Loading...
1// SPDX-License-Identifier: GPL-2.0
2/*
3 * mmconfig-shared.c - Low-level direct PCI config space access via
4 * MMCONFIG - common code between i386 and x86-64.
5 *
6 * This code does:
7 * - known chipset handling
8 * - ACPI decoding and validation
9 *
10 * Per-architecture code takes care of the mappings and accesses
11 * themselves.
12 */
13
14#include <linux/pci.h>
15#include <linux/init.h>
16#include <linux/sfi_acpi.h>
17#include <linux/bitmap.h>
18#include <linux/dmi.h>
19#include <linux/slab.h>
20#include <linux/mutex.h>
21#include <linux/rculist.h>
22#include <asm/e820/api.h>
23#include <asm/pci_x86.h>
24#include <asm/acpi.h>
25
26#define PREFIX "PCI: "
27
28/* Indicate if the mmcfg resources have been placed into the resource table. */
29static bool pci_mmcfg_running_state;
30static bool pci_mmcfg_arch_init_failed;
31static DEFINE_MUTEX(pci_mmcfg_lock);
32#define pci_mmcfg_lock_held() lock_is_held(&(pci_mmcfg_lock).dep_map)
33
34LIST_HEAD(pci_mmcfg_list);
35
36static void __init pci_mmconfig_remove(struct pci_mmcfg_region *cfg)
37{
38 if (cfg->res.parent)
39 release_resource(&cfg->res);
40 list_del(&cfg->list);
41 kfree(cfg);
42}
43
44static void __init free_all_mmcfg(void)
45{
46 struct pci_mmcfg_region *cfg, *tmp;
47
48 pci_mmcfg_arch_free();
49 list_for_each_entry_safe(cfg, tmp, &pci_mmcfg_list, list)
50 pci_mmconfig_remove(cfg);
51}
52
53static void list_add_sorted(struct pci_mmcfg_region *new)
54{
55 struct pci_mmcfg_region *cfg;
56
57 /* keep list sorted by segment and starting bus number */
58 list_for_each_entry_rcu(cfg, &pci_mmcfg_list, list, pci_mmcfg_lock_held()) {
59 if (cfg->segment > new->segment ||
60 (cfg->segment == new->segment &&
61 cfg->start_bus >= new->start_bus)) {
62 list_add_tail_rcu(&new->list, &cfg->list);
63 return;
64 }
65 }
66 list_add_tail_rcu(&new->list, &pci_mmcfg_list);
67}
68
69static struct pci_mmcfg_region *pci_mmconfig_alloc(int segment, int start,
70 int end, u64 addr)
71{
72 struct pci_mmcfg_region *new;
73 struct resource *res;
74
75 if (addr == 0)
76 return NULL;
77
78 new = kzalloc(sizeof(*new), GFP_KERNEL);
79 if (!new)
80 return NULL;
81
82 new->address = addr;
83 new->segment = segment;
84 new->start_bus = start;
85 new->end_bus = end;
86
87 res = &new->res;
88 res->start = addr + PCI_MMCFG_BUS_OFFSET(start);
89 res->end = addr + PCI_MMCFG_BUS_OFFSET(end + 1) - 1;
90 res->flags = IORESOURCE_MEM | IORESOURCE_BUSY;
91 snprintf(new->name, PCI_MMCFG_RESOURCE_NAME_LEN,
92 "PCI MMCONFIG %04x [bus %02x-%02x]", segment, start, end);
93 res->name = new->name;
94
95 return new;
96}
97
98struct pci_mmcfg_region *__init pci_mmconfig_add(int segment, int start,
99 int end, u64 addr)
100{
101 struct pci_mmcfg_region *new;
102
103 new = pci_mmconfig_alloc(segment, start, end, addr);
104 if (new) {
105 mutex_lock(&pci_mmcfg_lock);
106 list_add_sorted(new);
107 mutex_unlock(&pci_mmcfg_lock);
108
109 pr_info(PREFIX
110 "MMCONFIG for domain %04x [bus %02x-%02x] at %pR "
111 "(base %#lx)\n",
112 segment, start, end, &new->res, (unsigned long)addr);
113 }
114
115 return new;
116}
117
118struct pci_mmcfg_region *pci_mmconfig_lookup(int segment, int bus)
119{
120 struct pci_mmcfg_region *cfg;
121
122 list_for_each_entry_rcu(cfg, &pci_mmcfg_list, list, pci_mmcfg_lock_held())
123 if (cfg->segment == segment &&
124 cfg->start_bus <= bus && bus <= cfg->end_bus)
125 return cfg;
126
127 return NULL;
128}
129
130static const char *__init pci_mmcfg_e7520(void)
131{
132 u32 win;
133 raw_pci_ops->read(0, 0, PCI_DEVFN(0, 0), 0xce, 2, &win);
134
135 win = win & 0xf000;
136 if (win == 0x0000 || win == 0xf000)
137 return NULL;
138
139 if (pci_mmconfig_add(0, 0, 255, win << 16) == NULL)
140 return NULL;
141
142 return "Intel Corporation E7520 Memory Controller Hub";
143}
144
145static const char *__init pci_mmcfg_intel_945(void)
146{
147 u32 pciexbar, mask = 0, len = 0;
148
149 raw_pci_ops->read(0, 0, PCI_DEVFN(0, 0), 0x48, 4, &pciexbar);
150
151 /* Enable bit */
152 if (!(pciexbar & 1))
153 return NULL;
154
155 /* Size bits */
156 switch ((pciexbar >> 1) & 3) {
157 case 0:
158 mask = 0xf0000000U;
159 len = 0x10000000U;
160 break;
161 case 1:
162 mask = 0xf8000000U;
163 len = 0x08000000U;
164 break;
165 case 2:
166 mask = 0xfc000000U;
167 len = 0x04000000U;
168 break;
169 default:
170 return NULL;
171 }
172
173 /* Errata #2, things break when not aligned on a 256Mb boundary */
174 /* Can only happen in 64M/128M mode */
175
176 if ((pciexbar & mask) & 0x0fffffffU)
177 return NULL;
178
179 /* Don't hit the APIC registers and their friends */
180 if ((pciexbar & mask) >= 0xf0000000U)
181 return NULL;
182
183 if (pci_mmconfig_add(0, 0, (len >> 20) - 1, pciexbar & mask) == NULL)
184 return NULL;
185
186 return "Intel Corporation 945G/GZ/P/PL Express Memory Controller Hub";
187}
188
189static const char *__init pci_mmcfg_amd_fam10h(void)
190{
191 u32 low, high, address;
192 u64 base, msr;
193 int i;
194 unsigned segnbits = 0, busnbits, end_bus;
195
196 if (!(pci_probe & PCI_CHECK_ENABLE_AMD_MMCONF))
197 return NULL;
198
199 address = MSR_FAM10H_MMIO_CONF_BASE;
200 if (rdmsr_safe(address, &low, &high))
201 return NULL;
202
203 msr = high;
204 msr <<= 32;
205 msr |= low;
206
207 /* mmconfig is not enable */
208 if (!(msr & FAM10H_MMIO_CONF_ENABLE))
209 return NULL;
210
211 base = msr & (FAM10H_MMIO_CONF_BASE_MASK<<FAM10H_MMIO_CONF_BASE_SHIFT);
212
213 busnbits = (msr >> FAM10H_MMIO_CONF_BUSRANGE_SHIFT) &
214 FAM10H_MMIO_CONF_BUSRANGE_MASK;
215
216 /*
217 * only handle bus 0 ?
