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 printk(KERN_CONT "(usable)\n");
126 printk(KERN_CONT "(reserved)\n");
129 printk(KERN_CONT "(ACPI data)\n");
132 printk(KERN_CONT "(ACPI NVS)\n");
135 printk(KERN_CONT "type %u\n", e820.map[i].type);
142 * Sanitize the BIOS e820 map.
144 * Some e820 responses include overlapping entries. The following
145 * replaces the original e820 map with a new one, removing overlaps,
146 * and resolving conflicting memory types in favor of highest
149 * The input parameter biosmap points to an array of 'struct
150 * e820entry' which on entry has elements in the range [0, *pnr_map)
151 * valid, and which has space for up to max_nr_map entries.
152 * On return, the resulting sanitized e820 map entries will be in
153 * overwritten in the same location, starting at biosmap.
155 * The integer pointed to by pnr_map must be valid on entry (the
156 * current number of valid entries located at biosmap) and will
157 * be updated on return, with the new number of valid entries
158 * (something no more than max_nr_map.)
160 * The return value from sanitize_e820_map() is zero if it
161 * successfully 'sanitized' the map entries passed in, and is -1
162 * if it did nothing, which can happen if either of (1) it was
163 * only passed one map entry, or (2) any of the input map entries
164 * were invalid (start + size < start, meaning that the size was
165 * so big the described memory range wrapped around through zero.)
167 * Visually we're performing the following
168 * (1,2,3,4 = memory types)...
170 * Sample memory map (w/overlaps):
171 * ____22__________________
172 * ______________________4_
173 * ____1111________________
174 * _44_____________________
175 * 11111111________________
176 * ____________________33__
177 * ___________44___________
178 * __________33333_________
179 * ______________22________
180 * ___________________2222_
181 * _________111111111______
182 * _____________________11_
183 * _________________4______
185 * Sanitized equivalent (no overlap):
186 * 1_______________________
187 * _44_____________________
188 * ___1____________________
189 * ____22__________________
190 * ______11________________
191 * _________1______________
192 * __________3_____________
193 * ___________44___________
194 * _____________33_________
195 * _______________2________
196 * ________________1_______
197 * _________________4______
198 * ___________________2____
199 * ____________________33__
200 * ______________________4_
203 int __init sanitize_e820_map(struct e820entry *biosmap, int max_nr_map,
206 struct change_member {
207 struct e820entry *pbios; /* pointer to original bios entry */
208 unsigned long long addr; /* address for this change point */
210 static struct change_member change_point_list[2*E820_X_MAX] __initdata;
211 static struct change_member *change_point[2*E820_X_MAX] __initdata;
212 static struct e820entry *overlap_list[E820_X_MAX] __initdata;
213 static struct e820entry new_bios[E820_X_MAX] __initdata;
214 struct change_member *change_tmp;
215 unsigned long current_type, last_type;
216 unsigned long long last_addr;
217 int chgidx, still_changing;
220 int old_nr, new_nr, chg_nr;
223 /* if there's only one memory region, don't bother */
228 BUG_ON(old_nr > max_nr_map);
230 /* bail out if we find any unreasonable addresses in bios map */
231 for (i = 0; i < old_nr; i++)
232 if (biosmap[i].addr + biosmap[i].size < biosmap[i].addr)
235 /* create pointers for initial change-point information (for sorting) */
236 for (i = 0; i < 2 * old_nr; i++)
237 change_point[i] = &change_point_list[i];
239 /* record all known change-points (starting and ending addresses),
