2 * Intel IO-APIC support for multi-Pentium hosts.
4 * Copyright (C) 1997, 1998, 1999, 2000, 2009 Ingo Molnar, Hajnalka Szabo
6 * Many thanks to Stig Venaas for trying out countless experimental
7 * patches and reporting/debugging problems patiently!
9 * (c) 1999, Multiple IO-APIC support, developed by
10 * Ken-ichi Yaku <yaku@css1.kbnes.nec.co.jp> and
11 * Hidemi Kishimoto <kisimoto@css1.kbnes.nec.co.jp>,
12 * further tested and cleaned up by Zach Brown <zab@redhat.com>
13 * and Ingo Molnar <mingo@redhat.com>
16 * Maciej W. Rozycki : Bits for genuine 82489DX APICs;
17 * thanks to Eric Gilmore
19 * for testing these extensively
20 * Paul Diefenbaugh : Added full ACPI support
24 #include <linux/interrupt.h>
25 #include <linux/init.h>
26 #include <linux/delay.h>
27 #include <linux/sched.h>
28 #include <linux/pci.h>
29 #include <linux/mc146818rtc.h>
30 #include <linux/compiler.h>
31 #include <linux/acpi.h>
32 #include <linux/module.h>
33 #include <linux/sysdev.h>
34 #include <linux/msi.h>
35 #include <linux/htirq.h>
36 #include <linux/freezer.h>
37 #include <linux/kthread.h>
38 #include <linux/jiffies.h> /* time_after() */
39 #include <linux/slab.h>
41 #include <acpi/acpi_bus.h>
43 #include <linux/bootmem.h>
44 #include <linux/dmar.h>
45 #include <linux/hpet.h>
52 #include <asm/proto.h>
55 #include <asm/timer.h>
56 #include <asm/i8259.h>
58 #include <asm/msidef.h>
59 #include <asm/hypertransport.h>
60 #include <asm/setup.h>
61 #include <asm/irq_remapping.h>
63 #include <asm/hw_irq.h>
67 #define __apicdebuginit(type) static type __init
68 #define for_each_irq_pin(entry, head) \
69 for (entry = head; entry; entry = entry->next)
72 * Is the SiS APIC rmw bug present ?
73 * -1 = don't know, 0 = no, 1 = yes
75 int sis_apic_bug = -1;
77 static DEFINE_RAW_SPINLOCK(ioapic_lock);
78 static DEFINE_RAW_SPINLOCK(vector_lock);
81 * # of IRQ routing registers
83 int nr_ioapic_registers[MAX_IO_APICS];
85 /* I/O APIC entries */
86 struct mpc_ioapic mp_ioapics[MAX_IO_APICS];
89 /* IO APIC gsi routing info */
90 struct mp_ioapic_gsi mp_gsi_routing[MAX_IO_APICS];
92 /* The one past the highest gsi number used */
95 /* MP IRQ source entries */
96 struct mpc_intsrc mp_irqs[MAX_IRQ_SOURCES];
98 /* # of MP IRQ source entries */
102 static int nr_irqs_gsi = NR_IRQS_LEGACY;
104 #if defined (CONFIG_MCA) || defined (CONFIG_EISA)
105 int mp_bus_id_to_type[MAX_MP_BUSSES];
108 DECLARE_BITMAP(mp_bus_not_pci, MAX_MP_BUSSES);
110 int skip_ioapic_setup;
112 void arch_disable_smp_support(void)
116 noioapicreroute = -1;
118 skip_ioapic_setup = 1;
121 static int __init parse_noapic(char *str)
123 /* disable IO-APIC */
124 arch_disable_smp_support();
127 early_param("noapic", parse_noapic);
129 /* Will be called in mpparse/acpi/sfi codes for saving IRQ info */
130 void mp_save_irq(struct mpc_intsrc *m)
134 apic_printk(APIC_VERBOSE, "Int: type %d, pol %d, trig %d, bus %02x,"
135 " IRQ %02x, APIC ID %x, APIC INT %02x\n",
136 m->irqtype, m->irqflag & 3, (m->irqflag >> 2) & 3, m->srcbus,
137 m->srcbusirq, m->dstapic, m->dstirq);
139 for (i = 0; i < mp_irq_entries; i++) {
140 if (!memcmp(&mp_irqs[i], m, sizeof(*m)))
144 memcpy(&mp_irqs[mp_irq_entries], m, sizeof(*m));
145 if (++mp_irq_entries == MAX_IRQ_SOURCES)
146 panic("Max # of irq sources exceeded!!\n");
149 struct irq_pin_list {
151 struct irq_pin_list *next;
154 static struct irq_pin_list *alloc_irq_pin_list(int node)
156 return kzalloc_node(sizeof(struct irq_pin_list), GFP_KERNEL, node);
160 /* irq_cfg is indexed by the sum of all RTEs in all I/O APICs. */
161 #ifdef CONFIG_SPARSE_IRQ
162 static struct irq_cfg irq_cfgx[NR_IRQS_LEGACY];
164 static struct irq_cfg irq_cfgx[NR_IRQS];
167 int __init arch_early_irq_init(void)
172 if (!legacy_pic->nr_legacy_irqs) {
178 count = ARRAY_SIZE(irq_cfgx);
179 node = cpu_to_node(0);
181 /* Make sure the legacy interrupts are marked in the bitmap */
182 irq_reserve_irqs(0, legacy_pic->nr_legacy_irqs);
184 for (i = 0; i < count; i++) {
185 set_irq_chip_data(i, &cfg[i]);
186 zalloc_cpumask_var_node(&cfg[i].domain, GFP_KERNEL, node);
187 zalloc_cpumask_var_node(&cfg[i].old_domain, GFP_KERNEL, node);
189 * For legacy IRQ's, start with assigning irq0 to irq15 to
190 * IRQ0_VECTOR to IRQ15_VECTOR on cpu 0.
192 if (i < legacy_pic->nr_legacy_irqs) {
193 cfg[i].vector = IRQ0_VECTOR + i;
194 cpumask_set_cpu(0, cfg[i].domain);
201 #ifdef CONFIG_SPARSE_IRQ
202 static struct irq_cfg *irq_cfg(unsigned int irq)
204 return get_irq_chip_data(irq);
207 static struct irq_cfg *alloc_irq_cfg(unsigned int irq, int node)
211 cfg = kzalloc_node(sizeof(*cfg), GFP_KERNEL, node);
214 if (!zalloc_cpumask_var_node(&cfg->domain, GFP_KERNEL, node))
216 if (!zalloc_cpumask_var_node(&cfg->old_domain, GFP_KERNEL, node))
220 free_cpumask_var(cfg->domain);
226 static void free_irq_cfg(unsigned int at, struct irq_cfg *cfg)
230 set_irq_chip_data(at, NULL);
231 free_cpumask_var(cfg->domain);
232 free_cpumask_var(cfg->old_domain);
238 struct irq_cfg *irq_cfg(unsigned int irq)
240 return irq < nr_irqs ? irq_cfgx + irq : NULL;
243 static struct irq_cfg *alloc_irq_cfg(unsigned int irq, int node)
245 return irq_cfgx + irq;
248 static inline void free_irq_cfg(unsigned int at, struct irq_cfg *cfg) { }
252 static struct irq_cfg *alloc_irq_and_cfg_at(unsigned int at, int node)
254 int res = irq_alloc_desc_at(at, node);
260 cfg = get_irq_chip_data(at);
265 cfg = alloc_irq_cfg(at, node);
267 set_irq_chip_data(at, cfg);
273 static int alloc_irq_from(unsigned int from, int node)
275 return irq_alloc_desc_from(from, node);
278 static void free_irq_at(unsigned int at, struct irq_cfg *cfg)
280 free_irq_cfg(at, cfg);
286 unsigned int unused[3];
288 unsigned int unused2[11];
292 static __attribute_const__ struct io_apic __iomem *io_apic_base(int idx)
294 return (void __iomem *) __fix_to_virt(FIX_IO_APIC_BASE_0 + idx)
295 + (mp_ioapics[idx].apicaddr & ~PAGE_MASK);
298 static inline void io_apic_eoi(unsigned int apic, unsigned int vector)
300 struct io_apic __iomem *io_apic = io_apic_base(apic);
301 writel(vector, &io_apic->eoi);
304 static inline unsigned int io_apic_read(unsigned int apic, unsigned int reg)
306 struct io_apic __iomem *io_apic = io_apic_base(apic);
307 writel(reg, &io_apic->index);
308 return readl(&io_apic->data);
311 static inline void io_apic_write(unsigned int apic, unsigned int reg, unsigned int value)
313 struct io_apic __iomem *io_apic = io_apic_base(apic);
314 writel(reg, &io_apic->index);
315 writel(value, &io_apic->data);
319 * Re-write a value: to be used for read-modify-write
320 * cycles where the read already set up the index register.
322 * Older SiS APIC requires we rewrite the index register
324 static inline void io_apic_modify(unsigned int apic, unsigned int reg, unsigned int value)
326 struct io_apic __iomem *io_apic = io_apic_base(apic);
329 writel(reg, &io_apic->index);
330 writel(value, &io_apic->data);
333 static bool io_apic_level_ack_pending(struct irq_cfg *cfg)
335 struct irq_pin_list *entry;
338 raw_spin_lock_irqsave(&ioapic_lock, flags);
339 for_each_irq_pin(entry, cfg->irq_2_pin) {
344 reg = io_apic_read(entry->apic, 0x10 + pin*2);
345 /* Is the remote IRR bit set? */
346 if (reg & IO_APIC_REDIR_REMOTE_IRR) {
347 raw_spin_unlock_irqrestore(&ioapic_lock, flags);
351 raw_spin_unlock_irqrestore(&ioapic_lock, flags);
357 struct { u32 w1, w2; };
358 struct IO_APIC_route_entry entry;
361 static struct IO_APIC_route_entry ioapic_read_entry(int apic, int pin)
363 union entry_union eu;
365 raw_spin_lock_irqsave(&ioapic_lock, flags);
366 eu.w1 = io_apic_read(apic, 0x10 + 2 * pin);
367 eu.w2 = io_apic_read(apic, 0x11 + 2 * pin);
368 raw_spin_unlock_irqrestore(&ioapic_lock, flags);
373 * When we write a new IO APIC routing entry, we need to write the high
374 * word first! If the mask bit in the low word is clear, we will enable
375 * the interrupt, and we need to make sure the entry is fully populated
376 * before that happens.
379 __ioapic_write_entry(int apic, int pin, struct IO_APIC_route_entry e)
381 union entry_union eu = {{0, 0}};
384 io_apic_write(apic, 0x11 + 2*pin, eu.w2);
385 io_apic_write(apic, 0x10 + 2*pin, eu.w1);
388 static void ioapic_write_entry(int apic, int pin, struct IO_APIC_route_entry e)
391 raw_spin_lock_irqsave(&ioapic_lock, flags);
392 __ioapic_write_entry(apic, pin, e);
393 raw_spin_unlock_irqrestore(&ioapic_lock, flags);
397 * When we mask an IO APIC routing entry, we need to write the low
398 * word first, in order to set the mask bit before we change the
401 static void ioapic_mask_entry(int apic, int pin)
404 union entry_union eu = { .entry.mask = 1 };
406 raw_spin_lock_irqsave(&ioapic_lock, flags);
407 io_apic_write(apic, 0x10 + 2*pin, eu.w1);
408 io_apic_write(apic, 0x11 + 2*pin, eu.w2);
409 raw_spin_unlock_irqrestore(&ioapic_lock, flags);
413 * The common case is 1:1 IRQ<->pin mappings. Sometimes there are
414 * shared ISA-space IRQs, so we have to support them. We are super
415 * fast in the common case, and fast for shared ISA-space IRQs.
418 __add_pin_to_irq_node(struct irq_cfg *cfg, int node, int apic, int pin)
420 struct irq_pin_list **last, *entry;
422 /* don't allow duplicates */
423 last = &cfg->irq_2_pin;
424 for_each_irq_pin(entry, cfg->irq_2_pin) {
425 if (entry->apic == apic && entry->pin == pin)
430 entry = alloc_irq_pin_list(node);
432 printk(KERN_ERR "can not alloc irq_pin_list (%d,%d,%d)\n",
443 static void add_pin_to_irq_node(struct irq_cfg *cfg, int node, int apic, int pin)
445 if (__add_pin_to_irq_node(cfg, node, apic, pin))
446 panic("IO-APIC: failed to add irq-pin. Can not proceed\n");
450 * Reroute an IRQ to a different pin.
452 static void __init replace_pin_at_irq_node(struct irq_cfg *cfg, int node,
453 int oldapic, int oldpin,
454 int newapic, int newpin)
456 struct irq_pin_list *entry;
458 for_each_irq_pin(entry, cfg->irq_2_pin) {
459 if (entry->apic == oldapic && entry->pin == oldpin) {
460 entry->apic = newapic;
462 /* every one is different, right? */
467 /* old apic/pin didn't exist, so just add new ones */
468 add_pin_to_irq_node(cfg, node, newapic, newpin);
471 static void __io_apic_modify_irq(struct irq_pin_list *entry,
472 int mask_and, int mask_or,
473 void (*final)(struct irq_pin_list *entry))
475 unsigned int reg, pin;
478 reg = io_apic_read(entry->apic, 0x10 + pin * 2);
481 io_apic_modify(entry->apic, 0x10 + pin * 2, reg);
486 static void io_apic_modify_irq(struct irq_cfg *cfg,
487 int mask_and, int mask_or,
488 void (*final)(struct irq_pin_list *entry))
490 struct irq_pin_list *entry;
492 for_each_irq_pin(entry, cfg->irq_2_pin)
493 __io_apic_modify_irq(entry, mask_and, mask_or, final);
496 static void __mask_and_edge_IO_APIC_irq(struct irq_pin_list *entry)
498 __io_apic_modify_irq(entry, ~IO_APIC_REDIR_LEVEL_TRIGGER,
499 IO_APIC_REDIR_MASKED, NULL);
502 static void __unmask_and_level_IO_APIC_irq(struct irq_pin_list *entry)
504 __io_apic_modify_irq(entry, ~IO_APIC_REDIR_MASKED,
505 IO_APIC_REDIR_LEVEL_TRIGGER, NULL);
508 static void io_apic_sync(struct irq_pin_list *entry)
511 * Synchronize the IO-APIC and the CPU by doing
512 * a dummy read from the IO-APIC
514 struct io_apic __iomem *io_apic;
515 io_apic = io_apic_base(entry->apic);
516 readl(&io_apic->data);
519 static void mask_ioapic(struct irq_cfg *cfg)
523 raw_spin_lock_irqsave(&ioapic_lock, flags);
524 io_apic_modify_irq(cfg, ~0, IO_APIC_REDIR_MASKED, &io_apic_sync);
525 raw_spin_unlock_irqrestore(&ioapic_lock, flags);
528 static void mask_ioapic_irq(struct irq_data *data)
530 mask_ioapic(data->chip_data);
533 static void __unmask_ioapic(struct irq_cfg *cfg)
535 io_apic_modify_irq(cfg, ~IO_APIC_REDIR_MASKED, 0, NULL);
538 static void unmask_ioapic(struct irq_cfg *cfg)
542 raw_spin_lock_irqsave(&ioapic_lock, flags);
543 __unmask_ioapic(cfg);
544 raw_spin_unlock_irqrestore(&ioapic_lock, flags);
547 static void unmask_ioapic_irq(struct irq_data *data)
549 unmask_ioapic(data->chip_data);
552 static void clear_IO_APIC_pin(unsigned int apic, unsigned int pin)
554 struct IO_APIC_route_entry entry;
556 /* Check delivery_mode to be sure we're not clearing an SMI pin */
557 entry = ioapic_read_entry(apic, pin);
558 if (entry.delivery_mode == dest_SMI)
561 * Disable it in the IO-APIC irq-routing table:
563 ioapic_mask_entry(apic, pin);
566 static void clear_IO_APIC (void)
570 for (apic = 0; apic < nr_ioapics; apic++)
571 for (pin = 0; pin < nr_ioapic_registers[apic]; pin++)
572 clear_IO_APIC_pin(apic, pin);
577 * support for broken MP BIOSs, enables hand-redirection of PIRQ0-7 to
578 * specific CPU-side IRQs.