218 * need to skip it
219 */
220 if (!busnbits)
221 return NULL;
222
223 if (busnbits > 8) {
224 segnbits = busnbits - 8;
225 busnbits = 8;
226 }
227
228 end_bus = (1 << busnbits) - 1;
229 for (i = 0; i < (1 << segnbits); i++)
230 if (pci_mmconfig_add(i, 0, end_bus,
231 base + (1<<28) * i) == NULL) {
232 free_all_mmcfg();
233 return NULL;
234 }
235
236 return "AMD Family 10h NB";
237}
238
239static bool __initdata mcp55_checked;
240static const char *__init pci_mmcfg_nvidia_mcp55(void)
241{
242 int bus;
243 int mcp55_mmconf_found = 0;
244
245 static const u32 extcfg_regnum __initconst = 0x90;
246 static const u32 extcfg_regsize __initconst = 4;
247 static const u32 extcfg_enable_mask __initconst = 1 << 31;
248 static const u32 extcfg_start_mask __initconst = 0xff << 16;
249 static const int extcfg_start_shift __initconst = 16;
250 static const u32 extcfg_size_mask __initconst = 0x3 << 28;
251 static const int extcfg_size_shift __initconst = 28;
252 static const int extcfg_sizebus[] __initconst = {
253 0x100, 0x80, 0x40, 0x20
254 };
255 static const u32 extcfg_base_mask[] __initconst = {
256 0x7ff8, 0x7ffc, 0x7ffe, 0x7fff
257 };
258 static const int extcfg_base_lshift __initconst = 25;
259
260 /*
261 * do check if amd fam10h already took over
262 */
263 if (!acpi_disabled || !list_empty(&pci_mmcfg_list) || mcp55_checked)
264 return NULL;
265
266 mcp55_checked = true;
267 for (bus = 0; bus < 256; bus++) {
268 u64 base;
269 u32 l, extcfg;
270 u16 vendor, device;
271 int start, size_index, end;
272
273 raw_pci_ops->read(0, bus, PCI_DEVFN(0, 0), 0, 4, &l);
274 vendor = l & 0xffff;
275 device = (l >> 16) & 0xffff;
276
277 if (PCI_VENDOR_ID_NVIDIA != vendor || 0x0369 != device)
278 continue;
279
280 raw_pci_ops->read(0, bus, PCI_DEVFN(0, 0), extcfg_regnum,
281 extcfg_regsize, &extcfg);
282
283 if (!(extcfg & extcfg_enable_mask))
284 continue;
285
286 size_index = (extcfg & extcfg_size_mask) >> extcfg_size_shift;
287 base = extcfg & extcfg_base_mask[size_index];
288 /* base could > 4G */
289 base <<= extcfg_base_lshift;
290 start = (extcfg & extcfg_start_mask) >> extcfg_start_shift;
291 end = start + extcfg_sizebus[size_index] - 1;
292 if (pci_mmconfig_add(0, start, end, base) == NULL)
293 continue;
294 mcp55_mmconf_found++;
295 }
296
297 if (!mcp55_mmconf_found)
298 return NULL;
299
300 return "nVidia MCP55";
301}
302
303struct pci_mmcfg_hostbridge_probe {
304 u32 bus;
305 u32 devfn;
306 u32 vendor;
307 u32 device;
308 const char *(*probe)(void);
309};
310
311static const struct pci_mmcfg_hostbridge_probe pci_mmcfg_probes[] __initconst = {
312 { 0, PCI_DEVFN(0, 0), PCI_VENDOR_ID_INTEL,
313 PCI_DEVICE_ID_INTEL_E7520_MCH, pci_mmcfg_e7520 },
314 { 0, PCI_DEVFN(0, 0), PCI_VENDOR_ID_INTEL,
315 PCI_DEVICE_ID_INTEL_82945G_HB, pci_mmcfg_intel_945 },
316 { 0, PCI_DEVFN(0x18, 0), PCI_VENDOR_ID_AMD,
317 0x1200, pci_mmcfg_amd_fam10h },
318 { 0xff, PCI_DEVFN(0, 0), PCI_VENDOR_ID_AMD,
319 0x1200, pci_mmcfg_amd_fam10h },
320 { 0, PCI_DEVFN(0, 0), PCI_VENDOR_ID_NVIDIA,
321 0x0369, pci_mmcfg_nvidia_mcp55 },
322};
323
324static void __init pci_mmcfg_check_end_bus_number(void)
325{
326 struct pci_mmcfg_region *cfg, *cfgx;
327
328 /* Fixup overlaps */
329 list_for_each_entry(cfg, &pci_mmcfg_list, list) {
330 if (cfg->end_bus < cfg->start_bus)
331 cfg->end_bus = 255;
332
333 /* Don't access the list head ! */
334 if (cfg->list.next == &pci_mmcfg_list)
335 break;
336
337 cfgx = list_entry(cfg->list.next, typeof(*cfg), list);
338 if (cfg->end_bus >= cfgx->start_bus)
339 cfg->end_bus = cfgx->start_bus - 1;
340 }
341}
342
343static int __init pci_mmcfg_check_hostbridge(void)
344{
345 u32 l;
346 u32 bus, devfn;
347 u16 vendor, device;
348 int i;
349 const char *name;
350
351 if (!raw_pci_ops)
352 return 0;
353
354 free_all_mmcfg();
355
356 for (i = 0; i < ARRAY_SIZE(pci_mmcfg_probes); i++) {
357 bus = pci_mmcfg_probes[i].bus;
358 devfn = pci_mmcfg_probes[i].devfn;
359 raw_pci_ops->read(0, bus, devfn, 0, 4, &l);
360 vendor = l & 0xffff;
361 device = (l >> 16) & 0xffff;
362
363 name = NULL;
364 if (pci_mmcfg_probes[i].vendor == vendor &&
365 pci_mmcfg_probes[i].device == device)
366 name = pci_mmcfg_probes[i].probe();
367
368 if (name)
369 pr_info(PREFIX "%s with MMCONFIG support\n", name);
370 }
371
372 /* some end_bus_number is crazy, fix it */
373 pci_mmcfg_check_end_bus_number();
374
375 return !list_empty(&pci_mmcfg_list);
376}
377
378static acpi_status check_mcfg_resource(struct acpi_resource *res, void *data)
379{
380 struct resource *mcfg_res = data;
381 struct acpi_resource_address64 address;
382 acpi_status status;
383
384 if (res->type == ACPI_RESOURCE_TYPE_FIXED_MEMORY32) {
385 struct acpi_resource_fixed_memory32 *fixmem32 =
386 &res->data.fixed_memory32;
387 if (!fixmem32)
388 return AE_OK;
389 if ((mcfg_res->start >= fixmem32->address) &&
390 (mcfg_res->end < (fixmem32->address +
391 fixmem32->address_length))) {
392 mcfg_res->flags = 1;
393 return AE_CTRL_TERMINATE;
394 }
395 }
396 if ((res->type != ACPI_RESOURCE_TYPE_ADDRESS32) &&
397 (res->type != ACPI_RESOURCE_TYPE_ADDRESS64))
398 return AE_OK;
399
400 status = acpi_resource_to_address64(res, &address);
401 if (ACPI_FAILURE(status) ||
402 (address.address.address_length <= 0) ||
403 (address.resource_type != ACPI_MEMORY_RANGE))
404 return AE_OK;
405
406 if ((mcfg_res->start >= address.address.minimum) &&
407 (mcfg_res->end < (address.address.minimum + address.address.address_length))) {
