xref: /openbmc/linux/arch/x86/pci/mmconfig-shared.c (revision 5f32c314)
1 /*
2  * mmconfig-shared.c - Low-level direct PCI config space access via
3  *                     MMCONFIG - common code between i386 and x86-64.
4  *
5  * This code does:
6  * - known chipset handling
7  * - ACPI decoding and validation
8  *
9  * Per-architecture code takes care of the mappings and accesses
10  * themselves.
11  */
12 
13 #include <linux/pci.h>
14 #include <linux/init.h>
15 #include <linux/sfi_acpi.h>
16 #include <linux/bitmap.h>
17 #include <linux/dmi.h>
18 #include <linux/slab.h>
19 #include <linux/mutex.h>
20 #include <linux/rculist.h>
21 #include <asm/e820.h>
22 #include <asm/pci_x86.h>
23 #include <asm/acpi.h>
24 
25 #define PREFIX "PCI: "
26 
27 /* Indicate if the mmcfg resources have been placed into the resource table. */
28 static bool pci_mmcfg_running_state;
29 static bool pci_mmcfg_arch_init_failed;
30 static DEFINE_MUTEX(pci_mmcfg_lock);
31 
32 LIST_HEAD(pci_mmcfg_list);
33 
34 static __init void pci_mmconfig_remove(struct pci_mmcfg_region *cfg)
35 {
36 	if (cfg->res.parent)
37 		release_resource(&cfg->res);
38 	list_del(&cfg->list);
39 	kfree(cfg);
40 }
41 
42 static __init void free_all_mmcfg(void)
43 {
44 	struct pci_mmcfg_region *cfg, *tmp;
45 
46 	pci_mmcfg_arch_free();
47 	list_for_each_entry_safe(cfg, tmp, &pci_mmcfg_list, list)
48 		pci_mmconfig_remove(cfg);
49 }
50 
51 static void list_add_sorted(struct pci_mmcfg_region *new)
52 {
53 	struct pci_mmcfg_region *cfg;
54 
55 	/* keep list sorted by segment and starting bus number */
56 	list_for_each_entry_rcu(cfg, &pci_mmcfg_list, list) {
57 		if (cfg->segment > new->segment ||
58 		    (cfg->segment == new->segment &&
59 		     cfg->start_bus >= new->start_bus)) {
60 			list_add_tail_rcu(&new->list, &cfg->list);
61 			return;
62 		}
63 	}
64 	list_add_tail_rcu(&new->list, &pci_mmcfg_list);
65 }
66 
67 static struct pci_mmcfg_region *pci_mmconfig_alloc(int segment, int start,
68 						   int end, u64 addr)
69 {
70 	struct pci_mmcfg_region *new;
71 	struct resource *res;
72 
73 	if (addr == 0)
74 		return NULL;
75 
76 	new = kzalloc(sizeof(*new), GFP_KERNEL);
77 	if (!new)
78 		return NULL;
79 
80 	new->address = addr;
81 	new->segment = segment;
82 	new->start_bus = start;
83 	new->end_bus = end;
84 
85 	res = &new->res;
86 	res->start = addr + PCI_MMCFG_BUS_OFFSET(start);
87 	res->end = addr + PCI_MMCFG_BUS_OFFSET(end + 1) - 1;
88 	res->flags = IORESOURCE_MEM | IORESOURCE_BUSY;
89 	snprintf(new->name, PCI_MMCFG_RESOURCE_NAME_LEN,
90 		 "PCI MMCONFIG %04x [bus %02x-%02x]", segment, start, end);
91 	res->name = new->name;
92 
93 	return new;
94 }
95 
96 static __init struct pci_mmcfg_region *pci_mmconfig_add(int segment, int start,
97 							int end, u64 addr)
98 {
99 	struct pci_mmcfg_region *new;
100 
101 	new = pci_mmconfig_alloc(segment, start, end, addr);