240 omitting those that are for empty memory regions */
242 for (i = 0; i < old_nr; i++) {
243 if (biosmap[i].size != 0) {
244 change_point[chgidx]->addr = biosmap[i].addr;
245 change_point[chgidx++]->pbios = &biosmap[i];
246 change_point[chgidx]->addr = biosmap[i].addr +
248 change_point[chgidx++]->pbios = &biosmap[i];
253 /* sort change-point list by memory addresses (low -> high) */
255 while (still_changing) {
257 for (i = 1; i < chg_nr; i++) {
258 unsigned long long curaddr, lastaddr;
259 unsigned long long curpbaddr, lastpbaddr;
261 curaddr = change_point[i]->addr;
262 lastaddr = change_point[i - 1]->addr;
263 curpbaddr = change_point[i]->pbios->addr;
264 lastpbaddr = change_point[i - 1]->pbios->addr;
267 * swap entries, when:
269 * curaddr > lastaddr or
270 * curaddr == lastaddr and curaddr == curpbaddr and
271 * lastaddr != lastpbaddr
273 if (curaddr < lastaddr ||
274 (curaddr == lastaddr && curaddr == curpbaddr &&
275 lastaddr != lastpbaddr)) {
276 change_tmp = change_point[i];
277 change_point[i] = change_point[i-1];
278 change_point[i-1] = change_tmp;
284 /* create a new bios memory map, removing overlaps */
285 overlap_entries = 0; /* number of entries in the overlap table */
286 new_bios_entry = 0; /* index for creating new bios map entries */
287 last_type = 0; /* start with undefined memory type */
288 last_addr = 0; /* start with 0 as last starting address */
290 /* loop through change-points, determining affect on the new bios map */
291 for (chgidx = 0; chgidx < chg_nr; chgidx++) {
292 /* keep track of all overlapping bios entries */
293 if (change_point[chgidx]->addr ==
294 change_point[chgidx]->pbios->addr) {
296 * add map entry to overlap list (> 1 entry
297 * implies an overlap)
299 overlap_list[overlap_entries++] =
300 change_point[chgidx]->pbios;
303 * remove entry from list (order independent,
306 for (i = 0; i < overlap_entries; i++) {
307 if (overlap_list[i] ==
308 change_point[chgidx]->pbios)
310 overlap_list[overlap_entries-1];
315 * if there are overlapping entries, decide which
316 * "type" to use (larger value takes precedence --
317 * 1=usable, 2,3,4,4+=unusable)
320 for (i = 0; i < overlap_entries; i++)
321 if (overlap_list[i]->type > current_type)
322 current_type = overlap_list[i]->type;
324 * continue building up new bios map based on this
327 if (current_type != last_type) {
328 if (last_type != 0) {
329 new_bios[new_bios_entry].size =
330 change_point[chgidx]->addr - last_addr;
332 * move forward only if the new size
335 if (new_bios[new_bios_entry].size != 0)
337 * no more space left for new
340 if (++new_bios_entry >= max_nr_map)
343 if (current_type != 0) {
344 new_bios[new_bios_entry].addr =
345 change_point[chgidx]->addr;
346 new_bios[new_bios_entry].type = current_type;
347 last_addr = change_point[chgidx]->addr;
349 last_type = current_type;
352 /* retain count for new bios entries */
353 new_nr = new_bios_entry;
355 /* copy new bios mapping into original location */
356 memcpy(biosmap, new_bios, new_nr * sizeof(struct e820entry));
362 static int __init __copy_e820_map(struct e820entry *biosmap, int nr_map)
365 u64 start = biosmap->addr;
366 u64 size = biosmap->size;
367 u64 end = start + size;
368 u32 type = biosmap->type;
370 /* Overflow in 64 bits? Ignore the memory map. */
374 e820_add_region(start, size, type);
383 * Copy the BIOS e820 map into a safe place.
385 * Sanity-check it while we're at it..
387 * If we're lucky and live on a modern system, the setup code
388 * will have given us a memory map that we can use to properly
389 * set up memory. If we aren't, we'll fake a memory map.
