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/smp_lock.h>
29 #include <linux/pci.h>
30 #include <linux/mc146818rtc.h>
31 #include <linux/acpi.h>
32 #include <linux/sysdev.h>
33 #include <linux/msi.h>
34 #include <linux/htirq.h>
36 #include <acpi/acpi_bus.h>
42 #include <asm/proto.h>
43 #include <asm/mach_apic.h>
47 #include <asm/msidef.h>
48 #include <asm/hypertransport.h>
50 static int assign_irq_vector(int irq, cpumask_t mask, cpumask_t *result);
52 #define __apicdebuginit __init
54 int sis_apic_bug; /* not actually supported, dummy for compile */
56 static int no_timer_check;
58 static int disable_timer_pin_1 __initdata;
60 int timer_over_8254 __initdata = 1;
62 /* Where if anywhere is the i8259 connect in external int mode */
63 static struct { int pin, apic; } ioapic_i8259 = { -1, -1 };
65 static DEFINE_SPINLOCK(ioapic_lock);
66 DEFINE_SPINLOCK(vector_lock);
69 * # of IRQ routing registers
71 int nr_ioapic_registers[MAX_IO_APICS];
74 * Rough estimation of how many shared IRQs there are, can
77 #define MAX_PLUS_SHARED_IRQS NR_IRQ_VECTORS
78 #define PIN_MAP_SIZE (MAX_PLUS_SHARED_IRQS + NR_IRQS)
81 * This is performance-critical, we want to do it O(1)
83 * the indexing order of this array favors 1:1 mappings
84 * between pins and IRQs.
87 static struct irq_pin_list {
88 short apic, pin, next;
89 } irq_2_pin[PIN_MAP_SIZE];
93 unsigned int unused[3];
97 static __attribute_const__ struct io_apic __iomem *io_apic_base(int idx)
99 return (void __iomem *) __fix_to_virt(FIX_IO_APIC_BASE_0 + idx)
100 + (mp_ioapics[idx].mpc_apicaddr & ~PAGE_MASK);
103 static inline unsigned int io_apic_read(unsigned int apic, unsigned int reg)
105 struct io_apic __iomem *io_apic = io_apic_base(apic);
106 writel(reg, &io_apic->index);
107 return readl(&io_apic->data);
110 static inline void io_apic_write(unsigned int apic, unsigned int reg, unsigned int value)
112 struct io_apic __iomem *io_apic = io_apic_base(apic);
113 writel(reg, &io_apic->index);
114 writel(value, &io_apic->data);
118 * Re-write a value: to be used for read-modify-write
119 * cycles where the read already set up the index register.
121 static inline void io_apic_modify(unsigned int apic, unsigned int value)
123 struct io_apic __iomem *io_apic = io_apic_base(apic);
124 writel(value, &io_apic->data);
128 * Synchronize the IO-APIC and the CPU by doing
129 * a dummy read from the IO-APIC
131 static inline void io_apic_sync(unsigned int apic)
133 struct io_apic __iomem *io_apic = io_apic_base(apic);
134 readl(&io_apic->data);
137 #define __DO_ACTION(R, ACTION, FINAL) \
141 struct irq_pin_list *entry = irq_2_pin + irq; \
143 BUG_ON(irq >= NR_IRQS); \
149 reg = io_apic_read(entry->apic, 0x10 + R + pin*2); \
151 io_apic_modify(entry->apic, reg); \
154 entry = irq_2_pin + entry->next; \
160 struct { u32 w1, w2; };
161 struct IO_APIC_route_entry entry;
164 static struct IO_APIC_route_entry ioapic_read_entry(int apic, int pin)
166 union entry_union eu;
168 spin_lock_irqsave(&ioapic_lock, flags);
169 eu.w1 = io_apic_read(apic, 0x10 + 2 * pin);
170 eu.w2 = io_apic_read(apic, 0x11 + 2 * pin);
171 spin_unlock_irqrestore(&ioapic_lock, flags);
176 * When we write a new IO APIC routing entry, we need to write the high
177 * word first! If the mask bit in the low word is clear, we will enable
178 * the interrupt, and we need to make sure the entry is fully populated
179 * before that happens.
182 __ioapic_write_entry(int apic, int pin, struct IO_APIC_route_entry e)
184 union entry_union eu;
186 io_apic_write(apic, 0x11 + 2*pin, eu.w2);
187 io_apic_write(apic, 0x10 + 2*pin, eu.w1);
190 static void ioapic_write_entry(int apic, int pin, struct IO_APIC_route_entry e)
193 spin_lock_irqsave(&ioapic_lock, flags);
194 __ioapic_write_entry(apic, pin, e);
195 spin_unlock_irqrestore(&ioapic_lock, flags);
199 * When we mask an IO APIC routing entry, we need to write the low
200 * word first, in order to set the mask bit before we change the
203 static void ioapic_mask_entry(int apic, int pin)
206 union entry_union eu = { .entry.mask = 1 };
208 spin_lock_irqsave(&ioapic_lock, flags);
209 io_apic_write(apic, 0x10 + 2*pin, eu.w1);
210 io_apic_write(apic, 0x11 + 2*pin, eu.w2);
211 spin_unlock_irqrestore(&ioapic_lock, flags);
215 static void __target_IO_APIC_irq(unsigned int irq, unsigned int dest, u8 vector)
218 struct irq_pin_list *entry = irq_2_pin + irq;
220 BUG_ON(irq >= NR_IRQS);
227 io_apic_write(apic, 0x11 + pin*2, dest);
228 reg = io_apic_read(apic, 0x10 + pin*2);
231 io_apic_modify(apic, reg);
234 entry = irq_2_pin + entry->next;
238 static void set_ioapic_affinity_irq(unsigned int irq, cpumask_t mask)
245 cpus_and(tmp, mask, cpu_online_map);
249 cpus_and(mask, tmp, CPU_MASK_ALL);
251 vector = assign_irq_vector(irq, mask, &tmp);
255 dest = cpu_mask_to_apicid(tmp);
258 * Only the high 8 bits are valid.
260 dest = SET_APIC_LOGICAL_ID(dest);
262 spin_lock_irqsave(&ioapic_lock, flags);
263 __target_IO_APIC_irq(irq, dest, vector);
264 irq_desc[irq].affinity = mask;
265 spin_unlock_irqrestore(&ioapic_lock, flags);
270 * The common case is 1:1 IRQ<->pin mappings. Sometimes there are
271 * shared ISA-space IRQs, so we have to support them. We are super
272 * fast in the common case, and fast for shared ISA-space IRQs.
274 static void add_pin_to_irq(unsigned int irq, int apic, int pin)
276 static int first_free_entry = NR_IRQS;
277 struct irq_pin_list *entry = irq_2_pin + irq;
279 BUG_ON(irq >= NR_IRQS);
281 entry = irq_2_pin + entry->next;
283 if (entry->pin != -1) {
284 entry->next = first_free_entry;
285 entry = irq_2_pin + entry->next;
286 if (++first_free_entry >= PIN_MAP_SIZE)
287 panic("io_apic.c: ran out of irq_2_pin entries!");
294 #define DO_ACTION(name,R,ACTION, FINAL) \
296 static void name##_IO_APIC_irq (unsigned int irq) \
297 __DO_ACTION(R, ACTION, FINAL)
299 DO_ACTION( __mask, 0, |= 0x00010000, io_apic_sync(entry->apic) )
301 DO_ACTION( __unmask, 0, &= 0xfffeffff, )
304 static void mask_IO_APIC_irq (unsigned int irq)
308 spin_lock_irqsave(&ioapic_lock, flags);
309 __mask_IO_APIC_irq(irq);
310 spin_unlock_irqrestore(&ioapic_lock, flags);
313 static void unmask_IO_APIC_irq (unsigned int irq)
317 spin_lock_irqsave(&ioapic_lock, flags);
318 __unmask_IO_APIC_irq(irq);
319 spin_unlock_irqrestore(&ioapic_lock, flags);
322 static void clear_IO_APIC_pin(unsigned int apic, unsigned int pin)
324 struct IO_APIC_route_entry entry;
326 /* Check delivery_mode to be sure we're not clearing an SMI pin */
327 entry = ioapic_read_entry(apic, pin);
328 if (entry.delivery_mode == dest_SMI)
331 * Disable it in the IO-APIC irq-routing table:
333 ioapic_mask_entry(apic, pin);
336 static void clear_IO_APIC (void)
340 for (apic = 0; apic < nr_ioapics; apic++)
341 for (pin = 0; pin < nr_ioapic_registers[apic]; pin++)
342 clear_IO_APIC_pin(apic, pin);
345 int skip_ioapic_setup;
348 /* dummy parsing: see setup.c */
350 static int __init disable_ioapic_setup(char *str)
352 skip_ioapic_setup = 1;
355 early_param("noapic", disable_ioapic_setup);
357 /* Actually the next is obsolete, but keep it for paranoid reasons -AK */
358 static int __init disable_timer_pin_setup(char *arg)
360 disable_timer_pin_1 = 1;
363 __setup("disable_timer_pin_1", disable_timer_pin_setup);
365 static int __init setup_disable_8254_timer(char *s)
367 timer_over_8254 = -1;
370 static int __init setup_enable_8254_timer(char *s)
376 __setup("disable_8254_timer", setup_disable_8254_timer);
377 __setup("enable_8254_timer", setup_enable_8254_timer);
381 * Find the IRQ entry number of a certain pin.
