1 #include <linux/kernel.h>
2 #include <linux/types.h>
3 #include <linux/init.h>
4 #include <linux/bootmem.h>
5 #include <linux/ioport.h>
6 #include <linux/string.h>
7 #include <linux/kexec.h>
8 #include <linux/module.h>
10 #include <linux/efi.h>
11 #include <linux/pfn.h>
12 #include <linux/uaccess.h>
14 #include <asm/pgtable.h>
17 #include <asm/setup.h>
21 EXPORT_SYMBOL(efi_enabled
);
25 struct change_member
{
26 struct e820entry
*pbios
; /* pointer to original bios entry */
27 unsigned long long addr
; /* address for this change point */
29 static struct change_member change_point_list
[2*E820MAX
] __initdata
;
30 static struct change_member
*change_point
[2*E820MAX
] __initdata
;
31 static struct e820entry
*overlap_list
[E820MAX
] __initdata
;
32 static struct e820entry new_bios
[E820MAX
] __initdata
;
33 /* For PCI or other memory-mapped resources */
34 unsigned long pci_mem_start
= 0x10000000;
36 EXPORT_SYMBOL(pci_mem_start
);
38 extern int user_defined_memmap
;
39 struct resource data_resource
= {
40 .name
= "Kernel data",
43 .flags
= IORESOURCE_BUSY
| IORESOURCE_MEM
46 struct resource code_resource
= {
47 .name
= "Kernel code",
50 .flags
= IORESOURCE_BUSY
| IORESOURCE_MEM
53 static struct resource system_rom_resource
= {
57 .flags
= IORESOURCE_BUSY
| IORESOURCE_READONLY
| IORESOURCE_MEM
60 static struct resource extension_rom_resource
= {
61 .name
= "Extension ROM",
64 .flags
= IORESOURCE_BUSY
| IORESOURCE_READONLY
| IORESOURCE_MEM
67 static struct resource adapter_rom_resources
[] = { {
68 .name
= "Adapter ROM",
71 .flags
= IORESOURCE_BUSY
| IORESOURCE_READONLY
| IORESOURCE_MEM
73 .name
= "Adapter ROM",
76 .flags
= IORESOURCE_BUSY
| IORESOURCE_READONLY
| IORESOURCE_MEM
78 .name
= "Adapter ROM",
81 .flags
= IORESOURCE_BUSY
| IORESOURCE_READONLY
| IORESOURCE_MEM
83 .name
= "Adapter ROM",
86 .flags
= IORESOURCE_BUSY
| IORESOURCE_READONLY
| IORESOURCE_MEM
88 .name
= "Adapter ROM",
91 .flags
= IORESOURCE_BUSY
| IORESOURCE_READONLY
| IORESOURCE_MEM
93 .name
= "Adapter ROM",
96 .flags
= IORESOURCE_BUSY
| IORESOURCE_READONLY
| IORESOURCE_MEM
99 static struct resource video_rom_resource
= {
103 .flags
= IORESOURCE_BUSY
| IORESOURCE_READONLY
| IORESOURCE_MEM
106 static struct resource video_ram_resource
= {
107 .name
= "Video RAM area",
110 .flags
= IORESOURCE_BUSY
| IORESOURCE_MEM
113 static struct resource standard_io_resources
[] = { {
117 .flags
= IORESOURCE_BUSY
| IORESOURCE_IO
122 .flags
= IORESOURCE_BUSY
| IORESOURCE_IO
127 .flags
= IORESOURCE_BUSY
| IORESOURCE_IO
132 .flags
= IORESOURCE_BUSY
| IORESOURCE_IO
137 .flags
= IORESOURCE_BUSY
| IORESOURCE_IO
139 .name
= "dma page reg",
142 .flags
= IORESOURCE_BUSY
| IORESOURCE_IO
147 .flags
= IORESOURCE_BUSY
| IORESOURCE_IO
152 .flags
= IORESOURCE_BUSY
| IORESOURCE_IO
157 .flags
= IORESOURCE_BUSY
| IORESOURCE_IO
160 #define ROMSIGNATURE 0xaa55
162 static int __init
romsignature(const unsigned char *rom
)
166 return probe_kernel_address((const unsigned short *)rom
, sig
) == 0 &&
170 static int __init
romchecksum(unsigned char *rom
, unsigned long length
)
174 for (sum
= 0; length
; length
--)
179 static void __init
probe_roms(void)
