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];
91 #define __DO_ACTION(R, ACTION, FINAL) \
95 struct irq_pin_list *entry = irq_2_pin + irq; \
97 BUG_ON(irq >= NR_IRQS); \
103 reg = io_apic_read(entry->apic, 0x10 + R + pin*2); \
105 io_apic_modify(entry->apic, reg); \
108 entry = irq_2_pin + entry->next; \
114 struct { u32 w1, w2; };
115 struct IO_APIC_route_entry entry;
118 static struct IO_APIC_route_entry ioapic_read_entry(int apic, int pin)
120 union entry_union eu;
122 spin_lock_irqsave(&ioapic_lock, flags);
123 eu.w1 = io_apic_read(apic, 0x10 + 2 * pin);
124 eu.w2 = io_apic_read(apic, 0x11 + 2 * pin);
125 spin_unlock_irqrestore(&ioapic_lock, flags);
129 static void ioapic_write_entry(int apic, int pin, struct IO_APIC_route_entry e)
132 union entry_union eu;
134 spin_lock_irqsave(&ioapic_lock, flags);
135 io_apic_write(apic, 0x10 + 2*pin, eu.w1);
136 io_apic_write(apic, 0x11 + 2*pin, eu.w2);
137 spin_unlock_irqrestore(&ioapic_lock, flags);
141 static void __target_IO_APIC_irq(unsigned int irq, unsigned int dest, u8 vector)
144 struct irq_pin_list *entry = irq_2_pin + irq;
146 BUG_ON(irq >= NR_IRQS);
153 io_apic_write(apic, 0x11 + pin*2, dest);
154 reg = io_apic_read(apic, 0x10 + pin*2);
157 io_apic_modify(apic, reg);
160 entry = irq_2_pin + entry->next;
164 static void set_ioapic_affinity_irq(unsigned int irq, cpumask_t mask)
171 cpus_and(tmp, mask, cpu_online_map);
175 cpus_and(mask, tmp, CPU_MASK_ALL);
177 vector = assign_irq_vector(irq, mask, &tmp);
181 dest = cpu_mask_to_apicid(tmp);
184 * Only the high 8 bits are valid.
186 dest = SET_APIC_LOGICAL_ID(dest);
188 spin_lock_irqsave(&ioapic_lock, flags);
189 __target_IO_APIC_irq(irq, dest, vector);
190 set_native_irq_info(irq, mask);
191 spin_unlock_irqrestore(&ioapic_lock, flags);
196 * The common case is 1:1 IRQ<->pin mappings. Sometimes there are
197 * shared ISA-space IRQs, so we have to support them. We are super
198 * fast in the common case, and fast for shared ISA-space IRQs.
200 static void add_pin_to_irq(unsigned int irq, int apic, int pin)
202 static int first_free_entry = NR_IRQS;
203 struct irq_pin_list *entry = irq_2_pin + irq;
205 BUG_ON(irq >= NR_IRQS);
207 entry = irq_2_pin + entry->next;
209 if (entry->pin != -1) {
210 entry->next = first_free_entry;
211 entry = irq_2_pin + entry->next;
212 if (++first_free_entry >= PIN_MAP_SIZE)
213 panic("io_apic.c: ran out of irq_2_pin entries!");
220 #define DO_ACTION(name,R,ACTION, FINAL) \
222 static void name##_IO_APIC_irq (unsigned int irq) \
223 __DO_ACTION(R, ACTION, FINAL)
225 DO_ACTION( __mask, 0, |= 0x00010000, io_apic_sync(entry->apic) )
227 DO_ACTION( __unmask, 0, &= 0xfffeffff, )
230 static void mask_IO_APIC_irq (unsigned int irq)
234 spin_lock_irqsave(&ioapic_lock, flags);
235 __mask_IO_APIC_irq(irq);
236 spin_unlock_irqrestore(&ioapic_lock, flags);
239 static void unmask_IO_APIC_irq (unsigned int irq)
243 spin_lock_irqsave(&ioapic_lock, flags);
244 __unmask_IO_APIC_irq(irq);
245 spin_unlock_irqrestore(&ioapic_lock, flags);
248 static void clear_IO_APIC_pin(unsigned int apic, unsigned int pin)
250 struct IO_APIC_route_entry entry;
252 /* Check delivery_mode to be sure we're not clearing an SMI pin */
253 entry = ioapic_read_entry(apic, pin);
254 if (entry.delivery_mode == dest_SMI)
257 * Disable it in the IO-APIC irq-routing table:
259 memset(&entry, 0, sizeof(entry));
261 ioapic_write_entry(apic, pin, entry);
264 static void clear_IO_APIC (void)
268 for (apic = 0; apic < nr_ioapics; apic++)
269 for (pin = 0; pin < nr_ioapic_registers[apic]; pin++)
270 clear_IO_APIC_pin(apic, pin);
273 int skip_ioapic_setup;
276 /* dummy parsing: see setup.c */
278 static int __init disable_ioapic_setup(char *str)
280 skip_ioapic_setup = 1;
283 early_param("noapic", disable_ioapic_setup);
285 /* Actually the next is obsolete, but keep it for paranoid reasons -AK */
286 static int __init disable_timer_pin_setup(char *arg)
288 disable_timer_pin_1 = 1;
291 __setup("disable_timer_pin_1", disable_timer_pin_setup);
293 static int __init setup_disable_8254_timer(char *s)
295 timer_over_8254 = -1;
298 static int __init setup_enable_8254_timer(char *s)
304 __setup("disable_8254_timer", setup_disable_8254_timer);
305 __setup("enable_8254_timer", setup_enable_8254_timer);
309 * Find the IRQ entry number of a certain pin.
311 static int find_irq_entry(int apic, int pin, int type)
315 for (i = 0; i < mp_irq_entries; i++)
316 if (mp_irqs[i].mpc_irqtype == type &&
317 (mp_irqs[i].mpc_dstapic == mp_ioapics[apic].mpc_apicid ||
318 mp_irqs[i].mpc_dstapic == MP_APIC_ALL) &&
319 mp_irqs[i].mpc_dstirq == pin)
326 * Find the pin to which IRQ[irq] (ISA) is connected
328 static int __init find_isa_irq_pin(int irq, int type)
332 for (i = 0; i < mp_irq_entries; i++) {
333 int lbus = mp_irqs[i].mpc_srcbus;
335 if (test_bit(lbus, mp_bus_not_pci) &&
336 (mp_irqs[i].mpc_irqtype == type) &&
337 (mp_irqs[i].mpc_srcbusirq == irq))
339 return mp_irqs[i].mpc_dstirq;
344 static int __init find_isa_irq_apic(int irq, int type)
348 for (i = 0; i < mp_irq_entries; i++) {
349 int lbus = mp_irqs[i].mpc_srcbus;
351 if (test_bit(lbus, mp_bus_not_pci) &&
352 (mp_irqs[i].mpc_irqtype == type) &&
353 (mp_irqs[i].mpc_srcbusirq == irq))
356 if (i < mp_irq_entries) {
358 for(apic = 0; apic < nr_ioapics; apic++) {
359 if (mp_ioapics[apic].mpc_apicid == mp_irqs[i].mpc_dstapic)
368 * Find a specific PCI IRQ entry.
