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);
68 * Some device drivers need know if PCI is initiated.
69 * Basically, we think PCI is not initiated when there
70 * is no device to be found on the pci_bus_type.
72 int no_pci_devices(void)
77 dev = bus_find_next_device(&pci_bus_type, NULL);
78 no_devices = (dev == NULL);
82 EXPORT_SYMBOL(no_pci_devices);
87 static void release_pcibus_dev(struct device *dev)
89 struct pci_bus *pci_bus = to_pci_bus(dev);
91 put_device(pci_bus->bridge);
92 pci_bus_remove_resources(pci_bus);
93 pci_release_bus_of_node(pci_bus);
97 static struct class pcibus_class = {
99 .dev_release = &release_pcibus_dev,
100 .dev_groups = pcibus_groups,
103 static int __init pcibus_class_init(void)
105 return class_register(&pcibus_class);
107 postcore_initcall(pcibus_class_init);
109 static u64 pci_size(u64 base, u64 maxbase, u64 mask)
111 u64 size = mask & maxbase; /* Find the significant bits */
116 * Get the lowest of them to find the decode size, and from that
119 size = size & ~(size-1);
122 * base == maxbase can be valid only if the BAR has already been
123 * programmed with all 1s.
125 if (base == maxbase && ((base | (size - 1)) & mask) != mask)
131 static inline unsigned long decode_bar(struct pci_dev *dev, u32 bar)
136 if ((bar & PCI_BASE_ADDRESS_SPACE) == PCI_BASE_ADDRESS_SPACE_IO) {
137 flags = bar & ~PCI_BASE_ADDRESS_IO_MASK;
138 flags |= IORESOURCE_IO;
142 flags = bar & ~PCI_BASE_ADDRESS_MEM_MASK;
143 flags |= IORESOURCE_MEM;
144 if (flags & PCI_BASE_ADDRESS_MEM_PREFETCH)
145 flags |= IORESOURCE_PREFETCH;
147 mem_type = bar & PCI_BASE_ADDRESS_MEM_TYPE_MASK;
149 case PCI_BASE_ADDRESS_MEM_TYPE_32:
151 case PCI_BASE_ADDRESS_MEM_TYPE_1M:
152 /* 1M mem BAR treated as 32-bit BAR */
154 case PCI_BASE_ADDRESS_MEM_TYPE_64:
155 flags |= IORESOURCE_MEM_64;
158 /* mem unknown type treated as 32-bit BAR */
164 #define PCI_COMMAND_DECODE_ENABLE (PCI_COMMAND_MEMORY | PCI_COMMAND_IO)
167 * pci_read_base - Read a PCI BAR
168 * @dev: the PCI device
169 * @type: type of the BAR
170 * @res: resource buffer to be filled in
171 * @pos: BAR position in the config space
173 * Returns 1 if the BAR is 64-bit, or 0 if 32-bit.
175 int __pci_read_base(struct pci_dev *dev, enum pci_bar_type type,
176 struct resource *res, unsigned int pos)
178 u32 l = 0, sz = 0, mask;
179 u64 l64, sz64, mask64;
181 struct pci_bus_region region, inverted_region;
183 mask = type ? PCI_ROM_ADDRESS_MASK : ~0;
185 /* No printks while decoding is disabled! */
186 if (!dev->mmio_always_on) {
187 pci_read_config_word(dev, PCI_COMMAND, &orig_cmd);
188 if (orig_cmd & PCI_COMMAND_DECODE_ENABLE) {
189 pci_write_config_word(dev, PCI_COMMAND,
190 orig_cmd & ~PCI_COMMAND_DECODE_ENABLE);
194 res->name = pci_name(dev);
196 pci_read_config_dword(dev, pos, &l);
197 pci_write_config_dword(dev, pos, l | mask);
198 pci_read_config_dword(dev, pos, &sz);
199 pci_write_config_dword(dev, pos, l);
202 * All bits set in sz means the device isn't working properly.
203 * If the BAR isn't implemented, all bits must be 0. If it's a
204 * memory BAR or a ROM, bit 0 must be clear; if it's an io BAR, bit
207 if (sz == 0xffffffff)
211 * I don't know how l can have all bits set. Copied from old code.
212 * Maybe it fixes a bug on some ancient platform.
217 if (type == pci_bar_unknown) {
218 res->flags = decode_bar(dev, l);
219 res->flags |= IORESOURCE_SIZEALIGN;
220 if (res->flags & IORESOURCE_IO) {
221 l64 = l & PCI_BASE_ADDRESS_IO_MASK;
222 sz64 = sz & PCI_BASE_ADDRESS_IO_MASK;
223 mask64 = PCI_BASE_ADDRESS_IO_MASK & (u32)IO_SPACE_LIMIT;
225 l64 = l & PCI_BASE_ADDRESS_MEM_MASK;
226 sz64 = sz & PCI_BASE_ADDRESS_MEM_MASK;
227 mask64 = (u32)PCI_BASE_ADDRESS_MEM_MASK;
230 if (l & PCI_ROM_ADDRESS_ENABLE)
231 res->flags |= IORESOURCE_ROM_ENABLE;
232 l64 = l & PCI_ROM_ADDRESS_MASK;
233 sz64 = sz & PCI_ROM_ADDRESS_MASK;
234 mask64 = PCI_ROM_ADDRESS_MASK;
237 if (res->flags & IORESOURCE_MEM_64) {
238 pci_read_config_dword(dev, pos + 4, &l);
239 pci_write_config_dword(dev, pos + 4, ~0);
240 pci_read_config_dword(dev, pos + 4, &sz);
241 pci_write_config_dword(dev, pos + 4, l);
243 l64 |= ((u64)l << 32);
244 sz64 |= ((u64)sz << 32);
245 mask64 |= ((u64)~0 << 32);
248 if (!dev->mmio_always_on && (orig_cmd & PCI_COMMAND_DECODE_ENABLE))
249 pci_write_config_word(dev, PCI_COMMAND, orig_cmd);
254 sz64 = pci_size(l64, sz64, mask64);
256 pci_info(dev, FW_BUG "reg 0x%x: invalid BAR (can't size)\n",
261 if (res->flags & IORESOURCE_MEM_64) {
262 if ((sizeof(pci_bus_addr_t) < 8 || sizeof(resource_size_t) < 8)
263 && sz64 > 0x100000000ULL) {
264 res->flags |= IORESOURCE_UNSET | IORESOURCE_DISABLED;
267 pci_err(dev, "reg 0x%x: can't handle BAR larger than 4GB (size %#010llx)\n",
268 pos, (unsigned long long)sz64);
272 if ((sizeof(pci_bus_addr_t) < 8) && l) {
273 /* Above 32-bit boundary; try to reallocate */
274 res->flags |= IORESOURCE_UNSET;
277 pci_info(dev, "reg 0x%x: can't handle BAR above 4GB (bus address %#010llx)\n",
278 pos, (unsigned long long)l64);
284 region.end = l64 + sz64 - 1;
286 pcibios_bus_to_resource(dev->bus, res, ®ion);
287 pcibios_resource_to_bus(dev->bus, &inverted_region, res);
290 * If "A" is a BAR value (a bus address), "bus_to_resource(A)" is
291 * the corresponding resource address (the physical address used by
292 * the CPU. Converting that resource address back to a bus address
293 * should yield the original BAR value:
295 * resource_to_bus(bus_to_resource(A)) == A
297 * If it doesn't, CPU accesses to "bus_to_resource(A)" will not
298 * be claimed by the device.
300 if (inverted_region.start != region.start) {
301 res->flags |= IORESOURCE_UNSET;
303 res->end = region.end - region.start;
304 pci_info(dev, "reg 0x%x: initial BAR value %#010llx invalid\n",
305 pos, (unsigned long long)region.start);
315 pci_info(dev, "reg 0x%x: %pR\n", pos, res);
317 return (res->flags & IORESOURCE_MEM_64) ? 1 : 0;
320 static void pci_read_bases(struct pci_dev *dev, unsigned int howmany, int rom)
322 unsigned int pos, reg;
324 if (dev->non_compliant_bars)
327 /* Per PCIe r4.0, sec 9.3.4.1.11, the VF BARs are all RO Zero */
331 for (pos = 0; pos < howmany; pos++) {
332 struct resource *res = &dev->resource[pos];
333 reg = PCI_BASE_ADDRESS_0 + (pos << 2);
334 pos += __pci_read_base(dev, pci_bar_unknown, res, reg);
338 struct resource *res = &dev->resource[PCI_ROM_RESOURCE];
339 dev->rom_base_reg = rom;
340 res->flags = IORESOURCE_MEM | IORESOURCE_PREFETCH |
341 IORESOURCE_READONLY | IORESOURCE_SIZEALIGN;
342 __pci_read_base(dev, pci_bar_mem32, res, rom);
346 static void pci_read_bridge_windows(struct pci_dev *bridge)
351 pci_read_config_word(bridge, PCI_IO_BASE, &io);
353 pci_write_config_word(bridge, PCI_IO_BASE, 0xe0f0);
354 pci_read_config_word(bridge, PCI_IO_BASE, &io);
355 pci_write_config_word(bridge, PCI_IO_BASE, 0x0);
358 bridge->io_window = 1;
361 * DECchip 21050 pass 2 errata: the bridge may miss an address
362 * disconnect boundary by one PCI data phase. Workaround: do not
363 * use prefetching on this device.
365 if (bridge->vendor == PCI_VENDOR_ID_DEC && bridge->device == 0x0001)
368 pci_read_config_dword(bridge, PCI_PREF_MEMORY_BASE, &pmem);
370 pci_write_config_dword(bridge, PCI_PREF_MEMORY_BASE,
372 pci_read_config_dword(bridge, PCI_PREF_MEMORY_BASE, &pmem);
373 pci_write_config_dword(bridge, PCI_PREF_MEMORY_BASE, 0x0);
378 bridge->pref_window = 1;
380 if ((pmem & PCI_PREF_RANGE_TYPE_MASK) == PCI_PREF_RANGE_TYPE_64) {
383 * Bridge claims to have a 64-bit prefetchable memory
384 * window; verify that the upper bits are actually
387 pci_read_config_dword(bridge, PCI_PREF_BASE_UPPER32, &pmem);
388 pci_write_config_dword(bridge, PCI_PREF_BASE_UPPER32,
390 pci_read_config_dword(bridge, PCI_PREF_BASE_UPPER32, &tmp);
391 pci_write_config_dword(bridge, PCI_PREF_BASE_UPPER32, pmem);
393 bridge->pref_64_window = 1;
397 static void pci_read_bridge_io(struct pci_bus *child)
399 struct pci_dev *dev = child->self;
400 u8 io_base_lo, io_limit_lo;
401 unsigned long io_mask, io_granularity, base, limit;
402 struct pci_bus_region region;
403 struct resource *res;
405 io_mask = PCI_IO_RANGE_MASK;
406 io_granularity = 0x1000;
407 if (dev->io_window_1k) {
408 /* Support 1K I/O space granularity */
409 io_mask = PCI_IO_1K_RANGE_MASK;
410 io_granularity = 0x400;
413 res = child->resource[0];
414 pci_read_config_byte(dev, PCI_IO_BASE, &io_base_lo);
415 pci_read_config_byte(dev, PCI_IO_LIMIT, &io_limit_lo);
416 base = (io_base_lo & io_mask) << 8;
417 limit = (io_limit_lo & io_mask) << 8;
419 if ((io_base_lo & PCI_IO_RANGE_TYPE_MASK) == PCI_IO_RANGE_TYPE_32) {
420 u16 io_base_hi, io_limit_hi;
422 pci_read_config_word(dev, PCI_IO_BASE_UPPER16, &io_base_hi);
423 pci_read_config_word(dev, PCI_IO_LIMIT_UPPER16, &io_limit_hi);
424 base |= ((unsigned long) io_base_hi << 16);
425 limit |= ((unsigned long) io_limit_hi << 16);
429 res->flags = (io_base_lo & PCI_IO_RANGE_TYPE_MASK) | IORESOURCE_IO;
431 region.end = limit + io_granularity - 1;
432 pcibios_bus_to_resource(dev->bus, res, ®ion);
433 pci_info(dev, " bridge window %pR\n", res);
437 static void pci_read_bridge_mmio(struct pci_bus *child)
439 struct pci_dev *dev = child->self;
440 u16 mem_base_lo, mem_limit_lo;
441 unsigned long base, limit;
442 struct pci_bus_region region;
443 struct resource *res;
445 res = child->resource[1];
446 pci_read_config_word(dev, PCI_MEMORY_BASE, &mem_base_lo);
447 pci_read_config_word(dev, PCI_MEMORY_LIMIT, &mem_limit_lo);
448 base = ((unsigned long) mem_base_lo & PCI_MEMORY_RANGE_MASK) << 16;
449 limit = ((unsigned long) mem_limit_lo & PCI_MEMORY_RANGE_MASK) << 16;
451 res->flags = (mem_base_lo & PCI_MEMORY_RANGE_TYPE_MASK) | IORESOURCE_MEM;
453 region.end = limit + 0xfffff;
454 pcibios_bus_to_resource(dev->bus, res, ®ion);
455 pci_info(dev, " bridge window %pR\n", res);
459 static void pci_read_bridge_mmio_pref(struct pci_bus *child)
461 struct pci_dev *dev = child->self;
462 u16 mem_base_lo, mem_limit_lo;
464 pci_bus_addr_t base, limit;
465 struct pci_bus_region region;
466 struct resource *res;
468 res = child->resource[2];
469 pci_read_config_word(dev, PCI_PREF_MEMORY_BASE, &mem_base_lo);
470 pci_read_config_word(dev, PCI_PREF_MEMORY_LIMIT, &mem_limit_lo);
471 base64 = (mem_base_lo & PCI_PREF_RANGE_MASK) << 16;
472 limit64 = (mem_limit_lo & PCI_PREF_RANGE_MASK) << 16;
474 if ((mem_base_lo & PCI_PREF_RANGE_TYPE_MASK) == PCI_PREF_RANGE_TYPE_64) {
475 u32 mem_base_hi, mem_limit_hi;
477 pci_read_config_dword(dev, PCI_PREF_BASE_UPPER32, &mem_base_hi);
478 pci_read_config_dword(dev, PCI_PREF_LIMIT_UPPER32, &mem_limit_hi);
481 * Some bridges set the base > limit by default, and some
482 * (broken) BIOSes do not initialize them. If we find
483 * this, just assume they are not being used.
