blob: fa4c386c8cd84d439c2767bfc3c3ef37b5a00e42 [file] [log] [blame]
xjb04a4022021-11-25 15:01:52 +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
67static int find_anything(struct device *dev, void *data)
68{
69 return 1;
70}
71
72/*
73 * Some device drivers need know if PCI is initiated.
74 * Basically, we think PCI is not initiated when there
75 * is no device to be found on the pci_bus_type.
76 */
77int no_pci_devices(void)
78{
79 struct device *dev;
80 int no_devices;
81
82 dev = bus_find_device(&pci_bus_type, NULL, NULL, find_anything);
83 no_devices = (dev == NULL);
84 put_device(dev);
85 return no_devices;
86}
87EXPORT_SYMBOL(no_pci_devices);
88
89/*
90 * PCI Bus Class
91 */
92static void release_pcibus_dev(struct device *dev)
93{
94 struct pci_bus *pci_bus = to_pci_bus(dev);
95
96 put_device(pci_bus->bridge);
97 pci_bus_remove_resources(pci_bus);
98 pci_release_bus_of_node(pci_bus);
99 kfree(pci_bus);
100}
101
102static struct class pcibus_class = {
103 .name = "pci_bus",
104 .dev_release = &release_pcibus_dev,
105 .dev_groups = pcibus_groups,
106};
107
108static int __init pcibus_class_init(void)
109{
110 return class_register(&pcibus_class);
111}
112postcore_initcall(pcibus_class_init);
113
114static u64 pci_size(u64 base, u64 maxbase, u64 mask)
115{
116 u64 size = mask & maxbase; /* Find the significant bits */
117 if (!size)
118 return 0;
119
120 /*
121 * Get the lowest of them to find the decode size, and from that
122 * the extent.
123 */
124 size = (size & ~(size-1)) - 1;
125
126 /*
127 * base == maxbase can be valid only if the BAR has already been
128 * programmed with all 1s.
129 */
130 if (base == maxbase && ((base | size) & mask) != mask)
131 return 0;
132
133 return size;
134}
135
136static inline unsigned long decode_bar(struct pci_dev *dev, u32 bar)
137{
138 u32 mem_type;
139 unsigned long flags;
140
141 if ((bar & PCI_BASE_ADDRESS_SPACE) == PCI_BASE_ADDRESS_SPACE_IO) {
142 flags = bar & ~PCI_BASE_ADDRESS_IO_MASK;
143 flags |= IORESOURCE_IO;
144 return flags;
145 }
146
147 flags = bar & ~PCI_BASE_ADDRESS_MEM_MASK;
148 flags |= IORESOURCE_MEM;
149 if (flags & PCI_BASE_ADDRESS_MEM_PREFETCH)
150 flags |= IORESOURCE_PREFETCH;
151
152 mem_type = bar & PCI_BASE_ADDRESS_MEM_TYPE_MASK;
153 switch (mem_type) {
154 case PCI_BASE_ADDRESS_MEM_TYPE_32:
155 break;
156 case PCI_BASE_ADDRESS_MEM_TYPE_1M:
157 /* 1M mem BAR treated as 32-bit BAR */
158 break;
159 case PCI_BASE_ADDRESS_MEM_TYPE_64:
160 flags |= IORESOURCE_MEM_64;
161 break;
162 default:
163 /* mem unknown type treated as 32-bit BAR */
164 break;
165 }
166 return flags;
167}
168
169#define PCI_COMMAND_DECODE_ENABLE (PCI_COMMAND_MEMORY | PCI_COMMAND_IO)
170
171/**
172 * pci_read_base - Read a PCI BAR
173 * @dev: the PCI device
174 * @type: type of the BAR
175 * @res: resource buffer to be filled in
176 * @pos: BAR position in the config space
177 *
178 * Returns 1 if the BAR is 64-bit, or 0 if 32-bit.
179 */
180int __pci_read_base(struct pci_dev *dev, enum pci_bar_type type,
181 struct resource *res, unsigned int pos)
182{
183 u32 l = 0, sz = 0, mask;
184 u64 l64, sz64, mask64;
185 u16 orig_cmd;
186 struct pci_bus_region region, inverted_region;
187
188 mask = type ? PCI_ROM_ADDRESS_MASK : ~0;
189
190 /* No printks while decoding is disabled! */
191 if (!dev->mmio_always_on) {
192 pci_read_config_word(dev, PCI_COMMAND, &orig_cmd);
193 if (orig_cmd & PCI_COMMAND_DECODE_ENABLE) {
194 pci_write_config_word(dev, PCI_COMMAND,
195 orig_cmd & ~PCI_COMMAND_DECODE_ENABLE);
196 }
197 }
198
199 res->name = pci_name(dev);
200
201 pci_read_config_dword(dev, pos, &l);
202 pci_write_config_dword(dev, pos, l | mask);
203 pci_read_config_dword(dev, pos, &sz);
204 pci_write_config_dword(dev, pos, l);
205
206 /*
207 * All bits set in sz means the device isn't working properly.
208 * If the BAR isn't implemented, all bits must be 0. If it's a
209 * memory BAR or a ROM, bit 0 must be clear; if it's an io BAR, bit
210 * 1 must be clear.
211 */
212 if (sz == 0xffffffff)
213 sz = 0;
214
215 /*
216 * I don't know how l can have all bits set. Copied from old code.
217 * Maybe it fixes a bug on some ancient platform.
218 */
219 if (l == 0xffffffff)
220 l = 0;
221
222 if (type == pci_bar_unknown) {
223 res->flags = decode_bar(dev, l);
224 res->flags |= IORESOURCE_SIZEALIGN;
225 if (res->flags & IORESOURCE_IO) {
226 l64 = l & PCI_BASE_ADDRESS_IO_MASK;
227 sz64 = sz & PCI_BASE_ADDRESS_IO_MASK;
228 mask64 = PCI_BASE_ADDRESS_IO_MASK & (u32)IO_SPACE_LIMIT;
229 } else {
230 l64 = l & PCI_BASE_ADDRESS_MEM_MASK;
231 sz64 = sz & PCI_BASE_ADDRESS_MEM_MASK;
232 mask64 = (u32)PCI_BASE_ADDRESS_MEM_MASK;
233 }
234 } else {
235 if (l & PCI_ROM_ADDRESS_ENABLE)
236 res->flags |= IORESOURCE_ROM_ENABLE;
237 l64 = l & PCI_ROM_ADDRESS_MASK;
238 sz64 = sz & PCI_ROM_ADDRESS_MASK;
239 mask64 = PCI_ROM_ADDRESS_MASK;
240 }
241
242 if (res->flags & IORESOURCE_MEM_64) {
243 pci_read_config_dword(dev, pos + 4, &l);
244 pci_write_config_dword(dev, pos + 4, ~0);
245 pci_read_config_dword(dev, pos + 4, &sz);
246 pci_write_config_dword(dev, pos + 4, l);
247
248 l64 |= ((u64)l << 32);
249 sz64 |= ((u64)sz << 32);
250 mask64 |= ((u64)~0 << 32);
251 }
252
253 if (!dev->mmio_always_on && (orig_cmd & PCI_COMMAND_DECODE_ENABLE))
254 pci_write_config_word(dev, PCI_COMMAND, orig_cmd);
255
256 if (!sz64)
257 goto fail;
258
259 sz64 = pci_size(l64, sz64, mask64);
260 if (!sz64) {
261 pci_info(dev, FW_BUG "reg 0x%x: invalid BAR (can't size)\n",
262 pos);
263 goto fail;
264 }
265
266 if (res->flags & IORESOURCE_MEM_64) {
267 if ((sizeof(pci_bus_addr_t) < 8 || sizeof(resource_size_t) < 8)
268 && sz64 > 0x100000000ULL) {
269 res->flags |= IORESOURCE_UNSET | IORESOURCE_DISABLED;
270 res->start = 0;
271 res->end = 0;
272 pci_err(dev, "reg 0x%x: can't handle BAR larger than 4GB (size %#010llx)\n",
273 pos, (unsigned long long)sz64);
274 goto out;
275 }
276
277 if ((sizeof(pci_bus_addr_t) < 8) && l) {
278 /* Above 32-bit boundary; try to reallocate */
279 res->flags |= IORESOURCE_UNSET;
280 res->start = 0;
281 res->end = sz64;
282 pci_info(dev, "reg 0x%x: can't handle BAR above 4GB (bus address %#010llx)\n",
283 pos, (unsigned long long)l64);
284 goto out;
285 }
286 }
287
288 region.start = l64;
289 region.end = l64 + sz64;
290
291 pcibios_bus_to_resource(dev->bus, res, &region);
292 pcibios_resource_to_bus(dev->bus, &inverted_region, res);
293
294 /*
295 * If "A" is a BAR value (a bus address), "bus_to_resource(A)" is
296 * the corresponding resource address (the physical address used by
297 * the CPU. Converting that resource address back to a bus address
298 * should yield the original BAR value:
299 *
300 * resource_to_bus(bus_to_resource(A)) == A
301 *
302 * If it doesn't, CPU accesses to "bus_to_resource(A)" will not
303 * be claimed by the device.
304 */
305 if (inverted_region.start != region.start) {
306 res->flags |= IORESOURCE_UNSET;
307 res->start = 0;
308 res->end = region.end - region.start;
309 pci_info(dev, "reg 0x%x: initial BAR value %#010llx invalid\n",
310 pos, (unsigned long long)region.start);
311 }
312
313 goto out;
314
315
316fail:
317 res->flags = 0;
318out:
319 if (res->flags)
320 pci_printk(KERN_DEBUG, dev, "reg 0x%x: %pR\n", pos, res);
321
322 return (res->flags & IORESOURCE_MEM_64) ? 1 : 0;
323}
324
325static void pci_read_bases(struct pci_dev *dev, unsigned int howmany, int rom)
326{
327 unsigned int pos, reg;
328
329 if (dev->non_compliant_bars)
330 return;
331
332 /* Per PCIe r4.0, sec 9.3.4.1.11, the VF BARs are all RO Zero */
333 if (dev->is_virtfn)
334 return;
335
336 for (pos = 0; pos < howmany; pos++) {
337 struct resource *res = &dev->resource[pos];
338 reg = PCI_BASE_ADDRESS_0 + (pos << 2);
339 pos += __pci_read_base(dev, pci_bar_unknown, res, reg);
340 }
341
342 if (rom) {
343 struct resource *res = &dev->resource[PCI_ROM_RESOURCE];
344 dev->rom_base_reg = rom;
345 res->flags = IORESOURCE_MEM | IORESOURCE_PREFETCH |
346 IORESOURCE_READONLY | IORESOURCE_SIZEALIGN;
347 __pci_read_base(dev, pci_bar_mem32, res, rom);
348 }
349}
350
351static void pci_read_bridge_io(struct pci_bus *child)
352{
353 struct pci_dev *dev = child->self;
354 u8 io_base_lo, io_limit_lo;
355 unsigned long io_mask, io_granularity, base, limit;
356 struct pci_bus_region region;
357 struct resource *res;
358
359 io_mask = PCI_IO_RANGE_MASK;
360 io_granularity = 0x1000;
361 if (dev->io_window_1k) {
362 /* Support 1K I/O space granularity */
363 io_mask = PCI_IO_1K_RANGE_MASK;
364 io_granularity = 0x400;
365 }
366
367 res = child->resource[0];
368 pci_read_config_byte(dev, PCI_IO_BASE, &io_base_lo);
369 pci_read_config_byte(dev, PCI_IO_LIMIT, &io_limit_lo);
370 base = (io_base_lo & io_mask) << 8;
371 limit = (io_limit_lo & io_mask) << 8;
372
373 if ((io_base_lo & PCI_IO_RANGE_TYPE_MASK) == PCI_IO_RANGE_TYPE_32) {
374 u16 io_base_hi, io_limit_hi;
375
376 pci_read_config_word(dev, PCI_IO_BASE_UPPER16, &io_base_hi);
377 pci_read_config_word(dev, PCI_IO_LIMIT_UPPER16, &io_limit_hi);
378 base |= ((unsigned long) io_base_hi << 16);
379 limit |= ((unsigned long) io_limit_hi << 16);
380 }
381
382 if (base <= limit) {
383 res->flags = (io_base_lo & PCI_IO_RANGE_TYPE_MASK) | IORESOURCE_IO;
384 region.start = base;
385 region.end = limit + io_granularity - 1;
386 pcibios_bus_to_resource(dev->bus, res, &region);
387 pci_printk(KERN_DEBUG, dev, " bridge window %pR\n", res);
388 }
389}
390
391static void pci_read_bridge_mmio(struct pci_bus *child)
392{
393 struct pci_dev *dev = child->self;
394 u16 mem_base_lo, mem_limit_lo;
395 unsigned long base, limit;
396 struct pci_bus_region region;
397 struct resource *res;
398
399 res = child->resource[1];
400 pci_read_config_word(dev, PCI_MEMORY_BASE, &mem_base_lo);
401 pci_read_config_word(dev, PCI_MEMORY_LIMIT, &mem_limit_lo);
402 base = ((unsigned long) mem_base_lo & PCI_MEMORY_RANGE_MASK) << 16;
403 limit = ((unsigned long) mem_limit_lo & PCI_MEMORY_RANGE_MASK) << 16;
404 if (base <= limit) {
405 res->flags = (mem_base_lo & PCI_MEMORY_RANGE_TYPE_MASK) | IORESOURCE_MEM;
406 region.start = base;
407 region.end = limit + 0xfffff;
408 pcibios_bus_to_resource(dev->bus, res, &region);
409 pci_printk(KERN_DEBUG, dev, " bridge window %pR\n", res);
410 }
411}
412
413static void pci_read_bridge_mmio_pref(struct pci_bus *child)
414{
415 struct pci_dev *dev = child->self;
416 u16 mem_base_lo, mem_limit_lo;
417 u64 base64, limit64;
418 pci_bus_addr_t base, limit;
419 struct pci_bus_region region;
420 struct resource *res;
421
422 res = child->resource[2];
423 pci_read_config_word(dev, PCI_PREF_MEMORY_BASE, &mem_base_lo);
424 pci_read_config_word(dev, PCI_PREF_MEMORY_LIMIT, &mem_limit_lo);
425 base64 = (mem_base_lo & PCI_PREF_RANGE_MASK) << 16;
426 limit64 = (mem_limit_lo & PCI_PREF_RANGE_MASK) << 16;
427
428 if ((mem_base_lo & PCI_PREF_RANGE_TYPE_MASK) == PCI_PREF_RANGE_TYPE_64) {
429 u32 mem_base_hi, mem_limit_hi;
430
431 pci_read_config_dword(dev, PCI_PREF_BASE_UPPER32, &mem_base_hi);
432 pci_read_config_dword(dev, PCI_PREF_LIMIT_UPPER32, &mem_limit_hi);
433
434 /*
435 * Some bridges set the base > limit by default, and some
436 * (broken) BIOSes do not initialize them. If we find
437 * this, just assume they are not being used.
