aboutsummaryrefslogtreecommitdiffstats
path: root/arch/i386/kernel/io_apic.c
diff options
context:
space:
mode:
Diffstat (limited to 'arch/i386/kernel/io_apic.c')
-rw-r--r--arch/i386/kernel/io_apic.c2545
1 files changed, 2545 insertions, 0 deletions
diff --git a/arch/i386/kernel/io_apic.c b/arch/i386/kernel/io_apic.c
new file mode 100644
index 000000000000..9c1350e811d0
--- /dev/null
+++ b/arch/i386/kernel/io_apic.c
@@ -0,0 +1,2545 @@
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>
34
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}
815
816/*
817 * This function currently is only a helper for the i386 smp boot process where
818 * we need to reprogram the ioredtbls to cater for the cpus which have come online
819 * so mask in all cases should simply be TARGET_CPUS
820 */
821void __init setup_ioapic_dest(void)
822{
823 int pin, ioapic, irq, irq_entry;
824
825 if (skip_ioapic_setup == 1)
826 return;
827
828 for (ioapic = 0; ioapic < nr_ioapics; ioapic++) {
829 for (pin = 0; pin < nr_ioapic_registers[ioapic]; pin++) {
830 irq_entry = find_irq_entry(ioapic, pin, mp_INT);
831 if (irq_entry == -1)
832 continue;
833 irq = pin_2_irq(irq_entry, ioapic, pin);
834 set_ioapic_affinity_irq(irq, TARGET_CPUS);
835 }
836
837 }
838}
839
840/*
841 * EISA Edge/Level control register, ELCR
842 */
843static int EISA_ELCR(unsigned int irq)
844{
845 if (irq < 16) {
846 unsigned int port = 0x4d0 + (irq >> 3);
847 return (inb(port) >> (irq & 7)) & 1;
848 }
849 apic_printk(APIC_VERBOSE, KERN_INFO
850 "Broken MPtable reports ISA irq %d\n", irq);
851 return 0;
852}
853
854/* EISA interrupts are always polarity zero and can be edge or level
855 * trigger depending on the ELCR value. If an interrupt is listed as
856 * EISA conforming in the MP table, that means its trigger type must
857 * be read in from the ELCR */
858
859#define default_EISA_trigger(idx) (EISA_ELCR(mp_irqs[idx].mpc_srcbusirq))
860#define default_EISA_polarity(idx) (0)
861
862/* ISA interrupts are always polarity zero edge triggered,
863 * when listed as conforming in the MP table. */
864
865#define default_ISA_trigger(idx) (0)
866#define default_ISA_polarity(idx) (0)
867
868/* PCI interrupts are always polarity one level triggered,
869 * when listed as conforming in the MP table. */
870
871#define default_PCI_trigger(idx) (1)
872#define default_PCI_polarity(idx) (1)
873
874/* MCA interrupts are always polarity zero level triggered,
875 * when listed as conforming in the MP table. */
876
877#define default_MCA_trigger(idx) (1)
878#define default_MCA_polarity(idx) (0)
879
880/* NEC98 interrupts are always polarity zero edge triggered,
881 * when listed as conforming in the MP table. */
882
883#define default_NEC98_trigger(idx) (0)
884#define default_NEC98_polarity(idx) (0)
885
886static int __init MPBIOS_polarity(int idx)
887{
888 int bus = mp_irqs[idx].mpc_srcbus;
889 int polarity;
890
891 /*
892 * Determine IRQ line polarity (high active or low active):
893 */
894 switch (mp_irqs[idx].mpc_irqflag & 3)
895 {
896 case 0: /* conforms, ie. bus-type dependent polarity */
897 {
898 switch (mp_bus_id_to_type[bus])
899 {
900 case MP_BUS_ISA: /* ISA pin */
901 {
902 polarity = default_ISA_polarity(idx);
903 break;
904 }
905 case MP_BUS_EISA: /* EISA pin */
906 {
907 polarity = default_EISA_polarity(idx);
908 break;
909 }
910 case MP_BUS_PCI: /* PCI pin */
911 {
912 polarity = default_PCI_polarity(idx);
913 break;
914 }
915 case MP_BUS_MCA: /* MCA pin */
916 {
917 polarity = default_MCA_polarity(idx);
918 break;
919 }
920 case MP_BUS_NEC98: /* NEC 98 pin */
921 {
922 polarity = default_NEC98_polarity(idx);
923 break;
924 }
925 default:
926 {
927 printk(KERN_WARNING "broken BIOS!!\n");
928 polarity = 1;
929 break;
930 }
931 }
932 break;
933 }
934 case 1: /* high active */
935 {
936 polarity = 0;
937 break;
938 }
939 case 2: /* reserved */
940 {
941 printk(KERN_WARNING "broken BIOS!!\n");
942 polarity = 1;
943 break;
944 }
945 case 3: /* low active */
946 {
947 polarity = 1;
948 break;
949 }
950 default: /* invalid */
951 {
952 printk(KERN_WARNING "broken BIOS!!\n");
953 polarity = 1;
954 break;
955 }
956 }
957 return polarity;
958}
959
960static int MPBIOS_trigger(int idx)
961{
962 int bus = mp_irqs[idx].mpc_srcbus;
963 int trigger;
964
965 /*
966 * Determine IRQ trigger mode (edge or level sensitive):
967 */
968 switch ((mp_irqs[idx].mpc_irqflag>>2) & 3)
969 {
970 case 0: /* conforms, ie. bus-type dependent */
971 {
972 switch (mp_bus_id_to_type[bus])
973 {
974 case MP_BUS_ISA: /* ISA pin */
975 {
976 trigger = default_ISA_trigger(idx);
977 break;
978 }
979 case MP_BUS_EISA: /* EISA pin */
980 {
981 trigger = default_EISA_trigger(idx);
982 break;
983 }
984 case MP_BUS_PCI: /* PCI pin */
985 {
986 trigger = default_PCI_trigger(idx);
987 break;
988 }
989 case MP_BUS_MCA: /* MCA pin */
990 {
991 trigger = default_MCA_trigger(idx);
992 break;
993 }
994 case MP_BUS_NEC98: /* NEC 98 pin */
995 {
996 trigger = default_NEC98_trigger(idx);
997 break;
998 }
999 default:
1000 {
1001 printk(KERN_WARNING "broken BIOS!!\n");
1002 trigger = 1;
1003 break;
1004 }
1005 }
1006 break;
1007 }
1008 case 1: /* edge */
1009 {
1010 trigger = 0;
1011 break;
1012 }
1013 case 2: /* reserved */
1014 {
1015 printk(KERN_WARNING "broken BIOS!!\n");
1016 trigger = 1;
1017 break;
1018 }
1019 case 3: /* level */
1020 {
1021 trigger = 1;
1022 break;
1023 }
1024 default: /* invalid */
1025 {
1026 printk(KERN_WARNING "broken BIOS!!\n");
1027 trigger = 0;
1028 break;
1029 }
1030 }
1031 return trigger;
1032}
1033
1034static inline int irq_polarity(int idx)
1035{
1036 return MPBIOS_polarity(idx);
1037}
1038
1039static inline int irq_trigger(int idx)
1040{
1041 return MPBIOS_trigger(idx);
1042}
1043
1044static int pin_2_irq(int idx, int apic, int pin)
1045{
1046 int irq, i;
1047 int bus = mp_irqs[idx].mpc_srcbus;
1048
1049 /*
1050 * Debugging check, we are in big trouble if this message pops up!
1051 */
1052 if (mp_irqs[idx].mpc_dstirq != pin)
1053 printk(KERN_ERR "broken BIOS or MPTABLE parser, ayiee!!\n");
1054
1055 switch (mp_bus_id_to_type[bus])
1056 {
1057 case MP_BUS_ISA: /* ISA pin */
1058 case MP_BUS_EISA:
1059 case MP_BUS_MCA:
1060 case MP_BUS_NEC98:
1061 {
1062 irq = mp_irqs[idx].mpc_srcbusirq;
1063 break;
1064 }
1065 case MP_BUS_PCI: /* PCI pin */
1066 {
1067 /*
1068 * PCI IRQs are mapped in order
1069 */
1070 i = irq = 0;
1071 while (i < apic)
1072 irq += nr_ioapic_registers[i++];
1073 irq += pin;
1074
1075 /*
1076 * For MPS mode, so far only needed by ES7000 platform
1077 */
1078 if (ioapic_renumber_irq)
1079 irq = ioapic_renumber_irq(apic, irq);
1080
1081 break;
1082 }
1083 default:
1084 {
1085 printk(KERN_ERR "unknown bus type %d.\n",bus);
1086 irq = 0;
1087 break;
1088 }
1089 }
1090
1091 /*
1092 * PCI IRQ command line redirection. Yes, limits are hardcoded.
