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 add_memory_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));
363 * Copy the BIOS e820 map into a safe place.
365 * Sanity-check it while we're at it..
367 * If we're lucky and live on a modern system, the setup code
368 * will have given us a memory map that we can use to properly
369 * set up memory. If we aren't, we'll fake a memory map.
371 int __init copy_e820_map(struct e820entry *biosmap, int nr_map)
373 /* Only one memory region (or negative)? Ignore it */
378 u64 start = biosmap->addr;
379 u64 size = biosmap->size;
380 u64 end = start + size;
381 u32 type = biosmap->type;
383 /* Overflow in 64 bits? Ignore the memory map. */
387 add_memory_region(start, size, type);
388 } while (biosmap++, --nr_map);
392 u64 __init update_memory_range(u64 start, u64 size, unsigned old_type,
396 u64 real_updated_size = 0;
398 BUG_ON(old_type == new_type);
400 for (i = 0; i < e820.nr_map; i++) {
401 struct e820entry *ei = &e820.map[i];
402 u64 final_start, final_end;
403 if (ei->type != old_type)
405 /* totally covered? */
406 if (ei->addr >= start &&
407 (ei->addr + ei->size) <= (start + size)) {
409 real_updated_size += ei->size;
412 /* partially covered */
413 final_start = max(start, ei->addr);
414 final_end = min(start + size, ei->addr + ei->size);
415 if (final_start >= final_end)
417 add_memory_region(final_start, final_end - final_start,
419 real_updated_size += final_end - final_start;
421 return real_updated_size;
424 void __init update_e820(void)
428 nr_map = e820.nr_map;
429 if (sanitize_e820_map(e820.map, ARRAY_SIZE(e820.map), &nr_map))
431 e820.nr_map = nr_map;
432 printk(KERN_INFO "modified physical RAM map:\n");
433 e820_print_map("modified");
437 * Search for the biggest gap in the low 32 bits of the e820
438 * memory space. We pass this space to PCI to assign MMIO resources
439 * for hotplug or unconfigured devices in.
440 * Hopefully the BIOS let enough space left.
442 __init void e820_setup_gap(void)
444 unsigned long gapstart, gapsize, round;
445 unsigned long long last;
449 last = 0x100000000ull;
450 gapstart = 0x10000000;
454 unsigned long long start = e820.map[i].addr;
455 unsigned long long end = start + e820.map[i].size;
458 * Since "last" is at most 4GB, we know we'll
459 * fit in 32 bits if this condition is true
462 unsigned long gap = last - end;
476 gapstart = (end_pfn << PAGE_SHIFT) + 1024*1024;
477 printk(KERN_ERR "PCI: Warning: Cannot find a gap in the 32bit "
479 KERN_ERR "PCI: Unassigned devices with 32bit resource "
480 "registers may break!\n");
485 * See how much we want to round up: start off with
486 * rounding to the next 1MB area.
489 while ((gapsize >> 4) > round)
491 /* Fun with two's complement */
492 pci_mem_start = (gapstart + round) & -round;
495 "Allocating PCI resources starting at %lx (gap: %lx:%lx)\n",
496 pci_mem_start, gapstart, gapsize);
499 #if defined(CONFIG_X86_64) || \
500 (defined(CONFIG_X86_32) && defined(CONFIG_HIBERNATION))
502 * Find the ranges of physical addresses that do not correspond to
503 * e820 RAM areas and mark the corresponding pages as nosave for
504 * hibernation (32 bit) or software suspend and suspend to RAM (64 bit).
506 * This function requires the e820 map to be sorted and without any
507 * overlapping entries and assumes the first e820 area to be RAM.
509 void __init e820_mark_nosave_regions(unsigned long limit_pfn)
514 pfn = PFN_DOWN(e820.map[0].addr + e820.map[0].size);
515 for (i = 1; i < e820.nr_map; i++) {
516 struct e820entry *ei = &e820.map[i];
518 if (pfn < PFN_UP(ei->addr))
519 register_nosave_region(pfn, PFN_UP(ei->addr));
521 pfn = PFN_DOWN(ei->addr + ei->size);
522 if (ei->type != E820_RAM)
523 register_nosave_region(PFN_UP(ei->addr), pfn);
525 if (pfn >= limit_pfn)
532 * Early reserved memory areas.
