT_MASKED);
2794 static void unmask_lapic_irq(unsigned int irq)
2798 v = apic_read(APIC_LVT0);
2799 apic_write(APIC_LVT0, v & ~APIC_LVT_MASKED);
2802 static void ack_lapic_irq(unsigned int irq)
2807 static struct irq_chip lapic_chip __read_mostly = {
2808 .name = "local-APIC",
2809 .mask = mask_lapic_irq,
2810 .unmask = unmask_lapic_irq,
2811 .ack = ack_lapic_irq,
2814 static void lapic_register_intr(int irq, struct irq_desc *desc)
2816 desc->status &= ~IRQ_LEVEL;
2817 set_irq_chip_and_handler_name(irq, &lapic_chip, handle_edge_irq,
2821 static void __init setup_nmi(void)
2824 * Dirty trick to enable the NMI watchdog ...
2825 * We put the 8259A master into AEOI mode and
2826 * unmask on all local APICs LVT0 as NMI.
2828 * The idea to use the 8259A in AEOI mode ('8259A Virtual Wire')
2829 * is from Maciej W. Rozycki - so we do not have to EOI from
2830 * the NMI handler or the timer interrupt.
2832 apic_printk(APIC_VERBOSE, KERN_INFO "activating NMI Watchdog ...");
2834 enable_NMI_through_LVT0();
2836 apic_printk(APIC_VERBOSE, " done.\n");
2840 * This looks a bit hackish but it's about the only one way of sending
2841 * a few INTA cycles to 8259As and any associated glue logic. ICR does
2842 * not support the ExtINT mode, unfortunately. We need to send these
2843 * cycles as some i82489DX-based boards have glue logic that keeps the
2844 * 8259A interrupt line asserted until INTA. --macro
2846 static inline void __init unlock_ExtINT_logic(void)
2849 struct IO_APIC_route_entry entry0, entry1;
2850 unsigned char save_control, save_freq_select;
2852 pin = find_isa_irq_pin(8, mp_INT);
2857 apic = find_isa_irq_apic(8, mp_INT);
2863 entry0 = ioapic_read_entry(apic, pin);
2864 clear_IO_APIC_pin(apic, pin);
2866 memset(&entry1, 0, sizeof(entry1));
2868 entry1.dest_mode = 0; /* physical delivery */
2869 entry1.mask = 0; /* unmask IRQ now */
2870 entry1.dest = hard_smp_processor_id();
2871 entry1.delivery_mode = dest_ExtINT;
2872 entry1.polarity = entry0.polarity;
2876 ioapic_write_entry(apic, pin, entry1);
2878 save_control = CMOS_READ(RTC_CONTROL);
2879 save_freq_select = CMOS_READ(RTC_FREQ_SELECT);
2880 CMOS_WRITE((save_freq_select & ~RTC_RATE_SELECT) | 0x6,
2882 CMOS_WRITE(save_control | RTC_PIE, RTC_CONTROL);
2887 if ((CMOS_READ(RTC_INTR_FLAGS) & RTC_PF) == RTC_PF)
2891 CMOS_WRITE(save_control, RTC_CONTROL);
2892 CMOS_WRITE(save_freq_select, RTC_FREQ_SELECT);
2893 clear_IO_APIC_pin(apic, pin);
2895 ioapic_write_entry(apic, pin, entry0);
2898 static int disable_timer_pin_1 __initdata;
2899 /* Actually the next is obsolete, but keep it for paranoid reasons -AK */
2900 static int __init disable_timer_pin_setup(char *arg)
2902 disable_timer_pin_1 = 1;
2905 early_param("disable_timer_pin_1", disable_timer_pin_setup);
2907 int timer_through_8259 __initdata;
2910 * This code may look a bit paranoid, but it's supposed to cooperate with
2911 * a wide range of boards and BIOS bugs. Fortunately only the timer IRQ
2912 * is so screwy. Thanks to Brian Perkins for testing/hacking this beast
2913 * fanatically on his truly buggy board.
2915 * FIXME: really need to revamp this for all platforms.
2917 static inline void __init check_timer(void)
2919 struct irq_desc *desc = irq_to_desc(0);
2920 struct irq_cfg *cfg = desc->chip_data;
2921 int node = cpu_to_node(0);
2922 int apic1, pin1, apic2, pin2;
2923 unsigned long flags;
2926 local_irq_save(flags);
2929 * get/set the timer IRQ vector:
2931 legacy_pic->chip->mask(0);
2932 assign_irq_vector(0, cfg, apic->target_cpus());
2935 * As IRQ0 is to be enabled in the 8259A, the virtual
2936 * wire has to be disabled in the local APIC. Also
2937 * timer interrupts need to be acknowledged manually in
2938 * the 8259A for the i82489DX when using the NMI
2939 * watchdog as that APIC treats NMIs as level-triggered.
2940 * The AEOI mode will finish them in the 8259A
2943 apic_write(APIC_LVT0, APIC_LVT_MASKED | APIC_DM_EXTINT);
2944 legacy_pic->init(1);
2945 #ifdef CONFIG_X86_32
2949 ver = apic_read(APIC_LVR);
2950 ver = GET_APIC_VERSION(ver);
2951 timer_ack = (nmi_watchdog == NMI_IO_APIC && !APIC_INTEGRATED(ver));
2955 pin1 = find_isa_irq_pin(0, mp_INT);
2956 apic1 = find_isa_irq_apic(0, mp_INT);
2957 pin2 = ioapic_i8259.pin;
2958 apic2 = ioapic_i8259.apic;
2960 apic_printk(APIC_QUIET, KERN_INFO "..TIMER: vector=0x%02X "
2961 "apic1=%d pin1=%d apic2=%d pin2=%d\n",
2962 cfg->vector, apic1, pin1, apic2, pin2);
2965 * Some BIOS writers are clueless and report the ExtINTA
2966 * I/O APIC input from the cascaded 8259A as the timer
2967 * interrupt input. So just in case, if only one pin
2968 * was found above, try it both directly and through the
2972 if (intr_remapping_enabled)
2973 panic("BIOS bug: timer not connected to IO-APIC");
2977 } else if (pin2 == -1) {
2984 * Ok, does IRQ0 through the IOAPIC work?