485 if (mem_base_hi <= mem_limit_hi) {
486 base64 |= (u64) mem_base_hi << 32;
487 limit64 |= (u64) mem_limit_hi << 32;
491 base = (pci_bus_addr_t) base64;
492 limit = (pci_bus_addr_t) limit64;
494 if (base != base64) {
495 pci_err(dev, "can't handle bridge window above 4GB (bus address %#010llx)\n",
496 (unsigned long long) base64);
501 res->flags = (mem_base_lo & PCI_PREF_RANGE_TYPE_MASK) |
502 IORESOURCE_MEM | IORESOURCE_PREFETCH;
503 if (res->flags & PCI_PREF_RANGE_TYPE_64)
504 res->flags |= IORESOURCE_MEM_64;
506 region.end = limit + 0xfffff;
507 pcibios_bus_to_resource(dev->bus, res, ®ion);
508 pci_info(dev, " bridge window %pR\n", res);
512 void pci_read_bridge_bases(struct pci_bus *child)
514 struct pci_dev *dev = child->self;
515 struct resource *res;
518 if (pci_is_root_bus(child)) /* It's a host bus, nothing to read */
521 pci_info(dev, "PCI bridge to %pR%s\n",
523 dev->transparent ? " (subtractive decode)" : "");
525 pci_bus_remove_resources(child);
526 for (i = 0; i < PCI_BRIDGE_RESOURCE_NUM; i++)
527 child->resource[i] = &dev->resource[PCI_BRIDGE_RESOURCES+i];
529 pci_read_bridge_io(child);
530 pci_read_bridge_mmio(child);
531 pci_read_bridge_mmio_pref(child);
533 if (dev->transparent) {
534 pci_bus_for_each_resource(child->parent, res, i) {
535 if (res && res->flags) {
536 pci_bus_add_resource(child, res,
537 PCI_SUBTRACTIVE_DECODE);
538 pci_info(dev, " bridge window %pR (subtractive decode)\n",
545 static struct pci_bus *pci_alloc_bus(struct pci_bus *parent)
549 b = kzalloc(sizeof(*b), GFP_KERNEL);
553 INIT_LIST_HEAD(&b->node);
554 INIT_LIST_HEAD(&b->children);
555 INIT_LIST_HEAD(&b->devices);
556 INIT_LIST_HEAD(&b->slots);
557 INIT_LIST_HEAD(&b->resources);
558 b->max_bus_speed = PCI_SPEED_UNKNOWN;
559 b->cur_bus_speed = PCI_SPEED_UNKNOWN;
560 #ifdef CONFIG_PCI_DOMAINS_GENERIC
562 b->domain_nr = parent->domain_nr;
567 static void pci_release_host_bridge_dev(struct device *dev)
569 struct pci_host_bridge *bridge = to_pci_host_bridge(dev);
571 if (bridge->release_fn)
572 bridge->release_fn(bridge);
574 pci_free_resource_list(&bridge->windows);
575 pci_free_resource_list(&bridge->dma_ranges);
579 static void pci_init_host_bridge(struct pci_host_bridge *bridge)
581 INIT_LIST_HEAD(&bridge->windows);
582 INIT_LIST_HEAD(&bridge->dma_ranges);
585 * We assume we can manage these PCIe features. Some systems may
586 * reserve these for use by the platform itself, e.g., an ACPI BIOS
587 * may implement its own AER handling and use _OSC to prevent the
588 * OS from interfering.
590 bridge->native_aer = 1;
591 bridge->native_pcie_hotplug = 1;
592 bridge->native_shpc_hotplug = 1;
593 bridge->native_pme = 1;
594 bridge->native_ltr = 1;
596 device_initialize(&bridge->dev);
599 struct pci_host_bridge *pci_alloc_host_bridge(size_t priv)
601 struct pci_host_bridge *bridge;
603 bridge = kzalloc(sizeof(*bridge) + priv, GFP_KERNEL);
607 pci_init_host_bridge(bridge);
608 bridge->dev.release = pci_release_host_bridge_dev;
612 EXPORT_SYMBOL(pci_alloc_host_bridge);
614 static void devm_pci_alloc_host_bridge_release(void *data)
616 pci_free_host_bridge(data);
619 struct pci_host_bridge *devm_pci_alloc_host_bridge(struct device *dev,
623 struct pci_host_bridge *bridge;
625 bridge = pci_alloc_host_bridge(priv);
629 ret = devm_add_action_or_reset(dev, devm_pci_alloc_host_bridge_release,
636 EXPORT_SYMBOL(devm_pci_alloc_host_bridge);
638 void pci_free_host_bridge(struct pci_host_bridge *bridge)
640 put_device(&bridge->dev);
642 EXPORT_SYMBOL(pci_free_host_bridge);
644 static const unsigned char pcix_bus_speed[] = {
645 PCI_SPEED_UNKNOWN, /* 0 */
646 PCI_SPEED_66MHz_PCIX, /* 1 */
647 PCI_SPEED_100MHz_PCIX, /* 2 */
648 PCI_SPEED_133MHz_PCIX, /* 3 */
649 PCI_SPEED_UNKNOWN, /* 4 */
650 PCI_SPEED_66MHz_PCIX_ECC, /* 5 */
651 PCI_SPEED_100MHz_PCIX_ECC, /* 6 */
652 PCI_SPEED_133MHz_PCIX_ECC, /* 7 */
653 PCI_SPEED_UNKNOWN, /* 8 */
654 PCI_SPEED_66MHz_PCIX_266, /* 9 */
655 PCI_SPEED_100MHz_PCIX_266, /* A */
656 PCI_SPEED_133MHz_PCIX_266, /* B */
657 PCI_SPEED_UNKNOWN, /* C */
658 PCI_SPEED_66MHz_PCIX_533, /* D */
659 PCI_SPEED_100MHz_PCIX_533, /* E */
660 PCI_SPEED_133MHz_PCIX_533 /* F */
663 const unsigned char pcie_link_speed[] = {
664 PCI_SPEED_UNKNOWN, /* 0 */
665 PCIE_SPEED_2_5GT, /* 1 */
666 PCIE_SPEED_5_0GT, /* 2 */
667 PCIE_SPEED_8_0GT, /* 3 */
668 PCIE_SPEED_16_0GT, /* 4 */
669 PCIE_SPEED_32_0GT, /* 5 */
670 PCI_SPEED_UNKNOWN, /* 6 */
671 PCI_SPEED_UNKNOWN, /* 7 */
672 PCI_SPEED_UNKNOWN, /* 8 */
673 PCI_SPEED_UNKNOWN, /* 9 */
674 PCI_SPEED_UNKNOWN, /* A */
675 PCI_SPEED_UNKNOWN, /* B */
676 PCI_SPEED_UNKNOWN, /* C */
677 PCI_SPEED_UNKNOWN, /* D */
678 PCI_SPEED_UNKNOWN, /* E */
679 PCI_SPEED_UNKNOWN /* F */
682 void pcie_update_link_speed(struct pci_bus *bus, u16 linksta)
684 bus->cur_bus_speed = pcie_link_speed[linksta & PCI_EXP_LNKSTA_CLS];
686 EXPORT_SYMBOL_GPL(pcie_update_link_speed);
688 static unsigned char agp_speeds[] = {
696 static enum pci_bus_speed agp_speed(int agp3, int agpstat)
702 else if (agpstat & 2)
704 else if (agpstat & 1)
716 return agp_speeds[index];
719 static void pci_set_bus_speed(struct pci_bus *bus)
721 struct pci_dev *bridge = bus->self;
724 pos = pci_find_capability(bridge, PCI_CAP_ID_AGP);
726 pos = pci_find_capability(bridge, PCI_CAP_ID_AGP3);
730 pci_read_config_dword(bridge, pos + PCI_AGP_STATUS, &agpstat);
731 bus->max_bus_speed = agp_speed(agpstat & 8, agpstat & 7);
733 pci_read_config_dword(bridge, pos + PCI_AGP_COMMAND, &agpcmd);
734 bus->cur_bus_speed = agp_speed(agpstat & 8, agpcmd & 7);
737 pos = pci_find_capability(bridge, PCI_CAP_ID_PCIX);
740 enum pci_bus_speed max;
742 pci_read_config_word(bridge, pos + PCI_X_BRIDGE_SSTATUS,
745 if (status & PCI_X_SSTATUS_533MHZ) {
746 max = PCI_SPEED_133MHz_PCIX_533;
747 } else if (status & PCI_X_SSTATUS_266MHZ) {
748 max = PCI_SPEED_133MHz_PCIX_266;
749 } else if (status & PCI_X_SSTATUS_133MHZ) {
750 if ((status & PCI_X_SSTATUS_VERS) == PCI_X_SSTATUS_V2)
751 max = PCI_SPEED_133MHz_PCIX_ECC;
753 max = PCI_SPEED_133MHz_PCIX;
755 max = PCI_SPEED_66MHz_PCIX;
758 bus->max_bus_speed = max;
759 bus->cur_bus_speed = pcix_bus_speed[
760 (status & PCI_X_SSTATUS_FREQ) >> 6];
765 if (pci_is_pcie(bridge)) {
769 pcie_capability_read_dword(bridge, PCI_EXP_LNKCAP, &linkcap);
770 bus->max_bus_speed = pcie_link_speed[linkcap & PCI_EXP_LNKCAP_SLS];
771 bridge->link_active_reporting = !!(linkcap & PCI_EXP_LNKCAP_DLLLARC);
773 pcie_capability_read_word(bridge, PCI_EXP_LNKSTA, &linksta);
774 pcie_update_link_speed(bus, linksta);
778 static struct irq_domain *pci_host_bridge_msi_domain(struct pci_bus *bus)
780 struct irq_domain *d;
783 * Any firmware interface that can resolve the msi_domain
784 * should be called from here.
786 d = pci_host_bridge_of_msi_domain(bus);
788 d = pci_host_bridge_acpi_msi_domain(bus);
790 #ifdef CONFIG_PCI_MSI_IRQ_DOMAIN
792 * If no IRQ domain was found via the OF tree, try looking it up
793 * directly through the fwnode_handle.
