mfd: wm8350-i2c: Make sure the i2c regmap functions are compiled
[linux/fpc-iii.git] / arch / x86 / kernel / apic / io_apic.c
blobe63a5bd2a78f55de2c356b665751558a73bc18da
1 /*
2 * Intel IO-APIC support for multi-Pentium hosts.
4 * Copyright (C) 1997, 1998, 1999, 2000, 2009 Ingo Molnar, Hajnalka Szabo
6 * Many thanks to Stig Venaas for trying out countless experimental
7 * patches and reporting/debugging problems patiently!
9 * (c) 1999, Multiple IO-APIC support, developed by
10 * Ken-ichi Yaku <yaku@css1.kbnes.nec.co.jp> and
11 * Hidemi Kishimoto <kisimoto@css1.kbnes.nec.co.jp>,
12 * further tested and cleaned up by Zach Brown <zab@redhat.com>
13 * and Ingo Molnar <mingo@redhat.com>
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
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/compiler.h>
31 #include <linux/acpi.h>
32 #include <linux/module.h>
33 #include <linux/syscore_ops.h>
34 #include <linux/msi.h>
35 #include <linux/htirq.h>
36 #include <linux/freezer.h>
37 #include <linux/kthread.h>
38 #include <linux/jiffies.h> /* time_after() */
39 #include <linux/slab.h>
40 #ifdef CONFIG_ACPI
41 #include <acpi/acpi_bus.h>
42 #endif
43 #include <linux/bootmem.h>
44 #include <linux/dmar.h>
45 #include <linux/hpet.h>
47 #include <asm/idle.h>
48 #include <asm/io.h>
49 #include <asm/smp.h>
50 #include <asm/cpu.h>
51 #include <asm/desc.h>
52 #include <asm/proto.h>
53 #include <asm/acpi.h>
54 #include <asm/dma.h>
55 #include <asm/timer.h>
56 #include <asm/i8259.h>
57 #include <asm/msidef.h>
58 #include <asm/hypertransport.h>
59 #include <asm/setup.h>
60 #include <asm/irq_remapping.h>
61 #include <asm/hpet.h>
62 #include <asm/hw_irq.h>
64 #include <asm/apic.h>
66 #define __apicdebuginit(type) static type __init
68 #define for_each_irq_pin(entry, head) \
69 for (entry = head; entry; entry = entry->next)
72 * Is the SiS APIC rmw bug present ?
73 * -1 = don't know, 0 = no, 1 = yes
75 int sis_apic_bug = -1;
77 static DEFINE_RAW_SPINLOCK(ioapic_lock);
78 static DEFINE_RAW_SPINLOCK(vector_lock);
80 static struct ioapic {
82 * # of IRQ routing registers
84 int nr_registers;
86 * Saved state during suspend/resume, or while enabling intr-remap.
88 struct IO_APIC_route_entry *saved_registers;
89 /* I/O APIC config */
90 struct mpc_ioapic mp_config;
91 /* IO APIC gsi routing info */
92 struct mp_ioapic_gsi gsi_config;
93 DECLARE_BITMAP(pin_programmed, MP_MAX_IOAPIC_PIN + 1);
94 } ioapics[MAX_IO_APICS];
96 #define mpc_ioapic_ver(ioapic_idx) ioapics[ioapic_idx].mp_config.apicver
98 int mpc_ioapic_id(int ioapic_idx)
100 return ioapics[ioapic_idx].mp_config.apicid;
103 unsigned int mpc_ioapic_addr(int ioapic_idx)
105 return ioapics[ioapic_idx].mp_config.apicaddr;
108 struct mp_ioapic_gsi *mp_ioapic_gsi_routing(int ioapic_idx)
110 return &ioapics[ioapic_idx].gsi_config;
113 int nr_ioapics;
115 /* The one past the highest gsi number used */
116 u32 gsi_top;
118 /* MP IRQ source entries */
119 struct mpc_intsrc mp_irqs[MAX_IRQ_SOURCES];
121 /* # of MP IRQ source entries */
122 int mp_irq_entries;
124 /* GSI interrupts */
125 static int nr_irqs_gsi = NR_IRQS_LEGACY;
127 #ifdef CONFIG_EISA
128 int mp_bus_id_to_type[MAX_MP_BUSSES];
129 #endif
131 DECLARE_BITMAP(mp_bus_not_pci, MAX_MP_BUSSES);
133 int skip_ioapic_setup;
136 * disable_ioapic_support() - disables ioapic support at runtime
138 void disable_ioapic_support(void)
140 #ifdef CONFIG_PCI
141 noioapicquirk = 1;
142 noioapicreroute = -1;
143 #endif
144 skip_ioapic_setup = 1;
147 static int __init parse_noapic(char *str)
149 /* disable IO-APIC */
150 disable_ioapic_support();
151 return 0;
153 early_param("noapic", parse_noapic);
155 static int io_apic_setup_irq_pin(unsigned int irq, int node,
156 struct io_apic_irq_attr *attr);
158 /* Will be called in mpparse/acpi/sfi codes for saving IRQ info */
159 void mp_save_irq(struct mpc_intsrc *m)
161 int i;
163 apic_printk(APIC_VERBOSE, "Int: type %d, pol %d, trig %d, bus %02x,"
164 " IRQ %02x, APIC ID %x, APIC INT %02x\n",
165 m->irqtype, m->irqflag & 3, (m->irqflag >> 2) & 3, m->srcbus,
166 m->srcbusirq, m->dstapic, m->dstirq);
168 for (i = 0; i < mp_irq_entries; i++) {
169 if (!memcmp(&mp_irqs[i], m, sizeof(*m)))
170 return;
173 memcpy(&mp_irqs[mp_irq_entries], m, sizeof(*m));
174 if (++mp_irq_entries == MAX_IRQ_SOURCES)
175 panic("Max # of irq sources exceeded!!\n");
178 struct irq_pin_list {
179 int apic, pin;
180 struct irq_pin_list *next;
183 static struct irq_pin_list *alloc_irq_pin_list(int node)
185 return kzalloc_node(sizeof(struct irq_pin_list), GFP_KERNEL, node);
189 /* irq_cfg is indexed by the sum of all RTEs in all I/O APICs. */
190 static struct irq_cfg irq_cfgx[NR_IRQS_LEGACY];
192 int __init arch_early_irq_init(void)
194 struct irq_cfg *cfg;
195 int count, node, i;
197 if (!legacy_pic->nr_legacy_irqs)
198 io_apic_irqs = ~0UL;
200 for (i = 0; i < nr_ioapics; i++) {
201 ioapics[i].saved_registers =
202 kzalloc(sizeof(struct IO_APIC_route_entry) *
203 ioapics[i].nr_registers, GFP_KERNEL);
204 if (!ioapics[i].saved_registers)
205 pr_err("IOAPIC %d: suspend/resume impossible!\n", i);
208 cfg = irq_cfgx;
209 count = ARRAY_SIZE(irq_cfgx);
210 node = cpu_to_node(0);
212 /* Make sure the legacy interrupts are marked in the bitmap */
213 irq_reserve_irqs(0, legacy_pic->nr_legacy_irqs);
215 for (i = 0; i < count; i++) {
216 irq_set_chip_data(i, &cfg[i]);
217 zalloc_cpumask_var_node(&cfg[i].domain, GFP_KERNEL, node);
218 zalloc_cpumask_var_node(&cfg[i].old_domain, GFP_KERNEL, node);
220 * For legacy IRQ's, start with assigning irq0 to irq15 to
221 * IRQ0_VECTOR to IRQ15_VECTOR for all cpu's.
223 if (i < legacy_pic->nr_legacy_irqs) {
224 cfg[i].vector = IRQ0_VECTOR + i;
225 cpumask_setall(cfg[i].domain);
229 return 0;
232 static struct irq_cfg *irq_cfg(unsigned int irq)
234 return irq_get_chip_data(irq);
237 static struct irq_cfg *alloc_irq_cfg(unsigned int irq, int node)
239 struct irq_cfg *cfg;
241 cfg = kzalloc_node(sizeof(*cfg), GFP_KERNEL, node);
242 if (!cfg)
243 return NULL;
244 if (!zalloc_cpumask_var_node(&cfg->domain, GFP_KERNEL, node))
245 goto out_cfg;
246 if (!zalloc_cpumask_var_node(&cfg->old_domain, GFP_KERNEL, node))
247 goto out_domain;
248 return cfg;
249 out_domain:
250 free_cpumask_var(cfg->domain);
251 out_cfg:
252 kfree(cfg);
253 return NULL;
256 static void free_irq_cfg(unsigned int at, struct irq_cfg *cfg)
258 if (!cfg)
259 return;
260 irq_set_chip_data(at, NULL);
261 free_cpumask_var(cfg->domain);
262 free_cpumask_var(cfg->old_domain);
263 kfree(cfg);
266 static struct irq_cfg *alloc_irq_and_cfg_at(unsigned int at, int node)
268 int res = irq_alloc_desc_at(at, node);
269 struct irq_cfg *cfg;
271 if (res < 0) {
272 if (res != -EEXIST)
273 return NULL;
274 cfg = irq_get_chip_data(at);
275 if (cfg)
276 return cfg;
279 cfg = alloc_irq_cfg(at, node);
280 if (cfg)
281 irq_set_chip_data(at, cfg);
282 else
283 irq_free_desc(at);
284 return cfg;
287 static int alloc_irqs_from(unsigned int from, unsigned int count, int node)
289 return irq_alloc_descs_from(from, count, node);
292 static void free_irq_at(unsigned int at, struct irq_cfg *cfg)
294 free_irq_cfg(at, cfg);
295 irq_free_desc(at);
299 struct io_apic {
300 unsigned int index;
301 unsigned int unused[3];
302 unsigned int data;
303 unsigned int unused2[11];
304 unsigned int eoi;
307 static __attribute_const__ struct io_apic __iomem *io_apic_base(int idx)
309 return (void __iomem *) __fix_to_virt(FIX_IO_APIC_BASE_0 + idx)
310 + (mpc_ioapic_addr(idx) & ~PAGE_MASK);
313 void io_apic_eoi(unsigned int apic, unsigned int vector)
315 struct io_apic __iomem *io_apic = io_apic_base(apic);
316 writel(vector, &io_apic->eoi);
319 unsigned int native_io_apic_read(unsigned int apic, unsigned int reg)
321 struct io_apic __iomem *io_apic = io_apic_base(apic);
322 writel(reg, &io_apic->index);
323 return readl(&io_apic->data);
326 void native_io_apic_write(unsigned int apic, unsigned int reg, unsigned int value)
328 struct io_apic __iomem *io_apic = io_apic_base(apic);
330 writel(reg, &io_apic->index);
331 writel(value, &io_apic->data);
335 * Re-write a value: to be used for read-modify-write
336 * cycles where the read already set up the index register.
338 * Older SiS APIC requires we rewrite the index register
340 void native_io_apic_modify(unsigned int apic, unsigned int reg, unsigned int value)
342 struct io_apic __iomem *io_apic = io_apic_base(apic);
344 if (sis_apic_bug)
345 writel(reg, &io_apic->index);
346 writel(value, &io_apic->data);
349 union entry_union {
350 struct { u32 w1, w2; };
351 struct IO_APIC_route_entry entry;
354 static struct IO_APIC_route_entry __ioapic_read_entry(int apic, int pin)
356 union entry_union eu;
358 eu.w1 = io_apic_read(apic, 0x10 + 2 * pin);
359 eu.w2 = io_apic_read(apic, 0x11 + 2 * pin);
361 return eu.entry;
364 static struct IO_APIC_route_entry ioapic_read_entry(int apic, int pin)
366 union entry_union eu;
367 unsigned long flags;
369 raw_spin_lock_irqsave(&ioapic_lock, flags);
370 eu.entry = __ioapic_read_entry(apic, pin);
371 raw_spin_unlock_irqrestore(&ioapic_lock, flags);
373 return eu.entry;
377 * When we write a new IO APIC routing entry, we need to write the high
378 * word first! If the mask bit in the low word is clear, we will enable
379 * the interrupt, and we need to make sure the entry is fully populated
380 * before that happens.
382 static void __ioapic_write_entry(int apic, int pin, struct IO_APIC_route_entry e)
384 union entry_union eu = {{0, 0}};
386 eu.entry = e;
387 io_apic_write(apic, 0x11 + 2*pin, eu.w2);
388 io_apic_write(apic, 0x10 + 2*pin, eu.w1);
391 static void ioapic_write_entry(int apic, int pin, struct IO_APIC_route_entry e)
393 unsigned long flags;
395 raw_spin_lock_irqsave(&ioapic_lock, flags);
396 __ioapic_write_entry(apic, pin, e);
397 raw_spin_unlock_irqrestore(&ioapic_lock, flags);
401 * When we mask an IO APIC routing entry, we need to write the low
402 * word first, in order to set the mask bit before we change the
403 * high bits!
405 static void ioapic_mask_entry(int apic, int pin)
407 unsigned long flags;
408 union entry_union eu = { .entry.mask = 1 };
410 raw_spin_lock_irqsave(&ioapic_lock, flags);
411 io_apic_write(apic, 0x10 + 2*pin, eu.w1);
412 io_apic_write(apic, 0x11 + 2*pin, eu.w2);
413 raw_spin_unlock_irqrestore(&ioapic_lock, flags);
417 * The common case is 1:1 IRQ<->pin mappings. Sometimes there are
418 * shared ISA-space IRQs, so we have to support them. We are super
419 * fast in the common case, and fast for shared ISA-space IRQs.
421 static int __add_pin_to_irq_node(struct irq_cfg *cfg, int node, int apic, int pin)
423 struct irq_pin_list **last, *entry;
425 /* don't allow duplicates */
426 last = &cfg->irq_2_pin;
427 for_each_irq_pin(entry, cfg->irq_2_pin) {
428 if (entry->apic == apic && entry->pin == pin)
429 return 0;
430 last = &entry->next;
433 entry = alloc_irq_pin_list(node);
434 if (!entry) {
435 pr_err("can not alloc irq_pin_list (%d,%d,%d)\n",
436 node, apic, pin);
437 return -ENOMEM;
439 entry->apic = apic;
440 entry->pin = pin;
442 *last = entry;
443 return 0;
446 static void add_pin_to_irq_node(struct irq_cfg *cfg, int node, int apic, int pin)
448 if (__add_pin_to_irq_node(cfg, node, apic, pin))
449 panic("IO-APIC: failed to add irq-pin. Can not proceed\n");
453 * Reroute an IRQ to a different pin.
455 static void __init replace_pin_at_irq_node(struct irq_cfg *cfg, int node,
456 int oldapic, int oldpin,
457 int newapic, int newpin)
459 struct irq_pin_list *entry;
461 for_each_irq_pin(entry, cfg->irq_2_pin) {
462 if (entry->apic == oldapic && entry->pin == oldpin) {
463 entry->apic = newapic;
464 entry->pin = newpin;
465 /* every one is different, right? */
466 return;
470 /* old apic/pin didn't exist, so just add new ones */
471 add_pin_to_irq_node(cfg, node, newapic, newpin);
474 static void __io_apic_modify_irq(struct irq_pin_list *entry,
475 int mask_and, int mask_or,
476 void (*final)(struct irq_pin_list *entry))
478 unsigned int reg, pin;
480 pin = entry->pin;
481 reg = io_apic_read(entry->apic, 0x10 + pin * 2);
482 reg &= mask_and;
483 reg |= mask_or;
484 io_apic_modify(entry->apic, 0x10 + pin * 2, reg);
485 if (final)
486 final(entry);
489 static void io_apic_modify_irq(struct irq_cfg *cfg,
490 int mask_and, int mask_or,
491 void (*final)(struct irq_pin_list *entry))
493 struct irq_pin_list *entry;
495 for_each_irq_pin(entry, cfg->irq_2_pin)
496 __io_apic_modify_irq(entry, mask_and, mask_or, final);
499 static void io_apic_sync(struct irq_pin_list *entry)
502 * Synchronize the IO-APIC and the CPU by doing
503 * a dummy read from the IO-APIC
505 struct io_apic __iomem *io_apic;
507 io_apic = io_apic_base(entry->apic);
508 readl(&io_apic->data);
511 static void mask_ioapic(struct irq_cfg *cfg)
513 unsigned long flags;
515 raw_spin_lock_irqsave(&ioapic_lock, flags);
516 io_apic_modify_irq(cfg, ~0, IO_APIC_REDIR_MASKED, &io_apic_sync);
517 raw_spin_unlock_irqrestore(&ioapic_lock, flags);
520 static void mask_ioapic_irq(struct irq_data *data)
522 mask_ioapic(data->chip_data);
525 static void __unmask_ioapic(struct irq_cfg *cfg)
527 io_apic_modify_irq(cfg, ~IO_APIC_REDIR_MASKED, 0, NULL);
530 static void unmask_ioapic(struct irq_cfg *cfg)
532 unsigned long flags;
534 raw_spin_lock_irqsave(&ioapic_lock, flags);
535 __unmask_ioapic(cfg);
536 raw_spin_unlock_irqrestore(&ioapic_lock, flags);
539 static void unmask_ioapic_irq(struct irq_data *data)
541 unmask_ioapic(data->chip_data);
545 * IO-APIC versions below 0x20 don't support EOI register.
