projects / linux-2.6-block.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
Merge branch 'x86-fixes-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git...
[linux-2.6-block.git] / arch / x86 / 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>
1da177e4
LT		 24#include <linux/interrupt.h>
25#include <linux/init.h>
26#include <linux/delay.h>
27#include <linux/sched.h>
d4057bdb 28#include <linux/pci.h>
1da177e4
LT		 29#include <linux/mc146818rtc.h>
30#include <linux/compiler.h>
31#include <linux/acpi.h>
129f6946 32#include <linux/module.h>
1da177e4 33#include <linux/sysdev.h>
3b7d1921 34#include <linux/msi.h>
95d77884 35#include <linux/htirq.h>
7dfb7103 36#include <linux/freezer.h>
f26d6a2b 37#include <linux/kthread.h>
54168ed7 38#include <linux/jiffies.h> /* time_after() */
d4057bdb
YL		 39#ifdef CONFIG_ACPI
40#include <acpi/acpi_bus.h>
41#endif
42#include <linux/bootmem.h>
43#include <linux/dmar.h>
58ac1e76 44#include <linux/hpet.h>
54d5d424 45
d4057bdb 46#include <asm/idle.h>
1da177e4
LT		 47#include <asm/io.h>
48#include <asm/smp.h>
49#include <asm/desc.h>
d4057bdb
YL		 50#include <asm/proto.h>
51#include <asm/acpi.h>
52#include <asm/dma.h>
1da177e4 53#include <asm/timer.h>
306e440d 54#include <asm/i8259.h>
3e4ff115 55#include <asm/nmi.h>
2d3fcc1c 56#include <asm/msidef.h>
8b955b0d 57#include <asm/hypertransport.h>
a4dbc34d 58#include <asm/setup.h>
d4057bdb 59#include <asm/irq_remapping.h>
58ac1e76 60#include <asm/hpet.h>
4173a0e7
DN		 61#include <asm/uv/uv_hub.h>
62#include <asm/uv/uv_irq.h>
1da177e4 63
497c9a19 64#include <mach_ipi.h>
1da177e4 65#include <mach_apic.h>
874c4fe3 66#include <mach_apicdef.h>
1da177e4 67
32f71aff
MR		 68#define __apicdebuginit(type) static type __init
69
1da177e4 70/*
54168ed7
IM		 71 * Is the SiS APIC rmw bug present ?
72 * -1 = don't know, 0 = no, 1 = yes
1da177e4
LT		 73 */
74int sis_apic_bug = -1;
75
efa2559f
YL		 76static DEFINE_SPINLOCK(ioapic_lock);
77static DEFINE_SPINLOCK(vector_lock);
78
1da177e4
LT		 79/*
80 * # of IRQ routing registers
81 */
82int nr_ioapic_registers[MAX_IO_APICS];
83
9f640ccb 84/* I/O APIC entries */
ec2cd0a2 85struct mp_config_ioapic mp_ioapics[MAX_IO_APICS];
9f640ccb
AS		 86int nr_ioapics;
87
584f734d 88/* MP IRQ source entries */
2fddb6e2 89struct mp_config_intsrc mp_irqs[MAX_IRQ_SOURCES];
584f734d
AS		 90
91/* # of MP IRQ source entries */
92int mp_irq_entries;
93
8732fc4b
AS		 94#if defined (CONFIG_MCA) || defined (CONFIG_EISA)
95int mp_bus_id_to_type[MAX_MP_BUSSES];
96#endif
97
98DECLARE_BITMAP(mp_bus_not_pci, MAX_MP_BUSSES);
99
efa2559f
YL		 100int skip_ioapic_setup;
101
54168ed7 102static int __init parse_noapic(char *str)
efa2559f
YL		 103{
104 /* disable IO-APIC */
105 disable_ioapic_setup();
106 return 0;
107}
108early_param("noapic", parse_noapic);
66759a01 109
0f978f45 110struct irq_pin_list;
0b8f1efa
YL		 111
112/*
113 * This is performance-critical, we want to do it O(1)
114 *
115 * the indexing order of this array favors 1:1 mappings
116 * between pins and IRQs.
117 */
118
119struct irq_pin_list {
120 int apic, pin;
121 struct irq_pin_list *next;
122};
123
124static struct irq_pin_list *get_one_free_irq_2_pin(int cpu)
125{
126 struct irq_pin_list *pin;
127 int node;
128
129 node = cpu_to_node(cpu);
130
131 pin = kzalloc_node(sizeof(*pin), GFP_ATOMIC, node);
0b8f1efa
YL		 132
133 return pin;
134}
135
a1420f39 136struct irq_cfg {
0f978f45 137 struct irq_pin_list *irq_2_pin;
22f65d31
MT		 138 cpumask_var_t domain;
139 cpumask_var_t old_domain;
497c9a19 140 unsigned move_cleanup_count;
a1420f39 141 u8 vector;
497c9a19 142 u8 move_in_progress : 1;
48a1b10a
YL		 143#ifdef CONFIG_NUMA_MIGRATE_IRQ_DESC
144 u8 move_desc_pending : 1;
145#endif
a1420f39
YL		 146};
147
a1420f39 148/* irq_cfg is indexed by the sum of all RTEs in all I/O APICs. */
0b8f1efa
YL		 149#ifdef CONFIG_SPARSE_IRQ
150static struct irq_cfg irq_cfgx[] = {
151#else
d6c88a50 152static struct irq_cfg irq_cfgx[NR_IRQS] = {
0b8f1efa 153#endif
22f65d31
MT		 154 [0] = { .vector = IRQ0_VECTOR, },
155 [1] = { .vector = IRQ1_VECTOR, },
156 [2] = { .vector = IRQ2_VECTOR, },
157 [3] = { .vector = IRQ3_VECTOR, },
158 [4] = { .vector = IRQ4_VECTOR, },
159 [5] = { .vector = IRQ5_VECTOR, },
160 [6] = { .vector = IRQ6_VECTOR, },
161 [7] = { .vector = IRQ7_VECTOR, },
162 [8] = { .vector = IRQ8_VECTOR, },
163 [9] = { .vector = IRQ9_VECTOR, },
164 [10] = { .vector = IRQ10_VECTOR, },
165 [11] = { .vector = IRQ11_VECTOR, },
166 [12] = { .vector = IRQ12_VECTOR, },
167 [13] = { .vector = IRQ13_VECTOR, },
168 [14] = { .vector = IRQ14_VECTOR, },
169 [15] = { .vector = IRQ15_VECTOR, },
a1420f39
YL		 170};
171
13a0c3c2 172int __init arch_early_irq_init(void)
8f09cd20 173{
0b8f1efa
YL		 174 struct irq_cfg *cfg;
175 struct irq_desc *desc;
176 int count;
177 int i;
d6c88a50 178
0b8f1efa
YL		 179 cfg = irq_cfgx;
180 count = ARRAY_SIZE(irq_cfgx);
8f09cd20 181
0b8f1efa
YL		 182 for (i = 0; i < count; i++) {
183 desc = irq_to_desc(i);
184 desc->chip_data = &cfg[i];
22f65d31
MT		 185 alloc_bootmem_cpumask_var(&cfg[i].domain);
186 alloc_bootmem_cpumask_var(&cfg[i].old_domain);
187 if (i < NR_IRQS_LEGACY)
188 cpumask_setall(cfg[i].domain);
0b8f1efa 189 }
13a0c3c2
YL		 190
191 return 0;
0b8f1efa 192}
8f09cd20 193
0b8f1efa 194#ifdef CONFIG_SPARSE_IRQ
d6c88a50 195static struct irq_cfg *irq_cfg(unsigned int irq)
8f09cd20 196{
0b8f1efa
YL		 197 struct irq_cfg *cfg = NULL;
198 struct irq_desc *desc;
1da177e4 199
0b8f1efa
YL		 200 desc = irq_to_desc(irq);
201 if (desc)
202 cfg = desc->chip_data;
0f978f45 203
0b8f1efa 204 return cfg;
8f09cd20 205}
d6c88a50 206
0b8f1efa 207static struct irq_cfg *get_one_free_irq_cfg(int cpu)
8f09cd20 208{
0b8f1efa
YL		 209 struct irq_cfg *cfg;
210 int node;
211
212 node = cpu_to_node(cpu);
0f978f45 213
0b8f1efa 214 cfg = kzalloc_node(sizeof(*cfg), GFP_ATOMIC, node);
22f65d31 215 if (cfg) {
80855f73 216 if (!alloc_cpumask_var_node(&cfg->domain, GFP_ATOMIC, node)) {
22f65d31
MT		 217 kfree(cfg);
218 cfg = NULL;
80855f73
MT		 219 } else if (!alloc_cpumask_var_node(&cfg->old_domain,
220 GFP_ATOMIC, node)) {
22f65d31
MT		 221 free_cpumask_var(cfg->domain);
222 kfree(cfg);
223 cfg = NULL;
224 } else {
225 cpumask_clear(cfg->domain);
226 cpumask_clear(cfg->old_domain);
227 }
228 }
0f978f45 229
0b8f1efa 230 return cfg;
8f09cd20
YL		 231}
232
13a0c3c2 233int arch_init_chip_data(struct irq_desc *desc, int cpu)
0f978f45 234{
0b8f1efa 235 struct irq_cfg *cfg;
d6c88a50 236
0b8f1efa
YL		 237 cfg = desc->chip_data;
238 if (!cfg) {
239 desc->chip_data = get_one_free_irq_cfg(cpu);
240 if (!desc->chip_data) {
241 printk(KERN_ERR "can not alloc irq_cfg\n");
242 BUG_ON(1);
243 }
244 }
1da177e4 245
13a0c3c2 246 return 0;
0b8f1efa 247}
0f978f45 248
48a1b10a 249#ifdef CONFIG_NUMA_MIGRATE_IRQ_DESC
d6c88a50 250
48a1b10a
YL		 251static void
252init_copy_irq_2_pin(struct irq_cfg *old_cfg, struct irq_cfg *cfg, int cpu)
0f978f45 253{
48a1b10a
YL		 254 struct irq_pin_list *old_entry, *head, *tail, *entry;
255
256 cfg->irq_2_pin = NULL;
257 old_entry = old_cfg->irq_2_pin;
258 if (!old_entry)
259 return;
0f978f45 260
48a1b10a
YL		 261 entry = get_one_free_irq_2_pin(cpu);
262 if (!entry)
263 return;
0f978f45 264
48a1b10a
YL		 265 entry->apic = old_entry->apic;
266 entry->pin = old_entry->pin;
267 head = entry;
268 tail = entry;
269 old_entry = old_entry->next;
270 while (old_entry) {
271 entry = get_one_free_irq_2_pin(cpu);
272 if (!entry) {
273 entry = head;
274 while (entry) {
275 head = entry->next;
276 kfree(entry);
277 entry = head;
278 }
279 /* still use the old one */
280 return;
281 }
282 entry->apic = old_entry->apic;
283 entry->pin = old_entry->pin;
284 tail->next = entry;
285 tail = entry;
286 old_entry = old_entry->next;
287 }
0f978f45 288
48a1b10a
YL		 289 tail->next = NULL;
290 cfg->irq_2_pin = head;
0f978f45 291}
0f978f45 292
48a1b10a 293static void free_irq_2_pin(struct irq_cfg *old_cfg, struct irq_cfg *cfg)
0f978f45 294{
48a1b10a 295 struct irq_pin_list *entry, *next;
0f978f45 296
48a1b10a
YL		 297 if (old_cfg->irq_2_pin == cfg->irq_2_pin)
298 return;
301e6190 299
48a1b10a 300 entry = old_cfg->irq_2_pin;
0f978f45 301
48a1b10a
YL		 302 while (entry) {
303 next = entry->next;
304 kfree(entry);
305 entry = next;
306 }
307 old_cfg->irq_2_pin = NULL;
0f978f45 308}
0f978f45 309
48a1b10a
YL		 310void arch_init_copy_chip_data(struct irq_desc *old_desc,
311 struct irq_desc *desc, int cpu)
0f978f45 312{
48a1b10a
YL		 313 struct irq_cfg *cfg;
314 struct irq_cfg *old_cfg;
0f978f45 315
48a1b10a 316 cfg = get_one_free_irq_cfg(cpu);
301e6190 317
48a1b10a
YL		 318 if (!cfg)
319 return;
320
321 desc->chip_data = cfg;
322
323 old_cfg = old_desc->chip_data;
324
325 memcpy(cfg, old_cfg, sizeof(struct irq_cfg));
326
327 init_copy_irq_2_pin(old_cfg, cfg, cpu);
0f978f45 328}
1da177e4 329
48a1b10a
YL		 330static void free_irq_cfg(struct irq_cfg *old_cfg)
331{
332 kfree(old_cfg);
333}
334
335void arch_free_chip_data(struct irq_desc *old_desc, struct irq_desc *desc)
336{
337 struct irq_cfg *old_cfg, *cfg;
338
339 old_cfg = old_desc->chip_data;
340 cfg = desc->chip_data;
341
342 if (old_cfg == cfg)
343 return;
344
345 if (old_cfg) {
346 free_irq_2_pin(old_cfg, cfg);
347 free_irq_cfg(old_cfg);
348 old_desc->chip_data = NULL;
349 }
350}
351
d733e00d
IM		 352static void
353set_extra_move_desc(struct irq_desc *desc, const struct cpumask *mask)
48a1b10a
YL		 354{
355 struct irq_cfg *cfg = desc->chip_data;
356
357 if (!cfg->move_in_progress) {
358 /* it means that domain is not changed */
d733e00d 359 if (!cpumask_intersects(&desc->affinity, mask))
48a1b10a
YL		 360 cfg->move_desc_pending = 1;
361 }
0f978f45 362}
48a1b10a
YL		 363#endif
364
0b8f1efa
YL		 365#else
366static struct irq_cfg *irq_cfg(unsigned int irq)
367{
368 return irq < nr_irqs ? irq_cfgx + irq : NULL;
0f978f45 369}
1da177e4 370
0b8f1efa
YL		 371#endif
372
48a1b10a 373#ifndef CONFIG_NUMA_MIGRATE_IRQ_DESC
e7986739
MT		 374static inline void
375set_extra_move_desc(struct irq_desc *desc, const struct cpumask *mask)
3145e941
YL		 376{
377}
48a1b10a 378#endif
1da177e4 379
130fe05d
LT		 380struct io_apic {
381 unsigned int index;
382 unsigned int unused[3];
383 unsigned int data;
384};
385
386static __attribute_const__ struct io_apic __iomem *io_apic_base(int idx)
387{
388 return (void __iomem *) __fix_to_virt(FIX_IO_APIC_BASE_0 + idx)
ec2cd0a2 389 + (mp_ioapics[idx].mp_apicaddr & ~PAGE_MASK);
130fe05d
LT		 390}
391
392static inline unsigned int io_apic_read(unsigned int apic, unsigned int reg)
393{
394 struct io_apic __iomem *io_apic = io_apic_base(apic);
395 writel(reg, &io_apic->index);
396 return readl(&io_apic->data);
397}
398
399static inline void io_apic_write(unsigned int apic, unsigned int reg, unsigned int value)
400{
401 struct io_apic __iomem *io_apic = io_apic_base(apic);
402 writel(reg, &io_apic->index);
403 writel(value, &io_apic->data);
404}
405
406/*
407 * Re-write a value: to be used for read-modify-write
408 * cycles where the read already set up the index register.
409 *
410 * Older SiS APIC requires we rewrite the index register
411 */
412static inline void io_apic_modify(unsigned int apic, unsigned int reg, unsigned int value)
413{
54168ed7 414 struct io_apic __iomem *io_apic = io_apic_base(apic);
d6c88a50
TG		 415
416 if (sis_apic_bug)
417 writel(reg, &io_apic->index);
130fe05d
LT		 418 writel(value, &io_apic->data);
419}
420
3145e941 421static bool io_apic_level_ack_pending(struct irq_cfg *cfg)
047c8fdb
YL		 422{
423 struct irq_pin_list *entry;
424 unsigned long flags;
047c8fdb
YL		 425
426 spin_lock_irqsave(&ioapic_lock, flags);
427 entry = cfg->irq_2_pin;
428 for (;;) {
429 unsigned int reg;
430 int pin;
431
432 if (!entry)
433 break;
434 pin = entry->pin;
435 reg = io_apic_read(entry->apic, 0x10 + pin*2);
436 /* Is the remote IRR bit set? */
437 if (reg & IO_APIC_REDIR_REMOTE_IRR) {
438 spin_unlock_irqrestore(&ioapic_lock, flags);
439 return true;
440 }
441 if (!entry->next)
442 break;
443 entry = entry->next;
444 }
445 spin_unlock_irqrestore(&ioapic_lock, flags);
446
447 return false;
448}
047c8fdb 449
cf4c6a2f
AK		 450union entry_union {
451 struct { u32 w1, w2; };
452 struct IO_APIC_route_entry entry;
453};
454
455static struct IO_APIC_route_entry ioapic_read_entry(int apic, int pin)
456{
457 union entry_union eu;
458 unsigned long flags;
459 spin_lock_irqsave(&ioapic_lock, flags);
460 eu.w1 = io_apic_read(apic, 0x10 + 2 * pin);
461 eu.w2 = io_apic_read(apic, 0x11 + 2 * pin);
462 spin_unlock_irqrestore(&ioapic_lock, flags);
463 return eu.entry;
464}
465
f9dadfa7
LT		 466/*
467 * When we write a new IO APIC routing entry, we need to write the high
468 * word first! If the mask bit in the low word is clear, we will enable
469 * the interrupt, and we need to make sure the entry is fully populated
470 * before that happens.
471 */
d15512f4
AK		 472static void
473__ioapic_write_entry(int apic, int pin, struct IO_APIC_route_entry e)
cf4c6a2f 474{
cf4c6a2f
AK		 475 union entry_union eu;
476 eu.entry = e;
f9dadfa7
LT		 477 io_apic_write(apic, 0x11 + 2*pin, eu.w2);
478 io_apic_write(apic, 0x10 + 2*pin, eu.w1);
d15512f4
AK		 479}
480
481static void ioapic_write_entry(int apic, int pin, struct IO_APIC_route_entry e)
482{
483 unsigned long flags;
484 spin_lock_irqsave(&ioapic_lock, flags);
485 __ioapic_write_entry(apic, pin, e);
f9dadfa7
LT		 486 spin_unlock_irqrestore(&ioapic_lock, flags);
487}
488
489/*
490 * When we mask an IO APIC routing entry, we need to write the low
491 * word first, in order to set the mask bit before we change the
492 * high bits!
493 */
494static void ioapic_mask_entry(int apic, int pin)
495{
496 unsigned long flags;
497 union entry_union eu = { .entry.mask = 1 };
498
cf4c6a2f
AK		 499 spin_lock_irqsave(&ioapic_lock, flags);
500 io_apic_write(apic, 0x10 + 2*pin, eu.w1);
501 io_apic_write(apic, 0x11 + 2*pin, eu.w2);
502 spin_unlock_irqrestore(&ioapic_lock, flags);
503}
504
497c9a19 505#ifdef CONFIG_SMP
22f65d31
MT		 506static void send_cleanup_vector(struct irq_cfg *cfg)
507{
508 cpumask_var_t cleanup_mask;
509
510 if (unlikely(!alloc_cpumask_var(&cleanup_mask, GFP_ATOMIC))) {
511 unsigned int i;
512 cfg->move_cleanup_count = 0;
513 for_each_cpu_and(i, cfg->old_domain, cpu_online_mask)
514 cfg->move_cleanup_count++;
515 for_each_cpu_and(i, cfg->old_domain, cpu_online_mask)
516 send_IPI_mask(cpumask_of(i), IRQ_MOVE_CLEANUP_VECTOR);
517 } else {
518 cpumask_and(cleanup_mask, cfg->old_domain, cpu_online_mask);
519 cfg->move_cleanup_count = cpumask_weight(cleanup_mask);
520 send_IPI_mask(cleanup_mask, IRQ_MOVE_CLEANUP_VECTOR);
521 free_cpumask_var(cleanup_mask);
522 }
523 cfg->move_in_progress = 0;
524}
525
3145e941 526static void __target_IO_APIC_irq(unsigned int irq, unsigned int dest, struct irq_cfg *cfg)
497c9a19
YL		 527{
528 int apic, pin;
497c9a19 529 struct irq_pin_list *entry;
3145e941 530 u8 vector = cfg->vector;
497c9a19 531
497c9a19
YL		 532 entry = cfg->irq_2_pin;
533 for (;;) {
534 unsigned int reg;
535
536 if (!entry)
537 break;
538
539 apic = entry->apic;
540 pin = entry->pin;
54168ed7
IM		 541#ifdef CONFIG_INTR_REMAP
542 /*
543 * With interrupt-remapping, destination information comes
544 * from interrupt-remapping table entry.
545 */
546 if (!irq_remapped(irq))
547 io_apic_write(apic, 0x11 + pin*2, dest);
548#else
497c9a19 549 io_apic_write(apic, 0x11 + pin*2, dest);
54168ed7 550#endif
497c9a19
YL		 551 reg = io_apic_read(apic, 0x10 + pin*2);
552 reg &= ~IO_APIC_REDIR_VECTOR_MASK;
553 reg |= vector;
54168ed7 554 io_apic_modify(apic, 0x10 + pin*2, reg);
497c9a19
YL		 555 if (!entry->next)
556 break;
557 entry = entry->next;
558 }
559}
efa2559f 560
e7986739
MT		 561static int
562assign_irq_vector(int irq, struct irq_cfg *cfg, const struct cpumask *mask);
efa2559f 563
22f65d31
MT		 564/*
565 * Either sets desc->affinity to a valid value, and returns cpu_mask_to_apicid
566 * of that, or returns BAD_APICID and leaves desc->affinity untouched.
