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 static struct resource system_rom_resource = {
58 .flags = IORESOURCE_BUSY | IORESOURCE_READONLY | IORESOURCE_MEM
61 static struct resource extension_rom_resource = {
62 .name = "Extension ROM",
65 .flags = IORESOURCE_BUSY | IORESOURCE_READONLY | IORESOURCE_MEM
68 static struct resource adapter_rom_resources[] = { {
69 .name = "Adapter ROM",
72 .flags = IORESOURCE_BUSY | IORESOURCE_READONLY | IORESOURCE_MEM
74 .name = "Adapter ROM",
77 .flags = IORESOURCE_BUSY | IORESOURCE_READONLY | IORESOURCE_MEM
79 .name = "Adapter ROM",
82 .flags = IORESOURCE_BUSY | IORESOURCE_READONLY | IORESOURCE_MEM
84 .name = "Adapter ROM",
87 .flags = IORESOURCE_BUSY | IORESOURCE_READONLY | IORESOURCE_MEM
89 .name = "Adapter ROM",
92 .flags = IORESOURCE_BUSY | IORESOURCE_READONLY | IORESOURCE_MEM
94 .name = "Adapter ROM",
97 .flags = IORESOURCE_BUSY | IORESOURCE_READONLY | IORESOURCE_MEM
100 static struct resource video_rom_resource = {
104 .flags = IORESOURCE_BUSY | IORESOURCE_READONLY | IORESOURCE_MEM
107 static struct resource video_ram_resource = {
108 .name = "Video RAM area",
111 .flags = IORESOURCE_BUSY | IORESOURCE_MEM
114 static struct resource standard_io_resources[] = { {
118 .flags = IORESOURCE_BUSY | IORESOURCE_IO
123 .flags = IORESOURCE_BUSY | IORESOURCE_IO
128 .flags = IORESOURCE_BUSY | IORESOURCE_IO
133 .flags = IORESOURCE_BUSY | IORESOURCE_IO
138 .flags = IORESOURCE_BUSY | IORESOURCE_IO
140 .name = "dma page reg",
143 .flags = IORESOURCE_BUSY | IORESOURCE_IO
148 .flags = IORESOURCE_BUSY | IORESOURCE_IO
153 .flags = IORESOURCE_BUSY | IORESOURCE_IO
158 .flags = IORESOURCE_BUSY | IORESOURCE_IO
161 #define ROMSIGNATURE 0xaa55
163 static int __init romsignature(const unsigned char *rom)
165 const unsigned short * const ptr = (const unsigned short *)rom;
168 return probe_kernel_address(ptr, sig) == 0 && sig == ROMSIGNATURE;
171 static int __init romchecksum(const unsigned char *rom, unsigned long length)
173 unsigned char sum, c;
175 for (sum = 0; length && probe_kernel_address(rom++, c) == 0; length--)
177 return !length && !sum;
180 static void __init probe_roms(void)
182 const unsigned char *rom;
183 unsigned long start, length, upper;
188 upper = adapter_rom_resources[0].start;
189 for (start = video_rom_resource.start; start < upper; start += 2048) {
190 rom = isa_bus_to_virt(start);
191 if (!romsignature(rom))
194 video_rom_resource.start = start;
196 if (probe_kernel_address(rom + 2, c) != 0)
199 /* 0 < length <= 0x7f * 512, historically */
202 /* if checksum okay, trust length byte */
203 if (length && romchecksum(rom, length))
204 video_rom_resource.end = start + length - 1;
206 request_resource(&iomem_resource, &video_rom_resource);
210 start = (video_rom_resource.end + 1 + 2047) & ~2047UL;
215 request_resource(&iomem_resource, &system_rom_resource);
216 upper = system_rom_resource.start;
218 /* check for extension rom (ignore length byte!) */
219 rom = isa_bus_to_virt(extension_rom_resource.start);
220 if (romsignature(rom)) {
221 length = extension_rom_resource.end - extension_rom_resource.start + 1;
222 if (romchecksum(rom, length)) {
223 request_resource(&iomem_resource, &extension_rom_resource);
224 upper = extension_rom_resource.start;
228 /* check for adapter roms on 2k boundaries */
229 for (i = 0; i < ARRAY_SIZE(adapter_rom_resources) && start < upper; start += 2048) {
230 rom = isa_bus_to_virt(start);
231 if (!romsignature(rom))
234 if (probe_kernel_address(rom + 2, c) != 0)
237 /* 0 < length <= 0x7f * 512, historically */
240 /* but accept any length that fits if checksum okay */
241 if (!length || start + length > upper || !romchecksum(rom, length))
244 adapter_rom_resources[i].start = start;
245 adapter_rom_resources[i].end = start + length - 1;
246 request_resource(&iomem_resource, &adapter_rom_resources[i]);
248 start = adapter_rom_resources[i++].end & ~2047UL;
253 * Request address space for all standard RAM and ROM resources
254 * and also for regions reported as reserved by the e820.
