x86: make 32bit use irq_cfg_alloc, etc
[firewire-audio.git] / arch / x86 / kernel / io_apic_32.c
blob5a83d7f5b14717cd7b54276f1589cd47010e70aa
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/interrupt.h>
25 #include <linux/init.h>
26 #include <linux/delay.h>
27 #include <linux/sched.h>
28 #include <linux/bootmem.h>
29 #include <linux/mc146818rtc.h>
30 #include <linux/compiler.h>
31 #include <linux/acpi.h>
32 #include <linux/module.h>
33 #include <linux/sysdev.h>
34 #include <linux/pci.h>
35 #include <linux/msi.h>
36 #include <linux/htirq.h>
37 #include <linux/freezer.h>
38 #include <linux/kthread.h>
39 #include <linux/jiffies.h> /* time_after() */
41 #include <asm/io.h>
42 #include <asm/smp.h>
43 #include <asm/desc.h>
44 #include <asm/timer.h>
45 #include <asm/i8259.h>
46 #include <asm/nmi.h>
47 #include <asm/msidef.h>
48 #include <asm/hypertransport.h>
49 #include <asm/setup.h>
51 #include <mach_apic.h>
52 #include <mach_apicdef.h>
54 #define __apicdebuginit(type) static type __init
56 int (*ioapic_renumber_irq)(int ioapic, int irq);
57 atomic_t irq_mis_count;
59 /* Where if anywhere is the i8259 connect in external int mode */
60 static struct { int pin, apic; } ioapic_i8259 = { -1, -1 };
62 static DEFINE_SPINLOCK(ioapic_lock);
63 DEFINE_SPINLOCK(vector_lock);
65 int timer_through_8259 __initdata;
68 * Is the SiS APIC rmw bug present ?
69 * -1 = don't know, 0 = no, 1 = yes
71 int sis_apic_bug = -1;
73 int first_free_entry;
75 * # of IRQ routing registers
77 int nr_ioapic_registers[MAX_IO_APICS];
79 /* I/O APIC entries */
80 struct mp_config_ioapic mp_ioapics[MAX_IO_APICS];
81 int nr_ioapics;
83 /* MP IRQ source entries */
84 struct mp_config_intsrc mp_irqs[MAX_IRQ_SOURCES];
86 /* # of MP IRQ source entries */
87 int mp_irq_entries;
89 #if defined (CONFIG_MCA) || defined (CONFIG_EISA)
90 int mp_bus_id_to_type[MAX_MP_BUSSES];
91 #endif
93 DECLARE_BITMAP(mp_bus_not_pci, MAX_MP_BUSSES);
95 static int disable_timer_pin_1 __initdata;
97 struct irq_cfg;
99 struct irq_cfg {
100 unsigned int irq;
101 struct irq_cfg *next;
102 u8 vector;
106 /* irq_cfg is indexed by the sum of all RTEs in all I/O APICs. */
107 static struct irq_cfg irq_cfg_legacy[] __initdata = {
108 [0] = { .irq = 0, .vector = IRQ0_VECTOR, },
109 [1] = { .irq = 1, .vector = IRQ1_VECTOR, },
110 [2] = { .irq = 2, .vector = IRQ2_VECTOR, },
111 [3] = { .irq = 3, .vector = IRQ3_VECTOR, },
112 [4] = { .irq = 4, .vector = IRQ4_VECTOR, },
113 [5] = { .irq = 5, .vector = IRQ5_VECTOR, },
114 [6] = { .irq = 6, .vector = IRQ6_VECTOR, },
115 [7] = { .irq = 7, .vector = IRQ7_VECTOR, },
116 [8] = { .irq = 8, .vector = IRQ8_VECTOR, },
117 [9] = { .irq = 9, .vector = IRQ9_VECTOR, },
118 [10] = { .irq = 10, .vector = IRQ10_VECTOR, },
119 [11] = { .irq = 11, .vector = IRQ11_VECTOR, },
120 [12] = { .irq = 12, .vector = IRQ12_VECTOR, },
121 [13] = { .irq = 13, .vector = IRQ13_VECTOR, },
122 [14] = { .irq = 14, .vector = IRQ14_VECTOR, },
123 [15] = { .irq = 15, .vector = IRQ15_VECTOR, },
126 static struct irq_cfg irq_cfg_init = { .irq = -1U, };
127 /* need to be biger than size of irq_cfg_legacy */
128 static int nr_irq_cfg = 32;
130 static int __init parse_nr_irq_cfg(char *arg)
132 if (arg) {
133 nr_irq_cfg = simple_strtoul(arg, NULL, 0);
134 if (nr_irq_cfg < 32)
135 nr_irq_cfg = 32;
137 return 0;
140 early_param("nr_irq_cfg", parse_nr_irq_cfg);
142 static void init_one_irq_cfg(struct irq_cfg *cfg)
144 memcpy(cfg, &irq_cfg_init, sizeof(struct irq_cfg));
147 static struct irq_cfg *irq_cfgx;
148 static struct irq_cfg *irq_cfgx_free;
149 static void __init init_work(void *data)
151 struct dyn_array *da = data;
152 struct irq_cfg *cfg;
153 int legacy_count;
154 int i;
156 cfg = *da->name;
158 memcpy(cfg, irq_cfg_legacy, sizeof(irq_cfg_legacy));
160 legacy_count = sizeof(irq_cfg_legacy)/sizeof(irq_cfg_legacy[0]);
161 for (i = legacy_count; i < *da->nr; i++)
162 init_one_irq_cfg(&cfg[i]);
164 for (i = 1; i < *da->nr; i++)
165 cfg[i-1].next = &cfg[i];
167 irq_cfgx_free = &irq_cfgx[legacy_count];
168 irq_cfgx[legacy_count - 1].next = NULL;
171 #define for_each_irq_cfg(cfg) \
172 for (cfg = irq_cfgx; cfg; cfg = cfg->next)
174 DEFINE_DYN_ARRAY(irq_cfgx, sizeof(struct irq_cfg), nr_irq_cfg, PAGE_SIZE, init_work);
176 static struct irq_cfg *irq_cfg(unsigned int irq)
178 struct irq_cfg *cfg;
180 cfg = irq_cfgx;
181 while (cfg) {
182 if (cfg->irq == irq)
183 return cfg;
185 cfg = cfg->next;
188 return NULL;
191 static struct irq_cfg *irq_cfg_alloc(unsigned int irq)
193 struct irq_cfg *cfg, *cfg_pri;
194 int i;
195 int count = 0;
197 cfg_pri = cfg = irq_cfgx;
198 while (cfg) {
199 if (cfg->irq == irq)
200 return cfg;
202 cfg_pri = cfg;
203 cfg = cfg->next;
204 count++;
207 if (!irq_cfgx_free) {
208 unsigned long phys;
209 unsigned long total_bytes;
211 * we run out of pre-allocate ones, allocate more
213 printk(KERN_DEBUG "try to get more irq_cfg %d\n", nr_irq_cfg);
215 total_bytes = sizeof(struct irq_cfg) * nr_irq_cfg;
216 if (after_bootmem)
217 cfg = kzalloc(total_bytes, GFP_ATOMIC);
218 else
219 cfg = __alloc_bootmem_nopanic(total_bytes, PAGE_SIZE, 0);
221 if (!cfg)
222 panic("please boot with nr_irq_cfg= %d\n", count * 2);
224 phys = __pa(cfg);
225 printk(KERN_DEBUG "irq_irq ==> [%#lx - %#lx]\n", phys, phys + total_bytes);
227 for (i = 0; i < nr_irq_cfg; i++)
228 init_one_irq_cfg(&cfg[i]);
230 for (i = 1; i < nr_irq_cfg; i++)
231 cfg[i-1].next = &cfg[i];
233 irq_cfgx_free = cfg;
236 cfg = irq_cfgx_free;
237 irq_cfgx_free = irq_cfgx_free->next;
238 cfg->next = NULL;
239 if (cfg_pri)
240 cfg_pri->next = cfg;
241 else
242 irq_cfgx = cfg;
243 cfg->irq = irq;
244 printk(KERN_DEBUG "found new irq_cfg for irq %d\n", cfg->irq);
246 #ifdef CONFIG_HAVE_SPARSE_IRQ_DEBUG
248 /* dump the results */
249 struct irq_cfg *cfg;
250 unsigned long phys;
251 unsigned long bytes = sizeof(struct irq_cfg);
253 printk(KERN_DEBUG "=========================== %d\n", irq);
254 printk(KERN_DEBUG "irq_cfg dump after get that for %d\n", irq);
255 for_each_irq_cfg(cfg) {
256 phys = __pa(cfg);
257 printk(KERN_DEBUG "irq_cfg %d ==> [%#lx - %#lx]\n", cfg->irq, phys, phys + bytes);
259 printk(KERN_DEBUG "===========================\n");
261 #endif
262 return cfg;
266 * Rough estimation of how many shared IRQs there are, can
267 * be changed anytime.
269 int pin_map_size;
272 * This is performance-critical, we want to do it O(1)
274 * the indexing order of this array favors 1:1 mappings
275 * between pins and IRQs.
278 static struct irq_pin_list {
279 int apic, pin, next;
280 } *irq_2_pin;
282 DEFINE_DYN_ARRAY(irq_2_pin, sizeof(struct irq_pin_list), pin_map_size, 16, NULL);
284 struct io_apic {
285 unsigned int index;
286 unsigned int unused[3];
287 unsigned int data;
290 static __attribute_const__ struct io_apic __iomem *io_apic_base(int idx)
292 return (void __iomem *) __fix_to_virt(FIX_IO_APIC_BASE_0 + idx)
293 + (mp_ioapics[idx].mp_apicaddr & ~PAGE_MASK);
296 static inline unsigned int io_apic_read(unsigned int apic, unsigned int reg)
298 struct io_apic __iomem *io_apic = io_apic_base(apic);
299 writel(reg, &io_apic->index);
300 return readl(&io_apic->data);
303 static inline void io_apic_write(unsigned int apic, unsigned int reg, unsigned int value)
305 struct io_apic __iomem *io_apic = io_apic_base(apic);
306 writel(reg, &io_apic->index);
307 writel(value, &io_apic->data);
311 * Re-write a value: to be used for read-modify-write
312 * cycles where the read already set up the index register.
314 * Older SiS APIC requires we rewrite the index register
316 static inline void io_apic_modify(unsigned int apic, unsigned int reg, unsigned int value)
318 volatile struct io_apic __iomem *io_apic = io_apic_base(apic);
319 if (sis_apic_bug)
320 writel(reg, &io_apic->index);
321 writel(value, &io_apic->data);
324 union entry_union {
325 struct { u32 w1, w2; };
326 struct IO_APIC_route_entry entry;
329 static struct IO_APIC_route_entry ioapic_read_entry(int apic, int pin)
331 union entry_union eu;
332 unsigned long flags;
333 spin_lock_irqsave(&ioapic_lock, flags);
334 eu.w1 = io_apic_read(apic, 0x10 + 2 * pin);
335 eu.w2 = io_apic_read(apic, 0x11 + 2 * pin);
336 spin_unlock_irqrestore(&ioapic_lock, flags);
337 return eu.entry;
341 * When we write a new IO APIC routing entry, we need to write the high
342 * word first! If the mask bit in the low word is clear, we will enable
343 * the interrupt, and we need to make sure the entry is fully populated
344 * before that happens.
