blob: 6ae65a3da3384761942e434790fed1efa96bd5ac [file] [log] [blame]
b.liue9582032025-04-17 19:18:16 +08001// SPDX-License-Identifier: GPL-2.0
2/*
3 * PCI detection and setup code
4 */
5
6#include <linux/kernel.h>
7#include <linux/delay.h>
8#include <linux/init.h>
9#include <linux/pci.h>
10#include <linux/of_device.h>
11#include <linux/of_pci.h>
12#include <linux/pci_hotplug.h>
13#include <linux/slab.h>
14#include <linux/module.h>
15#include <linux/cpumask.h>
16#include <linux/aer.h>
17#include <linux/acpi.h>
18#include <linux/hypervisor.h>
19#include <linux/irqdomain.h>
20#include <linux/pm_runtime.h>
21#include "pci.h"
22
23#define CARDBUS_LATENCY_TIMER 176 /* secondary latency timer */
24#define CARDBUS_RESERVE_BUSNR 3
25
26static struct resource busn_resource = {
27 .name = "PCI busn",
28 .start = 0,
29 .end = 255,
30 .flags = IORESOURCE_BUS,
31};
32
33/* Ugh. Need to stop exporting this to modules. */
34LIST_HEAD(pci_root_buses);
35EXPORT_SYMBOL(pci_root_buses);
36
37static LIST_HEAD(pci_domain_busn_res_list);
38
39struct pci_domain_busn_res {
40 struct list_head list;
41 struct resource res;
42 int domain_nr;
43};
44
45static struct resource *get_pci_domain_busn_res(int domain_nr)
46{
47 struct pci_domain_busn_res *r;
48
49 list_for_each_entry(r, &pci_domain_busn_res_list, list)
50 if (r->domain_nr == domain_nr)
51 return &r->res;
52
53 r = kzalloc(sizeof(*r), GFP_KERNEL);
54 if (!r)
55 return NULL;
56
57 r->domain_nr = domain_nr;
58 r->res.start = 0;
59 r->res.end = 0xff;
60 r->res.flags = IORESOURCE_BUS | IORESOURCE_PCI_FIXED;
61
62 list_add_tail(&r->list, &pci_domain_busn_res_list);
63
64 return &r->res;
65}
66
67/*
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.
71 */
72int no_pci_devices(void)
73{
74 struct device *dev;
75 int no_devices;
76
77 dev = bus_find_next_device(&pci_bus_type, NULL);
78 no_devices = (dev == NULL);
79 put_device(dev);
80 return no_devices;
81}
82EXPORT_SYMBOL(no_pci_devices);
83
84/*
85 * PCI Bus Class
86 */
87static void release_pcibus_dev(struct device *dev)
88{
89 struct pci_bus *pci_bus = to_pci_bus(dev);
90
91 put_device(pci_bus->bridge);
92 pci_bus_remove_resources(pci_bus);
93 pci_release_bus_of_node(pci_bus);
94 kfree(pci_bus);
95}
96
97static struct class pcibus_class = {
98 .name = "pci_bus",
99 .dev_release = &release_pcibus_dev,
100 .dev_groups = pcibus_groups,
101};
102
103static int __init pcibus_class_init(void)
104{
105 return class_register(&pcibus_class);
106}
107postcore_initcall(pcibus_class_init);
108
109static u64 pci_size(u64 base, u64 maxbase, u64 mask)
110{
111 u64 size = mask & maxbase; /* Find the significant bits */
112 if (!size)
113 return 0;
114
115 /*
116 * Get the lowest of them to find the decode size, and from that
117 * the extent.
118 */
119 size = size & ~(size-1);
120
121 /*
122 * base == maxbase can be valid only if the BAR has already been
123 * programmed with all 1s.
124 */
125 if (base == maxbase && ((base | (size - 1)) & mask) != mask)
126 return 0;
127
128 return size;
129}
130
131static inline unsigned long decode_bar(struct pci_dev *dev, u32 bar)
132{
133 u32 mem_type;
134 unsigned long flags;
135
136 if ((bar & PCI_BASE_ADDRESS_SPACE) == PCI_BASE_ADDRESS_SPACE_IO) {
137 flags = bar & ~PCI_BASE_ADDRESS_IO_MASK;
138 flags |= IORESOURCE_IO;
139 return flags;
140 }
141
142 flags = bar & ~PCI_BASE_ADDRESS_MEM_MASK;
143 flags |= IORESOURCE_MEM;
144 if (flags & PCI_BASE_ADDRESS_MEM_PREFETCH)
145 flags |= IORESOURCE_PREFETCH;
146
147 mem_type = bar & PCI_BASE_ADDRESS_MEM_TYPE_MASK;
148 switch (mem_type) {
149 case PCI_BASE_ADDRESS_MEM_TYPE_32:
150 break;
151 case PCI_BASE_ADDRESS_MEM_TYPE_1M:
152 /* 1M mem BAR treated as 32-bit BAR */
153 break;
154 case PCI_BASE_ADDRESS_MEM_TYPE_64:
155 flags |= IORESOURCE_MEM_64;
156 break;
157 default:
158 /* mem unknown type treated as 32-bit BAR */
159 break;
160 }
161 return flags;
162}
163
164#define PCI_COMMAND_DECODE_ENABLE (PCI_COMMAND_MEMORY | PCI_COMMAND_IO)
165
166/**
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
172 *
173 * Returns 1 if the BAR is 64-bit, or 0 if 32-bit.
174 */
175int __pci_read_base(struct pci_dev *dev, enum pci_bar_type type,
176 struct resource *res, unsigned int pos)
177{
178 u32 l = 0, sz = 0, mask;
179 u64 l64, sz64, mask64;
180 u16 orig_cmd;
181 struct pci_bus_region region, inverted_region;
182
183 mask = type ? PCI_ROM_ADDRESS_MASK : ~0;
184
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);
191 }
192 }
193
194 res->name = pci_name(dev);
195
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);
200
201 /*
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
205 * 1 must be clear.
206 */
207 if (sz == 0xffffffff)
208 sz = 0;
209
210 /*
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.
213 */
214 if (l == 0xffffffff)
215 l = 0;
216
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;
224 } else {
225 l64 = l & PCI_BASE_ADDRESS_MEM_MASK;
226 sz64 = sz & PCI_BASE_ADDRESS_MEM_MASK;
227 mask64 = (u32)PCI_BASE_ADDRESS_MEM_MASK;
228 }
229 } else {
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;
235 }
236
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);
242
243 l64 |= ((u64)l << 32);
244 sz64 |= ((u64)sz << 32);
245 mask64 |= ((u64)~0 << 32);
246 }
247
248 if (!dev->mmio_always_on && (orig_cmd & PCI_COMMAND_DECODE_ENABLE))
249 pci_write_config_word(dev, PCI_COMMAND, orig_cmd);
250
251 if (!sz64)
252 goto fail;
253
254 sz64 = pci_size(l64, sz64, mask64);
255 if (!sz64) {
256 pci_info(dev, FW_BUG "reg 0x%x: invalid BAR (can't size)\n",
257 pos);
258 goto fail;
259 }
260
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;
265 res->start = 0;
266 res->end = 0;
267 pci_err(dev, "reg 0x%x: can't handle BAR larger than 4GB (size %#010llx)\n",
268 pos, (unsigned long long)sz64);
269 goto out;
270 }
271
272 if ((sizeof(pci_bus_addr_t) < 8) && l) {
273 /* Above 32-bit boundary; try to reallocate */
274 res->flags |= IORESOURCE_UNSET;
275 res->start = 0;
276 res->end = sz64 - 1;
277 pci_info(dev, "reg 0x%x: can't handle BAR above 4GB (bus address %#010llx)\n",
278 pos, (unsigned long long)l64);
279 goto out;
280 }
281 }
282
283 region.start = l64;
284 region.end = l64 + sz64 - 1;
285
286 pcibios_bus_to_resource(dev->bus, res, &region);
287 pcibios_resource_to_bus(dev->bus, &inverted_region, res);
288
289 /*
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:
294 *
295 * resource_to_bus(bus_to_resource(A)) == A
296 *
297 * If it doesn't, CPU accesses to "bus_to_resource(A)" will not
298 * be claimed by the device.
299 */
300 if (inverted_region.start != region.start) {
301 res->flags |= IORESOURCE_UNSET;
302 res->start = 0;
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);
306 }
307
308 goto out;
309
310
311fail:
312 res->flags = 0;
313out:
314 if (res->flags)
315 pci_info(dev, "reg 0x%x: %pR\n", pos, res);
316
317 return (res->flags & IORESOURCE_MEM_64) ? 1 : 0;
318}
319
320static void pci_read_bases(struct pci_dev *dev, unsigned int howmany, int rom)
321{
322 unsigned int pos, reg;
323
324 if (dev->non_compliant_bars)
325 return;
326
327 /* Per PCIe r4.0, sec 9.3.4.1.11, the VF BARs are all RO Zero */
328 if (dev->is_virtfn)
329 return;
330
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);
335 }
336
337 if (rom) {
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);
343 }
344}
345
346static void pci_read_bridge_windows(struct pci_dev *bridge)
347{
348 u16 io;
349 u32 pmem, tmp;
350
351 pci_read_config_word(bridge, PCI_IO_BASE, &io);
352 if (!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);
356 }
357 if (io)
358 bridge->io_window = 1;
359
360 /*
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.
364 */
365 if (bridge->vendor == PCI_VENDOR_ID_DEC && bridge->device == 0x0001)
366 return;
367
368 pci_read_config_dword(bridge, PCI_PREF_MEMORY_BASE, &pmem);
369 if (!pmem) {
370 pci_write_config_dword(bridge, PCI_PREF_MEMORY_BASE,
371 0xffe0fff0);
372 pci_read_config_dword(bridge, PCI_PREF_MEMORY_BASE, &pmem);
373 pci_write_config_dword(bridge, PCI_PREF_MEMORY_BASE, 0x0);
374 }
375 if (!pmem)
376 return;
377
378 bridge->pref_window = 1;
379
380 if ((pmem & PCI_PREF_RANGE_TYPE_MASK) == PCI_PREF_RANGE_TYPE_64) {
381
382 /*
383 * Bridge claims to have a 64-bit prefetchable memory
384 * window; verify that the upper bits are actually
385 * writable.
386 */
387 pci_read_config_dword(bridge, PCI_PREF_BASE_UPPER32, &pmem);
388 pci_write_config_dword(bridge, PCI_PREF_BASE_UPPER32,
389 0xffffffff);
390 pci_read_config_dword(bridge, PCI_PREF_BASE_UPPER32, &tmp);
391 pci_write_config_dword(bridge, PCI_PREF_BASE_UPPER32, pmem);
392 if (tmp)
393 bridge->pref_64_window = 1;
394 }
395}
396
397static void pci_read_bridge_io(struct pci_bus *child)
398{
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;
404
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;
411 }
412
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;
418
419 if ((io_base_lo & PCI_IO_RANGE_TYPE_MASK) == PCI_IO_RANGE_TYPE_32) {
420 u16 io_base_hi, io_limit_hi;
421
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);
426 }
427
428 if (base <= limit) {
429 res->flags = (io_base_lo & PCI_IO_RANGE_TYPE_MASK) | IORESOURCE_IO;
430 region.start = base;
431 region.end = limit + io_granularity - 1;
432 pcibios_bus_to_resource(dev->bus, res, &region);
433 pci_info(dev, " bridge window %pR\n", res);
434 }
435}
436
437static void pci_read_bridge_mmio(struct pci_bus *child)
438{
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;
444
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;
450 if (base <= limit) {
451 res->flags = (mem_base_lo & PCI_MEMORY_RANGE_TYPE_MASK) | IORESOURCE_MEM;
452 region.start = base;
453 region.end = limit + 0xfffff;
454 pcibios_bus_to_resource(dev->bus, res, &region);
455 pci_info(dev, " bridge window %pR\n", res);
456 }
457}
458
459static void pci_read_bridge_mmio_pref(struct pci_bus *child)
460{
461 struct pci_dev *dev = child->self;
462 u16 mem_base_lo, mem_limit_lo;
463 u64 base64, limit64;
464 pci_bus_addr_t base, limit;
465 struct pci_bus_region region;
466 struct resource *res;
467
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;
473
474 if ((mem_base_lo & PCI_PREF_RANGE_TYPE_MASK) == PCI_PREF_RANGE_TYPE_64) {
475 u32 mem_base_hi, mem_limit_hi;
476
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);
479
480 /*
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.