257 legacy_init_iomem_resources(struct resource *code_resource, struct resource *data_resource)
262 for (i = 0; i < e820.nr_map; i++) {
263 struct resource *res;
264 #ifndef CONFIG_RESOURCES_64BIT
265 if (e820.map[i].addr + e820.map[i].size > 0x100000000ULL)
268 res = kzalloc(sizeof(struct resource), GFP_ATOMIC);
269 switch (e820.map[i].type) {
270 case E820_RAM: res->name = "System RAM"; break;
271 case E820_ACPI: res->name = "ACPI Tables"; break;
272 case E820_NVS: res->name = "ACPI Non-volatile Storage"; break;
273 default: res->name = "reserved";
275 res->start = e820.map[i].addr;
276 res->end = res->start + e820.map[i].size - 1;
277 res->flags = IORESOURCE_MEM | IORESOURCE_BUSY;
278 if (request_resource(&iomem_resource, res)) {
282 if (e820.map[i].type == E820_RAM) {
284 * We don't know which RAM region contains kernel data,
285 * so we try it repeatedly and let the resource manager
288 request_resource(res, code_resource);
289 request_resource(res, data_resource);
291 request_resource(res, &crashk_res);
298 * Request address space for all standard resources
300 * This is called just before pcibios_init(), which is also a
301 * subsys_initcall, but is linked in later (in arch/i386/pci/common.c).
303 static int __init request_standard_resources(void)
307 printk("Setting up standard PCI resources\n");
309 efi_initialize_iomem_resources(&code_resource, &data_resource);
311 legacy_init_iomem_resources(&code_resource, &data_resource);
313 /* EFI systems may still have VGA */
314 request_resource(&iomem_resource, &video_ram_resource);
316 /* request I/O space for devices used on all i[345]86 PCs */
317 for (i = 0; i < ARRAY_SIZE(standard_io_resources); i++)
318 request_resource(&ioport_resource, &standard_io_resources[i]);
322 subsys_initcall(request_standard_resources);
324 #if defined(CONFIG_PM) && defined(CONFIG_HIBERNATION)
326 * e820_mark_nosave_regions - Find the ranges of physical addresses that do not
327 * correspond to e820 RAM areas and mark the corresponding pages as nosave for
330 * This function requires the e820 map to be sorted and without any
331 * overlapping entries and assumes the first e820 area to be RAM.
333 void __init e820_mark_nosave_regions(void)
338 pfn = PFN_DOWN(e820.map[0].addr + e820.map[0].size);
339 for (i = 1; i < e820.nr_map; i++) {
340 struct e820entry *ei = &e820.map[i];
342 if (pfn < PFN_UP(ei->addr))
343 register_nosave_region(pfn, PFN_UP(ei->addr));
345 pfn = PFN_DOWN(ei->addr + ei->size);
346 if (ei->type != E820_RAM)
347 register_nosave_region(PFN_UP(ei->addr), pfn);
349 if (pfn >= max_low_pfn)
355 void __init add_memory_region(unsigned long long start,
356 unsigned long long size, int type)
364 printk(KERN_ERR "Ooops! Too many entries in the memory map!\n");
368 e820.map[x].addr = start;
369 e820.map[x].size = size;
370 e820.map[x].type = type;
373 } /* add_memory_region */
376 * Sanitize the BIOS e820 map.
378 * Some e820 responses include overlapping entries. The following
379 * replaces the original e820 map with a new one, removing overlaps.
