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#include <asm/x86_init.h>
70
71#define for_each_ioapic(idx) \
72 for ((idx) = 0; (idx) < nr_ioapics; (idx)++)
73#define for_each_ioapic_reverse(idx) \
74 for ((idx) = nr_ioapics - 1; (idx) >= 0; (idx)--)
75#define for_each_pin(idx, pin) \
76 for ((pin) = 0; (pin) < ioapics[(idx)].nr_registers; (pin)++)
77#define for_each_ioapic_pin(idx, pin) \
78 for_each_ioapic((idx)) \
79 for_each_pin((idx), (pin))
80#define for_each_irq_pin(entry, head) \
81 list_for_each_entry(entry, &head, list)
82
83static DEFINE_RAW_SPINLOCK(ioapic_lock);
84static DEFINE_MUTEX(ioapic_mutex);
85static unsigned int ioapic_dynirq_base;
86static int ioapic_initialized;
87
88struct irq_pin_list {
89 struct list_head list;
90 int apic, pin;
91};
92
93struct mp_chip_data {
94 struct list_head irq_2_pin;
95 struct IO_APIC_route_entry entry;
96 bool is_level;
97 bool active_low;
98 bool isa_irq;
99 u32 count;
100};
101
102struct mp_ioapic_gsi {
103 u32 gsi_base;
104 u32 gsi_end;
105};
106
107static struct ioapic {
108 /*
109 * # of IRQ routing registers
110 */
111 int nr_registers;
112 /*
113 * Saved state during suspend/resume, or while enabling intr-remap.
114 */
115 struct IO_APIC_route_entry *saved_registers;
116 /* I/O APIC config */
117 struct mpc_ioapic mp_config;
118 /* IO APIC gsi routing info */
119 struct mp_ioapic_gsi gsi_config;
120 struct ioapic_domain_cfg irqdomain_cfg;
121 struct irq_domain *irqdomain;
122 struct resource *iomem_res;
123} ioapics[MAX_IO_APICS];
124
125#define mpc_ioapic_ver(ioapic_idx) ioapics[ioapic_idx].mp_config.apicver
126
127int mpc_ioapic_id(int ioapic_idx)
128{
129 return ioapics[ioapic_idx].mp_config.apicid;
130}
131
132unsigned int mpc_ioapic_addr(int ioapic_idx)
133{
134 return ioapics[ioapic_idx].mp_config.apicaddr;
135}
136
137static inline struct mp_ioapic_gsi *mp_ioapic_gsi_routing(int ioapic_idx)
138{
139 return &ioapics[ioapic_idx].gsi_config;
140}
141
142static inline int mp_ioapic_pin_count(int ioapic)
143{
144 struct mp_ioapic_gsi *gsi_cfg = mp_ioapic_gsi_routing(ioapic_idx: ioapic);
145
146 return gsi_cfg->gsi_end - gsi_cfg->gsi_base + 1;
147}
148
149static inline u32 mp_pin_to_gsi(int ioapic, int pin)
150{
151 return mp_ioapic_gsi_routing(ioapic_idx: ioapic)->gsi_base + pin;
152}
153
154static inline bool mp_is_legacy_irq(int irq)
155{
156 return irq >= 0 && irq < nr_legacy_irqs();
157}
158
159static inline struct irq_domain *mp_ioapic_irqdomain(int ioapic)
160{
161 return ioapics[ioapic].irqdomain;
162}
163
164int nr_ioapics;
165
166/* The one past the highest gsi number used */
167u32 gsi_top;
168
169/* MP IRQ source entries */
170struct mpc_intsrc mp_irqs[MAX_IRQ_SOURCES];
171
172/* # of MP IRQ source entries */
173int mp_irq_entries;
174
175#ifdef CONFIG_EISA
176int mp_bus_id_to_type[MAX_MP_BUSSES];
177#endif
178
179DECLARE_BITMAP(mp_bus_not_pci, MAX_MP_BUSSES);
180
181bool ioapic_is_disabled __ro_after_init;
182
183/**
184 * disable_ioapic_support() - disables ioapic support at runtime
185 */
186void disable_ioapic_support(void)
187{
188#ifdef CONFIG_PCI
189 noioapicquirk = 1;
190 noioapicreroute = -1;
191#endif
192 ioapic_is_disabled = true;
193}
194
195static int __init parse_noapic(char *str)
196{
197 /* disable IO-APIC */
198 disable_ioapic_support();
199 return 0;
200}
201early_param("noapic", parse_noapic);
202
203/* Will be called in mpparse/ACPI codes for saving IRQ info */
204void mp_save_irq(struct mpc_intsrc *m)
205{
206 int i;
207
208 apic_printk(APIC_VERBOSE, "Int: type %d, pol %d, trig %d, bus %02x,"
209 " IRQ %02x, APIC ID %x, APIC INT %02x\n",
210 m->irqtype, m->irqflag & 3, (m->irqflag >> 2) & 3, m->srcbus,
211 m->srcbusirq, m->dstapic, m->dstirq);
212
213 for (i = 0; i < mp_irq_entries; i++) {
214 if (!memcmp(p: &mp_irqs[i], q: m, size: sizeof(*m)))
215 return;
216 }
217
218 memcpy(&mp_irqs[mp_irq_entries], m, sizeof(*m));
219 if (++mp_irq_entries == MAX_IRQ_SOURCES)
220 panic(fmt: "Max # of irq sources exceeded!!\n");
221}
222
223static void alloc_ioapic_saved_registers(int idx)
224{
225 size_t size;
226
227 if (ioapics[idx].saved_registers)
228 return;
229
230 size = sizeof(struct IO_APIC_route_entry) * ioapics[idx].nr_registers;
231 ioapics[idx].saved_registers = kzalloc(size, GFP_KERNEL);
232 if (!ioapics[idx].saved_registers)
233 pr_err("IOAPIC %d: suspend/resume impossible!\n", idx);
234}
235
236static void free_ioapic_saved_registers(int idx)
237{
238 kfree(objp: ioapics[idx].saved_registers);
239 ioapics[idx].saved_registers = NULL;
240}
241
242int __init arch_early_ioapic_init(void)
243{
244 int i;
245
246 if (!nr_legacy_irqs())
247 io_apic_irqs = ~0UL;
248
249 for_each_ioapic(i)
250 alloc_ioapic_saved_registers(idx: i);
251
252 return 0;
253}
254
255struct io_apic {
256 unsigned int index;
257 unsigned int unused[3];
258 unsigned int data;
259 unsigned int unused2[11];
260 unsigned int eoi;
261};
262
263static __attribute_const__ struct io_apic __iomem *io_apic_base(int idx)
264{
265 return (void __iomem *) __fix_to_virt(FIX_IO_APIC_BASE_0 + idx)
266 + (mpc_ioapic_addr(ioapic_idx: idx) & ~PAGE_MASK);
267}
268
269static inline void io_apic_eoi(unsigned int apic, unsigned int vector)
270{
271 struct io_apic __iomem *io_apic = io_apic_base(idx: apic);
272 writel(val: vector, addr: &io_apic->eoi);
273}
274
275unsigned int native_io_apic_read(unsigned int apic, unsigned int reg)
276{
277 struct io_apic __iomem *io_apic = io_apic_base(idx: apic);
278 writel(val: reg, addr: &io_apic->index);
279 return readl(addr: &io_apic->data);
280}
281
282static void io_apic_write(unsigned int apic, unsigned int reg,
283 unsigned int value)
284{
285 struct io_apic __iomem *io_apic = io_apic_base(idx: apic);
286
287 writel(val: reg, addr: &io_apic->index);
288 writel(val: value, addr: &io_apic->data);
289}
290
291static struct IO_APIC_route_entry __ioapic_read_entry(int apic, int pin)
292{
293 struct IO_APIC_route_entry entry;
294
295 entry.w1 = io_apic_read(apic, reg: 0x10 + 2 * pin);
296 entry.w2 = io_apic_read(apic, reg: 0x11 + 2 * pin);
297
298 return entry;
299}
300
301static struct IO_APIC_route_entry ioapic_read_entry(int apic, int pin)
302{
303 struct IO_APIC_route_entry entry;
304 unsigned long flags;
305
306 raw_spin_lock_irqsave(&ioapic_lock, flags);
307 entry = __ioapic_read_entry(apic, pin);
308 raw_spin_unlock_irqrestore(&ioapic_lock, flags);
309
310 return entry;
311}
312
313/*
314 * When we write a new IO APIC routing entry, we need to write the high
315 * word first! If the mask bit in the low word is clear, we will enable
316 * the interrupt, and we need to make sure the entry is fully populated
317 * before that happens.
318 */
319static void __ioapic_write_entry(int apic, int pin, struct IO_APIC_route_entry e)
320{
321 io_apic_write(apic, reg: 0x11 + 2*pin, value: e.w2);
322 io_apic_write(apic, reg: 0x10 + 2*pin, value: e.w1);
323}
324
325static void ioapic_write_entry(int apic, int pin, struct IO_APIC_route_entry e)
326{
327 unsigned long flags;
328
329 raw_spin_lock_irqsave(&ioapic_lock, flags);
330 __ioapic_write_entry(apic, pin, e);
331 raw_spin_unlock_irqrestore(&ioapic_lock, flags);
332}
333
334/*
335 * When we mask an IO APIC routing entry, we need to write the low
336 * word first, in order to set the mask bit before we change the
337 * high bits!
338 */
339static void ioapic_mask_entry(int apic, int pin)
340{
341 struct IO_APIC_route_entry e = { .masked = true };
342 unsigned long flags;
343
344 raw_spin_lock_irqsave(&ioapic_lock, flags);
345 io_apic_write(apic, reg: 0x10 + 2*pin, value: e.w1);
346 io_apic_write(apic, reg: 0x11 + 2*pin, value: e.w2);
347 raw_spin_unlock_irqrestore(&ioapic_lock, flags);
348}
349
350/*
351 * The common case is 1:1 IRQ<->pin mappings. Sometimes there are
352 * shared ISA-space IRQs, so we have to support them. We are super
353 * fast in the common case, and fast for shared ISA-space IRQs.
354 */
355static int __add_pin_to_irq_node(struct mp_chip_data *data,
356 int node, int apic, int pin)
357{
358 struct irq_pin_list *entry;
359
360 /* don't allow duplicates */
361 for_each_irq_pin(entry, data->irq_2_pin)
362 if (entry->apic == apic && entry->pin == pin)
363 return 0;
364
365 entry = kzalloc_node(size: sizeof(struct irq_pin_list), GFP_ATOMIC, node);
366 if (!entry) {
367 pr_err("can not alloc irq_pin_list (%d,%d,%d)\n",
368 node, apic, pin);
369 return -ENOMEM;
370 }
371 entry->apic = apic;
372 entry->pin = pin;
373 list_add_tail(new: &entry->list, head: &data->irq_2_pin);
374
375 return 0;
376}
377
378static void __remove_pin_from_irq(struct mp_chip_data *data, int apic, int pin)
379{
380 struct irq_pin_list *tmp, *entry;
381
382 list_for_each_entry_safe(entry, tmp, &data->irq_2_pin, list)
383 if (entry->apic == apic && entry->pin == pin) {
384 list_del(entry: &entry->list);
385 kfree(objp: entry);
386 return;
387 }
388}
389
390static void add_pin_to_irq_node(struct mp_chip_data *data,
391 int node, int apic, int pin)
392{
393 if (__add_pin_to_irq_node(data, node, apic, pin))
394 panic(fmt: "IO-APIC: failed to add irq-pin. Can not proceed\n");
395}
396
397/*
398 * Reroute an IRQ to a different pin.
399 */
400static void __init replace_pin_at_irq_node(struct mp_chip_data *data, int node,
401 int oldapic, int oldpin,
402 int newapic, int newpin)
403{
404 struct irq_pin_list *entry;
405
406 for_each_irq_pin(entry, data->irq_2_pin) {
407 if (entry->apic == oldapic && entry->pin == oldpin) {
408 entry->apic = newapic;
409 entry->pin = newpin;
410 /* every one is different, right? */
411 return;
412 }
413 }
414
415 /* old apic/pin didn't exist, so just add new ones */
416 add_pin_to_irq_node(data, node, apic: newapic, pin: newpin);
417}
418
419static void io_apic_modify_irq(struct mp_chip_data *data, bool masked,
420 void (*final)(struct irq_pin_list *entry))
421{
422 struct irq_pin_list *entry;
423
424 data->entry.masked = masked;
425
426 for_each_irq_pin(entry, data->irq_2_pin) {
427 io_apic_write(apic: entry->apic, reg: 0x10 + 2 * entry->pin, value: data->entry.w1);
428 if (final)
429 final(entry);
430 }
431}
432
433static void io_apic_sync(struct irq_pin_list *entry)
434{
435 /*
436 * Synchronize the IO-APIC and the CPU by doing
437 * a dummy read from the IO-APIC
438 */
439 struct io_apic __iomem *io_apic;
440
441 io_apic = io_apic_base(idx: entry->apic);
442 readl(addr: &io_apic->data);
443}
444
445static void mask_ioapic_irq(struct irq_data *irq_data)
446{
447 struct mp_chip_data *data = irq_data->chip_data;
448 unsigned long flags;
449
450 raw_spin_lock_irqsave(&ioapic_lock, flags);
451 io_apic_modify_irq(data, masked: true, final: &io_apic_sync);
452 raw_spin_unlock_irqrestore(&ioapic_lock, flags);
453}
454
455static void __unmask_ioapic(struct mp_chip_data *data)
456{
457 io_apic_modify_irq(data, masked: false, NULL);
458}
459
460static void unmask_ioapic_irq(struct irq_data *irq_data)
461{
462 struct mp_chip_data *data = irq_data->chip_data;
463 unsigned long flags;
464
465 raw_spin_lock_irqsave(&ioapic_lock, flags);
466 __unmask_ioapic(data);
467 raw_spin_unlock_irqrestore(&ioapic_lock, flags);
468}
469
470/*
471 * IO-APIC versions below 0x20 don't support EOI register.
472 * For the record, here is the information about various versions:
473 * 0Xh 82489DX
474 * 1Xh I/OAPIC or I/O(x)APIC which are not PCI 2.2 Compliant
475 * 2Xh I/O(x)APIC which is PCI 2.2 Compliant
476 * 30h-FFh Reserved
477 *
478 * Some of the Intel ICH Specs (ICH2 to ICH5) documents the io-apic
479 * version as 0x2. This is an error with documentation and these ICH chips
480 * use io-apic's of version 0x20.
481 *
482 * For IO-APIC's with EOI register, we use that to do an explicit EOI.
483 * Otherwise, we simulate the EOI message manually by changing the trigger
484 * mode to edge and then back to level, with RTE being masked during this.
485 */
486static void __eoi_ioapic_pin(int apic, int pin, int vector)
487{
488 if (mpc_ioapic_ver(apic) >= 0x20) {
489 io_apic_eoi(apic, vector);
490 } else {
491 struct IO_APIC_route_entry entry, entry1;
492
493 entry = entry1 = __ioapic_read_entry(apic, pin);
494
495 /*
496 * Mask the entry and change the trigger mode to edge.
