2 * Intel IO-APIC support for multi-Pentium hosts.
4 * Copyright (C) 1997, 1998, 1999, 2000 Ingo Molnar, Hajnalka Szabo
6 * Many thanks to Stig Venaas for trying out countless experimental
7 * patches and reporting/debugging problems patiently!
9 * (c) 1999, Multiple IO-APIC support, developed by
10 * Ken-ichi Yaku <yaku@css1.kbnes.nec.co.jp> and
11 * Hidemi Kishimoto <kisimoto@css1.kbnes.nec.co.jp>,
12 * further tested and cleaned up by Zach Brown <zab@redhat.com>
13 * and Ingo Molnar <mingo@redhat.com>
16 * Maciej W. Rozycki : Bits for genuine 82489DX APICs;
17 * thanks to Eric Gilmore
19 * for testing these extensively
20 * Paul Diefenbaugh : Added full ACPI support
24 #include <linux/interrupt.h>
25 #include <linux/init.h>
26 #include <linux/delay.h>
27 #include <linux/sched.h>
28 #include <linux/smp_lock.h>
29 #include <linux/pci.h>
30 #include <linux/mc146818rtc.h>
31 #include <linux/acpi.h>
32 #include <linux/sysdev.h>
33 #include <linux/msi.h>
34 #include <linux/htirq.h>
36 #include <acpi/acpi_bus.h>
42 #include <asm/proto.h>
43 #include <asm/mach_apic.h>
47 #include <asm/msidef.h>
48 #include <asm/hypertransport.h>
50 static int assign_irq_vector(int irq, cpumask_t mask);
52 #define __apicdebuginit __init
54 int sis_apic_bug; /* not actually supported, dummy for compile */
56 static int no_timer_check;
58 static int disable_timer_pin_1 __initdata;
60 int timer_over_8254 __initdata = 0;
62 /* Where if anywhere is the i8259 connect in external int mode */
63 static struct { int pin, apic; } ioapic_i8259 = { -1, -1 };
65 static DEFINE_SPINLOCK(ioapic_lock);
66 static DEFINE_SPINLOCK(vector_lock);
69 * # of IRQ routing registers
71 int nr_ioapic_registers[MAX_IO_APICS];
74 * Rough estimation of how many shared IRQs there are, can
77 #define MAX_PLUS_SHARED_IRQS NR_IRQ_VECTORS
78 #define PIN_MAP_SIZE (MAX_PLUS_SHARED_IRQS + NR_IRQS)
81 * This is performance-critical, we want to do it O(1)
83 * the indexing order of this array favors 1:1 mappings
84 * between pins and IRQs.
87 static struct irq_pin_list {
88 short apic, pin, next;
89 } irq_2_pin[PIN_MAP_SIZE];
91 #define __DO_ACTION(R, ACTION, FINAL) \
95 struct irq_pin_list *entry = irq_2_pin + irq; \
97 BUG_ON(irq >= NR_IRQS); \
103 reg = io_apic_read(entry->apic, 0x10 + R + pin*2); \
105 io_apic_modify(entry->apic, reg); \
108 entry = irq_2_pin + entry->next; \
114 struct { u32 w1, w2; };
115 struct IO_APIC_route_entry entry;
118 static struct IO_APIC_route_entry ioapic_read_entry(int apic, int pin)
120 union entry_union eu;
122 spin_lock_irqsave(&ioapic_lock, flags);
123 eu.w1 = io_apic_read(apic, 0x10 + 2 * pin);
124 eu.w2 = io_apic_read(apic, 0x11 + 2 * pin);
125 spin_unlock_irqrestore(&ioapic_lock, flags);
129 static void ioapic_write_entry(int apic, int pin, struct IO_APIC_route_entry e)
132 union entry_union eu;
134 spin_lock_irqsave(&ioapic_lock, flags);
135 io_apic_write(apic, 0x10 + 2*pin, eu.w1);
136 io_apic_write(apic, 0x11 + 2*pin, eu.w2);
137 spin_unlock_irqrestore(&ioapic_lock, flags);
141 static void __target_IO_APIC_irq(unsigned int irq, unsigned int dest, u8 vector)
144 struct irq_pin_list *entry = irq_2_pin + irq;
146 BUG_ON(irq >= NR_IRQS);
153 io_apic_write(apic, 0x11 + pin*2, dest);
154 reg = io_apic_read(apic, 0x10 + pin*2);
157 io_apic_modify(apic, reg);
160 entry = irq_2_pin + entry->next;
164 static void set_ioapic_affinity_irq(unsigned int irq, cpumask_t mask)
171 cpus_and(tmp, mask, cpu_online_map);
175 cpus_and(mask, tmp, CPU_MASK_ALL);
177 vector = assign_irq_vector(irq, mask);
182 cpu_set(vector >> 8, tmp);
183 dest = cpu_mask_to_apicid(tmp);
186 * Only the high 8 bits are valid.
188 dest = SET_APIC_LOGICAL_ID(dest);
190 spin_lock_irqsave(&ioapic_lock, flags);
191 __target_IO_APIC_irq(irq, dest, vector & 0xff);
192 set_native_irq_info(irq, mask);
193 spin_unlock_irqrestore(&ioapic_lock, flags);
198 * The common case is 1:1 IRQ<->pin mappings. Sometimes there are
199 * shared ISA-space IRQs, so we have to support them. We are super
200 * fast in the common case, and fast for shared ISA-space IRQs.
202 static void add_pin_to_irq(unsigned int irq, int apic, int pin)
204 static int first_free_entry = NR_IRQS;
205 struct irq_pin_list *entry = irq_2_pin + irq;
207 BUG_ON(irq >= NR_IRQS);
209 entry = irq_2_pin + entry->next;
211 if (entry->pin != -1) {
212 entry->next = first_free_entry;
213 entry = irq_2_pin + entry->next;
214 if (++first_free_entry >= PIN_MAP_SIZE)
215 panic("io_apic.c: ran out of irq_2_pin entries!");
222 #define DO_ACTION(name,R,ACTION, FINAL) \
224 static void name##_IO_APIC_irq (unsigned int irq) \
225 __DO_ACTION(R, ACTION, FINAL)
227 DO_ACTION( __mask, 0, |= 0x00010000, io_apic_sync(entry->apic) )
229 DO_ACTION( __unmask, 0, &= 0xfffeffff, )
232 static void mask_IO_APIC_irq (unsigned int irq)
236 spin_lock_irqsave(&ioapic_lock, flags);
237 __mask_IO_APIC_irq(irq);
238 spin_unlock_irqrestore(&ioapic_lock, flags);
241 static void unmask_IO_APIC_irq (unsigned int irq)
245 spin_lock_irqsave(&ioapic_lock, flags);
246 __unmask_IO_APIC_irq(irq);
247 spin_unlock_irqrestore(&ioapic_lock, flags);
250 static void clear_IO_APIC_pin(unsigned int apic, unsigned int pin)
252 struct IO_APIC_route_entry entry;
254 /* Check delivery_mode to be sure we're not clearing an SMI pin */
255 entry = ioapic_read_entry(apic, pin);
256 if (entry.delivery_mode == dest_SMI)
259 * Disable it in the IO-APIC irq-routing table:
261 memset(&entry, 0, sizeof(entry));
263 ioapic_write_entry(apic, pin, entry);
266 static void clear_IO_APIC (void)
270 for (apic = 0; apic < nr_ioapics; apic++)
271 for (pin = 0; pin < nr_ioapic_registers[apic]; pin++)
272 clear_IO_APIC_pin(apic, pin);
275 int skip_ioapic_setup;
278 /* dummy parsing: see setup.c */
280 static int __init disable_ioapic_setup(char *str)
282 skip_ioapic_setup = 1;
285 early_param("noapic", disable_ioapic_setup);
287 /* Actually the next is obsolete, but keep it for paranoid reasons -AK */
288 static int __init disable_timer_pin_setup(char *arg)
290 disable_timer_pin_1 = 1;
293 __setup("disable_timer_pin_1", disable_timer_pin_setup);
295 static int __init setup_disable_8254_timer(char *s)
297 timer_over_8254 = -1;
300 static int __init setup_enable_8254_timer(char *s)
306 __setup("disable_8254_timer", setup_disable_8254_timer);
307 __setup("enable_8254_timer", setup_enable_8254_timer);
311 * Find the IRQ entry number of a certain pin.
