Loading...
1// SPDX-License-Identifier: GPL-2.0
2/*
3 * Intel IO-APIC support for multi-Pentium hosts.
4 *
5 * Copyright (C) 1997, 1998, 1999, 2000, 2009 Ingo Molnar, Hajnalka Szabo
6 *
7 * Many thanks to Stig Venaas for trying out countless experimental
8 * patches and reporting/debugging problems patiently!
9 *
10 * (c) 1999, Multiple IO-APIC support, developed by
11 * Ken-ichi Yaku <yaku@css1.kbnes.nec.co.jp> and
12 * Hidemi Kishimoto <kisimoto@css1.kbnes.nec.co.jp>,
13 * further tested and cleaned up by Zach Brown <zab@redhat.com>
14 * and Ingo Molnar <mingo@redhat.com>
15 *
16 * Fixes
17 * Maciej W. Rozycki : Bits for genuine 82489DX APICs;
18 * thanks to Eric Gilmore
19 * and Rolf G. Tews
20 * for testing these extensively
21 * Paul Diefenbaugh : Added full ACPI support
22 *
23 * Historical information which is worth to be preserved:
24 *
25 * - SiS APIC rmw bug:
26 *
27 * We used to have a workaround for a bug in SiS chips which
28 * required to rewrite the index register for a read-modify-write
29 * operation as the chip lost the index information which was
30 * setup for the read already. We cache the data now, so that
31 * workaround has been removed.
32 */
33
34#include <linux/mm.h>
35#include <linux/interrupt.h>
36#include <linux/irq.h>
37#include <linux/init.h>
38#include <linux/delay.h>
39#include <linux/sched.h>
40#include <linux/pci.h>
41#include <linux/mc146818rtc.h>
42#include <linux/compiler.h>
43#include <linux/acpi.h>
44#include <linux/export.h>
45#include <linux/syscore_ops.h>
46#include <linux/freezer.h>
47#include <linux/kthread.h>
48#include <linux/jiffies.h> /* time_after() */
49#include <linux/slab.h>
50#include <linux/memblock.h>
51#include <linux/msi.h>
52
53#include <asm/irqdomain.h>
54#include <asm/io.h>
55#include <asm/smp.h>
56#include <asm/cpu.h>
57#include <asm/desc.h>
58#include <asm/proto.h>
59#include <asm/acpi.h>
60#include <asm/dma.h>
61#include <asm/timer.h>
62#include <asm/time.h>
63#include <asm/i8259.h>
64#include <asm/setup.h>
65#include <asm/irq_remapping.h>
66#include <asm/hw_irq.h>
67#include <asm/apic.h>
68#include <asm/pgtable.h>
69
70#define for_each_ioapic(idx) \
71 for ((idx) = 0; (idx) < nr_ioapics; (idx)++)
72#define for_each_ioapic_reverse(idx) \
73 for ((idx) = nr_ioapics - 1; (idx) >= 0; (idx)--)
74#define for_each_pin(idx, pin) \
75 for ((pin) = 0; (pin) < ioapics[(idx)].nr_registers; (pin)++)
76#define for_each_ioapic_pin(idx, pin) \
77 for_each_ioapic((idx)) \
78 for_each_pin((idx), (pin))
79#define for_each_irq_pin(entry, head) \
80 list_for_each_entry(entry, &head, list)
81
82static DEFINE_RAW_SPINLOCK(ioapic_lock);
83static DEFINE_MUTEX(ioapic_mutex);
84static unsigned int ioapic_dynirq_base;
85static int ioapic_initialized;
86
87struct irq_pin_list {
88 struct list_head list;
89 int apic, pin;
90};
91
92struct mp_chip_data {
93 struct list_head irq_2_pin;
94 struct IO_APIC_route_entry entry;
95 bool is_level;
96 bool active_low;
97 bool isa_irq;
98 u32 count;
99};
100
101struct mp_ioapic_gsi {
102 u32 gsi_base;
103 u32 gsi_end;
104};
105
106static struct ioapic {
107 /*
108 * # of IRQ routing registers
109 */
110 int nr_registers;
111 /*
112 * Saved state during suspend/resume, or while enabling intr-remap.
113 */
114 struct IO_APIC_route_entry *saved_registers;
115 /* I/O APIC config */
116 struct mpc_ioapic mp_config;
117 /* IO APIC gsi routing info */
118 struct mp_ioapic_gsi gsi_config;
119 struct ioapic_domain_cfg irqdomain_cfg;
120 struct irq_domain *irqdomain;
121 struct resource *iomem_res;
122} ioapics[MAX_IO_APICS];
123
124#define mpc_ioapic_ver(ioapic_idx) ioapics[ioapic_idx].mp_config.apicver
125
126int mpc_ioapic_id(int ioapic_idx)
127{
128 return ioapics[ioapic_idx].mp_config.apicid;
129}
130
131unsigned int mpc_ioapic_addr(int ioapic_idx)
132{
133 return ioapics[ioapic_idx].mp_config.apicaddr;
134}
135
136static inline struct mp_ioapic_gsi *mp_ioapic_gsi_routing(int ioapic_idx)
137{
138 return &ioapics[ioapic_idx].gsi_config;
139}
140
141static inline int mp_ioapic_pin_count(int ioapic)
142{
143 struct mp_ioapic_gsi *gsi_cfg = mp_ioapic_gsi_routing(ioapic);
144
145 return gsi_cfg->gsi_end - gsi_cfg->gsi_base + 1;
146}
147
148static inline u32 mp_pin_to_gsi(int ioapic, int pin)
149{
150 return mp_ioapic_gsi_routing(ioapic)->gsi_base + pin;
151}
152
153static inline bool mp_is_legacy_irq(int irq)
154{
155 return irq >= 0 && irq < nr_legacy_irqs();
156}
157
158static inline struct irq_domain *mp_ioapic_irqdomain(int ioapic)
159{
160 return ioapics[ioapic].irqdomain;
161}
162
163int nr_ioapics;
164
165/* The one past the highest gsi number used */
166u32 gsi_top;
167
168/* MP IRQ source entries */
169struct mpc_intsrc mp_irqs[MAX_IRQ_SOURCES];
170
171/* # of MP IRQ source entries */
172int mp_irq_entries;
173
174#ifdef CONFIG_EISA
175int mp_bus_id_to_type[MAX_MP_BUSSES];
176#endif
177
178DECLARE_BITMAP(mp_bus_not_pci, MAX_MP_BUSSES);
179
180int skip_ioapic_setup;
181
182/**
183 * disable_ioapic_support() - disables ioapic support at runtime
184 */
185void disable_ioapic_support(void)
186{
187#ifdef CONFIG_PCI
188 noioapicquirk = 1;
189 noioapicreroute = -1;
190#endif
191 skip_ioapic_setup = 1;
192}
193
194static int __init parse_noapic(char *str)
195{
196 /* disable IO-APIC */
197 disable_ioapic_support();
198 return 0;
199}
200early_param("noapic", parse_noapic);
201
202/* Will be called in mpparse/ACPI codes for saving IRQ info */
203void mp_save_irq(struct mpc_intsrc *m)
204{
205 int i;
206
207 apic_printk(APIC_VERBOSE, "Int: type %d, pol %d, trig %d, bus %02x,"
208 " IRQ %02x, APIC ID %x, APIC INT %02x\n",
209 m->irqtype, m->irqflag & 3, (m->irqflag >> 2) & 3, m->srcbus,
210 m->srcbusirq, m->dstapic, m->dstirq);
211
212 for (i = 0; i < mp_irq_entries; i++) {
213 if (!memcmp(&mp_irqs[i], m, sizeof(*m)))
214 return;
215 }
216
217 memcpy(&mp_irqs[mp_irq_entries], m, sizeof(*m));
218 if (++mp_irq_entries == MAX_IRQ_SOURCES)
219 panic("Max # of irq sources exceeded!!\n");
220}
221
222static void alloc_ioapic_saved_registers(int idx)
223{
224 size_t size;
225
226 if (ioapics[idx].saved_registers)
227 return;
228
229 size = sizeof(struct IO_APIC_route_entry) * ioapics[idx].nr_registers;
230 ioapics[idx].saved_registers = kzalloc(size, GFP_KERNEL);
231 if (!ioapics[idx].saved_registers)
232 pr_err("IOAPIC %d: suspend/resume impossible!\n", idx);
233}
234
235static void free_ioapic_saved_registers(int idx)
236{
237 kfree(ioapics[idx].saved_registers);
238 ioapics[idx].saved_registers = NULL;
239}
240
241int __init arch_early_ioapic_init(void)
242{
243 int i;
244
245 if (!nr_legacy_irqs())
246 io_apic_irqs = ~0UL;
247
248 for_each_ioapic(i)
249 alloc_ioapic_saved_registers(i);
250
251 return 0;
252}
253
254struct io_apic {
255 unsigned int index;
256 unsigned int unused[3];
257 unsigned int data;
258 unsigned int unused2[11];
259 unsigned int eoi;
260};
261
262static __attribute_const__ struct io_apic __iomem *io_apic_base(int idx)
263{
264 return (void __iomem *) __fix_to_virt(FIX_IO_APIC_BASE_0 + idx)
265 + (mpc_ioapic_addr(idx) & ~PAGE_MASK);
266}
267
268static inline void io_apic_eoi(unsigned int apic, unsigned int vector)
269{
270 struct io_apic __iomem *io_apic = io_apic_base(apic);
271 writel(vector, &io_apic->eoi);
272}
273
274unsigned int native_io_apic_read(unsigned int apic, unsigned int reg)
275{
276 struct io_apic __iomem *io_apic = io_apic_base(apic);
277 writel(reg, &io_apic->index);
278 return readl(&io_apic->data);
279}
280
281static void io_apic_write(unsigned int apic, unsigned int reg,
282 unsigned int value)
283{
284 struct io_apic __iomem *io_apic = io_apic_base(apic);
285
286 writel(reg, &io_apic->index);
287 writel(value, &io_apic->data);
288}
289
290static struct IO_APIC_route_entry __ioapic_read_entry(int apic, int pin)
291{
292 struct IO_APIC_route_entry entry;
293
294 entry.w1 = io_apic_read(apic, 0x10 + 2 * pin);
295 entry.w2 = io_apic_read(apic, 0x11 + 2 * pin);
296
297 return entry;
298}
299
300static struct IO_APIC_route_entry ioapic_read_entry(int apic, int pin)
301{
302 struct IO_APIC_route_entry entry;
303 unsigned long flags;
304
305 raw_spin_lock_irqsave(&ioapic_lock, flags);
306 entry = __ioapic_read_entry(apic, pin);
307 raw_spin_unlock_irqrestore(&ioapic_lock, flags);
308
309 return entry;
310}
311
312/*
313 * When we write a new IO APIC routing entry, we need to write the high
314 * word first! If the mask bit in the low word is clear, we will enable
315 * the interrupt, and we need to make sure the entry is fully populated
316 * before that happens.
317 */
318static void __ioapic_write_entry(int apic, int pin, struct IO_APIC_route_entry e)
319{
320 io_apic_write(apic, 0x11 + 2*pin, e.w2);
321 io_apic_write(apic, 0x10 + 2*pin, e.w1);
322}
323
324static void ioapic_write_entry(int apic, int pin, struct IO_APIC_route_entry e)
325{
326 unsigned long flags;
327
328 raw_spin_lock_irqsave(&ioapic_lock, flags);
329 __ioapic_write_entry(apic, pin, e);
330 raw_spin_unlock_irqrestore(&ioapic_lock, flags);
331}
332
333/*
334 * When we mask an IO APIC routing entry, we need to write the low
335 * word first, in order to set the mask bit before we change the
336 * high bits!
337 */
338static void ioapic_mask_entry(int apic, int pin)
339{
340 struct IO_APIC_route_entry e = { .masked = true };
341 unsigned long flags;
342
343 raw_spin_lock_irqsave(&ioapic_lock, flags);
344 io_apic_write(apic, 0x10 + 2*pin, e.w1);
345 io_apic_write(apic, 0x11 + 2*pin, e.w2);
346 raw_spin_unlock_irqrestore(&ioapic_lock, flags);
347}
348
349/*
350 * The common case is 1:1 IRQ<->pin mappings. Sometimes there are
351 * shared ISA-space IRQs, so we have to support them. We are super
352 * fast in the common case, and fast for shared ISA-space IRQs.
353 */
354static int __add_pin_to_irq_node(struct mp_chip_data *data,
355 int node, int apic, int pin)
356{
357 struct irq_pin_list *entry;
358
359 /* don't allow duplicates */
360 for_each_irq_pin(entry, data->irq_2_pin)
361 if (entry->apic == apic && entry->pin == pin)
362 return 0;
363
364 entry = kzalloc_node(sizeof(struct irq_pin_list), GFP_ATOMIC, node);
365 if (!entry) {
366 pr_err("can not alloc irq_pin_list (%d,%d,%d)\n",
367 node, apic, pin);
368 return -ENOMEM;
369 }
370 entry->apic = apic;
371 entry->pin = pin;
372 list_add_tail(&entry->list, &data->irq_2_pin);
373
374 return 0;
375}
376
377static void __remove_pin_from_irq(struct mp_chip_data *data, int apic, int pin)
378{
379 struct irq_pin_list *tmp, *entry;
380
381 list_for_each_entry_safe(entry, tmp, &data->irq_2_pin, list)
382 if (entry->apic == apic && entry->pin == pin) {
383 list_del(&entry->list);
384 kfree(entry);
385 return;
386 }
387}
388
389static void add_pin_to_irq_node(struct mp_chip_data *data,
390 int node, int apic, int pin)
391{
392 if (__add_pin_to_irq_node(data, node, apic, pin))
393 panic("IO-APIC: failed to add irq-pin. Can not proceed\n");
394}
395
396/*
397 * Reroute an IRQ to a different pin.
398 */
399static void __init replace_pin_at_irq_node(struct mp_chip_data *data, int node,
400 int oldapic, int oldpin,
401 int newapic, int newpin)
402{
403 struct irq_pin_list *entry;
404
405 for_each_irq_pin(entry, data->irq_2_pin) {
406 if (entry->apic == oldapic && entry->pin == oldpin) {
407 entry->apic = newapic;
408 entry->pin = newpin;
409 /* every one is different, right? */
410 return;
411 }
412 }
413
414 /* old apic/pin didn't exist, so just add new ones */
415 add_pin_to_irq_node(data, node, newapic, newpin);
416}
417
418static void io_apic_modify_irq(struct mp_chip_data *data, bool masked,
419 void (*final)(struct irq_pin_list *entry))
420{
421 struct irq_pin_list *entry;
422
423 data->entry.masked = masked;
424
425 for_each_irq_pin(entry, data->irq_2_pin) {
426 io_apic_write(entry->apic, 0x10 + 2 * entry->pin, data->entry.w1);
427 if (final)
428 final(entry);
429 }
430}
431
432static void io_apic_sync(struct irq_pin_list *entry)
433{
434 /*
435 * Synchronize the IO-APIC and the CPU by doing
436 * a dummy read from the IO-APIC
437 */
438 struct io_apic __iomem *io_apic;
439
440 io_apic = io_apic_base(entry->apic);
441 readl(&io_apic->data);
442}
443
444static void mask_ioapic_irq(struct irq_data *irq_data)
445{
446 struct mp_chip_data *data = irq_data->chip_data;
447 unsigned long flags;
448
449 raw_spin_lock_irqsave(&ioapic_lock, flags);
450 io_apic_modify_irq(data, true, &io_apic_sync);
451 raw_spin_unlock_irqrestore(&ioapic_lock, flags);
452}
453
454static void __unmask_ioapic(struct mp_chip_data *data)
455{
456 io_apic_modify_irq(data, false, NULL);
457}
458
459static void unmask_ioapic_irq(struct irq_data *irq_data)
460{
461 struct mp_chip_data *data = irq_data->chip_data;
462 unsigned long flags;
463
464 raw_spin_lock_irqsave(&ioapic_lock, flags);
465 __unmask_ioapic(data);
466 raw_spin_unlock_irqrestore(&ioapic_lock, flags);
467}
468
469/*
470 * IO-APIC versions below 0x20 don't support EOI register.
471 * For the record, here is the information about various versions:
472 * 0Xh 82489DX
473 * 1Xh I/OAPIC or I/O(x)APIC which are not PCI 2.2 Compliant
474 * 2Xh I/O(x)APIC which is PCI 2.2 Compliant
475 * 30h-FFh Reserved
476 *
477 * Some of the Intel ICH Specs (ICH2 to ICH5) documents the io-apic
478 * version as 0x2. This is an error with documentation and these ICH chips
479 * use io-apic's of version 0x20.
480 *
481 * For IO-APIC's with EOI register, we use that to do an explicit EOI.
482 * Otherwise, we simulate the EOI message manually by changing the trigger
483 * mode to edge and then back to level, with RTE being masked during this.
484 */
485static void __eoi_ioapic_pin(int apic, int pin, int vector)
486{
487 if (mpc_ioapic_ver(apic) >= 0x20) {
488 io_apic_eoi(apic, vector);
489 } else {
490 struct IO_APIC_route_entry entry, entry1;
491
492 entry = entry1 = __ioapic_read_entry(apic, pin);
493
494 /*
495 * Mask the entry and change the trigger mode to edge.
496 */
497 entry1.masked = true;
498 entry1.is_level = false;
499
500 __ioapic_write_entry(apic, pin, entry1);
501
502 /*
503 * Restore the previous level triggered entry.
504 */
505 __ioapic_write_entry(apic, pin, entry);
506 }
507}
508
509static void eoi_ioapic_pin(int vector, struct mp_chip_data *data)
510{
511 unsigned long flags;
512 struct irq_pin_list *entry;
513
514 raw_spin_lock_irqsave(&ioapic_lock, flags);
515 for_each_irq_pin(entry, data->irq_2_pin)
516 __eoi_ioapic_pin(entry->apic, entry->pin, vector);
517 raw_spin_unlock_irqrestore(&ioapic_lock, flags);
518}
519
520static void clear_IO_APIC_pin(unsigned int apic, unsigned int pin)
521{
522 struct IO_APIC_route_entry entry;
523
524 /* Check delivery_mode to be sure we're not clearing an SMI pin */
525 entry = ioapic_read_entry(apic, pin);
526 if (entry.delivery_mode == APIC_DELIVERY_MODE_SMI)
527 return;
528
529 /*
530 * Make sure the entry is masked and re-read the contents to check
531 * if it is a level triggered pin and if the remote-IRR is set.
532 */
533 if (!entry.masked) {
534 entry.masked = true;
535 ioapic_write_entry(apic, pin, entry);
536 entry = ioapic_read_entry(apic, pin);
537 }
538
539 if (entry.irr) {
540 unsigned long flags;
541
542 /*
543 * Make sure the trigger mode is set to level. Explicit EOI
544 * doesn't clear the remote-IRR if the trigger mode is not
545 * set to level.
546 */
547 if (!entry.is_level) {
548 entry.is_level = true;
549 ioapic_write_entry(apic, pin, entry);
550 }
551 raw_spin_lock_irqsave(&ioapic_lock, flags);
552 __eoi_ioapic_pin(apic, pin, entry.vector);
553 raw_spin_unlock_irqrestore(&ioapic_lock, flags);
554 }
555
556 /*
557 * Clear the rest of the bits in the IO-APIC RTE except for the mask
558 * bit.
559 */
560 ioapic_mask_entry(apic, pin);
561 entry = ioapic_read_entry(apic, pin);
562 if (entry.irr)
563 pr_err("Unable to reset IRR for apic: %d, pin :%d\n",
564 mpc_ioapic_id(apic), pin);
565}
566
567void clear_IO_APIC (void)
568{
569 int apic, pin;
570
571 for_each_ioapic_pin(apic, pin)
572 clear_IO_APIC_pin(apic, pin);
573}
574
575#ifdef CONFIG_X86_32
576/*
577 * support for broken MP BIOSs, enables hand-redirection of PIRQ0-7 to
578 * specific CPU-side IRQs.
579 */
580
581#define MAX_PIRQS 8
582static int pirq_entries[MAX_PIRQS] = {
583 [0 ... MAX_PIRQS - 1] = -1
584};
585
586static int __init ioapic_pirq_setup(char *str)
587{
588 int i, max;
589 int ints[MAX_PIRQS+1];
590
591 get_options(str, ARRAY_SIZE(ints), ints);
592
593 apic_printk(APIC_VERBOSE, KERN_INFO
594 "PIRQ redirection, working around broken MP-BIOS.\n");
595 max = MAX_PIRQS;
596 if (ints[0] < MAX_PIRQS)
597 max = ints[0];
598
599 for (i = 0; i < max; i++) {
600 apic_printk(APIC_VERBOSE, KERN_DEBUG
601 "... PIRQ%d -> IRQ %d\n", i, ints[i+1]);
602 /*
603 * PIRQs are mapped upside down, usually.
604 */
605 pirq_entries[MAX_PIRQS-i-1] = ints[i+1];
606 }
607 return 1;
608}
609
610__setup("pirq=", ioapic_pirq_setup);
611#endif /* CONFIG_X86_32 */
612
613/*
614 * Saves all the IO-APIC RTE's
615 */
616int save_ioapic_entries(void)
617{
618 int apic, pin;
619 int err = 0;
620
621 for_each_ioapic(apic) {
622 if (!ioapics[apic].saved_registers) {
623 err = -ENOMEM;
624 continue;
625 }
626
627 for_each_pin(apic, pin)
628 ioapics[apic].saved_registers[pin] =
629 ioapic_read_entry(apic, pin);
630 }
631
632 return err;
633}
634
635/*
636 * Mask all IO APIC entries.
637 */
638void mask_ioapic_entries(void)
639{
640 int apic, pin;
641
642 for_each_ioapic(apic) {
643 if (!ioapics[apic].saved_registers)
644 continue;
645
646 for_each_pin(apic, pin) {
647 struct IO_APIC_route_entry entry;
648
649 entry = ioapics[apic].saved_registers[pin];
650 if (!entry.masked) {
651 entry.masked = true;
652 ioapic_write_entry(apic, pin, entry);
653 }
654 }
655 }
656}
657
658/*
659 * Restore IO APIC entries which was saved in the ioapic structure.
660 */
661int restore_ioapic_entries(void)
662{
663 int apic, pin;
664
665 for_each_ioapic(apic) {
666 if (!ioapics[apic].saved_registers)
667 continue;
668
669 for_each_pin(apic, pin)
670 ioapic_write_entry(apic, pin,
671 ioapics[apic].saved_registers[pin]);
672 }
673 return 0;
674}
675
676/*
677 * Find the IRQ entry number of a certain pin.
678 */
679static int find_irq_entry(int ioapic_idx, int pin, int type)
680{
681 int i;
682
683 for (i = 0; i < mp_irq_entries; i++)
684 if (mp_irqs[i].irqtype == type &&
685 (mp_irqs[i].dstapic == mpc_ioapic_id(ioapic_idx) ||
686 mp_irqs[i].dstapic == MP_APIC_ALL) &&
687 mp_irqs[i].dstirq == pin)
688 return i;
689
690 return -1;
691}
692
693/*
694 * Find the pin to which IRQ[irq] (ISA) is connected
695 */
696static int __init find_isa_irq_pin(int irq, int type)
697{
698 int i;
699
700 for (i = 0; i < mp_irq_entries; i++) {
701 int lbus = mp_irqs[i].srcbus;
702
703 if (test_bit(lbus, mp_bus_not_pci) &&
704 (mp_irqs[i].irqtype == type) &&
705 (mp_irqs[i].srcbusirq == irq))
706
707 return mp_irqs[i].dstirq;
708 }
709 return -1;
710}
711
712static int __init find_isa_irq_apic(int irq, int type)
713{
714 int i;
715
716 for (i = 0; i < mp_irq_entries; i++) {
717 int lbus = mp_irqs[i].srcbus;
718
719 if (test_bit(lbus, mp_bus_not_pci) &&
720 (mp_irqs[i].irqtype == type) &&
721 (mp_irqs[i].srcbusirq == irq))
722 break;
723 }
724
725 if (i < mp_irq_entries) {
726 int ioapic_idx;
727
728 for_each_ioapic(ioapic_idx)
729 if (mpc_ioapic_id(ioapic_idx) == mp_irqs[i].dstapic)
730 return ioapic_idx;
731 }
732
733 return -1;
734}
735
736static bool irq_active_low(int idx)
737{
738 int bus = mp_irqs[idx].srcbus;
739
740 /*
741 * Determine IRQ line polarity (high active or low active):
742 */
743 switch (mp_irqs[idx].irqflag & MP_IRQPOL_MASK) {
744 case MP_IRQPOL_DEFAULT:
745 /*
746 * Conforms to spec, ie. bus-type dependent polarity. PCI
747 * defaults to low active. [E]ISA defaults to high active.
748 */
749 return !test_bit(bus, mp_bus_not_pci);
750 case MP_IRQPOL_ACTIVE_HIGH:
751 return false;
752 case MP_IRQPOL_RESERVED:
753 pr_warn("IOAPIC: Invalid polarity: 2, defaulting to low\n");
754 fallthrough;
755 case MP_IRQPOL_ACTIVE_LOW:
756 default: /* Pointless default required due to do gcc stupidity */
757 return true;
758 }
759}
760
761#ifdef CONFIG_EISA
762/*
763 * EISA Edge/Level control register, ELCR
764 */
765static bool EISA_ELCR(unsigned int irq)
766{
767 if (irq < nr_legacy_irqs()) {
768 unsigned int port = PIC_ELCR1 + (irq >> 3);
769 return (inb(port) >> (irq & 7)) & 1;
770 }
771 apic_printk(APIC_VERBOSE, KERN_INFO
772 "Broken MPtable reports ISA irq %d\n", irq);
773 return false;
774}
775
776/*
777 * EISA interrupts are always active high and can be edge or level
778 * triggered depending on the ELCR value. If an interrupt is listed as
779 * EISA conforming in the MP table, that means its trigger type must be
780 * read in from the ELCR.
781 */
782static bool eisa_irq_is_level(int idx, int bus, bool level)
783{
784 switch (mp_bus_id_to_type[bus]) {
785 case MP_BUS_PCI:
786 case MP_BUS_ISA:
787 return level;
788 case MP_BUS_EISA:
789 return EISA_ELCR(mp_irqs[idx].srcbusirq);
790 }
791 pr_warn("IOAPIC: Invalid srcbus: %d defaulting to level\n", bus);
792 return true;
793}
794#else
795static inline int eisa_irq_is_level(int idx, int bus, bool level)
796{
797 return level;
798}
799#endif
800
801static bool irq_is_level(int idx)
802{
803 int bus = mp_irqs[idx].srcbus;
804 bool level;
805
806 /*
807 * Determine IRQ trigger mode (edge or level sensitive):
808 */
809 switch (mp_irqs[idx].irqflag & MP_IRQTRIG_MASK) {
810 case MP_IRQTRIG_DEFAULT:
811 /*
812 * Conforms to spec, ie. bus-type dependent trigger
813 * mode. PCI defaults to level, ISA to edge.
814 */
815 level = !test_bit(bus, mp_bus_not_pci);
816 /* Take EISA into account */
817 return eisa_irq_is_level(idx, bus, level);
818 case MP_IRQTRIG_EDGE:
819 return false;
820 case MP_IRQTRIG_RESERVED:
821 pr_warn("IOAPIC: Invalid trigger mode 2 defaulting to level\n");
822 fallthrough;
823 case MP_IRQTRIG_LEVEL:
824 default: /* Pointless default required due to do gcc stupidity */
825 return true;
826 }
827}
828
829static int __acpi_get_override_irq(u32 gsi, bool *trigger, bool *polarity)
830{
831 int ioapic, pin, idx;
832
833 if (skip_ioapic_setup)
834 return -1;
835
836 ioapic = mp_find_ioapic(gsi);
837 if (ioapic < 0)
838 return -1;
839
840 pin = mp_find_ioapic_pin(ioapic, gsi);
841 if (pin < 0)
842 return -1;
843
844 idx = find_irq_entry(ioapic, pin, mp_INT);
845 if (idx < 0)
846 return -1;
847
848 *trigger = irq_is_level(idx);
849 *polarity = irq_active_low(idx);
850 return 0;
851}
852
853#ifdef CONFIG_ACPI
854int acpi_get_override_irq(u32 gsi, int *is_level, int *active_low)
855{
856 *is_level = *active_low = 0;
857 return __acpi_get_override_irq(gsi, (bool *)is_level,
858 (bool *)active_low);
859}
860#endif
861
862void ioapic_set_alloc_attr(struct irq_alloc_info *info, int node,
863 int trigger, int polarity)
864{
865 init_irq_alloc_info(info, NULL);
866 info->type = X86_IRQ_ALLOC_TYPE_IOAPIC;
867 info->ioapic.node = node;
868 info->ioapic.is_level = trigger;
869 info->ioapic.active_low = polarity;
870 info->ioapic.valid = 1;
871}
872
873static void ioapic_copy_alloc_attr(struct irq_alloc_info *dst,
874 struct irq_alloc_info *src,
875 u32 gsi, int ioapic_idx, int pin)
876{
877 bool level, pol_low;
878
879 copy_irq_alloc_info(dst, src);
880 dst->type = X86_IRQ_ALLOC_TYPE_IOAPIC;
881 dst->devid = mpc_ioapic_id(ioapic_idx);
882 dst->ioapic.pin = pin;
883 dst->ioapic.valid = 1;
884 if (src && src->ioapic.valid) {
885 dst->ioapic.node = src->ioapic.node;
886 dst->ioapic.is_level = src->ioapic.is_level;
887 dst->ioapic.active_low = src->ioapic.active_low;
888 } else {
889 dst->ioapic.node = NUMA_NO_NODE;
890 if (__acpi_get_override_irq(gsi, &level, &pol_low) >= 0) {
891 dst->ioapic.is_level = level;
892 dst->ioapic.active_low = pol_low;
893 } else {
894 /*
895 * PCI interrupts are always active low level
896 * triggered.
