kvm tools, setup: Create private directory
[linux-2.6/next.git] / arch / x86 / pci / mmconfig-shared.c
blob750c346ef50a4251a1a4971cee1443b3f022b13a
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/slab.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 LIST_HEAD(pci_mmcfg_list);
31 static __init void pci_mmconfig_remove(struct pci_mmcfg_region *cfg)
33 if (cfg->res.parent)
34 release_resource(&cfg->res);
35 list_del(&cfg->list);
36 kfree(cfg);
39 static __init void free_all_mmcfg(void)
41 struct pci_mmcfg_region *cfg, *tmp;
43 pci_mmcfg_arch_free();
44 list_for_each_entry_safe(cfg, tmp, &pci_mmcfg_list, list)
45 pci_mmconfig_remove(cfg);
48 static __init void list_add_sorted(struct pci_mmcfg_region *new)
50 struct pci_mmcfg_region *cfg;
52 /* keep list sorted by segment and starting bus number */
53 list_for_each_entry(cfg, &pci_mmcfg_list, list) {
54 if (cfg->segment > new->segment ||
55 (cfg->segment == new->segment &&
56 cfg->start_bus >= new->start_bus)) {
57 list_add_tail(&new->list, &cfg->list);
58 return;
61 list_add_tail(&new->list, &pci_mmcfg_list);
64 static __init struct pci_mmcfg_region *pci_mmconfig_add(int segment, int start,
65 int end, u64 addr)
67 struct pci_mmcfg_region *new;
68 struct resource *res;
70 if (addr == 0)
71 return NULL;
73 new = kzalloc(sizeof(*new), GFP_KERNEL);
74 if (!new)
75 return NULL;
77 new->address = addr;
78 new->segment = segment;
79 new->start_bus = start;
80 new->end_bus = end;
82 list_add_sorted(new);
84 res = &new->res;
85 res->start = addr + PCI_MMCFG_BUS_OFFSET(start);
86 res->end = addr + PCI_MMCFG_BUS_OFFSET(end + 1) - 1;
87 res->flags = IORESOURCE_MEM | IORESOURCE_BUSY;
88 snprintf(new->name, PCI_MMCFG_RESOURCE_NAME_LEN,
89 "PCI MMCONFIG %04x [bus %02x-%02x]", segment, start, end);
90 res->name = new->name;
92 printk(KERN_INFO PREFIX "MMCONFIG for domain %04x [bus %02x-%02x] at "
93 "%pR (base %#lx)\n", segment, start, end, &new->res,
94 (unsigned long) addr);
96 return new;
99 struct pci_mmcfg_region *pci_mmconfig_lookup(int segment, int bus)
101 struct pci_mmcfg_region *cfg;
103 list_for_each_entry(cfg, &pci_mmcfg_list, list)
104 if (cfg->segment == segment &&
105 cfg->start_bus <= bus && bus <= cfg->end_bus)
106 return cfg;
108 return NULL;
111 static const char __init *pci_mmcfg_e7520(void)
113 u32 win;
114 raw_pci_ops->read(0, 0, PCI_DEVFN(0, 0), 0xce, 2, &win);
116 win = win & 0xf000;
117 if (win == 0x0000 || win == 0xf000)
118 return NULL;
120 if (pci_mmconfig_add(0, 0, 255, win << 16) == NULL)
121 return NULL;
123 return "Intel Corporation E7520 Memory Controller Hub";
126 static const char __init *pci_mmcfg_intel_945(void)
128 u32 pciexbar, mask = 0, len = 0;
130 raw_pci_ops->read(0, 0, PCI_DEVFN(0, 0), 0x48, 4, &pciexbar);
132 /* Enable bit */
133 if (!(pciexbar & 1))
134 return NULL;
136 /* Size bits */
137 switch ((pciexbar >> 1) & 3) {
138 case 0:
139 mask = 0xf0000000U;
140 len = 0x10000000U;
141 break;
142 case 1:
143 mask = 0xf8000000U;
144 len = 0x08000000U;
145 break;
146 case 2:
147 mask = 0xfc000000U;
148 len = 0x04000000U;
149 break;
150 default:
151 return NULL;
154 /* Errata #2, things break when not aligned on a 256Mb boundary */
