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/syscore_ops.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>
57 #include <asm/msidef.h>
58 #include <asm/hypertransport.h>
59 #include <asm/setup.h>
60 #include <asm/irq_remapping.h>
62 #include <asm/hw_irq.h>
66 #define __apicdebuginit(type) static type __init
67 #define for_each_irq_pin(entry, head) \
68 for (entry = head; entry; entry = entry->next)
71 * Is the SiS APIC rmw bug present ?
72 * -1 = don't know, 0 = no, 1 = yes
74 int sis_apic_bug = -1;
76 static DEFINE_RAW_SPINLOCK(ioapic_lock);
77 static DEFINE_RAW_SPINLOCK(vector_lock);
79 static struct ioapic {
81 * # of IRQ routing registers
85 * Saved state during suspend/resume, or while enabling intr-remap.
87 struct IO_APIC_route_entry *saved_registers;
89 struct mpc_ioapic mp_config;
90 /* IO APIC gsi routing info */
91 struct mp_ioapic_gsi gsi_config;
92 DECLARE_BITMAP(pin_programmed, MP_MAX_IOAPIC_PIN + 1);
93 } ioapics[MAX_IO_APICS];
95 #define mpc_ioapic_ver(id) ioapics[id].mp_config.apicver
97 int mpc_ioapic_id(int id)
99 return ioapics[id].mp_config.apicid;
102 unsigned int mpc_ioapic_addr(int id)
104 return ioapics[id].mp_config.apicaddr;
107 struct mp_ioapic_gsi *mp_ioapic_gsi_routing(int id)
109 return &ioapics[id].gsi_config;
114 /* The one past the highest gsi number used */
117 /* MP IRQ source entries */
118 struct mpc_intsrc mp_irqs[MAX_IRQ_SOURCES];
120 /* # of MP IRQ source entries */
124 static int nr_irqs_gsi = NR_IRQS_LEGACY;
126 #if defined (CONFIG_MCA) || defined (CONFIG_EISA)
127 int mp_bus_id_to_type[MAX_MP_BUSSES];
130 DECLARE_BITMAP(mp_bus_not_pci, MAX_MP_BUSSES);
132 int skip_ioapic_setup;
135 * disable_ioapic_support() - disables ioapic support at runtime
137 void disable_ioapic_support(void)
141 noioapicreroute = -1;
143 skip_ioapic_setup = 1;
146 static int __init parse_noapic(char *str)
148 /* disable IO-APIC */
149 disable_ioapic_support();
152 early_param("noapic", parse_noapic);
154 static int io_apic_setup_irq_pin(unsigned int irq, int node,
155 struct io_apic_irq_attr *attr);
157 /* Will be called in mpparse/acpi/sfi codes for saving IRQ info */
158 void mp_save_irq(struct mpc_intsrc *m)
162 apic_printk(APIC_VERBOSE, "Int: type %d, pol %d, trig %d, bus %02x,"
163 " IRQ %02x, APIC ID %x, APIC INT %02x\n",
164 m->irqtype, m->irqflag & 3, (m->irqflag >> 2) & 3, m->srcbus,
165 m->srcbusirq, m->dstapic, m->dstirq);
167 for (i = 0; i < mp_irq_entries; i++) {
168 if (!memcmp(&mp_irqs[i], m, sizeof(*m)))
172 memcpy(&mp_irqs[mp_irq_entries], m, sizeof(*m));
173 if (++mp_irq_entries == MAX_IRQ_SOURCES)
174 panic("Max # of irq sources exceeded!!\n");
177 struct irq_pin_list {
179 struct irq_pin_list *next;
182 static struct irq_pin_list *alloc_irq_pin_list(int node)
184 return kzalloc_node(sizeof(struct irq_pin_list), GFP_KERNEL, node);
188 /* irq_cfg is indexed by the sum of all RTEs in all I/O APICs. */
189 #ifdef CONFIG_SPARSE_IRQ
190 static struct irq_cfg irq_cfgx[NR_IRQS_LEGACY];
192 static struct irq_cfg irq_cfgx[NR_IRQS];
195 int __init arch_early_irq_init(void)
200 if (!legacy_pic->nr_legacy_irqs) {
205 for (i = 0; i < nr_ioapics; i++) {
206 ioapics[i].saved_registers =
207 kzalloc(sizeof(struct IO_APIC_route_entry) *
208 ioapics[i].nr_registers, GFP_KERNEL);
209 if (!ioapics[i].saved_registers)
210 pr_err("IOAPIC %d: suspend/resume impossible!\n", i);
214 count = ARRAY_SIZE(irq_cfgx);
215 node = cpu_to_node(0);
217 /* Make sure the legacy interrupts are marked in the bitmap */
218 irq_reserve_irqs(0, legacy_pic->nr_legacy_irqs);
220 for (i = 0; i < count; i++) {
221 irq_set_chip_data(i, &cfg[i]);
222 zalloc_cpumask_var_node(&cfg[i].domain, GFP_KERNEL, node);
223 zalloc_cpumask_var_node(&cfg[i].old_domain, GFP_KERNEL, node);
225 * For legacy IRQ's, start with assigning irq0 to irq15 to
226 * IRQ0_VECTOR to IRQ15_VECTOR on cpu 0.
228 if (i < legacy_pic->nr_legacy_irqs) {
229 cfg[i].vector = IRQ0_VECTOR + i;
230 cpumask_set_cpu(0, cfg[i].domain);
237 #ifdef CONFIG_SPARSE_IRQ
238 static struct irq_cfg *irq_cfg(unsigned int irq)
240 return irq_get_chip_data(irq);
243 static struct irq_cfg *alloc_irq_cfg(unsigned int irq, int node)
247 cfg = kzalloc_node(sizeof(*cfg), GFP_KERNEL, node);
250 if (!zalloc_cpumask_var_node(&cfg->domain, GFP_KERNEL, node))
252 if (!zalloc_cpumask_var_node(&cfg->old_domain, GFP_KERNEL, node))
256 free_cpumask_var(cfg->domain);
262 static void free_irq_cfg(unsigned int at, struct irq_cfg *cfg)
266 irq_set_chip_data(at, NULL);
267 free_cpumask_var(cfg->domain);
268 free_cpumask_var(cfg->old_domain);
274 struct irq_cfg *irq_cfg(unsigned int irq)
276 return irq < nr_irqs ? irq_cfgx + irq : NULL;
279 static struct irq_cfg *alloc_irq_cfg(unsigned int irq, int node)
281 return irq_cfgx + irq;
284 static inline void free_irq_cfg(unsigned int at, struct irq_cfg *cfg) { }
288 static struct irq_cfg *alloc_irq_and_cfg_at(unsigned int at, int node)
290 int res = irq_alloc_desc_at(at, node);
296 cfg = irq_get_chip_data(at);
301 cfg = alloc_irq_cfg(at, node);
303 irq_set_chip_data(at, cfg);
309 static int alloc_irq_from(unsigned int from, int node)
311 return irq_alloc_desc_from(from, node);
314 static void free_irq_at(unsigned int at, struct irq_cfg *cfg)
316 free_irq_cfg(at, cfg);
322 unsigned int unused[3];
324 unsigned int unused2[11];
328 static __attribute_const__ struct io_apic __iomem *io_apic_base(int idx)
330 return (void __iomem *) __fix_to_virt(FIX_IO_APIC_BASE_0 + idx)
331 + (mpc_ioapic_addr(idx) & ~PAGE_MASK);
334 static inline void io_apic_eoi(unsigned int apic, unsigned int vector)
336 struct io_apic __iomem *io_apic = io_apic_base(apic);
337 writel(vector, &io_apic->eoi);
340 static inline unsigned int io_apic_read(unsigned int apic, unsigned int reg)
342 struct io_apic __iomem *io_apic = io_apic_base(apic);
343 writel(reg, &io_apic->index);
344 return readl(&io_apic->data);
347 static inline void io_apic_write(unsigned int apic, unsigned int reg, unsigned int value)
349 struct io_apic __iomem *io_apic = io_apic_base(apic);
350 writel(reg, &io_apic->index);
351 writel(value, &io_apic->data);
355 * Re-write a value: to be used for read-modify-write
356 * cycles where the read already set up the index register.
358 * Older SiS APIC requires we rewrite the index register
360 static inline void io_apic_modify(unsigned int apic, unsigned int reg, unsigned int value)
362 struct io_apic __iomem *io_apic = io_apic_base(apic);
365 writel(reg, &io_apic->index);
366 writel(value, &io_apic->data);
369 static bool io_apic_level_ack_pending(struct irq_cfg *cfg)
371 struct irq_pin_list *entry;
374 raw_spin_lock_irqsave(&ioapic_lock, flags);
375 for_each_irq_pin(entry, cfg->irq_2_pin) {
380 reg = io_apic_read(entry->apic, 0x10 + pin*2);
381 /* Is the remote IRR bit set? */
382 if (reg & IO_APIC_REDIR_REMOTE_IRR) {
383 raw_spin_unlock_irqrestore(&ioapic_lock, flags);
387 raw_spin_unlock_irqrestore(&ioapic_lock, flags);
393 struct { u32 w1, w2; };
394 struct IO_APIC_route_entry entry;
397 static struct IO_APIC_route_entry ioapic_read_entry(int apic, int pin)
399 union entry_union eu;
401 raw_spin_lock_irqsave(&ioapic_lock, flags);
402 eu.w1 = io_apic_read(apic, 0x10 + 2 * pin);
403 eu.w2 = io_apic_read(apic, 0x11 + 2 * pin);
404 raw_spin_unlock_irqrestore(&ioapic_lock, flags);
409 * When we write a new IO APIC routing entry, we need to write the high
410 * word first! If the mask bit in the low word is clear, we will enable
411 * the interrupt, and we need to make sure the entry is fully populated
412 * before that happens.
415 __ioapic_write_entry(int apic, int pin, struct IO_APIC_route_entry e)
417 union entry_union eu = {{0, 0}};
420 io_apic_write(apic, 0x11 + 2*pin, eu.w2);
421 io_apic_write(apic, 0x10 + 2*pin, eu.w1);
424 static void ioapic_write_entry(int apic, int pin, struct IO_APIC_route_entry e)
427 raw_spin_lock_irqsave(&ioapic_lock, flags);
428 __ioapic_write_entry(apic, pin, e);
429 raw_spin_unlock_irqrestore(&ioapic_lock, flags);
433 * When we mask an IO APIC routing entry, we need to write the low
434 * word first, in order to set the mask bit before we change the
437 static void ioapic_mask_entry(int apic, int pin)
440 union entry_union eu = { .entry.mask = 1 };
442 raw_spin_lock_irqsave(&ioapic_lock, flags);
443 io_apic_write(apic, 0x10 + 2*pin, eu.w1);
444 io_apic_write(apic, 0x11 + 2*pin, eu.w2);
445 raw_spin_unlock_irqrestore(&ioapic_lock, flags);
449 * The common case is 1:1 IRQ<->pin mappings. Sometimes there are
450 * shared ISA-space IRQs, so we have to support them. We are super
451 * fast in the common case, and fast for shared ISA-space IRQs.
454 __add_pin_to_irq_node(struct irq_cfg *cfg, int node, int apic, int pin)
456 struct irq_pin_list **last, *entry;
458 /* don't allow duplicates */
459 last = &cfg->irq_2_pin;
460 for_each_irq_pin(entry, cfg->irq_2_pin) {
461 if (entry->apic == apic && entry->pin == pin)
466 entry = alloc_irq_pin_list(node);
468 printk(KERN_ERR "can not alloc irq_pin_list (%d,%d,%d)\n",
479 static void add_pin_to_irq_node(struct irq_cfg *cfg, int node, int apic, int pin)
481 if (__add_pin_to_irq_node(cfg, node, apic, pin))
482 panic("IO-APIC: failed to add irq-pin. Can not proceed\n");
486 * Reroute an IRQ to a different pin.
488 static void __init replace_pin_at_irq_node(struct irq_cfg *cfg, int node,
489 int oldapic, int oldpin,
490 int newapic, int newpin)
492 struct irq_pin_list *entry;
494 for_each_irq_pin(entry, cfg->irq_2_pin) {
495 if (entry->apic == oldapic && entry->pin == oldpin) {
496 entry->apic = newapic;
498 /* every one is different, right? */
503 /* old apic/pin didn't exist, so just add new ones */
504 add_pin_to_irq_node(cfg, node, newapic, newpin);
507 static void __io_apic_modify_irq(struct irq_pin_list *entry,
508 int mask_and, int mask_or,
509 void (*final)(struct irq_pin_list *entry))
511 unsigned int reg, pin;
514 reg = io_apic_read(entry->apic, 0x10 + pin * 2);
517 io_apic_modify(entry->apic, 0x10 + pin * 2, reg);
522 static void io_apic_modify_irq(struct irq_cfg *cfg,
523 int mask_and, int mask_or,
524 void (*final)(struct irq_pin_list *entry))
526 struct irq_pin_list *entry;
528 for_each_irq_pin(entry, cfg->irq_2_pin)
529 __io_apic_modify_irq(entry, mask_and, mask_or, final);
532 static void __mask_and_edge_IO_APIC_irq(struct irq_pin_list *entry)
534 __io_apic_modify_irq(entry, ~IO_APIC_REDIR_LEVEL_TRIGGER,
535 IO_APIC_REDIR_MASKED, NULL);
538 static void __unmask_and_level_IO_APIC_irq(struct irq_pin_list *entry)
540 __io_apic_modify_irq(entry, ~IO_APIC_REDIR_MASKED,
541 IO_APIC_REDIR_LEVEL_TRIGGER, NULL);
544 static void io_apic_sync(struct irq_pin_list *entry)
547 * Synchronize the IO-APIC and the CPU by doing
548 * a dummy read from the IO-APIC
550 struct io_apic __iomem *io_apic;
551 io_apic = io_apic_base(entry->apic);
552 readl(&io_apic->data);
555 static void mask_ioapic(struct irq_cfg *cfg)
559 raw_spin_lock_irqsave(&ioapic_lock, flags);
560 io_apic_modify_irq(cfg, ~0, IO_APIC_REDIR_MASKED, &io_apic_sync);
561 raw_spin_unlock_irqrestore(&ioapic_lock, flags);
564 static void mask_ioapic_irq(struct irq_data *data)
566 mask_ioapic(data->chip_data);
569 static void __unmask_ioapic(struct irq_cfg *cfg)
571 io_apic_modify_irq(cfg, ~IO_APIC_REDIR_MASKED, 0, NULL);
574 static void unmask_ioapic(struct irq_cfg *cfg)
578 raw_spin_lock_irqsave(&ioapic_lock, flags);
579 __unmask_ioapic(cfg);
580 raw_spin_unlock_irqrestore(&ioapic_lock, flags);
583 static void unmask_ioapic_irq(struct irq_data *data)
585 unmask_ioapic(data->chip_data);
588 static void clear_IO_APIC_pin(unsigned int apic, unsigned int pin)
590 struct IO_APIC_route_entry entry;
592 /* Check delivery_mode to be sure we're not clearing an SMI pin */
593 entry = ioapic_read_entry(apic, pin);
594 if (entry.delivery_mode == dest_SMI)
597 * Disable it in the IO-APIC irq-routing table:
599 ioapic_mask_entry(apic, pin);
602 static void clear_IO_APIC (void)
606 for (apic = 0; apic < nr_ioapics; apic++)
607 for (pin = 0; pin < ioapics[apic].nr_registers; pin++)
608 clear_IO_APIC_pin(apic, pin);
613 * support for broken MP BIOSs, enables hand-redirection of PIRQ0-7 to
614 * specific CPU-side IRQs.
