xref: /openbmc/linux/arch/x86/pci/mmconfig-shared.c (revision a09d2831)
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/acpi.h>
16 #include <linux/sfi_acpi.h>
17 #include <linux/bitmap.h>
18 #include <linux/dmi.h>
19 #include <asm/e820.h>
20 #include <asm/pci_x86.h>
21 #include <asm/acpi.h>
22 
23 #define PREFIX "PCI: "
24 
25 /* Indicate if the mmcfg resources have been placed into the resource table. */
26 static int __initdata pci_mmcfg_resources_inserted;
27 
28 LIST_HEAD(pci_mmcfg_list);
29 
30 static __init void pci_mmconfig_remove(struct pci_mmcfg_region *cfg)
31 {
32 	if (cfg->res.parent)
33 		release_resource(&cfg->res);
34 	list_del(&cfg->list);
35 	kfree(cfg);
36 }
37 
38 static __init void free_all_mmcfg(void)
39 {
40 	struct pci_mmcfg_region *cfg, *tmp;
41 
42 	pci_mmcfg_arch_free();
43 	list_for_each_entry_safe(cfg, tmp, &pci_mmcfg_list, list)
44 		pci_mmconfig_remove(cfg);
45 }
46 
47 static __init void list_add_sorted(struct pci_mmcfg_region *new)
48 {
49 	struct pci_mmcfg_region *cfg;
50 
51 	/* keep list sorted by segment and starting bus number */
52 	list_for_each_entry(cfg, &pci_mmcfg_list, list) {
53 		if (cfg->segment > new->segment ||
54 		    (cfg->segment == new->segment &&
55 		     cfg->start_bus >= new->start_bus)) {
56 			list_add_tail(&new->list, &cfg->list);
57 			return;
58 		}
59 	}
60 	list_add_tail(&new->list, &pci_mmcfg_list);
61 }
62 
63 static __init struct pci_mmcfg_region *pci_mmconfig_add(int segment, int start,
64 							int end, u64 addr)
65 {
66 	struct pci_mmcfg_region *new;
67 	int num_buses;
68 	struct resource *res;
69 
70 	if (addr == 0)
71 		return NULL;
72 
73 	new = kzalloc(sizeof(*new), GFP_KERNEL);
74 	if (!new)
75 		return NULL;
76 
77 	new->address = addr;
78 	new->segment = segment;
79 	new->start_bus = start;
80 	new->end_bus = end;
81 
82 	list_add_sorted(new);
83 
84 	num_buses = end - start + 1;
85 	res = &new->res;
86 	res->start = addr + PCI_MMCFG_BUS_OFFSET(start);
87 	res->end = addr + PCI_MMCFG_BUS_OFFSET(num_buses) - 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 	printk(KERN_INFO PREFIX "MMCONFIG for domain %04x [bus %02x-%02x] at "
94 	       "%pR (base %#lx)\n", segment, start, end, &new->res,
95 	       (unsigned long) addr);
96 
97 	return new;
98 }
99 
100 struct pci_mmcfg_region *pci_mmconfig_lookup(int segment, int bus)
101 {
102 	struct pci_mmcfg_region *cfg;
103 
104 	list_for_each_entry(cfg, &pci_mmcfg_list, list)
105 		if (cfg->segment == segment &&
106 		    cfg->start_bus <= bus && bus <= cfg->end_bus)
107 			return cfg;
108 
109 	return NULL;
110 }
111 
112 static const char __init *pci_mmcfg_e7520(void)
113 {
114 	u32 win;
115 	raw_pci_ops->read(0, 0, PCI_DEVFN(0, 0), 0xce, 2, &win);
116 
117 	win = win & 0xf000;
118 	if (win == 0x0000 || win == 0xf000)
119 		return NULL;
120 
121 	if (pci_mmconfig_add(0, 0, 255, win << 16) == NULL)
122 		return NULL;
123 
124 	return "Intel Corporation E7520 Memory Controller Hub";
