1 // SPDX-License-Identifier: GPL-2.0 2 /* 3 * PCI detection and setup code 4 */ 5 6 #include <linux/kernel.h> 7 #include <linux/delay.h> 8 #include <linux/init.h> 9 #include <linux/pci.h> 10 #include <linux/of_device.h> 11 #include <linux/of_pci.h> 12 #include <linux/pci_hotplug.h> 13 #include <linux/slab.h> 14 #include <linux/module.h> 15 #include <linux/cpumask.h> 16 #include <linux/aer.h> 17 #include <linux/acpi.h> 18 #include <linux/hypervisor.h> 19 #include <linux/irqdomain.h> 20 #include <linux/pm_runtime.h> 21 #include "pci.h" 22 23 #define CARDBUS_LATENCY_TIMER 176 /* secondary latency timer */ 24 #define CARDBUS_RESERVE_BUSNR 3 25 26 static struct resource busn_resource = { 27 .name = "PCI busn", 28 .start = 0, 29 .end = 255, 30 .flags = IORESOURCE_BUS, 31 }; 32 33 /* Ugh. Need to stop exporting this to modules. */ 34 LIST_HEAD(pci_root_buses); 35 EXPORT_SYMBOL(pci_root_buses); 36 37 static LIST_HEAD(pci_domain_busn_res_list); 38 39 struct pci_domain_busn_res { 40 struct list_head list; 41 struct resource res; 42 int domain_nr; 43 }; 44 45 static struct resource *get_pci_domain_busn_res(int domain_nr) 46 { 47 struct pci_domain_busn_res *r; 48 49 list_for_each_entry(r, &pci_domain_busn_res_list, list) 50 if (r->domain_nr == domain_nr) 51 return &r->res; 52 53 r = kzalloc(sizeof(*r), GFP_KERNEL); 54 if (!r) 55 return NULL; 56 57 r->domain_nr = domain_nr; 58 r->res.start = 0; 59 r->res.end = 0xff; 60 r->res.flags = IORESOURCE_BUS | IORESOURCE_PCI_FIXED; 61 62 list_add_tail(&r->list, &pci_domain_busn_res_list); 63 64 return &r->res; 65 } 66 67 static int find_anything(struct device *dev, void *data) 68 { 69 return 1; 70 } 71 72 /* 73 * Some device drivers need know if PCI is initiated. 74 * Basically, we think PCI is not initiated when there 75 * is no device to be found on the pci_bus_type. 76 */ 77 int no_pci_devices(void) 78 { 79 struct device *dev; 80 int no_devices; 81 82 dev = bus_find_device(&pci_bus_type, NULL, NULL, find_anything); 83 no_devices = (dev == NULL); 84 put_device(dev); 85 return no_devices; 86 } 87 EXPORT_SYMBOL(no_pci_devices); 88 89 /* 90 * PCI Bus Class 91 */ 92 static void release_pcibus_dev(struct device *dev) 93 { 94 struct pci_bus *pci_bus = to_pci_bus(dev); 95 96 put_device(pci_bus->bridge); 97 pci_bus_remove_resources(pci_bus); 98 pci_release_bus_of_node(pci_bus); 99 kfree(pci_bus); 100 } 101 102 static struct class pcibus_class = { 103 .name = "pci_bus", 104 .dev_release = &release_pcibus_dev, 105 .dev_groups = pcibus_groups, 106 }; 107 108 static int __init pcibus_class_init(void) 109 { 110 return class_register(&pcibus_class); 111 } 112 postcore_initcall(pcibus_class_init); 113 114 static u64 pci_size(u64 base, u64 maxbase, u64 mask) 115 { 116 u64 size = mask & maxbase; /* Find the significant bits */ 117 if (!size) 118 return 0; 119 120 /* 121 * Get the lowest of them to find the decode size, and from that 122 * the extent. 123 */ 124 size = (size & ~(size-1)) - 1; 125 126 /* 127 * base == maxbase can be valid only if the BAR has already been 128 * programmed with all 1s. 129 */ 130 if (base == maxbase && ((base | size) & mask) != mask) 131 return 0; 132 133 return size; 134 } 135 136 static inline unsigned long decode_bar(struct pci_dev *dev, u32 bar) 137 { 138 u32 mem_type; 139 unsigned long flags; 140 141 if ((bar & PCI_BASE_ADDRESS_SPACE) == PCI_BASE_ADDRESS_SPACE_IO) { 142 flags = bar & ~PCI_BASE_ADDRESS_IO_MASK; 143 flags |= IORESOURCE_IO; 144 return flags; 145 } 146 147 flags = bar & ~PCI_BASE_ADDRESS_MEM_MASK; 148 flags |= IORESOURCE_MEM; 149 if (flags & PCI_BASE_ADDRESS_MEM_PREFETCH) 150 flags |= IORESOURCE_PREFETCH; 151 152 mem_type = bar & PCI_BASE_ADDRESS_MEM_TYPE_MASK; 153 switch (mem_type) { 154 case PCI_BASE_ADDRESS_MEM_TYPE_32: 155 break; 156 case PCI_BASE_ADDRESS_MEM_TYPE_1M: 157 /* 1M mem BAR treated as 32-bit BAR */ 158 break; 159 case PCI_BASE_ADDRESS_MEM_TYPE_64: 160 flags |= IORESOURCE_MEM_64; 161 break; 162 default: 163 /* mem unknown type treated as 32-bit BAR */ 164 break; 165 } 166 return flags; 167 } 168 169 #define PCI_COMMAND_DECODE_ENABLE (PCI_COMMAND_MEMORY | PCI_COMMAND_IO) 170 171 /** 172 * pci_read_base - Read a PCI BAR 173 * @dev: the PCI device 174 * @type: type of the BAR 175 * @res: resource buffer to be filled in 176 * @pos: BAR position in the config space 177 * 178 * Returns 1 if the BAR is 64-bit, or 0 if 32-bit. 179 */ 180 int __pci_read_base(struct pci_dev *dev, enum pci_bar_type type, 181 struct resource *res, unsigned int pos) 182 { 183 u32 l = 0, sz = 0, mask; 184 u64 l64, sz64, mask64; 185 u16 orig_cmd; 186 struct pci_bus_region region, inverted_region; 187 188 mask = type ? PCI_ROM_ADDRESS_MASK : ~0; 189 190 /* No printks while decoding is disabled! */ 191 if (!dev->mmio_always_on) { 192 pci_read_config_word(dev, PCI_COMMAND, &orig_cmd); 193 if (orig_cmd & PCI_COMMAND_DECODE_ENABLE) { 194 pci_write_config_word(dev, PCI_COMMAND, 195 orig_cmd & ~PCI_COMMAND_DECODE_ENABLE); 196 } 197 } 198 199 res->name = pci_name(dev); 200 201 pci_read_config_dword(dev, pos, &l); 202 pci_write_config_dword(dev, pos, l | mask); 203 pci_read_config_dword(dev, pos, &sz); 204 pci_write_config_dword(dev, pos, l); 205 206 /* 207 * All bits set in sz means the device isn't working properly. 208 * If the BAR isn't implemented, all bits must be 0. If it's a 209 * memory BAR or a ROM, bit 0 must be clear; if it's an io BAR, bit 210 * 1 must be clear. 211 */ 212 if (sz == 0xffffffff) 213 sz = 0; 214 215 /* 216 * I don't know how l can have all bits set. Copied from old code. 217 * Maybe it fixes a bug on some ancient platform. 218 */ 219 if (l == 0xffffffff) 220 l = 0; 221 222 if (type == pci_bar_unknown) { 223 res->flags = decode_bar(dev, l); 224 res->flags |= IORESOURCE_SIZEALIGN; 225 if (res->flags & IORESOURCE_IO) { 226 l64 = l & PCI_BASE_ADDRESS_IO_MASK; 227 sz64 = sz & PCI_BASE_ADDRESS_IO_MASK; 228 mask64 = PCI_BASE_ADDRESS_IO_MASK & (u32)IO_SPACE_LIMIT; 229 } else { 230 l64 = l & PCI_BASE_ADDRESS_MEM_MASK; 231 sz64 = sz & PCI_BASE_ADDRESS_MEM_MASK; 232 mask64 = (u32)PCI_BASE_ADDRESS_MEM_MASK; 233 } 234 } else { 235 if (l & PCI_ROM_ADDRESS_ENABLE) 236 res->flags |= IORESOURCE_ROM_ENABLE; 237 l64 = l & PCI_ROM_ADDRESS_MASK; 238 sz64 = sz & PCI_ROM_ADDRESS_MASK; 239 mask64 = PCI_ROM_ADDRESS_MASK; 240 } 241 242 if (res->flags & IORESOURCE_MEM_64) { 243 pci_read_config_dword(dev, pos + 4, &l); 244 pci_write_config_dword(dev, pos + 4, ~0); 245 pci_read_config_dword(dev, pos + 4, &sz); 246 pci_write_config_dword(dev, pos + 4, l); 247 248 l64 |= ((u64)l << 32); 249 sz64 |= ((u64)sz << 32); 250 mask64 |= ((u64)~0 << 32); 251 } 252 253 if (!dev->mmio_always_on && (orig_cmd & PCI_COMMAND_DECODE_ENABLE)) 254 pci_write_config_word(dev, PCI_COMMAND, orig_cmd); 255 256 if (!sz64) 257 goto fail; 258 259 sz64 = pci_size(l64, sz64, mask64); 260 if (!sz64) { 261 pci_info(dev, FW_BUG "reg 0x%x: invalid BAR (can't size)\n", 262 pos); 263 goto fail; 264 } 265 266 if (res->flags & IORESOURCE_MEM_64) { 267 if ((sizeof(pci_bus_addr_t) < 8 || sizeof(resource_size_t) < 8) 268 && sz64 > 0x100000000ULL) { 269 res->flags |= IORESOURCE_UNSET | IORESOURCE_DISABLED; 270 res->start = 0; 271 res->end = 0; 272 pci_err(dev, "reg 0x%x: can't handle BAR larger than 4GB (size %#010llx)\n", 273 pos, (unsigned long long)sz64); 274 goto out; 275 } 276 277 if ((sizeof(pci_bus_addr_t) < 8) && l) { 278 /* Above 32-bit boundary; try to reallocate */ 279 res->flags |= IORESOURCE_UNSET; 280 res->start = 0; 281 res->end = sz64; 282 pci_info(dev, "reg 0x%x: can't handle BAR above 4GB (bus address %#010llx)\n", 283 pos, (unsigned long long)l64); 284 goto out; 285 } 286 } 287 288 region.start = l64; 289 region.end = l64 + sz64; 290 291 pcibios_bus_to_resource(dev->bus, res, ®ion); 292 pcibios_resource_to_bus(dev->bus, &inverted_region, res); 293 294 /* 295 * If "A" is a BAR value (a bus address), "bus_to_resource(A)" is 296 * the corresponding resource address (the physical address used by 297 * the CPU. Converting that resource address back to a bus address 298 * should yield the original BAR value: 299 * 300 * resource_to_bus(bus_to_resource(A)) == A 301 * 302 * If it doesn't, CPU accesses to "bus_to_resource(A)" will not 303 * be claimed by the device. 304 */ 305 if (inverted_region.start != region.start) { 306 res->flags |= IORESOURCE_UNSET; 307 res->start = 0; 308 res->end = region.end - region.start; 309 pci_info(dev, "reg 0x%x: initial BAR value %#010llx invalid\n", 310 pos, (unsigned long long)region.start); 311 } 312 313 goto out; 314 315 316 fail: 317 res->flags = 0; 318 out: 319 if (res->flags) 320 pci_printk(KERN_DEBUG, dev, "reg 0x%x: %pR\n", pos, res); 321 322 return (res->flags & IORESOURCE_MEM_64) ? 1 : 0; 323 } 324 325 static void pci_read_bases(struct pci_dev *dev, unsigned int howmany, int rom) 326 { 327 unsigned int pos, reg; 328 329 if (dev->non_compliant_bars) 330 return; 331 332 /* Per PCIe r4.0, sec 9.3.4.1.11, the VF BARs are all RO Zero */ 333 if (dev->is_virtfn) 334 return; 335 336 for (pos = 0; pos < howmany; pos++) { 337 struct resource *res = &dev->resource[pos]; 338 reg = PCI_BASE_ADDRESS_0 + (pos << 2); 339 pos += __pci_read_base(dev, pci_bar_unknown, res, reg); 340 } 341 342 if (rom) { 343 struct resource *res = &dev->resource[PCI_ROM_RESOURCE]; 344 dev->rom_base_reg = rom; 345 res->flags = IORESOURCE_MEM | IORESOURCE_PREFETCH | 346 IORESOURCE_READONLY | IORESOURCE_SIZEALIGN; 347 __pci_read_base(dev, pci_bar_mem32, res, rom); 348 } 349 } 350 351 static void pci_read_bridge_io(struct pci_bus *child) 352 { 353 struct pci_dev *dev = child->self; 354 u8 io_base_lo, io_limit_lo; 355 unsigned long io_mask, io_granularity, base, limit; 356 struct pci_bus_region region; 357 struct resource *res; 358 359 io_mask = PCI_IO_RANGE_MASK; 360 io_granularity = 0x1000; 361 if (dev->io_window_1k) { 362 /* Support 1K I/O space granularity */ 363 io_mask = PCI_IO_1K_RANGE_MASK; 364 io_granularity = 0x400; 365 } 366 367 res = child->resource[0]; 368 pci_read_config_byte(dev, PCI_IO_BASE, &io_base_lo); 369 pci_read_config_byte(dev, PCI_IO_LIMIT, &io_limit_lo); 370 base = (io_base_lo & io_mask) << 8; 371 limit = (io_limit_lo & io_mask) << 8; 372 373 if ((io_base_lo & PCI_IO_RANGE_TYPE_MASK) == PCI_IO_RANGE_TYPE_32) { 374 u16 io_base_hi, io_limit_hi; 375 376 pci_read_config_word(dev, PCI_IO_BASE_UPPER16, &io_base_hi); 377 pci_read_config_word(dev, PCI_IO_LIMIT_UPPER16, &io_limit_hi); 378 base |= ((unsigned long) io_base_hi << 16); 379 limit |= ((unsigned long) io_limit_hi << 16); 380 } 381 382 if (base <= limit) { 383 res->flags = (io_base_lo & PCI_IO_RANGE_TYPE_MASK) | IORESOURCE_IO; 384 region.start = base; 385 region.end = limit + io_granularity - 1; 386 pcibios_bus_to_resource(dev->bus, res, ®ion); 387 pci_printk(KERN_DEBUG, dev, " bridge window %pR\n", res); 388 } 389 } 390 391 static void pci_read_bridge_mmio(struct pci_bus *child) 392 { 393 struct pci_dev *dev = child->self; 394 u16 mem_base_lo, mem_limit_lo; 395 unsigned long base, limit; 396 struct pci_bus_region region; 397 struct resource *res; 398 399 res = child->resource[1]; 400 pci_read_config_word(dev, PCI_MEMORY_BASE, &mem_base_lo); 401 pci_read_config_word(dev, PCI_MEMORY_LIMIT, &mem_limit_lo); 402 base = ((unsigned long) mem_base_lo & PCI_MEMORY_RANGE_MASK) << 16; 403 limit = ((unsigned long) mem_limit_lo & PCI_MEMORY_RANGE_MASK) << 16; 404 if (base <= limit) { 405 res->flags = (mem_base_lo & PCI_MEMORY_RANGE_TYPE_MASK) | IORESOURCE_MEM; 406 region.start = base; 407 region.end = limit + 0xfffff; 408 pcibios_bus_to_resource(dev->bus, res, ®ion); 409 pci_printk(KERN_DEBUG, dev, " bridge window %pR\n", res); 410 } 411 } 412 413 static void pci_read_bridge_mmio_pref(struct pci_bus *child) 414 { 415 struct pci_dev *dev = child->self; 416 u16 mem_base_lo, mem_limit_lo; 417 