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/bootmem.h>
15 #include <linux/ioport.h>
16 #include <linux/string.h>
17 #include <linux/kexec.h>
18 #include <linux/module.h>
20 #include <linux/pfn.h>
21 #include <linux/suspend.h>
23 #include <asm/pgtable.h>
26 #include <asm/proto.h>
27 #include <asm/setup.h>
28 #include <asm/trampoline.h>
32 /* For PCI or other memory-mapped resources */
33 unsigned long pci_mem_start = 0xaeedbabe;
35 EXPORT_SYMBOL(pci_mem_start);
39 * This function checks if any part of the range <start,end> is mapped
43 e820_any_mapped(u64 start, u64 end, unsigned type)
47 for (i = 0; i < e820.nr_map; i++) {
48 struct e820entry *ei = &e820.map[i];
50 if (type && ei->type != type)
52 if (ei->addr >= end || ei->addr + ei->size <= start)
58 EXPORT_SYMBOL_GPL(e820_any_mapped);
61 * This function checks if the entire range <start,end> is mapped with type.
63 * Note: this function only works correct if the e820 table is sorted and
64 * not-overlapping, which is the case
66 int __init e820_all_mapped(u64 start, u64 end, unsigned type)
70 for (i = 0; i < e820.nr_map; i++) {
71 struct e820entry *ei = &e820.map[i];
73 if (type && ei->type != type)
75 /* is the region (part) in overlap with the current region ?*/
76 if (ei->addr >= end || ei->addr + ei->size <= start)
79 /* if the region is at the beginning of <start,end> we move
80 * start to the end of the region since it's ok until there
82 if (ei->addr <= start)
83 start = ei->addr + ei->size;
85 * if start is now at or beyond end, we're done, full
95 * Add a memory region to the kernel e820 map.
97 void __init e820_add_region(u64 start, u64 size, int type)
101 if (x == ARRAY_SIZE(e820.map)) {
102 printk(KERN_ERR "Ooops! Too many entries in the memory map!\n");
106 e820.map[x].addr = start;
107 e820.map[x].size = size;
108 e820.map[x].type = type;
112 void __init e820_print_map(char *who)
116 for (i = 0; i < e820.nr_map; i++) {
117 printk(KERN_INFO " %s: %016Lx - %016Lx ", who,
118 (unsigned long long) e820.map[i].addr,
120 (e820.map[i].addr + e820.map[i].size));
121 switch (e820.map[i].type) {
123 case E820_RESERVED_KERN:
124 printk(KERN_CONT "(usable)\n");
127 printk(KERN_CONT "(reserved)\n");
130 printk(KERN_CONT "(ACPI data)\n");
133 printk(KERN_CONT "(ACPI NVS)\n");
136 printk(KERN_CONT "type %u\n", e820.map[i].type);
143 * Sanitize the BIOS e820 map.
145 * Some e820 responses include overlapping entries. The following
146 * replaces the original e820 map with a new one, removing overlaps,
147 * and resolving conflicting memory types in favor of highest
150 * The input parameter biosmap points to an array of 'struct
151 * e820entry' which on entry has elements in the range [0, *pnr_map)
152 * valid, and which has space for up to max_nr_map entries.
153 * On return, the resulting sanitized e820 map entries will be in
154 * overwritten in the same location, starting at biosmap.
156 * The integer pointed to by pnr_map must be valid on entry (the
157 * current number of valid entries located at biosmap) and will
158 * be updated on return, with the new number of valid entries
159 * (something no more than max_nr_map.)
161 * The return value from sanitize_e820_map() is zero if it
162 * successfully 'sanitized' the map entries passed in, and is -1
163 * if it did nothing, which can happen if either of (1) it was
164 * only passed one map entry, or (2) any of the input map entries
165 * were invalid (start + size < start, meaning that the size was
166 * so big the described memory range wrapped around through zero.)
168 * Visually we're performing the following
169 * (1,2,3,4 = memory types)...
