2 * Intel IO-APIC support for multi-Pentium hosts.
4 * Copyright (C) 1997, 1998, 1999, 2000 Ingo Molnar, Hajnalka Szabo
6 * Many thanks to Stig Venaas for trying out countless experimental
7 * patches and reporting/debugging problems patiently!
9 * (c) 1999, Multiple IO-APIC support, developed by
10 * Ken-ichi Yaku <yaku@css1.kbnes.nec.co.jp> and
11 * Hidemi Kishimoto <kisimoto@css1.kbnes.nec.co.jp>,
12 * further tested and cleaned up by Zach Brown <zab@redhat.com>
13 * and Ingo Molnar <mingo@redhat.com>
16 * Maciej W. Rozycki : Bits for genuine 82489DX APICs;
17 * thanks to Eric Gilmore
19 * for testing these extensively
20 * Paul Diefenbaugh : Added full ACPI support
24 #include <linux/interrupt.h>
25 #include <linux/init.h>
26 #include <linux/delay.h>
27 #include <linux/sched.h>
28 #include <linux/pci.h>
29 #include <linux/mc146818rtc.h>
30 #include <linux/acpi.h>
31 #include <linux/sysdev.h>
32 #include <linux/msi.h>
33 #include <linux/htirq.h>
34 #include <linux/dmar.h>
35 #include <linux/jiffies.h>
37 #include <acpi/acpi_bus.h>
39 #include <linux/bootmem.h>
45 #include <asm/proto.h>
49 #include <asm/msidef.h>
50 #include <asm/hypertransport.h>
53 #include <mach_apic.h>
58 unsigned move_cleanup_count;
60 u8 move_in_progress : 1;
63 /* irq_cfg is indexed by the sum of all RTEs in all I/O APICs. */
64 struct irq_cfg irq_cfg[NR_IRQS] __read_mostly = {
65 [0] = { .domain = CPU_MASK_ALL, .vector = IRQ0_VECTOR, },
66 [1] = { .domain = CPU_MASK_ALL, .vector = IRQ1_VECTOR, },
67 [2] = { .domain = CPU_MASK_ALL, .vector = IRQ2_VECTOR, },
68 [3] = { .domain = CPU_MASK_ALL, .vector = IRQ3_VECTOR, },
69 [4] = { .domain = CPU_MASK_ALL, .vector = IRQ4_VECTOR, },
70 [5] = { .domain = CPU_MASK_ALL, .vector = IRQ5_VECTOR, },
71 [6] = { .domain = CPU_MASK_ALL, .vector = IRQ6_VECTOR, },
72 [7] = { .domain = CPU_MASK_ALL, .vector = IRQ7_VECTOR, },
73 [8] = { .domain = CPU_MASK_ALL, .vector = IRQ8_VECTOR, },
74 [9] = { .domain = CPU_MASK_ALL, .vector = IRQ9_VECTOR, },
75 [10] = { .domain = CPU_MASK_ALL, .vector = IRQ10_VECTOR, },
76 [11] = { .domain = CPU_MASK_ALL, .vector = IRQ11_VECTOR, },
77 [12] = { .domain = CPU_MASK_ALL, .vector = IRQ12_VECTOR, },
78 [13] = { .domain = CPU_MASK_ALL, .vector = IRQ13_VECTOR, },
79 [14] = { .domain = CPU_MASK_ALL, .vector = IRQ14_VECTOR, },
80 [15] = { .domain = CPU_MASK_ALL, .vector = IRQ15_VECTOR, },
83 static int assign_irq_vector(int irq, cpumask_t mask);
85 #define __apicdebuginit __init
87 int sis_apic_bug; /* not actually supported, dummy for compile */
89 static int no_timer_check;
91 static int disable_timer_pin_1 __initdata;
93 int timer_over_8254 __initdata = 1;
95 /* Where if anywhere is the i8259 connect in external int mode */
96 static struct { int pin, apic; } ioapic_i8259 = { -1, -1 };
98 static DEFINE_SPINLOCK(ioapic_lock);
99 DEFINE_SPINLOCK(vector_lock);
102 * # of IRQ routing registers
104 int nr_ioapic_registers[MAX_IO_APICS];
106 /* I/O APIC entries */
107 struct mpc_config_ioapic mp_ioapics[MAX_IO_APICS];
110 /* MP IRQ source entries */
111 struct mpc_config_intsrc mp_irqs[MAX_IRQ_SOURCES];
113 /* # of MP IRQ source entries */
117 * Rough estimation of how many shared IRQs there are, can
118 * be changed anytime.
120 #define MAX_PLUS_SHARED_IRQS NR_IRQS
121 #define PIN_MAP_SIZE (MAX_PLUS_SHARED_IRQS + NR_IRQS)
124 * This is performance-critical, we want to do it O(1)
126 * the indexing order of this array favors 1:1 mappings
127 * between pins and IRQs.
130 static struct irq_pin_list {
131 short apic, pin, next;
132 } irq_2_pin[PIN_MAP_SIZE];
136 unsigned int unused[3];
140 static __attribute_const__ struct io_apic __iomem *io_apic_base(int idx)
142 return (void __iomem *) __fix_to_virt(FIX_IO_APIC_BASE_0 + idx)
143 + (mp_ioapics[idx].mpc_apicaddr & ~PAGE_MASK);
146 static inline unsigned int io_apic_read(unsigned int apic, unsigned int reg)
148 struct io_apic __iomem *io_apic = io_apic_base(apic);
149 writel(reg, &io_apic->index);
150 return readl(&io_apic->data);
153 static inline void io_apic_write(unsigned int apic, unsigned int reg, unsigned int value)
155 struct io_apic __iomem *io_apic = io_apic_base(apic);
156 writel(reg, &io_apic->index);
157 writel(value, &io_apic->data);
161 * Re-write a value: to be used for read-modify-write
162 * cycles where the read already set up the index register.
164 static inline void io_apic_modify(unsigned int apic, unsigned int value)
166 struct io_apic __iomem *io_apic = io_apic_base(apic);
167 writel(value, &io_apic->data);
170 static bool io_apic_level_ack_pending(unsigned int irq)
172 struct irq_pin_list *entry;
175 spin_lock_irqsave(&ioapic_lock, flags);
176 entry = irq_2_pin + irq;
184 reg = io_apic_read(entry->apic, 0x10 + pin*2);
185 /* Is the remote IRR bit set? */
186 if ((reg >> 14) & 1) {
187 spin_unlock_irqrestore(&ioapic_lock, flags);
192 entry = irq_2_pin + entry->next;
194 spin_unlock_irqrestore(&ioapic_lock, flags);
200 * Synchronize the IO-APIC and the CPU by doing
201 * a dummy read from the IO-APIC
203 static inline void io_apic_sync(unsigned int apic)
205 struct io_apic __iomem *io_apic = io_apic_base(apic);
206 readl(&io_apic->data);
209 #define __DO_ACTION(R, ACTION, FINAL) \
213 struct irq_pin_list *entry = irq_2_pin + irq; \
215 BUG_ON(irq >= NR_IRQS); \
221 reg = io_apic_read(entry->apic, 0x10 + R + pin*2); \
223 io_apic_modify(entry->apic, reg); \
227 entry = irq_2_pin + entry->next; \
232 struct { u32 w1, w2; };
233 struct IO_APIC_route_entry entry;
236 static struct IO_APIC_route_entry ioapic_read_entry(int apic, int pin)
238 union entry_union eu;
240 spin_lock_irqsave(&ioapic_lock, flags);
241 eu.w1 = io_apic_read(apic, 0x10 + 2 * pin);
242 eu.w2 = io_apic_read(apic, 0x11 + 2 * pin);
243 spin_unlock_irqrestore(&ioapic_lock, flags);
248 * When we write a new IO APIC routing entry, we need to write the high
249 * word first! If the mask bit in the low word is clear, we will enable
250 * the interrupt, and we need to make sure the entry is fully populated
251 * before that happens.
254 __ioapic_write_entry(int apic, int pin, struct IO_APIC_route_entry e)
256 union entry_union eu;
258 io_apic_write(apic, 0x11 + 2*pin, eu.w2);
259 io_apic_write(apic, 0x10 + 2*pin, eu.w1);
262 static void ioapic_write_entry(int apic, int pin, struct IO_APIC_route_entry e)
265 spin_lock_irqsave(&ioapic_lock, flags);
266 __ioapic_write_entry(apic, pin, e);
267 spin_unlock_irqrestore(&ioapic_lock, flags);
271 * When we mask an IO APIC routing entry, we need to write the low
272 * word first, in order to set the mask bit before we change the
275 static void ioapic_mask_entry(int apic, int pin)
278 union entry_union eu = { .entry.mask = 1 };
280 spin_lock_irqsave(&ioapic_lock, flags);
281 io_apic_write(apic, 0x10 + 2*pin, eu.w1);
282 io_apic_write(apic, 0x11 + 2*pin, eu.w2);
283 spin_unlock_irqrestore(&ioapic_lock, flags);
287 static void __target_IO_APIC_irq(unsigned int irq, unsigned int dest, u8 vector)
290 struct irq_pin_list *entry = irq_2_pin + irq;
292 BUG_ON(irq >= NR_IRQS);
299 io_apic_write(apic, 0x11 + pin*2, dest);
300 reg = io_apic_read(apic, 0x10 + pin*2);
303 io_apic_modify(apic, reg);
306 entry = irq_2_pin + entry->next;
310 static void set_ioapic_affinity_irq(unsigned int irq, cpumask_t mask)
312 struct irq_cfg *cfg = irq_cfg + irq;
317 cpus_and(tmp, mask, cpu_online_map);
321 if (assign_irq_vector(irq, mask))
324 cpus_and(tmp, cfg->domain, mask);
325 dest = cpu_mask_to_apicid(tmp);
328 * Only the high 8 bits are valid.
