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>
13 #include <linux/suspend.h>
15 #include <asm/pgtable.h>
18 #include <asm/setup.h>
22 EXPORT_SYMBOL(efi_enabled);
26 struct change_member {
27 struct e820entry *pbios; /* pointer to original bios entry */
28 unsigned long long addr; /* address for this change point */
30 static struct change_member change_point_list[2*E820MAX] __initdata;
31 static struct change_member *change_point[2*E820MAX] __initdata;
32 static struct e820entry *overlap_list[E820MAX] __initdata;
33 static struct e820entry new_bios[E820MAX] __initdata;
34 /* For PCI or other memory-mapped resources */
35 unsigned long pci_mem_start = 0x10000000;
37 EXPORT_SYMBOL(pci_mem_start);
39 extern int user_defined_memmap;
40 struct resource data_resource = {
41 .name = "Kernel data",
44 .flags = IORESOURCE_BUSY | IORESOURCE_MEM
47 struct resource code_resource = {
48 .name = "Kernel code",
51 .flags = IORESOURCE_BUSY | IORESOURCE_MEM
54 struct resource bss_resource = {
58 .flags = IORESOURCE_BUSY | IORESOURCE_MEM
61 static struct resource system_rom_resource = {
65 .flags = IORESOURCE_BUSY | IORESOURCE_READONLY | IORESOURCE_MEM
68 static struct resource extension_rom_resource = {
69 .name = "Extension ROM",
72 .flags = IORESOURCE_BUSY | IORESOURCE_READONLY | IORESOURCE_MEM
75 static struct resource adapter_rom_resources[] = { {
76 .name = "Adapter ROM",
79 .flags = IORESOURCE_BUSY | IORESOURCE_READONLY | IORESOURCE_MEM
81 .name = "Adapter ROM",
84 .flags = IORESOURCE_BUSY | IORESOURCE_READONLY | IORESOURCE_MEM
86 .name = "Adapter ROM",
89 .flags = IORESOURCE_BUSY | IORESOURCE_READONLY | IORESOURCE_MEM
91 .name = "Adapter ROM",
94 .flags = IORESOURCE_BUSY | IORESOURCE_READONLY | IORESOURCE_MEM
96 .name = "Adapter ROM",
99 .flags = IORESOURCE_BUSY | IORESOURCE_READONLY | IORESOURCE_MEM
101 .name = "Adapter ROM",
104 .flags = IORESOURCE_BUSY | IORESOURCE_READONLY | IORESOURCE_MEM
107 static struct resource video_rom_resource = {
111 .flags = IORESOURCE_BUSY | IORESOURCE_READONLY | IORESOURCE_MEM
114 static struct resource video_ram_resource = {
115 .name = "Video RAM area",
118 .flags = IORESOURCE_BUSY | IORESOURCE_MEM
121 static struct resource standard_io_resources[] = { {
125 .flags = IORESOURCE_BUSY | IORESOURCE_IO
130 .flags = IORESOURCE_BUSY | IORESOURCE_IO
135 .flags = IORESOURCE_BUSY | IORESOURCE_IO
140 .flags = IORESOURCE_BUSY | IORESOURCE_IO
145 .flags = IORESOURCE_BUSY | IORESOURCE_IO
147 .name = "dma page reg",
150 .flags = IORESOURCE_BUSY | IORESOURCE_IO
155 .flags = IORESOURCE_BUSY | IORESOURCE_IO
160 .flags = IORESOURCE_BUSY | IORESOURCE_IO
165 .flags = IORESOURCE_BUSY | IORESOURCE_IO
168 #define ROMSIGNATURE 0xaa55
170 static int __init romsignature(const unsigned char *rom)
172 const unsigned short * const ptr = (const unsigned short *)rom;
175 return probe_kernel_address(ptr, sig) == 0 && sig == ROMSIGNATURE;
178 static int __init romchecksum(const unsigned char *rom, unsigned long length)
180 unsigned char sum, c;
182 for (sum = 0; length && probe_kernel_address(rom++, c) == 0; length--)
184 return !length && !sum;
187 static void __init probe_roms(void)
189 const unsigned char *rom;