313 static int find_irq_entry(int apic, int pin, int type)
317 for (i = 0; i < mp_irq_entries; i++)
318 if (mp_irqs[i].mpc_irqtype == type &&
319 (mp_irqs[i].mpc_dstapic == mp_ioapics[apic].mpc_apicid ||
320 mp_irqs[i].mpc_dstapic == MP_APIC_ALL) &&
321 mp_irqs[i].mpc_dstirq == pin)
328 * Find the pin to which IRQ[irq] (ISA) is connected
330 static int __init find_isa_irq_pin(int irq, int type)
334 for (i = 0; i < mp_irq_entries; i++) {
335 int lbus = mp_irqs[i].mpc_srcbus;
337 if (test_bit(lbus, mp_bus_not_pci) &&
338 (mp_irqs[i].mpc_irqtype == type) &&
339 (mp_irqs[i].mpc_srcbusirq == irq))
341 return mp_irqs[i].mpc_dstirq;
346 static int __init find_isa_irq_apic(int irq, int type)
350 for (i = 0; i < mp_irq_entries; i++) {
351 int lbus = mp_irqs[i].mpc_srcbus;
353 if (test_bit(lbus, mp_bus_not_pci) &&
354 (mp_irqs[i].mpc_irqtype == type) &&
355 (mp_irqs[i].mpc_srcbusirq == irq))
358 if (i < mp_irq_entries) {
360 for(apic = 0; apic < nr_ioapics; apic++) {
361 if (mp_ioapics[apic].mpc_apicid == mp_irqs[i].mpc_dstapic)
370 * Find a specific PCI IRQ entry.
371 * Not an __init, possibly needed by modules
373 static int pin_2_irq(int idx, int apic, int pin);
375 int IO_APIC_get_PCI_irq_vector(int bus, int slot, int pin)
377 int apic, i, best_guess = -1;
379 apic_printk(APIC_DEBUG, "querying PCI -> IRQ mapping bus:%d, slot:%d, pin:%d.\n",
381 if (mp_bus_id_to_pci_bus[bus] == -1) {
382 apic_printk(APIC_VERBOSE, "PCI BIOS passed nonexistent PCI bus %d!\n", bus);
385 for (i = 0; i < mp_irq_entries; i++) {
386 int lbus = mp_irqs[i].mpc_srcbus;
388 for (apic = 0; apic < nr_ioapics; apic++)
389 if (mp_ioapics[apic].mpc_apicid == mp_irqs[i].mpc_dstapic ||
390 mp_irqs[i].mpc_dstapic == MP_APIC_ALL)
393 if (!test_bit(lbus, mp_bus_not_pci) &&
394 !mp_irqs[i].mpc_irqtype &&
396 (slot == ((mp_irqs[i].mpc_srcbusirq >> 2) & 0x1f))) {
397 int irq = pin_2_irq(i,apic,mp_irqs[i].mpc_dstirq);
399 if (!(apic || IO_APIC_IRQ(irq)))
402 if (pin == (mp_irqs[i].mpc_srcbusirq & 3))
405 * Use the first all-but-pin matching entry as a
406 * best-guess fuzzy result for broken mptables.
412 BUG_ON(best_guess >= NR_IRQS);
416 /* ISA interrupts are always polarity zero edge triggered,
417 * when listed as conforming in the MP table. */
419 #define default_ISA_trigger(idx) (0)
420 #define default_ISA_polarity(idx) (0)
422 /* PCI interrupts are always polarity one level triggered,
423 * when listed as conforming in the MP table. */
425 #define default_PCI_trigger(idx) (1)
426 #define default_PCI_polarity(idx) (1)
428 static int __init MPBIOS_polarity(int idx)
430 int bus = mp_irqs[idx].mpc_srcbus;
434 * Determine IRQ line polarity (high active or low active):
436 switch (mp_irqs[idx].mpc_irqflag & 3)
438 case 0: /* conforms, ie. bus-type dependent polarity */
439 if (test_bit(bus, mp_bus_not_pci))
440 polarity = default_ISA_polarity(idx);
442 polarity = default_PCI_polarity(idx);
444 case 1: /* high active */
449 case 2: /* reserved */
451 printk(KERN_WARNING "broken BIOS!!\n");
455 case 3: /* low active */
460 default: /* invalid */
462 printk(KERN_WARNING "broken BIOS!!\n");
470 static int MPBIOS_trigger(int idx)
472 int bus = mp_irqs[idx].mpc_srcbus;
476 * Determine IRQ trigger mode (edge or level sensitive):
478 switch ((mp_irqs[idx].mpc_irqflag>>2) & 3)
480 case 0: /* conforms, ie. bus-type dependent */
481 if (test_bit(bus, mp_bus_not_pci))
482 trigger = default_ISA_trigger(idx);
484 trigger = default_PCI_trigger(idx);
491 case 2: /* reserved */
493 printk(KERN_WARNING "broken BIOS!!\n");
502 default: /* invalid */
504 printk(KERN_WARNING "broken BIOS!!\n");
512 static inline int irq_polarity(int idx)
514 return MPBIOS_polarity(idx);
517 static inline int irq_trigger(int idx)
519 return MPBIOS_trigger(idx);
522 static int pin_2_irq(int idx, int apic, int pin)
525 int bus = mp_irqs[idx].mpc_srcbus;
528 * Debugging check, we are in big trouble if this message pops up!
530 if (mp_irqs[idx].mpc_dstirq != pin)
531 printk(KERN_ERR "broken BIOS or MPTABLE parser, ayiee!!\n");
533 if (test_bit(bus, mp_bus_not_pci)) {
534 irq = mp_irqs[idx].mpc_srcbusirq;
537 * PCI IRQs are mapped in order
541 irq += nr_ioapic_registers[i++];
544 BUG_ON(irq >= NR_IRQS);
548 static inline int IO_APIC_irq_trigger(int irq)
552 for (apic = 0; apic < nr_ioapics; apic++) {
553 for (pin = 0; pin < nr_ioapic_registers[apic]; pin++) {
554 idx = find_irq_entry(apic,pin,mp_INT);
555 if ((idx != -1) && (irq == pin_2_irq(idx,apic,pin)))
556 return irq_trigger(idx);
560 * nonexistent IRQs are edge default
565 /* irq_vectors is indexed by the sum of all RTEs in all I/O APICs. */
566 unsigned int irq_vector[NR_IRQ_VECTORS] __read_mostly = { FIRST_EXTERNAL_VECTOR, 0 };
568 static int __assign_irq_vector(int irq, cpumask_t mask)
571 * NOTE! The local APIC isn't very good at handling
572 * multiple interrupts at the same interrupt level.
573 * As the interrupt level is determined by taking the
574 * vector number and shifting that right by 4, we
575 * want to spread these out a bit so that they don't
576 * all fall in the same interrupt level.
