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)
164 const unsigned short * const ptr = (const unsigned short *)rom;
167 return probe_kernel_address(ptr, sig) == 0 && sig == ROMSIGNATURE;
170 static int __init romchecksum(const unsigned char *rom, unsigned long length)
172 unsigned char sum, c;
174 for (sum = 0; length && probe_kernel_address(rom++, c) == 0; length--)
176 return !length && !sum;
179 static void __init probe_roms(void)
181 const unsigned char *rom;
182 unsigned long start, length, upper;
187 upper = adapter_rom_resources[0].start;
188 for (start = video_rom_resource.start; start < upper; start += 2048) {
189 rom = isa_bus_to_virt(start);
190 if (!romsignature(rom))
193 video_rom_resource.start = start;
195 if (probe_kernel_address(rom + 2, c) != 0)
198 /* 0 < length <= 0x7f * 512, historically */
201 /* if checksum okay, trust length byte */
202 if (length && romchecksum(rom, length))
203 video_rom_resource.end = start + length - 1;
205 request_resource(&iomem_resource, &video_rom_resource);
209 start = (video_rom_resource.end + 1 + 2047) & ~2047UL;
214 request_resource(&iomem_resource, &system_rom_resource);
215 upper = system_rom_resource.start;
217 /* check for extension rom (ignore length byte!) */
218 rom = isa_bus_to_virt(extension_rom_resource.start);
219 if (romsignature(rom)) {
220 length = extension_rom_resource.end - extension_rom_resource.start + 1;
221 if (romchecksum(rom, length)) {
222 request_resource(&iomem_resource, &extension_rom_resource);
223 upper = extension_rom_resource.start;
227 /* check for adapter roms on 2k boundaries */
228 for (i = 0; i < ARRAY_SIZE(adapter_rom_resources) && start < upper; start += 2048) {
229 rom = isa_bus_to_virt(start);
230 if (!romsignature(rom))
233 if (probe_kernel_address(rom + 2, c) != 0)
236 /* 0 < length <= 0x7f * 512, historically */
239 /* but accept any length that fits if checksum okay */
240 if (!length || start + length > upper || !romchecksum(rom, length))
243 adapter_rom_resources[i].start = start;
244 adapter_rom_resources[i].end = start + length - 1;
245 request_resource(&iomem_resource, &adapter_rom_resources[i]);
247 start = adapter_rom_resources[i++].end & ~2047UL;
252 * Request address space for all standard RAM and ROM resources
253 * and also for regions reported as reserved by the e820.
256 legacy_init_iomem_resources(struct resource *code_resource, struct resource *data_resource)
261 for (i = 0; i < e820.nr_map; i++) {
262 struct resource *res;
263 #ifndef CONFIG_RESOURCES_64BIT
264 if (e820.map[i].addr + e820.map[i].size > 0x100000000ULL)
267 res = kzalloc(sizeof(struct resource), GFP_ATOMIC);
268 switch (e820.map[i].type) {
269 case E820_RAM: res->name = "System RAM"; break;
270 case E820_ACPI: res->name = "ACPI Tables"; break;
271 case E820_NVS: res->name = "ACPI Non-volatile Storage"; break;
272 default: res->name = "reserved";
274 res->start = e820.map[i].addr;
275 res->end = res->start + e820.map[i].size - 1;
276 res->flags = IORESOURCE_MEM | IORESOURCE_BUSY;
277 if (request_resource(&iomem_resource, res)) {
281 if (e820.map[i].type == E820_RAM) {
283 * We don't know which RAM region contains kernel data,
284 * so we try it repeatedly and let the resource manager
287 request_resource(res, code_resource);
288 request_resource(res, data_resource);
290 request_resource(res, &crashk_res);
297 * Request address space for all standard resources
299 * This is called just before pcibios_init(), which is also a
300 * subsys_initcall, but is linked in later (in arch/i386/pci/common.c).
