1 // SPDX-License-Identifier: GPL-2.0
3 * PCI detection and setup code
6 #include <linux/kernel.h>
7 #include <linux/delay.h>
8 #include <linux/init.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>
23 #define CARDBUS_LATENCY_TIMER 176 /* secondary latency timer */
24 #define CARDBUS_RESERVE_BUSNR 3
26 static struct resource busn_resource = {
30 .flags = IORESOURCE_BUS,
33 /* Ugh. Need to stop exporting this to modules. */
34 LIST_HEAD(pci_root_buses);
35 EXPORT_SYMBOL(pci_root_buses);
37 static LIST_HEAD(pci_domain_busn_res_list);
39 struct pci_domain_busn_res {
40 struct list_head list;
45 static struct resource *get_pci_domain_busn_res(int domain_nr)
47 struct pci_domain_busn_res *r;
49 list_for_each_entry(r, &pci_domain_busn_res_list, list)
50 if (r->domain_nr == domain_nr)
53 r = kzalloc(sizeof(*r), GFP_KERNEL);
57 r->domain_nr = domain_nr;
60 r->res.flags = IORESOURCE_BUS | IORESOURCE_PCI_FIXED;
62 list_add_tail(&r->list, &pci_domain_busn_res_list);
67 static int find_anything(struct device *dev, void *data)
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.
77 int no_pci_devices(void)
82 dev = bus_find_device(&pci_bus_type, NULL, NULL, find_anything);
83 no_devices = (dev == NULL);
87 EXPORT_SYMBOL(no_pci_devices);
92 static void release_pcibus_dev(struct device *dev)
94 struct pci_bus *pci_bus = to_pci_bus(dev);
96 put_device(pci_bus->bridge);
97 pci_bus_remove_resources(pci_bus);
98 pci_release_bus_of_node(pci_bus);
102 static struct class pcibus_class = {
104 .dev_release = &release_pcibus_dev,
105 .dev_groups = pcibus_groups,
108 static int __init pcibus_class_init(void)
110 return class_register(&pcibus_class);
112 postcore_initcall(pcibus_class_init);
114 static u64 pci_size(u64 base, u64 maxbase, u64 mask)
116 u64 size = mask & maxbase; /* Find the significant bits */
121 * Get the lowest of them to find the decode size, and from that
124 size = (size & ~(size-1)) - 1;
127 * base == maxbase can be valid only if the BAR has already been
128 * programmed with all 1s.
130 if (base == maxbase && ((base | size) & mask) != mask)
136 static inline unsigned long decode_bar(struct pci_dev *dev, u32 bar)
141 if ((bar & PCI_BASE_ADDRESS_SPACE) == PCI_BASE_ADDRESS_SPACE_IO) {
142 flags = bar & ~PCI_BASE_ADDRESS_IO_MASK;
143 flags |= IORESOURCE_IO;
147 flags = bar & ~PCI_BASE_ADDRESS_MEM_MASK;
148 flags |= IORESOURCE_MEM;
149 if (flags & PCI_BASE_ADDRESS_MEM_PREFETCH)
150 flags |= IORESOURCE_PREFETCH;
152 mem_type = bar & PCI_BASE_ADDRESS_MEM_TYPE_MASK;
154 case PCI_BASE_ADDRESS_MEM_TYPE_32:
156 case PCI_BASE_ADDRESS_MEM_TYPE_1M:
157 /* 1M mem BAR treated as 32-bit BAR */
159 case PCI_BASE_ADDRESS_MEM_TYPE_64:
160 flags |= IORESOURCE_MEM_64;
163 /* mem unknown type treated as 32-bit BAR */
169 #define PCI_COMMAND_DECODE_ENABLE (PCI_COMMAND_MEMORY | PCI_COMMAND_IO)
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
178 * Returns 1 if the BAR is 64-bit, or 0 if 32-bit.
180 int __pci_read_base(struct pci_dev *dev, enum pci_bar_type type,
181 struct resource *res, unsigned int pos)
183 u32 l = 0, sz = 0, mask;
184 u64 l64, sz64, mask64;
186 struct pci_bus_region region, inverted_region;
188 mask = type ? PCI_ROM_ADDRESS_MASK : ~0;
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);
199 res->name = pci_name(dev);
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);
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
212 if (sz == 0xffffffff)
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.
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;
230 l64 = l & PCI_BASE_ADDRESS_MEM_MASK;
231 sz64 = sz & PCI_BASE_ADDRESS_MEM_MASK;
232 mask64 = (u32)PCI_BASE_ADDRESS_MEM_MASK;
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;
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);
248 l64 |= ((u64)l << 32);
249 sz64 |= ((u64)sz << 32);
250 mask64 |= ((u64)~0 << 32);
253 if (!dev->mmio_always_on && (orig_cmd & PCI_COMMAND_DECODE_ENABLE))
254 pci_write_config_word(dev, PCI_COMMAND, orig_cmd);
259 sz64 = pci_size(l64, sz64, mask64);
261 pci_info(dev, FW_BUG "reg 0x%x: invalid BAR (can't size)\n",
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;
272 pci_err(dev, "reg 0x%x: can't handle BAR larger than 4GB (size %#010llx)\n",
273 pos, (unsigned long long)sz64);
277 if ((sizeof(pci_bus_addr_t) < 8) && l) {
278 /* Above 32-bit boundary; try to reallocate */
279 res->flags |= IORESOURCE_UNSET;
282 pci_info(dev, "reg 0x%x: can't handle BAR above 4GB (bus address %#010llx)\n",
283 pos, (unsigned long long)l64);
289 region.end = l64 + sz64;
291 pcibios_bus_to_resource(dev->bus, res, ®ion);
292 pcibios_resource_to_bus(dev->bus, &inverted_region, res);
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:
300 * resource_to_bus(bus_to_resource(A)) == A
302 * If it doesn't, CPU accesses to "bus_to_resource(A)" will not
303 * be claimed by the device.
305 if (inverted_region.start != region.start) {
306 res->flags |= IORESOURCE_UNSET;
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);
320 pci_printk(KERN_DEBUG, dev, "reg 0x%x: %pR\n", pos, res);
322 return (res->flags & IORESOURCE_MEM_64) ? 1 : 0;
325 static void pci_read_bases(struct pci_dev *dev, unsigned int howmany, int rom)
327 unsigned int pos, reg;
329 if (dev->non_compliant_bars)
332 /* Per PCIe r4.0, sec 9.3.4.1.11, the VF BARs are all RO Zero */
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);
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);
351 static void pci_read_bridge_io(struct pci_bus *child)
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;
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;
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;
373 if ((io_base_lo & PCI_IO_RANGE_TYPE_MASK) == PCI_IO_RANGE_TYPE_32) {
374 u16 io_base_hi, io_limit_hi;
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);
383 res->flags = (io_base_lo & PCI_IO_RANGE_TYPE_MASK) | IORESOURCE_IO;
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);
391 static void pci_read_bridge_mmio(struct pci_bus *child)
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;
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;
405 res->flags = (mem_base_lo & PCI_MEMORY_RANGE_TYPE_MASK) | IORESOURCE_MEM;
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);
413 static void pci_read_bridge_mmio_pref(struct pci_bus *child)
415 struct pci_dev *dev = child->self;
416 u16 mem_base_lo, mem_limit_lo;
418 pci_bus_addr_t base, limit;
419 struct pci_bus_region region;
420 struct resource *res;
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;
428 if ((mem_base_lo & PCI_PREF_RANGE_TYPE_MASK) == PCI_PREF_RANGE_TYPE_64) {
429 u32 mem_base_hi, mem_limit_hi;
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);
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.
439 if (mem_base_hi <= mem_limit_hi) {
440 base64 |= (u64) mem_base_hi << 32;
441 limit64 |= (u64) mem_limit_hi << 32;
445 base = (pci_bus_addr_t) base64;
446 limit = (pci_bus_addr_t) limit64;
448 if (base != base64) {
449 pci_err(dev, "can't handle bridge window above 4GB (bus address %#010llx)\n",
450 (unsigned long long) base64);
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;
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);
466 void pci_read_bridge_bases(struct pci_bus *child)
468 struct pci_dev *dev = child->self;
469 struct resource *res;
472 if (pci_is_root_bus(child)) /* It's a host bus, nothing to read */
475 pci_info(dev, "PCI bridge to %pR%s\n",
477 dev->transparent ? " (subtractive decode)" : "");
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];
483 pci_read_bridge_io(child);
484 pci_read_bridge_mmio(child);
485 pci_read_bridge_mmio_pref(child);
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",
500 static struct pci_bus *pci_alloc_bus(struct pci_bus *parent)
504 b = kzalloc(sizeof(*b), GFP_KERNEL);
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
517 b->domain_nr = parent->domain_nr;
522 static void devm_pci_release_host_bridge_dev(struct device *dev)
524 struct pci_host_bridge *bridge = to_pci_host_bridge(dev);
526 if (bridge->release_fn)
527 bridge->release_fn(bridge);
529 pci_free_resource_list(&bridge->windows);
532 static void pci_release_host_bridge_dev(struct device *dev)
534 devm_pci_release_host_bridge_dev(dev);
535 kfree(to_pci_host_bridge(dev));
538 struct pci_host_bridge *pci_alloc_host_bridge(size_t priv)
540 struct pci_host_bridge *bridge;
542 bridge = kzalloc(sizeof(*bridge) + priv, GFP_KERNEL);
546 INIT_LIST_HEAD(&bridge->windows);
547 bridge->dev.release = pci_release_host_bridge_dev;
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.
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;
563 EXPORT_SYMBOL(pci_alloc_host_bridge);
565 struct pci_host_bridge *devm_pci_alloc_host_bridge(struct device *dev,
568 struct pci_host_bridge *bridge;
570 bridge = devm_kzalloc(dev, sizeof(*bridge) + priv, GFP_KERNEL);
574 INIT_LIST_HEAD(&bridge->windows);
575 bridge->dev.release = devm_pci_release_host_bridge_dev;
579 EXPORT_SYMBOL(devm_pci_alloc_host_bridge);
581 void pci_free_host_bridge(struct pci_host_bridge *bridge)
583 pci_free_resource_list(&bridge->windows);
587 EXPORT_SYMBOL(pci_free_host_bridge);
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 */
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 */
627 void pcie_update_link_speed(struct pci_bus *bus, u16 linksta)
629 bus->cur_bus_speed = pcie_link_speed[linksta & PCI_EXP_LNKSTA_CLS];
631 EXPORT_SYMBOL_GPL(pcie_update_link_speed);
633 static unsigned char agp_speeds[] = {
641 static enum pci_bus_speed agp_speed(int agp3, int agpstat)
647 else if (agpstat & 2)
649 else if (agpstat & 1)
661 return agp_speeds[index];
664 static void pci_set_bus_speed(struct pci_bus *bus)
666 struct pci_dev *bridge = bus->self;
669 pos = pci_find_capability(bridge, PCI_CAP_ID_AGP);
671 pos = pci_find_capability(bridge, PCI_CAP_ID_AGP3);
675 pci_read_config_dword(bridge, pos + PCI_AGP_STATUS, &agpstat);
676 bus->max_bus_speed = agp_speed(agpstat & 8, agpstat & 7);
678 pci_read_config_dword(bridge, pos + PCI_AGP_COMMAND, &agpcmd);
679 bus->cur_bus_speed = agp_speed(agpstat & 8, agpcmd & 7);
682 pos = pci_find_capability(bridge, PCI_CAP_ID_PCIX);
685 enum pci_bus_speed max;
687 pci_read_config_word(bridge, pos + PCI_X_BRIDGE_SSTATUS,
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;
698 max = PCI_SPEED_133MHz_PCIX;
700 max = PCI_SPEED_66MHz_PCIX;
703 bus->max_bus_speed = max;
704 bus->cur_bus_speed = pcix_bus_speed[
705 (status & PCI_X_SSTATUS_FREQ) >> 6];
710 if (pci_is_pcie(bridge)) {
714 pcie_capability_read_dword(bridge, PCI_EXP_LNKCAP, &linkcap);
715 bus->max_bus_speed = pcie_link_speed[linkcap & PCI_EXP_LNKCAP_SLS];
717 pcie_capability_read_word(bridge, PCI_EXP_LNKSTA, &linksta);
718 pcie_update_link_speed(bus, linksta);
722 static struct irq_domain *pci_host_bridge_msi_domain(struct pci_bus *bus)
724 struct irq_domain *d;
727 * Any firmware interface that can resolve the msi_domain
728 * should be called from here.