578 * Also, we've got to be careful not to trash gate
579 * 0x80, because int 0x80 is hm, kind of importantish. ;)
584 } pos[NR_CPUS] = { [ 0 ... NR_CPUS - 1] = {FIRST_DEVICE_VECTOR, 0} };
588 BUG_ON((unsigned)irq >= NR_IRQ_VECTORS);
590 if (IO_APIC_VECTOR(irq) > 0)
591 old_vector = IO_APIC_VECTOR(irq);
592 if ((old_vector > 0) && cpu_isset(old_vector >> 8, mask)) {
596 for_each_cpu_mask(cpu, mask) {
598 vector = pos[cpu].vector;
599 offset = pos[cpu].offset;
602 if (vector >= FIRST_SYSTEM_VECTOR) {
603 /* If we run out of vectors on large boxen, must share them. */
604 offset = (offset + 1) % 8;
605 vector = FIRST_DEVICE_VECTOR + offset;
607 if (unlikely(pos[cpu].vector == vector))
609 if (vector == IA32_SYSCALL_VECTOR)
611 if (per_cpu(vector_irq, cpu)[vector] != -1)
614 pos[cpu].vector = vector;
615 pos[cpu].offset = offset;
616 if (old_vector >= 0) {
617 int old_cpu = old_vector >> 8;
619 per_cpu(vector_irq, old_cpu)[old_vector] = -1;
621 per_cpu(vector_irq, cpu)[vector] = irq;
623 IO_APIC_VECTOR(irq) = vector;
629 static int assign_irq_vector(int irq, cpumask_t mask)
634 spin_lock_irqsave(&vector_lock, flags);
635 vector = __assign_irq_vector(irq, mask);
636 spin_unlock_irqrestore(&vector_lock, flags);
640 extern void (*interrupt[NR_IRQS])(void);
642 static struct irq_chip ioapic_chip;
644 #define IOAPIC_AUTO -1
645 #define IOAPIC_EDGE 0
646 #define IOAPIC_LEVEL 1
648 static void ioapic_register_intr(int irq, int vector, unsigned long trigger)
650 if ((trigger == IOAPIC_AUTO && IO_APIC_irq_trigger(irq)) ||
651 trigger == IOAPIC_LEVEL)
652 set_irq_chip_and_handler(irq, &ioapic_chip,
655 set_irq_chip_and_handler(irq, &ioapic_chip,
659 static void __init setup_IO_APIC_irqs(void)
661 struct IO_APIC_route_entry entry;
662 int apic, pin, idx, irq, first_notcon = 1, vector;
665 apic_printk(APIC_VERBOSE, KERN_DEBUG "init IO_APIC IRQs\n");
667 for (apic = 0; apic < nr_ioapics; apic++) {
668 for (pin = 0; pin < nr_ioapic_registers[apic]; pin++) {
671 * add it to the IO-APIC irq-routing table:
673 memset(&entry,0,sizeof(entry));
675 entry.delivery_mode = INT_DELIVERY_MODE;
676 entry.dest_mode = INT_DEST_MODE;
677 entry.mask = 0; /* enable IRQ */
678 entry.dest.logical.logical_dest = cpu_mask_to_apicid(TARGET_CPUS);
680 idx = find_irq_entry(apic,pin,mp_INT);
683 apic_printk(APIC_VERBOSE, KERN_DEBUG " IO-APIC (apicid-pin) %d-%d", mp_ioapics[apic].mpc_apicid, pin);
686 apic_printk(APIC_VERBOSE, ", %d-%d", mp_ioapics[apic].mpc_apicid, pin);
690 entry.trigger = irq_trigger(idx);
691 entry.polarity = irq_polarity(idx);
693 if (irq_trigger(idx)) {
696 entry.dest.logical.logical_dest = cpu_mask_to_apicid(TARGET_CPUS);
699 irq = pin_2_irq(idx, apic, pin);
700 add_pin_to_irq(irq, apic, pin);
702 if (!apic && !IO_APIC_IRQ(irq))
705 if (IO_APIC_IRQ(irq)) {
707 vector = assign_irq_vector(irq, TARGET_CPUS);
712 cpu_set(vector >> 8, mask);
713 entry.dest.logical.logical_dest = cpu_mask_to_apicid(mask);
714 entry.vector = vector & 0xff;
716 ioapic_register_intr(irq, vector, IOAPIC_AUTO);
717 if (!apic && (irq < 16))
718 disable_8259A_irq(irq);
720 ioapic_write_entry(apic, pin, entry);
722 spin_lock_irqsave(&ioapic_lock, flags);
723 set_native_irq_info(irq, TARGET_CPUS);
724 spin_unlock_irqrestore(&ioapic_lock, flags);
729 apic_printk(APIC_VERBOSE," not connected.\n");
733 * Set up the 8259A-master output pin as broadcast to all
736 static void __init setup_ExtINT_IRQ0_pin(unsigned int apic, unsigned int pin, int vector)
738 struct IO_APIC_route_entry entry;
741 memset(&entry,0,sizeof(entry));
743 disable_8259A_irq(0);
746 apic_write(APIC_LVT0, APIC_LVT_MASKED | APIC_DM_EXTINT);
749 * We use logical delivery to get the timer IRQ
752 entry.dest_mode = INT_DEST_MODE;
753 entry.mask = 0; /* unmask IRQ now */
754 entry.dest.logical.logical_dest = cpu_mask_to_apicid(TARGET_CPUS);
755 entry.delivery_mode = INT_DELIVERY_MODE;
758 entry.vector = vector;
761 * The timer IRQ doesn't have to know that behind the
762 * scene we have a 8259A-master in AEOI mode ...
764 set_irq_chip_and_handler(0, &ioapic_chip, handle_edge_irq);
767 * Add it to the IO-APIC irq-routing table:
769 spin_lock_irqsave(&ioapic_lock, flags);
770 io_apic_write(apic, 0x11+2*pin, *(((int *)&entry)+1));
771 io_apic_write(apic, 0x10+2*pin, *(((int *)&entry)+0));
772 spin_unlock_irqrestore(&ioapic_lock, flags);
777 void __init UNEXPECTED_IO_APIC(void)
781 void __apicdebuginit print_IO_APIC(void)
784 union IO_APIC_reg_00 reg_00;
785 union IO_APIC_reg_01 reg_01;
786 union IO_APIC_reg_02 reg_02;
789 if (apic_verbosity == APIC_QUIET)
792 printk(KERN_DEBUG "number of MP IRQ sources: %d.\n", mp_irq_entries);
793 for (i = 0; i < nr_ioapics; i++)
794 printk(KERN_DEBUG "number of IO-APIC #%d registers: %d.\n",
795 mp_ioapics[i].mpc_apicid, nr_ioapic_registers[i]);
798 * We are a bit conservative about what we expect. We have to
799 * know about every hardware change ASAP.