171 * Sample memory map (w/overlaps):
172 * ____22__________________
173 * ______________________4_
174 * ____1111________________
175 * _44_____________________
176 * 11111111________________
177 * ____________________33__
178 * ___________44___________
179 * __________33333_________
180 * ______________22________
181 * ___________________2222_
182 * _________111111111______
183 * _____________________11_
184 * _________________4______
186 * Sanitized equivalent (no overlap):
187 * 1_______________________
188 * _44_____________________
189 * ___1____________________
190 * ____22__________________
191 * ______11________________
192 * _________1______________
193 * __________3_____________
194 * ___________44___________
195 * _____________33_________
196 * _______________2________
197 * ________________1_______
198 * _________________4______
199 * ___________________2____
200 * ____________________33__
201 * ______________________4_
204 int __init sanitize_e820_map(struct e820entry *biosmap, int max_nr_map,
207 struct change_member {
208 struct e820entry *pbios; /* pointer to original bios entry */
209 unsigned long long addr; /* address for this change point */
211 static struct change_member change_point_list[2*E820_X_MAX] __initdata;
212 static struct change_member *change_point[2*E820_X_MAX] __initdata;
213 static struct e820entry *overlap_list[E820_X_MAX] __initdata;
214 static struct e820entry new_bios[E820_X_MAX] __initdata;
215 struct change_member *change_tmp;
216 unsigned long current_type, last_type;
217 unsigned long long last_addr;
218 int chgidx, still_changing;
221 int old_nr, new_nr, chg_nr;
224 /* if there's only one memory region, don't bother */
229 BUG_ON(old_nr > max_nr_map);
231 /* bail out if we find any unreasonable addresses in bios map */
232 for (i = 0; i < old_nr; i++)
233 if (biosmap[i].addr + biosmap[i].size < biosmap[i].addr)
236 /* create pointers for initial change-point information (for sorting) */
237 for (i = 0; i < 2 * old_nr; i++)
238 change_point[i] = &change_point_list[i];
240 /* record all known change-points (starting and ending addresses),
241 omitting those that are for empty memory regions */
243 for (i = 0; i < old_nr; i++) {
244 if (biosmap[i].size != 0) {
245 change_point[chgidx]->addr = biosmap[i].addr;
246 change_point[chgidx++]->pbios = &biosmap[i];
247 change_point[chgidx]->addr = biosmap[i].addr +
249 change_point[chgidx++]->pbios = &biosmap[i];
254 /* sort change-point list by memory addresses (low -> high) */
256 while (still_changing) {
258 for (i = 1; i < chg_nr; i++) {
259 unsigned long long curaddr, lastaddr;
260 unsigned long long curpbaddr, lastpbaddr;
262 curaddr = change_point[i]->addr;
263 lastaddr = change_point[i - 1]->addr;
264 curpbaddr = change_point[i]->pbios->addr;
265 lastpbaddr = change_point[i - 1]->pbios->addr;
268 * swap entries, when:
270 * curaddr > lastaddr or
271 * curaddr == lastaddr and curaddr == curpbaddr and
272 * lastaddr != lastpbaddr
274 if (curaddr < lastaddr ||
275 (curaddr == lastaddr && curaddr == curpbaddr &&
276 lastaddr != lastpbaddr)) {
277 change_tmp = change_point[i];
278 change_point[i] = change_point[i-1];
279 change_point[i-1] = change_tmp;
285 /* create a new bios memory map, removing overlaps */
286 overlap_entries = 0; /* number of entries in the overlap table */
287 new_bios_entry = 0; /* index for creating new bios map entries */
288 last_type = 0; /* start with undefined memory type */
289 last_addr = 0; /* start with 0 as last starting address */
291 /* loop through change-points, determining affect on the new bios map */
292 for (chgidx = 0; chgidx < chg_nr; chgidx++) {
293 /* keep track of all overlapping bios entries */
294 if (change_point[chgidx]->addr ==
295 change_point[chgidx]->pbios->addr) {
297 * add map entry to overlap list (> 1 entry
298 * implies an overlap)
300 overlap_list[overlap_entries++] =
301 change_point[chgidx]->pbios;
304 * remove entry from list (order independent,
307 for (i = 0; i < overlap_entries; i++) {
308 if (overlap_list[i] ==
309 change_point[chgidx]->pbios)
311 overlap_list[overlap_entries-1];
316 * if there are overlapping entries, decide which
317 * "type" to use (larger value takes precedence --
318 * 1=usable, 2,3,4,4+=unusable)
321 for (i = 0; i < overlap_entries; i++)
322 if (overlap_list[i]->type > current_type)
323 current_type = overlap_list[i]->type;
325 * continue building up new bios map based on this
328 if (current_type != last_type) {
329 if (last_type != 0) {
330 new_bios[new_bios_entry].size =
331 change_point[chgidx]->addr - last_addr;
333 * move forward only if the new size
336 if (new_bios[new_bios_entry].size != 0)
338 * no more space left for new
341 if (++new_bios_entry >= max_nr_map)
344 if (current_type != 0) {
345 new_bios[new_bios_entry].addr =
346 change_point[chgidx]->addr;
347 new_bios[new_bios_entry].type = current_type;
348 last_addr = change_point[chgidx]->addr;
350 last_type = current_type;
353 /* retain count for new bios entries */
354 new_nr = new_bios_entry;
356 /* copy new bios mapping into original location */
357 memcpy(biosmap, new_bios, new_nr * sizeof(struct e820entry));
363 static int __init __copy_e820_map(struct e820entry *biosmap, int nr_map)
366 u64 start = biosmap->addr;
367 u64 size = biosmap->size;
368 u64 end = start + size;
369 u32 type = biosmap->type;
371 /* Overflow in 64 bits? Ignore the memory map. */
375 e820_add_region(start, size, type);
384 * Copy the BIOS e820 map into a safe place.