346 static void
347 __ioapic_write_entry(int apic, int pin, struct IO_APIC_route_entry e)
349 union entry_union eu;
350 eu.entry = e;
351 io_apic_write(apic, 0x11 + 2*pin, eu.w2);
352 io_apic_write(apic, 0x10 + 2*pin, eu.w1);
355 static void ioapic_write_entry(int apic, int pin, struct IO_APIC_route_entry e)
357 unsigned long flags;
358 spin_lock_irqsave(&ioapic_lock, flags);
359 __ioapic_write_entry(apic, pin, e);
360 spin_unlock_irqrestore(&ioapic_lock, flags);
364 * When we mask an IO APIC routing entry, we need to write the low
365 * word first, in order to set the mask bit before we change the
366 * high bits!
368 static void ioapic_mask_entry(int apic, int pin)
370 unsigned long flags;
371 union entry_union eu = { .entry.mask = 1 };
373 spin_lock_irqsave(&ioapic_lock, flags);
374 io_apic_write(apic, 0x10 + 2*pin, eu.w1);
375 io_apic_write(apic, 0x11 + 2*pin, eu.w2);
376 spin_unlock_irqrestore(&ioapic_lock, flags);
380 * The common case is 1:1 IRQ<->pin mappings. Sometimes there are
381 * shared ISA-space IRQs, so we have to support them. We are super
382 * fast in the common case, and fast for shared ISA-space IRQs.
384 static void add_pin_to_irq(unsigned int irq, int apic, int pin)
386 struct irq_pin_list *entry = irq_2_pin + irq;
388 irq_cfg_alloc(irq);
389 while (entry->next)
390 entry = irq_2_pin + entry->next;
392 if (entry->pin != -1) {
393 entry->next = first_free_entry;
394 entry = irq_2_pin + entry->next;
395 if (++first_free_entry >= pin_map_size)
396 panic("io_apic.c: whoops");
398 entry->apic = apic;
399 entry->pin = pin;
403 * Reroute an IRQ to a different pin.
405 static void __init replace_pin_at_irq(unsigned int irq,
406 int oldapic, int oldpin,
407 int newapic, int newpin)
409 struct irq_pin_list *entry = irq_2_pin + irq;
411 while (1) {
412 if (entry->apic == oldapic && entry->pin == oldpin) {
413 entry->apic = newapic;
414 entry->pin = newpin;
416 if (!entry->next)
417 break;
418 entry = irq_2_pin + entry->next;
422 static void __modify_IO_APIC_irq(unsigned int irq, unsigned long enable, unsigned long disable)
424 struct irq_pin_list *entry = irq_2_pin + irq;
425 unsigned int pin, reg;
427 for (;;) {
428 pin = entry->pin;
429 if (pin == -1)
430 break;
431 reg = io_apic_read(entry->apic, 0x10 + pin*2);
432 reg &= ~disable;
433 reg |= enable;
434 io_apic_modify(entry->apic, 0x10 + pin*2, reg);
435 if (!entry->next)
436 break;
437 entry = irq_2_pin + entry->next;
441 /* mask = 1 */
442 static void __mask_IO_APIC_irq(unsigned int irq)
444 __modify_IO_APIC_irq(irq, IO_APIC_REDIR_MASKED, 0);
447 /* mask = 0 */
448 static void __unmask_IO_APIC_irq(unsigned int irq)
450 __modify_IO_APIC_irq(irq, 0, IO_APIC_REDIR_MASKED);
453 /* mask = 1, trigger = 0 */
454 static void __mask_and_edge_IO_APIC_irq(unsigned int irq)
456 __modify_IO_APIC_irq(irq, IO_APIC_REDIR_MASKED,
457 IO_APIC_REDIR_LEVEL_TRIGGER);
460 /* mask = 0, trigger = 1 */
461 static void __unmask_and_level_IO_APIC_irq(unsigned int irq)
463 __modify_IO_APIC_irq(irq, IO_APIC_REDIR_LEVEL_TRIGGER,
464 IO_APIC_REDIR_MASKED);
467 static void mask_IO_APIC_irq(unsigned int irq)
469 unsigned long flags;
471 spin_lock_irqsave(&ioapic_lock, flags);
472 __mask_IO_APIC_irq(irq);
473 spin_unlock_irqrestore(&ioapic_lock, flags);
476 static void unmask_IO_APIC_irq(unsigned int irq)
478 unsigned long flags;
480 spin_lock_irqsave(&ioapic_lock, flags);
481 __unmask_IO_APIC_irq(irq);
482 spin_unlock_irqrestore(&ioapic_lock, flags);
485 static void clear_IO_APIC_pin(unsigned int apic, unsigned int pin)
487 struct IO_APIC_route_entry entry;
489 /* Check delivery_mode to be sure we're not clearing an SMI pin */
490 entry = ioapic_read_entry(apic, pin);
491 if (entry.delivery_mode == dest_SMI)
492 return;
495 * Disable it in the IO-APIC irq-routing table:
497 ioapic_mask_entry(apic, pin);
500 static void clear_IO_APIC(void)
502 int apic, pin;
504 for (apic = 0; apic < nr_ioapics; apic++)
505 for (pin = 0; pin < nr_ioapic_registers[apic]; pin++)
506 clear_IO_APIC_pin(apic, pin);
509 #ifdef CONFIG_SMP
510 static void set_ioapic_affinity_irq(unsigned int irq, cpumask_t cpumask)
512 unsigned long flags;
513 int pin;
514 struct irq_pin_list *entry = irq_2_pin + irq;
515 unsigned int apicid_value;
516 cpumask_t tmp;
517 struct irq_desc *desc;
519 cpus_and(tmp, cpumask, cpu_online_map);
520 if (cpus_empty(tmp))
521 tmp = TARGET_CPUS;
523 cpus_and(cpumask, tmp, CPU_MASK_ALL);
525 apicid_value = cpu_mask_to_apicid(cpumask);
526 /* Prepare to do the io_apic_write */
527 apicid_value = apicid_value << 24;
528 spin_lock_irqsave(&ioapic_lock, flags);
529 for (;;) {
530 pin = entry->pin;
531 if (pin == -1)
532 break;
533 io_apic_write(entry->apic, 0x10 + 1 + pin*2, apicid_value);
534 if (!entry->next)
535 break;
536 entry = irq_2_pin + entry->next;
538 desc = irq_to_desc(irq);
539 desc->affinity = cpumask;
540 spin_unlock_irqrestore(&ioapic_lock, flags);
543 #endif /* CONFIG_SMP */
545 #ifndef CONFIG_SMP
546 void send_IPI_self(int vector)
548 unsigned int cfg;
551 * Wait for idle.
553 apic_wait_icr_idle();
554 cfg = APIC_DM_FIXED | APIC_DEST_SELF | vector | APIC_DEST_LOGICAL;
556 * Send the IPI. The write to APIC_ICR fires this off.
558 apic_write(APIC_ICR, cfg);
560 #endif /* !CONFIG_SMP */
564 * support for broken MP BIOSs, enables hand-redirection of PIRQ0-7 to
565 * specific CPU-side IRQs.
568 #define MAX_PIRQS 8
569 static int pirq_entries [MAX_PIRQS];
570 static int pirqs_enabled;
571 int skip_ioapic_setup;
573 static int __init ioapic_pirq_setup(char *str)
575 int i, max;
576 int ints[MAX_PIRQS+1];
578 get_options(str, ARRAY_SIZE(ints), ints);
580 for (i = 0; i < MAX_PIRQS; i++)
581 pirq_entries[i] = -1;
583 pirqs_enabled = 1;
584 apic_printk(APIC_VERBOSE, KERN_INFO
585 "PIRQ redirection, working around broken MP-BIOS.\n");
586 max = MAX_PIRQS;
587 if (ints[0] < MAX_PIRQS)
588 max = ints[0];
590 for (i = 0; i < max; i++) {
591 apic_printk(APIC_VERBOSE, KERN_DEBUG
592 "... PIRQ%d -> IRQ %d\n", i, ints[i+1]);
594 * PIRQs are mapped upside down, usually.
596 pirq_entries[MAX_PIRQS-i-1] = ints[i+1];
598 return 1;
601 __setup("pirq=", ioapic_pirq_setup);
604 * Find the IRQ entry number of a certain pin.
606 static int find_irq_entry(int apic, int pin, int type)
608 int i;
610 for (i = 0; i < mp_irq_entries; i++)
611 if (mp_irqs[i].mp_irqtype == type &&
612 (mp_irqs[i].mp_dstapic == mp_ioapics[apic].mp_apicid ||
613 mp_irqs[i].mp_dstapic == MP_APIC_ALL) &&
614 mp_irqs[i].mp_dstirq == pin)
615 return i;
617 return -1;
621 * Find the pin to which IRQ[irq] (ISA) is connected
623 static int __init find_isa_irq_pin(int irq, int type)
625 int i;
627 for (i = 0; i < mp_irq_entries; i++) {
628 int lbus = mp_irqs[i].mp_srcbus;
630 if (test_bit(lbus, mp_bus_not_pci) &&
631 (mp_irqs[i].mp_irqtype == type) &&
632 (mp_irqs[i].mp_srcbusirq == irq))
634 return mp_irqs[i].mp_dstirq;
636 return -1;
639 static int __init find_isa_irq_apic(int irq, int type)
641 int i;
643 for (i = 0; i < mp_irq_entries; i++) {
644 int lbus = mp_irqs[i].mp_srcbus;
646 if (test_bit(lbus, mp_bus_not_pci) &&
647 (mp_irqs[i].mp_irqtype == type) &&
648 (mp_irqs[i].mp_srcbusirq == irq))
649 break;
651 if (i < mp_irq_entries) {
652 int apic;
653 for (apic = 0; apic < nr_ioapics; apic++) {
654 if (mp_ioapics[apic].mp_apicid == mp_irqs[i].mp_dstapic)
655 return apic;
659 return -1;
663 * Find a specific PCI IRQ entry.
664 * Not an __init, possibly needed by modules
666 static int pin_2_irq(int idx, int apic, int pin);
668 int IO_APIC_get_PCI_irq_vector(int bus, int slot, int pin)
670 int apic, i, best_guess = -1;
672 apic_printk(APIC_DEBUG, "querying PCI -> IRQ mapping bus:%d, "
673 "slot:%d, pin:%d.\n", bus, slot, pin);
674 if (test_bit(bus, mp_bus_not_pci)) {
675 printk(KERN_WARNING "PCI BIOS passed nonexistent PCI bus %d!\n", bus);
676 return -1;
678 for (i = 0; i < mp_irq_entries; i++) {
679 int lbus = mp_irqs[i].mp_srcbus;
681 for (apic = 0; apic < nr_ioapics; apic++)
682 if (mp_ioapics[apic].mp_apicid == mp_irqs[i].mp_dstapic ||
683 mp_irqs[i].mp_dstapic == MP_APIC_ALL)
684 break;
686 if (!test_bit(lbus, mp_bus_not_pci) &&
687 !mp_irqs[i].mp_irqtype &&
688 (bus == lbus) &&
689 (slot == ((mp_irqs[i].mp_srcbusirq >> 2) & 0x1f))) {
690 int irq = pin_2_irq(i, apic, mp_irqs[i].mp_dstirq);
692 if (!(apic || IO_APIC_IRQ(irq)))
693 continue;
695 if (pin == (mp_irqs[i].mp_srcbusirq & 3))
696 return irq;
698 * Use the first all-but-pin matching entry as a
699 * best-guess fuzzy result for broken mptables.