582 static int pirq_entries[MAX_PIRQS] = {
583 [0 ... MAX_PIRQS - 1] = -1
586 static int __init ioapic_pirq_setup(char *str)
589 int ints[MAX_PIRQS+1];
591 get_options(str, ARRAY_SIZE(ints), ints);
593 apic_printk(APIC_VERBOSE, KERN_INFO
594 "PIRQ redirection, working around broken MP-BIOS.\n");
596 if (ints[0] < MAX_PIRQS)
599 for (i = 0; i < max; i++) {
600 apic_printk(APIC_VERBOSE, KERN_DEBUG
601 "... PIRQ%d -> IRQ %d\n", i, ints[i+1]);
603 * PIRQs are mapped upside down, usually.
605 pirq_entries[MAX_PIRQS-i-1] = ints[i+1];
610 __setup("pirq=", ioapic_pirq_setup);
611 #endif /* CONFIG_X86_32 */
613 struct IO_APIC_route_entry **alloc_ioapic_entries(void)
616 struct IO_APIC_route_entry **ioapic_entries;
618 ioapic_entries = kzalloc(sizeof(*ioapic_entries) * nr_ioapics,
623 for (apic = 0; apic < nr_ioapics; apic++) {
624 ioapic_entries[apic] =
625 kzalloc(sizeof(struct IO_APIC_route_entry) *
626 nr_ioapic_registers[apic], GFP_KERNEL);
627 if (!ioapic_entries[apic])
631 return ioapic_entries;
635 kfree(ioapic_entries[apic]);
636 kfree(ioapic_entries);
642 * Saves all the IO-APIC RTE's
644 int save_IO_APIC_setup(struct IO_APIC_route_entry **ioapic_entries)
651 for (apic = 0; apic < nr_ioapics; apic++) {
652 if (!ioapic_entries[apic])
655 for (pin = 0; pin < nr_ioapic_registers[apic]; pin++)
656 ioapic_entries[apic][pin] =
657 ioapic_read_entry(apic, pin);
664 * Mask all IO APIC entries.
666 void mask_IO_APIC_setup(struct IO_APIC_route_entry **ioapic_entries)
673 for (apic = 0; apic < nr_ioapics; apic++) {
674 if (!ioapic_entries[apic])
677 for (pin = 0; pin < nr_ioapic_registers[apic]; pin++) {
678 struct IO_APIC_route_entry entry;
680 entry = ioapic_entries[apic][pin];
683 ioapic_write_entry(apic, pin, entry);
690 * Restore IO APIC entries which was saved in ioapic_entries.
692 int restore_IO_APIC_setup(struct IO_APIC_route_entry **ioapic_entries)
699 for (apic = 0; apic < nr_ioapics; apic++) {
700 if (!ioapic_entries[apic])
703 for (pin = 0; pin < nr_ioapic_registers[apic]; pin++)
704 ioapic_write_entry(apic, pin,
705 ioapic_entries[apic][pin]);
710 void free_ioapic_entries(struct IO_APIC_route_entry **ioapic_entries)
714 for (apic = 0; apic < nr_ioapics; apic++)
715 kfree(ioapic_entries[apic]);
717 kfree(ioapic_entries);
721 * Find the IRQ entry number of a certain pin.
723 static int find_irq_entry(int apic, int pin, int type)
727 for (i = 0; i < mp_irq_entries; i++)
728 if (mp_irqs[i].irqtype == type &&
729 (mp_irqs[i].dstapic == mp_ioapics[apic].apicid ||
730 mp_irqs[i].dstapic == MP_APIC_ALL) &&
731 mp_irqs[i].dstirq == pin)
738 * Find the pin to which IRQ[irq] (ISA) is connected
740 static int __init find_isa_irq_pin(int irq, int type)
744 for (i = 0; i < mp_irq_entries; i++) {
745 int lbus = mp_irqs[i].srcbus;
747 if (test_bit(lbus, mp_bus_not_pci) &&
748 (mp_irqs[i].irqtype == type) &&
749 (mp_irqs[i].srcbusirq == irq))
751 return mp_irqs[i].dstirq;
756 static int __init find_isa_irq_apic(int irq, int type)
760 for (i = 0; i < mp_irq_entries; i++) {
761 int lbus = mp_irqs[i].srcbus;
763 if (test_bit(lbus, mp_bus_not_pci) &&
764 (mp_irqs[i].irqtype == type) &&
765 (mp_irqs[i].srcbusirq == irq))
768 if (i < mp_irq_entries) {
770 for(apic = 0; apic < nr_ioapics; apic++) {
771 if (mp_ioapics[apic].apicid == mp_irqs[i].dstapic)
779 #if defined(CONFIG_EISA) || defined(CONFIG_MCA)
781 * EISA Edge/Level control register, ELCR
783 static int EISA_ELCR(unsigned int irq)
785 if (irq < legacy_pic->nr_legacy_irqs) {
786 unsigned int port = 0x4d0 + (irq >> 3);
787 return (inb(port) >> (irq & 7)) & 1;
789 apic_printk(APIC_VERBOSE, KERN_INFO
790 "Broken MPtable reports ISA irq %d\n", irq);
796 /* ISA interrupts are always polarity zero edge triggered,
797 * when listed as conforming in the MP table. */
799 #define default_ISA_trigger(idx) (0)
800 #define default_ISA_polarity(idx) (0)
802 /* EISA interrupts are always polarity zero and can be edge or level
803 * trigger depending on the ELCR value. If an interrupt is listed as
804 * EISA conforming in the MP table, that means its trigger type must
805 * be read in from the ELCR */
807 #define default_EISA_trigger(idx) (EISA_ELCR(mp_irqs[idx].srcbusirq))
808 #define default_EISA_polarity(idx) default_ISA_polarity(idx)
810 /* PCI interrupts are always polarity one level triggered,
811 * when listed as conforming in the MP table. */
813 #define default_PCI_trigger(idx) (1)
814 #define default_PCI_polarity(idx) (1)
816 /* MCA interrupts are always polarity zero level triggered,
817 * when listed as conforming in the MP table. */
819 #define default_MCA_trigger(idx) (1)
820 #define default_MCA_polarity(idx) default_ISA_polarity(idx)
822 static int MPBIOS_polarity(int idx)
824 int bus = mp_irqs[idx].srcbus;
828 * Determine IRQ line polarity (high active or low active):
830 switch (mp_irqs[idx].irqflag & 3)
832 case 0: /* conforms, ie. bus-type dependent polarity */
833 if (test_bit(bus, mp_bus_not_pci))
834 polarity = default_ISA_polarity(idx);
836 polarity = default_PCI_polarity(idx);
838 case 1: /* high active */
843 case 2: /* reserved */
845 printk(KERN_WARNING "broken BIOS!!\n");
849 case 3: /* low active */
854 default: /* invalid */
856 printk(KERN_WARNING "broken BIOS!!\n");
864 static int MPBIOS_trigger(int idx)
866 int bus = mp_irqs[idx].srcbus;
870 * Determine IRQ trigger mode (edge or level sensitive):
872 switch ((mp_irqs[idx].irqflag>>2) & 3)
874 case 0: /* conforms, ie. bus-type dependent */
875 if (test_bit(bus, mp_bus_not_pci))
876 trigger = default_ISA_trigger(idx);
878 trigger = default_PCI_trigger(idx);
879 #if defined(CONFIG_EISA) || defined(CONFIG_MCA)
880 switch (mp_bus_id_to_type[bus]) {
881 case MP_BUS_ISA: /* ISA pin */
883 /* set before the switch */
886 case MP_BUS_EISA: /* EISA pin */
888 trigger = default_EISA_trigger(idx);
891 case MP_BUS_PCI: /* PCI pin */
893 /* set before the switch */
896 case MP_BUS_MCA: /* MCA pin */
898 trigger = default_MCA_trigger(idx);
903 printk(KERN_WARNING "broken BIOS!!\n");
915 case 2: /* reserved */
917 printk(KERN_WARNING "broken BIOS!!\n");
926 default: /* invalid */
928 printk(KERN_WARNING "broken BIOS!!\n");
936 static inline int irq_polarity(int idx)
938 return MPBIOS_polarity(idx);
941 static inline int irq_trigger(int idx)
943 return MPBIOS_trigger(idx);
946 static int pin_2_irq(int idx, int apic, int pin)
949 int bus = mp_irqs[idx].srcbus;
952 * Debugging check, we are in big trouble if this message pops up!
954 if (mp_irqs[idx].dstirq != pin)
955 printk(KERN_ERR "broken BIOS or MPTABLE parser, ayiee!!\n");
957 if (test_bit(bus, mp_bus_not_pci)) {
958 irq = mp_irqs[idx].srcbusirq;
960 u32 gsi = mp_gsi_routing[apic].gsi_base + pin;
962 if (gsi >= NR_IRQS_LEGACY)
970 * PCI IRQ command line redirection. Yes, limits are hardcoded.
972 if ((pin >= 16) && (pin <= 23)) {
973 if (pirq_entries[pin-16] != -1) {
974 if (!pirq_entries[pin-16]) {
975 apic_printk(APIC_VERBOSE, KERN_DEBUG
976 "disabling PIRQ%d\n", pin-16);
978 irq = pirq_entries[pin-16];
979 apic_printk(APIC_VERBOSE, KERN_DEBUG
980 "using PIRQ%d -> IRQ %d\n",
991 * Find a specific PCI IRQ entry.
992 * Not an __init, possibly needed by modules
994 int IO_APIC_get_PCI_irq_vector(int bus, int slot, int pin,
995 struct io_apic_irq_attr *irq_attr)
997 int apic, i, best_guess = -1;
999 apic_printk(APIC_DEBUG,
1000 "querying PCI -> IRQ mapping bus:%d, slot:%d, pin:%d.\n",
1002 if (test_bit(bus, mp_bus_not_pci)) {
1003 apic_printk(APIC_VERBOSE,
1004 "PCI BIOS passed nonexistent PCI bus %d!\n", bus);
1007 for (i = 0; i < mp_irq_entries; i++) {
1008 int lbus = mp_irqs[i].srcbus;
1010 for (apic = 0; apic < nr_ioapics; apic++)
1011 if (mp_ioapics[apic].apicid == mp_irqs[i].dstapic ||
1012 mp_irqs[i].dstapic == MP_APIC_ALL)
1015 if (!test_bit(lbus, mp_bus_not_pci) &&
1016 !mp_irqs[i].irqtype &&
1018 (slot == ((mp_irqs[i].srcbusirq >> 2) & 0x1f))) {
1019 int irq = pin_2_irq(i, apic, mp_irqs[i].dstirq);
1021 if (!(apic || IO_APIC_IRQ(irq)))
1024 if (pin == (mp_irqs[i].srcbusirq & 3)) {
1025 set_io_apic_irq_attr(irq_attr, apic,
1032 * Use the first all-but-pin matching entry as a
1033 * best-guess fuzzy result for broken mptables.
1035 if (best_guess < 0) {
1036 set_io_apic_irq_attr(irq_attr, apic,
1046 EXPORT_SYMBOL(IO_APIC_get_PCI_irq_vector);
1048 void lock_vector_lock(void)
1050 /* Used to the online set of cpus does not change
1051 * during assign_irq_vector.
1053 raw_spin_lock(&vector_lock);
1056 void unlock_vector_lock(void)
1058 raw_spin_unlock(&vector_lock);
1062 __assign_irq_vector(int irq, struct irq_cfg *cfg, const struct cpumask *mask)
1065 * NOTE! The local APIC isn't very good at handling
1066 * multiple interrupts at the same interrupt level.
1067 * As the interrupt level is determined by taking the
1068 * vector number and shifting that right by 4, we
1069 * want to spread these out a bit so that they don't
1070 * all fall in the same interrupt level.
1072 * Also, we've got to be careful not to trash gate
1073 * 0x80, because int 0x80 is hm, kind of importantish. ;)
1075 static int current_vector = FIRST_EXTERNAL_VECTOR + VECTOR_OFFSET_START;
1076 static int current_offset = VECTOR_OFFSET_START % 8;
1077 unsigned int old_vector;
1079 cpumask_var_t tmp_mask;
1081 if (cfg->move_in_progress)
1084 if (!alloc_cpumask_var(&tmp_mask, GFP_ATOMIC))
1087 old_vector = cfg->vector;
1089 cpumask_and(tmp_mask, mask, cpu_online_mask);
1090 cpumask_and(tmp_mask, cfg->domain, tmp_mask);
1091 if (!cpumask_empty(tmp_mask)) {
1092 free_cpumask_var(tmp_mask);
1097 /* Only try and allocate irqs on cpus that are present */
1099 for_each_cpu_and(cpu, mask, cpu_online_mask) {
1103 apic->vector_allocation_domain(cpu, tmp_mask);
1105 vector = current_vector;
1106 offset = current_offset;
1109 if (vector >= first_system_vector) {
1110 /* If out of vectors on large boxen, must share them. */
1111 offset = (offset + 1) % 8;
1112 vector = FIRST_EXTERNAL_VECTOR + offset;
1114 if (unlikely(current_vector == vector))
1117 if (test_bit(vector, used_vectors))
1120 for_each_cpu_and(new_cpu, tmp_mask, cpu_online_mask)
1121 if (per_cpu(vector_irq, new_cpu)[vector] != -1)
1124 current_vector = vector;
1125 current_offset = offset;
1127 cfg->move_in_progress = 1;
1128 cpumask_copy(cfg->old_domain, cfg->domain);
1130 for_each_cpu_and(new_cpu, tmp_mask, cpu_online_mask)
1131 per_cpu(vector_irq, new_cpu)[vector] = irq;
1132 cfg->vector = vector;
1133 cpumask_copy(cfg->domain, tmp_mask);
1137 free_cpumask_var(tmp_mask);
1141 int assign_irq_vector(int irq, struct irq_cfg *cfg, const struct cpumask *mask)
1144 unsigned long flags;
1146 raw_spin_lock_irqsave(&vector_lock, flags);
1147 err = __assign_irq_vector(irq, cfg, mask);
1148 raw_spin_unlock_irqrestore(&vector_lock, flags);
1152 static void __clear_irq_vector(int irq, struct irq_cfg *cfg)
1156 BUG_ON(!cfg->vector);
1158 vector = cfg->vector;
1159 for_each_cpu_and(cpu, cfg->domain, cpu_online_mask)
1160 per_cpu(vector_irq, cpu)[vector] = -1;
1163 cpumask_clear(cfg->domain);
1165 if (likely(!cfg->move_in_progress))
1167 for_each_cpu_and(cpu, cfg->old_domain, cpu_online_mask) {
1168 for (vector = FIRST_EXTERNAL_VECTOR; vector < NR_VECTORS;
1170 if (per_cpu(vector_irq, cpu)[vector] != irq)
1172 per_cpu(vector_irq, cpu)[vector] = -1;
1176 cfg->move_in_progress = 0;
1179 void __setup_vector_irq(int cpu)
1181 /* Initialize vector_irq on a new cpu */
1183 struct irq_cfg *cfg;
1186 * vector_lock will make sure that we don't run into irq vector
1187 * assignments that might be happening on another cpu in parallel,
1188 * while we setup our initial vector to irq mappings.