386 * Sanity-check it while we're at it..
388 * If we're lucky and live on a modern system, the setup code
389 * will have given us a memory map that we can use to properly
390 * set up memory. If we aren't, we'll fake a memory map.
392 int __init copy_e820_map(struct e820entry *biosmap, int nr_map)
394 /* Only one memory region (or negative)? Ignore it */
398 return __copy_e820_map(biosmap, nr_map);
401 u64 __init e820_update_range(u64 start, u64 size, unsigned old_type,
405 u64 real_updated_size = 0;
407 BUG_ON(old_type == new_type);
409 if (size > (ULLONG_MAX - start))
410 size = ULLONG_MAX - start;
412 for (i = 0; i < e820.nr_map; i++) {
413 struct e820entry *ei = &e820.map[i];
414 u64 final_start, final_end;
415 if (ei->type != old_type)
417 /* totally covered? */
418 if (ei->addr >= start &&
419 (ei->addr + ei->size) <= (start + size)) {
421 real_updated_size += ei->size;
424 /* partially covered */
425 final_start = max(start, ei->addr);
426 final_end = min(start + size, ei->addr + ei->size);
427 if (final_start >= final_end)
429 e820_add_region(final_start, final_end - final_start,
431 real_updated_size += final_end - final_start;
433 ei->size -= final_end - final_start;
434 if (ei->addr < final_start)
436 ei->addr = final_end;
438 return real_updated_size;
441 /* make e820 not cover the range */
442 u64 __init e820_remove_range(u64 start, u64 size, unsigned old_type,
446 u64 real_removed_size = 0;
448 if (size > (ULLONG_MAX - start))
449 size = ULLONG_MAX - start;
451 for (i = 0; i < e820.nr_map; i++) {
452 struct e820entry *ei = &e820.map[i];
453 u64 final_start, final_end;
455 if (checktype && ei->type != old_type)
457 /* totally covered? */
458 if (ei->addr >= start &&
459 (ei->addr + ei->size) <= (start + size)) {
460 real_removed_size += ei->size;
461 memset(ei, 0, sizeof(struct e820entry));
464 /* partially covered */
465 final_start = max(start, ei->addr);
466 final_end = min(start + size, ei->addr + ei->size);
467 if (final_start >= final_end)
469 real_removed_size += final_end - final_start;
471 ei->size -= final_end - final_start;
472 if (ei->addr < final_start)
474 ei->addr = final_end;
476 return real_removed_size;
479 void __init update_e820(void)
483 nr_map = e820.nr_map;
484 if (sanitize_e820_map(e820.map, ARRAY_SIZE(e820.map), &nr_map))
486 e820.nr_map = nr_map;
487 printk(KERN_INFO "modified physical RAM map:\n");
488 e820_print_map("modified");
490 #define MAX_GAP_END 0x100000000ull
492 * Search for a gap in the e820 memory space from start_addr to end_addr.
