2 * Intel IO-APIC support for multi-Pentium hosts.
4 * Copyright (C) 1997, 1998, 1999, 2000, 2009 Ingo Molnar, Hajnalka Szabo
6 * Many thanks to Stig Venaas for trying out countless experimental
7 * patches and reporting/debugging problems patiently!
9 * (c) 1999, Multiple IO-APIC support, developed by
10 * Ken-ichi Yaku <yaku@css1.kbnes.nec.co.jp> and
11 * Hidemi Kishimoto <kisimoto@css1.kbnes.nec.co.jp>,
12 * further tested and cleaned up by Zach Brown <zab@redhat.com>
13 * and Ingo Molnar <mingo@redhat.com>
16 * Maciej W. Rozycki : Bits for genuine 82489DX APICs;
17 * thanks to Eric Gilmore
19 * for testing these extensively
20 * Paul Diefenbaugh : Added full ACPI support
24 #include <linux/interrupt.h>
25 #include <linux/init.h>
26 #include <linux/delay.h>
27 #include <linux/sched.h>
28 #include <linux/pci.h>
29 #include <linux/mc146818rtc.h>
30 #include <linux/compiler.h>
31 #include <linux/acpi.h>
32 #include <linux/module.h>
33 #include <linux/sysdev.h>
34 #include <linux/msi.h>
35 #include <linux/htirq.h>
36 #include <linux/freezer.h>
37 #include <linux/kthread.h>
38 #include <linux/jiffies.h> /* time_after() */
40 #include <acpi/acpi_bus.h>
42 #include <linux/bootmem.h>
43 #include <linux/dmar.h>
44 #include <linux/hpet.h>
51 #include <asm/proto.h>
54 #include <asm/timer.h>
55 #include <asm/i8259.h>
57 #include <asm/msidef.h>
58 #include <asm/hypertransport.h>
59 #include <asm/setup.h>
60 #include <asm/irq_remapping.h>
62 #include <asm/uv/uv_hub.h>
63 #include <asm/uv/uv_irq.h>
67 #define __apicdebuginit(type) static type __init
70 * Is the SiS APIC rmw bug present ?
71 * -1 = don't know, 0 = no, 1 = yes
73 int sis_apic_bug = -1;
75 static DEFINE_SPINLOCK(ioapic_lock);
76 static DEFINE_SPINLOCK(vector_lock);
79 * # of IRQ routing registers
81 int nr_ioapic_registers[MAX_IO_APICS];
83 /* I/O APIC entries */
84 struct mpc_ioapic mp_ioapics[MAX_IO_APICS];
87 /* MP IRQ source entries */
88 struct mpc_intsrc mp_irqs[MAX_IRQ_SOURCES];
90 /* # of MP IRQ source entries */
93 #if defined (CONFIG_MCA) || defined (CONFIG_EISA)
94 int mp_bus_id_to_type[MAX_MP_BUSSES];
97 DECLARE_BITMAP(mp_bus_not_pci, MAX_MP_BUSSES);
99 int skip_ioapic_setup;
101 void arch_disable_smp_support(void)
105 noioapicreroute = -1;
107 skip_ioapic_setup = 1;
110 static int __init parse_noapic(char *str)
112 /* disable IO-APIC */
113 arch_disable_smp_support();
116 early_param("noapic", parse_noapic);
121 * This is performance-critical, we want to do it O(1)
123 * the indexing order of this array favors 1:1 mappings
124 * between pins and IRQs.
127 struct irq_pin_list {
129 struct irq_pin_list *next;
132 static struct irq_pin_list *get_one_free_irq_2_pin(int node)
134 struct irq_pin_list *pin;
136 pin = kzalloc_node(sizeof(*pin), GFP_ATOMIC, node);
142 struct irq_pin_list *irq_2_pin;
143 cpumask_var_t domain;
144 cpumask_var_t old_domain;
145 unsigned move_cleanup_count;
147 u8 move_in_progress : 1;
150 /* irq_cfg is indexed by the sum of all RTEs in all I/O APICs. */
151 #ifdef CONFIG_SPARSE_IRQ
152 static struct irq_cfg irq_cfgx[] = {
154 static struct irq_cfg irq_cfgx[NR_IRQS] = {
156 [0] = { .vector = IRQ0_VECTOR, },
157 [1] = { .vector = IRQ1_VECTOR, },
158 [2] = { .vector = IRQ2_VECTOR, },
159 [3] = { .vector = IRQ3_VECTOR, },
160 [4] = { .vector = IRQ4_VECTOR, },
161 [5] = { .vector = IRQ5_VECTOR, },
162 [6] = { .vector = IRQ6_VECTOR, },
163 [7] = { .vector = IRQ7_VECTOR, },
164 [8] = { .vector = IRQ8_VECTOR, },
165 [9] = { .vector = IRQ9_VECTOR, },
166 [10] = { .vector = IRQ10_VECTOR, },
167 [11] = { .vector = IRQ11_VECTOR, },
168 [12] = { .vector = IRQ12_VECTOR, },
169 [13] = { .vector = IRQ13_VECTOR, },
170 [14] = { .vector = IRQ14_VECTOR, },
171 [15] = { .vector = IRQ15_VECTOR, },
174 int __init arch_early_irq_init(void)
177 struct irq_desc *desc;
182 count = ARRAY_SIZE(irq_cfgx);
184 for (i = 0; i < count; i++) {
185 desc = irq_to_desc(i);
186 desc->chip_data = &cfg[i];
187 alloc_bootmem_cpumask_var(&cfg[i].domain);
188 alloc_bootmem_cpumask_var(&cfg[i].old_domain);
189 if (i < NR_IRQS_LEGACY)
190 cpumask_setall(cfg[i].domain);
196 #ifdef CONFIG_SPARSE_IRQ
197 static struct irq_cfg *irq_cfg(unsigned int irq)
199 struct irq_cfg *cfg = NULL;
200 struct irq_desc *desc;
202 desc = irq_to_desc(irq);
204 cfg = desc->chip_data;
209 static struct irq_cfg *get_one_free_irq_cfg(int node)
213 cfg = kzalloc_node(sizeof(*cfg), GFP_ATOMIC, node);
215 if (!alloc_cpumask_var_node(&cfg->domain, GFP_ATOMIC, node)) {
218 } else if (!alloc_cpumask_var_node(&cfg->old_domain,
220 free_cpumask_var(cfg->domain);
224 cpumask_clear(cfg->domain);
225 cpumask_clear(cfg->old_domain);
232 int arch_init_chip_data(struct irq_desc *desc, int node)
236 cfg = desc->chip_data;
238 desc->chip_data = get_one_free_irq_cfg(node);
239 if (!desc->chip_data) {
240 printk(KERN_ERR "can not alloc irq_cfg\n");
248 /* for move_irq_desc */
250 init_copy_irq_2_pin(struct irq_cfg *old_cfg, struct irq_cfg *cfg, int node)
252 struct irq_pin_list *old_entry, *head, *tail, *entry;
254 cfg->irq_2_pin = NULL;
255 old_entry = old_cfg->irq_2_pin;
259 entry = get_one_free_irq_2_pin(node);
263 entry->apic = old_entry->apic;
264 entry->pin = old_entry->pin;
267 old_entry = old_entry->next;
269 entry = get_one_free_irq_2_pin(node);
277 /* still use the old one */
280 entry->apic = old_entry->apic;
281 entry->pin = old_entry->pin;
284 old_entry = old_entry->next;
288 cfg->irq_2_pin = head;
291 static void free_irq_2_pin(struct irq_cfg *old_cfg, struct irq_cfg *cfg)
293 struct irq_pin_list *entry, *next;
295 if (old_cfg->irq_2_pin == cfg->irq_2_pin)
298 entry = old_cfg->irq_2_pin;
305 old_cfg->irq_2_pin = NULL;
308 void arch_init_copy_chip_data(struct irq_desc *old_desc,
309 struct irq_desc *desc, int node)
312 struct irq_cfg *old_cfg;
314 cfg = get_one_free_irq_cfg(node);
319 desc->chip_data = cfg;
321 old_cfg = old_desc->chip_data;
323 memcpy(cfg, old_cfg, sizeof(struct irq_cfg));
325 init_copy_irq_2_pin(old_cfg, cfg, node);
328 static void free_irq_cfg(struct irq_cfg *old_cfg)
333 void arch_free_chip_data(struct irq_desc *old_desc, struct irq_desc *desc)
335 struct irq_cfg *old_cfg, *cfg;
337 old_cfg = old_desc->chip_data;
338 cfg = desc->chip_data;
344 free_irq_2_pin(old_cfg, cfg);
345 free_irq_cfg(old_cfg);
346 old_desc->chip_data = NULL;
349 /* end for move_irq_desc */
352 static struct irq_cfg *irq_cfg(unsigned int irq)
354 return irq < nr_irqs ? irq_cfgx + irq : NULL;
361 unsigned int unused[3];
363 unsigned int unused2[11];
367 static __attribute_const__ struct io_apic __iomem *io_apic_base(int idx)
369 return (void __iomem *) __fix_to_virt(FIX_IO_APIC_BASE_0 + idx)
370 + (mp_ioapics[idx].apicaddr & ~PAGE_MASK);
373 static inline void io_apic_eoi(unsigned int apic, unsigned int vector)
375 struct io_apic __iomem *io_apic = io_apic_base(apic);
376 writel(vector, &io_apic->eoi);
379 static inline unsigned int io_apic_read(unsigned int apic, unsigned int reg)
381 struct io_apic __iomem *io_apic = io_apic_base(apic);
382 writel(reg, &io_apic->index);
383 return readl(&io_apic->data);
386 static inline void io_apic_write(unsigned int apic, unsigned int reg, unsigned int value)
388 struct io_apic __iomem *io_apic = io_apic_base(apic);
389 writel(reg, &io_apic->index);
390 writel(value, &io_apic->data);
394 * Re-write a value: to be used for read-modify-write
395 * cycles where the read already set up the index register.
397 * Older SiS APIC requires we rewrite the index register
399 static inline void io_apic_modify(unsigned int apic, unsigned int reg, unsigned int value)
401 struct io_apic __iomem *io_apic = io_apic_base(apic);
404 writel(reg, &io_apic->index);
405 writel(value, &io_apic->data);
408 static bool io_apic_level_ack_pending(struct irq_cfg *cfg)
410 struct irq_pin_list *entry;
413 spin_lock_irqsave(&ioapic_lock, flags);
414 entry = cfg->irq_2_pin;
422 reg = io_apic_read(entry->apic, 0x10 + pin*2);
423 /* Is the remote IRR bit set? */
424 if (reg & IO_APIC_REDIR_REMOTE_IRR) {
425 spin_unlock_irqrestore(&ioapic_lock, flags);
432 spin_unlock_irqrestore(&ioapic_lock, flags);
438 struct { u32 w1, w2; };
439 struct IO_APIC_route_entry entry;
442 static struct IO_APIC_route_entry ioapic_read_entry(int apic, int pin)
444 union entry_union eu;
446 spin_lock_irqsave(&ioapic_lock, flags);
447 eu.w1 = io_apic_read(apic, 0x10 + 2 * pin);
448 eu.w2 = io_apic_read(apic, 0x11 + 2 * pin);
449 spin_unlock_irqrestore(&ioapic_lock, flags);
454 * When we write a new IO APIC routing entry, we need to write the high
455 * word first! If the mask bit in the low word is clear, we will enable
456 * the interrupt, and we need to make sure the entry is fully populated
457 * before that happens.
460 __ioapic_write_entry(int apic, int pin, struct IO_APIC_route_entry e)
462 union entry_union eu;
464 io_apic_write(apic, 0x11 + 2*pin, eu.w2);
465 io_apic_write(apic, 0x10 + 2*pin, eu.w1);
468 void ioapic_write_entry(int apic, int pin, struct IO_APIC_route_entry e)
471 spin_lock_irqsave(&ioapic_lock, flags);
472 __ioapic_write_entry(apic, pin, e);
473 spin_unlock_irqrestore(&ioapic_lock, flags);
477 * When we mask an IO APIC routing entry, we need to write the low
478 * word first, in order to set the mask bit before we change the
481 static void ioapic_mask_entry(int apic, int pin)
484 union entry_union eu = { .entry.mask = 1 };
486 spin_lock_irqsave(&ioapic_lock, flags);
487 io_apic_write(apic, 0x10 + 2*pin, eu.w1);
488 io_apic_write(apic, 0x11 + 2*pin, eu.w2);
489 spin_unlock_irqrestore(&ioapic_lock, flags);
493 * The common case is 1:1 IRQ<->pin mappings. Sometimes there are
494 * shared ISA-space IRQs, so we have to support them. We are super
495 * fast in the common case, and fast for shared ISA-space IRQs.
497 static void add_pin_to_irq_node(struct irq_cfg *cfg, int node, int apic, int pin)
499 struct irq_pin_list *entry;
501 entry = cfg->irq_2_pin;
503 entry = get_one_free_irq_2_pin(node);
505 printk(KERN_ERR "can not alloc irq_2_pin to add %d - %d\n",
509 cfg->irq_2_pin = entry;
515 while (entry->next) {
516 /* not again, please */
517 if (entry->apic == apic && entry->pin == pin)
523 entry->next = get_one_free_irq_2_pin(node);
530 * Reroute an IRQ to a different pin.
