2 * Handle the memory map.
3 * The functions here do the job until bootmem takes over.
5 * Getting sanitize_e820_map() in sync with i386 version by applying change:
6 * - Provisions for empty E820 memory regions (reported by certain BIOSes).
7 * Alex Achenbach <xela@slit.de>, December 2002.
8 * Venkatesh Pallipadi <venkatesh.pallipadi@intel.com>
11 #include <linux/kernel.h>
12 #include <linux/types.h>
13 #include <linux/init.h>
14 #include <linux/crash_dump.h>
15 #include <linux/export.h>
16 #include <linux/bootmem.h>
17 #include <linux/pfn.h>
18 #include <linux/suspend.h>
19 #include <linux/acpi.h>
20 #include <linux/firmware-map.h>
21 #include <linux/memblock.h>
22 #include <linux/sort.h>
25 #include <asm/proto.h>
26 #include <asm/setup.h>
27 #include <asm/cpufeature.h>
30 * The e820 map is the map that gets modified e.g. with command line parameters
31 * and that is also registered with modifications in the kernel resource tree
32 * with the iomem_resource as parent.
34 * The e820_saved is directly saved after the BIOS-provided memory map is
35 * copied. It doesn't get modified afterwards. It's registered for the
36 * /sys/firmware/memmap interface.
38 * That memory map is not modified and is used as base for kexec. The kexec'd
39 * kernel should get the same memory map as the firmware provides. Then the
40 * user can e.g. boot the original kernel with mem=1G while still booting the
41 * next kernel with full memory.
43 static struct e820map initial_e820 __initdata;
44 static struct e820map initial_e820_saved __initdata;
45 struct e820map *e820 __refdata = &initial_e820;
46 struct e820map *e820_saved __refdata = &initial_e820_saved;
48 /* For PCI or other memory-mapped resources */
49 unsigned long pci_mem_start = 0xaeedbabe;
51 EXPORT_SYMBOL(pci_mem_start);
55 * This function checks if any part of the range <start,end> is mapped
59 e820_any_mapped(u64 start, u64 end, unsigned type)
63 for (i = 0; i < e820->nr_map; i++) {
64 struct e820entry *ei = &e820->map[i];
66 if (type && ei->type != type)
68 if (ei->addr >= end || ei->addr + ei->size <= start)
74 EXPORT_SYMBOL_GPL(e820_any_mapped);
77 * This function checks if the entire range <start,end> is mapped with type.
79 * Note: this function only works correct if the e820 table is sorted and
80 * not-overlapping, which is the case
82 int __init e820_all_mapped(u64 start, u64 end, unsigned type)
86 for (i = 0; i < e820->nr_map; i++) {
87 struct e820entry *ei = &e820->map[i];
89 if (type && ei->type != type)
91 /* is the region (part) in overlap with the current region ?*/
92 if (ei->addr >= end || ei->addr + ei->size <= start)
95 /* if the region is at the beginning of <start,end> we move
96 * start to the end of the region since it's ok until there
98 if (ei->addr <= start)
99 start = ei->addr + ei->size;
101 * if start is now at or beyond end, we're done, full
111 * Add a memory region to the kernel e820 map.
113 static void __init __e820_add_region(struct e820map *e820x, u64 start, u64 size,
116 int x = e820x->nr_map;
118 if (x >= ARRAY_SIZE(e820x->map)) {
119 printk(KERN_ERR "e820: too many entries; ignoring [mem %#010llx-%#010llx]\n",
120 (unsigned long long) start,
121 (unsigned long long) (start + size - 1));
125 e820x->map[x].addr = start;
126 e820x->map[x].size = size;
127 e820x->map[x].type = type;
131 void __init e820_add_region(u64 start, u64 size, int type)
133 __e820_add_region(e820, start, size, type);
136 static void __init e820_print_type(u32 type)
140 case E820_RESERVED_KERN:
141 printk(KERN_CONT "usable");
144 printk(KERN_CONT "reserved");
147 printk(KERN_CONT "ACPI data");
150 printk(KERN_CONT "ACPI NVS");
153 printk(KERN_CONT "unusable");
157 printk(KERN_CONT "persistent (type %u)", type);
160 printk(KERN_CONT "type %u", type);
165 void __init e820_print_map(char *who)
169 for (i = 0; i < e820->nr_map; i++) {
170 printk(KERN_INFO "%s: [mem %#018Lx-%#018Lx] ", who,
171 (unsigned long long) e820->map[i].addr,
173 (e820->map[i].addr + e820->map[i].size - 1));
174 e820_print_type(e820->map[i].type);
175 printk(KERN_CONT "\n");
180 * Sanitize the BIOS e820 map.
