2 * Intel IO-APIC support for multi-Pentium hosts.
4 * Copyright (C) 1997, 1998, 1999, 2000 Ingo Molnar, Hajnalka Szabo
6 * Many thanks to Stig Venaas for trying out countless experimental
7 * patches and reporting/debugging problems patiently!
9 * (c) 1999, Multiple IO-APIC support, developed by
10 * Ken-ichi Yaku <yaku@css1.kbnes.nec.co.jp> and
11 * Hidemi Kishimoto <kisimoto@css1.kbnes.nec.co.jp>,
12 * further tested and cleaned up by Zach Brown <zab@redhat.com>
13 * and Ingo Molnar <mingo@redhat.com>
16 * Maciej W. Rozycki : Bits for genuine 82489DX APICs;
17 * thanks to Eric Gilmore
19 * for testing these extensively
20 * Paul Diefenbaugh : Added full ACPI support
24 #include <linux/interrupt.h>
25 #include <linux/init.h>
26 #include <linux/delay.h>
27 #include <linux/sched.h>
28 #include <linux/smp_lock.h>
29 #include <linux/mc146818rtc.h>
30 #include <linux/acpi.h>
31 #include <linux/sysdev.h>
33 #include <acpi/acpi_bus.h>
39 #include <asm/proto.h>
40 #include <asm/mach_apic.h>
45 #define __apicdebuginit __init
47 int sis_apic_bug; /* not actually supported, dummy for compile */
49 static int no_timer_check;
51 int disable_timer_pin_1 __initdata;
53 int timer_over_8254 __initdata = 0;
55 /* Where if anywhere is the i8259 connect in external int mode */
56 static struct { int pin, apic; } ioapic_i8259 = { -1, -1 };
58 static DEFINE_SPINLOCK(ioapic_lock);
59 static DEFINE_SPINLOCK(vector_lock);
62 * # of IRQ routing registers
64 int nr_ioapic_registers[MAX_IO_APICS];
67 * Rough estimation of how many shared IRQs there are, can
70 #define MAX_PLUS_SHARED_IRQS NR_IRQ_VECTORS
71 #define PIN_MAP_SIZE (MAX_PLUS_SHARED_IRQS + NR_IRQS)
74 * This is performance-critical, we want to do it O(1)
76 * the indexing order of this array favors 1:1 mappings
77 * between pins and IRQs.
80 static struct irq_pin_list {
81 short apic, pin, next;
82 } irq_2_pin[PIN_MAP_SIZE];
84 int vector_irq[NR_VECTORS] __read_mostly = { [0 ... NR_VECTORS - 1] = -1};
86 #define vector_to_irq(vector) \
87 (platform_legacy_irq(vector) ? vector : vector_irq[vector])
89 #define vector_to_irq(vector) (vector)
92 #define __DO_ACTION(R, ACTION, FINAL) \
96 struct irq_pin_list *entry = irq_2_pin + irq; \
98 BUG_ON(irq >= NR_IRQS); \
104 reg = io_apic_read(entry->apic, 0x10 + R + pin*2); \
106 io_apic_modify(entry->apic, reg); \
109 entry = irq_2_pin + entry->next; \
115 static void set_ioapic_affinity_irq(unsigned int irq, cpumask_t mask)
121 cpus_and(tmp, mask, cpu_online_map);
125 cpus_and(mask, tmp, CPU_MASK_ALL);
127 dest = cpu_mask_to_apicid(mask);
130 * Only the high 8 bits are valid.
132 dest = SET_APIC_LOGICAL_ID(dest);
134 spin_lock_irqsave(&ioapic_lock, flags);
135 __DO_ACTION(1, = dest, )
136 set_irq_info(irq, mask);
137 spin_unlock_irqrestore(&ioapic_lock, flags);
141 static u8 gsi_2_irq[NR_IRQ_VECTORS] = { [0 ... NR_IRQ_VECTORS-1] = 0xFF };
144 * The common case is 1:1 IRQ<->pin mappings. Sometimes there are
145 * shared ISA-space IRQs, so we have to support them. We are super
146 * fast in the common case, and fast for shared ISA-space IRQs.
148 static void add_pin_to_irq(unsigned int irq, int apic, int pin)
150 static int first_free_entry = NR_IRQS;
151 struct irq_pin_list *entry = irq_2_pin + irq;
153 BUG_ON(irq >= NR_IRQS);
155 entry = irq_2_pin + entry->next;
157 if (entry->pin != -1) {
158 entry->next = first_free_entry;
159 entry = irq_2_pin + entry->next;
160 if (++first_free_entry >= PIN_MAP_SIZE)
161 panic("io_apic.c: ran out of irq_2_pin entries!");
168 #define DO_ACTION(name,R,ACTION, FINAL) \
170 static void name##_IO_APIC_irq (unsigned int irq) \
171 __DO_ACTION(R, ACTION, FINAL)
173 DO_ACTION( __mask, 0, |= 0x00010000, io_apic_sync(entry->apic) )
175 DO_ACTION( __unmask, 0, &= 0xfffeffff, )
178 static void mask_IO_APIC_irq (unsigned int irq)
182 spin_lock_irqsave(&ioapic_lock, flags);
183 __mask_IO_APIC_irq(irq);
184 spin_unlock_irqrestore(&ioapic_lock, flags);
187 static void unmask_IO_APIC_irq (unsigned int irq)
191 spin_lock_irqsave(&ioapic_lock, flags);
192 __unmask_IO_APIC_irq(irq);
193 spin_unlock_irqrestore(&ioapic_lock, flags);
196 static void clear_IO_APIC_pin(unsigned int apic, unsigned int pin)
198 struct IO_APIC_route_entry entry;
201 /* Check delivery_mode to be sure we're not clearing an SMI pin */
202 spin_lock_irqsave(&ioapic_lock, flags);
203 *(((int*)&entry) + 0) = io_apic_read(apic, 0x10 + 2 * pin);
204 *(((int*)&entry) + 1) = io_apic_read(apic, 0x11 + 2 * pin);
205 spin_unlock_irqrestore(&ioapic_lock, flags);
206 if (entry.delivery_mode == dest_SMI)
209 * Disable it in the IO-APIC irq-routing table:
211 memset(&entry, 0, sizeof(entry));
213 spin_lock_irqsave(&ioapic_lock, flags);
214 io_apic_write(apic, 0x10 + 2 * pin, *(((int *)&entry) + 0));
215 io_apic_write(apic, 0x11 + 2 * pin, *(((int *)&entry) + 1));
216 spin_unlock_irqrestore(&ioapic_lock, flags);
219 static void clear_IO_APIC (void)
223 for (apic = 0; apic < nr_ioapics; apic++)
224 for (pin = 0; pin < nr_ioapic_registers[apic]; pin++)
225 clear_IO_APIC_pin(apic, pin);
229 * support for broken MP BIOSs, enables hand-redirection of PIRQ0-7 to
230 * specific CPU-side IRQs.
234 static int pirq_entries [MAX_PIRQS];
235 static int pirqs_enabled;
236 int skip_ioapic_setup;
239 /* dummy parsing: see setup.c */
241 static int __init disable_ioapic_setup(char *str)
243 skip_ioapic_setup = 1;
247 static int __init enable_ioapic_setup(char *str)
250 skip_ioapic_setup = 0;
254 __setup("noapic", disable_ioapic_setup);
255 __setup("apic", enable_ioapic_setup);
257 static int __init setup_disable_8254_timer(char *s)
259 timer_over_8254 = -1;
262 static int __init setup_enable_8254_timer(char *s)
268 __setup("disable_8254_timer", setup_disable_8254_timer);
269 __setup("enable_8254_timer", setup_enable_8254_timer);
271 #include <asm/pci-direct.h>
272 #include <linux/pci_ids.h>
273 #include <linux/pci.h>
278 static int nvidia_hpet_detected __initdata;
280 static int __init nvidia_hpet_check(unsigned long phys, unsigned long size)
282 nvidia_hpet_detected = 1;
287 /* Temporary Hack. Nvidia and VIA boards currently only work with IO-APIC
288 off. Check for an Nvidia or VIA PCI bridge and turn it off.
289 Use pci direct infrastructure because this runs before the PCI subsystem.
291 Can be overwritten with "apic"
293 And another hack to disable the IOMMU on VIA chipsets.
295 ... and others. Really should move this somewhere else.