391 int __init copy_e820_map(struct e820entry *biosmap, int nr_map)
393 /* Only one memory region (or negative)? Ignore it */
397 return __copy_e820_map(biosmap, nr_map);
400 u64 __init e820_update_range(u64 start, u64 size, unsigned old_type,
404 u64 real_updated_size = 0;
406 BUG_ON(old_type == new_type);
408 if (size > (ULLONG_MAX - start))
409 size = ULLONG_MAX - start;
411 for (i = 0; i < e820.nr_map; i++) {
412 struct e820entry *ei = &e820.map[i];
413 u64 final_start, final_end;
414 if (ei->type != old_type)
416 /* totally covered? */
417 if (ei->addr >= start &&
418 (ei->addr + ei->size) <= (start + size)) {
420 real_updated_size += ei->size;
423 /* partially covered */
424 final_start = max(start, ei->addr);
425 final_end = min(start + size, ei->addr + ei->size);
426 if (final_start >= final_end)
428 e820_add_region(final_start, final_end - final_start,
430 real_updated_size += final_end - final_start;
432 ei->size -= final_end - final_start;
433 if (ei->addr < final_start)
435 ei->addr = final_end;
437 return real_updated_size;
440 /* make e820 not cover the range */
441 u64 __init e820_remove_range(u64 start, u64 size, unsigned old_type,
445 u64 real_removed_size = 0;
447 if (size > (ULLONG_MAX - start))
448 size = ULLONG_MAX - start;
450 for (i = 0; i < e820.nr_map; i++) {
451 struct e820entry *ei = &e820.map[i];
452 u64 final_start, final_end;
454 if (checktype && ei->type != old_type)
456 /* totally covered? */
457 if (ei->addr >= start &&
458 (ei->addr + ei->size) <= (start + size)) {
459 real_removed_size += ei->size;
460 memset(ei, 0, sizeof(struct e820entry));
463 /* partially covered */
464 final_start = max(start, ei->addr);
465 final_end = min(start + size, ei->addr + ei->size);
466 if (final_start >= final_end)
468 real_removed_size += final_end - final_start;
470 ei->size -= final_end - final_start;
471 if (ei->addr < final_start)
473 ei->addr = final_end;
475 return real_removed_size;
478 void __init update_e820(void)
482 nr_map = e820.nr_map;
483 if (sanitize_e820_map(e820.map, ARRAY_SIZE(e820.map), &nr_map))
485 e820.nr_map = nr_map;
486 printk(KERN_INFO "modified physical RAM map:\n");
487 e820_print_map("modified");
491 * Search for the biggest gap in the low 32 bits of the e820
492 * memory space. We pass this space to PCI to assign MMIO resources
493 * for hotplug or unconfigured devices in.
494 * Hopefully the BIOS let enough space left.
496 __init void e820_setup_gap(void)
498 unsigned long gapstart, gapsize, round;
499 unsigned long long last;
503 last = 0x100000000ull;
504 gapstart = 0x10000000;
508 unsigned long long start = e820.map[i].addr;
509 unsigned long long end = start + e820.map[i].size;
512 * Since "last" is at most 4GB, we know we'll
513 * fit in 32 bits if this condition is true
516 unsigned long gap = last - end;
530 gapstart = (end_pfn << PAGE_SHIFT) + 1024*1024;
531 printk(KERN_ERR "PCI: Warning: Cannot find a gap in the 32bit "
533 KERN_ERR "PCI: Unassigned devices with 32bit resource "
534 "registers may break!\n");
539 * See how much we want to round up: start off with
540 * rounding to the next 1MB area.
543 while ((gapsize >> 4) > round)
545 /* Fun with two's complement */
546 pci_mem_start = (gapstart + round) & -round;
549 "Allocating PCI resources starting at %lx (gap: %lx:%lx)\n",
550 pci_mem_start, gapstart, gapsize);
554 * Because of the size limitation of struct boot_params, only first
555 * 128 E820 memory entries are passed to kernel via
556 * boot_params.e820_map, others are passed via SETUP_E820_EXT node of
557 * linked list of struct setup_data, which is parsed here.
559 void __init parse_e820_ext(struct setup_data *sdata, unsigned long pa_data)
563 struct e820entry *extmap;
565 entries = sdata->len / sizeof(struct e820entry);
566 map_len = sdata->len + sizeof(struct setup_data);
567 if (map_len > PAGE_SIZE)
568 sdata = early_ioremap(pa_data, map_len);
569 extmap = (struct e820entry *)(sdata->data);
570 __copy_e820_map(extmap, entries);
571 sanitize_e820_map(e820.map, ARRAY_SIZE(e820.map), &e820.nr_map);
572 if (map_len > PAGE_SIZE)
573 early_iounmap(sdata, map_len);
574 printk(KERN_INFO "extended physical RAM map:\n");
575 e820_print_map("extended");
578 #if defined(CONFIG_X86_64) || \
579 (defined(CONFIG_X86_32) && defined(CONFIG_HIBERNATION))
581 * Find the ranges of physical addresses that do not correspond to
582 * e820 RAM areas and mark the corresponding pages as nosave for
583 * hibernation (32 bit) or software suspend and suspend to RAM (64 bit).
585 * This function requires the e820 map to be sorted and without any
586 * overlapping entries and assumes the first e820 area to be RAM.