494 __init int e820_search_gap(unsigned long *gapstart, unsigned long *gapsize,
495 unsigned long start_addr, unsigned long long end_addr)
497 unsigned long long last;
501 last = (end_addr && end_addr < MAX_GAP_END) ? end_addr : MAX_GAP_END;
504 unsigned long long start = e820.map[i].addr;
505 unsigned long long end = start + e820.map[i].size;
507 if (end < start_addr)
511 * Since "last" is at most 4GB, we know we'll
512 * fit in 32 bits if this condition is true
515 unsigned long gap = last - end;
517 if (gap >= *gapsize) {
530 * Search for the biggest gap in the low 32 bits of the e820
531 * memory space. We pass this space to PCI to assign MMIO resources
532 * for hotplug or unconfigured devices in.
533 * Hopefully the BIOS let enough space left.
535 __init void e820_setup_gap(void)
537 unsigned long gapstart, gapsize, round;
540 gapstart = 0x10000000;
542 found = e820_search_gap(&gapstart, &gapsize, 0, MAX_GAP_END);
546 gapstart = (max_pfn << PAGE_SHIFT) + 1024*1024;
547 printk(KERN_ERR "PCI: Warning: Cannot find a gap in the 32bit "
549 KERN_ERR "PCI: Unassigned devices with 32bit resource "
550 "registers may break!\n");
555 * See how much we want to round up: start off with
556 * rounding to the next 1MB area.
559 while ((gapsize >> 4) > round)
561 /* Fun with two's complement */
562 pci_mem_start = (gapstart + round) & -round;
565 "Allocating PCI resources starting at %lx (gap: %lx:%lx)\n",
566 pci_mem_start, gapstart, gapsize);
570 * Because of the size limitation of struct boot_params, only first
571 * 128 E820 memory entries are passed to kernel via
572 * boot_params.e820_map, others are passed via SETUP_E820_EXT node of
573 * linked list of struct setup_data, which is parsed here.
575 void __init parse_e820_ext(struct setup_data *sdata, unsigned long pa_data)
579 struct e820entry *extmap;
581 entries = sdata->len / sizeof(struct e820entry);
582 map_len = sdata->len + sizeof(struct setup_data);
583 if (map_len > PAGE_SIZE)
584 sdata = early_ioremap(pa_data, map_len);
585 extmap = (struct e820entry *)(sdata->data);
586 __copy_e820_map(extmap, entries);
587 sanitize_e820_map(e820.map, ARRAY_SIZE(e820.map), &e820.nr_map);
588 if (map_len > PAGE_SIZE)
589 early_iounmap(sdata, map_len);
590 printk(KERN_INFO "extended physical RAM map:\n");
591 e820_print_map("extended");
594 #if defined(CONFIG_X86_64) || \
595 (defined(CONFIG_X86_32) && defined(CONFIG_HIBERNATION))
597 * Find the ranges of physical addresses that do not correspond to
598 * e820 RAM areas and mark the corresponding pages as nosave for
599 * hibernation (32 bit) or software suspend and suspend to RAM (64 bit).
601 * This function requires the e820 map to be sorted and without any
602 * overlapping entries and assumes the first e820 area to be RAM.
604 void __init e820_mark_nosave_regions(unsigned long limit_pfn)
609 pfn = PFN_DOWN(e820.map[0].addr + e820.map[0].size);
610 for (i = 1; i < e820.nr_map; i++) {
611 struct e820entry *ei = &e820.map[i];
613 if (pfn < PFN_UP(ei->addr))
614 register_nosave_region(pfn, PFN_UP(ei->addr));
616 pfn = PFN_DOWN(ei->addr + ei->size);
617 if (ei->type != E820_RAM && ei->type != E820_RESERVED_KERN)
618 register_nosave_region(PFN_UP(ei->addr), pfn);
620 if (pfn >= limit_pfn)
627 * Early reserved memory areas.