330 dest = SET_APIC_LOGICAL_ID(dest);
332 spin_lock_irqsave(&ioapic_lock, flags);
333 __target_IO_APIC_irq(irq, dest, cfg->vector);
334 irq_desc[irq].affinity = mask;
335 spin_unlock_irqrestore(&ioapic_lock, flags);
340 * The common case is 1:1 IRQ<->pin mappings. Sometimes there are
341 * shared ISA-space IRQs, so we have to support them. We are super
342 * fast in the common case, and fast for shared ISA-space IRQs.
344 static void add_pin_to_irq(unsigned int irq, int apic, int pin)
346 static int first_free_entry = NR_IRQS;
347 struct irq_pin_list *entry = irq_2_pin + irq;
349 BUG_ON(irq >= NR_IRQS);
351 entry = irq_2_pin + entry->next;
353 if (entry->pin != -1) {
354 entry->next = first_free_entry;
355 entry = irq_2_pin + entry->next;
356 if (++first_free_entry >= PIN_MAP_SIZE)
357 panic("io_apic.c: ran out of irq_2_pin entries!");
364 #define DO_ACTION(name,R,ACTION, FINAL) \
366 static void name##_IO_APIC_irq (unsigned int irq) \
367 __DO_ACTION(R, ACTION, FINAL)
369 DO_ACTION( __mask, 0, |= 0x00010000, io_apic_sync(entry->apic) )
371 DO_ACTION( __unmask, 0, &= 0xfffeffff, )
374 static void mask_IO_APIC_irq (unsigned int irq)
378 spin_lock_irqsave(&ioapic_lock, flags);
379 __mask_IO_APIC_irq(irq);
380 spin_unlock_irqrestore(&ioapic_lock, flags);
383 static void unmask_IO_APIC_irq (unsigned int irq)
387 spin_lock_irqsave(&ioapic_lock, flags);
388 __unmask_IO_APIC_irq(irq);
389 spin_unlock_irqrestore(&ioapic_lock, flags);
392 static void clear_IO_APIC_pin(unsigned int apic, unsigned int pin)
394 struct IO_APIC_route_entry entry;
396 /* Check delivery_mode to be sure we're not clearing an SMI pin */
397 entry = ioapic_read_entry(apic, pin);
398 if (entry.delivery_mode == dest_SMI)
401 * Disable it in the IO-APIC irq-routing table:
403 ioapic_mask_entry(apic, pin);
406 static void clear_IO_APIC (void)
410 for (apic = 0; apic < nr_ioapics; apic++)
411 for (pin = 0; pin < nr_ioapic_registers[apic]; pin++)
412 clear_IO_APIC_pin(apic, pin);
415 int skip_ioapic_setup;
418 static int __init parse_noapic(char *str)
420 disable_ioapic_setup();
423 early_param("noapic", parse_noapic);
425 /* Actually the next is obsolete, but keep it for paranoid reasons -AK */
426 static int __init disable_timer_pin_setup(char *arg)
428 disable_timer_pin_1 = 1;
431 __setup("disable_timer_pin_1", disable_timer_pin_setup);
433 static int __init setup_disable_8254_timer(char *s)
435 timer_over_8254 = -1;
438 static int __init setup_enable_8254_timer(char *s)
444 __setup("disable_8254_timer", setup_disable_8254_timer);
445 __setup("enable_8254_timer", setup_enable_8254_timer);
449 * Find the IRQ entry number of a certain pin.
451 static int find_irq_entry(int apic, int pin, int type)
455 for (i = 0; i < mp_irq_entries; i++)
456 if (mp_irqs[i].mpc_irqtype == type &&
457 (mp_irqs[i].mpc_dstapic == mp_ioapics[apic].mpc_apicid ||
458 mp_irqs[i].mpc_dstapic == MP_APIC_ALL) &&
459 mp_irqs[i].mpc_dstirq == pin)
466 * Find the pin to which IRQ[irq] (ISA) is connected
468 static int __init find_isa_irq_pin(int irq, int type)
472 for (i = 0; i < mp_irq_entries; i++) {
473 int lbus = mp_irqs[i].mpc_srcbus;
475 if (test_bit(lbus, mp_bus_not_pci) &&
476 (mp_irqs[i].mpc_irqtype == type) &&
477 (mp_irqs[i].mpc_srcbusirq == irq))
479 return mp_irqs[i].mpc_dstirq;
484 static int __init find_isa_irq_apic(int irq, int type)
488 for (i = 0; i < mp_irq_entries; i++) {
489 int lbus = mp_irqs[i].mpc_srcbus;
491 if (test_bit(lbus, mp_bus_not_pci) &&
492 (mp_irqs[i].mpc_irqtype == type) &&
493 (mp_irqs[i].mpc_srcbusirq == irq))
496 if (i < mp_irq_entries) {
498 for(apic = 0; apic < nr_ioapics; apic++) {
499 if (mp_ioapics[apic].mpc_apicid == mp_irqs[i].mpc_dstapic)
508 * Find a specific PCI IRQ entry.
509 * Not an __init, possibly needed by modules
511 static int pin_2_irq(int idx, int apic, int pin);
513 int IO_APIC_get_PCI_irq_vector(int bus, int slot, int pin)
515 int apic, i, best_guess = -1;
517 apic_printk(APIC_DEBUG, "querying PCI -> IRQ mapping bus:%d, slot:%d, pin:%d.\n",
519 if (mp_bus_id_to_pci_bus[bus] == -1) {
520 apic_printk(APIC_VERBOSE, "PCI BIOS passed nonexistent PCI bus %d!\n", bus);
523 for (i = 0; i < mp_irq_entries; i++) {
524 int lbus = mp_irqs[i].mpc_srcbus;
526 for (apic = 0; apic < nr_ioapics; apic++)
527 if (mp_ioapics[apic].mpc_apicid == mp_irqs[i].mpc_dstapic ||
528 mp_irqs[i].mpc_dstapic == MP_APIC_ALL)
531 if (!test_bit(lbus, mp_bus_not_pci) &&
532 !mp_irqs[i].mpc_irqtype &&
534 (slot == ((mp_irqs[i].mpc_srcbusirq >> 2) & 0x1f))) {
535 int irq = pin_2_irq(i,apic,mp_irqs[i].mpc_dstirq);
537 if (!(apic || IO_APIC_IRQ(irq)))
540 if (pin == (mp_irqs[i].mpc_srcbusirq & 3))
543 * Use the first all-but-pin matching entry as a
544 * best-guess fuzzy result for broken mptables.
550 BUG_ON(best_guess >= NR_IRQS);
554 /* ISA interrupts are always polarity zero edge triggered,
555 * when listed as conforming in the MP table. */
557 #define default_ISA_trigger(idx) (0)
558 #define default_ISA_polarity(idx) (0)
560 /* PCI interrupts are always polarity one level triggered,
561 * when listed as conforming in the MP table. */
563 #define default_PCI_trigger(idx) (1)
564 #define default_PCI_polarity(idx) (1)
566 static int MPBIOS_polarity(int idx)
568 int bus = mp_irqs[idx].mpc_srcbus;
572 * Determine IRQ line polarity (high active or low active):
574 switch (mp_irqs[idx].mpc_irqflag & 3)
576 case 0: /* conforms, ie. bus-type dependent polarity */
577 if (test_bit(bus, mp_bus_not_pci))
578 polarity = default_ISA_polarity(idx);
580 polarity = default_PCI_polarity(idx);
582 case 1: /* high active */
587 case 2: /* reserved */
589 printk(KERN_WARNING "broken BIOS!!\n");
593 case 3: /* low active */
598 default: /* invalid */
600 printk(KERN_WARNING "broken BIOS!!\n");
608 static int MPBIOS_trigger(int idx)
610 int bus = mp_irqs[idx].mpc_srcbus;
614 * Determine IRQ trigger mode (edge or level sensitive):
616 switch ((mp_irqs[idx].mpc_irqflag>>2) & 3)
618 case 0: /* conforms, ie. bus-type dependent */
619 if (test_bit(bus, mp_bus_not_pci))
620 trigger = default_ISA_trigger(idx);
622 trigger = default_PCI_trigger(idx);
629 case 2: /* reserved */
631 printk(KERN_WARNING "broken BIOS!!\n");
640 default: /* invalid */
642 printk(KERN_WARNING "broken BIOS!!\n");
650 static inline int irq_polarity(int idx)
652 return MPBIOS_polarity(idx);
655 static inline int irq_trigger(int idx)
657 return MPBIOS_trigger(idx);
660 static int pin_2_irq(int idx, int apic, int pin)
663 int bus = mp_irqs[idx].mpc_srcbus;
666 * Debugging check, we are in big trouble if this message pops up!
