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>
36 #include <acpi/acpi_bus.h>
43 #include <asm/proto.h>
44 #include <asm/mach_apic.h>
48 #include <asm/msidef.h>
49 #include <asm/hypertransport.h>
54 unsigned move_cleanup_count;
56 u8 move_in_progress : 1;
59 /* irq_cfg is indexed by the sum of all RTEs in all I/O APICs. */
60 struct irq_cfg irq_cfg[NR_IRQS] __read_mostly = {
61 [0] = { .domain = CPU_MASK_ALL, .vector = IRQ0_VECTOR, },
62 [1] = { .domain = CPU_MASK_ALL, .vector = IRQ1_VECTOR, },
63 [2] = { .domain = CPU_MASK_ALL, .vector = IRQ2_VECTOR, },
64 [3] = { .domain = CPU_MASK_ALL, .vector = IRQ3_VECTOR, },
65 [4] = { .domain = CPU_MASK_ALL, .vector = IRQ4_VECTOR, },
66 [5] = { .domain = CPU_MASK_ALL, .vector = IRQ5_VECTOR, },
67 [6] = { .domain = CPU_MASK_ALL, .vector = IRQ6_VECTOR, },
68 [7] = { .domain = CPU_MASK_ALL, .vector = IRQ7_VECTOR, },
69 [8] = { .domain = CPU_MASK_ALL, .vector = IRQ8_VECTOR, },
70 [9] = { .domain = CPU_MASK_ALL, .vector = IRQ9_VECTOR, },
71 [10] = { .domain = CPU_MASK_ALL, .vector = IRQ10_VECTOR, },
72 [11] = { .domain = CPU_MASK_ALL, .vector = IRQ11_VECTOR, },
73 [12] = { .domain = CPU_MASK_ALL, .vector = IRQ12_VECTOR, },
74 [13] = { .domain = CPU_MASK_ALL, .vector = IRQ13_VECTOR, },
75 [14] = { .domain = CPU_MASK_ALL, .vector = IRQ14_VECTOR, },
76 [15] = { .domain = CPU_MASK_ALL, .vector = IRQ15_VECTOR, },
79 static int assign_irq_vector(int irq, cpumask_t mask);
81 #define __apicdebuginit __init
83 int sis_apic_bug; /* not actually supported, dummy for compile */
85 static int no_timer_check;
87 static int disable_timer_pin_1 __initdata;
89 int timer_over_8254 __initdata = 1;
91 /* Where if anywhere is the i8259 connect in external int mode */
92 static struct { int pin, apic; } ioapic_i8259 = { -1, -1 };
94 static DEFINE_SPINLOCK(ioapic_lock);
95 DEFINE_SPINLOCK(vector_lock);
98 * # of IRQ routing registers
100 int nr_ioapic_registers[MAX_IO_APICS];
103 * Rough estimation of how many shared IRQs there are, can
104 * be changed anytime.
106 #define MAX_PLUS_SHARED_IRQS NR_IRQS
107 #define PIN_MAP_SIZE (MAX_PLUS_SHARED_IRQS + NR_IRQS)
110 * This is performance-critical, we want to do it O(1)
112 * the indexing order of this array favors 1:1 mappings
113 * between pins and IRQs.
116 static struct irq_pin_list {
117 short apic, pin, next;
118 } irq_2_pin[PIN_MAP_SIZE];
122 unsigned int unused[3];
126 static __attribute_const__ struct io_apic __iomem *io_apic_base(int idx)
128 return (void __iomem *) __fix_to_virt(FIX_IO_APIC_BASE_0 + idx)
129 + (mp_ioapics[idx].mpc_apicaddr & ~PAGE_MASK);
132 static inline unsigned int io_apic_read(unsigned int apic, unsigned int reg)
134 struct io_apic __iomem *io_apic = io_apic_base(apic);
135 writel(reg, &io_apic->index);
136 return readl(&io_apic->data);
139 static inline void io_apic_write(unsigned int apic, unsigned int reg, unsigned int value)
141 struct io_apic __iomem *io_apic = io_apic_base(apic);
142 writel(reg, &io_apic->index);
143 writel(value, &io_apic->data);
147 * Re-write a value: to be used for read-modify-write
148 * cycles where the read already set up the index register.
150 static inline void io_apic_modify(unsigned int apic, unsigned int value)
152 struct io_apic __iomem *io_apic = io_apic_base(apic);
153 writel(value, &io_apic->data);
156 static int io_apic_level_ack_pending(unsigned int irq)
158 struct irq_pin_list *entry;
162 spin_lock_irqsave(&ioapic_lock, flags);
163 entry = irq_2_pin + irq;
171 reg = io_apic_read(entry->apic, 0x10 + pin*2);
172 /* Is the remote IRR bit set? */
173 pending |= (reg >> 14) & 1;
176 entry = irq_2_pin + entry->next;
178 spin_unlock_irqrestore(&ioapic_lock, flags);
183 * Synchronize the IO-APIC and the CPU by doing
184 * a dummy read from the IO-APIC
186 static inline void io_apic_sync(unsigned int apic)
188 struct io_apic __iomem *io_apic = io_apic_base(apic);
189 readl(&io_apic->data);
192 #define __DO_ACTION(R, ACTION, FINAL) \
196 struct irq_pin_list *entry = irq_2_pin + irq; \
198 BUG_ON(irq >= NR_IRQS); \
204 reg = io_apic_read(entry->apic, 0x10 + R + pin*2); \
206 io_apic_modify(entry->apic, reg); \
210 entry = irq_2_pin + entry->next; \
215 struct { u32 w1, w2; };
216 struct IO_APIC_route_entry entry;
219 static struct IO_APIC_route_entry ioapic_read_entry(int apic, int pin)
221 union entry_union eu;
223 spin_lock_irqsave(&ioapic_lock, flags);
224 eu.w1 = io_apic_read(apic, 0x10 + 2 * pin);
225 eu.w2 = io_apic_read(apic, 0x11 + 2 * pin);
226 spin_unlock_irqrestore(&ioapic_lock, flags);
231 * When we write a new IO APIC routing entry, we need to write the high
232 * word first! If the mask bit in the low word is clear, we will enable
233 * the interrupt, and we need to make sure the entry is fully populated
234 * before that happens.
237 __ioapic_write_entry(int apic, int pin, struct IO_APIC_route_entry e)
239 union entry_union eu;
241 io_apic_write(apic, 0x11 + 2*pin, eu.w2);
242 io_apic_write(apic, 0x10 + 2*pin, eu.w1);
245 static void ioapic_write_entry(int apic, int pin, struct IO_APIC_route_entry e)
248 spin_lock_irqsave(&ioapic_lock, flags);
249 __ioapic_write_entry(apic, pin, e);
250 spin_unlock_irqrestore(&ioapic_lock, flags);
254 * When we mask an IO APIC routing entry, we need to write the low
255 * word first, in order to set the mask bit before we change the
258 static void ioapic_mask_entry(int apic, int pin)
261 union entry_union eu = { .entry.mask = 1 };
263 spin_lock_irqsave(&ioapic_lock, flags);
264 io_apic_write(apic, 0x10 + 2*pin, eu.w1);
265 io_apic_write(apic, 0x11 + 2*pin, eu.w2);
266 spin_unlock_irqrestore(&ioapic_lock, flags);
270 static void __target_IO_APIC_irq(unsigned int irq, unsigned int dest, u8 vector)
273 struct irq_pin_list *entry = irq_2_pin + irq;
275 BUG_ON(irq >= NR_IRQS);
282 io_apic_write(apic, 0x11 + pin*2, dest);
283 reg = io_apic_read(apic, 0x10 + pin*2);
286 io_apic_modify(apic, reg);
289 entry = irq_2_pin + entry->next;
293 static void set_ioapic_affinity_irq(unsigned int irq, cpumask_t mask)
295 struct irq_cfg *cfg = irq_cfg + irq;
300 cpus_and(tmp, mask, cpu_online_map);
304 if (assign_irq_vector(irq, mask))
307 cpus_and(tmp, cfg->domain, mask);
308 dest = cpu_mask_to_apicid(tmp);
311 * Only the high 8 bits are valid.
313 dest = SET_APIC_LOGICAL_ID(dest);
315 spin_lock_irqsave(&ioapic_lock, flags);
316 __target_IO_APIC_irq(irq, dest, cfg->vector);
317 irq_desc[irq].affinity = mask;
318 spin_unlock_irqrestore(&ioapic_lock, flags);
323 * The common case is 1:1 IRQ<->pin mappings. Sometimes there are
324 * shared ISA-space IRQs, so we have to support them. We are super
325 * fast in the common case, and fast for shared ISA-space IRQs.
