xref: /linux/arch/x86/pci/mmconfig-shared.c (revision fcab107abe1ab5be9dbe874baa722372da8f4f73)
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/acpi.h>
26 #include <asm/e820/api.h>
27 #include <asm/msr.h>
28 #include <asm/pci_x86.h>
29 
30 /* Indicate if the ECAM resources have been placed into the resource table */
31 static bool pci_mmcfg_running_state;
32 static bool pci_mmcfg_arch_init_failed;
33 static DEFINE_MUTEX(pci_mmcfg_lock);
34 #define pci_mmcfg_lock_held() lock_is_held(&(pci_mmcfg_lock).dep_map)
35 
36 LIST_HEAD(pci_mmcfg_list);
37 
38 static void __init pci_mmconfig_remove(struct pci_mmcfg_region *cfg)
39 {
40 	if (cfg->res.parent)
41 		release_resource(&cfg->res);
42 	list_del(&cfg->list);
43 	kfree(cfg);
44 }
45 
46 static void __init free_all_mmcfg(void)
47 {
48 	struct pci_mmcfg_region *cfg, *tmp;
49 
50 	pci_mmcfg_arch_free();
51 	list_for_each_entry_safe(cfg, tmp, &pci_mmcfg_list, list)
52 		pci_mmconfig_remove(cfg);
53 }
54 
55 static void list_add_sorted(struct pci_mmcfg_region *new)
56 {
57 	struct pci_mmcfg_region *cfg;
58 
59 	/* keep list sorted by segment and starting bus number */
60 	list_for_each_entry_rcu(cfg, &pci_mmcfg_list, list, pci_mmcfg_lock_held()) {
61 		if (cfg->segment > new->segment ||
62 		    (cfg->segment == new->segment &&
63 		     cfg->start_bus >= new->start_bus)) {
64 			list_add_tail_rcu(&new->list, &cfg->list);
65 			return;
66 		}
67 	}
68 	list_add_tail_rcu(&new->list, &pci_mmcfg_list);
69 }
70 
71 static struct pci_mmcfg_region *pci_mmconfig_alloc(int segment, int start,
72 						   int end, u64 addr)
73 {
74 	struct pci_mmcfg_region *new;
75 	struct resource *res;
76 
77 	if (addr == 0)
78 		return NULL;
79 
80 	new = kzalloc(sizeof(*new), GFP_KERNEL);
81 	if (!new)
82 		return NULL;
83 
84 	new->address = addr;
85 	new->segment = segment;
86 	new->start_bus = start;
87 	new->end_bus = end;
88 
89 	res = &new->res;
90 	res->start = addr + PCI_MMCFG_BUS_OFFSET(start);
91 	res->end = addr + PCI_MMCFG_BUS_OFFSET(end + 1) - 1;
92 	res->flags = IORESOURCE_MEM | IORESOURCE_BUSY;
93 	snprintf(new->name, PCI_MMCFG_RESOURCE_NAME_LEN,
94 		 "PCI ECAM %04x [bus %02x-%02x]", segment, start, end);
95 	res->name = new->name;
96 
97 	return new;
98 }
99 
100 struct pci_mmcfg_region *__init pci_mmconfig_add(int segment, int start,
101 						 int end, u64 addr)
102 {
103 	struct pci_mmcfg_region *new;
104 
105 	new = pci_mmconfig_alloc(segment, start, end, addr);
106 	if (!new)
107 		return NULL;
108 
109 	mutex_lock(&pci_mmcfg_lock);
110 	list_add_sorted(new);
111 	mutex_unlock(&pci_mmcfg_lock);
112 
113 	pr_info("ECAM %pR (base %#lx) for domain %04x [bus %02x-%02x]\n",
