1 /* 2 * Intel IO-APIC support for multi-Pentium hosts. 3 * 4 * Copyright (C) 1997, 1998, 1999, 2000, 2009 Ingo Molnar, Hajnalka Szabo 5 * 6 * Many thanks to Stig Venaas for trying out countless experimental 7 * patches and reporting/debugging problems patiently! 8 * 9 * (c) 1999, Multiple IO-APIC support, developed by 10 * Ken-ichi Yaku <yaku@css1.kbnes.nec.co.jp> and 11 * Hidemi Kishimoto <kisimoto@css1.kbnes.nec.co.jp>, 12 * further tested and cleaned up by Zach Brown <zab@redhat.com> 13 * and Ingo Molnar <mingo@redhat.com> 14 * 15 * Fixes 16 * Maciej W. Rozycki : Bits for genuine 82489DX APICs; 17 * thanks to Eric Gilmore 18 * and Rolf G. Tews 19 * for testing these extensively 20 * Paul Diefenbaugh : Added full ACPI support 21 */ 22 23 #include <linux/mm.h> 24 #include <linux/interrupt.h> 25 #include <linux/init.h> 26 #include <linux/delay.h> 27 #include <linux/sched.h> 28 #include <linux/pci.h> 29 #include <linux/mc146818rtc.h> 30 #include <linux/compiler.h> 31 #include <linux/acpi.h> 32 #include <linux/module.h> 33 #include <linux/syscore_ops.h> 34 #include <linux/msi.h> 35 #include <linux/htirq.h> 36 #include <linux/freezer.h> 37 #include <linux/kthread.h> 38 #include <linux/jiffies.h> /* time_after() */ 39 #include <linux/slab.h> 40 #include <linux/bootmem.h> 41 #include <linux/dmar.h> 42 #include <linux/hpet.h> 43 44 #include <asm/idle.h> 45 #include <asm/io.h> 46 #include <asm/smp.h> 47 #include <asm/cpu.h> 48 #include <asm/desc.h> 49 #include <asm/proto.h> 50 #include <asm/acpi.h> 51 #include <asm/dma.h> 52 #include <asm/timer.h> 53 #include <asm/i8259.h> 54 #include <asm/msidef.h> 55 #include <asm/hypertransport.h> 56 #include <asm/setup.h> 57 #include <asm/irq_remapping.h> 58 #include <asm/hpet.h> 59 #include <asm/hw_irq.h> 60 61 #include <asm/apic.h> 62 63 #define __apicdebuginit(type) static type __init 64 65 #define for_each_ioapic(idx) \ 66 for ((idx) = 0; (idx) < nr_ioapics; (idx)++) 67 #define for_each_ioapic_reverse(idx) \ 68 for ((idx) = nr_ioapics - 1; (idx) >= 0; (idx)--) 69 #define for_each_pin(idx, pin) \ 70 for ((pin) = 0; (pin) < ioapics[(idx)].nr_registers; (pin)++) 71 #define for_each_ioapic_pin(idx, pin) \ 72 for_each_ioapic((idx)) \ 73 for_each_pin((idx), (pin)) 74 75 #define for_each_irq_pin(entry, head) \ 76 for (entry = head; entry; entry = entry->next) 77 78 /* 79 * Is the SiS APIC rmw bug present ? 80 * -1 = don't know, 0 = no, 1 = yes 81 */ 82 int sis_apic_bug = -1; 83 84 static DEFINE_RAW_SPINLOCK(ioapic_lock); 85 static DEFINE_RAW_SPINLOCK(vector_lock); 86 87 static struct ioapic { 88 /* 89 * # of IRQ routing registers 90 */ 91 int nr_registers; 92 /* 93 * Saved state during suspend/resume, or while enabling intr-remap. 94 */ 95 struct IO_APIC_route_entry *saved_registers; 96 /* I/O APIC config */ 97 struct mpc_ioapic mp_config; 98 /* IO APIC gsi routing info */ 99 struct mp_ioapic_gsi gsi_config; 100 DECLARE_BITMAP(pin_programmed, MP_MAX_IOAPIC_PIN + 1); 101 } ioapics[MAX_IO_APICS]; 102 103 #define mpc_ioapic_ver(ioapic_idx) ioapics[ioapic_idx].mp_config.apicver 104 105 int mpc_ioapic_id(int ioapic_idx) 106 { 107 return ioapics[ioapic_idx].mp_config.apicid; 108 } 109 110 unsigned int mpc_ioapic_addr(int ioapic_idx) 111 { 112 return ioapics[ioapic_idx].mp_config.apicaddr; 113 } 114 115 struct mp_ioapic_gsi *mp_ioapic_gsi_routing(int ioapic_idx) 116 { 117 return &ioapics[ioapic_idx].gsi_config; 118 } 119 120 static inline int mp_ioapic_pin_count(int ioapic) 121 { 122 struct mp_ioapic_gsi *gsi_cfg = mp_ioapic_gsi_routing(ioapic); 123 124 return gsi_cfg->gsi_end - gsi_cfg->gsi_base + 1; 125 } 126 127 u32 mp_pin_to_gsi(int ioapic, int pin) 128 { 129 return mp_ioapic_gsi_routing(ioapic)->gsi_base + pin; 130 } 131 132 /* Initialize all legacy IRQs and all pins on the first IOAPIC at boot */ 133 static inline int mp_init_irq_at_boot(int ioapic, int irq) 134 { 135 return ioapic == 0 || (irq >= 0 && irq < NR_IRQS_LEGACY); 136 } 137 138 int nr_ioapics; 139 140 /* The one past the highest gsi number used */ 141 u32 gsi_top; 142 143 /* MP IRQ source entries */ 144 struct mpc_intsrc mp_irqs[MAX_IRQ_SOURCES]; 145 146 /* # of MP IRQ source entries */ 147 int mp_irq_entries; 148 149 #ifdef CONFIG_EISA 150 int mp_bus_id_to_type[MAX_MP_BUSSES]; 151 #endif 152 153 DECLARE_BITMAP(mp_bus_not_pci, MAX_MP_BUSSES); 154 155 int skip_ioapic_setup; 156 157 /** 158 * disable_ioapic_support() - disables ioapic support at runtime 159 */ 160 void disable_ioapic_support(void) 161 { 162 #ifdef CONFIG_PCI 163 noioapicquirk = 1; 164 noioapicreroute = -1; 165 #endif 166 skip_ioapic_setup = 1; 167 } 168 169 static int __init parse_noapic(char *str) 170 { 171 /* disable IO-APIC */ 172 disable_ioapic_support(); 173 return 0; 174 } 175 early_param("noapic", parse_noapic); 176 177 static int io_apic_setup_irq_pin(unsigned int irq, int node, 178 struct io_apic_irq_attr *attr); 179 180 /* Will be called in mpparse/acpi/sfi codes for saving IRQ info */ 181 void mp_save_irq(struct mpc_intsrc *m) 182 { 183 int i; 184 185 apic_printk(APIC_VERBOSE, "Int: type %d, pol %d, trig %d, bus %02x," 186 " IRQ %02x, APIC ID %x, APIC INT %02x\n", 187 m->irqtype, m->irqflag & 3, (m->irqflag >> 2) & 3, m->srcbus, 188 m->srcbusirq, m->dstapic, m->dstirq); 189 190 for (i = 0; i < mp_irq_entries; i++) { 191 if (!memcmp(&mp_irqs[i], m, sizeof(*m))) 192 return; 193 } 194 195 memcpy(&mp_irqs[mp_irq_entries], m, sizeof(*m)); 196 if (++mp_irq_entries == MAX_IRQ_SOURCES) 197 panic("Max # of irq sources exceeded!!\n"); 198 } 199 200 struct irq_pin_list { 201 int apic, pin; 202 struct irq_pin_list *next; 203 }; 204 205 static struct irq_pin_list *alloc_irq_pin_list(int node) 206 { 207 return kzalloc_node(sizeof(struct irq_pin_list), GFP_KERNEL, node); 208 } 209 210 211 /* irq_cfg is indexed by the sum of all RTEs in all I/O APICs. */ 212 static struct irq_cfg irq_cfgx[NR_IRQS_LEGACY]; 213 214 int __init arch_early_irq_init(void) 215 { 216 struct irq_cfg *cfg; 217 int count, node, i; 218 219 if (!legacy_pic->nr_legacy_irqs) 220 io_apic_irqs = ~0UL; 221 222 for_each_ioapic(i) { 223 ioapics[i].saved_registers = 224 kzalloc(sizeof(struct IO_APIC_route_entry) * 225 ioapics[i].nr_registers, GFP_KERNEL); 226 if (!ioapics[i].saved_registers) 227 pr_err("IOAPIC %d: suspend/resume impossible!\n", i); 228 } 229 230 cfg = irq_cfgx; 231 count = ARRAY_SIZE(irq_cfgx); 232 node = cpu_to_node(0); 233 234 for (i = 0; i < count; i++) { 235 irq_set_chip_data(i, &cfg[i]); 236 zalloc_cpumask_var_node(&cfg[i].domain, GFP_KERNEL, node); 237 zalloc_cpumask_var_node(&cfg[i].old_domain, GFP_KERNEL, node); 238 /* 239 * For legacy IRQ's, start with assigning irq0 to irq15 to 240 * IRQ0_VECTOR to IRQ15_VECTOR for all cpu's. 241 */ 242 if (i < legacy_pic->nr_legacy_irqs) { 243 cfg[i].vector = IRQ0_VECTOR + i; 244 cpumask_setall(cfg[i].domain); 245 } 246 } 247 248 return 0; 249 } 250 251 static inline struct irq_cfg *irq_cfg(unsigned int irq) 252 { 253 return irq_get_chip_data(irq); 254 } 255 256 static struct irq_cfg *alloc_irq_cfg(unsigned int irq, int node) 257 { 258 struct irq_cfg *cfg; 259 260 cfg = kzalloc_node(sizeof(*cfg), GFP_KERNEL, node); 261 if (!cfg) 262 return NULL; 263 if (!zalloc_cpumask_var_node(&cfg->domain, GFP_KERNEL, node)) 264 goto out_cfg; 265 if (!zalloc_cpumask_var_node(&cfg->old_domain, GFP_KERNEL, node)) 266 goto out_domain; 267 return cfg; 268 out_domain: 269 free_cpumask_var(cfg->domain); 270 out_cfg: 271 kfree(cfg); 272 return NULL; 273 } 274 275 static void free_irq_cfg(unsigned int at, struct irq_cfg *cfg) 276 { 277 if (!cfg) 278 return; 279 irq_set_chip_data(at, NULL); 280 free_cpumask_var(cfg->domain); 281 free_cpumask_var(cfg->old_domain); 282 kfree(cfg); 283 } 284 285 static struct irq_cfg *alloc_irq_and_cfg_at(unsigned int at, int node) 286 { 287 int res = irq_alloc_desc_at(at, node); 288 struct irq_cfg *cfg; 289 290 if (res < 0) { 291 if (res != -EEXIST) 292 return NULL; 293 cfg = irq_cfg(at); 294 if (cfg) 295 return cfg; 296 } 297 298 cfg = alloc_irq_cfg(at, node); 299 if (cfg) 300 irq_set_chip_data(at, cfg); 301 else 302 irq_free_desc(at); 303 return cfg; 304 } 305 306 struct io_apic { 307 unsigned int index; 308 unsigned int unused[3]; 309 unsigned int data; 310 unsigned int unused2[11]; 311 unsigned int eoi; 312 }; 313 314 static __attribute_const__ struct io_apic __iomem *io_apic_base(int idx) 315 { 316 return (void __iomem *) __fix_to_virt(FIX_IO_APIC_BASE_0 + idx) 317 + (mpc_ioapic_addr(idx) & ~PAGE_MASK); 318 } 319 320 void io_apic_eoi(unsigned int apic, unsigned int vector) 321 { 322 struct io_apic __iomem *io_apic = io_apic_base(apic); 323 writel(vector, &io_apic->eoi); 324 } 325 326 unsigned int native_io_apic_read(unsigned int apic, unsigned int reg) 327 { 328 struct io_apic __iomem *io_apic = io_apic_base(apic); 329 writel(reg, &io_apic->index); 330 return readl(&io_apic->data); 331 } 332 333 void native_io_apic_write(unsigned int apic, unsigned int reg, unsigned int value) 334 { 335 struct io_apic __iomem *io_apic = io_apic_base(apic); 336 337 writel(reg, &io_apic->index); 338 writel(value, &io_apic->data); 339 } 340 341 /* 342 * Re-write a value: to be used for read-modify-write 343 * cycles where the read already set up the index register. 344 * 345 * Older SiS APIC requires we rewrite the index register 346 */ 347 void native_io_apic_modify(unsigned int apic, unsigned int reg, unsigned int value) 348 { 349 struct io_apic __iomem *io_apic = io_apic_base(apic); 350 351 if (sis_apic_bug) 352 writel(reg, &io_apic->index); 353 writel(value, &io_apic->data); 354 } 355 356 union entry_union { 357 struct { u32 w1, w2; }; 358 struct IO_APIC_route_entry entry; 359 }; 360 361 static struct IO_APIC_route_entry __ioapic_read_entry(int apic, int pin) 362 { 363 union entry_union eu; 364 365 eu.w1 = io_apic_read(apic, 0x10 + 2 * pin); 366 eu.w2 = io_apic_read(apic, 0x11 + 2 * pin); 367 368 return eu.entry; 369 } 370 371 static struct IO_APIC_route_entry ioapic_read_entry(int apic, int pin) 372 { 373 union entry_union eu; 374 unsigned long flags; 375 376 raw_spin_lock_irqsave(&ioapic_lock, flags); 377 eu.entry = __ioapic_read_entry(apic, pin); 378 raw_spin_unlock_irqrestore(&ioapic_lock, flags); 379 380 return eu.entry; 381 } 382 383 /* 384 * When we write a new IO APIC routing entry, we need to write the high 385 * word first! If the mask bit in the low word is clear, we will enable 386 * the interrupt, and we need to make sure the entry is fully populated 387 * before that happens. 388 */ 389 static void __ioapic_write_entry(int apic, int pin, struct IO_APIC_route_entry e) 390 { 391 union entry_union eu = {{0, 0}}; 392 393 eu.entry = e; 394 io_apic_write(apic, 0x11 + 2*pin, eu.w2); 395 io_apic_write(apic, 0x10 + 2*pin, eu.w1); 396 } 397 398 static void ioapic_write_entry(int apic, int pin, struct IO_APIC_route_entry e) 399 { 400 unsigned long flags; 401 402 raw_spin_lock_irqsave(&ioapic_lock, flags); 403 __ioapic_write_entry(apic, pin, e); 404 raw_spin_unlock_irqrestore(&ioapic_lock, flags); 405 } 406 407 /* 408 * When we mask an IO APIC routing entry, we need to write the low 409 * word first, in order to set the mask bit before we change the 410 * high bits! 411 */ 412 static void ioapic_mask_entry(int apic, int pin) 413 { 414 unsigned long flags; 415 union entry_union eu = { .entry.mask = 1 }; 416 417 raw_spin_lock_irqsave(&ioapic_lock, flags); 418 io_apic_write(apic, 0x10 + 2*pin, eu.w1); 419 io_apic_write(apic, 0x11 + 2*pin, eu.w2); 420 raw_spin_unlock_irqrestore(&ioapic_lock, flags); 421 } 422 423 /* 424 * The common case is 1:1 IRQ<->pin mappings. Sometimes there are 425 * shared ISA-space IRQs, so we have to support them. We are super 426 * fast in the common case, and fast for shared ISA-space IRQs. 427 */ 428 static int __add_pin_to_irq_node(struct irq_cfg *cfg, int node, int apic, int pin) 429 { 430 struct irq_pin_list **last, *entry; 431 432 /* don't allow duplicates */ 433 last = &cfg->irq_2_pin; 434 for_each_irq_pin(entry, cfg->irq_2_pin) { 435 if (entry->apic == apic && entry->pin == pin) 436 return 0; 437 last = &entry->next; 438 } 439 440 entry = alloc_irq_pin_list(node); 441 if (!entry) { 442 pr_err("can not alloc irq_pin_list (%d,%d,%d)\n", 443 node, apic, pin); 444 return -ENOMEM; 445 } 446 entry->apic = apic; 447 entry->pin = pin; 448 449 *last = entry; 450 return 0; 451 } 452 453 static void add_pin_to_irq_node(struct irq_cfg *cfg, int node, int apic, int pin) 454 { 455 if (__add_pin_to_irq_node(cfg, node, apic, pin)) 456 panic("IO-APIC: failed to add irq-pin. Can not proceed\n"); 457 } 458 459 /* 460 * Reroute an IRQ to a different pin. 461 */ 462 static void __init replace_pin_at_irq_node(struct irq_cfg *cfg, int node, 463 int oldapic, int oldpin, 464 int newapic, int newpin) 465 { 466 struct irq_pin_list *entry; 467 468 for_each_irq_pin(entry, cfg->irq_2_pin) { 469 if (entry->apic == oldapic && entry->pin == oldpin) { 470 entry->apic = newapic; 471 entry->pin = newpin; 472 /* every one is different, right? */ 473 return; 474 } 475 } 476 477 /* old apic/pin didn't exist, so just add new ones */ 478 add_pin_to_irq_node(cfg, node, newapic, newpin); 479 } 480 481 static void __io_apic_modify_irq(struct irq_pin_list *entry, 482 int mask_and, int mask_or, 483 void (*final)(struct irq_pin_list *entry)) 484 { 485 unsigned int reg, pin; 486 487 pin = entry->pin; 488 reg = io_apic_read(entry->apic, 0x10 + pin * 2); 489 reg &= mask_and; 490 reg |= mask_or; 491 io_apic_modify(entry->apic, 0x10 + pin * 2, reg); 492 if (final) 493 final(entry); 494 } 495 496 static void io_apic_modify_irq(struct irq_cfg *cfg, 497 int mask_and, int mask_or, 498 void (*final)(struct irq_pin_list *entry)) 499 { 500 struct irq_pin_list *entry; 501 502 for_each_irq_pin(entry, cfg->irq_2_pin) 503 __io_apic_modify_irq(entry, mask_and, mask_or, final); 504 } 505 506 static void io_apic_sync(struct irq_pin_list *entry) 507 { 508 /* 509 * Synchronize the IO-APIC and the CPU by doing 510 * a dummy read from the IO-APIC 511 */ 512 struct io_apic __iomem *io_apic; 513 514 io_apic = io_apic_base(entry->apic); 515 readl(&io_apic->data); 516 } 517 518 static void mask_ioapic(struct irq_cfg *cfg) 519 { 520 unsigned long flags; 521 522 raw_spin_lock_irqsave(&ioapic_lock, flags); 523 io_apic_modify_irq(cfg, ~0, IO_APIC_REDIR_MASKED, &io_apic_sync); 524 raw_spin_unlock_irqrestore(&ioapic_lock, flags); 525 } 526 527 static void mask_ioapic_irq(struct irq_data *data) 528 { 529 mask_ioapic(data->chip_data); 530 } 531 532 static void __unmask_ioapic(struct irq_cfg *cfg) 533 { 534 io_apic_modify_irq(cfg, ~IO_APIC_REDIR_MASKED, 0, NULL); 535 } 536 537 static void unmask_ioapic(struct irq_cfg *cfg) 538 { 539 unsigned long flags; 540 541 raw_spin_lock_irqsave(&ioapic_lock, flags); 542 __unmask_ioapic(cfg); 543 raw_spin_unlock_irqrestore(&ioapic_lock, flags); 544 } 545 546 static void unmask_ioapic_irq(struct irq_data *data) 547 { 548 unmask_ioapic(data->chip_data); 549 } 550 551 /* 552 * IO-APIC versions below 0x20 don't support EOI register. 