102 	if (new) {
103 		mutex_lock(&pci_mmcfg_lock);
104 		list_add_sorted(new);
105 		mutex_unlock(&pci_mmcfg_lock);
106 
107 		pr_info(PREFIX
108 		       "MMCONFIG for domain %04x [bus %02x-%02x] at %pR "
109 		       "(base %#lx)\n",
110 		       segment, start, end, &new->res, (unsigned long)addr);
111 	}
112 
113 	return new;
114 }
115 
116 struct pci_mmcfg_region *pci_mmconfig_lookup(int segment, int bus)
117 {
118 	struct pci_mmcfg_region *cfg;
119 
120 	list_for_each_entry_rcu(cfg, &pci_mmcfg_list, list)
121 		if (cfg->segment == segment &&
122 		    cfg->start_bus <= bus && bus <= cfg->end_bus)
123 			return cfg;
124 
125 	return NULL;
126 }
127 
128 static const char __init *pci_mmcfg_e7520(void)
129 {
130 	u32 win;
131 	raw_pci_ops->read(0, 0, PCI_DEVFN(0, 0), 0xce, 2, &win);
132 
133 	win = win & 0xf000;
134 	if (win == 0x0000 || win == 0xf000)
135 		return NULL;
136 
137 	if (pci_mmconfig_add(0, 0, 255, win << 16) == NULL)
138 		return NULL;
139 
140 	return "Intel Corporation E7520 Memory Controller Hub";
141 }
142 
143 static const char __init *pci_mmcfg_intel_945(void)
144 {
145 	u32 pciexbar, mask = 0, len = 0;
146 
147 	raw_pci_ops->read(0, 0, PCI_DEVFN(0, 0), 0x48, 4, &pciexbar);
148 
149 	/* Enable bit */
150 	if (!(pciexbar & 1))
151 		return NULL;
152 
153 	/* Size bits */
154 	switch ((pciexbar >> 1) & 3) {
155 	case 0:
156 		mask = 0xf0000000U;
157 		len  = 0x10000000U;
158 		break;
159 	case 1:
160 		mask = 0xf8000000U;
161 		len  = 0x08000000U;
162 		break;
163 	case 2:
164 		mask = 0xfc000000U;
165 		len  = 0x04000000U;
166 		break;
167 	default:
168 		return NULL;
169 	}
170 
171 	/* Errata #2, things break when not aligned on a 256Mb boundary */
172 	/* Can only happen in 64M/128M mode */
173 
174 	if ((pciexbar & mask) & 0x0fffffffU)
175 		return NULL;
176 
177 	/* Don't hit the APIC registers and their friends */
178 	if ((pciexbar & mask) >= 0xf0000000U)
179 		return NULL;
180 
181 	if (pci_mmconfig_add(0, 0, (len >> 20) - 1, pciexbar & mask) == NULL)
182 		return NULL;
183 
184 	return "Intel Corporation 945G/GZ/P/PL Express Memory Controller Hub";
185 }
186 
187 static const char __init *pci_mmcfg_amd_fam10h(void)
188 {
189 	u32 low, high, address;
190 	u64 base, msr;
191 	int i;
192 	unsigned segnbits = 0, busnbits, end_bus;
193 
194 	if (!(pci_probe & PCI_CHECK_ENABLE_AMD_MMCONF))
195 		return NULL;
196 
197 	address = MSR_FAM10H_MMIO_CONF_BASE;
198 	if (rdmsr_safe(address, &low, &high))
199 		return NULL;
200 
201 	msr = high;
202 	msr <<= 32;
203 	msr |= low;
204 
205 	/* mmconfig is not enable */
206 	if (!(msr & FAM10H_MMIO_CONF_ENABLE))
207 		return NULL;
208 
209 	base = msr & (FAM10H_MMIO_CONF_BASE_MASK<<FAM10H_MMIO_CONF_BASE_SHIFT);
210 
211 	busnbits = (msr >> FAM10H_MMIO_CONF_BUSRANGE_SHIFT) &
212 			 FAM10H_MMIO_CONF_BUSRANGE_MASK;
213 
214 	/*
215 	 * only handle bus 0 ?