546 * For the record, here is the information about various versions:
547 * 0Xh 82489DX
548 * 1Xh I/OAPIC or I/O(x)APIC which are not PCI 2.2 Compliant
549 * 2Xh I/O(x)APIC which is PCI 2.2 Compliant
550 * 30h-FFh Reserved
552 * Some of the Intel ICH Specs (ICH2 to ICH5) documents the io-apic
553 * version as 0x2. This is an error with documentation and these ICH chips
554 * use io-apic's of version 0x20.
556 * For IO-APIC's with EOI register, we use that to do an explicit EOI.
557 * Otherwise, we simulate the EOI message manually by changing the trigger
558 * mode to edge and then back to level, with RTE being masked during this.
560 void native_eoi_ioapic_pin(int apic, int pin, int vector)
562 if (mpc_ioapic_ver(apic) >= 0x20) {
563 io_apic_eoi(apic, vector);
564 } else {
565 struct IO_APIC_route_entry entry, entry1;
567 entry = entry1 = __ioapic_read_entry(apic, pin);
570 * Mask the entry and change the trigger mode to edge.
572 entry1.mask = 1;
573 entry1.trigger = IOAPIC_EDGE;
575 __ioapic_write_entry(apic, pin, entry1);
578 * Restore the previous level triggered entry.
580 __ioapic_write_entry(apic, pin, entry);
584 void eoi_ioapic_irq(unsigned int irq, struct irq_cfg *cfg)
586 struct irq_pin_list *entry;
587 unsigned long flags;
589 raw_spin_lock_irqsave(&ioapic_lock, flags);
590 for_each_irq_pin(entry, cfg->irq_2_pin)
591 x86_io_apic_ops.eoi_ioapic_pin(entry->apic, entry->pin,
592 cfg->vector);
593 raw_spin_unlock_irqrestore(&ioapic_lock, flags);
596 static void clear_IO_APIC_pin(unsigned int apic, unsigned int pin)
598 struct IO_APIC_route_entry entry;
600 /* Check delivery_mode to be sure we're not clearing an SMI pin */
601 entry = ioapic_read_entry(apic, pin);
602 if (entry.delivery_mode == dest_SMI)
603 return;
606 * Make sure the entry is masked and re-read the contents to check
607 * if it is a level triggered pin and if the remote-IRR is set.
609 if (!entry.mask) {
610 entry.mask = 1;
611 ioapic_write_entry(apic, pin, entry);
612 entry = ioapic_read_entry(apic, pin);
615 if (entry.irr) {
616 unsigned long flags;
619 * Make sure the trigger mode is set to level. Explicit EOI
620 * doesn't clear the remote-IRR if the trigger mode is not
621 * set to level.
623 if (!entry.trigger) {
624 entry.trigger = IOAPIC_LEVEL;
625 ioapic_write_entry(apic, pin, entry);
628 raw_spin_lock_irqsave(&ioapic_lock, flags);
629 x86_io_apic_ops.eoi_ioapic_pin(apic, pin, entry.vector);
630 raw_spin_unlock_irqrestore(&ioapic_lock, flags);
634 * Clear the rest of the bits in the IO-APIC RTE except for the mask
635 * bit.
637 ioapic_mask_entry(apic, pin);
638 entry = ioapic_read_entry(apic, pin);
639 if (entry.irr)
640 pr_err("Unable to reset IRR for apic: %d, pin :%d\n",
641 mpc_ioapic_id(apic), pin);
644 static void clear_IO_APIC (void)
646 int apic, pin;
648 for (apic = 0; apic < nr_ioapics; apic++)
649 for (pin = 0; pin < ioapics[apic].nr_registers; pin++)
650 clear_IO_APIC_pin(apic, pin);
653 #ifdef CONFIG_X86_32
655 * support for broken MP BIOSs, enables hand-redirection of PIRQ0-7 to
656 * specific CPU-side IRQs.
659 #define MAX_PIRQS 8
660 static int pirq_entries[MAX_PIRQS] = {
661 [0 ... MAX_PIRQS - 1] = -1
664 static int __init ioapic_pirq_setup(char *str)
666 int i, max;
667 int ints[MAX_PIRQS+1];
669 get_options(str, ARRAY_SIZE(ints), ints);
671 apic_printk(APIC_VERBOSE, KERN_INFO
672 "PIRQ redirection, working around broken MP-BIOS.\n");
673 max = MAX_PIRQS;
674 if (ints[0] < MAX_PIRQS)
675 max = ints[0];
677 for (i = 0; i < max; i++) {
678 apic_printk(APIC_VERBOSE, KERN_DEBUG
679 "... PIRQ%d -> IRQ %d\n", i, ints[i+1]);
681 * PIRQs are mapped upside down, usually.
683 pirq_entries[MAX_PIRQS-i-1] = ints[i+1];
685 return 1;
688 __setup("pirq=", ioapic_pirq_setup);
689 #endif /* CONFIG_X86_32 */
692 * Saves all the IO-APIC RTE's
694 int save_ioapic_entries(void)
696 int apic, pin;
697 int err = 0;
699 for (apic = 0; apic < nr_ioapics; apic++) {
700 if (!ioapics[apic].saved_registers) {
701 err = -ENOMEM;
702 continue;
705 for (pin = 0; pin < ioapics[apic].nr_registers; pin++)
706 ioapics[apic].saved_registers[pin] =
707 ioapic_read_entry(apic, pin);
710 return err;
714 * Mask all IO APIC entries.
716 void mask_ioapic_entries(void)
718 int apic, pin;
720 for (apic = 0; apic < nr_ioapics; apic++) {
721 if (!ioapics[apic].saved_registers)
722 continue;
724 for (pin = 0; pin < ioapics[apic].nr_registers; pin++) {
725 struct IO_APIC_route_entry entry;
727 entry = ioapics[apic].saved_registers[pin];
728 if (!entry.mask) {
729 entry.mask = 1;
730 ioapic_write_entry(apic, pin, entry);
737 * Restore IO APIC entries which was saved in the ioapic structure.
739 int restore_ioapic_entries(void)
741 int apic, pin;
743 for (apic = 0; apic < nr_ioapics; apic++) {
744 if (!ioapics[apic].saved_registers)
745 continue;
747 for (pin = 0; pin < ioapics[apic].nr_registers; pin++)
748 ioapic_write_entry(apic, pin,
749 ioapics[apic].saved_registers[pin]);
751 return 0;
755 * Find the IRQ entry number of a certain pin.
757 static int find_irq_entry(int ioapic_idx, int pin, int type)
759 int i;
761 for (i = 0; i < mp_irq_entries; i++)
762 if (mp_irqs[i].irqtype == type &&
763 (mp_irqs[i].dstapic == mpc_ioapic_id(ioapic_idx) ||
764 mp_irqs[i].dstapic == MP_APIC_ALL) &&
765 mp_irqs[i].dstirq == pin)
766 return i;
768 return -1;
772 * Find the pin to which IRQ[irq] (ISA) is connected
774 static int __init find_isa_irq_pin(int irq, int type)
776 int i;
778 for (i = 0; i < mp_irq_entries; i++) {
779 int lbus = mp_irqs[i].srcbus;
781 if (test_bit(lbus, mp_bus_not_pci) &&
782 (mp_irqs[i].irqtype == type) &&
783 (mp_irqs[i].srcbusirq == irq))
785 return mp_irqs[i].dstirq;
787 return -1;
790 static int __init find_isa_irq_apic(int irq, int type)
792 int i;
794 for (i = 0; i < mp_irq_entries; i++) {
795 int lbus = mp_irqs[i].srcbus;
797 if (test_bit(lbus, mp_bus_not_pci) &&
798 (mp_irqs[i].irqtype == type) &&
799 (mp_irqs[i].srcbusirq == irq))
800 break;
803 if (i < mp_irq_entries) {
804 int ioapic_idx;
806 for (ioapic_idx = 0; ioapic_idx < nr_ioapics; ioapic_idx++)
807 if (mpc_ioapic_id(ioapic_idx) == mp_irqs[i].dstapic)
808 return ioapic_idx;
811 return -1;
814 #ifdef CONFIG_EISA
816 * EISA Edge/Level control register, ELCR
818 static int EISA_ELCR(unsigned int irq)
820 if (irq < legacy_pic->nr_legacy_irqs) {
821 unsigned int port = 0x4d0 + (irq >> 3);
822 return (inb(port) >> (irq & 7)) & 1;
824 apic_printk(APIC_VERBOSE, KERN_INFO
825 "Broken MPtable reports ISA irq %d\n", irq);
826 return 0;
829 #endif
831 /* ISA interrupts are always polarity zero edge triggered,
832 * when listed as conforming in the MP table. */
834 #define default_ISA_trigger(idx) (0)
835 #define default_ISA_polarity(idx) (0)
837 /* EISA interrupts are always polarity zero and can be edge or level
838 * trigger depending on the ELCR value. If an interrupt is listed as
839 * EISA conforming in the MP table, that means its trigger type must
840 * be read in from the ELCR */
842 #define default_EISA_trigger(idx) (EISA_ELCR(mp_irqs[idx].srcbusirq))
843 #define default_EISA_polarity(idx) default_ISA_polarity(idx)
845 /* PCI interrupts are always polarity one level triggered,
846 * when listed as conforming in the MP table. */
848 #define default_PCI_trigger(idx) (1)
849 #define default_PCI_polarity(idx) (1)
851 static int irq_polarity(int idx)
853 int bus = mp_irqs[idx].srcbus;
854 int polarity;
857 * Determine IRQ line polarity (high active or low active):
859 switch (mp_irqs[idx].irqflag & 3)
861 case 0: /* conforms, ie. bus-type dependent polarity */
862 if (test_bit(bus, mp_bus_not_pci))
863 polarity = default_ISA_polarity(idx);
864 else
865 polarity = default_PCI_polarity(idx);
866 break;
867 case 1: /* high active */
869 polarity = 0;
870 break;
872 case 2: /* reserved */
874 pr_warn("broken BIOS!!\n");
875 polarity = 1;
876 break;
878 case 3: /* low active */
880 polarity = 1;
881 break;
883 default: /* invalid */
885 pr_warn("broken BIOS!!\n");
886 polarity = 1;
887 break;
890 return polarity;
893 static int irq_trigger(int idx)
895 int bus = mp_irqs[idx].srcbus;
896 int trigger;
899 * Determine IRQ trigger mode (edge or level sensitive):
901 switch ((mp_irqs[idx].irqflag>>2) & 3)
903 case 0: /* conforms, ie. bus-type dependent */
904 if (test_bit(bus, mp_bus_not_pci))
905 trigger = default_ISA_trigger(idx);
906 else
907 trigger = default_PCI_trigger(idx);
908 #ifdef CONFIG_EISA
909 switch (mp_bus_id_to_type[bus]) {
910 case MP_BUS_ISA: /* ISA pin */
912 /* set before the switch */
913 break;
915 case MP_BUS_EISA: /* EISA pin */
917 trigger = default_EISA_trigger(idx);
918 break;
920 case MP_BUS_PCI: /* PCI pin */
922 /* set before the switch */
923 break;
925 default:
927 pr_warn("broken BIOS!!\n");
928 trigger = 1;
929 break;
932 #endif
933 break;
934 case 1: /* edge */
936 trigger = 0;
937 break;
939 case 2: /* reserved */
941 pr_warn("broken BIOS!!\n");
942 trigger = 1;
943 break;
945 case 3: /* level */
947 trigger = 1;
948 break;
950 default: /* invalid */
952 pr_warn("broken BIOS!!\n");
953 trigger = 0;
954 break;
957 return trigger;
960 static int pin_2_irq(int idx, int apic, int pin)
962 int irq;
963 int bus = mp_irqs[idx].srcbus;
964 struct mp_ioapic_gsi *gsi_cfg = mp_ioapic_gsi_routing(apic);
967 * Debugging check, we are in big trouble if this message pops up!
969 if (mp_irqs[idx].dstirq != pin)
970 pr_err("broken BIOS or MPTABLE parser, ayiee!!\n");
972 if (test_bit(bus, mp_bus_not_pci)) {
973 irq = mp_irqs[idx].srcbusirq;
974 } else {
975 u32 gsi = gsi_cfg->gsi_base + pin;
977 if (gsi >= NR_IRQS_LEGACY)
978 irq = gsi;
979 else
980 irq = gsi_top + gsi;
983 #ifdef CONFIG_X86_32
985 * PCI IRQ command line redirection. Yes, limits are hardcoded.
987 if ((pin >= 16) && (pin <= 23)) {
988 if (pirq_entries[pin-16] != -1) {
989 if (!pirq_entries[pin-16]) {
990 apic_printk(APIC_VERBOSE, KERN_DEBUG
991 "disabling PIRQ%d\n", pin-16);
992 } else {
993 irq = pirq_entries[pin-16];
994 apic_printk(APIC_VERBOSE, KERN_DEBUG
995 "using PIRQ%d -> IRQ %d\n",
996 pin-16, irq);
1000 #endif
1002 return irq;
1006 * Find a specific PCI IRQ entry.
1007 * Not an __init, possibly needed by modules
1009 int IO_APIC_get_PCI_irq_vector(int bus, int slot, int pin,
1010 struct io_apic_irq_attr *irq_attr)
1012 int ioapic_idx, i, best_guess = -1;
1014 apic_printk(APIC_DEBUG,
1015 "querying PCI -> IRQ mapping bus:%d, slot:%d, pin:%d.\n",
1016 bus, slot, pin);
1017 if (test_bit(bus, mp_bus_not_pci)) {
1018 apic_printk(APIC_VERBOSE,
1019 "PCI BIOS passed nonexistent PCI bus %d!\n", bus);
1020 return -1;
1022 for (i = 0; i < mp_irq_entries; i++) {
1023 int lbus = mp_irqs[i].srcbus;
1025 for (ioapic_idx = 0; ioapic_idx < nr_ioapics; ioapic_idx++)
1026 if (mpc_ioapic_id(ioapic_idx) == mp_irqs[i].dstapic ||
1027 mp_irqs[i].dstapic == MP_APIC_ALL)
1028 break;
1030 if (!test_bit(lbus, mp_bus_not_pci) &&
1031 !mp_irqs[i].irqtype &&
1032 (bus == lbus) &&
1033 (slot == ((mp_irqs[i].srcbusirq >> 2) & 0x1f))) {
1034 int irq = pin_2_irq(i, ioapic_idx, mp_irqs[i].dstirq);
1036 if (!(ioapic_idx || IO_APIC_IRQ(irq)))
1037 continue;
1039 if (pin == (mp_irqs[i].srcbusirq & 3)) {
1040 set_io_apic_irq_attr(irq_attr, ioapic_idx,
1041 mp_irqs[i].dstirq,
1042 irq_trigger(i),
1043 irq_polarity(i));
1044 return irq;
1047 * Use the first all-but-pin matching entry as a
1048 * best-guess fuzzy result for broken mptables.
1050 if (best_guess < 0) {
1051 set_io_apic_irq_attr(irq_attr, ioapic_idx,
1052 mp_irqs[i].dstirq,
1053 irq_trigger(i),
1054 irq_polarity(i));
1055 best_guess = irq;
1059 return best_guess;
1061 EXPORT_SYMBOL(IO_APIC_get_PCI_irq_vector);
1063 void lock_vector_lock(void)
1065 /* Used to the online set of cpus does not change
1066 * during assign_irq_vector.
1068 raw_spin_lock(&vector_lock);
1071 void unlock_vector_lock(void)
1073 raw_spin_unlock(&vector_lock);
1076 static int
1077 __assign_irq_vector(int irq, struct irq_cfg *cfg, const struct cpumask *mask)
1080 * NOTE! The local APIC isn't very good at handling
1081 * multiple interrupts at the same interrupt level.
1082 * As the interrupt level is determined by taking the
1083 * vector number and shifting that right by 4, we
1084 * want to spread these out a bit so that they don't
1085 * all fall in the same interrupt level.