155 /* Can only happen in 64M/128M mode */
157 if ((pciexbar & mask) & 0x0fffffffU)
158 return NULL;
160 /* Don't hit the APIC registers and their friends */
161 if ((pciexbar & mask) >= 0xf0000000U)
162 return NULL;
164 if (pci_mmconfig_add(0, 0, (len >> 20) - 1, pciexbar & mask) == NULL)
165 return NULL;
167 return "Intel Corporation 945G/GZ/P/PL Express Memory Controller Hub";
170 static const char __init *pci_mmcfg_amd_fam10h(void)
172 u32 low, high, address;
173 u64 base, msr;
174 int i;
175 unsigned segnbits = 0, busnbits, end_bus;
177 if (!(pci_probe & PCI_CHECK_ENABLE_AMD_MMCONF))
178 return NULL;
180 address = MSR_FAM10H_MMIO_CONF_BASE;
181 if (rdmsr_safe(address, &low, &high))
182 return NULL;
184 msr = high;
185 msr <<= 32;
186 msr |= low;
188 /* mmconfig is not enable */
189 if (!(msr & FAM10H_MMIO_CONF_ENABLE))
190 return NULL;
192 base = msr & (FAM10H_MMIO_CONF_BASE_MASK<<FAM10H_MMIO_CONF_BASE_SHIFT);
194 busnbits = (msr >> FAM10H_MMIO_CONF_BUSRANGE_SHIFT) &
195 FAM10H_MMIO_CONF_BUSRANGE_MASK;
198 * only handle bus 0 ?
199 * need to skip it
201 if (!busnbits)
202 return NULL;
204 if (busnbits > 8) {
205 segnbits = busnbits - 8;
206 busnbits = 8;
209 end_bus = (1 << busnbits) - 1;
210 for (i = 0; i < (1 << segnbits); i++)
211 if (pci_mmconfig_add(i, 0, end_bus,
212 base + (1<<28) * i) == NULL) {
213 free_all_mmcfg();
214 return NULL;
217 return "AMD Family 10h NB";
220 static bool __initdata mcp55_checked;
221 static const char __init *pci_mmcfg_nvidia_mcp55(void)
223 int bus;
224 int mcp55_mmconf_found = 0;
226 static const u32 extcfg_regnum = 0x90;
227 static const u32 extcfg_regsize = 4;
228 static const u32 extcfg_enable_mask = 1<<31;
229 static const u32 extcfg_start_mask = 0xff<<16;
230 static const int extcfg_start_shift = 16;
231 static const u32 extcfg_size_mask = 0x3<<28;
232 static const int extcfg_size_shift = 28;
233 static const int extcfg_sizebus[] = {0x100, 0x80, 0x40, 0x20};
234 static const u32 extcfg_base_mask[] = {0x7ff8, 0x7ffc, 0x7ffe, 0x7fff};
235 static const int extcfg_base_lshift = 25;
238 * do check if amd fam10h already took over
240 if (!acpi_disabled || !list_empty(&pci_mmcfg_list) || mcp55_checked)
241 return NULL;
243 mcp55_checked = true;
244 for (bus = 0; bus < 256; bus++) {
245 u64 base;
246 u32 l, extcfg;
247 u16 vendor, device;
248 int start, size_index, end;
250 raw_pci_ops->read(0, bus, PCI_DEVFN(0, 0), 0, 4, &l);
251 vendor = l & 0xffff;
252 device = (l >> 16) & 0xffff;
254 if (PCI_VENDOR_ID_NVIDIA != vendor || 0x0369 != device)
255 continue;
257 raw_pci_ops->read(0, bus, PCI_DEVFN(0, 0), extcfg_regnum,
258 extcfg_regsize, &extcfg);
260 if (!(extcfg & extcfg_enable_mask))
261 continue;
263 size_index = (extcfg & extcfg_size_mask) >> extcfg_size_shift;
264 base = extcfg & extcfg_base_mask[size_index];
265 /* base could > 4G */
266 base <<= extcfg_base_lshift;
267 start = (extcfg & extcfg_start_mask) >> extcfg_start_shift;
268 end = start + extcfg_sizebus[size_index] - 1;
269 if (pci_mmconfig_add(0, start, end, base) == NULL)
270 continue;
271 mcp55_mmconf_found++;
274 if (!mcp55_mmconf_found)
275 return NULL;
277 return "nVidia MCP55";
280 struct pci_mmcfg_hostbridge_probe {
