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>
35 #include <acpi/acpi_bus.h>
42 #include <asm/proto.h>
43 #include <asm/mach_apic.h>
47 #include <asm/msidef.h>
48 #include <asm/hypertransport.h>
53 unsigned move_cleanup_count;
55 u8 move_in_progress : 1;
58 /* irq_cfg is indexed by the sum of all RTEs in all I/O APICs. */
59 struct irq_cfg irq_cfg[NR_IRQS] __read_mostly = {
60 [0] = { .domain = CPU_MASK_ALL, .vector = IRQ0_VECTOR, },
61 [1] = { .domain = CPU_MASK_ALL, .vector = IRQ1_VECTOR, },
62 [2] = { .domain = CPU_MASK_ALL, .vector = IRQ2_VECTOR, },
63 [3] = { .domain = CPU_MASK_ALL, .vector = IRQ3_VECTOR, },
64 [4] = { .domain = CPU_MASK_ALL, .vector = IRQ4_VECTOR, },
65 [5] = { .domain = CPU_MASK_ALL, .vector = IRQ5_VECTOR, },
66 [6] = { .domain = CPU_MASK_ALL, .vector = IRQ6_VECTOR, },
67 [7] = { .domain = CPU_MASK_ALL, .vector = IRQ7_VECTOR, },
68 [8] = { .domain = CPU_MASK_ALL, .vector = IRQ8_VECTOR, },
69 [9] = { .domain = CPU_MASK_ALL, .vector = IRQ9_VECTOR, },
70 [10] = { .domain = CPU_MASK_ALL, .vector = IRQ10_VECTOR, },
71 [11] = { .domain = CPU_MASK_ALL, .vector = IRQ11_VECTOR, },
72 [12] = { .domain = CPU_MASK_ALL, .vector = IRQ12_VECTOR, },
73 [13] = { .domain = CPU_MASK_ALL, .vector = IRQ13_VECTOR, },
74 [14] = { .domain = CPU_MASK_ALL, .vector = IRQ14_VECTOR, },
75 [15] = { .domain = CPU_MASK_ALL, .vector = IRQ15_VECTOR, },
78 static int assign_irq_vector(int irq, cpumask_t mask);
80 #define __apicdebuginit __init
82 int sis_apic_bug; /* not actually supported, dummy for compile */
84 static int no_timer_check;
86 static int disable_timer_pin_1 __initdata;
88 int timer_over_8254 __initdata = 1;
90 /* Where if anywhere is the i8259 connect in external int mode */
91 static struct { int pin, apic; } ioapic_i8259 = { -1, -1 };
93 static DEFINE_SPINLOCK(ioapic_lock);
94 DEFINE_SPINLOCK(vector_lock);
97 * # of IRQ routing registers
99 int nr_ioapic_registers[MAX_IO_APICS];
102 * Rough estimation of how many shared IRQs there are, can
103 * be changed anytime.
105 #define MAX_PLUS_SHARED_IRQS NR_IRQS
106 #define PIN_MAP_SIZE (MAX_PLUS_SHARED_IRQS + NR_IRQS)
109 * This is performance-critical, we want to do it O(1)
111 * the indexing order of this array favors 1:1 mappings
112 * between pins and IRQs.
115 static struct irq_pin_list {
116 short apic, pin, next;
117 } irq_2_pin[PIN_MAP_SIZE];
121 unsigned int unused[3];
125 static __attribute_const__ struct io_apic __iomem *io_apic_base(int idx)
127 return (void __iomem *) __fix_to_virt(FIX_IO_APIC_BASE_0 + idx)
128 + (mp_ioapics[idx].mpc_apicaddr & ~PAGE_MASK);
131 static inline unsigned int io_apic_read(unsigned int apic, unsigned int reg)
133 struct io_apic __iomem *io_apic = io_apic_base(apic);
134 writel(reg, &io_apic->index);
135 return readl(&io_apic->data);
138 static inline void io_apic_write(unsigned int apic, unsigned int reg, unsigned int value)
140 struct io_apic __iomem *io_apic = io_apic_base(apic);
141 writel(reg, &io_apic->index);
142 writel(value, &io_apic->data);
146 * Re-write a value: to be used for read-modify-write
147 * cycles where the read already set up the index register.
149 static inline void io_apic_modify(unsigned int apic, unsigned int value)
151 struct io_apic __iomem *io_apic = io_apic_base(apic);
152 writel(value, &io_apic->data);
155 static int io_apic_level_ack_pending(unsigned int irq)
157 struct irq_pin_list *entry;
161 spin_lock_irqsave(&ioapic_lock, flags);
162 entry = irq_2_pin + irq;
170 reg = io_apic_read(entry->apic, 0x10 + pin*2);
171 /* Is the remote IRR bit set? */
172 pending |= (reg >> 14) & 1;
175 entry = irq_2_pin + entry->next;
177 spin_unlock_irqrestore(&ioapic_lock, flags);
182 * Synchronize the IO-APIC and the CPU by doing
183 * a dummy read from the IO-APIC
185 static inline void io_apic_sync(unsigned int apic)
187 struct io_apic __iomem *io_apic = io_apic_base(apic);
188 readl(&io_apic->data);
191 #define __DO_ACTION(R, ACTION, FINAL) \
195 struct irq_pin_list *entry = irq_2_pin + irq; \
197 BUG_ON(irq >= NR_IRQS); \
203 reg = io_apic_read(entry->apic, 0x10 + R + pin*2); \
205 io_apic_modify(entry->apic, reg); \
209 entry = irq_2_pin + entry->next; \
214 struct { u32 w1, w2; };
215 struct IO_APIC_route_entry entry;
218 static struct IO_APIC_route_entry ioapic_read_entry(int apic, int pin)
220 union entry_union eu;
222 spin_lock_irqsave(&ioapic_lock, flags);
223 eu.w1 = io_apic_read(apic, 0x10 + 2 * pin);
224 eu.w2 = io_apic_read(apic, 0x11 + 2 * pin);
225 spin_unlock_irqrestore(&ioapic_lock, flags);
230 * When we write a new IO APIC routing entry, we need to write the high
231 * word first! If the mask bit in the low word is clear, we will enable
232 * the interrupt, and we need to make sure the entry is fully populated
233 * before that happens.
236 __ioapic_write_entry(int apic, int pin, struct IO_APIC_route_entry e)
238 union entry_union eu;
240 io_apic_write(apic, 0x11 + 2*pin, eu.w2);
241 io_apic_write(apic, 0x10 + 2*pin, eu.w1);
244 static void ioapic_write_entry(int apic, int pin, struct IO_APIC_route_entry e)
247 spin_lock_irqsave(&ioapic_lock, flags);
248 __ioapic_write_entry(apic, pin, e);
249 spin_unlock_irqrestore(&ioapic_lock, flags);
253 * When we mask an IO APIC routing entry, we need to write the low
254 * word first, in order to set the mask bit before we change the
257 static void ioapic_mask_entry(int apic, int pin)
260 union entry_union eu = { .entry.mask = 1 };
262 spin_lock_irqsave(&ioapic_lock, flags);
263 io_apic_write(apic, 0x10 + 2*pin, eu.w1);
264 io_apic_write(apic, 0x11 + 2*pin, eu.w2);
265 spin_unlock_irqrestore(&ioapic_lock, flags);
269 static void __target_IO_APIC_irq(unsigned int irq, unsigned int dest, u8 vector)
272 struct irq_pin_list *entry = irq_2_pin + irq;
274 BUG_ON(irq >= NR_IRQS);
281 io_apic_write(apic, 0x11 + pin*2, dest);
282 reg = io_apic_read(apic, 0x10 + pin*2);
285 io_apic_modify(apic, reg);
288 entry = irq_2_pin + entry->next;
292 static void set_ioapic_affinity_irq(unsigned int irq, cpumask_t mask)
294 struct irq_cfg *cfg = irq_cfg + irq;
299 cpus_and(tmp, mask, cpu_online_map);
303 if (assign_irq_vector(irq, mask))
306 cpus_and(tmp, cfg->domain, mask);
307 dest = cpu_mask_to_apicid(tmp);
310 * Only the high 8 bits are valid.