553 * For the record, here is the information about various versions: 554 * 0Xh 82489DX 555 * 1Xh I/OAPIC or I/O(x)APIC which are not PCI 2.2 Compliant 556 * 2Xh I/O(x)APIC which is PCI 2.2 Compliant 557 * 30h-FFh Reserved 558 * 559 * Some of the Intel ICH Specs (ICH2 to ICH5) documents the io-apic 560 * version as 0x2. This is an error with documentation and these ICH chips 561 * use io-apic's of version 0x20. 562 * 563 * For IO-APIC's with EOI register, we use that to do an explicit EOI. 564 * Otherwise, we simulate the EOI message manually by changing the trigger 565 * mode to edge and then back to level, with RTE being masked during this. 566 */ 567 void native_eoi_ioapic_pin(int apic, int pin, int vector) 568 { 569 if (mpc_ioapic_ver(apic) >= 0x20) { 570 io_apic_eoi(apic, vector); 571 } else { 572 struct IO_APIC_route_entry entry, entry1; 573 574 entry = entry1 = __ioapic_read_entry(apic, pin); 575 576 /* 577 * Mask the entry and change the trigger mode to edge. 578 */ 579 entry1.mask = 1; 580 entry1.trigger = IOAPIC_EDGE; 581 582 __ioapic_write_entry(apic, pin, entry1); 583 584 /* 585 * Restore the previous level triggered entry. 586 */ 587 __ioapic_write_entry(apic, pin, entry); 588 } 589 } 590 591 void eoi_ioapic_irq(unsigned int irq, struct irq_cfg *cfg) 592 { 593 struct irq_pin_list *entry; 594 unsigned long flags; 595 596 raw_spin_lock_irqsave(&ioapic_lock, flags); 597 for_each_irq_pin(entry, cfg->irq_2_pin) 598 x86_io_apic_ops.eoi_ioapic_pin(entry->apic, entry->pin, 599 cfg->vector); 600 raw_spin_unlock_irqrestore(&ioapic_lock, flags); 601 } 602 603 static void clear_IO_APIC_pin(unsigned int apic, unsigned int pin) 604 { 605 struct IO_APIC_route_entry entry; 606 607 /* Check delivery_mode to be sure we're not clearing an SMI pin */ 608 entry = ioapic_read_entry(apic, pin); 609 if (entry.delivery_mode == dest_SMI) 610 return; 611 612 /* 613 * Make sure the entry is masked and re-read the contents to check 614 * if it is a level triggered pin and if the remote-IRR is set. 615 */ 616 if (!entry.mask) { 617 entry.mask = 1; 618 ioapic_write_entry(apic, pin, entry); 619 entry = ioapic_read_entry(apic, pin); 620 } 621 622 if (entry.irr) { 623 unsigned long flags; 624 625 /* 626 * Make sure the trigger mode is set to level. Explicit EOI 627 * doesn't clear the remote-IRR if the trigger mode is not 628 * set to level. 629 */ 630 if (!entry.trigger) { 631 entry.trigger = IOAPIC_LEVEL; 632 ioapic_write_entry(apic, pin, entry); 633 } 634 635 raw_spin_lock_irqsave(&ioapic_lock, flags); 636 x86_io_apic_ops.eoi_ioapic_pin(apic, pin, entry.vector); 637 raw_spin_unlock_irqrestore(&ioapic_lock, flags); 638 } 639 640 /* 641 * Clear the rest of the bits in the IO-APIC RTE except for the mask 642 * bit. 643 */ 644 ioapic_mask_entry(apic, pin); 645 entry = ioapic_read_entry(apic, pin); 646 if (entry.irr) 647 pr_err("Unable to reset IRR for apic: %d, pin :%d\n", 648 mpc_ioapic_id(apic), pin); 649 } 650 651 static void clear_IO_APIC (void) 652 { 653 int apic, pin; 654 655 for_each_ioapic_pin(apic, pin) 656 clear_IO_APIC_pin(apic, pin); 657 } 658 659 #ifdef CONFIG_X86_32 660 /* 661 * support for broken MP BIOSs, enables hand-redirection of PIRQ0-7 to 662 * specific CPU-side IRQs. 663 */ 664 665 #define MAX_PIRQS 8 666 static int pirq_entries[MAX_PIRQS] = { 667 [0 ... MAX_PIRQS - 1] = -1 668 }; 669 670 static int __init ioapic_pirq_setup(char *str) 671 { 672 int i, max; 673 int ints[MAX_PIRQS+1]; 674 675 get_options(str, ARRAY_SIZE(ints), ints); 676 677 apic_printk(APIC_VERBOSE, KERN_INFO 678 "PIRQ redirection, working around broken MP-BIOS.\n"); 679 max = MAX_PIRQS; 680 if (ints[0] < MAX_PIRQS) 681 max = ints[0]; 682 683 for (i = 0; i < max; i++) { 684 apic_printk(APIC_VERBOSE, KERN_DEBUG 685 "... PIRQ%d -> IRQ %d\n", i, ints[i+1]); 686 /* 687 * PIRQs are mapped upside down, usually. 688 */ 689 pirq_entries[MAX_PIRQS-i-1] = ints[i+1]; 690 } 691 return 1; 692 } 693 694 __setup("pirq=", ioapic_pirq_setup); 695 #endif /* CONFIG_X86_32 */ 696 697 /* 698 * Saves all the IO-APIC RTE's 699 */ 700 int save_ioapic_entries(void) 701 { 702 int apic, pin; 703 int err = 0; 704 705 for_each_ioapic(apic) { 706 if (!ioapics[apic].saved_registers) { 707 err = -ENOMEM; 708 continue; 709 } 710 711 for_each_pin(apic, pin) 712 ioapics[apic].saved_registers[pin] = 713 ioapic_read_entry(apic, pin); 714 } 715 716 return err; 717 } 718 719 /* 720 * Mask all IO APIC entries. 721 */ 722 void mask_ioapic_entries(void) 723 { 724 int apic, pin; 725 726 for_each_ioapic(apic) { 727 if (!ioapics[apic].saved_registers) 728 continue; 729 730 for_each_pin(apic, pin) { 731 struct IO_APIC_route_entry entry; 732 733 entry = ioapics[apic].saved_registers[pin]; 734 if (!entry.mask) { 735 entry.mask = 1; 736 ioapic_write_entry(apic, pin, entry); 737 } 738 } 739 } 740 } 741 742 /* 743 * Restore IO APIC entries which was saved in the ioapic structure. 744 */ 745 int restore_ioapic_entries(void) 746 { 747 int apic, pin; 748 749 for_each_ioapic(apic) { 750 if (!ioapics[apic].saved_registers) 751 continue; 752 753 for_each_pin(apic, pin) 754 ioapic_write_entry(apic, pin, 755 ioapics[apic].saved_registers[pin]); 756 } 757 return 0; 758 } 759 760 /* 761 * Find the IRQ entry number of a certain pin. 762 */ 763 static int find_irq_entry(int ioapic_idx, int pin, int type) 764 { 765 int i; 766 767 for (i = 0; i < mp_irq_entries; i++) 768 if (mp_irqs[i].irqtype == type && 769 (mp_irqs[i].dstapic == mpc_ioapic_id(ioapic_idx) || 770 mp_irqs[i].dstapic == MP_APIC_ALL) && 771 mp_irqs[i].dstirq == pin) 772 return i; 773 774 return -1; 775 } 776 777 /* 778 * Find the pin to which IRQ[irq] (ISA) is connected 779 */ 780 static int __init find_isa_irq_pin(int irq, int type) 781 { 782 int i; 783 784 for (i = 0; i < mp_irq_entries; i++) { 785 int lbus = mp_irqs[i].srcbus; 786 787 if (test_bit(lbus, mp_bus_not_pci) && 788 (mp_irqs[i].irqtype == type) && 789 (mp_irqs[i].srcbusirq == irq)) 790 791 return mp_irqs[i].dstirq; 792 } 793 return -1; 794 } 795 796 static int __init find_isa_irq_apic(int irq, int type) 797 { 798 int i; 799 800 for (i = 0; i < mp_irq_entries; i++) { 801 int lbus = mp_irqs[i].srcbus; 802 803 if (test_bit(lbus, mp_bus_not_pci) && 804 (mp_irqs[i].irqtype == type) && 805 (mp_irqs[i].srcbusirq == irq)) 806 break; 807 } 808 809 if (i < mp_irq_entries) { 810 int ioapic_idx; 811 812 for_each_ioapic(ioapic_idx) 813 if (mpc_ioapic_id(ioapic_idx) == mp_irqs[i].dstapic) 814 return ioapic_idx; 815 } 816 817 return -1; 818 } 819 820 #ifdef CONFIG_EISA 821 /* 822 * EISA Edge/Level control register, ELCR 823 */ 824 static int EISA_ELCR(unsigned int irq) 825 { 826 if (irq < legacy_pic->nr_legacy_irqs) { 827 unsigned int port = 0x4d0 + (irq >> 3); 828 return (inb(port) >> (irq & 7)) & 1; 829 } 830 apic_printk(APIC_VERBOSE, KERN_INFO 831 "Broken MPtable reports ISA irq %d\n", irq); 832 return 0; 833 } 834 835 #endif 836 837 /* ISA interrupts are always polarity zero edge triggered, 838 * when listed as conforming in the MP table. */ 839 840 #define default_ISA_trigger(idx) (0) 841 #define default_ISA_polarity(idx) (0) 842 843 /* EISA interrupts are always polarity zero and can be edge or level 844 * trigger depending on the ELCR value. If an interrupt is listed as 845 * EISA conforming in the MP table, that means its trigger type must 846 * be read in from the ELCR */ 847 848 #define default_EISA_trigger(idx) (EISA_ELCR(mp_irqs[idx].srcbusirq)) 849 #define default_EISA_polarity(idx) default_ISA_polarity(idx) 850 851 /* PCI interrupts are always polarity one level triggered, 852 * when listed as conforming in the MP table. */ 853 854 #define default_PCI_trigger(idx) (1) 855 #define default_PCI_polarity(idx) (1) 856 857 static int irq_polarity(int idx) 858 { 859 int bus = mp_irqs[idx].srcbus; 860 int polarity; 861 862 /* 863 * Determine IRQ line polarity (high active or low active): 864 */ 865 switch (mp_irqs[idx].irqflag & 3) 866 { 867 case 0: /* conforms, ie. bus-type dependent polarity */ 868 if (test_bit(bus, mp_bus_not_pci)) 869 polarity = default_ISA_polarity(idx); 870 else 871 polarity = default_PCI_polarity(idx); 872 break; 873 case 1: /* high active */ 874 { 875 polarity = 0; 876 break; 877 } 878 case 2: /* reserved */ 879 { 880 pr_warn("broken BIOS!!\n"); 881 polarity = 1; 882 break; 883 } 884 case 3: /* low active */ 885 { 886 polarity = 1; 887 break; 888 } 889 default: /* invalid */ 890 { 891 pr_warn("broken BIOS!!\n"); 892 polarity = 1; 893 break; 894 } 895 } 896 return polarity; 897 } 898 899 static int irq_trigger(int idx) 900 { 901 int bus = mp_irqs[idx].srcbus; 902 int trigger; 903 904 /* 905 * Determine IRQ trigger mode (edge or level sensitive): 906 */ 907 switch ((mp_irqs[idx].irqflag>>2) & 3) 908 { 909 case 0: /* conforms, ie. bus-type dependent */ 910 if (test_bit(bus, mp_bus_not_pci)) 911 trigger = default_ISA_trigger(idx); 912 else 913 trigger = default_PCI_trigger(idx); 914 #ifdef CONFIG_EISA 915 switch (mp_bus_id_to_type[bus]) { 916 case MP_BUS_ISA: /* ISA pin */ 917 { 918 /* set before the switch */ 919 break; 920 } 921 case MP_BUS_EISA: /* EISA pin */ 922 { 923 trigger = default_EISA_trigger(idx); 924 break; 925 } 926 case MP_BUS_PCI: /* PCI pin */ 927 { 928 /* set before the switch */ 929 break; 930 } 931 default: 932 { 933 pr_warn("broken BIOS!!\n"); 934 trigger = 1; 935 break; 936 } 937 } 938 #endif 939 break; 940 case 1: /* edge */ 941 { 942 trigger = 0; 943 break; 944 } 945 case 2: /* reserved */ 946 { 947 pr_warn("broken BIOS!!\n"); 948 trigger = 1; 949 break; 950 } 951 case 3: /* level */ 952 { 953 trigger = 1; 954 break; 955 } 956 default: /* invalid */ 957 { 958 pr_warn("broken BIOS!!\n"); 959 trigger = 0; 960 break; 961 } 962 } 963 return trigger; 964 } 965 966 static int pin_2_irq(int idx, int apic, int pin) 967 { 968 int irq; 969 int bus = mp_irqs[idx].srcbus; 970 struct mp_ioapic_gsi *gsi_cfg = mp_ioapic_gsi_routing(apic); 971 972 /* 973 * Debugging check, we are in big trouble if this message pops up! 974 */ 975 if (mp_irqs[idx].dstirq != pin) 976 pr_err("broken BIOS or MPTABLE parser, ayiee!!\n"); 977 978 if (test_bit(bus, mp_bus_not_pci)) { 979 irq = mp_irqs[idx].srcbusirq; 980 } else { 981 u32 gsi = gsi_cfg->gsi_base + pin; 982 983 if (gsi >= NR_IRQS_LEGACY) 984 irq = gsi; 985 else 986 irq = gsi_top + gsi; 987 } 988 989 #ifdef CONFIG_X86_32 990 /* 991 * PCI IRQ command line redirection. Yes, limits are hardcoded. 992 */ 993 if ((pin >= 16) && (pin <= 23)) { 994 if (pirq_entries[pin-16] != -1) { 995 if (!pirq_entries[pin-16]) { 996 apic_printk(APIC_VERBOSE, KERN_DEBUG 997 "disabling PIRQ%d\n", pin-16); 998 } else { 999 irq = pirq_entries[pin-16]; 1000 apic_printk(APIC_VERBOSE, KERN_DEBUG 1001 "using PIRQ%d -> IRQ %d\n", 1002 pin-16, irq); 1003 } 1004 } 1005 } 1006 #endif 1007 1008 return irq; 1009 } 1010 1011 /* 1012 * Find a specific PCI IRQ entry. 1013 * Not an __init, possibly needed by modules 1014 */ 1015 int IO_APIC_get_PCI_irq_vector(int bus, int slot, int pin, 1016 struct io_apic_irq_attr *irq_attr) 1017 { 1018 int irq, i, best_guess = -1; 1019 1020 apic_printk(APIC_DEBUG, 1021 "querying PCI -> IRQ mapping bus:%d, slot:%d, pin:%d.\n", 1022 bus, slot, pin); 1023 if (test_bit(bus, mp_bus_not_pci)) { 1024 apic_printk(APIC_VERBOSE, 1025 "PCI BIOS passed nonexistent PCI bus %d!