[PATCH] x86_64: Remove unique APIC/IO-APIC ID check
[linux-2.6/zen-sources.git] / arch / x86_64 / kernel / io_apic.c
blobac7684324954ad77524852b77922d604e57a0cb0
1 /*
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>
15 * Fixes
16 * Maciej W. Rozycki : Bits for genuine 82489DX APICs;
17 * thanks to Eric Gilmore
18 * and Rolf G. Tews
19 * for testing these extensively
20 * Paul Diefenbaugh : Added full ACPI support
23 #include <linux/mm.h>
24 #include <linux/irq.h>
25 #include <linux/interrupt.h>
26 #include <linux/init.h>
27 #include <linux/delay.h>
28 #include <linux/sched.h>
29 #include <linux/config.h>
30 #include <linux/smp_lock.h>
31 #include <linux/mc146818rtc.h>
32 #include <linux/acpi.h>
33 #include <linux/sysdev.h>
35 #include <asm/io.h>
36 #include <asm/smp.h>
37 #include <asm/desc.h>
38 #include <asm/proto.h>
39 #include <asm/mach_apic.h>
41 #define __apicdebuginit __init
43 int sis_apic_bug; /* not actually supported, dummy for compile */
45 static DEFINE_SPINLOCK(ioapic_lock);
48 * # of IRQ routing registers
50 int nr_ioapic_registers[MAX_IO_APICS];
53 * Rough estimation of how many shared IRQs there are, can
54 * be changed anytime.
56 #define MAX_PLUS_SHARED_IRQS NR_IRQS
57 #define PIN_MAP_SIZE (MAX_PLUS_SHARED_IRQS + NR_IRQS)
60 * This is performance-critical, we want to do it O(1)
62 * the indexing order of this array favors 1:1 mappings
63 * between pins and IRQs.
66 static struct irq_pin_list {
67 short apic, pin, next;
68 } irq_2_pin[PIN_MAP_SIZE];
70 int vector_irq[NR_VECTORS] = { [0 ... NR_VECTORS - 1] = -1};
71 #ifdef CONFIG_PCI_MSI
72 #define vector_to_irq(vector) \
73 (platform_legacy_irq(vector) ? vector : vector_irq[vector])
74 #else
75 #define vector_to_irq(vector) (vector)
76 #endif
79 * The common case is 1:1 IRQ<->pin mappings. Sometimes there are
80 * shared ISA-space IRQs, so we have to support them. We are super
81 * fast in the common case, and fast for shared ISA-space IRQs.
83 static void add_pin_to_irq(unsigned int irq, int apic, int pin)
85 static int first_free_entry = NR_IRQS;
86 struct irq_pin_list *entry = irq_2_pin + irq;
88 while (entry->next)
89 entry = irq_2_pin + entry->next;
91 if (entry->pin != -1) {
92 entry->next = first_free_entry;
93 entry = irq_2_pin + entry->next;
94 if (++first_free_entry >= PIN_MAP_SIZE)
95 panic("io_apic.c: whoops");
97 entry->apic = apic;
98 entry->pin = pin;
101 #define __DO_ACTION(R, ACTION, FINAL) \
104 int pin; \
105 struct irq_pin_list *entry = irq_2_pin + irq; \
107 for (;;) { \
108 unsigned int reg; \
109 pin = entry->pin; \
110 if (pin == -1) \
111 break; \
112 reg = io_apic_read(entry->apic, 0x10 + R + pin*2); \
113 reg ACTION; \
114 io_apic_modify(entry->apic, reg); \
115 if (!entry->next) \
116 break; \
117 entry = irq_2_pin + entry->next; \
119 FINAL; \
122 #define DO_ACTION(name,R,ACTION, FINAL) \
124 static void name##_IO_APIC_irq (unsigned int irq) \
125 __DO_ACTION(R, ACTION, FINAL)
127 DO_ACTION( __mask, 0, |= 0x00010000, io_apic_sync(entry->apic) )
128 /* mask = 1 */
129 DO_ACTION( __unmask, 0, &= 0xfffeffff, )
130 /* mask = 0 */
132 static void mask_IO_APIC_irq (unsigned int irq)
134 unsigned long flags;
136 spin_lock_irqsave(&ioapic_lock, flags);
137 __mask_IO_APIC_irq(irq);
138 spin_unlock_irqrestore(&ioapic_lock, flags);
141 static void unmask_IO_APIC_irq (unsigned int irq)
143 unsigned long flags;
145 spin_lock_irqsave(&ioapic_lock, flags);
146 __unmask_IO_APIC_irq(irq);
147 spin_unlock_irqrestore(&ioapic_lock, flags);
150 static void clear_IO_APIC_pin(unsigned int apic, unsigned int pin)
152 struct IO_APIC_route_entry entry;
153 unsigned long flags;
155 /* Check delivery_mode to be sure we're not clearing an SMI pin */
156 spin_lock_irqsave(&ioapic_lock, flags);
157 *(((int*)&entry) + 0) = io_apic_read(apic, 0x10 + 2 * pin);
158 *(((int*)&entry) + 1) = io_apic_read(apic, 0x11 + 2 * pin);
159 spin_unlock_irqrestore(&ioapic_lock, flags);
160 if (entry.delivery_mode == dest_SMI)
161 return;
163 * Disable it in the IO-APIC irq-routing table:
165 memset(&entry, 0, sizeof(entry));
166 entry.mask = 1;
167 spin_lock_irqsave(&ioapic_lock, flags);
168 io_apic_write(apic, 0x10 + 2 * pin, *(((int *)&entry) + 0));
169 io_apic_write(apic, 0x11 + 2 * pin, *(((int *)&entry) + 1));
170 spin_unlock_irqrestore(&ioapic_lock, flags);
173 static void clear_IO_APIC (void)
175 int apic, pin;
177 for (apic = 0; apic < nr_ioapics; apic++)
178 for (pin = 0; pin < nr_ioapic_registers[apic]; pin++)
179 clear_IO_APIC_pin(apic, pin);
183 * support for broken MP BIOSs, enables hand-redirection of PIRQ0-7 to
184 * specific CPU-side IRQs.
187 #define MAX_PIRQS 8
188 static int pirq_entries [MAX_PIRQS];
189 static int pirqs_enabled;
190 int skip_ioapic_setup;
191 int ioapic_force;
193 /* dummy parsing: see setup.c */
195 static int __init disable_ioapic_setup(char *str)
197 skip_ioapic_setup = 1;
198 return 1;
201 static int __init enable_ioapic_setup(char *str)
203 ioapic_force = 1;
204 skip_ioapic_setup = 0;
205 return 1;
208 __setup("noapic", disable_ioapic_setup);
209 __setup("apic", enable_ioapic_setup);
211 #include <asm/pci-direct.h>
212 #include <linux/pci_ids.h>
213 #include <linux/pci.h>
215 /* Temporary Hack. Nvidia and VIA boards currently only work with IO-APIC
216 off. Check for an Nvidia or VIA PCI bridge and turn it off.
