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 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 struct pci_mmcfg_region
*pci_mmconfig_alloc(int segment
, int start
,
71 struct pci_mmcfg_region
*new;
77 new = kzalloc(sizeof(*new), GFP_KERNEL
);
82 new->segment
= segment
;
83 new->start_bus
= start
;
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
;
97 static __init
struct pci_mmcfg_region
*pci_mmconfig_add(int segment
, int start
,
100 struct pci_mmcfg_region
*new;
102 new = pci_mmconfig_alloc(segment
, start
, end
, addr
);
104 mutex_lock(&pci_mmcfg_lock
);
105 list_add_sorted(new);
106 mutex_unlock(&pci_mmcfg_lock
);
109 "MMCONFIG for domain %04x [bus %02x-%02x] at %pR "
111 segment
, start
, end
, &new->res
, (unsigned long)addr
);
117 struct pci_mmcfg_region
*pci_mmconfig_lookup(int segment
, int bus
)
119 struct pci_mmcfg_region
*cfg
;
121 list_for_each_entry_rcu(cfg
, &pci_mmcfg_list
, list
)
122 if (cfg
->segment
== segment
&&
123 cfg
->start_bus
<= bus
&& bus
<= cfg
->end_bus
)
129 static const char __init
*pci_mmcfg_e7520(void)
132 raw_pci_ops
->read(0, 0, PCI_DEVFN(0, 0), 0xce, 2, &win
);
135 if (win
== 0x0000 || win
== 0xf000)
138 if (pci_mmconfig_add(0, 0, 255, win
<< 16) == NULL
)
141 return "Intel Corporation E7520 Memory Controller Hub";
144 static const char __init
*pci_mmcfg_intel_945(void)
146 u32 pciexbar
, mask
= 0, len
= 0;
148 raw_pci_ops
->read(0, 0, PCI_DEVFN(0, 0), 0x48, 4, &pciexbar
);
155 switch ((pciexbar
>> 1) & 3) {
172 /* Errata #2, things break when not aligned on a 256Mb boundary */
173 /* Can only happen in 64M/128M mode */
175 if ((pciexbar
& mask
) & 0x0fffffffU
)
178 /* Don't hit the APIC registers and their friends */
179 if ((pciexbar
& mask
) >= 0xf0000000U
)
182 if (pci_mmconfig_add(0, 0, (len
>> 20) - 1, pciexbar
& mask
) == NULL
)
185 return "Intel Corporation 945G/GZ/P/PL Express Memory Controller Hub";
188 static const char __init
*pci_mmcfg_amd_fam10h(void)
190 u32 low
, high
, address
;
193 unsigned segnbits
= 0, busnbits
, end_bus
;
195 if (!(pci_probe
& PCI_CHECK_ENABLE_AMD_MMCONF
))
198 address
= MSR_FAM10H_MMIO_CONF_BASE
;
199 if (rdmsr_safe(address
, &low
, &high
))
206 /* mmconfig is not enable */
207 if (!(msr
& FAM10H_MMIO_CONF_ENABLE
))
210 base
= msr
& (FAM10H_MMIO_CONF_BASE_MASK
<<FAM10H_MMIO_CONF_BASE_SHIFT
);
212 busnbits
= (msr
>> FAM10H_MMIO_CONF_BUSRANGE_SHIFT
) &
213 FAM10H_MMIO_CONF_BUSRANGE_MASK
;
216 * only handle bus 0 ?