\n", bus); 1026 return -1; 1027 } 1028 1029 for (i = 0; i < mp_irq_entries; i++) { 1030 int lbus = mp_irqs[i].srcbus; 1031 int ioapic_idx, found = 0; 1032 1033 if (bus != lbus || mp_irqs[i].irqtype != mp_INT || 1034 slot != ((mp_irqs[i].srcbusirq >> 2) & 0x1f)) 1035 continue; 1036 1037 for_each_ioapic(ioapic_idx) 1038 if (mpc_ioapic_id(ioapic_idx) == mp_irqs[i].dstapic || 1039 mp_irqs[i].dstapic == MP_APIC_ALL) { 1040 found = 1; 1041 break; 1042 } 1043 if (!found) 1044 continue; 1045 1046 /* Skip ISA IRQs */ 1047 irq = pin_2_irq(i, ioapic_idx, mp_irqs[i].dstirq); 1048 if (ioapic_idx == 0 && !IO_APIC_IRQ(irq)) 1049 continue; 1050 1051 if (pin == (mp_irqs[i].srcbusirq & 3)) { 1052 set_io_apic_irq_attr(irq_attr, ioapic_idx, 1053 mp_irqs[i].dstirq, 1054 irq_trigger(i), 1055 irq_polarity(i)); 1056 return irq; 1057 } 1058 /* 1059 * Use the first all-but-pin matching entry as a 1060 * best-guess fuzzy result for broken mptables. 1061 */ 1062 if (best_guess < 0) { 1063 set_io_apic_irq_attr(irq_attr, ioapic_idx, 1064 mp_irqs[i].dstirq, 1065 irq_trigger(i), 1066 irq_polarity(i)); 1067 best_guess = irq; 1068 } 1069 } 1070 return best_guess; 1071 } 1072 EXPORT_SYMBOL(IO_APIC_get_PCI_irq_vector); 1073 1074 void lock_vector_lock(void) 1075 { 1076 /* Used to the online set of cpus does not change 1077 * during assign_irq_vector. 1078 */ 1079 raw_spin_lock(&vector_lock); 1080 } 1081 1082 void unlock_vector_lock(void) 1083 { 1084 raw_spin_unlock(&vector_lock); 1085 } 1086 1087 static int 1088 __assign_irq_vector(int irq, struct irq_cfg *cfg, const struct cpumask *mask) 1089 { 1090 /* 1091 * NOTE! The local APIC isn't very good at handling 1092 * multiple interrupts at the same interrupt level. 1093 * As the interrupt level is determined by taking the 1094 * vector number and shifting that right by 4, we 1095 * want to spread these out a bit so that they don't 1096 * all fall in the same interrupt level. 1097 * 1098 * Also, we've got to be careful not to trash gate 1099 * 0x80, because int 0x80 is hm, kind of importantish. ;) 1100 */ 1101 static int current_vector = FIRST_EXTERNAL_VECTOR + VECTOR_OFFSET_START; 1102 static int current_offset = VECTOR_OFFSET_START % 16; 1103 int cpu, err; 1104 cpumask_var_t tmp_mask; 1105 1106 if (cfg->move_in_progress) 1107 return -EBUSY; 1108 1109 if (!alloc_cpumask_var(&tmp_mask, GFP_ATOMIC)) 1110 return -ENOMEM; 1111 1112 /* Only try and allocate irqs on cpus that are present */ 1113 err = -ENOSPC; 1114 cpumask_clear(cfg->old_domain); 1115 cpu = cpumask_first_and(mask, cpu_online_mask); 1116 while (cpu < nr_cpu_ids) { 1117 int new_cpu, vector, offset; 1118 1119 apic->vector_allocation_domain(cpu, tmp_mask, mask); 1120 1121 if (cpumask_subset(tmp_mask, cfg->domain)) { 1122 err = 0; 1123 if (cpumask_equal(tmp_mask, cfg->domain)) 1124 break; 1125 /* 1126 * New cpumask using the vector is a proper subset of 1127 * the current in use mask. So cleanup the vector 1128 * allocation for the members that are not used anymore. 1129 */ 1130 cpumask_andnot(cfg->old_domain, cfg->domain, tmp_mask); 1131 cfg->move_in_progress = 1132 cpumask_intersects(cfg->old_domain, cpu_online_mask); 1133 cpumask_and(cfg->domain, cfg->domain, tmp_mask); 1134 break; 1135 } 1136 1137 vector = current_vector; 1138 offset = current_offset; 1139 next: 1140 vector += 16; 1141 if (vector >= first_system_vector) { 1142 offset = (offset + 1) % 16; 1143 vector = FIRST_EXTERNAL_VECTOR + offset; 1144 } 1145 1146 if (unlikely(current_vector == vector)) { 1147 cpumask_or(cfg->old_domain, cfg->old_domain, tmp_mask); 1148 cpumask_andnot(tmp_mask, mask, cfg->old_domain); 1149 cpu = cpumask_first_and(tmp_mask, cpu_online_mask); 1150 continue; 1151 } 1152 1153 if (test_bit(vector, used_vectors)) 1154 goto next; 1155 1156 for_each_cpu_and(new_cpu, tmp_mask, cpu_online_mask) { 1157 if (per_cpu(vector_irq, new_cpu)[vector] > VECTOR_UNDEFINED) 1158 goto next; 1159 } 1160 /* Found one! */ 1161 current_vector = vector; 1162 current_offset = offset; 1163 if (cfg->vector) { 1164 cpumask_copy(cfg->old_domain, cfg->domain); 1165 cfg->move_in_progress = 1166 cpumask_intersects(cfg->old_domain, cpu_online_mask); 1167 } 1168 for_each_cpu_and(new_cpu, tmp_mask, cpu_online_mask) 1169 per_cpu(vector_irq, new_cpu)[vector] = irq; 1170 cfg->vector = vector; 1171 cpumask_copy(cfg->domain, tmp_mask); 1172 err = 0; 1173 break; 1174 } 1175 free_cpumask_var(tmp_mask); 1176 return err; 1177 } 1178 1179 int assign_irq_vector(int irq, struct irq_cfg *cfg, const struct cpumask *mask) 1180 { 1181 int err; 1182 unsigned long flags; 1183 1184 raw_spin_lock_irqsave(&vector_lock, flags); 1185 err = __assign_irq_vector(irq, cfg, mask); 1186 raw_spin_unlock_irqrestore(&vector_lock, flags); 1187 return err; 1188 } 1189 1190 static void __clear_irq_vector(int irq, struct irq_cfg *cfg) 1191 { 1192 int cpu, vector; 1193 1194 BUG_ON(!cfg->vector); 1195 1196 vector = cfg->vector; 1197 for_each_cpu_and(cpu, cfg->domain, cpu_online_mask) 1198 per_cpu(vector_irq, cpu)[vector] = VECTOR_UNDEFINED; 1199 1200 cfg->vector = 0; 1201 cpumask_clear(cfg->domain); 1202 1203 if (likely(!cfg->move_in_progress)) 1204 return; 1205 for_each_cpu_and(cpu, cfg->old_domain, cpu_online_mask) { 1206 for (vector = FIRST_EXTERNAL_VECTOR; vector < NR_VECTORS; vector++) { 1207 if (per_cpu(vector_irq, cpu)[vector] != irq) 1208 continue; 1209 per_cpu(vector_irq, cpu)[vector] = VECTOR_UNDEFINED; 1210 break; 1211 } 1212 } 1213 cfg->move_in_progress = 0; 1214 } 1215 1216 void __setup_vector_irq(int cpu) 1217 { 1218 /* Initialize vector_irq on a new cpu */ 1219 int irq, vector; 1220 struct irq_cfg *cfg; 1221 1222 /* 1223 * vector_lock will make sure that we don't run into irq vector 1224 * assignments that might be happening on another cpu in parallel, 1225 * while we setup our initial vector to irq mappings. 1226 */ 1227 raw_spin_lock(&vector_lock); 1228 /* Mark the inuse vectors */ 1229 for_each_active_irq(irq) { 1230 cfg = irq_cfg(irq); 1231 if (!cfg) 1232 continue; 1233 1234 if (!cpumask_test_cpu(cpu, cfg->domain)) 1235 continue; 1236 vector = cfg->vector; 1237 per_cpu(vector_irq, cpu)[vector] = irq; 1238 } 1239 /* Mark the free vectors */ 1240 for (vector = 0; vector < NR_VECTORS; ++vector) { 1241 irq = per_cpu(vector_irq, cpu)[vector]; 1242 if (irq <= VECTOR_UNDEFINED) 1243 continue; 1244 1245 cfg = irq_cfg(irq); 1246 if (!cpumask_test_cpu(cpu, cfg->domain)) 1247 per_cpu(vector_irq, cpu)[vector] = VECTOR_UNDEFINED; 1248 } 1249 raw_spin_unlock(&vector_lock); 1250 } 1251 1252 static struct irq_chip ioapic_chip; 1253 1254 #ifdef CONFIG_X86_32 1255 static inline int IO_APIC_irq_trigger(int irq) 1256 { 1257 int apic, idx, pin; 1258 1259 for_each_ioapic_pin(apic, pin) { 1260 idx = find_irq_entry(apic, pin, mp_INT); 1261 if ((idx != -1) && (irq == pin_2_irq(idx, apic, pin))) 1262 return irq_trigger(idx); 1263 } 1264 /* 1265 * nonexistent IRQs are edge default 1266 */ 1267 return 0; 1268 } 1269 #else 1270 static inline int IO_APIC_irq_trigger(int irq) 1271 { 1272 return 1; 1273 } 1274 #endif 1275 1276 static void ioapic_register_intr(unsigned int irq, struct irq_cfg *cfg, 1277 unsigned long trigger) 1278 { 1279 struct irq_chip *chip = &ioapic_chip; 1280 irq_flow_handler_t hdl; 1281 bool fasteoi; 1282 1283 if ((trigger == IOAPIC_AUTO && IO_APIC_irq_trigger(irq)) || 1284 trigger == IOAPIC_LEVEL) { 1285 irq_set_status_flags(irq, IRQ_LEVEL); 1286 fasteoi = true; 1287 } else { 1288 irq_clear_status_flags(irq, IRQ_LEVEL); 1289 fasteoi = false; 1290 } 1291 1292 if (setup_remapped_irq(irq, cfg, chip)) 1293 fasteoi = trigger != 0; 1294 1295 hdl = fasteoi ? handle_fasteoi_irq : handle_edge_irq; 1296 irq_set_chip_and_handler_name(irq, chip, hdl, 1297 fasteoi ? "fasteoi" : "edge"); 1298 } 1299 1300 int native_setup_ioapic_entry(int irq, struct IO_APIC_route_entry *entry, 1301 unsigned int destination, int vector, 1302 struct io_apic_irq_attr *attr) 1303 { 1304 memset(entry, 0, sizeof(*entry)); 1305 1306 entry->delivery_mode = apic->irq_delivery_mode; 1307 entry->dest_mode = apic->irq_dest_mode; 1308 entry->dest = destination; 1309 entry->vector = vector; 1310 entry->mask = 0; /* enable IRQ */ 1311 entry->trigger = attr->trigger; 1312 entry->polarity = attr->polarity; 1313 1314 /* 1315 * Mask level triggered irqs. 1316 * Use IRQ_DELAYED_DISABLE for edge triggered irqs. 1317 */ 1318 if (attr->trigger) 1319 entry->mask = 1; 1320 1321 return 0; 1322 } 1323 1324 static void setup_ioapic_irq(unsigned int irq, struct irq_cfg *cfg, 1325 struct io_apic_irq_attr *attr) 1326 { 1327 struct IO_APIC_route_entry entry; 1328 unsigned int dest; 1329 1330 if (!IO_APIC_IRQ(irq)) 1331 return; 1332 1333 if (assign_irq_vector(irq, cfg, apic->target_cpus())) 1334 return; 1335 1336 if (apic->cpu_mask_to_apicid_and(cfg->domain, apic->target_cpus(), 1337 &dest)) { 1338 pr_warn("Failed to obtain apicid for ioapic %d, pin %d\n", 1339 mpc_ioapic_id(attr->ioapic), attr->ioapic_pin); 1340 __clear_irq_vector(irq, cfg); 1341 1342 return; 1343 } 1344 1345 apic_printk(APIC_VERBOSE,KERN_DEBUG 1346 "IOAPIC[%d]: Set routing entry (%d-%d -> 0x%x -> " 1347 "IRQ %d Mode:%i Active:%i Dest:%d)\n", 1348 attr->ioapic, mpc_ioapic_id(attr->ioapic), attr->ioapic_pin, 1349 cfg->vector, irq, attr->trigger, attr->polarity, dest); 1350 1351 if (x86_io_apic_ops.setup_entry(irq, &entry, dest, cfg->vector, attr)) { 1352 pr_warn("Failed to setup ioapic entry for ioapic %d, pin %d\n", 1353 mpc_ioapic_id(attr->ioapic), attr->ioapic_pin); 1354 __clear_irq_vector(irq, cfg); 1355 1356 return; 1357 } 1358 1359 ioapic_register_intr(irq, cfg, attr->trigger); 1360 if (irq < legacy_pic->nr_legacy_irqs) 1361 legacy_pic->mask(irq); 1362 1363 ioapic_write_entry(attr->ioapic, attr->ioapic_pin, entry); 1364 } 1365 1366 static bool __init io_apic_pin_not_connected(int idx, int ioapic_idx, int pin) 1367 { 1368 if (idx != -1) 1369 return false; 1370 1371 apic_printk(APIC_VERBOSE, KERN_DEBUG " apic %d pin %d not connected\n", 1372 mpc_ioapic_id(ioapic_idx), pin); 1373 return true; 1374 } 1375 1376 static void __init __io_apic_setup_irqs(unsigned int ioapic_idx) 1377 { 1378 int idx, node = cpu_to_node(0); 1379 struct io_apic_irq_attr attr; 1380 unsigned int pin, irq; 1381 1382 for_each_pin(ioapic_idx, pin) { 1383 idx = find_irq_entry(ioapic_idx, pin, mp_INT); 1384 if (io_apic_pin_not_connected(idx, ioapic_idx, pin)) 1385 continue; 1386 1387 irq = pin_2_irq(idx, ioapic_idx, pin); 1388 if (!mp_init_irq_at_boot(ioapic_idx, irq)) 1389 continue; 1390 1391 /* 1392 * Skip the timer IRQ if there's a quirk handler 1393 * installed and if it returns 1: 1394 */ 1395 if (apic->multi_timer_check && 1396 apic->multi_timer_check(ioapic_idx, irq)) 1397 continue; 1398 1399 set_io_apic_irq_attr(&attr, ioapic_idx, pin, irq_trigger(idx), 1400 irq_polarity(idx)); 1401 1402 io_apic_setup_irq_pin(irq, node, &attr); 1403 } 1404 } 1405 1406 static void __init setup_IO_APIC_irqs(void) 1407 { 1408 unsigned int ioapic_idx; 1409 1410 apic_printk(APIC_VERBOSE, KERN_DEBUG "init IO_APIC IRQs\n"); 1411 1412 for_each_ioapic(ioapic_idx) 1413 __io_apic_setup_irqs(ioapic_idx); 1414 } 1415 1416 /* 1417 * for the gsi that is not in first ioapic 1418 * but could not use acpi_register_gsi() 1419 * like some special sci in IBM x3330 1420 */ 1421 void setup_IO_APIC_irq_extra(u32 gsi) 1422 { 1423 int ioapic_idx = 0, pin, idx, irq, node = cpu_to_node(0); 1424 struct io_apic_irq_attr attr; 1425 1426 /* 1427 * Convert 'gsi' to 'ioapic.pin'. 1428 */ 1429 ioapic_idx = mp_find_ioapic(gsi); 1430 if (ioapic_idx < 0) 1431 return; 1432 1433 pin = mp_find_ioapic_pin(ioapic_idx, gsi); 1434 idx = find_irq_entry(ioapic_idx, pin, mp_INT); 1435 if (idx == -1) 1436 return; 1437 1438 irq = pin_2_irq(idx, ioapic_idx, pin); 1439 if (mp_init_irq_at_boot(ioapic_idx, irq)) 1440 return; 1441 1442 set_io_apic_irq_attr(&attr, ioapic_idx, pin, irq_trigger(idx), 1443 irq_polarity(idx)); 1444 1445 io_apic_setup_irq_pin_once(irq, node, &attr); 1446 } 1447 1448 /* 1449 * Set up the timer pin, possibly with the 8259A-master behind. 1450 */ 1451 static void __init setup_timer_IRQ0_pin(unsigned int ioapic_idx, 1452 unsigned int pin, int vector) 1453 { 1454 struct IO_APIC_route_entry entry; 1455 unsigned int dest; 1456 1457 memset(&entry, 0, sizeof(entry)); 1458 1459 /* 1460 * We use logical delivery to get the timer IRQ 1461 * to the first CPU. 1462 */ 1463 if (unlikely(apic->cpu_mask_to_apicid_and(apic->target_cpus(), 1464 apic->target_cpus(), &dest))) 1465 dest = BAD_APICID; 1466 1467 entry.dest_mode = apic->irq_dest_mode; 1468 entry.mask = 0; /* don't mask IRQ for edge */ 1469 entry.dest = dest; 1470 entry.delivery_mode = apic->irq_delivery_mode; 1471 entry.polarity = 0; 1472 entry.trigger = 0; 1473 entry.vector = vector; 1474 1475 /* 1476 * The timer IRQ doesn't have to know that behind the 1477 * scene we may have a 8259A-master in AEOI mode ... 