408 mcfg_res->flags = 1;
409 return AE_CTRL_TERMINATE;
410 }
411 return AE_OK;
412}
413
414static acpi_status find_mboard_resource(acpi_handle handle, u32 lvl,
415 void *context, void **rv)
416{
417 struct resource *mcfg_res = context;
418
419 acpi_walk_resources(handle, METHOD_NAME__CRS,
420 check_mcfg_resource, context);
421
422 if (mcfg_res->flags)
423 return AE_CTRL_TERMINATE;
424
425 return AE_OK;
426}
427
428static bool is_acpi_reserved(u64 start, u64 end, unsigned not_used)
429{
430 struct resource mcfg_res;
431
432 mcfg_res.start = start;
433 mcfg_res.end = end - 1;
434 mcfg_res.flags = 0;
435
436 acpi_get_devices("PNP0C01", find_mboard_resource, &mcfg_res, NULL);
437
438 if (!mcfg_res.flags)
439 acpi_get_devices("PNP0C02", find_mboard_resource, &mcfg_res,
440 NULL);
441
442 return mcfg_res.flags;
443}
444
445typedef bool (*check_reserved_t)(u64 start, u64 end, unsigned type);
446
447static bool __ref is_mmconf_reserved(check_reserved_t is_reserved,
448 struct pci_mmcfg_region *cfg,
449 struct device *dev, int with_e820)
450{
451 u64 addr = cfg->res.start;
452 u64 size = resource_size(&cfg->res);
453 u64 old_size = size;
454 int num_buses;
455 char *method = with_e820 ? "E820" : "ACPI motherboard resources";
456
457 while (!is_reserved(addr, addr + size, E820_TYPE_RESERVED)) {
458 size >>= 1;
459 if (size < (16UL<<20))
460 break;
461 }
462
463 if (size < (16UL<<20) && size != old_size)
464 return 0;
465
466 if (dev)
467 dev_info(dev, "MMCONFIG at %pR reserved in %s\n",
468 &cfg->res, method);
469 else
470 pr_info(PREFIX "MMCONFIG at %pR reserved in %s\n",
471 &cfg->res, method);
472
473 if (old_size != size) {
474 /* update end_bus */
475 cfg->end_bus = cfg->start_bus + ((size>>20) - 1);
476 num_buses = cfg->end_bus - cfg->start_bus + 1;
477 cfg->res.end = cfg->res.start +
478 PCI_MMCFG_BUS_OFFSET(num_buses) - 1;
479 snprintf(cfg->name, PCI_MMCFG_RESOURCE_NAME_LEN,
480 "PCI MMCONFIG %04x [bus %02x-%02x]",
481 cfg->segment, cfg->start_bus, cfg->end_bus);
482
483 if (dev)
484 dev_info(dev,
485 "MMCONFIG "
486 "at %pR (base %#lx) (size reduced!)\n",
487 &cfg->res, (unsigned long) cfg->address);
488 else
489 pr_info(PREFIX
490 "MMCONFIG for %04x [bus%02x-%02x] "
491 "at %pR (base %#lx) (size reduced!)\n",
492 cfg->segment, cfg->start_bus, cfg->end_bus,
493 &cfg->res, (unsigned long) cfg->address);
494 }
495
496 return 1;
497}
498
499static bool __ref
500pci_mmcfg_check_reserved(struct device *dev, struct pci_mmcfg_region *cfg, int early)
501{
502 if (!early && !acpi_disabled) {
503 if (is_mmconf_reserved(is_acpi_reserved, cfg, dev, 0))
504 return 1;
505
506 if (dev)
507 dev_info(dev, FW_INFO
508 "MMCONFIG at %pR not reserved in "
509 "ACPI motherboard resources\n",
510 &cfg->res);
511 else
512 pr_info(FW_INFO PREFIX
513 "MMCONFIG at %pR not reserved in "
514 "ACPI motherboard resources\n",
515 &cfg->res);
516 }
517
518 /*
519 * e820__mapped_all() is marked as __init.
520 * All entries from ACPI MCFG table have been checked at boot time.
521 * For MCFG information constructed from hotpluggable host bridge's
522 * _CBA method, just assume it's reserved.
523 */
524 if (pci_mmcfg_running_state)
525 return 1;
526
527 /* Don't try to do this check unless configuration
528 type 1 is available. how about type 2 ?*/
529 if (raw_pci_ops)
530 return is_mmconf_reserved(e820__mapped_all, cfg, dev, 1);
531
532 return 0;
533}
534
535static void __init pci_mmcfg_reject_broken(int early)
536{
537 struct pci_mmcfg_region *cfg;
538
539 list_for_each_entry(cfg, &pci_mmcfg_list, list) {
540 if (pci_mmcfg_check_reserved(NULL, cfg, early) == 0) {
541 pr_info(PREFIX "not using MMCONFIG\n");
542 free_all_mmcfg();
543 return;
544 }
545 }
546}
547
548static int __init acpi_mcfg_check_entry(struct acpi_table_mcfg *mcfg,
549 struct acpi_mcfg_allocation *cfg)
550{
551 if (cfg->address < 0xFFFFFFFF)
552 return 0;
553
554 if (!strncmp(mcfg->header.oem_id, "SGI", 3))
555 return 0;
556
557 if ((mcfg->header.revision >= 1) && (dmi_get_bios_year() >= 2010))
558 return 0;
559
560 pr_err(PREFIX "MCFG region for %04x [bus %02x-%02x] at %#llx "
561 "is above 4GB, ignored\n", cfg->pci_segment,
562 cfg->start_bus_number, cfg->end_bus_number, cfg->address);
563 return -EINVAL;
564}
565
566static int __init pci_parse_mcfg(struct acpi_table_header *header)
567{
568 struct acpi_table_mcfg *mcfg;
569 struct acpi_mcfg_allocation *cfg_table, *cfg;
570 unsigned long i;
571 int entries;
572
573 if (!header)
574 return -EINVAL;
575
576 mcfg = (struct acpi_table_mcfg *)header;
577
578 /* how many config structures do we have */
579 free_all_mmcfg();
580 entries = 0;
581 i = header->length - sizeof(struct acpi_table_mcfg);
582 while (i >= sizeof(struct acpi_mcfg_allocation)) {
583 entries++;
584 i -= sizeof(struct acpi_mcfg_allocation);
585 }
586 if (entries == 0) {
587 pr_err(PREFIX "MMCONFIG has no entries\n");
588 return -ENODEV;
589 }
590
591 cfg_table = (struct acpi_mcfg_allocation *) &mcfg[1];
592 for (i = 0; i < entries; i++) {
593 cfg = &cfg_table[i];
594 if (acpi_mcfg_check_entry(mcfg, cfg)) {
595 free_all_mmcfg();
596 return -ENODEV;
597 }
598
599 if (pci_mmconfig_add(cfg->pci_segment, cfg->start_bus_number,
600 cfg->end_bus_number, cfg->address) == NULL) {
601 pr_warn(PREFIX "no memory for MCFG entries\n");
602 free_all_mmcfg();
603 return -ENOMEM;
604 }
605 }
606
607 return 0;
608}
609
610#ifdef CONFIG_ACPI_APEI
611extern int (*arch_apei_filter_addr)(int (*func)(__u64 start, __u64 size,