1190 raw_spin_lock(&vector_lock);
1191 /* Mark the inuse vectors */
1192 for_each_active_irq(irq) {
1193 cfg = get_irq_chip_data(irq);
1197 * If it is a legacy IRQ handled by the legacy PIC, this cpu
1198 * will be part of the irq_cfg's domain.
1200 if (irq < legacy_pic->nr_legacy_irqs && !IO_APIC_IRQ(irq))
1201 cpumask_set_cpu(cpu, cfg->domain);
1203 if (!cpumask_test_cpu(cpu, cfg->domain))
1205 vector = cfg->vector;
1206 per_cpu(vector_irq, cpu)[vector] = irq;
1208 /* Mark the free vectors */
1209 for (vector = 0; vector < NR_VECTORS; ++vector) {
1210 irq = per_cpu(vector_irq, cpu)[vector];
1215 if (!cpumask_test_cpu(cpu, cfg->domain))
1216 per_cpu(vector_irq, cpu)[vector] = -1;
1218 raw_spin_unlock(&vector_lock);
1221 static struct irq_chip ioapic_chip;
1222 static struct irq_chip ir_ioapic_chip;
1224 #define IOAPIC_AUTO -1
1225 #define IOAPIC_EDGE 0
1226 #define IOAPIC_LEVEL 1
1228 #ifdef CONFIG_X86_32
1229 static inline int IO_APIC_irq_trigger(int irq)
1233 for (apic = 0; apic < nr_ioapics; apic++) {
1234 for (pin = 0; pin < nr_ioapic_registers[apic]; pin++) {
1235 idx = find_irq_entry(apic, pin, mp_INT);
1236 if ((idx != -1) && (irq == pin_2_irq(idx, apic, pin)))
1237 return irq_trigger(idx);
1241 * nonexistent IRQs are edge default
1246 static inline int IO_APIC_irq_trigger(int irq)
1252 static void ioapic_register_intr(unsigned int irq, unsigned long trigger)
1255 if ((trigger == IOAPIC_AUTO && IO_APIC_irq_trigger(irq)) ||
1256 trigger == IOAPIC_LEVEL)
1257 irq_set_status_flags(irq, IRQ_LEVEL);
1259 irq_clear_status_flags(irq, IRQ_LEVEL);
1261 if (irq_remapped(get_irq_chip_data(irq))) {
1262 irq_set_status_flags(irq, IRQ_MOVE_PCNTXT);
1264 set_irq_chip_and_handler_name(irq, &ir_ioapic_chip,
1268 set_irq_chip_and_handler_name(irq, &ir_ioapic_chip,
1269 handle_edge_irq, "edge");
1273 if ((trigger == IOAPIC_AUTO && IO_APIC_irq_trigger(irq)) ||
1274 trigger == IOAPIC_LEVEL)
1275 set_irq_chip_and_handler_name(irq, &ioapic_chip,
1279 set_irq_chip_and_handler_name(irq, &ioapic_chip,
1280 handle_edge_irq, "edge");
1283 static int setup_ioapic_entry(int apic_id, int irq,
1284 struct IO_APIC_route_entry *entry,
1285 unsigned int destination, int trigger,
1286 int polarity, int vector, int pin)
1289 * add it to the IO-APIC irq-routing table:
1291 memset(entry,0,sizeof(*entry));
1293 if (intr_remapping_enabled) {
1294 struct intel_iommu *iommu = map_ioapic_to_ir(apic_id);
1296 struct IR_IO_APIC_route_entry *ir_entry =
1297 (struct IR_IO_APIC_route_entry *) entry;
1301 panic("No mapping iommu for ioapic %d\n", apic_id);
1303 index = alloc_irte(iommu, irq, 1);
1305 panic("Failed to allocate IRTE for ioapic %d\n", apic_id);
1307 prepare_irte(&irte, vector, destination);
1309 /* Set source-id of interrupt request */
1310 set_ioapic_sid(&irte, apic_id);
1312 modify_irte(irq, &irte);
1314 ir_entry->index2 = (index >> 15) & 0x1;
1316 ir_entry->format = 1;
1317 ir_entry->index = (index & 0x7fff);
1319 * IO-APIC RTE will be configured with virtual vector.
1320 * irq handler will do the explicit EOI to the io-apic.
1322 ir_entry->vector = pin;
1324 entry->delivery_mode = apic->irq_delivery_mode;
1325 entry->dest_mode = apic->irq_dest_mode;
1326 entry->dest = destination;
1327 entry->vector = vector;
1330 entry->mask = 0; /* enable IRQ */
1331 entry->trigger = trigger;
1332 entry->polarity = polarity;
1334 /* Mask level triggered irqs.
1335 * Use IRQ_DELAYED_DISABLE for edge triggered irqs.
1342 static void setup_ioapic_irq(int apic_id, int pin, unsigned int irq,
1343 struct irq_cfg *cfg, int trigger, int polarity)
1345 struct IO_APIC_route_entry entry;
1348 if (!IO_APIC_IRQ(irq))
1351 * For legacy irqs, cfg->domain starts with cpu 0 for legacy
1352 * controllers like 8259. Now that IO-APIC can handle this irq, update
1355 if (irq < legacy_pic->nr_legacy_irqs && cpumask_test_cpu(0, cfg->domain))
1356 apic->vector_allocation_domain(0, cfg->domain);
1358 if (assign_irq_vector(irq, cfg, apic->target_cpus()))
1361 dest = apic->cpu_mask_to_apicid_and(cfg->domain, apic->target_cpus());
1363 apic_printk(APIC_VERBOSE,KERN_DEBUG
1364 "IOAPIC[%d]: Set routing entry (%d-%d -> 0x%x -> "
1365 "IRQ %d Mode:%i Active:%i)\n",
1366 apic_id, mp_ioapics[apic_id].apicid, pin, cfg->vector,
1367 irq, trigger, polarity);
1370 if (setup_ioapic_entry(mp_ioapics[apic_id].apicid, irq, &entry,
1371 dest, trigger, polarity, cfg->vector, pin)) {
1372 printk("Failed to setup ioapic entry for ioapic %d, pin %d\n",
1373 mp_ioapics[apic_id].apicid, pin);
1374 __clear_irq_vector(irq, cfg);
1378 ioapic_register_intr(irq, trigger);
1379 if (irq < legacy_pic->nr_legacy_irqs)
1380 legacy_pic->mask(irq);
1382 ioapic_write_entry(apic_id, pin, entry);
1386 DECLARE_BITMAP(pin_programmed, MP_MAX_IOAPIC_PIN + 1);
1387 } mp_ioapic_routing[MAX_IO_APICS];
1389 static void __init setup_IO_APIC_irqs(void)
1391 int apic_id, pin, idx, irq, notcon = 0;
1392 int node = cpu_to_node(0);
1393 struct irq_cfg *cfg;
1395 apic_printk(APIC_VERBOSE, KERN_DEBUG "init IO_APIC IRQs\n");
1397 for (apic_id = 0; apic_id < nr_ioapics; apic_id++)
1398 for (pin = 0; pin < nr_ioapic_registers[apic_id]; pin++) {
1399 idx = find_irq_entry(apic_id, pin, mp_INT);
1403 apic_printk(APIC_VERBOSE,
1404 KERN_DEBUG " %d-%d",
1405 mp_ioapics[apic_id].apicid, pin);
1407 apic_printk(APIC_VERBOSE, " %d-%d",
1408 mp_ioapics[apic_id].apicid, pin);
1412 apic_printk(APIC_VERBOSE,
1413 " (apicid-pin) not connected\n");
1417 irq = pin_2_irq(idx, apic_id, pin);
1419 if ((apic_id > 0) && (irq > 16))
1423 * Skip the timer IRQ if there's a quirk handler
1424 * installed and if it returns 1:
1426 if (apic->multi_timer_check &&
1427 apic->multi_timer_check(apic_id, irq))
1430 cfg = alloc_irq_and_cfg_at(irq, node);
1434 add_pin_to_irq_node(cfg, node, apic_id, pin);
1436 * don't mark it in pin_programmed, so later acpi could
1437 * set it correctly when irq < 16
1439 setup_ioapic_irq(apic_id, pin, irq, cfg, irq_trigger(idx),
1444 apic_printk(APIC_VERBOSE,
1445 " (apicid-pin) not connected\n");
1449 * for the gsit that is not in first ioapic
1450 * but could not use acpi_register_gsi()
1451 * like some special sci in IBM x3330
1453 void setup_IO_APIC_irq_extra(u32 gsi)
1455 int apic_id = 0, pin, idx, irq, node = cpu_to_node(0);
1456 struct irq_cfg *cfg;
1459 * Convert 'gsi' to 'ioapic.pin'.
1461 apic_id = mp_find_ioapic(gsi);
1465 pin = mp_find_ioapic_pin(apic_id, gsi);
1466 idx = find_irq_entry(apic_id, pin, mp_INT);
1470 irq = pin_2_irq(idx, apic_id, pin);
1472 /* Only handle the non legacy irqs on secondary ioapics */
1473 if (apic_id == 0 || irq < NR_IRQS_LEGACY)
1476 cfg = alloc_irq_and_cfg_at(irq, node);
1480 add_pin_to_irq_node(cfg, node, apic_id, pin);
1482 if (test_bit(pin, mp_ioapic_routing[apic_id].pin_programmed)) {
1483 pr_debug("Pin %d-%d already programmed\n",
1484 mp_ioapics[apic_id].apicid, pin);
1487 set_bit(pin, mp_ioapic_routing[apic_id].pin_programmed);
1489 setup_ioapic_irq(apic_id, pin, irq, cfg,
1490 irq_trigger(idx), irq_polarity(idx));
1494 * Set up the timer pin, possibly with the 8259A-master behind.
1496 static void __init setup_timer_IRQ0_pin(unsigned int apic_id, unsigned int pin,
1499 struct IO_APIC_route_entry entry;
1501 if (intr_remapping_enabled)
1504 memset(&entry, 0, sizeof(entry));
1507 * We use logical delivery to get the timer IRQ
1510 entry.dest_mode = apic->irq_dest_mode;
1511 entry.mask = 0; /* don't mask IRQ for edge */
1512 entry.dest = apic->cpu_mask_to_apicid(apic->target_cpus());
1513 entry.delivery_mode = apic->irq_delivery_mode;
1516 entry.vector = vector;
1519 * The timer IRQ doesn't have to know that behind the
1520 * scene we may have a 8259A-master in AEOI mode ...
1522 set_irq_chip_and_handler_name(0, &ioapic_chip, handle_edge_irq, "edge");
1525 * Add it to the IO-APIC irq-routing table:
1527 ioapic_write_entry(apic_id, pin, entry);
1531 __apicdebuginit(void) print_IO_APIC(void)
1534 union IO_APIC_reg_00 reg_00;
1535 union IO_APIC_reg_01 reg_01;
1536 union IO_APIC_reg_02 reg_02;
1537 union IO_APIC_reg_03 reg_03;
1538 unsigned long flags;
1539 struct irq_cfg *cfg;
1542 printk(KERN_DEBUG "number of MP IRQ sources: %d.\n", mp_irq_entries);
1543 for (i = 0; i < nr_ioapics; i++)
1544 printk(KERN_DEBUG "number of IO-APIC #%d registers: %d.\n",
1545 mp_ioapics[i].apicid, nr_ioapic_registers[i]);
1548 * We are a bit conservative about what we expect. We have to
1549 * know about every hardware change ASAP.
1551 printk(KERN_INFO "testing the IO APIC.......................\n");
1553 for (apic = 0; apic < nr_ioapics; apic++) {
1555 raw_spin_lock_irqsave(&ioapic_lock, flags);
1556 reg_00.raw = io_apic_read(apic, 0);
1557 reg_01.raw = io_apic_read(apic, 1);
1558 if (reg_01.bits.version >= 0x10)
1559 reg_02.raw = io_apic_read(apic, 2);
1560 if (reg_01.bits.version >= 0x20)
1561 reg_03.raw = io_apic_read(apic, 3);
1562 raw_spin_unlock_irqrestore(&ioapic_lock, flags);
1565 printk(KERN_DEBUG "IO APIC #%d......\n", mp_ioapics[apic].apicid);
1566 printk(KERN_DEBUG ".... register #00: %08X\n", reg_00.raw);
1567 printk(KERN_DEBUG "....... : physical APIC id: %02X\n", reg_00.bits.ID);
1568 printk(KERN_DEBUG "....... : Delivery Type: %X\n", reg_00.bits.delivery_type);
1569 printk(KERN_DEBUG "....... : LTS : %X\n", reg_00.bits.LTS);
1571 printk(KERN_DEBUG ".... register #01: %08X\n", *(int *)®_01);
1572 printk(KERN_DEBUG "....... : max redirection entries: %04X\n", reg_01.bits.entries);
1574 printk(KERN_DEBUG "....... : PRQ implemented: %X\n", reg_01.bits.PRQ);
1575 printk(KERN_DEBUG "....... : IO APIC version: %04X\n", reg_01.bits.version);
1578 * Some Intel chipsets with IO APIC VERSION of 0x1? don't have reg_02,
1579 * but the value of reg_02 is read as the previous read register
1580 * value, so ignore it if reg_02 == reg_01.
1582 if (reg_01.bits.version >= 0x10 && reg_02.raw != reg_01.raw) {
1583 printk(KERN_DEBUG ".... register #02: %08X\n", reg_02.raw);
1584 printk(KERN_DEBUG "....... : arbitration: %02X\n", reg_02.bits.arbitration);
1588 * Some Intel chipsets with IO APIC VERSION of 0x2? don't have reg_02
1589 * or reg_03, but the value of reg_0[23] is read as the previous read
1590 * register value, so ignore it if reg_03 == reg_0[12].
1592 if (reg_01.bits.version >= 0x20 && reg_03.raw != reg_02.raw &&
1593 reg_03.raw != reg_01.raw) {
1594 printk(KERN_DEBUG ".... register #03: %08X\n", reg_03.raw);
1595 printk(KERN_DEBUG "....... : Boot DT : %X\n", reg_03.bits.boot_DT);
1598 printk(KERN_DEBUG ".... IRQ redirection table:\n");
1600 printk(KERN_DEBUG " NR Dst Mask Trig IRR Pol"
1601 " Stat Dmod Deli Vect:\n");
1603 for (i = 0; i <= reg_01.bits.entries; i++) {
1604 struct IO_APIC_route_entry entry;
1606 entry = ioapic_read_entry(apic, i);
1608 printk(KERN_DEBUG " %02x %03X ",
1613 printk("%1d %1d %1d %1d %1d %1d %1d %02X\n",
1618 entry.delivery_status,
1620 entry.delivery_mode,
1625 printk(KERN_DEBUG "IRQ to pin mappings:\n");
1626 for_each_active_irq(irq) {
1627 struct irq_pin_list *entry;
1629 cfg = get_irq_chip_data(irq);
1632 entry = cfg->irq_2_pin;
1635 printk(KERN_DEBUG "IRQ%d ", irq);
1636 for_each_irq_pin(entry, cfg->irq_2_pin)
1637 printk("-> %d:%d", entry->apic, entry->pin);
1641 printk(KERN_INFO ".................................... done.\n");
1646 __apicdebuginit(void) print_APIC_field(int base)
1652 for (i = 0; i < 8; i++)
1653 printk(KERN_CONT "%08x", apic_read(base + i*0x10));
1655 printk(KERN_CONT "\n");
1658 __apicdebuginit(void) print_local_APIC(void *dummy)
1660 unsigned int i, v, ver, maxlvt;
1663 printk(KERN_DEBUG "printing local APIC contents on CPU#%d/%d:\n",
1664 smp_processor_id(), hard_smp_processor_id());
1665 v = apic_read(APIC_ID);
1666 printk(KERN_INFO "... APIC ID: %08x (%01x)\n", v, read_apic_id());
1667 v = apic_read(APIC_LVR);
1668 printk(KERN_INFO "... APIC VERSION: %08x\n", v);
1669 ver = GET_APIC_VERSION(v);
1670 maxlvt = lapic_get_maxlvt();
1672 v = apic_read(APIC_TASKPRI);
1673 printk(KERN_DEBUG "... APIC TASKPRI: %08x (%02x)\n", v, v & APIC_TPRI_MASK);
1675 if (APIC_INTEGRATED(ver)) { /* !82489DX */
1676 if (!APIC_XAPIC(ver)) {
1677 v = apic_read(APIC_ARBPRI);
1678 printk(KERN_DEBUG "... APIC ARBPRI: %08x (%02x)\n", v,
1679 v & APIC_ARBPRI_MASK);
1681 v = apic_read(APIC_PROCPRI);
1682 printk(KERN_DEBUG "... APIC PROCPRI: %08x\n", v);
1686 * Remote read supported only in the 82489DX and local APIC for
1687 * Pentium processors.
