projects / deliverable / linux.git / blob
? search:
summary | shortlog | log | commit | commitdiff | tree
blame | history | raw | HEAD
x86: move mp_irqs to io_apic_64.c
[deliverable/linux.git] / arch / x86 / kernel / io_apic_64.c
1 /*
2 * Intel IO-APIC support for multi-Pentium hosts.
3 *
4 * Copyright (C) 1997, 1998, 1999, 2000 Ingo Molnar, Hajnalka Szabo
5 *
6 * Many thanks to Stig Venaas for trying out countless experimental
7 * patches and reporting/debugging problems patiently!
8 *
9 * (c) 1999, Multiple IO-APIC support, developed by
10 * Ken-ichi Yaku <yaku@css1.kbnes.nec.co.jp> and
11 * Hidemi Kishimoto <kisimoto@css1.kbnes.nec.co.jp>,
12 * further tested and cleaned up by Zach Brown <zab@redhat.com>
13 * and Ingo Molnar <mingo@redhat.com>
14 *
15 * Fixes
16 * Maciej W. Rozycki : Bits for genuine 82489DX APICs;
17 * thanks to Eric Gilmore
18 * and Rolf G. Tews
19 * for testing these extensively
20 * Paul Diefenbaugh : Added full ACPI support
21 */
22
23 #include <linux/mm.h>
24 #include <linux/interrupt.h>
25 #include <linux/init.h>
26 #include <linux/delay.h>
27 #include <linux/sched.h>
28 #include <linux/pci.h>
29 #include <linux/mc146818rtc.h>
30 #include <linux/acpi.h>
31 #include <linux/sysdev.h>
32 #include <linux/msi.h>
33 #include <linux/htirq.h>
34 #include <linux/dmar.h>
35 #include <linux/jiffies.h>
36 #ifdef CONFIG_ACPI
37 #include <acpi/acpi_bus.h>
38 #endif
39 #include <linux/bootmem.h>
40
41 #include <asm/idle.h>
42 #include <asm/io.h>
43 #include <asm/smp.h>
44 #include <asm/desc.h>
45 #include <asm/proto.h>
46 #include <asm/acpi.h>
47 #include <asm/dma.h>
48 #include <asm/nmi.h>
49 #include <asm/msidef.h>
50 #include <asm/hypertransport.h>
51
52 #include <mach_ipi.h>
53 #include <mach_apic.h>
54
55 struct irq_cfg {
56 cpumask_t domain;
57 cpumask_t old_domain;
58 unsigned move_cleanup_count;
59 u8 vector;
60 u8 move_in_progress : 1;
61 };
62
63 /* irq_cfg is indexed by the sum of all RTEs in all I/O APICs. */
64 struct irq_cfg irq_cfg[NR_IRQS] __read_mostly = {
65 [0] = { .domain = CPU_MASK_ALL, .vector = IRQ0_VECTOR, },
66 [1] = { .domain = CPU_MASK_ALL, .vector = IRQ1_VECTOR, },
67 [2] = { .domain = CPU_MASK_ALL, .vector = IRQ2_VECTOR, },
68 [3] = { .domain = CPU_MASK_ALL, .vector = IRQ3_VECTOR, },
69 [4] = { .domain = CPU_MASK_ALL, .vector = IRQ4_VECTOR, },
70 [5] = { .domain = CPU_MASK_ALL, .vector = IRQ5_VECTOR, },
71 [6] = { .domain = CPU_MASK_ALL, .vector = IRQ6_VECTOR, },
72 [7] = { .domain = CPU_MASK_ALL, .vector = IRQ7_VECTOR, },
73 [8] = { .domain = CPU_MASK_ALL, .vector = IRQ8_VECTOR, },
74 [9] = { .domain = CPU_MASK_ALL, .vector = IRQ9_VECTOR, },
75 [10] = { .domain = CPU_MASK_ALL, .vector = IRQ10_VECTOR, },
76 [11] = { .domain = CPU_MASK_ALL, .vector = IRQ11_VECTOR, },
77 [12] = { .domain = CPU_MASK_ALL, .vector = IRQ12_VECTOR, },
78 [13] = { .domain = CPU_MASK_ALL, .vector = IRQ13_VECTOR, },
79 [14] = { .domain = CPU_MASK_ALL, .vector = IRQ14_VECTOR, },
80 [15] = { .domain = CPU_MASK_ALL, .vector = IRQ15_VECTOR, },
81 };
82
83 static int assign_irq_vector(int irq, cpumask_t mask);
84
85 #define __apicdebuginit __init
86
87 int sis_apic_bug; /* not actually supported, dummy for compile */
88
89 static int no_timer_check;
90
91 static int disable_timer_pin_1 __initdata;
92
93 int timer_over_8254 __initdata = 1;
94
95 /* Where if anywhere is the i8259 connect in external int mode */
96 static struct { int pin, apic; } ioapic_i8259 = { -1, -1 };
97
98 static DEFINE_SPINLOCK(ioapic_lock);
99 DEFINE_SPINLOCK(vector_lock);
100
101 /*
102 * # of IRQ routing registers
103 */
104 int nr_ioapic_registers[MAX_IO_APICS];
105
106 /* I/O APIC entries */
107 struct mpc_config_ioapic mp_ioapics[MAX_IO_APICS];
108 int nr_ioapics;
109
110 /* MP IRQ source entries */
111 struct mpc_config_intsrc mp_irqs[MAX_IRQ_SOURCES];
112
113 /* # of MP IRQ source entries */
114 int mp_irq_entries;
115
116 /*
117 * Rough estimation of how many shared IRQs there are, can
118 * be changed anytime.
119 */
120 #define MAX_PLUS_SHARED_IRQS NR_IRQS
121 #define PIN_MAP_SIZE (MAX_PLUS_SHARED_IRQS + NR_IRQS)
122
123 /*
124 * This is performance-critical, we want to do it O(1)
125 *
126 * the indexing order of this array favors 1:1 mappings
127 * between pins and IRQs.
128 */
129
130 static struct irq_pin_list {
131 short apic, pin, next;
132 } irq_2_pin[PIN_MAP_SIZE];
133
134 struct io_apic {
135 unsigned int index;
136 unsigned int unused[3];
137 unsigned int data;
138 };
139
140 static __attribute_const__ struct io_apic __iomem *io_apic_base(int idx)
141 {
142 return (void __iomem *) __fix_to_virt(FIX_IO_APIC_BASE_0 + idx)
143 + (mp_ioapics[idx].mpc_apicaddr & ~PAGE_MASK);
144 }
145
146 static inline unsigned int io_apic_read(unsigned int apic, unsigned int reg)
147 {
148 struct io_apic __iomem *io_apic = io_apic_base(apic);
149 writel(reg, &io_apic->index);
150 return readl(&io_apic->data);
151 }
152
153 static inline void io_apic_write(unsigned int apic, unsigned int reg, unsigned int value)
154 {
155 struct io_apic __iomem *io_apic = io_apic_base(apic);
156 writel(reg, &io_apic->index);
157 writel(value, &io_apic->data);
158 }
159
160 /*
161 * Re-write a value: to be used for read-modify-write
162 * cycles where the read already set up the index register.
163 */
164 static inline void io_apic_modify(unsigned int apic, unsigned int value)
165 {
166 struct io_apic __iomem *io_apic = io_apic_base(apic);
167 writel(value, &io_apic->data);
168 }
169
170 static int io_apic_level_ack_pending(unsigned int irq)
171 {
172 struct irq_pin_list *entry;
173 unsigned long flags;
174 int pending = 0;
175
176 spin_lock_irqsave(&ioapic_lock, flags);
177 entry = irq_2_pin + irq;
178 for (;;) {
179 unsigned int reg;
180 int pin;
181
182 pin = entry->pin;
183 if (pin == -1)
184 break;
185 reg = io_apic_read(entry->apic, 0x10 + pin*2);
186 /* Is the remote IRR bit set? */
187 pending |= (reg >> 14) & 1;
188 if (!entry->next)
189 break;
190 entry = irq_2_pin + entry->next;
191 }
192 spin_unlock_irqrestore(&ioapic_lock, flags);
193 return pending;
194 }
195
196 /*
197 * Synchronize the IO-APIC and the CPU by doing
198 * a dummy read from the IO-APIC
199 */
200 static inline void io_apic_sync(unsigned int apic)
201 {
202 struct io_apic __iomem *io_apic = io_apic_base(apic);
203 readl(&io_apic->data);
204 }
205
206 #define __DO_ACTION(R, ACTION, FINAL) \
207 \
208 { \
209 int pin; \
210 struct irq_pin_list *entry = irq_2_pin + irq; \
211 \
212 BUG_ON(irq >= NR_IRQS); \
213 for (;;) { \
214 unsigned int reg; \
215 pin = entry->pin; \
216 if (pin == -1) \
217 break; \
218 reg = io_apic_read(entry->apic, 0x10 + R + pin*2); \
219 reg ACTION; \
220 io_apic_modify(entry->apic, reg); \
221 FINAL; \
222 if (!entry->next) \
223 break; \
224 entry = irq_2_pin + entry->next; \
225 } \
226 }
227
228 union entry_union {
229 struct { u32 w1, w2; };
230 struct IO_APIC_route_entry entry;
231 };
232
233 static struct IO_APIC_route_entry ioapic_read_entry(int apic, int pin)
234 {
235 union entry_union eu;
236 unsigned long flags;
237 spin_lock_irqsave(&ioapic_lock, flags);
238 eu.w1 = io_apic_read(apic, 0x10 + 2 * pin);
239 eu.w2 = io_apic_read(apic, 0x11 + 2 * pin);
240 spin_unlock_irqrestore(&ioapic_lock, flags);
241 return eu.entry;
242 }
243
244 /*
245 * When we write a new IO APIC routing entry, we need to write the high
246 * word first! If the mask bit in the low word is clear, we will enable
247 * the interrupt, and we need to make sure the entry is fully populated
248 * before that happens.
249 */
250 static void
251 __ioapic_write_entry(int apic, int pin, struct IO_APIC_route_entry e)
252 {
253 union entry_union eu;
254 eu.entry = e;
255 io_apic_write(apic, 0x11 + 2*pin, eu.w2);
256 io_apic_write(apic, 0x10 + 2*pin, eu.w1);
257 }
258
259 static void ioapic_write_entry(int apic, int pin, struct IO_APIC_route_entry e)
260 {
261 unsigned long flags;
262 spin_lock_irqsave(&ioapic_lock, flags);
263 __ioapic_write_entry(apic, pin, e);
264 spin_unlock_irqrestore(&ioapic_lock, flags);
265 }
266
267 /*
268 * When we mask an IO APIC routing entry, we need to write the low
269 * word first, in order to set the mask bit before we change the
270 * high bits!