u64 base64, limit64; 418 pci_bus_addr_t base, limit; 419 struct pci_bus_region region; 420 struct resource *res; 421 422 res = child->resource[2]; 423 pci_read_config_word(dev, PCI_PREF_MEMORY_BASE, &mem_base_lo); 424 pci_read_config_word(dev, PCI_PREF_MEMORY_LIMIT, &mem_limit_lo); 425 base64 = (mem_base_lo & PCI_PREF_RANGE_MASK) << 16; 426 limit64 = (mem_limit_lo & PCI_PREF_RANGE_MASK) << 16; 427 428 if ((mem_base_lo & PCI_PREF_RANGE_TYPE_MASK) == PCI_PREF_RANGE_TYPE_64) { 429 u32 mem_base_hi, mem_limit_hi; 430 431 pci_read_config_dword(dev, PCI_PREF_BASE_UPPER32, &mem_base_hi); 432 pci_read_config_dword(dev, PCI_PREF_LIMIT_UPPER32, &mem_limit_hi); 433 434 /* 435 * Some bridges set the base > limit by default, and some 436 * (broken) BIOSes do not initialize them. If we find 437 * this, just assume they are not being used. 438 */ 439 if (mem_base_hi <= mem_limit_hi) { 440 base64 |= (u64) mem_base_hi << 32; 441 limit64 |= (u64) mem_limit_hi << 32; 442 } 443 } 444 445 base = (pci_bus_addr_t) base64; 446 limit = (pci_bus_addr_t) limit64; 447 448 if (base != base64) { 449 pci_err(dev, "can't handle bridge window above 4GB (bus address %#010llx)\n", 450 (unsigned long long) base64); 451 return; 452 } 453 454 if (base <= limit) { 455 res->flags = (mem_base_lo & PCI_PREF_RANGE_TYPE_MASK) | 456 IORESOURCE_MEM | IORESOURCE_PREFETCH; 457 if (res->flags & PCI_PREF_RANGE_TYPE_64) 458 res->flags |= IORESOURCE_MEM_64; 459 region.start = base; 460 region.end = limit + 0xfffff; 461 pcibios_bus_to_resource(dev->bus, res, ®ion); 462 pci_printk(KERN_DEBUG, dev, " bridge window %pR\n", res); 463 } 464 } 465 466 void pci_read_bridge_bases(struct pci_bus *child) 467 { 468 struct pci_dev *dev = child->self; 469 struct resource *res; 470 int i; 471 472 if (pci_is_root_bus(child)) /* It's a host bus, nothing to read */ 473 return; 474 475 pci_info(dev, "PCI bridge to %pR%s\n", 476 &child->busn_res, 477 dev->transparent ? " (subtractive decode)" : ""); 478 479 pci_bus_remove_resources(child); 480 for (i = 0; i < PCI_BRIDGE_RESOURCE_NUM; i++) 481 child->resource[i] = &dev->resource[PCI_BRIDGE_RESOURCES+i]; 482 483 pci_read_bridge_io(child); 484 pci_read_bridge_mmio(child); 485 pci_read_bridge_mmio_pref(child); 486 487 if (dev->transparent) { 488 pci_bus_for_each_resource(child->parent, res, i) { 489 if (res && res->flags) { 490 pci_bus_add_resource(child, res, 491 PCI_SUBTRACTIVE_DECODE); 492 pci_printk(KERN_DEBUG, dev, 493 " bridge window %pR (subtractive decode)\n", 494 res); 495 } 496 } 497 } 498 } 499 500 static struct pci_bus *pci_alloc_bus(struct pci_bus *parent) 501 { 502 struct pci_bus *b; 503 504 b = kzalloc(sizeof(*b), GFP_KERNEL); 505 if (!b) 506 return NULL; 507 508 INIT_LIST_HEAD(&b->node); 509 INIT_LIST_HEAD(&b->children); 510 INIT_LIST_HEAD(&b->devices); 511 INIT_LIST_HEAD(&b->slots); 512 INIT_LIST_HEAD(&b->resources); 513 b->max_bus_speed = PCI_SPEED_UNKNOWN; 514 b->cur_bus_speed = PCI_SPEED_UNKNOWN; 515 #ifdef CONFIG_PCI_DOMAINS_GENERIC 516 if (parent) 517 b->domain_nr = parent->domain_nr; 518 #endif 519 return b; 520 } 521 522 static void devm_pci_release_host_bridge_dev(struct device *dev) 523 { 524 struct pci_host_bridge *bridge = to_pci_host_bridge(dev); 525 526 if (bridge->release_fn) 527 bridge->release_fn(bridge); 528 529 pci_free_resource_list(&bridge->windows); 530 } 531 532 static void pci_release_host_bridge_dev(struct device *dev) 533 { 534 devm_pci_release_host_bridge_dev(dev); 535 kfree(to_pci_host_bridge(dev)); 536 } 537 538 struct pci_host_bridge *pci_alloc_host_bridge(size_t priv) 539 { 540 struct pci_host_bridge *bridge; 541 542 bridge = kzalloc(sizeof(*bridge) + priv, GFP_KERNEL); 543 if (!bridge) 544 return NULL; 545 546 INIT_LIST_HEAD(&bridge->windows); 547 bridge->dev.release = pci_release_host_bridge_dev; 548 549 /* 550 * We assume we can manage these PCIe features. Some systems may 551 * reserve these for use by the platform itself, e.g., an ACPI BIOS 552 * may implement its own AER handling and use _OSC to prevent the 553 * OS from interfering. 554 */ 555 bridge->native_aer = 1; 556 bridge->native_pcie_hotplug = 1; 557 bridge->native_shpc_hotplug = 1; 558 bridge->native_pme = 1; 559 bridge->native_ltr = 1; 560 561 return bridge; 562 } 563 EXPORT_SYMBOL(pci_alloc_host_bridge); 564 565 struct pci_host_bridge *devm_pci_alloc_host_bridge(struct device *dev, 566 size_t priv) 567 { 568 struct pci_host_bridge *bridge; 569 570 bridge = devm_kzalloc(dev, sizeof(*bridge) + priv, GFP_KERNEL); 571 if (!bridge) 572 return NULL; 573 574 INIT_LIST_HEAD(&bridge->windows); 575 bridge->dev.release = devm_pci_release_host_bridge_dev; 576 577 return bridge; 578 } 579 EXPORT_SYMBOL(devm_pci_alloc_host_bridge); 580 581 void pci_free_host_bridge(struct pci_host_bridge *bridge) 582 { 583 pci_free_resource_list(&bridge->windows); 584 585 kfree(bridge); 586 } 587 EXPORT_SYMBOL(pci_free_host_bridge); 588 589 static const unsigned char pcix_bus_speed[] = { 590 PCI_SPEED_UNKNOWN, /* 0 */ 591 PCI_SPEED_66MHz_PCIX, /* 1 */ 592 PCI_SPEED_100MHz_PCIX, /* 2 */ 593 PCI_SPEED_133MHz_PCIX, /* 3 */ 594 PCI_SPEED_UNKNOWN, /* 4 */ 595 PCI_SPEED_66MHz_PCIX_ECC, /* 5 */ 596 PCI_SPEED_100MHz_PCIX_ECC, /* 6 */ 597 PCI_SPEED_133MHz_PCIX_ECC, /* 7 */ 598 PCI_SPEED_UNKNOWN, /* 8 */ 599 PCI_SPEED_66MHz_PCIX_266, /* 9 */ 600 PCI_SPEED_100MHz_PCIX_266, /* A */ 601 PCI_SPEED_133MHz_PCIX_266, /* B */ 602 PCI_SPEED_UNKNOWN, /* C */ 603 PCI_SPEED_66MHz_PCIX_533, /* D */ 604 PCI_SPEED_100MHz_PCIX_533, /* E */ 605 PCI_SPEED_133MHz_PCIX_533 /* F */ 606 }; 607 608 const unsigned char pcie_link_speed[] = { 609 PCI_SPEED_UNKNOWN, /* 0 */ 610 PCIE_SPEED_2_5GT, /* 1 */ 611 PCIE_SPEED_5_0GT, /* 2 */ 612 PCIE_SPEED_8_0GT, /* 3 */ 613 PCIE_SPEED_16_0GT, /* 4 */ 614 PCI_SPEED_UNKNOWN, /* 5 */ 615 PCI_SPEED_UNKNOWN, /* 6 */ 616 PCI_SPEED_UNKNOWN, /* 7 */ 617 PCI_SPEED_UNKNOWN, /* 8 */ 618 PCI_SPEED_UNKNOWN, /* 9 */ 619 PCI_SPEED_UNKNOWN, /* A */ 620 PCI_SPEED_UNKNOWN, /* B */ 621 PCI_SPEED_UNKNOWN, /* C */ 622 PCI_SPEED_UNKNOWN, /* D */ 623 PCI_SPEED_UNKNOWN, /* E */ 624 PCI_SPEED_UNKNOWN /* F */ 625 }; 626 627 void pcie_update_link_speed(struct pci_bus *bus, u16 linksta) 628 { 629 bus->cur_bus_speed = pcie_link_speed[linksta & PCI_EXP_LNKSTA_CLS]; 630 } 631 EXPORT_SYMBOL_GPL(pcie_update_link_speed); 632 633 static unsigned char agp_speeds[] = { 634 AGP_UNKNOWN, 635 AGP_1X, 636 AGP_2X, 637 AGP_4X, 638 AGP_8X 639 }; 640 641 static enum pci_bus_speed agp_speed(int agp3, int agpstat) 642 { 643 int index = 0; 644 645 if (agpstat & 4) 646 index = 3; 647 else if (agpstat & 2) 648 index = 2; 649 else if (agpstat & 1) 650 index = 1; 651 else 652 goto out; 653 654 if (agp3) { 655 index += 2; 656 if (index == 5) 657 index = 0; 658 } 659 660 out: 661 return agp_speeds[index]; 662 } 663 664 static void pci_set_bus_speed(struct pci_bus *bus) 665 { 666 struct pci_dev *bridge = bus->self; 667 int pos; 668 669 pos = pci_find_capability(bridge, PCI_CAP_ID_AGP); 670 if (!pos) 671 pos = pci_find_capability(bridge, PCI_CAP_ID_AGP3); 672 if (pos) { 673 u32 agpstat, agpcmd; 674 675 pci_read_config_dword(bridge, pos + PCI_AGP_STATUS, &agpstat); 676 bus->max_bus_speed = agp_speed(agpstat & 8, agpstat & 7); 677 678 pci_read_config_dword(bridge, pos + PCI_AGP_COMMAND, &agpcmd); 679 bus->cur_bus_speed = agp_speed(agpstat & 8, agpcmd & 7); 680 } 681 682 pos = pci_find_capability(bridge, PCI_CAP_ID_PCIX); 683 if (pos) { 684 u16 status; 685 enum pci_bus_speed max; 686 687 pci_read_config_word(bridge, pos + PCI_X_BRIDGE_SSTATUS, 688 &status); 689 690 if (status & PCI_X_SSTATUS_533MHZ) { 691 max = PCI_SPEED_133MHz_PCIX_533; 692 } else if (status & PCI_X_SSTATUS_266MHZ) { 693 max = PCI_SPEED_133MHz_PCIX_266; 694 } else if (status & PCI_X_SSTATUS_133MHZ) { 695 if ((status & PCI_X_SSTATUS_VERS) == PCI_X_SSTATUS_V2) 696 max = PCI_SPEED_133MHz_PCIX_ECC; 697 else 698 max = PCI_SPEED_133MHz_PCIX; 699 } else { 700 max = PCI_SPEED_66MHz_PCIX; 701 } 702 703 bus->max_bus_speed = max; 704 bus->cur_bus_speed = pcix_bus_speed[ 705 (status & PCI_X_SSTATUS_FREQ) >> 6]; 706 707 return; 708 } 709 710 if (pci_is_pcie(bridge)) { 711 u32 linkcap; 712 u16 linksta; 713 714 pcie_capability_read_dword(bridge, PCI_EXP_LNKCAP, &linkcap); 715 bus->max_bus_speed = pcie_link_speed[linkcap & PCI_EXP_LNKCAP_SLS]; 716 bridge->link_active_reporting = !!(linkcap & PCI_EXP_LNKCAP_DLLLARC); 717 718 pcie_capability_read_word(bridge, PCI_EXP_LNKSTA, &linksta); 719 pcie_update_link_speed(bus, linksta); 720 } 721 } 722 723 static struct irq_domain *pci_host_bridge_msi_domain(struct pci_bus *bus) 724 { 725 struct irq_domain *d; 726 727 /* 728 * Any firmware interface that can resolve the msi_domain 729 * should be called from here. 730 */ 731 d = pci_host_bridge_of_msi_domain(bus); 732 if (!d) 733 d = pci_host_bridge_acpi_msi_domain(bus); 734 735 #ifdef CONFIG_PCI_MSI_IRQ_DOMAIN 736 /* 737 * If no IRQ domain was found via the OF tree, try looking it up 738 * directly through the fwnode_handle. 739 */ 740 if (!d) { 741 struct fwnode_handle *fwnode = pci_root_bus_fwnode(bus); 742 743 if (fwnode) 744 d = irq_find_matching_fwnode(fwnode, 745 DOMAIN_BUS_PCI_MSI); 746 } 747 #endif 748 749 return d; 750 } 751 752 static void pci_set_bus_msi_domain(struct pci_bus *bus) 753 { 754 struct irq_domain *d; 755 struct pci_bus *b; 756 757 /* 758 * The bus can be a root bus, a subordinate bus, or a virtual bus 759 * created by an SR-IOV device. Walk up to the first bridge device 760 * found or derive the domain from the host bridge. 761 */ 762 for (b = bus, d = NULL; !d && !pci_is_root_bus(b); b = b->parent) { 763 if (b->self) 764 d = dev_get_msi_domain(&b->self->dev); 765 } 766 767 if (!d) 768 d = pci_host_bridge_msi_domain(b); 769 770 dev_set_msi_domain(&bus->dev, d); 771 } 772 773 static int pci_register_host_bridge(struct pci_host_bridge *bridge) 774 { 775 struct device *parent = bridge->dev.parent; 776 struct resource_entry *window, *n; 777 struct pci_bus *bus, *b; 778 resource_size_t offset; 779 LIST_HEAD(resources); 780 struct resource *res; 781 char addr[64], *fmt; 782 const char *name; 783 int err; 784 785 bus = pci_alloc_bus(NULL); 786 if (!bus) 787 return -ENOMEM; 788 789 bridge->bus = bus; 790 791 /* Temporarily move resources off the list */ 792 list_splice_init(&bridge->windows, &resources); 793 bus->sysdata = bridge->sysdata; 794 bus->msi = bridge->msi; 795 bus->ops = bridge->ops; 796 bus->number = bus->busn_res.start = bridge->busnr; 797 #ifdef CONFIG_PCI_DOMAINS_GENERIC 798 bus->domain_nr = pci_bus_find_domain_nr(bus, parent); 799 #endif 800 801 b = pci_find_bus(pci_domain_nr(bus), bridge->busnr); 802 if (b) { 803 /* Ignore it if we already got here via a different bridge */ 804 dev_dbg(&b->dev, "bus already known\n"); 805 err = -EEXIST; 806 goto free; 807 } 808 809 dev_set_name(&bridge->dev, "pci%04x:%02x", pci_domain_nr(bus), 810 bridge->busnr); 811 812 err = pcibios_root_bridge_prepare(bridge); 813 if (err) 814 goto free; 815 816 err = device_register(&bridge->dev); 817 if (err) 818 put_device(&bridge->dev); 819 820 bus->bridge = get_device(&bridge->dev); 821 device_enable_async_suspend(bus->bridge); 822 pci_set_bus_of_node(bus); 823 pci_set_bus_msi_domain(bus); 824 825 if (!parent) 826 set_dev_node(bus->bridge, pcibus_to_node(bus)); 827 828 bus->dev.class = &pcibus_class; 829 bus->dev.parent = bus->bridge; 830 831 dev_set_name(&bus->dev, "%04x:%02x", pci_domain_nr(bus), bus->number); 832 name = dev_name(&bus->dev); 833 834 err = device_register(&bus->dev); 835 if (err) 836 goto unregister; 837 838 pcibios_add_bus(bus); 839 840 /* Create legacy_io and legacy_mem files for this bus */ 841 pci_create_legacy_files(bus); 842 843 if (parent) 844 dev_info(parent, "PCI host bridge to bus %s\n", name); 845 else 846 pr_info("PCI host bridge to bus %s\n", name); 847 848 /* Add initial resources to the bus */ 849 resource_list_for_each_entry_safe(window, n, &resources) { 850 list_move_tail(&window->node, &bridge->windows); 851 offset = window->offset; 852 res = window->res; 853 854 if (res->flags & IORESOURCE_BUS) 855 pci_bus_insert_busn_res(bus, bus->number, res->end); 856 else 857 pci_bus_add_resource(bus, res, 0); 858 859 if (offset) { 860 if (resource_type(res) == IORESOURCE_IO) 861 fmt = " (bus address [%#06llx-%#06llx])"; 862 else 863 fmt = " (bus address [%#010llx-%#010llx])"; 864 865 snprintf(addr, sizeof(addr), fmt, 866 (unsigned long long)(res->start - offset), 867 (unsigned long long)(res->end - offset)); 868 } else 869 addr[0] = '\0'; 870 871 dev_info(&bus->dev, "root bus resource %pR%s\n", res, addr); 872 } 873 874 down_write(&pci_bus_sem); 875 list_add_tail(&bus->node, &pci_root_buses); 876 up_write(&pci_bus_sem); 877 878 return 0; 879 880 unregister: 881 put_device(&bridge->dev); 882 device_unregister(&bridge->dev); 883 884 free: 885 kfree(bus); 886 return err; 887 } 888 889 static bool pci_bridge_child_ext_cfg_accessible(struct pci_dev *bridge) 890 { 891 int pos; 892 u32 status; 893 894 /* 895 * If extended config space isn't accessible on a bridge's primary 896 * bus, we certainly can't access it on the secondary bus. 897 */ 898 if (bridge->bus->bus_flags & PCI_BUS_FLAGS_NO_EXTCFG) 899 return false; 900 901 /* 902 * PCIe Root Ports and switch ports are PCIe on both sides, so if 903 * extended config space is accessible on the primary, it's also 904 * accessible on the secondary. 