1093 */
1094 if ((pin >= 16) && (pin <= 23)) {
1095 if (pirq_entries[pin-16] != -1) {
1096 if (!pirq_entries[pin-16]) {
1097 apic_printk(APIC_VERBOSE, KERN_DEBUG
1098 "disabling PIRQ%d\n", pin-16);
1099 } else {
1100 irq = pirq_entries[pin-16];
1101 apic_printk(APIC_VERBOSE, KERN_DEBUG
1102 "using PIRQ%d -> IRQ %d\n",
1103 pin-16, irq);
1104 }
1105 }
1106 }
1107 return irq;
1108}
1109
1110static inline int IO_APIC_irq_trigger(int irq)
1111{
1112 int apic, idx, pin;
1113
1114 for (apic = 0; apic < nr_ioapics; apic++) {
1115 for (pin = 0; pin < nr_ioapic_registers[apic]; pin++) {
1116 idx = find_irq_entry(apic,pin,mp_INT);
1117 if ((idx != -1) && (irq == pin_2_irq(idx,apic,pin)))
1118 return irq_trigger(idx);
1119 }
1120 }
1121 /*
1122 * nonexistent IRQs are edge default
1123 */
1124 return 0;
1125}
1126
1127/* irq_vectors is indexed by the sum of all RTEs in all I/O APICs. */
1128u8 irq_vector[NR_IRQ_VECTORS] = { FIRST_DEVICE_VECTOR , 0 };
1129
1130int assign_irq_vector(int irq)
1131{
1132 static int current_vector = FIRST_DEVICE_VECTOR, offset = 0;
1133
1134 BUG_ON(irq >= NR_IRQ_VECTORS);
1135 if (irq != AUTO_ASSIGN && IO_APIC_VECTOR(irq) > 0)
1136 return IO_APIC_VECTOR(irq);
1137next:
1138 current_vector += 8;
1139 if (current_vector == SYSCALL_VECTOR)
1140 goto next;
1141
1142 if (current_vector >= FIRST_SYSTEM_VECTOR) {
1143 offset++;
1144 if (!(offset%8))
1145 return -ENOSPC;
1146 current_vector = FIRST_DEVICE_VECTOR + offset;
1147 }
1148
1149 vector_irq[current_vector] = irq;
1150 if (irq != AUTO_ASSIGN)
1151 IO_APIC_VECTOR(irq) = current_vector;
1152
1153 return current_vector;
1154}
1155
1156static struct hw_interrupt_type ioapic_level_type;
1157static struct hw_interrupt_type ioapic_edge_type;
1158
1159#define IOAPIC_AUTO -1
1160#define IOAPIC_EDGE 0
1161#define IOAPIC_LEVEL 1
1162
1163static inline void ioapic_register_intr(int irq, int vector, unsigned long trigger)
1164{
1165 if (use_pci_vector() && !platform_legacy_irq(irq)) {
1166 if ((trigger == IOAPIC_AUTO && IO_APIC_irq_trigger(irq)) ||
1167 trigger == IOAPIC_LEVEL)
1168 irq_desc[vector].handler = &ioapic_level_type;
1169 else
1170 irq_desc[vector].handler = &ioapic_edge_type;
1171 set_intr_gate(vector, interrupt[vector]);
1172 } else {
1173 if ((trigger == IOAPIC_AUTO && IO_APIC_irq_trigger(irq)) ||
1174 trigger == IOAPIC_LEVEL)
1175 irq_desc[irq].handler = &ioapic_level_type;
1176 else
1177 irq_desc[irq].handler = &ioapic_edge_type;
1178 set_intr_gate(vector, interrupt[irq]);
1179 }
1180}
1181
1182static void __init setup_IO_APIC_irqs(void)
1183{
1184 struct IO_APIC_route_entry entry;
1185 int apic, pin, idx, irq, first_notcon = 1, vector;
1186 unsigned long flags;
1187
1188 apic_printk(APIC_VERBOSE, KERN_DEBUG "init IO_APIC IRQs\n");
1189
1190 for (apic = 0; apic < nr_ioapics; apic++) {
1191 for (pin = 0; pin < nr_ioapic_registers[apic]; pin++) {
1192
1193 /*
1194 * add it to the IO-APIC irq-routing table:
1195 */
1196 memset(&entry,0,sizeof(entry));
1197
1198 entry.delivery_mode = INT_DELIVERY_MODE;
1199 entry.dest_mode = INT_DEST_MODE;
1200 entry.mask = 0; /* enable IRQ */
1201 entry.dest.logical.logical_dest =
1202 cpu_mask_to_apicid(TARGET_CPUS);
1203
1204 idx = find_irq_entry(apic,pin,mp_INT);
1205 if (idx == -1) {
1206 if (first_notcon) {
1207 apic_printk(APIC_VERBOSE, KERN_DEBUG
1208 " IO-APIC (apicid-pin) %d-%d",
1209 mp_ioapics[apic].mpc_apicid,
1210 pin);
1211 first_notcon = 0;
1212 } else
1213 apic_printk(APIC_VERBOSE, ", %d-%d",
1214 mp_ioapics[apic].mpc_apicid, pin);
1215 continue;
1216 }
1217
1218 entry.trigger = irq_trigger(idx);
1219 entry.polarity = irq_polarity(idx);
1220
1221 if (irq_trigger(idx)) {
1222 entry.trigger = 1;
1223 entry.mask = 1;
1224 }
1225
1226 irq = pin_2_irq(idx, apic, pin);
1227 /*
1228 * skip adding the timer int on secondary nodes, which causes
1229 * a small but painful rift in the time-space continuum
1230 */
1231 if (multi_timer_check(apic, irq))
1232 continue;
1233 else
1234 add_pin_to_irq(irq, apic, pin);
1235
1236 if (!apic && !IO_APIC_IRQ(irq))
1237 continue;
1238
1239 if (IO_APIC_IRQ(irq)) {
1240 vector = assign_irq_vector(irq);
1241 entry.vector = vector;
1242 ioapic_register_intr(irq, vector, IOAPIC_AUTO);
1243
1244 if (!apic && (irq < 16))
1245 disable_8259A_irq(irq);
1246 }
1247 spin_lock_irqsave(&ioapic_lock, flags);
1248 io_apic_write(apic, 0x11+2*pin, *(((int *)&entry)+1));
1249 io_apic_write(apic, 0x10+2*pin, *(((int *)&entry)+0));
1250 spin_unlock_irqrestore(&ioapic_lock, flags);
1251 }
1252 }
1253
1254 if (!first_notcon)
1255 apic_printk(APIC_VERBOSE, " not connected.\n");
1256}
1257
1258/*
1259 * Set up the 8259A-master output pin:
1260 */
1261static void __init setup_ExtINT_IRQ0_pin(unsigned int pin, int vector)
1262{
1263 struct IO_APIC_route_entry entry;
1264 unsigned long flags;
1265
1266 memset(&entry,0,sizeof(entry));
1267
1268 disable_8259A_irq(0);
1269
1270 /* mask LVT0 */
1271 apic_write_around(APIC_LVT0, APIC_LVT_MASKED | APIC_DM_EXTINT);
1272
1273 /*
1274 * We use logical delivery to get the timer IRQ
1275 * to the first CPU.
1276 */
1277 entry.dest_mode = INT_DEST_MODE;
1278 entry.mask = 0; /* unmask IRQ now */
1279 entry.dest.logical.logical_dest = cpu_mask_to_apicid(TARGET_CPUS);
1280 entry.delivery_mode = INT_DELIVERY_MODE;
1281 entry.polarity = 0;
1282 entry.trigger = 0;
1283 entry.vector = vector;
1284
1285 /*
1286 * The timer IRQ doesn't have to know that behind the
1287 * scene we have a 8259A-master in AEOI mode ...
