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 if (crashk_res.start != crashk_res.end)
292 request_resource(res, &crashk_res);
299 * Request address space for all standard resources
301 * This is called just before pcibios_init(), which is also a
302 * subsys_initcall, but is linked in later (in arch/i386/pci/common.c).
304 static int __init request_standard_resources(void)
308 printk("Setting up standard PCI resources\n");
310 efi_initialize_iomem_resources(&code_resource, &data_resource);
312 legacy_init_iomem_resources(&code_resource, &data_resource);
314 /* EFI systems may still have VGA */
315 request_resource(&iomem_resource, &video_ram_resource);
317 /* request I/O space for devices used on all i[345]86 PCs */
318 for (i = 0; i < ARRAY_SIZE(standard_io_resources); i++)
319 request_resource(&ioport_resource, &standard_io_resources[i]);
323 subsys_initcall(request_standard_resources);
325 #if defined(CONFIG_PM) && defined(CONFIG_HIBERNATION)
327 * e820_mark_nosave_regions - Find the ranges of physical addresses that do not
328 * correspond to e820 RAM areas and mark the corresponding pages as nosave for
331 * This function requires the e820 map to be sorted and without any
332 * overlapping entries and assumes the first e820 area to be RAM.
334 void __init e820_mark_nosave_regions(void)
339 pfn = PFN_DOWN(e820.map[0].addr + e820.map[0].size);
340 for (i = 1; i < e820.nr_map; i++) {
341 struct e820entry *ei = &e820.map[i];
343 if (pfn < PFN_UP(ei->addr))
344 register_nosave_region(pfn, PFN_UP(ei->addr));
346 pfn = PFN_DOWN(ei->addr + ei->size);
347 if (ei->type != E820_RAM)
348 register_nosave_region(PFN_UP(ei->addr), pfn);
350 if (pfn >= max_low_pfn)
356 void __init add_memory_region(unsigned long long start,
357 unsigned long long size, int type)
365 printk(KERN_ERR "Ooops! Too many entries in the memory map!\n");
369 e820.map[x].addr = start;
370 e820.map[x].size = size;
371 e820.map[x].type = type;
374 } /* add_memory_region */
377 * Sanitize the BIOS e820 map.
379 * Some e820 responses include overlapping entries. The following
380 * replaces the original e820 map with a new one, removing overlaps.
383 int __init sanitize_e820_map(struct e820entry * biosmap, char * pnr_map)
385 struct change_member *change_tmp;
386 unsigned long current_type, last_type;
387 unsigned long long last_addr;
388 int chgidx, still_changing;
391 int old_nr, new_nr, chg_nr;
395 Visually we're performing the following (1,2,3,4 = memory types)...
397 Sample memory map (w/overlaps):
398 ____22__________________
399 ______________________4_
400 ____1111________________
401 _44_____________________
402 11111111________________
403 ____________________33__
404 ___________44___________
405 __________33333_________
406 ______________22________
407 ___________________2222_
408 _________111111111______
409 _____________________11_
410 _________________4______
412 Sanitized equivalent (no overlap):
413 1_______________________
414 _44_____________________
415 ___1____________________
416 ____22__________________
417 ______11________________
418 _________1______________
419 __________3_____________
420 ___________44___________
421 _____________33_________
422 _______________2________
423 ________________1_______
424 _________________4______
425 ___________________2____
426 ____________________33__
427 ______________________4_
429 /* if there's only one memory region, don't bother */
436 /* bail out if we find any unreasonable addresses in bios map */
437 for (i=0; i<old_nr; i++)
438 if (biosmap[i].addr + biosmap[i].size < biosmap[i].addr) {
442 /* create pointers for initial change-point information (for sorting) */
443 for (i=0; i < 2*old_nr; i++)
444 change_point[i] = &change_point_list[i];
446 /* record all known change-points (starting and ending addresses),
447 omitting those that are for empty memory regions */
449 for (i=0; i < old_nr; i++) {
450 if (biosmap[i].size != 0) {