114 		&new->res, (unsigned long)addr, segment, start, end);
115 
116 	return new;
117 }
118 
119 struct pci_mmcfg_region *pci_mmconfig_lookup(int segment, int bus)
120 {
121 	struct pci_mmcfg_region *cfg;
122 
123 	list_for_each_entry_rcu(cfg, &pci_mmcfg_list, list, pci_mmcfg_lock_held())
124 		if (cfg->segment == segment &&
125 		    cfg->start_bus <= bus && bus <= cfg->end_bus)
126 			return cfg;
127 
128 	return NULL;
129 }
130 
131 static const char *__init pci_mmcfg_e7520(void)
132 {
133 	u32 win;
134 	raw_pci_ops->read(0, 0, PCI_DEVFN(0, 0), 0xce, 2, &win);
135 
136 	win = win & 0xf000;
137 	if (win == 0x0000 || win == 0xf000)
138 		return NULL;
139 
140 	if (pci_mmconfig_add(0, 0, 255, win << 16) == NULL)
141 		return NULL;
142 
143 	return "Intel Corporation E7520 Memory Controller Hub";
144 }
145 
146 static const char *__init pci_mmcfg_intel_945(void)
147 {
148 	u32 pciexbar, mask = 0, len = 0;
149 
150 	raw_pci_ops->read(0, 0, PCI_DEVFN(0, 0), 0x48, 4, &pciexbar);
151 
152 	/* Enable bit */
153 	if (!(pciexbar & 1))
154 		return NULL;
155 
156 	/* Size bits */
157 	switch ((pciexbar >> 1) & 3) {
158 	case 0:
159 		mask = 0xf0000000U;
160 		len  = 0x10000000U;
161 		break;
162 	case 1:
163 		mask = 0xf8000000U;
164 		len  = 0x08000000U;
165 		break;
166 	case 2:
167 		mask = 0xfc000000U;
168 		len  = 0x04000000U;
169 		break;
170 	default:
171 		return NULL;
172 	}
173 
174 	/* Errata #2, things break when not aligned on a 256Mb boundary */
175 	/* Can only happen in 64M/128M mode */
176 
177 	if ((pciexbar & mask) & 0x0fffffffU)
178 		return NULL;
179 
180 	/* Don't hit the APIC registers and their friends */
181 	if ((pciexbar & mask) >= 0xf0000000U)
182 		return NULL;
183 
184 	if (pci_mmconfig_add(0, 0, (len >> 20) - 1, pciexbar & mask) == NULL)
185 		return NULL;
186 
187 	return "Intel Corporation 945G/GZ/P/PL Express Memory Controller Hub";
188 }
189 
190 static const char *__init pci_mmcfg_amd_fam10h(void)
191 {
192 	u32 low, high, address;
193 	u64 base, msr;
194 	int i;
195 	unsigned segnbits = 0, busnbits, end_bus;
196 
197 	if (!(pci_probe & PCI_CHECK_ENABLE_AMD_MMCONF))
198 		return NULL;
199 
200 	address = MSR_FAM10H_MMIO_CONF_BASE;
201 	if (rdmsr_safe(address, &low, &high))
202 		return NULL;
203 
204 	msr = high;
205 	msr <<= 32;
206 	msr |= low;
207 
208 	/* ECAM is not enabled */
209 	if (!(msr & FAM10H_MMIO_CONF_ENABLE))
210 		return NULL;
211 
212 	base = msr & (FAM10H_MMIO_CONF_BASE_MASK<<FAM10H_MMIO_CONF_BASE_SHIFT);
213 
214 	busnbits = (msr >> FAM10H_MMIO_CONF_BUSRANGE_SHIFT) &
215 			 FAM10H_MMIO_CONF_BUSRANGE_MASK;
216 
217 	/*
218 	 * only handle bus 0 ?
