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>
55 /* irq_cfg is indexed by the sum of all RTEs in all I/O APICs. */
56 struct irq_cfg irq_cfg[NR_IRQS] __read_mostly = {
57 [0] = { .domain = CPU_MASK_ALL, .vector = FIRST_EXTERNAL_VECTOR + 0 },
58 [1] = { .domain = CPU_MASK_ALL, .vector = FIRST_EXTERNAL_VECTOR + 1 },
59 [2] = { .domain = CPU_MASK_ALL, .vector = FIRST_EXTERNAL_VECTOR + 2 },
60 [3] = { .domain = CPU_MASK_ALL, .vector = FIRST_EXTERNAL_VECTOR + 3 },
61 [4] = { .domain = CPU_MASK_ALL, .vector = FIRST_EXTERNAL_VECTOR + 4 },
62 [5] = { .domain = CPU_MASK_ALL, .vector = FIRST_EXTERNAL_VECTOR + 5 },
63 [6] = { .domain = CPU_MASK_ALL, .vector = FIRST_EXTERNAL_VECTOR + 6 },
64 [7] = { .domain = CPU_MASK_ALL, .vector = FIRST_EXTERNAL_VECTOR + 7 },
65 [8] = { .domain = CPU_MASK_ALL, .vector = FIRST_EXTERNAL_VECTOR + 8 },
66 [9] = { .domain = CPU_MASK_ALL, .vector = FIRST_EXTERNAL_VECTOR + 9 },
67 [10] = { .domain = CPU_MASK_ALL, .vector = FIRST_EXTERNAL_VECTOR + 10 },
68 [11] = { .domain = CPU_MASK_ALL, .vector = FIRST_EXTERNAL_VECTOR + 11 },
69 [12] = { .domain = CPU_MASK_ALL, .vector = FIRST_EXTERNAL_VECTOR + 12 },
70 [13] = { .domain = CPU_MASK_ALL, .vector = FIRST_EXTERNAL_VECTOR + 13 },
71 [14] = { .domain = CPU_MASK_ALL, .vector = FIRST_EXTERNAL_VECTOR + 14 },
72 [15] = { .domain = CPU_MASK_ALL, .vector = FIRST_EXTERNAL_VECTOR + 15 },
75 static int assign_irq_vector(int irq, cpumask_t mask, cpumask_t *result);
77 #define __apicdebuginit __init
79 int sis_apic_bug; /* not actually supported, dummy for compile */
81 static int no_timer_check;
83 static int disable_timer_pin_1 __initdata;
85 int timer_over_8254 __initdata = 1;
87 /* Where if anywhere is the i8259 connect in external int mode */
88 static struct { int pin, apic; } ioapic_i8259 = { -1, -1 };
90 static DEFINE_SPINLOCK(ioapic_lock);
91 DEFINE_SPINLOCK(vector_lock);
94 * # of IRQ routing registers
96 int nr_ioapic_registers[MAX_IO_APICS];
99 * Rough estimation of how many shared IRQs there are, can
100 * be changed anytime.
102 #define MAX_PLUS_SHARED_IRQS NR_IRQS
103 #define PIN_MAP_SIZE (MAX_PLUS_SHARED_IRQS + NR_IRQS)
106 * This is performance-critical, we want to do it O(1)
108 * the indexing order of this array favors 1:1 mappings
109 * between pins and IRQs.
112 static struct irq_pin_list {
113 short apic, pin, next;
114 } irq_2_pin[PIN_MAP_SIZE];
118 unsigned int unused[3];
122 static __attribute_const__ struct io_apic __iomem *io_apic_base(int idx)
124 return (void __iomem *) __fix_to_virt(FIX_IO_APIC_BASE_0 + idx)
125 + (mp_ioapics[idx].mpc_apicaddr & ~PAGE_MASK);
128 static inline unsigned int io_apic_read(unsigned int apic, unsigned int reg)
130 struct io_apic __iomem *io_apic = io_apic_base(apic);
131 writel(reg, &io_apic->index);
132 return readl(&io_apic->data);
135 static inline void io_apic_write(unsigned int apic, unsigned int reg, unsigned int value)
137 struct io_apic __iomem *io_apic = io_apic_base(apic);
138 writel(reg, &io_apic->index);
139 writel(value, &io_apic->data);
143 * Re-write a value: to be used for read-modify-write
144 * cycles where the read already set up the index register.
146 static inline void io_apic_modify(unsigned int apic, unsigned int value)
148 struct io_apic __iomem *io_apic = io_apic_base(apic);
149 writel(value, &io_apic->data);
153 * Synchronize the IO-APIC and the CPU by doing
154 * a dummy read from the IO-APIC
156 static inline void io_apic_sync(unsigned int apic)
158 struct io_apic __iomem *io_apic = io_apic_base(apic);
159 readl(&io_apic->data);
162 #define __DO_ACTION(R, ACTION, FINAL) \
166 struct irq_pin_list *entry = irq_2_pin + irq; \
168 BUG_ON(irq >= NR_IRQS); \
174 reg = io_apic_read(entry->apic, 0x10 + R + pin*2); \
176 io_apic_modify(entry->apic, reg); \
180 entry = irq_2_pin + entry->next; \
185 struct { u32 w1, w2; };
186 struct IO_APIC_route_entry entry;
189 static struct IO_APIC_route_entry ioapic_read_entry(int apic, int pin)
191 union entry_union eu;
193 spin_lock_irqsave(&ioapic_lock, flags);
194 eu.w1 = io_apic_read(apic, 0x10 + 2 * pin);
195 eu.w2 = io_apic_read(apic, 0x11 + 2 * pin);
196 spin_unlock_irqrestore(&ioapic_lock, flags);
201 * When we write a new IO APIC routing entry, we need to write the high
202 * word first! If the mask bit in the low word is clear, we will enable
203 * the interrupt, and we need to make sure the entry is fully populated
204 * before that happens.
207 __ioapic_write_entry(int apic, int pin, struct IO_APIC_route_entry e)
209 union entry_union eu;
211 io_apic_write(apic, 0x11 + 2*pin, eu.w2);
212 io_apic_write(apic, 0x10 + 2*pin, eu.w1);
215 static void ioapic_write_entry(int apic, int pin, struct IO_APIC_route_entry e)
218 spin_lock_irqsave(&ioapic_lock, flags);
219 __ioapic_write_entry(apic, pin, e);
220 spin_unlock_irqrestore(&ioapic_lock, flags);
224 * When we mask an IO APIC routing entry, we need to write the low
225 * word first, in order to set the mask bit before we change the
228 static void ioapic_mask_entry(int apic, int pin)
231 union entry_union eu = { .entry.mask = 1 };
233 spin_lock_irqsave(&ioapic_lock, flags);
234 io_apic_write(apic, 0x10 + 2*pin, eu.w1);
235 io_apic_write(apic, 0x11 + 2*pin, eu.w2);
236 spin_unlock_irqrestore(&ioapic_lock, flags);
240 static void __target_IO_APIC_irq(unsigned int irq, unsigned int dest, u8 vector)
243 struct irq_pin_list *entry = irq_2_pin + irq;
245 BUG_ON(irq >= NR_IRQS);
252 io_apic_write(apic, 0x11 + pin*2, dest);
253 reg = io_apic_read(apic, 0x10 + pin*2);
256 io_apic_modify(apic, reg);
259 entry = irq_2_pin + entry->next;
263 static void set_ioapic_affinity_irq(unsigned int irq, cpumask_t mask)
270 cpus_and(tmp, mask, cpu_online_map);
274 vector = assign_irq_vector(irq, mask, &tmp);
278 dest = cpu_mask_to_apicid(tmp);
281 * Only the high 8 bits are valid.
283 dest = SET_APIC_LOGICAL_ID(dest);
285 spin_lock_irqsave(&ioapic_lock, flags);
286 __target_IO_APIC_irq(irq, dest, vector);
287 irq_desc[irq].affinity = mask;
288 spin_unlock_irqrestore(&ioapic_lock, flags);
293 * The common case is 1:1 IRQ<->pin mappings. Sometimes there are
294 * shared ISA-space IRQs, so we have to support them. We are super
295 * fast in the common case, and fast for shared ISA-space IRQs.