1478 */ 1479 irq_set_chip_and_handler_name(0, &ioapic_chip, handle_edge_irq, 1480 "edge"); 1481 1482 /* 1483 * Add it to the IO-APIC irq-routing table: 1484 */ 1485 ioapic_write_entry(ioapic_idx, pin, entry); 1486 } 1487 1488 void native_io_apic_print_entries(unsigned int apic, unsigned int nr_entries) 1489 { 1490 int i; 1491 1492 pr_debug(" NR Dst Mask Trig IRR Pol Stat Dmod Deli Vect:\n"); 1493 1494 for (i = 0; i <= nr_entries; i++) { 1495 struct IO_APIC_route_entry entry; 1496 1497 entry = ioapic_read_entry(apic, i); 1498 1499 pr_debug(" %02x %02X ", i, entry.dest); 1500 pr_cont("%1d %1d %1d %1d %1d " 1501 "%1d %1d %02X\n", 1502 entry.mask, 1503 entry.trigger, 1504 entry.irr, 1505 entry.polarity, 1506 entry.delivery_status, 1507 entry.dest_mode, 1508 entry.delivery_mode, 1509 entry.vector); 1510 } 1511 } 1512 1513 void intel_ir_io_apic_print_entries(unsigned int apic, 1514 unsigned int nr_entries) 1515 { 1516 int i; 1517 1518 pr_debug(" NR Indx Fmt Mask Trig IRR Pol Stat Indx2 Zero Vect:\n"); 1519 1520 for (i = 0; i <= nr_entries; i++) { 1521 struct IR_IO_APIC_route_entry *ir_entry; 1522 struct IO_APIC_route_entry entry; 1523 1524 entry = ioapic_read_entry(apic, i); 1525 1526 ir_entry = (struct IR_IO_APIC_route_entry *)&entry; 1527 1528 pr_debug(" %02x %04X ", i, ir_entry->index); 1529 pr_cont("%1d %1d %1d %1d %1d " 1530 "%1d %1d %X %02X\n", 1531 ir_entry->format, 1532 ir_entry->mask, 1533 ir_entry->trigger, 1534 ir_entry->irr, 1535 ir_entry->polarity, 1536 ir_entry->delivery_status, 1537 ir_entry->index2, 1538 ir_entry->zero, 1539 ir_entry->vector); 1540 } 1541 } 1542 1543 void ioapic_zap_locks(void) 1544 { 1545 raw_spin_lock_init(&ioapic_lock); 1546 } 1547 1548 __apicdebuginit(void) print_IO_APIC(int ioapic_idx) 1549 { 1550 union IO_APIC_reg_00 reg_00; 1551 union IO_APIC_reg_01 reg_01; 1552 union IO_APIC_reg_02 reg_02; 1553 union IO_APIC_reg_03 reg_03; 1554 unsigned long flags; 1555 1556 raw_spin_lock_irqsave(&ioapic_lock, flags); 1557 reg_00.raw = io_apic_read(ioapic_idx, 0); 1558 reg_01.raw = io_apic_read(ioapic_idx, 1); 1559 if (reg_01.bits.version >= 0x10) 1560 reg_02.raw = io_apic_read(ioapic_idx, 2); 1561 if (reg_01.bits.version >= 0x20) 1562 reg_03.raw = io_apic_read(ioapic_idx, 3); 1563 raw_spin_unlock_irqrestore(&ioapic_lock, flags); 1564 1565 printk(KERN_DEBUG "IO APIC #%d......\n", mpc_ioapic_id(ioapic_idx)); 1566 printk(KERN_DEBUG ".... register #00: %08X\n", reg_00.raw); 1567 printk(KERN_DEBUG "....... : physical APIC id: %02X\n", reg_00.bits.ID); 1568 printk(KERN_DEBUG "....... : Delivery Type: %X\n", reg_00.bits.delivery_type); 1569 printk(KERN_DEBUG "....... : LTS : %X\n", reg_00.bits.LTS); 1570 1571 printk(KERN_DEBUG ".... register #01: %08X\n", *(int *)®_01); 1572 printk(KERN_DEBUG "....... : max redirection entries: %02X\n", 1573 reg_01.bits.entries); 1574 1575 printk(KERN_DEBUG "....... : PRQ implemented: %X\n", reg_01.bits.PRQ); 1576 printk(KERN_DEBUG "....... : IO APIC version: %02X\n", 1577 reg_01.bits.version); 1578 1579 /* 1580 * Some Intel chipsets with IO APIC VERSION of 0x1? don't have reg_02, 1581 * but the value of reg_02 is read as the previous read register 1582 * value, so ignore it if reg_02 == reg_01. 1583 */ 1584 if (reg_01.bits.version >= 0x10 && reg_02.raw != reg_01.raw) { 1585 printk(KERN_DEBUG ".... register #02: %08X\n", reg_02.raw); 1586 printk(KERN_DEBUG "....... : arbitration: %02X\n", reg_02.bits.arbitration); 1587 } 1588 1589 /* 1590 * Some Intel chipsets with IO APIC VERSION of 0x2? don't have reg_02 1591 * or reg_03, but the value of reg_0[23] is read as the previous read 1592 * register value, so ignore it if reg_03 == reg_0[12]. 1593 */ 1594 if (reg_01.bits.version >= 0x20 && reg_03.raw != reg_02.raw && 1595 reg_03.raw != reg_01.raw) { 1596 printk(KERN_DEBUG ".... register #03: %08X\n", reg_03.raw); 1597 printk(KERN_DEBUG "....... : Boot DT : %X\n", reg_03.bits.boot_DT); 1598 } 1599 1600 printk(KERN_DEBUG ".... IRQ redirection table:\n"); 1601 1602 x86_io_apic_ops.print_entries(ioapic_idx, reg_01.bits.entries); 1603 } 1604 1605 __apicdebuginit(void) print_IO_APICs(void) 1606 { 1607 int ioapic_idx; 1608 struct irq_cfg *cfg; 1609 unsigned int irq; 1610 struct irq_chip *chip; 1611 1612 printk(KERN_DEBUG "number of MP IRQ sources: %d.\n", mp_irq_entries); 1613 for_each_ioapic(ioapic_idx) 1614 printk(KERN_DEBUG "number of IO-APIC #%d registers: %d.\n", 1615 mpc_ioapic_id(ioapic_idx), 1616 ioapics[ioapic_idx].nr_registers); 1617 1618 /* 1619 * We are a bit conservative about what we expect. We have to 1620 * know about every hardware change ASAP. 1621 */ 1622 printk(KERN_INFO "testing the IO APIC.......................\n"); 1623 1624 for_each_ioapic(ioapic_idx) 1625 print_IO_APIC(ioapic_idx); 1626 1627 printk(KERN_DEBUG "IRQ to pin mappings:\n"); 1628 for_each_active_irq(irq) { 1629 struct irq_pin_list *entry; 1630 1631 chip = irq_get_chip(irq); 1632 if (chip != &ioapic_chip) 1633 continue; 1634 1635 cfg = irq_cfg(irq); 1636 if (!cfg) 1637 continue; 1638 entry = cfg->irq_2_pin; 1639 if (!entry) 1640 continue; 1641 printk(KERN_DEBUG "IRQ%d ", irq); 1642 for_each_irq_pin(entry, cfg->irq_2_pin) 1643 pr_cont("-> %d:%d", entry->apic, entry->pin); 1644 pr_cont("\n"); 1645 } 1646 1647 printk(KERN_INFO ".................................... done.\n"); 1648 } 1649 1650 __apicdebuginit(void) print_APIC_field(int base) 1651 { 1652 int i; 1653 1654 printk(KERN_DEBUG); 1655 1656 for (i = 0; i < 8; i++) 1657 pr_cont("%08x", apic_read(base + i*0x10)); 1658 1659 pr_cont("\n"); 1660 } 1661 1662 __apicdebuginit(void) print_local_APIC(void *dummy) 1663 { 1664 unsigned int i, v, ver, maxlvt; 1665 u64 icr; 1666 1667 printk(KERN_DEBUG "printing local APIC contents on CPU#%d/%d:\n", 1668 smp_processor_id(), hard_smp_processor_id()); 1669 v = apic_read(APIC_ID); 1670 printk(KERN_INFO "... APIC ID: %08x (%01x)\n", v, read_apic_id()); 1671 v = apic_read(APIC_LVR); 1672 printk(KERN_INFO "... APIC VERSION: %08x\n", v); 1673 ver = GET_APIC_VERSION(v); 1674 maxlvt = lapic_get_maxlvt(); 1675 1676 v = apic_read(APIC_TASKPRI); 1677 printk(KERN_DEBUG "... APIC TASKPRI: %08x (%02x)\n", v, v & APIC_TPRI_MASK); 1678 1679 if (APIC_INTEGRATED(ver)) { /* !82489DX */ 1680 if (!APIC_XAPIC(ver)) { 1681 v = apic_read(APIC_ARBPRI); 1682 printk(KERN_DEBUG "... APIC ARBPRI: %08x (%02x)\n", v, 1683 v & APIC_ARBPRI_MASK); 1684 } 1685 v = apic_read(APIC_PROCPRI); 1686 printk(KERN_DEBUG "... APIC PROCPRI: %08x\n", v); 1687 } 1688 1689 /* 1690 * Remote read supported only in the 82489DX and local APIC for 1691 * Pentium processors. 1692 */ 1693 if (!APIC_INTEGRATED(ver) || maxlvt == 3) { 1694 v = apic_read(APIC_RRR); 1695 printk(KERN_DEBUG "... APIC RRR: %08x\n", v); 1696 } 1697 1698 v = apic_read(APIC_LDR); 1699 printk(KERN_DEBUG "... APIC LDR: %08x\n", v); 1700 if (!x2apic_enabled()) { 1701 v = apic_read(APIC_DFR); 1702 printk(KERN_DEBUG "... APIC DFR: %08x\n", v); 1703 } 1704 v = apic_read(APIC_SPIV); 1705 printk(KERN_DEBUG "... APIC SPIV: %08x\n", v); 1706 1707 printk(KERN_DEBUG "... APIC ISR field:\n"); 1708 print_APIC_field(APIC_ISR); 1709 printk(KERN_DEBUG "... APIC TMR field:\n"); 1710 print_APIC_field(APIC_TMR); 1711 printk(KERN_DEBUG "... APIC IRR field:\n"); 1712 print_APIC_field(APIC_IRR); 1713 1714 if (APIC_INTEGRATED(ver)) { /* !82489DX */ 1715 if (maxlvt > 3) /* Due to the Pentium erratum 3AP. */ 1716 apic_write(APIC_ESR, 0); 1717 1718 v = apic_read(APIC_ESR); 1719 printk(KERN_DEBUG "... APIC ESR: %08x\n", v); 1720 } 1721 1722 icr = apic_icr_read(); 1723 printk(KERN_DEBUG "... APIC ICR: %08x\n", (u32)icr); 1724 printk(KERN_DEBUG "... APIC ICR2: %08x\n", (u32)(icr >> 32)); 1725 1726 v = apic_read(APIC_LVTT); 1727 printk(KERN_DEBUG "... APIC LVTT: %08x\n", v); 1728 1729 if (maxlvt > 3) { /* PC is LVT#4. */ 1730 v = apic_read(APIC_LVTPC); 1731 printk(KERN_DEBUG "... APIC LVTPC: %08x\n", v); 1732 } 1733 v = apic_read(APIC_LVT0); 1734 printk(KERN_DEBUG "... APIC LVT0: %08x\n", v); 1735 v = apic_read(APIC_LVT1); 1736 printk(KERN_DEBUG "... APIC LVT1: %08x\n", v); 1737 1738 if (maxlvt > 2) { /* ERR is LVT#3. */ 1739 v = apic_read(APIC_LVTERR); 1740 printk(KERN_DEBUG "... APIC LVTERR: %08x\n", v); 1741 } 1742 1743 v = apic_read(APIC_TMICT); 1744 printk(KERN_DEBUG "... APIC TMICT: %08x\n", v); 1745 v = apic_read(APIC_TMCCT); 1746 printk(KERN_DEBUG "... APIC TMCCT: %08x\n", v); 1747 v = apic_read(APIC_TDCR); 1748 printk(KERN_DEBUG "... APIC TDCR: %08x\n", v); 1749 1750 if (boot_cpu_has(X86_FEATURE_EXTAPIC)) { 1751 v = apic_read(APIC_EFEAT); 1752 maxlvt = (v >> 16) & 0xff; 1753 printk(KERN_DEBUG "... APIC EFEAT: %08x\n", v); 1754 v = apic_read(APIC_ECTRL); 1755 printk(KERN_DEBUG "... APIC ECTRL: %08x\n", v); 1756 for (i = 0; i < maxlvt; i++) { 1757 v = apic_read(APIC_EILVTn(i)); 1758 printk(KERN_DEBUG "... APIC EILVT%d: %08x\n", i, v); 1759 } 1760 } 1761 pr_cont("\n"); 1762 } 1763 1764 __apicdebuginit(void) print_local_APICs(int maxcpu) 1765 { 1766 int cpu; 1767 1768 if (!maxcpu) 1769 return; 1770 1771 preempt_disable(); 1772 for_each_online_cpu(cpu) { 1773 if (cpu >= maxcpu) 1774 break; 1775 smp_call_function_single(cpu, print_local_APIC, NULL, 1); 1776 } 1777 preempt_enable(); 1778 } 1779 1780 __apicdebuginit(void) print_PIC(void) 1781 { 1782 unsigned int v; 1783 unsigned long flags; 1784 1785 if (!legacy_pic->nr_legacy_irqs) 1786 return; 1787 1788 printk(KERN_DEBUG "\nprinting PIC contents\n"); 1789 1790 raw_spin_lock_irqsave(&i8259A_lock, flags); 1791 1792 v = inb(0xa1) << 8 | inb(0x21); 1793 printk(KERN_DEBUG "... PIC IMR: %04x\n", v); 1794 1795 v = inb(0xa0) << 8 | inb(0x20); 1796 printk(KERN_DEBUG "... PIC IRR: %04x\n", v); 1797 1798 outb(0x0b,0xa0); 1799 outb(0x0b,0x20); 1800 v = inb(0xa0) << 8 | inb(0x20); 1801 outb(0x0a,0xa0); 1802 outb(0x0a,0x20); 1803 1804 raw_spin_unlock_irqrestore(&i8259A_lock, flags); 1805 1806 printk(KERN_DEBUG "... PIC ISR: %04x\n", v); 1807 1808 v = inb(0x4d1) << 8 | inb(0x4d0); 1809 printk(KERN_DEBUG "... PIC ELCR: %04x\n", v); 1810 } 1811 1812 static int __initdata show_lapic = 1; 1813 static __init int setup_show_lapic(char *arg) 1814 { 1815 int num = -1; 1816 1817 if (strcmp(arg, "all") == 0) { 1818 show_lapic = CONFIG_NR_CPUS; 1819 } else { 1820 get_option(&arg, &num); 1821 if (num >= 0) 1822 show_lapic = num; 1823 } 1824 1825 return 1; 1826 } 1827 __setup("show_lapic=", setup_show_lapic); 1828 1829 __apicdebuginit(int) print_ICs(void) 1830 { 1831 if (apic_verbosity == APIC_QUIET) 1832 return 0; 1833 1834 print_PIC(); 1835 1836 /* don't print out if apic is not there */ 1837 if (!cpu_has_apic && !apic_from_smp_config()) 1838 return 0; 1839 1840 print_local_APICs(show_lapic); 1841 print_IO_APICs(); 1842 1843 return 0; 1844 } 1845 1846 late_initcall(print_ICs); 1847 1848 1849 /* Where if anywhere is the i8259 connect in external int mode */ 1850 static struct { int pin, apic; } ioapic_i8259 = { -1, -1 }; 1851 1852 void __init enable_IO_APIC(void) 1853 { 1854 int i8259_apic, i8259_pin; 1855 int apic, pin; 1856 1857 if (!legacy_pic->nr_legacy_irqs) 1858 return; 1859 1860 for_each_ioapic_pin(apic, pin) { 1861 /* See if any of the pins is in ExtINT mode */ 1862 struct IO_APIC_route_entry entry = ioapic_read_entry(apic, pin); 1863 1864 /* If the interrupt line is enabled and in ExtInt mode 1865 * I have found the pin where the i8259 is connected. 1866 */ 1867 if ((entry.mask == 0) && (entry.delivery_mode == dest_ExtINT)) { 1868 ioapic_i8259.apic = apic; 1869 ioapic_i8259.pin = pin; 1870 goto found_i8259; 1871 } 1872 } 1873 found_i8259: 1874 /* Look to see what if the MP table has reported the ExtINT */ 1875 /* If we could not find the appropriate pin by looking at the ioapic 1876 * the i8259 probably is not connected the ioapic but give the 1877 * mptable a chance anyway. 1878 */ 1879 i8259_pin = find_isa_irq_pin(0, mp_ExtINT); 1880 i8259_apic = find_isa_irq_apic(0, mp_ExtINT); 1881 /* Trust the MP table if nothing is setup in the hardware */ 1882 if ((ioapic_i8259.pin == -1) && (i8259_pin >= 0)) { 1883 printk(KERN_WARNING "ExtINT not setup in hardware but reported by MP table\n"); 1884 ioapic_i8259.pin = i8259_pin; 1885 ioapic_i8259.apic = i8259_apic; 1886 } 1887 /* Complain if the MP table and the hardware disagree */ 1888 if (((ioapic_i8259.apic != i8259_apic) || (ioapic_i8259.pin != i8259_pin)) && 1889 (i8259_pin >= 0) && (ioapic_i8259.pin >= 0)) 1890 { 1891 printk(KERN_WARNING "ExtINT in hardware and MP table differ\n"); 1892 } 1893 1894 /* 1895 * Do not trust the IO-APIC being empty at bootup 1896 */ 1897 clear_IO_APIC(); 1898 } 1899 1900 void native_disable_io_apic(void) 1901 { 1902 /* 1903 * If the i8259 is routed through an IOAPIC 1904 * Put that IOAPIC in virtual wire mode 1905 * so legacy interrupts can be delivered. 1906 */ 1907 if (ioapic_i8259.pin != -1) { 1908 struct IO_APIC_route_entry entry; 1909 1910 memset(&entry, 0, sizeof(entry)); 1911 entry.mask = 0; /* Enabled */ 1912 entry.trigger = 0; /* Edge */ 1913 entry.irr = 0; 1914 entry.polarity = 0; /* High */ 1915 entry.delivery_status = 0; 1916 entry.dest_mode = 0; /* Physical */ 1917 entry.delivery_mode = dest_ExtINT; /* ExtInt */ 1918 entry.vector = 0; 1919 entry.dest = read_apic_id(); 1920 1921 /* 1922 * Add it to the IO-APIC irq-routing table: 1923 */ 1924 ioapic_write_entry(ioapic_i8259.apic, ioapic_i8259.pin, entry); 1925 } 1926 1927 if (cpu_has_apic || apic_from_smp_config()) 1928 disconnect_bsp_APIC(ioapic_i8259.pin != -1); 1929 1930 } 1931 1932 /* 1933 * Not an __init, needed by the reboot code 1934 */ 1935 void disable_IO_APIC(void) 1936 { 1937 /* 1938 * Clear the IO-APIC before rebooting: 1939 */ 1940 clear_IO_APIC(); 1941 1942 if (!legacy_pic->nr_legacy_irqs) 1943 return; 1944 1945 x86_io_apic_ops.disable(); 1946 } 1947 1948 #ifdef CONFIG_X86_32 1949 /* 1950 * function to set the IO-APIC physical IDs based on the 1951 * values stored in the MPC table. 1952 * 1953 * by Matt Domsch <Matt_Domsch@dell.com> Tue Dec 21 12:25:05 CST 1999 1954 */ 1955 void __init setup_ioapic_ids_from_mpc_nocheck(void) 1956 { 1957 union IO_APIC_reg_00 reg_00; 1958 physid_mask_t phys_id_present_map; 1959 int ioapic_idx; 1960 int i; 1961 unsigned char old_id; 1962 unsigned long flags; 1963 1964 /* 1965 * This is broken; anything with a real cpu count has to 1966 * circumvent this idiocy regardless. 