438 */
439 if (mem_base_hi <= mem_limit_hi) {
440 base64 |= (u64) mem_base_hi << 32;
441 limit64 |= (u64) mem_limit_hi << 32;
442 }
443 }
444
445 base = (pci_bus_addr_t) base64;
446 limit = (pci_bus_addr_t) limit64;
447
448 if (base != base64) {
449 pci_err(dev, "can't handle bridge window above 4GB (bus address %#010llx)\n",
450 (unsigned long long) base64);
451 return;
452 }
453
454 if (base <= limit) {
455 res->flags = (mem_base_lo & PCI_PREF_RANGE_TYPE_MASK) |
456 IORESOURCE_MEM | IORESOURCE_PREFETCH;
457 if (res->flags & PCI_PREF_RANGE_TYPE_64)
458 res->flags |= IORESOURCE_MEM_64;
459 region.start = base;
460 region.end = limit + 0xfffff;
461 pcibios_bus_to_resource(dev->bus, res, &region);
462 pci_printk(KERN_DEBUG, dev, " bridge window %pR\n", res);
463 }
464}
465
466void pci_read_bridge_bases(struct pci_bus *child)
467{
468 struct pci_dev *dev = child->self;
469 struct resource *res;
470 int i;
471
472 if (pci_is_root_bus(child)) /* It's a host bus, nothing to read */
473 return;
474
475 pci_info(dev, "PCI bridge to %pR%s\n",
476 &child->busn_res,
477 dev->transparent ? " (subtractive decode)" : "");
478
479 pci_bus_remove_resources(child);
480 for (i = 0; i < PCI_BRIDGE_RESOURCE_NUM; i++)
481 child->resource[i] = &dev->resource[PCI_BRIDGE_RESOURCES+i];
482
483 pci_read_bridge_io(child);
484 pci_read_bridge_mmio(child);
485 pci_read_bridge_mmio_pref(child);
486
487 if (dev->transparent) {
488 pci_bus_for_each_resource(child->parent, res, i) {
489 if (res && res->flags) {
490 pci_bus_add_resource(child, res,
491 PCI_SUBTRACTIVE_DECODE);
492 pci_printk(KERN_DEBUG, dev,
493 " bridge window %pR (subtractive decode)\n",
494 res);
495 }
496 }
497 }
498}
499
500static struct pci_bus *pci_alloc_bus(struct pci_bus *parent)
501{
502 struct pci_bus *b;
503
504 b = kzalloc(sizeof(*b), GFP_KERNEL);
505 if (!b)
506 return NULL;
507
508 INIT_LIST_HEAD(&b->node);
509 INIT_LIST_HEAD(&b->children);
510 INIT_LIST_HEAD(&b->devices);
511 INIT_LIST_HEAD(&b->slots);
512 INIT_LIST_HEAD(&b->resources);
513 b->max_bus_speed = PCI_SPEED_UNKNOWN;
514 b->cur_bus_speed = PCI_SPEED_UNKNOWN;
515#ifdef CONFIG_PCI_DOMAINS_GENERIC
516 if (parent)
517 b->domain_nr = parent->domain_nr;
518#endif
519 return b;
520}
521
522static void devm_pci_release_host_bridge_dev(struct device *dev)
523{
524 struct pci_host_bridge *bridge = to_pci_host_bridge(dev);
525
526 if (bridge->release_fn)
527 bridge->release_fn(bridge);
528
529 pci_free_resource_list(&bridge->windows);
530}
531
532static void pci_release_host_bridge_dev(struct device *dev)
533{
534 devm_pci_release_host_bridge_dev(dev);
535 kfree(to_pci_host_bridge(dev));
536}
537
538static void pci_init_host_bridge(struct pci_host_bridge *bridge)
539{
540 INIT_LIST_HEAD(&bridge->windows);
541
542 /*
543 * We assume we can manage these PCIe features. Some systems may
544 * reserve these for use by the platform itself, e.g., an ACPI BIOS
545 * may implement its own AER handling and use _OSC to prevent the
546 * OS from interfering.
547 */
548 bridge->native_aer = 1;
549 bridge->native_pcie_hotplug = 1;
550 bridge->native_shpc_hotplug = 1;
551 bridge->native_pme = 1;
552 bridge->native_ltr = 1;
553}
554
555struct pci_host_bridge *pci_alloc_host_bridge(size_t priv)
556{
557 struct pci_host_bridge *bridge;
558
559 bridge = kzalloc(sizeof(*bridge) + priv, GFP_KERNEL);
560 if (!bridge)
561 return NULL;
562
563 pci_init_host_bridge(bridge);
564 bridge->dev.release = pci_release_host_bridge_dev;
565
566 return bridge;
567}
568EXPORT_SYMBOL(pci_alloc_host_bridge);
569
570struct pci_host_bridge *devm_pci_alloc_host_bridge(struct device *dev,
571 size_t priv)
572{
573 struct pci_host_bridge *bridge;
574
575 bridge = devm_kzalloc(dev, sizeof(*bridge) + priv, GFP_KERNEL);
576 if (!bridge)
577 return NULL;
578
579 pci_init_host_bridge(bridge);
580 bridge->dev.release = devm_pci_release_host_bridge_dev;
581
582 return bridge;
583}
584EXPORT_SYMBOL(devm_pci_alloc_host_bridge);
585
586void pci_free_host_bridge(struct pci_host_bridge *bridge)
587{
588 pci_free_resource_list(&bridge->windows);
589
590 kfree(bridge);
591}
592EXPORT_SYMBOL(pci_free_host_bridge);
593
594static const unsigned char pcix_bus_speed[] = {
595 PCI_SPEED_UNKNOWN, /* 0 */
596 PCI_SPEED_66MHz_PCIX, /* 1 */
597 PCI_SPEED_100MHz_PCIX, /* 2 */
598 PCI_SPEED_133MHz_PCIX, /* 3 */
599 PCI_SPEED_UNKNOWN, /* 4 */
600 PCI_SPEED_66MHz_PCIX_ECC, /* 5 */
601 PCI_SPEED_100MHz_PCIX_ECC, /* 6 */
602 PCI_SPEED_133MHz_PCIX_ECC, /* 7 */
603 PCI_SPEED_UNKNOWN, /* 8 */
604 PCI_SPEED_66MHz_PCIX_266, /* 9 */
605 PCI_SPEED_100MHz_PCIX_266, /* A */
606 PCI_SPEED_133MHz_PCIX_266, /* B */
607 PCI_SPEED_UNKNOWN, /* C */
608 PCI_SPEED_66MHz_PCIX_533, /* D */
609 PCI_SPEED_100MHz_PCIX_533, /* E */
610 PCI_SPEED_133MHz_PCIX_533 /* F */
611};
612
613const unsigned char pcie_link_speed[] = {
614 PCI_SPEED_UNKNOWN, /* 0 */
615 PCIE_SPEED_2_5GT, /* 1 */
616 PCIE_SPEED_5_0GT, /* 2 */
617 PCIE_SPEED_8_0GT, /* 3 */
618 PCIE_SPEED_16_0GT, /* 4 */
619 PCI_SPEED_UNKNOWN, /* 5 */
620 PCI_SPEED_UNKNOWN, /* 6 */
621 PCI_SPEED_UNKNOWN, /* 7 */
622 PCI_SPEED_UNKNOWN, /* 8 */
623 PCI_SPEED_UNKNOWN, /* 9 */
624 PCI_SPEED_UNKNOWN, /* A */
625 PCI_SPEED_UNKNOWN, /* B */
626 PCI_SPEED_UNKNOWN, /* C */
627 PCI_SPEED_UNKNOWN, /* D */
628 PCI_SPEED_UNKNOWN, /* E */
629 PCI_SPEED_UNKNOWN /* F */
630};
631
632void pcie_update_link_speed(struct pci_bus *bus, u16 linksta)
633{
634 bus->cur_bus_speed = pcie_link_speed[linksta & PCI_EXP_LNKSTA_CLS];
635}
636EXPORT_SYMBOL_GPL(pcie_update_link_speed);
637
638static unsigned char agp_speeds[] = {
639 AGP_UNKNOWN,
640 AGP_1X,
641 AGP_2X,
642 AGP_4X,
643 AGP_8X
644};
645
646static enum pci_bus_speed agp_speed(int agp3, int agpstat)
647{
648 int index = 0;
649
650 if (agpstat & 4)
651 index = 3;
652 else if (agpstat & 2)
653 index = 2;
654 else if (agpstat & 1)
655 index = 1;
656 else
657 goto out;
658
659 if (agp3) {
660 index += 2;
661 if (index == 5)
662 index = 0;
663 }
664
665 out:
666 return agp_speeds[index];
667}
668
669static void pci_set_bus_speed(struct pci_bus *bus)
670{
671 struct pci_dev *bridge = bus->self;
672 int pos;
673
674 pos = pci_find_capability(bridge, PCI_CAP_ID_AGP);
675 if (!pos)
676 pos = pci_find_capability(bridge, PCI_CAP_ID_AGP3);
677 if (pos) {
678 u32 agpstat, agpcmd;
679
680 pci_read_config_dword(bridge, pos + PCI_AGP_STATUS, &agpstat);
681 bus->max_bus_speed = agp_speed(agpstat & 8, agpstat & 7);
682
683 pci_read_config_dword(bridge, pos + PCI_AGP_COMMAND, &agpcmd);
684 bus->cur_bus_speed = agp_speed(agpstat & 8, agpcmd & 7);
685 }
686
687 pos = pci_find_capability(bridge, PCI_CAP_ID_PCIX);
688 if (pos) {
689 u16 status;
690 enum pci_bus_speed max;
691
692 pci_read_config_word(bridge, pos + PCI_X_BRIDGE_SSTATUS,
693 &status);
694
695 if (status & PCI_X_SSTATUS_533MHZ) {
696 max = PCI_SPEED_133MHz_PCIX_533;
697 } else if (status & PCI_X_SSTATUS_266MHZ) {
698 max = PCI_SPEED_133MHz_PCIX_266;
699 } else if (status & PCI_X_SSTATUS_133MHZ) {
700 if ((status & PCI_X_SSTATUS_VERS) == PCI_X_SSTATUS_V2)
701 max = PCI_SPEED_133MHz_PCIX_ECC;
702 else
703 max = PCI_SPEED_133MHz_PCIX;
704 } else {
705 max = PCI_SPEED_66MHz_PCIX;
706 }
707
708 bus->max_bus_speed = max;
709 bus->cur_bus_speed = pcix_bus_speed[
710 (status & PCI_X_SSTATUS_FREQ) >> 6];
711
712 return;
713 }
714
715 if (pci_is_pcie(bridge)) {
716 u32 linkcap;
717 u16 linksta;
718
719 pcie_capability_read_dword(bridge, PCI_EXP_LNKCAP, &linkcap);
720 bus->max_bus_speed = pcie_link_speed[linkcap & PCI_EXP_LNKCAP_SLS];
721
722 pcie_capability_read_word(bridge, PCI_EXP_LNKSTA, &linksta);
723 pcie_update_link_speed(bus, linksta);
724 }
725}
726
727static struct irq_domain *pci_host_bridge_msi_domain(struct pci_bus *bus)
728{
729 struct irq_domain *d;
730
731 /*
732 * Any firmware interface that can resolve the msi_domain
733 * should be called from here.
734 */
735 d = pci_host_bridge_of_msi_domain(bus);
736 if (!d)
737 d = pci_host_bridge_acpi_msi_domain(bus);
738
739#ifdef CONFIG_PCI_MSI_IRQ_DOMAIN
740 /*
741 * If no IRQ domain was found via the OF tree, try looking it up
742 * directly through the fwnode_handle.
743 */
744 if (!d) {
745 struct fwnode_handle *fwnode = pci_root_bus_fwnode(bus);
746
747 if (fwnode)
748 d = irq_find_matching_fwnode(fwnode,
749 DOMAIN_BUS_PCI_MSI);
750 }
751#endif
752
753 return d;
754}
755
756static void pci_set_bus_msi_domain(struct pci_bus *bus)
757{
758 struct irq_domain *d;
759 struct pci_bus *b;
760
761 /*
762 * The bus can be a root bus, a subordinate bus, or a virtual bus
763 * created by an SR-IOV device. Walk up to the first bridge device
764 * found or derive the domain from the host bridge.
765 */
766 for (b = bus, d = NULL; !d && !pci_is_root_bus(b); b = b->parent) {
767 if (b->self)
768 d = dev_get_msi_domain(&b->self->dev);
769 }
770
771 if (!d)
772 d = pci_host_bridge_msi_domain(b);
773
774 dev_set_msi_domain(&bus->dev, d);
775}
776
777static int pci_register_host_bridge(struct pci_host_bridge *bridge)
778{
779 struct device *parent = bridge->dev.parent;
780 struct resource_entry *window, *n;
781 struct pci_bus *bus, *b;
782 resource_size_t offset;
783 LIST_HEAD(resources);
784 struct resource *res;
785 char addr[64], *fmt;
786 const char *name;
787 int err;
788
789 bus = pci_alloc_bus(NULL);
790 if (!bus)
791 return -ENOMEM;
792
793 bridge->bus = bus;
794
795 /* Temporarily move resources off the list */
796 list_splice_init(&bridge->windows, &resources);
797 bus->sysdata = bridge->sysdata;
798 bus->msi = bridge->msi;
799 bus->ops = bridge->ops;
800 bus->number = bus->busn_res.start = bridge->busnr;
801#ifdef CONFIG_PCI_DOMAINS_GENERIC
802 bus->domain_nr = pci_bus_find_domain_nr(bus, parent);
803#endif
804
805 b = pci_find_bus(pci_domain_nr(bus), bridge->busnr);
806 if (b) {
807 /* Ignore it if we already got here via a different bridge */
808 dev_dbg(&b->dev, "bus already known\n");
809 err = -EEXIST;
810 goto free;
811 }
812
813 dev_set_name(&bridge->dev, "pci%04x:%02x", pci_domain_nr(bus),
814 bridge->busnr);
815
816 err = pcibios_root_bridge_prepare(bridge);
817 if (err)
818 goto free;
819
820 err = device_register(&bridge->dev);
821 if (err)
822 put_device(&bridge->dev);
823
824 bus->bridge = get_device(&bridge->dev);
825 device_enable_async_suspend(bus->bridge);
826 pci_set_bus_of_node(bus);
827 pci_set_bus_msi_domain(bus);
828
829 if (!parent)
830 set_dev_node(bus->bridge, pcibus_to_node(bus));
831
832 bus->dev.class = &pcibus_class;
833 bus->dev.parent = bus->bridge;
834
835 dev_set_name(&bus->dev, "%04x:%02x", pci_domain_nr(bus), bus->number);
836 name = dev_name(&bus->dev);
837
838 err = device_register(&bus->dev);
839 if (err)
840 goto unregister;
841
842 pcibios_add_bus(bus);
843
844 /* Create legacy_io and legacy_mem files for this bus */
845 pci_create_legacy_files(bus);
846
847 if (parent)
848 dev_info(parent, "PCI host bridge to bus %s\n", name);
849 else
850 pr_info("PCI host bridge to bus %s\n", name);
851
852 /* Add initial resources to the bus */
853 resource_list_for_each_entry_safe(window, n, &resources) {
854 list_move_tail(&window->node, &bridge->windows);
855 offset = window->offset;
856 res = window->res;
857
858 if (res->flags & IORESOURCE_BUS)
859 pci_bus_insert_busn_res(bus, bus->number, res->end);
860 else
861 pci_bus_add_resource(bus, res, 0);
862
863 if (offset) {
864 if (resource_type(res) == IORESOURCE_IO)
865 fmt = " (bus address [%#06llx-%#06llx])";
866 else
867 fmt = " (bus address [%#010llx-%#010llx])";
868
869 snprintf(addr, sizeof(addr), fmt,
870 (unsigned long long)(res->start - offset),
871 (unsigned long long)(res->end - offset));
872 } else
873 addr[0] = '\0';
874
875 dev_info(&bus->dev, "root bus resource %pR%s\n", res, addr);
876 }
877
878 down_write(&pci_bus_sem);
879 list_add_tail(&bus->node, &pci_root_buses);
880 up_write(&pci_bus_sem);
881
882 return 0;
883
884unregister:
885 put_device(&bridge->dev);
886 device_unregister(&bridge->dev);
887
888free:
889 kfree(bus);
890 return err;
891}
892
893static bool pci_bridge_child_ext_cfg_accessible(struct pci_dev *bridge)
894{
895 int pos;
896 u32 status;
897
898 /*
899 * If extended config space isn't accessible on a bridge's primary
900 * bus, we certainly can't access it on the secondary bus.
