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>
22 #include <linux/firmware-map.h>
24 #include <asm/pgtable.h>
27 #include <asm/proto.h>
28 #include <asm/setup.h>
29 #include <asm/trampoline.h>
32 * The e820 map is the map that gets modified e.g. with command line parameters
33 * and that is also registered with modifications in the kernel resource tree
34 * with the iomem_resource as parent.
36 * The e820_saved is directly saved after the BIOS-provided memory map is
37 * copied. It doesn't get modified afterwards. It's registered for the
38 * /sys/firmware/memmap interface.
40 * That memory map is not modified and is used as base for kexec. The kexec'd
41 * kernel should get the same memory map as the firmware provides. Then the
42 * user can e.g. boot the original kernel with mem=1G while still booting the
43 * next kernel with full memory.
46 struct e820map 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 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
109 EXPORT_SYMBOL_GPL(e820_all_mapped);
112 * Add a memory region to the kernel e820 map.
114 static void __init __e820_add_region(struct e820map *e820x, u64 start, u64 size,
117 int x = e820x->nr_map;
119 if (x >= ARRAY_SIZE(e820x->map)) {
120 printk(KERN_ERR "Ooops! Too many entries in the memory map!\n");
124 e820x->map[x].addr = start;
125 e820x->map[x].size = size;
126 e820x->map[x].type = type;
130 void __init e820_add_region(u64 start, u64 size, int type)
132 __e820_add_region(&e820, start, size, type);
135 static void __init e820_print_type(u32 type)
139 case E820_RESERVED_KERN:
140 printk(KERN_CONT "(usable)");
143 printk(KERN_CONT "(reserved)");
146 printk(KERN_CONT "(ACPI data)");
149 printk(KERN_CONT "(ACPI NVS)");
152 printk(KERN_CONT "(unusable)");
155 printk(KERN_CONT "type %u", type);
160 void __init e820_print_map(char *who)
164 for (i = 0; i < e820.nr_map; i++) {
165 printk(KERN_INFO " %s: %016Lx - %016Lx ", who,
166 (unsigned long long) e820.map[i].addr,
168 (e820.map[i].addr + e820.map[i].size));
169 e820_print_type(e820.map[i].type);
170 printk(KERN_CONT "\n");
175 * Sanitize the BIOS e820 map.
177 * Some e820 responses include overlapping entries. The following
178 * replaces the original e820 map with a new one, removing overlaps,
179 * and resolving conflicting memory types in favor of highest
182 * The input parameter biosmap points to an array of 'struct
183 * e820entry' which on entry has elements in the range [0, *pnr_map)
184 * valid, and which has space for up to max_nr_map entries.
185 * On return, the resulting sanitized e820 map entries will be in
186 * overwritten in the same location, starting at biosmap.
188 * The integer pointed to by pnr_map must be valid on entry (the
189 * current number of valid entries located at biosmap) and will
190 * be updated on return, with the new number of valid entries
191 * (something no more than max_nr_map.)
193 * The return value from sanitize_e820_map() is zero if it
194 * successfully 'sanitized' the map entries passed in, and is -1
195 * if it did nothing, which can happen if either of (1) it was
196 * only passed one map entry, or (2) any of the input map entries
197 * were invalid (start + size < start, meaning that the size was
198 * so big the described memory range wrapped around through zero.)
200 * Visually we're performing the following
201 * (1,2,3,4 = memory types)...