484 */
485 if (mem_base_hi <= mem_limit_hi) {
486 base64 |= (u64) mem_base_hi << 32;
487 limit64 |= (u64) mem_limit_hi << 32;
488 }
489 }
490
491 base = (pci_bus_addr_t) base64;
492 limit = (pci_bus_addr_t) limit64;
493
494 if (base != base64) {
495 pci_err(dev, "can't handle bridge window above 4GB (bus address %#010llx)\n",
496 (unsigned long long) base64);
497 return;
498 }
499
500 if (base <= limit) {
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;
505 region.start = base;
506 region.end = limit + 0xfffff;
507 pcibios_bus_to_resource(dev->bus, res, &region);
508 pci_info(dev, " bridge window %pR\n", res);
509 }
510}
511
512void pci_read_bridge_bases(struct pci_bus *child)
513{
514 struct pci_dev *dev = child->self;
515 struct resource *res;
516 int i;
517
518 if (pci_is_root_bus(child)) /* It's a host bus, nothing to read */
519 return;
520
521 pci_info(dev, "PCI bridge to %pR%s\n",
522 &child->busn_res,
523 dev->transparent ? " (subtractive decode)" : "");
524
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];
528
529 pci_read_bridge_io(child);
530 pci_read_bridge_mmio(child);
531 pci_read_bridge_mmio_pref(child);
532
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",
539 res);
540 }
541 }
542 }
543}
544
545static struct pci_bus *pci_alloc_bus(struct pci_bus *parent)
546{
547 struct pci_bus *b;
548
549 b = kzalloc(sizeof(*b), GFP_KERNEL);
550 if (!b)
551 return NULL;
552
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
561 if (parent)
562 b->domain_nr = parent->domain_nr;
563#endif
564 return b;
565}
566
567static void pci_release_host_bridge_dev(struct device *dev)
568{
569 struct pci_host_bridge *bridge = to_pci_host_bridge(dev);
570
571 if (bridge->release_fn)
572 bridge->release_fn(bridge);
573
574 pci_free_resource_list(&bridge->windows);
575 pci_free_resource_list(&bridge->dma_ranges);
576 kfree(bridge);
577}
578
579static void pci_init_host_bridge(struct pci_host_bridge *bridge)
580{
581 INIT_LIST_HEAD(&bridge->windows);
582 INIT_LIST_HEAD(&bridge->dma_ranges);
583
584 /*
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.
589 */
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;
595
596 device_initialize(&bridge->dev);
597}
598
599struct pci_host_bridge *pci_alloc_host_bridge(size_t priv)
600{
601 struct pci_host_bridge *bridge;
602
603 bridge = kzalloc(sizeof(*bridge) + priv, GFP_KERNEL);
604 if (!bridge)
605 return NULL;
606
607 pci_init_host_bridge(bridge);
608 bridge->dev.release = pci_release_host_bridge_dev;
609
610 return bridge;
611}
612EXPORT_SYMBOL(pci_alloc_host_bridge);
613
614static void devm_pci_alloc_host_bridge_release(void *data)
615{
616 pci_free_host_bridge(data);
617}
618
619struct pci_host_bridge *devm_pci_alloc_host_bridge(struct device *dev,
620 size_t priv)
621{
622 int ret;
623 struct pci_host_bridge *bridge;
624
625 bridge = pci_alloc_host_bridge(priv);
626 if (!bridge)
627 return NULL;
628
629 ret = devm_add_action_or_reset(dev, devm_pci_alloc_host_bridge_release,
630 bridge);
631 if (ret)
632 return NULL;
633
634 return bridge;
635}
636EXPORT_SYMBOL(devm_pci_alloc_host_bridge);
637
638void pci_free_host_bridge(struct pci_host_bridge *bridge)
639{
640 put_device(&bridge->dev);
641}
642EXPORT_SYMBOL(pci_free_host_bridge);
643
644static 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 */
661};
662
663const 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 */
680};
681
682void pcie_update_link_speed(struct pci_bus *bus, u16 linksta)
683{
684 bus->cur_bus_speed = pcie_link_speed[linksta & PCI_EXP_LNKSTA_CLS];
685}
686EXPORT_SYMBOL_GPL(pcie_update_link_speed);
687
688static unsigned char agp_speeds[] = {
689 AGP_UNKNOWN,
690 AGP_1X,
691 AGP_2X,
692 AGP_4X,
693 AGP_8X
694};
695
696static enum pci_bus_speed agp_speed(int agp3, int agpstat)
697{
698 int index = 0;
699
700 if (agpstat & 4)
701 index = 3;
702 else if (agpstat & 2)
703 index = 2;
704 else if (agpstat & 1)
705 index = 1;
706 else
707 goto out;
708
709 if (agp3) {
710 index += 2;
711 if (index == 5)
712 index = 0;
713 }
714
715 out:
716 return agp_speeds[index];
717}
718
719static void pci_set_bus_speed(struct pci_bus *bus)
720{
721 struct pci_dev *bridge = bus->self;
722 int pos;
723
724 pos = pci_find_capability(bridge, PCI_CAP_ID_AGP);
725 if (!pos)
726 pos = pci_find_capability(bridge, PCI_CAP_ID_AGP3);
727 if (pos) {
728 u32 agpstat, agpcmd;
729
730 pci_read_config_dword(bridge, pos + PCI_AGP_STATUS, &agpstat);
731 bus->max_bus_speed = agp_speed(agpstat & 8, agpstat & 7);
732
733 pci_read_config_dword(bridge, pos + PCI_AGP_COMMAND, &agpcmd);
734 bus->cur_bus_speed = agp_speed(agpstat & 8, agpcmd & 7);
735 }
736
737 pos = pci_find_capability(bridge, PCI_CAP_ID_PCIX);
738 if (pos) {
739 u16 status;
740 enum pci_bus_speed max;
741
742 pci_read_config_word(bridge, pos + PCI_X_BRIDGE_SSTATUS,
743 &status);
744
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;
752 else
753 max = PCI_SPEED_133MHz_PCIX;
754 } else {
755 max = PCI_SPEED_66MHz_PCIX;
756 }
757
758 bus->max_bus_speed = max;
759 bus->cur_bus_speed = pcix_bus_speed[
760 (status & PCI_X_SSTATUS_FREQ) >> 6];
761
762 return;
763 }
764
765 if (pci_is_pcie(bridge)) {
766 u32 linkcap;
767 u16 linksta;
768
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);
772
773 pcie_capability_read_word(bridge, PCI_EXP_LNKSTA, &linksta);
774 pcie_update_link_speed(bus, linksta);
775 }
776}
777
778static struct irq_domain *pci_host_bridge_msi_domain(struct pci_bus *bus)
779{
780 struct irq_domain *d;
781
782 /*
783 * Any firmware interface that can resolve the msi_domain
784 * should be called from here.
785 */
786 d = pci_host_bridge_of_msi_domain(bus);
787 if (!d)
788 d = pci_host_bridge_acpi_msi_domain(bus);
789
790#ifdef CONFIG_PCI_MSI_IRQ_DOMAIN
791 /*
792 * If no IRQ domain was found via the OF tree, try looking it up
793 * directly through the fwnode_handle.
794 */
795 if (!d) {
796 struct fwnode_handle *fwnode = pci_root_bus_fwnode(bus);
797
798 if (fwnode)
799 d = irq_find_matching_fwnode(fwnode,
800 DOMAIN_BUS_PCI_MSI);
801 }
802#endif
803
804 return d;
805}
806
807static void pci_set_bus_msi_domain(struct pci_bus *bus)
808{
809 struct irq_domain *d;
810 struct pci_bus *b;
811
812 /*
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.
816 */
817 for (b = bus, d = NULL; !d && !pci_is_root_bus(b); b = b->parent) {
818 if (b->self)
819 d = dev_get_msi_domain(&b->self->dev);
820 }
821
822 if (!d)
823 d = pci_host_bridge_msi_domain(b);
824
825 dev_set_msi_domain(&bus->dev, d);
826}
827
828static int pci_register_host_bridge(struct pci_host_bridge *bridge)
829{
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;
836 char addr[64], *fmt;
837 const char *name;
838 int err;
839
840 bus = pci_alloc_bus(NULL);
841 if (!bus)
842 return -ENOMEM;
843
844 bridge->bus = bus;
845
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);
854#endif
855
856 b = pci_find_bus(pci_domain_nr(bus), bridge->busnr);
857 if (b) {
858 /* Ignore it if we already got here via a different bridge */
859 dev_dbg(&b->dev, "bus already known\n");
860 err = -EEXIST;
861 goto free;
862 }
863
864 dev_set_name(&bridge->dev, "pci%04x:%02x", pci_domain_nr(bus),
865 bridge->busnr);
866
867 err = pcibios_root_bridge_prepare(bridge);
868 if (err)
869 goto free;
870
871 err = device_add(&bridge->dev);
872 if (err) {
873 put_device(&bridge->dev);
874 goto free;
875 }
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);
880
881 if (!parent)
882 set_dev_node(bus->bridge, pcibus_to_node(bus));
883
884 bus->dev.class = &pcibus_class;
885 bus->dev.parent = bus->bridge;
886
887 dev_set_name(&bus->dev, "%04x:%02x", pci_domain_nr(bus), bus->number);
888 name = dev_name(&bus->dev);
889
890 err = device_register(&bus->dev);
891 if (err)
892 goto unregister;
893
894 pcibios_add_bus(bus);
895
896 /* Create legacy_io and legacy_mem files for this bus */
897 pci_create_legacy_files(bus);
898
899 if (parent)
900 dev_info(parent, "PCI host bridge to bus %s\n", name);
901 else
902 pr_info("PCI host bridge to bus %s\n", name);
903
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;
908 res = window->res;
909
910 if (res->flags & IORESOURCE_BUS)
911 pci_bus_insert_busn_res(bus, bus->number, res->end);
912 else
913 pci_bus_add_resource(bus, res, 0);
914
915 if (offset) {
916 if (resource_type(res) == IORESOURCE_IO)
917 fmt = " (bus address [%#06llx-%#06llx])";
918 else
919 fmt = " (bus address [%#010llx-%#010llx])";
920
921 snprintf(addr, sizeof(addr), fmt,
922 (unsigned long long)(res->start - offset),
923 (unsigned long long)(res->end - offset));
924 } else
925 addr[0] = '\0';
926
927 dev_info(&bus->dev, "root bus resource %pR%s\n", res, addr);
928 }
929
930 down_write(&pci_bus_sem);
931 list_add_tail(&bus->node, &pci_root_buses);
932 up_write(&pci_bus_sem);
933
934 return 0;
935
936unregister:
937 put_device(&bridge->dev);
938 device_del(&bridge->dev);
939
940free:
941 kfree(bus);
942 return err;
943}
944
945static bool pci_bridge_child_ext_cfg_accessible(struct pci_dev *bridge)
946{
947 int pos;
948 u32 status;
949
950 /*
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.
953 */
954 if (bridge->bus->bus_flags & PCI_BUS_FLAGS_NO_EXTCFG)
955 return false;
956
957 /*
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.
961 */
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))
966 return true;
967
968 /*
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.
975 */
976 pos = pci_find_capability(bridge, PCI_CAP_ID_PCIX);
977 if (!pos)
978 return false;
979
980 pci_read_config_dword(bridge, pos + PCI_X_STATUS, &status);
981 return status & (PCI_X_STATUS_266MHZ | PCI_X_STATUS_533MHZ);
982}
983
984static struct pci_bus *pci_alloc_child_bus(struct pci_bus *parent,
985 struct pci_dev *bridge, int busnr)
986{
987 struct pci_bus *child;
988 int i;
989 int ret;
990
991 /* Allocate a new bus and inherit stuff from the parent */
992 child = pci_alloc_bus(parent);
993 if (!child)
994 return NULL;
995
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;
1001
1002 /*
1003 * Initialize some portions of the bus device, but don't register
1004 * it now as the parent is not properly set up yet.
1005 */
1006 child->dev.class = &pcibus_class;
1007 dev_set_name(&child->dev, "%04x:%02x", pci_domain_nr(child), busnr);
1008
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;
1013
1014 if (!bridge) {
1015 child->dev.parent = parent->bridge;
1016 goto add_dev;
1017 }
1018
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);
1024
1025 /*
1026 * Check whether extended config space is accessible on the child
1027 * bus. Note that we currently assume it is always accessible on
1028 * the root bus.