382 int __init sanitize_e820_map(struct e820entry * biosmap, char * pnr_map)
384 struct change_member *change_tmp;
385 unsigned long current_type, last_type;
386 unsigned long long last_addr;
387 int chgidx, still_changing;
390 int old_nr, new_nr, chg_nr;
394 Visually we're performing the following (1,2,3,4 = memory types)...
396 Sample memory map (w/overlaps):
397 ____22__________________
398 ______________________4_
399 ____1111________________
400 _44_____________________
401 11111111________________
402 ____________________33__
403 ___________44___________
404 __________33333_________
405 ______________22________
406 ___________________2222_
407 _________111111111______
408 _____________________11_
409 _________________4______
411 Sanitized equivalent (no overlap):
412 1_______________________
413 _44_____________________
414 ___1____________________
415 ____22__________________
416 ______11________________
417 _________1______________
418 __________3_____________
419 ___________44___________
420 _____________33_________
421 _______________2________
422 ________________1_______
423 _________________4______
424 ___________________2____
425 ____________________33__
426 ______________________4_
428 /* if there's only one memory region, don't bother */
435 /* bail out if we find any unreasonable addresses in bios map */
436 for (i=0; i<old_nr; i++)
437 if (biosmap[i].addr + biosmap[i].size < biosmap[i].addr) {
441 /* create pointers for initial change-point information (for sorting) */
442 for (i=0; i < 2*old_nr; i++)
443 change_point[i] = &change_point_list[i];
445 /* record all known change-points (starting and ending addresses),
446 omitting those that are for empty memory regions */
448 for (i=0; i < old_nr; i++) {
449 if (biosmap[i].size != 0) {
450 change_point[chgidx]->addr = biosmap[i].addr;
451 change_point[chgidx++]->pbios = &biosmap[i];
452 change_point[chgidx]->addr = biosmap[i].addr + biosmap[i].size;
453 change_point[chgidx++]->pbios = &biosmap[i];
456 chg_nr = chgidx; /* true number of change-points */
458 /* sort change-point list by memory addresses (low -> high) */
460 while (still_changing) {
462 for (i=1; i < chg_nr; i++) {
463 /* if <current_addr> > <last_addr>, swap */
464 /* or, if current=<start_addr> & last=<end_addr>, swap */
465 if ((change_point[i]->addr < change_point[i-1]->addr) ||
466 ((change_point[i]->addr == change_point[i-1]->addr) &&
467 (change_point[i]->addr == change_point[i]->pbios->addr) &&
468 (change_point[i-1]->addr != change_point[i-1]->pbios->addr))
471 change_tmp = change_point[i];
472 change_point[i] = change_point[i-1];
473 change_point[i-1] = change_tmp;
479 /* create a new bios memory map, removing overlaps */
480 overlap_entries=0; /* number of entries in the overlap table */
481 new_bios_entry=0; /* index for creating new bios map entries */
482 last_type = 0; /* start with undefined memory type */
483 last_addr = 0; /* start with 0 as last starting address */
484 /* loop through change-points, determining affect on the new bios map */
485 for (chgidx=0; chgidx < chg_nr; chgidx++)
487 /* keep track of all overlapping bios entries */
488 if (change_point[chgidx]->addr == change_point[chgidx]->pbios->addr)
490 /* add map entry to overlap list (> 1 entry implies an overlap) */
491 overlap_list[overlap_entries++]=change_point[chgidx]->pbios;
495 /* remove entry from list (order independent, so swap with last) */
496 for (i=0; i<overlap_entries; i++)
498 if (overlap_list[i] == change_point[chgidx]->pbios)
499 overlap_list[i] = overlap_list[overlap_entries-1];
503 /* if there are overlapping entries, decide which "type" to use */
504 /* (larger value takes precedence -- 1=usable, 2,3,4,4+=unusable) */
506 for (i=0; i<overlap_entries; i++)
507 if (overlap_list[i]->type > current_type)
508 current_type = overlap_list[i]->type;
509 /* continue building up new bios map based on this information */
510 if (current_type != last_type) {
511 if (last_type != 0) {
512 new_bios[new_bios_entry].size =
513 change_point[chgidx]->addr - last_addr;
514 /* move forward only if the new size was non-zero */
515 if (new_bios[new_bios_entry].size != 0)
516 if (++new_bios_entry >= E820MAX)
517 break; /* no more space left for new bios entries */
519 if (current_type != 0) {
520 new_bios[new_bios_entry].addr = change_point[chgidx]->addr;
521 new_bios[new_bios_entry].type = current_type;
522 last_addr=change_point[chgidx]->addr;
524 last_type = current_type;
527 new_nr = new_bios_entry; /* retain count for new bios entries */
529 /* copy new bios mapping into original location */
530 memcpy(biosmap, new_bios, new_nr*sizeof(struct e820entry));
537 * Copy the BIOS e820 map into a safe place.
539 * Sanity-check it while we're at it..
541 * If we're lucky and live on a modern system, the setup code
542 * will have given us a memory map that we can use to properly
543 * set up memory. If we aren't, we'll fake a memory map.