1689 if (!APIC_INTEGRATED(ver) || maxlvt == 3) {
1690 v = apic_read(APIC_RRR);
1691 printk(KERN_DEBUG "... APIC RRR: %08x\n", v);
1694 v = apic_read(APIC_LDR);
1695 printk(KERN_DEBUG "... APIC LDR: %08x\n", v);
1696 if (!x2apic_enabled()) {
1697 v = apic_read(APIC_DFR);
1698 printk(KERN_DEBUG "... APIC DFR: %08x\n", v);
1700 v = apic_read(APIC_SPIV);
1701 printk(KERN_DEBUG "... APIC SPIV: %08x\n", v);
1703 printk(KERN_DEBUG "... APIC ISR field:\n");
1704 print_APIC_field(APIC_ISR);
1705 printk(KERN_DEBUG "... APIC TMR field:\n");
1706 print_APIC_field(APIC_TMR);
1707 printk(KERN_DEBUG "... APIC IRR field:\n");
1708 print_APIC_field(APIC_IRR);
1710 if (APIC_INTEGRATED(ver)) { /* !82489DX */
1711 if (maxlvt > 3) /* Due to the Pentium erratum 3AP. */
1712 apic_write(APIC_ESR, 0);
1714 v = apic_read(APIC_ESR);
1715 printk(KERN_DEBUG "... APIC ESR: %08x\n", v);
1718 icr = apic_icr_read();
1719 printk(KERN_DEBUG "... APIC ICR: %08x\n", (u32)icr);
1720 printk(KERN_DEBUG "... APIC ICR2: %08x\n", (u32)(icr >> 32));
1722 v = apic_read(APIC_LVTT);
1723 printk(KERN_DEBUG "... APIC LVTT: %08x\n", v);
1725 if (maxlvt > 3) { /* PC is LVT#4. */
1726 v = apic_read(APIC_LVTPC);
1727 printk(KERN_DEBUG "... APIC LVTPC: %08x\n", v);
1729 v = apic_read(APIC_LVT0);
1730 printk(KERN_DEBUG "... APIC LVT0: %08x\n", v);
1731 v = apic_read(APIC_LVT1);
1732 printk(KERN_DEBUG "... APIC LVT1: %08x\n", v);
1734 if (maxlvt > 2) { /* ERR is LVT#3. */
1735 v = apic_read(APIC_LVTERR);
1736 printk(KERN_DEBUG "... APIC LVTERR: %08x\n", v);
1739 v = apic_read(APIC_TMICT);
1740 printk(KERN_DEBUG "... APIC TMICT: %08x\n", v);
1741 v = apic_read(APIC_TMCCT);
1742 printk(KERN_DEBUG "... APIC TMCCT: %08x\n", v);
1743 v = apic_read(APIC_TDCR);
1744 printk(KERN_DEBUG "... APIC TDCR: %08x\n", v);
1746 if (boot_cpu_has(X86_FEATURE_EXTAPIC)) {
1747 v = apic_read(APIC_EFEAT);
1748 maxlvt = (v >> 16) & 0xff;
1749 printk(KERN_DEBUG "... APIC EFEAT: %08x\n", v);
1750 v = apic_read(APIC_ECTRL);
1751 printk(KERN_DEBUG "... APIC ECTRL: %08x\n", v);
1752 for (i = 0; i < maxlvt; i++) {
1753 v = apic_read(APIC_EILVTn(i));
1754 printk(KERN_DEBUG "... APIC EILVT%d: %08x\n", i, v);
1760 __apicdebuginit(void) print_local_APICs(int maxcpu)
1768 for_each_online_cpu(cpu) {
1771 smp_call_function_single(cpu, print_local_APIC, NULL, 1);
1776 __apicdebuginit(void) print_PIC(void)
1779 unsigned long flags;
1781 if (!legacy_pic->nr_legacy_irqs)
1784 printk(KERN_DEBUG "\nprinting PIC contents\n");
1786 raw_spin_lock_irqsave(&i8259A_lock, flags);
1788 v = inb(0xa1) << 8 | inb(0x21);
1789 printk(KERN_DEBUG "... PIC IMR: %04x\n", v);
1791 v = inb(0xa0) << 8 | inb(0x20);
1792 printk(KERN_DEBUG "... PIC IRR: %04x\n", v);
1796 v = inb(0xa0) << 8 | inb(0x20);
1800 raw_spin_unlock_irqrestore(&i8259A_lock, flags);
1802 printk(KERN_DEBUG "... PIC ISR: %04x\n", v);
1804 v = inb(0x4d1) << 8 | inb(0x4d0);
1805 printk(KERN_DEBUG "... PIC ELCR: %04x\n", v);
1808 static int __initdata show_lapic = 1;
1809 static __init int setup_show_lapic(char *arg)
1813 if (strcmp(arg, "all") == 0) {
1814 show_lapic = CONFIG_NR_CPUS;
1816 get_option(&arg, &num);
1823 __setup("show_lapic=", setup_show_lapic);
1825 __apicdebuginit(int) print_ICs(void)
1827 if (apic_verbosity == APIC_QUIET)
1832 /* don't print out if apic is not there */
1833 if (!cpu_has_apic && !apic_from_smp_config())
1836 print_local_APICs(show_lapic);
1842 fs_initcall(print_ICs);
1845 /* Where if anywhere is the i8259 connect in external int mode */
1846 static struct { int pin, apic; } ioapic_i8259 = { -1, -1 };
1848 void __init enable_IO_APIC(void)
1850 int i8259_apic, i8259_pin;
1853 if (!legacy_pic->nr_legacy_irqs)
1856 for(apic = 0; apic < nr_ioapics; apic++) {
1858 /* See if any of the pins is in ExtINT mode */
1859 for (pin = 0; pin < nr_ioapic_registers[apic]; pin++) {
1860 struct IO_APIC_route_entry entry;
1861 entry = ioapic_read_entry(apic, pin);
1863 /* If the interrupt line is enabled and in ExtInt mode
1864 * I have found the pin where the i8259 is connected.
1866 if ((entry.mask == 0) && (entry.delivery_mode == dest_ExtINT)) {
1867 ioapic_i8259.apic = apic;
1868 ioapic_i8259.pin = pin;
1874 /* Look to see what if the MP table has reported the ExtINT */
1875 /* If we could not find the appropriate pin by looking at the ioapic
1876 * the i8259 probably is not connected the ioapic but give the
1877 * mptable a chance anyway.
1879 i8259_pin = find_isa_irq_pin(0, mp_ExtINT);
1880 i8259_apic = find_isa_irq_apic(0, mp_ExtINT);
1881 /* Trust the MP table if nothing is setup in the hardware */
1882 if ((ioapic_i8259.pin == -1) && (i8259_pin >= 0)) {
1883 printk(KERN_WARNING "ExtINT not setup in hardware but reported by MP table\n");
1884 ioapic_i8259.pin = i8259_pin;
1885 ioapic_i8259.apic = i8259_apic;
1887 /* Complain if the MP table and the hardware disagree */
1888 if (((ioapic_i8259.apic != i8259_apic) || (ioapic_i8259.pin != i8259_pin)) &&
1889 (i8259_pin >= 0) && (ioapic_i8259.pin >= 0))
1891 printk(KERN_WARNING "ExtINT in hardware and MP table differ\n");
1895 * Do not trust the IO-APIC being empty at bootup
1901 * Not an __init, needed by the reboot code
1903 void disable_IO_APIC(void)
1906 * Clear the IO-APIC before rebooting:
1910 if (!legacy_pic->nr_legacy_irqs)
1914 * If the i8259 is routed through an IOAPIC
1915 * Put that IOAPIC in virtual wire mode
1916 * so legacy interrupts can be delivered.
1918 * With interrupt-remapping, for now we will use virtual wire A mode,
1919 * as virtual wire B is little complex (need to configure both
1920 * IOAPIC RTE aswell as interrupt-remapping table entry).
1921 * As this gets called during crash dump, keep this simple for now.
1923 if (ioapic_i8259.pin != -1 && !intr_remapping_enabled) {
1924 struct IO_APIC_route_entry entry;
1926 memset(&entry, 0, sizeof(entry));
1927 entry.mask = 0; /* Enabled */
1928 entry.trigger = 0; /* Edge */
1930 entry.polarity = 0; /* High */
1931 entry.delivery_status = 0;
1932 entry.dest_mode = 0; /* Physical */
1933 entry.delivery_mode = dest_ExtINT; /* ExtInt */
1935 entry.dest = read_apic_id();
1938 * Add it to the IO-APIC irq-routing table:
1940 ioapic_write_entry(ioapic_i8259.apic, ioapic_i8259.pin, entry);
1944 * Use virtual wire A mode when interrupt remapping is enabled.
1946 if (cpu_has_apic || apic_from_smp_config())
1947 disconnect_bsp_APIC(!intr_remapping_enabled &&
1948 ioapic_i8259.pin != -1);
1951 #ifdef CONFIG_X86_32
1953 * function to set the IO-APIC physical IDs based on the
1954 * values stored in the MPC table.
1956 * by Matt Domsch <Matt_Domsch@dell.com> Tue Dec 21 12:25:05 CST 1999
1958 void __init setup_ioapic_ids_from_mpc_nocheck(void)
1960 union IO_APIC_reg_00 reg_00;
1961 physid_mask_t phys_id_present_map;
1964 unsigned char old_id;
1965 unsigned long flags;
1968 * This is broken; anything with a real cpu count has to
1969 * circumvent this idiocy regardless.
1971 apic->ioapic_phys_id_map(&phys_cpu_present_map, &phys_id_present_map);
1974 * Set the IOAPIC ID to the value stored in the MPC table.
1976 for (apic_id = 0; apic_id < nr_ioapics; apic_id++) {
1978 /* Read the register 0 value */
1979 raw_spin_lock_irqsave(&ioapic_lock, flags);
1980 reg_00.raw = io_apic_read(apic_id, 0);
1981 raw_spin_unlock_irqrestore(&ioapic_lock, flags);
1983 old_id = mp_ioapics[apic_id].apicid;
1985 if (mp_ioapics[apic_id].apicid >= get_physical_broadcast()) {
1986 printk(KERN_ERR "BIOS bug, IO-APIC#%d ID is %d in the MPC table!...\n",
1987 apic_id, mp_ioapics[apic_id].apicid);
1988 printk(KERN_ERR "... fixing up to %d. (tell your hw vendor)\n",
1990 mp_ioapics[apic_id].apicid = reg_00.bits.ID;
1994 * Sanity check, is the ID really free? Every APIC in a
1995 * system must have a unique ID or we get lots of nice
1996 * 'stuck on smp_invalidate_needed IPI wait' messages.
1998 if (apic->check_apicid_used(&phys_id_present_map,
1999 mp_ioapics[apic_id].apicid)) {
2000 printk(KERN_ERR "BIOS bug, IO-APIC#%d ID %d is already used!...\n",
2001 apic_id, mp_ioapics[apic_id].apicid);
2002 for (i = 0; i < get_physical_broadcast(); i++)
2003 if (!physid_isset(i, phys_id_present_map))
2005 if (i >= get_physical_broadcast())
2006 panic("Max APIC ID exceeded!\n");
2007 printk(KERN_ERR "... fixing up to %d. (tell your hw vendor)\n",
2009 physid_set(i, phys_id_present_map);
2010 mp_ioapics[apic_id].apicid = i;
2013 apic->apicid_to_cpu_present(mp_ioapics[apic_id].apicid, &tmp);
2014 apic_printk(APIC_VERBOSE, "Setting %d in the "
2015 "phys_id_present_map\n",
2016 mp_ioapics[apic_id].apicid);
2017 physids_or(phys_id_present_map, phys_id_present_map, tmp);
2021 * We need to adjust the IRQ routing table
2022 * if the ID changed.
2024 if (old_id != mp_ioapics[apic_id].apicid)
2025 for (i = 0; i < mp_irq_entries; i++)
2026 if (mp_irqs[i].dstapic == old_id)
2028 = mp_ioapics[apic_id].apicid;
2031 * Update the ID register according to the right value
2032 * from the MPC table if they are different.
2034 if (mp_ioapics[apic_id].apicid == reg_00.bits.ID)
2037 apic_printk(APIC_VERBOSE, KERN_INFO
2038 "...changing IO-APIC physical APIC ID to %d ...",
2039 mp_ioapics[apic_id].apicid);
2041 reg_00.bits.ID = mp_ioapics[apic_id].apicid;
2042 raw_spin_lock_irqsave(&ioapic_lock, flags);
2043 io_apic_write(apic_id, 0, reg_00.raw);
2044 raw_spin_unlock_irqrestore(&ioapic_lock, flags);
2049 raw_spin_lock_irqsave(&ioapic_lock, flags);
2050 reg_00.raw = io_apic_read(apic_id, 0);
2051 raw_spin_unlock_irqrestore(&ioapic_lock, flags);
2052 if (reg_00.bits.ID != mp_ioapics[apic_id].apicid)
2053 printk("could not set ID!\n");
2055 apic_printk(APIC_VERBOSE, " ok.\n");
2059 void __init setup_ioapic_ids_from_mpc(void)
2065 * Don't check I/O APIC IDs for xAPIC systems. They have
2066 * no meaning without the serial APIC bus.
2068 if (!(boot_cpu_data.x86_vendor == X86_VENDOR_INTEL)
2069 || APIC_XAPIC(apic_version[boot_cpu_physical_apicid]))
2071 setup_ioapic_ids_from_mpc_nocheck();
2075 int no_timer_check __initdata;
2077 static int __init notimercheck(char *s)
2082 __setup("no_timer_check", notimercheck);
2085 * There is a nasty bug in some older SMP boards, their mptable lies
2086 * about the timer IRQ. We do the following to work around the situation:
2088 * - timer IRQ defaults to IO-APIC IRQ
2089 * - if this function detects that timer IRQs are defunct, then we fall
2090 * back to ISA timer IRQs
2092 static int __init timer_irq_works(void)
2094 unsigned long t1 = jiffies;
2095 unsigned long flags;
2100 local_save_flags(flags);
2102 /* Let ten ticks pass... */
2103 mdelay((10 * 1000) / HZ);
2104 local_irq_restore(flags);
2107 * Expect a few ticks at least, to be sure some possible
2108 * glue logic does not lock up after one or two first
2109 * ticks in a non-ExtINT mode. Also the local APIC
2110 * might have cached one ExtINT interrupt. Finally, at
2111 * least one tick may be lost due to delays.