618 static int pirq_entries[MAX_PIRQS] = {
619 [0 ... MAX_PIRQS - 1] = -1
622 static int __init ioapic_pirq_setup(char *str)
625 int ints[MAX_PIRQS+1];
627 get_options(str, ARRAY_SIZE(ints), ints);
629 apic_printk(APIC_VERBOSE, KERN_INFO
630 "PIRQ redirection, working around broken MP-BIOS.\n");
632 if (ints[0] < MAX_PIRQS)
635 for (i = 0; i < max; i++) {
636 apic_printk(APIC_VERBOSE, KERN_DEBUG
637 "... PIRQ%d -> IRQ %d\n", i, ints[i+1]);
639 * PIRQs are mapped upside down, usually.
641 pirq_entries[MAX_PIRQS-i-1] = ints[i+1];
646 __setup("pirq=", ioapic_pirq_setup);
647 #endif /* CONFIG_X86_32 */
650 * Saves all the IO-APIC RTE's
652 int save_ioapic_entries(void)
657 for (apic = 0; apic < nr_ioapics; apic++) {
658 if (!ioapics[apic].saved_registers) {
663 for (pin = 0; pin < ioapics[apic].nr_registers; pin++)
664 ioapics[apic].saved_registers[pin] =
665 ioapic_read_entry(apic, pin);
672 * Mask all IO APIC entries.
674 void mask_ioapic_entries(void)
678 for (apic = 0; apic < nr_ioapics; apic++) {
679 if (!ioapics[apic].saved_registers)
682 for (pin = 0; pin < ioapics[apic].nr_registers; pin++) {
683 struct IO_APIC_route_entry entry;
685 entry = ioapics[apic].saved_registers[pin];
688 ioapic_write_entry(apic, pin, entry);
695 * Restore IO APIC entries which was saved in the ioapic structure.
697 int restore_ioapic_entries(void)
701 for (apic = 0; apic < nr_ioapics; apic++) {
702 if (!ioapics[apic].saved_registers)
705 for (pin = 0; pin < ioapics[apic].nr_registers; pin++)
706 ioapic_write_entry(apic, pin,
707 ioapics[apic].saved_registers[pin]);
713 * Find the IRQ entry number of a certain pin.
715 static int find_irq_entry(int apic, int pin, int type)
719 for (i = 0; i < mp_irq_entries; i++)
720 if (mp_irqs[i].irqtype == type &&
721 (mp_irqs[i].dstapic == mpc_ioapic_id(apic) ||
722 mp_irqs[i].dstapic == MP_APIC_ALL) &&
723 mp_irqs[i].dstirq == pin)
730 * Find the pin to which IRQ[irq] (ISA) is connected
732 static int __init find_isa_irq_pin(int irq, int type)
736 for (i = 0; i < mp_irq_entries; i++) {
737 int lbus = mp_irqs[i].srcbus;
739 if (test_bit(lbus, mp_bus_not_pci) &&
740 (mp_irqs[i].irqtype == type) &&
741 (mp_irqs[i].srcbusirq == irq))
743 return mp_irqs[i].dstirq;
748 static int __init find_isa_irq_apic(int irq, int type)
752 for (i = 0; i < mp_irq_entries; i++) {
753 int lbus = mp_irqs[i].srcbus;
755 if (test_bit(lbus, mp_bus_not_pci) &&
756 (mp_irqs[i].irqtype == type) &&
757 (mp_irqs[i].srcbusirq == irq))
760 if (i < mp_irq_entries) {
762 for(apic = 0; apic < nr_ioapics; apic++) {
763 if (mpc_ioapic_id(apic) == mp_irqs[i].dstapic)
771 #if defined(CONFIG_EISA) || defined(CONFIG_MCA)
773 * EISA Edge/Level control register, ELCR
775 static int EISA_ELCR(unsigned int irq)
777 if (irq < legacy_pic->nr_legacy_irqs) {
778 unsigned int port = 0x4d0 + (irq >> 3);
779 return (inb(port) >> (irq & 7)) & 1;
781 apic_printk(APIC_VERBOSE, KERN_INFO
782 "Broken MPtable reports ISA irq %d\n", irq);
788 /* ISA interrupts are always polarity zero edge triggered,
789 * when listed as conforming in the MP table. */
791 #define default_ISA_trigger(idx) (0)
792 #define default_ISA_polarity(idx) (0)
794 /* EISA interrupts are always polarity zero and can be edge or level
795 * trigger depending on the ELCR value. If an interrupt is listed as
796 * EISA conforming in the MP table, that means its trigger type must
797 * be read in from the ELCR */
799 #define default_EISA_trigger(idx) (EISA_ELCR(mp_irqs[idx].srcbusirq))
800 #define default_EISA_polarity(idx) default_ISA_polarity(idx)
802 /* PCI interrupts are always polarity one level triggered,
803 * when listed as conforming in the MP table. */
805 #define default_PCI_trigger(idx) (1)
806 #define default_PCI_polarity(idx) (1)
808 /* MCA interrupts are always polarity zero level triggered,
809 * when listed as conforming in the MP table. */
811 #define default_MCA_trigger(idx) (1)
812 #define default_MCA_polarity(idx) default_ISA_polarity(idx)
814 static int irq_polarity(int idx)
816 int bus = mp_irqs[idx].srcbus;
820 * Determine IRQ line polarity (high active or low active):
822 switch (mp_irqs[idx].irqflag & 3)
824 case 0: /* conforms, ie. bus-type dependent polarity */
825 if (test_bit(bus, mp_bus_not_pci))
826 polarity = default_ISA_polarity(idx);
828 polarity = default_PCI_polarity(idx);
830 case 1: /* high active */
835 case 2: /* reserved */
837 printk(KERN_WARNING "broken BIOS!!\n");
841 case 3: /* low active */
846 default: /* invalid */
848 printk(KERN_WARNING "broken BIOS!!\n");
856 static int irq_trigger(int idx)
858 int bus = mp_irqs[idx].srcbus;
862 * Determine IRQ trigger mode (edge or level sensitive):
864 switch ((mp_irqs[idx].irqflag>>2) & 3)
866 case 0: /* conforms, ie. bus-type dependent */
867 if (test_bit(bus, mp_bus_not_pci))
868 trigger = default_ISA_trigger(idx);
870 trigger = default_PCI_trigger(idx);
871 #if defined(CONFIG_EISA) || defined(CONFIG_MCA)
872 switch (mp_bus_id_to_type[bus]) {
873 case MP_BUS_ISA: /* ISA pin */
875 /* set before the switch */
878 case MP_BUS_EISA: /* EISA pin */
880 trigger = default_EISA_trigger(idx);
883 case MP_BUS_PCI: /* PCI pin */
885 /* set before the switch */
888 case MP_BUS_MCA: /* MCA pin */
890 trigger = default_MCA_trigger(idx);
895 printk(KERN_WARNING "broken BIOS!!\n");
907 case 2: /* reserved */
909 printk(KERN_WARNING "broken BIOS!!\n");
918 default: /* invalid */
920 printk(KERN_WARNING "broken BIOS!!\n");
928 static int pin_2_irq(int idx, int apic, int pin)
931 int bus = mp_irqs[idx].srcbus;
932 struct mp_ioapic_gsi *gsi_cfg = mp_ioapic_gsi_routing(apic);
935 * Debugging check, we are in big trouble if this message pops up!
937 if (mp_irqs[idx].dstirq != pin)
938 printk(KERN_ERR "broken BIOS or MPTABLE parser, ayiee!!\n");
940 if (test_bit(bus, mp_bus_not_pci)) {
941 irq = mp_irqs[idx].srcbusirq;
943 u32 gsi = gsi_cfg->gsi_base + pin;
945 if (gsi >= NR_IRQS_LEGACY)
953 * PCI IRQ command line redirection. Yes, limits are hardcoded.
955 if ((pin >= 16) && (pin <= 23)) {
956 if (pirq_entries[pin-16] != -1) {
957 if (!pirq_entries[pin-16]) {
958 apic_printk(APIC_VERBOSE, KERN_DEBUG
959 "disabling PIRQ%d\n", pin-16);
961 irq = pirq_entries[pin-16];
962 apic_printk(APIC_VERBOSE, KERN_DEBUG
963 "using PIRQ%d -> IRQ %d\n",
974 * Find a specific PCI IRQ entry.
975 * Not an __init, possibly needed by modules
977 int IO_APIC_get_PCI_irq_vector(int bus, int slot, int pin,
978 struct io_apic_irq_attr *irq_attr)
980 int apic, i, best_guess = -1;
982 apic_printk(APIC_DEBUG,
983 "querying PCI -> IRQ mapping bus:%d, slot:%d, pin:%d.\n",
985 if (test_bit(bus, mp_bus_not_pci)) {
986 apic_printk(APIC_VERBOSE,
987 "PCI BIOS passed nonexistent PCI bus %d!\n", bus);
990 for (i = 0; i < mp_irq_entries; i++) {
991 int lbus = mp_irqs[i].srcbus;
993 for (apic = 0; apic < nr_ioapics; apic++)
994 if (mpc_ioapic_id(apic) == mp_irqs[i].dstapic ||
995 mp_irqs[i].dstapic == MP_APIC_ALL)
998 if (!test_bit(lbus, mp_bus_not_pci) &&
999 !mp_irqs[i].irqtype &&
1001 (slot == ((mp_irqs[i].srcbusirq >> 2) & 0x1f))) {
1002 int irq = pin_2_irq(i, apic, mp_irqs[i].dstirq);
1004 if (!(apic || IO_APIC_IRQ(irq)))
1007 if (pin == (mp_irqs[i].srcbusirq & 3)) {
1008 set_io_apic_irq_attr(irq_attr, apic,
1015 * Use the first all-but-pin matching entry as a
1016 * best-guess fuzzy result for broken mptables.
1018 if (best_guess < 0) {
1019 set_io_apic_irq_attr(irq_attr, apic,
1029 EXPORT_SYMBOL(IO_APIC_get_PCI_irq_vector);
1031 void lock_vector_lock(void)
1033 /* Used to the online set of cpus does not change
1034 * during assign_irq_vector.
1036 raw_spin_lock(&vector_lock);
1039 void unlock_vector_lock(void)
1041 raw_spin_unlock(&vector_lock);
1045 __assign_irq_vector(int irq, struct irq_cfg *cfg, const struct cpumask *mask)
1048 * NOTE! The local APIC isn't very good at handling
1049 * multiple interrupts at the same interrupt level.
1050 * As the interrupt level is determined by taking the
1051 * vector number and shifting that right by 4, we
1052 * want to spread these out a bit so that they don't
1053 * all fall in the same interrupt level.
1055 * Also, we've got to be careful not to trash gate
1056 * 0x80, because int 0x80 is hm, kind of importantish. ;)
1058 static int current_vector = FIRST_EXTERNAL_VECTOR + VECTOR_OFFSET_START;
1059 static int current_offset = VECTOR_OFFSET_START % 8;
1060 unsigned int old_vector;
1062 cpumask_var_t tmp_mask;
1064 if (cfg->move_in_progress)
1067 if (!alloc_cpumask_var(&tmp_mask, GFP_ATOMIC))
1070 old_vector = cfg->vector;
1072 cpumask_and(tmp_mask, mask, cpu_online_mask);
1073 cpumask_and(tmp_mask, cfg->domain, tmp_mask);
1074 if (!cpumask_empty(tmp_mask)) {
1075 free_cpumask_var(tmp_mask);
1080 /* Only try and allocate irqs on cpus that are present */
1082 for_each_cpu_and(cpu, mask, cpu_online_mask) {
1086 apic->vector_allocation_domain(cpu, tmp_mask);
1088 vector = current_vector;
1089 offset = current_offset;
1092 if (vector >= first_system_vector) {
1093 /* If out of vectors on large boxen, must share them. */
1094 offset = (offset + 1) % 8;
1095 vector = FIRST_EXTERNAL_VECTOR + offset;
1097 if (unlikely(current_vector == vector))
1100 if (test_bit(vector, used_vectors))
1103 for_each_cpu_and(new_cpu, tmp_mask, cpu_online_mask)
1104 if (per_cpu(vector_irq, new_cpu)[vector] != -1)
1107 current_vector = vector;
1108 current_offset = offset;
1110 cfg->move_in_progress = 1;
1111 cpumask_copy(cfg->old_domain, cfg->domain);
1113 for_each_cpu_and(new_cpu, tmp_mask, cpu_online_mask)
1114 per_cpu(vector_irq, new_cpu)[vector] = irq;
1115 cfg->vector = vector;
1116 cpumask_copy(cfg->domain, tmp_mask);
1120 free_cpumask_var(tmp_mask);
1124 int assign_irq_vector(int irq, struct irq_cfg *cfg, const struct cpumask *mask)
1127 unsigned long flags;
1129 raw_spin_lock_irqsave(&vector_lock, flags);
1130 err = __assign_irq_vector(irq, cfg, mask);
1131 raw_spin_unlock_irqrestore(&vector_lock, flags);
1135 static void __clear_irq_vector(int irq, struct irq_cfg *cfg)
1139 BUG_ON(!cfg->vector);
1141 vector = cfg->vector;
1142 for_each_cpu_and(cpu, cfg->domain, cpu_online_mask)
1143 per_cpu(vector_irq, cpu)[vector] = -1;
1146 cpumask_clear(cfg->domain);
1148 if (likely(!cfg->move_in_progress))
1150 for_each_cpu_and(cpu, cfg->old_domain, cpu_online_mask) {
1151 for (vector = FIRST_EXTERNAL_VECTOR; vector < NR_VECTORS;
1153 if (per_cpu(vector_irq, cpu)[vector] != irq)
1155 per_cpu(vector_irq, cpu)[vector] = -1;
1159 cfg->move_in_progress = 0;
1162 void __setup_vector_irq(int cpu)
1164 /* Initialize vector_irq on a new cpu */
1166 struct irq_cfg *cfg;
1169 * vector_lock will make sure that we don't run into irq vector
1170 * assignments that might be happening on another cpu in parallel,
1171 * while we setup our initial vector to irq mappings.
1173 raw_spin_lock(&vector_lock);
1174 /* Mark the inuse vectors */
1175 for_each_active_irq(irq) {
1176 cfg = irq_get_chip_data(irq);
1180 * If it is a legacy IRQ handled by the legacy PIC, this cpu
1181 * will be part of the irq_cfg's domain.