796 struct fwnode_handle *fwnode = pci_root_bus_fwnode(bus);
799 d = irq_find_matching_fwnode(fwnode,
807 static void pci_set_bus_msi_domain(struct pci_bus *bus)
809 struct irq_domain *d;
813 * The bus can be a root bus, a subordinate bus, or a virtual bus
814 * created by an SR-IOV device. Walk up to the first bridge device
815 * found or derive the domain from the host bridge.
817 for (b = bus, d = NULL; !d && !pci_is_root_bus(b); b = b->parent) {
819 d = dev_get_msi_domain(&b->self->dev);
823 d = pci_host_bridge_msi_domain(b);
825 dev_set_msi_domain(&bus->dev, d);
828 static int pci_register_host_bridge(struct pci_host_bridge *bridge)
830 struct device *parent = bridge->dev.parent;
831 struct resource_entry *window, *n;
832 struct pci_bus *bus, *b;
833 resource_size_t offset;
834 LIST_HEAD(resources);
835 struct resource *res;
840 bus = pci_alloc_bus(NULL);
846 /* Temporarily move resources off the list */
847 list_splice_init(&bridge->windows, &resources);
848 bus->sysdata = bridge->sysdata;
849 bus->msi = bridge->msi;
850 bus->ops = bridge->ops;
851 bus->number = bus->busn_res.start = bridge->busnr;
852 #ifdef CONFIG_PCI_DOMAINS_GENERIC
853 bus->domain_nr = pci_bus_find_domain_nr(bus, parent);
856 b = pci_find_bus(pci_domain_nr(bus), bridge->busnr);
858 /* Ignore it if we already got here via a different bridge */
859 dev_dbg(&b->dev, "bus already known\n");
864 dev_set_name(&bridge->dev, "pci%04x:%02x", pci_domain_nr(bus),
867 err = pcibios_root_bridge_prepare(bridge);
871 err = device_add(&bridge->dev);
873 put_device(&bridge->dev);
876 bus->bridge = get_device(&bridge->dev);
877 device_enable_async_suspend(bus->bridge);
878 pci_set_bus_of_node(bus);
879 pci_set_bus_msi_domain(bus);
882 set_dev_node(bus->bridge, pcibus_to_node(bus));
884 bus->dev.class = &pcibus_class;
885 bus->dev.parent = bus->bridge;
887 dev_set_name(&bus->dev, "%04x:%02x", pci_domain_nr(bus), bus->number);
888 name = dev_name(&bus->dev);
890 err = device_register(&bus->dev);
894 pcibios_add_bus(bus);
896 /* Create legacy_io and legacy_mem files for this bus */
897 pci_create_legacy_files(bus);
900 dev_info(parent, "PCI host bridge to bus %s\n", name);
902 pr_info("PCI host bridge to bus %s\n", name);
904 /* Add initial resources to the bus */
905 resource_list_for_each_entry_safe(window, n, &resources) {
906 list_move_tail(&window->node, &bridge->windows);
907 offset = window->offset;
910 if (res->flags & IORESOURCE_BUS)
911 pci_bus_insert_busn_res(bus, bus->number, res->end);
913 pci_bus_add_resource(bus, res, 0);
916 if (resource_type(res) == IORESOURCE_IO)
917 fmt = " (bus address [%#06llx-%#06llx])";
919 fmt = " (bus address [%#010llx-%#010llx])";
921 snprintf(addr, sizeof(addr), fmt,
922 (unsigned long long)(res->start - offset),
923 (unsigned long long)(res->end - offset));
927 dev_info(&bus->dev, "root bus resource %pR%s\n", res, addr);
930 down_write(&pci_bus_sem);
931 list_add_tail(&bus->node, &pci_root_buses);
932 up_write(&pci_bus_sem);
937 put_device(&bridge->dev);
938 device_del(&bridge->dev);
945 static bool pci_bridge_child_ext_cfg_accessible(struct pci_dev *bridge)
951 * If extended config space isn't accessible on a bridge's primary
952 * bus, we certainly can't access it on the secondary bus.
954 if (bridge->bus->bus_flags & PCI_BUS_FLAGS_NO_EXTCFG)
958 * PCIe Root Ports and switch ports are PCIe on both sides, so if
959 * extended config space is accessible on the primary, it's also
960 * accessible on the secondary.
962 if (pci_is_pcie(bridge) &&
963 (pci_pcie_type(bridge) == PCI_EXP_TYPE_ROOT_PORT ||
964 pci_pcie_type(bridge) == PCI_EXP_TYPE_UPSTREAM ||
965 pci_pcie_type(bridge) == PCI_EXP_TYPE_DOWNSTREAM))
969 * For the other bridge types:
970 * - PCI-to-PCI bridges
971 * - PCIe-to-PCI/PCI-X forward bridges
972 * - PCI/PCI-X-to-PCIe reverse bridges
973 * extended config space on the secondary side is only accessible
974 * if the bridge supports PCI-X Mode 2.
976 pos = pci_find_capability(bridge, PCI_CAP_ID_PCIX);
980 pci_read_config_dword(bridge, pos + PCI_X_STATUS, &status);
981 return status & (PCI_X_STATUS_266MHZ | PCI_X_STATUS_533MHZ);
984 static struct pci_bus *pci_alloc_child_bus(struct pci_bus *parent,
985 struct pci_dev *bridge, int busnr)
987 struct pci_bus *child;
991 /* Allocate a new bus and inherit stuff from the parent */
992 child = pci_alloc_bus(parent);
996 child->parent = parent;
997 child->ops = parent->ops;
998 child->msi = parent->msi;
999 child->sysdata = parent->sysdata;
1000 child->bus_flags = parent->bus_flags;
1003 * Initialize some portions of the bus device, but don't register
1004 * it now as the parent is not properly set up yet.
1006 child->dev.class = &pcibus_class;
1007 dev_set_name(&child->dev, "%04x:%02x", pci_domain_nr(child), busnr);
1009 /* Set up the primary, secondary and subordinate bus numbers */
1010 child->number = child->busn_res.start = busnr;
1011 child->primary = parent->busn_res.start;
1012 child->busn_res.end = 0xff;
1015 child->dev.parent = parent->bridge;
1019 child->self = bridge;
1020 child->bridge = get_device(&bridge->dev);
1021 child->dev.parent = child->bridge;
1022 pci_set_bus_of_node(child);
1023 pci_set_bus_speed(child);
1026 * Check whether extended config space is accessible on the child
1027 * bus. Note that we currently assume it is always accessible on
1030 if (!pci_bridge_child_ext_cfg_accessible(bridge)) {
1031 child->bus_flags |= PCI_BUS_FLAGS_NO_EXTCFG;
1032 pci_info(child, "extended config space not accessible\n");
1035 /* Set up default resource pointers and names */
1036 for (i = 0; i < PCI_BRIDGE_RESOURCE_NUM; i++) {
1037 child->resource[i] = &bridge->resource[PCI_BRIDGE_RESOURCES+i];
1038 child->resource[i]->name = child->name;
1040 bridge->subordinate = child;
1043 pci_set_bus_msi_domain(child);
1044 ret = device_register(&child->dev);
1047 pcibios_add_bus(child);
1049 if (child->ops->add_bus) {
1050 ret = child->ops->add_bus(child);
1051 if (WARN_ON(ret < 0))
1052 dev_err(&child->dev, "failed to add bus: %d\n", ret);
1055 /* Create legacy_io and legacy_mem files for this bus */
1056 pci_create_legacy_files(child);
1061 struct pci_bus *pci_add_new_bus(struct pci_bus *parent, struct pci_dev *dev,
1064 struct pci_bus *child;
1066 child = pci_alloc_child_bus(parent, dev, busnr);
1068 down_write(&pci_bus_sem);
1069 list_add_tail(&child->node, &parent->children);
1070 up_write(&pci_bus_sem);
1074 EXPORT_SYMBOL(pci_add_new_bus);
1076 static void pci_enable_crs(struct pci_dev *pdev)
1080 /* Enable CRS Software Visibility if supported */
1081 pcie_capability_read_word(pdev, PCI_EXP_RTCAP, &root_cap);
1082 if (root_cap & PCI_EXP_RTCAP_CRSVIS)
1083 pcie_capability_set_word(pdev, PCI_EXP_RTCTL,
1084 PCI_EXP_RTCTL_CRSSVE);
1087 static unsigned int pci_scan_child_bus_extend(struct pci_bus *bus,
1088 unsigned int available_buses);
1090 * pci_ea_fixed_busnrs() - Read fixed Secondary and Subordinate bus
1091 * numbers from EA capability.
1093 * @sec: updated with secondary bus number from EA
1094 * @sub: updated with subordinate bus number from EA
1096 * If @dev is a bridge with EA capability that specifies valid secondary
1097 * and subordinate bus numbers, return true with the bus numbers in @sec
1098 * and @sub. Otherwise return false.
1100 static bool pci_ea_fixed_busnrs(struct pci_dev *dev, u8 *sec, u8 *sub)
1106 if (dev->hdr_type != PCI_HEADER_TYPE_BRIDGE)
1109 /* find PCI EA capability in list */
1110 ea = pci_find_capability(dev, PCI_CAP_ID_EA);
1114 offset = ea + PCI_EA_FIRST_ENT;
1115 pci_read_config_dword(dev, offset, &dw);
1116 ea_sec = dw & PCI_EA_SEC_BUS_MASK;
1117 ea_sub = (dw & PCI_EA_SUB_BUS_MASK) >> PCI_EA_SUB_BUS_SHIFT;
1118 if (ea_sec == 0 || ea_sub < ea_sec)
1127 * pci_scan_bridge_extend() - Scan buses behind a bridge
1128 * @bus: Parent bus the bridge is on
1129 * @dev: Bridge itself
1130 * @max: Starting subordinate number of buses behind this bridge
1131 * @available_buses: Total number of buses available for this bridge and
1132 * the devices below. After the minimal bus space has
1133 * been allocated the remaining buses will be
1134 * distributed equally between hotplug-capable bridges.
1135 * @pass: Either %0 (scan already configured bridges) or %1 (scan bridges
1136 * that need to be reconfigured.
1138 * If it's a bridge, configure it and scan the bus behind it.
1139 * For CardBus bridges, we don't scan behind as the devices will
1140 * be handled by the bridge driver itself.
1142 * We need to process bridges in two passes -- first we scan those
1143 * already configured by the BIOS and after we are done with all of
1144 * them, we proceed to assigning numbers to the remaining buses in
1145 * order to avoid overlaps between old and new bus numbers.
1147 * Return: New subordinate number covering all buses behind this bridge.
1149 static int pci_scan_bridge_extend(struct pci_bus *bus, struct pci_dev *dev,
1150 int max, unsigned int available_buses,
1153 struct pci_bus *child;
1154 int is_cardbus = (dev->hdr_type == PCI_HEADER_TYPE_CARDBUS);
1155 u32 buses, i, j = 0;
1157 u8 primary, secondary, subordinate;
1160 u8 fixed_sec, fixed_sub;
1164 * Make sure the bridge is powered on to be able to access config
1165 * space of devices below it.
1167 pm_runtime_get_sync(&dev->dev);
1169 pci_read_config_dword(dev, PCI_PRIMARY_BUS, &buses);
1170 primary = buses & 0xFF;
1171 secondary = (buses >> 8) & 0xFF;
1172 subordinate = (buses >> 16) & 0xFF;
1174 pci_dbg(dev, "scanning [bus %02x-%02x] behind bridge, pass %d\n",
1175 secondary, subordinate, pass);
1177 if (!primary && (primary != bus->number) && secondary && subordinate) {
1178 pci_warn(dev, "Primary bus is hard wired to 0\n");
1179 primary = bus->number;
1182 /* Check if setup is sensible at all */
1184 (primary != bus->number || secondary <= bus->number ||
1185 secondary > subordinate)) {
1186 pci_info(dev, "bridge configuration invalid ([bus %02x-%02x]), reconfiguring\n",
1187 secondary, subordinate);
1192 * Disable Master-Abort Mode during probing to avoid reporting of
1193 * bus errors in some architectures.
