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/hw_irq.h>
63 #include <asm/uv/uv_hub.h>
64 #include <asm/uv/uv_irq.h>
68 #define __apicdebuginit(type) static type __init
71 * Is the SiS APIC rmw bug present ?
72 * -1 = don't know, 0 = no, 1 = yes
74 int sis_apic_bug = -1;
76 static DEFINE_SPINLOCK(ioapic_lock);
77 static DEFINE_SPINLOCK(vector_lock);
80 * # of IRQ routing registers
82 int nr_ioapic_registers[MAX_IO_APICS];
84 /* I/O APIC entries */
85 struct mpc_ioapic mp_ioapics[MAX_IO_APICS];
88 /* MP IRQ source entries */
89 struct mpc_intsrc mp_irqs[MAX_IRQ_SOURCES];
91 /* # of MP IRQ source entries */
94 #if defined (CONFIG_MCA) || defined (CONFIG_EISA)
95 int mp_bus_id_to_type[MAX_MP_BUSSES];
98 DECLARE_BITMAP(mp_bus_not_pci, MAX_MP_BUSSES);
100 int skip_ioapic_setup;
102 void arch_disable_smp_support(void)
106 noioapicreroute = -1;
108 skip_ioapic_setup = 1;
111 static int __init parse_noapic(char *str)
113 /* disable IO-APIC */
114 arch_disable_smp_support();
117 early_param("noapic", parse_noapic);
122 * This is performance-critical, we want to do it O(1)
124 * the indexing order of this array favors 1:1 mappings
125 * between pins and IRQs.
128 struct irq_pin_list {
130 struct irq_pin_list *next;
133 static struct irq_pin_list *get_one_free_irq_2_pin(int node)
135 struct irq_pin_list *pin;
137 pin = kzalloc_node(sizeof(*pin), GFP_ATOMIC, node);
143 struct irq_pin_list *irq_2_pin;
144 cpumask_var_t domain;
145 cpumask_var_t old_domain;
146 unsigned move_cleanup_count;
148 u8 move_in_progress : 1;
151 /* irq_cfg is indexed by the sum of all RTEs in all I/O APICs. */
152 #ifdef CONFIG_SPARSE_IRQ
153 static struct irq_cfg irq_cfgx[] = {
155 static struct irq_cfg irq_cfgx[NR_IRQS] = {
157 [0] = { .vector = IRQ0_VECTOR, },
158 [1] = { .vector = IRQ1_VECTOR, },
159 [2] = { .vector = IRQ2_VECTOR, },
160 [3] = { .vector = IRQ3_VECTOR, },
161 [4] = { .vector = IRQ4_VECTOR, },
162 [5] = { .vector = IRQ5_VECTOR, },
163 [6] = { .vector = IRQ6_VECTOR, },
164 [7] = { .vector = IRQ7_VECTOR, },
165 [8] = { .vector = IRQ8_VECTOR, },
166 [9] = { .vector = IRQ9_VECTOR, },
167 [10] = { .vector = IRQ10_VECTOR, },
168 [11] = { .vector = IRQ11_VECTOR, },
169 [12] = { .vector = IRQ12_VECTOR, },
170 [13] = { .vector = IRQ13_VECTOR, },
171 [14] = { .vector = IRQ14_VECTOR, },
172 [15] = { .vector = IRQ15_VECTOR, },
175 int __init arch_early_irq_init(void)
178 struct irq_desc *desc;
184 count = ARRAY_SIZE(irq_cfgx);
185 node= cpu_to_node(boot_cpu_id);
187 for (i = 0; i < count; i++) {
188 desc = irq_to_desc(i);
189 desc->chip_data = &cfg[i];
190 zalloc_cpumask_var_node(&cfg[i].domain, GFP_NOWAIT, node);
191 zalloc_cpumask_var_node(&cfg[i].old_domain, GFP_NOWAIT, node);
192 if (i < NR_IRQS_LEGACY)
193 cpumask_setall(cfg[i].domain);
199 #ifdef CONFIG_SPARSE_IRQ
200 static struct irq_cfg *irq_cfg(unsigned int irq)
202 struct irq_cfg *cfg = NULL;
203 struct irq_desc *desc;
205 desc = irq_to_desc(irq);
207 cfg = desc->chip_data;
212 static struct irq_cfg *get_one_free_irq_cfg(int node)
216 cfg = kzalloc_node(sizeof(*cfg), GFP_ATOMIC, node);
218 if (!alloc_cpumask_var_node(&cfg->domain, GFP_ATOMIC, node)) {
221 } else if (!alloc_cpumask_var_node(&cfg->old_domain,
223 free_cpumask_var(cfg->domain);
227 cpumask_clear(cfg->domain);
228 cpumask_clear(cfg->old_domain);
235 int arch_init_chip_data(struct irq_desc *desc, int node)
239 cfg = desc->chip_data;
241 desc->chip_data = get_one_free_irq_cfg(node);
242 if (!desc->chip_data) {
243 printk(KERN_ERR "can not alloc irq_cfg\n");
251 /* for move_irq_desc */
253 init_copy_irq_2_pin(struct irq_cfg *old_cfg, struct irq_cfg *cfg, int node)
255 struct irq_pin_list *old_entry, *head, *tail, *entry;
257 cfg->irq_2_pin = NULL;
258 old_entry = old_cfg->irq_2_pin;
262 entry = get_one_free_irq_2_pin(node);
266 entry->apic = old_entry->apic;
267 entry->pin = old_entry->pin;
270 old_entry = old_entry->next;
272 entry = get_one_free_irq_2_pin(node);
280 /* still use the old one */
283 entry->apic = old_entry->apic;
284 entry->pin = old_entry->pin;
287 old_entry = old_entry->next;
291 cfg->irq_2_pin = head;
294 static void free_irq_2_pin(struct irq_cfg *old_cfg, struct irq_cfg *cfg)
296 struct irq_pin_list *entry, *next;
298 if (old_cfg->irq_2_pin == cfg->irq_2_pin)
301 entry = old_cfg->irq_2_pin;
308 old_cfg->irq_2_pin = NULL;
311 void arch_init_copy_chip_data(struct irq_desc *old_desc,
312 struct irq_desc *desc, int node)
315 struct irq_cfg *old_cfg;
317 cfg = get_one_free_irq_cfg(node);
322 desc->chip_data = cfg;
324 old_cfg = old_desc->chip_data;
326 memcpy(cfg, old_cfg, sizeof(struct irq_cfg));
328 init_copy_irq_2_pin(old_cfg, cfg, node);
331 static void free_irq_cfg(struct irq_cfg *old_cfg)
336 void arch_free_chip_data(struct irq_desc *old_desc, struct irq_desc *desc)
338 struct irq_cfg *old_cfg, *cfg;
340 old_cfg = old_desc->chip_data;
341 cfg = desc->chip_data;
347 free_irq_2_pin(old_cfg, cfg);
348 free_irq_cfg(old_cfg);
349 old_desc->chip_data = NULL;
352 /* end for move_irq_desc */
355 static struct irq_cfg *irq_cfg(unsigned int irq)
357 return irq < nr_irqs ? irq_cfgx + irq : NULL;
364 unsigned int unused[3];
366 unsigned int unused2[11];
370 static __attribute_const__ struct io_apic __iomem *io_apic_base(int idx)
372 return (void __iomem *) __fix_to_virt(FIX_IO_APIC_BASE_0 + idx)
373 + (mp_ioapics[idx].apicaddr & ~PAGE_MASK);
376 static inline void io_apic_eoi(unsigned int apic, unsigned int vector)
378 struct io_apic __iomem *io_apic = io_apic_base(apic);
379 writel(vector, &io_apic->eoi);
382 static inline unsigned int io_apic_read(unsigned int apic, unsigned int reg)
384 struct io_apic __iomem *io_apic = io_apic_base(apic);
385 writel(reg, &io_apic->index);
386 return readl(&io_apic->data);
389 static inline void io_apic_write(unsigned int apic, unsigned int reg, unsigned int value)
391 struct io_apic __iomem *io_apic = io_apic_base(apic);
392 writel(reg, &io_apic->index);
393 writel(value, &io_apic->data);
397 * Re-write a value: to be used for read-modify-write
398 * cycles where the read already set up the index register.
400 * Older SiS APIC requires we rewrite the index register
402 static inline void io_apic_modify(unsigned int apic, unsigned int reg, unsigned int value)
404 struct io_apic __iomem *io_apic = io_apic_base(apic);
407 writel(reg, &io_apic->index);
408 writel(value, &io_apic->data);
411 static bool io_apic_level_ack_pending(struct irq_cfg *cfg)
413 struct irq_pin_list *entry;
416 spin_lock_irqsave(&ioapic_lock, flags);
417 entry = cfg->irq_2_pin;
425 reg = io_apic_read(entry->apic, 0x10 + pin*2);
426 /* Is the remote IRR bit set? */
427 if (reg & IO_APIC_REDIR_REMOTE_IRR) {
428 spin_unlock_irqrestore(&ioapic_lock, flags);
435 spin_unlock_irqrestore(&ioapic_lock, flags);
441 struct { u32 w1, w2; };
442 struct IO_APIC_route_entry entry;
445 static struct IO_APIC_route_entry ioapic_read_entry(int apic, int pin)
447 union entry_union eu;
449 spin_lock_irqsave(&ioapic_lock, flags);
450 eu.w1 = io_apic_read(apic, 0x10 + 2 * pin);
451 eu.w2 = io_apic_read(apic, 0x11 + 2 * pin);
452 spin_unlock_irqrestore(&ioapic_lock, flags);
457 * When we write a new IO APIC routing entry, we need to write the high
458 * word first! If the mask bit in the low word is clear, we will enable
459 * the interrupt, and we need to make sure the entry is fully populated
460 * before that happens.
463 __ioapic_write_entry(int apic, int pin, struct IO_APIC_route_entry e)
465 union entry_union eu = {{0, 0}};
468 io_apic_write(apic, 0x11 + 2*pin, eu.w2);
469 io_apic_write(apic, 0x10 + 2*pin, eu.w1);
472 void ioapic_write_entry(int apic, int pin, struct IO_APIC_route_entry e)
475 spin_lock_irqsave(&ioapic_lock, flags);
476 __ioapic_write_entry(apic, pin, e);
477 spin_unlock_irqrestore(&ioapic_lock, flags);
481 * When we mask an IO APIC routing entry, we need to write the low
482 * word first, in order to set the mask bit before we change the
485 static void ioapic_mask_entry(int apic, int pin)
488 union entry_union eu = { .entry.mask = 1 };
490 spin_lock_irqsave(&ioapic_lock, flags);
491 io_apic_write(apic, 0x10 + 2*pin, eu.w1);
492 io_apic_write(apic, 0x11 + 2*pin, eu.w2);
493 spin_unlock_irqrestore(&ioapic_lock, flags);
497 * The common case is 1:1 IRQ<->pin mappings. Sometimes there are
498 * shared ISA-space IRQs, so we have to support them. We are super
499 * fast in the common case, and fast for shared ISA-space IRQs.
501 static void add_pin_to_irq_node(struct irq_cfg *cfg, int node, int apic, int pin)
503 struct irq_pin_list *entry;
505 entry = cfg->irq_2_pin;
507 entry = get_one_free_irq_2_pin(node);
509 printk(KERN_ERR "can not alloc irq_2_pin to add %d - %d\n",
513 cfg->irq_2_pin = entry;
519 while (entry->next) {
520 /* not again, please */
521 if (entry->apic == apic && entry->pin == pin)
527 entry->next = get_one_free_irq_2_pin(node);
534 * Reroute an IRQ to a different pin.
536 static void __init replace_pin_at_irq_node(struct irq_cfg *cfg, int node,
537 int oldapic, int oldpin,
538 int newapic, int newpin)
540 struct irq_pin_list *entry = cfg->irq_2_pin;
544 if (entry->apic == oldapic && entry->pin == oldpin) {
545 entry->apic = newapic;
548 /* every one is different, right? */
554 /* why? call replace before add? */
556 add_pin_to_irq_node(cfg, node, newapic, newpin);
559 static inline void io_apic_modify_irq(struct irq_cfg *cfg,
560 int mask_and, int mask_or,
561 void (*final)(struct irq_pin_list *entry))
564 struct irq_pin_list *entry;
566 for (entry = cfg->irq_2_pin; entry != NULL; entry = entry->next) {
569 reg = io_apic_read(entry->apic, 0x10 + pin * 2);
572 io_apic_modify(entry->apic, 0x10 + pin * 2, reg);
578 static void __unmask_IO_APIC_irq(struct irq_cfg *cfg)
580 io_apic_modify_irq(cfg, ~IO_APIC_REDIR_MASKED, 0, NULL);
584 static void io_apic_sync(struct irq_pin_list *entry)
587 * Synchronize the IO-APIC and the CPU by doing
588 * a dummy read from the IO-APIC
590 struct io_apic __iomem *io_apic;
591 io_apic = io_apic_base(entry->apic);
592 readl(&io_apic->data);
595 static void __mask_IO_APIC_irq(struct irq_cfg *cfg)
597 io_apic_modify_irq(cfg, ~0, IO_APIC_REDIR_MASKED, &io_apic_sync);
599 #else /* CONFIG_X86_32 */
600 static void __mask_IO_APIC_irq(struct irq_cfg *cfg)
602 io_apic_modify_irq(cfg, ~0, IO_APIC_REDIR_MASKED, NULL);
605 static void __mask_and_edge_IO_APIC_irq(struct irq_cfg *cfg)
607 io_apic_modify_irq(cfg, ~IO_APIC_REDIR_LEVEL_TRIGGER,
608 IO_APIC_REDIR_MASKED, NULL);
611 static void __unmask_and_level_IO_APIC_irq(struct irq_cfg *cfg)
613 io_apic_modify_irq(cfg, ~IO_APIC_REDIR_MASKED,
614 IO_APIC_REDIR_LEVEL_TRIGGER, NULL);
616 #endif /* CONFIG_X86_32 */
618 static void mask_IO_APIC_irq_desc(struct irq_desc *desc)
620 struct irq_cfg *cfg = desc->chip_data;
625 spin_lock_irqsave(&ioapic_lock, flags);
626 __mask_IO_APIC_irq(cfg);
627 spin_unlock_irqrestore(&ioapic_lock, flags);
630 static void unmask_IO_APIC_irq_desc(struct irq_desc *desc)
632 struct irq_cfg *cfg = desc->chip_data;
635 spin_lock_irqsave(&ioapic_lock, flags);
636 __unmask_IO_APIC_irq(cfg);
637 spin_unlock_irqrestore(&ioapic_lock, flags);
640 static void mask_IO_APIC_irq(unsigned int irq)
642 struct irq_desc *desc = irq_to_desc(irq);
644 mask_IO_APIC_irq_desc(desc);
646 static void unmask_IO_APIC_irq(unsigned int irq)
648 struct irq_desc *desc = irq_to_desc(irq);
650 unmask_IO_APIC_irq_desc(desc);
653 static void clear_IO_APIC_pin(unsigned int apic, unsigned int pin)
655 struct IO_APIC_route_entry entry;
657 /* Check delivery_mode to be sure we're not clearing an SMI pin */
658 entry = ioapic_read_entry(apic, pin);
659 if (entry.delivery_mode == dest_SMI)
662 * Disable it in the IO-APIC irq-routing table:
664 ioapic_mask_entry(apic, pin);
667 static void clear_IO_APIC (void)
671 for (apic = 0; apic < nr_ioapics; apic++)
672 for (pin = 0; pin < nr_ioapic_registers[apic]; pin++)
673 clear_IO_APIC_pin(apic, pin);
678 * support for broken MP BIOSs, enables hand-redirection of PIRQ0-7 to
679 * specific CPU-side IRQs.