383 static int find_irq_entry(int apic, int pin, int type)
387 for (i = 0; i < mp_irq_entries; i++)
388 if (mp_irqs[i].mpc_irqtype == type &&
389 (mp_irqs[i].mpc_dstapic == mp_ioapics[apic].mpc_apicid ||
390 mp_irqs[i].mpc_dstapic == MP_APIC_ALL) &&
391 mp_irqs[i].mpc_dstirq == pin)
398 * Find the pin to which IRQ[irq] (ISA) is connected
400 static int __init find_isa_irq_pin(int irq, int type)
404 for (i = 0; i < mp_irq_entries; i++) {
405 int lbus = mp_irqs[i].mpc_srcbus;
407 if (test_bit(lbus, mp_bus_not_pci) &&
408 (mp_irqs[i].mpc_irqtype == type) &&
409 (mp_irqs[i].mpc_srcbusirq == irq))
411 return mp_irqs[i].mpc_dstirq;
416 static int __init find_isa_irq_apic(int irq, int type)
420 for (i = 0; i < mp_irq_entries; i++) {
421 int lbus = mp_irqs[i].mpc_srcbus;
423 if (test_bit(lbus, mp_bus_not_pci) &&
424 (mp_irqs[i].mpc_irqtype == type) &&
425 (mp_irqs[i].mpc_srcbusirq == irq))
428 if (i < mp_irq_entries) {
430 for(apic = 0; apic < nr_ioapics; apic++) {
431 if (mp_ioapics[apic].mpc_apicid == mp_irqs[i].mpc_dstapic)
440 * Find a specific PCI IRQ entry.
441 * Not an __init, possibly needed by modules
443 static int pin_2_irq(int idx, int apic, int pin);
445 int IO_APIC_get_PCI_irq_vector(int bus, int slot, int pin)
447 int apic, i, best_guess = -1;
449 apic_printk(APIC_DEBUG, "querying PCI -> IRQ mapping bus:%d, slot:%d, pin:%d.\n",
451 if (mp_bus_id_to_pci_bus[bus] == -1) {
452 apic_printk(APIC_VERBOSE, "PCI BIOS passed nonexistent PCI bus %d!\n", bus);
455 for (i = 0; i < mp_irq_entries; i++) {
456 int lbus = mp_irqs[i].mpc_srcbus;
458 for (apic = 0; apic < nr_ioapics; apic++)
459 if (mp_ioapics[apic].mpc_apicid == mp_irqs[i].mpc_dstapic ||
460 mp_irqs[i].mpc_dstapic == MP_APIC_ALL)
463 if (!test_bit(lbus, mp_bus_not_pci) &&
464 !mp_irqs[i].mpc_irqtype &&
466 (slot == ((mp_irqs[i].mpc_srcbusirq >> 2) & 0x1f))) {
467 int irq = pin_2_irq(i,apic,mp_irqs[i].mpc_dstirq);
469 if (!(apic || IO_APIC_IRQ(irq)))
472 if (pin == (mp_irqs[i].mpc_srcbusirq & 3))
475 * Use the first all-but-pin matching entry as a
476 * best-guess fuzzy result for broken mptables.
482 BUG_ON(best_guess >= NR_IRQS);
486 /* ISA interrupts are always polarity zero edge triggered,
487 * when listed as conforming in the MP table. */
489 #define default_ISA_trigger(idx) (0)
490 #define default_ISA_polarity(idx) (0)
492 /* PCI interrupts are always polarity one level triggered,
493 * when listed as conforming in the MP table. */
495 #define default_PCI_trigger(idx) (1)
496 #define default_PCI_polarity(idx) (1)
498 static int __init MPBIOS_polarity(int idx)
500 int bus = mp_irqs[idx].mpc_srcbus;
504 * Determine IRQ line polarity (high active or low active):
506 switch (mp_irqs[idx].mpc_irqflag & 3)
508 case 0: /* conforms, ie. bus-type dependent polarity */
509 if (test_bit(bus, mp_bus_not_pci))
510 polarity = default_ISA_polarity(idx);
512 polarity = default_PCI_polarity(idx);
514 case 1: /* high active */
519 case 2: /* reserved */
521 printk(KERN_WARNING "broken BIOS!!\n");
525 case 3: /* low active */
530 default: /* invalid */
532 printk(KERN_WARNING "broken BIOS!!\n");
540 static int MPBIOS_trigger(int idx)
542 int bus = mp_irqs[idx].mpc_srcbus;
546 * Determine IRQ trigger mode (edge or level sensitive):
548 switch ((mp_irqs[idx].mpc_irqflag>>2) & 3)
550 case 0: /* conforms, ie. bus-type dependent */
551 if (test_bit(bus, mp_bus_not_pci))
552 trigger = default_ISA_trigger(idx);
554 trigger = default_PCI_trigger(idx);
561 case 2: /* reserved */
563 printk(KERN_WARNING "broken BIOS!!\n");
572 default: /* invalid */
574 printk(KERN_WARNING "broken BIOS!!\n");
582 static inline int irq_polarity(int idx)
584 return MPBIOS_polarity(idx);
587 static inline int irq_trigger(int idx)
589 return MPBIOS_trigger(idx);
592 static int pin_2_irq(int idx, int apic, int pin)
595 int bus = mp_irqs[idx].mpc_srcbus;
598 * Debugging check, we are in big trouble if this message pops up!
600 if (mp_irqs[idx].mpc_dstirq != pin)
601 printk(KERN_ERR "broken BIOS or MPTABLE parser, ayiee!!\n");
603 if (test_bit(bus, mp_bus_not_pci)) {
604 irq = mp_irqs[idx].mpc_srcbusirq;
607 * PCI IRQs are mapped in order
611 irq += nr_ioapic_registers[i++];
614 BUG_ON(irq >= NR_IRQS);
618 static inline int IO_APIC_irq_trigger(int irq)
622 for (apic = 0; apic < nr_ioapics; apic++) {
623 for (pin = 0; pin < nr_ioapic_registers[apic]; pin++) {
624 idx = find_irq_entry(apic,pin,mp_INT);
625 if ((idx != -1) && (irq == pin_2_irq(idx,apic,pin)))
626 return irq_trigger(idx);
630 * nonexistent IRQs are edge default
635 /* irq_vectors is indexed by the sum of all RTEs in all I/O APICs. */
636 static u8 irq_vector[NR_IRQ_VECTORS] __read_mostly = {
637 [0] = FIRST_EXTERNAL_VECTOR + 0,
638 [1] = FIRST_EXTERNAL_VECTOR + 1,
639 [2] = FIRST_EXTERNAL_VECTOR + 2,
640 [3] = FIRST_EXTERNAL_VECTOR + 3,
641 [4] = FIRST_EXTERNAL_VECTOR + 4,
642 [5] = FIRST_EXTERNAL_VECTOR + 5,
643 [6] = FIRST_EXTERNAL_VECTOR + 6,
644 [7] = FIRST_EXTERNAL_VECTOR + 7,
645 [8] = FIRST_EXTERNAL_VECTOR + 8,
646 [9] = FIRST_EXTERNAL_VECTOR + 9,
647 [10] = FIRST_EXTERNAL_VECTOR + 10,
648 [11] = FIRST_EXTERNAL_VECTOR + 11,
649 [12] = FIRST_EXTERNAL_VECTOR + 12,
650 [13] = FIRST_EXTERNAL_VECTOR + 13,
651 [14] = FIRST_EXTERNAL_VECTOR + 14,
652 [15] = FIRST_EXTERNAL_VECTOR + 15,
655 static cpumask_t irq_domain[NR_IRQ_VECTORS] __read_mostly = {
674 static int __assign_irq_vector(int irq, cpumask_t mask, cpumask_t *result)
677 * NOTE! The local APIC isn't very good at handling
678 * multiple interrupts at the same interrupt level.