905 */ 906 if (pci_is_pcie(bridge) && 907 (pci_pcie_type(bridge) == PCI_EXP_TYPE_ROOT_PORT || 908 pci_pcie_type(bridge) == PCI_EXP_TYPE_UPSTREAM || 909 pci_pcie_type(bridge) == PCI_EXP_TYPE_DOWNSTREAM)) 910 return true; 911 912 /* 913 * For the other bridge types: 914 * - PCI-to-PCI bridges 915 * - PCIe-to-PCI/PCI-X forward bridges 916 * - PCI/PCI-X-to-PCIe reverse bridges 917 * extended config space on the secondary side is only accessible 918 * if the bridge supports PCI-X Mode 2. 919 */ 920 pos = pci_find_capability(bridge, PCI_CAP_ID_PCIX); 921 if (!pos) 922 return false; 923 924 pci_read_config_dword(bridge, pos + PCI_X_STATUS, &status); 925 return status & (PCI_X_STATUS_266MHZ | PCI_X_STATUS_533MHZ); 926 } 927 928 static struct pci_bus *pci_alloc_child_bus(struct pci_bus *parent, 929 struct pci_dev *bridge, int busnr) 930 { 931 struct pci_bus *child; 932 int i; 933 int ret; 934 935 /* Allocate a new bus and inherit stuff from the parent */ 936 child = pci_alloc_bus(parent); 937 if (!child) 938 return NULL; 939 940 child->parent = parent; 941 child->ops = parent->ops; 942 child->msi = parent->msi; 943 child->sysdata = parent->sysdata; 944 child->bus_flags = parent->bus_flags; 945 946 /* 947 * Initialize some portions of the bus device, but don't register 948 * it now as the parent is not properly set up yet. 949 */ 950 child->dev.class = &pcibus_class; 951 dev_set_name(&child->dev, "%04x:%02x", pci_domain_nr(child), busnr); 952 953 /* Set up the primary, secondary and subordinate bus numbers */ 954 child->number = child->busn_res.start = busnr; 955 child->primary = parent->busn_res.start; 956 child->busn_res.end = 0xff; 957 958 if (!bridge) { 959 child->dev.parent = parent->bridge; 960 goto add_dev; 961 } 962 963 child->self = bridge; 964 child->bridge = get_device(&bridge->dev); 965 child->dev.parent = child->bridge; 966 pci_set_bus_of_node(child); 967 pci_set_bus_speed(child); 968 969 /* 970 * Check whether extended config space is accessible on the child 971 * bus. Note that we currently assume it is always accessible on 972 * the root bus. 973 */ 974 if (!pci_bridge_child_ext_cfg_accessible(bridge)) { 975 child->bus_flags |= PCI_BUS_FLAGS_NO_EXTCFG; 976 pci_info(child, "extended config space not accessible\n"); 977 } 978 979 /* Set up default resource pointers and names */ 980 for (i = 0; i < PCI_BRIDGE_RESOURCE_NUM; i++) { 981 child->resource[i] = &bridge->resource[PCI_BRIDGE_RESOURCES+i]; 982 child->resource[i]->name = child->name; 983 } 984 bridge->subordinate = child; 985 986 add_dev: 987 pci_set_bus_msi_domain(child); 988 ret = device_register(&child->dev); 989 WARN_ON(ret < 0); 990 991 pcibios_add_bus(child); 992 993 if (child->ops->add_bus) { 994 ret = child->ops->add_bus(child); 995 if (WARN_ON(ret < 0)) 996 dev_err(&child->dev, "failed to add bus: %d\n", ret); 997 } 998 999 /* Create legacy_io and legacy_mem files for this bus */ 1000 pci_create_legacy_files(child); 1001 1002 return child; 1003 } 1004 1005 struct pci_bus *pci_add_new_bus(struct pci_bus *parent, struct pci_dev *dev, 1006 int busnr) 1007 { 1008 struct pci_bus *child; 1009 1010 child = pci_alloc_child_bus(parent, dev, busnr); 1011 if (child) { 1012 down_write(&pci_bus_sem); 1013 list_add_tail(&child->node, &parent->children); 1014 up_write(&pci_bus_sem); 1015 } 1016 return child; 1017 } 1018 EXPORT_SYMBOL(pci_add_new_bus); 1019 1020 static void pci_enable_crs(struct pci_dev *pdev) 1021 { 1022 u16 root_cap = 0; 1023 1024 /* Enable CRS Software Visibility if supported */ 1025 pcie_capability_read_word(pdev, PCI_EXP_RTCAP, &root_cap); 1026 if (root_cap & PCI_EXP_RTCAP_CRSVIS) 1027 pcie_capability_set_word(pdev, PCI_EXP_RTCTL, 1028 PCI_EXP_RTCTL_CRSSVE); 1029 } 1030 1031 static unsigned int pci_scan_child_bus_extend(struct pci_bus *bus, 1032 unsigned int available_buses); 1033 1034 /* 1035 * pci_scan_bridge_extend() - Scan buses behind a bridge 1036 * @bus: Parent bus the bridge is on 1037 * @dev: Bridge itself 1038 * @max: Starting subordinate number of buses behind this bridge 1039 * @available_buses: Total number of buses available for this bridge and 1040 * the devices below. After the minimal bus space has 1041 * been allocated the remaining buses will be 1042 * distributed equally between hotplug-capable bridges. 1043 * @pass: Either %0 (scan already configured bridges) or %1 (scan bridges 1044 * that need to be reconfigured. 1045 * 1046 * If it's a bridge, configure it and scan the bus behind it. 1047 * For CardBus bridges, we don't scan behind as the devices will 1048 * be handled by the bridge driver itself. 1049 * 1050 * We need to process bridges in two passes -- first we scan those 1051 * already configured by the BIOS and after we are done with all of 1052 * them, we proceed to assigning numbers to the remaining buses in 1053 * order to avoid overlaps between old and new bus numbers. 1054 * 1055 * Return: New subordinate number covering all buses behind this bridge. 1056 */ 1057 static int pci_scan_bridge_extend(struct pci_bus *bus, struct pci_dev *dev, 1058 int max, unsigned int available_buses, 1059 int pass) 1060 { 1061 struct pci_bus *child; 1062 int is_cardbus = (dev->hdr_type == PCI_HEADER_TYPE_CARDBUS); 1063 u32 buses, i, j = 0; 1064 u16 bctl; 1065 u8 primary, secondary, subordinate; 1066 int broken = 0; 1067 1068 /* 1069 * Make sure the bridge is powered on to be able to access config 1070 * space of devices below it. 1071 */ 1072 pm_runtime_get_sync(&dev->dev); 1073 1074 pci_read_config_dword(dev, PCI_PRIMARY_BUS, &buses); 1075 primary = buses & 0xFF; 1076 secondary = (buses >> 8) & 0xFF; 1077 subordinate = (buses >> 16) & 0xFF; 1078 1079 pci_dbg(dev, "scanning [bus %02x-%02x] behind bridge, pass %d\n", 1080 secondary, subordinate, pass); 1081 1082 if (!primary && (primary != bus->number) && secondary && subordinate) { 1083 pci_warn(dev, "Primary bus is hard wired to 0\n"); 1084 primary = bus->number; 1085 } 1086 1087 /* Check if setup is sensible at all */ 1088 if (!pass && 1089 (primary != bus->number || secondary <= bus->number || 1090 secondary > subordinate)) { 1091 pci_info(dev, "bridge configuration invalid ([bus %02x-%02x]), reconfiguring\n", 1092 secondary, subordinate); 1093 broken = 1; 1094 } 1095 1096 /* 1097 * Disable Master-Abort Mode during probing to avoid reporting of 1098 * bus errors in some architectures. 1099 */ 1100 pci_read_config_word(dev, PCI_BRIDGE_CONTROL, &bctl); 1101 pci_write_config_word(dev, PCI_BRIDGE_CONTROL, 1102 bctl & ~PCI_BRIDGE_CTL_MASTER_ABORT); 1103 1104 pci_enable_crs(dev); 1105 1106 if ((secondary || subordinate) && !pcibios_assign_all_busses() && 1107 !is_cardbus && !broken) { 1108 unsigned int cmax; 1109 1110 /* 1111 * Bus already configured by firmware, process it in the 1112 * first pass and just note the configuration. 1113 */ 1114 if (pass) 1115 goto out; 1116 1117 /* 1118 * The bus might already exist for two reasons: Either we 1119 * are rescanning the bus or the bus is reachable through 1120 * more than one bridge. The second case can happen with 1121 * the i450NX chipset. 1122 */ 1123 child = pci_find_bus(pci_domain_nr(bus), secondary); 1124 if (!child) { 1125 child = pci_add_new_bus(bus, dev, secondary); 1126 if (!child) 1127 goto out; 1128 child->primary = primary; 1129 pci_bus_insert_busn_res(child, secondary, subordinate); 1130 child->bridge_ctl = bctl; 1131 } 1132 1133 cmax = pci_scan_child_bus(child); 1134 if (cmax > subordinate) 1135 pci_warn(dev, "bridge has subordinate %02x but max busn %02x\n", 1136 subordinate, cmax); 1137 1138 /* Subordinate should equal child->busn_res.end */ 1139 if (subordinate > max) 1140 max = subordinate; 1141 } else { 1142 1143 /* 1144 * We need to assign a number to this bus which we always 1145 * do in the second pass. 1146 */ 1147 if (!pass) { 1148 if (pcibios_assign_all_busses() || broken || is_cardbus) 1149 1150 /* 1151 * Temporarily disable forwarding of the 1152 * configuration cycles on all bridges in 1153 * this bus segment to avoid possible 1154 * conflicts in the second pass between two 1155 * bridges programmed with overlapping bus 1156 * ranges. 1157 */ 1158 pci_write_config_dword(dev, PCI_PRIMARY_BUS, 1159 buses & ~0xffffff); 1160 goto out; 1161 } 1162 1163 /* Clear errors */ 1164 pci_write_config_word(dev, PCI_STATUS, 0xffff); 1165 1166 /* 1167 * Prevent assigning a bus number that already exists. 1168 * This can happen when a bridge is hot-plugged, so in this 1169 * case we only re-scan this bus. 1170 */ 1171 child = pci_find_bus(pci_domain_nr(bus), max+1); 1172 if (!child) { 1173 child = pci_add_new_bus(bus, dev, max+1); 1174 if (!child) 1175 goto out; 1176 pci_bus_insert_busn_res(child, max+1, 1177 bus->busn_res.end); 1178 } 1179 max++; 1180 if (available_buses) 1181 available_buses--; 1182 1183 buses = (buses & 0xff000000) 1184 | ((unsigned int)(child->primary) << 0) 1185 | ((unsigned int)(child->busn_res.start) << 8) 1186 | ((unsigned int)(child->busn_res.end) << 16); 1187 1188 /* 1189 * yenta.c forces a secondary latency timer of 176. 1190 * Copy that behaviour here. 1191 */ 1192 if (is_cardbus) { 1193 buses &= ~0xff000000; 1194 buses |= CARDBUS_LATENCY_TIMER << 24; 1195 } 1196 1197 /* We need to blast all three values with a single write */ 1198 pci_write_config_dword(dev, PCI_PRIMARY_BUS, buses); 1199 1200 if (!is_cardbus) { 1201 child->bridge_ctl = bctl; 1202 max = pci_scan_child_bus_extend(child, available_buses); 1203 } else { 1204 1205 /* 1206 * For CardBus bridges, we leave 4 bus numbers as 1207 * cards with a PCI-to-PCI bridge can be inserted 1208 * later. 