612 void *data), void *data);
613
614static int pci_mmcfg_for_each_region(int (*func)(__u64 start, __u64 size,
615 void *data), void *data)
616{
617 struct pci_mmcfg_region *cfg;
618 int rc;
619
620 if (list_empty(&pci_mmcfg_list))
621 return 0;
622
623 list_for_each_entry(cfg, &pci_mmcfg_list, list) {
624 rc = func(cfg->res.start, resource_size(&cfg->res), data);
625 if (rc)
626 return rc;
627 }
628
629 return 0;
630}
631#define set_apei_filter() (arch_apei_filter_addr = pci_mmcfg_for_each_region)
632#else
633#define set_apei_filter()
634#endif
635
636static void __init __pci_mmcfg_init(int early)
637{
638 pci_mmcfg_reject_broken(early);
639 if (list_empty(&pci_mmcfg_list))
640 return;
641
642 if (pcibios_last_bus < 0) {
643 const struct pci_mmcfg_region *cfg;
644
645 list_for_each_entry(cfg, &pci_mmcfg_list, list) {
646 if (cfg->segment)
647 break;
648 pcibios_last_bus = cfg->end_bus;
649 }
650 }
651
652 if (pci_mmcfg_arch_init())
653 pci_probe = (pci_probe & ~PCI_PROBE_MASK) | PCI_PROBE_MMCONF;
654 else {
655 free_all_mmcfg();
656 pci_mmcfg_arch_init_failed = true;
657 }
658}
659
660static int __initdata known_bridge;
661
662void __init pci_mmcfg_early_init(void)
663{
664 if (pci_probe & PCI_PROBE_MMCONF) {
665 if (pci_mmcfg_check_hostbridge())
666 known_bridge = 1;
667 else
668 acpi_sfi_table_parse(ACPI_SIG_MCFG, pci_parse_mcfg);
669 __pci_mmcfg_init(1);
670
671 set_apei_filter();
672 }
673}
674
675void __init pci_mmcfg_late_init(void)
676{
677 /* MMCONFIG disabled */
678 if ((pci_probe & PCI_PROBE_MMCONF) == 0)
679 return;
680
681 if (known_bridge)
682 return;
683
684 /* MMCONFIG hasn't been enabled yet, try again */
685 if (pci_probe & PCI_PROBE_MASK & ~PCI_PROBE_MMCONF) {
686 acpi_sfi_table_parse(ACPI_SIG_MCFG, pci_parse_mcfg);
687 __pci_mmcfg_init(0);
688 }
689}
690
691static int __init pci_mmcfg_late_insert_resources(void)
692{
693 struct pci_mmcfg_region *cfg;
694
695 pci_mmcfg_running_state = true;
696
697 /* If we are not using MMCONFIG, don't insert the resources. */
698 if ((pci_probe & PCI_PROBE_MMCONF) == 0)
699 return 1;
700
701 /*
702 * Attempt to insert the mmcfg resources but not with the busy flag
703 * marked so it won't cause request errors when __request_region is
704 * called.
705 */
706 list_for_each_entry(cfg, &pci_mmcfg_list, list)
707 if (!cfg->res.parent)
708 insert_resource(&iomem_resource, &cfg->res);
709
710 return 0;
711}
712
713/*
714 * Perform MMCONFIG resource insertion after PCI initialization to allow for
715 * misprogrammed MCFG tables that state larger sizes but actually conflict
716 * with other system resources.
717 */
718late_initcall(pci_mmcfg_late_insert_resources);
719
720/* Add MMCFG information for host bridges */
721int pci_mmconfig_insert(struct device *dev, u16 seg, u8 start, u8 end,
722 phys_addr_t addr)
723{
724 int rc;
725 struct resource *tmp = NULL;
726 struct pci_mmcfg_region *cfg;
727
728 if (!(pci_probe & PCI_PROBE_MMCONF) || pci_mmcfg_arch_init_failed)
729 return -ENODEV;
730
731 if (start > end)
732 return -EINVAL;
733
734 mutex_lock(&pci_mmcfg_lock);
735 cfg = pci_mmconfig_lookup(seg, start);
736 if (cfg) {
737 if (cfg->end_bus < end)
738 dev_info(dev, FW_INFO
739 "MMCONFIG for "
740 "domain %04x [bus %02x-%02x] "
741 "only partially covers this bridge\n",
742 cfg->segment, cfg->start_bus, cfg->end_bus);
743 mutex_unlock(&pci_mmcfg_lock);
744 return -EEXIST;
745 }
746
747 if (!addr) {
748 mutex_unlock(&pci_mmcfg_lock);
749 return -EINVAL;
750 }
751
752 rc = -EBUSY;
753 cfg = pci_mmconfig_alloc(seg, start, end, addr);
754 if (cfg == NULL) {
755 dev_warn(dev, "fail to add MMCONFIG (out of memory)\n");
756 rc = -ENOMEM;
757 } else if (!pci_mmcfg_check_reserved(dev, cfg, 0)) {
758 dev_warn(dev, FW_BUG "MMCONFIG %pR isn't reserved\n",
759 &cfg->res);
760 } else {
761 /* Insert resource if it's not in boot stage */
762 if (pci_mmcfg_running_state)
763 tmp = insert_resource_conflict(&iomem_resource,
764 &cfg->res);
765
766 if (tmp) {
767 dev_warn(dev,
768 "MMCONFIG %pR conflicts with "
769 "%s %pR\n",
770 &cfg->res, tmp->name, tmp);
771 } else if (pci_mmcfg_arch_map(cfg)) {
772 dev_warn(dev, "fail to map MMCONFIG %pR.\n",
773 &cfg->res);
774 } else {
775 list_add_sorted(cfg);
776 dev_info(dev, "MMCONFIG at %pR (base %#lx)\n",
777 &cfg->res, (unsigned long)addr);
778 cfg = NULL;
779 rc = 0;
780 }
781 }
782
783 if (cfg) {
784 if (cfg->res.parent)
785 release_resource(&cfg->res);
786 kfree(cfg);
787 }
788
789 mutex_unlock(&pci_mmcfg_lock);
790
791 return rc;
792}
793
794/* Delete MMCFG information for host bridges */
795int pci_mmconfig_delete(u16 seg, u8 start, u8 end)
796{
797 struct pci_mmcfg_region *cfg;
798
799 mutex_lock(&pci_mmcfg_lock);
800 list_for_each_entry_rcu(cfg, &pci_mmcfg_list, list)
801 if (cfg->segment == seg && cfg->start_bus == start &&
802 cfg->end_bus == end) {
803 list_del_rcu(&cfg->list);
804 synchronize_rcu();
805 pci_mmcfg_arch_unmap(cfg);
806 if (cfg->res.parent)
807 release_resource(&cfg->res);
808 mutex_unlock(&pci_mmcfg_lock);
809 kfree(cfg);
810 return 0;
811 }
812 mutex_unlock(&pci_mmcfg_lock);
813
814 return -ENOENT;
815}
1// SPDX-License-Identifier: GPL-2.0
2/*
3 * Low-level direct PCI config space access via ECAM - common code between
4 * i386 and x86-64.
5 *
6 * This code does:
7 * - known chipset handling
8 * - ACPI decoding and validation
9 *
10 * Per-architecture code takes care of the mappings and accesses
11 * themselves.