701 if (best_guess < 0)
702 best_guess = irq;
705 return best_guess;
707 EXPORT_SYMBOL(IO_APIC_get_PCI_irq_vector);
710 * This function currently is only a helper for the i386 smp boot process where
711 * we need to reprogram the ioredtbls to cater for the cpus which have come online
712 * so mask in all cases should simply be TARGET_CPUS
714 #ifdef CONFIG_SMP
715 void __init setup_ioapic_dest(void)
717 int pin, ioapic, irq, irq_entry;
719 if (skip_ioapic_setup == 1)
720 return;
722 for (ioapic = 0; ioapic < nr_ioapics; ioapic++) {
723 for (pin = 0; pin < nr_ioapic_registers[ioapic]; pin++) {
724 irq_entry = find_irq_entry(ioapic, pin, mp_INT);
725 if (irq_entry == -1)
726 continue;
727 irq = pin_2_irq(irq_entry, ioapic, pin);
728 set_ioapic_affinity_irq(irq, TARGET_CPUS);
733 #endif
735 #if defined(CONFIG_EISA) || defined(CONFIG_MCA)
737 * EISA Edge/Level control register, ELCR
739 static int EISA_ELCR(unsigned int irq)
741 if (irq < 16) {
742 unsigned int port = 0x4d0 + (irq >> 3);
743 return (inb(port) >> (irq & 7)) & 1;
745 apic_printk(APIC_VERBOSE, KERN_INFO
746 "Broken MPtable reports ISA irq %d\n", irq);
747 return 0;
749 #endif
751 /* ISA interrupts are always polarity zero edge triggered,
752 * when listed as conforming in the MP table. */
754 #define default_ISA_trigger(idx) (0)
755 #define default_ISA_polarity(idx) (0)
757 /* EISA interrupts are always polarity zero and can be edge or level
758 * trigger depending on the ELCR value. If an interrupt is listed as
759 * EISA conforming in the MP table, that means its trigger type must
760 * be read in from the ELCR */
762 #define default_EISA_trigger(idx) (EISA_ELCR(mp_irqs[idx].mp_srcbusirq))
763 #define default_EISA_polarity(idx) default_ISA_polarity(idx)
765 /* PCI interrupts are always polarity one level triggered,
766 * when listed as conforming in the MP table. */
768 #define default_PCI_trigger(idx) (1)
769 #define default_PCI_polarity(idx) (1)
771 /* MCA interrupts are always polarity zero level triggered,
772 * when listed as conforming in the MP table. */
774 #define default_MCA_trigger(idx) (1)
775 #define default_MCA_polarity(idx) default_ISA_polarity(idx)
777 static int MPBIOS_polarity(int idx)
779 int bus = mp_irqs[idx].mp_srcbus;
780 int polarity;
783 * Determine IRQ line polarity (high active or low active):
785 switch (mp_irqs[idx].mp_irqflag & 3) {
786 case 0: /* conforms, ie. bus-type dependent polarity */
788 polarity = test_bit(bus, mp_bus_not_pci)?
789 default_ISA_polarity(idx):
790 default_PCI_polarity(idx);
791 break;
793 case 1: /* high active */
795 polarity = 0;
796 break;
798 case 2: /* reserved */
800 printk(KERN_WARNING "broken BIOS!!\n");
801 polarity = 1;
802 break;
804 case 3: /* low active */
806 polarity = 1;
807 break;
809 default: /* invalid */
811 printk(KERN_WARNING "broken BIOS!!\n");
812 polarity = 1;
813 break;
816 return polarity;
819 static int MPBIOS_trigger(int idx)
821 int bus = mp_irqs[idx].mp_srcbus;
822 int trigger;
825 * Determine IRQ trigger mode (edge or level sensitive):
827 switch ((mp_irqs[idx].mp_irqflag>>2) & 3) {
828 case 0: /* conforms, ie. bus-type dependent */
830 trigger = test_bit(bus, mp_bus_not_pci)?
831 default_ISA_trigger(idx):
832 default_PCI_trigger(idx);
833 #if defined(CONFIG_EISA) || defined(CONFIG_MCA)
834 switch (mp_bus_id_to_type[bus]) {
835 case MP_BUS_ISA: /* ISA pin */
837 /* set before the switch */
838 break;
840 case MP_BUS_EISA: /* EISA pin */
842 trigger = default_EISA_trigger(idx);
843 break;
845 case MP_BUS_PCI: /* PCI pin */
847 /* set before the switch */
848 break;
850 case MP_BUS_MCA: /* MCA pin */
852 trigger = default_MCA_trigger(idx);
853 break;
855 default:
857 printk(KERN_WARNING "broken BIOS!!\n");
858 trigger = 1;
859 break;
862 #endif
863 break;
865 case 1: /* edge */
867 trigger = 0;
868 break;
870 case 2: /* reserved */
872 printk(KERN_WARNING "broken BIOS!!\n");
873 trigger = 1;
874 break;
876 case 3: /* level */
878 trigger = 1;
879 break;
881 default: /* invalid */
883 printk(KERN_WARNING "broken BIOS!!\n");
884 trigger = 0;
885 break;
888 return trigger;
891 static inline int irq_polarity(int idx)
893 return MPBIOS_polarity(idx);
896 static inline int irq_trigger(int idx)
898 return MPBIOS_trigger(idx);
901 static int pin_2_irq(int idx, int apic, int pin)
903 int irq, i;
904 int bus = mp_irqs[idx].mp_srcbus;
907 * Debugging check, we are in big trouble if this message pops up!
909 if (mp_irqs[idx].mp_dstirq != pin)
910 printk(KERN_ERR "broken BIOS or MPTABLE parser, ayiee!!\n");
912 if (test_bit(bus, mp_bus_not_pci))
913 irq = mp_irqs[idx].mp_srcbusirq;
914 else {
916 * PCI IRQs are mapped in order
918 i = irq = 0;
919 while (i < apic)
920 irq += nr_ioapic_registers[i++];
921 irq += pin;
924 * For MPS mode, so far only needed by ES7000 platform
926 if (ioapic_renumber_irq)
927 irq = ioapic_renumber_irq(apic, irq);
931 * PCI IRQ command line redirection. Yes, limits are hardcoded.
933 if ((pin >= 16) && (pin <= 23)) {
934 if (pirq_entries[pin-16] != -1) {
935 if (!pirq_entries[pin-16]) {
936 apic_printk(APIC_VERBOSE, KERN_DEBUG
937 "disabling PIRQ%d\n", pin-16);
938 } else {
939 irq = pirq_entries[pin-16];
940 apic_printk(APIC_VERBOSE, KERN_DEBUG
941 "using PIRQ%d -> IRQ %d\n",
942 pin-16, irq);
946 return irq;
949 static inline int IO_APIC_irq_trigger(int irq)
951 int apic, idx, pin;
953 for (apic = 0; apic < nr_ioapics; apic++) {
954 for (pin = 0; pin < nr_ioapic_registers[apic]; pin++) {
955 idx = find_irq_entry(apic, pin, mp_INT);
956 if ((idx != -1) && (irq == pin_2_irq(idx, apic, pin)))
957 return irq_trigger(idx);
961 * nonexistent IRQs are edge default
963 return 0;
967 static int __assign_irq_vector(int irq)
969 static int current_vector = FIRST_DEVICE_VECTOR, current_offset;
970 int vector, offset;
971 struct irq_cfg *cfg;
973 BUG_ON((unsigned)irq >= nr_irqs);
975 cfg = irq_cfg(irq);
976 if (cfg->vector > 0)
977 return cfg->vector;
979 vector = current_vector;
980 offset = current_offset;
981 next:
982 vector += 8;
983 if (vector >= first_system_vector) {
984 offset = (offset + 1) % 8;
985 vector = FIRST_DEVICE_VECTOR + offset;
987 if (vector == current_vector)
988 return -ENOSPC;
989 if (test_and_set_bit(vector, used_vectors))
990 goto next;
992 current_vector = vector;
993 current_offset = offset;
994 cfg->vector = vector;
996 return vector;
999 static int assign_irq_vector(int irq)
1001 unsigned long flags;
1002 int vector;
1004 spin_lock_irqsave(&vector_lock, flags);
1005 vector = __assign_irq_vector(irq);
1006 spin_unlock_irqrestore(&vector_lock, flags);
1008 return vector;
1011 static struct irq_chip ioapic_chip;
1013 #define IOAPIC_AUTO -1
1014 #define IOAPIC_EDGE 0
1015 #define IOAPIC_LEVEL 1
1017 static void ioapic_register_intr(int irq, int vector, unsigned long trigger)
1019 struct irq_desc *desc;
1021 desc = irq_to_desc(irq);
1022 if ((trigger == IOAPIC_AUTO && IO_APIC_irq_trigger(irq)) ||
1023 trigger == IOAPIC_LEVEL) {
1024 desc->status |= IRQ_LEVEL;
1025 set_irq_chip_and_handler_name(irq, &ioapic_chip,
1026 handle_fasteoi_irq, "fasteoi");
1027 } else {
1028 desc->status &= ~IRQ_LEVEL;
1029 set_irq_chip_and_handler_name(irq, &ioapic_chip,
1030 handle_edge_irq, "edge");
1032 set_intr_gate(vector, interrupt[irq]);
1035 static void __init setup_IO_APIC_irqs(void)
1037 struct IO_APIC_route_entry entry;
1038 int apic, pin, idx, irq, first_notcon = 1, vector;
1040 apic_printk(APIC_VERBOSE, KERN_DEBUG "init IO_APIC IRQs\n");
1042 for (apic = 0; apic < nr_ioapics; apic++) {
1043 for (pin = 0; pin < nr_ioapic_registers[apic]; pin++) {
1046 * add it to the IO-APIC irq-routing table:
1048 memset(&entry, 0, sizeof(entry));
1050 entry.delivery_mode = INT_DELIVERY_MODE;
1051 entry.dest_mode = INT_DEST_MODE;
1052 entry.mask = 0; /* enable IRQ */
1053 entry.dest.logical.logical_dest =
1054 cpu_mask_to_apicid(TARGET_CPUS);
1056 idx = find_irq_entry(apic, pin, mp_INT);
1057 if (idx == -1) {
1058 if (first_notcon) {
1059 apic_printk(APIC_VERBOSE, KERN_DEBUG
1060 " IO-APIC (apicid-pin) %d-%d",
1061 mp_ioapics[apic].mp_apicid,
1062 pin);
1063 first_notcon = 0;
1064 } else
1065 apic_printk(APIC_VERBOSE, ", %d-%d",
1066 mp_ioapics[apic].mp_apicid, pin);
1067 continue;
1070 if (!first_notcon) {
1071 apic_printk(APIC_VERBOSE, " not connected.\n");
1072 first_notcon = 1;
1075 entry.trigger = irq_trigger(idx);
1076 entry.polarity = irq_polarity(idx);
1078 if (irq_trigger(idx)) {
1079 entry.trigger = 1;
1080 entry.mask = 1;
1083 irq = pin_2_irq(idx, apic, pin);
1085 * skip adding the timer int on secondary nodes, which causes
1086 * a small but painful rift in the time-space continuum
1088 if (multi_timer_check(apic, irq))
1089 continue;
1090 else
1091 add_pin_to_irq(irq, apic, pin);
1093 if (!apic && !IO_APIC_IRQ(irq))
1094 continue;
1096 if (IO_APIC_IRQ(irq)) {
1097 vector = assign_irq_vector(irq);
1098 entry.vector = vector;
1099 ioapic_register_intr(irq, vector, IOAPIC_AUTO);
1101 if (!apic && (irq < 16))
1102 disable_8259A_irq(irq);
1104 ioapic_write_entry(apic, pin, entry);
1108 if (!first_notcon)
1109 apic_printk(APIC_VERBOSE, " not connected.\n");
1113 * Set up the timer pin, possibly with the 8259A-master behind.
1115 static void __init setup_timer_IRQ0_pin(unsigned int apic, unsigned int pin,
1116 int vector)
1118 struct IO_APIC_route_entry entry;
1120 memset(&entry, 0, sizeof(entry));
1123 * We use logical delivery to get the timer IRQ
1124 * to the first CPU.