1029 */
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");
1033 }
1034
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;
1039 }
1040 bridge->subordinate = child;
1041
1042add_dev:
1043 pci_set_bus_msi_domain(child);
1044 ret = device_register(&child->dev);
1045 WARN_ON(ret < 0);
1046
1047 pcibios_add_bus(child);
1048
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);
1053 }
1054
1055 /* Create legacy_io and legacy_mem files for this bus */
1056 pci_create_legacy_files(child);
1057
1058 return child;
1059}
1060
1061struct pci_bus *pci_add_new_bus(struct pci_bus *parent, struct pci_dev *dev,
1062 int busnr)
1063{
1064 struct pci_bus *child;
1065
1066 child = pci_alloc_child_bus(parent, dev, busnr);
1067 if (child) {
1068 down_write(&pci_bus_sem);
1069 list_add_tail(&child->node, &parent->children);
1070 up_write(&pci_bus_sem);
1071 }
1072 return child;
1073}
1074EXPORT_SYMBOL(pci_add_new_bus);
1075
1076static void pci_enable_crs(struct pci_dev *pdev)
1077{
1078 u16 root_cap = 0;
1079
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);
1085}
1086
1087static unsigned int pci_scan_child_bus_extend(struct pci_bus *bus,
1088 unsigned int available_buses);
1089/**
1090 * pci_ea_fixed_busnrs() - Read fixed Secondary and Subordinate bus
1091 * numbers from EA capability.
1092 * @dev: Bridge
1093 * @sec: updated with secondary bus number from EA
1094 * @sub: updated with subordinate bus number from EA
1095 *
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.
1099 */
1100static bool pci_ea_fixed_busnrs(struct pci_dev *dev, u8 *sec, u8 *sub)
1101{
1102 int ea, offset;
1103 u32 dw;
1104 u8 ea_sec, ea_sub;
1105
1106 if (dev->hdr_type != PCI_HEADER_TYPE_BRIDGE)
1107 return false;
1108
1109 /* find PCI EA capability in list */
1110 ea = pci_find_capability(dev, PCI_CAP_ID_EA);
1111 if (!ea)
1112 return false;
1113
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)
1119 return false;
1120
1121 *sec = ea_sec;
1122 *sub = ea_sub;
1123 return true;
1124}
1125
1126/*
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.
1137 *
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.
1141 *
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.
1146 *
1147 * Return: New subordinate number covering all buses behind this bridge.
1148 */
1149static int pci_scan_bridge_extend(struct pci_bus *bus, struct pci_dev *dev,
1150 int max, unsigned int available_buses,
1151 int pass)
1152{
1153 struct pci_bus *child;
1154 int is_cardbus = (dev->hdr_type == PCI_HEADER_TYPE_CARDBUS);
1155 u32 buses, i, j = 0;
1156 u16 bctl;
1157 u8 primary, secondary, subordinate;
1158 int broken = 0;
1159 bool fixed_buses;
1160 u8 fixed_sec, fixed_sub;
1161 int next_busnr;
1162
1163 /*
1164 * Make sure the bridge is powered on to be able to access config
1165 * space of devices below it.
1166 */
1167 pm_runtime_get_sync(&dev->dev);
1168
1169 pci_read_config_dword(dev, PCI_PRIMARY_BUS, &buses);
1170 primary = buses & 0xFF;
1171 secondary = (buses >> 8) & 0xFF;
1172 subordinate = (buses >> 16) & 0xFF;
1173
1174 pci_dbg(dev, "scanning [bus %02x-%02x] behind bridge, pass %d\n",
1175 secondary, subordinate, pass);
1176
1177 if (!primary && (primary != bus->number) && secondary && subordinate) {
1178 pci_warn(dev, "Primary bus is hard wired to 0\n");
1179 primary = bus->number;
1180 }
1181
1182 /* Check if setup is sensible at all */
1183 if (!pass &&
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);
1188 broken = 1;
1189 }
1190
1191 /*
1192 * Disable Master-Abort Mode during probing to avoid reporting of
1193 * bus errors in some architectures.
1194 */
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);
1198
1199 pci_enable_crs(dev);
1200
1201 if ((secondary || subordinate) && !pcibios_assign_all_busses() &&
1202 !is_cardbus && !broken) {
1203 unsigned int cmax;
1204
1205 /*
1206 * Bus already configured by firmware, process it in the
1207 * first pass and just note the configuration.
1208 */
1209 if (pass)
1210 goto out;
1211
1212 /*
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.
1217 */
1218 child = pci_find_bus(pci_domain_nr(bus), secondary);
1219 if (!child) {
1220 child = pci_add_new_bus(bus, dev, secondary);
1221 if (!child)
1222 goto out;
1223 child->primary = primary;
1224 pci_bus_insert_busn_res(child, secondary, subordinate);
1225 child->bridge_ctl = bctl;
1226 }
1227
1228 cmax = pci_scan_child_bus(child);
1229 if (cmax > subordinate)
1230 pci_warn(dev, "bridge has subordinate %02x but max busn %02x\n",
1231 subordinate, cmax);
1232
1233 /* Subordinate should equal child->busn_res.end */
1234 if (subordinate > max)
1235 max = subordinate;
1236 } else {
1237
1238 /*
1239 * We need to assign a number to this bus which we always
1240 * do in the second pass.
1241 */
1242 if (!pass) {
1243 if (pcibios_assign_all_busses() || broken || is_cardbus)
1244
1245 /*
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
1251 * ranges.
1252 */
1253 pci_write_config_dword(dev, PCI_PRIMARY_BUS,
1254 buses & ~0xffffff);
1255 goto out;
1256 }
1257
1258 /* Clear errors */
1259 pci_write_config_word(dev, PCI_STATUS, 0xffff);
1260
1261 /* Read bus numbers from EA Capability (if present) */
1262 fixed_buses = pci_ea_fixed_busnrs(dev, &fixed_sec, &fixed_sub);
1263 if (fixed_buses)
1264 next_busnr = fixed_sec;
1265 else
1266 next_busnr = max + 1;
1267
1268 /*
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.
1272 */
1273 child = pci_find_bus(pci_domain_nr(bus), next_busnr);
1274 if (!child) {
1275 child = pci_add_new_bus(bus, dev, next_busnr);
1276 if (!child)
1277 goto out;
1278 pci_bus_insert_busn_res(child, next_busnr,
1279 bus->busn_res.end);
1280 }
1281 max++;
1282 if (available_buses)
1283 available_buses--;
1284
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);
1289
1290 /*
1291 * yenta.c forces a secondary latency timer of 176.
1292 * Copy that behaviour here.
1293 */
1294 if (is_cardbus) {
1295 buses &= ~0xff000000;
1296 buses |= CARDBUS_LATENCY_TIMER << 24;
1297 }
1298
1299 /* We need to blast all three values with a single write */
1300 pci_write_config_dword(dev, PCI_PRIMARY_BUS, buses);
1301
1302 if (!is_cardbus) {
1303 child->bridge_ctl = bctl;
1304 max = pci_scan_child_bus_extend(child, available_buses);
1305 } else {
1306
1307 /*
1308 * For CardBus bridges, we leave 4 bus numbers as
1309 * cards with a PCI-to-PCI bridge can be inserted
1310 * later.
1311 */
1312 for (i = 0; i < CARDBUS_RESERVE_BUSNR; i++) {
1313 struct pci_bus *parent = bus;
1314 if (pci_find_bus(pci_domain_nr(bus),
1315 max+i+1))
1316 break;
1317 while (parent->parent) {
1318 if ((!pcibios_assign_all_busses()) &&
1319 (parent->busn_res.end > max) &&
1320 (parent->busn_res.end <= max+i)) {
1321 j = 1;
1322 }
1323 parent = parent->parent;
1324 }
1325 if (j) {
1326
1327 /*
1328 * Often, there are two CardBus
1329 * bridges -- try to leave one
1330 * valid bus number for each one.
1331 */
1332 i /= 2;
1333 break;
1334 }
1335 }
1336 max += i;
1337 }
1338
1339 /*
1340 * Set subordinate bus number to its real value.
1341 * If fixed subordinate bus number exists from EA
1342 * capability then use it.
1343 */
1344 if (fixed_buses)
1345 max = fixed_sub;
1346 pci_bus_update_busn_res_end(child, max);
1347 pci_write_config_byte(dev, PCI_SUBORDINATE_BUS, max);
1348 }
1349
1350 sprintf(child->name,
1351 (is_cardbus ? "PCI CardBus %04x:%02x" : "PCI Bus %04x:%02x"),
1352 pci_domain_nr(bus), child->number);
1353
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",
1361 &child->busn_res);
1362 break;
1363 }
1364 bus = bus->parent;
1365 }
1366
1367out:
1368 pci_write_config_word(dev, PCI_BRIDGE_CONTROL, bctl);
1369
1370 pm_runtime_put(&dev->dev);
1371
1372 return max;
1373}
1374
1375/*
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.
1382 *
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.
1386 *
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.
1391 *
1392 * Return: New subordinate number covering all buses behind this bridge.
1393 */
1394int pci_scan_bridge(struct pci_bus *bus, struct pci_dev *dev, int max, int pass)
1395{
1396 return pci_scan_bridge_extend(bus, dev, max, 0, pass);
1397}
1398EXPORT_SYMBOL(pci_scan_bridge);
1399
1400/*
1401 * Read interrupt line and base address registers.
1402 * The architecture-dependent code can tweak these, of course.
1403 */
1404static void pci_read_irq(struct pci_dev *dev)
1405{
1406 unsigned char irq;
1407
1408 /* VFs are not allowed to use INTx, so skip the config reads */
1409 if (dev->is_virtfn) {
1410 dev->pin = 0;
1411 dev->irq = 0;
1412 return;
1413 }
1414
1415 pci_read_config_byte(dev, PCI_INTERRUPT_PIN, &irq);
1416 dev->pin = irq;
1417 if (irq)
1418 pci_read_config_byte(dev, PCI_INTERRUPT_LINE, &irq);
1419 dev->irq = irq;
1420}
1421
1422void set_pcie_port_type(struct pci_dev *pdev)
1423{
1424 int pos;
1425 u16 reg16;
1426 int type;
1427 struct pci_dev *parent;
1428
1429 pos = pci_find_capability(pdev, PCI_CAP_ID_EXP);
1430 if (!pos)
1431 return;
1432
1433 pdev->pcie_cap = pos;
1434 pci_read_config_word(pdev, pos + PCI_EXP_FLAGS, &reg16);
1435 pdev->pcie_flags_reg = reg16;
1436 pci_read_config_word(pdev, pos + PCI_EXP_DEVCAP, &reg16);
1437 pdev->pcie_mpss = reg16 & PCI_EXP_DEVCAP_PAYLOAD;
1438
1439 parent = pci_upstream_bridge(pdev);
1440 if (!parent)
1441 return;
1442
1443 /*
1444 * Some systems do not identify their upstream/downstream ports
1445 * correctly so detect impossible configurations here and correct
1446 * the port type accordingly.
1447 */
1448 type = pci_pcie_type(pdev);
1449 if (type == PCI_EXP_TYPE_DOWNSTREAM) {
1450 /*
1451 * If pdev claims to be downstream port but the parent
1452 * device is also downstream port assume pdev is actually
1453 * upstream port.
1454 */
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;
1459 }
1460 } else if (type == PCI_EXP_TYPE_UPSTREAM) {
1461 /*
1462 * If pdev claims to be upstream port but the parent
1463 * device is also upstream port assume pdev is actually
1464 * downstream port.
1465 */
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;
1470 }
1471 }
1472}
1473
1474void set_pcie_hotplug_bridge(struct pci_dev *pdev)
1475{
1476 u32 reg32;
1477
1478 pcie_capability_read_dword(pdev, PCI_EXP_SLTCAP, &reg32);
1479 if (reg32 & PCI_EXP_SLTCAP_HPC)
1480 pdev->is_hotplug_bridge = 1;
1481}
1482
1483static void set_pcie_thunderbolt(struct pci_dev *dev)
1484{
1485 int vsec = 0;
1486 u32 header;
1487
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);
1491
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;
1496 return;
1497 }
1498 }
1499}
1500
1501static void set_pcie_untrusted(struct pci_dev *dev)
1502{
1503 struct pci_dev *parent;
1504
1505 /*
1506 * If the upstream bridge is untrusted we treat this device
1507 * untrusted as well.
1508 */
1509 parent = pci_upstream_bridge(dev);
1510 if (parent && parent->untrusted)
1511 dev->untrusted = true;
1512}
1513
1514/**
1515 * pci_ext_cfg_is_aliased - Is ext config space just an alias of std config?
1516 * @dev: PCI device
1517 *
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.