451 change_point[chgidx]->addr = biosmap[i].addr;
452 change_point[chgidx++]->pbios = &biosmap[i];
453 change_point[chgidx]->addr = biosmap[i].addr + biosmap[i].size;
454 change_point[chgidx++]->pbios = &biosmap[i];
457 chg_nr = chgidx; /* true number of change-points */
459 /* sort change-point list by memory addresses (low -> high) */
461 while (still_changing) {
463 for (i=1; i < chg_nr; i++) {
464 /* if <current_addr> > <last_addr>, swap */
465 /* or, if current=<start_addr> & last=<end_addr>, swap */
466 if ((change_point[i]->addr < change_point[i-1]->addr) ||
467 ((change_point[i]->addr == change_point[i-1]->addr) &&
468 (change_point[i]->addr == change_point[i]->pbios->addr) &&
469 (change_point[i-1]->addr != change_point[i-1]->pbios->addr))
472 change_tmp = change_point[i];
473 change_point[i] = change_point[i-1];
474 change_point[i-1] = change_tmp;
480 /* create a new bios memory map, removing overlaps */
481 overlap_entries=0; /* number of entries in the overlap table */
482 new_bios_entry=0; /* index for creating new bios map entries */
483 last_type = 0; /* start with undefined memory type */
484 last_addr = 0; /* start with 0 as last starting address */
485 /* loop through change-points, determining affect on the new bios map */
486 for (chgidx=0; chgidx < chg_nr; chgidx++)
488 /* keep track of all overlapping bios entries */
489 if (change_point[chgidx]->addr == change_point[chgidx]->pbios->addr)
491 /* add map entry to overlap list (> 1 entry implies an overlap) */
492 overlap_list[overlap_entries++]=change_point[chgidx]->pbios;
496 /* remove entry from list (order independent, so swap with last) */
497 for (i=0; i<overlap_entries; i++)
499 if (overlap_list[i] == change_point[chgidx]->pbios)
500 overlap_list[i] = overlap_list[overlap_entries-1];
504 /* if there are overlapping entries, decide which "type" to use */
505 /* (larger value takes precedence -- 1=usable, 2,3,4,4+=unusable) */
507 for (i=0; i<overlap_entries; i++)
508 if (overlap_list[i]->type > current_type)
509 current_type = overlap_list[i]->type;
510 /* continue building up new bios map based on this information */
511 if (current_type != last_type) {
512 if (last_type != 0) {
513 new_bios[new_bios_entry].size =
514 change_point[chgidx]->addr - last_addr;
515 /* move forward only if the new size was non-zero */
516 if (new_bios[new_bios_entry].size != 0)
517 if (++new_bios_entry >= E820MAX)
518 break; /* no more space left for new bios entries */
520 if (current_type != 0) {
521 new_bios[new_bios_entry].addr = change_point[chgidx]->addr;
522 new_bios[new_bios_entry].type = current_type;
523 last_addr=change_point[chgidx]->addr;
525 last_type = current_type;
528 new_nr = new_bios_entry; /* retain count for new bios entries */
530 /* copy new bios mapping into original location */
531 memcpy(biosmap, new_bios, new_nr*sizeof(struct e820entry));
538 * Copy the BIOS e820 map into a safe place.
540 * Sanity-check it while we're at it..
542 * If we're lucky and live on a modern system, the setup code
543 * will have given us a memory map that we can use to properly
544 * set up memory. If we aren't, we'll fake a memory map.
546 * We check to see that the memory map contains at least 2 elements
547 * before we'll use it, because the detection code in setup.S may
548 * not be perfect and most every PC known to man has two memory
549 * regions: one from 0 to 640k, and one from 1mb up. (The IBM
550 * thinkpad 560x, for example, does not cooperate with the memory
553 int __init copy_e820_map(struct e820entry * biosmap, int nr_map)
555 /* Only one memory region (or negative)? Ignore it */
560 unsigned long long start = biosmap->addr;
561 unsigned long long size = biosmap->size;
562 unsigned long long end = start + size;
563 unsigned long type = biosmap->type;
565 /* Overflow in 64 bits? Ignore the memory map. */
570 * Some BIOSes claim RAM in the 640k - 1M region.
571 * Not right. Fix it up.