369 * Not an __init, possibly needed by modules
371 static int pin_2_irq(int idx, int apic, int pin);
373 int IO_APIC_get_PCI_irq_vector(int bus, int slot, int pin)
375 int apic, i, best_guess = -1;
377 apic_printk(APIC_DEBUG, "querying PCI -> IRQ mapping bus:%d, slot:%d, pin:%d.\n",
379 if (mp_bus_id_to_pci_bus[bus] == -1) {
380 apic_printk(APIC_VERBOSE, "PCI BIOS passed nonexistent PCI bus %d!\n", bus);
383 for (i = 0; i < mp_irq_entries; i++) {
384 int lbus = mp_irqs[i].mpc_srcbus;
386 for (apic = 0; apic < nr_ioapics; apic++)
387 if (mp_ioapics[apic].mpc_apicid == mp_irqs[i].mpc_dstapic ||
388 mp_irqs[i].mpc_dstapic == MP_APIC_ALL)
391 if (!test_bit(lbus, mp_bus_not_pci) &&
392 !mp_irqs[i].mpc_irqtype &&
394 (slot == ((mp_irqs[i].mpc_srcbusirq >> 2) & 0x1f))) {
395 int irq = pin_2_irq(i,apic,mp_irqs[i].mpc_dstirq);
397 if (!(apic || IO_APIC_IRQ(irq)))
400 if (pin == (mp_irqs[i].mpc_srcbusirq & 3))
403 * Use the first all-but-pin matching entry as a
404 * best-guess fuzzy result for broken mptables.
410 BUG_ON(best_guess >= NR_IRQS);
414 /* ISA interrupts are always polarity zero edge triggered,
415 * when listed as conforming in the MP table. */
417 #define default_ISA_trigger(idx) (0)
418 #define default_ISA_polarity(idx) (0)
420 /* PCI interrupts are always polarity one level triggered,
421 * when listed as conforming in the MP table. */
423 #define default_PCI_trigger(idx) (1)
424 #define default_PCI_polarity(idx) (1)
426 static int __init MPBIOS_polarity(int idx)
428 int bus = mp_irqs[idx].mpc_srcbus;
432 * Determine IRQ line polarity (high active or low active):
434 switch (mp_irqs[idx].mpc_irqflag & 3)
436 case 0: /* conforms, ie. bus-type dependent polarity */
437 if (test_bit(bus, mp_bus_not_pci))
438 polarity = default_ISA_polarity(idx);
440 polarity = default_PCI_polarity(idx);
442 case 1: /* high active */
447 case 2: /* reserved */
449 printk(KERN_WARNING "broken BIOS!!\n");
453 case 3: /* low active */
458 default: /* invalid */
460 printk(KERN_WARNING "broken BIOS!!\n");
468 static int MPBIOS_trigger(int idx)
470 int bus = mp_irqs[idx].mpc_srcbus;
474 * Determine IRQ trigger mode (edge or level sensitive):
476 switch ((mp_irqs[idx].mpc_irqflag>>2) & 3)
478 case 0: /* conforms, ie. bus-type dependent */
479 if (test_bit(bus, mp_bus_not_pci))
480 trigger = default_ISA_trigger(idx);
482 trigger = default_PCI_trigger(idx);
489 case 2: /* reserved */
491 printk(KERN_WARNING "broken BIOS!!\n");
500 default: /* invalid */
502 printk(KERN_WARNING "broken BIOS!!\n");
510 static inline int irq_polarity(int idx)
512 return MPBIOS_polarity(idx);
515 static inline int irq_trigger(int idx)
517 return MPBIOS_trigger(idx);
520 static int pin_2_irq(int idx, int apic, int pin)
523 int bus = mp_irqs[idx].mpc_srcbus;
526 * Debugging check, we are in big trouble if this message pops up!
528 if (mp_irqs[idx].mpc_dstirq != pin)
529 printk(KERN_ERR "broken BIOS or MPTABLE parser, ayiee!!\n");
531 if (test_bit(bus, mp_bus_not_pci)) {
532 irq = mp_irqs[idx].mpc_srcbusirq;
535 * PCI IRQs are mapped in order
539 irq += nr_ioapic_registers[i++];
542 BUG_ON(irq >= NR_IRQS);
546 static inline int IO_APIC_irq_trigger(int irq)
550 for (apic = 0; apic < nr_ioapics; apic++) {
551 for (pin = 0; pin < nr_ioapic_registers[apic]; pin++) {
552 idx = find_irq_entry(apic,pin,mp_INT);
553 if ((idx != -1) && (irq == pin_2_irq(idx,apic,pin)))
554 return irq_trigger(idx);
558 * nonexistent IRQs are edge default
563 /* irq_vectors is indexed by the sum of all RTEs in all I/O APICs. */
564 static u8 irq_vector[NR_IRQ_VECTORS] __read_mostly = {
565 [0] = FIRST_EXTERNAL_VECTOR + 0,
566 [1] = FIRST_EXTERNAL_VECTOR + 1,
567 [2] = FIRST_EXTERNAL_VECTOR + 2,
568 [3] = FIRST_EXTERNAL_VECTOR + 3,
569 [4] = FIRST_EXTERNAL_VECTOR + 4,
570 [5] = FIRST_EXTERNAL_VECTOR + 5,
571 [6] = FIRST_EXTERNAL_VECTOR + 6,
572 [7] = FIRST_EXTERNAL_VECTOR + 7,
573 [8] = FIRST_EXTERNAL_VECTOR + 8,
574 [9] = FIRST_EXTERNAL_VECTOR + 9,
575 [10] = FIRST_EXTERNAL_VECTOR + 10,
576 [11] = FIRST_EXTERNAL_VECTOR + 11,
577 [12] = FIRST_EXTERNAL_VECTOR + 12,
578 [13] = FIRST_EXTERNAL_VECTOR + 13,
579 [14] = FIRST_EXTERNAL_VECTOR + 14,
580 [15] = FIRST_EXTERNAL_VECTOR + 15,
583 static cpumask_t irq_domain[NR_IRQ_VECTORS] __read_mostly = {
602 static int __assign_irq_vector(int irq, cpumask_t mask, cpumask_t *result)
605 * NOTE! The local APIC isn't very good at handling
606 * multiple interrupts at the same interrupt level.
607 * As the interrupt level is determined by taking the
608 * vector number and shifting that right by 4, we
609 * want to spread these out a bit so that they don't
610 * all fall in the same interrupt level.