1526 * Known offenders:
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)
1529 */
1530static bool pci_ext_cfg_is_aliased(struct pci_dev *dev)
1531{
1532#ifdef CONFIG_PCI_QUIRKS
1533 int pos;
1534 u32 header, tmp;
1535
1536 pci_read_config_dword(dev, PCI_VENDOR_ID, &header);
1537
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
1541 || header != tmp)
1542 return false;
1543 }
1544
1545 return true;
1546#else
1547 return false;
1548#endif
1549}
1550
1551/**
1552 * pci_cfg_space_size - Get the configuration space size of the PCI device
1553 * @dev: PCI device
1554 *
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.
1561 */
1562static int pci_cfg_space_size_ext(struct pci_dev *dev)
1563{
1564 u32 status;
1565 int pos = PCI_CFG_SPACE_SIZE;
1566
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;
1571
1572 return PCI_CFG_SPACE_EXP_SIZE;
1573}
1574
1575int pci_cfg_space_size(struct pci_dev *dev)
1576{
1577 int pos;
1578 u32 status;
1579 u16 class;
1580
1581#ifdef CONFIG_PCI_IOV
1582 /*
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.
1591 */
1592 if (dev->is_virtfn)
1593 return PCI_CFG_SPACE_EXP_SIZE;
1594#endif
1595
1596 if (dev->bus->bus_flags & PCI_BUS_FLAGS_NO_EXTCFG)
1597 return PCI_CFG_SPACE_SIZE;
1598
1599 class = dev->class >> 8;
1600 if (class == PCI_CLASS_BRIDGE_HOST)
1601 return pci_cfg_space_size_ext(dev);
1602
1603 if (pci_is_pcie(dev))
1604 return pci_cfg_space_size_ext(dev);
1605
1606 pos = pci_find_capability(dev, PCI_CAP_ID_PCIX);
1607 if (!pos)
1608 return PCI_CFG_SPACE_SIZE;
1609
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);
1613
1614 return PCI_CFG_SPACE_SIZE;
1615}
1616
1617static u32 pci_class(struct pci_dev *dev)
1618{
1619 u32 class;
1620
1621#ifdef CONFIG_PCI_IOV
1622 if (dev->is_virtfn)
1623 return dev->physfn->sriov->class;
1624#endif
1625 pci_read_config_dword(dev, PCI_CLASS_REVISION, &class);
1626 return class;
1627}
1628
1629static void pci_subsystem_ids(struct pci_dev *dev, u16 *vendor, u16 *device)
1630{
1631#ifdef CONFIG_PCI_IOV
1632 if (dev->is_virtfn) {
1633 *vendor = dev->physfn->sriov->subsystem_vendor;
1634 *device = dev->physfn->sriov->subsystem_device;
1635 return;
1636 }
1637#endif
1638 pci_read_config_word(dev, PCI_SUBSYSTEM_VENDOR_ID, vendor);
1639 pci_read_config_word(dev, PCI_SUBSYSTEM_ID, device);
1640}
1641
1642static u8 pci_hdr_type(struct pci_dev *dev)
1643{
1644 u8 hdr_type;
1645
1646#ifdef CONFIG_PCI_IOV
1647 if (dev->is_virtfn)
1648 return dev->physfn->sriov->hdr_type;
1649#endif
1650 pci_read_config_byte(dev, PCI_HEADER_TYPE, &hdr_type);
1651#ifdef CONFIG_PCIE_ASR1803
1652 /* for pcie8098, header type is 0x80 multifunction
1653 device, we set to normal device */
1654 if (hdr_type == 0x80)
1655 hdr_type = 0;
1656#endif
1657 return hdr_type;
1658}
1659
1660#define LEGACY_IO_RESOURCE (IORESOURCE_IO | IORESOURCE_PCI_FIXED)
1661
1662static void pci_msi_setup_pci_dev(struct pci_dev *dev)
1663{
1664 /*
1665 * Disable the MSI hardware to avoid screaming interrupts
1666 * during boot. This is the power on reset default so
1667 * usually this should be a noop.
1668 */
1669 dev->msi_cap = pci_find_capability(dev, PCI_CAP_ID_MSI);
1670 if (dev->msi_cap)
1671 pci_msi_set_enable(dev, 0);
1672
1673 dev->msix_cap = pci_find_capability(dev, PCI_CAP_ID_MSIX);
1674 if (dev->msix_cap)
1675 pci_msix_clear_and_set_ctrl(dev, PCI_MSIX_FLAGS_ENABLE, 0);
1676}
1677
1678/**
1679 * pci_intx_mask_broken - Test PCI_COMMAND_INTX_DISABLE writability
1680 * @dev: PCI device
1681 *
1682 * Test whether PCI_COMMAND_INTX_DISABLE is writable for @dev. Check this
1683 * at enumeration-time to avoid modifying PCI_COMMAND at run-time.
1684 */
1685static int pci_intx_mask_broken(struct pci_dev *dev)
1686{
1687 u16 orig, toggle, new;
1688
1689 pci_read_config_word(dev, PCI_COMMAND, &orig);
1690 toggle = orig ^ PCI_COMMAND_INTX_DISABLE;
1691 pci_write_config_word(dev, PCI_COMMAND, toggle);
1692 pci_read_config_word(dev, PCI_COMMAND, &new);
1693
1694 pci_write_config_word(dev, PCI_COMMAND, orig);
1695
1696 /*
1697 * PCI_COMMAND_INTX_DISABLE was reserved and read-only prior to PCI
1698 * r2.3, so strictly speaking, a device is not *broken* if it's not
1699 * writable. But we'll live with the misnomer for now.
1700 */
1701 if (new != toggle)
1702 return 1;
1703 return 0;
1704}
1705
1706static void early_dump_pci_device(struct pci_dev *pdev)
1707{
1708 u32 value[256 / 4];
1709 int i;
1710
1711 pci_info(pdev, "config space:\n");
1712
1713 for (i = 0; i < 256; i += 4)
1714 pci_read_config_dword(pdev, i, &value[i / 4]);
1715
1716 print_hex_dump(KERN_INFO, "", DUMP_PREFIX_OFFSET, 16, 1,
1717 value, 256, false);
1718}
1719
1720/**
1721 * pci_setup_device - Fill in class and map information of a device
1722 * @dev: the device structure to fill
1723 *
1724 * Initialize the device structure with information about the device's
1725 * vendor,class,memory and IO-space addresses, IRQ lines etc.
1726 * Called at initialisation of the PCI subsystem and by CardBus services.
1727 * Returns 0 on success and negative if unknown type of device (not normal,
1728 * bridge or CardBus).
1729 */
1730int pci_setup_device(struct pci_dev *dev)
1731{
1732 u32 class;
1733 u16 cmd;
1734 u8 hdr_type;
1735 int pos = 0;
1736 struct pci_bus_region region;
1737 struct resource *res;
1738
1739 hdr_type = pci_hdr_type(dev);
1740
1741 dev->sysdata = dev->bus->sysdata;
1742 dev->dev.parent = dev->bus->bridge;
1743 dev->dev.bus = &pci_bus_type;
1744 dev->hdr_type = hdr_type & 0x7f;
1745 dev->multifunction = !!(hdr_type & 0x80);
1746 dev->error_state = pci_channel_io_normal;
1747 set_pcie_port_type(dev);
1748
1749 pci_dev_assign_slot(dev);
1750
1751 /*
1752 * Assume 32-bit PCI; let 64-bit PCI cards (which are far rarer)
1753 * set this higher, assuming the system even supports it.
1754 */
1755 dev->dma_mask = 0xffffffff;
1756
1757 dev_set_name(&dev->dev, "%04x:%02x:%02x.%d", pci_domain_nr(dev->bus),
1758 dev->bus->number, PCI_SLOT(dev->devfn),
1759 PCI_FUNC(dev->devfn));
1760
1761 class = pci_class(dev);
1762
1763 dev->revision = class & 0xff;
1764 dev->class = class >> 8; /* upper 3 bytes */
1765
1766 pci_info(dev, "[%04x:%04x] type %02x class %#08x\n",
1767 dev->vendor, dev->device, dev->hdr_type, dev->class);
1768
1769 if (pci_early_dump)
1770 early_dump_pci_device(dev);
1771
1772 /* Need to have dev->class ready */
1773 dev->cfg_size = pci_cfg_space_size(dev);
1774
1775 /* Need to have dev->cfg_size ready */
1776 set_pcie_thunderbolt(dev);
1777
1778 set_pcie_untrusted(dev);
1779
1780 /* "Unknown power state" */
1781 dev->current_state = PCI_UNKNOWN;
1782
1783 /* Early fixups, before probing the BARs */
1784 pci_fixup_device(pci_fixup_early, dev);
1785
1786 /* Device class may be changed after fixup */
1787 class = dev->class >> 8;
1788
1789 if (dev->non_compliant_bars && !dev->mmio_always_on) {
1790 pci_read_config_word(dev, PCI_COMMAND, &cmd);
1791 if (cmd & (PCI_COMMAND_IO | PCI_COMMAND_MEMORY)) {
1792 pci_info(dev, "device has non-compliant BARs; disabling IO/MEM decoding\n");
1793 cmd &= ~PCI_COMMAND_IO;
1794 cmd &= ~PCI_COMMAND_MEMORY;
1795 pci_write_config_word(dev, PCI_COMMAND, cmd);
1796 }
1797 }
1798
1799 dev->broken_intx_masking = pci_intx_mask_broken(dev);
1800
1801 switch (dev->hdr_type) { /* header type */
1802 case PCI_HEADER_TYPE_NORMAL: /* standard header */
1803 if (class == PCI_CLASS_BRIDGE_PCI)
1804 goto bad;
1805 pci_read_irq(dev);
1806 pci_read_bases(dev, 6, PCI_ROM_ADDRESS);
1807
1808 pci_subsystem_ids(dev, &dev->subsystem_vendor, &dev->subsystem_device);
1809
1810 /*
1811 * Do the ugly legacy mode stuff here rather than broken chip
1812 * quirk code. Legacy mode ATA controllers have fixed
1813 * addresses. These are not always echoed in BAR0-3, and
1814 * BAR0-3 in a few cases contain junk!
1815 */
1816 if (class == PCI_CLASS_STORAGE_IDE) {
1817 u8 progif;
1818 pci_read_config_byte(dev, PCI_CLASS_PROG, &progif);
1819 if ((progif & 1) == 0) {
1820 region.start = 0x1F0;
1821 region.end = 0x1F7;
1822 res = &dev->resource[0];
1823 res->flags = LEGACY_IO_RESOURCE;
1824 pcibios_bus_to_resource(dev->bus, res, &region);
1825 pci_info(dev, "legacy IDE quirk: reg 0x10: %pR\n",
1826 res);
1827 region.start = 0x3F6;
1828 region.end = 0x3F6;
1829 res = &dev->resource[1];
1830 res->flags = LEGACY_IO_RESOURCE;
1831 pcibios_bus_to_resource(dev->bus, res, &region);
1832 pci_info(dev, "legacy IDE quirk: reg 0x14: %pR\n",
1833 res);
1834 }
1835 if ((progif & 4) == 0) {
1836 region.start = 0x170;
1837 region.end = 0x177;
1838 res = &dev->resource[2];
1839 res->flags = LEGACY_IO_RESOURCE;
1840 pcibios_bus_to_resource(dev->bus, res, &region);
1841 pci_info(dev, "legacy IDE quirk: reg 0x18: %pR\n",
1842 res);
1843 region.start = 0x376;
1844 region.end = 0x376;
1845 res = &dev->resource[3];
1846 res->flags = LEGACY_IO_RESOURCE;
1847 pcibios_bus_to_resource(dev->bus, res, &region);
1848 pci_info(dev, "legacy IDE quirk: reg 0x1c: %pR\n",
1849 res);
1850 }
1851 }
1852 break;
1853
1854 case PCI_HEADER_TYPE_BRIDGE: /* bridge header */
1855 /*
1856 * The PCI-to-PCI bridge spec requires that subtractive
1857 * decoding (i.e. transparent) bridge must have programming
1858 * interface code of 0x01.
