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 last 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(boot_cpu_id);
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 memcpy(cfg, old_cfg, sizeof(struct irq_cfg));
311 init_copy_irq_2_pin(old_cfg, cfg, node);
314 static void free_irq_cfg(struct irq_cfg *old_cfg)
319 void arch_free_chip_data(struct irq_desc *old_desc, struct irq_desc *desc)
321 struct irq_cfg *old_cfg, *cfg;
323 old_cfg = old_desc->chip_data;
324 cfg = desc->chip_data;
330 free_irq_2_pin(old_cfg, cfg);
331 free_irq_cfg(old_cfg);
332 old_desc->chip_data = NULL;
335 /* end for move_irq_desc */
338 struct irq_cfg *irq_cfg(unsigned int irq)
340 return irq < nr_irqs ? irq_cfgx + irq : NULL;
347 unsigned int unused[3];
349 unsigned int unused2[11];
353 static __attribute_const__ struct io_apic __iomem *io_apic_base(int idx)
355 return (void __iomem *) __fix_to_virt(FIX_IO_APIC_BASE_0 + idx)
356 + (mp_ioapics[idx].apicaddr & ~PAGE_MASK);
359 static inline void io_apic_eoi(unsigned int apic, unsigned int vector)
361 struct io_apic __iomem *io_apic = io_apic_base(apic);
362 writel(vector, &io_apic->eoi);
365 static inline unsigned int io_apic_read(unsigned int apic, unsigned int reg)
367 struct io_apic __iomem *io_apic = io_apic_base(apic);
368 writel(reg, &io_apic->index);
369 return readl(&io_apic->data);
372 static inline void io_apic_write(unsigned int apic, unsigned int reg, unsigned int value)
374 struct io_apic __iomem *io_apic = io_apic_base(apic);
375 writel(reg, &io_apic->index);
376 writel(value, &io_apic->data);
380 * Re-write a value: to be used for read-modify-write
381 * cycles where the read already set up the index register.
383 * Older SiS APIC requires we rewrite the index register
385 static inline void io_apic_modify(unsigned int apic, unsigned int reg, unsigned int value)
387 struct io_apic __iomem *io_apic = io_apic_base(apic);
390 writel(reg, &io_apic->index);
391 writel(value, &io_apic->data);
394 static bool io_apic_level_ack_pending(struct irq_cfg *cfg)
396 struct irq_pin_list *entry;
399 raw_spin_lock_irqsave(&ioapic_lock, flags);
400 for_each_irq_pin(entry, cfg->irq_2_pin) {
405 reg = io_apic_read(entry->apic, 0x10 + pin*2);
406 /* Is the remote IRR bit set? */
407 if (reg & IO_APIC_REDIR_REMOTE_IRR) {
408 raw_spin_unlock_irqrestore(&ioapic_lock, flags);
412 raw_spin_unlock_irqrestore(&ioapic_lock, flags);
418 struct { u32 w1, w2; };
419 struct IO_APIC_route_entry entry;
422 static struct IO_APIC_route_entry ioapic_read_entry(int apic, int pin)
424 union entry_union eu;
426 raw_spin_lock_irqsave(&ioapic_lock, flags);
427 eu.w1 = io_apic_read(apic, 0x10 + 2 * pin);
428 eu.w2 = io_apic_read(apic, 0x11 + 2 * pin);
429 raw_spin_unlock_irqrestore(&ioapic_lock, flags);
434 * When we write a new IO APIC routing entry, we need to write the high
435 * word first! If the mask bit in the low word is clear, we will enable
436 * the interrupt, and we need to make sure the entry is fully populated
437 * before that happens.
440 __ioapic_write_entry(int apic, int pin, struct IO_APIC_route_entry e)
442 union entry_union eu = {{0, 0}};
445 io_apic_write(apic, 0x11 + 2*pin, eu.w2);
446 io_apic_write(apic, 0x10 + 2*pin, eu.w1);
449 void ioapic_write_entry(int apic, int pin, struct IO_APIC_route_entry e)
452 raw_spin_lock_irqsave(&ioapic_lock, flags);
453 __ioapic_write_entry(apic, pin, e);
454 raw_spin_unlock_irqrestore(&ioapic_lock, flags);
458 * When we mask an IO APIC routing entry, we need to write the low
459 * word first, in order to set the mask bit before we change the
462 static void ioapic_mask_entry(int apic, int pin)
465 union entry_union eu = { .entry.mask = 1 };
467 raw_spin_lock_irqsave(&ioapic_lock, flags);
468 io_apic_write(apic, 0x10 + 2*pin, eu.w1);
469 io_apic_write(apic, 0x11 + 2*pin, eu.w2);
470 raw_spin_unlock_irqrestore(&ioapic_lock, flags);
474 * The common case is 1:1 IRQ<->pin mappings. Sometimes there are
475 * shared ISA-space IRQs, so we have to support them. We are super
476 * fast in the common case, and fast for shared ISA-space IRQs.
479 add_pin_to_irq_node_nopanic(struct irq_cfg *cfg, int node, int apic, int pin)
481 struct irq_pin_list **last, *entry;
483 /* don't allow duplicates */
484 last = &cfg->irq_2_pin;
485 for_each_irq_pin(entry, cfg->irq_2_pin) {
486 if (entry->apic == apic && entry->pin == pin)
491 entry = get_one_free_irq_2_pin(node);
493 printk(KERN_ERR "can not alloc irq_pin_list (%d,%d,%d)\n",
504 static void add_pin_to_irq_node(struct irq_cfg *cfg, int node, int apic, int pin)
506 if (add_pin_to_irq_node_nopanic(cfg, node, apic, pin))
507 panic("IO-APIC: failed to add irq-pin. Can not proceed\n");
511 * Reroute an IRQ to a different pin.
513 static void __init replace_pin_at_irq_node(struct irq_cfg *cfg, int node,
514 int oldapic, int oldpin,
515 int newapic, int newpin)
517 struct irq_pin_list *entry;
519 for_each_irq_pin(entry, cfg->irq_2_pin) {
520 if (entry->apic == oldapic && entry->pin == oldpin) {
521 entry->apic = newapic;
523 /* every one is different, right? */
528 /* old apic/pin didn't exist, so just add new ones */
529 add_pin_to_irq_node(cfg, node, newapic, newpin);
532 static void __io_apic_modify_irq(struct irq_pin_list *entry,
533 int mask_and, int mask_or,
534 void (*final)(struct irq_pin_list *entry))
536 unsigned int reg, pin;
539 reg = io_apic_read(entry->apic, 0x10 + pin * 2);
542 io_apic_modify(entry->apic, 0x10 + pin * 2, reg);
547 static void io_apic_modify_irq(struct irq_cfg *cfg,
548 int mask_and, int mask_or,
549 void (*final)(struct irq_pin_list *entry))
551 struct irq_pin_list *entry;
553 for_each_irq_pin(entry, cfg->irq_2_pin)
554 __io_apic_modify_irq(entry, mask_and, mask_or, final);
557 static void __mask_and_edge_IO_APIC_irq(struct irq_pin_list *entry)
559 __io_apic_modify_irq(entry, ~IO_APIC_REDIR_LEVEL_TRIGGER,
560 IO_APIC_REDIR_MASKED, NULL);
563 static void __unmask_and_level_IO_APIC_irq(struct irq_pin_list *entry)
565 __io_apic_modify_irq(entry, ~IO_APIC_REDIR_MASKED,
566 IO_APIC_REDIR_LEVEL_TRIGGER, NULL);
569 static void __unmask_IO_APIC_irq(struct irq_cfg *cfg)
571 io_apic_modify_irq(cfg, ~IO_APIC_REDIR_MASKED, 0, NULL);
574 static void io_apic_sync(struct irq_pin_list *entry)
577 * Synchronize the IO-APIC and the CPU by doing
578 * a dummy read from the IO-APIC
580 struct io_apic __iomem *io_apic;
581 io_apic = io_apic_base(entry->apic);
582 readl(&io_apic->data);
585 static void __mask_IO_APIC_irq(struct irq_cfg *cfg)
587 io_apic_modify_irq(cfg, ~0, IO_APIC_REDIR_MASKED, &io_apic_sync);
590 static void mask_IO_APIC_irq_desc(struct irq_desc *desc)
592 struct irq_cfg *cfg = desc->chip_data;
597 raw_spin_lock_irqsave(&ioapic_lock, flags);
598 __mask_IO_APIC_irq(cfg);
599 raw_spin_unlock_irqrestore(&ioapic_lock, flags);
602 static void unmask_IO_APIC_irq_desc(struct irq_desc *desc)
604 struct irq_cfg *cfg = desc->chip_data;
607 raw_spin_lock_irqsave(&ioapic_lock, flags);
608 __unmask_IO_APIC_irq(cfg);
609 raw_spin_unlock_irqrestore(&ioapic_lock, flags);
612 static void mask_IO_APIC_irq(unsigned int irq)
614 struct irq_desc *desc = irq_to_desc(irq);
616 mask_IO_APIC_irq_desc(desc);
618 static void unmask_IO_APIC_irq(unsigned int irq)
620 struct irq_desc *desc = irq_to_desc(irq);
622 unmask_IO_APIC_irq_desc(desc);
625 static void clear_IO_APIC_pin(unsigned int apic, unsigned int pin)
627 struct IO_APIC_route_entry entry;
629 /* Check delivery_mode to be sure we're not clearing an SMI pin */
630 entry = ioapic_read_entry(apic, pin);
631 if (entry.delivery_mode == dest_SMI)
634 * Disable it in the IO-APIC irq-routing table:
636 ioapic_mask_entry(apic, pin);
639 static void clear_IO_APIC (void)
643 for (apic = 0; apic < nr_ioapics; apic++)
644 for (pin = 0; pin < nr_ioapic_registers[apic]; pin++)
645 clear_IO_APIC_pin(apic, pin);
650 * support for broken MP BIOSs, enables hand-redirection of PIRQ0-7 to
651 * specific CPU-side IRQs.
655 static int pirq_entries[MAX_PIRQS] = {
656 [0 ... MAX_PIRQS - 1] = -1
659 static int __init ioapic_pirq_setup(char *str)
662 int ints[MAX_PIRQS+1];
664 get_options(str, ARRAY_SIZE(ints), ints);
666 apic_printk(APIC_VERBOSE, KERN_INFO
667 "PIRQ redirection, working around broken MP-BIOS.\n");
669 if (ints[0] < MAX_PIRQS)
672 for (i = 0; i < max; i++) {
673 apic_printk(APIC_VERBOSE, KERN_DEBUG
674 "... PIRQ%d -> IRQ %d\n", i, ints[i+1]);
676 * PIRQs are mapped upside down, usually.
678 pirq_entries[MAX_PIRQS-i-1] = ints[i+1];
683 __setup("pirq=", ioapic_pirq_setup);
684 #endif /* CONFIG_X86_32 */
686 struct IO_APIC_route_entry **alloc_ioapic_entries(void)
689 struct IO_APIC_route_entry **ioapic_entries;
691 ioapic_entries = kzalloc(sizeof(*ioapic_entries) * nr_ioapics,
696 for (apic = 0; apic < nr_ioapics; apic++) {
697 ioapic_entries[apic] =
698 kzalloc(sizeof(struct IO_APIC_route_entry) *
699 nr_ioapic_registers[apic], GFP_ATOMIC);
700 if (!ioapic_entries[apic])
704 return ioapic_entries;
708 kfree(ioapic_entries[apic]);
709 kfree(ioapic_entries);
715 * Saves all the IO-APIC RTE's
717 int save_IO_APIC_setup(struct IO_APIC_route_entry **ioapic_entries)
724 for (apic = 0; apic < nr_ioapics; apic++) {
725 if (!ioapic_entries[apic])
728 for (pin = 0; pin < nr_ioapic_registers[apic]; pin++)
729 ioapic_entries[apic][pin] =
730 ioapic_read_entry(apic, pin);
737 * Mask all IO APIC entries.
739 void mask_IO_APIC_setup(struct IO_APIC_route_entry **ioapic_entries)
746 for (apic = 0; apic < nr_ioapics; apic++) {
747 if (!ioapic_entries[apic])
750 for (pin = 0; pin < nr_ioapic_registers[apic]; pin++) {
751 struct IO_APIC_route_entry entry;
753 entry = ioapic_entries[apic][pin];
756 ioapic_write_entry(apic, pin, entry);
763 * Restore IO APIC entries which was saved in ioapic_entries.
765 int restore_IO_APIC_setup(struct IO_APIC_route_entry **ioapic_entries)
772 for (apic = 0; apic < nr_ioapics; apic++) {
773 if (!ioapic_entries[apic])
776 for (pin = 0; pin < nr_ioapic_registers[apic]; pin++)
777 ioapic_write_entry(apic, pin,
778 ioapic_entries[apic][pin]);
783 void free_ioapic_entries(struct IO_APIC_route_entry **ioapic_entries)
787 for (apic = 0; apic < nr_ioapics; apic++)
788 kfree(ioapic_entries[apic]);
790 kfree(ioapic_entries);
794 * Find the IRQ entry number of a certain pin.