1126 entry.dest_mode = INT_DEST_MODE;
1127 entry.mask = 1; /* mask IRQ now */
1128 entry.dest.logical.logical_dest = cpu_mask_to_apicid(TARGET_CPUS);
1129 entry.delivery_mode = INT_DELIVERY_MODE;
1130 entry.polarity = 0;
1131 entry.trigger = 0;
1132 entry.vector = vector;
1135 * The timer IRQ doesn't have to know that behind the
1136 * scene we may have a 8259A-master in AEOI mode ...
1138 ioapic_register_intr(0, vector, IOAPIC_EDGE);
1141 * Add it to the IO-APIC irq-routing table:
1143 ioapic_write_entry(apic, pin, entry);
1147 __apicdebuginit(void) print_IO_APIC(void)
1149 int apic, i;
1150 union IO_APIC_reg_00 reg_00;
1151 union IO_APIC_reg_01 reg_01;
1152 union IO_APIC_reg_02 reg_02;
1153 union IO_APIC_reg_03 reg_03;
1154 unsigned long flags;
1156 if (apic_verbosity == APIC_QUIET)
1157 return;
1159 printk(KERN_DEBUG "number of MP IRQ sources: %d.\n", mp_irq_entries);
1160 for (i = 0; i < nr_ioapics; i++)
1161 printk(KERN_DEBUG "number of IO-APIC #%d registers: %d.\n",
1162 mp_ioapics[i].mp_apicid, nr_ioapic_registers[i]);
1165 * We are a bit conservative about what we expect. We have to
1166 * know about every hardware change ASAP.
1168 printk(KERN_INFO "testing the IO APIC.......................\n");
1170 for (apic = 0; apic < nr_ioapics; apic++) {
1172 spin_lock_irqsave(&ioapic_lock, flags);
1173 reg_00.raw = io_apic_read(apic, 0);
1174 reg_01.raw = io_apic_read(apic, 1);
1175 if (reg_01.bits.version >= 0x10)
1176 reg_02.raw = io_apic_read(apic, 2);
1177 if (reg_01.bits.version >= 0x20)
1178 reg_03.raw = io_apic_read(apic, 3);
1179 spin_unlock_irqrestore(&ioapic_lock, flags);
1181 printk(KERN_DEBUG "IO APIC #%d......\n", mp_ioapics[apic].mp_apicid);
1182 printk(KERN_DEBUG ".... register #00: %08X\n", reg_00.raw);
1183 printk(KERN_DEBUG "....... : physical APIC id: %02X\n", reg_00.bits.ID);
1184 printk(KERN_DEBUG "....... : Delivery Type: %X\n", reg_00.bits.delivery_type);
1185 printk(KERN_DEBUG "....... : LTS : %X\n", reg_00.bits.LTS);
1187 printk(KERN_DEBUG ".... register #01: %08X\n", reg_01.raw);
1188 printk(KERN_DEBUG "....... : max redirection entries: %04X\n", reg_01.bits.entries);
1190 printk(KERN_DEBUG "....... : PRQ implemented: %X\n", reg_01.bits.PRQ);
1191 printk(KERN_DEBUG "....... : IO APIC version: %04X\n", reg_01.bits.version);
1194 * Some Intel chipsets with IO APIC VERSION of 0x1? don't have reg_02,
1195 * but the value of reg_02 is read as the previous read register
1196 * value, so ignore it if reg_02 == reg_01.
1198 if (reg_01.bits.version >= 0x10 && reg_02.raw != reg_01.raw) {
1199 printk(KERN_DEBUG ".... register #02: %08X\n", reg_02.raw);
1200 printk(KERN_DEBUG "....... : arbitration: %02X\n", reg_02.bits.arbitration);
1204 * Some Intel chipsets with IO APIC VERSION of 0x2? don't have reg_02
1205 * or reg_03, but the value of reg_0[23] is read as the previous read
1206 * register value, so ignore it if reg_03 == reg_0[12].
1208 if (reg_01.bits.version >= 0x20 && reg_03.raw != reg_02.raw &&
1209 reg_03.raw != reg_01.raw) {
1210 printk(KERN_DEBUG ".... register #03: %08X\n", reg_03.raw);
1211 printk(KERN_DEBUG "....... : Boot DT : %X\n", reg_03.bits.boot_DT);
1214 printk(KERN_DEBUG ".... IRQ redirection table:\n");
1216 printk(KERN_DEBUG " NR Log Phy Mask Trig IRR Pol"
1217 " Stat Dest Deli Vect: \n");
1219 for (i = 0; i <= reg_01.bits.entries; i++) {
1220 struct IO_APIC_route_entry entry;
1222 entry = ioapic_read_entry(apic, i);
1224 printk(KERN_DEBUG " %02x %03X %02X ",
1226 entry.dest.logical.logical_dest,
1227 entry.dest.physical.physical_dest
1230 printk("%1d %1d %1d %1d %1d %1d %1d %02X\n",
1231 entry.mask,
1232 entry.trigger,
1233 entry.irr,
1234 entry.polarity,
1235 entry.delivery_status,
1236 entry.dest_mode,
1237 entry.delivery_mode,
1238 entry.vector
1242 printk(KERN_DEBUG "IRQ to pin mappings:\n");
1243 for (i = 0; i < nr_irqs; i++) {
1244 struct irq_pin_list *entry = irq_2_pin + i;
1245 if (entry->pin < 0)
1246 continue;
1247 printk(KERN_DEBUG "IRQ%d ", i);
1248 for (;;) {
1249 printk("-> %d:%d", entry->apic, entry->pin);
1250 if (!entry->next)
1251 break;
1252 entry = irq_2_pin + entry->next;
1254 printk("\n");
1257 printk(KERN_INFO ".................................... done.\n");
1259 return;
1262 __apicdebuginit(void) print_APIC_bitfield(int base)
1264 unsigned int v;
1265 int i, j;
1267 if (apic_verbosity == APIC_QUIET)
1268 return;
1270 printk(KERN_DEBUG "0123456789abcdef0123456789abcdef\n" KERN_DEBUG);
1271 for (i = 0; i < 8; i++) {
1272 v = apic_read(base + i*0x10);
1273 for (j = 0; j < 32; j++) {
1274 if (v & (1<<j))
1275 printk("1");
1276 else
1277 printk("0");
1279 printk("\n");
1283 __apicdebuginit(void) print_local_APIC(void *dummy)
1285 unsigned int v, ver, maxlvt;
1286 u64 icr;
1288 if (apic_verbosity == APIC_QUIET)
1289 return;
1291 printk("\n" KERN_DEBUG "printing local APIC contents on CPU#%d/%d:\n",
1292 smp_processor_id(), hard_smp_processor_id());
1293 v = apic_read(APIC_ID);
1294 printk(KERN_INFO "... APIC ID: %08x (%01x)\n", v,
1295 GET_APIC_ID(v));
1296 v = apic_read(APIC_LVR);
1297 printk(KERN_INFO "... APIC VERSION: %08x\n", v);
1298 ver = GET_APIC_VERSION(v);
1299 maxlvt = lapic_get_maxlvt();
1301 v = apic_read(APIC_TASKPRI);
1302 printk(KERN_DEBUG "... APIC TASKPRI: %08x (%02x)\n", v, v & APIC_TPRI_MASK);
1304 if (APIC_INTEGRATED(ver)) { /* !82489DX */
1305 v = apic_read(APIC_ARBPRI);
1306 printk(KERN_DEBUG "... APIC ARBPRI: %08x (%02x)\n", v,
1307 v & APIC_ARBPRI_MASK);
1308 v = apic_read(APIC_PROCPRI);
1309 printk(KERN_DEBUG "... APIC PROCPRI: %08x\n", v);
1312 v = apic_read(APIC_EOI);
1313 printk(KERN_DEBUG "... APIC EOI: %08x\n", v);
1314 v = apic_read(APIC_RRR);
1315 printk(KERN_DEBUG "... APIC RRR: %08x\n", v);
1316 v = apic_read(APIC_LDR);
1317 printk(KERN_DEBUG "... APIC LDR: %08x\n", v);
1318 v = apic_read(APIC_DFR);
1319 printk(KERN_DEBUG "... APIC DFR: %08x\n", v);
1320 v = apic_read(APIC_SPIV);
1321 printk(KERN_DEBUG "... APIC SPIV: %08x\n", v);
1323 printk(KERN_DEBUG "... APIC ISR field:\n");
1324 print_APIC_bitfield(APIC_ISR);
1325 printk(KERN_DEBUG "... APIC TMR field:\n");
1326 print_APIC_bitfield(APIC_TMR);
1327 printk(KERN_DEBUG "... APIC IRR field:\n");
1328 print_APIC_bitfield(APIC_IRR);
1330 if (APIC_INTEGRATED(ver)) { /* !82489DX */
1331 if (maxlvt > 3) /* Due to the Pentium erratum 3AP. */
1332 apic_write(APIC_ESR, 0);
1333 v = apic_read(APIC_ESR);
1334 printk(KERN_DEBUG "... APIC ESR: %08x\n", v);
1337 icr = apic_icr_read();
1338 printk(KERN_DEBUG "... APIC ICR: %08x\n", icr);
1339 printk(KERN_DEBUG "... APIC ICR2: %08x\n", icr >> 32);
1341 v = apic_read(APIC_LVTT);
1342 printk(KERN_DEBUG "... APIC LVTT: %08x\n", v);
1344 if (maxlvt > 3) { /* PC is LVT#4. */
1345 v = apic_read(APIC_LVTPC);
1346 printk(KERN_DEBUG "... APIC LVTPC: %08x\n", v);
1348 v = apic_read(APIC_LVT0);
1349 printk(KERN_DEBUG "... APIC LVT0: %08x\n", v);
1350 v = apic_read(APIC_LVT1);
1351 printk(KERN_DEBUG "... APIC LVT1: %08x\n", v);
1353 if (maxlvt > 2) { /* ERR is LVT#3. */
1354 v = apic_read(APIC_LVTERR);
1355 printk(KERN_DEBUG "... APIC LVTERR: %08x\n", v);
1358 v = apic_read(APIC_TMICT);
1359 printk(KERN_DEBUG "... APIC TMICT: %08x\n", v);
1360 v = apic_read(APIC_TMCCT);
1361 printk(KERN_DEBUG "... APIC TMCCT: %08x\n", v);
1362 v = apic_read(APIC_TDCR);
1363 printk(KERN_DEBUG "... APIC TDCR: %08x\n", v);
1364 printk("\n");
1367 __apicdebuginit(void) print_all_local_APICs(void)
1369 on_each_cpu(print_local_APIC, NULL, 1);
1372 __apicdebuginit(void) print_PIC(void)
1374 unsigned int v;
1375 unsigned long flags;
1377 if (apic_verbosity == APIC_QUIET)
1378 return;
1380 printk(KERN_DEBUG "\nprinting PIC contents\n");
1382 spin_lock_irqsave(&i8259A_lock, flags);
1384 v = inb(0xa1) << 8 | inb(0x21);
1385 printk(KERN_DEBUG "... PIC IMR: %04x\n", v);
1387 v = inb(0xa0) << 8 | inb(0x20);
1388 printk(KERN_DEBUG "... PIC IRR: %04x\n", v);
1390 outb(0x0b, 0xa0);
1391 outb(0x0b, 0x20);
1392 v = inb(0xa0) << 8 | inb(0x20);
1393 outb(0x0a, 0xa0);
1394 outb(0x0a, 0x20);
1396 spin_unlock_irqrestore(&i8259A_lock, flags);
1398 printk(KERN_DEBUG "... PIC ISR: %04x\n", v);
1400 v = inb(0x4d1) << 8 | inb(0x4d0);
1401 printk(KERN_DEBUG "... PIC ELCR: %04x\n", v);
1404 __apicdebuginit(int) print_all_ICs(void)
1406 print_PIC();
1407 print_all_local_APICs();
1408 print_IO_APIC();
1410 return 0;
1413 fs_initcall(print_all_ICs);
1416 static void __init enable_IO_APIC(void)
1418 union IO_APIC_reg_01 reg_01;
1419 int i8259_apic, i8259_pin;
1420 int i, apic;
1421 unsigned long flags;
1423 for (i = 0; i < pin_map_size; i++) {
1424 irq_2_pin[i].pin = -1;
1425 irq_2_pin[i].next = 0;
1427 if (!pirqs_enabled)
1428 for (i = 0; i < MAX_PIRQS; i++)
1429 pirq_entries[i] = -1;
1432 * The number of IO-APIC IRQ registers (== #pins):
1434 for (apic = 0; apic < nr_ioapics; apic++) {
1435 spin_lock_irqsave(&ioapic_lock, flags);
1436 reg_01.raw = io_apic_read(apic, 1);
1437 spin_unlock_irqrestore(&ioapic_lock, flags);
1438 nr_ioapic_registers[apic] = reg_01.bits.entries+1;
1440 for (apic = 0; apic < nr_ioapics; apic++) {
1441 int pin;
1442 /* See if any of the pins is in ExtINT mode */
1443 for (pin = 0; pin < nr_ioapic_registers[apic]; pin++) {
1444 struct IO_APIC_route_entry entry;
1445 entry = ioapic_read_entry(apic, pin);
1448 /* If the interrupt line is enabled and in ExtInt mode
1449 * I have found the pin where the i8259 is connected.