567 */
568static unsigned int
569set_desc_affinity(struct irq_desc *desc, const struct cpumask *mask)
497c9a19
YL		 570{
571 struct irq_cfg *cfg;
3145e941 572 unsigned int irq;
497c9a19 573
0de26520 574 if (!cpumask_intersects(mask, cpu_online_mask))
22f65d31 575 return BAD_APICID;
497c9a19 576
3145e941
YL		 577 irq = desc->irq;
578 cfg = desc->chip_data;
579 if (assign_irq_vector(irq, cfg, mask))
22f65d31 580 return BAD_APICID;
497c9a19 581
22f65d31 582 cpumask_and(&desc->affinity, cfg->domain, mask);
3145e941 583 set_extra_move_desc(desc, mask);
22f65d31
MT		 584 return cpu_mask_to_apicid_and(&desc->affinity, cpu_online_mask);
585}
3145e941 586
22f65d31
MT		 587static void
588set_ioapic_affinity_irq_desc(struct irq_desc *desc, const struct cpumask *mask)
497c9a19
YL		 589{
590 struct irq_cfg *cfg;
591 unsigned long flags;
592 unsigned int dest;
22f65d31 593 unsigned int irq;
497c9a19 594
22f65d31
MT		 595 irq = desc->irq;
596 cfg = desc->chip_data;
497c9a19 597
497c9a19 598 spin_lock_irqsave(&ioapic_lock, flags);
22f65d31
MT		 599 dest = set_desc_affinity(desc, mask);
600 if (dest != BAD_APICID) {
601 /* Only the high 8 bits are valid. */
602 dest = SET_APIC_LOGICAL_ID(dest);
603 __target_IO_APIC_irq(irq, dest, cfg);
604 }
497c9a19
YL		 605 spin_unlock_irqrestore(&ioapic_lock, flags);
606}
497c9a19 607
22f65d31
MT		 608static void
609set_ioapic_affinity_irq(unsigned int irq, const struct cpumask *mask)
3145e941
YL		 610{
611 struct irq_desc *desc;
497c9a19 612
54168ed7 613 desc = irq_to_desc(irq);
3145e941
YL		 614
615 set_ioapic_affinity_irq_desc(desc, mask);
497c9a19 616}
497c9a19
YL		 617#endif /* CONFIG_SMP */
618
1da177e4
LT		 619/*
620 * The common case is 1:1 IRQ<->pin mappings. Sometimes there are
621 * shared ISA-space IRQs, so we have to support them. We are super
622 * fast in the common case, and fast for shared ISA-space IRQs.
623 */
3145e941 624static void add_pin_to_irq_cpu(struct irq_cfg *cfg, int cpu, int apic, int pin)
1da177e4 625{
0f978f45
YL		 626 struct irq_pin_list *entry;
627
0f978f45
YL		 628 entry = cfg->irq_2_pin;
629 if (!entry) {
0b8f1efa
YL		 630 entry = get_one_free_irq_2_pin(cpu);
631 if (!entry) {
632 printk(KERN_ERR "can not alloc irq_2_pin to add %d - %d\n",
633 apic, pin);
634 return;
635 }
0f978f45
YL		 636 cfg->irq_2_pin = entry;
637 entry->apic = apic;
638 entry->pin = pin;
0f978f45
YL		 639 return;
640 }
1da177e4 641
0f978f45
YL		 642 while (entry->next) {
643 /* not again, please */
644 if (entry->apic == apic && entry->pin == pin)
645 return;
1da177e4 646
0f978f45 647 entry = entry->next;
1da177e4 648 }
0f978f45 649
0b8f1efa 650 entry->next = get_one_free_irq_2_pin(cpu);
0f978f45 651 entry = entry->next;
1da177e4
LT		 652 entry->apic = apic;
653 entry->pin = pin;
654}
655
656/*
657 * Reroute an IRQ to a different pin.
658 */
3145e941 659static void __init replace_pin_at_irq_cpu(struct irq_cfg *cfg, int cpu,
1da177e4
LT		 660 int oldapic, int oldpin,
661 int newapic, int newpin)
662{
0f978f45
YL		 663 struct irq_pin_list *entry = cfg->irq_2_pin;
664 int replaced = 0;
1da177e4 665
0f978f45 666 while (entry) {
1da177e4
LT		 667 if (entry->apic == oldapic && entry->pin == oldpin) {
668 entry->apic = newapic;
669 entry->pin = newpin;
0f978f45
YL		 670 replaced = 1;
671 /* every one is different, right? */
1da177e4 672 break;
0f978f45
YL		 673 }
674 entry = entry->next;
1da177e4 675 }
0f978f45
YL		 676
677 /* why? call replace before add? */
678 if (!replaced)
3145e941 679 add_pin_to_irq_cpu(cfg, cpu, newapic, newpin);
1da177e4
LT		 680}
681
3145e941 682static inline void io_apic_modify_irq(struct irq_cfg *cfg,
87783be4
CG		 683 int mask_and, int mask_or,
684 void (*final)(struct irq_pin_list *entry))
685{
686 int pin;
87783be4 687 struct irq_pin_list *entry;
047c8fdb 688
87783be4
CG		 689 for (entry = cfg->irq_2_pin; entry != NULL; entry = entry->next) {
690 unsigned int reg;
691 pin = entry->pin;
692 reg = io_apic_read(entry->apic, 0x10 + pin * 2);
693 reg &= mask_and;
694 reg |= mask_or;
695 io_apic_modify(entry->apic, 0x10 + pin * 2, reg);
696 if (final)
697 final(entry);
698 }
699}
047c8fdb 700
3145e941 701static void __unmask_IO_APIC_irq(struct irq_cfg *cfg)
87783be4 702{
3145e941 703 io_apic_modify_irq(cfg, ~IO_APIC_REDIR_MASKED, 0, NULL);
87783be4 704}
047c8fdb 705
4e738e2f 706#ifdef CONFIG_X86_64
7f3e632f 707static void io_apic_sync(struct irq_pin_list *entry)
1da177e4 708{
87783be4
CG		 709 /*
710 * Synchronize the IO-APIC and the CPU by doing
711 * a dummy read from the IO-APIC
712 */
713 struct io_apic __iomem *io_apic;
714 io_apic = io_apic_base(entry->apic);
4e738e2f 715 readl(&io_apic->data);
1da177e4
LT		 716}
717
3145e941 718static void __mask_IO_APIC_irq(struct irq_cfg *cfg)
87783be4 719{
3145e941 720 io_apic_modify_irq(cfg, ~0, IO_APIC_REDIR_MASKED, &io_apic_sync);
87783be4
CG		 721}
722#else /* CONFIG_X86_32 */
3145e941 723static void __mask_IO_APIC_irq(struct irq_cfg *cfg)
87783be4 724{
3145e941 725 io_apic_modify_irq(cfg, ~0, IO_APIC_REDIR_MASKED, NULL);
87783be4 726}
1da177e4 727
3145e941 728static void __mask_and_edge_IO_APIC_irq(struct irq_cfg *cfg)
87783be4 729{
3145e941 730 io_apic_modify_irq(cfg, ~IO_APIC_REDIR_LEVEL_TRIGGER,
87783be4
CG		 731 IO_APIC_REDIR_MASKED, NULL);
732}
1da177e4 733
3145e941 734static void __unmask_and_level_IO_APIC_irq(struct irq_cfg *cfg)
87783be4 735{
3145e941 736 io_apic_modify_irq(cfg, ~IO_APIC_REDIR_MASKED,
87783be4
CG		 737 IO_APIC_REDIR_LEVEL_TRIGGER, NULL);
738}
739#endif /* CONFIG_X86_32 */
047c8fdb 740
3145e941 741static void mask_IO_APIC_irq_desc(struct irq_desc *desc)
1da177e4 742{
3145e941 743 struct irq_cfg *cfg = desc->chip_data;
1da177e4
LT		 744 unsigned long flags;
745
3145e941
YL		 746 BUG_ON(!cfg);
747
1da177e4 748 spin_lock_irqsave(&ioapic_lock, flags);
3145e941 749 __mask_IO_APIC_irq(cfg);
1da177e4
LT		 750 spin_unlock_irqrestore(&ioapic_lock, flags);
751}
752
3145e941 753static void unmask_IO_APIC_irq_desc(struct irq_desc *desc)
1da177e4 754{
3145e941 755 struct irq_cfg *cfg = desc->chip_data;
1da177e4
LT		 756 unsigned long flags;
757
758 spin_lock_irqsave(&ioapic_lock, flags);
3145e941 759 __unmask_IO_APIC_irq(cfg);
1da177e4
LT		 760 spin_unlock_irqrestore(&ioapic_lock, flags);
761}
762
3145e941
YL		 763static void mask_IO_APIC_irq(unsigned int irq)
764{
765 struct irq_desc *desc = irq_to_desc(irq);
766
767 mask_IO_APIC_irq_desc(desc);
768}
769static void unmask_IO_APIC_irq(unsigned int irq)
770{
771 struct irq_desc *desc = irq_to_desc(irq);
772
773 unmask_IO_APIC_irq_desc(desc);
774}
775
1da177e4
LT		 776static void clear_IO_APIC_pin(unsigned int apic, unsigned int pin)
777{
778 struct IO_APIC_route_entry entry;
36062448 779
1da177e4 780 /* Check delivery_mode to be sure we're not clearing an SMI pin */
cf4c6a2f 781 entry = ioapic_read_entry(apic, pin);
1da177e4
LT		 782 if (entry.delivery_mode == dest_SMI)
783 return;
1da177e4
LT		 784 /*
785 * Disable it in the IO-APIC irq-routing table:
786 */
f9dadfa7 787 ioapic_mask_entry(apic, pin);
1da177e4
LT		 788}
789
54168ed7 790static void clear_IO_APIC (void)
1da177e4
LT		 791{
792 int apic, pin;
793
794 for (apic = 0; apic < nr_ioapics; apic++)
795 for (pin = 0; pin < nr_ioapic_registers[apic]; pin++)
796 clear_IO_APIC_pin(apic, pin);
797}
798
54168ed7 799#if !defined(CONFIG_SMP) && defined(CONFIG_X86_32)
75604d7f 800void send_IPI_self(int vector)
1da177e4
LT		 801{
802 unsigned int cfg;
803
804 /*
805 * Wait for idle.
806 */
807 apic_wait_icr_idle();
808 cfg = APIC_DM_FIXED | APIC_DEST_SELF | vector | APIC_DEST_LOGICAL;
809 /*
810 * Send the IPI. The write to APIC_ICR fires this off.
811 */
593f4a78 812 apic_write(APIC_ICR, cfg);
1da177e4 813}
54168ed7 814#endif /* !CONFIG_SMP && CONFIG_X86_32*/
1da177e4 815
54168ed7 816#ifdef CONFIG_X86_32
1da177e4
LT		 817/*
818 * support for broken MP BIOSs, enables hand-redirection of PIRQ0-7 to
819 * specific CPU-side IRQs.
820 */
821
822#define MAX_PIRQS 8
823static int pirq_entries [MAX_PIRQS];
824static int pirqs_enabled;
1da177e4 825
1da177e4
LT		 826static int __init ioapic_pirq_setup(char *str)
827{
828 int i, max;
829 int ints[MAX_PIRQS+1];
830
831 get_options(str, ARRAY_SIZE(ints), ints);
832
833 for (i = 0; i < MAX_PIRQS; i++)
834 pirq_entries[i] = -1;
835
836 pirqs_enabled = 1;
837 apic_printk(APIC_VERBOSE, KERN_INFO
838 "PIRQ redirection, working around broken MP-BIOS.\n");
839 max = MAX_PIRQS;
840 if (ints[0] < MAX_PIRQS)
841 max = ints[0];
842
843 for (i = 0; i < max; i++) {
844 apic_printk(APIC_VERBOSE, KERN_DEBUG
845 "... PIRQ%d -> IRQ %d\n", i, ints[i+1]);
846 /*
847 * PIRQs are mapped upside down, usually.
848 */
849 pirq_entries[MAX_PIRQS-i-1] = ints[i+1];
850 }
851 return 1;
852}
853
854__setup("pirq=", ioapic_pirq_setup);
54168ed7
IM		 855#endif /* CONFIG_X86_32 */
856
857#ifdef CONFIG_INTR_REMAP
858/* I/O APIC RTE contents at the OS boot up */
859static struct IO_APIC_route_entry *early_ioapic_entries[MAX_IO_APICS];
860
861/*
862 * Saves and masks all the unmasked IO-APIC RTE's
863 */
864int save_mask_IO_APIC_setup(void)
865{
866 union IO_APIC_reg_01 reg_01;
867 unsigned long flags;
868 int apic, pin;
869
870 /*
871 * The number of IO-APIC IRQ registers (== #pins):
872 */
873 for (apic = 0; apic < nr_ioapics; apic++) {
874 spin_lock_irqsave(&ioapic_lock, flags);
875 reg_01.raw = io_apic_read(apic, 1);
876 spin_unlock_irqrestore(&ioapic_lock, flags);
877 nr_ioapic_registers[apic] = reg_01.bits.entries+1;
878 }
879
880 for (apic = 0; apic < nr_ioapics; apic++) {
881 early_ioapic_entries[apic] =
882 kzalloc(sizeof(struct IO_APIC_route_entry) *
883 nr_ioapic_registers[apic], GFP_KERNEL);
884 if (!early_ioapic_entries[apic])
5ffa4eb2 885 goto nomem;
54168ed7
IM		 886 }
887
888 for (apic = 0; apic < nr_ioapics; apic++)
889 for (pin = 0; pin < nr_ioapic_registers[apic]; pin++) {
890 struct IO_APIC_route_entry entry;
891
892 entry = early_ioapic_entries[apic][pin] =
893 ioapic_read_entry(apic, pin);
894 if (!entry.mask) {
895 entry.mask = 1;
896 ioapic_write_entry(apic, pin, entry);
897 }
898 }
5ffa4eb2 899
54168ed7 900 return 0;
5ffa4eb2
CG		 901
902nomem:
c1370b49
CG		 903 while (apic >= 0)
904 kfree(early_ioapic_entries[apic--]);
5ffa4eb2
CG		 905 memset(early_ioapic_entries, 0,
906 ARRAY_SIZE(early_ioapic_entries));
907
908 return -ENOMEM;
54168ed7
IM		 909}
910
911void restore_IO_APIC_setup(void)
912{
913 int apic, pin;
914
5ffa4eb2
CG		 915 for (apic = 0; apic < nr_ioapics; apic++) {
916 if (!early_ioapic_entries[apic])
917 break;
54168ed7
IM		 918 for (pin = 0; pin < nr_ioapic_registers[apic]; pin++)
919 ioapic_write_entry(apic, pin,
920 early_ioapic_entries[apic][pin]);
5ffa4eb2
CG		 921 kfree(early_ioapic_entries[apic]);
922 early_ioapic_entries[apic] = NULL;
923 }
54168ed7
IM		 924}
925
926void reinit_intr_remapped_IO_APIC(int intr_remapping)
927{
928 /*
929 * for now plain restore of previous settings.
930 * TBD: In the case of OS enabling interrupt-remapping,
931 * IO-APIC RTE's need to be setup to point to interrupt-remapping
932 * table entries. for now, do a plain restore, and wait for
933 * the setup_IO_APIC_irqs() to do proper initialization.
934 */
935 restore_IO_APIC_setup();
936}
937#endif
1da177e4
LT		 938
939/*
940 * Find the IRQ entry number of a certain pin.
941 */
942static int find_irq_entry(int apic, int pin, int type)
943{
944 int i;
945
946 for (i = 0; i < mp_irq_entries; i++)
2fddb6e2
AS		 947 if (mp_irqs[i].mp_irqtype == type &&
948 (mp_irqs[i].mp_dstapic == mp_ioapics[apic].mp_apicid ||
949 mp_irqs[i].mp_dstapic == MP_APIC_ALL) &&
950 mp_irqs[i].mp_dstirq == pin)
1da177e4
LT		 951 return i;
952
953 return -1;
954}
955
956/*
957 * Find the pin to which IRQ[irq] (ISA) is connected
958 */
fcfd636a 959static int __init find_isa_irq_pin(int irq, int type)
1da177e4
LT		 960{
961 int i;
962
963 for (i = 0; i < mp_irq_entries; i++) {
2fddb6e2 964 int lbus = mp_irqs[i].mp_srcbus;
1da177e4 965
d27e2b8e 966 if (test_bit(lbus, mp_bus_not_pci) &&
2fddb6e2
AS		 967 (mp_irqs[i].mp_irqtype == type) &&
968 (mp_irqs[i].mp_srcbusirq == irq))
1da177e4 969
2fddb6e2 970 return mp_irqs[i].mp_dstirq;
1da177e4
LT		 971 }
972 return -1;
973}
974
fcfd636a
EB		 975static int __init find_isa_irq_apic(int irq, int type)
976{
977 int i;
978
979 for (i = 0; i < mp_irq_entries; i++) {
2fddb6e2 980 int lbus = mp_irqs[i].mp_srcbus;
fcfd636a 981
73b2961b 982 if (test_bit(lbus, mp_bus_not_pci) &&
2fddb6e2
AS		 983 (mp_irqs[i].mp_irqtype == type) &&
984 (mp_irqs[i].mp_srcbusirq == irq))
fcfd636a
EB		 985 break;
986 }
987 if (i < mp_irq_entries) {
988 int apic;
54168ed7 989 for(apic = 0; apic < nr_ioapics; apic++) {
2fddb6e2 990 if (mp_ioapics[apic].mp_apicid == mp_irqs[i].mp_dstapic)
fcfd636a
EB		 991 return apic;
992 }
993 }
994
995 return -1;
996}
997
1da177e4
LT		 998/*
999 * Find a specific PCI IRQ entry.
1000 * Not an __init, possibly needed by modules
1001 */
1002static int pin_2_irq(int idx, int apic, int pin);
1003
1004int IO_APIC_get_PCI_irq_vector(int bus, int slot, int pin)
1005{
1006 int apic, i, best_guess = -1;
1007
54168ed7
IM		 1008 apic_printk(APIC_DEBUG, "querying PCI -> IRQ mapping bus:%d, slot:%d, pin:%d.\n",
1009 bus, slot, pin);
ce6444d3 1010 if (test_bit(bus, mp_bus_not_pci)) {
54168ed7 1011 apic_printk(APIC_VERBOSE, "PCI BIOS passed nonexistent PCI bus %d!\n", bus);
1da177e4
LT		 1012 return -1;
1013 }
1014 for (i = 0; i < mp_irq_entries; i++) {
2fddb6e2 1015 int lbus = mp_irqs[i].mp_srcbus;
1da177e4
LT		 1016
1017 for (apic = 0; apic < nr_ioapics; apic++)
2fddb6e2
AS		 1018 if (mp_ioapics[apic].mp_apicid == mp_irqs[i].mp_dstapic ||
1019 mp_irqs[i].mp_dstapic == MP_APIC_ALL)
1da177e4
LT		 1020 break;
1021
47cab822 1022 if (!test_bit(lbus, mp_bus_not_pci) &&
2fddb6e2 1023 !mp_irqs[i].mp_irqtype &&
1da177e4 1024 (bus == lbus) &&
2fddb6e2 1025 (slot == ((mp_irqs[i].mp_srcbusirq >> 2) & 0x1f))) {
54168ed7 1026 int irq = pin_2_irq(i,apic,mp_irqs[i].mp_dstirq);
1da177e4
LT		 1027
1028 if (!(apic || IO_APIC_IRQ(irq)))
1029 continue;
1030
2fddb6e2 1031 if (pin == (mp_irqs[i].mp_srcbusirq & 3))
1da177e4
LT		 1032 return irq;
1033 /*
1034 * Use the first all-but-pin matching entry as a
1035 * best-guess fuzzy result for broken mptables.
1036 */
1037 if (best_guess < 0)
1038 best_guess = irq;
1039 }
1040 }
1041 return best_guess;
1042}
54168ed7 1043
129f6946 1044EXPORT_SYMBOL(IO_APIC_get_PCI_irq_vector);
1da177e4 1045
c0a282c2 1046#if defined(CONFIG_EISA) || defined(CONFIG_MCA)
1da177e4
LT		 1047/*
1048 * EISA Edge/Level control register, ELCR
1049 */
1050static int EISA_ELCR(unsigned int irq)
1051{
99d093d1 1052 if (irq < NR_IRQS_LEGACY) {
1da177e4
LT		 1053 unsigned int port = 0x4d0 + (irq >> 3);
1054 return (inb(port) >> (irq & 7)) & 1;
1055 }
1056 apic_printk(APIC_VERBOSE, KERN_INFO
1057 "Broken MPtable reports ISA irq %d\n", irq);
1058 return 0;
1059}
54168ed7 1060
c0a282c2 1061#endif
1da177e4 1062
6728801d
AS		 1063/* ISA interrupts are always polarity zero edge triggered,
1064 * when listed as conforming in the MP table. */
1065
1066#define default_ISA_trigger(idx) (0)
1067#define default_ISA_polarity(idx) (0)
1068
1da177e4
LT		 1069/* EISA interrupts are always polarity zero and can be edge or level
1070 * trigger depending on the ELCR value. If an interrupt is listed as
1071 * EISA conforming in the MP table, that means its trigger type must
1072 * be read in from the ELCR */
1073
2fddb6e2 1074#define default_EISA_trigger(idx) (EISA_ELCR(mp_irqs[idx].mp_srcbusirq))
6728801d 1075#define default_EISA_polarity(idx) default_ISA_polarity(idx)
1da177e4
LT		 1076
1077/* PCI interrupts are always polarity one level triggered,
1078 * when listed as conforming in the MP table. */
1079
1080#define default_PCI_trigger(idx) (1)
1081#define default_PCI_polarity(idx) (1)
1082
1083/* MCA interrupts are always polarity zero level triggered,
1084 * when listed as conforming in the MP table. */
1085
1086#define default_MCA_trigger(idx) (1)
6728801d 1087#define default_MCA_polarity(idx) default_ISA_polarity(idx)
1da177e4 1088
61fd47e0 1089static int MPBIOS_polarity(int idx)
1da177e4 1090{
2fddb6e2 1091 int bus = mp_irqs[idx].mp_srcbus;
1da177e4
LT		 1092 int polarity;
1093
1094 /*
1095 * Determine IRQ line polarity (high active or low active):
1096 */
54168ed7 1097 switch (mp_irqs[idx].mp_irqflag & 3)
36062448 1098 {
54168ed7
IM		 1099 case 0: /* conforms, ie. bus-type dependent polarity */
1100 if (test_bit(bus, mp_bus_not_pci))
1101 polarity = default_ISA_polarity(idx);
1102 else
1103 polarity = default_PCI_polarity(idx);
1104 break;
1105 case 1: /* high active */
1106 {
1107 polarity = 0;
1108 break;
1109 }
1110 case 2: /* reserved */
1111 {
1112 printk(KERN_WARNING "broken BIOS!!\n");
1113 polarity = 1;
1114 break;
1115 }
1116 case 3: /* low active */
1117 {
1118 polarity = 1;
1119 break;
1120 }
1121 default: /* invalid */
1122 {
1123 printk(KERN_WARNING "broken BIOS!!\n");
1124 polarity = 1;
1125 break;
1126 }
1da177e4
LT		 1127 }
1128 return polarity;
1129}
1130
1131static int MPBIOS_trigger(int idx)
1132{
2fddb6e2 1133 int bus = mp_irqs[idx].mp_srcbus;
1da177e4
LT		 1134 int trigger;
1135
1136 /*
1137 * Determine IRQ trigger mode (edge or level sensitive):
1138 */
54168ed7 1139 switch ((mp_irqs[idx].mp_irqflag>>2) & 3)
1da177e4 1140 {
54168ed7
IM		 1141 case 0: /* conforms, ie. bus-type dependent */
1142 if (test_bit(bus, mp_bus_not_pci))
1143 trigger = default_ISA_trigger(idx);
1144 else
1145 trigger = default_PCI_trigger(idx);
c0a282c2 1146#if defined(CONFIG_EISA) || defined(CONFIG_MCA)
54168ed7
IM		 1147 switch (mp_bus_id_to_type[bus]) {
1148 case MP_BUS_ISA: /* ISA pin */
1149 {
1150 /* set before the switch */
1151 break;
1152 }
1153 case MP_BUS_EISA: /* EISA pin */
1154 {
1155 trigger = default_EISA_trigger(idx);
1156 break;
1157 }
1158 case MP_BUS_PCI: /* PCI pin */
1159 {
1160 /* set before the switch */
1161 break;
1162 }
1163 case MP_BUS_MCA: /* MCA pin */
1164 {
1165 trigger = default_MCA_trigger(idx);
1166 break;
1167 }
1168 default:
1169 {
1170 printk(KERN_WARNING "broken BIOS!!\n");
1171 trigger = 1;
1172 break;
1173 }
1174 }
1175#endif
1da177e4 1176 break;
54168ed7 1177 case 1: /* edge */
1da177e4 1178 {
54168ed7 1179 trigger = 0;
1da177e4
LT		 1180 break;
1181 }
54168ed7 1182 case 2: /* reserved */
1da177e4 1183 {
54168ed7
IM		 1184 printk(KERN_WARNING "broken BIOS!!\n");
1185 trigger = 1;
1da177e4
LT		 1186 break;
1187 }
54168ed7 1188 case 3: /* level */
1da177e4 1189 {
54168ed7 1190 trigger = 1;
1da177e4
LT		 1191 break;
1192 }
54168ed7 1193 default: /* invalid */
1da177e4
LT		 1194 {
1195 printk(KERN_WARNING "broken BIOS!!\n");
54168ed7 1196 trigger = 0;
1da177e4
LT		 1197 break;
1198 }
1199 }
1200 return trigger;
1201}
1202
1203static inline int irq_polarity(int idx)
1204{
1205 return MPBIOS_polarity(idx);
1206}
1207
1208static inline int irq_trigger(int idx)
1209{
1210 return MPBIOS_trigger(idx);
1211}
1212
efa2559f 1213int (*ioapic_renumber_irq)(int ioapic, int irq);
1da177e4
LT		 1214static int pin_2_irq(int idx, int apic, int pin)
1215{
1216 int irq, i;
2fddb6e2 1217 int bus = mp_irqs[idx].mp_srcbus;
1da177e4
LT		 1218
1219 /*
1220 * Debugging check, we are in big trouble if this message pops up!