1859 */
1860 pci_read_irq(dev);
1861 dev->transparent = ((dev->class & 0xff) == 1);
1862 pci_read_bases(dev, 2, PCI_ROM_ADDRESS1);
1863 pci_read_bridge_windows(dev);
1864 set_pcie_hotplug_bridge(dev);
1865 pos = pci_find_capability(dev, PCI_CAP_ID_SSVID);
1866 if (pos) {
1867 pci_read_config_word(dev, pos + PCI_SSVID_VENDOR_ID, &dev->subsystem_vendor);
1868 pci_read_config_word(dev, pos + PCI_SSVID_DEVICE_ID, &dev->subsystem_device);
1869 }
1870 break;
1871
1872 case PCI_HEADER_TYPE_CARDBUS: /* CardBus bridge header */
1873 if (class != PCI_CLASS_BRIDGE_CARDBUS)
1874 goto bad;
1875 pci_read_irq(dev);
1876 pci_read_bases(dev, 1, 0);
1877 pci_read_config_word(dev, PCI_CB_SUBSYSTEM_VENDOR_ID, &dev->subsystem_vendor);
1878 pci_read_config_word(dev, PCI_CB_SUBSYSTEM_ID, &dev->subsystem_device);
1879 break;
1880
1881 default: /* unknown header */
1882 pci_err(dev, "unknown header type %02x, ignoring device\n",
1883 dev->hdr_type);
1884 return -EIO;
1885
1886 bad:
1887 pci_err(dev, "ignoring class %#08x (doesn't match header type %02x)\n",
1888 dev->class, dev->hdr_type);
1889 dev->class = PCI_CLASS_NOT_DEFINED << 8;
1890 }
1891
1892 /* We found a fine healthy device, go go go... */
1893 return 0;
1894}
1895
1896static void pci_configure_mps(struct pci_dev *dev)
1897{
1898 struct pci_dev *bridge = pci_upstream_bridge(dev);
1899 int mps, mpss, p_mps, rc;
1900
1901 if (!pci_is_pcie(dev))
1902 return;
1903
1904 /* MPS and MRRS fields are of type 'RsvdP' for VFs, short-circuit out */
1905 if (dev->is_virtfn)
1906 return;
1907
1908 /*
1909 * For Root Complex Integrated Endpoints, program the maximum
1910 * supported value unless limited by the PCIE_BUS_PEER2PEER case.
1911 */
1912 if (pci_pcie_type(dev) == PCI_EXP_TYPE_RC_END) {
1913 if (pcie_bus_config == PCIE_BUS_PEER2PEER)
1914 mps = 128;
1915 else
1916 mps = 128 << dev->pcie_mpss;
1917 rc = pcie_set_mps(dev, mps);
1918 if (rc) {
1919 pci_warn(dev, "can't set Max Payload Size to %d; if necessary, use \"pci=pcie_bus_safe\" and report a bug\n",
1920 mps);
1921 }
1922 return;
1923 }
1924
1925 if (!bridge || !pci_is_pcie(bridge))
1926 return;
1927
1928 mps = pcie_get_mps(dev);
1929 p_mps = pcie_get_mps(bridge);
1930
1931 if (mps == p_mps)
1932 return;
1933
1934 if (pcie_bus_config == PCIE_BUS_TUNE_OFF) {
1935 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",
1936 mps, pci_name(bridge), p_mps);
1937 return;
1938 }
1939
1940 /*
1941 * Fancier MPS configuration is done later by
1942 * pcie_bus_configure_settings()
1943 */
1944 if (pcie_bus_config != PCIE_BUS_DEFAULT)
1945 return;
1946
1947 mpss = 128 << dev->pcie_mpss;
1948 if (mpss < p_mps && pci_pcie_type(bridge) == PCI_EXP_TYPE_ROOT_PORT) {
1949 pcie_set_mps(bridge, mpss);
1950 pci_info(dev, "Upstream bridge's Max Payload Size set to %d (was %d, max %d)\n",
1951 mpss, p_mps, 128 << bridge->pcie_mpss);
1952 p_mps = pcie_get_mps(bridge);
1953 }
1954
1955 rc = pcie_set_mps(dev, p_mps);
1956 if (rc) {
1957 pci_warn(dev, "can't set Max Payload Size to %d; if necessary, use \"pci=pcie_bus_safe\" and report a bug\n",
1958 p_mps);
1959 return;
1960 }
1961
1962 pci_info(dev, "Max Payload Size set to %d (was %d, max %d)\n",
1963 p_mps, mps, mpss);
1964}
1965
1966int pci_configure_extended_tags(struct pci_dev *dev, void *ign)
1967{
1968 struct pci_host_bridge *host;
1969 u32 cap;
1970 u16 ctl;
1971 int ret;
1972
1973 if (!pci_is_pcie(dev))
1974 return 0;
1975
1976 ret = pcie_capability_read_dword(dev, PCI_EXP_DEVCAP, &cap);
1977 if (ret)
1978 return 0;
1979
1980 if (!(cap & PCI_EXP_DEVCAP_EXT_TAG))
1981 return 0;
1982
1983 ret = pcie_capability_read_word(dev, PCI_EXP_DEVCTL, &ctl);
1984 if (ret)
1985 return 0;
1986
1987 host = pci_find_host_bridge(dev->bus);
1988 if (!host)
1989 return 0;
1990
1991 /*
1992 * If some device in the hierarchy doesn't handle Extended Tags
1993 * correctly, make sure they're disabled.
1994 */
1995 if (host->no_ext_tags) {
1996 if (ctl & PCI_EXP_DEVCTL_EXT_TAG) {
1997 pci_info(dev, "disabling Extended Tags\n");
1998 pcie_capability_clear_word(dev, PCI_EXP_DEVCTL,
1999 PCI_EXP_DEVCTL_EXT_TAG);
2000 }
2001 return 0;
2002 }
2003
2004 if (!(ctl & PCI_EXP_DEVCTL_EXT_TAG)) {
2005 pci_info(dev, "enabling Extended Tags\n");
2006 pcie_capability_set_word(dev, PCI_EXP_DEVCTL,
2007 PCI_EXP_DEVCTL_EXT_TAG);
2008 }
2009 return 0;
2010}
2011
2012/**
2013 * pcie_relaxed_ordering_enabled - Probe for PCIe relaxed ordering enable
2014 * @dev: PCI device to query
2015 *
2016 * Returns true if the device has enabled relaxed ordering attribute.
2017 */
2018bool pcie_relaxed_ordering_enabled(struct pci_dev *dev)
2019{
2020 u16 v;
2021
2022 pcie_capability_read_word(dev, PCI_EXP_DEVCTL, &v);
2023
2024 return !!(v & PCI_EXP_DEVCTL_RELAX_EN);
2025}
2026EXPORT_SYMBOL(pcie_relaxed_ordering_enabled);
2027
2028static void pci_configure_relaxed_ordering(struct pci_dev *dev)
2029{
2030 struct pci_dev *root;
2031
2032 /* PCI_EXP_DEVICE_RELAX_EN is RsvdP in VFs */
2033 if (dev->is_virtfn)
2034 return;
2035
2036 if (!pcie_relaxed_ordering_enabled(dev))
2037 return;
2038
2039 /*
2040 * For now, we only deal with Relaxed Ordering issues with Root
2041 * Ports. Peer-to-Peer DMA is another can of worms.
2042 */
2043 root = pci_find_pcie_root_port(dev);
2044 if (!root)
2045 return;
2046
2047 if (root->dev_flags & PCI_DEV_FLAGS_NO_RELAXED_ORDERING) {
2048 pcie_capability_clear_word(dev, PCI_EXP_DEVCTL,
2049 PCI_EXP_DEVCTL_RELAX_EN);
2050 pci_info(dev, "Relaxed Ordering disabled because the Root Port didn't support it\n");
2051 }
2052}
2053
2054static void pci_configure_ltr(struct pci_dev *dev)
2055{
2056#ifdef CONFIG_PCIEASPM
2057 struct pci_host_bridge *host = pci_find_host_bridge(dev->bus);
2058 struct pci_dev *bridge;
2059 u32 cap, ctl;
2060
2061 if (!pci_is_pcie(dev))
2062 return;
2063
2064 pcie_capability_read_dword(dev, PCI_EXP_DEVCAP2, &cap);
2065 if (!(cap & PCI_EXP_DEVCAP2_LTR))
2066 return;
2067
2068 pcie_capability_read_dword(dev, PCI_EXP_DEVCTL2, &ctl);
2069 if (ctl & PCI_EXP_DEVCTL2_LTR_EN) {
2070 if (pci_pcie_type(dev) == PCI_EXP_TYPE_ROOT_PORT) {
2071 dev->ltr_path = 1;
2072 return;
2073 }
2074
2075 bridge = pci_upstream_bridge(dev);
2076 if (bridge && bridge->ltr_path)
2077 dev->ltr_path = 1;
2078
2079 return;
2080 }
2081
2082 if (!host->native_ltr)
2083 return;
2084
2085 /*
2086 * Software must not enable LTR in an Endpoint unless the Root
2087 * Complex and all intermediate Switches indicate support for LTR.
2088 * PCIe r4.0, sec 6.18.
2089 */
2090 if (pci_pcie_type(dev) == PCI_EXP_TYPE_ROOT_PORT ||
2091 ((bridge = pci_upstream_bridge(dev)) &&
2092 bridge->ltr_path)) {
2093 pcie_capability_set_word(dev, PCI_EXP_DEVCTL2,
2094 PCI_EXP_DEVCTL2_LTR_EN);
2095 dev->ltr_path = 1;
2096 }
2097#endif
2098}
2099
2100static void pci_configure_eetlp_prefix(struct pci_dev *dev)
2101{
2102#ifdef CONFIG_PCI_PASID
2103 struct pci_dev *bridge;
2104 int pcie_type;
2105 u32 cap;
2106
2107 if (!pci_is_pcie(dev))
2108 return;
2109
2110 pcie_capability_read_dword(dev, PCI_EXP_DEVCAP2, &cap);
2111 if (!(cap & PCI_EXP_DEVCAP2_EE_PREFIX))
2112 return;
2113
2114 pcie_type = pci_pcie_type(dev);
2115 if (pcie_type == PCI_EXP_TYPE_ROOT_PORT ||
2116 pcie_type == PCI_EXP_TYPE_RC_END)
2117 dev->eetlp_prefix_path = 1;
2118 else {
2119 bridge = pci_upstream_bridge(dev);
2120 if (bridge && bridge->eetlp_prefix_path)
2121 dev->eetlp_prefix_path = 1;
2122 }
2123#endif
2124}
2125
2126static void pci_configure_serr(struct pci_dev *dev)
2127{
2128 u16 control;
2129
2130 if (dev->hdr_type == PCI_HEADER_TYPE_BRIDGE) {
2131
2132 /*
2133 * A bridge will not forward ERR_ messages coming from an
2134 * endpoint unless SERR# forwarding is enabled.
2135 */
2136 pci_read_config_word(dev, PCI_BRIDGE_CONTROL, &control);
2137 if (!(control & PCI_BRIDGE_CTL_SERR)) {
2138 control |= PCI_BRIDGE_CTL_SERR;
2139 pci_write_config_word(dev, PCI_BRIDGE_CONTROL, control);
2140 }
2141 }
2142}
2143
2144static void pci_configure_device(struct pci_dev *dev)
2145{
2146 pci_configure_mps(dev);
2147 pci_configure_extended_tags(dev, NULL);
2148 pci_configure_relaxed_ordering(dev);
2149 pci_configure_ltr(dev);
2150 pci_configure_eetlp_prefix(dev);
2151 pci_configure_serr(dev);
2152
2153 pci_acpi_program_hp_params(dev);
2154}
2155
2156static void pci_release_capabilities(struct pci_dev *dev)
2157{
2158 pci_aer_exit(dev);
2159 pci_vpd_release(dev);
2160 pci_iov_release(dev);
2161 pci_free_cap_save_buffers(dev);
2162}
2163
2164/**
2165 * pci_release_dev - Free a PCI device structure when all users of it are
2166 * finished
2167 * @dev: device that's been disconnected
2168 *
2169 * Will be called only by the device core when all users of this PCI device are
2170 * done.
