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 static int romsignature(const unsigned char *x)
164 if (probe_kernel_address((const unsigned short *)x, sig) == 0)
165 ret = (sig == 0xaa55);
169 static int __init romchecksum(unsigned char *rom, unsigned long length)
171 unsigned char *p, sum = 0;
173 for (p = rom; p < rom + length; p++)
178 static void __init probe_roms(void)
180 unsigned long start, length, upper;
185 upper = adapter_rom_resources[0].start;
186 for (start = video_rom_resource.start; start < upper; start += 2048) {
187 rom = isa_bus_to_virt(start);
188 if (!romsignature(rom))
191 video_rom_resource.start = start;
193 /* 0 < length <= 0x7f * 512, historically */
194 length = rom[2] * 512;
196 /* if checksum okay, trust length byte */
197 if (length && romchecksum(rom, length))
198 video_rom_resource.end = start + length - 1;
200 request_resource(&iomem_resource, &video_rom_resource);
204 start = (video_rom_resource.end + 1 + 2047) & ~2047UL;
209 request_resource(&iomem_resource, &system_rom_resource);
210 upper = system_rom_resource.start;
212 /* check for extension rom (ignore length byte!) */
213 rom = isa_bus_to_virt(extension_rom_resource.start);
214 if (romsignature(rom)) {
215 length = extension_rom_resource.end - extension_rom_resource.start + 1;
216 if (romchecksum(rom, length)) {
217 request_resource(&iomem_resource, &extension_rom_resource);
218 upper = extension_rom_resource.start;
222 /* check for adapter roms on 2k boundaries */
223 for (i = 0; i < ARRAY_SIZE(adapter_rom_resources) && start < upper; start += 2048) {
224 rom = isa_bus_to_virt(start);
225 if (!romsignature(rom))
228 /* 0 < length <= 0x7f * 512, historically */
229 length = rom[2] * 512;
231 /* but accept any length that fits if checksum okay */
232 if (!length || start + length > upper || !romchecksum(rom, length))
235 adapter_rom_resources[i].start = start;
236 adapter_rom_resources[i].end = start + length - 1;
237 request_resource(&iomem_resource, &adapter_rom_resources[i]);
239 start = adapter_rom_resources[i++].end & ~2047UL;
244 * Request address space for all standard RAM and ROM resources
245 * and also for regions reported as reserved by the e820.
248 legacy_init_iomem_resources(struct resource *code_resource, struct resource *data_resource)
253 for (i = 0; i < e820.nr_map; i++) {
254 struct resource *res;
255 #ifndef CONFIG_RESOURCES_64BIT
256 if (e820.map[i].addr + e820.map[i].size > 0x100000000ULL)
259 res = kzalloc(sizeof(struct resource), GFP_ATOMIC);
260 switch (e820.map[i].type) {
261 case E820_RAM: res->name = "System RAM"; break;
262 case E820_ACPI: res->name = "ACPI Tables"; break;
263 case E820_NVS: res->name = "ACPI Non-volatile Storage"; break;
264 default: res->name = "reserved";
266 res->start = e820.map[i].addr;
267 res->end = res->start + e820.map[i].size - 1;
268 res->flags = IORESOURCE_MEM | IORESOURCE_BUSY;
269 if (request_resource(&iomem_resource, res)) {
273 if (e820.map[i].type == E820_RAM) {
275 * We don't know which RAM region contains kernel data,
276 * so we try it repeatedly and let the resource manager
279 request_resource(res, code_resource);
280 request_resource(res, data_resource);
282 request_resource(res, &crashk_res);
289 * Request address space for all standard resources
291 * This is called just before pcibios_init(), which is also a
292 * subsys_initcall, but is linked in later (in arch/i386/pci/common.c).
294 static int __init request_standard_resources(void)
298 printk("Setting up standard PCI resources\n");
300 efi_initialize_iomem_resources(&code_resource, &data_resource);
302 legacy_init_iomem_resources(&code_resource, &data_resource);
304 /* EFI systems may still have VGA */
305 request_resource(&iomem_resource, &video_ram_resource);
307 /* request I/O space for devices used on all i[345]86 PCs */
308 for (i = 0; i < ARRAY_SIZE(standard_io_resources); i++)
309 request_resource(&ioport_resource, &standard_io_resources[i]);
313 subsys_initcall(request_standard_resources);
315 void __init add_memory_region(unsigned long long start,
316 unsigned long long size, int type)
324 printk(KERN_ERR "Ooops! Too many entries in the memory map!\n");
328 e820.map[x].addr = start;
329 e820.map[x].size = size;
330 e820.map[x].type = type;
333 } /* add_memory_region */
336 * Sanitize the BIOS e820 map.
338 * Some e820 responses include overlapping entries. The following
339 * replaces the original e820 map with a new one, removing overlaps.