2115 if (time_after(jiffies, t1 + 4))
2121 * In the SMP+IOAPIC case it might happen that there are an unspecified
2122 * number of pending IRQ events unhandled. These cases are very rare,
2123 * so we 'resend' these IRQs via IPIs, to the same CPU. It's much
2124 * better to do it this way as thus we do not have to be aware of
2125 * 'pending' interrupts in the IRQ path, except at this point.
2128 * Edge triggered needs to resend any interrupt
2129 * that was delayed but this is now handled in the device
2134 * Starting up a edge-triggered IO-APIC interrupt is
2135 * nasty - we need to make sure that we get the edge.
2136 * If it is already asserted for some reason, we need
2137 * return 1 to indicate that is was pending.
2139 * This is not complete - we should be able to fake
2140 * an edge even if it isn't on the 8259A...
2143 static unsigned int startup_ioapic_irq(struct irq_data *data)
2145 int was_pending = 0, irq = data->irq;
2146 unsigned long flags;
2148 raw_spin_lock_irqsave(&ioapic_lock, flags);
2149 if (irq < legacy_pic->nr_legacy_irqs) {
2150 legacy_pic->mask(irq);
2151 if (legacy_pic->irq_pending(irq))
2154 __unmask_ioapic(data->chip_data);
2155 raw_spin_unlock_irqrestore(&ioapic_lock, flags);
2160 static int ioapic_retrigger_irq(struct irq_data *data)
2162 struct irq_cfg *cfg = data->chip_data;
2163 unsigned long flags;
2165 raw_spin_lock_irqsave(&vector_lock, flags);
2166 apic->send_IPI_mask(cpumask_of(cpumask_first(cfg->domain)), cfg->vector);
2167 raw_spin_unlock_irqrestore(&vector_lock, flags);
2173 * Level and edge triggered IO-APIC interrupts need different handling,
2174 * so we use two separate IRQ descriptors. Edge triggered IRQs can be
2175 * handled with the level-triggered descriptor, but that one has slightly
2176 * more overhead. Level-triggered interrupts cannot be handled with the
2177 * edge-triggered handler, without risking IRQ storms and other ugly
2182 void send_cleanup_vector(struct irq_cfg *cfg)
2184 cpumask_var_t cleanup_mask;
2186 if (unlikely(!alloc_cpumask_var(&cleanup_mask, GFP_ATOMIC))) {
2188 for_each_cpu_and(i, cfg->old_domain, cpu_online_mask)
2189 apic->send_IPI_mask(cpumask_of(i), IRQ_MOVE_CLEANUP_VECTOR);
2191 cpumask_and(cleanup_mask, cfg->old_domain, cpu_online_mask);
2192 apic->send_IPI_mask(cleanup_mask, IRQ_MOVE_CLEANUP_VECTOR);
2193 free_cpumask_var(cleanup_mask);
2195 cfg->move_in_progress = 0;
2198 static void __target_IO_APIC_irq(unsigned int irq, unsigned int dest, struct irq_cfg *cfg)
2201 struct irq_pin_list *entry;
2202 u8 vector = cfg->vector;
2204 for_each_irq_pin(entry, cfg->irq_2_pin) {
2210 * With interrupt-remapping, destination information comes
2211 * from interrupt-remapping table entry.
2213 if (!irq_remapped(cfg))
2214 io_apic_write(apic, 0x11 + pin*2, dest);
2215 reg = io_apic_read(apic, 0x10 + pin*2);
2216 reg &= ~IO_APIC_REDIR_VECTOR_MASK;
2218 io_apic_modify(apic, 0x10 + pin*2, reg);
2223 * Either sets data->affinity to a valid value, and returns
2224 * ->cpu_mask_to_apicid of that in dest_id, or returns -1 and
2225 * leaves data->affinity untouched.
2227 int __ioapic_set_affinity(struct irq_data *data, const struct cpumask *mask,
2228 unsigned int *dest_id)
2230 struct irq_cfg *cfg = data->chip_data;
2232 if (!cpumask_intersects(mask, cpu_online_mask))
2235 if (assign_irq_vector(data->irq, data->chip_data, mask))
2238 cpumask_copy(data->affinity, mask);
2240 *dest_id = apic->cpu_mask_to_apicid_and(mask, cfg->domain);
2245 ioapic_set_affinity(struct irq_data *data, const struct cpumask *mask,
2248 unsigned int dest, irq = data->irq;
2249 unsigned long flags;
2252 raw_spin_lock_irqsave(&ioapic_lock, flags);
2253 ret = __ioapic_set_affinity(data, mask, &dest);
2255 /* Only the high 8 bits are valid. */
2256 dest = SET_APIC_LOGICAL_ID(dest);
2257 __target_IO_APIC_irq(irq, dest, data->chip_data);
2259 raw_spin_unlock_irqrestore(&ioapic_lock, flags);
2263 #ifdef CONFIG_INTR_REMAP
2266 * Migrate the IO-APIC irq in the presence of intr-remapping.
2268 * For both level and edge triggered, irq migration is a simple atomic
2269 * update(of vector and cpu destination) of IRTE and flush the hardware cache.
2271 * For level triggered, we eliminate the io-apic RTE modification (with the
2272 * updated vector information), by using a virtual vector (io-apic pin number).
2273 * Real vector that is used for interrupting cpu will be coming from
2274 * the interrupt-remapping table entry.
2277 ir_ioapic_set_affinity(struct irq_data *data, const struct cpumask *mask,
2280 struct irq_cfg *cfg = data->chip_data;
2281 unsigned int dest, irq = data->irq;
2284 if (!cpumask_intersects(mask, cpu_online_mask))
2287 if (get_irte(irq, &irte))
2290 if (assign_irq_vector(irq, cfg, mask))
2293 dest = apic->cpu_mask_to_apicid_and(cfg->domain, mask);
2295 irte.vector = cfg->vector;
2296 irte.dest_id = IRTE_DEST(dest);
2299 * Modified the IRTE and flushes the Interrupt entry cache.
2301 modify_irte(irq, &irte);
2303 if (cfg->move_in_progress)
2304 send_cleanup_vector(cfg);
2306 cpumask_copy(data->affinity, mask);
2312 ir_ioapic_set_affinity(struct irq_data *data, const struct cpumask *mask,
2319 asmlinkage void smp_irq_move_cleanup_interrupt(void)
2321 unsigned vector, me;
2327 me = smp_processor_id();
2328 for (vector = FIRST_EXTERNAL_VECTOR; vector < NR_VECTORS; vector++) {
2331 struct irq_desc *desc;
2332 struct irq_cfg *cfg;
2333 irq = __get_cpu_var(vector_irq)[vector];
2338 desc = irq_to_desc(irq);
2343 raw_spin_lock(&desc->lock);
2346 * Check if the irq migration is in progress. If so, we
2347 * haven't received the cleanup request yet for this irq.
2349 if (cfg->move_in_progress)
2352 if (vector == cfg->vector && cpumask_test_cpu(me, cfg->domain))
2355 irr = apic_read(APIC_IRR + (vector / 32 * 0x10));
2357 * Check if the vector that needs to be cleanedup is
2358 * registered at the cpu's IRR. If so, then this is not
2359 * the best time to clean it up. Lets clean it up in the
2360 * next attempt by sending another IRQ_MOVE_CLEANUP_VECTOR
2363 if (irr & (1 << (vector % 32))) {
2364 apic->send_IPI_self(IRQ_MOVE_CLEANUP_VECTOR);
2367 __get_cpu_var(vector_irq)[vector] = -1;
2369 raw_spin_unlock(&desc->lock);
2375 static void __irq_complete_move(struct irq_cfg *cfg, unsigned vector)
2379 if (likely(!cfg->move_in_progress))
2382 me = smp_processor_id();
2384 if (vector == cfg->vector && cpumask_test_cpu(me, cfg->domain))
2385 send_cleanup_vector(cfg);
2388 static void irq_complete_move(struct irq_cfg *cfg)
2390 __irq_complete_move(cfg, ~get_irq_regs()->orig_ax);
2393 void irq_force_complete_move(int irq)
2395 struct irq_cfg *cfg = get_irq_chip_data(irq);
2400 __irq_complete_move(cfg, cfg->vector);
2403 static inline void irq_complete_move(struct irq_cfg *cfg) { }
2406 static void ack_apic_edge(struct irq_data *data)
2408 irq_complete_move(data->chip_data);
2409 move_native_irq(data->irq);
2413 atomic_t irq_mis_count;
2416 * IO-APIC versions below 0x20 don't support EOI register.
2417 * For the record, here is the information about various versions:
2419 * 1Xh I/OAPIC or I/O(x)APIC which are not PCI 2.2 Compliant
2420 * 2Xh I/O(x)APIC which is PCI 2.2 Compliant
2423 * Some of the Intel ICH Specs (ICH2 to ICH5) documents the io-apic
2424 * version as 0x2. This is an error with documentation and these ICH chips
2425 * use io-apic's of version 0x20.
2427 * For IO-APIC's with EOI register, we use that to do an explicit EOI.
2428 * Otherwise, we simulate the EOI message manually by changing the trigger
2429 * mode to edge and then back to level, with RTE being masked during this.
2431 static void eoi_ioapic_irq(unsigned int irq, struct irq_cfg *cfg)
2433 struct irq_pin_list *entry;
2434 unsigned long flags;
2436 raw_spin_lock_irqsave(&ioapic_lock, flags);
2437 for_each_irq_pin(entry, cfg->irq_2_pin) {
2438 if (mp_ioapics[entry->apic].apicver >= 0x20) {
2440 * Intr-remapping uses pin number as the virtual vector
2441 * in the RTE. Actual vector is programmed in
2442 * intr-remapping table entry. Hence for the io-apic
2443 * EOI we use the pin number.
2445 if (irq_remapped(cfg))
2446 io_apic_eoi(entry->apic, entry->pin);
2448 io_apic_eoi(entry->apic, cfg->vector);
2450 __mask_and_edge_IO_APIC_irq(entry);
2451 __unmask_and_level_IO_APIC_irq(entry);
2454 raw_spin_unlock_irqrestore(&ioapic_lock, flags);
2457 static void ack_apic_level(struct irq_data *data)
2459 struct irq_cfg *cfg = data->chip_data;
2460 int i, do_unmask_irq = 0, irq = data->irq;
2461 struct irq_desc *desc = irq_to_desc(irq);
2464 irq_complete_move(cfg);
2465 #ifdef CONFIG_GENERIC_PENDING_IRQ
2466 /* If we are moving the irq we need to mask it */
2467 if (unlikely(desc->status & IRQ_MOVE_PENDING)) {
2474 * It appears there is an erratum which affects at least version 0x11
2475 * of I/O APIC (that's the 82093AA and cores integrated into various
2476 * chipsets). Under certain conditions a level-triggered interrupt is
2477 * erroneously delivered as edge-triggered one but the respective IRR
2478 * bit gets set nevertheless. As a result the I/O unit expects an EOI
2479 * message but it will never arrive and further interrupts are blocked
2480 * from the source. The exact reason is so far unknown, but the
2481 * phenomenon was observed when two consecutive interrupt requests
2482 * from a given source get delivered to the same CPU and the source is
2483 * temporarily disabled in between.
2485 * A workaround is to simulate an EOI message manually. We achieve it
2486 * by setting the trigger mode to edge and then to level when the edge
2487 * trigger mode gets detected in the TMR of a local APIC for a
2488 * level-triggered interrupt. We mask the source for the time of the
2489 * operation to prevent an edge-triggered interrupt escaping meanwhile.
2490 * The idea is from Manfred Spraul. --macro
2492 * Also in the case when cpu goes offline, fixup_irqs() will forward
2493 * any unhandled interrupt on the offlined cpu to the new cpu
2494 * destination that is handling the corresponding interrupt. This
2495 * interrupt forwarding is done via IPI's. Hence, in this case also
2496 * level-triggered io-apic interrupt will be seen as an edge
2497 * interrupt in the IRR. And we can't rely on the cpu's EOI
2498 * to be broadcasted to the IO-APIC's which will clear the remoteIRR
2499 * corresponding to the level-triggered interrupt. Hence on IO-APIC's
2500 * supporting EOI register, we do an explicit EOI to clear the
2501 * remote IRR and on IO-APIC's which don't have an EOI register,
2502 * we use the above logic (mask+edge followed by unmask+level) from
2503 * Manfred Spraul to clear the remote IRR.
2506 v = apic_read(APIC_TMR + ((i & ~0x1f) >> 1));
2509 * We must acknowledge the irq before we move it or the acknowledge will
2510 * not propagate properly.
2515 * Tail end of clearing remote IRR bit (either by delivering the EOI
2516 * message via io-apic EOI register write or simulating it using
2517 * mask+edge followed by unnask+level logic) manually when the
2518 * level triggered interrupt is seen as the edge triggered interrupt
2521 if (!(v & (1 << (i & 0x1f)))) {
2522 atomic_inc(&irq_mis_count);
2524 eoi_ioapic_irq(irq, cfg);
2527 /* Now we can move and renable the irq */
2528 if (unlikely(do_unmask_irq)) {
2529 /* Only migrate the irq if the ack has been received.
2531 * On rare occasions the broadcast level triggered ack gets
2532 * delayed going to ioapics, and if we reprogram the
2533 * vector while Remote IRR is still set the irq will never
2536 * To prevent this scenario we read the Remote IRR bit
2537 * of the ioapic. This has two effects.
2538 * - On any sane system the read of the ioapic will
2539 * flush writes (and acks) going to the ioapic from
2541 * - We get to see if the ACK has actually been delivered.
2543 * Based on failed experiments of reprogramming the
2544 * ioapic entry from outside of irq context starting
2545 * with masking the ioapic entry and then polling until
2546 * Remote IRR was clear before reprogramming the
2547 * ioapic I don't trust the Remote IRR bit to be
2548 * completey accurate.
2550 * However there appears to be no other way to plug
2551 * this race, so if the Remote IRR bit is not
2552 * accurate and is causing problems then it is a hardware bug
2553 * and you can go talk to the chipset vendor about it.
2555 if (!io_apic_level_ack_pending(cfg))
2556 move_masked_irq(irq);
2561 #ifdef CONFIG_INTR_REMAP
2562 static void ir_ack_apic_edge(struct irq_data *data)
2567 static void ir_ack_apic_level(struct irq_data *data)
2570 eoi_ioapic_irq(data->irq, data->chip_data);
2572 #endif /* CONFIG_INTR_REMAP */
2574 static struct irq_chip ioapic_chip __read_mostly = {
2576 .irq_startup = startup_ioapic_irq,
2577 .irq_mask = mask_ioapic_irq,
2578 .irq_unmask = unmask_ioapic_irq,
2579 .irq_ack = ack_apic_edge,
2580 .irq_eoi = ack_apic_level,
2582 .irq_set_affinity = ioapic_set_affinity,
2584 .irq_retrigger = ioapic_retrigger_irq,
2587 static struct irq_chip ir_ioapic_chip __read_mostly = {
2588 .name = "IR-IO-APIC",
2589 .irq_startup = startup_ioapic_irq,
2590 .irq_mask = mask_ioapic_irq,
2591 .irq_unmask = unmask_ioapic_irq,
2592 #ifdef CONFIG_INTR_REMAP
2593 .irq_ack = ir_ack_apic_edge,
2594 .irq_eoi = ir_ack_apic_level,
2596 .irq_set_affinity = ir_ioapic_set_affinity,
2599 .irq_retrigger = ioapic_retrigger_irq,
2602 static inline void init_IO_APIC_traps(void)
2604 struct irq_cfg *cfg;
2608 * NOTE! The local APIC isn't very good at handling
2609 * multiple interrupts at the same interrupt level.