1221 */
2fddb6e2 1222 if (mp_irqs[idx].mp_dstirq != pin)
1da177e4
LT		 1223 printk(KERN_ERR "broken BIOS or MPTABLE parser, ayiee!!\n");
1224
54168ed7 1225 if (test_bit(bus, mp_bus_not_pci)) {
2fddb6e2 1226 irq = mp_irqs[idx].mp_srcbusirq;
54168ed7 1227 } else {
643befed
AS		 1228 /*
1229 * PCI IRQs are mapped in order
1230 */
1231 i = irq = 0;
1232 while (i < apic)
1233 irq += nr_ioapic_registers[i++];
1234 irq += pin;
d6c88a50 1235 /*
54168ed7
IM		 1236 * For MPS mode, so far only needed by ES7000 platform
1237 */
d6c88a50
TG		 1238 if (ioapic_renumber_irq)
1239 irq = ioapic_renumber_irq(apic, irq);
1da177e4
LT		 1240 }
1241
54168ed7 1242#ifdef CONFIG_X86_32
1da177e4
LT		 1243 /*
1244 * PCI IRQ command line redirection. Yes, limits are hardcoded.
1245 */
1246 if ((pin >= 16) && (pin <= 23)) {
1247 if (pirq_entries[pin-16] != -1) {
1248 if (!pirq_entries[pin-16]) {
1249 apic_printk(APIC_VERBOSE, KERN_DEBUG
1250 "disabling PIRQ%d\n", pin-16);
1251 } else {
1252 irq = pirq_entries[pin-16];
1253 apic_printk(APIC_VERBOSE, KERN_DEBUG
1254 "using PIRQ%d -> IRQ %d\n",
1255 pin-16, irq);
1256 }
1257 }
1258 }
54168ed7
IM		 1259#endif
1260
1da177e4
LT		 1261 return irq;
1262}
1263
497c9a19
YL		 1264void lock_vector_lock(void)
1265{
1266 /* Used to the online set of cpus does not change
1267 * during assign_irq_vector.
1268 */
1269 spin_lock(&vector_lock);
1270}
1da177e4 1271
497c9a19 1272void unlock_vector_lock(void)
1da177e4 1273{
497c9a19
YL		 1274 spin_unlock(&vector_lock);
1275}
1da177e4 1276
e7986739
MT		 1277static int
1278__assign_irq_vector(int irq, struct irq_cfg *cfg, const struct cpumask *mask)
497c9a19 1279{
047c8fdb
YL		 1280 /*
1281 * NOTE! The local APIC isn't very good at handling
1282 * multiple interrupts at the same interrupt level.
1283 * As the interrupt level is determined by taking the
1284 * vector number and shifting that right by 4, we
1285 * want to spread these out a bit so that they don't
1286 * all fall in the same interrupt level.
1287 *
1288 * Also, we've got to be careful not to trash gate
1289 * 0x80, because int 0x80 is hm, kind of importantish. ;)
1290 */
54168ed7
IM		 1291 static int current_vector = FIRST_DEVICE_VECTOR, current_offset = 0;
1292 unsigned int old_vector;
22f65d31
MT		 1293 int cpu, err;
1294 cpumask_var_t tmp_mask;
ace80ab7 1295
54168ed7
IM		 1296 if ((cfg->move_in_progress) || cfg->move_cleanup_count)
1297 return -EBUSY;
0a1ad60d 1298
22f65d31
MT		 1299 if (!alloc_cpumask_var(&tmp_mask, GFP_ATOMIC))
1300 return -ENOMEM;
ace80ab7 1301
54168ed7
IM		 1302 old_vector = cfg->vector;
1303 if (old_vector) {
22f65d31
MT		 1304 cpumask_and(tmp_mask, mask, cpu_online_mask);
1305 cpumask_and(tmp_mask, cfg->domain, tmp_mask);
1306 if (!cpumask_empty(tmp_mask)) {
1307 free_cpumask_var(tmp_mask);
54168ed7 1308 return 0;
22f65d31 1309 }
54168ed7 1310 }
497c9a19 1311
e7986739 1312 /* Only try and allocate irqs on cpus that are present */
22f65d31
MT		 1313 err = -ENOSPC;
1314 for_each_cpu_and(cpu, mask, cpu_online_mask) {
54168ed7
IM		 1315 int new_cpu;
1316 int vector, offset;
497c9a19 1317
22f65d31 1318 vector_allocation_domain(cpu, tmp_mask);
497c9a19 1319
54168ed7
IM		 1320 vector = current_vector;
1321 offset = current_offset;
497c9a19 1322next:
54168ed7
IM		 1323 vector += 8;
1324 if (vector >= first_system_vector) {
e7986739 1325 /* If out of vectors on large boxen, must share them. */
54168ed7
IM		 1326 offset = (offset + 1) % 8;
1327 vector = FIRST_DEVICE_VECTOR + offset;
1328 }
1329 if (unlikely(current_vector == vector))
1330 continue;
b77b881f
YL		 1331
1332 if (test_bit(vector, used_vectors))
54168ed7 1333 goto next;
b77b881f 1334
22f65d31 1335 for_each_cpu_and(new_cpu, tmp_mask, cpu_online_mask)
54168ed7
IM		 1336 if (per_cpu(vector_irq, new_cpu)[vector] != -1)
1337 goto next;
1338 /* Found one! */
1339 current_vector = vector;
1340 current_offset = offset;
1341 if (old_vector) {
1342 cfg->move_in_progress = 1;
22f65d31 1343 cpumask_copy(cfg->old_domain, cfg->domain);
7a959cff 1344 }
22f65d31 1345 for_each_cpu_and(new_cpu, tmp_mask, cpu_online_mask)
54168ed7
IM		 1346 per_cpu(vector_irq, new_cpu)[vector] = irq;
1347 cfg->vector = vector;
22f65d31
MT		 1348 cpumask_copy(cfg->domain, tmp_mask);
1349 err = 0;
1350 break;
54168ed7 1351 }
22f65d31
MT		 1352 free_cpumask_var(tmp_mask);
1353 return err;
497c9a19
YL		 1354}
1355
e7986739
MT		 1356static int
1357assign_irq_vector(int irq, struct irq_cfg *cfg, const struct cpumask *mask)
497c9a19
YL		 1358{
1359 int err;
ace80ab7 1360 unsigned long flags;
ace80ab7
EB		 1361
1362 spin_lock_irqsave(&vector_lock, flags);
3145e941 1363 err = __assign_irq_vector(irq, cfg, mask);
26a3c49c 1364 spin_unlock_irqrestore(&vector_lock, flags);
497c9a19
YL		 1365 return err;
1366}
1367
3145e941 1368static void __clear_irq_vector(int irq, struct irq_cfg *cfg)
497c9a19 1369{
497c9a19
YL		 1370 int cpu, vector;
1371
497c9a19
YL		 1372 BUG_ON(!cfg->vector);
1373
1374 vector = cfg->vector;
22f65d31 1375 for_each_cpu_and(cpu, cfg->domain, cpu_online_mask)
497c9a19
YL		 1376 per_cpu(vector_irq, cpu)[vector] = -1;
1377
1378 cfg->vector = 0;
22f65d31 1379 cpumask_clear(cfg->domain);
0ca4b6b0
MW		 1380
1381 if (likely(!cfg->move_in_progress))
1382 return;
22f65d31 1383 for_each_cpu_and(cpu, cfg->old_domain, cpu_online_mask) {
0ca4b6b0
MW		 1384 for (vector = FIRST_EXTERNAL_VECTOR; vector < NR_VECTORS;
1385 vector++) {
1386 if (per_cpu(vector_irq, cpu)[vector] != irq)
1387 continue;
1388 per_cpu(vector_irq, cpu)[vector] = -1;
1389 break;
1390 }
1391 }
1392 cfg->move_in_progress = 0;
497c9a19
YL		 1393}
1394
1395void __setup_vector_irq(int cpu)
1396{
1397 /* Initialize vector_irq on a new cpu */
1398 /* This function must be called with vector_lock held */
1399 int irq, vector;
1400 struct irq_cfg *cfg;
0b8f1efa 1401 struct irq_desc *desc;
497c9a19
YL		 1402
1403 /* Mark the inuse vectors */
0b8f1efa 1404 for_each_irq_desc(irq, desc) {
0b8f1efa 1405 cfg = desc->chip_data;
22f65d31 1406 if (!cpumask_test_cpu(cpu, cfg->domain))
497c9a19
YL		 1407 continue;
1408 vector = cfg->vector;
497c9a19
YL		 1409 per_cpu(vector_irq, cpu)[vector] = irq;
1410 }
1411 /* Mark the free vectors */
1412 for (vector = 0; vector < NR_VECTORS; ++vector) {
1413 irq = per_cpu(vector_irq, cpu)[vector];
1414 if (irq < 0)
1415 continue;
1416
1417 cfg = irq_cfg(irq);
22f65d31 1418 if (!cpumask_test_cpu(cpu, cfg->domain))
497c9a19 1419 per_cpu(vector_irq, cpu)[vector] = -1;
54168ed7 1420 }
1da177e4 1421}
3fde6900 1422
f5b9ed7a 1423static struct irq_chip ioapic_chip;
54168ed7
IM		 1424#ifdef CONFIG_INTR_REMAP
1425static struct irq_chip ir_ioapic_chip;
1426#endif
1da177e4 1427
54168ed7
IM		 1428#define IOAPIC_AUTO -1
1429#define IOAPIC_EDGE 0
1430#define IOAPIC_LEVEL 1
1da177e4 1431
047c8fdb 1432#ifdef CONFIG_X86_32
1d025192
YL		 1433static inline int IO_APIC_irq_trigger(int irq)
1434{
d6c88a50 1435 int apic, idx, pin;
1d025192 1436
d6c88a50
TG		 1437 for (apic = 0; apic < nr_ioapics; apic++) {
1438 for (pin = 0; pin < nr_ioapic_registers[apic]; pin++) {
1439 idx = find_irq_entry(apic, pin, mp_INT);
1440 if ((idx != -1) && (irq == pin_2_irq(idx, apic, pin)))
1441 return irq_trigger(idx);
1442 }
1443 }
1444 /*
54168ed7
IM		 1445 * nonexistent IRQs are edge default
1446 */
d6c88a50 1447 return 0;
1d025192 1448}
047c8fdb
YL		 1449#else
1450static inline int IO_APIC_irq_trigger(int irq)
1451{
54168ed7 1452 return 1;
047c8fdb
YL		 1453}
1454#endif
1d025192 1455
3145e941 1456static void ioapic_register_intr(int irq, struct irq_desc *desc, unsigned long trigger)
1da177e4 1457{
199751d7 1458
6ebcc00e 1459 if ((trigger == IOAPIC_AUTO && IO_APIC_irq_trigger(irq)) ||
047c8fdb 1460 trigger == IOAPIC_LEVEL)
08678b08 1461 desc->status |= IRQ_LEVEL;
047c8fdb
YL		 1462 else
1463 desc->status &= ~IRQ_LEVEL;
1464
54168ed7
IM		 1465#ifdef CONFIG_INTR_REMAP
1466 if (irq_remapped(irq)) {
1467 desc->status |= IRQ_MOVE_PCNTXT;
1468 if (trigger)
1469 set_irq_chip_and_handler_name(irq, &ir_ioapic_chip,
1470 handle_fasteoi_irq,
1471 "fasteoi");
1472 else
1473 set_irq_chip_and_handler_name(irq, &ir_ioapic_chip,
1474 handle_edge_irq, "edge");
1475 return;
1476 }
1477#endif
047c8fdb
YL		 1478 if ((trigger == IOAPIC_AUTO && IO_APIC_irq_trigger(irq)) ||
1479 trigger == IOAPIC_LEVEL)
a460e745 1480 set_irq_chip_and_handler_name(irq, &ioapic_chip,
54168ed7
IM		 1481 handle_fasteoi_irq,
1482 "fasteoi");
047c8fdb 1483 else
a460e745 1484 set_irq_chip_and_handler_name(irq, &ioapic_chip,
54168ed7 1485 handle_edge_irq, "edge");
1da177e4
LT		 1486}
1487
497c9a19
YL		 1488static int setup_ioapic_entry(int apic, int irq,
1489 struct IO_APIC_route_entry *entry,
1490 unsigned int destination, int trigger,
1491 int polarity, int vector)
1da177e4 1492{
497c9a19
YL		 1493 /*
1494 * add it to the IO-APIC irq-routing table:
1495 */
1496 memset(entry,0,sizeof(*entry));
1497
54168ed7
IM		 1498#ifdef CONFIG_INTR_REMAP
1499 if (intr_remapping_enabled) {
1500 struct intel_iommu *iommu = map_ioapic_to_ir(apic);
1501 struct irte irte;
1502 struct IR_IO_APIC_route_entry *ir_entry =
1503 (struct IR_IO_APIC_route_entry *) entry;
1504 int index;
1505
1506 if (!iommu)
1507 panic("No mapping iommu for ioapic %d\n", apic);
1508
1509 index = alloc_irte(iommu, irq, 1);
1510 if (index < 0)
1511 panic("Failed to allocate IRTE for ioapic %d\n", apic);
1512
1513 memset(&irte, 0, sizeof(irte));
1514
1515 irte.present = 1;
1516 irte.dst_mode = INT_DEST_MODE;
1517 irte.trigger_mode = trigger;
1518 irte.dlvry_mode = INT_DELIVERY_MODE;
1519 irte.vector = vector;
1520 irte.dest_id = IRTE_DEST(destination);
1521
1522 modify_irte(irq, &irte);
1523
1524 ir_entry->index2 = (index >> 15) & 0x1;
1525 ir_entry->zero = 0;
1526 ir_entry->format = 1;
1527 ir_entry->index = (index & 0x7fff);
1528 } else
1529#endif
1530 {
1531 entry->delivery_mode = INT_DELIVERY_MODE;
1532 entry->dest_mode = INT_DEST_MODE;
1533 entry->dest = destination;
1534 }
497c9a19 1535
54168ed7 1536 entry->mask = 0; /* enable IRQ */
497c9a19
YL		 1537 entry->trigger = trigger;
1538 entry->polarity = polarity;
1539 entry->vector = vector;
1540
1541 /* Mask level triggered irqs.
1542 * Use IRQ_DELAYED_DISABLE for edge triggered irqs.
1543 */
1544 if (trigger)
1545 entry->mask = 1;
497c9a19
YL		 1546 return 0;
1547}
1548
3145e941 1549static void setup_IO_APIC_irq(int apic, int pin, unsigned int irq, struct irq_desc *desc,
54168ed7 1550 int trigger, int polarity)
497c9a19
YL		 1551{
1552 struct irq_cfg *cfg;
1da177e4 1553 struct IO_APIC_route_entry entry;
22f65d31 1554 unsigned int dest;
497c9a19
YL		 1555
1556 if (!IO_APIC_IRQ(irq))
1557 return;
1558
3145e941 1559 cfg = desc->chip_data;
497c9a19 1560
22f65d31 1561 if (assign_irq_vector(irq, cfg, TARGET_CPUS))
497c9a19
YL		 1562 return;
1563
22f65d31 1564 dest = cpu_mask_to_apicid_and(cfg->domain, TARGET_CPUS);
497c9a19
YL		 1565
1566 apic_printk(APIC_VERBOSE,KERN_DEBUG
1567 "IOAPIC[%d]: Set routing entry (%d-%d -> 0x%x -> "
1568 "IRQ %d Mode:%i Active:%i)\n",
1569 apic, mp_ioapics[apic].mp_apicid, pin, cfg->vector,
1570 irq, trigger, polarity);
1571
1572
1573 if (setup_ioapic_entry(mp_ioapics[apic].mp_apicid, irq, &entry,
22f65d31 1574 dest, trigger, polarity, cfg->vector)) {
497c9a19
YL		 1575 printk("Failed to setup ioapic entry for ioapic %d, pin %d\n",
1576 mp_ioapics[apic].mp_apicid, pin);
3145e941 1577 __clear_irq_vector(irq, cfg);
497c9a19
YL		 1578 return;
1579 }
1580
3145e941 1581 ioapic_register_intr(irq, desc, trigger);
99d093d1 1582 if (irq < NR_IRQS_LEGACY)
497c9a19
YL		 1583 disable_8259A_irq(irq);
1584
1585 ioapic_write_entry(apic, pin, entry);
1586}
1587
1588static void __init setup_IO_APIC_irqs(void)
1589{
3c2cbd24
CG		 1590 int apic, pin, idx, irq;
1591 int notcon = 0;
0b8f1efa 1592 struct irq_desc *desc;
3145e941 1593 struct irq_cfg *cfg;
0b8f1efa 1594 int cpu = boot_cpu_id;
1da177e4
LT		 1595
1596 apic_printk(APIC_VERBOSE, KERN_DEBUG "init IO_APIC IRQs\n");
1597
1598 for (apic = 0; apic < nr_ioapics; apic++) {
3c2cbd24 1599 for (pin = 0; pin < nr_ioapic_registers[apic]; pin++) {
20d225b9 1600
3c2cbd24
CG		 1601 idx = find_irq_entry(apic, pin, mp_INT);
1602 if (idx == -1) {
2a554fb1 1603 if (!notcon) {
3c2cbd24 1604 notcon = 1;
2a554fb1
CG		 1605 apic_printk(APIC_VERBOSE,
1606 KERN_DEBUG " %d-%d",
1607 mp_ioapics[apic].mp_apicid,
1608 pin);
1609 } else
1610 apic_printk(APIC_VERBOSE, " %d-%d",
1611 mp_ioapics[apic].mp_apicid,
1612 pin);
3c2cbd24
CG		 1613 continue;
1614 }
56ffa1a0
CG		 1615 if (notcon) {
1616 apic_printk(APIC_VERBOSE,
1617 " (apicid-pin) not connected\n");
1618 notcon = 0;
1619 }
3c2cbd24
CG		 1620
1621 irq = pin_2_irq(idx, apic, pin);
54168ed7 1622#ifdef CONFIG_X86_32
3c2cbd24
CG		 1623 if (multi_timer_check(apic, irq))
1624 continue;
54168ed7 1625#endif
0b8f1efa
YL		 1626 desc = irq_to_desc_alloc_cpu(irq, cpu);
1627 if (!desc) {
1628 printk(KERN_INFO "can not get irq_desc for %d\n", irq);
1629 continue;
1630 }
3145e941
YL		 1631 cfg = desc->chip_data;
1632 add_pin_to_irq_cpu(cfg, cpu, apic, pin);
36062448 1633
3145e941 1634 setup_IO_APIC_irq(apic, pin, irq, desc,
3c2cbd24
CG		 1635 irq_trigger(idx), irq_polarity(idx));
1636 }
1da177e4
LT		 1637 }
1638
3c2cbd24
CG		 1639 if (notcon)
1640 apic_printk(APIC_VERBOSE,
2a554fb1 1641 " (apicid-pin) not connected\n");
1da177e4
LT		 1642}
1643
1644/*
f7633ce5 1645 * Set up the timer pin, possibly with the 8259A-master behind.