1195 pci_read_config_word(dev, PCI_BRIDGE_CONTROL, &bctl);
1196 pci_write_config_word(dev, PCI_BRIDGE_CONTROL,
1197 bctl & ~PCI_BRIDGE_CTL_MASTER_ABORT);
1199 pci_enable_crs(dev);
1201 if ((secondary || subordinate) && !pcibios_assign_all_busses() &&
1202 !is_cardbus && !broken) {
1206 * Bus already configured by firmware, process it in the
1207 * first pass and just note the configuration.
1213 * The bus might already exist for two reasons: Either we
1214 * are rescanning the bus or the bus is reachable through
1215 * more than one bridge. The second case can happen with
1216 * the i450NX chipset.
1218 child = pci_find_bus(pci_domain_nr(bus), secondary);
1220 child = pci_add_new_bus(bus, dev, secondary);
1223 child->primary = primary;
1224 pci_bus_insert_busn_res(child, secondary, subordinate);
1225 child->bridge_ctl = bctl;
1228 cmax = pci_scan_child_bus(child);
1229 if (cmax > subordinate)
1230 pci_warn(dev, "bridge has subordinate %02x but max busn %02x\n",
1233 /* Subordinate should equal child->busn_res.end */
1234 if (subordinate > max)
1239 * We need to assign a number to this bus which we always
1240 * do in the second pass.
1243 if (pcibios_assign_all_busses() || broken || is_cardbus)
1246 * Temporarily disable forwarding of the
1247 * configuration cycles on all bridges in
1248 * this bus segment to avoid possible
1249 * conflicts in the second pass between two
1250 * bridges programmed with overlapping bus
1253 pci_write_config_dword(dev, PCI_PRIMARY_BUS,
1259 pci_write_config_word(dev, PCI_STATUS, 0xffff);
1261 /* Read bus numbers from EA Capability (if present) */
1262 fixed_buses = pci_ea_fixed_busnrs(dev, &fixed_sec, &fixed_sub);
1264 next_busnr = fixed_sec;
1266 next_busnr = max + 1;
1269 * Prevent assigning a bus number that already exists.
1270 * This can happen when a bridge is hot-plugged, so in this
1271 * case we only re-scan this bus.
1273 child = pci_find_bus(pci_domain_nr(bus), next_busnr);
1275 child = pci_add_new_bus(bus, dev, next_busnr);
1278 pci_bus_insert_busn_res(child, next_busnr,
1282 if (available_buses)
1285 buses = (buses & 0xff000000)
1286 | ((unsigned int)(child->primary) << 0)
1287 | ((unsigned int)(child->busn_res.start) << 8)
1288 | ((unsigned int)(child->busn_res.end) << 16);
1291 * yenta.c forces a secondary latency timer of 176.
1292 * Copy that behaviour here.
1295 buses &= ~0xff000000;
1296 buses |= CARDBUS_LATENCY_TIMER << 24;
1299 /* We need to blast all three values with a single write */
1300 pci_write_config_dword(dev, PCI_PRIMARY_BUS, buses);
1303 child->bridge_ctl = bctl;
1304 max = pci_scan_child_bus_extend(child, available_buses);
1308 * For CardBus bridges, we leave 4 bus numbers as
1309 * cards with a PCI-to-PCI bridge can be inserted
1312 for (i = 0; i < CARDBUS_RESERVE_BUSNR; i++) {
1313 struct pci_bus *parent = bus;
1314 if (pci_find_bus(pci_domain_nr(bus),
1317 while (parent->parent) {
1318 if ((!pcibios_assign_all_busses()) &&
1319 (parent->busn_res.end > max) &&
1320 (parent->busn_res.end <= max+i)) {
1323 parent = parent->parent;
1328 * Often, there are two CardBus
1329 * bridges -- try to leave one
1330 * valid bus number for each one.
1340 * Set subordinate bus number to its real value.
1341 * If fixed subordinate bus number exists from EA
1342 * capability then use it.
1346 pci_bus_update_busn_res_end(child, max);
1347 pci_write_config_byte(dev, PCI_SUBORDINATE_BUS, max);
1350 sprintf(child->name,
1351 (is_cardbus ? "PCI CardBus %04x:%02x" : "PCI Bus %04x:%02x"),
1352 pci_domain_nr(bus), child->number);
1354 /* Check that all devices are accessible */
1355 while (bus->parent) {
1356 if ((child->busn_res.end > bus->busn_res.end) ||
1357 (child->number > bus->busn_res.end) ||
1358 (child->number < bus->number) ||
1359 (child->busn_res.end < bus->number)) {
1360 dev_info(&dev->dev, "devices behind bridge are unusable because %pR cannot be assigned for them\n",
1368 pci_write_config_word(dev, PCI_BRIDGE_CONTROL, bctl);
1370 pm_runtime_put(&dev->dev);
1376 * pci_scan_bridge() - Scan buses behind a bridge
1377 * @bus: Parent bus the bridge is on
1378 * @dev: Bridge itself
1379 * @max: Starting subordinate number of buses behind this bridge
1380 * @pass: Either %0 (scan already configured bridges) or %1 (scan bridges
1381 * that need to be reconfigured.
1383 * If it's a bridge, configure it and scan the bus behind it.
1384 * For CardBus bridges, we don't scan behind as the devices will
1385 * be handled by the bridge driver itself.
1387 * We need to process bridges in two passes -- first we scan those
1388 * already configured by the BIOS and after we are done with all of
1389 * them, we proceed to assigning numbers to the remaining buses in
1390 * order to avoid overlaps between old and new bus numbers.
1392 * Return: New subordinate number covering all buses behind this bridge.
1394 int pci_scan_bridge(struct pci_bus *bus, struct pci_dev *dev, int max, int pass)
1396 return pci_scan_bridge_extend(bus, dev, max, 0, pass);
1398 EXPORT_SYMBOL(pci_scan_bridge);
1401 * Read interrupt line and base address registers.
1402 * The architecture-dependent code can tweak these, of course.
1404 static void pci_read_irq(struct pci_dev *dev)
1408 /* VFs are not allowed to use INTx, so skip the config reads */
1409 if (dev->is_virtfn) {
1415 pci_read_config_byte(dev, PCI_INTERRUPT_PIN, &irq);
1418 pci_read_config_byte(dev, PCI_INTERRUPT_LINE, &irq);
1422 void set_pcie_port_type(struct pci_dev *pdev)
1427 struct pci_dev *parent;
1429 pos = pci_find_capability(pdev, PCI_CAP_ID_EXP);
1433 pdev->pcie_cap = pos;
1434 pci_read_config_word(pdev, pos + PCI_EXP_FLAGS, ®16);
1435 pdev->pcie_flags_reg = reg16;
1436 pci_read_config_word(pdev, pos + PCI_EXP_DEVCAP, ®16);
1437 pdev->pcie_mpss = reg16 & PCI_EXP_DEVCAP_PAYLOAD;
1439 parent = pci_upstream_bridge(pdev);
1444 * Some systems do not identify their upstream/downstream ports
1445 * correctly so detect impossible configurations here and correct
1446 * the port type accordingly.
1448 type = pci_pcie_type(pdev);
1449 if (type == PCI_EXP_TYPE_DOWNSTREAM) {
1451 * If pdev claims to be downstream port but the parent
1452 * device is also downstream port assume pdev is actually
1455 if (pcie_downstream_port(parent)) {
1456 pci_info(pdev, "claims to be downstream port but is acting as upstream port, correcting type\n");
1457 pdev->pcie_flags_reg &= ~PCI_EXP_FLAGS_TYPE;
1458 pdev->pcie_flags_reg |= PCI_EXP_TYPE_UPSTREAM;
1460 } else if (type == PCI_EXP_TYPE_UPSTREAM) {
1462 * If pdev claims to be upstream port but the parent
1463 * device is also upstream port assume pdev is actually
1466 if (pci_pcie_type(parent) == PCI_EXP_TYPE_UPSTREAM) {
1467 pci_info(pdev, "claims to be upstream port but is acting as downstream port, correcting type\n");
1468 pdev->pcie_flags_reg &= ~PCI_EXP_FLAGS_TYPE;
1469 pdev->pcie_flags_reg |= PCI_EXP_TYPE_DOWNSTREAM;
1474 void set_pcie_hotplug_bridge(struct pci_dev *pdev)
1478 pcie_capability_read_dword(pdev, PCI_EXP_SLTCAP, ®32);
1479 if (reg32 & PCI_EXP_SLTCAP_HPC)
1480 pdev->is_hotplug_bridge = 1;
1483 static void set_pcie_thunderbolt(struct pci_dev *dev)
1488 while ((vsec = pci_find_next_ext_capability(dev, vsec,
1489 PCI_EXT_CAP_ID_VNDR))) {
1490 pci_read_config_dword(dev, vsec + PCI_VNDR_HEADER, &header);
1492 /* Is the device part of a Thunderbolt controller? */
1493 if (dev->vendor == PCI_VENDOR_ID_INTEL &&
1494 PCI_VNDR_HEADER_ID(header) == PCI_VSEC_ID_INTEL_TBT) {
1495 dev->is_thunderbolt = 1;
1501 static void set_pcie_untrusted(struct pci_dev *dev)
1503 struct pci_dev *parent;
1506 * If the upstream bridge is untrusted we treat this device
1507 * untrusted as well.
1509 parent = pci_upstream_bridge(dev);
1510 if (parent && parent->untrusted)
1511 dev->untrusted = true;
1515 * pci_ext_cfg_is_aliased - Is ext config space just an alias of std config?
1518 * PCI Express to PCI/PCI-X Bridge Specification, rev 1.0, 4.1.4 says that
1519 * when forwarding a type1 configuration request the bridge must check that
1520 * the extended register address field is zero. The bridge is not permitted
1521 * to forward the transactions and must handle it as an Unsupported Request.
1522 * Some bridges do not follow this rule and simply drop the extended register
1523 * bits, resulting in the standard config space being aliased, every 256
1524 * bytes across the entire configuration space. Test for this condition by
1525 * comparing the first dword of each potential alias to the vendor/device ID.
1527 * ASM1083/1085 PCIe-to-PCI Reversible Bridge (1b21:1080, rev 01 & 03)
1528 * AMD/ATI SBx00 PCI to PCI Bridge (1002:4384, rev 40)
1530 static bool pci_ext_cfg_is_aliased(struct pci_dev *dev)
1532 #ifdef CONFIG_PCI_QUIRKS
1536 pci_read_config_dword(dev, PCI_VENDOR_ID, &header);
1538 for (pos = PCI_CFG_SPACE_SIZE;
1539 pos < PCI_CFG_SPACE_EXP_SIZE; pos += PCI_CFG_SPACE_SIZE) {
1540 if (pci_read_config_dword(dev, pos, &tmp) != PCIBIOS_SUCCESSFUL
1552 * pci_cfg_space_size - Get the configuration space size of the PCI device
1555 * Regular PCI devices have 256 bytes, but PCI-X 2 and PCI Express devices
1556 * have 4096 bytes. Even if the device is capable, that doesn't mean we can
1557 * access it. Maybe we don't have a way to generate extended config space
1558 * accesses, or the device is behind a reverse Express bridge. So we try
1559 * reading the dword at 0x100 which must either be 0 or a valid extended
1560 * capability header.
1562 static int pci_cfg_space_size_ext(struct pci_dev *dev)
1565 int pos = PCI_CFG_SPACE_SIZE;
1567 if (pci_read_config_dword(dev, pos, &status) != PCIBIOS_SUCCESSFUL)
1568 return PCI_CFG_SPACE_SIZE;
1569 if (status == 0xffffffff || pci_ext_cfg_is_aliased(dev))
1570 return PCI_CFG_SPACE_SIZE;
1572 return PCI_CFG_SPACE_EXP_SIZE;
1575 int pci_cfg_space_size(struct pci_dev *dev)
1581 #ifdef CONFIG_PCI_IOV
1583 * Per the SR-IOV specification (rev 1.1, sec 3.5), VFs are required to
1584 * implement a PCIe capability and therefore must implement extended
1585 * config space. We can skip the NO_EXTCFG test below and the
1586 * reachability/aliasing test in pci_cfg_space_size_ext() by virtue of
1587 * the fact that the SR-IOV capability on the PF resides in extended
1588 * config space and must be accessible and non-aliased to have enabled
1589 * support for this VF. This is a micro performance optimization for
1590 * systems supporting many VFs.