573 if (type == E820_RAM) {
574 if (start < 0x100000ULL && end > 0xA0000ULL) {
575 if (start < 0xA0000ULL)
576 add_memory_region(start, 0xA0000ULL-start, type);
577 if (end <= 0x100000ULL)
583 add_memory_region(start, size, type);
584 } while (biosmap++,--nr_map);
589 * Callback for efi_memory_walk.
592 efi_find_max_pfn(unsigned long start, unsigned long end, void *arg)
594 unsigned long *max_pfn = arg, pfn;
597 pfn = PFN_UP(end -1);
605 efi_memory_present_wrapper(unsigned long start, unsigned long end, void *arg)
607 memory_present(0, PFN_UP(start), PFN_DOWN(end));
612 * Find the highest page frame number we have available
614 void __init find_max_pfn(void)
620 efi_memmap_walk(efi_find_max_pfn, &max_pfn);
621 efi_memmap_walk(efi_memory_present_wrapper, NULL);
625 for (i = 0; i < e820.nr_map; i++) {
626 unsigned long start, end;
628 if (e820.map[i].type != E820_RAM)
630 start = PFN_UP(e820.map[i].addr);
631 end = PFN_DOWN(e820.map[i].addr + e820.map[i].size);
636 memory_present(0, start, end);
641 * Free all available memory for boot time allocation. Used
642 * as a callback function by efi_memory_walk()
646 free_available_memory(unsigned long start, unsigned long end, void *arg)
648 /* check max_low_pfn */
649 if (start >= (max_low_pfn << PAGE_SHIFT))
651 if (end >= (max_low_pfn << PAGE_SHIFT))
652 end = max_low_pfn << PAGE_SHIFT;
654 free_bootmem(start, end - start);
659 * Register fully available low RAM pages with the bootmem allocator.
661 void __init register_bootmem_low_pages(unsigned long max_low_pfn)
666 efi_memmap_walk(free_available_memory, NULL);
669 for (i = 0; i < e820.nr_map; i++) {
670 unsigned long curr_pfn, last_pfn, size;
672 * Reserve usable low memory
674 if (e820.map[i].type != E820_RAM)
677 * We are rounding up the start address of usable memory:
679 curr_pfn = PFN_UP(e820.map[i].addr);
680 if (curr_pfn >= max_low_pfn)
683 * ... and at the end of the usable range downwards:
685 last_pfn = PFN_DOWN(e820.map[i].addr + e820.map[i].size);
687 if (last_pfn > max_low_pfn)
688 last_pfn = max_low_pfn;
691 * .. finally, did all the rounding and playing
692 * around just make the area go away?
694 if (last_pfn <= curr_pfn)
697 size = last_pfn - curr_pfn;
698 free_bootmem(PFN_PHYS(curr_pfn), PFN_PHYS(size));
702 void __init e820_register_memory(void)
704 unsigned long gapstart, gapsize, round;
705 unsigned long long last;
709 * Search for the biggest gap in the low 32 bits of the e820
712 last = 0x100000000ull;
713 gapstart = 0x10000000;
717 unsigned long long start = e820.map[i].addr;
718 unsigned long long end = start + e820.map[i].size;
721 * Since "last" is at most 4GB, we know we'll
722 * fit in 32 bits if this condition is true
725 unsigned long gap = last - end;
737 * See how much we want to round up: start off with
738 * rounding to the next 1MB area.