1967 */ 1968 apic->ioapic_phys_id_map(&phys_cpu_present_map, &phys_id_present_map); 1969 1970 /* 1971 * Set the IOAPIC ID to the value stored in the MPC table. 1972 */ 1973 for_each_ioapic(ioapic_idx) { 1974 /* Read the register 0 value */ 1975 raw_spin_lock_irqsave(&ioapic_lock, flags); 1976 reg_00.raw = io_apic_read(ioapic_idx, 0); 1977 raw_spin_unlock_irqrestore(&ioapic_lock, flags); 1978 1979 old_id = mpc_ioapic_id(ioapic_idx); 1980 1981 if (mpc_ioapic_id(ioapic_idx) >= get_physical_broadcast()) { 1982 printk(KERN_ERR "BIOS bug, IO-APIC#%d ID is %d in the MPC table!...\n", 1983 ioapic_idx, mpc_ioapic_id(ioapic_idx)); 1984 printk(KERN_ERR "... fixing up to %d. (tell your hw vendor)\n", 1985 reg_00.bits.ID); 1986 ioapics[ioapic_idx].mp_config.apicid = reg_00.bits.ID; 1987 } 1988 1989 /* 1990 * Sanity check, is the ID really free? Every APIC in a 1991 * system must have a unique ID or we get lots of nice 1992 * 'stuck on smp_invalidate_needed IPI wait' messages. 1993 */ 1994 if (apic->check_apicid_used(&phys_id_present_map, 1995 mpc_ioapic_id(ioapic_idx))) { 1996 printk(KERN_ERR "BIOS bug, IO-APIC#%d ID %d is already used!...\n", 1997 ioapic_idx, mpc_ioapic_id(ioapic_idx)); 1998 for (i = 0; i < get_physical_broadcast(); i++) 1999 if (!physid_isset(i, phys_id_present_map)) 2000 break; 2001 if (i >= get_physical_broadcast()) 2002 panic("Max APIC ID exceeded!\n"); 2003 printk(KERN_ERR "... fixing up to %d. (tell your hw vendor)\n", 2004 i); 2005 physid_set(i, phys_id_present_map); 2006 ioapics[ioapic_idx].mp_config.apicid = i; 2007 } else { 2008 physid_mask_t tmp; 2009 apic->apicid_to_cpu_present(mpc_ioapic_id(ioapic_idx), 2010 &tmp); 2011 apic_printk(APIC_VERBOSE, "Setting %d in the " 2012 "phys_id_present_map\n", 2013 mpc_ioapic_id(ioapic_idx)); 2014 physids_or(phys_id_present_map, phys_id_present_map, tmp); 2015 } 2016 2017 /* 2018 * We need to adjust the IRQ routing table 2019 * if the ID changed. 2020 */ 2021 if (old_id != mpc_ioapic_id(ioapic_idx)) 2022 for (i = 0; i < mp_irq_entries; i++) 2023 if (mp_irqs[i].dstapic == old_id) 2024 mp_irqs[i].dstapic 2025 = mpc_ioapic_id(ioapic_idx); 2026 2027 /* 2028 * Update the ID register according to the right value 2029 * from the MPC table if they are different. 2030 */ 2031 if (mpc_ioapic_id(ioapic_idx) == reg_00.bits.ID) 2032 continue; 2033 2034 apic_printk(APIC_VERBOSE, KERN_INFO 2035 "...changing IO-APIC physical APIC ID to %d ...", 2036 mpc_ioapic_id(ioapic_idx)); 2037 2038 reg_00.bits.ID = mpc_ioapic_id(ioapic_idx); 2039 raw_spin_lock_irqsave(&ioapic_lock, flags); 2040 io_apic_write(ioapic_idx, 0, reg_00.raw); 2041 raw_spin_unlock_irqrestore(&ioapic_lock, flags); 2042 2043 /* 2044 * Sanity check 2045 */ 2046 raw_spin_lock_irqsave(&ioapic_lock, flags); 2047 reg_00.raw = io_apic_read(ioapic_idx, 0); 2048 raw_spin_unlock_irqrestore(&ioapic_lock, flags); 2049 if (reg_00.bits.ID != mpc_ioapic_id(ioapic_idx)) 2050 pr_cont("could not set ID!\n"); 2051 else 2052 apic_printk(APIC_VERBOSE, " ok.\n"); 2053 } 2054 } 2055 2056 void __init setup_ioapic_ids_from_mpc(void) 2057 { 2058 2059 if (acpi_ioapic) 2060 return; 2061 /* 2062 * Don't check I/O APIC IDs for xAPIC systems. They have 2063 * no meaning without the serial APIC bus. 2064 */ 2065 if (!(boot_cpu_data.x86_vendor == X86_VENDOR_INTEL) 2066 || APIC_XAPIC(apic_version[boot_cpu_physical_apicid])) 2067 return; 2068 setup_ioapic_ids_from_mpc_nocheck(); 2069 } 2070 #endif 2071 2072 int no_timer_check __initdata; 2073 2074 static int __init notimercheck(char *s) 2075 { 2076 no_timer_check = 1; 2077 return 1; 2078 } 2079 __setup("no_timer_check", notimercheck); 2080 2081 /* 2082 * There is a nasty bug in some older SMP boards, their mptable lies 2083 * about the timer IRQ. We do the following to work around the situation: 2084 * 2085 * - timer IRQ defaults to IO-APIC IRQ 2086 * - if this function detects that timer IRQs are defunct, then we fall 2087 * back to ISA timer IRQs 2088 */ 2089 static int __init timer_irq_works(void) 2090 { 2091 unsigned long t1 = jiffies; 2092 unsigned long flags; 2093 2094 if (no_timer_check) 2095 return 1; 2096 2097 local_save_flags(flags); 2098 local_irq_enable(); 2099 /* Let ten ticks pass... */ 2100 mdelay((10 * 1000) / HZ); 2101 local_irq_restore(flags); 2102 2103 /* 2104 * Expect a few ticks at least, to be sure some possible 2105 * glue logic does not lock up after one or two first 2106 * ticks in a non-ExtINT mode. Also the local APIC 2107 * might have cached one ExtINT interrupt. Finally, at 2108 * least one tick may be lost due to delays. 2109 */ 2110 2111 /* jiffies wrap? */ 2112 if (time_after(jiffies, t1 + 4)) 2113 return 1; 2114 return 0; 2115 } 2116 2117 /* 2118 * In the SMP+IOAPIC case it might happen that there are an unspecified 2119 * number of pending IRQ events unhandled. These cases are very rare, 2120 * so we 'resend' these IRQs via IPIs, to the same CPU. It's much 2121 * better to do it this way as thus we do not have to be aware of 2122 * 'pending' interrupts in the IRQ path, except at this point. 2123 */ 2124 /* 2125 * Edge triggered needs to resend any interrupt 2126 * that was delayed but this is now handled in the device 2127 * independent code. 2128 */ 2129 2130 /* 2131 * Starting up a edge-triggered IO-APIC interrupt is 2132 * nasty - we need to make sure that we get the edge. 2133 * If it is already asserted for some reason, we need 2134 * return 1 to indicate that is was pending. 2135 * 2136 * This is not complete - we should be able to fake 2137 * an edge even if it isn't on the 8259A... 2138 */ 2139 2140 static unsigned int startup_ioapic_irq(struct irq_data *data) 2141 { 2142 int was_pending = 0, irq = data->irq; 2143 unsigned long flags; 2144 2145 raw_spin_lock_irqsave(&ioapic_lock, flags); 2146 if (irq < legacy_pic->nr_legacy_irqs) { 2147 legacy_pic->mask(irq); 2148 if (legacy_pic->irq_pending(irq)) 2149 was_pending = 1; 2150 } 2151 __unmask_ioapic(data->chip_data); 2152 raw_spin_unlock_irqrestore(&ioapic_lock, flags); 2153 2154 return was_pending; 2155 } 2156 2157 static int ioapic_retrigger_irq(struct irq_data *data) 2158 { 2159 struct irq_cfg *cfg = data->chip_data; 2160 unsigned long flags; 2161 int cpu; 2162 2163 raw_spin_lock_irqsave(&vector_lock, flags); 2164 cpu = cpumask_first_and(cfg->domain, cpu_online_mask); 2165 apic->send_IPI_mask(cpumask_of(cpu), cfg->vector); 2166 raw_spin_unlock_irqrestore(&vector_lock, flags); 2167 2168 return 1; 2169 } 2170 2171 /* 2172 * Level and edge triggered IO-APIC interrupts need different handling, 2173 * so we use two separate IRQ descriptors. Edge triggered IRQs can be 2174 * handled with the level-triggered descriptor, but that one has slightly 2175 * more overhead. Level-triggered interrupts cannot be handled with the 2176 * edge-triggered handler, without risking IRQ storms and other ugly 2177 * races. 2178 */ 2179 2180 #ifdef CONFIG_SMP 2181 void send_cleanup_vector(struct irq_cfg *cfg) 2182 { 2183 cpumask_var_t cleanup_mask; 2184 2185 if (unlikely(!alloc_cpumask_var(&cleanup_mask, GFP_ATOMIC))) { 2186 unsigned int i; 2187 for_each_cpu_and(i, cfg->old_domain, cpu_online_mask) 2188 apic->send_IPI_mask(cpumask_of(i), IRQ_MOVE_CLEANUP_VECTOR); 2189 } else { 2190 cpumask_and(cleanup_mask, cfg->old_domain, cpu_online_mask); 2191 apic->send_IPI_mask(cleanup_mask, IRQ_MOVE_CLEANUP_VECTOR); 2192 free_cpumask_var(cleanup_mask); 2193 } 2194 cfg->move_in_progress = 0; 2195 } 2196 2197 asmlinkage __visible void smp_irq_move_cleanup_interrupt(void) 2198 { 2199 unsigned vector, me; 2200 2201 ack_APIC_irq(); 2202 irq_enter(); 2203 exit_idle(); 2204 2205 me = smp_processor_id(); 2206 for (vector = FIRST_EXTERNAL_VECTOR; vector < NR_VECTORS; vector++) { 2207 int irq; 2208 unsigned int irr; 2209 struct irq_desc *desc; 2210 struct irq_cfg *cfg; 2211 irq = __this_cpu_read(vector_irq[vector]); 2212 2213 if (irq <= VECTOR_UNDEFINED) 2214 continue; 2215 2216 desc = irq_to_desc(irq); 2217 if (!desc) 2218 continue; 2219 2220 cfg = irq_cfg(irq); 2221 if (!cfg) 2222 continue; 2223 2224 raw_spin_lock(&desc->lock); 2225 2226 /* 2227 * Check if the irq migration is in progress. If so, we 2228 * haven't received the cleanup request yet for this irq. 2229 */ 2230 if (cfg->move_in_progress) 2231 goto unlock; 2232 2233 if (vector == cfg->vector && cpumask_test_cpu(me, cfg->domain)) 2234 goto unlock; 2235 2236 irr = apic_read(APIC_IRR + (vector / 32 * 0x10)); 2237 /* 2238 * Check if the vector that needs to be cleanedup is 2239 * registered at the cpu's IRR. If so, then this is not 2240 * the best time to clean it up. Lets clean it up in the 2241 * next attempt by sending another IRQ_MOVE_CLEANUP_VECTOR 2242 * to myself. 2243 */ 2244 if (irr & (1 << (vector % 32))) { 2245 apic->send_IPI_self(IRQ_MOVE_CLEANUP_VECTOR); 2246 goto unlock; 2247 } 2248 __this_cpu_write(vector_irq[vector], VECTOR_UNDEFINED); 2249 unlock: 2250 raw_spin_unlock(&desc->lock); 2251 } 2252 2253 irq_exit(); 2254 } 2255 2256 static void __irq_complete_move(struct irq_cfg *cfg, unsigned vector) 2257 { 2258 unsigned me; 2259 2260 if (likely(!cfg->move_in_progress)) 2261 return; 2262 2263 me = smp_processor_id(); 2264 2265 if (vector == cfg->vector && cpumask_test_cpu(me, cfg->domain)) 2266 send_cleanup_vector(cfg); 2267 } 2268 2269 static void irq_complete_move(struct irq_cfg *cfg) 2270 { 2271 __irq_complete_move(cfg, ~get_irq_regs()->orig_ax); 2272 } 2273 2274 void irq_force_complete_move(int irq) 2275 { 2276 struct irq_cfg *cfg = irq_cfg(irq); 2277 2278 if (!cfg) 2279 return; 2280 2281 __irq_complete_move(cfg, cfg->vector); 2282 } 2283 #else 2284 static inline void irq_complete_move(struct irq_cfg *cfg) { } 2285 #endif 2286 2287 static void __target_IO_APIC_irq(unsigned int irq, unsigned int dest, struct irq_cfg *cfg) 2288 { 2289 int apic, pin; 2290 struct irq_pin_list *entry; 2291 u8 vector = cfg->vector; 2292 2293 for_each_irq_pin(entry, cfg->irq_2_pin) { 2294 unsigned int reg; 2295 2296 apic = entry->apic; 2297 pin = entry->pin; 2298 2299 io_apic_write(apic, 0x11 + pin*2, dest); 2300 reg = io_apic_read(apic, 0x10 + pin*2); 2301 reg &= ~IO_APIC_REDIR_VECTOR_MASK; 2302 reg |= vector; 2303 io_apic_modify(apic, 0x10 + pin*2, reg); 2304 } 2305 } 2306 2307 /* 2308 * Either sets data->affinity to a valid value, and returns 2309 * ->cpu_mask_to_apicid of that in dest_id, or returns -1 and 2310 * leaves data->affinity untouched. 2311 */ 2312 int __ioapic_set_affinity(struct irq_data *data, const struct cpumask *mask, 2313 unsigned int *dest_id) 2314 { 2315 struct irq_cfg *cfg = data->chip_data; 2316 unsigned int irq = data->irq; 2317 int err; 2318 2319 if (!config_enabled(CONFIG_SMP)) 2320 return -EPERM; 2321 2322 if (!cpumask_intersects(mask, cpu_online_mask)) 2323 return -EINVAL; 2324 2325 err = assign_irq_vector(irq, cfg, mask); 2326 if (err) 2327 return err; 2328 2329 err = apic->cpu_mask_to_apicid_and(mask, cfg->domain, dest_id); 2330 if (err) { 2331 if (assign_irq_vector(irq, cfg, data->affinity)) 2332 pr_err("Failed to recover vector for irq %d\n", irq); 2333 return err; 2334 } 2335 2336 cpumask_copy(data->affinity, mask); 2337 2338 return 0; 2339 } 2340 2341 2342 int native_ioapic_set_affinity(struct irq_data *data, 2343 const struct cpumask *mask, 2344 bool force) 2345 { 2346 unsigned int dest, irq = data->irq; 2347 unsigned long flags; 2348 int ret; 2349 2350 if (!config_enabled(CONFIG_SMP)) 2351 return -EPERM; 2352 2353 raw_spin_lock_irqsave(&ioapic_lock, flags); 2354 ret = __ioapic_set_affinity(data, mask, &dest); 2355 if (!ret) { 2356 /* Only the high 8 bits are valid. */ 2357 dest = SET_APIC_LOGICAL_ID(dest); 2358 __target_IO_APIC_irq(irq, dest, data->chip_data); 2359 ret = IRQ_SET_MASK_OK_NOCOPY; 2360 } 2361 raw_spin_unlock_irqrestore(&ioapic_lock, flags); 2362 return ret; 2363 } 2364 2365 static void ack_apic_edge(struct irq_data *data) 2366 { 2367 irq_complete_move(data->chip_data); 2368 irq_move_irq(data); 2369 ack_APIC_irq(); 2370 } 2371 2372 atomic_t irq_mis_count; 2373 2374 #ifdef CONFIG_GENERIC_PENDING_IRQ 2375 static bool io_apic_level_ack_pending(struct irq_cfg *cfg) 2376 { 2377 struct irq_pin_list *entry; 2378 unsigned long flags; 2379 2380 raw_spin_lock_irqsave(&ioapic_lock, flags); 2381 for_each_irq_pin(entry, cfg->irq_2_pin) { 2382 unsigned int reg; 2383 int pin; 2384 2385 pin = entry->pin; 2386 reg = io_apic_read(entry->apic, 0x10 + pin*2); 2387 /* Is the remote IRR bit set? */ 2388 if (reg & IO_APIC_REDIR_REMOTE_IRR) { 2389 raw_spin_unlock_irqrestore(&ioapic_lock, flags); 2390 return true; 2391 } 2392 } 2393 raw_spin_unlock_irqrestore(&ioapic_lock, flags); 2394 2395 return false; 2396 } 2397 2398 static inline bool ioapic_irqd_mask(struct irq_data *data, struct irq_cfg *cfg) 2399 { 2400 /* If we are moving the irq we need to mask it */ 2401 if (unlikely(irqd_is_setaffinity_pending(data))) { 2402 mask_ioapic(cfg); 2403 return true; 2404 } 2405 return false; 2406 } 2407 2408 static inline void ioapic_irqd_unmask(struct irq_data *data, 2409 struct irq_cfg *cfg, bool masked) 2410 { 2411 if (unlikely(masked)) { 2412 /* Only migrate the irq if the ack has been received. 