683 static int pirq_entries[MAX_PIRQS] = {
684 [0 ... MAX_PIRQS - 1] = -1
687 static int __init ioapic_pirq_setup(char *str)
690 int ints[MAX_PIRQS+1];
692 get_options(str, ARRAY_SIZE(ints), ints);
694 apic_printk(APIC_VERBOSE, KERN_INFO
695 "PIRQ redirection, working around broken MP-BIOS.\n");
697 if (ints[0] < MAX_PIRQS)
700 for (i = 0; i < max; i++) {
701 apic_printk(APIC_VERBOSE, KERN_DEBUG
702 "... PIRQ%d -> IRQ %d\n", i, ints[i+1]);
704 * PIRQs are mapped upside down, usually.
706 pirq_entries[MAX_PIRQS-i-1] = ints[i+1];
711 __setup("pirq=", ioapic_pirq_setup);
712 #endif /* CONFIG_X86_32 */
714 struct IO_APIC_route_entry **alloc_ioapic_entries(void)
717 struct IO_APIC_route_entry **ioapic_entries;
719 ioapic_entries = kzalloc(sizeof(*ioapic_entries) * nr_ioapics,
724 for (apic = 0; apic < nr_ioapics; apic++) {
725 ioapic_entries[apic] =
726 kzalloc(sizeof(struct IO_APIC_route_entry) *
727 nr_ioapic_registers[apic], GFP_ATOMIC);
728 if (!ioapic_entries[apic])
732 return ioapic_entries;
736 kfree(ioapic_entries[apic]);
737 kfree(ioapic_entries);
743 * Saves all the IO-APIC RTE's
745 int save_IO_APIC_setup(struct IO_APIC_route_entry **ioapic_entries)
752 for (apic = 0; apic < nr_ioapics; apic++) {
753 if (!ioapic_entries[apic])
756 for (pin = 0; pin < nr_ioapic_registers[apic]; pin++)
757 ioapic_entries[apic][pin] =
758 ioapic_read_entry(apic, pin);
765 * Mask all IO APIC entries.
767 void mask_IO_APIC_setup(struct IO_APIC_route_entry **ioapic_entries)
774 for (apic = 0; apic < nr_ioapics; apic++) {
775 if (!ioapic_entries[apic])
778 for (pin = 0; pin < nr_ioapic_registers[apic]; pin++) {
779 struct IO_APIC_route_entry entry;
781 entry = ioapic_entries[apic][pin];
784 ioapic_write_entry(apic, pin, entry);
791 * Restore IO APIC entries which was saved in ioapic_entries.
793 int restore_IO_APIC_setup(struct IO_APIC_route_entry **ioapic_entries)
800 for (apic = 0; apic < nr_ioapics; apic++) {
801 if (!ioapic_entries[apic])
804 for (pin = 0; pin < nr_ioapic_registers[apic]; pin++)
805 ioapic_write_entry(apic, pin,
806 ioapic_entries[apic][pin]);
811 void free_ioapic_entries(struct IO_APIC_route_entry **ioapic_entries)
815 for (apic = 0; apic < nr_ioapics; apic++)
816 kfree(ioapic_entries[apic]);
818 kfree(ioapic_entries);
822 * Find the IRQ entry number of a certain pin.
824 static int find_irq_entry(int apic, int pin, int type)
828 for (i = 0; i < mp_irq_entries; i++)
829 if (mp_irqs[i].irqtype == type &&
830 (mp_irqs[i].dstapic == mp_ioapics[apic].apicid ||
831 mp_irqs[i].dstapic == MP_APIC_ALL) &&
832 mp_irqs[i].dstirq == pin)
839 * Find the pin to which IRQ[irq] (ISA) is connected
841 static int __init find_isa_irq_pin(int irq, int type)
845 for (i = 0; i < mp_irq_entries; i++) {
846 int lbus = mp_irqs[i].srcbus;
848 if (test_bit(lbus, mp_bus_not_pci) &&
849 (mp_irqs[i].irqtype == type) &&
850 (mp_irqs[i].srcbusirq == irq))
852 return mp_irqs[i].dstirq;
857 static int __init find_isa_irq_apic(int irq, int type)
861 for (i = 0; i < mp_irq_entries; i++) {
862 int lbus = mp_irqs[i].srcbus;
864 if (test_bit(lbus, mp_bus_not_pci) &&
865 (mp_irqs[i].irqtype == type) &&
866 (mp_irqs[i].srcbusirq == irq))
869 if (i < mp_irq_entries) {
871 for(apic = 0; apic < nr_ioapics; apic++) {
872 if (mp_ioapics[apic].apicid == mp_irqs[i].dstapic)
880 #if defined(CONFIG_EISA) || defined(CONFIG_MCA)
882 * EISA Edge/Level control register, ELCR
884 static int EISA_ELCR(unsigned int irq)
886 if (irq < NR_IRQS_LEGACY) {
887 unsigned int port = 0x4d0 + (irq >> 3);
888 return (inb(port) >> (irq & 7)) & 1;
890 apic_printk(APIC_VERBOSE, KERN_INFO
891 "Broken MPtable reports ISA irq %d\n", irq);
897 /* ISA interrupts are always polarity zero edge triggered,
898 * when listed as conforming in the MP table. */
900 #define default_ISA_trigger(idx) (0)
901 #define default_ISA_polarity(idx) (0)
903 /* EISA interrupts are always polarity zero and can be edge or level
904 * trigger depending on the ELCR value. If an interrupt is listed as
905 * EISA conforming in the MP table, that means its trigger type must
906 * be read in from the ELCR */
908 #define default_EISA_trigger(idx) (EISA_ELCR(mp_irqs[idx].srcbusirq))
909 #define default_EISA_polarity(idx) default_ISA_polarity(idx)
911 /* PCI interrupts are always polarity one level triggered,
912 * when listed as conforming in the MP table. */
914 #define default_PCI_trigger(idx) (1)
915 #define default_PCI_polarity(idx) (1)
917 /* MCA interrupts are always polarity zero level triggered,
918 * when listed as conforming in the MP table. */
920 #define default_MCA_trigger(idx) (1)
921 #define default_MCA_polarity(idx) default_ISA_polarity(idx)
923 static int MPBIOS_polarity(int idx)
925 int bus = mp_irqs[idx].srcbus;
929 * Determine IRQ line polarity (high active or low active):
931 switch (mp_irqs[idx].irqflag & 3)
933 case 0: /* conforms, ie. bus-type dependent polarity */
934 if (test_bit(bus, mp_bus_not_pci))
935 polarity = default_ISA_polarity(idx);
937 polarity = default_PCI_polarity(idx);
939 case 1: /* high active */
944 case 2: /* reserved */
946 printk(KERN_WARNING "broken BIOS!!\n");
950 case 3: /* low active */
955 default: /* invalid */
957 printk(KERN_WARNING "broken BIOS!!\n");
965 static int MPBIOS_trigger(int idx)
967 int bus = mp_irqs[idx].srcbus;
971 * Determine IRQ trigger mode (edge or level sensitive):
973 switch ((mp_irqs[idx].irqflag>>2) & 3)
975 case 0: /* conforms, ie. bus-type dependent */
976 if (test_bit(bus, mp_bus_not_pci))
977 trigger = default_ISA_trigger(idx);
979 trigger = default_PCI_trigger(idx);
980 #if defined(CONFIG_EISA) || defined(CONFIG_MCA)
981 switch (mp_bus_id_to_type[bus]) {
982 case MP_BUS_ISA: /* ISA pin */
984 /* set before the switch */
987 case MP_BUS_EISA: /* EISA pin */
989 trigger = default_EISA_trigger(idx);
992 case MP_BUS_PCI: /* PCI pin */
994 /* set before the switch */
997 case MP_BUS_MCA: /* MCA pin */
999 trigger = default_MCA_trigger(idx);
1004 printk(KERN_WARNING "broken BIOS!!\n");
1016 case 2: /* reserved */
1018 printk(KERN_WARNING "broken BIOS!!\n");
1027 default: /* invalid */
1029 printk(KERN_WARNING "broken BIOS!!\n");
1037 static inline int irq_polarity(int idx)
1039 return MPBIOS_polarity(idx);
1042 static inline int irq_trigger(int idx)
1044 return MPBIOS_trigger(idx);
1047 int (*ioapic_renumber_irq)(int ioapic, int irq);
1048 static int pin_2_irq(int idx, int apic, int pin)
1051 int bus = mp_irqs[idx].srcbus;
1054 * Debugging check, we are in big trouble if this message pops up!
1056 if (mp_irqs[idx].dstirq != pin)
1057 printk(KERN_ERR "broken BIOS or MPTABLE parser, ayiee!!\n");
1059 if (test_bit(bus, mp_bus_not_pci)) {
1060 irq = mp_irqs[idx].srcbusirq;
1063 * PCI IRQs are mapped in order
1067 irq += nr_ioapic_registers[i++];
1070 * For MPS mode, so far only needed by ES7000 platform
1072 if (ioapic_renumber_irq)
1073 irq = ioapic_renumber_irq(apic, irq);
1076 #ifdef CONFIG_X86_32
1078 * PCI IRQ command line redirection. Yes, limits are hardcoded.
1080 if ((pin >= 16) && (pin <= 23)) {
1081 if (pirq_entries[pin-16] != -1) {
1082 if (!pirq_entries[pin-16]) {
1083 apic_printk(APIC_VERBOSE, KERN_DEBUG
1084 "disabling PIRQ%d\n", pin-16);
1086 irq = pirq_entries[pin-16];
1087 apic_printk(APIC_VERBOSE, KERN_DEBUG
1088 "using PIRQ%d -> IRQ %d\n",
1099 * Find a specific PCI IRQ entry.
1100 * Not an __init, possibly needed by modules
1102 int IO_APIC_get_PCI_irq_vector(int bus, int slot, int pin,
1103 struct io_apic_irq_attr *irq_attr)
1105 int apic, i, best_guess = -1;
1107 apic_printk(APIC_DEBUG,
1108 "querying PCI -> IRQ mapping bus:%d, slot:%d, pin:%d.\n",
1110 if (test_bit(bus, mp_bus_not_pci)) {
1111 apic_printk(APIC_VERBOSE,
1112 "PCI BIOS passed nonexistent PCI bus %d!\n", bus);
1115 for (i = 0; i < mp_irq_entries; i++) {
1116 int lbus = mp_irqs[i].srcbus;
1118 for (apic = 0; apic < nr_ioapics; apic++)
1119 if (mp_ioapics[apic].apicid == mp_irqs[i].dstapic ||
1120 mp_irqs[i].dstapic == MP_APIC_ALL)
1123 if (!test_bit(lbus, mp_bus_not_pci) &&
1124 !mp_irqs[i].irqtype &&
1126 (slot == ((mp_irqs[i].srcbusirq >> 2) & 0x1f))) {
1127 int irq = pin_2_irq(i, apic, mp_irqs[i].dstirq);
1129 if (!(apic || IO_APIC_IRQ(irq)))
1132 if (pin == (mp_irqs[i].srcbusirq & 3)) {
1133 set_io_apic_irq_attr(irq_attr, apic,
1140 * Use the first all-but-pin matching entry as a
1141 * best-guess fuzzy result for broken mptables.
1143 if (best_guess < 0) {
1144 set_io_apic_irq_attr(irq_attr, apic,
1154 EXPORT_SYMBOL(IO_APIC_get_PCI_irq_vector);
1156 void lock_vector_lock(void)
1158 /* Used to the online set of cpus does not change
1159 * during assign_irq_vector.
1161 spin_lock(&vector_lock);
1164 void unlock_vector_lock(void)
1166 spin_unlock(&vector_lock);
1170 __assign_irq_vector(int irq, struct irq_cfg *cfg, const struct cpumask *mask)
1173 * NOTE! The local APIC isn't very good at handling
1174 * multiple interrupts at the same interrupt level.
1175 * As the interrupt level is determined by taking the
1176 * vector number and shifting that right by 4, we
1177 * want to spread these out a bit so that they don't
1178 * all fall in the same interrupt level.