497 */
498 entry1.masked = true;
499 entry1.is_level = false;
500
501 __ioapic_write_entry(apic, pin, e: entry1);
502
503 /*
504 * Restore the previous level triggered entry.
505 */
506 __ioapic_write_entry(apic, pin, e: entry);
507 }
508}
509
510static void eoi_ioapic_pin(int vector, struct mp_chip_data *data)
511{
512 unsigned long flags;
513 struct irq_pin_list *entry;
514
515 raw_spin_lock_irqsave(&ioapic_lock, flags);
516 for_each_irq_pin(entry, data->irq_2_pin)
517 __eoi_ioapic_pin(apic: entry->apic, pin: entry->pin, vector);
518 raw_spin_unlock_irqrestore(&ioapic_lock, flags);
519}
520
521static void clear_IO_APIC_pin(unsigned int apic, unsigned int pin)
522{
523 struct IO_APIC_route_entry entry;
524
525 /* Check delivery_mode to be sure we're not clearing an SMI pin */
526 entry = ioapic_read_entry(apic, pin);
527 if (entry.delivery_mode == APIC_DELIVERY_MODE_SMI)
528 return;
529
530 /*
531 * Make sure the entry is masked and re-read the contents to check
532 * if it is a level triggered pin and if the remote-IRR is set.
533 */
534 if (!entry.masked) {
535 entry.masked = true;
536 ioapic_write_entry(apic, pin, e: entry);
537 entry = ioapic_read_entry(apic, pin);
538 }
539
540 if (entry.irr) {
541 unsigned long flags;
542
543 /*
544 * Make sure the trigger mode is set to level. Explicit EOI
545 * doesn't clear the remote-IRR if the trigger mode is not
546 * set to level.
547 */
548 if (!entry.is_level) {
549 entry.is_level = true;
550 ioapic_write_entry(apic, pin, e: entry);
551 }
552 raw_spin_lock_irqsave(&ioapic_lock, flags);
553 __eoi_ioapic_pin(apic, pin, vector: entry.vector);
554 raw_spin_unlock_irqrestore(&ioapic_lock, flags);
555 }
556
557 /*
558 * Clear the rest of the bits in the IO-APIC RTE except for the mask
559 * bit.
560 */
561 ioapic_mask_entry(apic, pin);
562 entry = ioapic_read_entry(apic, pin);
563 if (entry.irr)
564 pr_err("Unable to reset IRR for apic: %d, pin :%d\n",
565 mpc_ioapic_id(apic), pin);
566}
567
568void clear_IO_APIC (void)
569{
570 int apic, pin;
571
572 for_each_ioapic_pin(apic, pin)
573 clear_IO_APIC_pin(apic, pin);
574}
575
576#ifdef CONFIG_X86_32
577/*
578 * support for broken MP BIOSs, enables hand-redirection of PIRQ0-7 to
579 * specific CPU-side IRQs.
580 */
581
582#define MAX_PIRQS 8
583static int pirq_entries[MAX_PIRQS] = {
584 [0 ... MAX_PIRQS - 1] = -1
585};
586
587static int __init ioapic_pirq_setup(char *str)
588{
589 int i, max;
590 int ints[MAX_PIRQS+1];
591
592 get_options(str, ARRAY_SIZE(ints), ints);
593
594 apic_printk(APIC_VERBOSE, KERN_INFO
595 "PIRQ redirection, working around broken MP-BIOS.\n");
596 max = MAX_PIRQS;
597 if (ints[0] < MAX_PIRQS)
598 max = ints[0];
599
600 for (i = 0; i < max; i++) {
601 apic_printk(APIC_VERBOSE, KERN_DEBUG
602 "... PIRQ%d -> IRQ %d\n", i, ints[i+1]);
603 /*
604 * PIRQs are mapped upside down, usually.
605 */
606 pirq_entries[MAX_PIRQS-i-1] = ints[i+1];
607 }
608 return 1;
609}
610
611__setup("pirq=", ioapic_pirq_setup);
612#endif /* CONFIG_X86_32 */
613
614/*
615 * Saves all the IO-APIC RTE's
616 */
617int save_ioapic_entries(void)
618{
619 int apic, pin;
620 int err = 0;
621
622 for_each_ioapic(apic) {
623 if (!ioapics[apic].saved_registers) {
624 err = -ENOMEM;
625 continue;
626 }
627
628 for_each_pin(apic, pin)
629 ioapics[apic].saved_registers[pin] =
630 ioapic_read_entry(apic, pin);
631 }
632
633 return err;
634}
635
636/*
637 * Mask all IO APIC entries.
638 */
639void mask_ioapic_entries(void)
640{
641 int apic, pin;
642
643 for_each_ioapic(apic) {
644 if (!ioapics[apic].saved_registers)
645 continue;
646
647 for_each_pin(apic, pin) {
648 struct IO_APIC_route_entry entry;
649
650 entry = ioapics[apic].saved_registers[pin];
651 if (!entry.masked) {
652 entry.masked = true;
653 ioapic_write_entry(apic, pin, e: entry);
654 }
655 }
656 }
657}
658
659/*
660 * Restore IO APIC entries which was saved in the ioapic structure.
661 */
662int restore_ioapic_entries(void)
663{
664 int apic, pin;
665
666 for_each_ioapic(apic) {
667 if (!ioapics[apic].saved_registers)
668 continue;
669
670 for_each_pin(apic, pin)
671 ioapic_write_entry(apic, pin,
672 e: ioapics[apic].saved_registers[pin]);
673 }
674 return 0;
675}
676
677/*
678 * Find the IRQ entry number of a certain pin.
679 */
680static int find_irq_entry(int ioapic_idx, int pin, int type)
681{
682 int i;
683
684 for (i = 0; i < mp_irq_entries; i++)
685 if (mp_irqs[i].irqtype == type &&
686 (mp_irqs[i].dstapic == mpc_ioapic_id(ioapic_idx) ||
687 mp_irqs[i].dstapic == MP_APIC_ALL) &&
688 mp_irqs[i].dstirq == pin)
689 return i;
690
691 return -1;
692}
693
694/*
695 * Find the pin to which IRQ[irq] (ISA) is connected
696 */
697static int __init find_isa_irq_pin(int irq, int type)
698{
699 int i;
700
701 for (i = 0; i < mp_irq_entries; i++) {
702 int lbus = mp_irqs[i].srcbus;
703
704 if (test_bit(lbus, mp_bus_not_pci) &&
705 (mp_irqs[i].irqtype == type) &&
706 (mp_irqs[i].srcbusirq == irq))
707
708 return mp_irqs[i].dstirq;
709 }
710 return -1;
711}
712
713static int __init find_isa_irq_apic(int irq, int type)
714{
715 int i;
716
717 for (i = 0; i < mp_irq_entries; i++) {
718 int lbus = mp_irqs[i].srcbus;
719
720 if (test_bit(lbus, mp_bus_not_pci) &&
721 (mp_irqs[i].irqtype == type) &&
722 (mp_irqs[i].srcbusirq == irq))
723 break;
724 }
725
726 if (i < mp_irq_entries) {
727 int ioapic_idx;
728
729 for_each_ioapic(ioapic_idx)
730 if (mpc_ioapic_id(ioapic_idx) == mp_irqs[i].dstapic)
731 return ioapic_idx;
732 }
733
734 return -1;
735}
736
737static bool irq_active_low(int idx)
738{
739 int bus = mp_irqs[idx].srcbus;
740
741 /*
742 * Determine IRQ line polarity (high active or low active):
743 */
744 switch (mp_irqs[idx].irqflag & MP_IRQPOL_MASK) {
745 case MP_IRQPOL_DEFAULT:
746 /*
747 * Conforms to spec, ie. bus-type dependent polarity. PCI
748 * defaults to low active. [E]ISA defaults to high active.
749 */
750 return !test_bit(bus, mp_bus_not_pci);
751 case MP_IRQPOL_ACTIVE_HIGH:
752 return false;
753 case MP_IRQPOL_RESERVED:
754 pr_warn("IOAPIC: Invalid polarity: 2, defaulting to low\n");
755 fallthrough;
756 case MP_IRQPOL_ACTIVE_LOW:
757 default: /* Pointless default required due to do gcc stupidity */
758 return true;
759 }
760}
761
762#ifdef CONFIG_EISA
763/*
764 * EISA Edge/Level control register, ELCR
765 */
766static bool EISA_ELCR(unsigned int irq)
767{
768 if (irq < nr_legacy_irqs()) {
769 unsigned int port = PIC_ELCR1 + (irq >> 3);
770 return (inb(port) >> (irq & 7)) & 1;
771 }
772 apic_printk(APIC_VERBOSE, KERN_INFO
773 "Broken MPtable reports ISA irq %d\n", irq);
774 return false;
775}
776
777/*
778 * EISA interrupts are always active high and can be edge or level
779 * triggered depending on the ELCR value. If an interrupt is listed as
780 * EISA conforming in the MP table, that means its trigger type must be
781 * read in from the ELCR.
782 */
783static bool eisa_irq_is_level(int idx, int bus, bool level)
784{
785 switch (mp_bus_id_to_type[bus]) {
786 case MP_BUS_PCI:
787 case MP_BUS_ISA:
788 return level;
789 case MP_BUS_EISA:
790 return EISA_ELCR(irq: mp_irqs[idx].srcbusirq);
791 }
792 pr_warn("IOAPIC: Invalid srcbus: %d defaulting to level\n", bus);
793 return true;
794}
795#else
796static inline int eisa_irq_is_level(int idx, int bus, bool level)
797{
798 return level;
799}
800#endif
801
802static bool irq_is_level(int idx)
803{
804 int bus = mp_irqs[idx].srcbus;
805 bool level;
806
807 /*
808 * Determine IRQ trigger mode (edge or level sensitive):
809 */
810 switch (mp_irqs[idx].irqflag & MP_IRQTRIG_MASK) {
811 case MP_IRQTRIG_DEFAULT:
812 /*
813 * Conforms to spec, ie. bus-type dependent trigger
814 * mode. PCI defaults to level, ISA to edge.
815 */
816 level = !test_bit(bus, mp_bus_not_pci);
817 /* Take EISA into account */
818 return eisa_irq_is_level(idx, bus, level);
819 case MP_IRQTRIG_EDGE:
820 return false;
821 case MP_IRQTRIG_RESERVED:
822 pr_warn("IOAPIC: Invalid trigger mode 2 defaulting to level\n");
823 fallthrough;
824 case MP_IRQTRIG_LEVEL:
825 default: /* Pointless default required due to do gcc stupidity */
826 return true;
827 }
828}
829
830static int __acpi_get_override_irq(u32 gsi, bool *trigger, bool *polarity)
831{
832 int ioapic, pin, idx;
833
834 if (ioapic_is_disabled)
835 return -1;
836
837 ioapic = mp_find_ioapic(gsi);
838 if (ioapic < 0)
839 return -1;
840
841 pin = mp_find_ioapic_pin(ioapic, gsi);
842 if (pin < 0)
843 return -1;
844
845 idx = find_irq_entry(ioapic_idx: ioapic, pin, type: mp_INT);
846 if (idx < 0)
847 return -1;
848
849 *trigger = irq_is_level(idx);
850 *polarity = irq_active_low(idx);
851 return 0;
852}
853
854#ifdef CONFIG_ACPI
855int acpi_get_override_irq(u32 gsi, int *is_level, int *active_low)
856{
857 *is_level = *active_low = 0;
858 return __acpi_get_override_irq(gsi, trigger: (bool *)is_level,
859 polarity: (bool *)active_low);
860}
861#endif
862
863void ioapic_set_alloc_attr(struct irq_alloc_info *info, int node,
864 int trigger, int polarity)
865{
866 init_irq_alloc_info(info, NULL);
867 info->type = X86_IRQ_ALLOC_TYPE_IOAPIC;
868 info->ioapic.node = node;
869 info->ioapic.is_level = trigger;
870 info->ioapic.active_low = polarity;
871 info->ioapic.valid = 1;
872}
873
874static void ioapic_copy_alloc_attr(struct irq_alloc_info *dst,
875 struct irq_alloc_info *src,
876 u32 gsi, int ioapic_idx, int pin)
877{
878 bool level, pol_low;
879
880 copy_irq_alloc_info(dst, src);
881 dst->type = X86_IRQ_ALLOC_TYPE_IOAPIC;
882 dst->devid = mpc_ioapic_id(ioapic_idx);
883 dst->ioapic.pin = pin;
884 dst->ioapic.valid = 1;
885 if (src && src->ioapic.valid) {
886 dst->ioapic.node = src->ioapic.node;
887 dst->ioapic.is_level = src->ioapic.is_level;
888 dst->ioapic.active_low = src->ioapic.active_low;
889 } else {
890 dst->ioapic.node = NUMA_NO_NODE;
891 if (__acpi_get_override_irq(gsi, trigger: &level, polarity: &pol_low) >= 0) {
892 dst->ioapic.is_level = level;
893 dst->ioapic.active_low = pol_low;
894 } else {
895 /*
896 * PCI interrupts are always active low level
897 * triggered.
898 */
899 dst->ioapic.is_level = true;
900 dst->ioapic.active_low = true;
901 }
902 }
903}
904
905static int ioapic_alloc_attr_node(struct irq_alloc_info *info)
906{
907 return (info && info->ioapic.valid) ? info->ioapic.node : NUMA_NO_NODE;
908}
909
910static void mp_register_handler(unsigned int irq, bool level)
911{
912 irq_flow_handler_t hdl;
913 bool fasteoi;
914
915 if (level) {
916 irq_set_status_flags(irq, set: IRQ_LEVEL);
917 fasteoi = true;
918 } else {
919 irq_clear_status_flags(irq, clr: IRQ_LEVEL);
920 fasteoi = false;
921 }
922
923 hdl = fasteoi ? handle_fasteoi_irq : handle_edge_irq;
924 __irq_set_handler(irq, handle: hdl, is_chained: 0, name: fasteoi ? "fasteoi" : "edge");
925}
926
927static bool mp_check_pin_attr(int irq, struct irq_alloc_info *info)
928{
929 struct mp_chip_data *data = irq_get_chip_data(irq);
930
931 /*
932 * setup_IO_APIC_irqs() programs all legacy IRQs with default trigger
933 * and polarity attributes. So allow the first user to reprogram the
934 * pin with real trigger and polarity attributes.
935 */
936 if (irq < nr_legacy_irqs() && data->count == 1) {
937 if (info->ioapic.is_level != data->is_level)
938 mp_register_handler(irq, level: info->ioapic.is_level);
939 data->entry.is_level = data->is_level = info->ioapic.is_level;
940 data->entry.active_low = data->active_low = info->ioapic.active_low;
941 }
942
943 return data->is_level == info->ioapic.is_level &&
944 data->active_low == info->ioapic.active_low;
945}
946
947static int alloc_irq_from_domain(struct irq_domain *domain, int ioapic, u32 gsi,
948 struct irq_alloc_info *info)
949{
950 bool legacy = false;
951 int irq = -1;
952 int type = ioapics[ioapic].irqdomain_cfg.type;
953
954 switch (type) {
955 case IOAPIC_DOMAIN_LEGACY:
956 /*
957 * Dynamically allocate IRQ number for non-ISA IRQs in the first
958 * 16 GSIs on some weird platforms.