801 printk(KERN_INFO "testing the IO APIC.......................\n");
803 for (apic = 0; apic < nr_ioapics; apic++) {
805 spin_lock_irqsave(&ioapic_lock, flags);
806 reg_00.raw = io_apic_read(apic, 0);
807 reg_01.raw = io_apic_read(apic, 1);
808 if (reg_01.bits.version >= 0x10)
809 reg_02.raw = io_apic_read(apic, 2);
810 spin_unlock_irqrestore(&ioapic_lock, flags);
813 printk(KERN_DEBUG "IO APIC #%d......\n", mp_ioapics[apic].mpc_apicid);
814 printk(KERN_DEBUG ".... register #00: %08X\n", reg_00.raw);
815 printk(KERN_DEBUG "....... : physical APIC id: %02X\n", reg_00.bits.ID);
816 if (reg_00.bits.__reserved_1 || reg_00.bits.__reserved_2)
817 UNEXPECTED_IO_APIC();
819 printk(KERN_DEBUG ".... register #01: %08X\n", *(int *)®_01);
820 printk(KERN_DEBUG "....... : max redirection entries: %04X\n", reg_01.bits.entries);
821 if ( (reg_01.bits.entries != 0x0f) && /* older (Neptune) boards */
822 (reg_01.bits.entries != 0x17) && /* typical ISA+PCI boards */
823 (reg_01.bits.entries != 0x1b) && /* Compaq Proliant boards */
824 (reg_01.bits.entries != 0x1f) && /* dual Xeon boards */
825 (reg_01.bits.entries != 0x22) && /* bigger Xeon boards */
826 (reg_01.bits.entries != 0x2E) &&
827 (reg_01.bits.entries != 0x3F) &&
828 (reg_01.bits.entries != 0x03)
830 UNEXPECTED_IO_APIC();
832 printk(KERN_DEBUG "....... : PRQ implemented: %X\n", reg_01.bits.PRQ);
833 printk(KERN_DEBUG "....... : IO APIC version: %04X\n", reg_01.bits.version);
834 if ( (reg_01.bits.version != 0x01) && /* 82489DX IO-APICs */
835 (reg_01.bits.version != 0x02) && /* 82801BA IO-APICs (ICH2) */
836 (reg_01.bits.version != 0x10) && /* oldest IO-APICs */
837 (reg_01.bits.version != 0x11) && /* Pentium/Pro IO-APICs */
838 (reg_01.bits.version != 0x13) && /* Xeon IO-APICs */
839 (reg_01.bits.version != 0x20) /* Intel P64H (82806 AA) */
841 UNEXPECTED_IO_APIC();
842 if (reg_01.bits.__reserved_1 || reg_01.bits.__reserved_2)
843 UNEXPECTED_IO_APIC();
845 if (reg_01.bits.version >= 0x10) {
846 printk(KERN_DEBUG ".... register #02: %08X\n", reg_02.raw);
847 printk(KERN_DEBUG "....... : arbitration: %02X\n", reg_02.bits.arbitration);
848 if (reg_02.bits.__reserved_1 || reg_02.bits.__reserved_2)
849 UNEXPECTED_IO_APIC();
852 printk(KERN_DEBUG ".... IRQ redirection table:\n");
854 printk(KERN_DEBUG " NR Log Phy Mask Trig IRR Pol"
855 " Stat Dest Deli Vect: \n");
857 for (i = 0; i <= reg_01.bits.entries; i++) {
858 struct IO_APIC_route_entry entry;
860 entry = ioapic_read_entry(apic, i);
862 printk(KERN_DEBUG " %02x %03X %02X ",
864 entry.dest.logical.logical_dest,
865 entry.dest.physical.physical_dest
868 printk("%1d %1d %1d %1d %1d %1d %1d %02X\n",
873 entry.delivery_status,
880 printk(KERN_DEBUG "IRQ to pin mappings:\n");
881 for (i = 0; i < NR_IRQS; i++) {
882 struct irq_pin_list *entry = irq_2_pin + i;
885 printk(KERN_DEBUG "IRQ%d ", i);
887 printk("-> %d:%d", entry->apic, entry->pin);
890 entry = irq_2_pin + entry->next;
895 printk(KERN_INFO ".................................... done.\n");
902 static __apicdebuginit void print_APIC_bitfield (int base)
907 if (apic_verbosity == APIC_QUIET)
910 printk(KERN_DEBUG "0123456789abcdef0123456789abcdef\n" KERN_DEBUG);
911 for (i = 0; i < 8; i++) {
912 v = apic_read(base + i*0x10);
913 for (j = 0; j < 32; j++) {
923 void __apicdebuginit print_local_APIC(void * dummy)
925 unsigned int v, ver, maxlvt;
927 if (apic_verbosity == APIC_QUIET)
930 printk("\n" KERN_DEBUG "printing local APIC contents on CPU#%d/%d:\n",
931 smp_processor_id(), hard_smp_processor_id());
932 v = apic_read(APIC_ID);
933 printk(KERN_INFO "... APIC ID: %08x (%01x)\n", v, GET_APIC_ID(v));
934 v = apic_read(APIC_LVR);
935 printk(KERN_INFO "... APIC VERSION: %08x\n", v);
936 ver = GET_APIC_VERSION(v);
937 maxlvt = get_maxlvt();
939 v = apic_read(APIC_TASKPRI);
940 printk(KERN_DEBUG "... APIC TASKPRI: %08x (%02x)\n", v, v & APIC_TPRI_MASK);
942 v = apic_read(APIC_ARBPRI);
943 printk(KERN_DEBUG "... APIC ARBPRI: %08x (%02x)\n", v,
944 v & APIC_ARBPRI_MASK);
945 v = apic_read(APIC_PROCPRI);
946 printk(KERN_DEBUG "... APIC PROCPRI: %08x\n", v);
948 v = apic_read(APIC_EOI);
949 printk(KERN_DEBUG "... APIC EOI: %08x\n", v);
950 v = apic_read(APIC_RRR);
951 printk(KERN_DEBUG "... APIC RRR: %08x\n", v);
952 v = apic_read(APIC_LDR);
953 printk(KERN_DEBUG "... APIC LDR: %08x\n", v);
954 v = apic_read(APIC_DFR);
955 printk(KERN_DEBUG "... APIC DFR: %08x\n", v);
956 v = apic_read(APIC_SPIV);
957 printk(KERN_DEBUG "... APIC SPIV: %08x\n", v);
959 printk(KERN_DEBUG "... APIC ISR field:\n");
960 print_APIC_bitfield(APIC_ISR);
961 printk(KERN_DEBUG "... APIC TMR field:\n");
962 print_APIC_bitfield(APIC_TMR);
963 printk(KERN_DEBUG "... APIC IRR field:\n");
964 print_APIC_bitfield(APIC_IRR);
966 v = apic_read(APIC_ESR);
967 printk(KERN_DEBUG "... APIC ESR: %08x\n", v);
969 v = apic_read(APIC_ICR);
970 printk(KERN_DEBUG "... APIC ICR: %08x\n", v);
971 v = apic_read(APIC_ICR2);
972 printk(KERN_DEBUG "... APIC ICR2: %08x\n", v);
974 v = apic_read(APIC_LVTT);
975 printk(KERN_DEBUG "... APIC LVTT: %08x\n", v);
977 if (maxlvt > 3) { /* PC is LVT#4. */
978 v = apic_read(APIC_LVTPC);
979 printk(KERN_DEBUG "... APIC LVTPC: %08x\n", v);
981 v = apic_read(APIC_LVT0);
982 printk(KERN_DEBUG "... APIC LVT0: %08x\n", v);
983 v = apic_read(APIC_LVT1);
984 printk(KERN_DEBUG "... APIC LVT1: %08x\n", v);
986 if (maxlvt > 2) { /* ERR is LVT#3. */
987 v = apic_read(APIC_LVTERR);
988 printk(KERN_DEBUG "... APIC LVTERR: %08x\n", v);
991 v = apic_read(APIC_TMICT);
992 printk(KERN_DEBUG "... APIC TMICT: %08x\n", v);
993 v = apic_read(APIC_TMCCT);
994 printk(KERN_DEBUG "... APIC TMCCT: %08x\n", v);
995 v = apic_read(APIC_TDCR);
996 printk(KERN_DEBUG "... APIC TDCR: %08x\n", v);
1000 void print_all_local_APICs (void)
1002 on_each_cpu(print_local_APIC, NULL, 1, 1);
1005 void __apicdebuginit print_PIC(void)
1008 unsigned long flags;
1010 if (apic_verbosity == APIC_QUIET)
1013 printk(KERN_DEBUG "\nprinting PIC contents\n");
1015 spin_lock_irqsave(&i8259A_lock, flags);
1017 v = inb(0xa1) << 8 | inb(0x21);
1018 printk(KERN_DEBUG "... PIC IMR: %04x\n", v);
1020 v = inb(0xa0) << 8 | inb(0x20);
1021 printk(KERN_DEBUG "... PIC IRR: %04x\n", v);
1025 v = inb(0xa0) << 8 | inb(0x20);
1029 spin_unlock_irqrestore(&i8259A_lock, flags);
1031 printk(KERN_DEBUG "... PIC ISR: %04x\n", v);
1033 v = inb(0x4d1) << 8 | inb(0x4d0);
1034 printk(KERN_DEBUG "... PIC ELCR: %04x\n", v);
1039 static void __init enable_IO_APIC(void)
1041 union IO_APIC_reg_01 reg_01;
1042 int i8259_apic, i8259_pin;
1044 unsigned long flags;
1046 for (i = 0; i < PIN_MAP_SIZE; i++) {
1047 irq_2_pin[i].pin = -1;
1048 irq_2_pin[i].next = 0;
1052 * The number of IO-APIC IRQ registers (== #pins):
1054 for (apic = 0; apic < nr_ioapics; apic++) {
1055 spin_lock_irqsave(&ioapic_lock, flags);
1056 reg_01.raw = io_apic_read(apic, 1);
1057 spin_unlock_irqrestore(&ioapic_lock, flags);
1058 nr_ioapic_registers[apic] = reg_01.bits.entries+1;
1060 for(apic = 0; apic < nr_ioapics; apic++) {
1062 /* See if any of the pins is in ExtINT mode */
1063 for (pin = 0; pin < nr_ioapic_registers[apic]; pin++) {
1064 struct IO_APIC_route_entry entry;
1065 entry = ioapic_read_entry(apic, pin);
1067 /* If the interrupt line is enabled and in ExtInt mode
1068 * I have found the pin where the i8259 is connected.