125 }
126 
127 static const char __init *pci_mmcfg_intel_945(void)
128 {
129 	u32 pciexbar, mask = 0, len = 0;
130 
131 	raw_pci_ops->read(0, 0, PCI_DEVFN(0, 0), 0x48, 4, &pciexbar);
132 
133 	/* Enable bit */
134 	if (!(pciexbar & 1))
135 		return NULL;
136 
137 	/* Size bits */
138 	switch ((pciexbar >> 1) & 3) {
139 	case 0:
140 		mask = 0xf0000000U;
141 		len  = 0x10000000U;
142 		break;
143 	case 1:
144 		mask = 0xf8000000U;
145 		len  = 0x08000000U;
146 		break;
147 	case 2:
148 		mask = 0xfc000000U;
149 		len  = 0x04000000U;
150 		break;
151 	default:
152 		return NULL;
153 	}
154 
155 	/* Errata #2, things break when not aligned on a 256Mb boundary */
156 	/* Can only happen in 64M/128M mode */
157 
158 	if ((pciexbar & mask) & 0x0fffffffU)
159 		return NULL;
160 
161 	/* Don't hit the APIC registers and their friends */
162 	if ((pciexbar & mask) >= 0xf0000000U)
163 		return NULL;
164 
165 	if (pci_mmconfig_add(0, 0, (len >> 20) - 1, pciexbar & mask) == NULL)
166 		return NULL;
167 
168 	return "Intel Corporation 945G/GZ/P/PL Express Memory Controller Hub";
169 }
170 
171 static const char __init *pci_mmcfg_amd_fam10h(void)
172 {
173 	u32 low, high, address;
174 	u64 base, msr;
175 	int i;
176 	unsigned segnbits = 0, busnbits, end_bus;
177 
178 	if (!(pci_probe & PCI_CHECK_ENABLE_AMD_MMCONF))
179 		return NULL;
180 
181 	address = MSR_FAM10H_MMIO_CONF_BASE;
182 	if (rdmsr_safe(address, &low, &high))
183 		return NULL;
184 
185 	msr = high;
186 	msr <<= 32;
187 	msr |= low;
188 
189 	/* mmconfig is not enable */
190 	if (!(msr & FAM10H_MMIO_CONF_ENABLE))
191 		return NULL;
192 
193 	base = msr & (FAM10H_MMIO_CONF_BASE_MASK<<FAM10H_MMIO_CONF_BASE_SHIFT);
194 
195 	busnbits = (msr >> FAM10H_MMIO_CONF_BUSRANGE_SHIFT) &
196 			 FAM10H_MMIO_CONF_BUSRANGE_MASK;
197 
198 	/*
199 	 * only handle bus 0 ?
200 	 * need to skip it
201 	 */
202 	if (!busnbits)
203 		return NULL;
204 
205 	if (busnbits > 8) {
206 		segnbits = busnbits - 8;
207 		busnbits = 8;
208 	}
209 
210 	end_bus = (1 << busnbits) - 1;
211 	for (i = 0; i < (1 << segnbits); i++)
212 		if (pci_mmconfig_add(i, 0, end_bus,
213 				     base + (1<<28) * i) == NULL) {
214 			free_all_mmcfg();
215 			return NULL;
216 		}
217 
218 	return "AMD Family 10h NB";
219 }
220 
221 static bool __initdata mcp55_checked;
222 static const char __init *pci_mmcfg_nvidia_mcp55(void)
223 {
224 	int bus;
225 	int mcp55_mmconf_found = 0;
226 
227 	static const u32 extcfg_regnum		= 0x90;
228 	static const u32 extcfg_regsize		= 4;
229 	static const u32 extcfg_enable_mask	= 1<<31;
230 	static const u32 extcfg_start_mask	= 0xff<<16;
231 	static const int extcfg_start_shift	= 16;
232 	static const u32 extcfg_size_mask	= 0x3<<28;
233 	static const int extcfg_size_shift	= 28;
234 	static const int extcfg_sizebus[]	= {0x100, 0x80, 0x40, 0x20};
235 	static const u32 extcfg_base_mask[]	= {0x7ff8, 0x7ffc, 0x7ffe, 0x7fff};
236 	static const int extcfg_base_lshift	= 25;
237 
238 	/*