216 	 * need to skip it
217 	 */
218 	if (!busnbits)
219 		return NULL;
220 
221 	if (busnbits > 8) {
222 		segnbits = busnbits - 8;
223 		busnbits = 8;
224 	}
225 
226 	end_bus = (1 << busnbits) - 1;
227 	for (i = 0; i < (1 << segnbits); i++)
228 		if (pci_mmconfig_add(i, 0, end_bus,
229 				     base + (1<<28) * i) == NULL) {
230 			free_all_mmcfg();
231 			return NULL;
232 		}
233 
234 	return "AMD Family 10h NB";
235 }
236 
237 static bool __initdata mcp55_checked;
238 static const char __init *pci_mmcfg_nvidia_mcp55(void)
239 {
240 	int bus;
241 	int mcp55_mmconf_found = 0;
242 
243 	static const u32 extcfg_regnum		= 0x90;
244 	static const u32 extcfg_regsize		= 4;
245 	static const u32 extcfg_enable_mask	= 1<<31;
246 	static const u32 extcfg_start_mask	= 0xff<<16;
247 	static const int extcfg_start_shift	= 16;
248 	static const u32 extcfg_size_mask	= 0x3<<28;
249 	static const int extcfg_size_shift	= 28;
250 	static const int extcfg_sizebus[]	= {0x100, 0x80, 0x40, 0x20};
251 	static const u32 extcfg_base_mask[]	= {0x7ff8, 0x7ffc, 0x7ffe, 0x7fff};
252 	static const int extcfg_base_lshift	= 25;
253 
254 	/*
255 	 * do check if amd fam10h already took over
256 	 */
257 	if (!acpi_disabled || !list_empty(&pci_mmcfg_list) || mcp55_checked)
258 		return NULL;
259 
260 	mcp55_checked = true;
261 	for (bus = 0; bus < 256; bus++) {
262 		u64 base;
263 		u32 l, extcfg;
264 		u16 vendor, device;
265 		int start, size_index, end;
266 
267 		raw_pci_ops->read(0, bus, PCI_DEVFN(0, 0), 0, 4, &l);
268 		vendor = l & 0xffff;
269 		device = (l >> 16) & 0xffff;
270 
271 		if (PCI_VENDOR_ID_NVIDIA != vendor || 0x0369 != device)
272 			continue;
273 
274 		raw_pci_ops->read(0, bus, PCI_DEVFN(0, 0), extcfg_regnum,
275 				  extcfg_regsize, &extcfg);
276 
277 		if (!(extcfg & extcfg_enable_mask))
278 			continue;
279 
280 		size_index = (extcfg & extcfg_size_mask) >> extcfg_size_shift;
281 		base = extcfg & extcfg_base_mask[size_index];
282 		/* base could > 4G */
283 		base <<= extcfg_base_lshift;
284 		start = (extcfg & extcfg_start_mask) >> extcfg_start_shift;
285 		end = start + extcfg_sizebus[size_index] - 1;
286 		if (pci_mmconfig_add(0, start, end, base) == NULL)
287 			continue;
288 		mcp55_mmconf_found++;
289 	}
290 
291 	if (!mcp55_mmconf_found)
292 		return NULL;
293 
294 	return "nVidia MCP55";
295 }
296 
297 struct pci_mmcfg_hostbridge_probe {
298 	u32 bus;
299 	u32 devfn;
300 	u32 vendor;
301 	u32 device;
302 	const char *(*probe)(void);
303 };
304 
305 static struct pci_mmcfg_hostbridge_probe pci_mmcfg_probes[] __initdata = {
306 	{ 0, PCI_DEVFN(0, 0), PCI_VENDOR_ID_INTEL,
307 	  PCI_DEVICE_ID_INTEL_E7520_MCH, pci_mmcfg_e7520 },
308 	{ 0, PCI_DEVFN(0, 0), PCI_VENDOR_ID_INTEL,
309 	  PCI_DEVICE_ID_INTEL_82945G_HB, pci_mmcfg_intel_945 },
310 	{ 0, PCI_DEVFN(0x18, 0), PCI_VENDOR_ID_AMD,
311 	  0x1200, pci_mmcfg_amd_fam10h },
312 	{ 0xff, PCI_DEVFN(0, 0), PCI_VENDOR_ID_AMD,
313 	  0x1200, pci_mmcfg_amd_fam10h },
314 	{ 0, PCI_DEVFN(0, 0), PCI_VENDOR_ID_NVIDIA,
315 	  0x0369, pci_mmcfg_nvidia_mcp55 },
316 };
317 
318 static void __init pci_mmcfg_check_end_bus_number(void)
319 {
320 	struct pci_mmcfg_region *cfg, *cfgx;
321 
322 	/* Fixup overlaps */
323 	list_for_each_entry(cfg, &pci_mmcfg_list, list) {
324 		if (cfg->end_bus < cfg->start_bus)
