1 // SPDX-License-Identifier: GPL-2.0
3 * mmconfig-shared.c - Low-level direct PCI config space access via
4 * MMCONFIG - common code between i386 and x86-64.
7 * - known chipset handling
8 * - ACPI decoding and validation
10 * Per-architecture code takes care of the mappings and accesses
14 #include <linux/pci.h>
15 #include <linux/init.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/api.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 void __init
pci_mmconfig_remove(struct pci_mmcfg_region
*cfg
)
38 release_resource(&cfg
->res
);
43 static void __init
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 struct pci_mmcfg_region
*__init
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 __initconst
= 0x90;
245 static const u32 extcfg_regsize __initconst
= 4;
246 static const u32 extcfg_enable_mask __initconst
= 1 << 31;
247 static const u32 extcfg_start_mask __initconst
= 0xff << 16;
248 static const int extcfg_start_shift __initconst
= 16;
249 static const u32 extcfg_size_mask __initconst
= 0x3 << 28;
250 static const int extcfg_size_shift __initconst
= 28;
251 static const int extcfg_sizebus
[] __initconst
= {
252 0x100, 0x80, 0x40, 0x20
254 static const u32 extcfg_base_mask
[] __initconst
= {
255 0x7ff8, 0x7ffc, 0x7ffe, 0x7fff
257 static const int extcfg_base_lshift __initconst
= 25;
260 * do check if amd fam10h already took over
262 if (!acpi_disabled
|| !list_empty(&pci_mmcfg_list
) || mcp55_checked
)
265 mcp55_checked
= true;
266 for (bus
= 0; bus
< 256; bus
++) {
270 int start
, size_index
, end
;
272 raw_pci_ops
->read(0, bus
, PCI_DEVFN(0, 0), 0, 4, &l
);
274 device
= (l
>> 16) & 0xffff;
276 if (PCI_VENDOR_ID_NVIDIA
!= vendor
|| 0x0369 != device
)
279 raw_pci_ops
->read(0, bus
, PCI_DEVFN(0, 0), extcfg_regnum
,
280 extcfg_regsize
, &extcfg
);
282 if (!(extcfg
& extcfg_enable_mask
))
285 size_index
= (extcfg
& extcfg_size_mask
) >> extcfg_size_shift
;
286 base
= extcfg
& extcfg_base_mask
[size_index
];
287 /* base could > 4G */
288 base
<<= extcfg_base_lshift
;
289 start
= (extcfg
& extcfg_start_mask
) >> extcfg_start_shift
;
290 end
= start
+ extcfg_sizebus
[size_index
] - 1;
291 if (pci_mmconfig_add(0, start
, end
, base
) == NULL
)
293 mcp55_mmconf_found
++;
296 if (!mcp55_mmconf_found
)
299 return "nVidia MCP55";
302 struct pci_mmcfg_hostbridge_probe
{
307 const char *(*probe
)(void);
310 static const struct pci_mmcfg_hostbridge_probe pci_mmcfg_probes
[] __initconst
= {
311 { 0, PCI_DEVFN(0, 0), PCI_VENDOR_ID_INTEL
,
312 PCI_DEVICE_ID_INTEL_E7520_MCH
, pci_mmcfg_e7520
},
313 { 0, PCI_DEVFN(0, 0), PCI_VENDOR_ID_INTEL
,
314 PCI_DEVICE_ID_INTEL_82945G_HB
, pci_mmcfg_intel_945
},
315 { 0, PCI_DEVFN(0x18, 0), PCI_VENDOR_ID_AMD
,
316 0x1200, pci_mmcfg_amd_fam10h
},
317 { 0xff, PCI_DEVFN(0, 0), PCI_VENDOR_ID_AMD
,
318 0x1200, pci_mmcfg_amd_fam10h
},
319 { 0, PCI_DEVFN(0, 0), PCI_VENDOR_ID_NVIDIA
,
320 0x0369, pci_mmcfg_nvidia_mcp55
},
323 static void __init
pci_mmcfg_check_end_bus_number(void)
325 struct pci_mmcfg_region
*cfg
, *cfgx
;
328 list_for_each_entry(cfg
, &pci_mmcfg_list
, list
) {
329 if (cfg
->end_bus
< cfg
->start_bus
)
332 /* Don't access the list head ! */
333 if (cfg
->list
.next
== &pci_mmcfg_list
)
336 cfgx
= list_entry(cfg
->list
.next
, typeof(*cfg
), list
);
337 if (cfg
->end_bus
>= cfgx
->start_bus
)
338 cfg
->end_bus
= cfgx
->start_bus
- 1;
342 static int __init
pci_mmcfg_check_hostbridge(void)
355 for (i
= 0; i
< ARRAY_SIZE(pci_mmcfg_probes
); i
++) {
356 bus
= pci_mmcfg_probes
[i
].bus
;
357 devfn
= pci_mmcfg_probes
[i
].devfn
;
358 raw_pci_ops
->read(0, bus
, devfn
, 0, 4, &l
);
360 device
= (l