217 Use pci direct infrastructure because this runs before the PCI subsystem.
219 Can be overwritten with "apic"
221 And another hack to disable the IOMMU on VIA chipsets.
223 Kludge-O-Rama. */
224 void __init check_ioapic(void)
226 int num,slot,func;
227 if (ioapic_force)
228 return;
230 /* Poor man's PCI discovery */
231 for (num = 0; num < 32; num++) {
232 for (slot = 0; slot < 32; slot++) {
233 for (func = 0; func < 8; func++) {
234 u32 class;
235 u32 vendor;
236 u8 type;
237 class = read_pci_config(num,slot,func,
238 PCI_CLASS_REVISION);
239 if (class == 0xffffffff)
240 break;
242 if ((class >> 16) != PCI_CLASS_BRIDGE_PCI)
243 continue;
245 vendor = read_pci_config(num, slot, func,
246 PCI_VENDOR_ID);
247 vendor &= 0xffff;
248 switch (vendor) {
249 case PCI_VENDOR_ID_VIA:
250 #ifdef CONFIG_GART_IOMMU
251 if ((end_pfn >= (0xffffffff>>PAGE_SHIFT) ||
252 force_iommu) &&
253 !iommu_aperture_allowed) {
254 printk(KERN_INFO
255 "Looks like a VIA chipset. Disabling IOMMU. Overwrite with \"iommu=allowed\"\n");
256 iommu_aperture_disabled = 1;
258 #endif
259 return;
260 case PCI_VENDOR_ID_NVIDIA:
261 #ifdef CONFIG_ACPI
262 /* All timer overrides on Nvidia
263 seem to be wrong. Skip them. */
264 acpi_skip_timer_override = 1;
265 printk(KERN_INFO
266 "Nvidia board detected. Ignoring ACPI timer override.\n");
267 #endif
268 /* RED-PEN skip them on mptables too? */
269 return;
272 /* No multi-function device? */
273 type = read_pci_config_byte(num,slot,func,
274 PCI_HEADER_TYPE);
275 if (!(type & 0x80))
276 break;
282 static int __init ioapic_pirq_setup(char *str)
284 int i, max;
285 int ints[MAX_PIRQS+1];
287 get_options(str, ARRAY_SIZE(ints), ints);
289 for (i = 0; i < MAX_PIRQS; i++)
290 pirq_entries[i] = -1;
292 pirqs_enabled = 1;
293 apic_printk(APIC_VERBOSE, "PIRQ redirection, working around broken MP-BIOS.\n");
294 max = MAX_PIRQS;
295 if (ints[0] < MAX_PIRQS)
296 max = ints[0];
298 for (i = 0; i < max; i++) {
299 apic_printk(APIC_VERBOSE, "... PIRQ%d -> IRQ %d\n", i, ints[i+1]);
301 * PIRQs are mapped upside down, usually.
303 pirq_entries[MAX_PIRQS-i-1] = ints[i+1];
305 return 1;
308 __setup("pirq=", ioapic_pirq_setup);
311 * Find the IRQ entry number of a certain pin.
313 static int find_irq_entry(int apic, int pin, int type)
315 int i;
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)
322 return i;
324 return -1;
328 * Find the pin to which IRQ[irq] (ISA) is connected
330 static int __init find_isa_irq_pin(int irq, int type)
332 int i;
334 for (i = 0; i < mp_irq_entries; i++) {
335 int lbus = mp_irqs[i].mpc_srcbus;
337 if ((mp_bus_id_to_type[lbus] == MP_BUS_ISA ||
338 mp_bus_id_to_type[lbus] == MP_BUS_EISA ||
339 mp_bus_id_to_type[lbus] == MP_BUS_MCA) &&
340 (mp_irqs[i].mpc_irqtype == type) &&
341 (mp_irqs[i].mpc_srcbusirq == irq))
343 return mp_irqs[i].mpc_dstirq;
345 return -1;
349 * Find a specific PCI IRQ entry.
350 * Not an __init, possibly needed by modules
352 static int pin_2_irq(int idx, int apic, int pin);
354 int IO_APIC_get_PCI_irq_vector(int bus, int slot, int pin)
356 int apic, i, best_guess = -1;
358 apic_printk(APIC_DEBUG, "querying PCI -> IRQ mapping bus:%d, slot:%d, pin:%d.\n",
359 bus, slot, pin);
360 if (mp_bus_id_to_pci_bus[bus] == -1) {
361 apic_printk(APIC_VERBOSE, "PCI BIOS passed nonexistent PCI bus %d!\n", bus);
362 return -1;
364 for (i = 0; i < mp_irq_entries; i++) {
365 int lbus = mp_irqs[i].mpc_srcbus;
367 for (apic = 0; apic < nr_ioapics; apic++)
368 if (mp_ioapics[apic].mpc_apicid == mp_irqs[i].mpc_dstapic ||
369 mp_irqs[i].mpc_dstapic == MP_APIC_ALL)
370 break;
372 if ((mp_bus_id_to_type[lbus] == MP_BUS_PCI) &&
373 !mp_irqs[i].mpc_irqtype &&
374 (bus == lbus) &&
375 (slot == ((mp_irqs[i].mpc_srcbusirq >> 2) & 0x1f))) {
376 int irq = pin_2_irq(i,apic,mp_irqs[i].mpc_dstirq);
378 if (!(apic || IO_APIC_IRQ(irq)))
379 continue;
381 if (pin == (mp_irqs[i].mpc_srcbusirq & 3))
382 return irq;
384 * Use the first all-but-pin matching entry as a
385 * best-guess fuzzy result for broken mptables.
387 if (best_guess < 0)
388 best_guess = irq;
391 return best_guess;
395 * EISA Edge/Level control register, ELCR
397 static int EISA_ELCR(unsigned int irq)
399 if (irq < 16) {
400 unsigned int port = 0x4d0 + (irq >> 3);
401 return (inb(port) >> (irq & 7)) & 1;
403 apic_printk(APIC_VERBOSE, "Broken MPtable reports ISA irq %d\n", irq);
404 return 0;
407 /* EISA interrupts are always polarity zero and can be edge or level
408 * trigger depending on the ELCR value. If an interrupt is listed as
409 * EISA conforming in the MP table, that means its trigger type must
410 * be read in from the ELCR */
412 #define default_EISA_trigger(idx) (EISA_ELCR(mp_irqs[idx].mpc_srcbusirq))
413 #define default_EISA_polarity(idx) (0)
415 /* ISA interrupts are always polarity zero edge triggered,
416 * when listed as conforming in the MP table. */
418 #define default_ISA_trigger(idx) (0)
419 #define default_ISA_polarity(idx) (0)
421 /* PCI interrupts are always polarity one level triggered,
422 * when listed as conforming in the MP table. */
424 #define default_PCI_trigger(idx) (1)
425 #define default_PCI_polarity(idx) (1)
427 /* MCA interrupts are always polarity zero level triggered,
428 * when listed as conforming in the MP table. */
430 #define default_MCA_trigger(idx) (1)
431 #define default_MCA_polarity(idx) (0)
433 static int __init MPBIOS_polarity(int idx)
435 int bus = mp_irqs[idx].mpc_srcbus;
436 int polarity;
439 * Determine IRQ line polarity (high active or low active):
441 switch (mp_irqs[idx].mpc_irqflag & 3)
443 case 0: /* conforms, ie. bus-type dependent polarity */
445 switch (mp_bus_id_to_type[bus])
447 case MP_BUS_ISA: /* ISA pin */
449 polarity = default_ISA_polarity(idx);
450 break;
452 case MP_BUS_EISA: /* EISA pin */
454 polarity = default_EISA_polarity(idx);
455 break;
457 case MP_BUS_PCI: /* PCI pin */
459 polarity = default_PCI_polarity(idx);
460 break;
462 case MP_BUS_MCA: /* MCA pin */
464 polarity = default_MCA_polarity(idx);
465 break;
467 default:
469 printk(KERN_WARNING "broken BIOS!!\n");
470 polarity = 1;
471 break;
474 break;
476 case 1: /* high active */
478 polarity = 0;
479 break;
481 case 2: /* reserved */
483 printk(KERN_WARNING "broken BIOS!!\n");
484 polarity = 1;
485 break;
487 case 3: /* low active */
489 polarity = 1;
490 break;
492 default: /* invalid */
494 printk(KERN_WARNING "broken BIOS!!\n");
495 polarity = 1;
496 break;
499 return polarity;
502 static int MPBIOS_trigger(int idx)
504 int bus = mp_irqs[idx].mpc_srcbus;
505 int trigger;
508 * Determine IRQ trigger mode (edge or level sensitive):
510 switch ((mp_irqs[idx].mpc_irqflag>>2) & 3)
512 case 0: /* conforms, ie. bus-type dependent */
514 switch (mp_bus_id_to_type[bus])
516 case MP_BUS_ISA: /* ISA pin */
518 trigger = default_ISA_trigger(idx);
519 break;
521 case MP_BUS_EISA: /* EISA pin */
523 trigger = default_EISA_trigger(idx);
524 break;
526 case MP_BUS_PCI: /* PCI pin */
528 trigger = default_PCI_trigger(idx);
529 break;
531 case MP_BUS_MCA: /* MCA pin */
533 trigger = default_MCA_trigger(idx);
534 break;
536 default:
538 printk(KERN_WARNING "broken BIOS!!\n");
539 trigger = 1;
540 break;
543 break;
545 case 1: /* edge */
547 trigger = 0;
548 break;
550 case 2: /* reserved */
552 printk(KERN_WARNING "broken BIOS!!\n");
553 trigger = 1;
554 break;
556 case 3: /* level */
558 trigger = 1;
559 break;
561 default: /* invalid */
563 printk(KERN_WARNING "broken BIOS!!\n");
564 trigger = 0;
565 break;
568 return trigger;
571 static inline int irq_polarity(int idx)
573 return MPBIOS_polarity(idx);
576 static inline int irq_trigger(int idx)
578 return MPBIOS_trigger(idx);
581 static int pin_2_irq(int idx, int apic, int pin)
583 int irq, i;
584 int bus = mp_irqs[idx].mpc_srcbus;
587 * Debugging check, we are in big trouble if this message pops up!