588 void __init e820_mark_nosave_regions(unsigned long limit_pfn)
593 pfn = PFN_DOWN(e820.map[0].addr + e820.map[0].size);
594 for (i = 1; i < e820.nr_map; i++) {
595 struct e820entry *ei = &e820.map[i];
597 if (pfn < PFN_UP(ei->addr))
598 register_nosave_region(pfn, PFN_UP(ei->addr));
600 pfn = PFN_DOWN(ei->addr + ei->size);
601 if (ei->type != E820_RAM)
602 register_nosave_region(PFN_UP(ei->addr), pfn);
604 if (pfn >= limit_pfn)
611 * Early reserved memory areas.
613 #define MAX_EARLY_RES 20
620 static struct early_res early_res[MAX_EARLY_RES] __initdata = {
621 { 0, PAGE_SIZE, "BIOS data page" }, /* BIOS data page */
622 #if defined(CONFIG_X86_64) && defined(CONFIG_X86_TRAMPOLINE)
623 { TRAMPOLINE_BASE, TRAMPOLINE_BASE + 2 * PAGE_SIZE, "TRAMPOLINE" },
625 #if defined(CONFIG_X86_32) && defined(CONFIG_SMP)
627 * But first pinch a few for the stack/trampoline stuff
628 * FIXME: Don't need the extra page at 4K, but need to fix
629 * trampoline before removing it. (see the GDT stuff)
631 { PAGE_SIZE, PAGE_SIZE + PAGE_SIZE, "EX TRAMPOLINE" },
633 * Has to be in very low memory so we can execute
636 { TRAMPOLINE_BASE, TRAMPOLINE_BASE + PAGE_SIZE, "TRAMPOLINE" },
641 static int __init find_overlapped_early(u64 start, u64 end)
646 for (i = 0; i < MAX_EARLY_RES && early_res[i].end; i++) {
648 if (end > r->start && start < r->end)
656 * Drop the i-th range from the early reservation map,
657 * by copying any higher ranges down one over it, and
658 * clearing what had been the last slot.
660 static void __init drop_range(int i)
664 for (j = i + 1; j < MAX_EARLY_RES && early_res[j].end; j++)
667 memmove(&early_res[i], &early_res[i + 1],
668 (j - 1 - i) * sizeof(struct early_res));
670 early_res[j - 1].end = 0;
674 * Split any existing ranges that:
675 * 1) are marked 'overlap_ok', and
676 * 2) overlap with the stated range [start, end)
677 * into whatever portion (if any) of the existing range is entirely
678 * below or entirely above the stated range. Drop the portion
679 * of the existing range that overlaps with the stated range,
680 * which will allow the caller of this routine to then add that
681 * stated range without conflicting with any existing range.
683 static void __init drop_overlaps_that_are_ok(u64 start, u64 end)
687 u64 lower_start, lower_end;
688 u64 upper_start, upper_end;
691 for (i = 0; i < MAX_EARLY_RES && early_res[i].end; i++) {
694 /* Continue past non-overlapping ranges */
695 if (end <= r->start || start >= r->end)
699 * Leave non-ok overlaps as is; let caller
700 * panic "Overlapping early reservations"
701 * when it hits this overlap.
707 * We have an ok overlap. We will drop it from the early
708 * reservation map, and add back in any non-overlapping
709 * portions (lower or upper) as separate, overlap_ok,
710 * non-overlapping ranges.
713 /* 1. Note any non-overlapping (lower or upper) ranges. */
714 strncpy(name, r->name, sizeof(name) - 1);
716 lower_start = lower_end = 0;
717 upper_start = upper_end = 0;
718 if (r->start < start) {
719 lower_start = r->start;
727 /* 2. Drop the original ok overlapping range */
730 i--; /* resume for-loop on copied down entry */
732 /* 3. Add back in any non-overlapping ranges. */
734 reserve_early_overlap_ok(lower_start, lower_end, name);
736 reserve_early_overlap_ok(upper_start, upper_end, name);
740 static void __init __reserve_early(u64 start, u64 end, char *name,
746 i = find_overlapped_early(start, end);
747 if (i >= MAX_EARLY_RES)
748 panic("Too many early reservations");
751 panic("Overlapping early reservations "
752 "%llx-%llx %s to %llx-%llx %s\n",
753 start, end - 1, name?name:"", r->start,
754 r->end - 1, r->name);
757 r->overlap_ok = overlap_ok;
759 strncpy(r->name, name, sizeof(r->name) - 1);
763 * A few early reservtations come here.
765 * The 'overlap_ok' in the name of this routine does -not- mean it
766 * is ok for these reservations to overlap an earlier reservation.