219 	 * need to skip it
220 	 */
221 	if (!busnbits)
222 		return NULL;
223 
224 	if (busnbits > 8) {
225 		segnbits = busnbits - 8;
226 		busnbits = 8;
227 	}
228 
229 	end_bus = (1 << busnbits) - 1;
230 	for (i = 0; i < (1 << segnbits); i++)
231 		if (pci_mmconfig_add(i, 0, end_bus,
232 				     base + (1<<28) * i) == NULL) {
233 			free_all_mmcfg();
234 			return NULL;
235 		}
236 
237 	return "AMD Family 10h NB";
238 }
239 
240 static bool __initdata mcp55_checked;
241 static const char *__init pci_mmcfg_nvidia_mcp55(void)
242 {
243 	int bus;
244 	int mcp55_mmconf_found = 0;
245 
246 	static const u32 extcfg_regnum __initconst	= 0x90;
247 	static const u32 extcfg_regsize __initconst	= 4;
248 	static const u32 extcfg_enable_mask __initconst	= 1 << 31;
249 	static const u32 extcfg_start_mask __initconst	= 0xff << 16;
250 	static const int extcfg_start_shift __initconst	= 16;
251 	static const u32 extcfg_size_mask __initconst	= 0x3 << 28;
252 	static const int extcfg_size_shift __initconst	= 28;
253 	static const int extcfg_sizebus[] __initconst	= {
254 		0x100, 0x80, 0x40, 0x20
255 	};
256 	static const u32 extcfg_base_mask[] __initconst	= {
257 		0x7ff8, 0x7ffc, 0x7ffe, 0x7fff
258 	};
259 	static const int extcfg_base_lshift __initconst	= 25;
260 
261 	/*
262 	 * do check if amd fam10h already took over
263 	 */
264 	if (!acpi_disabled || !list_empty(&pci_mmcfg_list) || mcp55_checked)
265 		return NULL;
266 
267 	mcp55_checked = true;
268 	for (bus = 0; bus < 256; bus++) {
269 		u64 base;
270 		u32 l, extcfg;
271 		u16 vendor, device;
272 		int start, size_index, end;
273 
274 		raw_pci_ops->read(0, bus, PCI_DEVFN(0, 0), 0, 4, &l);
275 		vendor = l & 0xffff;
276 		device = (l >> 16) & 0xffff;
277 
278 		if (PCI_VENDOR_ID_NVIDIA != vendor || 0x0369 != device)
279 			continue;
280 
281 		raw_pci_ops->read(0, bus, PCI_DEVFN(0, 0), extcfg_regnum,
282 				  extcfg_regsize, &extcfg);
283 
284 		if (!(extcfg & extcfg_enable_mask))
285 			continue;
286 
287 		size_index = (extcfg & extcfg_size_mask) >> extcfg_size_shift;
288 		base = extcfg & extcfg_base_mask[size_index];
289 		/* base could > 4G */
290 		base <<= extcfg_base_lshift;
291 		start = (extcfg & extcfg_start_mask) >> extcfg_start_shift;
292 		end = start + extcfg_sizebus[size_index] - 1;
293 		if (pci_mmconfig_add(0, start, end, base) == NULL)
294 			continue;
295 		mcp55_mmconf_found++;
296 	}
297 
298 	if (!mcp55_mmconf_found)
299 		return NULL;
300 
301 	return "nVidia MCP55";
302 }
303 
304 struct pci_mmcfg_hostbridge_probe {
305 	u32 bus;
306 	u32 devfn;
307 	u32 vendor;
308 	u32 device;
309 	const char *(*probe)(void);
310 };
311 
312 static const struct pci_mmcfg_hostbridge_probe pci_mmcfg_probes[] __initconst = {
313 	{ 0, PCI_DEVFN(0, 0), PCI_VENDOR_ID_INTEL,
314 	  PCI_DEVICE_ID_INTEL_E7520_MCH, pci_mmcfg_e7520 },
315 	{ 0, PCI_DEVFN(0, 0), PCI_VENDOR_ID_INTEL,
316 	  PCI_DEVICE_ID_INTEL_82945G_HB, pci_mmcfg_intel_945 },
317 	{ 0, PCI_DEVFN(0x18, 0), PCI_VENDOR_ID_AMD,
318 	  0x1200, pci_mmcfg_amd_fam10h },
319 	{ 0xff, PCI_DEVFN(0, 0), PCI_VENDOR_ID_AMD,
320 	  0x1200, pci_mmcfg_amd_fam10h },
321 	{ 0, PCI_DEVFN(0, 0), PCI_VENDOR_ID_NVIDIA,
322 	  0x0369, pci_mmcfg_nvidia_mcp55 },