897 */
898 dst->ioapic.is_level = true;
899 dst->ioapic.active_low = true;
900 }
901 }
902}
903
904static int ioapic_alloc_attr_node(struct irq_alloc_info *info)
905{
906 return (info && info->ioapic.valid) ? info->ioapic.node : NUMA_NO_NODE;
907}
908
909static void mp_register_handler(unsigned int irq, bool level)
910{
911 irq_flow_handler_t hdl;
912 bool fasteoi;
913
914 if (level) {
915 irq_set_status_flags(irq, IRQ_LEVEL);
916 fasteoi = true;
917 } else {
918 irq_clear_status_flags(irq, IRQ_LEVEL);
919 fasteoi = false;
920 }
921
922 hdl = fasteoi ? handle_fasteoi_irq : handle_edge_irq;
923 __irq_set_handler(irq, hdl, 0, fasteoi ? "fasteoi" : "edge");
924}
925
926static bool mp_check_pin_attr(int irq, struct irq_alloc_info *info)
927{
928 struct mp_chip_data *data = irq_get_chip_data(irq);
929
930 /*
931 * setup_IO_APIC_irqs() programs all legacy IRQs with default trigger
932 * and polarity attributes. So allow the first user to reprogram the
933 * pin with real trigger and polarity attributes.
934 */
935 if (irq < nr_legacy_irqs() && data->count == 1) {
936 if (info->ioapic.is_level != data->is_level)
937 mp_register_handler(irq, info->ioapic.is_level);
938 data->entry.is_level = data->is_level = info->ioapic.is_level;
939 data->entry.active_low = data->active_low = info->ioapic.active_low;
940 }
941
942 return data->is_level == info->ioapic.is_level &&
943 data->active_low == info->ioapic.active_low;
944}
945
946static int alloc_irq_from_domain(struct irq_domain *domain, int ioapic, u32 gsi,
947 struct irq_alloc_info *info)
948{
949 bool legacy = false;
950 int irq = -1;
951 int type = ioapics[ioapic].irqdomain_cfg.type;
952
953 switch (type) {
954 case IOAPIC_DOMAIN_LEGACY:
955 /*
956 * Dynamically allocate IRQ number for non-ISA IRQs in the first
957 * 16 GSIs on some weird platforms.
958 */
959 if (!ioapic_initialized || gsi >= nr_legacy_irqs())
960 irq = gsi;
961 legacy = mp_is_legacy_irq(irq);
962 break;
963 case IOAPIC_DOMAIN_STRICT:
964 irq = gsi;
965 break;
966 case IOAPIC_DOMAIN_DYNAMIC:
967 break;
968 default:
969 WARN(1, "ioapic: unknown irqdomain type %d\n", type);
970 return -1;
971 }
972
973 return __irq_domain_alloc_irqs(domain, irq, 1,
974 ioapic_alloc_attr_node(info),
975 info, legacy, NULL);
976}
977
978/*
979 * Need special handling for ISA IRQs because there may be multiple IOAPIC pins
980 * sharing the same ISA IRQ number and irqdomain only supports 1:1 mapping
981 * between IOAPIC pin and IRQ number. A typical IOAPIC has 24 pins, pin 0-15 are
982 * used for legacy IRQs and pin 16-23 are used for PCI IRQs (PIRQ A-H).
983 * When ACPI is disabled, only legacy IRQ numbers (IRQ0-15) are available, and
984 * some BIOSes may use MP Interrupt Source records to override IRQ numbers for
985 * PIRQs instead of reprogramming the interrupt routing logic. Thus there may be
986 * multiple pins sharing the same legacy IRQ number when ACPI is disabled.
987 */
988static int alloc_isa_irq_from_domain(struct irq_domain *domain,
989 int irq, int ioapic, int pin,
990 struct irq_alloc_info *info)
991{
992 struct mp_chip_data *data;
993 struct irq_data *irq_data = irq_get_irq_data(irq);
994 int node = ioapic_alloc_attr_node(info);
995
996 /*
997 * Legacy ISA IRQ has already been allocated, just add pin to
998 * the pin list associated with this IRQ and program the IOAPIC
999 * entry. The IOAPIC entry
1000 */
1001 if (irq_data && irq_data->parent_data) {
1002 if (!mp_check_pin_attr(irq, info))
1003 return -EBUSY;
1004 if (__add_pin_to_irq_node(irq_data->chip_data, node, ioapic,
1005 info->ioapic.pin))
1006 return -ENOMEM;
1007 } else {
1008 info->flags |= X86_IRQ_ALLOC_LEGACY;
1009 irq = __irq_domain_alloc_irqs(domain, irq, 1, node, info, true,
1010 NULL);
1011 if (irq >= 0) {
1012 irq_data = irq_domain_get_irq_data(domain, irq);
1013 data = irq_data->chip_data;
1014 data->isa_irq = true;
1015 }
1016 }
1017
1018 return irq;
1019}
1020
1021static int mp_map_pin_to_irq(u32 gsi, int idx, int ioapic, int pin,
1022 unsigned int flags, struct irq_alloc_info *info)
1023{
1024 int irq;
1025 bool legacy = false;
1026 struct irq_alloc_info tmp;
1027 struct mp_chip_data *data;
1028 struct irq_domain *domain = mp_ioapic_irqdomain(ioapic);
1029
1030 if (!domain)
1031 return -ENOSYS;
1032
1033 if (idx >= 0 && test_bit(mp_irqs[idx].srcbus, mp_bus_not_pci)) {
1034 irq = mp_irqs[idx].srcbusirq;
1035 legacy = mp_is_legacy_irq(irq);
1036 /*
1037 * IRQ2 is unusable for historical reasons on systems which
1038 * have a legacy PIC. See the comment vs. IRQ2 further down.
1039 *
1040 * If this gets removed at some point then the related code
1041 * in lapic_assign_system_vectors() needs to be adjusted as
1042 * well.
1043 */
1044 if (legacy && irq == PIC_CASCADE_IR)
1045 return -EINVAL;
1046 }
1047
1048 mutex_lock(&ioapic_mutex);
1049 if (!(flags & IOAPIC_MAP_ALLOC)) {
1050 if (!legacy) {
1051 irq = irq_find_mapping(domain, pin);
1052 if (irq == 0)
1053 irq = -ENOENT;
1054 }
1055 } else {
1056 ioapic_copy_alloc_attr(&tmp, info, gsi, ioapic, pin);
1057 if (legacy)
1058 irq = alloc_isa_irq_from_domain(domain, irq,
1059 ioapic, pin, &tmp);
1060 else if ((irq = irq_find_mapping(domain, pin)) == 0)
1061 irq = alloc_irq_from_domain(domain, ioapic, gsi, &tmp);
1062 else if (!mp_check_pin_attr(irq, &tmp))
1063 irq = -EBUSY;
1064 if (irq >= 0) {
1065 data = irq_get_chip_data(irq);
1066 data->count++;
1067 }
1068 }
1069 mutex_unlock(&ioapic_mutex);
1070
1071 return irq;
1072}
1073
1074static int pin_2_irq(int idx, int ioapic, int pin, unsigned int flags)
1075{
1076 u32 gsi = mp_pin_to_gsi(ioapic, pin);
1077
1078 /*
1079 * Debugging check, we are in big trouble if this message pops up!
1080 */
1081 if (mp_irqs[idx].dstirq != pin)
1082 pr_err("broken BIOS or MPTABLE parser, ayiee!!\n");
1083
1084#ifdef CONFIG_X86_32
1085 /*
1086 * PCI IRQ command line redirection. Yes, limits are hardcoded.
1087 */
1088 if ((pin >= 16) && (pin <= 23)) {
1089 if (pirq_entries[pin-16] != -1) {
1090 if (!pirq_entries[pin-16]) {
1091 apic_printk(APIC_VERBOSE, KERN_DEBUG
1092 "disabling PIRQ%d\n", pin-16);
1093 } else {
1094 int irq = pirq_entries[pin-16];
1095 apic_printk(APIC_VERBOSE, KERN_DEBUG
1096 "using PIRQ%d -> IRQ %d\n",
1097 pin-16, irq);
1098 return irq;
1099 }
1100 }
1101 }
1102#endif
1103
1104 return mp_map_pin_to_irq(gsi, idx, ioapic, pin, flags, NULL);
1105}
1106
1107int mp_map_gsi_to_irq(u32 gsi, unsigned int flags, struct irq_alloc_info *info)
1108{
1109 int ioapic, pin, idx;
1110
1111 ioapic = mp_find_ioapic(gsi);
1112 if (ioapic < 0)
1113 return -ENODEV;
1114
1115 pin = mp_find_ioapic_pin(ioapic, gsi);
1116 idx = find_irq_entry(ioapic, pin, mp_INT);
1117 if ((flags & IOAPIC_MAP_CHECK) && idx < 0)
1118 return -ENODEV;
1119
1120 return mp_map_pin_to_irq(gsi, idx, ioapic, pin, flags, info);
1121}
1122
1123void mp_unmap_irq(int irq)
1124{
1125 struct irq_data *irq_data = irq_get_irq_data(irq);
1126 struct mp_chip_data *data;
1127
1128 if (!irq_data || !irq_data->domain)
1129 return;
1130
1131 data = irq_data->chip_data;
1132 if (!data || data->isa_irq)
1133 return;
1134
1135 mutex_lock(&ioapic_mutex);
1136 if (--data->count == 0)
1137 irq_domain_free_irqs(irq, 1);
1138 mutex_unlock(&ioapic_mutex);
1139}
1140
1141/*
1142 * Find a specific PCI IRQ entry.
1143 * Not an __init, possibly needed by modules
1144 */
1145int IO_APIC_get_PCI_irq_vector(int bus, int slot, int pin)
1146{
1147 int irq, i, best_ioapic = -1, best_idx = -1;
1148
1149 apic_printk(APIC_DEBUG,
1150 "querying PCI -> IRQ mapping bus:%d, slot:%d, pin:%d.\n",
1151 bus, slot, pin);
1152 if (test_bit(bus, mp_bus_not_pci)) {
1153 apic_printk(APIC_VERBOSE,
1154 "PCI BIOS passed nonexistent PCI bus %d!\n", bus);
1155 return -1;
1156 }
1157
1158 for (i = 0; i < mp_irq_entries; i++) {
1159 int lbus = mp_irqs[i].srcbus;
1160 int ioapic_idx, found = 0;
1161
1162 if (bus != lbus || mp_irqs[i].irqtype != mp_INT ||
1163 slot != ((mp_irqs[i].srcbusirq >> 2) & 0x1f))
1164 continue;
1165
1166 for_each_ioapic(ioapic_idx)
1167 if (mpc_ioapic_id(ioapic_idx) == mp_irqs[i].dstapic ||
1168 mp_irqs[i].dstapic == MP_APIC_ALL) {
1169 found = 1;
1170 break;
1171 }
1172 if (!found)
1173 continue;
1174
1175 /* Skip ISA IRQs */
1176 irq = pin_2_irq(i, ioapic_idx, mp_irqs[i].dstirq, 0);
1177 if (irq > 0 && !IO_APIC_IRQ(irq))
1178 continue;
1179
1180 if (pin == (mp_irqs[i].srcbusirq & 3)) {
1181 best_idx = i;
1182 best_ioapic = ioapic_idx;
1183 goto out;
1184 }
1185
1186 /*
1187 * Use the first all-but-pin matching entry as a
1188 * best-guess fuzzy result for broken mptables.
1189 */
1190 if (best_idx < 0) {
1191 best_idx = i;
1192 best_ioapic = ioapic_idx;
1193 }
1194 }
1195 if (best_idx < 0)
1196 return -1;
1197
1198out:
1199 return pin_2_irq(best_idx, best_ioapic, mp_irqs[best_idx].dstirq,
1200 IOAPIC_MAP_ALLOC);
1201}
1202EXPORT_SYMBOL(IO_APIC_get_PCI_irq_vector);
1203
1204static struct irq_chip ioapic_chip, ioapic_ir_chip;
1205
1206static void __init setup_IO_APIC_irqs(void)
1207{
1208 unsigned int ioapic, pin;
1209 int idx;
1210
1211 apic_printk(APIC_VERBOSE, KERN_DEBUG "init IO_APIC IRQs\n");
1212
1213 for_each_ioapic_pin(ioapic, pin) {
1214 idx = find_irq_entry(ioapic, pin, mp_INT);
1215 if (idx < 0)
1216 apic_printk(APIC_VERBOSE,
1217 KERN_DEBUG " apic %d pin %d not connected\n",
1218 mpc_ioapic_id(ioapic), pin);
1219 else
1220 pin_2_irq(idx, ioapic, pin,
1221 ioapic ? 0 : IOAPIC_MAP_ALLOC);
1222 }
1223}
1224
1225void ioapic_zap_locks(void)
1226{
1227 raw_spin_lock_init(&ioapic_lock);
1228}
1229
1230static void io_apic_print_entries(unsigned int apic, unsigned int nr_entries)
1231{
1232 struct IO_APIC_route_entry entry;
1233 char buf[256];
1234 int i;
1235
1236 printk(KERN_DEBUG "IOAPIC %d:\n", apic);
1237 for (i = 0; i <= nr_entries; i++) {
1238 entry = ioapic_read_entry(apic, i);
1239 snprintf(buf, sizeof(buf),
1240 " pin%02x, %s, %s, %s, V(%02X), IRR(%1d), S(%1d)",
1241 i,
1242 entry.masked ? "disabled" : "enabled ",
1243 entry.is_level ? "level" : "edge ",
1244 entry.active_low ? "low " : "high",
1245 entry.vector, entry.irr, entry.delivery_status);
1246 if (entry.ir_format) {
1247 printk(KERN_DEBUG "%s, remapped, I(%04X), Z(%X)\n",
1248 buf,
1249 (entry.ir_index_15 << 15) | entry.ir_index_0_14,
1250 entry.ir_zero);
1251 } else {
1252 printk(KERN_DEBUG "%s, %s, D(%02X%02X), M(%1d)\n", buf,
1253 entry.dest_mode_logical ? "logical " : "physical",
1254 entry.virt_destid_8_14, entry.destid_0_7,
1255 entry.delivery_mode);
1256 }
1257 }
1258}
1259
1260static void __init print_IO_APIC(int ioapic_idx)
1261{
1262 union IO_APIC_reg_00 reg_00;
1263 union IO_APIC_reg_01 reg_01;
1264 union IO_APIC_reg_02 reg_02;
1265 union IO_APIC_reg_03 reg_03;
1266 unsigned long flags;
1267
1268 raw_spin_lock_irqsave(&ioapic_lock, flags);
1269 reg_00.raw = io_apic_read(ioapic_idx, 0);
1270 reg_01.raw = io_apic_read(ioapic_idx, 1);
1271 if (reg_01.bits.version >= 0x10)
1272 reg_02.raw = io_apic_read(ioapic_idx, 2);
1273 if (reg_01.bits.version >= 0x20)
1274 reg_03.raw = io_apic_read(ioapic_idx, 3);
1275 raw_spin_unlock_irqrestore(&ioapic_lock, flags);
1276
1277 printk(KERN_DEBUG "IO APIC #%d......\n", mpc_ioapic_id(ioapic_idx));
1278 printk(KERN_DEBUG ".... register #00: %08X\n", reg_00.raw);
1279 printk(KERN_DEBUG "....... : physical APIC id: %02X\n", reg_00.bits.ID);
1280 printk(KERN_DEBUG "....... : Delivery Type: %X\n", reg_00.bits.delivery_type);
1281 printk(KERN_DEBUG "....... : LTS : %X\n", reg_00.bits.LTS);
1282
1283 printk(KERN_DEBUG ".... register #01: %08X\n", *(int *)®_01);
1284 printk(KERN_DEBUG "....... : max redirection entries: %02X\n",
1285 reg_01.bits.entries);
1286
1287 printk(KERN_DEBUG "....... : PRQ implemented: %X\n", reg_01.bits.PRQ);
1288 printk(KERN_DEBUG "....... : IO APIC version: %02X\n",
1289 reg_01.bits.version);
1290
1291 /*
1292 * Some Intel chipsets with IO APIC VERSION of 0x1? don't have reg_02,
1293 * but the value of reg_02 is read as the previous read register
1294 * value, so ignore it if reg_02 == reg_01.
1295 */
1296 if (reg_01.bits.version >= 0x10 && reg_02.raw != reg_01.raw) {
1297 printk(KERN_DEBUG ".... register #02: %08X\n", reg_02.raw);
1298 printk(KERN_DEBUG "....... : arbitration: %02X\n", reg_02.bits.arbitration);
1299 }
1300
1301 /*
1302 * Some Intel chipsets with IO APIC VERSION of 0x2? don't have reg_02
1303 * or reg_03, but the value of reg_0[23] is read as the previous read
1304 * register value, so ignore it if reg_03 == reg_0[12].
1305 */
1306 if (reg_01.bits.version >= 0x20 && reg_03.raw != reg_02.raw &&
1307 reg_03.raw != reg_01.raw) {
1308 printk(KERN_DEBUG ".... register #03: %08X\n", reg_03.raw);
1309 printk(KERN_DEBUG "....... : Boot DT : %X\n", reg_03.bits.boot_DT);
1310 }
1311
1312 printk(KERN_DEBUG ".... IRQ redirection table:\n");
1313 io_apic_print_entries(ioapic_idx, reg_01.bits.entries);
1314}
1315
1316void __init print_IO_APICs(void)
1317{
1318 int ioapic_idx;
1319 unsigned int irq;
1320
1321 printk(KERN_DEBUG "number of MP IRQ sources: %d.\n", mp_irq_entries);
1322 for_each_ioapic(ioapic_idx)
1323 printk(KERN_DEBUG "number of IO-APIC #%d registers: %d.\n",
1324 mpc_ioapic_id(ioapic_idx),
1325 ioapics[ioapic_idx].nr_registers);
1326
1327 /*
1328 * We are a bit conservative about what we expect. We have to
1329 * know about every hardware change ASAP.
1330 */
1331 printk(KERN_INFO "testing the IO APIC.......................\n");
1332
1333 for_each_ioapic(ioapic_idx)
1334 print_IO_APIC(ioapic_idx);
1335
1336 printk(KERN_DEBUG "IRQ to pin mappings:\n");
1337 for_each_active_irq(irq) {
1338 struct irq_pin_list *entry;
1339 struct irq_chip *chip;
1340 struct mp_chip_data *data;
1341
1342 chip = irq_get_chip(irq);
1343 if (chip != &ioapic_chip && chip != &ioapic_ir_chip)
1344 continue;
1345 data = irq_get_chip_data(irq);
1346 if (!data)
1347 continue;
1348 if (list_empty(&data->irq_2_pin))
1349 continue;
1350
1351 printk(KERN_DEBUG "IRQ%d ", irq);
1352 for_each_irq_pin(entry, data->irq_2_pin)
1353 pr_cont("-> %d:%d", entry->apic, entry->pin);
1354 pr_cont("\n");
1355 }
1356
1357 printk(KERN_INFO ".................................... done.\n");
1358}
1359
1360/* Where if anywhere is the i8259 connect in external int mode */
1361static struct { int pin, apic; } ioapic_i8259 = { -1, -1 };
1362
1363void __init enable_IO_APIC(void)
1364{
1365 int i8259_apic, i8259_pin;
1366 int apic, pin;
1367
1368 if (skip_ioapic_setup)
1369 nr_ioapics = 0;
1370
1371 if (!nr_legacy_irqs() || !nr_ioapics)
1372 return;
1373
1374 for_each_ioapic_pin(apic, pin) {
1375 /* See if any of the pins is in ExtINT mode */
1376 struct IO_APIC_route_entry entry = ioapic_read_entry(apic, pin);
1377
1378 /* If the interrupt line is enabled and in ExtInt mode
1379 * I have found the pin where the i8259 is connected.
1380 */
1381 if (!entry.masked &&
1382 entry.delivery_mode == APIC_DELIVERY_MODE_EXTINT) {
1383 ioapic_i8259.apic = apic;
1384 ioapic_i8259.pin = pin;
1385 goto found_i8259;
1386 }
1387 }
1388 found_i8259:
1389 /* Look to see what if the MP table has reported the ExtINT */
1390 /* If we could not find the appropriate pin by looking at the ioapic
1391 * the i8259 probably is not connected the ioapic but give the
1392 * mptable a chance anyway.
1393 */
1394 i8259_pin = find_isa_irq_pin(0, mp_ExtINT);
1395 i8259_apic = find_isa_irq_apic(0, mp_ExtINT);
1396 /* Trust the MP table if nothing is setup in the hardware */
1397 if ((ioapic_i8259.pin == -1) && (i8259_pin >= 0)) {
1398 printk(KERN_WARNING "ExtINT not setup in hardware but reported by MP table\n");
1399 ioapic_i8259.pin = i8259_pin;
1400 ioapic_i8259.apic = i8259_apic;
1401 }
1402 /* Complain if the MP table and the hardware disagree */
1403 if (((ioapic_i8259.apic != i8259_apic) || (ioapic_i8259.pin != i8259_pin)) &&
1404 (i8259_pin >= 0) && (ioapic_i8259.pin >= 0))
1405 {
1406 printk(KERN_WARNING "ExtINT in hardware and MP table differ\n");
1407 }
1408
1409 /*
1410 * Do not trust the IO-APIC being empty at bootup
1411 */
1412 clear_IO_APIC();
1413}
1414
1415void native_restore_boot_irq_mode(void)
1416{
1417 /*
1418 * If the i8259 is routed through an IOAPIC
1419 * Put that IOAPIC in virtual wire mode
1420 * so legacy interrupts can be delivered.
1421 */
1422 if (ioapic_i8259.pin != -1) {
1423 struct IO_APIC_route_entry entry;
1424 u32 apic_id = read_apic_id();
1425
1426 memset(&entry, 0, sizeof(entry));
1427 entry.masked = false;
1428 entry.is_level = false;
1429 entry.active_low = false;
1430 entry.dest_mode_logical = false;
1431 entry.delivery_mode = APIC_DELIVERY_MODE_EXTINT;
1432 entry.destid_0_7 = apic_id & 0xFF;
1433 entry.virt_destid_8_14 = apic_id >> 8;
1434
1435 /*
1436 * Add it to the IO-APIC irq-routing table:
1437 */
1438 ioapic_write_entry(ioapic_i8259.apic, ioapic_i8259.pin, entry);
1439 }
1440
1441 if (boot_cpu_has(X86_FEATURE_APIC) || apic_from_smp_config())
1442 disconnect_bsp_APIC(ioapic_i8259.pin != -1);
1443}
1444
1445void restore_boot_irq_mode(void)
1446{
1447 if (!nr_legacy_irqs())
1448 return;
1449
1450 x86_apic_ops.restore();
1451}
1452
1453#ifdef CONFIG_X86_32
1454/*
1455 * function to set the IO-APIC physical IDs based on the
1456 * values stored in the MPC table.
1457 *
1458 * by Matt Domsch <Matt_Domsch@dell.com> Tue Dec 21 12:25:05 CST 1999
1459 */
1460void __init setup_ioapic_ids_from_mpc_nocheck(void)
1461{
1462 union IO_APIC_reg_00 reg_00;
1463 physid_mask_t phys_id_present_map;
1464 int ioapic_idx;
1465 int i;
1466 unsigned char old_id;
1467 unsigned long flags;
1468
1469 /*
1470 * This is broken; anything with a real cpu count has to
1471 * circumvent this idiocy regardless.
1472 */
1473 apic->ioapic_phys_id_map(&phys_cpu_present_map, &phys_id_present_map);
1474
1475 /*
1476 * Set the IOAPIC ID to the value stored in the MPC table.
1477 */
1478 for_each_ioapic(ioapic_idx) {
1479 /* Read the register 0 value */
1480 raw_spin_lock_irqsave(&ioapic_lock, flags);
1481 reg_00.raw = io_apic_read(ioapic_idx, 0);
1482 raw_spin_unlock_irqrestore(&ioapic_lock, flags);
1483
1484 old_id = mpc_ioapic_id(ioapic_idx);
1485
1486 if (mpc_ioapic_id(ioapic_idx) >= get_physical_broadcast()) {
1487 printk(KERN_ERR "BIOS bug, IO-APIC#%d ID is %d in the MPC table!...\n",
1488 ioapic_idx, mpc_ioapic_id(ioapic_idx));
1489 printk(KERN_ERR "... fixing up to %d. (tell your hw vendor)\n",
1490 reg_00.bits.ID);
1491 ioapics[ioapic_idx].mp_config.apicid = reg_00.bits.ID;
1492 }
1493
1494 /*
1495 * Sanity check, is the ID really free? Every APIC in a
1496 * system must have a unique ID or we get lots of nice
1497 * 'stuck on smp_invalidate_needed IPI wait' messages.
1498 */
1499 if (apic->check_apicid_used(&phys_id_present_map,
1500 mpc_ioapic_id(ioapic_idx))) {
1501 printk(KERN_ERR "BIOS bug, IO-APIC#%d ID %d is already used!...\n",
1502 ioapic_idx, mpc_ioapic_id(ioapic_idx));
1503 for (i = 0; i < get_physical_broadcast(); i++)
1504 if (!physid_isset(i, phys_id_present_map))
1505 break;
1506 if (i >= get_physical_broadcast())
1507 panic("Max APIC ID exceeded!\n");
1508 printk(KERN_ERR "... fixing up to %d. (tell your hw vendor)\n",
1509 i);
1510 physid_set(i, phys_id_present_map);
1511 ioapics[ioapic_idx].mp_config.apicid = i;
1512 } else {
1513 physid_mask_t tmp;
1514 apic->apicid_to_cpu_present(mpc_ioapic_id(ioapic_idx),
1515 &tmp);
1516 apic_printk(APIC_VERBOSE, "Setting %d in the "
1517 "phys_id_present_map\n",
1518 mpc_ioapic_id(ioapic_idx));
1519 physids_or(phys_id_present_map, phys_id_present_map, tmp);
1520 }
1521
1522 /*
1523 * We need to adjust the IRQ routing table
1524 * if the ID changed.
1525 */
1526 if (old_id != mpc_ioapic_id(ioapic_idx))
1527 for (i = 0; i < mp_irq_entries; i++)
1528 if (mp_irqs[i].dstapic == old_id)
1529 mp_irqs[i].dstapic
1530 = mpc_ioapic_id(ioapic_idx);
1531
1532 /*
1533 * Update the ID register according to the right value
1534 * from the MPC table if they are different.
1535 */
1536 if (mpc_ioapic_id(ioapic_idx) == reg_00.bits.ID)
1537 continue;
1538
1539 apic_printk(APIC_VERBOSE, KERN_INFO
1540 "...changing IO-APIC physical APIC ID to %d ...",
1541 mpc_ioapic_id(ioapic_idx));
1542
1543 reg_00.bits.ID = mpc_ioapic_id(ioapic_idx);
1544 raw_spin_lock_irqsave(&ioapic_lock, flags);
1545 io_apic_write(ioapic_idx, 0, reg_00.raw);
1546 raw_spin_unlock_irqrestore(&ioapic_lock, flags);
1547
1548 /*
1549 * Sanity check
1550 */
1551 raw_spin_lock_irqsave(&ioapic_lock, flags);
1552 reg_00.raw = io_apic_read(ioapic_idx, 0);
1553 raw_spin_unlock_irqrestore(&ioapic_lock, flags);
1554 if (reg_00.bits.ID != mpc_ioapic_id(ioapic_idx))
1555 pr_cont("could not set ID!\n");
1556 else
1557 apic_printk(APIC_VERBOSE, " ok.\n");
1558 }
1559}
1560
1561void __init setup_ioapic_ids_from_mpc(void)
1562{
1563
1564 if (acpi_ioapic)
1565 return;
1566 /*
1567 * Don't check I/O APIC IDs for xAPIC systems. They have
1568 * no meaning without the serial APIC bus.
1569 */
1570 if (!(boot_cpu_data.x86_vendor == X86_VENDOR_INTEL)
1571 || APIC_XAPIC(boot_cpu_apic_version))
1572 return;
1573 setup_ioapic_ids_from_mpc_nocheck();
1574}
1575#endif
1576
1577int no_timer_check __initdata;
1578
1579static int __init notimercheck(char *s)
1580{
1581 no_timer_check = 1;
1582 return 1;
1583}
1584__setup("no_timer_check", notimercheck);
1585
1586static void __init delay_with_tsc(void)
1587{
1588 unsigned long long start, now;
1589 unsigned long end = jiffies + 4;
1590
1591 start = rdtsc();
1592
1593 /*
1594 * We don't know the TSC frequency yet, but waiting for
1595 * 40000000000/HZ TSC cycles is safe:
1596 * 4 GHz == 10 jiffies
1597 * 1 GHz == 40 jiffies
1598 */
1599 do {
1600 rep_nop();
1601 now = rdtsc();
1602 } while ((now - start) < 40000000000ULL / HZ &&
1603 time_before_eq(jiffies, end));
1604}
1605
1606static void __init delay_without_tsc(void)
1607{
1608 unsigned long end = jiffies + 4;
1609 int band = 1;
1610
1611 /*
1612 * We don't know any frequency yet, but waiting for
1613 * 40940000000/HZ cycles is safe:
1614 * 4 GHz == 10 jiffies
1615 * 1 GHz == 40 jiffies
1616 * 1 << 1 + 1 << 2 +...+ 1 << 11 = 4094
1617 */
1618 do {
1619 __delay(((1U << band++) * 10000000UL) / HZ);
1620 } while (band < 12 && time_before_eq(jiffies, end));
1621}
1622
1623/*
1624 * There is a nasty bug in some older SMP boards, their mptable lies
1625 * about the timer IRQ. We do the following to work around the situation:
1626 *
1627 * - timer IRQ defaults to IO-APIC IRQ
1628 * - if this function detects that timer IRQs are defunct, then we fall
1629 * back to ISA timer IRQs
1630 */
1631static int __init timer_irq_works(void)
1632{
1633 unsigned long t1 = jiffies;
1634
1635 if (no_timer_check)
1636 return 1;
1637
1638 local_irq_enable();
1639 if (boot_cpu_has(X86_FEATURE_TSC))
1640 delay_with_tsc();
1641 else
1642 delay_without_tsc();
1643
1644 /*
1645 * Expect a few ticks at least, to be sure some possible
1646 * glue logic does not lock up after one or two first
1647 * ticks in a non-ExtINT mode. Also the local APIC
1648 * might have cached one ExtINT interrupt. Finally, at
1649 * least one tick may be lost due to delays.
