2 * Intel IO-APIC support for multi-Pentium hosts.
4 * Copyright (C) 1997, 1998, 1999, 2000 Ingo Molnar, Hajnalka Szabo
6 * Many thanks to Stig Venaas for trying out countless experimental
7 * patches and reporting/debugging problems patiently!
9 * (c) 1999, Multiple IO-APIC support, developed by
10 * Ken-ichi Yaku <yaku@css1.kbnes.nec.co.jp> and
11 * Hidemi Kishimoto <kisimoto@css1.kbnes.nec.co.jp>,
12 * further tested and cleaned up by Zach Brown <zab@redhat.com>
13 * and Ingo Molnar <mingo@redhat.com>
16 * Maciej W. Rozycki : Bits for genuine 82489DX APICs;
17 * thanks to Eric Gilmore
19 * for testing these extensively
20 * Paul Diefenbaugh : Added full ACPI support
24 #include <linux/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>
31 #include <linux/kdb.h>
32 #endif /* CONFIG_KDB */
33 #include <linux/smp_lock.h>
34 #include <linux/mc146818rtc.h>
35 #include <linux/compiler.h>
36 #include <linux/acpi.h>
38 #include <linux/sysdev.h>
42 #include <asm/timer.h>
44 #include <mach_apic.h>
48 static spinlock_t ioapic_lock = SPIN_LOCK_UNLOCKED;
51 * Is the SiS APIC rmw bug present ?
52 * -1 = don't know, 0 = no, 1 = yes
54 int sis_apic_bug = -1;
57 * # of IRQ routing registers
59 int nr_ioapic_registers[MAX_IO_APICS];
62 * Rough estimation of how many shared IRQs there are, can
65 #define MAX_PLUS_SHARED_IRQS NR_IRQS
66 #define PIN_MAP_SIZE (MAX_PLUS_SHARED_IRQS + NR_IRQS)
69 * This is performance-critical, we want to do it O(1)
71 * the indexing order of this array favors 1:1 mappings
72 * between pins and IRQs.
75 static struct irq_pin_list {
77 } irq_2_pin[PIN_MAP_SIZE];
79 int vector_irq[NR_VECTORS] = { [0 ... NR_VECTORS - 1] = -1};
81 #define vector_to_irq(vector) \
82 (platform_legacy_irq(vector) ? vector : vector_irq[vector])
84 #define vector_to_irq(vector) (vector)
88 * The common case is 1:1 IRQ<->pin mappings. Sometimes there are
89 * shared ISA-space IRQs, so we have to support them. We are super
90 * fast in the common case, and fast for shared ISA-space IRQs.
92 static void __init add_pin_to_irq(unsigned int irq, int apic, int pin)
94 static int first_free_entry = NR_IRQS;
95 struct irq_pin_list *entry = irq_2_pin + irq;
98 entry = irq_2_pin + entry->next;
100 if (entry->pin != -1) {
101 entry->next = first_free_entry;
102 entry = irq_2_pin + entry->next;
103 if (++first_free_entry >= PIN_MAP_SIZE)
104 panic("io_apic.c: whoops");
111 * Reroute an IRQ to a different pin.
113 static void __init replace_pin_at_irq(unsigned int irq,
114 int oldapic, int oldpin,
115 int newapic, int newpin)
117 struct irq_pin_list *entry = irq_2_pin + irq;
120 if (entry->apic == oldapic && entry->pin == oldpin) {
121 entry->apic = newapic;
126 entry = irq_2_pin + entry->next;
130 static void __modify_IO_APIC_irq (unsigned int irq, unsigned long enable, unsigned long disable)
132 struct irq_pin_list *entry = irq_2_pin + irq;
133 unsigned int pin, reg;
139 reg = io_apic_read(entry->apic, 0x10 + pin*2);
142 io_apic_modify(entry->apic, 0x10 + pin*2, reg);
145 entry = irq_2_pin + entry->next;
150 static void __mask_IO_APIC_irq (unsigned int irq)
152 __modify_IO_APIC_irq(irq, 0x00010000, 0);
156 static void __unmask_IO_APIC_irq (unsigned int irq)
158 __modify_IO_APIC_irq(irq, 0, 0x00010000);
161 /* mask = 1, trigger = 0 */
162 static void __mask_and_edge_IO_APIC_irq (unsigned int irq)
164 __modify_IO_APIC_irq(irq, 0x00010000, 0x00008000);
167 /* mask = 0, trigger = 1 */
168 static void __unmask_and_level_IO_APIC_irq (unsigned int irq)
170 __modify_IO_APIC_irq(irq, 0x00008000, 0x00010000);
173 static void mask_IO_APIC_irq (unsigned int irq)
177 spin_lock_irqsave(&ioapic_lock, flags);
178 __mask_IO_APIC_irq(irq);
179 spin_unlock_irqrestore(&ioapic_lock, flags);
182 static void unmask_IO_APIC_irq (unsigned int irq)
186 spin_lock_irqsave(&ioapic_lock, flags);
187 __unmask_IO_APIC_irq(irq);
188 spin_unlock_irqrestore(&ioapic_lock, flags);
191 void clear_IO_APIC_pin(unsigned int apic, unsigned int pin)
193 struct IO_APIC_route_entry entry;
196 /* Check delivery_mode to be sure we're not clearing an SMI pin */
197 spin_lock_irqsave(&ioapic_lock, flags);
198 *(((int*)&entry) + 0) = io_apic_read(apic, 0x10 + 2 * pin);
199 *(((int*)&entry) + 1) = io_apic_read(apic, 0x11 + 2 * pin);
200 spin_unlock_irqrestore(&ioapic_lock, flags);
201 if (entry.delivery_mode == dest_SMI)
205 * Disable it in the IO-APIC irq-routing table:
207 memset(&entry, 0, sizeof(entry));
209 spin_lock_irqsave(&ioapic_lock, flags);
210 io_apic_write(apic, 0x10 + 2 * pin, *(((int *)&entry) + 0));
211 io_apic_write(apic, 0x11 + 2 * pin, *(((int *)&entry) + 1));
212 spin_unlock_irqrestore(&ioapic_lock, flags);
215 static void clear_IO_APIC (void)
219 for (apic = 0; apic < nr_ioapics; apic++)
220 for (pin = 0; pin < nr_ioapic_registers[apic]; pin++)
221 clear_IO_APIC_pin(apic, pin);
224 static void set_ioapic_affinity_irq(unsigned int irq, cpumask_t cpumask)
228 struct irq_pin_list *entry = irq_2_pin + irq;
229 unsigned int apicid_value;
231 apicid_value = cpu_mask_to_apicid(cpumask);
232 /* Prepare to do the io_apic_write */
233 apicid_value = apicid_value << 24;
234 spin_lock_irqsave(&ioapic_lock, flags);
239 io_apic_write(entry->apic, 0x10 + 1 + pin*2, apicid_value);
242 entry = irq_2_pin + entry->next;
244 spin_unlock_irqrestore(&ioapic_lock, flags);
247 #if defined(CONFIG_IRQBALANCE)
248 # include <asm/processor.h> /* kernel_thread() */
249 # include <linux/kernel_stat.h> /* kstat */
250 # include <linux/slab.h> /* kmalloc() */
251 # include <linux/timer.h> /* time_after() */
253 # ifdef CONFIG_BALANCED_IRQ_DEBUG
254 # define TDprintk(x...) do { printk("<%ld:%s:%d>: ", jiffies, __FILE__, __LINE__); printk(x); } while (0)
255 # define Dprintk(x...) do { TDprintk(x); } while (0)
257 # define TDprintk(x...)
258 # define Dprintk(x...)