1183 if (irq < legacy_pic->nr_legacy_irqs && !IO_APIC_IRQ(irq))
1184 cpumask_set_cpu(cpu, cfg->domain);
1186 if (!cpumask_test_cpu(cpu, cfg->domain))
1188 vector = cfg->vector;
1189 per_cpu(vector_irq, cpu)[vector] = irq;
1191 /* Mark the free vectors */
1192 for (vector = 0; vector < NR_VECTORS; ++vector) {
1193 irq = per_cpu(vector_irq, cpu)[vector];
1198 if (!cpumask_test_cpu(cpu, cfg->domain))
1199 per_cpu(vector_irq, cpu)[vector] = -1;
1201 raw_spin_unlock(&vector_lock);
1204 static struct irq_chip ioapic_chip;
1205 static struct irq_chip ir_ioapic_chip;
1207 #ifdef CONFIG_X86_32
1208 static inline int IO_APIC_irq_trigger(int irq)
1212 for (apic = 0; apic < nr_ioapics; apic++) {
1213 for (pin = 0; pin < ioapics[apic].nr_registers; pin++) {
1214 idx = find_irq_entry(apic, pin, mp_INT);
1215 if ((idx != -1) && (irq == pin_2_irq(idx, apic, pin)))
1216 return irq_trigger(idx);
1220 * nonexistent IRQs are edge default
1225 static inline int IO_APIC_irq_trigger(int irq)
1231 static void ioapic_register_intr(unsigned int irq, struct irq_cfg *cfg,
1232 unsigned long trigger)
1234 struct irq_chip *chip = &ioapic_chip;
1235 irq_flow_handler_t hdl;
1238 if ((trigger == IOAPIC_AUTO && IO_APIC_irq_trigger(irq)) ||
1239 trigger == IOAPIC_LEVEL) {
1240 irq_set_status_flags(irq, IRQ_LEVEL);
1243 irq_clear_status_flags(irq, IRQ_LEVEL);
1247 if (irq_remapped(cfg)) {
1248 irq_set_status_flags(irq, IRQ_MOVE_PCNTXT);
1249 chip = &ir_ioapic_chip;
1250 fasteoi = trigger != 0;
1253 hdl = fasteoi ? handle_fasteoi_irq : handle_edge_irq;
1254 irq_set_chip_and_handler_name(irq, chip, hdl,
1255 fasteoi ? "fasteoi" : "edge");
1258 static int setup_ioapic_entry(int apic_id, int irq,
1259 struct IO_APIC_route_entry *entry,
1260 unsigned int destination, int trigger,
1261 int polarity, int vector, int pin)
1264 * add it to the IO-APIC irq-routing table:
1266 memset(entry,0,sizeof(*entry));
1268 if (intr_remapping_enabled) {
1269 struct intel_iommu *iommu = map_ioapic_to_ir(apic_id);
1271 struct IR_IO_APIC_route_entry *ir_entry =
1272 (struct IR_IO_APIC_route_entry *) entry;
1276 panic("No mapping iommu for ioapic %d\n", apic_id);
1278 index = alloc_irte(iommu, irq, 1);
1280 panic("Failed to allocate IRTE for ioapic %d\n", apic_id);
1282 prepare_irte(&irte, vector, destination);
1284 /* Set source-id of interrupt request */
1285 set_ioapic_sid(&irte, apic_id);
1287 modify_irte(irq, &irte);
1289 ir_entry->index2 = (index >> 15) & 0x1;
1291 ir_entry->format = 1;
1292 ir_entry->index = (index & 0x7fff);
1294 * IO-APIC RTE will be configured with virtual vector.
1295 * irq handler will do the explicit EOI to the io-apic.
1297 ir_entry->vector = pin;
1299 apic_printk(APIC_VERBOSE, KERN_DEBUG "IOAPIC[%d]: "
1300 "Set IRTE entry (P:%d FPD:%d Dst_Mode:%d "
1301 "Redir_hint:%d Trig_Mode:%d Dlvry_Mode:%X "
1302 "Avail:%X Vector:%02X Dest:%08X "
1303 "SID:%04X SQ:%X SVT:%X)\n",
1304 apic_id, irte.present, irte.fpd, irte.dst_mode,
1305 irte.redir_hint, irte.trigger_mode, irte.dlvry_mode,
1306 irte.avail, irte.vector, irte.dest_id,
1307 irte.sid, irte.sq, irte.svt);
1309 entry->delivery_mode = apic->irq_delivery_mode;
1310 entry->dest_mode = apic->irq_dest_mode;
1311 entry->dest = destination;
1312 entry->vector = vector;
1315 entry->mask = 0; /* enable IRQ */
1316 entry->trigger = trigger;
1317 entry->polarity = polarity;
1319 /* Mask level triggered irqs.
1320 * Use IRQ_DELAYED_DISABLE for edge triggered irqs.
1327 static void setup_ioapic_irq(unsigned int irq, struct irq_cfg *cfg,
1328 struct io_apic_irq_attr *attr)
1330 struct IO_APIC_route_entry entry;
1333 if (!IO_APIC_IRQ(irq))
1336 * For legacy irqs, cfg->domain starts with cpu 0 for legacy
1337 * controllers like 8259. Now that IO-APIC can handle this irq, update
1340 if (irq < legacy_pic->nr_legacy_irqs && cpumask_test_cpu(0, cfg->domain))
1341 apic->vector_allocation_domain(0, cfg->domain);
1343 if (assign_irq_vector(irq, cfg, apic->target_cpus()))
1346 dest = apic->cpu_mask_to_apicid_and(cfg->domain, apic->target_cpus());
1348 apic_printk(APIC_VERBOSE,KERN_DEBUG
1349 "IOAPIC[%d]: Set routing entry (%d-%d -> 0x%x -> "
1350 "IRQ %d Mode:%i Active:%i Dest:%d)\n",
1351 attr->ioapic, mpc_ioapic_id(attr->ioapic), attr->ioapic_pin,
1352 cfg->vector, irq, attr->trigger, attr->polarity, dest);
1355 if (setup_ioapic_entry(mpc_ioapic_id(attr->ioapic), irq, &entry,
1356 dest, attr->trigger, attr->polarity, cfg->vector,
1357 attr->ioapic_pin)) {
1358 printk("Failed to setup ioapic entry for ioapic %d, pin %d\n",
1359 mpc_ioapic_id(attr->ioapic), attr->ioapic_pin);
1360 __clear_irq_vector(irq, cfg);
1364 ioapic_register_intr(irq, cfg, attr->trigger);
1365 if (irq < legacy_pic->nr_legacy_irqs)
1366 legacy_pic->mask(irq);
1368 ioapic_write_entry(attr->ioapic, attr->ioapic_pin, entry);
1371 static bool __init io_apic_pin_not_connected(int idx, int apic_id, int pin)
1376 apic_printk(APIC_VERBOSE, KERN_DEBUG " apic %d pin %d not connected\n",
1377 mpc_ioapic_id(apic_id), pin);
1381 static void __init __io_apic_setup_irqs(unsigned int apic_id)
1383 int idx, node = cpu_to_node(0);
1384 struct io_apic_irq_attr attr;
1385 unsigned int pin, irq;
1387 for (pin = 0; pin < ioapics[apic_id].nr_registers; pin++) {
1388 idx = find_irq_entry(apic_id, pin, mp_INT);
1389 if (io_apic_pin_not_connected(idx, apic_id, pin))
1392 irq = pin_2_irq(idx, apic_id, pin);
1394 if ((apic_id > 0) && (irq > 16))
1398 * Skip the timer IRQ if there's a quirk handler
1399 * installed and if it returns 1:
1401 if (apic->multi_timer_check &&
1402 apic->multi_timer_check(apic_id, irq))
1405 set_io_apic_irq_attr(&attr, apic_id, pin, irq_trigger(idx),
1408 io_apic_setup_irq_pin(irq, node, &attr);
1412 static void __init setup_IO_APIC_irqs(void)
1414 unsigned int apic_id;
1416 apic_printk(APIC_VERBOSE, KERN_DEBUG "init IO_APIC IRQs\n");
1418 for (apic_id = 0; apic_id < nr_ioapics; apic_id++)
1419 __io_apic_setup_irqs(apic_id);
1423 * for the gsit that is not in first ioapic
1424 * but could not use acpi_register_gsi()
1425 * like some special sci in IBM x3330
1427 void setup_IO_APIC_irq_extra(u32 gsi)
1429 int apic_id = 0, pin, idx, irq, node = cpu_to_node(0);
1430 struct io_apic_irq_attr attr;
1433 * Convert 'gsi' to 'ioapic.pin'.
1435 apic_id = mp_find_ioapic(gsi);
1439 pin = mp_find_ioapic_pin(apic_id, gsi);
1440 idx = find_irq_entry(apic_id, pin, mp_INT);
1444 irq = pin_2_irq(idx, apic_id, pin);
1446 /* Only handle the non legacy irqs on secondary ioapics */
1447 if (apic_id == 0 || irq < NR_IRQS_LEGACY)
1450 set_io_apic_irq_attr(&attr, apic_id, pin, irq_trigger(idx),
1453 io_apic_setup_irq_pin_once(irq, node, &attr);
1457 * Set up the timer pin, possibly with the 8259A-master behind.
1459 static void __init setup_timer_IRQ0_pin(unsigned int apic_id, unsigned int pin,
1462 struct IO_APIC_route_entry entry;
1464 if (intr_remapping_enabled)
1467 memset(&entry, 0, sizeof(entry));
1470 * We use logical delivery to get the timer IRQ
1473 entry.dest_mode = apic->irq_dest_mode;
1474 entry.mask = 0; /* don't mask IRQ for edge */
1475 entry.dest = apic->cpu_mask_to_apicid(apic->target_cpus());
1476 entry.delivery_mode = apic->irq_delivery_mode;
1479 entry.vector = vector;
1482 * The timer IRQ doesn't have to know that behind the
1483 * scene we may have a 8259A-master in AEOI mode ...
1485 irq_set_chip_and_handler_name(0, &ioapic_chip, handle_edge_irq,
1489 * Add it to the IO-APIC irq-routing table:
1491 ioapic_write_entry(apic_id, pin, entry);
1495 __apicdebuginit(void) print_IO_APIC(void)
1498 union IO_APIC_reg_00 reg_00;
1499 union IO_APIC_reg_01 reg_01;
1500 union IO_APIC_reg_02 reg_02;
1501 union IO_APIC_reg_03 reg_03;
1502 unsigned long flags;
1503 struct irq_cfg *cfg;
1506 printk(KERN_DEBUG "number of MP IRQ sources: %d.\n", mp_irq_entries);
1507 for (i = 0; i < nr_ioapics; i++)
1508 printk(KERN_DEBUG "number of IO-APIC #%d registers: %d.\n",
1509 mpc_ioapic_id(i), ioapics[i].nr_registers);
1512 * We are a bit conservative about what we expect. We have to
1513 * know about every hardware change ASAP.
1515 printk(KERN_INFO "testing the IO APIC.......................\n");
1517 for (apic = 0; apic < nr_ioapics; apic++) {
1519 raw_spin_lock_irqsave(&ioapic_lock, flags);
1520 reg_00.raw = io_apic_read(apic, 0);
1521 reg_01.raw = io_apic_read(apic, 1);
1522 if (reg_01.bits.version >= 0x10)
1523 reg_02.raw = io_apic_read(apic, 2);
1524 if (reg_01.bits.version >= 0x20)
1525 reg_03.raw = io_apic_read(apic, 3);
1526 raw_spin_unlock_irqrestore(&ioapic_lock, flags);
1529 printk(KERN_DEBUG "IO APIC #%d......\n", mpc_ioapic_id(apic));
1530 printk(KERN_DEBUG ".... register #00: %08X\n", reg_00.raw);
1531 printk(KERN_DEBUG "....... : physical APIC id: %02X\n", reg_00.bits.ID);
1532 printk(KERN_DEBUG "....... : Delivery Type: %X\n", reg_00.bits.delivery_type);
1533 printk(KERN_DEBUG "....... : LTS : %X\n", reg_00.bits.LTS);
1535 printk(KERN_DEBUG ".... register #01: %08X\n", *(int *)®_01);
1536 printk(KERN_DEBUG "....... : max redirection entries: %02X\n",
1537 reg_01.bits.entries);
1539 printk(KERN_DEBUG "....... : PRQ implemented: %X\n", reg_01.bits.PRQ);
1540 printk(KERN_DEBUG "....... : IO APIC version: %02X\n",
1541 reg_01.bits.version);
1544 * Some Intel chipsets with IO APIC VERSION of 0x1? don't have reg_02,
1545 * but the value of reg_02 is read as the previous read register
1546 * value, so ignore it if reg_02 == reg_01.
1548 if (reg_01.bits.version >= 0x10 && reg_02.raw != reg_01.raw) {
1549 printk(KERN_DEBUG ".... register #02: %08X\n", reg_02.raw);
1550 printk(KERN_DEBUG "....... : arbitration: %02X\n", reg_02.bits.arbitration);
1554 * Some Intel chipsets with IO APIC VERSION of 0x2? don't have reg_02
1555 * or reg_03, but the value of reg_0[23] is read as the previous read
1556 * register value, so ignore it if reg_03 == reg_0[12].