203 * Sample memory map (w/overlaps):
204 * ____22__________________
205 * ______________________4_
206 * ____1111________________
207 * _44_____________________
208 * 11111111________________
209 * ____________________33__
210 * ___________44___________
211 * __________33333_________
212 * ______________22________
213 * ___________________2222_
214 * _________111111111______
215 * _____________________11_
216 * _________________4______
218 * Sanitized equivalent (no overlap):
219 * 1_______________________
220 * _44_____________________
221 * ___1____________________
222 * ____22__________________
223 * ______11________________
224 * _________1______________
225 * __________3_____________
226 * ___________44___________
227 * _____________33_________
228 * _______________2________
229 * ________________1_______
230 * _________________4______
231 * ___________________2____
232 * ____________________33__
233 * ______________________4_
236 int __init sanitize_e820_map(struct e820entry *biosmap, int max_nr_map,
239 struct change_member {
240 struct e820entry *pbios; /* pointer to original bios entry */
241 unsigned long long addr; /* address for this change point */
243 static struct change_member change_point_list[2*E820_X_MAX] __initdata;
244 static struct change_member *change_point[2*E820_X_MAX] __initdata;
245 static struct e820entry *overlap_list[E820_X_MAX] __initdata;
246 static struct e820entry new_bios[E820_X_MAX] __initdata;
247 struct change_member *change_tmp;
248 unsigned long current_type, last_type;
249 unsigned long long last_addr;
250 int chgidx, still_changing;
253 int old_nr, new_nr, chg_nr;
256 /* if there's only one memory region, don't bother */
261 BUG_ON(old_nr > max_nr_map);
263 /* bail out if we find any unreasonable addresses in bios map */
264 for (i = 0; i < old_nr; i++)
265 if (biosmap[i].addr + biosmap[i].size < biosmap[i].addr)
268 /* create pointers for initial change-point information (for sorting) */
269 for (i = 0; i < 2 * old_nr; i++)
270 change_point[i] = &change_point_list[i];
272 /* record all known change-points (starting and ending addresses),
273 omitting those that are for empty memory regions */
275 for (i = 0; i < old_nr; i++) {
276 if (biosmap[i].size != 0) {
277 change_point[chgidx]->addr = biosmap[i].addr;
278 change_point[chgidx++]->pbios = &biosmap[i];
279 change_point[chgidx]->addr = biosmap[i].addr +
281 change_point[chgidx++]->pbios = &biosmap[i];
286 /* sort change-point list by memory addresses (low -> high) */
288 while (still_changing) {
290 for (i = 1; i < chg_nr; i++) {
291 unsigned long long curaddr, lastaddr;
292 unsigned long long curpbaddr, lastpbaddr;
294 curaddr = change_point[i]->addr;
295 lastaddr = change_point[i - 1]->addr;
296 curpbaddr = change_point[i]->pbios->addr;
297 lastpbaddr = change_point[i - 1]->pbios->addr;
300 * swap entries, when:
302 * curaddr > lastaddr or
303 * curaddr == lastaddr and curaddr == curpbaddr and
304 * lastaddr != lastpbaddr
306 if (curaddr < lastaddr ||
307 (curaddr == lastaddr && curaddr == curpbaddr &&
308 lastaddr != lastpbaddr)) {
309 change_tmp = change_point[i];
310 change_point[i] = change_point[i-1];
311 change_point[i-1] = change_tmp;
317 /* create a new bios memory map, removing overlaps */
318 overlap_entries = 0; /* number of entries in the overlap table */
319 new_bios_entry = 0; /* index for creating new bios map entries */
320 last_type = 0; /* start with undefined memory type */
321 last_addr = 0; /* start with 0 as last starting address */
323 /* loop through change-points, determining affect on the new bios map */
324 for (chgidx = 0; chgidx < chg_nr; chgidx++) {
325 /* keep track of all overlapping bios entries */
326 if (change_point[chgidx]->addr ==
327 change_point[chgidx]->pbios->addr) {
329 * add map entry to overlap list (> 1 entry
330 * implies an overlap)
332 overlap_list[overlap_entries++] =
333 change_point[chgidx]->pbios;
336 * remove entry from list (order independent,
339 for (i = 0; i < overlap_entries; i++) {
340 if (overlap_list[i] ==
341 change_point[chgidx]->pbios)
343 overlap_list[overlap_entries-1];
348 * if there are overlapping entries, decide which
349 * "type" to use (larger value takes precedence --
350 * 1=usable, 2,3,4,4+=unusable)
353 for (i = 0; i < overlap_entries; i++)
354 if (overlap_list[i]->type > current_type)
355 current_type = overlap_list[i]->type;
357 * continue building up new bios map based on this
360 if (current_type != last_type) {
361 if (last_type != 0) {
362 new_bios[new_bios_entry].size =
363 change_point[chgidx]->addr - last_addr;
365 * move forward only if the new size
368 if (new_bios[new_bios_entry].size != 0)
370 * no more space left for new
373 if (++new_bios_entry >= max_nr_map)
376 if (current_type != 0) {
377 new_bios[new_bios_entry].addr =
378 change_point[chgidx]->addr;
379 new_bios[new_bios_entry].type = current_type;
380 last_addr = change_point[chgidx]->addr;
382 last_type = current_type;
385 /* retain count for new bios entries */
386 new_nr = new_bios_entry;
388 /* copy new bios mapping into original location */
389 memcpy(biosmap, new_bios, new_nr * sizeof(struct e820entry));
395 static int __init __append_e820_map(struct e820entry *biosmap, int nr_map)
398 u64 start = biosmap->addr;
399 u64 size = biosmap->size;
400 u64 end = start + size;
401 u32 type = biosmap->type;
403 /* Overflow in 64 bits? Ignore the memory map. */
407 e820_add_region(start, size, type);
416 * Copy the BIOS e820 map into a safe place.