181 unsigned long start
, length
, upper
;
186 upper
= adapter_rom_resources
[0].start
;
187 for (start
= video_rom_resource
.start
; start
< upper
; start
+= 2048) {
188 rom
= isa_bus_to_virt(start
);
189 if (!romsignature(rom
))
192 video_rom_resource
.start
= start
;
194 /* 0 < length <= 0x7f * 512, historically */
195 length
= rom
[2] * 512;
197 /* if checksum okay, trust length byte */
198 if (length
&& romchecksum(rom
, length
))
199 video_rom_resource
.end
= start
+ length
- 1;
201 request_resource(&iomem_resource
, &video_rom_resource
);
205 start
= (video_rom_resource
.end
+ 1 + 2047) & ~2047UL;
210 request_resource(&iomem_resource
, &system_rom_resource
);
211 upper
= system_rom_resource
.start
;
213 /* check for extension rom (ignore length byte!) */
214 rom
= isa_bus_to_virt(extension_rom_resource
.start
);
215 if (romsignature(rom
)) {
216 length
= extension_rom_resource
.end
- extension_rom_resource
.start
+ 1;
217 if (romchecksum(rom
, length
)) {
218 request_resource(&iomem_resource
, &extension_rom_resource
);
219 upper
= extension_rom_resource
.start
;
223 /* check for adapter roms on 2k boundaries */
224 for (i
= 0; i
< ARRAY_SIZE(adapter_rom_resources
) && start
< upper
; start
+= 2048) {
225 rom
= isa_bus_to_virt(start
);
226 if (!romsignature(rom
))
229 /* 0 < length <= 0x7f * 512, historically */
230 length
= rom
[2] * 512;
232 /* but accept any length that fits if checksum okay */
233 if (!length
|| start
+ length
> upper
|| !romchecksum(rom
, length
))
236 adapter_rom_resources
[i
].start
= start
;
237 adapter_rom_resources
[i
].end
= start
+ length
- 1;
238 request_resource(&iomem_resource
, &adapter_rom_resources
[i
]);
240 start
= adapter_rom_resources
[i
++].end
& ~2047UL;
245 * Request address space for all standard RAM and ROM resources
246 * and also for regions reported as reserved by the e820.
249 legacy_init_iomem_resources(struct resource
*code_resource
, struct resource
*data_resource
)
254 for (i
= 0; i
< e820
.nr_map
; i
++) {
255 struct resource
*res
;
256 #ifndef CONFIG_RESOURCES_64BIT
257 if (e820
.map
[i
].addr
+ e820
.map
[i
].size
> 0x100000000ULL
)
260 res
= kzalloc(sizeof(struct resource
), GFP_ATOMIC
);
261 switch (e820
.map
[i
].type
) {
262 case E820_RAM
: res
->name
= "System RAM"; break;
263 case E820_ACPI
: res
->name
= "ACPI Tables"; break;
264 case E820_NVS
: res
->name
= "ACPI Non-volatile Storage"; break;
265 default: res
->name
= "reserved";
267 res
->start
= e820
.map
[i
].addr
;
268 res
->end
= res
->start
+ e820
.map
[i
].size
- 1;
269 res
->flags
= IORESOURCE_MEM
| IORESOURCE_BUSY
;
270 if (request_resource(&iomem_resource
, res
)) {
274 if (e820
.map
[i
].type
== E820_RAM
) {
276 * We don't know which RAM region contains kernel data,
277 * so we try it repeatedly and let the resource manager
280 request_resource(res
, code_resource
);
281 request_resource(res
, data_resource
);
283 request_resource(res
, &crashk_res
);
290 * Request address space for all standard resources
292 * This is called just before pcibios_init(), which is also a
293 * subsys_initcall, but is linked in later (in arch/i386/pci/common.c).