1558 if (reg_01.bits.version >= 0x20 && reg_03.raw != reg_02.raw &&
1559 reg_03.raw != reg_01.raw) {
1560 printk(KERN_DEBUG ".... register #03: %08X\n", reg_03.raw);
1561 printk(KERN_DEBUG "....... : Boot DT : %X\n", reg_03.bits.boot_DT);
1564 printk(KERN_DEBUG ".... IRQ redirection table:\n");
1566 if (intr_remapping_enabled) {
1567 printk(KERN_DEBUG " NR Indx Fmt Mask Trig IRR"
1568 " Pol Stat Indx2 Zero Vect:\n");
1570 printk(KERN_DEBUG " NR Dst Mask Trig IRR Pol"
1571 " Stat Dmod Deli Vect:\n");
1574 for (i = 0; i <= reg_01.bits.entries; i++) {
1575 if (intr_remapping_enabled) {
1576 struct IO_APIC_route_entry entry;
1577 struct IR_IO_APIC_route_entry *ir_entry;
1579 entry = ioapic_read_entry(apic, i);
1580 ir_entry = (struct IR_IO_APIC_route_entry *) &entry;
1581 printk(KERN_DEBUG " %02x %04X ",
1585 printk("%1d %1d %1d %1d %1d "
1586 "%1d %1d %X %02X\n",
1592 ir_entry->delivery_status,
1598 struct IO_APIC_route_entry entry;
1600 entry = ioapic_read_entry(apic, i);
1601 printk(KERN_DEBUG " %02x %02X ",
1605 printk("%1d %1d %1d %1d %1d "
1611 entry.delivery_status,
1613 entry.delivery_mode,
1620 printk(KERN_DEBUG "IRQ to pin mappings:\n");
1621 for_each_active_irq(irq) {
1622 struct irq_pin_list *entry;
1624 cfg = irq_get_chip_data(irq);
1627 entry = cfg->irq_2_pin;
1630 printk(KERN_DEBUG "IRQ%d ", irq);
1631 for_each_irq_pin(entry, cfg->irq_2_pin)
1632 printk("-> %d:%d", entry->apic, entry->pin);
1636 printk(KERN_INFO ".................................... done.\n");
1641 __apicdebuginit(void) print_APIC_field(int base)
1647 for (i = 0; i < 8; i++)
1648 printk(KERN_CONT "%08x", apic_read(base + i*0x10));
1650 printk(KERN_CONT "\n");
1653 __apicdebuginit(void) print_local_APIC(void *dummy)
1655 unsigned int i, v, ver, maxlvt;
1658 printk(KERN_DEBUG "printing local APIC contents on CPU#%d/%d:\n",
1659 smp_processor_id(), hard_smp_processor_id());
1660 v = apic_read(APIC_ID);
1661 printk(KERN_INFO "... APIC ID: %08x (%01x)\n", v, read_apic_id());
1662 v = apic_read(APIC_LVR);
1663 printk(KERN_INFO "... APIC VERSION: %08x\n", v);
1664 ver = GET_APIC_VERSION(v);
1665 maxlvt = lapic_get_maxlvt();
1667 v = apic_read(APIC_TASKPRI);
1668 printk(KERN_DEBUG "... APIC TASKPRI: %08x (%02x)\n", v, v & APIC_TPRI_MASK);
1670 if (APIC_INTEGRATED(ver)) { /* !82489DX */
1671 if (!APIC_XAPIC(ver)) {
1672 v = apic_read(APIC_ARBPRI);
1673 printk(KERN_DEBUG "... APIC ARBPRI: %08x (%02x)\n", v,
1674 v & APIC_ARBPRI_MASK);
1676 v = apic_read(APIC_PROCPRI);
1677 printk(KERN_DEBUG "... APIC PROCPRI: %08x\n", v);
1681 * Remote read supported only in the 82489DX and local APIC for
1682 * Pentium processors.
1684 if (!APIC_INTEGRATED(ver) || maxlvt == 3) {
1685 v = apic_read(APIC_RRR);
1686 printk(KERN_DEBUG "... APIC RRR: %08x\n", v);
1689 v = apic_read(APIC_LDR);
1690 printk(KERN_DEBUG "... APIC LDR: %08x\n", v);
1691 if (!x2apic_enabled()) {
1692 v = apic_read(APIC_DFR);
1693 printk(KERN_DEBUG "... APIC DFR: %08x\n", v);
1695 v = apic_read(APIC_SPIV);
1696 printk(KERN_DEBUG "... APIC SPIV: %08x\n", v);
1698 printk(KERN_DEBUG "... APIC ISR field:\n");
1699 print_APIC_field(APIC_ISR);
1700 printk(KERN_DEBUG "... APIC TMR field:\n");
1701 print_APIC_field(APIC_TMR);
1702 printk(KERN_DEBUG "... APIC IRR field:\n");
1703 print_APIC_field(APIC_IRR);
1705 if (APIC_INTEGRATED(ver)) { /* !82489DX */
1706 if (maxlvt > 3) /* Due to the Pentium erratum 3AP. */
1707 apic_write(APIC_ESR, 0);
1709 v = apic_read(APIC_ESR);
1710 printk(KERN_DEBUG "... APIC ESR: %08x\n", v);
1713 icr = apic_icr_read();
1714 printk(KERN_DEBUG "... APIC ICR: %08x\n", (u32)icr);
1715 printk(KERN_DEBUG "... APIC ICR2: %08x\n", (u32)(icr >> 32));
1717 v = apic_read(APIC_LVTT);
1718 printk(KERN_DEBUG "... APIC LVTT: %08x\n", v);
1720 if (maxlvt > 3) { /* PC is LVT#4. */
1721 v = apic_read(APIC_LVTPC);
1722 printk(KERN_DEBUG "... APIC LVTPC: %08x\n", v);
1724 v = apic_read(APIC_LVT0);
1725 printk(KERN_DEBUG "... APIC LVT0: %08x\n", v);
1726 v = apic_read(APIC_LVT1);
1727 printk(KERN_DEBUG "... APIC LVT1: %08x\n", v);
1729 if (maxlvt > 2) { /* ERR is LVT#3. */
1730 v = apic_read(APIC_LVTERR);
1731 printk(KERN_DEBUG "... APIC LVTERR: %08x\n", v);
1734 v = apic_read(APIC_TMICT);
1735 printk(KERN_DEBUG "... APIC TMICT: %08x\n", v);
1736 v = apic_read(APIC_TMCCT);
1737 printk(KERN_DEBUG "... APIC TMCCT: %08x\n", v);
1738 v = apic_read(APIC_TDCR);
1739 printk(KERN_DEBUG "... APIC TDCR: %08x\n", v);
1741 if (boot_cpu_has(X86_FEATURE_EXTAPIC)) {
1742 v = apic_read(APIC_EFEAT);
1743 maxlvt = (v >> 16) & 0xff;
1744 printk(KERN_DEBUG "... APIC EFEAT: %08x\n", v);
1745 v = apic_read(APIC_ECTRL);
1746 printk(KERN_DEBUG "... APIC ECTRL: %08x\n", v);
1747 for (i = 0; i < maxlvt; i++) {
1748 v = apic_read(APIC_EILVTn(i));
1749 printk(KERN_DEBUG "... APIC EILVT%d: %08x\n", i, v);
1755 __apicdebuginit(void) print_local_APICs(int maxcpu)
1763 for_each_online_cpu(cpu) {
1766 smp_call_function_single(cpu, print_local_APIC, NULL, 1);
1771 __apicdebuginit(void) print_PIC(void)
1774 unsigned long flags;
1776 if (!legacy_pic->nr_legacy_irqs)
1779 printk(KERN_DEBUG "\nprinting PIC contents\n");
1781 raw_spin_lock_irqsave(&i8259A_lock, flags);
1783 v = inb(0xa1) << 8 | inb(0x21);
1784 printk(KERN_DEBUG "... PIC IMR: %04x\n", v);
1786 v = inb(0xa0) << 8 | inb(0x20);
1787 printk(KERN_DEBUG "... PIC IRR: %04x\n", v);
1791 v = inb(0xa0) << 8 | inb(0x20);
1795 raw_spin_unlock_irqrestore(&i8259A_lock, flags);
1797 printk(KERN_DEBUG "... PIC ISR: %04x\n", v);
1799 v = inb(0x4d1) << 8 | inb(0x4d0);
1800 printk(KERN_DEBUG "... PIC ELCR: %04x\n", v);
1803 static int __initdata show_lapic = 1;
1804 static __init int setup_show_lapic(char *arg)
1808 if (strcmp(arg, "all") == 0) {
1809 show_lapic = CONFIG_NR_CPUS;
1811 get_option(&arg, &num);
1818 __setup("show_lapic=", setup_show_lapic);
1820 __apicdebuginit(int) print_ICs(void)
1822 if (apic_verbosity == APIC_QUIET)
1827 /* don't print out if apic is not there */
1828 if (!cpu_has_apic && !apic_from_smp_config())
1831 print_local_APICs(show_lapic);
1837 late_initcall(print_ICs);
1840 /* Where if anywhere is the i8259 connect in external int mode */
1841 static struct { int pin, apic; } ioapic_i8259 = { -1, -1 };
1843 void __init enable_IO_APIC(void)
1845 int i8259_apic, i8259_pin;
1848 if (!legacy_pic->nr_legacy_irqs)
1851 for(apic = 0; apic < nr_ioapics; apic++) {
1853 /* See if any of the pins is in ExtINT mode */
1854 for (pin = 0; pin < ioapics[apic].nr_registers; pin++) {
1855 struct IO_APIC_route_entry entry;
1856 entry = ioapic_read_entry(apic, pin);
1858 /* If the interrupt line is enabled and in ExtInt mode
1859 * I have found the pin where the i8259 is connected.
1861 if ((entry.mask == 0) && (entry.delivery_mode == dest_ExtINT)) {
1862 ioapic_i8259.apic = apic;
1863 ioapic_i8259.pin = pin;
1869 /* Look to see what if the MP table has reported the ExtINT */
1870 /* If we could not find the appropriate pin by looking at the ioapic
1871 * the i8259 probably is not connected the ioapic but give the
1872 * mptable a chance anyway.
1874 i8259_pin = find_isa_irq_pin(0, mp_ExtINT);
1875 i8259_apic = find_isa_irq_apic(0, mp_ExtINT);
1876 /* Trust the MP table if nothing is setup in the hardware */
1877 if ((ioapic_i8259.pin == -1) && (i8259_pin >= 0)) {
1878 printk(KERN_WARNING "ExtINT not setup in hardware but reported by MP table\n");
1879 ioapic_i8259.pin = i8259_pin;
1880 ioapic_i8259.apic = i8259_apic;
1882 /* Complain if the MP table and the hardware disagree */
1883 if (((ioapic_i8259.apic != i8259_apic) || (ioapic_i8259.pin != i8259_pin)) &&
1884 (i8259_pin >= 0) && (ioapic_i8259.pin >= 0))
1886 printk(KERN_WARNING "ExtINT in hardware and MP table differ\n");
1890 * Do not trust the IO-APIC being empty at bootup
1896 * Not an __init, needed by the reboot code
1898 void disable_IO_APIC(void)
1901 * Clear the IO-APIC before rebooting:
1905 if (!legacy_pic->nr_legacy_irqs)
1909 * If the i8259 is routed through an IOAPIC
1910 * Put that IOAPIC in virtual wire mode
1911 * so legacy interrupts can be delivered.
1913 * With interrupt-remapping, for now we will use virtual wire A mode,
1914 * as virtual wire B is little complex (need to configure both
1915 * IOAPIC RTE as well as interrupt-remapping table entry).
1916 * As this gets called during crash dump, keep this simple for now.
1918 if (ioapic_i8259.pin != -1 && !intr_remapping_enabled) {
1919 struct IO_APIC_route_entry entry;
1921 memset(&entry, 0, sizeof(entry));
1922 entry.mask = 0; /* Enabled */
1923 entry.trigger = 0; /* Edge */
1925 entry.polarity = 0; /* High */
1926 entry.delivery_status = 0;
1927 entry.dest_mode = 0; /* Physical */
1928 entry.delivery_mode = dest_ExtINT; /* ExtInt */
1930 entry.dest = read_apic_id();
1933 * Add it to the IO-APIC irq-routing table:
1935 ioapic_write_entry(ioapic_i8259.apic, ioapic_i8259.pin, entry);
1939 * Use virtual wire A mode when interrupt remapping is enabled.
1941 if (cpu_has_apic || apic_from_smp_config())
1942 disconnect_bsp_APIC(!intr_remapping_enabled &&
1943 ioapic_i8259.pin != -1);
1946 #ifdef CONFIG_X86_32
1948 * function to set the IO-APIC physical IDs based on the
1949 * values stored in the MPC table.
1951 * by Matt Domsch <Matt_Domsch@dell.com> Tue Dec 21 12:25:05 CST 1999
1953 void __init setup_ioapic_ids_from_mpc_nocheck(void)
1955 union IO_APIC_reg_00 reg_00;
1956 physid_mask_t phys_id_present_map;
1959 unsigned char old_id;
1960 unsigned long flags;
1963 * This is broken; anything with a real cpu count has to
1964 * circumvent this idiocy regardless.
1966 apic->ioapic_phys_id_map(&phys_cpu_present_map, &phys_id_present_map);
1969 * Set the IOAPIC ID to the value stored in the MPC table.
1971 for (apic_id = 0; apic_id < nr_ioapics; apic_id++) {
1973 /* Read the register 0 value */
1974 raw_spin_lock_irqsave(&ioapic_lock, flags);
1975 reg_00.raw = io_apic_read(apic_id, 0);
1976 raw_spin_unlock_irqrestore(&ioapic_lock, flags);
1978 old_id = mpc_ioapic_id(apic_id);
1980 if (mpc_ioapic_id(apic_id) >= get_physical_broadcast()) {
1981 printk(KERN_ERR "BIOS bug, IO-APIC#%d ID is %d in the MPC table!...\n",
1982 apic_id, mpc_ioapic_id(apic_id));
1983 printk(KERN_ERR "... fixing up to %d. (tell your hw vendor)\n",
1985 ioapics[apic_id].mp_config.apicid = reg_00.bits.ID;
1989 * Sanity check, is the ID really free? Every APIC in a
1990 * system must have a unique ID or we get lots of nice
1991 * 'stuck on smp_invalidate_needed IPI wait' messages.
1993 if (apic->check_apicid_used(&phys_id_present_map,
1994 mpc_ioapic_id(apic_id))) {
1995 printk(KERN_ERR "BIOS bug, IO-APIC#%d ID %d is already used!...\n",
1996 apic_id, mpc_ioapic_id(apic_id));
1997 for (i = 0; i < get_physical_broadcast(); i++)
1998 if (!physid_isset(i, phys_id_present_map))
2000 if (i >= get_physical_broadcast())
2001 panic("Max APIC ID exceeded!\n");
2002 printk(KERN_ERR "... fixing up to %d. (tell your hw vendor)\n",
2004 physid_set(i, phys_id_present_map);
2005 ioapics[apic_id].mp_config.apicid = i;
2008 apic->apicid_to_cpu_present(mpc_ioapic_id(apic_id),
2010 apic_printk(APIC_VERBOSE, "Setting %d in the "
2011 "phys_id_present_map\n",
2012 mpc_ioapic_id(apic_id));
2013 physids_or(phys_id_present_map, phys_id_present_map, tmp);
2017 * We need to adjust the IRQ routing table
2018 * if the ID changed.
2020 if (old_id != mpc_ioapic_id(apic_id))
2021 for (i = 0; i < mp_irq_entries; i++)
2022 if (mp_irqs[i].dstapic == old_id)
2024 = mpc_ioapic_id(apic_id);
2027 * Update the ID register according to the right value
2028 * from the MPC table if they are different.
2030 if (mpc_ioapic_id(apic_id) == reg_00.bits.ID)
2033 apic_printk(APIC_VERBOSE, KERN_INFO
2034 "...changing IO-APIC physical APIC ID to %d ...",
2035 mpc_ioapic_id(apic_id));
2037 reg_00.bits.ID = mpc_ioapic_id(apic_id);
2038 raw_spin_lock_irqsave(&ioapic_lock, flags);
2039 io_apic_write(apic_id, 0, reg_00.raw);
2040 raw_spin_unlock_irqrestore(&ioapic_lock, flags);
2045 raw_spin_lock_irqsave(&ioapic_lock, flags);
2046 reg_00.raw = io_apic_read(apic_id, 0);
2047 raw_spin_unlock_irqrestore(&ioapic_lock, flags);
2048 if (reg_00.bits.ID != mpc_ioapic_id(apic_id))
2049 printk("could not set ID!\n");
2051 apic_printk(APIC_VERBOSE, " ok.\n");
2055 void __init setup_ioapic_ids_from_mpc(void)
2061 * Don't check I/O APIC IDs for xAPIC systems. They have
2062 * no meaning without the serial APIC bus.