2987 add_pin_to_irq_node(cfg, node, apic1, pin1);
2988 setup_timer_IRQ0_pin(apic1, pin1, cfg->vector);
2990 /* for edge trigger, setup_IO_APIC_irq already
2991 * leave it unmasked.
2992 * so only need to unmask if it is level-trigger
2993 * do we really have level trigger timer?
2996 idx = find_irq_entry(apic1, pin1, mp_INT);
2997 if (idx != -1 && irq_trigger(idx))
2998 unmask_IO_APIC_irq_desc(desc);
3000 if (timer_irq_works()) {
3001 if (nmi_watchdog == NMI_IO_APIC) {
3003 legacy_pic->chip->unmask(0);
3005 if (disable_timer_pin_1 > 0)
3006 clear_IO_APIC_pin(0, pin1);
3009 if (intr_remapping_enabled)
3010 panic("timer doesn't work through Interrupt-remapped IO-APIC");
3011 local_irq_disable();
3012 clear_IO_APIC_pin(apic1, pin1);
3014 apic_printk(APIC_QUIET, KERN_ERR "..MP-BIOS bug: "
3015 "8254 timer not connected to IO-APIC\n");
3017 apic_printk(APIC_QUIET, KERN_INFO "...trying to set up timer "
3018 "(IRQ0) through the 8259A ...\n");
3019 apic_printk(APIC_QUIET, KERN_INFO
3020 "..... (found apic %d pin %d) ...\n", apic2, pin2);
3022 * legacy devices should be connected to IO APIC #0
3024 replace_pin_at_irq_node(cfg, node, apic1, pin1, apic2, pin2);
3025 setup_timer_IRQ0_pin(apic2, pin2, cfg->vector);
3026 legacy_pic->chip->unmask(0);
3027 if (timer_irq_works()) {
3028 apic_printk(APIC_QUIET, KERN_INFO "....... works.\n");
3029 timer_through_8259 = 1;
3030 if (nmi_watchdog == NMI_IO_APIC) {
3031 legacy_pic->chip->mask(0);
3033 legacy_pic->chip->unmask(0);
3038 * Cleanup, just in case ...
3040 local_irq_disable();
3041 legacy_pic->chip->mask(0);
3042 clear_IO_APIC_pin(apic2, pin2);
3043 apic_printk(APIC_QUIET, KERN_INFO "....... failed.\n");
3046 if (nmi_watchdog == NMI_IO_APIC) {
3047 apic_printk(APIC_QUIET, KERN_WARNING "timer doesn't work "
3048 "through the IO-APIC - disabling NMI Watchdog!\n");
3049 nmi_watchdog = NMI_NONE;
3051 #ifdef CONFIG_X86_32
3055 apic_printk(APIC_QUIET, KERN_INFO
3056 "...trying to set up timer as Virtual Wire IRQ...\n");
3058 lapic_register_intr(0, desc);
3059 apic_write(APIC_LVT0, APIC_DM_FIXED | cfg->vector); /* Fixed mode */
3060 legacy_pic->chip->unmask(0);
3062 if (timer_irq_works()) {
3063 apic_printk(APIC_QUIET, KERN_INFO "..... works.\n");
3066 local_irq_disable();
3067 legacy_pic->chip->mask(0);
3068 apic_write(APIC_LVT0, APIC_LVT_MASKED | APIC_DM_FIXED | cfg->vector);
3069 apic_printk(APIC_QUIET, KERN_INFO "..... failed.\n");
3071 apic_printk(APIC_QUIET, KERN_INFO
3072 "...trying to set up timer as ExtINT IRQ...\n");
3074 legacy_pic->init(0);
3075 legacy_pic->make_irq(0);
3076 apic_write(APIC_LVT0, APIC_DM_EXTINT);
3078 unlock_ExtINT_logic();
3080 if (timer_irq_works()) {
3081 apic_printk(APIC_QUIET, KERN_INFO "..... works.\n");
3084 local_irq_disable();
3085 apic_printk(APIC_QUIET, KERN_INFO "..... failed :(.\n");
3086 panic("IO-APIC + timer doesn't work! Boot with apic=debug and send a "
3087 "report. Then try booting with the 'noapic' option.\n");
3089 local_irq_restore(flags);
3093 * Traditionally ISA IRQ2 is the cascade IRQ, and is not available
3094 * to devices. However there may be an I/O APIC pin available for
3095 * this interrupt regardless. The pin may be left unconnected, but
3096 * typically it will be reused as an ExtINT cascade interrupt for
3097 * the master 8259A. In the MPS case such a pin will normally be
3098 * reported as an ExtINT interrupt in the MP table. With ACPI
3099 * there is no provision for ExtINT interrupts, and in the absence
3100 * of an override it would be treated as an ordinary ISA I/O APIC
3101 * interrupt, that is edge-triggered and unmasked by default. We
3102 * used to do this, but it caused problems on some systems because
3103 * of the NMI watchdog and sometimes IRQ0 of the 8254 timer using
3104 * the same ExtINT cascade interrupt to drive the local APIC of the
3105 * bootstrap processor. Therefore we refrain from routing IRQ2 to
3106 * the I/O APIC in all cases now. No actual device should request
3107 * it anyway. --macro
3109 #define PIC_IRQS (1UL << PIC_CASCADE_IR)
3111 void __init setup_IO_APIC(void)
3115 * calling enable_IO_APIC() is moved to setup_local_APIC for BP
3117 io_apic_irqs = legacy_pic->nr_legacy_irqs ? ~PIC_IRQS : ~0UL;
3119 apic_printk(APIC_VERBOSE, "ENABLING IO-APIC IRQs\n");
3121 * Set up IO-APIC IRQ routing.