12 */
13
14#define pr_fmt(fmt) "PCI: " fmt
15
16#include <linux/acpi.h>
17#include <linux/efi.h>
18#include <linux/pci.h>
19#include <linux/init.h>
20#include <linux/bitmap.h>
21#include <linux/dmi.h>
22#include <linux/slab.h>
23#include <linux/mutex.h>
24#include <linux/rculist.h>
25#include <asm/e820/api.h>
26#include <asm/pci_x86.h>
27#include <asm/acpi.h>
28
29/* Indicate if the ECAM resources have been placed into the resource table */
30static bool pci_mmcfg_running_state;
31static bool pci_mmcfg_arch_init_failed;
32static DEFINE_MUTEX(pci_mmcfg_lock);
33#define pci_mmcfg_lock_held() lock_is_held(&(pci_mmcfg_lock).dep_map)
34
35LIST_HEAD(pci_mmcfg_list);
36
37static void __init pci_mmconfig_remove(struct pci_mmcfg_region *cfg)
38{
39 if (cfg->res.parent)
40 release_resource(&cfg->res);
41 list_del(&cfg->list);
42 kfree(cfg);
43}
44
45static void __init free_all_mmcfg(void)
46{
47 struct pci_mmcfg_region *cfg, *tmp;
48
49 pci_mmcfg_arch_free();
50 list_for_each_entry_safe(cfg, tmp, &pci_mmcfg_list, list)
51 pci_mmconfig_remove(cfg);
52}
53
54static void list_add_sorted(struct pci_mmcfg_region *new)
55{
56 struct pci_mmcfg_region *cfg;
57
58 /* keep list sorted by segment and starting bus number */
59 list_for_each_entry_rcu(cfg, &pci_mmcfg_list, list, pci_mmcfg_lock_held()) {
60 if (cfg->segment > new->segment ||
61 (cfg->segment == new->segment &&
62 cfg->start_bus >= new->start_bus)) {
63 list_add_tail_rcu(&new->list, &cfg->list);
64 return;
65 }
66 }
67 list_add_tail_rcu(&new->list, &pci_mmcfg_list);
68}
69
70static struct pci_mmcfg_region *pci_mmconfig_alloc(int segment, int start,
71 int end, u64 addr)
72{
73 struct pci_mmcfg_region *new;
74 struct resource *res;
75
76 if (addr == 0)
77 return NULL;
78
79 new = kzalloc(sizeof(*new), GFP_KERNEL);
80 if (!new)
81 return NULL;
82
83 new->address = addr;
84 new->segment = segment;
85 new->start_bus = start;
86 new->end_bus = end;
87
88 res = &new->res;
89 res->start = addr + PCI_MMCFG_BUS_OFFSET(start);
90 res->end = addr + PCI_MMCFG_BUS_OFFSET(end + 1) - 1;
91 res->flags = IORESOURCE_MEM | IORESOURCE_BUSY;
92 snprintf(new->name, PCI_MMCFG_RESOURCE_NAME_LEN,
93 "PCI ECAM %04x [bus %02x-%02x]", segment, start, end);
94 res->name = new->name;
95
96 return new;
97}
98
99struct pci_mmcfg_region *__init pci_mmconfig_add(int segment, int start,
100 int end, u64 addr)
101{
102 struct pci_mmcfg_region *new;
103
104 new = pci_mmconfig_alloc(segment, start, end, addr);
105 if (!new)
106 return NULL;
107
108 mutex_lock(&pci_mmcfg_lock);
109 list_add_sorted(new);
110 mutex_unlock(&pci_mmcfg_lock);
111
112 pr_info("ECAM %pR (base %#lx) for domain %04x [bus %02x-%02x]\n",
113 &new->res, (unsigned long)addr, segment, start, end);
114
115 return new;
116}
117
118struct pci_mmcfg_region *pci_mmconfig_lookup(int segment, int bus)
119{
120 struct pci_mmcfg_region *cfg;
121
122 list_for_each_entry_rcu(cfg, &pci_mmcfg_list, list, pci_mmcfg_lock_held())
123 if (cfg->segment == segment &&
124 cfg->start_bus <= bus && bus <= cfg->end_bus)
125 return cfg;
126
127 return NULL;
128}
129
130static const char *__init pci_mmcfg_e7520(void)
131{
132 u32 win;
133 raw_pci_ops->read(0, 0, PCI_DEVFN(0, 0), 0xce, 2, &win);
134
135 win = win & 0xf000;
136 if (win == 0x0000 || win == 0xf000)
137 return NULL;
138
139 if (pci_mmconfig_add(0, 0, 255, win << 16) == NULL)
140 return NULL;
141
142 return "Intel Corporation E7520 Memory Controller Hub";
143}
144
145static const char *__init pci_mmcfg_intel_945(void)
146{
147 u32 pciexbar, mask = 0, len = 0;
148
149 raw_pci_ops->read(0, 0, PCI_DEVFN(0, 0), 0x48, 4, &pciexbar);
150
151 /* Enable bit */
152 if (!(pciexbar & 1))
153 return NULL;
154
155 /* Size bits */
156 switch ((pciexbar >> 1) & 3) {
157 case 0:
158 mask = 0xf0000000U;
159 len = 0x10000000U;
160 break;
161 case 1:
162 mask = 0xf8000000U;
163 len = 0x08000000U;
164 break;
165 case 2:
166 mask = 0xfc000000U;
167 len = 0x04000000U;
168 break;
169 default:
170 return NULL;
171 }
172
173 /* Errata #2, things break when not aligned on a 256Mb boundary */
174 /* Can only happen in 64M/128M mode */
175
176 if ((pciexbar & mask) & 0x0fffffffU)
177 return NULL;
178
179 /* Don't hit the APIC registers and their friends */
180 if ((pciexbar & mask) >= 0xf0000000U)
181 return NULL;
182
183 if (pci_mmconfig_add(0, 0, (len >> 20) - 1, pciexbar & mask) == NULL)
184 return NULL;
185
186 return "Intel Corporation 945G/GZ/P/PL Express Memory Controller Hub";
187}
188
189static const char *__init pci_mmcfg_amd_fam10h(void)
190{
191 u32 low, high, address;
192 u64 base, msr;
193 int i;
194 unsigned segnbits = 0, busnbits, end_bus;
195
196 if (!(pci_probe & PCI_CHECK_ENABLE_AMD_MMCONF))
197 return NULL;
198
199 address = MSR_FAM10H_MMIO_CONF_BASE;
200 if (rdmsr_safe(address, &low, &high))
201 return NULL;
202
203 msr = high;
204 msr <<= 32;
205 msr |= low;
206
207 /* ECAM is not enabled */
208 if (!(msr & FAM10H_MMIO_CONF_ENABLE))
209 return NULL;
210
211 base = msr & (FAM10H_MMIO_CONF_BASE_MASK<<FAM10H_MMIO_CONF_BASE_SHIFT);
212
213 busnbits = (msr >> FAM10H_MMIO_CONF_BUSRANGE_SHIFT) &
214 FAM10H_MMIO_CONF_BUSRANGE_MASK;
215
216 /*
217 * only handle bus 0 ?