1451 if ((entry.mask == 0) && (entry.delivery_mode == dest_ExtINT)) {
1452 ioapic_i8259.apic = apic;
1453 ioapic_i8259.pin = pin;
1454 goto found_i8259;
1458 found_i8259:
1459 /* Look to see what if the MP table has reported the ExtINT */
1460 /* If we could not find the appropriate pin by looking at the ioapic
1461 * the i8259 probably is not connected the ioapic but give the
1462 * mptable a chance anyway.
1464 i8259_pin = find_isa_irq_pin(0, mp_ExtINT);
1465 i8259_apic = find_isa_irq_apic(0, mp_ExtINT);
1466 /* Trust the MP table if nothing is setup in the hardware */
1467 if ((ioapic_i8259.pin == -1) && (i8259_pin >= 0)) {
1468 printk(KERN_WARNING "ExtINT not setup in hardware but reported by MP table\n");
1469 ioapic_i8259.pin = i8259_pin;
1470 ioapic_i8259.apic = i8259_apic;
1472 /* Complain if the MP table and the hardware disagree */
1473 if (((ioapic_i8259.apic != i8259_apic) || (ioapic_i8259.pin != i8259_pin)) &&
1474 (i8259_pin >= 0) && (ioapic_i8259.pin >= 0))
1476 printk(KERN_WARNING "ExtINT in hardware and MP table differ\n");
1480 * Do not trust the IO-APIC being empty at bootup
1482 clear_IO_APIC();
1486 * Not an __init, needed by the reboot code
1488 void disable_IO_APIC(void)
1491 * Clear the IO-APIC before rebooting:
1493 clear_IO_APIC();
1496 * If the i8259 is routed through an IOAPIC
1497 * Put that IOAPIC in virtual wire mode
1498 * so legacy interrupts can be delivered.
1500 if (ioapic_i8259.pin != -1) {
1501 struct IO_APIC_route_entry entry;
1503 memset(&entry, 0, sizeof(entry));
1504 entry.mask = 0; /* Enabled */
1505 entry.trigger = 0; /* Edge */
1506 entry.irr = 0;
1507 entry.polarity = 0; /* High */
1508 entry.delivery_status = 0;
1509 entry.dest_mode = 0; /* Physical */
1510 entry.delivery_mode = dest_ExtINT; /* ExtInt */
1511 entry.vector = 0;
1512 entry.dest.physical.physical_dest = read_apic_id();
1515 * Add it to the IO-APIC irq-routing table:
1517 ioapic_write_entry(ioapic_i8259.apic, ioapic_i8259.pin, entry);
1519 disconnect_bsp_APIC(ioapic_i8259.pin != -1);
1523 * function to set the IO-APIC physical IDs based on the
1524 * values stored in the MPC table.
1526 * by Matt Domsch <Matt_Domsch@dell.com> Tue Dec 21 12:25:05 CST 1999
1529 static void __init setup_ioapic_ids_from_mpc(void)
1531 union IO_APIC_reg_00 reg_00;
1532 physid_mask_t phys_id_present_map;
1533 int apic;
1534 int i;
1535 unsigned char old_id;
1536 unsigned long flags;
1538 if (x86_quirks->setup_ioapic_ids && x86_quirks->setup_ioapic_ids())
1539 return;
1542 * Don't check I/O APIC IDs for xAPIC systems. They have
1543 * no meaning without the serial APIC bus.
1545 if (!(boot_cpu_data.x86_vendor == X86_VENDOR_INTEL)
1546 || APIC_XAPIC(apic_version[boot_cpu_physical_apicid]))
1547 return;
1549 * This is broken; anything with a real cpu count has to
1550 * circumvent this idiocy regardless.
1552 phys_id_present_map = ioapic_phys_id_map(phys_cpu_present_map);
1555 * Set the IOAPIC ID to the value stored in the MPC table.
1557 for (apic = 0; apic < nr_ioapics; apic++) {
1559 /* Read the register 0 value */
1560 spin_lock_irqsave(&ioapic_lock, flags);
1561 reg_00.raw = io_apic_read(apic, 0);
1562 spin_unlock_irqrestore(&ioapic_lock, flags);
1564 old_id = mp_ioapics[apic].mp_apicid;
1566 if (mp_ioapics[apic].mp_apicid >= get_physical_broadcast()) {
1567 printk(KERN_ERR "BIOS bug, IO-APIC#%d ID is %d in the MPC table!...\n",
1568 apic, mp_ioapics[apic].mp_apicid);
1569 printk(KERN_ERR "... fixing up to %d. (tell your hw vendor)\n",
1570 reg_00.bits.ID);
1571 mp_ioapics[apic].mp_apicid = reg_00.bits.ID;
1575 * Sanity check, is the ID really free? Every APIC in a
1576 * system must have a unique ID or we get lots of nice
1577 * 'stuck on smp_invalidate_needed IPI wait' messages.
1579 if (check_apicid_used(phys_id_present_map,
1580 mp_ioapics[apic].mp_apicid)) {
1581 printk(KERN_ERR "BIOS bug, IO-APIC#%d ID %d is already used!...\n",
1582 apic, mp_ioapics[apic].mp_apicid);
1583 for (i = 0; i < get_physical_broadcast(); i++)
1584 if (!physid_isset(i, phys_id_present_map))
1585 break;
1586 if (i >= get_physical_broadcast())
1587 panic("Max APIC ID exceeded!\n");
1588 printk(KERN_ERR "... fixing up to %d. (tell your hw vendor)\n",
1590 physid_set(i, phys_id_present_map);
1591 mp_ioapics[apic].mp_apicid = i;
1592 } else {
1593 physid_mask_t tmp;
1594 tmp = apicid_to_cpu_present(mp_ioapics[apic].mp_apicid);
1595 apic_printk(APIC_VERBOSE, "Setting %d in the "
1596 "phys_id_present_map\n",
1597 mp_ioapics[apic].mp_apicid);
1598 physids_or(phys_id_present_map, phys_id_present_map, tmp);
1603 * We need to adjust the IRQ routing table
1604 * if the ID changed.
1606 if (old_id != mp_ioapics[apic].mp_apicid)
1607 for (i = 0; i < mp_irq_entries; i++)
1608 if (mp_irqs[i].mp_dstapic == old_id)
1609 mp_irqs[i].mp_dstapic
1610 = mp_ioapics[apic].mp_apicid;
1613 * Read the right value from the MPC table and
1614 * write it into the ID register.
1616 apic_printk(APIC_VERBOSE, KERN_INFO
1617 "...changing IO-APIC physical APIC ID to %d ...",
1618 mp_ioapics[apic].mp_apicid);
1620 reg_00.bits.ID = mp_ioapics[apic].mp_apicid;
1621 spin_lock_irqsave(&ioapic_lock, flags);
1622 io_apic_write(apic, 0, reg_00.raw);
1623 spin_unlock_irqrestore(&ioapic_lock, flags);
1626 * Sanity check
1628 spin_lock_irqsave(&ioapic_lock, flags);
1629 reg_00.raw = io_apic_read(apic, 0);
1630 spin_unlock_irqrestore(&ioapic_lock, flags);
1631 if (reg_00.bits.ID != mp_ioapics[apic].mp_apicid)
1632 printk("could not set ID!\n");
1633 else
1634 apic_printk(APIC_VERBOSE, " ok.\n");
1638 int no_timer_check __initdata;
1640 static int __init notimercheck(char *s)
1642 no_timer_check = 1;
1643 return 1;
1645 __setup("no_timer_check", notimercheck);
1648 * There is a nasty bug in some older SMP boards, their mptable lies
1649 * about the timer IRQ. We do the following to work around the situation:
1651 * - timer IRQ defaults to IO-APIC IRQ
1652 * - if this function detects that timer IRQs are defunct, then we fall
1653 * back to ISA timer IRQs
1655 static int __init timer_irq_works(void)
1657 unsigned long t1 = jiffies;
1658 unsigned long flags;
1660 if (no_timer_check)
1661 return 1;
1663 local_save_flags(flags);
1664 local_irq_enable();
1665 /* Let ten ticks pass... */
1666 mdelay((10 * 1000) / HZ);
1667 local_irq_restore(flags);
1670 * Expect a few ticks at least, to be sure some possible
1671 * glue logic does not lock up after one or two first
1672 * ticks in a non-ExtINT mode. Also the local APIC
1673 * might have cached one ExtINT interrupt. Finally, at
1674 * least one tick may be lost due to delays.
1676 if (time_after(jiffies, t1 + 4))
1677 return 1;
1679 return 0;
1683 * In the SMP+IOAPIC case it might happen that there are an unspecified
1684 * number of pending IRQ events unhandled. These cases are very rare,
1685 * so we 'resend' these IRQs via IPIs, to the same CPU. It's much
1686 * better to do it this way as thus we do not have to be aware of
1687 * 'pending' interrupts in the IRQ path, except at this point.
1690 * Edge triggered needs to resend any interrupt
1691 * that was delayed but this is now handled in the device
1692 * independent code.
1696 * Startup quirk:
1698 * Starting up a edge-triggered IO-APIC interrupt is
1699 * nasty - we need to make sure that we get the edge.
1700 * If it is already asserted for some reason, we need
1701 * return 1 to indicate that is was pending.
1703 * This is not complete - we should be able to fake
1704 * an edge even if it isn't on the 8259A...
1706 * (We do this for level-triggered IRQs too - it cannot hurt.)
1708 static unsigned int startup_ioapic_irq(unsigned int irq)
1710 int was_pending = 0;
1711 unsigned long flags;
1713 spin_lock_irqsave(&ioapic_lock, flags);
1714 if (irq < 16) {
1715 disable_8259A_irq(irq);
1716 if (i8259A_irq_pending(irq))
1717 was_pending = 1;
1719 __unmask_IO_APIC_irq(irq);
1720 spin_unlock_irqrestore(&ioapic_lock, flags);
1722 return was_pending;
1725 static void ack_ioapic_irq(unsigned int irq)
1727 move_native_irq(irq);
1728 ack_APIC_irq();
1731 static void ack_ioapic_quirk_irq(unsigned int irq)
1733 unsigned long v;
1734 int i;
1736 move_native_irq(irq);
1738 * It appears there is an erratum which affects at least version 0x11
1739 * of I/O APIC (that's the 82093AA and cores integrated into various
1740 * chipsets). Under certain conditions a level-triggered interrupt is
1741 * erroneously delivered as edge-triggered one but the respective IRR
1742 * bit gets set nevertheless. As a result the I/O unit expects an EOI
1743 * message but it will never arrive and further interrupts are blocked
1744 * from the source. The exact reason is so far unknown, but the
1745 * phenomenon was observed when two consecutive interrupt requests
1746 * from a given source get delivered to the same CPU and the source is
1747 * temporarily disabled in between.