1180 * Also, we've got to be careful not to trash gate
1181 * 0x80, because int 0x80 is hm, kind of importantish. ;)
1183 static int current_vector = FIRST_DEVICE_VECTOR, current_offset = 0;
1184 unsigned int old_vector;
1186 cpumask_var_t tmp_mask;
1188 if ((cfg->move_in_progress) || cfg->move_cleanup_count)
1191 if (!alloc_cpumask_var(&tmp_mask, GFP_ATOMIC))
1194 old_vector = cfg->vector;
1196 cpumask_and(tmp_mask, mask, cpu_online_mask);
1197 cpumask_and(tmp_mask, cfg->domain, tmp_mask);
1198 if (!cpumask_empty(tmp_mask)) {
1199 free_cpumask_var(tmp_mask);
1204 /* Only try and allocate irqs on cpus that are present */
1206 for_each_cpu_and(cpu, mask, cpu_online_mask) {
1210 apic->vector_allocation_domain(cpu, tmp_mask);
1212 vector = current_vector;
1213 offset = current_offset;
1216 if (vector >= first_system_vector) {
1217 /* If out of vectors on large boxen, must share them. */
1218 offset = (offset + 1) % 8;
1219 vector = FIRST_DEVICE_VECTOR + offset;
1221 if (unlikely(current_vector == vector))
1224 if (test_bit(vector, used_vectors))
1227 for_each_cpu_and(new_cpu, tmp_mask, cpu_online_mask)
1228 if (per_cpu(vector_irq, new_cpu)[vector] != -1)
1231 current_vector = vector;
1232 current_offset = offset;
1234 cfg->move_in_progress = 1;
1235 cpumask_copy(cfg->old_domain, cfg->domain);
1237 for_each_cpu_and(new_cpu, tmp_mask, cpu_online_mask)
1238 per_cpu(vector_irq, new_cpu)[vector] = irq;
1239 cfg->vector = vector;
1240 cpumask_copy(cfg->domain, tmp_mask);
1244 free_cpumask_var(tmp_mask);
1249 assign_irq_vector(int irq, struct irq_cfg *cfg, const struct cpumask *mask)
1252 unsigned long flags;
1254 spin_lock_irqsave(&vector_lock, flags);
1255 err = __assign_irq_vector(irq, cfg, mask);
1256 spin_unlock_irqrestore(&vector_lock, flags);
1260 static void __clear_irq_vector(int irq, struct irq_cfg *cfg)
1264 BUG_ON(!cfg->vector);
1266 vector = cfg->vector;
1267 for_each_cpu_and(cpu, cfg->domain, cpu_online_mask)
1268 per_cpu(vector_irq, cpu)[vector] = -1;
1271 cpumask_clear(cfg->domain);
1273 if (likely(!cfg->move_in_progress))
1275 for_each_cpu_and(cpu, cfg->old_domain, cpu_online_mask) {
1276 for (vector = FIRST_EXTERNAL_VECTOR; vector < NR_VECTORS;
1278 if (per_cpu(vector_irq, cpu)[vector] != irq)
1280 per_cpu(vector_irq, cpu)[vector] = -1;
1284 cfg->move_in_progress = 0;
1287 void __setup_vector_irq(int cpu)
1289 /* Initialize vector_irq on a new cpu */
1290 /* This function must be called with vector_lock held */
1292 struct irq_cfg *cfg;
1293 struct irq_desc *desc;
1295 /* Mark the inuse vectors */
1296 for_each_irq_desc(irq, desc) {
1297 cfg = desc->chip_data;
1298 if (!cpumask_test_cpu(cpu, cfg->domain))
1300 vector = cfg->vector;
1301 per_cpu(vector_irq, cpu)[vector] = irq;
1303 /* Mark the free vectors */
1304 for (vector = 0; vector < NR_VECTORS; ++vector) {
1305 irq = per_cpu(vector_irq, cpu)[vector];
1310 if (!cpumask_test_cpu(cpu, cfg->domain))
1311 per_cpu(vector_irq, cpu)[vector] = -1;
1315 static struct irq_chip ioapic_chip;
1316 static struct irq_chip ir_ioapic_chip;
1318 #define IOAPIC_AUTO -1
1319 #define IOAPIC_EDGE 0
1320 #define IOAPIC_LEVEL 1
1322 #ifdef CONFIG_X86_32
1323 static inline int IO_APIC_irq_trigger(int irq)
1327 for (apic = 0; apic < nr_ioapics; apic++) {
1328 for (pin = 0; pin < nr_ioapic_registers[apic]; pin++) {
1329 idx = find_irq_entry(apic, pin, mp_INT);
1330 if ((idx != -1) && (irq == pin_2_irq(idx, apic, pin)))
1331 return irq_trigger(idx);
1335 * nonexistent IRQs are edge default
1340 static inline int IO_APIC_irq_trigger(int irq)
1346 static void ioapic_register_intr(int irq, struct irq_desc *desc, unsigned long trigger)
1349 if ((trigger == IOAPIC_AUTO && IO_APIC_irq_trigger(irq)) ||
1350 trigger == IOAPIC_LEVEL)
1351 desc->status |= IRQ_LEVEL;
1353 desc->status &= ~IRQ_LEVEL;
1355 if (irq_remapped(irq)) {
1356 desc->status |= IRQ_MOVE_PCNTXT;
1358 set_irq_chip_and_handler_name(irq, &ir_ioapic_chip,
1362 set_irq_chip_and_handler_name(irq, &ir_ioapic_chip,
1363 handle_edge_irq, "edge");
1367 if ((trigger == IOAPIC_AUTO && IO_APIC_irq_trigger(irq)) ||
1368 trigger == IOAPIC_LEVEL)
1369 set_irq_chip_and_handler_name(irq, &ioapic_chip,
1373 set_irq_chip_and_handler_name(irq, &ioapic_chip,
1374 handle_edge_irq, "edge");
1377 int setup_ioapic_entry(int apic_id, int irq,
1378 struct IO_APIC_route_entry *entry,
1379 unsigned int destination, int trigger,
1380 int polarity, int vector, int pin)
1383 * add it to the IO-APIC irq-routing table:
1385 memset(entry,0,sizeof(*entry));
1387 if (intr_remapping_enabled) {
1388 struct intel_iommu *iommu = map_ioapic_to_ir(apic_id);
1390 struct IR_IO_APIC_route_entry *ir_entry =
1391 (struct IR_IO_APIC_route_entry *) entry;
1395 panic("No mapping iommu for ioapic %d\n", apic_id);
1397 index = alloc_irte(iommu, irq, 1);
1399 panic("Failed to allocate IRTE for ioapic %d\n", apic_id);
1401 memset(&irte, 0, sizeof(irte));
1404 irte.dst_mode = apic->irq_dest_mode;
1406 * Trigger mode in the IRTE will always be edge, and the
1407 * actual level or edge trigger will be setup in the IO-APIC
1408 * RTE. This will help simplify level triggered irq migration.
1409 * For more details, see the comments above explainig IO-APIC
1410 * irq migration in the presence of interrupt-remapping.
1412 irte.trigger_mode = 0;
1413 irte.dlvry_mode = apic->irq_delivery_mode;
1414 irte.vector = vector;
1415 irte.dest_id = IRTE_DEST(destination);
1417 /* Set source-id of interrupt request */
1418 set_ioapic_sid(&irte, apic_id);
1420 modify_irte(irq, &irte);
1422 ir_entry->index2 = (index >> 15) & 0x1;
1424 ir_entry->format = 1;
1425 ir_entry->index = (index & 0x7fff);
1427 * IO-APIC RTE will be configured with virtual vector.
1428 * irq handler will do the explicit EOI to the io-apic.
1430 ir_entry->vector = pin;
1432 entry->delivery_mode = apic->irq_delivery_mode;
1433 entry->dest_mode = apic->irq_dest_mode;
1434 entry->dest = destination;
1435 entry->vector = vector;
1438 entry->mask = 0; /* enable IRQ */
1439 entry->trigger = trigger;
1440 entry->polarity = polarity;
1442 /* Mask level triggered irqs.
1443 * Use IRQ_DELAYED_DISABLE for edge triggered irqs.
1450 static void setup_IO_APIC_irq(int apic_id, int pin, unsigned int irq, struct irq_desc *desc,
1451 int trigger, int polarity)
1453 struct irq_cfg *cfg;
1454 struct IO_APIC_route_entry entry;
1457 if (!IO_APIC_IRQ(irq))
1460 cfg = desc->chip_data;
1462 if (assign_irq_vector(irq, cfg, apic->target_cpus()))
1465 dest = apic->cpu_mask_to_apicid_and(cfg->domain, apic->target_cpus());
1467 apic_printk(APIC_VERBOSE,KERN_DEBUG
1468 "IOAPIC[%d]: Set routing entry (%d-%d -> 0x%x -> "
1469 "IRQ %d Mode:%i Active:%i)\n",
1470 apic_id, mp_ioapics[apic_id].apicid, pin, cfg->vector,
1471 irq, trigger, polarity);
1474 if (setup_ioapic_entry(mp_ioapics[apic_id].apicid, irq, &entry,
1475 dest, trigger, polarity, cfg->vector, pin)) {
1476 printk("Failed to setup ioapic entry for ioapic %d, pin %d\n",
1477 mp_ioapics[apic_id].apicid, pin);
1478 __clear_irq_vector(irq, cfg);
1482 ioapic_register_intr(irq, desc, trigger);
1483 if (irq < NR_IRQS_LEGACY)
1484 disable_8259A_irq(irq);
1486 ioapic_write_entry(apic_id, pin, entry);
1490 DECLARE_BITMAP(pin_programmed, MP_MAX_IOAPIC_PIN + 1);
1491 } mp_ioapic_routing[MAX_IO_APICS];
1493 static void __init setup_IO_APIC_irqs(void)
1495 int apic_id = 0, pin, idx, irq;
1497 struct irq_desc *desc;
1498 struct irq_cfg *cfg;
1499 int node = cpu_to_node(boot_cpu_id);
1501 apic_printk(APIC_VERBOSE, KERN_DEBUG "init IO_APIC IRQs\n");
1504 if (!acpi_disabled && acpi_ioapic) {
1505 apic_id = mp_find_ioapic(0);
1511 for (pin = 0; pin < nr_ioapic_registers[apic_id]; pin++) {
1512 idx = find_irq_entry(apic_id, pin, mp_INT);
1516 apic_printk(APIC_VERBOSE,
1517 KERN_DEBUG " %d-%d",
1518 mp_ioapics[apic_id].apicid, pin);
1520 apic_printk(APIC_VERBOSE, " %d-%d",
1521 mp_ioapics[apic_id].apicid, pin);
1525 apic_printk(APIC_VERBOSE,
1526 " (apicid-pin) not connected\n");
1530 irq = pin_2_irq(idx, apic_id, pin);
1533 * Skip the timer IRQ if there's a quirk handler
1534 * installed and if it returns 1:
1536 if (apic->multi_timer_check &&
1537 apic->multi_timer_check(apic_id, irq))
1540 desc = irq_to_desc_alloc_node(irq, node);
1542 printk(KERN_INFO "can not get irq_desc for %d\n", irq);
1545 cfg = desc->chip_data;
1546 add_pin_to_irq_node(cfg, node, apic_id, pin);
1548 * don't mark it in pin_programmed, so later acpi could
1549 * set it correctly when irq < 16
1551 setup_IO_APIC_irq(apic_id, pin, irq, desc,
1552 irq_trigger(idx), irq_polarity(idx));
1556 apic_printk(APIC_VERBOSE,
1557 " (apicid-pin) not connected\n");
1561 * Set up the timer pin, possibly with the 8259A-master behind.
1563 static void __init setup_timer_IRQ0_pin(unsigned int apic_id, unsigned int pin,
1566 struct IO_APIC_route_entry entry;
1568 if (intr_remapping_enabled)
1571 memset(&entry, 0, sizeof(entry));
1574 * We use logical delivery to get the timer IRQ
1577 entry.dest_mode = apic->irq_dest_mode;
1578 entry.mask = 0; /* don't mask IRQ for edge */
1579 entry.dest = apic->cpu_mask_to_apicid(apic->target_cpus());
1580 entry.delivery_mode = apic->irq_delivery_mode;
1583 entry.vector = vector;
1586 * The timer IRQ doesn't have to know that behind the
1587 * scene we may have a 8259A-master in AEOI mode ...
1589 set_irq_chip_and_handler_name(0, &ioapic_chip, handle_edge_irq, "edge");
1592 * Add it to the IO-APIC irq-routing table:
1594 ioapic_write_entry(apic_id, pin, entry);
1598 __apicdebuginit(void) print_IO_APIC(void)
1601 union IO_APIC_reg_00 reg_00;
1602 union IO_APIC_reg_01 reg_01;
1603 union IO_APIC_reg_02 reg_02;
1604 union IO_APIC_reg_03 reg_03;
1605 unsigned long flags;
1606 struct irq_cfg *cfg;
1607 struct irq_desc *desc;
1610 if (apic_verbosity == APIC_QUIET)
1613 printk(KERN_DEBUG "number of MP IRQ sources: %d.\n", mp_irq_entries);
1614 for (i = 0; i < nr_ioapics; i++)
1615 printk(KERN_DEBUG "number of IO-APIC #%d registers: %d.\n",
1616 mp_ioapics[i].apicid, nr_ioapic_registers[i]);
1619 * We are a bit conservative about what we expect. We have to
1620 * know about every hardware change ASAP.
1622 printk(KERN_INFO "testing the IO APIC.......................\n");
1624 for (apic = 0; apic < nr_ioapics; apic++) {
1626 spin_lock_irqsave(&ioapic_lock, flags);
1627 reg_00.raw = io_apic_read(apic, 0);
1628 reg_01.raw = io_apic_read(apic, 1);
1629 if (reg_01.bits.version >= 0x10)
1630 reg_02.raw = io_apic_read(apic, 2);
1631 if (reg_01.bits.version >= 0x20)
1632 reg_03.raw = io_apic_read(apic, 3);
1633 spin_unlock_irqrestore(&ioapic_lock, flags);
1636 printk(KERN_DEBUG "IO APIC #%d......\n", mp_ioapics[apic].apicid);
1637 printk(KERN_DEBUG ".... register #00: %08X\n", reg_00.raw);
1638 printk(KERN_DEBUG "....... : physical APIC id: %02X\n", reg_00.bits.ID);
1639 printk(KERN_DEBUG "....... : Delivery Type: %X\n", reg_00.bits.delivery_type);
1640 printk(KERN_DEBUG "....... : LTS : %X\n", reg_00.bits.LTS);
1642 printk(KERN_DEBUG ".... register #01: %08X\n", *(int *)®_01);
1643 printk(KERN_DEBUG "....... : max redirection entries: %04X\n", reg_01.bits.entries);
1645 printk(KERN_DEBUG "....... : PRQ implemented: %X\n", reg_01.bits.PRQ);
1646 printk(KERN_DEBUG "....... : IO APIC version: %04X\n", reg_01.bits.version);
1649 * Some Intel chipsets with IO APIC VERSION of 0x1? don't have reg_02,
1650 * but the value of reg_02 is read as the previous read register
1651 * value, so ignore it if reg_02 == reg_01.
1653 if (reg_01.bits.version >= 0x10 && reg_02.raw != reg_01.raw) {
1654 printk(KERN_DEBUG ".... register #02: %08X\n", reg_02.raw);
1655 printk(KERN_DEBUG "....... : arbitration: %02X\n", reg_02.bits.arbitration);
1659 * Some Intel chipsets with IO APIC VERSION of 0x2? don't have reg_02
1660 * or reg_03, but the value of reg_0[23] is read as the previous read
1661 * register value, so ignore it if reg_03 == reg_0[12].
1663 if (reg_01.bits.version >= 0x20 && reg_03.raw != reg_02.raw &&
1664 reg_03.raw != reg_01.raw) {
1665 printk(KERN_DEBUG ".... register #03: %08X\n", reg_03.raw);
1666 printk(KERN_DEBUG "....... : Boot DT : %X\n", reg_03.bits.boot_DT);
1669 printk(KERN_DEBUG ".... IRQ redirection table:\n");
1671 printk(KERN_DEBUG " NR Dst Mask Trig IRR Pol"
1672 " Stat Dmod Deli Vect: \n");
1674 for (i = 0; i <= reg_01.bits.entries; i++) {
1675 struct IO_APIC_route_entry entry;
1677 entry = ioapic_read_entry(apic, i);
1679 printk(KERN_DEBUG " %02x %03X ",
1684 printk("%1d %1d %1d %1d %1d %1d %1d %02X\n",
1689 entry.delivery_status,
1691 entry.delivery_mode,
1696 printk(KERN_DEBUG "IRQ to pin mappings:\n");
1697 for_each_irq_desc(irq, desc) {
1698 struct irq_pin_list *entry;
1700 cfg = desc->chip_data;
1701 entry = cfg->irq_2_pin;
1704 printk(KERN_DEBUG "IRQ%d ", irq);
1706 printk("-> %d:%d", entry->apic, entry->pin);
1709 entry = entry->next;
1714 printk(KERN_INFO ".................................... done.\n");
1719 __apicdebuginit(void) print_APIC_field(int base)
1723 if (apic_verbosity == APIC_QUIET)
1728 for (i = 0; i < 8; i++)
1729 printk(KERN_CONT "%08x", apic_read(base + i*0x10));
1731 printk(KERN_CONT "\n");
1734 __apicdebuginit(void) print_local_APIC(void *dummy)
1736 unsigned int i, v, ver, maxlvt;
1739 if (apic_verbosity == APIC_QUIET)
1742 printk(KERN_DEBUG "\n");
1743 printk(KERN_DEBUG "printing local APIC contents on CPU#%d/%d:\n",
1744 smp_processor_id(), hard_smp_processor_id());
1745 v = apic_read(APIC_ID);
1746 printk(KERN_INFO "... APIC ID: %08x (%01x)\n", v, read_apic_id());
1747 v = apic_read(APIC_LVR);
1748 printk(KERN_INFO "... APIC VERSION: %08x\n", v);
1749 ver = GET_APIC_VERSION(v);
1750 maxlvt = lapic_get_maxlvt();
1752 v = apic_read(APIC_TASKPRI);
1753 printk(KERN_DEBUG "... APIC TASKPRI: %08x (%02x)\n", v, v & APIC_TPRI_MASK);
1755 if (APIC_INTEGRATED(ver)) { /* !82489DX */
1756 if (!APIC_XAPIC(ver)) {
1757 v = apic_read(APIC_ARBPRI);
1758 printk(KERN_DEBUG "... APIC ARBPRI: %08x (%02x)\n", v,
1759 v & APIC_ARBPRI_MASK);
1761 v = apic_read(APIC_PROCPRI);
1762 printk(KERN_DEBUG "... APIC PROCPRI: %08x\n", v);
1766 * Remote read supported only in the 82489DX and local APIC for
1767 * Pentium processors.