182 * Some e820 responses include overlapping entries. The following
183 * replaces the original e820 map with a new one, removing overlaps,
184 * and resolving conflicting memory types in favor of highest
187 * The input parameter biosmap points to an array of 'struct
188 * e820entry' which on entry has elements in the range [0, *pnr_map)
189 * valid, and which has space for up to max_nr_map entries.
190 * On return, the resulting sanitized e820 map entries will be in
191 * overwritten in the same location, starting at biosmap.
193 * The integer pointed to by pnr_map must be valid on entry (the
194 * current number of valid entries located at biosmap). If the
195 * sanitizing succeeds the *pnr_map will be updated with the new
196 * number of valid entries (something no more than max_nr_map).
198 * The return value from sanitize_e820_map() is zero if it
199 * successfully 'sanitized' the map entries passed in, and is -1
200 * if it did nothing, which can happen if either of (1) it was
201 * only passed one map entry, or (2) any of the input map entries
202 * were invalid (start + size < start, meaning that the size was
203 * so big the described memory range wrapped around through zero.)
205 * Visually we're performing the following
206 * (1,2,3,4 = memory types)...
208 * Sample memory map (w/overlaps):
209 * ____22__________________
210 * ______________________4_
211 * ____1111________________
212 * _44_____________________
213 * 11111111________________
214 * ____________________33__
215 * ___________44___________
216 * __________33333_________
217 * ______________22________
218 * ___________________2222_
219 * _________111111111______
220 * _____________________11_
221 * _________________4______
223 * Sanitized equivalent (no overlap):
224 * 1_______________________
225 * _44_____________________
226 * ___1____________________
227 * ____22__________________
228 * ______11________________
229 * _________1______________
230 * __________3_____________
231 * ___________44___________
232 * _____________33_________
233 * _______________2________
234 * ________________1_______
235 * _________________4______
236 * ___________________2____
237 * ____________________33__
238 * ______________________4_
240 struct change_member {
241 struct e820entry *pbios; /* pointer to original bios entry */
242 unsigned long long addr; /* address for this change point */
245 static int __init cpcompare(const void *a, const void *b)
247 struct change_member * const *app = a, * const *bpp = b;
248 const struct change_member *ap = *app, *bp = *bpp;
251 * Inputs are pointers to two elements of change_point[]. If their
252 * addresses are unequal, their difference dominates. If the addresses
253 * are equal, then consider one that represents the end of its region
254 * to be greater than one that does not.