545 * We check to see that the memory map contains at least 2 elements
546 * before we'll use it, because the detection code in setup.S may
547 * not be perfect and most every PC known to man has two memory
548 * regions: one from 0 to 640k, and one from 1mb up. (The IBM
549 * thinkpad 560x, for example, does not cooperate with the memory
552 int __init copy_e820_map(struct e820entry * biosmap, int nr_map)
554 /* Only one memory region (or negative)? Ignore it */
559 unsigned long long start = biosmap->addr;
560 unsigned long long size = biosmap->size;
561 unsigned long long end = start + size;
562 unsigned long type = biosmap->type;
564 /* Overflow in 64 bits? Ignore the memory map. */
569 * Some BIOSes claim RAM in the 640k - 1M region.
570 * Not right. Fix it up.
572 if (type == E820_RAM) {
573 if (start < 0x100000ULL && end > 0xA0000ULL) {
574 if (start < 0xA0000ULL)
575 add_memory_region(start, 0xA0000ULL-start, type);
576 if (end <= 0x100000ULL)
582 add_memory_region(start, size, type);
583 } while (biosmap++,--nr_map);
588 * Callback for efi_memory_walk.
591 efi_find_max_pfn(unsigned long start, unsigned long end, void *arg)
593 unsigned long *max_pfn = arg, pfn;
596 pfn = PFN_UP(end -1);
604 efi_memory_present_wrapper(unsigned long start, unsigned long end, void *arg)
606 memory_present(0, PFN_UP(start), PFN_DOWN(end));
611 * Find the highest page frame number we have available
613 void __init find_max_pfn(void)
619 efi_memmap_walk(efi_find_max_pfn, &max_pfn);
620 efi_memmap_walk(efi_memory_present_wrapper, NULL);
624 for (i = 0; i < e820.nr_map; i++) {
625 unsigned long start, end;
627 if (e820.map[i].type != E820_RAM)
629 start = PFN_UP(e820.map[i].addr);
630 end = PFN_DOWN(e820.map[i].addr + e820.map[i].size);
635 memory_present(0, start, end);
640 * Free all available memory for boot time allocation. Used
641 * as a callback function by efi_memory_walk()
645 free_available_memory(unsigned long start, unsigned long end, void *arg)
647 /* check max_low_pfn */
648 if (start >= (max_low_pfn << PAGE_SHIFT))
650 if (end >= (max_low_pfn << PAGE_SHIFT))
651 end = max_low_pfn << PAGE_SHIFT;
653 free_bootmem(start, end - start);
658 * Register fully available low RAM pages with the bootmem allocator.
660 void __init register_bootmem_low_pages(unsigned long max_low_pfn)
665 efi_memmap_walk(free_available_memory, NULL);
668 for (i = 0; i < e820.nr_map; i++) {
669 unsigned long curr_pfn, last_pfn, size;
671 * Reserve usable low memory
673 if (e820.map[i].type != E820_RAM)
676 * We are rounding up the start address of usable memory:
678 curr_pfn = PFN_UP(e820.map[i].addr);
679 if (curr_pfn >= max_low_pfn)
682 * ... and at the end of the usable range downwards:
684 last_pfn = PFN_DOWN(e820.map[i].addr + e820.map[i].size);
686 if (last_pfn > max_low_pfn)
687 last_pfn = max_low_pfn;
690 * .. finally, did all the rounding and playing
691 * around just make the area go away?
693 if (last_pfn <= curr_pfn)
696 size = last_pfn - curr_pfn;
697 free_bootmem(PFN_PHYS(curr_pfn), PFN_PHYS(size));
701 void __init e820_register_memory(void)
703 unsigned long gapstart, gapsize, round;
704 unsigned long long last;
708 * Search for the bigest gap in the low 32 bits of the e820
711 last = 0x100000000ull;
712 gapstart = 0x10000000;
716 unsigned long long start = e820.map[i].addr;
717 unsigned long long end = start + e820.map[i].size;
720 * Since "last" is at most 4GB, we know we'll
721 * fit in 32 bits if this condition is true
724 unsigned long gap = last - end;
736 * See how much we want to round up: start off with
737 * rounding to the next 1MB area.