297 static void add_pin_to_irq(unsigned int irq, int apic, int pin)
299 static int first_free_entry = NR_IRQS;
300 struct irq_pin_list *entry = irq_2_pin + irq;
302 BUG_ON(irq >= NR_IRQS);
304 entry = irq_2_pin + entry->next;
306 if (entry->pin != -1) {
307 entry->next = first_free_entry;
308 entry = irq_2_pin + entry->next;
309 if (++first_free_entry >= PIN_MAP_SIZE)
310 panic("io_apic.c: ran out of irq_2_pin entries!");
317 #define DO_ACTION(name,R,ACTION, FINAL) \
319 static void name##_IO_APIC_irq (unsigned int irq) \
320 __DO_ACTION(R, ACTION, FINAL)
322 DO_ACTION( __mask, 0, |= 0x00010000, io_apic_sync(entry->apic) )
324 DO_ACTION( __unmask, 0, &= 0xfffeffff, )
327 static void mask_IO_APIC_irq (unsigned int irq)
331 spin_lock_irqsave(&ioapic_lock, flags);
332 __mask_IO_APIC_irq(irq);
333 spin_unlock_irqrestore(&ioapic_lock, flags);
336 static void unmask_IO_APIC_irq (unsigned int irq)
340 spin_lock_irqsave(&ioapic_lock, flags);
341 __unmask_IO_APIC_irq(irq);
342 spin_unlock_irqrestore(&ioapic_lock, flags);
345 static void clear_IO_APIC_pin(unsigned int apic, unsigned int pin)
347 struct IO_APIC_route_entry entry;
349 /* Check delivery_mode to be sure we're not clearing an SMI pin */
350 entry = ioapic_read_entry(apic, pin);
351 if (entry.delivery_mode == dest_SMI)
354 * Disable it in the IO-APIC irq-routing table:
356 ioapic_mask_entry(apic, pin);
359 static void clear_IO_APIC (void)
363 for (apic = 0; apic < nr_ioapics; apic++)
364 for (pin = 0; pin < nr_ioapic_registers[apic]; pin++)
365 clear_IO_APIC_pin(apic, pin);
368 int skip_ioapic_setup;
371 /* dummy parsing: see setup.c */
373 static int __init disable_ioapic_setup(char *str)
375 skip_ioapic_setup = 1;
378 early_param("noapic", disable_ioapic_setup);
380 /* Actually the next is obsolete, but keep it for paranoid reasons -AK */
381 static int __init disable_timer_pin_setup(char *arg)
383 disable_timer_pin_1 = 1;
386 __setup("disable_timer_pin_1", disable_timer_pin_setup);
388 static int __init setup_disable_8254_timer(char *s)
390 timer_over_8254 = -1;
393 static int __init setup_enable_8254_timer(char *s)
399 __setup("disable_8254_timer", setup_disable_8254_timer);
400 __setup("enable_8254_timer", setup_enable_8254_timer);
404 * Find the IRQ entry number of a certain pin.
406 static int find_irq_entry(int apic, int pin, int type)
410 for (i = 0; i < mp_irq_entries; i++)
411 if (mp_irqs[i].mpc_irqtype == type &&
412 (mp_irqs[i].mpc_dstapic == mp_ioapics[apic].mpc_apicid ||
413 mp_irqs[i].mpc_dstapic == MP_APIC_ALL) &&
414 mp_irqs[i].mpc_dstirq == pin)
421 * Find the pin to which IRQ[irq] (ISA) is connected
423 static int __init find_isa_irq_pin(int irq, int type)
427 for (i = 0; i < mp_irq_entries; i++) {
428 int lbus = mp_irqs[i].mpc_srcbus;
430 if (test_bit(lbus, mp_bus_not_pci) &&
431 (mp_irqs[i].mpc_irqtype == type) &&
432 (mp_irqs[i].mpc_srcbusirq == irq))
434 return mp_irqs[i].mpc_dstirq;
439 static int __init find_isa_irq_apic(int irq, int type)
443 for (i = 0; i < mp_irq_entries; i++) {
444 int lbus = mp_irqs[i].mpc_srcbus;
446 if (test_bit(lbus, mp_bus_not_pci) &&
447 (mp_irqs[i].mpc_irqtype == type) &&
448 (mp_irqs[i].mpc_srcbusirq == irq))
451 if (i < mp_irq_entries) {
453 for(apic = 0; apic < nr_ioapics; apic++) {
454 if (mp_ioapics[apic].mpc_apicid == mp_irqs[i].mpc_dstapic)
463 * Find a specific PCI IRQ entry.
464 * Not an __init, possibly needed by modules
466 static int pin_2_irq(int idx, int apic, int pin);
468 int IO_APIC_get_PCI_irq_vector(int bus, int slot, int pin)
470 int apic, i, best_guess = -1;
472 apic_printk(APIC_DEBUG, "querying PCI -> IRQ mapping bus:%d, slot:%d, pin:%d.\n",
474 if (mp_bus_id_to_pci_bus[bus] == -1) {
475 apic_printk(APIC_VERBOSE, "PCI BIOS passed nonexistent PCI bus %d!\n", bus);
478 for (i = 0; i < mp_irq_entries; i++) {
479 int lbus = mp_irqs[i].mpc_srcbus;
481 for (apic = 0; apic < nr_ioapics; apic++)
482 if (mp_ioapics[apic].mpc_apicid == mp_irqs[i].mpc_dstapic ||
483 mp_irqs[i].mpc_dstapic == MP_APIC_ALL)
486 if (!test_bit(lbus, mp_bus_not_pci) &&
487 !mp_irqs[i].mpc_irqtype &&
489 (slot == ((mp_irqs[i].mpc_srcbusirq >> 2) & 0x1f))) {
490 int irq = pin_2_irq(i,apic,mp_irqs[i].mpc_dstirq);
492 if (!(apic || IO_APIC_IRQ(irq)))
495 if (pin == (mp_irqs[i].mpc_srcbusirq & 3))
498 * Use the first all-but-pin matching entry as a
499 * best-guess fuzzy result for broken mptables.
505 BUG_ON(best_guess >= NR_IRQS);
509 /* ISA interrupts are always polarity zero edge triggered,
510 * when listed as conforming in the MP table. */
512 #define default_ISA_trigger(idx) (0)
513 #define default_ISA_polarity(idx) (0)
515 /* PCI interrupts are always polarity one level triggered,
516 * when listed as conforming in the MP table. */
518 #define default_PCI_trigger(idx) (1)
519 #define default_PCI_polarity(idx) (1)
521 static int __init MPBIOS_polarity(int idx)
523 int bus = mp_irqs[idx].mpc_srcbus;
527 * Determine IRQ line polarity (high active or low active):
529 switch (mp_irqs[idx].mpc_irqflag & 3)
531 case 0: /* conforms, ie. bus-type dependent polarity */
532 if (test_bit(bus, mp_bus_not_pci))
533 polarity = default_ISA_polarity(idx);
535 polarity = default_PCI_polarity(idx);
537 case 1: /* high active */
542 case 2: /* reserved */
544 printk(KERN_WARNING "broken BIOS!!\n");
548 case 3: /* low active */
553 default: /* invalid */
555 printk(KERN_WARNING "broken BIOS!!\n");
563 static int MPBIOS_trigger(int idx)
565 int bus = mp_irqs[idx].mpc_srcbus;
569 * Determine IRQ trigger mode (edge or level sensitive):
571 switch ((mp_irqs[idx].mpc_irqflag>>2) & 3)
573 case 0: /* conforms, ie. bus-type dependent */
574 if (test_bit(bus, mp_bus_not_pci))
575 trigger = default_ISA_trigger(idx);
577 trigger = default_PCI_trigger(idx);
584 case 2: /* reserved */
586 printk(KERN_WARNING "broken BIOS!!\n");
595 default: /* invalid */
597 printk(KERN_WARNING "broken BIOS!!\n");
605 static inline int irq_polarity(int idx)
607 return MPBIOS_polarity(idx);
610 static inline int irq_trigger(int idx)
612 return MPBIOS_trigger(idx);
615 static int pin_2_irq(int idx, int apic, int pin)
618 int bus = mp_irqs[idx].mpc_srcbus;
621 * Debugging check, we are in big trouble if this message pops up!