629 #define MAX_EARLY_RES 20
636 static struct early_res early_res[MAX_EARLY_RES] __initdata = {
637 { 0, PAGE_SIZE, "BIOS data page" }, /* BIOS data page */
638 #if defined(CONFIG_X86_64) && defined(CONFIG_X86_TRAMPOLINE)
639 { TRAMPOLINE_BASE, TRAMPOLINE_BASE + 2 * PAGE_SIZE, "TRAMPOLINE" },
641 #if defined(CONFIG_X86_32) && defined(CONFIG_SMP)
643 * But first pinch a few for the stack/trampoline stuff
644 * FIXME: Don't need the extra page at 4K, but need to fix
645 * trampoline before removing it. (see the GDT stuff)
647 { PAGE_SIZE, PAGE_SIZE + PAGE_SIZE, "EX TRAMPOLINE" },
649 * Has to be in very low memory so we can execute
652 { TRAMPOLINE_BASE, TRAMPOLINE_BASE + PAGE_SIZE, "TRAMPOLINE" },
657 static int __init find_overlapped_early(u64 start, u64 end)
662 for (i = 0; i < MAX_EARLY_RES && early_res[i].end; i++) {
664 if (end > r->start && start < r->end)
672 * Drop the i-th range from the early reservation map,
673 * by copying any higher ranges down one over it, and
674 * clearing what had been the last slot.
676 static void __init drop_range(int i)
680 for (j = i + 1; j < MAX_EARLY_RES && early_res[j].end; j++)
683 memmove(&early_res[i], &early_res[i + 1],
684 (j - 1 - i) * sizeof(struct early_res));
686 early_res[j - 1].end = 0;
690 * Split any existing ranges that:
691 * 1) are marked 'overlap_ok', and
692 * 2) overlap with the stated range [start, end)
693 * into whatever portion (if any) of the existing range is entirely
694 * below or entirely above the stated range. Drop the portion
695 * of the existing range that overlaps with the stated range,
696 * which will allow the caller of this routine to then add that
697 * stated range without conflicting with any existing range.
699 static void __init drop_overlaps_that_are_ok(u64 start, u64 end)
703 u64 lower_start, lower_end;
704 u64 upper_start, upper_end;
707 for (i = 0; i < MAX_EARLY_RES && early_res[i].end; i++) {
710 /* Continue past non-overlapping ranges */
711 if (end <= r->start || start >= r->end)
715 * Leave non-ok overlaps as is; let caller
716 * panic "Overlapping early reservations"
717 * when it hits this overlap.
723 * We have an ok overlap. We will drop it from the early
724 * reservation map, and add back in any non-overlapping
725 * portions (lower or upper) as separate, overlap_ok,
726 * non-overlapping ranges.
729 /* 1. Note any non-overlapping (lower or upper) ranges. */
730 strncpy(name, r->name, sizeof(name) - 1);
732 lower_start = lower_end = 0;
733 upper_start = upper_end = 0;
734 if (r->start < start) {
735 lower_start = r->start;
743 /* 2. Drop the original ok overlapping range */
746 i--; /* resume for-loop on copied down entry */
748 /* 3. Add back in any non-overlapping ranges. */
750 reserve_early_overlap_ok(lower_start, lower_end, name);
752 reserve_early_overlap_ok(upper_start, upper_end, name);
756 static void __init __reserve_early(u64 start, u64 end, char *name,
762 i = find_overlapped_early(start, end);
763 if (i >= MAX_EARLY_RES)
764 panic("Too many early reservations");
767 panic("Overlapping early reservations "
768 "%llx-%llx %s to %llx-%llx %s\n",
769 start, end - 1, name?name:"", r->start,
770 r->end - 1, r->name);
773 r->overlap_ok = overlap_ok;
775 strncpy(r->name, name, sizeof(r->name) - 1);
779 * A few early reservtations come here.
781 * The 'overlap_ok' in the name of this routine does -not- mean it
782 * is ok for these reservations to overlap an earlier reservation.
783 * Rather it means that it is ok for subsequent reservations to
786 * Use this entry point to reserve early ranges when you are doing
787 * so out of "Paranoia", reserving perhaps more memory than you need,
788 * just in case, and don't mind a subsequent overlapping reservation
789 * that is known to be needed.
791 * The drop_overlaps_that_are_ok() call here isn't really needed.
792 * It would be needed if we had two colliding 'overlap_ok'
793 * reservations, so that the second such would not panic on the
794 * overlap with the first. We don't have any such as of this
795 * writing, but might as well tolerate such if it happens in
798 void __init reserve_early_overlap_ok(u64 start, u64 end, char *name)
800 drop_overlaps_that_are_ok(start, end);
801 __reserve_early(start, end, name, 1);
805 * Most early reservations come here.