668 if (mp_irqs[idx].mpc_dstirq != pin)
669 printk(KERN_ERR "broken BIOS or MPTABLE parser, ayiee!!\n");
671 if (test_bit(bus, mp_bus_not_pci)) {
672 irq = mp_irqs[idx].mpc_srcbusirq;
675 * PCI IRQs are mapped in order
679 irq += nr_ioapic_registers[i++];
682 BUG_ON(irq >= NR_IRQS);
686 static int __assign_irq_vector(int irq, cpumask_t mask)
689 * NOTE! The local APIC isn't very good at handling
690 * multiple interrupts at the same interrupt level.
691 * As the interrupt level is determined by taking the
692 * vector number and shifting that right by 4, we
693 * want to spread these out a bit so that they don't
694 * all fall in the same interrupt level.
696 * Also, we've got to be careful not to trash gate
697 * 0x80, because int 0x80 is hm, kind of importantish. ;)
699 static int current_vector = FIRST_DEVICE_VECTOR, current_offset = 0;
700 unsigned int old_vector;
704 BUG_ON((unsigned)irq >= NR_IRQS);
707 /* Only try and allocate irqs on cpus that are present */
708 cpus_and(mask, mask, cpu_online_map);
710 if ((cfg->move_in_progress) || cfg->move_cleanup_count)
713 old_vector = cfg->vector;
716 cpus_and(tmp, cfg->domain, mask);
717 if (!cpus_empty(tmp))
721 for_each_cpu_mask(cpu, mask) {
722 cpumask_t domain, new_mask;
726 domain = vector_allocation_domain(cpu);
727 cpus_and(new_mask, domain, cpu_online_map);
729 vector = current_vector;
730 offset = current_offset;
733 if (vector >= FIRST_SYSTEM_VECTOR) {
734 /* If we run out of vectors on large boxen, must share them. */
735 offset = (offset + 1) % 8;
736 vector = FIRST_DEVICE_VECTOR + offset;
738 if (unlikely(current_vector == vector))
740 if (vector == IA32_SYSCALL_VECTOR)
742 for_each_cpu_mask(new_cpu, new_mask)
743 if (per_cpu(vector_irq, new_cpu)[vector] != -1)
746 current_vector = vector;
747 current_offset = offset;
749 cfg->move_in_progress = 1;
750 cfg->old_domain = cfg->domain;
752 for_each_cpu_mask(new_cpu, new_mask)
753 per_cpu(vector_irq, new_cpu)[vector] = irq;
754 cfg->vector = vector;
755 cfg->domain = domain;
761 static int assign_irq_vector(int irq, cpumask_t mask)
766 spin_lock_irqsave(&vector_lock, flags);
767 err = __assign_irq_vector(irq, mask);
768 spin_unlock_irqrestore(&vector_lock, flags);
772 static void __clear_irq_vector(int irq)
778 BUG_ON((unsigned)irq >= NR_IRQS);
780 BUG_ON(!cfg->vector);
782 vector = cfg->vector;
783 cpus_and(mask, cfg->domain, cpu_online_map);
784 for_each_cpu_mask(cpu, mask)
785 per_cpu(vector_irq, cpu)[vector] = -1;
788 cpus_clear(cfg->domain);
791 void __setup_vector_irq(int cpu)
793 /* Initialize vector_irq on a new cpu */
794 /* This function must be called with vector_lock held */
797 /* Mark the inuse vectors */
798 for (irq = 0; irq < NR_IRQS; ++irq) {
799 if (!cpu_isset(cpu, irq_cfg[irq].domain))
801 vector = irq_cfg[irq].vector;
802 per_cpu(vector_irq, cpu)[vector] = irq;
804 /* Mark the free vectors */
805 for (vector = 0; vector < NR_VECTORS; ++vector) {
806 irq = per_cpu(vector_irq, cpu)[vector];
809 if (!cpu_isset(cpu, irq_cfg[irq].domain))
810 per_cpu(vector_irq, cpu)[vector] = -1;
815 static struct irq_chip ioapic_chip;
817 static void ioapic_register_intr(int irq, unsigned long trigger)
820 irq_desc[irq].status |= IRQ_LEVEL;
821 set_irq_chip_and_handler_name(irq, &ioapic_chip,
822 handle_fasteoi_irq, "fasteoi");
824 irq_desc[irq].status &= ~IRQ_LEVEL;
825 set_irq_chip_and_handler_name(irq, &ioapic_chip,
826 handle_edge_irq, "edge");
830 static void setup_IO_APIC_irq(int apic, int pin, unsigned int irq,
831 int trigger, int polarity)
833 struct irq_cfg *cfg = irq_cfg + irq;
834 struct IO_APIC_route_entry entry;
837 if (!IO_APIC_IRQ(irq))
841 if (assign_irq_vector(irq, mask))
844 cpus_and(mask, cfg->domain, mask);
846 apic_printk(APIC_VERBOSE,KERN_DEBUG
847 "IOAPIC[%d]: Set routing entry (%d-%d -> 0x%x -> "
848 "IRQ %d Mode:%i Active:%i)\n",
849 apic, mp_ioapics[apic].mpc_apicid, pin, cfg->vector,
850 irq, trigger, polarity);
853 * add it to the IO-APIC irq-routing table:
855 memset(&entry,0,sizeof(entry));
857 entry.delivery_mode = INT_DELIVERY_MODE;
858 entry.dest_mode = INT_DEST_MODE;
859 entry.dest = cpu_mask_to_apicid(mask);
860 entry.mask = 0; /* enable IRQ */
861 entry.trigger = trigger;
862 entry.polarity = polarity;
863 entry.vector = cfg->vector;
865 /* Mask level triggered irqs.
866 * Use IRQ_DELAYED_DISABLE for edge triggered irqs.
871 ioapic_register_intr(irq, trigger);
873 disable_8259A_irq(irq);
875 ioapic_write_entry(apic, pin, entry);
878 static void __init setup_IO_APIC_irqs(void)
880 int apic, pin, idx, irq, first_notcon = 1;
882 apic_printk(APIC_VERBOSE, KERN_DEBUG "init IO_APIC IRQs\n");
884 for (apic = 0; apic < nr_ioapics; apic++) {
885 for (pin = 0; pin < nr_ioapic_registers[apic]; pin++) {
887 idx = find_irq_entry(apic,pin,mp_INT);
890 apic_printk(APIC_VERBOSE, KERN_DEBUG " IO-APIC (apicid-pin) %d-%d", mp_ioapics[apic].mpc_apicid, pin);
893 apic_printk(APIC_VERBOSE, ", %d-%d", mp_ioapics[apic].mpc_apicid, pin);
897 apic_printk(APIC_VERBOSE, " not connected.\n");
901 irq = pin_2_irq(idx, apic, pin);
902 add_pin_to_irq(irq, apic, pin);
904 setup_IO_APIC_irq(apic, pin, irq,
905 irq_trigger(idx), irq_polarity(idx));
910 apic_printk(APIC_VERBOSE, " not connected.\n");
914 * Set up the 8259A-master output pin as broadcast to all
917 static void __init setup_ExtINT_IRQ0_pin(unsigned int apic, unsigned int pin, int vector)
919 struct IO_APIC_route_entry entry;
921 memset(&entry, 0, sizeof(entry));
923 disable_8259A_irq(0);
926 apic_write(APIC_LVT0, APIC_LVT_MASKED | APIC_DM_EXTINT);
929 * We use logical delivery to get the timer IRQ
932 entry.dest_mode = INT_DEST_MODE;
933 entry.mask = 0; /* unmask IRQ now */
934 entry.dest = cpu_mask_to_apicid(TARGET_CPUS);
935 entry.delivery_mode = INT_DELIVERY_MODE;
938 entry.vector = vector;
941 * The timer IRQ doesn't have to know that behind the
942 * scene we have a 8259A-master in AEOI mode ...
944 set_irq_chip_and_handler_name(0, &ioapic_chip, handle_edge_irq, "edge");
947 * Add it to the IO-APIC irq-routing table:
949 ioapic_write_entry(apic, pin, entry);
954 void __apicdebuginit print_IO_APIC(void)
957 union IO_APIC_reg_00 reg_00;
958 union IO_APIC_reg_01 reg_01;
959 union IO_APIC_reg_02 reg_02;
962 if (apic_verbosity == APIC_QUIET)
965 printk(KERN_DEBUG "number of MP IRQ sources: %d.\n", mp_irq_entries);
966 for (i = 0; i < nr_ioapics; i++)
967 printk(KERN_DEBUG "number of IO-APIC #%d registers: %d.\n",
968 mp_ioapics[i].mpc_apicid, nr_ioapic_registers[i]);
971 * We are a bit conservative about what we expect. We have to
972 * know about every hardware change ASAP.