271 */
272 static void ioapic_mask_entry(int apic, int pin)
273 {
274 unsigned long flags;
275 union entry_union eu = { .entry.mask = 1 };
276
277 spin_lock_irqsave(&ioapic_lock, flags);
278 io_apic_write(apic, 0x10 + 2*pin, eu.w1);
279 io_apic_write(apic, 0x11 + 2*pin, eu.w2);
280 spin_unlock_irqrestore(&ioapic_lock, flags);
281 }
282
283 #ifdef CONFIG_SMP
284 static void __target_IO_APIC_irq(unsigned int irq, unsigned int dest, u8 vector)
285 {
286 int apic, pin;
287 struct irq_pin_list *entry = irq_2_pin + irq;
288
289 BUG_ON(irq >= NR_IRQS);
290 for (;;) {
291 unsigned int reg;
292 apic = entry->apic;
293 pin = entry->pin;
294 if (pin == -1)
295 break;
296 io_apic_write(apic, 0x11 + pin*2, dest);
297 reg = io_apic_read(apic, 0x10 + pin*2);
298 reg &= ~0x000000ff;
299 reg |= vector;
300 io_apic_modify(apic, reg);
301 if (!entry->next)
302 break;
303 entry = irq_2_pin + entry->next;
304 }
305 }
306
307 static void set_ioapic_affinity_irq(unsigned int irq, cpumask_t mask)
308 {
309 struct irq_cfg *cfg = irq_cfg + irq;
310 unsigned long flags;
311 unsigned int dest;
312 cpumask_t tmp;
313
314 cpus_and(tmp, mask, cpu_online_map);
315 if (cpus_empty(tmp))
316 return;
317
318 if (assign_irq_vector(irq, mask))
319 return;
320
321 cpus_and(tmp, cfg->domain, mask);
322 dest = cpu_mask_to_apicid(tmp);
323
324 /*
325 * Only the high 8 bits are valid.
326 */
327 dest = SET_APIC_LOGICAL_ID(dest);
328
329 spin_lock_irqsave(&ioapic_lock, flags);
330 __target_IO_APIC_irq(irq, dest, cfg->vector);
331 irq_desc[irq].affinity = mask;
332 spin_unlock_irqrestore(&ioapic_lock, flags);
333 }
334 #endif
335
336 /*
337 * The common case is 1:1 IRQ<->pin mappings. Sometimes there are
338 * shared ISA-space IRQs, so we have to support them. We are super
339 * fast in the common case, and fast for shared ISA-space IRQs.
340 */
341 static void add_pin_to_irq(unsigned int irq, int apic, int pin)
342 {
343 static int first_free_entry = NR_IRQS;
344 struct irq_pin_list *entry = irq_2_pin + irq;
345
346 BUG_ON(irq >= NR_IRQS);
347 while (entry->next)
348 entry = irq_2_pin + entry->next;
349
350 if (entry->pin != -1) {
351 entry->next = first_free_entry;
352 entry = irq_2_pin + entry->next;
353 if (++first_free_entry >= PIN_MAP_SIZE)
354 panic("io_apic.c: ran out of irq_2_pin entries!");
355 }
356 entry->apic = apic;
357 entry->pin = pin;
358 }
359
360
361 #define DO_ACTION(name,R,ACTION, FINAL) \
362 \
363 static void name##_IO_APIC_irq (unsigned int irq) \
364 __DO_ACTION(R, ACTION, FINAL)
365
366 DO_ACTION( __mask, 0, |= 0x00010000, io_apic_sync(entry->apic) )
367 /* mask = 1 */
368 DO_ACTION( __unmask, 0, &= 0xfffeffff, )
369 /* mask = 0 */
370
371 static void mask_IO_APIC_irq (unsigned int irq)
372 {
373 unsigned long flags;
374
375 spin_lock_irqsave(&ioapic_lock, flags);
376 __mask_IO_APIC_irq(irq);
377 spin_unlock_irqrestore(&ioapic_lock, flags);
378 }
379
380 static void unmask_IO_APIC_irq (unsigned int irq)
381 {
382 unsigned long flags;
383
384 spin_lock_irqsave(&ioapic_lock, flags);
385 __unmask_IO_APIC_irq(irq);
386 spin_unlock_irqrestore(&ioapic_lock, flags);
387 }
388
389 static void clear_IO_APIC_pin(unsigned int apic, unsigned int pin)
390 {
391 struct IO_APIC_route_entry entry;
392
393 /* Check delivery_mode to be sure we're not clearing an SMI pin */
394 entry = ioapic_read_entry(apic, pin);
395 if (entry.delivery_mode == dest_SMI)
396 return;
397 /*
398 * Disable it in the IO-APIC irq-routing table:
399 */
400 ioapic_mask_entry(apic, pin);
401 }
402
403 static void clear_IO_APIC (void)
404 {
405 int apic, pin;
406
407 for (apic = 0; apic < nr_ioapics; apic++)
408 for (pin = 0; pin < nr_ioapic_registers[apic]; pin++)
409 clear_IO_APIC_pin(apic, pin);
410 }
411
412 int skip_ioapic_setup;
413 int ioapic_force;
414
415 static int __init parse_noapic(char *str)
416 {
417 disable_ioapic_setup();
418 return 0;
419 }
420 early_param("noapic", parse_noapic);
421
422 /* Actually the next is obsolete, but keep it for paranoid reasons -AK */
423 static int __init disable_timer_pin_setup(char *arg)
424 {
425 disable_timer_pin_1 = 1;
426 return 1;
427 }
428 __setup("disable_timer_pin_1", disable_timer_pin_setup);
429
430 static int __init setup_disable_8254_timer(char *s)
431 {
432 timer_over_8254 = -1;
433 return 1;
434 }
435 static int __init setup_enable_8254_timer(char *s)
436 {
437 timer_over_8254 = 2;
438 return 1;
439 }
440
441 __setup("disable_8254_timer", setup_disable_8254_timer);
442 __setup("enable_8254_timer", setup_enable_8254_timer);
443
444
445 /*
446 * Find the IRQ entry number of a certain pin.
447 */
448 static int find_irq_entry(int apic, int pin, int type)
449 {
450 int i;
451
452 for (i = 0; i < mp_irq_entries; i++)
453 if (mp_irqs[i].mpc_irqtype == type &&
454 (mp_irqs[i].mpc_dstapic == mp_ioapics[apic].mpc_apicid ||
455 mp_irqs[i].mpc_dstapic == MP_APIC_ALL) &&
456 mp_irqs[i].mpc_dstirq == pin)
457 return i;
458
459 return -1;
460 }
461
462 /*
463 * Find the pin to which IRQ[irq] (ISA) is connected
464 */
465 static int __init find_isa_irq_pin(int irq, int type)
466 {
467 int i;
468
469 for (i = 0; i < mp_irq_entries; i++) {
470 int lbus = mp_irqs[i].mpc_srcbus;
471
472 if (test_bit(lbus, mp_bus_not_pci) &&
473 (mp_irqs[i].mpc_irqtype == type) &&
474 (mp_irqs[i].mpc_srcbusirq == irq))
475
476 return mp_irqs[i].mpc_dstirq;
477 }
478 return -1;
479 }
480
481 static int __init find_isa_irq_apic(int irq, int type)
482 {
483 int i;
484
485 for (i = 0; i < mp_irq_entries; i++) {
486 int lbus = mp_irqs[i].mpc_srcbus;
487
488 if (test_bit(lbus, mp_bus_not_pci) &&
489 (mp_irqs[i].mpc_irqtype == type) &&
490 (mp_irqs[i].mpc_srcbusirq == irq))
491 break;
492 }
493 if (i < mp_irq_entries) {
494 int apic;
495 for(apic = 0; apic < nr_ioapics; apic++) {
496 if (mp_ioapics[apic].mpc_apicid == mp_irqs[i].mpc_dstapic)
497 return apic;
498 }
499 }
500
501 return -1;
502 }
503
504 /*
505 * Find a specific PCI IRQ entry.
506 * Not an __init, possibly needed by modules
507 */
508 static int pin_2_irq(int idx, int apic, int pin);
509
510 int IO_APIC_get_PCI_irq_vector(int bus, int slot, int pin)
511 {
512 int apic, i, best_guess = -1;
513
514 apic_printk(APIC_DEBUG, "querying PCI -> IRQ mapping bus:%d, slot:%d, pin:%d.\n",
515 bus, slot, pin);
516 if (mp_bus_id_to_pci_bus[bus] == -1) {
517 apic_printk(APIC_VERBOSE, "PCI BIOS passed nonexistent PCI bus %d!\n", bus);
518 return -1;
519 }
520 for (i = 0; i < mp_irq_entries; i++) {
521 int lbus = mp_irqs[i].mpc_srcbus;
522
523 for (apic = 0; apic < nr_ioapics; apic++)
524 if (mp_ioapics[apic].mpc_apicid == mp_irqs[i].mpc_dstapic ||
525 mp_irqs[i].mpc_dstapic == MP_APIC_ALL)
526 break;
527
528 if (!test_bit(lbus, mp_bus_not_pci) &&
529 !mp_irqs[i].mpc_irqtype &&
530 (bus == lbus) &&
531 (slot == ((mp_irqs[i].mpc_srcbusirq >> 2) & 0x1f))) {
532 int irq = pin_2_irq(i,apic,mp_irqs[i].mpc_dstirq);
533
534 if (!(apic || IO_APIC_IRQ(irq)))
535 continue;
536
537 if (pin == (mp_irqs[i].mpc_srcbusirq & 3))
538 return irq;
539 /*
540 * Use the first all-but-pin matching entry as a
541 * best-guess fuzzy result for broken mptables.
542 */
543 if (best_guess < 0)
544 best_guess = irq;
545 }
546 }
547 BUG_ON(best_guess >= NR_IRQS);
548 return best_guess;
549 }
550
551 /* ISA interrupts are always polarity zero edge triggered,
552 * when listed as conforming in the MP table. */
553
554 #define default_ISA_trigger(idx) (0)
555 #define default_ISA_polarity(idx) (0)
556
557 /* PCI interrupts are always polarity one level triggered,
558 * when listed as conforming in the MP table. */
559
560 #define default_PCI_trigger(idx) (1)
561 #define default_PCI_polarity(idx) (1)
562
563 static int MPBIOS_polarity(int idx)
564 {
565 int bus = mp_irqs[idx].mpc_srcbus;
566 int polarity;
567
568 /*
569 * Determine IRQ line polarity (high active or low active):
570 */
571 switch (mp_irqs[idx].mpc_irqflag & 3)
572 {
573 case 0: /* conforms, ie. bus-type dependent polarity */
574 if (test_bit(bus, mp_bus_not_pci))
575 polarity = default_ISA_polarity(idx);
576 else
577 polarity = default_PCI_polarity(idx);
578 break;
579 case 1: /* high active */
580 {
581 polarity = 0;
582 break;
583 }
584 case 2: /* reserved */
585 {
586 printk(KERN_WARNING "broken BIOS!!\n");
587 polarity = 1;
588 break;
589 }
590 case 3: /* low active */
591 {
592 polarity = 1;
593 break;
594 }
595 default: /* invalid */
596 {
597 printk(KERN_WARNING "broken BIOS!!\n");
598 polarity = 1;
599 break;
600 }
601 }
602 return polarity;
603 }
604
605 static int MPBIOS_trigger(int idx)
606 {
607 int bus = mp_irqs[idx].mpc_srcbus;
608 int trigger;
609
610 /*
611 * Determine IRQ trigger mode (edge or level sensitive):
612 */
613 switch ((mp_irqs[idx].mpc_irqflag>>2) & 3)
614 {
615 case 0: /* conforms, ie. bus-type dependent */
616 if (test_bit(bus, mp_bus_not_pci))
617 trigger = default_ISA_trigger(idx);
618 else
619 trigger = default_PCI_trigger(idx);
620 break;
621 case 1: /* edge */
622 {
623 trigger = 0;
624 break;
625 }
626 case 2: /* reserved */
627 {
628 printk(KERN_WARNING "broken BIOS!!\n");
629 trigger = 1;
630 break;
631 }
632 case 3: /* level */
633 {
634 trigger = 1;
635 break;
636 }
637 default: /* invalid */
638 {
639 printk(KERN_WARNING "broken BIOS!!\n");
640 trigger = 0;
641 break;
642 }
643 }
644 return trigger;
645 }
646
647 static inline int irq_polarity(int idx)
648 {
649 return MPBIOS_polarity(idx);
650 }
651
652 static inline int irq_trigger(int idx)
653 {
654 return MPBIOS_trigger(idx);
655 }
656
657 static int pin_2_irq(int idx, int apic, int pin)
658 {
659 int irq, i;
660 int bus = mp_irqs[idx].mpc_srcbus;
661
662 /*
663 * Debugging check, we are in big trouble if this message pops up!
