2 * Intel IO-APIC support for multi-Pentium hosts.
4 * Copyright (C) 1997, 1998, 1999, 2000 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/acpi.h>
31 #include <linux/sysdev.h>
32 #include <linux/msi.h>
33 #include <linux/htirq.h>
34 #include <linux/dmar.h>
35 #include <linux/jiffies.h>
37 #include <acpi/acpi_bus.h>
39 #include <linux/bootmem.h>
45 #include <asm/proto.h>
49 #include <asm/msidef.h>
50 #include <asm/hypertransport.h>
53 #include <mach_apic.h>
58 unsigned move_cleanup_count;
60 u8 move_in_progress : 1;
63 /* irq_cfg is indexed by the sum of all RTEs in all I/O APICs. */
64 struct irq_cfg irq_cfg[NR_IRQS] __read_mostly = {
65 [0] = { .domain = CPU_MASK_ALL, .vector = IRQ0_VECTOR, },
66 [1] = { .domain = CPU_MASK_ALL, .vector = IRQ1_VECTOR, },
67 [2] = { .domain = CPU_MASK_ALL, .vector = IRQ2_VECTOR, },
68 [3] = { .domain = CPU_MASK_ALL, .vector = IRQ3_VECTOR, },
69 [4] = { .domain = CPU_MASK_ALL, .vector = IRQ4_VECTOR, },
70 [5] = { .domain = CPU_MASK_ALL, .vector = IRQ5_VECTOR, },
71 [6] = { .domain = CPU_MASK_ALL, .vector = IRQ6_VECTOR, },
72 [7] = { .domain = CPU_MASK_ALL, .vector = IRQ7_VECTOR, },
73 [8] = { .domain = CPU_MASK_ALL, .vector = IRQ8_VECTOR, },
74 [9] = { .domain = CPU_MASK_ALL, .vector = IRQ9_VECTOR, },
75 [10] = { .domain = CPU_MASK_ALL, .vector = IRQ10_VECTOR, },
76 [11] = { .domain = CPU_MASK_ALL, .vector = IRQ11_VECTOR, },
77 [12] = { .domain = CPU_MASK_ALL, .vector = IRQ12_VECTOR, },
78 [13] = { .domain = CPU_MASK_ALL, .vector = IRQ13_VECTOR, },
79 [14] = { .domain = CPU_MASK_ALL, .vector = IRQ14_VECTOR, },
80 [15] = { .domain = CPU_MASK_ALL, .vector = IRQ15_VECTOR, },
83 static int assign_irq_vector(int irq, cpumask_t mask);
85 #define __apicdebuginit __init
87 int sis_apic_bug; /* not actually supported, dummy for compile */
89 static int no_timer_check;
91 static int disable_timer_pin_1 __initdata;
93 int timer_over_8254 __initdata = 1;
95 /* Where if anywhere is the i8259 connect in external int mode */
96 static struct { int pin, apic; } ioapic_i8259 = { -1, -1 };
98 static DEFINE_SPINLOCK(ioapic_lock);
99 DEFINE_SPINLOCK(vector_lock);
102 * # of IRQ routing registers
104 int nr_ioapic_registers[MAX_IO_APICS];
106 /* I/O APIC entries */
107 struct mp_config_ioapic mp_ioapics[MAX_IO_APICS];
110 /* MP IRQ source entries */
111 struct mp_config_intsrc mp_irqs[MAX_IRQ_SOURCES];
113 /* # of MP IRQ source entries */
116 DECLARE_BITMAP(mp_bus_not_pci, MAX_MP_BUSSES);
119 * Rough estimation of how many shared IRQs there are, can
120 * be changed anytime.
122 #define MAX_PLUS_SHARED_IRQS NR_IRQS
123 #define PIN_MAP_SIZE (MAX_PLUS_SHARED_IRQS + NR_IRQS)
126 * This is performance-critical, we want to do it O(1)
128 * the indexing order of this array favors 1:1 mappings
129 * between pins and IRQs.
132 static struct irq_pin_list {
133 short apic, pin, next;
134 } irq_2_pin[PIN_MAP_SIZE];
138 unsigned int unused[3];
142 static __attribute_const__ struct io_apic __iomem *io_apic_base(int idx)
144 return (void __iomem *) __fix_to_virt(FIX_IO_APIC_BASE_0 + idx)
145 + (mp_ioapics[idx].mp_apicaddr & ~PAGE_MASK);
148 static inline unsigned int io_apic_read(unsigned int apic, unsigned int reg)
150 struct io_apic __iomem *io_apic = io_apic_base(apic);
151 writel(reg, &io_apic->index);
152 return readl(&io_apic->data);
155 static inline void io_apic_write(unsigned int apic, unsigned int reg, unsigned int value)
157 struct io_apic __iomem *io_apic = io_apic_base(apic);
158 writel(reg, &io_apic->index);
159 writel(value, &io_apic->data);
163 * Re-write a value: to be used for read-modify-write
164 * cycles where the read already set up the index register.
166 static inline void io_apic_modify(unsigned int apic, unsigned int value)
168 struct io_apic __iomem *io_apic = io_apic_base(apic);
169 writel(value, &io_apic->data);
172 static bool io_apic_level_ack_pending(unsigned int irq)
174 struct irq_pin_list *entry;
177 spin_lock_irqsave(&ioapic_lock, flags);
178 entry = irq_2_pin + irq;
186 reg = io_apic_read(entry->apic, 0x10 + pin*2);
187 /* Is the remote IRR bit set? */
188 if ((reg >> 14) & 1) {
189 spin_unlock_irqrestore(&ioapic_lock, flags);
194 entry = irq_2_pin + entry->next;
196 spin_unlock_irqrestore(&ioapic_lock, flags);
202 * Synchronize the IO-APIC and the CPU by doing
203 * a dummy read from the IO-APIC
205 static inline void io_apic_sync(unsigned int apic)
207 struct io_apic __iomem *io_apic = io_apic_base(apic);
208 readl(&io_apic->data);
211 #define __DO_ACTION(R, ACTION, FINAL) \
215 struct irq_pin_list *entry = irq_2_pin + irq; \
217 BUG_ON(irq >= NR_IRQS); \
223 reg = io_apic_read(entry->apic, 0x10 + R + pin*2); \
225 io_apic_modify(entry->apic, reg); \
229 entry = irq_2_pin + entry->next; \
234 struct { u32 w1, w2; };
235 struct IO_APIC_route_entry entry;
238 static struct IO_APIC_route_entry ioapic_read_entry(int apic, int pin)
240 union entry_union eu;
242 spin_lock_irqsave(&ioapic_lock, flags);
243 eu.w1 = io_apic_read(apic, 0x10 + 2 * pin);
244 eu.w2 = io_apic_read(apic, 0x11 + 2 * pin);
245 spin_unlock_irqrestore(&ioapic_lock, flags);
250 * When we write a new IO APIC routing entry, we need to write the high
251 * word first! If the mask bit in the low word is clear, we will enable
252 * the interrupt, and we need to make sure the entry is fully populated
253 * before that happens.
256 __ioapic_write_entry(int apic, int pin, struct IO_APIC_route_entry e)
258 union entry_union eu;
260 io_apic_write(apic, 0x11 + 2*pin, eu.w2);
261 io_apic_write(apic, 0x10 + 2*pin, eu.w1);
264 static void ioapic_write_entry(int apic, int pin, struct IO_APIC_route_entry e)
267 spin_lock_irqsave(&ioapic_lock, flags);
268 __ioapic_write_entry(apic, pin, e);
269 spin_unlock_irqrestore(&ioapic_lock, flags);
273 * When we mask an IO APIC routing entry, we need to write the low
274 * word first, in order to set the mask bit before we change the
277 static void ioapic_mask_entry(int apic, int pin)
280 union entry_union eu = { .entry.mask = 1 };
282 spin_lock_irqsave(&ioapic_lock, flags);
283 io_apic_write(apic, 0x10 + 2*pin, eu.w1);
284 io_apic_write(apic, 0x11 + 2*pin, eu.w2);
285 spin_unlock_irqrestore(&ioapic_lock, flags);
289 static void __target_IO_APIC_irq(unsigned int irq, unsigned int dest, u8 vector)
292 struct irq_pin_list *entry = irq_2_pin + irq;
294 BUG_ON(irq >= NR_IRQS);
301 io_apic_write(apic, 0x11 + pin*2, dest);
302 reg = io_apic_read(apic, 0x10 + pin*2);
305 io_apic_modify(apic, reg);
308 entry = irq_2_pin + entry->next;
312 static void set_ioapic_affinity_irq(unsigned int irq, cpumask_t mask)
314 struct irq_cfg *cfg = irq_cfg + irq;
319 cpus_and(tmp, mask, cpu_online_map);
323 if (assign_irq_vector(irq, mask))
326 cpus_and(tmp, cfg->domain, mask);
327 dest = cpu_mask_to_apicid(tmp);
330 * Only the high 8 bits are valid.