534 #define MAX_EARLY_RES 20
540 static struct early_res early_res[MAX_EARLY_RES] __initdata = {
541 { 0, PAGE_SIZE, "BIOS data page" }, /* BIOS data page */
542 #if defined(CONFIG_X86_64) && defined(CONFIG_X86_TRAMPOLINE)
543 { TRAMPOLINE_BASE, TRAMPOLINE_BASE + 2 * PAGE_SIZE, "TRAMPOLINE" },
545 #if defined(CONFIG_X86_32) && defined(CONFIG_SMP)
547 * But first pinch a few for the stack/trampoline stuff
548 * FIXME: Don't need the extra page at 4K, but need to fix
549 * trampoline before removing it. (see the GDT stuff)
551 { PAGE_SIZE, PAGE_SIZE + PAGE_SIZE, "EX TRAMPOLINE" },
553 * Has to be in very low memory so we can execute
556 { TRAMPOLINE_BASE, TRAMPOLINE_BASE + PAGE_SIZE, "TRAMPOLINE" },
561 static int __init find_overlapped_early(u64 start, u64 end)
566 for (i = 0; i < MAX_EARLY_RES && early_res[i].end; i++) {
568 if (end > r->start && start < r->end)
575 void __init reserve_early(u64 start, u64 end, char *name)
580 i = find_overlapped_early(start, end);
581 if (i >= MAX_EARLY_RES)
582 panic("Too many early reservations");
585 panic("Overlapping early reservations "
586 "%llx-%llx %s to %llx-%llx %s\n",
587 start, end - 1, name?name:"", r->start,
588 r->end - 1, r->name);
592 strncpy(r->name, name, sizeof(r->name) - 1);
595 void __init free_early(u64 start, u64 end)
600 i = find_overlapped_early(start, end);
602 if (i >= MAX_EARLY_RES || r->end != end || r->start != start)
603 panic("free_early on not reserved area: %llx-%llx!",
606 for (j = i + 1; j < MAX_EARLY_RES && early_res[j].end; j++)
609 memmove(&early_res[i], &early_res[i + 1],
610 (j - 1 - i) * sizeof(struct early_res));
612 early_res[j - 1].end = 0;
615 int __init page_is_reserved_early(unsigned long pagenr)
617 u64 start = (u64)pagenr << PAGE_SHIFT;
621 i = find_overlapped_early(start, start + PAGE_SIZE);
623 return (i < MAX_EARLY_RES && r->end);
626 void __init early_res_to_bootmem(u64 start, u64 end)
629 u64 final_start, final_end;
630 for (i = 0; i < MAX_EARLY_RES && early_res[i].end; i++) {
631 struct early_res *r = &early_res[i];
632 final_start = max(start, r->start);
633 final_end = min(end, r->end);
634 if (final_start >= final_end)
636 printk(KERN_INFO " early res: %d [%llx-%llx] %s\n", i,
637 final_start, final_end - 1, r->name);
639 reserve_bootmem_generic(final_start, final_end - final_start);
641 reserve_bootmem(final_start, final_end - final_start,
647 /* Check for already reserved areas */
648 static inline int __init bad_addr(u64 *addrp, u64 size, u64 align)
655 i = find_overlapped_early(addr, addr + size);
657 if (i < MAX_EARLY_RES && r->end) {
658 *addrp = addr = round_up(r->end, align);
665 /* Check for already reserved areas */
666 static inline int __init bad_addr_size(u64 *addrp, u64 *sizep, u64 align)
669 u64 addr = *addrp, last;
674 for (i = 0; i < MAX_EARLY_RES && early_res[i].end; i++) {
675 struct early_res *r = &early_res[i];
676 if (last > r->start && addr < r->start) {
677 size = r->start - addr;
681 if (last > r->end && addr < r->end) {
682 addr = round_up(r->end, align);
687 if (last <= r->end && addr >= r->start) {
700 * Find a free area with specified alignment in a specific range.