807 * We first have drop_overlaps_that_are_ok() drop any pre-existing
808 * 'overlap_ok' ranges, so that we can then reserve this memory
809 * range without risk of panic'ing on an overlapping overlap_ok
812 void __init reserve_early(u64 start, u64 end, char *name)
814 drop_overlaps_that_are_ok(start, end);
815 __reserve_early(start, end, name, 0);
818 void __init free_early(u64 start, u64 end)
823 i = find_overlapped_early(start, end);
825 if (i >= MAX_EARLY_RES || r->end != end || r->start != start)
826 panic("free_early on not reserved area: %llx-%llx!",
832 void __init early_res_to_bootmem(u64 start, u64 end)
835 u64 final_start, final_end;
838 for (i = 0; i < MAX_EARLY_RES && early_res[i].end; i++)
841 printk(KERN_INFO "(%d early reservations) ==> bootmem\n", count);
842 for (i = 0; i < count; i++) {
843 struct early_res *r = &early_res[i];
844 printk(KERN_INFO " #%d [ %010llx - %010llx ] %16s", i,
845 r->start, r->end, r->name);
846 final_start = max(start, r->start);
847 final_end = min(end, r->end);
848 if (final_start >= final_end) {
849 printk(KERN_CONT "\n");
852 printk(KERN_CONT " ===> [ %010llx - %010llx ]\n",
853 final_start, final_end);
854 reserve_bootmem_generic(final_start, final_end - final_start,
859 /* Check for already reserved areas */
860 static inline int __init bad_addr(u64 *addrp, u64 size, u64 align)
867 i = find_overlapped_early(addr, addr + size);
869 if (i < MAX_EARLY_RES && r->end) {
870 *addrp = addr = round_up(r->end, align);
877 /* Check for already reserved areas */
878 static inline int __init bad_addr_size(u64 *addrp, u64 *sizep, u64 align)
881 u64 addr = *addrp, last;
886 for (i = 0; i < MAX_EARLY_RES && early_res[i].end; i++) {
887 struct early_res *r = &early_res[i];
888 if (last > r->start && addr < r->start) {
889 size = r->start - addr;
893 if (last > r->end && addr < r->end) {
894 addr = round_up(r->end, align);
899 if (last <= r->end && addr >= r->start) {
912 * Find a free area with specified alignment in a specific range.
914 u64 __init find_e820_area(u64 start, u64 end, u64 size, u64 align)
918 for (i = 0; i < e820.nr_map; i++) {
919 struct e820entry *ei = &e820.map[i];
923 if (ei->type != E820_RAM)
925 addr = round_up(ei->addr, align);
926 ei_last = ei->addr + ei->size;
928 addr = round_up(start, align);
931 while (bad_addr(&addr, size, align) && addr+size <= ei_last)
944 * Find next free range after *start
946 u64 __init find_e820_area_size(u64 start, u64 *sizep, u64 align)
950 for (i = 0; i < e820.nr_map; i++) {
951 struct e820entry *ei = &e820.map[i];
955 if (ei->type != E820_RAM)
957 addr = round_up(ei->addr, align);
958 ei_last = ei->addr + ei->size;
960 addr = round_up(start, align);
963 *sizep = ei_last - addr;
964 while (bad_addr_size(&addr, sizep, align) &&
965 addr + *sizep <= ei_last)
967 last = addr + *sizep;
977 * pre allocated 4k and reserved it in e820
979 u64 __init early_reserve_e820(u64 startt, u64 sizet, u64 align)
987 start = find_e820_area_size(start, &size, align);
992 addr = round_down(start + size - sizet, align);
993 e820_update_range(addr, sizet, E820_RAM, E820_RESERVED);
994 printk(KERN_INFO "update e820 for early_reserve_e820\n");
1000 #ifdef CONFIG_X86_32
1001 # ifdef CONFIG_X86_PAE
1002 # define MAX_ARCH_PFN (1ULL<<(36-PAGE_SHIFT))
1004 # define MAX_ARCH_PFN (1ULL<<(32-PAGE_SHIFT))
1006 #else /* CONFIG_X86_32 */
1007 # define MAX_ARCH_PFN MAXMEM>>PAGE_SHIFT
1011 * Last pfn which the user wants to use.