190 unsigned long start, length, upper;
195 upper = adapter_rom_resources[0].start;
196 for (start = video_rom_resource.start; start < upper; start += 2048) {
197 rom = isa_bus_to_virt(start);
198 if (!romsignature(rom))
201 video_rom_resource.start = start;
203 if (probe_kernel_address(rom + 2, c) != 0)
206 /* 0 < length <= 0x7f * 512, historically */
209 /* if checksum okay, trust length byte */
210 if (length && romchecksum(rom, length))
211 video_rom_resource.end = start + length - 1;
213 request_resource(&iomem_resource, &video_rom_resource);
217 start = (video_rom_resource.end + 1 + 2047) & ~2047UL;
222 request_resource(&iomem_resource, &system_rom_resource);
223 upper = system_rom_resource.start;
225 /* check for extension rom (ignore length byte!) */
226 rom = isa_bus_to_virt(extension_rom_resource.start);
227 if (romsignature(rom)) {
228 length = extension_rom_resource.end - extension_rom_resource.start + 1;
229 if (romchecksum(rom, length)) {
230 request_resource(&iomem_resource, &extension_rom_resource);
231 upper = extension_rom_resource.start;
235 /* check for adapter roms on 2k boundaries */
236 for (i = 0; i < ARRAY_SIZE(adapter_rom_resources) && start < upper; start += 2048) {
237 rom = isa_bus_to_virt(start);
238 if (!romsignature(rom))
241 if (probe_kernel_address(rom + 2, c) != 0)
244 /* 0 < length <= 0x7f * 512, historically */
247 /* but accept any length that fits if checksum okay */
248 if (!length || start + length > upper || !romchecksum(rom, length))
251 adapter_rom_resources[i].start = start;
252 adapter_rom_resources[i].end = start + length - 1;
253 request_resource(&iomem_resource, &adapter_rom_resources[i]);
255 start = adapter_rom_resources[i++].end & ~2047UL;
260 * Request address space for all standard RAM and ROM resources
261 * and also for regions reported as reserved by the e820.
264 legacy_init_iomem_resources(struct resource *code_resource,
265 struct resource *data_resource,
266 struct resource *bss_resource)
271 for (i = 0; i < e820.nr_map; i++) {
272 struct resource *res;
273 #ifndef CONFIG_RESOURCES_64BIT
274 if (e820.map[i].addr + e820.map[i].size > 0x100000000ULL)
277 res = kzalloc(sizeof(struct resource), GFP_ATOMIC);
278 switch (e820.map[i].type) {
279 case E820_RAM: res->name = "System RAM"; break;
280 case E820_ACPI: res->name = "ACPI Tables"; break;
281 case E820_NVS: res->name = "ACPI Non-volatile Storage"; break;
282 default: res->name = "reserved";
284 res->start = e820.map[i].addr;
285 res->end = res->start + e820.map[i].size - 1;
286 res->flags = IORESOURCE_MEM | IORESOURCE_BUSY;
287 if (request_resource(&iomem_resource, res)) {
291 if (e820.map[i].type == E820_RAM) {
293 * We don't know which RAM region contains kernel data,
294 * so we try it repeatedly and let the resource manager
297 request_resource(res, code_resource);
298 request_resource(res, data_resource);
299 request_resource(res, bss_resource);
301 if (crashk_res.start != crashk_res.end)
302 request_resource(res, &crashk_res);
309 * Request address space for all standard resources
311 * This is called just before pcibios_init(), which is also a
312 * subsys_initcall, but is linked in later (in arch/i386/pci/common.c).