3123 x86_init.mpparse.setup_ioapic_ids();
3126 setup_IO_APIC_irqs();
3127 init_IO_APIC_traps();
3128 if (legacy_pic->nr_legacy_irqs)
3133 * Called after all the initialization is done. If we didnt find any
3134 * APIC bugs then we can allow the modify fast path
3137 static int __init io_apic_bug_finalize(void)
3139 if (sis_apic_bug == -1)
3144 late_initcall(io_apic_bug_finalize);
3146 struct sysfs_ioapic_data {
3147 struct sys_device dev;
3148 struct IO_APIC_route_entry entry[0];
3150 static struct sysfs_ioapic_data * mp_ioapic_data[MAX_IO_APICS];
3152 static int ioapic_suspend(struct sys_device *dev, pm_message_t state)
3154 struct IO_APIC_route_entry *entry;
3155 struct sysfs_ioapic_data *data;
3158 data = container_of(dev, struct sysfs_ioapic_data, dev);
3159 entry = data->entry;
3160 for (i = 0; i < nr_ioapic_registers[dev->id]; i ++, entry ++ )
3161 *entry = ioapic_read_entry(dev->id, i);
3166 static int ioapic_resume(struct sys_device *dev)
3168 struct IO_APIC_route_entry *entry;
3169 struct sysfs_ioapic_data *data;
3170 unsigned long flags;
3171 union IO_APIC_reg_00 reg_00;
3174 data = container_of(dev, struct sysfs_ioapic_data, dev);
3175 entry = data->entry;
3177 raw_spin_lock_irqsave(&ioapic_lock, flags);
3178 reg_00.raw = io_apic_read(dev->id, 0);
3179 if (reg_00.bits.ID != mp_ioapics[dev->id].apicid) {
3180 reg_00.bits.ID = mp_ioapics[dev->id].apicid;
3181 io_apic_write(dev->id, 0, reg_00.raw);
3183 raw_spin_unlock_irqrestore(&ioapic_lock, flags);
3184 for (i = 0; i < nr_ioapic_registers[dev->id]; i++)
3185 ioapic_write_entry(dev->id, i, entry[i]);
3190 static struct sysdev_class ioapic_sysdev_class = {
3192 .suspend = ioapic_suspend,
3193 .resume = ioapic_resume,
3196 static int __init ioapic_init_sysfs(void)
3198 struct sys_device * dev;
3201 error = sysdev_class_register(&ioapic_sysdev_class);
3205 for (i = 0; i < nr_ioapics; i++ ) {
3206 size = sizeof(struct sys_device) + nr_ioapic_registers[i]
3207 * sizeof(struct IO_APIC_route_entry);
3208 mp_ioapic_data[i] = kzalloc(size, GFP_KERNEL);
3209 if (!mp_ioapic_data[i]) {
3210 printk(KERN_ERR "Can't suspend/resume IOAPIC %d\n", i);
3213 dev = &mp_ioapic_data[i]->dev;
3215 dev->cls = &ioapic_sysdev_class;
3216 error = sysdev_register(dev);
3218 kfree(mp_ioapic_data[i]);
3219 mp_ioapic_data[i] = NULL;
3220 printk(KERN_ERR "Can't suspend/resume IOAPIC %d\n", i);
3228 device_initcall(ioapic_init_sysfs);
3231 * Dynamic irq allocate and deallocation
3233 unsigned int create_irq_nr(unsigned int irq_want, int node)
3235 /* Allocate an unused irq */
3238 unsigned long flags;
3239 struct irq_cfg *cfg_new = NULL;
3240 struct irq_desc *desc_new = NULL;
3243 if (irq_want < nr_irqs_gsi)
3244 irq_want = nr_irqs_gsi;
3246 raw_spin_lock_irqsave(&vector_lock, flags);
3247 for (new = irq_want; new < nr_irqs; new++) {
3248 desc_new = irq_to_desc_alloc_node(new, node);
3250 printk(KERN_INFO "can not get irq_desc for %d\n", new);
3253 cfg_new = desc_new->chip_data;
3255 if (cfg_new->vector != 0)
3258 desc_new = move_irq_desc(desc_new, node);
3259 cfg_new = desc_new->chip_data;
3261 if (__assign_irq_vector(new, cfg_new, apic->target_cpus()) == 0)
3265 raw_spin_unlock_irqrestore(&vector_lock, flags);
3268 dynamic_irq_init_keep_chip_data(irq);
3273 int create_irq(void)
3275 int node = cpu_to_node(0);
3276 unsigned int irq_want;
3279 irq_want = nr_irqs_gsi;
3280 irq = create_irq_nr(irq_want, node);
3288 void destroy_irq(unsigned int irq)
3290 unsigned long flags;
3292 dynamic_irq_cleanup_keep_chip_data(irq);
3295 raw_spin_lock_irqsave(&vector_lock, flags);
3296 __clear_irq_vector(irq, get_irq_chip_data(irq));
3297 raw_spin_unlock_irqrestore(&vector_lock, flags);
3301 * MSI message composition
3303 #ifdef CONFIG_PCI_MSI
3304 static int msi_compose_msg(struct pci_dev *pdev, unsigned int irq,
3305 struct msi_msg *msg, u8 hpet_id)
3307 struct irq_cfg *cfg;
3315 err = assign_irq_vector(irq, cfg, apic->target_cpus());
3319 dest = apic->cpu_mask_to_apicid_and(cfg->domain, apic->target_cpus());
3321 if (irq_remapped(irq)) {
3326 ir_index = map_irq_to_irte_handle(irq, &sub_handle);
3327 BUG_ON(ir_index == -1);
3329 prepare_irte(&irte, cfg->vector, dest);
3331 /* Set source-id of interrupt request */
3333 set_msi_sid(&irte, pdev);
3335 set_hpet_sid(&irte, hpet_id);
3337 modify_irte(irq, &irte);
3339 msg->address_hi = MSI_ADDR_BASE_HI;
3340 msg->data = sub_handle;
3341 msg->address_lo = MSI_ADDR_BASE_LO | MSI_ADDR_IR_EXT_INT |
3343 MSI_ADDR_IR_INDEX1(ir_index) |
3344 MSI_ADDR_IR_INDEX2(ir_index);
3346 if (x2apic_enabled())
3347 msg->address_hi = MSI_ADDR_BASE_HI |
3348 MSI_ADDR_EXT_DEST_ID(dest);
3350 msg->address_hi = MSI_ADDR_BASE_HI;
3354 ((apic->irq_dest_mode == 0) ?