298 void __init check_ioapic(void)
301 /* Poor man's PCI discovery */
302 for (num = 0; num < 32; num++) {
303 for (slot = 0; slot < 32; slot++) {
304 for (func = 0; func < 8; func++) {
308 class = read_pci_config(num,slot,func,
310 if (class == 0xffffffff)
313 if ((class >> 16) != PCI_CLASS_BRIDGE_PCI)
316 vendor = read_pci_config(num, slot, func,
320 case PCI_VENDOR_ID_VIA:
322 if ((end_pfn > MAX_DMA32_PFN ||
324 !iommu_aperture_allowed) {
326 "Looks like a VIA chipset. Disabling IOMMU. Override with \"iommu=allowed\"\n");
327 iommu_aperture_disabled = 1;
331 case PCI_VENDOR_ID_NVIDIA:
334 * All timer overrides on Nvidia are
335 * wrong unless HPET is enabled.
337 nvidia_hpet_detected = 0;
338 acpi_table_parse(ACPI_HPET,
340 if (nvidia_hpet_detected == 0) {
341 acpi_skip_timer_override = 1;
342 printk(KERN_INFO "Nvidia board "
343 "detected. Ignoring ACPI "
344 "timer override.\n");
347 /* RED-PEN skip them on mptables too? */
350 /* This should be actually default, but
351 for 2.6.16 let's do it for ATI only where
352 it's really needed. */
353 case PCI_VENDOR_ID_ATI:
354 if (timer_over_8254 == 1) {
357 "ATI board detected. Disabling timer routing over 8254.\n");
363 /* No multi-function device? */
364 type = read_pci_config_byte(num,slot,func,
373 static int __init ioapic_pirq_setup(char *str)
376 int ints[MAX_PIRQS+1];
378 get_options(str, ARRAY_SIZE(ints), ints);
380 for (i = 0; i < MAX_PIRQS; i++)
381 pirq_entries[i] = -1;
384 apic_printk(APIC_VERBOSE, "PIRQ redirection, working around broken MP-BIOS.\n");
386 if (ints[0] < MAX_PIRQS)
389 for (i = 0; i < max; i++) {
390 apic_printk(APIC_VERBOSE, "... PIRQ%d -> IRQ %d\n", i, ints[i+1]);
392 * PIRQs are mapped upside down, usually.
394 pirq_entries[MAX_PIRQS-i-1] = ints[i+1];
399 __setup("pirq=", ioapic_pirq_setup);
402 * Find the IRQ entry number of a certain pin.
404 static int find_irq_entry(int apic, int pin, int type)
408 for (i = 0; i < mp_irq_entries; i++)
409 if (mp_irqs[i].mpc_irqtype == type &&
410 (mp_irqs[i].mpc_dstapic == mp_ioapics[apic].mpc_apicid ||
411 mp_irqs[i].mpc_dstapic == MP_APIC_ALL) &&
412 mp_irqs[i].mpc_dstirq == pin)
419 * Find the pin to which IRQ[irq] (ISA) is connected
421 static int __init find_isa_irq_pin(int irq, int type)
425 for (i = 0; i < mp_irq_entries; i++) {
426 int lbus = mp_irqs[i].mpc_srcbus;
428 if ((mp_bus_id_to_type[lbus] == MP_BUS_ISA ||
429 mp_bus_id_to_type[lbus] == MP_BUS_EISA ||
430 mp_bus_id_to_type[lbus] == MP_BUS_MCA) &&
431 (mp_irqs[i].mpc_irqtype == type) &&
432 (mp_irqs[i].mpc_srcbusirq == irq))
434 return mp_irqs[i].mpc_dstirq;
439 static int __init find_isa_irq_apic(int irq, int type)
443 for (i = 0; i < mp_irq_entries; i++) {
444 int lbus = mp_irqs[i].mpc_srcbus;
446 if ((mp_bus_id_to_type[lbus] == MP_BUS_ISA ||
447 mp_bus_id_to_type[lbus] == MP_BUS_EISA ||
448 mp_bus_id_to_type[lbus] == MP_BUS_MCA) &&
449 (mp_irqs[i].mpc_irqtype == type) &&
450 (mp_irqs[i].mpc_srcbusirq == irq))
453 if (i < mp_irq_entries) {
455 for(apic = 0; apic < nr_ioapics; apic++) {
456 if (mp_ioapics[apic].mpc_apicid == mp_irqs[i].mpc_dstapic)
465 * Find a specific PCI IRQ entry.
466 * Not an __init, possibly needed by modules
468 static int pin_2_irq(int idx, int apic, int pin);
470 int IO_APIC_get_PCI_irq_vector(int bus, int slot, int pin)
472 int apic, i, best_guess = -1;
474 apic_printk(APIC_DEBUG, "querying PCI -> IRQ mapping bus:%d, slot:%d, pin:%d.\n",
476 if (mp_bus_id_to_pci_bus[bus] == -1) {
477 apic_printk(APIC_VERBOSE, "PCI BIOS passed nonexistent PCI bus %d!\n", bus);
480 for (i = 0; i < mp_irq_entries; i++) {
481 int lbus = mp_irqs[i].mpc_srcbus;
483 for (apic = 0; apic < nr_ioapics; apic++)
484 if (mp_ioapics[apic].mpc_apicid == mp_irqs[i].mpc_dstapic ||
485 mp_irqs[i].mpc_dstapic == MP_APIC_ALL)
488 if ((mp_bus_id_to_type[lbus] == MP_BUS_PCI) &&
489 !mp_irqs[i].mpc_irqtype &&
491 (slot == ((mp_irqs[i].mpc_srcbusirq >> 2) & 0x1f))) {
492 int irq = pin_2_irq(i,apic,mp_irqs[i].mpc_dstirq);
494 if (!(apic || IO_APIC_IRQ(irq)))
497 if (pin == (mp_irqs[i].mpc_srcbusirq & 3))
500 * Use the first all-but-pin matching entry as a
501 * best-guess fuzzy result for broken mptables.
507 BUG_ON(best_guess >= NR_IRQS);
512 * EISA Edge/Level control register, ELCR
514 static int EISA_ELCR(unsigned int irq)
517 unsigned int port = 0x4d0 + (irq >> 3);
518 return (inb(port) >> (irq & 7)) & 1;
520 apic_printk(APIC_VERBOSE, "Broken MPtable reports ISA irq %d\n", irq);
524 /* EISA interrupts are always polarity zero and can be edge or level
525 * trigger depending on the ELCR value. If an interrupt is listed as
526 * EISA conforming in the MP table, that means its trigger type must
527 * be read in from the ELCR */
529 #define default_EISA_trigger(idx) (EISA_ELCR(mp_irqs[idx].mpc_srcbusirq))
530 #define default_EISA_polarity(idx) (0)
532 /* ISA interrupts are always polarity zero edge triggered,
533 * when listed as conforming in the MP table. */
535 #define default_ISA_trigger(idx) (0)
536 #define default_ISA_polarity(idx) (0)
538 /* PCI interrupts are always polarity one level triggered,
539 * when listed as conforming in the MP table. */
541 #define default_PCI_trigger(idx) (1)
542 #define default_PCI_polarity(idx) (1)
544 /* MCA interrupts are always polarity zero level triggered,
545 * when listed as conforming in the MP table. */
547 #define default_MCA_trigger(idx) (1)
548 #define default_MCA_polarity(idx) (0)
550 static int __init MPBIOS_polarity(int idx)
552 int bus = mp_irqs[idx].mpc_srcbus;
556 * Determine IRQ line polarity (high active or low active):
558 switch (mp_irqs[idx].mpc_irqflag & 3)
560 case 0: /* conforms, ie. bus-type dependent polarity */
562 switch (mp_bus_id_to_type[bus])
564 case MP_BUS_ISA: /* ISA pin */
566 polarity = default_ISA_polarity(idx);
569 case MP_BUS_EISA: /* EISA pin */
571 polarity = default_EISA_polarity(idx);
574 case MP_BUS_PCI: /* PCI pin */
576 polarity = default_PCI_polarity(idx);
579 case MP_BUS_MCA: /* MCA pin */
581 polarity = default_MCA_polarity(idx);
586 printk(KERN_WARNING "broken BIOS!!\n");
593 case 1: /* high active */
598 case 2: /* reserved */
600 printk(KERN_WARNING "broken BIOS!!\n");
604 case 3: /* low active */
609 default: /* invalid */
611 printk(KERN_WARNING "broken BIOS!!\n");
619 static int MPBIOS_trigger(int idx)
621 int bus = mp_irqs[idx].mpc_srcbus;
625 * Determine IRQ trigger mode (edge or level sensitive):
627 switch ((mp_irqs[idx].mpc_irqflag>>2) & 3)
629 case 0: /* conforms, ie. bus-type dependent */
631 switch (mp_bus_id_to_type[bus])
633 case MP_BUS_ISA: /* ISA pin */
635 trigger = default_ISA_trigger(idx);
638 case MP_BUS_EISA: /* EISA pin */
640 trigger = default_EISA_trigger(idx);
643 case MP_BUS_PCI: /* PCI pin */
645 trigger = default_PCI_trigger(idx);
648 case MP_BUS_MCA: /* MCA pin */
650 trigger = default_MCA_trigger(idx);
655 printk(KERN_WARNING "broken BIOS!!\n");
667 case 2: /* reserved */
669 printk(KERN_WARNING "broken BIOS!!\n");
678 default: /* invalid */
680 printk(KERN_WARNING "broken BIOS!!\n");
688 static inline int irq_polarity(int idx)
690 return MPBIOS_polarity(idx);
693 static inline int irq_trigger(int idx)
695 return MPBIOS_trigger(idx);
698 static int next_irq = 16;
701 * gsi_irq_sharing -- Name overload! "irq" can be either a legacy IRQ
702 * in the range 0-15, a linux IRQ in the range 0-223, or a GSI number
703 * from ACPI, which can reach 800 in large boxen.