612 * Also, we've got to be careful not to trash gate
613 * 0x80, because int 0x80 is hm, kind of importantish. ;)
615 static int current_vector = FIRST_DEVICE_VECTOR, current_offset = 0;
619 BUG_ON((unsigned)irq >= NR_IRQ_VECTORS);
621 /* Only try and allocate irqs on cpus that are present */
622 cpus_and(mask, mask, cpu_online_map);
624 if (irq_vector[irq] > 0)
625 old_vector = irq_vector[irq];
626 if (old_vector > 0) {
627 cpus_and(*result, irq_domain[irq], mask);
628 if (!cpus_empty(*result))
632 for_each_cpu_mask(cpu, mask) {
633 cpumask_t domain, new_mask;
637 domain = vector_allocation_domain(cpu);
638 cpus_and(new_mask, domain, cpu_online_map);
640 vector = current_vector;
641 offset = current_offset;
644 if (vector >= FIRST_SYSTEM_VECTOR) {
645 /* If we run out of vectors on large boxen, must share them. */
646 offset = (offset + 1) % 8;
647 vector = FIRST_DEVICE_VECTOR + offset;
649 if (unlikely(current_vector == vector))
651 if (vector == IA32_SYSCALL_VECTOR)
653 for_each_cpu_mask(new_cpu, new_mask)
654 if (per_cpu(vector_irq, new_cpu)[vector] != -1)
657 current_vector = vector;
658 current_offset = offset;
659 if (old_vector >= 0) {
662 cpus_and(old_mask, irq_domain[irq], cpu_online_map);
663 for_each_cpu_mask(old_cpu, old_mask)
664 per_cpu(vector_irq, old_cpu)[old_vector] = -1;
666 for_each_cpu_mask(new_cpu, new_mask)
667 per_cpu(vector_irq, new_cpu)[vector] = irq;
668 irq_vector[irq] = vector;
669 irq_domain[irq] = domain;
670 cpus_and(*result, domain, mask);
676 static int assign_irq_vector(int irq, cpumask_t mask, cpumask_t *result)
681 spin_lock_irqsave(&vector_lock, flags);
682 vector = __assign_irq_vector(irq, mask, result);
683 spin_unlock_irqrestore(&vector_lock, flags);
687 void __setup_vector_irq(int cpu)
689 /* Initialize vector_irq on a new cpu */
690 /* This function must be called with vector_lock held */
695 /* Mark the inuse vectors */
696 for (irq = 0; irq < NR_IRQ_VECTORS; ++irq) {
697 if (!cpu_isset(cpu, irq_domain[irq]))
699 vector = irq_vector[irq];
700 per_cpu(vector_irq, cpu)[vector] = irq;
702 /* Mark the free vectors */
703 for (vector = 0; vector < NR_VECTORS; ++vector) {
704 irq = per_cpu(vector_irq, cpu)[vector];
707 if (!cpu_isset(cpu, irq_domain[irq]))
708 per_cpu(vector_irq, cpu)[vector] = -1;
713 extern void (*interrupt[NR_IRQS])(void);
715 static struct irq_chip ioapic_chip;
717 #define IOAPIC_AUTO -1
718 #define IOAPIC_EDGE 0
719 #define IOAPIC_LEVEL 1
721 static void ioapic_register_intr(int irq, int vector, unsigned long trigger)
723 if ((trigger == IOAPIC_AUTO && IO_APIC_irq_trigger(irq)) ||
724 trigger == IOAPIC_LEVEL)
725 set_irq_chip_and_handler_name(irq, &ioapic_chip,
726 handle_fasteoi_irq, "fasteoi");
728 set_irq_chip_and_handler_name(irq, &ioapic_chip,
729 handle_edge_irq, "edge");
732 static void __init setup_IO_APIC_irqs(void)
734 struct IO_APIC_route_entry entry;
735 int apic, pin, idx, irq, first_notcon = 1, vector;
738 apic_printk(APIC_VERBOSE, KERN_DEBUG "init IO_APIC IRQs\n");
740 for (apic = 0; apic < nr_ioapics; apic++) {
741 for (pin = 0; pin < nr_ioapic_registers[apic]; pin++) {
744 * add it to the IO-APIC irq-routing table:
746 memset(&entry,0,sizeof(entry));
748 entry.delivery_mode = INT_DELIVERY_MODE;
749 entry.dest_mode = INT_DEST_MODE;
750 entry.mask = 0; /* enable IRQ */
751 entry.dest.logical.logical_dest = cpu_mask_to_apicid(TARGET_CPUS);
753 idx = find_irq_entry(apic,pin,mp_INT);
756 apic_printk(APIC_VERBOSE, KERN_DEBUG " IO-APIC (apicid-pin) %d-%d", mp_ioapics[apic].mpc_apicid, pin);
759 apic_printk(APIC_VERBOSE, ", %d-%d", mp_ioapics[apic].mpc_apicid, pin);
763 entry.trigger = irq_trigger(idx);
764 entry.polarity = irq_polarity(idx);
766 if (irq_trigger(idx)) {
769 entry.dest.logical.logical_dest = cpu_mask_to_apicid(TARGET_CPUS);
772 irq = pin_2_irq(idx, apic, pin);
773 add_pin_to_irq(irq, apic, pin);
775 if (!apic && !IO_APIC_IRQ(irq))
778 if (IO_APIC_IRQ(irq)) {
780 vector = assign_irq_vector(irq, TARGET_CPUS, &mask);
784 entry.dest.logical.logical_dest = cpu_mask_to_apicid(mask);
785 entry.vector = vector;
787 ioapic_register_intr(irq, vector, IOAPIC_AUTO);
788 if (!apic && (irq < 16))
789 disable_8259A_irq(irq);
791 ioapic_write_entry(apic, pin, entry);
793 spin_lock_irqsave(&ioapic_lock, flags);
794 set_native_irq_info(irq, TARGET_CPUS);
795 spin_unlock_irqrestore(&ioapic_lock, flags);
800 apic_printk(APIC_VERBOSE," not connected.\n");
804 * Set up the 8259A-master output pin as broadcast to all
807 static void __init setup_ExtINT_IRQ0_pin(unsigned int apic, unsigned int pin, int vector)
809 struct IO_APIC_route_entry entry;
812 memset(&entry,0,sizeof(entry));
814 disable_8259A_irq(0);
817 apic_write(APIC_LVT0, APIC_LVT_MASKED | APIC_DM_EXTINT);
820 * We use logical delivery to get the timer IRQ
823 entry.dest_mode = INT_DEST_MODE;
824 entry.mask = 0; /* unmask IRQ now */
825 entry.dest.logical.logical_dest = cpu_mask_to_apicid(TARGET_CPUS);
826 entry.delivery_mode = INT_DELIVERY_MODE;
829 entry.vector = vector;
832 * The timer IRQ doesn't have to know that behind the
833 * scene we have a 8259A-master in AEOI mode ...
835 set_irq_chip_and_handler_name(0, &ioapic_chip, handle_edge_irq, "edge");
838 * Add it to the IO-APIC irq-routing table:
840 spin_lock_irqsave(&ioapic_lock, flags);
841 io_apic_write(apic, 0x11+2*pin, *(((int *)&entry)+1));
842 io_apic_write(apic, 0x10+2*pin, *(((int *)&entry)+0));
843 spin_unlock_irqrestore(&ioapic_lock, flags);
848 void __init UNEXPECTED_IO_APIC(void)
852 void __apicdebuginit print_IO_APIC(void)
855 union IO_APIC_reg_00 reg_00;
856 union IO_APIC_reg_01 reg_01;
857 union IO_APIC_reg_02 reg_02;
860 if (apic_verbosity == APIC_QUIET)
863 printk(KERN_DEBUG "number of MP IRQ sources: %d.\n", mp_irq_entries);
864 for (i = 0; i < nr_ioapics; i++)
865 printk(KERN_DEBUG "number of IO-APIC #%d registers: %d.\n",
866 mp_ioapics[i].mpc_apicid, nr_ioapic_registers[i]);
869 * We are a bit conservative about what we expect. We have to
870 * know about every hardware change ASAP.