664 */
665 if (mp_irqs[idx].mpc_dstirq != pin)
666 printk(KERN_ERR "broken BIOS or MPTABLE parser, ayiee!!\n");
667
668 if (test_bit(bus, mp_bus_not_pci)) {
669 irq = mp_irqs[idx].mpc_srcbusirq;
670 } else {
671 /*
672 * PCI IRQs are mapped in order
673 */
674 i = irq = 0;
675 while (i < apic)
676 irq += nr_ioapic_registers[i++];
677 irq += pin;
678 }
679 BUG_ON(irq >= NR_IRQS);
680 return irq;
681 }
682
683 static int __assign_irq_vector(int irq, cpumask_t mask)
684 {
685 /*
686 * NOTE! The local APIC isn't very good at handling
687 * multiple interrupts at the same interrupt level.
688 * As the interrupt level is determined by taking the
689 * vector number and shifting that right by 4, we
690 * want to spread these out a bit so that they don't
691 * all fall in the same interrupt level.
692 *
693 * Also, we've got to be careful not to trash gate
694 * 0x80, because int 0x80 is hm, kind of importantish. ;)
695 */
696 static int current_vector = FIRST_DEVICE_VECTOR, current_offset = 0;
697 unsigned int old_vector;
698 int cpu;
699 struct irq_cfg *cfg;
700
701 BUG_ON((unsigned)irq >= NR_IRQS);
702 cfg = &irq_cfg[irq];
703
704 /* Only try and allocate irqs on cpus that are present */
705 cpus_and(mask, mask, cpu_online_map);
706
707 if ((cfg->move_in_progress) || cfg->move_cleanup_count)
708 return -EBUSY;
709
710 old_vector = cfg->vector;
711 if (old_vector) {
712 cpumask_t tmp;
713 cpus_and(tmp, cfg->domain, mask);
714 if (!cpus_empty(tmp))
715 return 0;
716 }
717
718 for_each_cpu_mask(cpu, mask) {
719 cpumask_t domain, new_mask;
720 int new_cpu;
721 int vector, offset;
722
723 domain = vector_allocation_domain(cpu);
724 cpus_and(new_mask, domain, cpu_online_map);
725
726 vector = current_vector;
727 offset = current_offset;
728 next:
729 vector += 8;
730 if (vector >= FIRST_SYSTEM_VECTOR) {
731 /* If we run out of vectors on large boxen, must share them. */
732 offset = (offset + 1) % 8;
733 vector = FIRST_DEVICE_VECTOR + offset;
734 }
735 if (unlikely(current_vector == vector))
736 continue;
737 if (vector == IA32_SYSCALL_VECTOR)
738 goto next;
739 for_each_cpu_mask(new_cpu, new_mask)
740 if (per_cpu(vector_irq, new_cpu)[vector] != -1)
741 goto next;
742 /* Found one! */
743 current_vector = vector;
744 current_offset = offset;
745 if (old_vector) {
746 cfg->move_in_progress = 1;
747 cfg->old_domain = cfg->domain;
748 }
749 for_each_cpu_mask(new_cpu, new_mask)
750 per_cpu(vector_irq, new_cpu)[vector] = irq;
751 cfg->vector = vector;
752 cfg->domain = domain;
753 return 0;
754 }
755 return -ENOSPC;
756 }
757
758 static int assign_irq_vector(int irq, cpumask_t mask)
759 {
760 int err;
761 unsigned long flags;
762
763 spin_lock_irqsave(&vector_lock, flags);
764 err = __assign_irq_vector(irq, mask);
765 spin_unlock_irqrestore(&vector_lock, flags);
766 return err;
767 }
768
769 static void __clear_irq_vector(int irq)
770 {
771 struct irq_cfg *cfg;
772 cpumask_t mask;
773 int cpu, vector;
774
775 BUG_ON((unsigned)irq >= NR_IRQS);
776 cfg = &irq_cfg[irq];
777 BUG_ON(!cfg->vector);
778
779 vector = cfg->vector;
780 cpus_and(mask, cfg->domain, cpu_online_map);
781 for_each_cpu_mask(cpu, mask)
782 per_cpu(vector_irq, cpu)[vector] = -1;
783
784 cfg->vector = 0;
785 cfg->domain = CPU_MASK_NONE;
786 }
787
788 void __setup_vector_irq(int cpu)
789 {
790 /* Initialize vector_irq on a new cpu */
791 /* This function must be called with vector_lock held */
792 int irq, vector;
793
794 /* Mark the inuse vectors */
795 for (irq = 0; irq < NR_IRQS; ++irq) {
796 if (!cpu_isset(cpu, irq_cfg[irq].domain))
797 continue;
798 vector = irq_cfg[irq].vector;
799 per_cpu(vector_irq, cpu)[vector] = irq;
800 }
801 /* Mark the free vectors */
802 for (vector = 0; vector < NR_VECTORS; ++vector) {
803 irq = per_cpu(vector_irq, cpu)[vector];
804 if (irq < 0)
805 continue;
806 if (!cpu_isset(cpu, irq_cfg[irq].domain))
807 per_cpu(vector_irq, cpu)[vector] = -1;
808 }
809 }
810
811
812 static struct irq_chip ioapic_chip;
813
814 static void ioapic_register_intr(int irq, unsigned long trigger)
815 {
816 if (trigger) {
817 irq_desc[irq].status |= IRQ_LEVEL;
818 set_irq_chip_and_handler_name(irq, &ioapic_chip,
819 handle_fasteoi_irq, "fasteoi");
820 } else {
821 irq_desc[irq].status &= ~IRQ_LEVEL;
822 set_irq_chip_and_handler_name(irq, &ioapic_chip,
823 handle_edge_irq, "edge");
824 }
825 }
826
827 static void setup_IO_APIC_irq(int apic, int pin, unsigned int irq,
828 int trigger, int polarity)
829 {
830 struct irq_cfg *cfg = irq_cfg + irq;
831 struct IO_APIC_route_entry entry;
832 cpumask_t mask;
833
834 if (!IO_APIC_IRQ(irq))
835 return;
836
837 mask = TARGET_CPUS;
838 if (assign_irq_vector(irq, mask))
839 return;
840
841 cpus_and(mask, cfg->domain, mask);
842
843 apic_printk(APIC_VERBOSE,KERN_DEBUG
844 "IOAPIC[%d]: Set routing entry (%d-%d -> 0x%x -> "
845 "IRQ %d Mode:%i Active:%i)\n",
846 apic, mp_ioapics[apic].mpc_apicid, pin, cfg->vector,
847 irq, trigger, polarity);
848
849 /*
850 * add it to the IO-APIC irq-routing table:
851 */
852 memset(&entry,0,sizeof(entry));
853
854 entry.delivery_mode = INT_DELIVERY_MODE;
855 entry.dest_mode = INT_DEST_MODE;
856 entry.dest = cpu_mask_to_apicid(mask);
857 entry.mask = 0; /* enable IRQ */
858 entry.trigger = trigger;
859 entry.polarity = polarity;
860 entry.vector = cfg->vector;
861
862 /* Mask level triggered irqs.
863 * Use IRQ_DELAYED_DISABLE for edge triggered irqs.
864 */
865 if (trigger)
866 entry.mask = 1;
867
868 ioapic_register_intr(irq, trigger);
869 if (irq < 16)
870 disable_8259A_irq(irq);
871
872 ioapic_write_entry(apic, pin, entry);
873 }
874
875 static void __init setup_IO_APIC_irqs(void)
876 {
877 int apic, pin, idx, irq, first_notcon = 1;
878
879 apic_printk(APIC_VERBOSE, KERN_DEBUG "init IO_APIC IRQs\n");
880
881 for (apic = 0; apic < nr_ioapics; apic++) {
882 for (pin = 0; pin < nr_ioapic_registers[apic]; pin++) {
883
884 idx = find_irq_entry(apic,pin,mp_INT);
885 if (idx == -1) {
886 if (first_notcon) {
887 apic_printk(APIC_VERBOSE, KERN_DEBUG " IO-APIC (apicid-pin) %d-%d", mp_ioapics[apic].mpc_apicid, pin);
888 first_notcon = 0;
889 } else
890 apic_printk(APIC_VERBOSE, ", %d-%d", mp_ioapics[apic].mpc_apicid, pin);
891 continue;
892 }
893 if (!first_notcon) {
894 apic_printk(APIC_VERBOSE, " not connected.\n");
895 first_notcon = 1;
896 }
897
898 irq = pin_2_irq(idx, apic, pin);
899 add_pin_to_irq(irq, apic, pin);
900
901 setup_IO_APIC_irq(apic, pin, irq,
902 irq_trigger(idx), irq_polarity(idx));
903 }
904 }
905
906 if (!first_notcon)
907 apic_printk(APIC_VERBOSE, " not connected.\n");
908 }
909
910 /*
911 * Set up the 8259A-master output pin as broadcast to all
912 * CPUs.
913 */
914 static void __init setup_ExtINT_IRQ0_pin(unsigned int apic, unsigned int pin, int vector)
915 {
916 struct IO_APIC_route_entry entry;
917 unsigned long flags;
918
919 memset(&entry,0,sizeof(entry));
920
921 disable_8259A_irq(0);
922
923 /* mask LVT0 */
924 apic_write(APIC_LVT0, APIC_LVT_MASKED | APIC_DM_EXTINT);
925
926 /*
927 * We use logical delivery to get the timer IRQ
928 * to the first CPU.
929 */
930 entry.dest_mode = INT_DEST_MODE;
931 entry.mask = 0; /* unmask IRQ now */
932 entry.dest = cpu_mask_to_apicid(TARGET_CPUS);
933 entry.delivery_mode = INT_DELIVERY_MODE;
934 entry.polarity = 0;
935 entry.trigger = 0;
936 entry.vector = vector;
937
938 /*
939 * The timer IRQ doesn't have to know that behind the
940 * scene we have a 8259A-master in AEOI mode ...