2610 * As the interrupt level is determined by taking the
2611 * vector number and shifting that right by 4, we
2612 * want to spread these out a bit so that they don't
2613 * all fall in the same interrupt level.
2615 * Also, we've got to be careful not to trash gate
2616 * 0x80, because int 0x80 is hm, kind of importantish. ;)
2618 for_each_active_irq(irq) {
2619 cfg = get_irq_chip_data(irq);
2620 if (IO_APIC_IRQ(irq) && cfg && !cfg->vector) {
2622 * Hmm.. We don't have an entry for this,
2623 * so default to an old-fashioned 8259
2624 * interrupt if we can..
2626 if (irq < legacy_pic->nr_legacy_irqs)
2627 legacy_pic->make_irq(irq);
2629 /* Strange. Oh, well.. */
2630 set_irq_chip(irq, &no_irq_chip);
2636 * The local APIC irq-chip implementation:
2639 static void mask_lapic_irq(struct irq_data *data)
2643 v = apic_read(APIC_LVT0);
2644 apic_write(APIC_LVT0, v | APIC_LVT_MASKED);
2647 static void unmask_lapic_irq(struct irq_data *data)
2651 v = apic_read(APIC_LVT0);
2652 apic_write(APIC_LVT0, v & ~APIC_LVT_MASKED);
2655 static void ack_lapic_irq(struct irq_data *data)
2660 static struct irq_chip lapic_chip __read_mostly = {
2661 .name = "local-APIC",
2662 .irq_mask = mask_lapic_irq,
2663 .irq_unmask = unmask_lapic_irq,
2664 .irq_ack = ack_lapic_irq,
2667 static void lapic_register_intr(int irq)
2669 irq_clear_status_flags(irq, IRQ_LEVEL);
2670 set_irq_chip_and_handler_name(irq, &lapic_chip, handle_edge_irq,
2674 static void __init setup_nmi(void)
2677 * Dirty trick to enable the NMI watchdog ...
2678 * We put the 8259A master into AEOI mode and
2679 * unmask on all local APICs LVT0 as NMI.
2681 * The idea to use the 8259A in AEOI mode ('8259A Virtual Wire')
2682 * is from Maciej W. Rozycki - so we do not have to EOI from
2683 * the NMI handler or the timer interrupt.
2685 apic_printk(APIC_VERBOSE, KERN_INFO "activating NMI Watchdog ...");
2687 enable_NMI_through_LVT0();
2689 apic_printk(APIC_VERBOSE, " done.\n");
2693 * This looks a bit hackish but it's about the only one way of sending
2694 * a few INTA cycles to 8259As and any associated glue logic. ICR does
2695 * not support the ExtINT mode, unfortunately. We need to send these
2696 * cycles as some i82489DX-based boards have glue logic that keeps the
2697 * 8259A interrupt line asserted until INTA. --macro
2699 static inline void __init unlock_ExtINT_logic(void)
2702 struct IO_APIC_route_entry entry0, entry1;
2703 unsigned char save_control, save_freq_select;
2705 pin = find_isa_irq_pin(8, mp_INT);
2710 apic = find_isa_irq_apic(8, mp_INT);
2716 entry0 = ioapic_read_entry(apic, pin);
2717 clear_IO_APIC_pin(apic, pin);
2719 memset(&entry1, 0, sizeof(entry1));
2721 entry1.dest_mode = 0; /* physical delivery */
2722 entry1.mask = 0; /* unmask IRQ now */
2723 entry1.dest = hard_smp_processor_id();
2724 entry1.delivery_mode = dest_ExtINT;
2725 entry1.polarity = entry0.polarity;
2729 ioapic_write_entry(apic, pin, entry1);
2731 save_control = CMOS_READ(RTC_CONTROL);
2732 save_freq_select = CMOS_READ(RTC_FREQ_SELECT);
2733 CMOS_WRITE((save_freq_select & ~RTC_RATE_SELECT) | 0x6,
2735 CMOS_WRITE(save_control | RTC_PIE, RTC_CONTROL);
2740 if ((CMOS_READ(RTC_INTR_FLAGS) & RTC_PF) == RTC_PF)
2744 CMOS_WRITE(save_control, RTC_CONTROL);
2745 CMOS_WRITE(save_freq_select, RTC_FREQ_SELECT);
2746 clear_IO_APIC_pin(apic, pin);
2748 ioapic_write_entry(apic, pin, entry0);
2751 static int disable_timer_pin_1 __initdata;
2752 /* Actually the next is obsolete, but keep it for paranoid reasons -AK */
2753 static int __init disable_timer_pin_setup(char *arg)
2755 disable_timer_pin_1 = 1;
2758 early_param("disable_timer_pin_1", disable_timer_pin_setup);
2760 int timer_through_8259 __initdata;
2763 * This code may look a bit paranoid, but it's supposed to cooperate with
2764 * a wide range of boards and BIOS bugs. Fortunately only the timer IRQ
2765 * is so screwy. Thanks to Brian Perkins for testing/hacking this beast
2766 * fanatically on his truly buggy board.
2768 * FIXME: really need to revamp this for all platforms.
2770 static inline void __init check_timer(void)
2772 struct irq_cfg *cfg = get_irq_chip_data(0);
2773 int node = cpu_to_node(0);
2774 int apic1, pin1, apic2, pin2;
2775 unsigned long flags;
2778 local_irq_save(flags);
2781 * get/set the timer IRQ vector:
2783 legacy_pic->mask(0);
2784 assign_irq_vector(0, cfg, apic->target_cpus());
2787 * As IRQ0 is to be enabled in the 8259A, the virtual
2788 * wire has to be disabled in the local APIC. Also
2789 * timer interrupts need to be acknowledged manually in
2790 * the 8259A for the i82489DX when using the NMI
2791 * watchdog as that APIC treats NMIs as level-triggered.
2792 * The AEOI mode will finish them in the 8259A
2795 apic_write(APIC_LVT0, APIC_LVT_MASKED | APIC_DM_EXTINT);
2796 legacy_pic->init(1);
2797 #ifdef CONFIG_X86_32
2801 ver = apic_read(APIC_LVR);
2802 ver = GET_APIC_VERSION(ver);
2803 timer_ack = (nmi_watchdog == NMI_IO_APIC && !APIC_INTEGRATED(ver));
2807 pin1 = find_isa_irq_pin(0, mp_INT);
2808 apic1 = find_isa_irq_apic(0, mp_INT);
2809 pin2 = ioapic_i8259.pin;
2810 apic2 = ioapic_i8259.apic;
2812 apic_printk(APIC_QUIET, KERN_INFO "..TIMER: vector=0x%02X "
2813 "apic1=%d pin1=%d apic2=%d pin2=%d\n",
2814 cfg->vector, apic1, pin1, apic2, pin2);
2817 * Some BIOS writers are clueless and report the ExtINTA
2818 * I/O APIC input from the cascaded 8259A as the timer
2819 * interrupt input. So just in case, if only one pin
2820 * was found above, try it both directly and through the
2824 if (intr_remapping_enabled)
2825 panic("BIOS bug: timer not connected to IO-APIC");
2829 } else if (pin2 == -1) {
2836 * Ok, does IRQ0 through the IOAPIC work?
2839 add_pin_to_irq_node(cfg, node, apic1, pin1);
2840 setup_timer_IRQ0_pin(apic1, pin1, cfg->vector);
2842 /* for edge trigger, setup_ioapic_irq already
2843 * leave it unmasked.
2844 * so only need to unmask if it is level-trigger
2845 * do we really have level trigger timer?
2848 idx = find_irq_entry(apic1, pin1, mp_INT);
2849 if (idx != -1 && irq_trigger(idx))
2852 if (timer_irq_works()) {
2853 if (nmi_watchdog == NMI_IO_APIC) {
2855 legacy_pic->unmask(0);
2857 if (disable_timer_pin_1 > 0)
2858 clear_IO_APIC_pin(0, pin1);
2861 if (intr_remapping_enabled)
2862 panic("timer doesn't work through Interrupt-remapped IO-APIC");
2863 local_irq_disable();
2864 clear_IO_APIC_pin(apic1, pin1);
2866 apic_printk(APIC_QUIET, KERN_ERR "..MP-BIOS bug: "
2867 "8254 timer not connected to IO-APIC\n");
2869 apic_printk(APIC_QUIET, KERN_INFO "...trying to set up timer "
2870 "(IRQ0) through the 8259A ...\n");
2871 apic_printk(APIC_QUIET, KERN_INFO
2872 "..... (found apic %d pin %d) ...\n", apic2, pin2);
2874 * legacy devices should be connected to IO APIC #0
2876 replace_pin_at_irq_node(cfg, node, apic1, pin1, apic2, pin2);
2877 setup_timer_IRQ0_pin(apic2, pin2, cfg->vector);
2878 legacy_pic->unmask(0);
2879 if (timer_irq_works()) {
2880 apic_printk(APIC_QUIET, KERN_INFO "....... works.\n");
2881 timer_through_8259 = 1;
2882 if (nmi_watchdog == NMI_IO_APIC) {
2883 legacy_pic->mask(0);
2885 legacy_pic->unmask(0);
2890 * Cleanup, just in case ...
2892 local_irq_disable();
2893 legacy_pic->mask(0);
2894 clear_IO_APIC_pin(apic2, pin2);
2895 apic_printk(APIC_QUIET, KERN_INFO "....... failed.\n");
2898 if (nmi_watchdog == NMI_IO_APIC) {
2899 apic_printk(APIC_QUIET, KERN_WARNING "timer doesn't work "
2900 "through the IO-APIC - disabling NMI Watchdog!\n");
2901 nmi_watchdog = NMI_NONE;
2903 #ifdef CONFIG_X86_32
2907 apic_printk(APIC_QUIET, KERN_INFO
2908 "...trying to set up timer as Virtual Wire IRQ...\n");
2910 lapic_register_intr(0);
2911 apic_write(APIC_LVT0, APIC_DM_FIXED | cfg->vector); /* Fixed mode */
2912 legacy_pic->unmask(0);
2914 if (timer_irq_works()) {
2915 apic_printk(APIC_QUIET, KERN_INFO "..... works.\n");
2918 local_irq_disable();
2919 legacy_pic->mask(0);
2920 apic_write(APIC_LVT0, APIC_LVT_MASKED | APIC_DM_FIXED | cfg->vector);
2921 apic_printk(APIC_QUIET, KERN_INFO "..... failed.\n");
2923 apic_printk(APIC_QUIET, KERN_INFO
2924 "...trying to set up timer as ExtINT IRQ...\n");
2926 legacy_pic->init(0);
2927 legacy_pic->make_irq(0);
2928 apic_write(APIC_LVT0, APIC_DM_EXTINT);
2930 unlock_ExtINT_logic();
2932 if (timer_irq_works()) {
2933 apic_printk(APIC_QUIET, KERN_INFO "..... works.\n");
2936 local_irq_disable();
2937 apic_printk(APIC_QUIET, KERN_INFO "..... failed :(.\n");
2938 panic("IO-APIC + timer doesn't work! Boot with apic=debug and send a "
2939 "report. Then try booting with the 'noapic' option.\n");
2941 local_irq_restore(flags);
2945 * Traditionally ISA IRQ2 is the cascade IRQ, and is not available
2946 * to devices. However there may be an I/O APIC pin available for
2947 * this interrupt regardless. The pin may be left unconnected, but
2948 * typically it will be reused as an ExtINT cascade interrupt for
2949 * the master 8259A. In the MPS case such a pin will normally be
2950 * reported as an ExtINT interrupt in the MP table. With ACPI
2951 * there is no provision for ExtINT interrupts, and in the absence
2952 * of an override it would be treated as an ordinary ISA I/O APIC
2953 * interrupt, that is edge-triggered and unmasked by default. We
2954 * used to do this, but it caused problems on some systems because
2955 * of the NMI watchdog and sometimes IRQ0 of the 8254 timer using
2956 * the same ExtINT cascade interrupt to drive the local APIC of the
2957 * bootstrap processor. Therefore we refrain from routing IRQ2 to
2958 * the I/O APIC in all cases now. No actual device should request
2959 * it anyway. --macro
2961 #define PIC_IRQS (1UL << PIC_CASCADE_IR)
2963 void __init setup_IO_APIC(void)
2967 * calling enable_IO_APIC() is moved to setup_local_APIC for BP
2969 io_apic_irqs = legacy_pic->nr_legacy_irqs ? ~PIC_IRQS : ~0UL;
2971 apic_printk(APIC_VERBOSE, "ENABLING IO-APIC IRQs\n");
2973 * Set up IO-APIC IRQ routing.