2064 if (!(boot_cpu_data.x86_vendor == X86_VENDOR_INTEL)
2065 || APIC_XAPIC(apic_version[boot_cpu_physical_apicid]))
2067 setup_ioapic_ids_from_mpc_nocheck();
2071 int no_timer_check __initdata;
2073 static int __init notimercheck(char *s)
2078 __setup("no_timer_check", notimercheck);
2081 * There is a nasty bug in some older SMP boards, their mptable lies
2082 * about the timer IRQ. We do the following to work around the situation:
2084 * - timer IRQ defaults to IO-APIC IRQ
2085 * - if this function detects that timer IRQs are defunct, then we fall
2086 * back to ISA timer IRQs
2088 static int __init timer_irq_works(void)
2090 unsigned long t1 = jiffies;
2091 unsigned long flags;
2096 local_save_flags(flags);
2098 /* Let ten ticks pass... */
2099 mdelay((10 * 1000) / HZ);
2100 local_irq_restore(flags);
2103 * Expect a few ticks at least, to be sure some possible
2104 * glue logic does not lock up after one or two first
2105 * ticks in a non-ExtINT mode. Also the local APIC
2106 * might have cached one ExtINT interrupt. Finally, at
2107 * least one tick may be lost due to delays.
2111 if (time_after(jiffies, t1 + 4))
2117 * In the SMP+IOAPIC case it might happen that there are an unspecified
2118 * number of pending IRQ events unhandled. These cases are very rare,
2119 * so we 'resend' these IRQs via IPIs, to the same CPU. It's much
2120 * better to do it this way as thus we do not have to be aware of
2121 * 'pending' interrupts in the IRQ path, except at this point.
2124 * Edge triggered needs to resend any interrupt
2125 * that was delayed but this is now handled in the device
2130 * Starting up a edge-triggered IO-APIC interrupt is
2131 * nasty - we need to make sure that we get the edge.
2132 * If it is already asserted for some reason, we need
2133 * return 1 to indicate that is was pending.
2135 * This is not complete - we should be able to fake
2136 * an edge even if it isn't on the 8259A...
2139 static unsigned int startup_ioapic_irq(struct irq_data *data)
2141 int was_pending = 0, irq = data->irq;
2142 unsigned long flags;
2144 raw_spin_lock_irqsave(&ioapic_lock, flags);
2145 if (irq < legacy_pic->nr_legacy_irqs) {
2146 legacy_pic->mask(irq);
2147 if (legacy_pic->irq_pending(irq))
2150 __unmask_ioapic(data->chip_data);
2151 raw_spin_unlock_irqrestore(&ioapic_lock, flags);
2156 static int ioapic_retrigger_irq(struct irq_data *data)
2158 struct irq_cfg *cfg = data->chip_data;
2159 unsigned long flags;
2161 raw_spin_lock_irqsave(&vector_lock, flags);
2162 apic->send_IPI_mask(cpumask_of(cpumask_first(cfg->domain)), cfg->vector);
2163 raw_spin_unlock_irqrestore(&vector_lock, flags);
2169 * Level and edge triggered IO-APIC interrupts need different handling,
2170 * so we use two separate IRQ descriptors. Edge triggered IRQs can be
2171 * handled with the level-triggered descriptor, but that one has slightly
2172 * more overhead. Level-triggered interrupts cannot be handled with the
2173 * edge-triggered handler, without risking IRQ storms and other ugly
2178 void send_cleanup_vector(struct irq_cfg *cfg)
2180 cpumask_var_t cleanup_mask;
2182 if (unlikely(!alloc_cpumask_var(&cleanup_mask, GFP_ATOMIC))) {
2184 for_each_cpu_and(i, cfg->old_domain, cpu_online_mask)
2185 apic->send_IPI_mask(cpumask_of(i), IRQ_MOVE_CLEANUP_VECTOR);
2187 cpumask_and(cleanup_mask, cfg->old_domain, cpu_online_mask);
2188 apic->send_IPI_mask(cleanup_mask, IRQ_MOVE_CLEANUP_VECTOR);
2189 free_cpumask_var(cleanup_mask);
2191 cfg->move_in_progress = 0;
2194 static void __target_IO_APIC_irq(unsigned int irq, unsigned int dest, struct irq_cfg *cfg)
2197 struct irq_pin_list *entry;
2198 u8 vector = cfg->vector;
2200 for_each_irq_pin(entry, cfg->irq_2_pin) {
2206 * With interrupt-remapping, destination information comes
2207 * from interrupt-remapping table entry.
2209 if (!irq_remapped(cfg))
2210 io_apic_write(apic, 0x11 + pin*2, dest);
2211 reg = io_apic_read(apic, 0x10 + pin*2);
2212 reg &= ~IO_APIC_REDIR_VECTOR_MASK;
2214 io_apic_modify(apic, 0x10 + pin*2, reg);
2219 * Either sets data->affinity to a valid value, and returns
2220 * ->cpu_mask_to_apicid of that in dest_id, or returns -1 and
2221 * leaves data->affinity untouched.
2223 int __ioapic_set_affinity(struct irq_data *data, const struct cpumask *mask,
2224 unsigned int *dest_id)
2226 struct irq_cfg *cfg = data->chip_data;
2228 if (!cpumask_intersects(mask, cpu_online_mask))
2231 if (assign_irq_vector(data->irq, data->chip_data, mask))
2234 cpumask_copy(data->affinity, mask);
2236 *dest_id = apic->cpu_mask_to_apicid_and(mask, cfg->domain);
2241 ioapic_set_affinity(struct irq_data *data, const struct cpumask *mask,
2244 unsigned int dest, irq = data->irq;
2245 unsigned long flags;
2248 raw_spin_lock_irqsave(&ioapic_lock, flags);
2249 ret = __ioapic_set_affinity(data, mask, &dest);
2251 /* Only the high 8 bits are valid. */
2252 dest = SET_APIC_LOGICAL_ID(dest);
2253 __target_IO_APIC_irq(irq, dest, data->chip_data);
2255 raw_spin_unlock_irqrestore(&ioapic_lock, flags);
2259 #ifdef CONFIG_INTR_REMAP
2262 * Migrate the IO-APIC irq in the presence of intr-remapping.
2264 * For both level and edge triggered, irq migration is a simple atomic
2265 * update(of vector and cpu destination) of IRTE and flush the hardware cache.
2267 * For level triggered, we eliminate the io-apic RTE modification (with the
2268 * updated vector information), by using a virtual vector (io-apic pin number).
2269 * Real vector that is used for interrupting cpu will be coming from
2270 * the interrupt-remapping table entry.
2273 ir_ioapic_set_affinity(struct irq_data *data, const struct cpumask *mask,
2276 struct irq_cfg *cfg = data->chip_data;
2277 unsigned int dest, irq = data->irq;
2280 if (!cpumask_intersects(mask, cpu_online_mask))
2283 if (get_irte(irq, &irte))
2286 if (assign_irq_vector(irq, cfg, mask))
2289 dest = apic->cpu_mask_to_apicid_and(cfg->domain, mask);
2291 irte.vector = cfg->vector;
2292 irte.dest_id = IRTE_DEST(dest);
2295 * Modified the IRTE and flushes the Interrupt entry cache.
2297 modify_irte(irq, &irte);
2299 if (cfg->move_in_progress)
2300 send_cleanup_vector(cfg);
2302 cpumask_copy(data->affinity, mask);
2308 ir_ioapic_set_affinity(struct irq_data *data, const struct cpumask *mask,
2315 asmlinkage void smp_irq_move_cleanup_interrupt(void)
2317 unsigned vector, me;
2323 me = smp_processor_id();
2324 for (vector = FIRST_EXTERNAL_VECTOR; vector < NR_VECTORS; vector++) {
2327 struct irq_desc *desc;
2328 struct irq_cfg *cfg;
2329 irq = __this_cpu_read(vector_irq[vector]);
2334 desc = irq_to_desc(irq);
2339 raw_spin_lock(&desc->lock);
2342 * Check if the irq migration is in progress. If so, we
2343 * haven't received the cleanup request yet for this irq.
2345 if (cfg->move_in_progress)
2348 if (vector == cfg->vector && cpumask_test_cpu(me, cfg->domain))
2351 irr = apic_read(APIC_IRR + (vector / 32 * 0x10));
2353 * Check if the vector that needs to be cleanedup is
2354 * registered at the cpu's IRR. If so, then this is not
2355 * the best time to clean it up. Lets clean it up in the
2356 * next attempt by sending another IRQ_MOVE_CLEANUP_VECTOR
2359 if (irr & (1 << (vector % 32))) {
2360 apic->send_IPI_self(IRQ_MOVE_CLEANUP_VECTOR);
2363 __this_cpu_write(vector_irq[vector], -1);
2365 raw_spin_unlock(&desc->lock);
2371 static void __irq_complete_move(struct irq_cfg *cfg, unsigned vector)
2375 if (likely(!cfg->move_in_progress))
2378 me = smp_processor_id();
2380 if (vector == cfg->vector && cpumask_test_cpu(me, cfg->domain))
2381 send_cleanup_vector(cfg);
2384 static void irq_complete_move(struct irq_cfg *cfg)
2386 __irq_complete_move(cfg, ~get_irq_regs()->orig_ax);
2389 void irq_force_complete_move(int irq)
2391 struct irq_cfg *cfg = irq_get_chip_data(irq);
2396 __irq_complete_move(cfg, cfg->vector);
2399 static inline void irq_complete_move(struct irq_cfg *cfg) { }
2402 static void ack_apic_edge(struct irq_data *data)
2404 irq_complete_move(data->chip_data);
2409 atomic_t irq_mis_count;
2412 * IO-APIC versions below 0x20 don't support EOI register.
2413 * For the record, here is the information about various versions:
2415 * 1Xh I/OAPIC or I/O(x)APIC which are not PCI 2.2 Compliant
2416 * 2Xh I/O(x)APIC which is PCI 2.2 Compliant
2419 * Some of the Intel ICH Specs (ICH2 to ICH5) documents the io-apic
2420 * version as 0x2. This is an error with documentation and these ICH chips
2421 * use io-apic's of version 0x20.
2423 * For IO-APIC's with EOI register, we use that to do an explicit EOI.
2424 * Otherwise, we simulate the EOI message manually by changing the trigger
2425 * mode to edge and then back to level, with RTE being masked during this.
2427 static void eoi_ioapic_irq(unsigned int irq, struct irq_cfg *cfg)
2429 struct irq_pin_list *entry;
2430 unsigned long flags;
2432 raw_spin_lock_irqsave(&ioapic_lock, flags);
2433 for_each_irq_pin(entry, cfg->irq_2_pin) {
2434 if (mpc_ioapic_ver(entry->apic) >= 0x20) {
2436 * Intr-remapping uses pin number as the virtual vector
2437 * in the RTE. Actual vector is programmed in
2438 * intr-remapping table entry. Hence for the io-apic
2439 * EOI we use the pin number.
2441 if (irq_remapped(cfg))
2442 io_apic_eoi(entry->apic, entry->pin);
2444 io_apic_eoi(entry->apic, cfg->vector);
2446 __mask_and_edge_IO_APIC_irq(entry);
2447 __unmask_and_level_IO_APIC_irq(entry);
2450 raw_spin_unlock_irqrestore(&ioapic_lock, flags);
2453 static void ack_apic_level(struct irq_data *data)
2455 struct irq_cfg *cfg = data->chip_data;
2456 int i, do_unmask_irq = 0, irq = data->irq;
2459 irq_complete_move(cfg);
2460 #ifdef CONFIG_GENERIC_PENDING_IRQ
2461 /* If we are moving the irq we need to mask it */
2462 if (unlikely(irqd_is_setaffinity_pending(data))) {
2469 * It appears there is an erratum which affects at least version 0x11
2470 * of I/O APIC (that's the 82093AA and cores integrated into various
2471 * chipsets). Under certain conditions a level-triggered interrupt is
2472 * erroneously delivered as edge-triggered one but the respective IRR
2473 * bit gets set nevertheless. As a result the I/O unit expects an EOI
2474 * message but it will never arrive and further interrupts are blocked
2475 * from the source. The exact reason is so far unknown, but the
2476 * phenomenon was observed when two consecutive interrupt requests
2477 * from a given source get delivered to the same CPU and the source is
2478 * temporarily disabled in between.
2480 * A workaround is to simulate an EOI message manually. We achieve it
2481 * by setting the trigger mode to edge and then to level when the edge
2482 * trigger mode gets detected in the TMR of a local APIC for a
2483 * level-triggered interrupt. We mask the source for the time of the
2484 * operation to prevent an edge-triggered interrupt escaping meanwhile.
2485 * The idea is from Manfred Spraul. --macro
2487 * Also in the case when cpu goes offline, fixup_irqs() will forward
2488 * any unhandled interrupt on the offlined cpu to the new cpu
2489 * destination that is handling the corresponding interrupt. This
2490 * interrupt forwarding is done via IPI's. Hence, in this case also
2491 * level-triggered io-apic interrupt will be seen as an edge
2492 * interrupt in the IRR. And we can't rely on the cpu's EOI
2493 * to be broadcasted to the IO-APIC's which will clear the remoteIRR
2494 * corresponding to the level-triggered interrupt. Hence on IO-APIC's
2495 * supporting EOI register, we do an explicit EOI to clear the
2496 * remote IRR and on IO-APIC's which don't have an EOI register,
2497 * we use the above logic (mask+edge followed by unmask+level) from
2498 * Manfred Spraul to clear the remote IRR.
2501 v = apic_read(APIC_TMR + ((i & ~0x1f) >> 1));
2504 * We must acknowledge the irq before we move it or the acknowledge will
2505 * not propagate properly.
2510 * Tail end of clearing remote IRR bit (either by delivering the EOI
2511 * message via io-apic EOI register write or simulating it using
2512 * mask+edge followed by unnask+level logic) manually when the
2513 * level triggered interrupt is seen as the edge triggered interrupt
2516 if (!(v & (1 << (i & 0x1f)))) {
2517 atomic_inc(&irq_mis_count);
2519 eoi_ioapic_irq(irq, cfg);
2522 /* Now we can move and renable the irq */
2523 if (unlikely(do_unmask_irq)) {
2524 /* Only migrate the irq if the ack has been received.
2526 * On rare occasions the broadcast level triggered ack gets
2527 * delayed going to ioapics, and if we reprogram the
2528 * vector while Remote IRR is still set the irq will never
2531 * To prevent this scenario we read the Remote IRR bit
2532 * of the ioapic. This has two effects.
2533 * - On any sane system the read of the ioapic will
2534 * flush writes (and acks) going to the ioapic from
2536 * - We get to see if the ACK has actually been delivered.
