]> git.proxmox.com Git - mirror_ubuntu-hirsute-kernel.git/blob - arch/x86/kernel/apic/io_apic.c
x86/ioapic: Cleanup the timer_works() irqflags mess
[mirror_ubuntu-hirsute-kernel.git] / arch / x86 / kernel / apic / io_apic.c
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
69 #define for_each_ioapic(idx) \
70 for ((idx) = 0; (idx) < nr_ioapics; (idx)++)
71 #define for_each_ioapic_reverse(idx) \
72 for ((idx) = nr_ioapics - 1; (idx) >= 0; (idx)--)
73 #define for_each_pin(idx, pin) \
74 for ((pin) = 0; (pin) < ioapics[(idx)].nr_registers; (pin)++)
75 #define for_each_ioapic_pin(idx, pin) \
76 for_each_ioapic((idx)) \
77 for_each_pin((idx), (pin))
78 #define for_each_irq_pin(entry, head) \
79 list_for_each_entry(entry, &head, list)
80
81 static DEFINE_RAW_SPINLOCK(ioapic_lock);
82 static DEFINE_MUTEX(ioapic_mutex);
83 static unsigned int ioapic_dynirq_base;
84 static int ioapic_initialized;
85
86 struct irq_pin_list {
87 struct list_head list;
88 int apic, pin;
89 };
90
91 struct mp_chip_data {
92 struct list_head irq_2_pin;
93 struct IO_APIC_route_entry entry;
94 bool is_level;
95 bool active_low;
96 bool isa_irq;
97 u32 count;
98 };
99
100 struct mp_ioapic_gsi {
101 u32 gsi_base;
102 u32 gsi_end;
103 };
104
105 static struct ioapic {
106 /*
107 * # of IRQ routing registers
108 */
109 int nr_registers;
110 /*
111 * Saved state during suspend/resume, or while enabling intr-remap.
112 */
113 struct IO_APIC_route_entry *saved_registers;
114 /* I/O APIC config */
115 struct mpc_ioapic mp_config;
116 /* IO APIC gsi routing info */
117 struct mp_ioapic_gsi gsi_config;
118 struct ioapic_domain_cfg irqdomain_cfg;
119 struct irq_domain *irqdomain;
120 struct resource *iomem_res;
121 } ioapics[MAX_IO_APICS];
122
123 #define mpc_ioapic_ver(ioapic_idx) ioapics[ioapic_idx].mp_config.apicver
124
125 int mpc_ioapic_id(int ioapic_idx)
126 {
127 return ioapics[ioapic_idx].mp_config.apicid;
128 }
129
130 unsigned int mpc_ioapic_addr(int ioapic_idx)
131 {
132 return ioapics[ioapic_idx].mp_config.apicaddr;
133 }
134
135 static inline struct mp_ioapic_gsi *mp_ioapic_gsi_routing(int ioapic_idx)
136 {
137 return &ioapics[ioapic_idx].gsi_config;
138 }
139
140 static inline int mp_ioapic_pin_count(int ioapic)
141 {
142 struct mp_ioapic_gsi *gsi_cfg = mp_ioapic_gsi_routing(ioapic);
143
144 return gsi_cfg->gsi_end - gsi_cfg->gsi_base + 1;
145 }
146
147 static inline u32 mp_pin_to_gsi(int ioapic, int pin)
148 {
149 return mp_ioapic_gsi_routing(ioapic)->gsi_base + pin;
150 }
151
152 static inline bool mp_is_legacy_irq(int irq)
153 {
154 return irq >= 0 && irq < nr_legacy_irqs();
155 }
156
157 static inline struct irq_domain *mp_ioapic_irqdomain(int ioapic)
158 {
159 return ioapics[ioapic].irqdomain;
160 }
161
162 int nr_ioapics;
163
164 /* The one past the highest gsi number used */
165 u32 gsi_top;
166
167 /* MP IRQ source entries */
168 struct mpc_intsrc mp_irqs[MAX_IRQ_SOURCES];
169
170 /* # of MP IRQ source entries */
171 int mp_irq_entries;
172
173 #ifdef CONFIG_EISA
174 int mp_bus_id_to_type[MAX_MP_BUSSES];
175 #endif
176
177 DECLARE_BITMAP(mp_bus_not_pci, MAX_MP_BUSSES);
178
179 int skip_ioapic_setup;
180
181 /**
182 * disable_ioapic_support() - disables ioapic support at runtime
183 */
184 void disable_ioapic_support(void)
185 {
186 #ifdef CONFIG_PCI
187 noioapicquirk = 1;
188 noioapicreroute = -1;
189 #endif
190 skip_ioapic_setup = 1;
191 }
192
193 static int __init parse_noapic(char *str)
194 {
195 /* disable IO-APIC */
196 disable_ioapic_support();
197 return 0;
198 }
199 early_param("noapic", parse_noapic);
200
201 /* Will be called in mpparse/acpi/sfi codes for saving IRQ info */
202 void mp_save_irq(struct mpc_intsrc *m)
203 {
204 int i;
205
206 apic_printk(APIC_VERBOSE, "Int: type %d, pol %d, trig %d, bus %02x,"
207 " IRQ %02x, APIC ID %x, APIC INT %02x\n",
208 m->irqtype, m->irqflag & 3, (m->irqflag >> 2) & 3, m->srcbus,
209 m->srcbusirq, m->dstapic, m->dstirq);
210
211 for (i = 0; i < mp_irq_entries; i++) {
212 if (!memcmp(&mp_irqs[i], m, sizeof(*m)))
213 return;
214 }
215
216 memcpy(&mp_irqs[mp_irq_entries], m, sizeof(*m));
217 if (++mp_irq_entries == MAX_IRQ_SOURCES)
218 panic("Max # of irq sources exceeded!!\n");
219 }
220
221 static void alloc_ioapic_saved_registers(int idx)
222 {
223 size_t size;
224
225 if (ioapics[idx].saved_registers)
226 return;
227
228 size = sizeof(struct IO_APIC_route_entry) * ioapics[idx].nr_registers;
229 ioapics[idx].saved_registers = kzalloc(size, GFP_KERNEL);
230 if (!ioapics[idx].saved_registers)
231 pr_err("IOAPIC %d: suspend/resume impossible!\n", idx);
232 }
233
234 static void free_ioapic_saved_registers(int idx)
235 {
236 kfree(ioapics[idx].saved_registers);
237 ioapics[idx].saved_registers = NULL;
238 }
239
240 int __init arch_early_ioapic_init(void)
241 {
242 int i;
243
244 if (!nr_legacy_irqs())
245 io_apic_irqs = ~0UL;
246
247 for_each_ioapic(i)
248 alloc_ioapic_saved_registers(i);
249
250 return 0;
251 }
252
253 struct io_apic {
254 unsigned int index;
255 unsigned int unused[3];
256 unsigned int data;
257 unsigned int unused2[11];
258 unsigned int eoi;
259 };
260
261 static __attribute_const__ struct io_apic __iomem *io_apic_base(int idx)
262 {
263 return (void __iomem *) __fix_to_virt(FIX_IO_APIC_BASE_0 + idx)
264 + (mpc_ioapic_addr(idx) & ~PAGE_MASK);
265 }
266
267 static inline void io_apic_eoi(unsigned int apic, unsigned int vector)
268 {
269 struct io_apic __iomem *io_apic = io_apic_base(apic);
270 writel(vector, &io_apic->eoi);
271 }
272
273 unsigned int native_io_apic_read(unsigned int apic, unsigned int reg)
274 {
275 struct io_apic __iomem *io_apic = io_apic_base(apic);
276 writel(reg, &io_apic->index);
277 return readl(&io_apic->data);
278 }
279
280 static void io_apic_write(unsigned int apic, unsigned int reg,
281 unsigned int value)
282 {
283 struct io_apic __iomem *io_apic = io_apic_base(apic);
284
285 writel(reg, &io_apic->index);
286 writel(value, &io_apic->data);
287 }
288
289 static struct IO_APIC_route_entry __ioapic_read_entry(int apic, int pin)
290 {
291 struct IO_APIC_route_entry entry;
292
293 entry.w1 = io_apic_read(apic, 0x10 + 2 * pin);
294 entry.w2 = io_apic_read(apic, 0x11 + 2 * pin);
295
296 return entry;
297 }
298
299 static struct IO_APIC_route_entry ioapic_read_entry(int apic, int pin)
300 {
301 struct IO_APIC_route_entry entry;
302 unsigned long flags;
303
304 raw_spin_lock_irqsave(&ioapic_lock, flags);
305 entry = __ioapic_read_entry(apic, pin);
306 raw_spin_unlock_irqrestore(&ioapic_lock, flags);
307
308 return entry;
309 }
310
311 /*
312 * When we write a new IO APIC routing entry, we need to write the high
313 * word first! If the mask bit in the low word is clear, we will enable
314 * the interrupt, and we need to make sure the entry is fully populated
315 * before that happens.
316 */
317 static void __ioapic_write_entry(int apic, int pin, struct IO_APIC_route_entry e)
318 {
319 io_apic_write(apic, 0x11 + 2*pin, e.w2);
320 io_apic_write(apic, 0x10 + 2*pin, e.w1);
321 }
322
323 static void ioapic_write_entry(int apic, int pin, struct IO_APIC_route_entry e)
324 {
325 unsigned long flags;
326
327 raw_spin_lock_irqsave(&ioapic_lock, flags);
328 __ioapic_write_entry(apic, pin, e);
329 raw_spin_unlock_irqrestore(&ioapic_lock, flags);
330 }
331
332 /*
333 * When we mask an IO APIC routing entry, we need to write the low
334 * word first, in order to set the mask bit before we change the
335 * high bits!
336 */
337 static void ioapic_mask_entry(int apic, int pin)
338 {
339 struct IO_APIC_route_entry e = { .masked = true };
340 unsigned long flags;
341
342 raw_spin_lock_irqsave(&ioapic_lock, flags);
343 io_apic_write(apic, 0x10 + 2*pin, e.w1);
344 io_apic_write(apic, 0x11 + 2*pin, e.w2);
345 raw_spin_unlock_irqrestore(&ioapic_lock, flags);
346 }
347
348 /*
349 * The common case is 1:1 IRQ<->pin mappings. Sometimes there are
350 * shared ISA-space IRQs, so we have to support them. We are super
351 * fast in the common case, and fast for shared ISA-space IRQs.
352 */
353 static int __add_pin_to_irq_node(struct mp_chip_data *data,
354 int node, int apic, int pin)
355 {
356 struct irq_pin_list *entry;
357
358 /* don't allow duplicates */
359 for_each_irq_pin(entry, data->irq_2_pin)
360 if (entry->apic == apic && entry->pin == pin)
361 return 0;
362
363 entry = kzalloc_node(sizeof(struct irq_pin_list), GFP_ATOMIC, node);
364 if (!entry) {
365 pr_err("can not alloc irq_pin_list (%d,%d,%d)\n",
366 node, apic, pin);
367 return -ENOMEM;
368 }
369 entry->apic = apic;
370 entry->pin = pin;
371 list_add_tail(&entry->list, &data->irq_2_pin);
372
373 return 0;
374 }
375
376 static void __remove_pin_from_irq(struct mp_chip_data *data, int apic, int pin)
377 {
378 struct irq_pin_list *tmp, *entry;
379
380 list_for_each_entry_safe(entry, tmp, &data->irq_2_pin, list)
381 if (entry->apic == apic && entry->pin == pin) {
382 list_del(&entry->list);
383 kfree(entry);
384 return;
385 }
386 }
387
388 static void add_pin_to_irq_node(struct mp_chip_data *data,
389 int node, int apic, int pin)
390 {
391 if (__add_pin_to_irq_node(data, node, apic, pin))
392 panic("IO-APIC: failed to add irq-pin. Can not proceed\n");
393 }
394
395 /*
396 * Reroute an IRQ to a different pin.
397 */
398 static void __init replace_pin_at_irq_node(struct mp_chip_data *data, int node,
399 int oldapic, int oldpin,
400 int newapic, int newpin)
401 {
402 struct irq_pin_list *entry;
403
404 for_each_irq_pin(entry, data->irq_2_pin) {
405 if (entry->apic == oldapic && entry->pin == oldpin) {
406 entry->apic = newapic;
407 entry->pin = newpin;
408 /* every one is different, right? */
409 return;
410 }
411 }
412
413 /* old apic/pin didn't exist, so just add new ones */
414 add_pin_to_irq_node(data, node, newapic, newpin);
415 }
416
417 static void io_apic_modify_irq(struct mp_chip_data *data, bool masked,
418 void (*final)(struct irq_pin_list *entry))
419 {
420 struct irq_pin_list *entry;
421
422 data->entry.masked = masked;
423
424 for_each_irq_pin(entry, data->irq_2_pin) {
425 io_apic_write(entry->apic, 0x10 + 2 * entry->pin, data->entry.w1);
426 if (final)
427 final(entry);
428 }
429 }
430
431 static void io_apic_sync(struct irq_pin_list *entry)
432 {
433 /*
434 * Synchronize the IO-APIC and the CPU by doing
435 * a dummy read from the IO-APIC
436 */
437 struct io_apic __iomem *io_apic;
438
439 io_apic = io_apic_base(entry->apic);
440 readl(&io_apic->data);
441 }
442
443 static void mask_ioapic_irq(struct irq_data *irq_data)
444 {
445 struct mp_chip_data *data = irq_data->chip_data;
446 unsigned long flags;
447
448 raw_spin_lock_irqsave(&ioapic_lock, flags);
449 io_apic_modify_irq(data, true, &io_apic_sync);
450 raw_spin_unlock_irqrestore(&ioapic_lock, flags);
451 }
452
453 static void __unmask_ioapic(struct mp_chip_data *data)
454 {
455 io_apic_modify_irq(data, false, NULL);
456 }
457
458 static void unmask_ioapic_irq(struct irq_data *irq_data)
459 {
460 struct mp_chip_data *data = irq_data->chip_data;
461 unsigned long flags;
462
463 raw_spin_lock_irqsave(&ioapic_lock, flags);
464 __unmask_ioapic(data);
465 raw_spin_unlock_irqrestore(&ioapic_lock, flags);
466 }
467
468 /*
469 * IO-APIC versions below 0x20 don't support EOI register.
470 * For the record, here is the information about various versions:
471 * 0Xh 82489DX
472 * 1Xh I/OAPIC or I/O(x)APIC which are not PCI 2.2 Compliant
473 * 2Xh I/O(x)APIC which is PCI 2.2 Compliant
474 * 30h-FFh Reserved
475 *
476 * Some of the Intel ICH Specs (ICH2 to ICH5) documents the io-apic
477 * version as 0x2. This is an error with documentation and these ICH chips
478 * use io-apic's of version 0x20.