342 int __init sanitize_e820_map(struct e820entry * biosmap, char * pnr_map)
344 struct change_member *change_tmp;
345 unsigned long current_type, last_type;
346 unsigned long long last_addr;
347 int chgidx, still_changing;
350 int old_nr, new_nr, chg_nr;
354 Visually we're performing the following (1,2,3,4 = memory types)...
356 Sample memory map (w/overlaps):
357 ____22__________________
358 ______________________4_
359 ____1111________________
360 _44_____________________
361 11111111________________
362 ____________________33__
363 ___________44___________
364 __________33333_________
365 ______________22________
366 ___________________2222_
367 _________111111111______
368 _____________________11_
369 _________________4______
371 Sanitized equivalent (no overlap):
372 1_______________________
373 _44_____________________
374 ___1____________________
375 ____22__________________
376 ______11________________
377 _________1______________
378 __________3_____________
379 ___________44___________
380 _____________33_________
381 _______________2________
382 ________________1_______
383 _________________4______
384 ___________________2____
385 ____________________33__
386 ______________________4_
388 printk("sanitize start\n");
389 /* if there's only one memory region, don't bother */
391 printk("sanitize bail 0\n");
397 /* bail out if we find any unreasonable addresses in bios map */
398 for (i=0; i<old_nr; i++)
399 if (biosmap[i].addr + biosmap[i].size < biosmap[i].addr) {
400 printk("sanitize bail 1\n");
404 /* create pointers for initial change-point information (for sorting) */
405 for (i=0; i < 2*old_nr; i++)
406 change_point[i] = &change_point_list[i];
408 /* record all known change-points (starting and ending addresses),
409 omitting those that are for empty memory regions */
411 for (i=0; i < old_nr; i++) {
412 if (biosmap[i].size != 0) {
413 change_point[chgidx]->addr = biosmap[i].addr;
414 change_point[chgidx++]->pbios = &biosmap[i];
415 change_point[chgidx]->addr = biosmap[i].addr + biosmap[i].size;
416 change_point[chgidx++]->pbios = &biosmap[i];
419 chg_nr = chgidx; /* true number of change-points */
421 /* sort change-point list by memory addresses (low -> high) */
423 while (still_changing) {
425 for (i=1; i < chg_nr; i++) {
426 /* if <current_addr> > <last_addr>, swap */
427 /* or, if current=<start_addr> & last=<end_addr>, swap */
428 if ((change_point[i]->addr < change_point[i-1]->addr) ||
429 ((change_point[i]->addr == change_point[i-1]->addr) &&
430 (change_point[i]->addr == change_point[i]->pbios->addr) &&
431 (change_point[i-1]->addr != change_point[i-1]->pbios->addr))
434 change_tmp = change_point[i];
435 change_point[i] = change_point[i-1];
436 change_point[i-1] = change_tmp;
442 /* create a new bios memory map, removing overlaps */
443 overlap_entries=0; /* number of entries in the overlap table */
444 new_bios_entry=0; /* index for creating new bios map entries */
445 last_type = 0; /* start with undefined memory type */
446 last_addr = 0; /* start with 0 as last starting address */
447 /* loop through change-points, determining affect on the new bios map */
448 for (chgidx=0; chgidx < chg_nr; chgidx++)
450 /* keep track of all overlapping bios entries */
451 if (change_point[chgidx]->addr == change_point[chgidx]->pbios->addr)
453 /* add map entry to overlap list (> 1 entry implies an overlap) */
454 overlap_list[overlap_entries++]=change_point[chgidx]->pbios;
458 /* remove entry from list (order independent, so swap with last) */
459 for (i=0; i<overlap_entries; i++)
461 if (overlap_list[i] == change_point[chgidx]->pbios)
462 overlap_list[i] = overlap_list[overlap_entries-1];
466 /* if there are overlapping entries, decide which "type" to use */
467 /* (larger value takes precedence -- 1=usable, 2,3,4,4+=unusable) */
469 for (i=0; i<overlap_entries; i++)
470 if (overlap_list[i]->type > current_type)
471 current_type = overlap_list[i]->type;
472 /* continue building up new bios map based on this information */
473 if (current_type != last_type) {
474 if (last_type != 0) {
475 new_bios[new_bios_entry].size =
476 change_point[chgidx]->addr - last_addr;
477 /* move forward only if the new size was non-zero */
478 if (new_bios[new_bios_entry].size != 0)
479 if (++new_bios_entry >= E820MAX)
480 break; /* no more space left for new bios entries */
482 if (current_type != 0) {
483 new_bios[new_bios_entry].addr = change_point[chgidx]->addr;
484 new_bios[new_bios_entry].type = current_type;
485 last_addr=change_point[chgidx]->addr;
487 last_type = current_type;
490 new_nr = new_bios_entry; /* retain count for new bios entries */
492 /* copy new bios mapping into original location */
493 memcpy(biosmap, new_bios, new_nr*sizeof(struct e820entry));
496 printk("sanitize end\n");
501 * Copy the BIOS e820 map into a safe place.
