projects / deliverable / linux.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
[PATCH] variable overflow after hundreds round of hotplug CPU
[deliverable/linux.git] / arch / i386 / kernel / io_apic.c
CommitLineData
1da177e4
LT		 1/*
2 * Intel IO-APIC support for multi-Pentium hosts.
3 *
4 * Copyright (C) 1997, 1998, 1999, 2000 Ingo Molnar, Hajnalka Szabo
5 *
6 * Many thanks to Stig Venaas for trying out countless experimental
7 * patches and reporting/debugging problems patiently!
8 *
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>
14 *
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
21 */
22
23#include <linux/mm.h>
24#include <linux/irq.h>
25#include <linux/interrupt.h>
26#include <linux/init.h>
27#include <linux/delay.h>
28#include <linux/sched.h>
29#include <linux/config.h>
30#include <linux/smp_lock.h>
31#include <linux/mc146818rtc.h>
32#include <linux/compiler.h>
33#include <linux/acpi.h>
129f6946 34#include <linux/module.h>
1da177e4
LT		 35#include <linux/sysdev.h>
36#include <asm/io.h>
37#include <asm/smp.h>
38#include <asm/desc.h>
39#include <asm/timer.h>
40
41#include <mach_apic.h>
42
43#include "io_ports.h"
44
45int (*ioapic_renumber_irq)(int ioapic, int irq);
46atomic_t irq_mis_count;
47
48static DEFINE_SPINLOCK(ioapic_lock);
49
50/*
51 * Is the SiS APIC rmw bug present ?
52 * -1 = don't know, 0 = no, 1 = yes
53 */
54int sis_apic_bug = -1;
55
56/*
57 * # of IRQ routing registers
58 */
59int nr_ioapic_registers[MAX_IO_APICS];
60
61/*
62 * Rough estimation of how many shared IRQs there are, can
63 * be changed anytime.
64 */
65#define MAX_PLUS_SHARED_IRQS NR_IRQS
66#define PIN_MAP_SIZE (MAX_PLUS_SHARED_IRQS + NR_IRQS)
67
68/*
69 * This is performance-critical, we want to do it O(1)
70 *
71 * the indexing order of this array favors 1:1 mappings
72 * between pins and IRQs.
73 */
74
75static struct irq_pin_list {
76 int apic, pin, next;
77} irq_2_pin[PIN_MAP_SIZE];
78
79int vector_irq[NR_VECTORS] = { [0 ... NR_VECTORS - 1] = -1};
80#ifdef CONFIG_PCI_MSI
81#define vector_to_irq(vector) \
82 (platform_legacy_irq(vector) ? vector : vector_irq[vector])
83#else
84#define vector_to_irq(vector) (vector)
85#endif
86
87/*
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.
91 */
92static void add_pin_to_irq(unsigned int irq, int apic, int pin)
93{
94 static int first_free_entry = NR_IRQS;
95 struct irq_pin_list *entry = irq_2_pin + irq;
96
97 while (entry->next)
98 entry = irq_2_pin + entry->next;
99
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");
105 }
106 entry->apic = apic;
107 entry->pin = pin;
108}
109
110/*
111 * Reroute an IRQ to a different pin.
112 */
113static void __init replace_pin_at_irq(unsigned int irq,
114 int oldapic, int oldpin,
115 int newapic, int newpin)
116{
117 struct irq_pin_list *entry = irq_2_pin + irq;
118
119 while (1) {
120 if (entry->apic == oldapic && entry->pin == oldpin) {
121 entry->apic = newapic;
122 entry->pin = newpin;
123 }
124 if (!entry->next)
125 break;
126 entry = irq_2_pin + entry->next;
127 }
128}
129
130static void __modify_IO_APIC_irq (unsigned int irq, unsigned long enable, unsigned long disable)
131{
132 struct irq_pin_list *entry = irq_2_pin + irq;
133 unsigned int pin, reg;
134
135 for (;;) {
136 pin = entry->pin;
137 if (pin == -1)
138 break;
139 reg = io_apic_read(entry->apic, 0x10 + pin*2);
140 reg &= ~disable;
141 reg |= enable;
142 io_apic_modify(entry->apic, 0x10 + pin*2, reg);
143 if (!entry->next)
144 break;
145 entry = irq_2_pin + entry->next;
146 }
147}
148
149/* mask = 1 */
150static void __mask_IO_APIC_irq (unsigned int irq)
151{
152 __modify_IO_APIC_irq(irq, 0x00010000, 0);
153}
154
155/* mask = 0 */
156static void __unmask_IO_APIC_irq (unsigned int irq)
157{
158 __modify_IO_APIC_irq(irq, 0, 0x00010000);
159}
160
161/* mask = 1, trigger = 0 */
162static void __mask_and_edge_IO_APIC_irq (unsigned int irq)
163{
164 __modify_IO_APIC_irq(irq, 0x00010000, 0x00008000);
165}
166
167/* mask = 0, trigger = 1 */
168static void __unmask_and_level_IO_APIC_irq (unsigned int irq)
169{
170 __modify_IO_APIC_irq(irq, 0x00008000, 0x00010000);
171}
172
173static void mask_IO_APIC_irq (unsigned int irq)
174{
175 unsigned long flags;
176
177 spin_lock_irqsave(&ioapic_lock, flags);
178 __mask_IO_APIC_irq(irq);
179 spin_unlock_irqrestore(&ioapic_lock, flags);
180}
181
182static void unmask_IO_APIC_irq (unsigned int irq)
183{
184 unsigned long flags;
185
186 spin_lock_irqsave(&ioapic_lock, flags);
187 __unmask_IO_APIC_irq(irq);
188 spin_unlock_irqrestore(&ioapic_lock, flags);
189}
190
191static void clear_IO_APIC_pin(unsigned int apic, unsigned int pin)
192{
193 struct IO_APIC_route_entry entry;
194 unsigned long flags;
195
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)
202 return;
203
204 /*
205 * Disable it in the IO-APIC irq-routing table:
206 */
207 memset(&entry, 0, sizeof(entry));
208 entry.mask = 1;
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);
213}
214
215static void clear_IO_APIC (void)
216{
217 int apic, pin;
218
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);
222}
223
224static void set_ioapic_affinity_irq(unsigned int irq, cpumask_t cpumask)
225{
226 unsigned long flags;
227 int pin;
228 struct irq_pin_list *entry = irq_2_pin + irq;
229 unsigned int apicid_value;
230
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);
235 for (;;) {
236 pin = entry->pin;
237 if (pin == -1)
238 break;
239 io_apic_write(entry->apic, 0x10 + 1 + pin*2, apicid_value);
240 if (!entry->next)
241 break;
242 entry = irq_2_pin + entry->next;
243 }
244 spin_unlock_irqrestore(&ioapic_lock, flags);
245}
246
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() */
252
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)
256# else
257# define TDprintk(x...)
258# define Dprintk(x...)
259# endif
260
261cpumask_t __cacheline_aligned pending_irq_balance_cpumask[NR_IRQS];
262
263#define IRQBALANCE_CHECK_ARCH -999
264static int irqbalance_disabled = IRQBALANCE_CHECK_ARCH;
265static int physical_balance = 0;
266
267static struct irq_cpu_info {
268 unsigned long * last_irq;
269 unsigned long * irq_delta;
270 unsigned long irq;
271} irq_cpu_data[NR_CPUS];
272
273#define CPU_IRQ(cpu) (irq_cpu_data[cpu].irq)
274#define LAST_CPU_IRQ(cpu,irq) (irq_cpu_data[cpu].last_irq[irq])
275#define IRQ_DELTA(cpu,irq) (irq_cpu_data[cpu].irq_delta[irq])
276
277#define IDLE_ENOUGH(cpu,now) \
278 (idle_cpu(cpu) && ((now) - per_cpu(irq_stat, (cpu)).idle_timestamp > 1))
279
280#define IRQ_ALLOWED(cpu, allowed_mask) cpu_isset(cpu, allowed_mask)
281
282#define CPU_TO_PACKAGEINDEX(i) (first_cpu(cpu_sibling_map[i]))
283
284#define MAX_BALANCED_IRQ_INTERVAL (5*HZ)
285#define MIN_BALANCED_IRQ_INTERVAL (HZ/2)
286#define BALANCED_IRQ_MORE_DELTA (HZ/10)
287#define BALANCED_IRQ_LESS_DELTA (HZ)
288
289static long balanced_irq_interval = MAX_BALANCED_IRQ_INTERVAL;
290
291static unsigned long move(int curr_cpu, cpumask_t allowed_mask,
292 unsigned long now, int direction)
293{
294 int search_idle = 1;
295 int cpu = curr_cpu;
296
297 goto inside;
298
299 do {
300 if (unlikely(cpu == curr_cpu))
301 search_idle = 0;
302inside:
303 if (direction == 1) {
304 cpu++;
305 if (cpu >= NR_CPUS)
306 cpu = 0;
307 } else {
308 cpu--;
309 if (cpu == -1)
310 cpu = NR_CPUS-1;
311 }
312 } while (!cpu_online(cpu) || !IRQ_ALLOWED(cpu,allowed_mask) ||
313 (search_idle && !IDLE_ENOUGH(cpu,now)));
314
315 return cpu;
316}
317
318static inline void balance_irq(int cpu, int irq)
319{
320 unsigned long now = jiffies;
321 cpumask_t allowed_mask;
322 unsigned int new_cpu;
323
324 if (irqbalance_disabled)
325 return;
326
327 cpus_and(allowed_mask, cpu_online_map, irq_affinity[irq]);
328 new_cpu = move(cpu, allowed_mask, now, 1);
329 if (cpu != new_cpu) {
330 irq_desc_t *desc = irq_desc + irq;
331 unsigned long flags;
332
333 spin_lock_irqsave(&desc->lock, flags);
334 pending_irq_balance_cpumask[irq] = cpumask_of_cpu(new_cpu);
335 spin_unlock_irqrestore(&desc->lock, flags);
336 }
337}
338
339static inline void rotate_irqs_among_cpus(unsigned long useful_load_threshold)
340{
341 int i, j;
342 Dprintk("Rotating IRQs among CPUs.\n");
343 for (i = 0; i < NR_CPUS; i++) {
344 for (j = 0; cpu_online(i) && (j < NR_IRQS); j++) {
345 if (!irq_desc[j].action)
346 continue;
347 /* Is it a significant load ? */
348 if (IRQ_DELTA(CPU_TO_PACKAGEINDEX(i),j) <
349 useful_load_threshold)
350 continue;
351 balance_irq(i, j);
352 }
353 }
354 balanced_irq_interval = max((long)MIN_BALANCED_IRQ_INTERVAL,
355 balanced_irq_interval - BALANCED_IRQ_LESS_DELTA);
356 return;
357}
358
359static void do_irq_balance(void)
360{
361 int i, j;
362 unsigned long max_cpu_irq = 0, min_cpu_irq = (~0);
363 unsigned long move_this_load = 0;
364 int max_loaded = 0, min_loaded = 0;
365 int load;
366 unsigned long useful_load_threshold = balanced_irq_interval + 10;
367 int selected_irq;
368 int tmp_loaded, first_attempt = 1;
369 unsigned long tmp_cpu_irq;
370 unsigned long imbalance = 0;
371 cpumask_t allowed_mask, target_cpu_mask, tmp;
372
373 for (i = 0; i < NR_CPUS; i++) {
374 int package_index;
375 CPU_IRQ(i) = 0;
376 if (!cpu_online(i))
377 continue;
378 package_index = CPU_TO_PACKAGEINDEX(i);
379 for (j = 0; j < NR_IRQS; j++) {
380 unsigned long value_now, delta;
381 /* Is this an active IRQ? */
382 if (!irq_desc[j].action)
383 continue;
384 if ( package_index == i )
385 IRQ_DELTA(package_index,j) = 0;
386 /* Determine the total count per processor per IRQ */
387 value_now = (unsigned long) kstat_cpu(i).irqs[j];
388
389 /* Determine the activity per processor per IRQ */
390 delta = value_now - LAST_CPU_IRQ(i,j);
391
392 /* Update last_cpu_irq[][] for the next time */
393 LAST_CPU_IRQ(i,j) = value_now;
394
395 /* Ignore IRQs whose rate is less than the clock */
396 if (delta < useful_load_threshold)
397 continue;
398 /* update the load for the processor or package total */
399 IRQ_DELTA(package_index,j) += delta;
400
401 /* Keep track of the higher numbered sibling as well */
402 if (i != package_index)
403 CPU_IRQ(i) += delta;
404 /*
405 * We have sibling A and sibling B in the package
406 *
407 * cpu_irq[A] = load for cpu A + load for cpu B
408 * cpu_irq[B] = load for cpu B
409 */
410 CPU_IRQ(package_index) += delta;
411 }
412 }
413 /* Find the least loaded processor package */
414 for (i = 0; i < NR_CPUS; i++) {
415 if (!cpu_online(i))
416 continue;
417 if (i != CPU_TO_PACKAGEINDEX(i))
418 continue;
419 if (min_cpu_irq > CPU_IRQ(i)) {
420 min_cpu_irq = CPU_IRQ(i);
421 min_loaded = i;
422 }
423 }
424 max_cpu_irq = ULONG_MAX;
425
426tryanothercpu:
427 /* Look for heaviest loaded processor.