418 * Sanity-check it while we're at it..
420 * If we're lucky and live on a modern system, the setup code
421 * will have given us a memory map that we can use to properly
422 * set up memory. If we aren't, we'll fake a memory map.
424 static int __init append_e820_map(struct e820entry *biosmap, int nr_map)
426 /* Only one memory region (or negative)? Ignore it */
430 return __append_e820_map(biosmap, nr_map);
433 static u64 __init __e820_update_range(struct e820map *e820x, u64 start,
434 u64 size, unsigned old_type,
439 u64 real_updated_size = 0;
441 BUG_ON(old_type == new_type);
443 if (size > (ULLONG_MAX - start))
444 size = ULLONG_MAX - start;
447 printk(KERN_DEBUG "e820 update range: %016Lx - %016Lx ",
448 (unsigned long long) start,
449 (unsigned long long) end);
450 e820_print_type(old_type);
451 printk(KERN_CONT " ==> ");
452 e820_print_type(new_type);
453 printk(KERN_CONT "\n");
455 for (i = 0; i < e820x->nr_map; i++) {
456 struct e820entry *ei = &e820x->map[i];
457 u64 final_start, final_end;
460 if (ei->type != old_type)
463 ei_end = ei->addr + ei->size;
464 /* totally covered by new range? */
465 if (ei->addr >= start && ei_end <= end) {
467 real_updated_size += ei->size;
471 /* new range is totally covered? */
472 if (ei->addr < start && ei_end > end) {
473 __e820_add_region(e820x, start, size, new_type);
474 __e820_add_region(e820x, end, ei_end - end, ei->type);
475 ei->size = start - ei->addr;
476 real_updated_size += size;
480 /* partially covered */
481 final_start = max(start, ei->addr);
482 final_end = min(end, ei_end);
483 if (final_start >= final_end)
486 __e820_add_region(e820x, final_start, final_end - final_start,
489 real_updated_size += final_end - final_start;
492 * left range could be head or tail, so need to update
495 ei->size -= final_end - final_start;
496 if (ei->addr < final_start)
498 ei->addr = final_end;
500 return real_updated_size;
503 u64 __init e820_update_range(u64 start, u64 size, unsigned old_type,
506 return __e820_update_range(&e820, start, size, old_type, new_type);
509 static u64 __init e820_update_range_saved(u64 start, u64 size,
510 unsigned old_type, unsigned new_type)
512 return __e820_update_range(&e820_saved, start, size, old_type,
516 /* make e820 not cover the range */
517 u64 __init e820_remove_range(u64 start, u64 size, unsigned old_type,
521 u64 real_removed_size = 0;
523 if (size > (ULLONG_MAX - start))
524 size = ULLONG_MAX - start;
526 for (i = 0; i < e820.nr_map; i++) {
527 struct e820entry *ei = &e820.map[i];
528 u64 final_start, final_end;
530 if (checktype && ei->type != old_type)
532 /* totally covered? */
533 if (ei->addr >= start &&
534 (ei->addr + ei->size) <= (start + size)) {
535 real_removed_size += ei->size;
536 memset(ei, 0, sizeof(struct e820entry));
539 /* partially covered */
540 final_start = max(start, ei->addr);
541 final_end = min(start + size, ei->addr + ei->size);
542 if (final_start >= final_end)
544 real_removed_size += final_end - final_start;
546 ei->size -= final_end - final_start;
547 if (ei->addr < final_start)
549 ei->addr = final_end;
551 return real_removed_size;
554 void __init update_e820(void)
558 nr_map = e820.nr_map;
559 if (sanitize_e820_map(e820.map, ARRAY_SIZE(e820.map), &nr_map))
561 e820.nr_map = nr_map;
562 printk(KERN_INFO "modified physical RAM map:\n");
563 e820_print_map("modified");
565 static void __init update_e820_saved(void)
569 nr_map = e820_saved.nr_map;
570 if (sanitize_e820_map(e820_saved.map, ARRAY_SIZE(e820_saved.map), &nr_map))
572 e820_saved.nr_map = nr_map;
574 #define MAX_GAP_END 0x100000000ull
576 * Search for a gap in the e820 memory space from start_addr to end_addr.