532 static void __init replace_pin_at_irq_node(struct irq_cfg *cfg, int node,
533 int oldapic, int oldpin,
534 int newapic, int newpin)
536 struct irq_pin_list *entry = cfg->irq_2_pin;
540 if (entry->apic == oldapic && entry->pin == oldpin) {
541 entry->apic = newapic;
544 /* every one is different, right? */
550 /* why? call replace before add? */
552 add_pin_to_irq_node(cfg, node, newapic, newpin);
555 static inline void io_apic_modify_irq(struct irq_cfg *cfg,
556 int mask_and, int mask_or,
557 void (*final)(struct irq_pin_list *entry))
560 struct irq_pin_list *entry;
562 for (entry = cfg->irq_2_pin; entry != NULL; entry = entry->next) {
565 reg = io_apic_read(entry->apic, 0x10 + pin * 2);
568 io_apic_modify(entry->apic, 0x10 + pin * 2, reg);
574 static void __unmask_IO_APIC_irq(struct irq_cfg *cfg)
576 io_apic_modify_irq(cfg, ~IO_APIC_REDIR_MASKED, 0, NULL);
580 static void io_apic_sync(struct irq_pin_list *entry)
583 * Synchronize the IO-APIC and the CPU by doing
584 * a dummy read from the IO-APIC
586 struct io_apic __iomem *io_apic;
587 io_apic = io_apic_base(entry->apic);
588 readl(&io_apic->data);
591 static void __mask_IO_APIC_irq(struct irq_cfg *cfg)
593 io_apic_modify_irq(cfg, ~0, IO_APIC_REDIR_MASKED, &io_apic_sync);
595 #else /* CONFIG_X86_32 */
596 static void __mask_IO_APIC_irq(struct irq_cfg *cfg)
598 io_apic_modify_irq(cfg, ~0, IO_APIC_REDIR_MASKED, NULL);
601 static void __mask_and_edge_IO_APIC_irq(struct irq_cfg *cfg)
603 io_apic_modify_irq(cfg, ~IO_APIC_REDIR_LEVEL_TRIGGER,
604 IO_APIC_REDIR_MASKED, NULL);
607 static void __unmask_and_level_IO_APIC_irq(struct irq_cfg *cfg)
609 io_apic_modify_irq(cfg, ~IO_APIC_REDIR_MASKED,
610 IO_APIC_REDIR_LEVEL_TRIGGER, NULL);
612 #endif /* CONFIG_X86_32 */
614 static void mask_IO_APIC_irq_desc(struct irq_desc *desc)
616 struct irq_cfg *cfg = desc->chip_data;
621 spin_lock_irqsave(&ioapic_lock, flags);
622 __mask_IO_APIC_irq(cfg);
623 spin_unlock_irqrestore(&ioapic_lock, flags);
626 static void unmask_IO_APIC_irq_desc(struct irq_desc *desc)
628 struct irq_cfg *cfg = desc->chip_data;
631 spin_lock_irqsave(&ioapic_lock, flags);
632 __unmask_IO_APIC_irq(cfg);
633 spin_unlock_irqrestore(&ioapic_lock, flags);
636 static void mask_IO_APIC_irq(unsigned int irq)
638 struct irq_desc *desc = irq_to_desc(irq);
640 mask_IO_APIC_irq_desc(desc);
642 static void unmask_IO_APIC_irq(unsigned int irq)
644 struct irq_desc *desc = irq_to_desc(irq);
646 unmask_IO_APIC_irq_desc(desc);
649 static void clear_IO_APIC_pin(unsigned int apic, unsigned int pin)
651 struct IO_APIC_route_entry entry;
653 /* Check delivery_mode to be sure we're not clearing an SMI pin */
654 entry = ioapic_read_entry(apic, pin);
655 if (entry.delivery_mode == dest_SMI)
658 * Disable it in the IO-APIC irq-routing table:
660 ioapic_mask_entry(apic, pin);
663 static void clear_IO_APIC (void)
667 for (apic = 0; apic < nr_ioapics; apic++)
668 for (pin = 0; pin < nr_ioapic_registers[apic]; pin++)
669 clear_IO_APIC_pin(apic, pin);
674 * support for broken MP BIOSs, enables hand-redirection of PIRQ0-7 to
675 * specific CPU-side IRQs.
679 static int pirq_entries[MAX_PIRQS] = {
680 [0 ... MAX_PIRQS - 1] = -1
683 static int __init ioapic_pirq_setup(char *str)
686 int ints[MAX_PIRQS+1];
688 get_options(str, ARRAY_SIZE(ints), ints);
690 apic_printk(APIC_VERBOSE, KERN_INFO
691 "PIRQ redirection, working around broken MP-BIOS.\n");
693 if (ints[0] < MAX_PIRQS)
696 for (i = 0; i < max; i++) {
697 apic_printk(APIC_VERBOSE, KERN_DEBUG
698 "... PIRQ%d -> IRQ %d\n", i, ints[i+1]);
700 * PIRQs are mapped upside down, usually.
702 pirq_entries[MAX_PIRQS-i-1] = ints[i+1];
707 __setup("pirq=", ioapic_pirq_setup);
708 #endif /* CONFIG_X86_32 */
710 struct IO_APIC_route_entry **alloc_ioapic_entries(void)
713 struct IO_APIC_route_entry **ioapic_entries;
715 ioapic_entries = kzalloc(sizeof(*ioapic_entries) * nr_ioapics,
720 for (apic = 0; apic < nr_ioapics; apic++) {
721 ioapic_entries[apic] =
722 kzalloc(sizeof(struct IO_APIC_route_entry) *
723 nr_ioapic_registers[apic], GFP_ATOMIC);
724 if (!ioapic_entries[apic])
728 return ioapic_entries;
732 kfree(ioapic_entries[apic]);
733 kfree(ioapic_entries);
739 * Saves all the IO-APIC RTE's
741 int save_IO_APIC_setup(struct IO_APIC_route_entry **ioapic_entries)
748 for (apic = 0; apic < nr_ioapics; apic++) {
749 if (!ioapic_entries[apic])
752 for (pin = 0; pin < nr_ioapic_registers[apic]; pin++)
753 ioapic_entries[apic][pin] =
754 ioapic_read_entry(apic, pin);
761 * Mask all IO APIC entries.
763 void mask_IO_APIC_setup(struct IO_APIC_route_entry **ioapic_entries)
770 for (apic = 0; apic < nr_ioapics; apic++) {
771 if (!ioapic_entries[apic])
774 for (pin = 0; pin < nr_ioapic_registers[apic]; pin++) {
775 struct IO_APIC_route_entry entry;
777 entry = ioapic_entries[apic][pin];
780 ioapic_write_entry(apic, pin, entry);
787 * Restore IO APIC entries which was saved in ioapic_entries.
789 int restore_IO_APIC_setup(struct IO_APIC_route_entry **ioapic_entries)
796 for (apic = 0; apic < nr_ioapics; apic++) {
797 if (!ioapic_entries[apic])
800 for (pin = 0; pin < nr_ioapic_registers[apic]; pin++)
801 ioapic_write_entry(apic, pin,
802 ioapic_entries[apic][pin]);
807 void free_ioapic_entries(struct IO_APIC_route_entry **ioapic_entries)
811 for (apic = 0; apic < nr_ioapics; apic++)
812 kfree(ioapic_entries[apic]);
814 kfree(ioapic_entries);
818 * Find the IRQ entry number of a certain pin.
820 static int find_irq_entry(int apic, int pin, int type)
824 for (i = 0; i < mp_irq_entries; i++)
825 if (mp_irqs[i].irqtype == type &&
826 (mp_irqs[i].dstapic == mp_ioapics[apic].apicid ||
827 mp_irqs[i].dstapic == MP_APIC_ALL) &&
828 mp_irqs[i].dstirq == pin)
835 * Find the pin to which IRQ[irq] (ISA) is connected
837 static int __init find_isa_irq_pin(int irq, int type)
841 for (i = 0; i < mp_irq_entries; i++) {
842 int lbus = mp_irqs[i].srcbus;
844 if (test_bit(lbus, mp_bus_not_pci) &&
845 (mp_irqs[i].irqtype == type) &&
846 (mp_irqs[i].srcbusirq == irq))
848 return mp_irqs[i].dstirq;
853 static int __init find_isa_irq_apic(int irq, int type)
857 for (i = 0; i < mp_irq_entries; i++) {
858 int lbus = mp_irqs[i].srcbus;
860 if (test_bit(lbus, mp_bus_not_pci) &&
861 (mp_irqs[i].irqtype == type) &&
862 (mp_irqs[i].srcbusirq == irq))
865 if (i < mp_irq_entries) {
867 for(apic = 0; apic < nr_ioapics; apic++) {
868 if (mp_ioapics[apic].apicid == mp_irqs[i].dstapic)
876 #if defined(CONFIG_EISA) || defined(CONFIG_MCA)
878 * EISA Edge/Level control register, ELCR
880 static int EISA_ELCR(unsigned int irq)
882 if (irq < NR_IRQS_LEGACY) {
883 unsigned int port = 0x4d0 + (irq >> 3);
884 return (inb(port) >> (irq & 7)) & 1;
886 apic_printk(APIC_VERBOSE, KERN_INFO
887 "Broken MPtable reports ISA irq %d\n", irq);
893 /* ISA interrupts are always polarity zero edge triggered,
894 * when listed as conforming in the MP table. */
896 #define default_ISA_trigger(idx) (0)
897 #define default_ISA_polarity(idx) (0)
899 /* EISA interrupts are always polarity zero and can be edge or level
900 * trigger depending on the ELCR value. If an interrupt is listed as
901 * EISA conforming in the MP table, that means its trigger type must
902 * be read in from the ELCR */
904 #define default_EISA_trigger(idx) (EISA_ELCR(mp_irqs[idx].srcbusirq))
905 #define default_EISA_polarity(idx) default_ISA_polarity(idx)
907 /* PCI interrupts are always polarity one level triggered,
908 * when listed as conforming in the MP table. */
910 #define default_PCI_trigger(idx) (1)
911 #define default_PCI_polarity(idx) (1)
913 /* MCA interrupts are always polarity zero level triggered,
914 * when listed as conforming in the MP table. */
916 #define default_MCA_trigger(idx) (1)
917 #define default_MCA_polarity(idx) default_ISA_polarity(idx)
919 static int MPBIOS_polarity(int idx)
921 int bus = mp_irqs[idx].srcbus;
925 * Determine IRQ line polarity (high active or low active):
927 switch (mp_irqs[idx].irqflag & 3)
929 case 0: /* conforms, ie. bus-type dependent polarity */
930 if (test_bit(bus, mp_bus_not_pci))
931 polarity = default_ISA_polarity(idx);
933 polarity = default_PCI_polarity(idx);
935 case 1: /* high active */
940 case 2: /* reserved */
942 printk(KERN_WARNING "broken BIOS!!\n");
946 case 3: /* low active */
951 default: /* invalid */
953 printk(KERN_WARNING "broken BIOS!!\n");
961 static int MPBIOS_trigger(int idx)
963 int bus = mp_irqs[idx].srcbus;
967 * Determine IRQ trigger mode (edge or level sensitive):
969 switch ((mp_irqs[idx].irqflag>>2) & 3)
971 case 0: /* conforms, ie. bus-type dependent */
972 if (test_bit(bus, mp_bus_not_pci))
973 trigger = default_ISA_trigger(idx);
975 trigger = default_PCI_trigger(idx);
976 #if defined(CONFIG_EISA) || defined(CONFIG_MCA)
977 switch (mp_bus_id_to_type[bus]) {
978 case MP_BUS_ISA: /* ISA pin */
980 /* set before the switch */
983 case MP_BUS_EISA: /* EISA pin */
985 trigger = default_EISA_trigger(idx);
988 case MP_BUS_PCI: /* PCI pin */
990 /* set before the switch */
993 case MP_BUS_MCA: /* MCA pin */
995 trigger = default_MCA_trigger(idx);
1000 printk(KERN_WARNING "broken BIOS!!\n");
1012 case 2: /* reserved */
1014 printk(KERN_WARNING "broken BIOS!!\n");
1023 default: /* invalid */
1025 printk(KERN_WARNING "broken BIOS!!\n");
1033 static inline int irq_polarity(int idx)
1035 return MPBIOS_polarity(idx);
1038 static inline int irq_trigger(int idx)
1040 return MPBIOS_trigger(idx);
1043 int (*ioapic_renumber_irq)(int ioapic, int irq);
1044 static int pin_2_irq(int idx, int apic, int pin)
1047 int bus = mp_irqs[idx].srcbus;
1050 * Debugging check, we are in big trouble if this message pops up!
1052 if (mp_irqs[idx].dstirq != pin)
1053 printk(KERN_ERR "broken BIOS or MPTABLE parser, ayiee!!\n");
1055 if (test_bit(bus, mp_bus_not_pci)) {
1056 irq = mp_irqs[idx].srcbusirq;
1059 * PCI IRQs are mapped in order
1063 irq += nr_ioapic_registers[i++];
1066 * For MPS mode, so far only needed by ES7000 platform
1068 if (ioapic_renumber_irq)
1069 irq = ioapic_renumber_irq(apic, irq);
1072 #ifdef CONFIG_X86_32
1074 * PCI IRQ command line redirection. Yes, limits are hardcoded.
1076 if ((pin >= 16) && (pin <= 23)) {
1077 if (pirq_entries[pin-16] != -1) {
1078 if (!pirq_entries[pin-16]) {
1079 apic_printk(APIC_VERBOSE, KERN_DEBUG
1080 "disabling PIRQ%d\n", pin-16);
1082 irq = pirq_entries[pin-16];
1083 apic_printk(APIC_VERBOSE, KERN_DEBUG
1084 "using PIRQ%d -> IRQ %d\n",
1095 * Find a specific PCI IRQ entry.
1096 * Not an __init, possibly needed by modules
1098 int IO_APIC_get_PCI_irq_vector(int bus, int slot, int pin,
1099 struct io_apic_irq_attr *irq_attr)
1101 int apic, i, best_guess = -1;
1103 apic_printk(APIC_DEBUG,
1104 "querying PCI -> IRQ mapping bus:%d, slot:%d, pin:%d.\n",
1106 if (test_bit(bus, mp_bus_not_pci)) {
1107 apic_printk(APIC_VERBOSE,
1108 "PCI BIOS passed nonexistent PCI bus %d!\n", bus);
1111 for (i = 0; i < mp_irq_entries; i++) {
1112 int lbus = mp_irqs[i].srcbus;
1114 for (apic = 0; apic < nr_ioapics; apic++)
1115 if (mp_ioapics[apic].apicid == mp_irqs[i].dstapic ||
1116 mp_irqs[i].dstapic == MP_APIC_ALL)
1119 if (!test_bit(lbus, mp_bus_not_pci) &&
1120 !mp_irqs[i].irqtype &&
1122 (slot == ((mp_irqs[i].srcbusirq >> 2) & 0x1f))) {
1123 int irq = pin_2_irq(i, apic, mp_irqs[i].dstirq);
1125 if (!(apic || IO_APIC_IRQ(irq)))
1128 if (pin == (mp_irqs[i].srcbusirq & 3)) {
1129 set_io_apic_irq_attr(irq_attr, apic,
1136 * Use the first all-but-pin matching entry as a
1137 * best-guess fuzzy result for broken mptables.
1139 if (best_guess < 0) {
1140 set_io_apic_irq_attr(irq_attr, apic,
1150 EXPORT_SYMBOL(IO_APIC_get_PCI_irq_vector);
1152 void lock_vector_lock(void)
1154 /* Used to the online set of cpus does not change
1155 * during assign_irq_vector.
1157 spin_lock(&vector_lock);
1160 void unlock_vector_lock(void)
1162 spin_unlock(&vector_lock);
1166 __assign_irq_vector(int irq, struct irq_cfg *cfg, const struct cpumask *mask)
1169 * NOTE! The local APIC isn't very good at handling
1170 * multiple interrupts at the same interrupt level.
1171 * As the interrupt level is determined by taking the
1172 * vector number and shifting that right by 4, we
1173 * want to spread these out a bit so that they don't
1174 * all fall in the same interrupt level.