901 */
902 if (bridge->bus->bus_flags & PCI_BUS_FLAGS_NO_EXTCFG)
903 return false;
904
905 /*
906 * PCIe Root Ports and switch ports are PCIe on both sides, so if
907 * extended config space is accessible on the primary, it's also
908 * accessible on the secondary.
909 */
910 if (pci_is_pcie(bridge) &&
911 (pci_pcie_type(bridge) == PCI_EXP_TYPE_ROOT_PORT ||
912 pci_pcie_type(bridge) == PCI_EXP_TYPE_UPSTREAM ||
913 pci_pcie_type(bridge) == PCI_EXP_TYPE_DOWNSTREAM))
914 return true;
915
916 /*
917 * For the other bridge types:
918 * - PCI-to-PCI bridges
919 * - PCIe-to-PCI/PCI-X forward bridges
920 * - PCI/PCI-X-to-PCIe reverse bridges
921 * extended config space on the secondary side is only accessible
922 * if the bridge supports PCI-X Mode 2.
923 */
924 pos = pci_find_capability(bridge, PCI_CAP_ID_PCIX);
925 if (!pos)
926 return false;
927
928 pci_read_config_dword(bridge, pos + PCI_X_STATUS, &status);
929 return status & (PCI_X_STATUS_266MHZ | PCI_X_STATUS_533MHZ);
930}
931
932static struct pci_bus *pci_alloc_child_bus(struct pci_bus *parent,
933 struct pci_dev *bridge, int busnr)
934{
935 struct pci_bus *child;
936 int i;
937 int ret;
938
939 /* Allocate a new bus and inherit stuff from the parent */
940 child = pci_alloc_bus(parent);
941 if (!child)
942 return NULL;
943
944 child->parent = parent;
945 child->ops = parent->ops;
946 child->msi = parent->msi;
947 child->sysdata = parent->sysdata;
948 child->bus_flags = parent->bus_flags;
949
950 /*
951 * Initialize some portions of the bus device, but don't register
952 * it now as the parent is not properly set up yet.
953 */
954 child->dev.class = &pcibus_class;
955 dev_set_name(&child->dev, "%04x:%02x", pci_domain_nr(child), busnr);
956
957 /* Set up the primary, secondary and subordinate bus numbers */
958 child->number = child->busn_res.start = busnr;
959 child->primary = parent->busn_res.start;
960 child->busn_res.end = 0xff;
961
962 if (!bridge) {
963 child->dev.parent = parent->bridge;
964 goto add_dev;
965 }
966
967 child->self = bridge;
968 child->bridge = get_device(&bridge->dev);
969 child->dev.parent = child->bridge;
970 pci_set_bus_of_node(child);
971 pci_set_bus_speed(child);
972
973 /*
974 * Check whether extended config space is accessible on the child
975 * bus. Note that we currently assume it is always accessible on
976 * the root bus.
977 */
978 if (!pci_bridge_child_ext_cfg_accessible(bridge)) {
979 child->bus_flags |= PCI_BUS_FLAGS_NO_EXTCFG;
980 pci_info(child, "extended config space not accessible\n");
981 }
982
983 /* Set up default resource pointers and names */
984 for (i = 0; i < PCI_BRIDGE_RESOURCE_NUM; i++) {
985 child->resource[i] = &bridge->resource[PCI_BRIDGE_RESOURCES+i];
986 child->resource[i]->name = child->name;
987 }
988 bridge->subordinate = child;
989
990add_dev:
991 pci_set_bus_msi_domain(child);
992 ret = device_register(&child->dev);
993 WARN_ON(ret < 0);
994
995 pcibios_add_bus(child);
996
997 if (child->ops->add_bus) {
998 ret = child->ops->add_bus(child);
999 if (WARN_ON(ret < 0))
1000 dev_err(&child->dev, "failed to add bus: %d\n", ret);
1001 }
1002
1003 /* Create legacy_io and legacy_mem files for this bus */
1004 pci_create_legacy_files(child);
1005
1006 return child;
1007}
1008
1009struct pci_bus *pci_add_new_bus(struct pci_bus *parent, struct pci_dev *dev,
1010 int busnr)
1011{
1012 struct pci_bus *child;
1013
1014 child = pci_alloc_child_bus(parent, dev, busnr);
1015 if (child) {
1016 down_write(&pci_bus_sem);
1017 list_add_tail(&child->node, &parent->children);
1018 up_write(&pci_bus_sem);
1019 }
1020 return child;
1021}
1022EXPORT_SYMBOL(pci_add_new_bus);
1023
1024static void pci_enable_crs(struct pci_dev *pdev)
1025{
1026 u16 root_cap = 0;
1027
1028 /* Enable CRS Software Visibility if supported */
1029 pcie_capability_read_word(pdev, PCI_EXP_RTCAP, &root_cap);
1030 if (root_cap & PCI_EXP_RTCAP_CRSVIS)
1031 pcie_capability_set_word(pdev, PCI_EXP_RTCTL,
1032 PCI_EXP_RTCTL_CRSSVE);
1033}
1034
1035static unsigned int pci_scan_child_bus_extend(struct pci_bus *bus,
1036 unsigned int available_buses);
1037
1038/*
1039 * pci_scan_bridge_extend() - Scan buses behind a bridge
1040 * @bus: Parent bus the bridge is on
1041 * @dev: Bridge itself
1042 * @max: Starting subordinate number of buses behind this bridge
1043 * @available_buses: Total number of buses available for this bridge and
1044 * the devices below. After the minimal bus space has
1045 * been allocated the remaining buses will be
1046 * distributed equally between hotplug-capable bridges.
1047 * @pass: Either %0 (scan already configured bridges) or %1 (scan bridges
1048 * that need to be reconfigured.
1049 *
1050 * If it's a bridge, configure it and scan the bus behind it.
1051 * For CardBus bridges, we don't scan behind as the devices will
1052 * be handled by the bridge driver itself.
1053 *
1054 * We need to process bridges in two passes -- first we scan those
1055 * already configured by the BIOS and after we are done with all of
1056 * them, we proceed to assigning numbers to the remaining buses in
1057 * order to avoid overlaps between old and new bus numbers.
1058 *
1059 * Return: New subordinate number covering all buses behind this bridge.
1060 */
1061static int pci_scan_bridge_extend(struct pci_bus *bus, struct pci_dev *dev,
1062 int max, unsigned int available_buses,
1063 int pass)
1064{
1065 struct pci_bus *child;
1066 int is_cardbus = (dev->hdr_type == PCI_HEADER_TYPE_CARDBUS);
1067 u32 buses, i, j = 0;
1068 u16 bctl;
1069 u8 primary, secondary, subordinate;
1070 int broken = 0;
1071
1072 /*
1073 * Make sure the bridge is powered on to be able to access config
1074 * space of devices below it.
1075 */
1076 pm_runtime_get_sync(&dev->dev);
1077
1078 pci_read_config_dword(dev, PCI_PRIMARY_BUS, &buses);
1079 primary = buses & 0xFF;
1080 secondary = (buses >> 8) & 0xFF;
1081 subordinate = (buses >> 16) & 0xFF;
1082
1083 pci_dbg(dev, "scanning [bus %02x-%02x] behind bridge, pass %d\n",
1084 secondary, subordinate, pass);
1085
1086 if (!primary && (primary != bus->number) && secondary && subordinate) {
1087 pci_warn(dev, "Primary bus is hard wired to 0\n");
1088 primary = bus->number;
1089 }
1090
1091 /* Check if setup is sensible at all */
1092 if (!pass &&
1093 (primary != bus->number || secondary <= bus->number ||
1094 secondary > subordinate)) {
1095 pci_info(dev, "bridge configuration invalid ([bus %02x-%02x]), reconfiguring\n",
1096 secondary, subordinate);
1097 broken = 1;
1098 }
1099
1100 /*
1101 * Disable Master-Abort Mode during probing to avoid reporting of
1102 * bus errors in some architectures.
1103 */
1104 pci_read_config_word(dev, PCI_BRIDGE_CONTROL, &bctl);
1105 pci_write_config_word(dev, PCI_BRIDGE_CONTROL,
1106 bctl & ~PCI_BRIDGE_CTL_MASTER_ABORT);
1107
1108 pci_enable_crs(dev);
1109
1110 if ((secondary || subordinate) && !pcibios_assign_all_busses() &&
1111 !is_cardbus && !broken) {
1112 unsigned int cmax;
1113
1114 /*
1115 * Bus already configured by firmware, process it in the
1116 * first pass and just note the configuration.
1117 */
1118 if (pass)
1119 goto out;
1120
1121 /*
1122 * The bus might already exist for two reasons: Either we
1123 * are rescanning the bus or the bus is reachable through
1124 * more than one bridge. The second case can happen with
1125 * the i450NX chipset.
1126 */
1127 child = pci_find_bus(pci_domain_nr(bus), secondary);
1128 if (!child) {
1129 child = pci_add_new_bus(bus, dev, secondary);
1130 if (!child)
1131 goto out;
1132 child->primary = primary;
1133 pci_bus_insert_busn_res(child, secondary, subordinate);
1134 child->bridge_ctl = bctl;
1135 }
1136
1137 cmax = pci_scan_child_bus(child);
1138 if (cmax > subordinate)
1139 pci_warn(dev, "bridge has subordinate %02x but max busn %02x\n",
1140 subordinate, cmax);
1141
1142 /* Subordinate should equal child->busn_res.end */
1143 if (subordinate > max)
1144 max = subordinate;
1145 } else {
1146
1147 /*
1148 * We need to assign a number to this bus which we always
1149 * do in the second pass.
1150 */
1151 if (!pass) {
1152 if (pcibios_assign_all_busses() || broken || is_cardbus)
1153
1154 /*
1155 * Temporarily disable forwarding of the
1156 * configuration cycles on all bridges in
1157 * this bus segment to avoid possible
1158 * conflicts in the second pass between two
1159 * bridges programmed with overlapping bus
1160 * ranges.
1161 */
1162 pci_write_config_dword(dev, PCI_PRIMARY_BUS,
1163 buses & ~0xffffff);
1164 goto out;
1165 }
1166
1167 /* Clear errors */
1168 pci_write_config_word(dev, PCI_STATUS, 0xffff);
1169
1170 /*
1171 * Prevent assigning a bus number that already exists.
1172 * This can happen when a bridge is hot-plugged, so in this
1173 * case we only re-scan this bus.
1174 */
1175 child = pci_find_bus(pci_domain_nr(bus), max+1);
1176 if (!child) {
1177 child = pci_add_new_bus(bus, dev, max+1);
1178 if (!child)
1179 goto out;
1180 pci_bus_insert_busn_res(child, max+1,
1181 bus->busn_res.end);
1182 }
1183 max++;
1184 if (available_buses)
1185 available_buses--;
1186
1187 buses = (buses & 0xff000000)
1188 | ((unsigned int)(child->primary) << 0)
1189 | ((unsigned int)(child->busn_res.start) << 8)
1190 | ((unsigned int)(child->busn_res.end) << 16);
1191
1192 /*
1193 * yenta.c forces a secondary latency timer of 176.
1194 * Copy that behaviour here.
1195 */
1196 if (is_cardbus) {
1197 buses &= ~0xff000000;
1198 buses |= CARDBUS_LATENCY_TIMER << 24;
1199 }
1200
1201 /* We need to blast all three values with a single write */
1202 pci_write_config_dword(dev, PCI_PRIMARY_BUS, buses);
1203
1204 if (!is_cardbus) {
1205 child->bridge_ctl = bctl;
1206 max = pci_scan_child_bus_extend(child, available_buses);
1207 } else {
1208
1209 /*
1210 * For CardBus bridges, we leave 4 bus numbers as
1211 * cards with a PCI-to-PCI bridge can be inserted
1212 * later.
1213 */
1214 for (i = 0; i < CARDBUS_RESERVE_BUSNR; i++) {
1215 struct pci_bus *parent = bus;
1216 if (pci_find_bus(pci_domain_nr(bus),
1217 max+i+1))
1218 break;
1219 while (parent->parent) {
1220 if ((!pcibios_assign_all_busses()) &&
1221 (parent->busn_res.end > max) &&
1222 (parent->busn_res.end <= max+i)) {
1223 j = 1;
1224 }
1225 parent = parent->parent;
1226 }
1227 if (j) {
1228
1229 /*
1230 * Often, there are two CardBus
1231 * bridges -- try to leave one
1232 * valid bus number for each one.
1233 */
1234 i /= 2;
1235 break;
1236 }
1237 }
1238 max += i;
1239 }
1240
1241 /* Set subordinate bus number to its real value */
1242 pci_bus_update_busn_res_end(child, max);
1243 pci_write_config_byte(dev, PCI_SUBORDINATE_BUS, max);
1244 }
1245
1246 sprintf(child->name,
1247 (is_cardbus ? "PCI CardBus %04x:%02x" : "PCI Bus %04x:%02x"),
1248 pci_domain_nr(bus), child->number);
1249
1250 /* Check that all devices are accessible */
1251 while (bus->parent) {
1252 if ((child->busn_res.end > bus->busn_res.end) ||
1253 (child->number > bus->busn_res.end) ||
1254 (child->number < bus->number) ||
1255 (child->busn_res.end < bus->number)) {
1256 dev_info(&dev->dev, "devices behind bridge are unusable because %pR cannot be assigned for them\n",
1257 &child->busn_res);
1258 break;
1259 }
1260 bus = bus->parent;
1261 }
1262
1263out:
1264 pci_write_config_word(dev, PCI_BRIDGE_CONTROL, bctl);
1265
1266 pm_runtime_put(&dev->dev);
1267
1268 return max;
1269}
1270
1271/*
1272 * pci_scan_bridge() - Scan buses behind a bridge
1273 * @bus: Parent bus the bridge is on
1274 * @dev: Bridge itself
1275 * @max: Starting subordinate number of buses behind this bridge
1276 * @pass: Either %0 (scan already configured bridges) or %1 (scan bridges
1277 * that need to be reconfigured.