1749 * A workaround is to simulate an EOI message manually. We achieve it
1750 * by setting the trigger mode to edge and then to level when the edge
1751 * trigger mode gets detected in the TMR of a local APIC for a
1752 * level-triggered interrupt. We mask the source for the time of the
1753 * operation to prevent an edge-triggered interrupt escaping meanwhile.
1754 * The idea is from Manfred Spraul. --macro
1756 i = irq_cfg(irq)->vector;
1758 v = apic_read(APIC_TMR + ((i & ~0x1f) >> 1));
1760 ack_APIC_irq();
1762 if (!(v & (1 << (i & 0x1f)))) {
1763 atomic_inc(&irq_mis_count);
1764 spin_lock(&ioapic_lock);
1765 __mask_and_edge_IO_APIC_irq(irq);
1766 __unmask_and_level_IO_APIC_irq(irq);
1767 spin_unlock(&ioapic_lock);
1771 static int ioapic_retrigger_irq(unsigned int irq)
1773 send_IPI_self(irq_cfg(irq)->vector);
1775 return 1;
1778 static struct irq_chip ioapic_chip __read_mostly = {
1779 .name = "IO-APIC",
1780 .startup = startup_ioapic_irq,
1781 .mask = mask_IO_APIC_irq,
1782 .unmask = unmask_IO_APIC_irq,
1783 .ack = ack_ioapic_irq,
1784 .eoi = ack_ioapic_quirk_irq,
1785 #ifdef CONFIG_SMP
1786 .set_affinity = set_ioapic_affinity_irq,
1787 #endif
1788 .retrigger = ioapic_retrigger_irq,
1792 static inline void init_IO_APIC_traps(void)
1794 int irq;
1795 struct irq_desc *desc;
1796 struct irq_cfg *cfg;
1799 * NOTE! The local APIC isn't very good at handling
1800 * multiple interrupts at the same interrupt level.
1801 * As the interrupt level is determined by taking the
1802 * vector number and shifting that right by 4, we
1803 * want to spread these out a bit so that they don't
1804 * all fall in the same interrupt level.
1806 * Also, we've got to be careful not to trash gate
1807 * 0x80, because int 0x80 is hm, kind of importantish. ;)
1809 for_each_irq_cfg(cfg) {
1810 irq = cfg->irq;
1811 if (IO_APIC_IRQ(irq) && !cfg->vector) {
1813 * Hmm.. We don't have an entry for this,
1814 * so default to an old-fashioned 8259
1815 * interrupt if we can..
1817 if (irq < 16)
1818 make_8259A_irq(irq);
1819 else {
1820 desc = irq_to_desc(irq);
1821 /* Strange. Oh, well.. */
1822 desc->chip = &no_irq_chip;
1829 * The local APIC irq-chip implementation:
1832 static void ack_lapic_irq(unsigned int irq)
1834 ack_APIC_irq();
1837 static void mask_lapic_irq(unsigned int irq)
1839 unsigned long v;
1841 v = apic_read(APIC_LVT0);
1842 apic_write(APIC_LVT0, v | APIC_LVT_MASKED);
1845 static void unmask_lapic_irq(unsigned int irq)
1847 unsigned long v;
1849 v = apic_read(APIC_LVT0);
1850 apic_write(APIC_LVT0, v & ~APIC_LVT_MASKED);
1853 static struct irq_chip lapic_chip __read_mostly = {
1854 .name = "local-APIC",
1855 .mask = mask_lapic_irq,
1856 .unmask = unmask_lapic_irq,
1857 .ack = ack_lapic_irq,
1860 static void lapic_register_intr(int irq, int vector)
1862 struct irq_desc *desc;
1864 desc = irq_to_desc(irq);
1865 desc->status &= ~IRQ_LEVEL;
1866 set_irq_chip_and_handler_name(irq, &lapic_chip, handle_edge_irq,
1867 "edge");
1868 set_intr_gate(vector, interrupt[irq]);
1871 static void __init setup_nmi(void)
1874 * Dirty trick to enable the NMI watchdog ...
1875 * We put the 8259A master into AEOI mode and
1876 * unmask on all local APICs LVT0 as NMI.
1878 * The idea to use the 8259A in AEOI mode ('8259A Virtual Wire')
1879 * is from Maciej W. Rozycki - so we do not have to EOI from
1880 * the NMI handler or the timer interrupt.
1882 apic_printk(APIC_VERBOSE, KERN_INFO "activating NMI Watchdog ...");
1884 enable_NMI_through_LVT0();
1886 apic_printk(APIC_VERBOSE, " done.\n");
1890 * This looks a bit hackish but it's about the only one way of sending
1891 * a few INTA cycles to 8259As and any associated glue logic. ICR does
1892 * not support the ExtINT mode, unfortunately. We need to send these
1893 * cycles as some i82489DX-based boards have glue logic that keeps the
1894 * 8259A interrupt line asserted until INTA. --macro
1896 static inline void __init unlock_ExtINT_logic(void)
1898 int apic, pin, i;
1899 struct IO_APIC_route_entry entry0, entry1;
1900 unsigned char save_control, save_freq_select;
1902 pin = find_isa_irq_pin(8, mp_INT);
1903 if (pin == -1) {
1904 WARN_ON_ONCE(1);
1905 return;
1907 apic = find_isa_irq_apic(8, mp_INT);
1908 if (apic == -1) {
1909 WARN_ON_ONCE(1);
1910 return;
1913 entry0 = ioapic_read_entry(apic, pin);
1914 clear_IO_APIC_pin(apic, pin);
1916 memset(&entry1, 0, sizeof(entry1));
1918 entry1.dest_mode = 0; /* physical delivery */
1919 entry1.mask = 0; /* unmask IRQ now */
1920 entry1.dest.physical.physical_dest = hard_smp_processor_id();
1921 entry1.delivery_mode = dest_ExtINT;
1922 entry1.polarity = entry0.polarity;
1923 entry1.trigger = 0;
1924 entry1.vector = 0;
1926 ioapic_write_entry(apic, pin, entry1);
1928 save_control = CMOS_READ(RTC_CONTROL);
1929 save_freq_select = CMOS_READ(RTC_FREQ_SELECT);
1930 CMOS_WRITE((save_freq_select & ~RTC_RATE_SELECT) | 0x6,
1931 RTC_FREQ_SELECT);
1932 CMOS_WRITE(save_control | RTC_PIE, RTC_CONTROL);
1934 i = 100;
1935 while (i-- > 0) {
1936 mdelay(10);
1937 if ((CMOS_READ(RTC_INTR_FLAGS) & RTC_PF) == RTC_PF)
1938 i -= 10;
1941 CMOS_WRITE(save_control, RTC_CONTROL);
1942 CMOS_WRITE(save_freq_select, RTC_FREQ_SELECT);
1943 clear_IO_APIC_pin(apic, pin);
1945 ioapic_write_entry(apic, pin, entry0);
1949 * This code may look a bit paranoid, but it's supposed to cooperate with
1950 * a wide range of boards and BIOS bugs. Fortunately only the timer IRQ
1951 * is so screwy. Thanks to Brian Perkins for testing/hacking this beast
1952 * fanatically on his truly buggy board.
1954 static inline void __init check_timer(void)
1956 int apic1, pin1, apic2, pin2;
1957 int no_pin1 = 0;
1958 int vector;
1959 unsigned int ver;
1960 unsigned long flags;
1962 local_irq_save(flags);
1964 ver = apic_read(APIC_LVR);
1965 ver = GET_APIC_VERSION(ver);
1968 * get/set the timer IRQ vector:
1970 disable_8259A_irq(0);
1971 vector = assign_irq_vector(0);
1972 set_intr_gate(vector, interrupt[0]);
1975 * As IRQ0 is to be enabled in the 8259A, the virtual
1976 * wire has to be disabled in the local APIC. Also
1977 * timer interrupts need to be acknowledged manually in
1978 * the 8259A for the i82489DX when using the NMI
1979 * watchdog as that APIC treats NMIs as level-triggered.
1980 * The AEOI mode will finish them in the 8259A
1981 * automatically.
1983 apic_write(APIC_LVT0, APIC_LVT_MASKED | APIC_DM_EXTINT);
1984 init_8259A(1);
1985 timer_ack = (nmi_watchdog == NMI_IO_APIC && !APIC_INTEGRATED(ver));
1987 pin1 = find_isa_irq_pin(0, mp_INT);
1988 apic1 = find_isa_irq_apic(0, mp_INT);
1989 pin2 = ioapic_i8259.pin;
1990 apic2 = ioapic_i8259.apic;
1992 apic_printk(APIC_QUIET, KERN_INFO "..TIMER: vector=0x%02X "
1993 "apic1=%d pin1=%d apic2=%d pin2=%d\n",
1994 vector, apic1, pin1, apic2, pin2);
1997 * Some BIOS writers are clueless and report the ExtINTA
1998 * I/O APIC input from the cascaded 8259A as the timer
1999 * interrupt input. So just in case, if only one pin
2000 * was found above, try it both directly and through the
2001 * 8259A.
2003 if (pin1 == -1) {
2004 pin1 = pin2;
2005 apic1 = apic2;
2006 no_pin1 = 1;
2007 } else if (pin2 == -1) {
2008 pin2 = pin1;
2009 apic2 = apic1;
2012 if (pin1 != -1) {
2014 * Ok, does IRQ0 through the IOAPIC work?
2016 if (no_pin1) {
2017 add_pin_to_irq(0, apic1, pin1);
2018 setup_timer_IRQ0_pin(apic1, pin1, vector);
2020 unmask_IO_APIC_irq(0);
2021 if (timer_irq_works()) {
2022 if (nmi_watchdog == NMI_IO_APIC) {
2023 setup_nmi();
2024 enable_8259A_irq(0);
2026 if (disable_timer_pin_1 > 0)
2027 clear_IO_APIC_pin(0, pin1);
2028 goto out;
2030 clear_IO_APIC_pin(apic1, pin1);
2031 if (!no_pin1)
2032 apic_printk(APIC_QUIET, KERN_ERR "..MP-BIOS bug: "
2033 "8254 timer not connected to IO-APIC\n");
2035 apic_printk(APIC_QUIET, KERN_INFO "...trying to set up timer "
2036 "(IRQ0) through the 8259A ...\n");
2037 apic_printk(APIC_QUIET, KERN_INFO
2038 "..... (found apic %d pin %d) ...\n", apic2, pin2);
2040 * legacy devices should be connected to IO APIC #0
2042 replace_pin_at_irq(0, apic1, pin1, apic2, pin2);
2043 setup_timer_IRQ0_pin(apic2, pin2, vector);
2044 unmask_IO_APIC_irq(0);
2045 enable_8259A_irq(0);
2046 if (timer_irq_works()) {
2047 apic_printk(APIC_QUIET, KERN_INFO "....... works.\n");
2048 timer_through_8259 = 1;
2049 if (nmi_watchdog == NMI_IO_APIC) {
2050 disable_8259A_irq(0);
2051 setup_nmi();
2052 enable_8259A_irq(0);
2054 goto out;
2057 * Cleanup, just in case ...
