2 * mmconfig-shared.c - Low-level direct PCI config space access via
3 * MMCONFIG - common code between i386 and x86-64.
6 * - known chipset handling
7 * - ACPI decoding and validation
9 * Per-architecture code takes care of the mappings and accesses
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 <linux/mutex.h>
21 #include <linux/rculist.h>
23 #include <asm/pci_x86.h>
26 #define PREFIX "PCI: "
28 /* Indicate if the mmcfg resources have been placed into the resource table. */
29 static bool pci_mmcfg_running_state;
30 static bool pci_mmcfg_arch_init_failed;
31 static DEFINE_MUTEX(pci_mmcfg_lock);
33 LIST_HEAD(pci_mmcfg_list);
35 static __init void pci_mmconfig_remove(struct pci_mmcfg_region *cfg)
38 release_resource(&cfg->res);
43 static __init void free_all_mmcfg(void)
45 struct pci_mmcfg_region *cfg, *tmp;
47 pci_mmcfg_arch_free();
48 list_for_each_entry_safe(cfg, tmp, &pci_mmcfg_list, list)
49 pci_mmconfig_remove(cfg);
52 static __devinit void list_add_sorted(struct pci_mmcfg_region *new)
54 struct pci_mmcfg_region *cfg;
56 /* keep list sorted by segment and starting bus number */
57 list_for_each_entry_rcu(cfg, &pci_mmcfg_list, list) {
58 if (cfg->segment > new->segment ||
59 (cfg->segment == new->segment &&
60 cfg->start_bus >= new->start_bus)) {
61 list_add_tail_rcu(&new->list, &cfg->list);
65 list_add_tail_rcu(&new->list, &pci_mmcfg_list);
68 static __devinit struct pci_mmcfg_region *pci_mmconfig_alloc(int segment,
72 struct pci_mmcfg_region *new;
78 new = kzalloc(sizeof(*new), GFP_KERNEL);
83 new->segment = segment;
84 new->start_bus = start;
88 res->start = addr + PCI_MMCFG_BUS_OFFSET(start);
89 res->end = addr + PCI_MMCFG_BUS_OFFSET(end + 1) - 1;
90 res->flags = IORESOURCE_MEM | IORESOURCE_BUSY;
91 snprintf(new->name, PCI_MMCFG_RESOURCE_NAME_LEN,
92 "PCI MMCONFIG %04x [bus %02x-%02x]", segment, start, end);
93 res->name = new->name;
98 static __init struct pci_mmcfg_region *pci_mmconfig_add(int segment, int start,
101 struct pci_mmcfg_region *new;
103 new = pci_mmconfig_alloc(segment, start, end, addr);
105 mutex_lock(&pci_mmcfg_lock);
106 list_add_sorted(new);
107 mutex_unlock(&pci_mmcfg_lock);
110 "MMCONFIG for domain %04x [bus %02x-%02x] at %pR "
112 segment, start, end, &new->res, (unsigned long)addr);
118 struct pci_mmcfg_region *pci_mmconfig_lookup(int segment, int bus)
120 struct pci_mmcfg_region *cfg;
122 list_for_each_entry_rcu(cfg, &pci_mmcfg_list, list)
123 if (cfg->segment == segment &&
124 cfg->start_bus <= bus && bus <= cfg->end_bus)
130 static const char __init *pci_mmcfg_e7520(void)
133 raw_pci_ops->read(0, 0, PCI_DEVFN(0, 0), 0xce, 2, &win);
136 if (win == 0x0000 || win == 0xf000)
139 if (pci_mmconfig_add(0, 0, 255, win << 16) == NULL)
142 return "Intel Corporation E7520 Memory Controller Hub";
145 static const char __init *pci_mmcfg_intel_945(void)
147 u32 pciexbar, mask = 0, len = 0;
149 raw_pci_ops->read(0, 0, PCI_DEVFN(0, 0), 0x48, 4, &pciexbar);
156 switch ((pciexbar >> 1) & 3) {
173 /* Errata #2, things break when not aligned on a 256Mb boundary */
174 /* Can only happen in 64M/128M mode */
176 if ((pciexbar & mask) & 0x0fffffffU)
179 /* Don't hit the APIC registers and their friends */
180 if ((pciexbar & mask) >= 0xf0000000U)
183 if (pci_mmconfig_add(0, 0, (len >> 20) - 1, pciexbar & mask) == NULL)
186 return "Intel Corporation 945G/GZ/P/PL Express Memory Controller Hub";
189 static const char __init *pci_mmcfg_amd_fam10h(void)
191 u32 low, high, address;
194 unsigned segnbits = 0, busnbits, end_bus;
196 if (!(pci_probe & PCI_CHECK_ENABLE_AMD_MMCONF))
199 address = MSR_FAM10H_MMIO_CONF_BASE;
200 if (rdmsr_safe(address, &low, &high))
207 /* mmconfig is not enable */
208 if (!(msr & FAM10H_MMIO_CONF_ENABLE))
211 base = msr & (FAM10H_MMIO_CONF_BASE_MASK<<FAM10H_MMIO_CONF_BASE_SHIFT);
213 busnbits = (msr >> FAM10H_MMIO_CONF_BUSRANGE_SHIFT) &
214 FAM10H_MMIO_CONF_BUSRANGE_MASK;
217 * only handle bus 0 ?