1176 * Also, we've got to be careful not to trash gate
1177 * 0x80, because int 0x80 is hm, kind of importantish. ;)
1179 static int current_vector = FIRST_DEVICE_VECTOR, current_offset = 0;
1180 unsigned int old_vector;
1182 cpumask_var_t tmp_mask;
1184 if ((cfg->move_in_progress) || cfg->move_cleanup_count)
1187 if (!alloc_cpumask_var(&tmp_mask, GFP_ATOMIC))
1190 old_vector = cfg->vector;
1192 cpumask_and(tmp_mask, mask, cpu_online_mask);
1193 cpumask_and(tmp_mask, cfg->domain, tmp_mask);
1194 if (!cpumask_empty(tmp_mask)) {
1195 free_cpumask_var(tmp_mask);
1200 /* Only try and allocate irqs on cpus that are present */
1202 for_each_cpu_and(cpu, mask, cpu_online_mask) {
1206 apic->vector_allocation_domain(cpu, tmp_mask);
1208 vector = current_vector;
1209 offset = current_offset;
1212 if (vector >= first_system_vector) {
1213 /* If out of vectors on large boxen, must share them. */
1214 offset = (offset + 1) % 8;
1215 vector = FIRST_DEVICE_VECTOR + offset;
1217 if (unlikely(current_vector == vector))
1220 if (test_bit(vector, used_vectors))
1223 for_each_cpu_and(new_cpu, tmp_mask, cpu_online_mask)
1224 if (per_cpu(vector_irq, new_cpu)[vector] != -1)
1227 current_vector = vector;
1228 current_offset = offset;
1230 cfg->move_in_progress = 1;
1231 cpumask_copy(cfg->old_domain, cfg->domain);
1233 for_each_cpu_and(new_cpu, tmp_mask, cpu_online_mask)
1234 per_cpu(vector_irq, new_cpu)[vector] = irq;
1235 cfg->vector = vector;
1236 cpumask_copy(cfg->domain, tmp_mask);
1240 free_cpumask_var(tmp_mask);
1245 assign_irq_vector(int irq, struct irq_cfg *cfg, const struct cpumask *mask)
1248 unsigned long flags;
1250 spin_lock_irqsave(&vector_lock, flags);
1251 err = __assign_irq_vector(irq, cfg, mask);
1252 spin_unlock_irqrestore(&vector_lock, flags);
1256 static void __clear_irq_vector(int irq, struct irq_cfg *cfg)
1260 BUG_ON(!cfg->vector);
1262 vector = cfg->vector;
1263 for_each_cpu_and(cpu, cfg->domain, cpu_online_mask)
1264 per_cpu(vector_irq, cpu)[vector] = -1;
1267 cpumask_clear(cfg->domain);
1269 if (likely(!cfg->move_in_progress))
1271 for_each_cpu_and(cpu, cfg->old_domain, cpu_online_mask) {
1272 for (vector = FIRST_EXTERNAL_VECTOR; vector < NR_VECTORS;
1274 if (per_cpu(vector_irq, cpu)[vector] != irq)
1276 per_cpu(vector_irq, cpu)[vector] = -1;
1280 cfg->move_in_progress = 0;
1283 void __setup_vector_irq(int cpu)
1285 /* Initialize vector_irq on a new cpu */
1286 /* This function must be called with vector_lock held */
1288 struct irq_cfg *cfg;
1289 struct irq_desc *desc;
1291 /* Mark the inuse vectors */
1292 for_each_irq_desc(irq, desc) {
1293 cfg = desc->chip_data;
1294 if (!cpumask_test_cpu(cpu, cfg->domain))
1296 vector = cfg->vector;
1297 per_cpu(vector_irq, cpu)[vector] = irq;
1299 /* Mark the free vectors */
1300 for (vector = 0; vector < NR_VECTORS; ++vector) {
1301 irq = per_cpu(vector_irq, cpu)[vector];
1306 if (!cpumask_test_cpu(cpu, cfg->domain))
1307 per_cpu(vector_irq, cpu)[vector] = -1;
1311 static struct irq_chip ioapic_chip;
1312 static struct irq_chip ir_ioapic_chip;
1314 #define IOAPIC_AUTO -1
1315 #define IOAPIC_EDGE 0
1316 #define IOAPIC_LEVEL 1
1318 #ifdef CONFIG_X86_32
1319 static inline int IO_APIC_irq_trigger(int irq)
1323 for (apic = 0; apic < nr_ioapics; apic++) {
1324 for (pin = 0; pin < nr_ioapic_registers[apic]; pin++) {
1325 idx = find_irq_entry(apic, pin, mp_INT);
1326 if ((idx != -1) && (irq == pin_2_irq(idx, apic, pin)))
1327 return irq_trigger(idx);
1331 * nonexistent IRQs are edge default
1336 static inline int IO_APIC_irq_trigger(int irq)
1342 static void ioapic_register_intr(int irq, struct irq_desc *desc, unsigned long trigger)
1345 if ((trigger == IOAPIC_AUTO && IO_APIC_irq_trigger(irq)) ||
1346 trigger == IOAPIC_LEVEL)
1347 desc->status |= IRQ_LEVEL;
1349 desc->status &= ~IRQ_LEVEL;
1351 if (irq_remapped(irq)) {
1352 desc->status |= IRQ_MOVE_PCNTXT;
1354 set_irq_chip_and_handler_name(irq, &ir_ioapic_chip,
1358 set_irq_chip_and_handler_name(irq, &ir_ioapic_chip,
1359 handle_edge_irq, "edge");
1363 if ((trigger == IOAPIC_AUTO && IO_APIC_irq_trigger(irq)) ||
1364 trigger == IOAPIC_LEVEL)
1365 set_irq_chip_and_handler_name(irq, &ioapic_chip,
1369 set_irq_chip_and_handler_name(irq, &ioapic_chip,
1370 handle_edge_irq, "edge");
1373 int setup_ioapic_entry(int apic_id, int irq,
1374 struct IO_APIC_route_entry *entry,
1375 unsigned int destination, int trigger,
1376 int polarity, int vector, int pin)
1379 * add it to the IO-APIC irq-routing table:
1381 memset(entry,0,sizeof(*entry));
1383 if (intr_remapping_enabled) {
1384 struct intel_iommu *iommu = map_ioapic_to_ir(apic_id);
1386 struct IR_IO_APIC_route_entry *ir_entry =
1387 (struct IR_IO_APIC_route_entry *) entry;
1391 panic("No mapping iommu for ioapic %d\n", apic_id);
1393 index = alloc_irte(iommu, irq, 1);
1395 panic("Failed to allocate IRTE for ioapic %d\n", apic_id);
1397 memset(&irte, 0, sizeof(irte));
1400 irte.dst_mode = apic->irq_dest_mode;
1402 * Trigger mode in the IRTE will always be edge, and the
1403 * actual level or edge trigger will be setup in the IO-APIC
1404 * RTE. This will help simplify level triggered irq migration.
1405 * For more details, see the comments above explainig IO-APIC
1406 * irq migration in the presence of interrupt-remapping.
1408 irte.trigger_mode = 0;
1409 irte.dlvry_mode = apic->irq_delivery_mode;
1410 irte.vector = vector;
1411 irte.dest_id = IRTE_DEST(destination);
1413 modify_irte(irq, &irte);
1415 ir_entry->index2 = (index >> 15) & 0x1;
1417 ir_entry->format = 1;
1418 ir_entry->index = (index & 0x7fff);
1420 * IO-APIC RTE will be configured with virtual vector.
1421 * irq handler will do the explicit EOI to the io-apic.
1423 ir_entry->vector = pin;
1425 entry->delivery_mode = apic->irq_delivery_mode;
1426 entry->dest_mode = apic->irq_dest_mode;
1427 entry->dest = destination;
1428 entry->vector = vector;
1431 entry->mask = 0; /* enable IRQ */
1432 entry->trigger = trigger;
1433 entry->polarity = polarity;
1435 /* Mask level triggered irqs.
1436 * Use IRQ_DELAYED_DISABLE for edge triggered irqs.
1443 static void setup_IO_APIC_irq(int apic_id, int pin, unsigned int irq, struct irq_desc *desc,
1444 int trigger, int polarity)
1446 struct irq_cfg *cfg;
1447 struct IO_APIC_route_entry entry;
1450 if (!IO_APIC_IRQ(irq))
1453 cfg = desc->chip_data;
1455 if (assign_irq_vector(irq, cfg, apic->target_cpus()))
1458 dest = apic->cpu_mask_to_apicid_and(cfg->domain, apic->target_cpus());
1460 apic_printk(APIC_VERBOSE,KERN_DEBUG
1461 "IOAPIC[%d]: Set routing entry (%d-%d -> 0x%x -> "
1462 "IRQ %d Mode:%i Active:%i)\n",
1463 apic_id, mp_ioapics[apic_id].apicid, pin, cfg->vector,
1464 irq, trigger, polarity);
1467 if (setup_ioapic_entry(mp_ioapics[apic_id].apicid, irq, &entry,
1468 dest, trigger, polarity, cfg->vector, pin)) {
1469 printk("Failed to setup ioapic entry for ioapic %d, pin %d\n",
1470 mp_ioapics[apic_id].apicid, pin);
1471 __clear_irq_vector(irq, cfg);
1475 ioapic_register_intr(irq, desc, trigger);
1476 if (irq < NR_IRQS_LEGACY)
1477 disable_8259A_irq(irq);
1479 ioapic_write_entry(apic_id, pin, entry);
1483 DECLARE_BITMAP(pin_programmed, MP_MAX_IOAPIC_PIN + 1);
1484 } mp_ioapic_routing[MAX_IO_APICS];
1486 static void __init setup_IO_APIC_irqs(void)
1488 int apic_id = 0, pin, idx, irq;
1490 struct irq_desc *desc;
1491 struct irq_cfg *cfg;
1492 int node = cpu_to_node(boot_cpu_id);
1494 apic_printk(APIC_VERBOSE, KERN_DEBUG "init IO_APIC IRQs\n");
1497 if (!acpi_disabled && acpi_ioapic) {
1498 apic_id = mp_find_ioapic(0);
1504 for (pin = 0; pin < nr_ioapic_registers[apic_id]; pin++) {
1505 idx = find_irq_entry(apic_id, pin, mp_INT);
1509 apic_printk(APIC_VERBOSE,
1510 KERN_DEBUG " %d-%d",
1511 mp_ioapics[apic_id].apicid, pin);
1513 apic_printk(APIC_VERBOSE, " %d-%d",
1514 mp_ioapics[apic_id].apicid, pin);
1518 apic_printk(APIC_VERBOSE,
1519 " (apicid-pin) not connected\n");
1523 irq = pin_2_irq(idx, apic_id, pin);
1526 * Skip the timer IRQ if there's a quirk handler
1527 * installed and if it returns 1:
1529 if (apic->multi_timer_check &&
1530 apic->multi_timer_check(apic_id, irq))
1533 desc = irq_to_desc_alloc_node(irq, node);
1535 printk(KERN_INFO "can not get irq_desc for %d\n", irq);
1538 cfg = desc->chip_data;
1539 add_pin_to_irq_node(cfg, node, apic_id, pin);
1541 * don't mark it in pin_programmed, so later acpi could
1542 * set it correctly when irq < 16
1544 setup_IO_APIC_irq(apic_id, pin, irq, desc,
1545 irq_trigger(idx), irq_polarity(idx));
1549 apic_printk(APIC_VERBOSE,
1550 " (apicid-pin) not connected\n");
1554 * Set up the timer pin, possibly with the 8259A-master behind.
1556 static void __init setup_timer_IRQ0_pin(unsigned int apic_id, unsigned int pin,
1559 struct IO_APIC_route_entry entry;
1561 if (intr_remapping_enabled)
1564 memset(&entry, 0, sizeof(entry));
1567 * We use logical delivery to get the timer IRQ
1570 entry.dest_mode = apic->irq_dest_mode;
1571 entry.mask = 0; /* don't mask IRQ for edge */
1572 entry.dest = apic->cpu_mask_to_apicid(apic->target_cpus());
1573 entry.delivery_mode = apic->irq_delivery_mode;
1576 entry.vector = vector;
1579 * The timer IRQ doesn't have to know that behind the
1580 * scene we may have a 8259A-master in AEOI mode ...
1582 set_irq_chip_and_handler_name(0, &ioapic_chip, handle_edge_irq, "edge");
1585 * Add it to the IO-APIC irq-routing table:
1587 ioapic_write_entry(apic_id, pin, entry);
1591 __apicdebuginit(void) print_IO_APIC(void)
1594 union IO_APIC_reg_00 reg_00;
1595 union IO_APIC_reg_01 reg_01;
1596 union IO_APIC_reg_02 reg_02;
1597 union IO_APIC_reg_03 reg_03;
1598 unsigned long flags;
1599 struct irq_cfg *cfg;
1600 struct irq_desc *desc;
1603 if (apic_verbosity == APIC_QUIET)
1606 printk(KERN_DEBUG "number of MP IRQ sources: %d.\n", mp_irq_entries);
1607 for (i = 0; i < nr_ioapics; i++)
1608 printk(KERN_DEBUG "number of IO-APIC #%d registers: %d.\n",
1609 mp_ioapics[i].apicid, nr_ioapic_registers[i]);
1612 * We are a bit conservative about what we expect. We have to
1613 * know about every hardware change ASAP.
1615 printk(KERN_INFO "testing the IO APIC.......................\n");
1617 for (apic = 0; apic < nr_ioapics; apic++) {
1619 spin_lock_irqsave(&ioapic_lock, flags);
1620 reg_00.raw = io_apic_read(apic, 0);
1621 reg_01.raw = io_apic_read(apic, 1);
1622 if (reg_01.bits.version >= 0x10)
1623 reg_02.raw = io_apic_read(apic, 2);
1624 if (reg_01.bits.version >= 0x20)
1625 reg_03.raw = io_apic_read(apic, 3);
1626 spin_unlock_irqrestore(&ioapic_lock, flags);
1629 printk(KERN_DEBUG "IO APIC #%d......\n", mp_ioapics[apic].apicid);
1630 printk(KERN_DEBUG ".... register #00: %08X\n", reg_00.raw);
1631 printk(KERN_DEBUG "....... : physical APIC id: %02X\n", reg_00.bits.ID);
1632 printk(KERN_DEBUG "....... : Delivery Type: %X\n", reg_00.bits.delivery_type);
1633 printk(KERN_DEBUG "....... : LTS : %X\n", reg_00.bits.LTS);
1635 printk(KERN_DEBUG ".... register #01: %08X\n", *(int *)®_01);
1636 printk(KERN_DEBUG "....... : max redirection entries: %04X\n", reg_01.bits.entries);
1638 printk(KERN_DEBUG "....... : PRQ implemented: %X\n", reg_01.bits.PRQ);
1639 printk(KERN_DEBUG "....... : IO APIC version: %04X\n", reg_01.bits.version);
1642 * Some Intel chipsets with IO APIC VERSION of 0x1? don't have reg_02,
1643 * but the value of reg_02 is read as the previous read register
1644 * value, so ignore it if reg_02 == reg_01.
1646 if (reg_01.bits.version >= 0x10 && reg_02.raw != reg_01.raw) {
1647 printk(KERN_DEBUG ".... register #02: %08X\n", reg_02.raw);
1648 printk(KERN_DEBUG "....... : arbitration: %02X\n", reg_02.bits.arbitration);
1652 * Some Intel chipsets with IO APIC VERSION of 0x2? don't have reg_02
1653 * or reg_03, but the value of reg_0[23] is read as the previous read
1654 * register value, so ignore it if reg_03 == reg_0[12].