1087 * Also, we've got to be careful not to trash gate
1088 * 0x80, because int 0x80 is hm, kind of importantish. ;)
1090 static int current_vector = FIRST_EXTERNAL_VECTOR + VECTOR_OFFSET_START;
1091 static int current_offset = VECTOR_OFFSET_START % 16;
1092 int cpu, err;
1093 cpumask_var_t tmp_mask;
1095 if (cfg->move_in_progress)
1096 return -EBUSY;
1098 if (!alloc_cpumask_var(&tmp_mask, GFP_ATOMIC))
1099 return -ENOMEM;
1101 /* Only try and allocate irqs on cpus that are present */
1102 err = -ENOSPC;
1103 cpumask_clear(cfg->old_domain);
1104 cpu = cpumask_first_and(mask, cpu_online_mask);
1105 while (cpu < nr_cpu_ids) {
1106 int new_cpu, vector, offset;
1108 apic->vector_allocation_domain(cpu, tmp_mask, mask);
1110 if (cpumask_subset(tmp_mask, cfg->domain)) {
1111 err = 0;
1112 if (cpumask_equal(tmp_mask, cfg->domain))
1113 break;
1115 * New cpumask using the vector is a proper subset of
1116 * the current in use mask. So cleanup the vector
1117 * allocation for the members that are not used anymore.
1119 cpumask_andnot(cfg->old_domain, cfg->domain, tmp_mask);
1120 cfg->move_in_progress =
1121 cpumask_intersects(cfg->old_domain, cpu_online_mask);
1122 cpumask_and(cfg->domain, cfg->domain, tmp_mask);
1123 break;
1126 vector = current_vector;
1127 offset = current_offset;
1128 next:
1129 vector += 16;
1130 if (vector >= first_system_vector) {
1131 offset = (offset + 1) % 16;
1132 vector = FIRST_EXTERNAL_VECTOR + offset;
1135 if (unlikely(current_vector == vector)) {
1136 cpumask_or(cfg->old_domain, cfg->old_domain, tmp_mask);
1137 cpumask_andnot(tmp_mask, mask, cfg->old_domain);
1138 cpu = cpumask_first_and(tmp_mask, cpu_online_mask);
1139 continue;
1142 if (test_bit(vector, used_vectors))
1143 goto next;
1145 for_each_cpu_and(new_cpu, tmp_mask, cpu_online_mask)
1146 if (per_cpu(vector_irq, new_cpu)[vector] != -1)
1147 goto next;
1148 /* Found one! */
1149 current_vector = vector;
1150 current_offset = offset;
1151 if (cfg->vector) {
1152 cpumask_copy(cfg->old_domain, cfg->domain);
1153 cfg->move_in_progress =
1154 cpumask_intersects(cfg->old_domain, cpu_online_mask);
1156 for_each_cpu_and(new_cpu, tmp_mask, cpu_online_mask)
1157 per_cpu(vector_irq, new_cpu)[vector] = irq;
1158 cfg->vector = vector;
1159 cpumask_copy(cfg->domain, tmp_mask);
1160 err = 0;
1161 break;
1163 free_cpumask_var(tmp_mask);
1164 return err;
1167 int assign_irq_vector(int irq, struct irq_cfg *cfg, const struct cpumask *mask)
1169 int err;
1170 unsigned long flags;
1172 raw_spin_lock_irqsave(&vector_lock, flags);
1173 err = __assign_irq_vector(irq, cfg, mask);
1174 raw_spin_unlock_irqrestore(&vector_lock, flags);
1175 return err;
1178 static void __clear_irq_vector(int irq, struct irq_cfg *cfg)
1180 int cpu, vector;
1182 BUG_ON(!cfg->vector);
1184 vector = cfg->vector;
1185 for_each_cpu_and(cpu, cfg->domain, cpu_online_mask)
1186 per_cpu(vector_irq, cpu)[vector] = -1;
1188 cfg->vector = 0;
1189 cpumask_clear(cfg->domain);
1191 if (likely(!cfg->move_in_progress))
1192 return;
1193 for_each_cpu_and(cpu, cfg->old_domain, cpu_online_mask) {
1194 for (vector = FIRST_EXTERNAL_VECTOR; vector < NR_VECTORS;
1195 vector++) {
1196 if (per_cpu(vector_irq, cpu)[vector] != irq)
1197 continue;
1198 per_cpu(vector_irq, cpu)[vector] = -1;
1199 break;
1202 cfg->move_in_progress = 0;
1205 void __setup_vector_irq(int cpu)
1207 /* Initialize vector_irq on a new cpu */
1208 int irq, vector;
1209 struct irq_cfg *cfg;
1212 * vector_lock will make sure that we don't run into irq vector
1213 * assignments that might be happening on another cpu in parallel,
1214 * while we setup our initial vector to irq mappings.
1216 raw_spin_lock(&vector_lock);
1217 /* Mark the inuse vectors */
1218 for_each_active_irq(irq) {
1219 cfg = irq_get_chip_data(irq);
1220 if (!cfg)
1221 continue;
1223 if (!cpumask_test_cpu(cpu, cfg->domain))
1224 continue;
1225 vector = cfg->vector;
1226 per_cpu(vector_irq, cpu)[vector] = irq;
1228 /* Mark the free vectors */
1229 for (vector = 0; vector < NR_VECTORS; ++vector) {
1230 irq = per_cpu(vector_irq, cpu)[vector];
1231 if (irq < 0)
1232 continue;
1234 cfg = irq_cfg(irq);
1235 if (!cpumask_test_cpu(cpu, cfg->domain))
1236 per_cpu(vector_irq, cpu)[vector] = -1;
1238 raw_spin_unlock(&vector_lock);
1241 static struct irq_chip ioapic_chip;
1243 #ifdef CONFIG_X86_32
1244 static inline int IO_APIC_irq_trigger(int irq)
1246 int apic, idx, pin;
1248 for (apic = 0; apic < nr_ioapics; apic++) {
1249 for (pin = 0; pin < ioapics[apic].nr_registers; pin++) {
1250 idx = find_irq_entry(apic, pin, mp_INT);
1251 if ((idx != -1) && (irq == pin_2_irq(idx, apic, pin)))
1252 return irq_trigger(idx);
1256 * nonexistent IRQs are edge default
1258 return 0;
1260 #else
1261 static inline int IO_APIC_irq_trigger(int irq)
1263 return 1;
1265 #endif
1267 static void ioapic_register_intr(unsigned int irq, struct irq_cfg *cfg,
1268 unsigned long trigger)
1270 struct irq_chip *chip = &ioapic_chip;
1271 irq_flow_handler_t hdl;
1272 bool fasteoi;
1274 if ((trigger == IOAPIC_AUTO && IO_APIC_irq_trigger(irq)) ||
1275 trigger == IOAPIC_LEVEL) {
1276 irq_set_status_flags(irq, IRQ_LEVEL);
1277 fasteoi = true;
1278 } else {
1279 irq_clear_status_flags(irq, IRQ_LEVEL);
1280 fasteoi = false;
1283 if (setup_remapped_irq(irq, cfg, chip))
1284 fasteoi = trigger != 0;
1286 hdl = fasteoi ? handle_fasteoi_irq : handle_edge_irq;
1287 irq_set_chip_and_handler_name(irq, chip, hdl,
1288 fasteoi ? "fasteoi" : "edge");
1291 int native_setup_ioapic_entry(int irq, struct IO_APIC_route_entry *entry,
1292 unsigned int destination, int vector,
1293 struct io_apic_irq_attr *attr)
1295 memset(entry, 0, sizeof(*entry));
1297 entry->delivery_mode = apic->irq_delivery_mode;
1298 entry->dest_mode = apic->irq_dest_mode;
1299 entry->dest = destination;
1300 entry->vector = vector;
1301 entry->mask = 0; /* enable IRQ */
1302 entry->trigger = attr->trigger;
1303 entry->polarity = attr->polarity;
1306 * Mask level triggered irqs.
1307 * Use IRQ_DELAYED_DISABLE for edge triggered irqs.
1309 if (attr->trigger)
1310 entry->mask = 1;
1312 return 0;
1315 static void setup_ioapic_irq(unsigned int irq, struct irq_cfg *cfg,
1316 struct io_apic_irq_attr *attr)
1318 struct IO_APIC_route_entry entry;
1319 unsigned int dest;
1321 if (!IO_APIC_IRQ(irq))
1322 return;
1324 if (assign_irq_vector(irq, cfg, apic->target_cpus()))
1325 return;
1327 if (apic->cpu_mask_to_apicid_and(cfg->domain, apic->target_cpus(),
1328 &dest)) {
1329 pr_warn("Failed to obtain apicid for ioapic %d, pin %d\n",
1330 mpc_ioapic_id(attr->ioapic), attr->ioapic_pin);
1331 __clear_irq_vector(irq, cfg);
1333 return;
1336 apic_printk(APIC_VERBOSE,KERN_DEBUG
1337 "IOAPIC[%d]: Set routing entry (%d-%d -> 0x%x -> "
1338 "IRQ %d Mode:%i Active:%i Dest:%d)\n",
1339 attr->ioapic, mpc_ioapic_id(attr->ioapic), attr->ioapic_pin,
1340 cfg->vector, irq, attr->trigger, attr->polarity, dest);
1342 if (x86_io_apic_ops.setup_entry(irq, &entry, dest, cfg->vector, attr)) {
1343 pr_warn("Failed to setup ioapic entry for ioapic %d, pin %d\n",
1344 mpc_ioapic_id(attr->ioapic), attr->ioapic_pin);
1345 __clear_irq_vector(irq, cfg);
1347 return;
1350 ioapic_register_intr(irq, cfg, attr->trigger);
1351 if (irq < legacy_pic->nr_legacy_irqs)
1352 legacy_pic->mask(irq);
1354 ioapic_write_entry(attr->ioapic, attr->ioapic_pin, entry);
1357 static bool __init io_apic_pin_not_connected(int idx, int ioapic_idx, int pin)
1359 if (idx != -1)
1360 return false;
1362 apic_printk(APIC_VERBOSE, KERN_DEBUG " apic %d pin %d not connected\n",
1363 mpc_ioapic_id(ioapic_idx), pin);
1364 return true;
1367 static void __init __io_apic_setup_irqs(unsigned int ioapic_idx)
1369 int idx, node = cpu_to_node(0);
1370 struct io_apic_irq_attr attr;
1371 unsigned int pin, irq;
1373 for (pin = 0; pin < ioapics[ioapic_idx].nr_registers; pin++) {
1374 idx = find_irq_entry(ioapic_idx, pin, mp_INT);
1375 if (io_apic_pin_not_connected(idx, ioapic_idx, pin))
1376 continue;
1378 irq = pin_2_irq(idx, ioapic_idx, pin);
1380 if ((ioapic_idx > 0) && (irq > 16))
1381 continue;
1384 * Skip the timer IRQ if there's a quirk handler
1385 * installed and if it returns 1:
1387 if (apic->multi_timer_check &&
1388 apic->multi_timer_check(ioapic_idx, irq))
1389 continue;
1391 set_io_apic_irq_attr(&attr, ioapic_idx, pin, irq_trigger(idx),
1392 irq_polarity(idx));
1394 io_apic_setup_irq_pin(irq, node, &attr);
1398 static void __init setup_IO_APIC_irqs(void)
1400 unsigned int ioapic_idx;
1402 apic_printk(APIC_VERBOSE, KERN_DEBUG "init IO_APIC IRQs\n");
1404 for (ioapic_idx = 0; ioapic_idx < nr_ioapics; ioapic_idx++)
1405 __io_apic_setup_irqs(ioapic_idx);
1409 * for the gsit that is not in first ioapic
1410 * but could not use acpi_register_gsi()
1411 * like some special sci in IBM x3330
1413 void setup_IO_APIC_irq_extra(u32 gsi)
1415 int ioapic_idx = 0, pin, idx, irq, node = cpu_to_node(0);
1416 struct io_apic_irq_attr attr;
1419 * Convert 'gsi' to 'ioapic.pin'.
1421 ioapic_idx = mp_find_ioapic(gsi);
1422 if (ioapic_idx < 0)
1423 return;
1425 pin = mp_find_ioapic_pin(ioapic_idx, gsi);
1426 idx = find_irq_entry(ioapic_idx, pin, mp_INT);
1427 if (idx == -1)
1428 return;
1430 irq = pin_2_irq(idx, ioapic_idx, pin);
1432 /* Only handle the non legacy irqs on secondary ioapics */
1433 if (ioapic_idx == 0 || irq < NR_IRQS_LEGACY)
1434 return;
1436 set_io_apic_irq_attr(&attr, ioapic_idx, pin, irq_trigger(idx),
1437 irq_polarity(idx));
1439 io_apic_setup_irq_pin_once(irq, node, &attr);
1443 * Set up the timer pin, possibly with the 8259A-master behind.
1445 static void __init setup_timer_IRQ0_pin(unsigned int ioapic_idx,
1446 unsigned int pin, int vector)
1448 struct IO_APIC_route_entry entry;
1449 unsigned int dest;
1451 memset(&entry, 0, sizeof(entry));
1454 * We use logical delivery to get the timer IRQ
1455 * to the first CPU.
1457 if (unlikely(apic->cpu_mask_to_apicid_and(apic->target_cpus(),
1458 apic->target_cpus(), &dest)))
1459 dest = BAD_APICID;
1461 entry.dest_mode = apic->irq_dest_mode;
1462 entry.mask = 0; /* don't mask IRQ for edge */
1463 entry.dest = dest;
1464 entry.delivery_mode = apic->irq_delivery_mode;
1465 entry.polarity = 0;
1466 entry.trigger = 0;
1467 entry.vector = vector;
1470 * The timer IRQ doesn't have to know that behind the
1471 * scene we may have a 8259A-master in AEOI mode ...
1473 irq_set_chip_and_handler_name(0, &ioapic_chip, handle_edge_irq,
1474 "edge");
1477 * Add it to the IO-APIC irq-routing table:
1479 ioapic_write_entry(ioapic_idx, pin, entry);
1482 void native_io_apic_print_entries(unsigned int apic, unsigned int nr_entries)
1484 int i;
1486 pr_debug(" NR Dst Mask Trig IRR Pol Stat Dmod Deli Vect:\n");
1488 for (i = 0; i <= nr_entries; i++) {
1489 struct IO_APIC_route_entry entry;
1491 entry = ioapic_read_entry(apic, i);
1493 pr_debug(" %02x %02X ", i, entry.dest);
1494 pr_cont("%1d %1d %1d %1d %1d "
1495 "%1d %1d %02X\n",
1496 entry.mask,
1497 entry.trigger,
1498 entry.irr,
1499 entry.polarity,
1500 entry.delivery_status,
1501 entry.dest_mode,
1502 entry.delivery_mode,
1503 entry.vector);
1507 void intel_ir_io_apic_print_entries(unsigned int apic,
1508 unsigned int nr_entries)
1510 int i;
1512 pr_debug(" NR Indx Fmt Mask Trig IRR Pol Stat Indx2 Zero Vect:\n");
1514 for (i = 0; i <= nr_entries; i++) {
1515 struct IR_IO_APIC_route_entry *ir_entry;
1516 struct IO_APIC_route_entry entry;
1518 entry = ioapic_read_entry(apic, i);
1520 ir_entry = (struct IR_IO_APIC_route_entry *)&entry;
1522 pr_debug(" %02x %04X ", i, ir_entry->index);
1523 pr_cont("%1d %1d %1d %1d %1d "
1524 "%1d %1d %X %02X\n",
1525 ir_entry->format,
1526 ir_entry->mask,
1527 ir_entry->trigger,
1528 ir_entry->irr,
1529 ir_entry->polarity,
1530 ir_entry->delivery_status,
1531 ir_entry->index2,
1532 ir_entry->zero,
1533 ir_entry->vector);
1537 void ioapic_zap_locks(void)
1539 raw_spin_lock_init(&ioapic_lock);
1542 __apicdebuginit(void) print_IO_APIC(int ioapic_idx)
1544 union IO_APIC_reg_00 reg_00;
1545 union IO_APIC_reg_01 reg_01;
1546 union IO_APIC_reg_02 reg_02;
1547 union IO_APIC_reg_03 reg_03;
1548 unsigned long flags;
1550 raw_spin_lock_irqsave(&ioapic_lock, flags);
1551 reg_00.raw = io_apic_read(ioapic_idx, 0);
1552 reg_01.raw = io_apic_read(ioapic_idx, 1);
1553 if (reg_01.bits.version >= 0x10)
1554 reg_02.raw = io_apic_read(ioapic_idx, 2);
1555 if (reg_01.bits.version >= 0x20)
1556 reg_03.raw = io_apic_read(ioapic_idx, 3);
1557 raw_spin_unlock_irqrestore(&ioapic_lock, flags);
1559 printk(KERN_DEBUG "IO APIC #%d......\n", mpc_ioapic_id(ioapic_idx));
1560 printk(KERN_DEBUG ".... register #00: %08X\n", reg_00.raw);
1561 printk(KERN_DEBUG "....... : physical APIC id: %02X\n", reg_00.bits.ID);
1562 printk(KERN_DEBUG "....... : Delivery Type: %X\n", reg_00.bits.delivery_type);
1563 printk(KERN_DEBUG "....... : LTS : %X\n", reg_00.bits.LTS);
1565 printk(KERN_DEBUG ".... register #01: %08X\n", *(int *)&reg_01);
1566 printk(KERN_DEBUG "....... : max redirection entries: %02X\n",
1567 reg_01.bits.entries);
1569 printk(KERN_DEBUG "....... : PRQ implemented: %X\n", reg_01.bits.PRQ);
1570 printk(KERN_DEBUG "....... : IO APIC version: %02X\n",
1571 reg_01.bits.version);
1574 * Some Intel chipsets with IO APIC VERSION of 0x1? don't have reg_02,
1575 * but the value of reg_02 is read as the previous read register
1576 * value, so ignore it if reg_02 == reg_01.