730 d = pci_host_bridge_of_msi_domain(bus);
732 d = pci_host_bridge_acpi_msi_domain(bus);
734 #ifdef CONFIG_PCI_MSI_IRQ_DOMAIN
736 * If no IRQ domain was found via the OF tree, try looking it up
737 * directly through the fwnode_handle.
740 struct fwnode_handle *fwnode = pci_root_bus_fwnode(bus);
743 d = irq_find_matching_fwnode(fwnode,
751 static void pci_set_bus_msi_domain(struct pci_bus *bus)
753 struct irq_domain *d;
757 * The bus can be a root bus, a subordinate bus, or a virtual bus
758 * created by an SR-IOV device. Walk up to the first bridge device
759 * found or derive the domain from the host bridge.
761 for (b = bus, d = NULL; !d && !pci_is_root_bus(b); b = b->parent) {
763 d = dev_get_msi_domain(&b->self->dev);
767 d = pci_host_bridge_msi_domain(b);
769 dev_set_msi_domain(&bus->dev, d);
772 static int pci_register_host_bridge(struct pci_host_bridge *bridge)
774 struct device *parent = bridge->dev.parent;
775 struct resource_entry *window, *n;
776 struct pci_bus *bus, *b;
777 resource_size_t offset;
778 LIST_HEAD(resources);
779 struct resource *res;
784 bus = pci_alloc_bus(NULL);
790 /* Temporarily move resources off the list */
791 list_splice_init(&bridge->windows, &resources);
792 bus->sysdata = bridge->sysdata;
793 bus->msi = bridge->msi;
794 bus->ops = bridge->ops;
795 bus->number = bus->busn_res.start = bridge->busnr;
796 #ifdef CONFIG_PCI_DOMAINS_GENERIC
797 bus->domain_nr = pci_bus_find_domain_nr(bus, parent);
800 b = pci_find_bus(pci_domain_nr(bus), bridge->busnr);
802 /* Ignore it if we already got here via a different bridge */
803 dev_dbg(&b->dev, "bus already known\n");
808 dev_set_name(&bridge->dev, "pci%04x:%02x", pci_domain_nr(bus),
811 err = pcibios_root_bridge_prepare(bridge);
815 err = device_register(&bridge->dev);
817 put_device(&bridge->dev);
819 bus->bridge = get_device(&bridge->dev);
820 device_enable_async_suspend(bus->bridge);
821 pci_set_bus_of_node(bus);
822 pci_set_bus_msi_domain(bus);
825 set_dev_node(bus->bridge, pcibus_to_node(bus));
827 bus->dev.class = &pcibus_class;
828 bus->dev.parent = bus->bridge;
830 dev_set_name(&bus->dev, "%04x:%02x", pci_domain_nr(bus), bus->number);
831 name = dev_name(&bus->dev);
833 err = device_register(&bus->dev);
837 pcibios_add_bus(bus);
839 /* Create legacy_io and legacy_mem files for this bus */
840 pci_create_legacy_files(bus);
843 dev_info(parent, "PCI host bridge to bus %s\n", name);
845 pr_info("PCI host bridge to bus %s\n", name);
847 /* Add initial resources to the bus */
848 resource_list_for_each_entry_safe(window, n, &resources) {
849 list_move_tail(&window->node, &bridge->windows);
850 offset = window->offset;
853 if (res->flags & IORESOURCE_BUS)
854 pci_bus_insert_busn_res(bus, bus->number, res->end);
856 pci_bus_add_resource(bus, res, 0);
859 if (resource_type(res) == IORESOURCE_IO)
860 fmt = " (bus address [%#06llx-%#06llx])";
862 fmt = " (bus address [%#010llx-%#010llx])";
864 snprintf(addr, sizeof(addr), fmt,
865 (unsigned long long)(res->start - offset),
866 (unsigned long long)(res->end - offset));
870 dev_info(&bus->dev, "root bus resource %pR%s\n", res, addr);
873 down_write(&pci_bus_sem);
874 list_add_tail(&bus->node, &pci_root_buses);
875 up_write(&pci_bus_sem);
880 put_device(&bridge->dev);
881 device_unregister(&bridge->dev);
888 static bool pci_bridge_child_ext_cfg_accessible(struct pci_dev *bridge)
894 * If extended config space isn't accessible on a bridge's primary
895 * bus, we certainly can't access it on the secondary bus.
897 if (bridge->bus->bus_flags & PCI_BUS_FLAGS_NO_EXTCFG)
901 * PCIe Root Ports and switch ports are PCIe on both sides, so if
902 * extended config space is accessible on the primary, it's also
903 * accessible on the secondary.
905 if (pci_is_pcie(bridge) &&
906 (pci_pcie_type(bridge) == PCI_EXP_TYPE_ROOT_PORT ||
907 pci_pcie_type(bridge) == PCI_EXP_TYPE_UPSTREAM ||
908 pci_pcie_type(bridge) == PCI_EXP_TYPE_DOWNSTREAM))
912 * For the other bridge types:
913 * - PCI-to-PCI bridges
914 * - PCIe-to-PCI/PCI-X forward bridges
915 * - PCI/PCI-X-to-PCIe reverse bridges
916 * extended config space on the secondary side is only accessible
917 * if the bridge supports PCI-X Mode 2.
919 pos = pci_find_capability(bridge, PCI_CAP_ID_PCIX);
923 pci_read_config_dword(bridge, pos + PCI_X_STATUS, &status);
924 return status & (PCI_X_STATUS_266MHZ | PCI_X_STATUS_533MHZ);
927 static struct pci_bus *pci_alloc_child_bus(struct pci_bus *parent,
928 struct pci_dev *bridge, int busnr)
930 struct pci_bus *child;
934 /* Allocate a new bus and inherit stuff from the parent */
935 child = pci_alloc_bus(parent);
939 child->parent = parent;
940 child->ops = parent->ops;
941 child->msi = parent->msi;
942 child->sysdata = parent->sysdata;
943 child->bus_flags = parent->bus_flags;
946 * Initialize some portions of the bus device, but don't register
947 * it now as the parent is not properly set up yet.
949 child->dev.class = &pcibus_class;
950 dev_set_name(&child->dev, "%04x:%02x", pci_domain_nr(child), busnr);
952 /* Set up the primary, secondary and subordinate bus numbers */
953 child->number = child->busn_res.start = busnr;
954 child->primary = parent->busn_res.start;
955 child->busn_res.end = 0xff;
958 child->dev.parent = parent->bridge;
962 child->self = bridge;
963 child->bridge = get_device(&bridge->dev);
964 child->dev.parent = child->bridge;
965 pci_set_bus_of_node(child);
966 pci_set_bus_speed(child);
969 * Check whether extended config space is accessible on the child
970 * bus. Note that we currently assume it is always accessible on
973 if (!pci_bridge_child_ext_cfg_accessible(bridge)) {
974 child->bus_flags |= PCI_BUS_FLAGS_NO_EXTCFG;
975 pci_info(child, "extended config space not accessible\n");
978 /* Set up default resource pointers and names */
979 for (i = 0; i < PCI_BRIDGE_RESOURCE_NUM; i++) {
980 child->resource[i] = &bridge->resource[PCI_BRIDGE_RESOURCES+i];
981 child->resource[i]->name = child->name;
983 bridge->subordinate = child;
986 pci_set_bus_msi_domain(child);
987 ret = device_register(&child->dev);
990 pcibios_add_bus(child);
992 if (child->ops->add_bus) {
993 ret = child->ops->add_bus(child);
994 if (WARN_ON(ret < 0))
995 dev_err(&child->dev, "failed to add bus: %d\n", ret);
998 /* Create legacy_io and legacy_mem files for this bus */
999 pci_create_legacy_files(child);
1004 struct pci_bus *pci_add_new_bus(struct pci_bus *parent, struct pci_dev *dev,
1007 struct pci_bus *child;
1009 child = pci_alloc_child_bus(parent, dev, busnr);
1011 down_write(&pci_bus_sem);
1012 list_add_tail(&child->node, &parent->children);
1013 up_write(&pci_bus_sem);
1017 EXPORT_SYMBOL(pci_add_new_bus);
1019 static void pci_enable_crs(struct pci_dev *pdev)
1023 /* Enable CRS Software Visibility if supported */
1024 pcie_capability_read_word(pdev, PCI_EXP_RTCAP, &root_cap);
1025 if (root_cap & PCI_EXP_RTCAP_CRSVIS)
1026 pcie_capability_set_word(pdev, PCI_EXP_RTCTL,
1027 PCI_EXP_RTCTL_CRSSVE);
1030 static unsigned int pci_scan_child_bus_extend(struct pci_bus *bus,
1031 unsigned int available_buses);
1034 * pci_scan_bridge_extend() - Scan buses behind a bridge
1035 * @bus: Parent bus the bridge is on
1036 * @dev: Bridge itself
1037 * @max: Starting subordinate number of buses behind this bridge
1038 * @available_buses: Total number of buses available for this bridge and
1039 * the devices below. After the minimal bus space has
1040 * been allocated the remaining buses will be
1041 * distributed equally between hotplug-capable bridges.
1042 * @pass: Either %0 (scan already configured bridges) or %1 (scan bridges
1043 * that need to be reconfigured.
1045 * If it's a bridge, configure it and scan the bus behind it.
1046 * For CardBus bridges, we don't scan behind as the devices will
1047 * be handled by the bridge driver itself.
1049 * We need to process bridges in two passes -- first we scan those
1050 * already configured by the BIOS and after we are done with all of
1051 * them, we proceed to assigning numbers to the remaining buses in
1052 * order to avoid overlaps between old and new bus numbers.