261 extern cpumask_t irq_affinity[NR_IRQS];
263 cpumask_t __cacheline_aligned pending_irq_balance_cpumask[NR_IRQS];
265 #define IRQBALANCE_CHECK_ARCH -999
266 static int irqbalance_disabled = IRQBALANCE_CHECK_ARCH;
267 static int physical_balance = 0;
269 struct irq_cpu_info {
270 unsigned long * last_irq;
271 unsigned long * irq_delta;
273 } irq_cpu_data[NR_CPUS];
275 #define CPU_IRQ(cpu) (irq_cpu_data[cpu].irq)
276 #define LAST_CPU_IRQ(cpu,irq) (irq_cpu_data[cpu].last_irq[irq])
277 #define IRQ_DELTA(cpu,irq) (irq_cpu_data[cpu].irq_delta[irq])
279 #define IDLE_ENOUGH(cpu,now) \
280 (idle_cpu(cpu) && ((now) - irq_stat[(cpu)].idle_timestamp > 1))
282 #define IRQ_ALLOWED(cpu, allowed_mask) cpu_isset(cpu, allowed_mask)
284 #define CPU_TO_PACKAGEINDEX(i) (first_cpu(cpu_sibling_map[i]))
286 #define MAX_BALANCED_IRQ_INTERVAL (5*HZ)
287 #define MIN_BALANCED_IRQ_INTERVAL (HZ/2)
288 #define BALANCED_IRQ_MORE_DELTA (HZ/10)
289 #define BALANCED_IRQ_LESS_DELTA (HZ)
291 long balanced_irq_interval = MAX_BALANCED_IRQ_INTERVAL;
293 static unsigned long move(int curr_cpu, cpumask_t allowed_mask,
294 unsigned long now, int direction)
302 if (unlikely(cpu == curr_cpu))
305 if (direction == 1) {
314 } while (!cpu_online(cpu) || !IRQ_ALLOWED(cpu,allowed_mask) ||
315 (search_idle && !IDLE_ENOUGH(cpu,now)));
320 static inline void balance_irq(int cpu, int irq)
322 unsigned long now = jiffies;
323 cpumask_t allowed_mask;
324 unsigned int new_cpu;
326 if (irqbalance_disabled)
329 cpus_and(allowed_mask, cpu_online_map, irq_affinity[irq]);
330 new_cpu = move(cpu, allowed_mask, now, 1);
331 if (cpu != new_cpu) {
332 irq_desc_t *desc = irq_desc + irq;
335 spin_lock_irqsave(&desc->lock, flags);
336 pending_irq_balance_cpumask[irq] = cpumask_of_cpu(new_cpu);
337 spin_unlock_irqrestore(&desc->lock, flags);
341 static inline void rotate_irqs_among_cpus(unsigned long useful_load_threshold)
344 Dprintk("Rotating IRQs among CPUs.\n");
345 for (i = 0; i < NR_CPUS; i++) {
346 for (j = 0; cpu_online(i) && (j < NR_IRQS); j++) {
347 if (!irq_desc[j].action)
349 /* Is it a significant load ? */
350 if (IRQ_DELTA(CPU_TO_PACKAGEINDEX(i),j) <
351 useful_load_threshold)
356 balanced_irq_interval = max((long)MIN_BALANCED_IRQ_INTERVAL,
357 balanced_irq_interval - BALANCED_IRQ_LESS_DELTA);
361 static void do_irq_balance(void)
364 unsigned long max_cpu_irq = 0, min_cpu_irq = (~0);
365 unsigned long move_this_load = 0;
366 int max_loaded = 0, min_loaded = 0;
368 unsigned long useful_load_threshold = balanced_irq_interval + 10;
370 int tmp_loaded, first_attempt = 1;
371 unsigned long tmp_cpu_irq;
372 unsigned long imbalance = 0;
373 cpumask_t allowed_mask, target_cpu_mask, tmp;
375 for (i = 0; i < NR_CPUS; i++) {
380 package_index = CPU_TO_PACKAGEINDEX(i);
381 for (j = 0; j < NR_IRQS; j++) {
382 unsigned long value_now, delta;
383 /* Is this an active IRQ? */
384 if (!irq_desc[j].action)
386 if ( package_index == i )
387 IRQ_DELTA(package_index,j) = 0;
388 /* Determine the total count per processor per IRQ */
389 value_now = (unsigned long) kstat_cpu(i).irqs[j];
391 /* Determine the activity per processor per IRQ */
392 delta = value_now - LAST_CPU_IRQ(i,j);
394 /* Update last_cpu_irq[][] for the next time */
395 LAST_CPU_IRQ(i,j) = value_now;
397 /* Ignore IRQs whose rate is less than the clock */
398 if (delta < useful_load_threshold)
400 /* update the load for the processor or package total */
401 IRQ_DELTA(package_index,j) += delta;
403 /* Keep track of the higher numbered sibling as well */
404 if (i != package_index)
407 * We have sibling A and sibling B in the package
409 * cpu_irq[A] = load for cpu A + load for cpu B
410 * cpu_irq[B] = load for cpu B
412 CPU_IRQ(package_index) += delta;
415 /* Find the least loaded processor package */
416 for (i = 0; i < NR_CPUS; i++) {
419 if (i != CPU_TO_PACKAGEINDEX(i))
421 if (min_cpu_irq > CPU_IRQ(i)) {
422 min_cpu_irq = CPU_IRQ(i);
426 max_cpu_irq = ULONG_MAX;
429 /* Look for heaviest loaded processor.
430 * We may come back to get the next heaviest loaded processor.
431 * Skip processors with trivial loads.
435 for (i = 0; i < NR_CPUS; i++) {
438 if (i != CPU_TO_PACKAGEINDEX(i))
440 if (max_cpu_irq <= CPU_IRQ(i))
442 if (tmp_cpu_irq < CPU_IRQ(i)) {
443 tmp_cpu_irq = CPU_IRQ(i);
448 if (tmp_loaded == -1) {
449 /* In the case of small number of heavy interrupt sources,
450 * loading some of the cpus too much. We use Ingo's original
451 * approach to rotate them around.
453 if (!first_attempt && imbalance >= useful_load_threshold) {
454 rotate_irqs_among_cpus(useful_load_threshold);
457 goto not_worth_the_effort;
460 first_attempt = 0; /* heaviest search */
461 max_cpu_irq = tmp_cpu_irq; /* load */
462 max_loaded = tmp_loaded; /* processor */
463 imbalance = (max_cpu_irq - min_cpu_irq) / 2;
465 Dprintk("max_loaded cpu = %d\n", max_loaded);
466 Dprintk("min_loaded cpu = %d\n", min_loaded);
467 Dprintk("max_cpu_irq load = %ld\n", max_cpu_irq);
468 Dprintk("min_cpu_irq load = %ld\n", min_cpu_irq);
469 Dprintk("load imbalance = %lu\n", imbalance);
471 /* if imbalance is less than approx 10% of max load, then
472 * observe diminishing returns action. - quit
474 if (imbalance < (max_cpu_irq >> 3)) {
475 Dprintk("Imbalance too trivial\n");
476 goto not_worth_the_effort;
480 /* if we select an IRQ to move that can't go where we want, then
481 * see if there is another one to try.
485 for (j = 0; j < NR_IRQS; j++) {
486 /* Is this an active IRQ? */
487 if (!irq_desc[j].action)
489 if (imbalance <= IRQ_DELTA(max_loaded,j))
491 /* Try to find the IRQ that is closest to the imbalance
492 * without going over.
494 if (move_this_load < IRQ_DELTA(max_loaded,j)) {
495 move_this_load = IRQ_DELTA(max_loaded,j);
499 if (selected_irq == -1) {
503 imbalance = move_this_load;
505 /* For physical_balance case, we accumlated both load
506 * values in the one of the siblings cpu_irq[],
507 * to use the same code for physical and logical processors
508 * as much as possible.
510 * NOTE: the cpu_irq[] array holds the sum of the load for
511 * sibling A and sibling B in the slot for the lowest numbered
512 * sibling (A), _AND_ the load for sibling B in the slot for
513 * the higher numbered sibling.
515 * We seek the least loaded sibling by making the comparison
518 load = CPU_IRQ(min_loaded) >> 1;
519 for_each_cpu_mask(j, cpu_sibling_map[min_loaded]) {
520 if (load > CPU_IRQ(j)) {
521 /* This won't change cpu_sibling_map[min_loaded] */
527 cpus_and(allowed_mask, cpu_online_map, irq_affinity[selected_irq]);
528 target_cpu_mask = cpumask_of_cpu(min_loaded);
529 cpus_and(tmp, target_cpu_mask, allowed_mask);
531 if (!cpus_empty(tmp)) {
532 irq_desc_t *desc = irq_desc + selected_irq;
535 Dprintk("irq = %d moved to cpu = %d\n",
536 selected_irq, min_loaded);
537 /* mark for change destination */
538 spin_lock_irqsave(&desc->lock, flags);
539 pending_irq_balance_cpumask[selected_irq] =
540 cpumask_of_cpu(min_loaded);
541 spin_unlock_irqrestore(&desc->lock, flags);
542 /* Since we made a change, come back sooner to
543 * check for more variation.
545 balanced_irq_interval = max((long)MIN_BALANCED_IRQ_INTERVAL,
546 balanced_irq_interval - BALANCED_IRQ_LESS_DELTA);
551 not_worth_the_effort:
553 * if we did not find an IRQ to move, then adjust the time interval
556 balanced_irq_interval = min((long)MAX_BALANCED_IRQ_INTERVAL,
557 balanced_irq_interval + BALANCED_IRQ_MORE_DELTA);
558 Dprintk("IRQ worth rotating not found\n");
562 static int balanced_irq(void *unused)
565 unsigned long prev_balance_time = jiffies;
566 long time_remaining = balanced_irq_interval;
570 /* push everything to CPU 0 to give us a starting point. */
571 for (i = 0 ; i < NR_IRQS ; i++) {
572 pending_irq_balance_cpumask[i] = cpumask_of_cpu(0);
576 set_current_state(TASK_INTERRUPTIBLE);
577 time_remaining = schedule_timeout(time_remaining);
578 if (time_after(jiffies,
579 prev_balance_time+balanced_irq_interval)) {
581 prev_balance_time = jiffies;
582 time_remaining = balanced_irq_interval;
588 static int __init balanced_irq_init(void)
591 struct cpuinfo_x86 *c;
594 cpus_shift_right(tmp, cpu_online_map, 2);
596 /* When not overwritten by the command line ask subarchitecture. */
597 if (irqbalance_disabled == IRQBALANCE_CHECK_ARCH)
598 irqbalance_disabled = NO_BALANCE_IRQ;
599 if (irqbalance_disabled)
602 /* disable irqbalance completely if there is only one processor online */
603 if (num_online_cpus() < 2) {
604 irqbalance_disabled = 1;
608 * Enable physical balance only if more than 1 physical processor
611 if (smp_num_siblings > 1 && !cpus_empty(tmp))
612 physical_balance = 1;
614 for (i = 0; i < NR_CPUS; i++) {
617 irq_cpu_data[i].irq_delta = kmalloc(sizeof(unsigned long) * NR_IRQS, GFP_KERNEL);
618 irq_cpu_data[i].last_irq = kmalloc(sizeof(unsigned long) * NR_IRQS, GFP_KERNEL);
619 if (irq_cpu_data[i].irq_delta == NULL || irq_cpu_data[i].last_irq == NULL) {
620 printk(KERN_ERR "balanced_irq_init: out of memory");
623 memset(irq_cpu_data[i].irq_delta,0,sizeof(unsigned long) * NR_IRQS);
624 memset(irq_cpu_data[i].last_irq,0,sizeof(unsigned long) * NR_IRQS);
627 printk(KERN_INFO "Starting balanced_irq\n");
628 if (kernel_thread(balanced_irq, NULL, CLONE_KERNEL) >= 0)
631 printk(KERN_ERR "balanced_irq_init: failed to spawn balanced_irq");
633 for (i = 0; i < NR_CPUS; i++) {
634 if(irq_cpu_data[i].irq_delta)
635 kfree(irq_cpu_data[i].irq_delta);
636 if(irq_cpu_data[i].last_irq)
637 kfree(irq_cpu_data[i].last_irq);
642 static int __init irqbalance_disable(char *str)
644 irqbalance_disabled = 1;
648 __setup("noirqbalance", irqbalance_disable);
650 static inline void move_irq(int irq)
652 /* note - we hold the desc->lock */
653 if (unlikely(!cpus_empty(pending_irq_balance_cpumask[irq]))) {
654 set_ioapic_affinity_irq(irq, pending_irq_balance_cpumask[irq]);
655 cpus_clear(pending_irq_balance_cpumask[irq]);
659 __initcall(balanced_irq_init);
661 #else /* !CONFIG_IRQBALANCE */
662 static inline void move_irq(int irq) { }
663 #endif /* CONFIG_IRQBALANCE */
666 void fastcall send_IPI_self(int vector)
673 apic_wait_icr_idle();
674 cfg = APIC_DM_FIXED | APIC_DEST_SELF | vector | APIC_DEST_LOGICAL;
676 * Send the IPI. The write to APIC_ICR fires this off.