1593 return PCI_CFG_SPACE_EXP_SIZE;
1596 if (dev->bus->bus_flags & PCI_BUS_FLAGS_NO_EXTCFG)
1597 return PCI_CFG_SPACE_SIZE;
1599 class = dev->class >> 8;
1600 if (class == PCI_CLASS_BRIDGE_HOST)
1601 return pci_cfg_space_size_ext(dev);
1603 if (pci_is_pcie(dev))
1604 return pci_cfg_space_size_ext(dev);
1606 pos = pci_find_capability(dev, PCI_CAP_ID_PCIX);
1608 return PCI_CFG_SPACE_SIZE;
1610 pci_read_config_dword(dev, pos + PCI_X_STATUS, &status);
1611 if (status & (PCI_X_STATUS_266MHZ | PCI_X_STATUS_533MHZ))
1612 return pci_cfg_space_size_ext(dev);
1614 return PCI_CFG_SPACE_SIZE;
1617 static u32 pci_class(struct pci_dev *dev)
1621 #ifdef CONFIG_PCI_IOV
1623 return dev->physfn->sriov->class;
1625 pci_read_config_dword(dev, PCI_CLASS_REVISION, &class);
1629 static void pci_subsystem_ids(struct pci_dev *dev, u16 *vendor, u16 *device)
1631 #ifdef CONFIG_PCI_IOV
1632 if (dev->is_virtfn) {
1633 *vendor = dev->physfn->sriov->subsystem_vendor;
1634 *device = dev->physfn->sriov->subsystem_device;
1638 pci_read_config_word(dev, PCI_SUBSYSTEM_VENDOR_ID, vendor);
1639 pci_read_config_word(dev, PCI_SUBSYSTEM_ID, device);
1642 static u8 pci_hdr_type(struct pci_dev *dev)
1646 #ifdef CONFIG_PCI_IOV
1648 return dev->physfn->sriov->hdr_type;
1650 pci_read_config_byte(dev, PCI_HEADER_TYPE, &hdr_type);
1654 #define LEGACY_IO_RESOURCE (IORESOURCE_IO | IORESOURCE_PCI_FIXED)
1656 static void pci_msi_setup_pci_dev(struct pci_dev *dev)
1659 * Disable the MSI hardware to avoid screaming interrupts
1660 * during boot. This is the power on reset default so
1661 * usually this should be a noop.
1663 dev->msi_cap = pci_find_capability(dev, PCI_CAP_ID_MSI);
1665 pci_msi_set_enable(dev, 0);
1667 dev->msix_cap = pci_find_capability(dev, PCI_CAP_ID_MSIX);
1669 pci_msix_clear_and_set_ctrl(dev, PCI_MSIX_FLAGS_ENABLE, 0);
1673 * pci_intx_mask_broken - Test PCI_COMMAND_INTX_DISABLE writability
1676 * Test whether PCI_COMMAND_INTX_DISABLE is writable for @dev. Check this
1677 * at enumeration-time to avoid modifying PCI_COMMAND at run-time.
1679 static int pci_intx_mask_broken(struct pci_dev *dev)
1681 u16 orig, toggle, new;
1683 pci_read_config_word(dev, PCI_COMMAND, &orig);
1684 toggle = orig ^ PCI_COMMAND_INTX_DISABLE;
1685 pci_write_config_word(dev, PCI_COMMAND, toggle);
1686 pci_read_config_word(dev, PCI_COMMAND, &new);
1688 pci_write_config_word(dev, PCI_COMMAND, orig);
1691 * PCI_COMMAND_INTX_DISABLE was reserved and read-only prior to PCI
1692 * r2.3, so strictly speaking, a device is not *broken* if it's not
1693 * writable. But we'll live with the misnomer for now.
1700 static void early_dump_pci_device(struct pci_dev *pdev)
1705 pci_info(pdev, "config space:\n");
1707 for (i = 0; i < 256; i += 4)
1708 pci_read_config_dword(pdev, i, &value[i / 4]);
1710 print_hex_dump(KERN_INFO, "", DUMP_PREFIX_OFFSET, 16, 1,
1715 * pci_setup_device - Fill in class and map information of a device
1716 * @dev: the device structure to fill
1718 * Initialize the device structure with information about the device's
1719 * vendor,class,memory and IO-space addresses, IRQ lines etc.
1720 * Called at initialisation of the PCI subsystem and by CardBus services.
1721 * Returns 0 on success and negative if unknown type of device (not normal,
1722 * bridge or CardBus).
1724 int pci_setup_device(struct pci_dev *dev)
1730 struct pci_bus_region region;
1731 struct resource *res;
1733 hdr_type = pci_hdr_type(dev);
1735 dev->sysdata = dev->bus->sysdata;
1736 dev->dev.parent = dev->bus->bridge;
1737 dev->dev.bus = &pci_bus_type;
1738 dev->hdr_type = hdr_type & 0x7f;
1739 dev->multifunction = !!(hdr_type & 0x80);
1740 dev->error_state = pci_channel_io_normal;
1741 set_pcie_port_type(dev);
1743 pci_dev_assign_slot(dev);
1746 * Assume 32-bit PCI; let 64-bit PCI cards (which are far rarer)
1747 * set this higher, assuming the system even supports it.
1749 dev->dma_mask = 0xffffffff;
1751 dev_set_name(&dev->dev, "%04x:%02x:%02x.%d", pci_domain_nr(dev->bus),
1752 dev->bus->number, PCI_SLOT(dev->devfn),
1753 PCI_FUNC(dev->devfn));
1755 class = pci_class(dev);
1757 dev->revision = class & 0xff;
1758 dev->class = class >> 8; /* upper 3 bytes */
1760 pci_info(dev, "[%04x:%04x] type %02x class %#08x\n",
1761 dev->vendor, dev->device, dev->hdr_type, dev->class);
1764 early_dump_pci_device(dev);
1766 /* Need to have dev->class ready */
1767 dev->cfg_size = pci_cfg_space_size(dev);
1769 /* Need to have dev->cfg_size ready */
1770 set_pcie_thunderbolt(dev);
1772 set_pcie_untrusted(dev);
1774 /* "Unknown power state" */
1775 dev->current_state = PCI_UNKNOWN;
1777 /* Early fixups, before probing the BARs */
1778 pci_fixup_device(pci_fixup_early, dev);
1780 /* Device class may be changed after fixup */
1781 class = dev->class >> 8;
1783 if (dev->non_compliant_bars && !dev->mmio_always_on) {
1784 pci_read_config_word(dev, PCI_COMMAND, &cmd);
1785 if (cmd & (PCI_COMMAND_IO | PCI_COMMAND_MEMORY)) {
1786 pci_info(dev, "device has non-compliant BARs; disabling IO/MEM decoding\n");
1787 cmd &= ~PCI_COMMAND_IO;
1788 cmd &= ~PCI_COMMAND_MEMORY;
1789 pci_write_config_word(dev, PCI_COMMAND, cmd);
1793 dev->broken_intx_masking = pci_intx_mask_broken(dev);
1795 switch (dev->hdr_type) { /* header type */
1796 case PCI_HEADER_TYPE_NORMAL: /* standard header */
1797 if (class == PCI_CLASS_BRIDGE_PCI)
1800 pci_read_bases(dev, 6, PCI_ROM_ADDRESS);
1802 pci_subsystem_ids(dev, &dev->subsystem_vendor, &dev->subsystem_device);
1805 * Do the ugly legacy mode stuff here rather than broken chip
1806 * quirk code. Legacy mode ATA controllers have fixed
1807 * addresses. These are not always echoed in BAR0-3, and
1808 * BAR0-3 in a few cases contain junk!
1810 if (class == PCI_CLASS_STORAGE_IDE) {
1812 pci_read_config_byte(dev, PCI_CLASS_PROG, &progif);
1813 if ((progif & 1) == 0) {
1814 region.start = 0x1F0;
1816 res = &dev->resource[0];
1817 res->flags = LEGACY_IO_RESOURCE;
1818 pcibios_bus_to_resource(dev->bus, res, ®ion);
1819 pci_info(dev, "legacy IDE quirk: reg 0x10: %pR\n",
1821 region.start = 0x3F6;
1823 res = &dev->resource[1];
1824 res->flags = LEGACY_IO_RESOURCE;
1825 pcibios_bus_to_resource(dev->bus, res, ®ion);
1826 pci_info(dev, "legacy IDE quirk: reg 0x14: %pR\n",
1829 if ((progif & 4) == 0) {
1830 region.start = 0x170;
1832 res = &dev->resource[2];
1833 res->flags = LEGACY_IO_RESOURCE;
1834 pcibios_bus_to_resource(dev->bus, res, ®ion);
1835 pci_info(dev, "legacy IDE quirk: reg 0x18: %pR\n",
1837 region.start = 0x376;
1839 res = &dev->resource[3];
1840 res->flags = LEGACY_IO_RESOURCE;
1841 pcibios_bus_to_resource(dev->bus, res, ®ion);
1842 pci_info(dev, "legacy IDE quirk: reg 0x1c: %pR\n",
1848 case PCI_HEADER_TYPE_BRIDGE: /* bridge header */
1850 * The PCI-to-PCI bridge spec requires that subtractive
1851 * decoding (i.e. transparent) bridge must have programming
1852 * interface code of 0x01.
1855 dev->transparent = ((dev->class & 0xff) == 1);
1856 pci_read_bases(dev, 2, PCI_ROM_ADDRESS1);
1857 pci_read_bridge_windows(dev);
1858 set_pcie_hotplug_bridge(dev);
1859 pos = pci_find_capability(dev, PCI_CAP_ID_SSVID);
1861 pci_read_config_word(dev, pos + PCI_SSVID_VENDOR_ID, &dev->subsystem_vendor);
1862 pci_read_config_word(dev, pos + PCI_SSVID_DEVICE_ID, &dev->subsystem_device);
1866 case PCI_HEADER_TYPE_CARDBUS: /* CardBus bridge header */
1867 if (class != PCI_CLASS_BRIDGE_CARDBUS)
1870 pci_read_bases(dev, 1, 0);
1871 pci_read_config_word(dev, PCI_CB_SUBSYSTEM_VENDOR_ID, &dev->subsystem_vendor);
1872 pci_read_config_word(dev, PCI_CB_SUBSYSTEM_ID, &dev->subsystem_device);
1875 default: /* unknown header */
1876 pci_err(dev, "unknown header type %02x, ignoring device\n",
1881 pci_err(dev, "ignoring class %#08x (doesn't match header type %02x)\n",
1882 dev->class, dev->hdr_type);
1883 dev->class = PCI_CLASS_NOT_DEFINED << 8;
1886 /* We found a fine healthy device, go go go... */
1890 static void pci_configure_mps(struct pci_dev *dev)
1892 struct pci_dev *bridge = pci_upstream_bridge(dev);
1893 int mps, mpss, p_mps, rc;
1895 if (!pci_is_pcie(dev))
1898 /* MPS and MRRS fields are of type 'RsvdP' for VFs, short-circuit out */
1903 * For Root Complex Integrated Endpoints, program the maximum
1904 * supported value unless limited by the PCIE_BUS_PEER2PEER case.
1906 if (pci_pcie_type(dev) == PCI_EXP_TYPE_RC_END) {
1907 if (pcie_bus_config == PCIE_BUS_PEER2PEER)
1910 mps = 128 << dev->pcie_mpss;
1911 rc = pcie_set_mps(dev, mps);
1913 pci_warn(dev, "can't set Max Payload Size to %d; if necessary, use \"pci=pcie_bus_safe\" and report a bug\n",
1919 if (!bridge || !pci_is_pcie(bridge))
1922 mps = pcie_get_mps(dev);
1923 p_mps = pcie_get_mps(bridge);
1928 if (pcie_bus_config == PCIE_BUS_TUNE_OFF) {
1929 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",
1930 mps, pci_name(bridge), p_mps);
1935 * Fancier MPS configuration is done later by
1936 * pcie_bus_configure_settings()
1938 if (pcie_bus_config != PCIE_BUS_DEFAULT)
1941 mpss = 128 << dev->pcie_mpss;
1942 if (mpss < p_mps && pci_pcie_type(bridge) == PCI_EXP_TYPE_ROOT_PORT) {
1943 pcie_set_mps(bridge, mpss);
1944 pci_info(dev, "Upstream bridge's Max Payload Size set to %d (was %d, max %d)\n",
1945 mpss, p_mps, 128 << bridge->pcie_mpss);
1946 p_mps = pcie_get_mps(bridge);
1949 rc = pcie_set_mps(dev, p_mps);
1951 pci_warn(dev, "can't set Max Payload Size to %d; if necessary, use \"pci=pcie_bus_safe\" and report a bug\n",
1956 pci_info(dev, "Max Payload Size set to %d (was %d, max %d)\n",
1960 int pci_configure_extended_tags(struct pci_dev *dev, void *ign)
1962 struct pci_host_bridge *host;
1967 if (!pci_is_pcie(dev))
1970 ret = pcie_capability_read_dword(dev, PCI_EXP_DEVCAP, &cap);
1974 if (!(cap & PCI_EXP_DEVCAP_EXT_TAG))
1977 ret = pcie_capability_read_word(dev, PCI_EXP_DEVCTL, &ctl);
1981 host = pci_find_host_bridge(dev->bus);
1986 * If some device in the hierarchy doesn't handle Extended Tags
1987 * correctly, make sure they're disabled.