479 *
480 * For IO-APIC's with EOI register, we use that to do an explicit EOI.
481 * Otherwise, we simulate the EOI message manually by changing the trigger
482 * mode to edge and then back to level, with RTE being masked during this.
483 */
484 static void __eoi_ioapic_pin(int apic, int pin, int vector)
485 {
486 if (mpc_ioapic_ver(apic) >= 0x20) {
487 io_apic_eoi(apic, vector);
488 } else {
489 struct IO_APIC_route_entry entry, entry1;
490
491 entry = entry1 = __ioapic_read_entry(apic, pin);
492
493 /*
494 * Mask the entry and change the trigger mode to edge.
495 */
496 entry1.masked = true;
497 entry1.is_level = false;
498
499 __ioapic_write_entry(apic, pin, entry1);
500
501 /*
502 * Restore the previous level triggered entry.
503 */
504 __ioapic_write_entry(apic, pin, entry);
505 }
506 }
507
508 static void eoi_ioapic_pin(int vector, struct mp_chip_data *data)
509 {
510 unsigned long flags;
511 struct irq_pin_list *entry;
512
513 raw_spin_lock_irqsave(&ioapic_lock, flags);
514 for_each_irq_pin(entry, data->irq_2_pin)
515 __eoi_ioapic_pin(entry->apic, entry->pin, vector);
516 raw_spin_unlock_irqrestore(&ioapic_lock, flags);
517 }
518
519 static void clear_IO_APIC_pin(unsigned int apic, unsigned int pin)
520 {
521 struct IO_APIC_route_entry entry;
522
523 /* Check delivery_mode to be sure we're not clearing an SMI pin */
524 entry = ioapic_read_entry(apic, pin);
525 if (entry.delivery_mode == APIC_DELIVERY_MODE_SMI)
526 return;
527
528 /*
529 * Make sure the entry is masked and re-read the contents to check
530 * if it is a level triggered pin and if the remote-IRR is set.
531 */
532 if (!entry.masked) {
533 entry.masked = true;
534 ioapic_write_entry(apic, pin, entry);
535 entry = ioapic_read_entry(apic, pin);
536 }
537
538 if (entry.irr) {
539 unsigned long flags;
540
541 /*
542 * Make sure the trigger mode is set to level. Explicit EOI
543 * doesn't clear the remote-IRR if the trigger mode is not
544 * set to level.
545 */
546 if (!entry.is_level) {
547 entry.is_level = true;
548 ioapic_write_entry(apic, pin, entry);
549 }
550 raw_spin_lock_irqsave(&ioapic_lock, flags);
551 __eoi_ioapic_pin(apic, pin, entry.vector);
552 raw_spin_unlock_irqrestore(&ioapic_lock, flags);
553 }
554
555 /*
556 * Clear the rest of the bits in the IO-APIC RTE except for the mask
557 * bit.
558 */
559 ioapic_mask_entry(apic, pin);
560 entry = ioapic_read_entry(apic, pin);
561 if (entry.irr)
562 pr_err("Unable to reset IRR for apic: %d, pin :%d\n",
563 mpc_ioapic_id(apic), pin);
564 }
565
566 void clear_IO_APIC (void)
567 {
568 int apic, pin;
569
570 for_each_ioapic_pin(apic, pin)
571 clear_IO_APIC_pin(apic, pin);
572 }
573
574 #ifdef CONFIG_X86_32
575 /*
576 * support for broken MP BIOSs, enables hand-redirection of PIRQ0-7 to
577 * specific CPU-side IRQs.
578 */
579
580 #define MAX_PIRQS 8
581 static int pirq_entries[MAX_PIRQS] = {
582 [0 ... MAX_PIRQS - 1] = -1
583 };
584
585 static int __init ioapic_pirq_setup(char *str)
586 {
587 int i, max;
588 int ints[MAX_PIRQS+1];
589
590 get_options(str, ARRAY_SIZE(ints), ints);
591
592 apic_printk(APIC_VERBOSE, KERN_INFO
593 "PIRQ redirection, working around broken MP-BIOS.\n");
594 max = MAX_PIRQS;
595 if (ints[0] < MAX_PIRQS)
596 max = ints[0];
597
598 for (i = 0; i < max; i++) {
599 apic_printk(APIC_VERBOSE, KERN_DEBUG
600 "... PIRQ%d -> IRQ %d\n", i, ints[i+1]);
601 /*
602 * PIRQs are mapped upside down, usually.
603 */
604 pirq_entries[MAX_PIRQS-i-1] = ints[i+1];
605 }
606 return 1;
607 }
608
609 __setup("pirq=", ioapic_pirq_setup);
610 #endif /* CONFIG_X86_32 */
611
612 /*
613 * Saves all the IO-APIC RTE's
614 */
615 int save_ioapic_entries(void)
616 {
617 int apic, pin;
618 int err = 0;
619
620 for_each_ioapic(apic) {
621 if (!ioapics[apic].saved_registers) {
622 err = -ENOMEM;
623 continue;
624 }
625
626 for_each_pin(apic, pin)
627 ioapics[apic].saved_registers[pin] =
628 ioapic_read_entry(apic, pin);
629 }
630
631 return err;
632 }
633
634 /*
635 * Mask all IO APIC entries.
636 */
637 void mask_ioapic_entries(void)
638 {
639 int apic, pin;
640
641 for_each_ioapic(apic) {
642 if (!ioapics[apic].saved_registers)
643 continue;
644
645 for_each_pin(apic, pin) {
646 struct IO_APIC_route_entry entry;
647
648 entry = ioapics[apic].saved_registers[pin];
649 if (!entry.masked) {
650 entry.masked = true;
651 ioapic_write_entry(apic, pin, entry);
652 }
653 }
654 }
655 }
656
657 /*
658 * Restore IO APIC entries which was saved in the ioapic structure.
659 */
660 int restore_ioapic_entries(void)
661 {
662 int apic, pin;
663
664 for_each_ioapic(apic) {
665 if (!ioapics[apic].saved_registers)
666 continue;
667
668 for_each_pin(apic, pin)
669 ioapic_write_entry(apic, pin,
670 ioapics[apic].saved_registers[pin]);
671 }
672 return 0;
673 }
674
675 /*
676 * Find the IRQ entry number of a certain pin.
677 */
678 static int find_irq_entry(int ioapic_idx, int pin, int type)
679 {
680 int i;
681
682 for (i = 0; i < mp_irq_entries; i++)
683 if (mp_irqs[i].irqtype == type &&
684 (mp_irqs[i].dstapic == mpc_ioapic_id(ioapic_idx) ||
685 mp_irqs[i].dstapic == MP_APIC_ALL) &&
686 mp_irqs[i].dstirq == pin)
687 return i;
688
689 return -1;
690 }
691
692 /*
693 * Find the pin to which IRQ[irq] (ISA) is connected
694 */
695 static int __init find_isa_irq_pin(int irq, int type)
696 {
697 int i;
698
699 for (i = 0; i < mp_irq_entries; i++) {
700 int lbus = mp_irqs[i].srcbus;
701
702 if (test_bit(lbus, mp_bus_not_pci) &&
703 (mp_irqs[i].irqtype == type) &&
704 (mp_irqs[i].srcbusirq == irq))
705
706 return mp_irqs[i].dstirq;
707 }
708 return -1;
709 }
710
711 static int __init find_isa_irq_apic(int irq, int type)
712 {
713 int i;
714
715 for (i = 0; i < mp_irq_entries; i++) {
716 int lbus = mp_irqs[i].srcbus;
717
718 if (test_bit(lbus, mp_bus_not_pci) &&
719 (mp_irqs[i].irqtype == type) &&
720 (mp_irqs[i].srcbusirq == irq))
721 break;
722 }
723
724 if (i < mp_irq_entries) {
725 int ioapic_idx;
726
727 for_each_ioapic(ioapic_idx)
728 if (mpc_ioapic_id(ioapic_idx) == mp_irqs[i].dstapic)
729 return ioapic_idx;
730 }
731
732 return -1;
733 }
734
735 static bool irq_active_low(int idx)
736 {
737 int bus = mp_irqs[idx].srcbus;
738
739 /*
740 * Determine IRQ line polarity (high active or low active):
741 */
742 switch (mp_irqs[idx].irqflag & MP_IRQPOL_MASK) {
743 case MP_IRQPOL_DEFAULT:
744 /*
745 * Conforms to spec, ie. bus-type dependent polarity. PCI
746 * defaults to low active. [E]ISA defaults to high active.
747 */
748 return !test_bit(bus, mp_bus_not_pci);
749 case MP_IRQPOL_ACTIVE_HIGH:
750 return false;
751 case MP_IRQPOL_RESERVED:
752 pr_warn("IOAPIC: Invalid polarity: 2, defaulting to low\n");
753 fallthrough;
754 case MP_IRQPOL_ACTIVE_LOW:
755 default: /* Pointless default required due to do gcc stupidity */
756 return true;
757 }
758 }
759
760 #ifdef CONFIG_EISA
761 /*
762 * EISA Edge/Level control register, ELCR
763 */
764 static bool EISA_ELCR(unsigned int irq)
765 {
766 if (irq < nr_legacy_irqs()) {
767 unsigned int port = 0x4d0 + (irq >> 3);
768 return (inb(port) >> (irq & 7)) & 1;
769 }
770 apic_printk(APIC_VERBOSE, KERN_INFO
771 "Broken MPtable reports ISA irq %d\n", irq);
772 return false;
773 }
774
775 /*
776 * EISA interrupts are always active high and can be edge or level
777 * triggered depending on the ELCR value. If an interrupt is listed as
778 * EISA conforming in the MP table, that means its trigger type must be
779 * read in from the ELCR.
780 */
781 static bool eisa_irq_is_level(int idx, int bus, bool level)
782 {
783 switch (mp_bus_id_to_type[bus]) {
784 case MP_BUS_PCI:
785 case MP_BUS_ISA:
786 return level;
787 case MP_BUS_EISA:
788 return EISA_ELCR(mp_irqs[idx].srcbusirq);
789 }
790 pr_warn("IOAPIC: Invalid srcbus: %d defaulting to level\n", bus);
791 return true;
792 }
793 #else
794 static inline int eisa_irq_is_level(int idx, int bus, bool level)
795 {
796 return level;
797 }
798 #endif
799
800 static bool irq_is_level(int idx)
801 {
802 int bus = mp_irqs[idx].srcbus;
803 bool level;
804
805 /*
806 * Determine IRQ trigger mode (edge or level sensitive):
807 */
808 switch (mp_irqs[idx].irqflag & MP_IRQTRIG_MASK) {
809 case MP_IRQTRIG_DEFAULT:
810 /*
811 * Conforms to spec, ie. bus-type dependent trigger
812 * mode. PCI defaults to level, ISA to edge.
813 */
814 level = !test_bit(bus, mp_bus_not_pci);
815 /* Take EISA into account */
816 return eisa_irq_is_level(idx, bus, level);
817 case MP_IRQTRIG_EDGE:
818 return false;
819 case MP_IRQTRIG_RESERVED:
820 pr_warn("IOAPIC: Invalid trigger mode 2 defaulting to level\n");
821 fallthrough;
822 case MP_IRQTRIG_LEVEL:
823 default: /* Pointless default required due to do gcc stupidity */
824 return true;
825 }
826 }
827
828 static int __acpi_get_override_irq(u32 gsi, bool *trigger, bool *polarity)
829 {
830 int ioapic, pin, idx;
831
832 if (skip_ioapic_setup)
833 return -1;
834
835 ioapic = mp_find_ioapic(gsi);
836 if (ioapic < 0)
837 return -1;
838
839 pin = mp_find_ioapic_pin(ioapic, gsi);
840 if (pin < 0)
841 return -1;
842
843 idx = find_irq_entry(ioapic, pin, mp_INT);
844 if (idx < 0)
845 return -1;
846
847 *trigger = irq_is_level(idx);
848 *polarity = irq_active_low(idx);
849 return 0;
850 }
851
852 #ifdef CONFIG_ACPI
853 int acpi_get_override_irq(u32 gsi, int *is_level, int *active_low)
854 {
855 *is_level = *active_low = 0;
856 return __acpi_get_override_irq(gsi, (bool *)is_level,
857 (bool *)active_low);
858 }
859 #endif
860
861 void ioapic_set_alloc_attr(struct irq_alloc_info *info, int node,
862 int trigger, int polarity)
863 {
864 init_irq_alloc_info(info, NULL);
865 info->type = X86_IRQ_ALLOC_TYPE_IOAPIC;
866 info->ioapic.node = node;
867 info->ioapic.is_level = trigger;
868 info->ioapic.active_low = polarity;
869 info->ioapic.valid = 1;
870 }
871
872 static void ioapic_copy_alloc_attr(struct irq_alloc_info *dst,
873 struct irq_alloc_info *src,
874 u32 gsi, int ioapic_idx, int pin)
875 {
876 bool level, pol_low;
877
878 copy_irq_alloc_info(dst, src);
879 dst->type = X86_IRQ_ALLOC_TYPE_IOAPIC;
880 dst->devid = mpc_ioapic_id(ioapic_idx);
881 dst->ioapic.pin = pin;
882 dst->ioapic.valid = 1;
883 if (src && src->ioapic.valid) {
884 dst->ioapic.node = src->ioapic.node;
885 dst->ioapic.is_level = src->ioapic.is_level;
886 dst->ioapic.active_low = src->ioapic.active_low;
887 } else {
888 dst->ioapic.node = NUMA_NO_NODE;
889 if (__acpi_get_override_irq(gsi, &level, &pol_low) >= 0) {
890 dst->ioapic.is_level = level;
891 dst->ioapic.active_low = pol_low;
892 } else {
893 /*
894 * PCI interrupts are always active low level
895 * triggered.
896 */
897 dst->ioapic.is_level = true;
898 dst->ioapic.active_low = true;
899 }
900 }
901 }
902
903 static int ioapic_alloc_attr_node(struct irq_alloc_info *info)
904 {
905 return (info && info->ioapic.valid) ? info->ioapic.node : NUMA_NO_NODE;
906 }
907
908 static void mp_register_handler(unsigned int irq, bool level)
909 {
910 irq_flow_handler_t hdl;
911 bool fasteoi;
912
913 if (level) {
914 irq_set_status_flags(irq, IRQ_LEVEL);
915 fasteoi = true;
916 } else {
917 irq_clear_status_flags(irq, IRQ_LEVEL);
918 fasteoi = false;
919 }
920
921 hdl = fasteoi ? handle_fasteoi_irq : handle_edge_irq;
922 __irq_set_handler(irq, hdl, 0, fasteoi ? "fasteoi" : "edge");
923 }
924
925 static bool mp_check_pin_attr(int irq, struct irq_alloc_info *info)
926 {
927 struct mp_chip_data *data = irq_get_chip_data(irq);
928
929 /*
930 * setup_IO_APIC_irqs() programs all legacy IRQs with default trigger
931 * and polarity attirbutes. So allow the first user to reprogram the
932 * pin with real trigger and polarity attributes.