623 if (mp_irqs[idx].mpc_dstirq != pin)
624 printk(KERN_ERR "broken BIOS or MPTABLE parser, ayiee!!\n");
626 if (test_bit(bus, mp_bus_not_pci)) {
627 irq = mp_irqs[idx].mpc_srcbusirq;
630 * PCI IRQs are mapped in order
634 irq += nr_ioapic_registers[i++];
637 BUG_ON(irq >= NR_IRQS);
641 static int __assign_irq_vector(int irq, cpumask_t mask, cpumask_t *result)
644 * NOTE! The local APIC isn't very good at handling
645 * multiple interrupts at the same interrupt level.
646 * As the interrupt level is determined by taking the
647 * vector number and shifting that right by 4, we
648 * want to spread these out a bit so that they don't
649 * all fall in the same interrupt level.
651 * Also, we've got to be careful not to trash gate
652 * 0x80, because int 0x80 is hm, kind of importantish. ;)
654 static int current_vector = FIRST_DEVICE_VECTOR, current_offset = 0;
655 cpumask_t old_mask = CPU_MASK_NONE;
660 BUG_ON((unsigned)irq >= NR_IRQS);
663 /* Only try and allocate irqs on cpus that are present */
664 cpus_and(mask, mask, cpu_online_map);
667 old_vector = cfg->vector;
668 if (old_vector > 0) {
669 cpus_and(*result, cfg->domain, mask);
670 if (!cpus_empty(*result))
672 cpus_and(old_mask, cfg->domain, cpu_online_map);
675 for_each_cpu_mask(cpu, mask) {
676 cpumask_t domain, new_mask;
677 int new_cpu, old_cpu;
680 domain = vector_allocation_domain(cpu);
681 cpus_and(new_mask, domain, cpu_online_map);
683 vector = current_vector;
684 offset = current_offset;
687 if (vector >= FIRST_SYSTEM_VECTOR) {
688 /* If we run out of vectors on large boxen, must share them. */
689 offset = (offset + 1) % 8;
690 vector = FIRST_DEVICE_VECTOR + offset;
692 if (unlikely(current_vector == vector))
694 if (vector == IA32_SYSCALL_VECTOR)
696 for_each_cpu_mask(new_cpu, new_mask)
697 if (per_cpu(vector_irq, new_cpu)[vector] != -1)
700 current_vector = vector;
701 current_offset = offset;
702 for_each_cpu_mask(old_cpu, old_mask)
703 per_cpu(vector_irq, old_cpu)[old_vector] = -1;
704 for_each_cpu_mask(new_cpu, new_mask)
705 per_cpu(vector_irq, new_cpu)[vector] = irq;
706 cfg->vector = vector;
707 cfg->domain = domain;
708 cpus_and(*result, domain, mask);
714 static int assign_irq_vector(int irq, cpumask_t mask, cpumask_t *result)
719 spin_lock_irqsave(&vector_lock, flags);
720 vector = __assign_irq_vector(irq, mask, result);
721 spin_unlock_irqrestore(&vector_lock, flags);
725 static void __clear_irq_vector(int irq)
731 BUG_ON((unsigned)irq >= NR_IRQS);
733 BUG_ON(!cfg->vector);
735 vector = cfg->vector;
736 cpus_and(mask, cfg->domain, cpu_online_map);
737 for_each_cpu_mask(cpu, mask)
738 per_cpu(vector_irq, cpu)[vector] = -1;
741 cfg->domain = CPU_MASK_NONE;
744 void __setup_vector_irq(int cpu)
746 /* Initialize vector_irq on a new cpu */
747 /* This function must be called with vector_lock held */
750 /* Mark the inuse vectors */
751 for (irq = 0; irq < NR_IRQS; ++irq) {
752 if (!cpu_isset(cpu, irq_cfg[irq].domain))
754 vector = irq_cfg[irq].vector;
755 per_cpu(vector_irq, cpu)[vector] = irq;
757 /* Mark the free vectors */
758 for (vector = 0; vector < NR_VECTORS; ++vector) {
759 irq = per_cpu(vector_irq, cpu)[vector];
762 if (!cpu_isset(cpu, irq_cfg[irq].domain))
763 per_cpu(vector_irq, cpu)[vector] = -1;
768 static struct irq_chip ioapic_chip;
770 static void ioapic_register_intr(int irq, unsigned long trigger)
773 set_irq_chip_and_handler_name(irq, &ioapic_chip,
774 handle_fasteoi_irq, "fasteoi");
776 set_irq_chip_and_handler_name(irq, &ioapic_chip,
777 handle_edge_irq, "edge");
780 static void setup_IO_APIC_irq(int apic, int pin, unsigned int irq,
781 int trigger, int polarity)
783 struct IO_APIC_route_entry entry;
788 if (!IO_APIC_IRQ(irq))
791 vector = assign_irq_vector(irq, TARGET_CPUS, &mask);
795 apic_printk(APIC_VERBOSE,KERN_DEBUG
796 "IOAPIC[%d]: Set routing entry (%d-%d -> 0x%x -> "
797 "IRQ %d Mode:%i Active:%i)\n",
798 apic, mp_ioapics[apic].mpc_apicid, pin, vector,
799 irq, trigger, polarity);
802 * add it to the IO-APIC irq-routing table:
804 memset(&entry,0,sizeof(entry));
806 entry.delivery_mode = INT_DELIVERY_MODE;
807 entry.dest_mode = INT_DEST_MODE;
808 entry.dest = cpu_mask_to_apicid(mask);
809 entry.mask = 0; /* enable IRQ */
810 entry.trigger = trigger;
811 entry.polarity = polarity;
812 entry.vector = vector;
814 /* Mask level triggered irqs.
815 * Use IRQ_DELAYED_DISABLE for edge triggered irqs.
820 ioapic_register_intr(irq, trigger);
822 disable_8259A_irq(irq);
824 ioapic_write_entry(apic, pin, entry);
826 spin_lock_irqsave(&ioapic_lock, flags);
827 irq_desc[irq].affinity = TARGET_CPUS;
828 spin_unlock_irqrestore(&ioapic_lock, flags);
831 static void __init setup_IO_APIC_irqs(void)
833 int apic, pin, idx, irq, first_notcon = 1;
835 apic_printk(APIC_VERBOSE, KERN_DEBUG "init IO_APIC IRQs\n");
837 for (apic = 0; apic < nr_ioapics; apic++) {
838 for (pin = 0; pin < nr_ioapic_registers[apic]; pin++) {
840 idx = find_irq_entry(apic,pin,mp_INT);
843 apic_printk(APIC_VERBOSE, KERN_DEBUG " IO-APIC (apicid-pin) %d-%d", mp_ioapics[apic].mpc_apicid, pin);
846 apic_printk(APIC_VERBOSE, ", %d-%d", mp_ioapics[apic].mpc_apicid, pin);
850 irq = pin_2_irq(idx, apic, pin);
851 add_pin_to_irq(irq, apic, pin);
853 setup_IO_APIC_irq(apic, pin, irq,
854 irq_trigger(idx), irq_polarity(idx));
859 apic_printk(APIC_VERBOSE," not connected.\n");
863 * Set up the 8259A-master output pin as broadcast to all
866 static void __init setup_ExtINT_IRQ0_pin(unsigned int apic, unsigned int pin, int vector)
868 struct IO_APIC_route_entry entry;
871 memset(&entry,0,sizeof(entry));
873 disable_8259A_irq(0);
876 apic_write(APIC_LVT0, APIC_LVT_MASKED | APIC_DM_EXTINT);
879 * We use logical delivery to get the timer IRQ
882 entry.dest_mode = INT_DEST_MODE;
883 entry.mask = 0; /* unmask IRQ now */
884 entry.dest = cpu_mask_to_apicid(TARGET_CPUS);
885 entry.delivery_mode = INT_DELIVERY_MODE;
888 entry.vector = vector;
891 * The timer IRQ doesn't have to know that behind the
892 * scene we have a 8259A-master in AEOI mode ...