959 */
960 if (!ioapic_initialized || gsi >= nr_legacy_irqs())
961 irq = gsi;
962 legacy = mp_is_legacy_irq(irq);
963 break;
964 case IOAPIC_DOMAIN_STRICT:
965 irq = gsi;
966 break;
967 case IOAPIC_DOMAIN_DYNAMIC:
968 break;
969 default:
970 WARN(1, "ioapic: unknown irqdomain type %d\n", type);
971 return -1;
972 }
973
974 return __irq_domain_alloc_irqs(domain, irq_base: irq, nr_irqs: 1,
975 node: ioapic_alloc_attr_node(info),
976 arg: info, realloc: legacy, NULL);
977}
978
979/*
980 * Need special handling for ISA IRQs because there may be multiple IOAPIC pins
981 * sharing the same ISA IRQ number and irqdomain only supports 1:1 mapping
982 * between IOAPIC pin and IRQ number. A typical IOAPIC has 24 pins, pin 0-15 are
983 * used for legacy IRQs and pin 16-23 are used for PCI IRQs (PIRQ A-H).
984 * When ACPI is disabled, only legacy IRQ numbers (IRQ0-15) are available, and
985 * some BIOSes may use MP Interrupt Source records to override IRQ numbers for
986 * PIRQs instead of reprogramming the interrupt routing logic. Thus there may be
987 * multiple pins sharing the same legacy IRQ number when ACPI is disabled.
988 */
989static int alloc_isa_irq_from_domain(struct irq_domain *domain,
990 int irq, int ioapic, int pin,
991 struct irq_alloc_info *info)
992{
993 struct mp_chip_data *data;
994 struct irq_data *irq_data = irq_get_irq_data(irq);
995 int node = ioapic_alloc_attr_node(info);
996
997 /*
998 * Legacy ISA IRQ has already been allocated, just add pin to
999 * the pin list associated with this IRQ and program the IOAPIC
1000 * entry. The IOAPIC entry
1001 */
1002 if (irq_data && irq_data->parent_data) {
1003 if (!mp_check_pin_attr(irq, info))
1004 return -EBUSY;
1005 if (__add_pin_to_irq_node(data: irq_data->chip_data, node, apic: ioapic,
1006 pin: info->ioapic.pin))
1007 return -ENOMEM;
1008 } else {
1009 info->flags |= X86_IRQ_ALLOC_LEGACY;
1010 irq = __irq_domain_alloc_irqs(domain, irq_base: irq, nr_irqs: 1, node, arg: info, realloc: true,
1011 NULL);
1012 if (irq >= 0) {
1013 irq_data = irq_domain_get_irq_data(domain, virq: irq);
1014 data = irq_data->chip_data;
1015 data->isa_irq = true;
1016 }
1017 }
1018
1019 return irq;
1020}
1021
1022static int mp_map_pin_to_irq(u32 gsi, int idx, int ioapic, int pin,
1023 unsigned int flags, struct irq_alloc_info *info)
1024{
1025 int irq;
1026 bool legacy = false;
1027 struct irq_alloc_info tmp;
1028 struct mp_chip_data *data;
1029 struct irq_domain *domain = mp_ioapic_irqdomain(ioapic);
1030
1031 if (!domain)
1032 return -ENOSYS;
1033
1034 if (idx >= 0 && test_bit(mp_irqs[idx].srcbus, mp_bus_not_pci)) {
1035 irq = mp_irqs[idx].srcbusirq;
1036 legacy = mp_is_legacy_irq(irq);
1037 /*
1038 * IRQ2 is unusable for historical reasons on systems which
1039 * have a legacy PIC. See the comment vs. IRQ2 further down.
1040 *
1041 * If this gets removed at some point then the related code
1042 * in lapic_assign_system_vectors() needs to be adjusted as
1043 * well.
1044 */
1045 if (legacy && irq == PIC_CASCADE_IR)
1046 return -EINVAL;
1047 }
1048
1049 mutex_lock(&ioapic_mutex);
1050 if (!(flags & IOAPIC_MAP_ALLOC)) {
1051 if (!legacy) {
1052 irq = irq_find_mapping(domain, hwirq: pin);
1053 if (irq == 0)
1054 irq = -ENOENT;
1055 }
1056 } else {
1057 ioapic_copy_alloc_attr(dst: &tmp, src: info, gsi, ioapic_idx: ioapic, pin);
1058 if (legacy)
1059 irq = alloc_isa_irq_from_domain(domain, irq,
1060 ioapic, pin, info: &tmp);
1061 else if ((irq = irq_find_mapping(domain, hwirq: pin)) == 0)
1062 irq = alloc_irq_from_domain(domain, ioapic, gsi, info: &tmp);
1063 else if (!mp_check_pin_attr(irq, info: &tmp))
1064 irq = -EBUSY;
1065 if (irq >= 0) {
1066 data = irq_get_chip_data(irq);
1067 data->count++;
1068 }
1069 }
1070 mutex_unlock(lock: &ioapic_mutex);
1071
1072 return irq;
1073}
1074
1075static int pin_2_irq(int idx, int ioapic, int pin, unsigned int flags)
1076{
1077 u32 gsi = mp_pin_to_gsi(ioapic, pin);
1078
1079 /*
1080 * Debugging check, we are in big trouble if this message pops up!
1081 */
1082 if (mp_irqs[idx].dstirq != pin)
1083 pr_err("broken BIOS or MPTABLE parser, ayiee!!\n");
1084
1085#ifdef CONFIG_X86_32
1086 /*
1087 * PCI IRQ command line redirection. Yes, limits are hardcoded.
1088 */
1089 if ((pin >= 16) && (pin <= 23)) {
1090 if (pirq_entries[pin-16] != -1) {
1091 if (!pirq_entries[pin-16]) {
1092 apic_printk(APIC_VERBOSE, KERN_DEBUG
1093 "disabling PIRQ%d\n", pin-16);
1094 } else {
1095 int irq = pirq_entries[pin-16];
1096 apic_printk(APIC_VERBOSE, KERN_DEBUG
1097 "using PIRQ%d -> IRQ %d\n",
1098 pin-16, irq);
1099 return irq;
1100 }
1101 }
1102 }
1103#endif
1104
1105 return mp_map_pin_to_irq(gsi, idx, ioapic, pin, flags, NULL);
1106}
1107
1108int mp_map_gsi_to_irq(u32 gsi, unsigned int flags, struct irq_alloc_info *info)
1109{
1110 int ioapic, pin, idx;
1111
1112 ioapic = mp_find_ioapic(gsi);
1113 if (ioapic < 0)
1114 return -ENODEV;
1115
1116 pin = mp_find_ioapic_pin(ioapic, gsi);
1117 idx = find_irq_entry(ioapic_idx: ioapic, pin, type: mp_INT);
1118 if ((flags & IOAPIC_MAP_CHECK) && idx < 0)
1119 return -ENODEV;
1120
1121 return mp_map_pin_to_irq(gsi, idx, ioapic, pin, flags, info);
1122}
1123
1124void mp_unmap_irq(int irq)
1125{
1126 struct irq_data *irq_data = irq_get_irq_data(irq);
1127 struct mp_chip_data *data;
1128
1129 if (!irq_data || !irq_data->domain)
1130 return;
1131
1132 data = irq_data->chip_data;
1133 if (!data || data->isa_irq)
1134 return;
1135
1136 mutex_lock(&ioapic_mutex);
1137 if (--data->count == 0)
1138 irq_domain_free_irqs(virq: irq, nr_irqs: 1);
1139 mutex_unlock(lock: &ioapic_mutex);
1140}
1141
1142/*
1143 * Find a specific PCI IRQ entry.
1144 * Not an __init, possibly needed by modules
1145 */
1146int IO_APIC_get_PCI_irq_vector(int bus, int slot, int pin)
1147{
1148 int irq, i, best_ioapic = -1, best_idx = -1;
1149
1150 apic_printk(APIC_DEBUG,
1151 "querying PCI -> IRQ mapping bus:%d, slot:%d, pin:%d.\n",
1152 bus, slot, pin);
1153 if (test_bit(bus, mp_bus_not_pci)) {
1154 apic_printk(APIC_VERBOSE,
1155 "PCI BIOS passed nonexistent PCI bus %d!\n", bus);
1156 return -1;
1157 }
1158
1159 for (i = 0; i < mp_irq_entries; i++) {
1160 int lbus = mp_irqs[i].srcbus;
1161 int ioapic_idx, found = 0;
1162
1163 if (bus != lbus || mp_irqs[i].irqtype != mp_INT ||
1164 slot != ((mp_irqs[i].srcbusirq >> 2) & 0x1f))
1165 continue;
1166
1167 for_each_ioapic(ioapic_idx)
1168 if (mpc_ioapic_id(ioapic_idx) == mp_irqs[i].dstapic ||
1169 mp_irqs[i].dstapic == MP_APIC_ALL) {
1170 found = 1;
1171 break;
1172 }
1173 if (!found)
1174 continue;
1175
1176 /* Skip ISA IRQs */
1177 irq = pin_2_irq(idx: i, ioapic: ioapic_idx, pin: mp_irqs[i].dstirq, flags: 0);
1178 if (irq > 0 && !IO_APIC_IRQ(irq))
1179 continue;
1180
1181 if (pin == (mp_irqs[i].srcbusirq & 3)) {
1182 best_idx = i;
1183 best_ioapic = ioapic_idx;
1184 goto out;
1185 }
1186
1187 /*
1188 * Use the first all-but-pin matching entry as a
1189 * best-guess fuzzy result for broken mptables.
1190 */
1191 if (best_idx < 0) {
1192 best_idx = i;
1193 best_ioapic = ioapic_idx;
1194 }
1195 }
1196 if (best_idx < 0)
1197 return -1;
1198
1199out:
1200 return pin_2_irq(idx: best_idx, ioapic: best_ioapic, pin: mp_irqs[best_idx].dstirq,
1201 IOAPIC_MAP_ALLOC);
1202}
1203EXPORT_SYMBOL(IO_APIC_get_PCI_irq_vector);
1204
1205static struct irq_chip ioapic_chip, ioapic_ir_chip;
1206
1207static void __init setup_IO_APIC_irqs(void)
1208{
1209 unsigned int ioapic, pin;
1210 int idx;
1211
1212 apic_printk(APIC_VERBOSE, KERN_DEBUG "init IO_APIC IRQs\n");
1213
1214 for_each_ioapic_pin(ioapic, pin) {
1215 idx = find_irq_entry(ioapic_idx: ioapic, pin, type: mp_INT);
1216 if (idx < 0)
1217 apic_printk(APIC_VERBOSE,
1218 KERN_DEBUG " apic %d pin %d not connected\n",
1219 mpc_ioapic_id(ioapic), pin);
1220 else
1221 pin_2_irq(idx, ioapic, pin,
1222 flags: ioapic ? 0 : IOAPIC_MAP_ALLOC);
1223 }
1224}
1225
1226void ioapic_zap_locks(void)
1227{
1228 raw_spin_lock_init(&ioapic_lock);
1229}
1230
1231static void io_apic_print_entries(unsigned int apic, unsigned int nr_entries)
1232{
1233 struct IO_APIC_route_entry entry;
1234 char buf[256];
1235 int i;
1236
1237 printk(KERN_DEBUG "IOAPIC %d:\n", apic);
1238 for (i = 0; i <= nr_entries; i++) {
1239 entry = ioapic_read_entry(apic, pin: i);
1240 snprintf(buf, size: sizeof(buf),
1241 fmt: " pin%02x, %s, %s, %s, V(%02X), IRR(%1d), S(%1d)",
1242 i,
1243 entry.masked ? "disabled" : "enabled ",
1244 entry.is_level ? "level" : "edge ",
1245 entry.active_low ? "low " : "high",
1246 entry.vector, entry.irr, entry.delivery_status);
1247 if (entry.ir_format) {
1248 printk(KERN_DEBUG "%s, remapped, I(%04X), Z(%X)\n",
1249 buf,
1250 (entry.ir_index_15 << 15) | entry.ir_index_0_14,
1251 entry.ir_zero);
1252 } else {
1253 printk(KERN_DEBUG "%s, %s, D(%02X%02X), M(%1d)\n", buf,
1254 entry.dest_mode_logical ? "logical " : "physical",
1255 entry.virt_destid_8_14, entry.destid_0_7,
1256 entry.delivery_mode);
1257 }
1258 }
1259}
1260
1261static void __init print_IO_APIC(int ioapic_idx)
1262{
1263 union IO_APIC_reg_00 reg_00;
1264 union IO_APIC_reg_01 reg_01;
1265 union IO_APIC_reg_02 reg_02;
1266 union IO_APIC_reg_03 reg_03;
1267 unsigned long flags;
1268
1269 raw_spin_lock_irqsave(&ioapic_lock, flags);
1270 reg_00.raw = io_apic_read(apic: ioapic_idx, reg: 0);
1271 reg_01.raw = io_apic_read(apic: ioapic_idx, reg: 1);
1272 if (reg_01.bits.version >= 0x10)
1273 reg_02.raw = io_apic_read(apic: ioapic_idx, reg: 2);
1274 if (reg_01.bits.version >= 0x20)
1275 reg_03.raw = io_apic_read(apic: ioapic_idx, reg: 3);
1276 raw_spin_unlock_irqrestore(&ioapic_lock, flags);
1277
1278 printk(KERN_DEBUG "IO APIC #%d......\n", mpc_ioapic_id(ioapic_idx));
1279 printk(KERN_DEBUG ".... register #00: %08X\n", reg_00.raw);
1280 printk(KERN_DEBUG "....... : physical APIC id: %02X\n", reg_00.bits.ID);
1281 printk(KERN_DEBUG "....... : Delivery Type: %X\n", reg_00.bits.delivery_type);
1282 printk(KERN_DEBUG "....... : LTS : %X\n", reg_00.bits.LTS);
1283
1284 printk(KERN_DEBUG ".... register #01: %08X\n", *(int *)&reg_01);
1285 printk(KERN_DEBUG "....... : max redirection entries: %02X\n",
1286 reg_01.bits.entries);
1287
1288 printk(KERN_DEBUG "....... : PRQ implemented: %X\n", reg_01.bits.PRQ);
1289 printk(KERN_DEBUG "....... : IO APIC version: %02X\n",
1290 reg_01.bits.version);
1291
1292 /*
1293 * Some Intel chipsets with IO APIC VERSION of 0x1? don't have reg_02,
1294 * but the value of reg_02 is read as the previous read register
1295 * value, so ignore it if reg_02 == reg_01.