302 static int __init request_standard_resources(void)
306 printk("Setting up standard PCI resources\n");
308 efi_initialize_iomem_resources(&code_resource, &data_resource);
310 legacy_init_iomem_resources(&code_resource, &data_resource);
312 /* EFI systems may still have VGA */
313 request_resource(&iomem_resource, &video_ram_resource);
315 /* request I/O space for devices used on all i[345]86 PCs */
316 for (i = 0; i < ARRAY_SIZE(standard_io_resources); i++)
317 request_resource(&ioport_resource, &standard_io_resources[i]);
321 subsys_initcall(request_standard_resources);
323 void __init add_memory_region(unsigned long long start,
324 unsigned long long size, int type)
332 printk(KERN_ERR "Ooops! Too many entries in the memory map!\n");
336 e820.map[x].addr = start;
337 e820.map[x].size = size;
338 e820.map[x].type = type;
341 } /* add_memory_region */
344 * Sanitize the BIOS e820 map.
346 * Some e820 responses include overlapping entries. The following
347 * replaces the original e820 map with a new one, removing overlaps.
350 int __init sanitize_e820_map(struct e820entry * biosmap, char * pnr_map)
352 struct change_member *change_tmp;
353 unsigned long current_type, last_type;
354 unsigned long long last_addr;
355 int chgidx, still_changing;
358 int old_nr, new_nr, chg_nr;
362 Visually we're performing the following (1,2,3,4 = memory types)...
364 Sample memory map (w/overlaps):
365 ____22__________________
366 ______________________4_
367 ____1111________________
368 _44_____________________
369 11111111________________
370 ____________________33__
371 ___________44___________
372 __________33333_________
373 ______________22________
374 ___________________2222_
375 _________111111111______
376 _____________________11_
377 _________________4______
379 Sanitized equivalent (no overlap):
380 1_______________________
381 _44_____________________
382 ___1____________________
383 ____22__________________
384 ______11________________
385 _________1______________
386 __________3_____________
387 ___________44___________
388 _____________33_________
389 _______________2________
390 ________________1_______
391 _________________4______
392 ___________________2____
393 ____________________33__
394 ______________________4_
396 printk("sanitize start\n");
397 /* if there's only one memory region, don't bother */
399 printk("sanitize bail 0\n");
405 /* bail out if we find any unreasonable addresses in bios map */
406 for (i=0; i<old_nr; i++)
407 if (biosmap[i].addr + biosmap[i].size < biosmap[i].addr) {
408 printk("sanitize bail 1\n");
412 /* create pointers for initial change-point information (for sorting) */
413 for (i=0; i < 2*old_nr; i++)
414 change_point[i] = &change_point_list[i];
416 /* record all known change-points (starting and ending addresses),
417 omitting those that are for empty memory regions */
419 for (i=0; i < old_nr; i++) {
420 if (biosmap[i].size != 0) {
421 change_point[chgidx]->addr = biosmap[i].addr;
422 change_point[chgidx++]->pbios = &biosmap[i];
423 change_point[chgidx]->addr = biosmap[i].addr + biosmap[i].size;
424 change_point[chgidx++]->pbios = &biosmap[i];
427 chg_nr = chgidx; /* true number of change-points */
429 /* sort change-point list by memory addresses (low -> high) */
431 while (still_changing) {
433 for (i=1; i < chg_nr; i++) {
434 /* if <current_addr> > <last_addr>, swap */
435 /* or, if current=<start_addr> & last=<end_addr>, swap */
436 if ((change_point[i]->addr < change_point[i-1]->addr) ||
437 ((change_point[i]->addr == change_point[i-1]->addr) &&
438 (change_point[i]->addr == change_point[i]->pbios->addr) &&
439 (change_point[i-1]->addr != change_point[i-1]->pbios->addr))
442 change_tmp = change_point[i];
443 change_point[i] = change_point[i-1];