1209 */ 1210 for (i = 0; i < CARDBUS_RESERVE_BUSNR; i++) { 1211 struct pci_bus *parent = bus; 1212 if (pci_find_bus(pci_domain_nr(bus), 1213 max+i+1)) 1214 break; 1215 while (parent->parent) { 1216 if ((!pcibios_assign_all_busses()) && 1217 (parent->busn_res.end > max) && 1218 (parent->busn_res.end <= max+i)) { 1219 j = 1; 1220 } 1221 parent = parent->parent; 1222 } 1223 if (j) { 1224 1225 /* 1226 * Often, there are two CardBus 1227 * bridges -- try to leave one 1228 * valid bus number for each one. 1229 */ 1230 i /= 2; 1231 break; 1232 } 1233 } 1234 max += i; 1235 } 1236 1237 /* Set subordinate bus number to its real value */ 1238 pci_bus_update_busn_res_end(child, max); 1239 pci_write_config_byte(dev, PCI_SUBORDINATE_BUS, max); 1240 } 1241 1242 sprintf(child->name, 1243 (is_cardbus ? "PCI CardBus %04x:%02x" : "PCI Bus %04x:%02x"), 1244 pci_domain_nr(bus), child->number); 1245 1246 /* Check that all devices are accessible */ 1247 while (bus->parent) { 1248 if ((child->busn_res.end > bus->busn_res.end) || 1249 (child->number > bus->busn_res.end) || 1250 (child->number < bus->number) || 1251 (child->busn_res.end < bus->number)) { 1252 dev_info(&dev->dev, "devices behind bridge are unusable because %pR cannot be assigned for them\n", 1253 &child->busn_res); 1254 break; 1255 } 1256 bus = bus->parent; 1257 } 1258 1259 out: 1260 pci_write_config_word(dev, PCI_BRIDGE_CONTROL, bctl); 1261 1262 pm_runtime_put(&dev->dev); 1263 1264 return max; 1265 } 1266 1267 /* 1268 * pci_scan_bridge() - Scan buses behind a bridge 1269 * @bus: Parent bus the bridge is on 1270 * @dev: Bridge itself 1271 * @max: Starting subordinate number of buses behind this bridge 1272 * @pass: Either %0 (scan already configured bridges) or %1 (scan bridges 1273 * that need to be reconfigured. 1274 * 1275 * If it's a bridge, configure it and scan the bus behind it. 1276 * For CardBus bridges, we don't scan behind as the devices will 1277 * be handled by the bridge driver itself. 1278 * 1279 * We need to process bridges in two passes -- first we scan those 1280 * already configured by the BIOS and after we are done with all of 1281 * them, we proceed to assigning numbers to the remaining buses in 1282 * order to avoid overlaps between old and new bus numbers. 1283 * 1284 * Return: New subordinate number covering all buses behind this bridge. 1285 */ 1286 int pci_scan_bridge(struct pci_bus *bus, struct pci_dev *dev, int max, int pass) 1287 { 1288 return pci_scan_bridge_extend(bus, dev, max, 0, pass); 1289 } 1290 EXPORT_SYMBOL(pci_scan_bridge); 1291 1292 /* 1293 * Read interrupt line and base address registers. 1294 * The architecture-dependent code can tweak these, of course. 1295 */ 1296 static void pci_read_irq(struct pci_dev *dev) 1297 { 1298 unsigned char irq; 1299 1300 /* VFs are not allowed to use INTx, so skip the config reads */ 1301 if (dev->is_virtfn) { 1302 dev->pin = 0; 1303 dev->irq = 0; 1304 return; 1305 } 1306 1307 pci_read_config_byte(dev, PCI_INTERRUPT_PIN, &irq); 1308 dev->pin = irq; 1309 if (irq) 1310 pci_read_config_byte(dev, PCI_INTERRUPT_LINE, &irq); 1311 dev->irq = irq; 1312 } 1313 1314 void set_pcie_port_type(struct pci_dev *pdev) 1315 { 1316 int pos; 1317 u16 reg16; 1318 int type; 1319 struct pci_dev *parent; 1320 1321 pos = pci_find_capability(pdev, PCI_CAP_ID_EXP); 1322 if (!pos) 1323 return; 1324 1325 pdev->pcie_cap = pos; 1326 pci_read_config_word(pdev, pos + PCI_EXP_FLAGS, ®16); 1327 pdev->pcie_flags_reg = reg16; 1328 pci_read_config_word(pdev, pos + PCI_EXP_DEVCAP, ®16); 1329 pdev->pcie_mpss = reg16 & PCI_EXP_DEVCAP_PAYLOAD; 1330 1331 /* 1332 * A Root Port or a PCI-to-PCIe bridge is always the upstream end 1333 * of a Link. No PCIe component has two Links. Two Links are 1334 * connected by a Switch that has a Port on each Link and internal 1335 * logic to connect the two Ports. 1336 */ 1337 type = pci_pcie_type(pdev); 1338 if (type == PCI_EXP_TYPE_ROOT_PORT || 1339 type == PCI_EXP_TYPE_PCIE_BRIDGE) 1340 pdev->has_secondary_link = 1; 1341 else if (type == PCI_EXP_TYPE_UPSTREAM || 1342 type == PCI_EXP_TYPE_DOWNSTREAM) { 1343 parent = pci_upstream_bridge(pdev); 1344 1345 /* 1346 * Usually there's an upstream device (Root Port or Switch 1347 * Downstream Port), but we can't assume one exists. 1348 */ 1349 if (parent && !parent->has_secondary_link) 1350 pdev->has_secondary_link = 1; 1351 } 1352 } 1353 1354 void set_pcie_hotplug_bridge(struct pci_dev *pdev) 1355 { 1356 u32 reg32; 1357 1358 pcie_capability_read_dword(pdev, PCI_EXP_SLTCAP, ®32); 1359 if (reg32 & PCI_EXP_SLTCAP_HPC) 1360 pdev->is_hotplug_bridge = 1; 1361 } 1362 1363 static void set_pcie_thunderbolt(struct pci_dev *dev) 1364 { 1365 int vsec = 0; 1366 u32 header; 1367 1368 while ((vsec = pci_find_next_ext_capability(dev, vsec, 1369 PCI_EXT_CAP_ID_VNDR))) { 1370 pci_read_config_dword(dev, vsec + PCI_VNDR_HEADER, &header); 1371 1372 /* Is the device part of a Thunderbolt controller? */ 1373 if (dev->vendor == PCI_VENDOR_ID_INTEL && 1374 PCI_VNDR_HEADER_ID(header) == PCI_VSEC_ID_INTEL_TBT) { 1375 dev->is_thunderbolt = 1; 1376 return; 1377 } 1378 } 1379 } 1380 1381 /** 1382 * pci_ext_cfg_is_aliased - Is ext config space just an alias of std config? 1383 * @dev: PCI device 1384 * 1385 * PCI Express to PCI/PCI-X Bridge Specification, rev 1.0, 4.1.4 says that 1386 * when forwarding a type1 configuration request the bridge must check that 1387 * the extended register address field is zero. The bridge is not permitted 1388 * to forward the transactions and must handle it as an Unsupported Request. 1389 * Some bridges do not follow this rule and simply drop the extended register 1390 * bits, resulting in the standard config space being aliased, every 256 1391 * bytes across the entire configuration space. Test for this condition by 1392 * comparing the first dword of each potential alias to the vendor/device ID. 1393 * Known offenders: 1394 * ASM1083/1085 PCIe-to-PCI Reversible Bridge (1b21:1080, rev 01 & 03) 1395 * AMD/ATI SBx00 PCI to PCI Bridge (1002:4384, rev 40) 1396 */ 1397 static bool pci_ext_cfg_is_aliased(struct pci_dev *dev) 1398 { 1399 #ifdef CONFIG_PCI_QUIRKS 1400 int pos; 1401 u32 header, tmp; 1402 1403 pci_read_config_dword(dev, PCI_VENDOR_ID, &header); 1404 1405 for (pos = PCI_CFG_SPACE_SIZE; 1406 pos < PCI_CFG_SPACE_EXP_SIZE; pos += PCI_CFG_SPACE_SIZE) { 1407 if (pci_read_config_dword(dev, pos, &tmp) != PCIBIOS_SUCCESSFUL 1408 || header != tmp) 1409 return false; 1410 } 1411 1412 return true; 1413 #else 1414 return false; 1415 #endif 1416 } 1417 1418 /** 1419 * pci_cfg_space_size - Get the configuration space size of the PCI device 1420 * @dev: PCI device 1421 * 1422 * Regular PCI devices have 256 bytes, but PCI-X 2 and PCI Express devices 1423 * have 4096 bytes. Even if the device is capable, that doesn't mean we can 1424 * access it. Maybe we don't have a way to generate extended config space 1425 * accesses, or the device is behind a reverse Express bridge. So we try 1426 * reading the dword at 0x100 which must either be 0 or a valid extended 1427 * capability header. 1428 */ 1429 static int pci_cfg_space_size_ext(struct pci_dev *dev) 1430 { 1431 u32 status; 1432 int pos = PCI_CFG_SPACE_SIZE; 1433 1434 if (pci_read_config_dword(dev, pos, &status) != PCIBIOS_SUCCESSFUL) 1435 return PCI_CFG_SPACE_SIZE; 1436 if (status == 0xffffffff || pci_ext_cfg_is_aliased(dev)) 1437 return PCI_CFG_SPACE_SIZE; 1438 1439 return PCI_CFG_SPACE_EXP_SIZE; 1440 } 1441 1442 #ifdef CONFIG_PCI_IOV 1443 static bool is_vf0(struct pci_dev *dev) 1444 { 1445 if (pci_iov_virtfn_devfn(dev->physfn, 0) == dev->devfn && 1446 pci_iov_virtfn_bus(dev->physfn, 0) == dev->bus->number) 1447 return true; 1448 1449 return false; 1450 } 1451 #endif 1452 1453 int pci_cfg_space_size(struct pci_dev *dev) 1454 { 1455 int pos; 1456 u32 status; 1457 u16 class; 1458 1459 #ifdef CONFIG_PCI_IOV 1460 /* Read cached value for all VFs except for VF0 */ 1461 if (dev->is_virtfn && !is_vf0(dev)) 1462 return dev->physfn->sriov->cfg_size; 1463 #endif 1464 1465 if (dev->bus->bus_flags & PCI_BUS_FLAGS_NO_EXTCFG) 1466 return PCI_CFG_SPACE_SIZE; 1467 1468 class = dev->class >> 8; 1469 if (class == PCI_CLASS_BRIDGE_HOST) 1470 return pci_cfg_space_size_ext(dev); 1471 1472 if (pci_is_pcie(dev)) 1473 return pci_cfg_space_size_ext(dev); 1474 1475 pos = pci_find_capability(dev, PCI_CAP_ID_PCIX); 1476 if (!pos) 1477 return PCI_CFG_SPACE_SIZE; 1478 1479 pci_read_config_dword(dev, pos + PCI_X_STATUS, &status); 1480 if (status & (PCI_X_STATUS_266MHZ | PCI_X_STATUS_533MHZ)) 1481 return pci_cfg_space_size_ext(dev); 1482 1483 return PCI_CFG_SPACE_SIZE; 1484 } 1485 1486 static u32 pci_class(struct pci_dev *dev) 1487 { 1488 u32 class; 1489 1490 #ifdef CONFIG_PCI_IOV 1491 if (dev->is_virtfn) 1492 return dev->physfn->sriov->class; 1493 #endif 1494 pci_read_config_dword(dev, PCI_CLASS_REVISION, &class); 1495 return class; 1496 } 1497 1498 static void pci_subsystem_ids(struct pci_dev *dev, u16 *vendor, u16 *device) 1499 { 1500 #ifdef CONFIG_PCI_IOV 1501 if (dev->is_virtfn) { 1502 *vendor = dev->physfn->sriov->subsystem_vendor; 1503 *device = dev->physfn->sriov->subsystem_device; 1504 return; 1505 } 1506 #endif 1507 pci_read_config_word(dev, PCI_SUBSYSTEM_VENDOR_ID, vendor); 1508 pci_read_config_word(dev, PCI_SUBSYSTEM_ID, device); 1509 } 1510 1511 static u8 pci_hdr_type(struct pci_dev *dev) 1512 { 1513 u8 hdr_type; 1514 1515 #ifdef CONFIG_PCI_IOV 1516 if (dev->is_virtfn) 1517 return dev->physfn->sriov->hdr_type; 1518 #endif 1519 pci_read_config_byte(dev, PCI_HEADER_TYPE, &hdr_type); 1520 return hdr_type; 1521 } 1522 1523 #define LEGACY_IO_RESOURCE (IORESOURCE_IO | IORESOURCE_PCI_FIXED) 1524 1525 static void pci_msi_setup_pci_dev(struct pci_dev *dev) 1526 { 1527 /* 1528 * Disable the MSI hardware to avoid screaming interrupts 1529 * during boot. This is the power on reset default so 1530 * usually this should be a noop. 1531 */ 1532 dev->msi_cap = pci_find_capability(dev, PCI_CAP_ID_MSI); 1533 if (dev->msi_cap) 1534 pci_msi_set_enable(dev, 0); 1535 1536 dev->msix_cap = pci_find_capability(dev, PCI_CAP_ID_MSIX); 1537 if (dev->msix_cap) 1538 pci_msix_clear_and_set_ctrl(dev, PCI_MSIX_FLAGS_ENABLE, 0); 1539 } 1540 1541 /** 1542 * pci_intx_mask_broken - Test PCI_COMMAND_INTX_DISABLE writability 1543 * @dev: PCI device 1544 * 1545 * Test whether PCI_COMMAND_INTX_DISABLE is writable for @dev. Check this 1546 * at enumeration-time to avoid modifying PCI_COMMAND at run-time. 1547 */ 1548 static int pci_intx_mask_broken(struct pci_dev *dev) 1549 { 1550 u16 orig, toggle, new; 1551 1552 pci_read_config_word(dev, PCI_COMMAND, &orig); 1553 toggle = orig ^ PCI_COMMAND_INTX_DISABLE; 1554 pci_write_config_word(dev, PCI_COMMAND, toggle); 1555 pci_read_config_word(dev, PCI_COMMAND, &new); 1556 1557 pci_write_config_word(dev, PCI_COMMAND, orig); 1558 1559 /* 1560 * PCI_COMMAND_INTX_DISABLE was reserved and read-only prior to PCI 1561 * r2.3, so strictly speaking, a device is not *broken* if it's not 1562 * writable. But we'll live with the misnomer for now. 1563 */ 1564 if (new != toggle) 1565 return 1; 1566 return 0; 1567 } 1568 1569 static void early_dump_pci_device(struct pci_dev *pdev) 1570 { 1571 u32 value[256 / 4]; 1572 int i; 1573 1574 pci_info(pdev, "config space:\n"); 1575 1576 for (i = 0; i < 256; i += 4) 1577 pci_read_config_dword(pdev, i, &value[i / 4]); 1578 1579 print_hex_dump(KERN_INFO, "", DUMP_PREFIX_OFFSET, 16, 1, 1580 value, 256, false); 1581 } 1582 1583 /** 1584 * pci_setup_device - Fill in class and map information of a device 1585 * @dev: the device structure to fill 1586 * 1587 * Initialize the device structure with information about the device's 1588 * vendor,class,memory and IO-space addresses, IRQ lines etc. 1589 * Called at initialisation of the PCI subsystem and by CardBus services. 1590 * Returns 0 on success and negative if unknown type of device (not normal, 1591 * bridge or CardBus). 1592 */ 1593 int pci_setup_device(struct pci_dev *dev) 1594 { 1595 u32 class; 1596 u16 cmd; 1597 u8 hdr_type; 1598 int pos = 0; 1599 struct pci_bus_region region; 1600 struct resource *res; 1601 1602 hdr_type = pci_hdr_type(dev); 1603 1604 dev->sysdata = dev->bus->sysdata; 1605 dev->dev.parent = dev->bus->bridge; 1606 dev->dev.bus = &pci_bus_type; 1607 dev->hdr_type = hdr_type & 0x7f; 1608 dev->multifunction = !!