323 };
324 
325 static void __init pci_mmcfg_check_end_bus_number(void)
326 {
327 	struct pci_mmcfg_region *cfg, *cfgx;
328 
329 	/* Fixup overlaps */
330 	list_for_each_entry(cfg, &pci_mmcfg_list, list) {
331 		if (cfg->end_bus < cfg->start_bus)
332 			cfg->end_bus = 255;
333 
334 		/* Don't access the list head ! */
335 		if (cfg->list.next == &pci_mmcfg_list)
336 			break;
337 
338 		cfgx = list_entry(cfg->list.next, typeof(*cfg), list);
339 		if (cfg->end_bus >= cfgx->start_bus)
340 			cfg->end_bus = cfgx->start_bus - 1;
341 	}
342 }
343 
344 static int __init pci_mmcfg_check_hostbridge(void)
345 {
346 	u32 l;
347 	u32 bus, devfn;
348 	u16 vendor, device;
349 	int i;
350 	const char *name;
351 
352 	if (!raw_pci_ops)
353 		return 0;
354 
355 	free_all_mmcfg();
356 
357 	for (i = 0; i < ARRAY_SIZE(pci_mmcfg_probes); i++) {
358 		bus =  pci_mmcfg_probes[i].bus;
359 		devfn = pci_mmcfg_probes[i].devfn;
360 		raw_pci_ops->read(0, bus, devfn, 0, 4, &l);
361 		vendor = l & 0xffff;
362 		device = (l >> 16) & 0xffff;
363 
364 		name = NULL;
365 		if (pci_mmcfg_probes[i].vendor == vendor &&
366 		    pci_mmcfg_probes[i].device == device)
367 			name = pci_mmcfg_probes[i].probe();
368 
369 		if (name)
370 			pr_info("%s with ECAM support\n", name);
371 	}
372 
373 	/* some end_bus_number is crazy, fix it */
374 	pci_mmcfg_check_end_bus_number();
375 
376 	return !list_empty(&pci_mmcfg_list);
377 }
378 
379 static acpi_status check_mcfg_resource(struct acpi_resource *res, void *data)
380 {
381 	struct resource *mcfg_res = data;
382 	struct acpi_resource_address64 address;
383 	acpi_status status;
384 
385 	if (res->type == ACPI_RESOURCE_TYPE_FIXED_MEMORY32) {
386 		struct acpi_resource_fixed_memory32 *fixmem32 =
387 			&res->data.fixed_memory32;
388 		if (!fixmem32)
389 			return AE_OK;
390 		if ((mcfg_res->start >= fixmem32->address) &&
391 		    (mcfg_res->end < (fixmem32->address +
392 				      fixmem32->address_length))) {
393 			mcfg_res->flags = 1;
394 			return AE_CTRL_TERMINATE;
395 		}
396 	}
397 	if ((res->type != ACPI_RESOURCE_TYPE_ADDRESS32) &&
398 	    (res->type != ACPI_RESOURCE_TYPE_ADDRESS64))
399 		return AE_OK;
400 
401 	status = acpi_resource_to_address64(res, &address);
402 	if (ACPI_FAILURE(status) ||
403 	   (address.address.address_length <= 0) ||
404 	   (address.resource_type != ACPI_MEMORY_RANGE))
405 		return AE_OK;
406 
407 	if ((mcfg_res->start >= address.address.minimum) &&
408 	    (mcfg_res->end < (address.address.minimum + address.address.address_length))) {
409 		mcfg_res->flags = 1;
410 		return AE_CTRL_TERMINATE;
411 	}
412 	return AE_OK;
413 }
414 
415 static acpi_status find_mboard_resource(acpi_handle handle, u32 lvl,
416 					void *context, void **rv)
417 {
418 	struct resource *mcfg_res = context;
419 
420 	acpi_walk_resources(handle, METHOD_NAME__CRS,
421 			    check_mcfg_resource, context);
422 
423 	if (mcfg_res->flags)
424 		return AE_CTRL_TERMINATE;
425 
426 	return AE_OK;
427 }
428 
429 static bool is_acpi_reserved(u64 start, u64 end, enum e820_type not_used)
430 {
431 	struct resource mcfg_res;
432 
433 	mcfg_res.start = start;
434 	mcfg_res.end = end - 1;
435 	mcfg_res.flags = 0;
436 