678 apic_write_around(APIC_ICR, cfg);
680 #endif /* !CONFIG_SMP */
684 * support for broken MP BIOSs, enables hand-redirection of PIRQ0-7 to
685 * specific CPU-side IRQs.
689 int pirq_entries [MAX_PIRQS];
692 int skip_ioapic_setup = 0;
694 int skip_ioapic_setup = 1;
697 static int __init ioapic_setup(char *str)
699 skip_ioapic_setup = 1;
703 __setup("noapic", ioapic_setup);
705 static int __init ioapic_pirq_setup(char *str)
708 int ints[MAX_PIRQS+1];
710 get_options(str, ARRAY_SIZE(ints), ints);
712 for (i = 0; i < MAX_PIRQS; i++)
713 pirq_entries[i] = -1;
716 printk(KERN_INFO "PIRQ redirection, working around broken MP-BIOS.\n");
718 if (ints[0] < MAX_PIRQS)
721 for (i = 0; i < max; i++) {
722 printk(KERN_DEBUG "... PIRQ%d -> IRQ %d\n", i, ints[i+1]);
724 * PIRQs are mapped upside down, usually.
726 pirq_entries[MAX_PIRQS-i-1] = ints[i+1];
731 __setup("pirq=", ioapic_pirq_setup);
734 * Find the IRQ entry number of a certain pin.
736 static int __init find_irq_entry(int apic, int pin, int type)
740 for (i = 0; i < mp_irq_entries; i++)
741 if (mp_irqs[i].mpc_irqtype == type &&
742 (mp_irqs[i].mpc_dstapic == mp_ioapics[apic].mpc_apicid ||
743 mp_irqs[i].mpc_dstapic == MP_APIC_ALL) &&
744 mp_irqs[i].mpc_dstirq == pin)
751 * Find the pin to which IRQ[irq] (ISA) is connected
753 static int __init find_isa_irq_pin(int irq, int type)
757 for (i = 0; i < mp_irq_entries; i++) {
758 int lbus = mp_irqs[i].mpc_srcbus;
760 if ((mp_bus_id_to_type[lbus] == MP_BUS_ISA ||
761 mp_bus_id_to_type[lbus] == MP_BUS_EISA ||
762 mp_bus_id_to_type[lbus] == MP_BUS_MCA ||
763 mp_bus_id_to_type[lbus] == MP_BUS_NEC98
765 (mp_irqs[i].mpc_irqtype == type) &&
766 (mp_irqs[i].mpc_srcbusirq == irq))
768 return mp_irqs[i].mpc_dstirq;
774 * Find a specific PCI IRQ entry.
775 * Not an __init, possibly needed by modules
777 static int pin_2_irq(int idx, int apic, int pin);
779 int IO_APIC_get_PCI_irq_vector(int bus, int slot, int pin)
781 int apic, i, best_guess = -1;
783 Dprintk("querying PCI -> IRQ mapping bus:%d, slot:%d, pin:%d.\n",
785 if (mp_bus_id_to_pci_bus[bus] == -1) {
786 printk(KERN_WARNING "PCI BIOS passed nonexistent PCI bus %d!\n", bus);
789 for (i = 0; i < mp_irq_entries; i++) {
790 int lbus = mp_irqs[i].mpc_srcbus;
792 for (apic = 0; apic < nr_ioapics; apic++)
793 if (mp_ioapics[apic].mpc_apicid == mp_irqs[i].mpc_dstapic ||
794 mp_irqs[i].mpc_dstapic == MP_APIC_ALL)
797 if ((mp_bus_id_to_type[lbus] == MP_BUS_PCI) &&
798 !mp_irqs[i].mpc_irqtype &&
800 (slot == ((mp_irqs[i].mpc_srcbusirq >> 2) & 0x1f))) {
801 int irq = pin_2_irq(i,apic,mp_irqs[i].mpc_dstirq);
803 if (!(apic || IO_APIC_IRQ(irq)))
806 if (pin == (mp_irqs[i].mpc_srcbusirq & 3))
809 * Use the first all-but-pin matching entry as a
810 * best-guess fuzzy result for broken mptables.
820 * This function currently is only a helper for the i386 smp boot process where
821 * we need to reprogram the ioredtbls to cater for the cpus which have come online
822 * so mask in all cases should simply be TARGET_CPUS
824 void __init setup_ioapic_dest(void)
826 int pin, ioapic, irq, irq_entry;
828 if (skip_ioapic_setup == 1)
831 for (ioapic = 0; ioapic < nr_ioapics; ioapic++) {
832 for (pin = 0; pin < nr_ioapic_registers[ioapic]; pin++) {
833 irq_entry = find_irq_entry(ioapic, pin, mp_INT);
836 irq = pin_2_irq(irq_entry, ioapic, pin);
837 set_ioapic_affinity_irq(irq, TARGET_CPUS);
844 * EISA Edge/Level control register, ELCR
846 static int __init EISA_ELCR(unsigned int irq)
849 unsigned int port = 0x4d0 + (irq >> 3);
850 return (inb(port) >> (irq & 7)) & 1;
852 printk(KERN_INFO "Broken MPtable reports ISA irq %d\n", irq);
856 /* EISA interrupts are always polarity zero and can be edge or level
857 * trigger depending on the ELCR value. If an interrupt is listed as
858 * EISA conforming in the MP table, that means its trigger type must
859 * be read in from the ELCR */
861 #define default_EISA_trigger(idx) (EISA_ELCR(mp_irqs[idx].mpc_srcbusirq))
862 #define default_EISA_polarity(idx) (0)
864 /* ISA interrupts are always polarity zero edge triggered,
865 * when listed as conforming in the MP table. */
867 #define default_ISA_trigger(idx) (0)
868 #define default_ISA_polarity(idx) (0)
870 /* PCI interrupts are always polarity one level triggered,
871 * when listed as conforming in the MP table. */
873 #define default_PCI_trigger(idx) (1)
874 #define default_PCI_polarity(idx) (1)
876 /* MCA interrupts are always polarity zero level triggered,
877 * when listed as conforming in the MP table. */
879 #define default_MCA_trigger(idx) (1)
880 #define default_MCA_polarity(idx) (0)
882 /* NEC98 interrupts are always polarity zero edge triggered,
883 * when listed as conforming in the MP table. */
885 #define default_NEC98_trigger(idx) (0)
886 #define default_NEC98_polarity(idx) (0)
888 static int __init MPBIOS_polarity(int idx)
890 int bus = mp_irqs[idx].mpc_srcbus;
894 * Determine IRQ line polarity (high active or low active):
896 switch (mp_irqs[idx].mpc_irqflag & 3)
898 case 0: /* conforms, ie. bus-type dependent polarity */
900 switch (mp_bus_id_to_type[bus])
902 case MP_BUS_ISA: /* ISA pin */
904 polarity = default_ISA_polarity(idx);
907 case MP_BUS_EISA: /* EISA pin */
909 polarity = default_EISA_polarity(idx);
912 case MP_BUS_PCI: /* PCI pin */
914 polarity = default_PCI_polarity(idx);
917 case MP_BUS_MCA: /* MCA pin */
919 polarity = default_MCA_polarity(idx);
922 case MP_BUS_NEC98: /* NEC 98 pin */
924 polarity = default_NEC98_polarity(idx);
929 printk(KERN_WARNING "broken BIOS!!\n");
936 case 1: /* high active */
941 case 2: /* reserved */
943 printk(KERN_WARNING "broken BIOS!!\n");
947 case 3: /* low active */
952 default: /* invalid */
954 printk(KERN_WARNING "broken BIOS!!\n");
962 static int __init MPBIOS_trigger(int idx)
964 int bus = mp_irqs[idx].mpc_srcbus;
968 * Determine IRQ trigger mode (edge or level sensitive):
970 switch ((mp_irqs[idx].mpc_irqflag>>2) & 3)
972 case 0: /* conforms, ie. bus-type dependent */
974 switch (mp_bus_id_to_type[bus])
976 case MP_BUS_ISA: /* ISA pin */
978 trigger = default_ISA_trigger(idx);
981 case MP_BUS_EISA: /* EISA pin */
983 trigger = default_EISA_trigger(idx);
986 case MP_BUS_PCI: /* PCI pin */
988 trigger = default_PCI_trigger(idx);
991 case MP_BUS_MCA: /* MCA pin */
993 trigger = default_MCA_trigger(idx);
996 case MP_BUS_NEC98: /* NEC 98 pin */
998 trigger = default_NEC98_trigger(idx);
1003 printk(KERN_WARNING "broken BIOS!!\n");
1015 case 2: /* reserved */
1017 printk(KERN_WARNING "broken BIOS!!\n");
1026 default: /* invalid */
1028 printk(KERN_WARNING "broken BIOS!!\n");
1036 static inline int irq_polarity(int idx)
1038 return MPBIOS_polarity(idx);
1041 static inline int irq_trigger(int idx)
1043 return MPBIOS_trigger(idx);
1046 static int pin_2_irq(int idx, int apic, int pin)
1049 int bus = mp_irqs[idx].mpc_srcbus;
1052 * Debugging check, we are in big trouble if this message pops up!
1054 if (mp_irqs[idx].mpc_dstirq != pin)
1055 printk(KERN_ERR "broken BIOS or MPTABLE parser, ayiee!!\n");
1057 switch (mp_bus_id_to_type[bus])
1059 case MP_BUS_ISA: /* ISA pin */
1064 irq = mp_irqs[idx].mpc_srcbusirq;
1067 case MP_BUS_PCI: /* PCI pin */
1070 * PCI IRQs are mapped in order
1074 irq += nr_ioapic_registers[i++];
1076 if ((!apic) && (irq < 16))
1082 printk(KERN_ERR "unknown bus type %d.\n",bus);
1089 * PCI IRQ command line redirection. Yes, limits are hardcoded.
