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>
20 EXPORT_SYMBOL(efi_enabled
);
24 struct change_member
{
25 struct e820entry
*pbios
; /* pointer to original bios entry */
26 unsigned long long addr
; /* address for this change point */
28 static struct change_member change_point_list
[2*E820MAX
] __initdata
;
29 static struct change_member
*change_point
[2*E820MAX
] __initdata
;
30 static struct e820entry
*overlap_list
[E820MAX
] __initdata
;
31 static struct e820entry new_bios
[E820MAX
] __initdata
;
32 /* For PCI or other memory-mapped resources */
33 unsigned long pci_mem_start
= 0x10000000;
35 EXPORT_SYMBOL(pci_mem_start
);
37 extern int user_defined_memmap
;
38 struct resource data_resource
= {
39 .name
= "Kernel data",
42 .flags
= IORESOURCE_BUSY
| IORESOURCE_MEM
45 struct resource code_resource
= {
46 .name
= "Kernel code",
49 .flags
= IORESOURCE_BUSY
| IORESOURCE_MEM
52 static struct resource system_rom_resource
= {
56 .flags
= IORESOURCE_BUSY
| IORESOURCE_READONLY
| IORESOURCE_MEM
59 static struct resource extension_rom_resource
= {
60 .name
= "Extension ROM",
63 .flags
= IORESOURCE_BUSY
| IORESOURCE_READONLY
| IORESOURCE_MEM
66 static struct resource adapter_rom_resources
[] = { {
67 .name
= "Adapter ROM",
70 .flags
= IORESOURCE_BUSY
| IORESOURCE_READONLY
| IORESOURCE_MEM
72 .name
= "Adapter ROM",
75 .flags
= IORESOURCE_BUSY
| IORESOURCE_READONLY
| IORESOURCE_MEM
77 .name
= "Adapter ROM",
80 .flags
= IORESOURCE_BUSY
| IORESOURCE_READONLY
| IORESOURCE_MEM
82 .name
= "Adapter ROM",
85 .flags
= IORESOURCE_BUSY
| IORESOURCE_READONLY
| IORESOURCE_MEM
87 .name
= "Adapter ROM",
90 .flags
= IORESOURCE_BUSY
| IORESOURCE_READONLY
| IORESOURCE_MEM
92 .name
= "Adapter ROM",
95 .flags
= IORESOURCE_BUSY
| IORESOURCE_READONLY
| IORESOURCE_MEM
98 static struct resource video_rom_resource
= {
102 .flags
= IORESOURCE_BUSY
| IORESOURCE_READONLY
| IORESOURCE_MEM
105 static struct resource video_ram_resource
= {
106 .name
= "Video RAM area",
109 .flags
= IORESOURCE_BUSY
| IORESOURCE_MEM
112 static struct resource standard_io_resources
[] = { {
116 .flags
= IORESOURCE_BUSY
| IORESOURCE_IO
121 .flags
= IORESOURCE_BUSY
| IORESOURCE_IO
126 .flags
= IORESOURCE_BUSY
| IORESOURCE_IO
131 .flags
= IORESOURCE_BUSY
| IORESOURCE_IO
136 .flags
= IORESOURCE_BUSY
| IORESOURCE_IO
138 .name
= "dma page reg",
141 .flags
= IORESOURCE_BUSY
| IORESOURCE_IO
146 .flags
= IORESOURCE_BUSY
| IORESOURCE_IO
151 .flags
= IORESOURCE_BUSY
| IORESOURCE_IO
156 .flags
= IORESOURCE_BUSY
| IORESOURCE_IO
159 static int romsignature(const unsigned char *x
)
163 if (probe_kernel_address((const unsigned short *)x
, sig
) == 0)
164 ret
= (sig
== 0xaa55);
168 static int __init
romchecksum(unsigned char *rom
, unsigned long length
)
170 unsigned char *p
, sum
= 0;
172 for (p
= rom
; p
< rom
+ length
; p
++)
177 static void __init
probe_roms(void)
179 unsigned long start
, length
, upper
;
184 upper
= adapter_rom_resources
[0].start
;
185 for (start
= video_rom_resource
.start
; start