2538 * Based on failed experiments of reprogramming the
2539 * ioapic entry from outside of irq context starting
2540 * with masking the ioapic entry and then polling until
2541 * Remote IRR was clear before reprogramming the
2542 * ioapic I don't trust the Remote IRR bit to be
2543 * completey accurate.
2545 * However there appears to be no other way to plug
2546 * this race, so if the Remote IRR bit is not
2547 * accurate and is causing problems then it is a hardware bug
2548 * and you can go talk to the chipset vendor about it.
2550 if (!io_apic_level_ack_pending(cfg))
2551 irq_move_masked_irq(data);
2556 #ifdef CONFIG_INTR_REMAP
2557 static void ir_ack_apic_edge(struct irq_data *data)
2562 static void ir_ack_apic_level(struct irq_data *data)
2565 eoi_ioapic_irq(data->irq, data->chip_data);
2567 #endif /* CONFIG_INTR_REMAP */
2569 static struct irq_chip ioapic_chip __read_mostly = {
2571 .irq_startup = startup_ioapic_irq,
2572 .irq_mask = mask_ioapic_irq,
2573 .irq_unmask = unmask_ioapic_irq,
2574 .irq_ack = ack_apic_edge,
2575 .irq_eoi = ack_apic_level,
2577 .irq_set_affinity = ioapic_set_affinity,
2579 .irq_retrigger = ioapic_retrigger_irq,
2582 static struct irq_chip ir_ioapic_chip __read_mostly = {
2583 .name = "IR-IO-APIC",
2584 .irq_startup = startup_ioapic_irq,
2585 .irq_mask = mask_ioapic_irq,
2586 .irq_unmask = unmask_ioapic_irq,
2587 #ifdef CONFIG_INTR_REMAP
2588 .irq_ack = ir_ack_apic_edge,
2589 .irq_eoi = ir_ack_apic_level,
2591 .irq_set_affinity = ir_ioapic_set_affinity,
2594 .irq_retrigger = ioapic_retrigger_irq,
2597 static inline void init_IO_APIC_traps(void)
2599 struct irq_cfg *cfg;
2603 * NOTE! The local APIC isn't very good at handling
2604 * multiple interrupts at the same interrupt level.
2605 * As the interrupt level is determined by taking the
2606 * vector number and shifting that right by 4, we
2607 * want to spread these out a bit so that they don't
2608 * all fall in the same interrupt level.
2610 * Also, we've got to be careful not to trash gate
2611 * 0x80, because int 0x80 is hm, kind of importantish. ;)
2613 for_each_active_irq(irq) {
2614 cfg = irq_get_chip_data(irq);
2615 if (IO_APIC_IRQ(irq) && cfg && !cfg->vector) {
2617 * Hmm.. We don't have an entry for this,
2618 * so default to an old-fashioned 8259
2619 * interrupt if we can..
2621 if (irq < legacy_pic->nr_legacy_irqs)
2622 legacy_pic->make_irq(irq);
2624 /* Strange. Oh, well.. */
2625 irq_set_chip(irq, &no_irq_chip);
2631 * The local APIC irq-chip implementation:
2634 static void mask_lapic_irq(struct irq_data *data)
2638 v = apic_read(APIC_LVT0);
2639 apic_write(APIC_LVT0, v | APIC_LVT_MASKED);
2642 static void unmask_lapic_irq(struct irq_data *data)
2646 v = apic_read(APIC_LVT0);
2647 apic_write(APIC_LVT0, v & ~APIC_LVT_MASKED);
2650 static void ack_lapic_irq(struct irq_data *data)
2655 static struct irq_chip lapic_chip __read_mostly = {
2656 .name = "local-APIC",
2657 .irq_mask = mask_lapic_irq,
2658 .irq_unmask = unmask_lapic_irq,
2659 .irq_ack = ack_lapic_irq,
2662 static void lapic_register_intr(int irq)
2664 irq_clear_status_flags(irq, IRQ_LEVEL);
2665 irq_set_chip_and_handler_name(irq, &lapic_chip, handle_edge_irq,
2670 * This looks a bit hackish but it's about the only one way of sending
2671 * a few INTA cycles to 8259As and any associated glue logic. ICR does
2672 * not support the ExtINT mode, unfortunately. We need to send these
2673 * cycles as some i82489DX-based boards have glue logic that keeps the
2674 * 8259A interrupt line asserted until INTA. --macro
2676 static inline void __init unlock_ExtINT_logic(void)
2679 struct IO_APIC_route_entry entry0, entry1;
2680 unsigned char save_control, save_freq_select;
2682 pin = find_isa_irq_pin(8, mp_INT);
2687 apic = find_isa_irq_apic(8, mp_INT);
2693 entry0 = ioapic_read_entry(apic, pin);
2694 clear_IO_APIC_pin(apic, pin);
2696 memset(&entry1, 0, sizeof(entry1));
2698 entry1.dest_mode = 0; /* physical delivery */
2699 entry1.mask = 0; /* unmask IRQ now */
2700 entry1.dest = hard_smp_processor_id();
2701 entry1.delivery_mode = dest_ExtINT;
2702 entry1.polarity = entry0.polarity;
2706 ioapic_write_entry(apic, pin, entry1);
2708 save_control = CMOS_READ(RTC_CONTROL);
2709 save_freq_select = CMOS_READ(RTC_FREQ_SELECT);
2710 CMOS_WRITE((save_freq_select & ~RTC_RATE_SELECT) | 0x6,
2712 CMOS_WRITE(save_control | RTC_PIE, RTC_CONTROL);
2717 if ((CMOS_READ(RTC_INTR_FLAGS) & RTC_PF) == RTC_PF)
2721 CMOS_WRITE(save_control, RTC_CONTROL);
2722 CMOS_WRITE(save_freq_select, RTC_FREQ_SELECT);
2723 clear_IO_APIC_pin(apic, pin);
2725 ioapic_write_entry(apic, pin, entry0);
2728 static int disable_timer_pin_1 __initdata;
2729 /* Actually the next is obsolete, but keep it for paranoid reasons -AK */
2730 static int __init disable_timer_pin_setup(char *arg)
2732 disable_timer_pin_1 = 1;
2735 early_param("disable_timer_pin_1", disable_timer_pin_setup);
2737 int timer_through_8259 __initdata;
2740 * This code may look a bit paranoid, but it's supposed to cooperate with
2741 * a wide range of boards and BIOS bugs. Fortunately only the timer IRQ
2742 * is so screwy. Thanks to Brian Perkins for testing/hacking this beast
2743 * fanatically on his truly buggy board.
2745 * FIXME: really need to revamp this for all platforms.
2747 static inline void __init check_timer(void)
2749 struct irq_cfg *cfg = irq_get_chip_data(0);
2750 int node = cpu_to_node(0);
2751 int apic1, pin1, apic2, pin2;
2752 unsigned long flags;
2755 local_irq_save(flags);
2758 * get/set the timer IRQ vector:
2760 legacy_pic->mask(0);
2761 assign_irq_vector(0, cfg, apic->target_cpus());
2764 * As IRQ0 is to be enabled in the 8259A, the virtual
2765 * wire has to be disabled in the local APIC. Also
2766 * timer interrupts need to be acknowledged manually in
2767 * the 8259A for the i82489DX when using the NMI
2768 * watchdog as that APIC treats NMIs as level-triggered.
2769 * The AEOI mode will finish them in the 8259A
2772 apic_write(APIC_LVT0, APIC_LVT_MASKED | APIC_DM_EXTINT);
2773 legacy_pic->init(1);
2775 pin1 = find_isa_irq_pin(0, mp_INT);
2776 apic1 = find_isa_irq_apic(0, mp_INT);
2777 pin2 = ioapic_i8259.pin;
2778 apic2 = ioapic_i8259.apic;
2780 apic_printk(APIC_QUIET, KERN_INFO "..TIMER: vector=0x%02X "
2781 "apic1=%d pin1=%d apic2=%d pin2=%d\n",
2782 cfg->vector, apic1, pin1, apic2, pin2);
2785 * Some BIOS writers are clueless and report the ExtINTA
2786 * I/O APIC input from the cascaded 8259A as the timer
2787 * interrupt input. So just in case, if only one pin
2788 * was found above, try it both directly and through the
2792 if (intr_remapping_enabled)
2793 panic("BIOS bug: timer not connected to IO-APIC");
2797 } else if (pin2 == -1) {
2804 * Ok, does IRQ0 through the IOAPIC work?
2807 add_pin_to_irq_node(cfg, node, apic1, pin1);
2808 setup_timer_IRQ0_pin(apic1, pin1, cfg->vector);
2810 /* for edge trigger, setup_ioapic_irq already
2811 * leave it unmasked.
2812 * so only need to unmask if it is level-trigger
2813 * do we really have level trigger timer?
2816 idx = find_irq_entry(apic1, pin1, mp_INT);
2817 if (idx != -1 && irq_trigger(idx))
2820 if (timer_irq_works()) {
2821 if (disable_timer_pin_1 > 0)
2822 clear_IO_APIC_pin(0, pin1);
2825 if (intr_remapping_enabled)
2826 panic("timer doesn't work through Interrupt-remapped IO-APIC");
2827 local_irq_disable();
2828 clear_IO_APIC_pin(apic1, pin1);
2830 apic_printk(APIC_QUIET, KERN_ERR "..MP-BIOS bug: "
2831 "8254 timer not connected to IO-APIC\n");
2833 apic_printk(APIC_QUIET, KERN_INFO "...trying to set up timer "
2834 "(IRQ0) through the 8259A ...\n");
2835 apic_printk(APIC_QUIET, KERN_INFO
2836 "..... (found apic %d pin %d) ...\n", apic2, pin2);
2838 * legacy devices should be connected to IO APIC #0
2840 replace_pin_at_irq_node(cfg, node, apic1, pin1, apic2, pin2);
2841 setup_timer_IRQ0_pin(apic2, pin2, cfg->vector);
2842 legacy_pic->unmask(0);
2843 if (timer_irq_works()) {
2844 apic_printk(APIC_QUIET, KERN_INFO "....... works.\n");
2845 timer_through_8259 = 1;
2849 * Cleanup, just in case ...
2851 local_irq_disable();
2852 legacy_pic->mask(0);
2853 clear_IO_APIC_pin(apic2, pin2);
2854 apic_printk(APIC_QUIET, KERN_INFO "....... failed.\n");
2857 apic_printk(APIC_QUIET, KERN_INFO
2858 "...trying to set up timer as Virtual Wire IRQ...\n");
2860 lapic_register_intr(0);
2861 apic_write(APIC_LVT0, APIC_DM_FIXED | cfg->vector); /* Fixed mode */
2862 legacy_pic->unmask(0);
2864 if (timer_irq_works()) {
2865 apic_printk(APIC_QUIET, KERN_INFO "..... works.\n");
2868 local_irq_disable();
2869 legacy_pic->mask(0);
2870 apic_write(APIC_LVT0, APIC_LVT_MASKED | APIC_DM_FIXED | cfg->vector);
2871 apic_printk(APIC_QUIET, KERN_INFO "..... failed.\n");
2873 apic_printk(APIC_QUIET, KERN_INFO
2874 "...trying to set up timer as ExtINT IRQ...\n");
2876 legacy_pic->init(0);
2877 legacy_pic->make_irq(0);
2878 apic_write(APIC_LVT0, APIC_DM_EXTINT);
2880 unlock_ExtINT_logic();
2882 if (timer_irq_works()) {
2883 apic_printk(APIC_QUIET, KERN_INFO "..... works.\n");
2886 local_irq_disable();
2887 apic_printk(APIC_QUIET, KERN_INFO "..... failed :(.\n");
2888 panic("IO-APIC + timer doesn't work! Boot with apic=debug and send a "
2889 "report. Then try booting with the 'noapic' option.\n");
2891 local_irq_restore(flags);
2895 * Traditionally ISA IRQ2 is the cascade IRQ, and is not available
2896 * to devices. However there may be an I/O APIC pin available for
2897 * this interrupt regardless. The pin may be left unconnected, but
2898 * typically it will be reused as an ExtINT cascade interrupt for
2899 * the master 8259A. In the MPS case such a pin will normally be
2900 * reported as an ExtINT interrupt in the MP table. With ACPI
2901 * there is no provision for ExtINT interrupts, and in the absence
2902 * of an override it would be treated as an ordinary ISA I/O APIC
2903 * interrupt, that is edge-triggered and unmasked by default. We
2904 * used to do this, but it caused problems on some systems because
2905 * of the NMI watchdog and sometimes IRQ0 of the 8254 timer using
2906 * the same ExtINT cascade interrupt to drive the local APIC of the
2907 * bootstrap processor. Therefore we refrain from routing IRQ2 to
2908 * the I/O APIC in all cases now. No actual device should request
2909 * it anyway. --macro
2911 #define PIC_IRQS (1UL << PIC_CASCADE_IR)
2913 void __init setup_IO_APIC(void)
2917 * calling enable_IO_APIC() is moved to setup_local_APIC for BP
2919 io_apic_irqs = legacy_pic->nr_legacy_irqs ? ~PIC_IRQS : ~0UL;
2921 apic_printk(APIC_VERBOSE, "ENABLING IO-APIC IRQs\n");
2923 * Set up IO-APIC IRQ routing.