796 static int find_irq_entry(int apic, int pin, int type)
800 for (i = 0; i < mp_irq_entries; i++)
801 if (mp_irqs[i].irqtype == type &&
802 (mp_irqs[i].dstapic == mp_ioapics[apic].apicid ||
803 mp_irqs[i].dstapic == MP_APIC_ALL) &&
804 mp_irqs[i].dstirq == pin)
811 * Find the pin to which IRQ[irq] (ISA) is connected
813 static int __init find_isa_irq_pin(int irq, int type)
817 for (i = 0; i < mp_irq_entries; i++) {
818 int lbus = mp_irqs[i].srcbus;
820 if (test_bit(lbus, mp_bus_not_pci) &&
821 (mp_irqs[i].irqtype == type) &&
822 (mp_irqs[i].srcbusirq == irq))
824 return mp_irqs[i].dstirq;
829 static int __init find_isa_irq_apic(int irq, int type)
833 for (i = 0; i < mp_irq_entries; i++) {
834 int lbus = mp_irqs[i].srcbus;
836 if (test_bit(lbus, mp_bus_not_pci) &&
837 (mp_irqs[i].irqtype == type) &&
838 (mp_irqs[i].srcbusirq == irq))
841 if (i < mp_irq_entries) {
843 for(apic = 0; apic < nr_ioapics; apic++) {
844 if (mp_ioapics[apic].apicid == mp_irqs[i].dstapic)
852 #if defined(CONFIG_EISA) || defined(CONFIG_MCA)
854 * EISA Edge/Level control register, ELCR
856 static int EISA_ELCR(unsigned int irq)
858 if (irq < legacy_pic->nr_legacy_irqs) {
859 unsigned int port = 0x4d0 + (irq >> 3);
860 return (inb(port) >> (irq & 7)) & 1;
862 apic_printk(APIC_VERBOSE, KERN_INFO
863 "Broken MPtable reports ISA irq %d\n", irq);
869 /* ISA interrupts are always polarity zero edge triggered,
870 * when listed as conforming in the MP table. */
872 #define default_ISA_trigger(idx) (0)
873 #define default_ISA_polarity(idx) (0)
875 /* EISA interrupts are always polarity zero and can be edge or level
876 * trigger depending on the ELCR value. If an interrupt is listed as
877 * EISA conforming in the MP table, that means its trigger type must
878 * be read in from the ELCR */
880 #define default_EISA_trigger(idx) (EISA_ELCR(mp_irqs[idx].srcbusirq))
881 #define default_EISA_polarity(idx) default_ISA_polarity(idx)
883 /* PCI interrupts are always polarity one level triggered,
884 * when listed as conforming in the MP table. */
886 #define default_PCI_trigger(idx) (1)
887 #define default_PCI_polarity(idx) (1)
889 /* MCA interrupts are always polarity zero level triggered,
890 * when listed as conforming in the MP table. */
892 #define default_MCA_trigger(idx) (1)
893 #define default_MCA_polarity(idx) default_ISA_polarity(idx)
895 static int MPBIOS_polarity(int idx)
897 int bus = mp_irqs[idx].srcbus;
901 * Determine IRQ line polarity (high active or low active):
903 switch (mp_irqs[idx].irqflag & 3)
905 case 0: /* conforms, ie. bus-type dependent polarity */
906 if (test_bit(bus, mp_bus_not_pci))
907 polarity = default_ISA_polarity(idx);
909 polarity = default_PCI_polarity(idx);
911 case 1: /* high active */
916 case 2: /* reserved */
918 printk(KERN_WARNING "broken BIOS!!\n");
922 case 3: /* low active */
927 default: /* invalid */
929 printk(KERN_WARNING "broken BIOS!!\n");
937 static int MPBIOS_trigger(int idx)
939 int bus = mp_irqs[idx].srcbus;
943 * Determine IRQ trigger mode (edge or level sensitive):
945 switch ((mp_irqs[idx].irqflag>>2) & 3)
947 case 0: /* conforms, ie. bus-type dependent */
948 if (test_bit(bus, mp_bus_not_pci))
949 trigger = default_ISA_trigger(idx);
951 trigger = default_PCI_trigger(idx);
952 #if defined(CONFIG_EISA) || defined(CONFIG_MCA)
953 switch (mp_bus_id_to_type[bus]) {
954 case MP_BUS_ISA: /* ISA pin */
956 /* set before the switch */
959 case MP_BUS_EISA: /* EISA pin */
961 trigger = default_EISA_trigger(idx);
964 case MP_BUS_PCI: /* PCI pin */
966 /* set before the switch */
969 case MP_BUS_MCA: /* MCA pin */
971 trigger = default_MCA_trigger(idx);
976 printk(KERN_WARNING "broken BIOS!!\n");
988 case 2: /* reserved */
990 printk(KERN_WARNING "broken BIOS!!\n");
999 default: /* invalid */
1001 printk(KERN_WARNING "broken BIOS!!\n");
1009 static inline int irq_polarity(int idx)
1011 return MPBIOS_polarity(idx);
1014 static inline int irq_trigger(int idx)
1016 return MPBIOS_trigger(idx);
1019 int (*ioapic_renumber_irq)(int ioapic, int irq);
1020 static int pin_2_irq(int idx, int apic, int pin)
1023 int bus = mp_irqs[idx].srcbus;
1026 * Debugging check, we are in big trouble if this message pops up!
1028 if (mp_irqs[idx].dstirq != pin)
1029 printk(KERN_ERR "broken BIOS or MPTABLE parser, ayiee!!\n");
1031 if (test_bit(bus, mp_bus_not_pci)) {
1032 irq = mp_irqs[idx].srcbusirq;
1034 u32 gsi = mp_gsi_routing[apic].gsi_base + pin;
1036 * For MPS mode, so far only needed by ES7000 platform
1038 if (ioapic_renumber_irq)
1039 gsi = ioapic_renumber_irq(apic, gsi);
1043 #ifdef CONFIG_X86_32
1045 * PCI IRQ command line redirection. Yes, limits are hardcoded.
1047 if ((pin >= 16) && (pin <= 23)) {
1048 if (pirq_entries[pin-16] != -1) {
1049 if (!pirq_entries[pin-16]) {
1050 apic_printk(APIC_VERBOSE, KERN_DEBUG
1051 "disabling PIRQ%d\n", pin-16);
1053 irq = pirq_entries[pin-16];
1054 apic_printk(APIC_VERBOSE, KERN_DEBUG
1055 "using PIRQ%d -> IRQ %d\n",
1066 * Find a specific PCI IRQ entry.
1067 * Not an __init, possibly needed by modules
1069 int IO_APIC_get_PCI_irq_vector(int bus, int slot, int pin,
1070 struct io_apic_irq_attr *irq_attr)
1072 int apic, i, best_guess = -1;
1074 apic_printk(APIC_DEBUG,
1075 "querying PCI -> IRQ mapping bus:%d, slot:%d, pin:%d.\n",
1077 if (test_bit(bus, mp_bus_not_pci)) {
1078 apic_printk(APIC_VERBOSE,
1079 "PCI BIOS passed nonexistent PCI bus %d!\n", bus);
1082 for (i = 0; i < mp_irq_entries; i++) {
1083 int lbus = mp_irqs[i].srcbus;
1085 for (apic = 0; apic < nr_ioapics; apic++)
1086 if (mp_ioapics[apic].apicid == mp_irqs[i].dstapic ||
1087 mp_irqs[i].dstapic == MP_APIC_ALL)
1090 if (!test_bit(lbus, mp_bus_not_pci) &&
1091 !mp_irqs[i].irqtype &&
1093 (slot == ((mp_irqs[i].srcbusirq >> 2) & 0x1f))) {
1094 int irq = pin_2_irq(i, apic, mp_irqs[i].dstirq);
1096 if (!(apic || IO_APIC_IRQ(irq)))
1099 if (pin == (mp_irqs[i].srcbusirq & 3)) {
1100 set_io_apic_irq_attr(irq_attr, apic,
1107 * Use the first all-but-pin matching entry as a
1108 * best-guess fuzzy result for broken mptables.
1110 if (best_guess < 0) {
1111 set_io_apic_irq_attr(irq_attr, apic,
1121 EXPORT_SYMBOL(IO_APIC_get_PCI_irq_vector);
1123 void lock_vector_lock(void)
1125 /* Used to the online set of cpus does not change
1126 * during assign_irq_vector.
1128 raw_spin_lock(&vector_lock);
1131 void unlock_vector_lock(void)
1133 raw_spin_unlock(&vector_lock);
1137 __assign_irq_vector(int irq, struct irq_cfg *cfg, const struct cpumask *mask)
1140 * NOTE! The local APIC isn't very good at handling
1141 * multiple interrupts at the same interrupt level.
1142 * As the interrupt level is determined by taking the
1143 * vector number and shifting that right by 4, we
1144 * want to spread these out a bit so that they don't
1145 * all fall in the same interrupt level.
1147 * Also, we've got to be careful not to trash gate
1148 * 0x80, because int 0x80 is hm, kind of importantish. ;)
1150 static int current_vector = FIRST_EXTERNAL_VECTOR + VECTOR_OFFSET_START;
1151 static int current_offset = VECTOR_OFFSET_START % 8;
1152 unsigned int old_vector;
1154 cpumask_var_t tmp_mask;
1156 if (cfg->move_in_progress)
1159 if (!alloc_cpumask_var(&tmp_mask, GFP_ATOMIC))
1162 old_vector = cfg->vector;
1164 cpumask_and(tmp_mask, mask, cpu_online_mask);
1165 cpumask_and(tmp_mask, cfg->domain, tmp_mask);
1166 if (!cpumask_empty(tmp_mask)) {
1167 free_cpumask_var(tmp_mask);
1172 /* Only try and allocate irqs on cpus that are present */
1174 for_each_cpu_and(cpu, mask, cpu_online_mask) {
1178 apic->vector_allocation_domain(cpu, tmp_mask);
1180 vector = current_vector;
1181 offset = current_offset;
1184 if (vector >= first_system_vector) {
1185 /* If out of vectors on large boxen, must share them. */
1186 offset = (offset + 1) % 8;
1187 vector = FIRST_EXTERNAL_VECTOR + offset;
1189 if (unlikely(current_vector == vector))
1192 if (test_bit(vector, used_vectors))
1195 for_each_cpu_and(new_cpu, tmp_mask, cpu_online_mask)
1196 if (per_cpu(vector_irq, new_cpu)[vector] != -1)
1199 current_vector = vector;
1200 current_offset = offset;
1202 cfg->move_in_progress = 1;
1203 cpumask_copy(cfg->old_domain, cfg->domain);
1205 for_each_cpu_and(new_cpu, tmp_mask, cpu_online_mask)
1206 per_cpu(vector_irq, new_cpu)[vector] = irq;
1207 cfg->vector = vector;
1208 cpumask_copy(cfg->domain, tmp_mask);
1212 free_cpumask_var(tmp_mask);
1216 int assign_irq_vector(int irq, struct irq_cfg *cfg, const struct cpumask *mask)
1219 unsigned long flags;
1221 raw_spin_lock_irqsave(&vector_lock, flags);
1222 err = __assign_irq_vector(irq, cfg, mask);
1223 raw_spin_unlock_irqrestore(&vector_lock, flags);
1227 static void __clear_irq_vector(int irq, struct irq_cfg *cfg)
1231 BUG_ON(!cfg->vector);
1233 vector = cfg->vector;
1234 for_each_cpu_and(cpu, cfg->domain, cpu_online_mask)
1235 per_cpu(vector_irq, cpu)[vector] = -1;
1238 cpumask_clear(cfg->domain);
1240 if (likely(!cfg->move_in_progress))
1242 for_each_cpu_and(cpu, cfg->old_domain, cpu_online_mask) {
1243 for (vector = FIRST_EXTERNAL_VECTOR; vector < NR_VECTORS;
1245 if (per_cpu(vector_irq, cpu)[vector] != irq)
1247 per_cpu(vector_irq, cpu)[vector] = -1;
1251 cfg->move_in_progress = 0;
1254 void __setup_vector_irq(int cpu)
1256 /* Initialize vector_irq on a new cpu */
1258 struct irq_cfg *cfg;
1259 struct irq_desc *desc;
1262 * vector_lock will make sure that we don't run into irq vector
1263 * assignments that might be happening on another cpu in parallel,
1264 * while we setup our initial vector to irq mappings.
1266 raw_spin_lock(&vector_lock);
1267 /* Mark the inuse vectors */
1268 for_each_irq_desc(irq, desc) {
1269 cfg = desc->chip_data;
1272 * If it is a legacy IRQ handled by the legacy PIC, this cpu
1273 * will be part of the irq_cfg's domain.
1275 if (irq < legacy_pic->nr_legacy_irqs && !IO_APIC_IRQ(irq))
1276 cpumask_set_cpu(cpu, cfg->domain);
1278 if (!cpumask_test_cpu(cpu, cfg->domain))
1280 vector = cfg->vector;
1281 per_cpu(vector_irq, cpu)[vector] = irq;
1283 /* Mark the free vectors */
1284 for (vector = 0; vector < NR_VECTORS; ++vector) {
1285 irq = per_cpu(vector_irq, cpu)[vector];
1290 if (!cpumask_test_cpu(cpu, cfg->domain))
1291 per_cpu(vector_irq, cpu)[vector] = -1;
1293 raw_spin_unlock(&vector_lock);
1296 static struct irq_chip ioapic_chip;
1297 static struct irq_chip ir_ioapic_chip;
1299 #define IOAPIC_AUTO -1
1300 #define IOAPIC_EDGE 0
1301 #define IOAPIC_LEVEL 1
1303 #ifdef CONFIG_X86_32
1304 static inline int IO_APIC_irq_trigger(int irq)
1308 for (apic = 0; apic < nr_ioapics; apic++) {
1309 for (pin = 0; pin < nr_ioapic_registers[apic]; pin++) {
1310 idx = find_irq_entry(apic, pin, mp_INT);
1311 if ((idx != -1) && (irq == pin_2_irq(idx, apic, pin)))
1312 return irq_trigger(idx);
1316 * nonexistent IRQs are edge default
1321 static inline int IO_APIC_irq_trigger(int irq)
1327 static void ioapic_register_intr(int irq, struct irq_desc *desc, unsigned long trigger)
1330 if ((trigger == IOAPIC_AUTO && IO_APIC_irq_trigger(irq)) ||
1331 trigger == IOAPIC_LEVEL)
1332 desc->status |= IRQ_LEVEL;
1334 desc->status &= ~IRQ_LEVEL;
1336 if (irq_remapped(irq)) {
1337 desc->status |= IRQ_MOVE_PCNTXT;
1339 set_irq_chip_and_handler_name(irq, &ir_ioapic_chip,
1343 set_irq_chip_and_handler_name(irq, &ir_ioapic_chip,
1344 handle_edge_irq, "edge");
1348 if ((trigger == IOAPIC_AUTO && IO_APIC_irq_trigger(irq)) ||
1349 trigger == IOAPIC_LEVEL)
1350 set_irq_chip_and_handler_name(irq, &ioapic_chip,
1354 set_irq_chip_and_handler_name(irq, &ioapic_chip,
1355 handle_edge_irq, "edge");
1358 int setup_ioapic_entry(int apic_id, int irq,
1359 struct IO_APIC_route_entry *entry,
1360 unsigned int destination, int trigger,
1361 int polarity, int vector, int pin)
1364 * add it to the IO-APIC irq-routing table:
1366 memset(entry,0,sizeof(*entry));
1368 if (intr_remapping_enabled) {
1369 struct intel_iommu *iommu = map_ioapic_to_ir(apic_id);
1371 struct IR_IO_APIC_route_entry *ir_entry =
1372 (struct IR_IO_APIC_route_entry *) entry;
1376 panic("No mapping iommu for ioapic %d\n", apic_id);
1378 index = alloc_irte(iommu, irq, 1);
1380 panic("Failed to allocate IRTE for ioapic %d\n", apic_id);
1382 memset(&irte, 0, sizeof(irte));
1385 irte.dst_mode = apic->irq_dest_mode;
1387 * Trigger mode in the IRTE will always be edge, and the
1388 * actual level or edge trigger will be setup in the IO-APIC
1389 * RTE. This will help simplify level triggered irq migration.
1390 * For more details, see the comments above explainig IO-APIC
1391 * irq migration in the presence of interrupt-remapping.
1393 irte.trigger_mode = 0;
1394 irte.dlvry_mode = apic->irq_delivery_mode;
1395 irte.vector = vector;
1396 irte.dest_id = IRTE_DEST(destination);
1398 /* Set source-id of interrupt request */
1399 set_ioapic_sid(&irte, apic_id);
1401 modify_irte(irq, &irte);
1403 ir_entry->index2 = (index >> 15) & 0x1;
1405 ir_entry->format = 1;
1406 ir_entry->index = (index & 0x7fff);
1408 * IO-APIC RTE will be configured with virtual vector.
1409 * irq handler will do the explicit EOI to the io-apic.
1411 ir_entry->vector = pin;
1413 entry->delivery_mode = apic->irq_delivery_mode;
1414 entry->dest_mode = apic->irq_dest_mode;
1415 entry->dest = destination;
1416 entry->vector = vector;
1419 entry->mask = 0; /* enable IRQ */
1420 entry->trigger = trigger;
1421 entry->polarity = polarity;
1423 /* Mask level triggered irqs.