437 	acpi_get_devices("PNP0C01", find_mboard_resource, &mcfg_res, NULL);
438 
439 	if (!mcfg_res.flags)
440 		acpi_get_devices("PNP0C02", find_mboard_resource, &mcfg_res,
441 				 NULL);
442 
443 	return mcfg_res.flags;
444 }
445 
446 static bool is_efi_mmio(struct resource *res)
447 {
448 #ifdef CONFIG_EFI
449 	u64 start = res->start;
450 	u64 end = res->start + resource_size(res);
451 	efi_memory_desc_t *md;
452 	u64 size, mmio_start, mmio_end;
453 
454 	for_each_efi_memory_desc(md) {
455 		if (md->type == EFI_MEMORY_MAPPED_IO) {
456 			size = md->num_pages << EFI_PAGE_SHIFT;
457 			mmio_start = md->phys_addr;
458 			mmio_end = mmio_start + size;
459 
460 			if (mmio_start <= start && end <= mmio_end)
461 				return true;
462 		}
463 	}
464 #endif
465 
466 	return false;
467 }
468 
469 typedef bool (*check_reserved_t)(u64 start, u64 end, enum e820_type type);
470 
471 static bool __ref is_mmconf_reserved(check_reserved_t is_reserved,
472 				     struct pci_mmcfg_region *cfg,
473 				     struct device *dev, const char *method)
474 {
475 	u64 addr = cfg->res.start;
476 	u64 size = resource_size(&cfg->res);
477 	u64 old_size = size;
478 	int num_buses;
479 
480 	while (!is_reserved(addr, addr + size, E820_TYPE_RESERVED)) {
481 		size >>= 1;
482 		if (size < (16UL<<20))
483 			break;
484 	}
485 
486 	if (size < (16UL<<20) && size != old_size)
487 		return false;
488 
489 	if (dev)
490 		dev_info(dev, "ECAM %pR reserved as %s\n",
491 			 &cfg->res, method);
492 	else
493 		pr_info("ECAM %pR reserved as %s\n", &cfg->res, method);
494 
495 	if (old_size != size) {
496 		/* update end_bus */
497 		cfg->end_bus = cfg->start_bus + ((size>>20) - 1);
498 		num_buses = cfg->end_bus - cfg->start_bus + 1;
499 		cfg->res.end = cfg->res.start +
500 		    PCI_MMCFG_BUS_OFFSET(num_buses) - 1;
501 		snprintf(cfg->name, PCI_MMCFG_RESOURCE_NAME_LEN,
502 			 "PCI ECAM %04x [bus %02x-%02x]",
503 			 cfg->segment, cfg->start_bus, cfg->end_bus);
504 
505 		if (dev)
506 			dev_info(dev, "ECAM %pR (base %#lx) (size reduced!)\n",
507 				 &cfg->res, (unsigned long) cfg->address);
508 		else
509 			pr_info("ECAM %pR (base %#lx) for %04x [bus%02x-%02x] (size reduced!)\n",
510 				&cfg->res, (unsigned long) cfg->address,
511 				cfg->segment, cfg->start_bus, cfg->end_bus);
512 	}
513 
514 	return true;
515 }
516 
517 static bool __ref pci_mmcfg_reserved(struct device *dev,
518 				     struct pci_mmcfg_region *cfg, int early)
519 {
520 	struct resource *conflict;
521 
522 	if (early) {
523 
524 		/*
525 		 * Don't try to do this check unless configuration type 1
526 		 * is available.  How about type 2?
527 		 */
528 
529 		/*
530 		 * 946f2ee5c731 ("Check that MCFG points to an e820
531 		 * reserved area") added this E820 check in 2006 to work
532 		 * around BIOS defects.
533 		 *
534 		 * Per PCI Firmware r3.3, sec 4.1.2, ECAM space must be
535 		 * reserved by a PNP0C02 resource, but it need not be
536 		 * mentioned in E820.  Before the ACPI interpreter is
537 		 * available, we can't check for PNP0C02 resources, so
538 		 * there's no reliable way to verify the region in this
539 		 * early check.  Keep it only for the old machines that
540 		 * motivated 946f2ee5c731.