256 if (ap->addr != bp->addr)
257 return ap->addr > bp->addr ? 1 : -1;
259 return (ap->addr != ap->pbios->addr) - (bp->addr != bp->pbios->addr);
262 int __init sanitize_e820_map(struct e820entry *biosmap, int max_nr_map,
265 static struct change_member change_point_list[2*E820_X_MAX] __initdata;
266 static struct change_member *change_point[2*E820_X_MAX] __initdata;
267 static struct e820entry *overlap_list[E820_X_MAX] __initdata;
268 static struct e820entry new_bios[E820_X_MAX] __initdata;
269 unsigned long current_type, last_type;
270 unsigned long long last_addr;
274 int old_nr, new_nr, chg_nr;
277 /* if there's only one memory region, don't bother */
282 BUG_ON(old_nr > max_nr_map);
284 /* bail out if we find any unreasonable addresses in bios map */
285 for (i = 0; i < old_nr; i++)
286 if (biosmap[i].addr + biosmap[i].size < biosmap[i].addr)
289 /* create pointers for initial change-point information (for sorting) */
290 for (i = 0; i < 2 * old_nr; i++)
291 change_point[i] = &change_point_list[i];
293 /* record all known change-points (starting and ending addresses),
294 omitting those that are for empty memory regions */
296 for (i = 0; i < old_nr; i++) {
297 if (biosmap[i].size != 0) {
298 change_point[chgidx]->addr = biosmap[i].addr;
299 change_point[chgidx++]->pbios = &biosmap[i];
300 change_point[chgidx]->addr = biosmap[i].addr +
302 change_point[chgidx++]->pbios = &biosmap[i];
307 /* sort change-point list by memory addresses (low -> high) */
308 sort(change_point, chg_nr, sizeof *change_point, cpcompare, NULL);
310 /* create a new bios memory map, removing overlaps */
311 overlap_entries = 0; /* number of entries in the overlap table */
312 new_bios_entry = 0; /* index for creating new bios map entries */
313 last_type = 0; /* start with undefined memory type */
314 last_addr = 0; /* start with 0 as last starting address */
316 /* loop through change-points, determining affect on the new bios map */
317 for (chgidx = 0; chgidx < chg_nr; chgidx++) {
318 /* keep track of all overlapping bios entries */
319 if (change_point[chgidx]->addr ==
320 change_point[chgidx]->pbios->addr) {
322 * add map entry to overlap list (> 1 entry
323 * implies an overlap)
325 overlap_list[overlap_entries++] =
326 change_point[chgidx]->pbios;
329 * remove entry from list (order independent,
332 for (i = 0; i < overlap_entries; i++) {
333 if (overlap_list[i] ==
334 change_point[chgidx]->pbios)
336 overlap_list[overlap_entries-1];
341 * if there are overlapping entries, decide which
342 * "type" to use (larger value takes precedence --
343 * 1=usable, 2,3,4,4+=unusable)
346 for (i = 0; i < overlap_entries; i++)
347 if (overlap_list[i]->type > current_type)
348 current_type = overlap_list[i]->type;
350 * continue building up new bios map based on this
353 if (current_type != last_type || current_type == E820_PRAM) {
354 if (last_type != 0) {
355 new_bios[new_bios_entry].size =
356 change_point[chgidx]->addr - last_addr;
358 * move forward only if the new size
361 if (new_bios[new_bios_entry].size != 0)
363 * no more space left for new
366 if (++new_bios_entry >= max_nr_map)
369 if (current_type != 0) {
370 new_bios[new_bios_entry].addr =
371 change_point[chgidx]->addr;
372 new_bios[new_bios_entry].type = current_type;
373 last_addr = change_point[chgidx]->addr;
375 last_type = current_type;
378 /* retain count for new bios entries */
379 new_nr = new_bios_entry;
381 /* copy new bios mapping into original location */
382 memcpy(biosmap, new_bios, new_nr * sizeof(struct e820entry));
388 static int __init __append_e820_map(struct e820entry *biosmap, int nr_map)
391 u64 start = biosmap->addr;
392 u64 size = biosmap->size;
393 u64 end = start + size - 1;
394 u32 type = biosmap->type;
396 /* Overflow in 64 bits? Ignore the memory map. */
397 if (start > end && likely(size))
400 e820_add_region(start, size, type);
409 * Copy the BIOS e820 map into a safe place.
411 * Sanity-check it while we're at it..
413 * If we're lucky and live on a modern system, the setup code
414 * will have given us a memory map that we can use to properly
415 * set up memory. If we aren't, we'll fake a memory map.