332 dest = SET_APIC_LOGICAL_ID(dest);
334 spin_lock_irqsave(&ioapic_lock, flags);
335 __target_IO_APIC_irq(irq, dest, cfg->vector);
336 irq_desc[irq].affinity = mask;
337 spin_unlock_irqrestore(&ioapic_lock, flags);
342 * The common case is 1:1 IRQ<->pin mappings. Sometimes there are
343 * shared ISA-space IRQs, so we have to support them. We are super
344 * fast in the common case, and fast for shared ISA-space IRQs.
346 static void add_pin_to_irq(unsigned int irq, int apic, int pin)
348 static int first_free_entry = NR_IRQS;
349 struct irq_pin_list *entry = irq_2_pin + irq;
351 BUG_ON(irq >= NR_IRQS);
353 entry = irq_2_pin + entry->next;
355 if (entry->pin != -1) {
356 entry->next = first_free_entry;
357 entry = irq_2_pin + entry->next;
358 if (++first_free_entry >= PIN_MAP_SIZE)
359 panic("io_apic.c: ran out of irq_2_pin entries!");
366 #define DO_ACTION(name,R,ACTION, FINAL) \
368 static void name##_IO_APIC_irq (unsigned int irq) \
369 __DO_ACTION(R, ACTION, FINAL)
371 DO_ACTION( __mask, 0, |= 0x00010000, io_apic_sync(entry->apic) )
373 DO_ACTION( __unmask, 0, &= 0xfffeffff, )
376 static void mask_IO_APIC_irq (unsigned int irq)
380 spin_lock_irqsave(&ioapic_lock, flags);
381 __mask_IO_APIC_irq(irq);
382 spin_unlock_irqrestore(&ioapic_lock, flags);
385 static void unmask_IO_APIC_irq (unsigned int irq)
389 spin_lock_irqsave(&ioapic_lock, flags);
390 __unmask_IO_APIC_irq(irq);
391 spin_unlock_irqrestore(&ioapic_lock, flags);
394 static void clear_IO_APIC_pin(unsigned int apic, unsigned int pin)
396 struct IO_APIC_route_entry entry;
398 /* Check delivery_mode to be sure we're not clearing an SMI pin */
399 entry = ioapic_read_entry(apic, pin);
400 if (entry.delivery_mode == dest_SMI)
403 * Disable it in the IO-APIC irq-routing table:
405 ioapic_mask_entry(apic, pin);
408 static void clear_IO_APIC (void)
412 for (apic = 0; apic < nr_ioapics; apic++)
413 for (pin = 0; pin < nr_ioapic_registers[apic]; pin++)
414 clear_IO_APIC_pin(apic, pin);
417 int skip_ioapic_setup;
420 static int __init parse_noapic(char *str)
422 disable_ioapic_setup();
425 early_param("noapic", parse_noapic);
427 /* Actually the next is obsolete, but keep it for paranoid reasons -AK */
428 static int __init disable_timer_pin_setup(char *arg)
430 disable_timer_pin_1 = 1;
433 __setup("disable_timer_pin_1", disable_timer_pin_setup);
435 static int __init setup_disable_8254_timer(char *s)
437 timer_over_8254 = -1;
440 static int __init setup_enable_8254_timer(char *s)
446 __setup("disable_8254_timer", setup_disable_8254_timer);
447 __setup("enable_8254_timer", setup_enable_8254_timer);
451 * Find the IRQ entry number of a certain pin.
453 static int find_irq_entry(int apic, int pin, int type)
457 for (i = 0; i < mp_irq_entries; i++)
458 if (mp_irqs[i].mp_irqtype == type &&
459 (mp_irqs[i].mp_dstapic == mp_ioapics[apic].mp_apicid ||
460 mp_irqs[i].mp_dstapic == MP_APIC_ALL) &&
461 mp_irqs[i].mp_dstirq == pin)
468 * Find the pin to which IRQ[irq] (ISA) is connected
470 static int __init find_isa_irq_pin(int irq, int type)
474 for (i = 0; i < mp_irq_entries; i++) {
475 int lbus = mp_irqs[i].mp_srcbus;
477 if (test_bit(lbus, mp_bus_not_pci) &&
478 (mp_irqs[i].mp_irqtype == type) &&
479 (mp_irqs[i].mp_srcbusirq == irq))
481 return mp_irqs[i].mp_dstirq;
486 static int __init find_isa_irq_apic(int irq, int type)
490 for (i = 0; i < mp_irq_entries; i++) {
491 int lbus = mp_irqs[i].mp_srcbus;
493 if (test_bit(lbus, mp_bus_not_pci) &&
494 (mp_irqs[i].mp_irqtype == type) &&
495 (mp_irqs[i].mp_srcbusirq == irq))
498 if (i < mp_irq_entries) {
500 for(apic = 0; apic < nr_ioapics; apic++) {
501 if (mp_ioapics[apic].mp_apicid == mp_irqs[i].mp_dstapic)
510 * Find a specific PCI IRQ entry.
511 * Not an __init, possibly needed by modules
513 static int pin_2_irq(int idx, int apic, int pin);
515 int IO_APIC_get_PCI_irq_vector(int bus, int slot, int pin)
517 int apic, i, best_guess = -1;
519 apic_printk(APIC_DEBUG, "querying PCI -> IRQ mapping bus:%d, slot:%d, pin:%d.\n",
521 if (test_bit(bus, mp_bus_not_pci)) {
522 apic_printk(APIC_VERBOSE, "PCI BIOS passed nonexistent PCI bus %d!\n", bus);
525 for (i = 0; i < mp_irq_entries; i++) {
526 int lbus = mp_irqs[i].mp_srcbus;
528 for (apic = 0; apic < nr_ioapics; apic++)
529 if (mp_ioapics[apic].mp_apicid == mp_irqs[i].mp_dstapic ||
530 mp_irqs[i].mp_dstapic == MP_APIC_ALL)
533 if (!test_bit(lbus, mp_bus_not_pci) &&
534 !mp_irqs[i].mp_irqtype &&
536 (slot == ((mp_irqs[i].mp_srcbusirq >> 2) & 0x1f))) {
537 int irq = pin_2_irq(i,apic,mp_irqs[i].mp_dstirq);
539 if (!(apic || IO_APIC_IRQ(irq)))
542 if (pin == (mp_irqs[i].mp_srcbusirq & 3))
545 * Use the first all-but-pin matching entry as a
546 * best-guess fuzzy result for broken mptables.
552 BUG_ON(best_guess >= NR_IRQS);
556 /* ISA interrupts are always polarity zero edge triggered,
557 * when listed as conforming in the MP table. */
559 #define default_ISA_trigger(idx) (0)
560 #define default_ISA_polarity(idx) (0)
562 /* PCI interrupts are always polarity one level triggered,
563 * when listed as conforming in the MP table. */
565 #define default_PCI_trigger(idx) (1)
566 #define default_PCI_polarity(idx) (1)
568 static int MPBIOS_polarity(int idx)
570 int bus = mp_irqs[idx].mp_srcbus;
574 * Determine IRQ line polarity (high active or low active):
576 switch (mp_irqs[idx].mp_irqflag & 3)
578 case 0: /* conforms, ie. bus-type dependent polarity */
579 if (test_bit(bus, mp_bus_not_pci))
580 polarity = default_ISA_polarity(idx);
582 polarity = default_PCI_polarity(idx);
584 case 1: /* high active */
589 case 2: /* reserved */
591 printk(KERN_WARNING "broken BIOS!!\n");
595 case 3: /* low active */
600 default: /* invalid */
602 printk(KERN_WARNING "broken BIOS!!\n");
610 static int MPBIOS_trigger(int idx)
612 int bus = mp_irqs[idx].mp_srcbus;
616 * Determine IRQ trigger mode (edge or level sensitive):
618 switch ((mp_irqs[idx].mp_irqflag>>2) & 3)
620 case 0: /* conforms, ie. bus-type dependent */
621 if (test_bit(bus, mp_bus_not_pci))
622 trigger = default_ISA_trigger(idx);
624 trigger = default_PCI_trigger(idx);
631 case 2: /* reserved */
633 printk(KERN_WARNING "broken BIOS!!\n");
642 default: /* invalid */
644 printk(KERN_WARNING "broken BIOS!!\n");
652 static inline int irq_polarity(int idx)
654 return MPBIOS_polarity(idx);
657 static inline int irq_trigger(int idx)
659 return MPBIOS_trigger(idx);
662 static int pin_2_irq(int idx, int apic, int pin)
665 int bus = mp_irqs[idx].mp_srcbus;
668 * Debugging check, we are in big trouble if this message pops up!