589 if (mp_irqs[idx].mpc_dstirq != pin)
590 printk(KERN_ERR "broken BIOS or MPTABLE parser, ayiee!!\n");
592 switch (mp_bus_id_to_type[bus])
594 case MP_BUS_ISA: /* ISA pin */
595 case MP_BUS_EISA:
596 case MP_BUS_MCA:
598 irq = mp_irqs[idx].mpc_srcbusirq;
599 break;
601 case MP_BUS_PCI: /* PCI pin */
604 * PCI IRQs are mapped in order
606 i = irq = 0;
607 while (i < apic)
608 irq += nr_ioapic_registers[i++];
609 irq += pin;
610 break;
612 default:
614 printk(KERN_ERR "unknown bus type %d.\n",bus);
615 irq = 0;
616 break;
621 * PCI IRQ command line redirection. Yes, limits are hardcoded.
623 if ((pin >= 16) && (pin <= 23)) {
624 if (pirq_entries[pin-16] != -1) {
625 if (!pirq_entries[pin-16]) {
626 apic_printk(APIC_VERBOSE, "disabling PIRQ%d\n", pin-16);
627 } else {
628 irq = pirq_entries[pin-16];
629 apic_printk(APIC_VERBOSE, "using PIRQ%d -> IRQ %d\n",
630 pin-16, irq);
634 return irq;
637 static inline int IO_APIC_irq_trigger(int irq)
639 int apic, idx, pin;
641 for (apic = 0; apic < nr_ioapics; apic++) {
642 for (pin = 0; pin < nr_ioapic_registers[apic]; pin++) {
643 idx = find_irq_entry(apic,pin,mp_INT);
644 if ((idx != -1) && (irq == pin_2_irq(idx,apic,pin)))
645 return irq_trigger(idx);
649 * nonexistent IRQs are edge default
651 return 0;
654 /* irq_vectors is indexed by the sum of all RTEs in all I/O APICs. */
655 u8 irq_vector[NR_IRQ_VECTORS] = { FIRST_DEVICE_VECTOR , 0 };
657 int assign_irq_vector(int irq)
659 static int current_vector = FIRST_DEVICE_VECTOR, offset = 0;
661 BUG_ON(irq >= NR_IRQ_VECTORS);
662 if (IO_APIC_VECTOR(irq) > 0)
663 return IO_APIC_VECTOR(irq);
664 next:
665 current_vector += 8;
666 if (current_vector == IA32_SYSCALL_VECTOR)
667 goto next;
669 if (current_vector >= FIRST_SYSTEM_VECTOR) {
670 offset++;
671 if (!(offset%8))
672 return -ENOSPC;
673 current_vector = FIRST_DEVICE_VECTOR + offset;
676 vector_irq[current_vector] = irq;
677 if (irq != AUTO_ASSIGN)
678 IO_APIC_VECTOR(irq) = current_vector;
680 return current_vector;
683 extern void (*interrupt[NR_IRQS])(void);
684 static struct hw_interrupt_type ioapic_level_type;
685 static struct hw_interrupt_type ioapic_edge_type;
687 #define IOAPIC_AUTO -1
688 #define IOAPIC_EDGE 0
689 #define IOAPIC_LEVEL 1
691 static inline void ioapic_register_intr(int irq, int vector, unsigned long trigger)
693 if (use_pci_vector() && !platform_legacy_irq(irq)) {
694 if ((trigger == IOAPIC_AUTO && IO_APIC_irq_trigger(irq)) ||
695 trigger == IOAPIC_LEVEL)
696 irq_desc[vector].handler = &ioapic_level_type;
697 else
698 irq_desc[vector].handler = &ioapic_edge_type;
699 set_intr_gate(vector, interrupt[vector]);
700 } else {
701 if ((trigger == IOAPIC_AUTO && IO_APIC_irq_trigger(irq)) ||
702 trigger == IOAPIC_LEVEL)
703 irq_desc[irq].handler = &ioapic_level_type;
704 else
705 irq_desc[irq].handler = &ioapic_edge_type;
706 set_intr_gate(vector, interrupt[irq]);
710 static void __init setup_IO_APIC_irqs(void)
712 struct IO_APIC_route_entry entry;
713 int apic, pin, idx, irq, first_notcon = 1, vector;
714 unsigned long flags;
716 apic_printk(APIC_VERBOSE, KERN_DEBUG "init IO_APIC IRQs\n");
718 for (apic = 0; apic < nr_ioapics; apic++) {
719 for (pin = 0; pin < nr_ioapic_registers[apic]; pin++) {
722 * add it to the IO-APIC irq-routing table:
724 memset(&entry,0,sizeof(entry));
726 entry.delivery_mode = INT_DELIVERY_MODE;
727 entry.dest_mode = INT_DEST_MODE;
728 entry.mask = 0; /* enable IRQ */
729 entry.dest.logical.logical_dest = cpu_mask_to_apicid(TARGET_CPUS);
731 idx = find_irq_entry(apic,pin,mp_INT);
732 if (idx == -1) {
733 if (first_notcon) {
734 apic_printk(APIC_VERBOSE, KERN_DEBUG " IO-APIC (apicid-pin) %d-%d", mp_ioapics[apic].mpc_apicid, pin);
735 first_notcon = 0;
736 } else
737 apic_printk(APIC_VERBOSE, ", %d-%d", mp_ioapics[apic].mpc_apicid, pin);
738 continue;
741 entry.trigger = irq_trigger(idx);
742 entry.polarity = irq_polarity(idx);
744 if (irq_trigger(idx)) {
745 entry.trigger = 1;
746 entry.mask = 1;
747 entry.dest.logical.logical_dest = cpu_mask_to_apicid(TARGET_CPUS);
750 irq = pin_2_irq(idx, apic, pin);
751 add_pin_to_irq(irq, apic, pin);
753 if (!apic && !IO_APIC_IRQ(irq))
754 continue;
756 if (IO_APIC_IRQ(irq)) {
757 vector = assign_irq_vector(irq);
758 entry.vector = vector;
760 ioapic_register_intr(irq, vector, IOAPIC_AUTO);
761 if (!apic && (irq < 16))
762 disable_8259A_irq(irq);
764 spin_lock_irqsave(&ioapic_lock, flags);
765 io_apic_write(apic, 0x11+2*pin, *(((int *)&entry)+1));
766 io_apic_write(apic, 0x10+2*pin, *(((int *)&entry)+0));
767 spin_unlock_irqrestore(&ioapic_lock, flags);
771 if (!first_notcon)
772 apic_printk(APIC_VERBOSE," not connected.\n");
776 * Set up the 8259A-master output pin as broadcast to all
777 * CPUs.
779 static void __init setup_ExtINT_IRQ0_pin(unsigned int pin, int vector)
781 struct IO_APIC_route_entry entry;
782 unsigned long flags;
784 memset(&entry,0,sizeof(entry));
786 disable_8259A_irq(0);
788 /* mask LVT0 */
789 apic_write_around(APIC_LVT0, APIC_LVT_MASKED | APIC_DM_EXTINT);
792 * We use logical delivery to get the timer IRQ
793 * to the first CPU.
795 entry.dest_mode = INT_DEST_MODE;
796 entry.mask = 0; /* unmask IRQ now */
797 entry.dest.logical.logical_dest = cpu_mask_to_apicid(TARGET_CPUS);
798 entry.delivery_mode = INT_DELIVERY_MODE;
799 entry.polarity = 0;
800 entry.trigger = 0;
801 entry.vector = vector;
804 * The timer IRQ doesn't have to know that behind the
805 * scene we have a 8259A-master in AEOI mode ...