417 static int __init append_e820_map(struct e820entry *biosmap, int nr_map)
419 /* Only one memory region (or negative)? Ignore it */
423 return __append_e820_map(biosmap, nr_map);
426 static u64 __init __e820_update_range(struct e820map *e820x, u64 start,
427 u64 size, unsigned old_type,
432 u64 real_updated_size = 0;
434 BUG_ON(old_type == new_type);
436 if (size > (ULLONG_MAX - start))
437 size = ULLONG_MAX - start;
440 printk(KERN_DEBUG "e820: update [mem %#010Lx-%#010Lx] ",
441 (unsigned long long) start, (unsigned long long) (end - 1));
442 e820_print_type(old_type);
443 printk(KERN_CONT " ==> ");
444 e820_print_type(new_type);
445 printk(KERN_CONT "\n");
447 for (i = 0; i < e820x->nr_map; i++) {
448 struct e820entry *ei = &e820x->map[i];
449 u64 final_start, final_end;
452 if (ei->type != old_type)
455 ei_end = ei->addr + ei->size;
456 /* totally covered by new range? */
457 if (ei->addr >= start && ei_end <= end) {
459 real_updated_size += ei->size;
463 /* new range is totally covered? */
464 if (ei->addr < start && ei_end > end) {
465 __e820_add_region(e820x, start, size, new_type);
466 __e820_add_region(e820x, end, ei_end - end, ei->type);
467 ei->size = start - ei->addr;
468 real_updated_size += size;
472 /* partially covered */
473 final_start = max(start, ei->addr);
474 final_end = min(end, ei_end);
475 if (final_start >= final_end)
478 __e820_add_region(e820x, final_start, final_end - final_start,
481 real_updated_size += final_end - final_start;
484 * left range could be head or tail, so need to update
487 ei->size -= final_end - final_start;
488 if (ei->addr < final_start)
490 ei->addr = final_end;
492 return real_updated_size;
495 u64 __init e820_update_range(u64 start, u64 size, unsigned old_type,
498 return __e820_update_range(e820, start, size, old_type, new_type);
501 static u64 __init e820_update_range_saved(u64 start, u64 size,
502 unsigned old_type, unsigned new_type)
504 return __e820_update_range(e820_saved, start, size, old_type,
508 /* make e820 not cover the range */
509 u64 __init e820_remove_range(u64 start, u64 size, unsigned old_type,
514 u64 real_removed_size = 0;
516 if (size > (ULLONG_MAX - start))
517 size = ULLONG_MAX - start;
520 printk(KERN_DEBUG "e820: remove [mem %#010Lx-%#010Lx] ",
521 (unsigned long long) start, (unsigned long long) (end - 1));
523 e820_print_type(old_type);
524 printk(KERN_CONT "\n");
526 for (i = 0; i < e820->nr_map; i++) {
527 struct e820entry *ei = &e820->map[i];
528 u64 final_start, final_end;
531 if (checktype && ei->type != old_type)
534 ei_end = ei->addr + ei->size;
535 /* totally covered? */
536 if (ei->addr >= start && ei_end <= end) {
537 real_removed_size += ei->size;
538 memset(ei, 0, sizeof(struct e820entry));
542 /* new range is totally covered? */
543 if (ei->addr < start && ei_end > end) {
544 e820_add_region(end, ei_end - end, ei->type);
545 ei->size = start - ei->addr;
546 real_removed_size += size;
550 /* partially covered */
551 final_start = max(start, ei->addr);
552 final_end = min(end, ei_end);
553 if (final_start >= final_end)
555 real_removed_size += final_end - final_start;
558 * left range could be head or tail, so need to update
561 ei->size -= final_end - final_start;
562 if (ei->addr < final_start)
564 ei->addr = final_end;
566 return real_removed_size;
569 void __init update_e820(void)
571 if (sanitize_e820_map(e820->map, ARRAY_SIZE(e820->map), &e820->nr_map))
573 printk(KERN_INFO "e820: modified physical RAM map:\n");
574 e820_print_map("modified");
576 static void __init update_e820_saved(void)
578 sanitize_e820_map(e820_saved->map, ARRAY_SIZE(e820_saved->map),
579 &e820_saved->nr_map);
581 #define MAX_GAP_END 0x100000000ull
583 * Search for a gap in the e820 memory space from start_addr to end_addr.