1650 */
1651
1652 local_irq_disable();
1653
1654 /* Did jiffies advance? */
1655 return time_after(jiffies, t1 + 4);
1656}
1657
1658/*
1659 * In the SMP+IOAPIC case it might happen that there are an unspecified
1660 * number of pending IRQ events unhandled. These cases are very rare,
1661 * so we 'resend' these IRQs via IPIs, to the same CPU. It's much
1662 * better to do it this way as thus we do not have to be aware of
1663 * 'pending' interrupts in the IRQ path, except at this point.
1664 */
1665/*
1666 * Edge triggered needs to resend any interrupt
1667 * that was delayed but this is now handled in the device
1668 * independent code.
1669 */
1670
1671/*
1672 * Starting up a edge-triggered IO-APIC interrupt is
1673 * nasty - we need to make sure that we get the edge.
1674 * If it is already asserted for some reason, we need
1675 * return 1 to indicate that is was pending.
1676 *
1677 * This is not complete - we should be able to fake
1678 * an edge even if it isn't on the 8259A...
1679 */
1680static unsigned int startup_ioapic_irq(struct irq_data *data)
1681{
1682 int was_pending = 0, irq = data->irq;
1683 unsigned long flags;
1684
1685 raw_spin_lock_irqsave(&ioapic_lock, flags);
1686 if (irq < nr_legacy_irqs()) {
1687 legacy_pic->mask(irq);
1688 if (legacy_pic->irq_pending(irq))
1689 was_pending = 1;
1690 }
1691 __unmask_ioapic(data->chip_data);
1692 raw_spin_unlock_irqrestore(&ioapic_lock, flags);
1693
1694 return was_pending;
1695}
1696
1697atomic_t irq_mis_count;
1698
1699#ifdef CONFIG_GENERIC_PENDING_IRQ
1700static bool io_apic_level_ack_pending(struct mp_chip_data *data)
1701{
1702 struct irq_pin_list *entry;
1703 unsigned long flags;
1704
1705 raw_spin_lock_irqsave(&ioapic_lock, flags);
1706 for_each_irq_pin(entry, data->irq_2_pin) {
1707 struct IO_APIC_route_entry e;
1708 int pin;
1709
1710 pin = entry->pin;
1711 e.w1 = io_apic_read(entry->apic, 0x10 + pin*2);
1712 /* Is the remote IRR bit set? */
1713 if (e.irr) {
1714 raw_spin_unlock_irqrestore(&ioapic_lock, flags);
1715 return true;
1716 }
1717 }
1718 raw_spin_unlock_irqrestore(&ioapic_lock, flags);
1719
1720 return false;
1721}
1722
1723static inline bool ioapic_prepare_move(struct irq_data *data)
1724{
1725 /* If we are moving the IRQ we need to mask it */
1726 if (unlikely(irqd_is_setaffinity_pending(data))) {
1727 if (!irqd_irq_masked(data))
1728 mask_ioapic_irq(data);
1729 return true;
1730 }
1731 return false;
1732}
1733
1734static inline void ioapic_finish_move(struct irq_data *data, bool moveit)
1735{
1736 if (unlikely(moveit)) {
1737 /* Only migrate the irq if the ack has been received.
1738 *
1739 * On rare occasions the broadcast level triggered ack gets
1740 * delayed going to ioapics, and if we reprogram the
1741 * vector while Remote IRR is still set the irq will never
1742 * fire again.
1743 *
1744 * To prevent this scenario we read the Remote IRR bit
1745 * of the ioapic. This has two effects.
1746 * - On any sane system the read of the ioapic will
1747 * flush writes (and acks) going to the ioapic from
1748 * this cpu.
1749 * - We get to see if the ACK has actually been delivered.
1750 *
1751 * Based on failed experiments of reprogramming the
1752 * ioapic entry from outside of irq context starting
1753 * with masking the ioapic entry and then polling until
1754 * Remote IRR was clear before reprogramming the
1755 * ioapic I don't trust the Remote IRR bit to be
1756 * completely accurate.
1757 *
1758 * However there appears to be no other way to plug
1759 * this race, so if the Remote IRR bit is not
1760 * accurate and is causing problems then it is a hardware bug
1761 * and you can go talk to the chipset vendor about it.
1762 */
1763 if (!io_apic_level_ack_pending(data->chip_data))
1764 irq_move_masked_irq(data);
1765 /* If the IRQ is masked in the core, leave it: */
1766 if (!irqd_irq_masked(data))
1767 unmask_ioapic_irq(data);
1768 }
1769}
1770#else
1771static inline bool ioapic_prepare_move(struct irq_data *data)
1772{
1773 return false;
1774}
1775static inline void ioapic_finish_move(struct irq_data *data, bool moveit)
1776{
1777}
1778#endif
1779
1780static void ioapic_ack_level(struct irq_data *irq_data)
1781{
1782 struct irq_cfg *cfg = irqd_cfg(irq_data);
1783 unsigned long v;
1784 bool moveit;
1785 int i;
1786
1787 irq_complete_move(cfg);
1788 moveit = ioapic_prepare_move(irq_data);
1789
1790 /*
1791 * It appears there is an erratum which affects at least version 0x11
1792 * of I/O APIC (that's the 82093AA and cores integrated into various
1793 * chipsets). Under certain conditions a level-triggered interrupt is
1794 * erroneously delivered as edge-triggered one but the respective IRR
1795 * bit gets set nevertheless. As a result the I/O unit expects an EOI
1796 * message but it will never arrive and further interrupts are blocked
1797 * from the source. The exact reason is so far unknown, but the
1798 * phenomenon was observed when two consecutive interrupt requests
1799 * from a given source get delivered to the same CPU and the source is
1800 * temporarily disabled in between.
1801 *
1802 * A workaround is to simulate an EOI message manually. We achieve it
1803 * by setting the trigger mode to edge and then to level when the edge
1804 * trigger mode gets detected in the TMR of a local APIC for a
1805 * level-triggered interrupt. We mask the source for the time of the
1806 * operation to prevent an edge-triggered interrupt escaping meanwhile.
1807 * The idea is from Manfred Spraul. --macro
1808 *
1809 * Also in the case when cpu goes offline, fixup_irqs() will forward
1810 * any unhandled interrupt on the offlined cpu to the new cpu
1811 * destination that is handling the corresponding interrupt. This
1812 * interrupt forwarding is done via IPI's. Hence, in this case also
1813 * level-triggered io-apic interrupt will be seen as an edge
1814 * interrupt in the IRR. And we can't rely on the cpu's EOI
1815 * to be broadcasted to the IO-APIC's which will clear the remoteIRR
1816 * corresponding to the level-triggered interrupt. Hence on IO-APIC's
1817 * supporting EOI register, we do an explicit EOI to clear the
1818 * remote IRR and on IO-APIC's which don't have an EOI register,
1819 * we use the above logic (mask+edge followed by unmask+level) from
1820 * Manfred Spraul to clear the remote IRR.
1821 */
1822 i = cfg->vector;
1823 v = apic_read(APIC_TMR + ((i & ~0x1f) >> 1));
1824
1825 /*
1826 * We must acknowledge the irq before we move it or the acknowledge will
1827 * not propagate properly.
1828 */
1829 ack_APIC_irq();
1830
1831 /*
1832 * Tail end of clearing remote IRR bit (either by delivering the EOI
1833 * message via io-apic EOI register write or simulating it using
1834 * mask+edge followed by unmask+level logic) manually when the
1835 * level triggered interrupt is seen as the edge triggered interrupt
1836 * at the cpu.
1837 */
1838 if (!(v & (1 << (i & 0x1f)))) {
1839 atomic_inc(&irq_mis_count);
1840 eoi_ioapic_pin(cfg->vector, irq_data->chip_data);
1841 }
1842
1843 ioapic_finish_move(irq_data, moveit);
1844}
1845
1846static void ioapic_ir_ack_level(struct irq_data *irq_data)
1847{
1848 struct mp_chip_data *data = irq_data->chip_data;
1849
1850 /*
1851 * Intr-remapping uses pin number as the virtual vector
1852 * in the RTE. Actual vector is programmed in
1853 * intr-remapping table entry. Hence for the io-apic
1854 * EOI we use the pin number.
1855 */
1856 apic_ack_irq(irq_data);
1857 eoi_ioapic_pin(data->entry.vector, data);
1858}
1859
1860/*
1861 * The I/OAPIC is just a device for generating MSI messages from legacy
1862 * interrupt pins. Various fields of the RTE translate into bits of the
1863 * resulting MSI which had a historical meaning.
1864 *
1865 * With interrupt remapping, many of those bits have different meanings
1866 * in the underlying MSI, but the way that the I/OAPIC transforms them
1867 * from its RTE to the MSI message is the same. This function allows
1868 * the parent IRQ domain to compose the MSI message, then takes the
1869 * relevant bits to put them in the appropriate places in the RTE in
1870 * order to generate that message when the IRQ happens.
1871 *
1872 * The setup here relies on a preconfigured route entry (is_level,
1873 * active_low, masked) because the parent domain is merely composing the
1874 * generic message routing information which is used for the MSI.
1875 */
1876static void ioapic_setup_msg_from_msi(struct irq_data *irq_data,
1877 struct IO_APIC_route_entry *entry)
1878{
1879 struct msi_msg msg;
1880
1881 /* Let the parent domain compose the MSI message */
1882 irq_chip_compose_msi_msg(irq_data, &msg);
1883
1884 /*
1885 * - Real vector
1886 * - DMAR/IR: 8bit subhandle (ioapic.pin)
1887 * - AMD/IR: 8bit IRTE index
1888 */
1889 entry->vector = msg.arch_data.vector;
1890 /* Delivery mode (for DMAR/IR all 0) */
1891 entry->delivery_mode = msg.arch_data.delivery_mode;
1892 /* Destination mode or DMAR/IR index bit 15 */
1893 entry->dest_mode_logical = msg.arch_addr_lo.dest_mode_logical;
1894 /* DMAR/IR: 1, 0 for all other modes */
1895 entry->ir_format = msg.arch_addr_lo.dmar_format;
1896 /*
1897 * - DMAR/IR: index bit 0-14.
1898 *
1899 * - Virt: If the host supports x2apic without a virtualized IR
1900 * unit then bit 0-6 of dmar_index_0_14 are providing bit
1901 * 8-14 of the destination id.
1902 *
1903 * All other modes have bit 0-6 of dmar_index_0_14 cleared and the
1904 * topmost 8 bits are destination id bit 0-7 (entry::destid_0_7).
1905 */
1906 entry->ir_index_0_14 = msg.arch_addr_lo.dmar_index_0_14;
1907}
1908
1909static void ioapic_configure_entry(struct irq_data *irqd)
1910{
1911 struct mp_chip_data *mpd = irqd->chip_data;
1912 struct irq_pin_list *entry;
1913
1914 ioapic_setup_msg_from_msi(irqd, &mpd->entry);
1915
1916 for_each_irq_pin(entry, mpd->irq_2_pin)
1917 __ioapic_write_entry(entry->apic, entry->pin, mpd->entry);
1918}
1919
1920static int ioapic_set_affinity(struct irq_data *irq_data,
1921 const struct cpumask *mask, bool force)
1922{
1923 struct irq_data *parent = irq_data->parent_data;
1924 unsigned long flags;
1925 int ret;
1926
1927 ret = parent->chip->irq_set_affinity(parent, mask, force);
1928 raw_spin_lock_irqsave(&ioapic_lock, flags);
1929 if (ret >= 0 && ret != IRQ_SET_MASK_OK_DONE)
1930 ioapic_configure_entry(irq_data);
1931 raw_spin_unlock_irqrestore(&ioapic_lock, flags);
1932
1933 return ret;
1934}
1935
1936/*
1937 * Interrupt shutdown masks the ioapic pin, but the interrupt might already
1938 * be in flight, but not yet serviced by the target CPU. That means
1939 * __synchronize_hardirq() would return and claim that everything is calmed
1940 * down. So free_irq() would proceed and deactivate the interrupt and free
1941 * resources.
1942 *
1943 * Once the target CPU comes around to service it it will find a cleared
1944 * vector and complain. While the spurious interrupt is harmless, the full
1945 * release of resources might prevent the interrupt from being acknowledged
1946 * which keeps the hardware in a weird state.
1947 *
1948 * Verify that the corresponding Remote-IRR bits are clear.
1949 */
1950static int ioapic_irq_get_chip_state(struct irq_data *irqd,
1951 enum irqchip_irq_state which,
1952 bool *state)
1953{
1954 struct mp_chip_data *mcd = irqd->chip_data;
1955 struct IO_APIC_route_entry rentry;
1956 struct irq_pin_list *p;
1957
1958 if (which != IRQCHIP_STATE_ACTIVE)
1959 return -EINVAL;
1960
1961 *state = false;
1962 raw_spin_lock(&ioapic_lock);
1963 for_each_irq_pin(p, mcd->irq_2_pin) {
1964 rentry = __ioapic_read_entry(p->apic, p->pin);
1965 /*
1966 * The remote IRR is only valid in level trigger mode. It's
1967 * meaning is undefined for edge triggered interrupts and
1968 * irrelevant because the IO-APIC treats them as fire and
1969 * forget.
1970 */
1971 if (rentry.irr && rentry.is_level) {
1972 *state = true;
1973 break;
1974 }
1975 }
1976 raw_spin_unlock(&ioapic_lock);
1977 return 0;
1978}
1979
1980static struct irq_chip ioapic_chip __read_mostly = {
1981 .name = "IO-APIC",
1982 .irq_startup = startup_ioapic_irq,
1983 .irq_mask = mask_ioapic_irq,
1984 .irq_unmask = unmask_ioapic_irq,
1985 .irq_ack = irq_chip_ack_parent,
1986 .irq_eoi = ioapic_ack_level,
1987 .irq_set_affinity = ioapic_set_affinity,
1988 .irq_retrigger = irq_chip_retrigger_hierarchy,
1989 .irq_get_irqchip_state = ioapic_irq_get_chip_state,
1990 .flags = IRQCHIP_SKIP_SET_WAKE |
1991 IRQCHIP_AFFINITY_PRE_STARTUP,
1992};
1993
1994static struct irq_chip ioapic_ir_chip __read_mostly = {
1995 .name = "IR-IO-APIC",
1996 .irq_startup = startup_ioapic_irq,
1997 .irq_mask = mask_ioapic_irq,
1998 .irq_unmask = unmask_ioapic_irq,
1999 .irq_ack = irq_chip_ack_parent,
2000 .irq_eoi = ioapic_ir_ack_level,
2001 .irq_set_affinity = ioapic_set_affinity,
2002 .irq_retrigger = irq_chip_retrigger_hierarchy,
2003 .irq_get_irqchip_state = ioapic_irq_get_chip_state,
2004 .flags = IRQCHIP_SKIP_SET_WAKE |
2005 IRQCHIP_AFFINITY_PRE_STARTUP,
2006};
2007
2008static inline void init_IO_APIC_traps(void)
2009{
2010 struct irq_cfg *cfg;
2011 unsigned int irq;
2012
2013 for_each_active_irq(irq) {
2014 cfg = irq_cfg(irq);
2015 if (IO_APIC_IRQ(irq) && cfg && !cfg->vector) {
2016 /*
2017 * Hmm.. We don't have an entry for this,
2018 * so default to an old-fashioned 8259
2019 * interrupt if we can..
2020 */
2021 if (irq < nr_legacy_irqs())
2022 legacy_pic->make_irq(irq);
2023 else
2024 /* Strange. Oh, well.. */
2025 irq_set_chip(irq, &no_irq_chip);
2026 }
2027 }
2028}
2029
2030/*
2031 * The local APIC irq-chip implementation:
2032 */
2033
2034static void mask_lapic_irq(struct irq_data *data)
2035{
2036 unsigned long v;
2037
2038 v = apic_read(APIC_LVT0);
2039 apic_write(APIC_LVT0, v | APIC_LVT_MASKED);
2040}
2041
2042static void unmask_lapic_irq(struct irq_data *data)
2043{
2044 unsigned long v;
2045
2046 v = apic_read(APIC_LVT0);
2047 apic_write(APIC_LVT0, v & ~APIC_LVT_MASKED);
2048}
2049
2050static void ack_lapic_irq(struct irq_data *data)
2051{
2052 ack_APIC_irq();
2053}
2054
2055static struct irq_chip lapic_chip __read_mostly = {
2056 .name = "local-APIC",
2057 .irq_mask = mask_lapic_irq,
2058 .irq_unmask = unmask_lapic_irq,
2059 .irq_ack = ack_lapic_irq,
2060};
2061
2062static void lapic_register_intr(int irq)
2063{
2064 irq_clear_status_flags(irq, IRQ_LEVEL);
2065 irq_set_chip_and_handler_name(irq, &lapic_chip, handle_edge_irq,
2066 "edge");
2067}
2068
2069/*
2070 * This looks a bit hackish but it's about the only one way of sending
2071 * a few INTA cycles to 8259As and any associated glue logic. ICR does
2072 * not support the ExtINT mode, unfortunately. We need to send these
2073 * cycles as some i82489DX-based boards have glue logic that keeps the
2074 * 8259A interrupt line asserted until INTA. --macro
2075 */
2076static inline void __init unlock_ExtINT_logic(void)
2077{
2078 int apic, pin, i;
2079 struct IO_APIC_route_entry entry0, entry1;
2080 unsigned char save_control, save_freq_select;
2081 u32 apic_id;
2082
2083 pin = find_isa_irq_pin(8, mp_INT);
2084 if (pin == -1) {
2085 WARN_ON_ONCE(1);
2086 return;
2087 }
2088 apic = find_isa_irq_apic(8, mp_INT);
2089 if (apic == -1) {
2090 WARN_ON_ONCE(1);
2091 return;
2092 }
2093
2094 entry0 = ioapic_read_entry(apic, pin);
2095 clear_IO_APIC_pin(apic, pin);
2096
2097 apic_id = hard_smp_processor_id();
2098 memset(&entry1, 0, sizeof(entry1));
2099
2100 entry1.dest_mode_logical = true;
2101 entry1.masked = false;
2102 entry1.destid_0_7 = apic_id & 0xFF;
2103 entry1.virt_destid_8_14 = apic_id >> 8;
2104 entry1.delivery_mode = APIC_DELIVERY_MODE_EXTINT;
2105 entry1.active_low = entry0.active_low;
2106 entry1.is_level = false;
2107 entry1.vector = 0;
2108
2109 ioapic_write_entry(apic, pin, entry1);
2110
2111 save_control = CMOS_READ(RTC_CONTROL);
2112 save_freq_select = CMOS_READ(RTC_FREQ_SELECT);
2113 CMOS_WRITE((save_freq_select & ~RTC_RATE_SELECT) | 0x6,
2114 RTC_FREQ_SELECT);
2115 CMOS_WRITE(save_control | RTC_PIE, RTC_CONTROL);
2116
2117 i = 100;
2118 while (i-- > 0) {
2119 mdelay(10);
2120 if ((CMOS_READ(RTC_INTR_FLAGS) & RTC_PF) == RTC_PF)
2121 i -= 10;
2122 }
2123
2124 CMOS_WRITE(save_control, RTC_CONTROL);
2125 CMOS_WRITE(save_freq_select, RTC_FREQ_SELECT);
2126 clear_IO_APIC_pin(apic, pin);
2127
2128 ioapic_write_entry(apic, pin, entry0);
2129}
2130
2131static int disable_timer_pin_1 __initdata;
2132/* Actually the next is obsolete, but keep it for paranoid reasons -AK */
2133static int __init disable_timer_pin_setup(char *arg)
2134{
2135 disable_timer_pin_1 = 1;
2136 return 0;
2137}
2138early_param("disable_timer_pin_1", disable_timer_pin_setup);
2139
2140static int mp_alloc_timer_irq(int ioapic, int pin)
2141{
2142 int irq = -1;
2143 struct irq_domain *domain = mp_ioapic_irqdomain(ioapic);
2144
2145 if (domain) {
2146 struct irq_alloc_info info;
2147
2148 ioapic_set_alloc_attr(&info, NUMA_NO_NODE, 0, 0);
2149 info.devid = mpc_ioapic_id(ioapic);
2150 info.ioapic.pin = pin;
2151 mutex_lock(&ioapic_mutex);
2152 irq = alloc_isa_irq_from_domain(domain, 0, ioapic, pin, &info);
2153 mutex_unlock(&ioapic_mutex);
2154 }
2155
2156 return irq;
2157}
2158
2159/*
2160 * This code may look a bit paranoid, but it's supposed to cooperate with
2161 * a wide range of boards and BIOS bugs. Fortunately only the timer IRQ
2162 * is so screwy. Thanks to Brian Perkins for testing/hacking this beast
2163 * fanatically on his truly buggy board.
2164 *
2165 * FIXME: really need to revamp this for all platforms.
2166 */
2167static inline void __init check_timer(void)
2168{
2169 struct irq_data *irq_data = irq_get_irq_data(0);
2170 struct mp_chip_data *data = irq_data->chip_data;
2171 struct irq_cfg *cfg = irqd_cfg(irq_data);
2172 int node = cpu_to_node(0);
2173 int apic1, pin1, apic2, pin2;
2174 int no_pin1 = 0;
2175
2176 if (!global_clock_event)
2177 return;
2178
2179 local_irq_disable();
2180
2181 /*
2182 * get/set the timer IRQ vector:
2183 */
2184 legacy_pic->mask(0);
2185
2186 /*
2187 * As IRQ0 is to be enabled in the 8259A, the virtual
2188 * wire has to be disabled in the local APIC. Also
2189 * timer interrupts need to be acknowledged manually in
2190 * the 8259A for the i82489DX when using the NMI
2191 * watchdog as that APIC treats NMIs as level-triggered.
2192 * The AEOI mode will finish them in the 8259A
2193 * automatically.
2194 */
2195 apic_write(APIC_LVT0, APIC_LVT_MASKED | APIC_DM_EXTINT);
2196 legacy_pic->init(1);
2197
2198 pin1 = find_isa_irq_pin(0, mp_INT);
2199 apic1 = find_isa_irq_apic(0, mp_INT);
2200 pin2 = ioapic_i8259.pin;
2201 apic2 = ioapic_i8259.apic;
2202
2203 apic_printk(APIC_QUIET, KERN_INFO "..TIMER: vector=0x%02X "
2204 "apic1=%d pin1=%d apic2=%d pin2=%d\n",
2205 cfg->vector, apic1, pin1, apic2, pin2);
2206
2207 /*
2208 * Some BIOS writers are clueless and report the ExtINTA
2209 * I/O APIC input from the cascaded 8259A as the timer
2210 * interrupt input. So just in case, if only one pin
2211 * was found above, try it both directly and through the
2212 * 8259A.
2213 */
2214 if (pin1 == -1) {
2215 panic_if_irq_remap("BIOS bug: timer not connected to IO-APIC");
2216 pin1 = pin2;
2217 apic1 = apic2;
2218 no_pin1 = 1;
2219 } else if (pin2 == -1) {
2220 pin2 = pin1;
2221 apic2 = apic1;
2222 }
2223
2224 if (pin1 != -1) {
2225 /* Ok, does IRQ0 through the IOAPIC work? */
2226 if (no_pin1) {
2227 mp_alloc_timer_irq(apic1, pin1);
2228 } else {
2229 /*
2230 * for edge trigger, it's already unmasked,
2231 * so only need to unmask if it is level-trigger
2232 * do we really have level trigger timer?
2233 */
2234 int idx = find_irq_entry(apic1, pin1, mp_INT);
2235
2236 if (idx != -1 && irq_is_level(idx))
2237 unmask_ioapic_irq(irq_get_irq_data(0));
2238 }
2239 irq_domain_deactivate_irq(irq_data);
2240 irq_domain_activate_irq(irq_data, false);
2241 if (timer_irq_works()) {
2242 if (disable_timer_pin_1 > 0)
2243 clear_IO_APIC_pin(0, pin1);
2244 goto out;
2245 }
2246 panic_if_irq_remap("timer doesn't work through Interrupt-remapped IO-APIC");
2247 clear_IO_APIC_pin(apic1, pin1);
2248 if (!no_pin1)
2249 apic_printk(APIC_QUIET, KERN_ERR "..MP-BIOS bug: "
2250 "8254 timer not connected to IO-APIC\n");
2251
2252 apic_printk(APIC_QUIET, KERN_INFO "...trying to set up timer "
2253 "(IRQ0) through the 8259A ...\n");
2254 apic_printk(APIC_QUIET, KERN_INFO
2255 "..... (found apic %d pin %d) ...\n", apic2, pin2);
2256 /*
2257 * legacy devices should be connected to IO APIC #0
2258 */
2259 replace_pin_at_irq_node(data, node, apic1, pin1, apic2, pin2);
2260 irq_domain_deactivate_irq(irq_data);
2261 irq_domain_activate_irq(irq_data, false);
2262 legacy_pic->unmask(0);
2263 if (timer_irq_works()) {
2264 apic_printk(APIC_QUIET, KERN_INFO "....... works.\n");
2265 goto out;
2266 }
2267 /*
2268 * Cleanup, just in case ...
2269 */
2270 legacy_pic->mask(0);
2271 clear_IO_APIC_pin(apic2, pin2);
2272 apic_printk(APIC_QUIET, KERN_INFO "....... failed.\n");
2273 }
2274
2275 apic_printk(APIC_QUIET, KERN_INFO
2276 "...trying to set up timer as Virtual Wire IRQ...\n");
2277
2278 lapic_register_intr(0);
2279 apic_write(APIC_LVT0, APIC_DM_FIXED | cfg->vector); /* Fixed mode */
2280 legacy_pic->unmask(0);
2281
2282 if (timer_irq_works()) {
2283 apic_printk(APIC_QUIET, KERN_INFO "..... works.\n");
2284 goto out;
2285 }
2286 legacy_pic->mask(0);
2287 apic_write(APIC_LVT0, APIC_LVT_MASKED | APIC_DM_FIXED | cfg->vector);
2288 apic_printk(APIC_QUIET, KERN_INFO "..... failed.\n");
2289
2290 apic_printk(APIC_QUIET, KERN_INFO
2291 "...trying to set up timer as ExtINT IRQ...\n");
2292
2293 legacy_pic->init(0);
2294 legacy_pic->make_irq(0);
2295 apic_write(APIC_LVT0, APIC_DM_EXTINT);
2296 legacy_pic->unmask(0);
2297
2298 unlock_ExtINT_logic();
2299
2300 if (timer_irq_works()) {
2301 apic_printk(APIC_QUIET, KERN_INFO "..... works.\n");
2302 goto out;
2303 }
2304 apic_printk(APIC_QUIET, KERN_INFO "..... failed :(.\n");
2305 if (apic_is_x2apic_enabled())
2306 apic_printk(APIC_QUIET, KERN_INFO
2307 "Perhaps problem with the pre-enabled x2apic mode\n"
2308 "Try booting with x2apic and interrupt-remapping disabled in the bios.\n");
2309 panic("IO-APIC + timer doesn't work! Boot with apic=debug and send a "
2310 "report. Then try booting with the 'noapic' option.\n");
2311out:
2312 local_irq_enable();
2313}
2314
2315/*
2316 * Traditionally ISA IRQ2 is the cascade IRQ, and is not available
2317 * to devices. However there may be an I/O APIC pin available for
2318 * this interrupt regardless. The pin may be left unconnected, but
2319 * typically it will be reused as an ExtINT cascade interrupt for
2320 * the master 8259A. In the MPS case such a pin will normally be
2321 * reported as an ExtINT interrupt in the MP table. With ACPI
2322 * there is no provision for ExtINT interrupts, and in the absence
2323 * of an override it would be treated as an ordinary ISA I/O APIC
2324 * interrupt, that is edge-triggered and unmasked by default. We
2325 * used to do this, but it caused problems on some systems because
2326 * of the NMI watchdog and sometimes IRQ0 of the 8254 timer using
2327 * the same ExtINT cascade interrupt to drive the local APIC of the
2328 * bootstrap processor. Therefore we refrain from routing IRQ2 to
2329 * the I/O APIC in all cases now. No actual device should request
2330 * it anyway. --macro
2331 */
2332#define PIC_IRQS (1UL << PIC_CASCADE_IR)
2333
2334static int mp_irqdomain_create(int ioapic)
2335{
2336 struct irq_domain *parent;
2337 int hwirqs = mp_ioapic_pin_count(ioapic);
2338 struct ioapic *ip = &ioapics[ioapic];
2339 struct ioapic_domain_cfg *cfg = &ip->irqdomain_cfg;
2340 struct mp_ioapic_gsi *gsi_cfg = mp_ioapic_gsi_routing(ioapic);
2341 struct fwnode_handle *fn;
2342 struct irq_fwspec fwspec;
2343
2344 if (cfg->type == IOAPIC_DOMAIN_INVALID)
2345 return 0;
2346
2347 /* Handle device tree enumerated APICs proper */
2348 if (cfg->dev) {
2349 fn = of_node_to_fwnode(cfg->dev);
2350 } else {
2351 fn = irq_domain_alloc_named_id_fwnode("IO-APIC", mpc_ioapic_id(ioapic));
2352 if (!fn)
2353 return -ENOMEM;
2354 }
2355
2356 fwspec.fwnode = fn;
2357 fwspec.param_count = 1;
2358 fwspec.param[0] = mpc_ioapic_id(ioapic);
2359
2360 parent = irq_find_matching_fwspec(&fwspec, DOMAIN_BUS_ANY);
2361 if (!parent) {
2362 if (!cfg->dev)
2363 irq_domain_free_fwnode(fn);
2364 return -ENODEV;
2365 }
2366
2367 ip->irqdomain = irq_domain_create_linear(fn, hwirqs, cfg->ops,
2368 (void *)(long)ioapic);
2369
2370 if (!ip->irqdomain) {
2371 /* Release fw handle if it was allocated above */
2372 if (!cfg->dev)
2373 irq_domain_free_fwnode(fn);
2374 return -ENOMEM;
2375 }
2376
2377 ip->irqdomain->parent = parent;
2378
2379 if (cfg->type == IOAPIC_DOMAIN_LEGACY ||
2380 cfg->type == IOAPIC_DOMAIN_STRICT)
2381 ioapic_dynirq_base = max(ioapic_dynirq_base,
2382 gsi_cfg->gsi_end + 1);
2383
2384 return 0;
2385}
2386
2387static void ioapic_destroy_irqdomain(int idx)
2388{
2389 struct ioapic_domain_cfg *cfg = &ioapics[idx].irqdomain_cfg;
2390 struct fwnode_handle *fn = ioapics[idx].irqdomain->fwnode;
2391
2392 if (ioapics[idx].irqdomain) {
2393 irq_domain_remove(ioapics[idx].irqdomain);
2394 if (!cfg->dev)
2395 irq_domain_free_fwnode(fn);
2396 ioapics[idx].irqdomain = NULL;
2397 }
2398}
2399
2400void __init setup_IO_APIC(void)
2401{
2402 int ioapic;
2403
2404 if (skip_ioapic_setup || !nr_ioapics)
2405 return;
2406
2407 io_apic_irqs = nr_legacy_irqs() ? ~PIC_IRQS : ~0UL;
2408
2409 apic_printk(APIC_VERBOSE, "ENABLING IO-APIC IRQs\n");
2410 for_each_ioapic(ioapic)
2411 BUG_ON(mp_irqdomain_create(ioapic));
2412
2413 /*
2414 * Set up IO-APIC IRQ routing.