< upper
; start
+= 2048) {
186 rom
= isa_bus_to_virt(start
);
187 if (!romsignature(rom
))
190 video_rom_resource
.start
= start
;
192 /* 0 < length <= 0x7f * 512, historically */
193 length
= rom
[2] * 512;
195 /* if checksum okay, trust length byte */
196 if (length
&& romchecksum(rom
, length
))
197 video_rom_resource
.end
= start
+ length
- 1;
199 request_resource(&iomem_resource
, &video_rom_resource
);
203 start
= (video_rom_resource
.end
+ 1 + 2047) & ~2047UL;
208 request_resource(&iomem_resource
, &system_rom_resource
);
209 upper
= system_rom_resource
.start
;
211 /* check for extension rom (ignore length byte!) */
212 rom
= isa_bus_to_virt(extension_rom_resource
.start
);
213 if (romsignature(rom
)) {
214 length
= extension_rom_resource
.end
- extension_rom_resource
.start
+ 1;
215 if (romchecksum(rom
, length
)) {
216 request_resource(&iomem_resource
, &extension_rom_resource
);
217 upper
= extension_rom_resource
.start
;
221 /* check for adapter roms on 2k boundaries */
222 for (i
= 0; i
< ARRAY_SIZE(adapter_rom_resources
) && start
< upper
; start
+= 2048) {
223 rom
= isa_bus_to_virt(start
);
224 if (!romsignature(rom
))
227 /* 0 < length <= 0x7f * 512, historically */
228 length
= rom
[2] * 512;
230 /* but accept any length that fits if checksum okay */
231 if (!length
|| start
+ length
> upper
|| !romchecksum(rom
, length
))
234 adapter_rom_resources
[i
].start
= start
;
235 adapter_rom_resources
[i
].end
= start
+ length
- 1;
236 request_resource(&iomem_resource
, &adapter_rom_resources
[i
]);
238 start
= adapter_rom_resources
[i
++].end
& ~2047UL;
243 * Request address space for all standard RAM and ROM resources
244 * and also for regions reported as reserved by the e820.
247 legacy_init_iomem_resources(struct resource
*code_resource
, struct resource
*data_resource
)
252 for (i
= 0; i
< e820
.nr_map
; i
++) {
253 struct resource
*res
;
254 #ifndef CONFIG_RESOURCES_64BIT
255 if (e820
.map
[i
].addr
+ e820
.map
[i
].size
> 0x100000000ULL
)
258 res
= kzalloc(sizeof(struct resource
), GFP_ATOMIC
);
259 switch (e820
.map
[i
].type
) {
260 case E820_RAM
: res
->name
= "System RAM"; break;
261 case E820_ACPI
: res
->name
= "ACPI Tables"; break;
262 case E820_NVS
: res
->name
= "ACPI Non-volatile Storage"; break;
263 default: res
->name
= "reserved";
265 res
->start
= e820
.map
[i
].addr
;
266 res
->end
= res
->start
+ e820
.map
[i
].size
- 1;
267 res
->flags
= IORESOURCE_MEM
| IORESOURCE_BUSY
;
268 if (request_resource(&iomem_resource
, res
)) {
272 if (e820
.map
[i
].type
== E820_RAM
) {
274 * We don't know which RAM region contains kernel data,
275 * so we try it repeatedly and let the resource manager
278 request_resource(res
, code_resource
);
279 request_resource(res
, data_resource
);
281 request_resource(res
, &crashk_res
);
288 * Request address space for all standard resources
290 * This is called just before pcibios_init(), which is also a
291 * subsys_initcall, but is linked in later (in arch/i386/pci/common.c).