224 segnbits = busnbits - 8;
228 end_bus = (1 << busnbits) - 1;
229 for (i = 0; i < (1 << segnbits); i++)
230 if (pci_mmconfig_add(i, 0, end_bus,
231 base + (1<<28) * i) == NULL) {
236 return "AMD Family 10h NB";
239 static bool __initdata mcp55_checked;
240 static const char __init *pci_mmcfg_nvidia_mcp55(void)
243 int mcp55_mmconf_found = 0;
245 static const u32 extcfg_regnum = 0x90;
246 static const u32 extcfg_regsize = 4;
247 static const u32 extcfg_enable_mask = 1<<31;
248 static const u32 extcfg_start_mask = 0xff<<16;
249 static const int extcfg_start_shift = 16;
250 static const u32 extcfg_size_mask = 0x3<<28;
251 static const int extcfg_size_shift = 28;
252 static const int extcfg_sizebus[] = {0x100, 0x80, 0x40, 0x20};
253 static const u32 extcfg_base_mask[] = {0x7ff8, 0x7ffc, 0x7ffe, 0x7fff};
254 static const int extcfg_base_lshift = 25;
257 * do check if amd fam10h already took over
259 if (!acpi_disabled || !list_empty(&pci_mmcfg_list) || mcp55_checked)
262 mcp55_checked = true;
263 for (bus = 0; bus < 256; bus++) {
267 int start, size_index, end;
269 raw_pci_ops->read(0, bus, PCI_DEVFN(0, 0), 0, 4, &l);
271 device = (l >> 16) & 0xffff;
273 if (PCI_VENDOR_ID_NVIDIA != vendor || 0x0369 != device)
276 raw_pci_ops->read(0, bus, PCI_DEVFN(0, 0), extcfg_regnum,
277 extcfg_regsize, &extcfg);
279 if (!(extcfg & extcfg_enable_mask))
282 size_index = (extcfg & extcfg_size_mask) >> extcfg_size_shift;
283 base = extcfg & extcfg_base_mask[size_index];
284 /* base could > 4G */
285 base <<= extcfg_base_lshift;
286 start = (extcfg & extcfg_start_mask) >> extcfg_start_shift;
287 end = start + extcfg_sizebus[size_index] - 1;
288 if (pci_mmconfig_add(0, start, end, base) == NULL)
290 mcp55_mmconf_found++;
293 if (!mcp55_mmconf_found)
296 return "nVidia MCP55";
299 struct pci_mmcfg_hostbridge_probe {
304 const char *(*probe)(void);
307 static struct pci_mmcfg_hostbridge_probe pci_mmcfg_probes[] __initdata = {
308 { 0, PCI_DEVFN(0, 0), PCI_VENDOR_ID_INTEL,
309 PCI_DEVICE_ID_INTEL_E7520_MCH, pci_mmcfg_e7520 },
310 { 0, PCI_DEVFN(0, 0), PCI_VENDOR_ID_INTEL,
311 PCI_DEVICE_ID_INTEL_82945G_HB, pci_mmcfg_intel_945 },
312 { 0, PCI_DEVFN(0x18, 0), PCI_VENDOR_ID_AMD,
313 0x1200, pci_mmcfg_amd_fam10h },
314 { 0xff, PCI_DEVFN(0, 0), PCI_VENDOR_ID_AMD,
315 0x1200, pci_mmcfg_amd_fam10h },
316 { 0, PCI_DEVFN(0, 0), PCI_VENDOR_ID_NVIDIA,
317 0x0369, pci_mmcfg_nvidia_mcp55 },
320 static void __init pci_mmcfg_check_end_bus_number(void)
322 struct pci_mmcfg_region *cfg, *cfgx;
325 list_for_each_entry(cfg, &pci_mmcfg_list, list) {
326 if (cfg->end_bus < cfg->start_bus)
329 /* Don't access the list head ! */
330 if (cfg->list.next == &pci_mmcfg_list)
333 cfgx = list_entry(cfg->list.next, typeof(*cfg), list);
334 if (cfg->end_bus >= cfgx->start_bus)
335 cfg->end_bus = cfgx->start_bus - 1;
339 static int __init pci_mmcfg_check_hostbridge(void)
352 for (i = 0; i < ARRAY_SIZE(pci_mmcfg_probes); i++) {
353 bus = pci_mmcfg_probes[i].bus;
354 devfn = pci_mmcfg_probes[i].devfn;