223 segnbits
= busnbits
- 8;
227 end_bus
= (1 << busnbits
) - 1;
228 for (i
= 0; i
< (1 << segnbits
); i
++)
229 if (pci_mmconfig_add(i
, 0, end_bus
,
230 base
+ (1<<28) * i
) == NULL
) {
235 return "AMD Family 10h NB";
238 static bool __initdata mcp55_checked
;
239 static const char __init
*pci_mmcfg_nvidia_mcp55(void)
242 int mcp55_mmconf_found
= 0;
244 static const u32 extcfg_regnum
= 0x90;
245 static const u32 extcfg_regsize
= 4;
246 static const u32 extcfg_enable_mask
= 1<<31;
247 static const u32 extcfg_start_mask
= 0xff<<16;
248 static const int extcfg_start_shift
= 16;
249 static const u32 extcfg_size_mask
= 0x3<<28;
250 static const int extcfg_size_shift
= 28;
251 static const int extcfg_sizebus
[] = {0x100, 0x80, 0x40, 0x20};
252 static const u32 extcfg_base_mask
[] = {0x7ff8, 0x7ffc, 0x7ffe, 0x7fff};
253 static const int extcfg_base_lshift
= 25;
256 * do check if amd fam10h already took over
258 if (!acpi_disabled
|| !list_empty(&pci_mmcfg_list
) || mcp55_checked
)
261 mcp55_checked
= true;
262 for (bus
= 0; bus
< 256; bus
++) {
266 int start
, size_index
, end
;
268 raw_pci_ops
->read(0, bus
, PCI_DEVFN(0, 0), 0, 4, &l
);
270 device
= (l
>> 16) & 0xffff;
272 if (PCI_VENDOR_ID_NVIDIA
!= vendor
|| 0x0369 != device
)
275 raw_pci_ops
->read(0, bus
, PCI_DEVFN(0, 0), extcfg_regnum
,
276 extcfg_regsize
, &extcfg
);
278 if (!(extcfg
& extcfg_enable_mask
))
281 size_index
= (extcfg
& extcfg_size_mask
) >> extcfg_size_shift
;
282 base
= extcfg
& extcfg_base_mask
[size_index
];
283 /* base could > 4G */
284 base
<<= extcfg_base_lshift
;
285 start
= (extcfg
& extcfg_start_mask
) >> extcfg_start_shift
;
286 end
= start
+ extcfg_sizebus
[size_index
] - 1;
287 if (pci_mmconfig_add(0, start
, end
, base
) == NULL
)
289 mcp55_mmconf_found
++;
292 if (!mcp55_mmconf_found
)
295 return "nVidia MCP55";
298 struct pci_mmcfg_hostbridge_probe
{
303 const char *(*probe
)(void);
306 static struct pci_mmcfg_hostbridge_probe pci_mmcfg_probes
[] __initdata
= {
307 { 0, PCI_DEVFN(0, 0), PCI_VENDOR_ID_INTEL
,
308 PCI_DEVICE_ID_INTEL_E7520_MCH
, pci_mmcfg_e7520
},
309 { 0, PCI_DEVFN(0, 0), PCI_VENDOR_ID_INTEL
,
310 PCI_DEVICE_ID_INTEL_82945G_HB
, pci_mmcfg_intel_945
},
311 { 0, PCI_DEVFN(0x18, 0), PCI_VENDOR_ID_AMD
,
312 0x1200, pci_mmcfg_amd_fam10h
},
313 { 0xff, PCI_DEVFN(0, 0), PCI_VENDOR_ID_AMD
,
314 0x1200, pci_mmcfg_amd_fam10h
},
315 { 0, PCI_DEVFN(0, 0), PCI_VENDOR_ID_NVIDIA
,
316 0x0369, pci_mmcfg_nvidia_mcp55
},
319 static void __init
pci_mmcfg_check_end_bus_number(void)
321 struct pci_mmcfg_region
*cfg
, *cfgx
;
324 list_for_each_entry(cfg
, &pci_mmcfg_list
, list
) {
325 if (cfg
->end_bus
< cfg
->start_bus
)
328 /* Don't access the list head ! */
329 if (cfg
->list
.next
== &pci_mmcfg_list
)
332 cfgx
= list_entry(cfg
->list
.next
, typeof(*cfg
), list
);
333 if (cfg
->end_bus
>= cfgx
->start_bus
)
334 cfg
->end_bus
= cfgx
->start_bus
- 1;
338 static int __init
pci_mmcfg_check_hostbridge(void)
351 for (i
= 0; i
< ARRAY_SIZE(pci_mmcfg_probes
); i
++) {
352 bus
= pci_mmcfg_probes
[i
].bus
;
353 devfn
= pci_mmcfg_probes
[i
].devfn
;
354 raw_pci_ops
->read(0, bus
, devfn
, 0, 4, &l
);
356 device
= (l
>> 16) & 0xffff;
359 if (pci_mmcfg_probes
[i
].vendor
== vendor
&&
360 pci_mmcfg_probes
[i
].device
== device
)
361 name
= pci_mmcfg_probes
[i
].probe();
364 pr_info(PREFIX
"%s with MMCONFIG support\n", name
);