933 */
934 if (irq < nr_legacy_irqs() && data->count == 1) {
935 if (info->ioapic.is_level != data->is_level)
936 mp_register_handler(irq, info->ioapic.is_level);
937 data->entry.is_level = data->is_level = info->ioapic.is_level;
938 data->entry.active_low = data->active_low = info->ioapic.active_low;
939 }
940
941 return data->is_level == info->ioapic.is_level &&
942 data->active_low == info->ioapic.active_low;
943 }
944
945 static int alloc_irq_from_domain(struct irq_domain *domain, int ioapic, u32 gsi,
946 struct irq_alloc_info *info)
947 {
948 bool legacy = false;
949 int irq = -1;
950 int type = ioapics[ioapic].irqdomain_cfg.type;
951
952 switch (type) {
953 case IOAPIC_DOMAIN_LEGACY:
954 /*
955 * Dynamically allocate IRQ number for non-ISA IRQs in the first
956 * 16 GSIs on some weird platforms.
957 */
958 if (!ioapic_initialized || gsi >= nr_legacy_irqs())
959 irq = gsi;
960 legacy = mp_is_legacy_irq(irq);
961 break;
962 case IOAPIC_DOMAIN_STRICT:
963 irq = gsi;
964 break;
965 case IOAPIC_DOMAIN_DYNAMIC:
966 break;
967 default:
968 WARN(1, "ioapic: unknown irqdomain type %d\n", type);
969 return -1;
970 }
971
972 return __irq_domain_alloc_irqs(domain, irq, 1,
973 ioapic_alloc_attr_node(info),
974 info, legacy, NULL);
975 }
976
977 /*
978 * Need special handling for ISA IRQs because there may be multiple IOAPIC pins
979 * sharing the same ISA IRQ number and irqdomain only supports 1:1 mapping
980 * between IOAPIC pin and IRQ number. A typical IOAPIC has 24 pins, pin 0-15 are
981 * used for legacy IRQs and pin 16-23 are used for PCI IRQs (PIRQ A-H).
982 * When ACPI is disabled, only legacy IRQ numbers (IRQ0-15) are available, and
983 * some BIOSes may use MP Interrupt Source records to override IRQ numbers for
984 * PIRQs instead of reprogramming the interrupt routing logic. Thus there may be
985 * multiple pins sharing the same legacy IRQ number when ACPI is disabled.
986 */
987 static int alloc_isa_irq_from_domain(struct irq_domain *domain,
988 int irq, int ioapic, int pin,
989 struct irq_alloc_info *info)
990 {
991 struct mp_chip_data *data;
992 struct irq_data *irq_data = irq_get_irq_data(irq);
993 int node = ioapic_alloc_attr_node(info);
994
995 /*
996 * Legacy ISA IRQ has already been allocated, just add pin to
997 * the pin list assoicated with this IRQ and program the IOAPIC
998 * entry. The IOAPIC entry
999 */
1000 if (irq_data && irq_data->parent_data) {
1001 if (!mp_check_pin_attr(irq, info))
1002 return -EBUSY;
1003 if (__add_pin_to_irq_node(irq_data->chip_data, node, ioapic,
1004 info->ioapic.pin))
1005 return -ENOMEM;
1006 } else {
1007 info->flags |= X86_IRQ_ALLOC_LEGACY;
1008 irq = __irq_domain_alloc_irqs(domain, irq, 1, node, info, true,
1009 NULL);
1010 if (irq >= 0) {
1011 irq_data = irq_domain_get_irq_data(domain, irq);
1012 data = irq_data->chip_data;
1013 data->isa_irq = true;
1014 }
1015 }
1016
1017 return irq;
1018 }
1019
1020 static int mp_map_pin_to_irq(u32 gsi, int idx, int ioapic, int pin,
1021 unsigned int flags, struct irq_alloc_info *info)
1022 {
1023 int irq;
1024 bool legacy = false;
1025 struct irq_alloc_info tmp;
1026 struct mp_chip_data *data;
1027 struct irq_domain *domain = mp_ioapic_irqdomain(ioapic);
1028
1029 if (!domain)
1030 return -ENOSYS;
1031
1032 if (idx >= 0 && test_bit(mp_irqs[idx].srcbus, mp_bus_not_pci)) {
1033 irq = mp_irqs[idx].srcbusirq;
1034 legacy = mp_is_legacy_irq(irq);
1035 }
1036
1037 mutex_lock(&ioapic_mutex);
1038 if (!(flags & IOAPIC_MAP_ALLOC)) {
1039 if (!legacy) {
1040 irq = irq_find_mapping(domain, pin);
1041 if (irq == 0)
1042 irq = -ENOENT;
1043 }
1044 } else {
1045 ioapic_copy_alloc_attr(&tmp, info, gsi, ioapic, pin);
1046 if (legacy)
1047 irq = alloc_isa_irq_from_domain(domain, irq,
1048 ioapic, pin, &tmp);
1049 else if ((irq = irq_find_mapping(domain, pin)) == 0)
1050 irq = alloc_irq_from_domain(domain, ioapic, gsi, &tmp);
1051 else if (!mp_check_pin_attr(irq, &tmp))
1052 irq = -EBUSY;
1053 if (irq >= 0) {
1054 data = irq_get_chip_data(irq);
1055 data->count++;
1056 }
1057 }
1058 mutex_unlock(&ioapic_mutex);
1059
1060 return irq;
1061 }
1062
1063 static int pin_2_irq(int idx, int ioapic, int pin, unsigned int flags)
1064 {
1065 u32 gsi = mp_pin_to_gsi(ioapic, pin);
1066
1067 /*
1068 * Debugging check, we are in big trouble if this message pops up!
1069 */
1070 if (mp_irqs[idx].dstirq != pin)
1071 pr_err("broken BIOS or MPTABLE parser, ayiee!!\n");
1072
1073 #ifdef CONFIG_X86_32
1074 /*
1075 * PCI IRQ command line redirection. Yes, limits are hardcoded.
1076 */
1077 if ((pin >= 16) && (pin <= 23)) {
1078 if (pirq_entries[pin-16] != -1) {
1079 if (!pirq_entries[pin-16]) {
1080 apic_printk(APIC_VERBOSE, KERN_DEBUG
1081 "disabling PIRQ%d\n", pin-16);
1082 } else {
1083 int irq = pirq_entries[pin-16];
1084 apic_printk(APIC_VERBOSE, KERN_DEBUG
1085 "using PIRQ%d -> IRQ %d\n",
1086 pin-16, irq);
1087 return irq;
1088 }
1089 }
1090 }
1091 #endif
1092
1093 return mp_map_pin_to_irq(gsi, idx, ioapic, pin, flags, NULL);
1094 }
1095
1096 int mp_map_gsi_to_irq(u32 gsi, unsigned int flags, struct irq_alloc_info *info)
1097 {
1098 int ioapic, pin, idx;
1099
1100 ioapic = mp_find_ioapic(gsi);
1101 if (ioapic < 0)
1102 return -ENODEV;
1103
1104 pin = mp_find_ioapic_pin(ioapic, gsi);
1105 idx = find_irq_entry(ioapic, pin, mp_INT);
1106 if ((flags & IOAPIC_MAP_CHECK) && idx < 0)
1107 return -ENODEV;
1108
1109 return mp_map_pin_to_irq(gsi, idx, ioapic, pin, flags, info);
1110 }
1111
1112 void mp_unmap_irq(int irq)
1113 {
1114 struct irq_data *irq_data = irq_get_irq_data(irq);
1115 struct mp_chip_data *data;
1116
1117 if (!irq_data || !irq_data->domain)
1118 return;
1119
1120 data = irq_data->chip_data;
1121 if (!data || data->isa_irq)
1122 return;
1123
1124 mutex_lock(&ioapic_mutex);
1125 if (--data->count == 0)
1126 irq_domain_free_irqs(irq, 1);
1127 mutex_unlock(&ioapic_mutex);
1128 }
1129
1130 /*
1131 * Find a specific PCI IRQ entry.
1132 * Not an __init, possibly needed by modules
1133 */
1134 int IO_APIC_get_PCI_irq_vector(int bus, int slot, int pin)
1135 {
1136 int irq, i, best_ioapic = -1, best_idx = -1;
1137
1138 apic_printk(APIC_DEBUG,
1139 "querying PCI -> IRQ mapping bus:%d, slot:%d, pin:%d.\n",
1140 bus, slot, pin);
1141 if (test_bit(bus, mp_bus_not_pci)) {
1142 apic_printk(APIC_VERBOSE,
1143 "PCI BIOS passed nonexistent PCI bus %d!\n", bus);
1144 return -1;
1145 }
1146
1147 for (i = 0; i < mp_irq_entries; i++) {
1148 int lbus = mp_irqs[i].srcbus;
1149 int ioapic_idx, found = 0;
1150
1151 if (bus != lbus || mp_irqs[i].irqtype != mp_INT ||
1152 slot != ((mp_irqs[i].srcbusirq >> 2) & 0x1f))
1153 continue;
1154
1155 for_each_ioapic(ioapic_idx)
1156 if (mpc_ioapic_id(ioapic_idx) == mp_irqs[i].dstapic ||
1157 mp_irqs[i].dstapic == MP_APIC_ALL) {
1158 found = 1;
1159 break;
1160 }
1161 if (!found)
1162 continue;
1163
1164 /* Skip ISA IRQs */
1165 irq = pin_2_irq(i, ioapic_idx, mp_irqs[i].dstirq, 0);
1166 if (irq > 0 && !IO_APIC_IRQ(irq))
1167 continue;
1168
1169 if (pin == (mp_irqs[i].srcbusirq & 3)) {
1170 best_idx = i;
1171 best_ioapic = ioapic_idx;
1172 goto out;
1173 }
1174
1175 /*
1176 * Use the first all-but-pin matching entry as a
1177 * best-guess fuzzy result for broken mptables.
1178 */
1179 if (best_idx < 0) {
1180 best_idx = i;
1181 best_ioapic = ioapic_idx;
1182 }
1183 }
1184 if (best_idx < 0)
1185 return -1;
1186
1187 out:
1188 return pin_2_irq(best_idx, best_ioapic, mp_irqs[best_idx].dstirq,
1189 IOAPIC_MAP_ALLOC);
1190 }
1191 EXPORT_SYMBOL(IO_APIC_get_PCI_irq_vector);
1192
1193 static struct irq_chip ioapic_chip, ioapic_ir_chip;
1194
1195 static void __init setup_IO_APIC_irqs(void)
1196 {
1197 unsigned int ioapic, pin;
1198 int idx;
1199
1200 apic_printk(APIC_VERBOSE, KERN_DEBUG "init IO_APIC IRQs\n");
1201
1202 for_each_ioapic_pin(ioapic, pin) {
1203 idx = find_irq_entry(ioapic, pin, mp_INT);
1204 if (idx < 0)
1205 apic_printk(APIC_VERBOSE,
1206 KERN_DEBUG " apic %d pin %d not connected\n",
1207 mpc_ioapic_id(ioapic), pin);
1208 else
1209 pin_2_irq(idx, ioapic, pin,
1210 ioapic ? 0 : IOAPIC_MAP_ALLOC);
1211 }
1212 }
1213
1214 void ioapic_zap_locks(void)
1215 {
1216 raw_spin_lock_init(&ioapic_lock);
1217 }
1218
1219 static void io_apic_print_entries(unsigned int apic, unsigned int nr_entries)
1220 {
1221 struct IO_APIC_route_entry entry;
1222 char buf[256];
1223 int i;
1224
1225 printk(KERN_DEBUG "IOAPIC %d:\n", apic);
1226 for (i = 0; i <= nr_entries; i++) {
1227 entry = ioapic_read_entry(apic, i);
1228 snprintf(buf, sizeof(buf),
1229 " pin%02x, %s, %s, %s, V(%02X), IRR(%1d), S(%1d)",
1230 i,
1231 entry.masked ? "disabled" : "enabled ",
1232 entry.is_level ? "level" : "edge ",
1233 entry.active_low ? "low " : "high",
1234 entry.vector, entry.irr, entry.delivery_status);
1235 if (entry.ir_format) {
1236 printk(KERN_DEBUG "%s, remapped, I(%04X), Z(%X)\n",
1237 buf,
1238 (entry.ir_index_15 << 15) | entry.ir_index_0_14,
1239 entry.ir_zero);
1240 } else {
1241 printk(KERN_DEBUG "%s, %s, D(%02X%02X), M(%1d)\n", buf,
1242 entry.dest_mode_logical ? "logical " : "physical",
1243 entry.virt_destid_8_14, entry.destid_0_7,
1244 entry.delivery_mode);
1245 }
1246 }
1247 }
1248
1249 static void __init print_IO_APIC(int ioapic_idx)
1250 {
1251 union IO_APIC_reg_00 reg_00;
1252 union IO_APIC_reg_01 reg_01;
1253 union IO_APIC_reg_02 reg_02;
1254 union IO_APIC_reg_03 reg_03;
1255 unsigned long flags;
1256
1257 raw_spin_lock_irqsave(&ioapic_lock, flags);
1258 reg_00.raw = io_apic_read(ioapic_idx, 0);
1259 reg_01.raw = io_apic_read(ioapic_idx, 1);
1260 if (reg_01.bits.version >= 0x10)
1261 reg_02.raw = io_apic_read(ioapic_idx, 2);
1262 if (reg_01.bits.version >= 0x20)
1263 reg_03.raw = io_apic_read(ioapic_idx, 3);
1264 raw_spin_unlock_irqrestore(&ioapic_lock, flags);
1265
1266 printk(KERN_DEBUG "IO APIC #%d......\n", mpc_ioapic_id(ioapic_idx));
1267 printk(KERN_DEBUG ".... register #00: %08X\n", reg_00.raw);
1268 printk(KERN_DEBUG "....... : physical APIC id: %02X\n", reg_00.bits.ID);
1269 printk(KERN_DEBUG "....... : Delivery Type: %X\n", reg_00.bits.delivery_type);
1270 printk(KERN_DEBUG "....... : LTS : %X\n", reg_00.bits.LTS);
1271
1272 printk(KERN_DEBUG ".... register #01: %08X\n", *(int *)&reg_01);
1273 printk(KERN_DEBUG "....... : max redirection entries: %02X\n",
1274 reg_01.bits.entries);
1275
1276 printk(KERN_DEBUG "....... : PRQ implemented: %X\n", reg_01.bits.PRQ);
1277 printk(KERN_DEBUG "....... : IO APIC version: %02X\n",
1278 reg_01.bits.version);
1279
1280 /*
1281 * Some Intel chipsets with IO APIC VERSION of 0x1? don't have reg_02,
1282 * but the value of reg_02 is read as the previous read register
1283 * value, so ignore it if reg_02 == reg_01.