679 * As the interrupt level is determined by taking the
680 * vector number and shifting that right by 4, we
681 * want to spread these out a bit so that they don't
682 * all fall in the same interrupt level.
684 * Also, we've got to be careful not to trash gate
685 * 0x80, because int 0x80 is hm, kind of importantish. ;)
687 static int current_vector = FIRST_DEVICE_VECTOR, current_offset = 0;
688 cpumask_t old_mask = CPU_MASK_NONE;
692 BUG_ON((unsigned)irq >= NR_IRQ_VECTORS);
694 /* Only try and allocate irqs on cpus that are present */
695 cpus_and(mask, mask, cpu_online_map);
697 if (irq_vector[irq] > 0)
698 old_vector = irq_vector[irq];
699 if (old_vector > 0) {
700 cpus_and(*result, irq_domain[irq], mask);
701 if (!cpus_empty(*result))
703 cpus_and(old_mask, irq_domain[irq], cpu_online_map);
706 for_each_cpu_mask(cpu, mask) {
707 cpumask_t domain, new_mask;
708 int new_cpu, old_cpu;
711 domain = vector_allocation_domain(cpu);
712 cpus_and(new_mask, domain, cpu_online_map);
714 vector = current_vector;
715 offset = current_offset;
718 if (vector >= FIRST_SYSTEM_VECTOR) {
719 /* If we run out of vectors on large boxen, must share them. */
720 offset = (offset + 1) % 8;
721 vector = FIRST_DEVICE_VECTOR + offset;
723 if (unlikely(current_vector == vector))
725 if (vector == IA32_SYSCALL_VECTOR)
727 for_each_cpu_mask(new_cpu, new_mask)
728 if (per_cpu(vector_irq, new_cpu)[vector] != -1)
731 current_vector = vector;
732 current_offset = offset;
733 for_each_cpu_mask(old_cpu, old_mask)
734 per_cpu(vector_irq, old_cpu)[old_vector] = -1;
735 for_each_cpu_mask(new_cpu, new_mask)
736 per_cpu(vector_irq, new_cpu)[vector] = irq;
737 irq_vector[irq] = vector;
738 irq_domain[irq] = domain;
739 cpus_and(*result, domain, mask);
745 static int assign_irq_vector(int irq, cpumask_t mask, cpumask_t *result)
750 spin_lock_irqsave(&vector_lock, flags);
751 vector = __assign_irq_vector(irq, mask, result);
752 spin_unlock_irqrestore(&vector_lock, flags);
756 static void __clear_irq_vector(int irq)
761 BUG_ON(!irq_vector[irq]);
763 vector = irq_vector[irq];
764 cpus_and(mask, irq_domain[irq], cpu_online_map);
765 for_each_cpu_mask(cpu, mask)
766 per_cpu(vector_irq, cpu)[vector] = -1;
769 irq_domain[irq] = CPU_MASK_NONE;
772 void __setup_vector_irq(int cpu)
774 /* Initialize vector_irq on a new cpu */
775 /* This function must be called with vector_lock held */
778 /* Mark the inuse vectors */
779 for (irq = 0; irq < NR_IRQ_VECTORS; ++irq) {
780 if (!cpu_isset(cpu, irq_domain[irq]))
782 vector = irq_vector[irq];
783 per_cpu(vector_irq, cpu)[vector] = irq;
785 /* Mark the free vectors */
786 for (vector = 0; vector < NR_VECTORS; ++vector) {
787 irq = per_cpu(vector_irq, cpu)[vector];
790 if (!cpu_isset(cpu, irq_domain[irq]))
791 per_cpu(vector_irq, cpu)[vector] = -1;
796 static struct irq_chip ioapic_chip;
798 #define IOAPIC_AUTO -1
799 #define IOAPIC_EDGE 0
800 #define IOAPIC_LEVEL 1
802 static void ioapic_register_intr(int irq, int vector, unsigned long trigger)
804 if ((trigger == IOAPIC_AUTO && IO_APIC_irq_trigger(irq)) ||
805 trigger == IOAPIC_LEVEL)
806 set_irq_chip_and_handler_name(irq, &ioapic_chip,
807 handle_fasteoi_irq, "fasteoi");
809 set_irq_chip_and_handler_name(irq, &ioapic_chip,
810 handle_edge_irq, "edge");
812 static void __init setup_IO_APIC_irq(int apic, int pin, int idx, int irq)
814 struct IO_APIC_route_entry entry;
820 * add it to the IO-APIC irq-routing table:
822 memset(&entry,0,sizeof(entry));
824 entry.delivery_mode = INT_DELIVERY_MODE;
825 entry.dest_mode = INT_DEST_MODE;
826 entry.mask = 0; /* enable IRQ */
827 entry.dest = cpu_mask_to_apicid(TARGET_CPUS);
829 entry.trigger = irq_trigger(idx);
830 entry.polarity = irq_polarity(idx);
832 if (irq_trigger(idx)) {
835 entry.dest = cpu_mask_to_apicid(TARGET_CPUS);
838 if (!apic && !IO_APIC_IRQ(irq))
841 if (IO_APIC_IRQ(irq)) {
843 vector = assign_irq_vector(irq, TARGET_CPUS, &mask);
847 entry.dest = cpu_mask_to_apicid(mask);
848 entry.vector = vector;
850 ioapic_register_intr(irq, vector, IOAPIC_AUTO);
851 if (!apic && (irq < 16))
852 disable_8259A_irq(irq);
855 ioapic_write_entry(apic, pin, entry);
857 spin_lock_irqsave(&ioapic_lock, flags);
858 irq_desc[irq].affinity = TARGET_CPUS;
859 spin_unlock_irqrestore(&ioapic_lock, flags);
863 static void __init setup_IO_APIC_irqs(void)
865 int apic, pin, idx, irq, first_notcon = 1;
867 apic_printk(APIC_VERBOSE, KERN_DEBUG "init IO_APIC IRQs\n");
869 for (apic = 0; apic < nr_ioapics; apic++) {
870 for (pin = 0; pin < nr_ioapic_registers[apic]; pin++) {
872 idx = find_irq_entry(apic,pin,mp_INT);
875 apic_printk(APIC_VERBOSE, KERN_DEBUG " IO-APIC (apicid-pin) %d-%d", mp_ioapics[apic].mpc_apicid, pin);
878 apic_printk(APIC_VERBOSE, ", %d-%d", mp_ioapics[apic].mpc_apicid, pin);
882 irq = pin_2_irq(idx, apic, pin);
883 add_pin_to_irq(irq, apic, pin);
885 setup_IO_APIC_irq(apic, pin, idx, irq);
891 apic_printk(APIC_VERBOSE," not connected.\n");
895 * Set up the 8259A-master output pin as broadcast to all
898 static void __init setup_ExtINT_IRQ0_pin(unsigned int apic, unsigned int pin, int vector)
900 struct IO_APIC_route_entry entry;
903 memset(&entry,0,sizeof(entry));
905 disable_8259A_irq(0);
908 apic_write(APIC_LVT0, APIC_LVT_MASKED | APIC_DM_EXTINT);
911 * We use logical delivery to get the timer IRQ
914 entry.dest_mode = INT_DEST_MODE;
915 entry.mask = 0; /* unmask IRQ now */
916 entry.dest = cpu_mask_to_apicid(TARGET_CPUS);
917 entry.delivery_mode = INT_DELIVERY_MODE;
920 entry.vector = vector;
923 * The timer IRQ doesn't have to know that behind the
924 * scene we have a 8259A-master in AEOI mode ...