807 irq_desc[0].handler = &ioapic_edge_type;
810 * Add it to the IO-APIC irq-routing table:
812 spin_lock_irqsave(&ioapic_lock, flags);
813 io_apic_write(0, 0x11+2*pin, *(((int *)&entry)+1));
814 io_apic_write(0, 0x10+2*pin, *(((int *)&entry)+0));
815 spin_unlock_irqrestore(&ioapic_lock, flags);
817 enable_8259A_irq(0);
820 void __init UNEXPECTED_IO_APIC(void)
824 void __apicdebuginit print_IO_APIC(void)
826 int apic, i;
827 union IO_APIC_reg_00 reg_00;
828 union IO_APIC_reg_01 reg_01;
829 union IO_APIC_reg_02 reg_02;
830 unsigned long flags;
832 if (apic_verbosity == APIC_QUIET)
833 return;
835 printk(KERN_DEBUG "number of MP IRQ sources: %d.\n", mp_irq_entries);
836 for (i = 0; i < nr_ioapics; i++)
837 printk(KERN_DEBUG "number of IO-APIC #%d registers: %d.\n",
838 mp_ioapics[i].mpc_apicid, nr_ioapic_registers[i]);
841 * We are a bit conservative about what we expect. We have to
842 * know about every hardware change ASAP.
844 printk(KERN_INFO "testing the IO APIC.......................\n");
846 for (apic = 0; apic < nr_ioapics; apic++) {
848 spin_lock_irqsave(&ioapic_lock, flags);
849 reg_00.raw = io_apic_read(apic, 0);
850 reg_01.raw = io_apic_read(apic, 1);
851 if (reg_01.bits.version >= 0x10)
852 reg_02.raw = io_apic_read(apic, 2);
853 spin_unlock_irqrestore(&ioapic_lock, flags);
855 printk("\n");
856 printk(KERN_DEBUG "IO APIC #%d......\n", mp_ioapics[apic].mpc_apicid);
857 printk(KERN_DEBUG ".... register #00: %08X\n", reg_00.raw);
858 printk(KERN_DEBUG "....... : physical APIC id: %02X\n", reg_00.bits.ID);
859 if (reg_00.bits.__reserved_1 || reg_00.bits.__reserved_2)
860 UNEXPECTED_IO_APIC();
862 printk(KERN_DEBUG ".... register #01: %08X\n", *(int *)&reg_01);
863 printk(KERN_DEBUG "....... : max redirection entries: %04X\n", reg_01.bits.entries);
864 if ( (reg_01.bits.entries != 0x0f) && /* older (Neptune) boards */
865 (reg_01.bits.entries != 0x17) && /* typical ISA+PCI boards */
866 (reg_01.bits.entries != 0x1b) && /* Compaq Proliant boards */
867 (reg_01.bits.entries != 0x1f) && /* dual Xeon boards */
868 (reg_01.bits.entries != 0x22) && /* bigger Xeon boards */
869 (reg_01.bits.entries != 0x2E) &&
870 (reg_01.bits.entries != 0x3F) &&
871 (reg_01.bits.entries != 0x03)
873 UNEXPECTED_IO_APIC();
875 printk(KERN_DEBUG "....... : PRQ implemented: %X\n", reg_01.bits.PRQ);
876 printk(KERN_DEBUG "....... : IO APIC version: %04X\n", reg_01.bits.version);
877 if ( (reg_01.bits.version != 0x01) && /* 82489DX IO-APICs */
878 (reg_01.bits.version != 0x02) && /* 82801BA IO-APICs (ICH2) */
879 (reg_01.bits.version != 0x10) && /* oldest IO-APICs */
880 (reg_01.bits.version != 0x11) && /* Pentium/Pro IO-APICs */
881 (reg_01.bits.version != 0x13) && /* Xeon IO-APICs */
882 (reg_01.bits.version != 0x20) /* Intel P64H (82806 AA) */
884 UNEXPECTED_IO_APIC();
885 if (reg_01.bits.__reserved_1 || reg_01.bits.__reserved_2)
886 UNEXPECTED_IO_APIC();
888 if (reg_01.bits.version >= 0x10) {
889 printk(KERN_DEBUG ".... register #02: %08X\n", reg_02.raw);
890 printk(KERN_DEBUG "....... : arbitration: %02X\n", reg_02.bits.arbitration);
891 if (reg_02.bits.__reserved_1 || reg_02.bits.__reserved_2)
892 UNEXPECTED_IO_APIC();
895 printk(KERN_DEBUG ".... IRQ redirection table:\n");
897 printk(KERN_DEBUG " NR Log Phy Mask Trig IRR Pol"
898 " Stat Dest Deli Vect: \n");
900 for (i = 0; i <= reg_01.bits.entries; i++) {
901 struct IO_APIC_route_entry entry;
903 spin_lock_irqsave(&ioapic_lock, flags);
904 *(((int *)&entry)+0) = io_apic_read(apic, 0x10+i*2);
905 *(((int *)&entry)+1) = io_apic_read(apic, 0x11+i*2);
906 spin_unlock_irqrestore(&ioapic_lock, flags);
908 printk(KERN_DEBUG " %02x %03X %02X ",
910 entry.dest.logical.logical_dest,
911 entry.dest.physical.physical_dest
914 printk("%1d %1d %1d %1d %1d %1d %1d %02X\n",
915 entry.mask,
916 entry.trigger,
917 entry.irr,
918 entry.polarity,
919 entry.delivery_status,
920 entry.dest_mode,
921 entry.delivery_mode,
922 entry.vector
926 if (use_pci_vector())
927 printk(KERN_INFO "Using vector-based indexing\n");
928 printk(KERN_DEBUG "IRQ to pin mappings:\n");
929 for (i = 0; i < NR_IRQS; i++) {
930 struct irq_pin_list *entry = irq_2_pin + i;
931 if (entry->pin < 0)
932 continue;
933 if (use_pci_vector() && !platform_legacy_irq(i))
934 printk(KERN_DEBUG "IRQ%d ", IO_APIC_VECTOR(i));
935 else
936 printk(KERN_DEBUG "IRQ%d ", i);
937 for (;;) {
938 printk("-> %d:%d", entry->apic, entry->pin);
939 if (!entry->next)
940 break;
941 entry = irq_2_pin + entry->next;
943 printk("\n");
946 printk(KERN_INFO ".................................... done.\n");
948 return;
951 #if 0
953 static __apicdebuginit void print_APIC_bitfield (int base)
955 unsigned int v;
956 int i, j;
958 if (apic_verbosity == APIC_QUIET)
959 return;
961 printk(KERN_DEBUG "0123456789abcdef0123456789abcdef\n" KERN_DEBUG);
962 for (i = 0; i < 8; i++) {
963 v = apic_read(base + i*0x10);
964 for (j = 0; j < 32; j++) {
965 if (v & (1<<j))
966 printk("1");
967 else
968 printk("0");
970 printk("\n");
974 void __apicdebuginit print_local_APIC(void * dummy)
976 unsigned int v, ver, maxlvt;
978 if (apic_verbosity == APIC_QUIET)
979 return;
981 printk("\n" KERN_DEBUG "printing local APIC contents on CPU#%d/%d:\n",
982 smp_processor_id(), hard_smp_processor_id());
983 v = apic_read(APIC_ID);
984 printk(KERN_INFO "... APIC ID: %08x (%01x)\n", v, GET_APIC_ID(v));
985 v = apic_read(APIC_LVR);
986 printk(KERN_INFO "... APIC VERSION: %08x\n", v);
987 ver = GET_APIC_VERSION(v);
988 maxlvt = get_maxlvt();
990 v = apic_read(APIC_TASKPRI);
991 printk(KERN_DEBUG "... APIC TASKPRI: %08x (%02x)\n", v, v & APIC_TPRI_MASK);
993 if (APIC_INTEGRATED(ver)) { /* !82489DX */
994 v = apic_read(APIC_ARBPRI);
995 printk(KERN_DEBUG "... APIC ARBPRI: %08x (%02x)\n", v,
996 v & APIC_ARBPRI_MASK);
997 v = apic_read(APIC_PROCPRI);
998 printk(KERN_DEBUG "... APIC PROCPRI: %08x\n", v);
1001 v = apic_read(APIC_EOI);
1002 printk(KERN_DEBUG "... APIC EOI: %08x\n", v);