281 u32 bus;
282 u32 devfn;
283 u32 vendor;
284 u32 device;
285 const char *(*probe)(void);
288 static struct pci_mmcfg_hostbridge_probe pci_mmcfg_probes[] __initdata = {
289 { 0, PCI_DEVFN(0, 0), PCI_VENDOR_ID_INTEL,
290 PCI_DEVICE_ID_INTEL_E7520_MCH, pci_mmcfg_e7520 },
291 { 0, PCI_DEVFN(0, 0), PCI_VENDOR_ID_INTEL,
292 PCI_DEVICE_ID_INTEL_82945G_HB, pci_mmcfg_intel_945 },
293 { 0, PCI_DEVFN(0x18, 0), PCI_VENDOR_ID_AMD,
294 0x1200, pci_mmcfg_amd_fam10h },
295 { 0xff, PCI_DEVFN(0, 0), PCI_VENDOR_ID_AMD,
296 0x1200, pci_mmcfg_amd_fam10h },
297 { 0, PCI_DEVFN(0, 0), PCI_VENDOR_ID_NVIDIA,
298 0x0369, pci_mmcfg_nvidia_mcp55 },
301 static void __init pci_mmcfg_check_end_bus_number(void)
303 struct pci_mmcfg_region *cfg, *cfgx;
305 /* Fixup overlaps */
306 list_for_each_entry(cfg, &pci_mmcfg_list, list) {
307 if (cfg->end_bus < cfg->start_bus)
308 cfg->end_bus = 255;
310 /* Don't access the list head ! */
311 if (cfg->list.next == &pci_mmcfg_list)
312 break;
314 cfgx = list_entry(cfg->list.next, typeof(*cfg), list);
315 if (cfg->end_bus >= cfgx->start_bus)
316 cfg->end_bus = cfgx->start_bus - 1;
320 static int __init pci_mmcfg_check_hostbridge(void)
322 u32 l;
323 u32 bus, devfn;
324 u16 vendor, device;
325 int i;
326 const char *name;
328 if (!raw_pci_ops)
329 return 0;
331 free_all_mmcfg();
333 for (i = 0; i < ARRAY_SIZE(pci_mmcfg_probes); i++) {
334 bus = pci_mmcfg_probes[i].bus;
335 devfn = pci_mmcfg_probes[i].devfn;
336 raw_pci_ops->read(0, bus, devfn, 0, 4, &l);
337 vendor = l & 0xffff;
338 device = (l >> 16) & 0xffff;
340 name = NULL;
341 if (pci_mmcfg_probes[i].vendor == vendor &&
342 pci_mmcfg_probes[i].device == device)
343 name = pci_mmcfg_probes[i].probe();
345 if (name)
346 printk(KERN_INFO PREFIX "%s with MMCONFIG support\n",
347 name);
350 /* some end_bus_number is crazy, fix it */
351 pci_mmcfg_check_end_bus_number();
353 return !list_empty(&pci_mmcfg_list);
356 static void __init pci_mmcfg_insert_resources(void)
358 struct pci_mmcfg_region *cfg;
360 list_for_each_entry(cfg, &pci_mmcfg_list, list)
361 insert_resource(&iomem_resource, &cfg->res);
363 /* Mark that the resources have been inserted. */
364 pci_mmcfg_resources_inserted = 1;
367 static acpi_status __init check_mcfg_resource(struct acpi_resource *res,
368 void *data)
370 struct resource *mcfg_res = data;
371 struct acpi_resource_address64 address;
372 acpi_status status;
374 if (res->type == ACPI_RESOURCE_TYPE_FIXED_MEMORY32) {
375 struct acpi_resource_fixed_memory32 *fixmem32 =
376 &res->data.fixed_memory32;
377 if (!fixmem32)
378 return AE_OK;
379 if ((mcfg_res->start >= fixmem32->address) &&
380 (mcfg_res->end < (fixmem32->address +
381 fixmem32->address_length))) {
382 mcfg_res->flags = 1;
383 return AE_CTRL_TERMINATE;
386 if ((res->type != ACPI_RESOURCE_TYPE_ADDRESS32) &&
387 (res->type != ACPI_RESOURCE_TYPE_ADDRESS64))
388 return AE_OK;
390 status = acpi_resource_to_address64(res, &address);
391 if (ACPI_FAILURE(status) ||
392 (address.address_length <= 0) ||
393 (address.resource_type != ACPI_MEMORY_RANGE))
394 return AE_OK;
396 if ((mcfg_res->start >= address.minimum) &&
397 (mcfg_res->end < (address.minimum + address.address_length))) {