941 */
942 set_irq_chip_and_handler_name(0, &ioapic_chip, handle_edge_irq, "edge");
943
944 /*
945 * Add it to the IO-APIC irq-routing table:
946 */
947 spin_lock_irqsave(&ioapic_lock, flags);
948 io_apic_write(apic, 0x11+2*pin, *(((int *)&entry)+1));
949 io_apic_write(apic, 0x10+2*pin, *(((int *)&entry)+0));
950 spin_unlock_irqrestore(&ioapic_lock, flags);
951
952 enable_8259A_irq(0);
953 }
954
955 void __apicdebuginit print_IO_APIC(void)
956 {
957 int apic, i;
958 union IO_APIC_reg_00 reg_00;
959 union IO_APIC_reg_01 reg_01;
960 union IO_APIC_reg_02 reg_02;
961 unsigned long flags;
962
963 if (apic_verbosity == APIC_QUIET)
964 return;
965
966 printk(KERN_DEBUG "number of MP IRQ sources: %d.\n", mp_irq_entries);
967 for (i = 0; i < nr_ioapics; i++)
968 printk(KERN_DEBUG "number of IO-APIC #%d registers: %d.\n",
969 mp_ioapics[i].mpc_apicid, nr_ioapic_registers[i]);
970
971 /*
972 * We are a bit conservative about what we expect. We have to
973 * know about every hardware change ASAP.
974 */
975 printk(KERN_INFO "testing the IO APIC.......................\n");
976
977 for (apic = 0; apic < nr_ioapics; apic++) {
978
979 spin_lock_irqsave(&ioapic_lock, flags);
980 reg_00.raw = io_apic_read(apic, 0);
981 reg_01.raw = io_apic_read(apic, 1);
982 if (reg_01.bits.version >= 0x10)
983 reg_02.raw = io_apic_read(apic, 2);
984 spin_unlock_irqrestore(&ioapic_lock, flags);
985
986 printk("\n");
987 printk(KERN_DEBUG "IO APIC #%d......\n", mp_ioapics[apic].mpc_apicid);
988 printk(KERN_DEBUG ".... register #00: %08X\n", reg_00.raw);
989 printk(KERN_DEBUG "....... : physical APIC id: %02X\n", reg_00.bits.ID);
990
991 printk(KERN_DEBUG ".... register #01: %08X\n", *(int *)&reg_01);
992 printk(KERN_DEBUG "....... : max redirection entries: %04X\n", reg_01.bits.entries);
993
994 printk(KERN_DEBUG "....... : PRQ implemented: %X\n", reg_01.bits.PRQ);
995 printk(KERN_DEBUG "....... : IO APIC version: %04X\n", reg_01.bits.version);
996
997 if (reg_01.bits.version >= 0x10) {
998 printk(KERN_DEBUG ".... register #02: %08X\n", reg_02.raw);
999 printk(KERN_DEBUG "....... : arbitration: %02X\n", reg_02.bits.arbitration);
1000 }
1001
1002 printk(KERN_DEBUG ".... IRQ redirection table:\n");
1003
1004 printk(KERN_DEBUG " NR Dst Mask Trig IRR Pol"
1005 " Stat Dmod Deli Vect: \n");
1006
1007 for (i = 0; i <= reg_01.bits.entries; i++) {
1008 struct IO_APIC_route_entry entry;
1009
1010 entry = ioapic_read_entry(apic, i);
1011
1012 printk(KERN_DEBUG " %02x %03X ",
1013 i,
1014 entry.dest
1015 );
1016
1017 printk("%1d %1d %1d %1d %1d %1d %1d %02X\n",
1018 entry.mask,
1019 entry.trigger,
1020 entry.irr,
1021 entry.polarity,
1022 entry.delivery_status,
1023 entry.dest_mode,
1024 entry.delivery_mode,
1025 entry.vector
1026 );
1027 }
1028 }
1029 printk(KERN_DEBUG "IRQ to pin mappings:\n");
1030 for (i = 0; i < NR_IRQS; i++) {
1031 struct irq_pin_list *entry = irq_2_pin + i;
1032 if (entry->pin < 0)
1033 continue;
1034 printk(KERN_DEBUG "IRQ%d ", i);
1035 for (;;) {
1036 printk("-> %d:%d", entry->apic, entry->pin);
1037 if (!entry->next)
1038 break;
1039 entry = irq_2_pin + entry->next;
1040 }
1041 printk("\n");
1042 }
1043
1044 printk(KERN_INFO ".................................... done.\n");
1045
1046 return;
1047 }
1048
1049 #if 0
1050
1051 static __apicdebuginit void print_APIC_bitfield (int base)
1052 {
1053 unsigned int v;
1054 int i, j;
1055
1056 if (apic_verbosity == APIC_QUIET)
1057 return;
1058
1059 printk(KERN_DEBUG "0123456789abcdef0123456789abcdef\n" KERN_DEBUG);
1060 for (i = 0; i < 8; i++) {
1061 v = apic_read(base + i*0x10);
1062 for (j = 0; j < 32; j++) {
1063 if (v & (1<<j))
1064 printk("1");
1065 else
1066 printk("0");
1067 }
1068 printk("\n");
1069 }
1070 }
1071
1072 void __apicdebuginit print_local_APIC(void * dummy)
1073 {
1074 unsigned int v, ver, maxlvt;
1075
1076 if (apic_verbosity == APIC_QUIET)
1077 return;
1078
1079 printk("\n" KERN_DEBUG "printing local APIC contents on CPU#%d/%d:\n",
1080 smp_processor_id(), hard_smp_processor_id());
1081 printk(KERN_INFO "... APIC ID: %08x (%01x)\n", v, GET_APIC_ID(read_apic_id()));
1082 v = apic_read(APIC_LVR);
1083 printk(KERN_INFO "... APIC VERSION: %08x\n", v);
1084 ver = GET_APIC_VERSION(v);
1085 maxlvt = lapic_get_maxlvt();
1086
1087 v = apic_read(APIC_TASKPRI);
1088 printk(KERN_DEBUG "... APIC TASKPRI: %08x (%02x)\n", v, v & APIC_TPRI_MASK);
1089
1090 v = apic_read(APIC_ARBPRI);
1091 printk(KERN_DEBUG "... APIC ARBPRI: %08x (%02x)\n", v,
1092 v & APIC_ARBPRI_MASK);
1093 v = apic_read(APIC_PROCPRI);
1094 printk(KERN_DEBUG "... APIC PROCPRI: %08x\n", v);
1095
1096 v = apic_read(APIC_EOI);
1097 printk(KERN_DEBUG "... APIC EOI: %08x\n", v);
1098 v = apic_read(APIC_RRR);
1099 printk(KERN_DEBUG "... APIC RRR: %08x\n", v);
1100 v = apic_read(APIC_LDR);
1101 printk(KERN_DEBUG "... APIC LDR: %08x\n", v);
1102 v = apic_read(APIC_DFR);
1103 printk(KERN_DEBUG "... APIC DFR: %08x\n", v);
1104 v = apic_read(APIC_SPIV);
1105 printk(KERN_DEBUG "... APIC SPIV: %08x\n", v);
1106
1107 printk(KERN_DEBUG "... APIC ISR field:\n");
1108 print_APIC_bitfield(APIC_ISR);
1109 printk(KERN_DEBUG "... APIC TMR field:\n");
1110 print_APIC_bitfield(APIC_TMR);
1111 printk(KERN_DEBUG "... APIC IRR field:\n");
1112 print_APIC_bitfield(APIC_IRR);
1113
1114 v = apic_read(APIC_ESR);
1115 printk(KERN_DEBUG "... APIC ESR: %08x\n", v);
1116
1117 v = apic_read(APIC_ICR);
1118 printk(KERN_DEBUG "... APIC ICR: %08x\n", v);
1119 v = apic_read(APIC_ICR2);
1120 printk(KERN_DEBUG "... APIC ICR2: %08x\n", v);
1121
1122 v = apic_read(APIC_LVTT);
1123 printk(KERN_DEBUG "... APIC LVTT: %08x\n", v);
1124
1125 if (maxlvt > 3) { /* PC is LVT#4. */
1126 v = apic_read(APIC_LVTPC);
1127 printk(KERN_DEBUG "... APIC LVTPC: %08x\n", v);
1128 }
1129 v = apic_read(APIC_LVT0);
1130 printk(KERN_DEBUG "... APIC LVT0: %08x\n", v);
1131 v = apic_read(APIC_LVT1);
1132 printk(KERN_DEBUG "... APIC LVT1: %08x\n", v);
1133
1134 if (maxlvt > 2) { /* ERR is LVT#3. */
1135 v = apic_read(APIC_LVTERR);
1136 printk(KERN_DEBUG "... APIC LVTERR: %08x\n", v);
1137 }
1138
1139 v = apic_read(APIC_TMICT);
1140 printk(KERN_DEBUG "... APIC TMICT: %08x\n", v);
1141 v = apic_read(APIC_TMCCT);
1142 printk(KERN_DEBUG "... APIC TMCCT: %08x\n", v);
1143 v = apic_read(APIC_TDCR);
1144 printk(KERN_DEBUG "... APIC TDCR: %08x\n", v);
1145 printk("\n");
1146 }
1147
1148 void print_all_local_APICs (void)
1149 {
1150 on_each_cpu(print_local_APIC, NULL, 1, 1);
1151 }
1152
1153 void __apicdebuginit print_PIC(void)
1154 {
1155 unsigned int v;
1156 unsigned long flags;
1157
1158 if (apic_verbosity == APIC_QUIET)
1159 return;
1160
1161 printk(KERN_DEBUG "\nprinting PIC contents\n");
1162
1163 spin_lock_irqsave(&i8259A_lock, flags);
1164
1165 v = inb(0xa1) << 8 | inb(0x21);
1166 printk(KERN_DEBUG "... PIC IMR: %04x\n", v);
1167
1168 v = inb(0xa0) << 8 | inb(0x20);
1169 printk(KERN_DEBUG "... PIC IRR: %04x\n", v);
1170
1171 outb(0x0b,0xa0);
1172 outb(0x0b,0x20);
1173 v = inb(0xa0) << 8 | inb(0x20);
1174 outb(0x0a,0xa0);
1175 outb(0x0a,0x20);
1176
1177 spin_unlock_irqrestore(&i8259A_lock, flags);
1178
1179 printk(KERN_DEBUG "... PIC ISR: %04x\n", v);
1180
1181 v = inb(0x4d1) << 8 | inb(0x4d0);
1182 printk(KERN_DEBUG "... PIC ELCR: %04x\n", v);
1183 }
1184
1185 #endif /* 0 */
1186
1187 void __init enable_IO_APIC(void)
1188 {
1189 union IO_APIC_reg_01 reg_01;
1190 int i8259_apic, i8259_pin;
1191 int i, apic;
1192 unsigned long flags;
1193
1194 for (i = 0; i < PIN_MAP_SIZE; i++) {
1195 irq_2_pin[i].pin = -1;
1196 irq_2_pin[i].next = 0;
1197 }
1198
1199 /*
1200 * The number of IO-APIC IRQ registers (== #pins):
1201 */
1202 for (apic = 0; apic < nr_ioapics; apic++) {
1203 spin_lock_irqsave(&ioapic_lock, flags);
1204 reg_01.raw = io_apic_read(apic, 1);
1205 spin_unlock_irqrestore(&ioapic_lock, flags);
1206 nr_ioapic_registers[apic] = reg_01.bits.entries+1;
1207 }
1208 for(apic = 0; apic < nr_ioapics; apic++) {
1209 int pin;
1210 /* See if any of the pins is in ExtINT mode */
1211 for (pin = 0; pin < nr_ioapic_registers[apic]; pin++) {
1212 struct IO_APIC_route_entry entry;
1213 entry = ioapic_read_entry(apic, pin);
1214
1215 /* If the interrupt line is enabled and in ExtInt mode
1216 * I have found the pin where the i8259 is connected.