314 static int __init request_standard_resources(void)
318 printk("Setting up standard PCI resources\n");
320 efi_initialize_iomem_resources(&code_resource,
321 &data_resource, &bss_resource);
323 legacy_init_iomem_resources(&code_resource,
324 &data_resource, &bss_resource);
326 /* EFI systems may still have VGA */
327 request_resource(&iomem_resource, &video_ram_resource);
329 /* request I/O space for devices used on all i[345]86 PCs */
330 for (i = 0; i < ARRAY_SIZE(standard_io_resources); i++)
331 request_resource(&ioport_resource, &standard_io_resources[i]);
335 subsys_initcall(request_standard_resources);
337 #if defined(CONFIG_PM) && defined(CONFIG_HIBERNATION)
339 * e820_mark_nosave_regions - Find the ranges of physical addresses that do not
340 * correspond to e820 RAM areas and mark the corresponding pages as nosave for
343 * This function requires the e820 map to be sorted and without any
344 * overlapping entries and assumes the first e820 area to be RAM.
346 void __init e820_mark_nosave_regions(void)
351 pfn = PFN_DOWN(e820.map[0].addr + e820.map[0].size);
352 for (i = 1; i < e820.nr_map; i++) {
353 struct e820entry *ei = &e820.map[i];
355 if (pfn < PFN_UP(ei->addr))
356 register_nosave_region(pfn, PFN_UP(ei->addr));
358 pfn = PFN_DOWN(ei->addr + ei->size);
359 if (ei->type != E820_RAM)
360 register_nosave_region(PFN_UP(ei->addr), pfn);
362 if (pfn >= max_low_pfn)
368 void __init add_memory_region(unsigned long long start,
369 unsigned long long size, int type)
377 printk(KERN_ERR "Ooops! Too many entries in the memory map!\n");
381 e820.map[x].addr = start;
382 e820.map[x].size = size;
383 e820.map[x].type = type;
386 } /* add_memory_region */
389 * Sanitize the BIOS e820 map.
391 * Some e820 responses include overlapping entries. The following
392 * replaces the original e820 map with a new one, removing overlaps.
395 int __init sanitize_e820_map(struct e820entry * biosmap, char * pnr_map)
397 struct change_member *change_tmp;
398 unsigned long current_type, last_type;
399 unsigned long long last_addr;
400 int chgidx, still_changing;
403 int old_nr, new_nr, chg_nr;
407 Visually we're performing the following (1,2,3,4 = memory types)...
409 Sample memory map (w/overlaps):
410 ____22__________________
411 ______________________4_
412 ____1111________________
413 _44_____________________
414 11111111________________
415 ____________________33__
416 ___________44___________
417 __________33333_________
418 ______________22________
419 ___________________2222_
420 _________111111111______
421 _____________________11_
422 _________________4______
424 Sanitized equivalent (no overlap):
425 1_______________________
426 _44_____________________
427 ___1____________________
428 ____22__________________
429 ______11________________
430 _________1______________
431 __________3_____________
432 ___________44___________
433 _____________33_________
434 _______________2________
435 ________________1_______
436 _________________4______
437 ___________________2____
438 ____________________33__
439 ______________________4_
441 /* if there's only one memory region, don't bother */
448 /* bail out if we find any unreasonable addresses in bios map */
449 for (i=0; i<old_nr; i++)
450 if (biosmap[i].addr + biosmap[i].size < biosmap[i].addr) {
454 /* create pointers for initial change-point information (for sorting) */
455 for (i=0; i < 2*old_nr; i++)
456 change_point[i] = &change_point_list[i];
458 /* record all known change-points (starting and ending addresses),
459 omitting those that are for empty memory regions */
461 for (i=0; i < old_nr; i++) {
462 if (biosmap[i].size != 0) {
463 change_point[chgidx]->addr = biosmap[i].addr;
464 change_point[chgidx++]->pbios = &biosmap[i];
465 change_point[chgidx]->addr = biosmap[i].addr + biosmap[i].size;
466 change_point[chgidx++]->pbios = &biosmap[i];
469 chg_nr = chgidx; /* true number of change-points */