312 dest = SET_APIC_LOGICAL_ID(dest);
314 spin_lock_irqsave(&ioapic_lock, flags);
315 __target_IO_APIC_irq(irq, dest, cfg->vector);
316 irq_desc[irq].affinity = mask;
317 spin_unlock_irqrestore(&ioapic_lock, flags);
322 * The common case is 1:1 IRQ<->pin mappings. Sometimes there are
323 * shared ISA-space IRQs, so we have to support them. We are super
324 * fast in the common case, and fast for shared ISA-space IRQs.
326 static void add_pin_to_irq(unsigned int irq, int apic, int pin)
328 static int first_free_entry = NR_IRQS;
329 struct irq_pin_list *entry = irq_2_pin + irq;
331 BUG_ON(irq >= NR_IRQS);
333 entry = irq_2_pin + entry->next;
335 if (entry->pin != -1) {
336 entry->next = first_free_entry;
337 entry = irq_2_pin + entry->next;
338 if (++first_free_entry >= PIN_MAP_SIZE)
339 panic("io_apic.c: ran out of irq_2_pin entries!");
346 #define DO_ACTION(name,R,ACTION, FINAL) \
348 static void name##_IO_APIC_irq (unsigned int irq) \
349 __DO_ACTION(R, ACTION, FINAL)
351 DO_ACTION( __mask, 0, |= 0x00010000, io_apic_sync(entry->apic) )
353 DO_ACTION( __unmask, 0, &= 0xfffeffff, )
356 static void mask_IO_APIC_irq (unsigned int irq)
360 spin_lock_irqsave(&ioapic_lock, flags);
361 __mask_IO_APIC_irq(irq);
362 spin_unlock_irqrestore(&ioapic_lock, flags);
365 static void unmask_IO_APIC_irq (unsigned int irq)
369 spin_lock_irqsave(&ioapic_lock, flags);
370 __unmask_IO_APIC_irq(irq);
371 spin_unlock_irqrestore(&ioapic_lock, flags);
374 static void clear_IO_APIC_pin(unsigned int apic, unsigned int pin)
376 struct IO_APIC_route_entry entry;
378 /* Check delivery_mode to be sure we're not clearing an SMI pin */
379 entry = ioapic_read_entry(apic, pin);
380 if (entry.delivery_mode == dest_SMI)
383 * Disable it in the IO-APIC irq-routing table:
385 ioapic_mask_entry(apic, pin);
388 static void clear_IO_APIC (void)
392 for (apic = 0; apic < nr_ioapics; apic++)
393 for (pin = 0; pin < nr_ioapic_registers[apic]; pin++)
394 clear_IO_APIC_pin(apic, pin);
397 int skip_ioapic_setup;
400 static int __init parse_noapic(char *str)
402 disable_ioapic_setup();
405 early_param("noapic", parse_noapic);
407 /* Actually the next is obsolete, but keep it for paranoid reasons -AK */
408 static int __init disable_timer_pin_setup(char *arg)
410 disable_timer_pin_1 = 1;
413 __setup("disable_timer_pin_1", disable_timer_pin_setup);
415 static int __init setup_disable_8254_timer(char *s)
417 timer_over_8254 = -1;
420 static int __init setup_enable_8254_timer(char *s)
426 __setup("disable_8254_timer", setup_disable_8254_timer);
427 __setup("enable_8254_timer", setup_enable_8254_timer);
431 * Find the IRQ entry number of a certain pin.
433 static int find_irq_entry(int apic, int pin, int type)
437 for (i = 0; i < mp_irq_entries; i++)
438 if (mp_irqs[i].mpc_irqtype == type &&
439 (mp_irqs[i].mpc_dstapic == mp_ioapics[apic].mpc_apicid ||
440 mp_irqs[i].mpc_dstapic == MP_APIC_ALL) &&
441 mp_irqs[i].mpc_dstirq == pin)
448 * Find the pin to which IRQ[irq] (ISA) is connected
450 static int __init find_isa_irq_pin(int irq, int type)
454 for (i = 0; i < mp_irq_entries; i++) {
455 int lbus = mp_irqs[i].mpc_srcbus;
457 if (test_bit(lbus, mp_bus_not_pci) &&
458 (mp_irqs[i].mpc_irqtype == type) &&
459 (mp_irqs[i].mpc_srcbusirq == irq))
461 return mp_irqs[i].mpc_dstirq;
466 static int __init find_isa_irq_apic(int irq, int type)
470 for (i = 0; i < mp_irq_entries; i++) {
471 int lbus = mp_irqs[i].mpc_srcbus;
473 if (test_bit(lbus, mp_bus_not_pci) &&
474 (mp_irqs[i].mpc_irqtype == type) &&
475 (mp_irqs[i].mpc_srcbusirq == irq))
478 if (i < mp_irq_entries) {
480 for(apic = 0; apic < nr_ioapics; apic++) {
481 if (mp_ioapics[apic].mpc_apicid == mp_irqs[i].mpc_dstapic)
490 * Find a specific PCI IRQ entry.
491 * Not an __init, possibly needed by modules
493 static int pin_2_irq(int idx, int apic, int pin);
495 int IO_APIC_get_PCI_irq_vector(int bus, int slot, int pin)
497 int apic, i, best_guess = -1;
499 apic_printk(APIC_DEBUG, "querying PCI -> IRQ mapping bus:%d, slot:%d, pin:%d.\n",
501 if (mp_bus_id_to_pci_bus[bus] == -1) {
502 apic_printk(APIC_VERBOSE, "PCI BIOS passed nonexistent PCI bus %d!\n", bus);
505 for (i = 0; i < mp_irq_entries; i++) {
506 int lbus = mp_irqs[i].mpc_srcbus;
508 for (apic = 0; apic < nr_ioapics; apic++)
509 if (mp_ioapics[apic].mpc_apicid == mp_irqs[i].mpc_dstapic ||
510 mp_irqs[i].mpc_dstapic == MP_APIC_ALL)
513 if (!test_bit(lbus, mp_bus_not_pci) &&
514 !mp_irqs[i].mpc_irqtype &&
516 (slot == ((mp_irqs[i].mpc_srcbusirq >> 2) & 0x1f))) {
517 int irq = pin_2_irq(i,apic,mp_irqs[i].mpc_dstirq);
519 if (!(apic || IO_APIC_IRQ(irq)))
522 if (pin == (mp_irqs[i].mpc_srcbusirq & 3))
525 * Use the first all-but-pin matching entry as a
526 * best-guess fuzzy result for broken mptables.
532 BUG_ON(best_guess >= NR_IRQS);
536 /* ISA interrupts are always polarity zero edge triggered,
537 * when listed as conforming in the MP table. */
539 #define default_ISA_trigger(idx) (0)
540 #define default_ISA_polarity(idx) (0)
542 /* PCI interrupts are always polarity one level triggered,
543 * when listed as conforming in the MP table. */
545 #define default_PCI_trigger(idx) (1)
546 #define default_PCI_polarity(idx) (1)
548 static int __init MPBIOS_polarity(int idx)
550 int bus = mp_irqs[idx].mpc_srcbus;
554 * Determine IRQ line polarity (high active or low active):
556 switch (mp_irqs[idx].mpc_irqflag & 3)
558 case 0: /* conforms, ie. bus-type dependent polarity */
559 if (test_bit(bus, mp_bus_not_pci))
560 polarity = default_ISA_polarity(idx);
562 polarity = default_PCI_polarity(idx);
564 case 1: /* high active */
569 case 2: /* reserved */
571 printk(KERN_WARNING "broken BIOS!!\n");
575 case 3: /* low active */
580 default: /* invalid */
582 printk(KERN_WARNING "broken BIOS!!\n");
590 static int MPBIOS_trigger(int idx)
592 int bus = mp_irqs[idx].mpc_srcbus;
596 * Determine IRQ trigger mode (edge or level sensitive):
598 switch ((mp_irqs[idx].mpc_irqflag>>2) & 3)
600 case 0: /* conforms, ie. bus-type dependent */
601 if (test_bit(bus, mp_bus_not_pci))
602 trigger = default_ISA_trigger(idx);
604 trigger = default_PCI_trigger(idx);
611 case 2: /* reserved */
613 printk(KERN_WARNING "broken BIOS!!\n");
622 default: /* invalid */
624 printk(KERN_WARNING "broken BIOS!!\n");
632 static inline int irq_polarity(int idx)
634 return MPBIOS_polarity(idx);
637 static inline int irq_trigger(int idx)
639 return MPBIOS_trigger(idx);
642 static int pin_2_irq(int idx, int apic, int pin)
645 int bus = mp_irqs[idx].mpc_srcbus;
648 * Debugging check, we are in big trouble if this message pops up!