1288 */
1289 irq_desc[0].handler = &ioapic_edge_type;
1290
1291 /*
1292 * Add it to the IO-APIC irq-routing table:
1293 */
1294 spin_lock_irqsave(&ioapic_lock, flags);
1295 io_apic_write(0, 0x11+2*pin, *(((int *)&entry)+1));
1296 io_apic_write(0, 0x10+2*pin, *(((int *)&entry)+0));
1297 spin_unlock_irqrestore(&ioapic_lock, flags);
1298
1299 enable_8259A_irq(0);
1300}
1301
1302static inline void UNEXPECTED_IO_APIC(void)
1303{
1304}
1305
1306void __init print_IO_APIC(void)
1307{
1308 int apic, i;
1309 union IO_APIC_reg_00 reg_00;
1310 union IO_APIC_reg_01 reg_01;
1311 union IO_APIC_reg_02 reg_02;
1312 union IO_APIC_reg_03 reg_03;
1313 unsigned long flags;
1314
1315 if (apic_verbosity == APIC_QUIET)
1316 return;
1317
1318 printk(KERN_DEBUG "number of MP IRQ sources: %d.\n", mp_irq_entries);
1319 for (i = 0; i < nr_ioapics; i++)
1320 printk(KERN_DEBUG "number of IO-APIC #%d registers: %d.\n",
1321 mp_ioapics[i].mpc_apicid, nr_ioapic_registers[i]);
1322
1323 /*
1324 * We are a bit conservative about what we expect. We have to
1325 * know about every hardware change ASAP.
1326 */
1327 printk(KERN_INFO "testing the IO APIC.......................\n");
1328
1329 for (apic = 0; apic < nr_ioapics; apic++) {
1330
1331 spin_lock_irqsave(&ioapic_lock, flags);
1332 reg_00.raw = io_apic_read(apic, 0);
1333 reg_01.raw = io_apic_read(apic, 1);
1334 if (reg_01.bits.version >= 0x10)
1335 reg_02.raw = io_apic_read(apic, 2);
1336 if (reg_01.bits.version >= 0x20)
1337 reg_03.raw = io_apic_read(apic, 3);
1338 spin_unlock_irqrestore(&ioapic_lock, flags);
1339
1340 printk(KERN_DEBUG "IO APIC #%d......\n", mp_ioapics[apic].mpc_apicid);
1341 printk(KERN_DEBUG ".... register #00: %08X\n", reg_00.raw);
1342 printk(KERN_DEBUG "....... : physical APIC id: %02X\n", reg_00.bits.ID);
1343 printk(KERN_DEBUG "....... : Delivery Type: %X\n", reg_00.bits.delivery_type);
1344 printk(KERN_DEBUG "....... : LTS : %X\n", reg_00.bits.LTS);
1345 if (reg_00.bits.ID >= get_physical_broadcast())
1346 UNEXPECTED_IO_APIC();
1347 if (reg_00.bits.__reserved_1 || reg_00.bits.__reserved_2)
1348 UNEXPECTED_IO_APIC();
1349
1350 printk(KERN_DEBUG ".... register #01: %08X\n", reg_01.raw);
1351 printk(KERN_DEBUG "....... : max redirection entries: %04X\n", reg_01.bits.entries);
1352 if ( (reg_01.bits.entries != 0x0f) && /* older (Neptune) boards */
1353 (reg_01.bits.entries != 0x17) && /* typical ISA+PCI boards */
1354 (reg_01.bits.entries != 0x1b) && /* Compaq Proliant boards */
1355 (reg_01.bits.entries != 0x1f) && /* dual Xeon boards */
1356 (reg_01.bits.entries != 0x22) && /* bigger Xeon boards */
1357 (reg_01.bits.entries != 0x2E) &&
1358 (reg_01.bits.entries != 0x3F)
1359 )
1360 UNEXPECTED_IO_APIC();
1361
1362 printk(KERN_DEBUG "....... : PRQ implemented: %X\n", reg_01.bits.PRQ);
1363 printk(KERN_DEBUG "....... : IO APIC version: %04X\n", reg_01.bits.version);
1364 if ( (reg_01.bits.version != 0x01) && /* 82489DX IO-APICs */
1365 (reg_01.bits.version != 0x10) && /* oldest IO-APICs */
1366 (reg_01.bits.version != 0x11) && /* Pentium/Pro IO-APICs */
1367 (reg_01.bits.version != 0x13) && /* Xeon IO-APICs */
1368 (reg_01.bits.version != 0x20) /* Intel P64H (82806 AA) */
1369 )
1370 UNEXPECTED_IO_APIC();
1371 if (reg_01.bits.__reserved_1 || reg_01.bits.__reserved_2)
1372 UNEXPECTED_IO_APIC();
1373
1374 /*
1375 * Some Intel chipsets with IO APIC VERSION of 0x1? don't have reg_02,
1376 * but the value of reg_02 is read as the previous read register
1377 * value, so ignore it if reg_02 == reg_01.
1378 */
1379 if (reg_01.bits.version >= 0x10 && reg_02.raw != reg_01.raw) {
1380 printk(KERN_DEBUG ".... register #02: %08X\n", reg_02.raw);
1381 printk(KERN_DEBUG "....... : arbitration: %02X\n", reg_02.bits.arbitration);
1382 if (reg_02.bits.__reserved_1 || reg_02.bits.__reserved_2)
1383 UNEXPECTED_IO_APIC();
1384 }
1385
1386 /*
1387 * Some Intel chipsets with IO APIC VERSION of 0x2? don't have reg_02
1388 * or reg_03, but the value of reg_0[23] is read as the previous read
1389 * register value, so ignore it if reg_03 == reg_0[12].
1390 */
1391 if (reg_01.bits.version >= 0x20 && reg_03.raw != reg_02.raw &&
1392 reg_03.raw != reg_01.raw) {
1393 printk(KERN_DEBUG ".... register #03: %08X\n", reg_03.raw);
1394 printk(KERN_DEBUG "....... : Boot DT : %X\n", reg_03.bits.boot_DT);
1395 if (reg_03.bits.__reserved_1)
1396 UNEXPECTED_IO_APIC();
1397 }
1398
1399 printk(KERN_DEBUG ".... IRQ redirection table:\n");
1400
1401 printk(KERN_DEBUG " NR Log Phy Mask Trig IRR Pol"
1402 " Stat Dest Deli Vect: \n");
1403
1404 for (i = 0; i <= reg_01.bits.entries; i++) {
1405 struct IO_APIC_route_entry entry;
1406
1407 spin_lock_irqsave(&ioapic_lock, flags);
1408 *(((int *)&entry)+0) = io_apic_read(apic, 0x10+i*2);
1409 *(((int *)&entry)+1) = io_apic_read(apic, 0x11+i*2);
1410 spin_unlock_irqrestore(&ioapic_lock, flags);
1411
1412 printk(KERN_DEBUG " %02x %03X %02X ",
1413 i,
1414 entry.dest.logical.logical_dest,
1415 entry.dest.physical.physical_dest
1416 );
1417
1418 printk("%1d %1d %1d %1d %1d %1d %1d %02X\n",
1419 entry.mask,
1420 entry.trigger,
1421 entry.irr,
1422 entry.polarity,
1423 entry.delivery_status,
1424 entry.dest_mode,
1425 entry.delivery_mode,
1426 entry.vector
1427 );
1428 }
1429 }
1430 if (use_pci_vector())
1431 printk(KERN_INFO "Using vector-based indexing\n");
1432 printk(KERN_DEBUG "IRQ to pin mappings:\n");
1433 for (i = 0; i < NR_IRQS; i++) {
1434 struct irq_pin_list *entry = irq_2_pin + i;
1435 if (entry->pin < 0)
1436 continue;
1437 if (use_pci_vector() && !platform_legacy_irq(i))
1438 printk(KERN_DEBUG "IRQ%d ", IO_APIC_VECTOR(i));
1439 else
1440 printk(KERN_DEBUG "IRQ%d ", i);
1441 for (;;) {
1442 printk("-> %d:%d", entry->apic, entry->pin);
1443 if (!entry->next)
1444 break;
1445 entry = irq_2_pin + entry->next;
1446 }
1447 printk("\n");
1448 }
1449
1450 printk(KERN_INFO ".................................... done.\n");
1451
1452 return;
1453}
1454
1455#if 0
1456
1457static void print_APIC_bitfield (int base)
1458{
1459 unsigned int v;
1460 int i, j;
1461
1462 if (apic_verbosity == APIC_QUIET)
1463 return;
1464
1465 printk(KERN_DEBUG "0123456789abcdef0123456789abcdef\n" KERN_DEBUG);
1466 for (i = 0; i < 8; i++) {
1467 v = apic_read(base + i*0x10);
1468 for (j = 0; j < 32; j++) {
1469 if (v & (1<<j))
1470 printk("1");
1471 else
1472 printk("0");
1473 }
1474 printk("\n");
1475 }
1476}
1477
1478void /*__init*/ print_local_APIC(void * dummy)