1054 * Return: New subordinate number covering all buses behind this bridge.
1056 static int pci_scan_bridge_extend(struct pci_bus *bus, struct pci_dev *dev,
1057 int max, unsigned int available_buses,
1060 struct pci_bus *child;
1061 int is_cardbus = (dev->hdr_type == PCI_HEADER_TYPE_CARDBUS);
1062 u32 buses, i, j = 0;
1064 u8 primary, secondary, subordinate;
1068 * Make sure the bridge is powered on to be able to access config
1069 * space of devices below it.
1071 pm_runtime_get_sync(&dev->dev);
1073 pci_read_config_dword(dev, PCI_PRIMARY_BUS, &buses);
1074 primary = buses & 0xFF;
1075 secondary = (buses >> 8) & 0xFF;
1076 subordinate = (buses >> 16) & 0xFF;
1078 pci_dbg(dev, "scanning [bus %02x-%02x] behind bridge, pass %d\n",
1079 secondary, subordinate, pass);
1081 if (!primary && (primary != bus->number) && secondary && subordinate) {
1082 pci_warn(dev, "Primary bus is hard wired to 0\n");
1083 primary = bus->number;
1086 /* Check if setup is sensible at all */
1088 (primary != bus->number || secondary <= bus->number ||
1089 secondary > subordinate)) {
1090 pci_info(dev, "bridge configuration invalid ([bus %02x-%02x]), reconfiguring\n",
1091 secondary, subordinate);
1096 * Disable Master-Abort Mode during probing to avoid reporting of
1097 * bus errors in some architectures.
1099 pci_read_config_word(dev, PCI_BRIDGE_CONTROL, &bctl);
1100 pci_write_config_word(dev, PCI_BRIDGE_CONTROL,
1101 bctl & ~PCI_BRIDGE_CTL_MASTER_ABORT);
1103 pci_enable_crs(dev);
1105 if ((secondary || subordinate) && !pcibios_assign_all_busses() &&
1106 !is_cardbus && !broken) {
1110 * Bus already configured by firmware, process it in the
1111 * first pass and just note the configuration.
1117 * The bus might already exist for two reasons: Either we
1118 * are rescanning the bus or the bus is reachable through
1119 * more than one bridge. The second case can happen with
1120 * the i450NX chipset.
1122 child = pci_find_bus(pci_domain_nr(bus), secondary);
1124 child = pci_add_new_bus(bus, dev, secondary);
1127 child->primary = primary;
1128 pci_bus_insert_busn_res(child, secondary, subordinate);
1129 child->bridge_ctl = bctl;
1132 cmax = pci_scan_child_bus(child);
1133 if (cmax > subordinate)
1134 pci_warn(dev, "bridge has subordinate %02x but max busn %02x\n",
1137 /* Subordinate should equal child->busn_res.end */
1138 if (subordinate > max)
1143 * We need to assign a number to this bus which we always
1144 * do in the second pass.
1147 if (pcibios_assign_all_busses() || broken || is_cardbus)
1150 * Temporarily disable forwarding of the
1151 * configuration cycles on all bridges in
1152 * this bus segment to avoid possible
1153 * conflicts in the second pass between two
1154 * bridges programmed with overlapping bus
1157 pci_write_config_dword(dev, PCI_PRIMARY_BUS,
1163 pci_write_config_word(dev, PCI_STATUS, 0xffff);
1166 * Prevent assigning a bus number that already exists.
1167 * This can happen when a bridge is hot-plugged, so in this
1168 * case we only re-scan this bus.
1170 child = pci_find_bus(pci_domain_nr(bus), max+1);
1172 child = pci_add_new_bus(bus, dev, max+1);
1175 pci_bus_insert_busn_res(child, max+1,
1179 if (available_buses)
1182 buses = (buses & 0xff000000)
1183 | ((unsigned int)(child->primary) << 0)
1184 | ((unsigned int)(child->busn_res.start) << 8)
1185 | ((unsigned int)(child->busn_res.end) << 16);
1188 * yenta.c forces a secondary latency timer of 176.
1189 * Copy that behaviour here.
1192 buses &= ~0xff000000;
1193 buses |= CARDBUS_LATENCY_TIMER << 24;
1196 /* We need to blast all three values with a single write */
1197 pci_write_config_dword(dev, PCI_PRIMARY_BUS, buses);
1200 child->bridge_ctl = bctl;
1201 max = pci_scan_child_bus_extend(child, available_buses);
1205 * For CardBus bridges, we leave 4 bus numbers as
1206 * cards with a PCI-to-PCI bridge can be inserted
1209 for (i = 0; i < CARDBUS_RESERVE_BUSNR; i++) {
1210 struct pci_bus *parent = bus;
1211 if (pci_find_bus(pci_domain_nr(bus),
1214 while (parent->parent) {
1215 if ((!pcibios_assign_all_busses()) &&
1216 (parent->busn_res.end > max) &&
1217 (parent->busn_res.end <= max+i)) {
1220 parent = parent->parent;
1225 * Often, there are two CardBus
1226 * bridges -- try to leave one
1227 * valid bus number for each one.
1236 /* Set subordinate bus number to its real value */
1237 pci_bus_update_busn_res_end(child, max);
1238 pci_write_config_byte(dev, PCI_SUBORDINATE_BUS, max);
1241 sprintf(child->name,
1242 (is_cardbus ? "PCI CardBus %04x:%02x" : "PCI Bus %04x:%02x"),
1243 pci_domain_nr(bus), child->number);
1245 /* Check that all devices are accessible */
1246 while (bus->parent) {
1247 if ((child->busn_res.end > bus->busn_res.end) ||
1248 (child->number > bus->busn_res.end) ||
1249 (child->number < bus->number) ||
1250 (child->busn_res.end < bus->number)) {
1251 dev_info(&dev->dev, "devices behind bridge are unusable because %pR cannot be assigned for them\n",
1259 pci_write_config_word(dev, PCI_BRIDGE_CONTROL, bctl);
1261 pm_runtime_put(&dev->dev);
1267 * pci_scan_bridge() - Scan buses behind a bridge
1268 * @bus: Parent bus the bridge is on
1269 * @dev: Bridge itself
1270 * @max: Starting subordinate number of buses behind this bridge
1271 * @pass: Either %0 (scan already configured bridges) or %1 (scan bridges
1272 * that need to be reconfigured.
1274 * If it's a bridge, configure it and scan the bus behind it.
1275 * For CardBus bridges, we don't scan behind as the devices will
1276 * be handled by the bridge driver itself.
1278 * We need to process bridges in two passes -- first we scan those
1279 * already configured by the BIOS and after we are done with all of
1280 * them, we proceed to assigning numbers to the remaining buses in
1281 * order to avoid overlaps between old and new bus numbers.
1283 * Return: New subordinate number covering all buses behind this bridge.
1285 int pci_scan_bridge(struct pci_bus *bus, struct pci_dev *dev, int max, int pass)
1287 return pci_scan_bridge_extend(bus, dev, max, 0, pass);
1289 EXPORT_SYMBOL(pci_scan_bridge);
1292 * Read interrupt line and base address registers.
1293 * The architecture-dependent code can tweak these, of course.
1295 static void pci_read_irq(struct pci_dev *dev)
1299 /* VFs are not allowed to use INTx, so skip the config reads */
1300 if (dev->is_virtfn) {
1306 pci_read_config_byte(dev, PCI_INTERRUPT_PIN, &irq);
1309 pci_read_config_byte(dev, PCI_INTERRUPT_LINE, &irq);
1313 void set_pcie_port_type(struct pci_dev *pdev)
1318 struct pci_dev *parent;
1320 pos = pci_find_capability(pdev, PCI_CAP_ID_EXP);
1324 pdev->pcie_cap = pos;
1325 pci_read_config_word(pdev, pos + PCI_EXP_FLAGS, ®16);
1326 pdev->pcie_flags_reg = reg16;
1327 pci_read_config_word(pdev, pos + PCI_EXP_DEVCAP, ®16);
1328 pdev->pcie_mpss = reg16 & PCI_EXP_DEVCAP_PAYLOAD;
1331 * A Root Port or a PCI-to-PCIe bridge is always the upstream end
1332 * of a Link. No PCIe component has two Links. Two Links are
1333 * connected by a Switch that has a Port on each Link and internal
1334 * logic to connect the two Ports.
1336 type = pci_pcie_type(pdev);
1337 if (type == PCI_EXP_TYPE_ROOT_PORT ||
1338 type == PCI_EXP_TYPE_PCIE_BRIDGE)
1339 pdev->has_secondary_link = 1;
1340 else if (type == PCI_EXP_TYPE_UPSTREAM ||
1341 type == PCI_EXP_TYPE_DOWNSTREAM) {
1342 parent = pci_upstream_bridge(pdev);
1345 * Usually there's an upstream device (Root Port or Switch
1346 * Downstream Port), but we can't assume one exists.
1348 if (parent && !parent->has_secondary_link)
1349 pdev->has_secondary_link = 1;
1353 void set_pcie_hotplug_bridge(struct pci_dev *pdev)
1357 pcie_capability_read_dword(pdev, PCI_EXP_SLTCAP, ®32);
1358 if (reg32 & PCI_EXP_SLTCAP_HPC)
1359 pdev->is_hotplug_bridge = 1;
1362 static void set_pcie_thunderbolt(struct pci_dev *dev)
1367 while ((vsec = pci_find_next_ext_capability(dev, vsec,
1368 PCI_EXT_CAP_ID_VNDR))) {
1369 pci_read_config_dword(dev, vsec + PCI_VNDR_HEADER, &header);
1371 /* Is the device part of a Thunderbolt controller? */
1372 if (dev->vendor == PCI_VENDOR_ID_INTEL &&
1373 PCI_VNDR_HEADER_ID(header) == PCI_VSEC_ID_INTEL_TBT) {
1374 dev->is_thunderbolt = 1;
1381 * pci_ext_cfg_is_aliased - Is ext config space just an alias of std config?
1384 * PCI Express to PCI/PCI-X Bridge Specification, rev 1.0, 4.1.4 says that
1385 * when forwarding a type1 configuration request the bridge must check that
1386 * the extended register address field is zero. The bridge is not permitted
1387 * to forward the transactions and must handle it as an Unsupported Request.
1388 * Some bridges do not follow this rule and simply drop the extended register
1389 * bits, resulting in the standard config space being aliased, every 256
1390 * bytes across the entire configuration space. Test for this condition by
1391 * comparing the first dword of each potential alias to the vendor/device ID.
