x86/PCI: use RCU list to protect mmconfig list
[deliverable/linux.git] / arch / x86 / pci / mmconfig-shared.c
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 <linux/slab.h>
20 #include <linux/mutex.h>
21 #include <linux/rculist.h>
22 #include <asm/e820.h>
23 #include <asm/pci_x86.h>
24 #include <asm/acpi.h>
25
26 #define PREFIX "PCI: "
27
28 /* Indicate if the mmcfg resources have been placed into the resource table. */
29 static int __initdata pci_mmcfg_resources_inserted;
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 __devinit 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 __devinit struct pci_mmcfg_region *pci_mmconfig_alloc(int segment,
68 int start,
69 int end, u64 addr)
70 {
71 struct pci_mmcfg_region *new;
72 struct resource *res;
73
74 if (addr == 0)
75 return NULL;
76
77 new = kzalloc(sizeof(*new), GFP_KERNEL);
78 if (!new)
79 return NULL;
80
81 new->address = addr;
82 new->segment = segment;
83 new->start_bus = start;
84 new->end_bus = end;
85
86 res = &new->res;
87 res->start = addr + PCI_MMCFG_BUS_OFFSET(start);
88 res->end = addr + PCI_MMCFG_BUS_OFFSET(end + 1) - 1;
89 res->flags = IORESOURCE_MEM | IORESOURCE_BUSY;
90 snprintf(new->name, PCI_MMCFG_RESOURCE_NAME_LEN,
91 "PCI MMCONFIG %04x [bus %02x-%02x]", segment, start, end);
92 res->name = new->name;
93
94 printk(KERN_INFO PREFIX "MMCONFIG for domain %04x [bus %02x-%02x] at "
95 "%pR (base %#lx)\n", segment, start, end, &new->res,
96 (unsigned long) addr);
97
98 return new;
99 }
100
101 static __init struct pci_mmcfg_region *pci_mmconfig_add(int segment, int start,
102 int end, u64 addr)
103 {
104 struct pci_mmcfg_region *new;
105
106 new = pci_mmconfig_alloc(segment, start, end, addr);
107 if (new) {
108 mutex_lock(&pci_mmcfg_lock);
109 list_add_sorted(new);
110 mutex_unlock(&pci_mmcfg_lock);
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 printk(KERN_INFO PREFIX "%s with MMCONFIG support\n",
364 name);
365 }
366
367 /* some end_bus_number is crazy, fix it */
368 pci_mmcfg_check_end_bus_number();
369
370 return !list_empty(&pci_mmcfg_list);
371 }
372
373 static void __init pci_mmcfg_insert_resources(void)
374 {
375 struct pci_mmcfg_region *cfg;
376
377 list_for_each_entry(cfg, &pci_mmcfg_list, list)
378 insert_resource(&iomem_resource, &cfg->res);
379
380 /* Mark that the resources have been inserted. */
381 pci_mmcfg_resources_inserted = 1;
382 }
383
384 static acpi_status __init check_mcfg_resource(struct acpi_resource *res,
385 void *data)
386 {
387 struct resource *mcfg_res = data;
388 struct acpi_resource_address64 address;
389 acpi_status status;
390
391 if (res->type == ACPI_RESOURCE_TYPE_FIXED_MEMORY32) {
392 struct acpi_resource_fixed_memory32 *fixmem32 =
393 &res->data.fixed_memory32;
394 if (!fixmem32)
395 return AE_OK;
396 if ((mcfg_res->start >= fixmem32->address) &&
397 (mcfg_res->end < (fixmem32->address +
398 fixmem32->address_length))) {
399 mcfg_res->flags = 1;
400 return AE_CTRL_TERMINATE;
401 }
402 }
403 if ((res->type != ACPI_RESOURCE_TYPE_ADDRESS32) &&
404 (res->type != ACPI_RESOURCE_TYPE_ADDRESS64))
405 return AE_OK;
406
407 status = acpi_resource_to_address64(res, &address);
408 if (ACPI_FAILURE(status) ||
409 (address.address_length <= 0) ||
410 (address.resource_type != ACPI_MEMORY_RANGE))
411 return AE_OK;
412
413 if ((mcfg_res->start >= address.minimum) &&
414 (mcfg_res->end < (address.minimum + address.address_length))) {
415 mcfg_res->flags = 1;