1091 if ((pin >= 16) && (pin <= 23)) {
1092 if (pirq_entries[pin-16] != -1) {
1093 if (!pirq_entries[pin-16]) {
1094 printk(KERN_DEBUG "disabling PIRQ%d\n", pin-16);
1096 irq = pirq_entries[pin-16];
1097 printk(KERN_DEBUG "using PIRQ%d -> IRQ %d\n",
1105 static inline int IO_APIC_irq_trigger(int irq)
1109 for (apic = 0; apic < nr_ioapics; apic++) {
1110 for (pin = 0; pin < nr_ioapic_registers[apic]; pin++) {
1111 idx = find_irq_entry(apic,pin,mp_INT);
1112 if ((idx != -1) && (irq == pin_2_irq(idx,apic,pin)))
1113 return irq_trigger(idx);
1117 * nonexistent IRQs are edge default
1122 /* irq_vectors is indexed by the sum of all RTEs in all I/O APICs. */
1123 u8 irq_vector[NR_IRQ_VECTORS] = { FIRST_DEVICE_VECTOR , 0 };
1125 #ifdef CONFIG_PCI_MSI
1126 int assign_irq_vector(int irq)
1128 int __init assign_irq_vector(int irq)
1131 static int current_vector = FIRST_DEVICE_VECTOR, offset = 0;
1133 BUG_ON(irq >= NR_IRQ_VECTORS);
1134 if (irq != AUTO_ASSIGN && IO_APIC_VECTOR(irq) > 0)
1135 return IO_APIC_VECTOR(irq);
1137 current_vector += 8;
1138 if (current_vector == SYSCALL_VECTOR)
1141 if (current_vector == KDBENTER_VECTOR)
1143 #endif /* CONFIG_KDB */
1145 if (current_vector >= FIRST_SYSTEM_VECTOR) {
1149 current_vector = FIRST_DEVICE_VECTOR + offset;
1152 vector_irq[current_vector] = irq;
1153 if (irq != AUTO_ASSIGN)
1154 IO_APIC_VECTOR(irq) = current_vector;
1156 return current_vector;
1159 static struct hw_interrupt_type ioapic_level_type;
1160 static struct hw_interrupt_type ioapic_edge_type;
1162 #define IOAPIC_AUTO -1
1163 #define IOAPIC_EDGE 0
1164 #define IOAPIC_LEVEL 1
1166 static inline void ioapic_register_intr(int irq, int vector, unsigned long trigger)
1168 if (use_pci_vector() && !platform_legacy_irq(irq)) {
1169 if ((trigger == IOAPIC_AUTO && IO_APIC_irq_trigger(irq)) ||
1170 trigger == IOAPIC_LEVEL)
1171 irq_desc[vector].handler = &ioapic_level_type;
1173 irq_desc[vector].handler = &ioapic_edge_type;
1174 set_intr_gate(vector, interrupt[vector]);
1176 if ((trigger == IOAPIC_AUTO && IO_APIC_irq_trigger(irq)) ||
1177 trigger == IOAPIC_LEVEL)
1178 irq_desc[irq].handler = &ioapic_level_type;
1180 irq_desc[irq].handler = &ioapic_edge_type;
1181 set_intr_gate(vector, interrupt[irq]);
1185 void __init setup_IO_APIC_irqs(void)
1187 struct IO_APIC_route_entry entry;
1188 int apic, pin, idx, irq, first_notcon = 1, vector;
1189 unsigned long flags;
1191 printk(KERN_DEBUG "init IO_APIC IRQs\n");
1193 for (apic = 0; apic < nr_ioapics; apic++) {
1194 for (pin = 0; pin < nr_ioapic_registers[apic]; pin++) {
1197 * add it to the IO-APIC irq-routing table:
1199 memset(&entry,0,sizeof(entry));
1201 entry.delivery_mode = INT_DELIVERY_MODE;
1202 entry.dest_mode = INT_DEST_MODE;
1203 entry.mask = 0; /* enable IRQ */
1204 entry.dest.logical.logical_dest =
1205 cpu_mask_to_apicid(TARGET_CPUS);
1207 idx = find_irq_entry(apic,pin,mp_INT);
1210 printk(KERN_DEBUG " IO-APIC (apicid-pin) %d-%d", mp_ioapics[apic].mpc_apicid, pin);
1213 printk(", %d-%d", mp_ioapics[apic].mpc_apicid, pin);
1217 entry.trigger = irq_trigger(idx);
1218 entry.polarity = irq_polarity(idx);
1220 if (irq_trigger(idx)) {
1225 irq = pin_2_irq(idx, apic, pin);
1227 * skip adding the timer int on secondary nodes, which causes
1228 * a small but painful rift in the time-space continuum
1230 if (multi_timer_check(apic, irq))
1233 add_pin_to_irq(irq, apic, pin);
1235 if (!apic && !IO_APIC_IRQ(irq))
1238 if (IO_APIC_IRQ(irq)) {
1239 vector = assign_irq_vector(irq);
1240 entry.vector = vector;
1241 ioapic_register_intr(irq, vector, IOAPIC_AUTO);
1243 if (!apic && (irq < 16))
1244 disable_8259A_irq(irq);
1246 spin_lock_irqsave(&ioapic_lock, flags);
1247 io_apic_write(apic, 0x11+2*pin, *(((int *)&entry)+1));
1248 io_apic_write(apic, 0x10+2*pin, *(((int *)&entry)+0));
1249 spin_unlock_irqrestore(&ioapic_lock, flags);
1254 printk(" not connected.\n");
1258 * Set up the 8259A-master output pin:
1260 void __init setup_ExtINT_IRQ0_pin(unsigned int pin, int vector)
1262 struct IO_APIC_route_entry entry;
1263 unsigned long flags;
1265 memset(&entry,0,sizeof(entry));
1267 disable_8259A_irq(0);
1270 apic_write_around(APIC_LVT0, APIC_LVT_MASKED | APIC_DM_EXTINT);
1273 * We use logical delivery to get the timer IRQ
1276 entry.dest_mode = INT_DEST_MODE;
1277 entry.mask = 0; /* unmask IRQ now */
1278 entry.dest.logical.logical_dest = cpu_mask_to_apicid(TARGET_CPUS);
1279 entry.delivery_mode = INT_DELIVERY_MODE;
1282 entry.vector = vector;
1285 * The timer IRQ doesn't have to know that behind the
1286 * scene we have a 8259A-master in AEOI mode ...
1288 irq_desc[0].handler = &ioapic_edge_type;
1291 * Add it to the IO-APIC irq-routing table:
1293 spin_lock_irqsave(&ioapic_lock, flags);
1294 io_apic_write(0, 0x11+2*pin, *(((int *)&entry)+1));
1295 io_apic_write(0, 0x10+2*pin, *(((int *)&entry)+0));
1296 spin_unlock_irqrestore(&ioapic_lock, flags);
1298 enable_8259A_irq(0);
1301 static inline void UNEXPECTED_IO_APIC(void)
1305 void __init print_IO_APIC(void)
1308 union IO_APIC_reg_00 reg_00;
1309 union IO_APIC_reg_01 reg_01;
1310 union IO_APIC_reg_02 reg_02;
1311 union IO_APIC_reg_03 reg_03;
1312 unsigned long flags;
1314 printk(KERN_DEBUG "number of MP IRQ sources: %d.\n", mp_irq_entries);
1315 for (i = 0; i < nr_ioapics; i++)
1316 printk(KERN_DEBUG "number of IO-APIC #%d registers: %d.\n",
1317 mp_ioapics[i].mpc_apicid, nr_ioapic_registers[i]);
1320 * We are a bit conservative about what we expect. We have to
1321 * know about every hardware change ASAP.
1323 printk(KERN_INFO "testing the IO APIC.......................\n");
1325 for (apic = 0; apic < nr_ioapics; apic++) {
1327 spin_lock_irqsave(&ioapic_lock, flags);
1328 reg_00.raw = io_apic_read(apic, 0);
1329 reg_01.raw = io_apic_read(apic, 1);
1330 if (reg_01.bits.version >= 0x10)
1331 reg_02.raw = io_apic_read(apic, 2);
1332 if (reg_01.bits.version >= 0x20)
1333 reg_03.raw = io_apic_read(apic, 3);
1334 spin_unlock_irqrestore(&ioapic_lock, flags);
1336 printk(KERN_DEBUG "IO APIC #%d......\n", mp_ioapics[apic].mpc_apicid);
1337 printk(KERN_DEBUG ".... register #00: %08X\n", reg_00.raw);
1338 printk(KERN_DEBUG "....... : physical APIC id: %02X\n", reg_00.bits.ID);
1339 printk(KERN_DEBUG "....... : Delivery Type: %X\n", reg_00.bits.delivery_type);
1340 printk(KERN_DEBUG "....... : LTS : %X\n", reg_00.bits.LTS);
1341 if (reg_00.bits.ID >= get_physical_broadcast())
1342 UNEXPECTED_IO_APIC();
1343 if (reg_00.bits.__reserved_1 || reg_00.bits.__reserved_2)
1344 UNEXPECTED_IO_APIC();
1346 printk(KERN_DEBUG ".... register #01: %08X\n", reg_01.raw);
1347 printk(KERN_DEBUG "....... : max redirection entries: %04X\n", reg_01.bits.entries);
1348 if ( (reg_01.bits.entries != 0x0f) && /* older (Neptune) boards */
1349 (reg_01.bits.entries != 0x17) && /* typical ISA+PCI boards */
1350 (reg_01.bits.entries != 0x1b) && /* Compaq Proliant boards */
1351 (reg_01.bits.entries != 0x1f) && /* dual Xeon boards */
1352 (reg_01.bits.entries != 0x22) && /* bigger Xeon boards */
1353 (reg_01.bits.entries != 0x2E) &&
1354 (reg_01.bits.entries != 0x3F)
1356 UNEXPECTED_IO_APIC();
1358 printk(KERN_DEBUG "....... : PRQ implemented: %X\n", reg_01.bits.PRQ);
1359 printk(KERN_DEBUG "....... : IO APIC version: %04X\n", reg_01.bits.version);
1360 if ( (reg_01.bits.version != 0x01) && /* 82489DX IO-APICs */
1361 (reg_01.bits.version != 0x10) && /* oldest IO-APICs */
1362 (reg_01.bits.version != 0x11) && /* Pentium/Pro IO-APICs */
1363 (reg_01.bits.version != 0x13) && /* Xeon IO-APICs */
1364 (reg_01.bits.version != 0x20) /* Intel P64H (82806 AA) */
1366 UNEXPECTED_IO_APIC();
1367 if (reg_01.bits.__reserved_1 || reg_01.bits.__reserved_2)
1368 UNEXPECTED_IO_APIC();
1371 * Some Intel chipsets with IO APIC VERSION of 0x1? don't have reg_02,
1372 * but the value of reg_02 is read as the previous read register
1373 * value, so ignore it if reg_02 == reg_01.