1769 if (!APIC_INTEGRATED(ver) || maxlvt == 3) {
1770 v = apic_read(APIC_RRR);
1771 printk(KERN_DEBUG "... APIC RRR: %08x\n", v);
1774 v = apic_read(APIC_LDR);
1775 printk(KERN_DEBUG "... APIC LDR: %08x\n", v);
1776 if (!x2apic_enabled()) {
1777 v = apic_read(APIC_DFR);
1778 printk(KERN_DEBUG "... APIC DFR: %08x\n", v);
1780 v = apic_read(APIC_SPIV);
1781 printk(KERN_DEBUG "... APIC SPIV: %08x\n", v);
1783 printk(KERN_DEBUG "... APIC ISR field:\n");
1784 print_APIC_field(APIC_ISR);
1785 printk(KERN_DEBUG "... APIC TMR field:\n");
1786 print_APIC_field(APIC_TMR);
1787 printk(KERN_DEBUG "... APIC IRR field:\n");
1788 print_APIC_field(APIC_IRR);
1790 if (APIC_INTEGRATED(ver)) { /* !82489DX */
1791 if (maxlvt > 3) /* Due to the Pentium erratum 3AP. */
1792 apic_write(APIC_ESR, 0);
1794 v = apic_read(APIC_ESR);
1795 printk(KERN_DEBUG "... APIC ESR: %08x\n", v);
1798 icr = apic_icr_read();
1799 printk(KERN_DEBUG "... APIC ICR: %08x\n", (u32)icr);
1800 printk(KERN_DEBUG "... APIC ICR2: %08x\n", (u32)(icr >> 32));
1802 v = apic_read(APIC_LVTT);
1803 printk(KERN_DEBUG "... APIC LVTT: %08x\n", v);
1805 if (maxlvt > 3) { /* PC is LVT#4. */
1806 v = apic_read(APIC_LVTPC);
1807 printk(KERN_DEBUG "... APIC LVTPC: %08x\n", v);
1809 v = apic_read(APIC_LVT0);
1810 printk(KERN_DEBUG "... APIC LVT0: %08x\n", v);
1811 v = apic_read(APIC_LVT1);
1812 printk(KERN_DEBUG "... APIC LVT1: %08x\n", v);
1814 if (maxlvt > 2) { /* ERR is LVT#3. */
1815 v = apic_read(APIC_LVTERR);
1816 printk(KERN_DEBUG "... APIC LVTERR: %08x\n", v);
1819 v = apic_read(APIC_TMICT);
1820 printk(KERN_DEBUG "... APIC TMICT: %08x\n", v);
1821 v = apic_read(APIC_TMCCT);
1822 printk(KERN_DEBUG "... APIC TMCCT: %08x\n", v);
1823 v = apic_read(APIC_TDCR);
1824 printk(KERN_DEBUG "... APIC TDCR: %08x\n", v);
1826 if (boot_cpu_has(X86_FEATURE_EXTAPIC)) {
1827 v = apic_read(APIC_EFEAT);
1828 maxlvt = (v >> 16) & 0xff;
1829 printk(KERN_DEBUG "... APIC EFEAT: %08x\n", v);
1830 v = apic_read(APIC_ECTRL);
1831 printk(KERN_DEBUG "... APIC ECTRL: %08x\n", v);
1832 for (i = 0; i < maxlvt; i++) {
1833 v = apic_read(APIC_EILVTn(i));
1834 printk(KERN_DEBUG "... APIC EILVT%d: %08x\n", i, v);
1840 __apicdebuginit(void) print_all_local_APICs(void)
1845 for_each_online_cpu(cpu)
1846 smp_call_function_single(cpu, print_local_APIC, NULL, 1);
1850 __apicdebuginit(void) print_PIC(void)
1853 unsigned long flags;
1855 if (apic_verbosity == APIC_QUIET)
1858 printk(KERN_DEBUG "\nprinting PIC contents\n");
1860 spin_lock_irqsave(&i8259A_lock, flags);
1862 v = inb(0xa1) << 8 | inb(0x21);
1863 printk(KERN_DEBUG "... PIC IMR: %04x\n", v);
1865 v = inb(0xa0) << 8 | inb(0x20);
1866 printk(KERN_DEBUG "... PIC IRR: %04x\n", v);
1870 v = inb(0xa0) << 8 | inb(0x20);
1874 spin_unlock_irqrestore(&i8259A_lock, flags);
1876 printk(KERN_DEBUG "... PIC ISR: %04x\n", v);
1878 v = inb(0x4d1) << 8 | inb(0x4d0);
1879 printk(KERN_DEBUG "... PIC ELCR: %04x\n", v);
1882 __apicdebuginit(int) print_all_ICs(void)
1886 /* don't print out if apic is not there */
1887 if (!cpu_has_apic || disable_apic)
1890 print_all_local_APICs();
1896 fs_initcall(print_all_ICs);
1899 /* Where if anywhere is the i8259 connect in external int mode */
1900 static struct { int pin, apic; } ioapic_i8259 = { -1, -1 };
1902 void __init enable_IO_APIC(void)
1904 union IO_APIC_reg_01 reg_01;
1905 int i8259_apic, i8259_pin;
1907 unsigned long flags;
1910 * The number of IO-APIC IRQ registers (== #pins):
1912 for (apic = 0; apic < nr_ioapics; apic++) {
1913 spin_lock_irqsave(&ioapic_lock, flags);
1914 reg_01.raw = io_apic_read(apic, 1);
1915 spin_unlock_irqrestore(&ioapic_lock, flags);
1916 nr_ioapic_registers[apic] = reg_01.bits.entries+1;
1918 for(apic = 0; apic < nr_ioapics; apic++) {
1920 /* See if any of the pins is in ExtINT mode */
1921 for (pin = 0; pin < nr_ioapic_registers[apic]; pin++) {
1922 struct IO_APIC_route_entry entry;
1923 entry = ioapic_read_entry(apic, pin);
1925 /* If the interrupt line is enabled and in ExtInt mode
1926 * I have found the pin where the i8259 is connected.
1928 if ((entry.mask == 0) && (entry.delivery_mode == dest_ExtINT)) {
1929 ioapic_i8259.apic = apic;
1930 ioapic_i8259.pin = pin;
1936 /* Look to see what if the MP table has reported the ExtINT */
1937 /* If we could not find the appropriate pin by looking at the ioapic
1938 * the i8259 probably is not connected the ioapic but give the
1939 * mptable a chance anyway.
1941 i8259_pin = find_isa_irq_pin(0, mp_ExtINT);
1942 i8259_apic = find_isa_irq_apic(0, mp_ExtINT);
1943 /* Trust the MP table if nothing is setup in the hardware */
1944 if ((ioapic_i8259.pin == -1) && (i8259_pin >= 0)) {
1945 printk(KERN_WARNING "ExtINT not setup in hardware but reported by MP table\n");
1946 ioapic_i8259.pin = i8259_pin;
1947 ioapic_i8259.apic = i8259_apic;
1949 /* Complain if the MP table and the hardware disagree */
1950 if (((ioapic_i8259.apic != i8259_apic) || (ioapic_i8259.pin != i8259_pin)) &&
1951 (i8259_pin >= 0) && (ioapic_i8259.pin >= 0))
1953 printk(KERN_WARNING "ExtINT in hardware and MP table differ\n");
1957 * Do not trust the IO-APIC being empty at bootup
1963 * Not an __init, needed by the reboot code
1965 void disable_IO_APIC(void)
1968 * Clear the IO-APIC before rebooting:
1973 * If the i8259 is routed through an IOAPIC
1974 * Put that IOAPIC in virtual wire mode
1975 * so legacy interrupts can be delivered.
1977 * With interrupt-remapping, for now we will use virtual wire A mode,
1978 * as virtual wire B is little complex (need to configure both
1979 * IOAPIC RTE aswell as interrupt-remapping table entry).
1980 * As this gets called during crash dump, keep this simple for now.
1982 if (ioapic_i8259.pin != -1 && !intr_remapping_enabled) {
1983 struct IO_APIC_route_entry entry;
1985 memset(&entry, 0, sizeof(entry));
1986 entry.mask = 0; /* Enabled */
1987 entry.trigger = 0; /* Edge */
1989 entry.polarity = 0; /* High */
1990 entry.delivery_status = 0;
1991 entry.dest_mode = 0; /* Physical */
1992 entry.delivery_mode = dest_ExtINT; /* ExtInt */
1994 entry.dest = read_apic_id();
1997 * Add it to the IO-APIC irq-routing table:
1999 ioapic_write_entry(ioapic_i8259.apic, ioapic_i8259.pin, entry);
2003 * Use virtual wire A mode when interrupt remapping is enabled.
2006 disconnect_bsp_APIC(!intr_remapping_enabled &&
2007 ioapic_i8259.pin != -1);
2010 #ifdef CONFIG_X86_32
2012 * function to set the IO-APIC physical IDs based on the
2013 * values stored in the MPC table.
2015 * by Matt Domsch <Matt_Domsch@dell.com> Tue Dec 21 12:25:05 CST 1999
2018 static void __init setup_ioapic_ids_from_mpc(void)
2020 union IO_APIC_reg_00 reg_00;
2021 physid_mask_t phys_id_present_map;
2024 unsigned char old_id;
2025 unsigned long flags;
2027 if (x86_quirks->setup_ioapic_ids && x86_quirks->setup_ioapic_ids())
2031 * Don't check I/O APIC IDs for xAPIC systems. They have
2032 * no meaning without the serial APIC bus.
2034 if (!(boot_cpu_data.x86_vendor == X86_VENDOR_INTEL)
2035 || APIC_XAPIC(apic_version[boot_cpu_physical_apicid]))
2038 * This is broken; anything with a real cpu count has to
2039 * circumvent this idiocy regardless.
2041 phys_id_present_map = apic->ioapic_phys_id_map(phys_cpu_present_map);
2044 * Set the IOAPIC ID to the value stored in the MPC table.
2046 for (apic_id = 0; apic_id < nr_ioapics; apic_id++) {
2048 /* Read the register 0 value */
2049 spin_lock_irqsave(&ioapic_lock, flags);
2050 reg_00.raw = io_apic_read(apic_id, 0);
2051 spin_unlock_irqrestore(&ioapic_lock, flags);
2053 old_id = mp_ioapics[apic_id].apicid;
2055 if (mp_ioapics[apic_id].apicid >= get_physical_broadcast()) {
2056 printk(KERN_ERR "BIOS bug, IO-APIC#%d ID is %d in the MPC table!...\n",
2057 apic_id, mp_ioapics[apic_id].apicid);
2058 printk(KERN_ERR "... fixing up to %d. (tell your hw vendor)\n",
2060 mp_ioapics[apic_id].apicid = reg_00.bits.ID;
2064 * Sanity check, is the ID really free? Every APIC in a
2065 * system must have a unique ID or we get lots of nice
2066 * 'stuck on smp_invalidate_needed IPI wait' messages.
2068 if (apic->check_apicid_used(phys_id_present_map,
2069 mp_ioapics[apic_id].apicid)) {
2070 printk(KERN_ERR "BIOS bug, IO-APIC#%d ID %d is already used!...\n",
2071 apic_id, mp_ioapics[apic_id].apicid);
2072 for (i = 0; i < get_physical_broadcast(); i++)
2073 if (!physid_isset(i, phys_id_present_map))
2075 if (i >= get_physical_broadcast())
2076 panic("Max APIC ID exceeded!\n");
2077 printk(KERN_ERR "... fixing up to %d. (tell your hw vendor)\n",
2079 physid_set(i, phys_id_present_map);
2080 mp_ioapics[apic_id].apicid = i;
2083 tmp = apic->apicid_to_cpu_present(mp_ioapics[apic_id].apicid);
2084 apic_printk(APIC_VERBOSE, "Setting %d in the "
2085 "phys_id_present_map\n",
2086 mp_ioapics[apic_id].apicid);
2087 physids_or(phys_id_present_map, phys_id_present_map, tmp);
2092 * We need to adjust the IRQ routing table
2093 * if the ID changed.
2095 if (old_id != mp_ioapics[apic_id].apicid)
2096 for (i = 0; i < mp_irq_entries; i++)
2097 if (mp_irqs[i].dstapic == old_id)
2099 = mp_ioapics[apic_id].apicid;
2102 * Read the right value from the MPC table and
2103 * write it into the ID register.
2105 apic_printk(APIC_VERBOSE, KERN_INFO
2106 "...changing IO-APIC physical APIC ID to %d ...",
2107 mp_ioapics[apic_id].apicid);
2109 reg_00.bits.ID = mp_ioapics[apic_id].apicid;
2110 spin_lock_irqsave(&ioapic_lock, flags);
2111 io_apic_write(apic_id, 0, reg_00.raw);
2112 spin_unlock_irqrestore(&ioapic_lock, flags);
2117 spin_lock_irqsave(&ioapic_lock, flags);
2118 reg_00.raw = io_apic_read(apic_id, 0);
2119 spin_unlock_irqrestore(&ioapic_lock, flags);
2120 if (reg_00.bits.ID != mp_ioapics[apic_id].apicid)
2121 printk("could not set ID!\n");
2123 apic_printk(APIC_VERBOSE, " ok.\n");
2128 int no_timer_check __initdata;
2130 static int __init notimercheck(char *s)
2135 __setup("no_timer_check", notimercheck);
2138 * There is a nasty bug in some older SMP boards, their mptable lies
2139 * about the timer IRQ. We do the following to work around the situation:
2141 * - timer IRQ defaults to IO-APIC IRQ
2142 * - if this function detects that timer IRQs are defunct, then we fall
2143 * back to ISA timer IRQs
2145 static int __init timer_irq_works(void)
2147 unsigned long t1 = jiffies;
2148 unsigned long flags;
2153 local_save_flags(flags);
2155 /* Let ten ticks pass... */
2156 mdelay((10 * 1000) / HZ);
2157 local_irq_restore(flags);
2160 * Expect a few ticks at least, to be sure some possible
2161 * glue logic does not lock up after one or two first
2162 * ticks in a non-ExtINT mode. Also the local APIC
2163 * might have cached one ExtINT interrupt. Finally, at
2164 * least one tick may be lost due to delays.