1284 */
1285 if (reg_01.bits.version >= 0x10 && reg_02.raw != reg_01.raw) {
1286 printk(KERN_DEBUG ".... register #02: %08X\n", reg_02.raw);
1287 printk(KERN_DEBUG "....... : arbitration: %02X\n", reg_02.bits.arbitration);
1288 }
1289
1290 /*
1291 * Some Intel chipsets with IO APIC VERSION of 0x2? don't have reg_02
1292 * or reg_03, but the value of reg_0[23] is read as the previous read
1293 * register value, so ignore it if reg_03 == reg_0[12].
1294 */
1295 if (reg_01.bits.version >= 0x20 && reg_03.raw != reg_02.raw &&
1296 reg_03.raw != reg_01.raw) {
1297 printk(KERN_DEBUG ".... register #03: %08X\n", reg_03.raw);
1298 printk(KERN_DEBUG "....... : Boot DT : %X\n", reg_03.bits.boot_DT);
1299 }
1300
1301 printk(KERN_DEBUG ".... IRQ redirection table:\n");
1302 io_apic_print_entries(ioapic_idx, reg_01.bits.entries);
1303 }
1304
1305 void __init print_IO_APICs(void)
1306 {
1307 int ioapic_idx;
1308 unsigned int irq;
1309
1310 printk(KERN_DEBUG "number of MP IRQ sources: %d.\n", mp_irq_entries);
1311 for_each_ioapic(ioapic_idx)
1312 printk(KERN_DEBUG "number of IO-APIC #%d registers: %d.\n",
1313 mpc_ioapic_id(ioapic_idx),
1314 ioapics[ioapic_idx].nr_registers);
1315
1316 /*
1317 * We are a bit conservative about what we expect. We have to
1318 * know about every hardware change ASAP.
1319 */
1320 printk(KERN_INFO "testing the IO APIC.......................\n");
1321
1322 for_each_ioapic(ioapic_idx)
1323 print_IO_APIC(ioapic_idx);
1324
1325 printk(KERN_DEBUG "IRQ to pin mappings:\n");
1326 for_each_active_irq(irq) {
1327 struct irq_pin_list *entry;
1328 struct irq_chip *chip;
1329 struct mp_chip_data *data;
1330
1331 chip = irq_get_chip(irq);
1332 if (chip != &ioapic_chip && chip != &ioapic_ir_chip)
1333 continue;
1334 data = irq_get_chip_data(irq);
1335 if (!data)
1336 continue;
1337 if (list_empty(&data->irq_2_pin))
1338 continue;
1339
1340 printk(KERN_DEBUG "IRQ%d ", irq);
1341 for_each_irq_pin(entry, data->irq_2_pin)
1342 pr_cont("-> %d:%d", entry->apic, entry->pin);
1343 pr_cont("\n");
1344 }
1345
1346 printk(KERN_INFO ".................................... done.\n");
1347 }
1348
1349 /* Where if anywhere is the i8259 connect in external int mode */
1350 static struct { int pin, apic; } ioapic_i8259 = { -1, -1 };
1351
1352 void __init enable_IO_APIC(void)
1353 {
1354 int i8259_apic, i8259_pin;
1355 int apic, pin;
1356
1357 if (skip_ioapic_setup)
1358 nr_ioapics = 0;
1359
1360 if (!nr_legacy_irqs() || !nr_ioapics)
1361 return;
1362
1363 for_each_ioapic_pin(apic, pin) {
1364 /* See if any of the pins is in ExtINT mode */
1365 struct IO_APIC_route_entry entry = ioapic_read_entry(apic, pin);
1366
1367 /* If the interrupt line is enabled and in ExtInt mode
1368 * I have found the pin where the i8259 is connected.
1369 */
1370 if (!entry.masked &&
1371 entry.delivery_mode == APIC_DELIVERY_MODE_EXTINT) {
1372 ioapic_i8259.apic = apic;
1373 ioapic_i8259.pin = pin;
1374 goto found_i8259;
1375 }
1376 }
1377 found_i8259:
1378 /* Look to see what if the MP table has reported the ExtINT */
1379 /* If we could not find the appropriate pin by looking at the ioapic
1380 * the i8259 probably is not connected the ioapic but give the
1381 * mptable a chance anyway.
1382 */
1383 i8259_pin = find_isa_irq_pin(0, mp_ExtINT);
1384 i8259_apic = find_isa_irq_apic(0, mp_ExtINT);
1385 /* Trust the MP table if nothing is setup in the hardware */
1386 if ((ioapic_i8259.pin == -1) && (i8259_pin >= 0)) {
1387 printk(KERN_WARNING "ExtINT not setup in hardware but reported by MP table\n");
1388 ioapic_i8259.pin = i8259_pin;
1389 ioapic_i8259.apic = i8259_apic;
1390 }
1391 /* Complain if the MP table and the hardware disagree */
1392 if (((ioapic_i8259.apic != i8259_apic) || (ioapic_i8259.pin != i8259_pin)) &&
1393 (i8259_pin >= 0) && (ioapic_i8259.pin >= 0))
1394 {
1395 printk(KERN_WARNING "ExtINT in hardware and MP table differ\n");
1396 }
1397
1398 /*
1399 * Do not trust the IO-APIC being empty at bootup
1400 */
1401 clear_IO_APIC();
1402 }
1403
1404 void native_restore_boot_irq_mode(void)
1405 {
1406 /*
1407 * If the i8259 is routed through an IOAPIC
1408 * Put that IOAPIC in virtual wire mode
1409 * so legacy interrupts can be delivered.
1410 */
1411 if (ioapic_i8259.pin != -1) {
1412 struct IO_APIC_route_entry entry;
1413 u32 apic_id = read_apic_id();
1414
1415 memset(&entry, 0, sizeof(entry));
1416 entry.masked = false;
1417 entry.is_level = false;
1418 entry.active_low = false;
1419 entry.dest_mode_logical = false;
1420 entry.delivery_mode = APIC_DELIVERY_MODE_EXTINT;
1421 entry.destid_0_7 = apic_id & 0xFF;
1422 entry.virt_destid_8_14 = apic_id >> 8;
1423
1424 /*
1425 * Add it to the IO-APIC irq-routing table:
1426 */
1427 ioapic_write_entry(ioapic_i8259.apic, ioapic_i8259.pin, entry);
1428 }
1429
1430 if (boot_cpu_has(X86_FEATURE_APIC) || apic_from_smp_config())
1431 disconnect_bsp_APIC(ioapic_i8259.pin != -1);
1432 }
1433
1434 void restore_boot_irq_mode(void)
1435 {
1436 if (!nr_legacy_irqs())
1437 return;
1438
1439 x86_apic_ops.restore();
1440 }
1441
1442 #ifdef CONFIG_X86_32
1443 /*
1444 * function to set the IO-APIC physical IDs based on the
1445 * values stored in the MPC table.
1446 *
1447 * by Matt Domsch <Matt_Domsch@dell.com> Tue Dec 21 12:25:05 CST 1999
1448 */
1449 void __init setup_ioapic_ids_from_mpc_nocheck(void)
1450 {
1451 union IO_APIC_reg_00 reg_00;
1452 physid_mask_t phys_id_present_map;
1453 int ioapic_idx;
1454 int i;
1455 unsigned char old_id;
1456 unsigned long flags;
1457
1458 /*
1459 * This is broken; anything with a real cpu count has to
1460 * circumvent this idiocy regardless.
1461 */
1462 apic->ioapic_phys_id_map(&phys_cpu_present_map, &phys_id_present_map);
1463
1464 /*
1465 * Set the IOAPIC ID to the value stored in the MPC table.
1466 */
1467 for_each_ioapic(ioapic_idx) {
1468 /* Read the register 0 value */
1469 raw_spin_lock_irqsave(&ioapic_lock, flags);
1470 reg_00.raw = io_apic_read(ioapic_idx, 0);
1471 raw_spin_unlock_irqrestore(&ioapic_lock, flags);
1472
1473 old_id = mpc_ioapic_id(ioapic_idx);
1474
1475 if (mpc_ioapic_id(ioapic_idx) >= get_physical_broadcast()) {
1476 printk(KERN_ERR "BIOS bug, IO-APIC#%d ID is %d in the MPC table!...\n",
1477 ioapic_idx, mpc_ioapic_id(ioapic_idx));
1478 printk(KERN_ERR "... fixing up to %d. (tell your hw vendor)\n",
1479 reg_00.bits.ID);
1480 ioapics[ioapic_idx].mp_config.apicid = reg_00.bits.ID;
1481 }
1482
1483 /*
1484 * Sanity check, is the ID really free? Every APIC in a
1485 * system must have a unique ID or we get lots of nice
1486 * 'stuck on smp_invalidate_needed IPI wait' messages.
1487 */
1488 if (apic->check_apicid_used(&phys_id_present_map,
1489 mpc_ioapic_id(ioapic_idx))) {
1490 printk(KERN_ERR "BIOS bug, IO-APIC#%d ID %d is already used!...\n",
1491 ioapic_idx, mpc_ioapic_id(ioapic_idx));
1492 for (i = 0; i < get_physical_broadcast(); i++)
1493 if (!physid_isset(i, phys_id_present_map))
1494 break;
1495 if (i >= get_physical_broadcast())
1496 panic("Max APIC ID exceeded!\n");
1497 printk(KERN_ERR "... fixing up to %d. (tell your hw vendor)\n",
1498 i);
1499 physid_set(i, phys_id_present_map);
1500 ioapics[ioapic_idx].mp_config.apicid = i;
1501 } else {
1502 physid_mask_t tmp;
1503 apic->apicid_to_cpu_present(mpc_ioapic_id(ioapic_idx),
1504 &tmp);
1505 apic_printk(APIC_VERBOSE, "Setting %d in the "
1506 "phys_id_present_map\n",
1507 mpc_ioapic_id(ioapic_idx));
1508 physids_or(phys_id_present_map, phys_id_present_map, tmp);
1509 }
1510
1511 /*
1512 * We need to adjust the IRQ routing table
1513 * if the ID changed.
1514 */
1515 if (old_id != mpc_ioapic_id(ioapic_idx))
1516 for (i = 0; i < mp_irq_entries; i++)
1517 if (mp_irqs[i].dstapic == old_id)
1518 mp_irqs[i].dstapic
1519 = mpc_ioapic_id(ioapic_idx);
1520
1521 /*
1522 * Update the ID register according to the right value
1523 * from the MPC table if they are different.
1524 */
1525 if (mpc_ioapic_id(ioapic_idx) == reg_00.bits.ID)
1526 continue;
1527
1528 apic_printk(APIC_VERBOSE, KERN_INFO
1529 "...changing IO-APIC physical APIC ID to %d ...",
1530 mpc_ioapic_id(ioapic_idx));
1531
1532 reg_00.bits.ID = mpc_ioapic_id(ioapic_idx);
1533 raw_spin_lock_irqsave(&ioapic_lock, flags);
1534 io_apic_write(ioapic_idx, 0, reg_00.raw);
1535 raw_spin_unlock_irqrestore(&ioapic_lock, flags);
1536
1537 /*
1538 * Sanity check
1539 */
1540 raw_spin_lock_irqsave(&ioapic_lock, flags);
1541 reg_00.raw = io_apic_read(ioapic_idx, 0);
1542 raw_spin_unlock_irqrestore(&ioapic_lock, flags);
1543 if (reg_00.bits.ID != mpc_ioapic_id(ioapic_idx))
1544 pr_cont("could not set ID!\n");
1545 else
1546 apic_printk(APIC_VERBOSE, " ok.\n");
1547 }
1548 }
1549
1550 void __init setup_ioapic_ids_from_mpc(void)
1551 {
1552
1553 if (acpi_ioapic)
1554 return;
1555 /*
1556 * Don't check I/O APIC IDs for xAPIC systems. They have
1557 * no meaning without the serial APIC bus.
1558 */
1559 if (!(boot_cpu_data.x86_vendor == X86_VENDOR_INTEL)
1560 || APIC_XAPIC(boot_cpu_apic_version))
1561 return;
1562 setup_ioapic_ids_from_mpc_nocheck();
1563 }
1564 #endif
1565
1566 int no_timer_check __initdata;
1567
1568 static int __init notimercheck(char *s)
1569 {
1570 no_timer_check = 1;
1571 return 1;
1572 }
1573 __setup("no_timer_check", notimercheck);
1574
1575 static void __init delay_with_tsc(void)
1576 {
1577 unsigned long long start, now;
1578 unsigned long end = jiffies + 4;
1579
1580 start = rdtsc();
1581
1582 /*
1583 * We don't know the TSC frequency yet, but waiting for
1584 * 40000000000/HZ TSC cycles is safe:
1585 * 4 GHz == 10 jiffies
1586 * 1 GHz == 40 jiffies
1587 */
1588 do {
1589 rep_nop();
1590 now = rdtsc();
1591 } while ((now - start) < 40000000000ULL / HZ &&
1592 time_before_eq(jiffies, end));
1593 }
1594
1595 static void __init delay_without_tsc(void)
1596 {
1597 unsigned long end = jiffies + 4;
1598 int band = 1;
1599
1600 /*
1601 * We don't know any frequency yet, but waiting for
1602 * 40940000000/HZ cycles is safe:
1603 * 4 GHz == 10 jiffies
1604 * 1 GHz == 40 jiffies
1605 * 1 << 1 + 1 << 2 +...+ 1 << 11 = 4094
1606 */
1607 do {
1608 __delay(((1U << band++) * 10000000UL) / HZ);
1609 } while (band < 12 && time_before_eq(jiffies, end));
1610 }
1611
1612 /*
1613 * There is a nasty bug in some older SMP boards, their mptable lies
1614 * about the timer IRQ. We do the following to work around the situation:
1615 *
1616 * - timer IRQ defaults to IO-APIC IRQ
1617 * - if this function detects that timer IRQs are defunct, then we fall
1618 * back to ISA timer IRQs
1619 */
1620 static int __init timer_irq_works(void)
1621 {
1622 unsigned long t1 = jiffies;
1623
1624 if (no_timer_check)
1625 return 1;
1626
1627 local_irq_enable();
1628 if (boot_cpu_has(X86_FEATURE_TSC))
1629 delay_with_tsc();
1630 else
1631 delay_without_tsc();
1632
1633 /*
1634 * Expect a few ticks at least, to be sure some possible
1635 * glue logic does not lock up after one or two first
1636 * ticks in a non-ExtINT mode. Also the local APIC
1637 * might have cached one ExtINT interrupt. Finally, at
1638 * least one tick may be lost due to delays.