1296 */
1297 if (reg_01.bits.version >= 0x10 && reg_02.raw != reg_01.raw) {
1298 printk(KERN_DEBUG ".... register #02: %08X\n", reg_02.raw);
1299 printk(KERN_DEBUG "....... : arbitration: %02X\n", reg_02.bits.arbitration);
1300 }
1301
1302 /*
1303 * Some Intel chipsets with IO APIC VERSION of 0x2? don't have reg_02
1304 * or reg_03, but the value of reg_0[23] is read as the previous read
1305 * register value, so ignore it if reg_03 == reg_0[12].
1306 */
1307 if (reg_01.bits.version >= 0x20 && reg_03.raw != reg_02.raw &&
1308 reg_03.raw != reg_01.raw) {
1309 printk(KERN_DEBUG ".... register #03: %08X\n", reg_03.raw);
1310 printk(KERN_DEBUG "....... : Boot DT : %X\n", reg_03.bits.boot_DT);
1311 }
1312
1313 printk(KERN_DEBUG ".... IRQ redirection table:\n");
1314 io_apic_print_entries(apic: ioapic_idx, nr_entries: reg_01.bits.entries);
1315}
1316
1317void __init print_IO_APICs(void)
1318{
1319 int ioapic_idx;
1320 unsigned int irq;
1321
1322 printk(KERN_DEBUG "number of MP IRQ sources: %d.\n", mp_irq_entries);
1323 for_each_ioapic(ioapic_idx)
1324 printk(KERN_DEBUG "number of IO-APIC #%d registers: %d.\n",
1325 mpc_ioapic_id(ioapic_idx),
1326 ioapics[ioapic_idx].nr_registers);
1327
1328 /*
1329 * We are a bit conservative about what we expect. We have to
1330 * know about every hardware change ASAP.
1331 */
1332 printk(KERN_INFO "testing the IO APIC.......................\n");
1333
1334 for_each_ioapic(ioapic_idx)
1335 print_IO_APIC(ioapic_idx);
1336
1337 printk(KERN_DEBUG "IRQ to pin mappings:\n");
1338 for_each_active_irq(irq) {
1339 struct irq_pin_list *entry;
1340 struct irq_chip *chip;
1341 struct mp_chip_data *data;
1342
1343 chip = irq_get_chip(irq);
1344 if (chip != &ioapic_chip && chip != &ioapic_ir_chip)
1345 continue;
1346 data = irq_get_chip_data(irq);
1347 if (!data)
1348 continue;
1349 if (list_empty(head: &data->irq_2_pin))
1350 continue;
1351
1352 printk(KERN_DEBUG "IRQ%d ", irq);
1353 for_each_irq_pin(entry, data->irq_2_pin)
1354 pr_cont("-> %d:%d", entry->apic, entry->pin);
1355 pr_cont("\n");
1356 }
1357
1358 printk(KERN_INFO ".................................... done.\n");
1359}
1360
1361/* Where if anywhere is the i8259 connect in external int mode */
1362static struct { int pin, apic; } ioapic_i8259 = { -1, -1 };
1363
1364void __init enable_IO_APIC(void)
1365{
1366 int i8259_apic, i8259_pin;
1367 int apic, pin;
1368
1369 if (ioapic_is_disabled)
1370 nr_ioapics = 0;
1371
1372 if (!nr_legacy_irqs() || !nr_ioapics)
1373 return;
1374
1375 for_each_ioapic_pin(apic, pin) {
1376 /* See if any of the pins is in ExtINT mode */
1377 struct IO_APIC_route_entry entry = ioapic_read_entry(apic, pin);
1378
1379 /* If the interrupt line is enabled and in ExtInt mode
1380 * I have found the pin where the i8259 is connected.
1381 */
1382 if (!entry.masked &&
1383 entry.delivery_mode == APIC_DELIVERY_MODE_EXTINT) {
1384 ioapic_i8259.apic = apic;
1385 ioapic_i8259.pin = pin;
1386 goto found_i8259;
1387 }
1388 }
1389 found_i8259:
1390 /* Look to see what if the MP table has reported the ExtINT */
1391 /* If we could not find the appropriate pin by looking at the ioapic
1392 * the i8259 probably is not connected the ioapic but give the
1393 * mptable a chance anyway.
1394 */
1395 i8259_pin = find_isa_irq_pin(irq: 0, type: mp_ExtINT);
1396 i8259_apic = find_isa_irq_apic(irq: 0, type: mp_ExtINT);
1397 /* Trust the MP table if nothing is setup in the hardware */
1398 if ((ioapic_i8259.pin == -1) && (i8259_pin >= 0)) {
1399 printk(KERN_WARNING "ExtINT not setup in hardware but reported by MP table\n");
1400 ioapic_i8259.pin = i8259_pin;
1401 ioapic_i8259.apic = i8259_apic;
1402 }
1403 /* Complain if the MP table and the hardware disagree */
1404 if (((ioapic_i8259.apic != i8259_apic) || (ioapic_i8259.pin != i8259_pin)) &&
1405 (i8259_pin >= 0) && (ioapic_i8259.pin >= 0))
1406 {
1407 printk(KERN_WARNING "ExtINT in hardware and MP table differ\n");
1408 }
1409
1410 /*
1411 * Do not trust the IO-APIC being empty at bootup
1412 */
1413 clear_IO_APIC();
1414}
1415
1416void native_restore_boot_irq_mode(void)
1417{
1418 /*
1419 * If the i8259 is routed through an IOAPIC
1420 * Put that IOAPIC in virtual wire mode
1421 * so legacy interrupts can be delivered.
1422 */
1423 if (ioapic_i8259.pin != -1) {
1424 struct IO_APIC_route_entry entry;
1425 u32 apic_id = read_apic_id();
1426
1427 memset(&entry, 0, sizeof(entry));
1428 entry.masked = false;
1429 entry.is_level = false;
1430 entry.active_low = false;
1431 entry.dest_mode_logical = false;
1432 entry.delivery_mode = APIC_DELIVERY_MODE_EXTINT;
1433 entry.destid_0_7 = apic_id & 0xFF;
1434 entry.virt_destid_8_14 = apic_id >> 8;
1435
1436 /*
1437 * Add it to the IO-APIC irq-routing table:
1438 */
1439 ioapic_write_entry(apic: ioapic_i8259.apic, pin: ioapic_i8259.pin, e: entry);
1440 }
1441
1442 if (boot_cpu_has(X86_FEATURE_APIC) || apic_from_smp_config())
1443 disconnect_bsp_APIC(virt_wire_setup: ioapic_i8259.pin != -1);
1444}
1445
1446void restore_boot_irq_mode(void)
1447{
1448 if (!nr_legacy_irqs())
1449 return;
1450
1451 x86_apic_ops.restore();
1452}
1453
1454#ifdef CONFIG_X86_32
1455/*
1456 * function to set the IO-APIC physical IDs based on the
1457 * values stored in the MPC table.
1458 *
1459 * by Matt Domsch <Matt_Domsch@dell.com> Tue Dec 21 12:25:05 CST 1999
1460 */
1461void __init setup_ioapic_ids_from_mpc_nocheck(void)
1462{
1463 union IO_APIC_reg_00 reg_00;
1464 physid_mask_t phys_id_present_map;
1465 int ioapic_idx;
1466 int i;
1467 unsigned char old_id;
1468 unsigned long flags;
1469
1470 /*
1471 * This is broken; anything with a real cpu count has to
1472 * circumvent this idiocy regardless.
1473 */
1474 apic->ioapic_phys_id_map(&phys_cpu_present_map, &phys_id_present_map);
1475
1476 /*
1477 * Set the IOAPIC ID to the value stored in the MPC table.
1478 */
1479 for_each_ioapic(ioapic_idx) {
1480 /* Read the register 0 value */
1481 raw_spin_lock_irqsave(&ioapic_lock, flags);
1482 reg_00.raw = io_apic_read(ioapic_idx, 0);
1483 raw_spin_unlock_irqrestore(&ioapic_lock, flags);
1484
1485 old_id = mpc_ioapic_id(ioapic_idx);
1486
1487 if (mpc_ioapic_id(ioapic_idx) >= get_physical_broadcast()) {
1488 printk(KERN_ERR "BIOS bug, IO-APIC#%d ID is %d in the MPC table!...\n",
1489 ioapic_idx, mpc_ioapic_id(ioapic_idx));
1490 printk(KERN_ERR "... fixing up to %d. (tell your hw vendor)\n",
1491 reg_00.bits.ID);
1492 ioapics[ioapic_idx].mp_config.apicid = reg_00.bits.ID;
1493 }
1494
1495 /*
1496 * Sanity check, is the ID really free? Every APIC in a
1497 * system must have a unique ID or we get lots of nice
1498 * 'stuck on smp_invalidate_needed IPI wait' messages.
1499 */
1500 if (apic->check_apicid_used(&phys_id_present_map,
1501 mpc_ioapic_id(ioapic_idx))) {
1502 printk(KERN_ERR "BIOS bug, IO-APIC#%d ID %d is already used!...\n",
1503 ioapic_idx, mpc_ioapic_id(ioapic_idx));
1504 for (i = 0; i < get_physical_broadcast(); i++)
1505 if (!physid_isset(i, phys_id_present_map))
1506 break;
1507 if (i >= get_physical_broadcast())
1508 panic("Max APIC ID exceeded!\n");
1509 printk(KERN_ERR "... fixing up to %d. (tell your hw vendor)\n",
1510 i);
1511 physid_set(i, phys_id_present_map);
1512 ioapics[ioapic_idx].mp_config.apicid = i;
1513 } else {
1514 apic_printk(APIC_VERBOSE, "Setting %d in the phys_id_present_map\n",
1515 mpc_ioapic_id(ioapic_idx));
1516 physid_set(mpc_ioapic_id(ioapic_idx), phys_id_present_map);
1517 }
1518
1519 /*
1520 * We need to adjust the IRQ routing table
1521 * if the ID changed.
1522 */
1523 if (old_id != mpc_ioapic_id(ioapic_idx))
1524 for (i = 0; i < mp_irq_entries; i++)
1525 if (mp_irqs[i].dstapic == old_id)
1526 mp_irqs[i].dstapic
1527 = mpc_ioapic_id(ioapic_idx);
1528
1529 /*
1530 * Update the ID register according to the right value
1531 * from the MPC table if they are different.
1532 */
1533 if (mpc_ioapic_id(ioapic_idx) == reg_00.bits.ID)
1534 continue;
1535
1536 apic_printk(APIC_VERBOSE, KERN_INFO
1537 "...changing IO-APIC physical APIC ID to %d ...",
1538 mpc_ioapic_id(ioapic_idx));
1539
1540 reg_00.bits.ID = mpc_ioapic_id(ioapic_idx);
1541 raw_spin_lock_irqsave(&ioapic_lock, flags);
1542 io_apic_write(ioapic_idx, 0, reg_00.raw);
1543 raw_spin_unlock_irqrestore(&ioapic_lock, flags);
1544
1545 /*
1546 * Sanity check
1547 */
1548 raw_spin_lock_irqsave(&ioapic_lock, flags);
1549 reg_00.raw = io_apic_read(ioapic_idx, 0);
1550 raw_spin_unlock_irqrestore(&ioapic_lock, flags);
1551 if (reg_00.bits.ID != mpc_ioapic_id(ioapic_idx))
1552 pr_cont("could not set ID!\n");
1553 else
1554 apic_printk(APIC_VERBOSE, " ok.\n");
1555 }
1556}
1557
1558void __init setup_ioapic_ids_from_mpc(void)
1559{
1560
1561 if (acpi_ioapic)
1562 return;
1563 /*
1564 * Don't check I/O APIC IDs for xAPIC systems. They have
1565 * no meaning without the serial APIC bus.
1566 */
1567 if (!(boot_cpu_data.x86_vendor == X86_VENDOR_INTEL)
1568 || APIC_XAPIC(boot_cpu_apic_version))
1569 return;
1570 setup_ioapic_ids_from_mpc_nocheck();
1571}
1572#endif
1573
1574int no_timer_check __initdata;
1575
1576static int __init notimercheck(char *s)
1577{
1578 no_timer_check = 1;
1579 return 1;
1580}
1581__setup("no_timer_check", notimercheck);
1582
1583static void __init delay_with_tsc(void)
1584{
1585 unsigned long long start, now;
1586 unsigned long end = jiffies + 4;
1587
1588 start = rdtsc();
1589
1590 /*
1591 * We don't know the TSC frequency yet, but waiting for
1592 * 40000000000/HZ TSC cycles is safe:
1593 * 4 GHz == 10 jiffies
1594 * 1 GHz == 40 jiffies
1595 */
1596 do {
1597 rep_nop();
1598 now = rdtsc();
1599 } while ((now - start) < 40000000000ULL / HZ &&
1600 time_before_eq(jiffies, end));
1601}
1602
1603static void __init delay_without_tsc(void)
1604{
1605 unsigned long end = jiffies + 4;
1606 int band = 1;
1607
1608 /*
1609 * We don't know any frequency yet, but waiting for
1610 * 40940000000/HZ cycles is safe:
1611 * 4 GHz == 10 jiffies
1612 * 1 GHz == 40 jiffies
1613 * 1 << 1 + 1 << 2 +...+ 1 << 11 = 4094
1614 */
1615 do {
1616 __delay(loops: ((1U << band++) * 10000000UL) / HZ);
1617 } while (band < 12 && time_before_eq(jiffies, end));
1618}
1619
1620/*
1621 * There is a nasty bug in some older SMP boards, their mptable lies
1622 * about the timer IRQ. We do the following to work around the situation:
1623 *
1624 * - timer IRQ defaults to IO-APIC IRQ
1625 * - if this function detects that timer IRQs are defunct, then we fall
1626 * back to ISA timer IRQs
1627 */
1628static int __init timer_irq_works(void)
1629{
1630 unsigned long t1 = jiffies;
1631
1632 if (no_timer_check)
1633 return 1;
1634
1635 local_irq_enable();
1636 if (boot_cpu_has(X86_FEATURE_TSC))
1637 delay_with_tsc();
1638 else
1639 delay_without_tsc();
1640
1641 /*
1642 * Expect a few ticks at least, to be sure some possible
1643 * glue logic does not lock up after one or two first
1644 * ticks in a non-ExtINT mode. Also the local APIC
1645 * might have cached one ExtINT interrupt. Finally, at
1646 * least one tick may be lost due to delays.
1647 */
1648
1649 local_irq_disable();
1650
1651 /* Did jiffies advance? */
1652 return time_after(jiffies, t1 + 4);
1653}
1654
1655/*
1656 * In the SMP+IOAPIC case it might happen that there are an unspecified
1657 * number of pending IRQ events unhandled. These cases are very rare,
1658 * so we 'resend' these IRQs via IPIs, to the same CPU. It's much
1659 * better to do it this way as thus we do not have to be aware of
1660 * 'pending' interrupts in the IRQ path, except at this point.