2059 disable_8259A_irq(0);
2060 clear_IO_APIC_pin(apic2, pin2);
2061 apic_printk(APIC_QUIET, KERN_INFO "....... failed.\n");
2064 if (nmi_watchdog == NMI_IO_APIC) {
2065 apic_printk(APIC_QUIET, KERN_WARNING "timer doesn't work "
2066 "through the IO-APIC - disabling NMI Watchdog!\n");
2067 nmi_watchdog = NMI_NONE;
2069 timer_ack = 0;
2071 apic_printk(APIC_QUIET, KERN_INFO
2072 "...trying to set up timer as Virtual Wire IRQ...\n");
2074 lapic_register_intr(0, vector);
2075 apic_write(APIC_LVT0, APIC_DM_FIXED | vector); /* Fixed mode */
2076 enable_8259A_irq(0);
2078 if (timer_irq_works()) {
2079 apic_printk(APIC_QUIET, KERN_INFO "..... works.\n");
2080 goto out;
2082 disable_8259A_irq(0);
2083 apic_write(APIC_LVT0, APIC_LVT_MASKED | APIC_DM_FIXED | vector);
2084 apic_printk(APIC_QUIET, KERN_INFO "..... failed.\n");
2086 apic_printk(APIC_QUIET, KERN_INFO
2087 "...trying to set up timer as ExtINT IRQ...\n");
2089 init_8259A(0);
2090 make_8259A_irq(0);
2091 apic_write(APIC_LVT0, APIC_DM_EXTINT);
2093 unlock_ExtINT_logic();
2095 if (timer_irq_works()) {
2096 apic_printk(APIC_QUIET, KERN_INFO "..... works.\n");
2097 goto out;
2099 apic_printk(APIC_QUIET, KERN_INFO "..... failed :(.\n");
2100 panic("IO-APIC + timer doesn't work! Boot with apic=debug and send a "
2101 "report. Then try booting with the 'noapic' option.\n");
2102 out:
2103 local_irq_restore(flags);
2107 * Traditionally ISA IRQ2 is the cascade IRQ, and is not available
2108 * to devices. However there may be an I/O APIC pin available for
2109 * this interrupt regardless. The pin may be left unconnected, but
2110 * typically it will be reused as an ExtINT cascade interrupt for
2111 * the master 8259A. In the MPS case such a pin will normally be
2112 * reported as an ExtINT interrupt in the MP table. With ACPI
2113 * there is no provision for ExtINT interrupts, and in the absence
2114 * of an override it would be treated as an ordinary ISA I/O APIC
2115 * interrupt, that is edge-triggered and unmasked by default. We
2116 * used to do this, but it caused problems on some systems because
2117 * of the NMI watchdog and sometimes IRQ0 of the 8254 timer using
2118 * the same ExtINT cascade interrupt to drive the local APIC of the
2119 * bootstrap processor. Therefore we refrain from routing IRQ2 to
2120 * the I/O APIC in all cases now. No actual device should request
2121 * it anyway. --macro
2123 #define PIC_IRQS (1 << PIC_CASCADE_IR)
2125 void __init setup_IO_APIC(void)
2127 int i;
2129 /* Reserve all the system vectors. */
2130 for (i = first_system_vector; i < NR_VECTORS; i++)
2131 set_bit(i, used_vectors);
2133 enable_IO_APIC();
2135 io_apic_irqs = ~PIC_IRQS;
2137 printk("ENABLING IO-APIC IRQs\n");
2140 * Set up IO-APIC IRQ routing.
2142 if (!acpi_ioapic)
2143 setup_ioapic_ids_from_mpc();
2144 sync_Arb_IDs();
2145 setup_IO_APIC_irqs();
2146 init_IO_APIC_traps();
2147 check_timer();
2151 * Called after all the initialization is done. If we didnt find any
2152 * APIC bugs then we can allow the modify fast path
2155 static int __init io_apic_bug_finalize(void)
2157 if (sis_apic_bug == -1)
2158 sis_apic_bug = 0;
2159 return 0;
2162 late_initcall(io_apic_bug_finalize);
2164 struct sysfs_ioapic_data {
2165 struct sys_device dev;
2166 struct IO_APIC_route_entry entry[0];
2168 static struct sysfs_ioapic_data *mp_ioapic_data[MAX_IO_APICS];
2170 static int ioapic_suspend(struct sys_device *dev, pm_message_t state)
2172 struct IO_APIC_route_entry *entry;
2173 struct sysfs_ioapic_data *data;
2174 int i;
2176 data = container_of(dev, struct sysfs_ioapic_data, dev);
2177 entry = data->entry;
2178 for (i = 0; i < nr_ioapic_registers[dev->id]; i++)
2179 entry[i] = ioapic_read_entry(dev->id, i);
2181 return 0;
2184 static int ioapic_resume(struct sys_device *dev)
2186 struct IO_APIC_route_entry *entry;
2187 struct sysfs_ioapic_data *data;
2188 unsigned long flags;
2189 union IO_APIC_reg_00 reg_00;
2190 int i;
2192 data = container_of(dev, struct sysfs_ioapic_data, dev);
2193 entry = data->entry;
2195 spin_lock_irqsave(&ioapic_lock, flags);
2196 reg_00.raw = io_apic_read(dev->id, 0);
2197 if (reg_00.bits.ID != mp_ioapics[dev->id].mp_apicid) {
2198 reg_00.bits.ID = mp_ioapics[dev->id].mp_apicid;
2199 io_apic_write(dev->id, 0, reg_00.raw);
2201 spin_unlock_irqrestore(&ioapic_lock, flags);
2202 for (i = 0; i < nr_ioapic_registers[dev->id]; i++)
2203 ioapic_write_entry(dev->id, i, entry[i]);
2205 return 0;
2208 static struct sysdev_class ioapic_sysdev_class = {
2209 .name = "ioapic",
2210 .suspend = ioapic_suspend,
2211 .resume = ioapic_resume,
2214 static int __init ioapic_init_sysfs(void)
2216 struct sys_device *dev;
2217 int i, size, error = 0;
2219 error = sysdev_class_register(&ioapic_sysdev_class);
2220 if (error)
2221 return error;
2223 for (i = 0; i < nr_ioapics; i++) {
2224 size = sizeof(struct sys_device) + nr_ioapic_registers[i]
2225 * sizeof(struct IO_APIC_route_entry);
2226 mp_ioapic_data[i] = kzalloc(size, GFP_KERNEL);
2227 if (!mp_ioapic_data[i]) {
2228 printk(KERN_ERR "Can't suspend/resume IOAPIC %d\n", i);
2229 continue;
2231 dev = &mp_ioapic_data[i]->dev;
2232 dev->id = i;
2233 dev->cls = &ioapic_sysdev_class;
2234 error = sysdev_register(dev);
2235 if (error) {
2236 kfree(mp_ioapic_data[i]);
2237 mp_ioapic_data[i] = NULL;
2238 printk(KERN_ERR "Can't suspend/resume IOAPIC %d\n", i);
2239 continue;
2243 return 0;
2246 device_initcall(ioapic_init_sysfs);
2249 * Dynamic irq allocate and deallocation
2251 int create_irq(void)
2253 /* Allocate an unused irq */
2254 int irq, new, vector = 0;
2255 unsigned long flags;
2256 struct irq_cfg *cfg_new;
2258 irq = -ENOSPC;
2259 spin_lock_irqsave(&vector_lock, flags);
2260 for (new = (nr_irqs - 1); new >= 0; new--) {
2261 if (platform_legacy_irq(new))
2262 continue;
2263 cfg_new = irq_cfg(new);
2264 if (cfg_new && cfg_new->vector != 0)
2265 continue;
2266 if (!cfg_new)
2267 cfg_new = irq_cfg_alloc(new);
2268 vector = __assign_irq_vector(new);
2269 if (likely(vector > 0))
2270 irq = new;
2271 break;
2273 spin_unlock_irqrestore(&vector_lock, flags);
2275 if (irq >= 0) {
2276 set_intr_gate(vector, interrupt[irq]);
2277 dynamic_irq_init(irq);
2279 return irq;
2282 void destroy_irq(unsigned int irq)
2284 unsigned long flags;
2286 dynamic_irq_cleanup(irq);
2288 spin_lock_irqsave(&vector_lock, flags);
2289 clear_bit(irq_cfg(irq)->vector, used_vectors);
2290 irq_cfg(irq)->vector = 0;
2291 spin_unlock_irqrestore(&vector_lock, flags);
2295 * MSI message composition
2297 #ifdef CONFIG_PCI_MSI
2298 static int msi_compose_msg(struct pci_dev *pdev, unsigned int irq, struct msi_msg *msg)
2300 int vector;
2301 unsigned dest;
2303 vector = assign_irq_vector(irq);
2304 if (vector >= 0) {
2305 dest = cpu_mask_to_apicid(TARGET_CPUS);
2307 msg->address_hi = MSI_ADDR_BASE_HI;
2308 msg->address_lo =
2309 MSI_ADDR_BASE_LO |
2310 ((INT_DEST_MODE == 0) ?
2311 MSI_ADDR_DEST_MODE_PHYSICAL:
2312 MSI_ADDR_DEST_MODE_LOGICAL) |
2313 ((INT_DELIVERY_MODE != dest_LowestPrio) ?
2314 MSI_ADDR_REDIRECTION_CPU:
2315 MSI_ADDR_REDIRECTION_LOWPRI) |
2316 MSI_ADDR_DEST_ID(dest);
2318 msg->data =
2319 MSI_DATA_TRIGGER_EDGE |
2320 MSI_DATA_LEVEL_ASSERT |
2321 ((INT_DELIVERY_MODE != dest_LowestPrio) ?
2322 MSI_DATA_DELIVERY_FIXED:
2323 MSI_DATA_DELIVERY_LOWPRI) |
2324 MSI_DATA_VECTOR(vector);
2326 return vector;
2329 #ifdef CONFIG_SMP
2330 static void set_msi_irq_affinity(unsigned int irq, cpumask_t mask)
2332 struct msi_msg msg;
2333 unsigned int dest;
2334 cpumask_t tmp;
2335 int vector;
2336 struct irq_desc *desc;
2338 cpus_and(tmp, mask, cpu_online_map);
2339 if (cpus_empty(tmp))
2340 tmp = TARGET_CPUS;
2342 vector = assign_irq_vector(irq);
2343 if (vector < 0)
2344 return;
2346 dest = cpu_mask_to_apicid(mask);
2348 read_msi_msg(irq, &msg);
2350 msg.data &= ~MSI_DATA_VECTOR_MASK;
2351 msg.data |= MSI_DATA_VECTOR(vector);
2352 msg.address_lo &= ~MSI_ADDR_DEST_ID_MASK;
2353 msg.address_lo |= MSI_ADDR_DEST_ID(dest);
2355 write_msi_msg(irq, &msg);
2356 desc = irq_to_desc(irq);
2357 desc->affinity = mask;
2359 #endif /* CONFIG_SMP */
2362 * IRQ Chip for MSI PCI/PCI-X/PCI-Express Devices,
2363 * which implement the MSI or MSI-X Capability Structure.