1da177e4 1646 */
f7633ce5
MR		 1647static void __init setup_timer_IRQ0_pin(unsigned int apic, unsigned int pin,
1648 int vector)
1da177e4
LT		 1649{
1650 struct IO_APIC_route_entry entry;
1da177e4 1651
54168ed7
IM		 1652#ifdef CONFIG_INTR_REMAP
1653 if (intr_remapping_enabled)
1654 return;
1655#endif
1656
36062448 1657 memset(&entry, 0, sizeof(entry));
1da177e4
LT		 1658
1659 /*
1660 * We use logical delivery to get the timer IRQ
1661 * to the first CPU.
1662 */
1663 entry.dest_mode = INT_DEST_MODE;
03be7505 1664 entry.mask = 1; /* mask IRQ now */
d83e94ac 1665 entry.dest = cpu_mask_to_apicid(TARGET_CPUS);
1da177e4
LT		 1666 entry.delivery_mode = INT_DELIVERY_MODE;
1667 entry.polarity = 0;
1668 entry.trigger = 0;
1669 entry.vector = vector;
1670
1671 /*
1672 * The timer IRQ doesn't have to know that behind the
f7633ce5 1673 * scene we may have a 8259A-master in AEOI mode ...
1da177e4 1674 */
54168ed7 1675 set_irq_chip_and_handler_name(0, &ioapic_chip, handle_edge_irq, "edge");
1da177e4
LT		 1676
1677 /*
1678 * Add it to the IO-APIC irq-routing table:
1679 */
cf4c6a2f 1680 ioapic_write_entry(apic, pin, entry);
1da177e4
LT		 1681}
1682
32f71aff
MR		 1683
1684__apicdebuginit(void) print_IO_APIC(void)
1da177e4
LT		 1685{
1686 int apic, i;
1687 union IO_APIC_reg_00 reg_00;
1688 union IO_APIC_reg_01 reg_01;
1689 union IO_APIC_reg_02 reg_02;
1690 union IO_APIC_reg_03 reg_03;
1691 unsigned long flags;
0f978f45 1692 struct irq_cfg *cfg;
0b8f1efa 1693 struct irq_desc *desc;
8f09cd20 1694 unsigned int irq;
1da177e4
LT		 1695
1696 if (apic_verbosity == APIC_QUIET)
1697 return;
1698
36062448 1699 printk(KERN_DEBUG "number of MP IRQ sources: %d.\n", mp_irq_entries);
1da177e4
LT		 1700 for (i = 0; i < nr_ioapics; i++)
1701 printk(KERN_DEBUG "number of IO-APIC #%d registers: %d.\n",
ec2cd0a2 1702 mp_ioapics[i].mp_apicid, nr_ioapic_registers[i]);
1da177e4
LT		 1703
1704 /*
1705 * We are a bit conservative about what we expect. We have to
1706 * know about every hardware change ASAP.
1707 */
1708 printk(KERN_INFO "testing the IO APIC.......................\n");
1709
1710 for (apic = 0; apic < nr_ioapics; apic++) {
1711
1712 spin_lock_irqsave(&ioapic_lock, flags);
1713 reg_00.raw = io_apic_read(apic, 0);
1714 reg_01.raw = io_apic_read(apic, 1);
1715 if (reg_01.bits.version >= 0x10)
1716 reg_02.raw = io_apic_read(apic, 2);
d6c88a50
TG		 1717 if (reg_01.bits.version >= 0x20)
1718 reg_03.raw = io_apic_read(apic, 3);
1da177e4
LT		 1719 spin_unlock_irqrestore(&ioapic_lock, flags);
1720
54168ed7 1721 printk("\n");
ec2cd0a2 1722 printk(KERN_DEBUG "IO APIC #%d......\n", mp_ioapics[apic].mp_apicid);
1da177e4
LT		 1723 printk(KERN_DEBUG ".... register #00: %08X\n", reg_00.raw);
1724 printk(KERN_DEBUG "....... : physical APIC id: %02X\n", reg_00.bits.ID);
1725 printk(KERN_DEBUG "....... : Delivery Type: %X\n", reg_00.bits.delivery_type);
1726 printk(KERN_DEBUG "....... : LTS : %X\n", reg_00.bits.LTS);
1da177e4 1727
54168ed7 1728 printk(KERN_DEBUG ".... register #01: %08X\n", *(int *)&reg_01);
1da177e4 1729 printk(KERN_DEBUG "....... : max redirection entries: %04X\n", reg_01.bits.entries);
1da177e4
LT		 1730
1731 printk(KERN_DEBUG "....... : PRQ implemented: %X\n", reg_01.bits.PRQ);
1732 printk(KERN_DEBUG "....... : IO APIC version: %04X\n", reg_01.bits.version);
1da177e4
LT		 1733
1734 /*
1735 * Some Intel chipsets with IO APIC VERSION of 0x1? don't have reg_02,
1736 * but the value of reg_02 is read as the previous read register
1737 * value, so ignore it if reg_02 == reg_01.
1738 */
1739 if (reg_01.bits.version >= 0x10 && reg_02.raw != reg_01.raw) {
1740 printk(KERN_DEBUG ".... register #02: %08X\n", reg_02.raw);
1741 printk(KERN_DEBUG "....... : arbitration: %02X\n", reg_02.bits.arbitration);
1da177e4
LT		 1742 }
1743
1744 /*
1745 * Some Intel chipsets with IO APIC VERSION of 0x2? don't have reg_02
1746 * or reg_03, but the value of reg_0[23] is read as the previous read
1747 * register value, so ignore it if reg_03 == reg_0[12].
1748 */
1749 if (reg_01.bits.version >= 0x20 && reg_03.raw != reg_02.raw &&
1750 reg_03.raw != reg_01.raw) {
1751 printk(KERN_DEBUG ".... register #03: %08X\n", reg_03.raw);
1752 printk(KERN_DEBUG "....... : Boot DT : %X\n", reg_03.bits.boot_DT);
1da177e4
LT		 1753 }
1754
1755 printk(KERN_DEBUG ".... IRQ redirection table:\n");
1756
d83e94ac
YL		 1757 printk(KERN_DEBUG " NR Dst Mask Trig IRR Pol"
1758 " Stat Dmod Deli Vect: \n");
1da177e4
LT		 1759
1760 for (i = 0; i <= reg_01.bits.entries; i++) {
1761 struct IO_APIC_route_entry entry;
1762
cf4c6a2f 1763 entry = ioapic_read_entry(apic, i);
1da177e4 1764
54168ed7
IM		 1765 printk(KERN_DEBUG " %02x %03X ",
1766 i,
1767 entry.dest
1768 );
1da177e4
LT		 1769
1770 printk("%1d %1d %1d %1d %1d %1d %1d %02X\n",
1771 entry.mask,
1772 entry.trigger,
1773 entry.irr,
1774 entry.polarity,
1775 entry.delivery_status,
1776 entry.dest_mode,
1777 entry.delivery_mode,
1778 entry.vector
1779 );
1780 }
1781 }
1da177e4 1782 printk(KERN_DEBUG "IRQ to pin mappings:\n");
0b8f1efa
YL		 1783 for_each_irq_desc(irq, desc) {
1784 struct irq_pin_list *entry;
1785
0b8f1efa
YL		 1786 cfg = desc->chip_data;
1787 entry = cfg->irq_2_pin;
0f978f45 1788 if (!entry)
1da177e4 1789 continue;
8f09cd20 1790 printk(KERN_DEBUG "IRQ%d ", irq);
1da177e4
LT		 1791 for (;;) {
1792 printk("-> %d:%d", entry->apic, entry->pin);
1793 if (!entry->next)
1794 break;
0f978f45 1795 entry = entry->next;
1da177e4
LT		 1796 }
1797 printk("\n");
1798 }
1799
1800 printk(KERN_INFO ".................................... done.\n");
1801
1802 return;
1803}
1804
32f71aff 1805__apicdebuginit(void) print_APIC_bitfield(int base)
1da177e4
LT		 1806{
1807 unsigned int v;
1808 int i, j;
1809
1810 if (apic_verbosity == APIC_QUIET)
1811 return;
1812
1813 printk(KERN_DEBUG "0123456789abcdef0123456789abcdef\n" KERN_DEBUG);
1814 for (i = 0; i < 8; i++) {
1815 v = apic_read(base + i*0x10);
1816 for (j = 0; j < 32; j++) {
1817 if (v & (1<<j))
1818 printk("1");
1819 else
1820 printk("0");
1821 }
1822 printk("\n");
1823 }
1824}
1825
32f71aff 1826__apicdebuginit(void) print_local_APIC(void *dummy)
1da177e4
LT		 1827{
1828 unsigned int v, ver, maxlvt;
7ab6af7a 1829 u64 icr;
1da177e4
LT		 1830
1831 if (apic_verbosity == APIC_QUIET)
1832 return;
1833
1834 printk("\n" KERN_DEBUG "printing local APIC contents on CPU#%d/%d:\n",
1835 smp_processor_id(), hard_smp_processor_id());
66823114 1836 v = apic_read(APIC_ID);
54168ed7 1837 printk(KERN_INFO "... APIC ID: %08x (%01x)\n", v, read_apic_id());
1da177e4
LT		 1838 v = apic_read(APIC_LVR);
1839 printk(KERN_INFO "... APIC VERSION: %08x\n", v);
1840 ver = GET_APIC_VERSION(v);
e05d723f 1841 maxlvt = lapic_get_maxlvt();
1da177e4
LT		 1842
1843 v = apic_read(APIC_TASKPRI);
1844 printk(KERN_DEBUG "... APIC TASKPRI: %08x (%02x)\n", v, v & APIC_TPRI_MASK);
1845
54168ed7 1846 if (APIC_INTEGRATED(ver)) { /* !82489DX */
a11b5abe
YL		 1847 if (!APIC_XAPIC(ver)) {
1848 v = apic_read(APIC_ARBPRI);
1849 printk(KERN_DEBUG "... APIC ARBPRI: %08x (%02x)\n", v,
1850 v & APIC_ARBPRI_MASK);
1851 }
1da177e4
LT		 1852 v = apic_read(APIC_PROCPRI);
1853 printk(KERN_DEBUG "... APIC PROCPRI: %08x\n", v);
1854 }
1855
a11b5abe
YL		 1856 /*
1857 * Remote read supported only in the 82489DX and local APIC for
1858 * Pentium processors.
1859 */
1860 if (!APIC_INTEGRATED(ver) || maxlvt == 3) {
1861 v = apic_read(APIC_RRR);
1862 printk(KERN_DEBUG "... APIC RRR: %08x\n", v);
1863 }
1864
1da177e4
LT		 1865 v = apic_read(APIC_LDR);
1866 printk(KERN_DEBUG "... APIC LDR: %08x\n", v);
a11b5abe
YL		 1867 if (!x2apic_enabled()) {
1868 v = apic_read(APIC_DFR);
1869 printk(KERN_DEBUG "... APIC DFR: %08x\n", v);
1870 }
1da177e4
LT		 1871 v = apic_read(APIC_SPIV);
1872 printk(KERN_DEBUG "... APIC SPIV: %08x\n", v);
1873
1874 printk(KERN_DEBUG "... APIC ISR field:\n");
1875 print_APIC_bitfield(APIC_ISR);
1876 printk(KERN_DEBUG "... APIC TMR field:\n");
1877 print_APIC_bitfield(APIC_TMR);
1878 printk(KERN_DEBUG "... APIC IRR field:\n");
1879 print_APIC_bitfield(APIC_IRR);
1880
54168ed7
IM		 1881 if (APIC_INTEGRATED(ver)) { /* !82489DX */
1882 if (maxlvt > 3) /* Due to the Pentium erratum 3AP. */
1da177e4 1883 apic_write(APIC_ESR, 0);
54168ed7 1884
1da177e4
LT		 1885 v = apic_read(APIC_ESR);
1886 printk(KERN_DEBUG "... APIC ESR: %08x\n", v);
1887 }
1888
7ab6af7a 1889 icr = apic_icr_read();
0c425cec
IM		 1890 printk(KERN_DEBUG "... APIC ICR: %08x\n", (u32)icr);
1891 printk(KERN_DEBUG "... APIC ICR2: %08x\n", (u32)(icr >> 32));
1da177e4
LT		 1892
1893 v = apic_read(APIC_LVTT);
1894 printk(KERN_DEBUG "... APIC LVTT: %08x\n", v);
1895
1896 if (maxlvt > 3) { /* PC is LVT#4. */
1897 v = apic_read(APIC_LVTPC);
1898 printk(KERN_DEBUG "... APIC LVTPC: %08x\n", v);
1899 }
1900 v = apic_read(APIC_LVT0);
1901 printk(KERN_DEBUG "... APIC LVT0: %08x\n", v);
1902 v = apic_read(APIC_LVT1);
1903 printk(KERN_DEBUG "... APIC LVT1: %08x\n", v);
1904
1905 if (maxlvt > 2) { /* ERR is LVT#3. */
1906 v = apic_read(APIC_LVTERR);
1907 printk(KERN_DEBUG "... APIC LVTERR: %08x\n", v);
1908 }
1909
1910 v = apic_read(APIC_TMICT);
1911 printk(KERN_DEBUG "... APIC TMICT: %08x\n", v);
1912 v = apic_read(APIC_TMCCT);
1913 printk(KERN_DEBUG "... APIC TMCCT: %08x\n", v);
1914 v = apic_read(APIC_TDCR);
1915 printk(KERN_DEBUG "... APIC TDCR: %08x\n", v);
1916 printk("\n");
1917}
1918
32f71aff 1919__apicdebuginit(void) print_all_local_APICs(void)
1da177e4 1920{
ffd5aae7
YL		 1921 int cpu;
1922
1923 preempt_disable();
1924 for_each_online_cpu(cpu)
1925 smp_call_function_single(cpu, print_local_APIC, NULL, 1);
1926 preempt_enable();
1da177e4
LT		 1927}
1928
32f71aff 1929__apicdebuginit(void) print_PIC(void)
1da177e4 1930{
1da177e4
LT		 1931 unsigned int v;
1932 unsigned long flags;
1933
1934 if (apic_verbosity == APIC_QUIET)
1935 return;
1936
1937 printk(KERN_DEBUG "\nprinting PIC contents\n");
1938
1939 spin_lock_irqsave(&i8259A_lock, flags);
1940
1941 v = inb(0xa1) << 8 | inb(0x21);
1942 printk(KERN_DEBUG "... PIC IMR: %04x\n", v);
1943
1944 v = inb(0xa0) << 8 | inb(0x20);
1945 printk(KERN_DEBUG "... PIC IRR: %04x\n", v);
1946
54168ed7
IM		 1947 outb(0x0b,0xa0);
1948 outb(0x0b,0x20);
1da177e4 1949 v = inb(0xa0) << 8 | inb(0x20);
54168ed7
IM		 1950 outb(0x0a,0xa0);
1951 outb(0x0a,0x20);
1da177e4
LT		 1952
1953 spin_unlock_irqrestore(&i8259A_lock, flags);
1954
1955 printk(KERN_DEBUG "... PIC ISR: %04x\n", v);
1956
1957 v = inb(0x4d1) << 8 | inb(0x4d0);
1958 printk(KERN_DEBUG "... PIC ELCR: %04x\n", v);
1959}
1960
32f71aff
MR		 1961__apicdebuginit(int) print_all_ICs(void)
1962{
1963 print_PIC();
1964 print_all_local_APICs();
1965 print_IO_APIC();
1966
1967 return 0;
1968}
1969
1970fs_initcall(print_all_ICs);
1971
1da177e4 1972
efa2559f
YL		 1973/* Where if anywhere is the i8259 connect in external int mode */
1974static struct { int pin, apic; } ioapic_i8259 = { -1, -1 };
1975
54168ed7 1976void __init enable_IO_APIC(void)
1da177e4
LT		 1977{
1978 union IO_APIC_reg_01 reg_01;
fcfd636a 1979 int i8259_apic, i8259_pin;
54168ed7 1980 int apic;
1da177e4
LT		 1981 unsigned long flags;
1982
54168ed7
IM		 1983#ifdef CONFIG_X86_32
1984 int i;
1da177e4
LT		 1985 if (!pirqs_enabled)
1986 for (i = 0; i < MAX_PIRQS; i++)
1987 pirq_entries[i] = -1;
54168ed7 1988#endif
1da177e4
LT		 1989
1990 /*
1991 * The number of IO-APIC IRQ registers (== #pins):
1992 */
fcfd636a 1993 for (apic = 0; apic < nr_ioapics; apic++) {
1da177e4 1994 spin_lock_irqsave(&ioapic_lock, flags);
fcfd636a 1995 reg_01.raw = io_apic_read(apic, 1);
1da177e4 1996 spin_unlock_irqrestore(&ioapic_lock, flags);
fcfd636a
EB		 1997 nr_ioapic_registers[apic] = reg_01.bits.entries+1;
1998 }
54168ed7 1999 for(apic = 0; apic < nr_ioapics; apic++) {
fcfd636a
EB		 2000 int pin;
2001 /* See if any of the pins is in ExtINT mode */
1008fddc 2002 for (pin = 0; pin < nr_ioapic_registers[apic]; pin++) {
fcfd636a 2003 struct IO_APIC_route_entry entry;
cf4c6a2f 2004 entry = ioapic_read_entry(apic, pin);
fcfd636a 2005
fcfd636a
EB		 2006 /* If the interrupt line is enabled and in ExtInt mode
2007 * I have found the pin where the i8259 is connected.
2008 */
2009 if ((entry.mask == 0) && (entry.delivery_mode == dest_ExtINT)) {
2010 ioapic_i8259.apic = apic;
2011 ioapic_i8259.pin = pin;
2012 goto found_i8259;
2013 }
2014 }
2015 }
2016 found_i8259:
2017 /* Look to see what if the MP table has reported the ExtINT */
2018 /* If we could not find the appropriate pin by looking at the ioapic
2019 * the i8259 probably is not connected the ioapic but give the
2020 * mptable a chance anyway.
2021 */
2022 i8259_pin = find_isa_irq_pin(0, mp_ExtINT);
2023 i8259_apic = find_isa_irq_apic(0, mp_ExtINT);
2024 /* Trust the MP table if nothing is setup in the hardware */
2025 if ((ioapic_i8259.pin == -1) && (i8259_pin >= 0)) {
2026 printk(KERN_WARNING "ExtINT not setup in hardware but reported by MP table\n");
2027 ioapic_i8259.pin = i8259_pin;
2028 ioapic_i8259.apic = i8259_apic;
2029 }
2030 /* Complain if the MP table and the hardware disagree */
2031 if (((ioapic_i8259.apic != i8259_apic) || (ioapic_i8259.pin != i8259_pin)) &&
2032 (i8259_pin >= 0) && (ioapic_i8259.pin >= 0))
2033 {
2034 printk(KERN_WARNING "ExtINT in hardware and MP table differ\n");
1da177e4
LT		 2035 }
2036
2037 /*
2038 * Do not trust the IO-APIC being empty at bootup
2039 */
2040 clear_IO_APIC();
2041}
2042
2043/*
2044 * Not an __init, needed by the reboot code
2045 */
2046void disable_IO_APIC(void)
2047{
2048 /*
2049 * Clear the IO-APIC before rebooting:
2050 */
2051 clear_IO_APIC();
2052
650927ef 2053 /*
0b968d23 2054 * If the i8259 is routed through an IOAPIC
650927ef 2055 * Put that IOAPIC in virtual wire mode
0b968d23 2056 * so legacy interrupts can be delivered.
650927ef 2057 */
fcfd636a 2058 if (ioapic_i8259.pin != -1) {
650927ef 2059 struct IO_APIC_route_entry entry;
650927ef
EB		 2060
2061 memset(&entry, 0, sizeof(entry));
2062 entry.mask = 0; /* Enabled */
2063 entry.trigger = 0; /* Edge */
2064 entry.irr = 0;
2065 entry.polarity = 0; /* High */
2066 entry.delivery_status = 0;
2067 entry.dest_mode = 0; /* Physical */
fcfd636a 2068 entry.delivery_mode = dest_ExtINT; /* ExtInt */
650927ef 2069 entry.vector = 0;
54168ed7 2070 entry.dest = read_apic_id();
650927ef
EB		 2071
2072 /*
2073 * Add it to the IO-APIC irq-routing table:
2074 */
cf4c6a2f 2075 ioapic_write_entry(ioapic_i8259.apic, ioapic_i8259.pin, entry);
650927ef 2076 }
54168ed7 2077
fcfd636a 2078 disconnect_bsp_APIC(ioapic_i8259.pin != -1);
1da177e4
LT		 2079}
2080
54168ed7 2081#ifdef CONFIG_X86_32
1da177e4
LT		 2082/*
2083 * function to set the IO-APIC physical IDs based on the
2084 * values stored in the MPC table.
2085 *
2086 * by Matt Domsch <Matt_Domsch@dell.com> Tue Dec 21 12:25:05 CST 1999
2087 */
2088
1da177e4
LT		 2089static void __init setup_ioapic_ids_from_mpc(void)
2090{
2091 union IO_APIC_reg_00 reg_00;
2092 physid_mask_t phys_id_present_map;
2093 int apic;
2094 int i;
2095 unsigned char old_id;
2096 unsigned long flags;
2097
a4dbc34d 2098 if (x86_quirks->setup_ioapic_ids && x86_quirks->setup_ioapic_ids())
d49c4288 2099 return;
d49c4288 2100
ca05fea6
NP		 2101 /*
2102 * Don't check I/O APIC IDs for xAPIC systems. They have
2103 * no meaning without the serial APIC bus.
2104 */
7c5c1e42
SL		 2105 if (!(boot_cpu_data.x86_vendor == X86_VENDOR_INTEL)
2106 || APIC_XAPIC(apic_version[boot_cpu_physical_apicid]))
ca05fea6 2107 return;
1da177e4
LT		 2108 /*
2109 * This is broken; anything with a real cpu count has to
2110 * circumvent this idiocy regardless.
2111 */
2112 phys_id_present_map = ioapic_phys_id_map(phys_cpu_present_map);
2113
2114 /*
2115 * Set the IOAPIC ID to the value stored in the MPC table.