2171 */
2172static void pci_release_dev(struct device *dev)
2173{
2174 struct pci_dev *pci_dev;
2175
2176 pci_dev = to_pci_dev(dev);
2177 pci_release_capabilities(pci_dev);
2178 pci_release_of_node(pci_dev);
2179 pcibios_release_device(pci_dev);
2180 pci_bus_put(pci_dev->bus);
2181 kfree(pci_dev->driver_override);
2182 bitmap_free(pci_dev->dma_alias_mask);
2183 kfree(pci_dev);
2184}
2185
2186struct pci_dev *pci_alloc_dev(struct pci_bus *bus)
2187{
2188 struct pci_dev *dev;
2189
2190 dev = kzalloc(sizeof(struct pci_dev), GFP_KERNEL);
2191 if (!dev)
2192 return NULL;
2193
2194 INIT_LIST_HEAD(&dev->bus_list);
2195 dev->dev.type = &pci_dev_type;
2196 dev->bus = pci_bus_get(bus);
2197
2198 return dev;
2199}
2200EXPORT_SYMBOL(pci_alloc_dev);
2201
2202static bool pci_bus_crs_vendor_id(u32 l)
2203{
2204 return (l & 0xffff) == 0x0001;
2205}
2206
2207static bool pci_bus_wait_crs(struct pci_bus *bus, int devfn, u32 *l,
2208 int timeout)
2209{
2210 int delay = 1;
2211
2212 if (!pci_bus_crs_vendor_id(*l))
2213 return true; /* not a CRS completion */
2214
2215 if (!timeout)
2216 return false; /* CRS, but caller doesn't want to wait */
2217
2218 /*
2219 * We got the reserved Vendor ID that indicates a completion with
2220 * Configuration Request Retry Status (CRS). Retry until we get a
2221 * valid Vendor ID or we time out.
2222 */
2223 while (pci_bus_crs_vendor_id(*l)) {
2224 if (delay > timeout) {
2225 pr_warn("pci %04x:%02x:%02x.%d: not ready after %dms; giving up\n",
2226 pci_domain_nr(bus), bus->number,
2227 PCI_SLOT(devfn), PCI_FUNC(devfn), delay - 1);
2228
2229 return false;
2230 }
2231 if (delay >= 1000)
2232 pr_info("pci %04x:%02x:%02x.%d: not ready after %dms; waiting\n",
2233 pci_domain_nr(bus), bus->number,
2234 PCI_SLOT(devfn), PCI_FUNC(devfn), delay - 1);
2235
2236 msleep(delay);
2237 delay *= 2;
2238
2239 if (pci_bus_read_config_dword(bus, devfn, PCI_VENDOR_ID, l))
2240 return false;
2241 }
2242
2243 if (delay >= 1000)
2244 pr_info("pci %04x:%02x:%02x.%d: ready after %dms\n",
2245 pci_domain_nr(bus), bus->number,
2246 PCI_SLOT(devfn), PCI_FUNC(devfn), delay - 1);
2247
2248 return true;
2249}
2250
2251bool pci_bus_generic_read_dev_vendor_id(struct pci_bus *bus, int devfn, u32 *l,
2252 int timeout)
2253{
2254 if (pci_bus_read_config_dword(bus, devfn, PCI_VENDOR_ID, l))
2255 return false;
2256
2257 /* Some broken boards return 0 or ~0 if a slot is empty: */
2258 if (*l == 0xffffffff || *l == 0x00000000 ||
2259 *l == 0x0000ffff || *l == 0xffff0000)
2260 return false;
2261
2262 if (pci_bus_crs_vendor_id(*l))
2263 return pci_bus_wait_crs(bus, devfn, l, timeout);
2264
2265 return true;
2266}
2267
2268bool pci_bus_read_dev_vendor_id(struct pci_bus *bus, int devfn, u32 *l,
2269 int timeout)
2270{
2271#ifdef CONFIG_PCI_QUIRKS
2272 struct pci_dev *bridge = bus->self;
2273
2274 /*
2275 * Certain IDT switches have an issue where they improperly trigger
2276 * ACS Source Validation errors on completions for config reads.
2277 */
2278 if (bridge && bridge->vendor == PCI_VENDOR_ID_IDT &&
2279 bridge->device == 0x80b5)
2280 return pci_idt_bus_quirk(bus, devfn, l, timeout);
2281#endif
2282
2283 return pci_bus_generic_read_dev_vendor_id(bus, devfn, l, timeout);
2284}
2285EXPORT_SYMBOL(pci_bus_read_dev_vendor_id);
2286
2287/*
2288 * Read the config data for a PCI device, sanity-check it,
2289 * and fill in the dev structure.
2290 */
2291static struct pci_dev *pci_scan_device(struct pci_bus *bus, int devfn)
2292{
2293 struct pci_dev *dev;
2294 u32 l;
2295
2296 if (!pci_bus_read_dev_vendor_id(bus, devfn, &l, 60*1000))
2297 return NULL;
2298
2299 dev = pci_alloc_dev(bus);
2300 if (!dev)
2301 return NULL;
2302
2303 dev->devfn = devfn;
2304 dev->vendor = l & 0xffff;
2305 dev->device = (l >> 16) & 0xffff;
2306
2307 pci_set_of_node(dev);
2308
2309 if (pci_setup_device(dev)) {
2310 pci_release_of_node(dev);
2311 pci_bus_put(dev->bus);
2312 kfree(dev);
2313 return NULL;
2314 }
2315
2316 return dev;
2317}
2318
2319void pcie_report_downtraining(struct pci_dev *dev)
2320{
2321 if (!pci_is_pcie(dev))
2322 return;
2323
2324 /* Look from the device up to avoid downstream ports with no devices */
2325 if ((pci_pcie_type(dev) != PCI_EXP_TYPE_ENDPOINT) &&
2326 (pci_pcie_type(dev) != PCI_EXP_TYPE_LEG_END) &&
2327 (pci_pcie_type(dev) != PCI_EXP_TYPE_UPSTREAM))
2328 return;
2329
2330 /* Multi-function PCIe devices share the same link/status */
2331 if (PCI_FUNC(dev->devfn) != 0 || dev->is_virtfn)
2332 return;
2333
2334 /* Print link status only if the device is constrained by the fabric */
2335 __pcie_print_link_status(dev, false);
2336}
2337
2338static void pci_init_capabilities(struct pci_dev *dev)
2339{
2340 /* Enhanced Allocation */
2341 pci_ea_init(dev);
2342
2343 /* Setup MSI caps & disable MSI/MSI-X interrupts */
2344 pci_msi_setup_pci_dev(dev);
2345
2346 /* Buffers for saving PCIe and PCI-X capabilities */
2347 pci_allocate_cap_save_buffers(dev);
2348
2349 /* Power Management */
2350 pci_pm_init(dev);
2351
2352 /* Vital Product Data */
2353 pci_vpd_init(dev);
2354
2355 /* Alternative Routing-ID Forwarding */
2356 pci_configure_ari(dev);
2357
2358 /* Single Root I/O Virtualization */
2359 pci_iov_init(dev);
2360
2361 /* Address Translation Services */
2362 pci_ats_init(dev);
2363
2364 /* Enable ACS P2P upstream forwarding */
2365 pci_enable_acs(dev);
2366
2367 /* Precision Time Measurement */
2368 pci_ptm_init(dev);
2369
2370 /* Advanced Error Reporting */
2371 pci_aer_init(dev);
2372
2373 pcie_report_downtraining(dev);
2374
2375 if (pci_probe_reset_function(dev) == 0)
2376 dev->reset_fn = 1;
2377}
2378
2379/*
2380 * This is the equivalent of pci_host_bridge_msi_domain() that acts on
2381 * devices. Firmware interfaces that can select the MSI domain on a
2382 * per-device basis should be called from here.
2383 */
2384static struct irq_domain *pci_dev_msi_domain(struct pci_dev *dev)
2385{
2386 struct irq_domain *d;
2387
2388 /*
2389 * If a domain has been set through the pcibios_add_device()
2390 * callback, then this is the one (platform code knows best).
2391 */
2392 d = dev_get_msi_domain(&dev->dev);
2393 if (d)
2394 return d;
2395
2396 /*
2397 * Let's see if we have a firmware interface able to provide
2398 * the domain.
2399 */
2400 d = pci_msi_get_device_domain(dev);
2401 if (d)
2402 return d;
2403
2404 return NULL;
2405}
2406
2407static void pci_set_msi_domain(struct pci_dev *dev)
2408{
2409 struct irq_domain *d;
2410
2411 /*
2412 * If the platform or firmware interfaces cannot supply a
2413 * device-specific MSI domain, then inherit the default domain
2414 * from the host bridge itself.
2415 */
2416 d = pci_dev_msi_domain(dev);
2417 if (!d)
2418 d = dev_get_msi_domain(&dev->bus->dev);
2419
2420 dev_set_msi_domain(&dev->dev, d);
2421}
2422
2423void pci_device_add(struct pci_dev *dev, struct pci_bus *bus)
2424{
2425 int ret;
2426
2427 pci_configure_device(dev);
2428
2429 device_initialize(&dev->dev);
2430 dev->dev.release = pci_release_dev;
2431
2432 set_dev_node(&dev->dev, pcibus_to_node(bus));
2433 dev->dev.dma_mask = &dev->dma_mask;
2434 dev->dev.dma_parms = &dev->dma_parms;
2435 dev->dev.coherent_dma_mask = 0xffffffffull;
2436
2437 dma_set_max_seg_size(&dev->dev, 65536);
2438 dma_set_seg_boundary(&dev->dev, 0xffffffff);
2439
2440 /* Fix up broken headers */
2441 pci_fixup_device(pci_fixup_header, dev);
2442
2443 /* Moved out from quirk header fixup code */
2444 pci_reassigndev_resource_alignment(dev);
2445
2446 /* Clear the state_saved flag */
2447 dev->state_saved = false;
2448
2449 /* Initialize various capabilities */
2450 pci_init_capabilities(dev);
2451
2452 /*
2453 * Add the device to our list of discovered devices
2454 * and the bus list for fixup functions, etc.
2455 */
2456 down_write(&pci_bus_sem);
2457 list_add_tail(&dev->bus_list, &bus->devices);
2458 up_write(&pci_bus_sem);
2459
2460 ret = pcibios_add_device(dev);
2461 WARN_ON(ret < 0);
2462
2463 /* Set up MSI IRQ domain */
2464 pci_set_msi_domain(dev);
2465
2466 /* Notifier could use PCI capabilities */
2467 dev->match_driver = false;
2468 ret = device_add(&dev->dev);
2469 WARN_ON(ret < 0);
2470}
2471
2472struct pci_dev *pci_scan_single_device(struct pci_bus *bus, int devfn)
2473{
2474 struct pci_dev *dev;
2475
2476 dev = pci_get_slot(bus, devfn);
2477 if (dev) {
2478 pci_dev_put(dev);
2479 return dev;
2480 }
2481
2482 dev = pci_scan_device(bus, devfn);
2483 if (!dev)
2484 return NULL;
2485
2486 pci_device_add(dev, bus);
2487
2488 return dev;
2489}
2490EXPORT_SYMBOL(pci_scan_single_device);
2491
2492static unsigned next_fn(struct pci_bus *bus, struct pci_dev *dev, unsigned fn)
2493{
2494 int pos;
2495 u16 cap = 0;
2496 unsigned next_fn;
2497
2498 if (pci_ari_enabled(bus)) {
2499 if (!dev)
2500 return 0;
2501 pos = pci_find_ext_capability(dev, PCI_EXT_CAP_ID_ARI);
2502 if (!pos)
2503 return 0;
2504
2505 pci_read_config_word(dev, pos + PCI_ARI_CAP, &cap);
2506 next_fn = PCI_ARI_CAP_NFN(cap);
2507 if (next_fn <= fn)
2508 return 0; /* protect against malformed list */
2509
2510 return next_fn;
2511 }
2512
2513 /* dev may be NULL for non-contiguous multifunction devices */
2514 if (!dev || dev->multifunction)
2515 return (fn + 1) % 8;
2516
2517 return 0;
2518}
2519
2520static int only_one_child(struct pci_bus *bus)
2521{
2522 struct pci_dev *bridge = bus->self;
2523
2524 /*
2525 * Systems with unusual topologies set PCI_SCAN_ALL_PCIE_DEVS so
2526 * we scan for all possible devices, not just Device 0.
2527 */
2528 if (pci_has_flag(PCI_SCAN_ALL_PCIE_DEVS))
2529 return 0;
2530
2531 /*
2532 * A PCIe Downstream Port normally leads to a Link with only Device
2533 * 0 on it (PCIe spec r3.1, sec 7.3.1). As an optimization, scan
2534 * only for Device 0 in that situation.
2535 */
2536 if (bridge && pci_is_pcie(bridge) && pcie_downstream_port(bridge))
2537 return 1;
2538
2539 return 0;
2540}
2541
2542/**
2543 * pci_scan_slot - Scan a PCI slot on a bus for devices
2544 * @bus: PCI bus to scan
2545 * @devfn: slot number to scan (must have zero function)
2546 *
2547 * Scan a PCI slot on the specified PCI bus for devices, adding
2548 * discovered devices to the @bus->devices list. New devices
2549 * will not have is_added set.
