x86/PCI: MMCONFIG: centralize MCFG structure management
[zen-stable.git] / arch / x86 / pci / mmconfig-shared.c
blob62a8ecd96980ebbf38b57cc2194c8a65dffaa0d5
1 /*
2 * mmconfig-shared.c - Low-level direct PCI config space access via
3 * MMCONFIG - common code between i386 and x86-64.
5 * This code does:
6 * - known chipset handling
7 * - ACPI decoding and validation
9 * Per-architecture code takes care of the mappings and accesses
10 * themselves.
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 <linux/sort.h>
20 #include <asm/e820.h>
21 #include <asm/pci_x86.h>
22 #include <asm/acpi.h>
24 #define PREFIX "PCI: "
26 /* Indicate if the mmcfg resources have been placed into the resource table. */
27 static int __initdata pci_mmcfg_resources_inserted;
29 static __init void free_all_mmcfg(void)
31 pci_mmcfg_arch_free();
32 pci_mmcfg_config_num = 0;
33 kfree(pci_mmcfg_config);
34 pci_mmcfg_config = NULL;
37 static __init struct acpi_mcfg_allocation *pci_mmconfig_add(int segment,
38 int start, int end, u64 addr)
40 struct acpi_mcfg_allocation *new;
41 int new_num = pci_mmcfg_config_num + 1;
42 int i = pci_mmcfg_config_num;
44 new = kzalloc(sizeof(pci_mmcfg_config[0]) * new_num, GFP_KERNEL);
45 if (!new)
46 return NULL;
48 if (pci_mmcfg_config) {
49 memcpy(new, pci_mmcfg_config,
50 sizeof(pci_mmcfg_config[0]) * new_num);
51 kfree(pci_mmcfg_config);
53 pci_mmcfg_config = new;
55 pci_mmcfg_config_num++;
56 pci_mmcfg_config[i].address = addr;
57 pci_mmcfg_config[i].pci_segment = segment;
58 pci_mmcfg_config[i].start_bus_number = start;
59 pci_mmcfg_config[i].end_bus_number = end;
61 return &pci_mmcfg_config[i];
64 static const char __init *pci_mmcfg_e7520(void)
66 u32 win;
67 raw_pci_ops->read(0, 0, PCI_DEVFN(0, 0), 0xce, 2, &win);
69 win = win & 0xf000;
70 if (win == 0x0000 || win == 0xf000)
71 return NULL;
73 if (pci_mmconfig_add(0, 0, 255, win << 16) == NULL)
74 return NULL;
76 return "Intel Corporation E7520 Memory Controller Hub";
79 static const char __init *pci_mmcfg_intel_945(void)
81 u32 pciexbar, mask = 0, len = 0;
83 raw_pci_ops->read(0, 0, PCI_DEVFN(0, 0), 0x48, 4, &pciexbar);
85 /* Enable bit */
86 if (!(pciexbar & 1))
87 return NULL;
89 /* Size bits */
90 switch ((pciexbar >> 1) & 3) {
91 case 0:
92 mask = 0xf0000000U;
93 len = 0x10000000U;
94 break;
95 case 1:
96 mask = 0xf8000000U;
97 len = 0x08000000U;
98 break;
99 case 2:
100 mask = 0xfc000000U;
101 len = 0x04000000U;
102 break;
103 default:
104 return NULL;
107 /* Errata #2, things break when not aligned on a 256Mb boundary */
108 /* Can only happen in 64M/128M mode */
110 if ((pciexbar & mask) & 0x0fffffffU)
111 return NULL;
113 /* Don't hit the APIC registers and their friends */
114 if ((pciexbar & mask) >= 0xf0000000U)
115 return NULL;
117 if (pci_mmconfig_add(0, 0, (len >> 20) - 1, pciexbar & mask) == NULL)
118 return NULL;
120 return "Intel Corporation 945G/GZ/P/PL Express Memory Controller Hub";
123 static const char __init *pci_mmcfg_amd_fam10h(void)
125 u32 low, high, address;
126 u64 base, msr;
127 int i;
128 unsigned segnbits = 0, busnbits, end_bus;
130 if (!(pci_probe & PCI_CHECK_ENABLE_AMD_MMCONF))
131 return NULL;
133 address = MSR_FAM10H_MMIO_CONF_BASE;
134 if (rdmsr_safe(address, &low, &high))
135 return NULL;
137 msr = high;
138 msr <<= 32;
139 msr |= low;
141 /* mmconfig is not enable */
142 if (!(msr & FAM10H_MMIO_CONF_ENABLE))
143 return NULL;
145 base = msr & (FAM10H_MMIO_CONF_BASE_MASK<<FAM10H_MMIO_CONF_BASE_SHIFT);
147 busnbits = (msr >> FAM10H_MMIO_CONF_BUSRANGE_SHIFT) &
148 FAM10H_MMIO_CONF_BUSRANGE_MASK;
151 * only handle bus 0 ?