2365 static struct irq_chip msi_chip = {
2366 .name = "PCI-MSI",
2367 .unmask = unmask_msi_irq,
2368 .mask = mask_msi_irq,
2369 .ack = ack_ioapic_irq,
2370 #ifdef CONFIG_SMP
2371 .set_affinity = set_msi_irq_affinity,
2372 #endif
2373 .retrigger = ioapic_retrigger_irq,
2376 int arch_setup_msi_irq(struct pci_dev *dev, struct msi_desc *desc)
2378 struct msi_msg msg;
2379 int irq, ret;
2380 irq = create_irq();
2381 if (irq < 0)
2382 return irq;
2384 ret = msi_compose_msg(dev, irq, &msg);
2385 if (ret < 0) {
2386 destroy_irq(irq);
2387 return ret;
2390 set_irq_msi(irq, desc);
2391 write_msi_msg(irq, &msg);
2393 set_irq_chip_and_handler_name(irq, &msi_chip, handle_edge_irq,
2394 "edge");
2396 return 0;
2399 void arch_teardown_msi_irq(unsigned int irq)
2401 destroy_irq(irq);
2404 #endif /* CONFIG_PCI_MSI */
2407 * Hypertransport interrupt support
2409 #ifdef CONFIG_HT_IRQ
2411 #ifdef CONFIG_SMP
2413 static void target_ht_irq(unsigned int irq, unsigned int dest)
2415 struct ht_irq_msg msg;
2416 fetch_ht_irq_msg(irq, &msg);
2418 msg.address_lo &= ~(HT_IRQ_LOW_DEST_ID_MASK);
2419 msg.address_hi &= ~(HT_IRQ_HIGH_DEST_ID_MASK);
2421 msg.address_lo |= HT_IRQ_LOW_DEST_ID(dest);
2422 msg.address_hi |= HT_IRQ_HIGH_DEST_ID(dest);
2424 write_ht_irq_msg(irq, &msg);
2427 static void set_ht_irq_affinity(unsigned int irq, cpumask_t mask)
2429 unsigned int dest;
2430 cpumask_t tmp;
2431 struct irq_desc *desc;
2433 cpus_and(tmp, mask, cpu_online_map);
2434 if (cpus_empty(tmp))
2435 tmp = TARGET_CPUS;
2437 cpus_and(mask, tmp, CPU_MASK_ALL);
2439 dest = cpu_mask_to_apicid(mask);
2441 target_ht_irq(irq, dest);
2442 desc = irq_to_desc(irq);
2443 desc->affinity = mask;
2445 #endif
2447 static struct irq_chip ht_irq_chip = {
2448 .name = "PCI-HT",
2449 .mask = mask_ht_irq,
2450 .unmask = unmask_ht_irq,
2451 .ack = ack_ioapic_irq,
2452 #ifdef CONFIG_SMP
2453 .set_affinity = set_ht_irq_affinity,
2454 #endif
2455 .retrigger = ioapic_retrigger_irq,
2458 int arch_setup_ht_irq(unsigned int irq, struct pci_dev *dev)
2460 int vector;
2462 vector = assign_irq_vector(irq);
2463 if (vector >= 0) {
2464 struct ht_irq_msg msg;
2465 unsigned dest;
2466 cpumask_t tmp;
2468 cpus_clear(tmp);
2469 cpu_set(vector >> 8, tmp);
2470 dest = cpu_mask_to_apicid(tmp);
2472 msg.address_hi = HT_IRQ_HIGH_DEST_ID(dest);
2474 msg.address_lo =
2475 HT_IRQ_LOW_BASE |
2476 HT_IRQ_LOW_DEST_ID(dest) |
2477 HT_IRQ_LOW_VECTOR(vector) |
2478 ((INT_DEST_MODE == 0) ?
2479 HT_IRQ_LOW_DM_PHYSICAL :
2480 HT_IRQ_LOW_DM_LOGICAL) |
2481 HT_IRQ_LOW_RQEOI_EDGE |
2482 ((INT_DELIVERY_MODE != dest_LowestPrio) ?
2483 HT_IRQ_LOW_MT_FIXED :
2484 HT_IRQ_LOW_MT_ARBITRATED) |
2485 HT_IRQ_LOW_IRQ_MASKED;
2487 write_ht_irq_msg(irq, &msg);
2489 set_irq_chip_and_handler_name(irq, &ht_irq_chip,
2490 handle_edge_irq, "edge");
2492 return vector;
2494 #endif /* CONFIG_HT_IRQ */
2496 /* --------------------------------------------------------------------------
2497 ACPI-based IOAPIC Configuration
2498 -------------------------------------------------------------------------- */
2500 #ifdef CONFIG_ACPI
2502 int __init io_apic_get_unique_id(int ioapic, int apic_id)
2504 union IO_APIC_reg_00 reg_00;
2505 static physid_mask_t apic_id_map = PHYSID_MASK_NONE;
2506 physid_mask_t tmp;
2507 unsigned long flags;
2508 int i = 0;
2511 * The P4 platform supports up to 256 APIC IDs on two separate APIC
2512 * buses (one for LAPICs, one for IOAPICs), where predecessors only
2513 * supports up to 16 on one shared APIC bus.
2515 * TBD: Expand LAPIC/IOAPIC support on P4-class systems to take full
2516 * advantage of new APIC bus architecture.
2519 if (physids_empty(apic_id_map))
2520 apic_id_map = ioapic_phys_id_map(phys_cpu_present_map);
2522 spin_lock_irqsave(&ioapic_lock, flags);
2523 reg_00.raw = io_apic_read(ioapic, 0);
2524 spin_unlock_irqrestore(&ioapic_lock, flags);
2526 if (apic_id >= get_physical_broadcast()) {
2527 printk(KERN_WARNING "IOAPIC[%d]: Invalid apic_id %d, trying "
2528 "%d\n", ioapic, apic_id, reg_00.bits.ID);
2529 apic_id = reg_00.bits.ID;
2533 * Every APIC in a system must have a unique ID or we get lots of nice
2534 * 'stuck on smp_invalidate_needed IPI wait' messages.
2536 if (check_apicid_used(apic_id_map, apic_id)) {
2538 for (i = 0; i < get_physical_broadcast(); i++) {
2539 if (!check_apicid_used(apic_id_map, i))
2540 break;
2543 if (i == get_physical_broadcast())
2544 panic("Max apic_id exceeded!\n");
2546 printk(KERN_WARNING "IOAPIC[%d]: apic_id %d already used, "
2547 "trying %d\n", ioapic, apic_id, i);
2549 apic_id = i;
2552 tmp = apicid_to_cpu_present(apic_id);
2553 physids_or(apic_id_map, apic_id_map, tmp);
2555 if (reg_00.bits.ID != apic_id) {
2556 reg_00.bits.ID = apic_id;
2558 spin_lock_irqsave(&ioapic_lock, flags);
2559 io_apic_write(ioapic, 0, reg_00.raw);
2560 reg_00.raw = io_apic_read(ioapic, 0);
2561 spin_unlock_irqrestore(&ioapic_lock, flags);
2563 /* Sanity check */
2564 if (reg_00.bits.ID != apic_id) {
2565 printk("IOAPIC[%d]: Unable to change apic_id!\n", ioapic);
2566 return -1;
2570 apic_printk(APIC_VERBOSE, KERN_INFO
2571 "IOAPIC[%d]: Assigned apic_id %d\n", ioapic, apic_id);
2573 return apic_id;
2577 int __init io_apic_get_version(int ioapic)
2579 union IO_APIC_reg_01 reg_01;
2580 unsigned long flags;
2582 spin_lock_irqsave(&ioapic_lock, flags);
2583 reg_01.raw = io_apic_read(ioapic, 1);
2584 spin_unlock_irqrestore(&ioapic_lock, flags);
2586 return reg_01.bits.version;
2590 int __init io_apic_get_redir_entries(int ioapic)
2592 union IO_APIC_reg_01 reg_01;
2593 unsigned long flags;
2595 spin_lock_irqsave(&ioapic_lock, flags);
2596 reg_01.raw = io_apic_read(ioapic, 1);
2597 spin_unlock_irqrestore(&ioapic_lock, flags);
2599 return reg_01.bits.entries;
2603 int io_apic_set_pci_routing(int ioapic, int pin, int irq, int edge_level, int active_high_low)
2605 struct IO_APIC_route_entry entry;
2607 if (!IO_APIC_IRQ(irq)) {
2608 printk(KERN_ERR "IOAPIC[%d]: Invalid reference to IRQ 0\n",
2609 ioapic);
2610 return -EINVAL;
2614 * Generate a PCI IRQ routing entry and program the IOAPIC accordingly.
2615 * Note that we mask (disable) IRQs now -- these get enabled when the
2616 * corresponding device driver registers for this IRQ.
2619 memset(&entry, 0, sizeof(entry));
2621 entry.delivery_mode = INT_DELIVERY_MODE;
2622 entry.dest_mode = INT_DEST_MODE;
2623 entry.dest.logical.logical_dest = cpu_mask_to_apicid(TARGET_CPUS);
2624 entry.trigger = edge_level;
2625 entry.polarity = active_high_low;
2626 entry.mask = 1;
2629 * IRQs < 16 are already in the irq_2_pin[] map
2631 if (irq >= 16)
2632 add_pin_to_irq(irq, ioapic, pin);
2634 entry.vector = assign_irq_vector(irq);
2636 apic_printk(APIC_DEBUG, KERN_DEBUG "IOAPIC[%d]: Set PCI routing entry "
2637 "(%d-%d -> 0x%x -> IRQ %d Mode:%i Active:%i)\n", ioapic,
2638 mp_ioapics[ioapic].mp_apicid, pin, entry.vector, irq,
2639 edge_level, active_high_low);
2641 ioapic_register_intr(irq, entry.vector, edge_level);
2643 if (!ioapic && (irq < 16))
2644 disable_8259A_irq(irq);
2646 ioapic_write_entry(ioapic, pin, entry);
2648 return 0;
2651 int acpi_get_override_irq(int bus_irq, int *trigger, int *polarity)
2653 int i;
2655 if (skip_ioapic_setup)
2656 return -1;
2658 for (i = 0; i < mp_irq_entries; i++)
2659 if (mp_irqs[i].mp_irqtype == mp_INT &&
2660 mp_irqs[i].mp_srcbusirq == bus_irq)
2661 break;
2662 if (i >= mp_irq_entries)
2663 return -1;
2665 *trigger = irq_trigger(i);
2666 *polarity = irq_polarity(i);
2667 return 0;
2670 #endif /* CONFIG_ACPI */
2672 static int __init parse_disable_timer_pin_1(char *arg)
2674 disable_timer_pin_1 = 1;
2675 return 0;
2677 early_param("disable_timer_pin_1", parse_disable_timer_pin_1);
2679 static int __init parse_enable_timer_pin_1(char *arg)
2681 disable_timer_pin_1 = -1;
2682 return 0;
2684 early_param("enable_timer_pin_1", parse_enable_timer_pin_1);
2686 static int __init parse_noapic(char *arg)
2688 /* disable IO-APIC */
2689 disable_ioapic_setup();
2690 return 0;
2692 early_param("noapic", parse_noapic);
2694 void __init ioapic_init_mappings(void)
2696 unsigned long ioapic_phys, idx = FIX_IO_APIC_BASE_0;
2697 int i;
2699 for (i = 0; i < nr_ioapics; i++) {
2700 if (smp_found_config) {
2701 ioapic_phys = mp_ioapics[i].mp_apicaddr;
2702 if (!ioapic_phys) {
2703 printk(KERN_ERR
2704 "WARNING: bogus zero IO-APIC "
2705 "address found in MPTABLE, "
2706 "disabling IO/APIC support!\n");
2707 smp_found_config = 0;
2708 skip_ioapic_setup = 1;
2709 goto fake_ioapic_page;
2711 } else {
2712 fake_ioapic_page:
2713 ioapic_phys = (unsigned long)
2714 alloc_bootmem_pages(PAGE_SIZE);
2715 ioapic_phys = __pa(ioapic_phys);
2717 set_fixmap_nocache(idx, ioapic_phys);
2718 printk(KERN_DEBUG "mapped IOAPIC to %08lx (%08lx)\n",
2719 __fix_to_virt(idx), ioapic_phys);
2720 idx++;