293 static int __init
request_standard_resources(void)
297 printk("Setting up standard PCI resources\n");
299 efi_initialize_iomem_resources(&code_resource
, &data_resource
);
301 legacy_init_iomem_resources(&code_resource
, &data_resource
);
303 /* EFI systems may still have VGA */
304 request_resource(&iomem_resource
, &video_ram_resource
);
306 /* request I/O space for devices used on all i[345]86 PCs */
307 for (i
= 0; i
< ARRAY_SIZE(standard_io_resources
); i
++)
308 request_resource(&ioport_resource
, &standard_io_resources
[i
]);
312 subsys_initcall(request_standard_resources
);
314 void __init
add_memory_region(unsigned long long start
,
315 unsigned long long size
, int type
)
323 printk(KERN_ERR
"Ooops! Too many entries in the memory map!\n");
327 e820
.map
[x
].addr
= start
;
328 e820
.map
[x
].size
= size
;
329 e820
.map
[x
].type
= type
;
332 } /* add_memory_region */
335 * Sanitize the BIOS e820 map.
337 * Some e820 responses include overlapping entries. The following
338 * replaces the original e820 map with a new one, removing overlaps.
341 int __init
sanitize_e820_map(struct e820entry
* biosmap
, char * pnr_map
)
343 struct change_member
*change_tmp
;
344 unsigned long current_type
, last_type
;
345 unsigned long long last_addr
;
346 int chgidx
, still_changing
;
349 int old_nr
, new_nr
, chg_nr
;
353 Visually we're performing the following (1,2,3,4 = memory types)...
355 Sample memory map (w/overlaps):
356 ____22__________________
357 ______________________4_
358 ____1111________________
359 _44_____________________
360 11111111________________
361 ____________________33__
362 ___________44___________
363 __________33333_________
364 ______________22________
365 ___________________2222_
366 _________111111111______
367 _____________________11_
368 _________________4______
370 Sanitized equivalent (no overlap):
371 1_______________________
372 _44_____________________
373 ___1____________________
374 ____22__________________
375 ______11________________
376 _________1______________
377 __________3_____________
378 ___________44___________
379 _____________33_________
380 _______________2________
381 ________________1_______
382 _________________4______
383 ___________________2____
384 ____________________33__
385 ______________________4_
387 printk("sanitize start\n");
388 /* if there's only one memory region, don't bother */
390 printk("sanitize bail 0\n");
396 /* bail out if we find any unreasonable addresses in bios map */
397 for (i
=0; i
<old_nr
; i
++)
398 if (biosmap
[i
].addr
+ biosmap
[i
].size
< biosmap
[i
].addr
) {
399 printk("sanitize bail 1\n");
403 /* create pointers for initial change-point information (for sorting) */
404 for (i
=0; i
< 2*old_nr
; i
++)
405 change_point
[i
] = &change_point_list
[i
];
407 /* record all known change-points (starting and ending addresses),
408 omitting those that are for empty memory regions */
410 for (i
=0; i
< old_nr
; i
++) {
411 if (biosmap
[i
].size
!= 0) {
412 change_point
[chgidx
]->addr
= biosmap
[i
].addr
;
413 change_point
[chgidx
++]->pbios
= &biosmap
[i
];
414 change_point
[chgidx
]->addr
= biosmap
[i
].addr
+ biosmap
[i
].size
;
415 change_point
[chgidx
++]->pbios
= &biosmap
[i
];
418 chg_nr
= chgidx
; /* true number of change-points */
420 /* sort change-point list by memory addresses (low -> high) */
422 while (still_changing
) {
424 for (i
=1; i
< chg_nr
; i
++) {
425 /* if <current_addr> > <last_addr>, swap */
426 /* or, if current=<start_addr> & last=<end_addr>, swap */
427 if ((change_point
[i
]->addr
< change_point
[i
-1]->addr
) ||
428 ((change_point
[i
]->addr
== change_point
[i
-1]->addr
) &&
429 (change_point
[i
]->addr
== change_point
[i
]->pbios
->addr
) &&
430 (change_point
[i
-1]->addr
!= change_point
[i
-1]->pbios
->addr
))
433 change_tmp
= change_point
[i
];
434 change_point
[i