2116 */
2117 for (apic = 0; apic < nr_ioapics; apic++) {
2118
2119 /* Read the register 0 value */
2120 spin_lock_irqsave(&ioapic_lock, flags);
2121 reg_00.raw = io_apic_read(apic, 0);
2122 spin_unlock_irqrestore(&ioapic_lock, flags);
36062448 2123
ec2cd0a2 2124 old_id = mp_ioapics[apic].mp_apicid;
1da177e4 2125
ec2cd0a2 2126 if (mp_ioapics[apic].mp_apicid >= get_physical_broadcast()) {
1da177e4 2127 printk(KERN_ERR "BIOS bug, IO-APIC#%d ID is %d in the MPC table!...\n",
ec2cd0a2 2128 apic, mp_ioapics[apic].mp_apicid);
1da177e4
LT		 2129 printk(KERN_ERR "... fixing up to %d. (tell your hw vendor)\n",
2130 reg_00.bits.ID);
ec2cd0a2 2131 mp_ioapics[apic].mp_apicid = reg_00.bits.ID;
1da177e4
LT		 2132 }
2133
1da177e4
LT		 2134 /*
2135 * Sanity check, is the ID really free? Every APIC in a
2136 * system must have a unique ID or we get lots of nice
2137 * 'stuck on smp_invalidate_needed IPI wait' messages.
2138 */
2139 if (check_apicid_used(phys_id_present_map,
ec2cd0a2 2140 mp_ioapics[apic].mp_apicid)) {
1da177e4 2141 printk(KERN_ERR "BIOS bug, IO-APIC#%d ID %d is already used!...\n",
ec2cd0a2 2142 apic, mp_ioapics[apic].mp_apicid);
1da177e4
LT		 2143 for (i = 0; i < get_physical_broadcast(); i++)
2144 if (!physid_isset(i, phys_id_present_map))
2145 break;
2146 if (i >= get_physical_broadcast())
2147 panic("Max APIC ID exceeded!\n");
2148 printk(KERN_ERR "... fixing up to %d. (tell your hw vendor)\n",
2149 i);
2150 physid_set(i, phys_id_present_map);
ec2cd0a2 2151 mp_ioapics[apic].mp_apicid = i;
1da177e4
LT		 2152 } else {
2153 physid_mask_t tmp;
ec2cd0a2 2154 tmp = apicid_to_cpu_present(mp_ioapics[apic].mp_apicid);
1da177e4
LT		 2155 apic_printk(APIC_VERBOSE, "Setting %d in the "
2156 "phys_id_present_map\n",
ec2cd0a2 2157 mp_ioapics[apic].mp_apicid);
1da177e4
LT		 2158 physids_or(phys_id_present_map, phys_id_present_map, tmp);
2159 }
2160
2161
2162 /*
2163 * We need to adjust the IRQ routing table
2164 * if the ID changed.
2165 */
ec2cd0a2 2166 if (old_id != mp_ioapics[apic].mp_apicid)
1da177e4 2167 for (i = 0; i < mp_irq_entries; i++)
2fddb6e2
AS		 2168 if (mp_irqs[i].mp_dstapic == old_id)
2169 mp_irqs[i].mp_dstapic
ec2cd0a2 2170 = mp_ioapics[apic].mp_apicid;
1da177e4
LT		 2171
2172 /*
2173 * Read the right value from the MPC table and
2174 * write it into the ID register.
36062448 2175 */
1da177e4
LT		 2176 apic_printk(APIC_VERBOSE, KERN_INFO
2177 "...changing IO-APIC physical APIC ID to %d ...",
ec2cd0a2 2178 mp_ioapics[apic].mp_apicid);
1da177e4 2179
ec2cd0a2 2180 reg_00.bits.ID = mp_ioapics[apic].mp_apicid;
1da177e4 2181 spin_lock_irqsave(&ioapic_lock, flags);
a2d332fa
YL		 2182 io_apic_write(apic, 0, reg_00.raw);
2183 spin_unlock_irqrestore(&ioapic_lock, flags);
1da177e4
LT		 2184
2185 /*
2186 * Sanity check
2187 */
2188 spin_lock_irqsave(&ioapic_lock, flags);
2189 reg_00.raw = io_apic_read(apic, 0);
2190 spin_unlock_irqrestore(&ioapic_lock, flags);
ec2cd0a2 2191 if (reg_00.bits.ID != mp_ioapics[apic].mp_apicid)
1da177e4
LT		 2192 printk("could not set ID!\n");
2193 else
2194 apic_printk(APIC_VERBOSE, " ok.\n");
2195 }
2196}
54168ed7 2197#endif
1da177e4 2198
7ce0bcfd 2199int no_timer_check __initdata;
8542b200
ZA		 2200
2201static int __init notimercheck(char *s)
2202{
2203 no_timer_check = 1;
2204 return 1;
2205}
2206__setup("no_timer_check", notimercheck);
2207
1da177e4
LT		 2208/*
2209 * There is a nasty bug in some older SMP boards, their mptable lies
2210 * about the timer IRQ. We do the following to work around the situation:
2211 *
2212 * - timer IRQ defaults to IO-APIC IRQ
2213 * - if this function detects that timer IRQs are defunct, then we fall
2214 * back to ISA timer IRQs
2215 */
f0a7a5c9 2216static int __init timer_irq_works(void)
1da177e4
LT		 2217{
2218 unsigned long t1 = jiffies;
4aae0702 2219 unsigned long flags;
1da177e4 2220
8542b200
ZA		 2221 if (no_timer_check)
2222 return 1;
2223
4aae0702 2224 local_save_flags(flags);
1da177e4
LT		 2225 local_irq_enable();
2226 /* Let ten ticks pass... */
2227 mdelay((10 * 1000) / HZ);
4aae0702 2228 local_irq_restore(flags);
1da177e4
LT		 2229
2230 /*
2231 * Expect a few ticks at least, to be sure some possible
2232 * glue logic does not lock up after one or two first
2233 * ticks in a non-ExtINT mode. Also the local APIC
2234 * might have cached one ExtINT interrupt. Finally, at
2235 * least one tick may be lost due to delays.
2236 */
54168ed7
IM		 2237
2238 /* jiffies wrap? */
1d16b53e 2239 if (time_after(jiffies, t1 + 4))
1da177e4 2240 return 1;
1da177e4
LT		 2241 return 0;
2242}
2243
2244/*
2245 * In the SMP+IOAPIC case it might happen that there are an unspecified
2246 * number of pending IRQ events unhandled. These cases are very rare,
2247 * so we 'resend' these IRQs via IPIs, to the same CPU. It's much
2248 * better to do it this way as thus we do not have to be aware of
2249 * 'pending' interrupts in the IRQ path, except at this point.
2250 */
2251/*
2252 * Edge triggered needs to resend any interrupt
2253 * that was delayed but this is now handled in the device
2254 * independent code.
2255 */
2256
2257/*
2258 * Starting up a edge-triggered IO-APIC interrupt is
2259 * nasty - we need to make sure that we get the edge.
2260 * If it is already asserted for some reason, we need
2261 * return 1 to indicate that is was pending.
2262 *
2263 * This is not complete - we should be able to fake
2264 * an edge even if it isn't on the 8259A...
2265 */
54168ed7 2266
f5b9ed7a 2267static unsigned int startup_ioapic_irq(unsigned int irq)
1da177e4
LT		 2268{
2269 int was_pending = 0;
2270 unsigned long flags;
0b8f1efa 2271 struct irq_cfg *cfg;
1da177e4
LT		 2272
2273 spin_lock_irqsave(&ioapic_lock, flags);
99d093d1 2274 if (irq < NR_IRQS_LEGACY) {
1da177e4
LT		 2275 disable_8259A_irq(irq);
2276 if (i8259A_irq_pending(irq))
2277 was_pending = 1;
2278 }
0b8f1efa 2279 cfg = irq_cfg(irq);
3145e941 2280 __unmask_IO_APIC_irq(cfg);
1da177e4
LT		 2281 spin_unlock_irqrestore(&ioapic_lock, flags);
2282
2283 return was_pending;
2284}
2285
54168ed7 2286#ifdef CONFIG_X86_64
ace80ab7 2287static int ioapic_retrigger_irq(unsigned int irq)
1da177e4 2288{
54168ed7
IM		 2289
2290 struct irq_cfg *cfg = irq_cfg(irq);
2291 unsigned long flags;
2292
2293 spin_lock_irqsave(&vector_lock, flags);
22f65d31 2294 send_IPI_mask(cpumask_of(cpumask_first(cfg->domain)), cfg->vector);
54168ed7 2295 spin_unlock_irqrestore(&vector_lock, flags);
c0ad90a3
IM		 2296
2297 return 1;
2298}
54168ed7
IM		 2299#else
2300static int ioapic_retrigger_irq(unsigned int irq)
497c9a19 2301{
d6c88a50 2302 send_IPI_self(irq_cfg(irq)->vector);
497c9a19 2303
d6c88a50 2304 return 1;
54168ed7
IM		 2305}
2306#endif
497c9a19 2307
54168ed7
IM		 2308/*
2309 * Level and edge triggered IO-APIC interrupts need different handling,
2310 * so we use two separate IRQ descriptors. Edge triggered IRQs can be
2311 * handled with the level-triggered descriptor, but that one has slightly
2312 * more overhead. Level-triggered interrupts cannot be handled with the
2313 * edge-triggered handler, without risking IRQ storms and other ugly
2314 * races.
2315 */
497c9a19 2316
54168ed7 2317#ifdef CONFIG_SMP
497c9a19 2318
54168ed7
IM		 2319#ifdef CONFIG_INTR_REMAP
2320static void ir_irq_migration(struct work_struct *work);
497c9a19 2321
54168ed7 2322static DECLARE_DELAYED_WORK(ir_migration_work, ir_irq_migration);
497c9a19 2323
54168ed7
IM		 2324/*
2325 * Migrate the IO-APIC irq in the presence of intr-remapping.
2326 *
2327 * For edge triggered, irq migration is a simple atomic update(of vector
2328 * and cpu destination) of IRTE and flush the hardware cache.
2329 *
2330 * For level triggered, we need to modify the io-apic RTE aswell with the update
2331 * vector information, along with modifying IRTE with vector and destination.
2332 * So irq migration for level triggered is little bit more complex compared to
2333 * edge triggered migration. But the good news is, we use the same algorithm
2334 * for level triggered migration as we have today, only difference being,
2335 * we now initiate the irq migration from process context instead of the
2336 * interrupt context.
2337 *
2338 * In future, when we do a directed EOI (combined with cpu EOI broadcast
2339 * suppression) to the IO-APIC, level triggered irq migration will also be
2340 * as simple as edge triggered migration and we can do the irq migration
2341 * with a simple atomic update to IO-APIC RTE.
2342 */
e7986739
MT		 2343static void
2344migrate_ioapic_irq_desc(struct irq_desc *desc, const struct cpumask *mask)
497c9a19 2345{
54168ed7 2346 struct irq_cfg *cfg;
54168ed7
IM		 2347 struct irte irte;
2348 int modify_ioapic_rte;
2349 unsigned int dest;
2350 unsigned long flags;
3145e941 2351 unsigned int irq;
497c9a19 2352
22f65d31 2353 if (!cpumask_intersects(mask, cpu_online_mask))
497c9a19
YL		 2354 return;
2355
3145e941 2356 irq = desc->irq;
54168ed7
IM		 2357 if (get_irte(irq, &irte))
2358 return;
497c9a19 2359
3145e941
YL		 2360 cfg = desc->chip_data;
2361 if (assign_irq_vector(irq, cfg, mask))
54168ed7
IM		 2362 return;
2363
3145e941
YL		 2364 set_extra_move_desc(desc, mask);
2365
22f65d31 2366 dest = cpu_mask_to_apicid_and(cfg->domain, mask);
54168ed7 2367
54168ed7
IM		 2368 modify_ioapic_rte = desc->status & IRQ_LEVEL;
2369 if (modify_ioapic_rte) {
2370 spin_lock_irqsave(&ioapic_lock, flags);
3145e941 2371 __target_IO_APIC_irq(irq, dest, cfg);
54168ed7
IM		 2372 spin_unlock_irqrestore(&ioapic_lock, flags);
2373 }
2374
2375 irte.vector = cfg->vector;
2376 irte.dest_id = IRTE_DEST(dest);
2377
2378 /*
2379 * Modified the IRTE and flushes the Interrupt entry cache.
2380 */
2381 modify_irte(irq, &irte);
2382
22f65d31
MT		 2383 if (cfg->move_in_progress)
2384 send_cleanup_vector(cfg);
54168ed7 2385
22f65d31 2386 cpumask_copy(&desc->affinity, mask);
54168ed7
IM		 2387}
2388
3145e941 2389static int migrate_irq_remapped_level_desc(struct irq_desc *desc)
54168ed7
IM		 2390{
2391 int ret = -1;
3145e941 2392 struct irq_cfg *cfg = desc->chip_data;
54168ed7 2393
3145e941 2394 mask_IO_APIC_irq_desc(desc);
54168ed7 2395
3145e941 2396 if (io_apic_level_ack_pending(cfg)) {
54168ed7 2397 /*
d6c88a50 2398 * Interrupt in progress. Migrating irq now will change the
54168ed7
IM		 2399 * vector information in the IO-APIC RTE and that will confuse
2400 * the EOI broadcast performed by cpu.
2401 * So, delay the irq migration to the next instance.
2402 */
2403 schedule_delayed_work(&ir_migration_work, 1);
2404 goto unmask;
2405 }
2406
2407 /* everthing is clear. we have right of way */
e7986739 2408 migrate_ioapic_irq_desc(desc, &desc->pending_mask);
54168ed7
IM		 2409
2410 ret = 0;
2411 desc->status &= ~IRQ_MOVE_PENDING;
22f65d31 2412 cpumask_clear(&desc->pending_mask);
54168ed7
IM		 2413
2414unmask:
3145e941
YL		 2415 unmask_IO_APIC_irq_desc(desc);
2416
54168ed7
IM		 2417 return ret;
2418}
2419
2420static void ir_irq_migration(struct work_struct *work)
2421{
2422 unsigned int irq;
2423 struct irq_desc *desc;
2424
2425 for_each_irq_desc(irq, desc) {
2426 if (desc->status & IRQ_MOVE_PENDING) {
2427 unsigned long flags;
2428
2429 spin_lock_irqsave(&desc->lock, flags);
2430 if (!desc->chip->set_affinity ||
2431 !(desc->status & IRQ_MOVE_PENDING)) {
2432 desc->status &= ~IRQ_MOVE_PENDING;
2433 spin_unlock_irqrestore(&desc->lock, flags);
2434 continue;
2435 }
2436
0de26520 2437 desc->chip->set_affinity(irq, &desc->pending_mask);
54168ed7
IM		 2438 spin_unlock_irqrestore(&desc->lock, flags);
2439 }
2440 }
2441}
2442
2443/*
2444 * Migrates the IRQ destination in the process context.
2445 */
968ea6d8
RR		 2446static void set_ir_ioapic_affinity_irq_desc(struct irq_desc *desc,
2447 const struct cpumask *mask)
54168ed7 2448{
54168ed7
IM		 2449 if (desc->status & IRQ_LEVEL) {
2450 desc->status |= IRQ_MOVE_PENDING;
0de26520 2451 cpumask_copy(&desc->pending_mask, mask);
3145e941 2452 migrate_irq_remapped_level_desc(desc);
54168ed7
IM		 2453 return;
2454 }
2455
3145e941
YL		 2456 migrate_ioapic_irq_desc(desc, mask);
2457}
968ea6d8
RR		 2458static void set_ir_ioapic_affinity_irq(unsigned int irq,
2459 const struct cpumask *mask)
3145e941
YL		 2460{
2461 struct irq_desc *desc = irq_to_desc(irq);
2462
2463 set_ir_ioapic_affinity_irq_desc(desc, mask);
54168ed7
IM		 2464}
2465#endif
2466
2467asmlinkage void smp_irq_move_cleanup_interrupt(void)
2468{
2469 unsigned vector, me;
8f2466f4 2470
54168ed7 2471 ack_APIC_irq();
54168ed7 2472 exit_idle();
54168ed7
IM		 2473 irq_enter();
2474
2475 me = smp_processor_id();
2476 for (vector = FIRST_EXTERNAL_VECTOR; vector < NR_VECTORS; vector++) {
2477 unsigned int irq;
2478 struct irq_desc *desc;
2479 struct irq_cfg *cfg;
2480 irq = __get_cpu_var(vector_irq)[vector];
2481
0b8f1efa
YL		 2482 if (irq == -1)
2483 continue;
2484
54168ed7
IM		 2485 desc = irq_to_desc(irq);
2486 if (!desc)
2487 continue;
2488
2489 cfg = irq_cfg(irq);
2490 spin_lock(&desc->lock);
2491 if (!cfg->move_cleanup_count)
2492 goto unlock;
2493
22f65d31 2494 if (vector == cfg->vector && cpumask_test_cpu(me, cfg->domain))
54168ed7
IM		 2495 goto unlock;
2496
2497 __get_cpu_var(vector_irq)[vector] = -1;
2498 cfg->move_cleanup_count--;
2499unlock:
2500 spin_unlock(&desc->lock);
2501 }
2502
2503 irq_exit();
2504}
2505
3145e941 2506static void irq_complete_move(struct irq_desc **descp)
54168ed7 2507{
3145e941
YL		 2508 struct irq_desc *desc = *descp;
2509 struct irq_cfg *cfg = desc->chip_data;
54168ed7
IM		 2510 unsigned vector, me;
2511
48a1b10a
YL		 2512 if (likely(!cfg->move_in_progress)) {
2513#ifdef CONFIG_NUMA_MIGRATE_IRQ_DESC
2514 if (likely(!cfg->move_desc_pending))
2515 return;
2516
b9098957 2517 /* domain has not changed, but affinity did */
48a1b10a
YL		 2518 me = smp_processor_id();
2519 if (cpu_isset(me, desc->affinity)) {
2520 *descp = desc = move_irq_desc(desc, me);
2521 /* get the new one */
2522 cfg = desc->chip_data;
2523 cfg->move_desc_pending = 0;
2524 }
2525#endif
54168ed7 2526 return;
48a1b10a 2527 }
54168ed7
IM		 2528
2529 vector = ~get_irq_regs()->orig_ax;
2530 me = smp_processor_id();
10b888d6
YL		 2531
2532 if (vector == cfg->vector && cpumask_test_cpu(me, cfg->domain)) {
48a1b10a
YL		 2533#ifdef CONFIG_NUMA_MIGRATE_IRQ_DESC
2534 *descp = desc = move_irq_desc(desc, me);
2535 /* get the new one */
2536 cfg = desc->chip_data;
2537#endif
22f65d31 2538 send_cleanup_vector(cfg);
10b888d6 2539 }
497c9a19
YL		 2540}
2541#else
3145e941 2542static inline void irq_complete_move(struct irq_desc **descp) {}
497c9a19 2543#endif
3145e941 2544
54168ed7
IM		 2545#ifdef CONFIG_INTR_REMAP
2546static void ack_x2apic_level(unsigned int irq)
2547{
2548 ack_x2APIC_irq();
2549}
2550
2551static void ack_x2apic_edge(unsigned int irq)
2552{
2553 ack_x2APIC_irq();
2554}
3145e941 2555
54168ed7 2556#endif
497c9a19 2557
1d025192
YL		 2558static void ack_apic_edge(unsigned int irq)
2559{
3145e941
YL		 2560 struct irq_desc *desc = irq_to_desc(irq);
2561
2562 irq_complete_move(&desc);
1d025192
YL		 2563 move_native_irq(irq);
2564 ack_APIC_irq();
2565}
2566
3eb2cce8 2567atomic_t irq_mis_count;
3eb2cce8 2568
047c8fdb
YL		 2569static void ack_apic_level(unsigned int irq)
2570{
3145e941
YL		 2571 struct irq_desc *desc = irq_to_desc(irq);
2572
3eb2cce8
YL		 2573#ifdef CONFIG_X86_32
2574 unsigned long v;
2575 int i;
2576#endif
3145e941 2577 struct irq_cfg *cfg;
54168ed7 2578 int do_unmask_irq = 0;
047c8fdb 2579
3145e941 2580 irq_complete_move(&desc);
047c8fdb 2581#ifdef CONFIG_GENERIC_PENDING_IRQ
54168ed7 2582 /* If we are moving the irq we need to mask it */
3145e941 2583 if (unlikely(desc->status & IRQ_MOVE_PENDING)) {
54168ed7 2584 do_unmask_irq = 1;
3145e941 2585 mask_IO_APIC_irq_desc(desc);
54168ed7 2586 }
047c8fdb
YL		 2587#endif
2588
3eb2cce8
YL		 2589#ifdef CONFIG_X86_32
2590 /*
2591 * It appears there is an erratum which affects at least version 0x11
2592 * of I/O APIC (that's the 82093AA and cores integrated into various
2593 * chipsets). Under certain conditions a level-triggered interrupt is
2594 * erroneously delivered as edge-triggered one but the respective IRR
2595 * bit gets set nevertheless. As a result the I/O unit expects an EOI
2596 * message but it will never arrive and further interrupts are blocked
2597 * from the source. The exact reason is so far unknown, but the
2598 * phenomenon was observed when two consecutive interrupt requests
2599 * from a given source get delivered to the same CPU and the source is
2600 * temporarily disabled in between.
2601 *
2602 * A workaround is to simulate an EOI message manually. We achieve it
2603 * by setting the trigger mode to edge and then to level when the edge
2604 * trigger mode gets detected in the TMR of a local APIC for a
2605 * level-triggered interrupt. We mask the source for the time of the
2606 * operation to prevent an edge-triggered interrupt escaping meanwhile.
2607 * The idea is from Manfred Spraul. --macro
2608 */
3145e941
YL		 2609 cfg = desc->chip_data;
2610 i = cfg->vector;
3eb2cce8
YL		 2611
2612 v = apic_read(APIC_TMR + ((i & ~0x1f) >> 1));
2613#endif
2614
54168ed7
IM		 2615 /*
2616 * We must acknowledge the irq before we move it or the acknowledge will
2617 * not propagate properly.
2618 */
2619 ack_APIC_irq();
2620
2621 /* Now we can move and renable the irq */
2622 if (unlikely(do_unmask_irq)) {
2623 /* Only migrate the irq if the ack has been received.
2624 *
2625 * On rare occasions the broadcast level triggered ack gets
2626 * delayed going to ioapics, and if we reprogram the
2627 * vector while Remote IRR is still set the irq will never
2628 * fire again.
2629 *
2630 * To prevent this scenario we read the Remote IRR bit
2631 * of the ioapic. This has two effects.
2632 * - On any sane system the read of the ioapic will
2633 * flush writes (and acks) going to the ioapic from
2634 * this cpu.
2635 * - We get to see if the ACK has actually been delivered.
2636 *
2637 * Based on failed experiments of reprogramming the
2638 * ioapic entry from outside of irq context starting
2639 * with masking the ioapic entry and then polling until
2640 * Remote IRR was clear before reprogramming the
2641 * ioapic I don't trust the Remote IRR bit to be
2642 * completey accurate.