239 	 * do check if amd fam10h already took over
240 	 */
241 	if (!acpi_disabled || !list_empty(&pci_mmcfg_list) || mcp55_checked)
242 		return NULL;
243 
244 	mcp55_checked = true;
245 	for (bus = 0; bus < 256; bus++) {
246 		u64 base;
247 		u32 l, extcfg;
248 		u16 vendor, device;
249 		int start, size_index, end;
250 
251 		raw_pci_ops->read(0, bus, PCI_DEVFN(0, 0), 0, 4, &l);
252 		vendor = l & 0xffff;
253 		device = (l >> 16) & 0xffff;
254 
255 		if (PCI_VENDOR_ID_NVIDIA != vendor || 0x0369 != device)
256 			continue;
257 
258 		raw_pci_ops->read(0, bus, PCI_DEVFN(0, 0), extcfg_regnum,
259 				  extcfg_regsize, &extcfg);
260 
261 		if (!(extcfg & extcfg_enable_mask))
262 			continue;
263 
264 		size_index = (extcfg & extcfg_size_mask) >> extcfg_size_shift;
265 		base = extcfg & extcfg_base_mask[size_index];
266 		/* base could > 4G */
267 		base <<= extcfg_base_lshift;
268 		start = (extcfg & extcfg_start_mask) >> extcfg_start_shift;
269 		end = start + extcfg_sizebus[size_index] - 1;
270 		if (pci_mmconfig_add(0, start, end, base) == NULL)
271 			continue;
272 		mcp55_mmconf_found++;
273 	}
274 
275 	if (!mcp55_mmconf_found)
276 		return NULL;
277 
278 	return "nVidia MCP55";
279 }
280 
281 struct pci_mmcfg_hostbridge_probe {
282 	u32 bus;
283 	u32 devfn;
284 	u32 vendor;
285 	u32 device;
286 	const char *(*probe)(void);
287 };
288 
289 static struct pci_mmcfg_hostbridge_probe pci_mmcfg_probes[] __initdata = {
290 	{ 0, PCI_DEVFN(0, 0), PCI_VENDOR_ID_INTEL,
291 	  PCI_DEVICE_ID_INTEL_E7520_MCH, pci_mmcfg_e7520 },
292 	{ 0, PCI_DEVFN(0, 0), PCI_VENDOR_ID_INTEL,
293 	  PCI_DEVICE_ID_INTEL_82945G_HB, pci_mmcfg_intel_945 },
294 	{ 0, PCI_DEVFN(0x18, 0), PCI_VENDOR_ID_AMD,
295 	  0x1200, pci_mmcfg_amd_fam10h },
296 	{ 0xff, PCI_DEVFN(0, 0), PCI_VENDOR_ID_AMD,
297 	  0x1200, pci_mmcfg_amd_fam10h },
298 	{ 0, PCI_DEVFN(0, 0), PCI_VENDOR_ID_NVIDIA,
299 	  0x0369, pci_mmcfg_nvidia_mcp55 },
300 };
301 
302 static void __init pci_mmcfg_check_end_bus_number(void)
303 {
304 	struct pci_mmcfg_region *cfg, *cfgx;
305 
306 	/* last one*/
307 	cfg = list_entry(pci_mmcfg_list.prev, typeof(*cfg), list);
308 	if (cfg)
309 		if (cfg->end_bus < cfg->start_bus)
310 			cfg->end_bus = 255;
311 
312 	if (list_is_singular(&pci_mmcfg_list))
313 		return;
314 
315 	/* don't overlap please */
316 	list_for_each_entry(cfg, &pci_mmcfg_list, list) {
317 		if (cfg->end_bus < cfg->start_bus)
318 			cfg->end_bus = 255;
319 
320 		cfgx = list_entry(cfg->list.next, typeof(*cfg), list);
321 		if (cfg != cfgx && cfg->end_bus >= cfgx->start_bus)
322 			cfg->end_bus = cfgx->start_bus - 1;
323 	}
324 }
325 
326 static int __init pci_mmcfg_check_hostbridge(void)
327 {
328 	u32 l;
329 	u32 bus, devfn;
330 	u16 vendor, device;
331 	int i;
332 	const char *name;
333 
334 	if (!raw_pci_ops)
335 		return 0;
336 
337 	free_all_mmcfg();
338 
339 	for (i = 0; i < ARRAY_SIZE(pci_mmcfg_probes); i++) {
340 		bus =  pci_mmcfg_probes[i].bus;
341 		devfn = pci_mmcfg_probes[i].devfn;