325 			cfg->end_bus = 255;
326 
327 		/* Don't access the list head ! */
328 		if (cfg->list.next == &pci_mmcfg_list)
329 			break;
330 
331 		cfgx = list_entry(cfg->list.next, typeof(*cfg), list);
332 		if (cfg->end_bus >= cfgx->start_bus)
333 			cfg->end_bus = cfgx->start_bus - 1;
334 	}
335 }
336 
337 static int __init pci_mmcfg_check_hostbridge(void)
338 {
339 	u32 l;
340 	u32 bus, devfn;
341 	u16 vendor, device;
342 	int i;
343 	const char *name;
344 
345 	if (!raw_pci_ops)
346 		return 0;
347 
348 	free_all_mmcfg();
349 
350 	for (i = 0; i < ARRAY_SIZE(pci_mmcfg_probes); i++) {
351 		bus =  pci_mmcfg_probes[i].bus;
352 		devfn = pci_mmcfg_probes[i].devfn;
353 		raw_pci_ops->read(0, bus, devfn, 0, 4, &l);
354 		vendor = l & 0xffff;
355 		device = (l >> 16) & 0xffff;
356 
357 		name = NULL;
358 		if (pci_mmcfg_probes[i].vendor == vendor &&
359 		    pci_mmcfg_probes[i].device == device)
360 			name = pci_mmcfg_probes[i].probe();
361 
362 		if (name)
363 			pr_info(PREFIX "%s with MMCONFIG support\n", name);
364 	}
365 
366 	/* some end_bus_number is crazy, fix it */
367 	pci_mmcfg_check_end_bus_number();
368 
369 	return !list_empty(&pci_mmcfg_list);
370 }
371 
372 static acpi_status check_mcfg_resource(struct acpi_resource *res, void *data)
373 {
374 	struct resource *mcfg_res = data;
375 	struct acpi_resource_address64 address;
376 	acpi_status status;
377 
378 	if (res->type == ACPI_RESOURCE_TYPE_FIXED_MEMORY32) {
379 		struct acpi_resource_fixed_memory32 *fixmem32 =
380 			&res->data.fixed_memory32;
381 		if (!fixmem32)
382 			return AE_OK;
383 		if ((mcfg_res->start >= fixmem32->address) &&
384 		    (mcfg_res->end < (fixmem32->address +
385 				      fixmem32->address_length))) {
386 			mcfg_res->flags = 1;
387 			return AE_CTRL_TERMINATE;
388 		}
389 	}
390 	if ((res->type != ACPI_RESOURCE_TYPE_ADDRESS32) &&
391 	    (res->type != ACPI_RESOURCE_TYPE_ADDRESS64))
392 		return AE_OK;
393 
394 	status = acpi_resource_to_address64(res, &address);
395 	if (ACPI_FAILURE(status) ||
396 	   (address.address_length <= 0) ||
397 	   (address.resource_type != ACPI_MEMORY_RANGE))
398 		return AE_OK;
399 
400 	if ((mcfg_res->start >= address.minimum) &&
401 	    (mcfg_res->end < (address.minimum + address.address_length))) {
402 		mcfg_res->flags = 1;
403 		return AE_CTRL_TERMINATE;
404 	}
405 	return AE_OK;
406 }
407 
408 static acpi_status find_mboard_resource(acpi_handle handle, u32 lvl,
409 					void *context, void **rv)
410 {
411 	struct resource *mcfg_res = context;
412 
413 	acpi_walk_resources(handle, METHOD_NAME__CRS,
414 			    check_mcfg_resource, context);
415 
416 	if (mcfg_res->flags)
417 		return AE_CTRL_TERMINATE;
418 
419 	return AE_OK;
420 }
421 
422 static int is_acpi_reserved(u64 start, u64 end, unsigned not_used)
423 {
424 	struct resource mcfg_res;
425 
426 	mcfg_res.start = start;
427 	mcfg_res.end = end - 1;
428 	mcfg_res.flags = 0;
429 
430 	acpi_get_devices("PNP0C01", find_mboard_resource, &mcfg_res, NULL);
431 
432 	if (!mcfg_res.flags)
433 		acpi_get_devices("PNP0C02", find_mboard_resource, &mcfg_res,
434 				 NULL);
435 
436 	return mcfg_res.flags;
437 }
438 
439 typedef int (*check_reserved_t)(u64 start, u64 end, unsigned type);
440 
441 static int __ref is_mmconf_reserved(check_reserved_t is_reserved,
