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
);
32 #define pci_mmcfg_lock_held() lock_is_held(&(pci_mmcfg_lock).dep_map)
34 LIST_HEAD(pci_mmcfg_list
);
36 static void __init
pci_mmconfig_remove(struct pci_mmcfg_region
*cfg
)
39 release_resource(&cfg
->res
);
44 static void __init
free_all_mmcfg(void)
46 struct pci_mmcfg_region
*cfg
, *tmp
;
48 pci_mmcfg_arch_free();
49 list_for_each_entry_safe(cfg
, tmp
, &pci_mmcfg_list
, list
)
50 pci_mmconfig_remove(cfg
);
53 static void list_add_sorted(struct pci_mmcfg_region
*new)
55 struct pci_mmcfg_region
*cfg
;
57 /* keep list sorted by segment and starting bus number */
58 list_for_each_entry_rcu(cfg
, &pci_mmcfg_list
, list
, pci_mmcfg_lock_held()) {
59 if (cfg
->segment
> new->segment
||
60 (cfg
->segment
== new->segment
&&
61 cfg
->start_bus
>= new->start_bus
)) {
62 list_add_tail_rcu(&new->list
, &cfg
->list
);
66 list_add_tail_rcu(&new->list
, &pci_mmcfg_list
);
69 static struct pci_mmcfg_region
*pci_mmconfig_alloc(int segment
, int start
,
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 struct pci_mmcfg_region
*__init
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
, pci_mmcfg_lock_held())
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 __initconst
= 0x90;
246 static const u32 extcfg_regsize __initconst
= 4;
247 static const u32 extcfg_enable_mask __initconst
= 1 << 31;
248 static const u32 extcfg_start_mask __initconst
= 0xff << 16;
249 static const int extcfg_start_shift __initconst
= 16;
250 static const u32 extcfg_size_mask __initconst
= 0x3 << 28;
251 static const int extcfg_size_shift __initconst
= 28;
252 static const int extcfg_sizebus
[] __initconst
= {
253 0x100, 0x80, 0x40, 0x20
255 static const u32 extcfg_base_mask
[] __initconst
= {
256 0x7ff8, 0x7ffc, 0x7ffe, 0x7fff
258 static const int extcfg_base_lshift __initconst
= 25;
261 * do check if amd fam10h already took over
263 if (!acpi_disabled
|| !list_empty(&pci_mmcfg_list
) || mcp55_checked
)
266 mcp55_checked
= true;
267 for (bus
= 0; bus
< 256; bus
++) {
271 int start
, size_index
, end
;
273 raw_pci_ops
->read(0, bus
, PCI_DEVFN(0, 0), 0, 4, &l
);
275 device
= (l
>> 16) & 0xffff;
277 if (PCI_VENDOR_ID_NVIDIA
!= vendor
|| 0x0369 != device
)
280 raw_pci_ops
->read(0, bus
, PCI_DEVFN(0, 0), extcfg_regnum
,
281 extcfg_regsize
, &extcfg
);
283 if (!(extcfg
& extcfg_enable_mask
))
286 size_index
= (extcfg
& extcfg_size_mask
) >> extcfg_size_shift
;
287 base
= extcfg
& extcfg_base_mask
[size_index
];
288 /* base could > 4G */
289 base
<<= extcfg_base_lshift
;
290 start
= (extcfg
& extcfg_start_mask
) >> extcfg_start_shift
;
291 end
= start
+ extcfg_sizebus
[size_index
] - 1;
292 if (pci_mmconfig_add(0, start
, end
, base
) == NULL
)
294 mcp55_mmconf_found
++;
297 if (!mcp55_mmconf_found
)
300 return "nVidia MCP55";
303 struct pci_mmcfg_hostbridge_probe
{
308 const char *(*probe
)(void);
311 static const struct pci_mmcfg_hostbridge_probe pci_mmcfg_probes
[] __initconst
= {
312 { 0, PCI_DEVFN(0, 0), PCI_VENDOR_ID_INTEL
,
313 PCI_DEVICE_ID_INTEL_E7520_MCH
, pci_mmcfg_e7520
},
314 { 0, PCI_DEVFN(0, 0), PCI_VENDOR_ID_INTEL
,
315 PCI_DEVICE_ID_INTEL_82945G_HB
, pci_mmcfg_intel_945
},
316 { 0, PCI_DEVFN(0x18, 0), PCI_VENDOR_ID_AMD