342 		raw_pci_ops->read(0, bus, devfn, 0, 4, &l);
343 		vendor = l & 0xffff;
344 		device = (l >> 16) & 0xffff;
345 
346 		name = NULL;
347 		if (pci_mmcfg_probes[i].vendor == vendor &&
348 		    pci_mmcfg_probes[i].device == device)
349 			name = pci_mmcfg_probes[i].probe();
350 
351 		if (name)
352 			printk(KERN_INFO PREFIX "%s with MMCONFIG support\n",
353 			       name);
354 	}
355 
356 	/* some end_bus_number is crazy, fix it */
357 	pci_mmcfg_check_end_bus_number();
358 
359 	return !list_empty(&pci_mmcfg_list);
360 }
361 
362 static void __init pci_mmcfg_insert_resources(void)
363 {
364 	struct pci_mmcfg_region *cfg;
365 
366 	list_for_each_entry(cfg, &pci_mmcfg_list, list)
367 		insert_resource(&iomem_resource, &cfg->res);
368 
369 	/* Mark that the resources have been inserted. */
370 	pci_mmcfg_resources_inserted = 1;
371 }
372 
373 static acpi_status __init check_mcfg_resource(struct acpi_resource *res,
374 					      void *data)
375 {
376 	struct resource *mcfg_res = data;
377 	struct acpi_resource_address64 address;
378 	acpi_status status;
379 
380 	if (res->type == ACPI_RESOURCE_TYPE_FIXED_MEMORY32) {
381 		struct acpi_resource_fixed_memory32 *fixmem32 =
382 			&res->data.fixed_memory32;
383 		if (!fixmem32)
384 			return AE_OK;
385 		if ((mcfg_res->start >= fixmem32->address) &&
386 		    (mcfg_res->end < (fixmem32->address +
387 				      fixmem32->address_length))) {
388 			mcfg_res->flags = 1;
389 			return AE_CTRL_TERMINATE;
390 		}
391 	}
392 	if ((res->type != ACPI_RESOURCE_TYPE_ADDRESS32) &&
393 	    (res->type != ACPI_RESOURCE_TYPE_ADDRESS64))
394 		return AE_OK;
395 
396 	status = acpi_resource_to_address64(res, &address);
397 	if (ACPI_FAILURE(status) ||
398 	   (address.address_length <= 0) ||
399 	   (address.resource_type != ACPI_MEMORY_RANGE))
400 		return AE_OK;
401 
402 	if ((mcfg_res->start >= address.minimum) &&
403 	    (mcfg_res->end < (address.minimum + address.address_length))) {
404 		mcfg_res->flags = 1;
405 		return AE_CTRL_TERMINATE;
406 	}
407 	return AE_OK;
408 }
409 
410 static acpi_status __init find_mboard_resource(acpi_handle handle, u32 lvl,
411 		void *context, void **rv)
412 {
413 	struct resource *mcfg_res = context;
414 
415 	acpi_walk_resources(handle, METHOD_NAME__CRS,
416 			    check_mcfg_resource, context);
417 
418 	if (mcfg_res->flags)
419 		return AE_CTRL_TERMINATE;
420 
421 	return AE_OK;
422 }
423 
424 static int __init is_acpi_reserved(u64 start, u64 end, unsigned not_used)
425 {
426 	struct resource mcfg_res;
427 
428 	mcfg_res.start = start;
429 	mcfg_res.end = end - 1;
430 	mcfg_res.flags = 0;
431 
432 	acpi_get_devices("PNP0C01", find_mboard_resource, &mcfg_res, NULL);
433 
434 	if (!mcfg_res.flags)
435 		acpi_get_devices("PNP0C02", find_mboard_resource, &mcfg_res,
436 				 NULL);
437 
438 	return mcfg_res.flags;
439 }
440 
441 typedef int (*check_reserved_t)(u64 start, u64 end, unsigned type);
442 