2643 *
2644 * However there appears to be no other way to plug
2645 * this race, so if the Remote IRR bit is not
2646 * accurate and is causing problems then it is a hardware bug
2647 * and you can go talk to the chipset vendor about it.
2648 */
3145e941
YL		 2649 cfg = desc->chip_data;
2650 if (!io_apic_level_ack_pending(cfg))
54168ed7 2651 move_masked_irq(irq);
3145e941 2652 unmask_IO_APIC_irq_desc(desc);
54168ed7 2653 }
1d025192 2654
3eb2cce8 2655#ifdef CONFIG_X86_32
1d025192
YL		 2656 if (!(v & (1 << (i & 0x1f)))) {
2657 atomic_inc(&irq_mis_count);
2658 spin_lock(&ioapic_lock);
3145e941
YL		 2659 __mask_and_edge_IO_APIC_irq(cfg);
2660 __unmask_and_level_IO_APIC_irq(cfg);
1d025192
YL		 2661 spin_unlock(&ioapic_lock);
2662 }
047c8fdb 2663#endif
3eb2cce8 2664}
1d025192 2665
f5b9ed7a 2666static struct irq_chip ioapic_chip __read_mostly = {
d6c88a50
TG		 2667 .name = "IO-APIC",
2668 .startup = startup_ioapic_irq,
2669 .mask = mask_IO_APIC_irq,
2670 .unmask = unmask_IO_APIC_irq,
2671 .ack = ack_apic_edge,
2672 .eoi = ack_apic_level,
54d5d424 2673#ifdef CONFIG_SMP
d6c88a50 2674 .set_affinity = set_ioapic_affinity_irq,
54d5d424 2675#endif
ace80ab7 2676 .retrigger = ioapic_retrigger_irq,
1da177e4
LT		 2677};
2678
54168ed7
IM		 2679#ifdef CONFIG_INTR_REMAP
2680static struct irq_chip ir_ioapic_chip __read_mostly = {
d6c88a50
TG		 2681 .name = "IR-IO-APIC",
2682 .startup = startup_ioapic_irq,
2683 .mask = mask_IO_APIC_irq,
2684 .unmask = unmask_IO_APIC_irq,
2685 .ack = ack_x2apic_edge,
2686 .eoi = ack_x2apic_level,
54168ed7 2687#ifdef CONFIG_SMP
d6c88a50 2688 .set_affinity = set_ir_ioapic_affinity_irq,
54168ed7
IM		 2689#endif
2690 .retrigger = ioapic_retrigger_irq,
2691};
2692#endif
1da177e4
LT		 2693
2694static inline void init_IO_APIC_traps(void)
2695{
2696 int irq;
08678b08 2697 struct irq_desc *desc;
da51a821 2698 struct irq_cfg *cfg;
1da177e4
LT		 2699
2700 /*
2701 * NOTE! The local APIC isn't very good at handling
2702 * multiple interrupts at the same interrupt level.
2703 * As the interrupt level is determined by taking the
2704 * vector number and shifting that right by 4, we
2705 * want to spread these out a bit so that they don't
2706 * all fall in the same interrupt level.
2707 *
2708 * Also, we've got to be careful not to trash gate
2709 * 0x80, because int 0x80 is hm, kind of importantish. ;)
2710 */
0b8f1efa 2711 for_each_irq_desc(irq, desc) {
0b8f1efa
YL		 2712 cfg = desc->chip_data;
2713 if (IO_APIC_IRQ(irq) && cfg && !cfg->vector) {
1da177e4
LT		 2714 /*
2715 * Hmm.. We don't have an entry for this,
2716 * so default to an old-fashioned 8259
2717 * interrupt if we can..
2718 */
99d093d1 2719 if (irq < NR_IRQS_LEGACY)
1da177e4 2720 make_8259A_irq(irq);
0b8f1efa 2721 else
1da177e4 2722 /* Strange. Oh, well.. */
08678b08 2723 desc->chip = &no_irq_chip;
1da177e4
LT		 2724 }
2725 }
2726}
2727
f5b9ed7a
IM		 2728/*
2729 * The local APIC irq-chip implementation:
2730 */
1da177e4 2731
36062448 2732static void mask_lapic_irq(unsigned int irq)
1da177e4
LT		 2733{
2734 unsigned long v;
2735
2736 v = apic_read(APIC_LVT0);
593f4a78 2737 apic_write(APIC_LVT0, v | APIC_LVT_MASKED);
1da177e4
LT		 2738}
2739
36062448 2740static void unmask_lapic_irq(unsigned int irq)
1da177e4 2741{
f5b9ed7a 2742 unsigned long v;
1da177e4 2743
f5b9ed7a 2744 v = apic_read(APIC_LVT0);
593f4a78 2745 apic_write(APIC_LVT0, v & ~APIC_LVT_MASKED);
f5b9ed7a 2746}
1da177e4 2747
3145e941 2748static void ack_lapic_irq(unsigned int irq)
1d025192
YL		 2749{
2750 ack_APIC_irq();
2751}
2752
f5b9ed7a 2753static struct irq_chip lapic_chip __read_mostly = {
9a1c6192 2754 .name = "local-APIC",
f5b9ed7a
IM		 2755 .mask = mask_lapic_irq,
2756 .unmask = unmask_lapic_irq,
c88ac1df 2757 .ack = ack_lapic_irq,
1da177e4
LT		 2758};
2759
3145e941 2760static void lapic_register_intr(int irq, struct irq_desc *desc)
c88ac1df 2761{
08678b08 2762 desc->status &= ~IRQ_LEVEL;
c88ac1df
MR		 2763 set_irq_chip_and_handler_name(irq, &lapic_chip, handle_edge_irq,
2764 "edge");
c88ac1df
MR		 2765}
2766
e9427101 2767static void __init setup_nmi(void)
1da177e4
LT		 2768{
2769 /*
36062448 2770 * Dirty trick to enable the NMI watchdog ...
1da177e4
LT		 2771 * We put the 8259A master into AEOI mode and
2772 * unmask on all local APICs LVT0 as NMI.
2773 *
2774 * The idea to use the 8259A in AEOI mode ('8259A Virtual Wire')
2775 * is from Maciej W. Rozycki - so we do not have to EOI from
2776 * the NMI handler or the timer interrupt.
36062448 2777 */
1da177e4
LT		 2778 apic_printk(APIC_VERBOSE, KERN_INFO "activating NMI Watchdog ...");
2779
e9427101 2780 enable_NMI_through_LVT0();
1da177e4
LT		 2781
2782 apic_printk(APIC_VERBOSE, " done.\n");
2783}
2784
2785/*
2786 * This looks a bit hackish but it's about the only one way of sending
2787 * a few INTA cycles to 8259As and any associated glue logic. ICR does
2788 * not support the ExtINT mode, unfortunately. We need to send these
2789 * cycles as some i82489DX-based boards have glue logic that keeps the
2790 * 8259A interrupt line asserted until INTA. --macro
2791 */
28acf285 2792static inline void __init unlock_ExtINT_logic(void)
1da177e4 2793{
fcfd636a 2794 int apic, pin, i;
1da177e4
LT		 2795 struct IO_APIC_route_entry entry0, entry1;
2796 unsigned char save_control, save_freq_select;
1da177e4 2797
fcfd636a 2798 pin = find_isa_irq_pin(8, mp_INT);
956fb531
AB		 2799 if (pin == -1) {
2800 WARN_ON_ONCE(1);
2801 return;
2802 }
fcfd636a 2803 apic = find_isa_irq_apic(8, mp_INT);
956fb531
AB		 2804 if (apic == -1) {
2805 WARN_ON_ONCE(1);
1da177e4 2806 return;
956fb531 2807 }
1da177e4 2808
cf4c6a2f 2809 entry0 = ioapic_read_entry(apic, pin);
fcfd636a 2810 clear_IO_APIC_pin(apic, pin);
1da177e4
LT		 2811
2812 memset(&entry1, 0, sizeof(entry1));
2813
2814 entry1.dest_mode = 0; /* physical delivery */
2815 entry1.mask = 0; /* unmask IRQ now */
d83e94ac 2816 entry1.dest = hard_smp_processor_id();
1da177e4
LT		 2817 entry1.delivery_mode = dest_ExtINT;
2818 entry1.polarity = entry0.polarity;
2819 entry1.trigger = 0;
2820 entry1.vector = 0;
2821
cf4c6a2f 2822 ioapic_write_entry(apic, pin, entry1);
1da177e4
LT		 2823
2824 save_control = CMOS_READ(RTC_CONTROL);
2825 save_freq_select = CMOS_READ(RTC_FREQ_SELECT);
2826 CMOS_WRITE((save_freq_select & ~RTC_RATE_SELECT) | 0x6,
2827 RTC_FREQ_SELECT);
2828 CMOS_WRITE(save_control | RTC_PIE, RTC_CONTROL);
2829
2830 i = 100;
2831 while (i-- > 0) {
2832 mdelay(10);
2833 if ((CMOS_READ(RTC_INTR_FLAGS) & RTC_PF) == RTC_PF)
2834 i -= 10;
2835 }
2836
2837 CMOS_WRITE(save_control, RTC_CONTROL);
2838 CMOS_WRITE(save_freq_select, RTC_FREQ_SELECT);
fcfd636a 2839 clear_IO_APIC_pin(apic, pin);
1da177e4 2840
cf4c6a2f 2841 ioapic_write_entry(apic, pin, entry0);
1da177e4
LT		 2842}
2843
efa2559f 2844static int disable_timer_pin_1 __initdata;
047c8fdb 2845/* Actually the next is obsolete, but keep it for paranoid reasons -AK */
54168ed7 2846static int __init disable_timer_pin_setup(char *arg)
efa2559f
YL		 2847{
2848 disable_timer_pin_1 = 1;
2849 return 0;
2850}
54168ed7 2851early_param("disable_timer_pin_1", disable_timer_pin_setup);
efa2559f
YL		 2852
2853int timer_through_8259 __initdata;
2854
1da177e4
LT		 2855/*
2856 * This code may look a bit paranoid, but it's supposed to cooperate with
2857 * a wide range of boards and BIOS bugs. Fortunately only the timer IRQ
2858 * is so screwy. Thanks to Brian Perkins for testing/hacking this beast
2859 * fanatically on his truly buggy board.
54168ed7
IM		 2860 *
2861 * FIXME: really need to revamp this for all platforms.
1da177e4 2862 */
8542b200 2863static inline void __init check_timer(void)
1da177e4 2864{
3145e941
YL		 2865 struct irq_desc *desc = irq_to_desc(0);
2866 struct irq_cfg *cfg = desc->chip_data;
2867 int cpu = boot_cpu_id;
fcfd636a 2868 int apic1, pin1, apic2, pin2;
4aae0702 2869 unsigned long flags;
047c8fdb
YL		 2870 unsigned int ver;
2871 int no_pin1 = 0;
4aae0702
IM		 2872
2873 local_irq_save(flags);
d4d25dec 2874
d6c88a50
TG		 2875 ver = apic_read(APIC_LVR);
2876 ver = GET_APIC_VERSION(ver);
6e908947 2877
1da177e4
LT		 2878 /*
2879 * get/set the timer IRQ vector:
2880 */
2881 disable_8259A_irq(0);
3145e941 2882 assign_irq_vector(0, cfg, TARGET_CPUS);
1da177e4
LT		 2883
2884 /*
d11d5794
MR		 2885 * As IRQ0 is to be enabled in the 8259A, the virtual
2886 * wire has to be disabled in the local APIC. Also
2887 * timer interrupts need to be acknowledged manually in
2888 * the 8259A for the i82489DX when using the NMI
2889 * watchdog as that APIC treats NMIs as level-triggered.
2890 * The AEOI mode will finish them in the 8259A
2891 * automatically.
1da177e4 2892 */
593f4a78 2893 apic_write(APIC_LVT0, APIC_LVT_MASKED | APIC_DM_EXTINT);
1da177e4 2894 init_8259A(1);
54168ed7 2895#ifdef CONFIG_X86_32
d11d5794 2896 timer_ack = (nmi_watchdog == NMI_IO_APIC && !APIC_INTEGRATED(ver));
54168ed7 2897#endif
1da177e4 2898
fcfd636a
EB		 2899 pin1 = find_isa_irq_pin(0, mp_INT);
2900 apic1 = find_isa_irq_apic(0, mp_INT);
2901 pin2 = ioapic_i8259.pin;
2902 apic2 = ioapic_i8259.apic;
1da177e4 2903
49a66a0b
MR		 2904 apic_printk(APIC_QUIET, KERN_INFO "..TIMER: vector=0x%02X "
2905 "apic1=%d pin1=%d apic2=%d pin2=%d\n",
497c9a19 2906 cfg->vector, apic1, pin1, apic2, pin2);
1da177e4 2907
691874fa
MR		 2908 /*
2909 * Some BIOS writers are clueless and report the ExtINTA
2910 * I/O APIC input from the cascaded 8259A as the timer
2911 * interrupt input. So just in case, if only one pin
2912 * was found above, try it both directly and through the
2913 * 8259A.
2914 */
2915 if (pin1 == -1) {
54168ed7
IM		 2916#ifdef CONFIG_INTR_REMAP
2917 if (intr_remapping_enabled)
2918 panic("BIOS bug: timer not connected to IO-APIC");
2919#endif
691874fa
MR		 2920 pin1 = pin2;
2921 apic1 = apic2;
2922 no_pin1 = 1;
2923 } else if (pin2 == -1) {
2924 pin2 = pin1;
2925 apic2 = apic1;
2926 }
2927
1da177e4
LT		 2928 if (pin1 != -1) {
2929 /*
2930 * Ok, does IRQ0 through the IOAPIC work?
2931 */
691874fa 2932 if (no_pin1) {
3145e941 2933 add_pin_to_irq_cpu(cfg, cpu, apic1, pin1);
497c9a19 2934 setup_timer_IRQ0_pin(apic1, pin1, cfg->vector);
691874fa 2935 }
3145e941 2936 unmask_IO_APIC_irq_desc(desc);
1da177e4
LT		 2937 if (timer_irq_works()) {
2938 if (nmi_watchdog == NMI_IO_APIC) {
1da177e4
LT		 2939 setup_nmi();
2940 enable_8259A_irq(0);
1da177e4 2941 }
66759a01
CE		 2942 if (disable_timer_pin_1 > 0)
2943 clear_IO_APIC_pin(0, pin1);
4aae0702 2944 goto out;
1da177e4 2945 }
54168ed7
IM		 2946#ifdef CONFIG_INTR_REMAP
2947 if (intr_remapping_enabled)
2948 panic("timer doesn't work through Interrupt-remapped IO-APIC");
2949#endif
fcfd636a 2950 clear_IO_APIC_pin(apic1, pin1);
691874fa 2951 if (!no_pin1)
49a66a0b
MR		 2952 apic_printk(APIC_QUIET, KERN_ERR "..MP-BIOS bug: "
2953 "8254 timer not connected to IO-APIC\n");
1da177e4 2954
49a66a0b
MR		 2955 apic_printk(APIC_QUIET, KERN_INFO "...trying to set up timer "
2956 "(IRQ0) through the 8259A ...\n");
2957 apic_printk(APIC_QUIET, KERN_INFO
2958 "..... (found apic %d pin %d) ...\n", apic2, pin2);
1da177e4
LT		 2959 /*
2960 * legacy devices should be connected to IO APIC #0
2961 */
3145e941 2962 replace_pin_at_irq_cpu(cfg, cpu, apic1, pin1, apic2, pin2);
497c9a19 2963 setup_timer_IRQ0_pin(apic2, pin2, cfg->vector);
3145e941 2964 unmask_IO_APIC_irq_desc(desc);
ecd29476 2965 enable_8259A_irq(0);
1da177e4 2966 if (timer_irq_works()) {
49a66a0b 2967 apic_printk(APIC_QUIET, KERN_INFO "....... works.\n");
35542c5e 2968 timer_through_8259 = 1;
1da177e4 2969 if (nmi_watchdog == NMI_IO_APIC) {
60134ebe 2970 disable_8259A_irq(0);
1da177e4 2971 setup_nmi();
60134ebe 2972 enable_8259A_irq(0);
1da177e4 2973 }
4aae0702 2974 goto out;
1da177e4
LT		 2975 }
2976 /*
2977 * Cleanup, just in case ...
2978 */
ecd29476 2979 disable_8259A_irq(0);
fcfd636a 2980 clear_IO_APIC_pin(apic2, pin2);
49a66a0b 2981 apic_printk(APIC_QUIET, KERN_INFO "....... failed.\n");
1da177e4 2982 }
1da177e4
LT		 2983
2984 if (nmi_watchdog == NMI_IO_APIC) {
49a66a0b
MR		 2985 apic_printk(APIC_QUIET, KERN_WARNING "timer doesn't work "
2986 "through the IO-APIC - disabling NMI Watchdog!\n");
067fa0ff 2987 nmi_watchdog = NMI_NONE;
1da177e4 2988 }
54168ed7 2989#ifdef CONFIG_X86_32
d11d5794 2990 timer_ack = 0;
54168ed7 2991#endif
1da177e4 2992
49a66a0b
MR		 2993 apic_printk(APIC_QUIET, KERN_INFO
2994 "...trying to set up timer as Virtual Wire IRQ...\n");
1da177e4 2995
3145e941 2996 lapic_register_intr(0, desc);
497c9a19 2997 apic_write(APIC_LVT0, APIC_DM_FIXED | cfg->vector); /* Fixed mode */
1da177e4
LT		 2998 enable_8259A_irq(0);
2999
3000 if (timer_irq_works()) {
49a66a0b 3001 apic_printk(APIC_QUIET, KERN_INFO "..... works.\n");
4aae0702 3002 goto out;
1da177e4 3003 }
e67465f1 3004 disable_8259A_irq(0);
497c9a19 3005 apic_write(APIC_LVT0, APIC_LVT_MASKED | APIC_DM_FIXED | cfg->vector);
49a66a0b 3006 apic_printk(APIC_QUIET, KERN_INFO "..... failed.\n");
1da177e4 3007
49a66a0b
MR		 3008 apic_printk(APIC_QUIET, KERN_INFO
3009 "...trying to set up timer as ExtINT IRQ...\n");
1da177e4 3010
1da177e4
LT		 3011 init_8259A(0);
3012 make_8259A_irq(0);
593f4a78 3013 apic_write(APIC_LVT0, APIC_DM_EXTINT);
1da177e4
LT		 3014
3015 unlock_ExtINT_logic();
3016
3017 if (timer_irq_works()) {
49a66a0b 3018 apic_printk(APIC_QUIET, KERN_INFO "..... works.\n");
4aae0702 3019 goto out;
1da177e4 3020 }
49a66a0b 3021 apic_printk(APIC_QUIET, KERN_INFO "..... failed :(.\n");
1da177e4 3022 panic("IO-APIC + timer doesn't work! Boot with apic=debug and send a "
49a66a0b 3023 "report. Then try booting with the 'noapic' option.\n");
4aae0702
IM		 3024out:
3025 local_irq_restore(flags);
1da177e4
LT		 3026}
3027
3028/*
af174783
MR		 3029 * Traditionally ISA IRQ2 is the cascade IRQ, and is not available
3030 * to devices. However there may be an I/O APIC pin available for
3031 * this interrupt regardless. The pin may be left unconnected, but
3032 * typically it will be reused as an ExtINT cascade interrupt for
3033 * the master 8259A. In the MPS case such a pin will normally be
3034 * reported as an ExtINT interrupt in the MP table. With ACPI
3035 * there is no provision for ExtINT interrupts, and in the absence
3036 * of an override it would be treated as an ordinary ISA I/O APIC
3037 * interrupt, that is edge-triggered and unmasked by default. We
3038 * used to do this, but it caused problems on some systems because
3039 * of the NMI watchdog and sometimes IRQ0 of the 8254 timer using
3040 * the same ExtINT cascade interrupt to drive the local APIC of the
3041 * bootstrap processor. Therefore we refrain from routing IRQ2 to
3042 * the I/O APIC in all cases now. No actual device should request
3043 * it anyway. --macro
1da177e4
LT		 3044 */
3045#define PIC_IRQS (1 << PIC_CASCADE_IR)
3046
3047void __init setup_IO_APIC(void)
3048{
54168ed7
IM		 3049
3050#ifdef CONFIG_X86_32
1da177e4 3051 enable_IO_APIC();
54168ed7
IM		 3052#else
3053 /*
3054 * calling enable_IO_APIC() is moved to setup_local_APIC for BP
3055 */
3056#endif
1da177e4 3057
af174783 3058 io_apic_irqs = ~PIC_IRQS;
1da177e4 3059
54168ed7 3060 apic_printk(APIC_VERBOSE, "ENABLING IO-APIC IRQs\n");
d6c88a50 3061 /*
54168ed7
IM		 3062 * Set up IO-APIC IRQ routing.