2975 x86_init.mpparse.setup_ioapic_ids();
2978 setup_IO_APIC_irqs();
2979 init_IO_APIC_traps();
2980 if (legacy_pic->nr_legacy_irqs)
2985 * Called after all the initialization is done. If we didnt find any
2986 * APIC bugs then we can allow the modify fast path
2989 static int __init io_apic_bug_finalize(void)
2991 if (sis_apic_bug == -1)
2996 late_initcall(io_apic_bug_finalize);
2998 struct sysfs_ioapic_data {
2999 struct sys_device dev;
3000 struct IO_APIC_route_entry entry[0];
3002 static struct sysfs_ioapic_data * mp_ioapic_data[MAX_IO_APICS];
3004 static int ioapic_suspend(struct sys_device *dev, pm_message_t state)
3006 struct IO_APIC_route_entry *entry;
3007 struct sysfs_ioapic_data *data;
3010 data = container_of(dev, struct sysfs_ioapic_data, dev);
3011 entry = data->entry;
3012 for (i = 0; i < nr_ioapic_registers[dev->id]; i ++, entry ++ )
3013 *entry = ioapic_read_entry(dev->id, i);
3018 static int ioapic_resume(struct sys_device *dev)
3020 struct IO_APIC_route_entry *entry;
3021 struct sysfs_ioapic_data *data;
3022 unsigned long flags;
3023 union IO_APIC_reg_00 reg_00;
3026 data = container_of(dev, struct sysfs_ioapic_data, dev);
3027 entry = data->entry;
3029 raw_spin_lock_irqsave(&ioapic_lock, flags);
3030 reg_00.raw = io_apic_read(dev->id, 0);
3031 if (reg_00.bits.ID != mp_ioapics[dev->id].apicid) {
3032 reg_00.bits.ID = mp_ioapics[dev->id].apicid;
3033 io_apic_write(dev->id, 0, reg_00.raw);
3035 raw_spin_unlock_irqrestore(&ioapic_lock, flags);
3036 for (i = 0; i < nr_ioapic_registers[dev->id]; i++)
3037 ioapic_write_entry(dev->id, i, entry[i]);
3042 static struct sysdev_class ioapic_sysdev_class = {
3044 .suspend = ioapic_suspend,
3045 .resume = ioapic_resume,
3048 static int __init ioapic_init_sysfs(void)
3050 struct sys_device * dev;
3053 error = sysdev_class_register(&ioapic_sysdev_class);
3057 for (i = 0; i < nr_ioapics; i++ ) {
3058 size = sizeof(struct sys_device) + nr_ioapic_registers[i]
3059 * sizeof(struct IO_APIC_route_entry);
3060 mp_ioapic_data[i] = kzalloc(size, GFP_KERNEL);
3061 if (!mp_ioapic_data[i]) {
3062 printk(KERN_ERR "Can't suspend/resume IOAPIC %d\n", i);
3065 dev = &mp_ioapic_data[i]->dev;
3067 dev->cls = &ioapic_sysdev_class;
3068 error = sysdev_register(dev);
3070 kfree(mp_ioapic_data[i]);
3071 mp_ioapic_data[i] = NULL;
3072 printk(KERN_ERR "Can't suspend/resume IOAPIC %d\n", i);
3080 device_initcall(ioapic_init_sysfs);
3083 * Dynamic irq allocate and deallocation
3085 unsigned int create_irq_nr(unsigned int from, int node)
3087 struct irq_cfg *cfg;
3088 unsigned long flags;
3089 unsigned int ret = 0;
3092 if (from < nr_irqs_gsi)
3095 irq = alloc_irq_from(from, node);
3098 cfg = alloc_irq_cfg(irq, node);
3100 free_irq_at(irq, NULL);
3104 raw_spin_lock_irqsave(&vector_lock, flags);
3105 if (!__assign_irq_vector(irq, cfg, apic->target_cpus()))
3107 raw_spin_unlock_irqrestore(&vector_lock, flags);
3110 set_irq_chip_data(irq, cfg);
3111 irq_clear_status_flags(irq, IRQ_NOREQUEST);
3113 free_irq_at(irq, cfg);
3118 int create_irq(void)
3120 int node = cpu_to_node(0);
3121 unsigned int irq_want;
3124 irq_want = nr_irqs_gsi;
3125 irq = create_irq_nr(irq_want, node);
3133 void destroy_irq(unsigned int irq)
3135 struct irq_cfg *cfg = get_irq_chip_data(irq);
3136 unsigned long flags;
3138 irq_set_status_flags(irq, IRQ_NOREQUEST|IRQ_NOPROBE);
3140 if (irq_remapped(cfg))
3142 raw_spin_lock_irqsave(&vector_lock, flags);
3143 __clear_irq_vector(irq, cfg);
3144 raw_spin_unlock_irqrestore(&vector_lock, flags);
3145 free_irq_at(irq, cfg);
3149 * MSI message composition
3151 #ifdef CONFIG_PCI_MSI
3152 static int msi_compose_msg(struct pci_dev *pdev, unsigned int irq,
3153 struct msi_msg *msg, u8 hpet_id)
3155 struct irq_cfg *cfg;
3163 err = assign_irq_vector(irq, cfg, apic->target_cpus());
3167 dest = apic->cpu_mask_to_apicid_and(cfg->domain, apic->target_cpus());
3169 if (irq_remapped(get_irq_chip_data(irq))) {
3174 ir_index = map_irq_to_irte_handle(irq, &sub_handle);
3175 BUG_ON(ir_index == -1);
3177 prepare_irte(&irte, cfg->vector, dest);
3179 /* Set source-id of interrupt request */
3181 set_msi_sid(&irte, pdev);
3183 set_hpet_sid(&irte, hpet_id);
3185 modify_irte(irq, &irte);
3187 msg->address_hi = MSI_ADDR_BASE_HI;
3188 msg->data = sub_handle;
3189 msg->address_lo = MSI_ADDR_BASE_LO | MSI_ADDR_IR_EXT_INT |
3191 MSI_ADDR_IR_INDEX1(ir_index) |
3192 MSI_ADDR_IR_INDEX2(ir_index);
3194 if (x2apic_enabled())
3195 msg->address_hi = MSI_ADDR_BASE_HI |
3196 MSI_ADDR_EXT_DEST_ID(dest);
3198 msg->address_hi = MSI_ADDR_BASE_HI;
3202 ((apic->irq_dest_mode == 0) ?
3203 MSI_ADDR_DEST_MODE_PHYSICAL:
3204 MSI_ADDR_DEST_MODE_LOGICAL) |
3205 ((apic->irq_delivery_mode != dest_LowestPrio) ?
3206 MSI_ADDR_REDIRECTION_CPU:
3207 MSI_ADDR_REDIRECTION_LOWPRI) |
3208 MSI_ADDR_DEST_ID(dest);
3211 MSI_DATA_TRIGGER_EDGE |
3212 MSI_DATA_LEVEL_ASSERT |
3213 ((apic->irq_delivery_mode != dest_LowestPrio) ?
3214 MSI_DATA_DELIVERY_FIXED:
3215 MSI_DATA_DELIVERY_LOWPRI) |
3216 MSI_DATA_VECTOR(cfg->vector);
3223 msi_set_affinity(struct irq_data *data, const struct cpumask *mask, bool force)
3225 struct irq_cfg *cfg = data->chip_data;
3229 if (__ioapic_set_affinity(data, mask, &dest))
3232 __get_cached_msi_msg(data->msi_desc, &msg);
3234 msg.data &= ~MSI_DATA_VECTOR_MASK;
3235 msg.data |= MSI_DATA_VECTOR(cfg->vector);
3236 msg.address_lo &= ~MSI_ADDR_DEST_ID_MASK;
3237 msg.address_lo |= MSI_ADDR_DEST_ID(dest);
3239 __write_msi_msg(data->msi_desc, &msg);
3243 #ifdef CONFIG_INTR_REMAP
3245 * Migrate the MSI irq to another cpumask. This migration is
3246 * done in the process context using interrupt-remapping hardware.
3249 ir_msi_set_affinity(struct irq_data *data, const struct cpumask *mask,
3252 struct irq_cfg *cfg = data->chip_data;
3253 unsigned int dest, irq = data->irq;
3256 if (get_irte(irq, &irte))
3259 if (__ioapic_set_affinity(data, mask, &dest))
3262 irte.vector = cfg->vector;
3263 irte.dest_id = IRTE_DEST(dest);
3266 * atomically update the IRTE with the new destination and vector.
3268 modify_irte(irq, &irte);
3271 * After this point, all the interrupts will start arriving
3272 * at the new destination. So, time to cleanup the previous
3273 * vector allocation.
3275 if (cfg->move_in_progress)
3276 send_cleanup_vector(cfg);
3282 #endif /* CONFIG_SMP */
3285 * IRQ Chip for MSI PCI/PCI-X/PCI-Express Devices,
3286 * which implement the MSI or MSI-X Capability Structure.
3288 static struct irq_chip msi_chip = {
3290 .irq_unmask = unmask_msi_irq,
3291 .irq_mask = mask_msi_irq,
3292 .irq_ack = ack_apic_edge,
3294 .irq_set_affinity = msi_set_affinity,
3296 .irq_retrigger = ioapic_retrigger_irq,
3299 static struct irq_chip msi_ir_chip = {
3300 .name = "IR-PCI-MSI",
3301 .irq_unmask = unmask_msi_irq,
3302 .irq_mask = mask_msi_irq,
3303 #ifdef CONFIG_INTR_REMAP
3304 .irq_ack = ir_ack_apic_edge,
3306 .irq_set_affinity = ir_msi_set_affinity,
3309 .irq_retrigger = ioapic_retrigger_irq,
3313 * Map the PCI dev to the corresponding remapping hardware unit
3314 * and allocate 'nvec' consecutive interrupt-remapping table entries
3317 static int msi_alloc_irte(struct pci_dev *dev, int irq, int nvec)
3319 struct intel_iommu *iommu;
3322 iommu = map_dev_to_ir(dev);
3325 "Unable to map PCI %s to iommu\n", pci_name(dev));
3329 index = alloc_irte(iommu, irq, nvec);
3332 "Unable to allocate %d IRTE for PCI %s\n", nvec,
3339 static int setup_msi_irq(struct pci_dev *dev, struct msi_desc *msidesc, int irq)
3344 ret = msi_compose_msg(dev, irq, &msg, -1);
3348 set_irq_msi(irq, msidesc);
3349 write_msi_msg(irq, &msg);
3351 if (irq_remapped(get_irq_chip_data(irq))) {
3352 irq_set_status_flags(irq, IRQ_MOVE_PCNTXT);
3353 set_irq_chip_and_handler_name(irq, &msi_ir_chip, handle_edge_irq, "edge");
3355 set_irq_chip_and_handler_name(irq, &msi_chip, handle_edge_irq, "edge");
3357 dev_printk(KERN_DEBUG, &dev->dev, "irq %d for MSI/MSI-X\n", irq);
3362 int native_setup_msi_irqs(struct pci_dev *dev, int nvec, int type)
3364 int node, ret, sub_handle, index = 0;
3365 unsigned int irq, irq_want;
3366 struct msi_desc *msidesc;
3367 struct intel_iommu *iommu = NULL;
3369 /* x86 doesn't support multiple MSI yet */
3370 if (type == PCI_CAP_ID_MSI && nvec > 1)
3373 node = dev_to_node(&dev->dev);
3374 irq_want = nr_irqs_gsi;
3376 list_for_each_entry(msidesc, &dev->msi_list, list) {
3377 irq = create_irq_nr(irq_want, node);
3381 if (!intr_remapping_enabled)
3386 * allocate the consecutive block of IRTE's
3389 index = msi_alloc_irte(dev, irq, nvec);
3395 iommu = map_dev_to_ir(dev);
3401 * setup the mapping between the irq and the IRTE
3402 * base index, the sub_handle pointing to the
3403 * appropriate interrupt remap table entry.
3405 set_irte_irq(irq, iommu, index, sub_handle);
3408 ret = setup_msi_irq(dev, msidesc, irq);
3420 void native_teardown_msi_irq(unsigned int irq)
3425 #if defined (CONFIG_DMAR) || defined (CONFIG_INTR_REMAP)
3428 dmar_msi_set_affinity(struct irq_data *data, const struct cpumask *mask,
3431 struct irq_cfg *cfg = data->chip_data;
3432 unsigned int dest, irq = data->irq;
3435 if (__ioapic_set_affinity(data, mask, &dest))
3438 dmar_msi_read(irq, &msg);
3440 msg.data &= ~MSI_DATA_VECTOR_MASK;
3441 msg.data |= MSI_DATA_VECTOR(cfg->vector);
3442 msg.address_lo &= ~MSI_ADDR_DEST_ID_MASK;
3443 msg.address_lo |= MSI_ADDR_DEST_ID(dest);
3445 dmar_msi_write(irq, &msg);
3450 #endif /* CONFIG_SMP */
3452 static struct irq_chip dmar_msi_type = {
3454 .irq_unmask = dmar_msi_unmask,
3455 .irq_mask = dmar_msi_mask,
3456 .irq_ack = ack_apic_edge,
3458 .irq_set_affinity = dmar_msi_set_affinity,
3460 .irq_retrigger = ioapic_retrigger_irq,
3463 int arch_setup_dmar_msi(unsigned int irq)
3468 ret = msi_compose_msg(NULL, irq, &msg, -1);
3471 dmar_msi_write(irq, &msg);
3472 set_irq_chip_and_handler_name(irq, &dmar_msi_type, handle_edge_irq,
3478 #ifdef CONFIG_HPET_TIMER
3481 static int hpet_msi_set_affinity(struct irq_data *data,
3482 const struct cpumask *mask, bool force)
3484 struct irq_cfg *cfg = data->chip_data;
3488 if (__ioapic_set_affinity(data, mask, &dest))
3491 hpet_msi_read(data->handler_data, &msg);
3493 msg.data &= ~MSI_DATA_VECTOR_MASK;
3494 msg.data |= MSI_DATA_VECTOR(cfg->vector);
3495 msg.address_lo &= ~MSI_ADDR_DEST_ID_MASK;
3496 msg.address_lo |= MSI_ADDR_DEST_ID(dest);
3498 hpet_msi_write(data->handler_data, &msg);
3503 #endif /* CONFIG_SMP */
3505 static struct irq_chip ir_hpet_msi_type = {
3506 .name = "IR-HPET_MSI",
3507 .irq_unmask = hpet_msi_unmask,
3508 .irq_mask = hpet_msi_mask,
3509 #ifdef CONFIG_INTR_REMAP
3510 .irq_ack = ir_ack_apic_edge,
3512 .irq_set_affinity = ir_msi_set_affinity,
3515 .irq_retrigger = ioapic_retrigger_irq,
3518 static struct irq_chip hpet_msi_type = {
3520 .irq_unmask = hpet_msi_unmask,
3521 .irq_mask = hpet_msi_mask,
3522 .irq_ack = ack_apic_edge,
3524 .irq_set_affinity = hpet_msi_set_affinity,
3526 .irq_retrigger = ioapic_retrigger_irq,
3529 int arch_setup_hpet_msi(unsigned int irq, unsigned int id)
3534 if (intr_remapping_enabled) {
3535 struct intel_iommu *iommu = map_hpet_to_ir(id);
3541 index = alloc_irte(iommu, irq, 1);
3546 ret = msi_compose_msg(NULL, irq, &msg, id);
3550 hpet_msi_write(get_irq_data(irq), &msg);
3551 irq_set_status_flags(irq, IRQ_MOVE_PCNTXT);
3552 if (irq_remapped(get_irq_chip_data(irq)))
3553 set_irq_chip_and_handler_name(irq, &ir_hpet_msi_type,
3554 handle_edge_irq, "edge");
3556 set_irq_chip_and_handler_name(irq, &hpet_msi_type,
3557 handle_edge_irq, "edge");
3563 #endif /* CONFIG_PCI_MSI */
3565 * Hypertransport interrupt support
3567 #ifdef CONFIG_HT_IRQ
3571 static void target_ht_irq(unsigned int irq, unsigned int dest, u8 vector)
3573 struct ht_irq_msg msg;
3574 fetch_ht_irq_msg(irq, &msg);
3576 msg.address_lo &= ~(HT_IRQ_LOW_VECTOR_MASK | HT_IRQ_LOW_DEST_ID_MASK);
3577 msg.address_hi &= ~(HT_IRQ_HIGH_DEST_ID_MASK);
3579 msg.address_lo |= HT_IRQ_LOW_VECTOR(vector) | HT_IRQ_LOW_DEST_ID(dest);
3580 msg.address_hi |= HT_IRQ_HIGH_DEST_ID(dest);
3582 write_ht_irq_msg(irq, &msg);
3586 ht_set_affinity(struct irq_data *data, const struct cpumask *mask, bool force)
3588 struct irq_cfg *cfg = data->chip_data;
3591 if (__ioapic_set_affinity(data, mask, &dest))
3594 target_ht_irq(data->irq, dest, cfg->vector);
3600 static struct irq_chip ht_irq_chip = {
3602 .irq_mask = mask_ht_irq,
3603 .irq_unmask = unmask_ht_irq,
3604 .irq_ack = ack_apic_edge,
3606 .irq_set_affinity = ht_set_affinity,
3608 .irq_retrigger = ioapic_retrigger_irq,
3611 int arch_setup_ht_irq(unsigned int irq, struct pci_dev *dev)
3613 struct irq_cfg *cfg;
3620 err = assign_irq_vector(irq, cfg, apic->target_cpus());
3622 struct ht_irq_msg msg;
3625 dest = apic->cpu_mask_to_apicid_and(cfg->domain,
3626 apic->target_cpus());
3628 msg.address_hi = HT_IRQ_HIGH_DEST_ID(dest);
3632 HT_IRQ_LOW_DEST_ID(dest) |
3633 HT_IRQ_LOW_VECTOR(cfg->vector) |
3634 ((apic->irq_dest_mode == 0) ?