1479{
1480 unsigned int v, ver, maxlvt;
1481
1482 if (apic_verbosity == APIC_QUIET)
1483 return;
1484
1485 printk("\n" KERN_DEBUG "printing local APIC contents on CPU#%d/%d:\n",
1486 smp_processor_id(), hard_smp_processor_id());
1487 v = apic_read(APIC_ID);
1488 printk(KERN_INFO "... APIC ID: %08x (%01x)\n", v, GET_APIC_ID(v));
1489 v = apic_read(APIC_LVR);
1490 printk(KERN_INFO "... APIC VERSION: %08x\n", v);
1491 ver = GET_APIC_VERSION(v);
1492 maxlvt = get_maxlvt();
1493
1494 v = apic_read(APIC_TASKPRI);
1495 printk(KERN_DEBUG "... APIC TASKPRI: %08x (%02x)\n", v, v & APIC_TPRI_MASK);
1496
1497 if (APIC_INTEGRATED(ver)) { /* !82489DX */
1498 v = apic_read(APIC_ARBPRI);
1499 printk(KERN_DEBUG "... APIC ARBPRI: %08x (%02x)\n", v,
1500 v & APIC_ARBPRI_MASK);
1501 v = apic_read(APIC_PROCPRI);
1502 printk(KERN_DEBUG "... APIC PROCPRI: %08x\n", v);
1503 }
1504
1505 v = apic_read(APIC_EOI);
1506 printk(KERN_DEBUG "... APIC EOI: %08x\n", v);
1507 v = apic_read(APIC_RRR);
1508 printk(KERN_DEBUG "... APIC RRR: %08x\n", v);
1509 v = apic_read(APIC_LDR);
1510 printk(KERN_DEBUG "... APIC LDR: %08x\n", v);
1511 v = apic_read(APIC_DFR);
1512 printk(KERN_DEBUG "... APIC DFR: %08x\n", v);
1513 v = apic_read(APIC_SPIV);
1514 printk(KERN_DEBUG "... APIC SPIV: %08x\n", v);
1515
1516 printk(KERN_DEBUG "... APIC ISR field:\n");
1517 print_APIC_bitfield(APIC_ISR);
1518 printk(KERN_DEBUG "... APIC TMR field:\n");
1519 print_APIC_bitfield(APIC_TMR);
1520 printk(KERN_DEBUG "... APIC IRR field:\n");
1521 print_APIC_bitfield(APIC_IRR);
1522
1523 if (APIC_INTEGRATED(ver)) { /* !82489DX */
1524 if (maxlvt > 3) /* Due to the Pentium erratum 3AP. */
1525 apic_write(APIC_ESR, 0);
1526 v = apic_read(APIC_ESR);
1527 printk(KERN_DEBUG "... APIC ESR: %08x\n", v);
1528 }
1529
1530 v = apic_read(APIC_ICR);
1531 printk(KERN_DEBUG "... APIC ICR: %08x\n", v);
1532 v = apic_read(APIC_ICR2);
1533 printk(KERN_DEBUG "... APIC ICR2: %08x\n", v);
1534
1535 v = apic_read(APIC_LVTT);
1536 printk(KERN_DEBUG "... APIC LVTT: %08x\n", v);
1537
1538 if (maxlvt > 3) { /* PC is LVT#4. */
1539 v = apic_read(APIC_LVTPC);
1540 printk(KERN_DEBUG "... APIC LVTPC: %08x\n", v);
1541 }
1542 v = apic_read(APIC_LVT0);
1543 printk(KERN_DEBUG "... APIC LVT0: %08x\n", v);
1544 v = apic_read(APIC_LVT1);
1545 printk(KERN_DEBUG "... APIC LVT1: %08x\n", v);
1546
1547 if (maxlvt > 2) { /* ERR is LVT#3. */
1548 v = apic_read(APIC_LVTERR);
1549 printk(KERN_DEBUG "... APIC LVTERR: %08x\n", v);
1550 }
1551
1552 v = apic_read(APIC_TMICT);
1553 printk(KERN_DEBUG "... APIC TMICT: %08x\n", v);
1554 v = apic_read(APIC_TMCCT);
1555 printk(KERN_DEBUG "... APIC TMCCT: %08x\n", v);
1556 v = apic_read(APIC_TDCR);
1557 printk(KERN_DEBUG "... APIC TDCR: %08x\n", v);
1558 printk("\n");
1559}
1560
1561void print_all_local_APICs (void)
1562{
1563 on_each_cpu(print_local_APIC, NULL, 1, 1);
1564}
1565
1566void /*__init*/ print_PIC(void)
1567{
1568 extern spinlock_t i8259A_lock;
1569 unsigned int v;
1570 unsigned long flags;
1571
1572 if (apic_verbosity == APIC_QUIET)
1573 return;
1574
1575 printk(KERN_DEBUG "\nprinting PIC contents\n");
1576
1577 spin_lock_irqsave(&i8259A_lock, flags);
1578
1579 v = inb(0xa1) << 8 | inb(0x21);
1580 printk(KERN_DEBUG "... PIC IMR: %04x\n", v);
1581
1582 v = inb(0xa0) << 8 | inb(0x20);
1583 printk(KERN_DEBUG "... PIC IRR: %04x\n", v);
1584
1585 outb(0x0b,0xa0);
1586 outb(0x0b,0x20);
1587 v = inb(0xa0) << 8 | inb(0x20);
1588 outb(0x0a,0xa0);
1589 outb(0x0a,0x20);
1590
1591 spin_unlock_irqrestore(&i8259A_lock, flags);
1592
1593 printk(KERN_DEBUG "... PIC ISR: %04x\n", v);
1594
1595 v = inb(0x4d1) << 8 | inb(0x4d0);
1596 printk(KERN_DEBUG "... PIC ELCR: %04x\n", v);
1597}
1598
1599#endif /* 0 */
1600
1601static void __init enable_IO_APIC(void)
1602{
1603 union IO_APIC_reg_01 reg_01;
1604 int i;
1605 unsigned long flags;
1606
1607 for (i = 0; i < PIN_MAP_SIZE; i++) {
1608 irq_2_pin[i].pin = -1;
1609 irq_2_pin[i].next = 0;
1610 }
1611 if (!pirqs_enabled)
1612 for (i = 0; i < MAX_PIRQS; i++)
1613 pirq_entries[i] = -1;
1614
1615 /*
1616 * The number of IO-APIC IRQ registers (== #pins):
1617 */
1618 for (i = 0; i < nr_ioapics; i++) {
1619 spin_lock_irqsave(&ioapic_lock, flags);
1620 reg_01.raw = io_apic_read(i, 1);
1621 spin_unlock_irqrestore(&ioapic_lock, flags);
1622 nr_ioapic_registers[i] = reg_01.bits.entries+1;
1623 }
1624
1625 /*
1626 * Do not trust the IO-APIC being empty at bootup
1627 */
1628 clear_IO_APIC();
1629}
1630
1631/*
1632 * Not an __init, needed by the reboot code
1633 */
1634void disable_IO_APIC(void)
1635{
1636 /*
1637 * Clear the IO-APIC before rebooting:
1638 */
1639 clear_IO_APIC();
1640
1641 disconnect_bsp_APIC();
1642}
1643
1644/*
1645 * function to set the IO-APIC physical IDs based on the
1646 * values stored in the MPC table.
1647 *
1648 * by Matt Domsch <Matt_Domsch@dell.com> Tue Dec 21 12:25:05 CST 1999
1649 */
1650
1651#ifndef CONFIG_X86_NUMAQ
1652static void __init setup_ioapic_ids_from_mpc(void)
1653{
1654 union IO_APIC_reg_00 reg_00;
1655 physid_mask_t phys_id_present_map;
1656 int apic;
1657 int i;
1658 unsigned char old_id;
1659 unsigned long flags;
1660
1661 /*
1662 * This is broken; anything with a real cpu count has to
1663 * circumvent this idiocy regardless.
1664 */
1665 phys_id_present_map = ioapic_phys_id_map(phys_cpu_present_map);
1666
1667 /*
1668 * Set the IOAPIC ID to the value stored in the MPC table.
1669 */
1670 for (apic = 0; apic < nr_ioapics; apic++) {
1671
1672 /* Read the register 0 value */
1673 spin_lock_irqsave(&ioapic_lock, flags);
1674 reg_00.raw = io_apic_read(apic, 0);
1675 spin_unlock_irqrestore(&ioapic_lock, flags);
1676
1677 old_id = mp_ioapics[apic].mpc_apicid;
1678
1679 if (mp_ioapics[apic].mpc_apicid >= get_physical_broadcast()) {
1680 printk(KERN_ERR "BIOS bug, IO-APIC#%d ID is %d in the MPC table!...\n",
1681 apic, mp_ioapics[apic].mpc_apicid);
1682 printk(KERN_ERR "... fixing up to %d. (tell your hw vendor)\n",
1683 reg_00.bits.ID);
1684 mp_ioapics[apic].mpc_apicid = reg_00.bits.ID;
1685 }
1686
1687 /* Don't check I/O APIC IDs for some xAPIC systems. They have
1688 * no meaning without the serial APIC bus. */
1689 if (NO_IOAPIC_CHECK)
1690 continue;
1691 /*
1692 * Sanity check, is the ID really free? Every APIC in a
1693 * system must have a unique ID or we get lots of nice
1694 * 'stuck on smp_invalidate_needed IPI wait' messages.