,
317 0x1200, pci_mmcfg_amd_fam10h
},
318 { 0xff, PCI_DEVFN(0, 0), PCI_VENDOR_ID_AMD
,
319 0x1200, pci_mmcfg_amd_fam10h
},
320 { 0, PCI_DEVFN(0, 0), PCI_VENDOR_ID_NVIDIA
,
321 0x0369, pci_mmcfg_nvidia_mcp55
},
324 static void __init
pci_mmcfg_check_end_bus_number(void)
326 struct pci_mmcfg_region
*cfg
, *cfgx
;
329 list_for_each_entry(cfg
, &pci_mmcfg_list
, list
) {
330 if (cfg
->end_bus
< cfg
->start_bus
)
333 /* Don't access the list head ! */
334 if (cfg
->list
.next
== &pci_mmcfg_list
)
337 cfgx
= list_entry(cfg
->list
.next
, typeof(*cfg
), list
);
338 if (cfg
->end_bus
>= cfgx
->start_bus
)
339 cfg
->end_bus
= cfgx
->start_bus
- 1;
343 static int __init
pci_mmcfg_check_hostbridge(void)
356 for (i
= 0; i
< ARRAY_SIZE(pci_mmcfg_probes
); i
++) {
357 bus
= pci_mmcfg_probes
[i
].bus
;
358 devfn
= pci_mmcfg_probes
[i
].devfn
;
359 raw_pci_ops
->read(0, bus
, devfn
, 0, 4, &l
);
361 device
= (l
>> 16) & 0xffff;
364 if (pci_mmcfg_probes
[i
].vendor
== vendor
&&
365 pci_mmcfg_probes
[i
].device
== device
)
366 name
= pci_mmcfg_probes
[i
].probe();
369 pr_info(PREFIX
"%s with MMCONFIG support\n", name
);
372 /* some end_bus_number is crazy, fix it */
373 pci_mmcfg_check_end_bus_number();
375 return !list_empty(&pci_mmcfg_list
);
378 static acpi_status
check_mcfg_resource(struct acpi_resource
*res
, void *data
)
380 struct resource
*mcfg_res
= data
;
381 struct acpi_resource_address64 address
;
384 if (res
->type
== ACPI_RESOURCE_TYPE_FIXED_MEMORY32
) {
385 struct acpi_resource_fixed_memory32
*fixmem32
=
386 &res
->data
.fixed_memory32
;
389 if ((mcfg_res
->start
>= fixmem32
->address
) &&
390 (mcfg_res
->end
< (fixmem32
->address
+
391 fixmem32
->address_length
))) {
393 return AE_CTRL_TERMINATE
;
396 if ((res
->type
!= ACPI_RESOURCE_TYPE_ADDRESS32
) &&
397 (res
->type
!= ACPI_RESOURCE_TYPE_ADDRESS64
))
400 status
= acpi_resource_to_address64(res
, &address
);
401 if (ACPI_FAILURE(status
) ||
402 (address
.address
.address_length
<= 0) ||
403 (address
.resource_type
!= ACPI_MEMORY_RANGE
))
406 if ((mcfg_res
->start
>= address
.address
.minimum
) &&
407 (mcfg_res
->end
< (address
.address
.minimum
+ address
.address
.address_length
))) {
409 return AE_CTRL_TERMINATE
;
414 static acpi_status
find_mboard_resource(acpi_handle handle
, u32 lvl
,
415 void *context
, void **rv
)
417 struct resource
*mcfg_res
= context
;
419 acpi_walk_resources(handle
, METHOD_NAME__CRS
,
420 check_mcfg_resource
, context
);
423 return AE_CTRL_TERMINATE
;
428 static bool is_acpi_reserved(u64 start
, u64 end
, enum e820_type not_used
)
430 struct resource mcfg_res
;
432 mcfg_res
.start
= start
;
433 mcfg_res
.end
= end
- 1;
436 acpi_get_devices("PNP0C01", find_mboard_resource
, &mcfg_res
, NULL
);
439 acpi_get_devices("PNP0C02", find_mboard_resource
, &mcfg_res
,
442 return mcfg_res
.flags
;
445 typedef bool (*check_reserved_t
)(u64 start
, u64 end
, enum e820_type type
);
447 static bool __ref
is_mmconf_reserved(check_reserved_t is_reserved
,
448 struct pci_mmcfg_region
*cfg
,
449 struct device
*dev
, int with_e820
)
451 u64 addr
= cfg
->res
.start
;
452 u64 size
= resource_size(&cfg
->res
);
455 char *method
= with_e820
? "E820" : "ACPI motherboard resources";