1639 */
1640
1641 local_irq_disable();
1642
1643 /* Did jiffies advance? */
1644 return time_after(jiffies, t1 + 4);
1645 }
1646
1647 /*
1648 * In the SMP+IOAPIC case it might happen that there are an unspecified
1649 * number of pending IRQ events unhandled. These cases are very rare,
1650 * so we 'resend' these IRQs via IPIs, to the same CPU. It's much
1651 * better to do it this way as thus we do not have to be aware of
1652 * 'pending' interrupts in the IRQ path, except at this point.
1653 */
1654 /*
1655 * Edge triggered needs to resend any interrupt
1656 * that was delayed but this is now handled in the device
1657 * independent code.
1658 */
1659
1660 /*
1661 * Starting up a edge-triggered IO-APIC interrupt is
1662 * nasty - we need to make sure that we get the edge.
1663 * If it is already asserted for some reason, we need
1664 * return 1 to indicate that is was pending.
1665 *
1666 * This is not complete - we should be able to fake
1667 * an edge even if it isn't on the 8259A...
1668 */
1669 static unsigned int startup_ioapic_irq(struct irq_data *data)
1670 {
1671 int was_pending = 0, irq = data->irq;
1672 unsigned long flags;
1673
1674 raw_spin_lock_irqsave(&ioapic_lock, flags);
1675 if (irq < nr_legacy_irqs()) {
1676 legacy_pic->mask(irq);
1677 if (legacy_pic->irq_pending(irq))
1678 was_pending = 1;
1679 }
1680 __unmask_ioapic(data->chip_data);
1681 raw_spin_unlock_irqrestore(&ioapic_lock, flags);
1682
1683 return was_pending;
1684 }
1685
1686 atomic_t irq_mis_count;
1687
1688 #ifdef CONFIG_GENERIC_PENDING_IRQ
1689 static bool io_apic_level_ack_pending(struct mp_chip_data *data)
1690 {
1691 struct irq_pin_list *entry;
1692 unsigned long flags;
1693
1694 raw_spin_lock_irqsave(&ioapic_lock, flags);
1695 for_each_irq_pin(entry, data->irq_2_pin) {
1696 struct IO_APIC_route_entry e;
1697 int pin;
1698
1699 pin = entry->pin;
1700 e.w1 = io_apic_read(entry->apic, 0x10 + pin*2);
1701 /* Is the remote IRR bit set? */
1702 if (e.irr) {
1703 raw_spin_unlock_irqrestore(&ioapic_lock, flags);
1704 return true;
1705 }
1706 }
1707 raw_spin_unlock_irqrestore(&ioapic_lock, flags);
1708
1709 return false;
1710 }
1711
1712 static inline bool ioapic_prepare_move(struct irq_data *data)
1713 {
1714 /* If we are moving the IRQ we need to mask it */
1715 if (unlikely(irqd_is_setaffinity_pending(data))) {
1716 if (!irqd_irq_masked(data))
1717 mask_ioapic_irq(data);
1718 return true;
1719 }
1720 return false;
1721 }
1722
1723 static inline void ioapic_finish_move(struct irq_data *data, bool moveit)
1724 {
1725 if (unlikely(moveit)) {
1726 /* Only migrate the irq if the ack has been received.
1727 *
1728 * On rare occasions the broadcast level triggered ack gets
1729 * delayed going to ioapics, and if we reprogram the
1730 * vector while Remote IRR is still set the irq will never
1731 * fire again.
1732 *
1733 * To prevent this scenario we read the Remote IRR bit
1734 * of the ioapic. This has two effects.
1735 * - On any sane system the read of the ioapic will
1736 * flush writes (and acks) going to the ioapic from
1737 * this cpu.
1738 * - We get to see if the ACK has actually been delivered.
1739 *
1740 * Based on failed experiments of reprogramming the
1741 * ioapic entry from outside of irq context starting
1742 * with masking the ioapic entry and then polling until
1743 * Remote IRR was clear before reprogramming the
1744 * ioapic I don't trust the Remote IRR bit to be
1745 * completey accurate.
1746 *
1747 * However there appears to be no other way to plug
1748 * this race, so if the Remote IRR bit is not
1749 * accurate and is causing problems then it is a hardware bug
1750 * and you can go talk to the chipset vendor about it.
1751 */
1752 if (!io_apic_level_ack_pending(data->chip_data))
1753 irq_move_masked_irq(data);
1754 /* If the IRQ is masked in the core, leave it: */
1755 if (!irqd_irq_masked(data))
1756 unmask_ioapic_irq(data);
1757 }
1758 }
1759 #else
1760 static inline bool ioapic_prepare_move(struct irq_data *data)
1761 {
1762 return false;
1763 }
1764 static inline void ioapic_finish_move(struct irq_data *data, bool moveit)
1765 {
1766 }
1767 #endif
1768
1769 static void ioapic_ack_level(struct irq_data *irq_data)
1770 {
1771 struct irq_cfg *cfg = irqd_cfg(irq_data);
1772 unsigned long v;
1773 bool moveit;
1774 int i;
1775
1776 irq_complete_move(cfg);
1777 moveit = ioapic_prepare_move(irq_data);
1778
1779 /*
1780 * It appears there is an erratum which affects at least version 0x11
1781 * of I/O APIC (that's the 82093AA and cores integrated into various
1782 * chipsets). Under certain conditions a level-triggered interrupt is
1783 * erroneously delivered as edge-triggered one but the respective IRR
1784 * bit gets set nevertheless. As a result the I/O unit expects an EOI
1785 * message but it will never arrive and further interrupts are blocked
1786 * from the source. The exact reason is so far unknown, but the
1787 * phenomenon was observed when two consecutive interrupt requests
1788 * from a given source get delivered to the same CPU and the source is
1789 * temporarily disabled in between.
1790 *
1791 * A workaround is to simulate an EOI message manually. We achieve it
1792 * by setting the trigger mode to edge and then to level when the edge
1793 * trigger mode gets detected in the TMR of a local APIC for a
1794 * level-triggered interrupt. We mask the source for the time of the
1795 * operation to prevent an edge-triggered interrupt escaping meanwhile.
1796 * The idea is from Manfred Spraul. --macro
1797 *
1798 * Also in the case when cpu goes offline, fixup_irqs() will forward
1799 * any unhandled interrupt on the offlined cpu to the new cpu
1800 * destination that is handling the corresponding interrupt. This
1801 * interrupt forwarding is done via IPI's. Hence, in this case also
1802 * level-triggered io-apic interrupt will be seen as an edge
1803 * interrupt in the IRR. And we can't rely on the cpu's EOI
1804 * to be broadcasted to the IO-APIC's which will clear the remoteIRR
1805 * corresponding to the level-triggered interrupt. Hence on IO-APIC's
1806 * supporting EOI register, we do an explicit EOI to clear the
1807 * remote IRR and on IO-APIC's which don't have an EOI register,
1808 * we use the above logic (mask+edge followed by unmask+level) from
1809 * Manfred Spraul to clear the remote IRR.
1810 */
1811 i = cfg->vector;
1812 v = apic_read(APIC_TMR + ((i & ~0x1f) >> 1));
1813
1814 /*
1815 * We must acknowledge the irq before we move it or the acknowledge will
1816 * not propagate properly.
1817 */
1818 ack_APIC_irq();
1819
1820 /*
1821 * Tail end of clearing remote IRR bit (either by delivering the EOI
1822 * message via io-apic EOI register write or simulating it using
1823 * mask+edge followed by unnask+level logic) manually when the
1824 * level triggered interrupt is seen as the edge triggered interrupt
1825 * at the cpu.
1826 */
1827 if (!(v & (1 << (i & 0x1f)))) {
1828 atomic_inc(&irq_mis_count);
1829 eoi_ioapic_pin(cfg->vector, irq_data->chip_data);
1830 }
1831
1832 ioapic_finish_move(irq_data, moveit);
1833 }
1834
1835 static void ioapic_ir_ack_level(struct irq_data *irq_data)
1836 {
1837 struct mp_chip_data *data = irq_data->chip_data;
1838
1839 /*
1840 * Intr-remapping uses pin number as the virtual vector
1841 * in the RTE. Actual vector is programmed in
1842 * intr-remapping table entry. Hence for the io-apic
1843 * EOI we use the pin number.
1844 */
1845 apic_ack_irq(irq_data);
1846 eoi_ioapic_pin(data->entry.vector, data);
1847 }
1848
1849 /*
1850 * The I/OAPIC is just a device for generating MSI messages from legacy
1851 * interrupt pins. Various fields of the RTE translate into bits of the
1852 * resulting MSI which had a historical meaning.
1853 *
1854 * With interrupt remapping, many of those bits have different meanings
1855 * in the underlying MSI, but the way that the I/OAPIC transforms them
1856 * from its RTE to the MSI message is the same. This function allows
1857 * the parent IRQ domain to compose the MSI message, then takes the
1858 * relevant bits to put them in the appropriate places in the RTE in
1859 * order to generate that message when the IRQ happens.
1860 *
1861 * The setup here relies on a preconfigured route entry (is_level,
1862 * active_low, masked) because the parent domain is merely composing the
1863 * generic message routing information which is used for the MSI.
1864 */
1865 static void ioapic_setup_msg_from_msi(struct irq_data *irq_data,
1866 struct IO_APIC_route_entry *entry)
1867 {
1868 struct msi_msg msg;
1869
1870 /* Let the parent domain compose the MSI message */
1871 irq_chip_compose_msi_msg(irq_data, &msg);
1872
1873 /*
1874 * - Real vector
1875 * - DMAR/IR: 8bit subhandle (ioapic.pin)
1876 * - AMD/IR: 8bit IRTE index
1877 */
1878 entry->vector = msg.arch_data.vector;
1879 /* Delivery mode (for DMAR/IR all 0) */
1880 entry->delivery_mode = msg.arch_data.delivery_mode;
1881 /* Destination mode or DMAR/IR index bit 15 */
1882 entry->dest_mode_logical = msg.arch_addr_lo.dest_mode_logical;
1883 /* DMAR/IR: 1, 0 for all other modes */
1884 entry->ir_format = msg.arch_addr_lo.dmar_format;
1885 /*
1886 * - DMAR/IR: index bit 0-14.
1887 *
1888 * - Virt: If the host supports x2apic without a virtualized IR
1889 * unit then bit 0-6 of dmar_index_0_14 are providing bit
1890 * 8-14 of the destination id.
1891 *
1892 * All other modes have bit 0-6 of dmar_index_0_14 cleared and the
1893 * topmost 8 bits are destination id bit 0-7 (entry::destid_0_7).
1894 */
1895 entry->ir_index_0_14 = msg.arch_addr_lo.dmar_index_0_14;
1896 }
1897
1898 static void ioapic_configure_entry(struct irq_data *irqd)
1899 {
1900 struct mp_chip_data *mpd = irqd->chip_data;
1901 struct irq_pin_list *entry;
1902
1903 ioapic_setup_msg_from_msi(irqd, &mpd->entry);
1904
1905 for_each_irq_pin(entry, mpd->irq_2_pin)
1906 __ioapic_write_entry(entry->apic, entry->pin, mpd->entry);
1907 }
1908
1909 static int ioapic_set_affinity(struct irq_data *irq_data,
1910 const struct cpumask *mask, bool force)
1911 {
1912 struct irq_data *parent = irq_data->parent_data;
1913 unsigned long flags;
1914 int ret;
1915
1916 ret = parent->chip->irq_set_affinity(parent, mask, force);
1917 raw_spin_lock_irqsave(&ioapic_lock, flags);
1918 if (ret >= 0 && ret != IRQ_SET_MASK_OK_DONE)
1919 ioapic_configure_entry(irq_data);
1920 raw_spin_unlock_irqrestore(&ioapic_lock, flags);
1921
1922 return ret;
1923 }
1924
1925 /*
1926 * Interrupt shutdown masks the ioapic pin, but the interrupt might already
1927 * be in flight, but not yet serviced by the target CPU. That means
1928 * __synchronize_hardirq() would return and claim that everything is calmed
1929 * down. So free_irq() would proceed and deactivate the interrupt and free
1930 * resources.
1931 *
1932 * Once the target CPU comes around to service it it will find a cleared
1933 * vector and complain. While the spurious interrupt is harmless, the full
1934 * release of resources might prevent the interrupt from being acknowledged
1935 * which keeps the hardware in a weird state.
1936 *
1937 * Verify that the corresponding Remote-IRR bits are clear.
1938 */
1939 static int ioapic_irq_get_chip_state(struct irq_data *irqd,
1940 enum irqchip_irq_state which,
1941 bool *state)
1942 {
1943 struct mp_chip_data *mcd = irqd->chip_data;
1944 struct IO_APIC_route_entry rentry;
1945 struct irq_pin_list *p;
1946
1947 if (which != IRQCHIP_STATE_ACTIVE)
1948 return -EINVAL;
1949
1950 *state = false;
1951 raw_spin_lock(&ioapic_lock);
1952 for_each_irq_pin(p, mcd->irq_2_pin) {
1953 rentry = __ioapic_read_entry(p->apic, p->pin);
1954 /*
1955 * The remote IRR is only valid in level trigger mode. It's
1956 * meaning is undefined for edge triggered interrupts and
1957 * irrelevant because the IO-APIC treats them as fire and
1958 * forget.
1959 */
1960 if (rentry.irr && rentry.is_level) {
1961 *state = true;
1962 break;
1963 }
1964 }
1965 raw_spin_unlock(&ioapic_lock);
1966 return 0;
1967 }
1968
1969 static struct irq_chip ioapic_chip __read_mostly = {
1970 .name = "IO-APIC",
1971 .irq_startup = startup_ioapic_irq,
1972 .irq_mask = mask_ioapic_irq,
1973 .irq_unmask = unmask_ioapic_irq,
1974 .irq_ack = irq_chip_ack_parent,
1975 .irq_eoi = ioapic_ack_level,
1976 .irq_set_affinity = ioapic_set_affinity,
1977 .irq_retrigger = irq_chip_retrigger_hierarchy,
1978 .irq_get_irqchip_state = ioapic_irq_get_chip_state,
1979 .flags = IRQCHIP_SKIP_SET_WAKE,
1980 };
1981
1982 static struct irq_chip ioapic_ir_chip __read_mostly = {
1983 .name = "IR-IO-APIC",
1984 .irq_startup = startup_ioapic_irq,
1985 .irq_mask = mask_ioapic_irq,
1986 .irq_unmask = unmask_ioapic_irq,
1987 .irq_ack = irq_chip_ack_parent,
1988 .irq_eoi = ioapic_ir_ack_level,
1989 .irq_set_affinity = ioapic_set_affinity,
1990 .irq_retrigger = irq_chip_retrigger_hierarchy,
1991 .irq_get_irqchip_state = ioapic_irq_get_chip_state,
1992 .flags = IRQCHIP_SKIP_SET_WAKE,
1993 };
1994
1995 static inline void init_IO_APIC_traps(void)
1996 {
1997 struct irq_cfg *cfg;
1998 unsigned int irq;
1999
2000 for_each_active_irq(irq) {
2001 cfg = irq_cfg(irq);
2002 if (IO_APIC_IRQ(irq) && cfg && !cfg->vector) {
2003 /*
2004 * Hmm.. We don't have an entry for this,
2005 * so default to an old-fashioned 8259
2006 * interrupt if we can..