327 static void add_pin_to_irq(unsigned int irq, int apic, int pin)
329 static int first_free_entry = NR_IRQS;
330 struct irq_pin_list *entry = irq_2_pin + irq;
332 BUG_ON(irq >= NR_IRQS);
334 entry = irq_2_pin + entry->next;
336 if (entry->pin != -1) {
337 entry->next = first_free_entry;
338 entry = irq_2_pin + entry->next;
339 if (++first_free_entry >= PIN_MAP_SIZE)
340 panic("io_apic.c: ran out of irq_2_pin entries!");
347 #define DO_ACTION(name,R,ACTION, FINAL) \
349 static void name##_IO_APIC_irq (unsigned int irq) \
350 __DO_ACTION(R, ACTION, FINAL)
352 DO_ACTION( __mask, 0, |= 0x00010000, io_apic_sync(entry->apic) )
354 DO_ACTION( __unmask, 0, &= 0xfffeffff, )
357 static void mask_IO_APIC_irq (unsigned int irq)
361 spin_lock_irqsave(&ioapic_lock, flags);
362 __mask_IO_APIC_irq(irq);
363 spin_unlock_irqrestore(&ioapic_lock, flags);
366 static void unmask_IO_APIC_irq (unsigned int irq)
370 spin_lock_irqsave(&ioapic_lock, flags);
371 __unmask_IO_APIC_irq(irq);
372 spin_unlock_irqrestore(&ioapic_lock, flags);
375 static void clear_IO_APIC_pin(unsigned int apic, unsigned int pin)
377 struct IO_APIC_route_entry entry;
379 /* Check delivery_mode to be sure we're not clearing an SMI pin */
380 entry = ioapic_read_entry(apic, pin);
381 if (entry.delivery_mode == dest_SMI)
384 * Disable it in the IO-APIC irq-routing table:
386 ioapic_mask_entry(apic, pin);
389 static void clear_IO_APIC (void)
393 for (apic = 0; apic < nr_ioapics; apic++)
394 for (pin = 0; pin < nr_ioapic_registers[apic]; pin++)
395 clear_IO_APIC_pin(apic, pin);
398 int skip_ioapic_setup;
401 static int __init parse_noapic(char *str)
403 disable_ioapic_setup();
406 early_param("noapic", parse_noapic);
408 /* Actually the next is obsolete, but keep it for paranoid reasons -AK */
409 static int __init disable_timer_pin_setup(char *arg)
411 disable_timer_pin_1 = 1;
414 __setup("disable_timer_pin_1", disable_timer_pin_setup);
416 static int __init setup_disable_8254_timer(char *s)
418 timer_over_8254 = -1;
421 static int __init setup_enable_8254_timer(char *s)
427 __setup("disable_8254_timer", setup_disable_8254_timer);
428 __setup("enable_8254_timer", setup_enable_8254_timer);
432 * Find the IRQ entry number of a certain pin.
434 static int find_irq_entry(int apic, int pin, int type)
438 for (i = 0; i < mp_irq_entries; i++)
439 if (mp_irqs[i].mpc_irqtype == type &&
440 (mp_irqs[i].mpc_dstapic == mp_ioapics[apic].mpc_apicid ||
441 mp_irqs[i].mpc_dstapic == MP_APIC_ALL) &&
442 mp_irqs[i].mpc_dstirq == pin)
449 * Find the pin to which IRQ[irq] (ISA) is connected
451 static int __init find_isa_irq_pin(int irq, int type)
455 for (i = 0; i < mp_irq_entries; i++) {
456 int lbus = mp_irqs[i].mpc_srcbus;
458 if (test_bit(lbus, mp_bus_not_pci) &&
459 (mp_irqs[i].mpc_irqtype == type) &&
460 (mp_irqs[i].mpc_srcbusirq == irq))
462 return mp_irqs[i].mpc_dstirq;
467 static int __init find_isa_irq_apic(int irq, int type)
471 for (i = 0; i < mp_irq_entries; i++) {
472 int lbus = mp_irqs[i].mpc_srcbus;
474 if (test_bit(lbus, mp_bus_not_pci) &&
475 (mp_irqs[i].mpc_irqtype == type) &&
476 (mp_irqs[i].mpc_srcbusirq == irq))
479 if (i < mp_irq_entries) {
481 for(apic = 0; apic < nr_ioapics; apic++) {
482 if (mp_ioapics[apic].mpc_apicid == mp_irqs[i].mpc_dstapic)
491 * Find a specific PCI IRQ entry.
492 * Not an __init, possibly needed by modules
494 static int pin_2_irq(int idx, int apic, int pin);
496 int IO_APIC_get_PCI_irq_vector(int bus, int slot, int pin)
498 int apic, i, best_guess = -1;
500 apic_printk(APIC_DEBUG, "querying PCI -> IRQ mapping bus:%d, slot:%d, pin:%d.\n",
502 if (mp_bus_id_to_pci_bus[bus] == -1) {
503 apic_printk(APIC_VERBOSE, "PCI BIOS passed nonexistent PCI bus %d!\n", bus);
506 for (i = 0; i < mp_irq_entries; i++) {
507 int lbus = mp_irqs[i].mpc_srcbus;
509 for (apic = 0; apic < nr_ioapics; apic++)
510 if (mp_ioapics[apic].mpc_apicid == mp_irqs[i].mpc_dstapic ||
511 mp_irqs[i].mpc_dstapic == MP_APIC_ALL)
514 if (!test_bit(lbus, mp_bus_not_pci) &&
515 !mp_irqs[i].mpc_irqtype &&
517 (slot == ((mp_irqs[i].mpc_srcbusirq >> 2) & 0x1f))) {
518 int irq = pin_2_irq(i,apic,mp_irqs[i].mpc_dstirq);
520 if (!(apic || IO_APIC_IRQ(irq)))
523 if (pin == (mp_irqs[i].mpc_srcbusirq & 3))
526 * Use the first all-but-pin matching entry as a
527 * best-guess fuzzy result for broken mptables.
533 BUG_ON(best_guess >= NR_IRQS);
537 /* ISA interrupts are always polarity zero edge triggered,
538 * when listed as conforming in the MP table. */
540 #define default_ISA_trigger(idx) (0)
541 #define default_ISA_polarity(idx) (0)
543 /* PCI interrupts are always polarity one level triggered,
544 * when listed as conforming in the MP table. */
546 #define default_PCI_trigger(idx) (1)
547 #define default_PCI_polarity(idx) (1)
549 static int MPBIOS_polarity(int idx)
551 int bus = mp_irqs[idx].mpc_srcbus;
555 * Determine IRQ line polarity (high active or low active):
557 switch (mp_irqs[idx].mpc_irqflag & 3)
559 case 0: /* conforms, ie. bus-type dependent polarity */
560 if (test_bit(bus, mp_bus_not_pci))
561 polarity = default_ISA_polarity(idx);
563 polarity = default_PCI_polarity(idx);
565 case 1: /* high active */
570 case 2: /* reserved */
572 printk(KERN_WARNING "broken BIOS!!\n");
576 case 3: /* low active */
581 default: /* invalid */
583 printk(KERN_WARNING "broken BIOS!!\n");
591 static int MPBIOS_trigger(int idx)
593 int bus = mp_irqs[idx].mpc_srcbus;
597 * Determine IRQ trigger mode (edge or level sensitive):
599 switch ((mp_irqs[idx].mpc_irqflag>>2) & 3)
601 case 0: /* conforms, ie. bus-type dependent */
602 if (test_bit(bus, mp_bus_not_pci))
603 trigger = default_ISA_trigger(idx);
605 trigger = default_PCI_trigger(idx);
612 case 2: /* reserved */
614 printk(KERN_WARNING "broken BIOS!!\n");
623 default: /* invalid */
625 printk(KERN_WARNING "broken BIOS!!\n");
633 static inline int irq_polarity(int idx)
635 return MPBIOS_polarity(idx);
638 static inline int irq_trigger(int idx)
640 return MPBIOS_trigger(idx);
643 static int pin_2_irq(int idx, int apic, int pin)
646 int bus = mp_irqs[idx].mpc_srcbus;
649 * Debugging check, we are in big trouble if this message pops up!
651 if (mp_irqs[idx].mpc_dstirq != pin)
652 printk(KERN_ERR "broken BIOS or MPTABLE parser, ayiee!!\n");
654 if (test_bit(bus, mp_bus_not_pci)) {
655 irq = mp_irqs[idx].mpc_srcbusirq;
658 * PCI IRQs are mapped in order
662 irq += nr_ioapic_registers[i++];
665 BUG_ON(irq >= NR_IRQS);
669 static int __assign_irq_vector(int irq, cpumask_t mask)
672 * NOTE! The local APIC isn't very good at handling
673 * multiple interrupts at the same interrupt level.