442 				    struct pci_mmcfg_region *cfg,
443 				    struct device *dev, int with_e820)
444 {
445 	u64 addr = cfg->res.start;
446 	u64 size = resource_size(&cfg->res);
447 	u64 old_size = size;
448 	int num_buses;
449 	char *method = with_e820 ? "E820" : "ACPI motherboard resources";
450 
451 	while (!is_reserved(addr, addr + size, E820_RESERVED)) {
452 		size >>= 1;
453 		if (size < (16UL<<20))
454 			break;
455 	}
456 
457 	if (size < (16UL<<20) && size != old_size)
458 		return 0;
459 
460 	if (dev)
461 		dev_info(dev, "MMCONFIG at %pR reserved in %s\n",
462 			 &cfg->res, method);
463 	else
464 		pr_info(PREFIX "MMCONFIG at %pR reserved in %s\n",
465 		       &cfg->res, method);
466 
467 	if (old_size != size) {
468 		/* update end_bus */
469 		cfg->end_bus = cfg->start_bus + ((size>>20) - 1);
470 		num_buses = cfg->end_bus - cfg->start_bus + 1;
471 		cfg->res.end = cfg->res.start +
472 		    PCI_MMCFG_BUS_OFFSET(num_buses) - 1;
473 		snprintf(cfg->name, PCI_MMCFG_RESOURCE_NAME_LEN,
474 			 "PCI MMCONFIG %04x [bus %02x-%02x]",
475 			 cfg->segment, cfg->start_bus, cfg->end_bus);
476 
477 		if (dev)
478 			dev_info(dev,
479 				"MMCONFIG "
480 				"at %pR (base %#lx) (size reduced!)\n",
481 				&cfg->res, (unsigned long) cfg->address);
482 		else
483 			pr_info(PREFIX
484 				"MMCONFIG for %04x [bus%02x-%02x] "
485 				"at %pR (base %#lx) (size reduced!)\n",
486 				cfg->segment, cfg->start_bus, cfg->end_bus,
487 				&cfg->res, (unsigned long) cfg->address);
488 	}
489 
490 	return 1;
491 }
492 
493 static int __ref pci_mmcfg_check_reserved(struct device *dev,
494 		  struct pci_mmcfg_region *cfg, int early)
495 {
496 	if (!early && !acpi_disabled) {
497 		if (is_mmconf_reserved(is_acpi_reserved, cfg, dev, 0))
498 			return 1;
499 
500 		if (dev)
501 			dev_info(dev, FW_INFO
502 				 "MMCONFIG at %pR not reserved in "
503 				 "ACPI motherboard resources\n",
504 				 &cfg->res);
505 		else
506 			pr_info(FW_INFO PREFIX
507 			       "MMCONFIG at %pR not reserved in "
508 			       "ACPI motherboard resources\n",
509 			       &cfg->res);
510 	}
511 
512 	/*
513 	 * e820_all_mapped() is marked as __init.
514 	 * All entries from ACPI MCFG table have been checked at boot time.
515 	 * For MCFG information constructed from hotpluggable host bridge's
516 	 * _CBA method, just assume it's reserved.
517 	 */
518 	if (pci_mmcfg_running_state)
519 		return 1;
520 
521 	/* Don't try to do this check unless configuration
522 	   type 1 is available. how about type 2 ?*/
523 	if (raw_pci_ops)
524 		return is_mmconf_reserved(e820_all_mapped, cfg, dev, 1);
525 
526 	return 0;
527 }
528 
529 static void __init pci_mmcfg_reject_broken(int early)
530 {
531 	struct pci_mmcfg_region *cfg;
532 
533 	list_for_each_entry(cfg, &pci_mmcfg_list, list) {
534 		if (pci_mmcfg_check_reserved(NULL, cfg, early) == 0) {
535 			pr_info(PREFIX "not using MMCONFIG\n");
536 			free_all_mmcfg();
537 			return;
538 		}
539 	}
540 }
541 
542 static int __init acpi_mcfg_check_entry(struct acpi_table_mcfg *mcfg,
543 					struct acpi_mcfg_allocation *cfg)
544 {
545 	int year;
546 
547 	if (cfg->address < 0xFFFFFFFF)
548 		return 0;
549 
550 	if (!strncmp(mcfg->header.oem_id, "SGI", 3))
551 		return 0;
552 
553 	if (mcfg->header.revision >= 1) {