428 * We may come back to get the next heaviest loaded processor.
429 * Skip processors with trivial loads.
430 */
431 tmp_cpu_irq = 0;
432 tmp_loaded = -1;
433 for (i = 0; i < NR_CPUS; i++) {
434 if (!cpu_online(i))
435 continue;
436 if (i != CPU_TO_PACKAGEINDEX(i))
437 continue;
438 if (max_cpu_irq <= CPU_IRQ(i))
439 continue;
440 if (tmp_cpu_irq < CPU_IRQ(i)) {
441 tmp_cpu_irq = CPU_IRQ(i);
442 tmp_loaded = i;
443 }
444 }
445
446 if (tmp_loaded == -1) {
447 /* In the case of small number of heavy interrupt sources,
448 * loading some of the cpus too much. We use Ingo's original
449 * approach to rotate them around.
450 */
451 if (!first_attempt && imbalance >= useful_load_threshold) {
452 rotate_irqs_among_cpus(useful_load_threshold);
453 return;
454 }
455 goto not_worth_the_effort;
456 }
457
458 first_attempt = 0; /* heaviest search */
459 max_cpu_irq = tmp_cpu_irq; /* load */
460 max_loaded = tmp_loaded; /* processor */
461 imbalance = (max_cpu_irq - min_cpu_irq) / 2;
462
463 Dprintk("max_loaded cpu = %d\n", max_loaded);
464 Dprintk("min_loaded cpu = %d\n", min_loaded);
465 Dprintk("max_cpu_irq load = %ld\n", max_cpu_irq);
466 Dprintk("min_cpu_irq load = %ld\n", min_cpu_irq);
467 Dprintk("load imbalance = %lu\n", imbalance);
468
469 /* if imbalance is less than approx 10% of max load, then
470 * observe diminishing returns action. - quit
471 */
472 if (imbalance < (max_cpu_irq >> 3)) {
473 Dprintk("Imbalance too trivial\n");
474 goto not_worth_the_effort;
475 }
476
477tryanotherirq:
478 /* if we select an IRQ to move that can't go where we want, then
479 * see if there is another one to try.
480 */
481 move_this_load = 0;
482 selected_irq = -1;
483 for (j = 0; j < NR_IRQS; j++) {
484 /* Is this an active IRQ? */
485 if (!irq_desc[j].action)
486 continue;
487 if (imbalance <= IRQ_DELTA(max_loaded,j))
488 continue;
489 /* Try to find the IRQ that is closest to the imbalance
490 * without going over.
491 */
492 if (move_this_load < IRQ_DELTA(max_loaded,j)) {
493 move_this_load = IRQ_DELTA(max_loaded,j);
494 selected_irq = j;
495 }
496 }
497 if (selected_irq == -1) {
498 goto tryanothercpu;
499 }
500
501 imbalance = move_this_load;
502
503 /* For physical_balance case, we accumlated both load
504 * values in the one of the siblings cpu_irq[],
505 * to use the same code for physical and logical processors
506 * as much as possible.
507 *
508 * NOTE: the cpu_irq[] array holds the sum of the load for
509 * sibling A and sibling B in the slot for the lowest numbered
510 * sibling (A), _AND_ the load for sibling B in the slot for
511 * the higher numbered sibling.
512 *
513 * We seek the least loaded sibling by making the comparison
514 * (A+B)/2 vs B
515 */
516 load = CPU_IRQ(min_loaded) >> 1;
517 for_each_cpu_mask(j, cpu_sibling_map[min_loaded]) {
518 if (load > CPU_IRQ(j)) {
519 /* This won't change cpu_sibling_map[min_loaded] */
520 load = CPU_IRQ(j);
521 min_loaded = j;
522 }
523 }
524
525 cpus_and(allowed_mask, cpu_online_map, irq_affinity[selected_irq]);
526 target_cpu_mask = cpumask_of_cpu(min_loaded);
527 cpus_and(tmp, target_cpu_mask, allowed_mask);
528
529 if (!cpus_empty(tmp)) {
530 irq_desc_t *desc = irq_desc + selected_irq;
531 unsigned long flags;
532
533 Dprintk("irq = %d moved to cpu = %d\n",
534 selected_irq, min_loaded);
535 /* mark for change destination */
536 spin_lock_irqsave(&desc->lock, flags);
537 pending_irq_balance_cpumask[selected_irq] =
538 cpumask_of_cpu(min_loaded);
539 spin_unlock_irqrestore(&desc->lock, flags);
540 /* Since we made a change, come back sooner to
541 * check for more variation.
542 */
543 balanced_irq_interval = max((long)MIN_BALANCED_IRQ_INTERVAL,
544 balanced_irq_interval - BALANCED_IRQ_LESS_DELTA);
545 return;
546 }
547 goto tryanotherirq;
548
549not_worth_the_effort:
550 /*
551 * if we did not find an IRQ to move, then adjust the time interval
552 * upward
553 */
554 balanced_irq_interval = min((long)MAX_BALANCED_IRQ_INTERVAL,
555 balanced_irq_interval + BALANCED_IRQ_MORE_DELTA);
556 Dprintk("IRQ worth rotating not found\n");
557 return;
558}
559
560static int balanced_irq(void *unused)
561{
562 int i;
563 unsigned long prev_balance_time = jiffies;
564 long time_remaining = balanced_irq_interval;
565
566 daemonize("kirqd");
567
568 /* push everything to CPU 0 to give us a starting point. */
569 for (i = 0 ; i < NR_IRQS ; i++) {
570 pending_irq_balance_cpumask[i] = cpumask_of_cpu(0);
571 }
572
573 for ( ; ; ) {
574 set_current_state(TASK_INTERRUPTIBLE);
575 time_remaining = schedule_timeout(time_remaining);
576 try_to_freeze(PF_FREEZE);
577 if (time_after(jiffies,
578 prev_balance_time+balanced_irq_interval)) {
579 do_irq_balance();
580 prev_balance_time = jiffies;
581 time_remaining = balanced_irq_interval;
582 }
583 }
584 return 0;
585}
586
587static int __init balanced_irq_init(void)
588{
589 int i;
590 struct cpuinfo_x86 *c;
591 cpumask_t tmp;
592
593 cpus_shift_right(tmp, cpu_online_map, 2);
594 c = &boot_cpu_data;
595 /* When not overwritten by the command line ask subarchitecture. */
596 if (irqbalance_disabled == IRQBALANCE_CHECK_ARCH)
597 irqbalance_disabled = NO_BALANCE_IRQ;
598 if (irqbalance_disabled)
599 return 0;
600
601 /* disable irqbalance completely if there is only one processor online */
602 if (num_online_cpus() < 2) {
603 irqbalance_disabled = 1;
604 return 0;
605 }
606 /*
607 * Enable physical balance only if more than 1 physical processor
608 * is present
609 */
610 if (smp_num_siblings > 1 && !cpus_empty(tmp))
611 physical_balance = 1;
612
613 for (i = 0; i < NR_CPUS; i++) {
614 if (!cpu_online(i))
615 continue;
616 irq_cpu_data[i].irq_delta = kmalloc(sizeof(unsigned long) * NR_IRQS, GFP_KERNEL);
617 irq_cpu_data[i].last_irq = kmalloc(sizeof(unsigned long) * NR_IRQS, GFP_KERNEL);
618 if (irq_cpu_data[i].irq_delta == NULL || irq_cpu_data[i].last_irq == NULL) {
619 printk(KERN_ERR "balanced_irq_init: out of memory");
620 goto failed;
621 }
622 memset(irq_cpu_data[i].irq_delta,0,sizeof(unsigned long) * NR_IRQS);
623 memset(irq_cpu_data[i].last_irq,0,sizeof(unsigned long) * NR_IRQS);
624 }
625
626 printk(KERN_INFO "Starting balanced_irq\n");
627 if (kernel_thread(balanced_irq, NULL, CLONE_KERNEL) >= 0)
628 return 0;
629 else
630 printk(KERN_ERR "balanced_irq_init: failed to spawn balanced_irq");
631failed:
632 for (i = 0; i < NR_CPUS; i++) {
633 if(irq_cpu_data[i].irq_delta)
634 kfree(irq_cpu_data[i].irq_delta);
635 if(irq_cpu_data[i].last_irq)
636 kfree(irq_cpu_data[i].last_irq);
637 }
638 return 0;
639}
640
641int __init irqbalance_disable(char *str)
642{
643 irqbalance_disabled = 1;
644 return 0;
645}
646
647__setup("noirqbalance", irqbalance_disable);
648
649static inline void move_irq(int irq)
650{
651 /* note - we hold the desc->lock */
652 if (unlikely(!cpus_empty(pending_irq_balance_cpumask[irq]))) {
653 set_ioapic_affinity_irq(irq, pending_irq_balance_cpumask[irq]);
654 cpus_clear(pending_irq_balance_cpumask[irq]);
655 }
656}
657
658late_initcall(balanced_irq_init);
659
660#else /* !CONFIG_IRQBALANCE */
661static inline void move_irq(int irq) { }
662#endif /* CONFIG_IRQBALANCE */
663
664#ifndef CONFIG_SMP
665void fastcall send_IPI_self(int vector)
666{
667 unsigned int cfg;
668
669 /*
670 * Wait for idle.
671 */
672 apic_wait_icr_idle();
673 cfg = APIC_DM_FIXED | APIC_DEST_SELF | vector | APIC_DEST_LOGICAL;
674 /*
675 * Send the IPI. The write to APIC_ICR fires this off.
676 */
677 apic_write_around(APIC_ICR, cfg);
678}
679#endif /* !CONFIG_SMP */
680
681
682/*
683 * support for broken MP BIOSs, enables hand-redirection of PIRQ0-7 to
684 * specific CPU-side IRQs.
685 */
686
687#define MAX_PIRQS 8
688static int pirq_entries [MAX_PIRQS];
689static int pirqs_enabled;
690int skip_ioapic_setup;
691
692static int __init ioapic_setup(char *str)
693{
694 skip_ioapic_setup = 1;
695 return 1;
696}
697
698__setup("noapic", ioapic_setup);
699
700static int __init ioapic_pirq_setup(char *str)
701{
702 int i, max;
703 int ints[MAX_PIRQS+1];
704
705 get_options(str, ARRAY_SIZE(ints), ints);
706
707 for (i = 0; i < MAX_PIRQS; i++)
708 pirq_entries[i] = -1;
709
710 pirqs_enabled = 1;
711 apic_printk(APIC_VERBOSE, KERN_INFO
712 "PIRQ redirection, working around broken MP-BIOS.\n");
713 max = MAX_PIRQS;
714 if (ints[0] < MAX_PIRQS)
715 max = ints[0];
716
717 for (i = 0; i < max; i++) {
718 apic_printk(APIC_VERBOSE, KERN_DEBUG
719 "... PIRQ%d -> IRQ %d\n", i, ints[i+1]);
720 /*
721 * PIRQs are mapped upside down, usually.