670 if (mp_irqs[idx].mp_dstirq != pin)
671 printk(KERN_ERR "broken BIOS or MPTABLE parser, ayiee!!\n");
673 if (test_bit(bus, mp_bus_not_pci)) {
674 irq = mp_irqs[idx].mp_srcbusirq;
677 * PCI IRQs are mapped in order
681 irq += nr_ioapic_registers[i++];
684 BUG_ON(irq >= NR_IRQS);
688 static int __assign_irq_vector(int irq, cpumask_t mask)
691 * NOTE! The local APIC isn't very good at handling
692 * multiple interrupts at the same interrupt level.
693 * As the interrupt level is determined by taking the
694 * vector number and shifting that right by 4, we
695 * want to spread these out a bit so that they don't
696 * all fall in the same interrupt level.
698 * Also, we've got to be careful not to trash gate
699 * 0x80, because int 0x80 is hm, kind of importantish. ;)
701 static int current_vector = FIRST_DEVICE_VECTOR, current_offset = 0;
702 unsigned int old_vector;
706 BUG_ON((unsigned)irq >= NR_IRQS);
709 /* Only try and allocate irqs on cpus that are present */
710 cpus_and(mask, mask, cpu_online_map);
712 if ((cfg->move_in_progress) || cfg->move_cleanup_count)
715 old_vector = cfg->vector;
718 cpus_and(tmp, cfg->domain, mask);
719 if (!cpus_empty(tmp))
723 for_each_cpu_mask(cpu, mask) {
724 cpumask_t domain, new_mask;
728 domain = vector_allocation_domain(cpu);
729 cpus_and(new_mask, domain, cpu_online_map);
731 vector = current_vector;
732 offset = current_offset;
735 if (vector >= FIRST_SYSTEM_VECTOR) {
736 /* If we run out of vectors on large boxen, must share them. */
737 offset = (offset + 1) % 8;
738 vector = FIRST_DEVICE_VECTOR + offset;
740 if (unlikely(current_vector == vector))
742 if (vector == IA32_SYSCALL_VECTOR)
744 for_each_cpu_mask(new_cpu, new_mask)
745 if (per_cpu(vector_irq, new_cpu)[vector] != -1)
748 current_vector = vector;
749 current_offset = offset;
751 cfg->move_in_progress = 1;
752 cfg->old_domain = cfg->domain;
754 for_each_cpu_mask(new_cpu, new_mask)
755 per_cpu(vector_irq, new_cpu)[vector] = irq;
756 cfg->vector = vector;
757 cfg->domain = domain;
763 static int assign_irq_vector(int irq, cpumask_t mask)
768 spin_lock_irqsave(&vector_lock, flags);
769 err = __assign_irq_vector(irq, mask);
770 spin_unlock_irqrestore(&vector_lock, flags);
774 static void __clear_irq_vector(int irq)
780 BUG_ON((unsigned)irq >= NR_IRQS);
782 BUG_ON(!cfg->vector);
784 vector = cfg->vector;
785 cpus_and(mask, cfg->domain, cpu_online_map);
786 for_each_cpu_mask(cpu, mask)
787 per_cpu(vector_irq, cpu)[vector] = -1;
790 cpus_clear(cfg->domain);
793 void __setup_vector_irq(int cpu)
795 /* Initialize vector_irq on a new cpu */
796 /* This function must be called with vector_lock held */
799 /* Mark the inuse vectors */
800 for (irq = 0; irq < NR_IRQS; ++irq) {
801 if (!cpu_isset(cpu, irq_cfg[irq].domain))
803 vector = irq_cfg[irq].vector;
804 per_cpu(vector_irq, cpu)[vector] = irq;
806 /* Mark the free vectors */
807 for (vector = 0; vector < NR_VECTORS; ++vector) {
808 irq = per_cpu(vector_irq, cpu)[vector];
811 if (!cpu_isset(cpu, irq_cfg[irq].domain))
812 per_cpu(vector_irq, cpu)[vector] = -1;
817 static struct irq_chip ioapic_chip;
819 static void ioapic_register_intr(int irq, unsigned long trigger)
822 irq_desc[irq].status |= IRQ_LEVEL;
823 set_irq_chip_and_handler_name(irq, &ioapic_chip,
824 handle_fasteoi_irq, "fasteoi");
826 irq_desc[irq].status &= ~IRQ_LEVEL;
827 set_irq_chip_and_handler_name(irq, &ioapic_chip,
828 handle_edge_irq, "edge");
832 static void setup_IO_APIC_irq(int apic, int pin, unsigned int irq,
833 int trigger, int polarity)
835 struct irq_cfg *cfg = irq_cfg + irq;
836 struct IO_APIC_route_entry entry;
839 if (!IO_APIC_IRQ(irq))
843 if (assign_irq_vector(irq, mask))
846 cpus_and(mask, cfg->domain, mask);
848 apic_printk(APIC_VERBOSE,KERN_DEBUG
849 "IOAPIC[%d]: Set routing entry (%d-%d -> 0x%x -> "
850 "IRQ %d Mode:%i Active:%i)\n",
851 apic, mp_ioapics[apic].mp_apicid, pin, cfg->vector,
852 irq, trigger, polarity);
855 * add it to the IO-APIC irq-routing table:
857 memset(&entry,0,sizeof(entry));
859 entry.delivery_mode = INT_DELIVERY_MODE;
860 entry.dest_mode = INT_DEST_MODE;
861 entry.dest = cpu_mask_to_apicid(mask);
862 entry.mask = 0; /* enable IRQ */
863 entry.trigger = trigger;
864 entry.polarity = polarity;
865 entry.vector = cfg->vector;
867 /* Mask level triggered irqs.
868 * Use IRQ_DELAYED_DISABLE for edge triggered irqs.
873 ioapic_register_intr(irq, trigger);
875 disable_8259A_irq(irq);
877 ioapic_write_entry(apic, pin, entry);
880 static void __init setup_IO_APIC_irqs(void)
882 int apic, pin, idx, irq, first_notcon = 1;
884 apic_printk(APIC_VERBOSE, KERN_DEBUG "init IO_APIC IRQs\n");
886 for (apic = 0; apic < nr_ioapics; apic++) {
887 for (pin = 0; pin < nr_ioapic_registers[apic]; pin++) {
889 idx = find_irq_entry(apic,pin,mp_INT);
892 apic_printk(APIC_VERBOSE, KERN_DEBUG " IO-APIC (apicid-pin) %d-%d", mp_ioapics[apic].mp_apicid, pin);
895 apic_printk(APIC_VERBOSE, ", %d-%d", mp_ioapics[apic].mp_apicid, pin);
899 apic_printk(APIC_VERBOSE, " not connected.\n");
903 irq = pin_2_irq(idx, apic, pin);
904 add_pin_to_irq(irq, apic, pin);
906 setup_IO_APIC_irq(apic, pin, irq,
907 irq_trigger(idx), irq_polarity(idx));
912 apic_printk(APIC_VERBOSE, " not connected.\n");
916 * Set up the 8259A-master output pin as broadcast to all
919 static void __init setup_ExtINT_IRQ0_pin(unsigned int apic, unsigned int pin, int vector)
921 struct IO_APIC_route_entry entry;
923 memset(&entry, 0, sizeof(entry));
925 disable_8259A_irq(0);
928 apic_write(APIC_LVT0, APIC_LVT_MASKED | APIC_DM_EXTINT);
931 * We use logical delivery to get the timer IRQ
934 entry.dest_mode = INT_DEST_MODE;
935 entry.mask = 0; /* unmask IRQ now */
936 entry.dest = cpu_mask_to_apicid(TARGET_CPUS);
937 entry.delivery_mode = INT_DELIVERY_MODE;
940 entry.vector = vector;
943 * The timer IRQ doesn't have to know that behind the
944 * scene we have a 8259A-master in AEOI mode ...
946 set_irq_chip_and_handler_name(0, &ioapic_chip, handle_edge_irq, "edge");
949 * Add it to the IO-APIC irq-routing table:
951 ioapic_write_entry(apic, pin, entry);
956 void __apicdebuginit print_IO_APIC(void)
959 union IO_APIC_reg_00 reg_00;
960 union IO_APIC_reg_01 reg_01;
961 union IO_APIC_reg_02 reg_02;
964 if (apic_verbosity == APIC_QUIET)
967 printk(KERN_DEBUG "number of MP IRQ sources: %d.\n", mp_irq_entries);
968 for (i = 0; i < nr_ioapics; i++)
969 printk(KERN_DEBUG "number of IO-APIC #%d registers: %d.\n",
970 mp_ioapics[i].mp_apicid, nr_ioapic_registers[i]);
973 * We are a bit conservative about what we expect. We have to
974 * know about every hardware change ASAP.