] = change_point
[i
-1];
435 change_point
[i
-1] = change_tmp
;
441 /* create a new bios memory map, removing overlaps */
442 overlap_entries
=0; /* number of entries in the overlap table */
443 new_bios_entry
=0; /* index for creating new bios map entries */
444 last_type
= 0; /* start with undefined memory type */
445 last_addr
= 0; /* start with 0 as last starting address */
446 /* loop through change-points, determining affect on the new bios map */
447 for (chgidx
=0; chgidx
< chg_nr
; chgidx
++)
449 /* keep track of all overlapping bios entries */
450 if (change_point
[chgidx
]->addr
== change_point
[chgidx
]->pbios
->addr
)
452 /* add map entry to overlap list (> 1 entry implies an overlap) */
453 overlap_list
[overlap_entries
++]=change_point
[chgidx
]->pbios
;
457 /* remove entry from list (order independent, so swap with last) */
458 for (i
=0; i
<overlap_entries
; i
++)
460 if (overlap_list
[i
] == change_point
[chgidx
]->pbios
)
461 overlap_list
[i
] = overlap_list
[overlap_entries
-1];
465 /* if there are overlapping entries, decide which "type" to use */
466 /* (larger value takes precedence -- 1=usable, 2,3,4,4+=unusable) */
468 for (i
=0; i
<overlap_entries
; i
++)
469 if (overlap_list
[i
]->type
> current_type
)
470 current_type
= overlap_list
[i
]->type
;
471 /* continue building up new bios map based on this information */
472 if (current_type
!= last_type
) {
473 if (last_type
!= 0) {
474 new_bios
[new_bios_entry
].size
=
475 change_point
[chgidx
]->addr
- last_addr
;
476 /* move forward only if the new size was non-zero */
477 if (new_bios
[new_bios_entry
].size
!= 0)
478 if (++new_bios_entry
>= E820MAX
)
479 break; /* no more space left for new bios entries */
481 if (current_type
!= 0) {
482 new_bios
[new_bios_entry
].addr
= change_point
[chgidx
]->addr
;
483 new_bios
[new_bios_entry
].type
= current_type
;
484 last_addr
=change_point
[chgidx
]->addr
;
486 last_type
= current_type
;
489 new_nr
= new_bios_entry
; /* retain count for new bios entries */
491 /* copy new bios mapping into original location */
492 memcpy(biosmap
, new_bios
, new_nr
*sizeof(struct e820entry
));
495 printk("sanitize end\n");
500 * Copy the BIOS e820 map into a safe place.
502 * Sanity-check it while we're at it..
504 * If we're lucky and live on a modern system, the setup code
505 * will have given us a memory map that we can use to properly
506 * set up memory. If we aren't, we'll fake a memory map.
508 * We check to see that the memory map contains at least 2 elements
509 * before we'll use it, because the detection code in setup.S may
510 * not be perfect and most every PC known to man has two memory
511 * regions: one from 0 to 640k, and one from 1mb up. (The IBM
512 * thinkpad 560x, for example, does not cooperate with the memory
515 int __init
copy_e820_map(struct e820entry
* biosmap
, int nr_map
)
517 /* Only one memory region (or negative)? Ignore it */
522 unsigned long long start
= biosmap
->addr
;
523 unsigned long long size
= biosmap
->size
;
524 unsigned long long end
= start
+ size
;
525 unsigned long type
= biosmap
->type
;
526 printk("copy_e820_map() start: %016Lx size: %016Lx end: %016Lx type: %ld\n", start
, size
, end
, type
);
528 /* Overflow in 64 bits? Ignore the memory map. */
533 * Some BIOSes claim RAM in the 640k - 1M region.
534 * Not right. Fix it up.
536 if (type
== E820_RAM
) {
537 printk("copy_e820_map() type is E820_RAM\n");
538 if (start
< 0x100000ULL
&& end
> 0xA0000ULL
) {
539 printk("copy_e820_map() lies in range...\n");
540 if (start
< 0xA0000ULL
) {
541 printk("copy_e820_map() start < 0xA0000ULL\n");
542 add_memory_region(start
, 0xA0000ULL
-start
, type
);
544 if (end
<= 0x100000ULL
) {
545 printk("copy_e820_map() end <= 0x100000ULL\n");
552 add_memory_region(start
, size
, type
);
553 } while (biosmap
++,--nr_map
);
558 * Callback for efi_memory_walk.