1278 *
1279 * If it's a bridge, configure it and scan the bus behind it.
1280 * For CardBus bridges, we don't scan behind as the devices will
1281 * be handled by the bridge driver itself.
1282 *
1283 * We need to process bridges in two passes -- first we scan those
1284 * already configured by the BIOS and after we are done with all of
1285 * them, we proceed to assigning numbers to the remaining buses in
1286 * order to avoid overlaps between old and new bus numbers.
1287 *
1288 * Return: New subordinate number covering all buses behind this bridge.
1289 */
1290int pci_scan_bridge(struct pci_bus *bus, struct pci_dev *dev, int max, int pass)
1291{
1292 return pci_scan_bridge_extend(bus, dev, max, 0, pass);
1293}
1294EXPORT_SYMBOL(pci_scan_bridge);
1295
1296/*
1297 * Read interrupt line and base address registers.
1298 * The architecture-dependent code can tweak these, of course.
1299 */
1300static void pci_read_irq(struct pci_dev *dev)
1301{
1302 unsigned char irq;
1303
1304 /* VFs are not allowed to use INTx, so skip the config reads */
1305 if (dev->is_virtfn) {
1306 dev->pin = 0;
1307 dev->irq = 0;
1308 return;
1309 }
1310
1311 pci_read_config_byte(dev, PCI_INTERRUPT_PIN, &irq);
1312 dev->pin = irq;
1313 if (irq)
1314 pci_read_config_byte(dev, PCI_INTERRUPT_LINE, &irq);
1315 dev->irq = irq;
1316}
1317
1318void set_pcie_port_type(struct pci_dev *pdev)
1319{
1320 int pos;
1321 u16 reg16;
1322 int type;
1323 struct pci_dev *parent;
1324
1325 pos = pci_find_capability(pdev, PCI_CAP_ID_EXP);
1326 if (!pos)
1327 return;
1328
1329 pdev->pcie_cap = pos;
1330 pci_read_config_word(pdev, pos + PCI_EXP_FLAGS, &reg16);
1331 pdev->pcie_flags_reg = reg16;
1332 pci_read_config_word(pdev, pos + PCI_EXP_DEVCAP, &reg16);
1333 pdev->pcie_mpss = reg16 & PCI_EXP_DEVCAP_PAYLOAD;
1334
1335 /*
1336 * A Root Port or a PCI-to-PCIe bridge is always the upstream end
1337 * of a Link. No PCIe component has two Links. Two Links are
1338 * connected by a Switch that has a Port on each Link and internal
1339 * logic to connect the two Ports.
1340 */
1341 type = pci_pcie_type(pdev);
1342 if (type == PCI_EXP_TYPE_ROOT_PORT ||
1343 type == PCI_EXP_TYPE_PCIE_BRIDGE)
1344 pdev->has_secondary_link = 1;
1345 else if (type == PCI_EXP_TYPE_UPSTREAM ||
1346 type == PCI_EXP_TYPE_DOWNSTREAM) {
1347 parent = pci_upstream_bridge(pdev);
1348
1349 /*
1350 * Usually there's an upstream device (Root Port or Switch
1351 * Downstream Port), but we can't assume one exists.
1352 */
1353 if (parent && !parent->has_secondary_link)
1354 pdev->has_secondary_link = 1;
1355 }
1356}
1357
1358void set_pcie_hotplug_bridge(struct pci_dev *pdev)
1359{
1360 u32 reg32;
1361
1362 pcie_capability_read_dword(pdev, PCI_EXP_SLTCAP, &reg32);
1363 if (reg32 & PCI_EXP_SLTCAP_HPC)
1364 pdev->is_hotplug_bridge = 1;
1365}
1366
1367static void set_pcie_thunderbolt(struct pci_dev *dev)
1368{
1369 int vsec = 0;
1370 u32 header;
1371
1372 while ((vsec = pci_find_next_ext_capability(dev, vsec,
1373 PCI_EXT_CAP_ID_VNDR))) {
1374 pci_read_config_dword(dev, vsec + PCI_VNDR_HEADER, &header);
1375
1376 /* Is the device part of a Thunderbolt controller? */
1377 if (dev->vendor == PCI_VENDOR_ID_INTEL &&
1378 PCI_VNDR_HEADER_ID(header) == PCI_VSEC_ID_INTEL_TBT) {
1379 dev->is_thunderbolt = 1;
1380 return;
1381 }
1382 }
1383}
1384
1385/**
1386 * pci_ext_cfg_is_aliased - Is ext config space just an alias of std config?
1387 * @dev: PCI device
1388 *
1389 * PCI Express to PCI/PCI-X Bridge Specification, rev 1.0, 4.1.4 says that
1390 * when forwarding a type1 configuration request the bridge must check that
1391 * the extended register address field is zero. The bridge is not permitted
1392 * to forward the transactions and must handle it as an Unsupported Request.
1393 * Some bridges do not follow this rule and simply drop the extended register
1394 * bits, resulting in the standard config space being aliased, every 256
1395 * bytes across the entire configuration space. Test for this condition by
1396 * comparing the first dword of each potential alias to the vendor/device ID.
1397 * Known offenders:
1398 * ASM1083/1085 PCIe-to-PCI Reversible Bridge (1b21:1080, rev 01 & 03)
1399 * AMD/ATI SBx00 PCI to PCI Bridge (1002:4384, rev 40)
1400 */
1401static bool pci_ext_cfg_is_aliased(struct pci_dev *dev)
1402{
1403#ifdef CONFIG_PCI_QUIRKS
1404 int pos;
1405 u32 header, tmp;
1406
1407 pci_read_config_dword(dev, PCI_VENDOR_ID, &header);
1408
1409 for (pos = PCI_CFG_SPACE_SIZE;
1410 pos < PCI_CFG_SPACE_EXP_SIZE; pos += PCI_CFG_SPACE_SIZE) {
1411 if (pci_read_config_dword(dev, pos, &tmp) != PCIBIOS_SUCCESSFUL
1412 || header != tmp)
1413 return false;
1414 }
1415
1416 return true;
1417#else
1418 return false;
1419#endif
1420}
1421
1422/**
1423 * pci_cfg_space_size - Get the configuration space size of the PCI device
1424 * @dev: PCI device
1425 *
1426 * Regular PCI devices have 256 bytes, but PCI-X 2 and PCI Express devices
1427 * have 4096 bytes. Even if the device is capable, that doesn't mean we can
1428 * access it. Maybe we don't have a way to generate extended config space
1429 * accesses, or the device is behind a reverse Express bridge. So we try
1430 * reading the dword at 0x100 which must either be 0 or a valid extended
1431 * capability header.
1432 */
1433static int pci_cfg_space_size_ext(struct pci_dev *dev)
1434{
1435 u32 status;
1436 int pos = PCI_CFG_SPACE_SIZE;
1437
1438 if (pci_read_config_dword(dev, pos, &status) != PCIBIOS_SUCCESSFUL)
1439 return PCI_CFG_SPACE_SIZE;
1440 if (status == 0xffffffff || pci_ext_cfg_is_aliased(dev))
1441 return PCI_CFG_SPACE_SIZE;
1442
1443 return PCI_CFG_SPACE_EXP_SIZE;
1444}
1445
1446int pci_cfg_space_size(struct pci_dev *dev)
1447{
1448 int pos;
1449 u32 status;
1450 u16 class;
1451
1452 if (dev->bus->bus_flags & PCI_BUS_FLAGS_NO_EXTCFG)
1453 return PCI_CFG_SPACE_SIZE;
1454
1455 class = dev->class >> 8;
1456 if (class == PCI_CLASS_BRIDGE_HOST)
1457 return pci_cfg_space_size_ext(dev);
1458
1459 if (pci_is_pcie(dev))
1460 return pci_cfg_space_size_ext(dev);
1461
1462 pos = pci_find_capability(dev, PCI_CAP_ID_PCIX);
1463 if (!pos)
1464 return PCI_CFG_SPACE_SIZE;
1465
1466 pci_read_config_dword(dev, pos + PCI_X_STATUS, &status);
1467 if (status & (PCI_X_STATUS_266MHZ | PCI_X_STATUS_533MHZ))
1468 return pci_cfg_space_size_ext(dev);
1469
1470 return PCI_CFG_SPACE_SIZE;
1471}
1472
1473static u32 pci_class(struct pci_dev *dev)
1474{
1475 u32 class;
1476
1477#ifdef CONFIG_PCI_IOV
1478 if (dev->is_virtfn)
1479 return dev->physfn->sriov->class;
1480#endif
1481 pci_read_config_dword(dev, PCI_CLASS_REVISION, &class);
1482 return class;
1483}
1484
1485static void pci_subsystem_ids(struct pci_dev *dev, u16 *vendor, u16 *device)
1486{
1487#ifdef CONFIG_PCI_IOV
1488 if (dev->is_virtfn) {
1489 *vendor = dev->physfn->sriov->subsystem_vendor;
1490 *device = dev->physfn->sriov->subsystem_device;
1491 return;
1492 }
1493#endif
1494 pci_read_config_word(dev, PCI_SUBSYSTEM_VENDOR_ID, vendor);
1495 pci_read_config_word(dev, PCI_SUBSYSTEM_ID, device);
1496}
1497
1498static u8 pci_hdr_type(struct pci_dev *dev)
1499{
1500 u8 hdr_type;
1501
1502#ifdef CONFIG_PCI_IOV
1503 if (dev->is_virtfn)
1504 return dev->physfn->sriov->hdr_type;
1505#endif
1506 pci_read_config_byte(dev, PCI_HEADER_TYPE, &hdr_type);
1507 return hdr_type;
1508}
1509
1510#define LEGACY_IO_RESOURCE (IORESOURCE_IO | IORESOURCE_PCI_FIXED)
1511
1512static void pci_msi_setup_pci_dev(struct pci_dev *dev)
1513{
1514 /*
1515 * Disable the MSI hardware to avoid screaming interrupts
1516 * during boot. This is the power on reset default so
1517 * usually this should be a noop.
1518 */
1519 dev->msi_cap = pci_find_capability(dev, PCI_CAP_ID_MSI);
1520 if (dev->msi_cap)
1521 pci_msi_set_enable(dev, 0);
1522
1523 dev->msix_cap = pci_find_capability(dev, PCI_CAP_ID_MSIX);
1524 if (dev->msix_cap)
1525 pci_msix_clear_and_set_ctrl(dev, PCI_MSIX_FLAGS_ENABLE, 0);
1526}
1527
1528/**
1529 * pci_intx_mask_broken - Test PCI_COMMAND_INTX_DISABLE writability
1530 * @dev: PCI device
1531 *
1532 * Test whether PCI_COMMAND_INTX_DISABLE is writable for @dev. Check this
1533 * at enumeration-time to avoid modifying PCI_COMMAND at run-time.
1534 */
1535static int pci_intx_mask_broken(struct pci_dev *dev)
1536{
1537 u16 orig, toggle, new;
1538
1539 pci_read_config_word(dev, PCI_COMMAND, &orig);
1540 toggle = orig ^ PCI_COMMAND_INTX_DISABLE;
1541 pci_write_config_word(dev, PCI_COMMAND, toggle);
1542 pci_read_config_word(dev, PCI_COMMAND, &new);
1543
1544 pci_write_config_word(dev, PCI_COMMAND, orig);
1545
1546 /*
1547 * PCI_COMMAND_INTX_DISABLE was reserved and read-only prior to PCI
1548 * r2.3, so strictly speaking, a device is not *broken* if it's not
1549 * writable. But we'll live with the misnomer for now.
1550 */
1551 if (new != toggle)
1552 return 1;
1553 return 0;
1554}
1555
1556static void early_dump_pci_device(struct pci_dev *pdev)
1557{
1558 u32 value[256 / 4];
1559 int i;
1560
1561 pci_info(pdev, "config space:\n");
1562
1563 for (i = 0; i < 256; i += 4)
1564 pci_read_config_dword(pdev, i, &value[i / 4]);
1565
1566 print_hex_dump(KERN_INFO, "", DUMP_PREFIX_OFFSET, 16, 1,
1567 value, 256, false);
1568}
1569
1570/**
1571 * pci_setup_device - Fill in class and map information of a device
1572 * @dev: the device structure to fill
1573 *
1574 * Initialize the device structure with information about the device's
1575 * vendor,class,memory and IO-space addresses, IRQ lines etc.
1576 * Called at initialisation of the PCI subsystem and by CardBus services.
1577 * Returns 0 on success and negative if unknown type of device (not normal,
1578 * bridge or CardBus).
1579 */
1580int pci_setup_device(struct pci_dev *dev)
1581{
1582 u32 class;
1583 u16 cmd;
1584 u8 hdr_type;
1585 int pos = 0;
1586 struct pci_bus_region region;
1587 struct resource *res;
1588
1589 hdr_type = pci_hdr_type(dev);
1590
1591 dev->sysdata = dev->bus->sysdata;
1592 dev->dev.parent = dev->bus->bridge;
1593 dev->dev.bus = &pci_bus_type;
1594 dev->hdr_type = hdr_type & 0x7f;
1595 dev->multifunction = !!(hdr_type & 0x80);
1596 dev->error_state = pci_channel_io_normal;
1597 set_pcie_port_type(dev);
1598
1599 pci_dev_assign_slot(dev);
1600
1601 /*
1602 * Assume 32-bit PCI; let 64-bit PCI cards (which are far rarer)
1603 * set this higher, assuming the system even supports it.