722 */
723 pirq_entries[MAX_PIRQS-i-1] = ints[i+1];
724 }
725 return 1;
726}
727
728__setup("pirq=", ioapic_pirq_setup);
729
730/*
731 * Find the IRQ entry number of a certain pin.
732 */
733static int find_irq_entry(int apic, int pin, int type)
734{
735 int i;
736
737 for (i = 0; i < mp_irq_entries; i++)
738 if (mp_irqs[i].mpc_irqtype == type &&
739 (mp_irqs[i].mpc_dstapic == mp_ioapics[apic].mpc_apicid ||
740 mp_irqs[i].mpc_dstapic == MP_APIC_ALL) &&
741 mp_irqs[i].mpc_dstirq == pin)
742 return i;
743
744 return -1;
745}
746
747/*
748 * Find the pin to which IRQ[irq] (ISA) is connected
749 */
750static int find_isa_irq_pin(int irq, int type)
751{
752 int i;
753
754 for (i = 0; i < mp_irq_entries; i++) {
755 int lbus = mp_irqs[i].mpc_srcbus;
756
757 if ((mp_bus_id_to_type[lbus] == MP_BUS_ISA ||
758 mp_bus_id_to_type[lbus] == MP_BUS_EISA ||
759 mp_bus_id_to_type[lbus] == MP_BUS_MCA ||
760 mp_bus_id_to_type[lbus] == MP_BUS_NEC98
761 ) &&
762 (mp_irqs[i].mpc_irqtype == type) &&
763 (mp_irqs[i].mpc_srcbusirq == irq))
764
765 return mp_irqs[i].mpc_dstirq;
766 }
767 return -1;
768}
769
770/*
771 * Find a specific PCI IRQ entry.
772 * Not an __init, possibly needed by modules
773 */
774static int pin_2_irq(int idx, int apic, int pin);
775
776int IO_APIC_get_PCI_irq_vector(int bus, int slot, int pin)
777{
778 int apic, i, best_guess = -1;
779
780 apic_printk(APIC_DEBUG, "querying PCI -> IRQ mapping bus:%d, "
781 "slot:%d, pin:%d.\n", bus, slot, pin);
782 if (mp_bus_id_to_pci_bus[bus] == -1) {
783 printk(KERN_WARNING "PCI BIOS passed nonexistent PCI bus %d!\n", bus);
784 return -1;
785 }
786 for (i = 0; i < mp_irq_entries; i++) {
787 int lbus = mp_irqs[i].mpc_srcbus;
788
789 for (apic = 0; apic < nr_ioapics; apic++)
790 if (mp_ioapics[apic].mpc_apicid == mp_irqs[i].mpc_dstapic ||
791 mp_irqs[i].mpc_dstapic == MP_APIC_ALL)
792 break;
793
794 if ((mp_bus_id_to_type[lbus] == MP_BUS_PCI) &&
795 !mp_irqs[i].mpc_irqtype &&
796 (bus == lbus) &&
797 (slot == ((mp_irqs[i].mpc_srcbusirq >> 2) & 0x1f))) {
798 int irq = pin_2_irq(i,apic,mp_irqs[i].mpc_dstirq);
799
800 if (!(apic || IO_APIC_IRQ(irq)))
801 continue;
802
803 if (pin == (mp_irqs[i].mpc_srcbusirq & 3))
804 return irq;
805 /*
806 * Use the first all-but-pin matching entry as a
807 * best-guess fuzzy result for broken mptables.
808 */
809 if (best_guess < 0)
810 best_guess = irq;
811 }
812 }
813 return best_guess;
814}
129f6946 815EXPORT_SYMBOL(IO_APIC_get_PCI_irq_vector);
1da177e4
LT		 816
817/*
818 * This function currently is only a helper for the i386 smp boot process where
819 * we need to reprogram the ioredtbls to cater for the cpus which have come online
820 * so mask in all cases should simply be TARGET_CPUS
821 */
822void __init setup_ioapic_dest(void)
823{
824 int pin, ioapic, irq, irq_entry;
825
826 if (skip_ioapic_setup == 1)
827 return;
828
829 for (ioapic = 0; ioapic < nr_ioapics; ioapic++) {
830 for (pin = 0; pin < nr_ioapic_registers[ioapic]; pin++) {
831 irq_entry = find_irq_entry(ioapic, pin, mp_INT);
832 if (irq_entry == -1)
833 continue;
834 irq = pin_2_irq(irq_entry, ioapic, pin);
835 set_ioapic_affinity_irq(irq, TARGET_CPUS);
836 }
837
838 }
839}
840
841/*
842 * EISA Edge/Level control register, ELCR
843 */
844static int EISA_ELCR(unsigned int irq)
845{
846 if (irq < 16) {
847 unsigned int port = 0x4d0 + (irq >> 3);
848 return (inb(port) >> (irq & 7)) & 1;
849 }
850 apic_printk(APIC_VERBOSE, KERN_INFO
851 "Broken MPtable reports ISA irq %d\n", irq);
852 return 0;
853}
854
855/* EISA interrupts are always polarity zero and can be edge or level
856 * trigger depending on the ELCR value. If an interrupt is listed as
857 * EISA conforming in the MP table, that means its trigger type must
858 * be read in from the ELCR */
859
860#define default_EISA_trigger(idx) (EISA_ELCR(mp_irqs[idx].mpc_srcbusirq))
861#define default_EISA_polarity(idx) (0)
862
863/* ISA interrupts are always polarity zero edge triggered,
864 * when listed as conforming in the MP table. */
865
866#define default_ISA_trigger(idx) (0)
867#define default_ISA_polarity(idx) (0)
868
869/* PCI interrupts are always polarity one level triggered,
870 * when listed as conforming in the MP table. */
871
872#define default_PCI_trigger(idx) (1)
873#define default_PCI_polarity(idx) (1)
874
875/* MCA interrupts are always polarity zero level triggered,
876 * when listed as conforming in the MP table. */
877
878#define default_MCA_trigger(idx) (1)
879#define default_MCA_polarity(idx) (0)
880
881/* NEC98 interrupts are always polarity zero edge triggered,
882 * when listed as conforming in the MP table. */
883
884#define default_NEC98_trigger(idx) (0)
885#define default_NEC98_polarity(idx) (0)
886
887static int __init MPBIOS_polarity(int idx)
888{
889 int bus = mp_irqs[idx].mpc_srcbus;
890 int polarity;
891
892 /*
893 * Determine IRQ line polarity (high active or low active):
894 */
895 switch (mp_irqs[idx].mpc_irqflag & 3)
896 {
897 case 0: /* conforms, ie. bus-type dependent polarity */
898 {
899 switch (mp_bus_id_to_type[bus])
900 {
901 case MP_BUS_ISA: /* ISA pin */
902 {
903 polarity = default_ISA_polarity(idx);
904 break;
905 }
906 case MP_BUS_EISA: /* EISA pin */
907 {
908 polarity = default_EISA_polarity(idx);
909 break;
910 }
911 case MP_BUS_PCI: /* PCI pin */
912 {
913 polarity = default_PCI_polarity(idx);
914 break;
915 }
916 case MP_BUS_MCA: /* MCA pin */
917 {
918 polarity = default_MCA_polarity(idx);
919 break;
920 }
921 case MP_BUS_NEC98: /* NEC 98 pin */
922 {
923 polarity = default_NEC98_polarity(idx);
924 break;
925 }
926 default:
927 {
928 printk(KERN_WARNING "broken BIOS!!\n");
929 polarity = 1;
930 break;
931 }
932 }
933 break;
934 }
935 case 1: /* high active */
936 {
937 polarity = 0;
938 break;
939 }
940 case 2: /* reserved */
941 {
942 printk(KERN_WARNING "broken BIOS!!\n");
943 polarity = 1;
944 break;
945 }
946 case 3: /* low active */
947 {
948 polarity = 1;
949 break;
950 }
951 default: /* invalid */
952 {
953 printk(KERN_WARNING "broken BIOS!!\n");
954 polarity = 1;
955 break;
956 }
957 }
958 return polarity;
959}
960
961static int MPBIOS_trigger(int idx)
962{
963 int bus = mp_irqs[idx].mpc_srcbus;
964 int trigger;
965
966 /*
967 * Determine IRQ trigger mode (edge or level sensitive):
968 */
969 switch ((mp_irqs[idx].mpc_irqflag>>2) & 3)
970 {
971 case 0: /* conforms, ie. bus-type dependent */
972 {
973 switch (mp_bus_id_to_type[bus])
974 {
975 case MP_BUS_ISA: /* ISA pin */
976 {
977 trigger = default_ISA_trigger(idx);
978 break;
979 }
980 case MP_BUS_EISA: /* EISA pin */
981 {
982 trigger = default_EISA_trigger(idx);
983 break;
984 }
985 case MP_BUS_PCI: /* PCI pin */
986 {
987 trigger = default_PCI_trigger(idx);
988 break;
989 }
990 case MP_BUS_MCA: /* MCA pin */
991 {
992 trigger = default_MCA_trigger(idx);
993 break;
994 }
995 case MP_BUS_NEC98: /* NEC 98 pin */
996 {
997 trigger = default_NEC98_trigger(idx);
998 break;
999 }
1000 default:
1001 {
1002 printk(KERN_WARNING "broken BIOS!!\n");
1003 trigger = 1;
1004 break;
1005 }
1006 }
1007 break;
1008 }
1009 case 1: /* edge */
1010 {
1011 trigger = 0;
1012 break;
1013 }
1014 case 2: /* reserved */
1015 {
1016 printk(KERN_WARNING "broken BIOS!!\n");
1017 trigger = 1;
1018 break;
1019 }
1020 case 3: /* level */
1021 {
1022 trigger = 1;
1023 break;
1024 }
1025 default: /* invalid */
1026 {
1027 printk(KERN_WARNING "broken BIOS!!\n");
1028 trigger = 0;
1029 break;
1030 }
1031 }
1032 return trigger;
1033}
1034
1035static inline int irq_polarity(int idx)
1036{
1037 return MPBIOS_polarity(idx);
1038}
1039
1040static inline int irq_trigger(int idx)
1041{
1042 return MPBIOS_trigger(idx);
1043}
1044
1045static int pin_2_irq(int idx, int apic, int pin)
1046{
1047 int irq, i;
1048 int bus = mp_irqs[idx].mpc_srcbus;
1049
1050 /*
1051 * Debugging check, we are in big trouble if this message pops up!
1052 */
1053 if (mp_irqs[idx].mpc_dstirq != pin)
1054 printk(KERN_ERR "broken BIOS or MPTABLE parser, ayiee!!\n");
1055
1056 switch (mp_bus_id_to_type[bus])
1057 {
1058 case MP_BUS_ISA: /* ISA pin */
1059 case MP_BUS_EISA:
1060 case MP_BUS_MCA:
1061 case MP_BUS_NEC98:
1062 {
1063 irq = mp_irqs[idx].mpc_srcbusirq;
1064 break;
1065 }
1066 case MP_BUS_PCI: /* PCI pin */
1067 {
1068 /*
1069 * PCI IRQs are mapped in order
1070 */
1071 i = irq = 0;
1072 while (i < apic)
1073 irq += nr_ioapic_registers[i++];
1074 irq += pin;
1075
1076 /*
1077 * For MPS mode, so far only needed by ES7000 platform
1078 */
1079 if (ioapic_renumber_irq)
1080 irq = ioapic_renumber_irq(apic, irq);
1081
1082 break;
1083 }
1084 default:
1085 {
1086 printk(KERN_ERR "unknown bus type %d.\n",bus);
1087 irq = 0;
1088 break;
1089 }
1090 }
1091
1092 /*
1093 * PCI IRQ command line redirection. Yes, limits are hardcoded.