976 printk(KERN_INFO "testing the IO APIC.......................\n");
978 for (apic = 0; apic < nr_ioapics; apic++) {
980 spin_lock_irqsave(&ioapic_lock, flags);
981 reg_00.raw = io_apic_read(apic, 0);
982 reg_01.raw = io_apic_read(apic, 1);
983 if (reg_01.bits.version >= 0x10)
984 reg_02.raw = io_apic_read(apic, 2);
985 spin_unlock_irqrestore(&ioapic_lock, flags);
988 printk(KERN_DEBUG "IO APIC #%d......\n", mp_ioapics[apic].mp_apicid);
989 printk(KERN_DEBUG ".... register #00: %08X\n", reg_00.raw);
990 printk(KERN_DEBUG "....... : physical APIC id: %02X\n", reg_00.bits.ID);
992 printk(KERN_DEBUG ".... register #01: %08X\n", *(int *)®_01);
993 printk(KERN_DEBUG "....... : max redirection entries: %04X\n", reg_01.bits.entries);
995 printk(KERN_DEBUG "....... : PRQ implemented: %X\n", reg_01.bits.PRQ);
996 printk(KERN_DEBUG "....... : IO APIC version: %04X\n", reg_01.bits.version);
998 if (reg_01.bits.version >= 0x10) {
999 printk(KERN_DEBUG ".... register #02: %08X\n", reg_02.raw);
1000 printk(KERN_DEBUG "....... : arbitration: %02X\n", reg_02.bits.arbitration);
1003 printk(KERN_DEBUG ".... IRQ redirection table:\n");
1005 printk(KERN_DEBUG " NR Dst Mask Trig IRR Pol"
1006 " Stat Dmod Deli Vect: \n");
1008 for (i = 0; i <= reg_01.bits.entries; i++) {
1009 struct IO_APIC_route_entry entry;
1011 entry = ioapic_read_entry(apic, i);
1013 printk(KERN_DEBUG " %02x %03X ",
1018 printk("%1d %1d %1d %1d %1d %1d %1d %02X\n",
1023 entry.delivery_status,
1025 entry.delivery_mode,
1030 printk(KERN_DEBUG "IRQ to pin mappings:\n");
1031 for (i = 0; i < NR_IRQS; i++) {
1032 struct irq_pin_list *entry = irq_2_pin + i;
1035 printk(KERN_DEBUG "IRQ%d ", i);
1037 printk("-> %d:%d", entry->apic, entry->pin);
1040 entry = irq_2_pin + entry->next;
1045 printk(KERN_INFO ".................................... done.\n");
1052 static __apicdebuginit void print_APIC_bitfield (int base)
1057 if (apic_verbosity == APIC_QUIET)
1060 printk(KERN_DEBUG "0123456789abcdef0123456789abcdef\n" KERN_DEBUG);
1061 for (i = 0; i < 8; i++) {
1062 v = apic_read(base + i*0x10);
1063 for (j = 0; j < 32; j++) {
1073 void __apicdebuginit print_local_APIC(void * dummy)
1075 unsigned int v, ver, maxlvt;
1077 if (apic_verbosity == APIC_QUIET)
1080 printk("\n" KERN_DEBUG "printing local APIC contents on CPU#%d/%d:\n",
1081 smp_processor_id(), hard_smp_processor_id());
1082 printk(KERN_INFO "... APIC ID: %08x (%01x)\n", v, GET_APIC_ID(read_apic_id()));
1083 v = apic_read(APIC_LVR);
1084 printk(KERN_INFO "... APIC VERSION: %08x\n", v);
1085 ver = GET_APIC_VERSION(v);
1086 maxlvt = lapic_get_maxlvt();
1088 v = apic_read(APIC_TASKPRI);
1089 printk(KERN_DEBUG "... APIC TASKPRI: %08x (%02x)\n", v, v & APIC_TPRI_MASK);
1091 v = apic_read(APIC_ARBPRI);
1092 printk(KERN_DEBUG "... APIC ARBPRI: %08x (%02x)\n", v,
1093 v & APIC_ARBPRI_MASK);
1094 v = apic_read(APIC_PROCPRI);
1095 printk(KERN_DEBUG "... APIC PROCPRI: %08x\n", v);
1097 v = apic_read(APIC_EOI);
1098 printk(KERN_DEBUG "... APIC EOI: %08x\n", v);
1099 v = apic_read(APIC_RRR);
1100 printk(KERN_DEBUG "... APIC RRR: %08x\n", v);
1101 v = apic_read(APIC_LDR);
1102 printk(KERN_DEBUG "... APIC LDR: %08x\n", v);
1103 v = apic_read(APIC_DFR);
1104 printk(KERN_DEBUG "... APIC DFR: %08x\n", v);
1105 v = apic_read(APIC_SPIV);
1106 printk(KERN_DEBUG "... APIC SPIV: %08x\n", v);
1108 printk(KERN_DEBUG "... APIC ISR field:\n");
1109 print_APIC_bitfield(APIC_ISR);
1110 printk(KERN_DEBUG "... APIC TMR field:\n");
1111 print_APIC_bitfield(APIC_TMR);
1112 printk(KERN_DEBUG "... APIC IRR field:\n");
1113 print_APIC_bitfield(APIC_IRR);
1115 v = apic_read(APIC_ESR);
1116 printk(KERN_DEBUG "... APIC ESR: %08x\n", v);
1118 v = apic_read(APIC_ICR);
1119 printk(KERN_DEBUG "... APIC ICR: %08x\n", v);
1120 v = apic_read(APIC_ICR2);
1121 printk(KERN_DEBUG "... APIC ICR2: %08x\n", v);
1123 v = apic_read(APIC_LVTT);
1124 printk(KERN_DEBUG "... APIC LVTT: %08x\n", v);
1126 if (maxlvt > 3) { /* PC is LVT#4. */
1127 v = apic_read(APIC_LVTPC);
1128 printk(KERN_DEBUG "... APIC LVTPC: %08x\n", v);
1130 v = apic_read(APIC_LVT0);
1131 printk(KERN_DEBUG "... APIC LVT0: %08x\n", v);
1132 v = apic_read(APIC_LVT1);
1133 printk(KERN_DEBUG "... APIC LVT1: %08x\n", v);
1135 if (maxlvt > 2) { /* ERR is LVT#3. */
1136 v = apic_read(APIC_LVTERR);
1137 printk(KERN_DEBUG "... APIC LVTERR: %08x\n", v);
1140 v = apic_read(APIC_TMICT);
1141 printk(KERN_DEBUG "... APIC TMICT: %08x\n", v);
1142 v = apic_read(APIC_TMCCT);
1143 printk(KERN_DEBUG "... APIC TMCCT: %08x\n", v);
1144 v = apic_read(APIC_TDCR);
1145 printk(KERN_DEBUG "... APIC TDCR: %08x\n", v);
1149 void print_all_local_APICs (void)
1151 on_each_cpu(print_local_APIC, NULL, 1, 1);
1154 void __apicdebuginit print_PIC(void)
1157 unsigned long flags;
1159 if (apic_verbosity == APIC_QUIET)
1162 printk(KERN_DEBUG "\nprinting PIC contents\n");
1164 spin_lock_irqsave(&i8259A_lock, flags);
1166 v = inb(0xa1) << 8 | inb(0x21);
1167 printk(KERN_DEBUG "... PIC IMR: %04x\n", v);
1169 v = inb(0xa0) << 8 | inb(0x20);
1170 printk(KERN_DEBUG "... PIC IRR: %04x\n", v);
1174 v = inb(0xa0) << 8 | inb(0x20);
1178 spin_unlock_irqrestore(&i8259A_lock, flags);
1180 printk(KERN_DEBUG "... PIC ISR: %04x\n", v);
1182 v = inb(0x4d1) << 8 | inb(0x4d0);
1183 printk(KERN_DEBUG "... PIC ELCR: %04x\n", v);
1188 void __init enable_IO_APIC(void)
1190 union IO_APIC_reg_01 reg_01;
1191 int i8259_apic, i8259_pin;
1193 unsigned long flags;
1195 for (i = 0; i < PIN_MAP_SIZE; i++) {
1196 irq_2_pin[i].pin = -1;
1197 irq_2_pin[i].next = 0;
1201 * The number of IO-APIC IRQ registers (== #pins):
1203 for (apic = 0; apic < nr_ioapics; apic++) {
1204 spin_lock_irqsave(&ioapic_lock, flags);
1205 reg_01.raw = io_apic_read(apic, 1);
1206 spin_unlock_irqrestore(&ioapic_lock, flags);
1207 nr_ioapic_registers[apic] = reg_01.bits.entries+1;
1209 for(apic = 0; apic < nr_ioapics; apic++) {
1211 /* See if any of the pins is in ExtINT mode */
1212 for (pin = 0; pin < nr_ioapic_registers[apic]; pin++) {
1213 struct IO_APIC_route_entry entry;
1214 entry = ioapic_read_entry(apic, pin);
1216 /* If the interrupt line is enabled and in ExtInt mode
1217 * I have found the pin where the i8259 is connected.