650 if (mp_irqs[idx].mpc_dstirq != pin)
651 printk(KERN_ERR "broken BIOS or MPTABLE parser, ayiee!!\n");
653 if (test_bit(bus, mp_bus_not_pci)) {
654 irq = mp_irqs[idx].mpc_srcbusirq;
657 * PCI IRQs are mapped in order
661 irq += nr_ioapic_registers[i++];
664 BUG_ON(irq >= NR_IRQS);
668 static int __assign_irq_vector(int irq, cpumask_t mask)
671 * NOTE! The local APIC isn't very good at handling
672 * multiple interrupts at the same interrupt level.
673 * As the interrupt level is determined by taking the
674 * vector number and shifting that right by 4, we
675 * want to spread these out a bit so that they don't
676 * all fall in the same interrupt level.
678 * Also, we've got to be careful not to trash gate
679 * 0x80, because int 0x80 is hm, kind of importantish. ;)
681 static int current_vector = FIRST_DEVICE_VECTOR, current_offset = 0;
682 unsigned int old_vector;
686 BUG_ON((unsigned)irq >= NR_IRQS);
689 /* Only try and allocate irqs on cpus that are present */
690 cpus_and(mask, mask, cpu_online_map);
692 if ((cfg->move_in_progress) || cfg->move_cleanup_count)
695 old_vector = cfg->vector;
698 cpus_and(tmp, cfg->domain, mask);
699 if (!cpus_empty(tmp))
703 for_each_cpu_mask(cpu, mask) {
704 cpumask_t domain, new_mask;
708 domain = vector_allocation_domain(cpu);
709 cpus_and(new_mask, domain, cpu_online_map);
711 vector = current_vector;
712 offset = current_offset;
715 if (vector >= FIRST_SYSTEM_VECTOR) {
716 /* If we run out of vectors on large boxen, must share them. */
717 offset = (offset + 1) % 8;
718 vector = FIRST_DEVICE_VECTOR + offset;
720 if (unlikely(current_vector == vector))
722 if (vector == IA32_SYSCALL_VECTOR)
724 for_each_cpu_mask(new_cpu, new_mask)
725 if (per_cpu(vector_irq, new_cpu)[vector] != -1)
728 current_vector = vector;
729 current_offset = offset;
731 cfg->move_in_progress = 1;
732 cfg->old_domain = cfg->domain;
734 for_each_cpu_mask(new_cpu, new_mask)
735 per_cpu(vector_irq, new_cpu)[vector] = irq;
736 cfg->vector = vector;
737 cfg->domain = domain;
743 static int assign_irq_vector(int irq, cpumask_t mask)
748 spin_lock_irqsave(&vector_lock, flags);
749 err = __assign_irq_vector(irq, mask);
750 spin_unlock_irqrestore(&vector_lock, flags);
754 static void __clear_irq_vector(int irq)
760 BUG_ON((unsigned)irq >= NR_IRQS);
762 BUG_ON(!cfg->vector);
764 vector = cfg->vector;
765 cpus_and(mask, cfg->domain, cpu_online_map);
766 for_each_cpu_mask(cpu, mask)
767 per_cpu(vector_irq, cpu)[vector] = -1;
770 cfg->domain = CPU_MASK_NONE;
773 void __setup_vector_irq(int cpu)
775 /* Initialize vector_irq on a new cpu */
776 /* This function must be called with vector_lock held */
779 /* Mark the inuse vectors */
780 for (irq = 0; irq < NR_IRQS; ++irq) {
781 if (!cpu_isset(cpu, irq_cfg[irq].domain))
783 vector = irq_cfg[irq].vector;
784 per_cpu(vector_irq, cpu)[vector] = irq;
786 /* Mark the free vectors */
787 for (vector = 0; vector < NR_VECTORS; ++vector) {
788 irq = per_cpu(vector_irq, cpu)[vector];
791 if (!cpu_isset(cpu, irq_cfg[irq].domain))
792 per_cpu(vector_irq, cpu)[vector] = -1;
797 static struct irq_chip ioapic_chip;
799 static void ioapic_register_intr(int irq, unsigned long trigger)
802 irq_desc[irq].status |= IRQ_LEVEL;
803 set_irq_chip_and_handler_name(irq, &ioapic_chip,
804 handle_fasteoi_irq, "fasteoi");
806 irq_desc[irq].status &= ~IRQ_LEVEL;
807 set_irq_chip_and_handler_name(irq, &ioapic_chip,
808 handle_edge_irq, "edge");
812 static void setup_IO_APIC_irq(int apic, int pin, unsigned int irq,
813 int trigger, int polarity)
815 struct irq_cfg *cfg = irq_cfg + irq;
816 struct IO_APIC_route_entry entry;
819 if (!IO_APIC_IRQ(irq))
823 if (assign_irq_vector(irq, mask))
826 cpus_and(mask, cfg->domain, mask);
828 apic_printk(APIC_VERBOSE,KERN_DEBUG
829 "IOAPIC[%d]: Set routing entry (%d-%d -> 0x%x -> "
830 "IRQ %d Mode:%i Active:%i)\n",
831 apic, mp_ioapics[apic].mpc_apicid, pin, cfg->vector,
832 irq, trigger, polarity);
835 * add it to the IO-APIC irq-routing table:
837 memset(&entry,0,sizeof(entry));
839 entry.delivery_mode = INT_DELIVERY_MODE;
840 entry.dest_mode = INT_DEST_MODE;
841 entry.dest = cpu_mask_to_apicid(mask);
842 entry.mask = 0; /* enable IRQ */
843 entry.trigger = trigger;
844 entry.polarity = polarity;
845 entry.vector = cfg->vector;
847 /* Mask level triggered irqs.
848 * Use IRQ_DELAYED_DISABLE for edge triggered irqs.
853 ioapic_register_intr(irq, trigger);
855 disable_8259A_irq(irq);
857 ioapic_write_entry(apic, pin, entry);
860 static void __init setup_IO_APIC_irqs(void)
862 int apic, pin, idx, irq, first_notcon = 1;
864 apic_printk(APIC_VERBOSE, KERN_DEBUG "init IO_APIC IRQs\n");
866 for (apic = 0; apic < nr_ioapics; apic++) {
867 for (pin = 0; pin < nr_ioapic_registers[apic]; pin++) {
869 idx = find_irq_entry(apic,pin,mp_INT);
872 apic_printk(APIC_VERBOSE, KERN_DEBUG " IO-APIC (apicid-pin) %d-%d", mp_ioapics[apic].mpc_apicid, pin);
875 apic_printk(APIC_VERBOSE, ", %d-%d", mp_ioapics[apic].mpc_apicid, pin);
879 irq = pin_2_irq(idx, apic, pin);
880 add_pin_to_irq(irq, apic, pin);
882 setup_IO_APIC_irq(apic, pin, irq,
883 irq_trigger(idx), irq_polarity(idx));
888 apic_printk(APIC_VERBOSE," not connected.\n");
892 * Set up the 8259A-master output pin as broadcast to all
895 static void __init setup_ExtINT_IRQ0_pin(unsigned int apic, unsigned int pin, int vector)
897 struct IO_APIC_route_entry entry;
900 memset(&entry,0,sizeof(entry));
902 disable_8259A_irq(0);
905 apic_write(APIC_LVT0, APIC_LVT_MASKED | APIC_DM_EXTINT);
908 * We use logical delivery to get the timer IRQ
911 entry.dest_mode = INT_DEST_MODE;
912 entry.mask = 0; /* unmask IRQ now */
913 entry.dest = cpu_mask_to_apicid(TARGET_CPUS);
914 entry.delivery_mode = INT_DELIVERY_MODE;
917 entry.vector = vector;
920 * The timer IRQ doesn't have to know that behind the
921 * scene we have a 8259A-master in AEOI mode ...