3355 MSI_ADDR_DEST_MODE_PHYSICAL:
3356 MSI_ADDR_DEST_MODE_LOGICAL) |
3357 ((apic->irq_delivery_mode != dest_LowestPrio) ?
3358 MSI_ADDR_REDIRECTION_CPU:
3359 MSI_ADDR_REDIRECTION_LOWPRI) |
3360 MSI_ADDR_DEST_ID(dest);
3363 MSI_DATA_TRIGGER_EDGE |
3364 MSI_DATA_LEVEL_ASSERT |
3365 ((apic->irq_delivery_mode != dest_LowestPrio) ?
3366 MSI_DATA_DELIVERY_FIXED:
3367 MSI_DATA_DELIVERY_LOWPRI) |
3368 MSI_DATA_VECTOR(cfg->vector);
3374 static int set_msi_irq_affinity(unsigned int irq, const struct cpumask *mask)
3376 struct irq_desc *desc = irq_to_desc(irq);
3377 struct irq_cfg *cfg;
3381 if (set_desc_affinity(desc, mask, &dest))
3384 cfg = desc->chip_data;
3386 get_cached_msi_msg_desc(desc, &msg);
3388 msg.data &= ~MSI_DATA_VECTOR_MASK;
3389 msg.data |= MSI_DATA_VECTOR(cfg->vector);
3390 msg.address_lo &= ~MSI_ADDR_DEST_ID_MASK;
3391 msg.address_lo |= MSI_ADDR_DEST_ID(dest);
3393 write_msi_msg_desc(desc, &msg);
3397 #ifdef CONFIG_INTR_REMAP
3399 * Migrate the MSI irq to another cpumask. This migration is
3400 * done in the process context using interrupt-remapping hardware.
3403 ir_set_msi_irq_affinity(unsigned int irq, const struct cpumask *mask)
3405 struct irq_desc *desc = irq_to_desc(irq);
3406 struct irq_cfg *cfg = desc->chip_data;
3410 if (get_irte(irq, &irte))
3413 if (set_desc_affinity(desc, mask, &dest))
3416 irte.vector = cfg->vector;
3417 irte.dest_id = IRTE_DEST(dest);
3420 * atomically update the IRTE with the new destination and vector.
3422 modify_irte(irq, &irte);
3425 * After this point, all the interrupts will start arriving
3426 * at the new destination. So, time to cleanup the previous
3427 * vector allocation.
3429 if (cfg->move_in_progress)
3430 send_cleanup_vector(cfg);
3436 #endif /* CONFIG_SMP */
3439 * IRQ Chip for MSI PCI/PCI-X/PCI-Express Devices,
3440 * which implement the MSI or MSI-X Capability Structure.
3442 static struct irq_chip msi_chip = {
3444 .unmask = unmask_msi_irq,
3445 .mask = mask_msi_irq,
3446 .ack = ack_apic_edge,
3448 .set_affinity = set_msi_irq_affinity,
3450 .retrigger = ioapic_retrigger_irq,
3453 static struct irq_chip msi_ir_chip = {
3454 .name = "IR-PCI-MSI",
3455 .unmask = unmask_msi_irq,
3456 .mask = mask_msi_irq,
3457 #ifdef CONFIG_INTR_REMAP
3458 .ack = ir_ack_apic_edge,
3460 .set_affinity = ir_set_msi_irq_affinity,
3463 .retrigger = ioapic_retrigger_irq,
3467 * Map the PCI dev to the corresponding remapping hardware unit
3468 * and allocate 'nvec' consecutive interrupt-remapping table entries
3471 static int msi_alloc_irte(struct pci_dev *dev, int irq, int nvec)
3473 struct intel_iommu *iommu;
3476 iommu = map_dev_to_ir(dev);
3479 "Unable to map PCI %s to iommu\n", pci_name(dev));
3483 index = alloc_irte(iommu, irq, nvec);
3486 "Unable to allocate %d IRTE for PCI %s\n", nvec,
3493 static int setup_msi_irq(struct pci_dev *dev, struct msi_desc *msidesc, int irq)
3498 ret = msi_compose_msg(dev, irq, &msg, -1);
3502 set_irq_msi(irq, msidesc);
3503 write_msi_msg(irq, &msg);
3505 if (irq_remapped(irq)) {
3506 struct irq_desc *desc = irq_to_desc(irq);
3508 * irq migration in process context
3510 desc->status |= IRQ_MOVE_PCNTXT;
3511 set_irq_chip_and_handler_name(irq, &msi_ir_chip, handle_edge_irq, "edge");
3513 set_irq_chip_and_handler_name(irq, &msi_chip, handle_edge_irq, "edge");
3515 dev_printk(KERN_DEBUG, &dev->dev, "irq %d for MSI/MSI-X\n", irq);
3520 int arch_setup_msi_irqs(struct pci_dev *dev, int nvec, int type)
3523 int ret, sub_handle;
3524 struct msi_desc *msidesc;
3525 unsigned int irq_want;
3526 struct intel_iommu *iommu = NULL;
3530 /* x86 doesn't support multiple MSI yet */
3531 if (type == PCI_CAP_ID_MSI && nvec > 1)
3534 node = dev_to_node(&dev->dev);
3535 irq_want = nr_irqs_gsi;
3537 list_for_each_entry(msidesc, &dev->msi_list, list) {
3538 irq = create_irq_nr(irq_want, node);
3542 if (!intr_remapping_enabled)
3547 * allocate the consecutive block of IRTE's
3550 index = msi_alloc_irte(dev, irq, nvec);
3556 iommu = map_dev_to_ir(dev);
3562 * setup the mapping between the irq and the IRTE
3563 * base index, the sub_handle pointing to the
3564 * appropriate interrupt remap table entry.