1393 * ASM1083/1085 PCIe-to-PCI Reversible Bridge (1b21:1080, rev 01 & 03)
1394 * AMD/ATI SBx00 PCI to PCI Bridge (1002:4384, rev 40)
1396 static bool pci_ext_cfg_is_aliased(struct pci_dev *dev)
1398 #ifdef CONFIG_PCI_QUIRKS
1402 pci_read_config_dword(dev, PCI_VENDOR_ID, &header);
1404 for (pos = PCI_CFG_SPACE_SIZE;
1405 pos < PCI_CFG_SPACE_EXP_SIZE; pos += PCI_CFG_SPACE_SIZE) {
1406 if (pci_read_config_dword(dev, pos, &tmp) != PCIBIOS_SUCCESSFUL
1418 * pci_cfg_space_size - Get the configuration space size of the PCI device
1421 * Regular PCI devices have 256 bytes, but PCI-X 2 and PCI Express devices
1422 * have 4096 bytes. Even if the device is capable, that doesn't mean we can
1423 * access it. Maybe we don't have a way to generate extended config space
1424 * accesses, or the device is behind a reverse Express bridge. So we try
1425 * reading the dword at 0x100 which must either be 0 or a valid extended
1426 * capability header.
1428 static int pci_cfg_space_size_ext(struct pci_dev *dev)
1431 int pos = PCI_CFG_SPACE_SIZE;
1433 if (pci_read_config_dword(dev, pos, &status) != PCIBIOS_SUCCESSFUL)
1434 return PCI_CFG_SPACE_SIZE;
1435 if (status == 0xffffffff || pci_ext_cfg_is_aliased(dev))
1436 return PCI_CFG_SPACE_SIZE;
1438 return PCI_CFG_SPACE_EXP_SIZE;
1441 int pci_cfg_space_size(struct pci_dev *dev)
1447 if (dev->bus->bus_flags & PCI_BUS_FLAGS_NO_EXTCFG)
1448 return PCI_CFG_SPACE_SIZE;
1450 class = dev->class >> 8;
1451 if (class == PCI_CLASS_BRIDGE_HOST)
1452 return pci_cfg_space_size_ext(dev);
1454 if (pci_is_pcie(dev))
1455 return pci_cfg_space_size_ext(dev);
1457 pos = pci_find_capability(dev, PCI_CAP_ID_PCIX);
1459 return PCI_CFG_SPACE_SIZE;
1461 pci_read_config_dword(dev, pos + PCI_X_STATUS, &status);
1462 if (status & (PCI_X_STATUS_266MHZ | PCI_X_STATUS_533MHZ))
1463 return pci_cfg_space_size_ext(dev);
1465 return PCI_CFG_SPACE_SIZE;
1468 static u32 pci_class(struct pci_dev *dev)
1472 #ifdef CONFIG_PCI_IOV
1474 return dev->physfn->sriov->class;
1476 pci_read_config_dword(dev, PCI_CLASS_REVISION, &class);
1480 static void pci_subsystem_ids(struct pci_dev *dev, u16 *vendor, u16 *device)
1482 #ifdef CONFIG_PCI_IOV
1483 if (dev->is_virtfn) {
1484 *vendor = dev->physfn->sriov->subsystem_vendor;
1485 *device = dev->physfn->sriov->subsystem_device;
1489 pci_read_config_word(dev, PCI_SUBSYSTEM_VENDOR_ID, vendor);
1490 pci_read_config_word(dev, PCI_SUBSYSTEM_ID, device);
1493 static u8 pci_hdr_type(struct pci_dev *dev)
1497 #ifdef CONFIG_PCI_IOV
1499 return dev->physfn->sriov->hdr_type;
1501 pci_read_config_byte(dev, PCI_HEADER_TYPE, &hdr_type);
1505 #define LEGACY_IO_RESOURCE (IORESOURCE_IO | IORESOURCE_PCI_FIXED)
1507 static void pci_msi_setup_pci_dev(struct pci_dev *dev)
1510 * Disable the MSI hardware to avoid screaming interrupts
1511 * during boot. This is the power on reset default so
1512 * usually this should be a noop.
1514 dev->msi_cap = pci_find_capability(dev, PCI_CAP_ID_MSI);
1516 pci_msi_set_enable(dev, 0);
1518 dev->msix_cap = pci_find_capability(dev, PCI_CAP_ID_MSIX);
1520 pci_msix_clear_and_set_ctrl(dev, PCI_MSIX_FLAGS_ENABLE, 0);
1524 * pci_intx_mask_broken - Test PCI_COMMAND_INTX_DISABLE writability
1527 * Test whether PCI_COMMAND_INTX_DISABLE is writable for @dev. Check this
1528 * at enumeration-time to avoid modifying PCI_COMMAND at run-time.
1530 static int pci_intx_mask_broken(struct pci_dev *dev)
1532 u16 orig, toggle, new;
1534 pci_read_config_word(dev, PCI_COMMAND, &orig);
1535 toggle = orig ^ PCI_COMMAND_INTX_DISABLE;
1536 pci_write_config_word(dev, PCI_COMMAND, toggle);
1537 pci_read_config_word(dev, PCI_COMMAND, &new);
1539 pci_write_config_word(dev, PCI_COMMAND, orig);
1542 * PCI_COMMAND_INTX_DISABLE was reserved and read-only prior to PCI
1543 * r2.3, so strictly speaking, a device is not *broken* if it's not
1544 * writable. But we'll live with the misnomer for now.
1551 static void early_dump_pci_device(struct pci_dev *pdev)
1556 pci_info(pdev, "config space:\n");
1558 for (i = 0; i < 256; i += 4)
1559 pci_read_config_dword(pdev, i, &value[i / 4]);
1561 print_hex_dump(KERN_INFO, "", DUMP_PREFIX_OFFSET, 16, 1,
1566 * pci_setup_device - Fill in class and map information of a device
1567 * @dev: the device structure to fill
1569 * Initialize the device structure with information about the device's
1570 * vendor,class,memory and IO-space addresses, IRQ lines etc.
1571 * Called at initialisation of the PCI subsystem and by CardBus services.
1572 * Returns 0 on success and negative if unknown type of device (not normal,
1573 * bridge or CardBus).
1575 int pci_setup_device(struct pci_dev *dev)
1581 struct pci_bus_region region;
1582 struct resource *res;
1584 hdr_type = pci_hdr_type(dev);
1586 dev->sysdata = dev->bus->sysdata;
1587 dev->dev.parent = dev->bus->bridge;
1588 dev->dev.bus = &pci_bus_type;
1589 dev->hdr_type = hdr_type & 0x7f;
1590 dev->multifunction = !!(hdr_type & 0x80);
1591 dev->error_state = pci_channel_io_normal;
1592 set_pcie_port_type(dev);
1594 pci_dev_assign_slot(dev);
1597 * Assume 32-bit PCI; let 64-bit PCI cards (which are far rarer)
1598 * set this higher, assuming the system even supports it.
1600 dev->dma_mask = 0xffffffff;
1602 dev_set_name(&dev->dev, "%04x:%02x:%02x.%d", pci_domain_nr(dev->bus),
1603 dev->bus->number, PCI_SLOT(dev->devfn),
1604 PCI_FUNC(dev->devfn));
1606 class = pci_class(dev);
1608 dev->revision = class & 0xff;
1609 dev->class = class >> 8; /* upper 3 bytes */
1611 pci_printk(KERN_DEBUG, dev, "[%04x:%04x] type %02x class %#08x\n",
1612 dev->vendor, dev->device, dev->hdr_type, dev->class);
1615 early_dump_pci_device(dev);
1617 /* Need to have dev->class ready */
1618 dev->cfg_size = pci_cfg_space_size(dev);
1620 /* Need to have dev->cfg_size ready */
1621 set_pcie_thunderbolt(dev);
1623 /* "Unknown power state" */
1624 dev->current_state = PCI_UNKNOWN;
1626 /* Early fixups, before probing the BARs */
1627 pci_fixup_device(pci_fixup_early, dev);
1629 /* Device class may be changed after fixup */
1630 class = dev->class >> 8;
1632 if (dev->non_compliant_bars) {
1633 pci_read_config_word(dev, PCI_COMMAND, &cmd);
1634 if (cmd & (PCI_COMMAND_IO | PCI_COMMAND_MEMORY)) {
1635 pci_info(dev, "device has non-compliant BARs; disabling IO/MEM decoding\n");
1636 cmd &= ~PCI_COMMAND_IO;
1637 cmd &= ~PCI_COMMAND_MEMORY;
1638 pci_write_config_word(dev, PCI_COMMAND, cmd);
1642 dev->broken_intx_masking = pci_intx_mask_broken(dev);
1644 switch (dev->hdr_type) { /* header type */
1645 case PCI_HEADER_TYPE_NORMAL: /* standard header */
1646 if (class == PCI_CLASS_BRIDGE_PCI)
1649 pci_read_bases(dev, 6, PCI_ROM_ADDRESS);
1651 pci_subsystem_ids(dev, &dev->subsystem_vendor, &dev->subsystem_device);
1654 * Do the ugly legacy mode stuff here rather than broken chip
1655 * quirk code. Legacy mode ATA controllers have fixed
1656 * addresses. These are not always echoed in BAR0-3, and
1657 * BAR0-3 in a few cases contain junk!
1659 if (class == PCI_CLASS_STORAGE_IDE) {
1661 pci_read_config_byte(dev, PCI_CLASS_PROG, &progif);
1662 if ((progif & 1) == 0) {
1663 region.start = 0x1F0;
1665 res = &dev->resource[0];
1666 res->flags = LEGACY_IO_RESOURCE;
1667 pcibios_bus_to_resource(dev->bus, res, ®ion);
1668 pci_info(dev, "legacy IDE quirk: reg 0x10: %pR\n",
1670 region.start = 0x3F6;
1672 res = &dev->resource[1];
1673 res->flags = LEGACY_IO_RESOURCE;
1674 pcibios_bus_to_resource(dev->bus, res, ®ion);
1675 pci_info(dev, "legacy IDE quirk: reg 0x14: %pR\n",
1678 if ((progif & 4) == 0) {
1679 region.start = 0x170;
1681 res = &dev->resource[2];
1682 res->flags = LEGACY_IO_RESOURCE;
1683 pcibios_bus_to_resource(dev->bus, res, ®ion);
1684 pci_info(dev, "legacy IDE quirk: reg 0x18: %pR\n",
1686 region.start = 0x376;
1688 res = &dev->resource[3];
1689 res->flags = LEGACY_IO_RESOURCE;
1690 pcibios_bus_to_resource(dev->bus, res, ®ion);
1691 pci_info(dev, "legacy IDE quirk: reg 0x1c: %pR\n",
1697 case PCI_HEADER_TYPE_BRIDGE: /* bridge header */
1698 if (class != PCI_CLASS_BRIDGE_PCI)
1702 * The PCI-to-PCI bridge spec requires that subtractive
1703 * decoding (i.e. transparent) bridge must have programming
1704 * interface code of 0x01.