1604 */
1605 dev->dma_mask = 0xffffffff;
1606
1607 dev_set_name(&dev->dev, "%04x:%02x:%02x.%d", pci_domain_nr(dev->bus),
1608 dev->bus->number, PCI_SLOT(dev->devfn),
1609 PCI_FUNC(dev->devfn));
1610
1611 class = pci_class(dev);
1612
1613 dev->revision = class & 0xff;
1614 dev->class = class >> 8; /* upper 3 bytes */
1615
1616 pci_printk(KERN_DEBUG, dev, "[%04x:%04x] type %02x class %#08x\n",
1617 dev->vendor, dev->device, dev->hdr_type, dev->class);
1618
1619 if (pci_early_dump)
1620 early_dump_pci_device(dev);
1621
1622 /* Need to have dev->class ready */
1623 dev->cfg_size = pci_cfg_space_size(dev);
1624
1625 /* Need to have dev->cfg_size ready */
1626 set_pcie_thunderbolt(dev);
1627
1628 /* "Unknown power state" */
1629 dev->current_state = PCI_UNKNOWN;
1630
1631 /* Early fixups, before probing the BARs */
1632 pci_fixup_device(pci_fixup_early, dev);
1633
1634 /* Device class may be changed after fixup */
1635 class = dev->class >> 8;
1636
1637 if (dev->non_compliant_bars) {
1638 pci_read_config_word(dev, PCI_COMMAND, &cmd);
1639 if (cmd & (PCI_COMMAND_IO | PCI_COMMAND_MEMORY)) {
1640 pci_info(dev, "device has non-compliant BARs; disabling IO/MEM decoding\n");
1641 cmd &= ~PCI_COMMAND_IO;
1642 cmd &= ~PCI_COMMAND_MEMORY;
1643 pci_write_config_word(dev, PCI_COMMAND, cmd);
1644 }
1645 }
1646
1647 dev->broken_intx_masking = pci_intx_mask_broken(dev);
1648
1649 switch (dev->hdr_type) { /* header type */
1650 case PCI_HEADER_TYPE_NORMAL: /* standard header */
1651 if (class == PCI_CLASS_BRIDGE_PCI)
1652 goto bad;
1653 pci_read_irq(dev);
1654 pci_read_bases(dev, 6, PCI_ROM_ADDRESS);
1655
1656 pci_subsystem_ids(dev, &dev->subsystem_vendor, &dev->subsystem_device);
1657
1658 /*
1659 * Do the ugly legacy mode stuff here rather than broken chip
1660 * quirk code. Legacy mode ATA controllers have fixed
1661 * addresses. These are not always echoed in BAR0-3, and
1662 * BAR0-3 in a few cases contain junk!
1663 */
1664 if (class == PCI_CLASS_STORAGE_IDE) {
1665 u8 progif;
1666 pci_read_config_byte(dev, PCI_CLASS_PROG, &progif);
1667 if ((progif & 1) == 0) {
1668 region.start = 0x1F0;
1669 region.end = 0x1F7;
1670 res = &dev->resource[0];
1671 res->flags = LEGACY_IO_RESOURCE;
1672 pcibios_bus_to_resource(dev->bus, res, &region);
1673 pci_info(dev, "legacy IDE quirk: reg 0x10: %pR\n",
1674 res);
1675 region.start = 0x3F6;
1676 region.end = 0x3F6;
1677 res = &dev->resource[1];
1678 res->flags = LEGACY_IO_RESOURCE;
1679 pcibios_bus_to_resource(dev->bus, res, &region);
1680 pci_info(dev, "legacy IDE quirk: reg 0x14: %pR\n",
1681 res);
1682 }
1683 if ((progif & 4) == 0) {
1684 region.start = 0x170;
1685 region.end = 0x177;
1686 res = &dev->resource[2];
1687 res->flags = LEGACY_IO_RESOURCE;
1688 pcibios_bus_to_resource(dev->bus, res, &region);
1689 pci_info(dev, "legacy IDE quirk: reg 0x18: %pR\n",
1690 res);
1691 region.start = 0x376;
1692 region.end = 0x376;
1693 res = &dev->resource[3];
1694 res->flags = LEGACY_IO_RESOURCE;
1695 pcibios_bus_to_resource(dev->bus, res, &region);
1696 pci_info(dev, "legacy IDE quirk: reg 0x1c: %pR\n",
1697 res);
1698 }
1699 }
1700 break;
1701
1702 case PCI_HEADER_TYPE_BRIDGE: /* bridge header */
1703 if (class != PCI_CLASS_BRIDGE_PCI)
1704 goto bad;
1705
1706 /*
1707 * The PCI-to-PCI bridge spec requires that subtractive
1708 * decoding (i.e. transparent) bridge must have programming
1709 * interface code of 0x01.
1710 */
1711 pci_read_irq(dev);
1712 dev->transparent = ((dev->class & 0xff) == 1);
1713 pci_read_bases(dev, 2, PCI_ROM_ADDRESS1);
1714 set_pcie_hotplug_bridge(dev);
1715 pos = pci_find_capability(dev, PCI_CAP_ID_SSVID);
1716 if (pos) {
1717 pci_read_config_word(dev, pos + PCI_SSVID_VENDOR_ID, &dev->subsystem_vendor);
1718 pci_read_config_word(dev, pos + PCI_SSVID_DEVICE_ID, &dev->subsystem_device);
1719 }
1720 break;
1721
1722 case PCI_HEADER_TYPE_CARDBUS: /* CardBus bridge header */
1723 if (class != PCI_CLASS_BRIDGE_CARDBUS)
1724 goto bad;
1725 pci_read_irq(dev);
1726 pci_read_bases(dev, 1, 0);
1727 pci_read_config_word(dev, PCI_CB_SUBSYSTEM_VENDOR_ID, &dev->subsystem_vendor);
1728 pci_read_config_word(dev, PCI_CB_SUBSYSTEM_ID, &dev->subsystem_device);
1729 break;
1730
1731 default: /* unknown header */
1732 pci_err(dev, "unknown header type %02x, ignoring device\n",
1733 dev->hdr_type);
1734 return -EIO;
1735
1736 bad:
1737 pci_err(dev, "ignoring class %#08x (doesn't match header type %02x)\n",
1738 dev->class, dev->hdr_type);
1739 dev->class = PCI_CLASS_NOT_DEFINED << 8;
1740 }
1741
1742 /* We found a fine healthy device, go go go... */
1743 return 0;
1744}
1745
1746static void pci_configure_mps(struct pci_dev *dev)
1747{
1748 struct pci_dev *bridge = pci_upstream_bridge(dev);
1749 int mps, mpss, p_mps, rc;
1750
1751 if (!pci_is_pcie(dev) || !bridge || !pci_is_pcie(bridge))
1752 return;
1753
1754 /* MPS and MRRS fields are of type 'RsvdP' for VFs, short-circuit out */
1755 if (dev->is_virtfn)
1756 return;
1757
1758 mps = pcie_get_mps(dev);
1759 p_mps = pcie_get_mps(bridge);
1760
1761 if (mps == p_mps)
1762 return;
1763
1764 if (pcie_bus_config == PCIE_BUS_TUNE_OFF) {
1765 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",
1766 mps, pci_name(bridge), p_mps);
1767 return;
1768 }
1769
1770 /*
1771 * Fancier MPS configuration is done later by
1772 * pcie_bus_configure_settings()
1773 */
1774 if (pcie_bus_config != PCIE_BUS_DEFAULT)
1775 return;
1776
1777 mpss = 128 << dev->pcie_mpss;
1778 if (mpss < p_mps && pci_pcie_type(bridge) == PCI_EXP_TYPE_ROOT_PORT) {
1779 pcie_set_mps(bridge, mpss);
1780 pci_info(dev, "Upstream bridge's Max Payload Size set to %d (was %d, max %d)\n",
1781 mpss, p_mps, 128 << bridge->pcie_mpss);
1782 p_mps = pcie_get_mps(bridge);
1783 }
1784
1785 rc = pcie_set_mps(dev, p_mps);
1786 if (rc) {
1787 pci_warn(dev, "can't set Max Payload Size to %d; if necessary, use \"pci=pcie_bus_safe\" and report a bug\n",
1788 p_mps);
1789 return;
1790 }
1791
1792 pci_info(dev, "Max Payload Size set to %d (was %d, max %d)\n",
1793 p_mps, mps, mpss);
1794}
1795
1796static struct hpp_type0 pci_default_type0 = {
1797 .revision = 1,
1798 .cache_line_size = 8,
1799 .latency_timer = 0x40,
1800 .enable_serr = 0,
1801 .enable_perr = 0,
1802};
1803
1804static void program_hpp_type0(struct pci_dev *dev, struct hpp_type0 *hpp)
1805{
1806 u16 pci_cmd, pci_bctl;
1807
1808 if (!hpp)
1809 hpp = &pci_default_type0;
1810
1811 if (hpp->revision > 1) {
1812 pci_warn(dev, "PCI settings rev %d not supported; using defaults\n",
1813 hpp->revision);
1814 hpp = &pci_default_type0;
1815 }
1816
1817 pci_write_config_byte(dev, PCI_CACHE_LINE_SIZE, hpp->cache_line_size);
1818 pci_write_config_byte(dev, PCI_LATENCY_TIMER, hpp->latency_timer);
1819 pci_read_config_word(dev, PCI_COMMAND, &pci_cmd);
1820 if (hpp->enable_serr)
1821 pci_cmd |= PCI_COMMAND_SERR;
1822 if (hpp->enable_perr)
1823 pci_cmd |= PCI_COMMAND_PARITY;
1824 pci_write_config_word(dev, PCI_COMMAND, pci_cmd);
1825
1826 /* Program bridge control value */
1827 if ((dev->class >> 8) == PCI_CLASS_BRIDGE_PCI) {
1828 pci_write_config_byte(dev, PCI_SEC_LATENCY_TIMER,
1829 hpp->latency_timer);
1830 pci_read_config_word(dev, PCI_BRIDGE_CONTROL, &pci_bctl);
1831 if (hpp->enable_serr)
1832 pci_bctl |= PCI_BRIDGE_CTL_SERR;
1833 if (hpp->enable_perr)
1834 pci_bctl |= PCI_BRIDGE_CTL_PARITY;
1835 pci_write_config_word(dev, PCI_BRIDGE_CONTROL, pci_bctl);
1836 }
1837}
1838
1839static void program_hpp_type1(struct pci_dev *dev, struct hpp_type1 *hpp)
1840{
1841 int pos;
1842
1843 if (!hpp)
1844 return;
1845
1846 pos = pci_find_capability(dev, PCI_CAP_ID_PCIX);
1847 if (!pos)
1848 return;
1849
1850 pci_warn(dev, "PCI-X settings not supported\n");
1851}
1852
1853static bool pcie_root_rcb_set(struct pci_dev *dev)
1854{
1855 struct pci_dev *rp = pcie_find_root_port(dev);
1856 u16 lnkctl;
1857
1858 if (!rp)
1859 return false;
1860
1861 pcie_capability_read_word(rp, PCI_EXP_LNKCTL, &lnkctl);
1862 if (lnkctl & PCI_EXP_LNKCTL_RCB)
1863 return true;
1864
1865 return false;
1866}
1867
1868static void program_hpp_type2(struct pci_dev *dev, struct hpp_type2 *hpp)
1869{
1870 int pos;
1871 u32 reg32;
1872
1873 if (!hpp)
1874 return;
1875
1876 if (!pci_is_pcie(dev))
1877 return;
1878
1879 if (hpp->revision > 1) {
1880 pci_warn(dev, "PCIe settings rev %d not supported\n",
1881 hpp->revision);
1882 return;
1883 }
1884
1885 /*
1886 * Don't allow _HPX to change MPS or MRRS settings. We manage
1887 * those to make sure they're consistent with the rest of the
1888 * platform.
1889 */
1890 hpp->pci_exp_devctl_and |= PCI_EXP_DEVCTL_PAYLOAD |
1891 PCI_EXP_DEVCTL_READRQ;
1892 hpp->pci_exp_devctl_or &= ~(PCI_EXP_DEVCTL_PAYLOAD |
1893 PCI_EXP_DEVCTL_READRQ);
1894
1895 /* Initialize Device Control Register */
1896 pcie_capability_clear_and_set_word(dev, PCI_EXP_DEVCTL,
1897 ~hpp->pci_exp_devctl_and, hpp->pci_exp_devctl_or);
1898
1899 /* Initialize Link Control Register */
1900 if (pcie_cap_has_lnkctl(dev)) {
1901
1902 /*
1903 * If the Root Port supports Read Completion Boundary of
1904 * 128, set RCB to 128. Otherwise, clear it.
1905 */
1906 hpp->pci_exp_lnkctl_and |= PCI_EXP_LNKCTL_RCB;
1907 hpp->pci_exp_lnkctl_or &= ~PCI_EXP_LNKCTL_RCB;
1908 if (pcie_root_rcb_set(dev))
1909 hpp->pci_exp_lnkctl_or |= PCI_EXP_LNKCTL_RCB;
1910
1911 pcie_capability_clear_and_set_word(dev, PCI_EXP_LNKCTL,
1912 ~hpp->pci_exp_lnkctl_and, hpp->pci_exp_lnkctl_or);
1913 }
1914
1915 /* Find Advanced Error Reporting Enhanced Capability */
1916 pos = pci_find_ext_capability(dev, PCI_EXT_CAP_ID_ERR);
1917 if (!pos)
1918 return;
1919
1920 /* Initialize Uncorrectable Error Mask Register */
1921 pci_read_config_dword(dev, pos + PCI_ERR_UNCOR_MASK, &reg32);
1922 reg32 = (reg32 & hpp->unc_err_mask_and) | hpp->unc_err_mask_or;
1923 pci_write_config_dword(dev, pos + PCI_ERR_UNCOR_MASK, reg32);
1924
1925 /* Initialize Uncorrectable Error Severity Register */
1926 pci_read_config_dword(dev, pos + PCI_ERR_UNCOR_SEVER, &reg32);
1927 reg32 = (reg32 & hpp->unc_err_sever_and) | hpp->unc_err_sever_or;
1928 pci_write_config_dword(dev, pos + PCI_ERR_UNCOR_SEVER, reg32);
1929
1930 /* Initialize Correctable Error Mask Register */
1931 pci_read_config_dword(dev, pos + PCI_ERR_COR_MASK, &reg32);
1932 reg32 = (reg32 & hpp->cor_err_mask_and) | hpp->cor_err_mask_or;
1933 pci_write_config_dword(dev, pos + PCI_ERR_COR_MASK, reg32);
1934
1935 /* Initialize Advanced Error Capabilities and Control Register */
1936 pci_read_config_dword(dev, pos + PCI_ERR_CAP, &reg32);
1937 reg32 = (reg32 & hpp->adv_err_cap_and) | hpp->adv_err_cap_or;
1938
1939 /* Don't enable ECRC generation or checking if unsupported */
1940 if (!(reg32 & PCI_ERR_CAP_ECRC_GENC))
1941 reg32 &= ~PCI_ERR_CAP_ECRC_GENE;
1942 if (!(reg32 & PCI_ERR_CAP_ECRC_CHKC))
1943 reg32 &= ~PCI_ERR_CAP_ECRC_CHKE;
1944 pci_write_config_dword(dev, pos + PCI_ERR_CAP, reg32);
1945
1946 /*
1947 * FIXME: The following two registers are not supported yet.
1948 *
1949 * o Secondary Uncorrectable Error Severity Register
1950 * o Secondary Uncorrectable Error Mask Register
1951 */
1952}
1953
1954int pci_configure_extended_tags(struct pci_dev *dev, void *ign)
1955{
1956 struct pci_host_bridge *host;
1957 u32 cap;
1958 u16 ctl;
1959 int ret;
1960
1961 if (!pci_is_pcie(dev))
1962 return 0;
1963
1964 ret = pcie_capability_read_dword(dev, PCI_EXP_DEVCAP, &cap);
1965 if (ret)
1966 return 0;
1967
1968 if (!(cap & PCI_EXP_DEVCAP_EXT_TAG))
1969 return 0;
1970
1971 ret = pcie_capability_read_word(dev, PCI_EXP_DEVCTL, &ctl);
1972 if (ret)
1973 return 0;
1974
1975 host = pci_find_host_bridge(dev->bus);
1976 if (!host)
1977 return 0;
1978
1979 /*
1980 * If some device in the hierarchy doesn't handle Extended Tags
1981 * correctly, make sure they're disabled.