444 change_point[i-1] = change_tmp;
450 /* create a new bios memory map, removing overlaps */
451 overlap_entries=0; /* number of entries in the overlap table */
452 new_bios_entry=0; /* index for creating new bios map entries */
453 last_type = 0; /* start with undefined memory type */
454 last_addr = 0; /* start with 0 as last starting address */
455 /* loop through change-points, determining affect on the new bios map */
456 for (chgidx=0; chgidx < chg_nr; chgidx++)
458 /* keep track of all overlapping bios entries */
459 if (change_point[chgidx]->addr == change_point[chgidx]->pbios->addr)
461 /* add map entry to overlap list (> 1 entry implies an overlap) */
462 overlap_list[overlap_entries++]=change_point[chgidx]->pbios;
466 /* remove entry from list (order independent, so swap with last) */
467 for (i=0; i<overlap_entries; i++)
469 if (overlap_list[i] == change_point[chgidx]->pbios)
470 overlap_list[i] = overlap_list[overlap_entries-1];
474 /* if there are overlapping entries, decide which "type" to use */
475 /* (larger value takes precedence -- 1=usable, 2,3,4,4+=unusable) */
477 for (i=0; i<overlap_entries; i++)
478 if (overlap_list[i]->type > current_type)
479 current_type = overlap_list[i]->type;
480 /* continue building up new bios map based on this information */
481 if (current_type != last_type) {
482 if (last_type != 0) {
483 new_bios[new_bios_entry].size =
484 change_point[chgidx]->addr - last_addr;
485 /* move forward only if the new size was non-zero */
486 if (new_bios[new_bios_entry].size != 0)
487 if (++new_bios_entry >= E820MAX)
488 break; /* no more space left for new bios entries */
490 if (current_type != 0) {
491 new_bios[new_bios_entry].addr = change_point[chgidx]->addr;
492 new_bios[new_bios_entry].type = current_type;
493 last_addr=change_point[chgidx]->addr;
495 last_type = current_type;
498 new_nr = new_bios_entry; /* retain count for new bios entries */
500 /* copy new bios mapping into original location */
501 memcpy(biosmap, new_bios, new_nr*sizeof(struct e820entry));
504 printk("sanitize end\n");
509 * Copy the BIOS e820 map into a safe place.
511 * Sanity-check it while we're at it..
513 * If we're lucky and live on a modern system, the setup code
514 * will have given us a memory map that we can use to properly
515 * set up memory. If we aren't, we'll fake a memory map.
517 * We check to see that the memory map contains at least 2 elements
518 * before we'll use it, because the detection code in setup.S may
519 * not be perfect and most every PC known to man has two memory
520 * regions: one from 0 to 640k, and one from 1mb up. (The IBM
521 * thinkpad 560x, for example, does not cooperate with the memory
524 int __init copy_e820_map(struct e820entry * biosmap, int nr_map)
526 /* Only one memory region (or negative)? Ignore it */
531 unsigned long long start = biosmap->addr;
532 unsigned long long size = biosmap->size;
533 unsigned long long end = start + size;
534 unsigned long type = biosmap->type;
535 printk("copy_e820_map() start: %016Lx size: %016Lx end: %016Lx type: %ld\n", start, size, end, type);
537 /* Overflow in 64 bits? Ignore the memory map. */
542 * Some BIOSes claim RAM in the 640k - 1M region.
543 * Not right. Fix it up.
545 if (type == E820_RAM) {
546 printk("copy_e820_map() type is E820_RAM\n");
547 if (start < 0x100000ULL && end > 0xA0000ULL) {
548 printk("copy_e820_map() lies in range...\n");
549 if (start < 0xA0000ULL) {
550 printk("copy_e820_map() start < 0xA0000ULL\n");
551 add_memory_region(start, 0xA0000ULL-start, type);
553 if (end <= 0x100000ULL) {
554 printk("copy_e820_map() end <= 0x100000ULL\n");
561 add_memory_region(start, size, type);
562 } while (biosmap++,--nr_map);
567 * Callback for efi_memory_walk.