416 return AE_CTRL_TERMINATE;
417 }
418 return AE_OK;
419 }
420
421 static acpi_status __init find_mboard_resource(acpi_handle handle, u32 lvl,
422 void *context, void **rv)
423 {
424 struct resource *mcfg_res = context;
425
426 acpi_walk_resources(handle, METHOD_NAME__CRS,
427 check_mcfg_resource, context);
428
429 if (mcfg_res->flags)
430 return AE_CTRL_TERMINATE;
431
432 return AE_OK;
433 }
434
435 static int __init is_acpi_reserved(u64 start, u64 end, unsigned not_used)
436 {
437 struct resource mcfg_res;
438
439 mcfg_res.start = start;
440 mcfg_res.end = end - 1;
441 mcfg_res.flags = 0;
442
443 acpi_get_devices("PNP0C01", find_mboard_resource, &mcfg_res, NULL);
444
445 if (!mcfg_res.flags)
446 acpi_get_devices("PNP0C02", find_mboard_resource, &mcfg_res,
447 NULL);
448
449 return mcfg_res.flags;
450 }
451
452 typedef int (*check_reserved_t)(u64 start, u64 end, unsigned type);
453
454 static int __init is_mmconf_reserved(check_reserved_t is_reserved,
455 struct pci_mmcfg_region *cfg, int with_e820)
456 {
457 u64 addr = cfg->res.start;
458 u64 size = resource_size(&cfg->res);
459 u64 old_size = size;
460 int valid = 0, num_buses;
461
462 while (!is_reserved(addr, addr + size, E820_RESERVED)) {
463 size >>= 1;
464 if (size < (16UL<<20))
465 break;
466 }
467
468 if (size >= (16UL<<20) || size == old_size) {
469 printk(KERN_INFO PREFIX "MMCONFIG at %pR reserved in %s\n",
470 &cfg->res,
471 with_e820 ? "E820" : "ACPI motherboard resources");
472 valid = 1;
473
474 if (old_size != size) {
475 /* update end_bus */
476 cfg->end_bus = cfg->start_bus + ((size>>20) - 1);
477 num_buses = cfg->end_bus - cfg->start_bus + 1;
478 cfg->res.end = cfg->res.start +
479 PCI_MMCFG_BUS_OFFSET(num_buses) - 1;
480 snprintf(cfg->name, PCI_MMCFG_RESOURCE_NAME_LEN,
481 "PCI MMCONFIG %04x [bus %02x-%02x]",
482 cfg->segment, cfg->start_bus, cfg->end_bus);
483 printk(KERN_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
491 return valid;
492 }
493
494 static int __devinit pci_mmcfg_check_reserved(struct pci_mmcfg_region *cfg,
495 int early)
496 {
497 if (!early && !acpi_disabled) {
498 if (is_mmconf_reserved(is_acpi_reserved, cfg, 0))
499 return 1;
500 else
501 printk(KERN_ERR FW_BUG PREFIX
502 "MMCONFIG at %pR not reserved in "
503 "ACPI motherboard resources\n",
504 &cfg->res);
505 }
506
507 /* Don't try to do this check unless configuration
508 type 1 is available. how about type 2 ?*/
509 if (raw_pci_ops)
510 return is_mmconf_reserved(e820_all_mapped, cfg, 1);
511
512 return 0;
513 }
514
515 static void __init pci_mmcfg_reject_broken(int early)
516 {
517 struct pci_mmcfg_region *cfg;
518
519 list_for_each_entry(cfg, &pci_mmcfg_list, list) {
520 if (pci_mmcfg_check_reserved(cfg, early) == 0) {
521 printk(KERN_INFO PREFIX "not using MMCONFIG\n");
522 free_all_mmcfg();
523 return;
524 }
525 }
526 }
527
528 static int __initdata known_bridge;
529
530 static int __init acpi_mcfg_check_entry(struct acpi_table_mcfg *mcfg,
531 struct acpi_mcfg_allocation *cfg)
532 {
533 int year;
534
535 if (cfg->address < 0xFFFFFFFF)
536 return 0;
537
538 if (!strcmp(mcfg->header.oem_id, "SGI") ||
539 !strcmp(mcfg->header.oem_id, "SGI2"))
540 return 0;
541
542 if (mcfg->header.revision >= 1) {
543 if (dmi_get_date(DMI_BIOS_DATE, &year, NULL, NULL) &&
544 year >= 2010)
545 return 0;
546 }
547
548 printk(KERN_ERR PREFIX "MCFG region for %04x [bus %02x-%02x] at %#llx "
549 "is above 4GB, ignored\n", cfg->pci_segment,