3566 set_irte_irq(irq, iommu, index, sub_handle);
3569 ret = setup_msi_irq(dev, msidesc, irq);
3581 void arch_teardown_msi_irq(unsigned int irq)
3586 #if defined (CONFIG_DMAR) || defined (CONFIG_INTR_REMAP)
3588 static int dmar_msi_set_affinity(unsigned int irq, const struct cpumask *mask)
3590 struct irq_desc *desc = irq_to_desc(irq);
3591 struct irq_cfg *cfg;
3595 if (set_desc_affinity(desc, mask, &dest))
3598 cfg = desc->chip_data;
3600 dmar_msi_read(irq, &msg);
3602 msg.data &= ~MSI_DATA_VECTOR_MASK;
3603 msg.data |= MSI_DATA_VECTOR(cfg->vector);
3604 msg.address_lo &= ~MSI_ADDR_DEST_ID_MASK;
3605 msg.address_lo |= MSI_ADDR_DEST_ID(dest);
3607 dmar_msi_write(irq, &msg);
3612 #endif /* CONFIG_SMP */
3614 static struct irq_chip dmar_msi_type = {
3616 .unmask = dmar_msi_unmask,
3617 .mask = dmar_msi_mask,
3618 .ack = ack_apic_edge,
3620 .set_affinity = dmar_msi_set_affinity,
3622 .retrigger = ioapic_retrigger_irq,
3625 int arch_setup_dmar_msi(unsigned int irq)
3630 ret = msi_compose_msg(NULL, irq, &msg, -1);
3633 dmar_msi_write(irq, &msg);
3634 set_irq_chip_and_handler_name(irq, &dmar_msi_type, handle_edge_irq,
3640 #ifdef CONFIG_HPET_TIMER
3643 static int hpet_msi_set_affinity(unsigned int irq, const struct cpumask *mask)
3645 struct irq_desc *desc = irq_to_desc(irq);
3646 struct irq_cfg *cfg;
3650 if (set_desc_affinity(desc, mask, &dest))
3653 cfg = desc->chip_data;
3655 hpet_msi_read(irq, &msg);
3657 msg.data &= ~MSI_DATA_VECTOR_MASK;
3658 msg.data |= MSI_DATA_VECTOR(cfg->vector);
3659 msg.address_lo &= ~MSI_ADDR_DEST_ID_MASK;
3660 msg.address_lo |= MSI_ADDR_DEST_ID(dest);
3662 hpet_msi_write(irq, &msg);
3667 #endif /* CONFIG_SMP */
3669 static struct irq_chip ir_hpet_msi_type = {
3670 .name = "IR-HPET_MSI",
3671 .unmask = hpet_msi_unmask,
3672 .mask = hpet_msi_mask,
3673 #ifdef CONFIG_INTR_REMAP
3674 .ack = ir_ack_apic_edge,
3676 .set_affinity = ir_set_msi_irq_affinity,
3679 .retrigger = ioapic_retrigger_irq,
3682 static struct irq_chip hpet_msi_type = {
3684 .unmask = hpet_msi_unmask,
3685 .mask = hpet_msi_mask,
3686 .ack = ack_apic_edge,
3688 .set_affinity = hpet_msi_set_affinity,
3690 .retrigger = ioapic_retrigger_irq,
3693 int arch_setup_hpet_msi(unsigned int irq, unsigned int id)
3697 struct irq_desc *desc = irq_to_desc(irq);
3699 if (intr_remapping_enabled) {
3700 struct intel_iommu *iommu = map_hpet_to_ir(id);
3706 index = alloc_irte(iommu, irq, 1);
3711 ret = msi_compose_msg(NULL, irq, &msg, id);
3715 hpet_msi_write(irq, &msg);
3716 desc->status |= IRQ_MOVE_PCNTXT;
3717 if (irq_remapped(irq))
3718 set_irq_chip_and_handler_name(irq, &ir_hpet_msi_type,
3719 handle_edge_irq, "edge");
3721 set_irq_chip_and_handler_name(irq, &hpet_msi_type,
3722 handle_edge_irq, "edge");
3728 #endif /* CONFIG_PCI_MSI */
3730 * Hypertransport interrupt support
3732 #ifdef CONFIG_HT_IRQ
3736 static void target_ht_irq(unsigned int irq, unsigned int dest, u8 vector)
3738 struct ht_irq_msg msg;
3739 fetch_ht_irq_msg(irq, &msg);
3741 msg.address_lo &= ~(HT_IRQ_LOW_VECTOR_MASK | HT_IRQ_LOW_DEST_ID_MASK);
3742 msg.address_hi &= ~(HT_IRQ_HIGH_DEST_ID_MASK);
3744 msg.address_lo |= HT_IRQ_LOW_VECTOR(vector) | HT_IRQ_LOW_DEST_ID(dest);
3745 msg.address_hi |= HT_IRQ_HIGH_DEST_ID(dest);
3747 write_ht_irq_msg(irq, &msg);
3750 static int set_ht_irq_affinity(unsigned int irq, const struct cpumask *mask)
3752 struct irq_desc *desc = irq_to_desc(irq);
3753 struct irq_cfg *cfg;
3756 if (set_desc_affinity(desc, mask, &dest))
3759 cfg = desc->chip_data;
3761 target_ht_irq(irq, dest, cfg->vector);
3768 static struct irq_chip ht_irq_chip = {
3770 .mask = mask_ht_irq,
3771 .unmask = unmask_ht_irq,
3772 .ack = ack_apic_edge,
3774 .set_affinity = set_ht_irq_affinity,
3776 .retrigger = ioapic_retrigger_irq,
3779 int arch_setup_ht_irq(unsigned int irq, struct pci_dev *dev)
3781 struct irq_cfg *cfg;
3788 err = assign_irq_vector(irq, cfg, apic->target_cpus());
3790 struct ht_irq_msg msg;
3793 dest = apic->cpu_mask_to_apicid_and(cfg->domain,
3794 apic->target_cpus());
3796 msg.address_hi = HT_IRQ_HIGH_DEST_ID(dest);
3800 HT_IRQ_LOW_DEST_ID(dest) |
3801 HT_IRQ_LOW_VECTOR(cfg->vector) |
3802 ((apic->irq_dest_mode == 0) ?