1982 */
1983 if (host->no_ext_tags) {
1984 if (ctl & PCI_EXP_DEVCTL_EXT_TAG) {
1985 pci_info(dev, "disabling Extended Tags\n");
1986 pcie_capability_clear_word(dev, PCI_EXP_DEVCTL,
1987 PCI_EXP_DEVCTL_EXT_TAG);
1988 }
1989 return 0;
1990 }
1991
1992 if (!(ctl & PCI_EXP_DEVCTL_EXT_TAG)) {
1993 pci_info(dev, "enabling Extended Tags\n");
1994 pcie_capability_set_word(dev, PCI_EXP_DEVCTL,
1995 PCI_EXP_DEVCTL_EXT_TAG);
1996 }
1997 return 0;
1998}
1999
2000/**
2001 * pcie_relaxed_ordering_enabled - Probe for PCIe relaxed ordering enable
2002 * @dev: PCI device to query
2003 *
2004 * Returns true if the device has enabled relaxed ordering attribute.
2005 */
2006bool pcie_relaxed_ordering_enabled(struct pci_dev *dev)
2007{
2008 u16 v;
2009
2010 pcie_capability_read_word(dev, PCI_EXP_DEVCTL, &v);
2011
2012 return !!(v & PCI_EXP_DEVCTL_RELAX_EN);
2013}
2014EXPORT_SYMBOL(pcie_relaxed_ordering_enabled);
2015
2016static void pci_configure_relaxed_ordering(struct pci_dev *dev)
2017{
2018 struct pci_dev *root;
2019
2020 /* PCI_EXP_DEVICE_RELAX_EN is RsvdP in VFs */
2021 if (dev->is_virtfn)
2022 return;
2023
2024 if (!pcie_relaxed_ordering_enabled(dev))
2025 return;
2026
2027 /*
2028 * For now, we only deal with Relaxed Ordering issues with Root
2029 * Ports. Peer-to-Peer DMA is another can of worms.
2030 */
2031 root = pci_find_pcie_root_port(dev);
2032 if (!root)
2033 return;
2034
2035 if (root->dev_flags & PCI_DEV_FLAGS_NO_RELAXED_ORDERING) {
2036 pcie_capability_clear_word(dev, PCI_EXP_DEVCTL,
2037 PCI_EXP_DEVCTL_RELAX_EN);
2038 pci_info(dev, "Relaxed Ordering disabled because the Root Port didn't support it\n");
2039 }
2040}
2041
2042static void pci_configure_ltr(struct pci_dev *dev)
2043{
2044#ifdef CONFIG_PCIEASPM
2045 struct pci_host_bridge *host = pci_find_host_bridge(dev->bus);
2046 struct pci_dev *bridge;
2047 u32 cap, ctl;
2048
2049 if (!pci_is_pcie(dev))
2050 return;
2051
2052 pcie_capability_read_dword(dev, PCI_EXP_DEVCAP2, &cap);
2053 if (!(cap & PCI_EXP_DEVCAP2_LTR))
2054 return;
2055
2056 pcie_capability_read_dword(dev, PCI_EXP_DEVCTL2, &ctl);
2057 if (ctl & PCI_EXP_DEVCTL2_LTR_EN) {
2058 if (pci_pcie_type(dev) == PCI_EXP_TYPE_ROOT_PORT) {
2059 dev->ltr_path = 1;
2060 return;
2061 }
2062
2063 bridge = pci_upstream_bridge(dev);
2064 if (bridge && bridge->ltr_path)
2065 dev->ltr_path = 1;
2066
2067 return;
2068 }
2069
2070 if (!host->native_ltr)
2071 return;
2072
2073 /*
2074 * Software must not enable LTR in an Endpoint unless the Root
2075 * Complex and all intermediate Switches indicate support for LTR.
2076 * PCIe r4.0, sec 6.18.
2077 */
2078 if (pci_pcie_type(dev) == PCI_EXP_TYPE_ROOT_PORT ||
2079 ((bridge = pci_upstream_bridge(dev)) &&
2080 bridge->ltr_path)) {
2081 pcie_capability_set_word(dev, PCI_EXP_DEVCTL2,
2082 PCI_EXP_DEVCTL2_LTR_EN);
2083 dev->ltr_path = 1;
2084 }
2085#endif
2086}
2087
2088static void pci_configure_eetlp_prefix(struct pci_dev *dev)
2089{
2090#ifdef CONFIG_PCI_PASID
2091 struct pci_dev *bridge;
2092 int pcie_type;
2093 u32 cap;
2094
2095 if (!pci_is_pcie(dev))
2096 return;
2097
2098 pcie_capability_read_dword(dev, PCI_EXP_DEVCAP2, &cap);
2099 if (!(cap & PCI_EXP_DEVCAP2_EE_PREFIX))
2100 return;
2101
2102 pcie_type = pci_pcie_type(dev);
2103 if (pcie_type == PCI_EXP_TYPE_ROOT_PORT ||
2104 pcie_type == PCI_EXP_TYPE_RC_END)
2105 dev->eetlp_prefix_path = 1;
2106 else {
2107 bridge = pci_upstream_bridge(dev);
2108 if (bridge && bridge->eetlp_prefix_path)
2109 dev->eetlp_prefix_path = 1;
2110 }
2111#endif
2112}
2113
2114static void pci_configure_device(struct pci_dev *dev)
2115{
2116 struct hotplug_params hpp;
2117 int ret;
2118
2119 pci_configure_mps(dev);
2120 pci_configure_extended_tags(dev, NULL);
2121 pci_configure_relaxed_ordering(dev);
2122 pci_configure_ltr(dev);
2123 pci_configure_eetlp_prefix(dev);
2124
2125 memset(&hpp, 0, sizeof(hpp));
2126 ret = pci_get_hp_params(dev, &hpp);
2127 if (ret)
2128 return;
2129
2130 program_hpp_type2(dev, hpp.t2);
2131 program_hpp_type1(dev, hpp.t1);
2132 program_hpp_type0(dev, hpp.t0);
2133}
2134
2135static void pci_release_capabilities(struct pci_dev *dev)
2136{
2137 pci_aer_exit(dev);
2138 pci_vpd_release(dev);
2139 pci_iov_release(dev);
2140 pci_free_cap_save_buffers(dev);
2141}
2142
2143/**
2144 * pci_release_dev - Free a PCI device structure when all users of it are
2145 * finished
2146 * @dev: device that's been disconnected
2147 *
2148 * Will be called only by the device core when all users of this PCI device are
2149 * done.
2150 */
2151static void pci_release_dev(struct device *dev)
2152{
2153 struct pci_dev *pci_dev;
2154
2155 pci_dev = to_pci_dev(dev);
2156 pci_release_capabilities(pci_dev);
2157 pci_release_of_node(pci_dev);
2158 pcibios_release_device(pci_dev);
2159 pci_bus_put(pci_dev->bus);
2160 kfree(pci_dev->driver_override);
2161 kfree(pci_dev->dma_alias_mask);
2162 kfree(pci_dev);
2163}
2164
2165struct pci_dev *pci_alloc_dev(struct pci_bus *bus)
2166{
2167 struct pci_dev *dev;
2168
2169 dev = kzalloc(sizeof(struct pci_dev), GFP_KERNEL);
2170 if (!dev)
2171 return NULL;
2172
2173 INIT_LIST_HEAD(&dev->bus_list);
2174 dev->dev.type = &pci_dev_type;
2175 dev->bus = pci_bus_get(bus);
2176
2177 return dev;
2178}
2179EXPORT_SYMBOL(pci_alloc_dev);
2180
2181static bool pci_bus_crs_vendor_id(u32 l)
2182{
2183 return (l & 0xffff) == 0x0001;
2184}
2185
2186static bool pci_bus_wait_crs(struct pci_bus *bus, int devfn, u32 *l,
2187 int timeout)
2188{
2189 int delay = 1;
2190
2191 if (!pci_bus_crs_vendor_id(*l))
2192 return true; /* not a CRS completion */
2193
2194 if (!timeout)
2195 return false; /* CRS, but caller doesn't want to wait */
2196
2197 /*
2198 * We got the reserved Vendor ID that indicates a completion with
2199 * Configuration Request Retry Status (CRS). Retry until we get a
2200 * valid Vendor ID or we time out.
2201 */
2202 while (pci_bus_crs_vendor_id(*l)) {
2203 if (delay > timeout) {
2204 pr_warn("pci %04x:%02x:%02x.%d: not ready after %dms; giving up\n",
2205 pci_domain_nr(bus), bus->number,
2206 PCI_SLOT(devfn), PCI_FUNC(devfn), delay - 1);
2207
2208 return false;
2209 }
2210 if (delay >= 1000)
2211 pr_info("pci %04x:%02x:%02x.%d: not ready after %dms; waiting\n",
2212 pci_domain_nr(bus), bus->number,
2213 PCI_SLOT(devfn), PCI_FUNC(devfn), delay - 1);
2214
2215 msleep(delay);
2216 delay *= 2;
2217
2218 if (pci_bus_read_config_dword(bus, devfn, PCI_VENDOR_ID, l))
2219 return false;
2220 }
2221
2222 if (delay >= 1000)
2223 pr_info("pci %04x:%02x:%02x.%d: ready after %dms\n",
2224 pci_domain_nr(bus), bus->number,
2225 PCI_SLOT(devfn), PCI_FUNC(devfn), delay - 1);
2226
2227 return true;
2228}
2229
2230bool pci_bus_generic_read_dev_vendor_id(struct pci_bus *bus, int devfn, u32 *l,
2231 int timeout)
2232{
2233 if (pci_bus_read_config_dword(bus, devfn, PCI_VENDOR_ID, l))
2234 return false;
2235
2236 /* Some broken boards return 0 or ~0 if a slot is empty: */
2237 if (*l == 0xffffffff || *l == 0x00000000 ||
2238 *l == 0x0000ffff || *l == 0xffff0000)
2239 return false;
2240
2241 if (pci_bus_crs_vendor_id(*l))
2242 return pci_bus_wait_crs(bus, devfn, l, timeout);
2243
2244 return true;
2245}
2246
2247bool pci_bus_read_dev_vendor_id(struct pci_bus *bus, int devfn, u32 *l,
2248 int timeout)
2249{
2250#ifdef CONFIG_PCI_QUIRKS
2251 struct pci_dev *bridge = bus->self;
2252
2253 /*
2254 * Certain IDT switches have an issue where they improperly trigger
2255 * ACS Source Validation errors on completions for config reads.
2256 */
2257 if (bridge && bridge->vendor == PCI_VENDOR_ID_IDT &&
2258 bridge->device == 0x80b5)
2259 return pci_idt_bus_quirk(bus, devfn, l, timeout);
2260#endif
2261
2262 return pci_bus_generic_read_dev_vendor_id(bus, devfn, l, timeout);
2263}
2264EXPORT_SYMBOL(pci_bus_read_dev_vendor_id);
2265
2266/*
2267 * Read the config data for a PCI device, sanity-check it,
2268 * and fill in the dev structure.
2269 */
2270static struct pci_dev *pci_scan_device(struct pci_bus *bus, int devfn)
2271{
2272 struct pci_dev *dev;
2273 u32 l;
2274
2275 if (!pci_bus_read_dev_vendor_id(bus, devfn, &l, 60*1000))
2276 return NULL;
2277
2278 dev = pci_alloc_dev(bus);
2279 if (!dev)
2280 return NULL;
2281
2282 dev->devfn = devfn;
2283 dev->vendor = l & 0xffff;
2284 dev->device = (l >> 16) & 0xffff;
2285
2286 pci_set_of_node(dev);
2287
2288 if (pci_setup_device(dev)) {
2289 pci_bus_put(dev->bus);
2290 kfree(dev);
2291 return NULL;
2292 }
2293
2294 return dev;
2295}
2296
2297static void pcie_report_downtraining(struct pci_dev *dev)
2298{
2299 if (!pci_is_pcie(dev))
2300 return;
2301
2302 /* Look from the device up to avoid downstream ports with no devices */
2303 if ((pci_pcie_type(dev) != PCI_EXP_TYPE_ENDPOINT) &&
2304 (pci_pcie_type(dev) != PCI_EXP_TYPE_LEG_END) &&
2305 (pci_pcie_type(dev) != PCI_EXP_TYPE_UPSTREAM))
2306 return;
2307
2308 /* Multi-function PCIe devices share the same link/status */
2309 if (PCI_FUNC(dev->devfn) != 0 || dev->is_virtfn)
2310 return;
2311
2312 /* Print link status only if the device is constrained by the fabric */
2313 __pcie_print_link_status(dev, false);
2314}
2315
2316static void pci_init_capabilities(struct pci_dev *dev)
2317{
2318 /* Enhanced Allocation */
2319 pci_ea_init(dev);
2320
2321 /* Setup MSI caps & disable MSI/MSI-X interrupts */
2322 pci_msi_setup_pci_dev(dev);
2323
2324 /* Buffers for saving PCIe and PCI-X capabilities */
2325 pci_allocate_cap_save_buffers(dev);
2326
2327 /* Power Management */
2328 pci_pm_init(dev);
2329
2330 /* Vital Product Data */
2331 pci_vpd_init(dev);
2332
2333 /* Alternative Routing-ID Forwarding */
2334 pci_configure_ari(dev);
2335
2336 /* Single Root I/O Virtualization */
2337 pci_iov_init(dev);
2338
2339 /* Address Translation Services */
2340 pci_ats_init(dev);
2341
2342 /* Enable ACS P2P upstream forwarding */
2343 pci_enable_acs(dev);
2344
2345 /* Precision Time Measurement */
2346 pci_ptm_init(dev);
2347
2348 /* Advanced Error Reporting */
2349 pci_aer_init(dev);
2350
2351 pcie_report_downtraining(dev);
2352
2353 if (pci_probe_reset_function(dev) == 0)
2354 dev->reset_fn = 1;
2355}
2356
2357/*
2358 * This is the equivalent of pci_host_bridge_msi_domain() that acts on
2359 * devices. Firmware interfaces that can select the MSI domain on a
2360 * per-device basis should be called from here.
2361 */
2362static struct irq_domain *pci_dev_msi_domain(struct pci_dev *dev)
2363{
2364 struct irq_domain *d;
2365
2366 /*
2367 * If a domain has been set through the pcibios_add_device()
2368 * callback, then this is the one (platform code knows best).
2369 */
2370 d = dev_get_msi_domain(&dev->dev);
2371 if (d)
2372 return d;
2373
2374 /*
2375 * Let's see if we have a firmware interface able to provide
2376 * the domain.