1661 */
1662/*
1663 * Edge triggered needs to resend any interrupt
1664 * that was delayed but this is now handled in the device
1665 * independent code.
1666 */
1667
1668/*
1669 * Starting up a edge-triggered IO-APIC interrupt is
1670 * nasty - we need to make sure that we get the edge.
1671 * If it is already asserted for some reason, we need
1672 * return 1 to indicate that is was pending.
1673 *
1674 * This is not complete - we should be able to fake
1675 * an edge even if it isn't on the 8259A...
1676 */
1677static unsigned int startup_ioapic_irq(struct irq_data *data)
1678{
1679 int was_pending = 0, irq = data->irq;
1680 unsigned long flags;
1681
1682 raw_spin_lock_irqsave(&ioapic_lock, flags);
1683 if (irq < nr_legacy_irqs()) {
1684 legacy_pic->mask(irq);
1685 if (legacy_pic->irq_pending(irq))
1686 was_pending = 1;
1687 }
1688 __unmask_ioapic(data: data->chip_data);
1689 raw_spin_unlock_irqrestore(&ioapic_lock, flags);
1690
1691 return was_pending;
1692}
1693
1694atomic_t irq_mis_count;
1695
1696#ifdef CONFIG_GENERIC_PENDING_IRQ
1697static bool io_apic_level_ack_pending(struct mp_chip_data *data)
1698{
1699 struct irq_pin_list *entry;
1700 unsigned long flags;
1701
1702 raw_spin_lock_irqsave(&ioapic_lock, flags);
1703 for_each_irq_pin(entry, data->irq_2_pin) {
1704 struct IO_APIC_route_entry e;
1705 int pin;
1706
1707 pin = entry->pin;
1708 e.w1 = io_apic_read(apic: entry->apic, reg: 0x10 + pin*2);
1709 /* Is the remote IRR bit set? */
1710 if (e.irr) {
1711 raw_spin_unlock_irqrestore(&ioapic_lock, flags);
1712 return true;
1713 }
1714 }
1715 raw_spin_unlock_irqrestore(&ioapic_lock, flags);
1716
1717 return false;
1718}
1719
1720static inline bool ioapic_prepare_move(struct irq_data *data)
1721{
1722 /* If we are moving the IRQ we need to mask it */
1723 if (unlikely(irqd_is_setaffinity_pending(data))) {
1724 if (!irqd_irq_masked(d: data))
1725 mask_ioapic_irq(irq_data: data);
1726 return true;
1727 }
1728 return false;
1729}
1730
1731static inline void ioapic_finish_move(struct irq_data *data, bool moveit)
1732{
1733 if (unlikely(moveit)) {
1734 /* Only migrate the irq if the ack has been received.
1735 *
1736 * On rare occasions the broadcast level triggered ack gets
1737 * delayed going to ioapics, and if we reprogram the
1738 * vector while Remote IRR is still set the irq will never
1739 * fire again.
1740 *
1741 * To prevent this scenario we read the Remote IRR bit
1742 * of the ioapic. This has two effects.
1743 * - On any sane system the read of the ioapic will
1744 * flush writes (and acks) going to the ioapic from
1745 * this cpu.
1746 * - We get to see if the ACK has actually been delivered.
1747 *
1748 * Based on failed experiments of reprogramming the
1749 * ioapic entry from outside of irq context starting
1750 * with masking the ioapic entry and then polling until
1751 * Remote IRR was clear before reprogramming the
1752 * ioapic I don't trust the Remote IRR bit to be
1753 * completely accurate.
1754 *
1755 * However there appears to be no other way to plug
1756 * this race, so if the Remote IRR bit is not
1757 * accurate and is causing problems then it is a hardware bug
1758 * and you can go talk to the chipset vendor about it.
1759 */
1760 if (!io_apic_level_ack_pending(data: data->chip_data))
1761 irq_move_masked_irq(data);
1762 /* If the IRQ is masked in the core, leave it: */
1763 if (!irqd_irq_masked(d: data))
1764 unmask_ioapic_irq(irq_data: data);
1765 }
1766}
1767#else
1768static inline bool ioapic_prepare_move(struct irq_data *data)
1769{
1770 return false;
1771}
1772static inline void ioapic_finish_move(struct irq_data *data, bool moveit)
1773{
1774}
1775#endif
1776
1777static void ioapic_ack_level(struct irq_data *irq_data)
1778{
1779 struct irq_cfg *cfg = irqd_cfg(irq_data);
1780 unsigned long v;
1781 bool moveit;
1782 int i;
1783
1784 irq_complete_move(cfg);
1785 moveit = ioapic_prepare_move(data: irq_data);
1786
1787 /*
1788 * It appears there is an erratum which affects at least version 0x11
1789 * of I/O APIC (that's the 82093AA and cores integrated into various
1790 * chipsets). Under certain conditions a level-triggered interrupt is
1791 * erroneously delivered as edge-triggered one but the respective IRR
1792 * bit gets set nevertheless. As a result the I/O unit expects an EOI
1793 * message but it will never arrive and further interrupts are blocked
1794 * from the source. The exact reason is so far unknown, but the
1795 * phenomenon was observed when two consecutive interrupt requests
1796 * from a given source get delivered to the same CPU and the source is
1797 * temporarily disabled in between.
1798 *
1799 * A workaround is to simulate an EOI message manually. We achieve it
1800 * by setting the trigger mode to edge and then to level when the edge
1801 * trigger mode gets detected in the TMR of a local APIC for a
1802 * level-triggered interrupt. We mask the source for the time of the
1803 * operation to prevent an edge-triggered interrupt escaping meanwhile.
1804 * The idea is from Manfred Spraul. --macro
1805 *
1806 * Also in the case when cpu goes offline, fixup_irqs() will forward
1807 * any unhandled interrupt on the offlined cpu to the new cpu
1808 * destination that is handling the corresponding interrupt. This
1809 * interrupt forwarding is done via IPI's. Hence, in this case also
1810 * level-triggered io-apic interrupt will be seen as an edge
1811 * interrupt in the IRR. And we can't rely on the cpu's EOI
1812 * to be broadcasted to the IO-APIC's which will clear the remoteIRR
1813 * corresponding to the level-triggered interrupt. Hence on IO-APIC's
1814 * supporting EOI register, we do an explicit EOI to clear the
1815 * remote IRR and on IO-APIC's which don't have an EOI register,
1816 * we use the above logic (mask+edge followed by unmask+level) from
1817 * Manfred Spraul to clear the remote IRR.
1818 */
1819 i = cfg->vector;
1820 v = apic_read(APIC_TMR + ((i & ~0x1f) >> 1));
1821
1822 /*
1823 * We must acknowledge the irq before we move it or the acknowledge will
1824 * not propagate properly.
1825 */
1826 apic_eoi();
1827
1828 /*
1829 * Tail end of clearing remote IRR bit (either by delivering the EOI
1830 * message via io-apic EOI register write or simulating it using
1831 * mask+edge followed by unmask+level logic) manually when the
1832 * level triggered interrupt is seen as the edge triggered interrupt
1833 * at the cpu.
1834 */
1835 if (!(v & (1 << (i & 0x1f)))) {
1836 atomic_inc(v: &irq_mis_count);
1837 eoi_ioapic_pin(vector: cfg->vector, data: irq_data->chip_data);
1838 }
1839
1840 ioapic_finish_move(data: irq_data, moveit);
1841}
1842
1843static void ioapic_ir_ack_level(struct irq_data *irq_data)
1844{
1845 struct mp_chip_data *data = irq_data->chip_data;
1846
1847 /*
1848 * Intr-remapping uses pin number as the virtual vector
1849 * in the RTE. Actual vector is programmed in
1850 * intr-remapping table entry. Hence for the io-apic
1851 * EOI we use the pin number.
1852 */
1853 apic_ack_irq(data: irq_data);
1854 eoi_ioapic_pin(vector: data->entry.vector, data);
1855}
1856
1857/*
1858 * The I/OAPIC is just a device for generating MSI messages from legacy
1859 * interrupt pins. Various fields of the RTE translate into bits of the
1860 * resulting MSI which had a historical meaning.
1861 *
1862 * With interrupt remapping, many of those bits have different meanings
1863 * in the underlying MSI, but the way that the I/OAPIC transforms them
1864 * from its RTE to the MSI message is the same. This function allows
1865 * the parent IRQ domain to compose the MSI message, then takes the
1866 * relevant bits to put them in the appropriate places in the RTE in
1867 * order to generate that message when the IRQ happens.
1868 *
1869 * The setup here relies on a preconfigured route entry (is_level,
1870 * active_low, masked) because the parent domain is merely composing the
1871 * generic message routing information which is used for the MSI.
1872 */
1873static void ioapic_setup_msg_from_msi(struct irq_data *irq_data,
1874 struct IO_APIC_route_entry *entry)
1875{
1876 struct msi_msg msg;
1877
1878 /* Let the parent domain compose the MSI message */
1879 irq_chip_compose_msi_msg(data: irq_data, msg: &msg);
1880
1881 /*
1882 * - Real vector
1883 * - DMAR/IR: 8bit subhandle (ioapic.pin)
1884 * - AMD/IR: 8bit IRTE index
1885 */
1886 entry->vector = msg.arch_data.vector;
1887 /* Delivery mode (for DMAR/IR all 0) */
1888 entry->delivery_mode = msg.arch_data.delivery_mode;
1889 /* Destination mode or DMAR/IR index bit 15 */
1890 entry->dest_mode_logical = msg.arch_addr_lo.dest_mode_logical;
1891 /* DMAR/IR: 1, 0 for all other modes */
1892 entry->ir_format = msg.arch_addr_lo.dmar_format;
1893 /*
1894 * - DMAR/IR: index bit 0-14.
1895 *
1896 * - Virt: If the host supports x2apic without a virtualized IR
1897 * unit then bit 0-6 of dmar_index_0_14 are providing bit
1898 * 8-14 of the destination id.
1899 *
1900 * All other modes have bit 0-6 of dmar_index_0_14 cleared and the
1901 * topmost 8 bits are destination id bit 0-7 (entry::destid_0_7).
1902 */
1903 entry->ir_index_0_14 = msg.arch_addr_lo.dmar_index_0_14;
1904}
1905
1906static void ioapic_configure_entry(struct irq_data *irqd)
1907{
1908 struct mp_chip_data *mpd = irqd->chip_data;
1909 struct irq_pin_list *entry;
1910
1911 ioapic_setup_msg_from_msi(irq_data: irqd, entry: &mpd->entry);
1912
1913 for_each_irq_pin(entry, mpd->irq_2_pin)
1914 __ioapic_write_entry(apic: entry->apic, pin: entry->pin, e: mpd->entry);
1915}
1916
1917static int ioapic_set_affinity(struct irq_data *irq_data,
1918 const struct cpumask *mask, bool force)
1919{
1920 struct irq_data *parent = irq_data->parent_data;
1921 unsigned long flags;
1922 int ret;
1923
1924 ret = parent->chip->irq_set_affinity(parent, mask, force);
1925 raw_spin_lock_irqsave(&ioapic_lock, flags);
1926 if (ret >= 0 && ret != IRQ_SET_MASK_OK_DONE)
1927 ioapic_configure_entry(irqd: irq_data);
1928 raw_spin_unlock_irqrestore(&ioapic_lock, flags);
1929
1930 return ret;
1931}
1932
1933/*
1934 * Interrupt shutdown masks the ioapic pin, but the interrupt might already
1935 * be in flight, but not yet serviced by the target CPU. That means
1936 * __synchronize_hardirq() would return and claim that everything is calmed
1937 * down. So free_irq() would proceed and deactivate the interrupt and free
1938 * resources.
1939 *
1940 * Once the target CPU comes around to service it it will find a cleared
1941 * vector and complain. While the spurious interrupt is harmless, the full
1942 * release of resources might prevent the interrupt from being acknowledged
1943 * which keeps the hardware in a weird state.
1944 *
1945 * Verify that the corresponding Remote-IRR bits are clear.
1946 */
1947static int ioapic_irq_get_chip_state(struct irq_data *irqd,
1948 enum irqchip_irq_state which,
1949 bool *state)
1950{
1951 struct mp_chip_data *mcd = irqd->chip_data;
1952 struct IO_APIC_route_entry rentry;
1953 struct irq_pin_list *p;
1954
1955 if (which != IRQCHIP_STATE_ACTIVE)
1956 return -EINVAL;
1957
1958 *state = false;
1959 raw_spin_lock(&ioapic_lock);
1960 for_each_irq_pin(p, mcd->irq_2_pin) {
1961 rentry = __ioapic_read_entry(apic: p->apic, pin: p->pin);
1962 /*
1963 * The remote IRR is only valid in level trigger mode. It's
1964 * meaning is undefined for edge triggered interrupts and
1965 * irrelevant because the IO-APIC treats them as fire and
1966 * forget.
1967 */
1968 if (rentry.irr && rentry.is_level) {
1969 *state = true;
1970 break;
1971 }
1972 }
1973 raw_spin_unlock(&ioapic_lock);
1974 return 0;
1975}
1976
1977static struct irq_chip ioapic_chip __read_mostly = {
1978 .name = "IO-APIC",
1979 .irq_startup = startup_ioapic_irq,
1980 .irq_mask = mask_ioapic_irq,
1981 .irq_unmask = unmask_ioapic_irq,
1982 .irq_ack = irq_chip_ack_parent,
1983 .irq_eoi = ioapic_ack_level,
1984 .irq_set_affinity = ioapic_set_affinity,
1985 .irq_retrigger = irq_chip_retrigger_hierarchy,
1986 .irq_get_irqchip_state = ioapic_irq_get_chip_state,
1987 .flags = IRQCHIP_SKIP_SET_WAKE |
1988 IRQCHIP_AFFINITY_PRE_STARTUP,
1989};
1990
1991static struct irq_chip ioapic_ir_chip __read_mostly = {
1992 .name = "IR-IO-APIC",
1993 .irq_startup = startup_ioapic_irq,
1994 .irq_mask = mask_ioapic_irq,
1995 .irq_unmask = unmask_ioapic_irq,
1996 .irq_ack = irq_chip_ack_parent,
1997 .irq_eoi = ioapic_ir_ack_level,
1998 .irq_set_affinity = ioapic_set_affinity,
1999 .irq_retrigger = irq_chip_retrigger_hierarchy,
2000 .irq_get_irqchip_state = ioapic_irq_get_chip_state,
2001 .flags = IRQCHIP_SKIP_SET_WAKE |
2002 IRQCHIP_AFFINITY_PRE_STARTUP,
2003};
2004
2005static inline void init_IO_APIC_traps(void)
2006{
2007 struct irq_cfg *cfg;
2008 unsigned int irq;
2009
2010 for_each_active_irq(irq) {
2011 cfg = irq_cfg(irq);
2012 if (IO_APIC_IRQ(irq) && cfg && !cfg->vector) {
2013 /*
2014 * Hmm.. We don't have an entry for this,
2015 * so default to an old-fashioned 8259
2016 * interrupt if we can..