295 static int __init
request_standard_resources(void)
299 printk("Setting up standard PCI resources\n");
301 efi_initialize_iomem_resources(&code_resource
, &data_resource
);
303 legacy_init_iomem_resources(&code_resource
, &data_resource
);
305 /* EFI systems may still have VGA */
306 request_resource(&iomem_resource
, &video_ram_resource
);
308 /* request I/O space for devices used on all i[345]86 PCs */
309 for (i
= 0; i
< ARRAY_SIZE(standard_io_resources
); i
++)
310 request_resource(&ioport_resource
, &standard_io_resources
[i
]);
314 subsys_initcall(request_standard_resources
);
316 void __init
add_memory_region(unsigned long long start
,
317 unsigned long long size
, int type
)
325 printk(KERN_ERR
"Ooops! Too many entries in the memory map!\n");
329 e820
.map
[x
].addr
= start
;
330 e820
.map
[x
].size
= size
;
331 e820
.map
[x
].type
= type
;
334 } /* add_memory_region */
337 * Sanitize the BIOS e820 map.
339 * Some e820 responses include overlapping entries. The following
340 * replaces the original e820 map with a new one, removing overlaps.
343 int __init
sanitize_e820_map(struct e820entry
* biosmap
, char * pnr_map
)
345 struct change_member
*change_tmp
;
346 unsigned long current_type
, last_type
;
347 unsigned long long last_addr
;
348 int chgidx
, still_changing
;
351 int old_nr
, new_nr
, chg_nr
;
355 Visually we're performing the following (1,2,3,4 = memory types)...
357 Sample memory map (w/overlaps):
358 ____22__________________
359 ______________________4_
360 ____1111________________
361 _44_____________________
362 11111111________________
363 ____________________33__
364 ___________44___________
365 __________33333_________
366 ______________22________
367 ___________________2222_
368 _________111111111______
369 _____________________11_
370 _________________4______
372 Sanitized equivalent (no overlap):
373 1_______________________
374 _44_____________________
375 ___1____________________
376 ____22__________________
377 ______11________________
378 _________1______________
379 __________3_____________
380 ___________44___________
381 _____________33_________
382 _______________2________
383 ________________1_______
384 _________________4______
385 ___________________2____
386 ____________________33__
387 ______________________4_
389 printk("sanitize start\n");
390 /* if there's only one memory region, don't bother */
392 printk("sanitize bail 0\n");
398 /* bail out if we find any unreasonable addresses in bios map */
399 for (i
=0; i
<old_nr
; i
++)
400 if (biosmap
[i
].addr
+ biosmap
[i
].size
< biosmap
[i
].addr
) {
401 printk("sanitize bail 1\n");
405 /* create pointers for initial change-point information (for sorting) */
406 for (i
=0; i
< 2*old_nr
; i
++)
407 change_point
[i
] = &change_point_list
[i
];
409 /* record all known change-points (starting and ending addresses),
410 omitting those that are for empty memory regions */
412 for (i
=0; i
< old_nr
; i
++) {
413 if (biosmap
[i
].size
!= 0) {
414 change_point
[chgidx
]->addr
= biosmap
[i
].addr
;
415 change_point
[chgidx
++]->pbios
= &biosmap
[i
];
416 change_point
[chgidx
]->addr
= biosmap
[i
].addr
+ biosmap
[i
].size
;
417 change_point
[chgidx
++]->pbios
= &biosmap
[i
];
420 chg_nr
= chgidx
; /* true number of change-points */
422 /* sort change-point list by memory addresses (low -> high) */
424 while (still_changing
) {
426 for (i
=1; i
< chg_nr
; i
++) {
427 /* if <current_addr> > <last_addr>, swap */
428 /* or, if current=<start_addr> & last=<end_addr>, swap */
429 if ((change_point
[i
]->addr
< change_point
[i
-1]->addr
) ||
430 ((change_point
[i
]->addr
== change_point
[i
-1]->addr
) &&
431 (change_point
[i
]->addr
== change_point
[i
]->pbios
->addr
) &&
432 (change_point
[i
-1]->addr
!= change_point
[i
-1]->pbios
->addr
))
435 change_tmp
= change_point
[i
];
436 change_point
[i