926 set_irq_chip_and_handler_name(0, &ioapic_chip, handle_edge_irq, "edge");
929 * Add it to the IO-APIC irq-routing table:
931 spin_lock_irqsave(&ioapic_lock, flags);
932 io_apic_write(apic, 0x11+2*pin, *(((int *)&entry)+1));
933 io_apic_write(apic, 0x10+2*pin, *(((int *)&entry)+0));
934 spin_unlock_irqrestore(&ioapic_lock, flags);
939 void __init UNEXPECTED_IO_APIC(void)
943 void __apicdebuginit print_IO_APIC(void)
946 union IO_APIC_reg_00 reg_00;
947 union IO_APIC_reg_01 reg_01;
948 union IO_APIC_reg_02 reg_02;
951 if (apic_verbosity == APIC_QUIET)
954 printk(KERN_DEBUG "number of MP IRQ sources: %d.\n", mp_irq_entries);
955 for (i = 0; i < nr_ioapics; i++)
956 printk(KERN_DEBUG "number of IO-APIC #%d registers: %d.\n",
957 mp_ioapics[i].mpc_apicid, nr_ioapic_registers[i]);
960 * We are a bit conservative about what we expect. We have to
961 * know about every hardware change ASAP.
963 printk(KERN_INFO "testing the IO APIC.......................\n");
965 for (apic = 0; apic < nr_ioapics; apic++) {
967 spin_lock_irqsave(&ioapic_lock, flags);
968 reg_00.raw = io_apic_read(apic, 0);
969 reg_01.raw = io_apic_read(apic, 1);
970 if (reg_01.bits.version >= 0x10)
971 reg_02.raw = io_apic_read(apic, 2);
972 spin_unlock_irqrestore(&ioapic_lock, flags);
975 printk(KERN_DEBUG "IO APIC #%d......\n", mp_ioapics[apic].mpc_apicid);
976 printk(KERN_DEBUG ".... register #00: %08X\n", reg_00.raw);
977 printk(KERN_DEBUG "....... : physical APIC id: %02X\n", reg_00.bits.ID);
978 if (reg_00.bits.__reserved_1 || reg_00.bits.__reserved_2)
979 UNEXPECTED_IO_APIC();
981 printk(KERN_DEBUG ".... register #01: %08X\n", *(int *)®_01);
982 printk(KERN_DEBUG "....... : max redirection entries: %04X\n", reg_01.bits.entries);
983 if ( (reg_01.bits.entries != 0x0f) && /* older (Neptune) boards */
984 (reg_01.bits.entries != 0x17) && /* typical ISA+PCI boards */
985 (reg_01.bits.entries != 0x1b) && /* Compaq Proliant boards */
986 (reg_01.bits.entries != 0x1f) && /* dual Xeon boards */
987 (reg_01.bits.entries != 0x22) && /* bigger Xeon boards */
988 (reg_01.bits.entries != 0x2E) &&
989 (reg_01.bits.entries != 0x3F) &&
990 (reg_01.bits.entries != 0x03)
992 UNEXPECTED_IO_APIC();
994 printk(KERN_DEBUG "....... : PRQ implemented: %X\n", reg_01.bits.PRQ);
995 printk(KERN_DEBUG "....... : IO APIC version: %04X\n", reg_01.bits.version);
996 if ( (reg_01.bits.version != 0x01) && /* 82489DX IO-APICs */
997 (reg_01.bits.version != 0x02) && /* 82801BA IO-APICs (ICH2) */
998 (reg_01.bits.version != 0x10) && /* oldest IO-APICs */
999 (reg_01.bits.version != 0x11) && /* Pentium/Pro IO-APICs */
1000 (reg_01.bits.version != 0x13) && /* Xeon IO-APICs */
1001 (reg_01.bits.version != 0x20) /* Intel P64H (82806 AA) */
1003 UNEXPECTED_IO_APIC();
1004 if (reg_01.bits.__reserved_1 || reg_01.bits.__reserved_2)
1005 UNEXPECTED_IO_APIC();
1007 if (reg_01.bits.version >= 0x10) {
1008 printk(KERN_DEBUG ".... register #02: %08X\n", reg_02.raw);
1009 printk(KERN_DEBUG "....... : arbitration: %02X\n", reg_02.bits.arbitration);
1010 if (reg_02.bits.__reserved_1 || reg_02.bits.__reserved_2)
1011 UNEXPECTED_IO_APIC();
1014 printk(KERN_DEBUG ".... IRQ redirection table:\n");
1016 printk(KERN_DEBUG " NR Dst Mask Trig IRR Pol"
1017 " Stat Dmod Deli Vect: \n");
1019 for (i = 0; i <= reg_01.bits.entries; i++) {
1020 struct IO_APIC_route_entry entry;
1022 entry = ioapic_read_entry(apic, i);
1024 printk(KERN_DEBUG " %02x %03X ",
1029 printk("%1d %1d %1d %1d %1d %1d %1d %02X\n",
1034 entry.delivery_status,
1036 entry.delivery_mode,
1041 printk(KERN_DEBUG "IRQ to pin mappings:\n");
1042 for (i = 0; i < NR_IRQS; i++) {
1043 struct irq_pin_list *entry = irq_2_pin + i;
1046 printk(KERN_DEBUG "IRQ%d ", i);
1048 printk("-> %d:%d", entry->apic, entry->pin);
1051 entry = irq_2_pin + entry->next;
1056 printk(KERN_INFO ".................................... done.\n");
1063 static __apicdebuginit void print_APIC_bitfield (int base)
1068 if (apic_verbosity == APIC_QUIET)
1071 printk(KERN_DEBUG "0123456789abcdef0123456789abcdef\n" KERN_DEBUG);
1072 for (i = 0; i < 8; i++) {
1073 v = apic_read(base + i*0x10);
1074 for (j = 0; j < 32; j++) {
1084 void __apicdebuginit print_local_APIC(void * dummy)
1086 unsigned int v, ver, maxlvt;
1088 if (apic_verbosity == APIC_QUIET)
1091 printk("\n" KERN_DEBUG "printing local APIC contents on CPU#%d/%d:\n",
1092 smp_processor_id(), hard_smp_processor_id());
1093 v = apic_read(APIC_ID);
1094 printk(KERN_INFO "... APIC ID: %08x (%01x)\n", v, GET_APIC_ID(v));
1095 v = apic_read(APIC_LVR);
1096 printk(KERN_INFO "... APIC VERSION: %08x\n", v);
1097 ver = GET_APIC_VERSION(v);
1098 maxlvt = get_maxlvt();
1100 v = apic_read(APIC_TASKPRI);
1101 printk(KERN_DEBUG "... APIC TASKPRI: %08x (%02x)\n", v, v & APIC_TPRI_MASK);
1103 v = apic_read(APIC_ARBPRI);
1104 printk(KERN_DEBUG "... APIC ARBPRI: %08x (%02x)\n", v,
1105 v & APIC_ARBPRI_MASK);
1106 v = apic_read(APIC_PROCPRI);
1107 printk(KERN_DEBUG "... APIC PROCPRI: %08x\n", v);
1109 v = apic_read(APIC_EOI);
1110 printk(KERN_DEBUG "... APIC EOI: %08x\n", v);
1111 v = apic_read(APIC_RRR);
1112 printk(KERN_DEBUG "... APIC RRR: %08x\n", v);
1113 v = apic_read(APIC_LDR);
1114 printk(KERN_DEBUG "... APIC LDR: %08x\n", v);
1115 v = apic_read(APIC_DFR);
1116 printk(KERN_DEBUG "... APIC DFR: %08x\n", v);
1117 v = apic_read(APIC_SPIV);
1118 printk(KERN_DEBUG "... APIC SPIV: %08x\n", v);
1120 printk(KERN_DEBUG "... APIC ISR field:\n");
1121 print_APIC_bitfield(APIC_ISR);
1122 printk(KERN_DEBUG "... APIC TMR field:\n");
1123 print_APIC_bitfield(APIC_TMR);
1124 printk(KERN_DEBUG "... APIC IRR field:\n");
1125 print_APIC_bitfield(APIC_IRR);
1127 v = apic_read(APIC_ESR);
1128 printk(KERN_DEBUG "... APIC ESR: %08x\n", v);
1130 v = apic_read(APIC_ICR);
1131 printk(KERN_DEBUG "... APIC ICR: %08x\n", v);
1132 v = apic_read(APIC_ICR2);
1133 printk(KERN_DEBUG "... APIC ICR2: %08x\n", v);
1135 v = apic_read(APIC_LVTT);
1136 printk(KERN_DEBUG "... APIC LVTT: %08x\n", v);
1138 if (maxlvt > 3) { /* PC is LVT#4. */
1139 v = apic_read(APIC_LVTPC);
1140 printk(KERN_DEBUG "... APIC LVTPC: %08x\n", v);
1142 v = apic_read(APIC_LVT0);
1143 printk(KERN_DEBUG "... APIC LVT0: %08x\n", v);
1144 v = apic_read(APIC_LVT1);
1145 printk(KERN_DEBUG "... APIC LVT1: %08x\n", v);
1147 if (maxlvt > 2) { /* ERR is LVT#3. */
1148 v = apic_read(APIC_LVTERR);
1149 printk(KERN_DEBUG "... APIC LVTERR: %08x\n", v);
1152 v = apic_read(APIC_TMICT);
1153 printk(KERN_DEBUG "... APIC TMICT: %08x\n", v);