503 * Sanity-check it while we're at it..
505 * If we're lucky and live on a modern system, the setup code
506 * will have given us a memory map that we can use to properly
507 * set up memory. If we aren't, we'll fake a memory map.
509 * We check to see that the memory map contains at least 2 elements
510 * before we'll use it, because the detection code in setup.S may
511 * not be perfect and most every PC known to man has two memory
512 * regions: one from 0 to 640k, and one from 1mb up. (The IBM
513 * thinkpad 560x, for example, does not cooperate with the memory
516 int __init copy_e820_map(struct e820entry * biosmap, int nr_map)
518 /* Only one memory region (or negative)? Ignore it */
523 unsigned long long start = biosmap->addr;
524 unsigned long long size = biosmap->size;
525 unsigned long long end = start + size;
526 unsigned long type = biosmap->type;
527 printk("copy_e820_map() start: %016Lx size: %016Lx end: %016Lx type: %ld\n", start, size, end, type);
529 /* Overflow in 64 bits? Ignore the memory map. */
534 * Some BIOSes claim RAM in the 640k - 1M region.
535 * Not right. Fix it up.
537 if (type == E820_RAM) {
538 printk("copy_e820_map() type is E820_RAM\n");
539 if (start < 0x100000ULL && end > 0xA0000ULL) {
540 printk("copy_e820_map() lies in range...\n");
541 if (start < 0xA0000ULL) {
542 printk("copy_e820_map() start < 0xA0000ULL\n");
543 add_memory_region(start, 0xA0000ULL-start, type);
545 if (end <= 0x100000ULL) {
546 printk("copy_e820_map() end <= 0x100000ULL\n");
553 add_memory_region(start, size, type);
554 } while (biosmap++,--nr_map);
559 * Callback for efi_memory_walk.
562 efi_find_max_pfn(unsigned long start, unsigned long end, void *arg)
564 unsigned long *max_pfn = arg, pfn;
567 pfn = PFN_UP(end -1);
575 efi_memory_present_wrapper(unsigned long start, unsigned long end, void *arg)
577 memory_present(0, PFN_UP(start), PFN_DOWN(end));
582 * Find the highest page frame number we have available
584 void __init find_max_pfn(void)
590 efi_memmap_walk(efi_find_max_pfn, &max_pfn);
591 efi_memmap_walk(efi_memory_present_wrapper, NULL);
595 for (i = 0; i < e820.nr_map; i++) {
596 unsigned long start, end;
598 if (e820.map[i].type != E820_RAM)
600 start = PFN_UP(e820.map[i].addr);
601 end = PFN_DOWN(e820.map[i].addr + e820.map[i].size);
606 memory_present(0, start, end);
611 * Free all available memory for boot time allocation. Used
612 * as a callback function by efi_memory_walk()
616 free_available_memory(unsigned long start, unsigned long end, void *arg)
618 /* check max_low_pfn */
619 if (start >= (max_low_pfn << PAGE_SHIFT))
621 if (end >= (max_low_pfn << PAGE_SHIFT))
622 end = max_low_pfn << PAGE_SHIFT;
624 free_bootmem(start, end - start);
629 * Register fully available low RAM pages with the bootmem allocator.
631 void __init register_bootmem_low_pages(unsigned long max_low_pfn)
636 efi_memmap_walk(free_available_memory, NULL);
639 for (i = 0; i < e820.nr_map; i++) {
640 unsigned long curr_pfn, last_pfn, size;
642 * Reserve usable low memory
644 if (e820.map[i].type != E820_RAM)
647 * We are rounding up the start address of usable memory:
649 curr_pfn = PFN_UP(e820.map[i].addr);
650 if (curr_pfn >= max_low_pfn)
653 * ... and at the end of the usable range downwards:
655 last_pfn = PFN_DOWN(e820.map[i].addr + e820.map[i].size);
657 if (last_pfn > max_low_pfn)
658 last_pfn = max_low_pfn;
661 * .. finally, did all the rounding and playing
662 * around just make the area go away?
664 if (last_pfn <= curr_pfn)
667 size = last_pfn - curr_pfn;
668 free_bootmem(PFN_PHYS(curr_pfn), PFN_PHYS(size));
672 void __init e820_register_memory(void)
674 unsigned long gapstart, gapsize, round;
675 unsigned long long last;
679 * Search for the bigest gap in the low 32 bits of the e820
682 last = 0x100000000ull;
683 gapstart = 0x10000000;
687 unsigned long long start = e820.map[i].addr;
688 unsigned long long end = start + e820.map[i].size;
691 * Since "last" is at most 4GB, we know we'll
692 * fit in 32 bits if this condition is true
695 unsigned long gap = last - end;
707 * See how much we want to round up: start off with
708 * rounding to the next 1MB area.