] = change_point
[i
-1];
437 change_point
[i
-1] = change_tmp
;
443 /* create a new bios memory map, removing overlaps */
444 overlap_entries
=0; /* number of entries in the overlap table */
445 new_bios_entry
=0; /* index for creating new bios map entries */
446 last_type
= 0; /* start with undefined memory type */
447 last_addr
= 0; /* start with 0 as last starting address */
448 /* loop through change-points, determining affect on the new bios map */
449 for (chgidx
=0; chgidx
< chg_nr
; chgidx
++)
451 /* keep track of all overlapping bios entries */
452 if (change_point
[chgidx
]->addr
== change_point
[chgidx
]->pbios
->addr
)
454 /* add map entry to overlap list (> 1 entry implies an overlap) */
455 overlap_list
[overlap_entries
++]=change_point
[chgidx
]->pbios
;
459 /* remove entry from list (order independent, so swap with last) */
460 for (i
=0; i
<overlap_entries
; i
++)
462 if (overlap_list
[i
] == change_point
[chgidx
]->pbios
)
463 overlap_list
[i
] = overlap_list
[overlap_entries
-1];
467 /* if there are overlapping entries, decide which "type" to use */
468 /* (larger value takes precedence -- 1=usable, 2,3,4,4+=unusable) */
470 for (i
=0; i
<overlap_entries
; i
++)
471 if (overlap_list
[i
]->type
> current_type
)
472 current_type
= overlap_list
[i
]->type
;
473 /* continue building up new bios map based on this information */
474 if (current_type
!= last_type
) {
475 if (last_type
!= 0) {
476 new_bios
[new_bios_entry
].size
=
477 change_point
[chgidx
]->addr
- last_addr
;
478 /* move forward only if the new size was non-zero */
479 if (new_bios
[new_bios_entry
].size
!= 0)
480 if (++new_bios_entry
>= E820MAX
)
481 break; /* no more space left for new bios entries */
483 if (current_type
!= 0) {
484 new_bios
[new_bios_entry
].addr
= change_point
[chgidx
]->addr
;
485 new_bios
[new_bios_entry
].type
= current_type
;
486 last_addr
=change_point
[chgidx
]->addr
;
488 last_type
= current_type
;
491 new_nr
= new_bios_entry
; /* retain count for new bios entries */
493 /* copy new bios mapping into original location */
494 memcpy(biosmap
, new_bios
, new_nr
*sizeof(struct e820entry
));
497 printk("sanitize end\n");
502 * Copy the BIOS e820 map into a safe place.
504 * Sanity-check it while we're at it..
506 * If we're lucky and live on a modern system, the setup code
507 * will have given us a memory map that we can use to properly
508 * set up memory. If we aren't, we'll fake a memory map.
510 * We check to see that the memory map contains at least 2 elements
511 * before we'll use it, because the detection code in setup.S may
512 * not be perfect and most every PC known to man has two memory
513 * regions: one from 0 to 640k, and one from 1mb up. (The IBM
514 * thinkpad 560x, for example, does not cooperate with the memory
517 int __init
copy_e820_map(struct e820entry
* biosmap
, int nr_map
)
519 /* Only one memory region (or negative)? Ignore it */
524 unsigned long long start
= biosmap
->addr
;
525 unsigned long long size
= biosmap
->size
;
526 unsigned long long end
= start
+ size
;
527 unsigned long type
= biosmap
->type
;
528 printk("copy_e820_map() start: %016Lx size: %016Lx end: %016Lx type: %ld\n", start
, size
, end
, type
);
530 /* Overflow in 64 bits? Ignore the memory map. */
535 * Some BIOSes claim RAM in the 640k - 1M region.
536 * Not right. Fix it up.
538 if (type
== E820_RAM
) {
539 printk("copy_e820_map() type is E820_RAM\n");
540 if (start
< 0x100000ULL
&& end
> 0xA0000ULL
) {
541 printk("copy_e820_map() lies in range...\n");
542 if (start
< 0xA0000ULL
) {
543 printk("copy_e820_map() start < 0xA0000ULL\n");
544 add_memory_region(start
, 0xA0000ULL
-start
, type
);
546 if (end
<= 0x100000ULL
) {
547 printk("copy_e820_map() end <= 0x100000ULL\n");
554 add_memory_region(start
, size
, type
);
555 } while (biosmap
++,--nr_map
);
560 * Callback for efi_memory_walk.