2413 * 2414 * On rare occasions the broadcast level triggered ack gets 2415 * delayed going to ioapics, and if we reprogram the 2416 * vector while Remote IRR is still set the irq will never 2417 * fire again. 2418 * 2419 * To prevent this scenario we read the Remote IRR bit 2420 * of the ioapic. This has two effects. 2421 * - On any sane system the read of the ioapic will 2422 * flush writes (and acks) going to the ioapic from 2423 * this cpu. 2424 * - We get to see if the ACK has actually been delivered. 2425 * 2426 * Based on failed experiments of reprogramming the 2427 * ioapic entry from outside of irq context starting 2428 * with masking the ioapic entry and then polling until 2429 * Remote IRR was clear before reprogramming the 2430 * ioapic I don't trust the Remote IRR bit to be 2431 * completey accurate. 2432 * 2433 * However there appears to be no other way to plug 2434 * this race, so if the Remote IRR bit is not 2435 * accurate and is causing problems then it is a hardware bug 2436 * and you can go talk to the chipset vendor about it. 2437 */ 2438 if (!io_apic_level_ack_pending(cfg)) 2439 irq_move_masked_irq(data); 2440 unmask_ioapic(cfg); 2441 } 2442 } 2443 #else 2444 static inline bool ioapic_irqd_mask(struct irq_data *data, struct irq_cfg *cfg) 2445 { 2446 return false; 2447 } 2448 static inline void ioapic_irqd_unmask(struct irq_data *data, 2449 struct irq_cfg *cfg, bool masked) 2450 { 2451 } 2452 #endif 2453 2454 static void ack_apic_level(struct irq_data *data) 2455 { 2456 struct irq_cfg *cfg = data->chip_data; 2457 int i, irq = data->irq; 2458 unsigned long v; 2459 bool masked; 2460 2461 irq_complete_move(cfg); 2462 masked = ioapic_irqd_mask(data, cfg); 2463 2464 /* 2465 * It appears there is an erratum which affects at least version 0x11 2466 * of I/O APIC (that's the 82093AA and cores integrated into various 2467 * chipsets). Under certain conditions a level-triggered interrupt is 2468 * erroneously delivered as edge-triggered one but the respective IRR 2469 * bit gets set nevertheless. As a result the I/O unit expects an EOI 2470 * message but it will never arrive and further interrupts are blocked 2471 * from the source. The exact reason is so far unknown, but the 2472 * phenomenon was observed when two consecutive interrupt requests 2473 * from a given source get delivered to the same CPU and the source is 2474 * temporarily disabled in between. 2475 * 2476 * A workaround is to simulate an EOI message manually. We achieve it 2477 * by setting the trigger mode to edge and then to level when the edge 2478 * trigger mode gets detected in the TMR of a local APIC for a 2479 * level-triggered interrupt. We mask the source for the time of the 2480 * operation to prevent an edge-triggered interrupt escaping meanwhile. 2481 * The idea is from Manfred Spraul. --macro 2482 * 2483 * Also in the case when cpu goes offline, fixup_irqs() will forward 2484 * any unhandled interrupt on the offlined cpu to the new cpu 2485 * destination that is handling the corresponding interrupt. This 2486 * interrupt forwarding is done via IPI's. Hence, in this case also 2487 * level-triggered io-apic interrupt will be seen as an edge 2488 * interrupt in the IRR. And we can't rely on the cpu's EOI 2489 * to be broadcasted to the IO-APIC's which will clear the remoteIRR 2490 * corresponding to the level-triggered interrupt. Hence on IO-APIC's 2491 * supporting EOI register, we do an explicit EOI to clear the 2492 * remote IRR and on IO-APIC's which don't have an EOI register, 2493 * we use the above logic (mask+edge followed by unmask+level) from 2494 * Manfred Spraul to clear the remote IRR. 2495 */ 2496 i = cfg->vector; 2497 v = apic_read(APIC_TMR + ((i & ~0x1f) >> 1)); 2498 2499 /* 2500 * We must acknowledge the irq before we move it or the acknowledge will 2501 * not propagate properly. 2502 */ 2503 ack_APIC_irq(); 2504 2505 /* 2506 * Tail end of clearing remote IRR bit (either by delivering the EOI 2507 * message via io-apic EOI register write or simulating it using 2508 * mask+edge followed by unnask+level logic) manually when the 2509 * level triggered interrupt is seen as the edge triggered interrupt 2510 * at the cpu. 2511 */ 2512 if (!(v & (1 << (i & 0x1f)))) { 2513 atomic_inc(&irq_mis_count); 2514 2515 eoi_ioapic_irq(irq, cfg); 2516 } 2517 2518 ioapic_irqd_unmask(data, cfg, masked); 2519 } 2520 2521 static struct irq_chip ioapic_chip __read_mostly = { 2522 .name = "IO-APIC", 2523 .irq_startup = startup_ioapic_irq, 2524 .irq_mask = mask_ioapic_irq, 2525 .irq_unmask = unmask_ioapic_irq, 2526 .irq_ack = ack_apic_edge, 2527 .irq_eoi = ack_apic_level, 2528 .irq_set_affinity = native_ioapic_set_affinity, 2529 .irq_retrigger = ioapic_retrigger_irq, 2530 }; 2531 2532 static inline void init_IO_APIC_traps(void) 2533 { 2534 struct irq_cfg *cfg; 2535 unsigned int irq; 2536 2537 for_each_active_irq(irq) { 2538 cfg = irq_cfg(irq); 2539 if (IO_APIC_IRQ(irq) && cfg && !cfg->vector) { 2540 /* 2541 * Hmm.. We don't have an entry for this, 2542 * so default to an old-fashioned 8259 2543 * interrupt if we can.. 2544 */ 2545 if (irq < legacy_pic->nr_legacy_irqs) 2546 legacy_pic->make_irq(irq); 2547 else 2548 /* Strange. Oh, well.. */ 2549 irq_set_chip(irq, &no_irq_chip); 2550 } 2551 } 2552 } 2553 2554 /* 2555 * The local APIC irq-chip implementation: 2556 */ 2557 2558 static void mask_lapic_irq(struct irq_data *data) 2559 { 2560 unsigned long v; 2561 2562 v = apic_read(APIC_LVT0); 2563 apic_write(APIC_LVT0, v | APIC_LVT_MASKED); 2564 } 2565 2566 static void unmask_lapic_irq(struct irq_data *data) 2567 { 2568 unsigned long v; 2569 2570 v = apic_read(APIC_LVT0); 2571 apic_write(APIC_LVT0, v & ~APIC_LVT_MASKED); 2572 } 2573 2574 static void ack_lapic_irq(struct irq_data *data) 2575 { 2576 ack_APIC_irq(); 2577 } 2578 2579 static struct irq_chip lapic_chip __read_mostly = { 2580 .name = "local-APIC", 2581 .irq_mask = mask_lapic_irq, 2582 .irq_unmask = unmask_lapic_irq, 2583 .irq_ack = ack_lapic_irq, 2584 }; 2585 2586 static void lapic_register_intr(int irq) 2587 { 2588 irq_clear_status_flags(irq, IRQ_LEVEL); 2589 irq_set_chip_and_handler_name(irq, &lapic_chip, handle_edge_irq, 2590 "edge"); 2591 } 2592 2593 /* 2594 * This looks a bit hackish but it's about the only one way of sending 2595 * a few INTA cycles to 8259As and any associated glue logic. ICR does 2596 * not support the ExtINT mode, unfortunately. We need to send these 2597 * cycles as some i82489DX-based boards have glue logic that keeps the 2598 * 8259A interrupt line asserted until INTA. --macro 2599 */ 2600 static inline void __init unlock_ExtINT_logic(void) 2601 { 2602 int apic, pin, i; 2603 struct IO_APIC_route_entry entry0, entry1; 2604 unsigned char save_control, save_freq_select; 2605 2606 pin = find_isa_irq_pin(8, mp_INT); 2607 if (pin == -1) { 2608 WARN_ON_ONCE(1); 2609 return; 2610 } 2611 apic = find_isa_irq_apic(8, mp_INT); 2612 if (apic == -1) { 2613 WARN_ON_ONCE(1); 2614 return; 2615 } 2616 2617 entry0 = ioapic_read_entry(apic, pin); 2618 clear_IO_APIC_pin(apic, pin); 2619 2620 memset(&entry1, 0, sizeof(entry1)); 2621 2622 entry1.dest_mode = 0; /* physical delivery */ 2623 entry1.mask = 0; /* unmask IRQ now */ 2624 entry1.dest = hard_smp_processor_id(); 2625 entry1.delivery_mode = dest_ExtINT; 2626 entry1.polarity = entry0.polarity; 2627 entry1.trigger = 0; 2628 entry1.vector = 0; 2629 2630 ioapic_write_entry(apic, pin, entry1); 2631 2632 save_control = CMOS_READ(RTC_CONTROL); 2633 save_freq_select = CMOS_READ(RTC_FREQ_SELECT); 2634 CMOS_WRITE((save_freq_select & ~RTC_RATE_SELECT) | 0x6, 2635 RTC_FREQ_SELECT); 2636 CMOS_WRITE(save_control | RTC_PIE, RTC_CONTROL); 2637 2638 i = 100; 2639 while (i-- > 0) { 2640 mdelay(10); 2641 if ((CMOS_READ(RTC_INTR_FLAGS) & RTC_PF) == RTC_PF) 2642 i -= 10; 2643 } 2644 2645 CMOS_WRITE(save_control, RTC_CONTROL); 2646 CMOS_WRITE(save_freq_select, RTC_FREQ_SELECT); 2647 clear_IO_APIC_pin(apic, pin); 2648 2649 ioapic_write_entry(apic, pin, entry0); 2650 } 2651 2652 static int disable_timer_pin_1 __initdata; 2653 /* Actually the next is obsolete, but keep it for paranoid reasons -AK */ 2654 static int __init disable_timer_pin_setup(char *arg) 2655 { 2656 disable_timer_pin_1 = 1; 2657 return 0; 2658 } 2659 early_param("disable_timer_pin_1", disable_timer_pin_setup); 2660 2661 /* 2662 * This code may look a bit paranoid, but it's supposed to cooperate with 2663 * a wide range of boards and BIOS bugs. Fortunately only the timer IRQ 2664 * is so screwy. Thanks to Brian Perkins for testing/hacking this beast 2665 * fanatically on his truly buggy board. 2666 * 2667 * FIXME: really need to revamp this for all platforms. 2668 */ 2669 static inline void __init check_timer(void) 2670 { 2671 struct irq_cfg *cfg = irq_cfg(0); 2672 int node = cpu_to_node(0); 2673 int apic1, pin1, apic2, pin2; 2674 unsigned long flags; 2675 int no_pin1 = 0; 2676 2677 local_irq_save(flags); 2678 2679 /* 2680 * get/set the timer IRQ vector: 2681 */ 2682 legacy_pic->mask(0); 2683 assign_irq_vector(0, cfg, apic->target_cpus()); 2684 2685 /* 2686 * As IRQ0 is to be enabled in the 8259A, the virtual 2687 * wire has to be disabled in the local APIC. Also 2688 * timer interrupts need to be acknowledged manually in 2689 * the 8259A for the i82489DX when using the NMI 2690 * watchdog as that APIC treats NMIs as level-triggered. 2691 * The AEOI mode will finish them in the 8259A 2692 * automatically. 2693 */ 2694 apic_write(APIC_LVT0, APIC_LVT_MASKED | APIC_DM_EXTINT); 2695 legacy_pic->init(1); 2696 2697 pin1 = find_isa_irq_pin(0, mp_INT); 2698 apic1 = find_isa_irq_apic(0, mp_INT); 2699 pin2 = ioapic_i8259.pin; 2700 apic2 = ioapic_i8259.apic; 2701 2702 apic_printk(APIC_QUIET, KERN_INFO "..TIMER: vector=0x%02X " 2703 "apic1=%d pin1=%d apic2=%d pin2=%d\n", 2704 cfg->vector, apic1, pin1, apic2, pin2); 2705 2706 /* 2707 * Some BIOS writers are clueless and report the ExtINTA 2708 * I/O APIC input from the cascaded 8259A as the timer 2709 * interrupt input. So just in case, if only one pin 2710 * was found above, try it both directly and through the 2711 * 8259A. 2712 */ 2713 if (pin1 == -1) { 2714 panic_if_irq_remap("BIOS bug: timer not connected to IO-APIC"); 2715 pin1 = pin2; 2716 apic1 = apic2; 2717 no_pin1 = 1; 2718 } else if (pin2 == -1) { 2719 pin2 = pin1; 2720 apic2 = apic1; 2721 } 2722 2723 if (pin1 != -1) { 2724 /* 2725 * Ok, does IRQ0 through the IOAPIC work? 2726 */ 2727 if (no_pin1) { 2728 add_pin_to_irq_node(cfg, node, apic1, pin1); 2729 setup_timer_IRQ0_pin(apic1, pin1, cfg->vector); 2730 } else { 2731 /* for edge trigger, setup_ioapic_irq already 2732 * leave it unmasked. 2733 * so only need to unmask if it is level-trigger 2734 * do we really have level trigger timer? 2735 */ 2736 int idx; 2737 idx = find_irq_entry(apic1, pin1, mp_INT); 2738 if (idx != -1 && irq_trigger(idx)) 2739 unmask_ioapic(cfg); 2740 } 2741 if (timer_irq_works()) { 2742 if (disable_timer_pin_1 > 0) 2743 clear_IO_APIC_pin(0, pin1); 2744 goto out; 2745 } 2746 panic_if_irq_remap("timer doesn't work through Interrupt-remapped IO-APIC"); 2747 local_irq_disable(); 2748 clear_IO_APIC_pin(apic1, pin1); 2749 if (!no_pin1) 2750 apic_printk(APIC_QUIET, KERN_ERR "..MP-BIOS bug: " 2751 "8254 timer not connected to IO-APIC\n"); 2752 2753 apic_printk(APIC_QUIET, KERN_INFO "...trying to set up timer " 2754 "(IRQ0) through the 8259A ...\n"); 2755 apic_printk(APIC_QUIET, KERN_INFO 2756 "..... (found apic %d pin %d) ...\n", apic2, pin2); 2757 /* 2758 * legacy devices should be connected to IO APIC #0 2759 */ 2760 replace_pin_at_irq_node(cfg, node, apic1, pin1, apic2, pin2); 2761 setup_timer_IRQ0_pin(apic2, pin2, cfg->vector); 2762 legacy_pic->unmask(0); 2763 if (timer_irq_works()) { 2764 apic_printk(APIC_QUIET, KERN_INFO "....... works.\n"); 2765 goto out; 2766 } 2767 /* 2768 * Cleanup, just in case ... 2769 */ 2770 local_irq_disable(); 2771 legacy_pic->mask(0); 2772 clear_IO_APIC_pin(apic2, pin2); 2773 apic_printk(APIC_QUIET, KERN_INFO "....... failed.\n"); 2774 } 2775 2776 apic_printk(APIC_QUIET, KERN_INFO 2777 "...trying to set up timer as Virtual Wire IRQ...\n"); 2778 2779 lapic_register_intr(0); 2780 apic_write(APIC_LVT0, APIC_DM_FIXED | cfg->vector); /* Fixed mode */ 2781 legacy_pic->unmask(0); 2782 2783 if (timer_irq_works()) { 2784 apic_printk(APIC_QUIET, KERN_INFO "..... works.\n"); 2785 goto out; 2786 } 2787 local_irq_disable(); 2788 legacy_pic->mask(0); 2789 apic_write(APIC_LVT0, APIC_LVT_MASKED | APIC_DM_FIXED | cfg->vector); 2790 apic_printk(APIC_QUIET, KERN_INFO "..... failed.\n"); 2791 2792 apic_printk(APIC_QUIET, KERN_INFO 2793 "...trying to set up timer as ExtINT IRQ...\n"); 2794 2795 legacy_pic->init(0); 2796 legacy_pic->make_irq(0); 2797 apic_write(APIC_LVT0, APIC_DM_EXTINT); 2798 2799 unlock_ExtINT_logic(); 2800 2801 if (timer_irq_works()) { 2802 apic_printk(APIC_QUIET, KERN_INFO "..... works.\n"); 2803 goto out; 2804 } 2805 local_irq_disable(); 2806 apic_printk(APIC_QUIET, KERN_INFO "..... failed :(.\n"); 2807 if (x2apic_preenabled) 2808 apic_printk(APIC_QUIET, KERN_INFO 2809 "Perhaps problem with the pre-enabled x2apic mode\n" 2810 "Try booting with x2apic and interrupt-remapping disabled in the bios.\n"); 2811 panic("IO-APIC + timer doesn't work! Boot with apic=debug and send a " 2812 "report. Then try booting with the 'noapic' option.