2168 if (time_after(jiffies, t1 + 4))
2174 * In the SMP+IOAPIC case it might happen that there are an unspecified
2175 * number of pending IRQ events unhandled. These cases are very rare,
2176 * so we 'resend' these IRQs via IPIs, to the same CPU. It's much
2177 * better to do it this way as thus we do not have to be aware of
2178 * 'pending' interrupts in the IRQ path, except at this point.
2181 * Edge triggered needs to resend any interrupt
2182 * that was delayed but this is now handled in the device
2187 * Starting up a edge-triggered IO-APIC interrupt is
2188 * nasty - we need to make sure that we get the edge.
2189 * If it is already asserted for some reason, we need
2190 * return 1 to indicate that is was pending.
2192 * This is not complete - we should be able to fake
2193 * an edge even if it isn't on the 8259A...
2196 static unsigned int startup_ioapic_irq(unsigned int irq)
2198 int was_pending = 0;
2199 unsigned long flags;
2200 struct irq_cfg *cfg;
2202 spin_lock_irqsave(&ioapic_lock, flags);
2203 if (irq < NR_IRQS_LEGACY) {
2204 disable_8259A_irq(irq);
2205 if (i8259A_irq_pending(irq))
2209 __unmask_IO_APIC_irq(cfg);
2210 spin_unlock_irqrestore(&ioapic_lock, flags);
2215 #ifdef CONFIG_X86_64
2216 static int ioapic_retrigger_irq(unsigned int irq)
2219 struct irq_cfg *cfg = irq_cfg(irq);
2220 unsigned long flags;
2222 spin_lock_irqsave(&vector_lock, flags);
2223 apic->send_IPI_mask(cpumask_of(cpumask_first(cfg->domain)), cfg->vector);
2224 spin_unlock_irqrestore(&vector_lock, flags);
2229 static int ioapic_retrigger_irq(unsigned int irq)
2231 apic->send_IPI_self(irq_cfg(irq)->vector);
2238 * Level and edge triggered IO-APIC interrupts need different handling,
2239 * so we use two separate IRQ descriptors. Edge triggered IRQs can be
2240 * handled with the level-triggered descriptor, but that one has slightly
2241 * more overhead. Level-triggered interrupts cannot be handled with the
2242 * edge-triggered handler, without risking IRQ storms and other ugly
2247 static void send_cleanup_vector(struct irq_cfg *cfg)
2249 cpumask_var_t cleanup_mask;
2251 if (unlikely(!alloc_cpumask_var(&cleanup_mask, GFP_ATOMIC))) {
2253 cfg->move_cleanup_count = 0;
2254 for_each_cpu_and(i, cfg->old_domain, cpu_online_mask)
2255 cfg->move_cleanup_count++;
2256 for_each_cpu_and(i, cfg->old_domain, cpu_online_mask)
2257 apic->send_IPI_mask(cpumask_of(i), IRQ_MOVE_CLEANUP_VECTOR);
2259 cpumask_and(cleanup_mask, cfg->old_domain, cpu_online_mask);
2260 cfg->move_cleanup_count = cpumask_weight(cleanup_mask);
2261 apic->send_IPI_mask(cleanup_mask, IRQ_MOVE_CLEANUP_VECTOR);
2262 free_cpumask_var(cleanup_mask);
2264 cfg->move_in_progress = 0;
2267 static void __target_IO_APIC_irq(unsigned int irq, unsigned int dest, struct irq_cfg *cfg)
2270 struct irq_pin_list *entry;
2271 u8 vector = cfg->vector;
2273 entry = cfg->irq_2_pin;
2283 * With interrupt-remapping, destination information comes
2284 * from interrupt-remapping table entry.
2286 if (!irq_remapped(irq))
2287 io_apic_write(apic, 0x11 + pin*2, dest);
2288 reg = io_apic_read(apic, 0x10 + pin*2);
2289 reg &= ~IO_APIC_REDIR_VECTOR_MASK;
2291 io_apic_modify(apic, 0x10 + pin*2, reg);
2294 entry = entry->next;
2299 assign_irq_vector(int irq, struct irq_cfg *cfg, const struct cpumask *mask);
2302 * Either sets desc->affinity to a valid value, and returns
2303 * ->cpu_mask_to_apicid of that, or returns BAD_APICID and
2304 * leaves desc->affinity untouched.
2307 set_desc_affinity(struct irq_desc *desc, const struct cpumask *mask)
2309 struct irq_cfg *cfg;
2312 if (!cpumask_intersects(mask, cpu_online_mask))
2316 cfg = desc->chip_data;
2317 if (assign_irq_vector(irq, cfg, mask))
2320 cpumask_copy(desc->affinity, mask);
2322 return apic->cpu_mask_to_apicid_and(desc->affinity, cfg->domain);
2326 set_ioapic_affinity_irq_desc(struct irq_desc *desc, const struct cpumask *mask)
2328 struct irq_cfg *cfg;
2329 unsigned long flags;
2335 cfg = desc->chip_data;
2337 spin_lock_irqsave(&ioapic_lock, flags);
2338 dest = set_desc_affinity(desc, mask);
2339 if (dest != BAD_APICID) {
2340 /* Only the high 8 bits are valid. */
2341 dest = SET_APIC_LOGICAL_ID(dest);
2342 __target_IO_APIC_irq(irq, dest, cfg);
2345 spin_unlock_irqrestore(&ioapic_lock, flags);
2351 set_ioapic_affinity_irq(unsigned int irq, const struct cpumask *mask)
2353 struct irq_desc *desc;
2355 desc = irq_to_desc(irq);
2357 return set_ioapic_affinity_irq_desc(desc, mask);
2360 #ifdef CONFIG_INTR_REMAP
2363 * Migrate the IO-APIC irq in the presence of intr-remapping.
2365 * For both level and edge triggered, irq migration is a simple atomic
2366 * update(of vector and cpu destination) of IRTE and flush the hardware cache.
2368 * For level triggered, we eliminate the io-apic RTE modification (with the
2369 * updated vector information), by using a virtual vector (io-apic pin number).
2370 * Real vector that is used for interrupting cpu will be coming from
2371 * the interrupt-remapping table entry.
2374 migrate_ioapic_irq_desc(struct irq_desc *desc, const struct cpumask *mask)
2376 struct irq_cfg *cfg;
2382 if (!cpumask_intersects(mask, cpu_online_mask))
2386 if (get_irte(irq, &irte))
2389 cfg = desc->chip_data;
2390 if (assign_irq_vector(irq, cfg, mask))
2393 dest = apic->cpu_mask_to_apicid_and(cfg->domain, mask);
2395 irte.vector = cfg->vector;
2396 irte.dest_id = IRTE_DEST(dest);
2399 * Modified the IRTE and flushes the Interrupt entry cache.
2401 modify_irte(irq, &irte);
2403 if (cfg->move_in_progress)
2404 send_cleanup_vector(cfg);
2406 cpumask_copy(desc->affinity, mask);
2412 * Migrates the IRQ destination in the process context.
2414 static int set_ir_ioapic_affinity_irq_desc(struct irq_desc *desc,
2415 const struct cpumask *mask)
2417 return migrate_ioapic_irq_desc(desc, mask);
2419 static int set_ir_ioapic_affinity_irq(unsigned int irq,
2420 const struct cpumask *mask)
2422 struct irq_desc *desc = irq_to_desc(irq);
2424 return set_ir_ioapic_affinity_irq_desc(desc, mask);
2427 static inline int set_ir_ioapic_affinity_irq_desc(struct irq_desc *desc,
2428 const struct cpumask *mask)
2434 asmlinkage void smp_irq_move_cleanup_interrupt(void)
2436 unsigned vector, me;
2442 me = smp_processor_id();
2443 for (vector = FIRST_EXTERNAL_VECTOR; vector < NR_VECTORS; vector++) {
2446 struct irq_desc *desc;
2447 struct irq_cfg *cfg;
2448 irq = __get_cpu_var(vector_irq)[vector];
2453 desc = irq_to_desc(irq);
2458 spin_lock(&desc->lock);
2459 if (!cfg->move_cleanup_count)
2462 if (vector == cfg->vector && cpumask_test_cpu(me, cfg->domain))
2465 irr = apic_read(APIC_IRR + (vector / 32 * 0x10));
2467 * Check if the vector that needs to be cleanedup is
2468 * registered at the cpu's IRR. If so, then this is not
2469 * the best time to clean it up. Lets clean it up in the
2470 * next attempt by sending another IRQ_MOVE_CLEANUP_VECTOR
2473 if (irr & (1 << (vector % 32))) {
2474 apic->send_IPI_self(IRQ_MOVE_CLEANUP_VECTOR);
2477 __get_cpu_var(vector_irq)[vector] = -1;
2478 cfg->move_cleanup_count--;
2480 spin_unlock(&desc->lock);
2486 static void irq_complete_move(struct irq_desc **descp)
2488 struct irq_desc *desc = *descp;
2489 struct irq_cfg *cfg = desc->chip_data;
2490 unsigned vector, me;
2492 if (likely(!cfg->move_in_progress))
2495 vector = ~get_irq_regs()->orig_ax;
2496 me = smp_processor_id();
2498 if (vector == cfg->vector && cpumask_test_cpu(me, cfg->domain))
2499 send_cleanup_vector(cfg);
2502 static inline void irq_complete_move(struct irq_desc **descp) {}
2505 static void ack_apic_edge(unsigned int irq)
2507 struct irq_desc *desc = irq_to_desc(irq);
2509 irq_complete_move(&desc);
2510 move_native_irq(irq);
2514 atomic_t irq_mis_count;
2516 static void ack_apic_level(unsigned int irq)
2518 struct irq_desc *desc = irq_to_desc(irq);
2520 #ifdef CONFIG_X86_32
2524 struct irq_cfg *cfg;
2525 int do_unmask_irq = 0;
2527 irq_complete_move(&desc);
2528 #ifdef CONFIG_GENERIC_PENDING_IRQ
2529 /* If we are moving the irq we need to mask it */
2530 if (unlikely(desc->status & IRQ_MOVE_PENDING)) {
2532 mask_IO_APIC_irq_desc(desc);
2536 #ifdef CONFIG_X86_32
2538 * It appears there is an erratum which affects at least version 0x11
2539 * of I/O APIC (that's the 82093AA and cores integrated into various
2540 * chipsets). Under certain conditions a level-triggered interrupt is
2541 * erroneously delivered as edge-triggered one but the respective IRR
2542 * bit gets set nevertheless. As a result the I/O unit expects an EOI
2543 * message but it will never arrive and further interrupts are blocked
2544 * from the source. The exact reason is so far unknown, but the
2545 * phenomenon was observed when two consecutive interrupt requests
2546 * from a given source get delivered to the same CPU and the source is
2547 * temporarily disabled in between.
2549 * A workaround is to simulate an EOI message manually. We achieve it
2550 * by setting the trigger mode to edge and then to level when the edge
2551 * trigger mode gets detected in the TMR of a local APIC for a
2552 * level-triggered interrupt. We mask the source for the time of the
2553 * operation to prevent an edge-triggered interrupt escaping meanwhile.
2554 * The idea is from Manfred Spraul. --macro
2556 cfg = desc->chip_data;
2559 v = apic_read(APIC_TMR + ((i & ~0x1f) >> 1));
2563 * We must acknowledge the irq before we move it or the acknowledge will
2564 * not propagate properly.
2568 /* Now we can move and renable the irq */
2569 if (unlikely(do_unmask_irq)) {
2570 /* Only migrate the irq if the ack has been received.
2572 * On rare occasions the broadcast level triggered ack gets
2573 * delayed going to ioapics, and if we reprogram the
2574 * vector while Remote IRR is still set the irq will never
2577 * To prevent this scenario we read the Remote IRR bit
2578 * of the ioapic. This has two effects.
2579 * - On any sane system the read of the ioapic will
2580 * flush writes (and acks) going to the ioapic from
2582 * - We get to see if the ACK has actually been delivered.
2584 * Based on failed experiments of reprogramming the
2585 * ioapic entry from outside of irq context starting
2586 * with masking the ioapic entry and then polling until
2587 * Remote IRR was clear before reprogramming the
2588 * ioapic I don't trust the Remote IRR bit to be
2589 * completey accurate.
2591 * However there appears to be no other way to plug
2592 * this race, so if the Remote IRR bit is not
2593 * accurate and is causing problems then it is a hardware bug
2594 * and you can go talk to the chipset vendor about it.
2596 cfg = desc->chip_data;
2597 if (!io_apic_level_ack_pending(cfg))
2598 move_masked_irq(irq);
2599 unmask_IO_APIC_irq_desc(desc);
2602 #ifdef CONFIG_X86_32
2603 if (!(v & (1 << (i & 0x1f)))) {
2604 atomic_inc(&irq_mis_count);
2605 spin_lock(&ioapic_lock);
2606 __mask_and_edge_IO_APIC_irq(cfg);
2607 __unmask_and_level_IO_APIC_irq(cfg);
2608 spin_unlock(&ioapic_lock);
2613 #ifdef CONFIG_INTR_REMAP
2614 static void __eoi_ioapic_irq(unsigned int irq, struct irq_cfg *cfg)
2617 struct irq_pin_list *entry;
2619 entry = cfg->irq_2_pin;
2627 io_apic_eoi(apic, pin);
2628 entry = entry->next;
2633 eoi_ioapic_irq(struct irq_desc *desc)
2635 struct irq_cfg *cfg;
2636 unsigned long flags;
2640 cfg = desc->chip_data;
2642 spin_lock_irqsave(&ioapic_lock, flags);
2643 __eoi_ioapic_irq(irq, cfg);
2644 spin_unlock_irqrestore(&ioapic_lock, flags);
2647 static void ir_ack_apic_edge(unsigned int irq)
2652 static void ir_ack_apic_level(unsigned int irq)
2654 struct irq_desc *desc = irq_to_desc(irq);
2657 eoi_ioapic_irq(desc);
2659 #endif /* CONFIG_INTR_REMAP */
2661 static struct irq_chip ioapic_chip __read_mostly = {
2663 .startup = startup_ioapic_irq,
2664 .mask = mask_IO_APIC_irq,
2665 .unmask = unmask_IO_APIC_irq,
2666 .ack = ack_apic_edge,
2667 .eoi = ack_apic_level,
2669 .set_affinity = set_ioapic_affinity_irq,
2671 .retrigger = ioapic_retrigger_irq,
2674 static struct irq_chip ir_ioapic_chip __read_mostly = {
2675 .name = "IR-IO-APIC",
2676 .startup = startup_ioapic_irq,
2677 .mask = mask_IO_APIC_irq,
2678 .unmask = unmask_IO_APIC_irq,
2679 #ifdef CONFIG_INTR_REMAP
2680 .ack = ir_ack_apic_edge,
2681 .eoi = ir_ack_apic_level,
2683 .set_affinity = set_ir_ioapic_affinity_irq,
2686 .retrigger = ioapic_retrigger_irq,
2689 static inline void init_IO_APIC_traps(void)
2692 struct irq_desc *desc;
2693 struct irq_cfg *cfg;
2696 * NOTE! The local APIC isn't very good at handling
2697 * multiple interrupts at the same interrupt level.