705 * Compact the sparse GSI space into a sequential IRQ series and reuse
706 * vectors if possible.
708 int gsi_irq_sharing(int gsi)
710 int i, tries, vector;
712 BUG_ON(gsi >= NR_IRQ_VECTORS);
714 if (platform_legacy_irq(gsi))
717 if (gsi_2_irq[gsi] != 0xFF)
718 return (int)gsi_2_irq[gsi];
722 vector = assign_irq_vector(gsi);
725 * Sharing vectors means sharing IRQs, so scan irq_vectors for previous
726 * use of vector and if found, return that IRQ. However, we never want
727 * to share legacy IRQs, which usually have a different trigger mode
730 for (i = 0; i < NR_IRQS; i++)
731 if (IO_APIC_VECTOR(i) == vector)
733 if (platform_legacy_irq(i)) {
735 IO_APIC_VECTOR(i) = 0;
738 panic("gsi_irq_sharing: didn't find an IRQ using vector 0x%02X for GSI %d", vector, gsi);
742 printk(KERN_INFO "GSI %d sharing vector 0x%02X and IRQ %d\n",
748 BUG_ON(i >= NR_IRQS);
750 IO_APIC_VECTOR(i) = vector;
751 printk(KERN_INFO "GSI %d assigned vector 0x%02X and IRQ %d\n",
756 static int pin_2_irq(int idx, int apic, int pin)
759 int bus = mp_irqs[idx].mpc_srcbus;
762 * Debugging check, we are in big trouble if this message pops up!
764 if (mp_irqs[idx].mpc_dstirq != pin)
765 printk(KERN_ERR "broken BIOS or MPTABLE parser, ayiee!!\n");
767 switch (mp_bus_id_to_type[bus])
769 case MP_BUS_ISA: /* ISA pin */
773 irq = mp_irqs[idx].mpc_srcbusirq;
776 case MP_BUS_PCI: /* PCI pin */
779 * PCI IRQs are mapped in order
783 irq += nr_ioapic_registers[i++];
785 irq = gsi_irq_sharing(irq);
790 printk(KERN_ERR "unknown bus type %d.\n",bus);
795 BUG_ON(irq >= NR_IRQS);
798 * PCI IRQ command line redirection. Yes, limits are hardcoded.
800 if ((pin >= 16) && (pin <= 23)) {
801 if (pirq_entries[pin-16] != -1) {
802 if (!pirq_entries[pin-16]) {
803 apic_printk(APIC_VERBOSE, "disabling PIRQ%d\n", pin-16);
805 irq = pirq_entries[pin-16];
806 apic_printk(APIC_VERBOSE, "using PIRQ%d -> IRQ %d\n",
811 BUG_ON(irq >= NR_IRQS);
815 static inline int IO_APIC_irq_trigger(int irq)
819 for (apic = 0; apic < nr_ioapics; apic++) {
820 for (pin = 0; pin < nr_ioapic_registers[apic]; pin++) {
821 idx = find_irq_entry(apic,pin,mp_INT);
822 if ((idx != -1) && (irq == pin_2_irq(idx,apic,pin)))
823 return irq_trigger(idx);
827 * nonexistent IRQs are edge default
832 /* irq_vectors is indexed by the sum of all RTEs in all I/O APICs. */
833 u8 irq_vector[NR_IRQ_VECTORS] __read_mostly = { FIRST_DEVICE_VECTOR , 0 };
835 int assign_irq_vector(int irq)
837 static int current_vector = FIRST_DEVICE_VECTOR, offset = 0;
841 BUG_ON(irq != AUTO_ASSIGN && (unsigned)irq >= NR_IRQ_VECTORS);
843 spin_lock_irqsave(&vector_lock, flags);
845 if (irq != AUTO_ASSIGN && IO_APIC_VECTOR(irq) > 0) {
846 spin_unlock_irqrestore(&vector_lock, flags);
847 return IO_APIC_VECTOR(irq);
851 if (current_vector == IA32_SYSCALL_VECTOR)
854 if (current_vector >= FIRST_SYSTEM_VECTOR) {
855 /* If we run out of vectors on large boxen, must share them. */
856 offset = (offset + 1) % 8;
857 current_vector = FIRST_DEVICE_VECTOR + offset;
860 vector = current_vector;
861 vector_irq[vector] = irq;
862 if (irq != AUTO_ASSIGN)
863 IO_APIC_VECTOR(irq) = vector;
865 spin_unlock_irqrestore(&vector_lock, flags);
870 extern void (*interrupt[NR_IRQS])(void);
871 static struct hw_interrupt_type ioapic_level_type;
872 static struct hw_interrupt_type ioapic_edge_type;
874 #define IOAPIC_AUTO -1
875 #define IOAPIC_EDGE 0
876 #define IOAPIC_LEVEL 1
878 static void ioapic_register_intr(int irq, int vector, unsigned long trigger)
882 idx = use_pci_vector() && !platform_legacy_irq(irq) ? vector : irq;
884 if ((trigger == IOAPIC_AUTO && IO_APIC_irq_trigger(irq)) ||
885 trigger == IOAPIC_LEVEL)
886 irq_desc[idx].chip = &ioapic_level_type;
888 irq_desc[idx].chip = &ioapic_edge_type;
889 set_intr_gate(vector, interrupt[idx]);
892 static void __init setup_IO_APIC_irqs(void)
894 struct IO_APIC_route_entry entry;
895 int apic, pin, idx, irq, first_notcon = 1, vector;
898 apic_printk(APIC_VERBOSE, KERN_DEBUG "init IO_APIC IRQs\n");
900 for (apic = 0; apic < nr_ioapics; apic++) {
901 for (pin = 0; pin < nr_ioapic_registers[apic]; pin++) {
904 * add it to the IO-APIC irq-routing table:
906 memset(&entry,0,sizeof(entry));
908 entry.delivery_mode = INT_DELIVERY_MODE;
909 entry.dest_mode = INT_DEST_MODE;
910 entry.mask = 0; /* enable IRQ */
911 entry.dest.logical.logical_dest = cpu_mask_to_apicid(TARGET_CPUS);
913 idx = find_irq_entry(apic,pin,mp_INT);
916 apic_printk(APIC_VERBOSE, KERN_DEBUG " IO-APIC (apicid-pin) %d-%d", mp_ioapics[apic].mpc_apicid, pin);
919 apic_printk(APIC_VERBOSE, ", %d-%d", mp_ioapics[apic].mpc_apicid, pin);
923 entry.trigger = irq_trigger(idx);
924 entry.polarity = irq_polarity(idx);
926 if (irq_trigger(idx)) {
929 entry.dest.logical.logical_dest = cpu_mask_to_apicid(TARGET_CPUS);
932 irq = pin_2_irq(idx, apic, pin);
933 add_pin_to_irq(irq, apic, pin);
935 if (!apic && !IO_APIC_IRQ(irq))
938 if (IO_APIC_IRQ(irq)) {
939 vector = assign_irq_vector(irq);
940 entry.vector = vector;
942 ioapic_register_intr(irq, vector, IOAPIC_AUTO);
943 if (!apic && (irq < 16))
944 disable_8259A_irq(irq);
946 spin_lock_irqsave(&ioapic_lock, flags);
947 io_apic_write(apic, 0x11+2*pin, *(((int *)&entry)+1));
948 io_apic_write(apic, 0x10+2*pin, *(((int *)&entry)+0));
949 set_native_irq_info(irq, TARGET_CPUS);
950 spin_unlock_irqrestore(&ioapic_lock, flags);
955 apic_printk(APIC_VERBOSE," not connected.\n");
959 * Set up the 8259A-master output pin as broadcast to all
962 static void __init setup_ExtINT_IRQ0_pin(unsigned int apic, unsigned int pin, int vector)
964 struct IO_APIC_route_entry entry;
967 memset(&entry,0,sizeof(entry));
969 disable_8259A_irq(0);
972 apic_write(APIC_LVT0, APIC_LVT_MASKED | APIC_DM_EXTINT);
975 * We use logical delivery to get the timer IRQ
978 entry.dest_mode = INT_DEST_MODE;
979 entry.mask = 0; /* unmask IRQ now */
980 entry.dest.logical.logical_dest = cpu_mask_to_apicid(TARGET_CPUS);
981 entry.delivery_mode = INT_DELIVERY_MODE;
984 entry.vector = vector;
987 * The timer IRQ doesn't have to know that behind the
988 * scene we have a 8259A-master in AEOI mode ...