674 * As the interrupt level is determined by taking the
675 * vector number and shifting that right by 4, we
676 * want to spread these out a bit so that they don't
677 * all fall in the same interrupt level.
679 * Also, we've got to be careful not to trash gate
680 * 0x80, because int 0x80 is hm, kind of importantish. ;)
682 static int current_vector = FIRST_DEVICE_VECTOR, current_offset = 0;
683 unsigned int old_vector;
687 BUG_ON((unsigned)irq >= NR_IRQS);
690 /* Only try and allocate irqs on cpus that are present */
691 cpus_and(mask, mask, cpu_online_map);
693 if ((cfg->move_in_progress) || cfg->move_cleanup_count)
696 old_vector = cfg->vector;
699 cpus_and(tmp, cfg->domain, mask);
700 if (!cpus_empty(tmp))
704 for_each_cpu_mask(cpu, mask) {
705 cpumask_t domain, new_mask;
709 domain = vector_allocation_domain(cpu);
710 cpus_and(new_mask, domain, cpu_online_map);
712 vector = current_vector;
713 offset = current_offset;
716 if (vector >= FIRST_SYSTEM_VECTOR) {
717 /* If we run out of vectors on large boxen, must share them. */
718 offset = (offset + 1) % 8;
719 vector = FIRST_DEVICE_VECTOR + offset;
721 if (unlikely(current_vector == vector))
723 if (vector == IA32_SYSCALL_VECTOR)
725 for_each_cpu_mask(new_cpu, new_mask)
726 if (per_cpu(vector_irq, new_cpu)[vector] != -1)
729 current_vector = vector;
730 current_offset = offset;
732 cfg->move_in_progress = 1;
733 cfg->old_domain = cfg->domain;
735 for_each_cpu_mask(new_cpu, new_mask)
736 per_cpu(vector_irq, new_cpu)[vector] = irq;
737 cfg->vector = vector;
738 cfg->domain = domain;
744 static int assign_irq_vector(int irq, cpumask_t mask)
749 spin_lock_irqsave(&vector_lock, flags);
750 err = __assign_irq_vector(irq, mask);
751 spin_unlock_irqrestore(&vector_lock, flags);
755 static void __clear_irq_vector(int irq)
761 BUG_ON((unsigned)irq >= NR_IRQS);
763 BUG_ON(!cfg->vector);
765 vector = cfg->vector;
766 cpus_and(mask, cfg->domain, cpu_online_map);
767 for_each_cpu_mask(cpu, mask)
768 per_cpu(vector_irq, cpu)[vector] = -1;
771 cfg->domain = CPU_MASK_NONE;
774 void __setup_vector_irq(int cpu)
776 /* Initialize vector_irq on a new cpu */
777 /* This function must be called with vector_lock held */
780 /* Mark the inuse vectors */
781 for (irq = 0; irq < NR_IRQS; ++irq) {
782 if (!cpu_isset(cpu, irq_cfg[irq].domain))
784 vector = irq_cfg[irq].vector;
785 per_cpu(vector_irq, cpu)[vector] = irq;
787 /* Mark the free vectors */
788 for (vector = 0; vector < NR_VECTORS; ++vector) {
789 irq = per_cpu(vector_irq, cpu)[vector];
792 if (!cpu_isset(cpu, irq_cfg[irq].domain))
793 per_cpu(vector_irq, cpu)[vector] = -1;
798 static struct irq_chip ioapic_chip;
800 static void ioapic_register_intr(int irq, unsigned long trigger)
803 irq_desc[irq].status |= IRQ_LEVEL;
804 set_irq_chip_and_handler_name(irq, &ioapic_chip,
805 handle_fasteoi_irq, "fasteoi");
807 irq_desc[irq].status &= ~IRQ_LEVEL;
808 set_irq_chip_and_handler_name(irq, &ioapic_chip,
809 handle_edge_irq, "edge");
813 static void setup_IO_APIC_irq(int apic, int pin, unsigned int irq,
814 int trigger, int polarity)
816 struct irq_cfg *cfg = irq_cfg + irq;
817 struct IO_APIC_route_entry entry;
820 if (!IO_APIC_IRQ(irq))
824 if (assign_irq_vector(irq, mask))
827 cpus_and(mask, cfg->domain, mask);
829 apic_printk(APIC_VERBOSE,KERN_DEBUG
830 "IOAPIC[%d]: Set routing entry (%d-%d -> 0x%x -> "
831 "IRQ %d Mode:%i Active:%i)\n",
832 apic, mp_ioapics[apic].mpc_apicid, pin, cfg->vector,
833 irq, trigger, polarity);
836 * add it to the IO-APIC irq-routing table:
838 memset(&entry,0,sizeof(entry));
840 entry.delivery_mode = INT_DELIVERY_MODE;
841 entry.dest_mode = INT_DEST_MODE;
842 entry.dest = cpu_mask_to_apicid(mask);
843 entry.mask = 0; /* enable IRQ */
844 entry.trigger = trigger;
845 entry.polarity = polarity;
846 entry.vector = cfg->vector;
848 /* Mask level triggered irqs.
849 * Use IRQ_DELAYED_DISABLE for edge triggered irqs.
854 ioapic_register_intr(irq, trigger);
856 disable_8259A_irq(irq);
858 ioapic_write_entry(apic, pin, entry);
861 static void __init setup_IO_APIC_irqs(void)
863 int apic, pin, idx, irq, first_notcon = 1;
865 apic_printk(APIC_VERBOSE, KERN_DEBUG "init IO_APIC IRQs\n");
867 for (apic = 0; apic < nr_ioapics; apic++) {
868 for (pin = 0; pin < nr_ioapic_registers[apic]; pin++) {
870 idx = find_irq_entry(apic,pin,mp_INT);
873 apic_printk(APIC_VERBOSE, KERN_DEBUG " IO-APIC (apicid-pin) %d-%d", mp_ioapics[apic].mpc_apicid, pin);
876 apic_printk(APIC_VERBOSE, ", %d-%d", mp_ioapics[apic].mpc_apicid, pin);
880 apic_printk(APIC_VERBOSE, " not connected.\n");
884 irq = pin_2_irq(idx, apic, pin);
885 add_pin_to_irq(irq, apic, pin);
887 setup_IO_APIC_irq(apic, pin, irq,
888 irq_trigger(idx), irq_polarity(idx));
893 apic_printk(APIC_VERBOSE, " not connected.\n");
897 * Set up the 8259A-master output pin as broadcast to all
900 static void __init setup_ExtINT_IRQ0_pin(unsigned int apic, unsigned int pin, int vector)
902 struct IO_APIC_route_entry entry;
905 memset(&entry,0,sizeof(entry));
907 disable_8259A_irq(0);
910 apic_write(APIC_LVT0, APIC_LVT_MASKED | APIC_DM_EXTINT);
913 * We use logical delivery to get the timer IRQ
916 entry.dest_mode = INT_DEST_MODE;
917 entry.mask = 0; /* unmask IRQ now */
918 entry.dest = cpu_mask_to_apicid(TARGET_CPUS);
919 entry.delivery_mode = INT_DELIVERY_MODE;
922 entry.vector = vector;
925 * The timer IRQ doesn't have to know that behind the
926 * scene we have a 8259A-master in AEOI mode ...
928 set_irq_chip_and_handler_name(0, &ioapic_chip, handle_edge_irq, "edge");
931 * Add it to the IO-APIC irq-routing table:
933 spin_lock_irqsave(&ioapic_lock, flags);
934 io_apic_write(apic, 0x11+2*pin, *(((int *)&entry)+1));
935 io_apic_write(apic, 0x10+2*pin, *(((int *)&entry)+0));
936 spin_unlock_irqrestore(&ioapic_lock, flags);
941 void __apicdebuginit print_IO_APIC(void)
944 union IO_APIC_reg_00 reg_00;
945 union IO_APIC_reg_01 reg_01;
946 union IO_APIC_reg_02 reg_02;
949 if (apic_verbosity == APIC_QUIET)
952 printk(KERN_DEBUG "number of MP IRQ sources: %d.\n", mp_irq_entries);
953 for (i = 0; i < nr_ioapics; i++)
954 printk(KERN_DEBUG "number of IO-APIC #%d registers: %d.\n",
955 mp_ioapics[i].mpc_apicid, nr_ioapic_registers[i]);
958 * We are a bit conservative about what we expect. We have to
959 * know about every hardware change ASAP.