3803 HT_IRQ_LOW_DM_PHYSICAL :
3804 HT_IRQ_LOW_DM_LOGICAL) |
3805 HT_IRQ_LOW_RQEOI_EDGE |
3806 ((apic->irq_delivery_mode != dest_LowestPrio) ?
3807 HT_IRQ_LOW_MT_FIXED :
3808 HT_IRQ_LOW_MT_ARBITRATED) |
3809 HT_IRQ_LOW_IRQ_MASKED;
3811 write_ht_irq_msg(irq, &msg);
3813 set_irq_chip_and_handler_name(irq, &ht_irq_chip,
3814 handle_edge_irq, "edge");
3816 dev_printk(KERN_DEBUG, &dev->dev, "irq %d for HT\n", irq);
3820 #endif /* CONFIG_HT_IRQ */
3822 int __init io_apic_get_redir_entries (int ioapic)
3824 union IO_APIC_reg_01 reg_01;
3825 unsigned long flags;
3827 raw_spin_lock_irqsave(&ioapic_lock, flags);
3828 reg_01.raw = io_apic_read(ioapic, 1);
3829 raw_spin_unlock_irqrestore(&ioapic_lock, flags);
3831 /* The register returns the maximum index redir index
3832 * supported, which is one less than the total number of redir
3835 return reg_01.bits.entries + 1;
3838 void __init probe_nr_irqs_gsi(void)
3842 nr = gsi_top + NR_IRQS_LEGACY;
3843 if (nr > nr_irqs_gsi)
3846 printk(KERN_DEBUG "nr_irqs_gsi: %d\n", nr_irqs_gsi);
3849 #ifdef CONFIG_SPARSE_IRQ
3850 int __init arch_probe_nr_irqs(void)
3854 if (nr_irqs > (NR_VECTORS * nr_cpu_ids))
3855 nr_irqs = NR_VECTORS * nr_cpu_ids;
3857 nr = nr_irqs_gsi + 8 * nr_cpu_ids;
3858 #if defined(CONFIG_PCI_MSI) || defined(CONFIG_HT_IRQ)
3860 * for MSI and HT dyn irq
3862 nr += nr_irqs_gsi * 16;
3871 static int __io_apic_set_pci_routing(struct device *dev, int irq,
3872 struct io_apic_irq_attr *irq_attr)
3874 struct irq_desc *desc;
3875 struct irq_cfg *cfg;
3878 int trigger, polarity;
3880 ioapic = irq_attr->ioapic;
3881 if (!IO_APIC_IRQ(irq)) {
3882 apic_printk(APIC_QUIET,KERN_ERR "IOAPIC[%d]: Invalid reference to IRQ 0\n",
3888 node = dev_to_node(dev);
3890 node = cpu_to_node(0);
3892 desc = irq_to_desc_alloc_node(irq, node);
3894 printk(KERN_INFO "can not get irq_desc %d\n", irq);
3898 pin = irq_attr->ioapic_pin;
3899 trigger = irq_attr->trigger;
3900 polarity = irq_attr->polarity;
3903 * IRQs < 16 are already in the irq_2_pin[] map
3905 if (irq >= legacy_pic->nr_legacy_irqs) {
3906 cfg = desc->chip_data;
3907 if (add_pin_to_irq_node_nopanic(cfg, node, ioapic, pin)) {
3908 printk(KERN_INFO "can not add pin %d for irq %d\n",
3914 setup_IO_APIC_irq(ioapic, pin, irq, desc, trigger, polarity);
3919 int io_apic_set_pci_routing(struct device *dev, int irq,
3920 struct io_apic_irq_attr *irq_attr)
3924 * Avoid pin reprogramming. PRTs typically include entries
3925 * with redundant pin->gsi mappings (but unique PCI devices);
3926 * we only program the IOAPIC on the first.