398 mcfg_res->flags = 1;
399 return AE_CTRL_TERMINATE;
401 return AE_OK;
404 static acpi_status __init find_mboard_resource(acpi_handle handle, u32 lvl,
405 void *context, void **rv)
407 struct resource *mcfg_res = context;
409 acpi_walk_resources(handle, METHOD_NAME__CRS,
410 check_mcfg_resource, context);
412 if (mcfg_res->flags)
413 return AE_CTRL_TERMINATE;
415 return AE_OK;
418 static int __init is_acpi_reserved(u64 start, u64 end, unsigned not_used)
420 struct resource mcfg_res;
422 mcfg_res.start = start;
423 mcfg_res.end = end - 1;
424 mcfg_res.flags = 0;
426 acpi_get_devices("PNP0C01", find_mboard_resource, &mcfg_res, NULL);
428 if (!mcfg_res.flags)
429 acpi_get_devices("PNP0C02", find_mboard_resource, &mcfg_res,
430 NULL);
432 return mcfg_res.flags;
435 typedef int (*check_reserved_t)(u64 start, u64 end, unsigned type);
437 static int __init is_mmconf_reserved(check_reserved_t is_reserved,
438 struct pci_mmcfg_region *cfg, int with_e820)
440 u64 addr = cfg->res.start;
441 u64 size = resource_size(&cfg->res);
442 u64 old_size = size;
443 int valid = 0, num_buses;
445 while (!is_reserved(addr, addr + size, E820_RESERVED)) {
446 size >>= 1;
447 if (size < (16UL<<20))
448 break;
451 if (size >= (16UL<<20) || size == old_size) {
452 printk(KERN_INFO PREFIX "MMCONFIG at %pR reserved in %s\n",
453 &cfg->res,
454 with_e820 ? "E820" : "ACPI motherboard resources");
455 valid = 1;
457 if (old_size != size) {
458 /* update end_bus */
459 cfg->end_bus = cfg->start_bus + ((size>>20) - 1);
460 num_buses = cfg->end_bus - cfg->start_bus + 1;
461 cfg->res.end = cfg->res.start +
462 PCI_MMCFG_BUS_OFFSET(num_buses) - 1;
463 snprintf(cfg->name, PCI_MMCFG_RESOURCE_NAME_LEN,
464 "PCI MMCONFIG %04x [bus %02x-%02x]",
465 cfg->segment, cfg->start_bus, cfg->end_bus);
466 printk(KERN_INFO PREFIX
467 "MMCONFIG for %04x [bus%02x-%02x] "
468 "at %pR (base %#lx) (size reduced!)\n",
469 cfg->segment, cfg->start_bus, cfg->end_bus,
470 &cfg->res, (unsigned long) cfg->address);
474 return valid;
477 static void __init pci_mmcfg_reject_broken(int early)
479 struct pci_mmcfg_region *cfg;
481 list_for_each_entry(cfg, &pci_mmcfg_list, list) {
482 int valid = 0;
484 if (!early && !acpi_disabled) {
485 valid = is_mmconf_reserved(is_acpi_reserved, cfg, 0);
487 if (valid)
488 continue;
489 else
490 printk(KERN_ERR FW_BUG PREFIX
491 "MMCONFIG at %pR not reserved in "
492 "ACPI motherboard resources\n",
493 &cfg->res);
496 /* Don't try to do this check unless configuration
497 type 1 is available. how about type 2 ?*/
498 if (raw_pci_ops)
499 valid = is_mmconf_reserved(e820_all_mapped, cfg, 1);
501 if (!valid)
502 goto reject;
505 return;
507 reject:
508 printk(KERN_INFO PREFIX "not using MMCONFIG\n");
509 free_all_mmcfg();
512 static int __initdata known_bridge;
514 static int __init acpi_mcfg_check_entry(struct acpi_table_mcfg *mcfg,
515 struct acpi_mcfg_allocation *cfg)
517 int year;
519 if (cfg->address < 0xFFFFFFFF)
520 return 0;
522 if (!strcmp(mcfg->header.oem_id, "SGI"))
523 return 0;
525 if (mcfg->header.revision >= 1) {
526 if (dmi_get_date(DMI_BIOS_DATE, &year, NULL, NULL) &&
527 year >= 2010)
528 return 0;