1217 */
1218 if ((entry.mask == 0) && (entry.delivery_mode == dest_ExtINT)) {
1219 ioapic_i8259.apic = apic;
1220 ioapic_i8259.pin = pin;
1221 goto found_i8259;
1222 }
1223 }
1224 }
1225 found_i8259:
1226 /* Look to see what if the MP table has reported the ExtINT */
1227 i8259_pin = find_isa_irq_pin(0, mp_ExtINT);
1228 i8259_apic = find_isa_irq_apic(0, mp_ExtINT);
1229 /* Trust the MP table if nothing is setup in the hardware */
1230 if ((ioapic_i8259.pin == -1) && (i8259_pin >= 0)) {
1231 printk(KERN_WARNING "ExtINT not setup in hardware but reported by MP table\n");
1232 ioapic_i8259.pin = i8259_pin;
1233 ioapic_i8259.apic = i8259_apic;
1234 }
1235 /* Complain if the MP table and the hardware disagree */
1236 if (((ioapic_i8259.apic != i8259_apic) || (ioapic_i8259.pin != i8259_pin)) &&
1237 (i8259_pin >= 0) && (ioapic_i8259.pin >= 0))
1238 {
1239 printk(KERN_WARNING "ExtINT in hardware and MP table differ\n");
1240 }
1241
1242 /*
1243 * Do not trust the IO-APIC being empty at bootup
1244 */
1245 clear_IO_APIC();
1246 }
1247
1248 /*
1249 * Not an __init, needed by the reboot code
1250 */
1251 void disable_IO_APIC(void)
1252 {
1253 /*
1254 * Clear the IO-APIC before rebooting:
1255 */
1256 clear_IO_APIC();
1257
1258 /*
1259 * If the i8259 is routed through an IOAPIC
1260 * Put that IOAPIC in virtual wire mode
1261 * so legacy interrupts can be delivered.
1262 */
1263 if (ioapic_i8259.pin != -1) {
1264 struct IO_APIC_route_entry entry;
1265
1266 memset(&entry, 0, sizeof(entry));
1267 entry.mask = 0; /* Enabled */
1268 entry.trigger = 0; /* Edge */
1269 entry.irr = 0;
1270 entry.polarity = 0; /* High */
1271 entry.delivery_status = 0;
1272 entry.dest_mode = 0; /* Physical */
1273 entry.delivery_mode = dest_ExtINT; /* ExtInt */
1274 entry.vector = 0;
1275 entry.dest = GET_APIC_ID(read_apic_id());
1276
1277 /*
1278 * Add it to the IO-APIC irq-routing table:
1279 */
1280 ioapic_write_entry(ioapic_i8259.apic, ioapic_i8259.pin, entry);
1281 }
1282
1283 disconnect_bsp_APIC(ioapic_i8259.pin != -1);
1284 }
1285
1286 /*
1287 * There is a nasty bug in some older SMP boards, their mptable lies
1288 * about the timer IRQ. We do the following to work around the situation:
1289 *
1290 * - timer IRQ defaults to IO-APIC IRQ
1291 * - if this function detects that timer IRQs are defunct, then we fall
1292 * back to ISA timer IRQs
1293 */
1294 static int __init timer_irq_works(void)
1295 {
1296 unsigned long t1 = jiffies;
1297 unsigned long flags;
1298
1299 local_save_flags(flags);
1300 local_irq_enable();
1301 /* Let ten ticks pass... */
1302 mdelay((10 * 1000) / HZ);
1303 local_irq_restore(flags);
1304
1305 /*
1306 * Expect a few ticks at least, to be sure some possible
1307 * glue logic does not lock up after one or two first
1308 * ticks in a non-ExtINT mode. Also the local APIC
1309 * might have cached one ExtINT interrupt. Finally, at
1310 * least one tick may be lost due to delays.
1311 */
1312
1313 /* jiffies wrap? */
1314 if (time_after(jiffies, t1 + 4))
1315 return 1;
1316 return 0;
1317 }
1318
1319 /*
1320 * In the SMP+IOAPIC case it might happen that there are an unspecified
1321 * number of pending IRQ events unhandled. These cases are very rare,
1322 * so we 'resend' these IRQs via IPIs, to the same CPU. It's much
1323 * better to do it this way as thus we do not have to be aware of
1324 * 'pending' interrupts in the IRQ path, except at this point.
1325 */
1326 /*
1327 * Edge triggered needs to resend any interrupt
1328 * that was delayed but this is now handled in the device
1329 * independent code.
1330 */
1331
1332 /*
1333 * Starting up a edge-triggered IO-APIC interrupt is
1334 * nasty - we need to make sure that we get the edge.
1335 * If it is already asserted for some reason, we need
1336 * return 1 to indicate that is was pending.
1337 *
1338 * This is not complete - we should be able to fake
1339 * an edge even if it isn't on the 8259A...
1340 */
1341
1342 static unsigned int startup_ioapic_irq(unsigned int irq)
1343 {
1344 int was_pending = 0;
1345 unsigned long flags;
1346
1347 spin_lock_irqsave(&ioapic_lock, flags);
1348 if (irq < 16) {
1349 disable_8259A_irq(irq);
1350 if (i8259A_irq_pending(irq))
1351 was_pending = 1;
1352 }
1353 __unmask_IO_APIC_irq(irq);
1354 spin_unlock_irqrestore(&ioapic_lock, flags);
1355
1356 return was_pending;
1357 }
1358
1359 static int ioapic_retrigger_irq(unsigned int irq)
1360 {
1361 struct irq_cfg *cfg = &irq_cfg[irq];
1362 cpumask_t mask;
1363 unsigned long flags;
1364
1365 spin_lock_irqsave(&vector_lock, flags);
1366 cpus_clear(mask);
1367 cpu_set(first_cpu(cfg->domain), mask);
1368
1369 send_IPI_mask(mask, cfg->vector);
1370 spin_unlock_irqrestore(&vector_lock, flags);
1371
1372 return 1;
1373 }
1374
1375 /*
1376 * Level and edge triggered IO-APIC interrupts need different handling,
1377 * so we use two separate IRQ descriptors. Edge triggered IRQs can be
1378 * handled with the level-triggered descriptor, but that one has slightly
1379 * more overhead. Level-triggered interrupts cannot be handled with the
1380 * edge-triggered handler, without risking IRQ storms and other ugly
1381 * races.
1382 */
1383
1384 #ifdef CONFIG_SMP
1385 asmlinkage void smp_irq_move_cleanup_interrupt(void)
1386 {
1387 unsigned vector, me;
1388 ack_APIC_irq();
1389 exit_idle();
1390 irq_enter();
1391
1392 me = smp_processor_id();
1393 for (vector = FIRST_EXTERNAL_VECTOR; vector < NR_VECTORS; vector++) {
1394 unsigned int irq;
1395 struct irq_desc *desc;
1396 struct irq_cfg *cfg;
1397 irq = __get_cpu_var(vector_irq)[vector];
1398 if (irq >= NR_IRQS)
1399 continue;
1400
1401 desc = irq_desc + irq;
1402 cfg = irq_cfg + irq;
1403 spin_lock(&desc->lock);
1404 if (!cfg->move_cleanup_count)
1405 goto unlock;
1406
1407 if ((vector == cfg->vector) && cpu_isset(me, cfg->domain))
1408 goto unlock;
1409
1410 __get_cpu_var(vector_irq)[vector] = -1;
1411 cfg->move_cleanup_count--;
1412 unlock:
1413 spin_unlock(&desc->lock);
1414 }
1415
1416 irq_exit();
1417 }
1418
1419 static void irq_complete_move(unsigned int irq)
1420 {
1421 struct irq_cfg *cfg = irq_cfg + irq;
1422 unsigned vector, me;
1423
1424 if (likely(!cfg->move_in_progress))
1425 return;
1426
1427 vector = ~get_irq_regs()->orig_ax;
1428 me = smp_processor_id();
1429 if ((vector == cfg->vector) && cpu_isset(me, cfg->domain)) {
1430 cpumask_t cleanup_mask;
1431
1432 cpus_and(cleanup_mask, cfg->old_domain, cpu_online_map);
1433 cfg->move_cleanup_count = cpus_weight(cleanup_mask);
1434 send_IPI_mask(cleanup_mask, IRQ_MOVE_CLEANUP_VECTOR);
1435 cfg->move_in_progress = 0;
1436 }
1437 }
1438 #else
1439 static inline void irq_complete_move(unsigned int irq) {}
1440 #endif
1441
1442 static void ack_apic_edge(unsigned int irq)
1443 {
1444 irq_complete_move(irq);
1445 move_native_irq(irq);
1446 ack_APIC_irq();
1447 }
1448
1449 static void ack_apic_level(unsigned int irq)
1450 {
1451 int do_unmask_irq = 0;
1452
1453 irq_complete_move(irq);
1454 #ifdef CONFIG_GENERIC_PENDING_IRQ
1455 /* If we are moving the irq we need to mask it */
1456 if (unlikely(irq_desc[irq].status & IRQ_MOVE_PENDING)) {
1457 do_unmask_irq = 1;
1458 mask_IO_APIC_irq(irq);
1459 }
1460 #endif
1461
1462 /*
1463 * We must acknowledge the irq before we move it or the acknowledge will
1464 * not propagate properly.
1465 */
1466 ack_APIC_irq();
1467
1468 /* Now we can move and renable the irq */
1469 if (unlikely(do_unmask_irq)) {
1470 /* Only migrate the irq if the ack has been received.
1471 *
1472 * On rare occasions the broadcast level triggered ack gets
1473 * delayed going to ioapics, and if we reprogram the
1474 * vector while Remote IRR is still set the irq will never
1475 * fire again.
1476 *
1477 * To prevent this scenario we read the Remote IRR bit
1478 * of the ioapic. This has two effects.
1479 * - On any sane system the read of the ioapic will
1480 * flush writes (and acks) going to the ioapic from
1481 * this cpu.
1482 * - We get to see if the ACK has actually been delivered.
1483 *
1484 * Based on failed experiments of reprogramming the
1485 * ioapic entry from outside of irq context starting
1486 * with masking the ioapic entry and then polling until
1487 * Remote IRR was clear before reprogramming the
1488 * ioapic I don't trust the Remote IRR bit to be
1489 * completey accurate.
1490 *
1491 * However there appears to be no other way to plug
1492 * this race, so if the Remote IRR bit is not
1493 * accurate and is causing problems then it is a hardware bug
1494 * and you can go talk to the chipset vendor about it.
1495 */
1496 if (!io_apic_level_ack_pending(irq))
1497 move_masked_irq(irq);
1498 unmask_IO_APIC_irq(irq);
1499 }
1500 }
1501
1502 static struct irq_chip ioapic_chip __read_mostly = {
1503 .name = "IO-APIC",
1504 .startup = startup_ioapic_irq,
1505 .mask = mask_IO_APIC_irq,
1506 .unmask = unmask_IO_APIC_irq,
1507 .ack = ack_apic_edge,
1508 .eoi = ack_apic_level,
1509 #ifdef CONFIG_SMP
1510 .set_affinity = set_ioapic_affinity_irq,
1511 #endif
1512 .retrigger = ioapic_retrigger_irq,
1513 };
1514
1515 static inline void init_IO_APIC_traps(void)
1516 {
1517 int irq;
1518
1519 /*
1520 * NOTE! The local APIC isn't very good at handling
1521 * multiple interrupts at the same interrupt level.