894 set_irq_chip_and_handler_name(0, &ioapic_chip, handle_edge_irq, "edge");
897 * Add it to the IO-APIC irq-routing table:
899 spin_lock_irqsave(&ioapic_lock, flags);
900 io_apic_write(apic, 0x11+2*pin, *(((int *)&entry)+1));
901 io_apic_write(apic, 0x10+2*pin, *(((int *)&entry)+0));
902 spin_unlock_irqrestore(&ioapic_lock, flags);
907 void __init UNEXPECTED_IO_APIC(void)
911 void __apicdebuginit print_IO_APIC(void)
914 union IO_APIC_reg_00 reg_00;
915 union IO_APIC_reg_01 reg_01;
916 union IO_APIC_reg_02 reg_02;
919 if (apic_verbosity == APIC_QUIET)
922 printk(KERN_DEBUG "number of MP IRQ sources: %d.\n", mp_irq_entries);
923 for (i = 0; i < nr_ioapics; i++)
924 printk(KERN_DEBUG "number of IO-APIC #%d registers: %d.\n",
925 mp_ioapics[i].mpc_apicid, nr_ioapic_registers[i]);
928 * We are a bit conservative about what we expect. We have to
929 * know about every hardware change ASAP.
931 printk(KERN_INFO "testing the IO APIC.......................\n");
933 for (apic = 0; apic < nr_ioapics; apic++) {
935 spin_lock_irqsave(&ioapic_lock, flags);
936 reg_00.raw = io_apic_read(apic, 0);
937 reg_01.raw = io_apic_read(apic, 1);
938 if (reg_01.bits.version >= 0x10)
939 reg_02.raw = io_apic_read(apic, 2);
940 spin_unlock_irqrestore(&ioapic_lock, flags);
943 printk(KERN_DEBUG "IO APIC #%d......\n", mp_ioapics[apic].mpc_apicid);
944 printk(KERN_DEBUG ".... register #00: %08X\n", reg_00.raw);
945 printk(KERN_DEBUG "....... : physical APIC id: %02X\n", reg_00.bits.ID);
946 if (reg_00.bits.__reserved_1 || reg_00.bits.__reserved_2)
947 UNEXPECTED_IO_APIC();
949 printk(KERN_DEBUG ".... register #01: %08X\n", *(int *)®_01);
950 printk(KERN_DEBUG "....... : max redirection entries: %04X\n", reg_01.bits.entries);
951 if ( (reg_01.bits.entries != 0x0f) && /* older (Neptune) boards */
952 (reg_01.bits.entries != 0x17) && /* typical ISA+PCI boards */
953 (reg_01.bits.entries != 0x1b) && /* Compaq Proliant boards */
954 (reg_01.bits.entries != 0x1f) && /* dual Xeon boards */
955 (reg_01.bits.entries != 0x22) && /* bigger Xeon boards */
956 (reg_01.bits.entries != 0x2E) &&
957 (reg_01.bits.entries != 0x3F) &&
958 (reg_01.bits.entries != 0x03)
960 UNEXPECTED_IO_APIC();
962 printk(KERN_DEBUG "....... : PRQ implemented: %X\n", reg_01.bits.PRQ);
963 printk(KERN_DEBUG "....... : IO APIC version: %04X\n", reg_01.bits.version);
964 if ( (reg_01.bits.version != 0x01) && /* 82489DX IO-APICs */
965 (reg_01.bits.version != 0x02) && /* 82801BA IO-APICs (ICH2) */
966 (reg_01.bits.version != 0x10) && /* oldest IO-APICs */
967 (reg_01.bits.version != 0x11) && /* Pentium/Pro IO-APICs */
968 (reg_01.bits.version != 0x13) && /* Xeon IO-APICs */
969 (reg_01.bits.version != 0x20) /* Intel P64H (82806 AA) */
971 UNEXPECTED_IO_APIC();
972 if (reg_01.bits.__reserved_1 || reg_01.bits.__reserved_2)
973 UNEXPECTED_IO_APIC();
975 if (reg_01.bits.version >= 0x10) {
976 printk(KERN_DEBUG ".... register #02: %08X\n", reg_02.raw);
977 printk(KERN_DEBUG "....... : arbitration: %02X\n", reg_02.bits.arbitration);
978 if (reg_02.bits.__reserved_1 || reg_02.bits.__reserved_2)
979 UNEXPECTED_IO_APIC();
982 printk(KERN_DEBUG ".... IRQ redirection table:\n");
984 printk(KERN_DEBUG " NR Dst Mask Trig IRR Pol"
985 " Stat Dmod Deli Vect: \n");
987 for (i = 0; i <= reg_01.bits.entries; i++) {
988 struct IO_APIC_route_entry entry;
990 entry = ioapic_read_entry(apic, i);
992 printk(KERN_DEBUG " %02x %03X ",
997 printk("%1d %1d %1d %1d %1d %1d %1d %02X\n",
1002 entry.delivery_status,
1004 entry.delivery_mode,
1009 printk(KERN_DEBUG "IRQ to pin mappings:\n");
1010 for (i = 0; i < NR_IRQS; i++) {
1011 struct irq_pin_list *entry = irq_2_pin + i;
1014 printk(KERN_DEBUG "IRQ%d ", i);
1016 printk("-> %d:%d", entry->apic, entry->pin);
1019 entry = irq_2_pin + entry->next;
1024 printk(KERN_INFO ".................................... done.\n");
1031 static __apicdebuginit void print_APIC_bitfield (int base)
1036 if (apic_verbosity == APIC_QUIET)
1039 printk(KERN_DEBUG "0123456789abcdef0123456789abcdef\n" KERN_DEBUG);
1040 for (i = 0; i < 8; i++) {
1041 v = apic_read(base + i*0x10);
1042 for (j = 0; j < 32; j++) {
1052 void __apicdebuginit print_local_APIC(void * dummy)
1054 unsigned int v, ver, maxlvt;
1056 if (apic_verbosity == APIC_QUIET)
1059 printk("\n" KERN_DEBUG "printing local APIC contents on CPU#%d/%d:\n",
1060 smp_processor_id(), hard_smp_processor_id());
1061 v = apic_read(APIC_ID);
1062 printk(KERN_INFO "... APIC ID: %08x (%01x)\n", v, GET_APIC_ID(v));
1063 v = apic_read(APIC_LVR);
1064 printk(KERN_INFO "... APIC VERSION: %08x\n", v);
1065 ver = GET_APIC_VERSION(v);
1066 maxlvt = get_maxlvt();
1068 v = apic_read(APIC_TASKPRI);
1069 printk(KERN_DEBUG "... APIC TASKPRI: %08x (%02x)\n", v, v & APIC_TPRI_MASK);
1071 v = apic_read(APIC_ARBPRI);
1072 printk(KERN_DEBUG "... APIC ARBPRI: %08x (%02x)\n", v,
1073 v & APIC_ARBPRI_MASK);
1074 v = apic_read(APIC_PROCPRI);
1075 printk(KERN_DEBUG "... APIC PROCPRI: %08x\n", v);
1077 v = apic_read(APIC_EOI);
1078 printk(KERN_DEBUG "... APIC EOI: %08x\n", v);
1079 v = apic_read(APIC_RRR);
1080 printk(KERN_DEBUG "... APIC RRR: %08x\n", v);
1081 v = apic_read(APIC_LDR);
1082 printk(KERN_DEBUG "... APIC LDR: %08x\n", v);
1083 v = apic_read(APIC_DFR);
1084 printk(KERN_DEBUG "... APIC DFR: %08x\n", v);
1085 v = apic_read(APIC_SPIV);
1086 printk(KERN_DEBUG "... APIC SPIV: %08x\n", v);
1088 printk(KERN_DEBUG "... APIC ISR field:\n");
1089 print_APIC_bitfield(APIC_ISR);
1090 printk(KERN_DEBUG "... APIC TMR field:\n");
1091 print_APIC_bitfield(APIC_TMR);
1092 printk(KERN_DEBUG "... APIC IRR field:\n");
1093 print_APIC_bitfield(APIC_IRR);
1095 v = apic_read(APIC_ESR);
1096 printk(KERN_DEBUG "... APIC ESR: %08x\n", v);
1098 v = apic_read(APIC_ICR);