457 while (!is_reserved(addr
, addr
+ size
, E820_TYPE_RESERVED
)) {
459 if (size
< (16UL<<20))
463 if (size
< (16UL<<20) && size
!= old_size
)
467 dev_info(dev
, "MMCONFIG at %pR reserved in %s\n",
470 pr_info(PREFIX
"MMCONFIG at %pR reserved in %s\n",
473 if (old_size
!= size
) {
475 cfg
->end_bus
= cfg
->start_bus
+ ((size
>>20) - 1);
476 num_buses
= cfg
->end_bus
- cfg
->start_bus
+ 1;
477 cfg
->res
.end
= cfg
->res
.start
+
478 PCI_MMCFG_BUS_OFFSET(num_buses
) - 1;
479 snprintf(cfg
->name
, PCI_MMCFG_RESOURCE_NAME_LEN
,
480 "PCI MMCONFIG %04x [bus %02x-%02x]",
481 cfg
->segment
, cfg
->start_bus
, cfg
->end_bus
);
486 "at %pR (base %#lx) (size reduced!)\n",
487 &cfg
->res
, (unsigned long) cfg
->address
);
490 "MMCONFIG for %04x [bus%02x-%02x] "
491 "at %pR (base %#lx) (size reduced!)\n",
492 cfg
->segment
, cfg
->start_bus
, cfg
->end_bus
,
493 &cfg
->res
, (unsigned long) cfg
->address
);
500 pci_mmcfg_check_reserved(struct device
*dev
, struct pci_mmcfg_region
*cfg
, int early
)
502 if (!early
&& !acpi_disabled
) {
503 if (is_mmconf_reserved(is_acpi_reserved
, cfg
, dev
, 0))
507 dev_info(dev
, FW_INFO
508 "MMCONFIG at %pR not reserved in "
509 "ACPI motherboard resources\n",
512 pr_info(FW_INFO PREFIX
513 "MMCONFIG at %pR not reserved in "
514 "ACPI motherboard resources\n",
519 * e820__mapped_all() is marked as __init.
520 * All entries from ACPI MCFG table have been checked at boot time.
521 * For MCFG information constructed from hotpluggable host bridge's
522 * _CBA method, just assume it's reserved.
524 if (pci_mmcfg_running_state
)
527 /* Don't try to do this check unless configuration
528 type 1 is available. how about type 2 ?*/
530 return is_mmconf_reserved(e820__mapped_all
, cfg
, dev
, 1);
535 static void __init
pci_mmcfg_reject_broken(int early
)
537 struct pci_mmcfg_region
*cfg
;
539 list_for_each_entry(cfg
, &pci_mmcfg_list
, list
) {
540 if (pci_mmcfg_check_reserved(NULL
, cfg
, early
) == 0) {
541 pr_info(PREFIX
"not using MMCONFIG\n");
548 static int __init
acpi_mcfg_check_entry(struct acpi_table_mcfg
*mcfg
,
549 struct acpi_mcfg_allocation
*cfg
)
551 if (cfg
->address
< 0xFFFFFFFF)
554 if (!strncmp(mcfg
->header
.oem_id
, "SGI", 3))
557 if ((mcfg
->header
.revision
>= 1) && (dmi_get_bios_year() >= 2010))
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 #ifdef CONFIG_ACPI_APEI
611 extern int (*arch_apei_filter_addr
)(int (*func
)(__u64 start
, __u64 size
,
612 void *data
), void *data
);
614 static int pci_mmcfg_for_each_region(int (*func
)(__u64 start
, __u64 size
,
615 void *data
), void *data
)
617 struct pci_mmcfg_region
*cfg
;
620 if (list_empty(&pci_mmcfg_list
))
623 list_for_each_entry(cfg
, &pci_mmcfg_list
, list
) {
624 rc
= func(cfg
->res
.start
, resource_size(&cfg
->res
), data
);
631 #define set_apei_filter() (arch_apei_filter_addr = pci_mmcfg_for_each_region)
633 #define set_apei_filter()
636 static void __init
__pci_mmcfg_init(int early
)
638 pci_mmcfg_reject_broken(early
);
639 if (list_empty(&pci_mmcfg_list
))
642 if (pcibios_last_bus
< 0) {
643 const struct pci_mmcfg_region
*cfg
;
645 list_for_each_entry(cfg
, &pci_mmcfg_list
, list
) {
648 pcibios_last_bus
= cfg
->end_bus
;
652 if (pci_mmcfg_arch_init())
653 pci_probe