578 __init int e820_search_gap(unsigned long *gapstart, unsigned long *gapsize,
579 unsigned long start_addr, unsigned long long end_addr)
581 unsigned long long last;
585 last = (end_addr && end_addr < MAX_GAP_END) ? end_addr : MAX_GAP_END;
588 unsigned long long start = e820.map[i].addr;
589 unsigned long long end = start + e820.map[i].size;
591 if (end < start_addr)
595 * Since "last" is at most 4GB, we know we'll
596 * fit in 32 bits if this condition is true
599 unsigned long gap = last - end;
601 if (gap >= *gapsize) {
614 * Search for the biggest gap in the low 32 bits of the e820
615 * memory space. We pass this space to PCI to assign MMIO resources
616 * for hotplug or unconfigured devices in.
617 * Hopefully the BIOS let enough space left.
619 __init void e820_setup_gap(void)
621 unsigned long gapstart, gapsize;
624 gapstart = 0x10000000;
626 found = e820_search_gap(&gapstart, &gapsize, 0, MAX_GAP_END);
630 gapstart = (max_pfn << PAGE_SHIFT) + 1024*1024;
632 "PCI: Warning: Cannot find a gap in the 32bit address range\n"
633 "PCI: Unassigned devices with 32bit resource registers may break!\n");
638 * e820_reserve_resources_late protect stolen RAM already
640 pci_mem_start = gapstart;
643 "Allocating PCI resources starting at %lx (gap: %lx:%lx)\n",
644 pci_mem_start, gapstart, gapsize);
648 * Because of the size limitation of struct boot_params, only first
649 * 128 E820 memory entries are passed to kernel via
650 * boot_params.e820_map, others are passed via SETUP_E820_EXT node of
651 * linked list of struct setup_data, which is parsed here.
653 void __init parse_e820_ext(struct setup_data *sdata, unsigned long pa_data)
657 struct e820entry *extmap;
659 entries = sdata->len / sizeof(struct e820entry);
660 map_len = sdata->len + sizeof(struct setup_data);
661 if (map_len > PAGE_SIZE)
662 sdata = early_ioremap(pa_data, map_len);
663 extmap = (struct e820entry *)(sdata->data);
664 __append_e820_map(extmap, entries);
665 sanitize_e820_map(e820.map, ARRAY_SIZE(e820.map), &e820.nr_map);
666 if (map_len > PAGE_SIZE)
667 early_iounmap(sdata, map_len);
668 printk(KERN_INFO "extended physical RAM map:\n");
669 e820_print_map("extended");
672 #if defined(CONFIG_X86_64) || \
673 (defined(CONFIG_X86_32) && defined(CONFIG_HIBERNATION))
675 * Find the ranges of physical addresses that do not correspond to
676 * e820 RAM areas and mark the corresponding pages as nosave for
677 * hibernation (32 bit) or software suspend and suspend to RAM (64 bit).
679 * This function requires the e820 map to be sorted and without any
680 * overlapping entries and assumes the first e820 area to be RAM.
682 void __init e820_mark_nosave_regions(unsigned long limit_pfn)
687 pfn = PFN_DOWN(e820.map[0].addr + e820.map[0].size);
688 for (i = 1; i < e820.nr_map; i++) {
689 struct e820entry *ei = &e820.map[i];
691 if (pfn < PFN_UP(ei->addr))
692 register_nosave_region(pfn, PFN_UP(ei->addr));
694 pfn = PFN_DOWN(ei->addr + ei->size);
695 if (ei->type != E820_RAM && ei->type != E820_RESERVED_KERN)
696 register_nosave_region(PFN_UP(ei->addr), pfn);
698 if (pfn >= limit_pfn)
704 #ifdef CONFIG_HIBERNATION
706 * Mark ACPI NVS memory region, so that we can save/restore it during
707 * hibernation and the subsequent resume.
709 static int __init e820_mark_nvs_memory(void)
713 for (i = 0; i < e820.nr_map; i++) {
714 struct e820entry *ei = &e820.map[i];
716 if (ei->type == E820_NVS)
717 hibernate_nvs_register(ei->addr, ei->size);
722 core_initcall(e820_mark_nvs_memory);
726 * Early reserved memory areas.
728 #define MAX_EARLY_RES 32
735 static struct early_res early_res[MAX_EARLY_RES] __initdata = {
736 { 0, PAGE_SIZE, "BIOS data page", 1 }, /* BIOS data page */
737 #if defined(CONFIG_X86_32) && defined(CONFIG_X86_TRAMPOLINE)
739 * But first pinch a few for the stack/trampoline stuff
740 * FIXME: Don't need the extra page at 4K, but need to fix
741 * trampoline before removing it. (see the GDT stuff)
743 { PAGE_SIZE, PAGE_SIZE + PAGE_SIZE, "EX TRAMPOLINE", 1 },
749 static int __init find_overlapped_early(u64 start, u64 end)
754 for (i = 0; i < MAX_EARLY_RES && early_res[i].end; i++) {
756 if (end > r->start && start < r->end)
764 * Drop the i-th range from the early reservation map,
765 * by copying any higher ranges down one over it, and
766 * clearing what had been the last slot.