2415 */
2416 x86_init.mpparse.setup_ioapic_ids();
2417
2418 sync_Arb_IDs();
2419 setup_IO_APIC_irqs();
2420 init_IO_APIC_traps();
2421 if (nr_legacy_irqs())
2422 check_timer();
2423
2424 ioapic_initialized = 1;
2425}
2426
2427static void resume_ioapic_id(int ioapic_idx)
2428{
2429 unsigned long flags;
2430 union IO_APIC_reg_00 reg_00;
2431
2432 raw_spin_lock_irqsave(&ioapic_lock, flags);
2433 reg_00.raw = io_apic_read(ioapic_idx, 0);
2434 if (reg_00.bits.ID != mpc_ioapic_id(ioapic_idx)) {
2435 reg_00.bits.ID = mpc_ioapic_id(ioapic_idx);
2436 io_apic_write(ioapic_idx, 0, reg_00.raw);
2437 }
2438 raw_spin_unlock_irqrestore(&ioapic_lock, flags);
2439}
2440
2441static void ioapic_resume(void)
2442{
2443 int ioapic_idx;
2444
2445 for_each_ioapic_reverse(ioapic_idx)
2446 resume_ioapic_id(ioapic_idx);
2447
2448 restore_ioapic_entries();
2449}
2450
2451static struct syscore_ops ioapic_syscore_ops = {
2452 .suspend = save_ioapic_entries,
2453 .resume = ioapic_resume,
2454};
2455
2456static int __init ioapic_init_ops(void)
2457{
2458 register_syscore_ops(&ioapic_syscore_ops);
2459
2460 return 0;
2461}
2462
2463device_initcall(ioapic_init_ops);
2464
2465static int io_apic_get_redir_entries(int ioapic)
2466{
2467 union IO_APIC_reg_01 reg_01;
2468 unsigned long flags;
2469
2470 raw_spin_lock_irqsave(&ioapic_lock, flags);
2471 reg_01.raw = io_apic_read(ioapic, 1);
2472 raw_spin_unlock_irqrestore(&ioapic_lock, flags);
2473
2474 /* The register returns the maximum index redir index
2475 * supported, which is one less than the total number of redir
2476 * entries.
2477 */
2478 return reg_01.bits.entries + 1;
2479}
2480
2481unsigned int arch_dynirq_lower_bound(unsigned int from)
2482{
2483 /*
2484 * dmar_alloc_hwirq() may be called before setup_IO_APIC(), so use
2485 * gsi_top if ioapic_dynirq_base hasn't been initialized yet.
2486 */
2487 if (!ioapic_initialized)
2488 return gsi_top;
2489 /*
2490 * For DT enabled machines ioapic_dynirq_base is irrelevant and not
2491 * updated. So simply return @from if ioapic_dynirq_base == 0.
2492 */
2493 return ioapic_dynirq_base ? : from;
2494}
2495
2496#ifdef CONFIG_X86_32
2497static int io_apic_get_unique_id(int ioapic, int apic_id)
2498{
2499 union IO_APIC_reg_00 reg_00;
2500 static physid_mask_t apic_id_map = PHYSID_MASK_NONE;
2501 physid_mask_t tmp;
2502 unsigned long flags;
2503 int i = 0;
2504
2505 /*
2506 * The P4 platform supports up to 256 APIC IDs on two separate APIC
2507 * buses (one for LAPICs, one for IOAPICs), where predecessors only
2508 * supports up to 16 on one shared APIC bus.
2509 *
2510 * TBD: Expand LAPIC/IOAPIC support on P4-class systems to take full
2511 * advantage of new APIC bus architecture.
2512 */
2513
2514 if (physids_empty(apic_id_map))
2515 apic->ioapic_phys_id_map(&phys_cpu_present_map, &apic_id_map);
2516
2517 raw_spin_lock_irqsave(&ioapic_lock, flags);
2518 reg_00.raw = io_apic_read(ioapic, 0);
2519 raw_spin_unlock_irqrestore(&ioapic_lock, flags);
2520
2521 if (apic_id >= get_physical_broadcast()) {
2522 printk(KERN_WARNING "IOAPIC[%d]: Invalid apic_id %d, trying "
2523 "%d\n", ioapic, apic_id, reg_00.bits.ID);
2524 apic_id = reg_00.bits.ID;
2525 }
2526
2527 /*
2528 * Every APIC in a system must have a unique ID or we get lots of nice
2529 * 'stuck on smp_invalidate_needed IPI wait' messages.
2530 */
2531 if (apic->check_apicid_used(&apic_id_map, apic_id)) {
2532
2533 for (i = 0; i < get_physical_broadcast(); i++) {
2534 if (!apic->check_apicid_used(&apic_id_map, i))
2535 break;
2536 }
2537
2538 if (i == get_physical_broadcast())
2539 panic("Max apic_id exceeded!\n");
2540
2541 printk(KERN_WARNING "IOAPIC[%d]: apic_id %d already used, "
2542 "trying %d\n", ioapic, apic_id, i);
2543
2544 apic_id = i;
2545 }
2546
2547 apic->apicid_to_cpu_present(apic_id, &tmp);
2548 physids_or(apic_id_map, apic_id_map, tmp);
2549
2550 if (reg_00.bits.ID != apic_id) {
2551 reg_00.bits.ID = apic_id;
2552
2553 raw_spin_lock_irqsave(&ioapic_lock, flags);
2554 io_apic_write(ioapic, 0, reg_00.raw);
2555 reg_00.raw = io_apic_read(ioapic, 0);
2556 raw_spin_unlock_irqrestore(&ioapic_lock, flags);
2557
2558 /* Sanity check */
2559 if (reg_00.bits.ID != apic_id) {
2560 pr_err("IOAPIC[%d]: Unable to change apic_id!\n",
2561 ioapic);
2562 return -1;
2563 }
2564 }
2565
2566 apic_printk(APIC_VERBOSE, KERN_INFO
2567 "IOAPIC[%d]: Assigned apic_id %d\n", ioapic, apic_id);
2568
2569 return apic_id;
2570}
2571
2572static u8 io_apic_unique_id(int idx, u8 id)
2573{
2574 if ((boot_cpu_data.x86_vendor == X86_VENDOR_INTEL) &&
2575 !APIC_XAPIC(boot_cpu_apic_version))
2576 return io_apic_get_unique_id(idx, id);
2577 else
2578 return id;
2579}
2580#else
2581static u8 io_apic_unique_id(int idx, u8 id)
2582{
2583 union IO_APIC_reg_00 reg_00;
2584 DECLARE_BITMAP(used, 256);
2585 unsigned long flags;
2586 u8 new_id;
2587 int i;
2588
2589 bitmap_zero(used, 256);
2590 for_each_ioapic(i)
2591 __set_bit(mpc_ioapic_id(i), used);
2592
2593 /* Hand out the requested id if available */
2594 if (!test_bit(id, used))
2595 return id;
2596
2597 /*
2598 * Read the current id from the ioapic and keep it if
2599 * available.
2600 */
2601 raw_spin_lock_irqsave(&ioapic_lock, flags);
2602 reg_00.raw = io_apic_read(idx, 0);
2603 raw_spin_unlock_irqrestore(&ioapic_lock, flags);
2604 new_id = reg_00.bits.ID;
2605 if (!test_bit(new_id, used)) {
2606 apic_printk(APIC_VERBOSE, KERN_INFO
2607 "IOAPIC[%d]: Using reg apic_id %d instead of %d\n",
2608 idx, new_id, id);
2609 return new_id;
2610 }
2611
2612 /*
2613 * Get the next free id and write it to the ioapic.
2614 */
2615 new_id = find_first_zero_bit(used, 256);
2616 reg_00.bits.ID = new_id;
2617 raw_spin_lock_irqsave(&ioapic_lock, flags);
2618 io_apic_write(idx, 0, reg_00.raw);
2619 reg_00.raw = io_apic_read(idx, 0);
2620 raw_spin_unlock_irqrestore(&ioapic_lock, flags);
2621 /* Sanity check */
2622 BUG_ON(reg_00.bits.ID != new_id);
2623
2624 return new_id;
2625}
2626#endif
2627
2628static int io_apic_get_version(int ioapic)
2629{
2630 union IO_APIC_reg_01 reg_01;
2631 unsigned long flags;
2632
2633 raw_spin_lock_irqsave(&ioapic_lock, flags);
2634 reg_01.raw = io_apic_read(ioapic, 1);
2635 raw_spin_unlock_irqrestore(&ioapic_lock, flags);
2636
2637 return reg_01.bits.version;
2638}
2639
2640/*
2641 * This function updates target affinity of IOAPIC interrupts to include
2642 * the CPUs which came online during SMP bringup.
2643 */
2644#define IOAPIC_RESOURCE_NAME_SIZE 11
2645
2646static struct resource *ioapic_resources;
2647
2648static struct resource * __init ioapic_setup_resources(void)
2649{
2650 unsigned long n;
2651 struct resource *res;
2652 char *mem;
2653 int i;
2654
2655 if (nr_ioapics == 0)
2656 return NULL;
2657
2658 n = IOAPIC_RESOURCE_NAME_SIZE + sizeof(struct resource);
2659 n *= nr_ioapics;
2660
2661 mem = memblock_alloc(n, SMP_CACHE_BYTES);
2662 if (!mem)
2663 panic("%s: Failed to allocate %lu bytes\n", __func__, n);
2664 res = (void *)mem;
2665
2666 mem += sizeof(struct resource) * nr_ioapics;
2667
2668 for_each_ioapic(i) {
2669 res[i].name = mem;
2670 res[i].flags = IORESOURCE_MEM | IORESOURCE_BUSY;
2671 snprintf(mem, IOAPIC_RESOURCE_NAME_SIZE, "IOAPIC %u", i);
2672 mem += IOAPIC_RESOURCE_NAME_SIZE;
2673 ioapics[i].iomem_res = &res[i];
2674 }
2675
2676 ioapic_resources = res;
2677
2678 return res;
2679}
2680
2681static void io_apic_set_fixmap(enum fixed_addresses idx, phys_addr_t phys)
2682{
2683 pgprot_t flags = FIXMAP_PAGE_NOCACHE;
2684
2685 /*
2686 * Ensure fixmaps for IOAPIC MMIO respect memory encryption pgprot
2687 * bits, just like normal ioremap():
2688 */
2689 flags = pgprot_decrypted(flags);
2690
2691 __set_fixmap(idx, phys, flags);
2692}
2693
2694void __init io_apic_init_mappings(void)
2695{
2696 unsigned long ioapic_phys, idx = FIX_IO_APIC_BASE_0;
2697 struct resource *ioapic_res;
2698 int i;
2699
2700 ioapic_res = ioapic_setup_resources();
2701 for_each_ioapic(i) {
2702 if (smp_found_config) {
2703 ioapic_phys = mpc_ioapic_addr(i);
2704#ifdef CONFIG_X86_32
2705 if (!ioapic_phys) {
2706 printk(KERN_ERR
2707 "WARNING: bogus zero IO-APIC "
2708 "address found in MPTABLE, "
2709 "disabling IO/APIC support!\n");
2710 smp_found_config = 0;
2711 skip_ioapic_setup = 1;
2712 goto fake_ioapic_page;
2713 }
2714#endif
2715 } else {
2716#ifdef CONFIG_X86_32
2717fake_ioapic_page:
2718#endif
2719 ioapic_phys = (unsigned long)memblock_alloc(PAGE_SIZE,
2720 PAGE_SIZE);
2721 if (!ioapic_phys)
2722 panic("%s: Failed to allocate %lu bytes align=0x%lx\n",
2723 __func__, PAGE_SIZE, PAGE_SIZE);
2724 ioapic_phys = __pa(ioapic_phys);
2725 }
2726 io_apic_set_fixmap(idx, ioapic_phys);
2727 apic_printk(APIC_VERBOSE, "mapped IOAPIC to %08lx (%08lx)\n",
2728 __fix_to_virt(idx) + (ioapic_phys & ~PAGE_MASK),
2729 ioapic_phys);
2730 idx++;
2731
2732 ioapic_res->start = ioapic_phys;
2733 ioapic_res->end = ioapic_phys + IO_APIC_SLOT_SIZE - 1;
2734 ioapic_res++;
2735 }
2736}
2737
2738void __init ioapic_insert_resources(void)
2739{
2740 int i;
2741 struct resource *r = ioapic_resources;
2742
2743 if (!r) {
2744 if (nr_ioapics > 0)
2745 printk(KERN_ERR
2746 "IO APIC resources couldn't be allocated.\n");
2747 return;
2748 }
2749
2750 for_each_ioapic(i) {
2751 insert_resource(&iomem_resource, r);
2752 r++;
2753 }
2754}
2755
2756int mp_find_ioapic(u32 gsi)
2757{
2758 int i;
2759
2760 if (nr_ioapics == 0)
2761 return -1;
2762
2763 /* Find the IOAPIC that manages this GSI. */
2764 for_each_ioapic(i) {
2765 struct mp_ioapic_gsi *gsi_cfg = mp_ioapic_gsi_routing(i);
2766 if (gsi >= gsi_cfg->gsi_base && gsi <= gsi_cfg->gsi_end)
2767 return i;
2768 }
2769
2770 printk(KERN_ERR "ERROR: Unable to locate IOAPIC for GSI %d\n", gsi);
2771 return -1;
2772}
2773
2774int mp_find_ioapic_pin(int ioapic, u32 gsi)
2775{
2776 struct mp_ioapic_gsi *gsi_cfg;
2777
2778 if (WARN_ON(ioapic < 0))
2779 return -1;
2780
2781 gsi_cfg = mp_ioapic_gsi_routing(ioapic);
2782 if (WARN_ON(gsi > gsi_cfg->gsi_end))
2783 return -1;
2784
2785 return gsi - gsi_cfg->gsi_base;
2786}
2787
2788static int bad_ioapic_register(int idx)
2789{
2790 union IO_APIC_reg_00 reg_00;
2791 union IO_APIC_reg_01 reg_01;
2792 union IO_APIC_reg_02 reg_02;
2793
2794 reg_00.raw = io_apic_read(idx, 0);
2795 reg_01.raw = io_apic_read(idx, 1);
2796 reg_02.raw = io_apic_read(idx, 2);
2797
2798 if (reg_00.raw == -1 && reg_01.raw == -1 && reg_02.raw == -1) {
2799 pr_warn("I/O APIC 0x%x registers return all ones, skipping!\n",
2800 mpc_ioapic_addr(idx));
2801 return 1;
2802 }
2803
2804 return 0;
2805}
2806
2807static int find_free_ioapic_entry(void)
2808{
2809 int idx;
2810
2811 for (idx = 0; idx < MAX_IO_APICS; idx++)
2812 if (ioapics[idx].nr_registers == 0)
2813 return idx;
2814
2815 return MAX_IO_APICS;
2816}
2817
2818/**
2819 * mp_register_ioapic - Register an IOAPIC device
2820 * @id: hardware IOAPIC ID
2821 * @address: physical address of IOAPIC register area
2822 * @gsi_base: base of GSI associated with the IOAPIC
2823 * @cfg: configuration information for the IOAPIC
2824 */
2825int mp_register_ioapic(int id, u32 address, u32 gsi_base,
2826 struct ioapic_domain_cfg *cfg)
2827{
2828 bool hotplug = !!ioapic_initialized;
2829 struct mp_ioapic_gsi *gsi_cfg;
2830 int idx, ioapic, entries;
2831 u32 gsi_end;
2832
2833 if (!address) {
2834 pr_warn("Bogus (zero) I/O APIC address found, skipping!\n");
2835 return -EINVAL;
2836 }
2837 for_each_ioapic(ioapic)
2838 if (ioapics[ioapic].mp_config.apicaddr == address) {
2839 pr_warn("address 0x%x conflicts with IOAPIC%d\n",
2840 address, ioapic);
2841 return -EEXIST;
2842 }
2843
2844 idx = find_free_ioapic_entry();
2845 if (idx >= MAX_IO_APICS) {
2846 pr_warn("Max # of I/O APICs (%d) exceeded (found %d), skipping\n",
2847 MAX_IO_APICS, idx);
2848 return -ENOSPC;
2849 }
2850
2851 ioapics[idx].mp_config.type = MP_IOAPIC;
2852 ioapics[idx].mp_config.flags = MPC_APIC_USABLE;
2853 ioapics[idx].mp_config.apicaddr = address;
2854
2855 io_apic_set_fixmap(FIX_IO_APIC_BASE_0 + idx, address);
2856 if (bad_ioapic_register(idx)) {
2857 clear_fixmap(FIX_IO_APIC_BASE_0 + idx);
2858 return -ENODEV;
2859 }
2860
2861 ioapics[idx].mp_config.apicid = io_apic_unique_id(idx, id);
2862 ioapics[idx].mp_config.apicver = io_apic_get_version(idx);
2863
2864 /*
2865 * Build basic GSI lookup table to facilitate gsi->io_apic lookups
2866 * and to prevent reprogramming of IOAPIC pins (PCI GSIs).
2867 */
2868 entries = io_apic_get_redir_entries(idx);
2869 gsi_end = gsi_base + entries - 1;
2870 for_each_ioapic(ioapic) {
2871 gsi_cfg = mp_ioapic_gsi_routing(ioapic);
2872 if ((gsi_base >= gsi_cfg->gsi_base &&
2873 gsi_base <= gsi_cfg->gsi_end) ||
2874 (gsi_end >= gsi_cfg->gsi_base &&
2875 gsi_end <= gsi_cfg->gsi_end)) {
2876 pr_warn("GSI range [%u-%u] for new IOAPIC conflicts with GSI[%u-%u]\n",
2877 gsi_base, gsi_end,
2878 gsi_cfg->gsi_base, gsi_cfg->gsi_end);
2879 clear_fixmap(FIX_IO_APIC_BASE_0 + idx);
2880 return -ENOSPC;
2881 }
2882 }
2883 gsi_cfg = mp_ioapic_gsi_routing(idx);
2884 gsi_cfg->gsi_base = gsi_base;
2885 gsi_cfg->gsi_end = gsi_end;
2886
2887 ioapics[idx].irqdomain = NULL;
2888 ioapics[idx].irqdomain_cfg = *cfg;
2889
2890 /*
2891 * If mp_register_ioapic() is called during early boot stage when
2892 * walking ACPI/DT tables, it's too early to create irqdomain,
2893 * we are still using bootmem allocator. So delay it to setup_IO_APIC().
2894 */
2895 if (hotplug) {
2896 if (mp_irqdomain_create(idx)) {
2897 clear_fixmap(FIX_IO_APIC_BASE_0 + idx);
2898 return -ENOMEM;
2899 }
2900 alloc_ioapic_saved_registers(idx);
2901 }
2902
2903 if (gsi_cfg->gsi_end >= gsi_top)
2904 gsi_top = gsi_cfg->gsi_end + 1;
2905 if (nr_ioapics <= idx)
2906 nr_ioapics = idx + 1;
2907
2908 /* Set nr_registers to mark entry present */
2909 ioapics[idx].nr_registers = entries;
2910
2911 pr_info("IOAPIC[%d]: apic_id %d, version %d, address 0x%x, GSI %d-%d\n",
2912 idx, mpc_ioapic_id(idx),
2913 mpc_ioapic_ver(idx), mpc_ioapic_addr(idx),
2914 gsi_cfg->gsi_base, gsi_cfg->gsi_end);
2915
2916 return 0;
2917}
2918
2919int mp_unregister_ioapic(u32 gsi_base)
2920{
2921 int ioapic, pin;
2922 int found = 0;
2923
2924 for_each_ioapic(ioapic)
2925 if (ioapics[ioapic].gsi_config.gsi_base == gsi_base) {
2926 found = 1;
2927 break;
2928 }
2929 if (!found) {
2930 pr_warn("can't find IOAPIC for GSI %d\n", gsi_base);
2931 return -ENODEV;
2932 }
2933
2934 for_each_pin(ioapic, pin) {
2935 u32 gsi = mp_pin_to_gsi(ioapic, pin);
2936 int irq = mp_map_gsi_to_irq(gsi, 0, NULL);
2937 struct mp_chip_data *data;
2938
2939 if (irq >= 0) {
2940 data = irq_get_chip_data(irq);
2941 if (data && data->count) {
2942 pr_warn("pin%d on IOAPIC%d is still in use.\n",
2943 pin, ioapic);
2944 return -EBUSY;
2945 }
2946 }
2947 }
2948
2949 /* Mark entry not present */
2950 ioapics[ioapic].nr_registers = 0;
2951 ioapic_destroy_irqdomain(ioapic);
2952 free_ioapic_saved_registers(ioapic);
2953 if (ioapics[ioapic].iomem_res)
2954 release_resource(ioapics[ioapic].iomem_res);
2955 clear_fixmap(FIX_IO_APIC_BASE_0 + ioapic);
2956 memset(&ioapics[ioapic], 0, sizeof(ioapics[ioapic]));
2957
2958 return 0;
2959}
2960
2961int mp_ioapic_registered(u32 gsi_base)
2962{
2963 int ioapic;
2964
2965 for_each_ioapic(ioapic)
2966 if (ioapics[ioapic].gsi_config.gsi_base == gsi_base)
2967 return 1;
2968
2969 return 0;
2970}
2971
2972static void mp_irqdomain_get_attr(u32 gsi, struct mp_chip_data *data,
2973 struct irq_alloc_info *info)
2974{
2975 if (info && info->ioapic.valid) {
2976 data->is_level = info->ioapic.is_level;
2977 data->active_low = info->ioapic.active_low;
2978 } else if (__acpi_get_override_irq(gsi, &data->is_level,
2979 &data->active_low) < 0) {
2980 /* PCI interrupts are always active low level triggered. */
2981 data->is_level = true;
2982 data->active_low = true;
2983 }
2984}
2985
2986/*
2987 * Configure the I/O-APIC specific fields in the routing entry.
2988 *
2989 * This is important to setup the I/O-APIC specific bits (is_level,
2990 * active_low, masked) because the underlying parent domain will only
2991 * provide the routing information and is oblivious of the I/O-APIC
2992 * specific bits.
2993 *
2994 * The entry is just preconfigured at this point and not written into the
2995 * RTE. This happens later during activation which will fill in the actual
2996 * routing information.
2997 */
2998static void mp_preconfigure_entry(struct mp_chip_data *data)
2999{
3000 struct IO_APIC_route_entry *entry = &data->entry;
3001
3002 memset(entry, 0, sizeof(*entry));
3003 entry->is_level = data->is_level;
3004 entry->active_low = data->active_low;
3005 /*
3006 * Mask level triggered irqs. Edge triggered irqs are masked
3007 * by the irq core code in case they fire.
3008 */
3009 entry->masked = data->is_level;
3010}
3011
3012int mp_irqdomain_alloc(struct irq_domain *domain, unsigned int virq,
3013 unsigned int nr_irqs, void *arg)
3014{
3015 struct irq_alloc_info *info = arg;
3016 struct mp_chip_data *data;
3017 struct irq_data *irq_data;
3018 int ret, ioapic, pin;
3019 unsigned long flags;
3020
3021 if (!info || nr_irqs > 1)
3022 return -EINVAL;
3023 irq_data = irq_domain_get_irq_data(domain, virq);
3024 if (!irq_data)
3025 return -EINVAL;
3026
3027 ioapic = mp_irqdomain_ioapic_idx(domain);
3028 pin = info->ioapic.pin;
3029 if (irq_find_mapping(domain, (irq_hw_number_t)pin) > 0)
3030 return -EEXIST;
3031
3032 data = kzalloc(sizeof(*data), GFP_KERNEL);
3033 if (!data)
3034 return -ENOMEM;
3035
3036 ret = irq_domain_alloc_irqs_parent(domain, virq, nr_irqs, info);
3037 if (ret < 0) {
3038 kfree(data);
3039 return ret;
3040 }
3041
3042 INIT_LIST_HEAD(&data->irq_2_pin);
3043 irq_data->hwirq = info->ioapic.pin;
3044 irq_data->chip = (domain->parent == x86_vector_domain) ?
3045 &ioapic_chip : &ioapic_ir_chip;
3046 irq_data->chip_data = data;
3047 mp_irqdomain_get_attr(mp_pin_to_gsi(ioapic, pin), data, info);
3048
3049 add_pin_to_irq_node(data, ioapic_alloc_attr_node(info), ioapic, pin);
3050
3051 mp_preconfigure_entry(data);
3052 mp_register_handler(virq, data->is_level);
3053
3054 local_irq_save(flags);
3055 if (virq < nr_legacy_irqs())
3056 legacy_pic->mask(virq);
3057 local_irq_restore(flags);
3058
3059 apic_printk(APIC_VERBOSE, KERN_DEBUG
3060 "IOAPIC[%d]: Preconfigured routing entry (%d-%d -> IRQ %d Level:%i ActiveLow:%i)\n",
3061 ioapic, mpc_ioapic_id(ioapic), pin, virq,
3062 data->is_level, data->active_low);
3063 return 0;
3064}
3065
3066void mp_irqdomain_free(struct irq_domain *domain, unsigned int virq,
3067 unsigned int nr_irqs)
3068{
3069 struct irq_data *irq_data;
3070 struct mp_chip_data *data;
3071
3072 BUG_ON(nr_irqs != 1);
3073 irq_data = irq_domain_get_irq_data(domain, virq);
3074 if (irq_data && irq_data->chip_data) {
3075 data = irq_data->chip_data;
3076 __remove_pin_from_irq(data, mp_irqdomain_ioapic_idx(domain),
3077 (int)irq_data->hwirq);
3078 WARN_ON(!list_empty(&data->irq_2_pin));
3079 kfree(irq_data->chip_data);
3080 }
3081 irq_domain_free_irqs_top(domain, virq, nr_irqs);
3082}
3083
3084int mp_irqdomain_activate(struct irq_domain *domain,
3085 struct irq_data *irq_data, bool reserve)
3086{
3087 unsigned long flags;
3088
3089 raw_spin_lock_irqsave(&ioapic_lock, flags);
3090 ioapic_configure_entry(irq_data);
3091 raw_spin_unlock_irqrestore(&ioapic_lock, flags);
3092 return 0;
3093}
3094
3095void mp_irqdomain_deactivate(struct irq_domain *domain,
3096 struct irq_data *irq_data)
3097{
3098 /* It won't be called for IRQ with multiple IOAPIC pins associated */
3099 ioapic_mask_entry(mp_irqdomain_ioapic_idx(domain),
3100 (int)irq_data->hwirq);
3101}
3102
3103int mp_irqdomain_ioapic_idx(struct irq_domain *domain)
3104{
3105 return (int)(long)domain->host_data;
3106}
3107
3108const struct irq_domain_ops mp_ioapic_irqdomain_ops = {
3109 .alloc = mp_irqdomain_alloc,
3110 .free = mp_irqdomain_free,
3111 .activate = mp_irqdomain_activate,
3112 .deactivate = mp_irqdomain_deactivate,
3113};
1// SPDX-License-Identifier: GPL-2.0
2/*
3 * Intel IO-APIC support for multi-Pentium hosts.