541 		 */
542 		if (dmi_get_bios_year() < 2016 && raw_pci_ops)
543 			return is_mmconf_reserved(e820__mapped_all, cfg, dev,
544 						  "E820 entry");
545 
546 		return true;
547 	}
548 
549 	if (!acpi_disabled) {
550 		if (is_mmconf_reserved(is_acpi_reserved, cfg, dev,
551 				       "ACPI motherboard resource"))
552 			return true;
553 
554 		if (dev)
555 			dev_info(dev, FW_INFO "ECAM %pR not reserved in ACPI motherboard resources\n",
556 				 &cfg->res);
557 		else
558 			pr_info(FW_INFO "ECAM %pR not reserved in ACPI motherboard resources\n",
559 			        &cfg->res);
560 
561 		if (is_efi_mmio(&cfg->res)) {
562 			pr_info("ECAM %pR is EfiMemoryMappedIO; assuming valid\n",
563 				&cfg->res);
564 			conflict = insert_resource_conflict(&iomem_resource,
565 							    &cfg->res);
566 			if (conflict)
567 				pr_warn("ECAM %pR conflicts with %s %pR\n",
568 					&cfg->res, conflict->name, conflict);
569 			else
570 				pr_info("ECAM %pR reserved to work around lack of ACPI motherboard _CRS\n",
571 					&cfg->res);
572 			return true;
573 		}
574 	}
575 
576 	/*
577 	 * e820__mapped_all() is marked as __init.
578 	 * All entries from ACPI MCFG table have been checked at boot time.
579 	 * For MCFG information constructed from hotpluggable host bridge's
580 	 * _CBA method, just assume it's reserved.
581 	 */
582 	return pci_mmcfg_running_state;
583 }
584 
585 static void __init pci_mmcfg_reject_broken(int early)
586 {
587 	struct pci_mmcfg_region *cfg;
588 
589 	list_for_each_entry(cfg, &pci_mmcfg_list, list) {
590 		if (!pci_mmcfg_reserved(NULL, cfg, early)) {
591 			pr_info("not using ECAM (%pR not reserved)\n",
592 				&cfg->res);
593 			free_all_mmcfg();
594 			return;
595 		}
596 	}
597 }
598 
599 static bool __init acpi_mcfg_valid_entry(struct acpi_table_mcfg *mcfg,
600 					 struct acpi_mcfg_allocation *cfg)
601 {
602 	if (cfg->address < 0xFFFFFFFF)
603 		return true;
604 
605 	if (!strncmp(mcfg->header.oem_id, "SGI", 3))
606 		return true;
607 
608 	if ((mcfg->header.revision >= 1) && (dmi_get_bios_year() >= 2010))
609 		return true;
610 
611 	pr_err("ECAM at %#llx for %04x [bus %02x-%02x] is above 4GB, ignored\n",
612 	       cfg->address, cfg->pci_segment, cfg->start_bus_number,
613 	       cfg->end_bus_number);
614 	return false;
615 }
616 
617 static int __init pci_parse_mcfg(struct acpi_table_header *header)
618 {
619 	struct acpi_table_mcfg *mcfg;
620 	struct acpi_mcfg_allocation *cfg_table, *cfg;
621 	unsigned long i;
622 	int entries;
623 
624 	if (!header)
625 		return -EINVAL;
626 
627 	mcfg = (struct acpi_table_mcfg *)header;
628 
629 	/* how many config structures do we have */
630 	free_all_mmcfg();
631 	entries = 0;
632 	i = header->length - sizeof(struct acpi_table_mcfg);
633 	while (i >= sizeof(struct acpi_mcfg_allocation)) {
634 		entries++;
635 		i -= sizeof(struct acpi_mcfg_allocation);
636 	}
637 	if (entries == 0) {
638 		pr_err("MCFG has no entries\n");
639 		return -ENODEV;
640 	}
641 
642 	cfg_table = (struct acpi_mcfg_allocation *) &mcfg[1];
643 	for (i = 0; i < entries; i++) {
644 		cfg = &cfg_table[i];
645 		if (!acpi_mcfg_valid_entry(mcfg, cfg)) {
646 			free_all_mmcfg();
647 			return -ENODEV;
648 		}
649 
650 		if (pci_mmconfig_add(cfg->pci_segment, cfg->start_bus_number,
651 				   cfg->end_bus_number, cfg->address) == NULL) {