218 * need to skip it
219 */
220 if (!busnbits)
221 return NULL;
222
223 if (busnbits > 8) {
224 segnbits = busnbits - 8;
225 busnbits = 8;
226 }
227
228 end_bus = (1 << busnbits) - 1;
229 for (i = 0; i < (1 << segnbits); i++)
230 if (pci_mmconfig_add(i, 0, end_bus,
231 base + (1<<28) * i) == NULL) {
232 free_all_mmcfg();
233 return NULL;
234 }
235
236 return "AMD Family 10h NB";
237}
238
239static bool __initdata mcp55_checked;
240static const char *__init pci_mmcfg_nvidia_mcp55(void)
241{
242 int bus;
243 int mcp55_mmconf_found = 0;
244
245 static const u32 extcfg_regnum __initconst = 0x90;
246 static const u32 extcfg_regsize __initconst = 4;
247 static const u32 extcfg_enable_mask __initconst = 1 << 31;
248 static const u32 extcfg_start_mask __initconst = 0xff << 16;
249 static const int extcfg_start_shift __initconst = 16;
250 static const u32 extcfg_size_mask __initconst = 0x3 << 28;
251 static const int extcfg_size_shift __initconst = 28;
252 static const int extcfg_sizebus[] __initconst = {
253 0x100, 0x80, 0x40, 0x20
254 };
255 static const u32 extcfg_base_mask[] __initconst = {
256 0x7ff8, 0x7ffc, 0x7ffe, 0x7fff
257 };
258 static const int extcfg_base_lshift __initconst = 25;
259
260 /*
261 * do check if amd fam10h already took over
262 */
263 if (!acpi_disabled || !list_empty(&pci_mmcfg_list) || mcp55_checked)
264 return NULL;
265
266 mcp55_checked = true;
267 for (bus = 0; bus < 256; bus++) {
268 u64 base;
269 u32 l, extcfg;
270 u16 vendor, device;
271 int start, size_index, end;
272
273 raw_pci_ops->read(0, bus, PCI_DEVFN(0, 0), 0, 4, &l);
274 vendor = l & 0xffff;
275 device = (l >> 16) & 0xffff;
276
277 if (PCI_VENDOR_ID_NVIDIA != vendor || 0x0369 != device)
278 continue;
279
280 raw_pci_ops->read(0, bus, PCI_DEVFN(0, 0), extcfg_regnum,
281 extcfg_regsize, &extcfg);
282
283 if (!(extcfg & extcfg_enable_mask))
284 continue;
285
286 size_index = (extcfg & extcfg_size_mask) >> extcfg_size_shift;
287 base = extcfg & extcfg_base_mask[size_index];
288 /* base could > 4G */
289 base <<= extcfg_base_lshift;
290 start = (extcfg & extcfg_start_mask) >> extcfg_start_shift;
291 end = start + extcfg_sizebus[size_index] - 1;
292 if (pci_mmconfig_add(0, start, end, base) == NULL)
293 continue;
294 mcp55_mmconf_found++;
295 }
296
297 if (!mcp55_mmconf_found)
298 return NULL;
299
300 return "nVidia MCP55";
301}
302
303struct pci_mmcfg_hostbridge_probe {
304 u32 bus;
305 u32 devfn;
306 u32 vendor;
307 u32 device;
308 const char *(*probe)(void);
309};
310
311static const struct pci_mmcfg_hostbridge_probe pci_mmcfg_probes[] __initconst = {
312 { 0, PCI_DEVFN(0, 0), PCI_VENDOR_ID_INTEL,
313 PCI_DEVICE_ID_INTEL_E7520_MCH, pci_mmcfg_e7520 },
314 { 0, PCI_DEVFN(0, 0), PCI_VENDOR_ID_INTEL,
315 PCI_DEVICE_ID_INTEL_82945G_HB, pci_mmcfg_intel_945 },
316 { 0, PCI_DEVFN(0x18, 0), PCI_VENDOR_ID_AMD,
317 0x1200, pci_mmcfg_amd_fam10h },
318 { 0xff, PCI_DEVFN(0, 0), PCI_VENDOR_ID_AMD,
319 0x1200, pci_mmcfg_amd_fam10h },
320 { 0, PCI_DEVFN(0, 0), PCI_VENDOR_ID_NVIDIA,
321 0x0369, pci_mmcfg_nvidia_mcp55 },
322};
323
324static void __init pci_mmcfg_check_end_bus_number(void)
325{
326 struct pci_mmcfg_region *cfg, *cfgx;
327
328 /* Fixup overlaps */
329 list_for_each_entry(cfg, &pci_mmcfg_list, list) {
330 if (cfg->end_bus < cfg->start_bus)
331 cfg->end_bus = 255;
332
333 /* Don't access the list head ! */
334 if (cfg->list.next == &pci_mmcfg_list)
335 break;
336
337 cfgx = list_entry(cfg->list.next, typeof(*cfg), list);
338 if (cfg->end_bus >= cfgx->start_bus)
339 cfg->end_bus = cfgx->start_bus - 1;
340 }
341}
342
343static int __init pci_mmcfg_check_hostbridge(void)
344{
345 u32 l;
346 u32 bus, devfn;
347 u16 vendor, device;
348 int i;
349 const char *name;
350
351 if (!raw_pci_ops)
352 return 0;
353
354 free_all_mmcfg();
355
356 for (i = 0; i < ARRAY_SIZE(pci_mmcfg_probes); i++) {
357 bus = pci_mmcfg_probes[i].bus;
358 devfn = pci_mmcfg_probes[i].devfn;
359 raw_pci_ops->read(0, bus, devfn, 0, 4, &l);
360 vendor = l & 0xffff;
361 device = (l >> 16) & 0xffff;
362
363 name = NULL;
364 if (pci_mmcfg_probes[i].vendor == vendor &&
365 pci_mmcfg_probes[i].device == device)
366 name = pci_mmcfg_probes[i].probe();
367
368 if (name)
369 pr_info("%s with ECAM support\n", name);
370 }
371
372 /* some end_bus_number is crazy, fix it */
373 pci_mmcfg_check_end_bus_number();
374
375 return !list_empty(&pci_mmcfg_list);
376}
377
378static acpi_status check_mcfg_resource(struct acpi_resource *res, void *data)
379{
380 struct resource *mcfg_res = data;
381 struct acpi_resource_address64 address;
382 acpi_status status;
383
384 if (res->type == ACPI_RESOURCE_TYPE_FIXED_MEMORY32) {
385 struct acpi_resource_fixed_memory32 *fixmem32 =
386 &res->data.fixed_memory32;
387 if (!fixmem32)
388 return AE_OK;
389 if ((mcfg_res->start >= fixmem32->address) &&
390 (mcfg_res->end < (fixmem32->address +
391 fixmem32->address_length))) {
392 mcfg_res->flags = 1;
393 return AE_CTRL_TERMINATE;
394 }
395 }
396 if ((res->type != ACPI_RESOURCE_TYPE_ADDRESS32) &&
397 (res->type != ACPI_RESOURCE_TYPE_ADDRESS64))
398 return AE_OK;
399
400 status = acpi_resource_to_address64(res, &address);
401 if (ACPI_FAILURE(status) ||
402 (address.address.address_length <= 0) ||
403 (address.resource_type != ACPI_MEMORY_RANGE))
404 return AE_OK;
405
406 if ((mcfg_res->start >= address.address.minimum) &&
407 (mcfg_res->end < (address.address.minimum + address.address.address_length))) {
408 mcfg_res->flags = 1;
409 return AE_CTRL_TERMINATE;
410 }
411 return AE_OK;
412}
413
414static acpi_status find_mboard_resource(acpi_handle handle, u32 lvl,
415 void *context, void **rv)
416{
417 struct resource *mcfg_res = context;
418
419 acpi_walk_resources(handle, METHOD_NAME__CRS,
420 check_mcfg_resource, context);
421
422 if (mcfg_res->flags)
423 return AE_CTRL_TERMINATE;
424
425 return AE_OK;
426}
427
428static bool is_acpi_reserved(u64 start, u64 end, enum e820_type not_used)
429{
430 struct resource mcfg_res;
431
432 mcfg_res.start = start;