1219 if ((entry.mask == 0) && (entry.delivery_mode == dest_ExtINT)) {
1220 ioapic_i8259.apic = apic;
1221 ioapic_i8259.pin = pin;
1227 /* Look to see what if the MP table has reported the ExtINT */
1228 i8259_pin = find_isa_irq_pin(0, mp_ExtINT);
1229 i8259_apic = find_isa_irq_apic(0, mp_ExtINT);
1230 /* Trust the MP table if nothing is setup in the hardware */
1231 if ((ioapic_i8259.pin == -1) && (i8259_pin >= 0)) {
1232 printk(KERN_WARNING "ExtINT not setup in hardware but reported by MP table\n");
1233 ioapic_i8259.pin = i8259_pin;
1234 ioapic_i8259.apic = i8259_apic;
1236 /* Complain if the MP table and the hardware disagree */
1237 if (((ioapic_i8259.apic != i8259_apic) || (ioapic_i8259.pin != i8259_pin)) &&
1238 (i8259_pin >= 0) && (ioapic_i8259.pin >= 0))
1240 printk(KERN_WARNING "ExtINT in hardware and MP table differ\n");
1244 * Do not trust the IO-APIC being empty at bootup
1250 * Not an __init, needed by the reboot code
1252 void disable_IO_APIC(void)
1255 * Clear the IO-APIC before rebooting:
1260 * If the i8259 is routed through an IOAPIC
1261 * Put that IOAPIC in virtual wire mode
1262 * so legacy interrupts can be delivered.
1264 if (ioapic_i8259.pin != -1) {
1265 struct IO_APIC_route_entry entry;
1267 memset(&entry, 0, sizeof(entry));
1268 entry.mask = 0; /* Enabled */
1269 entry.trigger = 0; /* Edge */
1271 entry.polarity = 0; /* High */
1272 entry.delivery_status = 0;
1273 entry.dest_mode = 0; /* Physical */
1274 entry.delivery_mode = dest_ExtINT; /* ExtInt */
1276 entry.dest = GET_APIC_ID(read_apic_id());
1279 * Add it to the IO-APIC irq-routing table:
1281 ioapic_write_entry(ioapic_i8259.apic, ioapic_i8259.pin, entry);
1284 disconnect_bsp_APIC(ioapic_i8259.pin != -1);
1288 * There is a nasty bug in some older SMP boards, their mptable lies
1289 * about the timer IRQ. We do the following to work around the situation:
1291 * - timer IRQ defaults to IO-APIC IRQ
1292 * - if this function detects that timer IRQs are defunct, then we fall
1293 * back to ISA timer IRQs
1295 static int __init timer_irq_works(void)
1297 unsigned long t1 = jiffies;
1298 unsigned long flags;
1300 local_save_flags(flags);
1302 /* Let ten ticks pass... */
1303 mdelay((10 * 1000) / HZ);
1304 local_irq_restore(flags);
1307 * Expect a few ticks at least, to be sure some possible
1308 * glue logic does not lock up after one or two first
1309 * ticks in a non-ExtINT mode. Also the local APIC
1310 * might have cached one ExtINT interrupt. Finally, at
1311 * least one tick may be lost due to delays.
1315 if (time_after(jiffies, t1 + 4))
1321 * In the SMP+IOAPIC case it might happen that there are an unspecified
1322 * number of pending IRQ events unhandled. These cases are very rare,
1323 * so we 'resend' these IRQs via IPIs, to the same CPU. It's much
1324 * better to do it this way as thus we do not have to be aware of
1325 * 'pending' interrupts in the IRQ path, except at this point.
1328 * Edge triggered needs to resend any interrupt
1329 * that was delayed but this is now handled in the device
1334 * Starting up a edge-triggered IO-APIC interrupt is
1335 * nasty - we need to make sure that we get the edge.
1336 * If it is already asserted for some reason, we need
1337 * return 1 to indicate that is was pending.
1339 * This is not complete - we should be able to fake
1340 * an edge even if it isn't on the 8259A...
1343 static unsigned int startup_ioapic_irq(unsigned int irq)
1345 int was_pending = 0;
1346 unsigned long flags;
1348 spin_lock_irqsave(&ioapic_lock, flags);
1350 disable_8259A_irq(irq);
1351 if (i8259A_irq_pending(irq))
1354 __unmask_IO_APIC_irq(irq);
1355 spin_unlock_irqrestore(&ioapic_lock, flags);
1360 static int ioapic_retrigger_irq(unsigned int irq)
1362 struct irq_cfg *cfg = &irq_cfg[irq];
1364 unsigned long flags;
1366 spin_lock_irqsave(&vector_lock, flags);
1367 mask = cpumask_of_cpu(first_cpu(cfg->domain));
1368 send_IPI_mask(mask, cfg->vector);
1369 spin_unlock_irqrestore(&vector_lock, flags);
1375 * Level and edge triggered IO-APIC interrupts need different handling,
1376 * so we use two separate IRQ descriptors. Edge triggered IRQs can be
1377 * handled with the level-triggered descriptor, but that one has slightly
1378 * more overhead. Level-triggered interrupts cannot be handled with the
1379 * edge-triggered handler, without risking IRQ storms and other ugly
1384 asmlinkage void smp_irq_move_cleanup_interrupt(void)
1386 unsigned vector, me;
1391 me = smp_processor_id();
1392 for (vector = FIRST_EXTERNAL_VECTOR; vector < NR_VECTORS; vector++) {
1394 struct irq_desc *desc;
1395 struct irq_cfg *cfg;
1396 irq = __get_cpu_var(vector_irq)[vector];
1400 desc = irq_desc + irq;
1401 cfg = irq_cfg + irq;
1402 spin_lock(&desc->lock);
1403 if (!cfg->move_cleanup_count)
1406 if ((vector == cfg->vector) && cpu_isset(me, cfg->domain))
1409 __get_cpu_var(vector_irq)[vector] = -1;
1410 cfg->move_cleanup_count--;
1412 spin_unlock(&desc->lock);
1418 static void irq_complete_move(unsigned int irq)
1420 struct irq_cfg *cfg = irq_cfg + irq;
1421 unsigned vector, me;
1423 if (likely(!cfg->move_in_progress))
1426 vector = ~get_irq_regs()->orig_ax;
1427 me = smp_processor_id();
1428 if ((vector == cfg->vector) && cpu_isset(me, cfg->domain)) {
1429 cpumask_t cleanup_mask;
1431 cpus_and(cleanup_mask, cfg->old_domain, cpu_online_map);
1432 cfg->move_cleanup_count = cpus_weight(cleanup_mask);
1433 send_IPI_mask(cleanup_mask, IRQ_MOVE_CLEANUP_VECTOR);
1434 cfg->move_in_progress = 0;
1438 static inline void irq_complete_move(unsigned int irq) {}
1441 static void ack_apic_edge(unsigned int irq)
1443 irq_complete_move(irq);
1444 move_native_irq(irq);
1448 static void ack_apic_level(unsigned int irq)
1450 int do_unmask_irq = 0;
1452 irq_complete_move(irq);
1453 #ifdef CONFIG_GENERIC_PENDING_IRQ
1454 /* If we are moving the irq we need to mask it */
1455 if (unlikely(irq_desc[irq].status & IRQ_MOVE_PENDING)) {
1457 mask_IO_APIC_irq(irq);
1462 * We must acknowledge the irq before we move it or the acknowledge will
1463 * not propagate properly.
1467 /* Now we can move and renable the irq */
1468 if (unlikely(do_unmask_irq)) {
1469 /* Only migrate the irq if the ack has been received.
1471 * On rare occasions the broadcast level triggered ack gets
1472 * delayed going to ioapics, and if we reprogram the
1473 * vector while Remote IRR is still set the irq will never
1476 * To prevent this scenario we read the Remote IRR bit
1477 * of the ioapic. This has two effects.
1478 * - On any sane system the read of the ioapic will
1479 * flush writes (and acks) going to the ioapic from
1481 * - We get to see if the ACK has actually been delivered.
1483 * Based on failed experiments of reprogramming the
1484 * ioapic entry from outside of irq context starting
1485 * with masking the ioapic entry and then polling until
1486 * Remote IRR was clear before reprogramming the
1487 * ioapic I don't trust the Remote IRR bit to be
1488 * completey accurate.
1490 * However there appears to be no other way to plug
1491 * this race, so if the Remote IRR bit is not
1492 * accurate and is causing problems then it is a hardware bug
1493 * and you can go talk to the chipset vendor about it.