1695 */
1696 if (check_apicid_used(phys_id_present_map,
1697 mp_ioapics[apic].mpc_apicid)) {
1698 printk(KERN_ERR "BIOS bug, IO-APIC#%d ID %d is already used!...\n",
1699 apic, mp_ioapics[apic].mpc_apicid);
1700 for (i = 0; i < get_physical_broadcast(); i++)
1701 if (!physid_isset(i, phys_id_present_map))
1702 break;
1703 if (i >= get_physical_broadcast())
1704 panic("Max APIC ID exceeded!\n");
1705 printk(KERN_ERR "... fixing up to %d. (tell your hw vendor)\n",
1706 i);
1707 physid_set(i, phys_id_present_map);
1708 mp_ioapics[apic].mpc_apicid = i;
1709 } else {
1710 physid_mask_t tmp;
1711 tmp = apicid_to_cpu_present(mp_ioapics[apic].mpc_apicid);
1712 apic_printk(APIC_VERBOSE, "Setting %d in the "
1713 "phys_id_present_map\n",
1714 mp_ioapics[apic].mpc_apicid);
1715 physids_or(phys_id_present_map, phys_id_present_map, tmp);
1716 }
1717
1718
1719 /*
1720 * We need to adjust the IRQ routing table
1721 * if the ID changed.
1722 */
1723 if (old_id != mp_ioapics[apic].mpc_apicid)
1724 for (i = 0; i < mp_irq_entries; i++)
1725 if (mp_irqs[i].mpc_dstapic == old_id)
1726 mp_irqs[i].mpc_dstapic
1727 = mp_ioapics[apic].mpc_apicid;
1728
1729 /*
1730 * Read the right value from the MPC table and
1731 * write it into the ID register.
1732 */
1733 apic_printk(APIC_VERBOSE, KERN_INFO
1734 "...changing IO-APIC physical APIC ID to %d ...",
1735 mp_ioapics[apic].mpc_apicid);
1736
1737 reg_00.bits.ID = mp_ioapics[apic].mpc_apicid;
1738 spin_lock_irqsave(&ioapic_lock, flags);
1739 io_apic_write(apic, 0, reg_00.raw);
1740 spin_unlock_irqrestore(&ioapic_lock, flags);
1741
1742 /*
1743 * Sanity check
1744 */
1745 spin_lock_irqsave(&ioapic_lock, flags);
1746 reg_00.raw = io_apic_read(apic, 0);
1747 spin_unlock_irqrestore(&ioapic_lock, flags);
1748 if (reg_00.bits.ID != mp_ioapics[apic].mpc_apicid)
1749 printk("could not set ID!\n");
1750 else
1751 apic_printk(APIC_VERBOSE, " ok.\n");
1752 }
1753}
1754#else
1755static void __init setup_ioapic_ids_from_mpc(void) { }
1756#endif
1757
1758/*
1759 * There is a nasty bug in some older SMP boards, their mptable lies
1760 * about the timer IRQ. We do the following to work around the situation:
1761 *
1762 * - timer IRQ defaults to IO-APIC IRQ
1763 * - if this function detects that timer IRQs are defunct, then we fall
1764 * back to ISA timer IRQs
1765 */
1766static int __init timer_irq_works(void)
1767{
1768 unsigned long t1 = jiffies;
1769
1770 local_irq_enable();
1771 /* Let ten ticks pass... */
1772 mdelay((10 * 1000) / HZ);
1773
1774 /*
1775 * Expect a few ticks at least, to be sure some possible
1776 * glue logic does not lock up after one or two first
1777 * ticks in a non-ExtINT mode. Also the local APIC
1778 * might have cached one ExtINT interrupt. Finally, at
1779 * least one tick may be lost due to delays.
1780 */
1781 if (jiffies - t1 > 4)
1782 return 1;
1783
1784 return 0;
1785}
1786
1787/*
1788 * In the SMP+IOAPIC case it might happen that there are an unspecified
1789 * number of pending IRQ events unhandled. These cases are very rare,
1790 * so we 'resend' these IRQs via IPIs, to the same CPU. It's much
1791 * better to do it this way as thus we do not have to be aware of
1792 * 'pending' interrupts in the IRQ path, except at this point.
1793 */
1794/*
1795 * Edge triggered needs to resend any interrupt
1796 * that was delayed but this is now handled in the device
1797 * independent code.
1798 */
1799
1800/*
1801 * Starting up a edge-triggered IO-APIC interrupt is
1802 * nasty - we need to make sure that we get the edge.
1803 * If it is already asserted for some reason, we need
1804 * return 1 to indicate that is was pending.
1805 *
1806 * This is not complete - we should be able to fake
1807 * an edge even if it isn't on the 8259A...
1808 */
1809static unsigned int startup_edge_ioapic_irq(unsigned int irq)
1810{
1811 int was_pending = 0;
1812 unsigned long flags;
1813
1814 spin_lock_irqsave(&ioapic_lock, flags);
1815 if (irq < 16) {
1816 disable_8259A_irq(irq);
1817 if (i8259A_irq_pending(irq))
1818 was_pending = 1;
1819 }
1820 __unmask_IO_APIC_irq(irq);
1821 spin_unlock_irqrestore(&ioapic_lock, flags);
1822
1823 return was_pending;
1824}
1825
1826/*
1827 * Once we have recorded IRQ_PENDING already, we can mask the
1828 * interrupt for real. This prevents IRQ storms from unhandled
1829 * devices.
1830 */
1831static void ack_edge_ioapic_irq(unsigned int irq)
1832{
1833 move_irq(irq);
1834 if ((irq_desc[irq].status & (IRQ_PENDING | IRQ_DISABLED))
1835 == (IRQ_PENDING | IRQ_DISABLED))
1836 mask_IO_APIC_irq(irq);
1837 ack_APIC_irq();
1838}
1839
1840/*
1841 * Level triggered interrupts can just be masked,
1842 * and shutting down and starting up the interrupt
1843 * is the same as enabling and disabling them -- except
1844 * with a startup need to return a "was pending" value.
1845 *
1846 * Level triggered interrupts are special because we
1847 * do not touch any IO-APIC register while handling
1848 * them. We ack the APIC in the end-IRQ handler, not
1849 * in the start-IRQ-handler. Protection against reentrance
1850 * from the same interrupt is still provided, both by the
1851 * generic IRQ layer and by the fact that an unacked local
1852 * APIC does not accept IRQs.
1853 */
1854static unsigned int startup_level_ioapic_irq (unsigned int irq)
1855{
1856 unmask_IO_APIC_irq(irq);
1857
1858 return 0; /* don't check for pending */
1859}
1860
1861static void end_level_ioapic_irq (unsigned int irq)
1862{
1863 unsigned long v;
1864 int i;
1865
1866 move_irq(irq);
1867/*
1868 * It appears there is an erratum which affects at least version 0x11
1869 * of I/O APIC (that's the 82093AA and cores integrated into various
1870 * chipsets). Under certain conditions a level-triggered interrupt is
1871 * erroneously delivered as edge-triggered one but the respective IRR
1872 * bit gets set nevertheless. As a result the I/O unit expects an EOI
1873 * message but it will never arrive and further interrupts are blocked
1874 * from the source. The exact reason is so far unknown, but the
1875 * phenomenon was observed when two consecutive interrupt requests
1876 * from a given source get delivered to the same CPU and the source is
1877 * temporarily disabled in between.
1878 *
1879 * A workaround is to simulate an EOI message manually. We achieve it
1880 * by setting the trigger mode to edge and then to level when the edge
1881 * trigger mode gets detected in the TMR of a local APIC for a
1882 * level-triggered interrupt. We mask the source for the time of the
1883 * operation to prevent an edge-triggered interrupt escaping meanwhile.