585 __init int e820_search_gap(unsigned long *gapstart, unsigned long *gapsize,
586 unsigned long start_addr, unsigned long long end_addr)
588 unsigned long long last;
589 int i = e820->nr_map;
592 last = (end_addr && end_addr < MAX_GAP_END) ? end_addr : MAX_GAP_END;
595 unsigned long long start = e820->map[i].addr;
596 unsigned long long end = start + e820->map[i].size;
598 if (end < start_addr)
602 * Since "last" is at most 4GB, we know we'll
603 * fit in 32 bits if this condition is true
606 unsigned long gap = last - end;
608 if (gap >= *gapsize) {
621 * Search for the biggest gap in the low 32 bits of the e820
622 * memory space. We pass this space to PCI to assign MMIO resources
623 * for hotplug or unconfigured devices in.
624 * Hopefully the BIOS let enough space left.
626 __init void e820_setup_gap(void)
628 unsigned long gapstart, gapsize;
631 gapstart = 0x10000000;
633 found = e820_search_gap(&gapstart, &gapsize, 0, MAX_GAP_END);
637 gapstart = (max_pfn << PAGE_SHIFT) + 1024*1024;
639 "e820: cannot find a gap in the 32bit address range\n"
640 "e820: PCI devices with unassigned 32bit BARs may break!\n");
645 * e820_reserve_resources_late protect stolen RAM already
647 pci_mem_start = gapstart;
650 "e820: [mem %#010lx-%#010lx] available for PCI devices\n",
651 gapstart, gapstart + gapsize - 1);
655 * Called late during init, in free_initmem().
657 * Initial e820 and e820_saved are largish __initdata arrays.
658 * Copy them to (usually much smaller) dynamically allocated area.
659 * This is done after all tweaks we ever do to them:
660 * all functions which modify them are __init functions,
661 * they won't exist after this point.
663 __init void e820_reallocate_tables(void)
668 size = offsetof(struct e820map, map) + sizeof(struct e820entry) * e820->nr_map;
669 n = kmalloc(size, GFP_KERNEL);
671 memcpy(n, e820, size);
674 size = offsetof(struct e820map, map) + sizeof(struct e820entry) * e820_saved->nr_map;
675 n = kmalloc(size, GFP_KERNEL);
677 memcpy(n, e820_saved, size);
682 * Because of the size limitation of struct boot_params, only first
683 * 128 E820 memory entries are passed to kernel via
684 * boot_params.e820_map, others are passed via SETUP_E820_EXT node of
685 * linked list of struct setup_data, which is parsed here.
687 void __init parse_e820_ext(u64 phys_addr, u32 data_len)
690 struct e820entry *extmap;
691 struct setup_data *sdata;
693 sdata = early_memremap(phys_addr, data_len);
694 entries = sdata->len / sizeof(struct e820entry);
695 extmap = (struct e820entry *)(sdata->data);
696 __append_e820_map(extmap, entries);
697 sanitize_e820_map(e820->map, ARRAY_SIZE(e820->map), &e820->nr_map);
698 early_memunmap(sdata, data_len);
699 printk(KERN_INFO "e820: extended physical RAM map:\n");
700 e820_print_map("extended");
703 #if defined(CONFIG_X86_64) || \
704 (defined(CONFIG_X86_32) && defined(CONFIG_HIBERNATION))
706 * Find the ranges of physical addresses that do not correspond to
707 * e820 RAM areas and mark the corresponding pages as nosave for
708 * hibernation (32 bit) or software suspend and suspend to RAM (64 bit).
710 * This function requires the e820 map to be sorted and without any
711 * overlapping entries.
713 void __init e820_mark_nosave_regions(unsigned long limit_pfn)
716 unsigned long pfn = 0;
718 for (i = 0; i < e820->nr_map; i++) {
719 struct e820entry *ei = &e820->map[i];
721 if (pfn < PFN_UP(ei->addr))
722 register_nosave_region(pfn, PFN_UP(ei->addr));
724 pfn = PFN_DOWN(ei->addr + ei->size);
726 if (ei->type != E820_RAM && ei->type != E820_RESERVED_KERN)
727 register_nosave_region(PFN_UP(ei->addr), pfn);
729 if (pfn >= limit_pfn)
737 * Mark ACPI NVS memory region, so that we can save/restore it during
738 * hibernation and the subsequent resume.