923 set_irq_chip_and_handler_name(0, &ioapic_chip, handle_edge_irq, "edge");
926 * Add it to the IO-APIC irq-routing table:
928 spin_lock_irqsave(&ioapic_lock, flags);
929 io_apic_write(apic, 0x11+2*pin, *(((int *)&entry)+1));
930 io_apic_write(apic, 0x10+2*pin, *(((int *)&entry)+0));
931 spin_unlock_irqrestore(&ioapic_lock, flags);
936 void __apicdebuginit print_IO_APIC(void)
939 union IO_APIC_reg_00 reg_00;
940 union IO_APIC_reg_01 reg_01;
941 union IO_APIC_reg_02 reg_02;
944 if (apic_verbosity == APIC_QUIET)
947 printk(KERN_DEBUG "number of MP IRQ sources: %d.\n", mp_irq_entries);
948 for (i = 0; i < nr_ioapics; i++)
949 printk(KERN_DEBUG "number of IO-APIC #%d registers: %d.\n",
950 mp_ioapics[i].mpc_apicid, nr_ioapic_registers[i]);
953 * We are a bit conservative about what we expect. We have to
954 * know about every hardware change ASAP.
956 printk(KERN_INFO "testing the IO APIC.......................\n");
958 for (apic = 0; apic < nr_ioapics; apic++) {
960 spin_lock_irqsave(&ioapic_lock, flags);
961 reg_00.raw = io_apic_read(apic, 0);
962 reg_01.raw = io_apic_read(apic, 1);
963 if (reg_01.bits.version >= 0x10)
964 reg_02.raw = io_apic_read(apic, 2);
965 spin_unlock_irqrestore(&ioapic_lock, flags);
968 printk(KERN_DEBUG "IO APIC #%d......\n", mp_ioapics[apic].mpc_apicid);
969 printk(KERN_DEBUG ".... register #00: %08X\n", reg_00.raw);
970 printk(KERN_DEBUG "....... : physical APIC id: %02X\n", reg_00.bits.ID);
972 printk(KERN_DEBUG ".... register #01: %08X\n", *(int *)®_01);
973 printk(KERN_DEBUG "....... : max redirection entries: %04X\n", reg_01.bits.entries);
975 printk(KERN_DEBUG "....... : PRQ implemented: %X\n", reg_01.bits.PRQ);
976 printk(KERN_DEBUG "....... : IO APIC version: %04X\n", reg_01.bits.version);
978 if (reg_01.bits.version >= 0x10) {
979 printk(KERN_DEBUG ".... register #02: %08X\n", reg_02.raw);
980 printk(KERN_DEBUG "....... : arbitration: %02X\n", reg_02.bits.arbitration);
983 printk(KERN_DEBUG ".... IRQ redirection table:\n");
985 printk(KERN_DEBUG " NR Dst Mask Trig IRR Pol"
986 " Stat Dmod Deli Vect: \n");
988 for (i = 0; i <= reg_01.bits.entries; i++) {
989 struct IO_APIC_route_entry entry;
991 entry = ioapic_read_entry(apic, i);
993 printk(KERN_DEBUG " %02x %03X ",
998 printk("%1d %1d %1d %1d %1d %1d %1d %02X\n",
1003 entry.delivery_status,
1005 entry.delivery_mode,
1010 printk(KERN_DEBUG "IRQ to pin mappings:\n");
1011 for (i = 0; i < NR_IRQS; i++) {
1012 struct irq_pin_list *entry = irq_2_pin + i;
1015 printk(KERN_DEBUG "IRQ%d ", i);
1017 printk("-> %d:%d", entry->apic, entry->pin);
1020 entry = irq_2_pin + entry->next;
1025 printk(KERN_INFO ".................................... done.\n");
1032 static __apicdebuginit void print_APIC_bitfield (int base)
1037 if (apic_verbosity == APIC_QUIET)
1040 printk(KERN_DEBUG "0123456789abcdef0123456789abcdef\n" KERN_DEBUG);
1041 for (i = 0; i < 8; i++) {
1042 v = apic_read(base + i*0x10);
1043 for (j = 0; j < 32; j++) {
1053 void __apicdebuginit print_local_APIC(void * dummy)
1055 unsigned int v, ver, maxlvt;
1057 if (apic_verbosity == APIC_QUIET)
1060 printk("\n" KERN_DEBUG "printing local APIC contents on CPU#%d/%d:\n",
1061 smp_processor_id(), hard_smp_processor_id());
1062 v = apic_read(APIC_ID);
1063 printk(KERN_INFO "... APIC ID: %08x (%01x)\n", v, GET_APIC_ID(v));
1064 v = apic_read(APIC_LVR);
1065 printk(KERN_INFO "... APIC VERSION: %08x\n", v);
1066 ver = GET_APIC_VERSION(v);
1067 maxlvt = get_maxlvt();
1069 v = apic_read(APIC_TASKPRI);
1070 printk(KERN_DEBUG "... APIC TASKPRI: %08x (%02x)\n", v, v & APIC_TPRI_MASK);
1072 v = apic_read(APIC_ARBPRI);
1073 printk(KERN_DEBUG "... APIC ARBPRI: %08x (%02x)\n", v,
1074 v & APIC_ARBPRI_MASK);
1075 v = apic_read(APIC_PROCPRI);
1076 printk(KERN_DEBUG "... APIC PROCPRI: %08x\n", v);
1078 v = apic_read(APIC_EOI);
1079 printk(KERN_DEBUG "... APIC EOI: %08x\n", v);
1080 v = apic_read(APIC_RRR);
1081 printk(KERN_DEBUG "... APIC RRR: %08x\n", v);
1082 v = apic_read(APIC_LDR);
1083 printk(KERN_DEBUG "... APIC LDR: %08x\n", v);
1084 v = apic_read(APIC_DFR);
1085 printk(KERN_DEBUG "... APIC DFR: %08x\n", v);
1086 v = apic_read(APIC_SPIV);
1087 printk(KERN_DEBUG "... APIC SPIV: %08x\n", v);
1089 printk(KERN_DEBUG "... APIC ISR field:\n");
1090 print_APIC_bitfield(APIC_ISR);
1091 printk(KERN_DEBUG "... APIC TMR field:\n");
1092 print_APIC_bitfield(APIC_TMR);
1093 printk(KERN_DEBUG "... APIC IRR field:\n");
1094 print_APIC_bitfield(APIC_IRR);
1096 v = apic_read(APIC_ESR);
1097 printk(KERN_DEBUG "... APIC ESR: %08x\n", v);
1099 v = apic_read(APIC_ICR);
1100 printk(KERN_DEBUG "... APIC ICR: %08x\n", v);
1101 v = apic_read(APIC_ICR2);
1102 printk(KERN_DEBUG "... APIC ICR2: %08x\n", v);
1104 v = apic_read(APIC_LVTT);
1105 printk(KERN_DEBUG "... APIC LVTT: %08x\n", v);
1107 if (maxlvt > 3) { /* PC is LVT#4. */
1108 v = apic_read(APIC_LVTPC);
1109 printk(KERN_DEBUG "... APIC LVTPC: %08x\n", v);
1111 v = apic_read(APIC_LVT0);
1112 printk(KERN_DEBUG "... APIC LVT0: %08x\n", v);
1113 v = apic_read(APIC_LVT1);
1114 printk(KERN_DEBUG "... APIC LVT1: %08x\n", v);
1116 if (maxlvt > 2) { /* ERR is LVT#3. */
1117 v = apic_read(APIC_LVTERR);
1118 printk(KERN_DEBUG "... APIC LVTERR: %08x\n", v);
1121 v = apic_read(APIC_TMICT);
1122 printk(KERN_DEBUG "... APIC TMICT: %08x\n", v);
1123 v = apic_read(APIC_TMCCT);
1124 printk(KERN_DEBUG "... APIC TMCCT: %08x\n", v);
1125 v = apic_read(APIC_TDCR);
1126 printk(KERN_DEBUG "... APIC TDCR: %08x\n", v);
1130 void print_all_local_APICs (void)
1132 on_each_cpu(print_local_APIC, NULL, 1, 1);
1135 void __apicdebuginit print_PIC(void)
1138 unsigned long flags;
1140 if (apic_verbosity == APIC_QUIET)
1143 printk(KERN_DEBUG "\nprinting PIC contents\n");
1145 spin_lock_irqsave(&i8259A_lock, flags);
1147 v = inb(0xa1) << 8 | inb(0x21);
1148 printk(KERN_DEBUG "... PIC IMR: %04x\n", v);
1150 v = inb(0xa0) << 8 | inb(0x20);
1151 printk(KERN_DEBUG "... PIC IRR: %04x\n", v);
1155 v = inb(0xa0) << 8 | inb(0x20);
1159 spin_unlock_irqrestore(&i8259A_lock, flags);
1161 printk(KERN_DEBUG "... PIC ISR: %04x\n", v);
1163 v = inb(0x4d1) << 8 | inb(0x4d0);
1164 printk(KERN_DEBUG "... PIC ELCR: %04x\n", v);
1169 static void __init enable_IO_APIC(void)
1171 union IO_APIC_reg_01 reg_01;
1172 int i8259_apic, i8259_pin;
1174 unsigned long flags;
1176 for (i = 0; i < PIN_MAP_SIZE; i++) {
1177 irq_2_pin[i].pin = -1;
1178 irq_2_pin[i].next = 0;
1182 * The number of IO-APIC IRQ registers (== #pins):
1184 for (apic = 0; apic < nr_ioapics; apic++) {
1185 spin_lock_irqsave(&ioapic_lock, flags);
1186 reg_01.raw = io_apic_read(apic, 1);
1187 spin_unlock_irqrestore(&ioapic_lock, flags);
1188 nr_ioapic_registers[apic] = reg_01.bits.entries+1;
1190 for(apic = 0; apic < nr_ioapics; apic++) {
1192 /* See if any of the pins is in ExtINT mode */
1193 for (pin = 0; pin < nr_ioapic_registers[apic]; pin++) {
1194 struct IO_APIC_route_entry entry;
1195 entry = ioapic_read_entry(apic, pin);
1197 /* If the interrupt line is enabled and in ExtInt mode
1198 * I have found the pin where the i8259 is connected.