702 u64 __init find_e820_area(u64 start, u64 end, u64 size, u64 align)
706 for (i = 0; i < e820.nr_map; i++) {
707 struct e820entry *ei = &e820.map[i];
711 if (ei->type != E820_RAM)
713 addr = round_up(ei->addr, align);
714 ei_last = ei->addr + ei->size;
716 addr = round_up(start, align);
719 while (bad_addr(&addr, size, align) && addr+size <= ei_last)
732 * Find next free range after *start
734 u64 __init find_e820_area_size(u64 start, u64 *sizep, u64 align)
738 for (i = 0; i < e820.nr_map; i++) {
739 struct e820entry *ei = &e820.map[i];
743 if (ei->type != E820_RAM)
745 addr = round_up(ei->addr, align);
746 ei_last = ei->addr + ei->size;
748 addr = round_up(start, align);
751 *sizep = ei_last - addr;
752 while (bad_addr_size(&addr, sizep, align) &&
753 addr + *sizep <= ei_last)
755 last = addr + *sizep;
765 * pre allocated 4k and reserved it in e820
767 u64 __init early_reserve_e820(u64 startt, u64 sizet, u64 align)
775 start = find_e820_area_size(start, &size, align);
780 addr = round_down(start + size - sizet, align);
781 update_memory_range(addr, sizet, E820_RAM, E820_RESERVED);
782 printk(KERN_INFO "update e820 for early_reserve_e820\n");
789 # ifdef CONFIG_X86_PAE
790 # define MAX_ARCH_PFN (1ULL<<(36-PAGE_SHIFT))
792 # define MAX_ARCH_PFN (1ULL<<(32-PAGE_SHIFT))
794 #else /* CONFIG_X86_32 */
795 # define MAX_ARCH_PFN MAXMEM>>PAGE_SHIFT
799 * Last pfn which the user wants to use.
801 unsigned long __initdata end_user_pfn = MAX_ARCH_PFN;
804 * Find the highest page frame number we have available
806 unsigned long __init e820_end_of_ram(void)
808 unsigned long last_pfn;
809 unsigned long max_arch_pfn = MAX_ARCH_PFN;
811 last_pfn = find_max_pfn_with_active_regions();
813 if (last_pfn > max_arch_pfn)
814 last_pfn = max_arch_pfn;
815 if (last_pfn > end_user_pfn)
816 last_pfn = end_user_pfn;
818 printk(KERN_INFO "last_pfn = %lu max_arch_pfn = %lu\n",
819 last_pfn, max_arch_pfn);
824 * Finds an active region in the address range from start_pfn to last_pfn and
825 * returns its range in ei_startpfn and ei_endpfn for the e820 entry.
827 int __init e820_find_active_region(const struct e820entry *ei,
828 unsigned long start_pfn,
829 unsigned long last_pfn,
830 unsigned long *ei_startpfn,
831 unsigned long *ei_endpfn)
833 u64 align = PAGE_SIZE;
835 *ei_startpfn = round_up(ei->addr, align) >> PAGE_SHIFT;
836 *ei_endpfn = round_down(ei->addr + ei->size, align) >> PAGE_SHIFT;
838 /* Skip map entries smaller than a page */
839 if (*ei_startpfn >= *ei_endpfn)
842 /* Skip if map is outside the node */
843 if (ei->type != E820_RAM || *ei_endpfn <= start_pfn ||
844 *ei_startpfn >= last_pfn)
847 /* Check for overlaps */
848 if (*ei_startpfn < start_pfn)
849 *ei_startpfn = start_pfn;
850 if (*ei_endpfn > last_pfn)
851 *ei_endpfn = last_pfn;
853 /* Obey end_user_pfn to save on memmap */
854 if (*ei_startpfn >= end_user_pfn)
856 if (*ei_endpfn > end_user_pfn)
857 *ei_endpfn = end_user_pfn;
862 /* Walk the e820 map and register active regions within a node */
863 void __init e820_register_active_regions(int nid, unsigned long start_pfn,
864 unsigned long last_pfn)
866 unsigned long ei_startpfn;
867 unsigned long ei_endpfn;
870 for (i = 0; i < e820.nr_map; i++)
871 if (e820_find_active_region(&e820.map[i],
873 &ei_startpfn, &ei_endpfn))
874 add_active_range(nid, ei_startpfn, ei_endpfn);
878 * Find the hole size (in bytes) in the memory range.
879 * @start: starting address of the memory range to scan
880 * @end: ending address of the memory range to scan
882 u64 __init e820_hole_size(u64 start, u64 end)
884 unsigned long start_pfn = start >> PAGE_SHIFT;
885 unsigned long last_pfn = end >> PAGE_SHIFT;
886 unsigned long ei_startpfn, ei_endpfn, ram = 0;
889 for (i = 0; i < e820.nr_map; i++) {
890 if (e820_find_active_region(&e820.map[i],
892 &ei_startpfn, &ei_endpfn))
893 ram += ei_endpfn - ei_startpfn;
895 return end - start - ((u64)ram << PAGE_SHIFT);