433 mcfg_res.end = end - 1;
434 mcfg_res.flags = 0;
435
436 acpi_get_devices("PNP0C01", find_mboard_resource, &mcfg_res, NULL);
437
438 if (!mcfg_res.flags)
439 acpi_get_devices("PNP0C02", find_mboard_resource, &mcfg_res,
440 NULL);
441
442 return mcfg_res.flags;
443}
444
445static bool is_efi_mmio(struct resource *res)
446{
447#ifdef CONFIG_EFI
448 u64 start = res->start;
449 u64 end = res->start + resource_size(res);
450 efi_memory_desc_t *md;
451 u64 size, mmio_start, mmio_end;
452
453 for_each_efi_memory_desc(md) {
454 if (md->type == EFI_MEMORY_MAPPED_IO) {
455 size = md->num_pages << EFI_PAGE_SHIFT;
456 mmio_start = md->phys_addr;
457 mmio_end = mmio_start + size;
458
459 if (mmio_start <= start && end <= mmio_end)
460 return true;
461 }
462 }
463#endif
464
465 return false;
466}
467
468typedef bool (*check_reserved_t)(u64 start, u64 end, enum e820_type type);
469
470static bool __ref is_mmconf_reserved(check_reserved_t is_reserved,
471 struct pci_mmcfg_region *cfg,
472 struct device *dev, const char *method)
473{
474 u64 addr = cfg->res.start;
475 u64 size = resource_size(&cfg->res);
476 u64 old_size = size;
477 int num_buses;
478
479 while (!is_reserved(addr, addr + size, E820_TYPE_RESERVED)) {
480 size >>= 1;
481 if (size < (16UL<<20))
482 break;
483 }
484
485 if (size < (16UL<<20) && size != old_size)
486 return false;
487
488 if (dev)
489 dev_info(dev, "ECAM %pR reserved as %s\n",
490 &cfg->res, method);
491 else
492 pr_info("ECAM %pR reserved as %s\n", &cfg->res, method);
493
494 if (old_size != size) {
495 /* update end_bus */
496 cfg->end_bus = cfg->start_bus + ((size>>20) - 1);
497 num_buses = cfg->end_bus - cfg->start_bus + 1;
498 cfg->res.end = cfg->res.start +
499 PCI_MMCFG_BUS_OFFSET(num_buses) - 1;
500 snprintf(cfg->name, PCI_MMCFG_RESOURCE_NAME_LEN,
501 "PCI ECAM %04x [bus %02x-%02x]",
502 cfg->segment, cfg->start_bus, cfg->end_bus);
503
504 if (dev)
505 dev_info(dev, "ECAM %pR (base %#lx) (size reduced!)\n",
506 &cfg->res, (unsigned long) cfg->address);
507 else
508 pr_info("ECAM %pR (base %#lx) for %04x [bus%02x-%02x] (size reduced!)\n",
509 &cfg->res, (unsigned long) cfg->address,
510 cfg->segment, cfg->start_bus, cfg->end_bus);
511 }
512
513 return true;
514}
515
516static bool __ref pci_mmcfg_reserved(struct device *dev,
517 struct pci_mmcfg_region *cfg, int early)
518{
519 struct resource *conflict;
520
521 if (early) {
522
523 /*
524 * Don't try to do this check unless configuration type 1
525 * is available. How about type 2?
526 */
527
528 /*
529 * 946f2ee5c731 ("Check that MCFG points to an e820
530 * reserved area") added this E820 check in 2006 to work
531 * around BIOS defects.
532 *
533 * Per PCI Firmware r3.3, sec 4.1.2, ECAM space must be
534 * reserved by a PNP0C02 resource, but it need not be
535 * mentioned in E820. Before the ACPI interpreter is
536 * available, we can't check for PNP0C02 resources, so
537 * there's no reliable way to verify the region in this
538 * early check. Keep it only for the old machines that
539 * motivated 946f2ee5c731.
540 */
541 if (dmi_get_bios_year() < 2016 && raw_pci_ops)
542 return is_mmconf_reserved(e820__mapped_all, cfg, dev,
543 "E820 entry");
544
545 return true;
546 }
547
548 if (!acpi_disabled) {
549 if (is_mmconf_reserved(is_acpi_reserved, cfg, dev,
550 "ACPI motherboard resource"))
551 return true;
552
553 if (dev)
554 dev_info(dev, FW_INFO "ECAM %pR not reserved in ACPI motherboard resources\n",
555 &cfg->res);
556 else
557 pr_info(FW_INFO "ECAM %pR not reserved in ACPI motherboard resources\n",
558 &cfg->res);
559
560 if (is_efi_mmio(&cfg->res)) {
561 pr_info("ECAM %pR is EfiMemoryMappedIO; assuming valid\n",
562 &cfg->res);
563 conflict = insert_resource_conflict(&iomem_resource,
564 &cfg->res);
565 if (conflict)
566 pr_warn("ECAM %pR conflicts with %s %pR\n",
567 &cfg->res, conflict->name, conflict);
568 else
569 pr_info("ECAM %pR reserved to work around lack of ACPI motherboard _CRS\n",
570 &cfg->res);
571 return true;
572 }
573 }
574
575 /*
576 * e820__mapped_all() is marked as __init.
577 * All entries from ACPI MCFG table have been checked at boot time.
578 * For MCFG information constructed from hotpluggable host bridge's
579 * _CBA method, just assume it's reserved.
580 */
581 return pci_mmcfg_running_state;
582}
583
584static void __init pci_mmcfg_reject_broken(int early)
585{
586 struct pci_mmcfg_region *cfg;
587
588 list_for_each_entry(cfg, &pci_mmcfg_list, list) {
589 if (!pci_mmcfg_reserved(NULL, cfg, early)) {
590 pr_info("not using ECAM (%pR not reserved)\n",
591 &cfg->res);
592 free_all_mmcfg();
593 return;
594 }
595 }
596}
597
598static bool __init acpi_mcfg_valid_entry(struct acpi_table_mcfg *mcfg,
599 struct acpi_mcfg_allocation *cfg)
600{
601 if (cfg->address < 0xFFFFFFFF)
602 return true;
603
604 if (!strncmp(mcfg->header.oem_id, "SGI", 3))
605 return true;
606
607 if ((mcfg->header.revision >= 1) && (dmi_get_bios_year() >= 2010))
608 return true;
609
610 pr_err("ECAM at %#llx for %04x [bus %02x-%02x] is above 4GB, ignored\n",
611 cfg->address, cfg->pci_segment, cfg->start_bus_number,
612 cfg->end_bus_number);
613 return false;
614}
615
616static int __init pci_parse_mcfg(struct acpi_table_header *header)
617{
618 struct acpi_table_mcfg *mcfg;
619 struct acpi_mcfg_allocation *cfg_table, *cfg;
620 unsigned long i;
621 int entries;
622
623 if (!header)
624 return -EINVAL;
625
626 mcfg = (struct acpi_table_mcfg *)header;
627
628 /* how many config structures do we have */
629 free_all_mmcfg();
630 entries = 0;
631 i = header->length - sizeof(struct acpi_table_mcfg);
632 while (i >= sizeof(struct acpi_mcfg_allocation)) {
633 entries++;
634 i -= sizeof(struct acpi_mcfg_allocation);
635 }
636 if (entries == 0) {
637 pr_err("MCFG has no entries\n");
638 return -ENODEV;
639 }
640
641 cfg_table = (struct acpi_mcfg_allocation *) &mcfg[1];
642 for (i = 0; i < entries; i++) {
643 cfg = &cfg_table[i];
644 if (!acpi_mcfg_valid_entry(mcfg, cfg)) {