1495 if (!io_apic_level_ack_pending(irq))
1496 move_masked_irq(irq);
1497 unmask_IO_APIC_irq(irq);
1501 static struct irq_chip ioapic_chip __read_mostly = {
1503 .startup = startup_ioapic_irq,
1504 .mask = mask_IO_APIC_irq,
1505 .unmask = unmask_IO_APIC_irq,
1506 .ack = ack_apic_edge,
1507 .eoi = ack_apic_level,
1509 .set_affinity = set_ioapic_affinity_irq,
1511 .retrigger = ioapic_retrigger_irq,
1514 static inline void init_IO_APIC_traps(void)
1519 * NOTE! The local APIC isn't very good at handling
1520 * multiple interrupts at the same interrupt level.
1521 * As the interrupt level is determined by taking the
1522 * vector number and shifting that right by 4, we
1523 * want to spread these out a bit so that they don't
1524 * all fall in the same interrupt level.
1526 * Also, we've got to be careful not to trash gate
1527 * 0x80, because int 0x80 is hm, kind of importantish. ;)
1529 for (irq = 0; irq < NR_IRQS ; irq++) {
1530 if (IO_APIC_IRQ(irq) && !irq_cfg[irq].vector) {
1532 * Hmm.. We don't have an entry for this,
1533 * so default to an old-fashioned 8259
1534 * interrupt if we can..
1537 make_8259A_irq(irq);
1539 /* Strange. Oh, well.. */
1540 irq_desc[irq].chip = &no_irq_chip;
1545 static void enable_lapic_irq (unsigned int irq)
1549 v = apic_read(APIC_LVT0);
1550 apic_write(APIC_LVT0, v & ~APIC_LVT_MASKED);
1553 static void disable_lapic_irq (unsigned int irq)
1557 v = apic_read(APIC_LVT0);
1558 apic_write(APIC_LVT0, v | APIC_LVT_MASKED);
1561 static void ack_lapic_irq (unsigned int irq)
1566 static void end_lapic_irq (unsigned int i) { /* nothing */ }
1568 static struct hw_interrupt_type lapic_irq_type __read_mostly = {
1569 .name = "local-APIC",
1570 .typename = "local-APIC-edge",
1571 .startup = NULL, /* startup_irq() not used for IRQ0 */
1572 .shutdown = NULL, /* shutdown_irq() not used for IRQ0 */
1573 .enable = enable_lapic_irq,
1574 .disable = disable_lapic_irq,
1575 .ack = ack_lapic_irq,
1576 .end = end_lapic_irq,
1579 static void __init setup_nmi(void)
1582 * Dirty trick to enable the NMI watchdog ...
1583 * We put the 8259A master into AEOI mode and
1584 * unmask on all local APICs LVT0 as NMI.
1586 * The idea to use the 8259A in AEOI mode ('8259A Virtual Wire')
1587 * is from Maciej W. Rozycki - so we do not have to EOI from
1588 * the NMI handler or the timer interrupt.
1590 printk(KERN_INFO "activating NMI Watchdog ...");
1592 enable_NMI_through_LVT0();
1598 * This looks a bit hackish but it's about the only one way of sending
1599 * a few INTA cycles to 8259As and any associated glue logic. ICR does
1600 * not support the ExtINT mode, unfortunately. We need to send these
1601 * cycles as some i82489DX-based boards have glue logic that keeps the
1602 * 8259A interrupt line asserted until INTA. --macro
1604 static inline void __init unlock_ExtINT_logic(void)
1607 struct IO_APIC_route_entry entry0, entry1;
1608 unsigned char save_control, save_freq_select;
1610 pin = find_isa_irq_pin(8, mp_INT);
1611 apic = find_isa_irq_apic(8, mp_INT);
1615 entry0 = ioapic_read_entry(apic, pin);
1617 clear_IO_APIC_pin(apic, pin);
1619 memset(&entry1, 0, sizeof(entry1));
1621 entry1.dest_mode = 0; /* physical delivery */
1622 entry1.mask = 0; /* unmask IRQ now */
1623 entry1.dest = hard_smp_processor_id();
1624 entry1.delivery_mode = dest_ExtINT;
1625 entry1.polarity = entry0.polarity;
1629 ioapic_write_entry(apic, pin, entry1);
1631 save_control = CMOS_READ(RTC_CONTROL);
1632 save_freq_select = CMOS_READ(RTC_FREQ_SELECT);
1633 CMOS_WRITE((save_freq_select & ~RTC_RATE_SELECT) | 0x6,
1635 CMOS_WRITE(save_control | RTC_PIE, RTC_CONTROL);
1640 if ((CMOS_READ(RTC_INTR_FLAGS) & RTC_PF) == RTC_PF)
1644 CMOS_WRITE(save_control, RTC_CONTROL);
1645 CMOS_WRITE(save_freq_select, RTC_FREQ_SELECT);
1646 clear_IO_APIC_pin(apic, pin);
1648 ioapic_write_entry(apic, pin, entry0);
1652 * This code may look a bit paranoid, but it's supposed to cooperate with
1653 * a wide range of boards and BIOS bugs. Fortunately only the timer IRQ
1654 * is so screwy. Thanks to Brian Perkins for testing/hacking this beast
1655 * fanatically on his truly buggy board.
1657 * FIXME: really need to revamp this for modern platforms only.
1659 static inline void __init check_timer(void)
1661 struct irq_cfg *cfg = irq_cfg + 0;
1662 int apic1, pin1, apic2, pin2;
1663 unsigned long flags;
1665 local_irq_save(flags);
1668 * get/set the timer IRQ vector:
1670 disable_8259A_irq(0);
1671 assign_irq_vector(0, TARGET_CPUS);
1674 * Subtle, code in do_timer_interrupt() expects an AEOI
1675 * mode for the 8259A whenever interrupts are routed
1676 * through I/O APICs. Also IRQ0 has to be enabled in
1677 * the 8259A which implies the virtual wire has to be
1678 * disabled in the local APIC.
1680 apic_write(APIC_LVT0, APIC_LVT_MASKED | APIC_DM_EXTINT);
1682 if (timer_over_8254 > 0)
1683 enable_8259A_irq(0);
1685 pin1 = find_isa_irq_pin(0, mp_INT);
1686 apic1 = find_isa_irq_apic(0, mp_INT);
1687 pin2 = ioapic_i8259.pin;
1688 apic2 = ioapic_i8259.apic;
1690 apic_printk(APIC_VERBOSE,KERN_INFO "..TIMER: vector=0x%02X apic1=%d pin1=%d apic2=%d pin2=%d\n",
1691 cfg->vector, apic1, pin1, apic2, pin2);
1695 * Ok, does IRQ0 through the IOAPIC work?
1697 unmask_IO_APIC_irq(0);
1698 if (!no_timer_check && timer_irq_works()) {
1699 nmi_watchdog_default();
1700 if (nmi_watchdog == NMI_IO_APIC) {
1701 disable_8259A_irq(0);
1703 enable_8259A_irq(0);
1705 if (disable_timer_pin_1 > 0)
1706 clear_IO_APIC_pin(0, pin1);
1709 clear_IO_APIC_pin(apic1, pin1);
1710 apic_printk(APIC_QUIET,KERN_ERR "..MP-BIOS bug: 8254 timer not "
1711 "connected to IO-APIC\n");
1714 apic_printk(APIC_VERBOSE,KERN_INFO "...trying to set up timer (IRQ0) "
1715 "through the 8259A ... ");
1717 apic_printk(APIC_VERBOSE,"\n..... (found apic %d pin %d) ...",
1720 * legacy devices should be connected to IO APIC #0
1722 setup_ExtINT_IRQ0_pin(apic2, pin2, cfg->vector);
1723 if (timer_irq_works()) {
1724 apic_printk(APIC_VERBOSE," works.\n");
1725 nmi_watchdog_default();
1726 if (nmi_watchdog == NMI_IO_APIC) {
1732 * Cleanup, just in case ...
1734 clear_IO_APIC_pin(apic2, pin2);
1736 apic_printk(APIC_VERBOSE," failed.\n");
1738 if (nmi_watchdog == NMI_IO_APIC) {
1739 printk(KERN_WARNING "timer doesn't work through the IO-APIC - disabling NMI Watchdog!\n");
1743 apic_printk(APIC_VERBOSE, KERN_INFO "...trying to set up timer as Virtual Wire IRQ...");
1745 disable_8259A_irq(0);
1746 irq_desc[0].chip = &lapic_irq_type;
1747 apic_write(APIC_LVT0, APIC_DM_FIXED | cfg->vector); /* Fixed mode */
1748 enable_8259A_irq(0);
1750 if (timer_irq_works()) {
1751 apic_printk(APIC_VERBOSE," works.\n");
1754 apic_write(APIC_LVT0, APIC_LVT_MASKED | APIC_DM_FIXED | cfg->vector);
1755 apic_printk(APIC_VERBOSE," failed.\n");
1757 apic_printk(APIC_VERBOSE, KERN_INFO "...trying to set up timer as ExtINT IRQ...");
1761 apic_write(APIC_LVT0, APIC_DM_EXTINT);
1763 unlock_ExtINT_logic();
1765 if (timer_irq_works()) {
1766 apic_printk(APIC_VERBOSE," works.\n");
1769 apic_printk(APIC_VERBOSE," failed :(.\n");
1770 panic("IO-APIC + timer doesn't work! Try using the 'noapic' kernel parameter\n");
1772 local_irq_restore(flags);
1775 static int __init notimercheck(char *s)
1780 __setup("no_timer_check", notimercheck);
1784 * IRQs that are handled by the PIC in the MPS IOAPIC case.