768 static void __init drop_range(int i)
772 for (j = i + 1; j < MAX_EARLY_RES && early_res[j].end; j++)
775 memmove(&early_res[i], &early_res[i + 1],
776 (j - 1 - i) * sizeof(struct early_res));
778 early_res[j - 1].end = 0;
782 * Split any existing ranges that:
783 * 1) are marked 'overlap_ok', and
784 * 2) overlap with the stated range [start, end)
785 * into whatever portion (if any) of the existing range is entirely
786 * below or entirely above the stated range. Drop the portion
787 * of the existing range that overlaps with the stated range,
788 * which will allow the caller of this routine to then add that
789 * stated range without conflicting with any existing range.
791 static void __init drop_overlaps_that_are_ok(u64 start, u64 end)
795 u64 lower_start, lower_end;
796 u64 upper_start, upper_end;
799 for (i = 0; i < MAX_EARLY_RES && early_res[i].end; i++) {
802 /* Continue past non-overlapping ranges */
803 if (end <= r->start || start >= r->end)
807 * Leave non-ok overlaps as is; let caller
808 * panic "Overlapping early reservations"
809 * when it hits this overlap.
815 * We have an ok overlap. We will drop it from the early
816 * reservation map, and add back in any non-overlapping
817 * portions (lower or upper) as separate, overlap_ok,
818 * non-overlapping ranges.
821 /* 1. Note any non-overlapping (lower or upper) ranges. */
822 strncpy(name, r->name, sizeof(name) - 1);
824 lower_start = lower_end = 0;
825 upper_start = upper_end = 0;
826 if (r->start < start) {
827 lower_start = r->start;
835 /* 2. Drop the original ok overlapping range */
838 i--; /* resume for-loop on copied down entry */
840 /* 3. Add back in any non-overlapping ranges. */
842 reserve_early_overlap_ok(lower_start, lower_end, name);
844 reserve_early_overlap_ok(upper_start, upper_end, name);
848 static void __init __reserve_early(u64 start, u64 end, char *name,
854 i = find_overlapped_early(start, end);
855 if (i >= MAX_EARLY_RES)
856 panic("Too many early reservations");
859 panic("Overlapping early reservations "
860 "%llx-%llx %s to %llx-%llx %s\n",
861 start, end - 1, name?name:"", r->start,
862 r->end - 1, r->name);
865 r->overlap_ok = overlap_ok;
867 strncpy(r->name, name, sizeof(r->name) - 1);
871 * A few early reservtations come here.
873 * The 'overlap_ok' in the name of this routine does -not- mean it
874 * is ok for these reservations to overlap an earlier reservation.
875 * Rather it means that it is ok for subsequent reservations to
878 * Use this entry point to reserve early ranges when you are doing
879 * so out of "Paranoia", reserving perhaps more memory than you need,
880 * just in case, and don't mind a subsequent overlapping reservation
881 * that is known to be needed.
883 * The drop_overlaps_that_are_ok() call here isn't really needed.
884 * It would be needed if we had two colliding 'overlap_ok'
885 * reservations, so that the second such would not panic on the
886 * overlap with the first. We don't have any such as of this
887 * writing, but might as well tolerate such if it happens in
890 void __init reserve_early_overlap_ok(u64 start, u64 end, char *name)
892 drop_overlaps_that_are_ok(start, end);
893 __reserve_early(start, end, name, 1);
897 * Most early reservations come here.