1154 v = apic_read(APIC_TMCCT);
1155 printk(KERN_DEBUG "... APIC TMCCT: %08x\n", v);
1156 v = apic_read(APIC_TDCR);
1157 printk(KERN_DEBUG "... APIC TDCR: %08x\n", v);
1161 void print_all_local_APICs (void)
1163 on_each_cpu(print_local_APIC, NULL, 1, 1);
1166 void __apicdebuginit print_PIC(void)
1169 unsigned long flags;
1171 if (apic_verbosity == APIC_QUIET)
1174 printk(KERN_DEBUG "\nprinting PIC contents\n");
1176 spin_lock_irqsave(&i8259A_lock, flags);
1178 v = inb(0xa1) << 8 | inb(0x21);
1179 printk(KERN_DEBUG "... PIC IMR: %04x\n", v);
1181 v = inb(0xa0) << 8 | inb(0x20);
1182 printk(KERN_DEBUG "... PIC IRR: %04x\n", v);
1186 v = inb(0xa0) << 8 | inb(0x20);
1190 spin_unlock_irqrestore(&i8259A_lock, flags);
1192 printk(KERN_DEBUG "... PIC ISR: %04x\n", v);
1194 v = inb(0x4d1) << 8 | inb(0x4d0);
1195 printk(KERN_DEBUG "... PIC ELCR: %04x\n", v);
1200 static void __init enable_IO_APIC(void)
1202 union IO_APIC_reg_01 reg_01;
1203 int i8259_apic, i8259_pin;
1205 unsigned long flags;
1207 for (i = 0; i < PIN_MAP_SIZE; i++) {
1208 irq_2_pin[i].pin = -1;
1209 irq_2_pin[i].next = 0;
1213 * The number of IO-APIC IRQ registers (== #pins):
1215 for (apic = 0; apic < nr_ioapics; apic++) {
1216 spin_lock_irqsave(&ioapic_lock, flags);
1217 reg_01.raw = io_apic_read(apic, 1);
1218 spin_unlock_irqrestore(&ioapic_lock, flags);
1219 nr_ioapic_registers[apic] = reg_01.bits.entries+1;
1221 for(apic = 0; apic < nr_ioapics; apic++) {
1223 /* See if any of the pins is in ExtINT mode */
1224 for (pin = 0; pin < nr_ioapic_registers[apic]; pin++) {
1225 struct IO_APIC_route_entry entry;
1226 entry = ioapic_read_entry(apic, pin);
1228 /* If the interrupt line is enabled and in ExtInt mode
1229 * I have found the pin where the i8259 is connected.
1231 if ((entry.mask == 0) && (entry.delivery_mode == dest_ExtINT)) {
1232 ioapic_i8259.apic = apic;
1233 ioapic_i8259.pin = pin;
1239 /* Look to see what if the MP table has reported the ExtINT */
1240 i8259_pin = find_isa_irq_pin(0, mp_ExtINT);
1241 i8259_apic = find_isa_irq_apic(0, mp_ExtINT);
1242 /* Trust the MP table if nothing is setup in the hardware */
1243 if ((ioapic_i8259.pin == -1) && (i8259_pin >= 0)) {
1244 printk(KERN_WARNING "ExtINT not setup in hardware but reported by MP table\n");
1245 ioapic_i8259.pin = i8259_pin;
1246 ioapic_i8259.apic = i8259_apic;
1248 /* Complain if the MP table and the hardware disagree */
1249 if (((ioapic_i8259.apic != i8259_apic) || (ioapic_i8259.pin != i8259_pin)) &&
1250 (i8259_pin >= 0) && (ioapic_i8259.pin >= 0))
1252 printk(KERN_WARNING "ExtINT in hardware and MP table differ\n");
1256 * Do not trust the IO-APIC being empty at bootup
1262 * Not an __init, needed by the reboot code
1264 void disable_IO_APIC(void)
1267 * Clear the IO-APIC before rebooting:
1272 * If the i8259 is routed through an IOAPIC
1273 * Put that IOAPIC in virtual wire mode
1274 * so legacy interrupts can be delivered.
1276 if (ioapic_i8259.pin != -1) {
1277 struct IO_APIC_route_entry entry;
1279 memset(&entry, 0, sizeof(entry));
1280 entry.mask = 0; /* Enabled */
1281 entry.trigger = 0; /* Edge */
1283 entry.polarity = 0; /* High */
1284 entry.delivery_status = 0;
1285 entry.dest_mode = 0; /* Physical */
1286 entry.delivery_mode = dest_ExtINT; /* ExtInt */
1288 entry.dest = GET_APIC_ID(apic_read(APIC_ID));
1291 * Add it to the IO-APIC irq-routing table:
1293 ioapic_write_entry(ioapic_i8259.apic, ioapic_i8259.pin, entry);
1296 disconnect_bsp_APIC(ioapic_i8259.pin != -1);
1300 * There is a nasty bug in some older SMP boards, their mptable lies
1301 * about the timer IRQ. We do the following to work around the situation:
1303 * - timer IRQ defaults to IO-APIC IRQ
1304 * - if this function detects that timer IRQs are defunct, then we fall
1305 * back to ISA timer IRQs
1307 static int __init timer_irq_works(void)
1309 unsigned long t1 = jiffies;
1312 /* Let ten ticks pass... */
1313 mdelay((10 * 1000) / HZ);
1316 * Expect a few ticks at least, to be sure some possible
1317 * glue logic does not lock up after one or two first
1318 * ticks in a non-ExtINT mode. Also the local APIC
1319 * might have cached one ExtINT interrupt. Finally, at
1320 * least one tick may be lost due to delays.
1324 if (jiffies - t1 > 4)
1330 * In the SMP+IOAPIC case it might happen that there are an unspecified
1331 * number of pending IRQ events unhandled. These cases are very rare,
1332 * so we 'resend' these IRQs via IPIs, to the same CPU. It's much
1333 * better to do it this way as thus we do not have to be aware of
1334 * 'pending' interrupts in the IRQ path, except at this point.
1337 * Edge triggered needs to resend any interrupt
1338 * that was delayed but this is now handled in the device
1343 * Starting up a edge-triggered IO-APIC interrupt is
1344 * nasty - we need to make sure that we get the edge.
1345 * If it is already asserted for some reason, we need
1346 * return 1 to indicate that is was pending.
1348 * This is not complete - we should be able to fake
1349 * an edge even if it isn't on the 8259A...
1352 static unsigned int startup_ioapic_irq(unsigned int irq)
1354 int was_pending = 0;
1355 unsigned long flags;
1357 spin_lock_irqsave(&ioapic_lock, flags);
1359 disable_8259A_irq(irq);
1360 if (i8259A_irq_pending(irq))
1363 __unmask_IO_APIC_irq(irq);
1364 spin_unlock_irqrestore(&ioapic_lock, flags);
1369 static int ioapic_retrigger_irq(unsigned int irq)
1373 unsigned long flags;
1375 spin_lock_irqsave(&vector_lock, flags);
1376 vector = irq_vector[irq];
1378 cpu_set(first_cpu(irq_domain[irq]), mask);
1380 send_IPI_mask(mask, vector);
1381 spin_unlock_irqrestore(&vector_lock, flags);
1387 * Level and edge triggered IO-APIC interrupts need different handling,
1388 * so we use two separate IRQ descriptors. Edge triggered IRQs can be
1389 * handled with the level-triggered descriptor, but that one has slightly
1390 * more overhead. Level-triggered interrupts cannot be handled with the
1391 * edge-triggered handler, without risking IRQ storms and other ugly
1395 static void ack_apic_edge(unsigned int irq)
1397 move_native_irq(irq);
1401 static void ack_apic_level(unsigned int irq)
1403 int do_unmask_irq = 0;
1405 #if defined(CONFIG_GENERIC_PENDING_IRQ) || defined(CONFIG_IRQBALANCE)
1406 /* If we are moving the irq we need to mask it */
1407 if (unlikely(irq_desc[irq].status & IRQ_MOVE_PENDING)) {
1409 mask_IO_APIC_irq(irq);
1414 * We must acknowledge the irq before we move it or the acknowledge will
1415 * not propogate properly.