1200 if ((entry.mask == 0) && (entry.delivery_mode == dest_ExtINT)) {
1201 ioapic_i8259.apic = apic;
1202 ioapic_i8259.pin = pin;
1208 /* Look to see what if the MP table has reported the ExtINT */
1209 i8259_pin = find_isa_irq_pin(0, mp_ExtINT);
1210 i8259_apic = find_isa_irq_apic(0, mp_ExtINT);
1211 /* Trust the MP table if nothing is setup in the hardware */
1212 if ((ioapic_i8259.pin == -1) && (i8259_pin >= 0)) {
1213 printk(KERN_WARNING "ExtINT not setup in hardware but reported by MP table\n");
1214 ioapic_i8259.pin = i8259_pin;
1215 ioapic_i8259.apic = i8259_apic;
1217 /* Complain if the MP table and the hardware disagree */
1218 if (((ioapic_i8259.apic != i8259_apic) || (ioapic_i8259.pin != i8259_pin)) &&
1219 (i8259_pin >= 0) && (ioapic_i8259.pin >= 0))
1221 printk(KERN_WARNING "ExtINT in hardware and MP table differ\n");
1225 * Do not trust the IO-APIC being empty at bootup
1231 * Not an __init, needed by the reboot code
1233 void disable_IO_APIC(void)
1236 * Clear the IO-APIC before rebooting:
1241 * If the i8259 is routed through an IOAPIC
1242 * Put that IOAPIC in virtual wire mode
1243 * so legacy interrupts can be delivered.
1245 if (ioapic_i8259.pin != -1) {
1246 struct IO_APIC_route_entry entry;
1248 memset(&entry, 0, sizeof(entry));
1249 entry.mask = 0; /* Enabled */
1250 entry.trigger = 0; /* Edge */
1252 entry.polarity = 0; /* High */
1253 entry.delivery_status = 0;
1254 entry.dest_mode = 0; /* Physical */
1255 entry.delivery_mode = dest_ExtINT; /* ExtInt */
1257 entry.dest = GET_APIC_ID(apic_read(APIC_ID));
1260 * Add it to the IO-APIC irq-routing table:
1262 ioapic_write_entry(ioapic_i8259.apic, ioapic_i8259.pin, entry);
1265 disconnect_bsp_APIC(ioapic_i8259.pin != -1);
1269 * There is a nasty bug in some older SMP boards, their mptable lies
1270 * about the timer IRQ. We do the following to work around the situation:
1272 * - timer IRQ defaults to IO-APIC IRQ
1273 * - if this function detects that timer IRQs are defunct, then we fall
1274 * back to ISA timer IRQs
1276 static int __init timer_irq_works(void)
1278 unsigned long t1 = jiffies;
1281 /* Let ten ticks pass... */
1282 mdelay((10 * 1000) / HZ);
1285 * Expect a few ticks at least, to be sure some possible
1286 * glue logic does not lock up after one or two first
1287 * ticks in a non-ExtINT mode. Also the local APIC
1288 * might have cached one ExtINT interrupt. Finally, at
1289 * least one tick may be lost due to delays.
1293 if (jiffies - t1 > 4)
1299 * In the SMP+IOAPIC case it might happen that there are an unspecified
1300 * number of pending IRQ events unhandled. These cases are very rare,
1301 * so we 'resend' these IRQs via IPIs, to the same CPU. It's much
1302 * better to do it this way as thus we do not have to be aware of
1303 * 'pending' interrupts in the IRQ path, except at this point.
1306 * Edge triggered needs to resend any interrupt
1307 * that was delayed but this is now handled in the device
1312 * Starting up a edge-triggered IO-APIC interrupt is
1313 * nasty - we need to make sure that we get the edge.
1314 * If it is already asserted for some reason, we need
1315 * return 1 to indicate that is was pending.
1317 * This is not complete - we should be able to fake
1318 * an edge even if it isn't on the 8259A...
1321 static unsigned int startup_ioapic_irq(unsigned int irq)
1323 int was_pending = 0;
1324 unsigned long flags;
1326 spin_lock_irqsave(&ioapic_lock, flags);
1328 disable_8259A_irq(irq);
1329 if (i8259A_irq_pending(irq))
1332 __unmask_IO_APIC_irq(irq);
1333 spin_unlock_irqrestore(&ioapic_lock, flags);
1338 static int ioapic_retrigger_irq(unsigned int irq)
1340 struct irq_cfg *cfg = &irq_cfg[irq];
1342 unsigned long flags;
1344 spin_lock_irqsave(&vector_lock, flags);
1346 cpu_set(first_cpu(cfg->domain), mask);
1348 send_IPI_mask(mask, cfg->vector);
1349 spin_unlock_irqrestore(&vector_lock, flags);
1355 * Level and edge triggered IO-APIC interrupts need different handling,
1356 * so we use two separate IRQ descriptors. Edge triggered IRQs can be
1357 * handled with the level-triggered descriptor, but that one has slightly
1358 * more overhead. Level-triggered interrupts cannot be handled with the
1359 * edge-triggered handler, without risking IRQ storms and other ugly
1364 asmlinkage void smp_irq_move_cleanup_interrupt(void)
1366 unsigned vector, me;
1371 me = smp_processor_id();
1372 for (vector = FIRST_EXTERNAL_VECTOR; vector < NR_VECTORS; vector++) {
1374 struct irq_desc *desc;
1375 struct irq_cfg *cfg;
1376 irq = __get_cpu_var(vector_irq)[vector];
1380 desc = irq_desc + irq;
1381 cfg = irq_cfg + irq;
1382 spin_lock(&desc->lock);
1383 if (!cfg->move_cleanup_count)
1386 if ((vector == cfg->vector) && cpu_isset(me, cfg->domain))
1389 __get_cpu_var(vector_irq)[vector] = -1;
1390 cfg->move_cleanup_count--;
1392 spin_unlock(&desc->lock);
1398 static void irq_complete_move(unsigned int irq)
1400 struct irq_cfg *cfg = irq_cfg + irq;
1401 unsigned vector, me;
1403 if (likely(!cfg->move_in_progress))
1406 vector = ~get_irq_regs()->orig_rax;
1407 me = smp_processor_id();
1408 if ((vector == cfg->vector) && cpu_isset(me, cfg->domain)) {
1409 cpumask_t cleanup_mask;
1411 cpus_and(cleanup_mask, cfg->old_domain, cpu_online_map);
1412 cfg->move_cleanup_count = cpus_weight(cleanup_mask);
1413 send_IPI_mask(cleanup_mask, IRQ_MOVE_CLEANUP_VECTOR);
1414 cfg->move_in_progress = 0;
1418 static inline void irq_complete_move(unsigned int irq) {}
1421 static void ack_apic_edge(unsigned int irq)
1423 irq_complete_move(irq);
1424 move_native_irq(irq);
1428 static void ack_apic_level(unsigned int irq)
1430 int do_unmask_irq = 0;
1432 irq_complete_move(irq);
1433 #if defined(CONFIG_GENERIC_PENDING_IRQ) || defined(CONFIG_IRQBALANCE)
1434 /* If we are moving the irq we need to mask it */
1435 if (unlikely(irq_desc[irq].status & IRQ_MOVE_PENDING)) {
1437 mask_IO_APIC_irq(irq);
1442 * We must acknowledge the irq before we move it or the acknowledge will
1443 * not propagate properly.