899 * We first have drop_overlaps_that_are_ok() drop any pre-existing
900 * 'overlap_ok' ranges, so that we can then reserve this memory
901 * range without risk of panic'ing on an overlapping overlap_ok
904 void __init reserve_early(u64 start, u64 end, char *name)
909 drop_overlaps_that_are_ok(start, end);
910 __reserve_early(start, end, name, 0);
913 void __init free_early(u64 start, u64 end)
918 i = find_overlapped_early(start, end);
920 if (i >= MAX_EARLY_RES || r->end != end || r->start != start)
921 panic("free_early on not reserved area: %llx-%llx!",
927 void __init early_res_to_bootmem(u64 start, u64 end)
930 u64 final_start, final_end;
933 for (i = 0; i < MAX_EARLY_RES && early_res[i].end; i++)
936 printk(KERN_INFO "(%d early reservations) ==> bootmem [%010llx - %010llx]\n",
938 for (i = 0; i < count; i++) {
939 struct early_res *r = &early_res[i];
940 printk(KERN_INFO " #%d [%010llx - %010llx] %16s", i,
941 r->start, r->end, r->name);
942 final_start = max(start, r->start);
943 final_end = min(end, r->end);
944 if (final_start >= final_end) {
945 printk(KERN_CONT "\n");
948 printk(KERN_CONT " ==> [%010llx - %010llx]\n",
949 final_start, final_end);
950 reserve_bootmem_generic(final_start, final_end - final_start,
955 /* Check for already reserved areas */
956 static inline int __init bad_addr(u64 *addrp, u64 size, u64 align)
963 i = find_overlapped_early(addr, addr + size);
965 if (i < MAX_EARLY_RES && r->end) {
966 *addrp = addr = round_up(r->end, align);
973 /* Check for already reserved areas */
974 static inline int __init bad_addr_size(u64 *addrp, u64 *sizep, u64 align)
977 u64 addr = *addrp, last;
982 for (i = 0; i < MAX_EARLY_RES && early_res[i].end; i++) {
983 struct early_res *r = &early_res[i];
984 if (last > r->start && addr < r->start) {
985 size = r->start - addr;
989 if (last > r->end && addr < r->end) {
990 addr = round_up(r->end, align);
995 if (last <= r->end && addr >= r->start) {
1008 * Find a free area with specified alignment in a specific range.
1010 u64 __init find_e820_area(u64 start, u64 end, u64 size, u64 align)
1014 for (i = 0; i < e820.nr_map; i++) {
1015 struct e820entry *ei = &e820.map[i];
1019 if (ei->type != E820_RAM)
1021 addr = round_up(ei->addr, align);
1022 ei_last = ei->addr + ei->size;
1024 addr = round_up(start, align);
1025 if (addr >= ei_last)
1027 while (bad_addr(&addr, size, align) && addr+size <= ei_last)
1040 * Find next free range after *start
1042 u64 __init find_e820_area_size(u64 start, u64 *sizep, u64 align)
1046 for (i = 0; i < e820.nr_map; i++) {
1047 struct e820entry *ei = &e820.map[i];
1051 if (ei->type != E820_RAM)
1053 addr = round_up(ei->addr, align);
1054 ei_last = ei->addr + ei->size;
1056 addr = round_up(start, align);
1057 if (addr >= ei_last)
1059 *sizep = ei_last - addr;
1060 while (bad_addr_size(&addr, sizep, align) &&
1061 addr + *sizep <= ei_last)
1063 last = addr + *sizep;
1073 * pre allocated 4k and reserved it in e820
1075 u64 __init early_reserve_e820(u64 startt, u64 sizet, u64 align)
1081 for (start = startt; ; start += size) {
1082 start = find_e820_area_size(start, &size, align);
1089 #ifdef CONFIG_X86_32
1090 if (start >= MAXMEM)
1092 if (start + size > MAXMEM)
1093 size = MAXMEM - start;
1096 addr = round_down(start + size - sizet, align);
1099 e820_update_range(addr, sizet, E820_RAM, E820_RESERVED);
1100 e820_update_range_saved(addr, sizet, E820_RAM, E820_RESERVED);
1101 printk(KERN_INFO "update e820 for early_reserve_e820\n");
1103 update_e820_saved();
1108 #ifdef CONFIG_X86_32
1109 # ifdef CONFIG_X86_PAE
1110 # define MAX_ARCH_PFN (1ULL<<(36-PAGE_SHIFT))
1112 # define MAX_ARCH_PFN (1ULL<<(32-PAGE_SHIFT))
1114 #else /* CONFIG_X86_32 */
1115 # define MAX_ARCH_PFN MAXMEM>>PAGE_SHIFT
1119 * Find the highest page frame number we have available
1121 static unsigned long __init e820_end_pfn(unsigned long limit_pfn, unsigned type)
1124 unsigned long last_pfn = 0;
1125 unsigned long max_arch_pfn = MAX_ARCH_PFN;
1127 for (i = 0; i < e820.nr_map; i++) {
1128 struct e820entry *ei = &e820.map[i];
1129 unsigned long start_pfn;
1130 unsigned long end_pfn;
1132 if (ei->type != type)
1135 start_pfn = ei->addr >> PAGE_SHIFT;
1136 end_pfn = (ei->addr + ei->size) >> PAGE_SHIFT;
1138 if (start_pfn >= limit_pfn)
1140 if (end_pfn > limit_pfn) {
1141 last_pfn = limit_pfn;
1144 if (end_pfn > last_pfn)
1148 if (last_pfn > max_arch_pfn)
1149 last_pfn = max_arch_pfn;
1151 printk(KERN_INFO "last_pfn = %#lx max_arch_pfn = %#lx\n",
1152 last_pfn, max_arch_pfn);
1155 unsigned long __init e820_end_of_ram_pfn(void)
1157 return e820_end_pfn(MAX_ARCH_PFN, E820_RAM);
1160 unsigned long __init e820_end_of_low_ram_pfn(void)
1162 return e820_end_pfn(1UL<<(32 - PAGE_SHIFT), E820_RAM);
1165 * Finds an active region in the address range from start_pfn to last_pfn and
1166 * returns its range in ei_startpfn and ei_endpfn for the e820 entry.