443 static int __init is_mmconf_reserved(check_reserved_t is_reserved,
444 				    struct pci_mmcfg_region *cfg, int with_e820)
445 {
446 	u64 addr = cfg->res.start;
447 	u64 size = resource_size(&cfg->res);
448 	u64 old_size = size;
449 	int valid = 0, num_buses;
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 		printk(KERN_INFO PREFIX "MMCONFIG at %pR reserved in %s\n",
459 		       &cfg->res,
460 		       with_e820 ? "E820" : "ACPI motherboard resources");
461 		valid = 1;
462 
463 		if (old_size != size) {
464 			/* update end_bus */
465 			cfg->end_bus = cfg->start_bus + ((size>>20) - 1);
466 			num_buses = cfg->end_bus - cfg->start_bus + 1;
467 			cfg->res.end = cfg->res.start +
468 			    PCI_MMCFG_BUS_OFFSET(num_buses) - 1;
469 			snprintf(cfg->name, PCI_MMCFG_RESOURCE_NAME_LEN,
470 				 "PCI MMCONFIG %04x [bus %02x-%02x]",
471 				 cfg->segment, cfg->start_bus, cfg->end_bus);
472 			printk(KERN_INFO PREFIX
473 			       "MMCONFIG for %04x [bus%02x-%02x] "
474 			       "at %pR (base %#lx) (size reduced!)\n",
475 			       cfg->segment, cfg->start_bus, cfg->end_bus,
476 			       &cfg->res, (unsigned long) cfg->address);
477 		}
478 	}
479 
480 	return valid;
481 }
482 
483 static void __init pci_mmcfg_reject_broken(int early)
484 {
485 	struct pci_mmcfg_region *cfg;
486 
487 	list_for_each_entry(cfg, &pci_mmcfg_list, list) {
488 		int valid = 0;
489 
490 		if (!early && !acpi_disabled)
491 			valid = is_mmconf_reserved(is_acpi_reserved, cfg, 0);
492 
493 		if (valid)
494 			continue;
495 
496 		if (!early)
497 			printk(KERN_ERR FW_BUG PREFIX
498 			       "MMCONFIG at %pR not reserved in "
499 			       "ACPI motherboard resources\n", &cfg->res);
500 
501 		/* Don't try to do this check unless configuration
502 		   type 1 is available. how about type 2 ?*/
503 		if (raw_pci_ops)
504 			valid = is_mmconf_reserved(e820_all_mapped, cfg, 1);
505 
506 		if (!valid)
507 			goto reject;
508 	}
509 
510 	return;
511 
512 reject:
513 	printk(KERN_INFO PREFIX "not using MMCONFIG\n");
514 	free_all_mmcfg();
515 }
516 
517 static int __initdata known_bridge;
518 
519 static int __init acpi_mcfg_check_entry(struct acpi_table_mcfg *mcfg,
520 					struct acpi_mcfg_allocation *cfg)
521 {
522 	int year;
523 
524 	if (cfg->address < 0xFFFFFFFF)
525 		return 0;
526 
527 	if (!strcmp(mcfg->header.oem_id, "SGI"))
528 		return 0;
529 
530 	if (mcfg->header.revision >= 1) {
531 		if (dmi_get_date(DMI_BIOS_DATE, &year, NULL, NULL) &&
532 		    year >= 2010)
533 			return 0;
534 	}
535 
536 	printk(KERN_ERR PREFIX "MCFG region for %04x [bus %02x-%02x] at %#llx "
537 	       "is above 4GB, ignored\n", cfg->pci_segment,
538 	       cfg->start_bus_number, cfg->end_bus_number, cfg->address);
539 	return -EINVAL;
540 }
541 
542 static int __init pci_parse_mcfg(struct acpi_table_header *header)
543 {
544 	struct acpi_table_mcfg *mcfg;
545 	struct acpi_mcfg_allocation *cfg_table, *cfg;
546 	unsigned long i;
547 	int entries;
548 
549 	if (!header)