1707 dev->transparent = ((dev->class & 0xff) == 1);
1708 pci_read_bases(dev, 2, PCI_ROM_ADDRESS1);
1709 set_pcie_hotplug_bridge(dev);
1710 pos = pci_find_capability(dev, PCI_CAP_ID_SSVID);
1712 pci_read_config_word(dev, pos + PCI_SSVID_VENDOR_ID, &dev->subsystem_vendor);
1713 pci_read_config_word(dev, pos + PCI_SSVID_DEVICE_ID, &dev->subsystem_device);
1717 case PCI_HEADER_TYPE_CARDBUS: /* CardBus bridge header */
1718 if (class != PCI_CLASS_BRIDGE_CARDBUS)
1721 pci_read_bases(dev, 1, 0);
1722 pci_read_config_word(dev, PCI_CB_SUBSYSTEM_VENDOR_ID, &dev->subsystem_vendor);
1723 pci_read_config_word(dev, PCI_CB_SUBSYSTEM_ID, &dev->subsystem_device);
1726 default: /* unknown header */
1727 pci_err(dev, "unknown header type %02x, ignoring device\n",
1732 pci_err(dev, "ignoring class %#08x (doesn't match header type %02x)\n",
1733 dev->class, dev->hdr_type);
1734 dev->class = PCI_CLASS_NOT_DEFINED << 8;
1737 /* We found a fine healthy device, go go go... */
1741 static void pci_configure_mps(struct pci_dev *dev)
1743 struct pci_dev *bridge = pci_upstream_bridge(dev);
1744 int mps, mpss, p_mps, rc;
1746 if (!pci_is_pcie(dev) || !bridge || !pci_is_pcie(bridge))
1749 /* MPS and MRRS fields are of type 'RsvdP' for VFs, short-circuit out */
1753 mps = pcie_get_mps(dev);
1754 p_mps = pcie_get_mps(bridge);
1759 if (pcie_bus_config == PCIE_BUS_TUNE_OFF) {
1760 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",
1761 mps, pci_name(bridge), p_mps);
1766 * Fancier MPS configuration is done later by
1767 * pcie_bus_configure_settings()
1769 if (pcie_bus_config != PCIE_BUS_DEFAULT)
1772 mpss = 128 << dev->pcie_mpss;
1773 if (mpss < p_mps && pci_pcie_type(bridge) == PCI_EXP_TYPE_ROOT_PORT) {
1774 pcie_set_mps(bridge, mpss);
1775 pci_info(dev, "Upstream bridge's Max Payload Size set to %d (was %d, max %d)\n",
1776 mpss, p_mps, 128 << bridge->pcie_mpss);
1777 p_mps = pcie_get_mps(bridge);
1780 rc = pcie_set_mps(dev, p_mps);
1782 pci_warn(dev, "can't set Max Payload Size to %d; if necessary, use \"pci=pcie_bus_safe\" and report a bug\n",
1787 pci_info(dev, "Max Payload Size set to %d (was %d, max %d)\n",
1791 static struct hpp_type0 pci_default_type0 = {
1793 .cache_line_size = 8,
1794 .latency_timer = 0x40,
1799 static void program_hpp_type0(struct pci_dev *dev, struct hpp_type0 *hpp)
1801 u16 pci_cmd, pci_bctl;
1804 hpp = &pci_default_type0;
1806 if (hpp->revision > 1) {
1807 pci_warn(dev, "PCI settings rev %d not supported; using defaults\n",
1809 hpp = &pci_default_type0;
1812 pci_write_config_byte(dev, PCI_CACHE_LINE_SIZE, hpp->cache_line_size);
1813 pci_write_config_byte(dev, PCI_LATENCY_TIMER, hpp->latency_timer);
1814 pci_read_config_word(dev, PCI_COMMAND, &pci_cmd);
1815 if (hpp->enable_serr)
1816 pci_cmd |= PCI_COMMAND_SERR;
1817 if (hpp->enable_perr)
1818 pci_cmd |= PCI_COMMAND_PARITY;
1819 pci_write_config_word(dev, PCI_COMMAND, pci_cmd);
1821 /* Program bridge control value */
1822 if ((dev->class >> 8) == PCI_CLASS_BRIDGE_PCI) {
1823 pci_write_config_byte(dev, PCI_SEC_LATENCY_TIMER,
1824 hpp->latency_timer);
1825 pci_read_config_word(dev, PCI_BRIDGE_CONTROL, &pci_bctl);
1826 if (hpp->enable_serr)
1827 pci_bctl |= PCI_BRIDGE_CTL_SERR;
1828 if (hpp->enable_perr)
1829 pci_bctl |= PCI_BRIDGE_CTL_PARITY;
1830 pci_write_config_word(dev, PCI_BRIDGE_CONTROL, pci_bctl);
1834 static void program_hpp_type1(struct pci_dev *dev, struct hpp_type1 *hpp)
1841 pos = pci_find_capability(dev, PCI_CAP_ID_PCIX);
1845 pci_warn(dev, "PCI-X settings not supported\n");
1848 static bool pcie_root_rcb_set(struct pci_dev *dev)
1850 struct pci_dev *rp = pcie_find_root_port(dev);
1856 pcie_capability_read_word(rp, PCI_EXP_LNKCTL, &lnkctl);
1857 if (lnkctl & PCI_EXP_LNKCTL_RCB)
1863 static void program_hpp_type2(struct pci_dev *dev, struct hpp_type2 *hpp)
1871 if (!pci_is_pcie(dev))
1874 if (hpp->revision > 1) {
1875 pci_warn(dev, "PCIe settings rev %d not supported\n",
1881 * Don't allow _HPX to change MPS or MRRS settings. We manage
1882 * those to make sure they're consistent with the rest of the
1885 hpp->pci_exp_devctl_and |= PCI_EXP_DEVCTL_PAYLOAD |
1886 PCI_EXP_DEVCTL_READRQ;
1887 hpp->pci_exp_devctl_or &= ~(PCI_EXP_DEVCTL_PAYLOAD |
1888 PCI_EXP_DEVCTL_READRQ);
1890 /* Initialize Device Control Register */
1891 pcie_capability_clear_and_set_word(dev, PCI_EXP_DEVCTL,
1892 ~hpp->pci_exp_devctl_and, hpp->pci_exp_devctl_or);
1894 /* Initialize Link Control Register */
1895 if (pcie_cap_has_lnkctl(dev)) {
1898 * If the Root Port supports Read Completion Boundary of
1899 * 128, set RCB to 128. Otherwise, clear it.
1901 hpp->pci_exp_lnkctl_and |= PCI_EXP_LNKCTL_RCB;
1902 hpp->pci_exp_lnkctl_or &= ~PCI_EXP_LNKCTL_RCB;
1903 if (pcie_root_rcb_set(dev))
1904 hpp->pci_exp_lnkctl_or |= PCI_EXP_LNKCTL_RCB;
1906 pcie_capability_clear_and_set_word(dev, PCI_EXP_LNKCTL,
1907 ~hpp->pci_exp_lnkctl_and, hpp->pci_exp_lnkctl_or);
1910 /* Find Advanced Error Reporting Enhanced Capability */
1911 pos = pci_find_ext_capability(dev, PCI_EXT_CAP_ID_ERR);
1915 /* Initialize Uncorrectable Error Mask Register */
1916 pci_read_config_dword(dev, pos + PCI_ERR_UNCOR_MASK, ®32);
1917 reg32 = (reg32 & hpp->unc_err_mask_and) | hpp->unc_err_mask_or;
1918 pci_write_config_dword(dev, pos + PCI_ERR_UNCOR_MASK, reg32);
1920 /* Initialize Uncorrectable Error Severity Register */
1921 pci_read_config_dword(dev, pos + PCI_ERR_UNCOR_SEVER, ®32);
1922 reg32 = (reg32 & hpp->unc_err_sever_and) | hpp->unc_err_sever_or;
1923 pci_write_config_dword(dev, pos + PCI_ERR_UNCOR_SEVER, reg32);
1925 /* Initialize Correctable Error Mask Register */
1926 pci_read_config_dword(dev, pos + PCI_ERR_COR_MASK, ®32);
1927 reg32 = (reg32 & hpp->cor_err_mask_and) | hpp->cor_err_mask_or;
1928 pci_write_config_dword(dev, pos + PCI_ERR_COR_MASK, reg32);
1930 /* Initialize Advanced Error Capabilities and Control Register */
1931 pci_read_config_dword(dev, pos + PCI_ERR_CAP, ®32);
1932 reg32 = (reg32 & hpp->adv_err_cap_and) | hpp->adv_err_cap_or;
1934 /* Don't enable ECRC generation or checking if unsupported */
1935 if (!(reg32 & PCI_ERR_CAP_ECRC_GENC))
1936 reg32 &= ~PCI_ERR_CAP_ECRC_GENE;
1937 if (!(reg32 & PCI_ERR_CAP_ECRC_CHKC))
1938 reg32 &= ~PCI_ERR_CAP_ECRC_CHKE;
1939 pci_write_config_dword(dev, pos + PCI_ERR_CAP, reg32);
1942 * FIXME: The following two registers are not supported yet.
1944 * o Secondary Uncorrectable Error Severity Register
1945 * o Secondary Uncorrectable Error Mask Register
1949 int pci_configure_extended_tags(struct pci_dev *dev, void *ign)
1951 struct pci_host_bridge *host;
1956 if (!pci_is_pcie(dev))
1959 ret = pcie_capability_read_dword(dev, PCI_EXP_DEVCAP, &cap);
1963 if (!(cap & PCI_EXP_DEVCAP_EXT_TAG))
1966 ret = pcie_capability_read_word(dev, PCI_EXP_DEVCTL, &ctl);
1970 host = pci_find_host_bridge(dev->bus);
1975 * If some device in the hierarchy doesn't handle Extended Tags
1976 * correctly, make sure they're disabled.
1978 if (host->no_ext_tags) {
1979 if (ctl & PCI_EXP_DEVCTL_EXT_TAG) {
1980 pci_info(dev, "disabling Extended Tags\n");
1981 pcie_capability_clear_word(dev, PCI_EXP_DEVCTL,
1982 PCI_EXP_DEVCTL_EXT_TAG);
1987 if (!(ctl & PCI_EXP_DEVCTL_EXT_TAG)) {
1988 pci_info(dev, "enabling Extended Tags\n");
1989 pcie_capability_set_word(dev, PCI_EXP_DEVCTL,
1990 PCI_EXP_DEVCTL_EXT_TAG);
1996 * pcie_relaxed_ordering_enabled - Probe for PCIe relaxed ordering enable
1997 * @dev: PCI device to query
1999 * Returns true if the device has enabled relaxed ordering attribute.
2001 bool pcie_relaxed_ordering_enabled(struct pci_dev *dev)
2005 pcie_capability_read_word(dev, PCI_EXP_DEVCTL, &v);
2007 return !!(v & PCI_EXP_DEVCTL_RELAX_EN);
2009 EXPORT_SYMBOL(pcie_relaxed_ordering_enabled);
2011 static void pci_configure_relaxed_ordering(struct pci_dev *dev)
2013 struct pci_dev *root;
2015 /* PCI_EXP_DEVICE_RELAX_EN is RsvdP in VFs */
2019 if (!pcie_relaxed_ordering_enabled(dev))
2023 * For now, we only deal with Relaxed Ordering issues with Root
2024 * Ports. Peer-to-Peer DMA is another can of worms.