471 /* sort change-point list by memory addresses (low -> high) */
473 while (still_changing) {
475 for (i=1; i < chg_nr; i++) {
476 /* if <current_addr> > <last_addr>, swap */
477 /* or, if current=<start_addr> & last=<end_addr>, swap */
478 if ((change_point[i]->addr < change_point[i-1]->addr) ||
479 ((change_point[i]->addr == change_point[i-1]->addr) &&
480 (change_point[i]->addr == change_point[i]->pbios->addr) &&
481 (change_point[i-1]->addr != change_point[i-1]->pbios->addr))
484 change_tmp = change_point[i];
485 change_point[i] = change_point[i-1];
486 change_point[i-1] = change_tmp;
492 /* create a new bios memory map, removing overlaps */
493 overlap_entries=0; /* number of entries in the overlap table */
494 new_bios_entry=0; /* index for creating new bios map entries */
495 last_type = 0; /* start with undefined memory type */
496 last_addr = 0; /* start with 0 as last starting address */
497 /* loop through change-points, determining affect on the new bios map */
498 for (chgidx=0; chgidx < chg_nr; chgidx++)
500 /* keep track of all overlapping bios entries */
501 if (change_point[chgidx]->addr == change_point[chgidx]->pbios->addr)
503 /* add map entry to overlap list (> 1 entry implies an overlap) */
504 overlap_list[overlap_entries++]=change_point[chgidx]->pbios;
508 /* remove entry from list (order independent, so swap with last) */
509 for (i=0; i<overlap_entries; i++)
511 if (overlap_list[i] == change_point[chgidx]->pbios)
512 overlap_list[i] = overlap_list[overlap_entries-1];
516 /* if there are overlapping entries, decide which "type" to use */
517 /* (larger value takes precedence -- 1=usable, 2,3,4,4+=unusable) */
519 for (i=0; i<overlap_entries; i++)
520 if (overlap_list[i]->type > current_type)
521 current_type = overlap_list[i]->type;
522 /* continue building up new bios map based on this information */
523 if (current_type != last_type) {
524 if (last_type != 0) {
525 new_bios[new_bios_entry].size =
526 change_point[chgidx]->addr - last_addr;
527 /* move forward only if the new size was non-zero */
528 if (new_bios[new_bios_entry].size != 0)
529 if (++new_bios_entry >= E820MAX)
530 break; /* no more space left for new bios entries */
532 if (current_type != 0) {
533 new_bios[new_bios_entry].addr = change_point[chgidx]->addr;
534 new_bios[new_bios_entry].type = current_type;
535 last_addr=change_point[chgidx]->addr;
537 last_type = current_type;
540 new_nr = new_bios_entry; /* retain count for new bios entries */
542 /* copy new bios mapping into original location */
543 memcpy(biosmap, new_bios, new_nr*sizeof(struct e820entry));
550 * Copy the BIOS e820 map into a safe place.
552 * Sanity-check it while we're at it..
554 * If we're lucky and live on a modern system, the setup code
555 * will have given us a memory map that we can use to properly
556 * set up memory. If we aren't, we'll fake a memory map.
558 * We check to see that the memory map contains at least 2 elements
559 * before we'll use it, because the detection code in setup.S may
560 * not be perfect and most every PC known to man has two memory
561 * regions: one from 0 to 640k, and one from 1mb up. (The IBM
562 * thinkpad 560x, for example, does not cooperate with the memory
565 int __init copy_e820_map(struct e820entry * biosmap, int nr_map)
567 /* Only one memory region (or negative)? Ignore it */
572 unsigned long long start = biosmap->addr;
573 unsigned long long size = biosmap->size;
574 unsigned long long end = start + size;
575 unsigned long type = biosmap->type;
577 /* Overflow in 64 bits? Ignore the memory map. */
582 * Some BIOSes claim RAM in the 640k - 1M region.
583 * Not right. Fix it up.
585 if (type == E820_RAM) {
586 if (start < 0x100000ULL && end > 0xA0000ULL) {
587 if (start < 0xA0000ULL)
588 add_memory_region(start, 0xA0000ULL-start, type);
589 if (end <= 0x100000ULL)
595 add_memory_region(start, size, type);
596 } while (biosmap++,--nr_map);
601 * Callback for efi_memory_walk.