1785 * - IRQ2 is the cascade IRQ, and cannot be a io-apic IRQ.
1786 * Linux doesn't really care, as it's not actually used
1787 * for any interrupt handling anyway.
1789 #define PIC_IRQS (1<<2)
1791 void __init setup_IO_APIC(void)
1795 * calling enable_IO_APIC() is moved to setup_local_APIC for BP
1799 io_apic_irqs = ~0; /* all IRQs go through IOAPIC */
1801 io_apic_irqs = ~PIC_IRQS;
1803 apic_printk(APIC_VERBOSE, "ENABLING IO-APIC IRQs\n");
1806 setup_IO_APIC_irqs();
1807 init_IO_APIC_traps();
1813 struct sysfs_ioapic_data {
1814 struct sys_device dev;
1815 struct IO_APIC_route_entry entry[0];
1817 static struct sysfs_ioapic_data * mp_ioapic_data[MAX_IO_APICS];
1819 static int ioapic_suspend(struct sys_device *dev, pm_message_t state)
1821 struct IO_APIC_route_entry *entry;
1822 struct sysfs_ioapic_data *data;
1825 data = container_of(dev, struct sysfs_ioapic_data, dev);
1826 entry = data->entry;
1827 for (i = 0; i < nr_ioapic_registers[dev->id]; i ++, entry ++ )
1828 *entry = ioapic_read_entry(dev->id, i);
1833 static int ioapic_resume(struct sys_device *dev)
1835 struct IO_APIC_route_entry *entry;
1836 struct sysfs_ioapic_data *data;
1837 unsigned long flags;
1838 union IO_APIC_reg_00 reg_00;
1841 data = container_of(dev, struct sysfs_ioapic_data, dev);
1842 entry = data->entry;
1844 spin_lock_irqsave(&ioapic_lock, flags);
1845 reg_00.raw = io_apic_read(dev->id, 0);
1846 if (reg_00.bits.ID != mp_ioapics[dev->id].mp_apicid) {
1847 reg_00.bits.ID = mp_ioapics[dev->id].mp_apicid;
1848 io_apic_write(dev->id, 0, reg_00.raw);
1850 spin_unlock_irqrestore(&ioapic_lock, flags);
1851 for (i = 0; i < nr_ioapic_registers[dev->id]; i++)
1852 ioapic_write_entry(dev->id, i, entry[i]);
1857 static struct sysdev_class ioapic_sysdev_class = {
1859 .suspend = ioapic_suspend,
1860 .resume = ioapic_resume,
1863 static int __init ioapic_init_sysfs(void)
1865 struct sys_device * dev;
1868 error = sysdev_class_register(&ioapic_sysdev_class);
1872 for (i = 0; i < nr_ioapics; i++ ) {
1873 size = sizeof(struct sys_device) + nr_ioapic_registers[i]
1874 * sizeof(struct IO_APIC_route_entry);
1875 mp_ioapic_data[i] = kzalloc(size, GFP_KERNEL);
1876 if (!mp_ioapic_data[i]) {
1877 printk(KERN_ERR "Can't suspend/resume IOAPIC %d\n", i);
1880 dev = &mp_ioapic_data[i]->dev;
1882 dev->cls = &ioapic_sysdev_class;
1883 error = sysdev_register(dev);
1885 kfree(mp_ioapic_data[i]);
1886 mp_ioapic_data[i] = NULL;
1887 printk(KERN_ERR "Can't suspend/resume IOAPIC %d\n", i);
1895 device_initcall(ioapic_init_sysfs);
1898 * Dynamic irq allocate and deallocation
1900 int create_irq(void)
1902 /* Allocate an unused irq */
1905 unsigned long flags;
1908 spin_lock_irqsave(&vector_lock, flags);
1909 for (new = (NR_IRQS - 1); new >= 0; new--) {
1910 if (platform_legacy_irq(new))
1912 if (irq_cfg[new].vector != 0)
1914 if (__assign_irq_vector(new, TARGET_CPUS) == 0)
1918 spin_unlock_irqrestore(&vector_lock, flags);
1921 dynamic_irq_init(irq);
1926 void destroy_irq(unsigned int irq)
1928 unsigned long flags;
1930 dynamic_irq_cleanup(irq);
1932 spin_lock_irqsave(&vector_lock, flags);
1933 __clear_irq_vector(irq);
1934 spin_unlock_irqrestore(&vector_lock, flags);
1938 * MSI message composition
1940 #ifdef CONFIG_PCI_MSI
1941 static int msi_compose_msg(struct pci_dev *pdev, unsigned int irq, struct msi_msg *msg)
1943 struct irq_cfg *cfg = irq_cfg + irq;
1949 err = assign_irq_vector(irq, tmp);
1951 cpus_and(tmp, cfg->domain, tmp);
1952 dest = cpu_mask_to_apicid(tmp);
1954 msg->address_hi = MSI_ADDR_BASE_HI;
1957 ((INT_DEST_MODE == 0) ?
1958 MSI_ADDR_DEST_MODE_PHYSICAL:
1959 MSI_ADDR_DEST_MODE_LOGICAL) |
1960 ((INT_DELIVERY_MODE != dest_LowestPrio) ?
1961 MSI_ADDR_REDIRECTION_CPU:
1962 MSI_ADDR_REDIRECTION_LOWPRI) |
1963 MSI_ADDR_DEST_ID(dest);
1966 MSI_DATA_TRIGGER_EDGE |
1967 MSI_DATA_LEVEL_ASSERT |
1968 ((INT_DELIVERY_MODE != dest_LowestPrio) ?
1969 MSI_DATA_DELIVERY_FIXED:
1970 MSI_DATA_DELIVERY_LOWPRI) |
1971 MSI_DATA_VECTOR(cfg->vector);
1977 static void set_msi_irq_affinity(unsigned int irq, cpumask_t mask)
1979 struct irq_cfg *cfg = irq_cfg + irq;
1984 cpus_and(tmp, mask, cpu_online_map);
1985 if (cpus_empty(tmp))
1988 if (assign_irq_vector(irq, mask))
1991 cpus_and(tmp, cfg->domain, mask);
1992 dest = cpu_mask_to_apicid(tmp);
1994 read_msi_msg(irq, &msg);
1996 msg.data &= ~MSI_DATA_VECTOR_MASK;
1997 msg.data |= MSI_DATA_VECTOR(cfg->vector);
1998 msg.address_lo &= ~MSI_ADDR_DEST_ID_MASK;
1999 msg.address_lo |= MSI_ADDR_DEST_ID(dest);
2001 write_msi_msg(irq, &msg);
2002 irq_desc[irq].affinity = mask;
2004 #endif /* CONFIG_SMP */
2007 * IRQ Chip for MSI PCI/PCI-X/PCI-Express Devices,
2008 * which implement the MSI or MSI-X Capability Structure.