1099 printk(KERN_DEBUG "... APIC ICR: %08x\n", v);
1100 v = apic_read(APIC_ICR2);
1101 printk(KERN_DEBUG "... APIC ICR2: %08x\n", v);
1103 v = apic_read(APIC_LVTT);
1104 printk(KERN_DEBUG "... APIC LVTT: %08x\n", v);
1106 if (maxlvt > 3) { /* PC is LVT#4. */
1107 v = apic_read(APIC_LVTPC);
1108 printk(KERN_DEBUG "... APIC LVTPC: %08x\n", v);
1110 v = apic_read(APIC_LVT0);
1111 printk(KERN_DEBUG "... APIC LVT0: %08x\n", v);
1112 v = apic_read(APIC_LVT1);
1113 printk(KERN_DEBUG "... APIC LVT1: %08x\n", v);
1115 if (maxlvt > 2) { /* ERR is LVT#3. */
1116 v = apic_read(APIC_LVTERR);
1117 printk(KERN_DEBUG "... APIC LVTERR: %08x\n", v);
1120 v = apic_read(APIC_TMICT);
1121 printk(KERN_DEBUG "... APIC TMICT: %08x\n", v);
1122 v = apic_read(APIC_TMCCT);
1123 printk(KERN_DEBUG "... APIC TMCCT: %08x\n", v);
1124 v = apic_read(APIC_TDCR);
1125 printk(KERN_DEBUG "... APIC TDCR: %08x\n", v);
1129 void print_all_local_APICs (void)
1131 on_each_cpu(print_local_APIC, NULL, 1, 1);
1134 void __apicdebuginit print_PIC(void)
1137 unsigned long flags;
1139 if (apic_verbosity == APIC_QUIET)
1142 printk(KERN_DEBUG "\nprinting PIC contents\n");
1144 spin_lock_irqsave(&i8259A_lock, flags);
1146 v = inb(0xa1) << 8 | inb(0x21);
1147 printk(KERN_DEBUG "... PIC IMR: %04x\n", v);
1149 v = inb(0xa0) << 8 | inb(0x20);
1150 printk(KERN_DEBUG "... PIC IRR: %04x\n", v);
1154 v = inb(0xa0) << 8 | inb(0x20);
1158 spin_unlock_irqrestore(&i8259A_lock, flags);
1160 printk(KERN_DEBUG "... PIC ISR: %04x\n", v);
1162 v = inb(0x4d1) << 8 | inb(0x4d0);
1163 printk(KERN_DEBUG "... PIC ELCR: %04x\n", v);
1168 static void __init enable_IO_APIC(void)
1170 union IO_APIC_reg_01 reg_01;
1171 int i8259_apic, i8259_pin;
1173 unsigned long flags;
1175 for (i = 0; i < PIN_MAP_SIZE; i++) {
1176 irq_2_pin[i].pin = -1;
1177 irq_2_pin[i].next = 0;
1181 * The number of IO-APIC IRQ registers (== #pins):
1183 for (apic = 0; apic < nr_ioapics; apic++) {
1184 spin_lock_irqsave(&ioapic_lock, flags);
1185 reg_01.raw = io_apic_read(apic, 1);
1186 spin_unlock_irqrestore(&ioapic_lock, flags);
1187 nr_ioapic_registers[apic] = reg_01.bits.entries+1;
1189 for(apic = 0; apic < nr_ioapics; apic++) {
1191 /* See if any of the pins is in ExtINT mode */
1192 for (pin = 0; pin < nr_ioapic_registers[apic]; pin++) {
1193 struct IO_APIC_route_entry entry;
1194 entry = ioapic_read_entry(apic, pin);
1196 /* If the interrupt line is enabled and in ExtInt mode
1197 * I have found the pin where the i8259 is connected.
1199 if ((entry.mask == 0) && (entry.delivery_mode == dest_ExtINT)) {
1200 ioapic_i8259.apic = apic;
1201 ioapic_i8259.pin = pin;
1207 /* Look to see what if the MP table has reported the ExtINT */
1208 i8259_pin = find_isa_irq_pin(0, mp_ExtINT);
1209 i8259_apic = find_isa_irq_apic(0, mp_ExtINT);
1210 /* Trust the MP table if nothing is setup in the hardware */
1211 if ((ioapic_i8259.pin == -1) && (i8259_pin >= 0)) {
1212 printk(KERN_WARNING "ExtINT not setup in hardware but reported by MP table\n");
1213 ioapic_i8259.pin = i8259_pin;
1214 ioapic_i8259.apic = i8259_apic;
1216 /* Complain if the MP table and the hardware disagree */
1217 if (((ioapic_i8259.apic != i8259_apic) || (ioapic_i8259.pin != i8259_pin)) &&
1218 (i8259_pin >= 0) && (ioapic_i8259.pin >= 0))
1220 printk(KERN_WARNING "ExtINT in hardware and MP table differ\n");
1224 * Do not trust the IO-APIC being empty at bootup
1230 * Not an __init, needed by the reboot code
1232 void disable_IO_APIC(void)
1235 * Clear the IO-APIC before rebooting:
1240 * If the i8259 is routed through an IOAPIC
1241 * Put that IOAPIC in virtual wire mode
1242 * so legacy interrupts can be delivered.
1244 if (ioapic_i8259.pin != -1) {
1245 struct IO_APIC_route_entry entry;
1247 memset(&entry, 0, sizeof(entry));
1248 entry.mask = 0; /* Enabled */
1249 entry.trigger = 0; /* Edge */
1251 entry.polarity = 0; /* High */
1252 entry.delivery_status = 0;
1253 entry.dest_mode = 0; /* Physical */
1254 entry.delivery_mode = dest_ExtINT; /* ExtInt */
1256 entry.dest = GET_APIC_ID(apic_read(APIC_ID));
1259 * Add it to the IO-APIC irq-routing table:
1261 ioapic_write_entry(ioapic_i8259.apic, ioapic_i8259.pin, entry);
1264 disconnect_bsp_APIC(ioapic_i8259.pin != -1);
1268 * There is a nasty bug in some older SMP boards, their mptable lies
1269 * about the timer IRQ. We do the following to work around the situation:
1271 * - timer IRQ defaults to IO-APIC IRQ
1272 * - if this function detects that timer IRQs are defunct, then we fall
1273 * back to ISA timer IRQs
1275 static int __init timer_irq_works(void)
1277 unsigned long t1 = jiffies;
1280 /* Let ten ticks pass... */
1281 mdelay((10 * 1000) / HZ);
1284 * Expect a few ticks at least, to be sure some possible
1285 * glue logic does not lock up after one or two first
1286 * ticks in a non-ExtINT mode. Also the local APIC
1287 * might have cached one ExtINT interrupt. Finally, at
1288 * least one tick may be lost due to delays.
1292 if (jiffies - t1 > 4)
1298 * In the SMP+IOAPIC case it might happen that there are an unspecified
1299 * number of pending IRQ events unhandled. These cases are very rare,
1300 * so we 'resend' these IRQs via IPIs, to the same CPU. It's much
1301 * better to do it this way as thus we do not have to be aware of
1302 * 'pending' interrupts in the IRQ path, except at this point.
1305 * Edge triggered needs to resend any interrupt
1306 * that was delayed but this is now handled in the device
1311 * Starting up a edge-triggered IO-APIC interrupt is
1312 * nasty - we need to make sure that we get the edge.
1313 * If it is already asserted for some reason, we need
1314 * return 1 to indicate that is was pending.
1316 * This is not complete - we should be able to fake
1317 * an edge even if it isn't on the 8259A...