1522 * As the interrupt level is determined by taking the
1523 * vector number and shifting that right by 4, we
1524 * want to spread these out a bit so that they don't
1525 * all fall in the same interrupt level.
1526 *
1527 * Also, we've got to be careful not to trash gate
1528 * 0x80, because int 0x80 is hm, kind of importantish. ;)
1529 */
1530 for (irq = 0; irq < NR_IRQS ; irq++) {
1531 int tmp = irq;
1532 if (IO_APIC_IRQ(tmp) && !irq_cfg[tmp].vector) {
1533 /*
1534 * Hmm.. We don't have an entry for this,
1535 * so default to an old-fashioned 8259
1536 * interrupt if we can..
1537 */
1538 if (irq < 16)
1539 make_8259A_irq(irq);
1540 else
1541 /* Strange. Oh, well.. */
1542 irq_desc[irq].chip = &no_irq_chip;
1543 }
1544 }
1545 }
1546
1547 static void enable_lapic_irq (unsigned int irq)
1548 {
1549 unsigned long v;
1550
1551 v = apic_read(APIC_LVT0);
1552 apic_write(APIC_LVT0, v & ~APIC_LVT_MASKED);
1553 }
1554
1555 static void disable_lapic_irq (unsigned int irq)
1556 {
1557 unsigned long v;
1558
1559 v = apic_read(APIC_LVT0);
1560 apic_write(APIC_LVT0, v | APIC_LVT_MASKED);
1561 }
1562
1563 static void ack_lapic_irq (unsigned int irq)
1564 {
1565 ack_APIC_irq();
1566 }
1567
1568 static void end_lapic_irq (unsigned int i) { /* nothing */ }
1569
1570 static struct hw_interrupt_type lapic_irq_type __read_mostly = {
1571 .name = "local-APIC",
1572 .typename = "local-APIC-edge",
1573 .startup = NULL, /* startup_irq() not used for IRQ0 */
1574 .shutdown = NULL, /* shutdown_irq() not used for IRQ0 */
1575 .enable = enable_lapic_irq,
1576 .disable = disable_lapic_irq,
1577 .ack = ack_lapic_irq,
1578 .end = end_lapic_irq,
1579 };
1580
1581 static void __init setup_nmi(void)
1582 {
1583 /*
1584 * Dirty trick to enable the NMI watchdog ...
1585 * We put the 8259A master into AEOI mode and
1586 * unmask on all local APICs LVT0 as NMI.
1587 *
1588 * The idea to use the 8259A in AEOI mode ('8259A Virtual Wire')
1589 * is from Maciej W. Rozycki - so we do not have to EOI from
1590 * the NMI handler or the timer interrupt.
1591 */
1592 printk(KERN_INFO "activating NMI Watchdog ...");
1593
1594 enable_NMI_through_LVT0();
1595
1596 printk(" done.\n");
1597 }
1598
1599 /*
1600 * This looks a bit hackish but it's about the only one way of sending
1601 * a few INTA cycles to 8259As and any associated glue logic. ICR does
1602 * not support the ExtINT mode, unfortunately. We need to send these
1603 * cycles as some i82489DX-based boards have glue logic that keeps the
1604 * 8259A interrupt line asserted until INTA. --macro
1605 */
1606 static inline void unlock_ExtINT_logic(void)
1607 {
1608 int apic, pin, i;
1609 struct IO_APIC_route_entry entry0, entry1;
1610 unsigned char save_control, save_freq_select;
1611 unsigned long flags;
1612
1613 pin = find_isa_irq_pin(8, mp_INT);
1614 apic = find_isa_irq_apic(8, mp_INT);
1615 if (pin == -1)
1616 return;
1617
1618 spin_lock_irqsave(&ioapic_lock, flags);
1619 *(((int *)&entry0) + 1) = io_apic_read(apic, 0x11 + 2 * pin);
1620 *(((int *)&entry0) + 0) = io_apic_read(apic, 0x10 + 2 * pin);
1621 spin_unlock_irqrestore(&ioapic_lock, flags);
1622 clear_IO_APIC_pin(apic, pin);
1623
1624 memset(&entry1, 0, sizeof(entry1));
1625
1626 entry1.dest_mode = 0; /* physical delivery */
1627 entry1.mask = 0; /* unmask IRQ now */
1628 entry1.dest = hard_smp_processor_id();
1629 entry1.delivery_mode = dest_ExtINT;
1630 entry1.polarity = entry0.polarity;
1631 entry1.trigger = 0;
1632 entry1.vector = 0;
1633
1634 spin_lock_irqsave(&ioapic_lock, flags);
1635 io_apic_write(apic, 0x11 + 2 * pin, *(((int *)&entry1) + 1));
1636 io_apic_write(apic, 0x10 + 2 * pin, *(((int *)&entry1) + 0));
1637 spin_unlock_irqrestore(&ioapic_lock, flags);
1638
1639 save_control = CMOS_READ(RTC_CONTROL);
1640 save_freq_select = CMOS_READ(RTC_FREQ_SELECT);
1641 CMOS_WRITE((save_freq_select & ~RTC_RATE_SELECT) | 0x6,
1642 RTC_FREQ_SELECT);
1643 CMOS_WRITE(save_control | RTC_PIE, RTC_CONTROL);
1644
1645 i = 100;
1646 while (i-- > 0) {
1647 mdelay(10);
1648 if ((CMOS_READ(RTC_INTR_FLAGS) & RTC_PF) == RTC_PF)
1649 i -= 10;
1650 }
1651
1652 CMOS_WRITE(save_control, RTC_CONTROL);
1653 CMOS_WRITE(save_freq_select, RTC_FREQ_SELECT);
1654 clear_IO_APIC_pin(apic, pin);
1655
1656 spin_lock_irqsave(&ioapic_lock, flags);
1657 io_apic_write(apic, 0x11 + 2 * pin, *(((int *)&entry0) + 1));
1658 io_apic_write(apic, 0x10 + 2 * pin, *(((int *)&entry0) + 0));
1659 spin_unlock_irqrestore(&ioapic_lock, flags);
1660 }
1661
1662 /*
1663 * This code may look a bit paranoid, but it's supposed to cooperate with
1664 * a wide range of boards and BIOS bugs. Fortunately only the timer IRQ
1665 * is so screwy. Thanks to Brian Perkins for testing/hacking this beast
1666 * fanatically on his truly buggy board.
1667 *
1668 * FIXME: really need to revamp this for modern platforms only.
1669 */
1670 static inline void __init check_timer(void)
1671 {
1672 struct irq_cfg *cfg = irq_cfg + 0;
1673 int apic1, pin1, apic2, pin2;
1674 unsigned long flags;
1675
1676 local_irq_save(flags);
1677
1678 /*
1679 * get/set the timer IRQ vector:
1680 */
1681 disable_8259A_irq(0);
1682 assign_irq_vector(0, TARGET_CPUS);
1683
1684 /*
1685 * Subtle, code in do_timer_interrupt() expects an AEOI
1686 * mode for the 8259A whenever interrupts are routed
1687 * through I/O APICs. Also IRQ0 has to be enabled in
1688 * the 8259A which implies the virtual wire has to be
1689 * disabled in the local APIC.
1690 */
1691 apic_write(APIC_LVT0, APIC_LVT_MASKED | APIC_DM_EXTINT);
1692 init_8259A(1);
1693 if (timer_over_8254 > 0)
1694 enable_8259A_irq(0);
1695
1696 pin1 = find_isa_irq_pin(0, mp_INT);
1697 apic1 = find_isa_irq_apic(0, mp_INT);
1698 pin2 = ioapic_i8259.pin;
1699 apic2 = ioapic_i8259.apic;
1700
1701 apic_printk(APIC_VERBOSE,KERN_INFO "..TIMER: vector=0x%02X apic1=%d pin1=%d apic2=%d pin2=%d\n",
1702 cfg->vector, apic1, pin1, apic2, pin2);
1703
1704 if (pin1 != -1) {
1705 /*
1706 * Ok, does IRQ0 through the IOAPIC work?
1707 */
1708 unmask_IO_APIC_irq(0);
1709 if (!no_timer_check && timer_irq_works()) {
1710 nmi_watchdog_default();
1711 if (nmi_watchdog == NMI_IO_APIC) {
1712 disable_8259A_irq(0);
1713 setup_nmi();
1714 enable_8259A_irq(0);
1715 }
1716 if (disable_timer_pin_1 > 0)
1717 clear_IO_APIC_pin(0, pin1);
1718 goto out;
1719 }
1720 clear_IO_APIC_pin(apic1, pin1);
1721 apic_printk(APIC_QUIET,KERN_ERR "..MP-BIOS bug: 8254 timer not "
1722 "connected to IO-APIC\n");
1723 }
1724
1725 apic_printk(APIC_VERBOSE,KERN_INFO "...trying to set up timer (IRQ0) "
1726 "through the 8259A ... ");
1727 if (pin2 != -1) {
1728 apic_printk(APIC_VERBOSE,"\n..... (found apic %d pin %d) ...",
1729 apic2, pin2);
1730 /*
1731 * legacy devices should be connected to IO APIC #0
1732 */
1733 setup_ExtINT_IRQ0_pin(apic2, pin2, cfg->vector);
1734 if (timer_irq_works()) {
1735 apic_printk(APIC_VERBOSE," works.\n");
1736 nmi_watchdog_default();
1737 if (nmi_watchdog == NMI_IO_APIC) {
1738 setup_nmi();
1739 }
1740 goto out;
1741 }
1742 /*
1743 * Cleanup, just in case ...
1744 */
1745 clear_IO_APIC_pin(apic2, pin2);
1746 }
1747 apic_printk(APIC_VERBOSE," failed.\n");
1748
1749 if (nmi_watchdog == NMI_IO_APIC) {
1750 printk(KERN_WARNING "timer doesn't work through the IO-APIC - disabling NMI Watchdog!\n");
1751 nmi_watchdog = 0;
1752 }
1753
1754 apic_printk(APIC_VERBOSE, KERN_INFO "...trying to set up timer as Virtual Wire IRQ...");
1755
1756 disable_8259A_irq(0);
1757 irq_desc[0].chip = &lapic_irq_type;
1758 apic_write(APIC_LVT0, APIC_DM_FIXED | cfg->vector); /* Fixed mode */
1759 enable_8259A_irq(0);
1760
1761 if (timer_irq_works()) {
1762 apic_printk(APIC_VERBOSE," works.\n");
1763 goto out;
1764 }
1765 apic_write(APIC_LVT0, APIC_LVT_MASKED | APIC_DM_FIXED | cfg->vector);
1766 apic_printk(APIC_VERBOSE," failed.\n");
1767
1768 apic_printk(APIC_VERBOSE, KERN_INFO "...trying to set up timer as ExtINT IRQ...");
1769
1770 init_8259A(0);
1771 make_8259A_irq(0);
1772 apic_write(APIC_LVT0, APIC_DM_EXTINT);
1773
1774 unlock_ExtINT_logic();
1775
1776 if (timer_irq_works()) {
1777 apic_printk(APIC_VERBOSE," works.\n");
1778 goto out;
1779 }
1780 apic_printk(APIC_VERBOSE," failed :(.\n");
1781 panic("IO-APIC + timer doesn't work! Try using the 'noapic' kernel parameter\n");
1782 out:
1783 local_irq_restore(flags);
1784 }
1785
1786 static int __init notimercheck(char *s)
1787 {
1788 no_timer_check = 1;
1789 return 1;
1790 }
1791 __setup("no_timer_check", notimercheck);
1792
1793 /*
1794 *
1795 * IRQs that are handled by the PIC in the MPS IOAPIC case.