2550 *
2551 * Returns the number of new devices found.
2552 */
2553int pci_scan_slot(struct pci_bus *bus, int devfn)
2554{
2555 unsigned fn, nr = 0;
2556 struct pci_dev *dev;
2557
2558 if (only_one_child(bus) && (devfn > 0))
2559 return 0; /* Already scanned the entire slot */
2560
2561 dev = pci_scan_single_device(bus, devfn);
2562 if (!dev)
2563 return 0;
2564 if (!pci_dev_is_added(dev))
2565 nr++;
2566
2567 for (fn = next_fn(bus, dev, 0); fn > 0; fn = next_fn(bus, dev, fn)) {
2568 dev = pci_scan_single_device(bus, devfn + fn);
2569 if (dev) {
2570 if (!pci_dev_is_added(dev))
2571 nr++;
2572 dev->multifunction = 1;
2573 }
2574 }
2575
2576 /* Only one slot has PCIe device */
2577 if (bus->self && nr)
2578 pcie_aspm_init_link_state(bus->self);
2579
2580 return nr;
2581}
2582EXPORT_SYMBOL(pci_scan_slot);
2583
2584static int pcie_find_smpss(struct pci_dev *dev, void *data)
2585{
2586 u8 *smpss = data;
2587
2588 if (!pci_is_pcie(dev))
2589 return 0;
2590
2591 /*
2592 * We don't have a way to change MPS settings on devices that have
2593 * drivers attached. A hot-added device might support only the minimum
2594 * MPS setting (MPS=128). Therefore, if the fabric contains a bridge
2595 * where devices may be hot-added, we limit the fabric MPS to 128 so
2596 * hot-added devices will work correctly.
2597 *
2598 * However, if we hot-add a device to a slot directly below a Root
2599 * Port, it's impossible for there to be other existing devices below
2600 * the port. We don't limit the MPS in this case because we can
2601 * reconfigure MPS on both the Root Port and the hot-added device,
2602 * and there are no other devices involved.
2603 *
2604 * Note that this PCIE_BUS_SAFE path assumes no peer-to-peer DMA.
2605 */
2606 if (dev->is_hotplug_bridge &&
2607 pci_pcie_type(dev) != PCI_EXP_TYPE_ROOT_PORT)
2608 *smpss = 0;
2609
2610 if (*smpss > dev->pcie_mpss)
2611 *smpss = dev->pcie_mpss;
2612
2613 return 0;
2614}
2615
2616static void pcie_write_mps(struct pci_dev *dev, int mps)
2617{
2618 int rc;
2619
2620 if (pcie_bus_config == PCIE_BUS_PERFORMANCE) {
2621 mps = 128 << dev->pcie_mpss;
2622
2623 if (pci_pcie_type(dev) != PCI_EXP_TYPE_ROOT_PORT &&
2624 dev->bus->self)
2625
2626 /*
2627 * For "Performance", the assumption is made that
2628 * downstream communication will never be larger than
2629 * the MRRS. So, the MPS only needs to be configured
2630 * for the upstream communication. This being the case,
2631 * walk from the top down and set the MPS of the child
2632 * to that of the parent bus.
2633 *
2634 * Configure the device MPS with the smaller of the
2635 * device MPSS or the bridge MPS (which is assumed to be
2636 * properly configured at this point to the largest
2637 * allowable MPS based on its parent bus).
2638 */
2639 mps = min(mps, pcie_get_mps(dev->bus->self));
2640 }
2641
2642 rc = pcie_set_mps(dev, mps);
2643 if (rc)
2644 pci_err(dev, "Failed attempting to set the MPS\n");
2645}
2646
2647static void pcie_write_mrrs(struct pci_dev *dev)
2648{
2649 int rc, mrrs;
2650
2651 /*
2652 * In the "safe" case, do not configure the MRRS. There appear to be
2653 * issues with setting MRRS to 0 on a number of devices.
2654 */
2655 if (pcie_bus_config != PCIE_BUS_PERFORMANCE)
2656 return;
2657
2658 /*
2659 * For max performance, the MRRS must be set to the largest supported
2660 * value. However, it cannot be configured larger than the MPS the
2661 * device or the bus can support. This should already be properly
2662 * configured by a prior call to pcie_write_mps().
2663 */
2664 mrrs = pcie_get_mps(dev);
2665
2666 /*
2667 * MRRS is a R/W register. Invalid values can be written, but a
2668 * subsequent read will verify if the value is acceptable or not.
2669 * If the MRRS value provided is not acceptable (e.g., too large),
2670 * shrink the value until it is acceptable to the HW.
2671 */
2672 while (mrrs != pcie_get_readrq(dev) && mrrs >= 128) {
2673 rc = pcie_set_readrq(dev, mrrs);
2674 if (!rc)
2675 break;
2676
2677 pci_warn(dev, "Failed attempting to set the MRRS\n");
2678 mrrs /= 2;
2679 }
2680
2681 if (mrrs < 128)
2682 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");
2683}
2684
2685static int pcie_bus_configure_set(struct pci_dev *dev, void *data)
2686{
2687 int mps, orig_mps;
2688
2689 if (!pci_is_pcie(dev))
2690 return 0;
2691
2692 if (pcie_bus_config == PCIE_BUS_TUNE_OFF ||
2693 pcie_bus_config == PCIE_BUS_DEFAULT)
2694 return 0;
2695
2696 mps = 128 << *(u8 *)data;
2697 orig_mps = pcie_get_mps(dev);
2698
2699 pcie_write_mps(dev, mps);
2700 pcie_write_mrrs(dev);
2701
2702 pci_info(dev, "Max Payload Size set to %4d/%4d (was %4d), Max Read Rq %4d\n",
2703 pcie_get_mps(dev), 128 << dev->pcie_mpss,
2704 orig_mps, pcie_get_readrq(dev));
2705
2706 return 0;
2707}
2708
2709/*
2710 * pcie_bus_configure_settings() requires that pci_walk_bus work in a top-down,
2711 * parents then children fashion. If this changes, then this code will not
2712 * work as designed.
2713 */
2714void pcie_bus_configure_settings(struct pci_bus *bus)
2715{
2716 u8 smpss = 0;
2717
2718 if (!bus->self)
2719 return;
2720
2721 if (!pci_is_pcie(bus->self))
2722 return;
2723
2724 /*
2725 * FIXME - Peer to peer DMA is possible, though the endpoint would need
2726 * to be aware of the MPS of the destination. To work around this,
2727 * simply force the MPS of the entire system to the smallest possible.
2728 */
2729 if (pcie_bus_config == PCIE_BUS_PEER2PEER)
2730 smpss = 0;
2731
2732 if (pcie_bus_config == PCIE_BUS_SAFE) {
2733 smpss = bus->self->pcie_mpss;
2734
2735 pcie_find_smpss(bus->self, &smpss);
2736 pci_walk_bus(bus, pcie_find_smpss, &smpss);
2737 }
2738
2739 pcie_bus_configure_set(bus->self, &smpss);
2740 pci_walk_bus(bus, pcie_bus_configure_set, &smpss);
2741}
2742EXPORT_SYMBOL_GPL(pcie_bus_configure_settings);
2743
2744/*
2745 * Called after each bus is probed, but before its children are examined. This
2746 * is marked as __weak because multiple architectures define it.
2747 */
2748void __weak pcibios_fixup_bus(struct pci_bus *bus)
2749{
2750 /* nothing to do, expected to be removed in the future */
2751}
2752
2753/**
2754 * pci_scan_child_bus_extend() - Scan devices below a bus
2755 * @bus: Bus to scan for devices
2756 * @available_buses: Total number of buses available (%0 does not try to
2757 * extend beyond the minimal)
2758 *
2759 * Scans devices below @bus including subordinate buses. Returns new
2760 * subordinate number including all the found devices. Passing
2761 * @available_buses causes the remaining bus space to be distributed
2762 * equally between hotplug-capable bridges to allow future extension of the
2763 * hierarchy.
2764 */
2765static unsigned int pci_scan_child_bus_extend(struct pci_bus *bus,
2766 unsigned int available_buses)
2767{
2768 unsigned int used_buses, normal_bridges = 0, hotplug_bridges = 0;
2769 unsigned int start = bus->busn_res.start;
2770 unsigned int devfn, fn, cmax, max = start;
2771 struct pci_dev *dev;
2772 int nr_devs;
2773
2774 dev_dbg(&bus->dev, "scanning bus\n");
2775
2776 /* Go find them, Rover! */
2777 for (devfn = 0; devfn < 256; devfn += 8) {
2778 nr_devs = pci_scan_slot(bus, devfn);
2779
2780 /*
2781 * The Jailhouse hypervisor may pass individual functions of a
2782 * multi-function device to a guest without passing function 0.
2783 * Look for them as well.
2784 */
2785 if (jailhouse_paravirt() && nr_devs == 0) {
2786 for (fn = 1; fn < 8; fn++) {
2787 dev = pci_scan_single_device(bus, devfn + fn);
2788 if (dev)
2789 dev->multifunction = 1;
2790 }
2791 }
2792 }
2793
2794 /* Reserve buses for SR-IOV capability */
2795 used_buses = pci_iov_bus_range(bus);
2796 max += used_buses;
2797
2798 /*
2799 * After performing arch-dependent fixup of the bus, look behind
2800 * all PCI-to-PCI bridges on this bus.
2801 */
2802 if (!bus->is_added) {
2803 dev_dbg(&bus->dev, "fixups for bus\n");
2804 pcibios_fixup_bus(bus);
2805 bus->is_added = 1;
2806 }
2807
2808 /*
2809 * Calculate how many hotplug bridges and normal bridges there
2810 * are on this bus. We will distribute the additional available
2811 * buses between hotplug bridges.
2812 */
2813 for_each_pci_bridge(dev, bus) {
2814 if (dev->is_hotplug_bridge)
2815 hotplug_bridges++;
2816 else
2817 normal_bridges++;
2818 }
2819
2820 /*
2821 * Scan bridges that are already configured. We don't touch them
2822 * unless they are misconfigured (which will be done in the second
2823 * scan below).
2824 */
2825 for_each_pci_bridge(dev, bus) {
2826 cmax = max;
2827 max = pci_scan_bridge_extend(bus, dev, max, 0, 0);
2828
2829 /*
2830 * Reserve one bus for each bridge now to avoid extending
2831 * hotplug bridges too much during the second scan below.
2832 */
2833 used_buses++;
2834 if (cmax - max > 1)
2835 used_buses += cmax - max - 1;
2836 }
2837
2838 /* Scan bridges that need to be reconfigured */
2839 for_each_pci_bridge(dev, bus) {
2840 unsigned int buses = 0;
2841
2842 if (!hotplug_bridges && normal_bridges == 1) {
2843
2844 /*
2845 * There is only one bridge on the bus (upstream
2846 * port) so it gets all available buses which it
2847 * can then distribute to the possible hotplug
2848 * bridges below.
2849 */
2850 buses = available_buses;
2851 } else if (dev->is_hotplug_bridge) {
2852
2853 /*
2854 * Distribute the extra buses between hotplug
2855 * bridges if any.
2856 */
2857 buses = available_buses / hotplug_bridges;
2858 buses = min(buses, available_buses - used_buses + 1);
2859 }
2860
2861 cmax = max;
2862 max = pci_scan_bridge_extend(bus, dev, cmax, buses, 1);
2863 /* One bus is already accounted so don't add it again */
2864 if (max - cmax > 1)
2865 used_buses += max - cmax - 1;
2866 }
2867
2868 /*
2869 * Make sure a hotplug bridge has at least the minimum requested
2870 * number of buses but allow it to grow up to the maximum available
2871 * bus number of there is room.
2872 */
2873 if (bus->self && bus->self->is_hotplug_bridge) {
2874 used_buses = max_t(unsigned int, available_buses,
2875 pci_hotplug_bus_size - 1);
2876 if (max - start < used_buses) {
2877 max = start + used_buses;
2878
2879 /* Do not allocate more buses than we have room left */
2880 if (max > bus->busn_res.end)
2881 max = bus->busn_res.end;
2882
2883 dev_dbg(&bus->dev, "%pR extended by %#02x\n",
2884 &bus->busn_res, max - start);
2885 }
2886 }
2887
2888 /*
2889 * We've scanned the bus and so we know all about what's on
2890 * the other side of any bridges that may be on this bus plus
2891 * any devices.
2892 *
2893 * Return how far we've got finding sub-buses.