1070 if ((entry.mask == 0) && (entry.delivery_mode == dest_ExtINT)) {
1071 ioapic_i8259.apic = apic;
1072 ioapic_i8259.pin = pin;
1078 /* Look to see what if the MP table has reported the ExtINT */
1079 i8259_pin = find_isa_irq_pin(0, mp_ExtINT);
1080 i8259_apic = find_isa_irq_apic(0, mp_ExtINT);
1081 /* Trust the MP table if nothing is setup in the hardware */
1082 if ((ioapic_i8259.pin == -1) && (i8259_pin >= 0)) {
1083 printk(KERN_WARNING "ExtINT not setup in hardware but reported by MP table\n");
1084 ioapic_i8259.pin = i8259_pin;
1085 ioapic_i8259.apic = i8259_apic;
1087 /* Complain if the MP table and the hardware disagree */
1088 if (((ioapic_i8259.apic != i8259_apic) || (ioapic_i8259.pin != i8259_pin)) &&
1089 (i8259_pin >= 0) && (ioapic_i8259.pin >= 0))
1091 printk(KERN_WARNING "ExtINT in hardware and MP table differ\n");
1095 * Do not trust the IO-APIC being empty at bootup
1101 * Not an __init, needed by the reboot code
1103 void disable_IO_APIC(void)
1106 * Clear the IO-APIC before rebooting:
1111 * If the i8259 is routed through an IOAPIC
1112 * Put that IOAPIC in virtual wire mode
1113 * so legacy interrupts can be delivered.
1115 if (ioapic_i8259.pin != -1) {
1116 struct IO_APIC_route_entry entry;
1118 memset(&entry, 0, sizeof(entry));
1119 entry.mask = 0; /* Enabled */
1120 entry.trigger = 0; /* Edge */
1122 entry.polarity = 0; /* High */
1123 entry.delivery_status = 0;
1124 entry.dest_mode = 0; /* Physical */
1125 entry.delivery_mode = dest_ExtINT; /* ExtInt */
1127 entry.dest.physical.physical_dest =
1128 GET_APIC_ID(apic_read(APIC_ID));
1131 * Add it to the IO-APIC irq-routing table:
1133 ioapic_write_entry(ioapic_i8259.apic, ioapic_i8259.pin, entry);
1136 disconnect_bsp_APIC(ioapic_i8259.pin != -1);
1140 * There is a nasty bug in some older SMP boards, their mptable lies
1141 * about the timer IRQ. We do the following to work around the situation:
1143 * - timer IRQ defaults to IO-APIC IRQ
1144 * - if this function detects that timer IRQs are defunct, then we fall
1145 * back to ISA timer IRQs
1147 static int __init timer_irq_works(void)
1149 unsigned long t1 = jiffies;
1152 /* Let ten ticks pass... */
1153 mdelay((10 * 1000) / HZ);
1156 * Expect a few ticks at least, to be sure some possible
1157 * glue logic does not lock up after one or two first
1158 * ticks in a non-ExtINT mode. Also the local APIC
1159 * might have cached one ExtINT interrupt. Finally, at
1160 * least one tick may be lost due to delays.
1164 if (jiffies - t1 > 4)
1170 * In the SMP+IOAPIC case it might happen that there are an unspecified
1171 * number of pending IRQ events unhandled. These cases are very rare,
1172 * so we 'resend' these IRQs via IPIs, to the same CPU. It's much
1173 * better to do it this way as thus we do not have to be aware of
1174 * 'pending' interrupts in the IRQ path, except at this point.
1177 * Edge triggered needs to resend any interrupt
1178 * that was delayed but this is now handled in the device
1183 * Starting up a edge-triggered IO-APIC interrupt is
1184 * nasty - we need to make sure that we get the edge.
1185 * If it is already asserted for some reason, we need
1186 * return 1 to indicate that is was pending.
1188 * This is not complete - we should be able to fake
1189 * an edge even if it isn't on the 8259A...
1192 static unsigned int startup_ioapic_irq(unsigned int irq)
1194 int was_pending = 0;
1195 unsigned long flags;
1197 spin_lock_irqsave(&ioapic_lock, flags);
1199 disable_8259A_irq(irq);
1200 if (i8259A_irq_pending(irq))
1203 __unmask_IO_APIC_irq(irq);
1204 spin_unlock_irqrestore(&ioapic_lock, flags);
1209 static int ioapic_retrigger_irq(unsigned int irq)
1214 vector = irq_vector[irq];
1216 cpu_set(vector >> 8, mask);
1218 send_IPI_mask(mask, vector & 0xff);
1224 * Level and edge triggered IO-APIC interrupts need different handling,
1225 * so we use two separate IRQ descriptors. Edge triggered IRQs can be
1226 * handled with the level-triggered descriptor, but that one has slightly
1227 * more overhead. Level-triggered interrupts cannot be handled with the
1228 * edge-triggered handler, without risking IRQ storms and other ugly
1232 static void ack_apic_edge(unsigned int irq)
1234 move_native_irq(irq);
1238 static void ack_apic_level(unsigned int irq)
1240 int do_unmask_irq = 0;
1242 #if defined(CONFIG_GENERIC_PENDING_IRQ) || defined(CONFIG_IRQBALANCE)
1243 /* If we are moving the irq we need to mask it */
1244 if (unlikely(irq_desc[irq].status & IRQ_MOVE_PENDING)) {
1246 mask_IO_APIC_irq(irq);
1251 * We must acknowledge the irq before we move it or the acknowledge will
1252 * not propogate properly.
1256 /* Now we can move and renable the irq */
1257 move_masked_irq(irq);
1258 if (unlikely(do_unmask_irq))
1259 unmask_IO_APIC_irq(irq);
1262 static struct irq_chip ioapic_chip __read_mostly = {
1264 .startup = startup_ioapic_irq,
1265 .mask = mask_IO_APIC_irq,
1266 .unmask = unmask_IO_APIC_irq,
1267 .ack = ack_apic_edge,
1268 .eoi = ack_apic_level,
1270 .set_affinity = set_ioapic_affinity_irq,
1272 .retrigger = ioapic_retrigger_irq,
1275 static inline void init_IO_APIC_traps(void)
1280 * NOTE! The local APIC isn't very good at handling
1281 * multiple interrupts at the same interrupt level.
1282 * As the interrupt level is determined by taking the
1283 * vector number and shifting that right by 4, we
1284 * want to spread these out a bit so that they don't
1285 * all fall in the same interrupt level.