\n"); 2813 out: 2814 local_irq_restore(flags); 2815 } 2816 2817 /* 2818 * Traditionally ISA IRQ2 is the cascade IRQ, and is not available 2819 * to devices. However there may be an I/O APIC pin available for 2820 * this interrupt regardless. The pin may be left unconnected, but 2821 * typically it will be reused as an ExtINT cascade interrupt for 2822 * the master 8259A. In the MPS case such a pin will normally be 2823 * reported as an ExtINT interrupt in the MP table. With ACPI 2824 * there is no provision for ExtINT interrupts, and in the absence 2825 * of an override it would be treated as an ordinary ISA I/O APIC 2826 * interrupt, that is edge-triggered and unmasked by default. We 2827 * used to do this, but it caused problems on some systems because 2828 * of the NMI watchdog and sometimes IRQ0 of the 8254 timer using 2829 * the same ExtINT cascade interrupt to drive the local APIC of the 2830 * bootstrap processor. Therefore we refrain from routing IRQ2 to 2831 * the I/O APIC in all cases now. No actual device should request 2832 * it anyway. --macro 2833 */ 2834 #define PIC_IRQS (1UL << PIC_CASCADE_IR) 2835 2836 void __init setup_IO_APIC(void) 2837 { 2838 2839 /* 2840 * calling enable_IO_APIC() is moved to setup_local_APIC for BP 2841 */ 2842 io_apic_irqs = legacy_pic->nr_legacy_irqs ? ~PIC_IRQS : ~0UL; 2843 2844 apic_printk(APIC_VERBOSE, "ENABLING IO-APIC IRQs\n"); 2845 /* 2846 * Set up IO-APIC IRQ routing. 2847 */ 2848 x86_init.mpparse.setup_ioapic_ids(); 2849 2850 sync_Arb_IDs(); 2851 setup_IO_APIC_irqs(); 2852 init_IO_APIC_traps(); 2853 if (legacy_pic->nr_legacy_irqs) 2854 check_timer(); 2855 } 2856 2857 /* 2858 * Called after all the initialization is done. If we didn't find any 2859 * APIC bugs then we can allow the modify fast path 2860 */ 2861 2862 static int __init io_apic_bug_finalize(void) 2863 { 2864 if (sis_apic_bug == -1) 2865 sis_apic_bug = 0; 2866 return 0; 2867 } 2868 2869 late_initcall(io_apic_bug_finalize); 2870 2871 static void resume_ioapic_id(int ioapic_idx) 2872 { 2873 unsigned long flags; 2874 union IO_APIC_reg_00 reg_00; 2875 2876 raw_spin_lock_irqsave(&ioapic_lock, flags); 2877 reg_00.raw = io_apic_read(ioapic_idx, 0); 2878 if (reg_00.bits.ID != mpc_ioapic_id(ioapic_idx)) { 2879 reg_00.bits.ID = mpc_ioapic_id(ioapic_idx); 2880 io_apic_write(ioapic_idx, 0, reg_00.raw); 2881 } 2882 raw_spin_unlock_irqrestore(&ioapic_lock, flags); 2883 } 2884 2885 static void ioapic_resume(void) 2886 { 2887 int ioapic_idx; 2888 2889 for_each_ioapic_reverse(ioapic_idx) 2890 resume_ioapic_id(ioapic_idx); 2891 2892 restore_ioapic_entries(); 2893 } 2894 2895 static struct syscore_ops ioapic_syscore_ops = { 2896 .suspend = save_ioapic_entries, 2897 .resume = ioapic_resume, 2898 }; 2899 2900 static int __init ioapic_init_ops(void) 2901 { 2902 register_syscore_ops(&ioapic_syscore_ops); 2903 2904 return 0; 2905 } 2906 2907 device_initcall(ioapic_init_ops); 2908 2909 /* 2910 * Dynamic irq allocate and deallocation. Should be replaced by irq domains! 2911 */ 2912 int arch_setup_hwirq(unsigned int irq, int node) 2913 { 2914 struct irq_cfg *cfg; 2915 unsigned long flags; 2916 int ret; 2917 2918 cfg = alloc_irq_cfg(irq, node); 2919 if (!cfg) 2920 return -ENOMEM; 2921 2922 raw_spin_lock_irqsave(&vector_lock, flags); 2923 ret = __assign_irq_vector(irq, cfg, apic->target_cpus()); 2924 raw_spin_unlock_irqrestore(&vector_lock, flags); 2925 2926 if (!ret) 2927 irq_set_chip_data(irq, cfg); 2928 else 2929 free_irq_cfg(irq, cfg); 2930 return ret; 2931 } 2932 2933 void arch_teardown_hwirq(unsigned int irq) 2934 { 2935 struct irq_cfg *cfg = irq_cfg(irq); 2936 unsigned long flags; 2937 2938 free_remapped_irq(irq); 2939 raw_spin_lock_irqsave(&vector_lock, flags); 2940 __clear_irq_vector(irq, cfg); 2941 raw_spin_unlock_irqrestore(&vector_lock, flags); 2942 free_irq_cfg(irq, cfg); 2943 } 2944 2945 /* 2946 * MSI message composition 2947 */ 2948 void native_compose_msi_msg(struct pci_dev *pdev, 2949 unsigned int irq, unsigned int dest, 2950 struct msi_msg *msg, u8 hpet_id) 2951 { 2952 struct irq_cfg *cfg = irq_cfg(irq); 2953 2954 msg->address_hi = MSI_ADDR_BASE_HI; 2955 2956 if (x2apic_enabled()) 2957 msg->address_hi |= MSI_ADDR_EXT_DEST_ID(dest); 2958 2959 msg->address_lo = 2960 MSI_ADDR_BASE_LO | 2961 ((apic->irq_dest_mode == 0) ? 2962 MSI_ADDR_DEST_MODE_PHYSICAL: 2963 MSI_ADDR_DEST_MODE_LOGICAL) | 2964 ((apic->irq_delivery_mode != dest_LowestPrio) ? 2965 MSI_ADDR_REDIRECTION_CPU: 2966 MSI_ADDR_REDIRECTION_LOWPRI) | 2967 MSI_ADDR_DEST_ID(dest); 2968 2969 msg->data = 2970 MSI_DATA_TRIGGER_EDGE | 2971 MSI_DATA_LEVEL_ASSERT | 2972 ((apic->irq_delivery_mode != dest_LowestPrio) ? 2973 MSI_DATA_DELIVERY_FIXED: 2974 MSI_DATA_DELIVERY_LOWPRI) | 2975 MSI_DATA_VECTOR(cfg->vector); 2976 } 2977 2978 #ifdef CONFIG_PCI_MSI 2979 static int msi_compose_msg(struct pci_dev *pdev, unsigned int irq, 2980 struct msi_msg *msg, u8 hpet_id) 2981 { 2982 struct irq_cfg *cfg; 2983 int err; 2984 unsigned dest; 2985 2986 if (disable_apic) 2987 return -ENXIO; 2988 2989 cfg = irq_cfg(irq); 2990 err = assign_irq_vector(irq, cfg, apic->target_cpus()); 2991 if (err) 2992 return err; 2993 2994 err = apic->cpu_mask_to_apicid_and(cfg->domain, 2995 apic->target_cpus(), &dest); 2996 if (err) 2997 return err; 2998 2999 x86_msi.compose_msi_msg(pdev, irq, dest, msg, hpet_id); 3000 3001 return 0; 3002 } 3003 3004 static int 3005 msi_set_affinity(struct irq_data *data, const struct cpumask *mask, bool force) 3006 { 3007 struct irq_cfg *cfg = data->chip_data; 3008 struct msi_msg msg; 3009 unsigned int dest; 3010 int ret; 3011 3012 ret = __ioapic_set_affinity(data, mask, &dest); 3013 if (ret) 3014 return ret; 3015 3016 __get_cached_msi_msg(data->msi_desc, &msg); 3017 3018 msg.data &= ~MSI_DATA_VECTOR_MASK; 3019 msg.data |= MSI_DATA_VECTOR(cfg->vector); 3020 msg.address_lo &= ~MSI_ADDR_DEST_ID_MASK; 3021 msg.address_lo |= MSI_ADDR_DEST_ID(dest); 3022 3023 __write_msi_msg(data->msi_desc, &msg); 3024 3025 return IRQ_SET_MASK_OK_NOCOPY; 3026 } 3027 3028 /* 3029 * IRQ Chip for MSI PCI/PCI-X/PCI-Express Devices, 3030 * which implement the MSI or MSI-X Capability Structure. 3031 */ 3032 static struct irq_chip msi_chip = { 3033 .name = "PCI-MSI", 3034 .irq_unmask = unmask_msi_irq, 3035 .irq_mask = mask_msi_irq, 3036 .irq_ack = ack_apic_edge, 3037 .irq_set_affinity = msi_set_affinity, 3038 .irq_retrigger = ioapic_retrigger_irq, 3039 }; 3040 3041 int setup_msi_irq(struct pci_dev *dev, struct msi_desc *msidesc, 3042 unsigned int irq_base, unsigned int irq_offset) 3043 { 3044 struct irq_chip *chip = &msi_chip; 3045 struct msi_msg msg; 3046 unsigned int irq = irq_base + irq_offset; 3047 int ret; 3048 3049 ret = msi_compose_msg(dev, irq, &msg, -1); 3050 if (ret < 0) 3051 return ret; 3052 3053 irq_set_msi_desc_off(irq_base, irq_offset, msidesc); 3054 3055 /* 3056 * MSI-X message is written per-IRQ, the offset is always 0. 3057 * MSI message denotes a contiguous group of IRQs, written for 0th IRQ. 3058 */ 3059 if (!irq_offset) 3060 write_msi_msg(irq, &msg); 3061 3062 setup_remapped_irq(irq, irq_cfg(irq), chip); 3063 3064 irq_set_chip_and_handler_name(irq, chip, handle_edge_irq, "edge"); 3065 3066 dev_printk(KERN_DEBUG, &dev->dev, "irq %d for MSI/MSI-X\n", irq); 3067 3068 return 0; 3069 } 3070 3071 int native_setup_msi_irqs(struct pci_dev *dev, int nvec, int type) 3072 { 3073 struct msi_desc *msidesc; 3074 unsigned int irq; 3075 int node, ret; 3076 3077 /* Multiple MSI vectors only supported with interrupt remapping */ 3078 if (type == PCI_CAP_ID_MSI && nvec > 1) 3079 return 1; 3080 3081 node = dev_to_node(&dev->dev); 3082 3083 list_for_each_entry(msidesc, &dev->msi_list, list) { 3084 irq = irq_alloc_hwirq(node); 3085 if (!irq) 3086 return -ENOSPC; 3087 3088 ret = setup_msi_irq(dev, msidesc, irq, 0); 3089 if (ret < 0) { 3090 irq_free_hwirq(irq); 3091 return ret; 3092 } 3093 3094 } 3095 return 0; 3096 } 3097 3098 void native_teardown_msi_irq(unsigned int irq) 3099 { 3100 irq_free_hwirq(irq); 3101 } 3102 3103 #ifdef CONFIG_DMAR_TABLE 3104 static int 3105 dmar_msi_set_affinity(struct irq_data *data, const struct cpumask *mask, 3106 bool force) 3107 { 3108 struct irq_cfg *cfg = data->chip_data; 3109 unsigned int dest, irq = data->irq; 3110 struct msi_msg msg; 3111 int ret; 3112 3113 ret = __ioapic_set_affinity(data, mask, &dest); 3114 if (ret) 3115 return ret; 3116 3117 dmar_msi_read(irq, &msg); 3118 3119 msg.data &= ~MSI_DATA_VECTOR_MASK; 3120 msg.data |= MSI_DATA_VECTOR(cfg->vector); 3121 msg.address_lo &= ~MSI_ADDR_DEST_ID_MASK; 3122 msg.address_lo |= MSI_ADDR_DEST_ID(dest); 3123 msg.address_hi = MSI_ADDR_BASE_HI | MSI_ADDR_EXT_DEST_ID(dest); 3124 3125 dmar_msi_write(irq, &msg); 3126 3127 return IRQ_SET_MASK_OK_NOCOPY; 3128 } 3129 3130 static struct irq_chip dmar_msi_type = { 3131 .name = "DMAR_MSI", 3132 .irq_unmask = dmar_msi_unmask, 3133 .irq_mask = dmar_msi_mask, 3134 .irq_ack = ack_apic_edge, 3135 .irq_set_affinity = dmar_msi_set_affinity, 3136 .irq_retrigger = ioapic_retrigger_irq, 3137 }; 3138 3139 int arch_setup_dmar_msi(unsigned int irq) 3140 { 3141 int ret; 3142 struct msi_msg msg; 3143 3144 ret = msi_compose_msg(NULL, irq, &msg, -1); 3145 if (ret < 0) 3146 return ret; 3147 dmar_msi_write(irq, &msg); 3148 irq_set_chip_and_handler_name(irq, &dmar_msi_type, handle_edge_irq, 3149 "edge"); 3150 return 0; 3151 } 3152 #endif 3153 3154 #ifdef CONFIG_HPET_TIMER 3155 3156 static int hpet_msi_set_affinity(struct irq_data *data, 3157 const struct cpumask *mask, bool force) 3158 { 3159 struct irq_cfg *cfg = data->chip_data; 3160 struct msi_msg msg; 3161 unsigned int dest; 3162 int ret; 3163 3164 ret = __ioapic_set_affinity(data, mask, &dest); 3165 if (ret) 3166 return ret; 3167 3168 hpet_msi_read(data->handler_data, &msg); 3169 3170 msg.data &= ~MSI_DATA_VECTOR_MASK; 3171 msg.data |= MSI_DATA_VECTOR(cfg->vector); 3172 msg.address_lo &= ~MSI_ADDR_DEST_ID_MASK; 3173 msg.address_lo |= MSI_ADDR_DEST_ID(dest); 3174 3175 hpet_msi_write(data->handler_data, &msg); 3176 3177 return IRQ_SET_MASK_OK_NOCOPY; 3178 } 3179 3180 static struct irq_chip hpet_msi_type = { 3181 .name = "HPET_MSI", 3182 .irq_unmask = hpet_msi_unmask, 3183 .irq_mask = hpet_msi_mask, 3184 .irq_ack = ack_apic_edge, 3185 .irq_set_affinity = hpet_msi_set_affinity, 3186 .irq_retrigger = ioapic_retrigger_irq, 3187 }; 3188 3189 int default_setup_hpet_msi(unsigned int irq, unsigned int id) 3190 { 3191 struct irq_chip *chip = &hpet_msi_type; 3192 struct msi_msg msg; 3193 int ret; 3194 3195 ret = msi_compose_msg(NULL, irq, &msg, id); 3196 if (ret < 0) 3197 return ret; 3198 3199 hpet_msi_write(irq_get_handler_data(irq), &msg); 3200 irq_set_status_flags(irq, IRQ_MOVE_PCNTXT); 3201 setup_remapped_irq(irq, irq_cfg(irq), chip); 3202 3203 irq_set_chip_and_handler_name(irq, chip, handle_edge_irq, "edge"); 3204 return 0; 3205 } 3206 #endif 3207 3208 #endif /* CONFIG_PCI_MSI */ 3209 /* 3210 * Hypertransport interrupt support 3211 */ 3212 #ifdef CONFIG_HT_IRQ 3213 3214 static void target_ht_irq(unsigned int irq, unsigned int dest, u8 vector) 3215 { 3216 struct ht_irq_msg msg; 3217 fetch_ht_irq_msg(irq, &msg); 3218 3219 msg.address_lo &= ~(HT_IRQ_LOW_VECTOR_MASK | HT_IRQ_LOW_DEST_ID_MASK); 3220 msg.address_hi &= ~(HT_IRQ_HIGH_DEST_ID_MASK); 3221 3222 msg.address_lo |= HT_IRQ_LOW_VECTOR(vector) | HT_IRQ_LOW_DEST_ID(dest); 3223 msg.address_hi |= HT_IRQ_HIGH_DEST_ID(dest); 3224 3225 write_ht_irq_msg(irq, &msg); 3226 } 3227 3228 static int 3229 ht_set_affinity(struct irq_data *data, const struct cpumask *mask, bool force) 3230 { 3231 struct irq_cfg *cfg = data->chip_data; 3232 unsigned int dest; 3233 int ret; 3234 3235 ret = __ioapic_set_affinity(data, mask, &dest); 3236 if (ret) 3237 return ret; 3238 3239 target_ht_irq(data->irq, dest, cfg->vector); 3240 return IRQ_SET_MASK_OK_NOCOPY; 3241 } 3242 3243 static struct irq_chip ht_irq_chip = { 3244 .name = "PCI-HT", 3245 .irq_mask = mask_ht_irq, 3246 .irq_unmask = unmask_ht_irq, 3247 .irq_ack = ack_apic_edge, 3248 .irq_set_affinity = ht_set_affinity, 3249 .irq_retrigger = ioapic_retrigger_irq, 3250 }; 3251 3252 int arch_setup_ht_irq(unsigned int irq, struct pci_dev *dev) 3253 { 3254 struct irq_cfg *cfg; 3255 struct ht_irq_msg msg; 3256 unsigned dest; 3257 int err; 3258 3259 if (disable_apic) 3260 return -ENXIO; 3261 3262 cfg = irq_cfg(irq); 3263 err = assign_irq_vector(irq, cfg, apic->target_cpus()); 3264 if (err) 3265 return err; 3266 3267 err = apic->cpu_mask_to_apicid_and(cfg->domain, 3268 apic->target_cpus(), &dest); 3269 if (err) 3270 return err; 3271 3272 msg.address_hi = HT_IRQ_HIGH_DEST_ID(dest); 3273 3274 msg.address_lo = 3275 HT_IRQ_LOW_BASE | 3276 HT_IRQ_LOW_DEST_ID(dest) | 3277 HT_IRQ_LOW_VECTOR(cfg->vector) | 3278 ((apic->irq_dest_mode == 0) ? 3279 HT_IRQ_LOW_DM_PHYSICAL : 3280 HT_IRQ_LOW_DM_LOGICAL) | 3281 HT_IRQ_LOW_RQEOI_EDGE | 3282 ((apic->irq_delivery_mode != dest_LowestPrio) ? 