1003 v = apic_read(APIC_RRR);
1004 printk(KERN_DEBUG "... APIC RRR: %08x\n", v);
1005 v = apic_read(APIC_LDR);
1006 printk(KERN_DEBUG "... APIC LDR: %08x\n", v);
1007 v = apic_read(APIC_DFR);
1008 printk(KERN_DEBUG "... APIC DFR: %08x\n", v);
1009 v = apic_read(APIC_SPIV);
1010 printk(KERN_DEBUG "... APIC SPIV: %08x\n", v);
1012 printk(KERN_DEBUG "... APIC ISR field:\n");
1013 print_APIC_bitfield(APIC_ISR);
1014 printk(KERN_DEBUG "... APIC TMR field:\n");
1015 print_APIC_bitfield(APIC_TMR);
1016 printk(KERN_DEBUG "... APIC IRR field:\n");
1017 print_APIC_bitfield(APIC_IRR);
1019 if (APIC_INTEGRATED(ver)) { /* !82489DX */
1020 if (maxlvt > 3) /* Due to the Pentium erratum 3AP. */
1021 apic_write(APIC_ESR, 0);
1022 v = apic_read(APIC_ESR);
1023 printk(KERN_DEBUG "... APIC ESR: %08x\n", v);
1026 v = apic_read(APIC_ICR);
1027 printk(KERN_DEBUG "... APIC ICR: %08x\n", v);
1028 v = apic_read(APIC_ICR2);
1029 printk(KERN_DEBUG "... APIC ICR2: %08x\n", v);
1031 v = apic_read(APIC_LVTT);
1032 printk(KERN_DEBUG "... APIC LVTT: %08x\n", v);
1034 if (maxlvt > 3) { /* PC is LVT#4. */
1035 v = apic_read(APIC_LVTPC);
1036 printk(KERN_DEBUG "... APIC LVTPC: %08x\n", v);
1038 v = apic_read(APIC_LVT0);
1039 printk(KERN_DEBUG "... APIC LVT0: %08x\n", v);
1040 v = apic_read(APIC_LVT1);
1041 printk(KERN_DEBUG "... APIC LVT1: %08x\n", v);
1043 if (maxlvt > 2) { /* ERR is LVT#3. */
1044 v = apic_read(APIC_LVTERR);
1045 printk(KERN_DEBUG "... APIC LVTERR: %08x\n", v);
1048 v = apic_read(APIC_TMICT);
1049 printk(KERN_DEBUG "... APIC TMICT: %08x\n", v);
1050 v = apic_read(APIC_TMCCT);
1051 printk(KERN_DEBUG "... APIC TMCCT: %08x\n", v);
1052 v = apic_read(APIC_TDCR);
1053 printk(KERN_DEBUG "... APIC TDCR: %08x\n", v);
1054 printk("\n");
1057 void print_all_local_APICs (void)
1059 on_each_cpu(print_local_APIC, NULL, 1, 1);
1062 void __apicdebuginit print_PIC(void)
1064 extern spinlock_t i8259A_lock;
1065 unsigned int v;
1066 unsigned long flags;
1068 if (apic_verbosity == APIC_QUIET)
1069 return;
1071 printk(KERN_DEBUG "\nprinting PIC contents\n");
1073 spin_lock_irqsave(&i8259A_lock, flags);
1075 v = inb(0xa1) << 8 | inb(0x21);
1076 printk(KERN_DEBUG "... PIC IMR: %04x\n", v);
1078 v = inb(0xa0) << 8 | inb(0x20);
1079 printk(KERN_DEBUG "... PIC IRR: %04x\n", v);
1081 outb(0x0b,0xa0);
1082 outb(0x0b,0x20);
1083 v = inb(0xa0) << 8 | inb(0x20);
1084 outb(0x0a,0xa0);
1085 outb(0x0a,0x20);
1087 spin_unlock_irqrestore(&i8259A_lock, flags);
1089 printk(KERN_DEBUG "... PIC ISR: %04x\n", v);
1091 v = inb(0x4d1) << 8 | inb(0x4d0);
1092 printk(KERN_DEBUG "... PIC ELCR: %04x\n", v);
1095 #endif /* 0 */
1097 static void __init enable_IO_APIC(void)
1099 union IO_APIC_reg_01 reg_01;
1100 int i;
1101 unsigned long flags;
1103 for (i = 0; i < PIN_MAP_SIZE; i++) {
1104 irq_2_pin[i].pin = -1;
1105 irq_2_pin[i].next = 0;
1107 if (!pirqs_enabled)
1108 for (i = 0; i < MAX_PIRQS; i++)
1109 pirq_entries[i] = -1;
1112 * The number of IO-APIC IRQ registers (== #pins):
1114 for (i = 0; i < nr_ioapics; i++) {
1115 spin_lock_irqsave(&ioapic_lock, flags);
1116 reg_01.raw = io_apic_read(i, 1);
1117 spin_unlock_irqrestore(&ioapic_lock, flags);
1118 nr_ioapic_registers[i] = reg_01.bits.entries+1;
1122 * Do not trust the IO-APIC being empty at bootup
1124 clear_IO_APIC();
1128 * Not an __init, needed by the reboot code
1130 void disable_IO_APIC(void)
1133 * Clear the IO-APIC before rebooting:
1135 clear_IO_APIC();
1137 disconnect_bsp_APIC();
1141 * function to set the IO-APIC physical IDs based on the
1142 * values stored in the MPC table.
1144 * by Matt Domsch <Matt_Domsch@dell.com> Tue Dec 21 12:25:05 CST 1999
1147 static void __init setup_ioapic_ids_from_mpc (void)
1149 union IO_APIC_reg_00 reg_00;
1150 int apic;
1151 int i;
1152 unsigned char old_id;
1153 unsigned long flags;
1156 * Set the IOAPIC ID to the value stored in the MPC table.
1158 for (apic = 0; apic < nr_ioapics; apic++) {
1160 /* Read the register 0 value */
1161 spin_lock_irqsave(&ioapic_lock, flags);
1162 reg_00.raw = io_apic_read(apic, 0);
1163 spin_unlock_irqrestore(&ioapic_lock, flags);
1165 old_id = mp_ioapics[apic].mpc_apicid;
1168 printk(KERN_INFO "Using IO-APIC %d\n", mp_ioapics[apic].mpc_apicid);
1172 * We need to adjust the IRQ routing table
1173 * if the ID changed.
1175 if (old_id != mp_ioapics[apic].mpc_apicid)
1176 for (i = 0; i < mp_irq_entries; i++)
1177 if (mp_irqs[i].mpc_dstapic == old_id)
1178 mp_irqs[i].mpc_dstapic
1179 = mp_ioapics[apic].mpc_apicid;
1182 * Read the right value from the MPC table and
1183 * write it into the ID register.
1185 apic_printk(APIC_VERBOSE,KERN_INFO "...changing IO-APIC physical APIC ID to %d ...",
1186 mp_ioapics[apic].mpc_apicid);
1188 reg_00.bits.ID = mp_ioapics[apic].mpc_apicid;
1189 spin_lock_irqsave(&ioapic_lock, flags);
1190 io_apic_write(apic, 0, reg_00.raw);
1191 spin_unlock_irqrestore(&ioapic_lock, flags);
1194 * Sanity check
1196 spin_lock_irqsave(&ioapic_lock, flags);
1197 reg_00.raw = io_apic_read(apic, 0);
1198 spin_unlock_irqrestore(&ioapic_lock, flags);
1199 if (reg_00.bits.ID != mp_ioapics[apic].mpc_apicid)
1200 printk("could not set ID!\n");
1201 else
1202 apic_printk(APIC_VERBOSE," ok.\n");
1207 * There is a nasty bug in some older SMP boards, their mptable lies
1208 * about the timer IRQ. We do the following to work around the situation:
1210 * - timer IRQ defaults to IO-APIC IRQ
1211 * - if this function detects that timer IRQs are defunct, then we fall
1212 * back to ISA timer IRQs
1214 static int __init timer_irq_works(void)
1216 unsigned long t1 = jiffies;
1218 local_irq_enable();
1219 /* Let ten ticks pass... */
1220 mdelay((10 * 1000) / HZ);
1223 * Expect a few ticks at least, to be sure some possible
1224 * glue logic does not lock up after one or two first
1225 * ticks in a non-ExtINT mode. Also the local APIC
1226 * might have cached one ExtINT interrupt. Finally, at
1227 * least one tick may be lost due to delays.