1989 if (host->no_ext_tags) {
1990 if (ctl & PCI_EXP_DEVCTL_EXT_TAG) {
1991 pci_info(dev, "disabling Extended Tags\n");
1992 pcie_capability_clear_word(dev, PCI_EXP_DEVCTL,
1993 PCI_EXP_DEVCTL_EXT_TAG);
1998 if (!(ctl & PCI_EXP_DEVCTL_EXT_TAG)) {
1999 pci_info(dev, "enabling Extended Tags\n");
2000 pcie_capability_set_word(dev, PCI_EXP_DEVCTL,
2001 PCI_EXP_DEVCTL_EXT_TAG);
2007 * pcie_relaxed_ordering_enabled - Probe for PCIe relaxed ordering enable
2008 * @dev: PCI device to query
2010 * Returns true if the device has enabled relaxed ordering attribute.
2012 bool pcie_relaxed_ordering_enabled(struct pci_dev *dev)
2016 pcie_capability_read_word(dev, PCI_EXP_DEVCTL, &v);
2018 return !!(v & PCI_EXP_DEVCTL_RELAX_EN);
2020 EXPORT_SYMBOL(pcie_relaxed_ordering_enabled);
2022 static void pci_configure_relaxed_ordering(struct pci_dev *dev)
2024 struct pci_dev *root;
2026 /* PCI_EXP_DEVICE_RELAX_EN is RsvdP in VFs */
2030 if (!pcie_relaxed_ordering_enabled(dev))
2034 * For now, we only deal with Relaxed Ordering issues with Root
2035 * Ports. Peer-to-Peer DMA is another can of worms.
2037 root = pci_find_pcie_root_port(dev);
2041 if (root->dev_flags & PCI_DEV_FLAGS_NO_RELAXED_ORDERING) {
2042 pcie_capability_clear_word(dev, PCI_EXP_DEVCTL,
2043 PCI_EXP_DEVCTL_RELAX_EN);
2044 pci_info(dev, "Relaxed Ordering disabled because the Root Port didn't support it\n");
2048 static void pci_configure_ltr(struct pci_dev *dev)
2050 #ifdef CONFIG_PCIEASPM
2051 struct pci_host_bridge *host = pci_find_host_bridge(dev->bus);
2052 struct pci_dev *bridge;
2055 if (!pci_is_pcie(dev))
2058 pcie_capability_read_dword(dev, PCI_EXP_DEVCAP2, &cap);
2059 if (!(cap & PCI_EXP_DEVCAP2_LTR))
2062 pcie_capability_read_dword(dev, PCI_EXP_DEVCTL2, &ctl);
2063 if (ctl & PCI_EXP_DEVCTL2_LTR_EN) {
2064 if (pci_pcie_type(dev) == PCI_EXP_TYPE_ROOT_PORT) {
2069 bridge = pci_upstream_bridge(dev);
2070 if (bridge && bridge->ltr_path)
2076 if (!host->native_ltr)
2080 * Software must not enable LTR in an Endpoint unless the Root
2081 * Complex and all intermediate Switches indicate support for LTR.
2082 * PCIe r4.0, sec 6.18.
2084 if (pci_pcie_type(dev) == PCI_EXP_TYPE_ROOT_PORT ||
2085 ((bridge = pci_upstream_bridge(dev)) &&
2086 bridge->ltr_path)) {
2087 pcie_capability_set_word(dev, PCI_EXP_DEVCTL2,
2088 PCI_EXP_DEVCTL2_LTR_EN);
2094 static void pci_configure_eetlp_prefix(struct pci_dev *dev)
2096 #ifdef CONFIG_PCI_PASID
2097 struct pci_dev *bridge;
2101 if (!pci_is_pcie(dev))
2104 pcie_capability_read_dword(dev, PCI_EXP_DEVCAP2, &cap);
2105 if (!(cap & PCI_EXP_DEVCAP2_EE_PREFIX))
2108 pcie_type = pci_pcie_type(dev);
2109 if (pcie_type == PCI_EXP_TYPE_ROOT_PORT ||
2110 pcie_type == PCI_EXP_TYPE_RC_END)
2111 dev->eetlp_prefix_path = 1;
2113 bridge = pci_upstream_bridge(dev);
2114 if (bridge && bridge->eetlp_prefix_path)
2115 dev->eetlp_prefix_path = 1;
2120 static void pci_configure_serr(struct pci_dev *dev)
2124 if (dev->hdr_type == PCI_HEADER_TYPE_BRIDGE) {
2127 * A bridge will not forward ERR_ messages coming from an
2128 * endpoint unless SERR# forwarding is enabled.
2130 pci_read_config_word(dev, PCI_BRIDGE_CONTROL, &control);
2131 if (!(control & PCI_BRIDGE_CTL_SERR)) {
2132 control |= PCI_BRIDGE_CTL_SERR;
2133 pci_write_config_word(dev, PCI_BRIDGE_CONTROL, control);
2138 static void pci_configure_device(struct pci_dev *dev)
2140 pci_configure_mps(dev);
2141 pci_configure_extended_tags(dev, NULL);
2142 pci_configure_relaxed_ordering(dev);
2143 pci_configure_ltr(dev);
2144 pci_configure_eetlp_prefix(dev);
2145 pci_configure_serr(dev);
2147 pci_acpi_program_hp_params(dev);
2150 static void pci_release_capabilities(struct pci_dev *dev)
2153 pci_vpd_release(dev);
2154 pci_iov_release(dev);
2155 pci_free_cap_save_buffers(dev);
2159 * pci_release_dev - Free a PCI device structure when all users of it are
2161 * @dev: device that's been disconnected
2163 * Will be called only by the device core when all users of this PCI device are
2166 static void pci_release_dev(struct device *dev)
2168 struct pci_dev *pci_dev;
2170 pci_dev = to_pci_dev(dev);
2171 pci_release_capabilities(pci_dev);
2172 pci_release_of_node(pci_dev);
2173 pcibios_release_device(pci_dev);
2174 pci_bus_put(pci_dev->bus);
2175 kfree(pci_dev->driver_override);
2176 bitmap_free(pci_dev->dma_alias_mask);
2180 struct pci_dev *pci_alloc_dev(struct pci_bus *bus)
2182 struct pci_dev *dev;
2184 dev = kzalloc(sizeof(struct pci_dev), GFP_KERNEL);
2188 INIT_LIST_HEAD(&dev->bus_list);
2189 dev->dev.type = &pci_dev_type;
2190 dev->bus = pci_bus_get(bus);
2194 EXPORT_SYMBOL(pci_alloc_dev);
2196 static bool pci_bus_crs_vendor_id(u32 l)
2198 return (l & 0xffff) == 0x0001;
2201 static bool pci_bus_wait_crs(struct pci_bus *bus, int devfn, u32 *l,
2206 if (!pci_bus_crs_vendor_id(*l))
2207 return true; /* not a CRS completion */
2210 return false; /* CRS, but caller doesn't want to wait */
2213 * We got the reserved Vendor ID that indicates a completion with
2214 * Configuration Request Retry Status (CRS). Retry until we get a
2215 * valid Vendor ID or we time out.
2217 while (pci_bus_crs_vendor_id(*l)) {
2218 if (delay > timeout) {
2219 pr_warn("pci %04x:%02x:%02x.%d: not ready after %dms; giving up\n",
2220 pci_domain_nr(bus), bus->number,
2221 PCI_SLOT(devfn), PCI_FUNC(devfn), delay - 1);
2226 pr_info("pci %04x:%02x:%02x.%d: not ready after %dms; waiting\n",
2227 pci_domain_nr(bus), bus->number,
2228 PCI_SLOT(devfn), PCI_FUNC(devfn), delay - 1);
2233 if (pci_bus_read_config_dword(bus, devfn, PCI_VENDOR_ID, l))
2238 pr_info("pci %04x:%02x:%02x.%d: ready after %dms\n",
2239 pci_domain_nr(bus), bus->number,
2240 PCI_SLOT(devfn), PCI_FUNC(devfn), delay - 1);
2245 bool pci_bus_generic_read_dev_vendor_id(struct pci_bus *bus, int devfn, u32 *l,
2248 if (pci_bus_read_config_dword(bus, devfn, PCI_VENDOR_ID, l))
2251 /* Some broken boards return 0 or ~0 if a slot is empty: */
2252 if (*l == 0xffffffff || *l == 0x00000000 ||
2253 *l == 0x0000ffff || *l == 0xffff0000)
2256 if (pci_bus_crs_vendor_id(*l))
2257 return pci_bus_wait_crs(bus, devfn, l, timeout);
2262 bool pci_bus_read_dev_vendor_id(struct pci_bus *bus, int devfn, u32 *l,
2265 #ifdef CONFIG_PCI_QUIRKS
2266 struct pci_dev *bridge = bus->self;
2269 * Certain IDT switches have an issue where they improperly trigger
2270 * ACS Source Validation errors on completions for config reads.
2272 if (bridge && bridge->vendor == PCI_VENDOR_ID_IDT &&
2273 bridge->device == 0x80b5)
2274 return pci_idt_bus_quirk(bus, devfn, l, timeout);
2277 return pci_bus_generic_read_dev_vendor_id(bus, devfn, l, timeout);
2279 EXPORT_SYMBOL(pci_bus_read_dev_vendor_id);
2282 * Read the config data for a PCI device, sanity-check it,
2283 * and fill in the dev structure.
2285 static struct pci_dev *pci_scan_device(struct pci_bus *bus, int devfn)
2287 struct pci_dev *dev;
2290 if (!pci_bus_read_dev_vendor_id(bus, devfn, &l, 60*1000))
2293 dev = pci_alloc_dev(bus);
2298 dev->vendor = l & 0xffff;
2299 dev->device = (l >> 16) & 0xffff;
2301 pci_set_of_node(dev);
2303 if (pci_setup_device(dev)) {
2304 pci_release_of_node(dev);
2305 pci_bus_put(dev->bus);
2313 void pcie_report_downtraining(struct pci_dev *dev)
2315 if (!pci_is_pcie(dev))
2318 /* Look from the device up to avoid downstream ports with no devices */
2319 if ((pci_pcie_type(dev) != PCI_EXP_TYPE_ENDPOINT) &&
2320 (pci_pcie_type(dev) != PCI_EXP_TYPE_LEG_END) &&
2321 (pci_pcie_type(dev) != PCI_EXP_TYPE_UPSTREAM))
2324 /* Multi-function PCIe devices share the same link/status */
2325 if (PCI_FUNC(dev->devfn) != 0 || dev->is_virtfn)
2328 /* Print link status only if the device is constrained by the fabric */
2329 __pcie_print_link_status(dev, false);
2332 static void pci_init_capabilities(struct pci_dev *dev)
2334 /* Enhanced Allocation */
2337 /* Setup MSI caps & disable MSI/MSI-X interrupts */
2338 pci_msi_setup_pci_dev(dev);
2340 /* Buffers for saving PCIe and PCI-X capabilities */
2341 pci_allocate_cap_save_buffers(dev);
2343 /* Power Management */
2346 /* Vital Product Data */
2349 /* Alternative Routing-ID Forwarding */
2350 pci_configure_ari(dev);
2352 /* Single Root I/O Virtualization */
2355 /* Address Translation Services */
2358 /* Enable ACS P2P upstream forwarding */
2359 pci_enable_acs(dev);
2361 /* Precision Time Measurement */
2364 /* Advanced Error Reporting */
2367 pcie_report_downtraining(dev);
2369 if (pci_probe_reset_function(dev) == 0)
2374 * This is the equivalent of pci_host_bridge_msi_domain() that acts on
2375 * devices. Firmware interfaces that can select the MSI domain on a
2376 * per-device basis should be called from here.