2010 static struct irq_chip msi_chip = {
2012 .unmask = unmask_msi_irq,
2013 .mask = mask_msi_irq,
2014 .ack = ack_apic_edge,
2016 .set_affinity = set_msi_irq_affinity,
2018 .retrigger = ioapic_retrigger_irq,
2021 int arch_setup_msi_irq(struct pci_dev *dev, struct msi_desc *desc)
2029 ret = msi_compose_msg(dev, irq, &msg);
2035 set_irq_msi(irq, desc);
2036 write_msi_msg(irq, &msg);
2038 set_irq_chip_and_handler_name(irq, &msi_chip, handle_edge_irq, "edge");
2043 void arch_teardown_msi_irq(unsigned int irq)
2050 static void dmar_msi_set_affinity(unsigned int irq, cpumask_t mask)
2052 struct irq_cfg *cfg = irq_cfg + irq;
2057 cpus_and(tmp, mask, cpu_online_map);
2058 if (cpus_empty(tmp))
2061 if (assign_irq_vector(irq, mask))
2064 cpus_and(tmp, cfg->domain, mask);
2065 dest = cpu_mask_to_apicid(tmp);
2067 dmar_msi_read(irq, &msg);
2069 msg.data &= ~MSI_DATA_VECTOR_MASK;
2070 msg.data |= MSI_DATA_VECTOR(cfg->vector);
2071 msg.address_lo &= ~MSI_ADDR_DEST_ID_MASK;
2072 msg.address_lo |= MSI_ADDR_DEST_ID(dest);
2074 dmar_msi_write(irq, &msg);
2075 irq_desc[irq].affinity = mask;
2077 #endif /* CONFIG_SMP */
2079 struct irq_chip dmar_msi_type = {
2081 .unmask = dmar_msi_unmask,
2082 .mask = dmar_msi_mask,
2083 .ack = ack_apic_edge,
2085 .set_affinity = dmar_msi_set_affinity,
2087 .retrigger = ioapic_retrigger_irq,
2090 int arch_setup_dmar_msi(unsigned int irq)
2095 ret = msi_compose_msg(NULL, irq, &msg);
2098 dmar_msi_write(irq, &msg);
2099 set_irq_chip_and_handler_name(irq, &dmar_msi_type, handle_edge_irq,
2105 #endif /* CONFIG_PCI_MSI */
2107 * Hypertransport interrupt support
2109 #ifdef CONFIG_HT_IRQ
2113 static void target_ht_irq(unsigned int irq, unsigned int dest, u8 vector)
2115 struct ht_irq_msg msg;
2116 fetch_ht_irq_msg(irq, &msg);
2118 msg.address_lo &= ~(HT_IRQ_LOW_VECTOR_MASK | HT_IRQ_LOW_DEST_ID_MASK);
2119 msg.address_hi &= ~(HT_IRQ_HIGH_DEST_ID_MASK);
2121 msg.address_lo |= HT_IRQ_LOW_VECTOR(vector) | HT_IRQ_LOW_DEST_ID(dest);
2122 msg.address_hi |= HT_IRQ_HIGH_DEST_ID(dest);
2124 write_ht_irq_msg(irq, &msg);
2127 static void set_ht_irq_affinity(unsigned int irq, cpumask_t mask)
2129 struct irq_cfg *cfg = irq_cfg + irq;
2133 cpus_and(tmp, mask, cpu_online_map);
2134 if (cpus_empty(tmp))
2137 if (assign_irq_vector(irq, mask))
2140 cpus_and(tmp, cfg->domain, mask);
2141 dest = cpu_mask_to_apicid(tmp);
2143 target_ht_irq(irq, dest, cfg->vector);
2144 irq_desc[irq].affinity = mask;
2148 static struct irq_chip ht_irq_chip = {
2150 .mask = mask_ht_irq,
2151 .unmask = unmask_ht_irq,
2152 .ack = ack_apic_edge,
2154 .set_affinity = set_ht_irq_affinity,
2156 .retrigger = ioapic_retrigger_irq,
2159 int arch_setup_ht_irq(unsigned int irq, struct pci_dev *dev)
2161 struct irq_cfg *cfg = irq_cfg + irq;
2166 err = assign_irq_vector(irq, tmp);
2168 struct ht_irq_msg msg;
2171 cpus_and(tmp, cfg->domain, tmp);
2172 dest = cpu_mask_to_apicid(tmp);
2174 msg.address_hi = HT_IRQ_HIGH_DEST_ID(dest);
2178 HT_IRQ_LOW_DEST_ID(dest) |
2179 HT_IRQ_LOW_VECTOR(cfg->vector) |
2180 ((INT_DEST_MODE == 0) ?
2181 HT_IRQ_LOW_DM_PHYSICAL :
2182 HT_IRQ_LOW_DM_LOGICAL) |
2183 HT_IRQ_LOW_RQEOI_EDGE |
2184 ((INT_DELIVERY_MODE != dest_LowestPrio) ?
2185 HT_IRQ_LOW_MT_FIXED :
2186 HT_IRQ_LOW_MT_ARBITRATED) |
2187 HT_IRQ_LOW_IRQ_MASKED;
2189 write_ht_irq_msg(irq, &msg);
2191 set_irq_chip_and_handler_name(irq, &ht_irq_chip,
2192 handle_edge_irq, "edge");
2196 #endif /* CONFIG_HT_IRQ */
2198 /* --------------------------------------------------------------------------
2199 ACPI-based IOAPIC Configuration
2200 -------------------------------------------------------------------------- */
2204 #define IO_APIC_MAX_ID 0xFE
2206 int __init io_apic_get_redir_entries (int ioapic)
2208 union IO_APIC_reg_01 reg_01;
2209 unsigned long flags;
2211 spin_lock_irqsave(&ioapic_lock, flags);
2212 reg_01.raw = io_apic_read(ioapic, 1);
2213 spin_unlock_irqrestore(&ioapic_lock, flags);
2215 return reg_01.bits.entries;
2219 int io_apic_set_pci_routing (int ioapic, int pin, int irq, int triggering, int polarity)
2221 if (!IO_APIC_IRQ(irq)) {
2222 apic_printk(APIC_QUIET,KERN_ERR "IOAPIC[%d]: Invalid reference to IRQ 0\n",
2228 * IRQs < 16 are already in the irq_2_pin[] map
2231 add_pin_to_irq(irq, ioapic, pin);
2233 setup_IO_APIC_irq(ioapic, pin, irq, triggering, polarity);
2239 int acpi_get_override_irq(int bus_irq, int *trigger, int *polarity)
2243 if (skip_ioapic_setup)
2246 for (i = 0; i < mp_irq_entries; i++)
2247 if (mp_irqs[i].mp_irqtype == mp_INT &&
2248 mp_irqs[i].mp_srcbusirq == bus_irq)
2250 if (i >= mp_irq_entries)
2253 *trigger = irq_trigger(i);
2254 *polarity = irq_polarity(i);
2258 #endif /* CONFIG_ACPI */
2261 * This function currently is only a helper for the i386 smp boot process where
2262 * we need to reprogram the ioredtbls to cater for the cpus which have come online
2263 * so mask in all cases should simply be TARGET_CPUS
2266 void __init setup_ioapic_dest(void)
2268 int pin, ioapic, irq, irq_entry;
2270 if (skip_ioapic_setup == 1)
2273 for (ioapic = 0; ioapic < nr_ioapics; ioapic++) {
2274 for (pin = 0; pin < nr_ioapic_registers[ioapic]; pin++) {
2275 irq_entry = find_irq_entry(ioapic, pin, mp_INT);
2276 if (irq_entry == -1)
2278 irq = pin_2_irq(irq_entry, ioapic, pin);
2280 /* setup_IO_APIC_irqs could fail to get vector for some device
2281 * when you have too many devices, because at that time only boot
2284 if (!irq_cfg[irq].vector)
2285 setup_IO_APIC_irq(ioapic, pin, irq,
2286 irq_trigger(irq_entry),
2287 irq_polarity(irq_entry));
2289 set_ioapic_affinity_irq(irq, TARGET_CPUS);
2296 #define IOAPIC_RESOURCE_NAME_SIZE 11
2298 static struct resource *ioapic_resources;
2300 static struct resource * __init ioapic_setup_resources(void)
2303 struct resource *res;
2307 if (nr_ioapics <= 0)
2310 n = IOAPIC_RESOURCE_NAME_SIZE + sizeof(struct resource);
2313 mem = alloc_bootmem(n);
2317 mem += sizeof(struct resource) * nr_ioapics;
2319 for (i = 0; i < nr_ioapics; i++) {
2321 res[i].flags = IORESOURCE_MEM | IORESOURCE_BUSY;
2322 sprintf(mem, "IOAPIC %u", i);
2323 mem += IOAPIC_RESOURCE_NAME_SIZE;
2327 ioapic_resources = res;
2332 void __init ioapic_init_mappings(void)
2334 unsigned long ioapic_phys, idx = FIX_IO_APIC_BASE_0;
2335 struct resource *ioapic_res;
2338 ioapic_res = ioapic_setup_resources();
2339 for (i = 0; i < nr_ioapics; i++) {
2340 if (smp_found_config) {
2341 ioapic_phys = mp_ioapics[i].mp_apicaddr;
2343 ioapic_phys = (unsigned long)
2344 alloc_bootmem_pages(PAGE_SIZE);
2345 ioapic_phys = __pa(ioapic_phys);
2347 set_fixmap_nocache(idx, ioapic_phys);
2348 apic_printk(APIC_VERBOSE,
2349 "mapped IOAPIC to %016lx (%016lx)\n",
2350 __fix_to_virt(idx), ioapic_phys);
2353 if (ioapic_res != NULL) {
2354 ioapic_res->start = ioapic_phys;
2355 ioapic_res->end = ioapic_phys + (4 * 1024) - 1;
2361 static int __init ioapic_insert_resources(void)
2364 struct resource *r = ioapic_resources;
2368 "IO APIC resources could be not be allocated.\n");
2372 for (i = 0; i < nr_ioapics; i++) {
2373 insert_resource(&iomem_resource, r);
2380 /* Insert the IO APIC resources after PCI initialization has occured to handle
2381 * IO APICS that are mapped in on a BAR in PCI space. */
2382 late_initcall(ioapic_insert_resources);