563 efi_find_max_pfn(unsigned long start
, unsigned long end
, void *arg
)
565 unsigned long *max_pfn
= arg
, pfn
;
568 pfn
= PFN_UP(end
-1);
576 efi_memory_present_wrapper(unsigned long start
, unsigned long end
, void *arg
)
578 memory_present(0, PFN_UP(start
), PFN_DOWN(end
));
583 * Find the highest page frame number we have available
585 void __init
find_max_pfn(void)
591 efi_memmap_walk(efi_find_max_pfn
, &max_pfn
);
592 efi_memmap_walk(efi_memory_present_wrapper
, NULL
);
596 for (i
= 0; i
< e820
.nr_map
; i
++) {
597 unsigned long start
, end
;
599 if (e820
.map
[i
].type
!= E820_RAM
)
601 start
= PFN_UP(e820
.map
[i
].addr
);
602 end
= PFN_DOWN(e820
.map
[i
].addr
+ e820
.map
[i
].size
);
607 memory_present(0, start
, end
);
612 * Free all available memory for boot time allocation. Used
613 * as a callback function by efi_memory_walk()
617 free_available_memory(unsigned long start
, unsigned long end
, void *arg
)
619 /* check max_low_pfn */
620 if (start
>= (max_low_pfn
<< PAGE_SHIFT
))
622 if (end
>= (max_low_pfn
<< PAGE_SHIFT
))
623 end
= max_low_pfn
<< PAGE_SHIFT
;
625 free_bootmem(start
, end
- start
);
630 * Register fully available low RAM pages with the bootmem allocator.
632 void __init
register_bootmem_low_pages(unsigned long max_low_pfn
)
637 efi_memmap_walk(free_available_memory
, NULL
);
640 for (i
= 0; i
< e820
.nr_map
; i
++) {
641 unsigned long curr_pfn
, last_pfn
, size
;
643 * Reserve usable low memory
645 if (e820
.map
[i
].type
!= E820_RAM
)
648 * We are rounding up the start address of usable memory:
650 curr_pfn
= PFN_UP(e820
.map
[i
].addr
);
651 if (curr_pfn
>= max_low_pfn
)
654 * ... and at the end of the usable range downwards:
656 last_pfn
= PFN_DOWN(e820
.map
[i
].addr
+ e820
.map
[i
].size
);
658 if (last_pfn
> max_low_pfn
)
659 last_pfn
= max_low_pfn
;
662 * .. finally, did all the rounding and playing
663 * around just make the area go away?
665 if (last_pfn
<= curr_pfn
)
668 size
= last_pfn
- curr_pfn
;
669 free_bootmem(PFN_PHYS(curr_pfn
), PFN_PHYS(size
));
673 void __init
e820_register_memory(void)
675 unsigned long gapstart
, gapsize
, round
;
676 unsigned long long last
;
680 * Search for the bigest gap in the low 32 bits of the e820
683 last
= 0x100000000ull
;
684 gapstart
= 0x10000000;
688 unsigned long long start
= e820
.map
[i
].addr
;
689 unsigned long long end
= start
+ e820
.map
[i
].size
;
692 * Since "last" is at most 4GB, we know we'll
693 * fit in 32 bits if this condition is true
696 unsigned long gap
= last
- end
;
708 * See how much we want to round up: start off with
709 * rounding to the next 1MB area.