767 * Rather it means that it is ok for subsequent reservations to
770 * Use this entry point to reserve early ranges when you are doing
771 * so out of "Paranoia", reserving perhaps more memory than you need,
772 * just in case, and don't mind a subsequent overlapping reservation
773 * that is known to be needed.
775 * The drop_overlaps_that_are_ok() call here isn't really needed.
776 * It would be needed if we had two colliding 'overlap_ok'
777 * reservations, so that the second such would not panic on the
778 * overlap with the first. We don't have any such as of this
779 * writing, but might as well tolerate such if it happens in
782 void __init reserve_early_overlap_ok(u64 start, u64 end, char *name)
784 drop_overlaps_that_are_ok(start, end);
785 __reserve_early(start, end, name, 1);
789 * Most early reservations come here.
791 * We first have drop_overlaps_that_are_ok() drop any pre-existing
792 * 'overlap_ok' ranges, so that we can then reserve this memory
793 * range without risk of panic'ing on an overlapping overlap_ok
796 void __init reserve_early(u64 start, u64 end, char *name)
798 drop_overlaps_that_are_ok(start, end);
799 __reserve_early(start, end, name, 0);
802 void __init free_early(u64 start, u64 end)
807 i = find_overlapped_early(start, end);
809 if (i >= MAX_EARLY_RES || r->end != end || r->start != start)
810 panic("free_early on not reserved area: %llx-%llx!",
816 void __init early_res_to_bootmem(u64 start, u64 end)
819 u64 final_start, final_end;
820 for (i = 0; i < MAX_EARLY_RES && early_res[i].end; i++) {
821 struct early_res *r = &early_res[i];
822 final_start = max(start, r->start);
823 final_end = min(end, r->end);
824 if (final_start >= final_end)
826 printk(KERN_INFO " early res: %d [%llx-%llx] %s\n", i,
827 final_start, final_end - 1, r->name);
828 reserve_bootmem_generic(final_start, final_end - final_start,
833 /* Check for already reserved areas */
834 static inline int __init bad_addr(u64 *addrp, u64 size, u64 align)
841 i = find_overlapped_early(addr, addr + size);
843 if (i < MAX_EARLY_RES && r->end) {
844 *addrp = addr = round_up(r->end, align);
851 /* Check for already reserved areas */
852 static inline int __init bad_addr_size(u64 *addrp, u64 *sizep, u64 align)
855 u64 addr = *addrp, last;
860 for (i = 0; i < MAX_EARLY_RES && early_res[i].end; i++) {
861 struct early_res *r = &early_res[i];
862 if (last > r->start && addr < r->start) {
863 size = r->start - addr;
867 if (last > r->end && addr < r->end) {
868 addr = round_up(r->end, align);
873 if (last <= r->end && addr >= r->start) {
886 * Find a free area with specified alignment in a specific range.
888 u64 __init find_e820_area(u64 start, u64 end, u64 size, u64 align)
892 for (i = 0; i < e820.nr_map; i++) {
893 struct e820entry *ei = &e820.map[i];
897 if (ei->type != E820_RAM)
899 addr = round_up(ei->addr, align);
900 ei_last = ei->addr + ei->size;
902 addr = round_up(start, align);
905 while (bad_addr(&addr, size, align) && addr+size <= ei_last)
918 * Find next free range after *start
920 u64 __init find_e820_area_size(u64 start, u64 *sizep, u64 align)
924 for (i = 0; i < e820.nr_map; i++) {
925 struct e820entry *ei = &e820.map[i];
929 if (ei->type != E820_RAM)
931 addr = round_up(ei->addr, align);
932 ei_last = ei->addr + ei->size;
934 addr = round_up(start, align);
937 *sizep = ei_last - addr;
938 while (bad_addr_size(&addr, sizep, align) &&
939 addr + *sizep <= ei_last)
941 last = addr + *sizep;
951 * pre allocated 4k and reserved it in e820
953 u64 __init early_reserve_e820(u64 startt, u64 sizet, u64 align)
961 start = find_e820_area_size(start, &size, align);
966 addr = round_down(start + size - sizet, align);
967 e820_update_range(addr, sizet, E820_RAM, E820_RESERVED);
968 printk(KERN_INFO "update e820 for early_reserve_e820\n");
975 # ifdef CONFIG_X86_PAE
976 # define MAX_ARCH_PFN (1ULL<<(36-PAGE_SHIFT))
978 # define MAX_ARCH_PFN (1ULL<<(32-PAGE_SHIFT))
980 #else /* CONFIG_X86_32 */
981 # define MAX_ARCH_PFN MAXMEM>>PAGE_SHIFT
985 * Last pfn which the user wants to use.