872 printk(KERN_INFO "testing the IO APIC.......................\n");
874 for (apic = 0; apic < nr_ioapics; apic++) {
876 spin_lock_irqsave(&ioapic_lock, flags);
877 reg_00.raw = io_apic_read(apic, 0);
878 reg_01.raw = io_apic_read(apic, 1);
879 if (reg_01.bits.version >= 0x10)
880 reg_02.raw = io_apic_read(apic, 2);
881 spin_unlock_irqrestore(&ioapic_lock, flags);
884 printk(KERN_DEBUG "IO APIC #%d......\n", mp_ioapics[apic].mpc_apicid);
885 printk(KERN_DEBUG ".... register #00: %08X\n", reg_00.raw);
886 printk(KERN_DEBUG "....... : physical APIC id: %02X\n", reg_00.bits.ID);
887 if (reg_00.bits.__reserved_1 || reg_00.bits.__reserved_2)
888 UNEXPECTED_IO_APIC();
890 printk(KERN_DEBUG ".... register #01: %08X\n", *(int *)®_01);
891 printk(KERN_DEBUG "....... : max redirection entries: %04X\n", reg_01.bits.entries);
892 if ( (reg_01.bits.entries != 0x0f) && /* older (Neptune) boards */
893 (reg_01.bits.entries != 0x17) && /* typical ISA+PCI boards */
894 (reg_01.bits.entries != 0x1b) && /* Compaq Proliant boards */
895 (reg_01.bits.entries != 0x1f) && /* dual Xeon boards */
896 (reg_01.bits.entries != 0x22) && /* bigger Xeon boards */
897 (reg_01.bits.entries != 0x2E) &&
898 (reg_01.bits.entries != 0x3F) &&
899 (reg_01.bits.entries != 0x03)
901 UNEXPECTED_IO_APIC();
903 printk(KERN_DEBUG "....... : PRQ implemented: %X\n", reg_01.bits.PRQ);
904 printk(KERN_DEBUG "....... : IO APIC version: %04X\n", reg_01.bits.version);
905 if ( (reg_01.bits.version != 0x01) && /* 82489DX IO-APICs */
906 (reg_01.bits.version != 0x02) && /* 82801BA IO-APICs (ICH2) */
907 (reg_01.bits.version != 0x10) && /* oldest IO-APICs */
908 (reg_01.bits.version != 0x11) && /* Pentium/Pro IO-APICs */
909 (reg_01.bits.version != 0x13) && /* Xeon IO-APICs */
910 (reg_01.bits.version != 0x20) /* Intel P64H (82806 AA) */
912 UNEXPECTED_IO_APIC();
913 if (reg_01.bits.__reserved_1 || reg_01.bits.__reserved_2)
914 UNEXPECTED_IO_APIC();
916 if (reg_01.bits.version >= 0x10) {
917 printk(KERN_DEBUG ".... register #02: %08X\n", reg_02.raw);
918 printk(KERN_DEBUG "....... : arbitration: %02X\n", reg_02.bits.arbitration);
919 if (reg_02.bits.__reserved_1 || reg_02.bits.__reserved_2)
920 UNEXPECTED_IO_APIC();
923 printk(KERN_DEBUG ".... IRQ redirection table:\n");
925 printk(KERN_DEBUG " NR Log Phy Mask Trig IRR Pol"
926 " Stat Dest Deli Vect: \n");
928 for (i = 0; i <= reg_01.bits.entries; i++) {
929 struct IO_APIC_route_entry entry;
931 entry = ioapic_read_entry(apic, i);
933 printk(KERN_DEBUG " %02x %03X %02X ",
935 entry.dest.logical.logical_dest,
936 entry.dest.physical.physical_dest
939 printk("%1d %1d %1d %1d %1d %1d %1d %02X\n",
944 entry.delivery_status,
951 printk(KERN_DEBUG "IRQ to pin mappings:\n");
952 for (i = 0; i < NR_IRQS; i++) {
953 struct irq_pin_list *entry = irq_2_pin + i;
956 printk(KERN_DEBUG "IRQ%d ", i);
958 printk("-> %d:%d", entry->apic, entry->pin);
961 entry = irq_2_pin + entry->next;
966 printk(KERN_INFO ".................................... done.\n");
973 static __apicdebuginit void print_APIC_bitfield (int base)
978 if (apic_verbosity == APIC_QUIET)
981 printk(KERN_DEBUG "0123456789abcdef0123456789abcdef\n" KERN_DEBUG);
982 for (i = 0; i < 8; i++) {
983 v = apic_read(base + i*0x10);
984 for (j = 0; j < 32; j++) {
994 void __apicdebuginit print_local_APIC(void * dummy)
996 unsigned int v, ver, maxlvt;
998 if (apic_verbosity == APIC_QUIET)
1001 printk("\n" KERN_DEBUG "printing local APIC contents on CPU#%d/%d:\n",
1002 smp_processor_id(), hard_smp_processor_id());
1003 v = apic_read(APIC_ID);
1004 printk(KERN_INFO "... APIC ID: %08x (%01x)\n", v, GET_APIC_ID(v));
1005 v = apic_read(APIC_LVR);
1006 printk(KERN_INFO "... APIC VERSION: %08x\n", v);
1007 ver = GET_APIC_VERSION(v);
1008 maxlvt = get_maxlvt();
1010 v = apic_read(APIC_TASKPRI);
1011 printk(KERN_DEBUG "... APIC TASKPRI: %08x (%02x)\n", v, v & APIC_TPRI_MASK);
1013 v = apic_read(APIC_ARBPRI);
1014 printk(KERN_DEBUG "... APIC ARBPRI: %08x (%02x)\n", v,
1015 v & APIC_ARBPRI_MASK);
1016 v = apic_read(APIC_PROCPRI);
1017 printk(KERN_DEBUG "... APIC PROCPRI: %08x\n", v);
1019 v = apic_read(APIC_EOI);
1020 printk(KERN_DEBUG "... APIC EOI: %08x\n", v);
1021 v = apic_read(APIC_RRR);
1022 printk(KERN_DEBUG "... APIC RRR: %08x\n", v);
1023 v = apic_read(APIC_LDR);
1024 printk(KERN_DEBUG "... APIC LDR: %08x\n", v);
1025 v = apic_read(APIC_DFR);
1026 printk(KERN_DEBUG "... APIC DFR: %08x\n", v);
1027 v = apic_read(APIC_SPIV);
1028 printk(KERN_DEBUG "... APIC SPIV: %08x\n", v);
1030 printk(KERN_DEBUG "... APIC ISR field:\n");
1031 print_APIC_bitfield(APIC_ISR);
1032 printk(KERN_DEBUG "... APIC TMR field:\n");
1033 print_APIC_bitfield(APIC_TMR);
1034 printk(KERN_DEBUG "... APIC IRR field:\n");
1035 print_APIC_bitfield(APIC_IRR);
1037 v = apic_read(APIC_ESR);
1038 printk(KERN_DEBUG "... APIC ESR: %08x\n", v);
1040 v = apic_read(APIC_ICR);
1041 printk(KERN_DEBUG "... APIC ICR: %08x\n", v);
1042 v = apic_read(APIC_ICR2);
1043 printk(KERN_DEBUG "... APIC ICR2: %08x\n", v);
1045 v = apic_read(APIC_LVTT);
1046 printk(KERN_DEBUG "... APIC LVTT: %08x\n", v);
1048 if (maxlvt > 3) { /* PC is LVT#4. */
1049 v = apic_read(APIC_LVTPC);
1050 printk(KERN_DEBUG "... APIC LVTPC: %08x\n", v);
1052 v = apic_read(APIC_LVT0);
1053 printk(KERN_DEBUG "... APIC LVT0: %08x\n", v);
1054 v = apic_read(APIC_LVT1);
1055 printk(KERN_DEBUG "... APIC LVT1: %08x\n", v);
1057 if (maxlvt > 2) { /* ERR is LVT#3. */
1058 v = apic_read(APIC_LVTERR);
1059 printk(KERN_DEBUG "... APIC LVTERR: %08x\n", v);
1062 v = apic_read(APIC_TMICT);
1063 printk(KERN_DEBUG "... APIC TMICT: %08x\n", v);
1064 v = apic_read(APIC_TMCCT);
1065 printk(KERN_DEBUG "... APIC TMCCT: %08x\n", v);
1066 v = apic_read(APIC_TDCR);
1067 printk(KERN_DEBUG "... APIC TDCR: %08x\n", v);
1071 void print_all_local_APICs (void)
1073 on_each_cpu(print_local_APIC, NULL, 1, 1);
1076 void __apicdebuginit print_PIC(void)
1079 unsigned long flags;
1081 if (apic_verbosity == APIC_QUIET)
1084 printk(KERN_DEBUG "\nprinting PIC contents\n");
1086 spin_lock_irqsave(&i8259A_lock, flags);
1088 v = inb(0xa1) << 8 | inb(0x21);
1089 printk(KERN_DEBUG "... PIC IMR: %04x\n", v);
1091 v = inb(0xa0) << 8 | inb(0x20);
1092 printk(KERN_DEBUG "... PIC IRR: %04x\n", v);
1096 v = inb(0xa0) << 8 | inb(0x20);
1100 spin_unlock_irqrestore(&i8259A_lock, flags);
1102 printk(KERN_DEBUG "... PIC ISR: %04x\n", v);
1104 v = inb(0x4d1) << 8 | inb(0x4d0);
1105 printk(KERN_DEBUG "... PIC ELCR: %04x\n", v);
1110 static void __init enable_IO_APIC(void)
1112 union IO_APIC_reg_01 reg_01;
1113 int i8259_apic, i8259_pin;
1115 unsigned long flags;
1117 for (i = 0; i < PIN_MAP_SIZE; i++) {
1118 irq_2_pin[i].pin = -1;
1119 irq_2_pin[i].next = 0;
1123 * The number of IO-APIC IRQ registers (== #pins):
1125 for (apic = 0; apic < nr_ioapics; apic++) {
1126 spin_lock_irqsave(&ioapic_lock, flags);
1127 reg_01.raw = io_apic_read(apic, 1);
1128 spin_unlock_irqrestore(&ioapic_lock, flags);
1129 nr_ioapic_registers[apic] = reg_01.bits.entries+1;
1131 for(apic = 0; apic < nr_ioapics; apic++) {
1133 /* See if any of the pins is in ExtINT mode */
1134 for (pin = 0; pin < nr_ioapic_registers[apic]; pin++) {
1135 struct IO_APIC_route_entry entry;
1136 entry = ioapic_read_entry(apic, pin);
1138 /* If the interrupt line is enabled and in ExtInt mode
1139 * I have found the pin where the i8259 is connected.