974 printk(KERN_INFO "testing the IO APIC.......................\n");
976 for (apic = 0; apic < nr_ioapics; apic++) {
978 spin_lock_irqsave(&ioapic_lock, flags);
979 reg_00.raw = io_apic_read(apic, 0);
980 reg_01.raw = io_apic_read(apic, 1);
981 if (reg_01.bits.version >= 0x10)
982 reg_02.raw = io_apic_read(apic, 2);
983 spin_unlock_irqrestore(&ioapic_lock, flags);
986 printk(KERN_DEBUG "IO APIC #%d......\n", mp_ioapics[apic].mpc_apicid);
987 printk(KERN_DEBUG ".... register #00: %08X\n", reg_00.raw);
988 printk(KERN_DEBUG "....... : physical APIC id: %02X\n", reg_00.bits.ID);
990 printk(KERN_DEBUG ".... register #01: %08X\n", *(int *)®_01);
991 printk(KERN_DEBUG "....... : max redirection entries: %04X\n", reg_01.bits.entries);
993 printk(KERN_DEBUG "....... : PRQ implemented: %X\n", reg_01.bits.PRQ);
994 printk(KERN_DEBUG "....... : IO APIC version: %04X\n", reg_01.bits.version);
996 if (reg_01.bits.version >= 0x10) {
997 printk(KERN_DEBUG ".... register #02: %08X\n", reg_02.raw);
998 printk(KERN_DEBUG "....... : arbitration: %02X\n", reg_02.bits.arbitration);
1001 printk(KERN_DEBUG ".... IRQ redirection table:\n");
1003 printk(KERN_DEBUG " NR Dst Mask Trig IRR Pol"
1004 " Stat Dmod Deli Vect: \n");
1006 for (i = 0; i <= reg_01.bits.entries; i++) {
1007 struct IO_APIC_route_entry entry;
1009 entry = ioapic_read_entry(apic, i);
1011 printk(KERN_DEBUG " %02x %03X ",
1016 printk("%1d %1d %1d %1d %1d %1d %1d %02X\n",
1021 entry.delivery_status,
1023 entry.delivery_mode,
1028 printk(KERN_DEBUG "IRQ to pin mappings:\n");
1029 for (i = 0; i < NR_IRQS; i++) {
1030 struct irq_pin_list *entry = irq_2_pin + i;
1033 printk(KERN_DEBUG "IRQ%d ", i);
1035 printk("-> %d:%d", entry->apic, entry->pin);
1038 entry = irq_2_pin + entry->next;
1043 printk(KERN_INFO ".................................... done.\n");
1050 static __apicdebuginit void print_APIC_bitfield (int base)
1055 if (apic_verbosity == APIC_QUIET)
1058 printk(KERN_DEBUG "0123456789abcdef0123456789abcdef\n" KERN_DEBUG);
1059 for (i = 0; i < 8; i++) {
1060 v = apic_read(base + i*0x10);
1061 for (j = 0; j < 32; j++) {
1071 void __apicdebuginit print_local_APIC(void * dummy)
1073 unsigned int v, ver, maxlvt;
1075 if (apic_verbosity == APIC_QUIET)
1078 printk("\n" KERN_DEBUG "printing local APIC contents on CPU#%d/%d:\n",
1079 smp_processor_id(), hard_smp_processor_id());
1080 printk(KERN_INFO "... APIC ID: %08x (%01x)\n", v, GET_APIC_ID(read_apic_id()));
1081 v = apic_read(APIC_LVR);
1082 printk(KERN_INFO "... APIC VERSION: %08x\n", v);
1083 ver = GET_APIC_VERSION(v);
1084 maxlvt = lapic_get_maxlvt();
1086 v = apic_read(APIC_TASKPRI);
1087 printk(KERN_DEBUG "... APIC TASKPRI: %08x (%02x)\n", v, v & APIC_TPRI_MASK);
1089 v = apic_read(APIC_ARBPRI);
1090 printk(KERN_DEBUG "... APIC ARBPRI: %08x (%02x)\n", v,
1091 v & APIC_ARBPRI_MASK);
1092 v = apic_read(APIC_PROCPRI);
1093 printk(KERN_DEBUG "... APIC PROCPRI: %08x\n", v);
1095 v = apic_read(APIC_EOI);
1096 printk(KERN_DEBUG "... APIC EOI: %08x\n", v);
1097 v = apic_read(APIC_RRR);
1098 printk(KERN_DEBUG "... APIC RRR: %08x\n", v);
1099 v = apic_read(APIC_LDR);
1100 printk(KERN_DEBUG "... APIC LDR: %08x\n", v);
1101 v = apic_read(APIC_DFR);
1102 printk(KERN_DEBUG "... APIC DFR: %08x\n", v);
1103 v = apic_read(APIC_SPIV);
1104 printk(KERN_DEBUG "... APIC SPIV: %08x\n", v);
1106 printk(KERN_DEBUG "... APIC ISR field:\n");
1107 print_APIC_bitfield(APIC_ISR);
1108 printk(KERN_DEBUG "... APIC TMR field:\n");
1109 print_APIC_bitfield(APIC_TMR);
1110 printk(KERN_DEBUG "... APIC IRR field:\n");
1111 print_APIC_bitfield(APIC_IRR);
1113 v = apic_read(APIC_ESR);
1114 printk(KERN_DEBUG "... APIC ESR: %08x\n", v);
1116 v = apic_read(APIC_ICR);
1117 printk(KERN_DEBUG "... APIC ICR: %08x\n", v);
1118 v = apic_read(APIC_ICR2);
1119 printk(KERN_DEBUG "... APIC ICR2: %08x\n", v);
1121 v = apic_read(APIC_LVTT);
1122 printk(KERN_DEBUG "... APIC LVTT: %08x\n", v);
1124 if (maxlvt > 3) { /* PC is LVT#4. */
1125 v = apic_read(APIC_LVTPC);
1126 printk(KERN_DEBUG "... APIC LVTPC: %08x\n", v);
1128 v = apic_read(APIC_LVT0);
1129 printk(KERN_DEBUG "... APIC LVT0: %08x\n", v);
1130 v = apic_read(APIC_LVT1);
1131 printk(KERN_DEBUG "... APIC LVT1: %08x\n", v);
1133 if (maxlvt > 2) { /* ERR is LVT#3. */
1134 v = apic_read(APIC_LVTERR);
1135 printk(KERN_DEBUG "... APIC LVTERR: %08x\n", v);
1138 v = apic_read(APIC_TMICT);
1139 printk(KERN_DEBUG "... APIC TMICT: %08x\n", v);
1140 v = apic_read(APIC_TMCCT);
1141 printk(KERN_DEBUG "... APIC TMCCT: %08x\n", v);
1142 v = apic_read(APIC_TDCR);
1143 printk(KERN_DEBUG "... APIC TDCR: %08x\n", v);
1147 void print_all_local_APICs (void)
1149 on_each_cpu(print_local_APIC, NULL, 1, 1);
1152 void __apicdebuginit print_PIC(void)
1155 unsigned long flags;
1157 if (apic_verbosity == APIC_QUIET)
1160 printk(KERN_DEBUG "\nprinting PIC contents\n");
1162 spin_lock_irqsave(&i8259A_lock, flags);
1164 v = inb(0xa1) << 8 | inb(0x21);
1165 printk(KERN_DEBUG "... PIC IMR: %04x\n", v);
1167 v = inb(0xa0) << 8 | inb(0x20);
1168 printk(KERN_DEBUG "... PIC IRR: %04x\n", v);
1172 v = inb(0xa0) << 8 | inb(0x20);
1176 spin_unlock_irqrestore(&i8259A_lock, flags);
1178 printk(KERN_DEBUG "... PIC ISR: %04x\n", v);
1180 v = inb(0x4d1) << 8 | inb(0x4d0);
1181 printk(KERN_DEBUG "... PIC ELCR: %04x\n", v);
1186 void __init enable_IO_APIC(void)
1188 union IO_APIC_reg_01 reg_01;
1189 int i8259_apic, i8259_pin;
1191 unsigned long flags;
1193 for (i = 0; i < PIN_MAP_SIZE; i++) {
1194 irq_2_pin[i].pin = -1;
1195 irq_2_pin[i].next = 0;
1199 * The number of IO-APIC IRQ registers (== #pins):
1201 for (apic = 0; apic < nr_ioapics; apic++) {
1202 spin_lock_irqsave(&ioapic_lock, flags);
1203 reg_01.raw = io_apic_read(apic, 1);
1204 spin_unlock_irqrestore(&ioapic_lock, flags);
1205 nr_ioapic_registers[apic] = reg_01.bits.entries+1;
1207 for(apic = 0; apic < nr_ioapics; apic++) {
1209 /* See if any of the pins is in ExtINT mode */
1210 for (pin = 0; pin < nr_ioapic_registers[apic]; pin++) {
1211 struct IO_APIC_route_entry entry;
1212 entry = ioapic_read_entry(apic, pin);
1214 /* If the interrupt line is enabled and in ExtInt mode
1215 * I have found the pin where the i8259 is connected.