645 free_all_mmcfg();
646 return -ENODEV;
647 }
648
649 if (pci_mmconfig_add(cfg->pci_segment, cfg->start_bus_number,
650 cfg->end_bus_number, cfg->address) == NULL) {
651 pr_warn("no memory for MCFG entries\n");
652 free_all_mmcfg();
653 return -ENOMEM;
654 }
655 }
656
657 return 0;
658}
659
660#ifdef CONFIG_ACPI_APEI
661extern int (*arch_apei_filter_addr)(int (*func)(__u64 start, __u64 size,
662 void *data), void *data);
663
664static int pci_mmcfg_for_each_region(int (*func)(__u64 start, __u64 size,
665 void *data), void *data)
666{
667 struct pci_mmcfg_region *cfg;
668 int rc;
669
670 if (list_empty(&pci_mmcfg_list))
671 return 0;
672
673 list_for_each_entry(cfg, &pci_mmcfg_list, list) {
674 rc = func(cfg->res.start, resource_size(&cfg->res), data);
675 if (rc)
676 return rc;
677 }
678
679 return 0;
680}
681#define set_apei_filter() (arch_apei_filter_addr = pci_mmcfg_for_each_region)
682#else
683#define set_apei_filter()
684#endif
685
686static void __init __pci_mmcfg_init(int early)
687{
688 pr_debug("%s(%s)\n", __func__, early ? "early" : "late");
689
690 pci_mmcfg_reject_broken(early);
691 if (list_empty(&pci_mmcfg_list))
692 return;
693
694 if (pcibios_last_bus < 0) {
695 const struct pci_mmcfg_region *cfg;
696
697 list_for_each_entry(cfg, &pci_mmcfg_list, list) {
698 if (cfg->segment)
699 break;
700 pcibios_last_bus = cfg->end_bus;
701 }
702 }
703
704 if (pci_mmcfg_arch_init())
705 pci_probe = (pci_probe & ~PCI_PROBE_MASK) | PCI_PROBE_MMCONF;
706 else {
707 free_all_mmcfg();
708 pci_mmcfg_arch_init_failed = true;
709 }
710}
711
712static int __initdata known_bridge;
713
714void __init pci_mmcfg_early_init(void)
715{
716 pr_debug("%s() pci_probe %#x\n", __func__, pci_probe);
717
718 if (pci_probe & PCI_PROBE_MMCONF) {
719 if (pci_mmcfg_check_hostbridge())
720 known_bridge = 1;
721 else
722 acpi_table_parse(ACPI_SIG_MCFG, pci_parse_mcfg);
723 __pci_mmcfg_init(1);
724
725 set_apei_filter();
726 }
727}
728
729void __init pci_mmcfg_late_init(void)
730{
731 pr_debug("%s() pci_probe %#x\n", __func__, pci_probe);
732
733 /* ECAM disabled */
734 if ((pci_probe & PCI_PROBE_MMCONF) == 0)
735 return;
736
737 if (known_bridge)
738 return;
739
740 /* ECAM hasn't been enabled yet, try again */
741 if (pci_probe & PCI_PROBE_MASK & ~PCI_PROBE_MMCONF) {
742 acpi_table_parse(ACPI_SIG_MCFG, pci_parse_mcfg);
743 __pci_mmcfg_init(0);
744 }
745}
746
747static int __init pci_mmcfg_late_insert_resources(void)
748{
749 struct pci_mmcfg_region *cfg;
750
751 pci_mmcfg_running_state = true;
752
753 pr_debug("%s() pci_probe %#x\n", __func__, pci_probe);
754
755 /* If we are not using ECAM, don't insert the resources. */
756 if ((pci_probe & PCI_PROBE_MMCONF) == 0)
757 return 1;
758
759 /*
760 * Attempt to insert the mmcfg resources but not with the busy flag
761 * marked so it won't cause request errors when __request_region is
762 * called.
763 */
764 list_for_each_entry(cfg, &pci_mmcfg_list, list) {
765 if (!cfg->res.parent) {
766 pr_debug("%s() insert %pR\n", __func__, &cfg->res);
767 insert_resource(&iomem_resource, &cfg->res);
768 }
769 }
770
771 return 0;
772}
773
774/*
775 * Perform ECAM resource insertion after PCI initialization to allow for
776 * misprogrammed MCFG tables that state larger sizes but actually conflict
777 * with other system resources.
778 */
779late_initcall(pci_mmcfg_late_insert_resources);
780
781/* Add ECAM information for host bridges */
782int pci_mmconfig_insert(struct device *dev, u16 seg, u8 start, u8 end,
783 phys_addr_t addr)
784{
785 int rc;
786 struct resource *tmp = NULL;
787 struct pci_mmcfg_region *cfg;
788
789 dev_dbg(dev, "%s(%04x [bus %02x-%02x])\n", __func__, seg, start, end);
790
791 if (!(pci_probe & PCI_PROBE_MMCONF) || pci_mmcfg_arch_init_failed)
792 return -ENODEV;
793
794 if (start > end)
795 return -EINVAL;
796
797 mutex_lock(&pci_mmcfg_lock);
798 cfg = pci_mmconfig_lookup(seg, start);
799 if (cfg) {
800 if (cfg->end_bus < end)
801 dev_info(dev, FW_INFO "ECAM %pR for domain %04x [bus %02x-%02x] only partially covers this bridge\n",
802 &cfg->res, cfg->segment, cfg->start_bus,
803 cfg->end_bus);
804 mutex_unlock(&pci_mmcfg_lock);
805 return -EEXIST;
806 }
807
808 /*
809 * Don't move earlier; we must return -EEXIST, not -EINVAL, if
810 * pci_mmconfig_lookup() finds something
811 */
812 if (!addr) {
813 mutex_unlock(&pci_mmcfg_lock);
814 return -EINVAL;
815 }
816
817 rc = -EBUSY;
818 cfg = pci_mmconfig_alloc(seg, start, end, addr);
819 if (cfg == NULL) {
820 dev_warn(dev, "fail to add ECAM (out of memory)\n");
821 rc = -ENOMEM;
822 } else if (!pci_mmcfg_reserved(dev, cfg, 0)) {
823 dev_warn(dev, FW_BUG "ECAM %pR isn't reserved\n",
824 &cfg->res);
825 } else {
826 /* Insert resource if it's not in boot stage */
827 if (pci_mmcfg_running_state)
828 tmp = insert_resource_conflict(&iomem_resource,
829 &cfg->res);
830
831 if (tmp) {
832 dev_warn(dev, "ECAM %pR conflicts with %s %pR\n",
833 &cfg->res, tmp->name, tmp);
834 } else if (pci_mmcfg_arch_map(cfg)) {
835 dev_warn(dev, "fail to map ECAM %pR\n", &cfg->res);
836 } else {
837 list_add_sorted(cfg);
838 dev_info(dev, "ECAM %pR (base %#lx)\n",
839 &cfg->res, (unsigned long)addr);
840 cfg = NULL;
841 rc = 0;
842 }
843 }
844
845 if (cfg) {
846 if (cfg->res.parent)
847 release_resource(&cfg->res);
848 kfree(cfg);
849 }
850
851 mutex_unlock(&pci_mmcfg_lock);
852
853 return rc;
854}
855
856/* Delete ECAM information for host bridges */
857int pci_mmconfig_delete(u16 seg, u8 start, u8 end)
858{
859 struct pci_mmcfg_region *cfg;
860
861 mutex_lock(&pci_mmcfg_lock);
862 list_for_each_entry_rcu(cfg, &pci_mmcfg_list, list)
863 if (cfg->segment == seg && cfg->start_bus == start &&
864 cfg->end_bus == end) {
865 list_del_rcu(&cfg->list);
866 synchronize_rcu();
867 pci_mmcfg_arch_unmap(cfg);
868 if (cfg->res.parent)
869 release_resource(&cfg->res);
870 mutex_unlock(&pci_mmcfg_lock);
871 kfree(cfg);
872 return 0;
873 }
874 mutex_unlock(&pci_mmcfg_lock);
875
876 return -ENOENT;
877}