1884 * The idea is from Manfred Spraul. --macro
1885 */
1886 i = IO_APIC_VECTOR(irq);
1887
1888 v = apic_read(APIC_TMR + ((i & ~0x1f) >> 1));
1889
1890 ack_APIC_irq();
1891
1892 if (!(v & (1 << (i & 0x1f)))) {
1893 atomic_inc(&irq_mis_count);
1894 spin_lock(&ioapic_lock);
1895 __mask_and_edge_IO_APIC_irq(irq);
1896 __unmask_and_level_IO_APIC_irq(irq);
1897 spin_unlock(&ioapic_lock);
1898 }
1899}
1900
1901#ifdef CONFIG_PCI_MSI
1902static unsigned int startup_edge_ioapic_vector(unsigned int vector)
1903{
1904 int irq = vector_to_irq(vector);
1905
1906 return startup_edge_ioapic_irq(irq);
1907}
1908
1909static void ack_edge_ioapic_vector(unsigned int vector)
1910{
1911 int irq = vector_to_irq(vector);
1912
1913 ack_edge_ioapic_irq(irq);
1914}
1915
1916static unsigned int startup_level_ioapic_vector (unsigned int vector)
1917{
1918 int irq = vector_to_irq(vector);
1919
1920 return startup_level_ioapic_irq (irq);
1921}
1922
1923static void end_level_ioapic_vector (unsigned int vector)
1924{
1925 int irq = vector_to_irq(vector);
1926
1927 end_level_ioapic_irq(irq);
1928}
1929
1930static void mask_IO_APIC_vector (unsigned int vector)
1931{
1932 int irq = vector_to_irq(vector);
1933
1934 mask_IO_APIC_irq(irq);
1935}
1936
1937static void unmask_IO_APIC_vector (unsigned int vector)
1938{
1939 int irq = vector_to_irq(vector);
1940
1941 unmask_IO_APIC_irq(irq);
1942}
1943
1944static void set_ioapic_affinity_vector (unsigned int vector,
1945 cpumask_t cpu_mask)
1946{
1947 int irq = vector_to_irq(vector);
1948
1949 set_ioapic_affinity_irq(irq, cpu_mask);
1950}
1951#endif
1952
1953/*
1954 * Level and edge triggered IO-APIC interrupts need different handling,
1955 * so we use two separate IRQ descriptors. Edge triggered IRQs can be
1956 * handled with the level-triggered descriptor, but that one has slightly
1957 * more overhead. Level-triggered interrupts cannot be handled with the
1958 * edge-triggered handler, without risking IRQ storms and other ugly
1959 * races.
1960 */
1961static struct hw_interrupt_type ioapic_edge_type = {
1962 .typename = "IO-APIC-edge",
1963 .startup = startup_edge_ioapic,
1964 .shutdown = shutdown_edge_ioapic,
1965 .enable = enable_edge_ioapic,
1966 .disable = disable_edge_ioapic,
1967 .ack = ack_edge_ioapic,
1968 .end = end_edge_ioapic,
1969 .set_affinity = set_ioapic_affinity,
1970};
1971
1972static struct hw_interrupt_type ioapic_level_type = {
1973 .typename = "IO-APIC-level",
1974 .startup = startup_level_ioapic,
1975 .shutdown = shutdown_level_ioapic,
1976 .enable = enable_level_ioapic,
1977 .disable = disable_level_ioapic,
1978 .ack = mask_and_ack_level_ioapic,
1979 .end = end_level_ioapic,
1980 .set_affinity = set_ioapic_affinity,
1981};
1982
1983static inline void init_IO_APIC_traps(void)
1984{
1985 int irq;
1986
1987 /*
1988 * NOTE! The local APIC isn't very good at handling
1989 * multiple interrupts at the same interrupt level.
1990 * As the interrupt level is determined by taking the
1991 * vector number and shifting that right by 4, we
1992 * want to spread these out a bit so that they don't
1993 * all fall in the same interrupt level.
1994 *
1995 * Also, we've got to be careful not to trash gate
1996 * 0x80, because int 0x80 is hm, kind of importantish. ;)
1997 */
1998 for (irq = 0; irq < NR_IRQS ; irq++) {
1999 int tmp = irq;
2000 if (use_pci_vector()) {
2001 if (!platform_legacy_irq(tmp))
2002 if ((tmp = vector_to_irq(tmp)) == -1)
2003 continue;
2004 }
2005 if (IO_APIC_IRQ(tmp) && !IO_APIC_VECTOR(tmp)) {
2006 /*
2007 * Hmm.. We don't have an entry for this,
2008 * so default to an old-fashioned 8259
2009 * interrupt if we can..
2010 */
2011 if (irq < 16)
2012 make_8259A_irq(irq);
2013 else
2014 /* Strange. Oh, well.. */
2015 irq_desc[irq].handler = &no_irq_type;
2016 }
2017 }
2018}
2019
2020static void enable_lapic_irq (unsigned int irq)
2021{
2022 unsigned long v;
2023
2024 v = apic_read(APIC_LVT0);
2025 apic_write_around(APIC_LVT0, v & ~APIC_LVT_MASKED);
2026}
2027
2028static void disable_lapic_irq (unsigned int irq)
2029{
2030 unsigned long v;
2031
2032 v = apic_read(APIC_LVT0);
2033 apic_write_around(APIC_LVT0, v | APIC_LVT_MASKED);
2034}
2035
2036static void ack_lapic_irq (unsigned int irq)
2037{
2038 ack_APIC_irq();
2039}
2040
2041static void end_lapic_irq (unsigned int i) { /* nothing */ }
2042
2043static struct hw_interrupt_type lapic_irq_type = {
2044 .typename = "local-APIC-edge",
2045 .startup = NULL, /* startup_irq() not used for IRQ0 */
2046 .shutdown = NULL, /* shutdown_irq() not used for IRQ0 */
2047 .enable = enable_lapic_irq,
2048 .disable = disable_lapic_irq,
2049 .ack = ack_lapic_irq,
2050 .end = end_lapic_irq
2051};
2052
2053static void setup_nmi (void)
2054{
2055 /*
2056 * Dirty trick to enable the NMI watchdog ...
2057 * We put the 8259A master into AEOI mode and
2058 * unmask on all local APICs LVT0 as NMI.
2059 *
2060 * The idea to use the 8259A in AEOI mode ('8259A Virtual Wire')
2061 * is from Maciej W. Rozycki - so we do not have to EOI from
2062 * the NMI handler or the timer interrupt.
2063 */
2064 apic_printk(APIC_VERBOSE, KERN_INFO "activating NMI Watchdog ...");
2065
2066 on_each_cpu(enable_NMI_through_LVT0, NULL, 1, 1);
2067
2068 apic_printk(APIC_VERBOSE, " done.\n");
2069}
2070
2071/*
2072 * This looks a bit hackish but it's about the only one way of sending
2073 * a few INTA cycles to 8259As and any associated glue logic. ICR does
2074 * not support the ExtINT mode, unfortunately. We need to send these
2075 * cycles as some i82489DX-based boards have glue logic that keeps the
2076 * 8259A interrupt line asserted until INTA. --macro
2077 */
2078static inline void unlock_ExtINT_logic(void)
2079{
2080 int pin, i;
2081 struct IO_APIC_route_entry entry0, entry1;
2082 unsigned char save_control, save_freq_select;
2083 unsigned long flags;
2084
2085 pin = find_isa_irq_pin(8, mp_INT);
2086 if (pin == -1)
2087 return;
2088
2089 spin_lock_irqsave(&ioapic_lock, flags);
2090 *(((int *)&entry0) + 1) = io_apic_read(0, 0x11 + 2 * pin);
2091 *(((int *)&entry0) + 0) = io_apic_read(0, 0x10 + 2 * pin);
2092 spin_unlock_irqrestore(&ioapic_lock, flags);
2093 clear_IO_APIC_pin(0, pin);
2094
2095 memset(&entry1, 0, sizeof(entry1));
2096
2097 entry1.dest_mode = 0; /* physical delivery */
2098 entry1.mask = 0; /* unmask IRQ now */
2099 entry1.dest.physical.physical_dest = hard_smp_processor_id();
2100 entry1.delivery_mode = dest_ExtINT;
2101 entry1.polarity = entry0.polarity;
2102 entry1.trigger = 0;
2103 entry1.vector = 0;
2104
2105 spin_lock_irqsave(&ioapic_lock, flags);
2106 io_apic_write(0, 0x11 + 2 * pin, *(((int *)&entry1) + 1));
2107 io_apic_write(0, 0x10 + 2 * pin, *(((int *)&entry1) + 0));
2108 spin_unlock_irqrestore(&ioapic_lock, flags);
2109
2110 save_control = CMOS_READ(RTC_CONTROL);
2111 save_freq_select = CMOS_READ(RTC_FREQ_SELECT);
2112 CMOS_WRITE((save_freq_select & ~RTC_RATE_SELECT) | 0x6,
2113 RTC_FREQ_SELECT);
2114 CMOS_WRITE(save_control | RTC_PIE, RTC_CONTROL);
2115
2116 i = 100;
2117 while (i-- > 0) {
2118 mdelay(10);
2119 if ((CMOS_READ(RTC_INTR_FLAGS) & RTC_PF) == RTC_PF)
2120 i -= 10;
2121 }
2122
2123 CMOS_WRITE(save_control, RTC_CONTROL);
2124 CMOS_WRITE(save_freq_select, RTC_FREQ_SELECT);
2125 clear_IO_APIC_pin(0, pin);
2126
2127 spin_lock_irqsave(&ioapic_lock, flags);
2128 io_apic_write(0, 0x11 + 2 * pin, *(((int *)&entry0) + 1));
2129 io_apic_write(0, 0x10 + 2 * pin, *(((int *)&entry0) + 0));
2130 spin_unlock_irqrestore(&ioapic_lock, flags);
2131}
2132
2133/*
2134 * This code may look a bit paranoid, but it's supposed to cooperate with
2135 * a wide range of boards and BIOS bugs. Fortunately only the timer IRQ
2136 * is so screwy. Thanks to Brian Perkins for testing/hacking this beast
2137 * fanatically on his truly buggy board.