561 efi_find_max_pfn(unsigned long start
, unsigned long end
, void *arg
)
563 unsigned long *max_pfn
= arg
, pfn
;
566 pfn
= PFN_UP(end
-1);
574 efi_memory_present_wrapper(unsigned long start
, unsigned long end
, void *arg
)
576 memory_present(0, PFN_UP(start
), PFN_DOWN(end
));
581 * Find the highest page frame number we have available
583 void __init
find_max_pfn(void)
589 efi_memmap_walk(efi_find_max_pfn
, &max_pfn
);
590 efi_memmap_walk(efi_memory_present_wrapper
, NULL
);
594 for (i
= 0; i
< e820
.nr_map
; i
++) {
595 unsigned long start
, end
;
597 if (e820
.map
[i
].type
!= E820_RAM
)
599 start
= PFN_UP(e820
.map
[i
].addr
);
600 end
= PFN_DOWN(e820
.map
[i
].addr
+ e820
.map
[i
].size
);
605 memory_present(0, start
, end
);
610 * Free all available memory for boot time allocation. Used
611 * as a callback function by efi_memory_walk()
615 free_available_memory(unsigned long start
, unsigned long end
, void *arg
)
617 /* check max_low_pfn */
618 if (start
>= (max_low_pfn
<< PAGE_SHIFT
))
620 if (end
>= (max_low_pfn
<< PAGE_SHIFT
))
621 end
= max_low_pfn
<< PAGE_SHIFT
;
623 free_bootmem(start
, end
- start
);
628 * Register fully available low RAM pages with the bootmem allocator.
630 void __init
register_bootmem_low_pages(unsigned long max_low_pfn
)
635 efi_memmap_walk(free_available_memory
, NULL
);
638 for (i
= 0; i
< e820
.nr_map
; i
++) {
639 unsigned long curr_pfn
, last_pfn
, size
;
641 * Reserve usable low memory
643 if (e820
.map
[i
].type
!= E820_RAM
)
646 * We are rounding up the start address of usable memory:
648 curr_pfn
= PFN_UP(e820
.map
[i
].addr
);
649 if (curr_pfn
>= max_low_pfn
)
652 * ... and at the end of the usable range downwards:
654 last_pfn
= PFN_DOWN(e820
.map
[i
].addr
+ e820
.map
[i
].size
);
656 if (last_pfn
> max_low_pfn
)
657 last_pfn
= max_low_pfn
;
660 * .. finally, did all the rounding and playing
661 * around just make the area go away?
663 if (last_pfn
<= curr_pfn
)
666 size
= last_pfn
- curr_pfn
;
667 free_bootmem(PFN_PHYS(curr_pfn
), PFN_PHYS(size
));
671 void __init
e820_register_memory(void)
673 unsigned long gapstart
, gapsize
, round
;
674 unsigned long long last
;
678 * Search for the bigest gap in the low 32 bits of the e820
681 last
= 0x100000000ull
;
682 gapstart
= 0x10000000;
686 unsigned long long start
= e820
.map
[i
].addr
;
687 unsigned long long end
= start
+ e820
.map
[i
].size
;
690 * Since "last" is at most 4GB, we know we'll
691 * fit in 32 bits if this condition is true
694 unsigned long gap
= last
- end
;
706 * See how much we want to round up: start off with
707 * rounding to the next 1MB area.