1419 /* Now we can move and renable the irq */
1420 move_masked_irq(irq);
1421 if (unlikely(do_unmask_irq))
1422 unmask_IO_APIC_irq(irq);
1425 static struct irq_chip ioapic_chip __read_mostly = {
1427 .startup = startup_ioapic_irq,
1428 .mask = mask_IO_APIC_irq,
1429 .unmask = unmask_IO_APIC_irq,
1430 .ack = ack_apic_edge,
1431 .eoi = ack_apic_level,
1433 .set_affinity = set_ioapic_affinity_irq,
1435 .retrigger = ioapic_retrigger_irq,
1438 static inline void init_IO_APIC_traps(void)
1443 * NOTE! The local APIC isn't very good at handling
1444 * multiple interrupts at the same interrupt level.
1445 * As the interrupt level is determined by taking the
1446 * vector number and shifting that right by 4, we
1447 * want to spread these out a bit so that they don't
1448 * all fall in the same interrupt level.
1450 * Also, we've got to be careful not to trash gate
1451 * 0x80, because int 0x80 is hm, kind of importantish. ;)
1453 for (irq = 0; irq < NR_IRQS ; irq++) {
1455 if (IO_APIC_IRQ(tmp) && !irq_vector[tmp]) {
1457 * Hmm.. We don't have an entry for this,
1458 * so default to an old-fashioned 8259
1459 * interrupt if we can..
1462 make_8259A_irq(irq);
1464 /* Strange. Oh, well.. */
1465 irq_desc[irq].chip = &no_irq_chip;
1470 static void enable_lapic_irq (unsigned int irq)
1474 v = apic_read(APIC_LVT0);
1475 apic_write(APIC_LVT0, v & ~APIC_LVT_MASKED);
1478 static void disable_lapic_irq (unsigned int irq)
1482 v = apic_read(APIC_LVT0);
1483 apic_write(APIC_LVT0, v | APIC_LVT_MASKED);
1486 static void ack_lapic_irq (unsigned int irq)
1491 static void end_lapic_irq (unsigned int i) { /* nothing */ }
1493 static struct hw_interrupt_type lapic_irq_type __read_mostly = {
1494 .typename = "local-APIC-edge",
1495 .startup = NULL, /* startup_irq() not used for IRQ0 */
1496 .shutdown = NULL, /* shutdown_irq() not used for IRQ0 */
1497 .enable = enable_lapic_irq,
1498 .disable = disable_lapic_irq,
1499 .ack = ack_lapic_irq,
1500 .end = end_lapic_irq,
1503 static void setup_nmi (void)
1506 * Dirty trick to enable the NMI watchdog ...
1507 * We put the 8259A master into AEOI mode and
1508 * unmask on all local APICs LVT0 as NMI.
1510 * The idea to use the 8259A in AEOI mode ('8259A Virtual Wire')
1511 * is from Maciej W. Rozycki - so we do not have to EOI from
1512 * the NMI handler or the timer interrupt.
1514 printk(KERN_INFO "activating NMI Watchdog ...");
1516 enable_NMI_through_LVT0(NULL);
1522 * This looks a bit hackish but it's about the only one way of sending
1523 * a few INTA cycles to 8259As and any associated glue logic. ICR does
1524 * not support the ExtINT mode, unfortunately. We need to send these
1525 * cycles as some i82489DX-based boards have glue logic that keeps the
1526 * 8259A interrupt line asserted until INTA. --macro
1528 static inline void unlock_ExtINT_logic(void)
1531 struct IO_APIC_route_entry entry0, entry1;
1532 unsigned char save_control, save_freq_select;
1533 unsigned long flags;
1535 pin = find_isa_irq_pin(8, mp_INT);
1536 apic = find_isa_irq_apic(8, mp_INT);
1540 spin_lock_irqsave(&ioapic_lock, flags);
1541 *(((int *)&entry0) + 1) = io_apic_read(apic, 0x11 + 2 * pin);
1542 *(((int *)&entry0) + 0) = io_apic_read(apic, 0x10 + 2 * pin);
1543 spin_unlock_irqrestore(&ioapic_lock, flags);
1544 clear_IO_APIC_pin(apic, pin);
1546 memset(&entry1, 0, sizeof(entry1));
1548 entry1.dest_mode = 0; /* physical delivery */
1549 entry1.mask = 0; /* unmask IRQ now */
1550 entry1.dest = hard_smp_processor_id();
1551 entry1.delivery_mode = dest_ExtINT;
1552 entry1.polarity = entry0.polarity;
1556 spin_lock_irqsave(&ioapic_lock, flags);
1557 io_apic_write(apic, 0x11 + 2 * pin, *(((int *)&entry1) + 1));
1558 io_apic_write(apic, 0x10 + 2 * pin, *(((int *)&entry1) + 0));
1559 spin_unlock_irqrestore(&ioapic_lock, flags);
1561 save_control = CMOS_READ(RTC_CONTROL);
1562 save_freq_select = CMOS_READ(RTC_FREQ_SELECT);
1563 CMOS_WRITE((save_freq_select & ~RTC_RATE_SELECT) | 0x6,
1565 CMOS_WRITE(save_control | RTC_PIE, RTC_CONTROL);
1570 if ((CMOS_READ(RTC_INTR_FLAGS) & RTC_PF) == RTC_PF)
1574 CMOS_WRITE(save_control, RTC_CONTROL);
1575 CMOS_WRITE(save_freq_select, RTC_FREQ_SELECT);
1576 clear_IO_APIC_pin(apic, pin);
1578 spin_lock_irqsave(&ioapic_lock, flags);
1579 io_apic_write(apic, 0x11 + 2 * pin, *(((int *)&entry0) + 1));
1580 io_apic_write(apic, 0x10 + 2 * pin, *(((int *)&entry0) + 0));
1581 spin_unlock_irqrestore(&ioapic_lock, flags);
1585 * This code may look a bit paranoid, but it's supposed to cooperate with
1586 * a wide range of boards and BIOS bugs. Fortunately only the timer IRQ
1587 * is so screwy. Thanks to Brian Perkins for testing/hacking this beast
1588 * fanatically on his truly buggy board.
1590 * FIXME: really need to revamp this for modern platforms only.
1592 static inline void check_timer(void)
1594 int apic1, pin1, apic2, pin2;
1599 * get/set the timer IRQ vector:
1601 disable_8259A_irq(0);
1602 vector = assign_irq_vector(0, TARGET_CPUS, &mask);
1605 * Subtle, code in do_timer_interrupt() expects an AEOI
1606 * mode for the 8259A whenever interrupts are routed
1607 * through I/O APICs. Also IRQ0 has to be enabled in
1608 * the 8259A which implies the virtual wire has to be
1609 * disabled in the local APIC.
1611 apic_write(APIC_LVT0, APIC_LVT_MASKED | APIC_DM_EXTINT);
1613 if (timer_over_8254 > 0)
1614 enable_8259A_irq(0);
1616 pin1 = find_isa_irq_pin(0, mp_INT);
1617 apic1 = find_isa_irq_apic(0, mp_INT);
1618 pin2 = ioapic_i8259.pin;
1619 apic2 = ioapic_i8259.apic;
1621 apic_printk(APIC_VERBOSE,KERN_INFO "..TIMER: vector=0x%02X apic1=%d pin1=%d apic2=%d pin2=%d\n",
1622 vector, apic1, pin1, apic2, pin2);
1626 * Ok, does IRQ0 through the IOAPIC work?