1217 if ((entry.mask == 0) && (entry.delivery_mode == dest_ExtINT)) {
1218 ioapic_i8259.apic = apic;
1219 ioapic_i8259.pin = pin;
1225 /* Look to see what if the MP table has reported the ExtINT */
1226 i8259_pin = find_isa_irq_pin(0, mp_ExtINT);
1227 i8259_apic = find_isa_irq_apic(0, mp_ExtINT);
1228 /* Trust the MP table if nothing is setup in the hardware */
1229 if ((ioapic_i8259.pin == -1) && (i8259_pin >= 0)) {
1230 printk(KERN_WARNING "ExtINT not setup in hardware but reported by MP table\n");
1231 ioapic_i8259.pin = i8259_pin;
1232 ioapic_i8259.apic = i8259_apic;
1234 /* Complain if the MP table and the hardware disagree */
1235 if (((ioapic_i8259.apic != i8259_apic) || (ioapic_i8259.pin != i8259_pin)) &&
1236 (i8259_pin >= 0) && (ioapic_i8259.pin >= 0))
1238 printk(KERN_WARNING "ExtINT in hardware and MP table differ\n");
1242 * Do not trust the IO-APIC being empty at bootup
1248 * Not an __init, needed by the reboot code
1250 void disable_IO_APIC(void)
1253 * Clear the IO-APIC before rebooting:
1258 * If the i8259 is routed through an IOAPIC
1259 * Put that IOAPIC in virtual wire mode
1260 * so legacy interrupts can be delivered.
1262 if (ioapic_i8259.pin != -1) {
1263 struct IO_APIC_route_entry entry;
1265 memset(&entry, 0, sizeof(entry));
1266 entry.mask = 0; /* Enabled */
1267 entry.trigger = 0; /* Edge */
1269 entry.polarity = 0; /* High */
1270 entry.delivery_status = 0;
1271 entry.dest_mode = 0; /* Physical */
1272 entry.delivery_mode = dest_ExtINT; /* ExtInt */
1274 entry.dest = GET_APIC_ID(read_apic_id());
1277 * Add it to the IO-APIC irq-routing table:
1279 ioapic_write_entry(ioapic_i8259.apic, ioapic_i8259.pin, entry);
1282 disconnect_bsp_APIC(ioapic_i8259.pin != -1);
1286 * There is a nasty bug in some older SMP boards, their mptable lies
1287 * about the timer IRQ. We do the following to work around the situation:
1289 * - timer IRQ defaults to IO-APIC IRQ
1290 * - if this function detects that timer IRQs are defunct, then we fall
1291 * back to ISA timer IRQs
1293 static int __init timer_irq_works(void)
1295 unsigned long t1 = jiffies;
1296 unsigned long flags;
1298 local_save_flags(flags);
1300 /* Let ten ticks pass... */
1301 mdelay((10 * 1000) / HZ);
1302 local_irq_restore(flags);
1305 * Expect a few ticks at least, to be sure some possible
1306 * glue logic does not lock up after one or two first
1307 * ticks in a non-ExtINT mode. Also the local APIC
1308 * might have cached one ExtINT interrupt. Finally, at
1309 * least one tick may be lost due to delays.
1313 if (time_after(jiffies, t1 + 4))
1319 * In the SMP+IOAPIC case it might happen that there are an unspecified
1320 * number of pending IRQ events unhandled. These cases are very rare,
1321 * so we 'resend' these IRQs via IPIs, to the same CPU. It's much
1322 * better to do it this way as thus we do not have to be aware of
1323 * 'pending' interrupts in the IRQ path, except at this point.
1326 * Edge triggered needs to resend any interrupt
1327 * that was delayed but this is now handled in the device
1332 * Starting up a edge-triggered IO-APIC interrupt is
1333 * nasty - we need to make sure that we get the edge.
1334 * If it is already asserted for some reason, we need
1335 * return 1 to indicate that is was pending.
1337 * This is not complete - we should be able to fake
1338 * an edge even if it isn't on the 8259A...
1341 static unsigned int startup_ioapic_irq(unsigned int irq)
1343 int was_pending = 0;
1344 unsigned long flags;
1346 spin_lock_irqsave(&ioapic_lock, flags);
1348 disable_8259A_irq(irq);
1349 if (i8259A_irq_pending(irq))
1352 __unmask_IO_APIC_irq(irq);
1353 spin_unlock_irqrestore(&ioapic_lock, flags);
1358 static int ioapic_retrigger_irq(unsigned int irq)
1360 struct irq_cfg *cfg = &irq_cfg[irq];
1362 unsigned long flags;
1364 spin_lock_irqsave(&vector_lock, flags);
1365 mask = cpumask_of_cpu(first_cpu(cfg->domain));
1366 send_IPI_mask(mask, cfg->vector);
1367 spin_unlock_irqrestore(&vector_lock, flags);
1373 * Level and edge triggered IO-APIC interrupts need different handling,
1374 * so we use two separate IRQ descriptors. Edge triggered IRQs can be
1375 * handled with the level-triggered descriptor, but that one has slightly
1376 * more overhead. Level-triggered interrupts cannot be handled with the
1377 * edge-triggered handler, without risking IRQ storms and other ugly
1382 asmlinkage void smp_irq_move_cleanup_interrupt(void)
1384 unsigned vector, me;
1389 me = smp_processor_id();
1390 for (vector = FIRST_EXTERNAL_VECTOR; vector < NR_VECTORS; vector++) {
1392 struct irq_desc *desc;
1393 struct irq_cfg *cfg;
1394 irq = __get_cpu_var(vector_irq)[vector];
1398 desc = irq_desc + irq;
1399 cfg = irq_cfg + irq;
1400 spin_lock(&desc->lock);
1401 if (!cfg->move_cleanup_count)
1404 if ((vector == cfg->vector) && cpu_isset(me, cfg->domain))
1407 __get_cpu_var(vector_irq)[vector] = -1;
1408 cfg->move_cleanup_count--;
1410 spin_unlock(&desc->lock);
1416 static void irq_complete_move(unsigned int irq)
1418 struct irq_cfg *cfg = irq_cfg + irq;
1419 unsigned vector, me;
1421 if (likely(!cfg->move_in_progress))
1424 vector = ~get_irq_regs()->orig_ax;
1425 me = smp_processor_id();
1426 if ((vector == cfg->vector) && cpu_isset(me, cfg->domain)) {
1427 cpumask_t cleanup_mask;
1429 cpus_and(cleanup_mask, cfg->old_domain, cpu_online_map);
1430 cfg->move_cleanup_count = cpus_weight(cleanup_mask);
1431 send_IPI_mask(cleanup_mask, IRQ_MOVE_CLEANUP_VECTOR);
1432 cfg->move_in_progress = 0;
1436 static inline void irq_complete_move(unsigned int irq) {}
1439 static void ack_apic_edge(unsigned int irq)
1441 irq_complete_move(irq);
1442 move_native_irq(irq);
1446 static void ack_apic_level(unsigned int irq)
1448 int do_unmask_irq = 0;
1450 irq_complete_move(irq);
1451 #ifdef CONFIG_GENERIC_PENDING_IRQ
1452 /* If we are moving the irq we need to mask it */
1453 if (unlikely(irq_desc[irq].status & IRQ_MOVE_PENDING)) {
1455 mask_IO_APIC_irq(irq);
1460 * We must acknowledge the irq before we move it or the acknowledge will
1461 * not propagate properly.
1465 /* Now we can move and renable the irq */
1466 if (unlikely(do_unmask_irq)) {
1467 /* Only migrate the irq if the ack has been received.
1469 * On rare occasions the broadcast level triggered ack gets
1470 * delayed going to ioapics, and if we reprogram the
1471 * vector while Remote IRR is still set the irq will never
1474 * To prevent this scenario we read the Remote IRR bit
1475 * of the ioapic. This has two effects.
1476 * - On any sane system the read of the ioapic will
1477 * flush writes (and acks) going to the ioapic from
1479 * - We get to see if the ACK has actually been delivered.
1481 * Based on failed experiments of reprogramming the
1482 * ioapic entry from outside of irq context starting
1483 * with masking the ioapic entry and then polling until
1484 * Remote IRR was clear before reprogramming the
1485 * ioapic I don't trust the Remote IRR bit to be
1486 * completey accurate.
1488 * However there appears to be no other way to plug
1489 * this race, so if the Remote IRR bit is not
1490 * accurate and is causing problems then it is a hardware bug
1491 * and you can go talk to the chipset vendor about it.