2026 root = pci_find_pcie_root_port(dev);
2030 if (root->dev_flags & PCI_DEV_FLAGS_NO_RELAXED_ORDERING) {
2031 pcie_capability_clear_word(dev, PCI_EXP_DEVCTL,
2032 PCI_EXP_DEVCTL_RELAX_EN);
2033 pci_info(dev, "Relaxed Ordering disabled because the Root Port didn't support it\n");
2037 static void pci_configure_ltr(struct pci_dev *dev)
2039 #ifdef CONFIG_PCIEASPM
2040 struct pci_host_bridge *host = pci_find_host_bridge(dev->bus);
2042 struct pci_dev *bridge;
2044 if (!host->native_ltr)
2047 if (!pci_is_pcie(dev))
2050 pcie_capability_read_dword(dev, PCI_EXP_DEVCAP2, &cap);
2051 if (!(cap & PCI_EXP_DEVCAP2_LTR))
2055 * Software must not enable LTR in an Endpoint unless the Root
2056 * Complex and all intermediate Switches indicate support for LTR.
2057 * PCIe r3.1, sec 6.18.
2059 if (pci_pcie_type(dev) == PCI_EXP_TYPE_ROOT_PORT)
2062 bridge = pci_upstream_bridge(dev);
2063 if (bridge && bridge->ltr_path)
2068 pcie_capability_set_word(dev, PCI_EXP_DEVCTL2,
2069 PCI_EXP_DEVCTL2_LTR_EN);
2073 static void pci_configure_eetlp_prefix(struct pci_dev *dev)
2075 #ifdef CONFIG_PCI_PASID
2076 struct pci_dev *bridge;
2079 if (!pci_is_pcie(dev))
2082 pcie_capability_read_dword(dev, PCI_EXP_DEVCAP2, &cap);
2083 if (!(cap & PCI_EXP_DEVCAP2_EE_PREFIX))
2086 if (pci_pcie_type(dev) == PCI_EXP_TYPE_ROOT_PORT)
2087 dev->eetlp_prefix_path = 1;
2089 bridge = pci_upstream_bridge(dev);
2090 if (bridge && bridge->eetlp_prefix_path)
2091 dev->eetlp_prefix_path = 1;
2096 static void pci_configure_device(struct pci_dev *dev)
2098 struct hotplug_params hpp;
2101 pci_configure_mps(dev);
2102 pci_configure_extended_tags(dev, NULL);
2103 pci_configure_relaxed_ordering(dev);
2104 pci_configure_ltr(dev);
2105 pci_configure_eetlp_prefix(dev);
2107 memset(&hpp, 0, sizeof(hpp));
2108 ret = pci_get_hp_params(dev, &hpp);
2112 program_hpp_type2(dev, hpp.t2);
2113 program_hpp_type1(dev, hpp.t1);
2114 program_hpp_type0(dev, hpp.t0);
2117 static void pci_release_capabilities(struct pci_dev *dev)
2120 pci_vpd_release(dev);
2121 pci_iov_release(dev);
2122 pci_free_cap_save_buffers(dev);
2126 * pci_release_dev - Free a PCI device structure when all users of it are
2128 * @dev: device that's been disconnected
2130 * Will be called only by the device core when all users of this PCI device are
2133 static void pci_release_dev(struct device *dev)
2135 struct pci_dev *pci_dev;
2137 pci_dev = to_pci_dev(dev);
2138 pci_release_capabilities(pci_dev);
2139 pci_release_of_node(pci_dev);
2140 pcibios_release_device(pci_dev);
2141 pci_bus_put(pci_dev->bus);
2142 kfree(pci_dev->driver_override);
2143 kfree(pci_dev->dma_alias_mask);
2147 struct pci_dev *pci_alloc_dev(struct pci_bus *bus)
2149 struct pci_dev *dev;
2151 dev = kzalloc(sizeof(struct pci_dev), GFP_KERNEL);
2155 INIT_LIST_HEAD(&dev->bus_list);
2156 dev->dev.type = &pci_dev_type;
2157 dev->bus = pci_bus_get(bus);
2161 EXPORT_SYMBOL(pci_alloc_dev);
2163 static bool pci_bus_crs_vendor_id(u32 l)
2165 return (l & 0xffff) == 0x0001;
2168 static bool pci_bus_wait_crs(struct pci_bus *bus, int devfn, u32 *l,
2173 if (!pci_bus_crs_vendor_id(*l))
2174 return true; /* not a CRS completion */
2177 return false; /* CRS, but caller doesn't want to wait */
2180 * We got the reserved Vendor ID that indicates a completion with
2181 * Configuration Request Retry Status (CRS). Retry until we get a
2182 * valid Vendor ID or we time out.
2184 while (pci_bus_crs_vendor_id(*l)) {
2185 if (delay > timeout) {
2186 pr_warn("pci %04x:%02x:%02x.%d: not ready after %dms; giving up\n",
2187 pci_domain_nr(bus), bus->number,
2188 PCI_SLOT(devfn), PCI_FUNC(devfn), delay - 1);
2193 pr_info("pci %04x:%02x:%02x.%d: not ready after %dms; waiting\n",
2194 pci_domain_nr(bus), bus->number,
2195 PCI_SLOT(devfn), PCI_FUNC(devfn), delay - 1);
2200 if (pci_bus_read_config_dword(bus, devfn, PCI_VENDOR_ID, l))
2205 pr_info("pci %04x:%02x:%02x.%d: ready after %dms\n",
2206 pci_domain_nr(bus), bus->number,
2207 PCI_SLOT(devfn), PCI_FUNC(devfn), delay - 1);
2212 bool pci_bus_generic_read_dev_vendor_id(struct pci_bus *bus, int devfn, u32 *l,
2215 if (pci_bus_read_config_dword(bus, devfn, PCI_VENDOR_ID, l))
2218 /* Some broken boards return 0 or ~0 if a slot is empty: */
2219 if (*l == 0xffffffff || *l == 0x00000000 ||
2220 *l == 0x0000ffff || *l == 0xffff0000)
2223 if (pci_bus_crs_vendor_id(*l))
2224 return pci_bus_wait_crs(bus, devfn, l, timeout);
2229 bool pci_bus_read_dev_vendor_id(struct pci_bus *bus, int devfn, u32 *l,
2232 #ifdef CONFIG_PCI_QUIRKS
2233 struct pci_dev *bridge = bus->self;
2236 * Certain IDT switches have an issue where they improperly trigger
2237 * ACS Source Validation errors on completions for config reads.
2239 if (bridge && bridge->vendor == PCI_VENDOR_ID_IDT &&
2240 bridge->device == 0x80b5)
2241 return pci_idt_bus_quirk(bus, devfn, l, timeout);
2244 return pci_bus_generic_read_dev_vendor_id(bus, devfn, l, timeout);
2246 EXPORT_SYMBOL(pci_bus_read_dev_vendor_id);
2249 * Read the config data for a PCI device, sanity-check it,
2250 * and fill in the dev structure.
2252 static struct pci_dev *pci_scan_device(struct pci_bus *bus, int devfn)
2254 struct pci_dev *dev;
2257 if (!pci_bus_read_dev_vendor_id(bus, devfn, &l, 60*1000))
2260 dev = pci_alloc_dev(bus);
2265 dev->vendor = l & 0xffff;
2266 dev->device = (l >> 16) & 0xffff;
2268 pci_set_of_node(dev);
2270 if (pci_setup_device(dev)) {
2271 pci_bus_put(dev->bus);
2279 static void pcie_report_downtraining(struct pci_dev *dev)
2281 if (!pci_is_pcie(dev))
2284 /* Look from the device up to avoid downstream ports with no devices */
2285 if ((pci_pcie_type(dev) != PCI_EXP_TYPE_ENDPOINT) &&
2286 (pci_pcie_type(dev) != PCI_EXP_TYPE_LEG_END) &&
2287 (pci_pcie_type(dev) != PCI_EXP_TYPE_UPSTREAM))
2290 /* Multi-function PCIe devices share the same link/status */
2291 if (PCI_FUNC(dev->devfn) != 0 || dev->is_virtfn)
2294 /* Print link status only if the device is constrained by the fabric */
2295 __pcie_print_link_status(dev, false);
2298 static void pci_init_capabilities(struct pci_dev *dev)
2300 /* Enhanced Allocation */
2303 /* Setup MSI caps & disable MSI/MSI-X interrupts */
2304 pci_msi_setup_pci_dev(dev);
2306 /* Buffers for saving PCIe and PCI-X capabilities */
2307 pci_allocate_cap_save_buffers(dev);
2309 /* Power Management */
2312 /* Vital Product Data */
2315 /* Alternative Routing-ID Forwarding */
2316 pci_configure_ari(dev);
2318 /* Single Root I/O Virtualization */
2321 /* Address Translation Services */
2324 /* Enable ACS P2P upstream forwarding */
2325 pci_enable_acs(dev);
2327 /* Precision Time Measurement */
2330 /* Advanced Error Reporting */
2333 pcie_report_downtraining(dev);
2335 if (pci_probe_reset_function(dev) == 0)
2340 * This is the equivalent of pci_host_bridge_msi_domain() that acts on
2341 * devices. Firmware interfaces that can select the MSI domain on a
2342 * per-device basis should be called from here.
2344 static struct irq_domain *pci_dev_msi_domain(struct pci_dev *dev)
2346 struct irq_domain *d;
2349 * If a domain has been set through the pcibios_add_device()
2350 * callback, then this is the one (platform code knows best).
2352 d = dev_get_msi_domain(&dev->dev);
2357 * Let's see if we have a firmware interface able to provide
2360 d = pci_msi_get_device_domain(dev);
2367 static void pci_set_msi_domain(struct pci_dev *dev)
2369 struct irq_domain *d;
2372 * If the platform or firmware interfaces cannot supply a
2373 * device-specific MSI domain, then inherit the default domain
2374 * from the host bridge itself.
2376 d = pci_dev_msi_domain(dev);
2378 d = dev_get_msi_domain(&dev->bus->dev);
2380 dev_set_msi_domain(&dev->dev, d);
2383 void pci_device_add(struct pci_dev *dev, struct pci_bus *bus)
2387 pci_configure_device(dev);
2389 device_initialize(&dev->dev);
2390 dev->dev.release = pci_release_dev;
2392 set_dev_node(&dev->dev, pcibus_to_node(bus));
2393 dev->dev.dma_mask = &dev->dma_mask;
2394 dev->dev.dma_parms = &dev->dma_parms;
2395 dev->dev.coherent_dma_mask = 0xffffffffull;
2397 pci_set_dma_max_seg_size(dev, 65536);
2398 pci_set_dma_seg_boundary(dev, 0xffffffff);
2400 /* Fix up broken headers */
2401 pci_fixup_device(pci_fixup_header, dev);
2403 /* Moved out from quirk header fixup code */
2404 pci_reassigndev_resource_alignment(dev);
2406 /* Clear the state_saved flag */
2407 dev->state_saved = false;
2409 /* Initialize various capabilities */
2410 pci_init_capabilities(dev);
2413 * Add the device to our list of discovered devices
2414 * and the bus list for fixup functions, etc.