1628 unmask_IO_APIC_irq(0);
1629 if (!no_timer_check && timer_irq_works()) {
1630 nmi_watchdog_default();
1631 if (nmi_watchdog == NMI_IO_APIC) {
1632 disable_8259A_irq(0);
1634 enable_8259A_irq(0);
1636 if (disable_timer_pin_1 > 0)
1637 clear_IO_APIC_pin(0, pin1);
1640 clear_IO_APIC_pin(apic1, pin1);
1641 apic_printk(APIC_QUIET,KERN_ERR "..MP-BIOS bug: 8254 timer not "
1642 "connected to IO-APIC\n");
1645 apic_printk(APIC_VERBOSE,KERN_INFO "...trying to set up timer (IRQ0) "
1646 "through the 8259A ... ");
1648 apic_printk(APIC_VERBOSE,"\n..... (found apic %d pin %d) ...",
1651 * legacy devices should be connected to IO APIC #0
1653 setup_ExtINT_IRQ0_pin(apic2, pin2, vector);
1654 if (timer_irq_works()) {
1655 apic_printk(APIC_VERBOSE," works.\n");
1656 nmi_watchdog_default();
1657 if (nmi_watchdog == NMI_IO_APIC) {
1663 * Cleanup, just in case ...
1665 clear_IO_APIC_pin(apic2, pin2);
1667 apic_printk(APIC_VERBOSE," failed.\n");
1669 if (nmi_watchdog == NMI_IO_APIC) {
1670 printk(KERN_WARNING "timer doesn't work through the IO-APIC - disabling NMI Watchdog!\n");
1674 apic_printk(APIC_VERBOSE, KERN_INFO "...trying to set up timer as Virtual Wire IRQ...");
1676 disable_8259A_irq(0);
1677 irq_desc[0].chip = &lapic_irq_type;
1678 apic_write(APIC_LVT0, APIC_DM_FIXED | vector); /* Fixed mode */
1679 enable_8259A_irq(0);
1681 if (timer_irq_works()) {
1682 apic_printk(APIC_VERBOSE," works.\n");
1685 apic_write(APIC_LVT0, APIC_LVT_MASKED | APIC_DM_FIXED | vector);
1686 apic_printk(APIC_VERBOSE," failed.\n");
1688 apic_printk(APIC_VERBOSE, KERN_INFO "...trying to set up timer as ExtINT IRQ...");
1692 apic_write(APIC_LVT0, APIC_DM_EXTINT);
1694 unlock_ExtINT_logic();
1696 if (timer_irq_works()) {
1697 apic_printk(APIC_VERBOSE," works.\n");
1700 apic_printk(APIC_VERBOSE," failed :(.\n");
1701 panic("IO-APIC + timer doesn't work! Try using the 'noapic' kernel parameter\n");
1704 static int __init notimercheck(char *s)
1709 __setup("no_timer_check", notimercheck);
1713 * IRQ's that are handled by the PIC in the MPS IOAPIC case.
1714 * - IRQ2 is the cascade IRQ, and cannot be a io-apic IRQ.
1715 * Linux doesn't really care, as it's not actually used
1716 * for any interrupt handling anyway.
1718 #define PIC_IRQS (1<<2)
1720 void __init setup_IO_APIC(void)
1725 io_apic_irqs = ~0; /* all IRQs go through IOAPIC */
1727 io_apic_irqs = ~PIC_IRQS;
1729 apic_printk(APIC_VERBOSE, "ENABLING IO-APIC IRQs\n");
1732 setup_IO_APIC_irqs();
1733 init_IO_APIC_traps();
1739 struct sysfs_ioapic_data {
1740 struct sys_device dev;
1741 struct IO_APIC_route_entry entry[0];
1743 static struct sysfs_ioapic_data * mp_ioapic_data[MAX_IO_APICS];
1745 static int ioapic_suspend(struct sys_device *dev, pm_message_t state)
1747 struct IO_APIC_route_entry *entry;
1748 struct sysfs_ioapic_data *data;
1751 data = container_of(dev, struct sysfs_ioapic_data, dev);
1752 entry = data->entry;
1753 for (i = 0; i < nr_ioapic_registers[dev->id]; i ++, entry ++ )
1754 *entry = ioapic_read_entry(dev->id, i);
1759 static int ioapic_resume(struct sys_device *dev)
1761 struct IO_APIC_route_entry *entry;
1762 struct sysfs_ioapic_data *data;
1763 unsigned long flags;
1764 union IO_APIC_reg_00 reg_00;
1767 data = container_of(dev, struct sysfs_ioapic_data, dev);
1768 entry = data->entry;
1770 spin_lock_irqsave(&ioapic_lock, flags);
1771 reg_00.raw = io_apic_read(dev->id, 0);
1772 if (reg_00.bits.ID != mp_ioapics[dev->id].mpc_apicid) {
1773 reg_00.bits.ID = mp_ioapics[dev->id].mpc_apicid;
1774 io_apic_write(dev->id, 0, reg_00.raw);
1776 spin_unlock_irqrestore(&ioapic_lock, flags);
1777 for (i = 0; i < nr_ioapic_registers[dev->id]; i++)
1778 ioapic_write_entry(dev->id, i, entry[i]);
1783 static struct sysdev_class ioapic_sysdev_class = {
1784 set_kset_name("ioapic"),
1785 .suspend = ioapic_suspend,
1786 .resume = ioapic_resume,
1789 static int __init ioapic_init_sysfs(void)
1791 struct sys_device * dev;
1792 int i, size, error = 0;
1794 error = sysdev_class_register(&ioapic_sysdev_class);
1798 for (i = 0; i < nr_ioapics; i++ ) {
1799 size = sizeof(struct sys_device) + nr_ioapic_registers[i]
1800 * sizeof(struct IO_APIC_route_entry);
1801 mp_ioapic_data[i] = kmalloc(size, GFP_KERNEL);
1802 if (!mp_ioapic_data[i]) {
1803 printk(KERN_ERR "Can't suspend/resume IOAPIC %d\n", i);
1806 memset(mp_ioapic_data[i], 0, size);
1807 dev = &mp_ioapic_data[i]->dev;
1809 dev->cls = &ioapic_sysdev_class;
1810 error = sysdev_register(dev);
1812 kfree(mp_ioapic_data[i]);
1813 mp_ioapic_data[i] = NULL;
1814 printk(KERN_ERR "Can't suspend/resume IOAPIC %d\n", i);
1822 device_initcall(ioapic_init_sysfs);
1825 * Dynamic irq allocate and deallocation
1827 int create_irq(void)
1829 /* Allocate an unused irq */
1833 unsigned long flags;
1837 spin_lock_irqsave(&vector_lock, flags);
1838 for (new = (NR_IRQS - 1); new >= 0; new--) {
1839 if (platform_legacy_irq(new))
1841 if (irq_vector[new] != 0)
1843 vector = __assign_irq_vector(new, TARGET_CPUS, &mask);
1844 if (likely(vector > 0))
1848 spin_unlock_irqrestore(&vector_lock, flags);
1851 dynamic_irq_init(irq);
1856 void destroy_irq(unsigned int irq)
1858 unsigned long flags;
1860 dynamic_irq_cleanup(irq);
1862 spin_lock_irqsave(&vector_lock, flags);
1863 __clear_irq_vector(irq);
1864 spin_unlock_irqrestore(&vector_lock, flags);
1868 * MSI mesage composition
1870 #ifdef CONFIG_PCI_MSI
1871 static int msi_compose_msg(struct pci_dev *pdev, unsigned int irq, struct msi_msg *msg)
1877 vector = assign_irq_vector(irq, TARGET_CPUS, &tmp);
1879 dest = cpu_mask_to_apicid(tmp);
1881 msg->address_hi = MSI_ADDR_BASE_HI;
1884 ((INT_DEST_MODE == 0) ?
1885 MSI_ADDR_DEST_MODE_PHYSICAL:
1886 MSI_ADDR_DEST_MODE_LOGICAL) |
1887 ((INT_DELIVERY_MODE != dest_LowestPrio) ?
1888 MSI_ADDR_REDIRECTION_CPU:
1889 MSI_ADDR_REDIRECTION_LOWPRI) |
1890 MSI_ADDR_DEST_ID(dest);
1893 MSI_DATA_TRIGGER_EDGE |
1894 MSI_DATA_LEVEL_ASSERT |
1895 ((INT_DELIVERY_MODE != dest_LowestPrio) ?