1656 if (reg_01.bits.version >= 0x20 && reg_03.raw != reg_02.raw &&
1657 reg_03.raw != reg_01.raw) {
1658 printk(KERN_DEBUG ".... register #03: %08X\n", reg_03.raw);
1659 printk(KERN_DEBUG "....... : Boot DT : %X\n", reg_03.bits.boot_DT);
1662 printk(KERN_DEBUG ".... IRQ redirection table:\n");
1664 printk(KERN_DEBUG " NR Dst Mask Trig IRR Pol"
1665 " Stat Dmod Deli Vect: \n");
1667 for (i = 0; i <= reg_01.bits.entries; i++) {
1668 struct IO_APIC_route_entry entry;
1670 entry = ioapic_read_entry(apic, i);
1672 printk(KERN_DEBUG " %02x %03X ",
1677 printk("%1d %1d %1d %1d %1d %1d %1d %02X\n",
1682 entry.delivery_status,
1684 entry.delivery_mode,
1689 printk(KERN_DEBUG "IRQ to pin mappings:\n");
1690 for_each_irq_desc(irq, desc) {
1691 struct irq_pin_list *entry;
1693 cfg = desc->chip_data;
1694 entry = cfg->irq_2_pin;
1697 printk(KERN_DEBUG "IRQ%d ", irq);
1699 printk("-> %d:%d", entry->apic, entry->pin);
1702 entry = entry->next;
1707 printk(KERN_INFO ".................................... done.\n");
1712 __apicdebuginit(void) print_APIC_bitfield(int base)
1717 if (apic_verbosity == APIC_QUIET)
1720 printk(KERN_DEBUG "0123456789abcdef0123456789abcdef\n" KERN_DEBUG);
1721 for (i = 0; i < 8; i++) {
1722 v = apic_read(base + i*0x10);
1723 for (j = 0; j < 32; j++) {
1733 __apicdebuginit(void) print_local_APIC(void *dummy)
1735 unsigned int i, v, ver, maxlvt;
1738 if (apic_verbosity == APIC_QUIET)
1741 printk("\n" KERN_DEBUG "printing local APIC contents on CPU#%d/%d:\n",
1742 smp_processor_id(), hard_smp_processor_id());
1743 v = apic_read(APIC_ID);
1744 printk(KERN_INFO "... APIC ID: %08x (%01x)\n", v, read_apic_id());
1745 v = apic_read(APIC_LVR);
1746 printk(KERN_INFO "... APIC VERSION: %08x\n", v);
1747 ver = GET_APIC_VERSION(v);
1748 maxlvt = lapic_get_maxlvt();
1750 v = apic_read(APIC_TASKPRI);
1751 printk(KERN_DEBUG "... APIC TASKPRI: %08x (%02x)\n", v, v & APIC_TPRI_MASK);
1753 if (APIC_INTEGRATED(ver)) { /* !82489DX */
1754 if (!APIC_XAPIC(ver)) {
1755 v = apic_read(APIC_ARBPRI);
1756 printk(KERN_DEBUG "... APIC ARBPRI: %08x (%02x)\n", v,
1757 v & APIC_ARBPRI_MASK);
1759 v = apic_read(APIC_PROCPRI);
1760 printk(KERN_DEBUG "... APIC PROCPRI: %08x\n", v);
1764 * Remote read supported only in the 82489DX and local APIC for
1765 * Pentium processors.
1767 if (!APIC_INTEGRATED(ver) || maxlvt == 3) {
1768 v = apic_read(APIC_RRR);
1769 printk(KERN_DEBUG "... APIC RRR: %08x\n", v);
1772 v = apic_read(APIC_LDR);
1773 printk(KERN_DEBUG "... APIC LDR: %08x\n", v);
1774 if (!x2apic_enabled()) {
1775 v = apic_read(APIC_DFR);
1776 printk(KERN_DEBUG "... APIC DFR: %08x\n", v);
1778 v = apic_read(APIC_SPIV);
1779 printk(KERN_DEBUG "... APIC SPIV: %08x\n", v);
1781 printk(KERN_DEBUG "... APIC ISR field:\n");
1782 print_APIC_bitfield(APIC_ISR);
1783 printk(KERN_DEBUG "... APIC TMR field:\n");
1784 print_APIC_bitfield(APIC_TMR);
1785 printk(KERN_DEBUG "... APIC IRR field:\n");
1786 print_APIC_bitfield(APIC_IRR);
1788 if (APIC_INTEGRATED(ver)) { /* !82489DX */
1789 if (maxlvt > 3) /* Due to the Pentium erratum 3AP. */
1790 apic_write(APIC_ESR, 0);
1792 v = apic_read(APIC_ESR);
1793 printk(KERN_DEBUG "... APIC ESR: %08x\n", v);
1796 icr = apic_icr_read();
1797 printk(KERN_DEBUG "... APIC ICR: %08x\n", (u32)icr);
1798 printk(KERN_DEBUG "... APIC ICR2: %08x\n", (u32)(icr >> 32));
1800 v = apic_read(APIC_LVTT);
1801 printk(KERN_DEBUG "... APIC LVTT: %08x\n", v);
1803 if (maxlvt > 3) { /* PC is LVT#4. */
1804 v = apic_read(APIC_LVTPC);
1805 printk(KERN_DEBUG "... APIC LVTPC: %08x\n", v);
1807 v = apic_read(APIC_LVT0);
1808 printk(KERN_DEBUG "... APIC LVT0: %08x\n", v);
1809 v = apic_read(APIC_LVT1);
1810 printk(KERN_DEBUG "... APIC LVT1: %08x\n", v);
1812 if (maxlvt > 2) { /* ERR is LVT#3. */
1813 v = apic_read(APIC_LVTERR);
1814 printk(KERN_DEBUG "... APIC LVTERR: %08x\n", v);
1817 v = apic_read(APIC_TMICT);
1818 printk(KERN_DEBUG "... APIC TMICT: %08x\n", v);
1819 v = apic_read(APIC_TMCCT);
1820 printk(KERN_DEBUG "... APIC TMCCT: %08x\n", v);
1821 v = apic_read(APIC_TDCR);
1822 printk(KERN_DEBUG "... APIC TDCR: %08x\n", v);
1824 if (boot_cpu_has(X86_FEATURE_EXTAPIC)) {
1825 v = apic_read(APIC_EFEAT);
1826 maxlvt = (v >> 16) & 0xff;
1827 printk(KERN_DEBUG "... APIC EFEAT: %08x\n", v);
1828 v = apic_read(APIC_ECTRL);
1829 printk(KERN_DEBUG "... APIC ECTRL: %08x\n", v);
1830 for (i = 0; i < maxlvt; i++) {
1831 v = apic_read(APIC_EILVTn(i));
1832 printk(KERN_DEBUG "... APIC EILVT%d: %08x\n", i, v);
1838 __apicdebuginit(void) print_all_local_APICs(void)
1843 for_each_online_cpu(cpu)
1844 smp_call_function_single(cpu, print_local_APIC, NULL, 1);
1848 __apicdebuginit(void) print_PIC(void)
1851 unsigned long flags;
1853 if (apic_verbosity == APIC_QUIET)
1856 printk(KERN_DEBUG "\nprinting PIC contents\n");
1858 spin_lock_irqsave(&i8259A_lock, flags);
1860 v = inb(0xa1) << 8 | inb(0x21);
1861 printk(KERN_DEBUG "... PIC IMR: %04x\n", v);
1863 v = inb(0xa0) << 8 | inb(0x20);
1864 printk(KERN_DEBUG "... PIC IRR: %04x\n", v);
1868 v = inb(0xa0) << 8 | inb(0x20);
1872 spin_unlock_irqrestore(&i8259A_lock, flags);
1874 printk(KERN_DEBUG "... PIC ISR: %04x\n", v);
1876 v = inb(0x4d1) << 8 | inb(0x4d0);
1877 printk(KERN_DEBUG "... PIC ELCR: %04x\n", v);
1880 __apicdebuginit(int) print_all_ICs(void)
1884 /* don't print out if apic is not there */
1885 if (!cpu_has_apic || disable_apic)
1888 print_all_local_APICs();
1894 fs_initcall(print_all_ICs);
1897 /* Where if anywhere is the i8259 connect in external int mode */
1898 static struct { int pin, apic; } ioapic_i8259 = { -1, -1 };
1900 void __init enable_IO_APIC(void)
1902 union IO_APIC_reg_01 reg_01;
1903 int i8259_apic, i8259_pin;
1905 unsigned long flags;
1908 * The number of IO-APIC IRQ registers (== #pins):
1910 for (apic = 0; apic < nr_ioapics; apic++) {
1911 spin_lock_irqsave(&ioapic_lock, flags);
1912 reg_01.raw = io_apic_read(apic, 1);
1913 spin_unlock_irqrestore(&ioapic_lock, flags);
1914 nr_ioapic_registers[apic] = reg_01.bits.entries+1;
1916 for(apic = 0; apic < nr_ioapics; apic++) {
1918 /* See if any of the pins is in ExtINT mode */
1919 for (pin = 0; pin < nr_ioapic_registers[apic]; pin++) {
1920 struct IO_APIC_route_entry entry;
1921 entry = ioapic_read_entry(apic, pin);
1923 /* If the interrupt line is enabled and in ExtInt mode
1924 * I have found the pin where the i8259 is connected.
1926 if ((entry.mask == 0) && (entry.delivery_mode == dest_ExtINT)) {
1927 ioapic_i8259.apic = apic;
1928 ioapic_i8259.pin = pin;
1934 /* Look to see what if the MP table has reported the ExtINT */
1935 /* If we could not find the appropriate pin by looking at the ioapic
1936 * the i8259 probably is not connected the ioapic but give the
1937 * mptable a chance anyway.
1939 i8259_pin = find_isa_irq_pin(0, mp_ExtINT);
1940 i8259_apic = find_isa_irq_apic(0, mp_ExtINT);
1941 /* Trust the MP table if nothing is setup in the hardware */
1942 if ((ioapic_i8259.pin == -1) && (i8259_pin >= 0)) {
1943 printk(KERN_WARNING "ExtINT not setup in hardware but reported by MP table\n");
1944 ioapic_i8259.pin = i8259_pin;
1945 ioapic_i8259.apic = i8259_apic;
1947 /* Complain if the MP table and the hardware disagree */
1948 if (((ioapic_i8259.apic != i8259_apic) || (ioapic_i8259.pin != i8259_pin)) &&
1949 (i8259_pin >= 0) && (ioapic_i8259.pin >= 0))
1951 printk(KERN_WARNING "ExtINT in hardware and MP table differ\n");
1955 * Do not trust the IO-APIC being empty at bootup
1961 * Not an __init, needed by the reboot code
1963 void disable_IO_APIC(void)
1966 * Clear the IO-APIC before rebooting:
1971 * If the i8259 is routed through an IOAPIC
1972 * Put that IOAPIC in virtual wire mode
1973 * so legacy interrupts can be delivered.
1975 * With interrupt-remapping, for now we will use virtual wire A mode,
1976 * as virtual wire B is little complex (need to configure both
1977 * IOAPIC RTE aswell as interrupt-remapping table entry).
1978 * As this gets called during crash dump, keep this simple for now.
1980 if (ioapic_i8259.pin != -1 && !intr_remapping_enabled) {
1981 struct IO_APIC_route_entry entry;
1983 memset(&entry, 0, sizeof(entry));
1984 entry.mask = 0; /* Enabled */
1985 entry.trigger = 0; /* Edge */
1987 entry.polarity = 0; /* High */
1988 entry.delivery_status = 0;
1989 entry.dest_mode = 0; /* Physical */
1990 entry.delivery_mode = dest_ExtINT; /* ExtInt */
1992 entry.dest = read_apic_id();
1995 * Add it to the IO-APIC irq-routing table:
1997 ioapic_write_entry(ioapic_i8259.apic, ioapic_i8259.pin, entry);
2001 * Use virtual wire A mode when interrupt remapping is enabled.
2003 disconnect_bsp_APIC(!intr_remapping_enabled && ioapic_i8259.pin != -1);
2006 #ifdef CONFIG_X86_32
2008 * function to set the IO-APIC physical IDs based on the
2009 * values stored in the MPC table.
2011 * by Matt Domsch <Matt_Domsch@dell.com> Tue Dec 21 12:25:05 CST 1999
2014 static void __init setup_ioapic_ids_from_mpc(void)
2016 union IO_APIC_reg_00 reg_00;
2017 physid_mask_t phys_id_present_map;
2020 unsigned char old_id;
2021 unsigned long flags;
2023 if (x86_quirks->setup_ioapic_ids && x86_quirks->setup_ioapic_ids())
2027 * Don't check I/O APIC IDs for xAPIC systems. They have
2028 * no meaning without the serial APIC bus.
2030 if (!(boot_cpu_data.x86_vendor == X86_VENDOR_INTEL)
2031 || APIC_XAPIC(apic_version[boot_cpu_physical_apicid]))
2034 * This is broken; anything with a real cpu count has to
2035 * circumvent this idiocy regardless.
2037 phys_id_present_map = apic->ioapic_phys_id_map(phys_cpu_present_map);
2040 * Set the IOAPIC ID to the value stored in the MPC table.
2042 for (apic_id = 0; apic_id < nr_ioapics; apic_id++) {
2044 /* Read the register 0 value */
2045 spin_lock_irqsave(&ioapic_lock, flags);
2046 reg_00.raw = io_apic_read(apic_id, 0);
2047 spin_unlock_irqrestore(&ioapic_lock, flags);
2049 old_id = mp_ioapics[apic_id].apicid;
2051 if (mp_ioapics[apic_id].apicid >= get_physical_broadcast()) {
2052 printk(KERN_ERR "BIOS bug, IO-APIC#%d ID is %d in the MPC table!...\n",
2053 apic_id, mp_ioapics[apic_id].apicid);
2054 printk(KERN_ERR "... fixing up to %d. (tell your hw vendor)\n",
2056 mp_ioapics[apic_id].apicid = reg_00.bits.ID;
2060 * Sanity check, is the ID really free? Every APIC in a
2061 * system must have a unique ID or we get lots of nice
2062 * 'stuck on smp_invalidate_needed IPI wait' messages.
2064 if (apic->check_apicid_used(phys_id_present_map,
2065 mp_ioapics[apic_id].apicid)) {
2066 printk(KERN_ERR "BIOS bug, IO-APIC#%d ID %d is already used!...\n",
2067 apic_id, mp_ioapics[apic_id].apicid);
2068 for (i = 0; i < get_physical_broadcast(); i++)
2069 if (!physid_isset(i, phys_id_present_map))
2071 if (i >= get_physical_broadcast())
2072 panic("Max APIC ID exceeded!\n");
2073 printk(KERN_ERR "... fixing up to %d. (tell your hw vendor)\n",
2075 physid_set(i, phys_id_present_map);
2076 mp_ioapics[apic_id].apicid = i;
2079 tmp = apic->apicid_to_cpu_present(mp_ioapics[apic_id].apicid);
2080 apic_printk(APIC_VERBOSE, "Setting %d in the "
2081 "phys_id_present_map\n",
2082 mp_ioapics[apic_id].apicid);
2083 physids_or(phys_id_present_map, phys_id_present_map, tmp);
2088 * We need to adjust the IRQ routing table
2089 * if the ID changed.
2091 if (old_id != mp_ioapics[apic_id].apicid)
2092 for (i = 0; i < mp_irq_entries; i++)
2093 if (mp_irqs[i].dstapic == old_id)
2095 = mp_ioapics[apic_id].apicid;
2098 * Read the right value from the MPC table and
2099 * write it into the ID register.