1141 if ((entry.mask == 0) && (entry.delivery_mode == dest_ExtINT)) {
1142 ioapic_i8259.apic = apic;
1143 ioapic_i8259.pin = pin;
1149 /* Look to see what if the MP table has reported the ExtINT */
1150 i8259_pin = find_isa_irq_pin(0, mp_ExtINT);
1151 i8259_apic = find_isa_irq_apic(0, mp_ExtINT);
1152 /* Trust the MP table if nothing is setup in the hardware */
1153 if ((ioapic_i8259.pin == -1) && (i8259_pin >= 0)) {
1154 printk(KERN_WARNING "ExtINT not setup in hardware but reported by MP table\n");
1155 ioapic_i8259.pin = i8259_pin;
1156 ioapic_i8259.apic = i8259_apic;
1158 /* Complain if the MP table and the hardware disagree */
1159 if (((ioapic_i8259.apic != i8259_apic) || (ioapic_i8259.pin != i8259_pin)) &&
1160 (i8259_pin >= 0) && (ioapic_i8259.pin >= 0))
1162 printk(KERN_WARNING "ExtINT in hardware and MP table differ\n");
1166 * Do not trust the IO-APIC being empty at bootup
1172 * Not an __init, needed by the reboot code
1174 void disable_IO_APIC(void)
1177 * Clear the IO-APIC before rebooting:
1182 * If the i8259 is routed through an IOAPIC
1183 * Put that IOAPIC in virtual wire mode
1184 * so legacy interrupts can be delivered.
1186 if (ioapic_i8259.pin != -1) {
1187 struct IO_APIC_route_entry entry;
1189 memset(&entry, 0, sizeof(entry));
1190 entry.mask = 0; /* Enabled */
1191 entry.trigger = 0; /* Edge */
1193 entry.polarity = 0; /* High */
1194 entry.delivery_status = 0;
1195 entry.dest_mode = 0; /* Physical */
1196 entry.delivery_mode = dest_ExtINT; /* ExtInt */
1198 entry.dest.physical.physical_dest =
1199 GET_APIC_ID(apic_read(APIC_ID));
1202 * Add it to the IO-APIC irq-routing table:
1204 ioapic_write_entry(ioapic_i8259.apic, ioapic_i8259.pin, entry);
1207 disconnect_bsp_APIC(ioapic_i8259.pin != -1);
1211 * There is a nasty bug in some older SMP boards, their mptable lies
1212 * about the timer IRQ. We do the following to work around the situation:
1214 * - timer IRQ defaults to IO-APIC IRQ
1215 * - if this function detects that timer IRQs are defunct, then we fall
1216 * back to ISA timer IRQs
1218 static int __init timer_irq_works(void)
1220 unsigned long t1 = jiffies;
1223 /* Let ten ticks pass... */
1224 mdelay((10 * 1000) / HZ);
1227 * Expect a few ticks at least, to be sure some possible
1228 * glue logic does not lock up after one or two first
1229 * ticks in a non-ExtINT mode. Also the local APIC
1230 * might have cached one ExtINT interrupt. Finally, at
1231 * least one tick may be lost due to delays.
1235 if (jiffies - t1 > 4)
1241 * In the SMP+IOAPIC case it might happen that there are an unspecified
1242 * number of pending IRQ events unhandled. These cases are very rare,
1243 * so we 'resend' these IRQs via IPIs, to the same CPU. It's much
1244 * better to do it this way as thus we do not have to be aware of
1245 * 'pending' interrupts in the IRQ path, except at this point.
1248 * Edge triggered needs to resend any interrupt
1249 * that was delayed but this is now handled in the device
1254 * Starting up a edge-triggered IO-APIC interrupt is
1255 * nasty - we need to make sure that we get the edge.
1256 * If it is already asserted for some reason, we need
1257 * return 1 to indicate that is was pending.
1259 * This is not complete - we should be able to fake
1260 * an edge even if it isn't on the 8259A...
1263 static unsigned int startup_ioapic_irq(unsigned int irq)
1265 int was_pending = 0;
1266 unsigned long flags;
1268 spin_lock_irqsave(&ioapic_lock, flags);
1270 disable_8259A_irq(irq);
1271 if (i8259A_irq_pending(irq))
1274 __unmask_IO_APIC_irq(irq);
1275 spin_unlock_irqrestore(&ioapic_lock, flags);
1280 static int ioapic_retrigger_irq(unsigned int irq)
1284 unsigned long flags;
1286 spin_lock_irqsave(&vector_lock, flags);
1287 vector = irq_vector[irq];
1289 cpu_set(first_cpu(irq_domain[irq]), mask);
1291 send_IPI_mask(mask, vector);
1292 spin_unlock_irqrestore(&vector_lock, flags);
1298 * Level and edge triggered IO-APIC interrupts need different handling,
1299 * so we use two separate IRQ descriptors. Edge triggered IRQs can be
1300 * handled with the level-triggered descriptor, but that one has slightly
1301 * more overhead. Level-triggered interrupts cannot be handled with the
1302 * edge-triggered handler, without risking IRQ storms and other ugly
1306 static void ack_apic_edge(unsigned int irq)
1308 move_native_irq(irq);
1312 static void ack_apic_level(unsigned int irq)
1314 int do_unmask_irq = 0;
1316 #if defined(CONFIG_GENERIC_PENDING_IRQ) || defined(CONFIG_IRQBALANCE)
1317 /* If we are moving the irq we need to mask it */
1318 if (unlikely(irq_desc[irq].status & IRQ_MOVE_PENDING)) {
1320 mask_IO_APIC_irq(irq);
1325 * We must acknowledge the irq before we move it or the acknowledge will
1326 * not propogate properly.
1330 /* Now we can move and renable the irq */
1331 move_masked_irq(irq);
1332 if (unlikely(do_unmask_irq))
1333 unmask_IO_APIC_irq(irq);
1336 static struct irq_chip ioapic_chip __read_mostly = {
1338 .startup = startup_ioapic_irq,
1339 .mask = mask_IO_APIC_irq,
1340 .unmask = unmask_IO_APIC_irq,
1341 .ack = ack_apic_edge,
1342 .eoi = ack_apic_level,
1344 .set_affinity = set_ioapic_affinity_irq,
1346 .retrigger = ioapic_retrigger_irq,
1349 static inline void init_IO_APIC_traps(void)
1354 * NOTE! The local APIC isn't very good at handling
1355 * multiple interrupts at the same interrupt level.