2138 */
2139static inline void check_timer(void)
2140{
2141 int pin1, pin2;
2142 int vector;
2143
2144 /*
2145 * get/set the timer IRQ vector:
2146 */
2147 disable_8259A_irq(0);
2148 vector = assign_irq_vector(0);
2149 set_intr_gate(vector, interrupt[0]);
2150
2151 /*
2152 * Subtle, code in do_timer_interrupt() expects an AEOI
2153 * mode for the 8259A whenever interrupts are routed
2154 * through I/O APICs. Also IRQ0 has to be enabled in
2155 * the 8259A which implies the virtual wire has to be
2156 * disabled in the local APIC.
2157 */
2158 apic_write_around(APIC_LVT0, APIC_LVT_MASKED | APIC_DM_EXTINT);
2159 init_8259A(1);
2160 timer_ack = 1;
2161 enable_8259A_irq(0);
2162
2163 pin1 = find_isa_irq_pin(0, mp_INT);
2164 pin2 = find_isa_irq_pin(0, mp_ExtINT);
2165
2166 printk(KERN_INFO "..TIMER: vector=0x%02X pin1=%d pin2=%d\n", vector, pin1, pin2);
2167
2168 if (pin1 != -1) {
2169 /*
2170 * Ok, does IRQ0 through the IOAPIC work?
2171 */
2172 unmask_IO_APIC_irq(0);
2173 if (timer_irq_works()) {
2174 if (nmi_watchdog == NMI_IO_APIC) {
2175 disable_8259A_irq(0);
2176 setup_nmi();
2177 enable_8259A_irq(0);
2178 check_nmi_watchdog();
2179 }
2180 return;
2181 }
2182 clear_IO_APIC_pin(0, pin1);
2183 printk(KERN_ERR "..MP-BIOS bug: 8254 timer not connected to IO-APIC\n");
2184 }
2185
2186 printk(KERN_INFO "...trying to set up timer (IRQ0) through the 8259A ... ");
2187 if (pin2 != -1) {
2188 printk("\n..... (found pin %d) ...", pin2);
2189 /*
2190 * legacy devices should be connected to IO APIC #0
2191 */
2192 setup_ExtINT_IRQ0_pin(pin2, vector);
2193 if (timer_irq_works()) {
2194 printk("works.\n");
2195 if (pin1 != -1)
2196 replace_pin_at_irq(0, 0, pin1, 0, pin2);
2197 else
2198 add_pin_to_irq(0, 0, pin2);
2199 if (nmi_watchdog == NMI_IO_APIC) {
2200 setup_nmi();
2201 check_nmi_watchdog();
2202 }
2203 return;
2204 }
2205 /*
2206 * Cleanup, just in case ...
2207 */
2208 clear_IO_APIC_pin(0, pin2);
2209 }
2210 printk(" failed.\n");
2211
2212 if (nmi_watchdog == NMI_IO_APIC) {
2213 printk(KERN_WARNING "timer doesn't work through the IO-APIC - disabling NMI Watchdog!\n");
2214 nmi_watchdog = 0;
2215 }
2216
2217 printk(KERN_INFO "...trying to set up timer as Virtual Wire IRQ...");
2218
2219 disable_8259A_irq(0);
2220 irq_desc[0].handler = &lapic_irq_type;
2221 apic_write_around(APIC_LVT0, APIC_DM_FIXED | vector); /* Fixed mode */
2222 enable_8259A_irq(0);
2223
2224 if (timer_irq_works()) {
2225 printk(" works.\n");
2226 return;
2227 }
2228 apic_write_around(APIC_LVT0, APIC_LVT_MASKED | APIC_DM_FIXED | vector);
2229 printk(" failed.\n");
2230
2231 printk(KERN_INFO "...trying to set up timer as ExtINT IRQ...");
2232
2233 timer_ack = 0;
2234 init_8259A(0);
2235 make_8259A_irq(0);
2236 apic_write_around(APIC_LVT0, APIC_DM_EXTINT);
2237
2238 unlock_ExtINT_logic();
2239
2240 if (timer_irq_works()) {
2241 printk(" works.\n");
2242 return;
2243 }
2244 printk(" failed :(.\n");
2245 panic("IO-APIC + timer doesn't work! Boot with apic=debug and send a "
2246 "report. Then try booting with the 'noapic' option");
2247}
2248
2249/*
2250 *
2251 * IRQ's that are handled by the PIC in the MPS IOAPIC case.
2252 * - IRQ2 is the cascade IRQ, and cannot be a io-apic IRQ.
2253 * Linux doesn't really care, as it's not actually used
2254 * for any interrupt handling anyway.
2255 */
2256#define PIC_IRQS (1 << PIC_CASCADE_IR)
2257
2258void __init setup_IO_APIC(void)
2259{
2260 enable_IO_APIC();
2261
2262 if (acpi_ioapic)
2263 io_apic_irqs = ~0; /* all IRQs go through IOAPIC */
2264 else
2265 io_apic_irqs = ~PIC_IRQS;
2266
2267 printk("ENABLING IO-APIC IRQs\n");
2268
2269 /*
2270 * Set up IO-APIC IRQ routing.