1447 /* Now we can move and renable the irq */
1448 if (unlikely(do_unmask_irq)) {
1449 /* Only migrate the irq if the ack has been received.
1451 * On rare occasions the broadcast level triggered ack gets
1452 * delayed going to ioapics, and if we reprogram the
1453 * vector while Remote IRR is still set the irq will never
1456 * To prevent this scenario we read the Remote IRR bit
1457 * of the ioapic. This has two effects.
1458 * - On any sane system the read of the ioapic will
1459 * flush writes (and acks) going to the ioapic from
1461 * - We get to see if the ACK has actually been delivered.
1463 * Based on failed experiments of reprogramming the
1464 * ioapic entry from outside of irq context starting
1465 * with masking the ioapic entry and then polling until
1466 * Remote IRR was clear before reprogramming the
1467 * ioapic I don't trust the Remote IRR bit to be
1468 * completey accurate.
1470 * However there appears to be no other way to plug
1471 * this race, so if the Remote IRR bit is not
1472 * accurate and is causing problems then it is a hardware bug
1473 * and you can go talk to the chipset vendor about it.
1475 if (!io_apic_level_ack_pending(irq))
1476 move_masked_irq(irq);
1477 unmask_IO_APIC_irq(irq);
1481 static struct irq_chip ioapic_chip __read_mostly = {
1483 .startup = startup_ioapic_irq,
1484 .mask = mask_IO_APIC_irq,
1485 .unmask = unmask_IO_APIC_irq,
1486 .ack = ack_apic_edge,
1487 .eoi = ack_apic_level,
1489 .set_affinity = set_ioapic_affinity_irq,
1491 .retrigger = ioapic_retrigger_irq,
1494 static inline void init_IO_APIC_traps(void)
1499 * NOTE! The local APIC isn't very good at handling
1500 * multiple interrupts at the same interrupt level.
1501 * As the interrupt level is determined by taking the
1502 * vector number and shifting that right by 4, we
1503 * want to spread these out a bit so that they don't
1504 * all fall in the same interrupt level.
1506 * Also, we've got to be careful not to trash gate
1507 * 0x80, because int 0x80 is hm, kind of importantish. ;)
1509 for (irq = 0; irq < NR_IRQS ; irq++) {
1511 if (IO_APIC_IRQ(tmp) && !irq_cfg[tmp].vector) {
1513 * Hmm.. We don't have an entry for this,
1514 * so default to an old-fashioned 8259
1515 * interrupt if we can..
1518 make_8259A_irq(irq);
1520 /* Strange. Oh, well.. */
1521 irq_desc[irq].chip = &no_irq_chip;
1526 static void enable_lapic_irq (unsigned int irq)
1530 v = apic_read(APIC_LVT0);
1531 apic_write(APIC_LVT0, v & ~APIC_LVT_MASKED);
1534 static void disable_lapic_irq (unsigned int irq)
1538 v = apic_read(APIC_LVT0);
1539 apic_write(APIC_LVT0, v | APIC_LVT_MASKED);
1542 static void ack_lapic_irq (unsigned int irq)
1547 static void end_lapic_irq (unsigned int i) { /* nothing */ }
1549 static struct hw_interrupt_type lapic_irq_type __read_mostly = {
1550 .name = "local-APIC",
1551 .typename = "local-APIC-edge",
1552 .startup = NULL, /* startup_irq() not used for IRQ0 */
1553 .shutdown = NULL, /* shutdown_irq() not used for IRQ0 */
1554 .enable = enable_lapic_irq,
1555 .disable = disable_lapic_irq,
1556 .ack = ack_lapic_irq,
1557 .end = end_lapic_irq,
1560 static void setup_nmi (void)
1563 * Dirty trick to enable the NMI watchdog ...
1564 * We put the 8259A master into AEOI mode and
1565 * unmask on all local APICs LVT0 as NMI.
1567 * The idea to use the 8259A in AEOI mode ('8259A Virtual Wire')
1568 * is from Maciej W. Rozycki - so we do not have to EOI from
1569 * the NMI handler or the timer interrupt.
1571 printk(KERN_INFO "activating NMI Watchdog ...");
1573 enable_NMI_through_LVT0(NULL);
1579 * This looks a bit hackish but it's about the only one way of sending
1580 * a few INTA cycles to 8259As and any associated glue logic. ICR does
1581 * not support the ExtINT mode, unfortunately. We need to send these
1582 * cycles as some i82489DX-based boards have glue logic that keeps the
1583 * 8259A interrupt line asserted until INTA. --macro
1585 static inline void unlock_ExtINT_logic(void)
1588 struct IO_APIC_route_entry entry0, entry1;
1589 unsigned char save_control, save_freq_select;
1590 unsigned long flags;
1592 pin = find_isa_irq_pin(8, mp_INT);
1593 apic = find_isa_irq_apic(8, mp_INT);
1597 spin_lock_irqsave(&ioapic_lock, flags);
1598 *(((int *)&entry0) + 1) = io_apic_read(apic, 0x11 + 2 * pin);
1599 *(((int *)&entry0) + 0) = io_apic_read(apic, 0x10 + 2 * pin);
1600 spin_unlock_irqrestore(&ioapic_lock, flags);
1601 clear_IO_APIC_pin(apic, pin);
1603 memset(&entry1, 0, sizeof(entry1));
1605 entry1.dest_mode = 0; /* physical delivery */
1606 entry1.mask = 0; /* unmask IRQ now */
1607 entry1.dest = hard_smp_processor_id();
1608 entry1.delivery_mode = dest_ExtINT;
1609 entry1.polarity = entry0.polarity;
1613 spin_lock_irqsave(&ioapic_lock, flags);
1614 io_apic_write(apic, 0x11 + 2 * pin, *(((int *)&entry1) + 1));
1615 io_apic_write(apic, 0x10 + 2 * pin, *(((int *)&entry1) + 0));
1616 spin_unlock_irqrestore(&ioapic_lock, flags);
1618 save_control = CMOS_READ(RTC_CONTROL);
1619 save_freq_select = CMOS_READ(RTC_FREQ_SELECT);
1620 CMOS_WRITE((save_freq_select & ~RTC_RATE_SELECT) | 0x6,
1622 CMOS_WRITE(save_control | RTC_PIE, RTC_CONTROL);
1627 if ((CMOS_READ(RTC_INTR_FLAGS) & RTC_PF) == RTC_PF)
1631 CMOS_WRITE(save_control, RTC_CONTROL);
1632 CMOS_WRITE(save_freq_select, RTC_FREQ_SELECT);
1633 clear_IO_APIC_pin(apic, pin);
1635 spin_lock_irqsave(&ioapic_lock, flags);
1636 io_apic_write(apic, 0x11 + 2 * pin, *(((int *)&entry0) + 1));
1637 io_apic_write(apic, 0x10 + 2 * pin, *(((int *)&entry0) + 0));
1638 spin_unlock_irqrestore(&ioapic_lock, flags);
1642 * This code may look a bit paranoid, but it's supposed to cooperate with
1643 * a wide range of boards and BIOS bugs. Fortunately only the timer IRQ
1644 * is so screwy. Thanks to Brian Perkins for testing/hacking this beast
1645 * fanatically on his truly buggy board.
1647 * FIXME: really need to revamp this for modern platforms only.
1649 static inline void check_timer(void)
1651 struct irq_cfg *cfg = irq_cfg + 0;
1652 int apic1, pin1, apic2, pin2;
1655 * get/set the timer IRQ vector:
1657 disable_8259A_irq(0);
1658 assign_irq_vector(0, TARGET_CPUS);
1661 * Subtle, code in do_timer_interrupt() expects an AEOI
1662 * mode for the 8259A whenever interrupts are routed
1663 * through I/O APICs. Also IRQ0 has to be enabled in
1664 * the 8259A which implies the virtual wire has to be
1665 * disabled in the local APIC.