1168 int __init e820_find_active_region(const struct e820entry *ei,
1169 unsigned long start_pfn,
1170 unsigned long last_pfn,
1171 unsigned long *ei_startpfn,
1172 unsigned long *ei_endpfn)
1174 u64 align = PAGE_SIZE;
1176 *ei_startpfn = round_up(ei->addr, align) >> PAGE_SHIFT;
1177 *ei_endpfn = round_down(ei->addr + ei->size, align) >> PAGE_SHIFT;
1179 /* Skip map entries smaller than a page */
1180 if (*ei_startpfn >= *ei_endpfn)
1183 /* Skip if map is outside the node */
1184 if (ei->type != E820_RAM || *ei_endpfn <= start_pfn ||
1185 *ei_startpfn >= last_pfn)
1188 /* Check for overlaps */
1189 if (*ei_startpfn < start_pfn)
1190 *ei_startpfn = start_pfn;
1191 if (*ei_endpfn > last_pfn)
1192 *ei_endpfn = last_pfn;
1197 /* Walk the e820 map and register active regions within a node */
1198 void __init e820_register_active_regions(int nid, unsigned long start_pfn,
1199 unsigned long last_pfn)
1201 unsigned long ei_startpfn;
1202 unsigned long ei_endpfn;
1205 for (i = 0; i < e820.nr_map; i++)
1206 if (e820_find_active_region(&e820.map[i],
1207 start_pfn, last_pfn,
1208 &ei_startpfn, &ei_endpfn))
1209 add_active_range(nid, ei_startpfn, ei_endpfn);
1213 * Find the hole size (in bytes) in the memory range.
1214 * @start: starting address of the memory range to scan
1215 * @end: ending address of the memory range to scan
1217 u64 __init e820_hole_size(u64 start, u64 end)
1219 unsigned long start_pfn = start >> PAGE_SHIFT;
1220 unsigned long last_pfn = end >> PAGE_SHIFT;
1221 unsigned long ei_startpfn, ei_endpfn, ram = 0;
1224 for (i = 0; i < e820.nr_map; i++) {
1225 if (e820_find_active_region(&e820.map[i],
1226 start_pfn, last_pfn,
1227 &ei_startpfn, &ei_endpfn))
1228 ram += ei_endpfn - ei_startpfn;
1230 return end - start - ((u64)ram << PAGE_SHIFT);
1233 static void early_panic(char *msg)
1239 static int userdef __initdata;
1241 /* "mem=nopentium" disables the 4MB page tables. */
1242 static int __init parse_memopt(char *p)
1249 #ifdef CONFIG_X86_32
1250 if (!strcmp(p, "nopentium")) {
1251 setup_clear_cpu_cap(X86_FEATURE_PSE);
1257 mem_size = memparse(p, &p);
1258 e820_remove_range(mem_size, ULLONG_MAX - mem_size, E820_RAM, 1);
1262 early_param("mem", parse_memopt);
1264 static int __init parse_memmap_opt(char *p)
1267 u64 start_at, mem_size;
1272 if (!strncmp(p, "exactmap", 8)) {
1273 #ifdef CONFIG_CRASH_DUMP
1275 * If we are doing a crash dump, we still need to know
1276 * the real mem size before original memory map is
1279 saved_max_pfn = e820_end_of_ram_pfn();
1287 mem_size = memparse(p, &p);
1293 start_at = memparse(p+1, &p);
1294 e820_add_region(start_at, mem_size, E820_RAM);
1295 } else if (*p == '#') {
1296 start_at = memparse(p+1, &p);
1297 e820_add_region(start_at, mem_size, E820_ACPI);
1298 } else if (*p == '$') {
1299 start_at = memparse(p+1, &p);
1300 e820_add_region(start_at, mem_size, E820_RESERVED);
1302 e820_remove_range(mem_size, ULLONG_MAX - mem_size, E820_RAM, 1);
1304 return *p == '\0' ? 0 : -EINVAL;
1306 early_param("memmap", parse_memmap_opt);
1308 void __init finish_e820_parsing(void)
1311 u32 nr = e820.nr_map;
1313 if (sanitize_e820_map(e820.map, ARRAY_SIZE(e820.map), &nr) < 0)
1314 early_panic("Invalid user supplied memory map");
1317 printk(KERN_INFO "user-defined physical RAM map:\n");
1318 e820_print_map("user");
1322 static inline const char *e820_type_to_string(int e820_type)
1324 switch (e820_type) {
1325 case E820_RESERVED_KERN:
1326 case E820_RAM: return "System RAM";
1327 case E820_ACPI: return "ACPI Tables";
1328 case E820_NVS: return "ACPI Non-volatile Storage";
1329 case E820_UNUSABLE: return "Unusable memory";
1330 default: return "reserved";
1335 * Mark e820 reserved areas as busy for the resource manager.