1287 * Also, we've got to be careful not to trash gate
1288 * 0x80, because int 0x80 is hm, kind of importantish. ;)
1290 for (irq = 0; irq < NR_IRQS ; irq++) {
1292 if (IO_APIC_IRQ(tmp) && !IO_APIC_VECTOR(tmp)) {
1294 * Hmm.. We don't have an entry for this,
1295 * so default to an old-fashioned 8259
1296 * interrupt if we can..
1299 make_8259A_irq(irq);
1301 /* Strange. Oh, well.. */
1302 irq_desc[irq].chip = &no_irq_chip;
1307 static void enable_lapic_irq (unsigned int irq)
1311 v = apic_read(APIC_LVT0);
1312 apic_write(APIC_LVT0, v & ~APIC_LVT_MASKED);
1315 static void disable_lapic_irq (unsigned int irq)
1319 v = apic_read(APIC_LVT0);
1320 apic_write(APIC_LVT0, v | APIC_LVT_MASKED);
1323 static void ack_lapic_irq (unsigned int irq)
1328 static void end_lapic_irq (unsigned int i) { /* nothing */ }
1330 static struct hw_interrupt_type lapic_irq_type __read_mostly = {
1331 .typename = "local-APIC-edge",
1332 .startup = NULL, /* startup_irq() not used for IRQ0 */
1333 .shutdown = NULL, /* shutdown_irq() not used for IRQ0 */
1334 .enable = enable_lapic_irq,
1335 .disable = disable_lapic_irq,
1336 .ack = ack_lapic_irq,
1337 .end = end_lapic_irq,
1340 static void setup_nmi (void)
1343 * Dirty trick to enable the NMI watchdog ...
1344 * We put the 8259A master into AEOI mode and
1345 * unmask on all local APICs LVT0 as NMI.
1347 * The idea to use the 8259A in AEOI mode ('8259A Virtual Wire')
1348 * is from Maciej W. Rozycki - so we do not have to EOI from
1349 * the NMI handler or the timer interrupt.
1351 printk(KERN_INFO "activating NMI Watchdog ...");
1353 enable_NMI_through_LVT0(NULL);
1359 * This looks a bit hackish but it's about the only one way of sending
1360 * a few INTA cycles to 8259As and any associated glue logic. ICR does
1361 * not support the ExtINT mode, unfortunately. We need to send these
1362 * cycles as some i82489DX-based boards have glue logic that keeps the
1363 * 8259A interrupt line asserted until INTA. --macro
1365 static inline void unlock_ExtINT_logic(void)
1368 struct IO_APIC_route_entry entry0, entry1;
1369 unsigned char save_control, save_freq_select;
1370 unsigned long flags;
1372 pin = find_isa_irq_pin(8, mp_INT);
1373 apic = find_isa_irq_apic(8, mp_INT);
1377 spin_lock_irqsave(&ioapic_lock, flags);
1378 *(((int *)&entry0) + 1) = io_apic_read(apic, 0x11 + 2 * pin);
1379 *(((int *)&entry0) + 0) = io_apic_read(apic, 0x10 + 2 * pin);
1380 spin_unlock_irqrestore(&ioapic_lock, flags);
1381 clear_IO_APIC_pin(apic, pin);
1383 memset(&entry1, 0, sizeof(entry1));
1385 entry1.dest_mode = 0; /* physical delivery */
1386 entry1.mask = 0; /* unmask IRQ now */
1387 entry1.dest.physical.physical_dest = hard_smp_processor_id();
1388 entry1.delivery_mode = dest_ExtINT;
1389 entry1.polarity = entry0.polarity;
1393 spin_lock_irqsave(&ioapic_lock, flags);
1394 io_apic_write(apic, 0x11 + 2 * pin, *(((int *)&entry1) + 1));
1395 io_apic_write(apic, 0x10 + 2 * pin, *(((int *)&entry1) + 0));
1396 spin_unlock_irqrestore(&ioapic_lock, flags);
1398 save_control = CMOS_READ(RTC_CONTROL);
1399 save_freq_select = CMOS_READ(RTC_FREQ_SELECT);
1400 CMOS_WRITE((save_freq_select & ~RTC_RATE_SELECT) | 0x6,
1402 CMOS_WRITE(save_control | RTC_PIE, RTC_CONTROL);
1407 if ((CMOS_READ(RTC_INTR_FLAGS) & RTC_PF) == RTC_PF)
1411 CMOS_WRITE(save_control, RTC_CONTROL);
1412 CMOS_WRITE(save_freq_select, RTC_FREQ_SELECT);
1413 clear_IO_APIC_pin(apic, pin);
1415 spin_lock_irqsave(&ioapic_lock, flags);
1416 io_apic_write(apic, 0x11 + 2 * pin, *(((int *)&entry0) + 1));
1417 io_apic_write(apic, 0x10 + 2 * pin, *(((int *)&entry0) + 0));
1418 spin_unlock_irqrestore(&ioapic_lock, flags);
1422 * This code may look a bit paranoid, but it's supposed to cooperate with
1423 * a wide range of boards and BIOS bugs. Fortunately only the timer IRQ
1424 * is so screwy. Thanks to Brian Perkins for testing/hacking this beast
1425 * fanatically on his truly buggy board.
1427 * FIXME: really need to revamp this for modern platforms only.
1429 static inline void check_timer(void)
1431 int apic1, pin1, apic2, pin2;
1435 * get/set the timer IRQ vector:
1437 disable_8259A_irq(0);
1438 vector = assign_irq_vector(0, TARGET_CPUS);
1441 * Subtle, code in do_timer_interrupt() expects an AEOI
1442 * mode for the 8259A whenever interrupts are routed
1443 * through I/O APICs. Also IRQ0 has to be enabled in
1444 * the 8259A which implies the virtual wire has to be
1445 * disabled in the local APIC.
1447 apic_write(APIC_LVT0, APIC_LVT_MASKED | APIC_DM_EXTINT);
1449 if (timer_over_8254 > 0)
1450 enable_8259A_irq(0);
1452 pin1 = find_isa_irq_pin(0, mp_INT);
1453 apic1 = find_isa_irq_apic(0, mp_INT);
1454 pin2 = ioapic_i8259.pin;
1455 apic2 = ioapic_i8259.apic;
1457 apic_printk(APIC_VERBOSE,KERN_INFO "..TIMER: vector=0x%02X apic1=%d pin1=%d apic2=%d pin2=%d\n",
1458 vector, apic1, pin1, apic2, pin2);
1462 * Ok, does IRQ0 through the IOAPIC work?
1464 unmask_IO_APIC_irq(0);
1465 if (!no_timer_check && timer_irq_works()) {
1466 nmi_watchdog_default();
1467 if (nmi_watchdog == NMI_IO_APIC) {
1468 disable_8259A_irq(0);
1470 enable_8259A_irq(0);
1472 if (disable_timer_pin_1 > 0)
1473 clear_IO_APIC_pin(0, pin1);
1476 clear_IO_APIC_pin(apic1, pin1);
1477 apic_printk(APIC_QUIET,KERN_ERR "..MP-BIOS bug: 8254 timer not "
1478 "connected to IO-APIC\n");
1481 apic_printk(APIC_VERBOSE,KERN_INFO "...trying to set up timer (IRQ0) "
1482 "through the 8259A ... ");
1484 apic_printk(APIC_VERBOSE,"\n..... (found apic %d pin %d) ...",
1487 * legacy devices should be connected to IO APIC #0
1489 setup_ExtINT_IRQ0_pin(apic2, pin2, vector);
1490 if (timer_irq_works()) {
1491 apic_printk(APIC_VERBOSE," works.\n");
1492 nmi_watchdog_default();
1493 if (nmi_watchdog == NMI_IO_APIC) {
1499 * Cleanup, just in case ...