4 *
5 * Copyright (C) 1997, 1998, 1999, 2000, 2009 Ingo Molnar, Hajnalka Szabo
6 *
7 * Many thanks to Stig Venaas for trying out countless experimental
8 * patches and reporting/debugging problems patiently!
9 *
10 * (c) 1999, Multiple IO-APIC support, developed by
11 * Ken-ichi Yaku <yaku@css1.kbnes.nec.co.jp> and
12 * Hidemi Kishimoto <kisimoto@css1.kbnes.nec.co.jp>,
13 * further tested and cleaned up by Zach Brown <zab@redhat.com>
14 * and Ingo Molnar <mingo@redhat.com>
15 *
16 * Fixes
17 * Maciej W. Rozycki : Bits for genuine 82489DX APICs;
18 * thanks to Eric Gilmore
19 * and Rolf G. Tews
20 * for testing these extensively
21 * Paul Diefenbaugh : Added full ACPI support
22 *
23 * Historical information which is worth to be preserved:
24 *
25 * - SiS APIC rmw bug:
26 *
27 * We used to have a workaround for a bug in SiS chips which
28 * required to rewrite the index register for a read-modify-write
29 * operation as the chip lost the index information which was
30 * setup for the read already. We cache the data now, so that
31 * workaround has been removed.
32 */
33
34#include <linux/mm.h>
35#include <linux/interrupt.h>
36#include <linux/irq.h>
37#include <linux/init.h>
38#include <linux/delay.h>
39#include <linux/sched.h>
40#include <linux/pci.h>
41#include <linux/mc146818rtc.h>
42#include <linux/compiler.h>
43#include <linux/acpi.h>
44#include <linux/export.h>
45#include <linux/syscore_ops.h>
46#include <linux/freezer.h>
47#include <linux/kthread.h>
48#include <linux/jiffies.h> /* time_after() */
49#include <linux/slab.h>
50#include <linux/memblock.h>
51
52#include <asm/irqdomain.h>
53#include <asm/io.h>
54#include <asm/smp.h>
55#include <asm/cpu.h>
56#include <asm/desc.h>
57#include <asm/proto.h>
58#include <asm/acpi.h>
59#include <asm/dma.h>
60#include <asm/timer.h>
61#include <asm/time.h>
62#include <asm/i8259.h>
63#include <asm/setup.h>
64#include <asm/irq_remapping.h>
65#include <asm/hw_irq.h>
66
67#include <asm/apic.h>
68
69#define for_each_ioapic(idx) \
70 for ((idx) = 0; (idx) < nr_ioapics; (idx)++)
71#define for_each_ioapic_reverse(idx) \
72 for ((idx) = nr_ioapics - 1; (idx) >= 0; (idx)--)
73#define for_each_pin(idx, pin) \
74 for ((pin) = 0; (pin) < ioapics[(idx)].nr_registers; (pin)++)
75#define for_each_ioapic_pin(idx, pin) \
76 for_each_ioapic((idx)) \
77 for_each_pin((idx), (pin))
78#define for_each_irq_pin(entry, head) \
79 list_for_each_entry(entry, &head, list)
80
81static DEFINE_RAW_SPINLOCK(ioapic_lock);
82static DEFINE_MUTEX(ioapic_mutex);
83static unsigned int ioapic_dynirq_base;
84static int ioapic_initialized;
85
86struct irq_pin_list {
87 struct list_head list;
88 int apic, pin;
89};
90
91struct mp_chip_data {
92 struct list_head irq_2_pin;
93 struct IO_APIC_route_entry entry;
94 int trigger;
95 int polarity;
96 u32 count;
97 bool isa_irq;
98};
99
100struct mp_ioapic_gsi {
101 u32 gsi_base;
102 u32 gsi_end;
103};
104
105static struct ioapic {
106 /*
107 * # of IRQ routing registers
108 */
109 int nr_registers;
110 /*
111 * Saved state during suspend/resume, or while enabling intr-remap.
112 */
113 struct IO_APIC_route_entry *saved_registers;
114 /* I/O APIC config */
115 struct mpc_ioapic mp_config;
116 /* IO APIC gsi routing info */
117 struct mp_ioapic_gsi gsi_config;
118 struct ioapic_domain_cfg irqdomain_cfg;
119 struct irq_domain *irqdomain;
120 struct resource *iomem_res;
121} ioapics[MAX_IO_APICS];
122
123#define mpc_ioapic_ver(ioapic_idx) ioapics[ioapic_idx].mp_config.apicver
124
125int mpc_ioapic_id(int ioapic_idx)
126{
127 return ioapics[ioapic_idx].mp_config.apicid;
128}
129
130unsigned int mpc_ioapic_addr(int ioapic_idx)
131{
132 return ioapics[ioapic_idx].mp_config.apicaddr;
133}
134
135static inline struct mp_ioapic_gsi *mp_ioapic_gsi_routing(int ioapic_idx)
136{
137 return &ioapics[ioapic_idx].gsi_config;
138}
139
140static inline int mp_ioapic_pin_count(int ioapic)
141{
142 struct mp_ioapic_gsi *gsi_cfg = mp_ioapic_gsi_routing(ioapic);
143
144 return gsi_cfg->gsi_end - gsi_cfg->gsi_base + 1;
145}
146
147static inline u32 mp_pin_to_gsi(int ioapic, int pin)
148{
149 return mp_ioapic_gsi_routing(ioapic)->gsi_base + pin;
150}
151
152static inline bool mp_is_legacy_irq(int irq)
153{
154 return irq >= 0 && irq < nr_legacy_irqs();
155}
156
157static inline struct irq_domain *mp_ioapic_irqdomain(int ioapic)
158{
159 return ioapics[ioapic].irqdomain;
160}
161
162int nr_ioapics;
163
164/* The one past the highest gsi number used */
165u32 gsi_top;
166
167/* MP IRQ source entries */
168struct mpc_intsrc mp_irqs[MAX_IRQ_SOURCES];
169
170/* # of MP IRQ source entries */
171int mp_irq_entries;
172
173#ifdef CONFIG_EISA
174int mp_bus_id_to_type[MAX_MP_BUSSES];
175#endif
176
177DECLARE_BITMAP(mp_bus_not_pci, MAX_MP_BUSSES);
178
179int skip_ioapic_setup;
180
181/**
182 * disable_ioapic_support() - disables ioapic support at runtime
183 */
184void disable_ioapic_support(void)
185{
186#ifdef CONFIG_PCI
187 noioapicquirk = 1;
188 noioapicreroute = -1;
189#endif
190 skip_ioapic_setup = 1;
191}
192
193static int __init parse_noapic(char *str)
194{
195 /* disable IO-APIC */
196 disable_ioapic_support();
197 return 0;
198}
199early_param("noapic", parse_noapic);
200
201/* Will be called in mpparse/acpi/sfi codes for saving IRQ info */
202void mp_save_irq(struct mpc_intsrc *m)
203{
204 int i;
205
206 apic_printk(APIC_VERBOSE, "Int: type %d, pol %d, trig %d, bus %02x,"
207 " IRQ %02x, APIC ID %x, APIC INT %02x\n",
208 m->irqtype, m->irqflag & 3, (m->irqflag >> 2) & 3, m->srcbus,
209 m->srcbusirq, m->dstapic, m->dstirq);
210
211 for (i = 0; i < mp_irq_entries; i++) {
212 if (!memcmp(&mp_irqs[i], m, sizeof(*m)))
213 return;
214 }
215
216 memcpy(&mp_irqs[mp_irq_entries], m, sizeof(*m));
217 if (++mp_irq_entries == MAX_IRQ_SOURCES)
218 panic("Max # of irq sources exceeded!!\n");
219}
220
221static void alloc_ioapic_saved_registers(int idx)
222{
223 size_t size;
224
225 if (ioapics[idx].saved_registers)
226 return;
227
228 size = sizeof(struct IO_APIC_route_entry) * ioapics[idx].nr_registers;
229 ioapics[idx].saved_registers = kzalloc(size, GFP_KERNEL);
230 if (!ioapics[idx].saved_registers)
231 pr_err("IOAPIC %d: suspend/resume impossible!\n", idx);
232}
233
234static void free_ioapic_saved_registers(int idx)
235{
236 kfree(ioapics[idx].saved_registers);
237 ioapics[idx].saved_registers = NULL;
238}
239
240int __init arch_early_ioapic_init(void)
241{
242 int i;
243
244 if (!nr_legacy_irqs())
245 io_apic_irqs = ~0UL;
246
247 for_each_ioapic(i)
248 alloc_ioapic_saved_registers(i);
249
250 return 0;
251}
252
253struct io_apic {
254 unsigned int index;
255 unsigned int unused[3];
256 unsigned int data;
257 unsigned int unused2[11];
258 unsigned int eoi;
259};
260
261static __attribute_const__ struct io_apic __iomem *io_apic_base(int idx)
262{
263 return (void __iomem *) __fix_to_virt(FIX_IO_APIC_BASE_0 + idx)
264 + (mpc_ioapic_addr(idx) & ~PAGE_MASK);
265}
266
267static inline void io_apic_eoi(unsigned int apic, unsigned int vector)
268{
269 struct io_apic __iomem *io_apic = io_apic_base(apic);
270 writel(vector, &io_apic->eoi);
271}
272
273unsigned int native_io_apic_read(unsigned int apic, unsigned int reg)
274{
275 struct io_apic __iomem *io_apic = io_apic_base(apic);
276 writel(reg, &io_apic->index);
277 return readl(&io_apic->data);
278}
279
280static void io_apic_write(unsigned int apic, unsigned int reg,
281 unsigned int value)
282{
283 struct io_apic __iomem *io_apic = io_apic_base(apic);
284
285 writel(reg, &io_apic->index);
286 writel(value, &io_apic->data);
287}
288
289union entry_union {
290 struct { u32 w1, w2; };
291 struct IO_APIC_route_entry entry;
292};
293
294static struct IO_APIC_route_entry __ioapic_read_entry(int apic, int pin)
295{
296 union entry_union eu;
297
298 eu.w1 = io_apic_read(apic, 0x10 + 2 * pin);
299 eu.w2 = io_apic_read(apic, 0x11 + 2 * pin);
300
301 return eu.entry;
302}
303
304static struct IO_APIC_route_entry ioapic_read_entry(int apic, int pin)
305{
306 union entry_union eu;
307 unsigned long flags;
308
309 raw_spin_lock_irqsave(&ioapic_lock, flags);
310 eu.entry = __ioapic_read_entry(apic, pin);
311 raw_spin_unlock_irqrestore(&ioapic_lock, flags);
312
313 return eu.entry;
314}
315
316/*
317 * When we write a new IO APIC routing entry, we need to write the high
318 * word first! If the mask bit in the low word is clear, we will enable
319 * the interrupt, and we need to make sure the entry is fully populated
320 * before that happens.
321 */
322static void __ioapic_write_entry(int apic, int pin, struct IO_APIC_route_entry e)
323{
324 union entry_union eu = {{0, 0}};
325
326 eu.entry = e;
327 io_apic_write(apic, 0x11 + 2*pin, eu.w2);
328 io_apic_write(apic, 0x10 + 2*pin, eu.w1);
329}
330
331static void ioapic_write_entry(int apic, int pin, struct IO_APIC_route_entry e)
332{
333 unsigned long flags;
334
335 raw_spin_lock_irqsave(&ioapic_lock, flags);
336 __ioapic_write_entry(apic, pin, e);
337 raw_spin_unlock_irqrestore(&ioapic_lock, flags);
338}
339
340/*
341 * When we mask an IO APIC routing entry, we need to write the low
342 * word first, in order to set the mask bit before we change the
343 * high bits!
344 */
345static void ioapic_mask_entry(int apic, int pin)
346{
347 unsigned long flags;
348 union entry_union eu = { .entry.mask = IOAPIC_MASKED };
349
350 raw_spin_lock_irqsave(&ioapic_lock, flags);
351 io_apic_write(apic, 0x10 + 2*pin, eu.w1);
352 io_apic_write(apic, 0x11 + 2*pin, eu.w2);
353 raw_spin_unlock_irqrestore(&ioapic_lock, flags);
354}
355
356/*
357 * The common case is 1:1 IRQ<->pin mappings. Sometimes there are
358 * shared ISA-space IRQs, so we have to support them. We are super
359 * fast in the common case, and fast for shared ISA-space IRQs.
360 */
361static int __add_pin_to_irq_node(struct mp_chip_data *data,
362 int node, int apic, int pin)
363{
364 struct irq_pin_list *entry;
365
366 /* don't allow duplicates */
367 for_each_irq_pin(entry, data->irq_2_pin)
368 if (entry->apic == apic && entry->pin == pin)
369 return 0;
370
371 entry = kzalloc_node(sizeof(struct irq_pin_list), GFP_ATOMIC, node);
372 if (!entry) {
373 pr_err("can not alloc irq_pin_list (%d,%d,%d)\n",
374 node, apic, pin);
375 return -ENOMEM;
376 }
377 entry->apic = apic;
378 entry->pin = pin;
379 list_add_tail(&entry->list, &data->irq_2_pin);
380
381 return 0;
382}
383
384static void __remove_pin_from_irq(struct mp_chip_data *data, int apic, int pin)
385{
386 struct irq_pin_list *tmp, *entry;
387
388 list_for_each_entry_safe(entry, tmp, &data->irq_2_pin, list)
389 if (entry->apic == apic && entry->pin == pin) {
390 list_del(&entry->list);
391 kfree(entry);
392 return;
393 }
394}
395
396static void add_pin_to_irq_node(struct mp_chip_data *data,
397 int node, int apic, int pin)
398{
399 if (__add_pin_to_irq_node(data, node, apic, pin))
400 panic("IO-APIC: failed to add irq-pin. Can not proceed\n");
401}
402
403/*
404 * Reroute an IRQ to a different pin.
405 */
406static void __init replace_pin_at_irq_node(struct mp_chip_data *data, int node,
407 int oldapic, int oldpin,
408 int newapic, int newpin)
409{
410 struct irq_pin_list *entry;
411
412 for_each_irq_pin(entry, data->irq_2_pin) {
413 if (entry->apic == oldapic && entry->pin == oldpin) {
414 entry->apic = newapic;
415 entry->pin = newpin;
416 /* every one is different, right? */
417 return;
418 }
419 }
420
421 /* old apic/pin didn't exist, so just add new ones */
422 add_pin_to_irq_node(data, node, newapic, newpin);
423}
424
425static void io_apic_modify_irq(struct mp_chip_data *data,
426 int mask_and, int mask_or,
427 void (*final)(struct irq_pin_list *entry))
428{
429 union entry_union eu;
430 struct irq_pin_list *entry;
431
432 eu.entry = data->entry;
433 eu.w1 &= mask_and;
434 eu.w1 |= mask_or;
435 data->entry = eu.entry;
436
437 for_each_irq_pin(entry, data->irq_2_pin) {
438 io_apic_write(entry->apic, 0x10 + 2 * entry->pin, eu.w1);
439 if (final)
440 final(entry);
441 }
442}
443
444static void io_apic_sync(struct irq_pin_list *entry)
445{
446 /*
447 * Synchronize the IO-APIC and the CPU by doing
448 * a dummy read from the IO-APIC
449 */
450 struct io_apic __iomem *io_apic;
451
452 io_apic = io_apic_base(entry->apic);
453 readl(&io_apic->data);
454}
455
456static void mask_ioapic_irq(struct irq_data *irq_data)
457{
458 struct mp_chip_data *data = irq_data->chip_data;
459 unsigned long flags;
460
461 raw_spin_lock_irqsave(&ioapic_lock, flags);
462 io_apic_modify_irq(data, ~0, IO_APIC_REDIR_MASKED, &io_apic_sync);
463 raw_spin_unlock_irqrestore(&ioapic_lock, flags);
464}
465
466static void __unmask_ioapic(struct mp_chip_data *data)
467{
468 io_apic_modify_irq(data, ~IO_APIC_REDIR_MASKED, 0, NULL);
469}
470
471static void unmask_ioapic_irq(struct irq_data *irq_data)
472{
473 struct mp_chip_data *data = irq_data->chip_data;
474 unsigned long flags;
475
476 raw_spin_lock_irqsave(&ioapic_lock, flags);
477 __unmask_ioapic(data);
478 raw_spin_unlock_irqrestore(&ioapic_lock, flags);
479}
480
481/*
482 * IO-APIC versions below 0x20 don't support EOI register.
483 * For the record, here is the information about various versions:
484 * 0Xh 82489DX
485 * 1Xh I/OAPIC or I/O(x)APIC which are not PCI 2.2 Compliant
486 * 2Xh I/O(x)APIC which is PCI 2.2 Compliant
487 * 30h-FFh Reserved
488 *
489 * Some of the Intel ICH Specs (ICH2 to ICH5) documents the io-apic
490 * version as 0x2. This is an error with documentation and these ICH chips
491 * use io-apic's of version 0x20.
492 *
493 * For IO-APIC's with EOI register, we use that to do an explicit EOI.
494 * Otherwise, we simulate the EOI message manually by changing the trigger
495 * mode to edge and then back to level, with RTE being masked during this.
496 */
497static void __eoi_ioapic_pin(int apic, int pin, int vector)
498{
499 if (mpc_ioapic_ver(apic) >= 0x20) {
500 io_apic_eoi(apic, vector);
501 } else {
502 struct IO_APIC_route_entry entry, entry1;
503
504 entry = entry1 = __ioapic_read_entry(apic, pin);
505
506 /*
507 * Mask the entry and change the trigger mode to edge.
508 */
509 entry1.mask = IOAPIC_MASKED;
510 entry1.trigger = IOAPIC_EDGE;
511
512 __ioapic_write_entry(apic, pin, entry1);
513
514 /*
515 * Restore the previous level triggered entry.
516 */
517 __ioapic_write_entry(apic, pin, entry);
518 }
519}
520
521static void eoi_ioapic_pin(int vector, struct mp_chip_data *data)
522{
523 unsigned long flags;
524 struct irq_pin_list *entry;
525
526 raw_spin_lock_irqsave(&ioapic_lock, flags);
527 for_each_irq_pin(entry, data->irq_2_pin)
528 __eoi_ioapic_pin(entry->apic, entry->pin, vector);
529 raw_spin_unlock_irqrestore(&ioapic_lock, flags);
530}
531
532static void clear_IO_APIC_pin(unsigned int apic, unsigned int pin)
533{
534 struct IO_APIC_route_entry entry;
535
536 /* Check delivery_mode to be sure we're not clearing an SMI pin */
537 entry = ioapic_read_entry(apic, pin);
538 if (entry.delivery_mode == dest_SMI)
539 return;
540
541 /*
542 * Make sure the entry is masked and re-read the contents to check
543 * if it is a level triggered pin and if the remote-IRR is set.
544 */
545 if (entry.mask == IOAPIC_UNMASKED) {
546 entry.mask = IOAPIC_MASKED;
547 ioapic_write_entry(apic, pin, entry);
548 entry = ioapic_read_entry(apic, pin);
549 }
550
551 if (entry.irr) {
552 unsigned long flags;
553
554 /*
555 * Make sure the trigger mode is set to level. Explicit EOI
556 * doesn't clear the remote-IRR if the trigger mode is not
557 * set to level.
558 */
559 if (entry.trigger == IOAPIC_EDGE) {
560 entry.trigger = IOAPIC_LEVEL;
561 ioapic_write_entry(apic, pin, entry);
562 }
563 raw_spin_lock_irqsave(&ioapic_lock, flags);
564 __eoi_ioapic_pin(apic, pin, entry.vector);
565 raw_spin_unlock_irqrestore(&ioapic_lock, flags);
566 }
567
568 /*
569 * Clear the rest of the bits in the IO-APIC RTE except for the mask
570 * bit.
571 */
572 ioapic_mask_entry(apic, pin);
573 entry = ioapic_read_entry(apic, pin);
574 if (entry.irr)
575 pr_err("Unable to reset IRR for apic: %d, pin :%d\n",
576 mpc_ioapic_id(apic), pin);
577}
578
579void clear_IO_APIC (void)
580{
581 int apic, pin;
582
583 for_each_ioapic_pin(apic, pin)
584 clear_IO_APIC_pin(apic, pin);
585}
586
587#ifdef CONFIG_X86_32
588/*
589 * support for broken MP BIOSs, enables hand-redirection of PIRQ0-7 to
590 * specific CPU-side IRQs.
591 */
592
593#define MAX_PIRQS 8
594static int pirq_entries[MAX_PIRQS] = {
595 [0 ... MAX_PIRQS - 1] = -1
596};
597
598static int __init ioapic_pirq_setup(char *str)
599{
600 int i, max;
601 int ints[MAX_PIRQS+1];
602
603 get_options(str, ARRAY_SIZE(ints), ints);
604
605 apic_printk(APIC_VERBOSE, KERN_INFO
606 "PIRQ redirection, working around broken MP-BIOS.\n");
607 max = MAX_PIRQS;
608 if (ints[0] < MAX_PIRQS)
609 max = ints[0];
610
611 for (i = 0; i < max; i++) {
612 apic_printk(APIC_VERBOSE, KERN_DEBUG
613 "... PIRQ%d -> IRQ %d\n", i, ints[i+1]);
614 /*
615 * PIRQs are mapped upside down, usually.
616 */
617 pirq_entries[MAX_PIRQS-i-1] = ints[i+1];
618 }
619 return 1;
620}
621
622__setup("pirq=", ioapic_pirq_setup);
623#endif /* CONFIG_X86_32 */
624
625/*
626 * Saves all the IO-APIC RTE's
627 */
628int save_ioapic_entries(void)
629{
630 int apic, pin;
631 int err = 0;
632
633 for_each_ioapic(apic) {
634 if (!ioapics[apic].saved_registers) {
635 err = -ENOMEM;
636 continue;
637 }
638
639 for_each_pin(apic, pin)
640 ioapics[apic].saved_registers[pin] =
641 ioapic_read_entry(apic, pin);
642 }
643
644 return err;
645}
646
647/*
648 * Mask all IO APIC entries.
649 */
650void mask_ioapic_entries(void)
651{
652 int apic, pin;
653
654 for_each_ioapic(apic) {
655 if (!ioapics[apic].saved_registers)
656 continue;
657
658 for_each_pin(apic, pin) {
659 struct IO_APIC_route_entry entry;
660
661 entry = ioapics[apic].saved_registers[pin];
662 if (entry.mask == IOAPIC_UNMASKED) {
663 entry.mask = IOAPIC_MASKED;
664 ioapic_write_entry(apic, pin, entry);
665 }
666 }
667 }
668}
669
670/*
671 * Restore IO APIC entries which was saved in the ioapic structure.
672 */
673int restore_ioapic_entries(void)
674{
675 int apic, pin;
676
677 for_each_ioapic(apic) {
678 if (!ioapics[apic].saved_registers)
679 continue;
680
681 for_each_pin(apic, pin)
682 ioapic_write_entry(apic, pin,
683 ioapics[apic].saved_registers[pin]);
684 }
685 return 0;
686}
687
688/*
689 * Find the IRQ entry number of a certain pin.
690 */
691static int find_irq_entry(int ioapic_idx, int pin, int type)
692{
693 int i;
694
695 for (i = 0; i < mp_irq_entries; i++)
696 if (mp_irqs[i].irqtype == type &&
697 (mp_irqs[i].dstapic == mpc_ioapic_id(ioapic_idx) ||
698 mp_irqs[i].dstapic == MP_APIC_ALL) &&
699 mp_irqs[i].dstirq == pin)
700 return i;
701
702 return -1;
703}
704
705/*
706 * Find the pin to which IRQ[irq] (ISA) is connected
707 */
708static int __init find_isa_irq_pin(int irq, int type)
709{
710 int i;
711
712 for (i = 0; i < mp_irq_entries; i++) {
713 int lbus = mp_irqs[i].srcbus;
714
715 if (test_bit(lbus, mp_bus_not_pci) &&
716 (mp_irqs[i].irqtype == type) &&
717 (mp_irqs[i].srcbusirq == irq))
718
719 return mp_irqs[i].dstirq;
720 }
721 return -1;
722}
723
724static int __init find_isa_irq_apic(int irq, int type)
725{
726 int i;
727
728 for (i = 0; i < mp_irq_entries; i++) {
729 int lbus = mp_irqs[i].srcbus;
730
731 if (test_bit(lbus, mp_bus_not_pci) &&
732 (mp_irqs[i].irqtype == type) &&
733 (mp_irqs[i].srcbusirq == irq))
734 break;
735 }
736
737 if (i < mp_irq_entries) {
738 int ioapic_idx;
739
740 for_each_ioapic(ioapic_idx)
741 if (mpc_ioapic_id(ioapic_idx) == mp_irqs[i].dstapic)
742 return ioapic_idx;
743 }
744
745 return -1;
746}
747
748#ifdef CONFIG_EISA
749/*
750 * EISA Edge/Level control register, ELCR
751 */
752static int EISA_ELCR(unsigned int irq)
753{
754 if (irq < nr_legacy_irqs()) {
755 unsigned int port = 0x4d0 + (irq >> 3);
756 return (inb(port) >> (irq & 7)) & 1;
757 }
758 apic_printk(APIC_VERBOSE, KERN_INFO
759 "Broken MPtable reports ISA irq %d\n", irq);
760 return 0;
761}
762
763#endif
764
765/* ISA interrupts are always active high edge triggered,
766 * when listed as conforming in the MP table. */
767
768#define default_ISA_trigger(idx) (IOAPIC_EDGE)
769#define default_ISA_polarity(idx) (IOAPIC_POL_HIGH)
770
771/* EISA interrupts are always polarity zero and can be edge or level
772 * trigger depending on the ELCR value. If an interrupt is listed as
773 * EISA conforming in the MP table, that means its trigger type must
774 * be read in from the ELCR */
775
776#define default_EISA_trigger(idx) (EISA_ELCR(mp_irqs[idx].srcbusirq))
777#define default_EISA_polarity(idx) default_ISA_polarity(idx)
778
779/* PCI interrupts are always active low level triggered,
780 * when listed as conforming in the MP table. */
781
782#define default_PCI_trigger(idx) (IOAPIC_LEVEL)
783#define default_PCI_polarity(idx) (IOAPIC_POL_LOW)
784
785static int irq_polarity(int idx)
786{
787 int bus = mp_irqs[idx].srcbus;
788
789 /*
790 * Determine IRQ line polarity (high active or low active):
791 */
792 switch (mp_irqs[idx].irqflag & MP_IRQPOL_MASK) {
793 case MP_IRQPOL_DEFAULT:
794 /* conforms to spec, ie. bus-type dependent polarity */
795 if (test_bit(bus, mp_bus_not_pci))
796 return default_ISA_polarity(idx);
797 else
798 return default_PCI_polarity(idx);
799 case MP_IRQPOL_ACTIVE_HIGH:
800 return IOAPIC_POL_HIGH;
801 case MP_IRQPOL_RESERVED:
802 pr_warn("IOAPIC: Invalid polarity: 2, defaulting to low\n");
803 fallthrough;
804 case MP_IRQPOL_ACTIVE_LOW:
805 default: /* Pointless default required due to do gcc stupidity */
806 return IOAPIC_POL_LOW;
807 }
808}
809
810#ifdef CONFIG_EISA
811static int eisa_irq_trigger(int idx, int bus, int trigger)
812{
813 switch (mp_bus_id_to_type[bus]) {
814 case MP_BUS_PCI:
815 case MP_BUS_ISA:
816 return trigger;
817 case MP_BUS_EISA:
818 return default_EISA_trigger(idx);
819 }
820 pr_warn("IOAPIC: Invalid srcbus: %d defaulting to level\n", bus);
821 return IOAPIC_LEVEL;
822}
823#else
824static inline int eisa_irq_trigger(int idx, int bus, int trigger)
825{
826 return trigger;
827}
828#endif
829
830static int irq_trigger(int idx)
831{
832 int bus = mp_irqs[idx].srcbus;
833 int trigger;
834
835 /*
836 * Determine IRQ trigger mode (edge or level sensitive):
837 */
838 switch (mp_irqs[idx].irqflag & MP_IRQTRIG_MASK) {
839 case MP_IRQTRIG_DEFAULT:
840 /* conforms to spec, ie. bus-type dependent trigger mode */
841 if (test_bit(bus, mp_bus_not_pci))
842 trigger = default_ISA_trigger(idx);
843 else
844 trigger = default_PCI_trigger(idx);
845 /* Take EISA into account */
846 return eisa_irq_trigger(idx, bus, trigger);
847 case MP_IRQTRIG_EDGE:
848 return IOAPIC_EDGE;
849 case MP_IRQTRIG_RESERVED:
850 pr_warn("IOAPIC: Invalid trigger mode 2 defaulting to level\n");
851 fallthrough;
852 case MP_IRQTRIG_LEVEL:
853 default: /* Pointless default required due to do gcc stupidity */
854 return IOAPIC_LEVEL;
855 }
856}
857
858void ioapic_set_alloc_attr(struct irq_alloc_info *info, int node,
859 int trigger, int polarity)
860{
861 init_irq_alloc_info(info, NULL);
862 info->type = X86_IRQ_ALLOC_TYPE_IOAPIC;
863 info->ioapic_node = node;
864 info->ioapic_trigger = trigger;
865 info->ioapic_polarity = polarity;
866 info->ioapic_valid = 1;
867}
868
869#ifndef CONFIG_ACPI
870int acpi_get_override_irq(u32 gsi, int *trigger, int *polarity);
871#endif
872
873static void ioapic_copy_alloc_attr(struct irq_alloc_info *dst,
874 struct irq_alloc_info *src,
875 u32 gsi, int ioapic_idx, int pin)
876{
877 int trigger, polarity;
878
879 copy_irq_alloc_info(dst, src);
880 dst->type = X86_IRQ_ALLOC_TYPE_IOAPIC;
881 dst->ioapic_id = mpc_ioapic_id(ioapic_idx);
882 dst->ioapic_pin = pin;
883 dst->ioapic_valid = 1;
884 if (src && src->ioapic_valid) {
885 dst->ioapic_node = src->ioapic_node;
886 dst->ioapic_trigger = src->ioapic_trigger;
887 dst->ioapic_polarity = src->ioapic_polarity;
888 } else {
889 dst->ioapic_node = NUMA_NO_NODE;
890 if (acpi_get_override_irq(gsi, &trigger, &polarity) >= 0) {
891 dst->ioapic_trigger = trigger;
892 dst->ioapic_polarity = polarity;
893 } else {
894 /*
895 * PCI interrupts are always active low level
896 * triggered.