2101 apic_printk(APIC_VERBOSE, KERN_INFO
2102 "...changing IO-APIC physical APIC ID to %d ...",
2103 mp_ioapics[apic_id].apicid);
2105 reg_00.bits.ID = mp_ioapics[apic_id].apicid;
2106 spin_lock_irqsave(&ioapic_lock, flags);
2107 io_apic_write(apic_id, 0, reg_00.raw);
2108 spin_unlock_irqrestore(&ioapic_lock, flags);
2113 spin_lock_irqsave(&ioapic_lock, flags);
2114 reg_00.raw = io_apic_read(apic_id, 0);
2115 spin_unlock_irqrestore(&ioapic_lock, flags);
2116 if (reg_00.bits.ID != mp_ioapics[apic_id].apicid)
2117 printk("could not set ID!\n");
2119 apic_printk(APIC_VERBOSE, " ok.\n");
2124 int no_timer_check __initdata;
2126 static int __init notimercheck(char *s)
2131 __setup("no_timer_check", notimercheck);
2134 * There is a nasty bug in some older SMP boards, their mptable lies
2135 * about the timer IRQ. We do the following to work around the situation:
2137 * - timer IRQ defaults to IO-APIC IRQ
2138 * - if this function detects that timer IRQs are defunct, then we fall
2139 * back to ISA timer IRQs
2141 static int __init timer_irq_works(void)
2143 unsigned long t1 = jiffies;
2144 unsigned long flags;
2149 local_save_flags(flags);
2151 /* Let ten ticks pass... */
2152 mdelay((10 * 1000) / HZ);
2153 local_irq_restore(flags);
2156 * Expect a few ticks at least, to be sure some possible
2157 * glue logic does not lock up after one or two first
2158 * ticks in a non-ExtINT mode. Also the local APIC
2159 * might have cached one ExtINT interrupt. Finally, at
2160 * least one tick may be lost due to delays.
2164 if (time_after(jiffies, t1 + 4))
2170 * In the SMP+IOAPIC case it might happen that there are an unspecified
2171 * number of pending IRQ events unhandled. These cases are very rare,
2172 * so we 'resend' these IRQs via IPIs, to the same CPU. It's much
2173 * better to do it this way as thus we do not have to be aware of
2174 * 'pending' interrupts in the IRQ path, except at this point.
2177 * Edge triggered needs to resend any interrupt
2178 * that was delayed but this is now handled in the device
2183 * Starting up a edge-triggered IO-APIC interrupt is
2184 * nasty - we need to make sure that we get the edge.
2185 * If it is already asserted for some reason, we need
2186 * return 1 to indicate that is was pending.
2188 * This is not complete - we should be able to fake
2189 * an edge even if it isn't on the 8259A...
2192 static unsigned int startup_ioapic_irq(unsigned int irq)
2194 int was_pending = 0;
2195 unsigned long flags;
2196 struct irq_cfg *cfg;
2198 spin_lock_irqsave(&ioapic_lock, flags);
2199 if (irq < NR_IRQS_LEGACY) {
2200 disable_8259A_irq(irq);
2201 if (i8259A_irq_pending(irq))
2205 __unmask_IO_APIC_irq(cfg);
2206 spin_unlock_irqrestore(&ioapic_lock, flags);
2211 #ifdef CONFIG_X86_64
2212 static int ioapic_retrigger_irq(unsigned int irq)
2215 struct irq_cfg *cfg = irq_cfg(irq);
2216 unsigned long flags;
2218 spin_lock_irqsave(&vector_lock, flags);
2219 apic->send_IPI_mask(cpumask_of(cpumask_first(cfg->domain)), cfg->vector);
2220 spin_unlock_irqrestore(&vector_lock, flags);
2225 static int ioapic_retrigger_irq(unsigned int irq)
2227 apic->send_IPI_self(irq_cfg(irq)->vector);
2234 * Level and edge triggered IO-APIC interrupts need different handling,
2235 * so we use two separate IRQ descriptors. Edge triggered IRQs can be
2236 * handled with the level-triggered descriptor, but that one has slightly
2237 * more overhead. Level-triggered interrupts cannot be handled with the
2238 * edge-triggered handler, without risking IRQ storms and other ugly
2243 static void send_cleanup_vector(struct irq_cfg *cfg)
2245 cpumask_var_t cleanup_mask;
2247 if (unlikely(!alloc_cpumask_var(&cleanup_mask, GFP_ATOMIC))) {
2249 cfg->move_cleanup_count = 0;
2250 for_each_cpu_and(i, cfg->old_domain, cpu_online_mask)
2251 cfg->move_cleanup_count++;
2252 for_each_cpu_and(i, cfg->old_domain, cpu_online_mask)
2253 apic->send_IPI_mask(cpumask_of(i), IRQ_MOVE_CLEANUP_VECTOR);
2255 cpumask_and(cleanup_mask, cfg->old_domain, cpu_online_mask);
2256 cfg->move_cleanup_count = cpumask_weight(cleanup_mask);
2257 apic->send_IPI_mask(cleanup_mask, IRQ_MOVE_CLEANUP_VECTOR);
2258 free_cpumask_var(cleanup_mask);
2260 cfg->move_in_progress = 0;
2263 static void __target_IO_APIC_irq(unsigned int irq, unsigned int dest, struct irq_cfg *cfg)
2266 struct irq_pin_list *entry;
2267 u8 vector = cfg->vector;
2269 entry = cfg->irq_2_pin;
2279 * With interrupt-remapping, destination information comes
2280 * from interrupt-remapping table entry.
2282 if (!irq_remapped(irq))
2283 io_apic_write(apic, 0x11 + pin*2, dest);
2284 reg = io_apic_read(apic, 0x10 + pin*2);
2285 reg &= ~IO_APIC_REDIR_VECTOR_MASK;
2287 io_apic_modify(apic, 0x10 + pin*2, reg);
2290 entry = entry->next;
2295 assign_irq_vector(int irq, struct irq_cfg *cfg, const struct cpumask *mask);
2298 * Either sets desc->affinity to a valid value, and returns
2299 * ->cpu_mask_to_apicid of that, or returns BAD_APICID and
2300 * leaves desc->affinity untouched.
2303 set_desc_affinity(struct irq_desc *desc, const struct cpumask *mask)
2305 struct irq_cfg *cfg;
2308 if (!cpumask_intersects(mask, cpu_online_mask))
2312 cfg = desc->chip_data;
2313 if (assign_irq_vector(irq, cfg, mask))
2316 cpumask_copy(desc->affinity, mask);
2318 return apic->cpu_mask_to_apicid_and(desc->affinity, cfg->domain);
2322 set_ioapic_affinity_irq_desc(struct irq_desc *desc, const struct cpumask *mask)
2324 struct irq_cfg *cfg;
2325 unsigned long flags;
2331 cfg = desc->chip_data;
2333 spin_lock_irqsave(&ioapic_lock, flags);
2334 dest = set_desc_affinity(desc, mask);
2335 if (dest != BAD_APICID) {
2336 /* Only the high 8 bits are valid. */
2337 dest = SET_APIC_LOGICAL_ID(dest);
2338 __target_IO_APIC_irq(irq, dest, cfg);
2341 spin_unlock_irqrestore(&ioapic_lock, flags);
2347 set_ioapic_affinity_irq(unsigned int irq, const struct cpumask *mask)
2349 struct irq_desc *desc;
2351 desc = irq_to_desc(irq);
2353 return set_ioapic_affinity_irq_desc(desc, mask);
2356 #ifdef CONFIG_INTR_REMAP
2359 * Migrate the IO-APIC irq in the presence of intr-remapping.
2361 * For both level and edge triggered, irq migration is a simple atomic
2362 * update(of vector and cpu destination) of IRTE and flush the hardware cache.
2364 * For level triggered, we eliminate the io-apic RTE modification (with the
2365 * updated vector information), by using a virtual vector (io-apic pin number).
2366 * Real vector that is used for interrupting cpu will be coming from
2367 * the interrupt-remapping table entry.
2370 migrate_ioapic_irq_desc(struct irq_desc *desc, const struct cpumask *mask)
2372 struct irq_cfg *cfg;
2378 if (!cpumask_intersects(mask, cpu_online_mask))
2382 if (get_irte(irq, &irte))
2385 cfg = desc->chip_data;
2386 if (assign_irq_vector(irq, cfg, mask))
2389 dest = apic->cpu_mask_to_apicid_and(cfg->domain, mask);
2391 irte.vector = cfg->vector;
2392 irte.dest_id = IRTE_DEST(dest);
2395 * Modified the IRTE and flushes the Interrupt entry cache.
2397 modify_irte(irq, &irte);
2399 if (cfg->move_in_progress)
2400 send_cleanup_vector(cfg);
2402 cpumask_copy(desc->affinity, mask);
2408 * Migrates the IRQ destination in the process context.
2410 static int set_ir_ioapic_affinity_irq_desc(struct irq_desc *desc,
2411 const struct cpumask *mask)
2413 return migrate_ioapic_irq_desc(desc, mask);
2415 static int set_ir_ioapic_affinity_irq(unsigned int irq,
2416 const struct cpumask *mask)
2418 struct irq_desc *desc = irq_to_desc(irq);
2420 return set_ir_ioapic_affinity_irq_desc(desc, mask);
2423 static inline int set_ir_ioapic_affinity_irq_desc(struct irq_desc *desc,
2424 const struct cpumask *mask)
2430 asmlinkage void smp_irq_move_cleanup_interrupt(void)
2432 unsigned vector, me;
2438 me = smp_processor_id();
2439 for (vector = FIRST_EXTERNAL_VECTOR; vector < NR_VECTORS; vector++) {
2442 struct irq_desc *desc;
2443 struct irq_cfg *cfg;
2444 irq = __get_cpu_var(vector_irq)[vector];
2449 desc = irq_to_desc(irq);
2454 spin_lock(&desc->lock);
2455 if (!cfg->move_cleanup_count)
2458 if (vector == cfg->vector && cpumask_test_cpu(me, cfg->domain))
2461 irr = apic_read(APIC_IRR + (vector / 32 * 0x10));
2463 * Check if the vector that needs to be cleanedup is
2464 * registered at the cpu's IRR. If so, then this is not
2465 * the best time to clean it up. Lets clean it up in the
2466 * next attempt by sending another IRQ_MOVE_CLEANUP_VECTOR
2469 if (irr & (1 << (vector % 32))) {
2470 apic->send_IPI_self(IRQ_MOVE_CLEANUP_VECTOR);
2473 __get_cpu_var(vector_irq)[vector] = -1;
2474 cfg->move_cleanup_count--;
2476 spin_unlock(&desc->lock);
2482 static void irq_complete_move(struct irq_desc **descp)
2484 struct irq_desc *desc = *descp;
2485 struct irq_cfg *cfg = desc->chip_data;
2486 unsigned vector, me;
2488 if (likely(!cfg->move_in_progress))
2491 vector = ~get_irq_regs()->orig_ax;
2492 me = smp_processor_id();
2494 if (vector == cfg->vector && cpumask_test_cpu(me, cfg->domain))
2495 send_cleanup_vector(cfg);
2498 static inline void irq_complete_move(struct irq_desc **descp) {}
2501 static void ack_apic_edge(unsigned int irq)
2503 struct irq_desc *desc = irq_to_desc(irq);
2505 irq_complete_move(&desc);
2506 move_native_irq(irq);
2510 atomic_t irq_mis_count;
2512 static void ack_apic_level(unsigned int irq)
2514 struct irq_desc *desc = irq_to_desc(irq);
2516 #ifdef CONFIG_X86_32
2520 struct irq_cfg *cfg;
2521 int do_unmask_irq = 0;
2523 irq_complete_move(&desc);
2524 #ifdef CONFIG_GENERIC_PENDING_IRQ
2525 /* If we are moving the irq we need to mask it */
2526 if (unlikely(desc->status & IRQ_MOVE_PENDING)) {
2528 mask_IO_APIC_irq_desc(desc);
2532 #ifdef CONFIG_X86_32
2534 * It appears there is an erratum which affects at least version 0x11
2535 * of I/O APIC (that's the 82093AA and cores integrated into various
2536 * chipsets). Under certain conditions a level-triggered interrupt is
2537 * erroneously delivered as edge-triggered one but the respective IRR
2538 * bit gets set nevertheless. As a result the I/O unit expects an EOI
2539 * message but it will never arrive and further interrupts are blocked
2540 * from the source. The exact reason is so far unknown, but the
2541 * phenomenon was observed when two consecutive interrupt requests
2542 * from a given source get delivered to the same CPU and the source is
2543 * temporarily disabled in between.
2545 * A workaround is to simulate an EOI message manually. We achieve it
2546 * by setting the trigger mode to edge and then to level when the edge
2547 * trigger mode gets detected in the TMR of a local APIC for a
2548 * level-triggered interrupt. We mask the source for the time of the
2549 * operation to prevent an edge-triggered interrupt escaping meanwhile.
2550 * The idea is from Manfred Spraul. --macro
2552 cfg = desc->chip_data;
2555 v = apic_read(APIC_TMR + ((i & ~0x1f) >> 1));
2559 * We must acknowledge the irq before we move it or the acknowledge will
2560 * not propagate properly.
2564 /* Now we can move and renable the irq */
2565 if (unlikely(do_unmask_irq)) {
2566 /* Only migrate the irq if the ack has been received.
2568 * On rare occasions the broadcast level triggered ack gets
2569 * delayed going to ioapics, and if we reprogram the
2570 * vector while Remote IRR is still set the irq will never
2573 * To prevent this scenario we read the Remote IRR bit
2574 * of the ioapic. This has two effects.
2575 * - On any sane system the read of the ioapic will
2576 * flush writes (and acks) going to the ioapic from
2578 * - We get to see if the ACK has actually been delivered.
2580 * Based on failed experiments of reprogramming the
2581 * ioapic entry from outside of irq context starting
2582 * with masking the ioapic entry and then polling until
2583 * Remote IRR was clear before reprogramming the
2584 * ioapic I don't trust the Remote IRR bit to be
2585 * completey accurate.
2587 * However there appears to be no other way to plug
2588 * this race, so if the Remote IRR bit is not
2589 * accurate and is causing problems then it is a hardware bug
2590 * and you can go talk to the chipset vendor about it.