1375 if (reg_01.bits.version >= 0x10 && reg_02.raw != reg_01.raw) {
1376 printk(KERN_DEBUG ".... register #02: %08X\n", reg_02.raw);
1377 printk(KERN_DEBUG "....... : arbitration: %02X\n", reg_02.bits.arbitration);
1378 if (reg_02.bits.__reserved_1 || reg_02.bits.__reserved_2)
1379 UNEXPECTED_IO_APIC();
1383 * Some Intel chipsets with IO APIC VERSION of 0x2? don't have reg_02
1384 * or reg_03, but the value of reg_0[23] is read as the previous read
1385 * register value, so ignore it if reg_03 == reg_0[12].
1387 if (reg_01.bits.version >= 0x20 && reg_03.raw != reg_02.raw &&
1388 reg_03.raw != reg_01.raw) {
1389 printk(KERN_DEBUG ".... register #03: %08X\n", reg_03.raw);
1390 printk(KERN_DEBUG "....... : Boot DT : %X\n", reg_03.bits.boot_DT);
1391 if (reg_03.bits.__reserved_1)
1392 UNEXPECTED_IO_APIC();
1395 printk(KERN_DEBUG ".... IRQ redirection table:\n");
1397 printk(KERN_DEBUG " NR Log Phy Mask Trig IRR Pol"
1398 " Stat Dest Deli Vect: \n");
1400 for (i = 0; i <= reg_01.bits.entries; i++) {
1401 struct IO_APIC_route_entry entry;
1403 spin_lock_irqsave(&ioapic_lock, flags);
1404 *(((int *)&entry)+0) = io_apic_read(apic, 0x10+i*2);
1405 *(((int *)&entry)+1) = io_apic_read(apic, 0x11+i*2);
1406 spin_unlock_irqrestore(&ioapic_lock, flags);
1408 printk(KERN_DEBUG " %02x %03X %02X ",
1410 entry.dest.logical.logical_dest,
1411 entry.dest.physical.physical_dest
1414 printk("%1d %1d %1d %1d %1d %1d %1d %02X\n",
1419 entry.delivery_status,
1421 entry.delivery_mode,
1426 if (use_pci_vector())
1427 printk(KERN_INFO "Using vector-based indexing\n");
1428 printk(KERN_DEBUG "IRQ to pin mappings:\n");
1429 for (i = 0; i < NR_IRQS; i++) {
1430 struct irq_pin_list *entry = irq_2_pin + i;
1433 if (use_pci_vector() && !platform_legacy_irq(i))
1434 printk(KERN_DEBUG "IRQ%d ", IO_APIC_VECTOR(i));
1436 printk(KERN_DEBUG "IRQ%d ", i);
1438 printk("-> %d:%d", entry->apic, entry->pin);
1441 entry = irq_2_pin + entry->next;
1446 printk(KERN_INFO ".................................... done.\n");
1451 static void print_APIC_bitfield (int base)
1456 printk(KERN_DEBUG "0123456789abcdef0123456789abcdef\n" KERN_DEBUG);
1457 for (i = 0; i < 8; i++) {
1458 v = apic_read(base + i*0x10);
1459 for (j = 0; j < 32; j++) {
1469 void /*__init*/ print_local_APIC(void * dummy)
1471 unsigned int v, ver, maxlvt;
1473 printk("\n" KERN_DEBUG "printing local APIC contents on CPU#%d/%d:\n",
1474 smp_processor_id(), hard_smp_processor_id());
1475 v = apic_read(APIC_ID);
1476 printk(KERN_INFO "... APIC ID: %08x (%01x)\n", v, GET_APIC_ID(v));
1477 v = apic_read(APIC_LVR);
1478 printk(KERN_INFO "... APIC VERSION: %08x\n", v);
1479 ver = GET_APIC_VERSION(v);
1480 maxlvt = get_maxlvt();
1482 v = apic_read(APIC_TASKPRI);
1483 printk(KERN_DEBUG "... APIC TASKPRI: %08x (%02x)\n", v, v & APIC_TPRI_MASK);
1485 if (APIC_INTEGRATED(ver)) { /* !82489DX */
1486 v = apic_read(APIC_ARBPRI);
1487 printk(KERN_DEBUG "... APIC ARBPRI: %08x (%02x)\n", v,
1488 v & APIC_ARBPRI_MASK);
1489 v = apic_read(APIC_PROCPRI);
1490 printk(KERN_DEBUG "... APIC PROCPRI: %08x\n", v);
1493 v = apic_read(APIC_EOI);
1494 printk(KERN_DEBUG "... APIC EOI: %08x\n", v);
1495 v = apic_read(APIC_RRR);
1496 printk(KERN_DEBUG "... APIC RRR: %08x\n", v);
1497 v = apic_read(APIC_LDR);
1498 printk(KERN_DEBUG "... APIC LDR: %08x\n", v);
1499 v = apic_read(APIC_DFR);
1500 printk(KERN_DEBUG "... APIC DFR: %08x\n", v);
1501 v = apic_read(APIC_SPIV);
1502 printk(KERN_DEBUG "... APIC SPIV: %08x\n", v);
1504 printk(KERN_DEBUG "... APIC ISR field:\n");
1505 print_APIC_bitfield(APIC_ISR);
1506 printk(KERN_DEBUG "... APIC TMR field:\n");
1507 print_APIC_bitfield(APIC_TMR);
1508 printk(KERN_DEBUG "... APIC IRR field:\n");
1509 print_APIC_bitfield(APIC_IRR);
1511 if (APIC_INTEGRATED(ver)) { /* !82489DX */
1512 if (maxlvt > 3) /* Due to the Pentium erratum 3AP. */
1513 apic_write(APIC_ESR, 0);
1514 v = apic_read(APIC_ESR);
1515 printk(KERN_DEBUG "... APIC ESR: %08x\n", v);
1518 v = apic_read(APIC_ICR);
1519 printk(KERN_DEBUG "... APIC ICR: %08x\n", v);
1520 v = apic_read(APIC_ICR2);
1521 printk(KERN_DEBUG "... APIC ICR2: %08x\n", v);
1523 v = apic_read(APIC_LVTT);
1524 printk(KERN_DEBUG "... APIC LVTT: %08x\n", v);
1526 if (maxlvt > 3) { /* PC is LVT#4. */
1527 v = apic_read(APIC_LVTPC);
1528 printk(KERN_DEBUG "... APIC LVTPC: %08x\n", v);
1530 v = apic_read(APIC_LVT0);
1531 printk(KERN_DEBUG "... APIC LVT0: %08x\n", v);
1532 v = apic_read(APIC_LVT1);
1533 printk(KERN_DEBUG "... APIC LVT1: %08x\n", v);
1535 if (maxlvt > 2) { /* ERR is LVT#3. */
1536 v = apic_read(APIC_LVTERR);
1537 printk(KERN_DEBUG "... APIC LVTERR: %08x\n", v);
1540 v = apic_read(APIC_TMICT);
1541 printk(KERN_DEBUG "... APIC TMICT: %08x\n", v);
1542 v = apic_read(APIC_TMCCT);
1543 printk(KERN_DEBUG "... APIC TMCCT: %08x\n", v);
1544 v = apic_read(APIC_TDCR);
1545 printk(KERN_DEBUG "... APIC TDCR: %08x\n", v);
1549 void print_all_local_APICs (void)
1551 on_each_cpu(print_local_APIC, NULL, 1, 1);
1554 void /*__init*/ print_PIC(void)
1556 extern spinlock_t i8259A_lock;
1558 unsigned long flags;
1560 printk(KERN_DEBUG "\nprinting PIC contents\n");
1562 spin_lock_irqsave(&i8259A_lock, flags);
1564 v = inb(0xa1) << 8 | inb(0x21);
1565 printk(KERN_DEBUG "... PIC IMR: %04x\n", v);
1567 v = inb(0xa0) << 8 | inb(0x20);
1568 printk(KERN_DEBUG "... PIC IRR: %04x\n", v);
1572 v = inb(0xa0) << 8 | inb(0x20);
1576 spin_unlock_irqrestore(&i8259A_lock, flags);
1578 printk(KERN_DEBUG "... PIC ISR: %04x\n", v);
1580 v = inb(0x4d1) << 8 | inb(0x4d0);
1581 printk(KERN_DEBUG "... PIC ELCR: %04x\n", v);
1584 static void __init enable_IO_APIC(void)
1586 union IO_APIC_reg_01 reg_01;
1588 unsigned long flags;
1590 for (i = 0; i < PIN_MAP_SIZE; i++) {
1591 irq_2_pin[i].pin = -1;
1592 irq_2_pin[i].next = 0;
1595 for (i = 0; i < MAX_PIRQS; i++)
1596 pirq_entries[i] = -1;
1599 * The number of IO-APIC IRQ registers (== #pins):
1601 for (i = 0; i < nr_ioapics; i++) {
1602 spin_lock_irqsave(&ioapic_lock, flags);
1603 reg_01.raw = io_apic_read(i, 1);
1604 spin_unlock_irqrestore(&ioapic_lock, flags);
1605 nr_ioapic_registers[i] = reg_01.bits.entries+1;
1609 * Do not trust the IO-APIC being empty at bootup
1615 * Not an __init, needed by the reboot code
1617 void disable_IO_APIC(void)
1620 * Clear the IO-APIC before rebooting:
1624 disconnect_bsp_APIC();
1628 * function to set the IO-APIC physical IDs based on the
1629 * values stored in the MPC table.