1424 * Use IRQ_DELAYED_DISABLE for edge triggered irqs.
1431 static void setup_IO_APIC_irq(int apic_id, int pin, unsigned int irq, struct irq_desc *desc,
1432 int trigger, int polarity)
1434 struct irq_cfg *cfg;
1435 struct IO_APIC_route_entry entry;
1438 if (!IO_APIC_IRQ(irq))
1441 cfg = desc->chip_data;
1444 * For legacy irqs, cfg->domain starts with cpu 0 for legacy
1445 * controllers like 8259. Now that IO-APIC can handle this irq, update
1448 if (irq < legacy_pic->nr_legacy_irqs && cpumask_test_cpu(0, cfg->domain))
1449 apic->vector_allocation_domain(0, cfg->domain);
1451 if (assign_irq_vector(irq, cfg, apic->target_cpus()))
1454 dest = apic->cpu_mask_to_apicid_and(cfg->domain, apic->target_cpus());
1456 apic_printk(APIC_VERBOSE,KERN_DEBUG
1457 "IOAPIC[%d]: Set routing entry (%d-%d -> 0x%x -> "
1458 "IRQ %d Mode:%i Active:%i)\n",
1459 apic_id, mp_ioapics[apic_id].apicid, pin, cfg->vector,
1460 irq, trigger, polarity);
1463 if (setup_ioapic_entry(mp_ioapics[apic_id].apicid, irq, &entry,
1464 dest, trigger, polarity, cfg->vector, pin)) {
1465 printk("Failed to setup ioapic entry for ioapic %d, pin %d\n",
1466 mp_ioapics[apic_id].apicid, pin);
1467 __clear_irq_vector(irq, cfg);
1471 ioapic_register_intr(irq, desc, trigger);
1472 if (irq < legacy_pic->nr_legacy_irqs)
1473 legacy_pic->chip->mask(irq);
1475 ioapic_write_entry(apic_id, pin, entry);
1479 DECLARE_BITMAP(pin_programmed, MP_MAX_IOAPIC_PIN + 1);
1480 } mp_ioapic_routing[MAX_IO_APICS];
1482 static void __init setup_IO_APIC_irqs(void)
1484 int apic_id, pin, idx, irq;
1486 struct irq_desc *desc;
1487 struct irq_cfg *cfg;
1488 int node = cpu_to_node(boot_cpu_id);
1490 apic_printk(APIC_VERBOSE, KERN_DEBUG "init IO_APIC IRQs\n");
1492 for (apic_id = 0; apic_id < nr_ioapics; apic_id++)
1493 for (pin = 0; pin < nr_ioapic_registers[apic_id]; pin++) {
1494 idx = find_irq_entry(apic_id, pin, mp_INT);
1498 apic_printk(APIC_VERBOSE,
1499 KERN_DEBUG " %d-%d",
1500 mp_ioapics[apic_id].apicid, pin);
1502 apic_printk(APIC_VERBOSE, " %d-%d",
1503 mp_ioapics[apic_id].apicid, pin);
1507 apic_printk(APIC_VERBOSE,
1508 " (apicid-pin) not connected\n");
1512 irq = pin_2_irq(idx, apic_id, pin);
1514 if ((apic_id > 0) && (irq > 16))
1518 * Skip the timer IRQ if there's a quirk handler
1519 * installed and if it returns 1:
1521 if (apic->multi_timer_check &&
1522 apic->multi_timer_check(apic_id, irq))
1525 desc = irq_to_desc_alloc_node(irq, node);
1527 printk(KERN_INFO "can not get irq_desc for %d\n", irq);
1530 cfg = desc->chip_data;
1531 add_pin_to_irq_node(cfg, node, apic_id, pin);
1533 * don't mark it in pin_programmed, so later acpi could
1534 * set it correctly when irq < 16
1536 setup_IO_APIC_irq(apic_id, pin, irq, desc,
1537 irq_trigger(idx), irq_polarity(idx));
1541 apic_printk(APIC_VERBOSE,
1542 " (apicid-pin) not connected\n");
1546 * for the gsit that is not in first ioapic
1547 * but could not use acpi_register_gsi()
1548 * like some special sci in IBM x3330
1550 void setup_IO_APIC_irq_extra(u32 gsi)
1552 int apic_id = 0, pin, idx, irq;
1553 int node = cpu_to_node(boot_cpu_id);
1554 struct irq_desc *desc;
1555 struct irq_cfg *cfg;
1558 * Convert 'gsi' to 'ioapic.pin'.
1560 apic_id = mp_find_ioapic(gsi);
1564 pin = mp_find_ioapic_pin(apic_id, gsi);
1565 idx = find_irq_entry(apic_id, pin, mp_INT);
1569 irq = pin_2_irq(idx, apic_id, pin);
1570 #ifdef CONFIG_SPARSE_IRQ
1571 desc = irq_to_desc(irq);
1575 desc = irq_to_desc_alloc_node(irq, node);
1577 printk(KERN_INFO "can not get irq_desc for %d\n", irq);
1581 cfg = desc->chip_data;
1582 add_pin_to_irq_node(cfg, node, apic_id, pin);
1584 if (test_bit(pin, mp_ioapic_routing[apic_id].pin_programmed)) {
1585 pr_debug("Pin %d-%d already programmed\n",
1586 mp_ioapics[apic_id].apicid, pin);
1589 set_bit(pin, mp_ioapic_routing[apic_id].pin_programmed);
1591 setup_IO_APIC_irq(apic_id, pin, irq, desc,
1592 irq_trigger(idx), irq_polarity(idx));
1596 * Set up the timer pin, possibly with the 8259A-master behind.
1598 static void __init setup_timer_IRQ0_pin(unsigned int apic_id, unsigned int pin,
1601 struct IO_APIC_route_entry entry;
1603 if (intr_remapping_enabled)
1606 memset(&entry, 0, sizeof(entry));
1609 * We use logical delivery to get the timer IRQ
1612 entry.dest_mode = apic->irq_dest_mode;
1613 entry.mask = 0; /* don't mask IRQ for edge */
1614 entry.dest = apic->cpu_mask_to_apicid(apic->target_cpus());
1615 entry.delivery_mode = apic->irq_delivery_mode;
1618 entry.vector = vector;
1621 * The timer IRQ doesn't have to know that behind the
1622 * scene we may have a 8259A-master in AEOI mode ...
1624 set_irq_chip_and_handler_name(0, &ioapic_chip, handle_edge_irq, "edge");
1627 * Add it to the IO-APIC irq-routing table:
1629 ioapic_write_entry(apic_id, pin, entry);
1633 __apicdebuginit(void) print_IO_APIC(void)
1636 union IO_APIC_reg_00 reg_00;
1637 union IO_APIC_reg_01 reg_01;
1638 union IO_APIC_reg_02 reg_02;
1639 union IO_APIC_reg_03 reg_03;
1640 unsigned long flags;
1641 struct irq_cfg *cfg;
1642 struct irq_desc *desc;
1645 printk(KERN_DEBUG "number of MP IRQ sources: %d.\n", mp_irq_entries);
1646 for (i = 0; i < nr_ioapics; i++)
1647 printk(KERN_DEBUG "number of IO-APIC #%d registers: %d.\n",
1648 mp_ioapics[i].apicid, nr_ioapic_registers[i]);
1651 * We are a bit conservative about what we expect. We have to
1652 * know about every hardware change ASAP.
1654 printk(KERN_INFO "testing the IO APIC.......................\n");
1656 for (apic = 0; apic < nr_ioapics; apic++) {
1658 raw_spin_lock_irqsave(&ioapic_lock, flags);
1659 reg_00.raw = io_apic_read(apic, 0);
1660 reg_01.raw = io_apic_read(apic, 1);
1661 if (reg_01.bits.version >= 0x10)
1662 reg_02.raw = io_apic_read(apic, 2);
1663 if (reg_01.bits.version >= 0x20)
1664 reg_03.raw = io_apic_read(apic, 3);
1665 raw_spin_unlock_irqrestore(&ioapic_lock, flags);
1668 printk(KERN_DEBUG "IO APIC #%d......\n", mp_ioapics[apic].apicid);
1669 printk(KERN_DEBUG ".... register #00: %08X\n", reg_00.raw);
1670 printk(KERN_DEBUG "....... : physical APIC id: %02X\n", reg_00.bits.ID);
1671 printk(KERN_DEBUG "....... : Delivery Type: %X\n", reg_00.bits.delivery_type);
1672 printk(KERN_DEBUG "....... : LTS : %X\n", reg_00.bits.LTS);
1674 printk(KERN_DEBUG ".... register #01: %08X\n", *(int *)®_01);
1675 printk(KERN_DEBUG "....... : max redirection entries: %04X\n", reg_01.bits.entries);
1677 printk(KERN_DEBUG "....... : PRQ implemented: %X\n", reg_01.bits.PRQ);
1678 printk(KERN_DEBUG "....... : IO APIC version: %04X\n", reg_01.bits.version);
1681 * Some Intel chipsets with IO APIC VERSION of 0x1? don't have reg_02,
1682 * but the value of reg_02 is read as the previous read register
1683 * value, so ignore it if reg_02 == reg_01.
1685 if (reg_01.bits.version >= 0x10 && reg_02.raw != reg_01.raw) {
1686 printk(KERN_DEBUG ".... register #02: %08X\n", reg_02.raw);
1687 printk(KERN_DEBUG "....... : arbitration: %02X\n", reg_02.bits.arbitration);
1691 * Some Intel chipsets with IO APIC VERSION of 0x2? don't have reg_02
1692 * or reg_03, but the value of reg_0[23] is read as the previous read
1693 * register value, so ignore it if reg_03 == reg_0[12].
1695 if (reg_01.bits.version >= 0x20 && reg_03.raw != reg_02.raw &&
1696 reg_03.raw != reg_01.raw) {
1697 printk(KERN_DEBUG ".... register #03: %08X\n", reg_03.raw);
1698 printk(KERN_DEBUG "....... : Boot DT : %X\n", reg_03.bits.boot_DT);
1701 printk(KERN_DEBUG ".... IRQ redirection table:\n");
1703 printk(KERN_DEBUG " NR Dst Mask Trig IRR Pol"
1704 " Stat Dmod Deli Vect:\n");
1706 for (i = 0; i <= reg_01.bits.entries; i++) {
1707 struct IO_APIC_route_entry entry;
1709 entry = ioapic_read_entry(apic, i);
1711 printk(KERN_DEBUG " %02x %03X ",
1716 printk("%1d %1d %1d %1d %1d %1d %1d %02X\n",
1721 entry.delivery_status,
1723 entry.delivery_mode,
1728 printk(KERN_DEBUG "IRQ to pin mappings:\n");
1729 for_each_irq_desc(irq, desc) {
1730 struct irq_pin_list *entry;
1732 cfg = desc->chip_data;
1733 entry = cfg->irq_2_pin;
1736 printk(KERN_DEBUG "IRQ%d ", irq);
1737 for_each_irq_pin(entry, cfg->irq_2_pin)
1738 printk("-> %d:%d", entry->apic, entry->pin);
1742 printk(KERN_INFO ".................................... done.\n");
1747 __apicdebuginit(void) print_APIC_field(int base)
1753 for (i = 0; i < 8; i++)
1754 printk(KERN_CONT "%08x", apic_read(base + i*0x10));
1756 printk(KERN_CONT "\n");
1759 __apicdebuginit(void) print_local_APIC(void *dummy)
1761 unsigned int i, v, ver, maxlvt;
1764 printk(KERN_DEBUG "printing local APIC contents on CPU#%d/%d:\n",
1765 smp_processor_id(), hard_smp_processor_id());
1766 v = apic_read(APIC_ID);
1767 printk(KERN_INFO "... APIC ID: %08x (%01x)\n", v, read_apic_id());
1768 v = apic_read(APIC_LVR);
1769 printk(KERN_INFO "... APIC VERSION: %08x\n", v);
1770 ver = GET_APIC_VERSION(v);
1771 maxlvt = lapic_get_maxlvt();
1773 v = apic_read(APIC_TASKPRI);
1774 printk(KERN_DEBUG "... APIC TASKPRI: %08x (%02x)\n", v, v & APIC_TPRI_MASK);
1776 if (APIC_INTEGRATED(ver)) { /* !82489DX */
1777 if (!APIC_XAPIC(ver)) {
1778 v = apic_read(APIC_ARBPRI);
1779 printk(KERN_DEBUG "... APIC ARBPRI: %08x (%02x)\n", v,
1780 v & APIC_ARBPRI_MASK);
1782 v = apic_read(APIC_PROCPRI);
1783 printk(KERN_DEBUG "... APIC PROCPRI: %08x\n", v);
1787 * Remote read supported only in the 82489DX and local APIC for
1788 * Pentium processors.