>> 16) & 0xffff;
363 if (pci_mmcfg_probes
[i
].vendor
== vendor
&&
364 pci_mmcfg_probes
[i
].device
== device
)
365 name
= pci_mmcfg_probes
[i
].probe();
368 pr_info(PREFIX
"%s with MMCONFIG support\n", name
);
371 /* some end_bus_number is crazy, fix it */
372 pci_mmcfg_check_end_bus_number();
374 return !list_empty(&pci_mmcfg_list
);
377 static acpi_status
check_mcfg_resource(struct acpi_resource
*res
, void *data
)
379 struct resource
*mcfg_res
= data
;
380 struct acpi_resource_address64 address
;
383 if (res
->type
== ACPI_RESOURCE_TYPE_FIXED_MEMORY32
) {
384 struct acpi_resource_fixed_memory32
*fixmem32
=
385 &res
->data
.fixed_memory32
;
388 if ((mcfg_res
->start
>= fixmem32
->address
) &&
389 (mcfg_res
->end
< (fixmem32
->address
+
390 fixmem32
->address_length
))) {
392 return AE_CTRL_TERMINATE
;
395 if ((res
->type
!= ACPI_RESOURCE_TYPE_ADDRESS32
) &&
396 (res
->type
!= ACPI_RESOURCE_TYPE_ADDRESS64
))
399 status
= acpi_resource_to_address64(res
, &address
);
400 if (ACPI_FAILURE(status
) ||
401 (address
.address
.address_length
<= 0) ||
402 (address
.resource_type
!= ACPI_MEMORY_RANGE
))
405 if ((mcfg_res
->start
>= address
.address
.minimum
) &&
406 (mcfg_res
->end
< (address
.address
.minimum
+ address
.address
.address_length
))) {
408 return AE_CTRL_TERMINATE
;
413 static acpi_status
find_mboard_resource(acpi_handle handle
, u32 lvl
,
414 void *context
, void **rv
)
416 struct resource
*mcfg_res
= context
;
418 acpi_walk_resources(handle
, METHOD_NAME__CRS
,
419 check_mcfg_resource
, context
);
422 return AE_CTRL_TERMINATE
;
427 static bool is_acpi_reserved(u64 start
, u64 end
, unsigned not_used
)
429 struct resource mcfg_res
;
431 mcfg_res
.start
= start
;
432 mcfg_res
.end
= end
- 1;
435 acpi_get_devices("PNP0C01", find_mboard_resource
, &mcfg_res
, NULL
);
438 acpi_get_devices("PNP0C02", find_mboard_resource
, &mcfg_res
,
441 return mcfg_res
.flags
;
444 typedef bool (*check_reserved_t
)(u64 start
, u64 end
, unsigned type
);
446 static bool __ref
is_mmconf_reserved(check_reserved_t is_reserved
,
447 struct pci_mmcfg_region
*cfg
,
448 struct device
*dev
, int with_e820
)
450 u64 addr
= cfg
->res
.start
;
451 u64 size
= resource_size(&cfg
->res
);
454 char *method
= with_e820
? "E820" : "ACPI motherboard resources";
456 while (!is_reserved(addr
, addr
+ size
, E820_TYPE_RESERVED
)) {
458 if (size
< (16UL<<20))
462 if (size
< (16UL<<20) && size
!= old_size
)
466 dev_info(dev
, "MMCONFIG at %pR reserved in %s\n",
469 pr_info(PREFIX
"MMCONFIG at %pR reserved in %s\n",
472 if (old_size
!= size
) {
474 cfg
->end_bus
= cfg
->start_bus
+ ((size
>>20) - 1);
475 num_buses
= cfg
->end_bus
- cfg
->start_bus
+ 1;
476 cfg
->res
.end
= cfg
->res
.start
+
477 PCI_MMCFG_BUS_OFFSET(num_buses
) - 1;
478 snprintf(cfg
->name
, PCI_MMCFG_RESOURCE_NAME_LEN
,
479 "PCI MMCONFIG %04x [bus %02x-%02x]",
480 cfg
->segment
, cfg
->start_bus
, cfg
->end_bus
);
485 "at %pR (base %#lx) (size reduced!)\n",
486 &cfg
->res
, (unsigned long) cfg
->address
);
489 "MMCONFIG for %04x [bus%02x-%02x] "
490 "at %pR (base %#lx) (size reduced!)\n",
491 cfg
->segment
, cfg
->start_bus
, cfg
->end_bus
,
492 &cfg
->res
, (unsigned long) cfg
->address
);
499 pci_mmcfg_check_reserved(struct device
*dev
, struct pci_mmcfg_region
*cfg
, int early
)
501 if (!early
&& !acpi_disabled
) {
502 if (is_mmconf_reserved(is_acpi_reserved
, cfg
, dev
, 0))
506 dev_info(dev
, FW_INFO
507 "MMCONFIG at %pR not reserved in "
508 "ACPI motherboard resources\n",
511 pr_info(FW_INFO PREFIX
512 "MMCONFIG at %pR not reserved in "
513 "ACPI motherboard resources\n",
518 * e820__mapped_all() is marked as __init.