(hdr_type & 0x80); 1609 dev->error_state = pci_channel_io_normal; 1610 set_pcie_port_type(dev); 1611 1612 pci_dev_assign_slot(dev); 1613 1614 /* 1615 * Assume 32-bit PCI; let 64-bit PCI cards (which are far rarer) 1616 * set this higher, assuming the system even supports it. 1617 */ 1618 dev->dma_mask = 0xffffffff; 1619 1620 dev_set_name(&dev->dev, "%04x:%02x:%02x.%d", pci_domain_nr(dev->bus), 1621 dev->bus->number, PCI_SLOT(dev->devfn), 1622 PCI_FUNC(dev->devfn)); 1623 1624 class = pci_class(dev); 1625 1626 dev->revision = class & 0xff; 1627 dev->class = class >> 8; /* upper 3 bytes */ 1628 1629 pci_printk(KERN_DEBUG, dev, "[%04x:%04x] type %02x class %#08x\n", 1630 dev->vendor, dev->device, dev->hdr_type, dev->class); 1631 1632 if (pci_early_dump) 1633 early_dump_pci_device(dev); 1634 1635 /* Need to have dev->class ready */ 1636 dev->cfg_size = pci_cfg_space_size(dev); 1637 1638 /* Need to have dev->cfg_size ready */ 1639 set_pcie_thunderbolt(dev); 1640 1641 /* "Unknown power state" */ 1642 dev->current_state = PCI_UNKNOWN; 1643 1644 /* Early fixups, before probing the BARs */ 1645 pci_fixup_device(pci_fixup_early, dev); 1646 1647 /* Device class may be changed after fixup */ 1648 class = dev->class >> 8; 1649 1650 if (dev->non_compliant_bars) { 1651 pci_read_config_word(dev, PCI_COMMAND, &cmd); 1652 if (cmd & (PCI_COMMAND_IO | PCI_COMMAND_MEMORY)) { 1653 pci_info(dev, "device has non-compliant BARs; disabling IO/MEM decoding\n"); 1654 cmd &= ~PCI_COMMAND_IO; 1655 cmd &= ~PCI_COMMAND_MEMORY; 1656 pci_write_config_word(dev, PCI_COMMAND, cmd); 1657 } 1658 } 1659 1660 dev->broken_intx_masking = pci_intx_mask_broken(dev); 1661 1662 switch (dev->hdr_type) { /* header type */ 1663 case PCI_HEADER_TYPE_NORMAL: /* standard header */ 1664 if (class == PCI_CLASS_BRIDGE_PCI) 1665 goto bad; 1666 pci_read_irq(dev); 1667 pci_read_bases(dev, 6, PCI_ROM_ADDRESS); 1668 1669 pci_subsystem_ids(dev, &dev->subsystem_vendor, &dev->subsystem_device); 1670 1671 /* 1672 * Do the ugly legacy mode stuff here rather than broken chip 1673 * quirk code. Legacy mode ATA controllers have fixed 1674 * addresses. These are not always echoed in BAR0-3, and 1675 * BAR0-3 in a few cases contain junk! 1676 */ 1677 if (class == PCI_CLASS_STORAGE_IDE) { 1678 u8 progif; 1679 pci_read_config_byte(dev, PCI_CLASS_PROG, &progif); 1680 if ((progif & 1) == 0) { 1681 region.start = 0x1F0; 1682 region.end = 0x1F7; 1683 res = &dev->resource[0]; 1684 res->flags = LEGACY_IO_RESOURCE; 1685 pcibios_bus_to_resource(dev->bus, res, ®ion); 1686 pci_info(dev, "legacy IDE quirk: reg 0x10: %pR\n", 1687 res); 1688 region.start = 0x3F6; 1689 region.end = 0x3F6; 1690 res = &dev->resource[1]; 1691 res->flags = LEGACY_IO_RESOURCE; 1692 pcibios_bus_to_resource(dev->bus, res, ®ion); 1693 pci_info(dev, "legacy IDE quirk: reg 0x14: %pR\n", 1694 res); 1695 } 1696 if ((progif & 4) == 0) { 1697 region.start = 0x170; 1698 region.end = 0x177; 1699 res = &dev->resource[2]; 1700 res->flags = LEGACY_IO_RESOURCE; 1701 pcibios_bus_to_resource(dev->bus, res, ®ion); 1702 pci_info(dev, "legacy IDE quirk: reg 0x18: %pR\n", 1703 res); 1704 region.start = 0x376; 1705 region.end = 0x376; 1706 res = &dev->resource[3]; 1707 res->flags = LEGACY_IO_RESOURCE; 1708 pcibios_bus_to_resource(dev->bus, res, ®ion); 1709 pci_info(dev, "legacy IDE quirk: reg 0x1c: %pR\n", 1710 res); 1711 } 1712 } 1713 break; 1714 1715 case PCI_HEADER_TYPE_BRIDGE: /* bridge header */ 1716 if (class != PCI_CLASS_BRIDGE_PCI) 1717 goto bad; 1718 1719 /* 1720 * The PCI-to-PCI bridge spec requires that subtractive 1721 * decoding (i.e. transparent) bridge must have programming 1722 * interface code of 0x01. 1723 */ 1724 pci_read_irq(dev); 1725 dev->transparent = ((dev->class & 0xff) == 1); 1726 pci_read_bases(dev, 2, PCI_ROM_ADDRESS1); 1727 set_pcie_hotplug_bridge(dev); 1728 pos = pci_find_capability(dev, PCI_CAP_ID_SSVID); 1729 if (pos) { 1730 pci_read_config_word(dev, pos + PCI_SSVID_VENDOR_ID, &dev->subsystem_vendor); 1731 pci_read_config_word(dev, pos + PCI_SSVID_DEVICE_ID, &dev->subsystem_device); 1732 } 1733 break; 1734 1735 case PCI_HEADER_TYPE_CARDBUS: /* CardBus bridge header */ 1736 if (class != PCI_CLASS_BRIDGE_CARDBUS) 1737 goto bad; 1738 pci_read_irq(dev); 1739 pci_read_bases(dev, 1, 0); 1740 pci_read_config_word(dev, PCI_CB_SUBSYSTEM_VENDOR_ID, &dev->subsystem_vendor); 1741 pci_read_config_word(dev, PCI_CB_SUBSYSTEM_ID, &dev->subsystem_device); 1742 break; 1743 1744 default: /* unknown header */ 1745 pci_err(dev, "unknown header type %02x, ignoring device\n", 1746 dev->hdr_type); 1747 return -EIO; 1748 1749 bad: 1750 pci_err(dev, "ignoring class %#08x (doesn't match header type %02x)\n", 1751 dev->class, dev->hdr_type); 1752 dev->class = PCI_CLASS_NOT_DEFINED << 8; 1753 } 1754 1755 /* We found a fine healthy device, go go go... */ 1756 return 0; 1757 } 1758 1759 static void pci_configure_mps(struct pci_dev *dev) 1760 { 1761 struct pci_dev *bridge = pci_upstream_bridge(dev); 1762 int mps, mpss, p_mps, rc; 1763 1764 if (!pci_is_pcie(dev) || !bridge || !pci_is_pcie(bridge)) 1765 return; 1766 1767 /* MPS and MRRS fields are of type 'RsvdP' for VFs, short-circuit out */ 1768 if (dev->is_virtfn) 1769 return; 1770 1771 mps = pcie_get_mps(dev); 1772 p_mps = pcie_get_mps(bridge); 1773 1774 if (mps == p_mps) 1775 return; 1776 1777 if (pcie_bus_config == PCIE_BUS_TUNE_OFF) { 1778 pci_warn(dev, "Max Payload Size %d, but upstream %s set to %d; if necessary, use \"pci=pcie_bus_safe\" and report a bug\n", 1779 mps, pci_name(bridge), p_mps); 1780 return; 1781 } 1782 1783 /* 1784 * Fancier MPS configuration is done later by 1785 * pcie_bus_configure_settings() 1786 */ 1787 if (pcie_bus_config != PCIE_BUS_DEFAULT) 1788 return; 1789 1790 mpss = 128 << dev->pcie_mpss; 1791 if (mpss < p_mps && pci_pcie_type(bridge) == PCI_EXP_TYPE_ROOT_PORT) { 1792 pcie_set_mps(bridge, mpss); 1793 pci_info(dev, "Upstream bridge's Max Payload Size set to %d (was %d, max %d)\n", 1794 mpss, p_mps, 128 << bridge->pcie_mpss); 1795 p_mps = pcie_get_mps(bridge); 1796 } 1797 1798 rc = pcie_set_mps(dev, p_mps); 1799 if (rc) { 1800 pci_warn(dev, "can't set Max Payload Size to %d; if necessary, use \"pci=pcie_bus_safe\" and report a bug\n", 1801 p_mps); 1802 return; 1803 } 1804 1805 pci_info(dev, "Max Payload Size set to %d (was %d, max %d)\n", 1806 p_mps, mps, mpss); 1807 } 1808 1809 static struct hpp_type0 pci_default_type0 = { 1810 .revision = 1, 1811 .cache_line_size = 8, 1812 .latency_timer = 0x40, 1813 .enable_serr = 0, 1814 .enable_perr = 0, 1815 }; 1816 1817 static void program_hpp_type0(struct pci_dev *dev, struct hpp_type0 *hpp) 1818 { 1819 u16 pci_cmd, pci_bctl; 1820 1821 if (!hpp) 1822 hpp = &pci_default_type0; 1823 1824 if (hpp->revision > 1) { 1825 pci_warn(dev, "PCI settings rev %d not supported; using defaults\n", 1826 hpp->revision); 1827 hpp = &pci_default_type0; 1828 } 1829 1830 pci_write_config_byte(dev, PCI_CACHE_LINE_SIZE, hpp->cache_line_size); 1831 pci_write_config_byte(dev, PCI_LATENCY_TIMER, hpp->latency_timer); 1832 pci_read_config_word(dev, PCI_COMMAND, &pci_cmd); 1833 if (hpp->enable_serr) 1834 pci_cmd |= PCI_COMMAND_SERR; 1835 if (hpp->enable_perr) 1836 pci_cmd |= PCI_COMMAND_PARITY; 1837 pci_write_config_word(dev, PCI_COMMAND, pci_cmd); 1838 1839 /* Program bridge control value */ 1840 if ((dev->class >> 8) == PCI_CLASS_BRIDGE_PCI) { 1841 pci_write_config_byte(dev, PCI_SEC_LATENCY_TIMER, 1842 hpp->latency_timer); 1843 pci_read_config_word(dev, PCI_BRIDGE_CONTROL, &pci_bctl); 1844 if (hpp->enable_serr) 1845 pci_bctl |= PCI_BRIDGE_CTL_SERR; 1846 if (hpp->enable_perr) 1847 pci_bctl |= PCI_BRIDGE_CTL_PARITY; 1848 pci_write_config_word(dev, PCI_BRIDGE_CONTROL, pci_bctl); 1849 } 1850 } 1851 1852 static void program_hpp_type1(struct pci_dev *dev, struct hpp_type1 *hpp) 1853 { 1854 int pos; 1855 1856 if (!hpp) 1857 return; 1858 1859 pos = pci_find_capability(dev, PCI_CAP_ID_PCIX); 1860 if (!pos) 1861 return; 1862 1863 pci_warn(dev, "PCI-X settings not supported\n"); 1864 } 1865 1866 static bool pcie_root_rcb_set(struct pci_dev *dev) 1867 { 1868 struct pci_dev *rp = pcie_find_root_port(dev); 1869 u16 lnkctl; 1870 1871 if (!rp) 1872 return false; 1873 1874 pcie_capability_read_word(rp, PCI_EXP_LNKCTL, &lnkctl); 1875 if (lnkctl & PCI_EXP_LNKCTL_RCB) 1876 return true; 1877 1878 return false; 1879 } 1880 1881 static void program_hpp_type2(struct pci_dev *dev, struct hpp_type2 *hpp) 1882 { 1883 int pos; 1884 u32 reg32; 1885 1886 if (!hpp) 1887 return; 1888 1889 if (!pci_is_pcie(dev)) 1890 return; 1891 1892 if (hpp->revision > 1) { 1893 pci_warn(dev, "PCIe settings rev %d not supported\n", 1894 hpp->revision); 1895 return; 1896 } 1897 1898 /* 1899 * Don't allow _HPX to change MPS or MRRS settings. We manage 1900 * those to make sure they're consistent with the rest of the 1901 * platform. 1902 */ 1903 hpp->pci_exp_devctl_and |= PCI_EXP_DEVCTL_PAYLOAD | 1904 PCI_EXP_DEVCTL_READRQ; 1905 hpp->pci_exp_devctl_or &= ~(PCI_EXP_DEVCTL_PAYLOAD | 1906 PCI_EXP_DEVCTL_READRQ); 1907 1908 /* Initialize Device Control Register */ 1909 pcie_capability_clear_and_set_word(dev, PCI_EXP_DEVCTL, 1910 ~hpp->pci_exp_devctl_and, hpp->pci_exp_devctl_or); 1911 1912 /* Initialize Link Control Register */ 1913 if (pcie_cap_has_lnkctl(dev)) { 1914 1915 /* 1916 * If the Root Port supports Read Completion Boundary of 1917 * 128, set RCB to 128. Otherwise, clear it. 1918 */ 1919 hpp->pci_exp_lnkctl_and |= PCI_EXP_LNKCTL_RCB; 1920 hpp->pci_exp_lnkctl_or &= ~PCI_EXP_LNKCTL_RCB; 1921 if (pcie_root_rcb_set(dev)) 1922 hpp->pci_exp_lnkctl_or |= PCI_EXP_LNKCTL_RCB; 1923 1924 pcie_capability_clear_and_set_word(dev, PCI_EXP_LNKCTL, 1925 ~hpp->pci_exp_lnkctl_and, hpp->pci_exp_lnkctl_or); 1926 } 1927 1928 /* Find Advanced Error Reporting Enhanced Capability */ 1929 pos = pci_find_ext_capability(dev, PCI_EXT_CAP_ID_ERR); 1930 if (!pos) 1931 return; 1932 1933 /* Initialize Uncorrectable Error Mask Register */ 1934 pci_read_config_dword(dev, pos + PCI_ERR_UNCOR_MASK, ®32); 1935 reg32 = (reg32 & hpp->unc_err_mask_and) | hpp->unc_err_mask_or; 1936 pci_write_config_dword(dev, pos + PCI_ERR_UNCOR_MASK, reg32); 1937 1938 /* Initialize Uncorrectable Error Severity Register */ 1939 pci_read_config_dword(dev, pos + PCI_ERR_UNCOR_SEVER, ®32); 1940 reg32 = (reg32 & hpp->unc_err_sever_and) | hpp->unc_err_sever_or; 1941 pci_write_config_dword(dev, pos + PCI_ERR_UNCOR_SEVER, reg32); 1942 1943 /* Initialize Correctable Error Mask Register */ 1944 pci_read_config_dword(dev, pos + PCI_ERR_COR_MASK, ®32); 1945 reg32 = (reg32 & hpp->cor_err_mask_and) | hpp->cor_err_mask_or; 1946 pci_write_config_dword(dev, pos + PCI_ERR_COR_MASK, reg32); 1947 1948 /* Initialize Advanced Error Capabilities and Control Register */ 1949 pci_read_config_dword(dev, pos + PCI_ERR_CAP, ®32); 1950 reg32 = (reg32 & hpp->adv_err_cap_and) | hpp->adv_err_cap_or; 1951 1952 /* Don't enable ECRC generation or checking if unsupported */ 1953 if (!(reg32 & PCI_ERR_CAP_ECRC_GENC)) 1954 reg32 &= ~PCI_ERR_CAP_ECRC_GENE; 1955 if (!(reg32 & PCI_ERR_CAP_ECRC_CHKC)) 1956 reg32 &= ~PCI_ERR_CAP_ECRC_CHKE; 1957 pci_write_config_dword(dev, pos + PCI_ERR_CAP, reg32); 1958 1959 /* 1960 * FIXME: The following two registers are not supported yet. 1961 * 1962 * o Secondary Uncorrectable Error Severity Register 1963 * o Secondary Uncorrectable Error Mask Register 1964 */ 1965 } 1966 1967 int pci_configure_extended_tags(struct pci_dev *dev, void *ign) 1968 { 1969 struct pci_host_bridge *host; 1970 u32 cap; 1971 u16 ctl; 1972 int ret; 1973 1974 if (!pci_is_pcie(dev)) 1975 return 0; 1976 1977 ret = pcie_capability_read_dword(dev, PCI_EXP_DEVCAP, &cap); 1978 if (ret) 1979 return 0; 1980 1981 if (!(cap & PCI_EXP_DEVCAP_EXT_TAG)) 1982 return 0; 1983 1984 ret = pcie_capability_read_word(dev, PCI_EXP_DEVCTL, &ctl); 1985 if (ret) 1986 return 0; 1987 1988 host = pci_find_host_bridge(dev->bus); 1989 if (!host) 1990 return 0; 1991 1992 /* 1993 * If some device in the hierarchy doesn't handle Extended Tags 1994 * correctly, make sure they're disabled. 