1790 if (!APIC_INTEGRATED(ver) || maxlvt == 3) {
1791 v = apic_read(APIC_RRR);
1792 printk(KERN_DEBUG "... APIC RRR: %08x\n", v);
1795 v = apic_read(APIC_LDR);
1796 printk(KERN_DEBUG "... APIC LDR: %08x\n", v);
1797 if (!x2apic_enabled()) {
1798 v = apic_read(APIC_DFR);
1799 printk(KERN_DEBUG "... APIC DFR: %08x\n", v);
1801 v = apic_read(APIC_SPIV);
1802 printk(KERN_DEBUG "... APIC SPIV: %08x\n", v);
1804 printk(KERN_DEBUG "... APIC ISR field:\n");
1805 print_APIC_field(APIC_ISR);
1806 printk(KERN_DEBUG "... APIC TMR field:\n");
1807 print_APIC_field(APIC_TMR);
1808 printk(KERN_DEBUG "... APIC IRR field:\n");
1809 print_APIC_field(APIC_IRR);
1811 if (APIC_INTEGRATED(ver)) { /* !82489DX */
1812 if (maxlvt > 3) /* Due to the Pentium erratum 3AP. */
1813 apic_write(APIC_ESR, 0);
1815 v = apic_read(APIC_ESR);
1816 printk(KERN_DEBUG "... APIC ESR: %08x\n", v);
1819 icr = apic_icr_read();
1820 printk(KERN_DEBUG "... APIC ICR: %08x\n", (u32)icr);
1821 printk(KERN_DEBUG "... APIC ICR2: %08x\n", (u32)(icr >> 32));
1823 v = apic_read(APIC_LVTT);
1824 printk(KERN_DEBUG "... APIC LVTT: %08x\n", v);
1826 if (maxlvt > 3) { /* PC is LVT#4. */
1827 v = apic_read(APIC_LVTPC);
1828 printk(KERN_DEBUG "... APIC LVTPC: %08x\n", v);
1830 v = apic_read(APIC_LVT0);
1831 printk(KERN_DEBUG "... APIC LVT0: %08x\n", v);
1832 v = apic_read(APIC_LVT1);
1833 printk(KERN_DEBUG "... APIC LVT1: %08x\n", v);
1835 if (maxlvt > 2) { /* ERR is LVT#3. */
1836 v = apic_read(APIC_LVTERR);
1837 printk(KERN_DEBUG "... APIC LVTERR: %08x\n", v);
1840 v = apic_read(APIC_TMICT);
1841 printk(KERN_DEBUG "... APIC TMICT: %08x\n", v);
1842 v = apic_read(APIC_TMCCT);
1843 printk(KERN_DEBUG "... APIC TMCCT: %08x\n", v);
1844 v = apic_read(APIC_TDCR);
1845 printk(KERN_DEBUG "... APIC TDCR: %08x\n", v);
1847 if (boot_cpu_has(X86_FEATURE_EXTAPIC)) {
1848 v = apic_read(APIC_EFEAT);
1849 maxlvt = (v >> 16) & 0xff;
1850 printk(KERN_DEBUG "... APIC EFEAT: %08x\n", v);
1851 v = apic_read(APIC_ECTRL);
1852 printk(KERN_DEBUG "... APIC ECTRL: %08x\n", v);
1853 for (i = 0; i < maxlvt; i++) {
1854 v = apic_read(APIC_EILVTn(i));
1855 printk(KERN_DEBUG "... APIC EILVT%d: %08x\n", i, v);
1861 __apicdebuginit(void) print_local_APICs(int maxcpu)
1869 for_each_online_cpu(cpu) {
1872 smp_call_function_single(cpu, print_local_APIC, NULL, 1);
1877 __apicdebuginit(void) print_PIC(void)
1880 unsigned long flags;
1882 if (!legacy_pic->nr_legacy_irqs)
1885 printk(KERN_DEBUG "\nprinting PIC contents\n");
1887 raw_spin_lock_irqsave(&i8259A_lock, flags);
1889 v = inb(0xa1) << 8 | inb(0x21);
1890 printk(KERN_DEBUG "... PIC IMR: %04x\n", v);
1892 v = inb(0xa0) << 8 | inb(0x20);
1893 printk(KERN_DEBUG "... PIC IRR: %04x\n", v);
1897 v = inb(0xa0) << 8 | inb(0x20);
1901 raw_spin_unlock_irqrestore(&i8259A_lock, flags);
1903 printk(KERN_DEBUG "... PIC ISR: %04x\n", v);
1905 v = inb(0x4d1) << 8 | inb(0x4d0);
1906 printk(KERN_DEBUG "... PIC ELCR: %04x\n", v);
1909 static int __initdata show_lapic = 1;
1910 static __init int setup_show_lapic(char *arg)
1914 if (strcmp(arg, "all") == 0) {
1915 show_lapic = CONFIG_NR_CPUS;
1917 get_option(&arg, &num);
1924 __setup("show_lapic=", setup_show_lapic);
1926 __apicdebuginit(int) print_ICs(void)
1928 if (apic_verbosity == APIC_QUIET)
1933 /* don't print out if apic is not there */
1934 if (!cpu_has_apic && !apic_from_smp_config())
1937 print_local_APICs(show_lapic);
1943 fs_initcall(print_ICs);
1946 /* Where if anywhere is the i8259 connect in external int mode */
1947 static struct { int pin, apic; } ioapic_i8259 = { -1, -1 };
1949 void __init enable_IO_APIC(void)
1951 int i8259_apic, i8259_pin;
1954 if (!legacy_pic->nr_legacy_irqs)
1957 for(apic = 0; apic < nr_ioapics; apic++) {
1959 /* See if any of the pins is in ExtINT mode */
1960 for (pin = 0; pin < nr_ioapic_registers[apic]; pin++) {
1961 struct IO_APIC_route_entry entry;
1962 entry = ioapic_read_entry(apic, pin);
1964 /* If the interrupt line is enabled and in ExtInt mode
1965 * I have found the pin where the i8259 is connected.
1967 if ((entry.mask == 0) && (entry.delivery_mode == dest_ExtINT)) {
1968 ioapic_i8259.apic = apic;
1969 ioapic_i8259.pin = pin;
1975 /* Look to see what if the MP table has reported the ExtINT */
1976 /* If we could not find the appropriate pin by looking at the ioapic
1977 * the i8259 probably is not connected the ioapic but give the
1978 * mptable a chance anyway.
1980 i8259_pin = find_isa_irq_pin(0, mp_ExtINT);
1981 i8259_apic = find_isa_irq_apic(0, mp_ExtINT);
1982 /* Trust the MP table if nothing is setup in the hardware */
1983 if ((ioapic_i8259.pin == -1) && (i8259_pin >= 0)) {
1984 printk(KERN_WARNING "ExtINT not setup in hardware but reported by MP table\n");
1985 ioapic_i8259.pin = i8259_pin;
1986 ioapic_i8259.apic = i8259_apic;
1988 /* Complain if the MP table and the hardware disagree */
1989 if (((ioapic_i8259.apic != i8259_apic) || (ioapic_i8259.pin != i8259_pin)) &&
1990 (i8259_pin >= 0) && (ioapic_i8259.pin >= 0))
1992 printk(KERN_WARNING "ExtINT in hardware and MP table differ\n");
1996 * Do not trust the IO-APIC being empty at bootup
2002 * Not an __init, needed by the reboot code
2004 void disable_IO_APIC(void)
2007 * Clear the IO-APIC before rebooting:
2011 if (!legacy_pic->nr_legacy_irqs)
2015 * If the i8259 is routed through an IOAPIC
2016 * Put that IOAPIC in virtual wire mode
2017 * so legacy interrupts can be delivered.
2019 * With interrupt-remapping, for now we will use virtual wire A mode,
2020 * as virtual wire B is little complex (need to configure both
2021 * IOAPIC RTE aswell as interrupt-remapping table entry).
2022 * As this gets called during crash dump, keep this simple for now.
2024 if (ioapic_i8259.pin != -1 && !intr_remapping_enabled) {
2025 struct IO_APIC_route_entry entry;
2027 memset(&entry, 0, sizeof(entry));
2028 entry.mask = 0; /* Enabled */
2029 entry.trigger = 0; /* Edge */
2031 entry.polarity = 0; /* High */
2032 entry.delivery_status = 0;
2033 entry.dest_mode = 0; /* Physical */
2034 entry.delivery_mode = dest_ExtINT; /* ExtInt */
2036 entry.dest = read_apic_id();
2039 * Add it to the IO-APIC irq-routing table:
2041 ioapic_write_entry(ioapic_i8259.apic, ioapic_i8259.pin, entry);
2045 * Use virtual wire A mode when interrupt remapping is enabled.
2047 if (cpu_has_apic || apic_from_smp_config())
2048 disconnect_bsp_APIC(!intr_remapping_enabled &&
2049 ioapic_i8259.pin != -1);
2052 #ifdef CONFIG_X86_32
2054 * function to set the IO-APIC physical IDs based on the
2055 * values stored in the MPC table.
2057 * by Matt Domsch <Matt_Domsch@dell.com> Tue Dec 21 12:25:05 CST 1999
2060 void __init setup_ioapic_ids_from_mpc(void)
2062 union IO_APIC_reg_00 reg_00;
2063 physid_mask_t phys_id_present_map;
2066 unsigned char old_id;
2067 unsigned long flags;
2072 * Don't check I/O APIC IDs for xAPIC systems. They have
2073 * no meaning without the serial APIC bus.
2075 if (!(boot_cpu_data.x86_vendor == X86_VENDOR_INTEL)
2076 || APIC_XAPIC(apic_version[boot_cpu_physical_apicid]))
2079 * This is broken; anything with a real cpu count has to
2080 * circumvent this idiocy regardless.
2082 apic->ioapic_phys_id_map(&phys_cpu_present_map, &phys_id_present_map);
2085 * Set the IOAPIC ID to the value stored in the MPC table.
2087 for (apic_id = 0; apic_id < nr_ioapics; apic_id++) {
2089 /* Read the register 0 value */
2090 raw_spin_lock_irqsave(&ioapic_lock, flags);
2091 reg_00.raw = io_apic_read(apic_id, 0);
2092 raw_spin_unlock_irqrestore(&ioapic_lock, flags);
2094 old_id = mp_ioapics[apic_id].apicid;
2096 if (mp_ioapics[apic_id].apicid >= get_physical_broadcast()) {
2097 printk(KERN_ERR "BIOS bug, IO-APIC#%d ID is %d in the MPC table!...\n",
2098 apic_id, mp_ioapics[apic_id].apicid);
2099 printk(KERN_ERR "... fixing up to %d. (tell your hw vendor)\n",
2101 mp_ioapics[apic_id].apicid = reg_00.bits.ID;
2105 * Sanity check, is the ID really free? Every APIC in a
2106 * system must have a unique ID or we get lots of nice
2107 * 'stuck on smp_invalidate_needed IPI wait' messages.
2109 if (apic->check_apicid_used(&phys_id_present_map,
2110 mp_ioapics[apic_id].apicid)) {
2111 printk(KERN_ERR "BIOS bug, IO-APIC#%d ID %d is already used!...\n",
2112 apic_id, mp_ioapics[apic_id].apicid);
2113 for (i = 0; i < get_physical_broadcast(); i++)
2114 if (!physid_isset(i, phys_id_present_map))
2116 if (i >= get_physical_broadcast())
2117 panic("Max APIC ID exceeded!\n");
2118 printk(KERN_ERR "... fixing up to %d. (tell your hw vendor)\n",
2120 physid_set(i, phys_id_present_map);
2121 mp_ioapics[apic_id].apicid = i;
2124 apic->apicid_to_cpu_present(mp_ioapics[apic_id].apicid, &tmp);
2125 apic_printk(APIC_VERBOSE, "Setting %d in the "
2126 "phys_id_present_map\n",
2127 mp_ioapics[apic_id].apicid);
2128 physids_or(phys_id_present_map, phys_id_present_map, tmp);
2133 * We need to adjust the IRQ routing table
2134 * if the ID changed.
2136 if (old_id != mp_ioapics[apic_id].apicid)
2137 for (i = 0; i < mp_irq_entries; i++)
2138 if (mp_irqs[i].dstapic == old_id)
2140 = mp_ioapics[apic_id].apicid;
2143 * Read the right value from the MPC table and
2144 * write it into the ID register.
2146 apic_printk(APIC_VERBOSE, KERN_INFO
2147 "...changing IO-APIC physical APIC ID to %d ...",
2148 mp_ioapics[apic_id].apicid);
2150 reg_00.bits.ID = mp_ioapics[apic_id].apicid;
2151 raw_spin_lock_irqsave(&ioapic_lock, flags);
2152 io_apic_write(apic_id, 0, reg_00.raw);
2153 raw_spin_unlock_irqrestore(&ioapic_lock, flags);
2158 raw_spin_lock_irqsave(&ioapic_lock, flags);
2159 reg_00.raw = io_apic_read(apic_id, 0);
2160 raw_spin_unlock_irqrestore(&ioapic_lock, flags);
2161 if (reg_00.bits.ID != mp_ioapics[apic_id].apicid)
2162 printk("could not set ID!\n");
2164 apic_printk(APIC_VERBOSE, " ok.\n");
2169 int no_timer_check __initdata;
2171 static int __init notimercheck(char *s)
2176 __setup("no_timer_check", notimercheck);
2179 * There is a nasty bug in some older SMP boards, their mptable lies
2180 * about the timer IRQ. We do the following to work around the situation:
2182 * - timer IRQ defaults to IO-APIC IRQ
2183 * - if this function detects that timer IRQs are defunct, then we fall
2184 * back to ISA timer IRQs
2186 static int __init timer_irq_works(void)
2188 unsigned long t1 = jiffies;
2189 unsigned long flags;
2194 local_save_flags(flags);
2196 /* Let ten ticks pass... */
2197 mdelay((10 * 1000) / HZ);
2198 local_irq_restore(flags);
2201 * Expect a few ticks at least, to be sure some possible
2202 * glue logic does not lock up after one or two first
2203 * ticks in a non-ExtINT mode. Also the local APIC
2204 * might have cached one ExtINT interrupt. Finally, at
2205 * least one tick may be lost due to delays.
2209 if (time_after(jiffies, t1 + 4))
2215 * In the SMP+IOAPIC case it might happen that there are an unspecified
2216 * number of pending IRQ events unhandled. These cases are very rare,
2217 * so we 'resend' these IRQs via IPIs, to the same CPU. It's much
2218 * better to do it this way as thus we do not have to be aware of
2219 * 'pending' interrupts in the IRQ path, except at this point.
2222 * Edge triggered needs to resend any interrupt
2223 * that was delayed but this is now handled in the device
2228 * Starting up a edge-triggered IO-APIC interrupt is
2229 * nasty - we need to make sure that we get the edge.
2230 * If it is already asserted for some reason, we need
2231 * return 1 to indicate that is was pending.
2233 * This is not complete - we should be able to fake
2234 * an edge even if it isn't on the 8259A...
2237 static unsigned int startup_ioapic_irq(unsigned int irq)
2239 int was_pending = 0;
2240 unsigned long flags;
2241 struct irq_cfg *cfg;
2243 raw_spin_lock_irqsave(&ioapic_lock, flags);
2244 if (irq < legacy_pic->nr_legacy_irqs) {
2245 legacy_pic->chip->mask(irq);
2246 if (legacy_pic->irq_pending(irq))
2250 __unmask_IO_APIC_irq(cfg);
2251 raw_spin_unlock_irqrestore(&ioapic_lock, flags);
2256 static int ioapic_retrigger_irq(unsigned int irq)
2259 struct irq_cfg *cfg = irq_cfg(irq);
2260 unsigned long flags;
2262 raw_spin_lock_irqsave(&vector_lock, flags);
2263 apic->send_IPI_mask(cpumask_of(cpumask_first(cfg->domain)), cfg->vector);
2264 raw_spin_unlock_irqrestore(&vector_lock, flags);
2270 * Level and edge triggered IO-APIC interrupts need different handling,
2271 * so we use two separate IRQ descriptors. Edge triggered IRQs can be
2272 * handled with the level-triggered descriptor, but that one has slightly
2273 * more overhead. Level-triggered interrupts cannot be handled with the
2274 * edge-triggered handler, without risking IRQ storms and other ugly
2279 void send_cleanup_vector(struct irq_cfg *cfg)
2281 cpumask_var_t cleanup_mask;
2283 if (unlikely(!alloc_cpumask_var(&cleanup_mask, GFP_ATOMIC))) {
2285 for_each_cpu_and(i, cfg->old_domain, cpu_online_mask)
2286 apic->send_IPI_mask(cpumask_of(i), IRQ_MOVE_CLEANUP_VECTOR);
2288 cpumask_and(cleanup_mask, cfg->old_domain, cpu_online_mask);
2289 apic->send_IPI_mask(cleanup_mask, IRQ_MOVE_CLEANUP_VECTOR);
2290 free_cpumask_var(cleanup_mask);
2292 cfg->move_in_progress = 0;
2295 static void __target_IO_APIC_irq(unsigned int irq, unsigned int dest, struct irq_cfg *cfg)
2298 struct irq_pin_list *entry;
2299 u8 vector = cfg->vector;
2301 for_each_irq_pin(entry, cfg->irq_2_pin) {
2307 * With interrupt-remapping, destination information comes
2308 * from interrupt-remapping table entry.