355 raw_pci_ops->read(0, bus, devfn, 0, 4, &l);
357 device = (l >> 16) & 0xffff;
360 if (pci_mmcfg_probes[i].vendor == vendor &&
361 pci_mmcfg_probes[i].device == device)
362 name = pci_mmcfg_probes[i].probe();
365 pr_info(PREFIX "%s with MMCONFIG support\n", name);
368 /* some end_bus_number is crazy, fix it */
369 pci_mmcfg_check_end_bus_number();
371 return !list_empty(&pci_mmcfg_list);
374 static acpi_status __devinit check_mcfg_resource(struct acpi_resource *res,
377 struct resource *mcfg_res = data;
378 struct acpi_resource_address64 address;
381 if (res->type == ACPI_RESOURCE_TYPE_FIXED_MEMORY32) {
382 struct acpi_resource_fixed_memory32 *fixmem32 =
383 &res->data.fixed_memory32;
386 if ((mcfg_res->start >= fixmem32->address) &&
387 (mcfg_res->end < (fixmem32->address +
388 fixmem32->address_length))) {
390 return AE_CTRL_TERMINATE;
393 if ((res->type != ACPI_RESOURCE_TYPE_ADDRESS32) &&
394 (res->type != ACPI_RESOURCE_TYPE_ADDRESS64))
397 status = acpi_resource_to_address64(res, &address);
398 if (ACPI_FAILURE(status) ||
399 (address.address_length <= 0) ||
400 (address.resource_type != ACPI_MEMORY_RANGE))
403 if ((mcfg_res->start >= address.minimum) &&
404 (mcfg_res->end < (address.minimum + address.address_length))) {
406 return AE_CTRL_TERMINATE;
411 static acpi_status __devinit find_mboard_resource(acpi_handle handle, u32 lvl,
412 void *context, void **rv)
414 struct resource *mcfg_res = context;
416 acpi_walk_resources(handle, METHOD_NAME__CRS,
417 check_mcfg_resource, context);
420 return AE_CTRL_TERMINATE;
425 static int __devinit is_acpi_reserved(u64 start, u64 end, unsigned not_used)
427 struct resource mcfg_res;
429 mcfg_res.start = start;
430 mcfg_res.end = end - 1;
433 acpi_get_devices("PNP0C01", find_mboard_resource, &mcfg_res, NULL);
436 acpi_get_devices("PNP0C02", find_mboard_resource, &mcfg_res,
439 return mcfg_res.flags;
442 typedef int (*check_reserved_t)(u64 start, u64 end, unsigned type);
444 static int __ref is_mmconf_reserved(check_reserved_t is_reserved,
445 struct pci_mmcfg_region *cfg,
446 struct device *dev, int with_e820)
448 u64 addr = cfg->res.start;
449 u64 size = resource_size(&cfg->res);
452 char *method = with_e820 ? "E820" : "ACPI motherboard resources";
454 while (!is_reserved(addr, addr + size, E820_RESERVED)) {
456 if (size < (16UL<<20))
460 if (size < (16UL<<20) && size != old_size)
464 dev_info(dev, "MMCONFIG at %pR reserved in %s\n",
467 pr_info(PREFIX "MMCONFIG at %pR reserved in %s\n",
470 if (old_size != size) {
472 cfg->end_bus = cfg->start_bus + ((size>>20) - 1);
473 num_buses = cfg->end_bus - cfg->start_bus + 1;
474 cfg->res.end = cfg->res.start +
475 PCI_MMCFG_BUS_OFFSET(num_buses) - 1;
476 snprintf(cfg->name, PCI_MMCFG_RESOURCE_NAME_LEN,
477 "PCI MMCONFIG %04x [bus %02x-%02x]",
478 cfg->segment, cfg->start_bus, cfg->end_bus);
483 "at %pR (base %#lx) (size reduced!)\n",
484 &cfg->res, (unsigned long) cfg->address);
487 "MMCONFIG for %04x [bus%02x-%02x] "