1995 */ 1996 if (host->no_ext_tags) { 1997 if (ctl & PCI_EXP_DEVCTL_EXT_TAG) { 1998 pci_info(dev, "disabling Extended Tags\n"); 1999 pcie_capability_clear_word(dev, PCI_EXP_DEVCTL, 2000 PCI_EXP_DEVCTL_EXT_TAG); 2001 } 2002 return 0; 2003 } 2004 2005 if (!(ctl & PCI_EXP_DEVCTL_EXT_TAG)) { 2006 pci_info(dev, "enabling Extended Tags\n"); 2007 pcie_capability_set_word(dev, PCI_EXP_DEVCTL, 2008 PCI_EXP_DEVCTL_EXT_TAG); 2009 } 2010 return 0; 2011 } 2012 2013 /** 2014 * pcie_relaxed_ordering_enabled - Probe for PCIe relaxed ordering enable 2015 * @dev: PCI device to query 2016 * 2017 * Returns true if the device has enabled relaxed ordering attribute. 2018 */ 2019 bool pcie_relaxed_ordering_enabled(struct pci_dev *dev) 2020 { 2021 u16 v; 2022 2023 pcie_capability_read_word(dev, PCI_EXP_DEVCTL, &v); 2024 2025 return !!(v & PCI_EXP_DEVCTL_RELAX_EN); 2026 } 2027 EXPORT_SYMBOL(pcie_relaxed_ordering_enabled); 2028 2029 static void pci_configure_relaxed_ordering(struct pci_dev *dev) 2030 { 2031 struct pci_dev *root; 2032 2033 /* PCI_EXP_DEVICE_RELAX_EN is RsvdP in VFs */ 2034 if (dev->is_virtfn) 2035 return; 2036 2037 if (!pcie_relaxed_ordering_enabled(dev)) 2038 return; 2039 2040 /* 2041 * For now, we only deal with Relaxed Ordering issues with Root 2042 * Ports. Peer-to-Peer DMA is another can of worms. 2043 */ 2044 root = pci_find_pcie_root_port(dev); 2045 if (!root) 2046 return; 2047 2048 if (root->dev_flags & PCI_DEV_FLAGS_NO_RELAXED_ORDERING) { 2049 pcie_capability_clear_word(dev, PCI_EXP_DEVCTL, 2050 PCI_EXP_DEVCTL_RELAX_EN); 2051 pci_info(dev, "Relaxed Ordering disabled because the Root Port didn't support it\n"); 2052 } 2053 } 2054 2055 static void pci_configure_ltr(struct pci_dev *dev) 2056 { 2057 #ifdef CONFIG_PCIEASPM 2058 struct pci_host_bridge *host = pci_find_host_bridge(dev->bus); 2059 u32 cap; 2060 struct pci_dev *bridge; 2061 2062 if (!host->native_ltr) 2063 return; 2064 2065 if (!pci_is_pcie(dev)) 2066 return; 2067 2068 pcie_capability_read_dword(dev, PCI_EXP_DEVCAP2, &cap); 2069 if (!(cap & PCI_EXP_DEVCAP2_LTR)) 2070 return; 2071 2072 /* 2073 * Software must not enable LTR in an Endpoint unless the Root 2074 * Complex and all intermediate Switches indicate support for LTR. 2075 * PCIe r3.1, sec 6.18. 2076 */ 2077 if (pci_pcie_type(dev) == PCI_EXP_TYPE_ROOT_PORT) 2078 dev->ltr_path = 1; 2079 else { 2080 bridge = pci_upstream_bridge(dev); 2081 if (bridge && bridge->ltr_path) 2082 dev->ltr_path = 1; 2083 } 2084 2085 if (dev->ltr_path) 2086 pcie_capability_set_word(dev, PCI_EXP_DEVCTL2, 2087 PCI_EXP_DEVCTL2_LTR_EN); 2088 #endif 2089 } 2090 2091 static void pci_configure_eetlp_prefix(struct pci_dev *dev) 2092 { 2093 #ifdef CONFIG_PCI_PASID 2094 struct pci_dev *bridge; 2095 int pcie_type; 2096 u32 cap; 2097 2098 if (!pci_is_pcie(dev)) 2099 return; 2100 2101 pcie_capability_read_dword(dev, PCI_EXP_DEVCAP2, &cap); 2102 if (!(cap & PCI_EXP_DEVCAP2_EE_PREFIX)) 2103 return; 2104 2105 pcie_type = pci_pcie_type(dev); 2106 if (pcie_type == PCI_EXP_TYPE_ROOT_PORT || 2107 pcie_type == PCI_EXP_TYPE_RC_END) 2108 dev->eetlp_prefix_path = 1; 2109 else { 2110 bridge = pci_upstream_bridge(dev); 2111 if (bridge && bridge->eetlp_prefix_path) 2112 dev->eetlp_prefix_path = 1; 2113 } 2114 #endif 2115 } 2116 2117 static void pci_configure_device(struct pci_dev *dev) 2118 { 2119 struct hotplug_params hpp; 2120 int ret; 2121 2122 pci_configure_mps(dev); 2123 pci_configure_extended_tags(dev, NULL); 2124 pci_configure_relaxed_ordering(dev); 2125 pci_configure_ltr(dev); 2126 pci_configure_eetlp_prefix(dev); 2127 2128 memset(&hpp, 0, sizeof(hpp)); 2129 ret = pci_get_hp_params(dev, &hpp); 2130 if (ret) 2131 return; 2132 2133 program_hpp_type2(dev, hpp.t2); 2134 program_hpp_type1(dev, hpp.t1); 2135 program_hpp_type0(dev, hpp.t0); 2136 } 2137 2138 static void pci_release_capabilities(struct pci_dev *dev) 2139 { 2140 pci_aer_exit(dev); 2141 pci_vpd_release(dev); 2142 pci_iov_release(dev); 2143 pci_free_cap_save_buffers(dev); 2144 } 2145 2146 /** 2147 * pci_release_dev - Free a PCI device structure when all users of it are 2148 * finished 2149 * @dev: device that's been disconnected 2150 * 2151 * Will be called only by the device core when all users of this PCI device are 2152 * done. 2153 */ 2154 static void pci_release_dev(struct device *dev) 2155 { 2156 struct pci_dev *pci_dev; 2157 2158 pci_dev = to_pci_dev(dev); 2159 pci_release_capabilities(pci_dev); 2160 pci_release_of_node(pci_dev); 2161 pcibios_release_device(pci_dev); 2162 pci_bus_put(pci_dev->bus); 2163 kfree(pci_dev->driver_override); 2164 bitmap_free(pci_dev->dma_alias_mask); 2165 kfree(pci_dev); 2166 } 2167 2168 struct pci_dev *pci_alloc_dev(struct pci_bus *bus) 2169 { 2170 struct pci_dev *dev; 2171 2172 dev = kzalloc(sizeof(struct pci_dev), GFP_KERNEL); 2173 if (!dev) 2174 return NULL; 2175 2176 INIT_LIST_HEAD(&dev->bus_list); 2177 dev->dev.type = &pci_dev_type; 2178 dev->bus = pci_bus_get(bus); 2179 2180 return dev; 2181 } 2182 EXPORT_SYMBOL(pci_alloc_dev); 2183 2184 static bool pci_bus_crs_vendor_id(u32 l) 2185 { 2186 return (l & 0xffff) == 0x0001; 2187 } 2188 2189 static bool pci_bus_wait_crs(struct pci_bus *bus, int devfn, u32 *l, 2190 int timeout) 2191 { 2192 int delay = 1; 2193 2194 if (!pci_bus_crs_vendor_id(*l)) 2195 return true; /* not a CRS completion */ 2196 2197 if (!timeout) 2198 return false; /* CRS, but caller doesn't want to wait */ 2199 2200 /* 2201 * We got the reserved Vendor ID that indicates a completion with 2202 * Configuration Request Retry Status (CRS). Retry until we get a 2203 * valid Vendor ID or we time out. 2204 */ 2205 while (pci_bus_crs_vendor_id(*l)) { 2206 if (delay > timeout) { 2207 pr_warn("pci %04x:%02x:%02x.%d: not ready after %dms; giving up\n", 2208 pci_domain_nr(bus), bus->number, 2209 PCI_SLOT(devfn), PCI_FUNC(devfn), delay - 1); 2210 2211 return false; 2212 } 2213 if (delay >= 1000) 2214 pr_info("pci %04x:%02x:%02x.%d: not ready after %dms; waiting\n", 2215 pci_domain_nr(bus), bus->number, 2216 PCI_SLOT(devfn), PCI_FUNC(devfn), delay - 1); 2217 2218 msleep(delay); 2219 delay *= 2; 2220 2221 if (pci_bus_read_config_dword(bus, devfn, PCI_VENDOR_ID, l)) 2222 return false; 2223 } 2224 2225 if (delay >= 1000) 2226 pr_info("pci %04x:%02x:%02x.%d: ready after %dms\n", 2227 pci_domain_nr(bus), bus->number, 2228 PCI_SLOT(devfn), PCI_FUNC(devfn), delay - 1); 2229 2230 return true; 2231 } 2232 2233 bool pci_bus_generic_read_dev_vendor_id(struct pci_bus *bus, int devfn, u32 *l, 2234 int timeout) 2235 { 2236 if (pci_bus_read_config_dword(bus, devfn, PCI_VENDOR_ID, l)) 2237 return false; 2238 2239 /* Some broken boards return 0 or ~0 if a slot is empty: */ 2240 if (*l == 0xffffffff || *l == 0x00000000 || 2241 *l == 0x0000ffff || *l == 0xffff0000) 2242 return false; 2243 2244 if (pci_bus_crs_vendor_id(*l)) 2245 return pci_bus_wait_crs(bus, devfn, l, timeout); 2246 2247 return true; 2248 } 2249 2250 bool pci_bus_read_dev_vendor_id(struct pci_bus *bus, int devfn, u32 *l, 2251 int timeout) 2252 { 2253 #ifdef CONFIG_PCI_QUIRKS 2254 struct pci_dev *bridge = bus->self; 2255 2256 /* 2257 * Certain IDT switches have an issue where they improperly trigger 2258 * ACS Source Validation errors on completions for config reads. 2259 */ 2260 if (bridge && bridge->vendor == PCI_VENDOR_ID_IDT && 2261 bridge->device == 0x80b5) 2262 return pci_idt_bus_quirk(bus, devfn, l, timeout); 2263 #endif 2264 2265 return pci_bus_generic_read_dev_vendor_id(bus, devfn, l, timeout); 2266 } 2267 EXPORT_SYMBOL(pci_bus_read_dev_vendor_id); 2268 2269 /* 2270 * Read the config data for a PCI device, sanity-check it, 2271 * and fill in the dev structure. 2272 */ 2273 static struct pci_dev *pci_scan_device(struct pci_bus *bus, int devfn) 2274 { 2275 struct pci_dev *dev; 2276 u32 l; 2277 2278 if (!pci_bus_read_dev_vendor_id(bus, devfn, &l, 60*1000)) 2279 return NULL; 2280 2281 dev = pci_alloc_dev(bus); 2282 if (!dev) 2283 return NULL; 2284 2285 dev->devfn = devfn; 2286 dev->vendor = l & 0xffff; 2287 dev->device = (l >> 16) & 0xffff; 2288 2289 pci_set_of_node(dev); 2290 2291 if (pci_setup_device(dev)) { 2292 pci_bus_put(dev->bus); 2293 kfree(dev); 2294 return NULL; 2295 } 2296 2297 return dev; 2298 } 2299 2300 static void pcie_report_downtraining(struct pci_dev *dev) 2301 { 2302 if (!pci_is_pcie(dev)) 2303 return; 2304 2305 /* Look from the device up to avoid downstream ports with no devices */ 2306 if ((pci_pcie_type(dev) != PCI_EXP_TYPE_ENDPOINT) && 2307 (pci_pcie_type(dev) != PCI_EXP_TYPE_LEG_END) && 2308 (pci_pcie_type(dev) != PCI_EXP_TYPE_UPSTREAM)) 2309 return; 2310 2311 /* Multi-function PCIe devices share the same link/status */ 2312 if (PCI_FUNC(dev->devfn) != 0 || dev->is_virtfn) 2313 return; 2314 2315 /* Print link status only if the device is constrained by the fabric */ 2316 __pcie_print_link_status(dev, false); 2317 } 2318 2319 static void pci_init_capabilities(struct pci_dev *dev) 2320 { 2321 /* Enhanced Allocation */ 2322 pci_ea_init(dev); 2323 2324 /* Setup MSI caps & disable MSI/MSI-X interrupts */ 2325 pci_msi_setup_pci_dev(dev); 2326 2327 /* Buffers for saving PCIe and PCI-X capabilities */ 2328 pci_allocate_cap_save_buffers(dev); 2329 2330 /* Power Management */ 2331 pci_pm_init(dev); 2332 2333 /* Vital Product Data */ 2334 pci_vpd_init(dev); 2335 2336 /* Alternative Routing-ID Forwarding */ 2337 pci_configure_ari(dev); 2338 2339 /* Single Root I/O Virtualization */ 2340 pci_iov_init(dev); 2341 2342 /* Address Translation Services */ 2343 pci_ats_init(dev); 2344 2345 /* Enable ACS P2P upstream forwarding */ 2346 pci_enable_acs(dev); 2347 2348 /* Precision Time Measurement */ 2349 pci_ptm_init(dev); 2350 2351 /* Advanced Error Reporting */ 2352 pci_aer_init(dev); 2353 2354 pcie_report_downtraining(dev); 2355 2356 if (pci_probe_reset_function(dev) == 0) 2357 dev->reset_fn = 1; 2358 } 2359 2360 /* 2361 * This is the equivalent of pci_host_bridge_msi_domain() that acts on 2362 * devices. Firmware interfaces that can select the MSI domain on a 2363 * per-device basis should be called from here. 2364 */ 2365 static struct irq_domain *pci_dev_msi_domain(struct pci_dev *dev) 2366 { 2367 struct irq_domain *d; 2368 2369 /* 2370 * If a domain has been set through the pcibios_add_device() 2371 * callback, then this is the one (platform code knows best). 2372 */ 2373 d = dev_get_msi_domain(&dev->dev); 2374 if (d) 2375 return d; 2376 2377 /* 2378 * Let's see if we have a firmware interface able to provide 2379 * the domain. 2380 */ 2381 d = pci_msi_get_device_domain(dev); 2382 if (d) 2383 return d; 2384 2385 return NULL; 2386 } 2387 2388 static void pci_set_msi_domain(struct pci_dev *dev) 2389 { 2390 struct irq_domain *d; 2391 2392 /* 2393 * If the platform or firmware interfaces cannot supply a 2394 * device-specific MSI domain, then inherit the default domain 2395 * from the host bridge itself. 