1631 * by Matt Domsch <Matt_Domsch@dell.com> Tue Dec 21 12:25:05 CST 1999
1634 #ifndef CONFIG_X86_NUMAQ
1635 static void __init setup_ioapic_ids_from_mpc(void)
1637 union IO_APIC_reg_00 reg_00;
1638 physid_mask_t phys_id_present_map;
1641 unsigned char old_id;
1642 unsigned long flags;
1645 * This is broken; anything with a real cpu count has to
1646 * circumvent this idiocy regardless.
1648 phys_id_present_map = ioapic_phys_id_map(phys_cpu_present_map);
1651 * Set the IOAPIC ID to the value stored in the MPC table.
1653 for (apic = 0; apic < nr_ioapics; apic++) {
1655 /* Read the register 0 value */
1656 spin_lock_irqsave(&ioapic_lock, flags);
1657 reg_00.raw = io_apic_read(apic, 0);
1658 spin_unlock_irqrestore(&ioapic_lock, flags);
1660 old_id = mp_ioapics[apic].mpc_apicid;
1662 if (mp_ioapics[apic].mpc_apicid >= get_physical_broadcast()) {
1663 printk(KERN_ERR "BIOS bug, IO-APIC#%d ID is %d in the MPC table!...\n",
1664 apic, mp_ioapics[apic].mpc_apicid);
1665 printk(KERN_ERR "... fixing up to %d. (tell your hw vendor)\n",
1667 mp_ioapics[apic].mpc_apicid = reg_00.bits.ID;
1670 /* Don't check I/O APIC IDs for some xAPIC systems. They have
1671 * no meaning without the serial APIC bus. */
1672 if (NO_IOAPIC_CHECK)
1675 * Sanity check, is the ID really free? Every APIC in a
1676 * system must have a unique ID or we get lots of nice
1677 * 'stuck on smp_invalidate_needed IPI wait' messages.
1679 if (check_apicid_used(phys_id_present_map,
1680 mp_ioapics[apic].mpc_apicid)) {
1681 printk(KERN_ERR "BIOS bug, IO-APIC#%d ID %d is already used!...\n",
1682 apic, mp_ioapics[apic].mpc_apicid);
1683 for (i = 0; i < get_physical_broadcast(); i++)
1684 if (!physid_isset(i, phys_id_present_map))
1686 if (i >= get_physical_broadcast())
1687 panic("Max APIC ID exceeded!\n");
1688 printk(KERN_ERR "... fixing up to %d. (tell your hw vendor)\n",
1690 physid_set(i, phys_id_present_map);
1691 mp_ioapics[apic].mpc_apicid = i;
1694 tmp = apicid_to_cpu_present(mp_ioapics[apic].mpc_apicid);
1695 printk("Setting %d in the phys_id_present_map\n", mp_ioapics[apic].mpc_apicid);
1696 physids_or(phys_id_present_map, phys_id_present_map, tmp);
1701 * We need to adjust the IRQ routing table
1702 * if the ID changed.
1704 if (old_id != mp_ioapics[apic].mpc_apicid)
1705 for (i = 0; i < mp_irq_entries; i++)
1706 if (mp_irqs[i].mpc_dstapic == old_id)
1707 mp_irqs[i].mpc_dstapic
1708 = mp_ioapics[apic].mpc_apicid;
1711 * Read the right value from the MPC table and
1712 * write it into the ID register.
1714 printk(KERN_INFO "...changing IO-APIC physical APIC ID to %d ...",
1715 mp_ioapics[apic].mpc_apicid);
1717 reg_00.bits.ID = mp_ioapics[apic].mpc_apicid;
1718 spin_lock_irqsave(&ioapic_lock, flags);
1719 io_apic_write(apic, 0, reg_00.raw);
1720 spin_unlock_irqrestore(&ioapic_lock, flags);
1725 spin_lock_irqsave(&ioapic_lock, flags);
1726 reg_00.raw = io_apic_read(apic, 0);
1727 spin_unlock_irqrestore(&ioapic_lock, flags);
1728 if (reg_00.bits.ID != mp_ioapics[apic].mpc_apicid)
1729 printk("could not set ID!\n");
1735 static void __init setup_ioapic_ids_from_mpc(void) { }
1739 * There is a nasty bug in some older SMP boards, their mptable lies
1740 * about the timer IRQ. We do the following to work around the situation:
1742 * - timer IRQ defaults to IO-APIC IRQ
1743 * - if this function detects that timer IRQs are defunct, then we fall
1744 * back to ISA timer IRQs
1746 static int __init timer_irq_works(void)
1748 unsigned long t1 = jiffies;
1751 /* Let ten ticks pass... */
1752 mdelay((10 * 1000) / HZ);
1755 * Expect a few ticks at least, to be sure some possible
1756 * glue logic does not lock up after one or two first
1757 * ticks in a non-ExtINT mode. Also the local APIC
1758 * might have cached one ExtINT interrupt. Finally, at
1759 * least one tick may be lost due to delays.
1761 if (jiffies - t1 > 4)
1768 * In the SMP+IOAPIC case it might happen that there are an unspecified
1769 * number of pending IRQ events unhandled. These cases are very rare,
1770 * so we 'resend' these IRQs via IPIs, to the same CPU. It's much
1771 * better to do it this way as thus we do not have to be aware of
1772 * 'pending' interrupts in the IRQ path, except at this point.
1775 * Edge triggered needs to resend any interrupt
1776 * that was delayed but this is now handled in the device
1781 * Starting up a edge-triggered IO-APIC interrupt is
1782 * nasty - we need to make sure that we get the edge.
1783 * If it is already asserted for some reason, we need
1784 * return 1 to indicate that is was pending.
1786 * This is not complete - we should be able to fake
1787 * an edge even if it isn't on the 8259A...
1789 static unsigned int startup_edge_ioapic_irq(unsigned int irq)
1791 int was_pending = 0;
1792 unsigned long flags;
1794 spin_lock_irqsave(&ioapic_lock, flags);
1796 disable_8259A_irq(irq);
1797 if (i8259A_irq_pending(irq))
1800 __unmask_IO_APIC_irq(irq);
1801 spin_unlock_irqrestore(&ioapic_lock, flags);
1807 * Once we have recorded IRQ_PENDING already, we can mask the
1808 * interrupt for real. This prevents IRQ storms from unhandled
1811 static void ack_edge_ioapic_irq(unsigned int irq)
1814 if ((irq_desc[irq].status & (IRQ_PENDING | IRQ_DISABLED))
1815 == (IRQ_PENDING | IRQ_DISABLED))
1816 mask_IO_APIC_irq(irq);
1821 * Level triggered interrupts can just be masked,
1822 * and shutting down and starting up the interrupt
1823 * is the same as enabling and disabling them -- except
1824 * with a startup need to return a "was pending" value.
1826 * Level triggered interrupts are special because we
1827 * do not touch any IO-APIC register while handling
1828 * them. We ack the APIC in the end-IRQ handler, not
1829 * in the start-IRQ-handler. Protection against reentrance
1830 * from the same interrupt is still provided, both by the
1831 * generic IRQ layer and by the fact that an unacked local
1832 * APIC does not accept IRQs.
1834 static unsigned int startup_level_ioapic_irq (unsigned int irq)
1836 unmask_IO_APIC_irq(irq);
1838 return 0; /* don't check for pending */
1841 static void end_level_ioapic_irq (unsigned int irq)
1848 * It appears there is an erratum which affects at least version 0x11
1849 * of I/O APIC (that's the 82093AA and cores integrated into various
1850 * chipsets). Under certain conditions a level-triggered interrupt is
1851 * erroneously delivered as edge-triggered one but the respective IRR
1852 * bit gets set nevertheless. As a result the I/O unit expects an EOI
1853 * message but it will never arrive and further interrupts are blocked
1854 * from the source. The exact reason is so far unknown, but the
1855 * phenomenon was observed when two consecutive interrupt requests
1856 * from a given source get delivered to the same CPU and the source is
1857 * temporarily disabled in between.
1859 * A workaround is to simulate an EOI message manually. We achieve it
1860 * by setting the trigger mode to edge and then to level when the edge
1861 * trigger mode gets detected in the TMR of a local APIC for a
1862 * level-triggered interrupt. We mask the source for the time of the
1863 * operation to prevent an edge-triggered interrupt escaping meanwhile.
1864 * The idea is from Manfred Spraul. --macro
1866 i = IO_APIC_VECTOR(irq);
1868 v = apic_read(APIC_TMR + ((i & ~0x1f) >> 1));
1872 if (!(v & (1 << (i & 0x1f)))) {
1873 #ifdef APIC_MISMATCH_DEBUG
1874 atomic_inc(&irq_mis_count);
1876 spin_lock(&ioapic_lock);
1877 __mask_and_edge_IO_APIC_irq(irq);
1878 __unmask_and_level_IO_APIC_irq(irq);
1879 spin_unlock(&ioapic_lock);
1883 #ifdef CONFIG_PCI_MSI
1884 static unsigned int startup_edge_ioapic_vector(unsigned int vector)
1886 int irq = vector_to_irq(vector);
1888 return startup_edge_ioapic_irq(irq);
1891 static void ack_edge_ioapic_vector(unsigned int vector)
1893 int irq = vector_to_irq(vector);
1895 ack_edge_ioapic_irq(irq);
1898 static unsigned int startup_level_ioapic_vector (unsigned int vector)
1900 int irq = vector_to_irq(vector);
1902 return startup_level_ioapic_irq (irq);
1905 static void end_level_ioapic_vector (unsigned int vector)
1907 int irq = vector_to_irq(vector);
1909 end_level_ioapic_irq(irq);
1912 static void mask_IO_APIC_vector (unsigned int vector)
1914 int irq = vector_to_irq(vector);
1916 mask_IO_APIC_irq(irq);
1919 static void unmask_IO_APIC_vector (unsigned int vector)
1921 int irq = vector_to_irq(vector);
1923 unmask_IO_APIC_irq(irq);
1926 static void set_ioapic_affinity_vector (unsigned int vector,
1929 int irq = vector_to_irq(vector);
1931 set_ioapic_affinity_irq(irq, cpu_mask);
1936 * Level and edge triggered IO-APIC interrupts need different handling,
1937 * so we use two separate IRQ descriptors. Edge triggered IRQs can be
1938 * handled with the level-triggered descriptor, but that one has slightly
1939 * more overhead. Level-triggered interrupts cannot be handled with the
1940 * edge-triggered handler, without risking IRQ storms and other ugly
1943 static struct hw_interrupt_type ioapic_edge_type = {
1944 .typename = "IO-APIC-edge",
1945 .startup = startup_edge_ioapic,
1946 .shutdown = shutdown_edge_ioapic,
1947 .enable = enable_edge_ioapic,
1948 .disable = disable_edge_ioapic,
1949 .ack = ack_edge_ioapic,
1950 .end = end_edge_ioapic,
1951 .set_affinity = set_ioapic_affinity,
1954 static struct hw_interrupt_type ioapic_level_type = {
1955 .typename = "IO-APIC-level",
1956 .startup = startup_level_ioapic,
1957 .shutdown = shutdown_level_ioapic,
1958 .enable = enable_level_ioapic,
1959 .disable = disable_level_ioapic,
1960 .ack = mask_and_ack_level_ioapic,
1961 .end = end_level_ioapic,
1962 .set_affinity = set_ioapic_affinity,
1965 static inline void init_IO_APIC_traps(void)
1970 * NOTE! The local APIC isn't very good at handling
1971 * multiple interrupts at the same interrupt level.