2271 */
2272 if (!acpi_ioapic)
2273 setup_ioapic_ids_from_mpc();
2274 sync_Arb_IDs();
2275 setup_IO_APIC_irqs();
2276 init_IO_APIC_traps();
2277 check_timer();
2278 if (!acpi_ioapic)
2279 print_IO_APIC();
2280}
2281
2282/*
2283 * Called after all the initialization is done. If we didnt find any
2284 * APIC bugs then we can allow the modify fast path
2285 */
2286
2287static int __init io_apic_bug_finalize(void)
2288{
2289 if(sis_apic_bug == -1)
2290 sis_apic_bug = 0;
2291 return 0;
2292}
2293
2294late_initcall(io_apic_bug_finalize);
2295
2296struct sysfs_ioapic_data {
2297 struct sys_device dev;
2298 struct IO_APIC_route_entry entry[0];
2299};
2300static struct sysfs_ioapic_data * mp_ioapic_data[MAX_IO_APICS];
2301
2302static int ioapic_suspend(struct sys_device *dev, u32 state)
2303{
2304 struct IO_APIC_route_entry *entry;
2305 struct sysfs_ioapic_data *data;
2306 unsigned long flags;
2307 int i;
2308
2309 data = container_of(dev, struct sysfs_ioapic_data, dev);
2310 entry = data->entry;
2311 spin_lock_irqsave(&ioapic_lock, flags);
2312 for (i = 0; i < nr_ioapic_registers[dev->id]; i ++, entry ++ ) {
2313 *(((int *)entry) + 1) = io_apic_read(dev->id, 0x11 + 2 * i);
2314 *(((int *)entry) + 0) = io_apic_read(dev->id, 0x10 + 2 * i);
2315 }
2316 spin_unlock_irqrestore(&ioapic_lock, flags);
2317
2318 return 0;
2319}
2320
2321static int ioapic_resume(struct sys_device *dev)
2322{
2323 struct IO_APIC_route_entry *entry;
2324 struct sysfs_ioapic_data *data;
2325 unsigned long flags;
2326 union IO_APIC_reg_00 reg_00;
2327 int i;
2328
2329 data = container_of(dev, struct sysfs_ioapic_data, dev);
2330 entry = data->entry;
2331
2332 spin_lock_irqsave(&ioapic_lock, flags);
2333 reg_00.raw = io_apic_read(dev->id, 0);
2334 if (reg_00.bits.ID != mp_ioapics[dev->id].mpc_apicid) {
2335 reg_00.bits.ID = mp_ioapics[dev->id].mpc_apicid;
2336 io_apic_write(dev->id, 0, reg_00.raw);
2337 }
2338 for (i = 0; i < nr_ioapic_registers[dev->id]; i ++, entry ++ ) {
2339 io_apic_write(dev->id, 0x11+2*i, *(((int *)entry)+1));
2340 io_apic_write(dev->id, 0x10+2*i, *(((int *)entry)+0));
2341 }
2342 spin_unlock_irqrestore(&ioapic_lock, flags);
2343
2344 return 0;
2345}
2346
2347static struct sysdev_class ioapic_sysdev_class = {
2348 set_kset_name("ioapic"),
2349 .suspend = ioapic_suspend,
2350 .resume = ioapic_resume,
2351};
2352
2353static int __init ioapic_init_sysfs(void)
2354{
2355 struct sys_device * dev;
2356 int i, size, error = 0;
2357
2358 error = sysdev_class_register(&ioapic_sysdev_class);
2359 if (error)
2360 return error;
2361
2362 for (i = 0; i < nr_ioapics; i++ ) {
2363 size = sizeof(struct sys_device) + nr_ioapic_registers[i]
2364 * sizeof(struct IO_APIC_route_entry);
2365 mp_ioapic_data[i] = kmalloc(size, GFP_KERNEL);
2366 if (!mp_ioapic_data[i]) {
2367 printk(KERN_ERR "Can't suspend/resume IOAPIC %d\n", i);
2368 continue;
2369 }
2370 memset(mp_ioapic_data[i], 0, size);
2371 dev = &mp_ioapic_data[i]->dev;
2372 dev->id = i;
2373 dev->cls = &ioapic_sysdev_class;
2374 error = sysdev_register(dev);
2375 if (error) {
2376 kfree(mp_ioapic_data[i]);
2377 mp_ioapic_data[i] = NULL;
2378 printk(KERN_ERR "Can't suspend/resume IOAPIC %d\n", i);
2379 continue;
2380 }
2381 }
2382
2383 return 0;
2384}
2385
2386device_initcall(ioapic_init_sysfs);
2387
2388/* --------------------------------------------------------------------------
2389 ACPI-based IOAPIC Configuration
2390 -------------------------------------------------------------------------- */
2391
2392#ifdef CONFIG_ACPI_BOOT
2393
2394int __init io_apic_get_unique_id (int ioapic, int apic_id)
2395{
2396 union IO_APIC_reg_00 reg_00;
2397 static physid_mask_t apic_id_map = PHYSID_MASK_NONE;
2398 physid_mask_t tmp;
2399 unsigned long flags;
2400 int i = 0;
2401
2402 /*
2403 * The P4 platform supports up to 256 APIC IDs on two separate APIC
2404 * buses (one for LAPICs, one for IOAPICs), where predecessors only
2405 * supports up to 16 on one shared APIC bus.
2406 *
2407 * TBD: Expand LAPIC/IOAPIC support on P4-class systems to take full
2408 * advantage of new APIC bus architecture.
2409 */
2410
2411 if (physids_empty(apic_id_map))
2412 apic_id_map = ioapic_phys_id_map(phys_cpu_present_map);
2413
2414 spin_lock_irqsave(&ioapic_lock, flags);
2415 reg_00.raw = io_apic_read(ioapic, 0);
2416 spin_unlock_irqrestore(&ioapic_lock, flags);
2417
2418 if (apic_id >= get_physical_broadcast()) {
2419 printk(KERN_WARNING "IOAPIC[%d]: Invalid apic_id %d, trying "
2420 "%d\n", ioapic, apic_id, reg_00.bits.ID);
2421 apic_id = reg_00.bits.ID;
2422 }
2423
2424 /*
2425 * Every APIC in a system must have a unique ID or we get lots of nice
2426 * 'stuck on smp_invalidate_needed IPI wait' messages.
2427 */
2428 if (check_apicid_used(apic_id_map, apic_id)) {
2429
2430 for (i = 0; i < get_physical_broadcast(); i++) {
2431 if (!check_apicid_used(apic_id_map, i))
2432 break;
2433 }
2434
2435 if (i == get_physical_broadcast())
2436 panic("Max apic_id exceeded!\n");
2437
2438 printk(KERN_WARNING "IOAPIC[%d]: apic_id %d already used, "
2439 "trying %d\n", ioapic, apic_id, i);
2440
2441 apic_id = i;
2442 }
2443
2444 tmp = apicid_to_cpu_present(apic_id);
2445 physids_or(apic_id_map, apic_id_map, tmp);
2446
2447 if (reg_00.bits.ID != apic_id) {
2448 reg_00.bits.ID = apic_id;
2449
2450 spin_lock_irqsave(&ioapic_lock, flags);
2451 io_apic_write(ioapic, 0, reg_00.raw);
2452 reg_00.raw = io_apic_read(ioapic, 0);
2453 spin_unlock_irqrestore(&ioapic_lock, flags);
2454
2455 /* Sanity check */
2456 if (reg_00.bits.ID != apic_id)
2457 panic("IOAPIC[%d]: Unable change apic_id!\n", ioapic);
2458 }
2459
2460 apic_printk(APIC_VERBOSE, KERN_INFO
2461 "IOAPIC[%d]: Assigned apic_id %d\n", ioapic, apic_id);
2462
2463 return apic_id;
2464}
2465
2466
2467int __init io_apic_get_version (int ioapic)
2468{
2469 union IO_APIC_reg_01 reg_01;
2470 unsigned long flags;
2471
2472 spin_lock_irqsave(&ioapic_lock, flags);
2473 reg_01.raw = io_apic_read(ioapic, 1);
2474 spin_unlock_irqrestore(&ioapic_lock, flags);
2475
2476 return reg_01.bits.version;
2477}
2478
2479
2480int __init io_apic_get_redir_entries (int ioapic)
2481{
2482 union IO_APIC_reg_01 reg_01;
2483 unsigned long flags;
2484
2485 spin_lock_irqsave(&ioapic_lock, flags);
2486 reg_01.raw = io_apic_read(ioapic, 1);
2487 spin_unlock_irqrestore(&ioapic_lock, flags);
2488
2489 return reg_01.bits.entries;
2490}
2491
2492
2493int io_apic_set_pci_routing (int ioapic, int pin, int irq, int edge_level, int active_high_low)
2494{
2495 struct IO_APIC_route_entry entry;
2496 unsigned long flags;
2497
2498 if (!IO_APIC_IRQ(irq)) {
2499 printk(KERN_ERR "IOAPIC[%d]: Invalid reference to IRQ 0\n",
2500 ioapic);
2501 return -EINVAL;
2502 }
2503
2504 /*
2505 * Generate a PCI IRQ routing entry and program the IOAPIC accordingly.
2506 * Note that we mask (disable) IRQs now -- these get enabled when the
2507 * corresponding device driver registers for this IRQ.
2508 */
2509
2510 memset(&entry,0,sizeof(entry));
2511
2512 entry.delivery_mode = INT_DELIVERY_MODE;
2513 entry.dest_mode = INT_DEST_MODE;
2514 entry.dest.logical.logical_dest = cpu_mask_to_apicid(TARGET_CPUS);
2515 entry.trigger = edge_level;
2516 entry.polarity = active_high_low;
2517 entry.mask = 1;
2518
2519 /*
2520 * IRQs < 16 are already in the irq_2_pin[] map
2521 */
2522 if (irq >= 16)
2523 add_pin_to_irq(irq, ioapic, pin);
2524
2525 entry.vector = assign_irq_vector(irq);
2526
2527 apic_printk(APIC_DEBUG, KERN_DEBUG "IOAPIC[%d]: Set PCI routing entry "
2528 "(%d-%d -> 0x%x -> IRQ %d Mode:%i Active:%i)\n", ioapic,
2529 mp_ioapics[ioapic].mpc_apicid, pin, entry.vector, irq,
2530 edge_level, active_high_low);
2531
2532 ioapic_register_intr(irq, entry.vector, edge_level);
2533
2534 if (!ioapic && (irq < 16))
2535 disable_8259A_irq(irq);
2536
2537 spin_lock_irqsave(&ioapic_lock, flags);
2538 io_apic_write(ioapic, 0x11+2*pin, *(((int *)&entry)+1));
2539 io_apic_write(ioapic, 0x10+2*pin, *(((int *)&entry)+0));
2540 spin_unlock_irqrestore(&ioapic_lock, flags);
2541
2542 return 0;
2543}
2544
2545#endif /*CONFIG_ACPI_BOOT*/
generated by cgit v1.2.2 (git 2.25.0) at 2024-11-07 14:08:34 -0500