1501 clear_IO_APIC_pin(apic2, pin2);
1503 apic_printk(APIC_VERBOSE," failed.\n");
1505 if (nmi_watchdog == NMI_IO_APIC) {
1506 printk(KERN_WARNING "timer doesn't work through the IO-APIC - disabling NMI Watchdog!\n");
1510 apic_printk(APIC_VERBOSE, KERN_INFO "...trying to set up timer as Virtual Wire IRQ...");
1512 disable_8259A_irq(0);
1513 irq_desc[0].chip = &lapic_irq_type;
1514 apic_write(APIC_LVT0, APIC_DM_FIXED | vector); /* Fixed mode */
1515 enable_8259A_irq(0);
1517 if (timer_irq_works()) {
1518 apic_printk(APIC_VERBOSE," works.\n");
1521 apic_write(APIC_LVT0, APIC_LVT_MASKED | APIC_DM_FIXED | vector);
1522 apic_printk(APIC_VERBOSE," failed.\n");
1524 apic_printk(APIC_VERBOSE, KERN_INFO "...trying to set up timer as ExtINT IRQ...");
1528 apic_write(APIC_LVT0, APIC_DM_EXTINT);
1530 unlock_ExtINT_logic();
1532 if (timer_irq_works()) {
1533 apic_printk(APIC_VERBOSE," works.\n");
1536 apic_printk(APIC_VERBOSE," failed :(.\n");
1537 panic("IO-APIC + timer doesn't work! Try using the 'noapic' kernel parameter\n");
1540 static int __init notimercheck(char *s)
1545 __setup("no_timer_check", notimercheck);
1549 * IRQ's that are handled by the PIC in the MPS IOAPIC case.
1550 * - IRQ2 is the cascade IRQ, and cannot be a io-apic IRQ.
1551 * Linux doesn't really care, as it's not actually used
1552 * for any interrupt handling anyway.
1554 #define PIC_IRQS (1<<2)
1556 void __init setup_IO_APIC(void)
1561 io_apic_irqs = ~0; /* all IRQs go through IOAPIC */
1563 io_apic_irqs = ~PIC_IRQS;
1565 apic_printk(APIC_VERBOSE, "ENABLING IO-APIC IRQs\n");
1568 setup_IO_APIC_irqs();
1569 init_IO_APIC_traps();
1575 struct sysfs_ioapic_data {
1576 struct sys_device dev;
1577 struct IO_APIC_route_entry entry[0];
1579 static struct sysfs_ioapic_data * mp_ioapic_data[MAX_IO_APICS];
1581 static int ioapic_suspend(struct sys_device *dev, pm_message_t state)
1583 struct IO_APIC_route_entry *entry;
1584 struct sysfs_ioapic_data *data;
1587 data = container_of(dev, struct sysfs_ioapic_data, dev);
1588 entry = data->entry;
1589 for (i = 0; i < nr_ioapic_registers[dev->id]; i ++, entry ++ )
1590 *entry = ioapic_read_entry(dev->id, i);
1595 static int ioapic_resume(struct sys_device *dev)
1597 struct IO_APIC_route_entry *entry;
1598 struct sysfs_ioapic_data *data;
1599 unsigned long flags;
1600 union IO_APIC_reg_00 reg_00;
1603 data = container_of(dev, struct sysfs_ioapic_data, dev);
1604 entry = data->entry;
1606 spin_lock_irqsave(&ioapic_lock, flags);
1607 reg_00.raw = io_apic_read(dev->id, 0);
1608 if (reg_00.bits.ID != mp_ioapics[dev->id].mpc_apicid) {
1609 reg_00.bits.ID = mp_ioapics[dev->id].mpc_apicid;
1610 io_apic_write(dev->id, 0, reg_00.raw);
1612 spin_unlock_irqrestore(&ioapic_lock, flags);
1613 for (i = 0; i < nr_ioapic_registers[dev->id]; i++)
1614 ioapic_write_entry(dev->id, i, entry[i]);
1619 static struct sysdev_class ioapic_sysdev_class = {
1620 set_kset_name("ioapic"),
1621 .suspend = ioapic_suspend,
1622 .resume = ioapic_resume,
1625 static int __init ioapic_init_sysfs(void)
1627 struct sys_device * dev;
1628 int i, size, error = 0;
1630 error = sysdev_class_register(&ioapic_sysdev_class);
1634 for (i = 0; i < nr_ioapics; i++ ) {
1635 size = sizeof(struct sys_device) + nr_ioapic_registers[i]
1636 * sizeof(struct IO_APIC_route_entry);
1637 mp_ioapic_data[i] = kmalloc(size, GFP_KERNEL);
1638 if (!mp_ioapic_data[i]) {
1639 printk(KERN_ERR "Can't suspend/resume IOAPIC %d\n", i);
1642 memset(mp_ioapic_data[i], 0, size);
1643 dev = &mp_ioapic_data[i]->dev;
1645 dev->cls = &ioapic_sysdev_class;
1646 error = sysdev_register(dev);
1648 kfree(mp_ioapic_data[i]);
1649 mp_ioapic_data[i] = NULL;
1650 printk(KERN_ERR "Can't suspend/resume IOAPIC %d\n", i);
1658 device_initcall(ioapic_init_sysfs);
1661 * Dynamic irq allocate and deallocation
1663 int create_irq(void)
1665 /* Allocate an unused irq */
1669 unsigned long flags;
1672 spin_lock_irqsave(&vector_lock, flags);
1673 for (new = (NR_IRQS - 1); new >= 0; new--) {
1674 if (platform_legacy_irq(new))
1676 if (irq_vector[new] != 0)
1678 vector = __assign_irq_vector(new, TARGET_CPUS);
1679 if (likely(vector > 0))
1683 spin_unlock_irqrestore(&vector_lock, flags);
1686 dynamic_irq_init(irq);
1691 void destroy_irq(unsigned int irq)
1693 unsigned long flags;
1695 dynamic_irq_cleanup(irq);
1697 spin_lock_irqsave(&vector_lock, flags);
1698 irq_vector[irq] = 0;
1699 spin_unlock_irqrestore(&vector_lock, flags);
1703 * MSI mesage composition
1705 #ifdef CONFIG_PCI_MSI
1706 static int msi_compose_msg(struct pci_dev *pdev, unsigned int irq, struct msi_msg *msg)
1711 vector = assign_irq_vector(irq, TARGET_CPUS);
1716 cpu_set(vector >> 8, tmp);
1717 dest = cpu_mask_to_apicid(tmp);
1719 msg->address_hi = MSI_ADDR_BASE_HI;
1722 ((INT_DEST_MODE == 0) ?
1723 MSI_ADDR_DEST_MODE_PHYSICAL:
1724 MSI_ADDR_DEST_MODE_LOGICAL) |
1725 ((INT_DELIVERY_MODE != dest_LowestPrio) ?
1726 MSI_ADDR_REDIRECTION_CPU:
1727 MSI_ADDR_REDIRECTION_LOWPRI) |
1728 MSI_ADDR_DEST_ID(dest);
1731 MSI_DATA_TRIGGER_EDGE |
1732 MSI_DATA_LEVEL_ASSERT |
1733 ((INT_DELIVERY_MODE != dest_LowestPrio) ?