2017 */
2018 if (irq < nr_legacy_irqs())
2019 legacy_pic->make_irq(irq);
2020 else
2021 /* Strange. Oh, well.. */
2022 irq_set_chip(irq, chip: &no_irq_chip);
2023 }
2024 }
2025}
2026
2027/*
2028 * The local APIC irq-chip implementation:
2029 */
2030
2031static void mask_lapic_irq(struct irq_data *data)
2032{
2033 unsigned long v;
2034
2035 v = apic_read(APIC_LVT0);
2036 apic_write(APIC_LVT0, val: v | APIC_LVT_MASKED);
2037}
2038
2039static void unmask_lapic_irq(struct irq_data *data)
2040{
2041 unsigned long v;
2042
2043 v = apic_read(APIC_LVT0);
2044 apic_write(APIC_LVT0, val: v & ~APIC_LVT_MASKED);
2045}
2046
2047static void ack_lapic_irq(struct irq_data *data)
2048{
2049 apic_eoi();
2050}
2051
2052static struct irq_chip lapic_chip __read_mostly = {
2053 .name = "local-APIC",
2054 .irq_mask = mask_lapic_irq,
2055 .irq_unmask = unmask_lapic_irq,
2056 .irq_ack = ack_lapic_irq,
2057};
2058
2059static void lapic_register_intr(int irq)
2060{
2061 irq_clear_status_flags(irq, clr: IRQ_LEVEL);
2062 irq_set_chip_and_handler_name(irq, chip: &lapic_chip, handle: handle_edge_irq,
2063 name: "edge");
2064}
2065
2066/*
2067 * This looks a bit hackish but it's about the only one way of sending
2068 * a few INTA cycles to 8259As and any associated glue logic. ICR does
2069 * not support the ExtINT mode, unfortunately. We need to send these
2070 * cycles as some i82489DX-based boards have glue logic that keeps the
2071 * 8259A interrupt line asserted until INTA. --macro
2072 */
2073static inline void __init unlock_ExtINT_logic(void)
2074{
2075 int apic, pin, i;
2076 struct IO_APIC_route_entry entry0, entry1;
2077 unsigned char save_control, save_freq_select;
2078 u32 apic_id;
2079
2080 pin = find_isa_irq_pin(irq: 8, type: mp_INT);
2081 if (pin == -1) {
2082 WARN_ON_ONCE(1);
2083 return;
2084 }
2085 apic = find_isa_irq_apic(irq: 8, type: mp_INT);
2086 if (apic == -1) {
2087 WARN_ON_ONCE(1);
2088 return;
2089 }
2090
2091 entry0 = ioapic_read_entry(apic, pin);
2092 clear_IO_APIC_pin(apic, pin);
2093
2094 apic_id = read_apic_id();
2095 memset(&entry1, 0, sizeof(entry1));
2096
2097 entry1.dest_mode_logical = true;
2098 entry1.masked = false;
2099 entry1.destid_0_7 = apic_id & 0xFF;
2100 entry1.virt_destid_8_14 = apic_id >> 8;
2101 entry1.delivery_mode = APIC_DELIVERY_MODE_EXTINT;
2102 entry1.active_low = entry0.active_low;
2103 entry1.is_level = false;
2104 entry1.vector = 0;
2105
2106 ioapic_write_entry(apic, pin, e: entry1);
2107
2108 save_control = CMOS_READ(RTC_CONTROL);
2109 save_freq_select = CMOS_READ(RTC_FREQ_SELECT);
2110 CMOS_WRITE((save_freq_select & ~RTC_RATE_SELECT) | 0x6,
2111 RTC_FREQ_SELECT);
2112 CMOS_WRITE(save_control | RTC_PIE, RTC_CONTROL);
2113
2114 i = 100;
2115 while (i-- > 0) {
2116 mdelay(10);
2117 if ((CMOS_READ(RTC_INTR_FLAGS) & RTC_PF) == RTC_PF)
2118 i -= 10;
2119 }
2120
2121 CMOS_WRITE(save_control, RTC_CONTROL);
2122 CMOS_WRITE(save_freq_select, RTC_FREQ_SELECT);
2123 clear_IO_APIC_pin(apic, pin);
2124
2125 ioapic_write_entry(apic, pin, e: entry0);
2126}
2127
2128static int disable_timer_pin_1 __initdata;
2129/* Actually the next is obsolete, but keep it for paranoid reasons -AK */
2130static int __init disable_timer_pin_setup(char *arg)
2131{
2132 disable_timer_pin_1 = 1;
2133 return 0;
2134}
2135early_param("disable_timer_pin_1", disable_timer_pin_setup);
2136
2137static int mp_alloc_timer_irq(int ioapic, int pin)
2138{
2139 int irq = -1;
2140 struct irq_domain *domain = mp_ioapic_irqdomain(ioapic);
2141
2142 if (domain) {
2143 struct irq_alloc_info info;
2144
2145 ioapic_set_alloc_attr(info: &info, NUMA_NO_NODE, trigger: 0, polarity: 0);
2146 info.devid = mpc_ioapic_id(ioapic_idx: ioapic);
2147 info.ioapic.pin = pin;
2148 mutex_lock(&ioapic_mutex);
2149 irq = alloc_isa_irq_from_domain(domain, irq: 0, ioapic, pin, info: &info);
2150 mutex_unlock(lock: &ioapic_mutex);
2151 }
2152
2153 return irq;
2154}
2155
2156/*
2157 * This code may look a bit paranoid, but it's supposed to cooperate with
2158 * a wide range of boards and BIOS bugs. Fortunately only the timer IRQ
2159 * is so screwy. Thanks to Brian Perkins for testing/hacking this beast
2160 * fanatically on his truly buggy board.
2161 *
2162 * FIXME: really need to revamp this for all platforms.
2163 */
2164static inline void __init check_timer(void)
2165{
2166 struct irq_data *irq_data = irq_get_irq_data(irq: 0);
2167 struct mp_chip_data *data = irq_data->chip_data;
2168 struct irq_cfg *cfg = irqd_cfg(irq_data);
2169 int node = cpu_to_node(cpu: 0);
2170 int apic1, pin1, apic2, pin2;
2171 int no_pin1 = 0;
2172
2173 if (!global_clock_event)
2174 return;
2175
2176 local_irq_disable();
2177
2178 /*
2179 * get/set the timer IRQ vector:
2180 */
2181 legacy_pic->mask(0);
2182
2183 /*
2184 * As IRQ0 is to be enabled in the 8259A, the virtual
2185 * wire has to be disabled in the local APIC. Also
2186 * timer interrupts need to be acknowledged manually in
2187 * the 8259A for the i82489DX when using the NMI
2188 * watchdog as that APIC treats NMIs as level-triggered.
2189 * The AEOI mode will finish them in the 8259A
2190 * automatically.
2191 */
2192 apic_write(APIC_LVT0, APIC_LVT_MASKED | APIC_DM_EXTINT);
2193 legacy_pic->init(1);
2194
2195 pin1 = find_isa_irq_pin(irq: 0, type: mp_INT);
2196 apic1 = find_isa_irq_apic(irq: 0, type: mp_INT);
2197 pin2 = ioapic_i8259.pin;
2198 apic2 = ioapic_i8259.apic;
2199
2200 apic_printk(APIC_QUIET, KERN_INFO "..TIMER: vector=0x%02X "
2201 "apic1=%d pin1=%d apic2=%d pin2=%d\n",
2202 cfg->vector, apic1, pin1, apic2, pin2);
2203
2204 /*
2205 * Some BIOS writers are clueless and report the ExtINTA
2206 * I/O APIC input from the cascaded 8259A as the timer
2207 * interrupt input. So just in case, if only one pin
2208 * was found above, try it both directly and through the
2209 * 8259A.
2210 */
2211 if (pin1 == -1) {
2212 panic_if_irq_remap(msg: "BIOS bug: timer not connected to IO-APIC");
2213 pin1 = pin2;
2214 apic1 = apic2;
2215 no_pin1 = 1;
2216 } else if (pin2 == -1) {
2217 pin2 = pin1;
2218 apic2 = apic1;
2219 }
2220
2221 if (pin1 != -1) {
2222 /* Ok, does IRQ0 through the IOAPIC work? */
2223 if (no_pin1) {
2224 mp_alloc_timer_irq(ioapic: apic1, pin: pin1);
2225 } else {
2226 /*
2227 * for edge trigger, it's already unmasked,
2228 * so only need to unmask if it is level-trigger
2229 * do we really have level trigger timer?
2230 */
2231 int idx = find_irq_entry(ioapic_idx: apic1, pin: pin1, type: mp_INT);
2232
2233 if (idx != -1 && irq_is_level(idx))
2234 unmask_ioapic_irq(irq_data: irq_get_irq_data(irq: 0));
2235 }
2236 irq_domain_deactivate_irq(irq_data);
2237 irq_domain_activate_irq(irq_data, early: false);
2238 if (timer_irq_works()) {
2239 if (disable_timer_pin_1 > 0)
2240 clear_IO_APIC_pin(apic: 0, pin: pin1);
2241 goto out;
2242 }
2243 panic_if_irq_remap(msg: "timer doesn't work through Interrupt-remapped IO-APIC");
2244 clear_IO_APIC_pin(apic: apic1, pin: pin1);
2245 if (!no_pin1)
2246 apic_printk(APIC_QUIET, KERN_ERR "..MP-BIOS bug: "
2247 "8254 timer not connected to IO-APIC\n");
2248
2249 apic_printk(APIC_QUIET, KERN_INFO "...trying to set up timer "
2250 "(IRQ0) through the 8259A ...\n");
2251 apic_printk(APIC_QUIET, KERN_INFO
2252 "..... (found apic %d pin %d) ...\n", apic2, pin2);
2253 /*
2254 * legacy devices should be connected to IO APIC #0
2255 */
2256 replace_pin_at_irq_node(data, node, oldapic: apic1, oldpin: pin1, newapic: apic2, newpin: pin2);
2257 irq_domain_deactivate_irq(irq_data);
2258 irq_domain_activate_irq(irq_data, early: false);
2259 legacy_pic->unmask(0);
2260 if (timer_irq_works()) {
2261 apic_printk(APIC_QUIET, KERN_INFO "....... works.\n");
2262 goto out;
2263 }
2264 /*
2265 * Cleanup, just in case ...
2266 */
2267 legacy_pic->mask(0);
2268 clear_IO_APIC_pin(apic: apic2, pin: pin2);
2269 apic_printk(APIC_QUIET, KERN_INFO "....... failed.\n");
2270 }
2271
2272 apic_printk(APIC_QUIET, KERN_INFO
2273 "...trying to set up timer as Virtual Wire IRQ...\n");
2274
2275 lapic_register_intr(irq: 0);
2276 apic_write(APIC_LVT0, APIC_DM_FIXED | cfg->vector); /* Fixed mode */
2277 legacy_pic->unmask(0);
2278
2279 if (timer_irq_works()) {
2280 apic_printk(APIC_QUIET, KERN_INFO "..... works.\n");
2281 goto out;
2282 }
2283 legacy_pic->mask(0);
2284 apic_write(APIC_LVT0, APIC_LVT_MASKED | APIC_DM_FIXED | cfg->vector);
2285 apic_printk(APIC_QUIET, KERN_INFO "..... failed.\n");
2286
2287 apic_printk(APIC_QUIET, KERN_INFO
2288 "...trying to set up timer as ExtINT IRQ...\n");
2289
2290 legacy_pic->init(0);
2291 legacy_pic->make_irq(0);
2292 apic_write(APIC_LVT0, APIC_DM_EXTINT);
2293 legacy_pic->unmask(0);
2294
2295 unlock_ExtINT_logic();
2296
2297 if (timer_irq_works()) {
2298 apic_printk(APIC_QUIET, KERN_INFO "..... works.\n");
2299 goto out;
2300 }
2301 apic_printk(APIC_QUIET, KERN_INFO "..... failed :(.\n");
2302 if (apic_is_x2apic_enabled())
2303 apic_printk(APIC_QUIET, KERN_INFO
2304 "Perhaps problem with the pre-enabled x2apic mode\n"
2305 "Try booting with x2apic and interrupt-remapping disabled in the bios.\n");
2306 panic(fmt: "IO-APIC + timer doesn't work! Boot with apic=debug and send a "
2307 "report. Then try booting with the 'noapic' option.\n");
2308out:
2309 local_irq_enable();
2310}
2311
2312/*
2313 * Traditionally ISA IRQ2 is the cascade IRQ, and is not available
2314 * to devices. However there may be an I/O APIC pin available for
2315 * this interrupt regardless. The pin may be left unconnected, but
2316 * typically it will be reused as an ExtINT cascade interrupt for
2317 * the master 8259A. In the MPS case such a pin will normally be
2318 * reported as an ExtINT interrupt in the MP table. With ACPI
2319 * there is no provision for ExtINT interrupts, and in the absence
2320 * of an override it would be treated as an ordinary ISA I/O APIC
2321 * interrupt, that is edge-triggered and unmasked by default. We
2322 * used to do this, but it caused problems on some systems because
2323 * of the NMI watchdog and sometimes IRQ0 of the 8254 timer using
2324 * the same ExtINT cascade interrupt to drive the local APIC of the
2325 * bootstrap processor. Therefore we refrain from routing IRQ2 to
2326 * the I/O APIC in all cases now. No actual device should request
2327 * it anyway. --macro
2328 */
2329#define PIC_IRQS (1UL << PIC_CASCADE_IR)
2330
2331static int mp_irqdomain_create(int ioapic)
2332{
2333 struct irq_domain *parent;
2334 int hwirqs = mp_ioapic_pin_count(ioapic);
2335 struct ioapic *ip = &ioapics[ioapic];
2336 struct ioapic_domain_cfg *cfg = &ip->irqdomain_cfg;
2337 struct mp_ioapic_gsi *gsi_cfg = mp_ioapic_gsi_routing(ioapic_idx: ioapic);
2338 struct fwnode_handle *fn;
2339 struct irq_fwspec fwspec;
2340
2341 if (cfg->type == IOAPIC_DOMAIN_INVALID)
2342 return 0;
2343
2344 /* Handle device tree enumerated APICs proper */
2345 if (cfg->dev) {
2346 fn = of_node_to_fwnode(node: cfg->dev);
2347 } else {
2348 fn = irq_domain_alloc_named_id_fwnode(name: "IO-APIC", id: mpc_ioapic_id(ioapic_idx: ioapic));
2349 if (!fn)
2350 return -ENOMEM;
2351 }
2352
2353 fwspec.fwnode = fn;
2354 fwspec.param_count = 1;
2355 fwspec.param[0] = mpc_ioapic_id(ioapic_idx: ioapic);
2356
2357 parent = irq_find_matching_fwspec(fwspec: &fwspec, bus_token: DOMAIN_BUS_ANY);
2358 if (!parent) {
2359 if (!cfg->dev)
2360 irq_domain_free_fwnode(fwnode: fn);
2361 return -ENODEV;
2362 }
2363
2364 ip->irqdomain = irq_domain_create_hierarchy(parent, flags: 0, size: hwirqs, fwnode: fn, ops: cfg->ops,
2365 host_data: (void *)(long)ioapic);
2366 if (!ip->irqdomain) {
2367 /* Release fw handle if it was allocated above */
2368 if (!cfg->dev)
2369 irq_domain_free_fwnode(fwnode: fn);
2370 return -ENOMEM;
2371 }
2372
2373 if (cfg->type == IOAPIC_DOMAIN_LEGACY ||
2374 cfg->type == IOAPIC_DOMAIN_STRICT)
2375 ioapic_dynirq_base = max(ioapic_dynirq_base,
2376 gsi_cfg->gsi_end + 1);
2377
2378 return 0;
2379}
2380
2381static void ioapic_destroy_irqdomain(int idx)
2382{
2383 struct ioapic_domain_cfg *cfg = &ioapics[idx].irqdomain_cfg;
2384 struct fwnode_handle *fn = ioapics[idx].irqdomain->fwnode;
2385
2386 if (ioapics[idx].irqdomain) {
2387 irq_domain_remove(host: ioapics[idx].irqdomain);
2388 if (!cfg->dev)
2389 irq_domain_free_fwnode(fwnode: fn);
2390 ioapics[idx].irqdomain = NULL;
2391 }
2392}
2393
2394void __init setup_IO_APIC(void)
2395{
2396 int ioapic;
2397
2398 if (ioapic_is_disabled || !nr_ioapics)
2399 return;
2400
2401 io_apic_irqs = nr_legacy_irqs() ? ~PIC_IRQS : ~0UL;
2402
2403 apic_printk(APIC_VERBOSE, "ENABLING IO-APIC IRQs\n");
2404 for_each_ioapic(ioapic)
2405 BUG_ON(mp_irqdomain_create(ioapic));
2406
2407 /*
2408 * Set up IO-APIC IRQ routing.