987 unsigned long __initdata end_user_pfn = MAX_ARCH_PFN;
990 * Find the highest page frame number we have available
992 unsigned long __init e820_end_of_ram(void)
994 unsigned long last_pfn;
995 unsigned long max_arch_pfn = MAX_ARCH_PFN;
997 last_pfn = find_max_pfn_with_active_regions();
999 if (last_pfn > max_arch_pfn)
1000 last_pfn = max_arch_pfn;
1001 if (last_pfn > end_user_pfn)
1002 last_pfn = end_user_pfn;
1004 printk(KERN_INFO "last_pfn = 0x%lx max_arch_pfn = 0x%lx\n",
1005 last_pfn, max_arch_pfn);
1010 * Finds an active region in the address range from start_pfn to last_pfn and
1011 * returns its range in ei_startpfn and ei_endpfn for the e820 entry.
1013 int __init e820_find_active_region(const struct e820entry *ei,
1014 unsigned long start_pfn,
1015 unsigned long last_pfn,
1016 unsigned long *ei_startpfn,
1017 unsigned long *ei_endpfn)
1019 u64 align = PAGE_SIZE;
1021 *ei_startpfn = round_up(ei->addr, align) >> PAGE_SHIFT;
1022 *ei_endpfn = round_down(ei->addr + ei->size, align) >> PAGE_SHIFT;
1024 /* Skip map entries smaller than a page */
1025 if (*ei_startpfn >= *ei_endpfn)
1028 /* Skip if map is outside the node */
1029 if (ei->type != E820_RAM || *ei_endpfn <= start_pfn ||
1030 *ei_startpfn >= last_pfn)
1033 /* Check for overlaps */
1034 if (*ei_startpfn < start_pfn)
1035 *ei_startpfn = start_pfn;
1036 if (*ei_endpfn > last_pfn)
1037 *ei_endpfn = last_pfn;
1039 /* Obey end_user_pfn to save on memmap */
1040 if (*ei_startpfn >= end_user_pfn)
1042 if (*ei_endpfn > end_user_pfn)
1043 *ei_endpfn = end_user_pfn;
1048 /* Walk the e820 map and register active regions within a node */
1049 void __init e820_register_active_regions(int nid, unsigned long start_pfn,
1050 unsigned long last_pfn)
1052 unsigned long ei_startpfn;
1053 unsigned long ei_endpfn;
1056 for (i = 0; i < e820.nr_map; i++)
1057 if (e820_find_active_region(&e820.map[i],
1058 start_pfn, last_pfn,
1059 &ei_startpfn, &ei_endpfn))
1060 add_active_range(nid, ei_startpfn, ei_endpfn);
1064 * Find the hole size (in bytes) in the memory range.
1065 * @start: starting address of the memory range to scan
1066 * @end: ending address of the memory range to scan
1068 u64 __init e820_hole_size(u64 start, u64 end)
1070 unsigned long start_pfn = start >> PAGE_SHIFT;
1071 unsigned long last_pfn = end >> PAGE_SHIFT;
1072 unsigned long ei_startpfn, ei_endpfn, ram = 0;
1075 for (i = 0; i < e820.nr_map; i++) {
1076 if (e820_find_active_region(&e820.map[i],
1077 start_pfn, last_pfn,
1078 &ei_startpfn, &ei_endpfn))
1079 ram += ei_endpfn - ei_startpfn;
1081 return end - start - ((u64)ram << PAGE_SHIFT);
1084 static void early_panic(char *msg)
1090 /* "mem=nopentium" disables the 4MB page tables. */
1091 static int __init parse_memopt(char *p)
1098 #ifdef CONFIG_X86_32
1099 if (!strcmp(p, "nopentium")) {
1100 setup_clear_cpu_cap(X86_FEATURE_PSE);
1105 mem_size = memparse(p, &p);
1106 end_user_pfn = mem_size>>PAGE_SHIFT;
1109 early_param("mem", parse_memopt);
1111 static int userdef __initdata;
1113 static int __init parse_memmap_opt(char *p)
1116 u64 start_at, mem_size;
1118 if (!strcmp(p, "exactmap")) {
1119 #ifdef CONFIG_CRASH_DUMP
1121 * If we are doing a crash dump, we still need to know
1122 * the real mem size before original memory map is
1125 e820_register_active_regions(0, 0, -1UL);
1126 saved_max_pfn = e820_end_of_ram();
1127 remove_all_active_ranges();
1135 mem_size = memparse(p, &p);
1141 start_at = memparse(p+1, &p);
1142 e820_add_region(start_at, mem_size, E820_RAM);
1143 } else if (*p == '#') {
1144 start_at = memparse(p+1, &p);
1145 e820_add_region(start_at, mem_size, E820_ACPI);
1146 } else if (*p == '$') {
1147 start_at = memparse(p+1, &p);
1148 e820_add_region(start_at, mem_size, E820_RESERVED);
1150 end_user_pfn = (mem_size >> PAGE_SHIFT);
1152 return *p == '\0' ? 0 : -EINVAL;
1154 early_param("memmap", parse_memmap_opt);
1156 void __init finish_e820_parsing(void)
1159 int nr = e820.nr_map;
1161 if (sanitize_e820_map(e820.map, ARRAY_SIZE(e820.map), &nr) < 0)
1162 early_panic("Invalid user supplied memory map");
1165 printk(KERN_INFO "user-defined physical RAM map:\n");
1166 e820_print_map("user");
1171 * Mark e820 reserved areas as busy for the resource manager.