3283 HT_IRQ_LOW_MT_FIXED : 3284 HT_IRQ_LOW_MT_ARBITRATED) | 3285 HT_IRQ_LOW_IRQ_MASKED; 3286 3287 write_ht_irq_msg(irq, &msg); 3288 3289 irq_set_chip_and_handler_name(irq, &ht_irq_chip, 3290 handle_edge_irq, "edge"); 3291 3292 dev_printk(KERN_DEBUG, &dev->dev, "irq %d for HT\n", irq); 3293 3294 return 0; 3295 } 3296 #endif /* CONFIG_HT_IRQ */ 3297 3298 static int 3299 io_apic_setup_irq_pin(unsigned int irq, int node, struct io_apic_irq_attr *attr) 3300 { 3301 struct irq_cfg *cfg = alloc_irq_and_cfg_at(irq, node); 3302 int ret; 3303 3304 if (!cfg) 3305 return -EINVAL; 3306 ret = __add_pin_to_irq_node(cfg, node, attr->ioapic, attr->ioapic_pin); 3307 if (!ret) 3308 setup_ioapic_irq(irq, cfg, attr); 3309 return ret; 3310 } 3311 3312 int io_apic_setup_irq_pin_once(unsigned int irq, int node, 3313 struct io_apic_irq_attr *attr) 3314 { 3315 unsigned int ioapic_idx = attr->ioapic, pin = attr->ioapic_pin; 3316 int ret; 3317 struct IO_APIC_route_entry orig_entry; 3318 3319 /* Avoid redundant programming */ 3320 if (test_bit(pin, ioapics[ioapic_idx].pin_programmed)) { 3321 pr_debug("Pin %d-%d already programmed\n", mpc_ioapic_id(ioapic_idx), pin); 3322 orig_entry = ioapic_read_entry(attr->ioapic, pin); 3323 if (attr->trigger == orig_entry.trigger && attr->polarity == orig_entry.polarity) 3324 return 0; 3325 return -EBUSY; 3326 } 3327 ret = io_apic_setup_irq_pin(irq, node, attr); 3328 if (!ret) 3329 set_bit(pin, ioapics[ioapic_idx].pin_programmed); 3330 return ret; 3331 } 3332 3333 static int __init io_apic_get_redir_entries(int ioapic) 3334 { 3335 union IO_APIC_reg_01 reg_01; 3336 unsigned long flags; 3337 3338 raw_spin_lock_irqsave(&ioapic_lock, flags); 3339 reg_01.raw = io_apic_read(ioapic, 1); 3340 raw_spin_unlock_irqrestore(&ioapic_lock, flags); 3341 3342 /* The register returns the maximum index redir index 3343 * supported, which is one less than the total number of redir 3344 * entries. 3345 */ 3346 return reg_01.bits.entries + 1; 3347 } 3348 3349 unsigned int arch_dynirq_lower_bound(unsigned int from) 3350 { 3351 unsigned int min = gsi_top + NR_IRQS_LEGACY; 3352 3353 return from < min ? min : from; 3354 } 3355 3356 int __init arch_probe_nr_irqs(void) 3357 { 3358 int nr; 3359 3360 if (nr_irqs > (NR_VECTORS * nr_cpu_ids)) 3361 nr_irqs = NR_VECTORS * nr_cpu_ids; 3362 3363 nr = (gsi_top + NR_IRQS_LEGACY) + 8 * nr_cpu_ids; 3364 #if defined(CONFIG_PCI_MSI) || defined(CONFIG_HT_IRQ) 3365 /* 3366 * for MSI and HT dyn irq 3367 */ 3368 nr += (gsi_top + NR_IRQS_LEGACY) * 16; 3369 #endif 3370 if (nr < nr_irqs) 3371 nr_irqs = nr; 3372 3373 return NR_IRQS_LEGACY; 3374 } 3375 3376 int io_apic_set_pci_routing(struct device *dev, int irq, 3377 struct io_apic_irq_attr *irq_attr) 3378 { 3379 int node; 3380 3381 if (!IO_APIC_IRQ(irq)) { 3382 apic_printk(APIC_QUIET,KERN_ERR "IOAPIC[%d]: Invalid reference to IRQ 0\n", 3383 irq_attr->ioapic); 3384 return -EINVAL; 3385 } 3386 3387 node = dev ? dev_to_node(dev) : cpu_to_node(0); 3388 3389 return io_apic_setup_irq_pin_once(irq, node, irq_attr); 3390 } 3391 3392 #ifdef CONFIG_X86_32 3393 static int __init io_apic_get_unique_id(int ioapic, int apic_id) 3394 { 3395 union IO_APIC_reg_00 reg_00; 3396 static physid_mask_t apic_id_map = PHYSID_MASK_NONE; 3397 physid_mask_t tmp; 3398 unsigned long flags; 3399 int i = 0; 3400 3401 /* 3402 * The P4 platform supports up to 256 APIC IDs on two separate APIC 3403 * buses (one for LAPICs, one for IOAPICs), where predecessors only 3404 * supports up to 16 on one shared APIC bus. 3405 * 3406 * TBD: Expand LAPIC/IOAPIC support on P4-class systems to take full 3407 * advantage of new APIC bus architecture. 3408 */ 3409 3410 if (physids_empty(apic_id_map)) 3411 apic->ioapic_phys_id_map(&phys_cpu_present_map, &apic_id_map); 3412 3413 raw_spin_lock_irqsave(&ioapic_lock, flags); 3414 reg_00.raw = io_apic_read(ioapic, 0); 3415 raw_spin_unlock_irqrestore(&ioapic_lock, flags); 3416 3417 if (apic_id >= get_physical_broadcast()) { 3418 printk(KERN_WARNING "IOAPIC[%d]: Invalid apic_id %d, trying " 3419 "%d\n", ioapic, apic_id, reg_00.bits.ID); 3420 apic_id = reg_00.bits.ID; 3421 } 3422 3423 /* 3424 * Every APIC in a system must have a unique ID or we get lots of nice 3425 * 'stuck on smp_invalidate_needed IPI wait' messages. 3426 */ 3427 if (apic->check_apicid_used(&apic_id_map, apic_id)) { 3428 3429 for (i = 0; i < get_physical_broadcast(); i++) { 3430 if (!apic->check_apicid_used(&apic_id_map, i)) 3431 break; 3432 } 3433 3434 if (i == get_physical_broadcast()) 3435 panic("Max apic_id exceeded!\n"); 3436 3437 printk(KERN_WARNING "IOAPIC[%d]: apic_id %d already used, " 3438 "trying %d\n", ioapic, apic_id, i); 3439 3440 apic_id = i; 3441 } 3442 3443 apic->apicid_to_cpu_present(apic_id, &tmp); 3444 physids_or(apic_id_map, apic_id_map, tmp); 3445 3446 if (reg_00.bits.ID != apic_id) { 3447 reg_00.bits.ID = apic_id; 3448 3449 raw_spin_lock_irqsave(&ioapic_lock, flags); 3450 io_apic_write(ioapic, 0, reg_00.raw); 3451 reg_00.raw = io_apic_read(ioapic, 0); 3452 raw_spin_unlock_irqrestore(&ioapic_lock, flags); 3453 3454 /* Sanity check */ 3455 if (reg_00.bits.ID != apic_id) { 3456 pr_err("IOAPIC[%d]: Unable to change apic_id!\n", 3457 ioapic); 3458 return -1; 3459 } 3460 } 3461 3462 apic_printk(APIC_VERBOSE, KERN_INFO 3463 "IOAPIC[%d]: Assigned apic_id %d\n", ioapic, apic_id); 3464 3465 return apic_id; 3466 } 3467 3468 static u8 __init io_apic_unique_id(u8 id) 3469 { 3470 if ((boot_cpu_data.x86_vendor == X86_VENDOR_INTEL) && 3471 !APIC_XAPIC(apic_version[boot_cpu_physical_apicid])) 3472 return io_apic_get_unique_id(nr_ioapics, id); 3473 else 3474 return id; 3475 } 3476 #else 3477 static u8 __init io_apic_unique_id(u8 id) 3478 { 3479 int i; 3480 DECLARE_BITMAP(used, 256); 3481 3482 bitmap_zero(used, 256); 3483 for_each_ioapic(i) 3484 __set_bit(mpc_ioapic_id(i), used); 3485 if (!test_bit(id, used)) 3486 return id; 3487 return find_first_zero_bit(used, 256); 3488 } 3489 #endif 3490 3491 static int __init io_apic_get_version(int ioapic) 3492 { 3493 union IO_APIC_reg_01 reg_01; 3494 unsigned long flags; 3495 3496 raw_spin_lock_irqsave(&ioapic_lock, flags); 3497 reg_01.raw = io_apic_read(ioapic, 1); 3498 raw_spin_unlock_irqrestore(&ioapic_lock, flags); 3499 3500 return reg_01.bits.version; 3501 } 3502 3503 int acpi_get_override_irq(u32 gsi, int *trigger, int *polarity) 3504 { 3505 int ioapic, pin, idx; 3506 3507 if (skip_ioapic_setup) 3508 return -1; 3509 3510 ioapic = mp_find_ioapic(gsi); 3511 if (ioapic < 0) 3512 return -1; 3513 3514 pin = mp_find_ioapic_pin(ioapic, gsi); 3515 if (pin < 0) 3516 return -1; 3517 3518 idx = find_irq_entry(ioapic, pin, mp_INT); 3519 if (idx < 0) 3520 return -1; 3521 3522 *trigger = irq_trigger(idx); 3523 *polarity = irq_polarity(idx); 3524 return 0; 3525 } 3526 3527 /* 3528 * This function currently is only a helper for the i386 smp boot process where 3529 * we need to reprogram the ioredtbls to cater for the cpus which have come online 3530 * so mask in all cases should simply be apic->target_cpus() 3531 */ 3532 #ifdef CONFIG_SMP 3533 void __init setup_ioapic_dest(void) 3534 { 3535 int pin, ioapic, irq, irq_entry; 3536 const struct cpumask *mask; 3537 struct irq_data *idata; 3538 3539 if (skip_ioapic_setup == 1) 3540 return; 3541 3542 for_each_ioapic_pin(ioapic, pin) { 3543 irq_entry = find_irq_entry(ioapic, pin, mp_INT); 3544 if (irq_entry == -1) 3545 continue; 3546 3547 irq = pin_2_irq(irq_entry, ioapic, pin); 3548 if (!mp_init_irq_at_boot(ioapic, irq)) 3549 continue; 3550 3551 idata = irq_get_irq_data(irq); 3552 3553 /* 3554 * Honour affinities which have been set in early boot 3555 */ 3556 if (!irqd_can_balance(idata) || irqd_affinity_was_set(idata)) 3557 mask = idata->affinity; 3558 else 3559 mask = apic->target_cpus(); 3560 3561 x86_io_apic_ops.set_affinity(idata, mask, false); 3562 } 3563 3564 } 3565 #endif 3566 3567 #define IOAPIC_RESOURCE_NAME_SIZE 11 3568 3569 static struct resource *ioapic_resources; 3570 3571 static struct resource * __init ioapic_setup_resources(void) 3572 { 3573 unsigned long n; 3574 struct resource *res; 3575 char *mem; 3576 int i, num = 0; 3577 3578 for_each_ioapic(i) 3579 num++; 3580 if (num == 0) 3581 return NULL; 3582 3583 n = IOAPIC_RESOURCE_NAME_SIZE + sizeof(struct resource); 3584 n *= num; 3585 3586 mem = alloc_bootmem(n); 3587 res = (void *)mem; 3588 3589 mem += sizeof(struct resource) * num; 3590 3591 num = 0; 3592 for_each_ioapic(i) { 3593 res[num].name = mem; 3594 res[num].flags = IORESOURCE_MEM | IORESOURCE_BUSY; 3595 snprintf(mem, IOAPIC_RESOURCE_NAME_SIZE, "IOAPIC %u", i); 3596 mem += IOAPIC_RESOURCE_NAME_SIZE; 3597 num++; 3598 } 3599 3600 ioapic_resources = res; 3601 3602 return res; 3603 } 3604 3605 void __init native_io_apic_init_mappings(void) 3606 { 3607 unsigned long ioapic_phys, idx = FIX_IO_APIC_BASE_0; 3608 struct resource *ioapic_res; 3609 int i; 3610 3611 ioapic_res = ioapic_setup_resources(); 3612 for_each_ioapic(i) { 3613 if (smp_found_config) { 3614 ioapic_phys = mpc_ioapic_addr(i); 3615 #ifdef CONFIG_X86_32 3616 if (!ioapic_phys) { 3617 printk(KERN_ERR 3618 "WARNING: bogus zero IO-APIC " 3619 "address found in MPTABLE, " 3620 "disabling IO/APIC support!\n"); 3621 smp_found_config = 0; 3622 skip_ioapic_setup = 1; 3623 goto fake_ioapic_page; 3624 } 3625 #endif 3626 } else { 3627 #ifdef CONFIG_X86_32 3628 fake_ioapic_page: 3629 #endif 3630 ioapic_phys = (unsigned long)alloc_bootmem_pages(PAGE_SIZE); 3631 ioapic_phys = __pa(ioapic_phys); 3632 } 3633 set_fixmap_nocache(idx, ioapic_phys); 3634 apic_printk(APIC_VERBOSE, "mapped IOAPIC to %08lx (%08lx)\n", 3635 __fix_to_virt(idx) + (ioapic_phys & ~PAGE_MASK), 3636 ioapic_phys); 3637 idx++; 3638 3639 ioapic_res->start = ioapic_phys; 3640 ioapic_res->end = ioapic_phys + IO_APIC_SLOT_SIZE - 1; 3641 ioapic_res++; 3642 } 3643 } 3644 3645 void __init ioapic_insert_resources(void) 3646 { 3647 int i; 3648 struct resource *r = ioapic_resources; 3649 3650 if (!r) { 3651 if (nr_ioapics > 0) 3652 printk(KERN_ERR 3653 "IO APIC resources couldn't be allocated.\n"); 3654 return; 3655 } 3656 3657 for_each_ioapic(i) { 3658 insert_resource(&iomem_resource, r); 3659 r++; 3660 } 3661 } 3662 3663 int mp_find_ioapic(u32 gsi) 3664 { 3665 int i; 3666 3667 if (nr_ioapics == 0) 3668 return -1; 3669 3670 /* Find the IOAPIC that manages this GSI. */ 3671 for_each_ioapic(i) { 3672 struct mp_ioapic_gsi *gsi_cfg = mp_ioapic_gsi_routing(i); 3673 if (gsi >= gsi_cfg->gsi_base && gsi <= gsi_cfg->gsi_end) 3674 return i; 3675 } 3676 3677 printk(KERN_ERR "ERROR: Unable to locate IOAPIC for GSI %d\n", gsi); 3678 return -1; 3679 } 3680 3681 int mp_find_ioapic_pin(int ioapic, u32 gsi) 3682 { 3683 struct mp_ioapic_gsi *gsi_cfg; 3684 3685 if (WARN_ON(ioapic < 0)) 3686 return -1; 3687 3688 gsi_cfg = mp_ioapic_gsi_routing(ioapic); 3689 if (WARN_ON(gsi > gsi_cfg->gsi_end)) 3690 return -1; 3691 3692 return gsi - gsi_cfg->gsi_base; 3693 } 3694 3695 static __init int bad_ioapic(unsigned long address) 3696 { 3697 if (nr_ioapics >= MAX_IO_APICS) { 3698 pr_warn("WARNING: Max # of I/O APICs (%d) exceeded (found %d), skipping\n", 3699 MAX_IO_APICS, nr_ioapics); 3700 return 1; 3701 } 3702 if (!address) { 3703 pr_warn("WARNING: Bogus (zero) I/O APIC address found in table, skipping!\n"); 3704 return 1; 3705 } 3706 return 0; 3707 } 3708 3709 static __init int bad_ioapic_register(int idx) 3710 { 3711 union IO_APIC_reg_00 reg_00; 3712 union IO_APIC_reg_01 reg_01; 3713 union IO_APIC_reg_02 reg_02; 3714 3715 reg_00.raw = io_apic_read(idx, 0); 3716 reg_01.raw = io_apic_read(idx, 1); 3717 reg_02.raw = io_apic_read(idx, 2); 3718 3719 if (reg_00.raw == -1 && reg_01.raw == -1 && reg_02.raw == -1) { 3720 pr_warn("I/O APIC 0x%x registers return all ones, skipping!\n", 3721 mpc_ioapic_addr(idx)); 3722 return 1; 3723 } 3724 3725 return 0; 3726 } 3727 3728 void __init mp_register_ioapic(int id, u32 address, u32 gsi_base) 3729 { 3730 int idx = 0; 3731 int entries; 3732 struct mp_ioapic_gsi *gsi_cfg; 3733 3734 if (bad_ioapic(address)) 3735 return; 3736 3737 idx = nr_ioapics; 3738 3739 ioapics[idx].mp_config.type = MP_IOAPIC; 3740 ioapics[idx].mp_config.flags = MPC_APIC_USABLE; 3741 ioapics[idx].mp_config.apicaddr = address; 3742 3743 set_fixmap_nocache(FIX_IO_APIC_BASE_0 + idx, address); 3744 3745 if (bad_ioapic_register(idx)) { 3746 clear_fixmap(FIX_IO_APIC_BASE_0 + idx); 3747 return; 3748 } 3749 3750 ioapics[idx].mp_config.apicid = io_apic_unique_id(id); 3751 ioapics[idx].mp_config.apicver = io_apic_get_version(idx); 3752 3753 /* 3754 * Build basic GSI lookup table to facilitate gsi->io_apic lookups 3755 * and to prevent reprogramming of IOAPIC pins (PCI GSIs). 3756 */ 3757 entries = io_apic_get_redir_entries(idx); 3758 gsi_cfg = mp_ioapic_gsi_routing(idx); 3759 gsi_cfg->gsi_base = gsi_base; 3760 gsi_cfg->gsi_end = gsi_base + entries - 1; 3761 3762 /* 3763 * The number of IO-APIC IRQ registers (== #pins): 3764 */ 3765 ioapics[idx].nr_registers = entries; 3766 3767 if (gsi_cfg->gsi_end >= gsi_top) 3768 gsi_top = gsi_cfg->gsi_end + 1; 3769 3770 pr_info("IOAPIC[%d]: apic_id %d, version %d, address 0x%x, GSI %d-%d\n", 3771 idx, mpc_ioapic_id(idx), 3772 mpc_ioapic_ver(idx), mpc_ioapic_addr(idx), 3773 gsi_cfg->gsi_base, gsi_cfg->gsi_end); 3774 3775 nr_ioapics++; 3776 } 3777 3778 /* Enable IOAPIC early just for system timer */ 3779 void __init pre_init_apic_IRQ0(void) 3780 { 3781 struct io_apic_irq_attr attr = { 0, 0, 0, 0 }; 3782 3783 printk(KERN_INFO "Early APIC setup for system timer0\n"); 3784 #ifndef CONFIG_SMP 3785 physid_set_mask_of_physid(boot_cpu_physical_apicid, 3786 &phys_cpu_present_map); 3787 #endif 3788 setup_local_APIC(); 3789 3790 io_apic_setup_irq_pin(0, 0, &attr); 3791 irq_set_chip_and_handler_name(0, &ioapic_chip, handle_edge_irq, 3792 "edge"); 3793 } 3794