1094 */
1095 if ((pin >= 16) && (pin <= 23)) {
1096 if (pirq_entries[pin-16] != -1) {
1097 if (!pirq_entries[pin-16]) {
1098 apic_printk(APIC_VERBOSE, KERN_DEBUG
1099 "disabling PIRQ%d\n", pin-16);
1100 } else {
1101 irq = pirq_entries[pin-16];
1102 apic_printk(APIC_VERBOSE, KERN_DEBUG
1103 "using PIRQ%d -> IRQ %d\n",
1104 pin-16, irq);
1105 }
1106 }
1107 }
1108 return irq;
1109}
1110
1111static inline int IO_APIC_irq_trigger(int irq)
1112{
1113 int apic, idx, pin;
1114
1115 for (apic = 0; apic < nr_ioapics; apic++) {
1116 for (pin = 0; pin < nr_ioapic_registers[apic]; pin++) {
1117 idx = find_irq_entry(apic,pin,mp_INT);
1118 if ((idx != -1) && (irq == pin_2_irq(idx,apic,pin)))
1119 return irq_trigger(idx);
1120 }
1121 }
1122 /*
1123 * nonexistent IRQs are edge default
1124 */
1125 return 0;
1126}
1127
1128/* irq_vectors is indexed by the sum of all RTEs in all I/O APICs. */
1129u8 irq_vector[NR_IRQ_VECTORS] = { FIRST_DEVICE_VECTOR , 0 };
1130
1131int assign_irq_vector(int irq)
1132{
1133 static int current_vector = FIRST_DEVICE_VECTOR, offset = 0;
1134
1135 BUG_ON(irq >= NR_IRQ_VECTORS);
1136 if (irq != AUTO_ASSIGN && IO_APIC_VECTOR(irq) > 0)
1137 return IO_APIC_VECTOR(irq);
1138next:
1139 current_vector += 8;
1140 if (current_vector == SYSCALL_VECTOR)
1141 goto next;
1142
1143 if (current_vector >= FIRST_SYSTEM_VECTOR) {
1144 offset++;
1145 if (!(offset%8))
1146 return -ENOSPC;
1147 current_vector = FIRST_DEVICE_VECTOR + offset;
1148 }
1149
1150 vector_irq[current_vector] = irq;
1151 if (irq != AUTO_ASSIGN)
1152 IO_APIC_VECTOR(irq) = current_vector;
1153
1154 return current_vector;
1155}
1156
1157static struct hw_interrupt_type ioapic_level_type;
1158static struct hw_interrupt_type ioapic_edge_type;
1159
1160#define IOAPIC_AUTO -1
1161#define IOAPIC_EDGE 0
1162#define IOAPIC_LEVEL 1
1163
1164static inline void ioapic_register_intr(int irq, int vector, unsigned long trigger)
1165{
1166 if (use_pci_vector() && !platform_legacy_irq(irq)) {
1167 if ((trigger == IOAPIC_AUTO && IO_APIC_irq_trigger(irq)) ||
1168 trigger == IOAPIC_LEVEL)
1169 irq_desc[vector].handler = &ioapic_level_type;
1170 else
1171 irq_desc[vector].handler = &ioapic_edge_type;
1172 set_intr_gate(vector, interrupt[vector]);
1173 } else {
1174 if ((trigger == IOAPIC_AUTO && IO_APIC_irq_trigger(irq)) ||
1175 trigger == IOAPIC_LEVEL)
1176 irq_desc[irq].handler = &ioapic_level_type;
1177 else
1178 irq_desc[irq].handler = &ioapic_edge_type;
1179 set_intr_gate(vector, interrupt[irq]);
1180 }
1181}
1182
1183static void __init setup_IO_APIC_irqs(void)
1184{
1185 struct IO_APIC_route_entry entry;
1186 int apic, pin, idx, irq, first_notcon = 1, vector;
1187 unsigned long flags;
1188
1189 apic_printk(APIC_VERBOSE, KERN_DEBUG "init IO_APIC IRQs\n");
1190
1191 for (apic = 0; apic < nr_ioapics; apic++) {
1192 for (pin = 0; pin < nr_ioapic_registers[apic]; pin++) {
1193
1194 /*
1195 * add it to the IO-APIC irq-routing table:
1196 */
1197 memset(&entry,0,sizeof(entry));
1198
1199 entry.delivery_mode = INT_DELIVERY_MODE;
1200 entry.dest_mode = INT_DEST_MODE;
1201 entry.mask = 0; /* enable IRQ */
1202 entry.dest.logical.logical_dest =
1203 cpu_mask_to_apicid(TARGET_CPUS);
1204
1205 idx = find_irq_entry(apic,pin,mp_INT);
1206 if (idx == -1) {
1207 if (first_notcon) {
1208 apic_printk(APIC_VERBOSE, KERN_DEBUG
1209 " IO-APIC (apicid-pin) %d-%d",
1210 mp_ioapics[apic].mpc_apicid,
1211 pin);
1212 first_notcon = 0;
1213 } else
1214 apic_printk(APIC_VERBOSE, ", %d-%d",
1215 mp_ioapics[apic].mpc_apicid, pin);
1216 continue;
1217 }
1218
1219 entry.trigger = irq_trigger(idx);
1220 entry.polarity = irq_polarity(idx);
1221
1222 if (irq_trigger(idx)) {
1223 entry.trigger = 1;
1224 entry.mask = 1;
1225 }
1226
1227 irq = pin_2_irq(idx, apic, pin);
1228 /*
1229 * skip adding the timer int on secondary nodes, which causes
1230 * a small but painful rift in the time-space continuum
1231 */
1232 if (multi_timer_check(apic, irq))
1233 continue;
1234 else
1235 add_pin_to_irq(irq, apic, pin);
1236
1237 if (!apic && !IO_APIC_IRQ(irq))
1238 continue;
1239
1240 if (IO_APIC_IRQ(irq)) {
1241 vector = assign_irq_vector(irq);
1242 entry.vector = vector;
1243 ioapic_register_intr(irq, vector, IOAPIC_AUTO);
1244
1245 if (!apic && (irq < 16))
1246 disable_8259A_irq(irq);
1247 }
1248 spin_lock_irqsave(&ioapic_lock, flags);
1249 io_apic_write(apic, 0x11+2*pin, *(((int *)&entry)+1));
1250 io_apic_write(apic, 0x10+2*pin, *(((int *)&entry)+0));
1251 spin_unlock_irqrestore(&ioapic_lock, flags);
1252 }
1253 }
1254
1255 if (!first_notcon)
1256 apic_printk(APIC_VERBOSE, " not connected.\n");
1257}
1258
1259/*
1260 * Set up the 8259A-master output pin:
1261 */
1262static void __init setup_ExtINT_IRQ0_pin(unsigned int pin, int vector)
1263{
1264 struct IO_APIC_route_entry entry;
1265 unsigned long flags;
1266
1267 memset(&entry,0,sizeof(entry));
1268
1269 disable_8259A_irq(0);
1270
1271 /* mask LVT0 */
1272 apic_write_around(APIC_LVT0, APIC_LVT_MASKED | APIC_DM_EXTINT);
1273
1274 /*
1275 * We use logical delivery to get the timer IRQ
1276 * to the first CPU.
1277 */
1278 entry.dest_mode = INT_DEST_MODE;
1279 entry.mask = 0; /* unmask IRQ now */
1280 entry.dest.logical.logical_dest = cpu_mask_to_apicid(TARGET_CPUS);
1281 entry.delivery_mode = INT_DELIVERY_MODE;
1282 entry.polarity = 0;
1283 entry.trigger = 0;
1284 entry.vector = vector;
1285
1286 /*
1287 * The timer IRQ doesn't have to know that behind the
1288 * scene we have a 8259A-master in AEOI mode ...
1289 */
1290 irq_desc[0].handler = &ioapic_edge_type;
1291
1292 /*
1293 * Add it to the IO-APIC irq-routing table:
1294 */
1295 spin_lock_irqsave(&ioapic_lock, flags);
1296 io_apic_write(0, 0x11+2*pin, *(((int *)&entry)+1));
1297 io_apic_write(0, 0x10+2*pin, *(((int *)&entry)+0));
1298 spin_unlock_irqrestore(&ioapic_lock, flags);
1299
1300 enable_8259A_irq(0);
1301}
1302
1303static inline void UNEXPECTED_IO_APIC(void)
1304{
1305}
1306
1307void __init print_IO_APIC(void)
1308{
1309 int apic, i;
1310 union IO_APIC_reg_00 reg_00;
1311 union IO_APIC_reg_01 reg_01;
1312 union IO_APIC_reg_02 reg_02;
1313 union IO_APIC_reg_03 reg_03;
1314 unsigned long flags;
1315
1316 if (apic_verbosity == APIC_QUIET)
1317 return;
1318
1319 printk(KERN_DEBUG "number of MP IRQ sources: %d.\n", mp_irq_entries);
1320 for (i = 0; i < nr_ioapics; i++)
1321 printk(KERN_DEBUG "number of IO-APIC #%d registers: %d.\n",
1322 mp_ioapics[i].mpc_apicid, nr_ioapic_registers[i]);
1323
1324 /*
1325 * We are a bit conservative about what we expect. We have to
1326 * know about every hardware change ASAP.
1327 */
1328 printk(KERN_INFO "testing the IO APIC.......................\n");
1329
1330 for (apic = 0; apic < nr_ioapics; apic++) {
1331
1332 spin_lock_irqsave(&ioapic_lock, flags);
1333 reg_00.raw = io_apic_read(apic, 0);
1334 reg_01.raw = io_apic_read(apic, 1);
1335 if (reg_01.bits.version >= 0x10)
1336 reg_02.raw = io_apic_read(apic, 2);
1337 if (reg_01.bits.version >= 0x20)
1338 reg_03.raw = io_apic_read(apic, 3);
1339 spin_unlock_irqrestore(&ioapic_lock, flags);
1340
1341 printk(KERN_DEBUG "IO APIC #%d......\n", mp_ioapics[apic].mpc_apicid);
1342 printk(KERN_DEBUG ".... register #00: %08X\n", reg_00.raw);
1343 printk(KERN_DEBUG "....... : physical APIC id: %02X\n", reg_00.bits.ID);
1344 printk(KERN_DEBUG "....... : Delivery Type: %X\n", reg_00.bits.delivery_type);
1345 printk(KERN_DEBUG "....... : LTS : %X\n", reg_00.bits.LTS);
1346 if (reg_00.bits.ID >= get_physical_broadcast())
1347 UNEXPECTED_IO_APIC();
1348 if (reg_00.bits.__reserved_1 || reg_00.bits.__reserved_2)
1349 UNEXPECTED_IO_APIC();
1350
1351 printk(KERN_DEBUG ".... register #01: %08X\n", reg_01.raw);
1352 printk(KERN_DEBUG "....... : max redirection entries: %04X\n", reg_01.bits.entries);
1353 if ( (reg_01.bits.entries != 0x0f) && /* older (Neptune) boards */
1354 (reg_01.bits.entries != 0x17) && /* typical ISA+PCI boards */
1355 (reg_01.bits.entries != 0x1b) && /* Compaq Proliant boards */
1356 (reg_01.bits.entries != 0x1f) && /* dual Xeon boards */
1357 (reg_01.bits.entries != 0x22) && /* bigger Xeon boards */
1358 (reg_01.bits.entries != 0x2E) &&
1359 (reg_01.bits.entries != 0x3F)
1360 )
1361 UNEXPECTED_IO_APIC();
1362
1363 printk(KERN_DEBUG "....... : PRQ implemented: %X\n", reg_01.bits.PRQ);
1364 printk(KERN_DEBUG "....... : IO APIC version: %04X\n", reg_01.bits.version);
1365 if ( (reg_01.bits.version != 0x01) && /* 82489DX IO-APICs */
1366 (reg_01.bits.version != 0x10) && /* oldest IO-APICs */
1367 (reg_01.bits.version != 0x11) && /* Pentium/Pro IO-APICs */
1368 (reg_01.bits.version != 0x13) && /* Xeon IO-APICs */
1369 (reg_01.bits.version != 0x20) /* Intel P64H (82806 AA) */
1370 )
1371 UNEXPECTED_IO_APIC();
1372 if (reg_01.bits.__reserved_1 || reg_01.bits.__reserved_2)
1373 UNEXPECTED_IO_APIC();
1374
1375 /*
1376 * Some Intel chipsets with IO APIC VERSION of 0x1? don't have reg_02,
1377 * but the value of reg_02 is read as the previous read register
1378 * value, so ignore it if reg_02 == reg_01.