2592 cfg = desc->chip_data;
2593 if (!io_apic_level_ack_pending(cfg))
2594 move_masked_irq(irq);
2595 unmask_IO_APIC_irq_desc(desc);
2598 #ifdef CONFIG_X86_32
2599 if (!(v & (1 << (i & 0x1f)))) {
2600 atomic_inc(&irq_mis_count);
2601 spin_lock(&ioapic_lock);
2602 __mask_and_edge_IO_APIC_irq(cfg);
2603 __unmask_and_level_IO_APIC_irq(cfg);
2604 spin_unlock(&ioapic_lock);
2609 #ifdef CONFIG_INTR_REMAP
2610 static void __eoi_ioapic_irq(unsigned int irq, struct irq_cfg *cfg)
2613 struct irq_pin_list *entry;
2615 entry = cfg->irq_2_pin;
2623 io_apic_eoi(apic, pin);
2624 entry = entry->next;
2629 eoi_ioapic_irq(struct irq_desc *desc)
2631 struct irq_cfg *cfg;
2632 unsigned long flags;
2636 cfg = desc->chip_data;
2638 spin_lock_irqsave(&ioapic_lock, flags);
2639 __eoi_ioapic_irq(irq, cfg);
2640 spin_unlock_irqrestore(&ioapic_lock, flags);
2643 static void ir_ack_apic_edge(unsigned int irq)
2648 static void ir_ack_apic_level(unsigned int irq)
2650 struct irq_desc *desc = irq_to_desc(irq);
2653 eoi_ioapic_irq(desc);
2655 #endif /* CONFIG_INTR_REMAP */
2657 static struct irq_chip ioapic_chip __read_mostly = {
2659 .startup = startup_ioapic_irq,
2660 .mask = mask_IO_APIC_irq,
2661 .unmask = unmask_IO_APIC_irq,
2662 .ack = ack_apic_edge,
2663 .eoi = ack_apic_level,
2665 .set_affinity = set_ioapic_affinity_irq,
2667 .retrigger = ioapic_retrigger_irq,
2670 static struct irq_chip ir_ioapic_chip __read_mostly = {
2671 .name = "IR-IO-APIC",
2672 .startup = startup_ioapic_irq,
2673 .mask = mask_IO_APIC_irq,
2674 .unmask = unmask_IO_APIC_irq,
2675 #ifdef CONFIG_INTR_REMAP
2676 .ack = ir_ack_apic_edge,
2677 .eoi = ir_ack_apic_level,
2679 .set_affinity = set_ir_ioapic_affinity_irq,
2682 .retrigger = ioapic_retrigger_irq,
2685 static inline void init_IO_APIC_traps(void)
2688 struct irq_desc *desc;
2689 struct irq_cfg *cfg;
2692 * NOTE! The local APIC isn't very good at handling
2693 * multiple interrupts at the same interrupt level.
2694 * As the interrupt level is determined by taking the
2695 * vector number and shifting that right by 4, we
2696 * want to spread these out a bit so that they don't
2697 * all fall in the same interrupt level.
2699 * Also, we've got to be careful not to trash gate
2700 * 0x80, because int 0x80 is hm, kind of importantish. ;)
2702 for_each_irq_desc(irq, desc) {
2703 cfg = desc->chip_data;
2704 if (IO_APIC_IRQ(irq) && cfg && !cfg->vector) {
2706 * Hmm.. We don't have an entry for this,
2707 * so default to an old-fashioned 8259
2708 * interrupt if we can..
2710 if (irq < NR_IRQS_LEGACY)
2711 make_8259A_irq(irq);
2713 /* Strange. Oh, well.. */
2714 desc->chip = &no_irq_chip;
2720 * The local APIC irq-chip implementation:
2723 static void mask_lapic_irq(unsigned int irq)
2727 v = apic_read(APIC_LVT0);
2728 apic_write(APIC_LVT0, v | APIC_LVT_MASKED);
2731 static void unmask_lapic_irq(unsigned int irq)
2735 v = apic_read(APIC_LVT0);
2736 apic_write(APIC_LVT0, v & ~APIC_LVT_MASKED);
2739 static void ack_lapic_irq(unsigned int irq)
2744 static struct irq_chip lapic_chip __read_mostly = {
2745 .name = "local-APIC",
2746 .mask = mask_lapic_irq,
2747 .unmask = unmask_lapic_irq,
2748 .ack = ack_lapic_irq,
2751 static void lapic_register_intr(int irq, struct irq_desc *desc)
2753 desc->status &= ~IRQ_LEVEL;
2754 set_irq_chip_and_handler_name(irq, &lapic_chip, handle_edge_irq,
2758 static void __init setup_nmi(void)
2761 * Dirty trick to enable the NMI watchdog ...
2762 * We put the 8259A master into AEOI mode and
2763 * unmask on all local APICs LVT0 as NMI.
2765 * The idea to use the 8259A in AEOI mode ('8259A Virtual Wire')
2766 * is from Maciej W. Rozycki - so we do not have to EOI from
2767 * the NMI handler or the timer interrupt.
2769 apic_printk(APIC_VERBOSE, KERN_INFO "activating NMI Watchdog ...");
2771 enable_NMI_through_LVT0();
2773 apic_printk(APIC_VERBOSE, " done.\n");
2777 * This looks a bit hackish but it's about the only one way of sending
2778 * a few INTA cycles to 8259As and any associated glue logic. ICR does
2779 * not support the ExtINT mode, unfortunately. We need to send these
2780 * cycles as some i82489DX-based boards have glue logic that keeps the
2781 * 8259A interrupt line asserted until INTA. --macro
2783 static inline void __init unlock_ExtINT_logic(void)
2786 struct IO_APIC_route_entry entry0, entry1;
2787 unsigned char save_control, save_freq_select;
2789 pin = find_isa_irq_pin(8, mp_INT);
2794 apic = find_isa_irq_apic(8, mp_INT);
2800 entry0 = ioapic_read_entry(apic, pin);
2801 clear_IO_APIC_pin(apic, pin);
2803 memset(&entry1, 0, sizeof(entry1));
2805 entry1.dest_mode = 0; /* physical delivery */
2806 entry1.mask = 0; /* unmask IRQ now */
2807 entry1.dest = hard_smp_processor_id();
2808 entry1.delivery_mode = dest_ExtINT;
2809 entry1.polarity = entry0.polarity;
2813 ioapic_write_entry(apic, pin, entry1);
2815 save_control = CMOS_READ(RTC_CONTROL);
2816 save_freq_select = CMOS_READ(RTC_FREQ_SELECT);
2817 CMOS_WRITE((save_freq_select & ~RTC_RATE_SELECT) | 0x6,
2819 CMOS_WRITE(save_control | RTC_PIE, RTC_CONTROL);
2824 if ((CMOS_READ(RTC_INTR_FLAGS) & RTC_PF) == RTC_PF)
2828 CMOS_WRITE(save_control, RTC_CONTROL);
2829 CMOS_WRITE(save_freq_select, RTC_FREQ_SELECT);
2830 clear_IO_APIC_pin(apic, pin);
2832 ioapic_write_entry(apic, pin, entry0);
2835 static int disable_timer_pin_1 __initdata;
2836 /* Actually the next is obsolete, but keep it for paranoid reasons -AK */
2837 static int __init disable_timer_pin_setup(char *arg)
2839 disable_timer_pin_1 = 1;
2842 early_param("disable_timer_pin_1", disable_timer_pin_setup);
2844 int timer_through_8259 __initdata;
2847 * This code may look a bit paranoid, but it's supposed to cooperate with
2848 * a wide range of boards and BIOS bugs. Fortunately only the timer IRQ
2849 * is so screwy. Thanks to Brian Perkins for testing/hacking this beast
2850 * fanatically on his truly buggy board.
2852 * FIXME: really need to revamp this for all platforms.
2854 static inline void __init check_timer(void)
2856 struct irq_desc *desc = irq_to_desc(0);
2857 struct irq_cfg *cfg = desc->chip_data;
2858 int node = cpu_to_node(boot_cpu_id);
2859 int apic1, pin1, apic2, pin2;
2860 unsigned long flags;
2863 local_irq_save(flags);
2866 * get/set the timer IRQ vector:
2868 disable_8259A_irq(0);
2869 assign_irq_vector(0, cfg, apic->target_cpus());
2872 * As IRQ0 is to be enabled in the 8259A, the virtual
2873 * wire has to be disabled in the local APIC. Also
2874 * timer interrupts need to be acknowledged manually in
2875 * the 8259A for the i82489DX when using the NMI
2876 * watchdog as that APIC treats NMIs as level-triggered.
2877 * The AEOI mode will finish them in the 8259A
2880 apic_write(APIC_LVT0, APIC_LVT_MASKED | APIC_DM_EXTINT);
2882 #ifdef CONFIG_X86_32
2886 ver = apic_read(APIC_LVR);
2887 ver = GET_APIC_VERSION(ver);
2888 timer_ack = (nmi_watchdog == NMI_IO_APIC && !APIC_INTEGRATED(ver));
2892 pin1 = find_isa_irq_pin(0, mp_INT);
2893 apic1 = find_isa_irq_apic(0, mp_INT);
2894 pin2 = ioapic_i8259.pin;
2895 apic2 = ioapic_i8259.apic;
2897 apic_printk(APIC_QUIET, KERN_INFO "..TIMER: vector=0x%02X "
2898 "apic1=%d pin1=%d apic2=%d pin2=%d\n",
2899 cfg->vector, apic1, pin1, apic2, pin2);
2902 * Some BIOS writers are clueless and report the ExtINTA
2903 * I/O APIC input from the cascaded 8259A as the timer
2904 * interrupt input. So just in case, if only one pin
2905 * was found above, try it both directly and through the
2909 if (intr_remapping_enabled)
2910 panic("BIOS bug: timer not connected to IO-APIC");
2914 } else if (pin2 == -1) {
2921 * Ok, does IRQ0 through the IOAPIC work?
2924 add_pin_to_irq_node(cfg, node, apic1, pin1);
2925 setup_timer_IRQ0_pin(apic1, pin1, cfg->vector);
2927 /* for edge trigger, setup_IO_APIC_irq already
2928 * leave it unmasked.
2929 * so only need to unmask if it is level-trigger
2930 * do we really have level trigger timer?
2933 idx = find_irq_entry(apic1, pin1, mp_INT);
2934 if (idx != -1 && irq_trigger(idx))
2935 unmask_IO_APIC_irq_desc(desc);
2937 if (timer_irq_works()) {
2938 if (nmi_watchdog == NMI_IO_APIC) {
2940 enable_8259A_irq(0);
2942 if (disable_timer_pin_1 > 0)
2943 clear_IO_APIC_pin(0, pin1);
2946 if (intr_remapping_enabled)
2947 panic("timer doesn't work through Interrupt-remapped IO-APIC");
2948 local_irq_disable();
2949 clear_IO_APIC_pin(apic1, pin1);
2951 apic_printk(APIC_QUIET, KERN_ERR "..MP-BIOS bug: "
2952 "8254 timer not connected to IO-APIC\n");
2954 apic_printk(APIC_QUIET, KERN_INFO "...trying to set up timer "
2955 "(IRQ0) through the 8259A ...\n");
2956 apic_printk(APIC_QUIET, KERN_INFO
2957 "..... (found apic %d pin %d) ...\n", apic2, pin2);
2959 * legacy devices should be connected to IO APIC #0
2961 replace_pin_at_irq_node(cfg, node, apic1, pin1, apic2, pin2);
2962 setup_timer_IRQ0_pin(apic2, pin2, cfg->vector);
2963 enable_8259A_irq(0);
2964 if (timer_irq_works()) {
2965 apic_printk(APIC_QUIET, KERN_INFO "....... works.\n");
2966 timer_through_8259 = 1;
2967 if (nmi_watchdog == NMI_IO_APIC) {
2968 disable_8259A_irq(0);
2970 enable_8259A_irq(0);
2975 * Cleanup, just in case ...
2977 local_irq_disable();
2978 disable_8259A_irq(0);
2979 clear_IO_APIC_pin(apic2, pin2);
2980 apic_printk(APIC_QUIET, KERN_INFO "....... failed.\n");
2983 if (nmi_watchdog == NMI_IO_APIC) {
2984 apic_printk(APIC_QUIET, KERN_WARNING "timer doesn't work "
2985 "through the IO-APIC - disabling NMI Watchdog!\n");
2986 nmi_watchdog = NMI_NONE;
2988 #ifdef CONFIG_X86_32
2992 apic_printk(APIC_QUIET, KERN_INFO
2993 "...trying to set up timer as Virtual Wire IRQ...\n");
2995 lapic_register_intr(0, desc);
2996 apic_write(APIC_LVT0, APIC_DM_FIXED | cfg->vector); /* Fixed mode */
2997 enable_8259A_irq(0);
2999 if (timer_irq_works()) {
3000 apic_printk(APIC_QUIET, KERN_INFO "..... works.\n");
3003 local_irq_disable();
3004 disable_8259A_irq(0);
3005 apic_write(APIC_LVT0, APIC_LVT_MASKED | APIC_DM_FIXED | cfg->vector);
3006 apic_printk(APIC_QUIET, KERN_INFO "..... failed.\n");
3008 apic_printk(APIC_QUIET, KERN_INFO
3009 "...trying to set up timer as ExtINT IRQ...\n");
3013 apic_write(APIC_LVT0, APIC_DM_EXTINT);
3015 unlock_ExtINT_logic();
3017 if (timer_irq_works()) {
3018 apic_printk(APIC_QUIET, KERN_INFO "..... works.\n");
3021 local_irq_disable();
3022 apic_printk(APIC_QUIET, KERN_INFO "..... failed :(.\n");
3023 panic("IO-APIC + timer doesn't work! Boot with apic=debug and send a "
3024 "report. Then try booting with the 'noapic' option.\n");
3026 local_irq_restore(flags);
3030 * Traditionally ISA IRQ2 is the cascade IRQ, and is not available
3031 * to devices. However there may be an I/O APIC pin available for
3032 * this interrupt regardless. The pin may be left unconnected, but
3033 * typically it will be reused as an ExtINT cascade interrupt for
3034 * the master 8259A. In the MPS case such a pin will normally be
3035 * reported as an ExtINT interrupt in the MP table. With ACPI
3036 * there is no provision for ExtINT interrupts, and in the absence
3037 * of an override it would be treated as an ordinary ISA I/O APIC
3038 * interrupt, that is edge-triggered and unmasked by default. We
3039 * used to do this, but it caused problems on some systems because
3040 * of the NMI watchdog and sometimes IRQ0 of the 8254 timer using
3041 * the same ExtINT cascade interrupt to drive the local APIC of the
3042 * bootstrap processor. Therefore we refrain from routing IRQ2 to
3043 * the I/O APIC in all cases now. No actual device should request
3044 * it anyway. --macro
3046 #define PIC_IRQS (1 << PIC_CASCADE_IR)
3048 void __init setup_IO_APIC(void)
3052 * calling enable_IO_APIC() is moved to setup_local_APIC for BP
3055 io_apic_irqs = ~PIC_IRQS;
3057 apic_printk(APIC_VERBOSE, "ENABLING IO-APIC IRQs\n");
3059 * Set up IO-APIC IRQ routing.