2310 if (!irq_remapped(irq))
2311 io_apic_write(apic, 0x11 + pin*2, dest);
2312 reg = io_apic_read(apic, 0x10 + pin*2);
2313 reg &= ~IO_APIC_REDIR_VECTOR_MASK;
2315 io_apic_modify(apic, 0x10 + pin*2, reg);
2320 * Either sets desc->affinity to a valid value, and returns
2321 * ->cpu_mask_to_apicid of that in dest_id, or returns -1 and
2322 * leaves desc->affinity untouched.
2325 set_desc_affinity(struct irq_desc *desc, const struct cpumask *mask,
2326 unsigned int *dest_id)
2328 struct irq_cfg *cfg;
2331 if (!cpumask_intersects(mask, cpu_online_mask))
2335 cfg = desc->chip_data;
2336 if (assign_irq_vector(irq, cfg, mask))
2339 cpumask_copy(desc->affinity, mask);
2341 *dest_id = apic->cpu_mask_to_apicid_and(desc->affinity, cfg->domain);
2346 set_ioapic_affinity_irq_desc(struct irq_desc *desc, const struct cpumask *mask)
2348 struct irq_cfg *cfg;
2349 unsigned long flags;
2355 cfg = desc->chip_data;
2357 raw_spin_lock_irqsave(&ioapic_lock, flags);
2358 ret = set_desc_affinity(desc, mask, &dest);
2360 /* Only the high 8 bits are valid. */
2361 dest = SET_APIC_LOGICAL_ID(dest);
2362 __target_IO_APIC_irq(irq, dest, cfg);
2364 raw_spin_unlock_irqrestore(&ioapic_lock, flags);
2370 set_ioapic_affinity_irq(unsigned int irq, const struct cpumask *mask)
2372 struct irq_desc *desc;
2374 desc = irq_to_desc(irq);
2376 return set_ioapic_affinity_irq_desc(desc, mask);
2379 #ifdef CONFIG_INTR_REMAP
2382 * Migrate the IO-APIC irq in the presence of intr-remapping.
2384 * For both level and edge triggered, irq migration is a simple atomic
2385 * update(of vector and cpu destination) of IRTE and flush the hardware cache.
2387 * For level triggered, we eliminate the io-apic RTE modification (with the
2388 * updated vector information), by using a virtual vector (io-apic pin number).
2389 * Real vector that is used for interrupting cpu will be coming from
2390 * the interrupt-remapping table entry.
2393 migrate_ioapic_irq_desc(struct irq_desc *desc, const struct cpumask *mask)
2395 struct irq_cfg *cfg;
2401 if (!cpumask_intersects(mask, cpu_online_mask))
2405 if (get_irte(irq, &irte))
2408 cfg = desc->chip_data;
2409 if (assign_irq_vector(irq, cfg, mask))
2412 dest = apic->cpu_mask_to_apicid_and(cfg->domain, mask);
2414 irte.vector = cfg->vector;
2415 irte.dest_id = IRTE_DEST(dest);
2418 * Modified the IRTE and flushes the Interrupt entry cache.
2420 modify_irte(irq, &irte);
2422 if (cfg->move_in_progress)
2423 send_cleanup_vector(cfg);
2425 cpumask_copy(desc->affinity, mask);
2431 * Migrates the IRQ destination in the process context.
2433 static int set_ir_ioapic_affinity_irq_desc(struct irq_desc *desc,
2434 const struct cpumask *mask)
2436 return migrate_ioapic_irq_desc(desc, mask);
2438 static int set_ir_ioapic_affinity_irq(unsigned int irq,
2439 const struct cpumask *mask)
2441 struct irq_desc *desc = irq_to_desc(irq);
2443 return set_ir_ioapic_affinity_irq_desc(desc, mask);
2446 static inline int set_ir_ioapic_affinity_irq_desc(struct irq_desc *desc,
2447 const struct cpumask *mask)
2453 asmlinkage void smp_irq_move_cleanup_interrupt(void)
2455 unsigned vector, me;
2461 me = smp_processor_id();
2462 for (vector = FIRST_EXTERNAL_VECTOR; vector < NR_VECTORS; vector++) {
2465 struct irq_desc *desc;
2466 struct irq_cfg *cfg;
2467 irq = __get_cpu_var(vector_irq)[vector];
2472 desc = irq_to_desc(irq);
2477 raw_spin_lock(&desc->lock);
2480 * Check if the irq migration is in progress. If so, we
2481 * haven't received the cleanup request yet for this irq.
2483 if (cfg->move_in_progress)
2486 if (vector == cfg->vector && cpumask_test_cpu(me, cfg->domain))
2489 irr = apic_read(APIC_IRR + (vector / 32 * 0x10));
2491 * Check if the vector that needs to be cleanedup is
2492 * registered at the cpu's IRR. If so, then this is not
2493 * the best time to clean it up. Lets clean it up in the
2494 * next attempt by sending another IRQ_MOVE_CLEANUP_VECTOR
2497 if (irr & (1 << (vector % 32))) {
2498 apic->send_IPI_self(IRQ_MOVE_CLEANUP_VECTOR);
2501 __get_cpu_var(vector_irq)[vector] = -1;
2503 raw_spin_unlock(&desc->lock);
2509 static void __irq_complete_move(struct irq_desc **descp, unsigned vector)
2511 struct irq_desc *desc = *descp;
2512 struct irq_cfg *cfg = desc->chip_data;
2515 if (likely(!cfg->move_in_progress))
2518 me = smp_processor_id();
2520 if (vector == cfg->vector && cpumask_test_cpu(me, cfg->domain))
2521 send_cleanup_vector(cfg);
2524 static void irq_complete_move(struct irq_desc **descp)
2526 __irq_complete_move(descp, ~get_irq_regs()->orig_ax);
2529 void irq_force_complete_move(int irq)
2531 struct irq_desc *desc = irq_to_desc(irq);
2532 struct irq_cfg *cfg = desc->chip_data;
2534 __irq_complete_move(&desc, cfg->vector);
2537 static inline void irq_complete_move(struct irq_desc **descp) {}
2540 static void ack_apic_edge(unsigned int irq)
2542 struct irq_desc *desc = irq_to_desc(irq);
2544 irq_complete_move(&desc);
2545 move_native_irq(irq);
2549 atomic_t irq_mis_count;
2552 * IO-APIC versions below 0x20 don't support EOI register.
2553 * For the record, here is the information about various versions:
2555 * 1Xh I/OAPIC or I/O(x)APIC which are not PCI 2.2 Compliant
2556 * 2Xh I/O(x)APIC which is PCI 2.2 Compliant
2559 * Some of the Intel ICH Specs (ICH2 to ICH5) documents the io-apic
2560 * version as 0x2. This is an error with documentation and these ICH chips
2561 * use io-apic's of version 0x20.
2563 * For IO-APIC's with EOI register, we use that to do an explicit EOI.
2564 * Otherwise, we simulate the EOI message manually by changing the trigger
2565 * mode to edge and then back to level, with RTE being masked during this.
2567 static void __eoi_ioapic_irq(unsigned int irq, struct irq_cfg *cfg)
2569 struct irq_pin_list *entry;
2571 for_each_irq_pin(entry, cfg->irq_2_pin) {
2572 if (mp_ioapics[entry->apic].apicver >= 0x20) {
2574 * Intr-remapping uses pin number as the virtual vector
2575 * in the RTE. Actual vector is programmed in
2576 * intr-remapping table entry. Hence for the io-apic
2577 * EOI we use the pin number.
2579 if (irq_remapped(irq))
2580 io_apic_eoi(entry->apic, entry->pin);
2582 io_apic_eoi(entry->apic, cfg->vector);
2584 __mask_and_edge_IO_APIC_irq(entry);
2585 __unmask_and_level_IO_APIC_irq(entry);
2590 static void eoi_ioapic_irq(struct irq_desc *desc)
2592 struct irq_cfg *cfg;
2593 unsigned long flags;
2597 cfg = desc->chip_data;
2599 raw_spin_lock_irqsave(&ioapic_lock, flags);
2600 __eoi_ioapic_irq(irq, cfg);
2601 raw_spin_unlock_irqrestore(&ioapic_lock, flags);
2604 static void ack_apic_level(unsigned int irq)
2606 struct irq_desc *desc = irq_to_desc(irq);
2609 struct irq_cfg *cfg;
2610 int do_unmask_irq = 0;
2612 irq_complete_move(&desc);
2613 #ifdef CONFIG_GENERIC_PENDING_IRQ
2614 /* If we are moving the irq we need to mask it */
2615 if (unlikely(desc->status & IRQ_MOVE_PENDING)) {
2617 mask_IO_APIC_irq_desc(desc);
2622 * It appears there is an erratum which affects at least version 0x11
2623 * of I/O APIC (that's the 82093AA and cores integrated into various
2624 * chipsets). Under certain conditions a level-triggered interrupt is
2625 * erroneously delivered as edge-triggered one but the respective IRR
2626 * bit gets set nevertheless. As a result the I/O unit expects an EOI
2627 * message but it will never arrive and further interrupts are blocked
2628 * from the source. The exact reason is so far unknown, but the
2629 * phenomenon was observed when two consecutive interrupt requests
2630 * from a given source get delivered to the same CPU and the source is
2631 * temporarily disabled in between.
2633 * A workaround is to simulate an EOI message manually. We achieve it
2634 * by setting the trigger mode to edge and then to level when the edge
2635 * trigger mode gets detected in the TMR of a local APIC for a
2636 * level-triggered interrupt. We mask the source for the time of the
2637 * operation to prevent an edge-triggered interrupt escaping meanwhile.
2638 * The idea is from Manfred Spraul. --macro
2640 * Also in the case when cpu goes offline, fixup_irqs() will forward
2641 * any unhandled interrupt on the offlined cpu to the new cpu
2642 * destination that is handling the corresponding interrupt. This
2643 * interrupt forwarding is done via IPI's. Hence, in this case also
2644 * level-triggered io-apic interrupt will be seen as an edge
2645 * interrupt in the IRR. And we can't rely on the cpu's EOI
2646 * to be broadcasted to the IO-APIC's which will clear the remoteIRR
2647 * corresponding to the level-triggered interrupt. Hence on IO-APIC's
2648 * supporting EOI register, we do an explicit EOI to clear the
2649 * remote IRR and on IO-APIC's which don't have an EOI register,
2650 * we use the above logic (mask+edge followed by unmask+level) from
2651 * Manfred Spraul to clear the remote IRR.
2653 cfg = desc->chip_data;
2655 v = apic_read(APIC_TMR + ((i & ~0x1f) >> 1));
2658 * We must acknowledge the irq before we move it or the acknowledge will
2659 * not propagate properly.
2664 * Tail end of clearing remote IRR bit (either by delivering the EOI
2665 * message via io-apic EOI register write or simulating it using
2666 * mask+edge followed by unnask+level logic) manually when the
2667 * level triggered interrupt is seen as the edge triggered interrupt
2670 if (!(v & (1 << (i & 0x1f)))) {
2671 atomic_inc(&irq_mis_count);
2673 eoi_ioapic_irq(desc);
2676 /* Now we can move and renable the irq */
2677 if (unlikely(do_unmask_irq)) {
2678 /* Only migrate the irq if the ack has been received.
2680 * On rare occasions the broadcast level triggered ack gets
2681 * delayed going to ioapics, and if we reprogram the
2682 * vector while Remote IRR is still set the irq will never
2685 * To prevent this scenario we read the Remote IRR bit
2686 * of the ioapic. This has two effects.
2687 * - On any sane system the read of the ioapic will
2688 * flush writes (and acks) going to the ioapic from
2690 * - We get to see if the ACK has actually been delivered.
2692 * Based on failed experiments of reprogramming the
2693 * ioapic entry from outside of irq context starting
2694 * with masking the ioapic entry and then polling until
2695 * Remote IRR was clear before reprogramming the
2696 * ioapic I don't trust the Remote IRR bit to be
2697 * completey accurate.
2699 * However there appears to be no other way to plug
2700 * this race, so if the Remote IRR bit is not
2701 * accurate and is causing problems then it is a hardware bug
2702 * and you can go talk to the chipset vendor about it.
2704 cfg = desc->chip_data;
2705 if (!io_apic_level_ack_pending(cfg))
2706 move_masked_irq(irq);
2707 unmask_IO_APIC_irq_desc(desc);
2711 #ifdef CONFIG_INTR_REMAP
2712 static void ir_ack_apic_edge(unsigned int irq)
2717 static void ir_ack_apic_level(unsigned int irq)
2719 struct irq_desc *desc = irq_to_desc(irq);
2722 eoi_ioapic_irq(desc);
2724 #endif /* CONFIG_INTR_REMAP */
2726 static struct irq_chip ioapic_chip __read_mostly = {
2728 .startup = startup_ioapic_irq,
2729 .mask = mask_IO_APIC_irq,
2730 .unmask = unmask_IO_APIC_irq,
2731 .ack = ack_apic_edge,
2732 .eoi = ack_apic_level,
2734 .set_affinity = set_ioapic_affinity_irq,
2736 .retrigger = ioapic_retrigger_irq,
2739 static struct irq_chip ir_ioapic_chip __read_mostly = {
2740 .name = "IR-IO-APIC",
2741 .startup = startup_ioapic_irq,
2742 .mask = mask_IO_APIC_irq,
2743 .unmask = unmask_IO_APIC_irq,
2744 #ifdef CONFIG_INTR_REMAP
2745 .ack = ir_ack_apic_edge,
2746 .eoi = ir_ack_apic_level,
2748 .set_affinity = set_ir_ioapic_affinity_irq,
2751 .retrigger = ioapic_retrigger_irq,
2754 static inline void init_IO_APIC_traps(void)
2757 struct irq_desc *desc;
2758 struct irq_cfg *cfg;
2761 * NOTE! The local APIC isn't very good at handling
2762 * multiple interrupts at the same interrupt level.
2763 * As the interrupt level is determined by taking the
2764 * vector number and shifting that right by 4, we
2765 * want to spread these out a bit so that they don't
2766 * all fall in the same interrupt level.
2768 * Also, we've got to be careful not to trash gate
2769 * 0x80, because int 0x80 is hm, kind of importantish. ;)
2771 for_each_irq_desc(irq, desc) {
2772 cfg = desc->chip_data;
2773 if (IO_APIC_IRQ(irq) && cfg && !cfg->vector) {
2775 * Hmm.. We don't have an entry for this,
2776 * so default to an old-fashioned 8259
2777 * interrupt if we can..
2779 if (irq < legacy_pic->nr_legacy_irqs)
2780 legacy_pic->make_irq(irq);
2782 /* Strange. Oh, well.. */
2783 desc->chip = &no_irq_chip;
2789 * The local APIC irq-chip implementation:
2792 static void mask_lapic_irq(unsigned int irq)
2796 v = apic_read(APIC_LVT0);
2797 apic_write(APIC_LVT0, v | APIC_LVT_MASKED);
2800 static void unmask_lapic_irq(unsigned int irq)
2804 v = apic_read(APIC_LVT0);
2805 apic_write(APIC_LVT0, v & ~APIC_LVT_MASKED);
2808 static void ack_lapic_irq(unsigned int irq)
2813 static struct irq_chip lapic_chip __read_mostly = {
2814 .name = "local-APIC",
2815 .mask = mask_lapic_irq,
2816 .unmask = unmask_lapic_irq,
2817 .ack = ack_lapic_irq,
2820 static void lapic_register_intr(int irq, struct irq_desc *desc)
2822 desc->status &= ~IRQ_LEVEL;
2823 set_irq_chip_and_handler_name(irq, &lapic_chip, handle_edge_irq,
2827 static void __init setup_nmi(void)
2830 * Dirty trick to enable the NMI watchdog ...