2894 */
2895 dev_dbg(&bus->dev, "bus scan returning with max=%02x\n", max);
2896 return max;
2897}
2898
2899/**
2900 * pci_scan_child_bus() - Scan devices below a bus
2901 * @bus: Bus to scan for devices
2902 *
2903 * Scans devices below @bus including subordinate buses. Returns new
2904 * subordinate number including all the found devices.
2905 */
2906unsigned int pci_scan_child_bus(struct pci_bus *bus)
2907{
2908 return pci_scan_child_bus_extend(bus, 0);
2909}
2910EXPORT_SYMBOL_GPL(pci_scan_child_bus);
2911
2912/**
2913 * pcibios_root_bridge_prepare - Platform-specific host bridge setup
2914 * @bridge: Host bridge to set up
2915 *
2916 * Default empty implementation. Replace with an architecture-specific setup
2917 * routine, if necessary.
2918 */
2919int __weak pcibios_root_bridge_prepare(struct pci_host_bridge *bridge)
2920{
2921 return 0;
2922}
2923
2924void __weak pcibios_add_bus(struct pci_bus *bus)
2925{
2926}
2927
2928void __weak pcibios_remove_bus(struct pci_bus *bus)
2929{
2930}
2931
2932struct pci_bus *pci_create_root_bus(struct device *parent, int bus,
2933 struct pci_ops *ops, void *sysdata, struct list_head *resources)
2934{
2935 int error;
2936 struct pci_host_bridge *bridge;
2937
2938 bridge = pci_alloc_host_bridge(0);
2939 if (!bridge)
2940 return NULL;
2941
2942 bridge->dev.parent = parent;
2943
2944 list_splice_init(resources, &bridge->windows);
2945 bridge->sysdata = sysdata;
2946 bridge->busnr = bus;
2947 bridge->ops = ops;
2948
2949 error = pci_register_host_bridge(bridge);
2950 if (error < 0)
2951 goto err_out;
2952
2953 return bridge->bus;
2954
2955err_out:
2956 put_device(&bridge->dev);
2957 return NULL;
2958}
2959EXPORT_SYMBOL_GPL(pci_create_root_bus);
2960
2961int pci_host_probe(struct pci_host_bridge *bridge)
2962{
2963 struct pci_bus *bus, *child;
2964 int ret;
2965
2966 ret = pci_scan_root_bus_bridge(bridge);
2967 if (ret < 0) {
2968 dev_err(bridge->dev.parent, "Scanning root bridge failed");
2969 return ret;
2970 }
2971
2972 bus = bridge->bus;
2973
2974 /*
2975 * We insert PCI resources into the iomem_resource and
2976 * ioport_resource trees in either pci_bus_claim_resources()
2977 * or pci_bus_assign_resources().
2978 */
2979 if (pci_has_flag(PCI_PROBE_ONLY)) {
2980 pci_bus_claim_resources(bus);
2981 } else {
2982 pci_bus_size_bridges(bus);
2983 pci_bus_assign_resources(bus);
2984
2985 list_for_each_entry(child, &bus->children, node)
2986 pcie_bus_configure_settings(child);
2987 }
2988
2989 pci_bus_add_devices(bus);
2990 return 0;
2991}
2992EXPORT_SYMBOL_GPL(pci_host_probe);
2993
2994int pci_bus_insert_busn_res(struct pci_bus *b, int bus, int bus_max)
2995{
2996 struct resource *res = &b->busn_res;
2997 struct resource *parent_res, *conflict;
2998
2999 res->start = bus;
3000 res->end = bus_max;
3001 res->flags = IORESOURCE_BUS;
3002
3003 if (!pci_is_root_bus(b))
3004 parent_res = &b->parent->busn_res;
3005 else {
3006 parent_res = get_pci_domain_busn_res(pci_domain_nr(b));
3007 res->flags |= IORESOURCE_PCI_FIXED;
3008 }
3009
3010 conflict = request_resource_conflict(parent_res, res);
3011
3012 if (conflict)
3013 dev_info(&b->dev,
3014 "busn_res: can not insert %pR under %s%pR (conflicts with %s %pR)\n",
3015 res, pci_is_root_bus(b) ? "domain " : "",
3016 parent_res, conflict->name, conflict);
3017
3018 return conflict == NULL;
3019}
3020
3021int pci_bus_update_busn_res_end(struct pci_bus *b, int bus_max)
3022{
3023 struct resource *res = &b->busn_res;
3024 struct resource old_res = *res;
3025 resource_size_t size;
3026 int ret;
3027
3028 if (res->start > bus_max)
3029 return -EINVAL;
3030
3031 size = bus_max - res->start + 1;
3032 ret = adjust_resource(res, res->start, size);
3033 dev_info(&b->dev, "busn_res: %pR end %s updated to %02x\n",
3034 &old_res, ret ? "can not be" : "is", bus_max);
3035
3036 if (!ret && !res->parent)
3037 pci_bus_insert_busn_res(b, res->start, res->end);
3038
3039 return ret;
3040}
3041
3042void pci_bus_release_busn_res(struct pci_bus *b)
3043{
3044 struct resource *res = &b->busn_res;
3045 int ret;
3046
3047 if (!res->flags || !res->parent)
3048 return;
3049
3050 ret = release_resource(res);
3051 dev_info(&b->dev, "busn_res: %pR %s released\n",
3052 res, ret ? "can not be" : "is");
3053}
3054
3055int pci_scan_root_bus_bridge(struct pci_host_bridge *bridge)
3056{
3057 struct resource_entry *window;
3058 bool found = false;
3059 struct pci_bus *b;
3060 int max, bus, ret;
3061
3062 if (!bridge)
3063 return -EINVAL;
3064
3065 resource_list_for_each_entry(window, &bridge->windows)
3066 if (window->res->flags & IORESOURCE_BUS) {
3067 found = true;
3068 break;
3069 }
3070
3071 ret = pci_register_host_bridge(bridge);
3072 if (ret < 0)
3073 return ret;
3074
3075 b = bridge->bus;
3076 bus = bridge->busnr;
3077
3078 if (!found) {
3079 dev_info(&b->dev,
3080 "No busn resource found for root bus, will use [bus %02x-ff]\n",
3081 bus);
3082 pci_bus_insert_busn_res(b, bus, 255);
3083 }
3084
3085 max = pci_scan_child_bus(b);
3086
3087 if (!found)
3088 pci_bus_update_busn_res_end(b, max);
3089
3090 return 0;
3091}
3092EXPORT_SYMBOL(pci_scan_root_bus_bridge);
3093
3094struct pci_bus *pci_scan_root_bus(struct device *parent, int bus,
3095 struct pci_ops *ops, void *sysdata, struct list_head *resources)
3096{
3097 struct resource_entry *window;
3098 bool found = false;
3099 struct pci_bus *b;
3100 int max;
3101
3102 resource_list_for_each_entry(window, resources)
3103 if (window->res->flags & IORESOURCE_BUS) {
3104 found = true;
3105 break;
3106 }
3107
3108 b = pci_create_root_bus(parent, bus, ops, sysdata, resources);
3109 if (!b)
3110 return NULL;
3111
3112 if (!found) {
3113 dev_info(&b->dev,
3114 "No busn resource found for root bus, will use [bus %02x-ff]\n",
3115 bus);
3116 pci_bus_insert_busn_res(b, bus, 255);
3117 }
3118
3119 max = pci_scan_child_bus(b);
3120
3121 if (!found)
3122 pci_bus_update_busn_res_end(b, max);
3123
3124 return b;
3125}
3126EXPORT_SYMBOL(pci_scan_root_bus);
3127
3128struct pci_bus *pci_scan_bus(int bus, struct pci_ops *ops,
3129 void *sysdata)
3130{
3131 LIST_HEAD(resources);
3132 struct pci_bus *b;
3133
3134 pci_add_resource(&resources, &ioport_resource);
3135 pci_add_resource(&resources, &iomem_resource);
3136 pci_add_resource(&resources, &busn_resource);
3137 b = pci_create_root_bus(NULL, bus, ops, sysdata, &resources);
3138 if (b) {
3139 pci_scan_child_bus(b);
3140 } else {
3141 pci_free_resource_list(&resources);
3142 }
3143 return b;
3144}
3145EXPORT_SYMBOL(pci_scan_bus);
3146
3147/**
3148 * pci_rescan_bus_bridge_resize - Scan a PCI bus for devices
3149 * @bridge: PCI bridge for the bus to scan
3150 *
3151 * Scan a PCI bus and child buses for new devices, add them,
3152 * and enable them, resizing bridge mmio/io resource if necessary
3153 * and possible. The caller must ensure the child devices are already
3154 * removed for resizing to occur.
3155 *
3156 * Returns the max number of subordinate bus discovered.
3157 */
3158unsigned int pci_rescan_bus_bridge_resize(struct pci_dev *bridge)
3159{
3160 unsigned int max;
3161 struct pci_bus *bus = bridge->subordinate;
3162
3163 max = pci_scan_child_bus(bus);
3164
3165 pci_assign_unassigned_bridge_resources(bridge);
3166
3167 pci_bus_add_devices(bus);
3168
3169 return max;
3170}
3171
3172/**
3173 * pci_rescan_bus - Scan a PCI bus for devices
3174 * @bus: PCI bus to scan
3175 *
3176 * Scan a PCI bus and child buses for new devices, add them,
3177 * and enable them.
3178 *
3179 * Returns the max number of subordinate bus discovered.
3180 */
3181unsigned int pci_rescan_bus(struct pci_bus *bus)
3182{
3183 unsigned int max;
3184
3185 max = pci_scan_child_bus(bus);
3186 pci_assign_unassigned_bus_resources(bus);
3187 pci_bus_add_devices(bus);
3188
3189 return max;
3190}
3191EXPORT_SYMBOL_GPL(pci_rescan_bus);
3192
3193/*
3194 * pci_rescan_bus(), pci_rescan_bus_bridge_resize() and PCI device removal
3195 * routines should always be executed under this mutex.
3196 */
3197static DEFINE_MUTEX(pci_rescan_remove_lock);
3198
3199void pci_lock_rescan_remove(void)
3200{
3201 mutex_lock(&pci_rescan_remove_lock);
3202}
3203EXPORT_SYMBOL_GPL(pci_lock_rescan_remove);
3204
3205void pci_unlock_rescan_remove(void)
3206{
3207 mutex_unlock(&pci_rescan_remove_lock);
3208}
3209EXPORT_SYMBOL_GPL(pci_unlock_rescan_remove);
3210
3211static int __init pci_sort_bf_cmp(const struct device *d_a,
3212 const struct device *d_b)
3213{
3214 const struct pci_dev *a = to_pci_dev(d_a);
3215 const struct pci_dev *b = to_pci_dev(d_b);
3216
3217 if (pci_domain_nr(a->bus) < pci_domain_nr(b->bus)) return -1;
3218 else if (pci_domain_nr(a->bus) > pci_domain_nr(b->bus)) return 1;
3219
3220 if (a->bus->number < b->bus->number) return -1;
3221 else if (a->bus->number > b->bus->number) return 1;
3222
3223 if (a->devfn < b->devfn) return -1;
3224 else if (a->devfn > b->devfn) return 1;
3225
3226 return 0;
3227}
3228
3229void __init pci_sort_breadthfirst(void)
3230{
3231 bus_sort_breadthfirst(&pci_bus_type, &pci_sort_bf_cmp);
3232}
3233
3234int pci_hp_add_bridge(struct pci_dev *dev)
3235{
3236 struct pci_bus *parent = dev->bus;
3237 int busnr, start = parent->busn_res.start;
3238 unsigned int available_buses = 0;
3239 int end = parent->busn_res.end;
3240
3241 for (busnr = start; busnr <= end; busnr++) {
3242 if (!pci_find_bus(pci_domain_nr(parent), busnr))
3243 break;
3244 }
3245 if (busnr-- > end) {
3246 pci_err(dev, "No bus number available for hot-added bridge\n");
3247 return -1;
3248 }
3249
3250 /* Scan bridges that are already configured */
3251 busnr = pci_scan_bridge(parent, dev, busnr, 0);
3252
3253 /*
3254 * Distribute the available bus numbers between hotplug-capable
3255 * bridges to make extending the chain later possible.
3256 */
3257 available_buses = end - busnr;
3258
3259 /* Scan bridges that need to be reconfigured */
3260 pci_scan_bridge_extend(parent, dev, busnr, available_buses, 1);
3261
3262 if (!dev->subordinate)
3263 return -1;
3264
3265 return 0;
3266}
3267EXPORT_SYMBOL_GPL(pci_hp_add_bridge);