1493 if (!io_apic_level_ack_pending(irq))
1494 move_masked_irq(irq);
1495 unmask_IO_APIC_irq(irq);
1499 static struct irq_chip ioapic_chip __read_mostly = {
1501 .startup = startup_ioapic_irq,
1502 .mask = mask_IO_APIC_irq,
1503 .unmask = unmask_IO_APIC_irq,
1504 .ack = ack_apic_edge,
1505 .eoi = ack_apic_level,
1507 .set_affinity = set_ioapic_affinity_irq,
1509 .retrigger = ioapic_retrigger_irq,
1512 static inline void init_IO_APIC_traps(void)
1517 * NOTE! The local APIC isn't very good at handling
1518 * multiple interrupts at the same interrupt level.
1519 * As the interrupt level is determined by taking the
1520 * vector number and shifting that right by 4, we
1521 * want to spread these out a bit so that they don't
1522 * all fall in the same interrupt level.
1524 * Also, we've got to be careful not to trash gate
1525 * 0x80, because int 0x80 is hm, kind of importantish. ;)
1527 for (irq = 0; irq < NR_IRQS ; irq++) {
1528 if (IO_APIC_IRQ(irq) && !irq_cfg[irq].vector) {
1530 * Hmm.. We don't have an entry for this,
1531 * so default to an old-fashioned 8259
1532 * interrupt if we can..
1535 make_8259A_irq(irq);
1537 /* Strange. Oh, well.. */
1538 irq_desc[irq].chip = &no_irq_chip;
1543 static void enable_lapic_irq (unsigned int irq)
1547 v = apic_read(APIC_LVT0);
1548 apic_write(APIC_LVT0, v & ~APIC_LVT_MASKED);
1551 static void disable_lapic_irq (unsigned int irq)
1555 v = apic_read(APIC_LVT0);
1556 apic_write(APIC_LVT0, v | APIC_LVT_MASKED);
1559 static void ack_lapic_irq (unsigned int irq)
1564 static void end_lapic_irq (unsigned int i) { /* nothing */ }
1566 static struct hw_interrupt_type lapic_irq_type __read_mostly = {
1567 .name = "local-APIC",
1568 .typename = "local-APIC-edge",
1569 .startup = NULL, /* startup_irq() not used for IRQ0 */
1570 .shutdown = NULL, /* shutdown_irq() not used for IRQ0 */
1571 .enable = enable_lapic_irq,
1572 .disable = disable_lapic_irq,
1573 .ack = ack_lapic_irq,
1574 .end = end_lapic_irq,
1577 static void __init setup_nmi(void)
1580 * Dirty trick to enable the NMI watchdog ...
1581 * We put the 8259A master into AEOI mode and
1582 * unmask on all local APICs LVT0 as NMI.
1584 * The idea to use the 8259A in AEOI mode ('8259A Virtual Wire')
1585 * is from Maciej W. Rozycki - so we do not have to EOI from
1586 * the NMI handler or the timer interrupt.
1588 printk(KERN_INFO "activating NMI Watchdog ...");
1590 enable_NMI_through_LVT0();
1596 * This looks a bit hackish but it's about the only one way of sending
1597 * a few INTA cycles to 8259As and any associated glue logic. ICR does
1598 * not support the ExtINT mode, unfortunately. We need to send these
1599 * cycles as some i82489DX-based boards have glue logic that keeps the
1600 * 8259A interrupt line asserted until INTA. --macro
1602 static inline void unlock_ExtINT_logic(void)
1605 struct IO_APIC_route_entry entry0, entry1;
1606 unsigned char save_control, save_freq_select;
1608 pin = find_isa_irq_pin(8, mp_INT);
1609 apic = find_isa_irq_apic(8, mp_INT);
1613 entry0 = ioapic_read_entry(apic, pin);
1615 clear_IO_APIC_pin(apic, pin);
1617 memset(&entry1, 0, sizeof(entry1));
1619 entry1.dest_mode = 0; /* physical delivery */
1620 entry1.mask = 0; /* unmask IRQ now */
1621 entry1.dest = hard_smp_processor_id();
1622 entry1.delivery_mode = dest_ExtINT;
1623 entry1.polarity = entry0.polarity;
1627 ioapic_write_entry(apic, pin, entry1);
1629 save_control = CMOS_READ(RTC_CONTROL);
1630 save_freq_select = CMOS_READ(RTC_FREQ_SELECT);
1631 CMOS_WRITE((save_freq_select & ~RTC_RATE_SELECT) | 0x6,
1633 CMOS_WRITE(save_control | RTC_PIE, RTC_CONTROL);
1638 if ((CMOS_READ(RTC_INTR_FLAGS) & RTC_PF) == RTC_PF)
1642 CMOS_WRITE(save_control, RTC_CONTROL);
1643 CMOS_WRITE(save_freq_select, RTC_FREQ_SELECT);
1644 clear_IO_APIC_pin(apic, pin);
1646 ioapic_write_entry(apic, pin, entry0);
1650 * This code may look a bit paranoid, but it's supposed to cooperate with
1651 * a wide range of boards and BIOS bugs. Fortunately only the timer IRQ
1652 * is so screwy. Thanks to Brian Perkins for testing/hacking this beast
1653 * fanatically on his truly buggy board.
1655 * FIXME: really need to revamp this for modern platforms only.
1657 static inline void __init check_timer(void)
1659 struct irq_cfg *cfg = irq_cfg + 0;
1660 int apic1, pin1, apic2, pin2;
1661 unsigned long flags;
1663 local_irq_save(flags);
1666 * get/set the timer IRQ vector:
1668 disable_8259A_irq(0);
1669 assign_irq_vector(0, TARGET_CPUS);
1672 * Subtle, code in do_timer_interrupt() expects an AEOI
1673 * mode for the 8259A whenever interrupts are routed
1674 * through I/O APICs. Also IRQ0 has to be enabled in
1675 * the 8259A which implies the virtual wire has to be
1676 * disabled in the local APIC.
1678 apic_write(APIC_LVT0, APIC_LVT_MASKED | APIC_DM_EXTINT);
1680 if (timer_over_8254 > 0)
1681 enable_8259A_irq(0);
1683 pin1 = find_isa_irq_pin(0, mp_INT);
1684 apic1 = find_isa_irq_apic(0, mp_INT);
1685 pin2 = ioapic_i8259.pin;
1686 apic2 = ioapic_i8259.apic;
1688 apic_printk(APIC_VERBOSE,KERN_INFO "..TIMER: vector=0x%02X apic1=%d pin1=%d apic2=%d pin2=%d\n",
1689 cfg->vector, apic1, pin1, apic2, pin2);
1693 * Ok, does IRQ0 through the IOAPIC work?
1695 unmask_IO_APIC_irq(0);
1696 if (!no_timer_check && timer_irq_works()) {
1697 nmi_watchdog_default();
1698 if (nmi_watchdog == NMI_IO_APIC) {
1699 disable_8259A_irq(0);
1701 enable_8259A_irq(0);
1703 if (disable_timer_pin_1 > 0)
1704 clear_IO_APIC_pin(0, pin1);
1707 clear_IO_APIC_pin(apic1, pin1);
1708 apic_printk(APIC_QUIET,KERN_ERR "..MP-BIOS bug: 8254 timer not "
1709 "connected to IO-APIC\n");
1712 apic_printk(APIC_VERBOSE,KERN_INFO "...trying to set up timer (IRQ0) "
1713 "through the 8259A ... ");
1715 apic_printk(APIC_VERBOSE,"\n..... (found apic %d pin %d) ...",
1718 * legacy devices should be connected to IO APIC #0
1720 setup_ExtINT_IRQ0_pin(apic2, pin2, cfg->vector);
1721 if (timer_irq_works()) {
1722 apic_printk(APIC_VERBOSE," works.\n");
1723 nmi_watchdog_default();
1724 if (nmi_watchdog == NMI_IO_APIC) {
1730 * Cleanup, just in case ...
1732 clear_IO_APIC_pin(apic2, pin2);
1734 apic_printk(APIC_VERBOSE," failed.\n");
1736 if (nmi_watchdog == NMI_IO_APIC) {
1737 printk(KERN_WARNING "timer doesn't work through the IO-APIC - disabling NMI Watchdog!\n");
1741 apic_printk(APIC_VERBOSE, KERN_INFO "...trying to set up timer as Virtual Wire IRQ...");
1743 disable_8259A_irq(0);
1744 irq_desc[0].chip = &lapic_irq_type;
1745 apic_write(APIC_LVT0, APIC_DM_FIXED | cfg->vector); /* Fixed mode */
1746 enable_8259A_irq(0);
1748 if (timer_irq_works()) {
1749 apic_printk(APIC_VERBOSE," works.\n");
1752 apic_write(APIC_LVT0, APIC_LVT_MASKED | APIC_DM_FIXED | cfg->vector);
1753 apic_printk(APIC_VERBOSE," failed.\n");
1755 apic_printk(APIC_VERBOSE, KERN_INFO "...trying to set up timer as ExtINT IRQ...");
1759 apic_write(APIC_LVT0, APIC_DM_EXTINT);
1761 unlock_ExtINT_logic();
1763 if (timer_irq_works()) {
1764 apic_printk(APIC_VERBOSE," works.\n");
1767 apic_printk(APIC_VERBOSE," failed :(.\n");
1768 panic("IO-APIC + timer doesn't work! Try using the 'noapic' kernel parameter\n");
1770 local_irq_restore(flags);
1773 static int __init notimercheck(char *s)
1778 __setup("no_timer_check", notimercheck);
1782 * IRQs that are handled by the PIC in the MPS IOAPIC case.