152 * need to skip it
154 if (!busnbits)
155 return NULL;
157 if (busnbits > 8) {
158 segnbits = busnbits - 8;
159 busnbits = 8;
162 end_bus = (1 << busnbits) - 1;
163 for (i = 0; i < (1 << segnbits); i++)
164 if (pci_mmconfig_add(i, 0, end_bus,
165 base + (1<<28) * i) == NULL) {
166 free_all_mmcfg();
167 return NULL;
170 return "AMD Family 10h NB";
173 static bool __initdata mcp55_checked;
174 static const char __init *pci_mmcfg_nvidia_mcp55(void)
176 int bus;
177 int mcp55_mmconf_found = 0;
179 static const u32 extcfg_regnum = 0x90;
180 static const u32 extcfg_regsize = 4;
181 static const u32 extcfg_enable_mask = 1<<31;
182 static const u32 extcfg_start_mask = 0xff<<16;
183 static const int extcfg_start_shift = 16;
184 static const u32 extcfg_size_mask = 0x3<<28;
185 static const int extcfg_size_shift = 28;
186 static const int extcfg_sizebus[] = {0x100, 0x80, 0x40, 0x20};
187 static const u32 extcfg_base_mask[] = {0x7ff8, 0x7ffc, 0x7ffe, 0x7fff};
188 static const int extcfg_base_lshift = 25;
191 * do check if amd fam10h already took over
193 if (!acpi_disabled || pci_mmcfg_config_num || mcp55_checked)
194 return NULL;
196 mcp55_checked = true;
197 for (bus = 0; bus < 256; bus++) {
198 u64 base;
199 u32 l, extcfg;
200 u16 vendor, device;
201 int start, size_index, end;
203 raw_pci_ops->read(0, bus, PCI_DEVFN(0, 0), 0, 4, &l);
204 vendor = l & 0xffff;
205 device = (l >> 16) & 0xffff;
207 if (PCI_VENDOR_ID_NVIDIA != vendor || 0x0369 != device)
208 continue;
210 raw_pci_ops->read(0, bus, PCI_DEVFN(0, 0), extcfg_regnum,
211 extcfg_regsize, &extcfg);
213 if (!(extcfg & extcfg_enable_mask))
214 continue;
216 size_index = (extcfg & extcfg_size_mask) >> extcfg_size_shift;
217 base = extcfg & extcfg_base_mask[size_index];
218 /* base could > 4G */
219 base <<= extcfg_base_lshift;
220 start = (extcfg & extcfg_start_mask) >> extcfg_start_shift;
221 end = start + extcfg_sizebus[size_index] - 1;
222 if (pci_mmconfig_add(0, start, end, base) == NULL)
223 continue;
224 mcp55_mmconf_found++;
227 if (!mcp55_mmconf_found)
228 return NULL;
230 return "nVidia MCP55";
233 struct pci_mmcfg_hostbridge_probe {
234 u32 bus;
235 u32 devfn;
236 u32 vendor;
237 u32 device;
238 const char *(*probe)(void);
241 static struct pci_mmcfg_hostbridge_probe pci_mmcfg_probes[] __initdata = {
242 { 0, PCI_DEVFN(0, 0), PCI_VENDOR_ID_INTEL,
243 PCI_DEVICE_ID_INTEL_E7520_MCH, pci_mmcfg_e7520 },
244 { 0, PCI_DEVFN(0, 0), PCI_VENDOR_ID_INTEL,
245 PCI_DEVICE_ID_INTEL_82945G_HB, pci_mmcfg_intel_945 },
246 { 0, PCI_DEVFN(0x18, 0), PCI_VENDOR_ID_AMD,
247 0x1200, pci_mmcfg_amd_fam10h },
248 { 0xff, PCI_DEVFN(0, 0), PCI_VENDOR_ID_AMD,
249 0x1200, pci_mmcfg_amd_fam10h },
250 { 0, PCI_DEVFN(0, 0), PCI_VENDOR_ID_NVIDIA,