1578 if (reg_01.bits.version >= 0x10 && reg_02.raw != reg_01.raw) {
1579 printk(KERN_DEBUG ".... register #02: %08X\n", reg_02.raw);
1580 printk(KERN_DEBUG "....... : arbitration: %02X\n", reg_02.bits.arbitration);
1584 * Some Intel chipsets with IO APIC VERSION of 0x2? don't have reg_02
1585 * or reg_03, but the value of reg_0[23] is read as the previous read
1586 * register value, so ignore it if reg_03 == reg_0[12].
1588 if (reg_01.bits.version >= 0x20 && reg_03.raw != reg_02.raw &&
1589 reg_03.raw != reg_01.raw) {
1590 printk(KERN_DEBUG ".... register #03: %08X\n", reg_03.raw);
1591 printk(KERN_DEBUG "....... : Boot DT : %X\n", reg_03.bits.boot_DT);
1594 printk(KERN_DEBUG ".... IRQ redirection table:\n");
1596 x86_io_apic_ops.print_entries(ioapic_idx, reg_01.bits.entries);
1599 __apicdebuginit(void) print_IO_APICs(void)
1601 int ioapic_idx;
1602 struct irq_cfg *cfg;
1603 unsigned int irq;
1604 struct irq_chip *chip;
1606 printk(KERN_DEBUG "number of MP IRQ sources: %d.\n", mp_irq_entries);
1607 for (ioapic_idx = 0; ioapic_idx < nr_ioapics; ioapic_idx++)
1608 printk(KERN_DEBUG "number of IO-APIC #%d registers: %d.\n",
1609 mpc_ioapic_id(ioapic_idx),
1610 ioapics[ioapic_idx].nr_registers);
1613 * We are a bit conservative about what we expect. We have to
1614 * know about every hardware change ASAP.
1616 printk(KERN_INFO "testing the IO APIC.......................\n");
1618 for (ioapic_idx = 0; ioapic_idx < nr_ioapics; ioapic_idx++)
1619 print_IO_APIC(ioapic_idx);
1621 printk(KERN_DEBUG "IRQ to pin mappings:\n");
1622 for_each_active_irq(irq) {
1623 struct irq_pin_list *entry;
1625 chip = irq_get_chip(irq);
1626 if (chip != &ioapic_chip)
1627 continue;
1629 cfg = irq_get_chip_data(irq);
1630 if (!cfg)
1631 continue;
1632 entry = cfg->irq_2_pin;
1633 if (!entry)
1634 continue;
1635 printk(KERN_DEBUG "IRQ%d ", irq);
1636 for_each_irq_pin(entry, cfg->irq_2_pin)
1637 pr_cont("-> %d:%d", entry->apic, entry->pin);
1638 pr_cont("\n");
1641 printk(KERN_INFO ".................................... done.\n");
1644 __apicdebuginit(void) print_APIC_field(int base)
1646 int i;
1648 printk(KERN_DEBUG);
1650 for (i = 0; i < 8; i++)
1651 pr_cont("%08x", apic_read(base + i*0x10));
1653 pr_cont("\n");
1656 __apicdebuginit(void) print_local_APIC(void *dummy)
1658 unsigned int i, v, ver, maxlvt;
1659 u64 icr;
1661 printk(KERN_DEBUG "printing local APIC contents on CPU#%d/%d:\n",
1662 smp_processor_id(), hard_smp_processor_id());
1663 v = apic_read(APIC_ID);
1664 printk(KERN_INFO "... APIC ID: %08x (%01x)\n", v, read_apic_id());
1665 v = apic_read(APIC_LVR);
1666 printk(KERN_INFO "... APIC VERSION: %08x\n", v);
1667 ver = GET_APIC_VERSION(v);
1668 maxlvt = lapic_get_maxlvt();
1670 v = apic_read(APIC_TASKPRI);
1671 printk(KERN_DEBUG "... APIC TASKPRI: %08x (%02x)\n", v, v & APIC_TPRI_MASK);
1673 if (APIC_INTEGRATED(ver)) { /* !82489DX */
1674 if (!APIC_XAPIC(ver)) {
1675 v = apic_read(APIC_ARBPRI);
1676 printk(KERN_DEBUG "... APIC ARBPRI: %08x (%02x)\n", v,
1677 v & APIC_ARBPRI_MASK);
1679 v = apic_read(APIC_PROCPRI);
1680 printk(KERN_DEBUG "... APIC PROCPRI: %08x\n", v);
1684 * Remote read supported only in the 82489DX and local APIC for
1685 * Pentium processors.
1687 if (!APIC_INTEGRATED(ver) || maxlvt == 3) {
1688 v = apic_read(APIC_RRR);
1689 printk(KERN_DEBUG "... APIC RRR: %08x\n", v);
1692 v = apic_read(APIC_LDR);
1693 printk(KERN_DEBUG "... APIC LDR: %08x\n", v);
1694 if (!x2apic_enabled()) {
1695 v = apic_read(APIC_DFR);
1696 printk(KERN_DEBUG "... APIC DFR: %08x\n", v);
1698 v = apic_read(APIC_SPIV);
1699 printk(KERN_DEBUG "... APIC SPIV: %08x\n", v);
1701 printk(KERN_DEBUG "... APIC ISR field:\n");
1702 print_APIC_field(APIC_ISR);
1703 printk(KERN_DEBUG "... APIC TMR field:\n");
1704 print_APIC_field(APIC_TMR);
1705 printk(KERN_DEBUG "... APIC IRR field:\n");
1706 print_APIC_field(APIC_IRR);
1708 if (APIC_INTEGRATED(ver)) { /* !82489DX */
1709 if (maxlvt > 3) /* Due to the Pentium erratum 3AP. */
1710 apic_write(APIC_ESR, 0);
1712 v = apic_read(APIC_ESR);
1713 printk(KERN_DEBUG "... APIC ESR: %08x\n", v);
1716 icr = apic_icr_read();
1717 printk(KERN_DEBUG "... APIC ICR: %08x\n", (u32)icr);
1718 printk(KERN_DEBUG "... APIC ICR2: %08x\n", (u32)(icr >> 32));
1720 v = apic_read(APIC_LVTT);
1721 printk(KERN_DEBUG "... APIC LVTT: %08x\n", v);
1723 if (maxlvt > 3) { /* PC is LVT#4. */
1724 v = apic_read(APIC_LVTPC);
1725 printk(KERN_DEBUG "... APIC LVTPC: %08x\n", v);
1727 v = apic_read(APIC_LVT0);
1728 printk(KERN_DEBUG "... APIC LVT0: %08x\n", v);
1729 v = apic_read(APIC_LVT1);
1730 printk(KERN_DEBUG "... APIC LVT1: %08x\n", v);
1732 if (maxlvt > 2) { /* ERR is LVT#3. */
1733 v = apic_read(APIC_LVTERR);
1734 printk(KERN_DEBUG "... APIC LVTERR: %08x\n", v);
1737 v = apic_read(APIC_TMICT);
1738 printk(KERN_DEBUG "... APIC TMICT: %08x\n", v);
1739 v = apic_read(APIC_TMCCT);
1740 printk(KERN_DEBUG "... APIC TMCCT: %08x\n", v);
1741 v = apic_read(APIC_TDCR);
1742 printk(KERN_DEBUG "... APIC TDCR: %08x\n", v);
1744 if (boot_cpu_has(X86_FEATURE_EXTAPIC)) {
1745 v = apic_read(APIC_EFEAT);
1746 maxlvt = (v >> 16) & 0xff;
1747 printk(KERN_DEBUG "... APIC EFEAT: %08x\n", v);
1748 v = apic_read(APIC_ECTRL);
1749 printk(KERN_DEBUG "... APIC ECTRL: %08x\n", v);
1750 for (i = 0; i < maxlvt; i++) {
1751 v = apic_read(APIC_EILVTn(i));
1752 printk(KERN_DEBUG "... APIC EILVT%d: %08x\n", i, v);
1755 pr_cont("\n");
1758 __apicdebuginit(void) print_local_APICs(int maxcpu)
1760 int cpu;
1762 if (!maxcpu)
1763 return;
1765 preempt_disable();
1766 for_each_online_cpu(cpu) {
1767 if (cpu >= maxcpu)
1768 break;
1769 smp_call_function_single(cpu, print_local_APIC, NULL, 1);
1771 preempt_enable();
1774 __apicdebuginit(void) print_PIC(void)
1776 unsigned int v;
1777 unsigned long flags;
1779 if (!legacy_pic->nr_legacy_irqs)
1780 return;
1782 printk(KERN_DEBUG "\nprinting PIC contents\n");
1784 raw_spin_lock_irqsave(&i8259A_lock, flags);
1786 v = inb(0xa1) << 8 | inb(0x21);
1787 printk(KERN_DEBUG "... PIC IMR: %04x\n", v);
1789 v = inb(0xa0) << 8 | inb(0x20);
1790 printk(KERN_DEBUG "... PIC IRR: %04x\n", v);
1792 outb(0x0b,0xa0);
1793 outb(0x0b,0x20);
1794 v = inb(0xa0) << 8 | inb(0x20);
1795 outb(0x0a,0xa0);
1796 outb(0x0a,0x20);
1798 raw_spin_unlock_irqrestore(&i8259A_lock, flags);
1800 printk(KERN_DEBUG "... PIC ISR: %04x\n", v);
1802 v = inb(0x4d1) << 8 | inb(0x4d0);
1803 printk(KERN_DEBUG "... PIC ELCR: %04x\n", v);
1806 static int __initdata show_lapic = 1;
1807 static __init int setup_show_lapic(char *arg)
1809 int num = -1;
1811 if (strcmp(arg, "all") == 0) {
1812 show_lapic = CONFIG_NR_CPUS;
1813 } else {
1814 get_option(&arg, &num);
1815 if (num >= 0)
1816 show_lapic = num;
1819 return 1;
1821 __setup("show_lapic=", setup_show_lapic);
1823 __apicdebuginit(int) print_ICs(void)
1825 if (apic_verbosity == APIC_QUIET)
1826 return 0;
1828 print_PIC();
1830 /* don't print out if apic is not there */
1831 if (!cpu_has_apic && !apic_from_smp_config())
1832 return 0;
1834 print_local_APICs(show_lapic);
1835 print_IO_APICs();
1837 return 0;
1840 late_initcall(print_ICs);
1843 /* Where if anywhere is the i8259 connect in external int mode */
1844 static struct { int pin, apic; } ioapic_i8259 = { -1, -1 };
1846 void __init enable_IO_APIC(void)
1848 int i8259_apic, i8259_pin;
1849 int apic;
1851 if (!legacy_pic->nr_legacy_irqs)
1852 return;
1854 for(apic = 0; apic < nr_ioapics; apic++) {
1855 int pin;
1856 /* See if any of the pins is in ExtINT mode */
1857 for (pin = 0; pin < ioapics[apic].nr_registers; pin++) {
1858 struct IO_APIC_route_entry entry;
1859 entry = ioapic_read_entry(apic, pin);
1861 /* If the interrupt line is enabled and in ExtInt mode
1862 * I have found the pin where the i8259 is connected.
1864 if ((entry.mask == 0) && (entry.delivery_mode == dest_ExtINT)) {
1865 ioapic_i8259.apic = apic;
1866 ioapic_i8259.pin = pin;
1867 goto found_i8259;
1871 found_i8259:
1872 /* Look to see what if the MP table has reported the ExtINT */
1873 /* If we could not find the appropriate pin by looking at the ioapic
1874 * the i8259 probably is not connected the ioapic but give the
1875 * mptable a chance anyway.
1877 i8259_pin = find_isa_irq_pin(0, mp_ExtINT);
1878 i8259_apic = find_isa_irq_apic(0, mp_ExtINT);
1879 /* Trust the MP table if nothing is setup in the hardware */
1880 if ((ioapic_i8259.pin == -1) && (i8259_pin >= 0)) {
1881 printk(KERN_WARNING "ExtINT not setup in hardware but reported by MP table\n");
1882 ioapic_i8259.pin = i8259_pin;
1883 ioapic_i8259.apic = i8259_apic;
1885 /* Complain if the MP table and the hardware disagree */
1886 if (((ioapic_i8259.apic != i8259_apic) || (ioapic_i8259.pin != i8259_pin)) &&
1887 (i8259_pin >= 0) && (ioapic_i8259.pin >= 0))
1889 printk(KERN_WARNING "ExtINT in hardware and MP table differ\n");
1893 * Do not trust the IO-APIC being empty at bootup
1895 clear_IO_APIC();
1898 void native_disable_io_apic(void)
1901 * If the i8259 is routed through an IOAPIC
1902 * Put that IOAPIC in virtual wire mode
1903 * so legacy interrupts can be delivered.
1905 if (ioapic_i8259.pin != -1) {
1906 struct IO_APIC_route_entry entry;
1908 memset(&entry, 0, sizeof(entry));
1909 entry.mask = 0; /* Enabled */
1910 entry.trigger = 0; /* Edge */
1911 entry.irr = 0;
1912 entry.polarity = 0; /* High */
1913 entry.delivery_status = 0;
1914 entry.dest_mode = 0; /* Physical */
1915 entry.delivery_mode = dest_ExtINT; /* ExtInt */
1916 entry.vector = 0;
1917 entry.dest = read_apic_id();
1920 * Add it to the IO-APIC irq-routing table:
1922 ioapic_write_entry(ioapic_i8259.apic, ioapic_i8259.pin, entry);
1925 if (cpu_has_apic || apic_from_smp_config())
1926 disconnect_bsp_APIC(ioapic_i8259.pin != -1);
1931 * Not an __init, needed by the reboot code
1933 void disable_IO_APIC(void)
1936 * Clear the IO-APIC before rebooting:
1938 clear_IO_APIC();
1940 if (!legacy_pic->nr_legacy_irqs)
1941 return;
1943 x86_io_apic_ops.disable();
1946 #ifdef CONFIG_X86_32
1948 * function to set the IO-APIC physical IDs based on the
1949 * values stored in the MPC table.
1951 * by Matt Domsch <Matt_Domsch@dell.com> Tue Dec 21 12:25:05 CST 1999
1953 void __init setup_ioapic_ids_from_mpc_nocheck(void)
1955 union IO_APIC_reg_00 reg_00;
1956 physid_mask_t phys_id_present_map;
1957 int ioapic_idx;
1958 int i;
1959 unsigned char old_id;
1960 unsigned long flags;
1963 * This is broken; anything with a real cpu count has to
1964 * circumvent this idiocy regardless.
1966 apic->ioapic_phys_id_map(&phys_cpu_present_map, &phys_id_present_map);
1969 * Set the IOAPIC ID to the value stored in the MPC table.
1971 for (ioapic_idx = 0; ioapic_idx < nr_ioapics; ioapic_idx++) {
1972 /* Read the register 0 value */
1973 raw_spin_lock_irqsave(&ioapic_lock, flags);
1974 reg_00.raw = io_apic_read(ioapic_idx, 0);
1975 raw_spin_unlock_irqrestore(&ioapic_lock, flags);
1977 old_id = mpc_ioapic_id(ioapic_idx);
1979 if (mpc_ioapic_id(ioapic_idx) >= get_physical_broadcast()) {
1980 printk(KERN_ERR "BIOS bug, IO-APIC#%d ID is %d in the MPC table!...\n",
1981 ioapic_idx, mpc_ioapic_id(ioapic_idx));
1982 printk(KERN_ERR "... fixing up to %d. (tell your hw vendor)\n",
1983 reg_00.bits.ID);
1984 ioapics[ioapic_idx].mp_config.apicid = reg_00.bits.ID;
1988 * Sanity check, is the ID really free? Every APIC in a
1989 * system must have a unique ID or we get lots of nice
1990 * 'stuck on smp_invalidate_needed IPI wait' messages.