3635 HT_IRQ_LOW_DM_PHYSICAL :
3636 HT_IRQ_LOW_DM_LOGICAL) |
3637 HT_IRQ_LOW_RQEOI_EDGE |
3638 ((apic->irq_delivery_mode != dest_LowestPrio) ?
3639 HT_IRQ_LOW_MT_FIXED :
3640 HT_IRQ_LOW_MT_ARBITRATED) |
3641 HT_IRQ_LOW_IRQ_MASKED;
3643 write_ht_irq_msg(irq, &msg);
3645 set_irq_chip_and_handler_name(irq, &ht_irq_chip,
3646 handle_edge_irq, "edge");
3648 dev_printk(KERN_DEBUG, &dev->dev, "irq %d for HT\n", irq);
3652 #endif /* CONFIG_HT_IRQ */
3654 int __init io_apic_get_redir_entries (int ioapic)
3656 union IO_APIC_reg_01 reg_01;
3657 unsigned long flags;
3659 raw_spin_lock_irqsave(&ioapic_lock, flags);
3660 reg_01.raw = io_apic_read(ioapic, 1);
3661 raw_spin_unlock_irqrestore(&ioapic_lock, flags);
3663 /* The register returns the maximum index redir index
3664 * supported, which is one less than the total number of redir
3667 return reg_01.bits.entries + 1;
3670 static void __init probe_nr_irqs_gsi(void)
3674 nr = gsi_top + NR_IRQS_LEGACY;
3675 if (nr > nr_irqs_gsi)
3678 printk(KERN_DEBUG "nr_irqs_gsi: %d\n", nr_irqs_gsi);
3681 int get_nr_irqs_gsi(void)
3686 #ifdef CONFIG_SPARSE_IRQ
3687 int __init arch_probe_nr_irqs(void)
3691 if (nr_irqs > (NR_VECTORS * nr_cpu_ids))
3692 nr_irqs = NR_VECTORS * nr_cpu_ids;
3694 nr = nr_irqs_gsi + 8 * nr_cpu_ids;
3695 #if defined(CONFIG_PCI_MSI) || defined(CONFIG_HT_IRQ)
3697 * for MSI and HT dyn irq
3699 nr += nr_irqs_gsi * 16;
3704 return NR_IRQS_LEGACY;
3708 static int __io_apic_set_pci_routing(struct device *dev, int irq,
3709 struct io_apic_irq_attr *irq_attr)
3711 struct irq_cfg *cfg;
3714 int trigger, polarity;
3716 ioapic = irq_attr->ioapic;
3717 if (!IO_APIC_IRQ(irq)) {
3718 apic_printk(APIC_QUIET,KERN_ERR "IOAPIC[%d]: Invalid reference to IRQ 0\n",
3724 node = dev_to_node(dev);
3726 node = cpu_to_node(0);
3728 cfg = alloc_irq_and_cfg_at(irq, node);
3732 pin = irq_attr->ioapic_pin;
3733 trigger = irq_attr->trigger;
3734 polarity = irq_attr->polarity;
3737 * IRQs < 16 are already in the irq_2_pin[] map
3739 if (irq >= legacy_pic->nr_legacy_irqs) {
3740 if (__add_pin_to_irq_node(cfg, node, ioapic, pin)) {
3741 printk(KERN_INFO "can not add pin %d for irq %d\n",
3747 setup_ioapic_irq(ioapic, pin, irq, cfg, trigger, polarity);
3752 int io_apic_set_pci_routing(struct device *dev, int irq,
3753 struct io_apic_irq_attr *irq_attr)
3757 * Avoid pin reprogramming. PRTs typically include entries
3758 * with redundant pin->gsi mappings (but unique PCI devices);
3759 * we only program the IOAPIC on the first.
3761 ioapic = irq_attr->ioapic;
3762 pin = irq_attr->ioapic_pin;
3763 if (test_bit(pin, mp_ioapic_routing[ioapic].pin_programmed)) {
3764 pr_debug("Pin %d-%d already programmed\n",
3765 mp_ioapics[ioapic].apicid, pin);
3768 set_bit(pin, mp_ioapic_routing[ioapic].pin_programmed);
3770 return __io_apic_set_pci_routing(dev, irq, irq_attr);
3773 u8 __init io_apic_unique_id(u8 id)
3775 #ifdef CONFIG_X86_32
3776 if ((boot_cpu_data.x86_vendor == X86_VENDOR_INTEL) &&
3777 !APIC_XAPIC(apic_version[boot_cpu_physical_apicid]))
3778 return io_apic_get_unique_id(nr_ioapics, id);
3783 DECLARE_BITMAP(used, 256);
3785 bitmap_zero(used, 256);
3786 for (i = 0; i < nr_ioapics; i++) {
3787 struct mpc_ioapic *ia = &mp_ioapics[i];
3788 __set_bit(ia->apicid, used);
3790 if (!test_bit(id, used))
3792 return find_first_zero_bit(used, 256);
3796 #ifdef CONFIG_X86_32
3797 int __init io_apic_get_unique_id(int ioapic, int apic_id)
3799 union IO_APIC_reg_00 reg_00;
3800 static physid_mask_t apic_id_map = PHYSID_MASK_NONE;
3802 unsigned long flags;
3806 * The P4 platform supports up to 256 APIC IDs on two separate APIC
3807 * buses (one for LAPICs, one for IOAPICs), where predecessors only
3808 * supports up to 16 on one shared APIC bus.
3810 * TBD: Expand LAPIC/IOAPIC support on P4-class systems to take full
3811 * advantage of new APIC bus architecture.
3814 if (physids_empty(apic_id_map))
3815 apic->ioapic_phys_id_map(&phys_cpu_present_map, &apic_id_map);
3817 raw_spin_lock_irqsave(&ioapic_lock, flags);
3818 reg_00.raw = io_apic_read(ioapic, 0);
3819 raw_spin_unlock_irqrestore(&ioapic_lock, flags);
3821 if (apic_id >= get_physical_broadcast()) {
3822 printk(KERN_WARNING "IOAPIC[%d]: Invalid apic_id %d, trying "
3823 "%d\n", ioapic, apic_id, reg_00.bits.ID);
3824 apic_id = reg_00.bits.ID;
3828 * Every APIC in a system must have a unique ID or we get lots of nice
3829 * 'stuck on smp_invalidate_needed IPI wait' messages.
3831 if (apic->check_apicid_used(&apic_id_map, apic_id)) {
3833 for (i = 0; i < get_physical_broadcast(); i++) {
3834 if (!apic->check_apicid_used(&apic_id_map, i))
3838 if (i == get_physical_broadcast())
3839 panic("Max apic_id exceeded!\n");
3841 printk(KERN_WARNING "IOAPIC[%d]: apic_id %d already used, "
3842 "trying %d\n", ioapic, apic_id, i);
3847 apic->apicid_to_cpu_present(apic_id, &tmp);
3848 physids_or(apic_id_map, apic_id_map, tmp);
3850 if (reg_00.bits.ID != apic_id) {
3851 reg_00.bits.ID = apic_id;
3853 raw_spin_lock_irqsave(&ioapic_lock, flags);
3854 io_apic_write(ioapic, 0, reg_00.raw);
3855 reg_00.raw = io_apic_read(ioapic, 0);
3856 raw_spin_unlock_irqrestore(&ioapic_lock, flags);
3859 if (reg_00.bits.ID != apic_id) {
3860 printk("IOAPIC[%d]: Unable to change apic_id!\n", ioapic);
3865 apic_printk(APIC_VERBOSE, KERN_INFO
3866 "IOAPIC[%d]: Assigned apic_id %d\n", ioapic, apic_id);
3872 int __init io_apic_get_version(int ioapic)
3874 union IO_APIC_reg_01 reg_01;
3875 unsigned long flags;
3877 raw_spin_lock_irqsave(&ioapic_lock, flags);
3878 reg_01.raw = io_apic_read(ioapic, 1);
3879 raw_spin_unlock_irqrestore(&ioapic_lock, flags);
3881 return reg_01.bits.version;
3884 int acpi_get_override_irq(u32 gsi, int *trigger, int *polarity)
3886 int ioapic, pin, idx;
3888 if (skip_ioapic_setup)
3891 ioapic = mp_find_ioapic(gsi);
3895 pin = mp_find_ioapic_pin(ioapic, gsi);
3899 idx = find_irq_entry(ioapic, pin, mp_INT);
3903 *trigger = irq_trigger(idx);
3904 *polarity = irq_polarity(idx);
3909 * This function currently is only a helper for the i386 smp boot process where
3910 * we need to reprogram the ioredtbls to cater for the cpus which have come online
3911 * so mask in all cases should simply be apic->target_cpus()
3914 void __init setup_ioapic_dest(void)
3916 int pin, ioapic, irq, irq_entry;
3917 struct irq_desc *desc;
3918 const struct cpumask *mask;
3920 if (skip_ioapic_setup == 1)
3923 for (ioapic = 0; ioapic < nr_ioapics; ioapic++)
3924 for (pin = 0; pin < nr_ioapic_registers[ioapic]; pin++) {
3925 irq_entry = find_irq_entry(ioapic, pin, mp_INT);
3926 if (irq_entry == -1)
3928 irq = pin_2_irq(irq_entry, ioapic, pin);
3930 if ((ioapic > 0) && (irq > 16))
3933 desc = irq_to_desc(irq);
3936 * Honour affinities which have been set in early boot
3939 (IRQ_NO_BALANCING | IRQ_AFFINITY_SET))
3940 mask = desc->irq_data.affinity;
3942 mask = apic->target_cpus();
3944 if (intr_remapping_enabled)
3945 ir_ioapic_set_affinity(&desc->irq_data, mask, false);
3947 ioapic_set_affinity(&desc->irq_data, mask, false);
3953 #define IOAPIC_RESOURCE_NAME_SIZE 11
3955 static struct resource *ioapic_resources;
3957 static struct resource * __init ioapic_setup_resources(int nr_ioapics)
3960 struct resource *res;
3964 if (nr_ioapics <= 0)
3967 n = IOAPIC_RESOURCE_NAME_SIZE + sizeof(struct resource);
3970 mem = alloc_bootmem(n);
3973 mem += sizeof(struct resource) * nr_ioapics;
3975 for (i = 0; i < nr_ioapics; i++) {
3977 res[i].flags = IORESOURCE_MEM | IORESOURCE_BUSY;
3978 snprintf(mem, IOAPIC_RESOURCE_NAME_SIZE, "IOAPIC %u", i);
3979 mem += IOAPIC_RESOURCE_NAME_SIZE;
3982 ioapic_resources = res;
3987 void __init ioapic_and_gsi_init(void)
3989 unsigned long ioapic_phys, idx = FIX_IO_APIC_BASE_0;
3990 struct resource *ioapic_res;
3993 ioapic_res = ioapic_setup_resources(nr_ioapics);
3994 for (i = 0; i < nr_ioapics; i++) {
3995 if (smp_found_config) {
3996 ioapic_phys = mp_ioapics[i].apicaddr;
3997 #ifdef CONFIG_X86_32
4000 "WARNING: bogus zero IO-APIC "
4001 "address found in MPTABLE, "
4002 "disabling IO/APIC support!\n");
4003 smp_found_config = 0;
4004 skip_ioapic_setup = 1;
4005 goto fake_ioapic_page;
4009 #ifdef CONFIG_X86_32
4012 ioapic_phys = (unsigned long)alloc_bootmem_pages(PAGE_SIZE);
4013 ioapic_phys = __pa(ioapic_phys);
4015 set_fixmap_nocache(idx, ioapic_phys);
4016 apic_printk(APIC_VERBOSE, "mapped IOAPIC to %08lx (%08lx)\n",
4017 __fix_to_virt(idx) + (ioapic_phys & ~PAGE_MASK),
4021 ioapic_res->start = ioapic_phys;
4022 ioapic_res->end = ioapic_phys + IO_APIC_SLOT_SIZE - 1;
4026 probe_nr_irqs_gsi();
4029 void __init ioapic_insert_resources(void)
4032 struct resource *r = ioapic_resources;
4037 "IO APIC resources couldn't be allocated.\n");
4041 for (i = 0; i < nr_ioapics; i++) {
4042 insert_resource(&iomem_resource, r);
4047 int mp_find_ioapic(u32 gsi)
4051 /* Find the IOAPIC that manages this GSI. */
4052 for (i = 0; i < nr_ioapics; i++) {
4053 if ((gsi >= mp_gsi_routing[i].gsi_base)
4054 && (gsi <= mp_gsi_routing[i].gsi_end))
4058 printk(KERN_ERR "ERROR: Unable to locate IOAPIC for GSI %d\n", gsi);
4062 int mp_find_ioapic_pin(int ioapic, u32 gsi)
4064 if (WARN_ON(ioapic == -1))
4066 if (WARN_ON(gsi > mp_gsi_routing[ioapic].gsi_end))
4069 return gsi - mp_gsi_routing[ioapic].gsi_base;
4072 static int bad_ioapic(unsigned long address)
4074 if (nr_ioapics >= MAX_IO_APICS) {
4075 printk(KERN_WARNING "WARING: Max # of I/O APICs (%d) exceeded "
4076 "(found %d), skipping\n", MAX_IO_APICS, nr_ioapics);
4080 printk(KERN_WARNING "WARNING: Bogus (zero) I/O APIC address"
4081 " found in table, skipping!\n");
4087 void __init mp_register_ioapic(int id, u32 address, u32 gsi_base)
4092 if (bad_ioapic(address))
4097 mp_ioapics[idx].type = MP_IOAPIC;
4098 mp_ioapics[idx].flags = MPC_APIC_USABLE;
4099 mp_ioapics[idx].apicaddr = address;
4101 set_fixmap_nocache(FIX_IO_APIC_BASE_0 + idx, address);
4102 mp_ioapics[idx].apicid = io_apic_unique_id(id);
4103 mp_ioapics[idx].apicver = io_apic_get_version(idx);
4106 * Build basic GSI lookup table to facilitate gsi->io_apic lookups
4107 * and to prevent reprogramming of IOAPIC pins (PCI GSIs).
4109 entries = io_apic_get_redir_entries(idx);
4110 mp_gsi_routing[idx].gsi_base = gsi_base;
4111 mp_gsi_routing[idx].gsi_end = gsi_base + entries - 1;
4114 * The number of IO-APIC IRQ registers (== #pins):
4116 nr_ioapic_registers[idx] = entries;
4118 if (mp_gsi_routing[idx].gsi_end >= gsi_top)
4119 gsi_top = mp_gsi_routing[idx].gsi_end + 1;
4121 printk(KERN_INFO "IOAPIC[%d]: apic_id %d, version %d, address 0x%x, "
4122 "GSI %d-%d\n", idx, mp_ioapics[idx].apicid,
4123 mp_ioapics[idx].apicver, mp_ioapics[idx].apicaddr,
4124 mp_gsi_routing[idx].gsi_base, mp_gsi_routing[idx].gsi_end);
4129 /* Enable IOAPIC early just for system timer */
4130 void __init pre_init_apic_IRQ0(void)
4132 struct irq_cfg *cfg;
4134 printk(KERN_INFO "Early APIC setup for system timer0\n");
4136 phys_cpu_present_map = physid_mask_of_physid(boot_cpu_physical_apicid);
4138 /* Make sure the irq descriptor is set up */
4139 cfg = alloc_irq_and_cfg_at(0, 0);
4143 add_pin_to_irq_node(cfg, 0, 0, 0);
4144 set_irq_chip_and_handler_name(0, &ioapic_chip, handle_edge_irq, "edge");
4146 setup_ioapic_irq(0, 0, 0, cfg, 0, 0);