1667 apic_write(APIC_LVT0, APIC_LVT_MASKED | APIC_DM_EXTINT);
1669 if (timer_over_8254 > 0)
1670 enable_8259A_irq(0);
1672 pin1 = find_isa_irq_pin(0, mp_INT);
1673 apic1 = find_isa_irq_apic(0, mp_INT);
1674 pin2 = ioapic_i8259.pin;
1675 apic2 = ioapic_i8259.apic;
1677 apic_printk(APIC_VERBOSE,KERN_INFO "..TIMER: vector=0x%02X apic1=%d pin1=%d apic2=%d pin2=%d\n",
1678 cfg->vector, apic1, pin1, apic2, pin2);
1682 * Ok, does IRQ0 through the IOAPIC work?
1684 unmask_IO_APIC_irq(0);
1685 if (!no_timer_check && timer_irq_works()) {
1686 nmi_watchdog_default();
1687 if (nmi_watchdog == NMI_IO_APIC) {
1688 disable_8259A_irq(0);
1690 enable_8259A_irq(0);
1692 if (disable_timer_pin_1 > 0)
1693 clear_IO_APIC_pin(0, pin1);
1696 clear_IO_APIC_pin(apic1, pin1);
1697 apic_printk(APIC_QUIET,KERN_ERR "..MP-BIOS bug: 8254 timer not "
1698 "connected to IO-APIC\n");
1701 apic_printk(APIC_VERBOSE,KERN_INFO "...trying to set up timer (IRQ0) "
1702 "through the 8259A ... ");
1704 apic_printk(APIC_VERBOSE,"\n..... (found apic %d pin %d) ...",
1707 * legacy devices should be connected to IO APIC #0
1709 setup_ExtINT_IRQ0_pin(apic2, pin2, cfg->vector);
1710 if (timer_irq_works()) {
1711 apic_printk(APIC_VERBOSE," works.\n");
1712 nmi_watchdog_default();
1713 if (nmi_watchdog == NMI_IO_APIC) {
1719 * Cleanup, just in case ...
1721 clear_IO_APIC_pin(apic2, pin2);
1723 apic_printk(APIC_VERBOSE," failed.\n");
1725 if (nmi_watchdog == NMI_IO_APIC) {
1726 printk(KERN_WARNING "timer doesn't work through the IO-APIC - disabling NMI Watchdog!\n");
1730 apic_printk(APIC_VERBOSE, KERN_INFO "...trying to set up timer as Virtual Wire IRQ...");
1732 disable_8259A_irq(0);
1733 irq_desc[0].chip = &lapic_irq_type;
1734 apic_write(APIC_LVT0, APIC_DM_FIXED | cfg->vector); /* Fixed mode */
1735 enable_8259A_irq(0);
1737 if (timer_irq_works()) {
1738 apic_printk(APIC_VERBOSE," works.\n");
1741 apic_write(APIC_LVT0, APIC_LVT_MASKED | APIC_DM_FIXED | cfg->vector);
1742 apic_printk(APIC_VERBOSE," failed.\n");
1744 apic_printk(APIC_VERBOSE, KERN_INFO "...trying to set up timer as ExtINT IRQ...");
1748 apic_write(APIC_LVT0, APIC_DM_EXTINT);
1750 unlock_ExtINT_logic();
1752 if (timer_irq_works()) {
1753 apic_printk(APIC_VERBOSE," works.\n");
1756 apic_printk(APIC_VERBOSE," failed :(.\n");
1757 panic("IO-APIC + timer doesn't work! Try using the 'noapic' kernel parameter\n");
1760 static int __init notimercheck(char *s)
1765 __setup("no_timer_check", notimercheck);
1769 * IRQ's that are handled by the PIC in the MPS IOAPIC case.
1770 * - IRQ2 is the cascade IRQ, and cannot be a io-apic IRQ.
1771 * Linux doesn't really care, as it's not actually used
1772 * for any interrupt handling anyway.
1774 #define PIC_IRQS (1<<2)
1776 void __init setup_IO_APIC(void)
1781 io_apic_irqs = ~0; /* all IRQs go through IOAPIC */
1783 io_apic_irqs = ~PIC_IRQS;
1785 apic_printk(APIC_VERBOSE, "ENABLING IO-APIC IRQs\n");
1788 setup_IO_APIC_irqs();
1789 init_IO_APIC_traps();
1795 struct sysfs_ioapic_data {
1796 struct sys_device dev;
1797 struct IO_APIC_route_entry entry[0];
1799 static struct sysfs_ioapic_data * mp_ioapic_data[MAX_IO_APICS];
1801 static int ioapic_suspend(struct sys_device *dev, pm_message_t state)
1803 struct IO_APIC_route_entry *entry;
1804 struct sysfs_ioapic_data *data;
1807 data = container_of(dev, struct sysfs_ioapic_data, dev);
1808 entry = data->entry;
1809 for (i = 0; i < nr_ioapic_registers[dev->id]; i ++, entry ++ )
1810 *entry = ioapic_read_entry(dev->id, i);
1815 static int ioapic_resume(struct sys_device *dev)
1817 struct IO_APIC_route_entry *entry;
1818 struct sysfs_ioapic_data *data;
1819 unsigned long flags;
1820 union IO_APIC_reg_00 reg_00;
1823 data = container_of(dev, struct sysfs_ioapic_data, dev);
1824 entry = data->entry;
1826 spin_lock_irqsave(&ioapic_lock, flags);
1827 reg_00.raw = io_apic_read(dev->id, 0);
1828 if (reg_00.bits.ID != mp_ioapics[dev->id].mpc_apicid) {
1829 reg_00.bits.ID = mp_ioapics[dev->id].mpc_apicid;
1830 io_apic_write(dev->id, 0, reg_00.raw);
1832 spin_unlock_irqrestore(&ioapic_lock, flags);
1833 for (i = 0; i < nr_ioapic_registers[dev->id]; i++)
1834 ioapic_write_entry(dev->id, i, entry[i]);
1839 static struct sysdev_class ioapic_sysdev_class = {
1840 set_kset_name("ioapic"),
1841 .suspend = ioapic_suspend,
1842 .resume = ioapic_resume,
1845 static int __init ioapic_init_sysfs(void)
1847 struct sys_device * dev;
1848 int i, size, error = 0;
1850 error = sysdev_class_register(&ioapic_sysdev_class);
1854 for (i = 0; i < nr_ioapics; i++ ) {
1855 size = sizeof(struct sys_device) + nr_ioapic_registers[i]
1856 * sizeof(struct IO_APIC_route_entry);
1857 mp_ioapic_data[i] = kmalloc(size, GFP_KERNEL);
1858 if (!mp_ioapic_data[i]) {
1859 printk(KERN_ERR "Can't suspend/resume IOAPIC %d\n", i);
1862 memset(mp_ioapic_data[i], 0, size);
1863 dev = &mp_ioapic_data[i]->dev;
1865 dev->cls = &ioapic_sysdev_class;
1866 error = sysdev_register(dev);
1868 kfree(mp_ioapic_data[i]);
1869 mp_ioapic_data[i] = NULL;
1870 printk(KERN_ERR "Can't suspend/resume IOAPIC %d\n", i);
1878 device_initcall(ioapic_init_sysfs);
1881 * Dynamic irq allocate and deallocation
1883 int create_irq(void)
1885 /* Allocate an unused irq */
1888 unsigned long flags;
1891 spin_lock_irqsave(&vector_lock, flags);
1892 for (new = (NR_IRQS - 1); new >= 0; new--) {
1893 if (platform_legacy_irq(new))
1895 if (irq_cfg[new].vector != 0)
1897 if (__assign_irq_vector(new, TARGET_CPUS) == 0)
1901 spin_unlock_irqrestore(&vector_lock, flags);
1904 dynamic_irq_init(irq);
1909 void destroy_irq(unsigned int irq)
1911 unsigned long flags;
1913 dynamic_irq_cleanup(irq);
1915 spin_lock_irqsave(&vector_lock, flags);
1916 __clear_irq_vector(irq);
1917 spin_unlock_irqrestore(&vector_lock, flags);
1921 * MSI mesage composition
1923 #ifdef CONFIG_PCI_MSI
1924 static int msi_compose_msg(struct pci_dev *pdev, unsigned int irq, struct msi_msg *msg)
1926 struct irq_cfg *cfg = irq_cfg + irq;
1932 err = assign_irq_vector(irq, tmp);
1934 cpus_and(tmp, cfg->domain, tmp);
1935 dest = cpu_mask_to_apicid(tmp);
1937 msg->address_hi = MSI_ADDR_BASE_HI;
1940 ((INT_DEST_MODE == 0) ?