519 * All entries from ACPI MCFG table have been checked at boot time.
520 * For MCFG information constructed from hotpluggable host bridge's
521 * _CBA method, just assume it's reserved.
523 if (pci_mmcfg_running_state
)
526 /* Don't try to do this check unless configuration
527 type 1 is available. how about type 2 ?*/
529 return is_mmconf_reserved(e820__mapped_all
, cfg
, dev
, 1);
534 static void __init
pci_mmcfg_reject_broken(int early
)
536 struct pci_mmcfg_region
*cfg
;
538 list_for_each_entry(cfg
, &pci_mmcfg_list
, list
) {
539 if (pci_mmcfg_check_reserved(NULL
, cfg
, early
) == 0) {
540 pr_info(PREFIX
"not using MMCONFIG\n");
547 static int __init
acpi_mcfg_check_entry(struct acpi_table_mcfg
*mcfg
,
548 struct acpi_mcfg_allocation
*cfg
)
550 if (cfg
->address
< 0xFFFFFFFF)
553 if (!strncmp(mcfg
->header
.oem_id
, "SGI", 3))
556 if ((mcfg
->header
.revision
>= 1) && (dmi_get_bios_year() >= 2010))
559 pr_err(PREFIX
"MCFG region for %04x [bus %02x-%02x] at %#llx "
560 "is above 4GB, ignored\n", cfg
->pci_segment
,
561 cfg
->start_bus_number
, cfg
->end_bus_number
, cfg
->address
);
565 static int __init
pci_parse_mcfg(struct acpi_table_header
*header
)
567 struct acpi_table_mcfg
*mcfg
;
568 struct acpi_mcfg_allocation
*cfg_table
, *cfg
;
575 mcfg
= (struct acpi_table_mcfg
*)header
;
577 /* how many config structures do we have */
580 i
= header
->length
- sizeof(struct acpi_table_mcfg
);
581 while (i
>= sizeof(struct acpi_mcfg_allocation
)) {
583 i
-= sizeof(struct acpi_mcfg_allocation
);
586 pr_err(PREFIX
"MMCONFIG has no entries\n");
590 cfg_table
= (struct acpi_mcfg_allocation
*) &mcfg
[1];
591 for (i
= 0; i
< entries
; i
++) {
593 if (acpi_mcfg_check_entry(mcfg
, cfg
)) {
598 if (pci_mmconfig_add(cfg
->pci_segment
, cfg
->start_bus_number
,
599 cfg
->end_bus_number
, cfg
->address
) == NULL
) {
600 pr_warn(PREFIX
"no memory for MCFG entries\n");
609 #ifdef CONFIG_ACPI_APEI
610 extern int (*arch_apei_filter_addr
)(int (*func
)(__u64 start
, __u64 size
,
611 void *data
), void *data
);
613 static int pci_mmcfg_for_each_region(int (*func
)(__u64 start
, __u64 size
,
614 void *data
), void *data
)
616 struct pci_mmcfg_region
*cfg
;
619 if (list_empty(&pci_mmcfg_list
))
622 list_for_each_entry(cfg
, &pci_mmcfg_list
, list
) {
623 rc
= func(cfg
->res
.start
, resource_size(&cfg
->res
), data
);
630 #define set_apei_filter() (arch_apei_filter_addr = pci_mmcfg_for_each_region)
632 #define set_apei_filter()
635 static void __init
__pci_mmcfg_init(int early
)
637 pci_mmcfg_reject_broken(early
);
638 if (list_empty(&pci_mmcfg_list
))
641 if (pcibios_last_bus
< 0) {
642 const struct pci_mmcfg_region
*cfg
;
644 list_for_each_entry(cfg
, &pci_mmcfg_list
, list
) {
647 pcibios_last_bus
= cfg
->end_bus
;
651 if (pci_mmcfg_arch_init())
652 pci_probe
= (pci_probe
& ~PCI_PROBE_MASK
) | PCI_PROBE_MMCONF
;
655 pci_mmcfg_arch_init_failed
= true;
659 static int __initdata known_bridge
;
661 void __init
pci_mmcfg_early_init(void)
663 if (pci_probe
& PCI_PROBE_MMCONF