531 printk(KERN_ERR PREFIX "MCFG region for %04x [bus %02x-%02x] at %#llx "
532 "is above 4GB, ignored\n", cfg->pci_segment,
533 cfg->start_bus_number, cfg->end_bus_number, cfg->address);
534 return -EINVAL;
537 static int __init pci_parse_mcfg(struct acpi_table_header *header)
539 struct acpi_table_mcfg *mcfg;
540 struct acpi_mcfg_allocation *cfg_table, *cfg;
541 unsigned long i;
542 int entries;
544 if (!header)
545 return -EINVAL;
547 mcfg = (struct acpi_table_mcfg *)header;
549 /* how many config structures do we have */
550 free_all_mmcfg();
551 entries = 0;
552 i = header->length - sizeof(struct acpi_table_mcfg);
553 while (i >= sizeof(struct acpi_mcfg_allocation)) {
554 entries++;
555 i -= sizeof(struct acpi_mcfg_allocation);
557 if (entries == 0) {
558 printk(KERN_ERR PREFIX "MMCONFIG has no entries\n");
559 return -ENODEV;
562 cfg_table = (struct acpi_mcfg_allocation *) &mcfg[1];
563 for (i = 0; i < entries; i++) {
564 cfg = &cfg_table[i];
565 if (acpi_mcfg_check_entry(mcfg, cfg)) {
566 free_all_mmcfg();
567 return -ENODEV;
570 if (pci_mmconfig_add(cfg->pci_segment, cfg->start_bus_number,
571 cfg->end_bus_number, cfg->address) == NULL) {
572 printk(KERN_WARNING PREFIX
573 "no memory for MCFG entries\n");
574 free_all_mmcfg();
575 return -ENOMEM;
579 return 0;
582 static void __init __pci_mmcfg_init(int early)
584 /* MMCONFIG disabled */
585 if ((pci_probe & PCI_PROBE_MMCONF) == 0)
586 return;
588 /* MMCONFIG already enabled */
589 if (!early && !(pci_probe & PCI_PROBE_MASK & ~PCI_PROBE_MMCONF))
590 return;
592 /* for late to exit */
593 if (known_bridge)
594 return;
596 if (early) {
597 if (pci_mmcfg_check_hostbridge())
598 known_bridge = 1;
601 if (!known_bridge)
602 acpi_sfi_table_parse(ACPI_SIG_MCFG, pci_parse_mcfg);
604 pci_mmcfg_reject_broken(early);
606 if (list_empty(&pci_mmcfg_list))
607 return;
609 if (pcibios_last_bus < 0) {
610 const struct pci_mmcfg_region *cfg;
612 list_for_each_entry(cfg, &pci_mmcfg_list, list) {
613 if (cfg->segment)
614 break;
615 pcibios_last_bus = cfg->end_bus;
619 if (pci_mmcfg_arch_init())
620 pci_probe = (pci_probe & ~PCI_PROBE_MASK) | PCI_PROBE_MMCONF;
621 else {
623 * Signal not to attempt to insert mmcfg resources because
624 * the architecture mmcfg setup could not initialize.
626 pci_mmcfg_resources_inserted = 1;
630 void __init pci_mmcfg_early_init(void)
632 __pci_mmcfg_init(1);
635 void __init pci_mmcfg_late_init(void)
637 __pci_mmcfg_init(0);
640 static int __init pci_mmcfg_late_insert_resources(void)
643 * If resources are already inserted or we are not using MMCONFIG,
644 * don't insert the resources.
646 if ((pci_mmcfg_resources_inserted == 1) ||
647 (pci_probe & PCI_PROBE_MMCONF) == 0 ||
648 list_empty(&pci_mmcfg_list))
649 return 1;
652 * Attempt to insert the mmcfg resources but not with the busy flag
653 * marked so it won't cause request errors when __request_region is
654 * called.
656 pci_mmcfg_insert_resources();
658 return 0;
662 * Perform MMCONFIG resource insertion after PCI initialization to allow for
663 * misprogrammed MCFG tables that state larger sizes but actually conflict
664 * with other system resources.
666 late_initcall(pci_mmcfg_late_insert_resources);