1230 /* jiffies wrap? */
1231 if (jiffies - t1 > 4)
1232 return 1;
1233 return 0;
1237 * In the SMP+IOAPIC case it might happen that there are an unspecified
1238 * number of pending IRQ events unhandled. These cases are very rare,
1239 * so we 'resend' these IRQs via IPIs, to the same CPU. It's much
1240 * better to do it this way as thus we do not have to be aware of
1241 * 'pending' interrupts in the IRQ path, except at this point.
1244 * Edge triggered needs to resend any interrupt
1245 * that was delayed but this is now handled in the device
1246 * independent code.
1250 * Starting up a edge-triggered IO-APIC interrupt is
1251 * nasty - we need to make sure that we get the edge.
1252 * If it is already asserted for some reason, we need
1253 * return 1 to indicate that is was pending.
1255 * This is not complete - we should be able to fake
1256 * an edge even if it isn't on the 8259A...
1259 static unsigned int startup_edge_ioapic_irq(unsigned int irq)
1261 int was_pending = 0;
1262 unsigned long flags;
1264 spin_lock_irqsave(&ioapic_lock, flags);
1265 if (irq < 16) {
1266 disable_8259A_irq(irq);
1267 if (i8259A_irq_pending(irq))
1268 was_pending = 1;
1270 __unmask_IO_APIC_irq(irq);
1271 spin_unlock_irqrestore(&ioapic_lock, flags);
1273 return was_pending;
1277 * Once we have recorded IRQ_PENDING already, we can mask the
1278 * interrupt for real. This prevents IRQ storms from unhandled
1279 * devices.
1281 static void ack_edge_ioapic_irq(unsigned int irq)
1283 if ((irq_desc[irq].status & (IRQ_PENDING | IRQ_DISABLED))
1284 == (IRQ_PENDING | IRQ_DISABLED))
1285 mask_IO_APIC_irq(irq);
1286 ack_APIC_irq();
1290 * Level triggered interrupts can just be masked,
1291 * and shutting down and starting up the interrupt
1292 * is the same as enabling and disabling them -- except
1293 * with a startup need to return a "was pending" value.
1295 * Level triggered interrupts are special because we
1296 * do not touch any IO-APIC register while handling
1297 * them. We ack the APIC in the end-IRQ handler, not
1298 * in the start-IRQ-handler. Protection against reentrance
1299 * from the same interrupt is still provided, both by the
1300 * generic IRQ layer and by the fact that an unacked local
1301 * APIC does not accept IRQs.
1303 static unsigned int startup_level_ioapic_irq (unsigned int irq)
1305 unmask_IO_APIC_irq(irq);
1307 return 0; /* don't check for pending */
1310 static void end_level_ioapic_irq (unsigned int irq)
1312 ack_APIC_irq();
1315 static void set_ioapic_affinity_irq(unsigned int irq, cpumask_t mask)
1317 unsigned long flags;
1318 unsigned int dest;
1320 dest = cpu_mask_to_apicid(mask);
1323 * Only the high 8 bits are valid.
1325 dest = SET_APIC_LOGICAL_ID(dest);
1327 spin_lock_irqsave(&ioapic_lock, flags);
1328 __DO_ACTION(1, = dest, )
1329 spin_unlock_irqrestore(&ioapic_lock, flags);
1332 #ifdef CONFIG_PCI_MSI
1333 static unsigned int startup_edge_ioapic_vector(unsigned int vector)
1335 int irq = vector_to_irq(vector);
1337 return startup_edge_ioapic_irq(irq);
1340 static void ack_edge_ioapic_vector(unsigned int vector)
1342 int irq = vector_to_irq(vector);
1344 ack_edge_ioapic_irq(irq);
1347 static unsigned int startup_level_ioapic_vector (unsigned int vector)
1349 int irq = vector_to_irq(vector);
1351 return startup_level_ioapic_irq (irq);
1354 static void end_level_ioapic_vector (unsigned int vector)
1356 int irq = vector_to_irq(vector);
1358 end_level_ioapic_irq(irq);
1361 static void mask_IO_APIC_vector (unsigned int vector)
1363 int irq = vector_to_irq(vector);
1365 mask_IO_APIC_irq(irq);
1368 static void unmask_IO_APIC_vector (unsigned int vector)
1370 int irq = vector_to_irq(vector);
1372 unmask_IO_APIC_irq(irq);
1375 static void set_ioapic_affinity_vector (unsigned int vector,
1376 cpumask_t cpu_mask)
1378 int irq = vector_to_irq(vector);
1380 set_ioapic_affinity_irq(irq, cpu_mask);
1382 #endif
1385 * Level and edge triggered IO-APIC interrupts need different handling,
1386 * so we use two separate IRQ descriptors. Edge triggered IRQs can be
1387 * handled with the level-triggered descriptor, but that one has slightly
1388 * more overhead. Level-triggered interrupts cannot be handled with the
1389 * edge-triggered handler, without risking IRQ storms and other ugly
1390 * races.
1393 static struct hw_interrupt_type ioapic_edge_type = {
1394 .typename = "IO-APIC-edge",
1395 .startup = startup_edge_ioapic,
1396 .shutdown = shutdown_edge_ioapic,
1397 .enable = enable_edge_ioapic,
1398 .disable = disable_edge_ioapic,
1399 .ack = ack_edge_ioapic,
1400 .end = end_edge_ioapic,
1401 .set_affinity = set_ioapic_affinity,
1404 static struct hw_interrupt_type ioapic_level_type = {
1405 .typename = "IO-APIC-level",
1406 .startup = startup_level_ioapic,
1407 .shutdown = shutdown_level_ioapic,
1408 .enable = enable_level_ioapic,
1409 .disable = disable_level_ioapic,
1410 .ack = mask_and_ack_level_ioapic,
1411 .end = end_level_ioapic,
1412 .set_affinity = set_ioapic_affinity,
1415 static inline void init_IO_APIC_traps(void)
1417 int irq;
1420 * NOTE! The local APIC isn't very good at handling
1421 * multiple interrupts at the same interrupt level.
1422 * As the interrupt level is determined by taking the
1423 * vector number and shifting that right by 4, we
1424 * want to spread these out a bit so that they don't
1425 * all fall in the same interrupt level.
1427 * Also, we've got to be careful not to trash gate
1428 * 0x80, because int 0x80 is hm, kind of importantish. ;)
1430 for (irq = 0; irq < NR_IRQS ; irq++) {
1431 int tmp = irq;
1432 if (use_pci_vector()) {
1433 if (!platform_legacy_irq(tmp))
1434 if ((tmp = vector_to_irq(tmp)) == -1)
1435 continue;
1437 if (IO_APIC_IRQ(tmp) && !IO_APIC_VECTOR(tmp)) {
1439 * Hmm.. We don't have an entry for this,
1440 * so default to an old-fashioned 8259
1441 * interrupt if we can..
1443 if (irq < 16)
1444 make_8259A_irq(irq);
1445 else
1446 /* Strange. Oh, well.. */
1447 irq_desc[irq].handler = &no_irq_type;
1452 static void enable_lapic_irq (unsigned int irq)
1454 unsigned long v;
1456 v = apic_read(APIC_LVT0);
1457 apic_write_around(APIC_LVT0, v & ~APIC_LVT_MASKED);
1460 static void disable_lapic_irq (unsigned int irq)
1462 unsigned long v;
1464 v = apic_read(APIC_LVT0);
1465 apic_write_around(APIC_LVT0, v | APIC_LVT_MASKED);
1468 static void ack_lapic_irq (unsigned int irq)
1470 ack_APIC_irq();
1473 static void end_lapic_irq (unsigned int i) { /* nothing */ }
1475 static struct hw_interrupt_type lapic_irq_type = {
1476 .typename = "local-APIC-edge",
1477 .startup = NULL, /* startup_irq() not used for IRQ0 */
1478 .shutdown = NULL, /* shutdown_irq() not used for IRQ0 */
1479 .enable = enable_lapic_irq,
1480 .disable = disable_lapic_irq,
1481 .ack = ack_lapic_irq,
1482 .end = end_lapic_irq,
1485 static void setup_nmi (void)
1488 * Dirty trick to enable the NMI watchdog ...