488 "at %pR (base %#lx) (size reduced!)\n",
489 cfg->segment, cfg->start_bus, cfg->end_bus,
490 &cfg->res, (unsigned long) cfg->address);
496 static int __ref pci_mmcfg_check_reserved(struct device *dev,
497 struct pci_mmcfg_region *cfg, int early)
499 if (!early && !acpi_disabled) {
500 if (is_mmconf_reserved(is_acpi_reserved, cfg, dev, 0))
504 dev_info(dev, FW_INFO
505 "MMCONFIG at %pR not reserved in "
506 "ACPI motherboard resources\n",
509 pr_info(FW_INFO PREFIX
510 "MMCONFIG at %pR not reserved in "
511 "ACPI motherboard resources\n",
516 * e820_all_mapped() is marked as __init.
517 * All entries from ACPI MCFG table have been checked at boot time.
518 * For MCFG information constructed from hotpluggable host bridge's
519 * _CBA method, just assume it's reserved.
521 if (pci_mmcfg_running_state)
524 /* Don't try to do this check unless configuration
525 type 1 is available. how about type 2 ?*/
527 return is_mmconf_reserved(e820_all_mapped, cfg, dev, 1);
532 static void __init pci_mmcfg_reject_broken(int early)
534 struct pci_mmcfg_region *cfg;
536 list_for_each_entry(cfg, &pci_mmcfg_list, list) {
537 if (pci_mmcfg_check_reserved(NULL, cfg, early) == 0) {
538 pr_info(PREFIX "not using MMCONFIG\n");
545 static int __init acpi_mcfg_check_entry(struct acpi_table_mcfg *mcfg,
546 struct acpi_mcfg_allocation *cfg)
550 if (cfg->address < 0xFFFFFFFF)
553 if (!strcmp(mcfg->header.oem_id, "SGI") ||
554 !strcmp(mcfg->header.oem_id, "SGI2"))
557 if (mcfg->header.revision >= 1) {
558 if (dmi_get_date(DMI_BIOS_DATE, &year, NULL, NULL) &&
563 pr_err(PREFIX "MCFG region for %04x [bus %02x-%02x] at %#llx "
564 "is above 4GB, ignored\n", cfg->pci_segment,
565 cfg->start_bus_number, cfg->end_bus_number, cfg->address);
569 static int __init pci_parse_mcfg(struct acpi_table_header *header)
571 struct acpi_table_mcfg *mcfg;
572 struct acpi_mcfg_allocation *cfg_table, *cfg;
579 mcfg = (struct acpi_table_mcfg *)header;
581 /* how many config structures do we have */
584 i = header->length - sizeof(struct acpi_table_mcfg);
585 while (i >= sizeof(struct acpi_mcfg_allocation)) {
587 i -= sizeof(struct acpi_mcfg_allocation);
590 pr_err(PREFIX "MMCONFIG has no entries\n");
594 cfg_table = (struct acpi_mcfg_allocation *) &mcfg[1];
595 for (i = 0; i < entries; i++) {
597 if (acpi_mcfg_check_entry(mcfg, cfg)) {
602 if (pci_mmconfig_add(cfg->pci_segment, cfg->start_bus_number,
603 cfg->end_bus_number, cfg->address) == NULL) {
604 pr_warn(PREFIX "no memory for MCFG entries\n");
613 static void __init __pci_mmcfg_init(int early)
615 pci_mmcfg_reject_broken(early);
616 if (list_empty(&pci_mmcfg_list))
619 if (pcibios_last_bus < 0) {
620 const struct pci_mmcfg_region *cfg;
622 list_for_each_entry(cfg, &pci_mmcfg_list, list) {
625 pcibios_last_bus = cfg->end_bus;
629 if (pci_mmcfg_arch_init())
630 pci_probe = (pci_probe & ~PCI_PROBE_MASK) | PCI_PROBE_MMCONF;
633 pci_mmcfg_arch_init_failed = true;
637 static int __initdata known_bridge;
639 void __init pci_mmcfg_early_init(void)