1013 unsigned long __initdata end_user_pfn = MAX_ARCH_PFN;
1016 * Find the highest page frame number we have available
1018 unsigned long __init e820_end_of_ram(void)
1020 unsigned long last_pfn;
1021 unsigned long max_arch_pfn = MAX_ARCH_PFN;
1023 last_pfn = find_max_pfn_with_active_regions();
1025 if (last_pfn > max_arch_pfn)
1026 last_pfn = max_arch_pfn;
1027 if (last_pfn > end_user_pfn)
1028 last_pfn = end_user_pfn;
1030 printk(KERN_INFO "last_pfn = %#lx max_arch_pfn = %#lx\n",
1031 last_pfn, max_arch_pfn);
1036 * Finds an active region in the address range from start_pfn to last_pfn and
1037 * returns its range in ei_startpfn and ei_endpfn for the e820 entry.
1039 int __init e820_find_active_region(const struct e820entry *ei,
1040 unsigned long start_pfn,
1041 unsigned long last_pfn,
1042 unsigned long *ei_startpfn,
1043 unsigned long *ei_endpfn)
1045 u64 align = PAGE_SIZE;
1047 *ei_startpfn = round_up(ei->addr, align) >> PAGE_SHIFT;
1048 *ei_endpfn = round_down(ei->addr + ei->size, align) >> PAGE_SHIFT;
1050 /* Skip map entries smaller than a page */
1051 if (*ei_startpfn >= *ei_endpfn)
1054 /* Skip if map is outside the node */
1055 if (ei->type != E820_RAM || *ei_endpfn <= start_pfn ||
1056 *ei_startpfn >= last_pfn)
1059 /* Check for overlaps */
1060 if (*ei_startpfn < start_pfn)
1061 *ei_startpfn = start_pfn;
1062 if (*ei_endpfn > last_pfn)
1063 *ei_endpfn = last_pfn;
1065 /* Obey end_user_pfn to save on memmap */
1066 if (*ei_startpfn >= end_user_pfn)
1068 if (*ei_endpfn > end_user_pfn)
1069 *ei_endpfn = end_user_pfn;
1074 /* Walk the e820 map and register active regions within a node */
1075 void __init e820_register_active_regions(int nid, unsigned long start_pfn,
1076 unsigned long last_pfn)
1078 unsigned long ei_startpfn;
1079 unsigned long ei_endpfn;
1082 for (i = 0; i < e820.nr_map; i++)
1083 if (e820_find_active_region(&e820.map[i],
1084 start_pfn, last_pfn,
1085 &ei_startpfn, &ei_endpfn))
1086 add_active_range(nid, ei_startpfn, ei_endpfn);
1090 * Find the hole size (in bytes) in the memory range.
1091 * @start: starting address of the memory range to scan
1092 * @end: ending address of the memory range to scan
1094 u64 __init e820_hole_size(u64 start, u64 end)
1096 unsigned long start_pfn = start >> PAGE_SHIFT;
1097 unsigned long last_pfn = end >> PAGE_SHIFT;
1098 unsigned long ei_startpfn, ei_endpfn, ram = 0;
1101 for (i = 0; i < e820.nr_map; i++) {
1102 if (e820_find_active_region(&e820.map[i],
1103 start_pfn, last_pfn,
1104 &ei_startpfn, &ei_endpfn))
1105 ram += ei_endpfn - ei_startpfn;
1107 return end - start - ((u64)ram << PAGE_SHIFT);
1110 static void early_panic(char *msg)
1116 /* "mem=nopentium" disables the 4MB page tables. */
1117 static int __init parse_memopt(char *p)
1124 #ifdef CONFIG_X86_32
1125 if (!strcmp(p, "nopentium")) {
1126 setup_clear_cpu_cap(X86_FEATURE_PSE);
1131 mem_size = memparse(p, &p);
1132 end_user_pfn = mem_size>>PAGE_SHIFT;
1133 e820_update_range(mem_size, ULLONG_MAX - mem_size,
1134 E820_RAM, E820_RESERVED);
1138 early_param("mem", parse_memopt);
1140 static int userdef __initdata;
1142 static int __init parse_memmap_opt(char *p)
1145 u64 start_at, mem_size;
1147 if (!strcmp(p, "exactmap")) {
1148 #ifdef CONFIG_CRASH_DUMP
1150 * If we are doing a crash dump, we still need to know
1151 * the real mem size before original memory map is
1154 e820_register_active_regions(0, 0, -1UL);
1155 saved_max_pfn = e820_end_of_ram();
1156 remove_all_active_ranges();
1164 mem_size = memparse(p, &p);
1170 start_at = memparse(p+1, &p);
1171 e820_add_region(start_at, mem_size, E820_RAM);
1172 } else if (*p == '#') {
1173 start_at = memparse(p+1, &p);
1174 e820_add_region(start_at, mem_size, E820_ACPI);
1175 } else if (*p == '$') {
1176 start_at = memparse(p+1, &p);
1177 e820_add_region(start_at, mem_size, E820_RESERVED);
1179 end_user_pfn = (mem_size >> PAGE_SHIFT);
1180 e820_update_range(mem_size, ULLONG_MAX - mem_size,
1181 E820_RAM, E820_RESERVED);
1183 return *p == '\0' ? 0 : -EINVAL;
1185 early_param("memmap", parse_memmap_opt);
1187 void __init finish_e820_parsing(void)
1190 int nr = e820.nr_map;
1192 if (sanitize_e820_map(e820.map, ARRAY_SIZE(e820.map), &nr) < 0)
1193 early_panic("Invalid user supplied memory map");
1196 printk(KERN_INFO "user-defined physical RAM map:\n");
1197 e820_print_map("user");
1202 * Mark e820 reserved areas as busy for the resource manager.