961 printk(KERN_INFO "testing the IO APIC.......................\n");
963 for (apic = 0; apic < nr_ioapics; apic++) {
965 spin_lock_irqsave(&ioapic_lock, flags);
966 reg_00.raw = io_apic_read(apic, 0);
967 reg_01.raw = io_apic_read(apic, 1);
968 if (reg_01.bits.version >= 0x10)
969 reg_02.raw = io_apic_read(apic, 2);
970 spin_unlock_irqrestore(&ioapic_lock, flags);
973 printk(KERN_DEBUG "IO APIC #%d......\n", mp_ioapics[apic].mpc_apicid);
974 printk(KERN_DEBUG ".... register #00: %08X\n", reg_00.raw);
975 printk(KERN_DEBUG "....... : physical APIC id: %02X\n", reg_00.bits.ID);
977 printk(KERN_DEBUG ".... register #01: %08X\n", *(int *)®_01);
978 printk(KERN_DEBUG "....... : max redirection entries: %04X\n", reg_01.bits.entries);
980 printk(KERN_DEBUG "....... : PRQ implemented: %X\n", reg_01.bits.PRQ);
981 printk(KERN_DEBUG "....... : IO APIC version: %04X\n", reg_01.bits.version);
983 if (reg_01.bits.version >= 0x10) {
984 printk(KERN_DEBUG ".... register #02: %08X\n", reg_02.raw);
985 printk(KERN_DEBUG "....... : arbitration: %02X\n", reg_02.bits.arbitration);
988 printk(KERN_DEBUG ".... IRQ redirection table:\n");
990 printk(KERN_DEBUG " NR Dst Mask Trig IRR Pol"
991 " Stat Dmod Deli Vect: \n");
993 for (i = 0; i <= reg_01.bits.entries; i++) {
994 struct IO_APIC_route_entry entry;
996 entry = ioapic_read_entry(apic, i);
998 printk(KERN_DEBUG " %02x %03X ",
1003 printk("%1d %1d %1d %1d %1d %1d %1d %02X\n",
1008 entry.delivery_status,
1010 entry.delivery_mode,
1015 printk(KERN_DEBUG "IRQ to pin mappings:\n");
1016 for (i = 0; i < NR_IRQS; i++) {
1017 struct irq_pin_list *entry = irq_2_pin + i;
1020 printk(KERN_DEBUG "IRQ%d ", i);
1022 printk("-> %d:%d", entry->apic, entry->pin);
1025 entry = irq_2_pin + entry->next;
1030 printk(KERN_INFO ".................................... done.\n");
1037 static __apicdebuginit void print_APIC_bitfield (int base)
1042 if (apic_verbosity == APIC_QUIET)
1045 printk(KERN_DEBUG "0123456789abcdef0123456789abcdef\n" KERN_DEBUG);
1046 for (i = 0; i < 8; i++) {
1047 v = apic_read(base + i*0x10);
1048 for (j = 0; j < 32; j++) {
1058 void __apicdebuginit print_local_APIC(void * dummy)
1060 unsigned int v, ver, maxlvt;
1062 if (apic_verbosity == APIC_QUIET)
1065 printk("\n" KERN_DEBUG "printing local APIC contents on CPU#%d/%d:\n",
1066 smp_processor_id(), hard_smp_processor_id());
1067 v = apic_read(APIC_ID);
1068 printk(KERN_INFO "... APIC ID: %08x (%01x)\n", v, GET_APIC_ID(v));
1069 v = apic_read(APIC_LVR);
1070 printk(KERN_INFO "... APIC VERSION: %08x\n", v);
1071 ver = GET_APIC_VERSION(v);
1072 maxlvt = get_maxlvt();
1074 v = apic_read(APIC_TASKPRI);
1075 printk(KERN_DEBUG "... APIC TASKPRI: %08x (%02x)\n", v, v & APIC_TPRI_MASK);
1077 v = apic_read(APIC_ARBPRI);
1078 printk(KERN_DEBUG "... APIC ARBPRI: %08x (%02x)\n", v,
1079 v & APIC_ARBPRI_MASK);
1080 v = apic_read(APIC_PROCPRI);
1081 printk(KERN_DEBUG "... APIC PROCPRI: %08x\n", v);
1083 v = apic_read(APIC_EOI);
1084 printk(KERN_DEBUG "... APIC EOI: %08x\n", v);
1085 v = apic_read(APIC_RRR);
1086 printk(KERN_DEBUG "... APIC RRR: %08x\n", v);
1087 v = apic_read(APIC_LDR);
1088 printk(KERN_DEBUG "... APIC LDR: %08x\n", v);
1089 v = apic_read(APIC_DFR);
1090 printk(KERN_DEBUG "... APIC DFR: %08x\n", v);
1091 v = apic_read(APIC_SPIV);
1092 printk(KERN_DEBUG "... APIC SPIV: %08x\n", v);
1094 printk(KERN_DEBUG "... APIC ISR field:\n");
1095 print_APIC_bitfield(APIC_ISR);
1096 printk(KERN_DEBUG "... APIC TMR field:\n");
1097 print_APIC_bitfield(APIC_TMR);
1098 printk(KERN_DEBUG "... APIC IRR field:\n");
1099 print_APIC_bitfield(APIC_IRR);
1101 v = apic_read(APIC_ESR);
1102 printk(KERN_DEBUG "... APIC ESR: %08x\n", v);
1104 v = apic_read(APIC_ICR);
1105 printk(KERN_DEBUG "... APIC ICR: %08x\n", v);
1106 v = apic_read(APIC_ICR2);
1107 printk(KERN_DEBUG "... APIC ICR2: %08x\n", v);
1109 v = apic_read(APIC_LVTT);
1110 printk(KERN_DEBUG "... APIC LVTT: %08x\n", v);
1112 if (maxlvt > 3) { /* PC is LVT#4. */
1113 v = apic_read(APIC_LVTPC);
1114 printk(KERN_DEBUG "... APIC LVTPC: %08x\n", v);
1116 v = apic_read(APIC_LVT0);
1117 printk(KERN_DEBUG "... APIC LVT0: %08x\n", v);
1118 v = apic_read(APIC_LVT1);
1119 printk(KERN_DEBUG "... APIC LVT1: %08x\n", v);
1121 if (maxlvt > 2) { /* ERR is LVT#3. */
1122 v = apic_read(APIC_LVTERR);
1123 printk(KERN_DEBUG "... APIC LVTERR: %08x\n", v);
1126 v = apic_read(APIC_TMICT);
1127 printk(KERN_DEBUG "... APIC TMICT: %08x\n", v);
1128 v = apic_read(APIC_TMCCT);
1129 printk(KERN_DEBUG "... APIC TMCCT: %08x\n", v);
1130 v = apic_read(APIC_TDCR);
1131 printk(KERN_DEBUG "... APIC TDCR: %08x\n", v);
1135 void print_all_local_APICs (void)
1137 on_each_cpu(print_local_APIC, NULL, 1, 1);
1140 void __apicdebuginit print_PIC(void)
1143 unsigned long flags;
1145 if (apic_verbosity == APIC_QUIET)
1148 printk(KERN_DEBUG "\nprinting PIC contents\n");
1150 spin_lock_irqsave(&i8259A_lock, flags);
1152 v = inb(0xa1) << 8 | inb(0x21);
1153 printk(KERN_DEBUG "... PIC IMR: %04x\n", v);
1155 v = inb(0xa0) << 8 | inb(0x20);
1156 printk(KERN_DEBUG "... PIC IRR: %04x\n", v);
1160 v = inb(0xa0) << 8 | inb(0x20);
1164 spin_unlock_irqrestore(&i8259A_lock, flags);
1166 printk(KERN_DEBUG "... PIC ISR: %04x\n", v);
1168 v = inb(0x4d1) << 8 | inb(0x4d0);
1169 printk(KERN_DEBUG "... PIC ELCR: %04x\n", v);
1174 static void __init enable_IO_APIC(void)
1176 union IO_APIC_reg_01 reg_01;
1177 int i8259_apic, i8259_pin;
1179 unsigned long flags;
1181 for (i = 0; i < PIN_MAP_SIZE; i++) {
1182 irq_2_pin[i].pin = -1;
1183 irq_2_pin[i].next = 0;
1187 * The number of IO-APIC IRQ registers (== #pins):
1189 for (apic = 0; apic < nr_ioapics; apic++) {
1190 spin_lock_irqsave(&ioapic_lock, flags);
1191 reg_01.raw = io_apic_read(apic, 1);
1192 spin_unlock_irqrestore(&ioapic_lock, flags);
1193 nr_ioapic_registers[apic] = reg_01.bits.entries+1;
1195 for(apic = 0; apic < nr_ioapics; apic++) {
1197 /* See if any of the pins is in ExtINT mode */
1198 for (pin = 0; pin < nr_ioapic_registers[apic]; pin++) {
1199 struct IO_APIC_route_entry entry;
1200 entry = ioapic_read_entry(apic, pin);
1202 /* If the interrupt line is enabled and in ExtInt mode
1203 * I have found the pin where the i8259 is connected.