550 		return -EINVAL;
551 
552 	mcfg = (struct acpi_table_mcfg *)header;
553 
554 	/* how many config structures do we have */
555 	free_all_mmcfg();
556 	entries = 0;
557 	i = header->length - sizeof(struct acpi_table_mcfg);
558 	while (i >= sizeof(struct acpi_mcfg_allocation)) {
559 		entries++;
560 		i -= sizeof(struct acpi_mcfg_allocation);
561 	};
562 	if (entries == 0) {
563 		printk(KERN_ERR PREFIX "MMCONFIG has no entries\n");
564 		return -ENODEV;
565 	}
566 
567 	cfg_table = (struct acpi_mcfg_allocation *) &mcfg[1];
568 	for (i = 0; i < entries; i++) {
569 		cfg = &cfg_table[i];
570 		if (acpi_mcfg_check_entry(mcfg, cfg)) {
571 			free_all_mmcfg();
572 			return -ENODEV;
573 		}
574 
575 		if (pci_mmconfig_add(cfg->pci_segment, cfg->start_bus_number,
576 				   cfg->end_bus_number, cfg->address) == NULL) {
577 			printk(KERN_WARNING PREFIX
578 			       "no memory for MCFG entries\n");
579 			free_all_mmcfg();
580 			return -ENOMEM;
581 		}
582 	}
583 
584 	return 0;
585 }
586 
587 static void __init __pci_mmcfg_init(int early)
588 {
589 	/* MMCONFIG disabled */
590 	if ((pci_probe & PCI_PROBE_MMCONF) == 0)
591 		return;
592 
593 	/* MMCONFIG already enabled */
594 	if (!early && !(pci_probe & PCI_PROBE_MASK & ~PCI_PROBE_MMCONF))
595 		return;
596 
597 	/* for late to exit */
598 	if (known_bridge)
599 		return;
600 
601 	if (early) {
602 		if (pci_mmcfg_check_hostbridge())
603 			known_bridge = 1;
604 	}
605 
606 	if (!known_bridge)
607 		acpi_sfi_table_parse(ACPI_SIG_MCFG, pci_parse_mcfg);
608 
609 	pci_mmcfg_reject_broken(early);
610 
611 	if (list_empty(&pci_mmcfg_list))
612 		return;
613 
614 	if (pci_mmcfg_arch_init())
615 		pci_probe = (pci_probe & ~PCI_PROBE_MASK) | PCI_PROBE_MMCONF;
616 	else {
617 		/*
618 		 * Signal not to attempt to insert mmcfg resources because
619 		 * the architecture mmcfg setup could not initialize.
620 		 */
621 		pci_mmcfg_resources_inserted = 1;
622 	}
623 }
624 
625 void __init pci_mmcfg_early_init(void)
626 {
627 	__pci_mmcfg_init(1);
628 }
629 
630 void __init pci_mmcfg_late_init(void)
631 {
632 	__pci_mmcfg_init(0);
633 }
634 
635 static int __init pci_mmcfg_late_insert_resources(void)
636 {
637 	/*
638 	 * If resources are already inserted or we are not using MMCONFIG,
639 	 * don't insert the resources.
640 	 */
641 	if ((pci_mmcfg_resources_inserted == 1) ||
642 	    (pci_probe & PCI_PROBE_MMCONF) == 0 ||
643 	    list_empty(&pci_mmcfg_list))
644 		return 1;
645 
646 	/*
647 	 * Attempt to insert the mmcfg resources but not with the busy flag
648 	 * marked so it won't cause request errors when __request_region is
649 	 * called.
650 	 */
651 	pci_mmcfg_insert_resources();
652 
653 	return 0;
654 }
655 
656 /*
657  * Perform MMCONFIG resource insertion after PCI initialization to allow for
658  * misprogrammed MCFG tables that state larger sizes but actually conflict
659  * with other system resources.
660  */
661 late_initcall(pci_mmcfg_late_insert_resources);
662