897 */
898 dst->ioapic_trigger = IOAPIC_LEVEL;
899 dst->ioapic_polarity = IOAPIC_POL_LOW;
900 }
901 }
902}
903
904static int ioapic_alloc_attr_node(struct irq_alloc_info *info)
905{
906 return (info && info->ioapic_valid) ? info->ioapic_node : NUMA_NO_NODE;
907}
908
909static void mp_register_handler(unsigned int irq, unsigned long trigger)
910{
911 irq_flow_handler_t hdl;
912 bool fasteoi;
913
914 if (trigger) {
915 irq_set_status_flags(irq, IRQ_LEVEL);
916 fasteoi = true;
917 } else {
918 irq_clear_status_flags(irq, IRQ_LEVEL);
919 fasteoi = false;
920 }
921
922 hdl = fasteoi ? handle_fasteoi_irq : handle_edge_irq;
923 __irq_set_handler(irq, hdl, 0, fasteoi ? "fasteoi" : "edge");
924}
925
926static bool mp_check_pin_attr(int irq, struct irq_alloc_info *info)
927{
928 struct mp_chip_data *data = irq_get_chip_data(irq);
929
930 /*
931 * setup_IO_APIC_irqs() programs all legacy IRQs with default trigger
932 * and polarity attirbutes. So allow the first user to reprogram the
933 * pin with real trigger and polarity attributes.
934 */
935 if (irq < nr_legacy_irqs() && data->count == 1) {
936 if (info->ioapic_trigger != data->trigger)
937 mp_register_handler(irq, info->ioapic_trigger);
938 data->entry.trigger = data->trigger = info->ioapic_trigger;
939 data->entry.polarity = data->polarity = info->ioapic_polarity;
940 }
941
942 return data->trigger == info->ioapic_trigger &&
943 data->polarity == info->ioapic_polarity;
944}
945
946static int alloc_irq_from_domain(struct irq_domain *domain, int ioapic, u32 gsi,
947 struct irq_alloc_info *info)
948{
949 bool legacy = false;
950 int irq = -1;
951 int type = ioapics[ioapic].irqdomain_cfg.type;
952
953 switch (type) {
954 case IOAPIC_DOMAIN_LEGACY:
955 /*
956 * Dynamically allocate IRQ number for non-ISA IRQs in the first
957 * 16 GSIs on some weird platforms.
958 */
959 if (!ioapic_initialized || gsi >= nr_legacy_irqs())
960 irq = gsi;
961 legacy = mp_is_legacy_irq(irq);
962 break;
963 case IOAPIC_DOMAIN_STRICT:
964 irq = gsi;
965 break;
966 case IOAPIC_DOMAIN_DYNAMIC:
967 break;
968 default:
969 WARN(1, "ioapic: unknown irqdomain type %d\n", type);
970 return -1;
971 }
972
973 return __irq_domain_alloc_irqs(domain, irq, 1,
974 ioapic_alloc_attr_node(info),
975 info, legacy, NULL);
976}
977
978/*
979 * Need special handling for ISA IRQs because there may be multiple IOAPIC pins
980 * sharing the same ISA IRQ number and irqdomain only supports 1:1 mapping
981 * between IOAPIC pin and IRQ number. A typical IOAPIC has 24 pins, pin 0-15 are
982 * used for legacy IRQs and pin 16-23 are used for PCI IRQs (PIRQ A-H).
983 * When ACPI is disabled, only legacy IRQ numbers (IRQ0-15) are available, and
984 * some BIOSes may use MP Interrupt Source records to override IRQ numbers for
985 * PIRQs instead of reprogramming the interrupt routing logic. Thus there may be
986 * multiple pins sharing the same legacy IRQ number when ACPI is disabled.
987 */
988static int alloc_isa_irq_from_domain(struct irq_domain *domain,
989 int irq, int ioapic, int pin,
990 struct irq_alloc_info *info)
991{
992 struct mp_chip_data *data;
993 struct irq_data *irq_data = irq_get_irq_data(irq);
994 int node = ioapic_alloc_attr_node(info);
995
996 /*
997 * Legacy ISA IRQ has already been allocated, just add pin to
998 * the pin list assoicated with this IRQ and program the IOAPIC
999 * entry. The IOAPIC entry
1000 */
1001 if (irq_data && irq_data->parent_data) {
1002 if (!mp_check_pin_attr(irq, info))
1003 return -EBUSY;
1004 if (__add_pin_to_irq_node(irq_data->chip_data, node, ioapic,
1005 info->ioapic_pin))
1006 return -ENOMEM;
1007 } else {
1008 info->flags |= X86_IRQ_ALLOC_LEGACY;
1009 irq = __irq_domain_alloc_irqs(domain, irq, 1, node, info, true,
1010 NULL);
1011 if (irq >= 0) {
1012 irq_data = irq_domain_get_irq_data(domain, irq);
1013 data = irq_data->chip_data;
1014 data->isa_irq = true;
1015 }
1016 }
1017
1018 return irq;
1019}
1020
1021static int mp_map_pin_to_irq(u32 gsi, int idx, int ioapic, int pin,
1022 unsigned int flags, struct irq_alloc_info *info)
1023{
1024 int irq;
1025 bool legacy = false;
1026 struct irq_alloc_info tmp;
1027 struct mp_chip_data *data;
1028 struct irq_domain *domain = mp_ioapic_irqdomain(ioapic);
1029
1030 if (!domain)
1031 return -ENOSYS;
1032
1033 if (idx >= 0 && test_bit(mp_irqs[idx].srcbus, mp_bus_not_pci)) {
1034 irq = mp_irqs[idx].srcbusirq;
1035 legacy = mp_is_legacy_irq(irq);
1036 }
1037
1038 mutex_lock(&ioapic_mutex);
1039 if (!(flags & IOAPIC_MAP_ALLOC)) {
1040 if (!legacy) {
1041 irq = irq_find_mapping(domain, pin);
1042 if (irq == 0)
1043 irq = -ENOENT;
1044 }
1045 } else {
1046 ioapic_copy_alloc_attr(&tmp, info, gsi, ioapic, pin);
1047 if (legacy)
1048 irq = alloc_isa_irq_from_domain(domain, irq,
1049 ioapic, pin, &tmp);
1050 else if ((irq = irq_find_mapping(domain, pin)) == 0)
1051 irq = alloc_irq_from_domain(domain, ioapic, gsi, &tmp);
1052 else if (!mp_check_pin_attr(irq, &tmp))
1053 irq = -EBUSY;
1054 if (irq >= 0) {
1055 data = irq_get_chip_data(irq);
1056 data->count++;
1057 }
1058 }
1059 mutex_unlock(&ioapic_mutex);
1060
1061 return irq;
1062}
1063
1064static int pin_2_irq(int idx, int ioapic, int pin, unsigned int flags)
1065{
1066 u32 gsi = mp_pin_to_gsi(ioapic, pin);
1067
1068 /*
1069 * Debugging check, we are in big trouble if this message pops up!
1070 */
1071 if (mp_irqs[idx].dstirq != pin)
1072 pr_err("broken BIOS or MPTABLE parser, ayiee!!\n");
1073
1074#ifdef CONFIG_X86_32
1075 /*
1076 * PCI IRQ command line redirection. Yes, limits are hardcoded.
1077 */
1078 if ((pin >= 16) && (pin <= 23)) {
1079 if (pirq_entries[pin-16] != -1) {
1080 if (!pirq_entries[pin-16]) {
1081 apic_printk(APIC_VERBOSE, KERN_DEBUG
1082 "disabling PIRQ%d\n", pin-16);
1083 } else {
1084 int irq = pirq_entries[pin-16];
1085 apic_printk(APIC_VERBOSE, KERN_DEBUG
1086 "using PIRQ%d -> IRQ %d\n",
1087 pin-16, irq);
1088 return irq;
1089 }
1090 }
1091 }
1092#endif
1093
1094 return mp_map_pin_to_irq(gsi, idx, ioapic, pin, flags, NULL);
1095}
1096
1097int mp_map_gsi_to_irq(u32 gsi, unsigned int flags, struct irq_alloc_info *info)
1098{
1099 int ioapic, pin, idx;
1100
1101 ioapic = mp_find_ioapic(gsi);
1102 if (ioapic < 0)
1103 return -ENODEV;
1104
1105 pin = mp_find_ioapic_pin(ioapic, gsi);
1106 idx = find_irq_entry(ioapic, pin, mp_INT);
1107 if ((flags & IOAPIC_MAP_CHECK) && idx < 0)
1108 return -ENODEV;
1109
1110 return mp_map_pin_to_irq(gsi, idx, ioapic, pin, flags, info);
1111}
1112
1113void mp_unmap_irq(int irq)
1114{
1115 struct irq_data *irq_data = irq_get_irq_data(irq);
1116 struct mp_chip_data *data;
1117
1118 if (!irq_data || !irq_data->domain)
1119 return;
1120
1121 data = irq_data->chip_data;
1122 if (!data || data->isa_irq)
1123 return;
1124
1125 mutex_lock(&ioapic_mutex);
1126 if (--data->count == 0)
1127 irq_domain_free_irqs(irq, 1);
1128 mutex_unlock(&ioapic_mutex);
1129}
1130
1131/*
1132 * Find a specific PCI IRQ entry.
1133 * Not an __init, possibly needed by modules
1134 */
1135int IO_APIC_get_PCI_irq_vector(int bus, int slot, int pin)
1136{
1137 int irq, i, best_ioapic = -1, best_idx = -1;
1138
1139 apic_printk(APIC_DEBUG,
1140 "querying PCI -> IRQ mapping bus:%d, slot:%d, pin:%d.\n",
1141 bus, slot, pin);
1142 if (test_bit(bus, mp_bus_not_pci)) {
1143 apic_printk(APIC_VERBOSE,
1144 "PCI BIOS passed nonexistent PCI bus %d!\n", bus);
1145 return -1;
1146 }
1147
1148 for (i = 0; i < mp_irq_entries; i++) {
1149 int lbus = mp_irqs[i].srcbus;
1150 int ioapic_idx, found = 0;
1151
1152 if (bus != lbus || mp_irqs[i].irqtype != mp_INT ||
1153 slot != ((mp_irqs[i].srcbusirq >> 2) & 0x1f))
1154 continue;
1155
1156 for_each_ioapic(ioapic_idx)
1157 if (mpc_ioapic_id(ioapic_idx) == mp_irqs[i].dstapic ||
1158 mp_irqs[i].dstapic == MP_APIC_ALL) {
1159 found = 1;
1160 break;
1161 }
1162 if (!found)
1163 continue;
1164
1165 /* Skip ISA IRQs */
1166 irq = pin_2_irq(i, ioapic_idx, mp_irqs[i].dstirq, 0);
1167 if (irq > 0 && !IO_APIC_IRQ(irq))
1168 continue;
1169
1170 if (pin == (mp_irqs[i].srcbusirq & 3)) {
1171 best_idx = i;
1172 best_ioapic = ioapic_idx;
1173 goto out;
1174 }
1175
1176 /*
1177 * Use the first all-but-pin matching entry as a
1178 * best-guess fuzzy result for broken mptables.
1179 */
1180 if (best_idx < 0) {
1181 best_idx = i;
1182 best_ioapic = ioapic_idx;
1183 }
1184 }
1185 if (best_idx < 0)
1186 return -1;
1187
1188out:
1189 return pin_2_irq(best_idx, best_ioapic, mp_irqs[best_idx].dstirq,
1190 IOAPIC_MAP_ALLOC);
1191}
1192EXPORT_SYMBOL(IO_APIC_get_PCI_irq_vector);
1193
1194static struct irq_chip ioapic_chip, ioapic_ir_chip;
1195
1196static void __init setup_IO_APIC_irqs(void)
1197{
1198 unsigned int ioapic, pin;
1199 int idx;
1200
1201 apic_printk(APIC_VERBOSE, KERN_DEBUG "init IO_APIC IRQs\n");
1202
1203 for_each_ioapic_pin(ioapic, pin) {
1204 idx = find_irq_entry(ioapic, pin, mp_INT);
1205 if (idx < 0)
1206 apic_printk(APIC_VERBOSE,
1207 KERN_DEBUG " apic %d pin %d not connected\n",
1208 mpc_ioapic_id(ioapic), pin);
1209 else
1210 pin_2_irq(idx, ioapic, pin,
1211 ioapic ? 0 : IOAPIC_MAP_ALLOC);
1212 }
1213}
1214
1215void ioapic_zap_locks(void)
1216{
1217 raw_spin_lock_init(&ioapic_lock);
1218}
1219
1220static void io_apic_print_entries(unsigned int apic, unsigned int nr_entries)
1221{
1222 int i;
1223 char buf[256];
1224 struct IO_APIC_route_entry entry;
1225 struct IR_IO_APIC_route_entry *ir_entry = (void *)&entry;
1226
1227 printk(KERN_DEBUG "IOAPIC %d:\n", apic);
1228 for (i = 0; i <= nr_entries; i++) {
1229 entry = ioapic_read_entry(apic, i);
1230 snprintf(buf, sizeof(buf),
1231 " pin%02x, %s, %s, %s, V(%02X), IRR(%1d), S(%1d)",
1232 i,
1233 entry.mask == IOAPIC_MASKED ? "disabled" : "enabled ",
1234 entry.trigger == IOAPIC_LEVEL ? "level" : "edge ",
1235 entry.polarity == IOAPIC_POL_LOW ? "low " : "high",
1236 entry.vector, entry.irr, entry.delivery_status);
1237 if (ir_entry->format)
1238 printk(KERN_DEBUG "%s, remapped, I(%04X), Z(%X)\n",
1239 buf, (ir_entry->index2 << 15) | ir_entry->index,
1240 ir_entry->zero);
1241 else
1242 printk(KERN_DEBUG "%s, %s, D(%02X), M(%1d)\n",
1243 buf,
1244 entry.dest_mode == IOAPIC_DEST_MODE_LOGICAL ?
1245 "logical " : "physical",
1246 entry.dest, entry.delivery_mode);
1247 }
1248}
1249
1250static void __init print_IO_APIC(int ioapic_idx)
1251{
1252 union IO_APIC_reg_00 reg_00;
1253 union IO_APIC_reg_01 reg_01;
1254 union IO_APIC_reg_02 reg_02;
1255 union IO_APIC_reg_03 reg_03;
1256 unsigned long flags;
1257
1258 raw_spin_lock_irqsave(&ioapic_lock, flags);
1259 reg_00.raw = io_apic_read(ioapic_idx, 0);
1260 reg_01.raw = io_apic_read(ioapic_idx, 1);
1261 if (reg_01.bits.version >= 0x10)
1262 reg_02.raw = io_apic_read(ioapic_idx, 2);
1263 if (reg_01.bits.version >= 0x20)
1264 reg_03.raw = io_apic_read(ioapic_idx, 3);
1265 raw_spin_unlock_irqrestore(&ioapic_lock, flags);
1266
1267 printk(KERN_DEBUG "IO APIC #%d......\n", mpc_ioapic_id(ioapic_idx));
1268 printk(KERN_DEBUG ".... register #00: %08X\n", reg_00.raw);
1269 printk(KERN_DEBUG "....... : physical APIC id: %02X\n", reg_00.bits.ID);
1270 printk(KERN_DEBUG "....... : Delivery Type: %X\n", reg_00.bits.delivery_type);
1271 printk(KERN_DEBUG "....... : LTS : %X\n", reg_00.bits.LTS);
1272
1273 printk(KERN_DEBUG ".... register #01: %08X\n", *(int *)®_01);
1274 printk(KERN_DEBUG "....... : max redirection entries: %02X\n",
1275 reg_01.bits.entries);
1276
1277 printk(KERN_DEBUG "....... : PRQ implemented: %X\n", reg_01.bits.PRQ);
1278 printk(KERN_DEBUG "....... : IO APIC version: %02X\n",
1279 reg_01.bits.version);
1280
1281 /*
1282 * Some Intel chipsets with IO APIC VERSION of 0x1? don't have reg_02,
1283 * but the value of reg_02 is read as the previous read register
1284 * value, so ignore it if reg_02 == reg_01.
1285 */
1286 if (reg_01.bits.version >= 0x10 && reg_02.raw != reg_01.raw) {
1287 printk(KERN_DEBUG ".... register #02: %08X\n", reg_02.raw);
1288 printk(KERN_DEBUG "....... : arbitration: %02X\n", reg_02.bits.arbitration);
1289 }
1290
1291 /*
1292 * Some Intel chipsets with IO APIC VERSION of 0x2? don't have reg_02
1293 * or reg_03, but the value of reg_0[23] is read as the previous read
1294 * register value, so ignore it if reg_03 == reg_0[12].
1295 */
1296 if (reg_01.bits.version >= 0x20 && reg_03.raw != reg_02.raw &&
1297 reg_03.raw != reg_01.raw) {
1298 printk(KERN_DEBUG ".... register #03: %08X\n", reg_03.raw);
1299 printk(KERN_DEBUG "....... : Boot DT : %X\n", reg_03.bits.boot_DT);
1300 }
1301
1302 printk(KERN_DEBUG ".... IRQ redirection table:\n");
1303 io_apic_print_entries(ioapic_idx, reg_01.bits.entries);
1304}
1305
1306void __init print_IO_APICs(void)
1307{
1308 int ioapic_idx;
1309 unsigned int irq;
1310
1311 printk(KERN_DEBUG "number of MP IRQ sources: %d.\n", mp_irq_entries);
1312 for_each_ioapic(ioapic_idx)
1313 printk(KERN_DEBUG "number of IO-APIC #%d registers: %d.\n",
1314 mpc_ioapic_id(ioapic_idx),
1315 ioapics[ioapic_idx].nr_registers);
1316
1317 /*
1318 * We are a bit conservative about what we expect. We have to
1319 * know about every hardware change ASAP.
1320 */
1321 printk(KERN_INFO "testing the IO APIC.......................\n");
1322
1323 for_each_ioapic(ioapic_idx)
1324 print_IO_APIC(ioapic_idx);
1325
1326 printk(KERN_DEBUG "IRQ to pin mappings:\n");
1327 for_each_active_irq(irq) {
1328 struct irq_pin_list *entry;
1329 struct irq_chip *chip;
1330 struct mp_chip_data *data;
1331
1332 chip = irq_get_chip(irq);
1333 if (chip != &ioapic_chip && chip != &ioapic_ir_chip)
1334 continue;
1335 data = irq_get_chip_data(irq);
1336 if (!data)
1337 continue;
1338 if (list_empty(&data->irq_2_pin))
1339 continue;
1340
1341 printk(KERN_DEBUG "IRQ%d ", irq);
1342 for_each_irq_pin(entry, data->irq_2_pin)
1343 pr_cont("-> %d:%d", entry->apic, entry->pin);
1344 pr_cont("\n");
1345 }
1346
1347 printk(KERN_INFO ".................................... done.\n");
1348}
1349
1350/* Where if anywhere is the i8259 connect in external int mode */
1351static struct { int pin, apic; } ioapic_i8259 = { -1, -1 };
1352
1353void __init enable_IO_APIC(void)
1354{
1355 int i8259_apic, i8259_pin;
1356 int apic, pin;
1357
1358 if (skip_ioapic_setup)
1359 nr_ioapics = 0;
1360
1361 if (!nr_legacy_irqs() || !nr_ioapics)
1362 return;
1363
1364 for_each_ioapic_pin(apic, pin) {
1365 /* See if any of the pins is in ExtINT mode */
1366 struct IO_APIC_route_entry entry = ioapic_read_entry(apic, pin);
1367
1368 /* If the interrupt line is enabled and in ExtInt mode
1369 * I have found the pin where the i8259 is connected.
1370 */
1371 if ((entry.mask == 0) && (entry.delivery_mode == dest_ExtINT)) {
1372 ioapic_i8259.apic = apic;
1373 ioapic_i8259.pin = pin;
1374 goto found_i8259;
1375 }
1376 }
1377 found_i8259:
1378 /* Look to see what if the MP table has reported the ExtINT */
1379 /* If we could not find the appropriate pin by looking at the ioapic
1380 * the i8259 probably is not connected the ioapic but give the
1381 * mptable a chance anyway.
1382 */
1383 i8259_pin = find_isa_irq_pin(0, mp_ExtINT);
1384 i8259_apic = find_isa_irq_apic(0, mp_ExtINT);
1385 /* Trust the MP table if nothing is setup in the hardware */
1386 if ((ioapic_i8259.pin == -1) && (i8259_pin >= 0)) {
1387 printk(KERN_WARNING "ExtINT not setup in hardware but reported by MP table\n");
1388 ioapic_i8259.pin = i8259_pin;
1389 ioapic_i8259.apic = i8259_apic;
1390 }
1391 /* Complain if the MP table and the hardware disagree */
1392 if (((ioapic_i8259.apic != i8259_apic) || (ioapic_i8259.pin != i8259_pin)) &&
1393 (i8259_pin >= 0) && (ioapic_i8259.pin >= 0))
1394 {
1395 printk(KERN_WARNING "ExtINT in hardware and MP table differ\n");
1396 }
1397
1398 /*
1399 * Do not trust the IO-APIC being empty at bootup
1400 */
1401 clear_IO_APIC();
1402}
1403
1404void native_restore_boot_irq_mode(void)
1405{
1406 /*
1407 * If the i8259 is routed through an IOAPIC
1408 * Put that IOAPIC in virtual wire mode
1409 * so legacy interrupts can be delivered.
1410 */
1411 if (ioapic_i8259.pin != -1) {
1412 struct IO_APIC_route_entry entry;
1413
1414 memset(&entry, 0, sizeof(entry));
1415 entry.mask = IOAPIC_UNMASKED;
1416 entry.trigger = IOAPIC_EDGE;
1417 entry.polarity = IOAPIC_POL_HIGH;
1418 entry.dest_mode = IOAPIC_DEST_MODE_PHYSICAL;
1419 entry.delivery_mode = dest_ExtINT;
1420 entry.dest = read_apic_id();
1421
1422 /*
1423 * Add it to the IO-APIC irq-routing table:
1424 */
1425 ioapic_write_entry(ioapic_i8259.apic, ioapic_i8259.pin, entry);
1426 }
1427
1428 if (boot_cpu_has(X86_FEATURE_APIC) || apic_from_smp_config())
1429 disconnect_bsp_APIC(ioapic_i8259.pin != -1);
1430}
1431
1432void restore_boot_irq_mode(void)
1433{
1434 if (!nr_legacy_irqs())
1435 return;
1436
1437 x86_apic_ops.restore();
1438}
1439
1440#ifdef CONFIG_X86_32
1441/*
1442 * function to set the IO-APIC physical IDs based on the
1443 * values stored in the MPC table.
1444 *
1445 * by Matt Domsch <Matt_Domsch@dell.com> Tue Dec 21 12:25:05 CST 1999
1446 */
1447void __init setup_ioapic_ids_from_mpc_nocheck(void)
1448{
1449 union IO_APIC_reg_00 reg_00;
1450 physid_mask_t phys_id_present_map;
1451 int ioapic_idx;
1452 int i;
1453 unsigned char old_id;
1454 unsigned long flags;
1455
1456 /*
1457 * This is broken; anything with a real cpu count has to
1458 * circumvent this idiocy regardless.
1459 */
1460 apic->ioapic_phys_id_map(&phys_cpu_present_map, &phys_id_present_map);
1461
1462 /*
1463 * Set the IOAPIC ID to the value stored in the MPC table.
1464 */
1465 for_each_ioapic(ioapic_idx) {
1466 /* Read the register 0 value */
1467 raw_spin_lock_irqsave(&ioapic_lock, flags);
1468 reg_00.raw = io_apic_read(ioapic_idx, 0);
1469 raw_spin_unlock_irqrestore(&ioapic_lock, flags);
1470
1471 old_id = mpc_ioapic_id(ioapic_idx);
1472
1473 if (mpc_ioapic_id(ioapic_idx) >= get_physical_broadcast()) {
1474 printk(KERN_ERR "BIOS bug, IO-APIC#%d ID is %d in the MPC table!...\n",
1475 ioapic_idx, mpc_ioapic_id(ioapic_idx));
1476 printk(KERN_ERR "... fixing up to %d. (tell your hw vendor)\n",
1477 reg_00.bits.ID);
1478 ioapics[ioapic_idx].mp_config.apicid = reg_00.bits.ID;
1479 }
1480
1481 /*
1482 * Sanity check, is the ID really free? Every APIC in a
1483 * system must have a unique ID or we get lots of nice
1484 * 'stuck on smp_invalidate_needed IPI wait' messages.
1485 */
1486 if (apic->check_apicid_used(&phys_id_present_map,
1487 mpc_ioapic_id(ioapic_idx))) {
1488 printk(KERN_ERR "BIOS bug, IO-APIC#%d ID %d is already used!...\n",
1489 ioapic_idx, mpc_ioapic_id(ioapic_idx));
1490 for (i = 0; i < get_physical_broadcast(); i++)
1491 if (!physid_isset(i, phys_id_present_map))
1492 break;
1493 if (i >= get_physical_broadcast())
1494 panic("Max APIC ID exceeded!\n");
1495 printk(KERN_ERR "... fixing up to %d. (tell your hw vendor)\n",
1496 i);
1497 physid_set(i, phys_id_present_map);
1498 ioapics[ioapic_idx].mp_config.apicid = i;
1499 } else {
1500 physid_mask_t tmp;
1501 apic->apicid_to_cpu_present(mpc_ioapic_id(ioapic_idx),
1502 &tmp);
1503 apic_printk(APIC_VERBOSE, "Setting %d in the "
1504 "phys_id_present_map\n",
1505 mpc_ioapic_id(ioapic_idx));
1506 physids_or(phys_id_present_map, phys_id_present_map, tmp);
1507 }
1508
1509 /*
1510 * We need to adjust the IRQ routing table
1511 * if the ID changed.
1512 */
1513 if (old_id != mpc_ioapic_id(ioapic_idx))
1514 for (i = 0; i < mp_irq_entries; i++)
1515 if (mp_irqs[i].dstapic == old_id)
1516 mp_irqs[i].dstapic
1517 = mpc_ioapic_id(ioapic_idx);
1518
1519 /*
1520 * Update the ID register according to the right value
1521 * from the MPC table if they are different.
1522 */
1523 if (mpc_ioapic_id(ioapic_idx) == reg_00.bits.ID)
1524 continue;
1525
1526 apic_printk(APIC_VERBOSE, KERN_INFO
1527 "...changing IO-APIC physical APIC ID to %d ...",
1528 mpc_ioapic_id(ioapic_idx));
1529
1530 reg_00.bits.ID = mpc_ioapic_id(ioapic_idx);
1531 raw_spin_lock_irqsave(&ioapic_lock, flags);
1532 io_apic_write(ioapic_idx, 0, reg_00.raw);
1533 raw_spin_unlock_irqrestore(&ioapic_lock, flags);
1534
1535 /*
1536 * Sanity check
1537 */
1538 raw_spin_lock_irqsave(&ioapic_lock, flags);
1539 reg_00.raw = io_apic_read(ioapic_idx, 0);
1540 raw_spin_unlock_irqrestore(&ioapic_lock, flags);
1541 if (reg_00.bits.ID != mpc_ioapic_id(ioapic_idx))
1542 pr_cont("could not set ID!\n");
1543 else
1544 apic_printk(APIC_VERBOSE, " ok.\n");
1545 }
1546}
1547
1548void __init setup_ioapic_ids_from_mpc(void)
1549{
1550
1551 if (acpi_ioapic)
1552 return;
1553 /*
1554 * Don't check I/O APIC IDs for xAPIC systems. They have
1555 * no meaning without the serial APIC bus.