641 if (pci_probe & PCI_PROBE_MMCONF) {
642 if (pci_mmcfg_check_hostbridge())
645 acpi_sfi_table_parse(ACPI_SIG_MCFG, pci_parse_mcfg);
650 void __init pci_mmcfg_late_init(void)
652 /* MMCONFIG disabled */
653 if ((pci_probe & PCI_PROBE_MMCONF) == 0)
659 /* MMCONFIG hasn't been enabled yet, try again */
660 if (pci_probe & PCI_PROBE_MASK & ~PCI_PROBE_MMCONF) {
661 acpi_sfi_table_parse(ACPI_SIG_MCFG, pci_parse_mcfg);
666 static int __init pci_mmcfg_late_insert_resources(void)
668 struct pci_mmcfg_region *cfg;
670 pci_mmcfg_running_state = true;
672 /* If we are not using MMCONFIG, don't insert the resources. */
673 if ((pci_probe & PCI_PROBE_MMCONF) == 0)
677 * Attempt to insert the mmcfg resources but not with the busy flag
678 * marked so it won't cause request errors when __request_region is
681 list_for_each_entry(cfg, &pci_mmcfg_list, list)
682 if (!cfg->res.parent)
683 insert_resource(&iomem_resource, &cfg->res);
689 * Perform MMCONFIG resource insertion after PCI initialization to allow for
690 * misprogrammed MCFG tables that state larger sizes but actually conflict
691 * with other system resources.
693 late_initcall(pci_mmcfg_late_insert_resources);
695 /* Add MMCFG information for host bridges */
696 int __devinit pci_mmconfig_insert(struct device *dev,
697 u16 seg, u8 start, u8 end,
701 struct resource *tmp = NULL;
702 struct pci_mmcfg_region *cfg;
704 if (!(pci_probe & PCI_PROBE_MMCONF) || pci_mmcfg_arch_init_failed)
710 mutex_lock(&pci_mmcfg_lock);
711 cfg = pci_mmconfig_lookup(seg, start);
713 if (cfg->end_bus < end)
714 dev_info(dev, FW_INFO
716 "domain %04x [bus %02x-%02x] "
717 "only partially covers this bridge\n",
718 cfg->segment, cfg->start_bus, cfg->end_bus);
719 mutex_unlock(&pci_mmcfg_lock);
724 mutex_unlock(&pci_mmcfg_lock);
729 cfg = pci_mmconfig_alloc(seg, start, end, addr);
731 dev_warn(dev, "fail to add MMCONFIG (out of memory)\n");
733 } else if (!pci_mmcfg_check_reserved(dev, cfg, 0)) {
734 dev_warn(dev, FW_BUG "MMCONFIG %pR isn't reserved\n",
737 /* Insert resource if it's not in boot stage */
738 if (pci_mmcfg_running_state)
739 tmp = insert_resource_conflict(&iomem_resource,
744 "MMCONFIG %pR conflicts with "
746 &cfg->res, tmp->name, tmp);
747 } else if (pci_mmcfg_arch_map(cfg)) {
748 dev_warn(dev, "fail to map MMCONFIG %pR.\n",
751 list_add_sorted(cfg);
752 dev_info(dev, "MMCONFIG at %pR (base %#lx)\n",
753 &cfg->res, (unsigned long)addr);
761 release_resource(&cfg->res);
765 mutex_unlock(&pci_mmcfg_lock);
770 /* Delete MMCFG information for host bridges */
771 int pci_mmconfig_delete(u16 seg, u8 start, u8 end)
773 struct pci_mmcfg_region *cfg;
775 mutex_lock(&pci_mmcfg_lock);
776 list_for_each_entry_rcu(cfg, &pci_mmcfg_list, list)
777 if (cfg->segment == seg && cfg->start_bus == start &&
778 cfg->end_bus == end) {
779 list_del_rcu(&cfg->list);
781 pci_mmcfg_arch_unmap(cfg);
783 release_resource(&cfg->res);
784 mutex_unlock(&pci_mmcfg_lock);
788 mutex_unlock(&pci_mmcfg_lock);