1992 if (apic->check_apicid_used(&phys_id_present_map,
1993 mpc_ioapic_id(ioapic_idx))) {
1994 printk(KERN_ERR "BIOS bug, IO-APIC#%d ID %d is already used!...\n",
1995 ioapic_idx, mpc_ioapic_id(ioapic_idx));
1996 for (i = 0; i < get_physical_broadcast(); i++)
1997 if (!physid_isset(i, phys_id_present_map))
1998 break;
1999 if (i >= get_physical_broadcast())
2000 panic("Max APIC ID exceeded!\n");
2001 printk(KERN_ERR "... fixing up to %d. (tell your hw vendor)\n",
2003 physid_set(i, phys_id_present_map);
2004 ioapics[ioapic_idx].mp_config.apicid = i;
2005 } else {
2006 physid_mask_t tmp;
2007 apic->apicid_to_cpu_present(mpc_ioapic_id(ioapic_idx),
2008 &tmp);
2009 apic_printk(APIC_VERBOSE, "Setting %d in the "
2010 "phys_id_present_map\n",
2011 mpc_ioapic_id(ioapic_idx));
2012 physids_or(phys_id_present_map, phys_id_present_map, tmp);
2016 * We need to adjust the IRQ routing table
2017 * if the ID changed.
2019 if (old_id != mpc_ioapic_id(ioapic_idx))
2020 for (i = 0; i < mp_irq_entries; i++)
2021 if (mp_irqs[i].dstapic == old_id)
2022 mp_irqs[i].dstapic
2023 = mpc_ioapic_id(ioapic_idx);
2026 * Update the ID register according to the right value
2027 * from the MPC table if they are different.
2029 if (mpc_ioapic_id(ioapic_idx) == reg_00.bits.ID)
2030 continue;
2032 apic_printk(APIC_VERBOSE, KERN_INFO
2033 "...changing IO-APIC physical APIC ID to %d ...",
2034 mpc_ioapic_id(ioapic_idx));
2036 reg_00.bits.ID = mpc_ioapic_id(ioapic_idx);
2037 raw_spin_lock_irqsave(&ioapic_lock, flags);
2038 io_apic_write(ioapic_idx, 0, reg_00.raw);
2039 raw_spin_unlock_irqrestore(&ioapic_lock, flags);
2042 * Sanity check
2044 raw_spin_lock_irqsave(&ioapic_lock, flags);
2045 reg_00.raw = io_apic_read(ioapic_idx, 0);
2046 raw_spin_unlock_irqrestore(&ioapic_lock, flags);
2047 if (reg_00.bits.ID != mpc_ioapic_id(ioapic_idx))
2048 pr_cont("could not set ID!\n");
2049 else
2050 apic_printk(APIC_VERBOSE, " ok.\n");
2054 void __init setup_ioapic_ids_from_mpc(void)
2057 if (acpi_ioapic)
2058 return;
2060 * Don't check I/O APIC IDs for xAPIC systems. They have
2061 * no meaning without the serial APIC bus.
2063 if (!(boot_cpu_data.x86_vendor == X86_VENDOR_INTEL)
2064 || APIC_XAPIC(apic_version[boot_cpu_physical_apicid]))
2065 return;
2066 setup_ioapic_ids_from_mpc_nocheck();
2068 #endif
2070 int no_timer_check __initdata;
2072 static int __init notimercheck(char *s)
2074 no_timer_check = 1;
2075 return 1;
2077 __setup("no_timer_check", notimercheck);
2080 * There is a nasty bug in some older SMP boards, their mptable lies
2081 * about the timer IRQ. We do the following to work around the situation:
2083 * - timer IRQ defaults to IO-APIC IRQ
2084 * - if this function detects that timer IRQs are defunct, then we fall
2085 * back to ISA timer IRQs
2087 static int __init timer_irq_works(void)
2089 unsigned long t1 = jiffies;
2090 unsigned long flags;
2092 if (no_timer_check)
2093 return 1;
2095 local_save_flags(flags);
2096 local_irq_enable();
2097 /* Let ten ticks pass... */
2098 mdelay((10 * 1000) / HZ);
2099 local_irq_restore(flags);
2102 * Expect a few ticks at least, to be sure some possible
2103 * glue logic does not lock up after one or two first
2104 * ticks in a non-ExtINT mode. Also the local APIC
2105 * might have cached one ExtINT interrupt. Finally, at
2106 * least one tick may be lost due to delays.
2109 /* jiffies wrap? */
2110 if (time_after(jiffies, t1 + 4))
2111 return 1;
2112 return 0;
2116 * In the SMP+IOAPIC case it might happen that there are an unspecified
2117 * number of pending IRQ events unhandled. These cases are very rare,
2118 * so we 'resend' these IRQs via IPIs, to the same CPU. It's much
2119 * better to do it this way as thus we do not have to be aware of
2120 * 'pending' interrupts in the IRQ path, except at this point.
2123 * Edge triggered needs to resend any interrupt
2124 * that was delayed but this is now handled in the device
2125 * independent code.
2129 * Starting up a edge-triggered IO-APIC interrupt is
2130 * nasty - we need to make sure that we get the edge.
2131 * If it is already asserted for some reason, we need
2132 * return 1 to indicate that is was pending.
2134 * This is not complete - we should be able to fake
2135 * an edge even if it isn't on the 8259A...
2138 static unsigned int startup_ioapic_irq(struct irq_data *data)
2140 int was_pending = 0, irq = data->irq;
2141 unsigned long flags;
2143 raw_spin_lock_irqsave(&ioapic_lock, flags);
2144 if (irq < legacy_pic->nr_legacy_irqs) {
2145 legacy_pic->mask(irq);
2146 if (legacy_pic->irq_pending(irq))
2147 was_pending = 1;
2149 __unmask_ioapic(data->chip_data);
2150 raw_spin_unlock_irqrestore(&ioapic_lock, flags);
2152 return was_pending;
2155 static int ioapic_retrigger_irq(struct irq_data *data)
2157 struct irq_cfg *cfg = data->chip_data;
2158 unsigned long flags;
2159 int cpu;
2161 raw_spin_lock_irqsave(&vector_lock, flags);
2162 cpu = cpumask_first_and(cfg->domain, cpu_online_mask);
2163 apic->send_IPI_mask(cpumask_of(cpu), cfg->vector);
2164 raw_spin_unlock_irqrestore(&vector_lock, flags);
2166 return 1;
2170 * Level and edge triggered IO-APIC interrupts need different handling,
2171 * so we use two separate IRQ descriptors. Edge triggered IRQs can be
2172 * handled with the level-triggered descriptor, but that one has slightly
2173 * more overhead. Level-triggered interrupts cannot be handled with the
2174 * edge-triggered handler, without risking IRQ storms and other ugly
2175 * races.
2178 #ifdef CONFIG_SMP
2179 void send_cleanup_vector(struct irq_cfg *cfg)
2181 cpumask_var_t cleanup_mask;
2183 if (unlikely(!alloc_cpumask_var(&cleanup_mask, GFP_ATOMIC))) {
2184 unsigned int i;
2185 for_each_cpu_and(i, cfg->old_domain, cpu_online_mask)
2186 apic->send_IPI_mask(cpumask_of(i), IRQ_MOVE_CLEANUP_VECTOR);
2187 } else {
2188 cpumask_and(cleanup_mask, cfg->old_domain, cpu_online_mask);
2189 apic->send_IPI_mask(cleanup_mask, IRQ_MOVE_CLEANUP_VECTOR);
2190 free_cpumask_var(cleanup_mask);
2192 cfg->move_in_progress = 0;
2195 asmlinkage void smp_irq_move_cleanup_interrupt(void)
2197 unsigned vector, me;
2199 ack_APIC_irq();
2200 irq_enter();
2201 exit_idle();
2203 me = smp_processor_id();
2204 for (vector = FIRST_EXTERNAL_VECTOR; vector < NR_VECTORS; vector++) {
2205 unsigned int irq;
2206 unsigned int irr;
2207 struct irq_desc *desc;
2208 struct irq_cfg *cfg;
2209 irq = __this_cpu_read(vector_irq[vector]);
2211 if (irq == -1)
2212 continue;
2214 desc = irq_to_desc(irq);
2215 if (!desc)
2216 continue;
2218 cfg = irq_cfg(irq);
2219 if (!cfg)
2220 continue;
2222 raw_spin_lock(&desc->lock);
2225 * Check if the irq migration is in progress. If so, we
2226 * haven't received the cleanup request yet for this irq.
2228 if (cfg->move_in_progress)
2229 goto unlock;
2231 if (vector == cfg->vector && cpumask_test_cpu(me, cfg->domain))
2232 goto unlock;
2234 irr = apic_read(APIC_IRR + (vector / 32 * 0x10));
2236 * Check if the vector that needs to be cleanedup is
2237 * registered at the cpu's IRR. If so, then this is not
2238 * the best time to clean it up. Lets clean it up in the
2239 * next attempt by sending another IRQ_MOVE_CLEANUP_VECTOR
2240 * to myself.
2242 if (irr & (1 << (vector % 32))) {
2243 apic->send_IPI_self(IRQ_MOVE_CLEANUP_VECTOR);
2244 goto unlock;
2246 __this_cpu_write(vector_irq[vector], -1);
2247 unlock:
2248 raw_spin_unlock(&desc->lock);
2251 irq_exit();
2254 static void __irq_complete_move(struct irq_cfg *cfg, unsigned vector)
2256 unsigned me;
2258 if (likely(!cfg->move_in_progress))
2259 return;
2261 me = smp_processor_id();
2263 if (vector == cfg->vector && cpumask_test_cpu(me, cfg->domain))
2264 send_cleanup_vector(cfg);
2267 static void irq_complete_move(struct irq_cfg *cfg)
2269 __irq_complete_move(cfg, ~get_irq_regs()->orig_ax);
2272 void irq_force_complete_move(int irq)
2274 struct irq_cfg *cfg = irq_get_chip_data(irq);
2276 if (!cfg)
2277 return;
2279 __irq_complete_move(cfg, cfg->vector);
2281 #else
2282 static inline void irq_complete_move(struct irq_cfg *cfg) { }
2283 #endif
2285 static void __target_IO_APIC_irq(unsigned int irq, unsigned int dest, struct irq_cfg *cfg)
2287 int apic, pin;
2288 struct irq_pin_list *entry;
2289 u8 vector = cfg->vector;
2291 for_each_irq_pin(entry, cfg->irq_2_pin) {
2292 unsigned int reg;
2294 apic = entry->apic;
2295 pin = entry->pin;
2297 io_apic_write(apic, 0x11 + pin*2, dest);
2298 reg = io_apic_read(apic, 0x10 + pin*2);
2299 reg &= ~IO_APIC_REDIR_VECTOR_MASK;
2300 reg |= vector;
2301 io_apic_modify(apic, 0x10 + pin*2, reg);
2306 * Either sets data->affinity to a valid value, and returns
2307 * ->cpu_mask_to_apicid of that in dest_id, or returns -1 and
2308 * leaves data->affinity untouched.
2310 int __ioapic_set_affinity(struct irq_data *data, const struct cpumask *mask,
2311 unsigned int *dest_id)
2313 struct irq_cfg *cfg = data->chip_data;
2314 unsigned int irq = data->irq;
2315 int err;
2317 if (!config_enabled(CONFIG_SMP))
2318 return -1;
2320 if (!cpumask_intersects(mask, cpu_online_mask))
2321 return -EINVAL;
2323 err = assign_irq_vector(irq, cfg, mask);
2324 if (err)
2325 return err;
2327 err = apic->cpu_mask_to_apicid_and(mask, cfg->domain, dest_id);
2328 if (err) {
2329 if (assign_irq_vector(irq, cfg, data->affinity))
2330 pr_err("Failed to recover vector for irq %d\n", irq);
2331 return err;
2334 cpumask_copy(data->affinity, mask);
2336 return 0;
2340 int native_ioapic_set_affinity(struct irq_data *data,
2341 const struct cpumask *mask,
2342 bool force)
2344 unsigned int dest, irq = data->irq;
2345 unsigned long flags;
2346 int ret;
2348 if (!config_enabled(CONFIG_SMP))
2349 return -1;
2351 raw_spin_lock_irqsave(&ioapic_lock, flags);
2352 ret = __ioapic_set_affinity(data, mask, &dest);
2353 if (!ret) {
2354 /* Only the high 8 bits are valid. */
2355 dest = SET_APIC_LOGICAL_ID(dest);
2356 __target_IO_APIC_irq(irq, dest, data->chip_data);
2357 ret = IRQ_SET_MASK_OK_NOCOPY;
2359 raw_spin_unlock_irqrestore(&ioapic_lock, flags);
2360 return ret;
2363 static void ack_apic_edge(struct irq_data *data)
2365 irq_complete_move(data->chip_data);
2366 irq_move_irq(data);
2367 ack_APIC_irq();
2370 atomic_t irq_mis_count;
2372 #ifdef CONFIG_GENERIC_PENDING_IRQ
2373 static bool io_apic_level_ack_pending(struct irq_cfg *cfg)
2375 struct irq_pin_list *entry;
2376 unsigned long flags;
2378 raw_spin_lock_irqsave(&ioapic_lock, flags);
2379 for_each_irq_pin(entry, cfg->irq_2_pin) {
2380 unsigned int reg;
2381 int pin;
2383 pin = entry->pin;
2384 reg = io_apic_read(entry->apic, 0x10 + pin*2);
2385 /* Is the remote IRR bit set? */
2386 if (reg & IO_APIC_REDIR_REMOTE_IRR) {
2387 raw_spin_unlock_irqrestore(&ioapic_lock, flags);
2388 return true;
2391 raw_spin_unlock_irqrestore(&ioapic_lock, flags);
2393 return false;
2396 static inline bool ioapic_irqd_mask(struct irq_data *data, struct irq_cfg *cfg)
2398 /* If we are moving the irq we need to mask it */
2399 if (unlikely(irqd_is_setaffinity_pending(data))) {
2400 mask_ioapic(cfg);
2401 return true;
2403 return false;
2406 static inline void ioapic_irqd_unmask(struct irq_data *data,
2407 struct irq_cfg *cfg, bool masked)
2409 if (unlikely(masked)) {
2410 /* Only migrate the irq if the ack has been received.
2412 * On rare occasions the broadcast level triggered ack gets
2413 * delayed going to ioapics, and if we reprogram the
2414 * vector while Remote IRR is still set the irq will never
2415 * fire again.
2417 * To prevent this scenario we read the Remote IRR bit
2418 * of the ioapic. This has two effects.
2419 * - On any sane system the read of the ioapic will
2420 * flush writes (and acks) going to the ioapic from
2421 * this cpu.
2422 * - We get to see if the ACK has actually been delivered.
2424 * Based on failed experiments of reprogramming the
2425 * ioapic entry from outside of irq context starting
2426 * with masking the ioapic entry and then polling until
2427 * Remote IRR was clear before reprogramming the
2428 * ioapic I don't trust the Remote IRR bit to be
2429 * completey accurate.
2431 * However there appears to be no other way to plug
2432 * this race, so if the Remote IRR bit is not
2433 * accurate and is causing problems then it is a hardware bug
2434 * and you can go talk to the chipset vendor about it.
2436 if (!io_apic_level_ack_pending(cfg))
2437 irq_move_masked_irq(data);
2438 unmask_ioapic(cfg);
2441 #else
2442 static inline bool ioapic_irqd_mask(struct irq_data *data, struct irq_cfg *cfg)
2444 return false;
2446 static inline void ioapic_irqd_unmask(struct irq_data *data,
2447 struct irq_cfg *cfg, bool masked)
2450 #endif
2452 static void ack_apic_level(struct irq_data *data)
2454 struct irq_cfg *cfg = data->chip_data;
2455 int i, irq = data->irq;
2456 unsigned long v;
2457 bool masked;
2459 irq_complete_move(cfg);
2460 masked = ioapic_irqd_mask(data, cfg);
2463 * It appears there is an erratum which affects at least version 0x11
2464 * of I/O APIC (that's the 82093AA and cores integrated into various
2465 * chipsets). Under certain conditions a level-triggered interrupt is
2466 * erroneously delivered as edge-triggered one but the respective IRR
2467 * bit gets set nevertheless. As a result the I/O unit expects an EOI
2468 * message but it will never arrive and further interrupts are blocked
2469 * from the source. The exact reason is so far unknown, but the
2470 * phenomenon was observed when two consecutive interrupt requests
2471 * from a given source get delivered to the same CPU and the source is
2472 * temporarily disabled in between.