2396 */ 2397 d = pci_dev_msi_domain(dev); 2398 if (!d) 2399 d = dev_get_msi_domain(&dev->bus->dev); 2400 2401 dev_set_msi_domain(&dev->dev, d); 2402 } 2403 2404 void pci_device_add(struct pci_dev *dev, struct pci_bus *bus) 2405 { 2406 int ret; 2407 2408 pci_configure_device(dev); 2409 2410 device_initialize(&dev->dev); 2411 dev->dev.release = pci_release_dev; 2412 2413 set_dev_node(&dev->dev, pcibus_to_node(bus)); 2414 dev->dev.dma_mask = &dev->dma_mask; 2415 dev->dev.dma_parms = &dev->dma_parms; 2416 dev->dev.coherent_dma_mask = 0xffffffffull; 2417 2418 dma_set_max_seg_size(&dev->dev, 65536); 2419 dma_set_seg_boundary(&dev->dev, 0xffffffff); 2420 2421 /* Fix up broken headers */ 2422 pci_fixup_device(pci_fixup_header, dev); 2423 2424 /* Moved out from quirk header fixup code */ 2425 pci_reassigndev_resource_alignment(dev); 2426 2427 /* Clear the state_saved flag */ 2428 dev->state_saved = false; 2429 2430 /* Initialize various capabilities */ 2431 pci_init_capabilities(dev); 2432 2433 /* 2434 * Add the device to our list of discovered devices 2435 * and the bus list for fixup functions, etc. 2436 */ 2437 down_write(&pci_bus_sem); 2438 list_add_tail(&dev->bus_list, &bus->devices); 2439 up_write(&pci_bus_sem); 2440 2441 ret = pcibios_add_device(dev); 2442 WARN_ON(ret < 0); 2443 2444 /* Set up MSI IRQ domain */ 2445 pci_set_msi_domain(dev); 2446 2447 /* Notifier could use PCI capabilities */ 2448 dev->match_driver = false; 2449 ret = device_add(&dev->dev); 2450 WARN_ON(ret < 0); 2451 } 2452 2453 struct pci_dev *pci_scan_single_device(struct pci_bus *bus, int devfn) 2454 { 2455 struct pci_dev *dev; 2456 2457 dev = pci_get_slot(bus, devfn); 2458 if (dev) { 2459 pci_dev_put(dev); 2460 return dev; 2461 } 2462 2463 dev = pci_scan_device(bus, devfn); 2464 if (!dev) 2465 return NULL; 2466 2467 pci_device_add(dev, bus); 2468 2469 return dev; 2470 } 2471 EXPORT_SYMBOL(pci_scan_single_device); 2472 2473 static unsigned next_fn(struct pci_bus *bus, struct pci_dev *dev, unsigned fn) 2474 { 2475 int pos; 2476 u16 cap = 0; 2477 unsigned next_fn; 2478 2479 if (pci_ari_enabled(bus)) { 2480 if (!dev) 2481 return 0; 2482 pos = pci_find_ext_capability(dev, PCI_EXT_CAP_ID_ARI); 2483 if (!pos) 2484 return 0; 2485 2486 pci_read_config_word(dev, pos + PCI_ARI_CAP, &cap); 2487 next_fn = PCI_ARI_CAP_NFN(cap); 2488 if (next_fn <= fn) 2489 return 0; /* protect against malformed list */ 2490 2491 return next_fn; 2492 } 2493 2494 /* dev may be NULL for non-contiguous multifunction devices */ 2495 if (!dev || dev->multifunction) 2496 return (fn + 1) % 8; 2497 2498 return 0; 2499 } 2500 2501 static int only_one_child(struct pci_bus *bus) 2502 { 2503 struct pci_dev *bridge = bus->self; 2504 2505 /* 2506 * Systems with unusual topologies set PCI_SCAN_ALL_PCIE_DEVS so 2507 * we scan for all possible devices, not just Device 0. 2508 */ 2509 if (pci_has_flag(PCI_SCAN_ALL_PCIE_DEVS)) 2510 return 0; 2511 2512 /* 2513 * A PCIe Downstream Port normally leads to a Link with only Device 2514 * 0 on it (PCIe spec r3.1, sec 7.3.1). As an optimization, scan 2515 * only for Device 0 in that situation. 2516 * 2517 * Checking has_secondary_link is a hack to identify Downstream 2518 * Ports because sometimes Switches are configured such that the 2519 * PCIe Port Type labels are backwards. 2520 */ 2521 if (bridge && pci_is_pcie(bridge) && bridge->has_secondary_link) 2522 return 1; 2523 2524 return 0; 2525 } 2526 2527 /** 2528 * pci_scan_slot - Scan a PCI slot on a bus for devices 2529 * @bus: PCI bus to scan 2530 * @devfn: slot number to scan (must have zero function) 2531 * 2532 * Scan a PCI slot on the specified PCI bus for devices, adding 2533 * discovered devices to the @bus->devices list. New devices 2534 * will not have is_added set. 2535 * 2536 * Returns the number of new devices found. 2537 */ 2538 int pci_scan_slot(struct pci_bus *bus, int devfn) 2539 { 2540 unsigned fn, nr = 0; 2541 struct pci_dev *dev; 2542 2543 if (only_one_child(bus) && (devfn > 0)) 2544 return 0; /* Already scanned the entire slot */ 2545 2546 dev = pci_scan_single_device(bus, devfn); 2547 if (!dev) 2548 return 0; 2549 if (!pci_dev_is_added(dev)) 2550 nr++; 2551 2552 for (fn = next_fn(bus, dev, 0); fn > 0; fn = next_fn(bus, dev, fn)) { 2553 dev = pci_scan_single_device(bus, devfn + fn); 2554 if (dev) { 2555 if (!pci_dev_is_added(dev)) 2556 nr++; 2557 dev->multifunction = 1; 2558 } 2559 } 2560 2561 /* Only one slot has PCIe device */ 2562 if (bus->self && nr) 2563 pcie_aspm_init_link_state(bus->self); 2564 2565 return nr; 2566 } 2567 EXPORT_SYMBOL(pci_scan_slot); 2568 2569 static int pcie_find_smpss(struct pci_dev *dev, void *data) 2570 { 2571 u8 *smpss = data; 2572 2573 if (!pci_is_pcie(dev)) 2574 return 0; 2575 2576 /* 2577 * We don't have a way to change MPS settings on devices that have 2578 * drivers attached. A hot-added device might support only the minimum 2579 * MPS setting (MPS=128). Therefore, if the fabric contains a bridge 2580 * where devices may be hot-added, we limit the fabric MPS to 128 so 2581 * hot-added devices will work correctly. 2582 * 2583 * However, if we hot-add a device to a slot directly below a Root 2584 * Port, it's impossible for there to be other existing devices below 2585 * the port. We don't limit the MPS in this case because we can 2586 * reconfigure MPS on both the Root Port and the hot-added device, 2587 * and there are no other devices involved. 2588 * 2589 * Note that this PCIE_BUS_SAFE path assumes no peer-to-peer DMA. 2590 */ 2591 if (dev->is_hotplug_bridge && 2592 pci_pcie_type(dev) != PCI_EXP_TYPE_ROOT_PORT) 2593 *smpss = 0; 2594 2595 if (*smpss > dev->pcie_mpss) 2596 *smpss = dev->pcie_mpss; 2597 2598 return 0; 2599 } 2600 2601 static void pcie_write_mps(struct pci_dev *dev, int mps) 2602 { 2603 int rc; 2604 2605 if (pcie_bus_config == PCIE_BUS_PERFORMANCE) { 2606 mps = 128 << dev->pcie_mpss; 2607 2608 if (pci_pcie_type(dev) != PCI_EXP_TYPE_ROOT_PORT && 2609 dev->bus->self) 2610 2611 /* 2612 * For "Performance", the assumption is made that 2613 * downstream communication will never be larger than 2614 * the MRRS. So, the MPS only needs to be configured 2615 * for the upstream communication. This being the case, 2616 * walk from the top down and set the MPS of the child 2617 * to that of the parent bus. 2618 * 2619 * Configure the device MPS with the smaller of the 2620 * device MPSS or the bridge MPS (which is assumed to be 2621 * properly configured at this point to the largest 2622 * allowable MPS based on its parent bus). 2623 */ 2624 mps = min(mps, pcie_get_mps(dev->bus->self)); 2625 } 2626 2627 rc = pcie_set_mps(dev, mps); 2628 if (rc) 2629 pci_err(dev, "Failed attempting to set the MPS\n"); 2630 } 2631 2632 static void pcie_write_mrrs(struct pci_dev *dev) 2633 { 2634 int rc, mrrs; 2635 2636 /* 2637 * In the "safe" case, do not configure the MRRS. There appear to be 2638 * issues with setting MRRS to 0 on a number of devices. 2639 */ 2640 if (pcie_bus_config != PCIE_BUS_PERFORMANCE) 2641 return; 2642 2643 /* 2644 * For max performance, the MRRS must be set to the largest supported 2645 * value. However, it cannot be configured larger than the MPS the 2646 * device or the bus can support. This should already be properly 2647 * configured by a prior call to pcie_write_mps(). 2648 */ 2649 mrrs = pcie_get_mps(dev); 2650 2651 /* 2652 * MRRS is a R/W register. Invalid values can be written, but a 2653 * subsequent read will verify if the value is acceptable or not. 2654 * If the MRRS value provided is not acceptable (e.g., too large), 2655 * shrink the value until it is acceptable to the HW. 2656 */ 2657 while (mrrs != pcie_get_readrq(dev) && mrrs >= 128) { 2658 rc = pcie_set_readrq(dev, mrrs); 2659 if (!rc) 2660 break; 2661 2662 pci_warn(dev, "Failed attempting to set the MRRS\n"); 2663 mrrs /= 2; 2664 } 2665 2666 if (mrrs < 128) 2667 pci_err(dev, "MRRS was unable to be configured with a safe value. If problems are experienced, try running with pci=pcie_bus_safe\n"); 2668 } 2669 2670 static int pcie_bus_configure_set(struct pci_dev *dev, void *data) 2671 { 2672 int mps, orig_mps; 2673 2674 if (!pci_is_pcie(dev)) 2675 return 0; 2676 2677 if (pcie_bus_config == PCIE_BUS_TUNE_OFF || 2678 pcie_bus_config == PCIE_BUS_DEFAULT) 2679 return 0; 2680 2681 mps = 128 << *(u8 *)data; 2682 orig_mps = pcie_get_mps(dev); 2683 2684 pcie_write_mps(dev, mps); 2685 pcie_write_mrrs(dev); 2686 2687 pci_info(dev, "Max Payload Size set to %4d/%4d (was %4d), Max Read Rq %4d\n", 2688 pcie_get_mps(dev), 128 << dev->pcie_mpss, 2689 orig_mps, pcie_get_readrq(dev)); 2690 2691 return 0; 2692 } 2693 2694 /* 2695 * pcie_bus_configure_settings() requires that pci_walk_bus work in a top-down, 2696 * parents then children fashion. If this changes, then this code will not 2697 * work as designed. 2698 */ 2699 void pcie_bus_configure_settings(struct pci_bus *bus) 2700 { 2701 u8 smpss = 0; 2702 2703 if (!bus->self) 2704 return; 2705 2706 if (!pci_is_pcie(bus->self)) 2707 return; 2708 2709 /* 2710 * FIXME - Peer to peer DMA is possible, though the endpoint would need 2711 * to be aware of the MPS of the destination. To work around this, 2712 * simply force the MPS of the entire system to the smallest possible. 2713 */ 2714 if (pcie_bus_config == PCIE_BUS_PEER2PEER) 2715 smpss = 0; 2716 2717 if (pcie_bus_config == PCIE_BUS_SAFE) { 2718 smpss = bus->self->pcie_mpss; 2719 2720 pcie_find_smpss(bus->self, &smpss); 2721 pci_walk_bus(bus, pcie_find_smpss, &smpss); 2722 } 2723 2724 pcie_bus_configure_set(bus->self, &smpss); 2725 pci_walk_bus(bus, pcie_bus_configure_set, &smpss); 2726 } 2727 EXPORT_SYMBOL_GPL(pcie_bus_configure_settings); 2728 2729 /* 2730 * Called after each bus is probed, but before its children are examined. This 2731 * is marked as __weak because multiple architectures define it. 2732 */ 2733 void __weak pcibios_fixup_bus(struct pci_bus *bus) 2734 { 2735 /* nothing to do, expected to be removed in the future */ 2736 } 2737 2738 /** 2739 * pci_scan_child_bus_extend() - Scan devices below a bus 2740 * @bus: Bus to scan for devices 2741 * @available_buses: Total number of buses available (%0 does not try to 2742 * extend beyond the minimal) 2743 * 2744 * Scans devices below @bus including subordinate buses. Returns new 2745 * subordinate number including all the found devices. Passing 2746 * @available_buses causes the remaining bus space to be distributed 2747 * equally between hotplug-capable bridges to allow future extension of the 2748 * hierarchy. 2749 */ 2750 static unsigned int pci_scan_child_bus_extend(struct pci_bus *bus, 2751 unsigned int available_buses) 2752 { 2753 unsigned int used_buses, normal_bridges = 0, hotplug_bridges = 0; 2754 unsigned int start = bus->busn_res.start; 2755 unsigned int devfn, fn, cmax, max = start; 2756 struct pci_dev *dev; 2757 int nr_devs; 2758 2759 dev_dbg(&bus->dev, "scanning bus\n"); 2760 2761 /* Go find them, Rover! */ 2762 for (devfn = 0; devfn < 256; devfn += 8) { 2763 nr_devs = pci_scan_slot(bus, devfn); 2764 2765 /* 2766 * The Jailhouse hypervisor may pass individual functions of a 2767 * multi-function device to a guest without passing function 0. 2768 * Look for them as well. 2769 */ 2770 if (jailhouse_paravirt() && nr_devs == 0) { 2771 for (fn = 1; fn < 8; fn++) { 2772 dev = pci_scan_single_device(bus, devfn + fn); 2773 if (dev) 2774 dev->multifunction = 1; 2775 } 2776 } 2777 } 2778 2779 /* Reserve buses for SR-IOV capability */ 2780 used_buses = pci_iov_bus_range(bus); 2781 max += used_buses; 2782 2783 /* 2784 * After performing arch-dependent fixup of the bus, look behind 2785 * all PCI-to-PCI bridges on this bus. 2786 */ 2787 if (!bus->is_added) { 2788 dev_dbg(&bus->dev, "fixups for bus\n"); 2789 pcibios_fixup_bus(bus); 2790 bus->is_added = 1; 2791 } 2792 2793 /* 2794 * Calculate how many hotplug bridges and normal bridges there 2795 * are on this bus. We will distribute the additional available 2796 * buses between hotplug bridges. 2797 */ 2798 for_each_pci_bridge(dev, bus) { 2799 if (dev->is_hotplug_bridge) 2800 hotplug_bridges++; 2801 else 2802 normal_bridges++; 2803 } 2804 2805 /* 2806 * Scan bridges that are already configured. We don't touch them 2807 * unless they are misconfigured (which will be done in the second 2808 * scan below). 2809 */ 2810 for_each_pci_bridge(dev, bus) { 2811 cmax = max; 2812 max = pci_scan_bridge_extend(bus, dev, max, 0, 0); 2813 2814 /* 2815 * Reserve one bus for each bridge now to avoid extending 2816 * hotplug bridges too much during the second scan below. 