990 irq_desc[0].chip = &ioapic_edge_type;
993 * Add it to the IO-APIC irq-routing table:
995 spin_lock_irqsave(&ioapic_lock, flags);
996 io_apic_write(apic, 0x11+2*pin, *(((int *)&entry)+1));
997 io_apic_write(apic, 0x10+2*pin, *(((int *)&entry)+0));
998 spin_unlock_irqrestore(&ioapic_lock, flags);
1000 enable_8259A_irq(0);
1003 void __init UNEXPECTED_IO_APIC(void)
1007 void __apicdebuginit print_IO_APIC(void)
1010 union IO_APIC_reg_00 reg_00;
1011 union IO_APIC_reg_01 reg_01;
1012 union IO_APIC_reg_02 reg_02;
1013 unsigned long flags;
1015 if (apic_verbosity == APIC_QUIET)
1018 printk(KERN_DEBUG "number of MP IRQ sources: %d.\n", mp_irq_entries);
1019 for (i = 0; i < nr_ioapics; i++)
1020 printk(KERN_DEBUG "number of IO-APIC #%d registers: %d.\n",
1021 mp_ioapics[i].mpc_apicid, nr_ioapic_registers[i]);
1024 * We are a bit conservative about what we expect. We have to
1025 * know about every hardware change ASAP.
1027 printk(KERN_INFO "testing the IO APIC.......................\n");
1029 for (apic = 0; apic < nr_ioapics; apic++) {
1031 spin_lock_irqsave(&ioapic_lock, flags);
1032 reg_00.raw = io_apic_read(apic, 0);
1033 reg_01.raw = io_apic_read(apic, 1);
1034 if (reg_01.bits.version >= 0x10)
1035 reg_02.raw = io_apic_read(apic, 2);
1036 spin_unlock_irqrestore(&ioapic_lock, flags);
1039 printk(KERN_DEBUG "IO APIC #%d......\n", mp_ioapics[apic].mpc_apicid);
1040 printk(KERN_DEBUG ".... register #00: %08X\n", reg_00.raw);
1041 printk(KERN_DEBUG "....... : physical APIC id: %02X\n", reg_00.bits.ID);
1042 if (reg_00.bits.__reserved_1 || reg_00.bits.__reserved_2)
1043 UNEXPECTED_IO_APIC();
1045 printk(KERN_DEBUG ".... register #01: %08X\n", *(int *)®_01);
1046 printk(KERN_DEBUG "....... : max redirection entries: %04X\n", reg_01.bits.entries);
1047 if ( (reg_01.bits.entries != 0x0f) && /* older (Neptune) boards */
1048 (reg_01.bits.entries != 0x17) && /* typical ISA+PCI boards */
1049 (reg_01.bits.entries != 0x1b) && /* Compaq Proliant boards */
1050 (reg_01.bits.entries != 0x1f) && /* dual Xeon boards */
1051 (reg_01.bits.entries != 0x22) && /* bigger Xeon boards */
1052 (reg_01.bits.entries != 0x2E) &&
1053 (reg_01.bits.entries != 0x3F) &&
1054 (reg_01.bits.entries != 0x03)
1056 UNEXPECTED_IO_APIC();
1058 printk(KERN_DEBUG "....... : PRQ implemented: %X\n", reg_01.bits.PRQ);
1059 printk(KERN_DEBUG "....... : IO APIC version: %04X\n", reg_01.bits.version);
1060 if ( (reg_01.bits.version != 0x01) && /* 82489DX IO-APICs */
1061 (reg_01.bits.version != 0x02) && /* 82801BA IO-APICs (ICH2) */
1062 (reg_01.bits.version != 0x10) && /* oldest IO-APICs */
1063 (reg_01.bits.version != 0x11) && /* Pentium/Pro IO-APICs */
1064 (reg_01.bits.version != 0x13) && /* Xeon IO-APICs */
1065 (reg_01.bits.version != 0x20) /* Intel P64H (82806 AA) */
1067 UNEXPECTED_IO_APIC();
1068 if (reg_01.bits.__reserved_1 || reg_01.bits.__reserved_2)
1069 UNEXPECTED_IO_APIC();
1071 if (reg_01.bits.version >= 0x10) {
1072 printk(KERN_DEBUG ".... register #02: %08X\n", reg_02.raw);
1073 printk(KERN_DEBUG "....... : arbitration: %02X\n", reg_02.bits.arbitration);
1074 if (reg_02.bits.__reserved_1 || reg_02.bits.__reserved_2)
1075 UNEXPECTED_IO_APIC();
1078 printk(KERN_DEBUG ".... IRQ redirection table:\n");
1080 printk(KERN_DEBUG " NR Log Phy Mask Trig IRR Pol"
1081 " Stat Dest Deli Vect: \n");
1083 for (i = 0; i <= reg_01.bits.entries; i++) {
1084 struct IO_APIC_route_entry entry;
1086 spin_lock_irqsave(&ioapic_lock, flags);
1087 *(((int *)&entry)+0) = io_apic_read(apic, 0x10+i*2);
1088 *(((int *)&entry)+1) = io_apic_read(apic, 0x11+i*2);
1089 spin_unlock_irqrestore(&ioapic_lock, flags);
1091 printk(KERN_DEBUG " %02x %03X %02X ",
1093 entry.dest.logical.logical_dest,
1094 entry.dest.physical.physical_dest
1097 printk("%1d %1d %1d %1d %1d %1d %1d %02X\n",
1102 entry.delivery_status,
1104 entry.delivery_mode,
1109 if (use_pci_vector())
1110 printk(KERN_INFO "Using vector-based indexing\n");
1111 printk(KERN_DEBUG "IRQ to pin mappings:\n");
1112 for (i = 0; i < NR_IRQS; i++) {
1113 struct irq_pin_list *entry = irq_2_pin + i;
1116 if (use_pci_vector() && !platform_legacy_irq(i))
1117 printk(KERN_DEBUG "IRQ%d ", IO_APIC_VECTOR(i));
1119 printk(KERN_DEBUG "IRQ%d ", i);
1121 printk("-> %d:%d", entry->apic, entry->pin);
1124 entry = irq_2_pin + entry->next;
1129 printk(KERN_INFO ".................................... done.\n");
1136 static __apicdebuginit void print_APIC_bitfield (int base)
1141 if (apic_verbosity == APIC_QUIET)
1144 printk(KERN_DEBUG "0123456789abcdef0123456789abcdef\n" KERN_DEBUG);
1145 for (i = 0; i < 8; i++) {
1146 v = apic_read(base + i*0x10);
1147 for (j = 0; j < 32; j++) {
1157 void __apicdebuginit print_local_APIC(void * dummy)
1159 unsigned int v, ver, maxlvt;
1161 if (apic_verbosity == APIC_QUIET)
1164 printk("\n" KERN_DEBUG "printing local APIC contents on CPU#%d/%d:\n",
1165 smp_processor_id(), hard_smp_processor_id());
1166 v = apic_read(APIC_ID);
1167 printk(KERN_INFO "... APIC ID: %08x (%01x)\n", v, GET_APIC_ID(v));
1168 v = apic_read(APIC_LVR);
1169 printk(KERN_INFO "... APIC VERSION: %08x\n", v);
1170 ver = GET_APIC_VERSION(v);
1171 maxlvt = get_maxlvt();
1173 v = apic_read(APIC_TASKPRI);
1174 printk(KERN_DEBUG "... APIC TASKPRI: %08x (%02x)\n", v, v & APIC_TPRI_MASK);
1176 v = apic_read(APIC_ARBPRI);
1177 printk(KERN_DEBUG "... APIC ARBPRI: %08x (%02x)\n", v,
1178 v & APIC_ARBPRI_MASK);
1179 v = apic_read(APIC_PROCPRI);
1180 printk(KERN_DEBUG "... APIC PROCPRI: %08x\n", v);
1182 v = apic_read(APIC_EOI);
1183 printk(KERN_DEBUG "... APIC EOI: %08x\n", v);
1184 v = apic_read(APIC_RRR);
1185 printk(KERN_DEBUG "... APIC RRR: %08x\n", v);
1186 v = apic_read(APIC_LDR);