251 0x0369, pci_mmcfg_nvidia_mcp55 },
254 static int __init cmp_mmcfg(const void *x1, const void *x2)
256 const typeof(pci_mmcfg_config[0]) *m1 = x1;
257 const typeof(pci_mmcfg_config[0]) *m2 = x2;
258 int start1, start2;
260 start1 = m1->start_bus_number;
261 start2 = m2->start_bus_number;
263 return start1 - start2;
266 static void __init pci_mmcfg_check_end_bus_number(void)
268 int i;
269 typeof(pci_mmcfg_config[0]) *cfg, *cfgx;
271 /* sort them at first */
272 sort(pci_mmcfg_config, pci_mmcfg_config_num,
273 sizeof(pci_mmcfg_config[0]), cmp_mmcfg, NULL);
275 /* last one*/
276 if (pci_mmcfg_config_num > 0) {
277 i = pci_mmcfg_config_num - 1;
278 cfg = &pci_mmcfg_config[i];
279 if (cfg->end_bus_number < cfg->start_bus_number)
280 cfg->end_bus_number = 255;
283 /* don't overlap please */
284 for (i = 0; i < pci_mmcfg_config_num - 1; i++) {
285 cfg = &pci_mmcfg_config[i];
286 cfgx = &pci_mmcfg_config[i+1];
288 if (cfg->end_bus_number < cfg->start_bus_number)
289 cfg->end_bus_number = 255;
291 if (cfg->end_bus_number >= cfgx->start_bus_number)
292 cfg->end_bus_number = cfgx->start_bus_number - 1;
296 static int __init pci_mmcfg_check_hostbridge(void)
298 u32 l;
299 u32 bus, devfn;
300 u16 vendor, device;
301 int i;
302 const char *name;
304 if (!raw_pci_ops)
305 return 0;
307 free_all_mmcfg();
309 for (i = 0; i < ARRAY_SIZE(pci_mmcfg_probes); i++) {
310 bus = pci_mmcfg_probes[i].bus;
311 devfn = pci_mmcfg_probes[i].devfn;
312 raw_pci_ops->read(0, bus, devfn, 0, 4, &l);
313 vendor = l & 0xffff;
314 device = (l >> 16) & 0xffff;
316 name = NULL;
317 if (pci_mmcfg_probes[i].vendor == vendor &&
318 pci_mmcfg_probes[i].device == device)
319 name = pci_mmcfg_probes[i].probe();
321 if (name)
322 printk(KERN_INFO "PCI: Found %s with MMCONFIG support.\n",
323 name);
326 /* some end_bus_number is crazy, fix it */
327 pci_mmcfg_check_end_bus_number();
329 return pci_mmcfg_config_num != 0;
332 static void __init pci_mmcfg_insert_resources(void)
334 #define PCI_MMCFG_RESOURCE_NAME_LEN 24
335 int i;
336 struct resource *res;
337 char *names;
338 unsigned num_buses;
340 res = kcalloc(PCI_MMCFG_RESOURCE_NAME_LEN + sizeof(*res),
341 pci_mmcfg_config_num, GFP_KERNEL);
342 if (!res) {
343 printk(KERN_ERR "PCI: Unable to allocate MMCONFIG resources\n");
344 return;
347 names = (void *)&res[pci_mmcfg_config_num];
348 for (i = 0; i < pci_mmcfg_config_num; i++, res++) {
349 struct acpi_mcfg_allocation *cfg = &pci_mmcfg_config[i];
350 num_buses = cfg->end_bus_number - cfg->start_bus_number + 1;
351 res->name = names;
352 snprintf(names, PCI_MMCFG_RESOURCE_NAME_LEN,
353 "PCI MMCONFIG %u [%02x-%02x]", cfg->pci_segment,
354 cfg->start_bus_number, cfg->end_bus_number);