2474 * A workaround is to simulate an EOI message manually. We achieve it
2475 * by setting the trigger mode to edge and then to level when the edge
2476 * trigger mode gets detected in the TMR of a local APIC for a
2477 * level-triggered interrupt. We mask the source for the time of the
2478 * operation to prevent an edge-triggered interrupt escaping meanwhile.
2479 * The idea is from Manfred Spraul. --macro
2481 * Also in the case when cpu goes offline, fixup_irqs() will forward
2482 * any unhandled interrupt on the offlined cpu to the new cpu
2483 * destination that is handling the corresponding interrupt. This
2484 * interrupt forwarding is done via IPI's. Hence, in this case also
2485 * level-triggered io-apic interrupt will be seen as an edge
2486 * interrupt in the IRR. And we can't rely on the cpu's EOI
2487 * to be broadcasted to the IO-APIC's which will clear the remoteIRR
2488 * corresponding to the level-triggered interrupt. Hence on IO-APIC's
2489 * supporting EOI register, we do an explicit EOI to clear the
2490 * remote IRR and on IO-APIC's which don't have an EOI register,
2491 * we use the above logic (mask+edge followed by unmask+level) from
2492 * Manfred Spraul to clear the remote IRR.
2494 i = cfg->vector;
2495 v = apic_read(APIC_TMR + ((i & ~0x1f) >> 1));
2498 * We must acknowledge the irq before we move it or the acknowledge will
2499 * not propagate properly.
2501 ack_APIC_irq();
2504 * Tail end of clearing remote IRR bit (either by delivering the EOI
2505 * message via io-apic EOI register write or simulating it using
2506 * mask+edge followed by unnask+level logic) manually when the
2507 * level triggered interrupt is seen as the edge triggered interrupt
2508 * at the cpu.
2510 if (!(v & (1 << (i & 0x1f)))) {
2511 atomic_inc(&irq_mis_count);
2513 eoi_ioapic_irq(irq, cfg);
2516 ioapic_irqd_unmask(data, cfg, masked);
2519 static struct irq_chip ioapic_chip __read_mostly = {
2520 .name = "IO-APIC",
2521 .irq_startup = startup_ioapic_irq,
2522 .irq_mask = mask_ioapic_irq,
2523 .irq_unmask = unmask_ioapic_irq,
2524 .irq_ack = ack_apic_edge,
2525 .irq_eoi = ack_apic_level,
2526 .irq_set_affinity = native_ioapic_set_affinity,
2527 .irq_retrigger = ioapic_retrigger_irq,
2530 static inline void init_IO_APIC_traps(void)
2532 struct irq_cfg *cfg;
2533 unsigned int irq;
2536 * NOTE! The local APIC isn't very good at handling
2537 * multiple interrupts at the same interrupt level.
2538 * As the interrupt level is determined by taking the
2539 * vector number and shifting that right by 4, we
2540 * want to spread these out a bit so that they don't
2541 * all fall in the same interrupt level.
2543 * Also, we've got to be careful not to trash gate
2544 * 0x80, because int 0x80 is hm, kind of importantish. ;)
2546 for_each_active_irq(irq) {
2547 cfg = irq_get_chip_data(irq);
2548 if (IO_APIC_IRQ(irq) && cfg && !cfg->vector) {
2550 * Hmm.. We don't have an entry for this,
2551 * so default to an old-fashioned 8259
2552 * interrupt if we can..
2554 if (irq < legacy_pic->nr_legacy_irqs)
2555 legacy_pic->make_irq(irq);
2556 else
2557 /* Strange. Oh, well.. */
2558 irq_set_chip(irq, &no_irq_chip);
2564 * The local APIC irq-chip implementation:
2567 static void mask_lapic_irq(struct irq_data *data)
2569 unsigned long v;
2571 v = apic_read(APIC_LVT0);
2572 apic_write(APIC_LVT0, v | APIC_LVT_MASKED);
2575 static void unmask_lapic_irq(struct irq_data *data)
2577 unsigned long v;
2579 v = apic_read(APIC_LVT0);
2580 apic_write(APIC_LVT0, v & ~APIC_LVT_MASKED);
2583 static void ack_lapic_irq(struct irq_data *data)
2585 ack_APIC_irq();
2588 static struct irq_chip lapic_chip __read_mostly = {
2589 .name = "local-APIC",
2590 .irq_mask = mask_lapic_irq,
2591 .irq_unmask = unmask_lapic_irq,
2592 .irq_ack = ack_lapic_irq,
2595 static void lapic_register_intr(int irq)
2597 irq_clear_status_flags(irq, IRQ_LEVEL);
2598 irq_set_chip_and_handler_name(irq, &lapic_chip, handle_edge_irq,
2599 "edge");
2603 * This looks a bit hackish but it's about the only one way of sending
2604 * a few INTA cycles to 8259As and any associated glue logic. ICR does
2605 * not support the ExtINT mode, unfortunately. We need to send these
2606 * cycles as some i82489DX-based boards have glue logic that keeps the
2607 * 8259A interrupt line asserted until INTA. --macro
2609 static inline void __init unlock_ExtINT_logic(void)
2611 int apic, pin, i;
2612 struct IO_APIC_route_entry entry0, entry1;
2613 unsigned char save_control, save_freq_select;
2615 pin = find_isa_irq_pin(8, mp_INT);
2616 if (pin == -1) {
2617 WARN_ON_ONCE(1);
2618 return;
2620 apic = find_isa_irq_apic(8, mp_INT);
2621 if (apic == -1) {
2622 WARN_ON_ONCE(1);
2623 return;
2626 entry0 = ioapic_read_entry(apic, pin);
2627 clear_IO_APIC_pin(apic, pin);
2629 memset(&entry1, 0, sizeof(entry1));
2631 entry1.dest_mode = 0; /* physical delivery */
2632 entry1.mask = 0; /* unmask IRQ now */
2633 entry1.dest = hard_smp_processor_id();
2634 entry1.delivery_mode = dest_ExtINT;
2635 entry1.polarity = entry0.polarity;
2636 entry1.trigger = 0;
2637 entry1.vector = 0;
2639 ioapic_write_entry(apic, pin, entry1);
2641 save_control = CMOS_READ(RTC_CONTROL);
2642 save_freq_select = CMOS_READ(RTC_FREQ_SELECT);
2643 CMOS_WRITE((save_freq_select & ~RTC_RATE_SELECT) | 0x6,
2644 RTC_FREQ_SELECT);
2645 CMOS_WRITE(save_control | RTC_PIE, RTC_CONTROL);
2647 i = 100;
2648 while (i-- > 0) {
2649 mdelay(10);
2650 if ((CMOS_READ(RTC_INTR_FLAGS) & RTC_PF) == RTC_PF)
2651 i -= 10;
2654 CMOS_WRITE(save_control, RTC_CONTROL);
2655 CMOS_WRITE(save_freq_select, RTC_FREQ_SELECT);
2656 clear_IO_APIC_pin(apic, pin);
2658 ioapic_write_entry(apic, pin, entry0);
2661 static int disable_timer_pin_1 __initdata;
2662 /* Actually the next is obsolete, but keep it for paranoid reasons -AK */
2663 static int __init disable_timer_pin_setup(char *arg)
2665 disable_timer_pin_1 = 1;
2666 return 0;
2668 early_param("disable_timer_pin_1", disable_timer_pin_setup);
2670 int timer_through_8259 __initdata;
2673 * This code may look a bit paranoid, but it's supposed to cooperate with
2674 * a wide range of boards and BIOS bugs. Fortunately only the timer IRQ
2675 * is so screwy. Thanks to Brian Perkins for testing/hacking this beast
2676 * fanatically on his truly buggy board.
2678 * FIXME: really need to revamp this for all platforms.
2680 static inline void __init check_timer(void)
2682 struct irq_cfg *cfg = irq_get_chip_data(0);
2683 int node = cpu_to_node(0);
2684 int apic1, pin1, apic2, pin2;
2685 unsigned long flags;
2686 int no_pin1 = 0;
2688 local_irq_save(flags);
2691 * get/set the timer IRQ vector:
2693 legacy_pic->mask(0);
2694 assign_irq_vector(0, cfg, apic->target_cpus());
2697 * As IRQ0 is to be enabled in the 8259A, the virtual
2698 * wire has to be disabled in the local APIC. Also
2699 * timer interrupts need to be acknowledged manually in
2700 * the 8259A for the i82489DX when using the NMI
2701 * watchdog as that APIC treats NMIs as level-triggered.
2702 * The AEOI mode will finish them in the 8259A
2703 * automatically.
2705 apic_write(APIC_LVT0, APIC_LVT_MASKED | APIC_DM_EXTINT);
2706 legacy_pic->init(1);
2708 pin1 = find_isa_irq_pin(0, mp_INT);
2709 apic1 = find_isa_irq_apic(0, mp_INT);
2710 pin2 = ioapic_i8259.pin;
2711 apic2 = ioapic_i8259.apic;
2713 apic_printk(APIC_QUIET, KERN_INFO "..TIMER: vector=0x%02X "
2714 "apic1=%d pin1=%d apic2=%d pin2=%d\n",
2715 cfg->vector, apic1, pin1, apic2, pin2);
2718 * Some BIOS writers are clueless and report the ExtINTA
2719 * I/O APIC input from the cascaded 8259A as the timer
2720 * interrupt input. So just in case, if only one pin
2721 * was found above, try it both directly and through the
2722 * 8259A.
2724 if (pin1 == -1) {
2725 panic_if_irq_remap("BIOS bug: timer not connected to IO-APIC");
2726 pin1 = pin2;
2727 apic1 = apic2;
2728 no_pin1 = 1;
2729 } else if (pin2 == -1) {
2730 pin2 = pin1;
2731 apic2 = apic1;
2734 if (pin1 != -1) {
2736 * Ok, does IRQ0 through the IOAPIC work?
2738 if (no_pin1) {
2739 add_pin_to_irq_node(cfg, node, apic1, pin1);
2740 setup_timer_IRQ0_pin(apic1, pin1, cfg->vector);
2741 } else {
2742 /* for edge trigger, setup_ioapic_irq already
2743 * leave it unmasked.
2744 * so only need to unmask if it is level-trigger
2745 * do we really have level trigger timer?
2747 int idx;
2748 idx = find_irq_entry(apic1, pin1, mp_INT);
2749 if (idx != -1 && irq_trigger(idx))
2750 unmask_ioapic(cfg);
2752 if (timer_irq_works()) {
2753 if (disable_timer_pin_1 > 0)
2754 clear_IO_APIC_pin(0, pin1);
2755 goto out;
2757 panic_if_irq_remap("timer doesn't work through Interrupt-remapped IO-APIC");
2758 local_irq_disable();
2759 clear_IO_APIC_pin(apic1, pin1);
2760 if (!no_pin1)
2761 apic_printk(APIC_QUIET, KERN_ERR "..MP-BIOS bug: "
2762 "8254 timer not connected to IO-APIC\n");
2764 apic_printk(APIC_QUIET, KERN_INFO "...trying to set up timer "
2765 "(IRQ0) through the 8259A ...\n");
2766 apic_printk(APIC_QUIET, KERN_INFO
2767 "..... (found apic %d pin %d) ...\n", apic2, pin2);
2769 * legacy devices should be connected to IO APIC #0
2771 replace_pin_at_irq_node(cfg, node, apic1, pin1, apic2, pin2);
2772 setup_timer_IRQ0_pin(apic2, pin2, cfg->vector);
2773 legacy_pic->unmask(0);
2774 if (timer_irq_works()) {
2775 apic_printk(APIC_QUIET, KERN_INFO "....... works.\n");
2776 timer_through_8259 = 1;
2777 goto out;
2780 * Cleanup, just in case ...
2782 local_irq_disable();
2783 legacy_pic->mask(0);
2784 clear_IO_APIC_pin(apic2, pin2);
2785 apic_printk(APIC_QUIET, KERN_INFO "....... failed.\n");
2788 apic_printk(APIC_QUIET, KERN_INFO
2789 "...trying to set up timer as Virtual Wire IRQ...\n");
2791 lapic_register_intr(0);
2792 apic_write(APIC_LVT0, APIC_DM_FIXED | cfg->vector); /* Fixed mode */
2793 legacy_pic->unmask(0);
2795 if (timer_irq_works()) {
2796 apic_printk(APIC_QUIET, KERN_INFO "..... works.\n");
2797 goto out;
2799 local_irq_disable();
2800 legacy_pic->mask(0);
2801 apic_write(APIC_LVT0, APIC_LVT_MASKED | APIC_DM_FIXED | cfg->vector);
2802 apic_printk(APIC_QUIET, KERN_INFO "..... failed.\n");
2804 apic_printk(APIC_QUIET, KERN_INFO
2805 "...trying to set up timer as ExtINT IRQ...\n");
2807 legacy_pic->init(0);
2808 legacy_pic->make_irq(0);
2809 apic_write(APIC_LVT0, APIC_DM_EXTINT);
2811 unlock_ExtINT_logic();
2813 if (timer_irq_works()) {
2814 apic_printk(APIC_QUIET, KERN_INFO "..... works.\n");
2815 goto out;
2817 local_irq_disable();
2818 apic_printk(APIC_QUIET, KERN_INFO "..... failed :(.\n");
2819 if (x2apic_preenabled)
2820 apic_printk(APIC_QUIET, KERN_INFO
2821 "Perhaps problem with the pre-enabled x2apic mode\n"
2822 "Try booting with x2apic and interrupt-remapping disabled in the bios.\n");
2823 panic("IO-APIC + timer doesn't work! Boot with apic=debug and send a "
2824 "report. Then try booting with the 'noapic' option.\n");
2825 out:
2826 local_irq_restore(flags);
2830 * Traditionally ISA IRQ2 is the cascade IRQ, and is not available
2831 * to devices. However there may be an I/O APIC pin available for
2832 * this interrupt regardless. The pin may be left unconnected, but
2833 * typically it will be reused as an ExtINT cascade interrupt for
2834 * the master 8259A. In the MPS case such a pin will normally be
2835 * reported as an ExtINT interrupt in the MP table. With ACPI
2836 * there is no provision for ExtINT interrupts, and in the absence
2837 * of an override it would be treated as an ordinary ISA I/O APIC
2838 * interrupt, that is edge-triggered and unmasked by default. We
2839 * used to do this, but it caused problems on some systems because
2840 * of the NMI watchdog and sometimes IRQ0 of the 8254 timer using
2841 * the same ExtINT cascade interrupt to drive the local APIC of the
2842 * bootstrap processor. Therefore we refrain from routing IRQ2 to
2843 * the I/O APIC in all cases now. No actual device should request
2844 * it anyway. --macro
2846 #define PIC_IRQS (1UL << PIC_CASCADE_IR)
2848 void __init setup_IO_APIC(void)
2852 * calling enable_IO_APIC() is moved to setup_local_APIC for BP
2854 io_apic_irqs = legacy_pic->nr_legacy_irqs ? ~PIC_IRQS : ~0UL;
2856 apic_printk(APIC_VERBOSE, "ENABLING IO-APIC IRQs\n");
2858 * Set up IO-APIC IRQ routing.