652 			pr_warn("no memory for MCFG entries\n");
653 			free_all_mmcfg();
654 			return -ENOMEM;
655 		}
656 	}
657 
658 	return 0;
659 }
660 
661 #ifdef CONFIG_ACPI_APEI
662 extern int (*arch_apei_filter_addr)(int (*func)(__u64 start, __u64 size,
663 				     void *data), void *data);
664 
665 static int pci_mmcfg_for_each_region(int (*func)(__u64 start, __u64 size,
666 				     void *data), void *data)
667 {
668 	struct pci_mmcfg_region *cfg;
669 	int rc;
670 
671 	if (list_empty(&pci_mmcfg_list))
672 		return 0;
673 
674 	list_for_each_entry(cfg, &pci_mmcfg_list, list) {
675 		rc = func(cfg->res.start, resource_size(&cfg->res), data);
676 		if (rc)
677 			return rc;
678 	}
679 
680 	return 0;
681 }
682 #define set_apei_filter() (arch_apei_filter_addr = pci_mmcfg_for_each_region)
683 #else
684 #define set_apei_filter()
685 #endif
686 
687 static void __init __pci_mmcfg_init(int early)
688 {
689 	pr_debug("%s(%s)\n", __func__, early ? "early" : "late");
690 
691 	pci_mmcfg_reject_broken(early);
692 	if (list_empty(&pci_mmcfg_list))
693 		return;
694 
695 	if (pcibios_last_bus < 0) {
696 		const struct pci_mmcfg_region *cfg;
697 
698 		list_for_each_entry(cfg, &pci_mmcfg_list, list) {
699 			if (cfg->segment)
700 				break;
701 			pcibios_last_bus = cfg->end_bus;
702 		}
703 	}
704 
705 	if (pci_mmcfg_arch_init())
706 		pci_probe = (pci_probe & ~PCI_PROBE_MASK) | PCI_PROBE_MMCONF;
707 	else {
708 		free_all_mmcfg();
709 		pci_mmcfg_arch_init_failed = true;
710 	}
711 }
712 
713 static int __initdata known_bridge;
714 
715 void __init pci_mmcfg_early_init(void)
716 {
717 	pr_debug("%s() pci_probe %#x\n", __func__, pci_probe);
718 
719 	if (pci_probe & PCI_PROBE_MMCONF) {
720 		if (pci_mmcfg_check_hostbridge())
721 			known_bridge = 1;
722 		else
723 			acpi_table_parse(ACPI_SIG_MCFG, pci_parse_mcfg);
724 		__pci_mmcfg_init(1);
725 
726 		set_apei_filter();
727 	}
728 }
729 
730 void __init pci_mmcfg_late_init(void)
731 {
732 	pr_debug("%s() pci_probe %#x\n", __func__, pci_probe);
733 
734 	/* ECAM disabled */
735 	if ((pci_probe & PCI_PROBE_MMCONF) == 0)
736 		return;
737 
738 	if (known_bridge)
739 		return;
740 
741 	/* ECAM hasn't been enabled yet, try again */
742 	if (pci_probe & PCI_PROBE_MASK & ~PCI_PROBE_MMCONF) {
743 		acpi_table_parse(ACPI_SIG_MCFG, pci_parse_mcfg);
744 		__pci_mmcfg_init(0);
745 	}
746 }
747 
748 static int __init pci_mmcfg_late_insert_resources(void)
749 {
750 	struct pci_mmcfg_region *cfg;
751 
752 	pci_mmcfg_running_state = true;
753 
754 	pr_debug("%s() pci_probe %#x\n", __func__, pci_probe);
755 
756 	/* If we are not using ECAM, don't insert the resources. */
757 	if ((pci_probe & PCI_PROBE_MMCONF) == 0)
758 		return 1;
759 
760 	/*
761 	 * Attempt to insert the mmcfg resources but not with the busy flag
762 	 * marked so it won't cause request errors when __request_region is
763 	 * called.
764 	 */
765 	list_for_each_entry(cfg, &pci_mmcfg_list, list) {
766 		if (!cfg->res.parent) {
767 			pr_debug("%s() insert %pR\n", __func__, &cfg->res);
768 			insert_resource(&iomem_resource, &cfg->res);
769 		}
770 	}
771 
772 	return 0;
773 }
774 
775 /*
776  * Perform ECAM resource insertion after PCI initialization to allow for
777  * misprogrammed MCFG tables that state larger sizes but actually conflict
778  * with other system resources.