3061 #ifdef CONFIG_X86_32
3063 setup_ioapic_ids_from_mpc();
3066 setup_IO_APIC_irqs();
3067 init_IO_APIC_traps();
3072 * Called after all the initialization is done. If we didnt find any
3073 * APIC bugs then we can allow the modify fast path
3076 static int __init io_apic_bug_finalize(void)
3078 if (sis_apic_bug == -1)
3083 late_initcall(io_apic_bug_finalize);
3085 struct sysfs_ioapic_data {
3086 struct sys_device dev;
3087 struct IO_APIC_route_entry entry[0];
3089 static struct sysfs_ioapic_data * mp_ioapic_data[MAX_IO_APICS];
3091 static int ioapic_suspend(struct sys_device *dev, pm_message_t state)
3093 struct IO_APIC_route_entry *entry;
3094 struct sysfs_ioapic_data *data;
3097 data = container_of(dev, struct sysfs_ioapic_data, dev);
3098 entry = data->entry;
3099 for (i = 0; i < nr_ioapic_registers[dev->id]; i ++, entry ++ )
3100 *entry = ioapic_read_entry(dev->id, i);
3105 static int ioapic_resume(struct sys_device *dev)
3107 struct IO_APIC_route_entry *entry;
3108 struct sysfs_ioapic_data *data;
3109 unsigned long flags;
3110 union IO_APIC_reg_00 reg_00;
3113 data = container_of(dev, struct sysfs_ioapic_data, dev);
3114 entry = data->entry;
3116 spin_lock_irqsave(&ioapic_lock, flags);
3117 reg_00.raw = io_apic_read(dev->id, 0);
3118 if (reg_00.bits.ID != mp_ioapics[dev->id].apicid) {
3119 reg_00.bits.ID = mp_ioapics[dev->id].apicid;
3120 io_apic_write(dev->id, 0, reg_00.raw);
3122 spin_unlock_irqrestore(&ioapic_lock, flags);
3123 for (i = 0; i < nr_ioapic_registers[dev->id]; i++)
3124 ioapic_write_entry(dev->id, i, entry[i]);
3129 static struct sysdev_class ioapic_sysdev_class = {
3131 .suspend = ioapic_suspend,
3132 .resume = ioapic_resume,
3135 static int __init ioapic_init_sysfs(void)
3137 struct sys_device * dev;
3140 error = sysdev_class_register(&ioapic_sysdev_class);
3144 for (i = 0; i < nr_ioapics; i++ ) {
3145 size = sizeof(struct sys_device) + nr_ioapic_registers[i]
3146 * sizeof(struct IO_APIC_route_entry);
3147 mp_ioapic_data[i] = kzalloc(size, GFP_KERNEL);
3148 if (!mp_ioapic_data[i]) {
3149 printk(KERN_ERR "Can't suspend/resume IOAPIC %d\n", i);
3152 dev = &mp_ioapic_data[i]->dev;
3154 dev->cls = &ioapic_sysdev_class;
3155 error = sysdev_register(dev);
3157 kfree(mp_ioapic_data[i]);
3158 mp_ioapic_data[i] = NULL;
3159 printk(KERN_ERR "Can't suspend/resume IOAPIC %d\n", i);
3167 device_initcall(ioapic_init_sysfs);
3169 static int nr_irqs_gsi = NR_IRQS_LEGACY;
3171 * Dynamic irq allocate and deallocation
3173 unsigned int create_irq_nr(unsigned int irq_want, int node)
3175 /* Allocate an unused irq */
3178 unsigned long flags;
3179 struct irq_cfg *cfg_new = NULL;
3180 struct irq_desc *desc_new = NULL;
3183 if (irq_want < nr_irqs_gsi)
3184 irq_want = nr_irqs_gsi;
3186 spin_lock_irqsave(&vector_lock, flags);
3187 for (new = irq_want; new < nr_irqs; new++) {
3188 desc_new = irq_to_desc_alloc_node(new, node);
3190 printk(KERN_INFO "can not get irq_desc for %d\n", new);
3193 cfg_new = desc_new->chip_data;
3195 if (cfg_new->vector != 0)
3198 desc_new = move_irq_desc(desc_new, node);
3200 if (__assign_irq_vector(new, cfg_new, apic->target_cpus()) == 0)
3204 spin_unlock_irqrestore(&vector_lock, flags);
3207 dynamic_irq_init(irq);
3208 /* restore it, in case dynamic_irq_init clear it */
3210 desc_new->chip_data = cfg_new;
3215 int create_irq(void)
3217 int node = cpu_to_node(boot_cpu_id);
3218 unsigned int irq_want;
3221 irq_want = nr_irqs_gsi;
3222 irq = create_irq_nr(irq_want, node);
3230 void destroy_irq(unsigned int irq)
3232 unsigned long flags;
3233 struct irq_cfg *cfg;
3234 struct irq_desc *desc;
3236 /* store it, in case dynamic_irq_cleanup clear it */
3237 desc = irq_to_desc(irq);
3238 cfg = desc->chip_data;
3239 dynamic_irq_cleanup(irq);
3240 /* connect back irq_cfg */
3242 desc->chip_data = cfg;
3245 spin_lock_irqsave(&vector_lock, flags);
3246 __clear_irq_vector(irq, cfg);
3247 spin_unlock_irqrestore(&vector_lock, flags);
3251 * MSI message composition
3253 #ifdef CONFIG_PCI_MSI
3254 static int msi_compose_msg(struct pci_dev *pdev, unsigned int irq, struct msi_msg *msg)
3256 struct irq_cfg *cfg;
3264 err = assign_irq_vector(irq, cfg, apic->target_cpus());
3268 dest = apic->cpu_mask_to_apicid_and(cfg->domain, apic->target_cpus());
3270 if (irq_remapped(irq)) {
3275 ir_index = map_irq_to_irte_handle(irq, &sub_handle);
3276 BUG_ON(ir_index == -1);
3278 memset (&irte, 0, sizeof(irte));
3281 irte.dst_mode = apic->irq_dest_mode;
3282 irte.trigger_mode = 0; /* edge */
3283 irte.dlvry_mode = apic->irq_delivery_mode;
3284 irte.vector = cfg->vector;
3285 irte.dest_id = IRTE_DEST(dest);
3287 modify_irte(irq, &irte);
3289 msg->address_hi = MSI_ADDR_BASE_HI;
3290 msg->data = sub_handle;
3291 msg->address_lo = MSI_ADDR_BASE_LO | MSI_ADDR_IR_EXT_INT |
3293 MSI_ADDR_IR_INDEX1(ir_index) |
3294 MSI_ADDR_IR_INDEX2(ir_index);
3296 if (x2apic_enabled())
3297 msg->address_hi = MSI_ADDR_BASE_HI |
3298 MSI_ADDR_EXT_DEST_ID(dest);
3300 msg->address_hi = MSI_ADDR_BASE_HI;
3304 ((apic->irq_dest_mode == 0) ?
3305 MSI_ADDR_DEST_MODE_PHYSICAL:
3306 MSI_ADDR_DEST_MODE_LOGICAL) |
3307 ((apic->irq_delivery_mode != dest_LowestPrio) ?
3308 MSI_ADDR_REDIRECTION_CPU:
3309 MSI_ADDR_REDIRECTION_LOWPRI) |
3310 MSI_ADDR_DEST_ID(dest);
3313 MSI_DATA_TRIGGER_EDGE |
3314 MSI_DATA_LEVEL_ASSERT |
3315 ((apic->irq_delivery_mode != dest_LowestPrio) ?
3316 MSI_DATA_DELIVERY_FIXED:
3317 MSI_DATA_DELIVERY_LOWPRI) |
3318 MSI_DATA_VECTOR(cfg->vector);
3324 static int set_msi_irq_affinity(unsigned int irq, const struct cpumask *mask)
3326 struct irq_desc *desc = irq_to_desc(irq);
3327 struct irq_cfg *cfg;
3331 dest = set_desc_affinity(desc, mask);
3332 if (dest == BAD_APICID)
3335 cfg = desc->chip_data;
3337 read_msi_msg_desc(desc, &msg);
3339 msg.data &= ~MSI_DATA_VECTOR_MASK;
3340 msg.data |= MSI_DATA_VECTOR(cfg->vector);
3341 msg.address_lo &= ~MSI_ADDR_DEST_ID_MASK;
3342 msg.address_lo |= MSI_ADDR_DEST_ID(dest);
3344 write_msi_msg_desc(desc, &msg);
3348 #ifdef CONFIG_INTR_REMAP
3350 * Migrate the MSI irq to another cpumask. This migration is
3351 * done in the process context using interrupt-remapping hardware.
3354 ir_set_msi_irq_affinity(unsigned int irq, const struct cpumask *mask)
3356 struct irq_desc *desc = irq_to_desc(irq);
3357 struct irq_cfg *cfg = desc->chip_data;
3361 if (get_irte(irq, &irte))
3364 dest = set_desc_affinity(desc, mask);
3365 if (dest == BAD_APICID)
3368 irte.vector = cfg->vector;
3369 irte.dest_id = IRTE_DEST(dest);
3372 * atomically update the IRTE with the new destination and vector.
3374 modify_irte(irq, &irte);
3377 * After this point, all the interrupts will start arriving
3378 * at the new destination. So, time to cleanup the previous
3379 * vector allocation.
3381 if (cfg->move_in_progress)
3382 send_cleanup_vector(cfg);
3388 #endif /* CONFIG_SMP */
3391 * IRQ Chip for MSI PCI/PCI-X/PCI-Express Devices,
3392 * which implement the MSI or MSI-X Capability Structure.
3394 static struct irq_chip msi_chip = {
3396 .unmask = unmask_msi_irq,
3397 .mask = mask_msi_irq,
3398 .ack = ack_apic_edge,
3400 .set_affinity = set_msi_irq_affinity,
3402 .retrigger = ioapic_retrigger_irq,
3405 static struct irq_chip msi_ir_chip = {
3406 .name = "IR-PCI-MSI",
3407 .unmask = unmask_msi_irq,
3408 .mask = mask_msi_irq,
3409 #ifdef CONFIG_INTR_REMAP
3410 .ack = ir_ack_apic_edge,
3412 .set_affinity = ir_set_msi_irq_affinity,
3415 .retrigger = ioapic_retrigger_irq,
3419 * Map the PCI dev to the corresponding remapping hardware unit
3420 * and allocate 'nvec' consecutive interrupt-remapping table entries
3423 static int msi_alloc_irte(struct pci_dev *dev, int irq, int nvec)
3425 struct intel_iommu *iommu;
3428 iommu = map_dev_to_ir(dev);
3431 "Unable to map PCI %s to iommu\n", pci_name(dev));
3435 index = alloc_irte(iommu, irq, nvec);
3438 "Unable to allocate %d IRTE for PCI %s\n", nvec,
3445 static int setup_msi_irq(struct pci_dev *dev, struct msi_desc *msidesc, int irq)
3450 ret = msi_compose_msg(dev, irq, &msg);
3454 set_irq_msi(irq, msidesc);
3455 write_msi_msg(irq, &msg);
3457 if (irq_remapped(irq)) {
3458 struct irq_desc *desc = irq_to_desc(irq);
3460 * irq migration in process context
3462 desc->status |= IRQ_MOVE_PCNTXT;
3463 set_irq_chip_and_handler_name(irq, &msi_ir_chip, handle_edge_irq, "edge");
3465 set_irq_chip_and_handler_name(irq, &msi_chip, handle_edge_irq, "edge");
3467 dev_printk(KERN_DEBUG, &dev->dev, "irq %d for MSI/MSI-X\n", irq);
3472 int arch_setup_msi_irqs(struct pci_dev *dev, int nvec, int type)
3475 int ret, sub_handle;
3476 struct msi_desc *msidesc;
3477 unsigned int irq_want;
3478 struct intel_iommu *iommu = NULL;
3482 /* x86 doesn't support multiple MSI yet */
3483 if (type == PCI_CAP_ID_MSI && nvec > 1)
3486 node = dev_to_node(&dev->dev);
3487 irq_want = nr_irqs_gsi;
3489 list_for_each_entry(msidesc, &dev->msi_list, list) {
3490 irq = create_irq_nr(irq_want, node);
3494 if (!intr_remapping_enabled)
3499 * allocate the consecutive block of IRTE's
3502 index = msi_alloc_irte(dev, irq, nvec);
3508 iommu = map_dev_to_ir(dev);
3514 * setup the mapping between the irq and the IRTE
3515 * base index, the sub_handle pointing to the
3516 * appropriate interrupt remap table entry.
3518 set_irte_irq(irq, iommu, index, sub_handle);
3521 ret = setup_msi_irq(dev, msidesc, irq);
3533 void arch_teardown_msi_irq(unsigned int irq)
3538 #if defined (CONFIG_DMAR) || defined (CONFIG_INTR_REMAP)
3540 static int dmar_msi_set_affinity(unsigned int irq, const struct cpumask *mask)
3542 struct irq_desc *desc = irq_to_desc(irq);
3543 struct irq_cfg *cfg;
3547 dest = set_desc_affinity(desc, mask);
3548 if (dest == BAD_APICID)
3551 cfg = desc->chip_data;
3553 dmar_msi_read(irq, &msg);
3555 msg.data &= ~MSI_DATA_VECTOR_MASK;
3556 msg.data |= MSI_DATA_VECTOR(cfg->vector);
3557 msg.address_lo &= ~MSI_ADDR_DEST_ID_MASK;
3558 msg.address_lo |= MSI_ADDR_DEST_ID(dest);
3560 dmar_msi_write(irq, &msg);
3565 #endif /* CONFIG_SMP */
3567 struct irq_chip dmar_msi_type = {
3569 .unmask = dmar_msi_unmask,
3570 .mask = dmar_msi_mask,
3571 .ack = ack_apic_edge,
3573 .set_affinity = dmar_msi_set_affinity,
3575 .retrigger = ioapic_retrigger_irq,
3578 int arch_setup_dmar_msi(unsigned int irq)
3583 ret = msi_compose_msg(NULL, irq, &msg);
3586 dmar_msi_write(irq, &msg);
3587 set_irq_chip_and_handler_name(irq, &dmar_msi_type, handle_edge_irq,
3593 #ifdef CONFIG_HPET_TIMER
3596 static int hpet_msi_set_affinity(unsigned int irq, const struct cpumask *mask)
3598 struct irq_desc *desc = irq_to_desc(irq);
3599 struct irq_cfg *cfg;
3603 dest = set_desc_affinity(desc, mask);
3604 if (dest == BAD_APICID)
3607 cfg = desc->chip_data;
3609 hpet_msi_read(irq, &msg);
3611 msg.data &= ~MSI_DATA_VECTOR_MASK;
3612 msg.data |= MSI_DATA_VECTOR(cfg->vector);
3613 msg.address_lo &= ~MSI_ADDR_DEST_ID_MASK;
3614 msg.address_lo |= MSI_ADDR_DEST_ID(dest);
3616 hpet_msi_write(irq, &msg);
3621 #endif /* CONFIG_SMP */
3623 static struct irq_chip hpet_msi_type = {
3625 .unmask = hpet_msi_unmask,
3626 .mask = hpet_msi_mask,
3627 .ack = ack_apic_edge,
3629 .set_affinity = hpet_msi_set_affinity,
3631 .retrigger = ioapic_retrigger_irq,
3634 int arch_setup_hpet_msi(unsigned int irq)
3638 struct irq_desc *desc = irq_to_desc(irq);
3640 ret = msi_compose_msg(NULL, irq, &msg);
3644 hpet_msi_write(irq, &msg);
3645 desc->status |= IRQ_MOVE_PCNTXT;
3646 set_irq_chip_and_handler_name(irq, &hpet_msi_type, handle_edge_irq,
3653 #endif /* CONFIG_PCI_MSI */
3655 * Hypertransport interrupt support
3657 #ifdef CONFIG_HT_IRQ
3661 static void target_ht_irq(unsigned int irq, unsigned int dest, u8 vector)
3663 struct ht_irq_msg msg;
3664 fetch_ht_irq_msg(irq, &msg);
3666 msg.address_lo &= ~(HT_IRQ_LOW_VECTOR_MASK | HT_IRQ_LOW_DEST_ID_MASK);
3667 msg.address_hi &= ~(HT_IRQ_HIGH_DEST_ID_MASK);
3669 msg.address_lo |= HT_IRQ_LOW_VECTOR(vector) | HT_IRQ_LOW_DEST_ID(dest);
3670 msg.address_hi |= HT_IRQ_HIGH_DEST_ID(dest);
3672 write_ht_irq_msg(irq, &msg);
3675 static int set_ht_irq_affinity(unsigned int irq, const struct cpumask *mask)
3677 struct irq_desc *desc = irq_to_desc(irq);
3678 struct irq_cfg *cfg;
3681 dest = set_desc_affinity(desc, mask);
3682 if (dest == BAD_APICID)
3685 cfg = desc->chip_data;
3687 target_ht_irq(irq, dest, cfg->vector);
3694 static struct irq_chip ht_irq_chip = {
3696 .mask = mask_ht_irq,
3697 .unmask = unmask_ht_irq,
3698 .ack = ack_apic_edge,
3700 .set_affinity = set_ht_irq_affinity,
3702 .retrigger = ioapic_retrigger_irq,
3705 int arch_setup_ht_irq(unsigned int irq, struct pci_dev *dev)
3707 struct irq_cfg *cfg;
3714 err = assign_irq_vector(irq, cfg, apic->target_cpus());
3716 struct ht_irq_msg msg;
3719 dest = apic->cpu_mask_to_apicid_and(cfg->domain,
3720 apic->target_cpus());
3722 msg.address_hi = HT_IRQ_HIGH_DEST_ID(dest);
3726 HT_IRQ_LOW_DEST_ID(dest) |
3727 HT_IRQ_LOW_VECTOR(cfg->vector) |
3728 ((apic->irq_dest_mode == 0) ?