355 res->start = cfg->address + (cfg->start_bus_number << 20);
356 res->end = res->start + (num_buses << 20) - 1;
357 res->flags = IORESOURCE_MEM | IORESOURCE_BUSY;
358 insert_resource(&iomem_resource, res);
359 names += PCI_MMCFG_RESOURCE_NAME_LEN;
362 /* Mark that the resources have been inserted. */
363 pci_mmcfg_resources_inserted = 1;
366 static acpi_status __init check_mcfg_resource(struct acpi_resource *res,
367 void *data)
369 struct resource *mcfg_res = data;
370 struct acpi_resource_address64 address;
371 acpi_status status;
373 if (res->type == ACPI_RESOURCE_TYPE_FIXED_MEMORY32) {
374 struct acpi_resource_fixed_memory32 *fixmem32 =
375 &res->data.fixed_memory32;
376 if (!fixmem32)
377 return AE_OK;
378 if ((mcfg_res->start >= fixmem32->address) &&
379 (mcfg_res->end < (fixmem32->address +
380 fixmem32->address_length))) {
381 mcfg_res->flags = 1;
382 return AE_CTRL_TERMINATE;
385 if ((res->type != ACPI_RESOURCE_TYPE_ADDRESS32) &&
386 (res->type != ACPI_RESOURCE_TYPE_ADDRESS64))
387 return AE_OK;
389 status = acpi_resource_to_address64(res, &address);
390 if (ACPI_FAILURE(status) ||
391 (address.address_length <= 0) ||
392 (address.resource_type != ACPI_MEMORY_RANGE))
393 return AE_OK;
395 if ((mcfg_res->start >= address.minimum) &&
396 (mcfg_res->end < (address.minimum + address.address_length))) {
397 mcfg_res->flags = 1;
398 return AE_CTRL_TERMINATE;
400 return AE_OK;
403 static acpi_status __init find_mboard_resource(acpi_handle handle, u32 lvl,
404 void *context, void **rv)
406 struct resource *mcfg_res = context;
408 acpi_walk_resources(handle, METHOD_NAME__CRS,
409 check_mcfg_resource, context);
411 if (mcfg_res->flags)
412 return AE_CTRL_TERMINATE;
414 return AE_OK;
417 static int __init is_acpi_reserved(u64 start, u64 end, unsigned not_used)
419 struct resource mcfg_res;
421 mcfg_res.start = start;
422 mcfg_res.end = end - 1;
423 mcfg_res.flags = 0;
425 acpi_get_devices("PNP0C01", find_mboard_resource, &mcfg_res, NULL);
427 if (!mcfg_res.flags)
428 acpi_get_devices("PNP0C02", find_mboard_resource, &mcfg_res,
429 NULL);
431 return mcfg_res.flags;
434 typedef int (*check_reserved_t)(u64 start, u64 end, unsigned type);
436 static int __init is_mmconf_reserved(check_reserved_t is_reserved,
437 u64 addr, u64 size, int i,
438 typeof(pci_mmcfg_config[0]) *cfg, int with_e820)
440 u64 old_size = size;
441 int valid = 0;
443 while (!is_reserved(addr, addr + size, E820_RESERVED)) {
444 size >>= 1;
445 if (size < (16UL<<20))
446 break;
449 if (size >= (16UL<<20) || size == old_size) {
450 printk(KERN_NOTICE
451 "PCI: MCFG area at %Lx reserved in %s\n",
452 addr, with_e820?"E820":"ACPI motherboard resources");
453 valid = 1;
455 if (old_size != size) {