3928 ioapic = irq_attr->ioapic;
3929 pin = irq_attr->ioapic_pin;
3930 if (test_bit(pin, mp_ioapic_routing[ioapic].pin_programmed)) {
3931 pr_debug("Pin %d-%d already programmed\n",
3932 mp_ioapics[ioapic].apicid, pin);
3935 set_bit(pin, mp_ioapic_routing[ioapic].pin_programmed);
3937 return __io_apic_set_pci_routing(dev, irq, irq_attr);
3940 u8 __init io_apic_unique_id(u8 id)
3942 #ifdef CONFIG_X86_32
3943 if ((boot_cpu_data.x86_vendor == X86_VENDOR_INTEL) &&
3944 !APIC_XAPIC(apic_version[boot_cpu_physical_apicid]))
3945 return io_apic_get_unique_id(nr_ioapics, id);
3950 DECLARE_BITMAP(used, 256);
3952 bitmap_zero(used, 256);
3953 for (i = 0; i < nr_ioapics; i++) {
3954 struct mpc_ioapic *ia = &mp_ioapics[i];
3955 __set_bit(ia->apicid, used);
3957 if (!test_bit(id, used))
3959 return find_first_zero_bit(used, 256);
3963 #ifdef CONFIG_X86_32
3964 int __init io_apic_get_unique_id(int ioapic, int apic_id)
3966 union IO_APIC_reg_00 reg_00;
3967 static physid_mask_t apic_id_map = PHYSID_MASK_NONE;
3969 unsigned long flags;
3973 * The P4 platform supports up to 256 APIC IDs on two separate APIC
3974 * buses (one for LAPICs, one for IOAPICs), where predecessors only
3975 * supports up to 16 on one shared APIC bus.
3977 * TBD: Expand LAPIC/IOAPIC support on P4-class systems to take full
3978 * advantage of new APIC bus architecture.
3981 if (physids_empty(apic_id_map))
3982 apic->ioapic_phys_id_map(&phys_cpu_present_map, &apic_id_map);
3984 raw_spin_lock_irqsave(&ioapic_lock, flags);
3985 reg_00.raw = io_apic_read(ioapic, 0);
3986 raw_spin_unlock_irqrestore(&ioapic_lock, flags);
3988 if (apic_id >= get_physical_broadcast()) {
3989 printk(KERN_WARNING "IOAPIC[%d]: Invalid apic_id %d, trying "
3990 "%d\n", ioapic, apic_id, reg_00.bits.ID);
3991 apic_id = reg_00.bits.ID;
3995 * Every APIC in a system must have a unique ID or we get lots of nice
3996 * 'stuck on smp_invalidate_needed IPI wait' messages.
3998 if (apic->check_apicid_used(&apic_id_map, apic_id)) {
4000 for (i = 0; i < get_physical_broadcast(); i++) {
4001 if (!apic->check_apicid_used(&apic_id_map, i))
4005 if (i == get_physical_broadcast())
4006 panic("Max apic_id exceeded!\n");
4008 printk(KERN_WARNING "IOAPIC[%d]: apic_id %d already used, "
4009 "trying %d\n", ioapic, apic_id, i);
4014 apic->apicid_to_cpu_present(apic_id, &tmp);
4015 physids_or(apic_id_map, apic_id_map, tmp);
4017 if (reg_00.bits.ID != apic_id) {
4018 reg_00.bits.ID = apic_id;
4020 raw_spin_lock_irqsave(&ioapic_lock, flags);
4021 io_apic_write(ioapic, 0, reg_00.raw);
4022 reg_00.raw = io_apic_read(ioapic, 0);
4023 raw_spin_unlock_irqrestore(&ioapic_lock, flags);
4026 if (reg_00.bits.ID != apic_id) {
4027 printk("IOAPIC[%d]: Unable to change apic_id!\n", ioapic);
4032 apic_printk(APIC_VERBOSE, KERN_INFO
4033 "IOAPIC[%d]: Assigned apic_id %d\n", ioapic, apic_id);
4039 int __init io_apic_get_version(int ioapic)
4041 union IO_APIC_reg_01 reg_01;
4042 unsigned long flags;
4044 raw_spin_lock_irqsave(&ioapic_lock, flags);
4045 reg_01.raw = io_apic_read(ioapic, 1);
4046 raw_spin_unlock_irqrestore(&ioapic_lock, flags);
4048 return reg_01.bits.version;
4051 int acpi_get_override_irq(u32 gsi, int *trigger, int *polarity)
4053 int ioapic, pin, idx;
4055 if (skip_ioapic_setup)
4058 ioapic = mp_find_ioapic(gsi);
4062 pin = mp_find_ioapic_pin(ioapic, gsi);
4066 idx = find_irq_entry(ioapic, pin, mp_INT);
4070 *trigger = irq_trigger(idx);
4071 *polarity = irq_polarity(idx);
4076 * This function currently is only a helper for the i386 smp boot process where
4077 * we need to reprogram the ioredtbls to cater for the cpus which have come online
4078 * so mask in all cases should simply be apic->target_cpus()
4081 void __init setup_ioapic_dest(void)
4083 int pin, ioapic, irq, irq_entry;
4084 struct irq_desc *desc;
4085 const struct cpumask *mask;
4087 if (skip_ioapic_setup == 1)
4090 for (ioapic = 0; ioapic < nr_ioapics; ioapic++)
4091 for (pin = 0; pin < nr_ioapic_registers[ioapic]; pin++) {
4092 irq_entry = find_irq_entry(ioapic, pin, mp_INT);
4093 if (irq_entry == -1)
4095 irq = pin_2_irq(irq_entry, ioapic, pin);
4097 if ((ioapic > 0) && (irq > 16))
4100 desc = irq_to_desc(irq);
4103 * Honour affinities which have been set in early boot
4106 (IRQ_NO_BALANCING | IRQ_AFFINITY_SET))
4107 mask = desc->affinity;