604 efi_find_max_pfn(unsigned long start, unsigned long end, void *arg)
606 unsigned long *max_pfn = arg, pfn;
609 pfn = PFN_UP(end -1);
617 efi_memory_present_wrapper(unsigned long start, unsigned long end, void *arg)
619 memory_present(0, PFN_UP(start), PFN_DOWN(end));
624 * Find the highest page frame number we have available
626 void __init find_max_pfn(void)
632 efi_memmap_walk(efi_find_max_pfn, &max_pfn);
633 efi_memmap_walk(efi_memory_present_wrapper, NULL);
637 for (i = 0; i < e820.nr_map; i++) {
638 unsigned long start, end;
640 if (e820.map[i].type != E820_RAM)
642 start = PFN_UP(e820.map[i].addr);
643 end = PFN_DOWN(e820.map[i].addr + e820.map[i].size);
648 memory_present(0, start, end);
653 * Free all available memory for boot time allocation. Used
654 * as a callback function by efi_memory_walk()
658 free_available_memory(unsigned long start, unsigned long end, void *arg)
660 /* check max_low_pfn */
661 if (start >= (max_low_pfn << PAGE_SHIFT))
663 if (end >= (max_low_pfn << PAGE_SHIFT))
664 end = max_low_pfn << PAGE_SHIFT;
666 free_bootmem(start, end - start);
671 * Register fully available low RAM pages with the bootmem allocator.
673 void __init register_bootmem_low_pages(unsigned long max_low_pfn)
678 efi_memmap_walk(free_available_memory, NULL);
681 for (i = 0; i < e820.nr_map; i++) {
682 unsigned long curr_pfn, last_pfn, size;
684 * Reserve usable low memory
686 if (e820.map[i].type != E820_RAM)
689 * We are rounding up the start address of usable memory:
691 curr_pfn = PFN_UP(e820.map[i].addr);
692 if (curr_pfn >= max_low_pfn)
695 * ... and at the end of the usable range downwards:
697 last_pfn = PFN_DOWN(e820.map[i].addr + e820.map[i].size);
699 if (last_pfn > max_low_pfn)
700 last_pfn = max_low_pfn;
703 * .. finally, did all the rounding and playing
704 * around just make the area go away?
706 if (last_pfn <= curr_pfn)
709 size = last_pfn - curr_pfn;
710 free_bootmem(PFN_PHYS(curr_pfn), PFN_PHYS(size));
714 void __init e820_register_memory(void)
716 unsigned long gapstart, gapsize, round;
717 unsigned long long last;
721 * Search for the biggest gap in the low 32 bits of the e820
724 last = 0x100000000ull;
725 gapstart = 0x10000000;
729 unsigned long long start = e820.map[i].addr;
730 unsigned long long end = start + e820.map[i].size;
733 * Since "last" is at most 4GB, we know we'll
734 * fit in 32 bits if this condition is true
737 unsigned long gap = last - end;
749 * See how much we want to round up: start off with
750 * rounding to the next 1MB area.