2416 down_write(&pci_bus_sem);
2417 list_add_tail(&dev->bus_list, &bus->devices);
2418 up_write(&pci_bus_sem);
2420 ret = pcibios_add_device(dev);
2423 /* Set up MSI IRQ domain */
2424 pci_set_msi_domain(dev);
2426 /* Notifier could use PCI capabilities */
2427 dev->match_driver = false;
2428 ret = device_add(&dev->dev);
2432 struct pci_dev *pci_scan_single_device(struct pci_bus *bus, int devfn)
2434 struct pci_dev *dev;
2436 dev = pci_get_slot(bus, devfn);
2442 dev = pci_scan_device(bus, devfn);
2446 pci_device_add(dev, bus);
2450 EXPORT_SYMBOL(pci_scan_single_device);
2452 static unsigned next_fn(struct pci_bus *bus, struct pci_dev *dev, unsigned fn)
2458 if (pci_ari_enabled(bus)) {
2461 pos = pci_find_ext_capability(dev, PCI_EXT_CAP_ID_ARI);
2465 pci_read_config_word(dev, pos + PCI_ARI_CAP, &cap);
2466 next_fn = PCI_ARI_CAP_NFN(cap);
2468 return 0; /* protect against malformed list */
2473 /* dev may be NULL for non-contiguous multifunction devices */
2474 if (!dev || dev->multifunction)
2475 return (fn + 1) % 8;
2480 static int only_one_child(struct pci_bus *bus)
2482 struct pci_dev *bridge = bus->self;
2485 * Systems with unusual topologies set PCI_SCAN_ALL_PCIE_DEVS so
2486 * we scan for all possible devices, not just Device 0.
2488 if (pci_has_flag(PCI_SCAN_ALL_PCIE_DEVS))
2492 * A PCIe Downstream Port normally leads to a Link with only Device
2493 * 0 on it (PCIe spec r3.1, sec 7.3.1). As an optimization, scan
2494 * only for Device 0 in that situation.
2496 * Checking has_secondary_link is a hack to identify Downstream
2497 * Ports because sometimes Switches are configured such that the
2498 * PCIe Port Type labels are backwards.
2500 if (bridge && pci_is_pcie(bridge) && bridge->has_secondary_link)
2507 * pci_scan_slot - Scan a PCI slot on a bus for devices
2508 * @bus: PCI bus to scan
2509 * @devfn: slot number to scan (must have zero function)
2511 * Scan a PCI slot on the specified PCI bus for devices, adding
2512 * discovered devices to the @bus->devices list. New devices
2513 * will not have is_added set.
2515 * Returns the number of new devices found.
2517 int pci_scan_slot(struct pci_bus *bus, int devfn)
2519 unsigned fn, nr = 0;
2520 struct pci_dev *dev;
2522 if (only_one_child(bus) && (devfn > 0))
2523 return 0; /* Already scanned the entire slot */
2525 dev = pci_scan_single_device(bus, devfn);
2528 if (!pci_dev_is_added(dev))
2531 for (fn = next_fn(bus, dev, 0); fn > 0; fn = next_fn(bus, dev, fn)) {
2532 dev = pci_scan_single_device(bus, devfn + fn);
2534 if (!pci_dev_is_added(dev))
2536 dev->multifunction = 1;
2540 /* Only one slot has PCIe device */
2541 if (bus->self && nr)
2542 pcie_aspm_init_link_state(bus->self);
2546 EXPORT_SYMBOL(pci_scan_slot);
2548 static int pcie_find_smpss(struct pci_dev *dev, void *data)
2552 if (!pci_is_pcie(dev))
2556 * We don't have a way to change MPS settings on devices that have
2557 * drivers attached. A hot-added device might support only the minimum
2558 * MPS setting (MPS=128). Therefore, if the fabric contains a bridge
2559 * where devices may be hot-added, we limit the fabric MPS to 128 so
2560 * hot-added devices will work correctly.
2562 * However, if we hot-add a device to a slot directly below a Root
2563 * Port, it's impossible for there to be other existing devices below
2564 * the port. We don't limit the MPS in this case because we can
2565 * reconfigure MPS on both the Root Port and the hot-added device,
2566 * and there are no other devices involved.
2568 * Note that this PCIE_BUS_SAFE path assumes no peer-to-peer DMA.
2570 if (dev->is_hotplug_bridge &&
2571 pci_pcie_type(dev) != PCI_EXP_TYPE_ROOT_PORT)
2574 if (*smpss > dev->pcie_mpss)
2575 *smpss = dev->pcie_mpss;
2580 static void pcie_write_mps(struct pci_dev *dev, int mps)
2584 if (pcie_bus_config == PCIE_BUS_PERFORMANCE) {
2585 mps = 128 << dev->pcie_mpss;
2587 if (pci_pcie_type(dev) != PCI_EXP_TYPE_ROOT_PORT &&
2591 * For "Performance", the assumption is made that
2592 * downstream communication will never be larger than
2593 * the MRRS. So, the MPS only needs to be configured
2594 * for the upstream communication. This being the case,
2595 * walk from the top down and set the MPS of the child
2596 * to that of the parent bus.
2598 * Configure the device MPS with the smaller of the
2599 * device MPSS or the bridge MPS (which is assumed to be
2600 * properly configured at this point to the largest
2601 * allowable MPS based on its parent bus).
2603 mps = min(mps, pcie_get_mps(dev->bus->self));
2606 rc = pcie_set_mps(dev, mps);
2608 pci_err(dev, "Failed attempting to set the MPS\n");
2611 static void pcie_write_mrrs(struct pci_dev *dev)
2616 * In the "safe" case, do not configure the MRRS. There appear to be
2617 * issues with setting MRRS to 0 on a number of devices.
2619 if (pcie_bus_config != PCIE_BUS_PERFORMANCE)
2623 * For max performance, the MRRS must be set to the largest supported
2624 * value. However, it cannot be configured larger than the MPS the
2625 * device or the bus can support. This should already be properly
2626 * configured by a prior call to pcie_write_mps().
2628 mrrs = pcie_get_mps(dev);
2631 * MRRS is a R/W register. Invalid values can be written, but a
2632 * subsequent read will verify if the value is acceptable or not.
2633 * If the MRRS value provided is not acceptable (e.g., too large),
2634 * shrink the value until it is acceptable to the HW.
2636 while (mrrs != pcie_get_readrq(dev) && mrrs >= 128) {
2637 rc = pcie_set_readrq(dev, mrrs);
2641 pci_warn(dev, "Failed attempting to set the MRRS\n");
2646 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");
2649 static int pcie_bus_configure_set(struct pci_dev *dev, void *data)
2653 if (!pci_is_pcie(dev))
2656 if (pcie_bus_config == PCIE_BUS_TUNE_OFF ||
2657 pcie_bus_config == PCIE_BUS_DEFAULT)
2660 mps = 128 << *(u8 *)data;
2661 orig_mps = pcie_get_mps(dev);
2663 pcie_write_mps(dev, mps);
2664 pcie_write_mrrs(dev);
2666 pci_info(dev, "Max Payload Size set to %4d/%4d (was %4d), Max Read Rq %4d\n",
2667 pcie_get_mps(dev), 128 << dev->pcie_mpss,
2668 orig_mps, pcie_get_readrq(dev));
2674 * pcie_bus_configure_settings() requires that pci_walk_bus work in a top-down,
2675 * parents then children fashion. If this changes, then this code will not
2678 void pcie_bus_configure_settings(struct pci_bus *bus)
2685 if (!pci_is_pcie(bus->self))
2689 * FIXME - Peer to peer DMA is possible, though the endpoint would need
2690 * to be aware of the MPS of the destination. To work around this,
2691 * simply force the MPS of the entire system to the smallest possible.
2693 if (pcie_bus_config == PCIE_BUS_PEER2PEER)
2696 if (pcie_bus_config == PCIE_BUS_SAFE) {
2697 smpss = bus->self->pcie_mpss;
2699 pcie_find_smpss(bus->self, &smpss);
2700 pci_walk_bus(bus, pcie_find_smpss, &smpss);
2703 pcie_bus_configure_set(bus->self, &smpss);
2704 pci_walk_bus(bus, pcie_bus_configure_set, &smpss);
2706 EXPORT_SYMBOL_GPL(pcie_bus_configure_settings);
2709 * Called after each bus is probed, but before its children are examined. This
2710 * is marked as __weak because multiple architectures define it.
2712 void __weak pcibios_fixup_bus(struct pci_bus *bus)
2714 /* nothing to do, expected to be removed in the future */
2718 * pci_scan_child_bus_extend() - Scan devices below a bus
2719 * @bus: Bus to scan for devices
2720 * @available_buses: Total number of buses available (%0 does not try to
2721 * extend beyond the minimal)
2723 * Scans devices below @bus including subordinate buses. Returns new
2724 * subordinate number including all the found devices. Passing
2725 * @available_buses causes the remaining bus space to be distributed
2726 * equally between hotplug-capable bridges to allow future extension of the
2729 static unsigned int pci_scan_child_bus_extend(struct pci_bus *bus,
2730 unsigned int available_buses)
2732 unsigned int used_buses, normal_bridges = 0, hotplug_bridges = 0;
2733 unsigned int start = bus->busn_res.start;
2734 unsigned int devfn, fn, cmax, max = start;
2735 struct pci_dev *dev;
2738 dev_dbg(&bus->dev, "scanning bus\n");
2740 /* Go find them, Rover! */
2741 for (devfn = 0; devfn < 256; devfn += 8) {
2742 nr_devs = pci_scan_slot(bus, devfn);
2745 * The Jailhouse hypervisor may pass individual functions of a
2746 * multi-function device to a guest without passing function 0.
2747 * Look for them as well.
2749 if (jailhouse_paravirt() && nr_devs == 0) {
2750 for (fn = 1; fn < 8; fn++) {
2751 dev = pci_scan_single_device(bus, devfn + fn);
2753 dev->multifunction = 1;
2758 /* Reserve buses for SR-IOV capability */
2759 used_buses = pci_iov_bus_range(bus);
2763 * After performing arch-dependent fixup of the bus, look behind
2764 * all PCI-to-PCI bridges on this bus.
2766 if (!bus->is_added) {
2767 dev_dbg(&bus->dev, "fixups for bus\n");
2768 pcibios_fixup_bus(bus);
2773 * Calculate how many hotplug bridges and normal bridges there
2774 * are on this bus. We will distribute the additional available
2775 * buses between hotplug bridges.
2777 for_each_pci_bridge(dev, bus) {
2778 if (dev->is_hotplug_bridge)
2785 * Scan bridges that are already configured. We don't touch them
2786 * unless they are misconfigured (which will be done in the second
2789 for_each_pci_bridge(dev, bus) {
2791 max = pci_scan_bridge_extend(bus, dev, max, 0, 0);
2794 * Reserve one bus for each bridge now to avoid extending
2795 * hotplug bridges too much during the second scan below.