1337 static struct resource __initdata *e820_res;
1338 void __init e820_reserve_resources(void)
1341 struct resource *res;
1344 res = alloc_bootmem(sizeof(struct resource) * e820.nr_map);
1346 for (i = 0; i < e820.nr_map; i++) {
1347 end = e820.map[i].addr + e820.map[i].size - 1;
1348 if (end != (resource_size_t)end) {
1352 res->name = e820_type_to_string(e820.map[i].type);
1353 res->start = e820.map[i].addr;
1356 res->flags = IORESOURCE_MEM;
1359 * don't register the region that could be conflicted with
1360 * pci device BAR resource and insert them later in
1361 * pcibios_resource_survey()
1363 if (e820.map[i].type != E820_RESERVED || res->start < (1ULL<<20)) {
1364 res->flags |= IORESOURCE_BUSY;
1365 insert_resource(&iomem_resource, res);
1370 for (i = 0; i < e820_saved.nr_map; i++) {
1371 struct e820entry *entry = &e820_saved.map[i];
1372 firmware_map_add_early(entry->addr,
1373 entry->addr + entry->size - 1,
1374 e820_type_to_string(entry->type));
1378 /* How much should we pad RAM ending depending on where it is? */
1379 static unsigned long ram_alignment(resource_size_t pos)
1381 unsigned long mb = pos >> 20;
1383 /* To 64kB in the first megabyte */
1387 /* To 1MB in the first 16MB */
1391 /* To 64MB for anything above that */
1392 return 64*1024*1024;
1395 #define MAX_RESOURCE_SIZE ((resource_size_t)-1)
1397 void __init e820_reserve_resources_late(void)
1400 struct resource *res;
1403 for (i = 0; i < e820.nr_map; i++) {
1404 if (!res->parent && res->end)
1405 insert_resource_expand_to_fit(&iomem_resource, res);
1410 * Try to bump up RAM regions to reasonable boundaries to
1413 for (i = 0; i < e820.nr_map; i++) {
1414 struct e820entry *entry = &e820.map[i];
1417 if (entry->type != E820_RAM)
1419 start = entry->addr + entry->size;
1420 end = round_up(start, ram_alignment(start)) - 1;
1421 if (end > MAX_RESOURCE_SIZE)
1422 end = MAX_RESOURCE_SIZE;
1425 reserve_region_with_split(&iomem_resource, start, end,
1430 char *__init default_machine_specific_memory_setup(void)
1432 char *who = "BIOS-e820";
1435 * Try to copy the BIOS-supplied E820-map.
1437 * Otherwise fake a memory map; one section from 0k->640k,
1438 * the next section from 1mb->appropriate_mem_k
1440 new_nr = boot_params.e820_entries;
1441 sanitize_e820_map(boot_params.e820_map,
1442 ARRAY_SIZE(boot_params.e820_map),
1444 boot_params.e820_entries = new_nr;
1445 if (append_e820_map(boot_params.e820_map, boot_params.e820_entries)
1449 /* compare results from other methods and take the greater */
1450 if (boot_params.alt_mem_k
1451 < boot_params.screen_info.ext_mem_k) {
1452 mem_size = boot_params.screen_info.ext_mem_k;
1455 mem_size = boot_params.alt_mem_k;
1460 e820_add_region(0, LOWMEMSIZE(), E820_RAM);
1461 e820_add_region(HIGH_MEMORY, mem_size << 10, E820_RAM);
1464 /* In case someone cares... */
1468 void __init setup_memory_map(void)
1472 who = x86_init.resources.memory_setup();
1473 memcpy(&e820_saved, &e820, sizeof(struct e820map));
1474 printk(KERN_INFO "BIOS-provided physical RAM map:\n");
1475 e820_print_map(who);