456 /* update end_bus_number */
457 cfg->end_bus_number = cfg->start_bus_number + ((size>>20) - 1);
458 printk(KERN_NOTICE "PCI: updated MCFG configuration %d: base %lx "
459 "segment %hu buses %u - %u\n",
460 i, (unsigned long)cfg->address, cfg->pci_segment,
461 (unsigned int)cfg->start_bus_number,
462 (unsigned int)cfg->end_bus_number);
466 return valid;
469 static void __init pci_mmcfg_reject_broken(int early)
471 typeof(pci_mmcfg_config[0]) *cfg;
472 int i;
474 if ((pci_mmcfg_config_num == 0) ||
475 (pci_mmcfg_config == NULL) ||
476 (pci_mmcfg_config[0].address == 0))
477 return;
479 for (i = 0; i < pci_mmcfg_config_num; i++) {
480 int valid = 0;
481 u64 addr, size;
483 cfg = &pci_mmcfg_config[i];
484 addr = cfg->start_bus_number;
485 addr <<= 20;
486 addr += cfg->address;
487 size = cfg->end_bus_number + 1 - cfg->start_bus_number;
488 size <<= 20;
489 printk(KERN_NOTICE "PCI: MCFG configuration %d: base %lx "
490 "segment %hu buses %u - %u\n",
491 i, (unsigned long)cfg->address, cfg->pci_segment,
492 (unsigned int)cfg->start_bus_number,
493 (unsigned int)cfg->end_bus_number);
495 if (!early && !acpi_disabled)
496 valid = is_mmconf_reserved(is_acpi_reserved, addr, size, i, cfg, 0);
498 if (valid)
499 continue;
501 if (!early)
502 printk(KERN_ERR "PCI: BIOS Bug: MCFG area at %Lx is not"
503 " reserved in ACPI motherboard resources\n",
504 cfg->address);
506 /* Don't try to do this check unless configuration
507 type 1 is available. how about type 2 ?*/
508 if (raw_pci_ops)
509 valid = is_mmconf_reserved(e820_all_mapped, addr, size, i, cfg, 1);
511 if (!valid)
512 goto reject;
515 return;
517 reject:
518 printk(KERN_INFO "PCI: Not using MMCONFIG.\n");
519 free_all_mmcfg();
522 static int __initdata known_bridge;
524 /* The physical address of the MMCONFIG aperture. Set from ACPI tables. */
525 struct acpi_mcfg_allocation *pci_mmcfg_config;
526 int pci_mmcfg_config_num;
528 static int __init acpi_mcfg_check_entry(struct acpi_table_mcfg *mcfg,
529 struct acpi_mcfg_allocation *cfg)
531 int year;
533 if (cfg->address < 0xFFFFFFFF)
534 return 0;
536 if (!strcmp(mcfg->header.oem_id, "SGI"))
537 return 0;
539 if (mcfg->header.revision >= 1) {
540 if (dmi_get_date(DMI_BIOS_DATE, &year, NULL, NULL) &&
541 year >= 2010)
542 return 0;
545 printk(KERN_ERR PREFIX "MCFG region for %04x:%02x-%02x at %#llx "
546 "is above 4GB, ignored\n", cfg->pci_segment,
547 cfg->start_bus_number, cfg->end_bus_number, cfg->address);
548 return -EINVAL;
551 static int __init pci_parse_mcfg(struct acpi_table_header *header)
553 struct acpi_table_mcfg *mcfg;
554 struct acpi_mcfg_allocation *cfg_table, *cfg;
555 unsigned long i;
556 int entries;
558 if (!header)
559 return -EINVAL;