1941 MSI_ADDR_DEST_MODE_PHYSICAL:
1942 MSI_ADDR_DEST_MODE_LOGICAL) |
1943 ((INT_DELIVERY_MODE != dest_LowestPrio) ?
1944 MSI_ADDR_REDIRECTION_CPU:
1945 MSI_ADDR_REDIRECTION_LOWPRI) |
1946 MSI_ADDR_DEST_ID(dest);
1949 MSI_DATA_TRIGGER_EDGE |
1950 MSI_DATA_LEVEL_ASSERT |
1951 ((INT_DELIVERY_MODE != dest_LowestPrio) ?
1952 MSI_DATA_DELIVERY_FIXED:
1953 MSI_DATA_DELIVERY_LOWPRI) |
1954 MSI_DATA_VECTOR(cfg->vector);
1960 static void set_msi_irq_affinity(unsigned int irq, cpumask_t mask)
1962 struct irq_cfg *cfg = irq_cfg + irq;
1967 cpus_and(tmp, mask, cpu_online_map);
1968 if (cpus_empty(tmp))
1971 if (assign_irq_vector(irq, mask))
1974 cpus_and(tmp, cfg->domain, mask);
1975 dest = cpu_mask_to_apicid(tmp);
1977 read_msi_msg(irq, &msg);
1979 msg.data &= ~MSI_DATA_VECTOR_MASK;
1980 msg.data |= MSI_DATA_VECTOR(cfg->vector);
1981 msg.address_lo &= ~MSI_ADDR_DEST_ID_MASK;
1982 msg.address_lo |= MSI_ADDR_DEST_ID(dest);
1984 write_msi_msg(irq, &msg);
1985 irq_desc[irq].affinity = mask;
1987 #endif /* CONFIG_SMP */
1990 * IRQ Chip for MSI PCI/PCI-X/PCI-Express Devices,
1991 * which implement the MSI or MSI-X Capability Structure.
1993 static struct irq_chip msi_chip = {
1995 .unmask = unmask_msi_irq,
1996 .mask = mask_msi_irq,
1997 .ack = ack_apic_edge,
1999 .set_affinity = set_msi_irq_affinity,
2001 .retrigger = ioapic_retrigger_irq,
2004 int arch_setup_msi_irq(struct pci_dev *dev, struct msi_desc *desc)
2012 ret = msi_compose_msg(dev, irq, &msg);
2018 set_irq_msi(irq, desc);
2019 write_msi_msg(irq, &msg);
2021 set_irq_chip_and_handler_name(irq, &msi_chip, handle_edge_irq, "edge");
2026 void arch_teardown_msi_irq(unsigned int irq)
2031 #endif /* CONFIG_PCI_MSI */
2034 * Hypertransport interrupt support
2036 #ifdef CONFIG_HT_IRQ
2040 static void target_ht_irq(unsigned int irq, unsigned int dest, u8 vector)
2042 struct ht_irq_msg msg;
2043 fetch_ht_irq_msg(irq, &msg);
2045 msg.address_lo &= ~(HT_IRQ_LOW_VECTOR_MASK | HT_IRQ_LOW_DEST_ID_MASK);
2046 msg.address_hi &= ~(HT_IRQ_HIGH_DEST_ID_MASK);
2048 msg.address_lo |= HT_IRQ_LOW_VECTOR(vector) | HT_IRQ_LOW_DEST_ID(dest);
2049 msg.address_hi |= HT_IRQ_HIGH_DEST_ID(dest);
2051 write_ht_irq_msg(irq, &msg);
2054 static void set_ht_irq_affinity(unsigned int irq, cpumask_t mask)
2056 struct irq_cfg *cfg = irq_cfg + irq;
2060 cpus_and(tmp, mask, cpu_online_map);
2061 if (cpus_empty(tmp))
2064 if (assign_irq_vector(irq, mask))
2067 cpus_and(tmp, cfg->domain, mask);
2068 dest = cpu_mask_to_apicid(tmp);
2070 target_ht_irq(irq, dest, cfg->vector);
2071 irq_desc[irq].affinity = mask;
2075 static struct irq_chip ht_irq_chip = {
2077 .mask = mask_ht_irq,
2078 .unmask = unmask_ht_irq,
2079 .ack = ack_apic_edge,
2081 .set_affinity = set_ht_irq_affinity,
2083 .retrigger = ioapic_retrigger_irq,
2086 int arch_setup_ht_irq(unsigned int irq, struct pci_dev *dev)
2088 struct irq_cfg *cfg = irq_cfg + irq;
2093 err = assign_irq_vector(irq, tmp);
2095 struct ht_irq_msg msg;
2098 cpus_and(tmp, cfg->domain, tmp);
2099 dest = cpu_mask_to_apicid(tmp);
2101 msg.address_hi = HT_IRQ_HIGH_DEST_ID(dest);
2105 HT_IRQ_LOW_DEST_ID(dest) |
2106 HT_IRQ_LOW_VECTOR(cfg->vector) |
2107 ((INT_DEST_MODE == 0) ?
2108 HT_IRQ_LOW_DM_PHYSICAL :
2109 HT_IRQ_LOW_DM_LOGICAL) |
2110 HT_IRQ_LOW_RQEOI_EDGE |
2111 ((INT_DELIVERY_MODE != dest_LowestPrio) ?
2112 HT_IRQ_LOW_MT_FIXED :
2113 HT_IRQ_LOW_MT_ARBITRATED) |
2114 HT_IRQ_LOW_IRQ_MASKED;
2116 write_ht_irq_msg(irq, &msg);
2118 set_irq_chip_and_handler_name(irq, &ht_irq_chip,
2119 handle_edge_irq, "edge");
2123 #endif /* CONFIG_HT_IRQ */
2125 /* --------------------------------------------------------------------------
2126 ACPI-based IOAPIC Configuration
2127 -------------------------------------------------------------------------- */
2131 #define IO_APIC_MAX_ID 0xFE
2133 int __init io_apic_get_redir_entries (int ioapic)
2135 union IO_APIC_reg_01 reg_01;
2136 unsigned long flags;
2138 spin_lock_irqsave(&ioapic_lock, flags);
2139 reg_01.raw = io_apic_read(ioapic, 1);
2140 spin_unlock_irqrestore(&ioapic_lock, flags);
2142 return reg_01.bits.entries;
2146 int io_apic_set_pci_routing (int ioapic, int pin, int irq, int triggering, int polarity)
2148 if (!IO_APIC_IRQ(irq)) {
2149 apic_printk(APIC_QUIET,KERN_ERR "IOAPIC[%d]: Invalid reference to IRQ 0\n",
2155 * IRQs < 16 are already in the irq_2_pin[] map
2158 add_pin_to_irq(irq, ioapic, pin);
2160 setup_IO_APIC_irq(ioapic, pin, irq, triggering, polarity);
2165 #endif /* CONFIG_ACPI */
2169 * This function currently is only a helper for the i386 smp boot process where
2170 * we need to reprogram the ioredtbls to cater for the cpus which have come online
2171 * so mask in all cases should simply be TARGET_CPUS
2174 void __init setup_ioapic_dest(void)
2176 int pin, ioapic, irq, irq_entry;
2178 if (skip_ioapic_setup == 1)
2181 for (ioapic = 0; ioapic < nr_ioapics; ioapic++) {
2182 for (pin = 0; pin < nr_ioapic_registers[ioapic]; pin++) {
2183 irq_entry = find_irq_entry(ioapic, pin, mp_INT);
2184 if (irq_entry == -1)
2186 irq = pin_2_irq(irq_entry, ioapic, pin);
2188 /* setup_IO_APIC_irqs could fail to get vector for some device
2189 * when you have too many devices, because at that time only boot
2192 if (!irq_cfg[irq].vector)
2193 setup_IO_APIC_irq(ioapic, pin, irq,
2194 irq_trigger(irq_entry),
2195 irq_polarity(irq_entry));
2197 set_ioapic_affinity_irq(irq, TARGET_CPUS);