1205 if ((entry.mask == 0) && (entry.delivery_mode == dest_ExtINT)) {
1206 ioapic_i8259.apic = apic;
1207 ioapic_i8259.pin = pin;
1213 /* Look to see what if the MP table has reported the ExtINT */
1214 i8259_pin = find_isa_irq_pin(0, mp_ExtINT);
1215 i8259_apic = find_isa_irq_apic(0, mp_ExtINT);
1216 /* Trust the MP table if nothing is setup in the hardware */
1217 if ((ioapic_i8259.pin == -1) && (i8259_pin >= 0)) {
1218 printk(KERN_WARNING "ExtINT not setup in hardware but reported by MP table\n");
1219 ioapic_i8259.pin = i8259_pin;
1220 ioapic_i8259.apic = i8259_apic;
1222 /* Complain if the MP table and the hardware disagree */
1223 if (((ioapic_i8259.apic != i8259_apic) || (ioapic_i8259.pin != i8259_pin)) &&
1224 (i8259_pin >= 0) && (ioapic_i8259.pin >= 0))
1226 printk(KERN_WARNING "ExtINT in hardware and MP table differ\n");
1230 * Do not trust the IO-APIC being empty at bootup
1236 * Not an __init, needed by the reboot code
1238 void disable_IO_APIC(void)
1241 * Clear the IO-APIC before rebooting:
1246 * If the i8259 is routed through an IOAPIC
1247 * Put that IOAPIC in virtual wire mode
1248 * so legacy interrupts can be delivered.
1250 if (ioapic_i8259.pin != -1) {
1251 struct IO_APIC_route_entry entry;
1253 memset(&entry, 0, sizeof(entry));
1254 entry.mask = 0; /* Enabled */
1255 entry.trigger = 0; /* Edge */
1257 entry.polarity = 0; /* High */
1258 entry.delivery_status = 0;
1259 entry.dest_mode = 0; /* Physical */
1260 entry.delivery_mode = dest_ExtINT; /* ExtInt */
1262 entry.dest = GET_APIC_ID(apic_read(APIC_ID));
1265 * Add it to the IO-APIC irq-routing table:
1267 ioapic_write_entry(ioapic_i8259.apic, ioapic_i8259.pin, entry);
1270 disconnect_bsp_APIC(ioapic_i8259.pin != -1);
1274 * There is a nasty bug in some older SMP boards, their mptable lies
1275 * about the timer IRQ. We do the following to work around the situation:
1277 * - timer IRQ defaults to IO-APIC IRQ
1278 * - if this function detects that timer IRQs are defunct, then we fall
1279 * back to ISA timer IRQs
1281 static int __init timer_irq_works(void)
1283 unsigned long t1 = jiffies;
1286 /* Let ten ticks pass... */
1287 mdelay((10 * 1000) / HZ);
1290 * Expect a few ticks at least, to be sure some possible
1291 * glue logic does not lock up after one or two first
1292 * ticks in a non-ExtINT mode. Also the local APIC
1293 * might have cached one ExtINT interrupt. Finally, at
1294 * least one tick may be lost due to delays.
1298 if (jiffies - t1 > 4)
1304 * In the SMP+IOAPIC case it might happen that there are an unspecified
1305 * number of pending IRQ events unhandled. These cases are very rare,
1306 * so we 'resend' these IRQs via IPIs, to the same CPU. It's much
1307 * better to do it this way as thus we do not have to be aware of
1308 * 'pending' interrupts in the IRQ path, except at this point.
1311 * Edge triggered needs to resend any interrupt
1312 * that was delayed but this is now handled in the device
1317 * Starting up a edge-triggered IO-APIC interrupt is
1318 * nasty - we need to make sure that we get the edge.
1319 * If it is already asserted for some reason, we need
1320 * return 1 to indicate that is was pending.
1322 * This is not complete - we should be able to fake
1323 * an edge even if it isn't on the 8259A...
1326 static unsigned int startup_ioapic_irq(unsigned int irq)
1328 int was_pending = 0;
1329 unsigned long flags;
1331 spin_lock_irqsave(&ioapic_lock, flags);
1333 disable_8259A_irq(irq);
1334 if (i8259A_irq_pending(irq))
1337 __unmask_IO_APIC_irq(irq);
1338 spin_unlock_irqrestore(&ioapic_lock, flags);
1343 static int ioapic_retrigger_irq(unsigned int irq)
1345 struct irq_cfg *cfg = &irq_cfg[irq];
1347 unsigned long flags;
1349 spin_lock_irqsave(&vector_lock, flags);
1351 cpu_set(first_cpu(cfg->domain), mask);
1353 send_IPI_mask(mask, cfg->vector);
1354 spin_unlock_irqrestore(&vector_lock, flags);
1360 * Level and edge triggered IO-APIC interrupts need different handling,
1361 * so we use two separate IRQ descriptors. Edge triggered IRQs can be
1362 * handled with the level-triggered descriptor, but that one has slightly
1363 * more overhead. Level-triggered interrupts cannot be handled with the
1364 * edge-triggered handler, without risking IRQ storms and other ugly
1369 asmlinkage void smp_irq_move_cleanup_interrupt(void)
1371 unsigned vector, me;
1376 me = smp_processor_id();
1377 for (vector = FIRST_EXTERNAL_VECTOR; vector < NR_VECTORS; vector++) {
1379 struct irq_desc *desc;
1380 struct irq_cfg *cfg;
1381 irq = __get_cpu_var(vector_irq)[vector];
1385 desc = irq_desc + irq;
1386 cfg = irq_cfg + irq;
1387 spin_lock(&desc->lock);
1388 if (!cfg->move_cleanup_count)
1391 if ((vector == cfg->vector) && cpu_isset(me, cfg->domain))
1394 __get_cpu_var(vector_irq)[vector] = -1;
1395 cfg->move_cleanup_count--;
1397 spin_unlock(&desc->lock);
1403 static void irq_complete_move(unsigned int irq)
1405 struct irq_cfg *cfg = irq_cfg + irq;
1406 unsigned vector, me;
1408 if (likely(!cfg->move_in_progress))
1411 vector = ~get_irq_regs()->orig_rax;
1412 me = smp_processor_id();
1413 if ((vector == cfg->vector) && cpu_isset(me, cfg->domain)) {
1414 cpumask_t cleanup_mask;
1416 cpus_and(cleanup_mask, cfg->old_domain, cpu_online_map);
1417 cfg->move_cleanup_count = cpus_weight(cleanup_mask);
1418 send_IPI_mask(cleanup_mask, IRQ_MOVE_CLEANUP_VECTOR);
1419 cfg->move_in_progress = 0;
1423 static inline void irq_complete_move(unsigned int irq) {}
1426 static void ack_apic_edge(unsigned int irq)
1428 irq_complete_move(irq);
1429 move_native_irq(irq);
1433 static void ack_apic_level(unsigned int irq)
1435 int do_unmask_irq = 0;
1437 irq_complete_move(irq);
1438 #if defined(CONFIG_GENERIC_PENDING_IRQ) || defined(CONFIG_IRQBALANCE)
1439 /* If we are moving the irq we need to mask it */
1440 if (unlikely(irq_desc[irq].status & IRQ_MOVE_PENDING)) {
1442 mask_IO_APIC_irq(irq);
1447 * We must acknowledge the irq before we move it or the acknowledge will
1448 * not propagate properly.
1452 /* Now we can move and renable the irq */
1453 if (unlikely(do_unmask_irq)) {
1454 /* Only migrate the irq if the ack has been received.
1456 * On rare occasions the broadcast level triggered ack gets
1457 * delayed going to ioapics, and if we reprogram the
1458 * vector while Remote IRR is still set the irq will never
1461 * To prevent this scenario we read the Remote IRR bit
1462 * of the ioapic. This has two effects.
1463 * - On any sane system the read of the ioapic will
1464 * flush writes (and acks) going to the ioapic from
1466 * - We get to see if the ACK has actually been delivered.