2925 x86_init.mpparse.setup_ioapic_ids();
2928 setup_IO_APIC_irqs();
2929 init_IO_APIC_traps();
2930 if (legacy_pic->nr_legacy_irqs)
2935 * Called after all the initialization is done. If we didn't find any
2936 * APIC bugs then we can allow the modify fast path
2939 static int __init io_apic_bug_finalize(void)
2941 if (sis_apic_bug == -1)
2946 late_initcall(io_apic_bug_finalize);
2948 static void resume_ioapic_id(int ioapic_id)
2950 unsigned long flags;
2951 union IO_APIC_reg_00 reg_00;
2954 raw_spin_lock_irqsave(&ioapic_lock, flags);
2955 reg_00.raw = io_apic_read(ioapic_id, 0);
2956 if (reg_00.bits.ID != mpc_ioapic_id(ioapic_id)) {
2957 reg_00.bits.ID = mpc_ioapic_id(ioapic_id);
2958 io_apic_write(ioapic_id, 0, reg_00.raw);
2960 raw_spin_unlock_irqrestore(&ioapic_lock, flags);
2963 static void ioapic_resume(void)
2967 for (ioapic_id = nr_ioapics - 1; ioapic_id >= 0; ioapic_id--)
2968 resume_ioapic_id(ioapic_id);
2970 restore_ioapic_entries();
2973 static struct syscore_ops ioapic_syscore_ops = {
2974 .suspend = save_ioapic_entries,
2975 .resume = ioapic_resume,
2978 static int __init ioapic_init_ops(void)
2980 register_syscore_ops(&ioapic_syscore_ops);
2985 device_initcall(ioapic_init_ops);
2988 * Dynamic irq allocate and deallocation
2990 unsigned int create_irq_nr(unsigned int from, int node)
2992 struct irq_cfg *cfg;
2993 unsigned long flags;
2994 unsigned int ret = 0;
2997 if (from < nr_irqs_gsi)
3000 irq = alloc_irq_from(from, node);
3003 cfg = alloc_irq_cfg(irq, node);
3005 free_irq_at(irq, NULL);
3009 raw_spin_lock_irqsave(&vector_lock, flags);
3010 if (!__assign_irq_vector(irq, cfg, apic->target_cpus()))
3012 raw_spin_unlock_irqrestore(&vector_lock, flags);
3015 irq_set_chip_data(irq, cfg);
3016 irq_clear_status_flags(irq, IRQ_NOREQUEST);
3018 free_irq_at(irq, cfg);
3023 int create_irq(void)
3025 int node = cpu_to_node(0);
3026 unsigned int irq_want;
3029 irq_want = nr_irqs_gsi;
3030 irq = create_irq_nr(irq_want, node);
3038 void destroy_irq(unsigned int irq)
3040 struct irq_cfg *cfg = irq_get_chip_data(irq);
3041 unsigned long flags;
3043 irq_set_status_flags(irq, IRQ_NOREQUEST|IRQ_NOPROBE);
3045 if (irq_remapped(cfg))
3047 raw_spin_lock_irqsave(&vector_lock, flags);
3048 __clear_irq_vector(irq, cfg);
3049 raw_spin_unlock_irqrestore(&vector_lock, flags);
3050 free_irq_at(irq, cfg);
3054 * MSI message composition
3056 #ifdef CONFIG_PCI_MSI
3057 static int msi_compose_msg(struct pci_dev *pdev, unsigned int irq,
3058 struct msi_msg *msg, u8 hpet_id)
3060 struct irq_cfg *cfg;
3068 err = assign_irq_vector(irq, cfg, apic->target_cpus());
3072 dest = apic->cpu_mask_to_apicid_and(cfg->domain, apic->target_cpus());
3074 if (irq_remapped(cfg)) {
3079 ir_index = map_irq_to_irte_handle(irq, &sub_handle);
3080 BUG_ON(ir_index == -1);
3082 prepare_irte(&irte, cfg->vector, dest);
3084 /* Set source-id of interrupt request */
3086 set_msi_sid(&irte, pdev);
3088 set_hpet_sid(&irte, hpet_id);
3090 modify_irte(irq, &irte);
3092 msg->address_hi = MSI_ADDR_BASE_HI;
3093 msg->data = sub_handle;
3094 msg->address_lo = MSI_ADDR_BASE_LO | MSI_ADDR_IR_EXT_INT |
3096 MSI_ADDR_IR_INDEX1(ir_index) |
3097 MSI_ADDR_IR_INDEX2(ir_index);
3099 if (x2apic_enabled())
3100 msg->address_hi = MSI_ADDR_BASE_HI |
3101 MSI_ADDR_EXT_DEST_ID(dest);
3103 msg->address_hi = MSI_ADDR_BASE_HI;
3107 ((apic->irq_dest_mode == 0) ?
3108 MSI_ADDR_DEST_MODE_PHYSICAL:
3109 MSI_ADDR_DEST_MODE_LOGICAL) |
3110 ((apic->irq_delivery_mode != dest_LowestPrio) ?
3111 MSI_ADDR_REDIRECTION_CPU:
3112 MSI_ADDR_REDIRECTION_LOWPRI) |
3113 MSI_ADDR_DEST_ID(dest);
3116 MSI_DATA_TRIGGER_EDGE |
3117 MSI_DATA_LEVEL_ASSERT |
3118 ((apic->irq_delivery_mode != dest_LowestPrio) ?
3119 MSI_DATA_DELIVERY_FIXED:
3120 MSI_DATA_DELIVERY_LOWPRI) |
3121 MSI_DATA_VECTOR(cfg->vector);
3128 msi_set_affinity(struct irq_data *data, const struct cpumask *mask, bool force)
3130 struct irq_cfg *cfg = data->chip_data;
3134 if (__ioapic_set_affinity(data, mask, &dest))
3137 __get_cached_msi_msg(data->msi_desc, &msg);
3139 msg.data &= ~MSI_DATA_VECTOR_MASK;
3140 msg.data |= MSI_DATA_VECTOR(cfg->vector);
3141 msg.address_lo &= ~MSI_ADDR_DEST_ID_MASK;
3142 msg.address_lo |= MSI_ADDR_DEST_ID(dest);
3144 __write_msi_msg(data->msi_desc, &msg);
3148 #ifdef CONFIG_INTR_REMAP
3150 * Migrate the MSI irq to another cpumask. This migration is
3151 * done in the process context using interrupt-remapping hardware.
3154 ir_msi_set_affinity(struct irq_data *data, const struct cpumask *mask,
3157 struct irq_cfg *cfg = data->chip_data;
3158 unsigned int dest, irq = data->irq;
3161 if (get_irte(irq, &irte))
3164 if (__ioapic_set_affinity(data, mask, &dest))
3167 irte.vector = cfg->vector;
3168 irte.dest_id = IRTE_DEST(dest);
3171 * atomically update the IRTE with the new destination and vector.
3173 modify_irte(irq, &irte);
3176 * After this point, all the interrupts will start arriving
3177 * at the new destination. So, time to cleanup the previous
3178 * vector allocation.
3180 if (cfg->move_in_progress)
3181 send_cleanup_vector(cfg);
3187 #endif /* CONFIG_SMP */
3190 * IRQ Chip for MSI PCI/PCI-X/PCI-Express Devices,
3191 * which implement the MSI or MSI-X Capability Structure.
3193 static struct irq_chip msi_chip = {
3195 .irq_unmask = unmask_msi_irq,
3196 .irq_mask = mask_msi_irq,
3197 .irq_ack = ack_apic_edge,
3199 .irq_set_affinity = msi_set_affinity,
3201 .irq_retrigger = ioapic_retrigger_irq,
3204 static struct irq_chip msi_ir_chip = {
3205 .name = "IR-PCI-MSI",
3206 .irq_unmask = unmask_msi_irq,
3207 .irq_mask = mask_msi_irq,
3208 #ifdef CONFIG_INTR_REMAP
3209 .irq_ack = ir_ack_apic_edge,
3211 .irq_set_affinity = ir_msi_set_affinity,
3214 .irq_retrigger = ioapic_retrigger_irq,
3218 * Map the PCI dev to the corresponding remapping hardware unit
3219 * and allocate 'nvec' consecutive interrupt-remapping table entries
3222 static int msi_alloc_irte(struct pci_dev *dev, int irq, int nvec)
3224 struct intel_iommu *iommu;
3227 iommu = map_dev_to_ir(dev);
3230 "Unable to map PCI %s to iommu\n", pci_name(dev));
3234 index = alloc_irte(iommu, irq, nvec);
3237 "Unable to allocate %d IRTE for PCI %s\n", nvec,
3244 static int setup_msi_irq(struct pci_dev *dev, struct msi_desc *msidesc, int irq)
3246 struct irq_chip *chip = &msi_chip;
3250 ret = msi_compose_msg(dev, irq, &msg, -1);
3254 irq_set_msi_desc(irq, msidesc);
3255 write_msi_msg(irq, &msg);
3257 if (irq_remapped(irq_get_chip_data(irq))) {
3258 irq_set_status_flags(irq, IRQ_MOVE_PCNTXT);
3259 chip = &msi_ir_chip;
3262 irq_set_chip_and_handler_name(irq, chip, handle_edge_irq, "edge");
3264 dev_printk(KERN_DEBUG, &dev->dev, "irq %d for MSI/MSI-X\n", irq);
3269 int native_setup_msi_irqs(struct pci_dev *dev, int nvec, int type)
3271 int node, ret, sub_handle, index = 0;
3272 unsigned int irq, irq_want;
3273 struct msi_desc *msidesc;
3274 struct intel_iommu *iommu = NULL;
3276 /* x86 doesn't support multiple MSI yet */
3277 if (type == PCI_CAP_ID_MSI && nvec > 1)
3280 node = dev_to_node(&dev->dev);
3281 irq_want = nr_irqs_gsi;
3283 list_for_each_entry(msidesc, &dev->msi_list, list) {
3284 irq = create_irq_nr(irq_want, node);
3288 if (!intr_remapping_enabled)
3293 * allocate the consecutive block of IRTE's
3296 index = msi_alloc_irte(dev, irq, nvec);
3302 iommu = map_dev_to_ir(dev);
3308 * setup the mapping between the irq and the IRTE
3309 * base index, the sub_handle pointing to the
3310 * appropriate interrupt remap table entry.
3312 set_irte_irq(irq, iommu, index, sub_handle);
3315 ret = setup_msi_irq(dev, msidesc, irq);
3327 void native_teardown_msi_irq(unsigned int irq)
3332 #if defined (CONFIG_DMAR) || defined (CONFIG_INTR_REMAP)
3335 dmar_msi_set_affinity(struct irq_data *data, const struct cpumask *mask,
3338 struct irq_cfg *cfg = data->chip_data;
3339 unsigned int dest, irq = data->irq;
3342 if (__ioapic_set_affinity(data, mask, &dest))
3345 dmar_msi_read(irq, &msg);
3347 msg.data &= ~MSI_DATA_VECTOR_MASK;
3348 msg.data |= MSI_DATA_VECTOR(cfg->vector);
3349 msg.address_lo &= ~MSI_ADDR_DEST_ID_MASK;
3350 msg.address_lo |= MSI_ADDR_DEST_ID(dest);
3351 msg.address_hi = MSI_ADDR_BASE_HI | MSI_ADDR_EXT_DEST_ID(dest);
3353 dmar_msi_write(irq, &msg);
3358 #endif /* CONFIG_SMP */
3360 static struct irq_chip dmar_msi_type = {
3362 .irq_unmask = dmar_msi_unmask,
3363 .irq_mask = dmar_msi_mask,
3364 .irq_ack = ack_apic_edge,
3366 .irq_set_affinity = dmar_msi_set_affinity,
3368 .irq_retrigger = ioapic_retrigger_irq,
3371 int arch_setup_dmar_msi(unsigned int irq)
3376 ret = msi_compose_msg(NULL, irq, &msg, -1);
3379 dmar_msi_write(irq, &msg);
3380 irq_set_chip_and_handler_name(irq, &dmar_msi_type, handle_edge_irq,
3386 #ifdef CONFIG_HPET_TIMER
3389 static int hpet_msi_set_affinity(struct irq_data *data,
3390 const struct cpumask *mask, bool force)
3392 struct irq_cfg *cfg = data->chip_data;
3396 if (__ioapic_set_affinity(data, mask, &dest))
3399 hpet_msi_read(data->handler_data, &msg);
3401 msg.data &= ~MSI_DATA_VECTOR_MASK;
3402 msg.data |= MSI_DATA_VECTOR(cfg->vector);
3403 msg.address_lo &= ~MSI_ADDR_DEST_ID_MASK;
3404 msg.address_lo |= MSI_ADDR_DEST_ID(dest);
3406 hpet_msi_write(data->handler_data, &msg);
3411 #endif /* CONFIG_SMP */
3413 static struct irq_chip ir_hpet_msi_type = {
3414 .name = "IR-HPET_MSI",
3415 .irq_unmask = hpet_msi_unmask,
3416 .irq_mask = hpet_msi_mask,
3417 #ifdef CONFIG_INTR_REMAP
3418 .irq_ack = ir_ack_apic_edge,
3420 .irq_set_affinity = ir_msi_set_affinity,
3423 .irq_retrigger = ioapic_retrigger_irq,
3426 static struct irq_chip hpet_msi_type = {
3428 .irq_unmask = hpet_msi_unmask,
3429 .irq_mask = hpet_msi_mask,
3430 .irq_ack = ack_apic_edge,
3432 .irq_set_affinity = hpet_msi_set_affinity,
3434 .irq_retrigger = ioapic_retrigger_irq,
3437 int arch_setup_hpet_msi(unsigned int irq, unsigned int id)
3439 struct irq_chip *chip = &hpet_msi_type;
3443 if (intr_remapping_enabled) {
3444 struct intel_iommu *iommu = map_hpet_to_ir(id);
3450 index = alloc_irte(iommu, irq, 1);
3455 ret = msi_compose_msg(NULL, irq, &msg, id);
3459 hpet_msi_write(irq_get_handler_data(irq), &msg);
3460 irq_set_status_flags(irq, IRQ_MOVE_PCNTXT);
3461 if (irq_remapped(irq_get_chip_data(irq)))
3462 chip = &ir_hpet_msi_type;
3464 irq_set_chip_and_handler_name(irq, chip, handle_edge_irq, "edge");
3469 #endif /* CONFIG_PCI_MSI */
3471 * Hypertransport interrupt support
3473 #ifdef CONFIG_HT_IRQ
3477 static void target_ht_irq(unsigned int irq, unsigned int dest, u8 vector)
3479 struct ht_irq_msg msg;
3480 fetch_ht_irq_msg(irq, &msg);
3482 msg.address_lo &= ~(HT_IRQ_LOW_VECTOR_MASK | HT_IRQ_LOW_DEST_ID_MASK);
3483 msg.address_hi &= ~(HT_IRQ_HIGH_DEST_ID_MASK);
3485 msg.address_lo |= HT_IRQ_LOW_VECTOR(vector) | HT_IRQ_LOW_DEST_ID(dest);
3486 msg.address_hi |= HT_IRQ_HIGH_DEST_ID(dest);
3488 write_ht_irq_msg(irq, &msg);
3492 ht_set_affinity(struct irq_data *data, const struct cpumask *mask, bool force)
3494 struct irq_cfg *cfg = data->chip_data;
3497 if (__ioapic_set_affinity(data, mask, &dest))
3500 target_ht_irq(data->irq, dest, cfg->vector);
3506 static struct irq_chip ht_irq_chip = {
3508 .irq_mask = mask_ht_irq,
3509 .irq_unmask = unmask_ht_irq,
3510 .irq_ack = ack_apic_edge,
3512 .irq_set_affinity = ht_set_affinity,
3514 .irq_retrigger = ioapic_retrigger_irq,
3517 int arch_setup_ht_irq(unsigned int irq, struct pci_dev *dev)
3519 struct irq_cfg *cfg;
3526 err = assign_irq_vector(irq, cfg, apic->target_cpus());
3528 struct ht_irq_msg msg;
3531 dest = apic->cpu_mask_to_apicid_and(cfg->domain,
3532 apic->target_cpus());
3534 msg.address_hi = HT_IRQ_HIGH_DEST_ID(dest);
3538 HT_IRQ_LOW_DEST_ID(dest) |
3539 HT_IRQ_LOW_VECTOR(cfg->vector) |
3540 ((apic->irq_dest_mode == 0) ?
3541 HT_IRQ_LOW_DM_PHYSICAL :
3542 HT_IRQ_LOW_DM_LOGICAL) |
3543 HT_IRQ_LOW_RQEOI_EDGE |
3544 ((apic->irq_delivery_mode != dest_LowestPrio) ?