1187 printk(KERN_DEBUG "... APIC LDR: %08x\n", v);
1188 v = apic_read(APIC_DFR);
1189 printk(KERN_DEBUG "... APIC DFR: %08x\n", v);
1190 v = apic_read(APIC_SPIV);
1191 printk(KERN_DEBUG "... APIC SPIV: %08x\n", v);
1193 printk(KERN_DEBUG "... APIC ISR field:\n");
1194 print_APIC_bitfield(APIC_ISR);
1195 printk(KERN_DEBUG "... APIC TMR field:\n");
1196 print_APIC_bitfield(APIC_TMR);
1197 printk(KERN_DEBUG "... APIC IRR field:\n");
1198 print_APIC_bitfield(APIC_IRR);
1200 v = apic_read(APIC_ESR);
1201 printk(KERN_DEBUG "... APIC ESR: %08x\n", v);
1203 v = apic_read(APIC_ICR);
1204 printk(KERN_DEBUG "... APIC ICR: %08x\n", v);
1205 v = apic_read(APIC_ICR2);
1206 printk(KERN_DEBUG "... APIC ICR2: %08x\n", v);
1208 v = apic_read(APIC_LVTT);
1209 printk(KERN_DEBUG "... APIC LVTT: %08x\n", v);
1211 if (maxlvt > 3) { /* PC is LVT#4. */
1212 v = apic_read(APIC_LVTPC);
1213 printk(KERN_DEBUG "... APIC LVTPC: %08x\n", v);
1215 v = apic_read(APIC_LVT0);
1216 printk(KERN_DEBUG "... APIC LVT0: %08x\n", v);
1217 v = apic_read(APIC_LVT1);
1218 printk(KERN_DEBUG "... APIC LVT1: %08x\n", v);
1220 if (maxlvt > 2) { /* ERR is LVT#3. */
1221 v = apic_read(APIC_LVTERR);
1222 printk(KERN_DEBUG "... APIC LVTERR: %08x\n", v);
1225 v = apic_read(APIC_TMICT);
1226 printk(KERN_DEBUG "... APIC TMICT: %08x\n", v);
1227 v = apic_read(APIC_TMCCT);
1228 printk(KERN_DEBUG "... APIC TMCCT: %08x\n", v);
1229 v = apic_read(APIC_TDCR);
1230 printk(KERN_DEBUG "... APIC TDCR: %08x\n", v);
1234 void print_all_local_APICs (void)
1236 on_each_cpu(print_local_APIC, NULL, 1, 1);
1239 void __apicdebuginit print_PIC(void)
1242 unsigned long flags;
1244 if (apic_verbosity == APIC_QUIET)
1247 printk(KERN_DEBUG "\nprinting PIC contents\n");
1249 spin_lock_irqsave(&i8259A_lock, flags);
1251 v = inb(0xa1) << 8 | inb(0x21);
1252 printk(KERN_DEBUG "... PIC IMR: %04x\n", v);
1254 v = inb(0xa0) << 8 | inb(0x20);
1255 printk(KERN_DEBUG "... PIC IRR: %04x\n", v);
1259 v = inb(0xa0) << 8 | inb(0x20);
1263 spin_unlock_irqrestore(&i8259A_lock, flags);
1265 printk(KERN_DEBUG "... PIC ISR: %04x\n", v);
1267 v = inb(0x4d1) << 8 | inb(0x4d0);
1268 printk(KERN_DEBUG "... PIC ELCR: %04x\n", v);
1273 static void __init enable_IO_APIC(void)
1275 union IO_APIC_reg_01 reg_01;
1276 int i8259_apic, i8259_pin;
1278 unsigned long flags;
1280 for (i = 0; i < PIN_MAP_SIZE; i++) {
1281 irq_2_pin[i].pin = -1;
1282 irq_2_pin[i].next = 0;
1285 for (i = 0; i < MAX_PIRQS; i++)
1286 pirq_entries[i] = -1;
1289 * The number of IO-APIC IRQ registers (== #pins):
1291 for (apic = 0; apic < nr_ioapics; apic++) {
1292 spin_lock_irqsave(&ioapic_lock, flags);
1293 reg_01.raw = io_apic_read(apic, 1);
1294 spin_unlock_irqrestore(&ioapic_lock, flags);
1295 nr_ioapic_registers[apic] = reg_01.bits.entries+1;
1297 for(apic = 0; apic < nr_ioapics; apic++) {
1299 /* See if any of the pins is in ExtINT mode */
1300 for (pin = 0; pin < nr_ioapic_registers[apic]; pin++) {
1301 struct IO_APIC_route_entry entry;
1302 spin_lock_irqsave(&ioapic_lock, flags);
1303 *(((int *)&entry) + 0) = io_apic_read(apic, 0x10 + 2 * pin);
1304 *(((int *)&entry) + 1) = io_apic_read(apic, 0x11 + 2 * pin);
1305 spin_unlock_irqrestore(&ioapic_lock, flags);
1308 /* If the interrupt line is enabled and in ExtInt mode
1309 * I have found the pin where the i8259 is connected.
1311 if ((entry.mask == 0) && (entry.delivery_mode == dest_ExtINT)) {
1312 ioapic_i8259.apic = apic;
1313 ioapic_i8259.pin = pin;
1319 /* Look to see what if the MP table has reported the ExtINT */
1320 i8259_pin = find_isa_irq_pin(0, mp_ExtINT);
1321 i8259_apic = find_isa_irq_apic(0, mp_ExtINT);
1322 /* Trust the MP table if nothing is setup in the hardware */
1323 if ((ioapic_i8259.pin == -1) && (i8259_pin >= 0)) {
1324 printk(KERN_WARNING "ExtINT not setup in hardware but reported by MP table\n");
1325 ioapic_i8259.pin = i8259_pin;
1326 ioapic_i8259.apic = i8259_apic;
1328 /* Complain if the MP table and the hardware disagree */
1329 if (((ioapic_i8259.apic != i8259_apic) || (ioapic_i8259.pin != i8259_pin)) &&
1330 (i8259_pin >= 0) && (ioapic_i8259.pin >= 0))
1332 printk(KERN_WARNING "ExtINT in hardware and MP table differ\n");
1336 * Do not trust the IO-APIC being empty at bootup
1342 * Not an __init, needed by the reboot code
1344 void disable_IO_APIC(void)
1347 * Clear the IO-APIC before rebooting:
1352 * If the i8259 is routed through an IOAPIC
1353 * Put that IOAPIC in virtual wire mode
1354 * so legacy interrupts can be delivered.
1356 if (ioapic_i8259.pin != -1) {
1357 struct IO_APIC_route_entry entry;
1358 unsigned long flags;
1360 memset(&entry, 0, sizeof(entry));
1361 entry.mask = 0; /* Enabled */
1362 entry.trigger = 0; /* Edge */
1364 entry.polarity = 0; /* High */
1365 entry.delivery_status = 0;
1366 entry.dest_mode = 0; /* Physical */
1367 entry.delivery_mode = dest_ExtINT; /* ExtInt */
1369 entry.dest.physical.physical_dest =
1370 GET_APIC_ID(apic_read(APIC_ID));
1373 * Add it to the IO-APIC irq-routing table:
1375 spin_lock_irqsave(&ioapic_lock, flags);
1376 io_apic_write(ioapic_i8259.apic, 0x11+2*ioapic_i8259.pin,
1377 *(((int *)&entry)+1));
1378 io_apic_write(ioapic_i8259.apic, 0x10+2*ioapic_i8259.pin,
1379 *(((int *)&entry)+0));
1380 spin_unlock_irqrestore(&ioapic_lock, flags);
1383 disconnect_bsp_APIC(ioapic_i8259.pin != -1);
1387 * function to set the IO-APIC physical IDs based on the
1388 * values stored in the MPC table.
1390 * by Matt Domsch <Matt_Domsch@dell.com> Tue Dec 21 12:25:05 CST 1999
1393 static void __init setup_ioapic_ids_from_mpc (void)
1395 union IO_APIC_reg_00 reg_00;
1398 unsigned char old_id;
1399 unsigned long flags;
1402 * Set the IOAPIC ID to the value stored in the MPC table.