367 /* some end_bus_number is crazy, fix it */
368 pci_mmcfg_check_end_bus_number();
370 return !list_empty(&pci_mmcfg_list
);
373 static acpi_status
check_mcfg_resource(struct acpi_resource
*res
, void *data
)
375 struct resource
*mcfg_res
= data
;
376 struct acpi_resource_address64 address
;
379 if (res
->type
== ACPI_RESOURCE_TYPE_FIXED_MEMORY32
) {
380 struct acpi_resource_fixed_memory32
*fixmem32
=
381 &res
->data
.fixed_memory32
;
384 if ((mcfg_res
->start
>= fixmem32
->address
) &&
385 (mcfg_res
->end
< (fixmem32
->address
+
386 fixmem32
->address_length
))) {
388 return AE_CTRL_TERMINATE
;
391 if ((res
->type
!= ACPI_RESOURCE_TYPE_ADDRESS32
) &&
392 (res
->type
!= ACPI_RESOURCE_TYPE_ADDRESS64
))
395 status
= acpi_resource_to_address64(res
, &address
);
396 if (ACPI_FAILURE(status
) ||
397 (address
.address_length
<= 0) ||
398 (address
.resource_type
!= ACPI_MEMORY_RANGE
))
401 if ((mcfg_res
->start
>= address
.minimum
) &&
402 (mcfg_res
->end
< (address
.minimum
+ address
.address_length
))) {
404 return AE_CTRL_TERMINATE
;
409 static acpi_status
find_mboard_resource(acpi_handle handle
, u32 lvl
,
410 void *context
, void **rv
)
412 struct resource
*mcfg_res
= context
;
414 acpi_walk_resources(handle
, METHOD_NAME__CRS
,
415 check_mcfg_resource
, context
);
418 return AE_CTRL_TERMINATE
;
423 static int is_acpi_reserved(u64 start
, u64 end
, unsigned not_used
)
425 struct resource mcfg_res
;
427 mcfg_res
.start
= start
;
428 mcfg_res
.end
= end
- 1;
431 acpi_get_devices("PNP0C01", find_mboard_resource
, &mcfg_res
, NULL
);
434 acpi_get_devices("PNP0C02", find_mboard_resource
, &mcfg_res
,
437 return mcfg_res
.flags
;
440 typedef int (*check_reserved_t
)(u64 start
, u64 end
, unsigned type
);
442 static int __ref
is_mmconf_reserved(check_reserved_t is_reserved
,
443 struct pci_mmcfg_region
*cfg
,
444 struct device
*dev
, int with_e820
)
446 u64 addr
= cfg
->res
.start
;
447 u64 size
= resource_size(&cfg
->res
);
450 char *method
= with_e820
? "E820" : "ACPI motherboard resources";
452 while (!is_reserved(addr
, addr
+ size
, E820_RESERVED
)) {
454 if (size
< (16UL<<20))
458 if (size
< (16UL<<20) && size
!= old_size
)
462 dev_info(dev
, "MMCONFIG at %pR reserved in %s\n",
465 pr_info(PREFIX
"MMCONFIG at %pR reserved in %s\n",
468 if (old_size
!= size
) {
470 cfg
->end_bus
= cfg
->start_bus
+ ((size
>>20) - 1);
471 num_buses
= cfg
->end_bus
- cfg
->start_bus
+ 1;
472 cfg
->res
.end
= cfg
->res
.start
+
473 PCI_MMCFG_BUS_OFFSET(num_buses
) - 1;
474 snprintf(cfg
->name
, PCI_MMCFG_RESOURCE_NAME_LEN
,
475 "PCI MMCONFIG %04x [bus %02x-%02x]",
476 cfg
->segment
, cfg
->start_bus
, cfg
->end_bus
);
481 "at %pR (base %#lx) (size reduced!)\n",
482 &cfg
->res
, (unsigned long) cfg
->address
);
485 "MMCONFIG for %04x [bus%02x-%02x] "
486 "at %pR (base %#lx) (size reduced!)\n",
487 cfg
->segment
, cfg
->start_bus
, cfg
->end_bus
,
488 &cfg
->res
, (unsigned long) cfg
->address
);
494 static int __ref
pci_mmcfg_check_reserved(struct device
*dev
,
495 struct pci_mmcfg_region
*cfg
, int early
)
497 if (!early
&& !acpi_disabled
) {
498 if (is_mmconf_reserved(is_acpi_reserved
, cfg
, dev
, 0))
502 dev_info(dev
, FW_INFO
503 "MMCONFIG at %pR not reserved in "
504 "ACPI motherboard resources\n",
507 pr_info(FW_INFO PREFIX
508 "MMCONFIG at %pR not reserved in "
509 "ACPI motherboard resources\n",
514 * e820_all_mapped() is marked as __init.