2698 * As the interrupt level is determined by taking the
2699 * vector number and shifting that right by 4, we
2700 * want to spread these out a bit so that they don't
2701 * all fall in the same interrupt level.
2703 * Also, we've got to be careful not to trash gate
2704 * 0x80, because int 0x80 is hm, kind of importantish. ;)
2706 for_each_irq_desc(irq, desc) {
2707 cfg = desc->chip_data;
2708 if (IO_APIC_IRQ(irq) && cfg && !cfg->vector) {
2710 * Hmm.. We don't have an entry for this,
2711 * so default to an old-fashioned 8259
2712 * interrupt if we can..
2714 if (irq < NR_IRQS_LEGACY)
2715 make_8259A_irq(irq);
2717 /* Strange. Oh, well.. */
2718 desc->chip = &no_irq_chip;
2724 * The local APIC irq-chip implementation:
2727 static void mask_lapic_irq(unsigned int irq)
2731 v = apic_read(APIC_LVT0);
2732 apic_write(APIC_LVT0, v | APIC_LVT_MASKED);
2735 static void unmask_lapic_irq(unsigned int irq)
2739 v = apic_read(APIC_LVT0);
2740 apic_write(APIC_LVT0, v & ~APIC_LVT_MASKED);
2743 static void ack_lapic_irq(unsigned int irq)
2748 static struct irq_chip lapic_chip __read_mostly = {
2749 .name = "local-APIC",
2750 .mask = mask_lapic_irq,
2751 .unmask = unmask_lapic_irq,
2752 .ack = ack_lapic_irq,
2755 static void lapic_register_intr(int irq, struct irq_desc *desc)
2757 desc->status &= ~IRQ_LEVEL;
2758 set_irq_chip_and_handler_name(irq, &lapic_chip, handle_edge_irq,
2762 static void __init setup_nmi(void)
2765 * Dirty trick to enable the NMI watchdog ...
2766 * We put the 8259A master into AEOI mode and
2767 * unmask on all local APICs LVT0 as NMI.
2769 * The idea to use the 8259A in AEOI mode ('8259A Virtual Wire')
2770 * is from Maciej W. Rozycki - so we do not have to EOI from
2771 * the NMI handler or the timer interrupt.
2773 apic_printk(APIC_VERBOSE, KERN_INFO "activating NMI Watchdog ...");
2775 enable_NMI_through_LVT0();
2777 apic_printk(APIC_VERBOSE, " done.\n");
2781 * This looks a bit hackish but it's about the only one way of sending
2782 * a few INTA cycles to 8259As and any associated glue logic. ICR does
2783 * not support the ExtINT mode, unfortunately. We need to send these
2784 * cycles as some i82489DX-based boards have glue logic that keeps the
2785 * 8259A interrupt line asserted until INTA. --macro
2787 static inline void __init unlock_ExtINT_logic(void)
2790 struct IO_APIC_route_entry entry0, entry1;
2791 unsigned char save_control, save_freq_select;
2793 pin = find_isa_irq_pin(8, mp_INT);
2798 apic = find_isa_irq_apic(8, mp_INT);
2804 entry0 = ioapic_read_entry(apic, pin);
2805 clear_IO_APIC_pin(apic, pin);
2807 memset(&entry1, 0, sizeof(entry1));
2809 entry1.dest_mode = 0; /* physical delivery */
2810 entry1.mask = 0; /* unmask IRQ now */
2811 entry1.dest = hard_smp_processor_id();
2812 entry1.delivery_mode = dest_ExtINT;
2813 entry1.polarity = entry0.polarity;
2817 ioapic_write_entry(apic, pin, entry1);
2819 save_control = CMOS_READ(RTC_CONTROL);
2820 save_freq_select = CMOS_READ(RTC_FREQ_SELECT);
2821 CMOS_WRITE((save_freq_select & ~RTC_RATE_SELECT) | 0x6,
2823 CMOS_WRITE(save_control | RTC_PIE, RTC_CONTROL);
2828 if ((CMOS_READ(RTC_INTR_FLAGS) & RTC_PF) == RTC_PF)
2832 CMOS_WRITE(save_control, RTC_CONTROL);
2833 CMOS_WRITE(save_freq_select, RTC_FREQ_SELECT);
2834 clear_IO_APIC_pin(apic, pin);
2836 ioapic_write_entry(apic, pin, entry0);
2839 static int disable_timer_pin_1 __initdata;
2840 /* Actually the next is obsolete, but keep it for paranoid reasons -AK */
2841 static int __init disable_timer_pin_setup(char *arg)
2843 disable_timer_pin_1 = 1;
2846 early_param("disable_timer_pin_1", disable_timer_pin_setup);
2848 int timer_through_8259 __initdata;
2851 * This code may look a bit paranoid, but it's supposed to cooperate with
2852 * a wide range of boards and BIOS bugs. Fortunately only the timer IRQ
2853 * is so screwy. Thanks to Brian Perkins for testing/hacking this beast
2854 * fanatically on his truly buggy board.
2856 * FIXME: really need to revamp this for all platforms.
2858 static inline void __init check_timer(void)
2860 struct irq_desc *desc = irq_to_desc(0);
2861 struct irq_cfg *cfg = desc->chip_data;
2862 int node = cpu_to_node(boot_cpu_id);
2863 int apic1, pin1, apic2, pin2;
2864 unsigned long flags;
2867 local_irq_save(flags);
2870 * get/set the timer IRQ vector:
2872 disable_8259A_irq(0);
2873 assign_irq_vector(0, cfg, apic->target_cpus());
2876 * As IRQ0 is to be enabled in the 8259A, the virtual
2877 * wire has to be disabled in the local APIC. Also
2878 * timer interrupts need to be acknowledged manually in
2879 * the 8259A for the i82489DX when using the NMI
2880 * watchdog as that APIC treats NMIs as level-triggered.
2881 * The AEOI mode will finish them in the 8259A
2884 apic_write(APIC_LVT0, APIC_LVT_MASKED | APIC_DM_EXTINT);
2886 #ifdef CONFIG_X86_32
2890 ver = apic_read(APIC_LVR);
2891 ver = GET_APIC_VERSION(ver);
2892 timer_ack = (nmi_watchdog == NMI_IO_APIC && !APIC_INTEGRATED(ver));
2896 pin1 = find_isa_irq_pin(0, mp_INT);
2897 apic1 = find_isa_irq_apic(0, mp_INT);
2898 pin2 = ioapic_i8259.pin;
2899 apic2 = ioapic_i8259.apic;
2901 apic_printk(APIC_QUIET, KERN_INFO "..TIMER: vector=0x%02X "
2902 "apic1=%d pin1=%d apic2=%d pin2=%d\n",
2903 cfg->vector, apic1, pin1, apic2, pin2);
2906 * Some BIOS writers are clueless and report the ExtINTA
2907 * I/O APIC input from the cascaded 8259A as the timer
2908 * interrupt input. So just in case, if only one pin
2909 * was found above, try it both directly and through the
2913 if (intr_remapping_enabled)
2914 panic("BIOS bug: timer not connected to IO-APIC");
2918 } else if (pin2 == -1) {
2925 * Ok, does IRQ0 through the IOAPIC work?
2928 add_pin_to_irq_node(cfg, node, apic1, pin1);
2929 setup_timer_IRQ0_pin(apic1, pin1, cfg->vector);
2931 /* for edge trigger, setup_IO_APIC_irq already
2932 * leave it unmasked.
2933 * so only need to unmask if it is level-trigger
2934 * do we really have level trigger timer?
2937 idx = find_irq_entry(apic1, pin1, mp_INT);
2938 if (idx != -1 && irq_trigger(idx))
2939 unmask_IO_APIC_irq_desc(desc);
2941 if (timer_irq_works()) {
2942 if (nmi_watchdog == NMI_IO_APIC) {
2944 enable_8259A_irq(0);
2946 if (disable_timer_pin_1 > 0)
2947 clear_IO_APIC_pin(0, pin1);
2950 if (intr_remapping_enabled)
2951 panic("timer doesn't work through Interrupt-remapped IO-APIC");
2952 local_irq_disable();
2953 clear_IO_APIC_pin(apic1, pin1);
2955 apic_printk(APIC_QUIET, KERN_ERR "..MP-BIOS bug: "
2956 "8254 timer not connected to IO-APIC\n");
2958 apic_printk(APIC_QUIET, KERN_INFO "...trying to set up timer "
2959 "(IRQ0) through the 8259A ...\n");
2960 apic_printk(APIC_QUIET, KERN_INFO
2961 "..... (found apic %d pin %d) ...\n", apic2, pin2);
2963 * legacy devices should be connected to IO APIC #0
2965 replace_pin_at_irq_node(cfg, node, apic1, pin1, apic2, pin2);
2966 setup_timer_IRQ0_pin(apic2, pin2, cfg->vector);
2967 enable_8259A_irq(0);
2968 if (timer_irq_works()) {
2969 apic_printk(APIC_QUIET, KERN_INFO "....... works.\n");
2970 timer_through_8259 = 1;
2971 if (nmi_watchdog == NMI_IO_APIC) {
2972 disable_8259A_irq(0);
2974 enable_8259A_irq(0);
2979 * Cleanup, just in case ...
2981 local_irq_disable();
2982 disable_8259A_irq(0);
2983 clear_IO_APIC_pin(apic2, pin2);
2984 apic_printk(APIC_QUIET, KERN_INFO "....... failed.\n");
2987 if (nmi_watchdog == NMI_IO_APIC) {
2988 apic_printk(APIC_QUIET, KERN_WARNING "timer doesn't work "
2989 "through the IO-APIC - disabling NMI Watchdog!\n");
2990 nmi_watchdog = NMI_NONE;
2992 #ifdef CONFIG_X86_32
2996 apic_printk(APIC_QUIET, KERN_INFO
2997 "...trying to set up timer as Virtual Wire IRQ...\n");
2999 lapic_register_intr(0, desc);
3000 apic_write(APIC_LVT0, APIC_DM_FIXED | cfg->vector); /* Fixed mode */
3001 enable_8259A_irq(0);
3003 if (timer_irq_works()) {
3004 apic_printk(APIC_QUIET, KERN_INFO "..... works.\n");
3007 local_irq_disable();
3008 disable_8259A_irq(0);
3009 apic_write(APIC_LVT0, APIC_LVT_MASKED | APIC_DM_FIXED | cfg->vector);
3010 apic_printk(APIC_QUIET, KERN_INFO "..... failed.\n");
3012 apic_printk(APIC_QUIET, KERN_INFO
3013 "...trying to set up timer as ExtINT IRQ...\n");
3017 apic_write(APIC_LVT0, APIC_DM_EXTINT);
3019 unlock_ExtINT_logic();
3021 if (timer_irq_works()) {
3022 apic_printk(APIC_QUIET, KERN_INFO "..... works.\n");
3025 local_irq_disable();
3026 apic_printk(APIC_QUIET, KERN_INFO "..... failed :(.\n");
3027 panic("IO-APIC + timer doesn't work! Boot with apic=debug and send a "
3028 "report. Then try booting with the 'noapic' option.\n");
3030 local_irq_restore(flags);
3034 * Traditionally ISA IRQ2 is the cascade IRQ, and is not available
3035 * to devices. However there may be an I/O APIC pin available for
3036 * this interrupt regardless. The pin may be left unconnected, but
3037 * typically it will be reused as an ExtINT cascade interrupt for
3038 * the master 8259A. In the MPS case such a pin will normally be
3039 * reported as an ExtINT interrupt in the MP table. With ACPI
3040 * there is no provision for ExtINT interrupts, and in the absence
3041 * of an override it would be treated as an ordinary ISA I/O APIC
3042 * interrupt, that is edge-triggered and unmasked by default. We
3043 * used to do this, but it caused problems on some systems because
3044 * of the NMI watchdog and sometimes IRQ0 of the 8254 timer using
3045 * the same ExtINT cascade interrupt to drive the local APIC of the
3046 * bootstrap processor. Therefore we refrain from routing IRQ2 to
3047 * the I/O APIC in all cases now. No actual device should request
3048 * it anyway. --macro
3050 #define PIC_IRQS (1 << PIC_CASCADE_IR)
3052 void __init setup_IO_APIC(void)
3056 * calling enable_IO_APIC() is moved to setup_local_APIC for BP
3059 io_apic_irqs = ~PIC_IRQS;
3061 apic_printk(APIC_VERBOSE, "ENABLING IO-APIC IRQs\n");
3063 * Set up IO-APIC IRQ routing.