1379 */
1380 if (reg_01.bits.version >= 0x10 && reg_02.raw != reg_01.raw) {
1381 printk(KERN_DEBUG ".... register #02: %08X\n", reg_02.raw);
1382 printk(KERN_DEBUG "....... : arbitration: %02X\n", reg_02.bits.arbitration);
1383 if (reg_02.bits.__reserved_1 || reg_02.bits.__reserved_2)
1384 UNEXPECTED_IO_APIC();
1385 }
1386
1387 /*
1388 * Some Intel chipsets with IO APIC VERSION of 0x2? don't have reg_02
1389 * or reg_03, but the value of reg_0[23] is read as the previous read
1390 * register value, so ignore it if reg_03 == reg_0[12].
1391 */
1392 if (reg_01.bits.version >= 0x20 && reg_03.raw != reg_02.raw &&
1393 reg_03.raw != reg_01.raw) {
1394 printk(KERN_DEBUG ".... register #03: %08X\n", reg_03.raw);
1395 printk(KERN_DEBUG "....... : Boot DT : %X\n", reg_03.bits.boot_DT);
1396 if (reg_03.bits.__reserved_1)
1397 UNEXPECTED_IO_APIC();
1398 }
1399
1400 printk(KERN_DEBUG ".... IRQ redirection table:\n");
1401
1402 printk(KERN_DEBUG " NR Log Phy Mask Trig IRR Pol"
1403 " Stat Dest Deli Vect: \n");
1404
1405 for (i = 0; i <= reg_01.bits.entries; i++) {
1406 struct IO_APIC_route_entry entry;
1407
1408 spin_lock_irqsave(&ioapic_lock, flags);
1409 *(((int *)&entry)+0) = io_apic_read(apic, 0x10+i*2);
1410 *(((int *)&entry)+1) = io_apic_read(apic, 0x11+i*2);
1411 spin_unlock_irqrestore(&ioapic_lock, flags);
1412
1413 printk(KERN_DEBUG " %02x %03X %02X ",
1414 i,
1415 entry.dest.logical.logical_dest,
1416 entry.dest.physical.physical_dest
1417 );
1418
1419 printk("%1d %1d %1d %1d %1d %1d %1d %02X\n",
1420 entry.mask,
1421 entry.trigger,
1422 entry.irr,
1423 entry.polarity,
1424 entry.delivery_status,
1425 entry.dest_mode,
1426 entry.delivery_mode,
1427 entry.vector
1428 );
1429 }
1430 }
1431 if (use_pci_vector())
1432 printk(KERN_INFO "Using vector-based indexing\n");
1433 printk(KERN_DEBUG "IRQ to pin mappings:\n");
1434 for (i = 0; i < NR_IRQS; i++) {
1435 struct irq_pin_list *entry = irq_2_pin + i;
1436 if (entry->pin < 0)
1437 continue;
1438 if (use_pci_vector() && !platform_legacy_irq(i))
1439 printk(KERN_DEBUG "IRQ%d ", IO_APIC_VECTOR(i));
1440 else
1441 printk(KERN_DEBUG "IRQ%d ", i);
1442 for (;;) {
1443 printk("-> %d:%d", entry->apic, entry->pin);
1444 if (!entry->next)
1445 break;
1446 entry = irq_2_pin + entry->next;
1447 }
1448 printk("\n");
1449 }
1450
1451 printk(KERN_INFO ".................................... done.\n");
1452
1453 return;
1454}
1455
1456#if 0
1457
1458static void print_APIC_bitfield (int base)
1459{
1460 unsigned int v;
1461 int i, j;
1462
1463 if (apic_verbosity == APIC_QUIET)
1464 return;
1465
1466 printk(KERN_DEBUG "0123456789abcdef0123456789abcdef\n" KERN_DEBUG);
1467 for (i = 0; i < 8; i++) {
1468 v = apic_read(base + i*0x10);
1469 for (j = 0; j < 32; j++) {
1470 if (v & (1<<j))
1471 printk("1");
1472 else
1473 printk("0");
1474 }
1475 printk("\n");
1476 }
1477}
1478
1479void /*__init*/ print_local_APIC(void * dummy)
1480{
1481 unsigned int v, ver, maxlvt;
1482
1483 if (apic_verbosity == APIC_QUIET)
1484 return;
1485
1486 printk("\n" KERN_DEBUG "printing local APIC contents on CPU#%d/%d:\n",
1487 smp_processor_id(), hard_smp_processor_id());
1488 v = apic_read(APIC_ID);
1489 printk(KERN_INFO "... APIC ID: %08x (%01x)\n", v, GET_APIC_ID(v));
1490 v = apic_read(APIC_LVR);
1491 printk(KERN_INFO "... APIC VERSION: %08x\n", v);
1492 ver = GET_APIC_VERSION(v);
1493 maxlvt = get_maxlvt();
1494
1495 v = apic_read(APIC_TASKPRI);
1496 printk(KERN_DEBUG "... APIC TASKPRI: %08x (%02x)\n", v, v & APIC_TPRI_MASK);
1497
1498 if (APIC_INTEGRATED(ver)) { /* !82489DX */
1499 v = apic_read(APIC_ARBPRI);
1500 printk(KERN_DEBUG "... APIC ARBPRI: %08x (%02x)\n", v,
1501 v & APIC_ARBPRI_MASK);
1502 v = apic_read(APIC_PROCPRI);
1503 printk(KERN_DEBUG "... APIC PROCPRI: %08x\n", v);
1504 }
1505
1506 v = apic_read(APIC_EOI);
1507 printk(KERN_DEBUG "... APIC EOI: %08x\n", v);
1508 v = apic_read(APIC_RRR);
1509 printk(KERN_DEBUG "... APIC RRR: %08x\n", v);
1510 v = apic_read(APIC_LDR);
1511 printk(KERN_DEBUG "... APIC LDR: %08x\n", v);
1512 v = apic_read(APIC_DFR);
1513 printk(KERN_DEBUG "... APIC DFR: %08x\n", v);
1514 v = apic_read(APIC_SPIV);
1515 printk(KERN_DEBUG "... APIC SPIV: %08x\n", v);
1516
1517 printk(KERN_DEBUG "... APIC ISR field:\n");
1518 print_APIC_bitfield(APIC_ISR);
1519 printk(KERN_DEBUG "... APIC TMR field:\n");
1520 print_APIC_bitfield(APIC_TMR);
1521 printk(KERN_DEBUG "... APIC IRR field:\n");
1522 print_APIC_bitfield(APIC_IRR);
1523
1524 if (APIC_INTEGRATED(ver)) { /* !82489DX */
1525 if (maxlvt > 3) /* Due to the Pentium erratum 3AP. */
1526 apic_write(APIC_ESR, 0);
1527 v = apic_read(APIC_ESR);
1528 printk(KERN_DEBUG "... APIC ESR: %08x\n", v);
1529 }
1530
1531 v = apic_read(APIC_ICR);
1532 printk(KERN_DEBUG "... APIC ICR: %08x\n", v);
1533 v = apic_read(APIC_ICR2);
1534 printk(KERN_DEBUG "... APIC ICR2: %08x\n", v);
1535
1536 v = apic_read(APIC_LVTT);
1537 printk(KERN_DEBUG "... APIC LVTT: %08x\n", v);
1538
1539 if (maxlvt > 3) { /* PC is LVT#4. */
1540 v = apic_read(APIC_LVTPC);
1541 printk(KERN_DEBUG "... APIC LVTPC: %08x\n", v);
1542 }
1543 v = apic_read(APIC_LVT0);
1544 printk(KERN_DEBUG "... APIC LVT0: %08x\n", v);
1545 v = apic_read(APIC_LVT1);
1546 printk(KERN_DEBUG "... APIC LVT1: %08x\n", v);
1547
1548 if (maxlvt > 2) { /* ERR is LVT#3. */
1549 v = apic_read(APIC_LVTERR);
1550 printk(KERN_DEBUG "... APIC LVTERR: %08x\n", v);
1551 }
1552
1553 v = apic_read(APIC_TMICT);
1554 printk(KERN_DEBUG "... APIC TMICT: %08x\n", v);
1555 v = apic_read(APIC_TMCCT);
1556 printk(KERN_DEBUG "... APIC TMCCT: %08x\n", v);
1557 v = apic_read(APIC_TDCR);
1558 printk(KERN_DEBUG "... APIC TDCR: %08x\n", v);
1559 printk("\n");
1560}
1561
1562void print_all_local_APICs (void)
1563{
1564 on_each_cpu(print_local_APIC, NULL, 1, 1);
1565}
1566
1567void /*__init*/ print_PIC(void)
1568{
1569 extern spinlock_t i8259A_lock;
1570 unsigned int v;
1571 unsigned long flags;
1572
1573 if (apic_verbosity == APIC_QUIET)
1574 return;
1575
1576 printk(KERN_DEBUG "\nprinting PIC contents\n");
1577
1578 spin_lock_irqsave(&i8259A_lock, flags);
1579
1580 v = inb(0xa1) << 8 | inb(0x21);
1581 printk(KERN_DEBUG "... PIC IMR: %04x\n", v);
1582
1583 v = inb(0xa0) << 8 | inb(0x20);
1584 printk(KERN_DEBUG "... PIC IRR: %04x\n", v);
1585
1586 outb(0x0b,0xa0);
1587 outb(0x0b,0x20);
1588 v = inb(0xa0) << 8 | inb(0x20);
1589 outb(0x0a,0xa0);
1590 outb(0x0a,0x20);
1591
1592 spin_unlock_irqrestore(&i8259A_lock, flags);
1593
1594 printk(KERN_DEBUG "... PIC ISR: %04x\n", v);
1595
1596 v = inb(0x4d1) << 8 | inb(0x4d0);
1597 printk(KERN_DEBUG "... PIC ELCR: %04x\n", v);
1598}
1599
1600#endif /* 0 */
1601
1602static void __init enable_IO_APIC(void)
1603{
1604 union IO_APIC_reg_01 reg_01;
1605 int i;
1606 unsigned long flags;
1607
1608 for (i = 0; i < PIN_MAP_SIZE; i++) {
1609 irq_2_pin[i].pin = -1;
1610 irq_2_pin[i].next = 0;
1611 }
1612 if (!pirqs_enabled)
1613 for (i = 0; i < MAX_PIRQS; i++)
1614 pirq_entries[i] = -1;
1615
1616 /*
1617 * The number of IO-APIC IRQ registers (== #pins):
1618 */
1619 for (i = 0; i < nr_ioapics; i++) {
1620 spin_lock_irqsave(&ioapic_lock, flags);
1621 reg_01.raw = io_apic_read(i, 1);
1622 spin_unlock_irqrestore(&ioapic_lock, flags);
1623 nr_ioapic_registers[i] = reg_01.bits.entries+1;
1624 }
1625
1626 /*
1627 * Do not trust the IO-APIC being empty at bootup
1628 */
1629 clear_IO_APIC();
1630}
1631
1632/*
1633 * Not an __init, needed by the reboot code
1634 */
1635void disable_IO_APIC(void)
1636{
1637 /*
1638 * Clear the IO-APIC before rebooting:
1639 */
1640 clear_IO_APIC();
1641
1642 disconnect_bsp_APIC();
1643}
1644
1645/*
1646 * function to set the IO-APIC physical IDs based on the
1647 * values stored in the MPC table.