740 while ((gapsize >> 4) > round)
742 /* Fun with two's complement */
743 pci_mem_start = (gapstart + round) & -round;
745 printk("Allocating PCI resources starting at %08lx (gap: %08lx:%08lx)\n",
746 pci_mem_start, gapstart, gapsize);
749 void __init print_memory_map(char *who)
753 for (i = 0; i < e820.nr_map; i++) {
754 printk(" %s: %016Lx - %016Lx ", who,
756 e820.map[i].addr + e820.map[i].size);
757 switch (e820.map[i].type) {
758 case E820_RAM: printk("(usable)\n");
761 printk("(reserved)\n");
764 printk("(ACPI data)\n");
767 printk("(ACPI NVS)\n");
769 default: printk("type %u\n", e820.map[i].type);
775 static __init __always_inline void efi_limit_regions(unsigned long long size)
777 unsigned long long current_addr = 0;
778 efi_memory_desc_t *md, *next_md;
784 for (p = p1, i = 0; p < memmap.map_end; p += memmap.desc_size, i++) {
787 current_addr = md->phys_addr +
788 PFN_PHYS(md->num_pages);
789 if (is_available_memory(md)) {
790 if (md->phys_addr >= size) continue;
791 memcpy(next_md, md, memmap.desc_size);
792 if (current_addr >= size) {
793 next_md->num_pages -=
794 PFN_UP(current_addr-size);
796 p1 += memmap.desc_size;
799 } else if ((md->attribute & EFI_MEMORY_RUNTIME) ==
800 EFI_MEMORY_RUNTIME) {
801 /* In order to make runtime services
802 * available we have to include runtime
803 * memory regions in memory map */
804 memcpy(next_md, md, memmap.desc_size);
805 p1 += memmap.desc_size;
811 memmap.map_end = memmap.map +
812 (memmap.nr_map * memmap.desc_size);
815 void __init limit_regions(unsigned long long size)
817 unsigned long long current_addr;
820 print_memory_map("limit_regions start");
822 efi_limit_regions(size);
825 for (i = 0; i < e820.nr_map; i++) {
826 current_addr = e820.map[i].addr + e820.map[i].size;
827 if (current_addr < size)
830 if (e820.map[i].type != E820_RAM)
833 if (e820.map[i].addr >= size) {
835 * This region starts past the end of the
836 * requested size, skip it completely.
841 e820.map[i].size -= current_addr - size;
843 print_memory_map("limit_regions endfor");
846 print_memory_map("limit_regions endfunc");
850 * This function checks if any part of the range <start,end> is mapped
854 e820_any_mapped(u64 start, u64 end, unsigned type)
857 for (i = 0; i < e820.nr_map; i++) {
858 const struct e820entry *ei = &e820.map[i];
859 if (type && ei->type != type)
861 if (ei->addr >= end || ei->addr + ei->size <= start)
867 EXPORT_SYMBOL_GPL(e820_any_mapped);
870 * This function checks if the entire range <start,end> is mapped with type.
872 * Note: this function only works correct if the e820 table is sorted and
873 * not-overlapping, which is the case
876 e820_all_mapped(unsigned long s, unsigned long e, unsigned type)
881 for (i = 0; i < e820.nr_map; i++) {
882 struct e820entry *ei = &e820.map[i];
883 if (type && ei->type != type)
885 /* is the region (part) in overlap with the current region ?*/
886 if (ei->addr >= end || ei->addr + ei->size <= start)
888 /* if the region is at the beginning of <start,end> we move
889 * start to the end of the region since it's ok until there
891 if (ei->addr <= start)
892 start = ei->addr + ei->size;
893 /* if start is now at or beyond end, we're done, full
896 return 1; /* we're done */
901 static int __init parse_memmap(char *arg)
906 if (strcmp(arg, "exactmap") == 0) {
907 #ifdef CONFIG_CRASH_DUMP
908 /* If we are doing a crash dump, we
909 * still need to know the real mem
910 * size before original memory map is
914 saved_max_pfn = max_pfn;
917 user_defined_memmap = 1;
919 /* If the user specifies memory size, we
920 * limit the BIOS-provided memory map to
921 * that size. exactmap can be used to specify
922 * the exact map. mem=number can be used to
923 * trim the existing memory map.
925 unsigned long long start_at, mem_size;
927 mem_size = memparse(arg, &arg);
929 start_at = memparse(arg+1, &arg);
930 add_memory_region(start_at, mem_size, E820_RAM);
931 } else if (*arg == '#') {
932 start_at = memparse(arg+1, &arg);
933 add_memory_region(start_at, mem_size, E820_ACPI);
934 } else if (*arg == '$') {
935 start_at = memparse(arg+1, &arg);
936 add_memory_region(start_at, mem_size, E820_RESERVED);
938 limit_regions(mem_size);
939 user_defined_memmap = 1;
944 early_param("memmap", parse_memmap);