711 while ((gapsize >> 4) > round)
713 /* Fun with two's complement */
714 pci_mem_start = (gapstart + round) & -round;
716 printk("Allocating PCI resources starting at %08lx (gap: %08lx:%08lx)\n",
717 pci_mem_start, gapstart, gapsize);
720 void __init print_memory_map(char *who)
724 for (i = 0; i < e820.nr_map; i++) {
725 printk(" %s: %016Lx - %016Lx ", who,
727 e820.map[i].addr + e820.map[i].size);
728 switch (e820.map[i].type) {
729 case E820_RAM: printk("(usable)\n");
732 printk("(reserved)\n");
735 printk("(ACPI data)\n");
738 printk("(ACPI NVS)\n");
740 default: printk("type %lu\n", e820.map[i].type);
746 static __init __always_inline void efi_limit_regions(unsigned long long size)
748 unsigned long long current_addr = 0;
749 efi_memory_desc_t *md, *next_md;
755 for (p = p1, i = 0; p < memmap.map_end; p += memmap.desc_size, i++) {
758 current_addr = md->phys_addr +
759 PFN_PHYS(md->num_pages);
760 if (is_available_memory(md)) {
761 if (md->phys_addr >= size) continue;
762 memcpy(next_md, md, memmap.desc_size);
763 if (current_addr >= size) {
764 next_md->num_pages -=
765 PFN_UP(current_addr-size);
767 p1 += memmap.desc_size;
770 } else if ((md->attribute & EFI_MEMORY_RUNTIME) ==
771 EFI_MEMORY_RUNTIME) {
772 /* In order to make runtime services
773 * available we have to include runtime
774 * memory regions in memory map */
775 memcpy(next_md, md, memmap.desc_size);
776 p1 += memmap.desc_size;
782 memmap.map_end = memmap.map +
783 (memmap.nr_map * memmap.desc_size);
786 void __init limit_regions(unsigned long long size)
788 unsigned long long current_addr;
791 print_memory_map("limit_regions start");
793 efi_limit_regions(size);
796 for (i = 0; i < e820.nr_map; i++) {
797 current_addr = e820.map[i].addr + e820.map[i].size;
798 if (current_addr < size)
801 if (e820.map[i].type != E820_RAM)
804 if (e820.map[i].addr >= size) {
806 * This region starts past the end of the
807 * requested size, skip it completely.
812 e820.map[i].size -= current_addr - size;
814 print_memory_map("limit_regions endfor");
817 print_memory_map("limit_regions endfunc");
821 * This function checks if the entire range <start,end> is mapped with type.
823 * Note: this function only works correct if the e820 table is sorted and
824 * not-overlapping, which is the case
827 e820_all_mapped(unsigned long s, unsigned long e, unsigned type)
832 for (i = 0; i < e820.nr_map; i++) {
833 struct e820entry *ei = &e820.map[i];
834 if (type && ei->type != type)
836 /* is the region (part) in overlap with the current region ?*/
837 if (ei->addr >= end || ei->addr + ei->size <= start)
839 /* if the region is at the beginning of <start,end> we move
840 * start to the end of the region since it's ok until there
842 if (ei->addr <= start)
843 start = ei->addr + ei->size;
844 /* if start is now at or beyond end, we're done, full
847 return 1; /* we're done */
852 static int __init parse_memmap(char *arg)
857 if (strcmp(arg, "exactmap") == 0) {
858 #ifdef CONFIG_CRASH_DUMP
859 /* If we are doing a crash dump, we
860 * still need to know the real mem
861 * size before original memory map is
865 saved_max_pfn = max_pfn;
868 user_defined_memmap = 1;
870 /* If the user specifies memory size, we
871 * limit the BIOS-provided memory map to
872 * that size. exactmap can be used to specify
873 * the exact map. mem=number can be used to
874 * trim the existing memory map.
876 unsigned long long start_at, mem_size;
878 mem_size = memparse(arg, &arg);
880 start_at = memparse(arg+1, &arg);
881 add_memory_region(start_at, mem_size, E820_RAM);
882 } else if (*arg == '#') {
883 start_at = memparse(arg+1, &arg);
884 add_memory_region(start_at, mem_size, E820_ACPI);
885 } else if (*arg == '$') {
886 start_at = memparse(arg+1, &arg);
887 add_memory_region(start_at, mem_size, E820_RESERVED);
889 limit_regions(mem_size);
890 user_defined_memmap = 1;
895 early_param("memmap", parse_memmap);