2860 x86_init.mpparse.setup_ioapic_ids();
2862 sync_Arb_IDs();
2863 setup_IO_APIC_irqs();
2864 init_IO_APIC_traps();
2865 if (legacy_pic->nr_legacy_irqs)
2866 check_timer();
2870 * Called after all the initialization is done. If we didn't find any
2871 * APIC bugs then we can allow the modify fast path
2874 static int __init io_apic_bug_finalize(void)
2876 if (sis_apic_bug == -1)
2877 sis_apic_bug = 0;
2878 return 0;
2881 late_initcall(io_apic_bug_finalize);
2883 static void resume_ioapic_id(int ioapic_idx)
2885 unsigned long flags;
2886 union IO_APIC_reg_00 reg_00;
2888 raw_spin_lock_irqsave(&ioapic_lock, flags);
2889 reg_00.raw = io_apic_read(ioapic_idx, 0);
2890 if (reg_00.bits.ID != mpc_ioapic_id(ioapic_idx)) {
2891 reg_00.bits.ID = mpc_ioapic_id(ioapic_idx);
2892 io_apic_write(ioapic_idx, 0, reg_00.raw);
2894 raw_spin_unlock_irqrestore(&ioapic_lock, flags);
2897 static void ioapic_resume(void)
2899 int ioapic_idx;
2901 for (ioapic_idx = nr_ioapics - 1; ioapic_idx >= 0; ioapic_idx--)
2902 resume_ioapic_id(ioapic_idx);
2904 restore_ioapic_entries();
2907 static struct syscore_ops ioapic_syscore_ops = {
2908 .suspend = save_ioapic_entries,
2909 .resume = ioapic_resume,
2912 static int __init ioapic_init_ops(void)
2914 register_syscore_ops(&ioapic_syscore_ops);
2916 return 0;
2919 device_initcall(ioapic_init_ops);
2922 * Dynamic irq allocate and deallocation
2924 unsigned int __create_irqs(unsigned int from, unsigned int count, int node)
2926 struct irq_cfg **cfg;
2927 unsigned long flags;
2928 int irq, i;
2930 if (from < nr_irqs_gsi)
2931 from = nr_irqs_gsi;
2933 cfg = kzalloc_node(count * sizeof(cfg[0]), GFP_KERNEL, node);
2934 if (!cfg)
2935 return 0;
2937 irq = alloc_irqs_from(from, count, node);
2938 if (irq < 0)
2939 goto out_cfgs;
2941 for (i = 0; i < count; i++) {
2942 cfg[i] = alloc_irq_cfg(irq + i, node);
2943 if (!cfg[i])
2944 goto out_irqs;
2947 raw_spin_lock_irqsave(&vector_lock, flags);
2948 for (i = 0; i < count; i++)
2949 if (__assign_irq_vector(irq + i, cfg[i], apic->target_cpus()))
2950 goto out_vecs;
2951 raw_spin_unlock_irqrestore(&vector_lock, flags);
2953 for (i = 0; i < count; i++) {
2954 irq_set_chip_data(irq + i, cfg[i]);
2955 irq_clear_status_flags(irq + i, IRQ_NOREQUEST);
2958 kfree(cfg);
2959 return irq;
2961 out_vecs:
2962 for (i--; i >= 0; i--)
2963 __clear_irq_vector(irq + i, cfg[i]);
2964 raw_spin_unlock_irqrestore(&vector_lock, flags);
2965 out_irqs:
2966 for (i = 0; i < count; i++)
2967 free_irq_at(irq + i, cfg[i]);
2968 out_cfgs:
2969 kfree(cfg);
2970 return 0;
2973 unsigned int create_irq_nr(unsigned int from, int node)
2975 return __create_irqs(from, 1, node);
2978 int create_irq(void)
2980 int node = cpu_to_node(0);
2981 unsigned int irq_want;
2982 int irq;
2984 irq_want = nr_irqs_gsi;
2985 irq = create_irq_nr(irq_want, node);
2987 if (irq == 0)
2988 irq = -1;
2990 return irq;
2993 void destroy_irq(unsigned int irq)
2995 struct irq_cfg *cfg = irq_get_chip_data(irq);
2996 unsigned long flags;
2998 irq_set_status_flags(irq, IRQ_NOREQUEST|IRQ_NOPROBE);
3000 free_remapped_irq(irq);
3002 raw_spin_lock_irqsave(&vector_lock, flags);
3003 __clear_irq_vector(irq, cfg);
3004 raw_spin_unlock_irqrestore(&vector_lock, flags);
3005 free_irq_at(irq, cfg);
3008 void destroy_irqs(unsigned int irq, unsigned int count)
3010 unsigned int i;
3012 for (i = 0; i < count; i++)
3013 destroy_irq(irq + i);
3017 * MSI message composition
3019 void native_compose_msi_msg(struct pci_dev *pdev,
3020 unsigned int irq, unsigned int dest,
3021 struct msi_msg *msg, u8 hpet_id)
3023 struct irq_cfg *cfg = irq_cfg(irq);
3025 msg->address_hi = MSI_ADDR_BASE_HI;
3027 if (x2apic_enabled())
3028 msg->address_hi |= MSI_ADDR_EXT_DEST_ID(dest);
3030 msg->address_lo =
3031 MSI_ADDR_BASE_LO |
3032 ((apic->irq_dest_mode == 0) ?
3033 MSI_ADDR_DEST_MODE_PHYSICAL:
3034 MSI_ADDR_DEST_MODE_LOGICAL) |
3035 ((apic->irq_delivery_mode != dest_LowestPrio) ?
3036 MSI_ADDR_REDIRECTION_CPU:
3037 MSI_ADDR_REDIRECTION_LOWPRI) |
3038 MSI_ADDR_DEST_ID(dest);
3040 msg->data =
3041 MSI_DATA_TRIGGER_EDGE |
3042 MSI_DATA_LEVEL_ASSERT |
3043 ((apic->irq_delivery_mode != dest_LowestPrio) ?
3044 MSI_DATA_DELIVERY_FIXED:
3045 MSI_DATA_DELIVERY_LOWPRI) |
3046 MSI_DATA_VECTOR(cfg->vector);
3049 #ifdef CONFIG_PCI_MSI
3050 static int msi_compose_msg(struct pci_dev *pdev, unsigned int irq,
3051 struct msi_msg *msg, u8 hpet_id)
3053 struct irq_cfg *cfg;
3054 int err;
3055 unsigned dest;
3057 if (disable_apic)
3058 return -ENXIO;
3060 cfg = irq_cfg(irq);
3061 err = assign_irq_vector(irq, cfg, apic->target_cpus());
3062 if (err)
3063 return err;
3065 err = apic->cpu_mask_to_apicid_and(cfg->domain,
3066 apic->target_cpus(), &dest);
3067 if (err)
3068 return err;
3070 x86_msi.compose_msi_msg(pdev, irq, dest, msg, hpet_id);
3072 return 0;
3075 static int
3076 msi_set_affinity(struct irq_data *data, const struct cpumask *mask, bool force)
3078 struct irq_cfg *cfg = data->chip_data;
3079 struct msi_msg msg;
3080 unsigned int dest;
3082 if (__ioapic_set_affinity(data, mask, &dest))
3083 return -1;
3085 __get_cached_msi_msg(data->msi_desc, &msg);
3087 msg.data &= ~MSI_DATA_VECTOR_MASK;
3088 msg.data |= MSI_DATA_VECTOR(cfg->vector);
3089 msg.address_lo &= ~MSI_ADDR_DEST_ID_MASK;
3090 msg.address_lo |= MSI_ADDR_DEST_ID(dest);
3092 __write_msi_msg(data->msi_desc, &msg);
3094 return IRQ_SET_MASK_OK_NOCOPY;
3098 * IRQ Chip for MSI PCI/PCI-X/PCI-Express Devices,
3099 * which implement the MSI or MSI-X Capability Structure.
3101 static struct irq_chip msi_chip = {
3102 .name = "PCI-MSI",
3103 .irq_unmask = unmask_msi_irq,
3104 .irq_mask = mask_msi_irq,
3105 .irq_ack = ack_apic_edge,
3106 .irq_set_affinity = msi_set_affinity,
3107 .irq_retrigger = ioapic_retrigger_irq,
3110 int setup_msi_irq(struct pci_dev *dev, struct msi_desc *msidesc,
3111 unsigned int irq_base, unsigned int irq_offset)
3113 struct irq_chip *chip = &msi_chip;
3114 struct msi_msg msg;
3115 unsigned int irq = irq_base + irq_offset;
3116 int ret;
3118 ret = msi_compose_msg(dev, irq, &msg, -1);
3119 if (ret < 0)
3120 return ret;
3122 irq_set_msi_desc_off(irq_base, irq_offset, msidesc);
3125 * MSI-X message is written per-IRQ, the offset is always 0.
3126 * MSI message denotes a contiguous group of IRQs, written for 0th IRQ.
3128 if (!irq_offset)
3129 write_msi_msg(irq, &msg);
3131 setup_remapped_irq(irq, irq_get_chip_data(irq), chip);
3133 irq_set_chip_and_handler_name(irq, chip, handle_edge_irq, "edge");
3135 dev_printk(KERN_DEBUG, &dev->dev, "irq %d for MSI/MSI-X\n", irq);
3137 return 0;
3140 int native_setup_msi_irqs(struct pci_dev *dev, int nvec, int type)
3142 unsigned int irq, irq_want;
3143 struct msi_desc *msidesc;
3144 int node, ret;
3146 /* Multiple MSI vectors only supported with interrupt remapping */
3147 if (type == PCI_CAP_ID_MSI && nvec > 1)
3148 return 1;
3150 node = dev_to_node(&dev->dev);
3151 irq_want = nr_irqs_gsi;
3152 list_for_each_entry(msidesc, &dev->msi_list, list) {
3153 irq = create_irq_nr(irq_want, node);
3154 if (irq == 0)
3155 return -ENOSPC;
3157 irq_want = irq + 1;
3159 ret = setup_msi_irq(dev, msidesc, irq, 0);
3160 if (ret < 0)
3161 goto error;
3163 return 0;
3165 error:
3166 destroy_irq(irq);
3167 return ret;
3170 void native_teardown_msi_irq(unsigned int irq)
3172 destroy_irq(irq);
3175 #ifdef CONFIG_DMAR_TABLE
3176 static int
3177 dmar_msi_set_affinity(struct irq_data *data, const struct cpumask *mask,
3178 bool force)
3180 struct irq_cfg *cfg = data->chip_data;
3181 unsigned int dest, irq = data->irq;
3182 struct msi_msg msg;
3184 if (__ioapic_set_affinity(data, mask, &dest))
3185 return -1;
3187 dmar_msi_read(irq, &msg);
3189 msg.data &= ~MSI_DATA_VECTOR_MASK;
3190 msg.data |= MSI_DATA_VECTOR(cfg->vector);
3191 msg.address_lo &= ~MSI_ADDR_DEST_ID_MASK;
3192 msg.address_lo |= MSI_ADDR_DEST_ID(dest);
3193 msg.address_hi = MSI_ADDR_BASE_HI | MSI_ADDR_EXT_DEST_ID(dest);
3195 dmar_msi_write(irq, &msg);
3197 return IRQ_SET_MASK_OK_NOCOPY;
3200 static struct irq_chip dmar_msi_type = {
3201 .name = "DMAR_MSI",
3202 .irq_unmask = dmar_msi_unmask,
3203 .irq_mask = dmar_msi_mask,
3204 .irq_ack = ack_apic_edge,
3205 .irq_set_affinity = dmar_msi_set_affinity,
3206 .irq_retrigger = ioapic_retrigger_irq,
3209 int arch_setup_dmar_msi(unsigned int irq)
3211 int ret;
3212 struct msi_msg msg;
3214 ret = msi_compose_msg(NULL, irq, &msg, -1);
3215 if (ret < 0)
3216 return ret;
3217 dmar_msi_write(irq, &msg);
3218 irq_set_chip_and_handler_name(irq, &dmar_msi_type, handle_edge_irq,
3219 "edge");
3220 return 0;
3222 #endif
3224 #ifdef CONFIG_HPET_TIMER
3226 static int hpet_msi_set_affinity(struct irq_data *data,
3227 const struct cpumask *mask, bool force)
3229 struct irq_cfg *cfg = data->chip_data;
3230 struct msi_msg msg;
3231 unsigned int dest;
3233 if (__ioapic_set_affinity(data, mask, &dest))
3234 return -1;
3236 hpet_msi_read(data->handler_data, &msg);
3238 msg.data &= ~MSI_DATA_VECTOR_MASK;
3239 msg.data |= MSI_DATA_VECTOR(cfg->vector);
3240 msg.address_lo &= ~MSI_ADDR_DEST_ID_MASK;
3241 msg.address_lo |= MSI_ADDR_DEST_ID(dest);
3243 hpet_msi_write(data->handler_data, &msg);
3245 return IRQ_SET_MASK_OK_NOCOPY;
3248 static struct irq_chip hpet_msi_type = {
3249 .name = "HPET_MSI",
3250 .irq_unmask = hpet_msi_unmask,
3251 .irq_mask = hpet_msi_mask,
3252 .irq_ack = ack_apic_edge,
3253 .irq_set_affinity = hpet_msi_set_affinity,
3254 .irq_retrigger = ioapic_retrigger_irq,
3257 int default_setup_hpet_msi(unsigned int irq, unsigned int id)
3259 struct irq_chip *chip = &hpet_msi_type;
3260 struct msi_msg msg;
3261 int ret;
3263 ret = msi_compose_msg(NULL, irq, &msg, id);
3264 if (ret < 0)
3265 return ret;
3267 hpet_msi_write(irq_get_handler_data(irq), &msg);
3268 irq_set_status_flags(irq, IRQ_MOVE_PCNTXT);
3269 setup_remapped_irq(irq, irq_get_chip_data(irq), chip);
3271 irq_set_chip_and_handler_name(irq, chip, handle_edge_irq, "edge");
3272 return 0;
3274 #endif
3276 #endif /* CONFIG_PCI_MSI */
3278 * Hypertransport interrupt support
3280 #ifdef CONFIG_HT_IRQ
3282 static void target_ht_irq(unsigned int irq, unsigned int dest, u8 vector)
3284 struct ht_irq_msg msg;
3285 fetch_ht_irq_msg(irq, &msg);
3287 msg.address_lo &= ~(HT_IRQ_LOW_VECTOR_MASK | HT_IRQ_LOW_DEST_ID_MASK);
3288 msg.address_hi &= ~(HT_IRQ_HIGH_DEST_ID_MASK);
3290 msg.address_lo |= HT_IRQ_LOW_VECTOR(vector) | HT_IRQ_LOW_DEST_ID(dest);
3291 msg.address_hi |= HT_IRQ_HIGH_DEST_ID(dest);
3293 write_ht_irq_msg(irq, &msg);
3296 static int
3297 ht_set_affinity(struct irq_data *data, const struct cpumask *mask, bool force)
3299 struct irq_cfg *cfg = data->chip_data;
3300 unsigned int dest;
3302 if (__ioapic_set_affinity(data, mask, &dest))
3303 return -1;
3305 target_ht_irq(data->irq, dest, cfg->vector);
3306 return IRQ_SET_MASK_OK_NOCOPY;
3309 static struct irq_chip ht_irq_chip = {
3310 .name = "PCI-HT",
3311 .irq_mask = mask_ht_irq,
3312 .irq_unmask = unmask_ht_irq,
3313 .irq_ack = ack_apic_edge,
3314 .irq_set_affinity = ht_set_affinity,
3315 .irq_retrigger = ioapic_retrigger_irq,
3318 int arch_setup_ht_irq(unsigned int irq, struct pci_dev *dev)
3320 struct irq_cfg *cfg;
3321 struct ht_irq_msg msg;
3322 unsigned dest;
3323 int err;
3325 if (disable_apic)
3326 return -ENXIO;
3328 cfg = irq_cfg(irq);
3329 err = assign_irq_vector(irq, cfg, apic->target_cpus());
3330 if (err)
3331 return err;
3333 err = apic->cpu_mask_to_apicid_and(cfg->domain,
3334 apic->target_cpus(), &dest);
3335 if (err)
3336 return err;
3338 msg.address_hi = HT_IRQ_HIGH_DEST_ID(dest);
3340 msg.address_lo =
3341 HT_IRQ_LOW_BASE |
3342 HT_IRQ_LOW_DEST_ID(dest) |
3343 HT_IRQ_LOW_VECTOR(cfg->vector) |
3344 ((apic->irq_dest_mode == 0) ?
3345 HT_IRQ_LOW_DM_PHYSICAL :
3346 HT_IRQ_LOW_DM_LOGICAL) |
3347 HT_IRQ_LOW_RQEOI_EDGE |
3348 ((apic->irq_delivery_mode != dest_LowestPrio) ?
3349 HT_IRQ_LOW_MT_FIXED :
3350 HT_IRQ_LOW_MT_ARBITRATED) |
3351 HT_IRQ_LOW_IRQ_MASKED;
3353 write_ht_irq_msg(irq, &msg);
3355 irq_set_chip_and_handler_name(irq, &ht_irq_chip,
3356 handle_edge_irq, "edge");
3358 dev_printk(KERN_DEBUG, &dev->dev, "irq %d for HT\n", irq);
3360 return 0;
3362 #endif /* CONFIG_HT_IRQ */
3364 static int
3365 io_apic_setup_irq_pin(unsigned int irq, int node, struct io_apic_irq_attr *attr)
3367 struct irq_cfg *cfg = alloc_irq_and_cfg_at(irq, node);
3368 int ret;
3370 if (!cfg)
3371 return -EINVAL;
3372 ret = __add_pin_to_irq_node(cfg, node, attr->ioapic, attr->ioapic_pin);
3373 if (!ret)
3374 setup_ioapic_irq(irq, cfg, attr);
3375 return ret;
3378 int io_apic_setup_irq_pin_once(unsigned int irq, int node,
3379 struct io_apic_irq_attr *attr)
3381 unsigned int ioapic_idx = attr->ioapic, pin = attr->ioapic_pin;
3382 int ret;
3383 struct IO_APIC_route_entry orig_entry;
3385 /* Avoid redundant programming */
3386 if (test_bit(pin, ioapics[ioapic_idx].pin_programmed)) {
3387 pr_debug("Pin %d-%d already programmed\n", mpc_ioapic_id(ioapic_idx), pin);
3388 orig_entry = ioapic_read_entry(attr->ioapic, pin);
3389 if (attr->trigger == orig_entry.trigger && attr->polarity == orig_entry.polarity)
3390 return 0;
3391 return -EBUSY;
3393 ret = io_apic_setup_irq_pin(irq, node, attr);
3394 if (!ret)
3395 set_bit(pin, ioapics[ioapic_idx].pin_programmed);
3396 return ret;
3399 static int __init io_apic_get_redir_entries(int ioapic)
3401 union IO_APIC_reg_01 reg_01;
3402 unsigned long flags;
3404 raw_spin_lock_irqsave(&ioapic_lock, flags);
3405 reg_01.raw = io_apic_read(ioapic, 1);
3406 raw_spin_unlock_irqrestore(&ioapic_lock, flags);
3408 /* The register returns the maximum index redir index
3409 * supported, which is one less than the total number of redir
3410 * entries.