554 		if (dmi_get_date(DMI_BIOS_DATE, &year, NULL, NULL) &&
555 		    year >= 2010)
556 			return 0;
557 	}
558 
559 	pr_err(PREFIX "MCFG region for %04x [bus %02x-%02x] at %#llx "
560 	       "is above 4GB, ignored\n", cfg->pci_segment,
561 	       cfg->start_bus_number, cfg->end_bus_number, cfg->address);
562 	return -EINVAL;
563 }
564 
565 static int __init pci_parse_mcfg(struct acpi_table_header *header)
566 {
567 	struct acpi_table_mcfg *mcfg;
568 	struct acpi_mcfg_allocation *cfg_table, *cfg;
569 	unsigned long i;
570 	int entries;
571 
572 	if (!header)
573 		return -EINVAL;
574 
575 	mcfg = (struct acpi_table_mcfg *)header;
576 
577 	/* how many config structures do we have */
578 	free_all_mmcfg();
579 	entries = 0;
580 	i = header->length - sizeof(struct acpi_table_mcfg);
581 	while (i >= sizeof(struct acpi_mcfg_allocation)) {
582 		entries++;
583 		i -= sizeof(struct acpi_mcfg_allocation);
584 	}
585 	if (entries == 0) {
586 		pr_err(PREFIX "MMCONFIG has no entries\n");
587 		return -ENODEV;
588 	}
589 
590 	cfg_table = (struct acpi_mcfg_allocation *) &mcfg[1];
591 	for (i = 0; i < entries; i++) {
592 		cfg = &cfg_table[i];
593 		if (acpi_mcfg_check_entry(mcfg, cfg)) {
594 			free_all_mmcfg();
595 			return -ENODEV;
596 		}
597 
598 		if (pci_mmconfig_add(cfg->pci_segment, cfg->start_bus_number,
599 				   cfg->end_bus_number, cfg->address) == NULL) {
600 			pr_warn(PREFIX "no memory for MCFG entries\n");
601 			free_all_mmcfg();
602 			return -ENOMEM;
603 		}
604 	}
605 
606 	return 0;
607 }
608 
609 static void __init __pci_mmcfg_init(int early)
610 {
611 	pci_mmcfg_reject_broken(early);
612 	if (list_empty(&pci_mmcfg_list))
613 		return;
614 
615 	if (pcibios_last_bus < 0) {
616 		const struct pci_mmcfg_region *cfg;
617 
618 		list_for_each_entry(cfg, &pci_mmcfg_list, list) {
619 			if (cfg->segment)
620 				break;
621 			pcibios_last_bus = cfg->end_bus;
622 		}
623 	}
624 
625 	if (pci_mmcfg_arch_init())
626 		pci_probe = (pci_probe & ~PCI_PROBE_MASK) | PCI_PROBE_MMCONF;
627 	else {
628 		free_all_mmcfg();
629 		pci_mmcfg_arch_init_failed = true;
630 	}
631 }
632 
633 static int __initdata known_bridge;
634 
635 void __init pci_mmcfg_early_init(void)
636 {
637 	if (pci_probe & PCI_PROBE_MMCONF) {
638 		if (pci_mmcfg_check_hostbridge())
639 			known_bridge = 1;
640 		else
641 			acpi_sfi_table_parse(ACPI_SIG_MCFG, pci_parse_mcfg);
642 		__pci_mmcfg_init(1);
643 	}
644 }
645 
646 void __init pci_mmcfg_late_init(void)
647 {
648 	/* MMCONFIG disabled */
649 	if ((pci_probe & PCI_PROBE_MMCONF) == 0)
650 		return;
651 
652 	if (known_bridge)
653 		return;
654 
655 	/* MMCONFIG hasn't been enabled yet, try again */
656 	if (pci_probe & PCI_PROBE_MASK & ~PCI_PROBE_MMCONF) {
657 		acpi_sfi_table_parse(ACPI_SIG_MCFG, pci_parse_mcfg);
658 		__pci_mmcfg_init(0);
659 	}
660 }
661 
662 static int __init pci_mmcfg_late_insert_resources(void)
663 {
664 	struct pci_mmcfg_region *cfg;
665 
666 	pci_mmcfg_running_state = true;
667 
668 	/* If we are not using MMCONFIG, don't insert the resources. */
669 	if ((pci_probe & PCI_PROBE_MMCONF) == 0)
670 		return 1;
671 
672 	/*
673 	 * Attempt to insert the mmcfg resources but not with the busy flag
674 	 * marked so it won't cause request errors when __request_region is
675 	 * called.