3545 HT_IRQ_LOW_MT_FIXED :
3546 HT_IRQ_LOW_MT_ARBITRATED) |
3547 HT_IRQ_LOW_IRQ_MASKED;
3549 write_ht_irq_msg(irq, &msg);
3551 irq_set_chip_and_handler_name(irq, &ht_irq_chip,
3552 handle_edge_irq, "edge");
3554 dev_printk(KERN_DEBUG, &dev->dev, "irq %d for HT\n", irq);
3558 #endif /* CONFIG_HT_IRQ */
3561 io_apic_setup_irq_pin(unsigned int irq, int node, struct io_apic_irq_attr *attr)
3563 struct irq_cfg *cfg = alloc_irq_and_cfg_at(irq, node);
3568 ret = __add_pin_to_irq_node(cfg, node, attr->ioapic, attr->ioapic_pin);
3570 setup_ioapic_irq(irq, cfg, attr);
3574 int io_apic_setup_irq_pin_once(unsigned int irq, int node,
3575 struct io_apic_irq_attr *attr)
3577 unsigned int id = attr->ioapic, pin = attr->ioapic_pin;
3580 /* Avoid redundant programming */
3581 if (test_bit(pin, ioapics[id].pin_programmed)) {
3582 pr_debug("Pin %d-%d already programmed\n",
3583 mpc_ioapic_id(id), pin);
3586 ret = io_apic_setup_irq_pin(irq, node, attr);
3588 set_bit(pin, ioapics[id].pin_programmed);
3592 static int __init io_apic_get_redir_entries(int ioapic)
3594 union IO_APIC_reg_01 reg_01;
3595 unsigned long flags;
3597 raw_spin_lock_irqsave(&ioapic_lock, flags);
3598 reg_01.raw = io_apic_read(ioapic, 1);
3599 raw_spin_unlock_irqrestore(&ioapic_lock, flags);
3601 /* The register returns the maximum index redir index
3602 * supported, which is one less than the total number of redir
3605 return reg_01.bits.entries + 1;
3608 static void __init probe_nr_irqs_gsi(void)
3612 nr = gsi_top + NR_IRQS_LEGACY;
3613 if (nr > nr_irqs_gsi)
3616 printk(KERN_DEBUG "nr_irqs_gsi: %d\n", nr_irqs_gsi);
3619 int get_nr_irqs_gsi(void)
3624 #ifdef CONFIG_SPARSE_IRQ
3625 int __init arch_probe_nr_irqs(void)
3629 if (nr_irqs > (NR_VECTORS * nr_cpu_ids))
3630 nr_irqs = NR_VECTORS * nr_cpu_ids;
3632 nr = nr_irqs_gsi + 8 * nr_cpu_ids;
3633 #if defined(CONFIG_PCI_MSI) || defined(CONFIG_HT_IRQ)
3635 * for MSI and HT dyn irq
3637 nr += nr_irqs_gsi * 16;
3642 return NR_IRQS_LEGACY;
3646 int io_apic_set_pci_routing(struct device *dev, int irq,
3647 struct io_apic_irq_attr *irq_attr)
3651 if (!IO_APIC_IRQ(irq)) {
3652 apic_printk(APIC_QUIET,KERN_ERR "IOAPIC[%d]: Invalid reference to IRQ 0\n",
3657 node = dev ? dev_to_node(dev) : cpu_to_node(0);
3659 return io_apic_setup_irq_pin_once(irq, node, irq_attr);
3662 #ifdef CONFIG_X86_32
3663 static int __init io_apic_get_unique_id(int ioapic, int apic_id)
3665 union IO_APIC_reg_00 reg_00;
3666 static physid_mask_t apic_id_map = PHYSID_MASK_NONE;
3668 unsigned long flags;
3672 * The P4 platform supports up to 256 APIC IDs on two separate APIC
3673 * buses (one for LAPICs, one for IOAPICs), where predecessors only
3674 * supports up to 16 on one shared APIC bus.
3676 * TBD: Expand LAPIC/IOAPIC support on P4-class systems to take full
3677 * advantage of new APIC bus architecture.
3680 if (physids_empty(apic_id_map))
3681 apic->ioapic_phys_id_map(&phys_cpu_present_map, &apic_id_map);
3683 raw_spin_lock_irqsave(&ioapic_lock, flags);
3684 reg_00.raw = io_apic_read(ioapic, 0);
3685 raw_spin_unlock_irqrestore(&ioapic_lock, flags);
3687 if (apic_id >= get_physical_broadcast()) {
3688 printk(KERN_WARNING "IOAPIC[%d]: Invalid apic_id %d, trying "
3689 "%d\n", ioapic, apic_id, reg_00.bits.ID);
3690 apic_id = reg_00.bits.ID;
3694 * Every APIC in a system must have a unique ID or we get lots of nice
3695 * 'stuck on smp_invalidate_needed IPI wait' messages.
3697 if (apic->check_apicid_used(&apic_id_map, apic_id)) {
3699 for (i = 0; i < get_physical_broadcast(); i++) {
3700 if (!apic->check_apicid_used(&apic_id_map, i))
3704 if (i == get_physical_broadcast())
3705 panic("Max apic_id exceeded!\n");
3707 printk(KERN_WARNING "IOAPIC[%d]: apic_id %d already used, "
3708 "trying %d\n", ioapic, apic_id, i);
3713 apic->apicid_to_cpu_present(apic_id, &tmp);
3714 physids_or(apic_id_map, apic_id_map, tmp);
3716 if (reg_00.bits.ID != apic_id) {
3717 reg_00.bits.ID = apic_id;
3719 raw_spin_lock_irqsave(&ioapic_lock, flags);
3720 io_apic_write(ioapic, 0, reg_00.raw);
3721 reg_00.raw = io_apic_read(ioapic, 0);
3722 raw_spin_unlock_irqrestore(&ioapic_lock, flags);
3725 if (reg_00.bits.ID != apic_id) {
3726 printk("IOAPIC[%d]: Unable to change apic_id!\n", ioapic);
3731 apic_printk(APIC_VERBOSE, KERN_INFO
3732 "IOAPIC[%d]: Assigned apic_id %d\n", ioapic, apic_id);
3737 static u8 __init io_apic_unique_id(u8 id)
3739 if ((boot_cpu_data.x86_vendor == X86_VENDOR_INTEL) &&
3740 !APIC_XAPIC(apic_version[boot_cpu_physical_apicid]))
3741 return io_apic_get_unique_id(nr_ioapics, id);
3746 static u8 __init io_apic_unique_id(u8 id)
3749 DECLARE_BITMAP(used, 256);
3751 bitmap_zero(used, 256);
3752 for (i = 0; i < nr_ioapics; i++) {
3753 __set_bit(mpc_ioapic_id(i), used);
3755 if (!test_bit(id, used))
3757 return find_first_zero_bit(used, 256);
3761 static int __init io_apic_get_version(int ioapic)
3763 union IO_APIC_reg_01 reg_01;
3764 unsigned long flags;
3766 raw_spin_lock_irqsave(&ioapic_lock, flags);
3767 reg_01.raw = io_apic_read(ioapic, 1);
3768 raw_spin_unlock_irqrestore(&ioapic_lock, flags);
3770 return reg_01.bits.version;
3773 int acpi_get_override_irq(u32 gsi, int *trigger, int *polarity)
3775 int ioapic, pin, idx;
3777 if (skip_ioapic_setup)
3780 ioapic = mp_find_ioapic(gsi);
3784 pin = mp_find_ioapic_pin(ioapic, gsi);
3788 idx = find_irq_entry(ioapic, pin, mp_INT);
3792 *trigger = irq_trigger(idx);
3793 *polarity = irq_polarity(idx);
3798 * This function currently is only a helper for the i386 smp boot process where
3799 * we need to reprogram the ioredtbls to cater for the cpus which have come online
3800 * so mask in all cases should simply be apic->target_cpus()
3803 void __init setup_ioapic_dest(void)
3805 int pin, ioapic, irq, irq_entry;
3806 const struct cpumask *mask;
3807 struct irq_data *idata;
3809 if (skip_ioapic_setup == 1)
3812 for (ioapic = 0; ioapic < nr_ioapics; ioapic++)
3813 for (pin = 0; pin < ioapics[ioapic].nr_registers; pin++) {
3814 irq_entry = find_irq_entry(ioapic, pin, mp_INT);
3815 if (irq_entry == -1)
3817 irq = pin_2_irq(irq_entry, ioapic, pin);
3819 if ((ioapic > 0) && (irq > 16))
3822 idata = irq_get_irq_data(irq);
3825 * Honour affinities which have been set in early boot
3827 if (!irqd_can_balance(idata) || irqd_affinity_was_set(idata))
3828 mask = idata->affinity;
3830 mask = apic->target_cpus();
3832 if (intr_remapping_enabled)
3833 ir_ioapic_set_affinity(idata, mask, false);
3835 ioapic_set_affinity(idata, mask, false);
3841 #define IOAPIC_RESOURCE_NAME_SIZE 11
3843 static struct resource *ioapic_resources;
3845 static struct resource * __init ioapic_setup_resources(int nr_ioapics)
3848 struct resource *res;
3852 if (nr_ioapics <= 0)
3855 n = IOAPIC_RESOURCE_NAME_SIZE + sizeof(struct resource);
3858 mem = alloc_bootmem(n);
3861 mem += sizeof(struct resource) * nr_ioapics;
3863 for (i = 0; i < nr_ioapics; i++) {
3865 res[i].flags = IORESOURCE_MEM | IORESOURCE_BUSY;
3866 snprintf(mem, IOAPIC_RESOURCE_NAME_SIZE, "IOAPIC %u", i);
3867 mem += IOAPIC_RESOURCE_NAME_SIZE;
3870 ioapic_resources = res;
3875 void __init ioapic_and_gsi_init(void)
3877 unsigned long ioapic_phys, idx = FIX_IO_APIC_BASE_0;
3878 struct resource *ioapic_res;
3881 ioapic_res = ioapic_setup_resources(nr_ioapics);
3882 for (i = 0; i < nr_ioapics; i++) {
3883 if (smp_found_config) {
3884 ioapic_phys = mpc_ioapic_addr(i);
3885 #ifdef CONFIG_X86_32
3888 "WARNING: bogus zero IO-APIC "
3889 "address found in MPTABLE, "
3890 "disabling IO/APIC support!\n");
3891 smp_found_config = 0;
3892 skip_ioapic_setup = 1;
3893 goto fake_ioapic_page;
3897 #ifdef CONFIG_X86_32
3900 ioapic_phys = (unsigned long)alloc_bootmem_pages(PAGE_SIZE);
3901 ioapic_phys = __pa(ioapic_phys);
3903 set_fixmap_nocache(idx, ioapic_phys);
3904 apic_printk(APIC_VERBOSE, "mapped IOAPIC to %08lx (%08lx)\n",
3905 __fix_to_virt(idx) + (ioapic_phys & ~PAGE_MASK),
3909 ioapic_res->start = ioapic_phys;
3910 ioapic_res->end = ioapic_phys + IO_APIC_SLOT_SIZE - 1;
3914 probe_nr_irqs_gsi();
3917 void __init ioapic_insert_resources(void)
3920 struct resource *r = ioapic_resources;
3925 "IO APIC resources couldn't be allocated.\n");
3929 for (i = 0; i < nr_ioapics; i++) {
3930 insert_resource(&iomem_resource, r);
3935 int mp_find_ioapic(u32 gsi)
3939 if (nr_ioapics == 0)
3942 /* Find the IOAPIC that manages this GSI. */
3943 for (i = 0; i < nr_ioapics; i++) {
3944 struct mp_ioapic_gsi *gsi_cfg = mp_ioapic_gsi_routing(i);
3945 if ((gsi >= gsi_cfg->gsi_base)
3946 && (gsi <= gsi_cfg->gsi_end))
3950 printk(KERN_ERR "ERROR: Unable to locate IOAPIC for GSI %d\n", gsi);
3954 int mp_find_ioapic_pin(int ioapic, u32 gsi)
3956 struct mp_ioapic_gsi *gsi_cfg;
3958 if (WARN_ON(ioapic == -1))
3961 gsi_cfg = mp_ioapic_gsi_routing(ioapic);
3962 if (WARN_ON(gsi > gsi_cfg->gsi_end))
3965 return gsi - gsi_cfg->gsi_base;
3968 static __init int bad_ioapic(unsigned long address)
3970 if (nr_ioapics >= MAX_IO_APICS) {
3971 printk(KERN_WARNING "WARNING: Max # of I/O APICs (%d) exceeded "
3972 "(found %d), skipping\n", MAX_IO_APICS, nr_ioapics);
3976 printk(KERN_WARNING "WARNING: Bogus (zero) I/O APIC address"
3977 " found in table, skipping!\n");
3983 void __init mp_register_ioapic(int id, u32 address, u32 gsi_base)
3987 struct mp_ioapic_gsi *gsi_cfg;
3989 if (bad_ioapic(address))
3994 ioapics[idx].mp_config.type = MP_IOAPIC;
3995 ioapics[idx].mp_config.flags = MPC_APIC_USABLE;
3996 ioapics[idx].mp_config.apicaddr = address;
3998 set_fixmap_nocache(FIX_IO_APIC_BASE_0 + idx, address);
3999 ioapics[idx].mp_config.apicid = io_apic_unique_id(id);
4000 ioapics[idx].mp_config.apicver = io_apic_get_version(idx);
4003 * Build basic GSI lookup table to facilitate gsi->io_apic lookups
4004 * and to prevent reprogramming of IOAPIC pins (PCI GSIs).
4006 entries = io_apic_get_redir_entries(idx);
4007 gsi_cfg = mp_ioapic_gsi_routing(idx);
4008 gsi_cfg->gsi_base = gsi_base;
4009 gsi_cfg->gsi_end = gsi_base + entries - 1;
4012 * The number of IO-APIC IRQ registers (== #pins):
4014 ioapics[idx].nr_registers = entries;
4016 if (gsi_cfg->gsi_end >= gsi_top)
4017 gsi_top = gsi_cfg->gsi_end + 1;
4019 printk(KERN_INFO "IOAPIC[%d]: apic_id %d, version %d, address 0x%x, "
4020 "GSI %d-%d\n", idx, mpc_ioapic_id(idx),
4021 mpc_ioapic_ver(idx), mpc_ioapic_addr(idx),
4022 gsi_cfg->gsi_base, gsi_cfg->gsi_end);
4027 /* Enable IOAPIC early just for system timer */
4028 void __init pre_init_apic_IRQ0(void)
4030 struct io_apic_irq_attr attr = { 0, 0, 0, 0 };
4032 printk(KERN_INFO "Early APIC setup for system timer0\n");
4034 physid_set_mask_of_physid(boot_cpu_physical_apicid,
4035 &phys_cpu_present_map);
4039 io_apic_setup_irq_pin(0, 0, &attr);
4040 irq_set_chip_and_handler_name(0, &ioapic_chip, handle_edge_irq,