515 * All entries from ACPI MCFG table have been checked at boot time.
516 * For MCFG information constructed from hotpluggable host bridge's
517 * _CBA method, just assume it's reserved.
519 if (pci_mmcfg_running_state
)
522 /* Don't try to do this check unless configuration
523 type 1 is available. how about type 2 ?*/
525 return is_mmconf_reserved(e820_all_mapped
, cfg
, dev
, 1);
530 static void __init
pci_mmcfg_reject_broken(int early
)
532 struct pci_mmcfg_region
*cfg
;
534 list_for_each_entry(cfg
, &pci_mmcfg_list
, list
) {
535 if (pci_mmcfg_check_reserved(NULL
, cfg
, early
) == 0) {
536 pr_info(PREFIX
"not using MMCONFIG\n");
543 static int __init
acpi_mcfg_check_entry(struct acpi_table_mcfg
*mcfg
,
544 struct acpi_mcfg_allocation
*cfg
)
548 if (cfg
->address
< 0xFFFFFFFF)
551 if (!strncmp(mcfg
->header
.oem_id
, "SGI", 3))
554 if (mcfg
->header
.revision
>= 1) {
555 if (dmi_get_date(DMI_BIOS_DATE
, &year
, NULL
, NULL
) &&
560 pr_err(PREFIX
"MCFG region for %04x [bus %02x-%02x] at %#llx "
561 "is above 4GB, ignored\n", cfg
->pci_segment
,
562 cfg
->start_bus_number
, cfg
->end_bus_number
, cfg
->address
);
566 static int __init
pci_parse_mcfg(struct acpi_table_header
*header
)
568 struct acpi_table_mcfg
*mcfg
;
569 struct acpi_mcfg_allocation
*cfg_table
, *cfg
;
576 mcfg
= (struct acpi_table_mcfg
*)header
;
578 /* how many config structures do we have */
581 i
= header
->length
- sizeof(struct acpi_table_mcfg
);
582 while (i
>= sizeof(struct acpi_mcfg_allocation
)) {
584 i
-= sizeof(struct acpi_mcfg_allocation
);
587 pr_err(PREFIX
"MMCONFIG has no entries\n");
591 cfg_table
= (struct acpi_mcfg_allocation
*) &mcfg
[1];
592 for (i
= 0; i
< entries
; i
++) {
594 if (acpi_mcfg_check_entry(mcfg
, cfg
)) {
599 if (pci_mmconfig_add(cfg
->pci_segment
, cfg
->start_bus_number
,
600 cfg
->end_bus_number
, cfg
->address
) == NULL
) {
601 pr_warn(PREFIX
"no memory for MCFG entries\n");
610 static void __init
__pci_mmcfg_init(int early
)
612 pci_mmcfg_reject_broken(early
);
613 if (list_empty(&pci_mmcfg_list
))
616 if (pcibios_last_bus
< 0) {
617 const struct pci_mmcfg_region
*cfg
;
619 list_for_each_entry(cfg
, &pci_mmcfg_list
, list
) {
622 pcibios_last_bus
= cfg
->end_bus
;
626 if (pci_mmcfg_arch_init())
627 pci_probe
= (pci_probe
& ~PCI_PROBE_MASK
) | PCI_PROBE_MMCONF
;
630 pci_mmcfg_arch_init_failed
= true;
634 static int __initdata known_bridge
;
636 void __init
pci_mmcfg_early_init(void)
638 if (pci_probe
& PCI_PROBE_MMCONF
) {
639 if (pci_mmcfg_check_hostbridge())
642 acpi_sfi_table_parse(ACPI_SIG_MCFG
, pci_parse_mcfg
);
647 void __init
pci_mmcfg_late_init(void)
649 /* MMCONFIG disabled */
650 if ((pci_probe
& PCI_PROBE_MMCONF
) == 0)
656 /* MMCONFIG hasn't been enabled yet, try again */