1556 */
1557 if (!(boot_cpu_data.x86_vendor == X86_VENDOR_INTEL)
1558 || APIC_XAPIC(boot_cpu_apic_version))
1559 return;
1560 setup_ioapic_ids_from_mpc_nocheck();
1561}
1562#endif
1563
1564int no_timer_check __initdata;
1565
1566static int __init notimercheck(char *s)
1567{
1568 no_timer_check = 1;
1569 return 1;
1570}
1571__setup("no_timer_check", notimercheck);
1572
1573static void __init delay_with_tsc(void)
1574{
1575 unsigned long long start, now;
1576 unsigned long end = jiffies + 4;
1577
1578 start = rdtsc();
1579
1580 /*
1581 * We don't know the TSC frequency yet, but waiting for
1582 * 40000000000/HZ TSC cycles is safe:
1583 * 4 GHz == 10 jiffies
1584 * 1 GHz == 40 jiffies
1585 */
1586 do {
1587 rep_nop();
1588 now = rdtsc();
1589 } while ((now - start) < 40000000000ULL / HZ &&
1590 time_before_eq(jiffies, end));
1591}
1592
1593static void __init delay_without_tsc(void)
1594{
1595 unsigned long end = jiffies + 4;
1596 int band = 1;
1597
1598 /*
1599 * We don't know any frequency yet, but waiting for
1600 * 40940000000/HZ cycles is safe:
1601 * 4 GHz == 10 jiffies
1602 * 1 GHz == 40 jiffies
1603 * 1 << 1 + 1 << 2 +...+ 1 << 11 = 4094
1604 */
1605 do {
1606 __delay(((1U << band++) * 10000000UL) / HZ);
1607 } while (band < 12 && time_before_eq(jiffies, end));
1608}
1609
1610/*
1611 * There is a nasty bug in some older SMP boards, their mptable lies
1612 * about the timer IRQ. We do the following to work around the situation:
1613 *
1614 * - timer IRQ defaults to IO-APIC IRQ
1615 * - if this function detects that timer IRQs are defunct, then we fall
1616 * back to ISA timer IRQs
1617 */
1618static int __init timer_irq_works(void)
1619{
1620 unsigned long t1 = jiffies;
1621 unsigned long flags;
1622
1623 if (no_timer_check)
1624 return 1;
1625
1626 local_save_flags(flags);
1627 local_irq_enable();
1628
1629 if (boot_cpu_has(X86_FEATURE_TSC))
1630 delay_with_tsc();
1631 else
1632 delay_without_tsc();
1633
1634 local_irq_restore(flags);
1635
1636 /*
1637 * Expect a few ticks at least, to be sure some possible
1638 * glue logic does not lock up after one or two first
1639 * ticks in a non-ExtINT mode. Also the local APIC
1640 * might have cached one ExtINT interrupt. Finally, at
1641 * least one tick may be lost due to delays.
1642 */
1643
1644 /* jiffies wrap? */
1645 if (time_after(jiffies, t1 + 4))
1646 return 1;
1647 return 0;
1648}
1649
1650/*
1651 * In the SMP+IOAPIC case it might happen that there are an unspecified
1652 * number of pending IRQ events unhandled. These cases are very rare,
1653 * so we 'resend' these IRQs via IPIs, to the same CPU. It's much
1654 * better to do it this way as thus we do not have to be aware of
1655 * 'pending' interrupts in the IRQ path, except at this point.
1656 */
1657/*
1658 * Edge triggered needs to resend any interrupt
1659 * that was delayed but this is now handled in the device
1660 * independent code.
1661 */
1662
1663/*
1664 * Starting up a edge-triggered IO-APIC interrupt is
1665 * nasty - we need to make sure that we get the edge.
1666 * If it is already asserted for some reason, we need
1667 * return 1 to indicate that is was pending.
1668 *
1669 * This is not complete - we should be able to fake
1670 * an edge even if it isn't on the 8259A...
1671 */
1672static unsigned int startup_ioapic_irq(struct irq_data *data)
1673{
1674 int was_pending = 0, irq = data->irq;
1675 unsigned long flags;
1676
1677 raw_spin_lock_irqsave(&ioapic_lock, flags);
1678 if (irq < nr_legacy_irqs()) {
1679 legacy_pic->mask(irq);
1680 if (legacy_pic->irq_pending(irq))
1681 was_pending = 1;
1682 }
1683 __unmask_ioapic(data->chip_data);
1684 raw_spin_unlock_irqrestore(&ioapic_lock, flags);
1685
1686 return was_pending;
1687}
1688
1689atomic_t irq_mis_count;
1690
1691#ifdef CONFIG_GENERIC_PENDING_IRQ
1692static bool io_apic_level_ack_pending(struct mp_chip_data *data)
1693{
1694 struct irq_pin_list *entry;
1695 unsigned long flags;
1696
1697 raw_spin_lock_irqsave(&ioapic_lock, flags);
1698 for_each_irq_pin(entry, data->irq_2_pin) {
1699 unsigned int reg;
1700 int pin;
1701
1702 pin = entry->pin;
1703 reg = io_apic_read(entry->apic, 0x10 + pin*2);
1704 /* Is the remote IRR bit set? */
1705 if (reg & IO_APIC_REDIR_REMOTE_IRR) {
1706 raw_spin_unlock_irqrestore(&ioapic_lock, flags);
1707 return true;
1708 }
1709 }
1710 raw_spin_unlock_irqrestore(&ioapic_lock, flags);
1711
1712 return false;
1713}
1714
1715static inline bool ioapic_prepare_move(struct irq_data *data)
1716{
1717 /* If we are moving the IRQ we need to mask it */
1718 if (unlikely(irqd_is_setaffinity_pending(data))) {
1719 if (!irqd_irq_masked(data))
1720 mask_ioapic_irq(data);
1721 return true;
1722 }
1723 return false;
1724}
1725
1726static inline void ioapic_finish_move(struct irq_data *data, bool moveit)
1727{
1728 if (unlikely(moveit)) {
1729 /* Only migrate the irq if the ack has been received.
1730 *
1731 * On rare occasions the broadcast level triggered ack gets
1732 * delayed going to ioapics, and if we reprogram the
1733 * vector while Remote IRR is still set the irq will never
1734 * fire again.
1735 *
1736 * To prevent this scenario we read the Remote IRR bit
1737 * of the ioapic. This has two effects.
1738 * - On any sane system the read of the ioapic will
1739 * flush writes (and acks) going to the ioapic from
1740 * this cpu.
1741 * - We get to see if the ACK has actually been delivered.
1742 *
1743 * Based on failed experiments of reprogramming the
1744 * ioapic entry from outside of irq context starting
1745 * with masking the ioapic entry and then polling until
1746 * Remote IRR was clear before reprogramming the
1747 * ioapic I don't trust the Remote IRR bit to be
1748 * completey accurate.
1749 *
1750 * However there appears to be no other way to plug
1751 * this race, so if the Remote IRR bit is not
1752 * accurate and is causing problems then it is a hardware bug
1753 * and you can go talk to the chipset vendor about it.
1754 */
1755 if (!io_apic_level_ack_pending(data->chip_data))
1756 irq_move_masked_irq(data);
1757 /* If the IRQ is masked in the core, leave it: */
1758 if (!irqd_irq_masked(data))
1759 unmask_ioapic_irq(data);
1760 }
1761}
1762#else
1763static inline bool ioapic_prepare_move(struct irq_data *data)
1764{
1765 return false;
1766}
1767static inline void ioapic_finish_move(struct irq_data *data, bool moveit)
1768{
1769}
1770#endif
1771
1772static void ioapic_ack_level(struct irq_data *irq_data)
1773{
1774 struct irq_cfg *cfg = irqd_cfg(irq_data);
1775 unsigned long v;
1776 bool moveit;
1777 int i;
1778
1779 irq_complete_move(cfg);
1780 moveit = ioapic_prepare_move(irq_data);
1781
1782 /*
1783 * It appears there is an erratum which affects at least version 0x11
1784 * of I/O APIC (that's the 82093AA and cores integrated into various
1785 * chipsets). Under certain conditions a level-triggered interrupt is
1786 * erroneously delivered as edge-triggered one but the respective IRR
1787 * bit gets set nevertheless. As a result the I/O unit expects an EOI
1788 * message but it will never arrive and further interrupts are blocked
1789 * from the source. The exact reason is so far unknown, but the
1790 * phenomenon was observed when two consecutive interrupt requests
1791 * from a given source get delivered to the same CPU and the source is
1792 * temporarily disabled in between.
1793 *
1794 * A workaround is to simulate an EOI message manually. We achieve it
1795 * by setting the trigger mode to edge and then to level when the edge
1796 * trigger mode gets detected in the TMR of a local APIC for a
1797 * level-triggered interrupt. We mask the source for the time of the
1798 * operation to prevent an edge-triggered interrupt escaping meanwhile.
1799 * The idea is from Manfred Spraul. --macro
1800 *
1801 * Also in the case when cpu goes offline, fixup_irqs() will forward
1802 * any unhandled interrupt on the offlined cpu to the new cpu
1803 * destination that is handling the corresponding interrupt. This
1804 * interrupt forwarding is done via IPI's. Hence, in this case also
1805 * level-triggered io-apic interrupt will be seen as an edge
1806 * interrupt in the IRR. And we can't rely on the cpu's EOI
1807 * to be broadcasted to the IO-APIC's which will clear the remoteIRR
1808 * corresponding to the level-triggered interrupt. Hence on IO-APIC's
1809 * supporting EOI register, we do an explicit EOI to clear the
1810 * remote IRR and on IO-APIC's which don't have an EOI register,
1811 * we use the above logic (mask+edge followed by unmask+level) from
1812 * Manfred Spraul to clear the remote IRR.
1813 */
1814 i = cfg->vector;
1815 v = apic_read(APIC_TMR + ((i & ~0x1f) >> 1));
1816
1817 /*
1818 * We must acknowledge the irq before we move it or the acknowledge will
1819 * not propagate properly.
1820 */
1821 ack_APIC_irq();
1822
1823 /*
1824 * Tail end of clearing remote IRR bit (either by delivering the EOI
1825 * message via io-apic EOI register write or simulating it using
1826 * mask+edge followed by unnask+level logic) manually when the
1827 * level triggered interrupt is seen as the edge triggered interrupt
1828 * at the cpu.
1829 */
1830 if (!(v & (1 << (i & 0x1f)))) {
1831 atomic_inc(&irq_mis_count);
1832 eoi_ioapic_pin(cfg->vector, irq_data->chip_data);
1833 }
1834
1835 ioapic_finish_move(irq_data, moveit);
1836}
1837
1838static void ioapic_ir_ack_level(struct irq_data *irq_data)
1839{
1840 struct mp_chip_data *data = irq_data->chip_data;
1841
1842 /*
1843 * Intr-remapping uses pin number as the virtual vector
1844 * in the RTE. Actual vector is programmed in
1845 * intr-remapping table entry. Hence for the io-apic
1846 * EOI we use the pin number.
1847 */
1848 apic_ack_irq(irq_data);
1849 eoi_ioapic_pin(data->entry.vector, data);
1850}
1851
1852static void ioapic_configure_entry(struct irq_data *irqd)
1853{
1854 struct mp_chip_data *mpd = irqd->chip_data;
1855 struct irq_cfg *cfg = irqd_cfg(irqd);
1856 struct irq_pin_list *entry;
1857
1858 /*
1859 * Only update when the parent is the vector domain, don't touch it
1860 * if the parent is the remapping domain. Check the installed
1861 * ioapic chip to verify that.
1862 */
1863 if (irqd->chip == &ioapic_chip) {
1864 mpd->entry.dest = cfg->dest_apicid;
1865 mpd->entry.vector = cfg->vector;
1866 }
1867 for_each_irq_pin(entry, mpd->irq_2_pin)
1868 __ioapic_write_entry(entry->apic, entry->pin, mpd->entry);
1869}
1870
1871static int ioapic_set_affinity(struct irq_data *irq_data,
1872 const struct cpumask *mask, bool force)
1873{
1874 struct irq_data *parent = irq_data->parent_data;
1875 unsigned long flags;
1876 int ret;
1877
1878 ret = parent->chip->irq_set_affinity(parent, mask, force);
1879 raw_spin_lock_irqsave(&ioapic_lock, flags);
1880 if (ret >= 0 && ret != IRQ_SET_MASK_OK_DONE)
1881 ioapic_configure_entry(irq_data);
1882 raw_spin_unlock_irqrestore(&ioapic_lock, flags);
1883
1884 return ret;
1885}
1886
1887/*
1888 * Interrupt shutdown masks the ioapic pin, but the interrupt might already
1889 * be in flight, but not yet serviced by the target CPU. That means
1890 * __synchronize_hardirq() would return and claim that everything is calmed
1891 * down. So free_irq() would proceed and deactivate the interrupt and free
1892 * resources.
1893 *
1894 * Once the target CPU comes around to service it it will find a cleared
1895 * vector and complain. While the spurious interrupt is harmless, the full
1896 * release of resources might prevent the interrupt from being acknowledged
1897 * which keeps the hardware in a weird state.
1898 *
1899 * Verify that the corresponding Remote-IRR bits are clear.
1900 */
1901static int ioapic_irq_get_chip_state(struct irq_data *irqd,
1902 enum irqchip_irq_state which,
1903 bool *state)
1904{
1905 struct mp_chip_data *mcd = irqd->chip_data;
1906 struct IO_APIC_route_entry rentry;
1907 struct irq_pin_list *p;
1908
1909 if (which != IRQCHIP_STATE_ACTIVE)
1910 return -EINVAL;
1911
1912 *state = false;
1913 raw_spin_lock(&ioapic_lock);
1914 for_each_irq_pin(p, mcd->irq_2_pin) {
1915 rentry = __ioapic_read_entry(p->apic, p->pin);
1916 /*
1917 * The remote IRR is only valid in level trigger mode. It's
1918 * meaning is undefined for edge triggered interrupts and
1919 * irrelevant because the IO-APIC treats them as fire and
1920 * forget.
1921 */
1922 if (rentry.irr && rentry.trigger) {
1923 *state = true;
1924 break;
1925 }
1926 }
1927 raw_spin_unlock(&ioapic_lock);
1928 return 0;
1929}
1930
1931static struct irq_chip ioapic_chip __read_mostly = {
1932 .name = "IO-APIC",
1933 .irq_startup = startup_ioapic_irq,
1934 .irq_mask = mask_ioapic_irq,
1935 .irq_unmask = unmask_ioapic_irq,
1936 .irq_ack = irq_chip_ack_parent,
1937 .irq_eoi = ioapic_ack_level,
1938 .irq_set_affinity = ioapic_set_affinity,
1939 .irq_retrigger = irq_chip_retrigger_hierarchy,
1940 .irq_get_irqchip_state = ioapic_irq_get_chip_state,
1941 .flags = IRQCHIP_SKIP_SET_WAKE,
1942};
1943
1944static struct irq_chip ioapic_ir_chip __read_mostly = {
1945 .name = "IR-IO-APIC",
1946 .irq_startup = startup_ioapic_irq,
1947 .irq_mask = mask_ioapic_irq,
1948 .irq_unmask = unmask_ioapic_irq,
1949 .irq_ack = irq_chip_ack_parent,
1950 .irq_eoi = ioapic_ir_ack_level,
1951 .irq_set_affinity = ioapic_set_affinity,
1952 .irq_retrigger = irq_chip_retrigger_hierarchy,
1953 .irq_get_irqchip_state = ioapic_irq_get_chip_state,
1954 .flags = IRQCHIP_SKIP_SET_WAKE,
1955};
1956
1957static inline void init_IO_APIC_traps(void)
1958{
1959 struct irq_cfg *cfg;
1960 unsigned int irq;
1961
1962 for_each_active_irq(irq) {
1963 cfg = irq_cfg(irq);
1964 if (IO_APIC_IRQ(irq) && cfg && !cfg->vector) {
1965 /*
1966 * Hmm.. We don't have an entry for this,
1967 * so default to an old-fashioned 8259
1968 * interrupt if we can..
1969 */
1970 if (irq < nr_legacy_irqs())
1971 legacy_pic->make_irq(irq);
1972 else
1973 /* Strange. Oh, well.. */
1974 irq_set_chip(irq, &no_irq_chip);
1975 }
1976 }
1977}
1978
1979/*
1980 * The local APIC irq-chip implementation:
1981 */
1982
1983static void mask_lapic_irq(struct irq_data *data)
1984{
1985 unsigned long v;
1986
1987 v = apic_read(APIC_LVT0);
1988 apic_write(APIC_LVT0, v | APIC_LVT_MASKED);
1989}
1990
1991static void unmask_lapic_irq(struct irq_data *data)
1992{
1993 unsigned long v;
1994
1995 v = apic_read(APIC_LVT0);
1996 apic_write(APIC_LVT0, v & ~APIC_LVT_MASKED);
1997}
1998
1999static void ack_lapic_irq(struct irq_data *data)
2000{
2001 ack_APIC_irq();
2002}
2003
2004static struct irq_chip lapic_chip __read_mostly = {
2005 .name = "local-APIC",
2006 .irq_mask = mask_lapic_irq,
2007 .irq_unmask = unmask_lapic_irq,
2008 .irq_ack = ack_lapic_irq,
2009};
2010
2011static void lapic_register_intr(int irq)
2012{
2013 irq_clear_status_flags(irq, IRQ_LEVEL);
2014 irq_set_chip_and_handler_name(irq, &lapic_chip, handle_edge_irq,
2015 "edge");
2016}
2017
2018/*
2019 * This looks a bit hackish but it's about the only one way of sending
2020 * a few INTA cycles to 8259As and any associated glue logic. ICR does
2021 * not support the ExtINT mode, unfortunately. We need to send these
2022 * cycles as some i82489DX-based boards have glue logic that keeps the
2023 * 8259A interrupt line asserted until INTA. --macro
2024 */
2025static inline void __init unlock_ExtINT_logic(void)
2026{
2027 int apic, pin, i;
2028 struct IO_APIC_route_entry entry0, entry1;
2029 unsigned char save_control, save_freq_select;
2030
2031 pin = find_isa_irq_pin(8, mp_INT);
2032 if (pin == -1) {
2033 WARN_ON_ONCE(1);
2034 return;
2035 }
2036 apic = find_isa_irq_apic(8, mp_INT);
2037 if (apic == -1) {
2038 WARN_ON_ONCE(1);
2039 return;
2040 }
2041
2042 entry0 = ioapic_read_entry(apic, pin);
2043 clear_IO_APIC_pin(apic, pin);
2044
2045 memset(&entry1, 0, sizeof(entry1));
2046
2047 entry1.dest_mode = IOAPIC_DEST_MODE_PHYSICAL;
2048 entry1.mask = IOAPIC_UNMASKED;
2049 entry1.dest = hard_smp_processor_id();
2050 entry1.delivery_mode = dest_ExtINT;
2051 entry1.polarity = entry0.polarity;
2052 entry1.trigger = IOAPIC_EDGE;
2053 entry1.vector = 0;
2054
2055 ioapic_write_entry(apic, pin, entry1);
2056
2057 save_control = CMOS_READ(RTC_CONTROL);
2058 save_freq_select = CMOS_READ(RTC_FREQ_SELECT);
2059 CMOS_WRITE((save_freq_select & ~RTC_RATE_SELECT) | 0x6,
2060 RTC_FREQ_SELECT);
2061 CMOS_WRITE(save_control | RTC_PIE, RTC_CONTROL);
2062
2063 i = 100;
2064 while (i-- > 0) {
2065 mdelay(10);
2066 if ((CMOS_READ(RTC_INTR_FLAGS) & RTC_PF) == RTC_PF)
2067 i -= 10;
2068 }
2069
2070 CMOS_WRITE(save_control, RTC_CONTROL);
2071 CMOS_WRITE(save_freq_select, RTC_FREQ_SELECT);
2072 clear_IO_APIC_pin(apic, pin);
2073
2074 ioapic_write_entry(apic, pin, entry0);
2075}
2076
2077static int disable_timer_pin_1 __initdata;
2078/* Actually the next is obsolete, but keep it for paranoid reasons -AK */
2079static int __init disable_timer_pin_setup(char *arg)
2080{
2081 disable_timer_pin_1 = 1;
2082 return 0;
2083}
2084early_param("disable_timer_pin_1", disable_timer_pin_setup);
2085
2086static int mp_alloc_timer_irq(int ioapic, int pin)
2087{
2088 int irq = -1;
2089 struct irq_domain *domain = mp_ioapic_irqdomain(ioapic);
2090
2091 if (domain) {
2092 struct irq_alloc_info info;
2093
2094 ioapic_set_alloc_attr(&info, NUMA_NO_NODE, 0, 0);
2095 info.ioapic_id = mpc_ioapic_id(ioapic);
2096 info.ioapic_pin = pin;
2097 mutex_lock(&ioapic_mutex);
2098 irq = alloc_isa_irq_from_domain(domain, 0, ioapic, pin, &info);
2099 mutex_unlock(&ioapic_mutex);
2100 }
2101
2102 return irq;
2103}
2104
2105/*
2106 * This code may look a bit paranoid, but it's supposed to cooperate with
2107 * a wide range of boards and BIOS bugs. Fortunately only the timer IRQ
2108 * is so screwy. Thanks to Brian Perkins for testing/hacking this beast
2109 * fanatically on his truly buggy board.
2110 *
2111 * FIXME: really need to revamp this for all platforms.
2112 */
2113static inline void __init check_timer(void)
2114{
2115 struct irq_data *irq_data = irq_get_irq_data(0);
2116 struct mp_chip_data *data = irq_data->chip_data;
2117 struct irq_cfg *cfg = irqd_cfg(irq_data);
2118 int node = cpu_to_node(0);
2119 int apic1, pin1, apic2, pin2;
2120 unsigned long flags;
2121 int no_pin1 = 0;
2122
2123 if (!global_clock_event)
2124 return;
2125
2126 local_irq_save(flags);
2127
2128 /*
2129 * get/set the timer IRQ vector:
2130 */
2131 legacy_pic->mask(0);
2132
2133 /*
2134 * As IRQ0 is to be enabled in the 8259A, the virtual
2135 * wire has to be disabled in the local APIC. Also
2136 * timer interrupts need to be acknowledged manually in
2137 * the 8259A for the i82489DX when using the NMI
2138 * watchdog as that APIC treats NMIs as level-triggered.
2139 * The AEOI mode will finish them in the 8259A
2140 * automatically.
2141 */
2142 apic_write(APIC_LVT0, APIC_LVT_MASKED | APIC_DM_EXTINT);
2143 legacy_pic->init(1);
2144
2145 pin1 = find_isa_irq_pin(0, mp_INT);
2146 apic1 = find_isa_irq_apic(0, mp_INT);
2147 pin2 = ioapic_i8259.pin;
2148 apic2 = ioapic_i8259.apic;
2149
2150 apic_printk(APIC_QUIET, KERN_INFO "..TIMER: vector=0x%02X "
2151 "apic1=%d pin1=%d apic2=%d pin2=%d\n",
2152 cfg->vector, apic1, pin1, apic2, pin2);
2153
2154 /*
2155 * Some BIOS writers are clueless and report the ExtINTA
2156 * I/O APIC input from the cascaded 8259A as the timer
2157 * interrupt input. So just in case, if only one pin
2158 * was found above, try it both directly and through the
2159 * 8259A.
2160 */
2161 if (pin1 == -1) {
2162 panic_if_irq_remap("BIOS bug: timer not connected to IO-APIC");
2163 pin1 = pin2;
2164 apic1 = apic2;
2165 no_pin1 = 1;
2166 } else if (pin2 == -1) {
2167 pin2 = pin1;
2168 apic2 = apic1;
2169 }
2170
2171 if (pin1 != -1) {
2172 /* Ok, does IRQ0 through the IOAPIC work? */
2173 if (no_pin1) {
2174 mp_alloc_timer_irq(apic1, pin1);
2175 } else {
2176 /*
2177 * for edge trigger, it's already unmasked,
2178 * so only need to unmask if it is level-trigger
2179 * do we really have level trigger timer?
2180 */
2181 int idx;
2182 idx = find_irq_entry(apic1, pin1, mp_INT);
2183 if (idx != -1 && irq_trigger(idx))
2184 unmask_ioapic_irq(irq_get_irq_data(0));
2185 }
2186 irq_domain_deactivate_irq(irq_data);
2187 irq_domain_activate_irq(irq_data, false);
2188 if (timer_irq_works()) {
2189 if (disable_timer_pin_1 > 0)
2190 clear_IO_APIC_pin(0, pin1);
2191 goto out;
2192 }
2193 panic_if_irq_remap("timer doesn't work through Interrupt-remapped IO-APIC");
2194 local_irq_disable();
2195 clear_IO_APIC_pin(apic1, pin1);
2196 if (!no_pin1)
2197 apic_printk(APIC_QUIET, KERN_ERR "..MP-BIOS bug: "
2198 "8254 timer not connected to IO-APIC\n");
2199
2200 apic_printk(APIC_QUIET, KERN_INFO "...trying to set up timer "
2201 "(IRQ0) through the 8259A ...\n");
2202 apic_printk(APIC_QUIET, KERN_INFO
2203 "..... (found apic %d pin %d) ...\n", apic2, pin2);
2204 /*
2205 * legacy devices should be connected to IO APIC #0
2206 */
2207 replace_pin_at_irq_node(data, node, apic1, pin1, apic2, pin2);
2208 irq_domain_deactivate_irq(irq_data);
2209 irq_domain_activate_irq(irq_data, false);
2210 legacy_pic->unmask(0);
2211 if (timer_irq_works()) {
2212 apic_printk(APIC_QUIET, KERN_INFO "....... works.\n");
2213 goto out;
2214 }
2215 /*
2216 * Cleanup, just in case ...
2217 */
2218 local_irq_disable();
2219 legacy_pic->mask(0);
2220 clear_IO_APIC_pin(apic2, pin2);
2221 apic_printk(APIC_QUIET, KERN_INFO "....... failed.\n");
2222 }
2223
2224 apic_printk(APIC_QUIET, KERN_INFO
2225 "...trying to set up timer as Virtual Wire IRQ...\n");
2226
2227 lapic_register_intr(0);
2228 apic_write(APIC_LVT0, APIC_DM_FIXED | cfg->vector); /* Fixed mode */
2229 legacy_pic->unmask(0);
2230
2231 if (timer_irq_works()) {
2232 apic_printk(APIC_QUIET, KERN_INFO "..... works.\n");
2233 goto out;
2234 }
2235 local_irq_disable();
2236 legacy_pic->mask(0);
2237 apic_write(APIC_LVT0, APIC_LVT_MASKED | APIC_DM_FIXED | cfg->vector);
2238 apic_printk(APIC_QUIET, KERN_INFO "..... failed.\n");
2239
2240 apic_printk(APIC_QUIET, KERN_INFO
2241 "...trying to set up timer as ExtINT IRQ...\n");
2242
2243 legacy_pic->init(0);
2244 legacy_pic->make_irq(0);
2245 apic_write(APIC_LVT0, APIC_DM_EXTINT);
2246 legacy_pic->unmask(0);
2247
2248 unlock_ExtINT_logic();
2249
2250 if (timer_irq_works()) {
2251 apic_printk(APIC_QUIET, KERN_INFO "..... works.\n");
2252 goto out;
2253 }
2254 local_irq_disable();
2255 apic_printk(APIC_QUIET, KERN_INFO "..... failed :(.\n");
2256 if (apic_is_x2apic_enabled())
2257 apic_printk(APIC_QUIET, KERN_INFO
2258 "Perhaps problem with the pre-enabled x2apic mode\n"
2259 "Try booting with x2apic and interrupt-remapping disabled in the bios.\n");
2260 panic("IO-APIC + timer doesn't work! Boot with apic=debug and send a "
2261 "report. Then try booting with the 'noapic' option.\n");
2262out:
2263 local_irq_restore(flags);
2264}
2265
2266/*
2267 * Traditionally ISA IRQ2 is the cascade IRQ, and is not available
2268 * to devices. However there may be an I/O APIC pin available for
2269 * this interrupt regardless. The pin may be left unconnected, but
2270 * typically it will be reused as an ExtINT cascade interrupt for
2271 * the master 8259A. In the MPS case such a pin will normally be
2272 * reported as an ExtINT interrupt in the MP table. With ACPI
2273 * there is no provision for ExtINT interrupts, and in the absence
2274 * of an override it would be treated as an ordinary ISA I/O APIC
2275 * interrupt, that is edge-triggered and unmasked by default. We
2276 * used to do this, but it caused problems on some systems because
2277 * of the NMI watchdog and sometimes IRQ0 of the 8254 timer using
2278 * the same ExtINT cascade interrupt to drive the local APIC of the
2279 * bootstrap processor. Therefore we refrain from routing IRQ2 to
2280 * the I/O APIC in all cases now. No actual device should request
2281 * it anyway. --macro
2282 */
2283#define PIC_IRQS (1UL << PIC_CASCADE_IR)
2284
2285static int mp_irqdomain_create(int ioapic)
2286{
2287 struct irq_alloc_info info;
2288 struct irq_domain *parent;
2289 int hwirqs = mp_ioapic_pin_count(ioapic);
2290 struct ioapic *ip = &ioapics[ioapic];
2291 struct ioapic_domain_cfg *cfg = &ip->irqdomain_cfg;
2292 struct mp_ioapic_gsi *gsi_cfg = mp_ioapic_gsi_routing(ioapic);
2293 struct fwnode_handle *fn;
2294 char *name = "IO-APIC";
2295
2296 if (cfg->type == IOAPIC_DOMAIN_INVALID)
2297 return 0;
2298
2299 init_irq_alloc_info(&info, NULL);
2300 info.type = X86_IRQ_ALLOC_TYPE_IOAPIC;
2301 info.ioapic_id = mpc_ioapic_id(ioapic);
2302 parent = irq_remapping_get_ir_irq_domain(&info);
2303 if (!parent)
2304 parent = x86_vector_domain;
2305 else
2306 name = "IO-APIC-IR";
2307
2308 /* Handle device tree enumerated APICs proper */
2309 if (cfg->dev) {
2310 fn = of_node_to_fwnode(cfg->dev);
2311 } else {
2312 fn = irq_domain_alloc_named_id_fwnode(name, ioapic);
2313 if (!fn)
2314 return -ENOMEM;
2315 }
2316
2317 ip->irqdomain = irq_domain_create_linear(fn, hwirqs, cfg->ops,
2318 (void *)(long)ioapic);
2319
2320 if (!ip->irqdomain) {
2321 /* Release fw handle if it was allocated above */
2322 if (!cfg->dev)
2323 irq_domain_free_fwnode(fn);
2324 return -ENOMEM;
2325 }
2326
2327 ip->irqdomain->parent = parent;
2328
2329 if (cfg->type == IOAPIC_DOMAIN_LEGACY ||
2330 cfg->type == IOAPIC_DOMAIN_STRICT)
2331 ioapic_dynirq_base = max(ioapic_dynirq_base,
2332 gsi_cfg->gsi_end + 1);
2333
2334 return 0;
2335}
2336
2337static void ioapic_destroy_irqdomain(int idx)
2338{
2339 struct ioapic_domain_cfg *cfg = &ioapics[idx].irqdomain_cfg;
2340 struct fwnode_handle *fn = ioapics[idx].irqdomain->fwnode;
2341
2342 if (ioapics[idx].irqdomain) {
2343 irq_domain_remove(ioapics[idx].irqdomain);
2344 if (!cfg->dev)
2345 irq_domain_free_fwnode(fn);
2346 ioapics[idx].irqdomain = NULL;
2347 }
2348}
2349
2350void __init setup_IO_APIC(void)
2351{
2352 int ioapic;
2353
2354 if (skip_ioapic_setup || !nr_ioapics)
2355 return;
2356
2357 io_apic_irqs = nr_legacy_irqs() ? ~PIC_IRQS : ~0UL;
2358
2359 apic_printk(APIC_VERBOSE, "ENABLING IO-APIC IRQs\n");
2360 for_each_ioapic(ioapic)
2361 BUG_ON(mp_irqdomain_create(ioapic));
2362
2363 /*
2364 * Set up IO-APIC IRQ routing.