1404 for (apic = 0; apic < nr_ioapics; apic++) {
1406 /* Read the register 0 value */
1407 spin_lock_irqsave(&ioapic_lock, flags);
1408 reg_00.raw = io_apic_read(apic, 0);
1409 spin_unlock_irqrestore(&ioapic_lock, flags);
1411 old_id = mp_ioapics[apic].mpc_apicid;
1414 printk(KERN_INFO "Using IO-APIC %d\n", mp_ioapics[apic].mpc_apicid);
1418 * We need to adjust the IRQ routing table
1419 * if the ID changed.
1421 if (old_id != mp_ioapics[apic].mpc_apicid)
1422 for (i = 0; i < mp_irq_entries; i++)
1423 if (mp_irqs[i].mpc_dstapic == old_id)
1424 mp_irqs[i].mpc_dstapic
1425 = mp_ioapics[apic].mpc_apicid;
1428 * Read the right value from the MPC table and
1429 * write it into the ID register.
1431 apic_printk(APIC_VERBOSE,KERN_INFO "...changing IO-APIC physical APIC ID to %d ...",
1432 mp_ioapics[apic].mpc_apicid);
1434 reg_00.bits.ID = mp_ioapics[apic].mpc_apicid;
1435 spin_lock_irqsave(&ioapic_lock, flags);
1436 io_apic_write(apic, 0, reg_00.raw);
1437 spin_unlock_irqrestore(&ioapic_lock, flags);
1442 spin_lock_irqsave(&ioapic_lock, flags);
1443 reg_00.raw = io_apic_read(apic, 0);
1444 spin_unlock_irqrestore(&ioapic_lock, flags);
1445 if (reg_00.bits.ID != mp_ioapics[apic].mpc_apicid)
1446 printk("could not set ID!\n");
1448 apic_printk(APIC_VERBOSE," ok.\n");
1453 * There is a nasty bug in some older SMP boards, their mptable lies
1454 * about the timer IRQ. We do the following to work around the situation:
1456 * - timer IRQ defaults to IO-APIC IRQ
1457 * - if this function detects that timer IRQs are defunct, then we fall
1458 * back to ISA timer IRQs
1460 static int __init timer_irq_works(void)
1462 unsigned long t1 = jiffies;
1465 /* Let ten ticks pass... */
1466 mdelay((10 * 1000) / HZ);
1469 * Expect a few ticks at least, to be sure some possible
1470 * glue logic does not lock up after one or two first
1471 * ticks in a non-ExtINT mode. Also the local APIC
1472 * might have cached one ExtINT interrupt. Finally, at
1473 * least one tick may be lost due to delays.
1477 if (jiffies - t1 > 4)
1483 * In the SMP+IOAPIC case it might happen that there are an unspecified
1484 * number of pending IRQ events unhandled. These cases are very rare,
1485 * so we 'resend' these IRQs via IPIs, to the same CPU. It's much
1486 * better to do it this way as thus we do not have to be aware of
1487 * 'pending' interrupts in the IRQ path, except at this point.
1490 * Edge triggered needs to resend any interrupt
1491 * that was delayed but this is now handled in the device
1496 * Starting up a edge-triggered IO-APIC interrupt is
1497 * nasty - we need to make sure that we get the edge.
1498 * If it is already asserted for some reason, we need
1499 * return 1 to indicate that is was pending.
1501 * This is not complete - we should be able to fake
1502 * an edge even if it isn't on the 8259A...
1505 static unsigned int startup_edge_ioapic_irq(unsigned int irq)
1507 int was_pending = 0;
1508 unsigned long flags;
1510 spin_lock_irqsave(&ioapic_lock, flags);
1512 disable_8259A_irq(irq);
1513 if (i8259A_irq_pending(irq))
1516 __unmask_IO_APIC_irq(irq);
1517 spin_unlock_irqrestore(&ioapic_lock, flags);
1523 * Once we have recorded IRQ_PENDING already, we can mask the
1524 * interrupt for real. This prevents IRQ storms from unhandled
1527 static void ack_edge_ioapic_irq(unsigned int irq)
1530 if ((irq_desc[irq].status & (IRQ_PENDING | IRQ_DISABLED))
1531 == (IRQ_PENDING | IRQ_DISABLED))
1532 mask_IO_APIC_irq(irq);
1537 * Level triggered interrupts can just be masked,
1538 * and shutting down and starting up the interrupt
1539 * is the same as enabling and disabling them -- except
1540 * with a startup need to return a "was pending" value.
1542 * Level triggered interrupts are special because we
1543 * do not touch any IO-APIC register while handling
1544 * them. We ack the APIC in the end-IRQ handler, not
1545 * in the start-IRQ-handler. Protection against reentrance
1546 * from the same interrupt is still provided, both by the
1547 * generic IRQ layer and by the fact that an unacked local
1548 * APIC does not accept IRQs.
1550 static unsigned int startup_level_ioapic_irq (unsigned int irq)
1552 unmask_IO_APIC_irq(irq);
1554 return 0; /* don't check for pending */
1557 static void end_level_ioapic_irq (unsigned int irq)
1563 #ifdef CONFIG_PCI_MSI
1564 static unsigned int startup_edge_ioapic_vector(unsigned int vector)
1566 int irq = vector_to_irq(vector);
1568 return startup_edge_ioapic_irq(irq);
1571 static void ack_edge_ioapic_vector(unsigned int vector)
1573 int irq = vector_to_irq(vector);
1575 move_native_irq(vector);
1576 ack_edge_ioapic_irq(irq);
1579 static unsigned int startup_level_ioapic_vector (unsigned int vector)
1581 int irq = vector_to_irq(vector);
1583 return startup_level_ioapic_irq (irq);
1586 static void end_level_ioapic_vector (unsigned int vector)
1588 int irq = vector_to_irq(vector);
1590 move_native_irq(vector);
1591 end_level_ioapic_irq(irq);
1594 static void mask_IO_APIC_vector (unsigned int vector)
1596 int irq = vector_to_irq(vector);
1598 mask_IO_APIC_irq(irq);
1601 static void unmask_IO_APIC_vector (unsigned int vector)
1603 int irq = vector_to_irq(vector);
1605 unmask_IO_APIC_irq(irq);
1609 static void set_ioapic_affinity_vector (unsigned int vector,
1612 int irq = vector_to_irq(vector);
1614 set_native_irq_info(vector, cpu_mask);
1615 set_ioapic_affinity_irq(irq, cpu_mask);
1617 #endif // CONFIG_SMP
1618 #endif // CONFIG_PCI_MSI
1620 static int ioapic_retrigger(unsigned int irq)
1622 send_IPI_self(IO_APIC_VECTOR(irq));
1628 * Level and edge triggered IO-APIC interrupts need different handling,
1629 * so we use two separate IRQ descriptors. Edge triggered IRQs can be
1630 * handled with the level-triggered descriptor, but that one has slightly
1631 * more overhead. Level-triggered interrupts cannot be handled with the
1632 * edge-triggered handler, without risking IRQ storms and other ugly
1636 static struct hw_interrupt_type ioapic_edge_type __read_mostly = {
1637 .typename = "IO-APIC-edge",
1638 .startup = startup_edge_ioapic,
1639 .shutdown = shutdown_edge_ioapic,
1640 .enable = enable_edge_ioapic,
1641 .disable = disable_edge_ioapic,
1642 .ack = ack_edge_ioapic,
1643 .end = end_edge_ioapic,
1645 .set_affinity = set_ioapic_affinity,
1647 .retrigger = ioapic_retrigger,
1650 static struct hw_interrupt_type ioapic_level_type __read_mostly = {
1651 .typename = "IO-APIC-level",
1652 .startup = startup_level_ioapic,
1653 .shutdown = shutdown_level_ioapic,
1654 .enable = enable_level_ioapic,
1655 .disable = disable_level_ioapic,
1656 .ack = mask_and_ack_level_ioapic,
1657 .end = end_level_ioapic,
1659 .set_affinity = set_ioapic_affinity,
1661 .retrigger = ioapic_retrigger,
1664 static inline void init_IO_APIC_traps(void)
1669 * NOTE! The local APIC isn't very good at handling
1670 * multiple interrupts at the same interrupt level.
1671 * As the interrupt level is determined by taking the
1672 * vector number and shifting that right by 4, we
1673 * want to spread these out a bit so that they don't
1674 * all fall in the same interrupt level.
1676 * Also, we've got to be careful not to trash gate
1677 * 0x80, because int 0x80 is hm, kind of importantish. ;)
1679 for (irq = 0; irq < NR_IRQS ; irq++) {
1681 if (use_pci_vector()) {
1682 if (!platform_legacy_irq(tmp))
1683 if ((tmp = vector_to_irq(tmp)) == -1)
1686 if (IO_APIC_IRQ(tmp) && !IO_APIC_VECTOR(tmp)) {
1688 * Hmm.. We don't have an entry for this,
1689 * so default to an old-fashioned 8259
1690 * interrupt if we can..