753 while ((gapsize >> 4) > round)
755 /* Fun with two's complement */
756 pci_mem_start = (gapstart + round) & -round;
758 printk("Allocating PCI resources starting at %08lx (gap: %08lx:%08lx)\n",
759 pci_mem_start, gapstart, gapsize);
762 void __init print_memory_map(char *who)
766 for (i = 0; i < e820.nr_map; i++) {
767 printk(" %s: %016Lx - %016Lx ", who,
769 e820.map[i].addr + e820.map[i].size);
770 switch (e820.map[i].type) {
771 case E820_RAM: printk("(usable)\n");
774 printk("(reserved)\n");
777 printk("(ACPI data)\n");
780 printk("(ACPI NVS)\n");
782 default: printk("type %u\n", e820.map[i].type);
788 static __init __always_inline void efi_limit_regions(unsigned long long size)
790 unsigned long long current_addr = 0;
791 efi_memory_desc_t *md, *next_md;
797 for (p = p1, i = 0; p < memmap.map_end; p += memmap.desc_size, i++) {
800 current_addr = md->phys_addr +
801 PFN_PHYS(md->num_pages);
802 if (is_available_memory(md)) {
803 if (md->phys_addr >= size) continue;
804 memcpy(next_md, md, memmap.desc_size);
805 if (current_addr >= size) {
806 next_md->num_pages -=
807 PFN_UP(current_addr-size);
809 p1 += memmap.desc_size;
812 } else if ((md->attribute & EFI_MEMORY_RUNTIME) ==
813 EFI_MEMORY_RUNTIME) {
814 /* In order to make runtime services
815 * available we have to include runtime
816 * memory regions in memory map */
817 memcpy(next_md, md, memmap.desc_size);
818 p1 += memmap.desc_size;
824 memmap.map_end = memmap.map +
825 (memmap.nr_map * memmap.desc_size);
828 void __init limit_regions(unsigned long long size)
830 unsigned long long current_addr;
833 print_memory_map("limit_regions start");
835 efi_limit_regions(size);
838 for (i = 0; i < e820.nr_map; i++) {
839 current_addr = e820.map[i].addr + e820.map[i].size;
840 if (current_addr < size)
843 if (e820.map[i].type != E820_RAM)
846 if (e820.map[i].addr >= size) {
848 * This region starts past the end of the
849 * requested size, skip it completely.
854 e820.map[i].size -= current_addr - size;
856 print_memory_map("limit_regions endfor");
859 print_memory_map("limit_regions endfunc");
863 * This function checks if any part of the range <start,end> is mapped
867 e820_any_mapped(u64 start, u64 end, unsigned type)
870 for (i = 0; i < e820.nr_map; i++) {
871 const struct e820entry *ei = &e820.map[i];
872 if (type && ei->type != type)
874 if (ei->addr >= end || ei->addr + ei->size <= start)
880 EXPORT_SYMBOL_GPL(e820_any_mapped);
883 * This function checks if the entire range <start,end> is mapped with type.
885 * Note: this function only works correct if the e820 table is sorted and
886 * not-overlapping, which is the case
889 e820_all_mapped(unsigned long s, unsigned long e, unsigned type)
894 for (i = 0; i < e820.nr_map; i++) {
895 struct e820entry *ei = &e820.map[i];
896 if (type && ei->type != type)
898 /* is the region (part) in overlap with the current region ?*/
899 if (ei->addr >= end || ei->addr + ei->size <= start)
901 /* if the region is at the beginning of <start,end> we move
902 * start to the end of the region since it's ok until there
904 if (ei->addr <= start)
905 start = ei->addr + ei->size;
906 /* if start is now at or beyond end, we're done, full
909 return 1; /* we're done */
914 static int __init parse_memmap(char *arg)
919 if (strcmp(arg, "exactmap") == 0) {
920 #ifdef CONFIG_CRASH_DUMP
921 /* If we are doing a crash dump, we
922 * still need to know the real mem
923 * size before original memory map is
927 saved_max_pfn = max_pfn;
930 user_defined_memmap = 1;
932 /* If the user specifies memory size, we
933 * limit the BIOS-provided memory map to
934 * that size. exactmap can be used to specify
935 * the exact map. mem=number can be used to
936 * trim the existing memory map.
938 unsigned long long start_at, mem_size;
940 mem_size = memparse(arg, &arg);
942 start_at = memparse(arg+1, &arg);
943 add_memory_region(start_at, mem_size, E820_RAM);
944 } else if (*arg == '#') {
945 start_at = memparse(arg+1, &arg);
946 add_memory_region(start_at, mem_size, E820_ACPI);
947 } else if (*arg == '$') {
948 start_at = memparse(arg+1, &arg);
949 add_memory_region(start_at, mem_size, E820_RESERVED);
951 limit_regions(mem_size);
952 user_defined_memmap = 1;
957 early_param("memmap", parse_memmap);