1896 MSI_DATA_DELIVERY_FIXED:
1897 MSI_DATA_DELIVERY_LOWPRI) |
1898 MSI_DATA_VECTOR(vector);
1904 static void set_msi_irq_affinity(unsigned int irq, cpumask_t mask)
1911 cpus_and(tmp, mask, cpu_online_map);
1912 if (cpus_empty(tmp))
1915 cpus_and(mask, tmp, CPU_MASK_ALL);
1917 vector = assign_irq_vector(irq, mask, &tmp);
1921 dest = cpu_mask_to_apicid(tmp);
1923 read_msi_msg(irq, &msg);
1925 msg.data &= ~MSI_DATA_VECTOR_MASK;
1926 msg.data |= MSI_DATA_VECTOR(vector);
1927 msg.address_lo &= ~MSI_ADDR_DEST_ID_MASK;
1928 msg.address_lo |= MSI_ADDR_DEST_ID(dest);
1930 write_msi_msg(irq, &msg);
1931 irq_desc[irq].affinity = mask;
1933 #endif /* CONFIG_SMP */
1936 * IRQ Chip for MSI PCI/PCI-X/PCI-Express Devices,
1937 * which implement the MSI or MSI-X Capability Structure.
1939 static struct irq_chip msi_chip = {
1941 .unmask = unmask_msi_irq,
1942 .mask = mask_msi_irq,
1943 .ack = ack_apic_edge,
1945 .set_affinity = set_msi_irq_affinity,
1947 .retrigger = ioapic_retrigger_irq,
1950 int arch_setup_msi_irq(struct pci_dev *dev, struct msi_desc *desc)
1958 set_irq_msi(irq, desc);
1959 ret = msi_compose_msg(dev, irq, &msg);
1965 write_msi_msg(irq, &msg);
1967 set_irq_chip_and_handler_name(irq, &msi_chip, handle_edge_irq, "edge");
1972 void arch_teardown_msi_irq(unsigned int irq)
1977 #endif /* CONFIG_PCI_MSI */
1980 * Hypertransport interrupt support
1982 #ifdef CONFIG_HT_IRQ
1986 static void target_ht_irq(unsigned int irq, unsigned int dest, u8 vector)
1988 struct ht_irq_msg msg;
1989 fetch_ht_irq_msg(irq, &msg);
1991 msg.address_lo &= ~(HT_IRQ_LOW_VECTOR_MASK | HT_IRQ_LOW_DEST_ID_MASK);
1992 msg.address_hi &= ~(HT_IRQ_HIGH_DEST_ID_MASK);
1994 msg.address_lo |= HT_IRQ_LOW_VECTOR(vector) | HT_IRQ_LOW_DEST_ID(dest);
1995 msg.address_hi |= HT_IRQ_HIGH_DEST_ID(dest);
1997 write_ht_irq_msg(irq, &msg);
2000 static void set_ht_irq_affinity(unsigned int irq, cpumask_t mask)
2006 cpus_and(tmp, mask, cpu_online_map);
2007 if (cpus_empty(tmp))
2010 cpus_and(mask, tmp, CPU_MASK_ALL);
2012 vector = assign_irq_vector(irq, mask, &tmp);
2016 dest = cpu_mask_to_apicid(tmp);
2018 target_ht_irq(irq, dest, vector);
2019 irq_desc[irq].affinity = mask;
2023 static struct irq_chip ht_irq_chip = {
2025 .mask = mask_ht_irq,
2026 .unmask = unmask_ht_irq,
2027 .ack = ack_apic_edge,
2029 .set_affinity = set_ht_irq_affinity,
2031 .retrigger = ioapic_retrigger_irq,
2034 int arch_setup_ht_irq(unsigned int irq, struct pci_dev *dev)
2039 vector = assign_irq_vector(irq, TARGET_CPUS, &tmp);
2041 struct ht_irq_msg msg;
2044 dest = cpu_mask_to_apicid(tmp);
2046 msg.address_hi = HT_IRQ_HIGH_DEST_ID(dest);
2050 HT_IRQ_LOW_DEST_ID(dest) |
2051 HT_IRQ_LOW_VECTOR(vector) |
2052 ((INT_DEST_MODE == 0) ?
2053 HT_IRQ_LOW_DM_PHYSICAL :
2054 HT_IRQ_LOW_DM_LOGICAL) |
2055 HT_IRQ_LOW_RQEOI_EDGE |
2056 ((INT_DELIVERY_MODE != dest_LowestPrio) ?
2057 HT_IRQ_LOW_MT_FIXED :
2058 HT_IRQ_LOW_MT_ARBITRATED) |
2059 HT_IRQ_LOW_IRQ_MASKED;
2061 write_ht_irq_msg(irq, &msg);
2063 set_irq_chip_and_handler_name(irq, &ht_irq_chip,
2064 handle_edge_irq, "edge");
2068 #endif /* CONFIG_HT_IRQ */
2070 /* --------------------------------------------------------------------------
2071 ACPI-based IOAPIC Configuration
2072 -------------------------------------------------------------------------- */
2076 #define IO_APIC_MAX_ID 0xFE
2078 int __init io_apic_get_redir_entries (int ioapic)
2080 union IO_APIC_reg_01 reg_01;
2081 unsigned long flags;
2083 spin_lock_irqsave(&ioapic_lock, flags);
2084 reg_01.raw = io_apic_read(ioapic, 1);
2085 spin_unlock_irqrestore(&ioapic_lock, flags);
2087 return reg_01.bits.entries;
2091 int io_apic_set_pci_routing (int ioapic, int pin, int irq, int triggering, int polarity)
2093 struct IO_APIC_route_entry entry;
2094 unsigned long flags;
2098 if (!IO_APIC_IRQ(irq)) {
2099 apic_printk(APIC_QUIET,KERN_ERR "IOAPIC[%d]: Invalid reference to IRQ 0\n",
2105 * IRQs < 16 are already in the irq_2_pin[] map
2108 add_pin_to_irq(irq, ioapic, pin);
2111 vector = assign_irq_vector(irq, TARGET_CPUS, &mask);
2116 * Generate a PCI IRQ routing entry and program the IOAPIC accordingly.
2117 * Note that we mask (disable) IRQs now -- these get enabled when the
2118 * corresponding device driver registers for this IRQ.
2121 memset(&entry,0,sizeof(entry));
2123 entry.delivery_mode = INT_DELIVERY_MODE;
2124 entry.dest_mode = INT_DEST_MODE;
2125 entry.dest = cpu_mask_to_apicid(mask);
2126 entry.trigger = triggering;
2127 entry.polarity = polarity;
2128 entry.mask = 1; /* Disabled (masked) */
2129 entry.vector = vector & 0xff;
2131 apic_printk(APIC_VERBOSE,KERN_DEBUG "IOAPIC[%d]: Set PCI routing entry (%d-%d -> 0x%x -> "
2132 "IRQ %d Mode:%i Active:%i)\n", ioapic,
2133 mp_ioapics[ioapic].mpc_apicid, pin, entry.vector, irq,
2134 triggering, polarity);
2136 ioapic_register_intr(irq, entry.vector, triggering);
2138 if (!ioapic && (irq < 16))
2139 disable_8259A_irq(irq);
2141 ioapic_write_entry(ioapic, pin, entry);
2143 spin_lock_irqsave(&ioapic_lock, flags);
2144 irq_desc[irq].affinity = TARGET_CPUS;
2145 spin_unlock_irqrestore(&ioapic_lock, flags);
2150 #endif /* CONFIG_ACPI */
2154 * This function currently is only a helper for the i386 smp boot process where
2155 * we need to reprogram the ioredtbls to cater for the cpus which have come online
2156 * so mask in all cases should simply be TARGET_CPUS
2159 void __init setup_ioapic_dest(void)
2161 int pin, ioapic, irq, irq_entry;
2163 if (skip_ioapic_setup == 1)
2166 for (ioapic = 0; ioapic < nr_ioapics; ioapic++) {
2167 for (pin = 0; pin < nr_ioapic_registers[ioapic]; pin++) {
2168 irq_entry = find_irq_entry(ioapic, pin, mp_INT);
2169 if (irq_entry == -1)
2171 irq = pin_2_irq(irq_entry, ioapic, pin);
2173 /* setup_IO_APIC_irqs could fail to get vector for some device
2174 * when you have too many devices, because at that time only boot
2177 if(!irq_vector[irq])
2178 setup_IO_APIC_irq(ioapic, pin, irq_entry, irq);
2180 set_ioapic_affinity_irq(irq, TARGET_CPUS);