1356 * As the interrupt level is determined by taking the
1357 * vector number and shifting that right by 4, we
1358 * want to spread these out a bit so that they don't
1359 * all fall in the same interrupt level.
1361 * Also, we've got to be careful not to trash gate
1362 * 0x80, because int 0x80 is hm, kind of importantish. ;)
1364 for (irq = 0; irq < NR_IRQS ; irq++) {
1366 if (IO_APIC_IRQ(tmp) && !irq_vector[tmp]) {
1368 * Hmm.. We don't have an entry for this,
1369 * so default to an old-fashioned 8259
1370 * interrupt if we can..
1373 make_8259A_irq(irq);
1375 /* Strange. Oh, well.. */
1376 irq_desc[irq].chip = &no_irq_chip;
1381 static void enable_lapic_irq (unsigned int irq)
1385 v = apic_read(APIC_LVT0);
1386 apic_write(APIC_LVT0, v & ~APIC_LVT_MASKED);
1389 static void disable_lapic_irq (unsigned int irq)
1393 v = apic_read(APIC_LVT0);
1394 apic_write(APIC_LVT0, v | APIC_LVT_MASKED);
1397 static void ack_lapic_irq (unsigned int irq)
1402 static void end_lapic_irq (unsigned int i) { /* nothing */ }
1404 static struct hw_interrupt_type lapic_irq_type __read_mostly = {
1405 .typename = "local-APIC-edge",
1406 .startup = NULL, /* startup_irq() not used for IRQ0 */
1407 .shutdown = NULL, /* shutdown_irq() not used for IRQ0 */
1408 .enable = enable_lapic_irq,
1409 .disable = disable_lapic_irq,
1410 .ack = ack_lapic_irq,
1411 .end = end_lapic_irq,
1414 static void setup_nmi (void)
1417 * Dirty trick to enable the NMI watchdog ...
1418 * We put the 8259A master into AEOI mode and
1419 * unmask on all local APICs LVT0 as NMI.
1421 * The idea to use the 8259A in AEOI mode ('8259A Virtual Wire')
1422 * is from Maciej W. Rozycki - so we do not have to EOI from
1423 * the NMI handler or the timer interrupt.
1425 printk(KERN_INFO "activating NMI Watchdog ...");
1427 enable_NMI_through_LVT0(NULL);
1433 * This looks a bit hackish but it's about the only one way of sending
1434 * a few INTA cycles to 8259As and any associated glue logic. ICR does
1435 * not support the ExtINT mode, unfortunately. We need to send these
1436 * cycles as some i82489DX-based boards have glue logic that keeps the
1437 * 8259A interrupt line asserted until INTA. --macro
1439 static inline void unlock_ExtINT_logic(void)
1442 struct IO_APIC_route_entry entry0, entry1;
1443 unsigned char save_control, save_freq_select;
1444 unsigned long flags;
1446 pin = find_isa_irq_pin(8, mp_INT);
1447 apic = find_isa_irq_apic(8, mp_INT);
1451 spin_lock_irqsave(&ioapic_lock, flags);
1452 *(((int *)&entry0) + 1) = io_apic_read(apic, 0x11 + 2 * pin);
1453 *(((int *)&entry0) + 0) = io_apic_read(apic, 0x10 + 2 * pin);
1454 spin_unlock_irqrestore(&ioapic_lock, flags);
1455 clear_IO_APIC_pin(apic, pin);
1457 memset(&entry1, 0, sizeof(entry1));
1459 entry1.dest_mode = 0; /* physical delivery */
1460 entry1.mask = 0; /* unmask IRQ now */
1461 entry1.dest.physical.physical_dest = hard_smp_processor_id();
1462 entry1.delivery_mode = dest_ExtINT;
1463 entry1.polarity = entry0.polarity;
1467 spin_lock_irqsave(&ioapic_lock, flags);
1468 io_apic_write(apic, 0x11 + 2 * pin, *(((int *)&entry1) + 1));
1469 io_apic_write(apic, 0x10 + 2 * pin, *(((int *)&entry1) + 0));
1470 spin_unlock_irqrestore(&ioapic_lock, flags);
1472 save_control = CMOS_READ(RTC_CONTROL);
1473 save_freq_select = CMOS_READ(RTC_FREQ_SELECT);
1474 CMOS_WRITE((save_freq_select & ~RTC_RATE_SELECT) | 0x6,
1476 CMOS_WRITE(save_control | RTC_PIE, RTC_CONTROL);
1481 if ((CMOS_READ(RTC_INTR_FLAGS) & RTC_PF) == RTC_PF)
1485 CMOS_WRITE(save_control, RTC_CONTROL);
1486 CMOS_WRITE(save_freq_select, RTC_FREQ_SELECT);
1487 clear_IO_APIC_pin(apic, pin);
1489 spin_lock_irqsave(&ioapic_lock, flags);
1490 io_apic_write(apic, 0x11 + 2 * pin, *(((int *)&entry0) + 1));
1491 io_apic_write(apic, 0x10 + 2 * pin, *(((int *)&entry0) + 0));
1492 spin_unlock_irqrestore(&ioapic_lock, flags);
1496 * This code may look a bit paranoid, but it's supposed to cooperate with
1497 * a wide range of boards and BIOS bugs. Fortunately only the timer IRQ
1498 * is so screwy. Thanks to Brian Perkins for testing/hacking this beast
1499 * fanatically on his truly buggy board.
1501 * FIXME: really need to revamp this for modern platforms only.
1503 static inline void check_timer(void)
1505 int apic1, pin1, apic2, pin2;
1510 * get/set the timer IRQ vector:
1512 disable_8259A_irq(0);
1513 vector = assign_irq_vector(0, TARGET_CPUS, &mask);
1516 * Subtle, code in do_timer_interrupt() expects an AEOI
1517 * mode for the 8259A whenever interrupts are routed
1518 * through I/O APICs. Also IRQ0 has to be enabled in
1519 * the 8259A which implies the virtual wire has to be
1520 * disabled in the local APIC.
1522 apic_write(APIC_LVT0, APIC_LVT_MASKED | APIC_DM_EXTINT);
1524 if (timer_over_8254 > 0)
1525 enable_8259A_irq(0);
1527 pin1 = find_isa_irq_pin(0, mp_INT);
1528 apic1 = find_isa_irq_apic(0, mp_INT);
1529 pin2 = ioapic_i8259.pin;
1530 apic2 = ioapic_i8259.apic;
1532 apic_printk(APIC_VERBOSE,KERN_INFO "..TIMER: vector=0x%02X apic1=%d pin1=%d apic2=%d pin2=%d\n",
1533 vector, apic1, pin1, apic2, pin2);
1537 * Ok, does IRQ0 through the IOAPIC work?
1539 unmask_IO_APIC_irq(0);
1540 if (!no_timer_check && timer_irq_works()) {
1541 nmi_watchdog_default();
1542 if (nmi_watchdog == NMI_IO_APIC) {
1543 disable_8259A_irq(0);
1545 enable_8259A_irq(0);
1547 if (disable_timer_pin_1 > 0)
1548 clear_IO_APIC_pin(0, pin1);
1551 clear_IO_APIC_pin(apic1, pin1);
1552 apic_printk(APIC_QUIET,KERN_ERR "..MP-BIOS bug: 8254 timer not "
1553 "connected to IO-APIC\n");
1556 apic_printk(APIC_VERBOSE,KERN_INFO "...trying to set up timer (IRQ0) "
1557 "through the 8259A ... ");
1559 apic_printk(APIC_VERBOSE,"\n..... (found apic %d pin %d) ...",
1562 * legacy devices should be connected to IO APIC #0
1564 setup_ExtINT_IRQ0_pin(apic2, pin2, vector);
1565 if (timer_irq_works()) {
1566 apic_printk(APIC_VERBOSE," works.\n");
1567 nmi_watchdog_default();
1568 if (nmi_watchdog == NMI_IO_APIC) {
1574 * Cleanup, just in case ...