1468 * Based on failed experiments of reprogramming the
1469 * ioapic entry from outside of irq context starting
1470 * with masking the ioapic entry and then polling until
1471 * Remote IRR was clear before reprogramming the
1472 * ioapic I don't trust the Remote IRR bit to be
1473 * completey accurate.
1475 * However there appears to be no other way to plug
1476 * this race, so if the Remote IRR bit is not
1477 * accurate and is causing problems then it is a hardware bug
1478 * and you can go talk to the chipset vendor about it.
1480 if (!io_apic_level_ack_pending(irq))
1481 move_masked_irq(irq);
1482 unmask_IO_APIC_irq(irq);
1486 static struct irq_chip ioapic_chip __read_mostly = {
1488 .startup = startup_ioapic_irq,
1489 .mask = mask_IO_APIC_irq,
1490 .unmask = unmask_IO_APIC_irq,
1491 .ack = ack_apic_edge,
1492 .eoi = ack_apic_level,
1494 .set_affinity = set_ioapic_affinity_irq,
1496 .retrigger = ioapic_retrigger_irq,
1499 static inline void init_IO_APIC_traps(void)
1504 * NOTE! The local APIC isn't very good at handling
1505 * multiple interrupts at the same interrupt level.
1506 * As the interrupt level is determined by taking the
1507 * vector number and shifting that right by 4, we
1508 * want to spread these out a bit so that they don't
1509 * all fall in the same interrupt level.
1511 * Also, we've got to be careful not to trash gate
1512 * 0x80, because int 0x80 is hm, kind of importantish. ;)
1514 for (irq = 0; irq < NR_IRQS ; irq++) {
1516 if (IO_APIC_IRQ(tmp) && !irq_cfg[tmp].vector) {
1518 * Hmm.. We don't have an entry for this,
1519 * so default to an old-fashioned 8259
1520 * interrupt if we can..
1523 make_8259A_irq(irq);
1525 /* Strange. Oh, well.. */
1526 irq_desc[irq].chip = &no_irq_chip;
1531 static void enable_lapic_irq (unsigned int irq)
1535 v = apic_read(APIC_LVT0);
1536 apic_write(APIC_LVT0, v & ~APIC_LVT_MASKED);
1539 static void disable_lapic_irq (unsigned int irq)
1543 v = apic_read(APIC_LVT0);
1544 apic_write(APIC_LVT0, v | APIC_LVT_MASKED);
1547 static void ack_lapic_irq (unsigned int irq)
1552 static void end_lapic_irq (unsigned int i) { /* nothing */ }
1554 static struct hw_interrupt_type lapic_irq_type __read_mostly = {
1555 .name = "local-APIC",
1556 .typename = "local-APIC-edge",
1557 .startup = NULL, /* startup_irq() not used for IRQ0 */
1558 .shutdown = NULL, /* shutdown_irq() not used for IRQ0 */
1559 .enable = enable_lapic_irq,
1560 .disable = disable_lapic_irq,
1561 .ack = ack_lapic_irq,
1562 .end = end_lapic_irq,
1565 static void setup_nmi (void)
1568 * Dirty trick to enable the NMI watchdog ...
1569 * We put the 8259A master into AEOI mode and
1570 * unmask on all local APICs LVT0 as NMI.
1572 * The idea to use the 8259A in AEOI mode ('8259A Virtual Wire')
1573 * is from Maciej W. Rozycki - so we do not have to EOI from
1574 * the NMI handler or the timer interrupt.
1576 printk(KERN_INFO "activating NMI Watchdog ...");
1578 enable_NMI_through_LVT0(NULL);
1584 * This looks a bit hackish but it's about the only one way of sending
1585 * a few INTA cycles to 8259As and any associated glue logic. ICR does
1586 * not support the ExtINT mode, unfortunately. We need to send these
1587 * cycles as some i82489DX-based boards have glue logic that keeps the
1588 * 8259A interrupt line asserted until INTA. --macro
1590 static inline void unlock_ExtINT_logic(void)
1593 struct IO_APIC_route_entry entry0, entry1;
1594 unsigned char save_control, save_freq_select;
1595 unsigned long flags;
1597 pin = find_isa_irq_pin(8, mp_INT);
1598 apic = find_isa_irq_apic(8, mp_INT);
1602 spin_lock_irqsave(&ioapic_lock, flags);
1603 *(((int *)&entry0) + 1) = io_apic_read(apic, 0x11 + 2 * pin);
1604 *(((int *)&entry0) + 0) = io_apic_read(apic, 0x10 + 2 * pin);
1605 spin_unlock_irqrestore(&ioapic_lock, flags);
1606 clear_IO_APIC_pin(apic, pin);
1608 memset(&entry1, 0, sizeof(entry1));
1610 entry1.dest_mode = 0; /* physical delivery */
1611 entry1.mask = 0; /* unmask IRQ now */
1612 entry1.dest = hard_smp_processor_id();
1613 entry1.delivery_mode = dest_ExtINT;
1614 entry1.polarity = entry0.polarity;
1618 spin_lock_irqsave(&ioapic_lock, flags);
1619 io_apic_write(apic, 0x11 + 2 * pin, *(((int *)&entry1) + 1));
1620 io_apic_write(apic, 0x10 + 2 * pin, *(((int *)&entry1) + 0));
1621 spin_unlock_irqrestore(&ioapic_lock, flags);
1623 save_control = CMOS_READ(RTC_CONTROL);
1624 save_freq_select = CMOS_READ(RTC_FREQ_SELECT);
1625 CMOS_WRITE((save_freq_select & ~RTC_RATE_SELECT) | 0x6,
1627 CMOS_WRITE(save_control | RTC_PIE, RTC_CONTROL);
1632 if ((CMOS_READ(RTC_INTR_FLAGS) & RTC_PF) == RTC_PF)
1636 CMOS_WRITE(save_control, RTC_CONTROL);
1637 CMOS_WRITE(save_freq_select, RTC_FREQ_SELECT);
1638 clear_IO_APIC_pin(apic, pin);
1640 spin_lock_irqsave(&ioapic_lock, flags);
1641 io_apic_write(apic, 0x11 + 2 * pin, *(((int *)&entry0) + 1));
1642 io_apic_write(apic, 0x10 + 2 * pin, *(((int *)&entry0) + 0));
1643 spin_unlock_irqrestore(&ioapic_lock, flags);
1647 * This code may look a bit paranoid, but it's supposed to cooperate with
1648 * a wide range of boards and BIOS bugs. Fortunately only the timer IRQ
1649 * is so screwy. Thanks to Brian Perkins for testing/hacking this beast
1650 * fanatically on his truly buggy board.
1652 * FIXME: really need to revamp this for modern platforms only.
1654 static inline void check_timer(void)
1656 struct irq_cfg *cfg = irq_cfg + 0;
1657 int apic1, pin1, apic2, pin2;
1660 * get/set the timer IRQ vector:
1662 disable_8259A_irq(0);
1663 assign_irq_vector(0, TARGET_CPUS);
1666 * Subtle, code in do_timer_interrupt() expects an AEOI
1667 * mode for the 8259A whenever interrupts are routed
1668 * through I/O APICs. Also IRQ0 has to be enabled in
1669 * the 8259A which implies the virtual wire has to be
1670 * disabled in the local APIC.
1672 apic_write(APIC_LVT0, APIC_LVT_MASKED | APIC_DM_EXTINT);
1674 if (timer_over_8254 > 0)
1675 enable_8259A_irq(0);
1677 pin1 = find_isa_irq_pin(0, mp_INT);
1678 apic1 = find_isa_irq_apic(0, mp_INT);
1679 pin2 = ioapic_i8259.pin;
1680 apic2 = ioapic_i8259.apic;
1682 apic_printk(APIC_VERBOSE,KERN_INFO "..TIMER: vector=0x%02X apic1=%d pin1=%d apic2=%d pin2=%d\n",
1683 cfg->vector, apic1, pin1, apic2, pin2);
1687 * Ok, does IRQ0 through the IOAPIC work?