1972 * As the interrupt level is determined by taking the
1973 * vector number and shifting that right by 4, we
1974 * want to spread these out a bit so that they don't
1975 * all fall in the same interrupt level.
1977 * Also, we've got to be careful not to trash gate
1978 * 0x80, because int 0x80 is hm, kind of importantish. ;)
1980 for (irq = 0; irq < NR_IRQS ; irq++) {
1982 if (use_pci_vector()) {
1983 if (!platform_legacy_irq(tmp))
1984 if ((tmp = vector_to_irq(tmp)) == -1)
1987 if (IO_APIC_IRQ(tmp) && !IO_APIC_VECTOR(tmp)) {
1989 * Hmm.. We don't have an entry for this,
1990 * so default to an old-fashioned 8259
1991 * interrupt if we can..
1994 make_8259A_irq(irq);
1996 /* Strange. Oh, well.. */
1997 irq_desc[irq].handler = &no_irq_type;
2002 static void enable_lapic_irq (unsigned int irq)
2006 v = apic_read(APIC_LVT0);
2007 apic_write_around(APIC_LVT0, v & ~APIC_LVT_MASKED);
2010 static void disable_lapic_irq (unsigned int irq)
2014 v = apic_read(APIC_LVT0);
2015 apic_write_around(APIC_LVT0, v | APIC_LVT_MASKED);
2018 static void ack_lapic_irq (unsigned int irq)
2023 static void end_lapic_irq (unsigned int i) { /* nothing */ }
2025 static struct hw_interrupt_type lapic_irq_type = {
2026 .typename = "local-APIC-edge",
2027 .startup = NULL, /* startup_irq() not used for IRQ0 */
2028 .shutdown = NULL, /* shutdown_irq() not used for IRQ0 */
2029 .enable = enable_lapic_irq,
2030 .disable = disable_lapic_irq,
2031 .ack = ack_lapic_irq,
2032 .end = end_lapic_irq
2035 static void setup_nmi (void)
2038 * Dirty trick to enable the NMI watchdog ...
2039 * We put the 8259A master into AEOI mode and
2040 * unmask on all local APICs LVT0 as NMI.
2042 * The idea to use the 8259A in AEOI mode ('8259A Virtual Wire')
2043 * is from Maciej W. Rozycki - so we do not have to EOI from
2044 * the NMI handler or the timer interrupt.
2046 printk(KERN_INFO "activating NMI Watchdog ...");
2048 on_each_cpu(enable_NMI_through_LVT0, NULL, 1, 1);
2054 * This looks a bit hackish but it's about the only one way of sending
2055 * a few INTA cycles to 8259As and any associated glue logic. ICR does
2056 * not support the ExtINT mode, unfortunately. We need to send these
2057 * cycles as some i82489DX-based boards have glue logic that keeps the
2058 * 8259A interrupt line asserted until INTA. --macro
2060 static inline void unlock_ExtINT_logic(void)
2063 struct IO_APIC_route_entry entry0, entry1;
2064 unsigned char save_control, save_freq_select;
2065 unsigned long flags;
2067 pin = find_isa_irq_pin(8, mp_INT);
2071 spin_lock_irqsave(&ioapic_lock, flags);
2072 *(((int *)&entry0) + 1) = io_apic_read(0, 0x11 + 2 * pin);
2073 *(((int *)&entry0) + 0) = io_apic_read(0, 0x10 + 2 * pin);
2074 spin_unlock_irqrestore(&ioapic_lock, flags);
2075 clear_IO_APIC_pin(0, pin);
2077 memset(&entry1, 0, sizeof(entry1));
2079 entry1.dest_mode = 0; /* physical delivery */
2080 entry1.mask = 0; /* unmask IRQ now */
2081 entry1.dest.physical.physical_dest = hard_smp_processor_id();
2082 entry1.delivery_mode = dest_ExtINT;
2083 entry1.polarity = entry0.polarity;
2087 spin_lock_irqsave(&ioapic_lock, flags);
2088 io_apic_write(0, 0x11 + 2 * pin, *(((int *)&entry1) + 1));
2089 io_apic_write(0, 0x10 + 2 * pin, *(((int *)&entry1) + 0));
2090 spin_unlock_irqrestore(&ioapic_lock, flags);
2092 save_control = CMOS_READ(RTC_CONTROL);
2093 save_freq_select = CMOS_READ(RTC_FREQ_SELECT);
2094 CMOS_WRITE((save_freq_select & ~RTC_RATE_SELECT) | 0x6,
2096 CMOS_WRITE(save_control | RTC_PIE, RTC_CONTROL);
2101 if ((CMOS_READ(RTC_INTR_FLAGS) & RTC_PF) == RTC_PF)
2105 CMOS_WRITE(save_control, RTC_CONTROL);
2106 CMOS_WRITE(save_freq_select, RTC_FREQ_SELECT);
2107 clear_IO_APIC_pin(0, pin);
2109 spin_lock_irqsave(&ioapic_lock, flags);
2110 io_apic_write(0, 0x11 + 2 * pin, *(((int *)&entry0) + 1));
2111 io_apic_write(0, 0x10 + 2 * pin, *(((int *)&entry0) + 0));
2112 spin_unlock_irqrestore(&ioapic_lock, flags);
2116 * This code may look a bit paranoid, but it's supposed to cooperate with
2117 * a wide range of boards and BIOS bugs. Fortunately only the timer IRQ
2118 * is so screwy. Thanks to Brian Perkins for testing/hacking this beast
2119 * fanatically on his truly buggy board.
2121 static inline void check_timer(void)
2127 * get/set the timer IRQ vector:
2129 disable_8259A_irq(0);
2130 vector = assign_irq_vector(0);
2131 set_intr_gate(vector, interrupt[0]);
2134 * Subtle, code in do_timer_interrupt() expects an AEOI
2135 * mode for the 8259A whenever interrupts are routed
2136 * through I/O APICs. Also IRQ0 has to be enabled in
2137 * the 8259A which implies the virtual wire has to be
2138 * disabled in the local APIC.
2140 apic_write_around(APIC_LVT0, APIC_LVT_MASKED | APIC_DM_EXTINT);
2143 enable_8259A_irq(0);
2145 pin1 = find_isa_irq_pin(0, mp_INT);
2146 pin2 = find_isa_irq_pin(0, mp_ExtINT);
2148 printk(KERN_INFO "..TIMER: vector=0x%02X pin1=%d pin2=%d\n", vector, pin1, pin2);
2152 * Ok, does IRQ0 through the IOAPIC work?
2154 unmask_IO_APIC_irq(0);
2155 if (timer_irq_works()) {
2156 if (nmi_watchdog == NMI_IO_APIC) {
2157 disable_8259A_irq(0);
2159 enable_8259A_irq(0);
2160 check_nmi_watchdog();
2164 clear_IO_APIC_pin(0, pin1);
2165 printk(KERN_ERR "..MP-BIOS bug: 8254 timer not connected to IO-APIC\n");
2168 printk(KERN_INFO "...trying to set up timer (IRQ0) through the 8259A ... ");
2170 printk("\n..... (found pin %d) ...", pin2);
2172 * legacy devices should be connected to IO APIC #0
2174 setup_ExtINT_IRQ0_pin(pin2, vector);
2175 if (timer_irq_works()) {
2178 replace_pin_at_irq(0, 0, pin1, 0, pin2);
2180 add_pin_to_irq(0, 0, pin2);
2181 if (nmi_watchdog == NMI_IO_APIC) {
2183 check_nmi_watchdog();
2188 * Cleanup, just in case ...
2190 clear_IO_APIC_pin(0, pin2);
2192 printk(" failed.\n");
2194 if (nmi_watchdog == NMI_IO_APIC) {
2195 printk(KERN_WARNING "timer doesn't work through the IO-APIC - disabling NMI Watchdog!\n");
2199 printk(KERN_INFO "...trying to set up timer as Virtual Wire IRQ...");
2201 disable_8259A_irq(0);
2202 irq_desc[0].handler = &lapic_irq_type;
2203 apic_write_around(APIC_LVT0, APIC_DM_FIXED | vector); /* Fixed mode */
2204 enable_8259A_irq(0);
2206 if (timer_irq_works()) {
2207 printk(" works.\n");
2210 apic_write_around(APIC_LVT0, APIC_LVT_MASKED | APIC_DM_FIXED | vector);
2211 printk(" failed.\n");
2213 printk(KERN_INFO "...trying to set up timer as ExtINT IRQ...");
2218 apic_write_around(APIC_LVT0, APIC_DM_EXTINT);
2220 unlock_ExtINT_logic();
2222 if (timer_irq_works()) {
2223 printk(" works.\n");
2226 printk(" failed :(.\n");
2227 panic("IO-APIC + timer doesn't work! Try using the 'noapic' kernel parameter\n");
2232 * IRQ's that are handled by the PIC in the MPS IOAPIC case.