2365 */
2366 x86_init.mpparse.setup_ioapic_ids();
2367
2368 sync_Arb_IDs();
2369 setup_IO_APIC_irqs();
2370 init_IO_APIC_traps();
2371 if (nr_legacy_irqs())
2372 check_timer();
2373
2374 ioapic_initialized = 1;
2375}
2376
2377static void resume_ioapic_id(int ioapic_idx)
2378{
2379 unsigned long flags;
2380 union IO_APIC_reg_00 reg_00;
2381
2382 raw_spin_lock_irqsave(&ioapic_lock, flags);
2383 reg_00.raw = io_apic_read(ioapic_idx, 0);
2384 if (reg_00.bits.ID != mpc_ioapic_id(ioapic_idx)) {
2385 reg_00.bits.ID = mpc_ioapic_id(ioapic_idx);
2386 io_apic_write(ioapic_idx, 0, reg_00.raw);
2387 }
2388 raw_spin_unlock_irqrestore(&ioapic_lock, flags);
2389}
2390
2391static void ioapic_resume(void)
2392{
2393 int ioapic_idx;
2394
2395 for_each_ioapic_reverse(ioapic_idx)
2396 resume_ioapic_id(ioapic_idx);
2397
2398 restore_ioapic_entries();
2399}
2400
2401static struct syscore_ops ioapic_syscore_ops = {
2402 .suspend = save_ioapic_entries,
2403 .resume = ioapic_resume,
2404};
2405
2406static int __init ioapic_init_ops(void)
2407{
2408 register_syscore_ops(&ioapic_syscore_ops);
2409
2410 return 0;
2411}
2412
2413device_initcall(ioapic_init_ops);
2414
2415static int io_apic_get_redir_entries(int ioapic)
2416{
2417 union IO_APIC_reg_01 reg_01;
2418 unsigned long flags;
2419
2420 raw_spin_lock_irqsave(&ioapic_lock, flags);
2421 reg_01.raw = io_apic_read(ioapic, 1);
2422 raw_spin_unlock_irqrestore(&ioapic_lock, flags);
2423
2424 /* The register returns the maximum index redir index
2425 * supported, which is one less than the total number of redir
2426 * entries.
2427 */
2428 return reg_01.bits.entries + 1;
2429}
2430
2431unsigned int arch_dynirq_lower_bound(unsigned int from)
2432{
2433 /*
2434 * dmar_alloc_hwirq() may be called before setup_IO_APIC(), so use
2435 * gsi_top if ioapic_dynirq_base hasn't been initialized yet.
2436 */
2437 if (!ioapic_initialized)
2438 return gsi_top;
2439 /*
2440 * For DT enabled machines ioapic_dynirq_base is irrelevant and not
2441 * updated. So simply return @from if ioapic_dynirq_base == 0.
2442 */
2443 return ioapic_dynirq_base ? : from;
2444}
2445
2446#ifdef CONFIG_X86_32
2447static int io_apic_get_unique_id(int ioapic, int apic_id)
2448{
2449 union IO_APIC_reg_00 reg_00;
2450 static physid_mask_t apic_id_map = PHYSID_MASK_NONE;
2451 physid_mask_t tmp;
2452 unsigned long flags;
2453 int i = 0;
2454
2455 /*
2456 * The P4 platform supports up to 256 APIC IDs on two separate APIC
2457 * buses (one for LAPICs, one for IOAPICs), where predecessors only
2458 * supports up to 16 on one shared APIC bus.
2459 *
2460 * TBD: Expand LAPIC/IOAPIC support on P4-class systems to take full
2461 * advantage of new APIC bus architecture.
2462 */
2463
2464 if (physids_empty(apic_id_map))
2465 apic->ioapic_phys_id_map(&phys_cpu_present_map, &apic_id_map);
2466
2467 raw_spin_lock_irqsave(&ioapic_lock, flags);
2468 reg_00.raw = io_apic_read(ioapic, 0);
2469 raw_spin_unlock_irqrestore(&ioapic_lock, flags);
2470
2471 if (apic_id >= get_physical_broadcast()) {
2472 printk(KERN_WARNING "IOAPIC[%d]: Invalid apic_id %d, trying "
2473 "%d\n", ioapic, apic_id, reg_00.bits.ID);
2474 apic_id = reg_00.bits.ID;
2475 }
2476
2477 /*
2478 * Every APIC in a system must have a unique ID or we get lots of nice
2479 * 'stuck on smp_invalidate_needed IPI wait' messages.
2480 */
2481 if (apic->check_apicid_used(&apic_id_map, apic_id)) {
2482
2483 for (i = 0; i < get_physical_broadcast(); i++) {
2484 if (!apic->check_apicid_used(&apic_id_map, i))
2485 break;
2486 }
2487
2488 if (i == get_physical_broadcast())
2489 panic("Max apic_id exceeded!\n");
2490
2491 printk(KERN_WARNING "IOAPIC[%d]: apic_id %d already used, "
2492 "trying %d\n", ioapic, apic_id, i);
2493
2494 apic_id = i;
2495 }
2496
2497 apic->apicid_to_cpu_present(apic_id, &tmp);
2498 physids_or(apic_id_map, apic_id_map, tmp);
2499
2500 if (reg_00.bits.ID != apic_id) {
2501 reg_00.bits.ID = apic_id;
2502
2503 raw_spin_lock_irqsave(&ioapic_lock, flags);
2504 io_apic_write(ioapic, 0, reg_00.raw);
2505 reg_00.raw = io_apic_read(ioapic, 0);
2506 raw_spin_unlock_irqrestore(&ioapic_lock, flags);
2507
2508 /* Sanity check */
2509 if (reg_00.bits.ID != apic_id) {
2510 pr_err("IOAPIC[%d]: Unable to change apic_id!\n",
2511 ioapic);
2512 return -1;
2513 }
2514 }
2515
2516 apic_printk(APIC_VERBOSE, KERN_INFO
2517 "IOAPIC[%d]: Assigned apic_id %d\n", ioapic, apic_id);
2518
2519 return apic_id;
2520}
2521
2522static u8 io_apic_unique_id(int idx, u8 id)
2523{
2524 if ((boot_cpu_data.x86_vendor == X86_VENDOR_INTEL) &&
2525 !APIC_XAPIC(boot_cpu_apic_version))
2526 return io_apic_get_unique_id(idx, id);
2527 else
2528 return id;
2529}
2530#else
2531static u8 io_apic_unique_id(int idx, u8 id)
2532{
2533 union IO_APIC_reg_00 reg_00;
2534 DECLARE_BITMAP(used, 256);
2535 unsigned long flags;
2536 u8 new_id;
2537 int i;
2538
2539 bitmap_zero(used, 256);
2540 for_each_ioapic(i)
2541 __set_bit(mpc_ioapic_id(i), used);
2542
2543 /* Hand out the requested id if available */
2544 if (!test_bit(id, used))
2545 return id;
2546
2547 /*
2548 * Read the current id from the ioapic and keep it if
2549 * available.
2550 */
2551 raw_spin_lock_irqsave(&ioapic_lock, flags);
2552 reg_00.raw = io_apic_read(idx, 0);
2553 raw_spin_unlock_irqrestore(&ioapic_lock, flags);
2554 new_id = reg_00.bits.ID;
2555 if (!test_bit(new_id, used)) {
2556 apic_printk(APIC_VERBOSE, KERN_INFO
2557 "IOAPIC[%d]: Using reg apic_id %d instead of %d\n",
2558 idx, new_id, id);
2559 return new_id;
2560 }
2561
2562 /*
2563 * Get the next free id and write it to the ioapic.
2564 */
2565 new_id = find_first_zero_bit(used, 256);
2566 reg_00.bits.ID = new_id;
2567 raw_spin_lock_irqsave(&ioapic_lock, flags);
2568 io_apic_write(idx, 0, reg_00.raw);
2569 reg_00.raw = io_apic_read(idx, 0);
2570 raw_spin_unlock_irqrestore(&ioapic_lock, flags);
2571 /* Sanity check */
2572 BUG_ON(reg_00.bits.ID != new_id);
2573
2574 return new_id;
2575}
2576#endif
2577
2578static int io_apic_get_version(int ioapic)
2579{
2580 union IO_APIC_reg_01 reg_01;
2581 unsigned long flags;
2582
2583 raw_spin_lock_irqsave(&ioapic_lock, flags);
2584 reg_01.raw = io_apic_read(ioapic, 1);
2585 raw_spin_unlock_irqrestore(&ioapic_lock, flags);
2586
2587 return reg_01.bits.version;
2588}
2589
2590int acpi_get_override_irq(u32 gsi, int *trigger, int *polarity)
2591{
2592 int ioapic, pin, idx;
2593
2594 if (skip_ioapic_setup)
2595 return -1;
2596
2597 ioapic = mp_find_ioapic(gsi);
2598 if (ioapic < 0)
2599 return -1;
2600
2601 pin = mp_find_ioapic_pin(ioapic, gsi);
2602 if (pin < 0)
2603 return -1;
2604
2605 idx = find_irq_entry(ioapic, pin, mp_INT);
2606 if (idx < 0)
2607 return -1;
2608
2609 *trigger = irq_trigger(idx);
2610 *polarity = irq_polarity(idx);
2611 return 0;
2612}
2613
2614/*
2615 * This function updates target affinity of IOAPIC interrupts to include
2616 * the CPUs which came online during SMP bringup.
2617 */
2618#define IOAPIC_RESOURCE_NAME_SIZE 11
2619
2620static struct resource *ioapic_resources;
2621
2622static struct resource * __init ioapic_setup_resources(void)
2623{
2624 unsigned long n;
2625 struct resource *res;
2626 char *mem;
2627 int i;
2628
2629 if (nr_ioapics == 0)
2630 return NULL;
2631
2632 n = IOAPIC_RESOURCE_NAME_SIZE + sizeof(struct resource);
2633 n *= nr_ioapics;
2634
2635 mem = memblock_alloc(n, SMP_CACHE_BYTES);
2636 if (!mem)
2637 panic("%s: Failed to allocate %lu bytes\n", __func__, n);
2638 res = (void *)mem;
2639
2640 mem += sizeof(struct resource) * nr_ioapics;
2641
2642 for_each_ioapic(i) {
2643 res[i].name = mem;
2644 res[i].flags = IORESOURCE_MEM | IORESOURCE_BUSY;
2645 snprintf(mem, IOAPIC_RESOURCE_NAME_SIZE, "IOAPIC %u", i);
2646 mem += IOAPIC_RESOURCE_NAME_SIZE;
2647 ioapics[i].iomem_res = &res[i];
2648 }
2649
2650 ioapic_resources = res;
2651
2652 return res;
2653}
2654
2655void __init io_apic_init_mappings(void)
2656{
2657 unsigned long ioapic_phys, idx = FIX_IO_APIC_BASE_0;
2658 struct resource *ioapic_res;
2659 int i;
2660
2661 ioapic_res = ioapic_setup_resources();
2662 for_each_ioapic(i) {
2663 if (smp_found_config) {
2664 ioapic_phys = mpc_ioapic_addr(i);
2665#ifdef CONFIG_X86_32
2666 if (!ioapic_phys) {
2667 printk(KERN_ERR
2668 "WARNING: bogus zero IO-APIC "
2669 "address found in MPTABLE, "
2670 "disabling IO/APIC support!\n");
2671 smp_found_config = 0;
2672 skip_ioapic_setup = 1;
2673 goto fake_ioapic_page;
2674 }
2675#endif
2676 } else {
2677#ifdef CONFIG_X86_32
2678fake_ioapic_page:
2679#endif
2680 ioapic_phys = (unsigned long)memblock_alloc(PAGE_SIZE,
2681 PAGE_SIZE);
2682 if (!ioapic_phys)
2683 panic("%s: Failed to allocate %lu bytes align=0x%lx\n",
2684 __func__, PAGE_SIZE, PAGE_SIZE);
2685 ioapic_phys = __pa(ioapic_phys);
2686 }
2687 set_fixmap_nocache(idx, ioapic_phys);
2688 apic_printk(APIC_VERBOSE, "mapped IOAPIC to %08lx (%08lx)\n",
2689 __fix_to_virt(idx) + (ioapic_phys & ~PAGE_MASK),
2690 ioapic_phys);
2691 idx++;
2692
2693 ioapic_res->start = ioapic_phys;
2694 ioapic_res->end = ioapic_phys + IO_APIC_SLOT_SIZE - 1;
2695 ioapic_res++;
2696 }
2697}
2698
2699void __init ioapic_insert_resources(void)
2700{
2701 int i;
2702 struct resource *r = ioapic_resources;
2703
2704 if (!r) {
2705 if (nr_ioapics > 0)
2706 printk(KERN_ERR
2707 "IO APIC resources couldn't be allocated.\n");
2708 return;
2709 }
2710
2711 for_each_ioapic(i) {
2712 insert_resource(&iomem_resource, r);
2713 r++;
2714 }
2715}
2716
2717int mp_find_ioapic(u32 gsi)
2718{
2719 int i;
2720
2721 if (nr_ioapics == 0)
2722 return -1;
2723
2724 /* Find the IOAPIC that manages this GSI. */
2725 for_each_ioapic(i) {
2726 struct mp_ioapic_gsi *gsi_cfg = mp_ioapic_gsi_routing(i);
2727 if (gsi >= gsi_cfg->gsi_base && gsi <= gsi_cfg->gsi_end)
2728 return i;
2729 }
2730
2731 printk(KERN_ERR "ERROR: Unable to locate IOAPIC for GSI %d\n", gsi);
2732 return -1;
2733}
2734
2735int mp_find_ioapic_pin(int ioapic, u32 gsi)
2736{
2737 struct mp_ioapic_gsi *gsi_cfg;
2738
2739 if (WARN_ON(ioapic < 0))
2740 return -1;
2741
2742 gsi_cfg = mp_ioapic_gsi_routing(ioapic);
2743 if (WARN_ON(gsi > gsi_cfg->gsi_end))
2744 return -1;
2745
2746 return gsi - gsi_cfg->gsi_base;
2747}
2748
2749static int bad_ioapic_register(int idx)
2750{
2751 union IO_APIC_reg_00 reg_00;
2752 union IO_APIC_reg_01 reg_01;
2753 union IO_APIC_reg_02 reg_02;
2754
2755 reg_00.raw = io_apic_read(idx, 0);
2756 reg_01.raw = io_apic_read(idx, 1);
2757 reg_02.raw = io_apic_read(idx, 2);
2758
2759 if (reg_00.raw == -1 && reg_01.raw == -1 && reg_02.raw == -1) {
2760 pr_warn("I/O APIC 0x%x registers return all ones, skipping!\n",
2761 mpc_ioapic_addr(idx));
2762 return 1;
2763 }
2764
2765 return 0;
2766}
2767
2768static int find_free_ioapic_entry(void)
2769{
2770 int idx;
2771
2772 for (idx = 0; idx < MAX_IO_APICS; idx++)
2773 if (ioapics[idx].nr_registers == 0)
2774 return idx;
2775
2776 return MAX_IO_APICS;
2777}
2778
2779/**
2780 * mp_register_ioapic - Register an IOAPIC device
2781 * @id: hardware IOAPIC ID
2782 * @address: physical address of IOAPIC register area
2783 * @gsi_base: base of GSI associated with the IOAPIC
2784 * @cfg: configuration information for the IOAPIC
2785 */
2786int mp_register_ioapic(int id, u32 address, u32 gsi_base,
2787 struct ioapic_domain_cfg *cfg)
2788{
2789 bool hotplug = !!ioapic_initialized;
2790 struct mp_ioapic_gsi *gsi_cfg;
2791 int idx, ioapic, entries;
2792 u32 gsi_end;
2793
2794 if (!address) {
2795 pr_warn("Bogus (zero) I/O APIC address found, skipping!\n");
2796 return -EINVAL;
2797 }
2798 for_each_ioapic(ioapic)
2799 if (ioapics[ioapic].mp_config.apicaddr == address) {
2800 pr_warn("address 0x%x conflicts with IOAPIC%d\n",
2801 address, ioapic);
2802 return -EEXIST;
2803 }
2804
2805 idx = find_free_ioapic_entry();
2806 if (idx >= MAX_IO_APICS) {
2807 pr_warn("Max # of I/O APICs (%d) exceeded (found %d), skipping\n",
2808 MAX_IO_APICS, idx);
2809 return -ENOSPC;
2810 }
2811
2812 ioapics[idx].mp_config.type = MP_IOAPIC;
2813 ioapics[idx].mp_config.flags = MPC_APIC_USABLE;
2814 ioapics[idx].mp_config.apicaddr = address;
2815
2816 set_fixmap_nocache(FIX_IO_APIC_BASE_0 + idx, address);
2817 if (bad_ioapic_register(idx)) {
2818 clear_fixmap(FIX_IO_APIC_BASE_0 + idx);
2819 return -ENODEV;
2820 }
2821
2822 ioapics[idx].mp_config.apicid = io_apic_unique_id(idx, id);
2823 ioapics[idx].mp_config.apicver = io_apic_get_version(idx);
2824
2825 /*
2826 * Build basic GSI lookup table to facilitate gsi->io_apic lookups
2827 * and to prevent reprogramming of IOAPIC pins (PCI GSIs).
2828 */
2829 entries = io_apic_get_redir_entries(idx);
2830 gsi_end = gsi_base + entries - 1;
2831 for_each_ioapic(ioapic) {
2832 gsi_cfg = mp_ioapic_gsi_routing(ioapic);
2833 if ((gsi_base >= gsi_cfg->gsi_base &&
2834 gsi_base <= gsi_cfg->gsi_end) ||
2835 (gsi_end >= gsi_cfg->gsi_base &&
2836 gsi_end <= gsi_cfg->gsi_end)) {
2837 pr_warn("GSI range [%u-%u] for new IOAPIC conflicts with GSI[%u-%u]\n",
2838 gsi_base, gsi_end,
2839 gsi_cfg->gsi_base, gsi_cfg->gsi_end);
2840 clear_fixmap(FIX_IO_APIC_BASE_0 + idx);
2841 return -ENOSPC;
2842 }
2843 }
2844 gsi_cfg = mp_ioapic_gsi_routing(idx);
2845 gsi_cfg->gsi_base = gsi_base;
2846 gsi_cfg->gsi_end = gsi_end;
2847
2848 ioapics[idx].irqdomain = NULL;
2849 ioapics[idx].irqdomain_cfg = *cfg;
2850
2851 /*
2852 * If mp_register_ioapic() is called during early boot stage when
2853 * walking ACPI/SFI/DT tables, it's too early to create irqdomain,
2854 * we are still using bootmem allocator. So delay it to setup_IO_APIC().
2855 */
2856 if (hotplug) {
2857 if (mp_irqdomain_create(idx)) {
2858 clear_fixmap(FIX_IO_APIC_BASE_0 + idx);
2859 return -ENOMEM;
2860 }
2861 alloc_ioapic_saved_registers(idx);
2862 }
2863
2864 if (gsi_cfg->gsi_end >= gsi_top)
2865 gsi_top = gsi_cfg->gsi_end + 1;
2866 if (nr_ioapics <= idx)
2867 nr_ioapics = idx + 1;
2868
2869 /* Set nr_registers to mark entry present */
2870 ioapics[idx].nr_registers = entries;
2871
2872 pr_info("IOAPIC[%d]: apic_id %d, version %d, address 0x%x, GSI %d-%d\n",
2873 idx, mpc_ioapic_id(idx),
2874 mpc_ioapic_ver(idx), mpc_ioapic_addr(idx),
2875 gsi_cfg->gsi_base, gsi_cfg->gsi_end);
2876
2877 return 0;
2878}
2879
2880int mp_unregister_ioapic(u32 gsi_base)
2881{
2882 int ioapic, pin;
2883 int found = 0;
2884
2885 for_each_ioapic(ioapic)
2886 if (ioapics[ioapic].gsi_config.gsi_base == gsi_base) {
2887 found = 1;
2888 break;
2889 }
2890 if (!found) {
2891 pr_warn("can't find IOAPIC for GSI %d\n", gsi_base);
2892 return -ENODEV;
2893 }
2894
2895 for_each_pin(ioapic, pin) {
2896 u32 gsi = mp_pin_to_gsi(ioapic, pin);
2897 int irq = mp_map_gsi_to_irq(gsi, 0, NULL);
2898 struct mp_chip_data *data;
2899
2900 if (irq >= 0) {
2901 data = irq_get_chip_data(irq);
2902 if (data && data->count) {
2903 pr_warn("pin%d on IOAPIC%d is still in use.\n",
2904 pin, ioapic);
2905 return -EBUSY;
2906 }
2907 }
2908 }
2909
2910 /* Mark entry not present */
2911 ioapics[ioapic].nr_registers = 0;
2912 ioapic_destroy_irqdomain(ioapic);
2913 free_ioapic_saved_registers(ioapic);
2914 if (ioapics[ioapic].iomem_res)
2915 release_resource(ioapics[ioapic].iomem_res);
2916 clear_fixmap(FIX_IO_APIC_BASE_0 + ioapic);
2917 memset(&ioapics[ioapic], 0, sizeof(ioapics[ioapic]));
2918
2919 return 0;
2920}
2921
2922int mp_ioapic_registered(u32 gsi_base)
2923{
2924 int ioapic;
2925
2926 for_each_ioapic(ioapic)
2927 if (ioapics[ioapic].gsi_config.gsi_base == gsi_base)
2928 return 1;
2929
2930 return 0;
2931}
2932
2933static void mp_irqdomain_get_attr(u32 gsi, struct mp_chip_data *data,
2934 struct irq_alloc_info *info)
2935{
2936 if (info && info->ioapic_valid) {
2937 data->trigger = info->ioapic_trigger;
2938 data->polarity = info->ioapic_polarity;
2939 } else if (acpi_get_override_irq(gsi, &data->trigger,
2940 &data->polarity) < 0) {
2941 /* PCI interrupts are always active low level triggered. */
2942 data->trigger = IOAPIC_LEVEL;
2943 data->polarity = IOAPIC_POL_LOW;
2944 }
2945}
2946
2947static void mp_setup_entry(struct irq_cfg *cfg, struct mp_chip_data *data,
2948 struct IO_APIC_route_entry *entry)
2949{
2950 memset(entry, 0, sizeof(*entry));
2951 entry->delivery_mode = apic->irq_delivery_mode;
2952 entry->dest_mode = apic->irq_dest_mode;
2953 entry->dest = cfg->dest_apicid;
2954 entry->vector = cfg->vector;
2955 entry->trigger = data->trigger;
2956 entry->polarity = data->polarity;
2957 /*
2958 * Mask level triggered irqs. Edge triggered irqs are masked
2959 * by the irq core code in case they fire.
2960 */
2961 if (data->trigger == IOAPIC_LEVEL)
2962 entry->mask = IOAPIC_MASKED;
2963 else
2964 entry->mask = IOAPIC_UNMASKED;
2965}
2966
2967int mp_irqdomain_alloc(struct irq_domain *domain, unsigned int virq,
2968 unsigned int nr_irqs, void *arg)
2969{
2970 int ret, ioapic, pin;
2971 struct irq_cfg *cfg;
2972 struct irq_data *irq_data;
2973 struct mp_chip_data *data;
2974 struct irq_alloc_info *info = arg;
2975 unsigned long flags;
2976
2977 if (!info || nr_irqs > 1)
2978 return -EINVAL;
2979 irq_data = irq_domain_get_irq_data(domain, virq);
2980 if (!irq_data)
2981 return -EINVAL;
2982
2983 ioapic = mp_irqdomain_ioapic_idx(domain);
2984 pin = info->ioapic_pin;
2985 if (irq_find_mapping(domain, (irq_hw_number_t)pin) > 0)
2986 return -EEXIST;
2987
2988 data = kzalloc(sizeof(*data), GFP_KERNEL);
2989 if (!data)
2990 return -ENOMEM;
2991
2992 info->ioapic_entry = &data->entry;
2993 ret = irq_domain_alloc_irqs_parent(domain, virq, nr_irqs, info);
2994 if (ret < 0) {
2995 kfree(data);
2996 return ret;
2997 }
2998
2999 INIT_LIST_HEAD(&data->irq_2_pin);
3000 irq_data->hwirq = info->ioapic_pin;
3001 irq_data->chip = (domain->parent == x86_vector_domain) ?
3002 &ioapic_chip : &ioapic_ir_chip;
3003 irq_data->chip_data = data;
3004 mp_irqdomain_get_attr(mp_pin_to_gsi(ioapic, pin), data, info);
3005
3006 cfg = irqd_cfg(irq_data);
3007 add_pin_to_irq_node(data, ioapic_alloc_attr_node(info), ioapic, pin);
3008
3009 local_irq_save(flags);
3010 if (info->ioapic_entry)
3011 mp_setup_entry(cfg, data, info->ioapic_entry);
3012 mp_register_handler(virq, data->trigger);
3013 if (virq < nr_legacy_irqs())
3014 legacy_pic->mask(virq);
3015 local_irq_restore(flags);
3016
3017 apic_printk(APIC_VERBOSE, KERN_DEBUG
3018 "IOAPIC[%d]: Set routing entry (%d-%d -> 0x%x -> IRQ %d Mode:%i Active:%i Dest:%d)\n",
3019 ioapic, mpc_ioapic_id(ioapic), pin, cfg->vector,
3020 virq, data->trigger, data->polarity, cfg->dest_apicid);
3021
3022 return 0;
3023}
3024
3025void mp_irqdomain_free(struct irq_domain *domain, unsigned int virq,
3026 unsigned int nr_irqs)
3027{
3028 struct irq_data *irq_data;
3029 struct mp_chip_data *data;
3030
3031 BUG_ON(nr_irqs != 1);
3032 irq_data = irq_domain_get_irq_data(domain, virq);
3033 if (irq_data && irq_data->chip_data) {
3034 data = irq_data->chip_data;
3035 __remove_pin_from_irq(data, mp_irqdomain_ioapic_idx(domain),
3036 (int)irq_data->hwirq);
3037 WARN_ON(!list_empty(&data->irq_2_pin));
3038 kfree(irq_data->chip_data);
3039 }
3040 irq_domain_free_irqs_top(domain, virq, nr_irqs);
3041}
3042
3043int mp_irqdomain_activate(struct irq_domain *domain,
3044 struct irq_data *irq_data, bool reserve)
3045{
3046 unsigned long flags;
3047
3048 raw_spin_lock_irqsave(&ioapic_lock, flags);
3049 ioapic_configure_entry(irq_data);
3050 raw_spin_unlock_irqrestore(&ioapic_lock, flags);
3051 return 0;
3052}
3053
3054void mp_irqdomain_deactivate(struct irq_domain *domain,
3055 struct irq_data *irq_data)
3056{
3057 /* It won't be called for IRQ with multiple IOAPIC pins associated */
3058 ioapic_mask_entry(mp_irqdomain_ioapic_idx(domain),
3059 (int)irq_data->hwirq);
3060}
3061
3062int mp_irqdomain_ioapic_idx(struct irq_domain *domain)
3063{
3064 return (int)(long)domain->host_data;
3065}
3066
3067const struct irq_domain_ops mp_ioapic_irqdomain_ops = {
3068 .alloc = mp_irqdomain_alloc,
3069 .free = mp_irqdomain_free,
3070 .activate = mp_irqdomain_activate,
3071 .deactivate = mp_irqdomain_deactivate,
3072};