1489 * We put the 8259A master into AEOI mode and
1490 * unmask on all local APICs LVT0 as NMI.
1492 * The idea to use the 8259A in AEOI mode ('8259A Virtual Wire')
1493 * is from Maciej W. Rozycki - so we do not have to EOI from
1494 * the NMI handler or the timer interrupt.
1496 printk(KERN_INFO "activating NMI Watchdog ...");
1498 enable_NMI_through_LVT0(NULL);
1500 printk(" done.\n");
1504 * This looks a bit hackish but it's about the only one way of sending
1505 * a few INTA cycles to 8259As and any associated glue logic. ICR does
1506 * not support the ExtINT mode, unfortunately. We need to send these
1507 * cycles as some i82489DX-based boards have glue logic that keeps the
1508 * 8259A interrupt line asserted until INTA. --macro
1510 static inline void unlock_ExtINT_logic(void)
1512 int pin, i;
1513 struct IO_APIC_route_entry entry0, entry1;
1514 unsigned char save_control, save_freq_select;
1515 unsigned long flags;
1517 pin = find_isa_irq_pin(8, mp_INT);
1518 if (pin == -1)
1519 return;
1521 spin_lock_irqsave(&ioapic_lock, flags);
1522 *(((int *)&entry0) + 1) = io_apic_read(0, 0x11 + 2 * pin);
1523 *(((int *)&entry0) + 0) = io_apic_read(0, 0x10 + 2 * pin);
1524 spin_unlock_irqrestore(&ioapic_lock, flags);
1525 clear_IO_APIC_pin(0, pin);
1527 memset(&entry1, 0, sizeof(entry1));
1529 entry1.dest_mode = 0; /* physical delivery */
1530 entry1.mask = 0; /* unmask IRQ now */
1531 entry1.dest.physical.physical_dest = hard_smp_processor_id();
1532 entry1.delivery_mode = dest_ExtINT;
1533 entry1.polarity = entry0.polarity;
1534 entry1.trigger = 0;
1535 entry1.vector = 0;
1537 spin_lock_irqsave(&ioapic_lock, flags);
1538 io_apic_write(0, 0x11 + 2 * pin, *(((int *)&entry1) + 1));
1539 io_apic_write(0, 0x10 + 2 * pin, *(((int *)&entry1) + 0));
1540 spin_unlock_irqrestore(&ioapic_lock, flags);
1542 save_control = CMOS_READ(RTC_CONTROL);
1543 save_freq_select = CMOS_READ(RTC_FREQ_SELECT);
1544 CMOS_WRITE((save_freq_select & ~RTC_RATE_SELECT) | 0x6,
1545 RTC_FREQ_SELECT);
1546 CMOS_WRITE(save_control | RTC_PIE, RTC_CONTROL);
1548 i = 100;
1549 while (i-- > 0) {
1550 mdelay(10);
1551 if ((CMOS_READ(RTC_INTR_FLAGS) & RTC_PF) == RTC_PF)
1552 i -= 10;
1555 CMOS_WRITE(save_control, RTC_CONTROL);
1556 CMOS_WRITE(save_freq_select, RTC_FREQ_SELECT);
1557 clear_IO_APIC_pin(0, pin);
1559 spin_lock_irqsave(&ioapic_lock, flags);
1560 io_apic_write(0, 0x11 + 2 * pin, *(((int *)&entry0) + 1));
1561 io_apic_write(0, 0x10 + 2 * pin, *(((int *)&entry0) + 0));
1562 spin_unlock_irqrestore(&ioapic_lock, flags);
1566 * This code may look a bit paranoid, but it's supposed to cooperate with
1567 * a wide range of boards and BIOS bugs. Fortunately only the timer IRQ
1568 * is so screwy. Thanks to Brian Perkins for testing/hacking this beast
1569 * fanatically on his truly buggy board.
1571 static inline void check_timer(void)
1573 int pin1, pin2;
1574 int vector;
1577 * get/set the timer IRQ vector:
1579 disable_8259A_irq(0);
1580 vector = assign_irq_vector(0);
1581 set_intr_gate(vector, interrupt[0]);
1584 * Subtle, code in do_timer_interrupt() expects an AEOI
1585 * mode for the 8259A whenever interrupts are routed
1586 * through I/O APICs. Also IRQ0 has to be enabled in
1587 * the 8259A which implies the virtual wire has to be
1588 * disabled in the local APIC.
1590 apic_write_around(APIC_LVT0, APIC_LVT_MASKED | APIC_DM_EXTINT);
1591 init_8259A(1);
1592 enable_8259A_irq(0);
1594 pin1 = find_isa_irq_pin(0, mp_INT);
1595 pin2 = find_isa_irq_pin(0, mp_ExtINT);
1597 apic_printk(APIC_VERBOSE,KERN_INFO "..TIMER: vector=0x%02X pin1=%d pin2=%d\n", vector, pin1, pin2);
1599 if (pin1 != -1) {
1601 * Ok, does IRQ0 through the IOAPIC work?
1603 unmask_IO_APIC_irq(0);
1604 if (timer_irq_works()) {
1605 nmi_watchdog_default();
1606 if (nmi_watchdog == NMI_IO_APIC) {
1607 disable_8259A_irq(0);
1608 setup_nmi();
1609 enable_8259A_irq(0);
1611 return;
1613 clear_IO_APIC_pin(0, pin1);
1614 apic_printk(APIC_QUIET,KERN_ERR "..MP-BIOS bug: 8254 timer not connected to IO-APIC\n");
1617 apic_printk(APIC_VERBOSE,KERN_INFO "...trying to set up timer (IRQ0) through the 8259A ... ");
1618 if (pin2 != -1) {
1619 apic_printk(APIC_VERBOSE,"\n..... (found pin %d) ...", pin2);
1621 * legacy devices should be connected to IO APIC #0
1623 setup_ExtINT_IRQ0_pin(pin2, vector);
1624 if (timer_irq_works()) {
1625 printk("works.\n");
1626 nmi_watchdog_default();
1627 if (nmi_watchdog == NMI_IO_APIC) {
1628 setup_nmi();
1630 return;
1633 * Cleanup, just in case ...
1635 clear_IO_APIC_pin(0, pin2);
1637 printk(" failed.\n");
1639 if (nmi_watchdog) {
1640 printk(KERN_WARNING "timer doesn't work through the IO-APIC - disabling NMI Watchdog!\n");
1641 nmi_watchdog = 0;
1644 apic_printk(APIC_VERBOSE, KERN_INFO "...trying to set up timer as Virtual Wire IRQ...");
1646 disable_8259A_irq(0);
1647 irq_desc[0].handler = &lapic_irq_type;
1648 apic_write_around(APIC_LVT0, APIC_DM_FIXED | vector); /* Fixed mode */
1649 enable_8259A_irq(0);
1651 if (timer_irq_works()) {
1652 apic_printk(APIC_QUIET, " works.\n");
1653 return;
1655 apic_write_around(APIC_LVT0, APIC_LVT_MASKED | APIC_DM_FIXED | vector);
1656 apic_printk(APIC_VERBOSE," failed.\n");
1658 apic_printk(APIC_VERBOSE, KERN_INFO "...trying to set up timer as ExtINT IRQ...");
1660 init_8259A(0);
1661 make_8259A_irq(0);
1662 apic_write_around(APIC_LVT0, APIC_DM_EXTINT);
1664 unlock_ExtINT_logic();
1666 if (timer_irq_works()) {
1667 apic_printk(APIC_VERBOSE," works.\n");
1668 return;
1670 apic_printk(APIC_VERBOSE," failed :(.\n");
1671 panic("IO-APIC + timer doesn't work! Try using the 'noapic' kernel parameter\n");
1676 * IRQ's that are handled by the PIC in the MPS IOAPIC case.
1677 * - IRQ2 is the cascade IRQ, and cannot be a io-apic IRQ.
1678 * Linux doesn't really care, as it's not actually used
1679 * for any interrupt handling anyway.