1648 *
1649 * by Matt Domsch <Matt_Domsch@dell.com> Tue Dec 21 12:25:05 CST 1999
1650 */
1651
1652#ifndef CONFIG_X86_NUMAQ
1653static void __init setup_ioapic_ids_from_mpc(void)
1654{
1655 union IO_APIC_reg_00 reg_00;
1656 physid_mask_t phys_id_present_map;
1657 int apic;
1658 int i;
1659 unsigned char old_id;
1660 unsigned long flags;
1661
ca05fea6
NP		 1662 /*
1663 * Don't check I/O APIC IDs for xAPIC systems. They have
1664 * no meaning without the serial APIC bus.
1665 */
1666 if (!(boot_cpu_data.x86_vendor == X86_VENDOR_INTEL && boot_cpu_data.x86 < 15))
1667 return;
1da177e4
LT		 1668 /*
1669 * This is broken; anything with a real cpu count has to
1670 * circumvent this idiocy regardless.
1671 */
1672 phys_id_present_map = ioapic_phys_id_map(phys_cpu_present_map);
1673
1674 /*
1675 * Set the IOAPIC ID to the value stored in the MPC table.
1676 */
1677 for (apic = 0; apic < nr_ioapics; apic++) {
1678
1679 /* Read the register 0 value */
1680 spin_lock_irqsave(&ioapic_lock, flags);
1681 reg_00.raw = io_apic_read(apic, 0);
1682 spin_unlock_irqrestore(&ioapic_lock, flags);
1683
1684 old_id = mp_ioapics[apic].mpc_apicid;
1685
1686 if (mp_ioapics[apic].mpc_apicid >= get_physical_broadcast()) {
1687 printk(KERN_ERR "BIOS bug, IO-APIC#%d ID is %d in the MPC table!...\n",
1688 apic, mp_ioapics[apic].mpc_apicid);
1689 printk(KERN_ERR "... fixing up to %d. (tell your hw vendor)\n",
1690 reg_00.bits.ID);
1691 mp_ioapics[apic].mpc_apicid = reg_00.bits.ID;
1692 }
1693
1da177e4
LT		 1694 /*
1695 * Sanity check, is the ID really free? Every APIC in a
1696 * system must have a unique ID or we get lots of nice
1697 * 'stuck on smp_invalidate_needed IPI wait' messages.
1698 */
1699 if (check_apicid_used(phys_id_present_map,
1700 mp_ioapics[apic].mpc_apicid)) {
1701 printk(KERN_ERR "BIOS bug, IO-APIC#%d ID %d is already used!...\n",
1702 apic, mp_ioapics[apic].mpc_apicid);
1703 for (i = 0; i < get_physical_broadcast(); i++)
1704 if (!physid_isset(i, phys_id_present_map))
1705 break;
1706 if (i >= get_physical_broadcast())
1707 panic("Max APIC ID exceeded!\n");
1708 printk(KERN_ERR "... fixing up to %d. (tell your hw vendor)\n",
1709 i);
1710 physid_set(i, phys_id_present_map);
1711 mp_ioapics[apic].mpc_apicid = i;
1712 } else {
1713 physid_mask_t tmp;
1714 tmp = apicid_to_cpu_present(mp_ioapics[apic].mpc_apicid);
1715 apic_printk(APIC_VERBOSE, "Setting %d in the "
1716 "phys_id_present_map\n",
1717 mp_ioapics[apic].mpc_apicid);
1718 physids_or(phys_id_present_map, phys_id_present_map, tmp);
1719 }
1720
1721
1722 /*
1723 * We need to adjust the IRQ routing table
1724 * if the ID changed.
1725 */
1726 if (old_id != mp_ioapics[apic].mpc_apicid)
1727 for (i = 0; i < mp_irq_entries; i++)
1728 if (mp_irqs[i].mpc_dstapic == old_id)
1729 mp_irqs[i].mpc_dstapic
1730 = mp_ioapics[apic].mpc_apicid;
1731
1732 /*
1733 * Read the right value from the MPC table and
1734 * write it into the ID register.
1735 */
1736 apic_printk(APIC_VERBOSE, KERN_INFO
1737 "...changing IO-APIC physical APIC ID to %d ...",
1738 mp_ioapics[apic].mpc_apicid);
1739
1740 reg_00.bits.ID = mp_ioapics[apic].mpc_apicid;
1741 spin_lock_irqsave(&ioapic_lock, flags);
1742 io_apic_write(apic, 0, reg_00.raw);
1743 spin_unlock_irqrestore(&ioapic_lock, flags);
1744
1745 /*
1746 * Sanity check
1747 */
1748 spin_lock_irqsave(&ioapic_lock, flags);
1749 reg_00.raw = io_apic_read(apic, 0);
1750 spin_unlock_irqrestore(&ioapic_lock, flags);
1751 if (reg_00.bits.ID != mp_ioapics[apic].mpc_apicid)
1752 printk("could not set ID!\n");
1753 else
1754 apic_printk(APIC_VERBOSE, " ok.\n");
1755 }
1756}
1757#else
1758static void __init setup_ioapic_ids_from_mpc(void) { }
1759#endif
1760
1761/*
1762 * There is a nasty bug in some older SMP boards, their mptable lies
1763 * about the timer IRQ. We do the following to work around the situation:
1764 *
1765 * - timer IRQ defaults to IO-APIC IRQ
1766 * - if this function detects that timer IRQs are defunct, then we fall
1767 * back to ISA timer IRQs
1768 */
1769static int __init timer_irq_works(void)
1770{
1771 unsigned long t1 = jiffies;
1772
1773 local_irq_enable();
1774 /* Let ten ticks pass... */
1775 mdelay((10 * 1000) / HZ);
1776
1777 /*
1778 * Expect a few ticks at least, to be sure some possible
1779 * glue logic does not lock up after one or two first
1780 * ticks in a non-ExtINT mode. Also the local APIC
1781 * might have cached one ExtINT interrupt. Finally, at
1782 * least one tick may be lost due to delays.
1783 */
1784 if (jiffies - t1 > 4)
1785 return 1;
1786
1787 return 0;
1788}
1789
1790/*
1791 * In the SMP+IOAPIC case it might happen that there are an unspecified
1792 * number of pending IRQ events unhandled. These cases are very rare,
1793 * so we 'resend' these IRQs via IPIs, to the same CPU. It's much
1794 * better to do it this way as thus we do not have to be aware of
1795 * 'pending' interrupts in the IRQ path, except at this point.
1796 */
1797/*
1798 * Edge triggered needs to resend any interrupt
1799 * that was delayed but this is now handled in the device
1800 * independent code.
1801 */
1802
1803/*
1804 * Starting up a edge-triggered IO-APIC interrupt is
1805 * nasty - we need to make sure that we get the edge.
1806 * If it is already asserted for some reason, we need
1807 * return 1 to indicate that is was pending.
1808 *
1809 * This is not complete - we should be able to fake
1810 * an edge even if it isn't on the 8259A...
1811 */
1812static unsigned int startup_edge_ioapic_irq(unsigned int irq)
1813{
1814 int was_pending = 0;
1815 unsigned long flags;
1816
1817 spin_lock_irqsave(&ioapic_lock, flags);
1818 if (irq < 16) {
1819 disable_8259A_irq(irq);
1820 if (i8259A_irq_pending(irq))
1821 was_pending = 1;
1822 }
1823 __unmask_IO_APIC_irq(irq);
1824 spin_unlock_irqrestore(&ioapic_lock, flags);
1825
1826 return was_pending;
1827}
1828
1829/*
1830 * Once we have recorded IRQ_PENDING already, we can mask the
1831 * interrupt for real. This prevents IRQ storms from unhandled
1832 * devices.
1833 */
1834static void ack_edge_ioapic_irq(unsigned int irq)
1835{
1836 move_irq(irq);
1837 if ((irq_desc[irq].status & (IRQ_PENDING | IRQ_DISABLED))
1838 == (IRQ_PENDING | IRQ_DISABLED))
1839 mask_IO_APIC_irq(irq);
1840 ack_APIC_irq();
1841}
1842
1843/*
1844 * Level triggered interrupts can just be masked,
1845 * and shutting down and starting up the interrupt
1846 * is the same as enabling and disabling them -- except
1847 * with a startup need to return a "was pending" value.
1848 *
1849 * Level triggered interrupts are special because we
1850 * do not touch any IO-APIC register while handling
1851 * them. We ack the APIC in the end-IRQ handler, not
1852 * in the start-IRQ-handler. Protection against reentrance
1853 * from the same interrupt is still provided, both by the
1854 * generic IRQ layer and by the fact that an unacked local
1855 * APIC does not accept IRQs.
1856 */
1857static unsigned int startup_level_ioapic_irq (unsigned int irq)
1858{
1859 unmask_IO_APIC_irq(irq);
1860
1861 return 0; /* don't check for pending */
1862}
1863
1864static void end_level_ioapic_irq (unsigned int irq)
1865{
1866 unsigned long v;
1867 int i;
1868
1869 move_irq(irq);
1870/*
1871 * It appears there is an erratum which affects at least version 0x11
1872 * of I/O APIC (that's the 82093AA and cores integrated into various
1873 * chipsets). Under certain conditions a level-triggered interrupt is
1874 * erroneously delivered as edge-triggered one but the respective IRR
1875 * bit gets set nevertheless. As a result the I/O unit expects an EOI
1876 * message but it will never arrive and further interrupts are blocked
1877 * from the source. The exact reason is so far unknown, but the
1878 * phenomenon was observed when two consecutive interrupt requests
1879 * from a given source get delivered to the same CPU and the source is
1880 * temporarily disabled in between.
1881 *
1882 * A workaround is to simulate an EOI message manually. We achieve it
1883 * by setting the trigger mode to edge and then to level when the edge
1884 * trigger mode gets detected in the TMR of a local APIC for a
1885 * level-triggered interrupt. We mask the source for the time of the
1886 * operation to prevent an edge-triggered interrupt escaping meanwhile.
1887 * The idea is from Manfred Spraul. --macro
1888 */
1889 i = IO_APIC_VECTOR(irq);
1890
1891 v = apic_read(APIC_TMR + ((i & ~0x1f) >> 1));
1892
1893 ack_APIC_irq();
1894
1895 if (!(v & (1 << (i & 0x1f)))) {
1896 atomic_inc(&irq_mis_count);
1897 spin_lock(&ioapic_lock);
1898 __mask_and_edge_IO_APIC_irq(irq);
1899 __unmask_and_level_IO_APIC_irq(irq);
1900 spin_unlock(&ioapic_lock);
1901 }
1902}
1903
1904#ifdef CONFIG_PCI_MSI
1905static unsigned int startup_edge_ioapic_vector(unsigned int vector)
1906{
1907 int irq = vector_to_irq(vector);
1908
1909 return startup_edge_ioapic_irq(irq);
1910}
1911
1912static void ack_edge_ioapic_vector(unsigned int vector)
1913{
1914 int irq = vector_to_irq(vector);
1915
1916 ack_edge_ioapic_irq(irq);
1917}
1918
1919static unsigned int startup_level_ioapic_vector (unsigned int vector)
1920{
1921 int irq = vector_to_irq(vector);
1922
1923 return startup_level_ioapic_irq (irq);
1924}
1925
1926static void end_level_ioapic_vector (unsigned int vector)
1927{
1928 int irq = vector_to_irq(vector);
1929
1930 end_level_ioapic_irq(irq);
1931}
1932
1933static void mask_IO_APIC_vector (unsigned int vector)
1934{
1935 int irq = vector_to_irq(vector);
1936
1937 mask_IO_APIC_irq(irq);
1938}
1939
1940static void unmask_IO_APIC_vector (unsigned int vector)
1941{
1942 int irq = vector_to_irq(vector);
1943
1944 unmask_IO_APIC_irq(irq);
1945}
1946
1947static void set_ioapic_affinity_vector (unsigned int vector,
1948 cpumask_t cpu_mask)
1949{
1950 int irq = vector_to_irq(vector);
1951
1952 set_ioapic_affinity_irq(irq, cpu_mask);
1953}
1954#endif
1955
1956/*
1957 * Level and edge triggered IO-APIC interrupts need different handling,
1958 * so we use two separate IRQ descriptors. Edge triggered IRQs can be
1959 * handled with the level-triggered descriptor, but that one has slightly
1960 * more overhead. Level-triggered interrupts cannot be handled with the
1961 * edge-triggered handler, without risking IRQ storms and other ugly
1962 * races.