2409 */
2410 x86_init.mpparse.setup_ioapic_ids();
2411
2412 sync_Arb_IDs();
2413 setup_IO_APIC_irqs();
2414 init_IO_APIC_traps();
2415 if (nr_legacy_irqs())
2416 check_timer();
2417
2418 ioapic_initialized = 1;
2419}
2420
2421static void resume_ioapic_id(int ioapic_idx)
2422{
2423 unsigned long flags;
2424 union IO_APIC_reg_00 reg_00;
2425
2426 raw_spin_lock_irqsave(&ioapic_lock, flags);
2427 reg_00.raw = io_apic_read(apic: ioapic_idx, reg: 0);
2428 if (reg_00.bits.ID != mpc_ioapic_id(ioapic_idx)) {
2429 reg_00.bits.ID = mpc_ioapic_id(ioapic_idx);
2430 io_apic_write(apic: ioapic_idx, reg: 0, value: reg_00.raw);
2431 }
2432 raw_spin_unlock_irqrestore(&ioapic_lock, flags);
2433}
2434
2435static void ioapic_resume(void)
2436{
2437 int ioapic_idx;
2438
2439 for_each_ioapic_reverse(ioapic_idx)
2440 resume_ioapic_id(ioapic_idx);
2441
2442 restore_ioapic_entries();
2443}
2444
2445static struct syscore_ops ioapic_syscore_ops = {
2446 .suspend = save_ioapic_entries,
2447 .resume = ioapic_resume,
2448};
2449
2450static int __init ioapic_init_ops(void)
2451{
2452 register_syscore_ops(ops: &ioapic_syscore_ops);
2453
2454 return 0;
2455}
2456
2457device_initcall(ioapic_init_ops);
2458
2459static int io_apic_get_redir_entries(int ioapic)
2460{
2461 union IO_APIC_reg_01 reg_01;
2462 unsigned long flags;
2463
2464 raw_spin_lock_irqsave(&ioapic_lock, flags);
2465 reg_01.raw = io_apic_read(apic: ioapic, reg: 1);
2466 raw_spin_unlock_irqrestore(&ioapic_lock, flags);
2467
2468 /* The register returns the maximum index redir index
2469 * supported, which is one less than the total number of redir
2470 * entries.
2471 */
2472 return reg_01.bits.entries + 1;
2473}
2474
2475unsigned int arch_dynirq_lower_bound(unsigned int from)
2476{
2477 unsigned int ret;
2478
2479 /*
2480 * dmar_alloc_hwirq() may be called before setup_IO_APIC(), so use
2481 * gsi_top if ioapic_dynirq_base hasn't been initialized yet.
2482 */
2483 ret = ioapic_dynirq_base ? : gsi_top;
2484
2485 /*
2486 * For DT enabled machines ioapic_dynirq_base is irrelevant and
2487 * always 0. gsi_top can be 0 if there is no IO/APIC registered.
2488 * 0 is an invalid interrupt number for dynamic allocations. Return
2489 * @from instead.
2490 */
2491 return ret ? : from;
2492}
2493
2494#ifdef CONFIG_X86_32
2495static int io_apic_get_unique_id(int ioapic, int apic_id)
2496{
2497 union IO_APIC_reg_00 reg_00;
2498 static physid_mask_t apic_id_map = PHYSID_MASK_NONE;
2499 physid_mask_t tmp;
2500 unsigned long flags;
2501 int i = 0;
2502
2503 /*
2504 * The P4 platform supports up to 256 APIC IDs on two separate APIC
2505 * buses (one for LAPICs, one for IOAPICs), where predecessors only
2506 * supports up to 16 on one shared APIC bus.
2507 *
2508 * TBD: Expand LAPIC/IOAPIC support on P4-class systems to take full
2509 * advantage of new APIC bus architecture.
2510 */
2511
2512 if (physids_empty(apic_id_map))
2513 apic->ioapic_phys_id_map(&phys_cpu_present_map, &apic_id_map);
2514
2515 raw_spin_lock_irqsave(&ioapic_lock, flags);
2516 reg_00.raw = io_apic_read(ioapic, 0);
2517 raw_spin_unlock_irqrestore(&ioapic_lock, flags);
2518
2519 if (apic_id >= get_physical_broadcast()) {
2520 printk(KERN_WARNING "IOAPIC[%d]: Invalid apic_id %d, trying "
2521 "%d\n", ioapic, apic_id, reg_00.bits.ID);
2522 apic_id = reg_00.bits.ID;
2523 }
2524
2525 /*
2526 * Every APIC in a system must have a unique ID or we get lots of nice
2527 * 'stuck on smp_invalidate_needed IPI wait' messages.
2528 */
2529 if (apic->check_apicid_used(&apic_id_map, apic_id)) {
2530
2531 for (i = 0; i < get_physical_broadcast(); i++) {
2532 if (!apic->check_apicid_used(&apic_id_map, i))
2533 break;
2534 }
2535
2536 if (i == get_physical_broadcast())
2537 panic("Max apic_id exceeded!\n");
2538
2539 printk(KERN_WARNING "IOAPIC[%d]: apic_id %d already used, "
2540 "trying %d\n", ioapic, apic_id, i);
2541
2542 apic_id = i;
2543 }
2544
2545 physid_set_mask_of_physid(apic_id, &tmp);
2546 physids_or(apic_id_map, apic_id_map, tmp);
2547
2548 if (reg_00.bits.ID != apic_id) {
2549 reg_00.bits.ID = apic_id;
2550
2551 raw_spin_lock_irqsave(&ioapic_lock, flags);
2552 io_apic_write(ioapic, 0, reg_00.raw);
2553 reg_00.raw = io_apic_read(ioapic, 0);
2554 raw_spin_unlock_irqrestore(&ioapic_lock, flags);
2555
2556 /* Sanity check */
2557 if (reg_00.bits.ID != apic_id) {
2558 pr_err("IOAPIC[%d]: Unable to change apic_id!\n",
2559 ioapic);
2560 return -1;
2561 }
2562 }
2563
2564 apic_printk(APIC_VERBOSE, KERN_INFO
2565 "IOAPIC[%d]: Assigned apic_id %d\n", ioapic, apic_id);
2566
2567 return apic_id;
2568}
2569
2570static u8 io_apic_unique_id(int idx, u8 id)
2571{
2572 if ((boot_cpu_data.x86_vendor == X86_VENDOR_INTEL) &&
2573 !APIC_XAPIC(boot_cpu_apic_version))
2574 return io_apic_get_unique_id(idx, id);
2575 else
2576 return id;
2577}
2578#else
2579static u8 io_apic_unique_id(int idx, u8 id)
2580{
2581 union IO_APIC_reg_00 reg_00;
2582 DECLARE_BITMAP(used, 256);
2583 unsigned long flags;
2584 u8 new_id;
2585 int i;
2586
2587 bitmap_zero(dst: used, nbits: 256);
2588 for_each_ioapic(i)
2589 __set_bit(mpc_ioapic_id(i), used);
2590
2591 /* Hand out the requested id if available */
2592 if (!test_bit(id, used))
2593 return id;
2594
2595 /*
2596 * Read the current id from the ioapic and keep it if
2597 * available.
2598 */
2599 raw_spin_lock_irqsave(&ioapic_lock, flags);
2600 reg_00.raw = io_apic_read(apic: idx, reg: 0);
2601 raw_spin_unlock_irqrestore(&ioapic_lock, flags);
2602 new_id = reg_00.bits.ID;
2603 if (!test_bit(new_id, used)) {
2604 apic_printk(APIC_VERBOSE, KERN_INFO
2605 "IOAPIC[%d]: Using reg apic_id %d instead of %d\n",
2606 idx, new_id, id);
2607 return new_id;
2608 }
2609
2610 /*
2611 * Get the next free id and write it to the ioapic.
2612 */
2613 new_id = find_first_zero_bit(addr: used, size: 256);
2614 reg_00.bits.ID = new_id;
2615 raw_spin_lock_irqsave(&ioapic_lock, flags);
2616 io_apic_write(apic: idx, reg: 0, value: reg_00.raw);
2617 reg_00.raw = io_apic_read(apic: idx, reg: 0);
2618 raw_spin_unlock_irqrestore(&ioapic_lock, flags);
2619 /* Sanity check */
2620 BUG_ON(reg_00.bits.ID != new_id);
2621
2622 return new_id;
2623}
2624#endif
2625
2626static int io_apic_get_version(int ioapic)
2627{
2628 union IO_APIC_reg_01 reg_01;
2629 unsigned long flags;
2630
2631 raw_spin_lock_irqsave(&ioapic_lock, flags);
2632 reg_01.raw = io_apic_read(apic: ioapic, reg: 1);
2633 raw_spin_unlock_irqrestore(&ioapic_lock, flags);
2634
2635 return reg_01.bits.version;
2636}
2637
2638/*
2639 * This function updates target affinity of IOAPIC interrupts to include
2640 * the CPUs which came online during SMP bringup.