2378 static struct irq_domain *pci_dev_msi_domain(struct pci_dev *dev)
2380 struct irq_domain *d;
2383 * If a domain has been set through the pcibios_add_device()
2384 * callback, then this is the one (platform code knows best).
2386 d = dev_get_msi_domain(&dev->dev);
2391 * Let's see if we have a firmware interface able to provide
2394 d = pci_msi_get_device_domain(dev);
2401 static void pci_set_msi_domain(struct pci_dev *dev)
2403 struct irq_domain *d;
2406 * If the platform or firmware interfaces cannot supply a
2407 * device-specific MSI domain, then inherit the default domain
2408 * from the host bridge itself.
2410 d = pci_dev_msi_domain(dev);
2412 d = dev_get_msi_domain(&dev->bus->dev);
2414 dev_set_msi_domain(&dev->dev, d);
2417 void pci_device_add(struct pci_dev *dev, struct pci_bus *bus)
2421 pci_configure_device(dev);
2423 device_initialize(&dev->dev);
2424 dev->dev.release = pci_release_dev;
2426 set_dev_node(&dev->dev, pcibus_to_node(bus));
2427 dev->dev.dma_mask = &dev->dma_mask;
2428 dev->dev.dma_parms = &dev->dma_parms;
2429 dev->dev.coherent_dma_mask = 0xffffffffull;
2431 dma_set_max_seg_size(&dev->dev, 65536);
2432 dma_set_seg_boundary(&dev->dev, 0xffffffff);
2434 /* Fix up broken headers */
2435 pci_fixup_device(pci_fixup_header, dev);
2437 /* Moved out from quirk header fixup code */
2438 pci_reassigndev_resource_alignment(dev);
2440 /* Clear the state_saved flag */
2441 dev->state_saved = false;
2443 /* Initialize various capabilities */
2444 pci_init_capabilities(dev);
2447 * Add the device to our list of discovered devices
2448 * and the bus list for fixup functions, etc.
2450 down_write(&pci_bus_sem);
2451 list_add_tail(&dev->bus_list, &bus->devices);
2452 up_write(&pci_bus_sem);
2454 ret = pcibios_add_device(dev);
2457 /* Set up MSI IRQ domain */
2458 pci_set_msi_domain(dev);
2460 /* Notifier could use PCI capabilities */
2461 dev->match_driver = false;
2462 ret = device_add(&dev->dev);
2466 struct pci_dev *pci_scan_single_device(struct pci_bus *bus, int devfn)
2468 struct pci_dev *dev;
2470 dev = pci_get_slot(bus, devfn);
2476 dev = pci_scan_device(bus, devfn);
2480 pci_device_add(dev, bus);
2484 EXPORT_SYMBOL(pci_scan_single_device);
2486 static unsigned next_fn(struct pci_bus *bus, struct pci_dev *dev, unsigned fn)
2492 if (pci_ari_enabled(bus)) {
2495 pos = pci_find_ext_capability(dev, PCI_EXT_CAP_ID_ARI);
2499 pci_read_config_word(dev, pos + PCI_ARI_CAP, &cap);
2500 next_fn = PCI_ARI_CAP_NFN(cap);
2502 return 0; /* protect against malformed list */
2507 /* dev may be NULL for non-contiguous multifunction devices */
2508 if (!dev || dev->multifunction)
2509 return (fn + 1) % 8;
2514 static int only_one_child(struct pci_bus *bus)
2516 struct pci_dev *bridge = bus->self;
2519 * Systems with unusual topologies set PCI_SCAN_ALL_PCIE_DEVS so
2520 * we scan for all possible devices, not just Device 0.
2522 if (pci_has_flag(PCI_SCAN_ALL_PCIE_DEVS))
2526 * A PCIe Downstream Port normally leads to a Link with only Device
2527 * 0 on it (PCIe spec r3.1, sec 7.3.1). As an optimization, scan
2528 * only for Device 0 in that situation.
2530 if (bridge && pci_is_pcie(bridge) && pcie_downstream_port(bridge))
2537 * pci_scan_slot - Scan a PCI slot on a bus for devices
2538 * @bus: PCI bus to scan
2539 * @devfn: slot number to scan (must have zero function)
2541 * Scan a PCI slot on the specified PCI bus for devices, adding
2542 * discovered devices to the @bus->devices list. New devices
2543 * will not have is_added set.
2545 * Returns the number of new devices found.
2547 int pci_scan_slot(struct pci_bus *bus, int devfn)
2549 unsigned fn, nr = 0;
2550 struct pci_dev *dev;
2552 if (only_one_child(bus) && (devfn > 0))
2553 return 0; /* Already scanned the entire slot */
2555 dev = pci_scan_single_device(bus, devfn);
2558 if (!pci_dev_is_added(dev))
2561 for (fn = next_fn(bus, dev, 0); fn > 0; fn = next_fn(bus, dev, fn)) {
2562 dev = pci_scan_single_device(bus, devfn + fn);
2564 if (!pci_dev_is_added(dev))
2566 dev->multifunction = 1;
2570 /* Only one slot has PCIe device */
2571 if (bus->self && nr)
2572 pcie_aspm_init_link_state(bus->self);
2576 EXPORT_SYMBOL(pci_scan_slot);
2578 static int pcie_find_smpss(struct pci_dev *dev, void *data)
2582 if (!pci_is_pcie(dev))
2586 * We don't have a way to change MPS settings on devices that have
2587 * drivers attached. A hot-added device might support only the minimum
2588 * MPS setting (MPS=128). Therefore, if the fabric contains a bridge
2589 * where devices may be hot-added, we limit the fabric MPS to 128 so
2590 * hot-added devices will work correctly.
2592 * However, if we hot-add a device to a slot directly below a Root
2593 * Port, it's impossible for there to be other existing devices below
2594 * the port. We don't limit the MPS in this case because we can
2595 * reconfigure MPS on both the Root Port and the hot-added device,
2596 * and there are no other devices involved.
2598 * Note that this PCIE_BUS_SAFE path assumes no peer-to-peer DMA.
2600 if (dev->is_hotplug_bridge &&
2601 pci_pcie_type(dev) != PCI_EXP_TYPE_ROOT_PORT)
2604 if (*smpss > dev->pcie_mpss)
2605 *smpss = dev->pcie_mpss;
2610 static void pcie_write_mps(struct pci_dev *dev, int mps)
2614 if (pcie_bus_config == PCIE_BUS_PERFORMANCE) {
2615 mps = 128 << dev->pcie_mpss;
2617 if (pci_pcie_type(dev) != PCI_EXP_TYPE_ROOT_PORT &&
2621 * For "Performance", the assumption is made that
2622 * downstream communication will never be larger than
2623 * the MRRS. So, the MPS only needs to be configured
2624 * for the upstream communication. This being the case,
2625 * walk from the top down and set the MPS of the child
2626 * to that of the parent bus.
2628 * Configure the device MPS with the smaller of the
2629 * device MPSS or the bridge MPS (which is assumed to be
2630 * properly configured at this point to the largest
2631 * allowable MPS based on its parent bus).
2633 mps = min(mps, pcie_get_mps(dev->bus->self));
2636 rc = pcie_set_mps(dev, mps);
2638 pci_err(dev, "Failed attempting to set the MPS\n");
2641 static void pcie_write_mrrs(struct pci_dev *dev)
2646 * In the "safe" case, do not configure the MRRS. There appear to be
2647 * issues with setting MRRS to 0 on a number of devices.
2649 if (pcie_bus_config != PCIE_BUS_PERFORMANCE)
2653 * For max performance, the MRRS must be set to the largest supported
2654 * value. However, it cannot be configured larger than the MPS the
2655 * device or the bus can support. This should already be properly
2656 * configured by a prior call to pcie_write_mps().
2658 mrrs = pcie_get_mps(dev);
2661 * MRRS is a R/W register. Invalid values can be written, but a
2662 * subsequent read will verify if the value is acceptable or not.
2663 * If the MRRS value provided is not acceptable (e.g., too large),
2664 * shrink the value until it is acceptable to the HW.
2666 while (mrrs != pcie_get_readrq(dev) && mrrs >= 128) {
2667 rc = pcie_set_readrq(dev, mrrs);
2671 pci_warn(dev, "Failed attempting to set the MRRS\n");
2676 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");
2679 static int pcie_bus_configure_set(struct pci_dev *dev, void *data)
2683 if (!pci_is_pcie(dev))
2686 if (pcie_bus_config == PCIE_BUS_TUNE_OFF ||
2687 pcie_bus_config == PCIE_BUS_DEFAULT)
2690 mps = 128 << *(u8 *)data;
2691 orig_mps = pcie_get_mps(dev);
2693 pcie_write_mps(dev, mps);
2694 pcie_write_mrrs(dev);
2696 pci_info(dev, "Max Payload Size set to %4d/%4d (was %4d), Max Read Rq %4d\n",
2697 pcie_get_mps(dev), 128 << dev->pcie_mpss,
2698 orig_mps, pcie_get_readrq(dev));
2704 * pcie_bus_configure_settings() requires that pci_walk_bus work in a top-down,
2705 * parents then children fashion. If this changes, then this code will not
2708 void pcie_bus_configure_settings(struct pci_bus *bus)
2715 if (!pci_is_pcie(bus->self))
2719 * FIXME - Peer to peer DMA is possible, though the endpoint would need
2720 * to be aware of the MPS of the destination. To work around this,
2721 * simply force the MPS of the entire system to the smallest possible.
2723 if (pcie_bus_config == PCIE_BUS_PEER2PEER)
2726 if (pcie_bus_config == PCIE_BUS_SAFE) {
2727 smpss = bus->self->pcie_mpss;
2729 pcie_find_smpss(bus->self, &smpss);
2730 pci_walk_bus(bus, pcie_find_smpss, &smpss);
2733 pcie_bus_configure_set(bus->self, &smpss);
2734 pci_walk_bus(bus, pcie_bus_configure_set, &smpss);
2736 EXPORT_SYMBOL_GPL(pcie_bus_configure_settings);
2739 * Called after each bus is probed, but before its children are examined. This
2740 * is marked as __weak because multiple architectures define it.
2742 void __weak pcibios_fixup_bus(struct pci_bus *bus)
2744 /* nothing to do, expected to be removed in the future */
2748 * pci_scan_child_bus_extend() - Scan devices below a bus
2749 * @bus: Bus to scan for devices
2750 * @available_buses: Total number of buses available (%0 does not try to
2751 * extend beyond the minimal)
2753 * Scans devices below @bus including subordinate buses. Returns new
2754 * subordinate number including all the found devices. Passing
2755 * @available_buses causes the remaining bus space to be distributed
2756 * equally between hotplug-capable bridges to allow future extension of the
2759 static unsigned int pci_scan_child_bus_extend(struct pci_bus *bus,
2760 unsigned int available_buses)
2762 unsigned int used_buses, normal_bridges = 0, hotplug_bridges = 0;
2763 unsigned int start = bus->busn_res.start;
2764 unsigned int devfn, fn, cmax, max = start;
2765 struct pci_dev *dev;
2768 dev_dbg(&bus->dev, "scanning bus\n");
2770 /* Go find them, Rover! */
2771 for (devfn = 0; devfn < 256; devfn += 8) {
2772 nr_devs = pci_scan_slot(bus, devfn);
2775 * The Jailhouse hypervisor may pass individual functions of a
2776 * multi-function device to a guest without passing function 0.
2777 * Look for them as well.
2779 if (jailhouse_paravirt() && nr_devs == 0) {
2780 for (fn = 1; fn < 8; fn++) {
2781 dev = pci_scan_single_device(bus, devfn + fn);
2783 dev->multifunction = 1;
2788 /* Reserve buses for SR-IOV capability */
2789 used_buses = pci_iov_bus_range(bus);
2793 * After performing arch-dependent fixup of the bus, look behind
2794 * all PCI-to-PCI bridges on this bus.