1681 #define PIC_IRQS (1<<2)
1683 void __init setup_IO_APIC(void)
1685 enable_IO_APIC();
1687 if (acpi_ioapic)
1688 io_apic_irqs = ~0; /* all IRQs go through IOAPIC */
1689 else
1690 io_apic_irqs = ~PIC_IRQS;
1692 apic_printk(APIC_VERBOSE, "ENABLING IO-APIC IRQs\n");
1695 * Set up the IO-APIC IRQ routing table.
1697 if (!acpi_ioapic)
1698 setup_ioapic_ids_from_mpc();
1699 sync_Arb_IDs();
1700 setup_IO_APIC_irqs();
1701 init_IO_APIC_traps();
1702 check_timer();
1703 if (!acpi_ioapic)
1704 print_IO_APIC();
1707 struct sysfs_ioapic_data {
1708 struct sys_device dev;
1709 struct IO_APIC_route_entry entry[0];
1711 static struct sysfs_ioapic_data * mp_ioapic_data[MAX_IO_APICS];
1713 static int ioapic_suspend(struct sys_device *dev, pm_message_t state)
1715 struct IO_APIC_route_entry *entry;
1716 struct sysfs_ioapic_data *data;
1717 unsigned long flags;
1718 int i;
1720 data = container_of(dev, struct sysfs_ioapic_data, dev);
1721 entry = data->entry;
1722 spin_lock_irqsave(&ioapic_lock, flags);
1723 for (i = 0; i < nr_ioapic_registers[dev->id]; i ++, entry ++ ) {
1724 *(((int *)entry) + 1) = io_apic_read(dev->id, 0x11 + 2 * i);
1725 *(((int *)entry) + 0) = io_apic_read(dev->id, 0x10 + 2 * i);
1727 spin_unlock_irqrestore(&ioapic_lock, flags);
1729 return 0;
1732 static int ioapic_resume(struct sys_device *dev)
1734 struct IO_APIC_route_entry *entry;
1735 struct sysfs_ioapic_data *data;
1736 unsigned long flags;
1737 union IO_APIC_reg_00 reg_00;
1738 int i;
1740 data = container_of(dev, struct sysfs_ioapic_data, dev);
1741 entry = data->entry;
1743 spin_lock_irqsave(&ioapic_lock, flags);
1744 reg_00.raw = io_apic_read(dev->id, 0);
1745 if (reg_00.bits.ID != mp_ioapics[dev->id].mpc_apicid) {
1746 reg_00.bits.ID = mp_ioapics[dev->id].mpc_apicid;
1747 io_apic_write(dev->id, 0, reg_00.raw);
1749 for (i = 0; i < nr_ioapic_registers[dev->id]; i ++, entry ++ ) {
1750 io_apic_write(dev->id, 0x11+2*i, *(((int *)entry)+1));
1751 io_apic_write(dev->id, 0x10+2*i, *(((int *)entry)+0));
1753 spin_unlock_irqrestore(&ioapic_lock, flags);
1755 return 0;
1758 static struct sysdev_class ioapic_sysdev_class = {
1759 set_kset_name("ioapic"),
1760 .suspend = ioapic_suspend,
1761 .resume = ioapic_resume,
1764 static int __init ioapic_init_sysfs(void)
1766 struct sys_device * dev;
1767 int i, size, error = 0;
1769 error = sysdev_class_register(&ioapic_sysdev_class);
1770 if (error)
1771 return error;
1773 for (i = 0; i < nr_ioapics; i++ ) {
1774 size = sizeof(struct sys_device) + nr_ioapic_registers[i]
1775 * sizeof(struct IO_APIC_route_entry);
1776 mp_ioapic_data[i] = kmalloc(size, GFP_KERNEL);
1777 if (!mp_ioapic_data[i]) {
1778 printk(KERN_ERR "Can't suspend/resume IOAPIC %d\n", i);
1779 continue;
1781 memset(mp_ioapic_data[i], 0, size);
1782 dev = &mp_ioapic_data[i]->dev;
1783 dev->id = i;
1784 dev->cls = &ioapic_sysdev_class;
1785 error = sysdev_register(dev);
1786 if (error) {
1787 kfree(mp_ioapic_data[i]);
1788 mp_ioapic_data[i] = NULL;
1789 printk(KERN_ERR "Can't suspend/resume IOAPIC %d\n", i);
1790 continue;
1794 return 0;
1797 device_initcall(ioapic_init_sysfs);
1799 /* --------------------------------------------------------------------------
1800 ACPI-based IOAPIC Configuration
1801 -------------------------------------------------------------------------- */
1803 #ifdef CONFIG_ACPI_BOOT
1805 #define IO_APIC_MAX_ID 0xFE
1807 int __init io_apic_get_version (int ioapic)
1809 union IO_APIC_reg_01 reg_01;
1810 unsigned long flags;
1812 spin_lock_irqsave(&ioapic_lock, flags);
1813 reg_01.raw = io_apic_read(ioapic, 1);
1814 spin_unlock_irqrestore(&ioapic_lock, flags);
1816 return reg_01.bits.version;
1820 int __init io_apic_get_redir_entries (int ioapic)
1822 union IO_APIC_reg_01 reg_01;
1823 unsigned long flags;
1825 spin_lock_irqsave(&ioapic_lock, flags);
1826 reg_01.raw = io_apic_read(ioapic, 1);
1827 spin_unlock_irqrestore(&ioapic_lock, flags);
1829 return reg_01.bits.entries;
1833 int io_apic_set_pci_routing (int ioapic, int pin, int irq, int edge_level, int active_high_low)
1835 struct IO_APIC_route_entry entry;
1836 unsigned long flags;
1838 if (!IO_APIC_IRQ(irq)) {
1839 apic_printk(APIC_QUIET,KERN_ERR "IOAPIC[%d]: Invalid reference to IRQ 0\n",
1840 ioapic);
1841 return -EINVAL;
1845 * Generate a PCI IRQ routing entry and program the IOAPIC accordingly.
1846 * Note that we mask (disable) IRQs now -- these get enabled when the
1847 * corresponding device driver registers for this IRQ.
1850 memset(&entry,0,sizeof(entry));
1852 entry.delivery_mode = INT_DELIVERY_MODE;
1853 entry.dest_mode = INT_DEST_MODE;
1854 entry.dest.logical.logical_dest = cpu_mask_to_apicid(TARGET_CPUS);
1855 entry.trigger = edge_level;
1856 entry.polarity = active_high_low;
1857 entry.mask = 1; /* Disabled (masked) */
1860 * IRQs < 16 are already in the irq_2_pin[] map
1862 if (irq >= 16)
1863 add_pin_to_irq(irq, ioapic, pin);
1865 entry.vector = assign_irq_vector(irq);
1867 apic_printk(APIC_VERBOSE,KERN_DEBUG "IOAPIC[%d]: Set PCI routing entry (%d-%d -> 0x%x -> "
1868 "IRQ %d Mode:%i Active:%i)\n", ioapic,
1869 mp_ioapics[ioapic].mpc_apicid, pin, entry.vector, irq,
1870 edge_level, active_high_low);
1872 ioapic_register_intr(irq, entry.vector, edge_level);
1874 if (!ioapic && (irq < 16))
1875 disable_8259A_irq(irq);
1877 spin_lock_irqsave(&ioapic_lock, flags);
1878 io_apic_write(ioapic, 0x11+2*pin, *(((int *)&entry)+1));
1879 io_apic_write(ioapic, 0x10+2*pin, *(((int *)&entry)+0));
1880 spin_unlock_irqrestore(&ioapic_lock, flags);
1882 return 0;
1885 #endif /*CONFIG_ACPI_BOOT*/
1889 * This function currently is only a helper for the i386 smp boot process where
1890 * we need to reprogram the ioredtbls to cater for the cpus which have come online
1891 * so mask in all cases should simply be TARGET_CPUS
1893 void __init setup_ioapic_dest(void)
1895 int pin, ioapic, irq, irq_entry;
1897 if (skip_ioapic_setup == 1)
1898 return;
1900 for (ioapic = 0; ioapic < nr_ioapics; ioapic++) {
1901 for (pin = 0; pin < nr_ioapic_registers[ioapic]; pin++) {
1902 irq_entry = find_irq_entry(ioapic, pin, mp_INT);
1903 if (irq_entry == -1)
1904 continue;
1905 irq = pin_2_irq(irq_entry, ioapic, pin);
1906 set_ioapic_affinity_irq(irq, TARGET_CPUS);