2377 */
2378 d = pci_msi_get_device_domain(dev);
2379 if (d)
2380 return d;
2381
2382 return NULL;
2383}
2384
2385static void pci_set_msi_domain(struct pci_dev *dev)
2386{
2387 struct irq_domain *d;
2388
2389 /*
2390 * If the platform or firmware interfaces cannot supply a
2391 * device-specific MSI domain, then inherit the default domain
2392 * from the host bridge itself.
2393 */
2394 d = pci_dev_msi_domain(dev);
2395 if (!d)
2396 d = dev_get_msi_domain(&dev->bus->dev);
2397
2398 dev_set_msi_domain(&dev->dev, d);
2399}
2400
2401void pci_device_add(struct pci_dev *dev, struct pci_bus *bus)
2402{
2403 int ret;
2404
2405 pci_configure_device(dev);
2406
2407 device_initialize(&dev->dev);
2408 dev->dev.release = pci_release_dev;
2409
2410 set_dev_node(&dev->dev, pcibus_to_node(bus));
2411 dev->dev.dma_mask = &dev->dma_mask;
2412 dev->dev.dma_parms = &dev->dma_parms;
2413 dev->dev.coherent_dma_mask = 0xffffffffull;
2414
2415 pci_set_dma_max_seg_size(dev, 65536);
2416 pci_set_dma_seg_boundary(dev, 0xffffffff);
2417
2418 /* Fix up broken headers */
2419 pci_fixup_device(pci_fixup_header, dev);
2420
2421 /* Moved out from quirk header fixup code */
2422 pci_reassigndev_resource_alignment(dev);
2423
2424 /* Clear the state_saved flag */
2425 dev->state_saved = false;
2426
2427 /* Initialize various capabilities */
2428 pci_init_capabilities(dev);
2429
2430 /*
2431 * Add the device to our list of discovered devices
2432 * and the bus list for fixup functions, etc.
2433 */
2434 down_write(&pci_bus_sem);
2435 list_add_tail(&dev->bus_list, &bus->devices);
2436 up_write(&pci_bus_sem);
2437
2438 ret = pcibios_add_device(dev);
2439 WARN_ON(ret < 0);
2440
2441 /* Set up MSI IRQ domain */
2442 pci_set_msi_domain(dev);
2443
2444 /* Notifier could use PCI capabilities */
2445 dev->match_driver = false;
2446 ret = device_add(&dev->dev);
2447 WARN_ON(ret < 0);
2448}
2449
2450struct pci_dev *pci_scan_single_device(struct pci_bus *bus, int devfn)
2451{
2452 struct pci_dev *dev;
2453
2454 dev = pci_get_slot(bus, devfn);
2455 if (dev) {
2456 pci_dev_put(dev);
2457 return dev;
2458 }
2459
2460 dev = pci_scan_device(bus, devfn);
2461 if (!dev)
2462 return NULL;
2463
2464 pci_device_add(dev, bus);
2465
2466 return dev;
2467}
2468EXPORT_SYMBOL(pci_scan_single_device);
2469
2470static unsigned next_fn(struct pci_bus *bus, struct pci_dev *dev, unsigned fn)
2471{
2472 int pos;
2473 u16 cap = 0;
2474 unsigned next_fn;
2475
2476 if (pci_ari_enabled(bus)) {
2477 if (!dev)
2478 return 0;
2479 pos = pci_find_ext_capability(dev, PCI_EXT_CAP_ID_ARI);
2480 if (!pos)
2481 return 0;
2482
2483 pci_read_config_word(dev, pos + PCI_ARI_CAP, &cap);
2484 next_fn = PCI_ARI_CAP_NFN(cap);
2485 if (next_fn <= fn)
2486 return 0; /* protect against malformed list */
2487
2488 return next_fn;
2489 }
2490
2491 /* dev may be NULL for non-contiguous multifunction devices */
2492 if (!dev || dev->multifunction)
2493 return (fn + 1) % 8;
2494
2495 return 0;
2496}
2497
2498static int only_one_child(struct pci_bus *bus)
2499{
2500 struct pci_dev *bridge = bus->self;
2501
2502 /*
2503 * Systems with unusual topologies set PCI_SCAN_ALL_PCIE_DEVS so
2504 * we scan for all possible devices, not just Device 0.
2505 */
2506 if (pci_has_flag(PCI_SCAN_ALL_PCIE_DEVS))
2507 return 0;
2508
2509 /*
2510 * A PCIe Downstream Port normally leads to a Link with only Device
2511 * 0 on it (PCIe spec r3.1, sec 7.3.1). As an optimization, scan
2512 * only for Device 0 in that situation.
2513 *
2514 * Checking has_secondary_link is a hack to identify Downstream
2515 * Ports because sometimes Switches are configured such that the
2516 * PCIe Port Type labels are backwards.
2517 */
2518 if (bridge && pci_is_pcie(bridge) && bridge->has_secondary_link)
2519 return 1;
2520
2521 return 0;
2522}
2523
2524/**
2525 * pci_scan_slot - Scan a PCI slot on a bus for devices
2526 * @bus: PCI bus to scan
2527 * @devfn: slot number to scan (must have zero function)
2528 *
2529 * Scan a PCI slot on the specified PCI bus for devices, adding
2530 * discovered devices to the @bus->devices list. New devices
2531 * will not have is_added set.
2532 *
2533 * Returns the number of new devices found.
2534 */
2535int pci_scan_slot(struct pci_bus *bus, int devfn)
2536{
2537 unsigned fn, nr = 0;
2538 struct pci_dev *dev;
2539
2540 if (only_one_child(bus) && (devfn > 0))
2541 return 0; /* Already scanned the entire slot */
2542
2543 dev = pci_scan_single_device(bus, devfn);
2544 if (!dev)
2545 return 0;
2546 if (!pci_dev_is_added(dev))
2547 nr++;
2548
2549 for (fn = next_fn(bus, dev, 0); fn > 0; fn = next_fn(bus, dev, fn)) {
2550 dev = pci_scan_single_device(bus, devfn + fn);
2551 if (dev) {
2552 if (!pci_dev_is_added(dev))
2553 nr++;
2554 dev->multifunction = 1;
2555 }
2556 }
2557
2558 /* Only one slot has PCIe device */
2559 if (bus->self && nr)
2560 pcie_aspm_init_link_state(bus->self);
2561
2562 return nr;
2563}
2564EXPORT_SYMBOL(pci_scan_slot);
2565
2566static int pcie_find_smpss(struct pci_dev *dev, void *data)
2567{
2568 u8 *smpss = data;
2569
2570 if (!pci_is_pcie(dev))
2571 return 0;
2572
2573 /*
2574 * We don't have a way to change MPS settings on devices that have
2575 * drivers attached. A hot-added device might support only the minimum
2576 * MPS setting (MPS=128). Therefore, if the fabric contains a bridge
2577 * where devices may be hot-added, we limit the fabric MPS to 128 so
2578 * hot-added devices will work correctly.
2579 *
2580 * However, if we hot-add a device to a slot directly below a Root
2581 * Port, it's impossible for there to be other existing devices below
2582 * the port. We don't limit the MPS in this case because we can
2583 * reconfigure MPS on both the Root Port and the hot-added device,
2584 * and there are no other devices involved.
2585 *
2586 * Note that this PCIE_BUS_SAFE path assumes no peer-to-peer DMA.
2587 */
2588 if (dev->is_hotplug_bridge &&
2589 pci_pcie_type(dev) != PCI_EXP_TYPE_ROOT_PORT)
2590 *smpss = 0;
2591
2592 if (*smpss > dev->pcie_mpss)
2593 *smpss = dev->pcie_mpss;
2594
2595 return 0;
2596}
2597
2598static void pcie_write_mps(struct pci_dev *dev, int mps)
2599{
2600 int rc;
2601
2602 if (pcie_bus_config == PCIE_BUS_PERFORMANCE) {
2603 mps = 128 << dev->pcie_mpss;
2604
2605 if (pci_pcie_type(dev) != PCI_EXP_TYPE_ROOT_PORT &&
2606 dev->bus->self)
2607
2608 /*
2609 * For "Performance", the assumption is made that
2610 * downstream communication will never be larger than
2611 * the MRRS. So, the MPS only needs to be configured
2612 * for the upstream communication. This being the case,
2613 * walk from the top down and set the MPS of the child
2614 * to that of the parent bus.
2615 *
2616 * Configure the device MPS with the smaller of the
2617 * device MPSS or the bridge MPS (which is assumed to be
2618 * properly configured at this point to the largest
2619 * allowable MPS based on its parent bus).
2620 */
2621 mps = min(mps, pcie_get_mps(dev->bus->self));
2622 }
2623
2624 rc = pcie_set_mps(dev, mps);
2625 if (rc)
2626 pci_err(dev, "Failed attempting to set the MPS\n");
2627}
2628
2629static void pcie_write_mrrs(struct pci_dev *dev)
2630{
2631 int rc, mrrs;
2632
2633 /*
2634 * In the "safe" case, do not configure the MRRS. There appear to be
2635 * issues with setting MRRS to 0 on a number of devices.
2636 */
2637 if (pcie_bus_config != PCIE_BUS_PERFORMANCE)
2638 return;
2639
2640 /*
2641 * For max performance, the MRRS must be set to the largest supported
2642 * value. However, it cannot be configured larger than the MPS the
2643 * device or the bus can support. This should already be properly
2644 * configured by a prior call to pcie_write_mps().
2645 */
2646 mrrs = pcie_get_mps(dev);
2647
2648 /*
2649 * MRRS is a R/W register. Invalid values can be written, but a
2650 * subsequent read will verify if the value is acceptable or not.
2651 * If the MRRS value provided is not acceptable (e.g., too large),
2652 * shrink the value until it is acceptable to the HW.
2653 */
2654 while (mrrs != pcie_get_readrq(dev) && mrrs >= 128) {
2655 rc = pcie_set_readrq(dev, mrrs);
2656 if (!rc)
2657 break;
2658
2659 pci_warn(dev, "Failed attempting to set the MRRS\n");
2660 mrrs /= 2;
2661 }
2662
2663 if (mrrs < 128)
2664 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");
2665}
2666
2667static int pcie_bus_configure_set(struct pci_dev *dev, void *data)
2668{
2669 int mps, orig_mps;
2670
2671 if (!pci_is_pcie(dev))
2672 return 0;
2673
2674 if (pcie_bus_config == PCIE_BUS_TUNE_OFF ||
2675 pcie_bus_config == PCIE_BUS_DEFAULT)
2676 return 0;
2677
2678 mps = 128 << *(u8 *)data;
2679 orig_mps = pcie_get_mps(dev);
2680
2681 pcie_write_mps(dev, mps);
2682 pcie_write_mrrs(dev);
2683
2684 pci_info(dev, "Max Payload Size set to %4d/%4d (was %4d), Max Read Rq %4d\n",
2685 pcie_get_mps(dev), 128 << dev->pcie_mpss,
2686 orig_mps, pcie_get_readrq(dev));
2687
2688 return 0;
2689}
2690
2691/*
2692 * pcie_bus_configure_settings() requires that pci_walk_bus work in a top-down,
2693 * parents then children fashion. If this changes, then this code will not
2694 * work as designed.
2695 */
2696void pcie_bus_configure_settings(struct pci_bus *bus)
2697{
2698 u8 smpss = 0;
2699
2700 if (!bus->self)
2701 return;
2702
2703 if (!pci_is_pcie(bus->self))
2704 return;
2705
2706 /*
2707 * FIXME - Peer to peer DMA is possible, though the endpoint would need
2708 * to be aware of the MPS of the destination. To work around this,
2709 * simply force the MPS of the entire system to the smallest possible.
2710 */
2711 if (pcie_bus_config == PCIE_BUS_PEER2PEER)
2712 smpss = 0;
2713
2714 if (pcie_bus_config == PCIE_BUS_SAFE) {
2715 smpss = bus->self->pcie_mpss;
2716
2717 pcie_find_smpss(bus->self, &smpss);
2718 pci_walk_bus(bus, pcie_find_smpss, &smpss);
2719 }
2720
2721 pcie_bus_configure_set(bus->self, &smpss);
2722 pci_walk_bus(bus, pcie_bus_configure_set, &smpss);
2723}
2724EXPORT_SYMBOL_GPL(pcie_bus_configure_settings);
2725
2726/*
2727 * Called after each bus is probed, but before its children are examined. This
2728 * is marked as __weak because multiple architectures define it.
2729 */
2730void __weak pcibios_fixup_bus(struct pci_bus *bus)
2731{
2732 /* nothing to do, expected to be removed in the future */
2733}
2734
2735/**
2736 * pci_scan_child_bus_extend() - Scan devices below a bus
2737 * @bus: Bus to scan for devices
2738 * @available_buses: Total number of buses available (%0 does not try to
2739 * extend beyond the minimal)
2740 *
2741 * Scans devices below @bus including subordinate buses. Returns new
2742 * subordinate number including all the found devices. Passing
2743 * @available_buses causes the remaining bus space to be distributed
2744 * equally between hotplug-capable bridges to allow future extension of the
2745 * hierarchy.
2746 */
2747static unsigned int pci_scan_child_bus_extend(struct pci_bus *bus,
2748 unsigned int available_buses)
2749{
2750 unsigned int used_buses, normal_bridges = 0, hotplug_bridges = 0;
2751 unsigned int start = bus->busn_res.start;
2752 unsigned int devfn, fn, cmax, max = start;
2753 struct pci_dev *dev;
2754 int nr_devs;
2755
2756 dev_dbg(&bus->dev, "scanning bus\n");
2757
2758 /* Go find them, Rover! */
2759 for (devfn = 0; devfn < 256; devfn += 8) {
2760 nr_devs = pci_scan_slot(bus, devfn);
2761
2762 /*
2763 * The Jailhouse hypervisor may pass individual functions of a
2764 * multi-function device to a guest without passing function 0.
2765 * Look for them as well.
2766 */
2767 if (jailhouse_paravirt() && nr_devs == 0) {
2768 for (fn = 1; fn < 8; fn++) {
2769 dev = pci_scan_single_device(bus, devfn + fn);
2770 if (dev)
2771 dev->multifunction = 1;
2772 }
2773 }
2774 }
2775
2776 /* Reserve buses for SR-IOV capability */
2777 used_buses = pci_iov_bus_range(bus);
2778 max += used_buses;
2779
2780 /*
2781 * After performing arch-dependent fixup of the bus, look behind
2782 * all PCI-to-PCI bridges on this bus.
2783 */
2784 if (!bus->is_added) {
2785 dev_dbg(&bus->dev, "fixups for bus\n");
2786 pcibios_fixup_bus(bus);
2787 bus->is_added = 1;
2788 }
2789
2790 /*
2791 * Calculate how many hotplug bridges and normal bridges there
2792 * are on this bus. We will distribute the additional available
2793 * buses between hotplug bridges.
2794 */
2795 for_each_pci_bridge(dev, bus) {
2796 if (dev->is_hotplug_bridge)
2797 hotplug_bridges++;
2798 else
2799 normal_bridges++;
2800 }
2801
2802 /*
2803 * Scan bridges that are already configured. We don't touch them
2804 * unless they are misconfigured (which will be done in the second
2805 * scan below).