2831 * We put the 8259A master into AEOI mode and
2832 * unmask on all local APICs LVT0 as NMI.
2834 * The idea to use the 8259A in AEOI mode ('8259A Virtual Wire')
2835 * is from Maciej W. Rozycki - so we do not have to EOI from
2836 * the NMI handler or the timer interrupt.
2838 apic_printk(APIC_VERBOSE, KERN_INFO "activating NMI Watchdog ...");
2840 enable_NMI_through_LVT0();
2842 apic_printk(APIC_VERBOSE, " done.\n");
2846 * This looks a bit hackish but it's about the only one way of sending
2847 * a few INTA cycles to 8259As and any associated glue logic. ICR does
2848 * not support the ExtINT mode, unfortunately. We need to send these
2849 * cycles as some i82489DX-based boards have glue logic that keeps the
2850 * 8259A interrupt line asserted until INTA. --macro
2852 static inline void __init unlock_ExtINT_logic(void)
2855 struct IO_APIC_route_entry entry0, entry1;
2856 unsigned char save_control, save_freq_select;
2858 pin = find_isa_irq_pin(8, mp_INT);
2863 apic = find_isa_irq_apic(8, mp_INT);
2869 entry0 = ioapic_read_entry(apic, pin);
2870 clear_IO_APIC_pin(apic, pin);
2872 memset(&entry1, 0, sizeof(entry1));
2874 entry1.dest_mode = 0; /* physical delivery */
2875 entry1.mask = 0; /* unmask IRQ now */
2876 entry1.dest = hard_smp_processor_id();
2877 entry1.delivery_mode = dest_ExtINT;
2878 entry1.polarity = entry0.polarity;
2882 ioapic_write_entry(apic, pin, entry1);
2884 save_control = CMOS_READ(RTC_CONTROL);
2885 save_freq_select = CMOS_READ(RTC_FREQ_SELECT);
2886 CMOS_WRITE((save_freq_select & ~RTC_RATE_SELECT) | 0x6,
2888 CMOS_WRITE(save_control | RTC_PIE, RTC_CONTROL);
2893 if ((CMOS_READ(RTC_INTR_FLAGS) & RTC_PF) == RTC_PF)
2897 CMOS_WRITE(save_control, RTC_CONTROL);
2898 CMOS_WRITE(save_freq_select, RTC_FREQ_SELECT);
2899 clear_IO_APIC_pin(apic, pin);
2901 ioapic_write_entry(apic, pin, entry0);
2904 static int disable_timer_pin_1 __initdata;
2905 /* Actually the next is obsolete, but keep it for paranoid reasons -AK */
2906 static int __init disable_timer_pin_setup(char *arg)
2908 disable_timer_pin_1 = 1;
2911 early_param("disable_timer_pin_1", disable_timer_pin_setup);
2913 int timer_through_8259 __initdata;
2916 * This code may look a bit paranoid, but it's supposed to cooperate with
2917 * a wide range of boards and BIOS bugs. Fortunately only the timer IRQ
2918 * is so screwy. Thanks to Brian Perkins for testing/hacking this beast
2919 * fanatically on his truly buggy board.
2921 * FIXME: really need to revamp this for all platforms.
2923 static inline void __init check_timer(void)
2925 struct irq_desc *desc = irq_to_desc(0);
2926 struct irq_cfg *cfg = desc->chip_data;
2927 int node = cpu_to_node(boot_cpu_id);
2928 int apic1, pin1, apic2, pin2;
2929 unsigned long flags;
2932 local_irq_save(flags);
2935 * get/set the timer IRQ vector:
2937 legacy_pic->chip->mask(0);
2938 assign_irq_vector(0, cfg, apic->target_cpus());
2941 * As IRQ0 is to be enabled in the 8259A, the virtual
2942 * wire has to be disabled in the local APIC. Also
2943 * timer interrupts need to be acknowledged manually in
2944 * the 8259A for the i82489DX when using the NMI
2945 * watchdog as that APIC treats NMIs as level-triggered.
2946 * The AEOI mode will finish them in the 8259A
2949 apic_write(APIC_LVT0, APIC_LVT_MASKED | APIC_DM_EXTINT);
2950 legacy_pic->init(1);
2951 #ifdef CONFIG_X86_32
2955 ver = apic_read(APIC_LVR);
2956 ver = GET_APIC_VERSION(ver);
2957 timer_ack = (nmi_watchdog == NMI_IO_APIC && !APIC_INTEGRATED(ver));
2961 pin1 = find_isa_irq_pin(0, mp_INT);
2962 apic1 = find_isa_irq_apic(0, mp_INT);
2963 pin2 = ioapic_i8259.pin;
2964 apic2 = ioapic_i8259.apic;
2966 apic_printk(APIC_QUIET, KERN_INFO "..TIMER: vector=0x%02X "
2967 "apic1=%d pin1=%d apic2=%d pin2=%d\n",
2968 cfg->vector, apic1, pin1, apic2, pin2);
2971 * Some BIOS writers are clueless and report the ExtINTA
2972 * I/O APIC input from the cascaded 8259A as the timer
2973 * interrupt input. So just in case, if only one pin
2974 * was found above, try it both directly and through the
2978 if (intr_remapping_enabled)
2979 panic("BIOS bug: timer not connected to IO-APIC");
2983 } else if (pin2 == -1) {
2990 * Ok, does IRQ0 through the IOAPIC work?
2993 add_pin_to_irq_node(cfg, node, apic1, pin1);
2994 setup_timer_IRQ0_pin(apic1, pin1, cfg->vector);
2996 /* for edge trigger, setup_IO_APIC_irq already
2997 * leave it unmasked.
2998 * so only need to unmask if it is level-trigger
2999 * do we really have level trigger timer?
3002 idx = find_irq_entry(apic1, pin1, mp_INT);
3003 if (idx != -1 && irq_trigger(idx))
3004 unmask_IO_APIC_irq_desc(desc);
3006 if (timer_irq_works()) {
3007 if (nmi_watchdog == NMI_IO_APIC) {
3009 legacy_pic->chip->unmask(0);
3011 if (disable_timer_pin_1 > 0)
3012 clear_IO_APIC_pin(0, pin1);
3015 if (intr_remapping_enabled)
3016 panic("timer doesn't work through Interrupt-remapped IO-APIC");
3017 local_irq_disable();
3018 clear_IO_APIC_pin(apic1, pin1);
3020 apic_printk(APIC_QUIET, KERN_ERR "..MP-BIOS bug: "
3021 "8254 timer not connected to IO-APIC\n");
3023 apic_printk(APIC_QUIET, KERN_INFO "...trying to set up timer "
3024 "(IRQ0) through the 8259A ...\n");
3025 apic_printk(APIC_QUIET, KERN_INFO
3026 "..... (found apic %d pin %d) ...\n", apic2, pin2);
3028 * legacy devices should be connected to IO APIC #0
3030 replace_pin_at_irq_node(cfg, node, apic1, pin1, apic2, pin2);
3031 setup_timer_IRQ0_pin(apic2, pin2, cfg->vector);
3032 legacy_pic->chip->unmask(0);
3033 if (timer_irq_works()) {
3034 apic_printk(APIC_QUIET, KERN_INFO "....... works.\n");
3035 timer_through_8259 = 1;
3036 if (nmi_watchdog == NMI_IO_APIC) {
3037 legacy_pic->chip->mask(0);
3039 legacy_pic->chip->unmask(0);
3044 * Cleanup, just in case ...
3046 local_irq_disable();
3047 legacy_pic->chip->mask(0);
3048 clear_IO_APIC_pin(apic2, pin2);
3049 apic_printk(APIC_QUIET, KERN_INFO "....... failed.\n");
3052 if (nmi_watchdog == NMI_IO_APIC) {
3053 apic_printk(APIC_QUIET, KERN_WARNING "timer doesn't work "
3054 "through the IO-APIC - disabling NMI Watchdog!\n");
3055 nmi_watchdog = NMI_NONE;
3057 #ifdef CONFIG_X86_32
3061 apic_printk(APIC_QUIET, KERN_INFO
3062 "...trying to set up timer as Virtual Wire IRQ...\n");
3064 lapic_register_intr(0, desc);
3065 apic_write(APIC_LVT0, APIC_DM_FIXED | cfg->vector); /* Fixed mode */
3066 legacy_pic->chip->unmask(0);
3068 if (timer_irq_works()) {
3069 apic_printk(APIC_QUIET, KERN_INFO "..... works.\n");
3072 local_irq_disable();
3073 legacy_pic->chip->mask(0);
3074 apic_write(APIC_LVT0, APIC_LVT_MASKED | APIC_DM_FIXED | cfg->vector);
3075 apic_printk(APIC_QUIET, KERN_INFO "..... failed.\n");
3077 apic_printk(APIC_QUIET, KERN_INFO
3078 "...trying to set up timer as ExtINT IRQ...\n");
3080 legacy_pic->init(0);
3081 legacy_pic->make_irq(0);
3082 apic_write(APIC_LVT0, APIC_DM_EXTINT);
3084 unlock_ExtINT_logic();
3086 if (timer_irq_works()) {
3087 apic_printk(APIC_QUIET, KERN_INFO "..... works.\n");
3090 local_irq_disable();
3091 apic_printk(APIC_QUIET, KERN_INFO "..... failed :(.\n");
3092 panic("IO-APIC + timer doesn't work! Boot with apic=debug and send a "
3093 "report. Then try booting with the 'noapic' option.\n");
3095 local_irq_restore(flags);
3099 * Traditionally ISA IRQ2 is the cascade IRQ, and is not available
3100 * to devices. However there may be an I/O APIC pin available for
3101 * this interrupt regardless. The pin may be left unconnected, but
3102 * typically it will be reused as an ExtINT cascade interrupt for
3103 * the master 8259A. In the MPS case such a pin will normally be
3104 * reported as an ExtINT interrupt in the MP table. With ACPI
3105 * there is no provision for ExtINT interrupts, and in the absence
3106 * of an override it would be treated as an ordinary ISA I/O APIC
3107 * interrupt, that is edge-triggered and unmasked by default. We
3108 * used to do this, but it caused problems on some systems because
3109 * of the NMI watchdog and sometimes IRQ0 of the 8254 timer using
3110 * the same ExtINT cascade interrupt to drive the local APIC of the
3111 * bootstrap processor. Therefore we refrain from routing IRQ2 to
3112 * the I/O APIC in all cases now. No actual device should request
3113 * it anyway. --macro
3115 #define PIC_IRQS (1UL << PIC_CASCADE_IR)
3117 void __init setup_IO_APIC(void)
3121 * calling enable_IO_APIC() is moved to setup_local_APIC for BP
3123 io_apic_irqs = legacy_pic->nr_legacy_irqs ? ~PIC_IRQS : ~0UL;
3125 apic_printk(APIC_VERBOSE, "ENABLING IO-APIC IRQs\n");
3127 * Set up IO-APIC IRQ routing.
3129 x86_init.mpparse.setup_ioapic_ids();
3132 setup_IO_APIC_irqs();
3133 init_IO_APIC_traps();
3134 if (legacy_pic->nr_legacy_irqs)
3139 * Called after all the initialization is done. If we didnt find any
3140 * APIC bugs then we can allow the modify fast path
3143 static int __init io_apic_bug_finalize(void)
3145 if (sis_apic_bug == -1)
3150 late_initcall(io_apic_bug_finalize);
3152 struct sysfs_ioapic_data {
3153 struct sys_device dev;
3154 struct IO_APIC_route_entry entry[0];
3156 static struct sysfs_ioapic_data * mp_ioapic_data[MAX_IO_APICS];
3158 static int ioapic_suspend(struct sys_device *dev, pm_message_t state)
3160 struct IO_APIC_route_entry *entry;
3161 struct sysfs_ioapic_data *data;
3164 data = container_of(dev, struct sysfs_ioapic_data, dev);
3165 entry = data->entry;
3166 for (i = 0; i < nr_ioapic_registers[dev->id]; i ++, entry ++ )
3167 *entry = ioapic_read_entry(dev->id, i);
3172 static int ioapic_resume(struct sys_device *dev)
3174 struct IO_APIC_route_entry *entry;
3175 struct sysfs_ioapic_data *data;
3176 unsigned long flags;
3177 union IO_APIC_reg_00 reg_00;
3180 data = container_of(dev, struct sysfs_ioapic_data, dev);
3181 entry = data->entry;
3183 raw_spin_lock_irqsave(&ioapic_lock, flags);
3184 reg_00.raw = io_apic_read(dev->id, 0);
3185 if (reg_00.bits.ID != mp_ioapics[dev->id].apicid) {
3186 reg_00.bits.ID = mp_ioapics[dev->id].apicid;
3187 io_apic_write(dev->id, 0, reg_00.raw);
3189 raw_spin_unlock_irqrestore(&ioapic_lock, flags);
3190 for (i = 0; i < nr_ioapic_registers[dev->id]; i++)
3191 ioapic_write_entry(dev->id, i, entry[i]);
3196 static struct sysdev_class ioapic_sysdev_class = {
3198 .suspend = ioapic_suspend,
3199 .resume = ioapic_resume,
3202 static int __init ioapic_init_sysfs(void)
3204 struct sys_device * dev;
3207 error = sysdev_class_register(&ioapic_sysdev_class);
3211 for (i = 0; i < nr_ioapics; i++ ) {
3212 size = sizeof(struct sys_device) + nr_ioapic_registers[i]
3213 * sizeof(struct IO_APIC_route_entry);
3214 mp_ioapic_data[i] = kzalloc(size, GFP_KERNEL);
3215 if (!mp_ioapic_data[i]) {
3216 printk(KERN_ERR "Can't suspend/resume IOAPIC %d\n", i);
3219 dev = &mp_ioapic_data[i]->dev;
3221 dev->cls = &ioapic_sysdev_class;
3222 error = sysdev_register(dev);
3224 kfree(mp_ioapic_data[i]);
3225 mp_ioapic_data[i] = NULL;
3226 printk(KERN_ERR "Can't suspend/resume IOAPIC %d\n", i);
3234 device_initcall(ioapic_init_sysfs);
3237 * Dynamic irq allocate and deallocation
3239 unsigned int create_irq_nr(unsigned int irq_want, int node)
3241 /* Allocate an unused irq */
3244 unsigned long flags;
3245 struct irq_cfg *cfg_new = NULL;
3246 struct irq_desc *desc_new = NULL;
3249 if (irq_want < nr_irqs_gsi)
3250 irq_want = nr_irqs_gsi;
3252 raw_spin_lock_irqsave(&vector_lock, flags);
3253 for (new = irq_want; new < nr_irqs; new++) {
3254 desc_new = irq_to_desc_alloc_node(new, node);
3256 printk(KERN_INFO "can not get irq_desc for %d\n", new);
3259 cfg_new = desc_new->chip_data;
3261 if (cfg_new->vector != 0)
3264 desc_new = move_irq_desc(desc_new, node);
3265 cfg_new = desc_new->chip_data;
3267 if (__assign_irq_vector(new, cfg_new, apic->target_cpus()) == 0)
3271 raw_spin_unlock_irqrestore(&vector_lock, flags);
3274 dynamic_irq_init_keep_chip_data(irq);
3279 int create_irq(void)
3281 int node = cpu_to_node(boot_cpu_id);
3282 unsigned int irq_want;
3285 irq_want = nr_irqs_gsi;
3286 irq = create_irq_nr(irq_want, node);
3294 void destroy_irq(unsigned int irq)
3296 unsigned long flags;
3298 dynamic_irq_cleanup_keep_chip_data(irq);
3301 raw_spin_lock_irqsave(&vector_lock, flags);
3302 __clear_irq_vector(irq, get_irq_chip_data(irq));
3303 raw_spin_unlock_irqrestore(&vector_lock, flags);
3307 * MSI message composition
3309 #ifdef CONFIG_PCI_MSI
3310 static int msi_compose_msg(struct pci_dev *pdev, unsigned int irq,
3311 struct msi_msg *msg, u8 hpet_id)
3313 struct irq_cfg *cfg;
3321 err = assign_irq_vector(irq, cfg, apic->target_cpus());
3325 dest = apic->cpu_mask_to_apicid_and(cfg->domain, apic->target_cpus());
3327 if (irq_remapped(irq)) {
3332 ir_index = map_irq_to_irte_handle(irq, &sub_handle);
3333 BUG_ON(ir_index == -1);
3335 memset (&irte, 0, sizeof(irte));
3338 irte.dst_mode = apic->irq_dest_mode;
3339 irte.trigger_mode = 0; /* edge */
3340 irte.dlvry_mode = apic->irq_delivery_mode;
3341 irte.vector = cfg->vector;
3342 irte.dest_id = IRTE_DEST(dest);
3344 /* Set source-id of interrupt request */
3346 set_msi_sid(&irte, pdev);
3348 set_hpet_sid(&irte, hpet_id);
3350 modify_irte(irq, &irte);
3352 msg->address_hi = MSI_ADDR_BASE_HI;
3353 msg->data = sub_handle;
3354 msg->address_lo = MSI_ADDR_BASE_LO | MSI_ADDR_IR_EXT_INT |
3356 MSI_ADDR_IR_INDEX1(ir_index) |
3357 MSI_ADDR_IR_INDEX2(ir_index);
3359 if (x2apic_enabled())
3360 msg->address_hi = MSI_ADDR_BASE_HI |
3361 MSI_ADDR_EXT_DEST_ID(dest);
3363 msg->address_hi = MSI_ADDR_BASE_HI;
3367 ((apic->irq_dest_mode == 0) ?