740 static int __init e820_mark_nvs_memory(void)
744 for (i = 0; i < e820->nr_map; i++) {
745 struct e820entry *ei = &e820->map[i];
747 if (ei->type == E820_NVS)
748 acpi_nvs_register(ei->addr, ei->size);
753 core_initcall(e820_mark_nvs_memory);
757 * pre allocated 4k and reserved it in memblock and e820_saved
759 u64 __init early_reserve_e820(u64 size, u64 align)
763 addr = __memblock_alloc_base(size, align, MEMBLOCK_ALLOC_ACCESSIBLE);
765 e820_update_range_saved(addr, size, E820_RAM, E820_RESERVED);
766 printk(KERN_INFO "e820: update e820_saved for early_reserve_e820\n");
774 # ifdef CONFIG_X86_PAE
775 # define MAX_ARCH_PFN (1ULL<<(36-PAGE_SHIFT))
777 # define MAX_ARCH_PFN (1ULL<<(32-PAGE_SHIFT))
779 #else /* CONFIG_X86_32 */
780 # define MAX_ARCH_PFN MAXMEM>>PAGE_SHIFT
784 * Find the highest page frame number we have available
786 static unsigned long __init e820_end_pfn(unsigned long limit_pfn, unsigned type)
789 unsigned long last_pfn = 0;
790 unsigned long max_arch_pfn = MAX_ARCH_PFN;
792 for (i = 0; i < e820->nr_map; i++) {
793 struct e820entry *ei = &e820->map[i];
794 unsigned long start_pfn;
795 unsigned long end_pfn;
797 if (ei->type != type)
800 start_pfn = ei->addr >> PAGE_SHIFT;
801 end_pfn = (ei->addr + ei->size) >> PAGE_SHIFT;
803 if (start_pfn >= limit_pfn)
805 if (end_pfn > limit_pfn) {
806 last_pfn = limit_pfn;
809 if (end_pfn > last_pfn)
813 if (last_pfn > max_arch_pfn)
814 last_pfn = max_arch_pfn;
816 printk(KERN_INFO "e820: last_pfn = %#lx max_arch_pfn = %#lx\n",
817 last_pfn, max_arch_pfn);
820 unsigned long __init e820_end_of_ram_pfn(void)
822 return e820_end_pfn(MAX_ARCH_PFN, E820_RAM);
825 unsigned long __init e820_end_of_low_ram_pfn(void)
827 return e820_end_pfn(1UL << (32 - PAGE_SHIFT), E820_RAM);
830 static void __init early_panic(char *msg)
836 static int userdef __initdata;
838 /* "mem=nopentium" disables the 4MB page tables. */
839 static int __init parse_memopt(char *p)
846 if (!strcmp(p, "nopentium")) {
848 setup_clear_cpu_cap(X86_FEATURE_PSE);
851 printk(KERN_WARNING "mem=nopentium ignored! (only supported on x86_32)\n");
857 mem_size = memparse(p, &p);
858 /* don't remove all of memory when handling "mem={invalid}" param */
861 e820_remove_range(mem_size, ULLONG_MAX - mem_size, E820_RAM, 1);
865 early_param("mem", parse_memopt);
867 static int __init parse_memmap_one(char *p)
870 u64 start_at, mem_size;
875 if (!strncmp(p, "exactmap", 8)) {
876 #ifdef CONFIG_CRASH_DUMP
878 * If we are doing a crash dump, we still need to know
879 * the real mem size before original memory map is
882 saved_max_pfn = e820_end_of_ram_pfn();
890 mem_size = memparse(p, &p);
896 start_at = memparse(p+1, &p);
897 e820_add_region(start_at, mem_size, E820_RAM);
898 } else if (*p == '#') {
899 start_at = memparse(p+1, &p);
900 e820_add_region(start_at, mem_size, E820_ACPI);
901 } else if (*p == '$') {
902 start_at = memparse(p+1, &p);
903 e820_add_region(start_at, mem_size, E820_RESERVED);
904 } else if (*p == '!') {
905 start_at = memparse(p+1, &p);
906 e820_add_region(start_at, mem_size, E820_PRAM);
908 e820_remove_range(mem_size, ULLONG_MAX - mem_size, E820_RAM, 1);
910 return *p == '\0' ? 0 : -EINVAL;
912 static int __init parse_memmap_opt(char *str)
915 char *k = strchr(str, ',');
920 parse_memmap_one(str);
926 early_param("memmap", parse_memmap_opt);
928 void __init finish_e820_parsing(void)
931 if (sanitize_e820_map(e820->map, ARRAY_SIZE(e820->map),
933 early_panic("Invalid user supplied memory map");