561 mcfg = (struct acpi_table_mcfg *)header;
563 /* how many config structures do we have */
564 free_all_mmcfg();
565 entries = 0;
566 i = header->length - sizeof(struct acpi_table_mcfg);
567 while (i >= sizeof(struct acpi_mcfg_allocation)) {
568 entries++;
569 i -= sizeof(struct acpi_mcfg_allocation);
571 if (entries == 0) {
572 printk(KERN_ERR PREFIX "MMCONFIG has no entries\n");
573 return -ENODEV;
576 cfg_table = (struct acpi_mcfg_allocation *) &mcfg[1];
577 for (i = 0; i < entries; i++) {
578 cfg = &cfg_table[i];
579 if (acpi_mcfg_check_entry(mcfg, cfg)) {
580 free_all_mmcfg();
581 return -ENODEV;
584 if (pci_mmconfig_add(cfg->pci_segment, cfg->start_bus_number,
585 cfg->end_bus_number, cfg->address) == NULL) {
586 printk(KERN_WARNING PREFIX
587 "no memory for MCFG entries\n");
588 free_all_mmcfg();
589 return -ENOMEM;
593 return 0;
596 static void __init __pci_mmcfg_init(int early)
598 /* MMCONFIG disabled */
599 if ((pci_probe & PCI_PROBE_MMCONF) == 0)
600 return;
602 /* MMCONFIG already enabled */
603 if (!early && !(pci_probe & PCI_PROBE_MASK & ~PCI_PROBE_MMCONF))
604 return;
606 /* for late to exit */
607 if (known_bridge)
608 return;
610 if (early) {
611 if (pci_mmcfg_check_hostbridge())
612 known_bridge = 1;
615 if (!known_bridge)
616 acpi_sfi_table_parse(ACPI_SIG_MCFG, pci_parse_mcfg);
618 pci_mmcfg_reject_broken(early);
620 if ((pci_mmcfg_config_num == 0) ||
621 (pci_mmcfg_config == NULL) ||
622 (pci_mmcfg_config[0].address == 0))
623 return;
625 if (pci_mmcfg_arch_init())
626 pci_probe = (pci_probe & ~PCI_PROBE_MASK) | PCI_PROBE_MMCONF;
627 else {
629 * Signal not to attempt to insert mmcfg resources because
630 * the architecture mmcfg setup could not initialize.
632 pci_mmcfg_resources_inserted = 1;
636 void __init pci_mmcfg_early_init(void)
638 __pci_mmcfg_init(1);
641 void __init pci_mmcfg_late_init(void)
643 __pci_mmcfg_init(0);
646 static int __init pci_mmcfg_late_insert_resources(void)
649 * If resources are already inserted or we are not using MMCONFIG,
650 * don't insert the resources.
652 if ((pci_mmcfg_resources_inserted == 1) ||
653 (pci_probe & PCI_PROBE_MMCONF) == 0 ||
654 (pci_mmcfg_config_num == 0) ||
655 (pci_mmcfg_config == NULL) ||
656 (pci_mmcfg_config[0].address == 0))
657 return 1;
660 * Attempt to insert the mmcfg resources but not with the busy flag
661 * marked so it won't cause request errors when __request_region is
662 * called.
664 pci_mmcfg_insert_resources();
666 return 0;
670 * Perform MMCONFIG resource insertion after PCI initialization to allow for
671 * misprogrammed MCFG tables that state larger sizes but actually conflict
672 * with other system resources.
674 late_initcall(pci_mmcfg_late_insert_resources);