3368 MSI_ADDR_DEST_MODE_PHYSICAL:
3369 MSI_ADDR_DEST_MODE_LOGICAL) |
3370 ((apic->irq_delivery_mode != dest_LowestPrio) ?
3371 MSI_ADDR_REDIRECTION_CPU:
3372 MSI_ADDR_REDIRECTION_LOWPRI) |
3373 MSI_ADDR_DEST_ID(dest);
3376 MSI_DATA_TRIGGER_EDGE |
3377 MSI_DATA_LEVEL_ASSERT |
3378 ((apic->irq_delivery_mode != dest_LowestPrio) ?
3379 MSI_DATA_DELIVERY_FIXED:
3380 MSI_DATA_DELIVERY_LOWPRI) |
3381 MSI_DATA_VECTOR(cfg->vector);
3387 static int set_msi_irq_affinity(unsigned int irq, const struct cpumask *mask)
3389 struct irq_desc *desc = irq_to_desc(irq);
3390 struct irq_cfg *cfg;
3394 if (set_desc_affinity(desc, mask, &dest))
3397 cfg = desc->chip_data;
3399 read_msi_msg_desc(desc, &msg);
3401 msg.data &= ~MSI_DATA_VECTOR_MASK;
3402 msg.data |= MSI_DATA_VECTOR(cfg->vector);
3403 msg.address_lo &= ~MSI_ADDR_DEST_ID_MASK;
3404 msg.address_lo |= MSI_ADDR_DEST_ID(dest);
3406 write_msi_msg_desc(desc, &msg);
3410 #ifdef CONFIG_INTR_REMAP
3412 * Migrate the MSI irq to another cpumask. This migration is
3413 * done in the process context using interrupt-remapping hardware.
3416 ir_set_msi_irq_affinity(unsigned int irq, const struct cpumask *mask)
3418 struct irq_desc *desc = irq_to_desc(irq);
3419 struct irq_cfg *cfg = desc->chip_data;
3423 if (get_irte(irq, &irte))
3426 if (set_desc_affinity(desc, mask, &dest))
3429 irte.vector = cfg->vector;
3430 irte.dest_id = IRTE_DEST(dest);
3433 * atomically update the IRTE with the new destination and vector.
3435 modify_irte(irq, &irte);
3438 * After this point, all the interrupts will start arriving
3439 * at the new destination. So, time to cleanup the previous
3440 * vector allocation.
3442 if (cfg->move_in_progress)
3443 send_cleanup_vector(cfg);
3449 #endif /* CONFIG_SMP */
3452 * IRQ Chip for MSI PCI/PCI-X/PCI-Express Devices,
3453 * which implement the MSI or MSI-X Capability Structure.
3455 static struct irq_chip msi_chip = {
3457 .unmask = unmask_msi_irq,
3458 .mask = mask_msi_irq,
3459 .ack = ack_apic_edge,
3461 .set_affinity = set_msi_irq_affinity,
3463 .retrigger = ioapic_retrigger_irq,
3466 static struct irq_chip msi_ir_chip = {
3467 .name = "IR-PCI-MSI",
3468 .unmask = unmask_msi_irq,
3469 .mask = mask_msi_irq,
3470 #ifdef CONFIG_INTR_REMAP
3471 .ack = ir_ack_apic_edge,
3473 .set_affinity = ir_set_msi_irq_affinity,
3476 .retrigger = ioapic_retrigger_irq,
3480 * Map the PCI dev to the corresponding remapping hardware unit
3481 * and allocate 'nvec' consecutive interrupt-remapping table entries
3484 static int msi_alloc_irte(struct pci_dev *dev, int irq, int nvec)
3486 struct intel_iommu *iommu;
3489 iommu = map_dev_to_ir(dev);
3492 "Unable to map PCI %s to iommu\n", pci_name(dev));
3496 index = alloc_irte(iommu, irq, nvec);
3499 "Unable to allocate %d IRTE for PCI %s\n", nvec,
3506 static int setup_msi_irq(struct pci_dev *dev, struct msi_desc *msidesc, int irq)
3511 ret = msi_compose_msg(dev, irq, &msg, -1);
3515 set_irq_msi(irq, msidesc);
3516 write_msi_msg(irq, &msg);
3518 if (irq_remapped(irq)) {
3519 struct irq_desc *desc = irq_to_desc(irq);
3521 * irq migration in process context
3523 desc->status |= IRQ_MOVE_PCNTXT;
3524 set_irq_chip_and_handler_name(irq, &msi_ir_chip, handle_edge_irq, "edge");
3526 set_irq_chip_and_handler_name(irq, &msi_chip, handle_edge_irq, "edge");
3528 dev_printk(KERN_DEBUG, &dev->dev, "irq %d for MSI/MSI-X\n", irq);
3533 int arch_setup_msi_irqs(struct pci_dev *dev, int nvec, int type)
3536 int ret, sub_handle;
3537 struct msi_desc *msidesc;
3538 unsigned int irq_want;
3539 struct intel_iommu *iommu = NULL;
3543 /* x86 doesn't support multiple MSI yet */
3544 if (type == PCI_CAP_ID_MSI && nvec > 1)
3547 node = dev_to_node(&dev->dev);
3548 irq_want = nr_irqs_gsi;
3550 list_for_each_entry(msidesc, &dev->msi_list, list) {
3551 irq = create_irq_nr(irq_want, node);
3555 if (!intr_remapping_enabled)
3560 * allocate the consecutive block of IRTE's
3563 index = msi_alloc_irte(dev, irq, nvec);
3569 iommu = map_dev_to_ir(dev);
3575 * setup the mapping between the irq and the IRTE
3576 * base index, the sub_handle pointing to the
3577 * appropriate interrupt remap table entry.
3579 set_irte_irq(irq, iommu, index, sub_handle);
3582 ret = setup_msi_irq(dev, msidesc, irq);
3594 void arch_teardown_msi_irq(unsigned int irq)
3599 #if defined (CONFIG_DMAR) || defined (CONFIG_INTR_REMAP)
3601 static int dmar_msi_set_affinity(unsigned int irq, const struct cpumask *mask)
3603 struct irq_desc *desc = irq_to_desc(irq);
3604 struct irq_cfg *cfg;
3608 if (set_desc_affinity(desc, mask, &dest))
3611 cfg = desc->chip_data;
3613 dmar_msi_read(irq, &msg);
3615 msg.data &= ~MSI_DATA_VECTOR_MASK;
3616 msg.data |= MSI_DATA_VECTOR(cfg->vector);
3617 msg.address_lo &= ~MSI_ADDR_DEST_ID_MASK;
3618 msg.address_lo |= MSI_ADDR_DEST_ID(dest);
3620 dmar_msi_write(irq, &msg);
3625 #endif /* CONFIG_SMP */
3627 static struct irq_chip dmar_msi_type = {
3629 .unmask = dmar_msi_unmask,
3630 .mask = dmar_msi_mask,
3631 .ack = ack_apic_edge,
3633 .set_affinity = dmar_msi_set_affinity,
3635 .retrigger = ioapic_retrigger_irq,
3638 int arch_setup_dmar_msi(unsigned int irq)
3643 ret = msi_compose_msg(NULL, irq, &msg, -1);
3646 dmar_msi_write(irq, &msg);
3647 set_irq_chip_and_handler_name(irq, &dmar_msi_type, handle_edge_irq,
3653 #ifdef CONFIG_HPET_TIMER
3656 static int hpet_msi_set_affinity(unsigned int irq, const struct cpumask *mask)
3658 struct irq_desc *desc = irq_to_desc(irq);
3659 struct irq_cfg *cfg;
3663 if (set_desc_affinity(desc, mask, &dest))
3666 cfg = desc->chip_data;
3668 hpet_msi_read(irq, &msg);
3670 msg.data &= ~MSI_DATA_VECTOR_MASK;
3671 msg.data |= MSI_DATA_VECTOR(cfg->vector);
3672 msg.address_lo &= ~MSI_ADDR_DEST_ID_MASK;
3673 msg.address_lo |= MSI_ADDR_DEST_ID(dest);
3675 hpet_msi_write(irq, &msg);
3680 #endif /* CONFIG_SMP */
3682 static struct irq_chip ir_hpet_msi_type = {
3683 .name = "IR-HPET_MSI",
3684 .unmask = hpet_msi_unmask,
3685 .mask = hpet_msi_mask,
3686 #ifdef CONFIG_INTR_REMAP
3687 .ack = ir_ack_apic_edge,
3689 .set_affinity = ir_set_msi_irq_affinity,
3692 .retrigger = ioapic_retrigger_irq,
3695 static struct irq_chip hpet_msi_type = {
3697 .unmask = hpet_msi_unmask,
3698 .mask = hpet_msi_mask,
3699 .ack = ack_apic_edge,
3701 .set_affinity = hpet_msi_set_affinity,
3703 .retrigger = ioapic_retrigger_irq,
3706 int arch_setup_hpet_msi(unsigned int irq, unsigned int id)
3710 struct irq_desc *desc = irq_to_desc(irq);
3712 if (intr_remapping_enabled) {
3713 struct intel_iommu *iommu = map_hpet_to_ir(id);
3719 index = alloc_irte(iommu, irq, 1);
3724 ret = msi_compose_msg(NULL, irq, &msg, id);
3728 hpet_msi_write(irq, &msg);
3729 desc->status |= IRQ_MOVE_PCNTXT;
3730 if (irq_remapped(irq))
3731 set_irq_chip_and_handler_name(irq, &ir_hpet_msi_type,
3732 handle_edge_irq, "edge");
3734 set_irq_chip_and_handler_name(irq, &hpet_msi_type,
3735 handle_edge_irq, "edge");
3741 #endif /* CONFIG_PCI_MSI */
3743 * Hypertransport interrupt support
3745 #ifdef CONFIG_HT_IRQ
3749 static void target_ht_irq(unsigned int irq, unsigned int dest, u8 vector)
3751 struct ht_irq_msg msg;
3752 fetch_ht_irq_msg(irq, &msg);
3754 msg.address_lo &= ~(HT_IRQ_LOW_VECTOR_MASK | HT_IRQ_LOW_DEST_ID_MASK);
3755 msg.address_hi &= ~(HT_IRQ_HIGH_DEST_ID_MASK);
3757 msg.address_lo |= HT_IRQ_LOW_VECTOR(vector) | HT_IRQ_LOW_DEST_ID(dest);
3758 msg.address_hi |= HT_IRQ_HIGH_DEST_ID(dest);
3760 write_ht_irq_msg(irq, &msg);
3763 static int set_ht_irq_affinity(unsigned int irq, const struct cpumask *mask)
3765 struct irq_desc *desc = irq_to_desc(irq);
3766 struct irq_cfg *cfg;
3769 if (set_desc_affinity(desc, mask, &dest))
3772 cfg = desc->chip_data;
3774 target_ht_irq(irq, dest, cfg->vector);
3781 static struct irq_chip ht_irq_chip = {
3783 .mask = mask_ht_irq,
3784 .unmask = unmask_ht_irq,
3785 .ack = ack_apic_edge,
3787 .set_affinity = set_ht_irq_affinity,
3789 .retrigger = ioapic_retrigger_irq,
3792 int arch_setup_ht_irq(unsigned int irq, struct pci_dev *dev)
3794 struct irq_cfg *cfg;
3801 err = assign_irq_vector(irq, cfg, apic->target_cpus());
3803 struct ht_irq_msg msg;
3806 dest = apic->cpu_mask_to_apicid_and(cfg->domain,
3807 apic->target_cpus());
3809 msg.address_hi = HT_IRQ_HIGH_DEST_ID(dest);
3813 HT_IRQ_LOW_DEST_ID(dest) |
3814 HT_IRQ_LOW_VECTOR(cfg->vector) |
3815 ((apic->irq_dest_mode == 0) ?
3816 HT_IRQ_LOW_DM_PHYSICAL :
3817 HT_IRQ_LOW_DM_LOGICAL) |
3818 HT_IRQ_LOW_RQEOI_EDGE |
3819 ((apic->irq_delivery_mode != dest_LowestPrio) ?