) {
664 if (pci_mmcfg_check_hostbridge())
667 acpi_sfi_table_parse(ACPI_SIG_MCFG
, pci_parse_mcfg
);
674 void __init
pci_mmcfg_late_init(void)
676 /* MMCONFIG disabled */
677 if ((pci_probe
& PCI_PROBE_MMCONF
) == 0)
683 /* MMCONFIG hasn't been enabled yet, try again */
684 if (pci_probe
& PCI_PROBE_MASK
& ~PCI_PROBE_MMCONF
) {
685 acpi_sfi_table_parse(ACPI_SIG_MCFG
, pci_parse_mcfg
);
690 static int __init
pci_mmcfg_late_insert_resources(void)
692 struct pci_mmcfg_region
*cfg
;
694 pci_mmcfg_running_state
= true;
696 /* If we are not using MMCONFIG, don't insert the resources. */
697 if ((pci_probe
& PCI_PROBE_MMCONF
) == 0)
701 * Attempt to insert the mmcfg resources but not with the busy flag
702 * marked so it won't cause request errors when __request_region is
705 list_for_each_entry(cfg
, &pci_mmcfg_list
, list
)
706 if (!cfg
->res
.parent
)
707 insert_resource(&iomem_resource
, &cfg
->res
);
713 * Perform MMCONFIG resource insertion after PCI initialization to allow for
714 * misprogrammed MCFG tables that state larger sizes but actually conflict
715 * with other system resources.
717 late_initcall(pci_mmcfg_late_insert_resources
);
719 /* Add MMCFG information for host bridges */
720 int pci_mmconfig_insert(struct device
*dev
, u16 seg
, u8 start
, u8 end
,
724 struct resource
*tmp
= NULL
;
725 struct pci_mmcfg_region
*cfg
;
727 if (!(pci_probe
& PCI_PROBE_MMCONF
) || pci_mmcfg_arch_init_failed
)
733 mutex_lock(&pci_mmcfg_lock
);
734 cfg
= pci_mmconfig_lookup(seg
, start
);
736 if (cfg
->end_bus
< end
)
737 dev_info(dev
, FW_INFO
739 "domain %04x [bus %02x-%02x] "
740 "only partially covers this bridge\n",
741 cfg
->segment
, cfg
->start_bus
, cfg
->end_bus
);
742 mutex_unlock(&pci_mmcfg_lock
);
747 mutex_unlock(&pci_mmcfg_lock
);
752 cfg
= pci_mmconfig_alloc(seg
, start
, end
, addr
);
754 dev_warn(dev
, "fail to add MMCONFIG (out of memory)\n");
756 } else if (!pci_mmcfg_check_reserved(dev
, cfg
, 0)) {
757 dev_warn(dev
, FW_BUG
"MMCONFIG %pR isn't reserved\n",
760 /* Insert resource if it's not in boot stage */
761 if (pci_mmcfg_running_state
)
762 tmp
= insert_resource_conflict(&iomem_resource
,
767 "MMCONFIG %pR conflicts with "
769 &cfg
->res
, tmp
->name
, tmp
);
770 } else if (pci_mmcfg_arch_map(cfg
)) {
771 dev_warn(dev
, "fail to map MMCONFIG %pR.\n",
774 list_add_sorted(cfg
);
775 dev_info(dev
, "MMCONFIG at %pR (base %#lx)\n",
776 &cfg
->res
, (unsigned long)addr
);
784 release_resource(&cfg
->res
);
788 mutex_unlock(&pci_mmcfg_lock
);
793 /* Delete MMCFG information for host bridges */
794 int pci_mmconfig_delete(u16 seg
, u8 start
, u8 end
)
796 struct pci_mmcfg_region
*cfg
;
798 mutex_lock(&pci_mmcfg_lock
);
799 list_for_each_entry_rcu(cfg
, &pci_mmcfg_list
, list
)
800 if (cfg
->segment
== seg
&& cfg
->start_bus
== start
&&
801 cfg
->end_bus
== end
) {
802 list_del_rcu(&cfg
->list
);
804 pci_mmcfg_arch_unmap(cfg
);
806 release_resource(&cfg
->res
);
807 mutex_unlock(&pci_mmcfg_lock
);
811 mutex_unlock(&pci_mmcfg_lock
);