779  */
780 late_initcall(pci_mmcfg_late_insert_resources);
781 
782 /* Add ECAM information for host bridges */
783 int pci_mmconfig_insert(struct device *dev, u16 seg, u8 start, u8 end,
784 			phys_addr_t addr)
785 {
786 	int rc;
787 	struct resource *tmp = NULL;
788 	struct pci_mmcfg_region *cfg;
789 
790 	dev_dbg(dev, "%s(%04x [bus %02x-%02x])\n", __func__, seg, start, end);
791 
792 	if (!(pci_probe & PCI_PROBE_MMCONF) || pci_mmcfg_arch_init_failed)
793 		return -ENODEV;
794 
795 	if (start > end)
796 		return -EINVAL;
797 
798 	mutex_lock(&pci_mmcfg_lock);
799 	cfg = pci_mmconfig_lookup(seg, start);
800 	if (cfg) {
801 		if (cfg->end_bus < end)
802 			dev_info(dev, FW_INFO "ECAM %pR for domain %04x [bus %02x-%02x] only partially covers this bridge\n",
803 				 &cfg->res, cfg->segment, cfg->start_bus,
804 				 cfg->end_bus);
805 		mutex_unlock(&pci_mmcfg_lock);
806 		return -EEXIST;
807 	}
808 
809 	/*
810 	 * Don't move earlier; we must return -EEXIST, not -EINVAL, if
811 	 * pci_mmconfig_lookup() finds something
812 	 */
813 	if (!addr) {
814 		mutex_unlock(&pci_mmcfg_lock);
815 		return -EINVAL;
816 	}
817 
818 	rc = -EBUSY;
819 	cfg = pci_mmconfig_alloc(seg, start, end, addr);
820 	if (cfg == NULL) {
821 		dev_warn(dev, "fail to add ECAM (out of memory)\n");
822 		rc = -ENOMEM;
823 	} else if (!pci_mmcfg_reserved(dev, cfg, 0)) {
824 		dev_warn(dev, FW_BUG "ECAM %pR isn't reserved\n",
825 			 &cfg->res);
826 	} else {
827 		/* Insert resource if it's not in boot stage */
828 		if (pci_mmcfg_running_state)
829 			tmp = insert_resource_conflict(&iomem_resource,
830 						       &cfg->res);
831 
832 		if (tmp) {
833 			dev_warn(dev, "ECAM %pR conflicts with %s %pR\n",
834 				 &cfg->res, tmp->name, tmp);
835 		} else if (pci_mmcfg_arch_map(cfg)) {
836 			dev_warn(dev, "fail to map ECAM %pR\n", &cfg->res);
837 		} else {
838 			list_add_sorted(cfg);
839 			dev_info(dev, "ECAM %pR (base %#lx)\n",
840 				 &cfg->res, (unsigned long)addr);
841 			cfg = NULL;
842 			rc = 0;
843 		}
844 	}
845 
846 	if (cfg) {
847 		if (cfg->res.parent)
848 			release_resource(&cfg->res);
849 		kfree(cfg);
850 	}
851 
852 	mutex_unlock(&pci_mmcfg_lock);
853 
854 	return rc;
855 }
856 
857 /* Delete ECAM information for host bridges */
858 int pci_mmconfig_delete(u16 seg, u8 start, u8 end)
859 {
860 	struct pci_mmcfg_region *cfg;
861 
862 	mutex_lock(&pci_mmcfg_lock);
863 	list_for_each_entry_rcu(cfg, &pci_mmcfg_list, list)
864 		if (cfg->segment == seg && cfg->start_bus == start &&
865 		    cfg->end_bus == end) {
866 			list_del_rcu(&cfg->list);
867 			synchronize_rcu();
868 			pci_mmcfg_arch_unmap(cfg);
869 			if (cfg->res.parent)
870 				release_resource(&cfg->res);
871 			mutex_unlock(&pci_mmcfg_lock);
872 			kfree(cfg);
873 			return 0;
874 		}
875 	mutex_unlock(&pci_mmcfg_lock);
876 
877 	return -ENOENT;
878 }
879