550 cfg->start_bus_number, cfg->end_bus_number, cfg->address);
551 return -EINVAL;
552 }
553
554 static int __init pci_parse_mcfg(struct acpi_table_header *header)
555 {
556 struct acpi_table_mcfg *mcfg;
557 struct acpi_mcfg_allocation *cfg_table, *cfg;
558 unsigned long i;
559 int entries;
560
561 if (!header)
562 return -EINVAL;
563
564 mcfg = (struct acpi_table_mcfg *)header;
565
566 /* how many config structures do we have */
567 free_all_mmcfg();
568 entries = 0;
569 i = header->length - sizeof(struct acpi_table_mcfg);
570 while (i >= sizeof(struct acpi_mcfg_allocation)) {
571 entries++;
572 i -= sizeof(struct acpi_mcfg_allocation);
573 };
574 if (entries == 0) {
575 printk(KERN_ERR PREFIX "MMCONFIG has no entries\n");
576 return -ENODEV;
577 }
578
579 cfg_table = (struct acpi_mcfg_allocation *) &mcfg[1];
580 for (i = 0; i < entries; i++) {
581 cfg = &cfg_table[i];
582 if (acpi_mcfg_check_entry(mcfg, cfg)) {
583 free_all_mmcfg();
584 return -ENODEV;
585 }
586
587 if (pci_mmconfig_add(cfg->pci_segment, cfg->start_bus_number,
588 cfg->end_bus_number, cfg->address) == NULL) {
589 printk(KERN_WARNING PREFIX
590 "no memory for MCFG entries\n");
591 free_all_mmcfg();
592 return -ENOMEM;
593 }
594 }
595
596 return 0;
597 }
598
599 static void __init __pci_mmcfg_init(int early)
600 {
601 /* MMCONFIG disabled */
602 if ((pci_probe & PCI_PROBE_MMCONF) == 0)
603 return;
604
605 /* MMCONFIG already enabled */
606 if (!early && !(pci_probe & PCI_PROBE_MASK & ~PCI_PROBE_MMCONF))
607 return;
608
609 /* for late to exit */
610 if (known_bridge)
611 return;
612
613 if (early) {
614 if (pci_mmcfg_check_hostbridge())
615 known_bridge = 1;
616 }
617
618 if (!known_bridge)
619 acpi_sfi_table_parse(ACPI_SIG_MCFG, pci_parse_mcfg);
620
621 pci_mmcfg_reject_broken(early);
622
623 if (list_empty(&pci_mmcfg_list))
624 return;
625
626 if (pcibios_last_bus < 0) {
627 const struct pci_mmcfg_region *cfg;
628
629 list_for_each_entry(cfg, &pci_mmcfg_list, list) {
630 if (cfg->segment)
631 break;
632 pcibios_last_bus = cfg->end_bus;
633 }
634 }
635
636 if (pci_mmcfg_arch_init())
637 pci_probe = (pci_probe & ~PCI_PROBE_MASK) | PCI_PROBE_MMCONF;
638 else {
639 /*
640 * Signal not to attempt to insert mmcfg resources because
641 * the architecture mmcfg setup could not initialize.
642 */
643 pci_mmcfg_resources_inserted = 1;
644 }
645 }
646
647 void __init pci_mmcfg_early_init(void)
648 {
649 __pci_mmcfg_init(1);
650 }
651
652 void __init pci_mmcfg_late_init(void)
653 {
654 __pci_mmcfg_init(0);
655 }
656
657 static int __init pci_mmcfg_late_insert_resources(void)
658 {
659 /*
660 * If resources are already inserted or we are not using MMCONFIG,
661 * don't insert the resources.
662 */
663 if ((pci_mmcfg_resources_inserted == 1) ||
664 (pci_probe & PCI_PROBE_MMCONF) == 0 ||
665 list_empty(&pci_mmcfg_list))
666 return 1;
667
668 /*
669 * Attempt to insert the mmcfg resources but not with the busy flag
670 * marked so it won't cause request errors when __request_region is
671 * called.
672 */
673 pci_mmcfg_insert_resources();
674
675 return 0;
676 }
677
678 /*
679 * Perform MMCONFIG resource insertion after PCI initialization to allow for
680 * misprogrammed MCFG tables that state larger sizes but actually conflict
681 * with other system resources.
682 */
683 late_initcall(pci_mmcfg_late_insert_resources);
This page took 0.04437 seconds and 5 git commands to generate.