1796 * - IRQ2 is the cascade IRQ, and cannot be a io-apic IRQ.
1797 * Linux doesn't really care, as it's not actually used
1798 * for any interrupt handling anyway.
1799 */
1800 #define PIC_IRQS (1<<2)
1801
1802 void __init setup_IO_APIC(void)
1803 {
1804
1805 /*
1806 * calling enable_IO_APIC() is moved to setup_local_APIC for BP
1807 */
1808
1809 if (acpi_ioapic)
1810 io_apic_irqs = ~0; /* all IRQs go through IOAPIC */
1811 else
1812 io_apic_irqs = ~PIC_IRQS;
1813
1814 apic_printk(APIC_VERBOSE, "ENABLING IO-APIC IRQs\n");
1815
1816 sync_Arb_IDs();
1817 setup_IO_APIC_irqs();
1818 init_IO_APIC_traps();
1819 check_timer();
1820 if (!acpi_ioapic)
1821 print_IO_APIC();
1822 }
1823
1824 struct sysfs_ioapic_data {
1825 struct sys_device dev;
1826 struct IO_APIC_route_entry entry[0];
1827 };
1828 static struct sysfs_ioapic_data * mp_ioapic_data[MAX_IO_APICS];
1829
1830 static int ioapic_suspend(struct sys_device *dev, pm_message_t state)
1831 {
1832 struct IO_APIC_route_entry *entry;
1833 struct sysfs_ioapic_data *data;
1834 int i;
1835
1836 data = container_of(dev, struct sysfs_ioapic_data, dev);
1837 entry = data->entry;
1838 for (i = 0; i < nr_ioapic_registers[dev->id]; i ++, entry ++ )
1839 *entry = ioapic_read_entry(dev->id, i);
1840
1841 return 0;
1842 }
1843
1844 static int ioapic_resume(struct sys_device *dev)
1845 {
1846 struct IO_APIC_route_entry *entry;
1847 struct sysfs_ioapic_data *data;
1848 unsigned long flags;
1849 union IO_APIC_reg_00 reg_00;
1850 int i;
1851
1852 data = container_of(dev, struct sysfs_ioapic_data, dev);
1853 entry = data->entry;
1854
1855 spin_lock_irqsave(&ioapic_lock, flags);
1856 reg_00.raw = io_apic_read(dev->id, 0);
1857 if (reg_00.bits.ID != mp_ioapics[dev->id].mpc_apicid) {
1858 reg_00.bits.ID = mp_ioapics[dev->id].mpc_apicid;
1859 io_apic_write(dev->id, 0, reg_00.raw);
1860 }
1861 spin_unlock_irqrestore(&ioapic_lock, flags);
1862 for (i = 0; i < nr_ioapic_registers[dev->id]; i++)
1863 ioapic_write_entry(dev->id, i, entry[i]);
1864
1865 return 0;
1866 }
1867
1868 static struct sysdev_class ioapic_sysdev_class = {
1869 .name = "ioapic",
1870 .suspend = ioapic_suspend,
1871 .resume = ioapic_resume,
1872 };
1873
1874 static int __init ioapic_init_sysfs(void)
1875 {
1876 struct sys_device * dev;
1877 int i, size, error;
1878
1879 error = sysdev_class_register(&ioapic_sysdev_class);
1880 if (error)
1881 return error;
1882
1883 for (i = 0; i < nr_ioapics; i++ ) {
1884 size = sizeof(struct sys_device) + nr_ioapic_registers[i]
1885 * sizeof(struct IO_APIC_route_entry);
1886 mp_ioapic_data[i] = kzalloc(size, GFP_KERNEL);
1887 if (!mp_ioapic_data[i]) {
1888 printk(KERN_ERR "Can't suspend/resume IOAPIC %d\n", i);
1889 continue;
1890 }
1891 dev = &mp_ioapic_data[i]->dev;
1892 dev->id = i;
1893 dev->cls = &ioapic_sysdev_class;
1894 error = sysdev_register(dev);
1895 if (error) {
1896 kfree(mp_ioapic_data[i]);
1897 mp_ioapic_data[i] = NULL;
1898 printk(KERN_ERR "Can't suspend/resume IOAPIC %d\n", i);
1899 continue;
1900 }
1901 }
1902
1903 return 0;
1904 }
1905
1906 device_initcall(ioapic_init_sysfs);
1907
1908 /*
1909 * Dynamic irq allocate and deallocation
1910 */
1911 int create_irq(void)
1912 {
1913 /* Allocate an unused irq */
1914 int irq;
1915 int new;
1916 unsigned long flags;
1917
1918 irq = -ENOSPC;
1919 spin_lock_irqsave(&vector_lock, flags);
1920 for (new = (NR_IRQS - 1); new >= 0; new--) {
1921 if (platform_legacy_irq(new))
1922 continue;
1923 if (irq_cfg[new].vector != 0)
1924 continue;
1925 if (__assign_irq_vector(new, TARGET_CPUS) == 0)
1926 irq = new;
1927 break;
1928 }
1929 spin_unlock_irqrestore(&vector_lock, flags);
1930
1931 if (irq >= 0) {
1932 dynamic_irq_init(irq);
1933 }
1934 return irq;
1935 }
1936
1937 void destroy_irq(unsigned int irq)
1938 {
1939 unsigned long flags;
1940
1941 dynamic_irq_cleanup(irq);
1942
1943 spin_lock_irqsave(&vector_lock, flags);
1944 __clear_irq_vector(irq);
1945 spin_unlock_irqrestore(&vector_lock, flags);
1946 }
1947
1948 /*
1949 * MSI message composition
1950 */
1951 #ifdef CONFIG_PCI_MSI
1952 static int msi_compose_msg(struct pci_dev *pdev, unsigned int irq, struct msi_msg *msg)
1953 {
1954 struct irq_cfg *cfg = irq_cfg + irq;
1955 int err;
1956 unsigned dest;
1957 cpumask_t tmp;
1958
1959 tmp = TARGET_CPUS;
1960 err = assign_irq_vector(irq, tmp);
1961 if (!err) {
1962 cpus_and(tmp, cfg->domain, tmp);
1963 dest = cpu_mask_to_apicid(tmp);
1964
1965 msg->address_hi = MSI_ADDR_BASE_HI;
1966 msg->address_lo =
1967 MSI_ADDR_BASE_LO |
1968 ((INT_DEST_MODE == 0) ?
1969 MSI_ADDR_DEST_MODE_PHYSICAL:
1970 MSI_ADDR_DEST_MODE_LOGICAL) |
1971 ((INT_DELIVERY_MODE != dest_LowestPrio) ?
1972 MSI_ADDR_REDIRECTION_CPU:
1973 MSI_ADDR_REDIRECTION_LOWPRI) |
1974 MSI_ADDR_DEST_ID(dest);
1975
1976 msg->data =
1977 MSI_DATA_TRIGGER_EDGE |
1978 MSI_DATA_LEVEL_ASSERT |
1979 ((INT_DELIVERY_MODE != dest_LowestPrio) ?
1980 MSI_DATA_DELIVERY_FIXED:
1981 MSI_DATA_DELIVERY_LOWPRI) |
1982 MSI_DATA_VECTOR(cfg->vector);
1983 }
1984 return err;
1985 }
1986
1987 #ifdef CONFIG_SMP
1988 static void set_msi_irq_affinity(unsigned int irq, cpumask_t mask)
1989 {
1990 struct irq_cfg *cfg = irq_cfg + irq;
1991 struct msi_msg msg;
1992 unsigned int dest;
1993 cpumask_t tmp;
1994
1995 cpus_and(tmp, mask, cpu_online_map);
1996 if (cpus_empty(tmp))
1997 return;
1998
1999 if (assign_irq_vector(irq, mask))
2000 return;
2001
2002 cpus_and(tmp, cfg->domain, mask);
2003 dest = cpu_mask_to_apicid(tmp);
2004
2005 read_msi_msg(irq, &msg);
2006
2007 msg.data &= ~MSI_DATA_VECTOR_MASK;
2008 msg.data |= MSI_DATA_VECTOR(cfg->vector);
2009 msg.address_lo &= ~MSI_ADDR_DEST_ID_MASK;
2010 msg.address_lo |= MSI_ADDR_DEST_ID(dest);
2011
2012 write_msi_msg(irq, &msg);
2013 irq_desc[irq].affinity = mask;
2014 }
2015 #endif /* CONFIG_SMP */
2016
2017 /*
2018 * IRQ Chip for MSI PCI/PCI-X/PCI-Express Devices,
2019 * which implement the MSI or MSI-X Capability Structure.