712 while ((gapsize
>> 4) > round
)
714 /* Fun with two's complement */
715 pci_mem_start
= (gapstart
+ round
) & -round
;
717 printk("Allocating PCI resources starting at %08lx (gap: %08lx:%08lx)\n",
718 pci_mem_start
, gapstart
, gapsize
);
721 void __init
print_memory_map(char *who
)
725 for (i
= 0; i
< e820
.nr_map
; i
++) {
726 printk(" %s: %016Lx - %016Lx ", who
,
728 e820
.map
[i
].addr
+ e820
.map
[i
].size
);
729 switch (e820
.map
[i
].type
) {
730 case E820_RAM
: printk("(usable)\n");
733 printk("(reserved)\n");
736 printk("(ACPI data)\n");
739 printk("(ACPI NVS)\n");
741 default: printk("type %lu\n", e820
.map
[i
].type
);
747 static __init __always_inline
void efi_limit_regions(unsigned long long size
)
749 unsigned long long current_addr
= 0;
750 efi_memory_desc_t
*md
, *next_md
;
756 for (p
= p1
, i
= 0; p
< memmap
.map_end
; p
+= memmap
.desc_size
, i
++) {
759 current_addr
= md
->phys_addr
+
760 PFN_PHYS(md
->num_pages
);
761 if (is_available_memory(md
)) {
762 if (md
->phys_addr
>= size
) continue;
763 memcpy(next_md
, md
, memmap
.desc_size
);
764 if (current_addr
>= size
) {
765 next_md
->num_pages
-=
766 PFN_UP(current_addr
-size
);
768 p1
+= memmap
.desc_size
;
771 } else if ((md
->attribute
& EFI_MEMORY_RUNTIME
) ==
772 EFI_MEMORY_RUNTIME
) {
773 /* In order to make runtime services
774 * available we have to include runtime
775 * memory regions in memory map */
776 memcpy(next_md
, md
, memmap
.desc_size
);
777 p1
+= memmap
.desc_size
;
783 memmap
.map_end
= memmap
.map
+
784 (memmap
.nr_map
* memmap
.desc_size
);
787 void __init
limit_regions(unsigned long long size
)
789 unsigned long long current_addr
;
792 print_memory_map("limit_regions start");
794 efi_limit_regions(size
);
797 for (i
= 0; i
< e820
.nr_map
; i
++) {
798 current_addr
= e820
.map
[i
].addr
+ e820
.map
[i
].size
;
799 if (current_addr
< size
)
802 if (e820
.map
[i
].type
!= E820_RAM
)
805 if (e820
.map
[i
].addr
>= size
) {
807 * This region starts past the end of the
808 * requested size, skip it completely.
813 e820
.map
[i
].size
-= current_addr
- size
;
815 print_memory_map("limit_regions endfor");
818 print_memory_map("limit_regions endfunc");
822 * This function checks if the entire range <start,end> is mapped with type.
824 * Note: this function only works correct if the e820 table is sorted and
825 * not-overlapping, which is the case
828 e820_all_mapped(unsigned long s
, unsigned long e
, unsigned type
)
833 for (i
= 0; i
< e820
.nr_map
; i
++) {
834 struct e820entry
*ei
= &e820
.map
[i
];
835 if (type
&& ei
->type
!= type
)
837 /* is the region (part) in overlap with the current region ?*/
838 if (ei
->addr
>= end
|| ei
->addr
+ ei
->size
<= start
)
840 /* if the region is at the beginning of <start,end> we move
841 * start to the end of the region since it's ok until there
843 if (ei
->addr
<= start
)
844 start
= ei
->addr
+ ei
->size
;
845 /* if start is now at or beyond end, we're done, full
848 return 1; /* we're done */
853 static int __init
parse_memmap(char *arg
)
858 if (strcmp(arg
, "exactmap") == 0) {
859 #ifdef CONFIG_CRASH_DUMP
860 /* If we are doing a crash dump, we
861 * still need to know the real mem
862 * size before original memory map is
866 saved_max_pfn
= max_pfn
;
869 user_defined_memmap
= 1;
871 /* If the user specifies memory size, we
872 * limit the BIOS-provided memory map to
873 * that size. exactmap can be used to specify
874 * the exact map. mem=number can be used to
875 * trim the existing memory map.
877 unsigned long long start_at
, mem_size
;
879 mem_size
= memparse(arg
, &arg
);
881 start_at
= memparse(arg
+1, &arg
);
882 add_memory_region(start_at
, mem_size
, E820_RAM
);
883 } else if (*arg
== '#') {
884 start_at
= memparse(arg
+1, &arg
);
885 add_memory_region(start_at
, mem_size
, E820_ACPI
);
886 } else if (*arg
== '$') {
887 start_at
= memparse(arg
+1, &arg
);
888 add_memory_region(start_at
, mem_size
, E820_RESERVED
);
890 limit_regions(mem_size
);
891 user_defined_memmap
= 1;
896 early_param("memmap", parse_memmap
);