3065 #ifdef CONFIG_X86_32
3067 setup_ioapic_ids_from_mpc();
3070 setup_IO_APIC_irqs();
3071 init_IO_APIC_traps();
3076 * Called after all the initialization is done. If we didnt find any
3077 * APIC bugs then we can allow the modify fast path
3080 static int __init io_apic_bug_finalize(void)
3082 if (sis_apic_bug == -1)
3087 late_initcall(io_apic_bug_finalize);
3089 struct sysfs_ioapic_data {
3090 struct sys_device dev;
3091 struct IO_APIC_route_entry entry[0];
3093 static struct sysfs_ioapic_data * mp_ioapic_data[MAX_IO_APICS];
3095 static int ioapic_suspend(struct sys_device *dev, pm_message_t state)
3097 struct IO_APIC_route_entry *entry;
3098 struct sysfs_ioapic_data *data;
3101 data = container_of(dev, struct sysfs_ioapic_data, dev);
3102 entry = data->entry;
3103 for (i = 0; i < nr_ioapic_registers[dev->id]; i ++, entry ++ )
3104 *entry = ioapic_read_entry(dev->id, i);
3109 static int ioapic_resume(struct sys_device *dev)
3111 struct IO_APIC_route_entry *entry;
3112 struct sysfs_ioapic_data *data;
3113 unsigned long flags;
3114 union IO_APIC_reg_00 reg_00;
3117 data = container_of(dev, struct sysfs_ioapic_data, dev);
3118 entry = data->entry;
3120 spin_lock_irqsave(&ioapic_lock, flags);
3121 reg_00.raw = io_apic_read(dev->id, 0);
3122 if (reg_00.bits.ID != mp_ioapics[dev->id].apicid) {
3123 reg_00.bits.ID = mp_ioapics[dev->id].apicid;
3124 io_apic_write(dev->id, 0, reg_00.raw);
3126 spin_unlock_irqrestore(&ioapic_lock, flags);
3127 for (i = 0; i < nr_ioapic_registers[dev->id]; i++)
3128 ioapic_write_entry(dev->id, i, entry[i]);
3133 static struct sysdev_class ioapic_sysdev_class = {
3135 .suspend = ioapic_suspend,
3136 .resume = ioapic_resume,
3139 static int __init ioapic_init_sysfs(void)
3141 struct sys_device * dev;
3144 error = sysdev_class_register(&ioapic_sysdev_class);
3148 for (i = 0; i < nr_ioapics; i++ ) {
3149 size = sizeof(struct sys_device) + nr_ioapic_registers[i]
3150 * sizeof(struct IO_APIC_route_entry);
3151 mp_ioapic_data[i] = kzalloc(size, GFP_KERNEL);
3152 if (!mp_ioapic_data[i]) {
3153 printk(KERN_ERR "Can't suspend/resume IOAPIC %d\n", i);
3156 dev = &mp_ioapic_data[i]->dev;
3158 dev->cls = &ioapic_sysdev_class;
3159 error = sysdev_register(dev);
3161 kfree(mp_ioapic_data[i]);
3162 mp_ioapic_data[i] = NULL;
3163 printk(KERN_ERR "Can't suspend/resume IOAPIC %d\n", i);
3171 device_initcall(ioapic_init_sysfs);
3173 static int nr_irqs_gsi = NR_IRQS_LEGACY;
3175 * Dynamic irq allocate and deallocation
3177 unsigned int create_irq_nr(unsigned int irq_want, int node)
3179 /* Allocate an unused irq */
3182 unsigned long flags;
3183 struct irq_cfg *cfg_new = NULL;
3184 struct irq_desc *desc_new = NULL;
3187 if (irq_want < nr_irqs_gsi)
3188 irq_want = nr_irqs_gsi;
3190 spin_lock_irqsave(&vector_lock, flags);
3191 for (new = irq_want; new < nr_irqs; new++) {
3192 desc_new = irq_to_desc_alloc_node(new, node);
3194 printk(KERN_INFO "can not get irq_desc for %d\n", new);
3197 cfg_new = desc_new->chip_data;
3199 if (cfg_new->vector != 0)
3202 desc_new = move_irq_desc(desc_new, node);
3204 if (__assign_irq_vector(new, cfg_new, apic->target_cpus()) == 0)
3208 spin_unlock_irqrestore(&vector_lock, flags);
3211 dynamic_irq_init(irq);
3212 /* restore it, in case dynamic_irq_init clear it */
3214 desc_new->chip_data = cfg_new;
3219 int create_irq(void)
3221 int node = cpu_to_node(boot_cpu_id);
3222 unsigned int irq_want;
3225 irq_want = nr_irqs_gsi;
3226 irq = create_irq_nr(irq_want, node);
3234 void destroy_irq(unsigned int irq)
3236 unsigned long flags;
3237 struct irq_cfg *cfg;
3238 struct irq_desc *desc;
3240 /* store it, in case dynamic_irq_cleanup clear it */
3241 desc = irq_to_desc(irq);
3242 cfg = desc->chip_data;
3243 dynamic_irq_cleanup(irq);
3244 /* connect back irq_cfg */
3246 desc->chip_data = cfg;
3249 spin_lock_irqsave(&vector_lock, flags);
3250 __clear_irq_vector(irq, cfg);
3251 spin_unlock_irqrestore(&vector_lock, flags);
3255 * MSI message composition
3257 #ifdef CONFIG_PCI_MSI
3258 static int msi_compose_msg(struct pci_dev *pdev, unsigned int irq, struct msi_msg *msg)
3260 struct irq_cfg *cfg;
3268 err = assign_irq_vector(irq, cfg, apic->target_cpus());
3272 dest = apic->cpu_mask_to_apicid_and(cfg->domain, apic->target_cpus());
3274 if (irq_remapped(irq)) {
3279 ir_index = map_irq_to_irte_handle(irq, &sub_handle);
3280 BUG_ON(ir_index == -1);
3282 memset (&irte, 0, sizeof(irte));
3285 irte.dst_mode = apic->irq_dest_mode;
3286 irte.trigger_mode = 0; /* edge */
3287 irte.dlvry_mode = apic->irq_delivery_mode;
3288 irte.vector = cfg->vector;
3289 irte.dest_id = IRTE_DEST(dest);
3291 /* Set source-id of interrupt request */
3292 set_msi_sid(&irte, pdev);
3294 modify_irte(irq, &irte);
3296 msg->address_hi = MSI_ADDR_BASE_HI;
3297 msg->data = sub_handle;
3298 msg->address_lo = MSI_ADDR_BASE_LO | MSI_ADDR_IR_EXT_INT |
3300 MSI_ADDR_IR_INDEX1(ir_index) |
3301 MSI_ADDR_IR_INDEX2(ir_index);
3303 if (x2apic_enabled())
3304 msg->address_hi = MSI_ADDR_BASE_HI |
3305 MSI_ADDR_EXT_DEST_ID(dest);
3307 msg->address_hi = MSI_ADDR_BASE_HI;
3311 ((apic->irq_dest_mode == 0) ?
3312 MSI_ADDR_DEST_MODE_PHYSICAL:
3313 MSI_ADDR_DEST_MODE_LOGICAL) |
3314 ((apic->irq_delivery_mode != dest_LowestPrio) ?
3315 MSI_ADDR_REDIRECTION_CPU:
3316 MSI_ADDR_REDIRECTION_LOWPRI) |
3317 MSI_ADDR_DEST_ID(dest);
3320 MSI_DATA_TRIGGER_EDGE |
3321 MSI_DATA_LEVEL_ASSERT |
3322 ((apic->irq_delivery_mode != dest_LowestPrio) ?
3323 MSI_DATA_DELIVERY_FIXED:
3324 MSI_DATA_DELIVERY_LOWPRI) |
3325 MSI_DATA_VECTOR(cfg->vector);
3331 static int set_msi_irq_affinity(unsigned int irq, const struct cpumask *mask)
3333 struct irq_desc *desc = irq_to_desc(irq);
3334 struct irq_cfg *cfg;
3338 dest = set_desc_affinity(desc, mask);
3339 if (dest == BAD_APICID)
3342 cfg = desc->chip_data;
3344 read_msi_msg_desc(desc, &msg);
3346 msg.data &= ~MSI_DATA_VECTOR_MASK;
3347 msg.data |= MSI_DATA_VECTOR(cfg->vector);
3348 msg.address_lo &= ~MSI_ADDR_DEST_ID_MASK;
3349 msg.address_lo |= MSI_ADDR_DEST_ID(dest);
3351 write_msi_msg_desc(desc, &msg);
3355 #ifdef CONFIG_INTR_REMAP
3357 * Migrate the MSI irq to another cpumask. This migration is
3358 * done in the process context using interrupt-remapping hardware.
3361 ir_set_msi_irq_affinity(unsigned int irq, const struct cpumask *mask)
3363 struct irq_desc *desc = irq_to_desc(irq);
3364 struct irq_cfg *cfg = desc->chip_data;
3368 if (get_irte(irq, &irte))
3371 dest = set_desc_affinity(desc, mask);
3372 if (dest == BAD_APICID)
3375 irte.vector = cfg->vector;
3376 irte.dest_id = IRTE_DEST(dest);
3379 * atomically update the IRTE with the new destination and vector.
3381 modify_irte(irq, &irte);
3384 * After this point, all the interrupts will start arriving
3385 * at the new destination. So, time to cleanup the previous
3386 * vector allocation.
3388 if (cfg->move_in_progress)
3389 send_cleanup_vector(cfg);
3395 #endif /* CONFIG_SMP */
3398 * IRQ Chip for MSI PCI/PCI-X/PCI-Express Devices,
3399 * which implement the MSI or MSI-X Capability Structure.
3401 static struct irq_chip msi_chip = {
3403 .unmask = unmask_msi_irq,
3404 .mask = mask_msi_irq,
3405 .ack = ack_apic_edge,
3407 .set_affinity = set_msi_irq_affinity,
3409 .retrigger = ioapic_retrigger_irq,
3412 static struct irq_chip msi_ir_chip = {
3413 .name = "IR-PCI-MSI",
3414 .unmask = unmask_msi_irq,
3415 .mask = mask_msi_irq,
3416 #ifdef CONFIG_INTR_REMAP
3417 .ack = ir_ack_apic_edge,
3419 .set_affinity = ir_set_msi_irq_affinity,
3422 .retrigger = ioapic_retrigger_irq,
3426 * Map the PCI dev to the corresponding remapping hardware unit
3427 * and allocate 'nvec' consecutive interrupt-remapping table entries
3430 static int msi_alloc_irte(struct pci_dev *dev, int irq, int nvec)
3432 struct intel_iommu *iommu;
3435 iommu = map_dev_to_ir(dev);
3438 "Unable to map PCI %s to iommu\n", pci_name(dev));
3442 index = alloc_irte(iommu, irq, nvec);
3445 "Unable to allocate %d IRTE for PCI %s\n", nvec,
3452 static int setup_msi_irq(struct pci_dev *dev, struct msi_desc *msidesc, int irq)
3457 ret = msi_compose_msg(dev, irq, &msg);
3461 set_irq_msi(irq, msidesc);
3462 write_msi_msg(irq, &msg);
3464 if (irq_remapped(irq)) {
3465 struct irq_desc *desc = irq_to_desc(irq);
3467 * irq migration in process context
3469 desc->status |= IRQ_MOVE_PCNTXT;
3470 set_irq_chip_and_handler_name(irq, &msi_ir_chip, handle_edge_irq, "edge");
3472 set_irq_chip_and_handler_name(irq, &msi_chip, handle_edge_irq, "edge");
3474 dev_printk(KERN_DEBUG, &dev->dev, "irq %d for MSI/MSI-X\n", irq);
3479 int arch_setup_msi_irqs(struct pci_dev *dev, int nvec, int type)
3482 int ret, sub_handle;
3483 struct msi_desc *msidesc;
3484 unsigned int irq_want;
3485 struct intel_iommu *iommu = NULL;
3489 /* x86 doesn't support multiple MSI yet */
3490 if (type == PCI_CAP_ID_MSI && nvec > 1)
3493 node = dev_to_node(&dev->dev);
3494 irq_want = nr_irqs_gsi;
3496 list_for_each_entry(msidesc, &dev->msi_list, list) {
3497 irq = create_irq_nr(irq_want, node);
3501 if (!intr_remapping_enabled)
3506 * allocate the consecutive block of IRTE's
3509 index = msi_alloc_irte(dev, irq, nvec);
3515 iommu = map_dev_to_ir(dev);
3521 * setup the mapping between the irq and the IRTE
3522 * base index, the sub_handle pointing to the
3523 * appropriate interrupt remap table entry.
3525 set_irte_irq(irq, iommu, index, sub_handle);
3528 ret = setup_msi_irq(dev, msidesc, irq);
3540 void arch_teardown_msi_irq(unsigned int irq)
3545 #if defined (CONFIG_DMAR) || defined (CONFIG_INTR_REMAP)
3547 static int dmar_msi_set_affinity(unsigned int irq, const struct cpumask *mask)
3549 struct irq_desc *desc = irq_to_desc(irq);
3550 struct irq_cfg *cfg;
3554 dest = set_desc_affinity(desc, mask);
3555 if (dest == BAD_APICID)
3558 cfg = desc->chip_data;
3560 dmar_msi_read(irq, &msg);
3562 msg.data &= ~MSI_DATA_VECTOR_MASK;
3563 msg.data |= MSI_DATA_VECTOR(cfg->vector);
3564 msg.address_lo &= ~MSI_ADDR_DEST_ID_MASK;
3565 msg.address_lo |= MSI_ADDR_DEST_ID(dest);
3567 dmar_msi_write(irq, &msg);
3572 #endif /* CONFIG_SMP */
3574 static struct irq_chip dmar_msi_type = {
3576 .unmask = dmar_msi_unmask,
3577 .mask = dmar_msi_mask,
3578 .ack = ack_apic_edge,
3580 .set_affinity = dmar_msi_set_affinity,
3582 .retrigger = ioapic_retrigger_irq,
3585 int arch_setup_dmar_msi(unsigned int irq)
3590 ret = msi_compose_msg(NULL, irq, &msg);
3593 dmar_msi_write(irq, &msg);
3594 set_irq_chip_and_handler_name(irq, &dmar_msi_type, handle_edge_irq,
3600 #ifdef CONFIG_HPET_TIMER
3603 static int hpet_msi_set_affinity(unsigned int irq, const struct cpumask *mask)
3605 struct irq_desc *desc = irq_to_desc(irq);
3606 struct irq_cfg *cfg;
3610 dest = set_desc_affinity(desc, mask);
3611 if (dest == BAD_APICID)
3614 cfg = desc->chip_data;
3616 hpet_msi_read(irq, &msg);
3618 msg.data &= ~MSI_DATA_VECTOR_MASK;
3619 msg.data |= MSI_DATA_VECTOR(cfg->vector);
3620 msg.address_lo &= ~MSI_ADDR_DEST_ID_MASK;
3621 msg.address_lo |= MSI_ADDR_DEST_ID(dest);
3623 hpet_msi_write(irq, &msg);
3628 #endif /* CONFIG_SMP */
3630 static struct irq_chip hpet_msi_type = {
3632 .unmask = hpet_msi_unmask,
3633 .mask = hpet_msi_mask,
3634 .ack = ack_apic_edge,
3636 .set_affinity = hpet_msi_set_affinity,
3638 .retrigger = ioapic_retrigger_irq,
3641 int arch_setup_hpet_msi(unsigned int irq)
3645 struct irq_desc *desc = irq_to_desc(irq);
3647 ret = msi_compose_msg(NULL, irq, &msg);
3651 hpet_msi_write(irq, &msg);
3652 desc->status |= IRQ_MOVE_PCNTXT;
3653 set_irq_chip_and_handler_name(irq, &hpet_msi_type, handle_edge_irq,
3660 #endif /* CONFIG_PCI_MSI */
3662 * Hypertransport interrupt support
3664 #ifdef CONFIG_HT_IRQ
3668 static void target_ht_irq(unsigned int irq, unsigned int dest, u8 vector)
3670 struct ht_irq_msg msg;
3671 fetch_ht_irq_msg(irq, &msg);
3673 msg.address_lo &= ~(HT_IRQ_LOW_VECTOR_MASK | HT_IRQ_LOW_DEST_ID_MASK);
3674 msg.address_hi &= ~(HT_IRQ_HIGH_DEST_ID_MASK);
3676 msg.address_lo |= HT_IRQ_LOW_VECTOR(vector) | HT_IRQ_LOW_DEST_ID(dest);
3677 msg.address_hi |= HT_IRQ_HIGH_DEST_ID(dest);
3679 write_ht_irq_msg(irq, &msg);
3682 static int set_ht_irq_affinity(unsigned int irq, const struct cpumask *mask)
3684 struct irq_desc *desc = irq_to_desc(irq);
3685 struct irq_cfg *cfg;
3688 dest = set_desc_affinity(desc, mask);
3689 if (dest == BAD_APICID)
3692 cfg = desc->chip_data;
3694 target_ht_irq(irq, dest, cfg->vector);
3701 static struct irq_chip ht_irq_chip = {
3703 .mask = mask_ht_irq,
3704 .unmask = unmask_ht_irq,
3705 .ack = ack_apic_edge,
3707 .set_affinity = set_ht_irq_affinity,
3709 .retrigger = ioapic_retrigger_irq,
3712 int arch_setup_ht_irq(unsigned int irq, struct pci_dev *dev)
3714 struct irq_cfg *cfg;
3721 err = assign_irq_vector(irq, cfg, apic->target_cpus());
3723 struct ht_irq_msg msg;
3726 dest = apic->cpu_mask_to_apicid_and(cfg->domain,
3727 apic->target_cpus());
3729 msg.address_hi = HT_IRQ_HIGH_DEST_ID(dest);
3733 HT_IRQ_LOW_DEST_ID(dest) |
3734 HT_IRQ_LOW_VECTOR(cfg->vector) |
3735 ((apic->irq_dest_mode == 0) ?