710 while ((gapsize
>> 4) > round
)
712 /* Fun with two's complement */
713 pci_mem_start
= (gapstart
+ round
) & -round
;
715 printk("Allocating PCI resources starting at %08lx (gap: %08lx:%08lx)\n",
716 pci_mem_start
, gapstart
, gapsize
);
719 void __init
print_memory_map(char *who
)
723 for (i
= 0; i
< e820
.nr_map
; i
++) {
724 printk(" %s: %016Lx - %016Lx ", who
,
726 e820
.map
[i
].addr
+ e820
.map
[i
].size
);
727 switch (e820
.map
[i
].type
) {
728 case E820_RAM
: printk("(usable)\n");
731 printk("(reserved)\n");
734 printk("(ACPI data)\n");
737 printk("(ACPI NVS)\n");
739 default: printk("type %lu\n", e820
.map
[i
].type
);
745 static __init __always_inline
void efi_limit_regions(unsigned long long size
)
747 unsigned long long current_addr
= 0;
748 efi_memory_desc_t
*md
, *next_md
;
754 for (p
= p1
, i
= 0; p
< memmap
.map_end
; p
+= memmap
.desc_size
, i
++) {
757 current_addr
= md
->phys_addr
+
758 PFN_PHYS(md
->num_pages
);
759 if (is_available_memory(md
)) {
760 if (md
->phys_addr
>= size
) continue;
761 memcpy(next_md
, md
, memmap
.desc_size
);
762 if (current_addr
>= size
) {
763 next_md
->num_pages
-=
764 PFN_UP(current_addr
-size
);
766 p1
+= memmap
.desc_size
;
769 } else if ((md
->attribute
& EFI_MEMORY_RUNTIME
) ==
770 EFI_MEMORY_RUNTIME
) {
771 /* In order to make runtime services
772 * available we have to include runtime
773 * memory regions in memory map */
774 memcpy(next_md
, md
, memmap
.desc_size
);
775 p1
+= memmap
.desc_size
;
781 memmap
.map_end
= memmap
.map
+
782 (memmap
.nr_map
* memmap
.desc_size
);
785 void __init
limit_regions(unsigned long long size
)
787 unsigned long long current_addr
;
790 print_memory_map("limit_regions start");
792 efi_limit_regions(size
);
795 for (i
= 0; i
< e820
.nr_map
; i
++) {
796 current_addr
= e820
.map
[i
].addr
+ e820
.map
[i
].size
;
797 if (current_addr
< size
)
800 if (e820
.map
[i
].type
!= E820_RAM
)
803 if (e820
.map
[i
].addr
>= size
) {
805 * This region starts past the end of the
806 * requested size, skip it completely.
811 e820
.map
[i
].size
-= current_addr
- size
;
813 print_memory_map("limit_regions endfor");
816 print_memory_map("limit_regions endfunc");
820 * This function checks if the entire range <start,end> is mapped with type.
822 * Note: this function only works correct if the e820 table is sorted and
823 * not-overlapping, which is the case
826 e820_all_mapped(unsigned long s
, unsigned long e
, unsigned type
)
831 for (i
= 0; i
< e820
.nr_map
; i
++) {
832 struct e820entry
*ei
= &e820
.map
[i
];
833 if (type
&& ei
->type
!= type
)
835 /* is the region (part) in overlap with the current region ?*/
836 if (ei
->addr
>= end
|| ei
->addr
+ ei
->size
<= start
)
838 /* if the region is at the beginning of <start,end> we move
839 * start to the end of the region since it's ok until there
841 if (ei
->addr
<= start
)
842 start
= ei
->addr
+ ei
->size
;
843 /* if start is now at or beyond end, we're done, full
846 return 1; /* we're done */
851 static int __init
parse_memmap(char *arg
)
856 if (strcmp(arg
, "exactmap") == 0) {
857 #ifdef CONFIG_CRASH_DUMP
858 /* If we are doing a crash dump, we
859 * still need to know the real mem
860 * size before original memory map is
864 saved_max_pfn
= max_pfn
;
867 user_defined_memmap
= 1;
869 /* If the user specifies memory size, we
870 * limit the BIOS-provided memory map to
871 * that size. exactmap can be used to specify
872 * the exact map. mem=number can be used to
873 * trim the existing memory map.
875 unsigned long long start_at
, mem_size
;
877 mem_size
= memparse(arg
, &arg
);
879 start_at
= memparse(arg
+1, &arg
);
880 add_memory_region(start_at
, mem_size
, E820_RAM
);
881 } else if (*arg
== '#') {
882 start_at
= memparse(arg
+1, &arg
);
883 add_memory_region(start_at
, mem_size
, E820_ACPI
);
884 } else if (*arg
== '$') {
885 start_at
= memparse(arg
+1, &arg
);
886 add_memory_region(start_at
, mem_size
, E820_RESERVED
);
888 limit_regions(mem_size
);
889 user_defined_memmap
= 1;
894 early_param("memmap", parse_memmap
);