2020 */
2021 static struct irq_chip msi_chip = {
2022 .name = "PCI-MSI",
2023 .unmask = unmask_msi_irq,
2024 .mask = mask_msi_irq,
2025 .ack = ack_apic_edge,
2026 #ifdef CONFIG_SMP
2027 .set_affinity = set_msi_irq_affinity,
2028 #endif
2029 .retrigger = ioapic_retrigger_irq,
2030 };
2031
2032 int arch_setup_msi_irq(struct pci_dev *dev, struct msi_desc *desc)
2033 {
2034 struct msi_msg msg;
2035 int irq, ret;
2036 irq = create_irq();
2037 if (irq < 0)
2038 return irq;
2039
2040 ret = msi_compose_msg(dev, irq, &msg);
2041 if (ret < 0) {
2042 destroy_irq(irq);
2043 return ret;
2044 }
2045
2046 set_irq_msi(irq, desc);
2047 write_msi_msg(irq, &msg);
2048
2049 set_irq_chip_and_handler_name(irq, &msi_chip, handle_edge_irq, "edge");
2050
2051 return 0;
2052 }
2053
2054 void arch_teardown_msi_irq(unsigned int irq)
2055 {
2056 destroy_irq(irq);
2057 }
2058
2059 #ifdef CONFIG_DMAR
2060 #ifdef CONFIG_SMP
2061 static void dmar_msi_set_affinity(unsigned int irq, cpumask_t mask)
2062 {
2063 struct irq_cfg *cfg = irq_cfg + irq;
2064 struct msi_msg msg;
2065 unsigned int dest;
2066 cpumask_t tmp;
2067
2068 cpus_and(tmp, mask, cpu_online_map);
2069 if (cpus_empty(tmp))
2070 return;
2071
2072 if (assign_irq_vector(irq, mask))
2073 return;
2074
2075 cpus_and(tmp, cfg->domain, mask);
2076 dest = cpu_mask_to_apicid(tmp);
2077
2078 dmar_msi_read(irq, &msg);
2079
2080 msg.data &= ~MSI_DATA_VECTOR_MASK;
2081 msg.data |= MSI_DATA_VECTOR(cfg->vector);
2082 msg.address_lo &= ~MSI_ADDR_DEST_ID_MASK;
2083 msg.address_lo |= MSI_ADDR_DEST_ID(dest);
2084
2085 dmar_msi_write(irq, &msg);
2086 irq_desc[irq].affinity = mask;
2087 }
2088 #endif /* CONFIG_SMP */
2089
2090 struct irq_chip dmar_msi_type = {
2091 .name = "DMAR_MSI",
2092 .unmask = dmar_msi_unmask,
2093 .mask = dmar_msi_mask,
2094 .ack = ack_apic_edge,
2095 #ifdef CONFIG_SMP
2096 .set_affinity = dmar_msi_set_affinity,
2097 #endif
2098 .retrigger = ioapic_retrigger_irq,
2099 };
2100
2101 int arch_setup_dmar_msi(unsigned int irq)
2102 {
2103 int ret;
2104 struct msi_msg msg;
2105
2106 ret = msi_compose_msg(NULL, irq, &msg);
2107 if (ret < 0)
2108 return ret;
2109 dmar_msi_write(irq, &msg);
2110 set_irq_chip_and_handler_name(irq, &dmar_msi_type, handle_edge_irq,
2111 "edge");
2112 return 0;
2113 }
2114 #endif
2115
2116 #endif /* CONFIG_PCI_MSI */
2117 /*
2118 * Hypertransport interrupt support
2119 */
2120 #ifdef CONFIG_HT_IRQ
2121
2122 #ifdef CONFIG_SMP
2123
2124 static void target_ht_irq(unsigned int irq, unsigned int dest, u8 vector)
2125 {
2126 struct ht_irq_msg msg;
2127 fetch_ht_irq_msg(irq, &msg);
2128
2129 msg.address_lo &= ~(HT_IRQ_LOW_VECTOR_MASK | HT_IRQ_LOW_DEST_ID_MASK);
2130 msg.address_hi &= ~(HT_IRQ_HIGH_DEST_ID_MASK);
2131
2132 msg.address_lo |= HT_IRQ_LOW_VECTOR(vector) | HT_IRQ_LOW_DEST_ID(dest);
2133 msg.address_hi |= HT_IRQ_HIGH_DEST_ID(dest);
2134
2135 write_ht_irq_msg(irq, &msg);
2136 }
2137
2138 static void set_ht_irq_affinity(unsigned int irq, cpumask_t mask)
2139 {
2140 struct irq_cfg *cfg = irq_cfg + irq;
2141 unsigned int dest;
2142 cpumask_t tmp;
2143
2144 cpus_and(tmp, mask, cpu_online_map);
2145 if (cpus_empty(tmp))
2146 return;
2147
2148 if (assign_irq_vector(irq, mask))
2149 return;
2150
2151 cpus_and(tmp, cfg->domain, mask);
2152 dest = cpu_mask_to_apicid(tmp);
2153
2154 target_ht_irq(irq, dest, cfg->vector);
2155 irq_desc[irq].affinity = mask;
2156 }
2157 #endif
2158
2159 static struct irq_chip ht_irq_chip = {
2160 .name = "PCI-HT",
2161 .mask = mask_ht_irq,
2162 .unmask = unmask_ht_irq,
2163 .ack = ack_apic_edge,
2164 #ifdef CONFIG_SMP
2165 .set_affinity = set_ht_irq_affinity,
2166 #endif
2167 .retrigger = ioapic_retrigger_irq,
2168 };
2169
2170 int arch_setup_ht_irq(unsigned int irq, struct pci_dev *dev)
2171 {
2172 struct irq_cfg *cfg = irq_cfg + irq;
2173 int err;
2174 cpumask_t tmp;
2175
2176 tmp = TARGET_CPUS;
2177 err = assign_irq_vector(irq, tmp);
2178 if (!err) {
2179 struct ht_irq_msg msg;
2180 unsigned dest;
2181
2182 cpus_and(tmp, cfg->domain, tmp);
2183 dest = cpu_mask_to_apicid(tmp);
2184
2185 msg.address_hi = HT_IRQ_HIGH_DEST_ID(dest);
2186
2187 msg.address_lo =
2188 HT_IRQ_LOW_BASE |
2189 HT_IRQ_LOW_DEST_ID(dest) |
2190 HT_IRQ_LOW_VECTOR(cfg->vector) |
2191 ((INT_DEST_MODE == 0) ?
2192 HT_IRQ_LOW_DM_PHYSICAL :
2193 HT_IRQ_LOW_DM_LOGICAL) |
2194 HT_IRQ_LOW_RQEOI_EDGE |
2195 ((INT_DELIVERY_MODE != dest_LowestPrio) ?
2196 HT_IRQ_LOW_MT_FIXED :
2197 HT_IRQ_LOW_MT_ARBITRATED) |
2198 HT_IRQ_LOW_IRQ_MASKED;
2199
2200 write_ht_irq_msg(irq, &msg);
2201
2202 set_irq_chip_and_handler_name(irq, &ht_irq_chip,
2203 handle_edge_irq, "edge");
2204 }
2205 return err;
2206 }
2207 #endif /* CONFIG_HT_IRQ */
2208
2209 /* --------------------------------------------------------------------------
2210 ACPI-based IOAPIC Configuration
2211 -------------------------------------------------------------------------- */
2212
2213 #ifdef CONFIG_ACPI
2214
2215 #define IO_APIC_MAX_ID 0xFE
2216
2217 int __init io_apic_get_redir_entries (int ioapic)
2218 {
2219 union IO_APIC_reg_01 reg_01;
2220 unsigned long flags;
2221
2222 spin_lock_irqsave(&ioapic_lock, flags);
2223 reg_01.raw = io_apic_read(ioapic, 1);
2224 spin_unlock_irqrestore(&ioapic_lock, flags);
2225
2226 return reg_01.bits.entries;
2227 }
2228
2229
2230 int io_apic_set_pci_routing (int ioapic, int pin, int irq, int triggering, int polarity)
2231 {
2232 if (!IO_APIC_IRQ(irq)) {
2233 apic_printk(APIC_QUIET,KERN_ERR "IOAPIC[%d]: Invalid reference to IRQ 0\n",
2234 ioapic);
2235 return -EINVAL;
2236 }
2237
2238 /*
2239 * IRQs < 16 are already in the irq_2_pin[] map
2240 */
2241 if (irq >= 16)
2242 add_pin_to_irq(irq, ioapic, pin);
2243
2244 setup_IO_APIC_irq(ioapic, pin, irq, triggering, polarity);
2245
2246 return 0;
2247 }
2248
2249
2250 int acpi_get_override_irq(int bus_irq, int *trigger, int *polarity)
2251 {
2252 int i;
2253
2254 if (skip_ioapic_setup)
2255 return -1;
2256
2257 for (i = 0; i < mp_irq_entries; i++)
2258 if (mp_irqs[i].mpc_irqtype == mp_INT &&
2259 mp_irqs[i].mpc_srcbusirq == bus_irq)
2260 break;
2261 if (i >= mp_irq_entries)
2262 return -1;
2263
2264 *trigger = irq_trigger(i);
2265 *polarity = irq_polarity(i);
2266 return 0;
2267 }
2268
2269 #endif /* CONFIG_ACPI */
2270
2271 /*
2272 * This function currently is only a helper for the i386 smp boot process where
2273 * we need to reprogram the ioredtbls to cater for the cpus which have come online
2274 * so mask in all cases should simply be TARGET_CPUS
2275 */
2276 #ifdef CONFIG_SMP
2277 void __init setup_ioapic_dest(void)
2278 {
2279 int pin, ioapic, irq, irq_entry;
2280
2281 if (skip_ioapic_setup == 1)
2282 return;
2283
2284 for (ioapic = 0; ioapic < nr_ioapics; ioapic++) {
2285 for (pin = 0; pin < nr_ioapic_registers[ioapic]; pin++) {
2286 irq_entry = find_irq_entry(ioapic, pin, mp_INT);
2287 if (irq_entry == -1)
2288 continue;
2289 irq = pin_2_irq(irq_entry, ioapic, pin);
2290
2291 /* setup_IO_APIC_irqs could fail to get vector for some device
2292 * when you have too many devices, because at that time only boot
2293 * cpu is online.
2294 */
2295 if (!irq_cfg[irq].vector)
2296 setup_IO_APIC_irq(ioapic, pin, irq,
2297 irq_trigger(irq_entry),
2298 irq_polarity(irq_entry));
2299 else
2300 set_ioapic_affinity_irq(irq, TARGET_CPUS);
2301 }
2302
2303 }
2304 }
2305 #endif
2306
2307 #define IOAPIC_RESOURCE_NAME_SIZE 11
2308
2309 static struct resource *ioapic_resources;
2310
2311 static struct resource * __init ioapic_setup_resources(void)
2312 {
2313 unsigned long n;
2314 struct resource *res;
2315 char *mem;
2316 int i;
2317
2318 if (nr_ioapics <= 0)
2319 return NULL;
2320
2321 n = IOAPIC_RESOURCE_NAME_SIZE + sizeof(struct resource);
2322 n *= nr_ioapics;
2323
2324 mem = alloc_bootmem(n);
2325 res = (void *)mem;
2326
2327 if (mem != NULL) {
2328 memset(mem, 0, n);
2329 mem += sizeof(struct resource) * nr_ioapics;
2330
2331 for (i = 0; i < nr_ioapics; i++) {
2332 res[i].name = mem;
2333 res[i].flags = IORESOURCE_MEM | IORESOURCE_BUSY;
2334 sprintf(mem, "IOAPIC %u", i);
2335 mem += IOAPIC_RESOURCE_NAME_SIZE;
2336 }
2337 }
2338
2339 ioapic_resources = res;
2340
2341 return res;
2342 }
2343
2344 void __init ioapic_init_mappings(void)
2345 {
2346 unsigned long ioapic_phys, idx = FIX_IO_APIC_BASE_0;
2347 struct resource *ioapic_res;
2348 int i;
2349
2350 ioapic_res = ioapic_setup_resources();
2351 for (i = 0; i < nr_ioapics; i++) {
2352 if (smp_found_config) {
2353 ioapic_phys = mp_ioapics[i].mpc_apicaddr;
2354 } else {
2355 ioapic_phys = (unsigned long)
2356 alloc_bootmem_pages(PAGE_SIZE);
2357 ioapic_phys = __pa(ioapic_phys);
2358 }
2359 set_fixmap_nocache(idx, ioapic_phys);
2360 apic_printk(APIC_VERBOSE,
2361 "mapped IOAPIC to %016lx (%016lx)\n",
2362 __fix_to_virt(idx), ioapic_phys);
2363 idx++;
2364
2365 if (ioapic_res != NULL) {
2366 ioapic_res->start = ioapic_phys;
2367 ioapic_res->end = ioapic_phys + (4 * 1024) - 1;
2368 ioapic_res++;
2369 }
2370 }
2371 }
2372
2373 static int __init ioapic_insert_resources(void)
2374 {
2375 int i;
2376 struct resource *r = ioapic_resources;
2377
2378 if (!r) {
2379 printk(KERN_ERR
2380 "IO APIC resources could be not be allocated.\n");
2381 return -1;
2382 }
2383
2384 for (i = 0; i < nr_ioapics; i++) {
2385 insert_resource(&iomem_resource, r);
2386 r++;
2387 }
2388
2389 return 0;
2390 }
2391
2392 /* Insert the IO APIC resources after PCI initialization has occured to handle
2393 * IO APICS that are mapped in on a BAR in PCI space. */
2394 late_initcall(ioapic_insert_resources);
2395
Linux kernel - Deliverables
This page took 0.09386 seconds and 5 git commands to generate.