2641 */
2642#define IOAPIC_RESOURCE_NAME_SIZE 11
2643
2644static struct resource *ioapic_resources;
2645
2646static struct resource * __init ioapic_setup_resources(void)
2647{
2648 unsigned long n;
2649 struct resource *res;
2650 char *mem;
2651 int i;
2652
2653 if (nr_ioapics == 0)
2654 return NULL;
2655
2656 n = IOAPIC_RESOURCE_NAME_SIZE + sizeof(struct resource);
2657 n *= nr_ioapics;
2658
2659 mem = memblock_alloc(size: n, SMP_CACHE_BYTES);
2660 if (!mem)
2661 panic(fmt: "%s: Failed to allocate %lu bytes\n", __func__, n);
2662 res = (void *)mem;
2663
2664 mem += sizeof(struct resource) * nr_ioapics;
2665
2666 for_each_ioapic(i) {
2667 res[i].name = mem;
2668 res[i].flags = IORESOURCE_MEM | IORESOURCE_BUSY;
2669 snprintf(buf: mem, IOAPIC_RESOURCE_NAME_SIZE, fmt: "IOAPIC %u", i);
2670 mem += IOAPIC_RESOURCE_NAME_SIZE;
2671 ioapics[i].iomem_res = &res[i];
2672 }
2673
2674 ioapic_resources = res;
2675
2676 return res;
2677}
2678
2679static void io_apic_set_fixmap(enum fixed_addresses idx, phys_addr_t phys)
2680{
2681 pgprot_t flags = FIXMAP_PAGE_NOCACHE;
2682
2683 /*
2684 * Ensure fixmaps for IO-APIC MMIO respect memory encryption pgprot
2685 * bits, just like normal ioremap():
2686 */
2687 if (cc_platform_has(attr: CC_ATTR_GUEST_MEM_ENCRYPT)) {
2688 if (x86_platform.hyper.is_private_mmio(phys))
2689 flags = pgprot_encrypted(flags);
2690 else
2691 flags = pgprot_decrypted(flags);
2692 }
2693
2694 __set_fixmap(idx, phys, flags);
2695}
2696
2697void __init io_apic_init_mappings(void)
2698{
2699 unsigned long ioapic_phys, idx = FIX_IO_APIC_BASE_0;
2700 struct resource *ioapic_res;
2701 int i;
2702
2703 ioapic_res = ioapic_setup_resources();
2704 for_each_ioapic(i) {
2705 if (smp_found_config) {
2706 ioapic_phys = mpc_ioapic_addr(ioapic_idx: i);
2707#ifdef CONFIG_X86_32
2708 if (!ioapic_phys) {
2709 printk(KERN_ERR
2710 "WARNING: bogus zero IO-APIC "
2711 "address found in MPTABLE, "
2712 "disabling IO/APIC support!\n");
2713 smp_found_config = 0;
2714 ioapic_is_disabled = true;
2715 goto fake_ioapic_page;
2716 }
2717#endif
2718 } else {
2719#ifdef CONFIG_X86_32
2720fake_ioapic_page:
2721#endif
2722 ioapic_phys = (unsigned long)memblock_alloc(PAGE_SIZE,
2723 PAGE_SIZE);
2724 if (!ioapic_phys)
2725 panic(fmt: "%s: Failed to allocate %lu bytes align=0x%lx\n",
2726 __func__, PAGE_SIZE, PAGE_SIZE);
2727 ioapic_phys = __pa(ioapic_phys);
2728 }
2729 io_apic_set_fixmap(idx, phys: ioapic_phys);
2730 apic_printk(APIC_VERBOSE, "mapped IOAPIC to %08lx (%08lx)\n",
2731 __fix_to_virt(idx) + (ioapic_phys & ~PAGE_MASK),
2732 ioapic_phys);
2733 idx++;
2734
2735 ioapic_res->start = ioapic_phys;
2736 ioapic_res->end = ioapic_phys + IO_APIC_SLOT_SIZE - 1;
2737 ioapic_res++;
2738 }
2739}
2740
2741void __init ioapic_insert_resources(void)
2742{
2743 int i;
2744 struct resource *r = ioapic_resources;
2745
2746 if (!r) {
2747 if (nr_ioapics > 0)
2748 printk(KERN_ERR
2749 "IO APIC resources couldn't be allocated.\n");
2750 return;
2751 }
2752
2753 for_each_ioapic(i) {
2754 insert_resource(parent: &iomem_resource, new: r);
2755 r++;
2756 }
2757}
2758
2759int mp_find_ioapic(u32 gsi)
2760{
2761 int i;
2762
2763 if (nr_ioapics == 0)
2764 return -1;
2765
2766 /* Find the IOAPIC that manages this GSI. */
2767 for_each_ioapic(i) {
2768 struct mp_ioapic_gsi *gsi_cfg = mp_ioapic_gsi_routing(ioapic_idx: i);
2769 if (gsi >= gsi_cfg->gsi_base && gsi <= gsi_cfg->gsi_end)
2770 return i;
2771 }
2772
2773 printk(KERN_ERR "ERROR: Unable to locate IOAPIC for GSI %d\n", gsi);
2774 return -1;
2775}
2776
2777int mp_find_ioapic_pin(int ioapic, u32 gsi)
2778{
2779 struct mp_ioapic_gsi *gsi_cfg;
2780
2781 if (WARN_ON(ioapic < 0))
2782 return -1;
2783
2784 gsi_cfg = mp_ioapic_gsi_routing(ioapic_idx: ioapic);
2785 if (WARN_ON(gsi > gsi_cfg->gsi_end))
2786 return -1;
2787
2788 return gsi - gsi_cfg->gsi_base;
2789}
2790
2791static int bad_ioapic_register(int idx)
2792{
2793 union IO_APIC_reg_00 reg_00;
2794 union IO_APIC_reg_01 reg_01;
2795 union IO_APIC_reg_02 reg_02;
2796
2797 reg_00.raw = io_apic_read(apic: idx, reg: 0);
2798 reg_01.raw = io_apic_read(apic: idx, reg: 1);
2799 reg_02.raw = io_apic_read(apic: idx, reg: 2);
2800
2801 if (reg_00.raw == -1 && reg_01.raw == -1 && reg_02.raw == -1) {
2802 pr_warn("I/O APIC 0x%x registers return all ones, skipping!\n",
2803 mpc_ioapic_addr(idx));
2804 return 1;
2805 }
2806
2807 return 0;
2808}
2809
2810static int find_free_ioapic_entry(void)
2811{
2812 int idx;
2813
2814 for (idx = 0; idx < MAX_IO_APICS; idx++)
2815 if (ioapics[idx].nr_registers == 0)
2816 return idx;
2817
2818 return MAX_IO_APICS;
2819}
2820
2821/**
2822 * mp_register_ioapic - Register an IOAPIC device
2823 * @id: hardware IOAPIC ID
2824 * @address: physical address of IOAPIC register area
2825 * @gsi_base: base of GSI associated with the IOAPIC
2826 * @cfg: configuration information for the IOAPIC
2827 */
2828int mp_register_ioapic(int id, u32 address, u32 gsi_base,
2829 struct ioapic_domain_cfg *cfg)
2830{
2831 bool hotplug = !!ioapic_initialized;
2832 struct mp_ioapic_gsi *gsi_cfg;
2833 int idx, ioapic, entries;
2834 u32 gsi_end;
2835
2836 if (!address) {
2837 pr_warn("Bogus (zero) I/O APIC address found, skipping!\n");
2838 return -EINVAL;
2839 }
2840 for_each_ioapic(ioapic)
2841 if (ioapics[ioapic].mp_config.apicaddr == address) {
2842 pr_warn("address 0x%x conflicts with IOAPIC%d\n",
2843 address, ioapic);
2844 return -EEXIST;
2845 }
2846
2847 idx = find_free_ioapic_entry();
2848 if (idx >= MAX_IO_APICS) {
2849 pr_warn("Max # of I/O APICs (%d) exceeded (found %d), skipping\n",
2850 MAX_IO_APICS, idx);
2851 return -ENOSPC;
2852 }
2853
2854 ioapics[idx].mp_config.type = MP_IOAPIC;
2855 ioapics[idx].mp_config.flags = MPC_APIC_USABLE;
2856 ioapics[idx].mp_config.apicaddr = address;
2857
2858 io_apic_set_fixmap(idx: FIX_IO_APIC_BASE_0 + idx, phys: address);
2859 if (bad_ioapic_register(idx)) {
2860 clear_fixmap(FIX_IO_APIC_BASE_0 + idx);
2861 return -ENODEV;
2862 }
2863
2864 ioapics[idx].mp_config.apicid = io_apic_unique_id(idx, id);
2865 ioapics[idx].mp_config.apicver = io_apic_get_version(ioapic: idx);
2866
2867 /*
2868 * Build basic GSI lookup table to facilitate gsi->io_apic lookups
2869 * and to prevent reprogramming of IOAPIC pins (PCI GSIs).
2870 */
2871 entries = io_apic_get_redir_entries(ioapic: idx);
2872 gsi_end = gsi_base + entries - 1;
2873 for_each_ioapic(ioapic) {
2874 gsi_cfg = mp_ioapic_gsi_routing(ioapic_idx: ioapic);
2875 if ((gsi_base >= gsi_cfg->gsi_base &&
2876 gsi_base <= gsi_cfg->gsi_end) ||
2877 (gsi_end >= gsi_cfg->gsi_base &&
2878 gsi_end <= gsi_cfg->gsi_end)) {
2879 pr_warn("GSI range [%u-%u] for new IOAPIC conflicts with GSI[%u-%u]\n",
2880 gsi_base, gsi_end,
2881 gsi_cfg->gsi_base, gsi_cfg->gsi_end);
2882 clear_fixmap(FIX_IO_APIC_BASE_0 + idx);
2883 return -ENOSPC;
2884 }
2885 }
2886 gsi_cfg = mp_ioapic_gsi_routing(ioapic_idx: idx);
2887 gsi_cfg->gsi_base = gsi_base;
2888 gsi_cfg->gsi_end = gsi_end;
2889
2890 ioapics[idx].irqdomain = NULL;
2891 ioapics[idx].irqdomain_cfg = *cfg;
2892
2893 /*
2894 * If mp_register_ioapic() is called during early boot stage when
2895 * walking ACPI/DT tables, it's too early to create irqdomain,
2896 * we are still using bootmem allocator. So delay it to setup_IO_APIC().
2897 */
2898 if (hotplug) {
2899 if (mp_irqdomain_create(ioapic: idx)) {
2900 clear_fixmap(FIX_IO_APIC_BASE_0 + idx);
2901 return -ENOMEM;
2902 }
2903 alloc_ioapic_saved_registers(idx);
2904 }
2905
2906 if (gsi_cfg->gsi_end >= gsi_top)
2907 gsi_top = gsi_cfg->gsi_end + 1;
2908 if (nr_ioapics <= idx)
2909 nr_ioapics = idx + 1;
2910
2911 /* Set nr_registers to mark entry present */
2912 ioapics[idx].nr_registers = entries;
2913
2914 pr_info("IOAPIC[%d]: apic_id %d, version %d, address 0x%x, GSI %d-%d\n",
2915 idx, mpc_ioapic_id(idx),
2916 mpc_ioapic_ver(idx), mpc_ioapic_addr(idx),
2917 gsi_cfg->gsi_base, gsi_cfg->gsi_end);
2918
2919 return 0;
2920}
2921
2922int mp_unregister_ioapic(u32 gsi_base)
2923{
2924 int ioapic, pin;
2925 int found = 0;
2926
2927 for_each_ioapic(ioapic)
2928 if (ioapics[ioapic].gsi_config.gsi_base == gsi_base) {
2929 found = 1;
2930 break;
2931 }
2932 if (!found) {
2933 pr_warn("can't find IOAPIC for GSI %d\n", gsi_base);
2934 return -ENODEV;
2935 }
2936
2937 for_each_pin(ioapic, pin) {
2938 u32 gsi = mp_pin_to_gsi(ioapic, pin);
2939 int irq = mp_map_gsi_to_irq(gsi, flags: 0, NULL);
2940 struct mp_chip_data *data;
2941
2942 if (irq >= 0) {
2943 data = irq_get_chip_data(irq);
2944 if (data && data->count) {
2945 pr_warn("pin%d on IOAPIC%d is still in use.\n",
2946 pin, ioapic);
2947 return -EBUSY;
2948 }
2949 }
2950 }
2951
2952 /* Mark entry not present */
2953 ioapics[ioapic].nr_registers = 0;
2954 ioapic_destroy_irqdomain(idx: ioapic);
2955 free_ioapic_saved_registers(idx: ioapic);
2956 if (ioapics[ioapic].iomem_res)
2957 release_resource(new: ioapics[ioapic].iomem_res);
2958 clear_fixmap(FIX_IO_APIC_BASE_0 + ioapic);
2959 memset(&ioapics[ioapic], 0, sizeof(ioapics[ioapic]));
2960
2961 return 0;
2962}
2963
2964int mp_ioapic_registered(u32 gsi_base)
2965{
2966 int ioapic;
2967
2968 for_each_ioapic(ioapic)
2969 if (ioapics[ioapic].gsi_config.gsi_base == gsi_base)
2970 return 1;
2971
2972 return 0;
2973}
2974
2975static void mp_irqdomain_get_attr(u32 gsi, struct mp_chip_data *data,
2976 struct irq_alloc_info *info)
2977{
2978 if (info && info->ioapic.valid) {
2979 data->is_level = info->ioapic.is_level;
2980 data->active_low = info->ioapic.active_low;
2981 } else if (__acpi_get_override_irq(gsi, trigger: &data->is_level,
2982 polarity: &data->active_low) < 0) {
2983 /* PCI interrupts are always active low level triggered. */
2984 data->is_level = true;
2985 data->active_low = true;
2986 }
2987}
2988
2989/*
2990 * Configure the I/O-APIC specific fields in the routing entry.
2991 *
2992 * This is important to setup the I/O-APIC specific bits (is_level,
2993 * active_low, masked) because the underlying parent domain will only
2994 * provide the routing information and is oblivious of the I/O-APIC
2995 * specific bits.
2996 *
2997 * The entry is just preconfigured at this point and not written into the
2998 * RTE. This happens later during activation which will fill in the actual
2999 * routing information.
3000 */
3001static void mp_preconfigure_entry(struct mp_chip_data *data)
3002{
3003 struct IO_APIC_route_entry *entry = &data->entry;
3004
3005 memset(entry, 0, sizeof(*entry));
3006 entry->is_level = data->is_level;
3007 entry->active_low = data->active_low;
3008 /*
3009 * Mask level triggered irqs. Edge triggered irqs are masked
3010 * by the irq core code in case they fire.
3011 */
3012 entry->masked = data->is_level;
3013}
3014
3015int mp_irqdomain_alloc(struct irq_domain *domain, unsigned int virq,
3016 unsigned int nr_irqs, void *arg)
3017{
3018 struct irq_alloc_info *info = arg;
3019 struct mp_chip_data *data;
3020 struct irq_data *irq_data;
3021 int ret, ioapic, pin;
3022 unsigned long flags;
3023
3024 if (!info || nr_irqs > 1)
3025 return -EINVAL;
3026 irq_data = irq_domain_get_irq_data(domain, virq);
3027 if (!irq_data)
3028 return -EINVAL;
3029
3030 ioapic = mp_irqdomain_ioapic_idx(domain);
3031 pin = info->ioapic.pin;
3032 if (irq_find_mapping(domain, hwirq: (irq_hw_number_t)pin) > 0)
3033 return -EEXIST;
3034
3035 data = kzalloc(size: sizeof(*data), GFP_KERNEL);
3036 if (!data)
3037 return -ENOMEM;
3038
3039 ret = irq_domain_alloc_irqs_parent(domain, irq_base: virq, nr_irqs, arg: info);
3040 if (ret < 0) {
3041 kfree(objp: data);
3042 return ret;
3043 }
3044
3045 INIT_LIST_HEAD(list: &data->irq_2_pin);
3046 irq_data->hwirq = info->ioapic.pin;
3047 irq_data->chip = (domain->parent == x86_vector_domain) ?
3048 &ioapic_chip : &ioapic_ir_chip;
3049 irq_data->chip_data = data;
3050 mp_irqdomain_get_attr(gsi: mp_pin_to_gsi(ioapic, pin), data, info);
3051
3052 add_pin_to_irq_node(data, node: ioapic_alloc_attr_node(info), apic: ioapic, pin);
3053
3054 mp_preconfigure_entry(data);
3055 mp_register_handler(irq: virq, level: data->is_level);
3056
3057 local_irq_save(flags);
3058 if (virq < nr_legacy_irqs())
3059 legacy_pic->mask(virq);
3060 local_irq_restore(flags);
3061
3062 apic_printk(APIC_VERBOSE, KERN_DEBUG
3063 "IOAPIC[%d]: Preconfigured routing entry (%d-%d -> IRQ %d Level:%i ActiveLow:%i)\n",
3064 ioapic, mpc_ioapic_id(ioapic), pin, virq,
3065 data->is_level, data->active_low);
3066 return 0;
3067}
3068
3069void mp_irqdomain_free(struct irq_domain *domain, unsigned int virq,
3070 unsigned int nr_irqs)
3071{
3072 struct irq_data *irq_data;
3073 struct mp_chip_data *data;
3074
3075 BUG_ON(nr_irqs != 1);
3076 irq_data = irq_domain_get_irq_data(domain, virq);
3077 if (irq_data && irq_data->chip_data) {
3078 data = irq_data->chip_data;
3079 __remove_pin_from_irq(data, apic: mp_irqdomain_ioapic_idx(domain),
3080 pin: (int)irq_data->hwirq);
3081 WARN_ON(!list_empty(&data->irq_2_pin));
3082 kfree(objp: irq_data->chip_data);
3083 }
3084 irq_domain_free_irqs_top(domain, virq, nr_irqs);
3085}
3086
3087int mp_irqdomain_activate(struct irq_domain *domain,
3088 struct irq_data *irq_data, bool reserve)
3089{
3090 unsigned long flags;
3091
3092 raw_spin_lock_irqsave(&ioapic_lock, flags);
3093 ioapic_configure_entry(irqd: irq_data);
3094 raw_spin_unlock_irqrestore(&ioapic_lock, flags);
3095 return 0;
3096}
3097
3098void mp_irqdomain_deactivate(struct irq_domain *domain,
3099 struct irq_data *irq_data)
3100{
3101 /* It won't be called for IRQ with multiple IOAPIC pins associated */
3102 ioapic_mask_entry(apic: mp_irqdomain_ioapic_idx(domain),
3103 pin: (int)irq_data->hwirq);
3104}
3105
3106int mp_irqdomain_ioapic_idx(struct irq_domain *domain)
3107{
3108 return (int)(long)domain->host_data;
3109}
3110
3111const struct irq_domain_ops mp_ioapic_irqdomain_ops = {
3112 .alloc = mp_irqdomain_alloc,
3113 .free = mp_irqdomain_free,
3114 .activate = mp_irqdomain_activate,
3115 .deactivate = mp_irqdomain_deactivate,
3116};
3117

source code of linux/arch/x86/kernel/apic/io_apic.c