570 efi_find_max_pfn(unsigned long start, unsigned long end, void *arg)
572 unsigned long *max_pfn = arg, pfn;
575 pfn = PFN_UP(end -1);
583 efi_memory_present_wrapper(unsigned long start, unsigned long end, void *arg)
585 memory_present(0, PFN_UP(start), PFN_DOWN(end));
590 * Find the highest page frame number we have available
592 void __init find_max_pfn(void)
598 efi_memmap_walk(efi_find_max_pfn, &max_pfn);
599 efi_memmap_walk(efi_memory_present_wrapper, NULL);
603 for (i = 0; i < e820.nr_map; i++) {
604 unsigned long start, end;
606 if (e820.map[i].type != E820_RAM)
608 start = PFN_UP(e820.map[i].addr);
609 end = PFN_DOWN(e820.map[i].addr + e820.map[i].size);
614 memory_present(0, start, end);
619 * Free all available memory for boot time allocation. Used
620 * as a callback function by efi_memory_walk()
624 free_available_memory(unsigned long start, unsigned long end, void *arg)
626 /* check max_low_pfn */
627 if (start >= (max_low_pfn << PAGE_SHIFT))
629 if (end >= (max_low_pfn << PAGE_SHIFT))
630 end = max_low_pfn << PAGE_SHIFT;
632 free_bootmem(start, end - start);
637 * Register fully available low RAM pages with the bootmem allocator.
639 void __init register_bootmem_low_pages(unsigned long max_low_pfn)
644 efi_memmap_walk(free_available_memory, NULL);
647 for (i = 0; i < e820.nr_map; i++) {
648 unsigned long curr_pfn, last_pfn, size;
650 * Reserve usable low memory
652 if (e820.map[i].type != E820_RAM)
655 * We are rounding up the start address of usable memory:
657 curr_pfn = PFN_UP(e820.map[i].addr);
658 if (curr_pfn >= max_low_pfn)
661 * ... and at the end of the usable range downwards:
663 last_pfn = PFN_DOWN(e820.map[i].addr + e820.map[i].size);
665 if (last_pfn > max_low_pfn)
666 last_pfn = max_low_pfn;
669 * .. finally, did all the rounding and playing
670 * around just make the area go away?
672 if (last_pfn <= curr_pfn)
675 size = last_pfn - curr_pfn;
676 free_bootmem(PFN_PHYS(curr_pfn), PFN_PHYS(size));
680 void __init e820_register_memory(void)
682 unsigned long gapstart, gapsize, round;
683 unsigned long long last;
687 * Search for the bigest gap in the low 32 bits of the e820
690 last = 0x100000000ull;
691 gapstart = 0x10000000;
695 unsigned long long start = e820.map[i].addr;
696 unsigned long long end = start + e820.map[i].size;
699 * Since "last" is at most 4GB, we know we'll
700 * fit in 32 bits if this condition is true
703 unsigned long gap = last - end;
715 * See how much we want to round up: start off with
716 * rounding to the next 1MB area.