1963 */
1964static struct hw_interrupt_type ioapic_edge_type = {
1965 .typename = "IO-APIC-edge",
1966 .startup = startup_edge_ioapic,
1967 .shutdown = shutdown_edge_ioapic,
1968 .enable = enable_edge_ioapic,
1969 .disable = disable_edge_ioapic,
1970 .ack = ack_edge_ioapic,
1971 .end = end_edge_ioapic,
1972 .set_affinity = set_ioapic_affinity,
1973};
1974
1975static struct hw_interrupt_type ioapic_level_type = {
1976 .typename = "IO-APIC-level",
1977 .startup = startup_level_ioapic,
1978 .shutdown = shutdown_level_ioapic,
1979 .enable = enable_level_ioapic,
1980 .disable = disable_level_ioapic,
1981 .ack = mask_and_ack_level_ioapic,
1982 .end = end_level_ioapic,
1983 .set_affinity = set_ioapic_affinity,
1984};
1985
1986static inline void init_IO_APIC_traps(void)
1987{
1988 int irq;
1989
1990 /*
1991 * NOTE! The local APIC isn't very good at handling
1992 * multiple interrupts at the same interrupt level.
1993 * As the interrupt level is determined by taking the
1994 * vector number and shifting that right by 4, we
1995 * want to spread these out a bit so that they don't
1996 * all fall in the same interrupt level.
1997 *
1998 * Also, we've got to be careful not to trash gate
1999 * 0x80, because int 0x80 is hm, kind of importantish. ;)
2000 */
2001 for (irq = 0; irq < NR_IRQS ; irq++) {
2002 int tmp = irq;
2003 if (use_pci_vector()) {
2004 if (!platform_legacy_irq(tmp))
2005 if ((tmp = vector_to_irq(tmp)) == -1)
2006 continue;
2007 }
2008 if (IO_APIC_IRQ(tmp) && !IO_APIC_VECTOR(tmp)) {
2009 /*
2010 * Hmm.. We don't have an entry for this,
2011 * so default to an old-fashioned 8259
2012 * interrupt if we can..
2013 */
2014 if (irq < 16)
2015 make_8259A_irq(irq);
2016 else
2017 /* Strange. Oh, well.. */
2018 irq_desc[irq].handler = &no_irq_type;
2019 }
2020 }
2021}
2022
2023static void enable_lapic_irq (unsigned int irq)
2024{
2025 unsigned long v;
2026
2027 v = apic_read(APIC_LVT0);
2028 apic_write_around(APIC_LVT0, v & ~APIC_LVT_MASKED);
2029}
2030
2031static void disable_lapic_irq (unsigned int irq)
2032{
2033 unsigned long v;
2034
2035 v = apic_read(APIC_LVT0);
2036 apic_write_around(APIC_LVT0, v | APIC_LVT_MASKED);
2037}
2038
2039static void ack_lapic_irq (unsigned int irq)
2040{
2041 ack_APIC_irq();
2042}
2043
2044static void end_lapic_irq (unsigned int i) { /* nothing */ }
2045
2046static struct hw_interrupt_type lapic_irq_type = {
2047 .typename = "local-APIC-edge",
2048 .startup = NULL, /* startup_irq() not used for IRQ0 */
2049 .shutdown = NULL, /* shutdown_irq() not used for IRQ0 */
2050 .enable = enable_lapic_irq,
2051 .disable = disable_lapic_irq,
2052 .ack = ack_lapic_irq,
2053 .end = end_lapic_irq
2054};
2055
2056static void setup_nmi (void)
2057{
2058 /*
2059 * Dirty trick to enable the NMI watchdog ...
2060 * We put the 8259A master into AEOI mode and
2061 * unmask on all local APICs LVT0 as NMI.
2062 *
2063 * The idea to use the 8259A in AEOI mode ('8259A Virtual Wire')
2064 * is from Maciej W. Rozycki - so we do not have to EOI from
2065 * the NMI handler or the timer interrupt.
2066 */
2067 apic_printk(APIC_VERBOSE, KERN_INFO "activating NMI Watchdog ...");
2068
2069 on_each_cpu(enable_NMI_through_LVT0, NULL, 1, 1);
2070
2071 apic_printk(APIC_VERBOSE, " done.\n");
2072}
2073
2074/*
2075 * This looks a bit hackish but it's about the only one way of sending
2076 * a few INTA cycles to 8259As and any associated glue logic. ICR does
2077 * not support the ExtINT mode, unfortunately. We need to send these
2078 * cycles as some i82489DX-based boards have glue logic that keeps the
2079 * 8259A interrupt line asserted until INTA. --macro
2080 */
2081static inline void unlock_ExtINT_logic(void)
2082{
2083 int pin, i;
2084 struct IO_APIC_route_entry entry0, entry1;
2085 unsigned char save_control, save_freq_select;
2086 unsigned long flags;
2087
2088 pin = find_isa_irq_pin(8, mp_INT);
2089 if (pin == -1)
2090 return;
2091
2092 spin_lock_irqsave(&ioapic_lock, flags);
2093 *(((int *)&entry0) + 1) = io_apic_read(0, 0x11 + 2 * pin);
2094 *(((int *)&entry0) + 0) = io_apic_read(0, 0x10 + 2 * pin);
2095 spin_unlock_irqrestore(&ioapic_lock, flags);
2096 clear_IO_APIC_pin(0, pin);
2097
2098 memset(&entry1, 0, sizeof(entry1));
2099
2100 entry1.dest_mode = 0; /* physical delivery */
2101 entry1.mask = 0; /* unmask IRQ now */
2102 entry1.dest.physical.physical_dest = hard_smp_processor_id();
2103 entry1.delivery_mode = dest_ExtINT;
2104 entry1.polarity = entry0.polarity;
2105 entry1.trigger = 0;
2106 entry1.vector = 0;
2107
2108 spin_lock_irqsave(&ioapic_lock, flags);
2109 io_apic_write(0, 0x11 + 2 * pin, *(((int *)&entry1) + 1));
2110 io_apic_write(0, 0x10 + 2 * pin, *(((int *)&entry1) + 0));
2111 spin_unlock_irqrestore(&ioapic_lock, flags);
2112
2113 save_control = CMOS_READ(RTC_CONTROL);
2114 save_freq_select = CMOS_READ(RTC_FREQ_SELECT);
2115 CMOS_WRITE((save_freq_select & ~RTC_RATE_SELECT) | 0x6,
2116 RTC_FREQ_SELECT);
2117 CMOS_WRITE(save_control | RTC_PIE, RTC_CONTROL);
2118
2119 i = 100;
2120 while (i-- > 0) {
2121 mdelay(10);
2122 if ((CMOS_READ(RTC_INTR_FLAGS) & RTC_PF) == RTC_PF)
2123 i -= 10;
2124 }
2125
2126 CMOS_WRITE(save_control, RTC_CONTROL);
2127 CMOS_WRITE(save_freq_select, RTC_FREQ_SELECT);
2128 clear_IO_APIC_pin(0, pin);
2129
2130 spin_lock_irqsave(&ioapic_lock, flags);
2131 io_apic_write(0, 0x11 + 2 * pin, *(((int *)&entry0) + 1));
2132 io_apic_write(0, 0x10 + 2 * pin, *(((int *)&entry0) + 0));
2133 spin_unlock_irqrestore(&ioapic_lock, flags);
2134}
2135
2136/*
2137 * This code may look a bit paranoid, but it's supposed to cooperate with
2138 * a wide range of boards and BIOS bugs. Fortunately only the timer IRQ
2139 * is so screwy. Thanks to Brian Perkins for testing/hacking this beast
2140 * fanatically on his truly buggy board.
2141 */
2142static inline void check_timer(void)
2143{
2144 int pin1, pin2;
2145 int vector;
2146
2147 /*
2148 * get/set the timer IRQ vector:
2149 */
2150 disable_8259A_irq(0);
2151 vector = assign_irq_vector(0);
2152 set_intr_gate(vector, interrupt[0]);
2153
2154 /*
2155 * Subtle, code in do_timer_interrupt() expects an AEOI
2156 * mode for the 8259A whenever interrupts are routed
2157 * through I/O APICs. Also IRQ0 has to be enabled in
2158 * the 8259A which implies the virtual wire has to be
2159 * disabled in the local APIC.
2160 */
2161 apic_write_around(APIC_LVT0, APIC_LVT_MASKED | APIC_DM_EXTINT);
2162 init_8259A(1);
2163 timer_ack = 1;
2164 enable_8259A_irq(0);
2165
2166 pin1 = find_isa_irq_pin(0, mp_INT);
2167 pin2 = find_isa_irq_pin(0, mp_ExtINT);
2168
2169 printk(KERN_INFO "..TIMER: vector=0x%02X pin1=%d pin2=%d\n", vector, pin1, pin2);
2170
2171 if (pin1 != -1) {
2172 /*
2173 * Ok, does IRQ0 through the IOAPIC work?
2174 */
2175 unmask_IO_APIC_irq(0);
2176 if (timer_irq_works()) {
2177 if (nmi_watchdog == NMI_IO_APIC) {
2178 disable_8259A_irq(0);
2179 setup_nmi();
2180 enable_8259A_irq(0);
1da177e4
LT		 2181 }
2182 return;
2183 }
2184 clear_IO_APIC_pin(0, pin1);
2185 printk(KERN_ERR "..MP-BIOS bug: 8254 timer not connected to IO-APIC\n");
2186 }
2187
2188 printk(KERN_INFO "...trying to set up timer (IRQ0) through the 8259A ... ");
2189 if (pin2 != -1) {
2190 printk("\n..... (found pin %d) ...", pin2);
2191 /*
2192 * legacy devices should be connected to IO APIC #0
2193 */
2194 setup_ExtINT_IRQ0_pin(pin2, vector);
2195 if (timer_irq_works()) {
2196 printk("works.\n");
2197 if (pin1 != -1)
2198 replace_pin_at_irq(0, 0, pin1, 0, pin2);
2199 else
2200 add_pin_to_irq(0, 0, pin2);
2201 if (nmi_watchdog == NMI_IO_APIC) {
2202 setup_nmi();
1da177e4
LT		 2203 }
2204 return;
2205 }
2206 /*
2207 * Cleanup, just in case ...
2208 */
2209 clear_IO_APIC_pin(0, pin2);
2210 }
2211 printk(" failed.\n");
2212
2213 if (nmi_watchdog == NMI_IO_APIC) {
2214 printk(KERN_WARNING "timer doesn't work through the IO-APIC - disabling NMI Watchdog!\n");
2215 nmi_watchdog = 0;
2216 }
2217
2218 printk(KERN_INFO "...trying to set up timer as Virtual Wire IRQ...");
2219
2220 disable_8259A_irq(0);
2221 irq_desc[0].handler = &lapic_irq_type;
2222 apic_write_around(APIC_LVT0, APIC_DM_FIXED | vector); /* Fixed mode */
2223 enable_8259A_irq(0);
2224
2225 if (timer_irq_works()) {
2226 printk(" works.\n");
2227 return;
2228 }
2229 apic_write_around(APIC_LVT0, APIC_LVT_MASKED | APIC_DM_FIXED | vector);
2230 printk(" failed.\n");
2231
2232 printk(KERN_INFO "...trying to set up timer as ExtINT IRQ...");
2233
2234 timer_ack = 0;
2235 init_8259A(0);
2236 make_8259A_irq(0);
2237 apic_write_around(APIC_LVT0, APIC_DM_EXTINT);
2238
2239 unlock_ExtINT_logic();
2240
2241 if (timer_irq_works()) {
2242 printk(" works.\n");
2243 return;
2244 }
2245 printk(" failed :(.\n");
2246 panic("IO-APIC + timer doesn't work! Boot with apic=debug and send a "
2247 "report. Then try booting with the 'noapic' option");
2248}
2249
2250/*
2251 *
2252 * IRQ's that are handled by the PIC in the MPS IOAPIC case.