935 printk(KERN_INFO "e820: user-defined physical RAM map:\n");
936 e820_print_map("user");
940 static const char *__init e820_type_to_string(int e820_type)
943 case E820_RESERVED_KERN:
944 case E820_RAM: return "System RAM";
945 case E820_ACPI: return "ACPI Tables";
946 case E820_NVS: return "ACPI Non-volatile Storage";
947 case E820_UNUSABLE: return "Unusable memory";
948 case E820_PRAM: return "Persistent Memory (legacy)";
949 case E820_PMEM: return "Persistent Memory";
950 default: return "reserved";
954 static unsigned long __init e820_type_to_iomem_type(int e820_type)
957 case E820_RESERVED_KERN:
959 return IORESOURCE_SYSTEM_RAM;
966 return IORESOURCE_MEM;
970 static unsigned long __init e820_type_to_iores_desc(int e820_type)
974 return IORES_DESC_ACPI_TABLES;
976 return IORES_DESC_ACPI_NV_STORAGE;
978 return IORES_DESC_PERSISTENT_MEMORY;
980 return IORES_DESC_PERSISTENT_MEMORY_LEGACY;
981 case E820_RESERVED_KERN:
985 return IORES_DESC_NONE;
989 static bool __init do_mark_busy(u32 type, struct resource *res)
991 /* this is the legacy bios/dos rom-shadow + mmio region */
992 if (res->start < (1ULL<<20))
996 * Treat persistent memory like device memory, i.e. reserve it
997 * for exclusive use of a driver
1010 * Mark e820 reserved areas as busy for the resource manager.
1012 static struct resource __initdata *e820_res;
1013 void __init e820_reserve_resources(void)
1016 struct resource *res;
1019 res = alloc_bootmem(sizeof(struct resource) * e820->nr_map);
1021 for (i = 0; i < e820->nr_map; i++) {
1022 end = e820->map[i].addr + e820->map[i].size - 1;
1023 if (end != (resource_size_t)end) {
1027 res->name = e820_type_to_string(e820->map[i].type);
1028 res->start = e820->map[i].addr;
1031 res->flags = e820_type_to_iomem_type(e820->map[i].type);
1032 res->desc = e820_type_to_iores_desc(e820->map[i].type);
1035 * don't register the region that could be conflicted with
1036 * pci device BAR resource and insert them later in
1037 * pcibios_resource_survey()
1039 if (do_mark_busy(e820->map[i].type, res)) {
1040 res->flags |= IORESOURCE_BUSY;
1041 insert_resource(&iomem_resource, res);
1046 for (i = 0; i < e820_saved->nr_map; i++) {
1047 struct e820entry *entry = &e820_saved->map[i];
1048 firmware_map_add_early(entry->addr,
1049 entry->addr + entry->size,
1050 e820_type_to_string(entry->type));
1054 /* How much should we pad RAM ending depending on where it is? */
1055 static unsigned long __init ram_alignment(resource_size_t pos)
1057 unsigned long mb = pos >> 20;
1059 /* To 64kB in the first megabyte */
1063 /* To 1MB in the first 16MB */
1067 /* To 64MB for anything above that */
1068 return 64*1024*1024;
1071 #define MAX_RESOURCE_SIZE ((resource_size_t)-1)
1073 void __init e820_reserve_resources_late(void)
1076 struct resource *res;
1079 for (i = 0; i < e820->nr_map; i++) {
1080 if (!res->parent && res->end)
1081 insert_resource_expand_to_fit(&iomem_resource, res);
1086 * Try to bump up RAM regions to reasonable boundaries to
1089 for (i = 0; i < e820->nr_map; i++) {
1090 struct e820entry *entry = &e820->map[i];
1093 if (entry->type != E820_RAM)
1095 start = entry->addr + entry->size;
1096 end = round_up(start, ram_alignment(start)) - 1;
1097 if (end > MAX_RESOURCE_SIZE)
1098 end = MAX_RESOURCE_SIZE;
1102 "e820: reserve RAM buffer [mem %#010llx-%#010llx]\n",
1104 reserve_region_with_split(&iomem_resource, start, end,
1109 char *__init default_machine_specific_memory_setup(void)
1111 char *who = "BIOS-e820";
1114 * Try to copy the BIOS-supplied E820-map.