1320 static unsigned int startup_ioapic_irq(unsigned int irq)
1322 int was_pending = 0;
1323 unsigned long flags;
1325 spin_lock_irqsave(&ioapic_lock, flags);
1327 disable_8259A_irq(irq);
1328 if (i8259A_irq_pending(irq))
1331 __unmask_IO_APIC_irq(irq);
1332 spin_unlock_irqrestore(&ioapic_lock, flags);
1337 static int ioapic_retrigger_irq(unsigned int irq)
1339 struct irq_cfg *cfg = &irq_cfg[irq];
1341 unsigned long flags;
1343 spin_lock_irqsave(&vector_lock, flags);
1345 cpu_set(first_cpu(cfg->domain), mask);
1347 send_IPI_mask(mask, cfg->vector);
1348 spin_unlock_irqrestore(&vector_lock, flags);
1354 * Level and edge triggered IO-APIC interrupts need different handling,
1355 * so we use two separate IRQ descriptors. Edge triggered IRQs can be
1356 * handled with the level-triggered descriptor, but that one has slightly
1357 * more overhead. Level-triggered interrupts cannot be handled with the
1358 * edge-triggered handler, without risking IRQ storms and other ugly
1362 static void ack_apic_edge(unsigned int irq)
1364 move_native_irq(irq);
1368 static void ack_apic_level(unsigned int irq)
1370 int do_unmask_irq = 0;
1372 #if defined(CONFIG_GENERIC_PENDING_IRQ) || defined(CONFIG_IRQBALANCE)
1373 /* If we are moving the irq we need to mask it */
1374 if (unlikely(irq_desc[irq].status & IRQ_MOVE_PENDING)) {
1376 mask_IO_APIC_irq(irq);
1381 * We must acknowledge the irq before we move it or the acknowledge will
1382 * not propogate properly.
1386 /* Now we can move and renable the irq */
1387 move_masked_irq(irq);
1388 if (unlikely(do_unmask_irq))
1389 unmask_IO_APIC_irq(irq);
1392 static struct irq_chip ioapic_chip __read_mostly = {
1394 .startup = startup_ioapic_irq,
1395 .mask = mask_IO_APIC_irq,
1396 .unmask = unmask_IO_APIC_irq,
1397 .ack = ack_apic_edge,
1398 .eoi = ack_apic_level,
1400 .set_affinity = set_ioapic_affinity_irq,
1402 .retrigger = ioapic_retrigger_irq,
1405 static inline void init_IO_APIC_traps(void)
1410 * NOTE! The local APIC isn't very good at handling
1411 * multiple interrupts at the same interrupt level.
1412 * As the interrupt level is determined by taking the
1413 * vector number and shifting that right by 4, we
1414 * want to spread these out a bit so that they don't
1415 * all fall in the same interrupt level.
1417 * Also, we've got to be careful not to trash gate
1418 * 0x80, because int 0x80 is hm, kind of importantish. ;)
1420 for (irq = 0; irq < NR_IRQS ; irq++) {
1422 if (IO_APIC_IRQ(tmp) && !irq_cfg[tmp].vector) {
1424 * Hmm.. We don't have an entry for this,
1425 * so default to an old-fashioned 8259
1426 * interrupt if we can..
1429 make_8259A_irq(irq);
1431 /* Strange. Oh, well.. */
1432 irq_desc[irq].chip = &no_irq_chip;
1437 static void enable_lapic_irq (unsigned int irq)
1441 v = apic_read(APIC_LVT0);
1442 apic_write(APIC_LVT0, v & ~APIC_LVT_MASKED);
1445 static void disable_lapic_irq (unsigned int irq)
1449 v = apic_read(APIC_LVT0);
1450 apic_write(APIC_LVT0, v | APIC_LVT_MASKED);
1453 static void ack_lapic_irq (unsigned int irq)
1458 static void end_lapic_irq (unsigned int i) { /* nothing */ }
1460 static struct hw_interrupt_type lapic_irq_type __read_mostly = {
1461 .typename = "local-APIC-edge",
1462 .startup = NULL, /* startup_irq() not used for IRQ0 */
1463 .shutdown = NULL, /* shutdown_irq() not used for IRQ0 */
1464 .enable = enable_lapic_irq,
1465 .disable = disable_lapic_irq,
1466 .ack = ack_lapic_irq,
1467 .end = end_lapic_irq,
1470 static void setup_nmi (void)
1473 * Dirty trick to enable the NMI watchdog ...
1474 * We put the 8259A master into AEOI mode and
1475 * unmask on all local APICs LVT0 as NMI.
1477 * The idea to use the 8259A in AEOI mode ('8259A Virtual Wire')
1478 * is from Maciej W. Rozycki - so we do not have to EOI from
1479 * the NMI handler or the timer interrupt.
1481 printk(KERN_INFO "activating NMI Watchdog ...");
1483 enable_NMI_through_LVT0(NULL);
1489 * This looks a bit hackish but it's about the only one way of sending
1490 * a few INTA cycles to 8259As and any associated glue logic. ICR does
1491 * not support the ExtINT mode, unfortunately. We need to send these
1492 * cycles as some i82489DX-based boards have glue logic that keeps the
1493 * 8259A interrupt line asserted until INTA. --macro
1495 static inline void unlock_ExtINT_logic(void)
1498 struct IO_APIC_route_entry entry0, entry1;
1499 unsigned char save_control, save_freq_select;
1500 unsigned long flags;
1502 pin = find_isa_irq_pin(8, mp_INT);
1503 apic = find_isa_irq_apic(8, mp_INT);
1507 spin_lock_irqsave(&ioapic_lock, flags);
1508 *(((int *)&entry0) + 1) = io_apic_read(apic, 0x11 + 2 * pin);
1509 *(((int *)&entry0) + 0) = io_apic_read(apic, 0x10 + 2 * pin);
1510 spin_unlock_irqrestore(&ioapic_lock, flags);
1511 clear_IO_APIC_pin(apic, pin);
1513 memset(&entry1, 0, sizeof(entry1));
1515 entry1.dest_mode = 0; /* physical delivery */
1516 entry1.mask = 0; /* unmask IRQ now */
1517 entry1.dest = hard_smp_processor_id();
1518 entry1.delivery_mode = dest_ExtINT;
1519 entry1.polarity = entry0.polarity;
1523 spin_lock_irqsave(&ioapic_lock, flags);
1524 io_apic_write(apic, 0x11 + 2 * pin, *(((int *)&entry1) + 1));
1525 io_apic_write(apic, 0x10 + 2 * pin, *(((int *)&entry1) + 0));
1526 spin_unlock_irqrestore(&ioapic_lock, flags);
1528 save_control = CMOS_READ(RTC_CONTROL);
1529 save_freq_select = CMOS_READ(RTC_FREQ_SELECT);
1530 CMOS_WRITE((save_freq_select & ~RTC_RATE_SELECT) | 0x6,
1532 CMOS_WRITE(save_control | RTC_PIE, RTC_CONTROL);
1537 if ((CMOS_READ(RTC_INTR_FLAGS) & RTC_PF) == RTC_PF)
1541 CMOS_WRITE(save_control, RTC_CONTROL);
1542 CMOS_WRITE(save_freq_select, RTC_FREQ_SELECT);
1543 clear_IO_APIC_pin(apic, pin);
1545 spin_lock_irqsave(&ioapic_lock, flags);
1546 io_apic_write(apic, 0x11 + 2 * pin, *(((int *)&entry0) + 1));
1547 io_apic_write(apic, 0x10 + 2 * pin, *(((int *)&entry0) + 0));
1548 spin_unlock_irqrestore(&ioapic_lock, flags);
1552 * This code may look a bit paranoid, but it's supposed to cooperate with
1553 * a wide range of boards and BIOS bugs. Fortunately only the timer IRQ
1554 * is so screwy. Thanks to Brian Perkins for testing/hacking this beast
1555 * fanatically on his truly buggy board.
1557 * FIXME: really need to revamp this for modern platforms only.
1559 static inline void check_timer(void)
1561 int apic1, pin1, apic2, pin2;
1566 * get/set the timer IRQ vector:
1568 disable_8259A_irq(0);
1569 vector = assign_irq_vector(0, TARGET_CPUS, &mask);
1572 * Subtle, code in do_timer_interrupt() expects an AEOI
1573 * mode for the 8259A whenever interrupts are routed
1574 * through I/O APICs. Also IRQ0 has to be enabled in
1575 * the 8259A which implies the virtual wire has to be
1576 * disabled in the local APIC.