719 while ((gapsize >> 4) > round)
721 /* Fun with two's complement */
722 pci_mem_start = (gapstart + round) & -round;
724 printk("Allocating PCI resources starting at %08lx (gap: %08lx:%08lx)\n",
725 pci_mem_start, gapstart, gapsize);
728 void __init print_memory_map(char *who)
732 for (i = 0; i < e820.nr_map; i++) {
733 printk(" %s: %016Lx - %016Lx ", who,
735 e820.map[i].addr + e820.map[i].size);
736 switch (e820.map[i].type) {
737 case E820_RAM: printk("(usable)\n");
740 printk("(reserved)\n");
743 printk("(ACPI data)\n");
746 printk("(ACPI NVS)\n");
748 default: printk("type %lu\n", e820.map[i].type);
754 static __init __always_inline void efi_limit_regions(unsigned long long size)
756 unsigned long long current_addr = 0;
757 efi_memory_desc_t *md, *next_md;
763 for (p = p1, i = 0; p < memmap.map_end; p += memmap.desc_size, i++) {
766 current_addr = md->phys_addr +
767 PFN_PHYS(md->num_pages);
768 if (is_available_memory(md)) {
769 if (md->phys_addr >= size) continue;
770 memcpy(next_md, md, memmap.desc_size);
771 if (current_addr >= size) {
772 next_md->num_pages -=
773 PFN_UP(current_addr-size);
775 p1 += memmap.desc_size;
778 } else if ((md->attribute & EFI_MEMORY_RUNTIME) ==
779 EFI_MEMORY_RUNTIME) {
780 /* In order to make runtime services
781 * available we have to include runtime
782 * memory regions in memory map */
783 memcpy(next_md, md, memmap.desc_size);
784 p1 += memmap.desc_size;
790 memmap.map_end = memmap.map +
791 (memmap.nr_map * memmap.desc_size);
794 void __init limit_regions(unsigned long long size)
796 unsigned long long current_addr;
799 print_memory_map("limit_regions start");
801 efi_limit_regions(size);
804 for (i = 0; i < e820.nr_map; i++) {
805 current_addr = e820.map[i].addr + e820.map[i].size;
806 if (current_addr < size)
809 if (e820.map[i].type != E820_RAM)
812 if (e820.map[i].addr >= size) {
814 * This region starts past the end of the
815 * requested size, skip it completely.
820 e820.map[i].size -= current_addr - size;
822 print_memory_map("limit_regions endfor");
825 print_memory_map("limit_regions endfunc");
829 * This function checks if any part of the range <start,end> is mapped
833 e820_any_mapped(u64 start, u64 end, unsigned type)
836 for (i = 0; i < e820.nr_map; i++) {
837 const struct e820entry *ei = &e820.map[i];
838 if (type && ei->type != type)
840 if (ei->addr >= end || ei->addr + ei->size <= start)
846 EXPORT_SYMBOL_GPL(e820_any_mapped);
849 * This function checks if the entire range <start,end> is mapped with type.
851 * Note: this function only works correct if the e820 table is sorted and
852 * not-overlapping, which is the case
855 e820_all_mapped(unsigned long s, unsigned long e, unsigned type)
860 for (i = 0; i < e820.nr_map; i++) {
861 struct e820entry *ei = &e820.map[i];
862 if (type && ei->type != type)
864 /* is the region (part) in overlap with the current region ?*/
865 if (ei->addr >= end || ei->addr + ei->size <= start)
867 /* if the region is at the beginning of <start,end> we move
868 * start to the end of the region since it's ok until there
870 if (ei->addr <= start)
871 start = ei->addr + ei->size;
872 /* if start is now at or beyond end, we're done, full
875 return 1; /* we're done */
880 static int __init parse_memmap(char *arg)
885 if (strcmp(arg, "exactmap") == 0) {
886 #ifdef CONFIG_CRASH_DUMP
887 /* If we are doing a crash dump, we
888 * still need to know the real mem
889 * size before original memory map is
893 saved_max_pfn = max_pfn;
896 user_defined_memmap = 1;
898 /* If the user specifies memory size, we
899 * limit the BIOS-provided memory map to
900 * that size. exactmap can be used to specify
901 * the exact map. mem=number can be used to
902 * trim the existing memory map.
904 unsigned long long start_at, mem_size;
906 mem_size = memparse(arg, &arg);
908 start_at = memparse(arg+1, &arg);
909 add_memory_region(start_at, mem_size, E820_RAM);
910 } else if (*arg == '#') {
911 start_at = memparse(arg+1, &arg);
912 add_memory_region(start_at, mem_size, E820_ACPI);
913 } else if (*arg == '$') {
914 start_at = memparse(arg+1, &arg);
915 add_memory_region(start_at, mem_size, E820_RESERVED);
917 limit_regions(mem_size);
918 user_defined_memmap = 1;
923 early_param("memmap", parse_memmap);