741 while ((gapsize >> 4) > round)
743 /* Fun with two's complement */
744 pci_mem_start = (gapstart + round) & -round;
746 printk("Allocating PCI resources starting at %08lx (gap: %08lx:%08lx)\n",
747 pci_mem_start, gapstart, gapsize);
750 void __init print_memory_map(char *who)
754 for (i = 0; i < e820.nr_map; i++) {
755 printk(" %s: %016Lx - %016Lx ", who,
757 e820.map[i].addr + e820.map[i].size);
758 switch (e820.map[i].type) {
759 case E820_RAM: printk("(usable)\n");
762 printk("(reserved)\n");
765 printk("(ACPI data)\n");
768 printk("(ACPI NVS)\n");
770 default: printk("type %u\n", e820.map[i].type);
776 static __init __always_inline void efi_limit_regions(unsigned long long size)
778 unsigned long long current_addr = 0;
779 efi_memory_desc_t *md, *next_md;
785 for (p = p1, i = 0; p < memmap.map_end; p += memmap.desc_size, i++) {
788 current_addr = md->phys_addr +
789 PFN_PHYS(md->num_pages);
790 if (is_available_memory(md)) {
791 if (md->phys_addr >= size) continue;
792 memcpy(next_md, md, memmap.desc_size);
793 if (current_addr >= size) {
794 next_md->num_pages -=
795 PFN_UP(current_addr-size);
797 p1 += memmap.desc_size;
800 } else if ((md->attribute & EFI_MEMORY_RUNTIME) ==
801 EFI_MEMORY_RUNTIME) {
802 /* In order to make runtime services
803 * available we have to include runtime
804 * memory regions in memory map */
805 memcpy(next_md, md, memmap.desc_size);
806 p1 += memmap.desc_size;
812 memmap.map_end = memmap.map +
813 (memmap.nr_map * memmap.desc_size);
816 void __init limit_regions(unsigned long long size)
818 unsigned long long current_addr;
821 print_memory_map("limit_regions start");
823 efi_limit_regions(size);
826 for (i = 0; i < e820.nr_map; i++) {
827 current_addr = e820.map[i].addr + e820.map[i].size;
828 if (current_addr < size)
831 if (e820.map[i].type != E820_RAM)
834 if (e820.map[i].addr >= size) {
836 * This region starts past the end of the
837 * requested size, skip it completely.
842 e820.map[i].size -= current_addr - size;
844 print_memory_map("limit_regions endfor");
847 print_memory_map("limit_regions endfunc");
851 * This function checks if any part of the range <start,end> is mapped
855 e820_any_mapped(u64 start, u64 end, unsigned type)
858 for (i = 0; i < e820.nr_map; i++) {
859 const struct e820entry *ei = &e820.map[i];
860 if (type && ei->type != type)
862 if (ei->addr >= end || ei->addr + ei->size <= start)
868 EXPORT_SYMBOL_GPL(e820_any_mapped);
871 * This function checks if the entire range <start,end> is mapped with type.
873 * Note: this function only works correct if the e820 table is sorted and
874 * not-overlapping, which is the case
877 e820_all_mapped(unsigned long s, unsigned long e, unsigned type)
882 for (i = 0; i < e820.nr_map; i++) {
883 struct e820entry *ei = &e820.map[i];
884 if (type && ei->type != type)
886 /* is the region (part) in overlap with the current region ?*/
887 if (ei->addr >= end || ei->addr + ei->size <= start)
889 /* if the region is at the beginning of <start,end> we move
890 * start to the end of the region since it's ok until there
892 if (ei->addr <= start)
893 start = ei->addr + ei->size;
894 /* if start is now at or beyond end, we're done, full
897 return 1; /* we're done */
902 static int __init parse_memmap(char *arg)
907 if (strcmp(arg, "exactmap") == 0) {
908 #ifdef CONFIG_CRASH_DUMP
909 /* If we are doing a crash dump, we
910 * still need to know the real mem
911 * size before original memory map is
915 saved_max_pfn = max_pfn;
918 user_defined_memmap = 1;
920 /* If the user specifies memory size, we
921 * limit the BIOS-provided memory map to
922 * that size. exactmap can be used to specify
923 * the exact map. mem=number can be used to
924 * trim the existing memory map.
926 unsigned long long start_at, mem_size;
928 mem_size = memparse(arg, &arg);
930 start_at = memparse(arg+1, &arg);
931 add_memory_region(start_at, mem_size, E820_RAM);
932 } else if (*arg == '#') {
933 start_at = memparse(arg+1, &arg);
934 add_memory_region(start_at, mem_size, E820_ACPI);
935 } else if (*arg == '$') {
936 start_at = memparse(arg+1, &arg);
937 add_memory_region(start_at, mem_size, E820_RESERVED);
939 limit_regions(mem_size);
940 user_defined_memmap = 1;
945 early_param("memmap", parse_memmap);