1689 unmask_IO_APIC_irq(0);
1690 if (!no_timer_check && timer_irq_works()) {
1691 nmi_watchdog_default();
1692 if (nmi_watchdog == NMI_IO_APIC) {
1693 disable_8259A_irq(0);
1695 enable_8259A_irq(0);
1697 if (disable_timer_pin_1 > 0)
1698 clear_IO_APIC_pin(0, pin1);
1701 clear_IO_APIC_pin(apic1, pin1);
1702 apic_printk(APIC_QUIET,KERN_ERR "..MP-BIOS bug: 8254 timer not "
1703 "connected to IO-APIC\n");
1706 apic_printk(APIC_VERBOSE,KERN_INFO "...trying to set up timer (IRQ0) "
1707 "through the 8259A ... ");
1709 apic_printk(APIC_VERBOSE,"\n..... (found apic %d pin %d) ...",
1712 * legacy devices should be connected to IO APIC #0
1714 setup_ExtINT_IRQ0_pin(apic2, pin2, cfg->vector);
1715 if (timer_irq_works()) {
1716 apic_printk(APIC_VERBOSE," works.\n");
1717 nmi_watchdog_default();
1718 if (nmi_watchdog == NMI_IO_APIC) {
1724 * Cleanup, just in case ...
1726 clear_IO_APIC_pin(apic2, pin2);
1728 apic_printk(APIC_VERBOSE," failed.\n");
1730 if (nmi_watchdog == NMI_IO_APIC) {
1731 printk(KERN_WARNING "timer doesn't work through the IO-APIC - disabling NMI Watchdog!\n");
1735 apic_printk(APIC_VERBOSE, KERN_INFO "...trying to set up timer as Virtual Wire IRQ...");
1737 disable_8259A_irq(0);
1738 irq_desc[0].chip = &lapic_irq_type;
1739 apic_write(APIC_LVT0, APIC_DM_FIXED | cfg->vector); /* Fixed mode */
1740 enable_8259A_irq(0);
1742 if (timer_irq_works()) {
1743 apic_printk(APIC_VERBOSE," works.\n");
1746 apic_write(APIC_LVT0, APIC_LVT_MASKED | APIC_DM_FIXED | cfg->vector);
1747 apic_printk(APIC_VERBOSE," failed.\n");
1749 apic_printk(APIC_VERBOSE, KERN_INFO "...trying to set up timer as ExtINT IRQ...");
1753 apic_write(APIC_LVT0, APIC_DM_EXTINT);
1755 unlock_ExtINT_logic();
1757 if (timer_irq_works()) {
1758 apic_printk(APIC_VERBOSE," works.\n");
1761 apic_printk(APIC_VERBOSE," failed :(.\n");
1762 panic("IO-APIC + timer doesn't work! Try using the 'noapic' kernel parameter\n");
1765 static int __init notimercheck(char *s)
1770 __setup("no_timer_check", notimercheck);
1774 * IRQs that are handled by the PIC in the MPS IOAPIC case.
1775 * - IRQ2 is the cascade IRQ, and cannot be a io-apic IRQ.
1776 * Linux doesn't really care, as it's not actually used
1777 * for any interrupt handling anyway.
1779 #define PIC_IRQS (1<<2)
1781 void __init setup_IO_APIC(void)
1786 io_apic_irqs = ~0; /* all IRQs go through IOAPIC */
1788 io_apic_irqs = ~PIC_IRQS;
1790 apic_printk(APIC_VERBOSE, "ENABLING IO-APIC IRQs\n");
1793 setup_IO_APIC_irqs();
1794 init_IO_APIC_traps();
1800 struct sysfs_ioapic_data {
1801 struct sys_device dev;
1802 struct IO_APIC_route_entry entry[0];
1804 static struct sysfs_ioapic_data * mp_ioapic_data[MAX_IO_APICS];
1806 static int ioapic_suspend(struct sys_device *dev, pm_message_t state)
1808 struct IO_APIC_route_entry *entry;
1809 struct sysfs_ioapic_data *data;
1812 data = container_of(dev, struct sysfs_ioapic_data, dev);
1813 entry = data->entry;
1814 for (i = 0; i < nr_ioapic_registers[dev->id]; i ++, entry ++ )
1815 *entry = ioapic_read_entry(dev->id, i);
1820 static int ioapic_resume(struct sys_device *dev)
1822 struct IO_APIC_route_entry *entry;
1823 struct sysfs_ioapic_data *data;
1824 unsigned long flags;
1825 union IO_APIC_reg_00 reg_00;
1828 data = container_of(dev, struct sysfs_ioapic_data, dev);
1829 entry = data->entry;
1831 spin_lock_irqsave(&ioapic_lock, flags);
1832 reg_00.raw = io_apic_read(dev->id, 0);
1833 if (reg_00.bits.ID != mp_ioapics[dev->id].mpc_apicid) {
1834 reg_00.bits.ID = mp_ioapics[dev->id].mpc_apicid;
1835 io_apic_write(dev->id, 0, reg_00.raw);
1837 spin_unlock_irqrestore(&ioapic_lock, flags);
1838 for (i = 0; i < nr_ioapic_registers[dev->id]; i++)
1839 ioapic_write_entry(dev->id, i, entry[i]);
1844 static struct sysdev_class ioapic_sysdev_class = {
1845 set_kset_name("ioapic"),
1846 .suspend = ioapic_suspend,
1847 .resume = ioapic_resume,
1850 static int __init ioapic_init_sysfs(void)
1852 struct sys_device * dev;
1855 error = sysdev_class_register(&ioapic_sysdev_class);
1859 for (i = 0; i < nr_ioapics; i++ ) {
1860 size = sizeof(struct sys_device) + nr_ioapic_registers[i]
1861 * sizeof(struct IO_APIC_route_entry);
1862 mp_ioapic_data[i] = kzalloc(size, GFP_KERNEL);
1863 if (!mp_ioapic_data[i]) {
1864 printk(KERN_ERR "Can't suspend/resume IOAPIC %d\n", i);
1867 dev = &mp_ioapic_data[i]->dev;
1869 dev->cls = &ioapic_sysdev_class;
1870 error = sysdev_register(dev);
1872 kfree(mp_ioapic_data[i]);
1873 mp_ioapic_data[i] = NULL;
1874 printk(KERN_ERR "Can't suspend/resume IOAPIC %d\n", i);
1882 device_initcall(ioapic_init_sysfs);
1885 * Dynamic irq allocate and deallocation
1887 int create_irq(void)
1889 /* Allocate an unused irq */
1892 unsigned long flags;
1895 spin_lock_irqsave(&vector_lock, flags);
1896 for (new = (NR_IRQS - 1); new >= 0; new--) {
1897 if (platform_legacy_irq(new))
1899 if (irq_cfg[new].vector != 0)
1901 if (__assign_irq_vector(new, TARGET_CPUS) == 0)
1905 spin_unlock_irqrestore(&vector_lock, flags);
1908 dynamic_irq_init(irq);
1913 void destroy_irq(unsigned int irq)
1915 unsigned long flags;
1917 dynamic_irq_cleanup(irq);
1919 spin_lock_irqsave(&vector_lock, flags);
1920 __clear_irq_vector(irq);
1921 spin_unlock_irqrestore(&vector_lock, flags);
1925 * MSI message composition
1927 #ifdef CONFIG_PCI_MSI
1928 static int msi_compose_msg(struct pci_dev *pdev, unsigned int irq, struct msi_msg *msg)
1930 struct irq_cfg *cfg = irq_cfg + irq;
1936 err = assign_irq_vector(irq, tmp);
1938 cpus_and(tmp, cfg->domain, tmp);
1939 dest = cpu_mask_to_apicid(tmp);
1941 msg->address_hi = MSI_ADDR_BASE_HI;
1944 ((INT_DEST_MODE == 0) ?
1945 MSI_ADDR_DEST_MODE_PHYSICAL:
1946 MSI_ADDR_DEST_MODE_LOGICAL) |
1947 ((INT_DELIVERY_MODE != dest_LowestPrio) ?
1948 MSI_ADDR_REDIRECTION_CPU:
1949 MSI_ADDR_REDIRECTION_LOWPRI) |
1950 MSI_ADDR_DEST_ID(dest);
1953 MSI_DATA_TRIGGER_EDGE |
1954 MSI_DATA_LEVEL_ASSERT |
1955 ((INT_DELIVERY_MODE != dest_LowestPrio) ?
1956 MSI_DATA_DELIVERY_FIXED:
1957 MSI_DATA_DELIVERY_LOWPRI) |
1958 MSI_DATA_VECTOR(cfg->vector);
1964 static void set_msi_irq_affinity(unsigned int irq, cpumask_t mask)
1966 struct irq_cfg *cfg = irq_cfg + irq;
1971 cpus_and(tmp, mask, cpu_online_map);
1972 if (cpus_empty(tmp))
1975 if (assign_irq_vector(irq, mask))
1978 cpus_and(tmp, cfg->domain, mask);
1979 dest = cpu_mask_to_apicid(tmp);
1981 read_msi_msg(irq, &msg);
1983 msg.data &= ~MSI_DATA_VECTOR_MASK;
1984 msg.data |= MSI_DATA_VECTOR(cfg->vector);
1985 msg.address_lo &= ~MSI_ADDR_DEST_ID_MASK;
1986 msg.address_lo |= MSI_ADDR_DEST_ID(dest);
1988 write_msi_msg(irq, &msg);
1989 irq_desc[irq].affinity = mask;
1991 #endif /* CONFIG_SMP */
1994 * IRQ Chip for MSI PCI/PCI-X/PCI-Express Devices,
1995 * which implement the MSI or MSI-X Capability Structure.