3820 HT_IRQ_LOW_MT_FIXED :
3821 HT_IRQ_LOW_MT_ARBITRATED) |
3822 HT_IRQ_LOW_IRQ_MASKED;
3824 write_ht_irq_msg(irq, &msg);
3826 set_irq_chip_and_handler_name(irq, &ht_irq_chip,
3827 handle_edge_irq, "edge");
3829 dev_printk(KERN_DEBUG, &dev->dev, "irq %d for HT\n", irq);
3833 #endif /* CONFIG_HT_IRQ */
3835 int __init io_apic_get_redir_entries (int ioapic)
3837 union IO_APIC_reg_01 reg_01;
3838 unsigned long flags;
3840 raw_spin_lock_irqsave(&ioapic_lock, flags);
3841 reg_01.raw = io_apic_read(ioapic, 1);
3842 raw_spin_unlock_irqrestore(&ioapic_lock, flags);
3844 /* The register returns the maximum index redir index
3845 * supported, which is one less than the total number of redir
3848 return reg_01.bits.entries + 1;
3851 void __init probe_nr_irqs_gsi(void)
3856 if (nr > nr_irqs_gsi)
3859 printk(KERN_DEBUG "nr_irqs_gsi: %d\n", nr_irqs_gsi);
3862 #ifdef CONFIG_SPARSE_IRQ
3863 int __init arch_probe_nr_irqs(void)
3867 if (nr_irqs > (NR_VECTORS * nr_cpu_ids))
3868 nr_irqs = NR_VECTORS * nr_cpu_ids;
3870 nr = nr_irqs_gsi + 8 * nr_cpu_ids;
3871 #if defined(CONFIG_PCI_MSI) || defined(CONFIG_HT_IRQ)
3873 * for MSI and HT dyn irq
3875 nr += nr_irqs_gsi * 16;
3884 static int __io_apic_set_pci_routing(struct device *dev, int irq,
3885 struct io_apic_irq_attr *irq_attr)
3887 struct irq_desc *desc;
3888 struct irq_cfg *cfg;
3891 int trigger, polarity;
3893 ioapic = irq_attr->ioapic;
3894 if (!IO_APIC_IRQ(irq)) {
3895 apic_printk(APIC_QUIET,KERN_ERR "IOAPIC[%d]: Invalid reference to IRQ 0\n",
3901 node = dev_to_node(dev);
3903 node = cpu_to_node(boot_cpu_id);
3905 desc = irq_to_desc_alloc_node(irq, node);
3907 printk(KERN_INFO "can not get irq_desc %d\n", irq);
3911 pin = irq_attr->ioapic_pin;
3912 trigger = irq_attr->trigger;
3913 polarity = irq_attr->polarity;
3916 * IRQs < 16 are already in the irq_2_pin[] map
3918 if (irq >= legacy_pic->nr_legacy_irqs) {
3919 cfg = desc->chip_data;
3920 if (add_pin_to_irq_node_nopanic(cfg, node, ioapic, pin)) {
3921 printk(KERN_INFO "can not add pin %d for irq %d\n",
3927 setup_IO_APIC_irq(ioapic, pin, irq, desc, trigger, polarity);
3932 int io_apic_set_pci_routing(struct device *dev, int irq,
3933 struct io_apic_irq_attr *irq_attr)
3937 * Avoid pin reprogramming. PRTs typically include entries
3938 * with redundant pin->gsi mappings (but unique PCI devices);
3939 * we only program the IOAPIC on the first.
3941 ioapic = irq_attr->ioapic;
3942 pin = irq_attr->ioapic_pin;
3943 if (test_bit(pin, mp_ioapic_routing[ioapic].pin_programmed)) {
3944 pr_debug("Pin %d-%d already programmed\n",
3945 mp_ioapics[ioapic].apicid, pin);
3948 set_bit(pin, mp_ioapic_routing[ioapic].pin_programmed);
3950 return __io_apic_set_pci_routing(dev, irq, irq_attr);
3953 u8 __init io_apic_unique_id(u8 id)
3955 #ifdef CONFIG_X86_32
3956 if ((boot_cpu_data.x86_vendor == X86_VENDOR_INTEL) &&
3957 !APIC_XAPIC(apic_version[boot_cpu_physical_apicid]))
3958 return io_apic_get_unique_id(nr_ioapics, id);
3963 DECLARE_BITMAP(used, 256);
3965 bitmap_zero(used, 256);
3966 for (i = 0; i < nr_ioapics; i++) {
3967 struct mpc_ioapic *ia = &mp_ioapics[i];
3968 __set_bit(ia->apicid, used);
3970 if (!test_bit(id, used))
3972 return find_first_zero_bit(used, 256);
3976 #ifdef CONFIG_X86_32
3977 int __init io_apic_get_unique_id(int ioapic, int apic_id)
3979 union IO_APIC_reg_00 reg_00;
3980 static physid_mask_t apic_id_map = PHYSID_MASK_NONE;
3982 unsigned long flags;
3986 * The P4 platform supports up to 256 APIC IDs on two separate APIC
3987 * buses (one for LAPICs, one for IOAPICs), where predecessors only
3988 * supports up to 16 on one shared APIC bus.
3990 * TBD: Expand LAPIC/IOAPIC support on P4-class systems to take full
3991 * advantage of new APIC bus architecture.
3994 if (physids_empty(apic_id_map))
3995 apic->ioapic_phys_id_map(&phys_cpu_present_map, &apic_id_map);
3997 raw_spin_lock_irqsave(&ioapic_lock, flags);
3998 reg_00.raw = io_apic_read(ioapic, 0);
3999 raw_spin_unlock_irqrestore(&ioapic_lock, flags);
4001 if (apic_id >= get_physical_broadcast()) {
4002 printk(KERN_WARNING "IOAPIC[%d]: Invalid apic_id %d, trying "
4003 "%d\n", ioapic, apic_id, reg_00.bits.ID);
4004 apic_id = reg_00.bits.ID;
4008 * Every APIC in a system must have a unique ID or we get lots of nice
4009 * 'stuck on smp_invalidate_needed IPI wait' messages.
4011 if (apic->check_apicid_used(&apic_id_map, apic_id)) {
4013 for (i = 0; i < get_physical_broadcast(); i++) {
4014 if (!apic->check_apicid_used(&apic_id_map, i))
4018 if (i == get_physical_broadcast())
4019 panic("Max apic_id exceeded!\n");
4021 printk(KERN_WARNING "IOAPIC[%d]: apic_id %d already used, "
4022 "trying %d\n", ioapic, apic_id, i);
4027 apic->apicid_to_cpu_present(apic_id, &tmp);
4028 physids_or(apic_id_map, apic_id_map, tmp);
4030 if (reg_00.bits.ID != apic_id) {
4031 reg_00.bits.ID = apic_id;
4033 raw_spin_lock_irqsave(&ioapic_lock, flags);
4034 io_apic_write(ioapic, 0, reg_00.raw);
4035 reg_00.raw = io_apic_read(ioapic, 0);
4036 raw_spin_unlock_irqrestore(&ioapic_lock, flags);
4039 if (reg_00.bits.ID != apic_id) {
4040 printk("IOAPIC[%d]: Unable to change apic_id!\n", ioapic);
4045 apic_printk(APIC_VERBOSE, KERN_INFO
4046 "IOAPIC[%d]: Assigned apic_id %d\n", ioapic, apic_id);
4052 int __init io_apic_get_version(int ioapic)
4054 union IO_APIC_reg_01 reg_01;
4055 unsigned long flags;
4057 raw_spin_lock_irqsave(&ioapic_lock, flags);
4058 reg_01.raw = io_apic_read(ioapic, 1);
4059 raw_spin_unlock_irqrestore(&ioapic_lock, flags);
4061 return reg_01.bits.version;
4064 int acpi_get_override_irq(u32 gsi, int *trigger, int *polarity)
4066 int ioapic, pin, idx;
4068 if (skip_ioapic_setup)
4071 ioapic = mp_find_ioapic(gsi);
4075 pin = mp_find_ioapic_pin(ioapic, gsi);
4079 idx = find_irq_entry(ioapic, pin, mp_INT);
4083 *trigger = irq_trigger(idx);
4084 *polarity = irq_polarity(idx);
4089 * This function currently is only a helper for the i386 smp boot process where
4090 * we need to reprogram the ioredtbls to cater for the cpus which have come online
4091 * so mask in all cases should simply be apic->target_cpus()
4094 void __init setup_ioapic_dest(void)
4096 int pin, ioapic, irq, irq_entry;
4097 struct irq_desc *desc;
4098 const struct cpumask *mask;
4100 if (skip_ioapic_setup == 1)
4103 for (ioapic = 0; ioapic < nr_ioapics; ioapic++)
4104 for (pin = 0; pin < nr_ioapic_registers[ioapic]; pin++) {
4105 irq_entry = find_irq_entry(ioapic, pin, mp_INT);
4106 if (irq_entry == -1)
4108 irq = pin_2_irq(irq_entry, ioapic, pin);
4110 if ((ioapic > 0) && (irq > 16))
4113 desc = irq_to_desc(irq);
4116 * Honour affinities which have been set in early boot
4119 (IRQ_NO_BALANCING | IRQ_AFFINITY_SET))
4120 mask = desc->affinity;
4122 mask = apic->target_cpus();
4124 if (intr_remapping_enabled)
4125 set_ir_ioapic_affinity_irq_desc(desc, mask);
4127 set_ioapic_affinity_irq_desc(desc, mask);
4133 #define IOAPIC_RESOURCE_NAME_SIZE 11
4135 static struct resource *ioapic_resources;
4137 static struct resource * __init ioapic_setup_resources(int nr_ioapics)
4140 struct resource *res;
4144 if (nr_ioapics <= 0)
4147 n = IOAPIC_RESOURCE_NAME_SIZE + sizeof(struct resource);
4150 mem = alloc_bootmem(n);
4153 mem += sizeof(struct resource) * nr_ioapics;
4155 for (i = 0; i < nr_ioapics; i++) {
4157 res[i].flags = IORESOURCE_MEM | IORESOURCE_BUSY;
4158 snprintf(mem, IOAPIC_RESOURCE_NAME_SIZE, "IOAPIC %u", i);
4159 mem += IOAPIC_RESOURCE_NAME_SIZE;
4162 ioapic_resources = res;
4167 void __init ioapic_init_mappings(void)
4169 unsigned long ioapic_phys, idx = FIX_IO_APIC_BASE_0;
4170 struct resource *ioapic_res;
4173 ioapic_res = ioapic_setup_resources(nr_ioapics);
4174 for (i = 0; i < nr_ioapics; i++) {
4175 if (smp_found_config) {
4176 ioapic_phys = mp_ioapics[i].apicaddr;
4177 #ifdef CONFIG_X86_32
4180 "WARNING: bogus zero IO-APIC "
4181 "address found in MPTABLE, "
4182 "disabling IO/APIC support!\n");
4183 smp_found_config = 0;
4184 skip_ioapic_setup = 1;
4185 goto fake_ioapic_page;
4189 #ifdef CONFIG_X86_32
4192 ioapic_phys = (unsigned long)alloc_bootmem_pages(PAGE_SIZE);
4193 ioapic_phys = __pa(ioapic_phys);
4195 set_fixmap_nocache(idx, ioapic_phys);
4196 apic_printk(APIC_VERBOSE, "mapped IOAPIC to %08lx (%08lx)\n",
4197 __fix_to_virt(idx) + (ioapic_phys & ~PAGE_MASK),
4201 ioapic_res->start = ioapic_phys;
4202 ioapic_res->end = ioapic_phys + IO_APIC_SLOT_SIZE - 1;
4207 void __init ioapic_insert_resources(void)
4210 struct resource *r = ioapic_resources;
4215 "IO APIC resources couldn't be allocated.\n");
4219 for (i = 0; i < nr_ioapics; i++) {
4220 insert_resource(&iomem_resource, r);
4225 int mp_find_ioapic(u32 gsi)
4229 /* Find the IOAPIC that manages this GSI. */
4230 for (i = 0; i < nr_ioapics; i++) {
4231 if ((gsi >= mp_gsi_routing[i].gsi_base)
4232 && (gsi <= mp_gsi_routing[i].gsi_end))
4236 printk(KERN_ERR "ERROR: Unable to locate IOAPIC for GSI %d\n", gsi);
4240 int mp_find_ioapic_pin(int ioapic, u32 gsi)
4242 if (WARN_ON(ioapic == -1))
4244 if (WARN_ON(gsi > mp_gsi_routing[ioapic].gsi_end))
4247 return gsi - mp_gsi_routing[ioapic].gsi_base;
4250 static int bad_ioapic(unsigned long address)
4252 if (nr_ioapics >= MAX_IO_APICS) {
4253 printk(KERN_WARNING "WARING: Max # of I/O APICs (%d) exceeded "
4254 "(found %d), skipping\n", MAX_IO_APICS, nr_ioapics);
4258 printk(KERN_WARNING "WARNING: Bogus (zero) I/O APIC address"
4259 " found in table, skipping!\n");
4265 void __init mp_register_ioapic(int id, u32 address, u32 gsi_base)
4270 if (bad_ioapic(address))
4275 mp_ioapics[idx].type = MP_IOAPIC;
4276 mp_ioapics[idx].flags = MPC_APIC_USABLE;
4277 mp_ioapics[idx].apicaddr = address;
4279 set_fixmap_nocache(FIX_IO_APIC_BASE_0 + idx, address);
4280 mp_ioapics[idx].apicid = io_apic_unique_id(id);
4281 mp_ioapics[idx].apicver = io_apic_get_version(idx);
4284 * Build basic GSI lookup table to facilitate gsi->io_apic lookups
4285 * and to prevent reprogramming of IOAPIC pins (PCI GSIs).
4287 entries = io_apic_get_redir_entries(idx);
4288 mp_gsi_routing[idx].gsi_base = gsi_base;
4289 mp_gsi_routing[idx].gsi_end = gsi_base + entries - 1;
4292 * The number of IO-APIC IRQ registers (== #pins):
4294 nr_ioapic_registers[idx] = entries;
4296 if (mp_gsi_routing[idx].gsi_end > gsi_end)
4297 gsi_end = mp_gsi_routing[idx].gsi_end;
4299 printk(KERN_INFO "IOAPIC[%d]: apic_id %d, version %d, address 0x%x, "
4300 "GSI %d-%d\n", idx, mp_ioapics[idx].apicid,
4301 mp_ioapics[idx].apicver, mp_ioapics[idx].apicaddr,
4302 mp_gsi_routing[idx].gsi_base, mp_gsi_routing[idx].gsi_end);
4307 /* Enable IOAPIC early just for system timer */
4308 void __init pre_init_apic_IRQ0(void)
4310 struct irq_cfg *cfg;
4311 struct irq_desc *desc;
4313 printk(KERN_INFO "Early APIC setup for system timer0\n");
4315 phys_cpu_present_map = physid_mask_of_physid(boot_cpu_physical_apicid);
4317 desc = irq_to_desc_alloc_node(0, 0);
4322 add_pin_to_irq_node(cfg, 0, 0, 0);
4323 set_irq_chip_and_handler_name(0, &ioapic_chip, handle_edge_irq, "edge");
4325 setup_IO_APIC_irq(0, 0, 0, desc, 0, 0);