2007 */
2008 if (irq < nr_legacy_irqs())
2009 legacy_pic->make_irq(irq);
2010 else
2011 /* Strange. Oh, well.. */
2012 irq_set_chip(irq, &no_irq_chip);
2013 }
2014 }
2015 }
2016
2017 /*
2018 * The local APIC irq-chip implementation:
2019 */
2020
2021 static void mask_lapic_irq(struct irq_data *data)
2022 {
2023 unsigned long v;
2024
2025 v = apic_read(APIC_LVT0);
2026 apic_write(APIC_LVT0, v | APIC_LVT_MASKED);
2027 }
2028
2029 static void unmask_lapic_irq(struct irq_data *data)
2030 {
2031 unsigned long v;
2032
2033 v = apic_read(APIC_LVT0);
2034 apic_write(APIC_LVT0, v & ~APIC_LVT_MASKED);
2035 }
2036
2037 static void ack_lapic_irq(struct irq_data *data)
2038 {
2039 ack_APIC_irq();
2040 }
2041
2042 static struct irq_chip lapic_chip __read_mostly = {
2043 .name = "local-APIC",
2044 .irq_mask = mask_lapic_irq,
2045 .irq_unmask = unmask_lapic_irq,
2046 .irq_ack = ack_lapic_irq,
2047 };
2048
2049 static void lapic_register_intr(int irq)
2050 {
2051 irq_clear_status_flags(irq, IRQ_LEVEL);
2052 irq_set_chip_and_handler_name(irq, &lapic_chip, handle_edge_irq,
2053 "edge");
2054 }
2055
2056 /*
2057 * This looks a bit hackish but it's about the only one way of sending
2058 * a few INTA cycles to 8259As and any associated glue logic. ICR does
2059 * not support the ExtINT mode, unfortunately. We need to send these
2060 * cycles as some i82489DX-based boards have glue logic that keeps the
2061 * 8259A interrupt line asserted until INTA. --macro
2062 */
2063 static inline void __init unlock_ExtINT_logic(void)
2064 {
2065 int apic, pin, i;
2066 struct IO_APIC_route_entry entry0, entry1;
2067 unsigned char save_control, save_freq_select;
2068 u32 apic_id;
2069
2070 pin = find_isa_irq_pin(8, mp_INT);
2071 if (pin == -1) {
2072 WARN_ON_ONCE(1);
2073 return;
2074 }
2075 apic = find_isa_irq_apic(8, mp_INT);
2076 if (apic == -1) {
2077 WARN_ON_ONCE(1);
2078 return;
2079 }
2080
2081 entry0 = ioapic_read_entry(apic, pin);
2082 clear_IO_APIC_pin(apic, pin);
2083
2084 apic_id = hard_smp_processor_id();
2085 memset(&entry1, 0, sizeof(entry1));
2086
2087 entry1.dest_mode_logical = true;
2088 entry1.masked = false;
2089 entry1.destid_0_7 = apic_id & 0xFF;
2090 entry1.virt_destid_8_14 = apic_id >> 8;
2091 entry1.delivery_mode = APIC_DELIVERY_MODE_EXTINT;
2092 entry1.active_low = entry0.active_low;
2093 entry1.is_level = false;
2094 entry1.vector = 0;
2095
2096 ioapic_write_entry(apic, pin, entry1);
2097
2098 save_control = CMOS_READ(RTC_CONTROL);
2099 save_freq_select = CMOS_READ(RTC_FREQ_SELECT);
2100 CMOS_WRITE((save_freq_select & ~RTC_RATE_SELECT) | 0x6,
2101 RTC_FREQ_SELECT);
2102 CMOS_WRITE(save_control | RTC_PIE, RTC_CONTROL);
2103
2104 i = 100;
2105 while (i-- > 0) {
2106 mdelay(10);
2107 if ((CMOS_READ(RTC_INTR_FLAGS) & RTC_PF) == RTC_PF)
2108 i -= 10;
2109 }
2110
2111 CMOS_WRITE(save_control, RTC_CONTROL);
2112 CMOS_WRITE(save_freq_select, RTC_FREQ_SELECT);
2113 clear_IO_APIC_pin(apic, pin);
2114
2115 ioapic_write_entry(apic, pin, entry0);
2116 }
2117
2118 static int disable_timer_pin_1 __initdata;
2119 /* Actually the next is obsolete, but keep it for paranoid reasons -AK */
2120 static int __init disable_timer_pin_setup(char *arg)
2121 {
2122 disable_timer_pin_1 = 1;
2123 return 0;
2124 }
2125 early_param("disable_timer_pin_1", disable_timer_pin_setup);
2126
2127 static int mp_alloc_timer_irq(int ioapic, int pin)
2128 {
2129 int irq = -1;
2130 struct irq_domain *domain = mp_ioapic_irqdomain(ioapic);
2131
2132 if (domain) {
2133 struct irq_alloc_info info;
2134
2135 ioapic_set_alloc_attr(&info, NUMA_NO_NODE, 0, 0);
2136 info.devid = mpc_ioapic_id(ioapic);
2137 info.ioapic.pin = pin;
2138 mutex_lock(&ioapic_mutex);
2139 irq = alloc_isa_irq_from_domain(domain, 0, ioapic, pin, &info);
2140 mutex_unlock(&ioapic_mutex);
2141 }
2142
2143 return irq;
2144 }
2145
2146 /*
2147 * This code may look a bit paranoid, but it's supposed to cooperate with
2148 * a wide range of boards and BIOS bugs. Fortunately only the timer IRQ
2149 * is so screwy. Thanks to Brian Perkins for testing/hacking this beast
2150 * fanatically on his truly buggy board.
2151 *
2152 * FIXME: really need to revamp this for all platforms.
2153 */
2154 static inline void __init check_timer(void)
2155 {
2156 struct irq_data *irq_data = irq_get_irq_data(0);
2157 struct mp_chip_data *data = irq_data->chip_data;
2158 struct irq_cfg *cfg = irqd_cfg(irq_data);
2159 int node = cpu_to_node(0);
2160 int apic1, pin1, apic2, pin2;
2161 int no_pin1 = 0;
2162
2163 if (!global_clock_event)
2164 return;
2165
2166 local_irq_disable();
2167
2168 /*
2169 * get/set the timer IRQ vector:
2170 */
2171 legacy_pic->mask(0);
2172
2173 /*
2174 * As IRQ0 is to be enabled in the 8259A, the virtual
2175 * wire has to be disabled in the local APIC. Also
2176 * timer interrupts need to be acknowledged manually in
2177 * the 8259A for the i82489DX when using the NMI
2178 * watchdog as that APIC treats NMIs as level-triggered.
2179 * The AEOI mode will finish them in the 8259A
2180 * automatically.
2181 */
2182 apic_write(APIC_LVT0, APIC_LVT_MASKED | APIC_DM_EXTINT);
2183 legacy_pic->init(1);
2184
2185 pin1 = find_isa_irq_pin(0, mp_INT);
2186 apic1 = find_isa_irq_apic(0, mp_INT);
2187 pin2 = ioapic_i8259.pin;
2188 apic2 = ioapic_i8259.apic;
2189
2190 apic_printk(APIC_QUIET, KERN_INFO "..TIMER: vector=0x%02X "
2191 "apic1=%d pin1=%d apic2=%d pin2=%d\n",
2192 cfg->vector, apic1, pin1, apic2, pin2);
2193
2194 /*
2195 * Some BIOS writers are clueless and report the ExtINTA
2196 * I/O APIC input from the cascaded 8259A as the timer
2197 * interrupt input. So just in case, if only one pin
2198 * was found above, try it both directly and through the
2199 * 8259A.
2200 */
2201 if (pin1 == -1) {
2202 panic_if_irq_remap("BIOS bug: timer not connected to IO-APIC");
2203 pin1 = pin2;
2204 apic1 = apic2;
2205 no_pin1 = 1;
2206 } else if (pin2 == -1) {
2207 pin2 = pin1;
2208 apic2 = apic1;
2209 }
2210
2211 if (pin1 != -1) {
2212 /* Ok, does IRQ0 through the IOAPIC work? */
2213 if (no_pin1) {
2214 mp_alloc_timer_irq(apic1, pin1);
2215 } else {
2216 /*
2217 * for edge trigger, it's already unmasked,
2218 * so only need to unmask if it is level-trigger
2219 * do we really have level trigger timer?
2220 */
2221 int idx = find_irq_entry(apic1, pin1, mp_INT);
2222
2223 if (idx != -1 && irq_is_level(idx))
2224 unmask_ioapic_irq(irq_get_irq_data(0));
2225 }
2226 irq_domain_deactivate_irq(irq_data);
2227 irq_domain_activate_irq(irq_data, false);
2228 if (timer_irq_works()) {
2229 if (disable_timer_pin_1 > 0)
2230 clear_IO_APIC_pin(0, pin1);
2231 goto out;
2232 }
2233 panic_if_irq_remap("timer doesn't work through Interrupt-remapped IO-APIC");
2234 clear_IO_APIC_pin(apic1, pin1);
2235 if (!no_pin1)
2236 apic_printk(APIC_QUIET, KERN_ERR "..MP-BIOS bug: "
2237 "8254 timer not connected to IO-APIC\n");
2238
2239 apic_printk(APIC_QUIET, KERN_INFO "...trying to set up timer "
2240 "(IRQ0) through the 8259A ...\n");
2241 apic_printk(APIC_QUIET, KERN_INFO
2242 "..... (found apic %d pin %d) ...\n", apic2, pin2);
2243 /*
2244 * legacy devices should be connected to IO APIC #0
2245 */
2246 replace_pin_at_irq_node(data, node, apic1, pin1, apic2, pin2);
2247 irq_domain_deactivate_irq(irq_data);
2248 irq_domain_activate_irq(irq_data, false);
2249 legacy_pic->unmask(0);
2250 if (timer_irq_works()) {
2251 apic_printk(APIC_QUIET, KERN_INFO "....... works.\n");
2252 goto out;
2253 }
2254 /*
2255 * Cleanup, just in case ...
2256 */
2257 legacy_pic->mask(0);
2258 clear_IO_APIC_pin(apic2, pin2);
2259 apic_printk(APIC_QUIET, KERN_INFO "....... failed.\n");
2260 }
2261
2262 apic_printk(APIC_QUIET, KERN_INFO
2263 "...trying to set up timer as Virtual Wire IRQ...\n");
2264
2265 lapic_register_intr(0);
2266 apic_write(APIC_LVT0, APIC_DM_FIXED | cfg->vector); /* Fixed mode */
2267 legacy_pic->unmask(0);
2268
2269 if (timer_irq_works()) {
2270 apic_printk(APIC_QUIET, KERN_INFO "..... works.\n");
2271 goto out;
2272 }
2273 legacy_pic->mask(0);
2274 apic_write(APIC_LVT0, APIC_LVT_MASKED | APIC_DM_FIXED | cfg->vector);
2275 apic_printk(APIC_QUIET, KERN_INFO "..... failed.\n");
2276
2277 apic_printk(APIC_QUIET, KERN_INFO
2278 "...trying to set up timer as ExtINT IRQ...\n");
2279
2280 legacy_pic->init(0);
2281 legacy_pic->make_irq(0);
2282 apic_write(APIC_LVT0, APIC_DM_EXTINT);
2283 legacy_pic->unmask(0);
2284
2285 unlock_ExtINT_logic();
2286
2287 if (timer_irq_works()) {
2288 apic_printk(APIC_QUIET, KERN_INFO "..... works.\n");
2289 goto out;
2290 }
2291 apic_printk(APIC_QUIET, KERN_INFO "..... failed :(.\n");
2292 if (apic_is_x2apic_enabled())
2293 apic_printk(APIC_QUIET, KERN_INFO
2294 "Perhaps problem with the pre-enabled x2apic mode\n"
2295 "Try booting with x2apic and interrupt-remapping disabled in the bios.\n");
2296 panic("IO-APIC + timer doesn't work! Boot with apic=debug and send a "
2297 "report. Then try booting with the 'noapic' option.\n");
2298 out:
2299 local_irq_enable();
2300 }
2301
2302 /*
2303 * Traditionally ISA IRQ2 is the cascade IRQ, and is not available
2304 * to devices. However there may be an I/O APIC pin available for
2305 * this interrupt regardless. The pin may be left unconnected, but
2306 * typically it will be reused as an ExtINT cascade interrupt for
2307 * the master 8259A. In the MPS case such a pin will normally be
2308 * reported as an ExtINT interrupt in the MP table. With ACPI
2309 * there is no provision for ExtINT interrupts, and in the absence
2310 * of an override it would be treated as an ordinary ISA I/O APIC
2311 * interrupt, that is edge-triggered and unmasked by default. We
2312 * used to do this, but it caused problems on some systems because
2313 * of the NMI watchdog and sometimes IRQ0 of the 8254 timer using
2314 * the same ExtINT cascade interrupt to drive the local APIC of the
2315 * bootstrap processor. Therefore we refrain from routing IRQ2 to
2316 * the I/O APIC in all cases now. No actual device should request
2317 * it anyway. --macro
2318 */
2319 #define PIC_IRQS (1UL << PIC_CASCADE_IR)
2320
2321 static int mp_irqdomain_create(int ioapic)
2322 {
2323 struct irq_domain *parent;
2324 int hwirqs = mp_ioapic_pin_count(ioapic);
2325 struct ioapic *ip = &ioapics[ioapic];
2326 struct ioapic_domain_cfg *cfg = &ip->irqdomain_cfg;
2327 struct mp_ioapic_gsi *gsi_cfg = mp_ioapic_gsi_routing(ioapic);
2328 struct fwnode_handle *fn;
2329 struct irq_fwspec fwspec;
2330
2331 if (cfg->type == IOAPIC_DOMAIN_INVALID)
2332 return 0;
2333
2334 /* Handle device tree enumerated APICs proper */
2335 if (cfg->dev) {
2336 fn = of_node_to_fwnode(cfg->dev);
2337 } else {
2338 fn = irq_domain_alloc_named_id_fwnode("IO-APIC", mpc_ioapic_id(ioapic));
2339 if (!fn)
2340 return -ENOMEM;
2341 }
2342
2343 fwspec.fwnode = fn;
2344 fwspec.param_count = 1;
2345 fwspec.param[0] = mpc_ioapic_id(ioapic);
2346
2347 parent = irq_find_matching_fwspec(&fwspec, DOMAIN_BUS_ANY);
2348 if (!parent) {
2349 if (!cfg->dev)
2350 irq_domain_free_fwnode(fn);
2351 return -ENODEV;
2352 }
2353
2354 ip->irqdomain = irq_domain_create_linear(fn, hwirqs, cfg->ops,
2355 (void *)(long)ioapic);
2356
2357 if (!ip->irqdomain) {
2358 /* Release fw handle if it was allocated above */
2359 if (!cfg->dev)
2360 irq_domain_free_fwnode(fn);
2361 return -ENOMEM;
2362 }
2363
2364 ip->irqdomain->parent = parent;
2365
2366 if (cfg->type == IOAPIC_DOMAIN_LEGACY ||
2367 cfg->type == IOAPIC_DOMAIN_STRICT)
2368 ioapic_dynirq_base = max(ioapic_dynirq_base,
2369 gsi_cfg->gsi_end + 1);
2370
2371 return 0;
2372 }
2373
2374 static void ioapic_destroy_irqdomain(int idx)
2375 {
2376 struct ioapic_domain_cfg *cfg = &ioapics[idx].irqdomain_cfg;
2377 struct fwnode_handle *fn = ioapics[idx].irqdomain->fwnode;
2378
2379 if (ioapics[idx].irqdomain) {
2380 irq_domain_remove(ioapics[idx].irqdomain);
2381 if (!cfg->dev)
2382 irq_domain_free_fwnode(fn);
2383 ioapics[idx].irqdomain = NULL;
2384 }
2385 }
2386
2387 void __init setup_IO_APIC(void)
2388 {
2389 int ioapic;
2390
2391 if (skip_ioapic_setup || !nr_ioapics)
2392 return;
2393
2394 io_apic_irqs = nr_legacy_irqs() ? ~PIC_IRQS : ~0UL;
2395
2396 apic_printk(APIC_VERBOSE, "ENABLING IO-APIC IRQs\n");
2397 for_each_ioapic(ioapic)
2398 BUG_ON(mp_irqdomain_create(ioapic));
2399
2400 /*
2401 * Set up IO-APIC IRQ routing.