1693 make_8259A_irq(irq);
1695 /* Strange. Oh, well.. */
1696 irq_desc[irq].chip = &no_irq_type;
1701 static void enable_lapic_irq (unsigned int irq)
1705 v = apic_read(APIC_LVT0);
1706 apic_write(APIC_LVT0, v & ~APIC_LVT_MASKED);
1709 static void disable_lapic_irq (unsigned int irq)
1713 v = apic_read(APIC_LVT0);
1714 apic_write(APIC_LVT0, v | APIC_LVT_MASKED);
1717 static void ack_lapic_irq (unsigned int irq)
1722 static void end_lapic_irq (unsigned int i) { /* nothing */ }
1724 static struct hw_interrupt_type lapic_irq_type __read_mostly = {
1725 .typename = "local-APIC-edge",
1726 .startup = NULL, /* startup_irq() not used for IRQ0 */
1727 .shutdown = NULL, /* shutdown_irq() not used for IRQ0 */
1728 .enable = enable_lapic_irq,
1729 .disable = disable_lapic_irq,
1730 .ack = ack_lapic_irq,
1731 .end = end_lapic_irq,
1734 static void setup_nmi (void)
1737 * Dirty trick to enable the NMI watchdog ...
1738 * We put the 8259A master into AEOI mode and
1739 * unmask on all local APICs LVT0 as NMI.
1741 * The idea to use the 8259A in AEOI mode ('8259A Virtual Wire')
1742 * is from Maciej W. Rozycki - so we do not have to EOI from
1743 * the NMI handler or the timer interrupt.
1745 printk(KERN_INFO "activating NMI Watchdog ...");
1747 enable_NMI_through_LVT0(NULL);
1753 * This looks a bit hackish but it's about the only one way of sending
1754 * a few INTA cycles to 8259As and any associated glue logic. ICR does
1755 * not support the ExtINT mode, unfortunately. We need to send these
1756 * cycles as some i82489DX-based boards have glue logic that keeps the
1757 * 8259A interrupt line asserted until INTA. --macro
1759 static inline void unlock_ExtINT_logic(void)
1762 struct IO_APIC_route_entry entry0, entry1;
1763 unsigned char save_control, save_freq_select;
1764 unsigned long flags;
1766 pin = find_isa_irq_pin(8, mp_INT);
1767 apic = find_isa_irq_apic(8, mp_INT);
1771 spin_lock_irqsave(&ioapic_lock, flags);
1772 *(((int *)&entry0) + 1) = io_apic_read(apic, 0x11 + 2 * pin);
1773 *(((int *)&entry0) + 0) = io_apic_read(apic, 0x10 + 2 * pin);
1774 spin_unlock_irqrestore(&ioapic_lock, flags);
1775 clear_IO_APIC_pin(apic, pin);
1777 memset(&entry1, 0, sizeof(entry1));
1779 entry1.dest_mode = 0; /* physical delivery */
1780 entry1.mask = 0; /* unmask IRQ now */
1781 entry1.dest.physical.physical_dest = hard_smp_processor_id();
1782 entry1.delivery_mode = dest_ExtINT;
1783 entry1.polarity = entry0.polarity;
1787 spin_lock_irqsave(&ioapic_lock, flags);
1788 io_apic_write(apic, 0x11 + 2 * pin, *(((int *)&entry1) + 1));
1789 io_apic_write(apic, 0x10 + 2 * pin, *(((int *)&entry1) + 0));
1790 spin_unlock_irqrestore(&ioapic_lock, flags);
1792 save_control = CMOS_READ(RTC_CONTROL);
1793 save_freq_select = CMOS_READ(RTC_FREQ_SELECT);
1794 CMOS_WRITE((save_freq_select & ~RTC_RATE_SELECT) | 0x6,
1796 CMOS_WRITE(save_control | RTC_PIE, RTC_CONTROL);
1801 if ((CMOS_READ(RTC_INTR_FLAGS) & RTC_PF) == RTC_PF)
1805 CMOS_WRITE(save_control, RTC_CONTROL);
1806 CMOS_WRITE(save_freq_select, RTC_FREQ_SELECT);
1807 clear_IO_APIC_pin(apic, pin);
1809 spin_lock_irqsave(&ioapic_lock, flags);
1810 io_apic_write(apic, 0x11 + 2 * pin, *(((int *)&entry0) + 1));
1811 io_apic_write(apic, 0x10 + 2 * pin, *(((int *)&entry0) + 0));
1812 spin_unlock_irqrestore(&ioapic_lock, flags);
1815 int timer_uses_ioapic_pin_0;
1818 * This code may look a bit paranoid, but it's supposed to cooperate with
1819 * a wide range of boards and BIOS bugs. Fortunately only the timer IRQ
1820 * is so screwy. Thanks to Brian Perkins for testing/hacking this beast
1821 * fanatically on his truly buggy board.
1823 * FIXME: really need to revamp this for modern platforms only.
1825 static inline void check_timer(void)
1827 int apic1, pin1, apic2, pin2;
1831 * get/set the timer IRQ vector:
1833 disable_8259A_irq(0);
1834 vector = assign_irq_vector(0);
1835 set_intr_gate(vector, interrupt[0]);
1838 * Subtle, code in do_timer_interrupt() expects an AEOI
1839 * mode for the 8259A whenever interrupts are routed
1840 * through I/O APICs. Also IRQ0 has to be enabled in
1841 * the 8259A which implies the virtual wire has to be
1842 * disabled in the local APIC.
1844 apic_write(APIC_LVT0, APIC_LVT_MASKED | APIC_DM_EXTINT);
1846 if (timer_over_8254 > 0)
1847 enable_8259A_irq(0);
1849 pin1 = find_isa_irq_pin(0, mp_INT);
1850 apic1 = find_isa_irq_apic(0, mp_INT);
1851 pin2 = ioapic_i8259.pin;
1852 apic2 = ioapic_i8259.apic;
1855 timer_uses_ioapic_pin_0 = 1;
1857 apic_printk(APIC_VERBOSE,KERN_INFO "..TIMER: vector=0x%02X apic1=%d pin1=%d apic2=%d pin2=%d\n",
1858 vector, apic1, pin1, apic2, pin2);
1862 * Ok, does IRQ0 through the IOAPIC work?
1864 unmask_IO_APIC_irq(0);
1865 if (!no_timer_check && timer_irq_works()) {
1866 nmi_watchdog_default();
1867 if (nmi_watchdog == NMI_IO_APIC) {
1868 disable_8259A_irq(0);
1870 enable_8259A_irq(0);
1872 if (disable_timer_pin_1 > 0)
1873 clear_IO_APIC_pin(0, pin1);
1876 clear_IO_APIC_pin(apic1, pin1);
1877 apic_printk(APIC_QUIET,KERN_ERR "..MP-BIOS bug: 8254 timer not "
1878 "connected to IO-APIC\n");
1881 apic_printk(APIC_VERBOSE,KERN_INFO "...trying to set up timer (IRQ0) "
1882 "through the 8259A ... ");
1884 apic_printk(APIC_VERBOSE,"\n..... (found apic %d pin %d) ...",
1887 * legacy devices should be connected to IO APIC #0
1889 setup_ExtINT_IRQ0_pin(apic2, pin2, vector);
1890 if (timer_irq_works()) {
1891 apic_printk(APIC_VERBOSE," works.\n");
1892 nmi_watchdog_default();
1893 if (nmi_watchdog == NMI_IO_APIC) {
1899 * Cleanup, just in case ...
1901 clear_IO_APIC_pin(apic2, pin2);
1903 apic_printk(APIC_VERBOSE," failed.\n");
1905 if (nmi_watchdog == NMI_IO_APIC) {
1906 printk(KERN_WARNING "timer doesn't work through the IO-APIC - disabling NMI Watchdog!\n");
1910 apic_printk(APIC_VERBOSE, KERN_INFO "...trying to set up timer as Virtual Wire IRQ...");
1912 disable_8259A_irq(0);
1913 irq_desc[0].chip = &lapic_irq_type;
1914 apic_write(APIC_LVT0, APIC_DM_FIXED | vector); /* Fixed mode */
1915 enable_8259A_irq(0);
1917 if (timer_irq_works()) {
1918 apic_printk(APIC_VERBOSE," works.\n");
1921 apic_write(APIC_LVT0, APIC_LVT_MASKED | APIC_DM_FIXED | vector);
1922 apic_printk(APIC_VERBOSE," failed.\n");
1924 apic_printk(APIC_VERBOSE, KERN_INFO "...trying to set up timer as ExtINT IRQ...");
1928 apic_write(APIC_LVT0, APIC_DM_EXTINT);
1930 unlock_ExtINT_logic();
1932 if (timer_irq_works()) {
1933 apic_printk(APIC_VERBOSE," works.\n");
1936 apic_printk(APIC_VERBOSE," failed :(.\n");
1937 panic("IO-APIC + timer doesn't work! Try using the 'noapic' kernel parameter\n");
1940 static int __init notimercheck(char *s)
1945 __setup("no_timer_check", notimercheck);
1949 * IRQ's that are handled by the PIC in the MPS IOAPIC case.