1576 clear_IO_APIC_pin(apic2, pin2);
1578 apic_printk(APIC_VERBOSE," failed.\n");
1580 if (nmi_watchdog == NMI_IO_APIC) {
1581 printk(KERN_WARNING "timer doesn't work through the IO-APIC - disabling NMI Watchdog!\n");
1585 apic_printk(APIC_VERBOSE, KERN_INFO "...trying to set up timer as Virtual Wire IRQ...");
1587 disable_8259A_irq(0);
1588 irq_desc[0].chip = &lapic_irq_type;
1589 apic_write(APIC_LVT0, APIC_DM_FIXED | vector); /* Fixed mode */
1590 enable_8259A_irq(0);
1592 if (timer_irq_works()) {
1593 apic_printk(APIC_VERBOSE," works.\n");
1596 apic_write(APIC_LVT0, APIC_LVT_MASKED | APIC_DM_FIXED | vector);
1597 apic_printk(APIC_VERBOSE," failed.\n");
1599 apic_printk(APIC_VERBOSE, KERN_INFO "...trying to set up timer as ExtINT IRQ...");
1603 apic_write(APIC_LVT0, APIC_DM_EXTINT);
1605 unlock_ExtINT_logic();
1607 if (timer_irq_works()) {
1608 apic_printk(APIC_VERBOSE," works.\n");
1611 apic_printk(APIC_VERBOSE," failed :(.\n");
1612 panic("IO-APIC + timer doesn't work! Try using the 'noapic' kernel parameter\n");
1615 static int __init notimercheck(char *s)
1620 __setup("no_timer_check", notimercheck);
1624 * IRQ's that are handled by the PIC in the MPS IOAPIC case.
1625 * - IRQ2 is the cascade IRQ, and cannot be a io-apic IRQ.
1626 * Linux doesn't really care, as it's not actually used
1627 * for any interrupt handling anyway.
1629 #define PIC_IRQS (1<<2)
1631 void __init setup_IO_APIC(void)
1636 io_apic_irqs = ~0; /* all IRQs go through IOAPIC */
1638 io_apic_irqs = ~PIC_IRQS;
1640 apic_printk(APIC_VERBOSE, "ENABLING IO-APIC IRQs\n");
1643 setup_IO_APIC_irqs();
1644 init_IO_APIC_traps();
1650 struct sysfs_ioapic_data {
1651 struct sys_device dev;
1652 struct IO_APIC_route_entry entry[0];
1654 static struct sysfs_ioapic_data * mp_ioapic_data[MAX_IO_APICS];
1656 static int ioapic_suspend(struct sys_device *dev, pm_message_t state)
1658 struct IO_APIC_route_entry *entry;
1659 struct sysfs_ioapic_data *data;
1662 data = container_of(dev, struct sysfs_ioapic_data, dev);
1663 entry = data->entry;
1664 for (i = 0; i < nr_ioapic_registers[dev->id]; i ++, entry ++ )
1665 *entry = ioapic_read_entry(dev->id, i);
1670 static int ioapic_resume(struct sys_device *dev)
1672 struct IO_APIC_route_entry *entry;
1673 struct sysfs_ioapic_data *data;
1674 unsigned long flags;
1675 union IO_APIC_reg_00 reg_00;
1678 data = container_of(dev, struct sysfs_ioapic_data, dev);
1679 entry = data->entry;
1681 spin_lock_irqsave(&ioapic_lock, flags);
1682 reg_00.raw = io_apic_read(dev->id, 0);
1683 if (reg_00.bits.ID != mp_ioapics[dev->id].mpc_apicid) {
1684 reg_00.bits.ID = mp_ioapics[dev->id].mpc_apicid;
1685 io_apic_write(dev->id, 0, reg_00.raw);
1687 spin_unlock_irqrestore(&ioapic_lock, flags);
1688 for (i = 0; i < nr_ioapic_registers[dev->id]; i++)
1689 ioapic_write_entry(dev->id, i, entry[i]);
1694 static struct sysdev_class ioapic_sysdev_class = {
1695 set_kset_name("ioapic"),
1696 .suspend = ioapic_suspend,
1697 .resume = ioapic_resume,
1700 static int __init ioapic_init_sysfs(void)
1702 struct sys_device * dev;
1703 int i, size, error = 0;
1705 error = sysdev_class_register(&ioapic_sysdev_class);
1709 for (i = 0; i < nr_ioapics; i++ ) {
1710 size = sizeof(struct sys_device) + nr_ioapic_registers[i]
1711 * sizeof(struct IO_APIC_route_entry);
1712 mp_ioapic_data[i] = kmalloc(size, GFP_KERNEL);
1713 if (!mp_ioapic_data[i]) {
1714 printk(KERN_ERR "Can't suspend/resume IOAPIC %d\n", i);
1717 memset(mp_ioapic_data[i], 0, size);
1718 dev = &mp_ioapic_data[i]->dev;
1720 dev->cls = &ioapic_sysdev_class;
1721 error = sysdev_register(dev);
1723 kfree(mp_ioapic_data[i]);
1724 mp_ioapic_data[i] = NULL;
1725 printk(KERN_ERR "Can't suspend/resume IOAPIC %d\n", i);
1733 device_initcall(ioapic_init_sysfs);
1736 * Dynamic irq allocate and deallocation
1738 int create_irq(void)
1740 /* Allocate an unused irq */
1744 unsigned long flags;
1748 spin_lock_irqsave(&vector_lock, flags);
1749 for (new = (NR_IRQS - 1); new >= 0; new--) {
1750 if (platform_legacy_irq(new))
1752 if (irq_vector[new] != 0)
1754 vector = __assign_irq_vector(new, TARGET_CPUS, &mask);
1755 if (likely(vector > 0))
1759 spin_unlock_irqrestore(&vector_lock, flags);
1762 dynamic_irq_init(irq);
1767 void destroy_irq(unsigned int irq)
1769 unsigned long flags;
1771 dynamic_irq_cleanup(irq);
1773 spin_lock_irqsave(&vector_lock, flags);
1774 irq_vector[irq] = 0;
1775 spin_unlock_irqrestore(&vector_lock, flags);
1779 * MSI mesage composition
1781 #ifdef CONFIG_PCI_MSI
1782 static int msi_compose_msg(struct pci_dev *pdev, unsigned int irq, struct msi_msg *msg)
1788 vector = assign_irq_vector(irq, TARGET_CPUS, &tmp);
1790 dest = cpu_mask_to_apicid(tmp);
1792 msg->address_hi = MSI_ADDR_BASE_HI;
1795 ((INT_DEST_MODE == 0) ?
1796 MSI_ADDR_DEST_MODE_PHYSICAL:
1797 MSI_ADDR_DEST_MODE_LOGICAL) |
1798 ((INT_DELIVERY_MODE != dest_LowestPrio) ?
1799 MSI_ADDR_REDIRECTION_CPU:
1800 MSI_ADDR_REDIRECTION_LOWPRI) |
1801 MSI_ADDR_DEST_ID(dest);
1804 MSI_DATA_TRIGGER_EDGE |
1805 MSI_DATA_LEVEL_ASSERT |
1806 ((INT_DELIVERY_MODE != dest_LowestPrio) ?