1783 * - IRQ2 is the cascade IRQ, and cannot be a io-apic IRQ.
1784 * Linux doesn't really care, as it's not actually used
1785 * for any interrupt handling anyway.
1787 #define PIC_IRQS (1<<2)
1789 void __init setup_IO_APIC(void)
1793 * calling enable_IO_APIC() is moved to setup_local_APIC for BP
1797 io_apic_irqs = ~0; /* all IRQs go through IOAPIC */
1799 io_apic_irqs = ~PIC_IRQS;
1801 apic_printk(APIC_VERBOSE, "ENABLING IO-APIC IRQs\n");
1804 setup_IO_APIC_irqs();
1805 init_IO_APIC_traps();
1811 struct sysfs_ioapic_data {
1812 struct sys_device dev;
1813 struct IO_APIC_route_entry entry[0];
1815 static struct sysfs_ioapic_data * mp_ioapic_data[MAX_IO_APICS];
1817 static int ioapic_suspend(struct sys_device *dev, pm_message_t state)
1819 struct IO_APIC_route_entry *entry;
1820 struct sysfs_ioapic_data *data;
1823 data = container_of(dev, struct sysfs_ioapic_data, dev);
1824 entry = data->entry;
1825 for (i = 0; i < nr_ioapic_registers[dev->id]; i ++, entry ++ )
1826 *entry = ioapic_read_entry(dev->id, i);
1831 static int ioapic_resume(struct sys_device *dev)
1833 struct IO_APIC_route_entry *entry;
1834 struct sysfs_ioapic_data *data;
1835 unsigned long flags;
1836 union IO_APIC_reg_00 reg_00;
1839 data = container_of(dev, struct sysfs_ioapic_data, dev);
1840 entry = data->entry;
1842 spin_lock_irqsave(&ioapic_lock, flags);
1843 reg_00.raw = io_apic_read(dev->id, 0);
1844 if (reg_00.bits.ID != mp_ioapics[dev->id].mpc_apicid) {
1845 reg_00.bits.ID = mp_ioapics[dev->id].mpc_apicid;
1846 io_apic_write(dev->id, 0, reg_00.raw);
1848 spin_unlock_irqrestore(&ioapic_lock, flags);
1849 for (i = 0; i < nr_ioapic_registers[dev->id]; i++)
1850 ioapic_write_entry(dev->id, i, entry[i]);
1855 static struct sysdev_class ioapic_sysdev_class = {
1857 .suspend = ioapic_suspend,
1858 .resume = ioapic_resume,
1861 static int __init ioapic_init_sysfs(void)
1863 struct sys_device * dev;
1866 error = sysdev_class_register(&ioapic_sysdev_class);
1870 for (i = 0; i < nr_ioapics; i++ ) {
1871 size = sizeof(struct sys_device) + nr_ioapic_registers[i]
1872 * sizeof(struct IO_APIC_route_entry);
1873 mp_ioapic_data[i] = kzalloc(size, GFP_KERNEL);
1874 if (!mp_ioapic_data[i]) {
1875 printk(KERN_ERR "Can't suspend/resume IOAPIC %d\n", i);
1878 dev = &mp_ioapic_data[i]->dev;
1880 dev->cls = &ioapic_sysdev_class;
1881 error = sysdev_register(dev);
1883 kfree(mp_ioapic_data[i]);
1884 mp_ioapic_data[i] = NULL;
1885 printk(KERN_ERR "Can't suspend/resume IOAPIC %d\n", i);
1893 device_initcall(ioapic_init_sysfs);
1896 * Dynamic irq allocate and deallocation
1898 int create_irq(void)
1900 /* Allocate an unused irq */
1903 unsigned long flags;
1906 spin_lock_irqsave(&vector_lock, flags);
1907 for (new = (NR_IRQS - 1); new >= 0; new--) {
1908 if (platform_legacy_irq(new))
1910 if (irq_cfg[new].vector != 0)
1912 if (__assign_irq_vector(new, TARGET_CPUS) == 0)
1916 spin_unlock_irqrestore(&vector_lock, flags);
1919 dynamic_irq_init(irq);
1924 void destroy_irq(unsigned int irq)
1926 unsigned long flags;
1928 dynamic_irq_cleanup(irq);
1930 spin_lock_irqsave(&vector_lock, flags);
1931 __clear_irq_vector(irq);
1932 spin_unlock_irqrestore(&vector_lock, flags);
1936 * MSI message composition
1938 #ifdef CONFIG_PCI_MSI
1939 static int msi_compose_msg(struct pci_dev *pdev, unsigned int irq, struct msi_msg *msg)
1941 struct irq_cfg *cfg = irq_cfg + irq;
1947 err = assign_irq_vector(irq, tmp);
1949 cpus_and(tmp, cfg->domain, tmp);
1950 dest = cpu_mask_to_apicid(tmp);
1952 msg->address_hi = MSI_ADDR_BASE_HI;
1955 ((INT_DEST_MODE == 0) ?
1956 MSI_ADDR_DEST_MODE_PHYSICAL:
1957 MSI_ADDR_DEST_MODE_LOGICAL) |
1958 ((INT_DELIVERY_MODE != dest_LowestPrio) ?
1959 MSI_ADDR_REDIRECTION_CPU:
1960 MSI_ADDR_REDIRECTION_LOWPRI) |
1961 MSI_ADDR_DEST_ID(dest);
1964 MSI_DATA_TRIGGER_EDGE |
1965 MSI_DATA_LEVEL_ASSERT |
1966 ((INT_DELIVERY_MODE != dest_LowestPrio) ?
1967 MSI_DATA_DELIVERY_FIXED:
1968 MSI_DATA_DELIVERY_LOWPRI) |
1969 MSI_DATA_VECTOR(cfg->vector);
1975 static void set_msi_irq_affinity(unsigned int irq, cpumask_t mask)
1977 struct irq_cfg *cfg = irq_cfg + irq;
1982 cpus_and(tmp, mask, cpu_online_map);
1983 if (cpus_empty(tmp))
1986 if (assign_irq_vector(irq, mask))
1989 cpus_and(tmp, cfg->domain, mask);
1990 dest = cpu_mask_to_apicid(tmp);
1992 read_msi_msg(irq, &msg);
1994 msg.data &= ~MSI_DATA_VECTOR_MASK;
1995 msg.data |= MSI_DATA_VECTOR(cfg->vector);
1996 msg.address_lo &= ~MSI_ADDR_DEST_ID_MASK;
1997 msg.address_lo |= MSI_ADDR_DEST_ID(dest);
1999 write_msi_msg(irq, &msg);
2000 irq_desc[irq].affinity = mask;
2002 #endif /* CONFIG_SMP */
2005 * IRQ Chip for MSI PCI/PCI-X/PCI-Express Devices,
2006 * which implement the MSI or MSI-X Capability Structure.