2402 */
2403 x86_init.mpparse.setup_ioapic_ids();
2404
2405 sync_Arb_IDs();
2406 setup_IO_APIC_irqs();
2407 init_IO_APIC_traps();
2408 if (nr_legacy_irqs())
2409 check_timer();
2410
2411 ioapic_initialized = 1;
2412 }
2413
2414 static void resume_ioapic_id(int ioapic_idx)
2415 {
2416 unsigned long flags;
2417 union IO_APIC_reg_00 reg_00;
2418
2419 raw_spin_lock_irqsave(&ioapic_lock, flags);
2420 reg_00.raw = io_apic_read(ioapic_idx, 0);
2421 if (reg_00.bits.ID != mpc_ioapic_id(ioapic_idx)) {
2422 reg_00.bits.ID = mpc_ioapic_id(ioapic_idx);
2423 io_apic_write(ioapic_idx, 0, reg_00.raw);
2424 }
2425 raw_spin_unlock_irqrestore(&ioapic_lock, flags);
2426 }
2427
2428 static void ioapic_resume(void)
2429 {
2430 int ioapic_idx;
2431
2432 for_each_ioapic_reverse(ioapic_idx)
2433 resume_ioapic_id(ioapic_idx);
2434
2435 restore_ioapic_entries();
2436 }
2437
2438 static struct syscore_ops ioapic_syscore_ops = {
2439 .suspend = save_ioapic_entries,
2440 .resume = ioapic_resume,
2441 };
2442
2443 static int __init ioapic_init_ops(void)
2444 {
2445 register_syscore_ops(&ioapic_syscore_ops);
2446
2447 return 0;
2448 }
2449
2450 device_initcall(ioapic_init_ops);
2451
2452 static int io_apic_get_redir_entries(int ioapic)
2453 {
2454 union IO_APIC_reg_01 reg_01;
2455 unsigned long flags;
2456
2457 raw_spin_lock_irqsave(&ioapic_lock, flags);
2458 reg_01.raw = io_apic_read(ioapic, 1);
2459 raw_spin_unlock_irqrestore(&ioapic_lock, flags);
2460
2461 /* The register returns the maximum index redir index
2462 * supported, which is one less than the total number of redir
2463 * entries.
2464 */
2465 return reg_01.bits.entries + 1;
2466 }
2467
2468 unsigned int arch_dynirq_lower_bound(unsigned int from)
2469 {
2470 /*
2471 * dmar_alloc_hwirq() may be called before setup_IO_APIC(), so use
2472 * gsi_top if ioapic_dynirq_base hasn't been initialized yet.
2473 */
2474 if (!ioapic_initialized)
2475 return gsi_top;
2476 /*
2477 * For DT enabled machines ioapic_dynirq_base is irrelevant and not
2478 * updated. So simply return @from if ioapic_dynirq_base == 0.
2479 */
2480 return ioapic_dynirq_base ? : from;
2481 }
2482
2483 #ifdef CONFIG_X86_32
2484 static int io_apic_get_unique_id(int ioapic, int apic_id)
2485 {
2486 union IO_APIC_reg_00 reg_00;
2487 static physid_mask_t apic_id_map = PHYSID_MASK_NONE;
2488 physid_mask_t tmp;
2489 unsigned long flags;
2490 int i = 0;
2491
2492 /*
2493 * The P4 platform supports up to 256 APIC IDs on two separate APIC
2494 * buses (one for LAPICs, one for IOAPICs), where predecessors only
2495 * supports up to 16 on one shared APIC bus.
2496 *
2497 * TBD: Expand LAPIC/IOAPIC support on P4-class systems to take full
2498 * advantage of new APIC bus architecture.
2499 */
2500
2501 if (physids_empty(apic_id_map))
2502 apic->ioapic_phys_id_map(&phys_cpu_present_map, &apic_id_map);
2503
2504 raw_spin_lock_irqsave(&ioapic_lock, flags);
2505 reg_00.raw = io_apic_read(ioapic, 0);
2506 raw_spin_unlock_irqrestore(&ioapic_lock, flags);
2507
2508 if (apic_id >= get_physical_broadcast()) {
2509 printk(KERN_WARNING "IOAPIC[%d]: Invalid apic_id %d, trying "
2510 "%d\n", ioapic, apic_id, reg_00.bits.ID);
2511 apic_id = reg_00.bits.ID;
2512 }
2513
2514 /*
2515 * Every APIC in a system must have a unique ID or we get lots of nice
2516 * 'stuck on smp_invalidate_needed IPI wait' messages.
2517 */
2518 if (apic->check_apicid_used(&apic_id_map, apic_id)) {
2519
2520 for (i = 0; i < get_physical_broadcast(); i++) {
2521 if (!apic->check_apicid_used(&apic_id_map, i))
2522 break;
2523 }
2524
2525 if (i == get_physical_broadcast())
2526 panic("Max apic_id exceeded!\n");
2527
2528 printk(KERN_WARNING "IOAPIC[%d]: apic_id %d already used, "
2529 "trying %d\n", ioapic, apic_id, i);
2530
2531 apic_id = i;
2532 }
2533
2534 apic->apicid_to_cpu_present(apic_id, &tmp);
2535 physids_or(apic_id_map, apic_id_map, tmp);
2536
2537 if (reg_00.bits.ID != apic_id) {
2538 reg_00.bits.ID = apic_id;
2539
2540 raw_spin_lock_irqsave(&ioapic_lock, flags);
2541 io_apic_write(ioapic, 0, reg_00.raw);
2542 reg_00.raw = io_apic_read(ioapic, 0);
2543 raw_spin_unlock_irqrestore(&ioapic_lock, flags);
2544
2545 /* Sanity check */
2546 if (reg_00.bits.ID != apic_id) {
2547 pr_err("IOAPIC[%d]: Unable to change apic_id!\n",
2548 ioapic);
2549 return -1;
2550 }
2551 }
2552
2553 apic_printk(APIC_VERBOSE, KERN_INFO
2554 "IOAPIC[%d]: Assigned apic_id %d\n", ioapic, apic_id);
2555
2556 return apic_id;
2557 }
2558
2559 static u8 io_apic_unique_id(int idx, u8 id)
2560 {
2561 if ((boot_cpu_data.x86_vendor == X86_VENDOR_INTEL) &&
2562 !APIC_XAPIC(boot_cpu_apic_version))
2563 return io_apic_get_unique_id(idx, id);
2564 else
2565 return id;
2566 }
2567 #else
2568 static u8 io_apic_unique_id(int idx, u8 id)
2569 {
2570 union IO_APIC_reg_00 reg_00;
2571 DECLARE_BITMAP(used, 256);
2572 unsigned long flags;
2573 u8 new_id;
2574 int i;
2575
2576 bitmap_zero(used, 256);
2577 for_each_ioapic(i)
2578 __set_bit(mpc_ioapic_id(i), used);
2579
2580 /* Hand out the requested id if available */
2581 if (!test_bit(id, used))
2582 return id;
2583
2584 /*
2585 * Read the current id from the ioapic and keep it if
2586 * available.
2587 */
2588 raw_spin_lock_irqsave(&ioapic_lock, flags);
2589 reg_00.raw = io_apic_read(idx, 0);
2590 raw_spin_unlock_irqrestore(&ioapic_lock, flags);
2591 new_id = reg_00.bits.ID;
2592 if (!test_bit(new_id, used)) {
2593 apic_printk(APIC_VERBOSE, KERN_INFO
2594 "IOAPIC[%d]: Using reg apic_id %d instead of %d\n",
2595 idx, new_id, id);
2596 return new_id;
2597 }
2598
2599 /*
2600 * Get the next free id and write it to the ioapic.
2601 */
2602 new_id = find_first_zero_bit(used, 256);
2603 reg_00.bits.ID = new_id;
2604 raw_spin_lock_irqsave(&ioapic_lock, flags);
2605 io_apic_write(idx, 0, reg_00.raw);
2606 reg_00.raw = io_apic_read(idx, 0);
2607 raw_spin_unlock_irqrestore(&ioapic_lock, flags);
2608 /* Sanity check */
2609 BUG_ON(reg_00.bits.ID != new_id);
2610
2611 return new_id;
2612 }
2613 #endif
2614
2615 static int io_apic_get_version(int ioapic)
2616 {
2617 union IO_APIC_reg_01 reg_01;
2618 unsigned long flags;
2619
2620 raw_spin_lock_irqsave(&ioapic_lock, flags);
2621 reg_01.raw = io_apic_read(ioapic, 1);
2622 raw_spin_unlock_irqrestore(&ioapic_lock, flags);
2623
2624 return reg_01.bits.version;
2625 }
2626
2627 /*
2628 * This function updates target affinity of IOAPIC interrupts to include
2629 * the CPUs which came online during SMP bringup.