= (pci_probe
& ~PCI_PROBE_MASK
) | PCI_PROBE_MMCONF
;
656 pci_mmcfg_arch_init_failed
= true;
660 static int __initdata known_bridge
;
662 void __init
pci_mmcfg_early_init(void)
664 if (pci_probe
& PCI_PROBE_MMCONF
) {
665 if (pci_mmcfg_check_hostbridge())
668 acpi_sfi_table_parse(ACPI_SIG_MCFG
, pci_parse_mcfg
);
675 void __init
pci_mmcfg_late_init(void)
677 /* MMCONFIG disabled */
678 if ((pci_probe
& PCI_PROBE_MMCONF
) == 0)
684 /* MMCONFIG hasn't been enabled yet, try again */
685 if (pci_probe
& PCI_PROBE_MASK
& ~PCI_PROBE_MMCONF
) {
686 acpi_sfi_table_parse(ACPI_SIG_MCFG
, pci_parse_mcfg
);
691 static int __init
pci_mmcfg_late_insert_resources(void)
693 struct pci_mmcfg_region
*cfg
;
695 pci_mmcfg_running_state
= true;
697 /* If we are not using MMCONFIG, don't insert the resources. */
698 if ((pci_probe
& PCI_PROBE_MMCONF
) == 0)
702 * Attempt to insert the mmcfg resources but not with the busy flag
703 * marked so it won't cause request errors when __request_region is
706 list_for_each_entry(cfg
, &pci_mmcfg_list
, list
)
707 if (!cfg
->res
.parent
)
708 insert_resource(&iomem_resource
, &cfg
->res
);
714 * Perform MMCONFIG resource insertion after PCI initialization to allow for
715 * misprogrammed MCFG tables that state larger sizes but actually conflict
716 * with other system resources.
718 late_initcall(pci_mmcfg_late_insert_resources
);
720 /* Add MMCFG information for host bridges */
721 int pci_mmconfig_insert(struct device
*dev
, u16 seg
, u8 start
, u8 end
,
725 struct resource
*tmp
= NULL
;
726 struct pci_mmcfg_region
*cfg
;
728 if (!(pci_probe
& PCI_PROBE_MMCONF
) || pci_mmcfg_arch_init_failed
)
734 mutex_lock(&pci_mmcfg_lock
);
735 cfg
= pci_mmconfig_lookup(seg
, start
);
737 if (cfg
->end_bus
< end
)
738 dev_info(dev
, FW_INFO
740 "domain %04x [bus %02x-%02x] "
741 "only partially covers this bridge\n",
742 cfg
->segment
, cfg
->start_bus
, cfg
->end_bus
);
743 mutex_unlock(&pci_mmcfg_lock
);
748 mutex_unlock(&pci_mmcfg_lock
);
753 cfg
= pci_mmconfig_alloc(seg
, start
, end
, addr
);
755 dev_warn(dev
, "fail to add MMCONFIG (out of memory)\n");
757 } else if (!pci_mmcfg_check_reserved(dev
, cfg
, 0)) {
758 dev_warn(dev
, FW_BUG
"MMCONFIG %pR isn't reserved\n",
761 /* Insert resource if it's not in boot stage */
762 if (pci_mmcfg_running_state
)
763 tmp
= insert_resource_conflict(&iomem_resource
,
768 "MMCONFIG %pR conflicts with "
770 &cfg
->res
, tmp
->name
, tmp
);
771 } else if (pci_mmcfg_arch_map(cfg
)) {
772 dev_warn(dev
, "fail to map MMCONFIG %pR.\n",
775 list_add_sorted(cfg
);
776 dev_info(dev
, "MMCONFIG at %pR (base %#lx)\n",
777 &cfg
->res
, (unsigned long)addr
);
785 release_resource(&cfg
->res
);
789 mutex_unlock(&pci_mmcfg_lock
);
794 /* Delete MMCFG information for host bridges */
795 int pci_mmconfig_delete(u16 seg
, u8 start
, u8 end
)
797 struct pci_mmcfg_region
*cfg
;
799 mutex_lock(&pci_mmcfg_lock
);
800 list_for_each_entry_rcu(cfg
, &pci_mmcfg_list
, list
)
801 if (cfg
->segment
== seg
&& cfg
->start_bus
== start
&&
802 cfg
->end_bus
== end
) {
803 list_del_rcu(&cfg
->list
);
805 pci_mmcfg_arch_unmap(cfg
);
807 release_resource(&cfg
->res
);
808 mutex_unlock(&pci_mmcfg_lock
);
812 mutex_unlock(&pci_mmcfg_lock
);