3412 return reg_01.bits.entries + 1;
3415 static void __init probe_nr_irqs_gsi(void)
3417 int nr;
3419 nr = gsi_top + NR_IRQS_LEGACY;
3420 if (nr > nr_irqs_gsi)
3421 nr_irqs_gsi = nr;
3423 printk(KERN_DEBUG "nr_irqs_gsi: %d\n", nr_irqs_gsi);
3426 int get_nr_irqs_gsi(void)
3428 return nr_irqs_gsi;
3431 int __init arch_probe_nr_irqs(void)
3433 int nr;
3435 if (nr_irqs > (NR_VECTORS * nr_cpu_ids))
3436 nr_irqs = NR_VECTORS * nr_cpu_ids;
3438 nr = nr_irqs_gsi + 8 * nr_cpu_ids;
3439 #if defined(CONFIG_PCI_MSI) || defined(CONFIG_HT_IRQ)
3441 * for MSI and HT dyn irq
3443 nr += nr_irqs_gsi * 16;
3444 #endif
3445 if (nr < nr_irqs)
3446 nr_irqs = nr;
3448 return NR_IRQS_LEGACY;
3451 int io_apic_set_pci_routing(struct device *dev, int irq,
3452 struct io_apic_irq_attr *irq_attr)
3454 int node;
3456 if (!IO_APIC_IRQ(irq)) {
3457 apic_printk(APIC_QUIET,KERN_ERR "IOAPIC[%d]: Invalid reference to IRQ 0\n",
3458 irq_attr->ioapic);
3459 return -EINVAL;
3462 node = dev ? dev_to_node(dev) : cpu_to_node(0);
3464 return io_apic_setup_irq_pin_once(irq, node, irq_attr);
3467 #ifdef CONFIG_X86_32
3468 static int __init io_apic_get_unique_id(int ioapic, int apic_id)
3470 union IO_APIC_reg_00 reg_00;
3471 static physid_mask_t apic_id_map = PHYSID_MASK_NONE;
3472 physid_mask_t tmp;
3473 unsigned long flags;
3474 int i = 0;
3477 * The P4 platform supports up to 256 APIC IDs on two separate APIC
3478 * buses (one for LAPICs, one for IOAPICs), where predecessors only
3479 * supports up to 16 on one shared APIC bus.
3481 * TBD: Expand LAPIC/IOAPIC support on P4-class systems to take full
3482 * advantage of new APIC bus architecture.
3485 if (physids_empty(apic_id_map))
3486 apic->ioapic_phys_id_map(&phys_cpu_present_map, &apic_id_map);
3488 raw_spin_lock_irqsave(&ioapic_lock, flags);
3489 reg_00.raw = io_apic_read(ioapic, 0);
3490 raw_spin_unlock_irqrestore(&ioapic_lock, flags);
3492 if (apic_id >= get_physical_broadcast()) {
3493 printk(KERN_WARNING "IOAPIC[%d]: Invalid apic_id %d, trying "
3494 "%d\n", ioapic, apic_id, reg_00.bits.ID);
3495 apic_id = reg_00.bits.ID;
3499 * Every APIC in a system must have a unique ID or we get lots of nice
3500 * 'stuck on smp_invalidate_needed IPI wait' messages.
3502 if (apic->check_apicid_used(&apic_id_map, apic_id)) {
3504 for (i = 0; i < get_physical_broadcast(); i++) {
3505 if (!apic->check_apicid_used(&apic_id_map, i))
3506 break;
3509 if (i == get_physical_broadcast())
3510 panic("Max apic_id exceeded!\n");
3512 printk(KERN_WARNING "IOAPIC[%d]: apic_id %d already used, "
3513 "trying %d\n", ioapic, apic_id, i);
3515 apic_id = i;
3518 apic->apicid_to_cpu_present(apic_id, &tmp);
3519 physids_or(apic_id_map, apic_id_map, tmp);
3521 if (reg_00.bits.ID != apic_id) {
3522 reg_00.bits.ID = apic_id;
3524 raw_spin_lock_irqsave(&ioapic_lock, flags);
3525 io_apic_write(ioapic, 0, reg_00.raw);
3526 reg_00.raw = io_apic_read(ioapic, 0);
3527 raw_spin_unlock_irqrestore(&ioapic_lock, flags);
3529 /* Sanity check */
3530 if (reg_00.bits.ID != apic_id) {
3531 pr_err("IOAPIC[%d]: Unable to change apic_id!\n",
3532 ioapic);
3533 return -1;
3537 apic_printk(APIC_VERBOSE, KERN_INFO
3538 "IOAPIC[%d]: Assigned apic_id %d\n", ioapic, apic_id);
3540 return apic_id;
3543 static u8 __init io_apic_unique_id(u8 id)
3545 if ((boot_cpu_data.x86_vendor == X86_VENDOR_INTEL) &&
3546 !APIC_XAPIC(apic_version[boot_cpu_physical_apicid]))
3547 return io_apic_get_unique_id(nr_ioapics, id);
3548 else
3549 return id;
3551 #else
3552 static u8 __init io_apic_unique_id(u8 id)
3554 int i;
3555 DECLARE_BITMAP(used, 256);
3557 bitmap_zero(used, 256);
3558 for (i = 0; i < nr_ioapics; i++) {
3559 __set_bit(mpc_ioapic_id(i), used);
3561 if (!test_bit(id, used))
3562 return id;
3563 return find_first_zero_bit(used, 256);
3565 #endif
3567 static int __init io_apic_get_version(int ioapic)
3569 union IO_APIC_reg_01 reg_01;
3570 unsigned long flags;
3572 raw_spin_lock_irqsave(&ioapic_lock, flags);
3573 reg_01.raw = io_apic_read(ioapic, 1);
3574 raw_spin_unlock_irqrestore(&ioapic_lock, flags);
3576 return reg_01.bits.version;
3579 int acpi_get_override_irq(u32 gsi, int *trigger, int *polarity)
3581 int ioapic, pin, idx;
3583 if (skip_ioapic_setup)
3584 return -1;
3586 ioapic = mp_find_ioapic(gsi);
3587 if (ioapic < 0)
3588 return -1;
3590 pin = mp_find_ioapic_pin(ioapic, gsi);
3591 if (pin < 0)
3592 return -1;
3594 idx = find_irq_entry(ioapic, pin, mp_INT);
3595 if (idx < 0)
3596 return -1;
3598 *trigger = irq_trigger(idx);
3599 *polarity = irq_polarity(idx);
3600 return 0;
3604 * This function currently is only a helper for the i386 smp boot process where
3605 * we need to reprogram the ioredtbls to cater for the cpus which have come online
3606 * so mask in all cases should simply be apic->target_cpus()
3608 #ifdef CONFIG_SMP
3609 void __init setup_ioapic_dest(void)
3611 int pin, ioapic, irq, irq_entry;
3612 const struct cpumask *mask;
3613 struct irq_data *idata;
3615 if (skip_ioapic_setup == 1)
3616 return;
3618 for (ioapic = 0; ioapic < nr_ioapics; ioapic++)
3619 for (pin = 0; pin < ioapics[ioapic].nr_registers; pin++) {
3620 irq_entry = find_irq_entry(ioapic, pin, mp_INT);
3621 if (irq_entry == -1)
3622 continue;
3623 irq = pin_2_irq(irq_entry, ioapic, pin);
3625 if ((ioapic > 0) && (irq > 16))
3626 continue;
3628 idata = irq_get_irq_data(irq);
3631 * Honour affinities which have been set in early boot
3633 if (!irqd_can_balance(idata) || irqd_affinity_was_set(idata))
3634 mask = idata->affinity;
3635 else
3636 mask = apic->target_cpus();
3638 x86_io_apic_ops.set_affinity(idata, mask, false);
3642 #endif
3644 #define IOAPIC_RESOURCE_NAME_SIZE 11
3646 static struct resource *ioapic_resources;
3648 static struct resource * __init ioapic_setup_resources(int nr_ioapics)
3650 unsigned long n;
3651 struct resource *res;
3652 char *mem;
3653 int i;
3655 if (nr_ioapics <= 0)
3656 return NULL;
3658 n = IOAPIC_RESOURCE_NAME_SIZE + sizeof(struct resource);
3659 n *= nr_ioapics;
3661 mem = alloc_bootmem(n);
3662 res = (void *)mem;
3664 mem += sizeof(struct resource) * nr_ioapics;
3666 for (i = 0; i < nr_ioapics; i++) {
3667 res[i].name = mem;
3668 res[i].flags = IORESOURCE_MEM | IORESOURCE_BUSY;
3669 snprintf(mem, IOAPIC_RESOURCE_NAME_SIZE, "IOAPIC %u", i);
3670 mem += IOAPIC_RESOURCE_NAME_SIZE;
3673 ioapic_resources = res;
3675 return res;
3678 void __init native_io_apic_init_mappings(void)
3680 unsigned long ioapic_phys, idx = FIX_IO_APIC_BASE_0;
3681 struct resource *ioapic_res;
3682 int i;
3684 ioapic_res = ioapic_setup_resources(nr_ioapics);
3685 for (i = 0; i < nr_ioapics; i++) {
3686 if (smp_found_config) {
3687 ioapic_phys = mpc_ioapic_addr(i);
3688 #ifdef CONFIG_X86_32
3689 if (!ioapic_phys) {
3690 printk(KERN_ERR
3691 "WARNING: bogus zero IO-APIC "
3692 "address found in MPTABLE, "
3693 "disabling IO/APIC support!\n");
3694 smp_found_config = 0;
3695 skip_ioapic_setup = 1;
3696 goto fake_ioapic_page;
3698 #endif
3699 } else {
3700 #ifdef CONFIG_X86_32
3701 fake_ioapic_page:
3702 #endif
3703 ioapic_phys = (unsigned long)alloc_bootmem_pages(PAGE_SIZE);
3704 ioapic_phys = __pa(ioapic_phys);
3706 set_fixmap_nocache(idx, ioapic_phys);
3707 apic_printk(APIC_VERBOSE, "mapped IOAPIC to %08lx (%08lx)\n",
3708 __fix_to_virt(idx) + (ioapic_phys & ~PAGE_MASK),
3709 ioapic_phys);
3710 idx++;
3712 ioapic_res->start = ioapic_phys;
3713 ioapic_res->end = ioapic_phys + IO_APIC_SLOT_SIZE - 1;
3714 ioapic_res++;
3717 probe_nr_irqs_gsi();
3720 void __init ioapic_insert_resources(void)
3722 int i;
3723 struct resource *r = ioapic_resources;
3725 if (!r) {
3726 if (nr_ioapics > 0)
3727 printk(KERN_ERR
3728 "IO APIC resources couldn't be allocated.\n");
3729 return;
3732 for (i = 0; i < nr_ioapics; i++) {
3733 insert_resource(&iomem_resource, r);
3734 r++;
3738 int mp_find_ioapic(u32 gsi)
3740 int i = 0;
3742 if (nr_ioapics == 0)
3743 return -1;
3745 /* Find the IOAPIC that manages this GSI. */
3746 for (i = 0; i < nr_ioapics; i++) {
3747 struct mp_ioapic_gsi *gsi_cfg = mp_ioapic_gsi_routing(i);
3748 if ((gsi >= gsi_cfg->gsi_base)
3749 && (gsi <= gsi_cfg->gsi_end))
3750 return i;
3753 printk(KERN_ERR "ERROR: Unable to locate IOAPIC for GSI %d\n", gsi);
3754 return -1;
3757 int mp_find_ioapic_pin(int ioapic, u32 gsi)
3759 struct mp_ioapic_gsi *gsi_cfg;
3761 if (WARN_ON(ioapic == -1))
3762 return -1;
3764 gsi_cfg = mp_ioapic_gsi_routing(ioapic);
3765 if (WARN_ON(gsi > gsi_cfg->gsi_end))
3766 return -1;
3768 return gsi - gsi_cfg->gsi_base;
3771 static __init int bad_ioapic(unsigned long address)
3773 if (nr_ioapics >= MAX_IO_APICS) {
3774 pr_warn("WARNING: Max # of I/O APICs (%d) exceeded (found %d), skipping\n",
3775 MAX_IO_APICS, nr_ioapics);
3776 return 1;
3778 if (!address) {
3779 pr_warn("WARNING: Bogus (zero) I/O APIC address found in table, skipping!\n");
3780 return 1;
3782 return 0;
3785 static __init int bad_ioapic_register(int idx)
3787 union IO_APIC_reg_00 reg_00;
3788 union IO_APIC_reg_01 reg_01;
3789 union IO_APIC_reg_02 reg_02;
3791 reg_00.raw = io_apic_read(idx, 0);
3792 reg_01.raw = io_apic_read(idx, 1);
3793 reg_02.raw = io_apic_read(idx, 2);
3795 if (reg_00.raw == -1 && reg_01.raw == -1 && reg_02.raw == -1) {
3796 pr_warn("I/O APIC 0x%x registers return all ones, skipping!\n",
3797 mpc_ioapic_addr(idx));
3798 return 1;
3801 return 0;
3804 void __init mp_register_ioapic(int id, u32 address, u32 gsi_base)
3806 int idx = 0;
3807 int entries;
3808 struct mp_ioapic_gsi *gsi_cfg;
3810 if (bad_ioapic(address))
3811 return;
3813 idx = nr_ioapics;
3815 ioapics[idx].mp_config.type = MP_IOAPIC;
3816 ioapics[idx].mp_config.flags = MPC_APIC_USABLE;
3817 ioapics[idx].mp_config.apicaddr = address;
3819 set_fixmap_nocache(FIX_IO_APIC_BASE_0 + idx, address);
3821 if (bad_ioapic_register(idx)) {
3822 clear_fixmap(FIX_IO_APIC_BASE_0 + idx);
3823 return;
3826 ioapics[idx].mp_config.apicid = io_apic_unique_id(id);
3827 ioapics[idx].mp_config.apicver = io_apic_get_version(idx);
3830 * Build basic GSI lookup table to facilitate gsi->io_apic lookups
3831 * and to prevent reprogramming of IOAPIC pins (PCI GSIs).
3833 entries = io_apic_get_redir_entries(idx);
3834 gsi_cfg = mp_ioapic_gsi_routing(idx);
3835 gsi_cfg->gsi_base = gsi_base;
3836 gsi_cfg->gsi_end = gsi_base + entries - 1;
3839 * The number of IO-APIC IRQ registers (== #pins):
3841 ioapics[idx].nr_registers = entries;
3843 if (gsi_cfg->gsi_end >= gsi_top)
3844 gsi_top = gsi_cfg->gsi_end + 1;
3846 pr_info("IOAPIC[%d]: apic_id %d, version %d, address 0x%x, GSI %d-%d\n",
3847 idx, mpc_ioapic_id(idx),
3848 mpc_ioapic_ver(idx), mpc_ioapic_addr(idx),
3849 gsi_cfg->gsi_base, gsi_cfg->gsi_end);
3851 nr_ioapics++;
3854 /* Enable IOAPIC early just for system timer */
3855 void __init pre_init_apic_IRQ0(void)
3857 struct io_apic_irq_attr attr = { 0, 0, 0, 0 };
3859 printk(KERN_INFO "Early APIC setup for system timer0\n");
3860 #ifndef CONFIG_SMP
3861 physid_set_mask_of_physid(boot_cpu_physical_apicid,
3862 &phys_cpu_present_map);
3863 #endif
3864 setup_local_APIC();
3866 io_apic_setup_irq_pin(0, 0, &attr);
3867 irq_set_chip_and_handler_name(0, &ioapic_chip, handle_edge_irq,
3868 "edge");