1950 * - IRQ2 is the cascade IRQ, and cannot be a io-apic IRQ.
1951 * Linux doesn't really care, as it's not actually used
1952 * for any interrupt handling anyway.
1954 #define PIC_IRQS (1<<2)
1956 void __init setup_IO_APIC(void)
1961 io_apic_irqs = ~0; /* all IRQs go through IOAPIC */
1963 io_apic_irqs = ~PIC_IRQS;
1965 apic_printk(APIC_VERBOSE, "ENABLING IO-APIC IRQs\n");
1968 * Set up the IO-APIC IRQ routing table.
1971 setup_ioapic_ids_from_mpc();
1973 setup_IO_APIC_irqs();
1974 init_IO_APIC_traps();
1980 struct sysfs_ioapic_data {
1981 struct sys_device dev;
1982 struct IO_APIC_route_entry entry[0];
1984 static struct sysfs_ioapic_data * mp_ioapic_data[MAX_IO_APICS];
1986 static int ioapic_suspend(struct sys_device *dev, pm_message_t state)
1988 struct IO_APIC_route_entry *entry;
1989 struct sysfs_ioapic_data *data;
1990 unsigned long flags;
1993 data = container_of(dev, struct sysfs_ioapic_data, dev);
1994 entry = data->entry;
1995 spin_lock_irqsave(&ioapic_lock, flags);
1996 for (i = 0; i < nr_ioapic_registers[dev->id]; i ++, entry ++ ) {
1997 *(((int *)entry) + 1) = io_apic_read(dev->id, 0x11 + 2 * i);
1998 *(((int *)entry) + 0) = io_apic_read(dev->id, 0x10 + 2 * i);
2000 spin_unlock_irqrestore(&ioapic_lock, flags);
2005 static int ioapic_resume(struct sys_device *dev)
2007 struct IO_APIC_route_entry *entry;
2008 struct sysfs_ioapic_data *data;
2009 unsigned long flags;
2010 union IO_APIC_reg_00 reg_00;
2013 data = container_of(dev, struct sysfs_ioapic_data, dev);
2014 entry = data->entry;
2016 spin_lock_irqsave(&ioapic_lock, flags);
2017 reg_00.raw = io_apic_read(dev->id, 0);
2018 if (reg_00.bits.ID != mp_ioapics[dev->id].mpc_apicid) {
2019 reg_00.bits.ID = mp_ioapics[dev->id].mpc_apicid;
2020 io_apic_write(dev->id, 0, reg_00.raw);
2022 for (i = 0; i < nr_ioapic_registers[dev->id]; i ++, entry ++ ) {
2023 io_apic_write(dev->id, 0x11+2*i, *(((int *)entry)+1));
2024 io_apic_write(dev->id, 0x10+2*i, *(((int *)entry)+0));
2026 spin_unlock_irqrestore(&ioapic_lock, flags);
2031 static struct sysdev_class ioapic_sysdev_class = {
2032 set_kset_name("ioapic"),
2033 .suspend = ioapic_suspend,
2034 .resume = ioapic_resume,
2037 static int __init ioapic_init_sysfs(void)
2039 struct sys_device * dev;
2040 int i, size, error = 0;
2042 error = sysdev_class_register(&ioapic_sysdev_class);
2046 for (i = 0; i < nr_ioapics; i++ ) {
2047 size = sizeof(struct sys_device) + nr_ioapic_registers[i]
2048 * sizeof(struct IO_APIC_route_entry);
2049 mp_ioapic_data[i] = kmalloc(size, GFP_KERNEL);
2050 if (!mp_ioapic_data[i]) {
2051 printk(KERN_ERR "Can't suspend/resume IOAPIC %d\n", i);
2054 memset(mp_ioapic_data[i], 0, size);
2055 dev = &mp_ioapic_data[i]->dev;
2057 dev->cls = &ioapic_sysdev_class;
2058 error = sysdev_register(dev);
2060 kfree(mp_ioapic_data[i]);
2061 mp_ioapic_data[i] = NULL;
2062 printk(KERN_ERR "Can't suspend/resume IOAPIC %d\n", i);
2070 device_initcall(ioapic_init_sysfs);
2072 /* --------------------------------------------------------------------------
2073 ACPI-based IOAPIC Configuration
2074 -------------------------------------------------------------------------- */
2078 #define IO_APIC_MAX_ID 0xFE
2080 int __init io_apic_get_version (int ioapic)
2082 union IO_APIC_reg_01 reg_01;
2083 unsigned long flags;
2085 spin_lock_irqsave(&ioapic_lock, flags);
2086 reg_01.raw = io_apic_read(ioapic, 1);
2087 spin_unlock_irqrestore(&ioapic_lock, flags);
2089 return reg_01.bits.version;
2093 int __init io_apic_get_redir_entries (int ioapic)
2095 union IO_APIC_reg_01 reg_01;
2096 unsigned long flags;
2098 spin_lock_irqsave(&ioapic_lock, flags);
2099 reg_01.raw = io_apic_read(ioapic, 1);
2100 spin_unlock_irqrestore(&ioapic_lock, flags);
2102 return reg_01.bits.entries;
2106 int io_apic_set_pci_routing (int ioapic, int pin, int irq, int triggering, int polarity)
2108 struct IO_APIC_route_entry entry;
2109 unsigned long flags;
2111 if (!IO_APIC_IRQ(irq)) {
2112 apic_printk(APIC_QUIET,KERN_ERR "IOAPIC[%d]: Invalid reference to IRQ 0\n",
2118 * Generate a PCI IRQ routing entry and program the IOAPIC accordingly.
2119 * Note that we mask (disable) IRQs now -- these get enabled when the
2120 * corresponding device driver registers for this IRQ.
2123 memset(&entry,0,sizeof(entry));
2125 entry.delivery_mode = INT_DELIVERY_MODE;
2126 entry.dest_mode = INT_DEST_MODE;
2127 entry.dest.logical.logical_dest = cpu_mask_to_apicid(TARGET_CPUS);
2128 entry.trigger = triggering;
2129 entry.polarity = polarity;
2130 entry.mask = 1; /* Disabled (masked) */
2132 irq = gsi_irq_sharing(irq);
2134 * IRQs < 16 are already in the irq_2_pin[] map
2137 add_pin_to_irq(irq, ioapic, pin);
2139 entry.vector = assign_irq_vector(irq);
2141 apic_printk(APIC_VERBOSE,KERN_DEBUG "IOAPIC[%d]: Set PCI routing entry (%d-%d -> 0x%x -> "
2142 "IRQ %d Mode:%i Active:%i)\n", ioapic,
2143 mp_ioapics[ioapic].mpc_apicid, pin, entry.vector, irq,
2144 triggering, polarity);
2146 ioapic_register_intr(irq, entry.vector, triggering);
2148 if (!ioapic && (irq < 16))
2149 disable_8259A_irq(irq);
2151 spin_lock_irqsave(&ioapic_lock, flags);
2152 io_apic_write(ioapic, 0x11+2*pin, *(((int *)&entry)+1));
2153 io_apic_write(ioapic, 0x10+2*pin, *(((int *)&entry)+0));
2154 set_native_irq_info(use_pci_vector() ? entry.vector : irq, TARGET_CPUS);
2155 spin_unlock_irqrestore(&ioapic_lock, flags);
2160 #endif /* CONFIG_ACPI */
2164 * This function currently is only a helper for the i386 smp boot process where
2165 * we need to reprogram the ioredtbls to cater for the cpus which have come online
2166 * so mask in all cases should simply be TARGET_CPUS
2169 void __init setup_ioapic_dest(void)
2171 int pin, ioapic, irq, irq_entry;
2173 if (skip_ioapic_setup == 1)
2176 for (ioapic = 0; ioapic < nr_ioapics; ioapic++) {
2177 for (pin = 0; pin < nr_ioapic_registers[ioapic]; pin++) {
2178 irq_entry = find_irq_entry(ioapic, pin, mp_INT);
2179 if (irq_entry == -1)
2181 irq = pin_2_irq(irq_entry, ioapic, pin);
2182 set_ioapic_affinity_irq(irq, TARGET_CPUS);