2253 * - IRQ2 is the cascade IRQ, and cannot be a io-apic IRQ.
2254 * Linux doesn't really care, as it's not actually used
2255 * for any interrupt handling anyway.
2256 */
2257#define PIC_IRQS (1 << PIC_CASCADE_IR)
2258
2259void __init setup_IO_APIC(void)
2260{
2261 enable_IO_APIC();
2262
2263 if (acpi_ioapic)
2264 io_apic_irqs = ~0; /* all IRQs go through IOAPIC */
2265 else
2266 io_apic_irqs = ~PIC_IRQS;
2267
2268 printk("ENABLING IO-APIC IRQs\n");
2269
2270 /*
2271 * Set up IO-APIC IRQ routing.
2272 */
2273 if (!acpi_ioapic)
2274 setup_ioapic_ids_from_mpc();
2275 sync_Arb_IDs();
2276 setup_IO_APIC_irqs();
2277 init_IO_APIC_traps();
2278 check_timer();
2279 if (!acpi_ioapic)
2280 print_IO_APIC();
2281}
2282
2283/*
2284 * Called after all the initialization is done. If we didnt find any
2285 * APIC bugs then we can allow the modify fast path
2286 */
2287
2288static int __init io_apic_bug_finalize(void)
2289{
2290 if(sis_apic_bug == -1)
2291 sis_apic_bug = 0;
2292 return 0;
2293}
2294
2295late_initcall(io_apic_bug_finalize);
2296
2297struct sysfs_ioapic_data {
2298 struct sys_device dev;
2299 struct IO_APIC_route_entry entry[0];
2300};
2301static struct sysfs_ioapic_data * mp_ioapic_data[MAX_IO_APICS];
2302
438510f6 2303static int ioapic_suspend(struct sys_device *dev, pm_message_t state)
1da177e4
LT		 2304{
2305 struct IO_APIC_route_entry *entry;
2306 struct sysfs_ioapic_data *data;
2307 unsigned long flags;
2308 int i;
2309
2310 data = container_of(dev, struct sysfs_ioapic_data, dev);
2311 entry = data->entry;
2312 spin_lock_irqsave(&ioapic_lock, flags);
2313 for (i = 0; i < nr_ioapic_registers[dev->id]; i ++, entry ++ ) {
2314 *(((int *)entry) + 1) = io_apic_read(dev->id, 0x11 + 2 * i);
2315 *(((int *)entry) + 0) = io_apic_read(dev->id, 0x10 + 2 * i);
2316 }
2317 spin_unlock_irqrestore(&ioapic_lock, flags);
2318
2319 return 0;
2320}
2321
2322static int ioapic_resume(struct sys_device *dev)
2323{
2324 struct IO_APIC_route_entry *entry;
2325 struct sysfs_ioapic_data *data;
2326 unsigned long flags;
2327 union IO_APIC_reg_00 reg_00;
2328 int i;
2329
2330 data = container_of(dev, struct sysfs_ioapic_data, dev);
2331 entry = data->entry;
2332
2333 spin_lock_irqsave(&ioapic_lock, flags);
2334 reg_00.raw = io_apic_read(dev->id, 0);
2335 if (reg_00.bits.ID != mp_ioapics[dev->id].mpc_apicid) {
2336 reg_00.bits.ID = mp_ioapics[dev->id].mpc_apicid;
2337 io_apic_write(dev->id, 0, reg_00.raw);
2338 }
2339 for (i = 0; i < nr_ioapic_registers[dev->id]; i ++, entry ++ ) {
2340 io_apic_write(dev->id, 0x11+2*i, *(((int *)entry)+1));
2341 io_apic_write(dev->id, 0x10+2*i, *(((int *)entry)+0));
2342 }
2343 spin_unlock_irqrestore(&ioapic_lock, flags);
2344
2345 return 0;
2346}
2347
2348static struct sysdev_class ioapic_sysdev_class = {
2349 set_kset_name("ioapic"),
2350 .suspend = ioapic_suspend,
2351 .resume = ioapic_resume,
2352};
2353
2354static int __init ioapic_init_sysfs(void)
2355{
2356 struct sys_device * dev;
2357 int i, size, error = 0;
2358
2359 error = sysdev_class_register(&ioapic_sysdev_class);
2360 if (error)
2361 return error;
2362
2363 for (i = 0; i < nr_ioapics; i++ ) {
2364 size = sizeof(struct sys_device) + nr_ioapic_registers[i]
2365 * sizeof(struct IO_APIC_route_entry);
2366 mp_ioapic_data[i] = kmalloc(size, GFP_KERNEL);
2367 if (!mp_ioapic_data[i]) {
2368 printk(KERN_ERR "Can't suspend/resume IOAPIC %d\n", i);
2369 continue;
2370 }
2371 memset(mp_ioapic_data[i], 0, size);
2372 dev = &mp_ioapic_data[i]->dev;
2373 dev->id = i;
2374 dev->cls = &ioapic_sysdev_class;
2375 error = sysdev_register(dev);
2376 if (error) {
2377 kfree(mp_ioapic_data[i]);
2378 mp_ioapic_data[i] = NULL;
2379 printk(KERN_ERR "Can't suspend/resume IOAPIC %d\n", i);
2380 continue;
2381 }
2382 }
2383
2384 return 0;
2385}
2386
2387device_initcall(ioapic_init_sysfs);
2388
2389/* --------------------------------------------------------------------------
2390 ACPI-based IOAPIC Configuration
2391 -------------------------------------------------------------------------- */
2392
2393#ifdef CONFIG_ACPI_BOOT
2394
2395int __init io_apic_get_unique_id (int ioapic, int apic_id)
2396{
2397 union IO_APIC_reg_00 reg_00;
2398 static physid_mask_t apic_id_map = PHYSID_MASK_NONE;
2399 physid_mask_t tmp;
2400 unsigned long flags;
2401 int i = 0;
2402
2403 /*
2404 * The P4 platform supports up to 256 APIC IDs on two separate APIC
2405 * buses (one for LAPICs, one for IOAPICs), where predecessors only
2406 * supports up to 16 on one shared APIC bus.
2407 *
2408 * TBD: Expand LAPIC/IOAPIC support on P4-class systems to take full
2409 * advantage of new APIC bus architecture.
2410 */
2411
2412 if (physids_empty(apic_id_map))
2413 apic_id_map = ioapic_phys_id_map(phys_cpu_present_map);
2414
2415 spin_lock_irqsave(&ioapic_lock, flags);
2416 reg_00.raw = io_apic_read(ioapic, 0);
2417 spin_unlock_irqrestore(&ioapic_lock, flags);
2418
2419 if (apic_id >= get_physical_broadcast()) {
2420 printk(KERN_WARNING "IOAPIC[%d]: Invalid apic_id %d, trying "
2421 "%d\n", ioapic, apic_id, reg_00.bits.ID);
2422 apic_id = reg_00.bits.ID;
2423 }
2424
2425 /*
2426 * Every APIC in a system must have a unique ID or we get lots of nice
2427 * 'stuck on smp_invalidate_needed IPI wait' messages.
2428 */
2429 if (check_apicid_used(apic_id_map, apic_id)) {
2430
2431 for (i = 0; i < get_physical_broadcast(); i++) {
2432 if (!check_apicid_used(apic_id_map, i))
2433 break;
2434 }
2435
2436 if (i == get_physical_broadcast())
2437 panic("Max apic_id exceeded!\n");
2438
2439 printk(KERN_WARNING "IOAPIC[%d]: apic_id %d already used, "
2440 "trying %d\n", ioapic, apic_id, i);
2441
2442 apic_id = i;
2443 }
2444
2445 tmp = apicid_to_cpu_present(apic_id);
2446 physids_or(apic_id_map, apic_id_map, tmp);
2447
2448 if (reg_00.bits.ID != apic_id) {
2449 reg_00.bits.ID = apic_id;
2450
2451 spin_lock_irqsave(&ioapic_lock, flags);
2452 io_apic_write(ioapic, 0, reg_00.raw);
2453 reg_00.raw = io_apic_read(ioapic, 0);
2454 spin_unlock_irqrestore(&ioapic_lock, flags);
2455
2456 /* Sanity check */
2457 if (reg_00.bits.ID != apic_id)
2458 panic("IOAPIC[%d]: Unable change apic_id!\n", ioapic);
2459 }
2460
2461 apic_printk(APIC_VERBOSE, KERN_INFO
2462 "IOAPIC[%d]: Assigned apic_id %d\n", ioapic, apic_id);
2463
2464 return apic_id;
2465}
2466
2467
2468int __init io_apic_get_version (int ioapic)
2469{
2470 union IO_APIC_reg_01 reg_01;
2471 unsigned long flags;
2472
2473 spin_lock_irqsave(&ioapic_lock, flags);
2474 reg_01.raw = io_apic_read(ioapic, 1);
2475 spin_unlock_irqrestore(&ioapic_lock, flags);
2476
2477 return reg_01.bits.version;
2478}
2479
2480
2481int __init io_apic_get_redir_entries (int ioapic)
2482{
2483 union IO_APIC_reg_01 reg_01;
2484 unsigned long flags;
2485
2486 spin_lock_irqsave(&ioapic_lock, flags);
2487 reg_01.raw = io_apic_read(ioapic, 1);
2488 spin_unlock_irqrestore(&ioapic_lock, flags);
2489
2490 return reg_01.bits.entries;
2491}
2492
2493
2494int io_apic_set_pci_routing (int ioapic, int pin, int irq, int edge_level, int active_high_low)
2495{
2496 struct IO_APIC_route_entry entry;
2497 unsigned long flags;
2498
2499 if (!IO_APIC_IRQ(irq)) {
2500 printk(KERN_ERR "IOAPIC[%d]: Invalid reference to IRQ 0\n",
2501 ioapic);
2502 return -EINVAL;
2503 }
2504
2505 /*
2506 * Generate a PCI IRQ routing entry and program the IOAPIC accordingly.
2507 * Note that we mask (disable) IRQs now -- these get enabled when the
2508 * corresponding device driver registers for this IRQ.
2509 */
2510
2511 memset(&entry,0,sizeof(entry));
2512
2513 entry.delivery_mode = INT_DELIVERY_MODE;
2514 entry.dest_mode = INT_DEST_MODE;
2515 entry.dest.logical.logical_dest = cpu_mask_to_apicid(TARGET_CPUS);
2516 entry.trigger = edge_level;
2517 entry.polarity = active_high_low;
2518 entry.mask = 1;
2519
2520 /*
2521 * IRQs < 16 are already in the irq_2_pin[] map
2522 */
2523 if (irq >= 16)
2524 add_pin_to_irq(irq, ioapic, pin);
2525
2526 entry.vector = assign_irq_vector(irq);
2527
2528 apic_printk(APIC_DEBUG, KERN_DEBUG "IOAPIC[%d]: Set PCI routing entry "
2529 "(%d-%d -> 0x%x -> IRQ %d Mode:%i Active:%i)\n", ioapic,
2530 mp_ioapics[ioapic].mpc_apicid, pin, entry.vector, irq,
2531 edge_level, active_high_low);
2532
2533 ioapic_register_intr(irq, entry.vector, edge_level);
2534
2535 if (!ioapic && (irq < 16))
2536 disable_8259A_irq(irq);
2537
2538 spin_lock_irqsave(&ioapic_lock, flags);
2539 io_apic_write(ioapic, 0x11+2*pin, *(((int *)&entry)+1));
2540 io_apic_write(ioapic, 0x10+2*pin, *(((int *)&entry)+0));
2541 spin_unlock_irqrestore(&ioapic_lock, flags);
2542
2543 return 0;
2544}
2545
2546#endif /*CONFIG_ACPI_BOOT*/
Linux kernel - Deliverables
This page took 0.254989 seconds and 5 git commands to generate.