1807 MSI_DATA_DELIVERY_FIXED:
1808 MSI_DATA_DELIVERY_LOWPRI) |
1809 MSI_DATA_VECTOR(vector);
1815 static void set_msi_irq_affinity(unsigned int irq, cpumask_t mask)
1822 cpus_and(tmp, mask, cpu_online_map);
1823 if (cpus_empty(tmp))
1826 cpus_and(mask, tmp, CPU_MASK_ALL);
1828 vector = assign_irq_vector(irq, mask, &tmp);
1832 dest = cpu_mask_to_apicid(tmp);
1834 read_msi_msg(irq, &msg);
1836 msg.data &= ~MSI_DATA_VECTOR_MASK;
1837 msg.data |= MSI_DATA_VECTOR(vector);
1838 msg.address_lo &= ~MSI_ADDR_DEST_ID_MASK;
1839 msg.address_lo |= MSI_ADDR_DEST_ID(dest);
1841 write_msi_msg(irq, &msg);
1842 set_native_irq_info(irq, mask);
1844 #endif /* CONFIG_SMP */
1847 * IRQ Chip for MSI PCI/PCI-X/PCI-Express Devices,
1848 * which implement the MSI or MSI-X Capability Structure.
1850 static struct irq_chip msi_chip = {
1852 .unmask = unmask_msi_irq,
1853 .mask = mask_msi_irq,
1854 .ack = ack_apic_edge,
1856 .set_affinity = set_msi_irq_affinity,
1858 .retrigger = ioapic_retrigger_irq,
1861 int arch_setup_msi_irq(unsigned int irq, struct pci_dev *dev)
1865 ret = msi_compose_msg(dev, irq, &msg);
1869 write_msi_msg(irq, &msg);
1871 set_irq_chip_and_handler_name(irq, &msi_chip, handle_edge_irq, "edge");
1876 void arch_teardown_msi_irq(unsigned int irq)
1881 #endif /* CONFIG_PCI_MSI */
1884 * Hypertransport interrupt support
1886 #ifdef CONFIG_HT_IRQ
1890 static void target_ht_irq(unsigned int irq, unsigned int dest, u8 vector)
1893 low = read_ht_irq_low(irq);
1894 high = read_ht_irq_high(irq);
1896 low &= ~(HT_IRQ_LOW_VECTOR_MASK | HT_IRQ_LOW_DEST_ID_MASK);
1897 high &= ~(HT_IRQ_HIGH_DEST_ID_MASK);
1899 low |= HT_IRQ_LOW_VECTOR(vector) | HT_IRQ_LOW_DEST_ID(dest);
1900 high |= HT_IRQ_HIGH_DEST_ID(dest);
1902 write_ht_irq_low(irq, low);
1903 write_ht_irq_high(irq, high);
1906 static void set_ht_irq_affinity(unsigned int irq, cpumask_t mask)
1912 cpus_and(tmp, mask, cpu_online_map);
1913 if (cpus_empty(tmp))
1916 cpus_and(mask, tmp, CPU_MASK_ALL);
1918 vector = assign_irq_vector(irq, mask, &tmp);
1922 dest = cpu_mask_to_apicid(tmp);
1924 target_ht_irq(irq, dest, vector & 0xff);
1925 set_native_irq_info(irq, mask);
1929 static struct irq_chip ht_irq_chip = {
1931 .mask = mask_ht_irq,
1932 .unmask = unmask_ht_irq,
1933 .ack = ack_apic_edge,
1935 .set_affinity = set_ht_irq_affinity,
1937 .retrigger = ioapic_retrigger_irq,
1940 int arch_setup_ht_irq(unsigned int irq, struct pci_dev *dev)
1945 vector = assign_irq_vector(irq, TARGET_CPUS, &tmp);
1950 dest = cpu_mask_to_apicid(tmp);
1952 high = HT_IRQ_HIGH_DEST_ID(dest);
1954 low = HT_IRQ_LOW_BASE |
1955 HT_IRQ_LOW_DEST_ID(dest) |
1956 HT_IRQ_LOW_VECTOR(vector) |
1957 ((INT_DEST_MODE == 0) ?
1958 HT_IRQ_LOW_DM_PHYSICAL :
1959 HT_IRQ_LOW_DM_LOGICAL) |
1960 HT_IRQ_LOW_RQEOI_EDGE |
1961 ((INT_DELIVERY_MODE != dest_LowestPrio) ?
1962 HT_IRQ_LOW_MT_FIXED :
1963 HT_IRQ_LOW_MT_ARBITRATED);
1965 write_ht_irq_low(irq, low);
1966 write_ht_irq_high(irq, high);
1968 set_irq_chip_and_handler_name(irq, &ht_irq_chip,
1969 handle_edge_irq, "edge");
1973 #endif /* CONFIG_HT_IRQ */
1975 /* --------------------------------------------------------------------------
1976 ACPI-based IOAPIC Configuration
1977 -------------------------------------------------------------------------- */
1981 #define IO_APIC_MAX_ID 0xFE
1983 int __init io_apic_get_redir_entries (int ioapic)
1985 union IO_APIC_reg_01 reg_01;
1986 unsigned long flags;
1988 spin_lock_irqsave(&ioapic_lock, flags);
1989 reg_01.raw = io_apic_read(ioapic, 1);
1990 spin_unlock_irqrestore(&ioapic_lock, flags);
1992 return reg_01.bits.entries;
1996 int io_apic_set_pci_routing (int ioapic, int pin, int irq, int triggering, int polarity)
1998 struct IO_APIC_route_entry entry;
1999 unsigned long flags;
2003 if (!IO_APIC_IRQ(irq)) {
2004 apic_printk(APIC_QUIET,KERN_ERR "IOAPIC[%d]: Invalid reference to IRQ 0\n",
2010 * IRQs < 16 are already in the irq_2_pin[] map
2013 add_pin_to_irq(irq, ioapic, pin);
2016 vector = assign_irq_vector(irq, TARGET_CPUS, &mask);
2021 * Generate a PCI IRQ routing entry and program the IOAPIC accordingly.
2022 * Note that we mask (disable) IRQs now -- these get enabled when the
2023 * corresponding device driver registers for this IRQ.
2026 memset(&entry,0,sizeof(entry));
2028 entry.delivery_mode = INT_DELIVERY_MODE;
2029 entry.dest_mode = INT_DEST_MODE;
2030 entry.dest.logical.logical_dest = cpu_mask_to_apicid(mask);
2031 entry.trigger = triggering;
2032 entry.polarity = polarity;
2033 entry.mask = 1; /* Disabled (masked) */
2034 entry.vector = vector & 0xff;
2036 apic_printk(APIC_VERBOSE,KERN_DEBUG "IOAPIC[%d]: Set PCI routing entry (%d-%d -> 0x%x -> "
2037 "IRQ %d Mode:%i Active:%i)\n", ioapic,
2038 mp_ioapics[ioapic].mpc_apicid, pin, entry.vector, irq,
2039 triggering, polarity);
2041 ioapic_register_intr(irq, entry.vector, triggering);
2043 if (!ioapic && (irq < 16))
2044 disable_8259A_irq(irq);
2046 ioapic_write_entry(ioapic, pin, entry);
2048 spin_lock_irqsave(&ioapic_lock, flags);
2049 set_native_irq_info(irq, TARGET_CPUS);
2050 spin_unlock_irqrestore(&ioapic_lock, flags);
2055 #endif /* CONFIG_ACPI */
2059 * This function currently is only a helper for the i386 smp boot process where
2060 * we need to reprogram the ioredtbls to cater for the cpus which have come online
2061 * so mask in all cases should simply be TARGET_CPUS
2064 void __init setup_ioapic_dest(void)
2066 int pin, ioapic, irq, irq_entry;
2068 if (skip_ioapic_setup == 1)
2071 for (ioapic = 0; ioapic < nr_ioapics; ioapic++) {
2072 for (pin = 0; pin < nr_ioapic_registers[ioapic]; pin++) {
2073 irq_entry = find_irq_entry(ioapic, pin, mp_INT);
2074 if (irq_entry == -1)
2076 irq = pin_2_irq(irq_entry, ioapic, pin);
2077 set_ioapic_affinity_irq(irq, TARGET_CPUS);