3063 */
3064#ifdef CONFIG_X86_32
d6c88a50
TG		 3065 if (!acpi_ioapic)
3066 setup_ioapic_ids_from_mpc();
54168ed7 3067#endif
1da177e4
LT		 3068 sync_Arb_IDs();
3069 setup_IO_APIC_irqs();
3070 init_IO_APIC_traps();
1e4c85f9 3071 check_timer();
1da177e4
LT		 3072}
3073
3074/*
54168ed7
IM		 3075 * Called after all the initialization is done. If we didnt find any
3076 * APIC bugs then we can allow the modify fast path
1da177e4 3077 */
36062448 3078
1da177e4
LT		 3079static int __init io_apic_bug_finalize(void)
3080{
d6c88a50
TG		 3081 if (sis_apic_bug == -1)
3082 sis_apic_bug = 0;
3083 return 0;
1da177e4
LT		 3084}
3085
3086late_initcall(io_apic_bug_finalize);
3087
3088struct sysfs_ioapic_data {
3089 struct sys_device dev;
3090 struct IO_APIC_route_entry entry[0];
3091};
54168ed7 3092static struct sysfs_ioapic_data * mp_ioapic_data[MAX_IO_APICS];
1da177e4 3093
438510f6 3094static int ioapic_suspend(struct sys_device *dev, pm_message_t state)
1da177e4
LT		 3095{
3096 struct IO_APIC_route_entry *entry;
3097 struct sysfs_ioapic_data *data;
1da177e4 3098 int i;
36062448 3099
1da177e4
LT		 3100 data = container_of(dev, struct sysfs_ioapic_data, dev);
3101 entry = data->entry;
54168ed7
IM		 3102 for (i = 0; i < nr_ioapic_registers[dev->id]; i ++, entry ++ )
3103 *entry = ioapic_read_entry(dev->id, i);
1da177e4
LT		 3104
3105 return 0;
3106}
3107
3108static int ioapic_resume(struct sys_device *dev)
3109{
3110 struct IO_APIC_route_entry *entry;
3111 struct sysfs_ioapic_data *data;
3112 unsigned long flags;
3113 union IO_APIC_reg_00 reg_00;
3114 int i;
36062448 3115
1da177e4
LT		 3116 data = container_of(dev, struct sysfs_ioapic_data, dev);
3117 entry = data->entry;
3118
3119 spin_lock_irqsave(&ioapic_lock, flags);
3120 reg_00.raw = io_apic_read(dev->id, 0);
ec2cd0a2
AS		 3121 if (reg_00.bits.ID != mp_ioapics[dev->id].mp_apicid) {
3122 reg_00.bits.ID = mp_ioapics[dev->id].mp_apicid;
1da177e4
LT		 3123 io_apic_write(dev->id, 0, reg_00.raw);
3124 }
1da177e4 3125 spin_unlock_irqrestore(&ioapic_lock, flags);
36062448 3126 for (i = 0; i < nr_ioapic_registers[dev->id]; i++)
cf4c6a2f 3127 ioapic_write_entry(dev->id, i, entry[i]);
1da177e4
LT		 3128
3129 return 0;
3130}
3131
3132static struct sysdev_class ioapic_sysdev_class = {
af5ca3f4 3133 .name = "ioapic",
1da177e4
LT		 3134 .suspend = ioapic_suspend,
3135 .resume = ioapic_resume,
3136};
3137
3138static int __init ioapic_init_sysfs(void)
3139{
54168ed7
IM		 3140 struct sys_device * dev;
3141 int i, size, error;
1da177e4
LT		 3142
3143 error = sysdev_class_register(&ioapic_sysdev_class);
3144 if (error)
3145 return error;
3146
54168ed7 3147 for (i = 0; i < nr_ioapics; i++ ) {
36062448 3148 size = sizeof(struct sys_device) + nr_ioapic_registers[i]
1da177e4 3149 * sizeof(struct IO_APIC_route_entry);
25556c16 3150 mp_ioapic_data[i] = kzalloc(size, GFP_KERNEL);
1da177e4
LT		 3151 if (!mp_ioapic_data[i]) {
3152 printk(KERN_ERR "Can't suspend/resume IOAPIC %d\n", i);
3153 continue;
3154 }
1da177e4 3155 dev = &mp_ioapic_data[i]->dev;
36062448 3156 dev->id = i;
1da177e4
LT		 3157 dev->cls = &ioapic_sysdev_class;
3158 error = sysdev_register(dev);
3159 if (error) {
3160 kfree(mp_ioapic_data[i]);
3161 mp_ioapic_data[i] = NULL;
3162 printk(KERN_ERR "Can't suspend/resume IOAPIC %d\n", i);
3163 continue;
3164 }
3165 }
3166
3167 return 0;
3168}
3169
3170device_initcall(ioapic_init_sysfs);
3171
3fc471ed 3172/*
95d77884 3173 * Dynamic irq allocate and deallocation
3fc471ed 3174 */
199751d7 3175unsigned int create_irq_nr(unsigned int irq_want)
3fc471ed 3176{
ace80ab7 3177 /* Allocate an unused irq */
54168ed7
IM		 3178 unsigned int irq;
3179 unsigned int new;
3fc471ed 3180 unsigned long flags;
0b8f1efa
YL		 3181 struct irq_cfg *cfg_new = NULL;
3182 int cpu = boot_cpu_id;
3183 struct irq_desc *desc_new = NULL;
199751d7
YL		 3184
3185 irq = 0;
ace80ab7 3186 spin_lock_irqsave(&vector_lock, flags);
be5d5350 3187 for (new = irq_want; new < NR_IRQS; new++) {
ace80ab7
EB		 3188 if (platform_legacy_irq(new))
3189 continue;
0b8f1efa
YL		 3190
3191 desc_new = irq_to_desc_alloc_cpu(new, cpu);
3192 if (!desc_new) {
3193 printk(KERN_INFO "can not get irq_desc for %d\n", new);
ace80ab7 3194 continue;
0b8f1efa
YL		 3195 }
3196 cfg_new = desc_new->chip_data;
3197
3198 if (cfg_new->vector != 0)
ace80ab7 3199 continue;
3145e941 3200 if (__assign_irq_vector(new, cfg_new, TARGET_CPUS) == 0)
ace80ab7
EB		 3201 irq = new;
3202 break;
3203 }
3204 spin_unlock_irqrestore(&vector_lock, flags);
3fc471ed 3205
199751d7 3206 if (irq > 0) {
3fc471ed 3207 dynamic_irq_init(irq);
0b8f1efa
YL		 3208 /* restore it, in case dynamic_irq_init clear it */
3209 if (desc_new)
3210 desc_new->chip_data = cfg_new;
3fc471ed
EB		 3211 }
3212 return irq;
3213}
3214
be5d5350 3215static int nr_irqs_gsi = NR_IRQS_LEGACY;
199751d7
YL		 3216int create_irq(void)
3217{
be5d5350 3218 unsigned int irq_want;
54168ed7
IM		 3219 int irq;
3220
be5d5350
YL		 3221 irq_want = nr_irqs_gsi;
3222 irq = create_irq_nr(irq_want);
54168ed7
IM		 3223
3224 if (irq == 0)
3225 irq = -1;
3226
3227 return irq;
199751d7
YL		 3228}
3229
3fc471ed
EB		 3230void destroy_irq(unsigned int irq)
3231{
3232 unsigned long flags;
0b8f1efa
YL		 3233 struct irq_cfg *cfg;
3234 struct irq_desc *desc;
3fc471ed 3235
0b8f1efa
YL		 3236 /* store it, in case dynamic_irq_cleanup clear it */
3237 desc = irq_to_desc(irq);
3238 cfg = desc->chip_data;
3fc471ed 3239 dynamic_irq_cleanup(irq);
0b8f1efa
YL		 3240 /* connect back irq_cfg */
3241 if (desc)
3242 desc->chip_data = cfg;
3fc471ed 3243
54168ed7
IM		 3244#ifdef CONFIG_INTR_REMAP
3245 free_irte(irq);
3246#endif
3fc471ed 3247 spin_lock_irqsave(&vector_lock, flags);
3145e941 3248 __clear_irq_vector(irq, cfg);
3fc471ed
EB		 3249 spin_unlock_irqrestore(&vector_lock, flags);
3250}
3fc471ed 3251
2d3fcc1c 3252/*
27b46d76 3253 * MSI message composition
2d3fcc1c
EB		 3254 */
3255#ifdef CONFIG_PCI_MSI
3b7d1921 3256static int msi_compose_msg(struct pci_dev *pdev, unsigned int irq, struct msi_msg *msg)
2d3fcc1c 3257{
497c9a19
YL		 3258 struct irq_cfg *cfg;
3259 int err;
2d3fcc1c
EB		 3260 unsigned dest;
3261
3145e941 3262 cfg = irq_cfg(irq);
22f65d31 3263 err = assign_irq_vector(irq, cfg, TARGET_CPUS);
497c9a19
YL		 3264 if (err)
3265 return err;
2d3fcc1c 3266
22f65d31 3267 dest = cpu_mask_to_apicid_and(cfg->domain, TARGET_CPUS);
497c9a19 3268
54168ed7
IM		 3269#ifdef CONFIG_INTR_REMAP
3270 if (irq_remapped(irq)) {
3271 struct irte irte;
3272 int ir_index;
3273 u16 sub_handle;
3274
3275 ir_index = map_irq_to_irte_handle(irq, &sub_handle);
3276 BUG_ON(ir_index == -1);
3277
3278 memset (&irte, 0, sizeof(irte));
3279
3280 irte.present = 1;
3281 irte.dst_mode = INT_DEST_MODE;
3282 irte.trigger_mode = 0; /* edge */
3283 irte.dlvry_mode = INT_DELIVERY_MODE;
3284 irte.vector = cfg->vector;
3285 irte.dest_id = IRTE_DEST(dest);
3286
3287 modify_irte(irq, &irte);
3288
3289 msg->address_hi = MSI_ADDR_BASE_HI;
3290 msg->data = sub_handle;
3291 msg->address_lo = MSI_ADDR_BASE_LO | MSI_ADDR_IR_EXT_INT |
3292 MSI_ADDR_IR_SHV |
3293 MSI_ADDR_IR_INDEX1(ir_index) |
3294 MSI_ADDR_IR_INDEX2(ir_index);
3295 } else
3296#endif
3297 {
3298 msg->address_hi = MSI_ADDR_BASE_HI;
3299 msg->address_lo =
3300 MSI_ADDR_BASE_LO |
3301 ((INT_DEST_MODE == 0) ?
3302 MSI_ADDR_DEST_MODE_PHYSICAL:
3303 MSI_ADDR_DEST_MODE_LOGICAL) |
3304 ((INT_DELIVERY_MODE != dest_LowestPrio) ?
3305 MSI_ADDR_REDIRECTION_CPU:
3306 MSI_ADDR_REDIRECTION_LOWPRI) |
3307 MSI_ADDR_DEST_ID(dest);
497c9a19 3308
54168ed7
IM		 3309 msg->data =
3310 MSI_DATA_TRIGGER_EDGE |
3311 MSI_DATA_LEVEL_ASSERT |
3312 ((INT_DELIVERY_MODE != dest_LowestPrio) ?
3313 MSI_DATA_DELIVERY_FIXED:
3314 MSI_DATA_DELIVERY_LOWPRI) |
3315 MSI_DATA_VECTOR(cfg->vector);
3316 }
497c9a19 3317 return err;
2d3fcc1c
EB		 3318}
3319
3b7d1921 3320#ifdef CONFIG_SMP
0de26520 3321static void set_msi_irq_affinity(unsigned int irq, const struct cpumask *mask)
2d3fcc1c 3322{
3145e941 3323 struct irq_desc *desc = irq_to_desc(irq);
497c9a19 3324 struct irq_cfg *cfg;
3b7d1921
EB		 3325 struct msi_msg msg;
3326 unsigned int dest;
3b7d1921 3327
22f65d31
MT		 3328 dest = set_desc_affinity(desc, mask);
3329 if (dest == BAD_APICID)
497c9a19 3330 return;
2d3fcc1c 3331
3145e941 3332 cfg = desc->chip_data;
2d3fcc1c 3333
3145e941 3334 read_msi_msg_desc(desc, &msg);
3b7d1921
EB		 3335
3336 msg.data &= ~MSI_DATA_VECTOR_MASK;
497c9a19 3337 msg.data |= MSI_DATA_VECTOR(cfg->vector);
3b7d1921
EB		 3338 msg.address_lo &= ~MSI_ADDR_DEST_ID_MASK;
3339 msg.address_lo |= MSI_ADDR_DEST_ID(dest);
3340
3145e941 3341 write_msi_msg_desc(desc, &msg);
2d3fcc1c 3342}
54168ed7
IM		 3343#ifdef CONFIG_INTR_REMAP
3344/*
3345 * Migrate the MSI irq to another cpumask. This migration is
3346 * done in the process context using interrupt-remapping hardware.
3347 */
e7986739
MT		 3348static void
3349ir_set_msi_irq_affinity(unsigned int irq, const struct cpumask *mask)
54168ed7 3350{
3145e941 3351 struct irq_desc *desc = irq_to_desc(irq);
a7883dec 3352 struct irq_cfg *cfg = desc->chip_data;
54168ed7 3353 unsigned int dest;
54168ed7 3354 struct irte irte;
54168ed7
IM		 3355
3356 if (get_irte(irq, &irte))
3357 return;
3358
22f65d31
MT		 3359 dest = set_desc_affinity(desc, mask);
3360 if (dest == BAD_APICID)
54168ed7
IM		 3361 return;
3362
54168ed7
IM		 3363 irte.vector = cfg->vector;
3364 irte.dest_id = IRTE_DEST(dest);
3365
3366 /*
3367 * atomically update the IRTE with the new destination and vector.
3368 */
3369 modify_irte(irq, &irte);
3370
3371 /*
3372 * After this point, all the interrupts will start arriving
3373 * at the new destination. So, time to cleanup the previous
3374 * vector allocation.
3375 */
22f65d31
MT		 3376 if (cfg->move_in_progress)
3377 send_cleanup_vector(cfg);
54168ed7 3378}
3145e941 3379
54168ed7 3380#endif
3b7d1921 3381#endif /* CONFIG_SMP */
2d3fcc1c 3382
3b7d1921
EB		 3383/*
3384 * IRQ Chip for MSI PCI/PCI-X/PCI-Express Devices,
3385 * which implement the MSI or MSI-X Capability Structure.
3386 */
3387static struct irq_chip msi_chip = {
3388 .name = "PCI-MSI",
3389 .unmask = unmask_msi_irq,
3390 .mask = mask_msi_irq,
1d025192 3391 .ack = ack_apic_edge,
3b7d1921
EB		 3392#ifdef CONFIG_SMP
3393 .set_affinity = set_msi_irq_affinity,
3394#endif
3395 .retrigger = ioapic_retrigger_irq,
2d3fcc1c
EB		 3396};
3397
54168ed7
IM		 3398#ifdef CONFIG_INTR_REMAP
3399static struct irq_chip msi_ir_chip = {
3400 .name = "IR-PCI-MSI",
3401 .unmask = unmask_msi_irq,
3402 .mask = mask_msi_irq,
3403 .ack = ack_x2apic_edge,
3404#ifdef CONFIG_SMP
3405 .set_affinity = ir_set_msi_irq_affinity,
3406#endif
3407 .retrigger = ioapic_retrigger_irq,
3408};
3409
3410/*
3411 * Map the PCI dev to the corresponding remapping hardware unit
3412 * and allocate 'nvec' consecutive interrupt-remapping table entries
3413 * in it.
3414 */
3415static int msi_alloc_irte(struct pci_dev *dev, int irq, int nvec)
3416{
3417 struct intel_iommu *iommu;
3418 int index;
3419
3420 iommu = map_dev_to_ir(dev);
3421 if (!iommu) {
3422 printk(KERN_ERR
3423 "Unable to map PCI %s to iommu\n", pci_name(dev));
3424 return -ENOENT;
3425 }
3426
3427 index = alloc_irte(iommu, irq, nvec);
3428 if (index < 0) {
3429 printk(KERN_ERR
3430 "Unable to allocate %d IRTE for PCI %s\n", nvec,
d6c88a50 3431 pci_name(dev));
54168ed7
IM		 3432 return -ENOSPC;
3433 }
3434 return index;
3435}
3436#endif
1d025192 3437
3145e941 3438static int setup_msi_irq(struct pci_dev *dev, struct msi_desc *msidesc, int irq)
1d025192
YL		 3439{
3440 int ret;
3441 struct msi_msg msg;
3442
3443 ret = msi_compose_msg(dev, irq, &msg);
3444 if (ret < 0)
3445 return ret;
3446
3145e941 3447 set_irq_msi(irq, msidesc);
1d025192
YL		 3448 write_msi_msg(irq, &msg);
3449
54168ed7
IM		 3450#ifdef CONFIG_INTR_REMAP
3451 if (irq_remapped(irq)) {
3452 struct irq_desc *desc = irq_to_desc(irq);
3453 /*
3454 * irq migration in process context
3455 */
3456 desc->status |= IRQ_MOVE_PCNTXT;
3457 set_irq_chip_and_handler_name(irq, &msi_ir_chip, handle_edge_irq, "edge");
3458 } else
3459#endif
3460 set_irq_chip_and_handler_name(irq, &msi_chip, handle_edge_irq, "edge");
1d025192 3461
c81bba49
YL		 3462 dev_printk(KERN_DEBUG, &dev->dev, "irq %d for MSI/MSI-X\n", irq);
3463
1d025192
YL		 3464 return 0;
3465}
3466
0b8f1efa 3467int arch_setup_msi_irq(struct pci_dev *dev, struct msi_desc *msidesc)
3b7d1921 3468{
54168ed7
IM		 3469 unsigned int irq;
3470 int ret;
199751d7
YL		 3471 unsigned int irq_want;
3472
be5d5350 3473 irq_want = nr_irqs_gsi;
199751d7 3474 irq = create_irq_nr(irq_want);
199751d7
YL		 3475 if (irq == 0)
3476 return -1;
f7feaca7 3477
54168ed7
IM		 3478#ifdef CONFIG_INTR_REMAP
3479 if (!intr_remapping_enabled)
3480 goto no_ir;
3481
3482 ret = msi_alloc_irte(dev, irq, 1);
3483 if (ret < 0)
3484 goto error;
3485no_ir:
3486#endif
0b8f1efa 3487 ret = setup_msi_irq(dev, msidesc, irq);
f7feaca7
EB		 3488 if (ret < 0) {
3489 destroy_irq(irq);
3b7d1921 3490 return ret;
54168ed7 3491 }
7fe3730d 3492 return 0;
54168ed7
IM		 3493
3494#ifdef CONFIG_INTR_REMAP
3495error:
3496 destroy_irq(irq);
3497 return ret;
3498#endif
3b7d1921
EB		 3499}
3500
047c8fdb
YL		 3501int arch_setup_msi_irqs(struct pci_dev *dev, int nvec, int type)
3502{
54168ed7
IM		 3503 unsigned int irq;
3504 int ret, sub_handle;
0b8f1efa 3505 struct msi_desc *msidesc;
54168ed7
IM		 3506 unsigned int irq_want;
3507
3508#ifdef CONFIG_INTR_REMAP
3509 struct intel_iommu *iommu = 0;
3510 int index = 0;
3511#endif
3512
be5d5350 3513 irq_want = nr_irqs_gsi;
54168ed7 3514 sub_handle = 0;
0b8f1efa
YL		 3515 list_for_each_entry(msidesc, &dev->msi_list, list) {
3516 irq = create_irq_nr(irq_want);
be5d5350 3517 irq_want++;
54168ed7
IM		 3518 if (irq == 0)
3519 return -1;
3520#ifdef CONFIG_INTR_REMAP
3521 if (!intr_remapping_enabled)
3522 goto no_ir;
3523
3524 if (!sub_handle) {
3525 /*
3526 * allocate the consecutive block of IRTE's
3527 * for 'nvec'
3528 */
3529 index = msi_alloc_irte(dev, irq, nvec);
3530 if (index < 0) {
3531 ret = index;
3532 goto error;
3533 }
3534 } else {
3535 iommu = map_dev_to_ir(dev);
3536 if (!iommu) {
3537 ret = -ENOENT;
3538 goto error;
3539 }
3540 /*
3541 * setup the mapping between the irq and the IRTE
3542 * base index, the sub_handle pointing to the
3543 * appropriate interrupt remap table entry.