1997 static struct irq_chip msi_chip = {
1999 .unmask = unmask_msi_irq,
2000 .mask = mask_msi_irq,
2001 .ack = ack_apic_edge,
2003 .set_affinity = set_msi_irq_affinity,
2005 .retrigger = ioapic_retrigger_irq,
2008 int arch_setup_msi_irq(struct pci_dev *dev, struct msi_desc *desc)
2016 ret = msi_compose_msg(dev, irq, &msg);
2022 set_irq_msi(irq, desc);
2023 write_msi_msg(irq, &msg);
2025 set_irq_chip_and_handler_name(irq, &msi_chip, handle_edge_irq, "edge");
2030 void arch_teardown_msi_irq(unsigned int irq)
2037 static void dmar_msi_set_affinity(unsigned int irq, cpumask_t mask)
2039 struct irq_cfg *cfg = irq_cfg + irq;
2044 cpus_and(tmp, mask, cpu_online_map);
2045 if (cpus_empty(tmp))
2048 if (assign_irq_vector(irq, mask))
2051 cpus_and(tmp, cfg->domain, mask);
2052 dest = cpu_mask_to_apicid(tmp);
2054 dmar_msi_read(irq, &msg);
2056 msg.data &= ~MSI_DATA_VECTOR_MASK;
2057 msg.data |= MSI_DATA_VECTOR(cfg->vector);
2058 msg.address_lo &= ~MSI_ADDR_DEST_ID_MASK;
2059 msg.address_lo |= MSI_ADDR_DEST_ID(dest);
2061 dmar_msi_write(irq, &msg);
2062 irq_desc[irq].affinity = mask;
2064 #endif /* CONFIG_SMP */
2066 struct irq_chip dmar_msi_type = {
2068 .unmask = dmar_msi_unmask,
2069 .mask = dmar_msi_mask,
2070 .ack = ack_apic_edge,
2072 .set_affinity = dmar_msi_set_affinity,
2074 .retrigger = ioapic_retrigger_irq,
2077 int arch_setup_dmar_msi(unsigned int irq)
2082 ret = msi_compose_msg(NULL, irq, &msg);
2085 dmar_msi_write(irq, &msg);
2086 set_irq_chip_and_handler_name(irq, &dmar_msi_type, handle_edge_irq,
2092 #endif /* CONFIG_PCI_MSI */
2094 * Hypertransport interrupt support
2096 #ifdef CONFIG_HT_IRQ
2100 static void target_ht_irq(unsigned int irq, unsigned int dest, u8 vector)
2102 struct ht_irq_msg msg;
2103 fetch_ht_irq_msg(irq, &msg);
2105 msg.address_lo &= ~(HT_IRQ_LOW_VECTOR_MASK | HT_IRQ_LOW_DEST_ID_MASK);
2106 msg.address_hi &= ~(HT_IRQ_HIGH_DEST_ID_MASK);
2108 msg.address_lo |= HT_IRQ_LOW_VECTOR(vector) | HT_IRQ_LOW_DEST_ID(dest);
2109 msg.address_hi |= HT_IRQ_HIGH_DEST_ID(dest);
2111 write_ht_irq_msg(irq, &msg);
2114 static void set_ht_irq_affinity(unsigned int irq, cpumask_t mask)
2116 struct irq_cfg *cfg = irq_cfg + irq;
2120 cpus_and(tmp, mask, cpu_online_map);
2121 if (cpus_empty(tmp))
2124 if (assign_irq_vector(irq, mask))
2127 cpus_and(tmp, cfg->domain, mask);
2128 dest = cpu_mask_to_apicid(tmp);
2130 target_ht_irq(irq, dest, cfg->vector);
2131 irq_desc[irq].affinity = mask;
2135 static struct irq_chip ht_irq_chip = {
2137 .mask = mask_ht_irq,
2138 .unmask = unmask_ht_irq,
2139 .ack = ack_apic_edge,
2141 .set_affinity = set_ht_irq_affinity,
2143 .retrigger = ioapic_retrigger_irq,
2146 int arch_setup_ht_irq(unsigned int irq, struct pci_dev *dev)
2148 struct irq_cfg *cfg = irq_cfg + irq;
2153 err = assign_irq_vector(irq, tmp);
2155 struct ht_irq_msg msg;
2158 cpus_and(tmp, cfg->domain, tmp);
2159 dest = cpu_mask_to_apicid(tmp);
2161 msg.address_hi = HT_IRQ_HIGH_DEST_ID(dest);
2165 HT_IRQ_LOW_DEST_ID(dest) |
2166 HT_IRQ_LOW_VECTOR(cfg->vector) |
2167 ((INT_DEST_MODE == 0) ?
2168 HT_IRQ_LOW_DM_PHYSICAL :
2169 HT_IRQ_LOW_DM_LOGICAL) |
2170 HT_IRQ_LOW_RQEOI_EDGE |
2171 ((INT_DELIVERY_MODE != dest_LowestPrio) ?
2172 HT_IRQ_LOW_MT_FIXED :
2173 HT_IRQ_LOW_MT_ARBITRATED) |
2174 HT_IRQ_LOW_IRQ_MASKED;
2176 write_ht_irq_msg(irq, &msg);
2178 set_irq_chip_and_handler_name(irq, &ht_irq_chip,
2179 handle_edge_irq, "edge");
2183 #endif /* CONFIG_HT_IRQ */
2185 /* --------------------------------------------------------------------------
2186 ACPI-based IOAPIC Configuration
2187 -------------------------------------------------------------------------- */
2191 #define IO_APIC_MAX_ID 0xFE
2193 int __init io_apic_get_redir_entries (int ioapic)
2195 union IO_APIC_reg_01 reg_01;
2196 unsigned long flags;
2198 spin_lock_irqsave(&ioapic_lock, flags);
2199 reg_01.raw = io_apic_read(ioapic, 1);
2200 spin_unlock_irqrestore(&ioapic_lock, flags);
2202 return reg_01.bits.entries;
2206 int io_apic_set_pci_routing (int ioapic, int pin, int irq, int triggering, int polarity)
2208 if (!IO_APIC_IRQ(irq)) {
2209 apic_printk(APIC_QUIET,KERN_ERR "IOAPIC[%d]: Invalid reference to IRQ 0\n",
2215 * IRQs < 16 are already in the irq_2_pin[] map
2218 add_pin_to_irq(irq, ioapic, pin);
2220 setup_IO_APIC_irq(ioapic, pin, irq, triggering, polarity);
2226 int acpi_get_override_irq(int bus_irq, int *trigger, int *polarity)
2230 if (skip_ioapic_setup)
2233 for (i = 0; i < mp_irq_entries; i++)
2234 if (mp_irqs[i].mpc_irqtype == mp_INT &&
2235 mp_irqs[i].mpc_srcbusirq == bus_irq)
2237 if (i >= mp_irq_entries)
2240 *trigger = irq_trigger(i);
2241 *polarity = irq_polarity(i);
2245 #endif /* CONFIG_ACPI */
2248 * This function currently is only a helper for the i386 smp boot process where
2249 * we need to reprogram the ioredtbls to cater for the cpus which have come online
2250 * so mask in all cases should simply be TARGET_CPUS
2253 void __init setup_ioapic_dest(void)
2255 int pin, ioapic, irq, irq_entry;
2257 if (skip_ioapic_setup == 1)
2260 for (ioapic = 0; ioapic < nr_ioapics; ioapic++) {
2261 for (pin = 0; pin < nr_ioapic_registers[ioapic]; pin++) {
2262 irq_entry = find_irq_entry(ioapic, pin, mp_INT);
2263 if (irq_entry == -1)
2265 irq = pin_2_irq(irq_entry, ioapic, pin);
2267 /* setup_IO_APIC_irqs could fail to get vector for some device
2268 * when you have too many devices, because at that time only boot
2271 if (!irq_cfg[irq].vector)
2272 setup_IO_APIC_irq(ioapic, pin, irq,
2273 irq_trigger(irq_entry),
2274 irq_polarity(irq_entry));
2276 set_ioapic_affinity_irq(irq, TARGET_CPUS);