1578 apic_write(APIC_LVT0, APIC_LVT_MASKED | APIC_DM_EXTINT);
1580 if (timer_over_8254 > 0)
1581 enable_8259A_irq(0);
1583 pin1 = find_isa_irq_pin(0, mp_INT);
1584 apic1 = find_isa_irq_apic(0, mp_INT);
1585 pin2 = ioapic_i8259.pin;
1586 apic2 = ioapic_i8259.apic;
1588 apic_printk(APIC_VERBOSE,KERN_INFO "..TIMER: vector=0x%02X apic1=%d pin1=%d apic2=%d pin2=%d\n",
1589 vector, apic1, pin1, apic2, pin2);
1593 * Ok, does IRQ0 through the IOAPIC work?
1595 unmask_IO_APIC_irq(0);
1596 if (!no_timer_check && timer_irq_works()) {
1597 nmi_watchdog_default();
1598 if (nmi_watchdog == NMI_IO_APIC) {
1599 disable_8259A_irq(0);
1601 enable_8259A_irq(0);
1603 if (disable_timer_pin_1 > 0)
1604 clear_IO_APIC_pin(0, pin1);
1607 clear_IO_APIC_pin(apic1, pin1);
1608 apic_printk(APIC_QUIET,KERN_ERR "..MP-BIOS bug: 8254 timer not "
1609 "connected to IO-APIC\n");
1612 apic_printk(APIC_VERBOSE,KERN_INFO "...trying to set up timer (IRQ0) "
1613 "through the 8259A ... ");
1615 apic_printk(APIC_VERBOSE,"\n..... (found apic %d pin %d) ...",
1618 * legacy devices should be connected to IO APIC #0
1620 setup_ExtINT_IRQ0_pin(apic2, pin2, vector);
1621 if (timer_irq_works()) {
1622 apic_printk(APIC_VERBOSE," works.\n");
1623 nmi_watchdog_default();
1624 if (nmi_watchdog == NMI_IO_APIC) {
1630 * Cleanup, just in case ...
1632 clear_IO_APIC_pin(apic2, pin2);
1634 apic_printk(APIC_VERBOSE," failed.\n");
1636 if (nmi_watchdog == NMI_IO_APIC) {
1637 printk(KERN_WARNING "timer doesn't work through the IO-APIC - disabling NMI Watchdog!\n");
1641 apic_printk(APIC_VERBOSE, KERN_INFO "...trying to set up timer as Virtual Wire IRQ...");
1643 disable_8259A_irq(0);
1644 irq_desc[0].chip = &lapic_irq_type;
1645 apic_write(APIC_LVT0, APIC_DM_FIXED | vector); /* Fixed mode */
1646 enable_8259A_irq(0);
1648 if (timer_irq_works()) {
1649 apic_printk(APIC_VERBOSE," works.\n");
1652 apic_write(APIC_LVT0, APIC_LVT_MASKED | APIC_DM_FIXED | vector);
1653 apic_printk(APIC_VERBOSE," failed.\n");
1655 apic_printk(APIC_VERBOSE, KERN_INFO "...trying to set up timer as ExtINT IRQ...");
1659 apic_write(APIC_LVT0, APIC_DM_EXTINT);
1661 unlock_ExtINT_logic();
1663 if (timer_irq_works()) {
1664 apic_printk(APIC_VERBOSE," works.\n");
1667 apic_printk(APIC_VERBOSE," failed :(.\n");
1668 panic("IO-APIC + timer doesn't work! Try using the 'noapic' kernel parameter\n");
1671 static int __init notimercheck(char *s)
1676 __setup("no_timer_check", notimercheck);
1680 * IRQ's that are handled by the PIC in the MPS IOAPIC case.
1681 * - IRQ2 is the cascade IRQ, and cannot be a io-apic IRQ.
1682 * Linux doesn't really care, as it's not actually used
1683 * for any interrupt handling anyway.
1685 #define PIC_IRQS (1<<2)
1687 void __init setup_IO_APIC(void)
1692 io_apic_irqs = ~0; /* all IRQs go through IOAPIC */
1694 io_apic_irqs = ~PIC_IRQS;
1696 apic_printk(APIC_VERBOSE, "ENABLING IO-APIC IRQs\n");
1699 setup_IO_APIC_irqs();
1700 init_IO_APIC_traps();
1706 struct sysfs_ioapic_data {
1707 struct sys_device dev;
1708 struct IO_APIC_route_entry entry[0];
1710 static struct sysfs_ioapic_data * mp_ioapic_data[MAX_IO_APICS];
1712 static int ioapic_suspend(struct sys_device *dev, pm_message_t state)
1714 struct IO_APIC_route_entry *entry;
1715 struct sysfs_ioapic_data *data;
1718 data = container_of(dev, struct sysfs_ioapic_data, dev);
1719 entry = data->entry;
1720 for (i = 0; i < nr_ioapic_registers[dev->id]; i ++, entry ++ )
1721 *entry = ioapic_read_entry(dev->id, i);
1726 static int ioapic_resume(struct sys_device *dev)
1728 struct IO_APIC_route_entry *entry;
1729 struct sysfs_ioapic_data *data;
1730 unsigned long flags;
1731 union IO_APIC_reg_00 reg_00;
1734 data = container_of(dev, struct sysfs_ioapic_data, dev);
1735 entry = data->entry;
1737 spin_lock_irqsave(&ioapic_lock, flags);
1738 reg_00.raw = io_apic_read(dev->id, 0);
1739 if (reg_00.bits.ID != mp_ioapics[dev->id].mpc_apicid) {
1740 reg_00.bits.ID = mp_ioapics[dev->id].mpc_apicid;
1741 io_apic_write(dev->id, 0, reg_00.raw);
1743 spin_unlock_irqrestore(&ioapic_lock, flags);
1744 for (i = 0; i < nr_ioapic_registers[dev->id]; i++)
1745 ioapic_write_entry(dev->id, i, entry[i]);
1750 static struct sysdev_class ioapic_sysdev_class = {
1751 set_kset_name("ioapic"),
1752 .suspend = ioapic_suspend,
1753 .resume = ioapic_resume,
1756 static int __init ioapic_init_sysfs(void)
1758 struct sys_device * dev;
1759 int i, size, error = 0;
1761 error = sysdev_class_register(&ioapic_sysdev_class);
1765 for (i = 0; i < nr_ioapics; i++ ) {
1766 size = sizeof(struct sys_device) + nr_ioapic_registers[i]
1767 * sizeof(struct IO_APIC_route_entry);
1768 mp_ioapic_data[i] = kmalloc(size, GFP_KERNEL);
1769 if (!mp_ioapic_data[i]) {
1770 printk(KERN_ERR "Can't suspend/resume IOAPIC %d\n", i);
1773 memset(mp_ioapic_data[i], 0, size);
1774 dev = &mp_ioapic_data[i]->dev;
1776 dev->cls = &ioapic_sysdev_class;
1777 error = sysdev_register(dev);
1779 kfree(mp_ioapic_data[i]);
1780 mp_ioapic_data[i] = NULL;
1781 printk(KERN_ERR "Can't suspend/resume IOAPIC %d\n", i);
1789 device_initcall(ioapic_init_sysfs);
1792 * Dynamic irq allocate and deallocation
1794 int create_irq(void)
1796 /* Allocate an unused irq */
1800 unsigned long flags;
1804 spin_lock_irqsave(&vector_lock, flags);
1805 for (new = (NR_IRQS - 1); new >= 0; new--) {
1806 if (platform_legacy_irq(new))
1808 if (irq_cfg[new].vector != 0)
1810 vector = __assign_irq_vector(new, TARGET_CPUS, &mask);
1811 if (likely(vector > 0))
1815 spin_unlock_irqrestore(&vector_lock, flags);
1818 dynamic_irq_init(irq);
1823 void destroy_irq(unsigned int irq)
1825 unsigned long flags;
1827 dynamic_irq_cleanup(irq);
1829 spin_lock_irqsave(&vector_lock, flags);
1830 __clear_irq_vector(irq);
1831 spin_unlock_irqrestore(&vector_lock, flags);
1835 * MSI mesage composition
1837 #ifdef CONFIG_PCI_MSI
1838 static int msi_compose_msg(struct pci_dev *pdev, unsigned int irq, struct msi_msg *msg)
1844 vector = assign_irq_vector(irq, TARGET_CPUS, &tmp);
1846 dest = cpu_mask_to_apicid(tmp);
1848 msg->address_hi = MSI_ADDR_BASE_HI;
1851 ((INT_DEST_MODE == 0) ?