676 	 */
677 	list_for_each_entry(cfg, &pci_mmcfg_list, list)
678 		if (!cfg->res.parent)
679 			insert_resource(&iomem_resource, &cfg->res);
680 
681 	return 0;
682 }
683 
684 /*
685  * Perform MMCONFIG resource insertion after PCI initialization to allow for
686  * misprogrammed MCFG tables that state larger sizes but actually conflict
687  * with other system resources.
688  */
689 late_initcall(pci_mmcfg_late_insert_resources);
690 
691 /* Add MMCFG information for host bridges */
692 int pci_mmconfig_insert(struct device *dev, u16 seg, u8 start, u8 end,
693 			phys_addr_t addr)
694 {
695 	int rc;
696 	struct resource *tmp = NULL;
697 	struct pci_mmcfg_region *cfg;
698 
699 	if (!(pci_probe & PCI_PROBE_MMCONF) || pci_mmcfg_arch_init_failed)
700 		return -ENODEV;
701 
702 	if (start > end)
703 		return -EINVAL;
704 
705 	mutex_lock(&pci_mmcfg_lock);
706 	cfg = pci_mmconfig_lookup(seg, start);
707 	if (cfg) {
708 		if (cfg->end_bus < end)
709 			dev_info(dev, FW_INFO
710 				 "MMCONFIG for "
711 				 "domain %04x [bus %02x-%02x] "
712 				 "only partially covers this bridge\n",
713 				  cfg->segment, cfg->start_bus, cfg->end_bus);
714 		mutex_unlock(&pci_mmcfg_lock);
715 		return -EEXIST;
716 	}
717 
718 	if (!addr) {
719 		mutex_unlock(&pci_mmcfg_lock);
720 		return -EINVAL;
721 	}
722 
723 	rc = -EBUSY;
724 	cfg = pci_mmconfig_alloc(seg, start, end, addr);
725 	if (cfg == NULL) {
726 		dev_warn(dev, "fail to add MMCONFIG (out of memory)\n");
727 		rc = -ENOMEM;
728 	} else if (!pci_mmcfg_check_reserved(dev, cfg, 0)) {
729 		dev_warn(dev, FW_BUG "MMCONFIG %pR isn't reserved\n",
730 			 &cfg->res);
731 	} else {
732 		/* Insert resource if it's not in boot stage */
733 		if (pci_mmcfg_running_state)
734 			tmp = insert_resource_conflict(&iomem_resource,
735 						       &cfg->res);
736 
737 		if (tmp) {
738 			dev_warn(dev,
739 				 "MMCONFIG %pR conflicts with "
740 				 "%s %pR\n",
741 				 &cfg->res, tmp->name, tmp);
742 		} else if (pci_mmcfg_arch_map(cfg)) {
743 			dev_warn(dev, "fail to map MMCONFIG %pR.\n",
744 				 &cfg->res);
745 		} else {
746 			list_add_sorted(cfg);
747 			dev_info(dev, "MMCONFIG at %pR (base %#lx)\n",
748 				 &cfg->res, (unsigned long)addr);
749 			cfg = NULL;
750 			rc = 0;
751 		}
752 	}
753 
754 	if (cfg) {
755 		if (cfg->res.parent)
756 			release_resource(&cfg->res);
757 		kfree(cfg);
758 	}
759 
760 	mutex_unlock(&pci_mmcfg_lock);
761 
762 	return rc;
763 }
764 
765 /* Delete MMCFG information for host bridges */
766 int pci_mmconfig_delete(u16 seg, u8 start, u8 end)
767 {
768 	struct pci_mmcfg_region *cfg;
769 
770 	mutex_lock(&pci_mmcfg_lock);
771 	list_for_each_entry_rcu(cfg, &pci_mmcfg_list, list)
772 		if (cfg->segment == seg && cfg->start_bus == start &&
773 		    cfg->end_bus == end) {
774 			list_del_rcu(&cfg->list);
775 			synchronize_rcu();
776 			pci_mmcfg_arch_unmap(cfg);
777 			if (cfg->res.parent)
778 				release_resource(&cfg->res);
779 			mutex_unlock(&pci_mmcfg_lock);
780 			kfree(cfg);
781 			return 0;
782 		}
783 	mutex_unlock(&pci_mmcfg_lock);
784 
785 	return -ENOENT;
786 }
787