3736 HT_IRQ_LOW_DM_PHYSICAL :
3737 HT_IRQ_LOW_DM_LOGICAL) |
3738 HT_IRQ_LOW_RQEOI_EDGE |
3739 ((apic->irq_delivery_mode != dest_LowestPrio) ?
3740 HT_IRQ_LOW_MT_FIXED :
3741 HT_IRQ_LOW_MT_ARBITRATED) |
3742 HT_IRQ_LOW_IRQ_MASKED;
3744 write_ht_irq_msg(irq, &msg);
3746 set_irq_chip_and_handler_name(irq, &ht_irq_chip,
3747 handle_edge_irq, "edge");
3749 dev_printk(KERN_DEBUG, &dev->dev, "irq %d for HT\n", irq);
3753 #endif /* CONFIG_HT_IRQ */
3755 #ifdef CONFIG_X86_UV
3757 * Re-target the irq to the specified CPU and enable the specified MMR located
3758 * on the specified blade to allow the sending of MSIs to the specified CPU.
3760 int arch_enable_uv_irq(char *irq_name, unsigned int irq, int cpu, int mmr_blade,
3761 unsigned long mmr_offset)
3763 const struct cpumask *eligible_cpu = cpumask_of(cpu);
3764 struct irq_cfg *cfg;
3766 unsigned long mmr_value;
3767 struct uv_IO_APIC_route_entry *entry;
3768 unsigned long flags;
3771 BUILD_BUG_ON(sizeof(struct uv_IO_APIC_route_entry) != sizeof(unsigned long));
3775 err = assign_irq_vector(irq, cfg, eligible_cpu);
3779 spin_lock_irqsave(&vector_lock, flags);
3780 set_irq_chip_and_handler_name(irq, &uv_irq_chip, handle_percpu_irq,
3782 spin_unlock_irqrestore(&vector_lock, flags);
3785 entry = (struct uv_IO_APIC_route_entry *)&mmr_value;
3786 entry->vector = cfg->vector;
3787 entry->delivery_mode = apic->irq_delivery_mode;
3788 entry->dest_mode = apic->irq_dest_mode;
3789 entry->polarity = 0;
3792 entry->dest = apic->cpu_mask_to_apicid(eligible_cpu);
3794 mmr_pnode = uv_blade_to_pnode(mmr_blade);
3795 uv_write_global_mmr64(mmr_pnode, mmr_offset, mmr_value);
3801 * Disable the specified MMR located on the specified blade so that MSIs are
3802 * longer allowed to be sent.
3804 void arch_disable_uv_irq(int mmr_blade, unsigned long mmr_offset)
3806 unsigned long mmr_value;
3807 struct uv_IO_APIC_route_entry *entry;
3810 BUILD_BUG_ON(sizeof(struct uv_IO_APIC_route_entry) != sizeof(unsigned long));
3813 entry = (struct uv_IO_APIC_route_entry *)&mmr_value;
3816 mmr_pnode = uv_blade_to_pnode(mmr_blade);
3817 uv_write_global_mmr64(mmr_pnode, mmr_offset, mmr_value);
3819 #endif /* CONFIG_X86_64 */
3821 int __init io_apic_get_redir_entries (int ioapic)
3823 union IO_APIC_reg_01 reg_01;
3824 unsigned long flags;
3826 spin_lock_irqsave(&ioapic_lock, flags);
3827 reg_01.raw = io_apic_read(ioapic, 1);
3828 spin_unlock_irqrestore(&ioapic_lock, flags);
3830 return reg_01.bits.entries;
3833 void __init probe_nr_irqs_gsi(void)
3837 nr = acpi_probe_gsi();
3838 if (nr > nr_irqs_gsi) {
3841 /* for acpi=off or acpi is not compiled in */
3845 for (idx = 0; idx < nr_ioapics; idx++)
3846 nr += io_apic_get_redir_entries(idx) + 1;
3848 if (nr > nr_irqs_gsi)
3852 printk(KERN_DEBUG "nr_irqs_gsi: %d\n", nr_irqs_gsi);
3855 #ifdef CONFIG_SPARSE_IRQ
3856 int __init arch_probe_nr_irqs(void)
3860 if (nr_irqs > (NR_VECTORS * nr_cpu_ids))
3861 nr_irqs = NR_VECTORS * nr_cpu_ids;
3863 nr = nr_irqs_gsi + 8 * nr_cpu_ids;
3864 #if defined(CONFIG_PCI_MSI) || defined(CONFIG_HT_IRQ)
3866 * for MSI and HT dyn irq
3868 nr += nr_irqs_gsi * 16;
3877 static int __io_apic_set_pci_routing(struct device *dev, int irq,
3878 struct io_apic_irq_attr *irq_attr)
3880 struct irq_desc *desc;
3881 struct irq_cfg *cfg;
3884 int trigger, polarity;
3886 ioapic = irq_attr->ioapic;
3887 if (!IO_APIC_IRQ(irq)) {
3888 apic_printk(APIC_QUIET,KERN_ERR "IOAPIC[%d]: Invalid reference to IRQ 0\n",
3894 node = dev_to_node(dev);
3896 node = cpu_to_node(boot_cpu_id);
3898 desc = irq_to_desc_alloc_node(irq, node);
3900 printk(KERN_INFO "can not get irq_desc %d\n", irq);
3904 pin = irq_attr->ioapic_pin;
3905 trigger = irq_attr->trigger;
3906 polarity = irq_attr->polarity;
3909 * IRQs < 16 are already in the irq_2_pin[] map
3911 if (irq >= NR_IRQS_LEGACY) {
3912 cfg = desc->chip_data;
3913 add_pin_to_irq_node(cfg, node, ioapic, pin);
3916 setup_IO_APIC_irq(ioapic, pin, irq, desc, trigger, polarity);
3921 int io_apic_set_pci_routing(struct device *dev, int irq,
3922 struct io_apic_irq_attr *irq_attr)
3926 * Avoid pin reprogramming. PRTs typically include entries
3927 * with redundant pin->gsi mappings (but unique PCI devices);
3928 * we only program the IOAPIC on the first.
3930 ioapic = irq_attr->ioapic;
3931 pin = irq_attr->ioapic_pin;
3932 if (test_bit(pin, mp_ioapic_routing[ioapic].pin_programmed)) {
3933 pr_debug("Pin %d-%d already programmed\n",
3934 mp_ioapics[ioapic].apicid, pin);
3937 set_bit(pin, mp_ioapic_routing[ioapic].pin_programmed);
3939 return __io_apic_set_pci_routing(dev, irq, irq_attr);
3942 /* --------------------------------------------------------------------------
3943 ACPI-based IOAPIC Configuration
3944 -------------------------------------------------------------------------- */
3948 #ifdef CONFIG_X86_32
3949 int __init io_apic_get_unique_id(int ioapic, int apic_id)
3951 union IO_APIC_reg_00 reg_00;
3952 static physid_mask_t apic_id_map = PHYSID_MASK_NONE;
3954 unsigned long flags;
3958 * The P4 platform supports up to 256 APIC IDs on two separate APIC
3959 * buses (one for LAPICs, one for IOAPICs), where predecessors only
3960 * supports up to 16 on one shared APIC bus.
3962 * TBD: Expand LAPIC/IOAPIC support on P4-class systems to take full
3963 * advantage of new APIC bus architecture.
3966 if (physids_empty(apic_id_map))
3967 apic_id_map = apic->ioapic_phys_id_map(phys_cpu_present_map);
3969 spin_lock_irqsave(&ioapic_lock, flags);
3970 reg_00.raw = io_apic_read(ioapic, 0);
3971 spin_unlock_irqrestore(&ioapic_lock, flags);
3973 if (apic_id >= get_physical_broadcast()) {
3974 printk(KERN_WARNING "IOAPIC[%d]: Invalid apic_id %d, trying "
3975 "%d\n", ioapic, apic_id, reg_00.bits.ID);
3976 apic_id = reg_00.bits.ID;
3980 * Every APIC in a system must have a unique ID or we get lots of nice
3981 * 'stuck on smp_invalidate_needed IPI wait' messages.
3983 if (apic->check_apicid_used(apic_id_map, apic_id)) {
3985 for (i = 0; i < get_physical_broadcast(); i++) {
3986 if (!apic->check_apicid_used(apic_id_map, i))
3990 if (i == get_physical_broadcast())
3991 panic("Max apic_id exceeded!\n");
3993 printk(KERN_WARNING "IOAPIC[%d]: apic_id %d already used, "
3994 "trying %d\n", ioapic, apic_id, i);
3999 tmp = apic->apicid_to_cpu_present(apic_id);
4000 physids_or(apic_id_map, apic_id_map, tmp);
4002 if (reg_00.bits.ID != apic_id) {
4003 reg_00.bits.ID = apic_id;
4005 spin_lock_irqsave(&ioapic_lock, flags);
4006 io_apic_write(ioapic, 0, reg_00.raw);
4007 reg_00.raw = io_apic_read(ioapic, 0);
4008 spin_unlock_irqrestore(&ioapic_lock, flags);
4011 if (reg_00.bits.ID != apic_id) {
4012 printk("IOAPIC[%d]: Unable to change apic_id!\n", ioapic);
4017 apic_printk(APIC_VERBOSE, KERN_INFO
4018 "IOAPIC[%d]: Assigned apic_id %d\n", ioapic, apic_id);
4024 int __init io_apic_get_version(int ioapic)
4026 union IO_APIC_reg_01 reg_01;
4027 unsigned long flags;
4029 spin_lock_irqsave(&ioapic_lock, flags);
4030 reg_01.raw = io_apic_read(ioapic, 1);
4031 spin_unlock_irqrestore(&ioapic_lock, flags);
4033 return reg_01.bits.version;
4036 int acpi_get_override_irq(int bus_irq, int *trigger, int *polarity)
4040 if (skip_ioapic_setup)
4043 for (i = 0; i < mp_irq_entries; i++)
4044 if (mp_irqs[i].irqtype == mp_INT &&
4045 mp_irqs[i].srcbusirq == bus_irq)
4047 if (i >= mp_irq_entries)
4050 *trigger = irq_trigger(i);
4051 *polarity = irq_polarity(i);
4055 #endif /* CONFIG_ACPI */
4058 * This function currently is only a helper for the i386 smp boot process where
4059 * we need to reprogram the ioredtbls to cater for the cpus which have come online
4060 * so mask in all cases should simply be apic->target_cpus()
4063 void __init setup_ioapic_dest(void)
4065 int pin, ioapic = 0, irq, irq_entry;
4066 struct irq_desc *desc;
4067 const struct cpumask *mask;
4069 if (skip_ioapic_setup == 1)
4073 if (!acpi_disabled && acpi_ioapic) {
4074 ioapic = mp_find_ioapic(0);
4080 for (pin = 0; pin < nr_ioapic_registers[ioapic]; pin++) {
4081 irq_entry = find_irq_entry(ioapic, pin, mp_INT);
4082 if (irq_entry == -1)
4084 irq = pin_2_irq(irq_entry, ioapic, pin);
4086 desc = irq_to_desc(irq);
4089 * Honour affinities which have been set in early boot
4092 (IRQ_NO_BALANCING | IRQ_AFFINITY_SET))
4093 mask = desc->affinity;
4095 mask = apic->target_cpus();
4097 if (intr_remapping_enabled)
4098 set_ir_ioapic_affinity_irq_desc(desc, mask);
4100 set_ioapic_affinity_irq_desc(desc, mask);
4106 #define IOAPIC_RESOURCE_NAME_SIZE 11
4108 static struct resource *ioapic_resources;
4110 static struct resource * __init ioapic_setup_resources(void)
4113 struct resource *res;
4117 if (nr_ioapics <= 0)
4120 n = IOAPIC_RESOURCE_NAME_SIZE + sizeof(struct resource);
4123 mem = alloc_bootmem(n);
4127 mem += sizeof(struct resource) * nr_ioapics;
4129 for (i = 0; i < nr_ioapics; i++) {
4131 res[i].flags = IORESOURCE_MEM | IORESOURCE_BUSY;
4132 sprintf(mem, "IOAPIC %u", i);
4133 mem += IOAPIC_RESOURCE_NAME_SIZE;
4137 ioapic_resources = res;
4142 void __init ioapic_init_mappings(void)
4144 unsigned long ioapic_phys, idx = FIX_IO_APIC_BASE_0;
4145 struct resource *ioapic_res;
4148 ioapic_res = ioapic_setup_resources();
4149 for (i = 0; i < nr_ioapics; i++) {
4150 if (smp_found_config) {
4151 ioapic_phys = mp_ioapics[i].apicaddr;
4152 #ifdef CONFIG_X86_32
4155 "WARNING: bogus zero IO-APIC "
4156 "address found in MPTABLE, "
4157 "disabling IO/APIC support!\n");
4158 smp_found_config = 0;
4159 skip_ioapic_setup = 1;
4160 goto fake_ioapic_page;
4164 #ifdef CONFIG_X86_32
4167 ioapic_phys = (unsigned long)
4168 alloc_bootmem_pages(PAGE_SIZE);
4169 ioapic_phys = __pa(ioapic_phys);
4171 set_fixmap_nocache(idx, ioapic_phys);
4172 apic_printk(APIC_VERBOSE,
4173 "mapped IOAPIC to %08lx (%08lx)\n",
4174 __fix_to_virt(idx), ioapic_phys);
4177 if (ioapic_res != NULL) {
4178 ioapic_res->start = ioapic_phys;
4179 ioapic_res->end = ioapic_phys + (4 * 1024) - 1;
4185 static int __init ioapic_insert_resources(void)
4188 struct resource *r = ioapic_resources;
4191 if (nr_ioapics > 0) {
4193 "IO APIC resources couldn't be allocated.\n");
4199 for (i = 0; i < nr_ioapics; i++) {
4200 insert_resource(&iomem_resource, r);
4207 /* Insert the IO APIC resources after PCI initialization has occured to handle
4208 * IO APICS that are mapped in on a BAR in PCI space. */
4209 late_initcall(ioapic_insert_resources);