2008 static struct irq_chip msi_chip = {
2010 .unmask = unmask_msi_irq,
2011 .mask = mask_msi_irq,
2012 .ack = ack_apic_edge,
2014 .set_affinity = set_msi_irq_affinity,
2016 .retrigger = ioapic_retrigger_irq,
2019 int arch_setup_msi_irq(struct pci_dev *dev, struct msi_desc *desc)
2027 ret = msi_compose_msg(dev, irq, &msg);
2033 set_irq_msi(irq, desc);
2034 write_msi_msg(irq, &msg);
2036 set_irq_chip_and_handler_name(irq, &msi_chip, handle_edge_irq, "edge");
2041 void arch_teardown_msi_irq(unsigned int irq)
2048 static void dmar_msi_set_affinity(unsigned int irq, cpumask_t mask)
2050 struct irq_cfg *cfg = irq_cfg + irq;
2055 cpus_and(tmp, mask, cpu_online_map);
2056 if (cpus_empty(tmp))
2059 if (assign_irq_vector(irq, mask))
2062 cpus_and(tmp, cfg->domain, mask);
2063 dest = cpu_mask_to_apicid(tmp);
2065 dmar_msi_read(irq, &msg);
2067 msg.data &= ~MSI_DATA_VECTOR_MASK;
2068 msg.data |= MSI_DATA_VECTOR(cfg->vector);
2069 msg.address_lo &= ~MSI_ADDR_DEST_ID_MASK;
2070 msg.address_lo |= MSI_ADDR_DEST_ID(dest);
2072 dmar_msi_write(irq, &msg);
2073 irq_desc[irq].affinity = mask;
2075 #endif /* CONFIG_SMP */
2077 struct irq_chip dmar_msi_type = {
2079 .unmask = dmar_msi_unmask,
2080 .mask = dmar_msi_mask,
2081 .ack = ack_apic_edge,
2083 .set_affinity = dmar_msi_set_affinity,
2085 .retrigger = ioapic_retrigger_irq,
2088 int arch_setup_dmar_msi(unsigned int irq)
2093 ret = msi_compose_msg(NULL, irq, &msg);
2096 dmar_msi_write(irq, &msg);
2097 set_irq_chip_and_handler_name(irq, &dmar_msi_type, handle_edge_irq,
2103 #endif /* CONFIG_PCI_MSI */
2105 * Hypertransport interrupt support
2107 #ifdef CONFIG_HT_IRQ
2111 static void target_ht_irq(unsigned int irq, unsigned int dest, u8 vector)
2113 struct ht_irq_msg msg;
2114 fetch_ht_irq_msg(irq, &msg);
2116 msg.address_lo &= ~(HT_IRQ_LOW_VECTOR_MASK | HT_IRQ_LOW_DEST_ID_MASK);
2117 msg.address_hi &= ~(HT_IRQ_HIGH_DEST_ID_MASK);
2119 msg.address_lo |= HT_IRQ_LOW_VECTOR(vector) | HT_IRQ_LOW_DEST_ID(dest);
2120 msg.address_hi |= HT_IRQ_HIGH_DEST_ID(dest);
2122 write_ht_irq_msg(irq, &msg);
2125 static void set_ht_irq_affinity(unsigned int irq, cpumask_t mask)
2127 struct irq_cfg *cfg = irq_cfg + irq;
2131 cpus_and(tmp, mask, cpu_online_map);
2132 if (cpus_empty(tmp))
2135 if (assign_irq_vector(irq, mask))
2138 cpus_and(tmp, cfg->domain, mask);
2139 dest = cpu_mask_to_apicid(tmp);
2141 target_ht_irq(irq, dest, cfg->vector);
2142 irq_desc[irq].affinity = mask;
2146 static struct irq_chip ht_irq_chip = {
2148 .mask = mask_ht_irq,
2149 .unmask = unmask_ht_irq,
2150 .ack = ack_apic_edge,
2152 .set_affinity = set_ht_irq_affinity,
2154 .retrigger = ioapic_retrigger_irq,
2157 int arch_setup_ht_irq(unsigned int irq, struct pci_dev *dev)
2159 struct irq_cfg *cfg = irq_cfg + irq;
2164 err = assign_irq_vector(irq, tmp);
2166 struct ht_irq_msg msg;
2169 cpus_and(tmp, cfg->domain, tmp);
2170 dest = cpu_mask_to_apicid(tmp);
2172 msg.address_hi = HT_IRQ_HIGH_DEST_ID(dest);
2176 HT_IRQ_LOW_DEST_ID(dest) |
2177 HT_IRQ_LOW_VECTOR(cfg->vector) |
2178 ((INT_DEST_MODE == 0) ?
2179 HT_IRQ_LOW_DM_PHYSICAL :
2180 HT_IRQ_LOW_DM_LOGICAL) |
2181 HT_IRQ_LOW_RQEOI_EDGE |
2182 ((INT_DELIVERY_MODE != dest_LowestPrio) ?
2183 HT_IRQ_LOW_MT_FIXED :
2184 HT_IRQ_LOW_MT_ARBITRATED) |
2185 HT_IRQ_LOW_IRQ_MASKED;
2187 write_ht_irq_msg(irq, &msg);
2189 set_irq_chip_and_handler_name(irq, &ht_irq_chip,
2190 handle_edge_irq, "edge");
2194 #endif /* CONFIG_HT_IRQ */
2196 /* --------------------------------------------------------------------------
2197 ACPI-based IOAPIC Configuration
2198 -------------------------------------------------------------------------- */
2202 #define IO_APIC_MAX_ID 0xFE
2204 int __init io_apic_get_redir_entries (int ioapic)
2206 union IO_APIC_reg_01 reg_01;
2207 unsigned long flags;
2209 spin_lock_irqsave(&ioapic_lock, flags);
2210 reg_01.raw = io_apic_read(ioapic, 1);
2211 spin_unlock_irqrestore(&ioapic_lock, flags);
2213 return reg_01.bits.entries;
2217 int io_apic_set_pci_routing (int ioapic, int pin, int irq, int triggering, int polarity)
2219 if (!IO_APIC_IRQ(irq)) {
2220 apic_printk(APIC_QUIET,KERN_ERR "IOAPIC[%d]: Invalid reference to IRQ 0\n",
2226 * IRQs < 16 are already in the irq_2_pin[] map
2229 add_pin_to_irq(irq, ioapic, pin);
2231 setup_IO_APIC_irq(ioapic, pin, irq, triggering, polarity);
2237 int acpi_get_override_irq(int bus_irq, int *trigger, int *polarity)
2241 if (skip_ioapic_setup)
2244 for (i = 0; i < mp_irq_entries; i++)
2245 if (mp_irqs[i].mpc_irqtype == mp_INT &&
2246 mp_irqs[i].mpc_srcbusirq == bus_irq)
2248 if (i >= mp_irq_entries)
2251 *trigger = irq_trigger(i);
2252 *polarity = irq_polarity(i);
2256 #endif /* CONFIG_ACPI */
2259 * This function currently is only a helper for the i386 smp boot process where
2260 * we need to reprogram the ioredtbls to cater for the cpus which have come online
2261 * so mask in all cases should simply be TARGET_CPUS
2264 void __init setup_ioapic_dest(void)
2266 int pin, ioapic, irq, irq_entry;
2268 if (skip_ioapic_setup == 1)
2271 for (ioapic = 0; ioapic < nr_ioapics; ioapic++) {
2272 for (pin = 0; pin < nr_ioapic_registers[ioapic]; pin++) {
2273 irq_entry = find_irq_entry(ioapic, pin, mp_INT);
2274 if (irq_entry == -1)
2276 irq = pin_2_irq(irq_entry, ioapic, pin);
2278 /* setup_IO_APIC_irqs could fail to get vector for some device
2279 * when you have too many devices, because at that time only boot
2282 if (!irq_cfg[irq].vector)
2283 setup_IO_APIC_irq(ioapic, pin, irq,
2284 irq_trigger(irq_entry),
2285 irq_polarity(irq_entry));
2287 set_ioapic_affinity_irq(irq, TARGET_CPUS);
2294 #define IOAPIC_RESOURCE_NAME_SIZE 11
2296 static struct resource *ioapic_resources;
2298 static struct resource * __init ioapic_setup_resources(void)
2301 struct resource *res;
2305 if (nr_ioapics <= 0)
2308 n = IOAPIC_RESOURCE_NAME_SIZE + sizeof(struct resource);
2311 mem = alloc_bootmem(n);
2315 mem += sizeof(struct resource) * nr_ioapics;
2317 for (i = 0; i < nr_ioapics; i++) {
2319 res[i].flags = IORESOURCE_MEM | IORESOURCE_BUSY;
2320 sprintf(mem, "IOAPIC %u", i);
2321 mem += IOAPIC_RESOURCE_NAME_SIZE;
2325 ioapic_resources = res;
2330 void __init ioapic_init_mappings(void)
2332 unsigned long ioapic_phys, idx = FIX_IO_APIC_BASE_0;
2333 struct resource *ioapic_res;
2336 ioapic_res = ioapic_setup_resources();
2337 for (i = 0; i < nr_ioapics; i++) {
2338 if (smp_found_config) {
2339 ioapic_phys = mp_ioapics[i].mpc_apicaddr;
2341 ioapic_phys = (unsigned long)
2342 alloc_bootmem_pages(PAGE_SIZE);
2343 ioapic_phys = __pa(ioapic_phys);
2345 set_fixmap_nocache(idx, ioapic_phys);
2346 apic_printk(APIC_VERBOSE,
2347 "mapped IOAPIC to %016lx (%016lx)\n",
2348 __fix_to_virt(idx), ioapic_phys);
2351 if (ioapic_res != NULL) {
2352 ioapic_res->start = ioapic_phys;
2353 ioapic_res->end = ioapic_phys + (4 * 1024) - 1;
2359 static int __init ioapic_insert_resources(void)
2362 struct resource *r = ioapic_resources;
2366 "IO APIC resources could be not be allocated.\n");
2370 for (i = 0; i < nr_ioapics; i++) {
2371 insert_resource(&iomem_resource, r);
2378 /* Insert the IO APIC resources after PCI initialization has occured to handle
2379 * IO APICS that are mapped in on a BAR in PCI space. */
2380 late_initcall(ioapic_insert_resources);