2817 */ 2818 used_buses++; 2819 if (cmax - max > 1) 2820 used_buses += cmax - max - 1; 2821 } 2822 2823 /* Scan bridges that need to be reconfigured */ 2824 for_each_pci_bridge(dev, bus) { 2825 unsigned int buses = 0; 2826 2827 if (!hotplug_bridges && normal_bridges == 1) { 2828 2829 /* 2830 * There is only one bridge on the bus (upstream 2831 * port) so it gets all available buses which it 2832 * can then distribute to the possible hotplug 2833 * bridges below. 2834 */ 2835 buses = available_buses; 2836 } else if (dev->is_hotplug_bridge) { 2837 2838 /* 2839 * Distribute the extra buses between hotplug 2840 * bridges if any. 2841 */ 2842 buses = available_buses / hotplug_bridges; 2843 buses = min(buses, available_buses - used_buses + 1); 2844 } 2845 2846 cmax = max; 2847 max = pci_scan_bridge_extend(bus, dev, cmax, buses, 1); 2848 /* One bus is already accounted so don't add it again */ 2849 if (max - cmax > 1) 2850 used_buses += max - cmax - 1; 2851 } 2852 2853 /* 2854 * Make sure a hotplug bridge has at least the minimum requested 2855 * number of buses but allow it to grow up to the maximum available 2856 * bus number of there is room. 2857 */ 2858 if (bus->self && bus->self->is_hotplug_bridge) { 2859 used_buses = max_t(unsigned int, available_buses, 2860 pci_hotplug_bus_size - 1); 2861 if (max - start < used_buses) { 2862 max = start + used_buses; 2863 2864 /* Do not allocate more buses than we have room left */ 2865 if (max > bus->busn_res.end) 2866 max = bus->busn_res.end; 2867 2868 dev_dbg(&bus->dev, "%pR extended by %#02x\n", 2869 &bus->busn_res, max - start); 2870 } 2871 } 2872 2873 /* 2874 * We've scanned the bus and so we know all about what's on 2875 * the other side of any bridges that may be on this bus plus 2876 * any devices. 2877 * 2878 * Return how far we've got finding sub-buses. 2879 */ 2880 dev_dbg(&bus->dev, "bus scan returning with max=%02x\n", max); 2881 return max; 2882 } 2883 2884 /** 2885 * pci_scan_child_bus() - Scan devices below a bus 2886 * @bus: Bus to scan for devices 2887 * 2888 * Scans devices below @bus including subordinate buses. Returns new 2889 * subordinate number including all the found devices. 2890 */ 2891 unsigned int pci_scan_child_bus(struct pci_bus *bus) 2892 { 2893 return pci_scan_child_bus_extend(bus, 0); 2894 } 2895 EXPORT_SYMBOL_GPL(pci_scan_child_bus); 2896 2897 /** 2898 * pcibios_root_bridge_prepare - Platform-specific host bridge setup 2899 * @bridge: Host bridge to set up 2900 * 2901 * Default empty implementation. Replace with an architecture-specific setup 2902 * routine, if necessary. 2903 */ 2904 int __weak pcibios_root_bridge_prepare(struct pci_host_bridge *bridge) 2905 { 2906 return 0; 2907 } 2908 2909 void __weak pcibios_add_bus(struct pci_bus *bus) 2910 { 2911 } 2912 2913 void __weak pcibios_remove_bus(struct pci_bus *bus) 2914 { 2915 } 2916 2917 struct pci_bus *pci_create_root_bus(struct device *parent, int bus, 2918 struct pci_ops *ops, void *sysdata, struct list_head *resources) 2919 { 2920 int error; 2921 struct pci_host_bridge *bridge; 2922 2923 bridge = pci_alloc_host_bridge(0); 2924 if (!bridge) 2925 return NULL; 2926 2927 bridge->dev.parent = parent; 2928 2929 list_splice_init(resources, &bridge->windows); 2930 bridge->sysdata = sysdata; 2931 bridge->busnr = bus; 2932 bridge->ops = ops; 2933 2934 error = pci_register_host_bridge(bridge); 2935 if (error < 0) 2936 goto err_out; 2937 2938 return bridge->bus; 2939 2940 err_out: 2941 kfree(bridge); 2942 return NULL; 2943 } 2944 EXPORT_SYMBOL_GPL(pci_create_root_bus); 2945 2946 int pci_host_probe(struct pci_host_bridge *bridge) 2947 { 2948 struct pci_bus *bus, *child; 2949 int ret; 2950 2951 ret = pci_scan_root_bus_bridge(bridge); 2952 if (ret < 0) { 2953 dev_err(bridge->dev.parent, "Scanning root bridge failed"); 2954 return ret; 2955 } 2956 2957 bus = bridge->bus; 2958 2959 /* 2960 * We insert PCI resources into the iomem_resource and 2961 * ioport_resource trees in either pci_bus_claim_resources() 2962 * or pci_bus_assign_resources(). 2963 */ 2964 if (pci_has_flag(PCI_PROBE_ONLY)) { 2965 pci_bus_claim_resources(bus); 2966 } else { 2967 pci_bus_size_bridges(bus); 2968 pci_bus_assign_resources(bus); 2969 2970 list_for_each_entry(child, &bus->children, node) 2971 pcie_bus_configure_settings(child); 2972 } 2973 2974 pci_bus_add_devices(bus); 2975 return 0; 2976 } 2977 EXPORT_SYMBOL_GPL(pci_host_probe); 2978 2979 int pci_bus_insert_busn_res(struct pci_bus *b, int bus, int bus_max) 2980 { 2981 struct resource *res = &b->busn_res; 2982 struct resource *parent_res, *conflict; 2983 2984 res->start = bus; 2985 res->end = bus_max; 2986 res->flags = IORESOURCE_BUS; 2987 2988 if (!pci_is_root_bus(b)) 2989 parent_res = &b->parent->busn_res; 2990 else { 2991 parent_res = get_pci_domain_busn_res(pci_domain_nr(b)); 2992 res->flags |= IORESOURCE_PCI_FIXED; 2993 } 2994 2995 conflict = request_resource_conflict(parent_res, res); 2996 2997 if (conflict) 2998 dev_printk(KERN_DEBUG, &b->dev, 2999 "busn_res: can not insert %pR under %s%pR (conflicts with %s %pR)\n", 3000 res, pci_is_root_bus(b) ? "domain " : "", 3001 parent_res, conflict->name, conflict); 3002 3003 return conflict == NULL; 3004 } 3005 3006 int pci_bus_update_busn_res_end(struct pci_bus *b, int bus_max) 3007 { 3008 struct resource *res = &b->busn_res; 3009 struct resource old_res = *res; 3010 resource_size_t size; 3011 int ret; 3012 3013 if (res->start > bus_max) 3014 return -EINVAL; 3015 3016 size = bus_max - res->start + 1; 3017 ret = adjust_resource(res, res->start, size); 3018 dev_printk(KERN_DEBUG, &b->dev, 3019 "busn_res: %pR end %s updated to %02x\n", 3020 &old_res, ret ? "can not be" : "is", bus_max); 3021 3022 if (!ret && !res->parent) 3023 pci_bus_insert_busn_res(b, res->start, res->end); 3024 3025 return ret; 3026 } 3027 3028 void pci_bus_release_busn_res(struct pci_bus *b) 3029 { 3030 struct resource *res = &b->busn_res; 3031 int ret; 3032 3033 if (!res->flags || !res->parent) 3034 return; 3035 3036 ret = release_resource(res); 3037 dev_printk(KERN_DEBUG, &b->dev, 3038 "busn_res: %pR %s released\n", 3039 res, ret ? "can not be" : "is"); 3040 } 3041 3042 int pci_scan_root_bus_bridge(struct pci_host_bridge *bridge) 3043 { 3044 struct resource_entry *window; 3045 bool found = false; 3046 struct pci_bus *b; 3047 int max, bus, ret; 3048 3049 if (!bridge) 3050 return -EINVAL; 3051 3052 resource_list_for_each_entry(window, &bridge->windows) 3053 if (window->res->flags & IORESOURCE_BUS) { 3054 found = true; 3055 break; 3056 } 3057 3058 ret = pci_register_host_bridge(bridge); 3059 if (ret < 0) 3060 return ret; 3061 3062 b = bridge->bus; 3063 bus = bridge->busnr; 3064 3065 if (!found) { 3066 dev_info(&b->dev, 3067 "No busn resource found for root bus, will use [bus %02x-ff]\n", 3068 bus); 3069 pci_bus_insert_busn_res(b, bus, 255); 3070 } 3071 3072 max = pci_scan_child_bus(b); 3073 3074 if (!found) 3075 pci_bus_update_busn_res_end(b, max); 3076 3077 return 0; 3078 } 3079 EXPORT_SYMBOL(pci_scan_root_bus_bridge); 3080 3081 struct pci_bus *pci_scan_root_bus(struct device *parent, int bus, 3082 struct pci_ops *ops, void *sysdata, struct list_head *resources) 3083 { 3084 struct resource_entry *window; 3085 bool found = false; 3086 struct pci_bus *b; 3087 int max; 3088 3089 resource_list_for_each_entry(window, resources) 3090 if (window->res->flags & IORESOURCE_BUS) { 3091 found = true; 3092 break; 3093 } 3094 3095 b = pci_create_root_bus(parent, bus, ops, sysdata, resources); 3096 if (!b) 3097 return NULL; 3098 3099 if (!found) { 3100 dev_info(&b->dev, 3101 "No busn resource found for root bus, will use [bus %02x-ff]\n", 3102 bus); 3103 pci_bus_insert_busn_res(b, bus, 255); 3104 } 3105 3106 max = pci_scan_child_bus(b); 3107 3108 if (!found) 3109 pci_bus_update_busn_res_end(b, max); 3110 3111 return b; 3112 } 3113 EXPORT_SYMBOL(pci_scan_root_bus); 3114 3115 struct pci_bus *pci_scan_bus(int bus, struct pci_ops *ops, 3116 void *sysdata) 3117 { 3118 LIST_HEAD(resources); 3119 struct pci_bus *b; 3120 3121 pci_add_resource(&resources, &ioport_resource); 3122 pci_add_resource(&resources, &iomem_resource); 3123 pci_add_resource(&resources, &busn_resource); 3124 b = pci_create_root_bus(NULL, bus, ops, sysdata, &resources); 3125 if (b) { 3126 pci_scan_child_bus(b); 3127 } else { 3128 pci_free_resource_list(&resources); 3129 } 3130 return b; 3131 } 3132 EXPORT_SYMBOL(pci_scan_bus); 3133 3134 /** 3135 * pci_rescan_bus_bridge_resize - Scan a PCI bus for devices 3136 * @bridge: PCI bridge for the bus to scan 3137 * 3138 * Scan a PCI bus and child buses for new devices, add them, 3139 * and enable them, resizing bridge mmio/io resource if necessary 3140 * and possible. The caller must ensure the child devices are already 3141 * removed for resizing to occur. 3142 * 3143 * Returns the max number of subordinate bus discovered. 3144 */ 3145 unsigned int pci_rescan_bus_bridge_resize(struct pci_dev *bridge) 3146 { 3147 unsigned int max; 3148 struct pci_bus *bus = bridge->subordinate; 3149 3150 max = pci_scan_child_bus(bus); 3151 3152 pci_assign_unassigned_bridge_resources(bridge); 3153 3154 pci_bus_add_devices(bus); 3155 3156 return max; 3157 } 3158 3159 /** 3160 * pci_rescan_bus - Scan a PCI bus for devices 3161 * @bus: PCI bus to scan 3162 * 3163 * Scan a PCI bus and child buses for new devices, add them, 3164 * and enable them. 3165 * 3166 * Returns the max number of subordinate bus discovered. 3167 */ 3168 unsigned int pci_rescan_bus(struct pci_bus *bus) 3169 { 3170 unsigned int max; 3171 3172 max = pci_scan_child_bus(bus); 3173 pci_assign_unassigned_bus_resources(bus); 3174 pci_bus_add_devices(bus); 3175 3176 return max; 3177 } 3178 EXPORT_SYMBOL_GPL(pci_rescan_bus); 3179 3180 /* 3181 * pci_rescan_bus(), pci_rescan_bus_bridge_resize() and PCI device removal 3182 * routines should always be executed under this mutex. 3183 */ 3184 static DEFINE_MUTEX(pci_rescan_remove_lock); 3185 3186 void pci_lock_rescan_remove(void) 3187 { 3188 mutex_lock(&pci_rescan_remove_lock); 3189 } 3190 EXPORT_SYMBOL_GPL(pci_lock_rescan_remove); 3191 3192 void pci_unlock_rescan_remove(void) 3193 { 3194 mutex_unlock(&pci_rescan_remove_lock); 3195 } 3196 EXPORT_SYMBOL_GPL(pci_unlock_rescan_remove); 3197 3198 static int __init pci_sort_bf_cmp(const struct device *d_a, 3199 const struct device *d_b) 3200 { 3201 const struct pci_dev *a = to_pci_dev(d_a); 3202 const struct pci_dev *b = to_pci_dev(d_b); 3203 3204 if (pci_domain_nr(a->bus) < pci_domain_nr(b->bus)) return -1; 3205 else if (pci_domain_nr(a->bus) > pci_domain_nr(b->bus)) return 1; 3206 3207 if (a->bus->number < b->bus->number) return -1; 3208 else if (a->bus->number > b->bus->number) return 1; 3209 3210 if (a->devfn < b->devfn) return -1; 3211 else if (a->devfn > b->devfn) return 1; 3212 3213 return 0; 3214 } 3215 3216 void __init pci_sort_breadthfirst(void) 3217 { 3218 bus_sort_breadthfirst(&pci_bus_type, &pci_sort_bf_cmp); 3219 } 3220 3221 int pci_hp_add_bridge(struct pci_dev *dev) 3222 { 3223 struct pci_bus *parent = dev->bus; 3224 int busnr, start = parent->busn_res.start; 3225 unsigned int available_buses = 0; 3226 int end = parent->busn_res.end; 3227 3228 for (busnr = start; busnr <= end; busnr++) { 3229 if (!pci_find_bus(pci_domain_nr(parent), busnr)) 3230 break; 3231 } 3232 if (busnr-- > end) { 3233 pci_err(dev, "No bus number available for hot-added bridge\n"); 3234 return -1; 3235 } 3236 3237 /* Scan bridges that are already configured */ 3238 busnr = pci_scan_bridge(parent, dev, busnr, 0); 3239 3240 /* 3241 * Distribute the available bus numbers between hotplug-capable 3242 * bridges to make extending the chain later possible. 3243 */ 3244 available_buses = end - busnr; 3245 3246 /* Scan bridges that need to be reconfigured */ 3247 pci_scan_bridge_extend(parent, dev, busnr, available_buses, 1); 3248 3249 if (!dev->subordinate) 3250 return -1; 3251 3252 return 0; 3253 } 3254 EXPORT_SYMBOL_GPL(pci_hp_add_bridge); 3255