1173 void __init e820_reserve_resources(void)
1176 struct resource *res;
1178 res = alloc_bootmem_low(sizeof(struct resource) * e820.nr_map);
1179 for (i = 0; i < e820.nr_map; i++) {
1180 switch (e820.map[i].type) {
1181 case E820_RAM: res->name = "System RAM"; break;
1182 case E820_ACPI: res->name = "ACPI Tables"; break;
1183 case E820_NVS: res->name = "ACPI Non-volatile Storage"; break;
1184 default: res->name = "reserved";
1186 res->start = e820.map[i].addr;
1187 res->end = res->start + e820.map[i].size - 1;
1188 #ifndef CONFIG_RESOURCES_64BIT
1189 if (res->end > 0x100000000ULL) {
1194 res->flags = IORESOURCE_MEM | IORESOURCE_BUSY;
1195 insert_resource(&iomem_resource, res);
1200 char *__init default_machine_specific_memory_setup(void)
1202 char *who = "BIOS-e820";
1205 * Try to copy the BIOS-supplied E820-map.
1207 * Otherwise fake a memory map; one section from 0k->640k,
1208 * the next section from 1mb->appropriate_mem_k
1210 new_nr = boot_params.e820_entries;
1211 sanitize_e820_map(boot_params.e820_map,
1212 ARRAY_SIZE(boot_params.e820_map),
1214 boot_params.e820_entries = new_nr;
1215 if (copy_e820_map(boot_params.e820_map, boot_params.e820_entries) < 0) {
1218 /* compare results from other methods and take the greater */
1219 if (boot_params.alt_mem_k
1220 < boot_params.screen_info.ext_mem_k) {
1221 mem_size = boot_params.screen_info.ext_mem_k;
1224 mem_size = boot_params.alt_mem_k;
1229 e820_add_region(0, LOWMEMSIZE(), E820_RAM);
1230 e820_add_region(HIGH_MEMORY, mem_size << 10, E820_RAM);
1233 /* In case someone cares... */
1237 char *__init __attribute__((weak)) machine_specific_memory_setup(void)
1239 return default_machine_specific_memory_setup();
1242 /* Overridden in paravirt.c if CONFIG_PARAVIRT */
1243 char * __init __attribute__((weak)) memory_setup(void)
1245 return machine_specific_memory_setup();
1248 void __init setup_memory_map(void)
1250 printk(KERN_INFO "BIOS-provided physical RAM map:\n");
1251 e820_print_map(memory_setup());
1254 #ifdef CONFIG_X86_64
1255 int __init arch_get_ram_range(int slot, u64 *addr, u64 *size)
1259 if (slot < 0 || slot >= e820.nr_map)
1261 for (i = slot; i < e820.nr_map; i++) {
1262 if (e820.map[i].type != E820_RAM)
1266 if (i == e820.nr_map || e820.map[i].addr > (max_pfn << PAGE_SHIFT))
1268 *addr = e820.map[i].addr;
1269 *size = min_t(u64, e820.map[i].size + e820.map[i].addr,
1270 max_pfn << PAGE_SHIFT) - *addr;