1852 MSI_ADDR_DEST_MODE_PHYSICAL:
1853 MSI_ADDR_DEST_MODE_LOGICAL) |
1854 ((INT_DELIVERY_MODE != dest_LowestPrio) ?
1855 MSI_ADDR_REDIRECTION_CPU:
1856 MSI_ADDR_REDIRECTION_LOWPRI) |
1857 MSI_ADDR_DEST_ID(dest);
1860 MSI_DATA_TRIGGER_EDGE |
1861 MSI_DATA_LEVEL_ASSERT |
1862 ((INT_DELIVERY_MODE != dest_LowestPrio) ?
1863 MSI_DATA_DELIVERY_FIXED:
1864 MSI_DATA_DELIVERY_LOWPRI) |
1865 MSI_DATA_VECTOR(vector);
1871 static void set_msi_irq_affinity(unsigned int irq, cpumask_t mask)
1878 cpus_and(tmp, mask, cpu_online_map);
1879 if (cpus_empty(tmp))
1882 vector = assign_irq_vector(irq, mask, &tmp);
1886 dest = cpu_mask_to_apicid(tmp);
1888 read_msi_msg(irq, &msg);
1890 msg.data &= ~MSI_DATA_VECTOR_MASK;
1891 msg.data |= MSI_DATA_VECTOR(vector);
1892 msg.address_lo &= ~MSI_ADDR_DEST_ID_MASK;
1893 msg.address_lo |= MSI_ADDR_DEST_ID(dest);
1895 write_msi_msg(irq, &msg);
1896 irq_desc[irq].affinity = mask;
1898 #endif /* CONFIG_SMP */
1901 * IRQ Chip for MSI PCI/PCI-X/PCI-Express Devices,
1902 * which implement the MSI or MSI-X Capability Structure.
1904 static struct irq_chip msi_chip = {
1906 .unmask = unmask_msi_irq,
1907 .mask = mask_msi_irq,
1908 .ack = ack_apic_edge,
1910 .set_affinity = set_msi_irq_affinity,
1912 .retrigger = ioapic_retrigger_irq,
1915 int arch_setup_msi_irq(struct pci_dev *dev, struct msi_desc *desc)
1923 set_irq_msi(irq, desc);
1924 ret = msi_compose_msg(dev, irq, &msg);
1930 write_msi_msg(irq, &msg);
1932 set_irq_chip_and_handler_name(irq, &msi_chip, handle_edge_irq, "edge");
1937 void arch_teardown_msi_irq(unsigned int irq)
1942 #endif /* CONFIG_PCI_MSI */
1945 * Hypertransport interrupt support
1947 #ifdef CONFIG_HT_IRQ
1951 static void target_ht_irq(unsigned int irq, unsigned int dest, u8 vector)
1953 struct ht_irq_msg msg;
1954 fetch_ht_irq_msg(irq, &msg);
1956 msg.address_lo &= ~(HT_IRQ_LOW_VECTOR_MASK | HT_IRQ_LOW_DEST_ID_MASK);
1957 msg.address_hi &= ~(HT_IRQ_HIGH_DEST_ID_MASK);
1959 msg.address_lo |= HT_IRQ_LOW_VECTOR(vector) | HT_IRQ_LOW_DEST_ID(dest);
1960 msg.address_hi |= HT_IRQ_HIGH_DEST_ID(dest);
1962 write_ht_irq_msg(irq, &msg);
1965 static void set_ht_irq_affinity(unsigned int irq, cpumask_t mask)
1971 cpus_and(tmp, mask, cpu_online_map);
1972 if (cpus_empty(tmp))
1975 vector = assign_irq_vector(irq, mask, &tmp);
1979 dest = cpu_mask_to_apicid(tmp);
1981 target_ht_irq(irq, dest, vector);
1982 irq_desc[irq].affinity = mask;
1986 static struct irq_chip ht_irq_chip = {
1988 .mask = mask_ht_irq,
1989 .unmask = unmask_ht_irq,
1990 .ack = ack_apic_edge,
1992 .set_affinity = set_ht_irq_affinity,
1994 .retrigger = ioapic_retrigger_irq,
1997 int arch_setup_ht_irq(unsigned int irq, struct pci_dev *dev)
2002 vector = assign_irq_vector(irq, TARGET_CPUS, &tmp);
2004 struct ht_irq_msg msg;
2007 dest = cpu_mask_to_apicid(tmp);
2009 msg.address_hi = HT_IRQ_HIGH_DEST_ID(dest);
2013 HT_IRQ_LOW_DEST_ID(dest) |
2014 HT_IRQ_LOW_VECTOR(vector) |
2015 ((INT_DEST_MODE == 0) ?
2016 HT_IRQ_LOW_DM_PHYSICAL :
2017 HT_IRQ_LOW_DM_LOGICAL) |
2018 HT_IRQ_LOW_RQEOI_EDGE |
2019 ((INT_DELIVERY_MODE != dest_LowestPrio) ?
2020 HT_IRQ_LOW_MT_FIXED :
2021 HT_IRQ_LOW_MT_ARBITRATED) |
2022 HT_IRQ_LOW_IRQ_MASKED;
2024 write_ht_irq_msg(irq, &msg);
2026 set_irq_chip_and_handler_name(irq, &ht_irq_chip,
2027 handle_edge_irq, "edge");
2031 #endif /* CONFIG_HT_IRQ */
2033 /* --------------------------------------------------------------------------
2034 ACPI-based IOAPIC Configuration
2035 -------------------------------------------------------------------------- */
2039 #define IO_APIC_MAX_ID 0xFE
2041 int __init io_apic_get_redir_entries (int ioapic)
2043 union IO_APIC_reg_01 reg_01;
2044 unsigned long flags;
2046 spin_lock_irqsave(&ioapic_lock, flags);
2047 reg_01.raw = io_apic_read(ioapic, 1);
2048 spin_unlock_irqrestore(&ioapic_lock, flags);
2050 return reg_01.bits.entries;
2054 int io_apic_set_pci_routing (int ioapic, int pin, int irq, int triggering, int polarity)
2056 if (!IO_APIC_IRQ(irq)) {
2057 apic_printk(APIC_QUIET,KERN_ERR "IOAPIC[%d]: Invalid reference to IRQ 0\n",
2063 * IRQs < 16 are already in the irq_2_pin[] map
2066 add_pin_to_irq(irq, ioapic, pin);
2068 setup_IO_APIC_irq(ioapic, pin, irq, triggering, polarity);
2073 #endif /* CONFIG_ACPI */
2077 * This function currently is only a helper for the i386 smp boot process where
2078 * we need to reprogram the ioredtbls to cater for the cpus which have come online
2079 * so mask in all cases should simply be TARGET_CPUS
2082 void __init setup_ioapic_dest(void)
2084 int pin, ioapic, irq, irq_entry;
2086 if (skip_ioapic_setup == 1)
2089 for (ioapic = 0; ioapic < nr_ioapics; ioapic++) {
2090 for (pin = 0; pin < nr_ioapic_registers[ioapic]; pin++) {
2091 irq_entry = find_irq_entry(ioapic, pin, mp_INT);
2092 if (irq_entry == -1)
2094 irq = pin_2_irq(irq_entry, ioapic, pin);
2096 /* setup_IO_APIC_irqs could fail to get vector for some device
2097 * when you have too many devices, because at that time only boot
2100 if (!irq_cfg[irq].vector)
2101 setup_IO_APIC_irq(ioapic, pin, irq,
2102 irq_trigger(irq_entry),
2103 irq_polarity(irq_entry));
2105 set_ioapic_affinity_irq(irq, TARGET_CPUS);