1204 void __init e820_reserve_resources(void)
1207 struct resource *res;
1210 res = alloc_bootmem_low(sizeof(struct resource) * e820.nr_map);
1211 for (i = 0; i < e820.nr_map; i++) {
1212 switch (e820.map[i].type) {
1213 case E820_RESERVED_KERN:
1214 case E820_RAM: res->name = "System RAM"; break;
1215 case E820_ACPI: res->name = "ACPI Tables"; break;
1216 case E820_NVS: res->name = "ACPI Non-volatile Storage"; break;
1217 default: res->name = "reserved";
1219 end = e820.map[i].addr + e820.map[i].size - 1;
1220 #ifndef CONFIG_RESOURCES_64BIT
1221 if (end > 0x100000000ULL) {
1226 res->start = e820.map[i].addr;
1229 res->flags = IORESOURCE_MEM | IORESOURCE_BUSY;
1230 insert_resource(&iomem_resource, res);
1235 char *__init default_machine_specific_memory_setup(void)
1237 char *who = "BIOS-e820";
1240 * Try to copy the BIOS-supplied E820-map.
1242 * Otherwise fake a memory map; one section from 0k->640k,
1243 * the next section from 1mb->appropriate_mem_k
1245 new_nr = boot_params.e820_entries;
1246 sanitize_e820_map(boot_params.e820_map,
1247 ARRAY_SIZE(boot_params.e820_map),
1249 boot_params.e820_entries = new_nr;
1250 if (copy_e820_map(boot_params.e820_map, boot_params.e820_entries) < 0) {
1253 /* compare results from other methods and take the greater */
1254 if (boot_params.alt_mem_k
1255 < boot_params.screen_info.ext_mem_k) {
1256 mem_size = boot_params.screen_info.ext_mem_k;
1259 mem_size = boot_params.alt_mem_k;
1264 e820_add_region(0, LOWMEMSIZE(), E820_RAM);
1265 e820_add_region(HIGH_MEMORY, mem_size << 10, E820_RAM);
1268 /* In case someone cares... */
1272 char *__init __attribute__((weak)) machine_specific_memory_setup(void)
1274 return default_machine_specific_memory_setup();
1277 /* Overridden in paravirt.c if CONFIG_PARAVIRT */
1278 char * __init __attribute__((weak)) memory_setup(void)
1280 return machine_specific_memory_setup();
1283 void __init setup_memory_map(void)
1285 printk(KERN_INFO "BIOS-provided physical RAM map:\n");
1286 e820_print_map(memory_setup());
1289 #ifdef CONFIG_X86_64
1290 int __init arch_get_ram_range(int slot, u64 *addr, u64 *size)
1294 if (slot < 0 || slot >= e820.nr_map)
1296 for (i = slot; i < e820.nr_map; i++) {
1297 if (e820.map[i].type != E820_RAM)
1301 if (i == e820.nr_map || e820.map[i].addr > (max_pfn << PAGE_SHIFT))
1303 *addr = e820.map[i].addr;
1304 *size = min_t(u64, e820.map[i].size + e820.map[i].addr,
1305 max_pfn << PAGE_SHIFT) - *addr;