1734 MSI_DATA_DELIVERY_FIXED:
1735 MSI_DATA_DELIVERY_LOWPRI) |
1736 MSI_DATA_VECTOR(vector);
1742 static void set_msi_irq_affinity(unsigned int irq, cpumask_t mask)
1749 cpus_and(tmp, mask, cpu_online_map);
1750 if (cpus_empty(tmp))
1753 cpus_and(mask, tmp, CPU_MASK_ALL);
1755 vector = assign_irq_vector(irq, mask);
1760 cpu_set(vector >> 8, tmp);
1761 dest = cpu_mask_to_apicid(tmp);
1763 read_msi_msg(irq, &msg);
1765 msg.data &= ~MSI_DATA_VECTOR_MASK;
1766 msg.data |= MSI_DATA_VECTOR(vector);
1767 msg.address_lo &= ~MSI_ADDR_DEST_ID_MASK;
1768 msg.address_lo |= MSI_ADDR_DEST_ID(dest);
1770 write_msi_msg(irq, &msg);
1771 set_native_irq_info(irq, mask);
1773 #endif /* CONFIG_SMP */
1776 * IRQ Chip for MSI PCI/PCI-X/PCI-Express Devices,
1777 * which implement the MSI or MSI-X Capability Structure.
1779 static struct irq_chip msi_chip = {
1781 .unmask = unmask_msi_irq,
1782 .mask = mask_msi_irq,
1783 .ack = ack_apic_edge,
1785 .set_affinity = set_msi_irq_affinity,
1787 .retrigger = ioapic_retrigger_irq,
1790 int arch_setup_msi_irq(unsigned int irq, struct pci_dev *dev)
1794 ret = msi_compose_msg(dev, irq, &msg);
1798 write_msi_msg(irq, &msg);
1800 set_irq_chip_and_handler(irq, &msi_chip, handle_edge_irq);
1805 void arch_teardown_msi_irq(unsigned int irq)
1810 #endif /* CONFIG_PCI_MSI */
1813 * Hypertransport interrupt support
1815 #ifdef CONFIG_HT_IRQ
1819 static void target_ht_irq(unsigned int irq, unsigned int dest, u8 vector)
1822 low = read_ht_irq_low(irq);
1823 high = read_ht_irq_high(irq);
1825 low &= ~(HT_IRQ_LOW_VECTOR_MASK | HT_IRQ_LOW_DEST_ID_MASK);
1826 high &= ~(HT_IRQ_HIGH_DEST_ID_MASK);
1828 low |= HT_IRQ_LOW_VECTOR(vector) | HT_IRQ_LOW_DEST_ID(dest);
1829 high |= HT_IRQ_HIGH_DEST_ID(dest);
1831 write_ht_irq_low(irq, low);
1832 write_ht_irq_high(irq, high);
1835 static void set_ht_irq_affinity(unsigned int irq, cpumask_t mask)
1841 cpus_and(tmp, mask, cpu_online_map);
1842 if (cpus_empty(tmp))
1845 cpus_and(mask, tmp, CPU_MASK_ALL);
1847 vector = assign_irq_vector(irq, mask);
1852 cpu_set(vector >> 8, tmp);
1853 dest = cpu_mask_to_apicid(tmp);
1855 target_ht_irq(irq, dest, vector & 0xff);
1856 set_native_irq_info(irq, mask);
1860 static struct hw_interrupt_type ht_irq_chip = {
1862 .mask = mask_ht_irq,
1863 .unmask = unmask_ht_irq,
1864 .ack = ack_apic_edge,
1866 .set_affinity = set_ht_irq_affinity,
1868 .retrigger = ioapic_retrigger_irq,
1871 int arch_setup_ht_irq(unsigned int irq, struct pci_dev *dev)
1875 vector = assign_irq_vector(irq, TARGET_CPUS);
1882 cpu_set(vector >> 8, tmp);
1883 dest = cpu_mask_to_apicid(tmp);
1885 high = HT_IRQ_HIGH_DEST_ID(dest);
1887 low = HT_IRQ_LOW_BASE |
1888 HT_IRQ_LOW_DEST_ID(dest) |
1889 HT_IRQ_LOW_VECTOR(vector) |
1890 ((INT_DEST_MODE == 0) ?
1891 HT_IRQ_LOW_DM_PHYSICAL :
1892 HT_IRQ_LOW_DM_LOGICAL) |
1893 HT_IRQ_LOW_RQEOI_EDGE |
1894 ((INT_DELIVERY_MODE != dest_LowestPrio) ?
1895 HT_IRQ_LOW_MT_FIXED :
1896 HT_IRQ_LOW_MT_ARBITRATED);
1898 write_ht_irq_low(irq, low);
1899 write_ht_irq_high(irq, high);
1901 set_irq_chip_and_handler(irq, &ht_irq_chip, handle_edge_irq);
1905 #endif /* CONFIG_HT_IRQ */
1907 /* --------------------------------------------------------------------------
1908 ACPI-based IOAPIC Configuration
1909 -------------------------------------------------------------------------- */
1913 #define IO_APIC_MAX_ID 0xFE
1915 int __init io_apic_get_redir_entries (int ioapic)
1917 union IO_APIC_reg_01 reg_01;
1918 unsigned long flags;
1920 spin_lock_irqsave(&ioapic_lock, flags);
1921 reg_01.raw = io_apic_read(ioapic, 1);
1922 spin_unlock_irqrestore(&ioapic_lock, flags);
1924 return reg_01.bits.entries;
1928 int io_apic_set_pci_routing (int ioapic, int pin, int irq, int triggering, int polarity)
1930 struct IO_APIC_route_entry entry;
1931 unsigned long flags;
1935 if (!IO_APIC_IRQ(irq)) {
1936 apic_printk(APIC_QUIET,KERN_ERR "IOAPIC[%d]: Invalid reference to IRQ 0\n",
1942 * IRQs < 16 are already in the irq_2_pin[] map
1945 add_pin_to_irq(irq, ioapic, pin);
1948 vector = assign_irq_vector(irq, TARGET_CPUS);
1953 cpu_set(vector >> 8, mask);
1956 * Generate a PCI IRQ routing entry and program the IOAPIC accordingly.
1957 * Note that we mask (disable) IRQs now -- these get enabled when the
1958 * corresponding device driver registers for this IRQ.
1961 memset(&entry,0,sizeof(entry));
1963 entry.delivery_mode = INT_DELIVERY_MODE;
1964 entry.dest_mode = INT_DEST_MODE;
1965 entry.dest.logical.logical_dest = cpu_mask_to_apicid(mask);
1966 entry.trigger = triggering;
1967 entry.polarity = polarity;
1968 entry.mask = 1; /* Disabled (masked) */
1969 entry.vector = vector & 0xff;
1971 apic_printk(APIC_VERBOSE,KERN_DEBUG "IOAPIC[%d]: Set PCI routing entry (%d-%d -> 0x%x -> "
1972 "IRQ %d Mode:%i Active:%i)\n", ioapic,
1973 mp_ioapics[ioapic].mpc_apicid, pin, entry.vector, irq,
1974 triggering, polarity);
1976 ioapic_register_intr(irq, entry.vector, triggering);
1978 if (!ioapic && (irq < 16))
1979 disable_8259A_irq(irq);
1981 ioapic_write_entry(ioapic, pin, entry);
1983 spin_lock_irqsave(&ioapic_lock, flags);
1984 set_native_irq_info(irq, TARGET_CPUS);
1985 spin_unlock_irqrestore(&ioapic_lock, flags);
1990 #endif /* CONFIG_ACPI */
1994 * This function currently is only a helper for the i386 smp boot process where
1995 * we need to reprogram the ioredtbls to cater for the cpus which have come online
1996 * so mask in all cases should simply be TARGET_CPUS
1999 void __init setup_ioapic_dest(void)
2001 int pin, ioapic, irq, irq_entry;
2003 if (skip_ioapic_setup == 1)
2006 for (ioapic = 0; ioapic < nr_ioapics; ioapic++) {
2007 for (pin = 0; pin < nr_ioapic_registers[ioapic]; pin++) {
2008 irq_entry = find_irq_entry(ioapic, pin, mp_INT);
2009 if (irq_entry == -1)
2011 irq = pin_2_irq(irq_entry, ioapic, pin);
2012 set_ioapic_affinity_irq(irq, TARGET_CPUS);