1116 * Otherwise fake a memory map; one section from 0k->640k,
1117 * the next section from 1mb->appropriate_mem_k
1119 new_nr = boot_params.e820_entries;
1120 sanitize_e820_map(boot_params.e820_map,
1121 ARRAY_SIZE(boot_params.e820_map),
1123 boot_params.e820_entries = new_nr;
1124 if (append_e820_map(boot_params.e820_map, boot_params.e820_entries)
1128 /* compare results from other methods and take the greater */
1129 if (boot_params.alt_mem_k
1130 < boot_params.screen_info.ext_mem_k) {
1131 mem_size = boot_params.screen_info.ext_mem_k;
1134 mem_size = boot_params.alt_mem_k;
1139 e820_add_region(0, LOWMEMSIZE(), E820_RAM);
1140 e820_add_region(HIGH_MEMORY, mem_size << 10, E820_RAM);
1143 /* In case someone cares... */
1147 void __init setup_memory_map(void)
1151 who = x86_init.resources.memory_setup();
1152 memcpy(e820_saved, e820, sizeof(struct e820map));
1153 printk(KERN_INFO "e820: BIOS-provided physical RAM map:\n");
1154 e820_print_map(who);
1157 void __init memblock_x86_fill(void)
1163 * EFI may have more than 128 entries
1164 * We are safe to enable resizing, beause memblock_x86_fill()
1165 * is rather later for x86
1167 memblock_allow_resize();
1169 for (i = 0; i < e820->nr_map; i++) {
1170 struct e820entry *ei = &e820->map[i];
1172 end = ei->addr + ei->size;
1173 if (end != (resource_size_t)end)
1176 if (ei->type != E820_RAM && ei->type != E820_RESERVED_KERN)
1179 memblock_add(ei->addr, ei->size);
1182 /* throw away partial pages */
1183 memblock_trim_memory(PAGE_SIZE);
1185 memblock_dump_all();
1188 void __init memblock_find_dma_reserve(void)
1190 #ifdef CONFIG_X86_64
1191 u64 nr_pages = 0, nr_free_pages = 0;
1192 unsigned long start_pfn, end_pfn;
1193 phys_addr_t start, end;
1198 * need to find out used area below MAX_DMA_PFN
1199 * need to use memblock to get free size in [0, MAX_DMA_PFN]
1200 * at first, and assume boot_mem will not take below MAX_DMA_PFN
1202 for_each_mem_pfn_range(i, MAX_NUMNODES, &start_pfn, &end_pfn, NULL) {
1203 start_pfn = min(start_pfn, MAX_DMA_PFN);
1204 end_pfn = min(end_pfn, MAX_DMA_PFN);
1205 nr_pages += end_pfn - start_pfn;
1208 for_each_free_mem_range(u, NUMA_NO_NODE, MEMBLOCK_NONE, &start, &end,
1210 start_pfn = min_t(unsigned long, PFN_UP(start), MAX_DMA_PFN);
1211 end_pfn = min_t(unsigned long, PFN_DOWN(end), MAX_DMA_PFN);
1212 if (start_pfn < end_pfn)
1213 nr_free_pages += end_pfn - start_pfn;
1216 set_dma_reserve(nr_pages - nr_free_pages);