2799 used_buses += cmax - max - 1;
2802 /* Scan bridges that need to be reconfigured */
2803 for_each_pci_bridge(dev, bus) {
2804 unsigned int buses = 0;
2806 if (!hotplug_bridges && normal_bridges == 1) {
2809 * There is only one bridge on the bus (upstream
2810 * port) so it gets all available buses which it
2811 * can then distribute to the possible hotplug
2814 buses = available_buses;
2815 } else if (dev->is_hotplug_bridge) {
2818 * Distribute the extra buses between hotplug
2821 buses = available_buses / hotplug_bridges;
2822 buses = min(buses, available_buses - used_buses + 1);
2826 max = pci_scan_bridge_extend(bus, dev, cmax, buses, 1);
2827 /* One bus is already accounted so don't add it again */
2829 used_buses += max - cmax - 1;
2833 * Make sure a hotplug bridge has at least the minimum requested
2834 * number of buses but allow it to grow up to the maximum available
2835 * bus number of there is room.
2837 if (bus->self && bus->self->is_hotplug_bridge) {
2838 used_buses = max_t(unsigned int, available_buses,
2839 pci_hotplug_bus_size - 1);
2840 if (max - start < used_buses) {
2841 max = start + used_buses;
2843 /* Do not allocate more buses than we have room left */
2844 if (max > bus->busn_res.end)
2845 max = bus->busn_res.end;
2847 dev_dbg(&bus->dev, "%pR extended by %#02x\n",
2848 &bus->busn_res, max - start);
2853 * We've scanned the bus and so we know all about what's on
2854 * the other side of any bridges that may be on this bus plus
2857 * Return how far we've got finding sub-buses.
2859 dev_dbg(&bus->dev, "bus scan returning with max=%02x\n", max);
2864 * pci_scan_child_bus() - Scan devices below a bus
2865 * @bus: Bus to scan for devices
2867 * Scans devices below @bus including subordinate buses. Returns new
2868 * subordinate number including all the found devices.
2870 unsigned int pci_scan_child_bus(struct pci_bus *bus)
2872 return pci_scan_child_bus_extend(bus, 0);
2874 EXPORT_SYMBOL_GPL(pci_scan_child_bus);
2877 * pcibios_root_bridge_prepare - Platform-specific host bridge setup
2878 * @bridge: Host bridge to set up
2880 * Default empty implementation. Replace with an architecture-specific setup
2881 * routine, if necessary.
2883 int __weak pcibios_root_bridge_prepare(struct pci_host_bridge *bridge)
2888 void __weak pcibios_add_bus(struct pci_bus *bus)
2892 void __weak pcibios_remove_bus(struct pci_bus *bus)
2896 struct pci_bus *pci_create_root_bus(struct device *parent, int bus,
2897 struct pci_ops *ops, void *sysdata, struct list_head *resources)
2900 struct pci_host_bridge *bridge;
2902 bridge = pci_alloc_host_bridge(0);
2906 bridge->dev.parent = parent;
2908 list_splice_init(resources, &bridge->windows);
2909 bridge->sysdata = sysdata;
2910 bridge->busnr = bus;
2913 error = pci_register_host_bridge(bridge);
2923 EXPORT_SYMBOL_GPL(pci_create_root_bus);
2925 int pci_host_probe(struct pci_host_bridge *bridge)
2927 struct pci_bus *bus, *child;
2930 ret = pci_scan_root_bus_bridge(bridge);
2932 dev_err(bridge->dev.parent, "Scanning root bridge failed");
2939 * We insert PCI resources into the iomem_resource and
2940 * ioport_resource trees in either pci_bus_claim_resources()
2941 * or pci_bus_assign_resources().
2943 if (pci_has_flag(PCI_PROBE_ONLY)) {
2944 pci_bus_claim_resources(bus);
2946 pci_bus_size_bridges(bus);
2947 pci_bus_assign_resources(bus);
2949 list_for_each_entry(child, &bus->children, node)
2950 pcie_bus_configure_settings(child);
2953 pci_bus_add_devices(bus);
2956 EXPORT_SYMBOL_GPL(pci_host_probe);
2958 int pci_bus_insert_busn_res(struct pci_bus *b, int bus, int bus_max)
2960 struct resource *res = &b->busn_res;
2961 struct resource *parent_res, *conflict;
2965 res->flags = IORESOURCE_BUS;
2967 if (!pci_is_root_bus(b))
2968 parent_res = &b->parent->busn_res;
2970 parent_res = get_pci_domain_busn_res(pci_domain_nr(b));
2971 res->flags |= IORESOURCE_PCI_FIXED;
2974 conflict = request_resource_conflict(parent_res, res);
2977 dev_printk(KERN_DEBUG, &b->dev,
2978 "busn_res: can not insert %pR under %s%pR (conflicts with %s %pR)\n",
2979 res, pci_is_root_bus(b) ? "domain " : "",
2980 parent_res, conflict->name, conflict);
2982 return conflict == NULL;
2985 int pci_bus_update_busn_res_end(struct pci_bus *b, int bus_max)
2987 struct resource *res = &b->busn_res;
2988 struct resource old_res = *res;
2989 resource_size_t size;
2992 if (res->start > bus_max)
2995 size = bus_max - res->start + 1;
2996 ret = adjust_resource(res, res->start, size);
2997 dev_printk(KERN_DEBUG, &b->dev,
2998 "busn_res: %pR end %s updated to %02x\n",
2999 &old_res, ret ? "can not be" : "is", bus_max);
3001 if (!ret && !res->parent)
3002 pci_bus_insert_busn_res(b, res->start, res->end);
3007 void pci_bus_release_busn_res(struct pci_bus *b)
3009 struct resource *res = &b->busn_res;
3012 if (!res->flags || !res->parent)
3015 ret = release_resource(res);
3016 dev_printk(KERN_DEBUG, &b->dev,
3017 "busn_res: %pR %s released\n",
3018 res, ret ? "can not be" : "is");
3021 int pci_scan_root_bus_bridge(struct pci_host_bridge *bridge)
3023 struct resource_entry *window;
3031 resource_list_for_each_entry(window, &bridge->windows)
3032 if (window->res->flags & IORESOURCE_BUS) {
3037 ret = pci_register_host_bridge(bridge);
3042 bus = bridge->busnr;
3046 "No busn resource found for root bus, will use [bus %02x-ff]\n",
3048 pci_bus_insert_busn_res(b, bus, 255);
3051 max = pci_scan_child_bus(b);
3054 pci_bus_update_busn_res_end(b, max);
3058 EXPORT_SYMBOL(pci_scan_root_bus_bridge);
3060 struct pci_bus *pci_scan_root_bus(struct device *parent, int bus,
3061 struct pci_ops *ops, void *sysdata, struct list_head *resources)
3063 struct resource_entry *window;
3068 resource_list_for_each_entry(window, resources)
3069 if (window->res->flags & IORESOURCE_BUS) {
3074 b = pci_create_root_bus(parent, bus, ops, sysdata, resources);
3080 "No busn resource found for root bus, will use [bus %02x-ff]\n",
3082 pci_bus_insert_busn_res(b, bus, 255);
3085 max = pci_scan_child_bus(b);
3088 pci_bus_update_busn_res_end(b, max);
3092 EXPORT_SYMBOL(pci_scan_root_bus);
3094 struct pci_bus *pci_scan_bus(int bus, struct pci_ops *ops,
3097 LIST_HEAD(resources);
3100 pci_add_resource(&resources, &ioport_resource);
3101 pci_add_resource(&resources, &iomem_resource);
3102 pci_add_resource(&resources, &busn_resource);
3103 b = pci_create_root_bus(NULL, bus, ops, sysdata, &resources);
3105 pci_scan_child_bus(b);
3107 pci_free_resource_list(&resources);
3111 EXPORT_SYMBOL(pci_scan_bus);
3114 * pci_rescan_bus_bridge_resize - Scan a PCI bus for devices
3115 * @bridge: PCI bridge for the bus to scan
3117 * Scan a PCI bus and child buses for new devices, add them,
3118 * and enable them, resizing bridge mmio/io resource if necessary
3119 * and possible. The caller must ensure the child devices are already
3120 * removed for resizing to occur.
3122 * Returns the max number of subordinate bus discovered.
3124 unsigned int pci_rescan_bus_bridge_resize(struct pci_dev *bridge)
3127 struct pci_bus *bus = bridge->subordinate;
3129 max = pci_scan_child_bus(bus);
3131 pci_assign_unassigned_bridge_resources(bridge);
3133 pci_bus_add_devices(bus);
3139 * pci_rescan_bus - Scan a PCI bus for devices
3140 * @bus: PCI bus to scan
3142 * Scan a PCI bus and child buses for new devices, add them,
3145 * Returns the max number of subordinate bus discovered.
3147 unsigned int pci_rescan_bus(struct pci_bus *bus)
3151 max = pci_scan_child_bus(bus);
3152 pci_assign_unassigned_bus_resources(bus);
3153 pci_bus_add_devices(bus);
3157 EXPORT_SYMBOL_GPL(pci_rescan_bus);
3160 * pci_rescan_bus(), pci_rescan_bus_bridge_resize() and PCI device removal
3161 * routines should always be executed under this mutex.
3163 static DEFINE_MUTEX(pci_rescan_remove_lock);
3165 void pci_lock_rescan_remove(void)
3167 mutex_lock(&pci_rescan_remove_lock);
3169 EXPORT_SYMBOL_GPL(pci_lock_rescan_remove);
3171 void pci_unlock_rescan_remove(void)
3173 mutex_unlock(&pci_rescan_remove_lock);
3175 EXPORT_SYMBOL_GPL(pci_unlock_rescan_remove);
3177 static int __init pci_sort_bf_cmp(const struct device *d_a,
3178 const struct device *d_b)
3180 const struct pci_dev *a = to_pci_dev(d_a);
3181 const struct pci_dev *b = to_pci_dev(d_b);
3183 if (pci_domain_nr(a->bus) < pci_domain_nr(b->bus)) return -1;
3184 else if (pci_domain_nr(a->bus) > pci_domain_nr(b->bus)) return 1;
3186 if (a->bus->number < b->bus->number) return -1;
3187 else if (a->bus->number > b->bus->number) return 1;
3189 if (a->devfn < b->devfn) return -1;
3190 else if (a->devfn > b->devfn) return 1;
3195 void __init pci_sort_breadthfirst(void)
3197 bus_sort_breadthfirst(&pci_bus_type, &pci_sort_bf_cmp);
3200 int pci_hp_add_bridge(struct pci_dev *dev)
3202 struct pci_bus *parent = dev->bus;
3203 int busnr, start = parent->busn_res.start;
3204 unsigned int available_buses = 0;
3205 int end = parent->busn_res.end;
3207 for (busnr = start; busnr <= end; busnr++) {
3208 if (!pci_find_bus(pci_domain_nr(parent), busnr))
3211 if (busnr-- > end) {
3212 pci_err(dev, "No bus number available for hot-added bridge\n");
3216 /* Scan bridges that are already configured */
3217 busnr = pci_scan_bridge(parent, dev, busnr, 0);
3220 * Distribute the available bus numbers between hotplug-capable
3221 * bridges to make extending the chain later possible.
3223 available_buses = end - busnr;
3225 /* Scan bridges that need to be reconfigured */
3226 pci_scan_bridge_extend(parent, dev, busnr, available_buses, 1);
3228 if (!dev->subordinate)
3233 EXPORT_SYMBOL_GPL(pci_hp_add_bridge);