3544 */
3545 set_irte_irq(irq, iommu, index, sub_handle);
3546 }
3547no_ir:
3548#endif
0b8f1efa 3549 ret = setup_msi_irq(dev, msidesc, irq);
54168ed7
IM		 3550 if (ret < 0)
3551 goto error;
3552 sub_handle++;
3553 }
3554 return 0;
047c8fdb
YL		 3555
3556error:
54168ed7
IM		 3557 destroy_irq(irq);
3558 return ret;
047c8fdb
YL		 3559}
3560
3b7d1921
EB		 3561void arch_teardown_msi_irq(unsigned int irq)
3562{
f7feaca7 3563 destroy_irq(irq);
3b7d1921
EB		 3564}
3565
54168ed7
IM		 3566#ifdef CONFIG_DMAR
3567#ifdef CONFIG_SMP
22f65d31 3568static void dmar_msi_set_affinity(unsigned int irq, const struct cpumask *mask)
54168ed7 3569{
3145e941 3570 struct irq_desc *desc = irq_to_desc(irq);
54168ed7
IM		 3571 struct irq_cfg *cfg;
3572 struct msi_msg msg;
3573 unsigned int dest;
54168ed7 3574
22f65d31
MT		 3575 dest = set_desc_affinity(desc, mask);
3576 if (dest == BAD_APICID)
54168ed7
IM		 3577 return;
3578
3145e941 3579 cfg = desc->chip_data;
54168ed7
IM		 3580
3581 dmar_msi_read(irq, &msg);
3582
3583 msg.data &= ~MSI_DATA_VECTOR_MASK;
3584 msg.data |= MSI_DATA_VECTOR(cfg->vector);
3585 msg.address_lo &= ~MSI_ADDR_DEST_ID_MASK;
3586 msg.address_lo |= MSI_ADDR_DEST_ID(dest);
3587
3588 dmar_msi_write(irq, &msg);
54168ed7 3589}
3145e941 3590
54168ed7
IM		 3591#endif /* CONFIG_SMP */
3592
3593struct irq_chip dmar_msi_type = {
3594 .name = "DMAR_MSI",
3595 .unmask = dmar_msi_unmask,
3596 .mask = dmar_msi_mask,
3597 .ack = ack_apic_edge,
3598#ifdef CONFIG_SMP
3599 .set_affinity = dmar_msi_set_affinity,
3600#endif
3601 .retrigger = ioapic_retrigger_irq,
3602};
3603
3604int arch_setup_dmar_msi(unsigned int irq)
3605{
3606 int ret;
3607 struct msi_msg msg;
2d3fcc1c 3608
54168ed7
IM		 3609 ret = msi_compose_msg(NULL, irq, &msg);
3610 if (ret < 0)
3611 return ret;
3612 dmar_msi_write(irq, &msg);
3613 set_irq_chip_and_handler_name(irq, &dmar_msi_type, handle_edge_irq,
3614 "edge");
3615 return 0;
3616}
3617#endif
3618
58ac1e76 3619#ifdef CONFIG_HPET_TIMER
3620
3621#ifdef CONFIG_SMP
22f65d31 3622static void hpet_msi_set_affinity(unsigned int irq, const struct cpumask *mask)
58ac1e76 3623{
3145e941 3624 struct irq_desc *desc = irq_to_desc(irq);
58ac1e76 3625 struct irq_cfg *cfg;
58ac1e76 3626 struct msi_msg msg;
3627 unsigned int dest;
58ac1e76 3628
22f65d31
MT		 3629 dest = set_desc_affinity(desc, mask);
3630 if (dest == BAD_APICID)
58ac1e76 3631 return;
3632
3145e941 3633 cfg = desc->chip_data;
58ac1e76 3634
3635 hpet_msi_read(irq, &msg);
3636
3637 msg.data &= ~MSI_DATA_VECTOR_MASK;
3638 msg.data |= MSI_DATA_VECTOR(cfg->vector);
3639 msg.address_lo &= ~MSI_ADDR_DEST_ID_MASK;
3640 msg.address_lo |= MSI_ADDR_DEST_ID(dest);
3641
3642 hpet_msi_write(irq, &msg);
58ac1e76 3643}
3145e941 3644
58ac1e76 3645#endif /* CONFIG_SMP */
3646
3647struct irq_chip hpet_msi_type = {
3648 .name = "HPET_MSI",
3649 .unmask = hpet_msi_unmask,
3650 .mask = hpet_msi_mask,
3651 .ack = ack_apic_edge,
3652#ifdef CONFIG_SMP
3653 .set_affinity = hpet_msi_set_affinity,
3654#endif
3655 .retrigger = ioapic_retrigger_irq,
3656};
3657
3658int arch_setup_hpet_msi(unsigned int irq)
3659{
3660 int ret;
3661 struct msi_msg msg;
3662
3663 ret = msi_compose_msg(NULL, irq, &msg);
3664 if (ret < 0)
3665 return ret;
3666
3667 hpet_msi_write(irq, &msg);
3668 set_irq_chip_and_handler_name(irq, &hpet_msi_type, handle_edge_irq,
3669 "edge");
c81bba49 3670
58ac1e76 3671 return 0;
3672}
3673#endif
3674
54168ed7 3675#endif /* CONFIG_PCI_MSI */
8b955b0d
EB		 3676/*
3677 * Hypertransport interrupt support
3678 */
3679#ifdef CONFIG_HT_IRQ
3680
3681#ifdef CONFIG_SMP
3682
497c9a19 3683static void target_ht_irq(unsigned int irq, unsigned int dest, u8 vector)
8b955b0d 3684{
ec68307c
EB		 3685 struct ht_irq_msg msg;
3686 fetch_ht_irq_msg(irq, &msg);
8b955b0d 3687
497c9a19 3688 msg.address_lo &= ~(HT_IRQ_LOW_VECTOR_MASK | HT_IRQ_LOW_DEST_ID_MASK);
ec68307c 3689 msg.address_hi &= ~(HT_IRQ_HIGH_DEST_ID_MASK);
8b955b0d 3690
497c9a19 3691 msg.address_lo |= HT_IRQ_LOW_VECTOR(vector) | HT_IRQ_LOW_DEST_ID(dest);
ec68307c 3692 msg.address_hi |= HT_IRQ_HIGH_DEST_ID(dest);
8b955b0d 3693
ec68307c 3694 write_ht_irq_msg(irq, &msg);
8b955b0d
EB		 3695}
3696
22f65d31 3697static void set_ht_irq_affinity(unsigned int irq, const struct cpumask *mask)
8b955b0d 3698{
3145e941 3699 struct irq_desc *desc = irq_to_desc(irq);
497c9a19 3700 struct irq_cfg *cfg;
8b955b0d 3701 unsigned int dest;
8b955b0d 3702
22f65d31
MT		 3703 dest = set_desc_affinity(desc, mask);
3704 if (dest == BAD_APICID)
497c9a19 3705 return;
8b955b0d 3706
3145e941 3707 cfg = desc->chip_data;
8b955b0d 3708
497c9a19 3709 target_ht_irq(irq, dest, cfg->vector);
8b955b0d 3710}
3145e941 3711
8b955b0d
EB		 3712#endif
3713
c37e108d 3714static struct irq_chip ht_irq_chip = {
8b955b0d
EB		 3715 .name = "PCI-HT",
3716 .mask = mask_ht_irq,
3717 .unmask = unmask_ht_irq,
1d025192 3718 .ack = ack_apic_edge,
8b955b0d
EB		 3719#ifdef CONFIG_SMP
3720 .set_affinity = set_ht_irq_affinity,
3721#endif
3722 .retrigger = ioapic_retrigger_irq,
3723};
3724
3725int arch_setup_ht_irq(unsigned int irq, struct pci_dev *dev)
3726{
497c9a19
YL		 3727 struct irq_cfg *cfg;
3728 int err;
8b955b0d 3729
3145e941 3730 cfg = irq_cfg(irq);
e7986739 3731 err = assign_irq_vector(irq, cfg, TARGET_CPUS);
54168ed7 3732 if (!err) {
ec68307c 3733 struct ht_irq_msg msg;
8b955b0d 3734 unsigned dest;
8b955b0d 3735
22f65d31 3736 dest = cpu_mask_to_apicid_and(cfg->domain, TARGET_CPUS);
8b955b0d 3737
ec68307c 3738 msg.address_hi = HT_IRQ_HIGH_DEST_ID(dest);
8b955b0d 3739
ec68307c
EB		 3740 msg.address_lo =
3741 HT_IRQ_LOW_BASE |
8b955b0d 3742 HT_IRQ_LOW_DEST_ID(dest) |
497c9a19 3743 HT_IRQ_LOW_VECTOR(cfg->vector) |
8b955b0d
EB		 3744 ((INT_DEST_MODE == 0) ?
3745 HT_IRQ_LOW_DM_PHYSICAL :
3746 HT_IRQ_LOW_DM_LOGICAL) |
3747 HT_IRQ_LOW_RQEOI_EDGE |
3748 ((INT_DELIVERY_MODE != dest_LowestPrio) ?
3749 HT_IRQ_LOW_MT_FIXED :
3750 HT_IRQ_LOW_MT_ARBITRATED) |
3751 HT_IRQ_LOW_IRQ_MASKED;
3752
ec68307c 3753 write_ht_irq_msg(irq, &msg);
8b955b0d 3754
a460e745
IM		 3755 set_irq_chip_and_handler_name(irq, &ht_irq_chip,
3756 handle_edge_irq, "edge");
c81bba49
YL		 3757
3758 dev_printk(KERN_DEBUG, &dev->dev, "irq %d for HT\n", irq);
8b955b0d 3759 }
497c9a19 3760 return err;
8b955b0d
EB		 3761}
3762#endif /* CONFIG_HT_IRQ */
3763
4173a0e7
DN		 3764#ifdef CONFIG_X86_64
3765/*
3766 * Re-target the irq to the specified CPU and enable the specified MMR located
3767 * on the specified blade to allow the sending of MSIs to the specified CPU.
3768 */
3769int arch_enable_uv_irq(char *irq_name, unsigned int irq, int cpu, int mmr_blade,
3770 unsigned long mmr_offset)
3771{
22f65d31 3772 const struct cpumask *eligible_cpu = cpumask_of(cpu);
4173a0e7
DN		 3773 struct irq_cfg *cfg;
3774 int mmr_pnode;
3775 unsigned long mmr_value;
3776 struct uv_IO_APIC_route_entry *entry;
3777 unsigned long flags;
3778 int err;
3779
3145e941
YL		 3780 cfg = irq_cfg(irq);
3781
e7986739 3782 err = assign_irq_vector(irq, cfg, eligible_cpu);
4173a0e7
DN		 3783 if (err != 0)
3784 return err;
3785
3786 spin_lock_irqsave(&vector_lock, flags);
3787 set_irq_chip_and_handler_name(irq, &uv_irq_chip, handle_percpu_irq,
3788 irq_name);
3789 spin_unlock_irqrestore(&vector_lock, flags);
3790
4173a0e7
DN		 3791 mmr_value = 0;
3792 entry = (struct uv_IO_APIC_route_entry *)&mmr_value;
3793 BUG_ON(sizeof(struct uv_IO_APIC_route_entry) != sizeof(unsigned long));
3794
3795 entry->vector = cfg->vector;
3796 entry->delivery_mode = INT_DELIVERY_MODE;
3797 entry->dest_mode = INT_DEST_MODE;
3798 entry->polarity = 0;
3799 entry->trigger = 0;
3800 entry->mask = 0;
e7986739 3801 entry->dest = cpu_mask_to_apicid(eligible_cpu);
4173a0e7
DN		 3802
3803 mmr_pnode = uv_blade_to_pnode(mmr_blade);
3804 uv_write_global_mmr64(mmr_pnode, mmr_offset, mmr_value);
3805
3806 return irq;
3807}
3808
3809/*
3810 * Disable the specified MMR located on the specified blade so that MSIs are
3811 * longer allowed to be sent.
3812 */
3813void arch_disable_uv_irq(int mmr_blade, unsigned long mmr_offset)
3814{
3815 unsigned long mmr_value;
3816 struct uv_IO_APIC_route_entry *entry;
3817 int mmr_pnode;
3818
3819 mmr_value = 0;
3820 entry = (struct uv_IO_APIC_route_entry *)&mmr_value;
3821 BUG_ON(sizeof(struct uv_IO_APIC_route_entry) != sizeof(unsigned long));
3822
3823 entry->mask = 1;
3824
3825 mmr_pnode = uv_blade_to_pnode(mmr_blade);
3826 uv_write_global_mmr64(mmr_pnode, mmr_offset, mmr_value);
3827}
3828#endif /* CONFIG_X86_64 */
3829
9d6a4d08
YL		 3830int __init io_apic_get_redir_entries (int ioapic)
3831{
3832 union IO_APIC_reg_01 reg_01;
3833 unsigned long flags;
3834
3835 spin_lock_irqsave(&ioapic_lock, flags);
3836 reg_01.raw = io_apic_read(ioapic, 1);
3837 spin_unlock_irqrestore(&ioapic_lock, flags);
3838
3839 return reg_01.bits.entries;
3840}
3841
be5d5350 3842void __init probe_nr_irqs_gsi(void)
9d6a4d08 3843{
be5d5350
YL		 3844 int idx;
3845 int nr = 0;
3846
3847 for (idx = 0; idx < nr_ioapics; idx++)
3848 nr += io_apic_get_redir_entries(idx) + 1;
3849
3850 if (nr > nr_irqs_gsi)
3851 nr_irqs_gsi = nr;
9d6a4d08
YL		 3852}
3853
1da177e4 3854/* --------------------------------------------------------------------------
54168ed7 3855 ACPI-based IOAPIC Configuration
1da177e4
LT		 3856 -------------------------------------------------------------------------- */
3857
888ba6c6 3858#ifdef CONFIG_ACPI
1da177e4 3859
54168ed7 3860#ifdef CONFIG_X86_32
36062448 3861int __init io_apic_get_unique_id(int ioapic, int apic_id)
1da177e4
LT		 3862{
3863 union IO_APIC_reg_00 reg_00;
3864 static physid_mask_t apic_id_map = PHYSID_MASK_NONE;
3865 physid_mask_t tmp;
3866 unsigned long flags;
3867 int i = 0;
3868
3869 /*
36062448
PC		 3870 * The P4 platform supports up to 256 APIC IDs on two separate APIC
3871 * buses (one for LAPICs, one for IOAPICs), where predecessors only
1da177e4 3872 * supports up to 16 on one shared APIC bus.
36062448 3873 *
1da177e4
LT		 3874 * TBD: Expand LAPIC/IOAPIC support on P4-class systems to take full
3875 * advantage of new APIC bus architecture.
3876 */
3877
3878 if (physids_empty(apic_id_map))
3879 apic_id_map = ioapic_phys_id_map(phys_cpu_present_map);
3880
3881 spin_lock_irqsave(&ioapic_lock, flags);
3882 reg_00.raw = io_apic_read(ioapic, 0);
3883 spin_unlock_irqrestore(&ioapic_lock, flags);
3884
3885 if (apic_id >= get_physical_broadcast()) {
3886 printk(KERN_WARNING "IOAPIC[%d]: Invalid apic_id %d, trying "
3887 "%d\n", ioapic, apic_id, reg_00.bits.ID);
3888 apic_id = reg_00.bits.ID;
3889 }
3890
3891 /*
36062448 3892 * Every APIC in a system must have a unique ID or we get lots of nice
1da177e4
LT		 3893 * 'stuck on smp_invalidate_needed IPI wait' messages.
3894 */
3895 if (check_apicid_used(apic_id_map, apic_id)) {
3896
3897 for (i = 0; i < get_physical_broadcast(); i++) {
3898 if (!check_apicid_used(apic_id_map, i))
3899 break;
3900 }
3901
3902 if (i == get_physical_broadcast())
3903 panic("Max apic_id exceeded!\n");
3904
3905 printk(KERN_WARNING "IOAPIC[%d]: apic_id %d already used, "
3906 "trying %d\n", ioapic, apic_id, i);
3907
3908 apic_id = i;
36062448 3909 }
1da177e4
LT		 3910
3911 tmp = apicid_to_cpu_present(apic_id);
3912 physids_or(apic_id_map, apic_id_map, tmp);
3913
3914 if (reg_00.bits.ID != apic_id) {
3915 reg_00.bits.ID = apic_id;
3916
3917 spin_lock_irqsave(&ioapic_lock, flags);
3918 io_apic_write(ioapic, 0, reg_00.raw);
3919 reg_00.raw = io_apic_read(ioapic, 0);
3920 spin_unlock_irqrestore(&ioapic_lock, flags);
3921
3922 /* Sanity check */
6070f9ec
AD		 3923 if (reg_00.bits.ID != apic_id) {
3924 printk("IOAPIC[%d]: Unable to change apic_id!\n", ioapic);
3925 return -1;
3926 }
1da177e4
LT		 3927 }
3928
3929 apic_printk(APIC_VERBOSE, KERN_INFO
3930 "IOAPIC[%d]: Assigned apic_id %d\n", ioapic, apic_id);
3931
3932 return apic_id;
3933}
3934
36062448 3935int __init io_apic_get_version(int ioapic)
1da177e4
LT		 3936{
3937 union IO_APIC_reg_01 reg_01;
3938 unsigned long flags;
3939
3940 spin_lock_irqsave(&ioapic_lock, flags);
3941 reg_01.raw = io_apic_read(ioapic, 1);
3942 spin_unlock_irqrestore(&ioapic_lock, flags);
3943
3944 return reg_01.bits.version;
3945}
54168ed7 3946#endif
1da177e4 3947
54168ed7 3948int io_apic_set_pci_routing (int ioapic, int pin, int irq, int triggering, int polarity)
1da177e4 3949{
0b8f1efa
YL		 3950 struct irq_desc *desc;
3951 struct irq_cfg *cfg;
3952 int cpu = boot_cpu_id;
3953
1da177e4 3954 if (!IO_APIC_IRQ(irq)) {
54168ed7 3955 apic_printk(APIC_QUIET,KERN_ERR "IOAPIC[%d]: Invalid reference to IRQ 0\n",
1da177e4
LT		 3956 ioapic);
3957 return -EINVAL;
3958 }
3959
0b8f1efa
YL		 3960 desc = irq_to_desc_alloc_cpu(irq, cpu);
3961 if (!desc) {
3962 printk(KERN_INFO "can not get irq_desc %d\n", irq);
3963 return 0;
3964 }
3965
1da177e4
LT		 3966 /*
3967 * IRQs < 16 are already in the irq_2_pin[] map
3968 */
99d093d1 3969 if (irq >= NR_IRQS_LEGACY) {
0b8f1efa 3970 cfg = desc->chip_data;
3145e941 3971 add_pin_to_irq_cpu(cfg, cpu, ioapic, pin);
0b8f1efa 3972 }
1da177e4 3973
3145e941 3974 setup_IO_APIC_irq(ioapic, pin, irq, desc, triggering, polarity);
1da177e4
LT		 3975
3976 return 0;
3977}
3978
54168ed7 3979
61fd47e0
SL		 3980int acpi_get_override_irq(int bus_irq, int *trigger, int *polarity)
3981{
3982 int i;
3983
3984 if (skip_ioapic_setup)
3985 return -1;
3986
3987 for (i = 0; i < mp_irq_entries; i++)
2fddb6e2
AS		 3988 if (mp_irqs[i].mp_irqtype == mp_INT &&
3989 mp_irqs[i].mp_srcbusirq == bus_irq)
61fd47e0
SL		 3990 break;
3991 if (i >= mp_irq_entries)
3992 return -1;
3993
3994 *trigger = irq_trigger(i);
3995 *polarity = irq_polarity(i);
3996 return 0;
3997}
3998
888ba6c6 3999#endif /* CONFIG_ACPI */
1a3f239d 4000
497c9a19
YL		 4001/*
4002 * This function currently is only a helper for the i386 smp boot process where
4003 * we need to reprogram the ioredtbls to cater for the cpus which have come online
4004 * so mask in all cases should simply be TARGET_CPUS
4005 */
4006#ifdef CONFIG_SMP
4007void __init setup_ioapic_dest(void)
4008{
4009 int pin, ioapic, irq, irq_entry;
6c2e9403 4010 struct irq_desc *desc;
497c9a19 4011 struct irq_cfg *cfg;
22f65d31 4012 const struct cpumask *mask;
497c9a19
YL		 4013
4014 if (skip_ioapic_setup == 1)
4015 return;
4016
4017 for (ioapic = 0; ioapic < nr_ioapics; ioapic++) {
4018 for (pin = 0; pin < nr_ioapic_registers[ioapic]; pin++) {
4019 irq_entry = find_irq_entry(ioapic, pin, mp_INT);
4020 if (irq_entry == -1)
4021 continue;
4022 irq = pin_2_irq(irq_entry, ioapic, pin);
4023
4024 /* setup_IO_APIC_irqs could fail to get vector for some device
4025 * when you have too many devices, because at that time only boot
4026 * cpu is online.
4027 */
0b8f1efa
YL		 4028 desc = irq_to_desc(irq);
4029 cfg = desc->chip_data;
6c2e9403 4030 if (!cfg->vector) {
3145e941 4031 setup_IO_APIC_irq(ioapic, pin, irq, desc,
497c9a19
YL		 4032 irq_trigger(irq_entry),
4033 irq_polarity(irq_entry));
6c2e9403
TG		 4034 continue;
4035
4036 }
4037
4038 /*
4039 * Honour affinities which have been set in early boot
4040 */
6c2e9403
TG		 4041 if (desc->status &
4042 (IRQ_NO_BALANCING | IRQ_AFFINITY_SET))
e7986739 4043 mask = &desc->affinity;
6c2e9403
TG		 4044 else
4045 mask = TARGET_CPUS;
4046
54168ed7 4047#ifdef CONFIG_INTR_REMAP
6c2e9403 4048 if (intr_remapping_enabled)
3145e941 4049 set_ir_ioapic_affinity_irq_desc(desc, mask);
54168ed7 4050 else
6c2e9403 4051#endif
3145e941 4052 set_ioapic_affinity_irq_desc(desc, mask);
497c9a19
YL		 4053 }
4054
4055 }
4056}
4057#endif
4058
54168ed7
IM		 4059#define IOAPIC_RESOURCE_NAME_SIZE 11
4060
4061static struct resource *ioapic_resources;
4062
4063static struct resource * __init ioapic_setup_resources(void)
4064{
4065 unsigned long n;
4066 struct resource *res;
4067 char *mem;
4068 int i;
4069
4070 if (nr_ioapics <= 0)
4071 return NULL;
4072
4073 n = IOAPIC_RESOURCE_NAME_SIZE + sizeof(struct resource);
4074 n *= nr_ioapics;
4075
4076 mem = alloc_bootmem(n);
4077 res = (void *)mem;
4078
4079 if (mem != NULL) {
4080 mem += sizeof(struct resource) * nr_ioapics;
4081
4082 for (i = 0; i < nr_ioapics; i++) {
4083 res[i].name = mem;
4084 res[i].flags = IORESOURCE_MEM | IORESOURCE_BUSY;
4085 sprintf(mem, "IOAPIC %u", i);
4086 mem += IOAPIC_RESOURCE_NAME_SIZE;
4087 }
4088 }
4089
4090 ioapic_resources = res;
4091
4092 return res;
4093}
54168ed7 4094
f3294a33
YL		 4095void __init ioapic_init_mappings(void)
4096{
4097 unsigned long ioapic_phys, idx = FIX_IO_APIC_BASE_0;
54168ed7 4098 struct resource *ioapic_res;
d6c88a50 4099 int i;
f3294a33 4100
54168ed7 4101 ioapic_res = ioapic_setup_resources();
f3294a33
YL		 4102 for (i = 0; i < nr_ioapics; i++) {
4103 if (smp_found_config) {
4104 ioapic_phys = mp_ioapics[i].mp_apicaddr;
54168ed7 4105#ifdef CONFIG_X86_32
d6c88a50
TG		 4106 if (!ioapic_phys) {
4107 printk(KERN_ERR
4108 "WARNING: bogus zero IO-APIC "
4109 "address found in MPTABLE, "
4110 "disabling IO/APIC support!\n");
4111 smp_found_config = 0;
4112 skip_ioapic_setup = 1;
4113 goto fake_ioapic_page;
4114 }
54168ed7 4115#endif
f3294a33 4116 } else {
54168ed7 4117#ifdef CONFIG_X86_32
f3294a33 4118fake_ioapic_page:
54168ed7 4119#endif
f3294a33 4120 ioapic_phys = (unsigned long)
54168ed7 4121 alloc_bootmem_pages(PAGE_SIZE);
f3294a33
YL		 4122 ioapic_phys = __pa(ioapic_phys);
4123 }
4124 set_fixmap_nocache(idx, ioapic_phys);
54168ed7
IM		 4125 apic_printk(APIC_VERBOSE,
4126 "mapped IOAPIC to %08lx (%08lx)\n",
4127 __fix_to_virt(idx), ioapic_phys);
f3294a33 4128 idx++;
54168ed7 4129
54168ed7
IM		 4130 if (ioapic_res != NULL) {
4131 ioapic_res->start = ioapic_phys;
4132 ioapic_res->end = ioapic_phys + (4 * 1024) - 1;
4133 ioapic_res++;
4134 }
f3294a33
YL		 4135 }
4136}
4137
54168ed7
IM		 4138static int __init ioapic_insert_resources(void)
4139{
4140 int i;
4141 struct resource *r = ioapic_resources;
4142
4143 if (!r) {
4144 printk(KERN_ERR
4145 "IO APIC resources could be not be allocated.\n");
4146 return -1;
4147 }
4148
4149 for (i = 0; i < nr_ioapics; i++) {
4150 insert_resource(&iomem_resource, r);
4151 r++;
4152 }
4153
4154 return 0;
4155}
4156
4157/* Insert the IO APIC resources after PCI initialization has occured to handle
4158 * IO APICS that are mapped in on a BAR in PCI space. */
4159late_initcall(ioapic_insert_resources);
Linux 6.x block layer and io_uring tree(s)