2806 */
2807 for_each_pci_bridge(dev, bus) {
2808 cmax = max;
2809 max = pci_scan_bridge_extend(bus, dev, max, 0, 0);
2810
2811 /*
2812 * Reserve one bus for each bridge now to avoid extending
2813 * hotplug bridges too much during the second scan below.
2814 */
2815 used_buses++;
2816 if (cmax - max > 1)
2817 used_buses += cmax - max - 1;
2818 }
2819
2820 /* Scan bridges that need to be reconfigured */
2821 for_each_pci_bridge(dev, bus) {
2822 unsigned int buses = 0;
2823
2824 if (!hotplug_bridges && normal_bridges == 1) {
2825
2826 /*
2827 * There is only one bridge on the bus (upstream
2828 * port) so it gets all available buses which it
2829 * can then distribute to the possible hotplug
2830 * bridges below.
2831 */
2832 buses = available_buses;
2833 } else if (dev->is_hotplug_bridge) {
2834
2835 /*
2836 * Distribute the extra buses between hotplug
2837 * bridges if any.
2838 */
2839 buses = available_buses / hotplug_bridges;
2840 buses = min(buses, available_buses - used_buses + 1);
2841 }
2842
2843 cmax = max;
2844 max = pci_scan_bridge_extend(bus, dev, cmax, buses, 1);
2845 /* One bus is already accounted so don't add it again */
2846 if (max - cmax > 1)
2847 used_buses += max - cmax - 1;
2848 }
2849
2850 /*
2851 * Make sure a hotplug bridge has at least the minimum requested
2852 * number of buses but allow it to grow up to the maximum available
2853 * bus number of there is room.
2854 */
2855 if (bus->self && bus->self->is_hotplug_bridge) {
2856 used_buses = max_t(unsigned int, available_buses,
2857 pci_hotplug_bus_size - 1);
2858 if (max - start < used_buses) {
2859 max = start + used_buses;
2860
2861 /* Do not allocate more buses than we have room left */
2862 if (max > bus->busn_res.end)
2863 max = bus->busn_res.end;
2864
2865 dev_dbg(&bus->dev, "%pR extended by %#02x\n",
2866 &bus->busn_res, max - start);
2867 }
2868 }
2869
2870 /*
2871 * We've scanned the bus and so we know all about what's on
2872 * the other side of any bridges that may be on this bus plus
2873 * any devices.
2874 *
2875 * Return how far we've got finding sub-buses.
2876 */
2877 dev_dbg(&bus->dev, "bus scan returning with max=%02x\n", max);
2878 return max;
2879}
2880
2881/**
2882 * pci_scan_child_bus() - Scan devices below a bus
2883 * @bus: Bus to scan for devices
2884 *
2885 * Scans devices below @bus including subordinate buses. Returns new
2886 * subordinate number including all the found devices.
2887 */
2888unsigned int pci_scan_child_bus(struct pci_bus *bus)
2889{
2890 return pci_scan_child_bus_extend(bus, 0);
2891}
2892EXPORT_SYMBOL_GPL(pci_scan_child_bus);
2893
2894/**
2895 * pcibios_root_bridge_prepare - Platform-specific host bridge setup
2896 * @bridge: Host bridge to set up
2897 *
2898 * Default empty implementation. Replace with an architecture-specific setup
2899 * routine, if necessary.
2900 */
2901int __weak pcibios_root_bridge_prepare(struct pci_host_bridge *bridge)
2902{
2903 return 0;
2904}
2905
2906void __weak pcibios_add_bus(struct pci_bus *bus)
2907{
2908}
2909
2910void __weak pcibios_remove_bus(struct pci_bus *bus)
2911{
2912}
2913
2914struct pci_bus *pci_create_root_bus(struct device *parent, int bus,
2915 struct pci_ops *ops, void *sysdata, struct list_head *resources)
2916{
2917 int error;
2918 struct pci_host_bridge *bridge;
2919
2920 bridge = pci_alloc_host_bridge(0);
2921 if (!bridge)
2922 return NULL;
2923
2924 bridge->dev.parent = parent;
2925
2926 list_splice_init(resources, &bridge->windows);
2927 bridge->sysdata = sysdata;
2928 bridge->busnr = bus;
2929 bridge->ops = ops;
2930
2931 error = pci_register_host_bridge(bridge);
2932 if (error < 0)
2933 goto err_out;
2934
2935 return bridge->bus;
2936
2937err_out:
2938 kfree(bridge);
2939 return NULL;
2940}
2941EXPORT_SYMBOL_GPL(pci_create_root_bus);
2942
2943int pci_host_probe(struct pci_host_bridge *bridge)
2944{
2945 struct pci_bus *bus, *child;
2946 int ret;
2947
2948 ret = pci_scan_root_bus_bridge(bridge);
2949 if (ret < 0) {
2950 dev_err(bridge->dev.parent, "Scanning root bridge failed");
2951 return ret;
2952 }
2953
2954 bus = bridge->bus;
2955
2956 /*
2957 * We insert PCI resources into the iomem_resource and
2958 * ioport_resource trees in either pci_bus_claim_resources()
2959 * or pci_bus_assign_resources().
2960 */
2961 if (pci_has_flag(PCI_PROBE_ONLY)) {
2962 pci_bus_claim_resources(bus);
2963 } else {
2964 pci_bus_size_bridges(bus);
2965 pci_bus_assign_resources(bus);
2966
2967 list_for_each_entry(child, &bus->children, node)
2968 pcie_bus_configure_settings(child);
2969 }
2970
2971 pci_bus_add_devices(bus);
2972 return 0;
2973}
2974EXPORT_SYMBOL_GPL(pci_host_probe);
2975
2976int pci_bus_insert_busn_res(struct pci_bus *b, int bus, int bus_max)
2977{
2978 struct resource *res = &b->busn_res;
2979 struct resource *parent_res, *conflict;
2980
2981 res->start = bus;
2982 res->end = bus_max;
2983 res->flags = IORESOURCE_BUS;
2984
2985 if (!pci_is_root_bus(b))
2986 parent_res = &b->parent->busn_res;
2987 else {
2988 parent_res = get_pci_domain_busn_res(pci_domain_nr(b));
2989 res->flags |= IORESOURCE_PCI_FIXED;
2990 }
2991
2992 conflict = request_resource_conflict(parent_res, res);
2993
2994 if (conflict)
2995 dev_printk(KERN_DEBUG, &b->dev,
2996 "busn_res: can not insert %pR under %s%pR (conflicts with %s %pR)\n",
2997 res, pci_is_root_bus(b) ? "domain " : "",
2998 parent_res, conflict->name, conflict);
2999
3000 return conflict == NULL;
3001}
3002
3003int pci_bus_update_busn_res_end(struct pci_bus *b, int bus_max)
3004{
3005 struct resource *res = &b->busn_res;
3006 struct resource old_res = *res;
3007 resource_size_t size;
3008 int ret;
3009
3010 if (res->start > bus_max)
3011 return -EINVAL;
3012
3013 size = bus_max - res->start + 1;
3014 ret = adjust_resource(res, res->start, size);
3015 dev_printk(KERN_DEBUG, &b->dev,
3016 "busn_res: %pR end %s updated to %02x\n",
3017 &old_res, ret ? "can not be" : "is", bus_max);
3018
3019 if (!ret && !res->parent)
3020 pci_bus_insert_busn_res(b, res->start, res->end);
3021
3022 return ret;
3023}
3024
3025void pci_bus_release_busn_res(struct pci_bus *b)
3026{
3027 struct resource *res = &b->busn_res;
3028 int ret;
3029
3030 if (!res->flags || !res->parent)
3031 return;
3032
3033 ret = release_resource(res);
3034 dev_printk(KERN_DEBUG, &b->dev,
3035 "busn_res: %pR %s released\n",
3036 res, ret ? "can not be" : "is");
3037}
3038
3039int pci_scan_root_bus_bridge(struct pci_host_bridge *bridge)
3040{
3041 struct resource_entry *window;
3042 bool found = false;
3043 struct pci_bus *b;
3044 int max, bus, ret;
3045
3046 if (!bridge)
3047 return -EINVAL;
3048
3049 resource_list_for_each_entry(window, &bridge->windows)
3050 if (window->res->flags & IORESOURCE_BUS) {
3051 found = true;
3052 break;
3053 }
3054
3055 ret = pci_register_host_bridge(bridge);
3056 if (ret < 0)
3057 return ret;
3058
3059 b = bridge->bus;
3060 bus = bridge->busnr;
3061
3062 if (!found) {
3063 dev_info(&b->dev,
3064 "No busn resource found for root bus, will use [bus %02x-ff]\n",
3065 bus);
3066 pci_bus_insert_busn_res(b, bus, 255);
3067 }
3068
3069 max = pci_scan_child_bus(b);
3070
3071 if (!found)
3072 pci_bus_update_busn_res_end(b, max);
3073
3074 return 0;
3075}
3076EXPORT_SYMBOL(pci_scan_root_bus_bridge);
3077
3078struct pci_bus *pci_scan_root_bus(struct device *parent, int bus,
3079 struct pci_ops *ops, void *sysdata, struct list_head *resources)
3080{
3081 struct resource_entry *window;
3082 bool found = false;
3083 struct pci_bus *b;
3084 int max;
3085
3086 resource_list_for_each_entry(window, resources)
3087 if (window->res->flags & IORESOURCE_BUS) {
3088 found = true;
3089 break;
3090 }
3091
3092 b = pci_create_root_bus(parent, bus, ops, sysdata, resources);
3093 if (!b)
3094 return NULL;
3095
3096 if (!found) {
3097 dev_info(&b->dev,
3098 "No busn resource found for root bus, will use [bus %02x-ff]\n",
3099 bus);
3100 pci_bus_insert_busn_res(b, bus, 255);
3101 }
3102
3103 max = pci_scan_child_bus(b);
3104
3105 if (!found)
3106 pci_bus_update_busn_res_end(b, max);
3107
3108 return b;
3109}
3110EXPORT_SYMBOL(pci_scan_root_bus);
3111
3112struct pci_bus *pci_scan_bus(int bus, struct pci_ops *ops,
3113 void *sysdata)
3114{
3115 LIST_HEAD(resources);
3116 struct pci_bus *b;
3117
3118 pci_add_resource(&resources, &ioport_resource);
3119 pci_add_resource(&resources, &iomem_resource);
3120 pci_add_resource(&resources, &busn_resource);
3121 b = pci_create_root_bus(NULL, bus, ops, sysdata, &resources);
3122 if (b) {
3123 pci_scan_child_bus(b);
3124 } else {
3125 pci_free_resource_list(&resources);
3126 }
3127 return b;
3128}
3129EXPORT_SYMBOL(pci_scan_bus);
3130
3131/**
3132 * pci_rescan_bus_bridge_resize - Scan a PCI bus for devices
3133 * @bridge: PCI bridge for the bus to scan
3134 *
3135 * Scan a PCI bus and child buses for new devices, add them,
3136 * and enable them, resizing bridge mmio/io resource if necessary
3137 * and possible. The caller must ensure the child devices are already
3138 * removed for resizing to occur.
3139 *
3140 * Returns the max number of subordinate bus discovered.
3141 */
3142unsigned int pci_rescan_bus_bridge_resize(struct pci_dev *bridge)
3143{
3144 unsigned int max;
3145 struct pci_bus *bus = bridge->subordinate;
3146
3147 max = pci_scan_child_bus(bus);
3148
3149 pci_assign_unassigned_bridge_resources(bridge);
3150
3151 pci_bus_add_devices(bus);
3152
3153 return max;
3154}
3155
3156/**
3157 * pci_rescan_bus - Scan a PCI bus for devices
3158 * @bus: PCI bus to scan
3159 *
3160 * Scan a PCI bus and child buses for new devices, add them,
3161 * and enable them.
3162 *
3163 * Returns the max number of subordinate bus discovered.
3164 */
3165unsigned int pci_rescan_bus(struct pci_bus *bus)
3166{
3167 unsigned int max;
3168
3169 max = pci_scan_child_bus(bus);
3170 pci_assign_unassigned_bus_resources(bus);
3171 pci_bus_add_devices(bus);
3172
3173 return max;
3174}
3175EXPORT_SYMBOL_GPL(pci_rescan_bus);
3176
3177/*
3178 * pci_rescan_bus(), pci_rescan_bus_bridge_resize() and PCI device removal
3179 * routines should always be executed under this mutex.
3180 */
3181static DEFINE_MUTEX(pci_rescan_remove_lock);
3182
3183void pci_lock_rescan_remove(void)
3184{
3185 mutex_lock(&pci_rescan_remove_lock);
3186}
3187EXPORT_SYMBOL_GPL(pci_lock_rescan_remove);
3188
3189void pci_unlock_rescan_remove(void)
3190{
3191 mutex_unlock(&pci_rescan_remove_lock);
3192}
3193EXPORT_SYMBOL_GPL(pci_unlock_rescan_remove);
3194
3195static int __init pci_sort_bf_cmp(const struct device *d_a,
3196 const struct device *d_b)
3197{
3198 const struct pci_dev *a = to_pci_dev(d_a);
3199 const struct pci_dev *b = to_pci_dev(d_b);
3200
3201 if (pci_domain_nr(a->bus) < pci_domain_nr(b->bus)) return -1;
3202 else if (pci_domain_nr(a->bus) > pci_domain_nr(b->bus)) return 1;
3203
3204 if (a->bus->number < b->bus->number) return -1;
3205 else if (a->bus->number > b->bus->number) return 1;
3206
3207 if (a->devfn < b->devfn) return -1;
3208 else if (a->devfn > b->devfn) return 1;
3209
3210 return 0;
3211}
3212
3213void __init pci_sort_breadthfirst(void)
3214{
3215 bus_sort_breadthfirst(&pci_bus_type, &pci_sort_bf_cmp);
3216}
3217
3218int pci_hp_add_bridge(struct pci_dev *dev)
3219{
3220 struct pci_bus *parent = dev->bus;
3221 int busnr, start = parent->busn_res.start;
3222 unsigned int available_buses = 0;
3223 int end = parent->busn_res.end;
3224
3225 for (busnr = start; busnr <= end; busnr++) {
3226 if (!pci_find_bus(pci_domain_nr(parent), busnr))
3227 break;
3228 }
3229 if (busnr-- > end) {
3230 pci_err(dev, "No bus number available for hot-added bridge\n");
3231 return -1;
3232 }
3233
3234 /* Scan bridges that are already configured */
3235 busnr = pci_scan_bridge(parent, dev, busnr, 0);
3236
3237 /*
3238 * Distribute the available bus numbers between hotplug-capable
3239 * bridges to make extending the chain later possible.
3240 */
3241 available_buses = end - busnr;
3242
3243 /* Scan bridges that need to be reconfigured */
3244 pci_scan_bridge_extend(parent, dev, busnr, available_buses, 1);
3245
3246 if (!dev->subordinate)
3247 return -1;
3248
3249 return 0;
3250}
3251EXPORT_SYMBOL_GPL(pci_hp_add_bridge);