2630 */
2631 #define IOAPIC_RESOURCE_NAME_SIZE 11
2632
2633 static struct resource *ioapic_resources;
2634
2635 static struct resource * __init ioapic_setup_resources(void)
2636 {
2637 unsigned long n;
2638 struct resource *res;
2639 char *mem;
2640 int i;
2641
2642 if (nr_ioapics == 0)
2643 return NULL;
2644
2645 n = IOAPIC_RESOURCE_NAME_SIZE + sizeof(struct resource);
2646 n *= nr_ioapics;
2647
2648 mem = memblock_alloc(n, SMP_CACHE_BYTES);
2649 if (!mem)
2650 panic("%s: Failed to allocate %lu bytes\n", __func__, n);
2651 res = (void *)mem;
2652
2653 mem += sizeof(struct resource) * nr_ioapics;
2654
2655 for_each_ioapic(i) {
2656 res[i].name = mem;
2657 res[i].flags = IORESOURCE_MEM | IORESOURCE_BUSY;
2658 snprintf(mem, IOAPIC_RESOURCE_NAME_SIZE, "IOAPIC %u", i);
2659 mem += IOAPIC_RESOURCE_NAME_SIZE;
2660 ioapics[i].iomem_res = &res[i];
2661 }
2662
2663 ioapic_resources = res;
2664
2665 return res;
2666 }
2667
2668 void __init io_apic_init_mappings(void)
2669 {
2670 unsigned long ioapic_phys, idx = FIX_IO_APIC_BASE_0;
2671 struct resource *ioapic_res;
2672 int i;
2673
2674 ioapic_res = ioapic_setup_resources();
2675 for_each_ioapic(i) {
2676 if (smp_found_config) {
2677 ioapic_phys = mpc_ioapic_addr(i);
2678 #ifdef CONFIG_X86_32
2679 if (!ioapic_phys) {
2680 printk(KERN_ERR
2681 "WARNING: bogus zero IO-APIC "
2682 "address found in MPTABLE, "
2683 "disabling IO/APIC support!\n");
2684 smp_found_config = 0;
2685 skip_ioapic_setup = 1;
2686 goto fake_ioapic_page;
2687 }
2688 #endif
2689 } else {
2690 #ifdef CONFIG_X86_32
2691 fake_ioapic_page:
2692 #endif
2693 ioapic_phys = (unsigned long)memblock_alloc(PAGE_SIZE,
2694 PAGE_SIZE);
2695 if (!ioapic_phys)
2696 panic("%s: Failed to allocate %lu bytes align=0x%lx\n",
2697 __func__, PAGE_SIZE, PAGE_SIZE);
2698 ioapic_phys = __pa(ioapic_phys);
2699 }
2700 set_fixmap_nocache(idx, ioapic_phys);
2701 apic_printk(APIC_VERBOSE, "mapped IOAPIC to %08lx (%08lx)\n",
2702 __fix_to_virt(idx) + (ioapic_phys & ~PAGE_MASK),
2703 ioapic_phys);
2704 idx++;
2705
2706 ioapic_res->start = ioapic_phys;
2707 ioapic_res->end = ioapic_phys + IO_APIC_SLOT_SIZE - 1;
2708 ioapic_res++;
2709 }
2710 }
2711
2712 void __init ioapic_insert_resources(void)
2713 {
2714 int i;
2715 struct resource *r = ioapic_resources;
2716
2717 if (!r) {
2718 if (nr_ioapics > 0)
2719 printk(KERN_ERR
2720 "IO APIC resources couldn't be allocated.\n");
2721 return;
2722 }
2723
2724 for_each_ioapic(i) {
2725 insert_resource(&iomem_resource, r);
2726 r++;
2727 }
2728 }
2729
2730 int mp_find_ioapic(u32 gsi)
2731 {
2732 int i;
2733
2734 if (nr_ioapics == 0)
2735 return -1;
2736
2737 /* Find the IOAPIC that manages this GSI. */
2738 for_each_ioapic(i) {
2739 struct mp_ioapic_gsi *gsi_cfg = mp_ioapic_gsi_routing(i);
2740 if (gsi >= gsi_cfg->gsi_base && gsi <= gsi_cfg->gsi_end)
2741 return i;
2742 }
2743
2744 printk(KERN_ERR "ERROR: Unable to locate IOAPIC for GSI %d\n", gsi);
2745 return -1;
2746 }
2747
2748 int mp_find_ioapic_pin(int ioapic, u32 gsi)
2749 {
2750 struct mp_ioapic_gsi *gsi_cfg;
2751
2752 if (WARN_ON(ioapic < 0))
2753 return -1;
2754
2755 gsi_cfg = mp_ioapic_gsi_routing(ioapic);
2756 if (WARN_ON(gsi > gsi_cfg->gsi_end))
2757 return -1;
2758
2759 return gsi - gsi_cfg->gsi_base;
2760 }
2761
2762 static int bad_ioapic_register(int idx)
2763 {
2764 union IO_APIC_reg_00 reg_00;
2765 union IO_APIC_reg_01 reg_01;
2766 union IO_APIC_reg_02 reg_02;
2767
2768 reg_00.raw = io_apic_read(idx, 0);
2769 reg_01.raw = io_apic_read(idx, 1);
2770 reg_02.raw = io_apic_read(idx, 2);
2771
2772 if (reg_00.raw == -1 && reg_01.raw == -1 && reg_02.raw == -1) {
2773 pr_warn("I/O APIC 0x%x registers return all ones, skipping!\n",
2774 mpc_ioapic_addr(idx));
2775 return 1;
2776 }
2777
2778 return 0;
2779 }
2780
2781 static int find_free_ioapic_entry(void)
2782 {
2783 int idx;
2784
2785 for (idx = 0; idx < MAX_IO_APICS; idx++)
2786 if (ioapics[idx].nr_registers == 0)
2787 return idx;
2788
2789 return MAX_IO_APICS;
2790 }
2791
2792 /**
2793 * mp_register_ioapic - Register an IOAPIC device
2794 * @id: hardware IOAPIC ID
2795 * @address: physical address of IOAPIC register area
2796 * @gsi_base: base of GSI associated with the IOAPIC
2797 * @cfg: configuration information for the IOAPIC
2798 */
2799 int mp_register_ioapic(int id, u32 address, u32 gsi_base,
2800 struct ioapic_domain_cfg *cfg)
2801 {
2802 bool hotplug = !!ioapic_initialized;
2803 struct mp_ioapic_gsi *gsi_cfg;
2804 int idx, ioapic, entries;
2805 u32 gsi_end;
2806
2807 if (!address) {
2808 pr_warn("Bogus (zero) I/O APIC address found, skipping!\n");
2809 return -EINVAL;
2810 }
2811 for_each_ioapic(ioapic)
2812 if (ioapics[ioapic].mp_config.apicaddr == address) {
2813 pr_warn("address 0x%x conflicts with IOAPIC%d\n",
2814 address, ioapic);
2815 return -EEXIST;
2816 }
2817
2818 idx = find_free_ioapic_entry();
2819 if (idx >= MAX_IO_APICS) {
2820 pr_warn("Max # of I/O APICs (%d) exceeded (found %d), skipping\n",
2821 MAX_IO_APICS, idx);
2822 return -ENOSPC;
2823 }
2824
2825 ioapics[idx].mp_config.type = MP_IOAPIC;
2826 ioapics[idx].mp_config.flags = MPC_APIC_USABLE;
2827 ioapics[idx].mp_config.apicaddr = address;
2828
2829 set_fixmap_nocache(FIX_IO_APIC_BASE_0 + idx, address);
2830 if (bad_ioapic_register(idx)) {
2831 clear_fixmap(FIX_IO_APIC_BASE_0 + idx);
2832 return -ENODEV;
2833 }
2834
2835 ioapics[idx].mp_config.apicid = io_apic_unique_id(idx, id);
2836 ioapics[idx].mp_config.apicver = io_apic_get_version(idx);
2837
2838 /*
2839 * Build basic GSI lookup table to facilitate gsi->io_apic lookups
2840 * and to prevent reprogramming of IOAPIC pins (PCI GSIs).
2841 */
2842 entries = io_apic_get_redir_entries(idx);
2843 gsi_end = gsi_base + entries - 1;
2844 for_each_ioapic(ioapic) {
2845 gsi_cfg = mp_ioapic_gsi_routing(ioapic);
2846 if ((gsi_base >= gsi_cfg->gsi_base &&
2847 gsi_base <= gsi_cfg->gsi_end) ||
2848 (gsi_end >= gsi_cfg->gsi_base &&
2849 gsi_end <= gsi_cfg->gsi_end)) {
2850 pr_warn("GSI range [%u-%u] for new IOAPIC conflicts with GSI[%u-%u]\n",
2851 gsi_base, gsi_end,
2852 gsi_cfg->gsi_base, gsi_cfg->gsi_end);
2853 clear_fixmap(FIX_IO_APIC_BASE_0 + idx);
2854 return -ENOSPC;
2855 }
2856 }
2857 gsi_cfg = mp_ioapic_gsi_routing(idx);
2858 gsi_cfg->gsi_base = gsi_base;
2859 gsi_cfg->gsi_end = gsi_end;
2860
2861 ioapics[idx].irqdomain = NULL;
2862 ioapics[idx].irqdomain_cfg = *cfg;
2863
2864 /*
2865 * If mp_register_ioapic() is called during early boot stage when
2866 * walking ACPI/SFI/DT tables, it's too early to create irqdomain,
2867 * we are still using bootmem allocator. So delay it to setup_IO_APIC().
2868 */
2869 if (hotplug) {
2870 if (mp_irqdomain_create(idx)) {
2871 clear_fixmap(FIX_IO_APIC_BASE_0 + idx);
2872 return -ENOMEM;
2873 }
2874 alloc_ioapic_saved_registers(idx);
2875 }
2876
2877 if (gsi_cfg->gsi_end >= gsi_top)
2878 gsi_top = gsi_cfg->gsi_end + 1;
2879 if (nr_ioapics <= idx)
2880 nr_ioapics = idx + 1;
2881
2882 /* Set nr_registers to mark entry present */
2883 ioapics[idx].nr_registers = entries;
2884
2885 pr_info("IOAPIC[%d]: apic_id %d, version %d, address 0x%x, GSI %d-%d\n",
2886 idx, mpc_ioapic_id(idx),
2887 mpc_ioapic_ver(idx), mpc_ioapic_addr(idx),
2888 gsi_cfg->gsi_base, gsi_cfg->gsi_end);
2889
2890 return 0;
2891 }
2892
2893 int mp_unregister_ioapic(u32 gsi_base)
2894 {
2895 int ioapic, pin;
2896 int found = 0;
2897
2898 for_each_ioapic(ioapic)
2899 if (ioapics[ioapic].gsi_config.gsi_base == gsi_base) {
2900 found = 1;
2901 break;
2902 }
2903 if (!found) {
2904 pr_warn("can't find IOAPIC for GSI %d\n", gsi_base);
2905 return -ENODEV;
2906 }
2907
2908 for_each_pin(ioapic, pin) {
2909 u32 gsi = mp_pin_to_gsi(ioapic, pin);
2910 int irq = mp_map_gsi_to_irq(gsi, 0, NULL);
2911 struct mp_chip_data *data;
2912
2913 if (irq >= 0) {
2914 data = irq_get_chip_data(irq);
2915 if (data && data->count) {
2916 pr_warn("pin%d on IOAPIC%d is still in use.\n",
2917 pin, ioapic);
2918 return -EBUSY;
2919 }
2920 }
2921 }
2922
2923 /* Mark entry not present */
2924 ioapics[ioapic].nr_registers = 0;
2925 ioapic_destroy_irqdomain(ioapic);
2926 free_ioapic_saved_registers(ioapic);
2927 if (ioapics[ioapic].iomem_res)
2928 release_resource(ioapics[ioapic].iomem_res);
2929 clear_fixmap(FIX_IO_APIC_BASE_0 + ioapic);
2930 memset(&ioapics[ioapic], 0, sizeof(ioapics[ioapic]));
2931
2932 return 0;
2933 }
2934
2935 int mp_ioapic_registered(u32 gsi_base)
2936 {
2937 int ioapic;
2938
2939 for_each_ioapic(ioapic)
2940 if (ioapics[ioapic].gsi_config.gsi_base == gsi_base)
2941 return 1;
2942
2943 return 0;
2944 }
2945
2946 static void mp_irqdomain_get_attr(u32 gsi, struct mp_chip_data *data,
2947 struct irq_alloc_info *info)
2948 {
2949 if (info && info->ioapic.valid) {
2950 data->is_level = info->ioapic.is_level;
2951 data->active_low = info->ioapic.active_low;
2952 } else if (__acpi_get_override_irq(gsi, &data->is_level,
2953 &data->active_low) < 0) {
2954 /* PCI interrupts are always active low level triggered. */
2955 data->is_level = true;
2956 data->active_low = true;
2957 }
2958 }
2959
2960 /*
2961 * Configure the I/O-APIC specific fields in the routing entry.
2962 *
2963 * This is important to setup the I/O-APIC specific bits (is_level,
2964 * active_low, masked) because the underlying parent domain will only
2965 * provide the routing information and is oblivious of the I/O-APIC
2966 * specific bits.
2967 *
2968 * The entry is just preconfigured at this point and not written into the
2969 * RTE. This happens later during activation which will fill in the actual
2970 * routing information.
2971 */
2972 static void mp_preconfigure_entry(struct mp_chip_data *data)
2973 {
2974 struct IO_APIC_route_entry *entry = &data->entry;
2975
2976 memset(entry, 0, sizeof(*entry));
2977 entry->is_level = data->is_level;
2978 entry->active_low = data->active_low;
2979 /*
2980 * Mask level triggered irqs. Edge triggered irqs are masked
2981 * by the irq core code in case they fire.
2982 */
2983 entry->masked = data->is_level;
2984 }
2985
2986 int mp_irqdomain_alloc(struct irq_domain *domain, unsigned int virq,
2987 unsigned int nr_irqs, void *arg)
2988 {
2989 struct irq_alloc_info *info = arg;
2990 struct mp_chip_data *data;
2991 struct irq_data *irq_data;
2992 int ret, ioapic, pin;
2993 unsigned long flags;
2994
2995 if (!info || nr_irqs > 1)
2996 return -EINVAL;
2997 irq_data = irq_domain_get_irq_data(domain, virq);
2998 if (!irq_data)
2999 return -EINVAL;
3000
3001 ioapic = mp_irqdomain_ioapic_idx(domain);
3002 pin = info->ioapic.pin;
3003 if (irq_find_mapping(domain, (irq_hw_number_t)pin) > 0)
3004 return -EEXIST;
3005
3006 data = kzalloc(sizeof(*data), GFP_KERNEL);
3007 if (!data)
3008 return -ENOMEM;
3009
3010 ret = irq_domain_alloc_irqs_parent(domain, virq, nr_irqs, info);
3011 if (ret < 0) {
3012 kfree(data);
3013 return ret;
3014 }
3015
3016 INIT_LIST_HEAD(&data->irq_2_pin);
3017 irq_data->hwirq = info->ioapic.pin;
3018 irq_data->chip = (domain->parent == x86_vector_domain) ?
3019 &ioapic_chip : &ioapic_ir_chip;
3020 irq_data->chip_data = data;
3021 mp_irqdomain_get_attr(mp_pin_to_gsi(ioapic, pin), data, info);
3022
3023 add_pin_to_irq_node(data, ioapic_alloc_attr_node(info), ioapic, pin);
3024
3025 mp_preconfigure_entry(data);
3026 mp_register_handler(virq, data->is_level);
3027
3028 local_irq_save(flags);
3029 if (virq < nr_legacy_irqs())
3030 legacy_pic->mask(virq);
3031 local_irq_restore(flags);
3032
3033 apic_printk(APIC_VERBOSE, KERN_DEBUG
3034 "IOAPIC[%d]: Preconfigured routing entry (%d-%d -> IRQ %d Level:%i ActiveLow:%i)\n",
3035 ioapic, mpc_ioapic_id(ioapic), pin, virq,
3036 data->is_level, data->active_low);
3037 return 0;
3038 }
3039
3040 void mp_irqdomain_free(struct irq_domain *domain, unsigned int virq,
3041 unsigned int nr_irqs)
3042 {
3043 struct irq_data *irq_data;
3044 struct mp_chip_data *data;
3045
3046 BUG_ON(nr_irqs != 1);
3047 irq_data = irq_domain_get_irq_data(domain, virq);
3048 if (irq_data && irq_data->chip_data) {
3049 data = irq_data->chip_data;
3050 __remove_pin_from_irq(data, mp_irqdomain_ioapic_idx(domain),
3051 (int)irq_data->hwirq);
3052 WARN_ON(!list_empty(&data->irq_2_pin));
3053 kfree(irq_data->chip_data);
3054 }
3055 irq_domain_free_irqs_top(domain, virq, nr_irqs);
3056 }
3057
3058 int mp_irqdomain_activate(struct irq_domain *domain,
3059 struct irq_data *irq_data, bool reserve)
3060 {
3061 unsigned long flags;
3062
3063 raw_spin_lock_irqsave(&ioapic_lock, flags);
3064 ioapic_configure_entry(irq_data);
3065 raw_spin_unlock_irqrestore(&ioapic_lock, flags);
3066 return 0;
3067 }
3068
3069 void mp_irqdomain_deactivate(struct irq_domain *domain,
3070 struct irq_data *irq_data)
3071 {
3072 /* It won't be called for IRQ with multiple IOAPIC pins associated */
3073 ioapic_mask_entry(mp_irqdomain_ioapic_idx(domain),
3074 (int)irq_data->hwirq);
3075 }
3076
3077 int mp_irqdomain_ioapic_idx(struct irq_domain *domain)
3078 {
3079 return (int)(long)domain->host_data;
3080 }
3081
3082 const struct irq_domain_ops mp_ioapic_irqdomain_ops = {
3083 .alloc = mp_irqdomain_alloc,
3084 .free = mp_irqdomain_free,
3085 .activate = mp_irqdomain_activate,
3086 .deactivate = mp_irqdomain_deactivate,
3087 };