657 if (pci_probe
& PCI_PROBE_MASK
& ~PCI_PROBE_MMCONF
) {
658 acpi_sfi_table_parse(ACPI_SIG_MCFG
, pci_parse_mcfg
);
663 static int __init
pci_mmcfg_late_insert_resources(void)
665 struct pci_mmcfg_region
*cfg
;
667 pci_mmcfg_running_state
= true;
669 /* If we are not using MMCONFIG, don't insert the resources. */
670 if ((pci_probe
& PCI_PROBE_MMCONF
) == 0)
674 * Attempt to insert the mmcfg resources but not with the busy flag
675 * marked so it won't cause request errors when __request_region is
678 list_for_each_entry(cfg
, &pci_mmcfg_list
, list
)
679 if (!cfg
->res
.parent
)
680 insert_resource(&iomem_resource
, &cfg
->res
);
686 * Perform MMCONFIG resource insertion after PCI initialization to allow for
687 * misprogrammed MCFG tables that state larger sizes but actually conflict
688 * with other system resources.
690 late_initcall(pci_mmcfg_late_insert_resources
);
692 /* Add MMCFG information for host bridges */
693 int pci_mmconfig_insert(struct device
*dev
, u16 seg
, u8 start
, u8 end
,
697 struct resource
*tmp
= NULL
;
698 struct pci_mmcfg_region
*cfg
;
700 if (!(pci_probe
& PCI_PROBE_MMCONF
) || pci_mmcfg_arch_init_failed
)
706 mutex_lock(&pci_mmcfg_lock
);
707 cfg
= pci_mmconfig_lookup(seg
, start
);
709 if (cfg
->end_bus
< end
)
710 dev_info(dev
, FW_INFO
712 "domain %04x [bus %02x-%02x] "
713 "only partially covers this bridge\n",
714 cfg
->segment
, cfg
->start_bus
, cfg
->end_bus
);
715 mutex_unlock(&pci_mmcfg_lock
);
720 mutex_unlock(&pci_mmcfg_lock
);
725 cfg
= pci_mmconfig_alloc(seg
, start
, end
, addr
);
727 dev_warn(dev
, "fail to add MMCONFIG (out of memory)\n");
729 } else if (!pci_mmcfg_check_reserved(dev
, cfg
, 0)) {
730 dev_warn(dev
, FW_BUG
"MMCONFIG %pR isn't reserved\n",
733 /* Insert resource if it's not in boot stage */
734 if (pci_mmcfg_running_state
)
735 tmp
= insert_resource_conflict(&iomem_resource
,
740 "MMCONFIG %pR conflicts with "
742 &cfg
->res
, tmp
->name
, tmp
);
743 } else if (pci_mmcfg_arch_map(cfg
)) {
744 dev_warn(dev
, "fail to map MMCONFIG %pR.\n",
747 list_add_sorted(cfg
);
748 dev_info(dev
, "MMCONFIG at %pR (base %#lx)\n",
749 &cfg
->res
, (unsigned long)addr
);
757 release_resource(&cfg
->res
);
761 mutex_unlock(&pci_mmcfg_lock
);
766 /* Delete MMCFG information for host bridges */
767 int pci_mmconfig_delete(u16 seg
, u8 start
, u8 end
)
769 struct pci_mmcfg_region
*cfg
;
771 mutex_lock(&pci_mmcfg_lock
);
772 list_for_each_entry_rcu(cfg
, &pci_mmcfg_list
, list
)
773 if (cfg
->segment
== seg
&& cfg
->start_bus
== start
&&
774 cfg
->end_bus
== end
) {
775 list_del_rcu(&cfg
->list
);
777 pci_mmcfg_arch_unmap(cfg
);
779 release_resource(&cfg
->res
);
780 mutex_unlock(&pci_mmcfg_lock
);
784 mutex_unlock(&pci_mmcfg_lock
);