4109 mask = apic->target_cpus();
4111 if (intr_remapping_enabled)
4112 set_ir_ioapic_affinity_irq_desc(desc, mask);
4114 set_ioapic_affinity_irq_desc(desc, mask);
4120 #define IOAPIC_RESOURCE_NAME_SIZE 11
4122 static struct resource *ioapic_resources;
4124 static struct resource * __init ioapic_setup_resources(int nr_ioapics)
4127 struct resource *res;
4131 if (nr_ioapics <= 0)
4134 n = IOAPIC_RESOURCE_NAME_SIZE + sizeof(struct resource);
4137 mem = alloc_bootmem(n);
4140 mem += sizeof(struct resource) * nr_ioapics;
4142 for (i = 0; i < nr_ioapics; i++) {
4144 res[i].flags = IORESOURCE_MEM | IORESOURCE_BUSY;
4145 snprintf(mem, IOAPIC_RESOURCE_NAME_SIZE, "IOAPIC %u", i);
4146 mem += IOAPIC_RESOURCE_NAME_SIZE;
4149 ioapic_resources = res;
4154 void __init ioapic_init_mappings(void)
4156 unsigned long ioapic_phys, idx = FIX_IO_APIC_BASE_0;
4157 struct resource *ioapic_res;
4160 ioapic_res = ioapic_setup_resources(nr_ioapics);
4161 for (i = 0; i < nr_ioapics; i++) {
4162 if (smp_found_config) {
4163 ioapic_phys = mp_ioapics[i].apicaddr;
4164 #ifdef CONFIG_X86_32
4167 "WARNING: bogus zero IO-APIC "
4168 "address found in MPTABLE, "
4169 "disabling IO/APIC support!\n");
4170 smp_found_config = 0;
4171 skip_ioapic_setup = 1;
4172 goto fake_ioapic_page;
4176 #ifdef CONFIG_X86_32
4179 ioapic_phys = (unsigned long)alloc_bootmem_pages(PAGE_SIZE);
4180 ioapic_phys = __pa(ioapic_phys);
4182 set_fixmap_nocache(idx, ioapic_phys);
4183 apic_printk(APIC_VERBOSE, "mapped IOAPIC to %08lx (%08lx)\n",
4184 __fix_to_virt(idx) + (ioapic_phys & ~PAGE_MASK),
4188 ioapic_res->start = ioapic_phys;
4189 ioapic_res->end = ioapic_phys + IO_APIC_SLOT_SIZE - 1;
4194 void __init ioapic_insert_resources(void)
4197 struct resource *r = ioapic_resources;
4202 "IO APIC resources couldn't be allocated.\n");
4206 for (i = 0; i < nr_ioapics; i++) {
4207 insert_resource(&iomem_resource, r);
4212 int mp_find_ioapic(u32 gsi)
4216 /* Find the IOAPIC that manages this GSI. */
4217 for (i = 0; i < nr_ioapics; i++) {
4218 if ((gsi >= mp_gsi_routing[i].gsi_base)
4219 && (gsi <= mp_gsi_routing[i].gsi_end))
4223 printk(KERN_ERR "ERROR: Unable to locate IOAPIC for GSI %d\n", gsi);
4227 int mp_find_ioapic_pin(int ioapic, u32 gsi)
4229 if (WARN_ON(ioapic == -1))
4231 if (WARN_ON(gsi > mp_gsi_routing[ioapic].gsi_end))
4234 return gsi - mp_gsi_routing[ioapic].gsi_base;
4237 static int bad_ioapic(unsigned long address)
4239 if (nr_ioapics >= MAX_IO_APICS) {
4240 printk(KERN_WARNING "WARING: Max # of I/O APICs (%d) exceeded "
4241 "(found %d), skipping\n", MAX_IO_APICS, nr_ioapics);
4245 printk(KERN_WARNING "WARNING: Bogus (zero) I/O APIC address"
4246 " found in table, skipping!\n");
4252 void __init mp_register_ioapic(int id, u32 address, u32 gsi_base)
4257 if (bad_ioapic(address))
4262 mp_ioapics[idx].type = MP_IOAPIC;
4263 mp_ioapics[idx].flags = MPC_APIC_USABLE;
4264 mp_ioapics[idx].apicaddr = address;
4266 set_fixmap_nocache(FIX_IO_APIC_BASE_0 + idx, address);
4267 mp_ioapics[idx].apicid = io_apic_unique_id(id);
4268 mp_ioapics[idx].apicver = io_apic_get_version(idx);
4271 * Build basic GSI lookup table to facilitate gsi->io_apic lookups
4272 * and to prevent reprogramming of IOAPIC pins (PCI GSIs).
4274 entries = io_apic_get_redir_entries(idx);
4275 mp_gsi_routing[idx].gsi_base = gsi_base;
4276 mp_gsi_routing[idx].gsi_end = gsi_base + entries - 1;
4279 * The number of IO-APIC IRQ registers (== #pins):
4281 nr_ioapic_registers[idx] = entries;
4283 if (mp_gsi_routing[idx].gsi_end >= gsi_top)
4284 gsi_top = mp_gsi_routing[idx].gsi_end + 1;
4286 printk(KERN_INFO "IOAPIC[%d]: apic_id %d, version %d, address 0x%x, "
4287 "GSI %d-%d\n", idx, mp_ioapics[idx].apicid,
4288 mp_ioapics[idx].apicver, mp_ioapics[idx].apicaddr,
4289 mp_gsi_routing[idx].gsi_base, mp_gsi_routing[idx].gsi_end);
4294 /* Enable IOAPIC early just for system timer */
4295 void __init pre_init_apic_IRQ0(void)
4297 struct irq_cfg *cfg;
4298 struct irq_desc *desc;
4300 printk(KERN_INFO "Early APIC setup for system timer0\n");
4302 phys_cpu_present_map = physid_mask_of_physid(boot_cpu_physical_apicid);
4304 desc = irq_to_desc_alloc_node(0, 0);
4309 add_pin_to_irq_node(cfg, 0, 0, 0);
4310 set_irq_chip_and_handler_name(0, &ioapic_chip, handle_edge_irq, "edge");
4312 setup_IO_APIC_irq(0, 0, 0, desc, 0, 0);