2796 if (!bus->is_added) {
2797 dev_dbg(&bus->dev, "fixups for bus\n");
2798 pcibios_fixup_bus(bus);
2803 * Calculate how many hotplug bridges and normal bridges there
2804 * are on this bus. We will distribute the additional available
2805 * buses between hotplug bridges.
2807 for_each_pci_bridge(dev, bus) {
2808 if (dev->is_hotplug_bridge)
2815 * Scan bridges that are already configured. We don't touch them
2816 * unless they are misconfigured (which will be done in the second
2819 for_each_pci_bridge(dev, bus) {
2821 max = pci_scan_bridge_extend(bus, dev, max, 0, 0);
2824 * Reserve one bus for each bridge now to avoid extending
2825 * hotplug bridges too much during the second scan below.
2829 used_buses += cmax - max - 1;
2832 /* Scan bridges that need to be reconfigured */
2833 for_each_pci_bridge(dev, bus) {
2834 unsigned int buses = 0;
2836 if (!hotplug_bridges && normal_bridges == 1) {
2839 * There is only one bridge on the bus (upstream
2840 * port) so it gets all available buses which it
2841 * can then distribute to the possible hotplug
2844 buses = available_buses;
2845 } else if (dev->is_hotplug_bridge) {
2848 * Distribute the extra buses between hotplug
2851 buses = available_buses / hotplug_bridges;
2852 buses = min(buses, available_buses - used_buses + 1);
2856 max = pci_scan_bridge_extend(bus, dev, cmax, buses, 1);
2857 /* One bus is already accounted so don't add it again */
2859 used_buses += max - cmax - 1;
2863 * Make sure a hotplug bridge has at least the minimum requested
2864 * number of buses but allow it to grow up to the maximum available
2865 * bus number of there is room.
2867 if (bus->self && bus->self->is_hotplug_bridge) {
2868 used_buses = max_t(unsigned int, available_buses,
2869 pci_hotplug_bus_size - 1);
2870 if (max - start < used_buses) {
2871 max = start + used_buses;
2873 /* Do not allocate more buses than we have room left */
2874 if (max > bus->busn_res.end)
2875 max = bus->busn_res.end;
2877 dev_dbg(&bus->dev, "%pR extended by %#02x\n",
2878 &bus->busn_res, max - start);
2883 * We've scanned the bus and so we know all about what's on
2884 * the other side of any bridges that may be on this bus plus
2887 * Return how far we've got finding sub-buses.
2889 dev_dbg(&bus->dev, "bus scan returning with max=%02x\n", max);
2894 * pci_scan_child_bus() - Scan devices below a bus
2895 * @bus: Bus to scan for devices
2897 * Scans devices below @bus including subordinate buses. Returns new
2898 * subordinate number including all the found devices.
2900 unsigned int pci_scan_child_bus(struct pci_bus *bus)
2902 return pci_scan_child_bus_extend(bus, 0);
2904 EXPORT_SYMBOL_GPL(pci_scan_child_bus);
2907 * pcibios_root_bridge_prepare - Platform-specific host bridge setup
2908 * @bridge: Host bridge to set up
2910 * Default empty implementation. Replace with an architecture-specific setup
2911 * routine, if necessary.
2913 int __weak pcibios_root_bridge_prepare(struct pci_host_bridge *bridge)
2918 void __weak pcibios_add_bus(struct pci_bus *bus)
2922 void __weak pcibios_remove_bus(struct pci_bus *bus)
2926 struct pci_bus *pci_create_root_bus(struct device *parent, int bus,
2927 struct pci_ops *ops, void *sysdata, struct list_head *resources)
2930 struct pci_host_bridge *bridge;
2932 bridge = pci_alloc_host_bridge(0);
2936 bridge->dev.parent = parent;
2938 list_splice_init(resources, &bridge->windows);
2939 bridge->sysdata = sysdata;
2940 bridge->busnr = bus;
2943 error = pci_register_host_bridge(bridge);
2950 put_device(&bridge->dev);
2953 EXPORT_SYMBOL_GPL(pci_create_root_bus);
2955 int pci_host_probe(struct pci_host_bridge *bridge)
2957 struct pci_bus *bus, *child;
2960 ret = pci_scan_root_bus_bridge(bridge);
2962 dev_err(bridge->dev.parent, "Scanning root bridge failed");
2969 * We insert PCI resources into the iomem_resource and
2970 * ioport_resource trees in either pci_bus_claim_resources()
2971 * or pci_bus_assign_resources().
2973 if (pci_has_flag(PCI_PROBE_ONLY)) {
2974 pci_bus_claim_resources(bus);
2976 pci_bus_size_bridges(bus);
2977 pci_bus_assign_resources(bus);
2979 list_for_each_entry(child, &bus->children, node)
2980 pcie_bus_configure_settings(child);
2983 pci_bus_add_devices(bus);
2986 EXPORT_SYMBOL_GPL(pci_host_probe);
2988 int pci_bus_insert_busn_res(struct pci_bus *b, int bus, int bus_max)
2990 struct resource *res = &b->busn_res;
2991 struct resource *parent_res, *conflict;
2995 res->flags = IORESOURCE_BUS;
2997 if (!pci_is_root_bus(b))
2998 parent_res = &b->parent->busn_res;
3000 parent_res = get_pci_domain_busn_res(pci_domain_nr(b));
3001 res->flags |= IORESOURCE_PCI_FIXED;
3004 conflict = request_resource_conflict(parent_res, res);
3008 "busn_res: can not insert %pR under %s%pR (conflicts with %s %pR)\n",
3009 res, pci_is_root_bus(b) ? "domain " : "",
3010 parent_res, conflict->name, conflict);
3012 return conflict == NULL;
3015 int pci_bus_update_busn_res_end(struct pci_bus *b, int bus_max)
3017 struct resource *res = &b->busn_res;
3018 struct resource old_res = *res;
3019 resource_size_t size;
3022 if (res->start > bus_max)
3025 size = bus_max - res->start + 1;
3026 ret = adjust_resource(res, res->start, size);
3027 dev_info(&b->dev, "busn_res: %pR end %s updated to %02x\n",
3028 &old_res, ret ? "can not be" : "is", bus_max);
3030 if (!ret && !res->parent)
3031 pci_bus_insert_busn_res(b, res->start, res->end);
3036 void pci_bus_release_busn_res(struct pci_bus *b)
3038 struct resource *res = &b->busn_res;
3041 if (!res->flags || !res->parent)
3044 ret = release_resource(res);
3045 dev_info(&b->dev, "busn_res: %pR %s released\n",
3046 res, ret ? "can not be" : "is");
3049 int pci_scan_root_bus_bridge(struct pci_host_bridge *bridge)
3051 struct resource_entry *window;
3059 resource_list_for_each_entry(window, &bridge->windows)
3060 if (window->res->flags & IORESOURCE_BUS) {
3065 ret = pci_register_host_bridge(bridge);
3070 bus = bridge->busnr;
3074 "No busn resource found for root bus, will use [bus %02x-ff]\n",
3076 pci_bus_insert_busn_res(b, bus, 255);
3079 max = pci_scan_child_bus(b);
3082 pci_bus_update_busn_res_end(b, max);
3086 EXPORT_SYMBOL(pci_scan_root_bus_bridge);
3088 struct pci_bus *pci_scan_root_bus(struct device *parent, int bus,
3089 struct pci_ops *ops, void *sysdata, struct list_head *resources)
3091 struct resource_entry *window;
3096 resource_list_for_each_entry(window, resources)
3097 if (window->res->flags & IORESOURCE_BUS) {
3102 b = pci_create_root_bus(parent, bus, ops, sysdata, resources);
3108 "No busn resource found for root bus, will use [bus %02x-ff]\n",
3110 pci_bus_insert_busn_res(b, bus, 255);
3113 max = pci_scan_child_bus(b);
3116 pci_bus_update_busn_res_end(b, max);
3120 EXPORT_SYMBOL(pci_scan_root_bus);
3122 struct pci_bus *pci_scan_bus(int bus, struct pci_ops *ops,
3125 LIST_HEAD(resources);
3128 pci_add_resource(&resources, &ioport_resource);
3129 pci_add_resource(&resources, &iomem_resource);
3130 pci_add_resource(&resources, &busn_resource);
3131 b = pci_create_root_bus(NULL, bus, ops, sysdata, &resources);
3133 pci_scan_child_bus(b);
3135 pci_free_resource_list(&resources);
3139 EXPORT_SYMBOL(pci_scan_bus);
3142 * pci_rescan_bus_bridge_resize - Scan a PCI bus for devices
3143 * @bridge: PCI bridge for the bus to scan
3145 * Scan a PCI bus and child buses for new devices, add them,
3146 * and enable them, resizing bridge mmio/io resource if necessary
3147 * and possible. The caller must ensure the child devices are already
3148 * removed for resizing to occur.
3150 * Returns the max number of subordinate bus discovered.
3152 unsigned int pci_rescan_bus_bridge_resize(struct pci_dev *bridge)
3155 struct pci_bus *bus = bridge->subordinate;
3157 max = pci_scan_child_bus(bus);
3159 pci_assign_unassigned_bridge_resources(bridge);
3161 pci_bus_add_devices(bus);
3167 * pci_rescan_bus - Scan a PCI bus for devices
3168 * @bus: PCI bus to scan
3170 * Scan a PCI bus and child buses for new devices, add them,
3173 * Returns the max number of subordinate bus discovered.
3175 unsigned int pci_rescan_bus(struct pci_bus *bus)
3179 max = pci_scan_child_bus(bus);
3180 pci_assign_unassigned_bus_resources(bus);
3181 pci_bus_add_devices(bus);
3185 EXPORT_SYMBOL_GPL(pci_rescan_bus);
3188 * pci_rescan_bus(), pci_rescan_bus_bridge_resize() and PCI device removal
3189 * routines should always be executed under this mutex.
3191 static DEFINE_MUTEX(pci_rescan_remove_lock);
3193 void pci_lock_rescan_remove(void)
3195 mutex_lock(&pci_rescan_remove_lock);
3197 EXPORT_SYMBOL_GPL(pci_lock_rescan_remove);
3199 void pci_unlock_rescan_remove(void)
3201 mutex_unlock(&pci_rescan_remove_lock);
3203 EXPORT_SYMBOL_GPL(pci_unlock_rescan_remove);
3205 static int __init pci_sort_bf_cmp(const struct device *d_a,
3206 const struct device *d_b)
3208 const struct pci_dev *a = to_pci_dev(d_a);
3209 const struct pci_dev *b = to_pci_dev(d_b);
3211 if (pci_domain_nr(a->bus) < pci_domain_nr(b->bus)) return -1;
3212 else if (pci_domain_nr(a->bus) > pci_domain_nr(b->bus)) return 1;
3214 if (a->bus->number < b->bus->number) return -1;
3215 else if (a->bus->number > b->bus->number) return 1;
3217 if (a->devfn < b->devfn) return -1;
3218 else if (a->devfn > b->devfn) return 1;
3223 void __init pci_sort_breadthfirst(void)
3225 bus_sort_breadthfirst(&pci_bus_type, &pci_sort_bf_cmp);
3228 int pci_hp_add_bridge(struct pci_dev *dev)
3230 struct pci_bus *parent = dev->bus;
3231 int busnr, start = parent->busn_res.start;
3232 unsigned int available_buses = 0;
3233 int end = parent->busn_res.end;
3235 for (busnr = start; busnr <= end; busnr++) {
3236 if (!pci_find_bus(pci_domain_nr(parent), busnr))
3239 if (busnr-- > end) {
3240 pci_err(dev, "No bus number available for hot-added bridge\n");
3244 /* Scan bridges that are already configured */
3245 busnr = pci_scan_bridge(parent, dev, busnr, 0);
3248 * Distribute the available bus numbers between hotplug-capable
3249 * bridges to make extending the chain later possible.
3251 available_buses = end - busnr;
3253 /* Scan bridges that need to be reconfigured */
3254 pci_scan_bridge_extend(parent, dev, busnr, available_buses, 1);
3256 if (!dev->subordinate)
3261 EXPORT_SYMBOL_GPL(pci_hp_add_bridge);