3729 HT_IRQ_LOW_DM_PHYSICAL :
3730 HT_IRQ_LOW_DM_LOGICAL) |
3731 HT_IRQ_LOW_RQEOI_EDGE |
3732 ((apic->irq_delivery_mode != dest_LowestPrio) ?
3733 HT_IRQ_LOW_MT_FIXED :
3734 HT_IRQ_LOW_MT_ARBITRATED) |
3735 HT_IRQ_LOW_IRQ_MASKED;
3737 write_ht_irq_msg(irq, &msg);
3739 set_irq_chip_and_handler_name(irq, &ht_irq_chip,
3740 handle_edge_irq, "edge");
3742 dev_printk(KERN_DEBUG, &dev->dev, "irq %d for HT\n", irq);
3746 #endif /* CONFIG_HT_IRQ */
3748 #ifdef CONFIG_X86_UV
3750 * Re-target the irq to the specified CPU and enable the specified MMR located
3751 * on the specified blade to allow the sending of MSIs to the specified CPU.
3753 int arch_enable_uv_irq(char *irq_name, unsigned int irq, int cpu, int mmr_blade,
3754 unsigned long mmr_offset)
3756 const struct cpumask *eligible_cpu = cpumask_of(cpu);
3757 struct irq_cfg *cfg;
3759 unsigned long mmr_value;
3760 struct uv_IO_APIC_route_entry *entry;
3761 unsigned long flags;
3764 BUILD_BUG_ON(sizeof(struct uv_IO_APIC_route_entry) != sizeof(unsigned long));
3768 err = assign_irq_vector(irq, cfg, eligible_cpu);
3772 spin_lock_irqsave(&vector_lock, flags);
3773 set_irq_chip_and_handler_name(irq, &uv_irq_chip, handle_percpu_irq,
3775 spin_unlock_irqrestore(&vector_lock, flags);
3778 entry = (struct uv_IO_APIC_route_entry *)&mmr_value;
3779 entry->vector = cfg->vector;
3780 entry->delivery_mode = apic->irq_delivery_mode;
3781 entry->dest_mode = apic->irq_dest_mode;
3782 entry->polarity = 0;
3785 entry->dest = apic->cpu_mask_to_apicid(eligible_cpu);
3787 mmr_pnode = uv_blade_to_pnode(mmr_blade);
3788 uv_write_global_mmr64(mmr_pnode, mmr_offset, mmr_value);
3794 * Disable the specified MMR located on the specified blade so that MSIs are
3795 * longer allowed to be sent.
3797 void arch_disable_uv_irq(int mmr_blade, unsigned long mmr_offset)
3799 unsigned long mmr_value;
3800 struct uv_IO_APIC_route_entry *entry;
3803 BUILD_BUG_ON(sizeof(struct uv_IO_APIC_route_entry) != sizeof(unsigned long));
3806 entry = (struct uv_IO_APIC_route_entry *)&mmr_value;
3809 mmr_pnode = uv_blade_to_pnode(mmr_blade);
3810 uv_write_global_mmr64(mmr_pnode, mmr_offset, mmr_value);
3812 #endif /* CONFIG_X86_64 */
3814 int __init io_apic_get_redir_entries (int ioapic)
3816 union IO_APIC_reg_01 reg_01;
3817 unsigned long flags;
3819 spin_lock_irqsave(&ioapic_lock, flags);
3820 reg_01.raw = io_apic_read(ioapic, 1);
3821 spin_unlock_irqrestore(&ioapic_lock, flags);
3823 return reg_01.bits.entries;
3826 void __init probe_nr_irqs_gsi(void)
3830 nr = acpi_probe_gsi();
3831 if (nr > nr_irqs_gsi) {
3834 /* for acpi=off or acpi is not compiled in */
3838 for (idx = 0; idx < nr_ioapics; idx++)
3839 nr += io_apic_get_redir_entries(idx) + 1;
3841 if (nr > nr_irqs_gsi)
3845 printk(KERN_DEBUG "nr_irqs_gsi: %d\n", nr_irqs_gsi);
3848 #ifdef CONFIG_SPARSE_IRQ
3849 int __init arch_probe_nr_irqs(void)
3853 if (nr_irqs > (NR_VECTORS * nr_cpu_ids))
3854 nr_irqs = NR_VECTORS * nr_cpu_ids;
3856 nr = nr_irqs_gsi + 8 * nr_cpu_ids;
3857 #if defined(CONFIG_PCI_MSI) || defined(CONFIG_HT_IRQ)
3859 * for MSI and HT dyn irq
3861 nr += nr_irqs_gsi * 16;
3870 static int __io_apic_set_pci_routing(struct device *dev, int irq,
3871 struct io_apic_irq_attr *irq_attr)
3873 struct irq_desc *desc;
3874 struct irq_cfg *cfg;
3877 int trigger, polarity;
3879 ioapic = irq_attr->ioapic;
3880 if (!IO_APIC_IRQ(irq)) {
3881 apic_printk(APIC_QUIET,KERN_ERR "IOAPIC[%d]: Invalid reference to IRQ 0\n",
3887 node = dev_to_node(dev);
3889 node = cpu_to_node(boot_cpu_id);
3891 desc = irq_to_desc_alloc_node(irq, node);
3893 printk(KERN_INFO "can not get irq_desc %d\n", irq);
3897 pin = irq_attr->ioapic_pin;
3898 trigger = irq_attr->trigger;
3899 polarity = irq_attr->polarity;
3902 * IRQs < 16 are already in the irq_2_pin[] map
3904 if (irq >= NR_IRQS_LEGACY) {
3905 cfg = desc->chip_data;
3906 add_pin_to_irq_node(cfg, node, ioapic, pin);
3909 setup_IO_APIC_irq(ioapic, pin, irq, desc, trigger, polarity);
3914 int io_apic_set_pci_routing(struct device *dev, int irq,
3915 struct io_apic_irq_attr *irq_attr)
3919 * Avoid pin reprogramming. PRTs typically include entries
3920 * with redundant pin->gsi mappings (but unique PCI devices);
3921 * we only program the IOAPIC on the first.
3923 ioapic = irq_attr->ioapic;
3924 pin = irq_attr->ioapic_pin;
3925 if (test_bit(pin, mp_ioapic_routing[ioapic].pin_programmed)) {
3926 pr_debug("Pin %d-%d already programmed\n",
3927 mp_ioapics[ioapic].apicid, pin);
3930 set_bit(pin, mp_ioapic_routing[ioapic].pin_programmed);
3932 return __io_apic_set_pci_routing(dev, irq, irq_attr);
3935 /* --------------------------------------------------------------------------
3936 ACPI-based IOAPIC Configuration
3937 -------------------------------------------------------------------------- */
3941 #ifdef CONFIG_X86_32
3942 int __init io_apic_get_unique_id(int ioapic, int apic_id)
3944 union IO_APIC_reg_00 reg_00;
3945 static physid_mask_t apic_id_map = PHYSID_MASK_NONE;
3947 unsigned long flags;
3951 * The P4 platform supports up to 256 APIC IDs on two separate APIC
3952 * buses (one for LAPICs, one for IOAPICs), where predecessors only
3953 * supports up to 16 on one shared APIC bus.
3955 * TBD: Expand LAPIC/IOAPIC support on P4-class systems to take full
3956 * advantage of new APIC bus architecture.
3959 if (physids_empty(apic_id_map))
3960 apic_id_map = apic->ioapic_phys_id_map(phys_cpu_present_map);
3962 spin_lock_irqsave(&ioapic_lock, flags);
3963 reg_00.raw = io_apic_read(ioapic, 0);
3964 spin_unlock_irqrestore(&ioapic_lock, flags);
3966 if (apic_id >= get_physical_broadcast()) {
3967 printk(KERN_WARNING "IOAPIC[%d]: Invalid apic_id %d, trying "
3968 "%d\n", ioapic, apic_id, reg_00.bits.ID);
3969 apic_id = reg_00.bits.ID;
3973 * Every APIC in a system must have a unique ID or we get lots of nice
3974 * 'stuck on smp_invalidate_needed IPI wait' messages.
3976 if (apic->check_apicid_used(apic_id_map, apic_id)) {
3978 for (i = 0; i < get_physical_broadcast(); i++) {
3979 if (!apic->check_apicid_used(apic_id_map, i))
3983 if (i == get_physical_broadcast())
3984 panic("Max apic_id exceeded!\n");
3986 printk(KERN_WARNING "IOAPIC[%d]: apic_id %d already used, "
3987 "trying %d\n", ioapic, apic_id, i);
3992 tmp = apic->apicid_to_cpu_present(apic_id);
3993 physids_or(apic_id_map, apic_id_map, tmp);
3995 if (reg_00.bits.ID != apic_id) {
3996 reg_00.bits.ID = apic_id;
3998 spin_lock_irqsave(&ioapic_lock, flags);
3999 io_apic_write(ioapic, 0, reg_00.raw);
4000 reg_00.raw = io_apic_read(ioapic, 0);
4001 spin_unlock_irqrestore(&ioapic_lock, flags);
4004 if (reg_00.bits.ID != apic_id) {
4005 printk("IOAPIC[%d]: Unable to change apic_id!\n", ioapic);
4010 apic_printk(APIC_VERBOSE, KERN_INFO
4011 "IOAPIC[%d]: Assigned apic_id %d\n", ioapic, apic_id);
4016 int __init io_apic_get_version(int ioapic)
4018 union IO_APIC_reg_01 reg_01;
4019 unsigned long flags;
4021 spin_lock_irqsave(&ioapic_lock, flags);
4022 reg_01.raw = io_apic_read(ioapic, 1);
4023 spin_unlock_irqrestore(&ioapic_lock, flags);
4025 return reg_01.bits.version;
4029 int acpi_get_override_irq(int bus_irq, int *trigger, int *polarity)
4033 if (skip_ioapic_setup)
4036 for (i = 0; i < mp_irq_entries; i++)
4037 if (mp_irqs[i].irqtype == mp_INT &&
4038 mp_irqs[i].srcbusirq == bus_irq)
4040 if (i >= mp_irq_entries)
4043 *trigger = irq_trigger(i);
4044 *polarity = irq_polarity(i);
4048 #endif /* CONFIG_ACPI */
4051 * This function currently is only a helper for the i386 smp boot process where
4052 * we need to reprogram the ioredtbls to cater for the cpus which have come online
4053 * so mask in all cases should simply be apic->target_cpus()
4056 void __init setup_ioapic_dest(void)
4058 int pin, ioapic = 0, irq, irq_entry;
4059 struct irq_desc *desc;
4060 const struct cpumask *mask;
4062 if (skip_ioapic_setup == 1)
4066 if (!acpi_disabled && acpi_ioapic) {
4067 ioapic = mp_find_ioapic(0);
4073 for (pin = 0; pin < nr_ioapic_registers[ioapic]; pin++) {
4074 irq_entry = find_irq_entry(ioapic, pin, mp_INT);
4075 if (irq_entry == -1)
4077 irq = pin_2_irq(irq_entry, ioapic, pin);
4079 desc = irq_to_desc(irq);
4082 * Honour affinities which have been set in early boot
4085 (IRQ_NO_BALANCING | IRQ_AFFINITY_SET))
4086 mask = desc->affinity;
4088 mask = apic->target_cpus();
4090 if (intr_remapping_enabled)
4091 set_ir_ioapic_affinity_irq_desc(desc, mask);
4093 set_ioapic_affinity_irq_desc(desc, mask);
4099 #define IOAPIC_RESOURCE_NAME_SIZE 11
4101 static struct resource *ioapic_resources;
4103 static struct resource * __init ioapic_setup_resources(void)
4106 struct resource *res;
4110 if (nr_ioapics <= 0)
4113 n = IOAPIC_RESOURCE_NAME_SIZE + sizeof(struct resource);
4116 mem = alloc_bootmem(n);
4120 mem += sizeof(struct resource) * nr_ioapics;
4122 for (i = 0; i < nr_ioapics; i++) {
4124 res[i].flags = IORESOURCE_MEM | IORESOURCE_BUSY;
4125 sprintf(mem, "IOAPIC %u", i);
4126 mem += IOAPIC_RESOURCE_NAME_SIZE;
4130 ioapic_resources = res;
4135 void __init ioapic_init_mappings(void)
4137 unsigned long ioapic_phys, idx = FIX_IO_APIC_BASE_0;
4138 struct resource *ioapic_res;
4141 ioapic_res = ioapic_setup_resources();
4142 for (i = 0; i < nr_ioapics; i++) {
4143 if (smp_found_config) {
4144 ioapic_phys = mp_ioapics[i].apicaddr;
4145 #ifdef CONFIG_X86_32
4148 "WARNING: bogus zero IO-APIC "
4149 "address found in MPTABLE, "
4150 "disabling IO/APIC support!\n");
4151 smp_found_config = 0;
4152 skip_ioapic_setup = 1;
4153 goto fake_ioapic_page;
4157 #ifdef CONFIG_X86_32
4160 ioapic_phys = (unsigned long)
4161 alloc_bootmem_pages(PAGE_SIZE);
4162 ioapic_phys = __pa(ioapic_phys);
4164 set_fixmap_nocache(idx, ioapic_phys);
4165 apic_printk(APIC_VERBOSE,
4166 "mapped IOAPIC to %08lx (%08lx)\n",
4167 __fix_to_virt(idx), ioapic_phys);
4170 if (ioapic_res != NULL) {
4171 ioapic_res->start = ioapic_phys;
4172 ioapic_res->end = ioapic_phys + (4 * 1024) - 1;
4178 static int __init ioapic_insert_resources(void)
4181 struct resource *r = ioapic_resources;
4184 if (nr_ioapics > 0) {
4186 "IO APIC resources couldn't be allocated.\n");
4192 for (i = 0; i < nr_ioapics; i++) {
4193 insert_resource(&iomem_resource, r);
4200 /* Insert the IO APIC resources after PCI initialization has occured to handle
4201 * IO APICS that are mapped in on a BAR in PCI space. */
4202 late_initcall(ioapic_insert_resources);