2233 * - IRQ2 is the cascade IRQ, and cannot be a io-apic IRQ.
2234 * Linux doesn't really care, as it's not actually used
2235 * for any interrupt handling anyway.
2237 #define PIC_IRQS (1 << PIC_CASCADE_IR)
2239 void __init setup_IO_APIC(void)
2244 io_apic_irqs = ~0; /* all IRQs go through IOAPIC */
2246 io_apic_irqs = ~PIC_IRQS;
2248 printk("ENABLING IO-APIC IRQs\n");
2251 * Set up IO-APIC IRQ routing.
2254 setup_ioapic_ids_from_mpc();
2256 setup_IO_APIC_irqs();
2257 init_IO_APIC_traps();
2264 * Called after all the initialization is done. If we didnt find any
2265 * APIC bugs then we can allow the modify fast path
2268 static int __init io_apic_bug_finalize(void)
2270 if(sis_apic_bug == -1)
2275 late_initcall(io_apic_bug_finalize);
2277 struct sysfs_ioapic_data {
2278 struct sys_device dev;
2279 struct IO_APIC_route_entry entry[0];
2281 static struct sysfs_ioapic_data * mp_ioapic_data[MAX_IO_APICS];
2283 static int ioapic_suspend(struct sys_device *dev, u32 state)
2285 struct IO_APIC_route_entry *entry;
2286 struct sysfs_ioapic_data *data;
2287 unsigned long flags;
2290 data = container_of(dev, struct sysfs_ioapic_data, dev);
2291 entry = data->entry;
2292 spin_lock_irqsave(&ioapic_lock, flags);
2293 for (i = 0; i < nr_ioapic_registers[dev->id]; i ++, entry ++ ) {
2294 *(((int *)entry) + 1) = io_apic_read(dev->id, 0x11 + 2 * i);
2295 *(((int *)entry) + 0) = io_apic_read(dev->id, 0x10 + 2 * i);
2297 spin_unlock_irqrestore(&ioapic_lock, flags);
2302 static int ioapic_resume(struct sys_device *dev)
2304 struct IO_APIC_route_entry *entry;
2305 struct sysfs_ioapic_data *data;
2306 unsigned long flags;
2307 union IO_APIC_reg_00 reg_00;
2310 data = container_of(dev, struct sysfs_ioapic_data, dev);
2311 entry = data->entry;
2313 spin_lock_irqsave(&ioapic_lock, flags);
2314 reg_00.raw = io_apic_read(dev->id, 0);
2315 if (reg_00.bits.ID != mp_ioapics[dev->id].mpc_apicid) {
2316 reg_00.bits.ID = mp_ioapics[dev->id].mpc_apicid;
2317 io_apic_write(dev->id, 0, reg_00.raw);
2319 for (i = 0; i < nr_ioapic_registers[dev->id]; i ++, entry ++ ) {
2320 io_apic_write(dev->id, 0x11+2*i, *(((int *)entry)+1));
2321 io_apic_write(dev->id, 0x10+2*i, *(((int *)entry)+0));
2323 spin_unlock_irqrestore(&ioapic_lock, flags);
2328 static struct sysdev_class ioapic_sysdev_class = {
2329 set_kset_name("ioapic"),
2330 .suspend = ioapic_suspend,
2331 .resume = ioapic_resume,
2334 static int __init ioapic_init_sysfs(void)
2336 struct sys_device * dev;
2337 int i, size, error = 0;
2339 error = sysdev_class_register(&ioapic_sysdev_class);
2343 for (i = 0; i < nr_ioapics; i++ ) {
2344 size = sizeof(struct sys_device) + nr_ioapic_registers[i]
2345 * sizeof(struct IO_APIC_route_entry);
2346 mp_ioapic_data[i] = kmalloc(size, GFP_KERNEL);
2347 if (!mp_ioapic_data[i]) {
2348 printk(KERN_ERR "Can't suspend/resume IOAPIC %d\n", i);
2351 memset(mp_ioapic_data[i], 0, size);
2352 dev = &mp_ioapic_data[i]->dev;
2354 dev->cls = &ioapic_sysdev_class;
2355 error = sysdev_register(dev);
2357 kfree(mp_ioapic_data[i]);
2358 mp_ioapic_data[i] = NULL;
2359 printk(KERN_ERR "Can't suspend/resume IOAPIC %d\n", i);
2367 device_initcall(ioapic_init_sysfs);
2369 /* --------------------------------------------------------------------------
2370 ACPI-based IOAPIC Configuration
2371 -------------------------------------------------------------------------- */
2373 #ifdef CONFIG_ACPI_BOOT
2375 int __init io_apic_get_unique_id (int ioapic, int apic_id)
2377 union IO_APIC_reg_00 reg_00;
2378 static physid_mask_t apic_id_map = PHYSID_MASK_NONE;
2380 unsigned long flags;
2384 * The P4 platform supports up to 256 APIC IDs on two separate APIC
2385 * buses (one for LAPICs, one for IOAPICs), where predecessors only
2386 * supports up to 16 on one shared APIC bus.
2388 * TBD: Expand LAPIC/IOAPIC support on P4-class systems to take full
2389 * advantage of new APIC bus architecture.
2392 if (physids_empty(apic_id_map))
2393 apic_id_map = ioapic_phys_id_map(phys_cpu_present_map);
2395 spin_lock_irqsave(&ioapic_lock, flags);
2396 reg_00.raw = io_apic_read(ioapic, 0);
2397 spin_unlock_irqrestore(&ioapic_lock, flags);
2399 if (apic_id >= get_physical_broadcast()) {
2400 printk(KERN_WARNING "IOAPIC[%d]: Invalid apic_id %d, trying "
2401 "%d\n", ioapic, apic_id, reg_00.bits.ID);
2402 apic_id = reg_00.bits.ID;
2406 * Every APIC in a system must have a unique ID or we get lots of nice
2407 * 'stuck on smp_invalidate_needed IPI wait' messages.
2409 if (check_apicid_used(apic_id_map, apic_id)) {
2411 for (i = 0; i < get_physical_broadcast(); i++) {
2412 if (!check_apicid_used(apic_id_map, i))
2416 if (i == get_physical_broadcast())
2417 panic("Max apic_id exceeded!\n");
2419 printk(KERN_WARNING "IOAPIC[%d]: apic_id %d already used, "
2420 "trying %d\n", ioapic, apic_id, i);
2425 tmp = apicid_to_cpu_present(apic_id);
2426 physids_or(apic_id_map, apic_id_map, tmp);
2428 if (reg_00.bits.ID != apic_id) {
2429 reg_00.bits.ID = apic_id;
2431 spin_lock_irqsave(&ioapic_lock, flags);
2432 io_apic_write(ioapic, 0, reg_00.raw);
2433 reg_00.raw = io_apic_read(ioapic, 0);
2434 spin_unlock_irqrestore(&ioapic_lock, flags);
2437 if (reg_00.bits.ID != apic_id)
2438 panic("IOAPIC[%d]: Unable change apic_id!\n", ioapic);
2441 printk(KERN_INFO "IOAPIC[%d]: Assigned apic_id %d\n", ioapic, apic_id);
2447 int __init io_apic_get_version (int ioapic)
2449 union IO_APIC_reg_01 reg_01;
2450 unsigned long flags;
2452 spin_lock_irqsave(&ioapic_lock, flags);
2453 reg_01.raw = io_apic_read(ioapic, 1);
2454 spin_unlock_irqrestore(&ioapic_lock, flags);
2456 return reg_01.bits.version;
2460 int __init io_apic_get_redir_entries (int ioapic)
2462 union IO_APIC_reg_01 reg_01;
2463 unsigned long flags;
2465 spin_lock_irqsave(&ioapic_lock, flags);
2466 reg_01.raw = io_apic_read(ioapic, 1);
2467 spin_unlock_irqrestore(&ioapic_lock, flags);
2469 return reg_01.bits.entries;
2473 int io_apic_set_pci_routing (int ioapic, int pin, int irq, int edge_level, int active_high_low)
2475 struct IO_APIC_route_entry entry;
2476 unsigned long flags;
2478 if (!IO_APIC_IRQ(irq)) {
2479 printk(KERN_ERR "IOAPIC[%d]: Invalid reference to IRQ 0\n",
2485 * Generate a PCI IRQ routing entry and program the IOAPIC accordingly.
2486 * Note that we mask (disable) IRQs now -- these get enabled when the
2487 * corresponding device driver registers for this IRQ.
2490 memset(&entry,0,sizeof(entry));
2492 entry.delivery_mode = INT_DELIVERY_MODE;
2493 entry.dest_mode = INT_DEST_MODE;
2494 entry.dest.logical.logical_dest = cpu_mask_to_apicid(TARGET_CPUS);
2495 entry.trigger = edge_level;
2496 entry.polarity = active_high_low;
2500 * IRQs < 16 are already in the irq_2_pin[] map
2503 add_pin_to_irq(irq, ioapic, pin);
2505 entry.vector = assign_irq_vector(irq);
2507 Dprintk(KERN_DEBUG "IOAPIC[%d]: Set PCI routing entry (%d-%d -> 0x%x -> "
2508 "IRQ %d Mode:%i Active:%i)\n", ioapic,
2509 mp_ioapics[ioapic].mpc_apicid, pin, entry.vector, irq, edge_level, active_high_low);
2511 if (use_pci_vector() && !platform_legacy_irq(irq))
2512 irq = IO_APIC_VECTOR(irq);
2514 irq_desc[irq].handler = &ioapic_level_type;
2516 irq_desc[irq].handler = &ioapic_edge_type;
2519 set_intr_gate(entry.vector, interrupt[irq]);
2521 if (!ioapic && (irq < 16))
2522 disable_8259A_irq(irq);
2524 spin_lock_irqsave(&ioapic_lock, flags);
2525 io_apic_write(ioapic, 0x11+2*pin, *(((int *)&entry)+1));
2526 io_apic_write(ioapic, 0x10+2*pin, *(((int *)&entry)+0));
2527 spin_unlock_irqrestore(&ioapic_lock, flags);
2532 #endif /*CONFIG_ACPI_BOOT*/